xref: /netbsd-src/sys/netinet6/nd6.c (revision 404fbe5fb94ca1e054339640cabb2801ce52dd30)
1 /*	$NetBSD: nd6.c,v 1.131 2008/11/07 00:20:18 dyoung Exp $	*/
2 /*	$KAME: nd6.c,v 1.279 2002/06/08 11:16:51 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 #include <sys/cdefs.h>
34 __KERNEL_RCSID(0, "$NetBSD: nd6.c,v 1.131 2008/11/07 00:20:18 dyoung Exp $");
35 
36 #include "opt_ipsec.h"
37 
38 #include <sys/param.h>
39 #include <sys/systm.h>
40 #include <sys/callout.h>
41 #include <sys/malloc.h>
42 #include <sys/mbuf.h>
43 #include <sys/socket.h>
44 #include <sys/socketvar.h>
45 #include <sys/sockio.h>
46 #include <sys/time.h>
47 #include <sys/kernel.h>
48 #include <sys/protosw.h>
49 #include <sys/errno.h>
50 #include <sys/ioctl.h>
51 #include <sys/syslog.h>
52 #include <sys/queue.h>
53 
54 #include <net/if.h>
55 #include <net/if_dl.h>
56 #include <net/if_types.h>
57 #include <net/route.h>
58 #include <net/if_ether.h>
59 #include <net/if_fddi.h>
60 #include <net/if_arc.h>
61 
62 #include <netinet/in.h>
63 #include <netinet6/in6_var.h>
64 #include <netinet/ip6.h>
65 #include <netinet6/ip6_var.h>
66 #include <netinet6/scope6_var.h>
67 #include <netinet6/nd6.h>
68 #include <netinet/icmp6.h>
69 #include <netinet6/icmp6_private.h>
70 
71 #ifdef IPSEC
72 #include <netinet6/ipsec.h>
73 #endif
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 /* timer values */
81 int	nd6_prune	= 1;	/* walk list every 1 seconds */
82 int	nd6_delay	= 5;	/* delay first probe time 5 second */
83 int	nd6_umaxtries	= 3;	/* maximum unicast query */
84 int	nd6_mmaxtries	= 3;	/* maximum multicast query */
85 int	nd6_useloopback = 1;	/* use loopback interface for local traffic */
86 int	nd6_gctimer	= (60 * 60 * 24); /* 1 day: garbage collection timer */
87 
88 /* preventing too many loops in ND option parsing */
89 int nd6_maxndopt = 10;	/* max # of ND options allowed */
90 
91 int nd6_maxnudhint = 0;	/* max # of subsequent upper layer hints */
92 
93 int nd6_maxqueuelen = 1; /* max # of packets cached in unresolved ND entries */
94 
95 #ifdef ND6_DEBUG
96 int nd6_debug = 1;
97 #else
98 int nd6_debug = 0;
99 #endif
100 
101 /* for debugging? */
102 static int nd6_inuse, nd6_allocated;
103 
104 struct llinfo_nd6 llinfo_nd6 = {
105 	.ln_prev = &llinfo_nd6,
106 	.ln_next = &llinfo_nd6,
107 };
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 const struct sockaddr_in6 all1_sa = {
113 	  .sin6_family = AF_INET6
114 	, .sin6_len = sizeof(struct sockaddr_in6)
115 	, .sin6_addr = {.s6_addr = {0xff, 0xff, 0xff, 0xff,
116 				    0xff, 0xff, 0xff, 0xff,
117 				    0xff, 0xff, 0xff, 0xff,
118 				    0xff, 0xff, 0xff, 0xff}}
119 };
120 
121 static void nd6_setmtu0(struct ifnet *, struct nd_ifinfo *);
122 static void nd6_slowtimo(void *);
123 static int regen_tmpaddr(struct in6_ifaddr *);
124 static struct llinfo_nd6 *nd6_free(struct rtentry *, int);
125 static void nd6_llinfo_timer(void *);
126 static void clear_llinfo_pqueue(struct llinfo_nd6 *);
127 
128 callout_t nd6_slowtimo_ch;
129 callout_t nd6_timer_ch;
130 extern callout_t in6_tmpaddrtimer_ch;
131 
132 static int fill_drlist(void *, size_t *, size_t);
133 static int fill_prlist(void *, size_t *, size_t);
134 
135 MALLOC_DEFINE(M_IP6NDP, "NDP", "IPv6 Neighbour Discovery");
136 
137 void
138 nd6_init(void)
139 {
140 	static int nd6_init_done = 0;
141 
142 	if (nd6_init_done) {
143 		log(LOG_NOTICE, "nd6_init called more than once(ignored)\n");
144 		return;
145 	}
146 
147 	/* initialization of the default router list */
148 	TAILQ_INIT(&nd_defrouter);
149 
150 	nd6_init_done = 1;
151 
152 	callout_init(&nd6_slowtimo_ch, CALLOUT_MPSAFE);
153 	callout_init(&nd6_timer_ch, CALLOUT_MPSAFE);
154 
155 	/* start timer */
156 	callout_reset(&nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
157 	    nd6_slowtimo, NULL);
158 }
159 
160 struct nd_ifinfo *
161 nd6_ifattach(struct ifnet *ifp)
162 {
163 	struct nd_ifinfo *nd;
164 
165 	nd = (struct nd_ifinfo *)malloc(sizeof(*nd), M_IP6NDP, M_WAITOK);
166 	memset(nd, 0, sizeof(*nd));
167 
168 	nd->initialized = 1;
169 
170 	nd->chlim = IPV6_DEFHLIM;
171 	nd->basereachable = REACHABLE_TIME;
172 	nd->reachable = ND_COMPUTE_RTIME(nd->basereachable);
173 	nd->retrans = RETRANS_TIMER;
174 	/*
175 	 * Note that the default value of ip6_accept_rtadv is 0, which means
176 	 * we won't accept RAs by default even if we set ND6_IFF_ACCEPT_RTADV
177 	 * here.
178 	 */
179 	nd->flags = (ND6_IFF_PERFORMNUD | ND6_IFF_ACCEPT_RTADV);
180 
181 	/* XXX: we cannot call nd6_setmtu since ifp is not fully initialized */
182 	nd6_setmtu0(ifp, nd);
183 
184 	return nd;
185 }
186 
187 void
188 nd6_ifdetach(struct nd_ifinfo *nd)
189 {
190 
191 	free(nd, M_IP6NDP);
192 }
193 
194 void
195 nd6_setmtu(struct ifnet *ifp)
196 {
197 	nd6_setmtu0(ifp, ND_IFINFO(ifp));
198 }
199 
200 void
201 nd6_setmtu0(struct ifnet *ifp, struct nd_ifinfo *ndi)
202 {
203 	u_int32_t omaxmtu;
204 
205 	omaxmtu = ndi->maxmtu;
206 
207 	switch (ifp->if_type) {
208 	case IFT_ARCNET:
209 		ndi->maxmtu = MIN(ARC_PHDS_MAXMTU, ifp->if_mtu); /* RFC2497 */
210 		break;
211 	case IFT_FDDI:
212 		ndi->maxmtu = MIN(FDDIIPMTU, ifp->if_mtu);
213 		break;
214 	default:
215 		ndi->maxmtu = ifp->if_mtu;
216 		break;
217 	}
218 
219 	/*
220 	 * Decreasing the interface MTU under IPV6 minimum MTU may cause
221 	 * undesirable situation.  We thus notify the operator of the change
222 	 * explicitly.  The check for omaxmtu is necessary to restrict the
223 	 * log to the case of changing the MTU, not initializing it.
224 	 */
225 	if (omaxmtu >= IPV6_MMTU && ndi->maxmtu < IPV6_MMTU) {
226 		log(LOG_NOTICE, "nd6_setmtu0: new link MTU on %s (%lu) is too"
227 		    " small for IPv6 which needs %lu\n",
228 		    if_name(ifp), (unsigned long)ndi->maxmtu, (unsigned long)
229 		    IPV6_MMTU);
230 	}
231 
232 	if (ndi->maxmtu > in6_maxmtu)
233 		in6_setmaxmtu(); /* check all interfaces just in case */
234 }
235 
236 void
237 nd6_option_init(void *opt, int icmp6len, union nd_opts *ndopts)
238 {
239 
240 	memset(ndopts, 0, sizeof(*ndopts));
241 	ndopts->nd_opts_search = (struct nd_opt_hdr *)opt;
242 	ndopts->nd_opts_last
243 		= (struct nd_opt_hdr *)(((u_char *)opt) + icmp6len);
244 
245 	if (icmp6len == 0) {
246 		ndopts->nd_opts_done = 1;
247 		ndopts->nd_opts_search = NULL;
248 	}
249 }
250 
251 /*
252  * Take one ND option.
253  */
254 struct nd_opt_hdr *
255 nd6_option(union nd_opts *ndopts)
256 {
257 	struct nd_opt_hdr *nd_opt;
258 	int olen;
259 
260 	if (ndopts == NULL)
261 		panic("ndopts == NULL in nd6_option");
262 	if (ndopts->nd_opts_last == NULL)
263 		panic("uninitialized ndopts in nd6_option");
264 	if (ndopts->nd_opts_search == NULL)
265 		return NULL;
266 	if (ndopts->nd_opts_done)
267 		return NULL;
268 
269 	nd_opt = ndopts->nd_opts_search;
270 
271 	/* make sure nd_opt_len is inside the buffer */
272 	if ((void *)&nd_opt->nd_opt_len >= (void *)ndopts->nd_opts_last) {
273 		memset(ndopts, 0, sizeof(*ndopts));
274 		return NULL;
275 	}
276 
277 	olen = nd_opt->nd_opt_len << 3;
278 	if (olen == 0) {
279 		/*
280 		 * Message validation requires that all included
281 		 * options have a length that is greater than zero.
282 		 */
283 		memset(ndopts, 0, sizeof(*ndopts));
284 		return NULL;
285 	}
286 
287 	ndopts->nd_opts_search = (struct nd_opt_hdr *)((char *)nd_opt + olen);
288 	if (ndopts->nd_opts_search > ndopts->nd_opts_last) {
289 		/* option overruns the end of buffer, invalid */
290 		memset(ndopts, 0, sizeof(*ndopts));
291 		return NULL;
292 	} else if (ndopts->nd_opts_search == ndopts->nd_opts_last) {
293 		/* reached the end of options chain */
294 		ndopts->nd_opts_done = 1;
295 		ndopts->nd_opts_search = NULL;
296 	}
297 	return nd_opt;
298 }
299 
300 /*
301  * Parse multiple ND options.
302  * This function is much easier to use, for ND routines that do not need
303  * multiple options of the same type.
304  */
305 int
306 nd6_options(union nd_opts *ndopts)
307 {
308 	struct nd_opt_hdr *nd_opt;
309 	int i = 0;
310 
311 	if (ndopts == NULL)
312 		panic("ndopts == NULL in nd6_options");
313 	if (ndopts->nd_opts_last == NULL)
314 		panic("uninitialized ndopts in nd6_options");
315 	if (ndopts->nd_opts_search == NULL)
316 		return 0;
317 
318 	while (1) {
319 		nd_opt = nd6_option(ndopts);
320 		if (nd_opt == NULL && ndopts->nd_opts_last == NULL) {
321 			/*
322 			 * Message validation requires that all included
323 			 * options have a length that is greater than zero.
324 			 */
325 			ICMP6_STATINC(ICMP6_STAT_ND_BADOPT);
326 			memset(ndopts, 0, sizeof(*ndopts));
327 			return -1;
328 		}
329 
330 		if (nd_opt == NULL)
331 			goto skip1;
332 
333 		switch (nd_opt->nd_opt_type) {
334 		case ND_OPT_SOURCE_LINKADDR:
335 		case ND_OPT_TARGET_LINKADDR:
336 		case ND_OPT_MTU:
337 		case ND_OPT_REDIRECTED_HEADER:
338 			if (ndopts->nd_opt_array[nd_opt->nd_opt_type]) {
339 				nd6log((LOG_INFO,
340 				    "duplicated ND6 option found (type=%d)\n",
341 				    nd_opt->nd_opt_type));
342 				/* XXX bark? */
343 			} else {
344 				ndopts->nd_opt_array[nd_opt->nd_opt_type]
345 					= nd_opt;
346 			}
347 			break;
348 		case ND_OPT_PREFIX_INFORMATION:
349 			if (ndopts->nd_opt_array[nd_opt->nd_opt_type] == 0) {
350 				ndopts->nd_opt_array[nd_opt->nd_opt_type]
351 					= nd_opt;
352 			}
353 			ndopts->nd_opts_pi_end =
354 				(struct nd_opt_prefix_info *)nd_opt;
355 			break;
356 		default:
357 			/*
358 			 * Unknown options must be silently ignored,
359 			 * to accommodate future extension to the protocol.
360 			 */
361 			nd6log((LOG_DEBUG,
362 			    "nd6_options: unsupported option %d - "
363 			    "option ignored\n", nd_opt->nd_opt_type));
364 		}
365 
366 skip1:
367 		i++;
368 		if (i > nd6_maxndopt) {
369 			ICMP6_STATINC(ICMP6_STAT_ND_TOOMANYOPT);
370 			nd6log((LOG_INFO, "too many loop in nd opt\n"));
371 			break;
372 		}
373 
374 		if (ndopts->nd_opts_done)
375 			break;
376 	}
377 
378 	return 0;
379 }
380 
381 /*
382  * ND6 timer routine to handle ND6 entries
383  */
384 void
385 nd6_llinfo_settimer(struct llinfo_nd6 *ln, long xtick)
386 {
387 	int s;
388 
389 	s = splsoftnet();
390 
391 	if (xtick < 0) {
392 		ln->ln_expire = 0;
393 		ln->ln_ntick = 0;
394 		callout_stop(&ln->ln_timer_ch);
395 	} else {
396 		ln->ln_expire = time_second + xtick / hz;
397 		if (xtick > INT_MAX) {
398 			ln->ln_ntick = xtick - INT_MAX;
399 			callout_reset(&ln->ln_timer_ch, INT_MAX,
400 			    nd6_llinfo_timer, ln);
401 		} else {
402 			ln->ln_ntick = 0;
403 			callout_reset(&ln->ln_timer_ch, xtick,
404 			    nd6_llinfo_timer, ln);
405 		}
406 	}
407 
408 	splx(s);
409 }
410 
411 static void
412 nd6_llinfo_timer(void *arg)
413 {
414 	struct llinfo_nd6 *ln;
415 	struct rtentry *rt;
416 	const struct sockaddr_in6 *dst;
417 	struct ifnet *ifp;
418 	struct nd_ifinfo *ndi = NULL;
419 
420 	mutex_enter(softnet_lock);
421 	KERNEL_LOCK(1, NULL);
422 
423 	ln = (struct llinfo_nd6 *)arg;
424 
425 	if (ln->ln_ntick > 0) {
426 		nd6_llinfo_settimer(ln, ln->ln_ntick);
427 		KERNEL_UNLOCK_ONE(NULL);
428 		mutex_exit(softnet_lock);
429 		return;
430 	}
431 
432 	if ((rt = ln->ln_rt) == NULL)
433 		panic("ln->ln_rt == NULL");
434 	if ((ifp = rt->rt_ifp) == NULL)
435 		panic("ln->ln_rt->rt_ifp == NULL");
436 	ndi = ND_IFINFO(ifp);
437 	dst = satocsin6(rt_getkey(rt));
438 
439 	/* sanity check */
440 	if (rt->rt_llinfo && (struct llinfo_nd6 *)rt->rt_llinfo != ln)
441 		panic("rt_llinfo(%p) is not equal to ln(%p)",
442 		      rt->rt_llinfo, ln);
443 	if (!dst)
444 		panic("dst=0 in nd6_timer(ln=%p)", ln);
445 
446 	switch (ln->ln_state) {
447 	case ND6_LLINFO_INCOMPLETE:
448 		if (ln->ln_asked < nd6_mmaxtries) {
449 			ln->ln_asked++;
450 			nd6_llinfo_settimer(ln, (long)ndi->retrans * hz / 1000);
451 			nd6_ns_output(ifp, NULL, &dst->sin6_addr, ln, 0);
452 		} else {
453 			struct mbuf *m = ln->ln_hold;
454 			if (m) {
455 				struct mbuf *m0;
456 
457 				/*
458 				 * assuming every packet in ln_hold has
459 				 * the same IP header
460 				 */
461 				m0 = m->m_nextpkt;
462 				m->m_nextpkt = NULL;
463 				icmp6_error2(m, ICMP6_DST_UNREACH,
464 				    ICMP6_DST_UNREACH_ADDR, 0, rt->rt_ifp);
465 
466 				ln->ln_hold = m0;
467 				clear_llinfo_pqueue(ln);
468  			}
469 			(void)nd6_free(rt, 0);
470 			ln = NULL;
471 		}
472 		break;
473 	case ND6_LLINFO_REACHABLE:
474 		if (!ND6_LLINFO_PERMANENT(ln)) {
475 			ln->ln_state = ND6_LLINFO_STALE;
476 			nd6_llinfo_settimer(ln, (long)nd6_gctimer * hz);
477 		}
478 		break;
479 
480 	case ND6_LLINFO_STALE:
481 		/* Garbage Collection(RFC 2461 5.3) */
482 		if (!ND6_LLINFO_PERMANENT(ln)) {
483 			(void)nd6_free(rt, 1);
484 			ln = NULL;
485 		}
486 		break;
487 
488 	case ND6_LLINFO_DELAY:
489 		if (ndi && (ndi->flags & ND6_IFF_PERFORMNUD) != 0) {
490 			/* We need NUD */
491 			ln->ln_asked = 1;
492 			ln->ln_state = ND6_LLINFO_PROBE;
493 			nd6_llinfo_settimer(ln, (long)ndi->retrans * hz / 1000);
494 			nd6_ns_output(ifp, &dst->sin6_addr,
495 			    &dst->sin6_addr, ln, 0);
496 		} else {
497 			ln->ln_state = ND6_LLINFO_STALE; /* XXX */
498 			nd6_llinfo_settimer(ln, (long)nd6_gctimer * hz);
499 		}
500 		break;
501 	case ND6_LLINFO_PROBE:
502 		if (ln->ln_asked < nd6_umaxtries) {
503 			ln->ln_asked++;
504 			nd6_llinfo_settimer(ln, (long)ndi->retrans * hz / 1000);
505 			nd6_ns_output(ifp, &dst->sin6_addr,
506 			    &dst->sin6_addr, ln, 0);
507 		} else {
508 			(void)nd6_free(rt, 0);
509 			ln = NULL;
510 		}
511 		break;
512 	}
513 
514 	KERNEL_UNLOCK_ONE(NULL);
515 	mutex_exit(softnet_lock);
516 }
517 
518 /*
519  * ND6 timer routine to expire default route list and prefix list
520  */
521 void
522 nd6_timer(void *ignored_arg)
523 {
524 	struct nd_defrouter *next_dr, *dr;
525 	struct nd_prefix *next_pr, *pr;
526 	struct in6_ifaddr *ia6, *nia6;
527 	struct in6_addrlifetime *lt6;
528 
529 	callout_reset(&nd6_timer_ch, nd6_prune * hz,
530 	    nd6_timer, NULL);
531 
532 	mutex_enter(softnet_lock);
533 	KERNEL_LOCK(1, NULL);
534 
535 	/* expire default router list */
536 
537 	for (dr = TAILQ_FIRST(&nd_defrouter); dr != NULL; dr = next_dr) {
538 		next_dr = TAILQ_NEXT(dr, dr_entry);
539 		if (dr->expire && dr->expire < time_second) {
540 			defrtrlist_del(dr);
541 		}
542 	}
543 
544 	/*
545 	 * expire interface addresses.
546 	 * in the past the loop was inside prefix expiry processing.
547 	 * However, from a stricter speci-confrmance standpoint, we should
548 	 * rather separate address lifetimes and prefix lifetimes.
549 	 */
550   addrloop:
551 	for (ia6 = in6_ifaddr; ia6; ia6 = nia6) {
552 		nia6 = ia6->ia_next;
553 		/* check address lifetime */
554 		lt6 = &ia6->ia6_lifetime;
555 		if (IFA6_IS_INVALID(ia6)) {
556 			int regen = 0;
557 
558 			/*
559 			 * If the expiring address is temporary, try
560 			 * regenerating a new one.  This would be useful when
561 			 * we suspended a laptop PC, then turned it on after a
562 			 * period that could invalidate all temporary
563 			 * addresses.  Although we may have to restart the
564 			 * loop (see below), it must be after purging the
565 			 * address.  Otherwise, we'd see an infinite loop of
566 			 * regeneration.
567 			 */
568 			if (ip6_use_tempaddr &&
569 			    (ia6->ia6_flags & IN6_IFF_TEMPORARY) != 0) {
570 				if (regen_tmpaddr(ia6) == 0)
571 					regen = 1;
572 			}
573 
574  			in6_purgeaddr(&ia6->ia_ifa);
575 
576 			if (regen)
577 				goto addrloop; /* XXX: see below */
578 		} else if (IFA6_IS_DEPRECATED(ia6)) {
579 			int oldflags = ia6->ia6_flags;
580 
581  			ia6->ia6_flags |= IN6_IFF_DEPRECATED;
582 
583 			/*
584 			 * If a temporary address has just become deprecated,
585 			 * regenerate a new one if possible.
586 			 */
587 			if (ip6_use_tempaddr &&
588 			    (ia6->ia6_flags & IN6_IFF_TEMPORARY) != 0 &&
589 			    (oldflags & IN6_IFF_DEPRECATED) == 0) {
590 
591 				if (regen_tmpaddr(ia6) == 0) {
592 					/*
593 					 * A new temporary address is
594 					 * generated.
595 					 * XXX: this means the address chain
596 					 * has changed while we are still in
597 					 * the loop.  Although the change
598 					 * would not cause disaster (because
599 					 * it's not a deletion, but an
600 					 * addition,) we'd rather restart the
601 					 * loop just for safety.  Or does this
602 					 * significantly reduce performance??
603 					 */
604 					goto addrloop;
605 				}
606 			}
607 		} else {
608 			/*
609 			 * A new RA might have made a deprecated address
610 			 * preferred.
611 			 */
612 			ia6->ia6_flags &= ~IN6_IFF_DEPRECATED;
613 		}
614 	}
615 
616 	/* expire prefix list */
617 	for (pr = LIST_FIRST(&nd_prefix); pr != NULL; pr = next_pr) {
618 		next_pr = LIST_NEXT(pr, ndpr_entry);
619 		/*
620 		 * check prefix lifetime.
621 		 * since pltime is just for autoconf, pltime processing for
622 		 * prefix is not necessary.
623 		 */
624 		if (pr->ndpr_vltime != ND6_INFINITE_LIFETIME &&
625 		    time_second - pr->ndpr_lastupdate > pr->ndpr_vltime) {
626 
627 			/*
628 			 * address expiration and prefix expiration are
629 			 * separate.  NEVER perform in6_purgeaddr here.
630 			 */
631 
632 			prelist_remove(pr);
633 		}
634 	}
635 
636 	KERNEL_UNLOCK_ONE(NULL);
637 	mutex_exit(softnet_lock);
638 }
639 
640 /* ia6: deprecated/invalidated temporary address */
641 static int
642 regen_tmpaddr(struct in6_ifaddr *ia6)
643 {
644 	struct ifaddr *ifa;
645 	struct ifnet *ifp;
646 	struct in6_ifaddr *public_ifa6 = NULL;
647 
648 	ifp = ia6->ia_ifa.ifa_ifp;
649 	IFADDR_FOREACH(ifa, ifp) {
650 		struct in6_ifaddr *it6;
651 
652 		if (ifa->ifa_addr->sa_family != AF_INET6)
653 			continue;
654 
655 		it6 = (struct in6_ifaddr *)ifa;
656 
657 		/* ignore no autoconf addresses. */
658 		if ((it6->ia6_flags & IN6_IFF_AUTOCONF) == 0)
659 			continue;
660 
661 		/* ignore autoconf addresses with different prefixes. */
662 		if (it6->ia6_ndpr == NULL || it6->ia6_ndpr != ia6->ia6_ndpr)
663 			continue;
664 
665 		/*
666 		 * Now we are looking at an autoconf address with the same
667 		 * prefix as ours.  If the address is temporary and is still
668 		 * preferred, do not create another one.  It would be rare, but
669 		 * could happen, for example, when we resume a laptop PC after
670 		 * a long period.
671 		 */
672 		if ((it6->ia6_flags & IN6_IFF_TEMPORARY) != 0 &&
673 		    !IFA6_IS_DEPRECATED(it6)) {
674 			public_ifa6 = NULL;
675 			break;
676 		}
677 
678 		/*
679 		 * This is a public autoconf address that has the same prefix
680 		 * as ours.  If it is preferred, keep it.  We can't break the
681 		 * loop here, because there may be a still-preferred temporary
682 		 * address with the prefix.
683 		 */
684 		if (!IFA6_IS_DEPRECATED(it6))
685 		    public_ifa6 = it6;
686 	}
687 
688 	if (public_ifa6 != NULL) {
689 		int e;
690 
691 		/*
692 		 * Random factor is introduced in the preferred lifetime, so
693 		 * we do not need additional delay (3rd arg to in6_tmpifadd).
694 		 */
695 		if ((e = in6_tmpifadd(public_ifa6, 0, 0)) != 0) {
696 			log(LOG_NOTICE, "regen_tmpaddr: failed to create a new"
697 			    " tmp addr, errno=%d\n", e);
698 			return -1;
699 		}
700 		return 0;
701 	}
702 
703 	return -1;
704 }
705 
706 /*
707  * Nuke neighbor cache/prefix/default router management table, right before
708  * ifp goes away.
709  */
710 void
711 nd6_purge(struct ifnet *ifp)
712 {
713 	struct llinfo_nd6 *ln, *nln;
714 	struct nd_defrouter *dr, *ndr;
715 	struct nd_prefix *pr, *npr;
716 
717 	/*
718 	 * Nuke default router list entries toward ifp.
719 	 * We defer removal of default router list entries that is installed
720 	 * in the routing table, in order to keep additional side effects as
721 	 * small as possible.
722 	 */
723 	for (dr = TAILQ_FIRST(&nd_defrouter); dr != NULL; dr = ndr) {
724 		ndr = TAILQ_NEXT(dr, dr_entry);
725 		if (dr->installed)
726 			continue;
727 
728 		if (dr->ifp == ifp)
729 			defrtrlist_del(dr);
730 	}
731 	for (dr = TAILQ_FIRST(&nd_defrouter); dr != NULL; dr = ndr) {
732 		ndr = TAILQ_NEXT(dr, dr_entry);
733 		if (!dr->installed)
734 			continue;
735 
736 		if (dr->ifp == ifp)
737 			defrtrlist_del(dr);
738 	}
739 
740 	/* Nuke prefix list entries toward ifp */
741 	for (pr = LIST_FIRST(&nd_prefix); pr != NULL; pr = npr) {
742 		npr = LIST_NEXT(pr, ndpr_entry);
743 		if (pr->ndpr_ifp == ifp) {
744 			/*
745 			 * Because if_detach() does *not* release prefixes
746 			 * while purging addresses the reference count will
747 			 * still be above zero. We therefore reset it to
748 			 * make sure that the prefix really gets purged.
749 			 */
750 			pr->ndpr_refcnt = 0;
751 			/*
752 			 * Previously, pr->ndpr_addr is removed as well,
753 			 * but I strongly believe we don't have to do it.
754 			 * nd6_purge() is only called from in6_ifdetach(),
755 			 * which removes all the associated interface addresses
756 			 * by itself.
757 			 * (jinmei@kame.net 20010129)
758 			 */
759 			prelist_remove(pr);
760 		}
761 	}
762 
763 	/* cancel default outgoing interface setting */
764 	if (nd6_defifindex == ifp->if_index)
765 		nd6_setdefaultiface(0);
766 
767 	if (!ip6_forwarding && ip6_accept_rtadv) { /* XXX: too restrictive? */
768 		/* refresh default router list */
769 		defrouter_select();
770 	}
771 
772 	/*
773 	 * Nuke neighbor cache entries for the ifp.
774 	 * Note that rt->rt_ifp may not be the same as ifp,
775 	 * due to KAME goto ours hack.  See RTM_RESOLVE case in
776 	 * nd6_rtrequest(), and ip6_input().
777 	 */
778 	ln = llinfo_nd6.ln_next;
779 	while (ln != NULL && ln != &llinfo_nd6) {
780 		struct rtentry *rt;
781 		const struct sockaddr_dl *sdl;
782 
783 		nln = ln->ln_next;
784 		rt = ln->ln_rt;
785 		if (rt && rt->rt_gateway &&
786 		    rt->rt_gateway->sa_family == AF_LINK) {
787 			sdl = satocsdl(rt->rt_gateway);
788 			if (sdl->sdl_index == ifp->if_index)
789 				nln = nd6_free(rt, 0);
790 		}
791 		ln = nln;
792 	}
793 }
794 
795 struct rtentry *
796 nd6_lookup(const struct in6_addr *addr6, int create, struct ifnet *ifp)
797 {
798 	struct rtentry *rt;
799 	struct sockaddr_in6 sin6;
800 
801 	sockaddr_in6_init(&sin6, addr6, 0, 0, 0);
802 	rt = rtalloc1((struct sockaddr *)&sin6, create);
803 	if (rt != NULL && (rt->rt_flags & RTF_LLINFO) == 0) {
804 		/*
805 		 * This is the case for the default route.
806 		 * If we want to create a neighbor cache for the address, we
807 		 * should free the route for the destination and allocate an
808 		 * interface route.
809 		 */
810 		if (create) {
811 			RTFREE(rt);
812 			rt = NULL;
813 		}
814 	}
815 	if (rt != NULL)
816 		;
817 	else 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, (const struct sockaddr *)&sin6,
839 		    ifa->ifa_addr, (const struct sockaddr *)&all1_sa,
840 		    (ifa->ifa_flags | RTF_HOST | RTF_LLINFO) &
841 		    ~RTF_CLONING, &rt)) != 0) {
842 #if 0
843 			log(LOG_ERR,
844 			    "nd6_lookup: failed to add route for a "
845 			    "neighbor(%s), errno=%d\n",
846 			    ip6_sprintf(addr6), e);
847 #endif
848 			return NULL;
849 		}
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 	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 	 * Note also that ifa_ifp and ifp may differ when we connect two
872 	 * interfaces to a same link, install a link prefix to an interface,
873 	 * and try to install a neighbor cache on an interface that does not
874 	 * have a route to the prefix.
875 	 */
876 	if ((rt->rt_flags & RTF_GATEWAY) || (rt->rt_flags & RTF_LLINFO) == 0 ||
877 	    rt->rt_gateway->sa_family != AF_LINK || rt->rt_llinfo == NULL ||
878 	    (ifp && rt->rt_ifa->ifa_ifp != ifp)) {
879 		if (create) {
880 			nd6log((LOG_DEBUG,
881 			    "nd6_lookup: failed to lookup %s (if = %s)\n",
882 			    ip6_sprintf(addr6),
883 			    ifp ? if_name(ifp) : "unspec"));
884 		}
885 		return NULL;
886 	}
887 	return rt;
888 }
889 
890 /*
891  * Detect if a given IPv6 address identifies a neighbor on a given link.
892  * XXX: should take care of the destination of a p2p link?
893  */
894 int
895 nd6_is_addr_neighbor(const struct sockaddr_in6 *addr, struct ifnet *ifp)
896 {
897 	struct nd_prefix *pr;
898 
899 	/*
900 	 * A link-local address is always a neighbor.
901 	 * XXX: a link does not necessarily specify a single interface.
902 	 */
903 	if (IN6_IS_ADDR_LINKLOCAL(&addr->sin6_addr)) {
904 		struct sockaddr_in6 sin6_copy;
905 		u_int32_t zone;
906 
907 		/*
908 		 * We need sin6_copy since sa6_recoverscope() may modify the
909 		 * content (XXX).
910 		 */
911 		sin6_copy = *addr;
912 		if (sa6_recoverscope(&sin6_copy))
913 			return 0; /* XXX: should be impossible */
914 		if (in6_setscope(&sin6_copy.sin6_addr, ifp, &zone))
915 			return 0;
916 		if (sin6_copy.sin6_scope_id == zone)
917 			return 1;
918 		else
919 			return 0;
920 	}
921 
922 	/*
923 	 * If the address matches one of our on-link prefixes, it should be a
924 	 * neighbor.
925 	 */
926 	LIST_FOREACH(pr, &nd_prefix, ndpr_entry) {
927 		if (pr->ndpr_ifp != ifp)
928 			continue;
929 
930 		if (!(pr->ndpr_stateflags & NDPRF_ONLINK))
931 			continue;
932 
933 		if (IN6_ARE_MASKED_ADDR_EQUAL(&pr->ndpr_prefix.sin6_addr,
934 		    &addr->sin6_addr, &pr->ndpr_mask))
935 			return 1;
936 	}
937 
938 	/*
939 	 * If the default router list is empty, all addresses are regarded
940 	 * as on-link, and thus, as a neighbor.
941 	 * XXX: we restrict the condition to hosts, because routers usually do
942 	 * not have the "default router list".
943 	 */
944 	if (!ip6_forwarding && TAILQ_FIRST(&nd_defrouter) == NULL &&
945 	    nd6_defifindex == ifp->if_index) {
946 		return 1;
947 	}
948 
949 	/*
950 	 * Even if the address matches none of our addresses, it might be
951 	 * in the neighbor cache.
952 	 */
953 	if (nd6_lookup(&addr->sin6_addr, 0, ifp) != NULL)
954 		return 1;
955 
956 	return 0;
957 }
958 
959 /*
960  * Free an nd6 llinfo entry.
961  * Since the function would cause significant changes in the kernel, DO NOT
962  * make it global, unless you have a strong reason for the change, and are sure
963  * that the change is safe.
964  */
965 static struct llinfo_nd6 *
966 nd6_free(struct rtentry *rt, int gc)
967 {
968 	struct llinfo_nd6 *ln = (struct llinfo_nd6 *)rt->rt_llinfo, *next;
969 	struct in6_addr in6 = satocsin6(rt_getkey(rt))->sin6_addr;
970 	struct nd_defrouter *dr;
971 
972 	/*
973 	 * we used to have pfctlinput(PRC_HOSTDEAD) here.
974 	 * even though it is not harmful, it was not really necessary.
975 	 */
976 
977 	/* cancel timer */
978 	nd6_llinfo_settimer(ln, -1);
979 
980 	if (!ip6_forwarding) {
981 		int s;
982 		s = splsoftnet();
983 		dr = defrouter_lookup(&satocsin6(rt_getkey(rt))->sin6_addr,
984 		    rt->rt_ifp);
985 
986 		if (dr != NULL && dr->expire &&
987 		    ln->ln_state == ND6_LLINFO_STALE && gc) {
988 			/*
989 			 * If the reason for the deletion is just garbage
990 			 * collection, and the neighbor is an active default
991 			 * router, do not delete it.  Instead, reset the GC
992 			 * timer using the router's lifetime.
993 			 * Simply deleting the entry would affect default
994 			 * router selection, which is not necessarily a good
995 			 * thing, especially when we're using router preference
996 			 * values.
997 			 * XXX: the check for ln_state would be redundant,
998 			 *      but we intentionally keep it just in case.
999 			 */
1000 			if (dr->expire > time_second)
1001 				nd6_llinfo_settimer(ln,
1002 				    (dr->expire - time_second) * hz);
1003 			else
1004 				nd6_llinfo_settimer(ln, (long)nd6_gctimer * hz);
1005 			splx(s);
1006 			return ln->ln_next;
1007 		}
1008 
1009 		if (ln->ln_router || dr) {
1010 			/*
1011 			 * rt6_flush must be called whether or not the neighbor
1012 			 * is in the Default Router List.
1013 			 * See a corresponding comment in nd6_na_input().
1014 			 */
1015 			rt6_flush(&in6, rt->rt_ifp);
1016 		}
1017 
1018 		if (dr) {
1019 			/*
1020 			 * Unreachablity of a router might affect the default
1021 			 * router selection and on-link detection of advertised
1022 			 * prefixes.
1023 			 */
1024 
1025 			/*
1026 			 * Temporarily fake the state to choose a new default
1027 			 * router and to perform on-link determination of
1028 			 * prefixes correctly.
1029 			 * Below the state will be set correctly,
1030 			 * or the entry itself will be deleted.
1031 			 */
1032 			ln->ln_state = ND6_LLINFO_INCOMPLETE;
1033 
1034 			/*
1035 			 * Since defrouter_select() does not affect the
1036 			 * on-link determination and MIP6 needs the check
1037 			 * before the default router selection, we perform
1038 			 * the check now.
1039 			 */
1040 			pfxlist_onlink_check();
1041 
1042 			/*
1043 			 * refresh default router list
1044 			 */
1045 			defrouter_select();
1046 		}
1047 		splx(s);
1048 	}
1049 
1050 	/*
1051 	 * Before deleting the entry, remember the next entry as the
1052 	 * return value.  We need this because pfxlist_onlink_check() above
1053 	 * might have freed other entries (particularly the old next entry) as
1054 	 * a side effect (XXX).
1055 	 */
1056 	next = ln->ln_next;
1057 
1058 	/*
1059 	 * Detach the route from the routing tree and the list of neighbor
1060 	 * caches, and disable the route entry not to be used in already
1061 	 * cached routes.
1062 	 */
1063 	rtrequest(RTM_DELETE, rt_getkey(rt), NULL, rt_mask(rt), 0, NULL);
1064 
1065 	return next;
1066 }
1067 
1068 /*
1069  * Upper-layer reachability hint for Neighbor Unreachability Detection.
1070  *
1071  * XXX cost-effective methods?
1072  */
1073 void
1074 nd6_nud_hint(struct rtentry *rt, struct in6_addr *dst6, int force)
1075 {
1076 	struct llinfo_nd6 *ln;
1077 
1078 	/*
1079 	 * If the caller specified "rt", use that.  Otherwise, resolve the
1080 	 * routing table by supplied "dst6".
1081 	 */
1082 	if (rt == NULL) {
1083 		if (dst6 == NULL)
1084 			return;
1085 		if ((rt = nd6_lookup(dst6, 0, NULL)) == NULL)
1086 			return;
1087 	}
1088 
1089 	if ((rt->rt_flags & RTF_GATEWAY) != 0 ||
1090 	    (rt->rt_flags & RTF_LLINFO) == 0 ||
1091 	    !rt->rt_llinfo || !rt->rt_gateway ||
1092 	    rt->rt_gateway->sa_family != AF_LINK) {
1093 		/* This is not a host route. */
1094 		return;
1095 	}
1096 
1097 	ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1098 	if (ln->ln_state < ND6_LLINFO_REACHABLE)
1099 		return;
1100 
1101 	/*
1102 	 * if we get upper-layer reachability confirmation many times,
1103 	 * it is possible we have false information.
1104 	 */
1105 	if (!force) {
1106 		ln->ln_byhint++;
1107 		if (ln->ln_byhint > nd6_maxnudhint)
1108 			return;
1109 	}
1110 
1111 	ln->ln_state = ND6_LLINFO_REACHABLE;
1112 	if (!ND6_LLINFO_PERMANENT(ln)) {
1113 		nd6_llinfo_settimer(ln,
1114 		    (long)ND_IFINFO(rt->rt_ifp)->reachable * hz);
1115 	}
1116 }
1117 
1118 void
1119 nd6_rtrequest(int req, struct rtentry *rt, const struct rt_addrinfo *info)
1120 {
1121 	struct sockaddr *gate = rt->rt_gateway;
1122 	struct llinfo_nd6 *ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1123 	struct ifnet *ifp = rt->rt_ifp;
1124 	uint8_t namelen = strlen(ifp->if_xname), addrlen = ifp->if_addrlen;
1125 	struct ifaddr *ifa;
1126 
1127 	RT_DPRINTF("rt->_rt_key = %p\n", (void *)rt->_rt_key);
1128 
1129 	if (req == RTM_LLINFO_UPD) {
1130 		int rc;
1131 		struct in6_addr *in6;
1132 		struct in6_addr in6_all;
1133 		int anycast;
1134 
1135 		if ((ifa = info->rti_ifa) == NULL)
1136 			return;
1137 
1138 		in6 = &ifatoia6(ifa)->ia_addr.sin6_addr;
1139 		anycast = ifatoia6(ifa)->ia6_flags & IN6_IFF_ANYCAST;
1140 
1141 		in6_all = in6addr_linklocal_allnodes;
1142 		if ((rc = in6_setscope(&in6_all, ifa->ifa_ifp, NULL)) != 0) {
1143 			log(LOG_ERR, "%s: failed to set scope %s "
1144 			    "(errno=%d)\n", __func__, if_name(ifp), rc);
1145 			return;
1146 		}
1147 
1148 		/* XXX don't set Override for proxy addresses */
1149 		nd6_na_output(ifa->ifa_ifp, &in6_all, in6,
1150 		    (anycast ? 0 : ND_NA_FLAG_OVERRIDE)
1151 #if 0
1152 		    | (ip6_forwarding ? ND_NA_FLAG_ROUTER : 0)
1153 #endif
1154 		    , 1, NULL);
1155 		return;
1156 	}
1157 
1158 	if ((rt->rt_flags & RTF_GATEWAY) != 0)
1159 		return;
1160 
1161 	if (nd6_need_cache(ifp) == 0 && (rt->rt_flags & RTF_HOST) == 0) {
1162 		RT_DPRINTF("rt->_rt_key = %p\n", (void *)rt->_rt_key);
1163 		/*
1164 		 * This is probably an interface direct route for a link
1165 		 * which does not need neighbor caches (e.g. fe80::%lo0/64).
1166 		 * We do not need special treatment below for such a route.
1167 		 * Moreover, the RTF_LLINFO flag which would be set below
1168 		 * would annoy the ndp(8) command.
1169 		 */
1170 		return;
1171 	}
1172 
1173 	if (req == RTM_RESOLVE &&
1174 	    (nd6_need_cache(ifp) == 0 || /* stf case */
1175 	     !nd6_is_addr_neighbor(satocsin6(rt_getkey(rt)), ifp))) {
1176 		RT_DPRINTF("rt->_rt_key = %p\n", (void *)rt->_rt_key);
1177 		/*
1178 		 * FreeBSD and BSD/OS often make a cloned host route based
1179 		 * on a less-specific route (e.g. the default route).
1180 		 * If the less specific route does not have a "gateway"
1181 		 * (this is the case when the route just goes to a p2p or an
1182 		 * stf interface), we'll mistakenly make a neighbor cache for
1183 		 * the host route, and will see strange neighbor solicitation
1184 		 * for the corresponding destination.  In order to avoid the
1185 		 * confusion, we check if the destination of the route is
1186 		 * a neighbor in terms of neighbor discovery, and stop the
1187 		 * process if not.  Additionally, we remove the LLINFO flag
1188 		 * so that ndp(8) will not try to get the neighbor information
1189 		 * of the destination.
1190 		 */
1191 		rt->rt_flags &= ~RTF_LLINFO;
1192 		return;
1193 	}
1194 
1195 	switch (req) {
1196 	case RTM_ADD:
1197 		RT_DPRINTF("rt->_rt_key = %p\n", (void *)rt->_rt_key);
1198 		/*
1199 		 * There is no backward compatibility :)
1200 		 *
1201 		 * if ((rt->rt_flags & RTF_HOST) == 0 &&
1202 		 *     SIN(rt_mask(rt))->sin_addr.s_addr != 0xffffffff)
1203 		 *	   rt->rt_flags |= RTF_CLONING;
1204 		 */
1205 		if ((rt->rt_flags & RTF_CLONING) ||
1206 		    ((rt->rt_flags & RTF_LLINFO) && ln == NULL)) {
1207 			union {
1208 				struct sockaddr sa;
1209 				struct sockaddr_dl sdl;
1210 				struct sockaddr_storage ss;
1211 			} u;
1212 			/*
1213 			 * Case 1: This route should come from a route to
1214 			 * interface (RTF_CLONING case) or the route should be
1215 			 * treated as on-link but is currently not
1216 			 * (RTF_LLINFO && ln == NULL case).
1217 			 */
1218 			sockaddr_dl_init(&u.sdl, sizeof(u.ss),
1219 			    ifp->if_index, ifp->if_type,
1220 			    NULL, namelen, NULL, addrlen);
1221 			rt_setgate(rt, &u.sa);
1222 			gate = rt->rt_gateway;
1223 			RT_DPRINTF("rt->_rt_key = %p\n", (void *)rt->_rt_key);
1224 			if (ln != NULL)
1225 				nd6_llinfo_settimer(ln, 0);
1226 			RT_DPRINTF("rt->_rt_key = %p\n", (void *)rt->_rt_key);
1227 			if ((rt->rt_flags & RTF_CLONING) != 0)
1228 				break;
1229 		}
1230 		RT_DPRINTF("rt->_rt_key = %p\n", (void *)rt->_rt_key);
1231 		/*
1232 		 * In IPv4 code, we try to annonuce new RTF_ANNOUNCE entry here.
1233 		 * We don't do that here since llinfo is not ready yet.
1234 		 *
1235 		 * There are also couple of other things to be discussed:
1236 		 * - unsolicited NA code needs improvement beforehand
1237 		 * - RFC2461 says we MAY send multicast unsolicited NA
1238 		 *   (7.2.6 paragraph 4), however, it also says that we
1239 		 *   SHOULD provide a mechanism to prevent multicast NA storm.
1240 		 *   we don't have anything like it right now.
1241 		 *   note that the mechanism needs a mutual agreement
1242 		 *   between proxies, which means that we need to implement
1243 		 *   a new protocol, or a new kludge.
1244 		 * - from RFC2461 6.2.4, host MUST NOT send an unsolicited NA.
1245 		 *   we need to check ip6forwarding before sending it.
1246 		 *   (or should we allow proxy ND configuration only for
1247 		 *   routers?  there's no mention about proxy ND from hosts)
1248 		 */
1249 #if 0
1250 		/* XXX it does not work */
1251 		if (rt->rt_flags & RTF_ANNOUNCE)
1252 			nd6_na_output(ifp,
1253 			      &satocsin6(rt_getkey(rt))->sin6_addr,
1254 			      &satocsin6(rt_getkey(rt))->sin6_addr,
1255 			      ip6_forwarding ? ND_NA_FLAG_ROUTER : 0,
1256 			      1, NULL);
1257 #endif
1258 		/* FALLTHROUGH */
1259 	case RTM_RESOLVE:
1260 		if ((ifp->if_flags & (IFF_POINTOPOINT | IFF_LOOPBACK)) == 0) {
1261 			RT_DPRINTF("rt->_rt_key = %p\n", (void *)rt->_rt_key);
1262 			/*
1263 			 * Address resolution isn't necessary for a point to
1264 			 * point link, so we can skip this test for a p2p link.
1265 			 */
1266 			if (gate->sa_family != AF_LINK ||
1267 			    gate->sa_len <
1268 			    sockaddr_dl_measure(namelen, addrlen)) {
1269 				log(LOG_DEBUG,
1270 				    "nd6_rtrequest: bad gateway value: %s\n",
1271 				    if_name(ifp));
1272 				break;
1273 			}
1274 			satosdl(gate)->sdl_type = ifp->if_type;
1275 			satosdl(gate)->sdl_index = ifp->if_index;
1276 			RT_DPRINTF("rt->_rt_key = %p\n", (void *)rt->_rt_key);
1277 		}
1278 		if (ln != NULL)
1279 			break;	/* This happens on a route change */
1280 		RT_DPRINTF("rt->_rt_key = %p\n", (void *)rt->_rt_key);
1281 		/*
1282 		 * Case 2: This route may come from cloning, or a manual route
1283 		 * add with a LL address.
1284 		 */
1285 		R_Malloc(ln, struct llinfo_nd6 *, sizeof(*ln));
1286 		rt->rt_llinfo = ln;
1287 		RT_DPRINTF("rt->_rt_key = %p\n", (void *)rt->_rt_key);
1288 		if (ln == NULL) {
1289 			log(LOG_DEBUG, "nd6_rtrequest: malloc failed\n");
1290 			break;
1291 		}
1292 		RT_DPRINTF("rt->_rt_key = %p\n", (void *)rt->_rt_key);
1293 		nd6_inuse++;
1294 		nd6_allocated++;
1295 		memset(ln, 0, sizeof(*ln));
1296 		ln->ln_rt = rt;
1297 		callout_init(&ln->ln_timer_ch, CALLOUT_MPSAFE);
1298 		/* this is required for "ndp" command. - shin */
1299 		if (req == RTM_ADD) {
1300 		        /*
1301 			 * gate should have some valid AF_LINK entry,
1302 			 * and ln->ln_expire should have some lifetime
1303 			 * which is specified by ndp command.
1304 			 */
1305 			ln->ln_state = ND6_LLINFO_REACHABLE;
1306 			ln->ln_byhint = 0;
1307 		} else {
1308 		        /*
1309 			 * When req == RTM_RESOLVE, rt is created and
1310 			 * initialized in rtrequest(), so rt_expire is 0.
1311 			 */
1312 			ln->ln_state = ND6_LLINFO_NOSTATE;
1313 			nd6_llinfo_settimer(ln, 0);
1314 		}
1315 		RT_DPRINTF("rt->_rt_key = %p\n", (void *)rt->_rt_key);
1316 		rt->rt_flags |= RTF_LLINFO;
1317 		ln->ln_next = llinfo_nd6.ln_next;
1318 		llinfo_nd6.ln_next = ln;
1319 		ln->ln_prev = &llinfo_nd6;
1320 		ln->ln_next->ln_prev = ln;
1321 
1322 		RT_DPRINTF("rt->_rt_key = %p\n", (void *)rt->_rt_key);
1323 		/*
1324 		 * check if rt_getkey(rt) is an address assigned
1325 		 * to the interface.
1326 		 */
1327 		ifa = (struct ifaddr *)in6ifa_ifpwithaddr(ifp,
1328 		    &satocsin6(rt_getkey(rt))->sin6_addr);
1329 		RT_DPRINTF("rt->_rt_key = %p\n", (void *)rt->_rt_key);
1330 		if (ifa != NULL) {
1331 			const void *mac;
1332 			nd6_llinfo_settimer(ln, -1);
1333 			ln->ln_state = ND6_LLINFO_REACHABLE;
1334 			ln->ln_byhint = 0;
1335 			if ((mac = nd6_ifptomac(ifp)) != NULL) {
1336 				/* XXX check for error */
1337 				(void)sockaddr_dl_setaddr(satosdl(gate),
1338 				    gate->sa_len, mac, ifp->if_addrlen);
1339 			}
1340 			if (nd6_useloopback) {
1341 				ifp = rt->rt_ifp = lo0ifp;	/* XXX */
1342 				/*
1343 				 * Make sure rt_ifa be equal to the ifaddr
1344 				 * corresponding to the address.
1345 				 * We need this because when we refer
1346 				 * rt_ifa->ia6_flags in ip6_input, we assume
1347 				 * that the rt_ifa points to the address instead
1348 				 * of the loopback address.
1349 				 */
1350 				if (ifa != rt->rt_ifa)
1351 					rt_replace_ifa(rt, ifa);
1352 				rt->rt_flags &= ~RTF_CLONED;
1353 			}
1354 		} else if (rt->rt_flags & RTF_ANNOUNCE) {
1355 			nd6_llinfo_settimer(ln, -1);
1356 			ln->ln_state = ND6_LLINFO_REACHABLE;
1357 			ln->ln_byhint = 0;
1358 
1359 			/* join solicited node multicast for proxy ND */
1360 			if (ifp->if_flags & IFF_MULTICAST) {
1361 				struct in6_addr llsol;
1362 				int error;
1363 
1364 				llsol = satocsin6(rt_getkey(rt))->sin6_addr;
1365 				llsol.s6_addr32[0] = htonl(0xff020000);
1366 				llsol.s6_addr32[1] = 0;
1367 				llsol.s6_addr32[2] = htonl(1);
1368 				llsol.s6_addr8[12] = 0xff;
1369 				if (in6_setscope(&llsol, ifp, NULL))
1370 					break;
1371 				if (!in6_addmulti(&llsol, ifp, &error, 0)) {
1372 					nd6log((LOG_ERR, "%s: failed to join "
1373 					    "%s (errno=%d)\n", if_name(ifp),
1374 					    ip6_sprintf(&llsol), error));
1375 				}
1376 			}
1377 		}
1378 		break;
1379 
1380 	case RTM_DELETE:
1381 		if (ln == NULL)
1382 			break;
1383 		/* leave from solicited node multicast for proxy ND */
1384 		if ((rt->rt_flags & RTF_ANNOUNCE) != 0 &&
1385 		    (ifp->if_flags & IFF_MULTICAST) != 0) {
1386 			struct in6_addr llsol;
1387 			struct in6_multi *in6m;
1388 
1389 			llsol = satocsin6(rt_getkey(rt))->sin6_addr;
1390 			llsol.s6_addr32[0] = htonl(0xff020000);
1391 			llsol.s6_addr32[1] = 0;
1392 			llsol.s6_addr32[2] = htonl(1);
1393 			llsol.s6_addr8[12] = 0xff;
1394 			if (in6_setscope(&llsol, ifp, NULL) == 0) {
1395 				IN6_LOOKUP_MULTI(llsol, ifp, in6m);
1396 				if (in6m)
1397 					in6_delmulti(in6m);
1398 			}
1399 		}
1400 		nd6_inuse--;
1401 		ln->ln_next->ln_prev = ln->ln_prev;
1402 		ln->ln_prev->ln_next = ln->ln_next;
1403 		ln->ln_prev = NULL;
1404 		nd6_llinfo_settimer(ln, -1);
1405 		rt->rt_llinfo = 0;
1406 		rt->rt_flags &= ~RTF_LLINFO;
1407 		clear_llinfo_pqueue(ln);
1408 		Free(ln);
1409 	}
1410 }
1411 
1412 int
1413 nd6_ioctl(u_long cmd, void *data, struct ifnet *ifp)
1414 {
1415 	struct in6_drlist *drl = (struct in6_drlist *)data;
1416 	struct in6_oprlist *oprl = (struct in6_oprlist *)data;
1417 	struct in6_ndireq *ndi = (struct in6_ndireq *)data;
1418 	struct in6_nbrinfo *nbi = (struct in6_nbrinfo *)data;
1419 	struct in6_ndifreq *ndif = (struct in6_ndifreq *)data;
1420 	struct nd_defrouter *dr;
1421 	struct nd_prefix *pr;
1422 	struct rtentry *rt;
1423 	int i = 0, error = 0;
1424 	int s;
1425 
1426 	switch (cmd) {
1427 	case SIOCGDRLST_IN6:
1428 		/*
1429 		 * obsolete API, use sysctl under net.inet6.icmp6
1430 		 */
1431 		memset(drl, 0, sizeof(*drl));
1432 		s = splsoftnet();
1433 		TAILQ_FOREACH(dr, &nd_defrouter, dr_entry) {
1434 			if (i >= DRLSTSIZ)
1435 				break;
1436 			drl->defrouter[i].rtaddr = dr->rtaddr;
1437 			in6_clearscope(&drl->defrouter[i].rtaddr);
1438 
1439 			drl->defrouter[i].flags = dr->flags;
1440 			drl->defrouter[i].rtlifetime = dr->rtlifetime;
1441 			drl->defrouter[i].expire = dr->expire;
1442 			drl->defrouter[i].if_index = dr->ifp->if_index;
1443 			i++;
1444 		}
1445 		splx(s);
1446 		break;
1447 	case SIOCGPRLST_IN6:
1448 		/*
1449 		 * obsolete API, use sysctl under net.inet6.icmp6
1450 		 *
1451 		 * XXX the structure in6_prlist was changed in backward-
1452 		 * incompatible manner.  in6_oprlist is used for SIOCGPRLST_IN6,
1453 		 * in6_prlist is used for nd6_sysctl() - fill_prlist().
1454 		 */
1455 		/*
1456 		 * XXX meaning of fields, especialy "raflags", is very
1457 		 * differnet between RA prefix list and RR/static prefix list.
1458 		 * how about separating ioctls into two?
1459 		 */
1460 		memset(oprl, 0, sizeof(*oprl));
1461 		s = splsoftnet();
1462 		LIST_FOREACH(pr, &nd_prefix, ndpr_entry) {
1463 			struct nd_pfxrouter *pfr;
1464 			int j;
1465 
1466 			if (i >= PRLSTSIZ)
1467 				break;
1468 			oprl->prefix[i].prefix = pr->ndpr_prefix.sin6_addr;
1469 			oprl->prefix[i].raflags = pr->ndpr_raf;
1470 			oprl->prefix[i].prefixlen = pr->ndpr_plen;
1471 			oprl->prefix[i].vltime = pr->ndpr_vltime;
1472 			oprl->prefix[i].pltime = pr->ndpr_pltime;
1473 			oprl->prefix[i].if_index = pr->ndpr_ifp->if_index;
1474 			if (pr->ndpr_vltime == ND6_INFINITE_LIFETIME)
1475 				oprl->prefix[i].expire = 0;
1476 			else {
1477 				time_t maxexpire;
1478 
1479 				/* XXX: we assume time_t is signed. */
1480 				maxexpire = (-1) &
1481 				    ~((time_t)1 <<
1482 				    ((sizeof(maxexpire) * 8) - 1));
1483 				if (pr->ndpr_vltime <
1484 				    maxexpire - pr->ndpr_lastupdate) {
1485 					oprl->prefix[i].expire =
1486 						 pr->ndpr_lastupdate +
1487 						pr->ndpr_vltime;
1488 				} else
1489 					oprl->prefix[i].expire = maxexpire;
1490 			}
1491 
1492 			j = 0;
1493 			LIST_FOREACH(pfr, &pr->ndpr_advrtrs, pfr_entry) {
1494 				if (j < DRLSTSIZ) {
1495 #define RTRADDR oprl->prefix[i].advrtr[j]
1496 					RTRADDR = pfr->router->rtaddr;
1497 					in6_clearscope(&RTRADDR);
1498 #undef RTRADDR
1499 				}
1500 				j++;
1501 			}
1502 			oprl->prefix[i].advrtrs = j;
1503 			oprl->prefix[i].origin = PR_ORIG_RA;
1504 
1505 			i++;
1506 		}
1507 		splx(s);
1508 
1509 		break;
1510 	case OSIOCGIFINFO_IN6:
1511 #define ND	ndi->ndi
1512 		/* XXX: old ndp(8) assumes a positive value for linkmtu. */
1513 		memset(&ND, 0, sizeof(ND));
1514 		ND.linkmtu = IN6_LINKMTU(ifp);
1515 		ND.maxmtu = ND_IFINFO(ifp)->maxmtu;
1516 		ND.basereachable = ND_IFINFO(ifp)->basereachable;
1517 		ND.reachable = ND_IFINFO(ifp)->reachable;
1518 		ND.retrans = ND_IFINFO(ifp)->retrans;
1519 		ND.flags = ND_IFINFO(ifp)->flags;
1520 		ND.recalctm = ND_IFINFO(ifp)->recalctm;
1521 		ND.chlim = ND_IFINFO(ifp)->chlim;
1522 		break;
1523 	case SIOCGIFINFO_IN6:
1524 		ND = *ND_IFINFO(ifp);
1525 		break;
1526 	case SIOCSIFINFO_IN6:
1527 		/*
1528 		 * used to change host variables from userland.
1529 		 * intented for a use on router to reflect RA configurations.
1530 		 */
1531 		/* 0 means 'unspecified' */
1532 		if (ND.linkmtu != 0) {
1533 			if (ND.linkmtu < IPV6_MMTU ||
1534 			    ND.linkmtu > IN6_LINKMTU(ifp)) {
1535 				error = EINVAL;
1536 				break;
1537 			}
1538 			ND_IFINFO(ifp)->linkmtu = ND.linkmtu;
1539 		}
1540 
1541 		if (ND.basereachable != 0) {
1542 			int obasereachable = ND_IFINFO(ifp)->basereachable;
1543 
1544 			ND_IFINFO(ifp)->basereachable = ND.basereachable;
1545 			if (ND.basereachable != obasereachable)
1546 				ND_IFINFO(ifp)->reachable =
1547 				    ND_COMPUTE_RTIME(ND.basereachable);
1548 		}
1549 		if (ND.retrans != 0)
1550 			ND_IFINFO(ifp)->retrans = ND.retrans;
1551 		if (ND.chlim != 0)
1552 			ND_IFINFO(ifp)->chlim = ND.chlim;
1553 		/* FALLTHROUGH */
1554 	case SIOCSIFINFO_FLAGS:
1555 		ND_IFINFO(ifp)->flags = ND.flags;
1556 		break;
1557 #undef ND
1558 	case SIOCSNDFLUSH_IN6:	/* XXX: the ioctl name is confusing... */
1559 		/* sync kernel routing table with the default router list */
1560 		defrouter_reset();
1561 		defrouter_select();
1562 		break;
1563 	case SIOCSPFXFLUSH_IN6:
1564 	{
1565 		/* flush all the prefix advertised by routers */
1566 		struct nd_prefix *pfx, *next;
1567 
1568 		s = splsoftnet();
1569 		for (pfx = LIST_FIRST(&nd_prefix); pfx; pfx = next) {
1570 			struct in6_ifaddr *ia, *ia_next;
1571 
1572 			next = LIST_NEXT(pfx, ndpr_entry);
1573 
1574 			if (IN6_IS_ADDR_LINKLOCAL(&pfx->ndpr_prefix.sin6_addr))
1575 				continue; /* XXX */
1576 
1577 			/* do we really have to remove addresses as well? */
1578 			for (ia = in6_ifaddr; ia; ia = ia_next) {
1579 				/* ia might be removed.  keep the next ptr. */
1580 				ia_next = ia->ia_next;
1581 
1582 				if ((ia->ia6_flags & IN6_IFF_AUTOCONF) == 0)
1583 					continue;
1584 
1585 				if (ia->ia6_ndpr == pfx)
1586 					in6_purgeaddr(&ia->ia_ifa);
1587 			}
1588 			prelist_remove(pfx);
1589 		}
1590 		splx(s);
1591 		break;
1592 	}
1593 	case SIOCSRTRFLUSH_IN6:
1594 	{
1595 		/* flush all the default routers */
1596 		struct nd_defrouter *drtr, *next;
1597 
1598 		s = splsoftnet();
1599 		defrouter_reset();
1600 		for (drtr = TAILQ_FIRST(&nd_defrouter); drtr; drtr = next) {
1601 			next = TAILQ_NEXT(drtr, dr_entry);
1602 			defrtrlist_del(drtr);
1603 		}
1604 		defrouter_select();
1605 		splx(s);
1606 		break;
1607 	}
1608 	case SIOCGNBRINFO_IN6:
1609 	{
1610 		struct llinfo_nd6 *ln;
1611 		struct in6_addr nb_addr = nbi->addr; /* make local for safety */
1612 
1613 		if ((error = in6_setscope(&nb_addr, ifp, NULL)) != 0)
1614 			return error;
1615 
1616 		s = splsoftnet();
1617 		if ((rt = nd6_lookup(&nb_addr, 0, ifp)) == NULL ||
1618 		    (ln = (struct llinfo_nd6 *)rt->rt_llinfo) == NULL) {
1619 			error = EINVAL;
1620 			splx(s);
1621 			break;
1622 		}
1623 		nbi->state = ln->ln_state;
1624 		nbi->asked = ln->ln_asked;
1625 		nbi->isrouter = ln->ln_router;
1626 		nbi->expire = ln->ln_expire;
1627 		splx(s);
1628 
1629 		break;
1630 	}
1631 	case SIOCGDEFIFACE_IN6:	/* XXX: should be implemented as a sysctl? */
1632 		ndif->ifindex = nd6_defifindex;
1633 		break;
1634 	case SIOCSDEFIFACE_IN6:	/* XXX: should be implemented as a sysctl? */
1635 		return nd6_setdefaultiface(ndif->ifindex);
1636 	}
1637 	return error;
1638 }
1639 
1640 void
1641 nd6_llinfo_release_pkts(struct llinfo_nd6 *ln, struct ifnet *ifp,
1642     struct rtentry *rt)
1643 {
1644 	struct mbuf *m_hold, *m_hold_next;
1645 
1646 	for (m_hold = ln->ln_hold, ln->ln_hold = NULL;
1647 	     m_hold != NULL;
1648 	     m_hold = m_hold_next) {
1649 		m_hold_next = m_hold->m_nextpkt;
1650 		m_hold->m_nextpkt = NULL;
1651 
1652 		/*
1653 		 * we assume ifp is not a p2p here, so
1654 		 * just set the 2nd argument as the
1655 		 * 1st one.
1656 		 */
1657 		nd6_output(ifp, ifp, m_hold, satocsin6(rt_getkey(rt)), rt);
1658 	}
1659 }
1660 
1661 /*
1662  * Create neighbor cache entry and cache link-layer address,
1663  * on reception of inbound ND6 packets.  (RS/RA/NS/redirect)
1664  */
1665 struct rtentry *
1666 nd6_cache_lladdr(
1667     struct ifnet *ifp,
1668     struct in6_addr *from,
1669     char *lladdr,
1670     int lladdrlen,
1671     int type,	/* ICMP6 type */
1672     int code	/* type dependent information */
1673 )
1674 {
1675 	struct rtentry *rt = NULL;
1676 	struct llinfo_nd6 *ln = NULL;
1677 	int is_newentry;
1678 	struct sockaddr_dl *sdl = NULL;
1679 	int do_update;
1680 	int olladdr;
1681 	int llchange;
1682 	int newstate = 0;
1683 
1684 	if (ifp == NULL)
1685 		panic("ifp == NULL in nd6_cache_lladdr");
1686 	if (from == NULL)
1687 		panic("from == NULL in nd6_cache_lladdr");
1688 
1689 	/* nothing must be updated for unspecified address */
1690 	if (IN6_IS_ADDR_UNSPECIFIED(from))
1691 		return NULL;
1692 
1693 	/*
1694 	 * Validation about ifp->if_addrlen and lladdrlen must be done in
1695 	 * the caller.
1696 	 *
1697 	 * XXX If the link does not have link-layer adderss, what should
1698 	 * we do? (ifp->if_addrlen == 0)
1699 	 * Spec says nothing in sections for RA, RS and NA.  There's small
1700 	 * description on it in NS section (RFC 2461 7.2.3).
1701 	 */
1702 
1703 	rt = nd6_lookup(from, 0, ifp);
1704 	if (rt == NULL) {
1705 #if 0
1706 		/* nothing must be done if there's no lladdr */
1707 		if (!lladdr || !lladdrlen)
1708 			return NULL;
1709 #endif
1710 
1711 		rt = nd6_lookup(from, 1, ifp);
1712 		is_newentry = 1;
1713 	} else {
1714 		/* do nothing if static ndp is set */
1715 		if (rt->rt_flags & RTF_STATIC)
1716 			return NULL;
1717 		is_newentry = 0;
1718 	}
1719 
1720 	if (rt == NULL)
1721 		return NULL;
1722 	if ((rt->rt_flags & (RTF_GATEWAY | RTF_LLINFO)) != RTF_LLINFO) {
1723 fail:
1724 		(void)nd6_free(rt, 0);
1725 		return NULL;
1726 	}
1727 	ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1728 	if (ln == NULL)
1729 		goto fail;
1730 	if (rt->rt_gateway == NULL)
1731 		goto fail;
1732 	if (rt->rt_gateway->sa_family != AF_LINK)
1733 		goto fail;
1734 	sdl = satosdl(rt->rt_gateway);
1735 
1736 	olladdr = (sdl->sdl_alen) ? 1 : 0;
1737 	if (olladdr && lladdr) {
1738 		if (memcmp(lladdr, CLLADDR(sdl), ifp->if_addrlen))
1739 			llchange = 1;
1740 		else
1741 			llchange = 0;
1742 	} else
1743 		llchange = 0;
1744 
1745 	/*
1746 	 * newentry olladdr  lladdr  llchange	(*=record)
1747 	 *	0	n	n	--	(1)
1748 	 *	0	y	n	--	(2)
1749 	 *	0	n	y	--	(3) * STALE
1750 	 *	0	y	y	n	(4) *
1751 	 *	0	y	y	y	(5) * STALE
1752 	 *	1	--	n	--	(6)   NOSTATE(= PASSIVE)
1753 	 *	1	--	y	--	(7) * STALE
1754 	 */
1755 
1756 	if (lladdr) {		/* (3-5) and (7) */
1757 		/*
1758 		 * Record source link-layer address
1759 		 * XXX is it dependent to ifp->if_type?
1760 		 */
1761 		/* XXX check for error */
1762 		(void)sockaddr_dl_setaddr(sdl, sdl->sdl_len, lladdr,
1763 		    ifp->if_addrlen);
1764 	}
1765 
1766 	if (!is_newentry) {
1767 		if ((!olladdr && lladdr) ||		/* (3) */
1768 		    (olladdr && lladdr && llchange)) {	/* (5) */
1769 			do_update = 1;
1770 			newstate = ND6_LLINFO_STALE;
1771 		} else					/* (1-2,4) */
1772 			do_update = 0;
1773 	} else {
1774 		do_update = 1;
1775 		if (lladdr == NULL)			/* (6) */
1776 			newstate = ND6_LLINFO_NOSTATE;
1777 		else					/* (7) */
1778 			newstate = ND6_LLINFO_STALE;
1779 	}
1780 
1781 	if (do_update) {
1782 		/*
1783 		 * Update the state of the neighbor cache.
1784 		 */
1785 		ln->ln_state = newstate;
1786 
1787 		if (ln->ln_state == ND6_LLINFO_STALE) {
1788 			/*
1789 			 * XXX: since nd6_output() below will cause
1790 			 * state tansition to DELAY and reset the timer,
1791 			 * we must set the timer now, although it is actually
1792 			 * meaningless.
1793 			 */
1794 			nd6_llinfo_settimer(ln, (long)nd6_gctimer * hz);
1795 
1796 			nd6_llinfo_release_pkts(ln, ifp, rt);
1797 		} else if (ln->ln_state == ND6_LLINFO_INCOMPLETE) {
1798 			/* probe right away */
1799 			nd6_llinfo_settimer((void *)ln, 0);
1800 		}
1801 	}
1802 
1803 	/*
1804 	 * ICMP6 type dependent behavior.
1805 	 *
1806 	 * NS: clear IsRouter if new entry
1807 	 * RS: clear IsRouter
1808 	 * RA: set IsRouter if there's lladdr
1809 	 * redir: clear IsRouter if new entry
1810 	 *
1811 	 * RA case, (1):
1812 	 * The spec says that we must set IsRouter in the following cases:
1813 	 * - If lladdr exist, set IsRouter.  This means (1-5).
1814 	 * - If it is old entry (!newentry), set IsRouter.  This means (7).
1815 	 * So, based on the spec, in (1-5) and (7) cases we must set IsRouter.
1816 	 * A quetion arises for (1) case.  (1) case has no lladdr in the
1817 	 * neighbor cache, this is similar to (6).
1818 	 * This case is rare but we figured that we MUST NOT set IsRouter.
1819 	 *
1820 	 * newentry olladdr  lladdr  llchange	    NS  RS  RA	redir
1821 	 *							D R
1822 	 *	0	n	n	--	(1)	c   ?     s
1823 	 *	0	y	n	--	(2)	c   s     s
1824 	 *	0	n	y	--	(3)	c   s     s
1825 	 *	0	y	y	n	(4)	c   s     s
1826 	 *	0	y	y	y	(5)	c   s     s
1827 	 *	1	--	n	--	(6) c	c 	c s
1828 	 *	1	--	y	--	(7) c	c   s	c s
1829 	 *
1830 	 *					(c=clear s=set)
1831 	 */
1832 	switch (type & 0xff) {
1833 	case ND_NEIGHBOR_SOLICIT:
1834 		/*
1835 		 * New entry must have is_router flag cleared.
1836 		 */
1837 		if (is_newentry)	/* (6-7) */
1838 			ln->ln_router = 0;
1839 		break;
1840 	case ND_REDIRECT:
1841 		/*
1842 		 * If the icmp is a redirect to a better router, always set the
1843 		 * is_router flag.  Otherwise, if the entry is newly created,
1844 		 * clear the flag.  [RFC 2461, sec 8.3]
1845 		 */
1846 		if (code == ND_REDIRECT_ROUTER)
1847 			ln->ln_router = 1;
1848 		else if (is_newentry) /* (6-7) */
1849 			ln->ln_router = 0;
1850 		break;
1851 	case ND_ROUTER_SOLICIT:
1852 		/*
1853 		 * is_router flag must always be cleared.
1854 		 */
1855 		ln->ln_router = 0;
1856 		break;
1857 	case ND_ROUTER_ADVERT:
1858 		/*
1859 		 * Mark an entry with lladdr as a router.
1860 		 */
1861 		if ((!is_newentry && (olladdr || lladdr)) ||	/* (2-5) */
1862 		    (is_newentry && lladdr)) {			/* (7) */
1863 			ln->ln_router = 1;
1864 		}
1865 		break;
1866 	}
1867 
1868 	/*
1869 	 * When the link-layer address of a router changes, select the
1870 	 * best router again.  In particular, when the neighbor entry is newly
1871 	 * created, it might affect the selection policy.
1872 	 * Question: can we restrict the first condition to the "is_newentry"
1873 	 * case?
1874 	 * XXX: when we hear an RA from a new router with the link-layer
1875 	 * address option, defrouter_select() is called twice, since
1876 	 * defrtrlist_update called the function as well.  However, I believe
1877 	 * we can compromise the overhead, since it only happens the first
1878 	 * time.
1879 	 * XXX: although defrouter_select() should not have a bad effect
1880 	 * for those are not autoconfigured hosts, we explicitly avoid such
1881 	 * cases for safety.
1882 	 */
1883 	if (do_update && ln->ln_router && !ip6_forwarding && ip6_accept_rtadv)
1884 		defrouter_select();
1885 
1886 	return rt;
1887 }
1888 
1889 static void
1890 nd6_slowtimo(void *ignored_arg)
1891 {
1892 	struct nd_ifinfo *nd6if;
1893 	struct ifnet *ifp;
1894 
1895 	mutex_enter(softnet_lock);
1896 	KERNEL_LOCK(1, NULL);
1897       	callout_reset(&nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
1898 	    nd6_slowtimo, NULL);
1899 	TAILQ_FOREACH(ifp, &ifnet, if_list) {
1900 		nd6if = ND_IFINFO(ifp);
1901 		if (nd6if->basereachable && /* already initialized */
1902 		    (nd6if->recalctm -= ND6_SLOWTIMER_INTERVAL) <= 0) {
1903 			/*
1904 			 * Since reachable time rarely changes by router
1905 			 * advertisements, we SHOULD insure that a new random
1906 			 * value gets recomputed at least once every few hours.
1907 			 * (RFC 2461, 6.3.4)
1908 			 */
1909 			nd6if->recalctm = nd6_recalc_reachtm_interval;
1910 			nd6if->reachable = ND_COMPUTE_RTIME(nd6if->basereachable);
1911 		}
1912 	}
1913 	KERNEL_UNLOCK_ONE(NULL);
1914 	mutex_exit(softnet_lock);
1915 }
1916 
1917 #define senderr(e) { error = (e); goto bad;}
1918 int
1919 nd6_output(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *m0,
1920     const struct sockaddr_in6 *dst, struct rtentry *rt0)
1921 {
1922 	struct mbuf *m = m0;
1923 	struct rtentry *rt = rt0;
1924 	struct sockaddr_in6 *gw6 = NULL;
1925 	struct llinfo_nd6 *ln = NULL;
1926 	int error = 0;
1927 
1928 	if (IN6_IS_ADDR_MULTICAST(&dst->sin6_addr))
1929 		goto sendpkt;
1930 
1931 	if (nd6_need_cache(ifp) == 0)
1932 		goto sendpkt;
1933 
1934 	/*
1935 	 * next hop determination.  This routine is derived from ether_output.
1936 	 */
1937 	if (rt) {
1938 		if ((rt->rt_flags & RTF_UP) == 0) {
1939 			if ((rt0 = rt = rtalloc1(sin6tocsa(dst), 1)) != NULL) {
1940 				rt->rt_refcnt--;
1941 				if (rt->rt_ifp != ifp)
1942 					senderr(EHOSTUNREACH);
1943 			} else
1944 				senderr(EHOSTUNREACH);
1945 		}
1946 
1947 		if (rt->rt_flags & RTF_GATEWAY) {
1948 			gw6 = (struct sockaddr_in6 *)rt->rt_gateway;
1949 
1950 			/*
1951 			 * We skip link-layer address resolution and NUD
1952 			 * if the gateway is not a neighbor from ND point
1953 			 * of view, regardless of the value of nd_ifinfo.flags.
1954 			 * The second condition is a bit tricky; we skip
1955 			 * if the gateway is our own address, which is
1956 			 * sometimes used to install a route to a p2p link.
1957 			 */
1958 			if (!nd6_is_addr_neighbor(gw6, ifp) ||
1959 			    in6ifa_ifpwithaddr(ifp, &gw6->sin6_addr)) {
1960 				/*
1961 				 * We allow this kind of tricky route only
1962 				 * when the outgoing interface is p2p.
1963 				 * XXX: we may need a more generic rule here.
1964 				 */
1965 				if ((ifp->if_flags & IFF_POINTOPOINT) == 0)
1966 					senderr(EHOSTUNREACH);
1967 
1968 				goto sendpkt;
1969 			}
1970 
1971 			if (rt->rt_gwroute == NULL)
1972 				goto lookup;
1973 			if (((rt = rt->rt_gwroute)->rt_flags & RTF_UP) == 0) {
1974 				rtfree(rt); rt = rt0;
1975 			lookup:
1976 				rt->rt_gwroute = rtalloc1(rt->rt_gateway, 1);
1977 				if ((rt = rt->rt_gwroute) == NULL)
1978 					senderr(EHOSTUNREACH);
1979 				/* the "G" test below also prevents rt == rt0 */
1980 				if ((rt->rt_flags & RTF_GATEWAY) ||
1981 				    (rt->rt_ifp != ifp)) {
1982 					rt->rt_refcnt--;
1983 					rt0->rt_gwroute = NULL;
1984 					senderr(EHOSTUNREACH);
1985 				}
1986 			}
1987 		}
1988 	}
1989 
1990 	/*
1991 	 * Address resolution or Neighbor Unreachability Detection
1992 	 * for the next hop.
1993 	 * At this point, the destination of the packet must be a unicast
1994 	 * or an anycast address(i.e. not a multicast).
1995 	 */
1996 
1997 	/* Look up the neighbor cache for the nexthop */
1998 	if (rt != NULL && (rt->rt_flags & RTF_LLINFO) != 0)
1999 		ln = (struct llinfo_nd6 *)rt->rt_llinfo;
2000 	else {
2001 		/*
2002 		 * Since nd6_is_addr_neighbor() internally calls nd6_lookup(),
2003 		 * the condition below is not very efficient.  But we believe
2004 		 * it is tolerable, because this should be a rare case.
2005 		 */
2006 		if (nd6_is_addr_neighbor(dst, ifp) &&
2007 		    (rt = nd6_lookup(&dst->sin6_addr, 1, ifp)) != NULL)
2008 			ln = (struct llinfo_nd6 *)rt->rt_llinfo;
2009 	}
2010 	if (ln == NULL || rt == NULL) {
2011 		if ((ifp->if_flags & IFF_POINTOPOINT) == 0 &&
2012 		    !(ND_IFINFO(ifp)->flags & ND6_IFF_PERFORMNUD)) {
2013 			log(LOG_DEBUG,
2014 			    "nd6_output: can't allocate llinfo for %s "
2015 			    "(ln=%p, rt=%p)\n",
2016 			    ip6_sprintf(&dst->sin6_addr), ln, rt);
2017 			senderr(EIO);	/* XXX: good error? */
2018 		}
2019 
2020 		goto sendpkt;	/* send anyway */
2021 	}
2022 
2023 	/* We don't have to do link-layer address resolution on a p2p link. */
2024 	if ((ifp->if_flags & IFF_POINTOPOINT) != 0 &&
2025 	    ln->ln_state < ND6_LLINFO_REACHABLE) {
2026 		ln->ln_state = ND6_LLINFO_STALE;
2027 		nd6_llinfo_settimer(ln, (long)nd6_gctimer * hz);
2028 	}
2029 
2030 	/*
2031 	 * The first time we send a packet to a neighbor whose entry is
2032 	 * STALE, we have to change the state to DELAY and a sets a timer to
2033 	 * expire in DELAY_FIRST_PROBE_TIME seconds to ensure do
2034 	 * neighbor unreachability detection on expiration.
2035 	 * (RFC 2461 7.3.3)
2036 	 */
2037 	if (ln->ln_state == ND6_LLINFO_STALE) {
2038 		ln->ln_asked = 0;
2039 		ln->ln_state = ND6_LLINFO_DELAY;
2040 		nd6_llinfo_settimer(ln, (long)nd6_delay * hz);
2041 	}
2042 
2043 	/*
2044 	 * If the neighbor cache entry has a state other than INCOMPLETE
2045 	 * (i.e. its link-layer address is already resolved), just
2046 	 * send the packet.
2047 	 */
2048 	if (ln->ln_state > ND6_LLINFO_INCOMPLETE)
2049 		goto sendpkt;
2050 
2051 	/*
2052 	 * There is a neighbor cache entry, but no ethernet address
2053 	 * response yet.  Append this latest packet to the end of the
2054 	 * packet queue in the mbuf, unless the number of the packet
2055 	 * does not exceed nd6_maxqueuelen.  When it exceeds nd6_maxqueuelen,
2056 	 * the oldest packet in the queue will be removed.
2057 	 */
2058 	if (ln->ln_state == ND6_LLINFO_NOSTATE)
2059 		ln->ln_state = ND6_LLINFO_INCOMPLETE;
2060 	if (ln->ln_hold) {
2061 		struct mbuf *m_hold;
2062 		int i;
2063 
2064 		i = 0;
2065 		for (m_hold = ln->ln_hold; m_hold; m_hold = m_hold->m_nextpkt) {
2066 			i++;
2067 			if (m_hold->m_nextpkt == NULL) {
2068 				m_hold->m_nextpkt = m;
2069 				break;
2070 			}
2071 		}
2072 		while (i >= nd6_maxqueuelen) {
2073 			m_hold = ln->ln_hold;
2074 			ln->ln_hold = ln->ln_hold->m_nextpkt;
2075 			m_freem(m_hold);
2076 			i--;
2077 		}
2078 	} else {
2079 		ln->ln_hold = m;
2080 	}
2081 
2082 	/*
2083 	 * If there has been no NS for the neighbor after entering the
2084 	 * INCOMPLETE state, send the first solicitation.
2085 	 */
2086 	if (!ND6_LLINFO_PERMANENT(ln) && ln->ln_asked == 0) {
2087 		ln->ln_asked++;
2088 		nd6_llinfo_settimer(ln,
2089 		    (long)ND_IFINFO(ifp)->retrans * hz / 1000);
2090 		nd6_ns_output(ifp, NULL, &dst->sin6_addr, ln, 0);
2091 	}
2092 	return 0;
2093 
2094   sendpkt:
2095 	/* discard the packet if IPv6 operation is disabled on the interface */
2096 	if ((ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED)) {
2097 		error = ENETDOWN; /* better error? */
2098 		goto bad;
2099 	}
2100 
2101 #ifdef IPSEC
2102 	/* clean ipsec history once it goes out of the node */
2103 	ipsec_delaux(m);
2104 #endif
2105 	if ((ifp->if_flags & IFF_LOOPBACK) != 0)
2106 		return (*ifp->if_output)(origifp, m, sin6tocsa(dst), rt);
2107 	return (*ifp->if_output)(ifp, m, sin6tocsa(dst), rt);
2108 
2109   bad:
2110 	if (m != NULL)
2111 		m_freem(m);
2112 	return error;
2113 }
2114 #undef senderr
2115 
2116 int
2117 nd6_need_cache(struct ifnet *ifp)
2118 {
2119 	/*
2120 	 * XXX: we currently do not make neighbor cache on any interface
2121 	 * other than ARCnet, Ethernet, FDDI and GIF.
2122 	 *
2123 	 * RFC2893 says:
2124 	 * - unidirectional tunnels needs no ND
2125 	 */
2126 	switch (ifp->if_type) {
2127 	case IFT_ARCNET:
2128 	case IFT_ETHER:
2129 	case IFT_FDDI:
2130 	case IFT_IEEE1394:
2131 	case IFT_CARP:
2132 	case IFT_GIF:		/* XXX need more cases? */
2133 	case IFT_PPP:
2134 	case IFT_TUNNEL:
2135 		return 1;
2136 	default:
2137 		return 0;
2138 	}
2139 }
2140 
2141 int
2142 nd6_storelladdr(const struct ifnet *ifp, const struct rtentry *rt,
2143     struct mbuf *m, const struct sockaddr *dst, uint8_t *lldst,
2144     size_t dstsize)
2145 {
2146 	const struct sockaddr_dl *sdl;
2147 
2148 	if (m->m_flags & M_MCAST) {
2149 		switch (ifp->if_type) {
2150 		case IFT_ETHER:
2151 		case IFT_FDDI:
2152 			ETHER_MAP_IPV6_MULTICAST(&satocsin6(dst)->sin6_addr,
2153 			    lldst);
2154 			return 1;
2155 		case IFT_IEEE1394:
2156 			memcpy(lldst, ifp->if_broadcastaddr,
2157 			    MIN(dstsize, ifp->if_addrlen));
2158 			return 1;
2159 		case IFT_ARCNET:
2160 			*lldst = 0;
2161 			return 1;
2162 		default:
2163 			m_freem(m);
2164 			return 0;
2165 		}
2166 	}
2167 
2168 	if (rt == NULL) {
2169 		/* this could happen, if we could not allocate memory */
2170 		m_freem(m);
2171 		return 0;
2172 	}
2173 	if (rt->rt_gateway->sa_family != AF_LINK) {
2174 		printf("%s: something odd happens\n", __func__);
2175 		m_freem(m);
2176 		return 0;
2177 	}
2178 	sdl = satocsdl(rt->rt_gateway);
2179 	if (sdl->sdl_alen == 0 || sdl->sdl_alen > dstsize) {
2180 		/* this should be impossible, but we bark here for debugging */
2181 		printf("%s: sdl_alen == 0, dst=%s, if=%s\n", __func__,
2182 		    ip6_sprintf(&satocsin6(dst)->sin6_addr), if_name(ifp));
2183 		m_freem(m);
2184 		return 0;
2185 	}
2186 
2187 	memcpy(lldst, CLLADDR(sdl), MIN(dstsize, sdl->sdl_alen));
2188 	return 1;
2189 }
2190 
2191 static void
2192 clear_llinfo_pqueue(struct llinfo_nd6 *ln)
2193 {
2194 	struct mbuf *m_hold, *m_hold_next;
2195 
2196 	for (m_hold = ln->ln_hold; m_hold; m_hold = m_hold_next) {
2197 		m_hold_next = m_hold->m_nextpkt;
2198 		m_hold->m_nextpkt = NULL;
2199 		m_freem(m_hold);
2200 	}
2201 
2202 	ln->ln_hold = NULL;
2203 	return;
2204 }
2205 
2206 int
2207 nd6_sysctl(
2208     int name,
2209     void *oldp,	/* syscall arg, need copyout */
2210     size_t *oldlenp,
2211     void *newp,	/* syscall arg, need copyin */
2212     size_t newlen
2213 )
2214 {
2215 	void *p;
2216 	size_t ol;
2217 	int error;
2218 
2219 	error = 0;
2220 
2221 	if (newp)
2222 		return EPERM;
2223 	if (oldp && !oldlenp)
2224 		return EINVAL;
2225 	ol = oldlenp ? *oldlenp : 0;
2226 
2227 	if (oldp) {
2228 		p = malloc(*oldlenp, M_TEMP, M_WAITOK);
2229 		if (p == NULL)
2230 			return ENOMEM;
2231 	} else
2232 		p = NULL;
2233 	switch (name) {
2234 	case ICMPV6CTL_ND6_DRLIST:
2235 		error = fill_drlist(p, oldlenp, ol);
2236 		if (!error && p != NULL && oldp != NULL)
2237 			error = copyout(p, oldp, *oldlenp);
2238 		break;
2239 
2240 	case ICMPV6CTL_ND6_PRLIST:
2241 		error = fill_prlist(p, oldlenp, ol);
2242 		if (!error && p != NULL && oldp != NULL)
2243 			error = copyout(p, oldp, *oldlenp);
2244 		break;
2245 
2246 	case ICMPV6CTL_ND6_MAXQLEN:
2247 		break;
2248 
2249 	default:
2250 		error = ENOPROTOOPT;
2251 		break;
2252 	}
2253 	if (p)
2254 		free(p, M_TEMP);
2255 
2256 	return error;
2257 }
2258 
2259 static int
2260 fill_drlist(void *oldp, size_t *oldlenp, size_t ol)
2261 {
2262 	int error = 0, s;
2263 	struct in6_defrouter *d = NULL, *de = NULL;
2264 	struct nd_defrouter *dr;
2265 	size_t l;
2266 
2267 	s = splsoftnet();
2268 
2269 	if (oldp) {
2270 		d = (struct in6_defrouter *)oldp;
2271 		de = (struct in6_defrouter *)((char *)oldp + *oldlenp);
2272 	}
2273 	l = 0;
2274 
2275 	TAILQ_FOREACH(dr, &nd_defrouter, dr_entry) {
2276 
2277 		if (oldp && d + 1 <= de) {
2278 			memset(d, 0, sizeof(*d));
2279 			sockaddr_in6_init(&d->rtaddr, &dr->rtaddr, 0, 0, 0);
2280 			if (sa6_recoverscope(&d->rtaddr)) {
2281 				log(LOG_ERR,
2282 				    "scope error in router list (%s)\n",
2283 				    ip6_sprintf(&d->rtaddr.sin6_addr));
2284 				/* XXX: press on... */
2285 			}
2286 			d->flags = dr->flags;
2287 			d->rtlifetime = dr->rtlifetime;
2288 			d->expire = dr->expire;
2289 			d->if_index = dr->ifp->if_index;
2290 		}
2291 
2292 		l += sizeof(*d);
2293 		if (d)
2294 			d++;
2295 	}
2296 
2297 	if (oldp) {
2298 		if (l > ol)
2299 			error = ENOMEM;
2300 	}
2301 	if (oldlenp)
2302 		*oldlenp = l;	/* (void *)d - (void *)oldp */
2303 
2304 	splx(s);
2305 
2306 	return error;
2307 }
2308 
2309 static int
2310 fill_prlist(void *oldp, size_t *oldlenp, size_t ol)
2311 {
2312 	int error = 0, s;
2313 	struct nd_prefix *pr;
2314 	struct in6_prefix *p = NULL;
2315 	struct in6_prefix *pe = NULL;
2316 	size_t l;
2317 
2318 	s = splsoftnet();
2319 
2320 	if (oldp) {
2321 		p = (struct in6_prefix *)oldp;
2322 		pe = (struct in6_prefix *)((char *)oldp + *oldlenp);
2323 	}
2324 	l = 0;
2325 
2326 	LIST_FOREACH(pr, &nd_prefix, ndpr_entry) {
2327 		u_short advrtrs;
2328 		size_t advance;
2329 		struct sockaddr_in6 *sin6;
2330 		struct sockaddr_in6 *s6;
2331 		struct nd_pfxrouter *pfr;
2332 
2333 		if (oldp && p + 1 <= pe)
2334 		{
2335 			memset(p, 0, sizeof(*p));
2336 			sin6 = (struct sockaddr_in6 *)(p + 1);
2337 
2338 			p->prefix = pr->ndpr_prefix;
2339 			if (sa6_recoverscope(&p->prefix)) {
2340 				log(LOG_ERR,
2341 				    "scope error in prefix list (%s)\n",
2342 				    ip6_sprintf(&p->prefix.sin6_addr));
2343 				/* XXX: press on... */
2344 			}
2345 			p->raflags = pr->ndpr_raf;
2346 			p->prefixlen = pr->ndpr_plen;
2347 			p->vltime = pr->ndpr_vltime;
2348 			p->pltime = pr->ndpr_pltime;
2349 			p->if_index = pr->ndpr_ifp->if_index;
2350 			if (pr->ndpr_vltime == ND6_INFINITE_LIFETIME)
2351 				p->expire = 0;
2352 			else {
2353 				time_t maxexpire;
2354 
2355 				/* XXX: we assume time_t is signed. */
2356 				maxexpire = (-1) &
2357 				    ~((time_t)1 <<
2358 				    ((sizeof(maxexpire) * 8) - 1));
2359 				if (pr->ndpr_vltime <
2360 				    maxexpire - pr->ndpr_lastupdate) {
2361 					p->expire = pr->ndpr_lastupdate +
2362 						pr->ndpr_vltime;
2363 				} else
2364 					p->expire = maxexpire;
2365 			}
2366 			p->refcnt = pr->ndpr_refcnt;
2367 			p->flags = pr->ndpr_stateflags;
2368 			p->origin = PR_ORIG_RA;
2369 			advrtrs = 0;
2370 			LIST_FOREACH(pfr, &pr->ndpr_advrtrs, pfr_entry) {
2371 				if ((void *)&sin6[advrtrs + 1] > (void *)pe) {
2372 					advrtrs++;
2373 					continue;
2374 				}
2375 				s6 = &sin6[advrtrs];
2376 				sockaddr_in6_init(s6, &pfr->router->rtaddr,
2377 				    0, 0, 0);
2378 				if (sa6_recoverscope(s6)) {
2379 					log(LOG_ERR,
2380 					    "scope error in "
2381 					    "prefix list (%s)\n",
2382 					    ip6_sprintf(&pfr->router->rtaddr));
2383 				}
2384 				advrtrs++;
2385 			}
2386 			p->advrtrs = advrtrs;
2387 		}
2388 		else {
2389 			advrtrs = 0;
2390 			LIST_FOREACH(pfr, &pr->ndpr_advrtrs, pfr_entry)
2391 				advrtrs++;
2392 		}
2393 
2394 		advance = sizeof(*p) + sizeof(*sin6) * advrtrs;
2395 		l += advance;
2396 		if (p)
2397 			p = (struct in6_prefix *)((char *)p + advance);
2398 	}
2399 
2400 	if (oldp) {
2401 		*oldlenp = l;	/* (void *)d - (void *)oldp */
2402 		if (l > ol)
2403 			error = ENOMEM;
2404 	} else
2405 		*oldlenp = l;
2406 
2407 	splx(s);
2408 
2409 	return error;
2410 }
2411