xref: /freebsd/sys/netinet6/nd6.c (revision c11e094d96120a2e0e726ed9705ae0ec08db49b6)
1 /*	$FreeBSD$	*/
2 /*	$KAME: nd6.c,v 1.144 2001/05/24 07:44:00 itojun 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 "opt_inet.h"
42 #include "opt_inet6.h"
43 
44 #include <sys/param.h>
45 #include <sys/systm.h>
46 #include <sys/callout.h>
47 #include <sys/malloc.h>
48 #include <sys/mbuf.h>
49 #include <sys/socket.h>
50 #include <sys/sockio.h>
51 #include <sys/time.h>
52 #include <sys/kernel.h>
53 #include <sys/protosw.h>
54 #include <sys/errno.h>
55 #include <sys/syslog.h>
56 #include <sys/queue.h>
57 #include <sys/sysctl.h>
58 
59 #include <net/if.h>
60 #include <net/if_dl.h>
61 #include <net/if_types.h>
62 #include <net/if_atm.h>
63 #include <net/fddi.h>
64 #include <net/route.h>
65 
66 #include <netinet/in.h>
67 #include <netinet/if_ether.h>
68 #include <netinet6/in6_var.h>
69 #include <netinet/ip6.h>
70 #include <netinet6/ip6_var.h>
71 #include <netinet6/nd6.h>
72 #include <netinet6/in6_prefix.h>
73 #include <netinet/icmp6.h>
74 
75 #include <net/net_osdep.h>
76 
77 #define ND6_SLOWTIMER_INTERVAL (60 * 60) /* 1 hour */
78 #define ND6_RECALC_REACHTM_INTERVAL (60 * 120) /* 2 hours */
79 
80 #define SIN6(s) ((struct sockaddr_in6 *)s)
81 #define SDL(s) ((struct sockaddr_dl *)s)
82 
83 /* timer values */
84 int	nd6_prune	= 1;	/* walk list every 1 seconds */
85 int	nd6_delay	= 5;	/* delay first probe time 5 second */
86 int	nd6_umaxtries	= 3;	/* maximum unicast query */
87 int	nd6_mmaxtries	= 3;	/* maximum multicast query */
88 int	nd6_useloopback = 1;	/* use loopback interface for local traffic */
89 int	nd6_gctimer	= (60 * 60 * 24); /* 1 day: garbage collection timer */
90 
91 /* preventing too many loops in ND option parsing */
92 int nd6_maxndopt = 10;	/* max # of ND options allowed */
93 
94 int nd6_maxnudhint = 0;	/* max # of subsequent upper layer hints */
95 
96 #ifdef ND6_DEBUG
97 int nd6_debug = 1;
98 #else
99 int nd6_debug = 0;
100 #endif
101 
102 /* for debugging? */
103 static int nd6_inuse, nd6_allocated;
104 
105 struct llinfo_nd6 llinfo_nd6 = {&llinfo_nd6, &llinfo_nd6};
106 static size_t nd_ifinfo_indexlim = 8;
107 struct nd_ifinfo *nd_ifinfo = NULL;
108 struct nd_drhead nd_defrouter;
109 struct nd_prhead nd_prefix = { 0 };
110 
111 int nd6_recalc_reachtm_interval = ND6_RECALC_REACHTM_INTERVAL;
112 static struct sockaddr_in6 all1_sa;
113 
114 static void nd6_slowtimo __P((void *));
115 static int regen_tmpaddr __P((struct in6_ifaddr *));
116 
117 struct callout nd6_slowtimo_ch;
118 struct callout nd6_timer_ch;
119 extern struct callout in6_tmpaddrtimer_ch;
120 
121 void
122 nd6_init()
123 {
124 	static int nd6_init_done = 0;
125 	int i;
126 
127 	if (nd6_init_done) {
128 		log(LOG_NOTICE, "nd6_init called more than once(ignored)\n");
129 		return;
130 	}
131 
132 	all1_sa.sin6_family = AF_INET6;
133 	all1_sa.sin6_len = sizeof(struct sockaddr_in6);
134 	for (i = 0; i < sizeof(all1_sa.sin6_addr); i++)
135 		all1_sa.sin6_addr.s6_addr[i] = 0xff;
136 
137 	/* initialization of the default router list */
138 	TAILQ_INIT(&nd_defrouter);
139 
140 	nd6_init_done = 1;
141 
142 	/* start timer */
143 	callout_reset(&nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
144 	    nd6_slowtimo, NULL);
145 }
146 
147 void
148 nd6_ifattach(ifp)
149 	struct ifnet *ifp;
150 {
151 
152 	/*
153 	 * We have some arrays that should be indexed by if_index.
154 	 * since if_index will grow dynamically, they should grow too.
155 	 */
156 	if (nd_ifinfo == NULL || if_index >= nd_ifinfo_indexlim) {
157 		size_t n;
158 		caddr_t q;
159 
160 		while (if_index >= nd_ifinfo_indexlim)
161 			nd_ifinfo_indexlim <<= 1;
162 
163 		/* grow nd_ifinfo */
164 		n = nd_ifinfo_indexlim * sizeof(struct nd_ifinfo);
165 		q = (caddr_t)malloc(n, M_IP6NDP, M_WAITOK);
166 		bzero(q, n);
167 		if (nd_ifinfo) {
168 			bcopy((caddr_t)nd_ifinfo, q, n/2);
169 			free((caddr_t)nd_ifinfo, M_IP6NDP);
170 		}
171 		nd_ifinfo = (struct nd_ifinfo *)q;
172 	}
173 
174 #define ND nd_ifinfo[ifp->if_index]
175 
176 	/*
177 	 * Don't initialize if called twice.
178 	 * XXX: to detect this, we should choose a member that is never set
179 	 * before initialization of the ND structure itself.  We formaly used
180 	 * the linkmtu member, which was not suitable because it could be
181 	 * initialized via "ifconfig mtu".
182 	 */
183 	if (ND.basereachable)
184 		return;
185 
186 	ND.linkmtu = ifnet_byindex(ifp->if_index)->if_mtu;
187 	ND.chlim = IPV6_DEFHLIM;
188 	ND.basereachable = REACHABLE_TIME;
189 	ND.reachable = ND_COMPUTE_RTIME(ND.basereachable);
190 	ND.retrans = RETRANS_TIMER;
191 	ND.receivedra = 0;
192 	ND.flags = ND6_IFF_PERFORMNUD;
193 	nd6_setmtu(ifp);
194 #undef ND
195 }
196 
197 /*
198  * Reset ND level link MTU. This function is called when the physical MTU
199  * changes, which means we might have to adjust the ND level MTU.
200  */
201 void
202 nd6_setmtu(ifp)
203 	struct ifnet *ifp;
204 {
205 #define MIN(a,b) ((a) < (b) ? (a) : (b))
206 	struct nd_ifinfo *ndi = &nd_ifinfo[ifp->if_index];
207 	u_long oldmaxmtu = ndi->maxmtu;
208 	u_long oldlinkmtu = ndi->linkmtu;
209 
210 	switch (ifp->if_type) {
211 	case IFT_ARCNET:	/* XXX MTU handling needs more work */
212 		ndi->maxmtu = MIN(60480, ifp->if_mtu);
213 		break;
214 	case IFT_ETHER:
215 		ndi->maxmtu = MIN(ETHERMTU, ifp->if_mtu);
216 		break;
217 	case IFT_FDDI:
218 		ndi->maxmtu = MIN(FDDIIPMTU, ifp->if_mtu);
219 		break;
220 	case IFT_ATM:
221 		ndi->maxmtu = MIN(ATMMTU, ifp->if_mtu);
222 		break;
223 	case IFT_IEEE1394:	/* XXX should be IEEE1394MTU(1500) */
224 		ndi->maxmtu = MIN(ETHERMTU, ifp->if_mtu);
225 		break;
226 #ifdef IFT_IEEE80211
227 	case IFT_IEEE80211:	/* XXX should be IEEE80211MTU(1500) */
228 		ndi->maxmtu = MIN(ETHERMTU, ifp->if_mtu);
229 		break;
230 #endif
231 	default:
232 		ndi->maxmtu = ifp->if_mtu;
233 		break;
234 	}
235 
236 	if (oldmaxmtu != ndi->maxmtu) {
237 		/*
238 		 * If the ND level MTU is not set yet, or if the maxmtu
239 		 * is reset to a smaller value than the ND level MTU,
240 		 * also reset the ND level MTU.
241 		 */
242 		if (ndi->linkmtu == 0 ||
243 		    ndi->maxmtu < ndi->linkmtu) {
244 			ndi->linkmtu = ndi->maxmtu;
245 			/* also adjust in6_maxmtu if necessary. */
246 			if (oldlinkmtu == 0) {
247 				/*
248 				 * XXX: the case analysis is grotty, but
249 				 * it is not efficient to call in6_setmaxmtu()
250 				 * here when we are during the initialization
251 				 * procedure.
252 				 */
253 				if (in6_maxmtu < ndi->linkmtu)
254 					in6_maxmtu = ndi->linkmtu;
255 			} else
256 				in6_setmaxmtu();
257 		}
258 	}
259 #undef MIN
260 }
261 
262 void
263 nd6_option_init(opt, icmp6len, ndopts)
264 	void *opt;
265 	int icmp6len;
266 	union nd_opts *ndopts;
267 {
268 	bzero(ndopts, sizeof(*ndopts));
269 	ndopts->nd_opts_search = (struct nd_opt_hdr *)opt;
270 	ndopts->nd_opts_last
271 		= (struct nd_opt_hdr *)(((u_char *)opt) + icmp6len);
272 
273 	if (icmp6len == 0) {
274 		ndopts->nd_opts_done = 1;
275 		ndopts->nd_opts_search = NULL;
276 	}
277 }
278 
279 /*
280  * Take one ND option.
281  */
282 struct nd_opt_hdr *
283 nd6_option(ndopts)
284 	union nd_opts *ndopts;
285 {
286 	struct nd_opt_hdr *nd_opt;
287 	int olen;
288 
289 	if (!ndopts)
290 		panic("ndopts == NULL in nd6_option\n");
291 	if (!ndopts->nd_opts_last)
292 		panic("uninitialized ndopts in nd6_option\n");
293 	if (!ndopts->nd_opts_search)
294 		return NULL;
295 	if (ndopts->nd_opts_done)
296 		return NULL;
297 
298 	nd_opt = ndopts->nd_opts_search;
299 
300 	/* make sure nd_opt_len is inside the buffer */
301 	if ((caddr_t)&nd_opt->nd_opt_len >= (caddr_t)ndopts->nd_opts_last) {
302 		bzero(ndopts, sizeof(*ndopts));
303 		return NULL;
304 	}
305 
306 	olen = nd_opt->nd_opt_len << 3;
307 	if (olen == 0) {
308 		/*
309 		 * Message validation requires that all included
310 		 * options have a length that is greater than zero.
311 		 */
312 		bzero(ndopts, sizeof(*ndopts));
313 		return NULL;
314 	}
315 
316 	ndopts->nd_opts_search = (struct nd_opt_hdr *)((caddr_t)nd_opt + olen);
317 	if (ndopts->nd_opts_search > ndopts->nd_opts_last) {
318 		/* option overruns the end of buffer, invalid */
319 		bzero(ndopts, sizeof(*ndopts));
320 		return NULL;
321 	} else if (ndopts->nd_opts_search == ndopts->nd_opts_last) {
322 		/* reached the end of options chain */
323 		ndopts->nd_opts_done = 1;
324 		ndopts->nd_opts_search = NULL;
325 	}
326 	return nd_opt;
327 }
328 
329 /*
330  * Parse multiple ND options.
331  * This function is much easier to use, for ND routines that do not need
332  * multiple options of the same type.
333  */
334 int
335 nd6_options(ndopts)
336 	union nd_opts *ndopts;
337 {
338 	struct nd_opt_hdr *nd_opt;
339 	int i = 0;
340 
341 	if (!ndopts)
342 		panic("ndopts == NULL in nd6_options\n");
343 	if (!ndopts->nd_opts_last)
344 		panic("uninitialized ndopts in nd6_options\n");
345 	if (!ndopts->nd_opts_search)
346 		return 0;
347 
348 	while (1) {
349 		nd_opt = nd6_option(ndopts);
350 		if (!nd_opt && !ndopts->nd_opts_last) {
351 			/*
352 			 * Message validation requires that all included
353 			 * options have a length that is greater than zero.
354 			 */
355 			icmp6stat.icp6s_nd_badopt++;
356 			bzero(ndopts, sizeof(*ndopts));
357 			return -1;
358 		}
359 
360 		if (!nd_opt)
361 			goto skip1;
362 
363 		switch (nd_opt->nd_opt_type) {
364 		case ND_OPT_SOURCE_LINKADDR:
365 		case ND_OPT_TARGET_LINKADDR:
366 		case ND_OPT_MTU:
367 		case ND_OPT_REDIRECTED_HEADER:
368 			if (ndopts->nd_opt_array[nd_opt->nd_opt_type]) {
369 				nd6log((LOG_INFO,
370 				    "duplicated ND6 option found (type=%d)\n",
371 				    nd_opt->nd_opt_type));
372 				/* XXX bark? */
373 			} else {
374 				ndopts->nd_opt_array[nd_opt->nd_opt_type]
375 					= nd_opt;
376 			}
377 			break;
378 		case ND_OPT_PREFIX_INFORMATION:
379 			if (ndopts->nd_opt_array[nd_opt->nd_opt_type] == 0) {
380 				ndopts->nd_opt_array[nd_opt->nd_opt_type]
381 					= nd_opt;
382 			}
383 			ndopts->nd_opts_pi_end =
384 				(struct nd_opt_prefix_info *)nd_opt;
385 			break;
386 		default:
387 			/*
388 			 * Unknown options must be silently ignored,
389 			 * to accomodate future extension to the protocol.
390 			 */
391 			nd6log((LOG_DEBUG,
392 			    "nd6_options: unsupported option %d - "
393 			    "option ignored\n", nd_opt->nd_opt_type));
394 		}
395 
396 skip1:
397 		i++;
398 		if (i > nd6_maxndopt) {
399 			icmp6stat.icp6s_nd_toomanyopt++;
400 			nd6log((LOG_INFO, "too many loop in nd opt\n"));
401 			break;
402 		}
403 
404 		if (ndopts->nd_opts_done)
405 			break;
406 	}
407 
408 	return 0;
409 }
410 
411 /*
412  * ND6 timer routine to expire default route list and prefix list
413  */
414 void
415 nd6_timer(ignored_arg)
416 	void	*ignored_arg;
417 {
418 	int s;
419 	struct llinfo_nd6 *ln;
420 	struct nd_defrouter *dr;
421 	struct nd_prefix *pr;
422 	struct ifnet *ifp;
423 	struct in6_ifaddr *ia6, *nia6;
424 	struct in6_addrlifetime *lt6;
425 
426 	s = splnet();
427 	callout_reset(&nd6_timer_ch, nd6_prune * hz,
428 		      nd6_timer, NULL);
429 
430 	ln = llinfo_nd6.ln_next;
431 	while (ln && ln != &llinfo_nd6) {
432 		struct rtentry *rt;
433 		struct sockaddr_in6 *dst;
434 		struct llinfo_nd6 *next = ln->ln_next;
435 		/* XXX: used for the DELAY case only: */
436 		struct nd_ifinfo *ndi = NULL;
437 
438 		if ((rt = ln->ln_rt) == NULL) {
439 			ln = next;
440 			continue;
441 		}
442 		if ((ifp = rt->rt_ifp) == NULL) {
443 			ln = next;
444 			continue;
445 		}
446 		ndi = &nd_ifinfo[ifp->if_index];
447 		dst = (struct sockaddr_in6 *)rt_key(rt);
448 
449 		if (ln->ln_expire > time_second) {
450 			ln = next;
451 			continue;
452 		}
453 
454 		/* sanity check */
455 		if (!rt)
456 			panic("rt=0 in nd6_timer(ln=%p)\n", ln);
457 		if (rt->rt_llinfo && (struct llinfo_nd6 *)rt->rt_llinfo != ln)
458 			panic("rt_llinfo(%p) is not equal to ln(%p)\n",
459 			      rt->rt_llinfo, ln);
460 		if (!dst)
461 			panic("dst=0 in nd6_timer(ln=%p)\n", ln);
462 
463 		switch (ln->ln_state) {
464 		case ND6_LLINFO_INCOMPLETE:
465 			if (ln->ln_asked < nd6_mmaxtries) {
466 				ln->ln_asked++;
467 				ln->ln_expire = time_second +
468 					nd_ifinfo[ifp->if_index].retrans / 1000;
469 				nd6_ns_output(ifp, NULL, &dst->sin6_addr,
470 					ln, 0);
471 			} else {
472 				struct mbuf *m = ln->ln_hold;
473 				if (m) {
474 					if (rt->rt_ifp) {
475 						/*
476 						 * Fake rcvif to make ICMP error
477 						 * more helpful in diagnosing
478 						 * for the receiver.
479 						 * XXX: should we consider
480 						 * older rcvif?
481 						 */
482 						m->m_pkthdr.rcvif = rt->rt_ifp;
483 					}
484 					icmp6_error(m, ICMP6_DST_UNREACH,
485 						    ICMP6_DST_UNREACH_ADDR, 0);
486 					ln->ln_hold = NULL;
487 				}
488 				next = nd6_free(rt);
489 			}
490 			break;
491 		case ND6_LLINFO_REACHABLE:
492 			if (ln->ln_expire) {
493 				ln->ln_state = ND6_LLINFO_STALE;
494 				ln->ln_expire = time_second + nd6_gctimer;
495 			}
496 			break;
497 
498 		case ND6_LLINFO_STALE:
499 			/* Garbage Collection(RFC 2461 5.3) */
500 			if (ln->ln_expire)
501 				next = nd6_free(rt);
502 			break;
503 
504 		case ND6_LLINFO_DELAY:
505 			if (ndi && (ndi->flags & ND6_IFF_PERFORMNUD) != 0) {
506 				/* We need NUD */
507 				ln->ln_asked = 1;
508 				ln->ln_state = ND6_LLINFO_PROBE;
509 				ln->ln_expire = time_second +
510 					ndi->retrans / 1000;
511 				nd6_ns_output(ifp, &dst->sin6_addr,
512 					      &dst->sin6_addr,
513 					      ln, 0);
514 			} else {
515 				ln->ln_state = ND6_LLINFO_STALE; /* XXX */
516 				ln->ln_expire = time_second + nd6_gctimer;
517 			}
518 			break;
519 		case ND6_LLINFO_PROBE:
520 			if (ln->ln_asked < nd6_umaxtries) {
521 				ln->ln_asked++;
522 				ln->ln_expire = time_second +
523 					nd_ifinfo[ifp->if_index].retrans / 1000;
524 				nd6_ns_output(ifp, &dst->sin6_addr,
525 					       &dst->sin6_addr, ln, 0);
526 			} else {
527 				next = nd6_free(rt);
528 			}
529 			break;
530 		}
531 		ln = next;
532 	}
533 
534 	/* expire default router list */
535 	dr = TAILQ_FIRST(&nd_defrouter);
536 	while (dr) {
537 		if (dr->expire && dr->expire < time_second) {
538 			struct nd_defrouter *t;
539 			t = TAILQ_NEXT(dr, dr_entry);
540 			defrtrlist_del(dr);
541 			dr = t;
542 		} else {
543 			dr = TAILQ_NEXT(dr, dr_entry);
544 		}
545 	}
546 
547 	/*
548 	 * expire interface addresses.
549 	 * in the past the loop was inside prefix expiry processing.
550 	 * However, from a stricter speci-confrmance standpoint, we should
551 	 * rather separate address lifetimes and prefix lifetimes.
552 	 */
553   addrloop:
554 	for (ia6 = in6_ifaddr; ia6; ia6 = nia6) {
555 		nia6 = ia6->ia_next;
556 		/* check address lifetime */
557 		lt6 = &ia6->ia6_lifetime;
558 		if (IFA6_IS_INVALID(ia6)) {
559 			int regen = 0;
560 
561 			/*
562 			 * If the expiring address is temporary, try
563 			 * regenerating a new one.  This would be useful when
564 			 * we suspended a laptop PC, then turned it on after a
565 			 * period that could invalidate all temporary
566 			 * addresses.  Although we may have to restart the
567 			 * loop (see below), it must be after purging the
568 			 * address.  Otherwise, we'd see an infinite loop of
569 			 * regeneration.
570 			 */
571 			if (ip6_use_tempaddr &&
572 			    (ia6->ia6_flags & IN6_IFF_TEMPORARY) != 0) {
573 				if (regen_tmpaddr(ia6) == 0)
574 					regen = 1;
575 			}
576 
577 			in6_purgeaddr(&ia6->ia_ifa);
578 
579 			if (regen)
580 				goto addrloop; /* XXX: see below */
581 		}
582 		if (IFA6_IS_DEPRECATED(ia6)) {
583 			int oldflags = ia6->ia6_flags;
584 
585 			ia6->ia6_flags |= IN6_IFF_DEPRECATED;
586 
587 			/*
588 			 * If a temporary address has just become deprecated,
589 			 * regenerate a new one if possible.
590 			 */
591 			if (ip6_use_tempaddr &&
592 			    (ia6->ia6_flags & IN6_IFF_TEMPORARY) != 0 &&
593 			    (oldflags & IN6_IFF_DEPRECATED) == 0) {
594 
595 				if (regen_tmpaddr(ia6) == 0) {
596 					/*
597 					 * A new temporary address is
598 					 * generated.
599 					 * XXX: this means the address chain
600 					 * has changed while we are still in
601 					 * the loop.  Although the change
602 					 * would not cause disaster (because
603 					 * it's not a deletion, but an
604 					 * addition,) we'd rather restart the
605 					 * loop just for safety.  Or does this
606 					 * significantly reduce performance??
607 					 */
608 					goto addrloop;
609 				}
610 			}
611 		} else {
612 			/*
613 			 * A new RA might have made a deprecated address
614 			 * preferred.
615 			 */
616 			ia6->ia6_flags &= ~IN6_IFF_DEPRECATED;
617 		}
618 	}
619 
620 	/* expire prefix list */
621 	pr = nd_prefix.lh_first;
622 	while (pr) {
623 		/*
624 		 * check prefix lifetime.
625 		 * since pltime is just for autoconf, pltime processing for
626 		 * prefix is not necessary.
627 		 */
628 		if (pr->ndpr_expire && pr->ndpr_expire < time_second) {
629 			struct nd_prefix *t;
630 			t = pr->ndpr_next;
631 
632 			/*
633 			 * address expiration and prefix expiration are
634 			 * separate.  NEVER perform in6_purgeaddr here.
635 			 */
636 
637 			prelist_remove(pr);
638 			pr = t;
639 		} else
640 			pr = pr->ndpr_next;
641 	}
642 	splx(s);
643 }
644 
645 static int
646 regen_tmpaddr(ia6)
647 	struct in6_ifaddr *ia6; /* deprecated/invalidated temporary address */
648 {
649 	struct ifaddr *ifa;
650 	struct ifnet *ifp;
651 	struct in6_ifaddr *public_ifa6 = NULL;
652 
653 	ifp = ia6->ia_ifa.ifa_ifp;
654 	for (ifa = ifp->if_addrlist.tqh_first; ifa;
655 	     ifa = ifa->ifa_list.tqe_next)
656 	{
657 		struct in6_ifaddr *it6;
658 
659 		if (ifa->ifa_addr->sa_family != AF_INET6)
660 			continue;
661 
662 		it6 = (struct in6_ifaddr *)ifa;
663 
664 		/* ignore no autoconf addresses. */
665 		if ((it6->ia6_flags & IN6_IFF_AUTOCONF) == 0)
666 			continue;
667 
668 		/* ignore autoconf addresses with different prefixes. */
669 		if (it6->ia6_ndpr == NULL || it6->ia6_ndpr != ia6->ia6_ndpr)
670 			continue;
671 
672 		/*
673 		 * Now we are looking at an autoconf address with the same
674 		 * prefix as ours.  If the address is temporary and is still
675 		 * preferred, do not create another one.  It would be rare, but
676 		 * could happen, for example, when we resume a laptop PC after
677 		 * a long period.
678 		 */
679 		if ((it6->ia6_flags & IN6_IFF_TEMPORARY) != 0 &&
680 		    !IFA6_IS_DEPRECATED(it6)) {
681 			public_ifa6 = NULL;
682 			break;
683 		}
684 
685 		/*
686 		 * This is a public autoconf address that has the same prefix
687 		 * as ours.  If it is preferred, keep it.  We can't break the
688 		 * loop here, because there may be a still-preferred temporary
689 		 * address with the prefix.
690 		 */
691 		if (!IFA6_IS_DEPRECATED(it6))
692 		    public_ifa6 = it6;
693 	}
694 
695 	if (public_ifa6 != NULL) {
696 		int e;
697 
698 		if ((e = in6_tmpifadd(public_ifa6, 0)) != 0) {
699 			log(LOG_NOTICE, "regen_tmpaddr: failed to create a new"
700 			    " tmp addr,errno=%d\n", e);
701 			return(-1);
702 		}
703 		return(0);
704 	}
705 
706 	return(-1);
707 }
708 
709 /*
710  * Nuke neighbor cache/prefix/default router management table, right before
711  * ifp goes away.
712  */
713 void
714 nd6_purge(ifp)
715 	struct ifnet *ifp;
716 {
717 	struct llinfo_nd6 *ln, *nln;
718 	struct nd_defrouter *dr, *ndr, drany;
719 	struct nd_prefix *pr, *npr;
720 
721 	/* Nuke default router list entries toward ifp */
722 	if ((dr = TAILQ_FIRST(&nd_defrouter)) != NULL) {
723 		/*
724 		 * The first entry of the list may be stored in
725 		 * the routing table, so we'll delete it later.
726 		 */
727 		for (dr = TAILQ_NEXT(dr, dr_entry); dr; dr = ndr) {
728 			ndr = TAILQ_NEXT(dr, dr_entry);
729 			if (dr->ifp == ifp)
730 				defrtrlist_del(dr);
731 		}
732 		dr = TAILQ_FIRST(&nd_defrouter);
733 		if (dr->ifp == ifp)
734 			defrtrlist_del(dr);
735 	}
736 
737 	/* Nuke prefix list entries toward ifp */
738 	for (pr = nd_prefix.lh_first; pr; pr = npr) {
739 		npr = pr->ndpr_next;
740 		if (pr->ndpr_ifp == ifp) {
741 			/*
742 			 * Previously, pr->ndpr_addr is removed as well,
743 			 * but I strongly believe we don't have to do it.
744 			 * nd6_purge() is only called from in6_ifdetach(),
745 			 * which removes all the associated interface addresses
746 			 * by itself.
747 			 * (jinmei@kame.net 20010129)
748 			 */
749 			prelist_remove(pr);
750 		}
751 	}
752 
753 	/* cancel default outgoing interface setting */
754 	if (nd6_defifindex == ifp->if_index)
755 		nd6_setdefaultiface(0);
756 
757 	if (!ip6_forwarding && ip6_accept_rtadv) { /* XXX: too restrictive? */
758 		/* refresh default router list */
759 		bzero(&drany, sizeof(drany));
760 		defrouter_delreq(&drany, 0);
761 		defrouter_select();
762 	}
763 
764 	/*
765 	 * Nuke neighbor cache entries for the ifp.
766 	 * Note that rt->rt_ifp may not be the same as ifp,
767 	 * due to KAME goto ours hack.  See RTM_RESOLVE case in
768 	 * nd6_rtrequest(), and ip6_input().
769 	 */
770 	ln = llinfo_nd6.ln_next;
771 	while (ln && ln != &llinfo_nd6) {
772 		struct rtentry *rt;
773 		struct sockaddr_dl *sdl;
774 
775 		nln = ln->ln_next;
776 		rt = ln->ln_rt;
777 		if (rt && rt->rt_gateway &&
778 		    rt->rt_gateway->sa_family == AF_LINK) {
779 			sdl = (struct sockaddr_dl *)rt->rt_gateway;
780 			if (sdl->sdl_index == ifp->if_index)
781 				nln = nd6_free(rt);
782 		}
783 		ln = nln;
784 	}
785 }
786 
787 struct rtentry *
788 nd6_lookup(addr6, create, ifp)
789 	struct in6_addr *addr6;
790 	int create;
791 	struct ifnet *ifp;
792 {
793 	struct rtentry *rt;
794 	struct sockaddr_in6 sin6;
795 
796 	bzero(&sin6, sizeof(sin6));
797 	sin6.sin6_len = sizeof(struct sockaddr_in6);
798 	sin6.sin6_family = AF_INET6;
799 	sin6.sin6_addr = *addr6;
800 #ifdef SCOPEDROUTING
801 	sin6.sin6_scope_id = in6_addr2scopeid(ifp, addr6);
802 #endif
803 	rt = rtalloc1((struct sockaddr *)&sin6, create, 0UL);
804 	if (rt && (rt->rt_flags & RTF_LLINFO) == 0) {
805 		/*
806 		 * This is the case for the default route.
807 		 * If we want to create a neighbor cache for the address, we
808 		 * should free the route for the destination and allocate an
809 		 * interface route.
810 		 */
811 		if (create) {
812 			RTFREE(rt);
813 			rt = 0;
814 		}
815 	}
816 	if (!rt) {
817 		if (create && ifp) {
818 			int e;
819 
820 			/*
821 			 * If no route is available and create is set,
822 			 * we allocate a host route for the destination
823 			 * and treat it like an interface route.
824 			 * This hack is necessary for a neighbor which can't
825 			 * be covered by our own prefix.
826 			 */
827 			struct ifaddr *ifa =
828 				ifaof_ifpforaddr((struct sockaddr *)&sin6, ifp);
829 			if (ifa == NULL)
830 				return(NULL);
831 
832 			/*
833 			 * Create a new route.  RTF_LLINFO is necessary
834 			 * to create a Neighbor Cache entry for the
835 			 * destination in nd6_rtrequest which will be
836 			 * called in rtrequest via ifa->ifa_rtrequest.
837 			 */
838 			if ((e = rtrequest(RTM_ADD, (struct sockaddr *)&sin6,
839 					   ifa->ifa_addr,
840 					   (struct sockaddr *)&all1_sa,
841 					   (ifa->ifa_flags |
842 					    RTF_HOST | RTF_LLINFO) &
843 					   ~RTF_CLONING,
844 					   &rt)) != 0)
845 				log(LOG_ERR,
846 				    "nd6_lookup: failed to add route for a "
847 				    "neighbor(%s), errno=%d\n",
848 				    ip6_sprintf(addr6), e);
849 			if (rt == NULL)
850 				return(NULL);
851 			if (rt->rt_llinfo) {
852 				struct llinfo_nd6 *ln =
853 					(struct llinfo_nd6 *)rt->rt_llinfo;
854 				ln->ln_state = ND6_LLINFO_NOSTATE;
855 			}
856 		} else
857 			return(NULL);
858 	}
859 	rt->rt_refcnt--;
860 	/*
861 	 * Validation for the entry.
862 	 * Note that the check for rt_llinfo is necessary because a cloned
863 	 * route from a parent route that has the L flag (e.g. the default
864 	 * route to a p2p interface) may have the flag, too, while the
865 	 * destination is not actually a neighbor.
866 	 * XXX: we can't use rt->rt_ifp to check for the interface, since
867 	 *      it might be the loopback interface if the entry is for our
868 	 *      own address on a non-loopback interface. Instead, we should
869 	 *      use rt->rt_ifa->ifa_ifp, which would specify the REAL
870 	 *      interface.
871 	 */
872 	if ((rt->rt_flags & RTF_GATEWAY) || (rt->rt_flags & RTF_LLINFO) == 0 ||
873 	    rt->rt_gateway->sa_family != AF_LINK || rt->rt_llinfo == NULL ||
874 	    (ifp && rt->rt_ifa->ifa_ifp != ifp)) {
875 		if (create) {
876 			log(LOG_DEBUG, "nd6_lookup: failed to lookup %s (if = %s)\n",
877 			    ip6_sprintf(addr6), ifp ? if_name(ifp) : "unspec");
878 			/* xxx more logs... kazu */
879 		}
880 		return(NULL);
881 	}
882 	return(rt);
883 }
884 
885 /*
886  * Detect if a given IPv6 address identifies a neighbor on a given link.
887  * XXX: should take care of the destination of a p2p link?
888  */
889 int
890 nd6_is_addr_neighbor(addr, ifp)
891 	struct sockaddr_in6 *addr;
892 	struct ifnet *ifp;
893 {
894 	struct ifaddr *ifa;
895 	int i;
896 
897 #define IFADDR6(a) ((((struct in6_ifaddr *)(a))->ia_addr).sin6_addr)
898 #define IFMASK6(a) ((((struct in6_ifaddr *)(a))->ia_prefixmask).sin6_addr)
899 
900 	/*
901 	 * A link-local address is always a neighbor.
902 	 * XXX: we should use the sin6_scope_id field rather than the embedded
903 	 * interface index.
904 	 */
905 	if (IN6_IS_ADDR_LINKLOCAL(&addr->sin6_addr) &&
906 	    ntohs(*(u_int16_t *)&addr->sin6_addr.s6_addr[2]) == ifp->if_index)
907 		return(1);
908 
909 	/*
910 	 * If the address matches one of our addresses,
911 	 * it should be a neighbor.
912 	 */
913 	for (ifa = ifp->if_addrlist.tqh_first;
914 	     ifa;
915 	     ifa = ifa->ifa_list.tqe_next)
916 	{
917 		if (ifa->ifa_addr->sa_family != AF_INET6)
918 			next: continue;
919 
920 		for (i = 0; i < 4; i++) {
921 			if ((IFADDR6(ifa).s6_addr32[i] ^
922 			     addr->sin6_addr.s6_addr32[i]) &
923 			    IFMASK6(ifa).s6_addr32[i])
924 				goto next;
925 		}
926 		return(1);
927 	}
928 
929 	/*
930 	 * Even if the address matches none of our addresses, it might be
931 	 * in the neighbor cache.
932 	 */
933 	if (nd6_lookup(&addr->sin6_addr, 0, ifp) != NULL)
934 		return(1);
935 
936 	return(0);
937 #undef IFADDR6
938 #undef IFMASK6
939 }
940 
941 /*
942  * Free an nd6 llinfo entry.
943  */
944 struct llinfo_nd6 *
945 nd6_free(rt)
946 	struct rtentry *rt;
947 {
948 	struct llinfo_nd6 *ln = (struct llinfo_nd6 *)rt->rt_llinfo, *next;
949 	struct in6_addr in6 = ((struct sockaddr_in6 *)rt_key(rt))->sin6_addr;
950 	struct nd_defrouter *dr;
951 
952 	/*
953 	 * we used to have pfctlinput(PRC_HOSTDEAD) here.
954 	 * even though it is not harmful, it was not really necessary.
955 	 */
956 
957 	if (!ip6_forwarding && ip6_accept_rtadv) { /* XXX: too restrictive? */
958 		int s;
959 		s = splnet();
960 		dr = defrouter_lookup(&((struct sockaddr_in6 *)rt_key(rt))->sin6_addr,
961 				      rt->rt_ifp);
962 
963 		if (ln->ln_router || dr) {
964 			/*
965 			 * rt6_flush must be called whether or not the neighbor
966 			 * is in the Default Router List.
967 			 * See a corresponding comment in nd6_na_input().
968 			 */
969 			rt6_flush(&in6, rt->rt_ifp);
970 		}
971 
972 		if (dr) {
973 			/*
974 			 * Unreachablity of a router might affect the default
975 			 * router selection and on-link detection of advertised
976 			 * prefixes.
977 			 */
978 
979 			/*
980 			 * Temporarily fake the state to choose a new default
981 			 * router and to perform on-link determination of
982 			 * prefixes correctly.
983 			 * Below the state will be set correctly,
984 			 * or the entry itself will be deleted.
985 			 */
986 			ln->ln_state = ND6_LLINFO_INCOMPLETE;
987 
988 			/*
989 			 * Since defrouter_select() does not affect the
990 			 * on-link determination and MIP6 needs the check
991 			 * before the default router selection, we perform
992 			 * the check now.
993 			 */
994 			pfxlist_onlink_check();
995 
996 			if (dr == TAILQ_FIRST(&nd_defrouter)) {
997 				/*
998 				 * It is used as the current default router,
999 				 * so we have to move it to the end of the
1000 				 * list and choose a new one.
1001 				 * XXX: it is not very efficient if this is
1002 				 *      the only router.
1003 				 */
1004 				TAILQ_REMOVE(&nd_defrouter, dr, dr_entry);
1005 				TAILQ_INSERT_TAIL(&nd_defrouter, dr, dr_entry);
1006 
1007 				defrouter_select();
1008 			}
1009 		}
1010 		splx(s);
1011 	}
1012 
1013 	/*
1014 	 * Before deleting the entry, remember the next entry as the
1015 	 * return value.  We need this because pfxlist_onlink_check() above
1016 	 * might have freed other entries (particularly the old next entry) as
1017 	 * a side effect (XXX).
1018 	 */
1019 	next = ln->ln_next;
1020 
1021 	/*
1022 	 * Detach the route from the routing tree and the list of neighbor
1023 	 * caches, and disable the route entry not to be used in already
1024 	 * cached routes.
1025 	 */
1026 	rtrequest(RTM_DELETE, rt_key(rt), (struct sockaddr *)0,
1027 		  rt_mask(rt), 0, (struct rtentry **)0);
1028 
1029 	return(next);
1030 }
1031 
1032 /*
1033  * Upper-layer reachability hint for Neighbor Unreachability Detection.
1034  *
1035  * XXX cost-effective metods?
1036  */
1037 void
1038 nd6_nud_hint(rt, dst6, force)
1039 	struct rtentry *rt;
1040 	struct in6_addr *dst6;
1041 	int force;
1042 {
1043 	struct llinfo_nd6 *ln;
1044 
1045 	/*
1046 	 * If the caller specified "rt", use that.  Otherwise, resolve the
1047 	 * routing table by supplied "dst6".
1048 	 */
1049 	if (!rt) {
1050 		if (!dst6)
1051 			return;
1052 		if (!(rt = nd6_lookup(dst6, 0, NULL)))
1053 			return;
1054 	}
1055 
1056 	if ((rt->rt_flags & RTF_GATEWAY) != 0 ||
1057 	    (rt->rt_flags & RTF_LLINFO) == 0 ||
1058 	    !rt->rt_llinfo || !rt->rt_gateway ||
1059 	    rt->rt_gateway->sa_family != AF_LINK) {
1060 		/* This is not a host route. */
1061 		return;
1062 	}
1063 
1064 	ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1065 	if (ln->ln_state < ND6_LLINFO_REACHABLE)
1066 		return;
1067 
1068 	/*
1069 	 * if we get upper-layer reachability confirmation many times,
1070 	 * it is possible we have false information.
1071 	 */
1072 	if (!force) {
1073 		ln->ln_byhint++;
1074 		if (ln->ln_byhint > nd6_maxnudhint)
1075 			return;
1076 	}
1077 
1078 	ln->ln_state = ND6_LLINFO_REACHABLE;
1079 	if (ln->ln_expire)
1080 		ln->ln_expire = time_second +
1081 			nd_ifinfo[rt->rt_ifp->if_index].reachable;
1082 }
1083 
1084 void
1085 nd6_rtrequest(req, rt, info)
1086 	int	req;
1087 	struct rtentry *rt;
1088 	struct rt_addrinfo *info; /* xxx unused */
1089 {
1090 	struct sockaddr *gate = rt->rt_gateway;
1091 	struct llinfo_nd6 *ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1092 	static struct sockaddr_dl null_sdl = {sizeof(null_sdl), AF_LINK};
1093 	struct ifnet *ifp = rt->rt_ifp;
1094 	struct ifaddr *ifa;
1095 
1096 	if ((rt->rt_flags & RTF_GATEWAY))
1097 		return;
1098 
1099 	if (nd6_need_cache(ifp) == 0 && (rt->rt_flags & RTF_HOST) == 0) {
1100 		/*
1101 		 * This is probably an interface direct route for a link
1102 		 * which does not need neighbor caches (e.g. fe80::%lo0/64).
1103 		 * We do not need special treatment below for such a route.
1104 		 * Moreover, the RTF_LLINFO flag which would be set below
1105 		 * would annoy the ndp(8) command.
1106 		 */
1107 		return;
1108 	}
1109 
1110 	if (req == RTM_RESOLVE &&
1111 	    (nd6_need_cache(ifp) == 0 || /* stf case */
1112 	     !nd6_is_addr_neighbor((struct sockaddr_in6 *)rt_key(rt), ifp))) {
1113 		/*
1114 		 * FreeBSD and BSD/OS often make a cloned host route based
1115 		 * on a less-specific route (e.g. the default route).
1116 		 * If the less specific route does not have a "gateway"
1117 		 * (this is the case when the route just goes to a p2p or an
1118 		 * stf interface), we'll mistakenly make a neighbor cache for
1119 		 * the host route, and will see strange neighbor solicitation
1120 		 * for the corresponding destination.  In order to avoid the
1121 		 * confusion, we check if the destination of the route is
1122 		 * a neighbor in terms of neighbor discovery, and stop the
1123 		 * process if not.  Additionally, we remove the LLINFO flag
1124 		 * so that ndp(8) will not try to get the neighbor information
1125 		 * of the destination.
1126 		 */
1127 		rt->rt_flags &= ~RTF_LLINFO;
1128 		return;
1129 	}
1130 
1131 	switch (req) {
1132 	case RTM_ADD:
1133 		/*
1134 		 * There is no backward compatibility :)
1135 		 *
1136 		 * if ((rt->rt_flags & RTF_HOST) == 0 &&
1137 		 *     SIN(rt_mask(rt))->sin_addr.s_addr != 0xffffffff)
1138 		 *	   rt->rt_flags |= RTF_CLONING;
1139 		 */
1140 		if (rt->rt_flags & (RTF_CLONING | RTF_LLINFO)) {
1141 			/*
1142 			 * Case 1: This route should come from
1143 			 * a route to interface.  RTF_LLINFO flag is set
1144 			 * for a host route whose destination should be
1145 			 * treated as on-link.
1146 			 */
1147 			rt_setgate(rt, rt_key(rt),
1148 				   (struct sockaddr *)&null_sdl);
1149 			gate = rt->rt_gateway;
1150 			SDL(gate)->sdl_type = ifp->if_type;
1151 			SDL(gate)->sdl_index = ifp->if_index;
1152 			if (ln)
1153 				ln->ln_expire = time_second;
1154 #if 1
1155 			if (ln && ln->ln_expire == 0) {
1156 				/* kludge for desktops */
1157 #if 0
1158 				printf("nd6_rtequest: time.tv_sec is zero; "
1159 				       "treat it as 1\n");
1160 #endif
1161 				ln->ln_expire = 1;
1162 			}
1163 #endif
1164 			if ((rt->rt_flags & RTF_CLONING))
1165 				break;
1166 		}
1167 		/*
1168 		 * In IPv4 code, we try to annonuce new RTF_ANNOUNCE entry here.
1169 		 * We don't do that here since llinfo is not ready yet.
1170 		 *
1171 		 * There are also couple of other things to be discussed:
1172 		 * - unsolicited NA code needs improvement beforehand
1173 		 * - RFC2461 says we MAY send multicast unsolicited NA
1174 		 *   (7.2.6 paragraph 4), however, it also says that we
1175 		 *   SHOULD provide a mechanism to prevent multicast NA storm.
1176 		 *   we don't have anything like it right now.
1177 		 *   note that the mechanism needs a mutual agreement
1178 		 *   between proxies, which means that we need to implement
1179 		 *   a new protocol, or a new kludge.
1180 		 * - from RFC2461 6.2.4, host MUST NOT send an unsolicited NA.
1181 		 *   we need to check ip6forwarding before sending it.
1182 		 *   (or should we allow proxy ND configuration only for
1183 		 *   routers?  there's no mention about proxy ND from hosts)
1184 		 */
1185 #if 0
1186 		/* XXX it does not work */
1187 		if (rt->rt_flags & RTF_ANNOUNCE)
1188 			nd6_na_output(ifp,
1189 			      &SIN6(rt_key(rt))->sin6_addr,
1190 			      &SIN6(rt_key(rt))->sin6_addr,
1191 			      ip6_forwarding ? ND_NA_FLAG_ROUTER : 0,
1192 			      1, NULL);
1193 #endif
1194 		/* FALLTHROUGH */
1195 	case RTM_RESOLVE:
1196 		if ((ifp->if_flags & (IFF_POINTOPOINT | IFF_LOOPBACK)) == 0) {
1197 			/*
1198 			 * Address resolution isn't necessary for a point to
1199 			 * point link, so we can skip this test for a p2p link.
1200 			 */
1201 			if (gate->sa_family != AF_LINK ||
1202 			    gate->sa_len < sizeof(null_sdl)) {
1203 				log(LOG_DEBUG,
1204 				    "nd6_rtrequest: bad gateway value: %s\n",
1205 				    if_name(ifp));
1206 				break;
1207 			}
1208 			SDL(gate)->sdl_type = ifp->if_type;
1209 			SDL(gate)->sdl_index = ifp->if_index;
1210 		}
1211 		if (ln != NULL)
1212 			break;	/* This happens on a route change */
1213 		/*
1214 		 * Case 2: This route may come from cloning, or a manual route
1215 		 * add with a LL address.
1216 		 */
1217 		R_Malloc(ln, struct llinfo_nd6 *, sizeof(*ln));
1218 		rt->rt_llinfo = (caddr_t)ln;
1219 		if (!ln) {
1220 			log(LOG_DEBUG, "nd6_rtrequest: malloc failed\n");
1221 			break;
1222 		}
1223 		nd6_inuse++;
1224 		nd6_allocated++;
1225 		Bzero(ln, sizeof(*ln));
1226 		ln->ln_rt = rt;
1227 		/* this is required for "ndp" command. - shin */
1228 		if (req == RTM_ADD) {
1229 		        /*
1230 			 * gate should have some valid AF_LINK entry,
1231 			 * and ln->ln_expire should have some lifetime
1232 			 * which is specified by ndp command.
1233 			 */
1234 			ln->ln_state = ND6_LLINFO_REACHABLE;
1235 			ln->ln_byhint = 0;
1236 		} else {
1237 		        /*
1238 			 * When req == RTM_RESOLVE, rt is created and
1239 			 * initialized in rtrequest(), so rt_expire is 0.
1240 			 */
1241 			ln->ln_state = ND6_LLINFO_NOSTATE;
1242 			ln->ln_expire = time_second;
1243 		}
1244 		rt->rt_flags |= RTF_LLINFO;
1245 		ln->ln_next = llinfo_nd6.ln_next;
1246 		llinfo_nd6.ln_next = ln;
1247 		ln->ln_prev = &llinfo_nd6;
1248 		ln->ln_next->ln_prev = ln;
1249 
1250 		/*
1251 		 * check if rt_key(rt) is one of my address assigned
1252 		 * to the interface.
1253 		 */
1254 		ifa = (struct ifaddr *)in6ifa_ifpwithaddr(rt->rt_ifp,
1255 					  &SIN6(rt_key(rt))->sin6_addr);
1256 		if (ifa) {
1257 			caddr_t macp = nd6_ifptomac(ifp);
1258 			ln->ln_expire = 0;
1259 			ln->ln_state = ND6_LLINFO_REACHABLE;
1260 			ln->ln_byhint = 0;
1261 			if (macp) {
1262 				Bcopy(macp, LLADDR(SDL(gate)), ifp->if_addrlen);
1263 				SDL(gate)->sdl_alen = ifp->if_addrlen;
1264 			}
1265 			if (nd6_useloopback) {
1266 				rt->rt_ifp = &loif[0];	/* XXX */
1267 				/*
1268 				 * Make sure rt_ifa be equal to the ifaddr
1269 				 * corresponding to the address.
1270 				 * We need this because when we refer
1271 				 * rt_ifa->ia6_flags in ip6_input, we assume
1272 				 * that the rt_ifa points to the address instead
1273 				 * of the loopback address.
1274 				 */
1275 				if (ifa != rt->rt_ifa) {
1276 					IFAFREE(rt->rt_ifa);
1277 					IFAREF(ifa);
1278 					rt->rt_ifa = ifa;
1279 				}
1280 			}
1281 		} else if (rt->rt_flags & RTF_ANNOUNCE) {
1282 			ln->ln_expire = 0;
1283 			ln->ln_state = ND6_LLINFO_REACHABLE;
1284 			ln->ln_byhint = 0;
1285 
1286 			/* join solicited node multicast for proxy ND */
1287 			if (ifp->if_flags & IFF_MULTICAST) {
1288 				struct in6_addr llsol;
1289 				int error;
1290 
1291 				llsol = SIN6(rt_key(rt))->sin6_addr;
1292 				llsol.s6_addr16[0] = htons(0xff02);
1293 				llsol.s6_addr16[1] = htons(ifp->if_index);
1294 				llsol.s6_addr32[1] = 0;
1295 				llsol.s6_addr32[2] = htonl(1);
1296 				llsol.s6_addr8[12] = 0xff;
1297 
1298 				if (!in6_addmulti(&llsol, ifp, &error)) {
1299 					nd6log((LOG_ERR, "%s: failed to join "
1300 					    "%s (errno=%d)\n", if_name(ifp),
1301 					    ip6_sprintf(&llsol), error));
1302 				}
1303 			}
1304 		}
1305 		break;
1306 
1307 	case RTM_DELETE:
1308 		if (!ln)
1309 			break;
1310 		/* leave from solicited node multicast for proxy ND */
1311 		if ((rt->rt_flags & RTF_ANNOUNCE) != 0 &&
1312 		    (ifp->if_flags & IFF_MULTICAST) != 0) {
1313 			struct in6_addr llsol;
1314 			struct in6_multi *in6m;
1315 
1316 			llsol = SIN6(rt_key(rt))->sin6_addr;
1317 			llsol.s6_addr16[0] = htons(0xff02);
1318 			llsol.s6_addr16[1] = htons(ifp->if_index);
1319 			llsol.s6_addr32[1] = 0;
1320 			llsol.s6_addr32[2] = htonl(1);
1321 			llsol.s6_addr8[12] = 0xff;
1322 
1323 			IN6_LOOKUP_MULTI(llsol, ifp, in6m);
1324 			if (in6m)
1325 				in6_delmulti(in6m);
1326 		}
1327 		nd6_inuse--;
1328 		ln->ln_next->ln_prev = ln->ln_prev;
1329 		ln->ln_prev->ln_next = ln->ln_next;
1330 		ln->ln_prev = NULL;
1331 		rt->rt_llinfo = 0;
1332 		rt->rt_flags &= ~RTF_LLINFO;
1333 		if (ln->ln_hold)
1334 			m_freem(ln->ln_hold);
1335 		Free((caddr_t)ln);
1336 	}
1337 }
1338 
1339 int
1340 nd6_ioctl(cmd, data, ifp)
1341 	u_long cmd;
1342 	caddr_t	data;
1343 	struct ifnet *ifp;
1344 {
1345 	struct in6_drlist *drl = (struct in6_drlist *)data;
1346 	struct in6_prlist *prl = (struct in6_prlist *)data;
1347 	struct in6_ndireq *ndi = (struct in6_ndireq *)data;
1348 	struct in6_nbrinfo *nbi = (struct in6_nbrinfo *)data;
1349 	struct in6_ndifreq *ndif = (struct in6_ndifreq *)data;
1350 	struct nd_defrouter *dr, any;
1351 	struct nd_prefix *pr;
1352 	struct rtentry *rt;
1353 	int i = 0, error = 0;
1354 	int s;
1355 
1356 	switch (cmd) {
1357 	case SIOCGDRLST_IN6:
1358 		/*
1359 		 * obsolete API, use sysctl under net.inet6.icmp6
1360 		 */
1361 		bzero(drl, sizeof(*drl));
1362 		s = splnet();
1363 		dr = TAILQ_FIRST(&nd_defrouter);
1364 		while (dr && i < DRLSTSIZ) {
1365 			drl->defrouter[i].rtaddr = dr->rtaddr;
1366 			if (IN6_IS_ADDR_LINKLOCAL(&drl->defrouter[i].rtaddr)) {
1367 				/* XXX: need to this hack for KAME stack */
1368 				drl->defrouter[i].rtaddr.s6_addr16[1] = 0;
1369 			} else
1370 				log(LOG_ERR,
1371 				    "default router list contains a "
1372 				    "non-linklocal address(%s)\n",
1373 				    ip6_sprintf(&drl->defrouter[i].rtaddr));
1374 
1375 			drl->defrouter[i].flags = dr->flags;
1376 			drl->defrouter[i].rtlifetime = dr->rtlifetime;
1377 			drl->defrouter[i].expire = dr->expire;
1378 			drl->defrouter[i].if_index = dr->ifp->if_index;
1379 			i++;
1380 			dr = TAILQ_NEXT(dr, dr_entry);
1381 		}
1382 		splx(s);
1383 		break;
1384 	case SIOCGPRLST_IN6:
1385 		/*
1386 		 * obsolete API, use sysctl under net.inet6.icmp6
1387 		 */
1388 		/*
1389 		 * XXX meaning of fields, especialy "raflags", is very
1390 		 * differnet between RA prefix list and RR/static prefix list.
1391 		 * how about separating ioctls into two?
1392 		 */
1393 		bzero(prl, sizeof(*prl));
1394 		s = splnet();
1395 		pr = nd_prefix.lh_first;
1396 		while (pr && i < PRLSTSIZ) {
1397 			struct nd_pfxrouter *pfr;
1398 			int j;
1399 
1400 			(void)in6_embedscope(&prl->prefix[i].prefix,
1401 			    &pr->ndpr_prefix, NULL, NULL);
1402 			prl->prefix[i].raflags = pr->ndpr_raf;
1403 			prl->prefix[i].prefixlen = pr->ndpr_plen;
1404 			prl->prefix[i].vltime = pr->ndpr_vltime;
1405 			prl->prefix[i].pltime = pr->ndpr_pltime;
1406 			prl->prefix[i].if_index = pr->ndpr_ifp->if_index;
1407 			prl->prefix[i].expire = pr->ndpr_expire;
1408 
1409 			pfr = pr->ndpr_advrtrs.lh_first;
1410 			j = 0;
1411 			while (pfr) {
1412 				if (j < DRLSTSIZ) {
1413 #define RTRADDR prl->prefix[i].advrtr[j]
1414 					RTRADDR = pfr->router->rtaddr;
1415 					if (IN6_IS_ADDR_LINKLOCAL(&RTRADDR)) {
1416 						/* XXX: hack for KAME */
1417 						RTRADDR.s6_addr16[1] = 0;
1418 					} else
1419 						log(LOG_ERR,
1420 						    "a router(%s) advertises "
1421 						    "a prefix with "
1422 						    "non-link local address\n",
1423 						    ip6_sprintf(&RTRADDR));
1424 #undef RTRADDR
1425 				}
1426 				j++;
1427 				pfr = pfr->pfr_next;
1428 			}
1429 			prl->prefix[i].advrtrs = j;
1430 			prl->prefix[i].origin = PR_ORIG_RA;
1431 
1432 			i++;
1433 			pr = pr->ndpr_next;
1434 		}
1435 	      {
1436 		struct rr_prefix *rpp;
1437 
1438 		for (rpp = LIST_FIRST(&rr_prefix); rpp;
1439 		     rpp = LIST_NEXT(rpp, rp_entry)) {
1440 			if (i >= PRLSTSIZ)
1441 				break;
1442 			(void)in6_embedscope(&prl->prefix[i].prefix,
1443 			    &pr->ndpr_prefix, NULL, NULL);
1444 			prl->prefix[i].raflags = rpp->rp_raf;
1445 			prl->prefix[i].prefixlen = rpp->rp_plen;
1446 			prl->prefix[i].vltime = rpp->rp_vltime;
1447 			prl->prefix[i].pltime = rpp->rp_pltime;
1448 			prl->prefix[i].if_index = rpp->rp_ifp->if_index;
1449 			prl->prefix[i].expire = rpp->rp_expire;
1450 			prl->prefix[i].advrtrs = 0;
1451 			prl->prefix[i].origin = rpp->rp_origin;
1452 			i++;
1453 		}
1454 	      }
1455 		splx(s);
1456 
1457 		break;
1458 	case OSIOCGIFINFO_IN6:
1459 		if (!nd_ifinfo || i >= nd_ifinfo_indexlim) {
1460 			error = EINVAL;
1461 			break;
1462 		}
1463 		ndi->ndi.linkmtu = nd_ifinfo[ifp->if_index].linkmtu;
1464 		ndi->ndi.maxmtu = nd_ifinfo[ifp->if_index].maxmtu;
1465 		ndi->ndi.basereachable =
1466 		    nd_ifinfo[ifp->if_index].basereachable;
1467 		ndi->ndi.reachable = nd_ifinfo[ifp->if_index].reachable;
1468 		ndi->ndi.retrans = nd_ifinfo[ifp->if_index].retrans;
1469 		ndi->ndi.flags = nd_ifinfo[ifp->if_index].flags;
1470 		ndi->ndi.recalctm = nd_ifinfo[ifp->if_index].recalctm;
1471 		ndi->ndi.chlim = nd_ifinfo[ifp->if_index].chlim;
1472 		ndi->ndi.receivedra = nd_ifinfo[ifp->if_index].receivedra;
1473 		break;
1474 	case SIOCGIFINFO_IN6:
1475 		if (!nd_ifinfo || i >= nd_ifinfo_indexlim) {
1476 			error = EINVAL;
1477 			break;
1478 		}
1479 		ndi->ndi = nd_ifinfo[ifp->if_index];
1480 		break;
1481 	case SIOCSIFINFO_FLAGS:
1482 		/* XXX: almost all other fields of ndi->ndi is unused */
1483 		if (!nd_ifinfo || i >= nd_ifinfo_indexlim) {
1484 			error = EINVAL;
1485 			break;
1486 		}
1487 		nd_ifinfo[ifp->if_index].flags = ndi->ndi.flags;
1488 		break;
1489 	case SIOCSNDFLUSH_IN6:	/* XXX: the ioctl name is confusing... */
1490 		/* flush default router list */
1491 		/*
1492 		 * xxx sumikawa: should not delete route if default
1493 		 * route equals to the top of default router list
1494 		 */
1495 		bzero(&any, sizeof(any));
1496 		defrouter_delreq(&any, 0);
1497 		defrouter_select();
1498 		/* xxx sumikawa: flush prefix list */
1499 		break;
1500 	case SIOCSPFXFLUSH_IN6:
1501 	    {
1502 		/* flush all the prefix advertised by routers */
1503 		struct nd_prefix *pr, *next;
1504 
1505 		s = splnet();
1506 		for (pr = nd_prefix.lh_first; pr; pr = next) {
1507 			struct in6_ifaddr *ia, *ia_next;
1508 
1509 			next = pr->ndpr_next;
1510 
1511 			if (IN6_IS_ADDR_LINKLOCAL(&pr->ndpr_prefix.sin6_addr))
1512 				continue; /* XXX */
1513 
1514 			/* do we really have to remove addresses as well? */
1515 			for (ia = in6_ifaddr; ia; ia = ia_next) {
1516 				/* ia might be removed.  keep the next ptr. */
1517 				ia_next = ia->ia_next;
1518 
1519 				if ((ia->ia6_flags & IN6_IFF_AUTOCONF) == 0)
1520 					continue;
1521 
1522 				if (ia->ia6_ndpr == pr)
1523 					in6_purgeaddr(&ia->ia_ifa);
1524 			}
1525 			prelist_remove(pr);
1526 		}
1527 		splx(s);
1528 		break;
1529 	    }
1530 	case SIOCSRTRFLUSH_IN6:
1531 	    {
1532 		/* flush all the default routers */
1533 		struct nd_defrouter *dr, *next;
1534 
1535 		s = splnet();
1536 		if ((dr = TAILQ_FIRST(&nd_defrouter)) != NULL) {
1537 			/*
1538 			 * The first entry of the list may be stored in
1539 			 * the routing table, so we'll delete it later.
1540 			 */
1541 			for (dr = TAILQ_NEXT(dr, dr_entry); dr; dr = next) {
1542 				next = TAILQ_NEXT(dr, dr_entry);
1543 				defrtrlist_del(dr);
1544 			}
1545 			defrtrlist_del(TAILQ_FIRST(&nd_defrouter));
1546 		}
1547 		splx(s);
1548 		break;
1549 	    }
1550 	case SIOCGNBRINFO_IN6:
1551 	    {
1552 		struct llinfo_nd6 *ln;
1553 		struct in6_addr nb_addr = nbi->addr; /* make local for safety */
1554 
1555 		/*
1556 		 * XXX: KAME specific hack for scoped addresses
1557 		 *      XXXX: for other scopes than link-local?
1558 		 */
1559 		if (IN6_IS_ADDR_LINKLOCAL(&nbi->addr) ||
1560 		    IN6_IS_ADDR_MC_LINKLOCAL(&nbi->addr)) {
1561 			u_int16_t *idp = (u_int16_t *)&nb_addr.s6_addr[2];
1562 
1563 			if (*idp == 0)
1564 				*idp = htons(ifp->if_index);
1565 		}
1566 
1567 		s = splnet();
1568 		if ((rt = nd6_lookup(&nb_addr, 0, ifp)) == NULL) {
1569 			error = EINVAL;
1570 			splx(s);
1571 			break;
1572 		}
1573 		ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1574 		nbi->state = ln->ln_state;
1575 		nbi->asked = ln->ln_asked;
1576 		nbi->isrouter = ln->ln_router;
1577 		nbi->expire = ln->ln_expire;
1578 		splx(s);
1579 
1580 		break;
1581 	    }
1582 	case SIOCGDEFIFACE_IN6:	/* XXX: should be implemented as a sysctl? */
1583 		ndif->ifindex = nd6_defifindex;
1584 		break;
1585 	case SIOCSDEFIFACE_IN6:	/* XXX: should be implemented as a sysctl? */
1586 		return(nd6_setdefaultiface(ndif->ifindex));
1587 		break;
1588 	}
1589 	return(error);
1590 }
1591 
1592 /*
1593  * Create neighbor cache entry and cache link-layer address,
1594  * on reception of inbound ND6 packets. (RS/RA/NS/redirect)
1595  */
1596 struct rtentry *
1597 nd6_cache_lladdr(ifp, from, lladdr, lladdrlen, type, code)
1598 	struct ifnet *ifp;
1599 	struct in6_addr *from;
1600 	char *lladdr;
1601 	int lladdrlen;
1602 	int type;	/* ICMP6 type */
1603 	int code;	/* type dependent information */
1604 {
1605 	struct rtentry *rt = NULL;
1606 	struct llinfo_nd6 *ln = NULL;
1607 	int is_newentry;
1608 	struct sockaddr_dl *sdl = NULL;
1609 	int do_update;
1610 	int olladdr;
1611 	int llchange;
1612 	int newstate = 0;
1613 
1614 	if (!ifp)
1615 		panic("ifp == NULL in nd6_cache_lladdr");
1616 	if (!from)
1617 		panic("from == NULL in nd6_cache_lladdr");
1618 
1619 	/* nothing must be updated for unspecified address */
1620 	if (IN6_IS_ADDR_UNSPECIFIED(from))
1621 		return NULL;
1622 
1623 	/*
1624 	 * Validation about ifp->if_addrlen and lladdrlen must be done in
1625 	 * the caller.
1626 	 *
1627 	 * XXX If the link does not have link-layer adderss, what should
1628 	 * we do? (ifp->if_addrlen == 0)
1629 	 * Spec says nothing in sections for RA, RS and NA.  There's small
1630 	 * description on it in NS section (RFC 2461 7.2.3).
1631 	 */
1632 
1633 	rt = nd6_lookup(from, 0, ifp);
1634 	if (!rt) {
1635 #if 0
1636 		/* nothing must be done if there's no lladdr */
1637 		if (!lladdr || !lladdrlen)
1638 			return NULL;
1639 #endif
1640 
1641 		rt = nd6_lookup(from, 1, ifp);
1642 		is_newentry = 1;
1643 	} else {
1644 		/* do nothing if static ndp is set */
1645 		if (rt->rt_flags & RTF_STATIC)
1646 			return NULL;
1647 		is_newentry = 0;
1648 	}
1649 
1650 	if (!rt)
1651 		return NULL;
1652 	if ((rt->rt_flags & (RTF_GATEWAY | RTF_LLINFO)) != RTF_LLINFO) {
1653 fail:
1654 		(void)nd6_free(rt);
1655 		return NULL;
1656 	}
1657 	ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1658 	if (!ln)
1659 		goto fail;
1660 	if (!rt->rt_gateway)
1661 		goto fail;
1662 	if (rt->rt_gateway->sa_family != AF_LINK)
1663 		goto fail;
1664 	sdl = SDL(rt->rt_gateway);
1665 
1666 	olladdr = (sdl->sdl_alen) ? 1 : 0;
1667 	if (olladdr && lladdr) {
1668 		if (bcmp(lladdr, LLADDR(sdl), ifp->if_addrlen))
1669 			llchange = 1;
1670 		else
1671 			llchange = 0;
1672 	} else
1673 		llchange = 0;
1674 
1675 	/*
1676 	 * newentry olladdr  lladdr  llchange	(*=record)
1677 	 *	0	n	n	--	(1)
1678 	 *	0	y	n	--	(2)
1679 	 *	0	n	y	--	(3) * STALE
1680 	 *	0	y	y	n	(4) *
1681 	 *	0	y	y	y	(5) * STALE
1682 	 *	1	--	n	--	(6)   NOSTATE(= PASSIVE)
1683 	 *	1	--	y	--	(7) * STALE
1684 	 */
1685 
1686 	if (lladdr) {		/* (3-5) and (7) */
1687 		/*
1688 		 * Record source link-layer address
1689 		 * XXX is it dependent to ifp->if_type?
1690 		 */
1691 		sdl->sdl_alen = ifp->if_addrlen;
1692 		bcopy(lladdr, LLADDR(sdl), ifp->if_addrlen);
1693 	}
1694 
1695 	if (!is_newentry) {
1696 		if ((!olladdr && lladdr)		/* (3) */
1697 		 || (olladdr && lladdr && llchange)) {	/* (5) */
1698 			do_update = 1;
1699 			newstate = ND6_LLINFO_STALE;
1700 		} else					/* (1-2,4) */
1701 			do_update = 0;
1702 	} else {
1703 		do_update = 1;
1704 		if (!lladdr)				/* (6) */
1705 			newstate = ND6_LLINFO_NOSTATE;
1706 		else					/* (7) */
1707 			newstate = ND6_LLINFO_STALE;
1708 	}
1709 
1710 	if (do_update) {
1711 		/*
1712 		 * Update the state of the neighbor cache.
1713 		 */
1714 		ln->ln_state = newstate;
1715 
1716 		if (ln->ln_state == ND6_LLINFO_STALE) {
1717 			/*
1718 			 * XXX: since nd6_output() below will cause
1719 			 * state tansition to DELAY and reset the timer,
1720 			 * we must set the timer now, although it is actually
1721 			 * meaningless.
1722 			 */
1723 			ln->ln_expire = time_second + nd6_gctimer;
1724 
1725 			if (ln->ln_hold) {
1726 				/*
1727 				 * we assume ifp is not a p2p here, so just
1728 				 * set the 2nd argument as the 1st one.
1729 				 */
1730 				nd6_output(ifp, ifp, ln->ln_hold,
1731 					   (struct sockaddr_in6 *)rt_key(rt),
1732 					   rt);
1733 				ln->ln_hold = NULL;
1734 			}
1735 		} else if (ln->ln_state == ND6_LLINFO_INCOMPLETE) {
1736 			/* probe right away */
1737 			ln->ln_expire = time_second;
1738 		}
1739 	}
1740 
1741 	/*
1742 	 * ICMP6 type dependent behavior.
1743 	 *
1744 	 * NS: clear IsRouter if new entry
1745 	 * RS: clear IsRouter
1746 	 * RA: set IsRouter if there's lladdr
1747 	 * redir: clear IsRouter if new entry
1748 	 *
1749 	 * RA case, (1):
1750 	 * The spec says that we must set IsRouter in the following cases:
1751 	 * - If lladdr exist, set IsRouter.  This means (1-5).
1752 	 * - If it is old entry (!newentry), set IsRouter.  This means (7).
1753 	 * So, based on the spec, in (1-5) and (7) cases we must set IsRouter.
1754 	 * A quetion arises for (1) case.  (1) case has no lladdr in the
1755 	 * neighbor cache, this is similar to (6).
1756 	 * This case is rare but we figured that we MUST NOT set IsRouter.
1757 	 *
1758 	 * newentry olladdr  lladdr  llchange	    NS  RS  RA	redir
1759 	 *							D R
1760 	 *	0	n	n	--	(1)	c   ?     s
1761 	 *	0	y	n	--	(2)	c   s     s
1762 	 *	0	n	y	--	(3)	c   s     s
1763 	 *	0	y	y	n	(4)	c   s     s
1764 	 *	0	y	y	y	(5)	c   s     s
1765 	 *	1	--	n	--	(6) c	c 	c s
1766 	 *	1	--	y	--	(7) c	c   s	c s
1767 	 *
1768 	 *					(c=clear s=set)
1769 	 */
1770 	switch (type & 0xff) {
1771 	case ND_NEIGHBOR_SOLICIT:
1772 		/*
1773 		 * New entry must have is_router flag cleared.
1774 		 */
1775 		if (is_newentry)	/* (6-7) */
1776 			ln->ln_router = 0;
1777 		break;
1778 	case ND_REDIRECT:
1779 		/*
1780 		 * If the icmp is a redirect to a better router, always set the
1781 		 * is_router flag. Otherwise, if the entry is newly created,
1782 		 * clear the flag. [RFC 2461, sec 8.3]
1783 		 */
1784 		if (code == ND_REDIRECT_ROUTER)
1785 			ln->ln_router = 1;
1786 		else if (is_newentry) /* (6-7) */
1787 			ln->ln_router = 0;
1788 		break;
1789 	case ND_ROUTER_SOLICIT:
1790 		/*
1791 		 * is_router flag must always be cleared.
1792 		 */
1793 		ln->ln_router = 0;
1794 		break;
1795 	case ND_ROUTER_ADVERT:
1796 		/*
1797 		 * Mark an entry with lladdr as a router.
1798 		 */
1799 		if ((!is_newentry && (olladdr || lladdr))	/* (2-5) */
1800 		 || (is_newentry && lladdr)) {			/* (7) */
1801 			ln->ln_router = 1;
1802 		}
1803 		break;
1804 	}
1805 
1806 	/*
1807 	 * When the link-layer address of a router changes, select the
1808 	 * best router again.  In particular, when the neighbor entry is newly
1809 	 * created, it might affect the selection policy.
1810 	 * Question: can we restrict the first condition to the "is_newentry"
1811 	 * case?
1812 	 * XXX: when we hear an RA from a new router with the link-layer
1813 	 * address option, defrouter_select() is called twice, since
1814 	 * defrtrlist_update called the function as well.  However, I believe
1815 	 * we can compromise the overhead, since it only happens the first
1816 	 * time.
1817 	 * XXX: although defrouter_select() should not have a bad effect
1818 	 * for those are not autoconfigured hosts, we explicitly avoid such
1819 	 * cases for safety.
1820 	 */
1821 	if (do_update && ln->ln_router && !ip6_forwarding && ip6_accept_rtadv)
1822 		defrouter_select();
1823 
1824 	return rt;
1825 }
1826 
1827 static void
1828 nd6_slowtimo(ignored_arg)
1829     void *ignored_arg;
1830 {
1831 	int s = splnet();
1832 	int i;
1833 	struct nd_ifinfo *nd6if;
1834 
1835 	callout_reset(&nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
1836 	    nd6_slowtimo, NULL);
1837 	for (i = 1; i < if_index + 1; i++) {
1838 		if (!nd_ifinfo || i >= nd_ifinfo_indexlim)
1839 			continue;
1840 		nd6if = &nd_ifinfo[i];
1841 		if (nd6if->basereachable && /* already initialized */
1842 		    (nd6if->recalctm -= ND6_SLOWTIMER_INTERVAL) <= 0) {
1843 			/*
1844 			 * Since reachable time rarely changes by router
1845 			 * advertisements, we SHOULD insure that a new random
1846 			 * value gets recomputed at least once every few hours.
1847 			 * (RFC 2461, 6.3.4)
1848 			 */
1849 			nd6if->recalctm = nd6_recalc_reachtm_interval;
1850 			nd6if->reachable = ND_COMPUTE_RTIME(nd6if->basereachable);
1851 		}
1852 	}
1853 	splx(s);
1854 }
1855 
1856 #define senderr(e) { error = (e); goto bad;}
1857 int
1858 nd6_output(ifp, origifp, m0, dst, rt0)
1859 	struct ifnet *ifp;
1860 	struct ifnet *origifp;
1861 	struct mbuf *m0;
1862 	struct sockaddr_in6 *dst;
1863 	struct rtentry *rt0;
1864 {
1865 	struct mbuf *m = m0;
1866 	struct rtentry *rt = rt0;
1867 	struct sockaddr_in6 *gw6 = NULL;
1868 	struct llinfo_nd6 *ln = NULL;
1869 	int error = 0;
1870 
1871 	if (IN6_IS_ADDR_MULTICAST(&dst->sin6_addr))
1872 		goto sendpkt;
1873 
1874 	if (nd6_need_cache(ifp) == 0)
1875 		goto sendpkt;
1876 
1877 	/*
1878 	 * next hop determination.  This routine is derived from ether_outpout.
1879 	 */
1880 	if (rt) {
1881 		if ((rt->rt_flags & RTF_UP) == 0) {
1882 			if ((rt0 = rt = rtalloc1((struct sockaddr *)dst, 1, 0UL)) !=
1883 				NULL)
1884 			{
1885 				rt->rt_refcnt--;
1886 				if (rt->rt_ifp != ifp) {
1887 					/* XXX: loop care? */
1888 					return nd6_output(ifp, origifp, m0,
1889 							  dst, rt);
1890 				}
1891 			} else
1892 				senderr(EHOSTUNREACH);
1893 		}
1894 
1895 		if (rt->rt_flags & RTF_GATEWAY) {
1896 			gw6 = (struct sockaddr_in6 *)rt->rt_gateway;
1897 
1898 			/*
1899 			 * We skip link-layer address resolution and NUD
1900 			 * if the gateway is not a neighbor from ND point
1901 			 * of view, regardless of the value of nd_ifinfo.flags.
1902 			 * The second condition is a bit tricky; we skip
1903 			 * if the gateway is our own address, which is
1904 			 * sometimes used to install a route to a p2p link.
1905 			 */
1906 			if (!nd6_is_addr_neighbor(gw6, ifp) ||
1907 			    in6ifa_ifpwithaddr(ifp, &gw6->sin6_addr)) {
1908 				/*
1909 				 * We allow this kind of tricky route only
1910 				 * when the outgoing interface is p2p.
1911 				 * XXX: we may need a more generic rule here.
1912 				 */
1913 				if ((ifp->if_flags & IFF_POINTOPOINT) == 0)
1914 					senderr(EHOSTUNREACH);
1915 
1916 				goto sendpkt;
1917 			}
1918 
1919 			if (rt->rt_gwroute == 0)
1920 				goto lookup;
1921 			if (((rt = rt->rt_gwroute)->rt_flags & RTF_UP) == 0) {
1922 				rtfree(rt); rt = rt0;
1923 			lookup: rt->rt_gwroute = rtalloc1(rt->rt_gateway, 1, 0UL);
1924 				if ((rt = rt->rt_gwroute) == 0)
1925 					senderr(EHOSTUNREACH);
1926 			}
1927 		}
1928 	}
1929 
1930 	/*
1931 	 * Address resolution or Neighbor Unreachability Detection
1932 	 * for the next hop.
1933 	 * At this point, the destination of the packet must be a unicast
1934 	 * or an anycast address(i.e. not a multicast).
1935 	 */
1936 
1937 	/* Look up the neighbor cache for the nexthop */
1938 	if (rt && (rt->rt_flags & RTF_LLINFO) != 0)
1939 		ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1940 	else {
1941 		/*
1942 		 * Since nd6_is_addr_neighbor() internally calls nd6_lookup(),
1943 		 * the condition below is not very efficient.  But we believe
1944 		 * it is tolerable, because this should be a rare case.
1945 		 */
1946 		if (nd6_is_addr_neighbor(dst, ifp) &&
1947 		    (rt = nd6_lookup(&dst->sin6_addr, 1, ifp)) != NULL)
1948 			ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1949 	}
1950 	if (!ln || !rt) {
1951 		if ((ifp->if_flags & IFF_POINTOPOINT) == 0 &&
1952 		    !(nd_ifinfo[ifp->if_index].flags & ND6_IFF_PERFORMNUD)) {
1953 			log(LOG_DEBUG,
1954 			    "nd6_output: can't allocate llinfo for %s "
1955 			    "(ln=%p, rt=%p)\n",
1956 			    ip6_sprintf(&dst->sin6_addr), ln, rt);
1957 			senderr(EIO);	/* XXX: good error? */
1958 		}
1959 
1960 		goto sendpkt;	/* send anyway */
1961 	}
1962 
1963 	/* We don't have to do link-layer address resolution on a p2p link. */
1964 	if ((ifp->if_flags & IFF_POINTOPOINT) != 0 &&
1965 	    ln->ln_state < ND6_LLINFO_REACHABLE) {
1966 		ln->ln_state = ND6_LLINFO_STALE;
1967 		ln->ln_expire = time_second + nd6_gctimer;
1968 	}
1969 
1970 	/*
1971 	 * The first time we send a packet to a neighbor whose entry is
1972 	 * STALE, we have to change the state to DELAY and a sets a timer to
1973 	 * expire in DELAY_FIRST_PROBE_TIME seconds to ensure do
1974 	 * neighbor unreachability detection on expiration.
1975 	 * (RFC 2461 7.3.3)
1976 	 */
1977 	if (ln->ln_state == ND6_LLINFO_STALE) {
1978 		ln->ln_asked = 0;
1979 		ln->ln_state = ND6_LLINFO_DELAY;
1980 		ln->ln_expire = time_second + nd6_delay;
1981 	}
1982 
1983 	/*
1984 	 * If the neighbor cache entry has a state other than INCOMPLETE
1985 	 * (i.e. its link-layer address is already resolved), just
1986 	 * send the packet.
1987 	 */
1988 	if (ln->ln_state > ND6_LLINFO_INCOMPLETE)
1989 		goto sendpkt;
1990 
1991 	/*
1992 	 * There is a neighbor cache entry, but no ethernet address
1993 	 * response yet.  Replace the held mbuf (if any) with this
1994 	 * latest one.
1995 	 *
1996 	 * This code conforms to the rate-limiting rule described in Section
1997 	 * 7.2.2 of RFC 2461, because the timer is set correctly after sending
1998 	 * an NS below.
1999 	 */
2000 	if (ln->ln_state == ND6_LLINFO_NOSTATE)
2001 		ln->ln_state = ND6_LLINFO_INCOMPLETE;
2002 	if (ln->ln_hold)
2003 		m_freem(ln->ln_hold);
2004 	ln->ln_hold = m;
2005 	if (ln->ln_expire) {
2006 		if (ln->ln_asked < nd6_mmaxtries &&
2007 		    ln->ln_expire < time_second) {
2008 			ln->ln_asked++;
2009 			ln->ln_expire = time_second +
2010 				nd_ifinfo[ifp->if_index].retrans / 1000;
2011 			nd6_ns_output(ifp, NULL, &dst->sin6_addr, ln, 0);
2012 		}
2013 	}
2014 	return(0);
2015 
2016   sendpkt:
2017 
2018 	if ((ifp->if_flags & IFF_LOOPBACK) != 0) {
2019 		return((*ifp->if_output)(origifp, m, (struct sockaddr *)dst,
2020 					 rt));
2021 	}
2022 	return((*ifp->if_output)(ifp, m, (struct sockaddr *)dst, rt));
2023 
2024   bad:
2025 	if (m)
2026 		m_freem(m);
2027 	return (error);
2028 }
2029 #undef senderr
2030 
2031 int
2032 nd6_need_cache(ifp)
2033 	struct ifnet *ifp;
2034 {
2035 	/*
2036 	 * XXX: we currently do not make neighbor cache on any interface
2037 	 * other than ARCnet, Ethernet, FDDI and GIF.
2038 	 *
2039 	 * RFC2893 says:
2040 	 * - unidirectional tunnels needs no ND
2041 	 */
2042 	switch (ifp->if_type) {
2043 	case IFT_ARCNET:
2044 	case IFT_ETHER:
2045 	case IFT_FDDI:
2046 	case IFT_IEEE1394:
2047 #ifdef IFT_L2VLAN
2048 	case IFT_L2VLAN:
2049 #endif
2050 #ifdef IFT_IEEE80211
2051 	case IFT_IEEE80211:
2052 #endif
2053 	case IFT_GIF:		/* XXX need more cases? */
2054 		return(1);
2055 	default:
2056 		return(0);
2057 	}
2058 }
2059 
2060 int
2061 nd6_storelladdr(ifp, rt, m, dst, desten)
2062 	struct ifnet *ifp;
2063 	struct rtentry *rt;
2064 	struct mbuf *m;
2065 	struct sockaddr *dst;
2066 	u_char *desten;
2067 {
2068 	int i;
2069 	struct sockaddr_dl *sdl;
2070 
2071 	if (m->m_flags & M_MCAST) {
2072 		switch (ifp->if_type) {
2073 		case IFT_ETHER:
2074 		case IFT_FDDI:
2075 #ifdef IFT_L2VLAN
2076 	case IFT_L2VLAN:
2077 #endif
2078 #ifdef IFT_IEEE80211
2079 		case IFT_IEEE80211:
2080 #endif
2081 			ETHER_MAP_IPV6_MULTICAST(&SIN6(dst)->sin6_addr,
2082 						 desten);
2083 			return(1);
2084 		case IFT_IEEE1394:
2085 			/*
2086 			 * netbsd can use if_broadcastaddr, but we don't do so
2087 			 * to reduce # of ifdef.
2088 			 */
2089 			for (i = 0; i < ifp->if_addrlen; i++)
2090 				desten[i] = ~0;
2091 			return(1);
2092 		case IFT_ARCNET:
2093 			*desten = 0;
2094 			return(1);
2095 		default:
2096 			m_freem(m);
2097 			return(0);
2098 		}
2099 	}
2100 
2101 	if (rt == NULL) {
2102 		/* this could happen, if we could not allocate memory */
2103 		m_freem(m);
2104 		return(0);
2105 	}
2106 	if (rt->rt_gateway->sa_family != AF_LINK) {
2107 		printf("nd6_storelladdr: something odd happens\n");
2108 		m_freem(m);
2109 		return(0);
2110 	}
2111 	sdl = SDL(rt->rt_gateway);
2112 	if (sdl->sdl_alen == 0) {
2113 		/* this should be impossible, but we bark here for debugging */
2114 		printf("nd6_storelladdr: sdl_alen == 0\n");
2115 		m_freem(m);
2116 		return(0);
2117 	}
2118 
2119 	bcopy(LLADDR(sdl), desten, sdl->sdl_alen);
2120 	return(1);
2121 }
2122 
2123 static int nd6_sysctl_drlist(SYSCTL_HANDLER_ARGS);
2124 static int nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS);
2125 #ifdef SYSCTL_DECL
2126 SYSCTL_DECL(_net_inet6_icmp6);
2127 #endif
2128 SYSCTL_NODE(_net_inet6_icmp6, ICMPV6CTL_ND6_DRLIST, nd6_drlist,
2129 	CTLFLAG_RD, nd6_sysctl_drlist, "");
2130 SYSCTL_NODE(_net_inet6_icmp6, ICMPV6CTL_ND6_PRLIST, nd6_prlist,
2131 	CTLFLAG_RD, nd6_sysctl_prlist, "");
2132 
2133 static int
2134 nd6_sysctl_drlist(SYSCTL_HANDLER_ARGS)
2135 {
2136 	int error;
2137 	char buf[1024];
2138 	struct in6_defrouter *d, *de;
2139 	struct nd_defrouter *dr;
2140 
2141 	if (req->newptr)
2142 		return EPERM;
2143 	error = 0;
2144 
2145 	for (dr = TAILQ_FIRST(&nd_defrouter);
2146 	     dr;
2147 	     dr = TAILQ_NEXT(dr, dr_entry)) {
2148 		d = (struct in6_defrouter *)buf;
2149 		de = (struct in6_defrouter *)(buf + sizeof(buf));
2150 
2151 		if (d + 1 <= de) {
2152 			bzero(d, sizeof(*d));
2153 			d->rtaddr.sin6_family = AF_INET6;
2154 			d->rtaddr.sin6_len = sizeof(d->rtaddr);
2155 			if (in6_recoverscope(&d->rtaddr, &dr->rtaddr,
2156 			    dr->ifp) != 0)
2157 				log(LOG_ERR,
2158 				    "scope error in "
2159 				    "default router list (%s)\n",
2160 				    ip6_sprintf(&dr->rtaddr));
2161 			d->flags = dr->flags;
2162 			d->rtlifetime = dr->rtlifetime;
2163 			d->expire = dr->expire;
2164 			d->if_index = dr->ifp->if_index;
2165 		} else
2166 			panic("buffer too short");
2167 
2168 		error = SYSCTL_OUT(req, buf, sizeof(*d));
2169 		if (error)
2170 			break;
2171 	}
2172 	return error;
2173 }
2174 
2175 static int
2176 nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS)
2177 {
2178 	int error;
2179 	char buf[1024];
2180 	struct in6_prefix *p, *pe;
2181 	struct nd_prefix *pr;
2182 
2183 	if (req->newptr)
2184 		return EPERM;
2185 	error = 0;
2186 
2187 	for (pr = nd_prefix.lh_first; pr; pr = pr->ndpr_next) {
2188 		u_short advrtrs;
2189 		size_t advance;
2190 		struct sockaddr_in6 *sin6, *s6;
2191 		struct nd_pfxrouter *pfr;
2192 
2193 		p = (struct in6_prefix *)buf;
2194 		pe = (struct in6_prefix *)(buf + sizeof(buf));
2195 
2196 		if (p + 1 <= pe) {
2197 			bzero(p, sizeof(*p));
2198 			sin6 = (struct sockaddr_in6 *)(p + 1);
2199 
2200 			p->prefix = pr->ndpr_prefix;
2201 			if (in6_recoverscope(&p->prefix,
2202 			    &p->prefix.sin6_addr, pr->ndpr_ifp) != 0)
2203 				log(LOG_ERR,
2204 				    "scope error in prefix list (%s)\n",
2205 				    ip6_sprintf(&p->prefix.sin6_addr));
2206 			p->raflags = pr->ndpr_raf;
2207 			p->prefixlen = pr->ndpr_plen;
2208 			p->vltime = pr->ndpr_vltime;
2209 			p->pltime = pr->ndpr_pltime;
2210 			p->if_index = pr->ndpr_ifp->if_index;
2211 			p->expire = pr->ndpr_expire;
2212 			p->refcnt = pr->ndpr_refcnt;
2213 			p->flags = pr->ndpr_stateflags;
2214 			p->origin = PR_ORIG_RA;
2215 			advrtrs = 0;
2216 			for (pfr = pr->ndpr_advrtrs.lh_first;
2217 			     pfr;
2218 			     pfr = pfr->pfr_next) {
2219 				if ((void *)&sin6[advrtrs + 1] >
2220 				    (void *)pe) {
2221 					advrtrs++;
2222 					continue;
2223 				}
2224 				s6 = &sin6[advrtrs];
2225 				bzero(s6, sizeof(*s6));
2226 				s6->sin6_family = AF_INET6;
2227 				s6->sin6_len = sizeof(*sin6);
2228 				if (in6_recoverscope(s6,
2229 				    &pfr->router->rtaddr,
2230 				    pfr->router->ifp) != 0)
2231 					log(LOG_ERR,
2232 					    "scope error in "
2233 					    "prefix list (%s)\n",
2234 					    ip6_sprintf(&pfr->router->rtaddr));
2235 				advrtrs++;
2236 			}
2237 			p->advrtrs = advrtrs;
2238 		} else
2239 			panic("buffer too short");
2240 
2241 		advance = sizeof(*p) + sizeof(*sin6) * advrtrs;
2242 		error = SYSCTL_OUT(req, buf, advance);
2243 		if (error)
2244 			break;
2245 	}
2246 	return error;
2247 }
2248