xref: /openbsd-src/sys/netinet6/nd6.c (revision b2ea75c1b17e1a9a339660e7ed45cd24946b230e)
1 /*	$OpenBSD: nd6.c,v 1.37 2001/06/29 16:10:36 itojun Exp $	*/
2 /*	$KAME: nd6.c,v 1.151 2001/06/19 14:24:41 sumikawa Exp $	*/
3 
4 /*
5  * Copyright (C) 1995, 1996, 1997, and 1998 WIDE Project.
6  * All rights reserved.
7  *
8  * Redistribution and use in source and binary forms, with or without
9  * modification, are permitted provided that the following conditions
10  * are met:
11  * 1. Redistributions of source code must retain the above copyright
12  *    notice, this list of conditions and the following disclaimer.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in the
15  *    documentation and/or other materials provided with the distribution.
16  * 3. Neither the name of the project nor the names of its contributors
17  *    may be used to endorse or promote products derived from this software
18  *    without specific prior written permission.
19  *
20  * THIS SOFTWARE IS PROVIDED BY THE PROJECT AND CONTRIBUTORS ``AS IS'' AND
21  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
22  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
23  * ARE DISCLAIMED.  IN NO EVENT SHALL THE PROJECT OR CONTRIBUTORS BE LIABLE
24  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
25  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
26  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
27  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
28  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
29  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
30  * SUCH DAMAGE.
31  */
32 
33 /*
34  * XXX
35  * KAME 970409 note:
36  * BSD/OS version heavily modifies this code, related to llinfo.
37  * Since we don't have BSD/OS version of net/route.c in our hand,
38  * I left the code mostly as it was in 970310.  -- itojun
39  */
40 
41 #include <sys/param.h>
42 #include <sys/systm.h>
43 #include <sys/timeout.h>
44 #include <sys/malloc.h>
45 #include <sys/mbuf.h>
46 #include <sys/socket.h>
47 #include <sys/sockio.h>
48 #include <sys/time.h>
49 #include <sys/kernel.h>
50 #include <sys/protosw.h>
51 #include <sys/errno.h>
52 #include <sys/ioctl.h>
53 #include <sys/syslog.h>
54 #include <sys/queue.h>
55 #include <dev/rndvar.h>
56 
57 #include <net/if.h>
58 #include <net/if_dl.h>
59 #include <net/if_types.h>
60 #include <net/if_atm.h>
61 #include <net/route.h>
62 
63 #include <netinet/in.h>
64 #include <netinet/if_ether.h>
65 #include <netinet/ip_ipsp.h>
66 
67 #include <netinet6/in6_var.h>
68 #include <netinet/ip6.h>
69 #include <netinet6/ip6_var.h>
70 #include <netinet6/nd6.h>
71 #include <netinet6/in6_prefix.h>
72 #include <netinet/icmp6.h>
73 
74 #define ND6_SLOWTIMER_INTERVAL (60 * 60) /* 1 hour */
75 #define ND6_RECALC_REACHTM_INTERVAL (60 * 120) /* 2 hours */
76 
77 #define SIN6(s) ((struct sockaddr_in6 *)s)
78 #define SDL(s) ((struct sockaddr_dl *)s)
79 
80 /* timer values */
81 int	nd6_prune	= 1;	/* walk list every 1 seconds */
82 int	nd6_delay	= 5;	/* delay first probe time 5 second */
83 int	nd6_umaxtries	= 3;	/* maximum unicast query */
84 int	nd6_mmaxtries	= 3;	/* maximum multicast query */
85 int	nd6_useloopback = 1;	/* use loopback interface for local traffic */
86 int	nd6_gctimer	= (60 * 60 * 24); /* 1 day: garbage collection timer */
87 
88 /* preventing too many loops in ND option parsing */
89 int nd6_maxndopt = 10;	/* max # of ND options allowed */
90 
91 int nd6_maxnudhint = 0;	/* max # of subsequent upper layer hints */
92 
93 #ifdef ND6_DEBUG
94 int nd6_debug = 1;
95 #else
96 int nd6_debug = 0;
97 #endif
98 
99 /* for debugging? */
100 static int nd6_inuse, nd6_allocated;
101 
102 struct llinfo_nd6 llinfo_nd6 = {&llinfo_nd6, &llinfo_nd6};
103 static size_t nd_ifinfo_indexlim = 8;
104 struct nd_ifinfo *nd_ifinfo = NULL;
105 struct nd_drhead nd_defrouter;
106 struct nd_prhead nd_prefix = { 0 };
107 
108 int nd6_recalc_reachtm_interval = ND6_RECALC_REACHTM_INTERVAL;
109 static struct sockaddr_in6 all1_sa;
110 
111 static void nd6_slowtimo __P((void *));
112 
113 struct timeout nd6_slowtimo_ch;
114 struct timeout nd6_timer_ch;
115 extern struct timeout in6_tmpaddrtimer_ch;
116 
117 void
118 nd6_init()
119 {
120 	static int nd6_init_done = 0;
121 	int i;
122 
123 	if (nd6_init_done) {
124 		log(LOG_NOTICE, "nd6_init called more than once(ignored)\n");
125 		return;
126 	}
127 
128 	all1_sa.sin6_family = AF_INET6;
129 	all1_sa.sin6_len = sizeof(struct sockaddr_in6);
130 	for (i = 0; i < sizeof(all1_sa.sin6_addr); i++)
131 		all1_sa.sin6_addr.s6_addr[i] = 0xff;
132 
133 	/* initialization of the default router list */
134 	TAILQ_INIT(&nd_defrouter);
135 
136 	nd6_init_done = 1;
137 
138 	/* start timer */
139 	timeout_set(&nd6_slowtimo_ch, nd6_slowtimo, NULL);
140 	timeout_add(&nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz);
141 }
142 
143 void
144 nd6_ifattach(ifp)
145 	struct ifnet *ifp;
146 {
147 
148 	/*
149 	 * We have some arrays that should be indexed by if_index.
150 	 * since if_index will grow dynamically, they should grow too.
151 	 */
152 	if (nd_ifinfo == NULL || if_index >= nd_ifinfo_indexlim) {
153 		size_t n;
154 		caddr_t q;
155 
156 		while (if_index >= nd_ifinfo_indexlim)
157 			nd_ifinfo_indexlim <<= 1;
158 
159 		/* grow nd_ifinfo */
160 		n = nd_ifinfo_indexlim * sizeof(struct nd_ifinfo);
161 		q = (caddr_t)malloc(n, M_IP6NDP, M_WAITOK);
162 		bzero(q, n);
163 		if (nd_ifinfo) {
164 			bcopy((caddr_t)nd_ifinfo, q, n/2);
165 			free((caddr_t)nd_ifinfo, M_IP6NDP);
166 		}
167 		nd_ifinfo = (struct nd_ifinfo *)q;
168 	}
169 
170 #define ND nd_ifinfo[ifp->if_index]
171 
172 	/*
173 	 * Don't initialize if called twice.
174 	 * XXX: to detect this, we should choose a member that is never set
175 	 * before initialization of the ND structure itself.  We formaly used
176 	 * the linkmtu member, which was not suitable because it could be
177 	 * initialized via "ifconfig mtu".
178 	 */
179 	if (ND.basereachable)
180 		return;
181 
182 	ND.linkmtu = ifindex2ifnet[ifp->if_index]->if_mtu;
183 	ND.chlim = IPV6_DEFHLIM;
184 	ND.basereachable = REACHABLE_TIME;
185 	ND.reachable = ND_COMPUTE_RTIME(ND.basereachable);
186 	ND.retrans = RETRANS_TIMER;
187 	ND.receivedra = 0;
188 	ND.flags = ND6_IFF_PERFORMNUD;
189 	nd6_setmtu(ifp);
190 #undef ND
191 }
192 
193 /*
194  * Reset ND level link MTU. This function is called when the physical MTU
195  * changes, which means we might have to adjust the ND level MTU.
196  */
197 void
198 nd6_setmtu(ifp)
199 	struct ifnet *ifp;
200 {
201 	struct nd_ifinfo *ndi = &nd_ifinfo[ifp->if_index];
202 	u_long oldmaxmtu = ndi->maxmtu;
203 	u_long oldlinkmtu = ndi->linkmtu;
204 
205 	switch (ifp->if_type) {
206 	case IFT_ARCNET:	/* XXX MTU handling needs more work */
207 		ndi->maxmtu = MIN(60480, ifp->if_mtu);
208 		break;
209 	case IFT_ETHER:
210 		ndi->maxmtu = MIN(ETHERMTU, ifp->if_mtu);
211 		break;
212 	case IFT_ATM:
213 		ndi->maxmtu = MIN(ATMMTU, ifp->if_mtu);
214 		break;
215 	default:
216 		ndi->maxmtu = ifp->if_mtu;
217 		break;
218 	}
219 
220 	if (oldmaxmtu != ndi->maxmtu) {
221 		/*
222 		 * If the ND level MTU is not set yet, or if the maxmtu
223 		 * is reset to a smaller value than the ND level MTU,
224 		 * also reset the ND level MTU.
225 		 */
226 		if (ndi->linkmtu == 0 ||
227 		    ndi->maxmtu < ndi->linkmtu) {
228 			ndi->linkmtu = ndi->maxmtu;
229 			/* also adjust in6_maxmtu if necessary. */
230 			if (oldlinkmtu == 0) {
231 				/*
232 				 * XXX: the case analysis is grotty, but
233 				 * it is not efficient to call in6_setmaxmtu()
234 				 * here when we are during the initialization
235 				 * procedure.
236 				 */
237 				if (in6_maxmtu < ndi->linkmtu)
238 					in6_maxmtu = ndi->linkmtu;
239 			} else
240 				in6_setmaxmtu();
241 		}
242 	}
243 #undef MIN
244 }
245 
246 void
247 nd6_option_init(opt, icmp6len, ndopts)
248 	void *opt;
249 	int icmp6len;
250 	union nd_opts *ndopts;
251 {
252 	bzero(ndopts, sizeof(*ndopts));
253 	ndopts->nd_opts_search = (struct nd_opt_hdr *)opt;
254 	ndopts->nd_opts_last
255 		= (struct nd_opt_hdr *)(((u_char *)opt) + icmp6len);
256 
257 	if (icmp6len == 0) {
258 		ndopts->nd_opts_done = 1;
259 		ndopts->nd_opts_search = NULL;
260 	}
261 }
262 
263 /*
264  * Take one ND option.
265  */
266 struct nd_opt_hdr *
267 nd6_option(ndopts)
268 	union nd_opts *ndopts;
269 {
270 	struct nd_opt_hdr *nd_opt;
271 	int olen;
272 
273 	if (!ndopts)
274 		panic("ndopts == NULL in nd6_option\n");
275 	if (!ndopts->nd_opts_last)
276 		panic("uninitialized ndopts in nd6_option\n");
277 	if (!ndopts->nd_opts_search)
278 		return NULL;
279 	if (ndopts->nd_opts_done)
280 		return NULL;
281 
282 	nd_opt = ndopts->nd_opts_search;
283 
284 	/* make sure nd_opt_len is inside the buffer */
285 	if ((caddr_t)&nd_opt->nd_opt_len >= (caddr_t)ndopts->nd_opts_last) {
286 		bzero(ndopts, sizeof(*ndopts));
287 		return NULL;
288 	}
289 
290 	olen = nd_opt->nd_opt_len << 3;
291 	if (olen == 0) {
292 		/*
293 		 * Message validation requires that all included
294 		 * options have a length that is greater than zero.
295 		 */
296 		bzero(ndopts, sizeof(*ndopts));
297 		return NULL;
298 	}
299 
300 	ndopts->nd_opts_search = (struct nd_opt_hdr *)((caddr_t)nd_opt + olen);
301 	if (ndopts->nd_opts_search > ndopts->nd_opts_last) {
302 		/* option overruns the end of buffer, invalid */
303 		bzero(ndopts, sizeof(*ndopts));
304 		return NULL;
305 	} else if (ndopts->nd_opts_search == ndopts->nd_opts_last) {
306 		/* reached the end of options chain */
307 		ndopts->nd_opts_done = 1;
308 		ndopts->nd_opts_search = NULL;
309 	}
310 	return nd_opt;
311 }
312 
313 /*
314  * Parse multiple ND options.
315  * This function is much easier to use, for ND routines that do not need
316  * multiple options of the same type.
317  */
318 int
319 nd6_options(ndopts)
320 	union nd_opts *ndopts;
321 {
322 	struct nd_opt_hdr *nd_opt;
323 	int i = 0;
324 
325 	if (!ndopts)
326 		panic("ndopts == NULL in nd6_options\n");
327 	if (!ndopts->nd_opts_last)
328 		panic("uninitialized ndopts in nd6_options\n");
329 	if (!ndopts->nd_opts_search)
330 		return 0;
331 
332 	while (1) {
333 		nd_opt = nd6_option(ndopts);
334 		if (!nd_opt && !ndopts->nd_opts_last) {
335 			/*
336 			 * Message validation requires that all included
337 			 * options have a length that is greater than zero.
338 			 */
339 			icmp6stat.icp6s_nd_badopt++;
340 			bzero(ndopts, sizeof(*ndopts));
341 			return -1;
342 		}
343 
344 		if (!nd_opt)
345 			goto skip1;
346 
347 		switch (nd_opt->nd_opt_type) {
348 		case ND_OPT_SOURCE_LINKADDR:
349 		case ND_OPT_TARGET_LINKADDR:
350 		case ND_OPT_MTU:
351 		case ND_OPT_REDIRECTED_HEADER:
352 			if (ndopts->nd_opt_array[nd_opt->nd_opt_type]) {
353 				nd6log((LOG_INFO,
354 				    "duplicated ND6 option found (type=%d)\n",
355 				    nd_opt->nd_opt_type));
356 				/* XXX bark? */
357 			} else {
358 				ndopts->nd_opt_array[nd_opt->nd_opt_type]
359 					= nd_opt;
360 			}
361 			break;
362 		case ND_OPT_PREFIX_INFORMATION:
363 			if (ndopts->nd_opt_array[nd_opt->nd_opt_type] == 0) {
364 				ndopts->nd_opt_array[nd_opt->nd_opt_type]
365 					= nd_opt;
366 			}
367 			ndopts->nd_opts_pi_end =
368 				(struct nd_opt_prefix_info *)nd_opt;
369 			break;
370 		default:
371 			/*
372 			 * Unknown options must be silently ignored,
373 			 * to accomodate future extension to the protocol.
374 			 */
375 			nd6log((LOG_DEBUG,
376 			    "nd6_options: unsupported option %d - "
377 			    "option ignored\n", nd_opt->nd_opt_type));
378 		}
379 
380 skip1:
381 		i++;
382 		if (i > nd6_maxndopt) {
383 			icmp6stat.icp6s_nd_toomanyopt++;
384 			nd6log((LOG_INFO, "too many loop in nd opt\n"));
385 			break;
386 		}
387 
388 		if (ndopts->nd_opts_done)
389 			break;
390 	}
391 
392 	return 0;
393 }
394 
395 /*
396  * ND6 timer routine to expire default route list and prefix list
397  */
398 void
399 nd6_timer(ignored_arg)
400 	void	*ignored_arg;
401 {
402 	int s;
403 	struct llinfo_nd6 *ln;
404 	struct nd_defrouter *dr;
405 	struct nd_prefix *pr;
406 	long time_second = time.tv_sec;
407 
408 	s = splnet();
409 
410 	timeout_set(&nd6_timer_ch, nd6_timer, NULL);
411 	timeout_add(&nd6_timer_ch, nd6_prune * hz);
412 
413 	ln = llinfo_nd6.ln_next;
414 	/* XXX BSD/OS separates this code -- itojun */
415 	while (ln && ln != &llinfo_nd6) {
416 		struct rtentry *rt;
417 		struct ifnet *ifp;
418 		struct sockaddr_in6 *dst;
419 		struct llinfo_nd6 *next = ln->ln_next;
420 		/* XXX: used for the DELAY case only: */
421 		struct nd_ifinfo *ndi = NULL;
422 
423 		if ((rt = ln->ln_rt) == NULL) {
424 			ln = next;
425 			continue;
426 		}
427 		if ((ifp = rt->rt_ifp) == NULL) {
428 			ln = next;
429 			continue;
430 		}
431 		ndi = &nd_ifinfo[ifp->if_index];
432 		dst = (struct sockaddr_in6 *)rt_key(rt);
433 
434 		if (ln->ln_expire > time_second) {
435 			ln = next;
436 			continue;
437 		}
438 
439 		/* sanity check */
440 		if (!rt)
441 			panic("rt=0 in nd6_timer(ln=%p)\n", ln);
442 		if (rt->rt_llinfo && (struct llinfo_nd6 *)rt->rt_llinfo != ln)
443 			panic("rt_llinfo(%p) is not equal to ln(%p)\n",
444 			      rt->rt_llinfo, ln);
445 		if (!dst)
446 			panic("dst=0 in nd6_timer(ln=%p)\n", ln);
447 
448 		switch (ln->ln_state) {
449 		case ND6_LLINFO_INCOMPLETE:
450 			if (ln->ln_asked < nd6_mmaxtries) {
451 				ln->ln_asked++;
452 				ln->ln_expire = time_second +
453 					nd_ifinfo[ifp->if_index].retrans / 1000;
454 				nd6_ns_output(ifp, NULL, &dst->sin6_addr,
455 					ln, 0);
456 			} else {
457 				struct mbuf *m = ln->ln_hold;
458 				if (m) {
459 					if (rt->rt_ifp) {
460 						/*
461 						 * Fake rcvif to make ICMP error
462 						 * more helpful in diagnosing
463 						 * for the receiver.
464 						 * XXX: should we consider
465 						 * older rcvif?
466 						 */
467 						m->m_pkthdr.rcvif = rt->rt_ifp;
468 					}
469 					icmp6_error(m, ICMP6_DST_UNREACH,
470 						    ICMP6_DST_UNREACH_ADDR, 0);
471 					ln->ln_hold = NULL;
472 				}
473 				next = nd6_free(rt);
474 			}
475 			break;
476 		case ND6_LLINFO_REACHABLE:
477 			if (ln->ln_expire) {
478 				ln->ln_state = ND6_LLINFO_STALE;
479 				ln->ln_expire = time_second + nd6_gctimer;
480 			}
481 			break;
482 
483 		case ND6_LLINFO_STALE:
484 			/* Garbage Collection(RFC 2461 5.3) */
485 			if (ln->ln_expire)
486 				next = nd6_free(rt);
487 			break;
488 
489 		case ND6_LLINFO_DELAY:
490 			if (ndi && (ndi->flags & ND6_IFF_PERFORMNUD) != 0) {
491 				/* We need NUD */
492 				ln->ln_asked = 1;
493 				ln->ln_state = ND6_LLINFO_PROBE;
494 				ln->ln_expire = time_second +
495 					ndi->retrans / 1000;
496 				nd6_ns_output(ifp, &dst->sin6_addr,
497 					      &dst->sin6_addr,
498 					      ln, 0);
499 			} else {
500 				ln->ln_state = ND6_LLINFO_STALE; /* XXX */
501 				ln->ln_expire = time_second + nd6_gctimer;
502 			}
503 			break;
504 		case ND6_LLINFO_PROBE:
505 			if (ln->ln_asked < nd6_umaxtries) {
506 				ln->ln_asked++;
507 				ln->ln_expire = time_second +
508 					nd_ifinfo[ifp->if_index].retrans / 1000;
509 				nd6_ns_output(ifp, &dst->sin6_addr,
510 					       &dst->sin6_addr, ln, 0);
511 			} else
512 				next = nd6_free(rt);
513 			break;
514 		}
515 		ln = next;
516 	}
517 
518 	/* expire default router list */
519 	dr = TAILQ_FIRST(&nd_defrouter);
520 	while (dr) {
521 		if (dr->expire && dr->expire < time_second) {
522 			struct nd_defrouter *t;
523 			t = TAILQ_NEXT(dr, dr_entry);
524 			defrtrlist_del(dr);
525 			dr = t;
526 		} else
527 			dr = TAILQ_NEXT(dr, dr_entry);
528 	}
529 	pr = nd_prefix.lh_first;
530 	while (pr) {
531 		struct in6_ifaddr *ia6;
532 		struct in6_addrlifetime *lt6;
533 
534 		if (IN6_IS_ADDR_UNSPECIFIED(&pr->ndpr_addr))
535 			ia6 = NULL;
536 		else
537 			ia6 = in6ifa_ifpwithaddr(pr->ndpr_ifp, &pr->ndpr_addr);
538 
539 		if (ia6) {
540 			/* check address lifetime */
541 			lt6 = &ia6->ia6_lifetime;
542 			if (lt6->ia6t_preferred && lt6->ia6t_preferred < time_second)
543 				ia6->ia6_flags |= IN6_IFF_DEPRECATED;
544 			if (lt6->ia6t_expire && lt6->ia6t_expire < time_second) {
545 				if (!IN6_IS_ADDR_UNSPECIFIED(&pr->ndpr_addr))
546 					in6_ifdel(pr->ndpr_ifp, &pr->ndpr_addr);
547 				/* xxx ND_OPT_PI_FLAG_ONLINK processing */
548 			}
549 		}
550 
551 		/*
552 		 * check prefix lifetime.
553 		 * since pltime is just for autoconf, pltime processing for
554 		 * prefix is not necessary.
555 		 *
556 		 * we offset expire time by NDPR_KEEP_EXPIRE, so that we
557 		 * can use the old prefix information to validate the
558 		 * next prefix information to come.  See prelist_update()
559 		 * for actual validation.
560 		 */
561 		if (pr->ndpr_expire
562 		 && pr->ndpr_expire + NDPR_KEEP_EXPIRED < time_second) {
563 			struct nd_prefix *t;
564 			t = pr->ndpr_next;
565 
566 			/*
567 			 * address expiration and prefix expiration are
568 			 * separate.  NEVER perform in6_ifdel here.
569 			 */
570 
571 			prelist_remove(pr);
572 			pr = t;
573 		} else
574 			pr = pr->ndpr_next;
575 	}
576 	splx(s);
577 }
578 
579 /*
580  * Nuke neighbor cache/prefix/default router management table, right before
581  * ifp goes away.
582  */
583 void
584 nd6_purge(ifp)
585 	struct ifnet *ifp;
586 {
587 	struct llinfo_nd6 *ln, *nln;
588 	struct nd_defrouter *dr, *ndr, drany;
589 	struct nd_prefix *pr, *npr;
590 
591 	/* Nuke default router list entries toward ifp */
592 	if ((dr = TAILQ_FIRST(&nd_defrouter)) != NULL) {
593 		/*
594 		 * The first entry of the list may be stored in
595 		 * the routing table, so we'll delete it later.
596 		 */
597 		for (dr = TAILQ_NEXT(dr, dr_entry); dr; dr = ndr) {
598 			ndr = TAILQ_NEXT(dr, dr_entry);
599 			if (dr->ifp == ifp)
600 				defrtrlist_del(dr);
601 		}
602 		dr = TAILQ_FIRST(&nd_defrouter);
603 		if (dr->ifp == ifp)
604 			defrtrlist_del(dr);
605 	}
606 
607 	/* Nuke prefix list entries toward ifp */
608 	for (pr = nd_prefix.lh_first; pr; pr = npr) {
609 		npr = pr->ndpr_next;
610 		if (pr->ndpr_ifp == ifp) {
611 			if (!IN6_IS_ADDR_UNSPECIFIED(&pr->ndpr_addr))
612 				in6_ifdel(pr->ndpr_ifp, &pr->ndpr_addr);
613 			prelist_remove(pr);
614 		}
615 	}
616 
617 	/* cancel default outgoing interface setting */
618 	if (nd6_defifindex == ifp->if_index)
619 		nd6_setdefaultiface(0);
620 
621 	if (!ip6_forwarding && ip6_accept_rtadv) { /* XXX: too restrictive? */
622 		/* refresh default router list */
623 		bzero(&drany, sizeof(drany));
624 		defrouter_delreq(&drany, 0);
625 		defrouter_select();
626 	}
627 
628 	/*
629 	 * Nuke neighbor cache entries for the ifp.
630 	 * Note that rt->rt_ifp may not be the same as ifp,
631 	 * due to KAME goto ours hack.  See RTM_RESOLVE case in
632 	 * nd6_rtrequest(), and ip6_input().
633 	 */
634 	ln = llinfo_nd6.ln_next;
635 	while (ln && ln != &llinfo_nd6) {
636 		struct rtentry *rt;
637 		struct sockaddr_dl *sdl;
638 
639 		nln = ln->ln_next;
640 		rt = ln->ln_rt;
641 		if (rt && rt->rt_gateway &&
642 		    rt->rt_gateway->sa_family == AF_LINK) {
643 			sdl = (struct sockaddr_dl *)rt->rt_gateway;
644 			if (sdl->sdl_index == ifp->if_index)
645 				nln = nd6_free(rt);
646 		}
647 		ln = nln;
648 	}
649 }
650 
651 struct rtentry *
652 nd6_lookup(addr6, create, ifp)
653 	struct in6_addr *addr6;
654 	int create;
655 	struct ifnet *ifp;
656 {
657 	struct rtentry *rt;
658 	struct sockaddr_in6 sin6;
659 
660 	bzero(&sin6, sizeof(sin6));
661 	sin6.sin6_len = sizeof(struct sockaddr_in6);
662 	sin6.sin6_family = AF_INET6;
663 	sin6.sin6_addr = *addr6;
664 	rt = rtalloc1((struct sockaddr *)&sin6, create);
665 	if (rt && (rt->rt_flags & RTF_LLINFO) == 0) {
666 		/*
667 		 * This is the case for the default route.
668 		 * If we want to create a neighbor cache for the address, we
669 		 * should free the route for the destination and allocate an
670 		 * interface route.
671 		 */
672 		if (create) {
673 			RTFREE(rt);
674 			rt = 0;
675 		}
676 	}
677 	if (!rt) {
678 		if (create && ifp) {
679 			int e;
680 
681 			/*
682 			 * If no route is available and create is set,
683 			 * we allocate a host route for the destination
684 			 * and treat it like an interface route.
685 			 * This hack is necessary for a neighbor which can't
686 			 * be covered by our own prefix.
687 			 */
688 			struct ifaddr *ifa =
689 				ifaof_ifpforaddr((struct sockaddr *)&sin6, ifp);
690 			if (ifa == NULL)
691 				return(NULL);
692 
693 			/*
694 			 * Create a new route. RTF_LLINFO is necessary
695 			 * to create a Neighbor Cache entry for the
696 			 * destination in nd6_rtrequest which will be
697 			 * called in rtequest via ifa->ifa_rtrequest.
698 			 */
699 			if ((e = rtrequest(RTM_ADD, (struct sockaddr *)&sin6,
700 					   ifa->ifa_addr,
701 					   (struct sockaddr *)&all1_sa,
702 					   (ifa->ifa_flags |
703 					    RTF_HOST | RTF_LLINFO) &
704 					   ~RTF_CLONING,
705 					   &rt)) != 0)
706 				log(LOG_ERR,
707 				    "nd6_lookup: failed to add route for a "
708 				    "neighbor(%s), errno=%d\n",
709 				    ip6_sprintf(addr6), e);
710 			if (rt == NULL)
711 				return(NULL);
712 			if (rt->rt_llinfo) {
713 				struct llinfo_nd6 *ln =
714 					(struct llinfo_nd6 *)rt->rt_llinfo;
715 				ln->ln_state = ND6_LLINFO_NOSTATE;
716 			}
717 		} else
718 			return(NULL);
719 	}
720 	rt->rt_refcnt--;
721 	/*
722 	 * Validation for the entry.
723 	 * XXX: we can't use rt->rt_ifp to check for the interface, since
724 	 *      it might be the loopback interface if the entry is for our
725 	 *      own address on a non-loopback interface. Instead, we should
726 	 *      use rt->rt_ifa->ifa_ifp, which would specify the REAL interface.
727 	 */
728 	if ((rt->rt_flags & RTF_GATEWAY) || (rt->rt_flags & RTF_LLINFO) == 0 ||
729 	    rt->rt_gateway->sa_family != AF_LINK ||
730 	    (ifp && rt->rt_ifa->ifa_ifp != ifp)) {
731 		if (create) {
732 			log(LOG_DEBUG, "nd6_lookup: failed to lookup %s (if = %s)\n",
733 			    ip6_sprintf(addr6), ifp ? ifp->if_xname : "unspec");
734 			/* xxx more logs... kazu */
735 		}
736 		return(0);
737 	}
738 	return(rt);
739 }
740 
741 /*
742  * Detect if a given IPv6 address identifies a neighbor on a given link.
743  * XXX: should take care of the destination of a p2p link?
744  */
745 int
746 nd6_is_addr_neighbor(addr, ifp)
747 	struct sockaddr_in6 *addr;
748 	struct ifnet *ifp;
749 {
750 	struct ifaddr *ifa;
751 	int i;
752 
753 #define IFADDR6(a) ((((struct in6_ifaddr *)(a))->ia_addr).sin6_addr)
754 #define IFMASK6(a) ((((struct in6_ifaddr *)(a))->ia_prefixmask).sin6_addr)
755 
756 	/*
757 	 * A link-local address is always a neighbor.
758 	 * XXX: we should use the sin6_scope_id field rather than the embedded
759 	 * interface index.
760 	 */
761 	if (IN6_IS_ADDR_LINKLOCAL(&addr->sin6_addr) &&
762 	    ntohs(*(u_int16_t *)&addr->sin6_addr.s6_addr[2]) == ifp->if_index)
763 		return(1);
764 
765 	/*
766 	 * If the address matches one of our addresses,
767 	 * it should be a neighbor.
768 	 */
769 	for (ifa = ifp->if_addrlist.tqh_first;
770 	     ifa;
771 	     ifa = ifa->ifa_list.tqe_next)
772 	{
773 		if (ifa->ifa_addr->sa_family != AF_INET6)
774 			next: continue;
775 
776 		for (i = 0; i < 4; i++) {
777 			if ((IFADDR6(ifa).s6_addr32[i] ^
778 			     addr->sin6_addr.s6_addr32[i]) &
779 			    IFMASK6(ifa).s6_addr32[i])
780 				goto next;
781 		}
782 		return(1);
783 	}
784 
785 	/*
786 	 * Even if the address matches none of our addresses, it might be
787 	 * in the neighbor cache.
788 	 */
789 	if (nd6_lookup(&addr->sin6_addr, 0, ifp))
790 		return(1);
791 
792 	return(0);
793 #undef IFADDR6
794 #undef IFMASK6
795 }
796 
797 /*
798  * Free an nd6 llinfo entry.
799  */
800 struct llinfo_nd6 *
801 nd6_free(rt)
802 	struct rtentry *rt;
803 {
804 	struct llinfo_nd6 *ln = (struct llinfo_nd6 *)rt->rt_llinfo, *next;
805 	struct in6_addr in6 = ((struct sockaddr_in6 *)rt_key(rt))->sin6_addr;
806 	struct nd_defrouter *dr;
807 
808 	/*
809 	 * Clear all destination cache entries for the neighbor.
810 	 * XXX: is it better to restrict this to hosts?
811 	 */
812 	pfctlinput(PRC_HOSTDEAD, rt_key(rt));
813 
814 	if (!ip6_forwarding && ip6_accept_rtadv) { /* XXX: too restrictive? */
815 		int s;
816 		s = splnet();
817 		dr = defrouter_lookup(&((struct sockaddr_in6 *)rt_key(rt))->sin6_addr,
818 				      rt->rt_ifp);
819 		if (ln->ln_router || dr) {
820 			/*
821 			 * rt6_flush must be called whether or not the neighbor
822 			 * is in the Default Router List.
823 			 * See a corresponding comment in nd6_na_input().
824 			 */
825 			rt6_flush(&in6, rt->rt_ifp);
826 		}
827 
828 		if (dr) {
829 			/*
830 			 * Unreachablity of a router might affect the default
831 			 * router selection and on-link detection of advertised
832 			 * prefixes.
833 			 */
834 
835 			/*
836 			 * Temporarily fake the state to choose a new default
837 			 * router and to perform on-link determination of
838 			 * prefixes coreectly.
839 			 * Below the state will be set correctly,
840 			 * or the entry itself will be deleted.
841 			 */
842 			ln->ln_state = ND6_LLINFO_INCOMPLETE;
843 
844 			if (dr == TAILQ_FIRST(&nd_defrouter)) {
845 				/*
846 				 * It is used as the current default router,
847 				 * so we have to move it to the end of the
848 				 * list and choose a new one.
849 				 * XXX: it is not very efficient if this is
850 				 *      the only router.
851 				 */
852 				TAILQ_REMOVE(&nd_defrouter, dr, dr_entry);
853 				TAILQ_INSERT_TAIL(&nd_defrouter, dr, dr_entry);
854 
855 				defrouter_select();
856 			}
857 			pfxlist_onlink_check();
858 		}
859 		splx(s);
860 	}
861 
862 	/*
863 	 * Before deleting the entry, remember the next entry as the
864 	 * return value.  We need this because pfxlist_onlink_check() above
865 	 * might have freed other entries (particularly the old next entry) as
866 	 * a side effect (XXX).
867 	 */
868 	next = ln->ln_next;
869 
870 	/*
871 	 * Detach the route from the routing tree and the list of neighbor
872 	 * caches, and disable the route entry not to be used in already
873 	 * cached routes.
874 	 */
875 	rtrequest(RTM_DELETE, rt_key(rt), (struct sockaddr *)0,
876 		  rt_mask(rt), 0, (struct rtentry **)0);
877 
878 	return(next);
879 }
880 
881 /*
882  * Upper-layer reachability hint for Neighbor Unreachability Detection.
883  *
884  * XXX cost-effective metods?
885  */
886 void
887 nd6_nud_hint(rt, dst6, force)
888 	struct rtentry *rt;
889 	struct in6_addr *dst6;
890 	int force;
891 {
892 	struct llinfo_nd6 *ln;
893 	long time_second = time.tv_sec;
894 
895 	/*
896 	 * If the caller specified "rt", use that.  Otherwise, resolve the
897 	 * routing table by supplied "dst6".
898 	 */
899 	if (!rt) {
900 		if (!dst6)
901 			return;
902 		if (!(rt = nd6_lookup(dst6, 0, NULL)))
903 			return;
904 	}
905 
906 	if ((rt->rt_flags & RTF_GATEWAY) != 0 ||
907 	    (rt->rt_flags & RTF_LLINFO) == 0 ||
908 	    !rt->rt_llinfo || !rt->rt_gateway ||
909 	    rt->rt_gateway->sa_family != AF_LINK) {
910 		/* This is not a host route. */
911 		return;
912 	}
913 
914 	ln = (struct llinfo_nd6 *)rt->rt_llinfo;
915 	if (ln->ln_state < ND6_LLINFO_REACHABLE)
916 		return;
917 
918 	/*
919 	 * if we get upper-layer reachability confirmation many times,
920 	 * it is possible we have false information.
921 	 */
922 	if (!force) {
923 		ln->ln_byhint++;
924 		if (ln->ln_byhint > nd6_maxnudhint)
925 			return;
926 	}
927 
928 	ln->ln_state = ND6_LLINFO_REACHABLE;
929 	if (ln->ln_expire)
930 		ln->ln_expire = time_second +
931 			nd_ifinfo[rt->rt_ifp->if_index].reachable;
932 }
933 
934 #ifdef OLDIP6OUTPUT
935 /*
936  * Resolve an IP6 address into an ethernet address. If success,
937  * desten is filled in. If there is no entry in ndptab,
938  * set one up and multicast a solicitation for the IP6 address.
939  * Hold onto this mbuf and resend it once the address
940  * is finally resolved. A return value of 1 indicates
941  * that desten has been filled in and the packet should be sent
942  * normally; a 0 return indicates that the packet has been
943  * taken over here, either now or for later transmission.
944  */
945 int
946 nd6_resolve(ifp, rt, m, dst, desten)
947 	struct ifnet *ifp;
948 	struct rtentry *rt;
949 	struct mbuf *m;
950 	struct sockaddr *dst;
951 	u_char *desten;
952 {
953 	struct llinfo_nd6 *ln = (struct llinfo_nd6 *)NULL;
954 	struct sockaddr_dl *sdl;
955 	long time_second = time.tv_sec;
956 
957 	if (m->m_flags & M_MCAST) {
958 		switch (ifp->if_type) {
959 		case IFT_ETHER:
960 		case IFT_FDDI:
961 			ETHER_MAP_IPV6_MULTICAST(&SIN6(dst)->sin6_addr,
962 						 desten);
963 			return(1);
964 			break;
965 		case IFT_ARCNET:
966 			*desten = 0;
967 			return(1);
968 		default:
969 			m_freem(m);
970 			return(0);
971 		}
972 	}
973 	if (rt && (rt->rt_flags & RTF_LLINFO) != 0)
974 		ln = (struct llinfo_nd6 *)rt->rt_llinfo;
975 	else {
976 		if ((rt = nd6_lookup(&(SIN6(dst)->sin6_addr), 1, ifp)) != NULL)
977 			ln = (struct llinfo_nd6 *)rt->rt_llinfo;
978 	}
979 	if (!ln || !rt) {
980 		log(LOG_DEBUG, "nd6_resolve: can't allocate llinfo for %s\n",
981 			ip6_sprintf(&(SIN6(dst)->sin6_addr)));
982 		m_freem(m);
983 		return(0);
984 	}
985 	sdl = SDL(rt->rt_gateway);
986 	/*
987 	 * Ckeck the address family and length is valid, the address
988 	 * is resolved; otherwise, try to resolve.
989 	 */
990 	if (ln->ln_state >= ND6_LLINFO_REACHABLE
991 	   && sdl->sdl_family == AF_LINK
992 	   && sdl->sdl_alen != 0) {
993 		bcopy(LLADDR(sdl), desten, sdl->sdl_alen);
994 		if (ln->ln_state == ND6_LLINFO_STALE) {
995 			ln->ln_asked = 0;
996 			ln->ln_state = ND6_LLINFO_DELAY;
997 			ln->ln_expire = time_second + nd6_delay;
998 		}
999 		return(1);
1000 	}
1001 	/*
1002 	 * There is an ndp entry, but no ethernet address
1003 	 * response yet. Replace the held mbuf with this
1004 	 * latest one.
1005 	 *
1006 	 * XXX Does the code conform to rate-limiting rule?
1007 	 * (RFC 2461 7.2.2)
1008 	 */
1009 	if (ln->ln_state == ND6_LLINFO_NOSTATE)
1010 		ln->ln_state = ND6_LLINFO_INCOMPLETE;
1011 	if (ln->ln_hold)
1012 		m_freem(ln->ln_hold);
1013 	ln->ln_hold = m;
1014 	if (ln->ln_expire) {
1015 		if (ln->ln_asked < nd6_mmaxtries &&
1016 		    ln->ln_expire < time_second) {
1017 			ln->ln_asked++;
1018 			ln->ln_expire = time_second +
1019 				nd_ifinfo[ifp->if_index].retrans / 1000;
1020 			nd6_ns_output(ifp, NULL, &(SIN6(dst)->sin6_addr),
1021 				ln, 0);
1022 		}
1023 	}
1024 	/* do not free mbuf here, it is queued into llinfo_nd6 */
1025 	return(0);
1026 }
1027 #endif /* OLDIP6OUTPUT */
1028 
1029 void
1030 nd6_rtrequest(req, rt, info)
1031 	int	req;
1032 	struct rtentry *rt;
1033 	struct rt_addrinfo *info; /* xxx unused */
1034 {
1035 	struct sockaddr *gate = rt->rt_gateway;
1036 	struct llinfo_nd6 *ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1037 	static struct sockaddr_dl null_sdl = {sizeof(null_sdl), AF_LINK};
1038 	struct ifnet *ifp = rt->rt_ifp;
1039 	struct ifaddr *ifa;
1040 	long time_second = time.tv_sec;
1041 
1042 	if (rt->rt_flags & RTF_GATEWAY)
1043 		return;
1044 
1045 	switch (req) {
1046 	case RTM_ADD:
1047 		/*
1048 		 * There is no backward compatibility :)
1049 		 *
1050 		 * if ((rt->rt_flags & RTF_HOST) == 0 &&
1051 		 *     SIN(rt_mask(rt))->sin_addr.s_addr != 0xffffffff)
1052 		 *	   rt->rt_flags |= RTF_CLONING;
1053 		 */
1054 		if (rt->rt_flags & (RTF_CLONING | RTF_LLINFO)) {
1055 			/*
1056 			 * Case 1: This route should come from
1057 			 * a route to interface. RTF_LLINFO flag is set
1058 			 * for a host route whose destination should be
1059 			 * treated as on-link.
1060 			 */
1061 			rt_setgate(rt, rt_key(rt),
1062 				   (struct sockaddr *)&null_sdl);
1063 			gate = rt->rt_gateway;
1064 			SDL(gate)->sdl_type = ifp->if_type;
1065 			SDL(gate)->sdl_index = ifp->if_index;
1066 			if (ln)
1067 				ln->ln_expire = time_second;
1068 #if 1
1069 			if (ln && ln->ln_expire == 0) {
1070 				/* kludge for desktops */
1071 #if 0
1072 				printf("nd6_request: time.tv_sec is zero; "
1073 				       "treat it as 1\n");
1074 #endif
1075 				ln->ln_expire = 1;
1076 			}
1077 #endif
1078 			if (rt->rt_flags & RTF_CLONING)
1079 				break;
1080 		}
1081 		/*
1082 		 * In IPv4 code, we try to annonuce new RTF_ANNOUNCE entry here.
1083 		 * We don't do that here since llinfo is not ready yet.
1084 		 *
1085 		 * There are also couple of other things to be discussed:
1086 		 * - unsolicited NA code needs improvement beforehand
1087 		 * - RFC2461 says we MAY send multicast unsolicited NA
1088 		 *   (7.2.6 paragraph 4), however, it also says that we
1089 		 *   SHOULD provide a mechanism to prevent multicast NA storm.
1090 		 *   we don't have anything like it right now.
1091 		 *   note that the mechanism needs a mutual agreement
1092 		 *   between proxies, which means that we need to implement
1093 		 *   a new protocol, or a new kludge.
1094 		 * - from RFC2461 6.2.4, host MUST NOT send an unsolicited NA.
1095 		 *   we need to check ip6forwarding before sending it.
1096 		 *   (or should we allow proxy ND configuration only for
1097 		 *   routers?  there's no mention about proxy ND from hosts)
1098 		 */
1099 #if 0
1100 		/* XXX it does not work */
1101 		if (rt->rt_flags & RTF_ANNOUNCE)
1102 			nd6_na_output(ifp,
1103 			      &SIN6(rt_key(rt))->sin6_addr,
1104 			      &SIN6(rt_key(rt))->sin6_addr,
1105 			      ip6_forwarding ? ND_NA_FLAG_ROUTER : 0,
1106 			      1, NULL);
1107 #endif
1108 		/* FALLTHROUGH */
1109 	case RTM_RESOLVE:
1110 		if ((ifp->if_flags & IFF_POINTOPOINT) == 0) {
1111 			/*
1112 			 * Address resolution isn't necessary for a point to
1113 			 * point link, so we can skip this test for a p2p link.
1114 			 */
1115 			if (gate->sa_family != AF_LINK ||
1116 			    gate->sa_len < sizeof(null_sdl)) {
1117 				log(LOG_DEBUG,
1118 				    "nd6_rtrequest: bad gateway value: %s\n",
1119 				    ifp->if_xname);
1120 				break;
1121 			}
1122 			SDL(gate)->sdl_type = ifp->if_type;
1123 			SDL(gate)->sdl_index = ifp->if_index;
1124 		}
1125 		if (ln != NULL)
1126 			break;	/* This happens on a route change */
1127 		/*
1128 		 * Case 2: This route may come from cloning, or a manual route
1129 		 * add with a LL address.
1130 		 */
1131 		R_Malloc(ln, struct llinfo_nd6 *, sizeof(*ln));
1132 		rt->rt_llinfo = (caddr_t)ln;
1133 		if (!ln) {
1134 			log(LOG_DEBUG, "nd6_rtrequest: malloc failed\n");
1135 			break;
1136 		}
1137 		nd6_inuse++;
1138 		nd6_allocated++;
1139 		Bzero(ln, sizeof(*ln));
1140 		ln->ln_rt = rt;
1141 		/* this is required for "ndp" command. - shin */
1142 		if (req == RTM_ADD) {
1143 		        /*
1144 			 * gate should have some valid AF_LINK entry,
1145 			 * and ln->ln_expire should have some lifetime
1146 			 * which is specified by ndp command.
1147 			 */
1148 			ln->ln_state = ND6_LLINFO_REACHABLE;
1149 			ln->ln_byhint = 0;
1150 		} else {
1151 		        /*
1152 			 * When req == RTM_RESOLVE, rt is created and
1153 			 * initialized in rtrequest(), so rt_expire is 0.
1154 			 */
1155 			ln->ln_state = ND6_LLINFO_NOSTATE;
1156 			ln->ln_expire = time_second;
1157 		}
1158 		rt->rt_flags |= RTF_LLINFO;
1159 		ln->ln_next = llinfo_nd6.ln_next;
1160 		llinfo_nd6.ln_next = ln;
1161 		ln->ln_prev = &llinfo_nd6;
1162 		ln->ln_next->ln_prev = ln;
1163 
1164 		/*
1165 		 * check if rt_key(rt) is one of my address assigned
1166 		 * to the interface.
1167 		 */
1168 		ifa = (struct ifaddr *)in6ifa_ifpwithaddr(rt->rt_ifp,
1169 					  &SIN6(rt_key(rt))->sin6_addr);
1170 		if (ifa) {
1171 			caddr_t macp = nd6_ifptomac(ifp);
1172 			ln->ln_expire = 0;
1173 			ln->ln_state = ND6_LLINFO_REACHABLE;
1174 			ln->ln_byhint = 0;
1175 			if (macp) {
1176 				Bcopy(macp, LLADDR(SDL(gate)), ifp->if_addrlen);
1177 				SDL(gate)->sdl_alen = ifp->if_addrlen;
1178 			}
1179 			if (nd6_useloopback) {
1180 				rt->rt_ifp = lo0ifp;	/*XXX*/
1181 				/*
1182 				 * Make sure rt_ifa be equal to the ifaddr
1183 				 * corresponding to the address.
1184 				 * We need this because when we refer
1185 				 * rt_ifa->ia6_flags in ip6_input, we assume
1186 				 * that the rt_ifa points to the address instead
1187 				 * of the loopback address.
1188 				 */
1189 				if (ifa != rt->rt_ifa) {
1190 					IFAFREE(rt->rt_ifa);
1191 					ifa->ifa_refcnt++;
1192 					rt->rt_ifa = ifa;
1193 				}
1194 			}
1195 		} else if (rt->rt_flags & RTF_ANNOUNCE) {
1196 			ln->ln_expire = 0;
1197 			ln->ln_state = ND6_LLINFO_REACHABLE;
1198 			ln->ln_byhint = 0;
1199 
1200 			/* join solicited node multicast for proxy ND */
1201 			if (ifp->if_flags & IFF_MULTICAST) {
1202 				struct in6_addr llsol;
1203 				int error;
1204 
1205 				llsol = SIN6(rt_key(rt))->sin6_addr;
1206 				llsol.s6_addr16[0] = htons(0xff02);
1207 				llsol.s6_addr16[1] = htons(ifp->if_index);
1208 				llsol.s6_addr32[1] = 0;
1209 				llsol.s6_addr32[2] = htonl(1);
1210 				llsol.s6_addr8[12] = 0xff;
1211 
1212 				(void)in6_addmulti(&llsol, ifp, &error);
1213 				if (error)
1214 					printf(
1215 "nd6_rtrequest: could not join solicited node multicast (errno=%d)\n", error);
1216 			}
1217 		}
1218 		break;
1219 
1220 	case RTM_DELETE:
1221 		if (!ln)
1222 			break;
1223 		/* leave from solicited node multicast for proxy ND */
1224 		if ((rt->rt_flags & RTF_ANNOUNCE) != 0 &&
1225 		    (ifp->if_flags & IFF_MULTICAST) != 0) {
1226 			struct in6_addr llsol;
1227 			struct in6_multi *in6m;
1228 
1229 			llsol = SIN6(rt_key(rt))->sin6_addr;
1230 			llsol.s6_addr16[0] = htons(0xff02);
1231 			llsol.s6_addr16[1] = htons(ifp->if_index);
1232 			llsol.s6_addr32[1] = 0;
1233 			llsol.s6_addr32[2] = htonl(1);
1234 			llsol.s6_addr8[12] = 0xff;
1235 
1236 			IN6_LOOKUP_MULTI(llsol, ifp, in6m);
1237 			if (in6m)
1238 				in6_delmulti(in6m);
1239 		}
1240 		nd6_inuse--;
1241 		ln->ln_next->ln_prev = ln->ln_prev;
1242 		ln->ln_prev->ln_next = ln->ln_next;
1243 		ln->ln_prev = NULL;
1244 		rt->rt_llinfo = 0;
1245 		rt->rt_flags &= ~RTF_LLINFO;
1246 		if (ln->ln_hold)
1247 			m_freem(ln->ln_hold);
1248 		Free((caddr_t)ln);
1249 	}
1250 }
1251 
1252 void
1253 nd6_p2p_rtrequest(req, rt, info)
1254 	int	req;
1255 	struct rtentry *rt;
1256 	struct rt_addrinfo *info; /* xxx unused */
1257 {
1258 	struct sockaddr *gate = rt->rt_gateway;
1259 	static struct sockaddr_dl null_sdl = {sizeof(null_sdl), AF_LINK};
1260 	struct ifnet *ifp = rt->rt_ifp;
1261 	struct ifaddr *ifa;
1262 
1263 	if (rt->rt_flags & RTF_GATEWAY)
1264 		return;
1265 
1266 	switch (req) {
1267 	case RTM_ADD:
1268 		/*
1269 		 * There is no backward compatibility :)
1270 		 *
1271 		 * if ((rt->rt_flags & RTF_HOST) == 0 &&
1272 		 *     SIN(rt_mask(rt))->sin_addr.s_addr != 0xffffffff)
1273 		 *	   rt->rt_flags |= RTF_CLONING;
1274 		 */
1275 		if (rt->rt_flags & RTF_CLONING) {
1276 			/*
1277 			 * Case 1: This route should come from
1278 			 * a route to interface.
1279 			 */
1280 			rt_setgate(rt, rt_key(rt),
1281 				   (struct sockaddr *)&null_sdl);
1282 			gate = rt->rt_gateway;
1283 			SDL(gate)->sdl_type = ifp->if_type;
1284 			SDL(gate)->sdl_index = ifp->if_index;
1285 			break;
1286 		}
1287 		/* Announce a new entry if requested. */
1288 		if (rt->rt_flags & RTF_ANNOUNCE)
1289 			nd6_na_output(ifp,
1290 				      &SIN6(rt_key(rt))->sin6_addr,
1291 				      &SIN6(rt_key(rt))->sin6_addr,
1292 				      ip6_forwarding ? ND_NA_FLAG_ROUTER : 0,
1293 				      1, NULL);
1294 		/* FALLTHROUGH */
1295 	case RTM_RESOLVE:
1296 		/*
1297 		 * check if rt_key(rt) is one of my address assigned
1298 		 * to the interface.
1299 		 */
1300  		ifa = (struct ifaddr *)in6ifa_ifpwithaddr(rt->rt_ifp,
1301 					  &SIN6(rt_key(rt))->sin6_addr);
1302 		if (ifa) {
1303 			if (nd6_useloopback) {
1304 				rt->rt_ifp = lo0ifp;	/*XXX*/
1305 			}
1306 		}
1307 		break;
1308 	}
1309 }
1310 
1311 int
1312 nd6_ioctl(cmd, data, ifp)
1313 	u_long cmd;
1314 	caddr_t	data;
1315 	struct ifnet *ifp;
1316 {
1317 	struct in6_drlist *drl = (struct in6_drlist *)data;
1318 	struct in6_prlist *prl = (struct in6_prlist *)data;
1319 	struct in6_ndireq *ndi = (struct in6_ndireq *)data;
1320 	struct in6_nbrinfo *nbi = (struct in6_nbrinfo *)data;
1321 	struct in6_ndifreq *ndif = (struct in6_ndifreq *)data;
1322 	struct nd_defrouter *dr, any;
1323 	struct nd_prefix *pr;
1324 	struct rtentry *rt;
1325 	int i = 0, error = 0;
1326 	int s;
1327 
1328 	switch (cmd) {
1329 	case SIOCGDRLST_IN6:
1330 		bzero(drl, sizeof(*drl));
1331 		s = splnet();
1332 		dr = TAILQ_FIRST(&nd_defrouter);
1333 		while (dr && i < DRLSTSIZ) {
1334 			drl->defrouter[i].rtaddr = dr->rtaddr;
1335 			if (IN6_IS_ADDR_LINKLOCAL(&drl->defrouter[i].rtaddr)) {
1336 				/* XXX: need to this hack for KAME stack */
1337 				drl->defrouter[i].rtaddr.s6_addr16[1] = 0;
1338 			} else
1339 				log(LOG_ERR,
1340 				    "default router list contains a "
1341 				    "non-linklocal address(%s)\n",
1342 				    ip6_sprintf(&drl->defrouter[i].rtaddr));
1343 
1344 			drl->defrouter[i].flags = dr->flags;
1345 			drl->defrouter[i].rtlifetime = dr->rtlifetime;
1346 			drl->defrouter[i].expire = dr->expire;
1347 			drl->defrouter[i].if_index = dr->ifp->if_index;
1348 			i++;
1349 			dr = TAILQ_NEXT(dr, dr_entry);
1350 		}
1351 		splx(s);
1352 		break;
1353 	case SIOCGPRLST_IN6:
1354 		/*
1355 		 * XXX meaning of fields, especialy "raflags", is very
1356 		 * differnet between RA prefix list and RR/static prefix list.
1357 		 * how about separating ioctls into two?
1358 		 */
1359 		bzero(prl, sizeof(*prl));
1360 		s = splnet();
1361 		pr = nd_prefix.lh_first;
1362 		while (pr && i < PRLSTSIZ) {
1363 			struct nd_pfxrouter *pfr;
1364 			int j;
1365 
1366 			prl->prefix[i].prefix = pr->ndpr_prefix.sin6_addr;
1367 			prl->prefix[i].raflags = pr->ndpr_raf;
1368 			prl->prefix[i].prefixlen = pr->ndpr_plen;
1369 			prl->prefix[i].vltime = pr->ndpr_vltime;
1370 			prl->prefix[i].pltime = pr->ndpr_pltime;
1371 			prl->prefix[i].if_index = pr->ndpr_ifp->if_index;
1372 			prl->prefix[i].expire = pr->ndpr_expire;
1373 
1374 			pfr = pr->ndpr_advrtrs.lh_first;
1375 			j = 0;
1376 			while(pfr) {
1377 				if (j < DRLSTSIZ) {
1378 #define RTRADDR prl->prefix[i].advrtr[j]
1379 					RTRADDR = pfr->router->rtaddr;
1380 					if (IN6_IS_ADDR_LINKLOCAL(&RTRADDR)) {
1381 						/* XXX: hack for KAME */
1382 						RTRADDR.s6_addr16[1] = 0;
1383 					} else
1384 						log(LOG_ERR,
1385 						    "a router(%s) advertises "
1386 						    "a prefix with "
1387 						    "non-link local address\n",
1388 						    ip6_sprintf(&RTRADDR));
1389 #undef RTRADDR
1390 				}
1391 				j++;
1392 				pfr = pfr->pfr_next;
1393 			}
1394 			prl->prefix[i].advrtrs = j;
1395 			prl->prefix[i].origin = PR_ORIG_RA;
1396 
1397 			i++;
1398 			pr = pr->ndpr_next;
1399 		}
1400 	      {
1401 		struct rr_prefix *rpp;
1402 
1403 		for (rpp = LIST_FIRST(&rr_prefix); rpp;
1404 		     rpp = LIST_NEXT(rpp, rp_entry)) {
1405 			if (i >= PRLSTSIZ)
1406 				break;
1407 			prl->prefix[i].prefix = rpp->rp_prefix.sin6_addr;
1408 			prl->prefix[i].raflags = rpp->rp_raf;
1409 			prl->prefix[i].prefixlen = rpp->rp_plen;
1410 			prl->prefix[i].vltime = rpp->rp_vltime;
1411 			prl->prefix[i].pltime = rpp->rp_pltime;
1412 			prl->prefix[i].if_index = rpp->rp_ifp->if_index;
1413 			prl->prefix[i].expire = rpp->rp_expire;
1414 			prl->prefix[i].advrtrs = 0;
1415 			prl->prefix[i].origin = rpp->rp_origin;
1416 			i++;
1417 		}
1418 	      }
1419 		splx(s);
1420 
1421 		break;
1422 	case SIOCGIFINFO_IN6:
1423 		if (!nd_ifinfo || i >= nd_ifinfo_indexlim) {
1424 			error = EINVAL;
1425 			break;
1426 		}
1427 		ndi->ndi = nd_ifinfo[ifp->if_index];
1428 		break;
1429 	case SIOCSIFINFO_FLAGS:
1430 		/* XXX: almost all other fields of ndi->ndi is unused */
1431 		if (!nd_ifinfo || i >= nd_ifinfo_indexlim) {
1432 			error = EINVAL;
1433 			break;
1434 		}
1435 		nd_ifinfo[ifp->if_index].flags = ndi->ndi.flags;
1436 		break;
1437 	case SIOCSNDFLUSH_IN6:	/* XXX: the ioctl name is confusing... */
1438 		/* flush default router list */
1439 		/*
1440 		 * xxx sumikawa: should not delete route if default
1441 		 * route equals to the top of default router list
1442 		 */
1443 		bzero(&any, sizeof(any));
1444 		defrouter_delreq(&any, 0);
1445 		defrouter_select();
1446 		/* xxx sumikawa: flush prefix list */
1447 		break;
1448 	case SIOCSPFXFLUSH_IN6:
1449 	    {
1450 		/* flush all the prefix advertised by routers */
1451 		struct nd_prefix *pr, *next;
1452 
1453 		s = splnet();
1454 
1455 		for (pr = nd_prefix.lh_first; pr; pr = next) {
1456 			next = pr->ndpr_next;
1457 			if (!IN6_IS_ADDR_UNSPECIFIED(&pr->ndpr_addr))
1458 				in6_ifdel(pr->ndpr_ifp, &pr->ndpr_addr);
1459 			prelist_remove(pr);
1460 		}
1461 		splx(s);
1462 		break;
1463 	    }
1464 	case SIOCSRTRFLUSH_IN6:
1465 	    {
1466 		/* flush all the default routers */
1467 		struct nd_defrouter *dr, *next;
1468 
1469 		s = splnet();
1470 		if ((dr = TAILQ_FIRST(&nd_defrouter)) != NULL) {
1471 			/*
1472 			 * The first entry of the list may be stored in
1473 			 * the routing table, so we'll delete it later.
1474 			 */
1475 			for (dr = TAILQ_NEXT(dr, dr_entry); dr; dr = next) {
1476 				next = TAILQ_NEXT(dr, dr_entry);
1477 				defrtrlist_del(dr);
1478 			}
1479 			defrtrlist_del(TAILQ_FIRST(&nd_defrouter));
1480 		}
1481 		splx(s);
1482 		break;
1483 	    }
1484 	case SIOCGNBRINFO_IN6:
1485 	    {
1486 		struct llinfo_nd6 *ln;
1487 		struct in6_addr nb_addr = nbi->addr; /* make local for safety */
1488 
1489 		/*
1490 		 * XXX: KAME specific hack for scoped addresses
1491 		 *      XXXX: for other scopes than link-local?
1492 		 */
1493 		if (IN6_IS_ADDR_LINKLOCAL(&nbi->addr) ||
1494 		    IN6_IS_ADDR_MC_LINKLOCAL(&nbi->addr)) {
1495 			u_int16_t *idp = (u_int16_t *)&nb_addr.s6_addr[2];
1496 
1497 			if (*idp == 0)
1498 				*idp = htons(ifp->if_index);
1499 		}
1500 
1501 		s = splnet();
1502 
1503 		if ((rt = nd6_lookup(&nb_addr, 0, ifp)) == NULL) {
1504 			error = EINVAL;
1505 			splx(s);
1506 			break;
1507 		}
1508 		ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1509 		nbi->state = ln->ln_state;
1510 		nbi->asked = ln->ln_asked;
1511 		nbi->isrouter = ln->ln_router;
1512 		nbi->expire = ln->ln_expire;
1513 		splx(s);
1514 
1515 		break;
1516 	    }
1517 	case SIOCGDEFIFACE_IN6:	/* XXX: should be implemented as a sysctl? */
1518 		ndif->ifindex = nd6_defifindex;
1519 		break;
1520 	case SIOCSDEFIFACE_IN6:	/* XXX: should be implemented as a sysctl? */
1521 		return(nd6_setdefaultiface(ndif->ifindex));
1522 		break;
1523 	}
1524 	return(error);
1525 }
1526 
1527 /*
1528  * Create neighbor cache entry and cache link-layer address,
1529  * on reception of inbound ND6 packets. (RS/RA/NS/redirect)
1530  */
1531 struct rtentry *
1532 nd6_cache_lladdr(ifp, from, lladdr, lladdrlen, type, code)
1533 	struct ifnet *ifp;
1534 	struct in6_addr *from;
1535 	char *lladdr;
1536 	int lladdrlen;
1537 	int type;	/* ICMP6 type */
1538 	int code;	/* type dependent information */
1539 {
1540 	struct rtentry *rt = NULL;
1541 	struct llinfo_nd6 *ln = NULL;
1542 	int is_newentry;
1543 	struct sockaddr_dl *sdl = NULL;
1544 	int do_update;
1545 	int olladdr;
1546 	int llchange;
1547 	int newstate = 0;
1548 	long time_second = time.tv_sec;
1549 
1550 	if (!ifp)
1551 		panic("ifp == NULL in nd6_cache_lladdr");
1552 	if (!from)
1553 		panic("from == NULL in nd6_cache_lladdr");
1554 
1555 	/* nothing must be updated for unspecified address */
1556 	if (IN6_IS_ADDR_UNSPECIFIED(from))
1557 		return NULL;
1558 
1559 	/*
1560 	 * Validation about ifp->if_addrlen and lladdrlen must be done in
1561 	 * the caller.
1562 	 *
1563 	 * XXX If the link does not have link-layer adderss, what should
1564 	 * we do? (ifp->if_addrlen == 0)
1565 	 * Spec says nothing in sections for RA, RS and NA.  There's small
1566 	 * description on it in NS section (RFC 2461 7.2.3).
1567 	 */
1568 
1569 	rt = nd6_lookup(from, 0, ifp);
1570 	if (!rt) {
1571 #if 0
1572 		/* nothing must be done if there's no lladdr */
1573 		if (!lladdr || !lladdrlen)
1574 			return NULL;
1575 #endif
1576 
1577 		rt = nd6_lookup(from, 1, ifp);
1578 		is_newentry = 1;
1579 	} else {
1580 		/* do nothing if static ndp is set */
1581 		if (rt->rt_flags & RTF_STATIC)
1582 			return NULL;
1583 		is_newentry = 0;
1584 	}
1585 
1586 	if (!rt)
1587 		return NULL;
1588 	if ((rt->rt_flags & (RTF_GATEWAY | RTF_LLINFO)) != RTF_LLINFO) {
1589 fail:
1590 		(void)nd6_free(rt);
1591 		return NULL;
1592 	}
1593 	ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1594 	if (!ln)
1595 		goto fail;
1596 	if (!rt->rt_gateway)
1597 		goto fail;
1598 	if (rt->rt_gateway->sa_family != AF_LINK)
1599 		goto fail;
1600 	sdl = SDL(rt->rt_gateway);
1601 
1602 	olladdr = (sdl->sdl_alen) ? 1 : 0;
1603 	if (olladdr && lladdr) {
1604 		if (bcmp(lladdr, LLADDR(sdl), ifp->if_addrlen))
1605 			llchange = 1;
1606 		else
1607 			llchange = 0;
1608 	} else
1609 		llchange = 0;
1610 
1611 	/*
1612 	 * newentry olladdr  lladdr  llchange	(*=record)
1613 	 *	0	n	n	--	(1)
1614 	 *	0	y	n	--	(2)
1615 	 *	0	n	y	--	(3) * STALE
1616 	 *	0	y	y	n	(4) *
1617 	 *	0	y	y	y	(5) * STALE
1618 	 *	1	--	n	--	(6)   NOSTATE(= PASSIVE)
1619 	 *	1	--	y	--	(7) * STALE
1620 	 */
1621 
1622 	if (lladdr) {		/*(3-5) and (7)*/
1623 		/*
1624 		 * Record source link-layer address
1625 		 * XXX is it dependent to ifp->if_type?
1626 		 */
1627 		sdl->sdl_alen = ifp->if_addrlen;
1628 		bcopy(lladdr, LLADDR(sdl), ifp->if_addrlen);
1629 	}
1630 
1631 	if (!is_newentry) {
1632 		if ((!olladdr && lladdr)		/*(3)*/
1633 		 || (olladdr && lladdr && llchange)) {	/*(5)*/
1634 			do_update = 1;
1635 			newstate = ND6_LLINFO_STALE;
1636 		} else					/*(1-2,4)*/
1637 			do_update = 0;
1638 	} else {
1639 		do_update = 1;
1640 		if (!lladdr)				/*(6)*/
1641 			newstate = ND6_LLINFO_NOSTATE;
1642 		else					/*(7)*/
1643 			newstate = ND6_LLINFO_STALE;
1644 	}
1645 
1646 	if (do_update) {
1647 		/*
1648 		 * Update the state of the neighbor cache.
1649 		 */
1650 		ln->ln_state = newstate;
1651 
1652 		if (ln->ln_state == ND6_LLINFO_STALE) {
1653 			/*
1654 			 * XXX: since nd6_output() below will cause
1655 			 * state tansition to DELAY and reset the timer,
1656 			 * we must set the timer now, although it is actually
1657 			 * meaningless.
1658 			 */
1659 			ln->ln_expire = time_second + nd6_gctimer;
1660 
1661 			if (ln->ln_hold) {
1662 #ifdef OLDIP6OUTPUT
1663 				ln->ln_asked = 0;
1664 				ln->ln_state = ND6_LLINFO_DELAY;
1665 				ln->ln_expire = time_second + nd6_delay;
1666 				(*ifp->if_output)(ifp, ln->ln_hold,
1667 						  rt_key(rt), rt);
1668 #else
1669 				/*
1670 				 * we assume ifp is not a p2p here, so just
1671 				 * set the 2nd argument as the 1st one.
1672 				 */
1673 				nd6_output(ifp, ifp, ln->ln_hold,
1674 					   (struct sockaddr_in6 *)rt_key(rt),
1675 					   rt);
1676 #endif
1677 				ln->ln_hold = NULL;
1678 			}
1679 		} else if (ln->ln_state == ND6_LLINFO_INCOMPLETE) {
1680 			/* probe right away */
1681 			ln->ln_expire = time_second;
1682 		}
1683 	}
1684 
1685 	/*
1686 	 * ICMP6 type dependent behavior.
1687 	 *
1688 	 * NS: clear IsRouter if new entry
1689 	 * RS: clear IsRouter
1690 	 * RA: set IsRouter if there's lladdr
1691 	 * redir: clear IsRouter if new entry
1692 	 *
1693 	 * RA case, (1):
1694 	 * The spec says that we must set IsRouter in the following cases:
1695 	 * - If lladdr exist, set IsRouter.  This means (1-5).
1696 	 * - If it is old entry (!newentry), set IsRouter.  This means (7).
1697 	 * So, based on the spec, in (1-5) and (7) cases we must set IsRouter.
1698 	 * A quetion arises for (1) case.  (1) case has no lladdr in the
1699 	 * neighbor cache, this is similar to (6).
1700 	 * This case is rare but we figured that we MUST NOT set IsRouter.
1701 	 *
1702 	 * newentry olladdr  lladdr  llchange	    NS  RS  RA	redir
1703 	 *							D R
1704 	 *	0	n	n	--	(1)	c   ?     s
1705 	 *	0	y	n	--	(2)	c   s     s
1706 	 *	0	n	y	--	(3)	c   s     s
1707 	 *	0	y	y	n	(4)	c   s     s
1708 	 *	0	y	y	y	(5)	c   s     s
1709 	 *	1	--	n	--	(6) c	c 	c s
1710 	 *	1	--	y	--	(7) c	c   s	c s
1711 	 *
1712 	 *					(c=clear s=set)
1713 	 */
1714 	switch (type & 0xff) {
1715 	case ND_NEIGHBOR_SOLICIT:
1716 		/*
1717 		 * New entry must have is_router flag cleared.
1718 		 */
1719 		if (is_newentry)	/*(6-7)*/
1720 			ln->ln_router = 0;
1721 		break;
1722 	case ND_REDIRECT:
1723 		/*
1724 		 * If the icmp is a redirect to a better router, always set the
1725 		 * is_router flag. Otherwise, if the entry is newly created,
1726 		 * clear the flag. [RFC 2461, sec 8.3]
1727 		 */
1728 		if (code == ND_REDIRECT_ROUTER)
1729 			ln->ln_router = 1;
1730 		else if (is_newentry) /*(6-7)*/
1731 			ln->ln_router = 0;
1732 		break;
1733 	case ND_ROUTER_SOLICIT:
1734 		/*
1735 		 * is_router flag must always be cleared.
1736 		 */
1737 		ln->ln_router = 0;
1738 		break;
1739 	case ND_ROUTER_ADVERT:
1740 		/*
1741 		 * Mark an entry with lladdr as a router.
1742 		 */
1743 		if ((!is_newentry && (olladdr || lladdr))	/*(2-5)*/
1744 		 || (is_newentry && lladdr)) {			/*(7)*/
1745 			ln->ln_router = 1;
1746 		}
1747 		break;
1748 	}
1749 
1750 	/*
1751 	 * When the link-layer address of a router changes, select the
1752 	 * best router again.  In particular, when the neighbor entry is newly
1753 	 * created, it might affect the selection policy.
1754 	 * Question: can we restrict the first condition to the "is_newentry"
1755 	 * case?
1756 	 * XXX: when we hear an RA from a new router with the link-layer
1757 	 * address option, defrouter_select() is called twice, since
1758 	 * defrtrlist_update called the function as well.  However, I believe
1759 	 * we can compromise the overhead, since it only happens the first
1760 	 * time.
1761 	 */
1762 	if (do_update && ln->ln_router && !ip6_forwarding && ip6_accept_rtadv)
1763 		defrouter_select();
1764 
1765 	return rt;
1766 }
1767 
1768 static void
1769 nd6_slowtimo(ignored_arg)
1770     void *ignored_arg;
1771 {
1772 	int s = splnet();
1773 	int i;
1774 	struct nd_ifinfo *nd6if;
1775 
1776 	timeout_set(&nd6_slowtimo_ch, nd6_slowtimo, NULL);
1777 	timeout_add(&nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz);
1778 	for (i = 1; i < if_index + 1; i++) {
1779 		if (!nd_ifinfo || i >= nd_ifinfo_indexlim)
1780 			continue;
1781 		nd6if = &nd_ifinfo[i];
1782 		if (nd6if->basereachable && /* already initialized */
1783 		    (nd6if->recalctm -= ND6_SLOWTIMER_INTERVAL) <= 0) {
1784 			/*
1785 			 * Since reachable time rarely changes by router
1786 			 * advertisements, we SHOULD insure that a new random
1787 			 * value gets recomputed at least once every few hours.
1788 			 * (RFC 2461, 6.3.4)
1789 			 */
1790 			nd6if->recalctm = nd6_recalc_reachtm_interval;
1791 			nd6if->reachable = ND_COMPUTE_RTIME(nd6if->basereachable);
1792 		}
1793 	}
1794 	splx(s);
1795 }
1796 
1797 #define senderr(e) { error = (e); goto bad;}
1798 int
1799 nd6_output(ifp, origifp, m0, dst, rt0)
1800 	struct ifnet *ifp;
1801 	struct ifnet *origifp;
1802 	struct mbuf *m0;
1803 	struct sockaddr_in6 *dst;
1804 	struct rtentry *rt0;
1805 {
1806 	struct mbuf *m = m0;
1807 	struct rtentry *rt = rt0;
1808 	struct sockaddr_in6 *gw6 = NULL;
1809 	struct llinfo_nd6 *ln = NULL;
1810 	int error = 0;
1811 	long time_second = time.tv_sec;
1812 #ifdef IPSEC
1813 	struct m_tag *mtag;
1814 #endif /* IPSEC */
1815 
1816 	if (IN6_IS_ADDR_MULTICAST(&dst->sin6_addr))
1817 		goto sendpkt;
1818 
1819 	/*
1820 	 * XXX: we currently do not make neighbor cache on any interface
1821 	 * other than ARCnet, Ethernet, FDDI and GIF.
1822 	 *
1823 	 * RFC2893 says:
1824 	 * - unidirectional tunnels needs no ND
1825 	 */
1826 	switch (ifp->if_type) {
1827 	case IFT_ARCNET:
1828 	case IFT_ETHER:
1829 	case IFT_FDDI:
1830 	case IFT_GIF:		/* XXX need more cases? */
1831 		break;
1832 	default:
1833 		goto sendpkt;
1834 	}
1835 
1836 	/*
1837 	 * next hop determination. This routine is derived from ether_outpout.
1838 	 */
1839 	if (rt) {
1840 		if ((rt->rt_flags & RTF_UP) == 0) {
1841 			if ((rt0 = rt = rtalloc1((struct sockaddr *)dst, 1)) !=
1842 				NULL)
1843 			{
1844 				rt->rt_refcnt--;
1845 				if (rt->rt_ifp != ifp) {
1846 					/* XXX: loop care? */
1847 					return nd6_output(ifp, origifp, m0,
1848 							  dst, rt);
1849 				}
1850 			} else
1851 				senderr(EHOSTUNREACH);
1852 		}
1853 
1854 		if (rt->rt_flags & RTF_GATEWAY) {
1855 			gw6 = (struct sockaddr_in6 *)rt->rt_gateway;
1856 
1857 			/*
1858 			 * We skip link-layer address resolution and NUD
1859 			 * if the gateway is not a neighbor from ND point
1860 			 * of view, regardless the value of the
1861 			 * nd_ifinfo.flags.
1862 			 * The second condition is a bit tricky: we skip
1863 			 * if the gateway is our own address, which is
1864 			 * sometimes used to install a route to a p2p link.
1865 			 */
1866 			if (!nd6_is_addr_neighbor(gw6, ifp) ||
1867 			    in6ifa_ifpwithaddr(ifp, &gw6->sin6_addr)) {
1868 				/*
1869 				 * We allow this kind of tricky route only
1870 				 * when the outgoing interface is p2p.
1871 				 * XXX: we may need a more generic rule here.
1872 				 */
1873 				if ((ifp->if_flags & IFF_POINTOPOINT) == 0)
1874 					senderr(EHOSTUNREACH);
1875 
1876 				goto sendpkt;
1877 			}
1878 
1879 			if (rt->rt_gwroute == 0)
1880 				goto lookup;
1881 			if (((rt = rt->rt_gwroute)->rt_flags & RTF_UP) == 0) {
1882 				rtfree(rt); rt = rt0;
1883 			lookup: rt->rt_gwroute = rtalloc1(rt->rt_gateway, 1);
1884 				if ((rt = rt->rt_gwroute) == 0)
1885 					senderr(EHOSTUNREACH);
1886 			}
1887 		}
1888 	}
1889 
1890 	/*
1891 	 * Address resolution or Neighbor Unreachability Detection
1892 	 * for the next hop.
1893 	 * At this point, the destination of the packet must be a unicast
1894 	 * or an anycast address(i.e. not a multicast).
1895 	 */
1896 
1897 	/* Look up the neighbor cache for the nexthop */
1898 	if (rt && (rt->rt_flags & RTF_LLINFO) != 0)
1899 		ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1900 	else {
1901 		/*
1902 		 * Since nd6_is_addr_neighbor() internally calls nd6_lookup(),
1903 		 * the condition below is not very efficient. But we believe
1904 		 * it is tolerable, because this should be a rare case.
1905 		 */
1906 		if (nd6_is_addr_neighbor(dst, ifp) &&
1907 		    (rt = nd6_lookup(&dst->sin6_addr, 1, ifp)) != NULL)
1908 			ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1909 	}
1910 	if (!ln || !rt) {
1911 		if ((ifp->if_flags & IFF_POINTOPOINT) == 0 &&
1912 		    !(nd_ifinfo[ifp->if_index].flags & ND6_IFF_PERFORMNUD)) {
1913 			log(LOG_DEBUG,
1914 			    "nd6_output: can't allocate llinfo for %s "
1915 			    "(ln=%p, rt=%p)\n",
1916 			    ip6_sprintf(&dst->sin6_addr), ln, rt);
1917 			senderr(EIO);	/* XXX: good error? */
1918 		}
1919 
1920 		goto sendpkt;	/* send anyway */
1921 	}
1922 
1923 	/* We don't have to do link-layer address resolution on a p2p link. */
1924 	if ((ifp->if_flags & IFF_POINTOPOINT) != 0 &&
1925 	    ln->ln_state < ND6_LLINFO_REACHABLE) {
1926 		ln->ln_state = ND6_LLINFO_STALE;
1927 		ln->ln_expire = time_second + nd6_gctimer;
1928 	}
1929 
1930 	/*
1931 	 * The first time we send a packet to a neighbor whose entry is
1932 	 * STALE, we have to change the state to DELAY and a sets a timer to
1933 	 * expire in DELAY_FIRST_PROBE_TIME seconds to ensure do
1934 	 * neighbor unreachability detection on expiration.
1935 	 * (RFC 2461 7.3.3)
1936 	 */
1937 	if (ln->ln_state == ND6_LLINFO_STALE) {
1938 		ln->ln_asked = 0;
1939 		ln->ln_state = ND6_LLINFO_DELAY;
1940 		ln->ln_expire = time_second + nd6_delay;
1941 	}
1942 
1943 	/*
1944 	 * If the neighbor cache entry has a state other than INCOMPLETE
1945 	 * (i.e. its link-layer address is already reloved), just
1946 	 * send the packet.
1947 	 */
1948 	if (ln->ln_state > ND6_LLINFO_INCOMPLETE)
1949 		goto sendpkt;
1950 
1951 	/*
1952 	 * There is a neighbor cache entry, but no ethernet address
1953 	 * response yet. Replace the held mbuf (if any) with this
1954 	 * latest one.
1955 	 *
1956 	 * XXX Does the code conform to rate-limiting rule?
1957 	 * (RFC 2461 7.2.2)
1958 	 */
1959 	if (ln->ln_state == ND6_LLINFO_NOSTATE)
1960 		ln->ln_state = ND6_LLINFO_INCOMPLETE;
1961 	if (ln->ln_hold)
1962 		m_freem(ln->ln_hold);
1963 	ln->ln_hold = m;
1964 	if (ln->ln_expire) {
1965 		if (ln->ln_asked < nd6_mmaxtries &&
1966 		    ln->ln_expire < time_second) {
1967 			ln->ln_asked++;
1968 			ln->ln_expire = time_second +
1969 				nd_ifinfo[ifp->if_index].retrans / 1000;
1970 			nd6_ns_output(ifp, NULL, &dst->sin6_addr, ln, 0);
1971 		}
1972 	}
1973 	return(0);
1974 
1975   sendpkt:
1976 #ifdef IPSEC
1977 	/*
1978 	 * If the packet needs outgoing IPsec crypto processing and the
1979 	 * interface doesn't support it, drop it.
1980 	 */
1981 	mtag = m_tag_find(m, PACKET_TAG_IPSEC_OUT_CRYPTO_NEEDED, NULL);
1982 #endif /* IPSEC */
1983 
1984 	if ((ifp->if_flags & IFF_LOOPBACK) != 0) {
1985 #ifdef IPSEC
1986 		if (mtag != NULL &&
1987 		    (origifp->if_capabilities & IFCAP_IPSEC) == 0) {
1988 			/* Tell IPsec to do its own crypto. */
1989 			ipsp_skipcrypto_unmark((struct tdb_ident *)(mtag + 1));
1990 			error = EACCES;
1991 			goto bad;
1992 		}
1993 #endif /* IPSEC */
1994 		return((*ifp->if_output)(origifp, m, (struct sockaddr *)dst,
1995 					 rt));
1996 	}
1997 #ifdef IPSEC
1998 	if (mtag != NULL &&
1999 	    (ifp->if_capabilities & IFCAP_IPSEC) == 0) {
2000 		/* Tell IPsec to do its own crypto. */
2001 		ipsp_skipcrypto_unmark((struct tdb_ident *)(mtag + 1));
2002 		error = EACCES;
2003 		goto bad;
2004 	}
2005 #endif /* IPSEC */
2006 	return((*ifp->if_output)(ifp, m, (struct sockaddr *)dst, rt));
2007 
2008   bad:
2009 	if (m)
2010 		m_freem(m);
2011 	return (error);
2012 }
2013 #undef senderr
2014 
2015 int
2016 nd6_storelladdr(ifp, rt, m, dst, desten)
2017 	struct ifnet *ifp;
2018 	struct rtentry *rt;
2019 	struct mbuf *m;
2020 	struct sockaddr *dst;
2021 	u_char *desten;
2022 {
2023 	struct sockaddr_dl *sdl;
2024 
2025 	if (m->m_flags & M_MCAST) {
2026 		switch (ifp->if_type) {
2027 		case IFT_ETHER:
2028 		case IFT_FDDI:
2029 			ETHER_MAP_IPV6_MULTICAST(&SIN6(dst)->sin6_addr,
2030 						 desten);
2031 			return(1);
2032 			break;
2033 		case IFT_ARCNET:
2034 			*desten = 0;
2035 			return(1);
2036 		default:
2037 			m_freem(m);
2038 			return(0);
2039 		}
2040 	}
2041 
2042 	if (rt == NULL) {
2043 		/* this could happen, if we could not allocate memory */
2044 		m_freem(m);
2045 		return(0);
2046 	}
2047 	if (rt->rt_gateway->sa_family != AF_LINK) {
2048 		printf("nd6_storelladdr: something odd happens\n");
2049 		m_freem(m);
2050 		return(0);
2051 	}
2052 	sdl = SDL(rt->rt_gateway);
2053 	if (sdl->sdl_alen == 0) {
2054 		/* this should be impossible, but we bark here for debugging */
2055 		printf("nd6_storelladdr: sdl_alen == 0\n");
2056 		m_freem(m);
2057 		return(0);
2058 	}
2059 
2060 	bcopy(LLADDR(sdl), desten, sdl->sdl_alen);
2061 	return(1);
2062 }
2063