xref: /freebsd/sys/netinet6/nd6.c (revision 6780ab54325a71e7e70112b11657973edde8655e)
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 #include "opt_mac.h"
44 
45 #include <sys/param.h>
46 #include <sys/systm.h>
47 #include <sys/callout.h>
48 #include <sys/mac.h>
49 #include <sys/malloc.h>
50 #include <sys/mbuf.h>
51 #include <sys/socket.h>
52 #include <sys/sockio.h>
53 #include <sys/time.h>
54 #include <sys/kernel.h>
55 #include <sys/protosw.h>
56 #include <sys/errno.h>
57 #include <sys/syslog.h>
58 #include <sys/queue.h>
59 #include <sys/sysctl.h>
60 
61 #include <net/if.h>
62 #include <net/if_dl.h>
63 #include <net/if_types.h>
64 #include <net/if_atm.h>
65 #include <net/fddi.h>
66 #include <net/route.h>
67 
68 #include <netinet/in.h>
69 #include <netinet/if_ether.h>
70 #include <netinet6/in6_var.h>
71 #include <netinet/ip6.h>
72 #include <netinet6/ip6_var.h>
73 #include <netinet6/nd6.h>
74 #include <netinet6/in6_prefix.h>
75 #include <netinet/icmp6.h>
76 
77 #include <net/net_osdep.h>
78 
79 #define ND6_SLOWTIMER_INTERVAL (60 * 60) /* 1 hour */
80 #define ND6_RECALC_REACHTM_INTERVAL (60 * 120) /* 2 hours */
81 
82 #define SIN6(s) ((struct sockaddr_in6 *)s)
83 #define SDL(s) ((struct sockaddr_dl *)s)
84 
85 /* timer values */
86 int	nd6_prune	= 1;	/* walk list every 1 seconds */
87 int	nd6_delay	= 5;	/* delay first probe time 5 second */
88 int	nd6_umaxtries	= 3;	/* maximum unicast query */
89 int	nd6_mmaxtries	= 3;	/* maximum multicast query */
90 int	nd6_useloopback = 1;	/* use loopback interface for local traffic */
91 int	nd6_gctimer	= (60 * 60 * 24); /* 1 day: garbage collection timer */
92 
93 /* preventing too many loops in ND option parsing */
94 int nd6_maxndopt = 10;	/* max # of ND options allowed */
95 
96 int nd6_maxnudhint = 0;	/* max # of subsequent upper layer hints */
97 
98 #ifdef ND6_DEBUG
99 int nd6_debug = 1;
100 #else
101 int nd6_debug = 0;
102 #endif
103 
104 /* for debugging? */
105 static int nd6_inuse, nd6_allocated;
106 
107 struct llinfo_nd6 llinfo_nd6 = {&llinfo_nd6, &llinfo_nd6};
108 static size_t nd_ifinfo_indexlim = 8;
109 struct nd_ifinfo *nd_ifinfo = NULL;
110 struct nd_drhead nd_defrouter;
111 struct nd_prhead nd_prefix = { 0 };
112 
113 int nd6_recalc_reachtm_interval = ND6_RECALC_REACHTM_INTERVAL;
114 static struct sockaddr_in6 all1_sa;
115 
116 static void nd6_slowtimo __P((void *));
117 static int regen_tmpaddr __P((struct in6_ifaddr *));
118 
119 struct callout nd6_slowtimo_ch;
120 struct callout nd6_timer_ch;
121 extern struct callout in6_tmpaddrtimer_ch;
122 
123 void
124 nd6_init()
125 {
126 	static int nd6_init_done = 0;
127 	int i;
128 
129 	if (nd6_init_done) {
130 		log(LOG_NOTICE, "nd6_init called more than once(ignored)\n");
131 		return;
132 	}
133 
134 	all1_sa.sin6_family = AF_INET6;
135 	all1_sa.sin6_len = sizeof(struct sockaddr_in6);
136 	for (i = 0; i < sizeof(all1_sa.sin6_addr); i++)
137 		all1_sa.sin6_addr.s6_addr[i] = 0xff;
138 
139 	/* initialization of the default router list */
140 	TAILQ_INIT(&nd_defrouter);
141 
142 	nd6_init_done = 1;
143 
144 	/* start timer */
145 	callout_reset(&nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
146 	    nd6_slowtimo, NULL);
147 }
148 
149 void
150 nd6_ifattach(ifp)
151 	struct ifnet *ifp;
152 {
153 
154 	/*
155 	 * We have some arrays that should be indexed by if_index.
156 	 * since if_index will grow dynamically, they should grow too.
157 	 */
158 	if (nd_ifinfo == NULL || if_index >= nd_ifinfo_indexlim) {
159 		size_t n;
160 		caddr_t q;
161 
162 		while (if_index >= nd_ifinfo_indexlim)
163 			nd_ifinfo_indexlim <<= 1;
164 
165 		/* grow nd_ifinfo */
166 		n = nd_ifinfo_indexlim * sizeof(struct nd_ifinfo);
167 		q = (caddr_t)malloc(n, M_IP6NDP, 0);
168 		bzero(q, n);
169 		if (nd_ifinfo) {
170 			bcopy((caddr_t)nd_ifinfo, q, n/2);
171 			free((caddr_t)nd_ifinfo, M_IP6NDP);
172 		}
173 		nd_ifinfo = (struct nd_ifinfo *)q;
174 	}
175 
176 #define ND nd_ifinfo[ifp->if_index]
177 
178 	/*
179 	 * Don't initialize if called twice.
180 	 * XXX: to detect this, we should choose a member that is never set
181 	 * before initialization of the ND structure itself.  We formaly used
182 	 * the linkmtu member, which was not suitable because it could be
183 	 * initialized via "ifconfig mtu".
184 	 */
185 	if (ND.basereachable)
186 		return;
187 
188 	ND.linkmtu = ifnet_byindex(ifp->if_index)->if_mtu;
189 	ND.chlim = IPV6_DEFHLIM;
190 	ND.basereachable = REACHABLE_TIME;
191 	ND.reachable = ND_COMPUTE_RTIME(ND.basereachable);
192 	ND.retrans = RETRANS_TIMER;
193 	ND.receivedra = 0;
194 	ND.flags = ND6_IFF_PERFORMNUD;
195 	nd6_setmtu(ifp);
196 #undef ND
197 }
198 
199 /*
200  * Reset ND level link MTU. This function is called when the physical MTU
201  * changes, which means we might have to adjust the ND level MTU.
202  */
203 void
204 nd6_setmtu(ifp)
205 	struct ifnet *ifp;
206 {
207 	struct nd_ifinfo *ndi = &nd_ifinfo[ifp->if_index];
208 	u_long oldmaxmtu = ndi->maxmtu;
209 	u_long oldlinkmtu = ndi->linkmtu;
210 
211 	switch (ifp->if_type) {
212 	case IFT_ARCNET:	/* XXX MTU handling needs more work */
213 		ndi->maxmtu = MIN(60480, ifp->if_mtu);
214 		break;
215 	case IFT_ETHER:
216 		ndi->maxmtu = MIN(ETHERMTU, ifp->if_mtu);
217 		break;
218 	case IFT_FDDI:
219 		ndi->maxmtu = MIN(FDDIIPMTU, ifp->if_mtu);
220 		break;
221 	case IFT_ATM:
222 		ndi->maxmtu = MIN(ATMMTU, ifp->if_mtu);
223 		break;
224 	case IFT_IEEE1394:	/* XXX should be IEEE1394MTU(1500) */
225 		ndi->maxmtu = MIN(ETHERMTU, ifp->if_mtu);
226 		break;
227 #ifdef IFT_IEEE80211
228 	case IFT_IEEE80211:	/* XXX should be IEEE80211MTU(1500) */
229 		ndi->maxmtu = MIN(ETHERMTU, ifp->if_mtu);
230 		break;
231 #endif
232 	default:
233 		ndi->maxmtu = ifp->if_mtu;
234 		break;
235 	}
236 
237 	if (oldmaxmtu != ndi->maxmtu) {
238 		/*
239 		 * If the ND level MTU is not set yet, or if the maxmtu
240 		 * is reset to a smaller value than the ND level MTU,
241 		 * also reset the ND level MTU.
242 		 */
243 		if (ndi->linkmtu == 0 ||
244 		    ndi->maxmtu < ndi->linkmtu) {
245 			ndi->linkmtu = ndi->maxmtu;
246 			/* also adjust in6_maxmtu if necessary. */
247 			if (oldlinkmtu == 0) {
248 				/*
249 				 * XXX: the case analysis is grotty, but
250 				 * it is not efficient to call in6_setmaxmtu()
251 				 * here when we are during the initialization
252 				 * procedure.
253 				 */
254 				if (in6_maxmtu < ndi->linkmtu)
255 					in6_maxmtu = ndi->linkmtu;
256 			} else
257 				in6_setmaxmtu();
258 		}
259 	}
260 #undef MIN
261 }
262 
263 void
264 nd6_option_init(opt, icmp6len, ndopts)
265 	void *opt;
266 	int icmp6len;
267 	union nd_opts *ndopts;
268 {
269 	bzero(ndopts, sizeof(*ndopts));
270 	ndopts->nd_opts_search = (struct nd_opt_hdr *)opt;
271 	ndopts->nd_opts_last
272 		= (struct nd_opt_hdr *)(((u_char *)opt) + icmp6len);
273 
274 	if (icmp6len == 0) {
275 		ndopts->nd_opts_done = 1;
276 		ndopts->nd_opts_search = NULL;
277 	}
278 }
279 
280 /*
281  * Take one ND option.
282  */
283 struct nd_opt_hdr *
284 nd6_option(ndopts)
285 	union nd_opts *ndopts;
286 {
287 	struct nd_opt_hdr *nd_opt;
288 	int olen;
289 
290 	if (!ndopts)
291 		panic("ndopts == NULL in nd6_option\n");
292 	if (!ndopts->nd_opts_last)
293 		panic("uninitialized ndopts in nd6_option\n");
294 	if (!ndopts->nd_opts_search)
295 		return NULL;
296 	if (ndopts->nd_opts_done)
297 		return NULL;
298 
299 	nd_opt = ndopts->nd_opts_search;
300 
301 	/* make sure nd_opt_len is inside the buffer */
302 	if ((caddr_t)&nd_opt->nd_opt_len >= (caddr_t)ndopts->nd_opts_last) {
303 		bzero(ndopts, sizeof(*ndopts));
304 		return NULL;
305 	}
306 
307 	olen = nd_opt->nd_opt_len << 3;
308 	if (olen == 0) {
309 		/*
310 		 * Message validation requires that all included
311 		 * options have a length that is greater than zero.
312 		 */
313 		bzero(ndopts, sizeof(*ndopts));
314 		return NULL;
315 	}
316 
317 	ndopts->nd_opts_search = (struct nd_opt_hdr *)((caddr_t)nd_opt + olen);
318 	if (ndopts->nd_opts_search > ndopts->nd_opts_last) {
319 		/* option overruns the end of buffer, invalid */
320 		bzero(ndopts, sizeof(*ndopts));
321 		return NULL;
322 	} else if (ndopts->nd_opts_search == ndopts->nd_opts_last) {
323 		/* reached the end of options chain */
324 		ndopts->nd_opts_done = 1;
325 		ndopts->nd_opts_search = NULL;
326 	}
327 	return nd_opt;
328 }
329 
330 /*
331  * Parse multiple ND options.
332  * This function is much easier to use, for ND routines that do not need
333  * multiple options of the same type.
334  */
335 int
336 nd6_options(ndopts)
337 	union nd_opts *ndopts;
338 {
339 	struct nd_opt_hdr *nd_opt;
340 	int i = 0;
341 
342 	if (!ndopts)
343 		panic("ndopts == NULL in nd6_options\n");
344 	if (!ndopts->nd_opts_last)
345 		panic("uninitialized ndopts in nd6_options\n");
346 	if (!ndopts->nd_opts_search)
347 		return 0;
348 
349 	while (1) {
350 		nd_opt = nd6_option(ndopts);
351 		if (!nd_opt && !ndopts->nd_opts_last) {
352 			/*
353 			 * Message validation requires that all included
354 			 * options have a length that is greater than zero.
355 			 */
356 			icmp6stat.icp6s_nd_badopt++;
357 			bzero(ndopts, sizeof(*ndopts));
358 			return -1;
359 		}
360 
361 		if (!nd_opt)
362 			goto skip1;
363 
364 		switch (nd_opt->nd_opt_type) {
365 		case ND_OPT_SOURCE_LINKADDR:
366 		case ND_OPT_TARGET_LINKADDR:
367 		case ND_OPT_MTU:
368 		case ND_OPT_REDIRECTED_HEADER:
369 			if (ndopts->nd_opt_array[nd_opt->nd_opt_type]) {
370 				nd6log((LOG_INFO,
371 				    "duplicated ND6 option found (type=%d)\n",
372 				    nd_opt->nd_opt_type));
373 				/* XXX bark? */
374 			} else {
375 				ndopts->nd_opt_array[nd_opt->nd_opt_type]
376 					= nd_opt;
377 			}
378 			break;
379 		case ND_OPT_PREFIX_INFORMATION:
380 			if (ndopts->nd_opt_array[nd_opt->nd_opt_type] == 0) {
381 				ndopts->nd_opt_array[nd_opt->nd_opt_type]
382 					= nd_opt;
383 			}
384 			ndopts->nd_opts_pi_end =
385 				(struct nd_opt_prefix_info *)nd_opt;
386 			break;
387 		default:
388 			/*
389 			 * Unknown options must be silently ignored,
390 			 * to accomodate future extension to the protocol.
391 			 */
392 			nd6log((LOG_DEBUG,
393 			    "nd6_options: unsupported option %d - "
394 			    "option ignored\n", nd_opt->nd_opt_type));
395 		}
396 
397 skip1:
398 		i++;
399 		if (i > nd6_maxndopt) {
400 			icmp6stat.icp6s_nd_toomanyopt++;
401 			nd6log((LOG_INFO, "too many loop in nd opt\n"));
402 			break;
403 		}
404 
405 		if (ndopts->nd_opts_done)
406 			break;
407 	}
408 
409 	return 0;
410 }
411 
412 /*
413  * ND6 timer routine to expire default route list and prefix list
414  */
415 void
416 nd6_timer(ignored_arg)
417 	void	*ignored_arg;
418 {
419 	int s;
420 	struct llinfo_nd6 *ln;
421 	struct nd_defrouter *dr;
422 	struct nd_prefix *pr;
423 	struct ifnet *ifp;
424 	struct in6_ifaddr *ia6, *nia6;
425 	struct in6_addrlifetime *lt6;
426 
427 	s = splnet();
428 	callout_reset(&nd6_timer_ch, nd6_prune * hz,
429 		      nd6_timer, NULL);
430 
431 	ln = llinfo_nd6.ln_next;
432 	while (ln && ln != &llinfo_nd6) {
433 		struct rtentry *rt;
434 		struct sockaddr_in6 *dst;
435 		struct llinfo_nd6 *next = ln->ln_next;
436 		/* XXX: used for the DELAY case only: */
437 		struct nd_ifinfo *ndi = NULL;
438 
439 		if ((rt = ln->ln_rt) == NULL) {
440 			ln = next;
441 			continue;
442 		}
443 		if ((ifp = rt->rt_ifp) == NULL) {
444 			ln = next;
445 			continue;
446 		}
447 		ndi = &nd_ifinfo[ifp->if_index];
448 		dst = (struct sockaddr_in6 *)rt_key(rt);
449 
450 		if (ln->ln_expire > time_second) {
451 			ln = next;
452 			continue;
453 		}
454 
455 		/* sanity check */
456 		if (!rt)
457 			panic("rt=0 in nd6_timer(ln=%p)\n", ln);
458 		if (rt->rt_llinfo && (struct llinfo_nd6 *)rt->rt_llinfo != ln)
459 			panic("rt_llinfo(%p) is not equal to ln(%p)\n",
460 			      rt->rt_llinfo, ln);
461 		if (!dst)
462 			panic("dst=0 in nd6_timer(ln=%p)\n", ln);
463 
464 		switch (ln->ln_state) {
465 		case ND6_LLINFO_INCOMPLETE:
466 			if (ln->ln_asked < nd6_mmaxtries) {
467 				ln->ln_asked++;
468 				ln->ln_expire = time_second +
469 					nd_ifinfo[ifp->if_index].retrans / 1000;
470 				nd6_ns_output(ifp, NULL, &dst->sin6_addr,
471 					ln, 0);
472 			} else {
473 				struct mbuf *m = ln->ln_hold;
474 				if (m) {
475 					if (rt->rt_ifp) {
476 						/*
477 						 * Fake rcvif to make ICMP error
478 						 * more helpful in diagnosing
479 						 * for the receiver.
480 						 * XXX: should we consider
481 						 * older rcvif?
482 						 */
483 						m->m_pkthdr.rcvif = rt->rt_ifp;
484 					}
485 					icmp6_error(m, ICMP6_DST_UNREACH,
486 						    ICMP6_DST_UNREACH_ADDR, 0);
487 					ln->ln_hold = NULL;
488 				}
489 				next = nd6_free(rt);
490 			}
491 			break;
492 		case ND6_LLINFO_REACHABLE:
493 			if (ln->ln_expire) {
494 				ln->ln_state = ND6_LLINFO_STALE;
495 				ln->ln_expire = time_second + nd6_gctimer;
496 			}
497 			break;
498 
499 		case ND6_LLINFO_STALE:
500 			/* Garbage Collection(RFC 2461 5.3) */
501 			if (ln->ln_expire)
502 				next = nd6_free(rt);
503 			break;
504 
505 		case ND6_LLINFO_DELAY:
506 			if (ndi && (ndi->flags & ND6_IFF_PERFORMNUD) != 0) {
507 				/* We need NUD */
508 				ln->ln_asked = 1;
509 				ln->ln_state = ND6_LLINFO_PROBE;
510 				ln->ln_expire = time_second +
511 					ndi->retrans / 1000;
512 				nd6_ns_output(ifp, &dst->sin6_addr,
513 					      &dst->sin6_addr,
514 					      ln, 0);
515 			} else {
516 				ln->ln_state = ND6_LLINFO_STALE; /* XXX */
517 				ln->ln_expire = time_second + nd6_gctimer;
518 			}
519 			break;
520 		case ND6_LLINFO_PROBE:
521 			if (ln->ln_asked < nd6_umaxtries) {
522 				ln->ln_asked++;
523 				ln->ln_expire = time_second +
524 					nd_ifinfo[ifp->if_index].retrans / 1000;
525 				nd6_ns_output(ifp, &dst->sin6_addr,
526 					       &dst->sin6_addr, ln, 0);
527 			} else {
528 				next = nd6_free(rt);
529 			}
530 			break;
531 		}
532 		ln = next;
533 	}
534 
535 	/* expire default router list */
536 	dr = TAILQ_FIRST(&nd_defrouter);
537 	while (dr) {
538 		if (dr->expire && dr->expire < time_second) {
539 			struct nd_defrouter *t;
540 			t = TAILQ_NEXT(dr, dr_entry);
541 			defrtrlist_del(dr);
542 			dr = t;
543 		} else {
544 			dr = TAILQ_NEXT(dr, dr_entry);
545 		}
546 	}
547 
548 	/*
549 	 * expire interface addresses.
550 	 * in the past the loop was inside prefix expiry processing.
551 	 * However, from a stricter speci-confrmance standpoint, we should
552 	 * rather separate address lifetimes and prefix lifetimes.
553 	 */
554   addrloop:
555 	for (ia6 = in6_ifaddr; ia6; ia6 = nia6) {
556 		nia6 = ia6->ia_next;
557 		/* check address lifetime */
558 		lt6 = &ia6->ia6_lifetime;
559 		if (IFA6_IS_INVALID(ia6)) {
560 			int regen = 0;
561 
562 			/*
563 			 * If the expiring address is temporary, try
564 			 * regenerating a new one.  This would be useful when
565 			 * we suspended a laptop PC, then turned it on after a
566 			 * period that could invalidate all temporary
567 			 * addresses.  Although we may have to restart the
568 			 * loop (see below), it must be after purging the
569 			 * address.  Otherwise, we'd see an infinite loop of
570 			 * regeneration.
571 			 */
572 			if (ip6_use_tempaddr &&
573 			    (ia6->ia6_flags & IN6_IFF_TEMPORARY) != 0) {
574 				if (regen_tmpaddr(ia6) == 0)
575 					regen = 1;
576 			}
577 
578 			in6_purgeaddr(&ia6->ia_ifa);
579 
580 			if (regen)
581 				goto addrloop; /* XXX: see below */
582 		}
583 		if (IFA6_IS_DEPRECATED(ia6)) {
584 			int oldflags = ia6->ia6_flags;
585 
586 			ia6->ia6_flags |= IN6_IFF_DEPRECATED;
587 
588 			/*
589 			 * If a temporary address has just become deprecated,
590 			 * regenerate a new one if possible.
591 			 */
592 			if (ip6_use_tempaddr &&
593 			    (ia6->ia6_flags & IN6_IFF_TEMPORARY) != 0 &&
594 			    (oldflags & IN6_IFF_DEPRECATED) == 0) {
595 
596 				if (regen_tmpaddr(ia6) == 0) {
597 					/*
598 					 * A new temporary address is
599 					 * generated.
600 					 * XXX: this means the address chain
601 					 * has changed while we are still in
602 					 * the loop.  Although the change
603 					 * would not cause disaster (because
604 					 * it's not a deletion, but an
605 					 * addition,) we'd rather restart the
606 					 * loop just for safety.  Or does this
607 					 * significantly reduce performance??
608 					 */
609 					goto addrloop;
610 				}
611 			}
612 		} else {
613 			/*
614 			 * A new RA might have made a deprecated address
615 			 * preferred.
616 			 */
617 			ia6->ia6_flags &= ~IN6_IFF_DEPRECATED;
618 		}
619 	}
620 
621 	/* expire prefix list */
622 	pr = nd_prefix.lh_first;
623 	while (pr) {
624 		/*
625 		 * check prefix lifetime.
626 		 * since pltime is just for autoconf, pltime processing for
627 		 * prefix is not necessary.
628 		 */
629 		if (pr->ndpr_expire && pr->ndpr_expire < time_second) {
630 			struct nd_prefix *t;
631 			t = pr->ndpr_next;
632 
633 			/*
634 			 * address expiration and prefix expiration are
635 			 * separate.  NEVER perform in6_purgeaddr here.
636 			 */
637 
638 			prelist_remove(pr);
639 			pr = t;
640 		} else
641 			pr = pr->ndpr_next;
642 	}
643 	splx(s);
644 }
645 
646 static int
647 regen_tmpaddr(ia6)
648 	struct in6_ifaddr *ia6; /* deprecated/invalidated temporary address */
649 {
650 	struct ifaddr *ifa;
651 	struct ifnet *ifp;
652 	struct in6_ifaddr *public_ifa6 = NULL;
653 
654 	ifp = ia6->ia_ifa.ifa_ifp;
655 	for (ifa = ifp->if_addrlist.tqh_first; ifa;
656 	     ifa = ifa->ifa_list.tqe_next)
657 	{
658 		struct in6_ifaddr *it6;
659 
660 		if (ifa->ifa_addr->sa_family != AF_INET6)
661 			continue;
662 
663 		it6 = (struct in6_ifaddr *)ifa;
664 
665 		/* ignore no autoconf addresses. */
666 		if ((it6->ia6_flags & IN6_IFF_AUTOCONF) == 0)
667 			continue;
668 
669 		/* ignore autoconf addresses with different prefixes. */
670 		if (it6->ia6_ndpr == NULL || it6->ia6_ndpr != ia6->ia6_ndpr)
671 			continue;
672 
673 		/*
674 		 * Now we are looking at an autoconf address with the same
675 		 * prefix as ours.  If the address is temporary and is still
676 		 * preferred, do not create another one.  It would be rare, but
677 		 * could happen, for example, when we resume a laptop PC after
678 		 * a long period.
679 		 */
680 		if ((it6->ia6_flags & IN6_IFF_TEMPORARY) != 0 &&
681 		    !IFA6_IS_DEPRECATED(it6)) {
682 			public_ifa6 = NULL;
683 			break;
684 		}
685 
686 		/*
687 		 * This is a public autoconf address that has the same prefix
688 		 * as ours.  If it is preferred, keep it.  We can't break the
689 		 * loop here, because there may be a still-preferred temporary
690 		 * address with the prefix.
691 		 */
692 		if (!IFA6_IS_DEPRECATED(it6))
693 		    public_ifa6 = it6;
694 	}
695 
696 	if (public_ifa6 != NULL) {
697 		int e;
698 
699 		if ((e = in6_tmpifadd(public_ifa6, 0)) != 0) {
700 			log(LOG_NOTICE, "regen_tmpaddr: failed to create a new"
701 			    " tmp addr,errno=%d\n", e);
702 			return(-1);
703 		}
704 		return(0);
705 	}
706 
707 	return(-1);
708 }
709 
710 /*
711  * Nuke neighbor cache/prefix/default router management table, right before
712  * ifp goes away.
713  */
714 void
715 nd6_purge(ifp)
716 	struct ifnet *ifp;
717 {
718 	struct llinfo_nd6 *ln, *nln;
719 	struct nd_defrouter *dr, *ndr, drany;
720 	struct nd_prefix *pr, *npr;
721 
722 	/* Nuke default router list entries toward ifp */
723 	if ((dr = TAILQ_FIRST(&nd_defrouter)) != NULL) {
724 		/*
725 		 * The first entry of the list may be stored in
726 		 * the routing table, so we'll delete it later.
727 		 */
728 		for (dr = TAILQ_NEXT(dr, dr_entry); dr; dr = ndr) {
729 			ndr = TAILQ_NEXT(dr, dr_entry);
730 			if (dr->ifp == ifp)
731 				defrtrlist_del(dr);
732 		}
733 		dr = TAILQ_FIRST(&nd_defrouter);
734 		if (dr->ifp == ifp)
735 			defrtrlist_del(dr);
736 	}
737 
738 	/* Nuke prefix list entries toward ifp */
739 	for (pr = nd_prefix.lh_first; pr; pr = npr) {
740 		npr = pr->ndpr_next;
741 		if (pr->ndpr_ifp == ifp) {
742 			/*
743 			 * Previously, pr->ndpr_addr is removed as well,
744 			 * but I strongly believe we don't have to do it.
745 			 * nd6_purge() is only called from in6_ifdetach(),
746 			 * which removes all the associated interface addresses
747 			 * by itself.
748 			 * (jinmei@kame.net 20010129)
749 			 */
750 			prelist_remove(pr);
751 		}
752 	}
753 
754 	/* cancel default outgoing interface setting */
755 	if (nd6_defifindex == ifp->if_index)
756 		nd6_setdefaultiface(0);
757 
758 	if (!ip6_forwarding && ip6_accept_rtadv) { /* XXX: too restrictive? */
759 		/* refresh default router list */
760 		bzero(&drany, sizeof(drany));
761 		defrouter_delreq(&drany, 0);
762 		defrouter_select();
763 	}
764 
765 	/*
766 	 * Nuke neighbor cache entries for the ifp.
767 	 * Note that rt->rt_ifp may not be the same as ifp,
768 	 * due to KAME goto ours hack.  See RTM_RESOLVE case in
769 	 * nd6_rtrequest(), and ip6_input().
770 	 */
771 	ln = llinfo_nd6.ln_next;
772 	while (ln && ln != &llinfo_nd6) {
773 		struct rtentry *rt;
774 		struct sockaddr_dl *sdl;
775 
776 		nln = ln->ln_next;
777 		rt = ln->ln_rt;
778 		if (rt && rt->rt_gateway &&
779 		    rt->rt_gateway->sa_family == AF_LINK) {
780 			sdl = (struct sockaddr_dl *)rt->rt_gateway;
781 			if (sdl->sdl_index == ifp->if_index)
782 				nln = nd6_free(rt);
783 		}
784 		ln = nln;
785 	}
786 }
787 
788 struct rtentry *
789 nd6_lookup(addr6, create, ifp)
790 	struct in6_addr *addr6;
791 	int create;
792 	struct ifnet *ifp;
793 {
794 	struct rtentry *rt;
795 	struct sockaddr_in6 sin6;
796 
797 	bzero(&sin6, sizeof(sin6));
798 	sin6.sin6_len = sizeof(struct sockaddr_in6);
799 	sin6.sin6_family = AF_INET6;
800 	sin6.sin6_addr = *addr6;
801 #ifdef SCOPEDROUTING
802 	sin6.sin6_scope_id = in6_addr2scopeid(ifp, addr6);
803 #endif
804 	rt = rtalloc1((struct sockaddr *)&sin6, create, 0UL);
805 	if (rt && (rt->rt_flags & RTF_LLINFO) == 0) {
806 		/*
807 		 * This is the case for the default route.
808 		 * If we want to create a neighbor cache for the address, we
809 		 * should free the route for the destination and allocate an
810 		 * interface route.
811 		 */
812 		if (create) {
813 			RTFREE(rt);
814 			rt = 0;
815 		}
816 	}
817 	if (!rt) {
818 		if (create && ifp) {
819 			int e;
820 
821 			/*
822 			 * If no route is available and create is set,
823 			 * we allocate a host route for the destination
824 			 * and treat it like an interface route.
825 			 * This hack is necessary for a neighbor which can't
826 			 * be covered by our own prefix.
827 			 */
828 			struct ifaddr *ifa =
829 				ifaof_ifpforaddr((struct sockaddr *)&sin6, ifp);
830 			if (ifa == NULL)
831 				return(NULL);
832 
833 			/*
834 			 * Create a new route.  RTF_LLINFO is necessary
835 			 * to create a Neighbor Cache entry for the
836 			 * destination in nd6_rtrequest which will be
837 			 * called in rtrequest via ifa->ifa_rtrequest.
838 			 */
839 			if ((e = rtrequest(RTM_ADD, (struct sockaddr *)&sin6,
840 					   ifa->ifa_addr,
841 					   (struct sockaddr *)&all1_sa,
842 					   (ifa->ifa_flags |
843 					    RTF_HOST | RTF_LLINFO) &
844 					   ~RTF_CLONING,
845 					   &rt)) != 0)
846 				log(LOG_ERR,
847 				    "nd6_lookup: failed to add route for a "
848 				    "neighbor(%s), errno=%d\n",
849 				    ip6_sprintf(addr6), e);
850 			if (rt == NULL)
851 				return(NULL);
852 			if (rt->rt_llinfo) {
853 				struct llinfo_nd6 *ln =
854 					(struct llinfo_nd6 *)rt->rt_llinfo;
855 				ln->ln_state = ND6_LLINFO_NOSTATE;
856 			}
857 		} else
858 			return(NULL);
859 	}
860 	rt->rt_refcnt--;
861 	/*
862 	 * Validation for the entry.
863 	 * Note that the check for rt_llinfo is necessary because a cloned
864 	 * route from a parent route that has the L flag (e.g. the default
865 	 * route to a p2p interface) may have the flag, too, while the
866 	 * destination is not actually a neighbor.
867 	 * XXX: we can't use rt->rt_ifp to check for the interface, since
868 	 *      it might be the loopback interface if the entry is for our
869 	 *      own address on a non-loopback interface. Instead, we should
870 	 *      use rt->rt_ifa->ifa_ifp, which would specify the REAL
871 	 *      interface.
872 	 */
873 	if ((rt->rt_flags & RTF_GATEWAY) || (rt->rt_flags & RTF_LLINFO) == 0 ||
874 	    rt->rt_gateway->sa_family != AF_LINK || rt->rt_llinfo == NULL ||
875 	    (ifp && rt->rt_ifa->ifa_ifp != ifp)) {
876 		if (create) {
877 			log(LOG_DEBUG, "nd6_lookup: failed to lookup %s (if = %s)\n",
878 			    ip6_sprintf(addr6), ifp ? if_name(ifp) : "unspec");
879 			/* xxx more logs... kazu */
880 		}
881 		return(NULL);
882 	}
883 	return(rt);
884 }
885 
886 /*
887  * Detect if a given IPv6 address identifies a neighbor on a given link.
888  * XXX: should take care of the destination of a p2p link?
889  */
890 int
891 nd6_is_addr_neighbor(addr, ifp)
892 	struct sockaddr_in6 *addr;
893 	struct ifnet *ifp;
894 {
895 	struct ifaddr *ifa;
896 	int i;
897 
898 #define IFADDR6(a) ((((struct in6_ifaddr *)(a))->ia_addr).sin6_addr)
899 #define IFMASK6(a) ((((struct in6_ifaddr *)(a))->ia_prefixmask).sin6_addr)
900 
901 	/*
902 	 * A link-local address is always a neighbor.
903 	 * XXX: we should use the sin6_scope_id field rather than the embedded
904 	 * interface index.
905 	 */
906 	if (IN6_IS_ADDR_LINKLOCAL(&addr->sin6_addr) &&
907 	    ntohs(*(u_int16_t *)&addr->sin6_addr.s6_addr[2]) == ifp->if_index)
908 		return(1);
909 
910 	/*
911 	 * If the address matches one of our addresses,
912 	 * it should be a neighbor.
913 	 */
914 	for (ifa = ifp->if_addrlist.tqh_first;
915 	     ifa;
916 	     ifa = ifa->ifa_list.tqe_next)
917 	{
918 		if (ifa->ifa_addr->sa_family != AF_INET6)
919 			next: continue;
920 
921 		for (i = 0; i < 4; i++) {
922 			if ((IFADDR6(ifa).s6_addr32[i] ^
923 			     addr->sin6_addr.s6_addr32[i]) &
924 			    IFMASK6(ifa).s6_addr32[i])
925 				goto next;
926 		}
927 		return(1);
928 	}
929 
930 	/*
931 	 * Even if the address matches none of our addresses, it might be
932 	 * in the neighbor cache.
933 	 */
934 	if (nd6_lookup(&addr->sin6_addr, 0, ifp) != NULL)
935 		return(1);
936 
937 	return(0);
938 #undef IFADDR6
939 #undef IFMASK6
940 }
941 
942 /*
943  * Free an nd6 llinfo entry.
944  */
945 struct llinfo_nd6 *
946 nd6_free(rt)
947 	struct rtentry *rt;
948 {
949 	struct llinfo_nd6 *ln = (struct llinfo_nd6 *)rt->rt_llinfo, *next;
950 	struct in6_addr in6 = ((struct sockaddr_in6 *)rt_key(rt))->sin6_addr;
951 	struct nd_defrouter *dr;
952 
953 	/*
954 	 * we used to have pfctlinput(PRC_HOSTDEAD) here.
955 	 * even though it is not harmful, it was not really necessary.
956 	 */
957 
958 	if (!ip6_forwarding && ip6_accept_rtadv) { /* XXX: too restrictive? */
959 		int s;
960 		s = splnet();
961 		dr = defrouter_lookup(&((struct sockaddr_in6 *)rt_key(rt))->sin6_addr,
962 				      rt->rt_ifp);
963 
964 		if (ln->ln_router || dr) {
965 			/*
966 			 * rt6_flush must be called whether or not the neighbor
967 			 * is in the Default Router List.
968 			 * See a corresponding comment in nd6_na_input().
969 			 */
970 			rt6_flush(&in6, rt->rt_ifp);
971 		}
972 
973 		if (dr) {
974 			/*
975 			 * Unreachablity of a router might affect the default
976 			 * router selection and on-link detection of advertised
977 			 * prefixes.
978 			 */
979 
980 			/*
981 			 * Temporarily fake the state to choose a new default
982 			 * router and to perform on-link determination of
983 			 * prefixes correctly.
984 			 * Below the state will be set correctly,
985 			 * or the entry itself will be deleted.
986 			 */
987 			ln->ln_state = ND6_LLINFO_INCOMPLETE;
988 
989 			/*
990 			 * Since defrouter_select() does not affect the
991 			 * on-link determination and MIP6 needs the check
992 			 * before the default router selection, we perform
993 			 * the check now.
994 			 */
995 			pfxlist_onlink_check();
996 
997 			if (dr == TAILQ_FIRST(&nd_defrouter)) {
998 				/*
999 				 * It is used as the current default router,
1000 				 * so we have to move it to the end of the
1001 				 * list and choose a new one.
1002 				 * XXX: it is not very efficient if this is
1003 				 *      the only router.
1004 				 */
1005 				TAILQ_REMOVE(&nd_defrouter, dr, dr_entry);
1006 				TAILQ_INSERT_TAIL(&nd_defrouter, dr, dr_entry);
1007 
1008 				defrouter_select();
1009 			}
1010 		}
1011 		splx(s);
1012 	}
1013 
1014 	/*
1015 	 * Before deleting the entry, remember the next entry as the
1016 	 * return value.  We need this because pfxlist_onlink_check() above
1017 	 * might have freed other entries (particularly the old next entry) as
1018 	 * a side effect (XXX).
1019 	 */
1020 	next = ln->ln_next;
1021 
1022 	/*
1023 	 * Detach the route from the routing tree and the list of neighbor
1024 	 * caches, and disable the route entry not to be used in already
1025 	 * cached routes.
1026 	 */
1027 	rtrequest(RTM_DELETE, rt_key(rt), (struct sockaddr *)0,
1028 		  rt_mask(rt), 0, (struct rtentry **)0);
1029 
1030 	return(next);
1031 }
1032 
1033 /*
1034  * Upper-layer reachability hint for Neighbor Unreachability Detection.
1035  *
1036  * XXX cost-effective metods?
1037  */
1038 void
1039 nd6_nud_hint(rt, dst6, force)
1040 	struct rtentry *rt;
1041 	struct in6_addr *dst6;
1042 	int force;
1043 {
1044 	struct llinfo_nd6 *ln;
1045 
1046 	/*
1047 	 * If the caller specified "rt", use that.  Otherwise, resolve the
1048 	 * routing table by supplied "dst6".
1049 	 */
1050 	if (!rt) {
1051 		if (!dst6)
1052 			return;
1053 		if (!(rt = nd6_lookup(dst6, 0, NULL)))
1054 			return;
1055 	}
1056 
1057 	if ((rt->rt_flags & RTF_GATEWAY) != 0 ||
1058 	    (rt->rt_flags & RTF_LLINFO) == 0 ||
1059 	    !rt->rt_llinfo || !rt->rt_gateway ||
1060 	    rt->rt_gateway->sa_family != AF_LINK) {
1061 		/* This is not a host route. */
1062 		return;
1063 	}
1064 
1065 	ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1066 	if (ln->ln_state < ND6_LLINFO_REACHABLE)
1067 		return;
1068 
1069 	/*
1070 	 * if we get upper-layer reachability confirmation many times,
1071 	 * it is possible we have false information.
1072 	 */
1073 	if (!force) {
1074 		ln->ln_byhint++;
1075 		if (ln->ln_byhint > nd6_maxnudhint)
1076 			return;
1077 	}
1078 
1079 	ln->ln_state = ND6_LLINFO_REACHABLE;
1080 	if (ln->ln_expire)
1081 		ln->ln_expire = time_second +
1082 			nd_ifinfo[rt->rt_ifp->if_index].reachable;
1083 }
1084 
1085 void
1086 nd6_rtrequest(req, rt, info)
1087 	int	req;
1088 	struct rtentry *rt;
1089 	struct rt_addrinfo *info; /* xxx unused */
1090 {
1091 	struct sockaddr *gate = rt->rt_gateway;
1092 	struct llinfo_nd6 *ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1093 	static struct sockaddr_dl null_sdl = {sizeof(null_sdl), AF_LINK};
1094 	struct ifnet *ifp = rt->rt_ifp;
1095 	struct ifaddr *ifa;
1096 
1097 	if ((rt->rt_flags & RTF_GATEWAY))
1098 		return;
1099 
1100 	if (nd6_need_cache(ifp) == 0 && (rt->rt_flags & RTF_HOST) == 0) {
1101 		/*
1102 		 * This is probably an interface direct route for a link
1103 		 * which does not need neighbor caches (e.g. fe80::%lo0/64).
1104 		 * We do not need special treatment below for such a route.
1105 		 * Moreover, the RTF_LLINFO flag which would be set below
1106 		 * would annoy the ndp(8) command.
1107 		 */
1108 		return;
1109 	}
1110 
1111 	if (req == RTM_RESOLVE &&
1112 	    (nd6_need_cache(ifp) == 0 || /* stf case */
1113 	     !nd6_is_addr_neighbor((struct sockaddr_in6 *)rt_key(rt), ifp))) {
1114 		/*
1115 		 * FreeBSD and BSD/OS often make a cloned host route based
1116 		 * on a less-specific route (e.g. the default route).
1117 		 * If the less specific route does not have a "gateway"
1118 		 * (this is the case when the route just goes to a p2p or an
1119 		 * stf interface), we'll mistakenly make a neighbor cache for
1120 		 * the host route, and will see strange neighbor solicitation
1121 		 * for the corresponding destination.  In order to avoid the
1122 		 * confusion, we check if the destination of the route is
1123 		 * a neighbor in terms of neighbor discovery, and stop the
1124 		 * process if not.  Additionally, we remove the LLINFO flag
1125 		 * so that ndp(8) will not try to get the neighbor information
1126 		 * of the destination.
1127 		 */
1128 		rt->rt_flags &= ~RTF_LLINFO;
1129 		return;
1130 	}
1131 
1132 	switch (req) {
1133 	case RTM_ADD:
1134 		/*
1135 		 * There is no backward compatibility :)
1136 		 *
1137 		 * if ((rt->rt_flags & RTF_HOST) == 0 &&
1138 		 *     SIN(rt_mask(rt))->sin_addr.s_addr != 0xffffffff)
1139 		 *	   rt->rt_flags |= RTF_CLONING;
1140 		 */
1141 		if (rt->rt_flags & (RTF_CLONING | RTF_LLINFO)) {
1142 			/*
1143 			 * Case 1: This route should come from
1144 			 * a route to interface.  RTF_LLINFO flag is set
1145 			 * for a host route whose destination should be
1146 			 * treated as on-link.
1147 			 */
1148 			rt_setgate(rt, rt_key(rt),
1149 				   (struct sockaddr *)&null_sdl);
1150 			gate = rt->rt_gateway;
1151 			SDL(gate)->sdl_type = ifp->if_type;
1152 			SDL(gate)->sdl_index = ifp->if_index;
1153 			if (ln)
1154 				ln->ln_expire = time_second;
1155 #if 1
1156 			if (ln && ln->ln_expire == 0) {
1157 				/* kludge for desktops */
1158 #if 0
1159 				printf("nd6_rtequest: time.tv_sec is zero; "
1160 				       "treat it as 1\n");
1161 #endif
1162 				ln->ln_expire = 1;
1163 			}
1164 #endif
1165 			if ((rt->rt_flags & RTF_CLONING))
1166 				break;
1167 		}
1168 		/*
1169 		 * In IPv4 code, we try to annonuce new RTF_ANNOUNCE entry here.
1170 		 * We don't do that here since llinfo is not ready yet.
1171 		 *
1172 		 * There are also couple of other things to be discussed:
1173 		 * - unsolicited NA code needs improvement beforehand
1174 		 * - RFC2461 says we MAY send multicast unsolicited NA
1175 		 *   (7.2.6 paragraph 4), however, it also says that we
1176 		 *   SHOULD provide a mechanism to prevent multicast NA storm.
1177 		 *   we don't have anything like it right now.
1178 		 *   note that the mechanism needs a mutual agreement
1179 		 *   between proxies, which means that we need to implement
1180 		 *   a new protocol, or a new kludge.
1181 		 * - from RFC2461 6.2.4, host MUST NOT send an unsolicited NA.
1182 		 *   we need to check ip6forwarding before sending it.
1183 		 *   (or should we allow proxy ND configuration only for
1184 		 *   routers?  there's no mention about proxy ND from hosts)
1185 		 */
1186 #if 0
1187 		/* XXX it does not work */
1188 		if (rt->rt_flags & RTF_ANNOUNCE)
1189 			nd6_na_output(ifp,
1190 			      &SIN6(rt_key(rt))->sin6_addr,
1191 			      &SIN6(rt_key(rt))->sin6_addr,
1192 			      ip6_forwarding ? ND_NA_FLAG_ROUTER : 0,
1193 			      1, NULL);
1194 #endif
1195 		/* FALLTHROUGH */
1196 	case RTM_RESOLVE:
1197 		if ((ifp->if_flags & (IFF_POINTOPOINT | IFF_LOOPBACK)) == 0) {
1198 			/*
1199 			 * Address resolution isn't necessary for a point to
1200 			 * point link, so we can skip this test for a p2p link.
1201 			 */
1202 			if (gate->sa_family != AF_LINK ||
1203 			    gate->sa_len < sizeof(null_sdl)) {
1204 				log(LOG_DEBUG,
1205 				    "nd6_rtrequest: bad gateway value: %s\n",
1206 				    if_name(ifp));
1207 				break;
1208 			}
1209 			SDL(gate)->sdl_type = ifp->if_type;
1210 			SDL(gate)->sdl_index = ifp->if_index;
1211 		}
1212 		if (ln != NULL)
1213 			break;	/* This happens on a route change */
1214 		/*
1215 		 * Case 2: This route may come from cloning, or a manual route
1216 		 * add with a LL address.
1217 		 */
1218 		R_Malloc(ln, struct llinfo_nd6 *, sizeof(*ln));
1219 		rt->rt_llinfo = (caddr_t)ln;
1220 		if (!ln) {
1221 			log(LOG_DEBUG, "nd6_rtrequest: malloc failed\n");
1222 			break;
1223 		}
1224 		nd6_inuse++;
1225 		nd6_allocated++;
1226 		Bzero(ln, sizeof(*ln));
1227 		ln->ln_rt = rt;
1228 		/* this is required for "ndp" command. - shin */
1229 		if (req == RTM_ADD) {
1230 		        /*
1231 			 * gate should have some valid AF_LINK entry,
1232 			 * and ln->ln_expire should have some lifetime
1233 			 * which is specified by ndp command.
1234 			 */
1235 			ln->ln_state = ND6_LLINFO_REACHABLE;
1236 			ln->ln_byhint = 0;
1237 		} else {
1238 		        /*
1239 			 * When req == RTM_RESOLVE, rt is created and
1240 			 * initialized in rtrequest(), so rt_expire is 0.
1241 			 */
1242 			ln->ln_state = ND6_LLINFO_NOSTATE;
1243 			ln->ln_expire = time_second;
1244 		}
1245 		rt->rt_flags |= RTF_LLINFO;
1246 		ln->ln_next = llinfo_nd6.ln_next;
1247 		llinfo_nd6.ln_next = ln;
1248 		ln->ln_prev = &llinfo_nd6;
1249 		ln->ln_next->ln_prev = ln;
1250 
1251 		/*
1252 		 * check if rt_key(rt) is one of my address assigned
1253 		 * to the interface.
1254 		 */
1255 		ifa = (struct ifaddr *)in6ifa_ifpwithaddr(rt->rt_ifp,
1256 					  &SIN6(rt_key(rt))->sin6_addr);
1257 		if (ifa) {
1258 			caddr_t macp = nd6_ifptomac(ifp);
1259 			ln->ln_expire = 0;
1260 			ln->ln_state = ND6_LLINFO_REACHABLE;
1261 			ln->ln_byhint = 0;
1262 			if (macp) {
1263 				Bcopy(macp, LLADDR(SDL(gate)), ifp->if_addrlen);
1264 				SDL(gate)->sdl_alen = ifp->if_addrlen;
1265 			}
1266 			if (nd6_useloopback) {
1267 				rt->rt_ifp = &loif[0];	/* XXX */
1268 				/*
1269 				 * Make sure rt_ifa be equal to the ifaddr
1270 				 * corresponding to the address.
1271 				 * We need this because when we refer
1272 				 * rt_ifa->ia6_flags in ip6_input, we assume
1273 				 * that the rt_ifa points to the address instead
1274 				 * of the loopback address.
1275 				 */
1276 				if (ifa != rt->rt_ifa) {
1277 					IFAFREE(rt->rt_ifa);
1278 					IFAREF(ifa);
1279 					rt->rt_ifa = ifa;
1280 				}
1281 			}
1282 		} else if (rt->rt_flags & RTF_ANNOUNCE) {
1283 			ln->ln_expire = 0;
1284 			ln->ln_state = ND6_LLINFO_REACHABLE;
1285 			ln->ln_byhint = 0;
1286 
1287 			/* join solicited node multicast for proxy ND */
1288 			if (ifp->if_flags & IFF_MULTICAST) {
1289 				struct in6_addr llsol;
1290 				int error;
1291 
1292 				llsol = SIN6(rt_key(rt))->sin6_addr;
1293 				llsol.s6_addr16[0] = htons(0xff02);
1294 				llsol.s6_addr16[1] = htons(ifp->if_index);
1295 				llsol.s6_addr32[1] = 0;
1296 				llsol.s6_addr32[2] = htonl(1);
1297 				llsol.s6_addr8[12] = 0xff;
1298 
1299 				if (!in6_addmulti(&llsol, ifp, &error)) {
1300 					nd6log((LOG_ERR, "%s: failed to join "
1301 					    "%s (errno=%d)\n", if_name(ifp),
1302 					    ip6_sprintf(&llsol), error));
1303 				}
1304 			}
1305 		}
1306 		break;
1307 
1308 	case RTM_DELETE:
1309 		if (!ln)
1310 			break;
1311 		/* leave from solicited node multicast for proxy ND */
1312 		if ((rt->rt_flags & RTF_ANNOUNCE) != 0 &&
1313 		    (ifp->if_flags & IFF_MULTICAST) != 0) {
1314 			struct in6_addr llsol;
1315 			struct in6_multi *in6m;
1316 
1317 			llsol = SIN6(rt_key(rt))->sin6_addr;
1318 			llsol.s6_addr16[0] = htons(0xff02);
1319 			llsol.s6_addr16[1] = htons(ifp->if_index);
1320 			llsol.s6_addr32[1] = 0;
1321 			llsol.s6_addr32[2] = htonl(1);
1322 			llsol.s6_addr8[12] = 0xff;
1323 
1324 			IN6_LOOKUP_MULTI(llsol, ifp, in6m);
1325 			if (in6m)
1326 				in6_delmulti(in6m);
1327 		}
1328 		nd6_inuse--;
1329 		ln->ln_next->ln_prev = ln->ln_prev;
1330 		ln->ln_prev->ln_next = ln->ln_next;
1331 		ln->ln_prev = NULL;
1332 		rt->rt_llinfo = 0;
1333 		rt->rt_flags &= ~RTF_LLINFO;
1334 		if (ln->ln_hold)
1335 			m_freem(ln->ln_hold);
1336 		Free((caddr_t)ln);
1337 	}
1338 }
1339 
1340 int
1341 nd6_ioctl(cmd, data, ifp)
1342 	u_long cmd;
1343 	caddr_t	data;
1344 	struct ifnet *ifp;
1345 {
1346 	struct in6_drlist *drl = (struct in6_drlist *)data;
1347 	struct in6_prlist *prl = (struct in6_prlist *)data;
1348 	struct in6_ndireq *ndi = (struct in6_ndireq *)data;
1349 	struct in6_nbrinfo *nbi = (struct in6_nbrinfo *)data;
1350 	struct in6_ndifreq *ndif = (struct in6_ndifreq *)data;
1351 	struct nd_defrouter *dr, any;
1352 	struct nd_prefix *pr;
1353 	struct rtentry *rt;
1354 	int i = 0, error = 0;
1355 	int s;
1356 
1357 	switch (cmd) {
1358 	case SIOCGDRLST_IN6:
1359 		/*
1360 		 * obsolete API, use sysctl under net.inet6.icmp6
1361 		 */
1362 		bzero(drl, sizeof(*drl));
1363 		s = splnet();
1364 		dr = TAILQ_FIRST(&nd_defrouter);
1365 		while (dr && i < DRLSTSIZ) {
1366 			drl->defrouter[i].rtaddr = dr->rtaddr;
1367 			if (IN6_IS_ADDR_LINKLOCAL(&drl->defrouter[i].rtaddr)) {
1368 				/* XXX: need to this hack for KAME stack */
1369 				drl->defrouter[i].rtaddr.s6_addr16[1] = 0;
1370 			} else
1371 				log(LOG_ERR,
1372 				    "default router list contains a "
1373 				    "non-linklocal address(%s)\n",
1374 				    ip6_sprintf(&drl->defrouter[i].rtaddr));
1375 
1376 			drl->defrouter[i].flags = dr->flags;
1377 			drl->defrouter[i].rtlifetime = dr->rtlifetime;
1378 			drl->defrouter[i].expire = dr->expire;
1379 			drl->defrouter[i].if_index = dr->ifp->if_index;
1380 			i++;
1381 			dr = TAILQ_NEXT(dr, dr_entry);
1382 		}
1383 		splx(s);
1384 		break;
1385 	case SIOCGPRLST_IN6:
1386 		/*
1387 		 * obsolete API, use sysctl under net.inet6.icmp6
1388 		 */
1389 		/*
1390 		 * XXX meaning of fields, especialy "raflags", is very
1391 		 * differnet between RA prefix list and RR/static prefix list.
1392 		 * how about separating ioctls into two?
1393 		 */
1394 		bzero(prl, sizeof(*prl));
1395 		s = splnet();
1396 		pr = nd_prefix.lh_first;
1397 		while (pr && i < PRLSTSIZ) {
1398 			struct nd_pfxrouter *pfr;
1399 			int j;
1400 
1401 			(void)in6_embedscope(&prl->prefix[i].prefix,
1402 			    &pr->ndpr_prefix, NULL, NULL);
1403 			prl->prefix[i].raflags = pr->ndpr_raf;
1404 			prl->prefix[i].prefixlen = pr->ndpr_plen;
1405 			prl->prefix[i].vltime = pr->ndpr_vltime;
1406 			prl->prefix[i].pltime = pr->ndpr_pltime;
1407 			prl->prefix[i].if_index = pr->ndpr_ifp->if_index;
1408 			prl->prefix[i].expire = pr->ndpr_expire;
1409 
1410 			pfr = pr->ndpr_advrtrs.lh_first;
1411 			j = 0;
1412 			while (pfr) {
1413 				if (j < DRLSTSIZ) {
1414 #define RTRADDR prl->prefix[i].advrtr[j]
1415 					RTRADDR = pfr->router->rtaddr;
1416 					if (IN6_IS_ADDR_LINKLOCAL(&RTRADDR)) {
1417 						/* XXX: hack for KAME */
1418 						RTRADDR.s6_addr16[1] = 0;
1419 					} else
1420 						log(LOG_ERR,
1421 						    "a router(%s) advertises "
1422 						    "a prefix with "
1423 						    "non-link local address\n",
1424 						    ip6_sprintf(&RTRADDR));
1425 #undef RTRADDR
1426 				}
1427 				j++;
1428 				pfr = pfr->pfr_next;
1429 			}
1430 			prl->prefix[i].advrtrs = j;
1431 			prl->prefix[i].origin = PR_ORIG_RA;
1432 
1433 			i++;
1434 			pr = pr->ndpr_next;
1435 		}
1436 	      {
1437 		struct rr_prefix *rpp;
1438 
1439 		for (rpp = LIST_FIRST(&rr_prefix); rpp;
1440 		     rpp = LIST_NEXT(rpp, rp_entry)) {
1441 			if (i >= PRLSTSIZ)
1442 				break;
1443 			(void)in6_embedscope(&prl->prefix[i].prefix,
1444 			    &pr->ndpr_prefix, NULL, NULL);
1445 			prl->prefix[i].raflags = rpp->rp_raf;
1446 			prl->prefix[i].prefixlen = rpp->rp_plen;
1447 			prl->prefix[i].vltime = rpp->rp_vltime;
1448 			prl->prefix[i].pltime = rpp->rp_pltime;
1449 			prl->prefix[i].if_index = rpp->rp_ifp->if_index;
1450 			prl->prefix[i].expire = rpp->rp_expire;
1451 			prl->prefix[i].advrtrs = 0;
1452 			prl->prefix[i].origin = rpp->rp_origin;
1453 			i++;
1454 		}
1455 	      }
1456 		splx(s);
1457 
1458 		break;
1459 	case OSIOCGIFINFO_IN6:
1460 		if (!nd_ifinfo || i >= nd_ifinfo_indexlim) {
1461 			error = EINVAL;
1462 			break;
1463 		}
1464 		ndi->ndi.linkmtu = nd_ifinfo[ifp->if_index].linkmtu;
1465 		ndi->ndi.maxmtu = nd_ifinfo[ifp->if_index].maxmtu;
1466 		ndi->ndi.basereachable =
1467 		    nd_ifinfo[ifp->if_index].basereachable;
1468 		ndi->ndi.reachable = nd_ifinfo[ifp->if_index].reachable;
1469 		ndi->ndi.retrans = nd_ifinfo[ifp->if_index].retrans;
1470 		ndi->ndi.flags = nd_ifinfo[ifp->if_index].flags;
1471 		ndi->ndi.recalctm = nd_ifinfo[ifp->if_index].recalctm;
1472 		ndi->ndi.chlim = nd_ifinfo[ifp->if_index].chlim;
1473 		ndi->ndi.receivedra = nd_ifinfo[ifp->if_index].receivedra;
1474 		break;
1475 	case SIOCGIFINFO_IN6:
1476 		if (!nd_ifinfo || i >= nd_ifinfo_indexlim) {
1477 			error = EINVAL;
1478 			break;
1479 		}
1480 		ndi->ndi = nd_ifinfo[ifp->if_index];
1481 		break;
1482 	case SIOCSIFINFO_FLAGS:
1483 		/* XXX: almost all other fields of ndi->ndi is unused */
1484 		if (!nd_ifinfo || i >= nd_ifinfo_indexlim) {
1485 			error = EINVAL;
1486 			break;
1487 		}
1488 		nd_ifinfo[ifp->if_index].flags = ndi->ndi.flags;
1489 		break;
1490 	case SIOCSNDFLUSH_IN6:	/* XXX: the ioctl name is confusing... */
1491 		/* flush default router list */
1492 		/*
1493 		 * xxx sumikawa: should not delete route if default
1494 		 * route equals to the top of default router list
1495 		 */
1496 		bzero(&any, sizeof(any));
1497 		defrouter_delreq(&any, 0);
1498 		defrouter_select();
1499 		/* xxx sumikawa: flush prefix list */
1500 		break;
1501 	case SIOCSPFXFLUSH_IN6:
1502 	    {
1503 		/* flush all the prefix advertised by routers */
1504 		struct nd_prefix *pr, *next;
1505 
1506 		s = splnet();
1507 		for (pr = nd_prefix.lh_first; pr; pr = next) {
1508 			struct in6_ifaddr *ia, *ia_next;
1509 
1510 			next = pr->ndpr_next;
1511 
1512 			if (IN6_IS_ADDR_LINKLOCAL(&pr->ndpr_prefix.sin6_addr))
1513 				continue; /* XXX */
1514 
1515 			/* do we really have to remove addresses as well? */
1516 			for (ia = in6_ifaddr; ia; ia = ia_next) {
1517 				/* ia might be removed.  keep the next ptr. */
1518 				ia_next = ia->ia_next;
1519 
1520 				if ((ia->ia6_flags & IN6_IFF_AUTOCONF) == 0)
1521 					continue;
1522 
1523 				if (ia->ia6_ndpr == pr)
1524 					in6_purgeaddr(&ia->ia_ifa);
1525 			}
1526 			prelist_remove(pr);
1527 		}
1528 		splx(s);
1529 		break;
1530 	    }
1531 	case SIOCSRTRFLUSH_IN6:
1532 	    {
1533 		/* flush all the default routers */
1534 		struct nd_defrouter *dr, *next;
1535 
1536 		s = splnet();
1537 		if ((dr = TAILQ_FIRST(&nd_defrouter)) != NULL) {
1538 			/*
1539 			 * The first entry of the list may be stored in
1540 			 * the routing table, so we'll delete it later.
1541 			 */
1542 			for (dr = TAILQ_NEXT(dr, dr_entry); dr; dr = next) {
1543 				next = TAILQ_NEXT(dr, dr_entry);
1544 				defrtrlist_del(dr);
1545 			}
1546 			defrtrlist_del(TAILQ_FIRST(&nd_defrouter));
1547 		}
1548 		splx(s);
1549 		break;
1550 	    }
1551 	case SIOCGNBRINFO_IN6:
1552 	    {
1553 		struct llinfo_nd6 *ln;
1554 		struct in6_addr nb_addr = nbi->addr; /* make local for safety */
1555 
1556 		/*
1557 		 * XXX: KAME specific hack for scoped addresses
1558 		 *      XXXX: for other scopes than link-local?
1559 		 */
1560 		if (IN6_IS_ADDR_LINKLOCAL(&nbi->addr) ||
1561 		    IN6_IS_ADDR_MC_LINKLOCAL(&nbi->addr)) {
1562 			u_int16_t *idp = (u_int16_t *)&nb_addr.s6_addr[2];
1563 
1564 			if (*idp == 0)
1565 				*idp = htons(ifp->if_index);
1566 		}
1567 
1568 		s = splnet();
1569 		if ((rt = nd6_lookup(&nb_addr, 0, ifp)) == NULL) {
1570 			error = EINVAL;
1571 			splx(s);
1572 			break;
1573 		}
1574 		ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1575 		nbi->state = ln->ln_state;
1576 		nbi->asked = ln->ln_asked;
1577 		nbi->isrouter = ln->ln_router;
1578 		nbi->expire = ln->ln_expire;
1579 		splx(s);
1580 
1581 		break;
1582 	    }
1583 	case SIOCGDEFIFACE_IN6:	/* XXX: should be implemented as a sysctl? */
1584 		ndif->ifindex = nd6_defifindex;
1585 		break;
1586 	case SIOCSDEFIFACE_IN6:	/* XXX: should be implemented as a sysctl? */
1587 		return(nd6_setdefaultiface(ndif->ifindex));
1588 		break;
1589 	}
1590 	return(error);
1591 }
1592 
1593 /*
1594  * Create neighbor cache entry and cache link-layer address,
1595  * on reception of inbound ND6 packets. (RS/RA/NS/redirect)
1596  */
1597 struct rtentry *
1598 nd6_cache_lladdr(ifp, from, lladdr, lladdrlen, type, code)
1599 	struct ifnet *ifp;
1600 	struct in6_addr *from;
1601 	char *lladdr;
1602 	int lladdrlen;
1603 	int type;	/* ICMP6 type */
1604 	int code;	/* type dependent information */
1605 {
1606 	struct rtentry *rt = NULL;
1607 	struct llinfo_nd6 *ln = NULL;
1608 	int is_newentry;
1609 	struct sockaddr_dl *sdl = NULL;
1610 	int do_update;
1611 	int olladdr;
1612 	int llchange;
1613 	int newstate = 0;
1614 
1615 	if (!ifp)
1616 		panic("ifp == NULL in nd6_cache_lladdr");
1617 	if (!from)
1618 		panic("from == NULL in nd6_cache_lladdr");
1619 
1620 	/* nothing must be updated for unspecified address */
1621 	if (IN6_IS_ADDR_UNSPECIFIED(from))
1622 		return NULL;
1623 
1624 	/*
1625 	 * Validation about ifp->if_addrlen and lladdrlen must be done in
1626 	 * the caller.
1627 	 *
1628 	 * XXX If the link does not have link-layer adderss, what should
1629 	 * we do? (ifp->if_addrlen == 0)
1630 	 * Spec says nothing in sections for RA, RS and NA.  There's small
1631 	 * description on it in NS section (RFC 2461 7.2.3).
1632 	 */
1633 
1634 	rt = nd6_lookup(from, 0, ifp);
1635 	if (!rt) {
1636 #if 0
1637 		/* nothing must be done if there's no lladdr */
1638 		if (!lladdr || !lladdrlen)
1639 			return NULL;
1640 #endif
1641 
1642 		rt = nd6_lookup(from, 1, ifp);
1643 		is_newentry = 1;
1644 	} else {
1645 		/* do nothing if static ndp is set */
1646 		if (rt->rt_flags & RTF_STATIC)
1647 			return NULL;
1648 		is_newentry = 0;
1649 	}
1650 
1651 	if (!rt)
1652 		return NULL;
1653 	if ((rt->rt_flags & (RTF_GATEWAY | RTF_LLINFO)) != RTF_LLINFO) {
1654 fail:
1655 		(void)nd6_free(rt);
1656 		return NULL;
1657 	}
1658 	ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1659 	if (!ln)
1660 		goto fail;
1661 	if (!rt->rt_gateway)
1662 		goto fail;
1663 	if (rt->rt_gateway->sa_family != AF_LINK)
1664 		goto fail;
1665 	sdl = SDL(rt->rt_gateway);
1666 
1667 	olladdr = (sdl->sdl_alen) ? 1 : 0;
1668 	if (olladdr && lladdr) {
1669 		if (bcmp(lladdr, LLADDR(sdl), ifp->if_addrlen))
1670 			llchange = 1;
1671 		else
1672 			llchange = 0;
1673 	} else
1674 		llchange = 0;
1675 
1676 	/*
1677 	 * newentry olladdr  lladdr  llchange	(*=record)
1678 	 *	0	n	n	--	(1)
1679 	 *	0	y	n	--	(2)
1680 	 *	0	n	y	--	(3) * STALE
1681 	 *	0	y	y	n	(4) *
1682 	 *	0	y	y	y	(5) * STALE
1683 	 *	1	--	n	--	(6)   NOSTATE(= PASSIVE)
1684 	 *	1	--	y	--	(7) * STALE
1685 	 */
1686 
1687 	if (lladdr) {		/* (3-5) and (7) */
1688 		/*
1689 		 * Record source link-layer address
1690 		 * XXX is it dependent to ifp->if_type?
1691 		 */
1692 		sdl->sdl_alen = ifp->if_addrlen;
1693 		bcopy(lladdr, LLADDR(sdl), ifp->if_addrlen);
1694 	}
1695 
1696 	if (!is_newentry) {
1697 		if ((!olladdr && lladdr)		/* (3) */
1698 		 || (olladdr && lladdr && llchange)) {	/* (5) */
1699 			do_update = 1;
1700 			newstate = ND6_LLINFO_STALE;
1701 		} else					/* (1-2,4) */
1702 			do_update = 0;
1703 	} else {
1704 		do_update = 1;
1705 		if (!lladdr)				/* (6) */
1706 			newstate = ND6_LLINFO_NOSTATE;
1707 		else					/* (7) */
1708 			newstate = ND6_LLINFO_STALE;
1709 	}
1710 
1711 	if (do_update) {
1712 		/*
1713 		 * Update the state of the neighbor cache.
1714 		 */
1715 		ln->ln_state = newstate;
1716 
1717 		if (ln->ln_state == ND6_LLINFO_STALE) {
1718 			/*
1719 			 * XXX: since nd6_output() below will cause
1720 			 * state tansition to DELAY and reset the timer,
1721 			 * we must set the timer now, although it is actually
1722 			 * meaningless.
1723 			 */
1724 			ln->ln_expire = time_second + nd6_gctimer;
1725 
1726 			if (ln->ln_hold) {
1727 				/*
1728 				 * we assume ifp is not a p2p here, so just
1729 				 * set the 2nd argument as the 1st one.
1730 				 */
1731 				nd6_output(ifp, ifp, ln->ln_hold,
1732 					   (struct sockaddr_in6 *)rt_key(rt),
1733 					   rt);
1734 				ln->ln_hold = NULL;
1735 			}
1736 		} else if (ln->ln_state == ND6_LLINFO_INCOMPLETE) {
1737 			/* probe right away */
1738 			ln->ln_expire = time_second;
1739 		}
1740 	}
1741 
1742 	/*
1743 	 * ICMP6 type dependent behavior.
1744 	 *
1745 	 * NS: clear IsRouter if new entry
1746 	 * RS: clear IsRouter
1747 	 * RA: set IsRouter if there's lladdr
1748 	 * redir: clear IsRouter if new entry
1749 	 *
1750 	 * RA case, (1):
1751 	 * The spec says that we must set IsRouter in the following cases:
1752 	 * - If lladdr exist, set IsRouter.  This means (1-5).
1753 	 * - If it is old entry (!newentry), set IsRouter.  This means (7).
1754 	 * So, based on the spec, in (1-5) and (7) cases we must set IsRouter.
1755 	 * A quetion arises for (1) case.  (1) case has no lladdr in the
1756 	 * neighbor cache, this is similar to (6).
1757 	 * This case is rare but we figured that we MUST NOT set IsRouter.
1758 	 *
1759 	 * newentry olladdr  lladdr  llchange	    NS  RS  RA	redir
1760 	 *							D R
1761 	 *	0	n	n	--	(1)	c   ?     s
1762 	 *	0	y	n	--	(2)	c   s     s
1763 	 *	0	n	y	--	(3)	c   s     s
1764 	 *	0	y	y	n	(4)	c   s     s
1765 	 *	0	y	y	y	(5)	c   s     s
1766 	 *	1	--	n	--	(6) c	c 	c s
1767 	 *	1	--	y	--	(7) c	c   s	c s
1768 	 *
1769 	 *					(c=clear s=set)
1770 	 */
1771 	switch (type & 0xff) {
1772 	case ND_NEIGHBOR_SOLICIT:
1773 		/*
1774 		 * New entry must have is_router flag cleared.
1775 		 */
1776 		if (is_newentry)	/* (6-7) */
1777 			ln->ln_router = 0;
1778 		break;
1779 	case ND_REDIRECT:
1780 		/*
1781 		 * If the icmp is a redirect to a better router, always set the
1782 		 * is_router flag. Otherwise, if the entry is newly created,
1783 		 * clear the flag. [RFC 2461, sec 8.3]
1784 		 */
1785 		if (code == ND_REDIRECT_ROUTER)
1786 			ln->ln_router = 1;
1787 		else if (is_newentry) /* (6-7) */
1788 			ln->ln_router = 0;
1789 		break;
1790 	case ND_ROUTER_SOLICIT:
1791 		/*
1792 		 * is_router flag must always be cleared.
1793 		 */
1794 		ln->ln_router = 0;
1795 		break;
1796 	case ND_ROUTER_ADVERT:
1797 		/*
1798 		 * Mark an entry with lladdr as a router.
1799 		 */
1800 		if ((!is_newentry && (olladdr || lladdr))	/* (2-5) */
1801 		 || (is_newentry && lladdr)) {			/* (7) */
1802 			ln->ln_router = 1;
1803 		}
1804 		break;
1805 	}
1806 
1807 	/*
1808 	 * When the link-layer address of a router changes, select the
1809 	 * best router again.  In particular, when the neighbor entry is newly
1810 	 * created, it might affect the selection policy.
1811 	 * Question: can we restrict the first condition to the "is_newentry"
1812 	 * case?
1813 	 * XXX: when we hear an RA from a new router with the link-layer
1814 	 * address option, defrouter_select() is called twice, since
1815 	 * defrtrlist_update called the function as well.  However, I believe
1816 	 * we can compromise the overhead, since it only happens the first
1817 	 * time.
1818 	 * XXX: although defrouter_select() should not have a bad effect
1819 	 * for those are not autoconfigured hosts, we explicitly avoid such
1820 	 * cases for safety.
1821 	 */
1822 	if (do_update && ln->ln_router && !ip6_forwarding && ip6_accept_rtadv)
1823 		defrouter_select();
1824 
1825 	return rt;
1826 }
1827 
1828 static void
1829 nd6_slowtimo(ignored_arg)
1830     void *ignored_arg;
1831 {
1832 	int s = splnet();
1833 	int i;
1834 	struct nd_ifinfo *nd6if;
1835 
1836 	callout_reset(&nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
1837 	    nd6_slowtimo, NULL);
1838 	for (i = 1; i < if_index + 1; i++) {
1839 		if (!nd_ifinfo || i >= nd_ifinfo_indexlim)
1840 			continue;
1841 		nd6if = &nd_ifinfo[i];
1842 		if (nd6if->basereachable && /* already initialized */
1843 		    (nd6if->recalctm -= ND6_SLOWTIMER_INTERVAL) <= 0) {
1844 			/*
1845 			 * Since reachable time rarely changes by router
1846 			 * advertisements, we SHOULD insure that a new random
1847 			 * value gets recomputed at least once every few hours.
1848 			 * (RFC 2461, 6.3.4)
1849 			 */
1850 			nd6if->recalctm = nd6_recalc_reachtm_interval;
1851 			nd6if->reachable = ND_COMPUTE_RTIME(nd6if->basereachable);
1852 		}
1853 	}
1854 	splx(s);
1855 }
1856 
1857 #define senderr(e) { error = (e); goto bad;}
1858 int
1859 nd6_output(ifp, origifp, m0, dst, rt0)
1860 	struct ifnet *ifp;
1861 	struct ifnet *origifp;
1862 	struct mbuf *m0;
1863 	struct sockaddr_in6 *dst;
1864 	struct rtentry *rt0;
1865 {
1866 	struct mbuf *m = m0;
1867 	struct rtentry *rt = rt0;
1868 	struct sockaddr_in6 *gw6 = NULL;
1869 	struct llinfo_nd6 *ln = NULL;
1870 	int error = 0;
1871 
1872 	if (IN6_IS_ADDR_MULTICAST(&dst->sin6_addr))
1873 		goto sendpkt;
1874 
1875 	if (nd6_need_cache(ifp) == 0)
1876 		goto sendpkt;
1877 
1878 	/*
1879 	 * next hop determination.  This routine is derived from ether_outpout.
1880 	 */
1881 	if (rt) {
1882 		if ((rt->rt_flags & RTF_UP) == 0) {
1883 			if ((rt0 = rt = rtalloc1((struct sockaddr *)dst, 1, 0UL)) !=
1884 				NULL)
1885 			{
1886 				rt->rt_refcnt--;
1887 				if (rt->rt_ifp != ifp) {
1888 					/* XXX: loop care? */
1889 					return nd6_output(ifp, origifp, m0,
1890 							  dst, rt);
1891 				}
1892 			} else
1893 				senderr(EHOSTUNREACH);
1894 		}
1895 
1896 		if (rt->rt_flags & RTF_GATEWAY) {
1897 			gw6 = (struct sockaddr_in6 *)rt->rt_gateway;
1898 
1899 			/*
1900 			 * We skip link-layer address resolution and NUD
1901 			 * if the gateway is not a neighbor from ND point
1902 			 * of view, regardless of the value of nd_ifinfo.flags.
1903 			 * The second condition is a bit tricky; we skip
1904 			 * if the gateway is our own address, which is
1905 			 * sometimes used to install a route to a p2p link.
1906 			 */
1907 			if (!nd6_is_addr_neighbor(gw6, ifp) ||
1908 			    in6ifa_ifpwithaddr(ifp, &gw6->sin6_addr)) {
1909 				/*
1910 				 * We allow this kind of tricky route only
1911 				 * when the outgoing interface is p2p.
1912 				 * XXX: we may need a more generic rule here.
1913 				 */
1914 				if ((ifp->if_flags & IFF_POINTOPOINT) == 0)
1915 					senderr(EHOSTUNREACH);
1916 
1917 				goto sendpkt;
1918 			}
1919 
1920 			if (rt->rt_gwroute == 0)
1921 				goto lookup;
1922 			if (((rt = rt->rt_gwroute)->rt_flags & RTF_UP) == 0) {
1923 				rtfree(rt); rt = rt0;
1924 			lookup: rt->rt_gwroute = rtalloc1(rt->rt_gateway, 1, 0UL);
1925 				if ((rt = rt->rt_gwroute) == 0)
1926 					senderr(EHOSTUNREACH);
1927 			}
1928 		}
1929 	}
1930 
1931 	/*
1932 	 * Address resolution or Neighbor Unreachability Detection
1933 	 * for the next hop.
1934 	 * At this point, the destination of the packet must be a unicast
1935 	 * or an anycast address(i.e. not a multicast).
1936 	 */
1937 
1938 	/* Look up the neighbor cache for the nexthop */
1939 	if (rt && (rt->rt_flags & RTF_LLINFO) != 0)
1940 		ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1941 	else {
1942 		/*
1943 		 * Since nd6_is_addr_neighbor() internally calls nd6_lookup(),
1944 		 * the condition below is not very efficient.  But we believe
1945 		 * it is tolerable, because this should be a rare case.
1946 		 */
1947 		if (nd6_is_addr_neighbor(dst, ifp) &&
1948 		    (rt = nd6_lookup(&dst->sin6_addr, 1, ifp)) != NULL)
1949 			ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1950 	}
1951 	if (!ln || !rt) {
1952 		if ((ifp->if_flags & IFF_POINTOPOINT) == 0 &&
1953 		    !(nd_ifinfo[ifp->if_index].flags & ND6_IFF_PERFORMNUD)) {
1954 			log(LOG_DEBUG,
1955 			    "nd6_output: can't allocate llinfo for %s "
1956 			    "(ln=%p, rt=%p)\n",
1957 			    ip6_sprintf(&dst->sin6_addr), ln, rt);
1958 			senderr(EIO);	/* XXX: good error? */
1959 		}
1960 
1961 		goto sendpkt;	/* send anyway */
1962 	}
1963 
1964 	/* We don't have to do link-layer address resolution on a p2p link. */
1965 	if ((ifp->if_flags & IFF_POINTOPOINT) != 0 &&
1966 	    ln->ln_state < ND6_LLINFO_REACHABLE) {
1967 		ln->ln_state = ND6_LLINFO_STALE;
1968 		ln->ln_expire = time_second + nd6_gctimer;
1969 	}
1970 
1971 	/*
1972 	 * The first time we send a packet to a neighbor whose entry is
1973 	 * STALE, we have to change the state to DELAY and a sets a timer to
1974 	 * expire in DELAY_FIRST_PROBE_TIME seconds to ensure do
1975 	 * neighbor unreachability detection on expiration.
1976 	 * (RFC 2461 7.3.3)
1977 	 */
1978 	if (ln->ln_state == ND6_LLINFO_STALE) {
1979 		ln->ln_asked = 0;
1980 		ln->ln_state = ND6_LLINFO_DELAY;
1981 		ln->ln_expire = time_second + nd6_delay;
1982 	}
1983 
1984 	/*
1985 	 * If the neighbor cache entry has a state other than INCOMPLETE
1986 	 * (i.e. its link-layer address is already resolved), just
1987 	 * send the packet.
1988 	 */
1989 	if (ln->ln_state > ND6_LLINFO_INCOMPLETE)
1990 		goto sendpkt;
1991 
1992 	/*
1993 	 * There is a neighbor cache entry, but no ethernet address
1994 	 * response yet.  Replace the held mbuf (if any) with this
1995 	 * latest one.
1996 	 *
1997 	 * This code conforms to the rate-limiting rule described in Section
1998 	 * 7.2.2 of RFC 2461, because the timer is set correctly after sending
1999 	 * an NS below.
2000 	 */
2001 	if (ln->ln_state == ND6_LLINFO_NOSTATE)
2002 		ln->ln_state = ND6_LLINFO_INCOMPLETE;
2003 	if (ln->ln_hold)
2004 		m_freem(ln->ln_hold);
2005 	ln->ln_hold = m;
2006 	if (ln->ln_expire) {
2007 		if (ln->ln_asked < nd6_mmaxtries &&
2008 		    ln->ln_expire < time_second) {
2009 			ln->ln_asked++;
2010 			ln->ln_expire = time_second +
2011 				nd_ifinfo[ifp->if_index].retrans / 1000;
2012 			nd6_ns_output(ifp, NULL, &dst->sin6_addr, ln, 0);
2013 		}
2014 	}
2015 	return(0);
2016 
2017   sendpkt:
2018 
2019 #ifdef MAC
2020 	mac_create_mbuf_linklayer(ifp, m);
2021 #endif
2022 	if ((ifp->if_flags & IFF_LOOPBACK) != 0) {
2023 		return((*ifp->if_output)(origifp, m, (struct sockaddr *)dst,
2024 					 rt));
2025 	}
2026 	return((*ifp->if_output)(ifp, m, (struct sockaddr *)dst, rt));
2027 
2028   bad:
2029 	if (m)
2030 		m_freem(m);
2031 	return (error);
2032 }
2033 #undef senderr
2034 
2035 int
2036 nd6_need_cache(ifp)
2037 	struct ifnet *ifp;
2038 {
2039 	/*
2040 	 * XXX: we currently do not make neighbor cache on any interface
2041 	 * other than ARCnet, Ethernet, FDDI and GIF.
2042 	 *
2043 	 * RFC2893 says:
2044 	 * - unidirectional tunnels needs no ND
2045 	 */
2046 	switch (ifp->if_type) {
2047 	case IFT_ARCNET:
2048 	case IFT_ETHER:
2049 	case IFT_FDDI:
2050 	case IFT_IEEE1394:
2051 #ifdef IFT_L2VLAN
2052 	case IFT_L2VLAN:
2053 #endif
2054 #ifdef IFT_IEEE80211
2055 	case IFT_IEEE80211:
2056 #endif
2057 	case IFT_GIF:		/* XXX need more cases? */
2058 		return(1);
2059 	default:
2060 		return(0);
2061 	}
2062 }
2063 
2064 int
2065 nd6_storelladdr(ifp, rt, m, dst, desten)
2066 	struct ifnet *ifp;
2067 	struct rtentry *rt;
2068 	struct mbuf *m;
2069 	struct sockaddr *dst;
2070 	u_char *desten;
2071 {
2072 	int i;
2073 	struct sockaddr_dl *sdl;
2074 
2075 	if (m->m_flags & M_MCAST) {
2076 		switch (ifp->if_type) {
2077 		case IFT_ETHER:
2078 		case IFT_FDDI:
2079 #ifdef IFT_L2VLAN
2080 	case IFT_L2VLAN:
2081 #endif
2082 #ifdef IFT_IEEE80211
2083 		case IFT_IEEE80211:
2084 #endif
2085 			ETHER_MAP_IPV6_MULTICAST(&SIN6(dst)->sin6_addr,
2086 						 desten);
2087 			return(1);
2088 		case IFT_IEEE1394:
2089 			/*
2090 			 * netbsd can use if_broadcastaddr, but we don't do so
2091 			 * to reduce # of ifdef.
2092 			 */
2093 			for (i = 0; i < ifp->if_addrlen; i++)
2094 				desten[i] = ~0;
2095 			return(1);
2096 		case IFT_ARCNET:
2097 			*desten = 0;
2098 			return(1);
2099 		default:
2100 			m_freem(m);
2101 			return(0);
2102 		}
2103 	}
2104 
2105 	if (rt == NULL) {
2106 		/* this could happen, if we could not allocate memory */
2107 		m_freem(m);
2108 		return(0);
2109 	}
2110 	if (rt->rt_gateway->sa_family != AF_LINK) {
2111 		printf("nd6_storelladdr: something odd happens\n");
2112 		m_freem(m);
2113 		return(0);
2114 	}
2115 	sdl = SDL(rt->rt_gateway);
2116 	if (sdl->sdl_alen == 0) {
2117 		/* this should be impossible, but we bark here for debugging */
2118 		printf("nd6_storelladdr: sdl_alen == 0\n");
2119 		m_freem(m);
2120 		return(0);
2121 	}
2122 
2123 	bcopy(LLADDR(sdl), desten, sdl->sdl_alen);
2124 	return(1);
2125 }
2126 
2127 static int nd6_sysctl_drlist(SYSCTL_HANDLER_ARGS);
2128 static int nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS);
2129 #ifdef SYSCTL_DECL
2130 SYSCTL_DECL(_net_inet6_icmp6);
2131 #endif
2132 SYSCTL_NODE(_net_inet6_icmp6, ICMPV6CTL_ND6_DRLIST, nd6_drlist,
2133 	CTLFLAG_RD, nd6_sysctl_drlist, "");
2134 SYSCTL_NODE(_net_inet6_icmp6, ICMPV6CTL_ND6_PRLIST, nd6_prlist,
2135 	CTLFLAG_RD, nd6_sysctl_prlist, "");
2136 
2137 static int
2138 nd6_sysctl_drlist(SYSCTL_HANDLER_ARGS)
2139 {
2140 	int error;
2141 	char buf[1024];
2142 	struct in6_defrouter *d, *de;
2143 	struct nd_defrouter *dr;
2144 
2145 	if (req->newptr)
2146 		return EPERM;
2147 	error = 0;
2148 
2149 	for (dr = TAILQ_FIRST(&nd_defrouter);
2150 	     dr;
2151 	     dr = TAILQ_NEXT(dr, dr_entry)) {
2152 		d = (struct in6_defrouter *)buf;
2153 		de = (struct in6_defrouter *)(buf + sizeof(buf));
2154 
2155 		if (d + 1 <= de) {
2156 			bzero(d, sizeof(*d));
2157 			d->rtaddr.sin6_family = AF_INET6;
2158 			d->rtaddr.sin6_len = sizeof(d->rtaddr);
2159 			if (in6_recoverscope(&d->rtaddr, &dr->rtaddr,
2160 			    dr->ifp) != 0)
2161 				log(LOG_ERR,
2162 				    "scope error in "
2163 				    "default router list (%s)\n",
2164 				    ip6_sprintf(&dr->rtaddr));
2165 			d->flags = dr->flags;
2166 			d->rtlifetime = dr->rtlifetime;
2167 			d->expire = dr->expire;
2168 			d->if_index = dr->ifp->if_index;
2169 		} else
2170 			panic("buffer too short");
2171 
2172 		error = SYSCTL_OUT(req, buf, sizeof(*d));
2173 		if (error)
2174 			break;
2175 	}
2176 	return error;
2177 }
2178 
2179 static int
2180 nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS)
2181 {
2182 	int error;
2183 	char buf[1024];
2184 	struct in6_prefix *p, *pe;
2185 	struct nd_prefix *pr;
2186 
2187 	if (req->newptr)
2188 		return EPERM;
2189 	error = 0;
2190 
2191 	for (pr = nd_prefix.lh_first; pr; pr = pr->ndpr_next) {
2192 		u_short advrtrs;
2193 		size_t advance;
2194 		struct sockaddr_in6 *sin6, *s6;
2195 		struct nd_pfxrouter *pfr;
2196 
2197 		p = (struct in6_prefix *)buf;
2198 		pe = (struct in6_prefix *)(buf + sizeof(buf));
2199 
2200 		if (p + 1 <= pe) {
2201 			bzero(p, sizeof(*p));
2202 			sin6 = (struct sockaddr_in6 *)(p + 1);
2203 
2204 			p->prefix = pr->ndpr_prefix;
2205 			if (in6_recoverscope(&p->prefix,
2206 			    &p->prefix.sin6_addr, pr->ndpr_ifp) != 0)
2207 				log(LOG_ERR,
2208 				    "scope error in prefix list (%s)\n",
2209 				    ip6_sprintf(&p->prefix.sin6_addr));
2210 			p->raflags = pr->ndpr_raf;
2211 			p->prefixlen = pr->ndpr_plen;
2212 			p->vltime = pr->ndpr_vltime;
2213 			p->pltime = pr->ndpr_pltime;
2214 			p->if_index = pr->ndpr_ifp->if_index;
2215 			p->expire = pr->ndpr_expire;
2216 			p->refcnt = pr->ndpr_refcnt;
2217 			p->flags = pr->ndpr_stateflags;
2218 			p->origin = PR_ORIG_RA;
2219 			advrtrs = 0;
2220 			for (pfr = pr->ndpr_advrtrs.lh_first;
2221 			     pfr;
2222 			     pfr = pfr->pfr_next) {
2223 				if ((void *)&sin6[advrtrs + 1] >
2224 				    (void *)pe) {
2225 					advrtrs++;
2226 					continue;
2227 				}
2228 				s6 = &sin6[advrtrs];
2229 				bzero(s6, sizeof(*s6));
2230 				s6->sin6_family = AF_INET6;
2231 				s6->sin6_len = sizeof(*sin6);
2232 				if (in6_recoverscope(s6,
2233 				    &pfr->router->rtaddr,
2234 				    pfr->router->ifp) != 0)
2235 					log(LOG_ERR,
2236 					    "scope error in "
2237 					    "prefix list (%s)\n",
2238 					    ip6_sprintf(&pfr->router->rtaddr));
2239 				advrtrs++;
2240 			}
2241 			p->advrtrs = advrtrs;
2242 		} else
2243 			panic("buffer too short");
2244 
2245 		advance = sizeof(*p) + sizeof(*sin6) * advrtrs;
2246 		error = SYSCTL_OUT(req, buf, advance);
2247 		if (error)
2248 			break;
2249 	}
2250 	return error;
2251 }
2252