xref: /freebsd/sys/netinet6/nd6.c (revision 9fd69f37d28cfd7438cac3eeb45fe9dd46b4d7dd)
1 /*-
2  * Copyright (C) 1995, 1996, 1997, and 1998 WIDE Project.
3  * All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice, this list of conditions and the following disclaimer.
10  * 2. Redistributions in binary form must reproduce the above copyright
11  *    notice, this list of conditions and the following disclaimer in the
12  *    documentation and/or other materials provided with the distribution.
13  * 3. Neither the name of the project nor the names of its contributors
14  *    may be used to endorse or promote products derived from this software
15  *    without specific prior written permission.
16  *
17  * THIS SOFTWARE IS PROVIDED BY THE PROJECT AND CONTRIBUTORS ``AS IS'' AND
18  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20  * ARE DISCLAIMED.  IN NO EVENT SHALL THE PROJECT OR CONTRIBUTORS BE LIABLE
21  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27  * SUCH DAMAGE.
28  *
29  *	$KAME: nd6.c,v 1.144 2001/05/24 07:44:00 itojun Exp $
30  */
31 
32 #include <sys/cdefs.h>
33 __FBSDID("$FreeBSD$");
34 
35 #include "opt_inet.h"
36 #include "opt_inet6.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/sockio.h>
45 #include <sys/time.h>
46 #include <sys/kernel.h>
47 #include <sys/protosw.h>
48 #include <sys/errno.h>
49 #include <sys/syslog.h>
50 #include <sys/lock.h>
51 #include <sys/rwlock.h>
52 #include <sys/queue.h>
53 #include <sys/sysctl.h>
54 
55 #include <net/if.h>
56 #include <net/if_arc.h>
57 #include <net/if_dl.h>
58 #include <net/if_types.h>
59 #include <net/iso88025.h>
60 #include <net/fddi.h>
61 #include <net/route.h>
62 #include <net/vnet.h>
63 
64 #include <netinet/in.h>
65 #include <net/if_llatbl.h>
66 #define	L3_ADDR_SIN6(le)	((struct sockaddr_in6 *) L3_ADDR(le))
67 #include <netinet/if_ether.h>
68 #include <netinet6/in6_var.h>
69 #include <netinet/ip6.h>
70 #include <netinet6/ip6_var.h>
71 #include <netinet6/scope6_var.h>
72 #include <netinet6/nd6.h>
73 #include <netinet6/in6_ifattach.h>
74 #include <netinet/icmp6.h>
75 
76 #include <sys/limits.h>
77 
78 #include <security/mac/mac_framework.h>
79 
80 #define ND6_SLOWTIMER_INTERVAL (60 * 60) /* 1 hour */
81 #define ND6_RECALC_REACHTM_INTERVAL (60 * 120) /* 2 hours */
82 
83 #define SIN6(s) ((struct sockaddr_in6 *)s)
84 
85 VNET_DEFINE(int, nd6_prune);
86 VNET_DEFINE(int, nd6_delay);
87 VNET_DEFINE(int, nd6_umaxtries);
88 VNET_DEFINE(int, nd6_mmaxtries);
89 VNET_DEFINE(int, nd6_useloopback);
90 VNET_DEFINE(int, nd6_gctimer);
91 
92 /* preventing too many loops in ND option parsing */
93 static VNET_DEFINE(int, nd6_maxndopt);
94 VNET_DEFINE(int, nd6_maxnudhint);
95 static VNET_DEFINE(int, nd6_maxqueuelen);
96 #define	V_nd6_maxndopt			VNET(nd6_maxndopt)
97 #define	V_nd6_maxqueuelen		VNET(nd6_maxqueuelen)
98 
99 VNET_DEFINE(int, nd6_debug);
100 
101 /* for debugging? */
102 #if 0
103 static int nd6_inuse, nd6_allocated;
104 #endif
105 
106 VNET_DEFINE(struct nd_drhead, nd_defrouter);
107 VNET_DEFINE(struct nd_prhead, nd_prefix);
108 
109 VNET_DEFINE(int, nd6_recalc_reachtm_interval);
110 #define	V_nd6_recalc_reachtm_interval	VNET(nd6_recalc_reachtm_interval)
111 
112 static struct sockaddr_in6 all1_sa;
113 
114 static int nd6_is_new_addr_neighbor __P((struct sockaddr_in6 *,
115 	struct ifnet *));
116 static void nd6_setmtu0(struct ifnet *, struct nd_ifinfo *);
117 static void nd6_slowtimo(void *);
118 static int regen_tmpaddr(struct in6_ifaddr *);
119 static struct llentry *nd6_free(struct llentry *, int);
120 static void nd6_llinfo_timer(void *);
121 static void clear_llinfo_pqueue(struct llentry *);
122 
123 static VNET_DEFINE(struct callout, nd6_slowtimo_ch);
124 #define	V_nd6_slowtimo_ch		VNET(nd6_slowtimo_ch)
125 
126 VNET_DEFINE(struct callout, nd6_timer_ch);
127 
128 VNET_DECLARE(int, dad_ignore_ns);
129 VNET_DECLARE(int, dad_maxtry);
130 #define	V_dad_ignore_ns			VNET(dad_ignore_ns)
131 #define	V_dad_maxtry			VNET(dad_maxtry)
132 
133 void
134 nd6_init(void)
135 {
136 	int i;
137 
138 	V_nd6_prune	= 1;	/* walk list every 1 seconds */
139 	V_nd6_delay	= 5;	/* delay first probe time 5 second */
140 	V_nd6_umaxtries	= 3;	/* maximum unicast query */
141 	V_nd6_mmaxtries	= 3;	/* maximum multicast query */
142 	V_nd6_useloopback = 1;	/* use loopback interface for local traffic */
143 	V_nd6_gctimer	= (60 * 60 * 24); /* 1 day: garbage collection timer */
144 
145 	/* preventing too many loops in ND option parsing */
146 	V_nd6_maxndopt = 10;	/* max # of ND options allowed */
147 
148 	V_nd6_maxnudhint = 0;	/* max # of subsequent upper layer hints */
149 	V_nd6_maxqueuelen = 1;	/* max pkts cached in unresolved ND entries */
150 
151 #ifdef ND6_DEBUG
152 	V_nd6_debug = 1;
153 #else
154 	V_nd6_debug = 0;
155 #endif
156 
157 	V_nd6_recalc_reachtm_interval = ND6_RECALC_REACHTM_INTERVAL;
158 
159 	V_dad_ignore_ns = 0;	/* ignore NS in DAD - specwise incorrect*/
160 	V_dad_maxtry = 15;	/* max # of *tries* to transmit DAD packet */
161 
162 	/*
163 	 * XXX just to get this to compile KMM
164 	 */
165 #ifdef notyet
166 	V_llinfo_nd6.ln_next = &V_llinfo_nd6;
167 	V_llinfo_nd6.ln_prev = &V_llinfo_nd6;
168 #endif
169 	LIST_INIT(&V_nd_prefix);
170 
171 	V_ip6_use_tempaddr = 0;
172 	V_ip6_temp_preferred_lifetime = DEF_TEMP_PREFERRED_LIFETIME;
173 	V_ip6_temp_valid_lifetime = DEF_TEMP_VALID_LIFETIME;
174 	V_ip6_temp_regen_advance = TEMPADDR_REGEN_ADVANCE;
175 
176 	V_ip6_desync_factor = 0;
177 
178 	all1_sa.sin6_family = AF_INET6;
179 	all1_sa.sin6_len = sizeof(struct sockaddr_in6);
180 	for (i = 0; i < sizeof(all1_sa.sin6_addr); i++)
181 		all1_sa.sin6_addr.s6_addr[i] = 0xff;
182 
183 	/* initialization of the default router list */
184 	TAILQ_INIT(&V_nd_defrouter);
185 	/* start timer */
186 	callout_init(&V_nd6_slowtimo_ch, 0);
187 	callout_reset(&V_nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
188 	    nd6_slowtimo, curvnet);
189 }
190 
191 
192 #ifdef VIMAGE
193 void
194 nd6_destroy()
195 {
196 
197 	callout_drain(&V_nd6_slowtimo_ch);
198 	callout_drain(&V_nd6_timer_ch);
199 }
200 #endif
201 
202 struct nd_ifinfo *
203 nd6_ifattach(struct ifnet *ifp)
204 {
205 	struct nd_ifinfo *nd;
206 
207 	nd = (struct nd_ifinfo *)malloc(sizeof(*nd), M_IP6NDP, M_WAITOK);
208 	bzero(nd, sizeof(*nd));
209 
210 	nd->initialized = 1;
211 
212 	nd->chlim = IPV6_DEFHLIM;
213 	nd->basereachable = REACHABLE_TIME;
214 	nd->reachable = ND_COMPUTE_RTIME(nd->basereachable);
215 	nd->retrans = RETRANS_TIMER;
216 
217 	nd->flags = ND6_IFF_PERFORMNUD;
218 
219 	/* A loopback interface always has ND6_IFF_AUTO_LINKLOCAL. */
220 	if (V_ip6_auto_linklocal || (ifp->if_flags & IFF_LOOPBACK))
221 		nd->flags |= ND6_IFF_AUTO_LINKLOCAL;
222 
223 	/* A loopback interface does not need to accept RTADV. */
224 	if (V_ip6_accept_rtadv && !(ifp->if_flags & IFF_LOOPBACK))
225 		nd->flags |= ND6_IFF_ACCEPT_RTADV;
226 
227 	/* XXX: we cannot call nd6_setmtu since ifp is not fully initialized */
228 	nd6_setmtu0(ifp, nd);
229 
230 	return nd;
231 }
232 
233 void
234 nd6_ifdetach(struct nd_ifinfo *nd)
235 {
236 
237 	free(nd, M_IP6NDP);
238 }
239 
240 /*
241  * Reset ND level link MTU. This function is called when the physical MTU
242  * changes, which means we might have to adjust the ND level MTU.
243  */
244 void
245 nd6_setmtu(struct ifnet *ifp)
246 {
247 
248 	nd6_setmtu0(ifp, ND_IFINFO(ifp));
249 }
250 
251 /* XXX todo: do not maintain copy of ifp->if_mtu in ndi->maxmtu */
252 void
253 nd6_setmtu0(struct ifnet *ifp, struct nd_ifinfo *ndi)
254 {
255 	u_int32_t omaxmtu;
256 
257 	omaxmtu = ndi->maxmtu;
258 
259 	switch (ifp->if_type) {
260 	case IFT_ARCNET:
261 		ndi->maxmtu = MIN(ARC_PHDS_MAXMTU, ifp->if_mtu); /* RFC2497 */
262 		break;
263 	case IFT_FDDI:
264 		ndi->maxmtu = MIN(FDDIIPMTU, ifp->if_mtu); /* RFC2467 */
265 		break;
266 	case IFT_ISO88025:
267 		 ndi->maxmtu = MIN(ISO88025_MAX_MTU, ifp->if_mtu);
268 		 break;
269 	default:
270 		ndi->maxmtu = ifp->if_mtu;
271 		break;
272 	}
273 
274 	/*
275 	 * Decreasing the interface MTU under IPV6 minimum MTU may cause
276 	 * undesirable situation.  We thus notify the operator of the change
277 	 * explicitly.  The check for omaxmtu is necessary to restrict the
278 	 * log to the case of changing the MTU, not initializing it.
279 	 */
280 	if (omaxmtu >= IPV6_MMTU && ndi->maxmtu < IPV6_MMTU) {
281 		log(LOG_NOTICE, "nd6_setmtu0: "
282 		    "new link MTU on %s (%lu) is too small for IPv6\n",
283 		    if_name(ifp), (unsigned long)ndi->maxmtu);
284 	}
285 
286 	if (ndi->maxmtu > V_in6_maxmtu)
287 		in6_setmaxmtu(); /* check all interfaces just in case */
288 
289 }
290 
291 void
292 nd6_option_init(void *opt, int icmp6len, union nd_opts *ndopts)
293 {
294 
295 	bzero(ndopts, sizeof(*ndopts));
296 	ndopts->nd_opts_search = (struct nd_opt_hdr *)opt;
297 	ndopts->nd_opts_last
298 		= (struct nd_opt_hdr *)(((u_char *)opt) + icmp6len);
299 
300 	if (icmp6len == 0) {
301 		ndopts->nd_opts_done = 1;
302 		ndopts->nd_opts_search = NULL;
303 	}
304 }
305 
306 /*
307  * Take one ND option.
308  */
309 struct nd_opt_hdr *
310 nd6_option(union nd_opts *ndopts)
311 {
312 	struct nd_opt_hdr *nd_opt;
313 	int olen;
314 
315 	if (ndopts == NULL)
316 		panic("ndopts == NULL in nd6_option");
317 	if (ndopts->nd_opts_last == NULL)
318 		panic("uninitialized ndopts in nd6_option");
319 	if (ndopts->nd_opts_search == NULL)
320 		return NULL;
321 	if (ndopts->nd_opts_done)
322 		return NULL;
323 
324 	nd_opt = ndopts->nd_opts_search;
325 
326 	/* make sure nd_opt_len is inside the buffer */
327 	if ((caddr_t)&nd_opt->nd_opt_len >= (caddr_t)ndopts->nd_opts_last) {
328 		bzero(ndopts, sizeof(*ndopts));
329 		return NULL;
330 	}
331 
332 	olen = nd_opt->nd_opt_len << 3;
333 	if (olen == 0) {
334 		/*
335 		 * Message validation requires that all included
336 		 * options have a length that is greater than zero.
337 		 */
338 		bzero(ndopts, sizeof(*ndopts));
339 		return NULL;
340 	}
341 
342 	ndopts->nd_opts_search = (struct nd_opt_hdr *)((caddr_t)nd_opt + olen);
343 	if (ndopts->nd_opts_search > ndopts->nd_opts_last) {
344 		/* option overruns the end of buffer, invalid */
345 		bzero(ndopts, sizeof(*ndopts));
346 		return NULL;
347 	} else if (ndopts->nd_opts_search == ndopts->nd_opts_last) {
348 		/* reached the end of options chain */
349 		ndopts->nd_opts_done = 1;
350 		ndopts->nd_opts_search = NULL;
351 	}
352 	return nd_opt;
353 }
354 
355 /*
356  * Parse multiple ND options.
357  * This function is much easier to use, for ND routines that do not need
358  * multiple options of the same type.
359  */
360 int
361 nd6_options(union nd_opts *ndopts)
362 {
363 	struct nd_opt_hdr *nd_opt;
364 	int i = 0;
365 
366 	if (ndopts == NULL)
367 		panic("ndopts == NULL in nd6_options");
368 	if (ndopts->nd_opts_last == NULL)
369 		panic("uninitialized ndopts in nd6_options");
370 	if (ndopts->nd_opts_search == NULL)
371 		return 0;
372 
373 	while (1) {
374 		nd_opt = nd6_option(ndopts);
375 		if (nd_opt == NULL && ndopts->nd_opts_last == NULL) {
376 			/*
377 			 * Message validation requires that all included
378 			 * options have a length that is greater than zero.
379 			 */
380 			ICMP6STAT_INC(icp6s_nd_badopt);
381 			bzero(ndopts, sizeof(*ndopts));
382 			return -1;
383 		}
384 
385 		if (nd_opt == NULL)
386 			goto skip1;
387 
388 		switch (nd_opt->nd_opt_type) {
389 		case ND_OPT_SOURCE_LINKADDR:
390 		case ND_OPT_TARGET_LINKADDR:
391 		case ND_OPT_MTU:
392 		case ND_OPT_REDIRECTED_HEADER:
393 			if (ndopts->nd_opt_array[nd_opt->nd_opt_type]) {
394 				nd6log((LOG_INFO,
395 				    "duplicated ND6 option found (type=%d)\n",
396 				    nd_opt->nd_opt_type));
397 				/* XXX bark? */
398 			} else {
399 				ndopts->nd_opt_array[nd_opt->nd_opt_type]
400 					= nd_opt;
401 			}
402 			break;
403 		case ND_OPT_PREFIX_INFORMATION:
404 			if (ndopts->nd_opt_array[nd_opt->nd_opt_type] == 0) {
405 				ndopts->nd_opt_array[nd_opt->nd_opt_type]
406 					= nd_opt;
407 			}
408 			ndopts->nd_opts_pi_end =
409 				(struct nd_opt_prefix_info *)nd_opt;
410 			break;
411 		default:
412 			/*
413 			 * Unknown options must be silently ignored,
414 			 * to accomodate future extension to the protocol.
415 			 */
416 			nd6log((LOG_DEBUG,
417 			    "nd6_options: unsupported option %d - "
418 			    "option ignored\n", nd_opt->nd_opt_type));
419 		}
420 
421 skip1:
422 		i++;
423 		if (i > V_nd6_maxndopt) {
424 			ICMP6STAT_INC(icp6s_nd_toomanyopt);
425 			nd6log((LOG_INFO, "too many loop in nd opt\n"));
426 			break;
427 		}
428 
429 		if (ndopts->nd_opts_done)
430 			break;
431 	}
432 
433 	return 0;
434 }
435 
436 /*
437  * ND6 timer routine to handle ND6 entries
438  */
439 void
440 nd6_llinfo_settimer_locked(struct llentry *ln, long tick)
441 {
442 	int canceled;
443 
444 	if (tick < 0) {
445 		ln->la_expire = 0;
446 		ln->ln_ntick = 0;
447 		canceled = callout_stop(&ln->ln_timer_ch);
448 	} else {
449 		ln->la_expire = time_second + tick / hz;
450 		LLE_ADDREF(ln);
451 		if (tick > INT_MAX) {
452 			ln->ln_ntick = tick - INT_MAX;
453 			canceled = callout_reset(&ln->ln_timer_ch, INT_MAX,
454 			    nd6_llinfo_timer, ln);
455 		} else {
456 			ln->ln_ntick = 0;
457 			canceled = callout_reset(&ln->ln_timer_ch, tick,
458 			    nd6_llinfo_timer, ln);
459 		}
460 	}
461 	if (canceled)
462 		LLE_REMREF(ln);
463 }
464 
465 void
466 nd6_llinfo_settimer(struct llentry *ln, long tick)
467 {
468 
469 	LLE_WLOCK(ln);
470 	nd6_llinfo_settimer_locked(ln, tick);
471 	LLE_WUNLOCK(ln);
472 }
473 
474 static void
475 nd6_llinfo_timer(void *arg)
476 {
477 	struct llentry *ln;
478 	struct in6_addr *dst;
479 	struct ifnet *ifp;
480 	struct nd_ifinfo *ndi = NULL;
481 
482 	ln = (struct llentry *)arg;
483 	if (ln == NULL) {
484 		panic("%s: NULL entry!\n", __func__);
485 		return;
486 	}
487 
488 	if ((ifp = ((ln->lle_tbl != NULL) ? ln->lle_tbl->llt_ifp : NULL)) == NULL)
489 		panic("ln ifp == NULL");
490 
491 	CURVNET_SET(ifp->if_vnet);
492 
493 	if (ln->ln_ntick > 0) {
494 		if (ln->ln_ntick > INT_MAX) {
495 			ln->ln_ntick -= INT_MAX;
496 			nd6_llinfo_settimer(ln, INT_MAX);
497 		} else {
498 			ln->ln_ntick = 0;
499 			nd6_llinfo_settimer(ln, ln->ln_ntick);
500 		}
501 		goto done;
502 	}
503 
504 	ndi = ND_IFINFO(ifp);
505 	dst = &L3_ADDR_SIN6(ln)->sin6_addr;
506 	if (ln->la_flags & LLE_STATIC) {
507 		goto done;
508 	}
509 
510 	if (ln->la_flags & LLE_DELETED) {
511 		(void)nd6_free(ln, 0);
512 		ln = NULL;
513 		goto done;
514 	}
515 
516 	switch (ln->ln_state) {
517 	case ND6_LLINFO_INCOMPLETE:
518 		if (ln->la_asked < V_nd6_mmaxtries) {
519 			ln->la_asked++;
520 			nd6_llinfo_settimer(ln, (long)ndi->retrans * hz / 1000);
521 			nd6_ns_output(ifp, NULL, dst, ln, 0);
522 		} else {
523 			struct mbuf *m = ln->la_hold;
524 			if (m) {
525 				struct mbuf *m0;
526 
527 				/*
528 				 * assuming every packet in la_hold has the
529 				 * same IP header
530 				 */
531 				m0 = m->m_nextpkt;
532 				m->m_nextpkt = NULL;
533 				icmp6_error2(m, ICMP6_DST_UNREACH,
534 				    ICMP6_DST_UNREACH_ADDR, 0, ifp);
535 
536 				ln->la_hold = m0;
537 				clear_llinfo_pqueue(ln);
538 			}
539 			(void)nd6_free(ln, 0);
540 			ln = NULL;
541 		}
542 		break;
543 	case ND6_LLINFO_REACHABLE:
544 		if (!ND6_LLINFO_PERMANENT(ln)) {
545 			ln->ln_state = ND6_LLINFO_STALE;
546 			nd6_llinfo_settimer(ln, (long)V_nd6_gctimer * hz);
547 		}
548 		break;
549 
550 	case ND6_LLINFO_STALE:
551 		/* Garbage Collection(RFC 2461 5.3) */
552 		if (!ND6_LLINFO_PERMANENT(ln)) {
553 			(void)nd6_free(ln, 1);
554 			ln = NULL;
555 		}
556 		break;
557 
558 	case ND6_LLINFO_DELAY:
559 		if (ndi && (ndi->flags & ND6_IFF_PERFORMNUD) != 0) {
560 			/* We need NUD */
561 			ln->la_asked = 1;
562 			ln->ln_state = ND6_LLINFO_PROBE;
563 			nd6_llinfo_settimer(ln, (long)ndi->retrans * hz / 1000);
564 			nd6_ns_output(ifp, dst, dst, ln, 0);
565 		} else {
566 			ln->ln_state = ND6_LLINFO_STALE; /* XXX */
567 			nd6_llinfo_settimer(ln, (long)V_nd6_gctimer * hz);
568 		}
569 		break;
570 	case ND6_LLINFO_PROBE:
571 		if (ln->la_asked < V_nd6_umaxtries) {
572 			ln->la_asked++;
573 			nd6_llinfo_settimer(ln, (long)ndi->retrans * hz / 1000);
574 			nd6_ns_output(ifp, dst, dst, ln, 0);
575 		} else {
576 			(void)nd6_free(ln, 0);
577 			ln = NULL;
578 		}
579 		break;
580 	}
581 done:
582 	if (ln != NULL)
583 		LLE_FREE(ln);
584 	CURVNET_RESTORE();
585 }
586 
587 
588 /*
589  * ND6 timer routine to expire default route list and prefix list
590  */
591 void
592 nd6_timer(void *arg)
593 {
594 	CURVNET_SET((struct vnet *) arg);
595 	int s;
596 	struct nd_defrouter *dr;
597 	struct nd_prefix *pr;
598 	struct in6_ifaddr *ia6, *nia6;
599 	struct in6_addrlifetime *lt6;
600 
601 	callout_reset(&V_nd6_timer_ch, V_nd6_prune * hz,
602 	    nd6_timer, curvnet);
603 
604 	/* expire default router list */
605 	s = splnet();
606 	dr = TAILQ_FIRST(&V_nd_defrouter);
607 	while (dr) {
608 		if (dr->expire && dr->expire < time_second) {
609 			struct nd_defrouter *t;
610 			t = TAILQ_NEXT(dr, dr_entry);
611 			defrtrlist_del(dr);
612 			dr = t;
613 		} else {
614 			dr = TAILQ_NEXT(dr, dr_entry);
615 		}
616 	}
617 
618 	/*
619 	 * expire interface addresses.
620 	 * in the past the loop was inside prefix expiry processing.
621 	 * However, from a stricter speci-confrmance standpoint, we should
622 	 * rather separate address lifetimes and prefix lifetimes.
623 	 *
624 	 * XXXRW: in6_ifaddrhead locking.
625 	 */
626   addrloop:
627 	TAILQ_FOREACH_SAFE(ia6, &V_in6_ifaddrhead, ia_link, nia6) {
628 		/* check address lifetime */
629 		lt6 = &ia6->ia6_lifetime;
630 		if (IFA6_IS_INVALID(ia6)) {
631 			int regen = 0;
632 
633 			/*
634 			 * If the expiring address is temporary, try
635 			 * regenerating a new one.  This would be useful when
636 			 * we suspended a laptop PC, then turned it on after a
637 			 * period that could invalidate all temporary
638 			 * addresses.  Although we may have to restart the
639 			 * loop (see below), it must be after purging the
640 			 * address.  Otherwise, we'd see an infinite loop of
641 			 * regeneration.
642 			 */
643 			if (V_ip6_use_tempaddr &&
644 			    (ia6->ia6_flags & IN6_IFF_TEMPORARY) != 0) {
645 				if (regen_tmpaddr(ia6) == 0)
646 					regen = 1;
647 			}
648 
649 			in6_purgeaddr(&ia6->ia_ifa);
650 
651 			if (regen)
652 				goto addrloop; /* XXX: see below */
653 		} else if (IFA6_IS_DEPRECATED(ia6)) {
654 			int oldflags = ia6->ia6_flags;
655 
656 			ia6->ia6_flags |= IN6_IFF_DEPRECATED;
657 
658 			/*
659 			 * If a temporary address has just become deprecated,
660 			 * regenerate a new one if possible.
661 			 */
662 			if (V_ip6_use_tempaddr &&
663 			    (ia6->ia6_flags & IN6_IFF_TEMPORARY) != 0 &&
664 			    (oldflags & IN6_IFF_DEPRECATED) == 0) {
665 
666 				if (regen_tmpaddr(ia6) == 0) {
667 					/*
668 					 * A new temporary address is
669 					 * generated.
670 					 * XXX: this means the address chain
671 					 * has changed while we are still in
672 					 * the loop.  Although the change
673 					 * would not cause disaster (because
674 					 * it's not a deletion, but an
675 					 * addition,) we'd rather restart the
676 					 * loop just for safety.  Or does this
677 					 * significantly reduce performance??
678 					 */
679 					goto addrloop;
680 				}
681 			}
682 		} else {
683 			/*
684 			 * A new RA might have made a deprecated address
685 			 * preferred.
686 			 */
687 			ia6->ia6_flags &= ~IN6_IFF_DEPRECATED;
688 		}
689 	}
690 
691 	/* expire prefix list */
692 	pr = V_nd_prefix.lh_first;
693 	while (pr) {
694 		/*
695 		 * check prefix lifetime.
696 		 * since pltime is just for autoconf, pltime processing for
697 		 * prefix is not necessary.
698 		 */
699 		if (pr->ndpr_vltime != ND6_INFINITE_LIFETIME &&
700 		    time_second - pr->ndpr_lastupdate > pr->ndpr_vltime) {
701 			struct nd_prefix *t;
702 			t = pr->ndpr_next;
703 
704 			/*
705 			 * address expiration and prefix expiration are
706 			 * separate.  NEVER perform in6_purgeaddr here.
707 			 */
708 
709 			prelist_remove(pr);
710 			pr = t;
711 		} else
712 			pr = pr->ndpr_next;
713 	}
714 	splx(s);
715 	CURVNET_RESTORE();
716 }
717 
718 /*
719  * ia6 - deprecated/invalidated temporary address
720  */
721 static int
722 regen_tmpaddr(struct in6_ifaddr *ia6)
723 {
724 	struct ifaddr *ifa;
725 	struct ifnet *ifp;
726 	struct in6_ifaddr *public_ifa6 = NULL;
727 
728 	ifp = ia6->ia_ifa.ifa_ifp;
729 	IF_ADDR_LOCK(ifp);
730 	TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
731 		struct in6_ifaddr *it6;
732 
733 		if (ifa->ifa_addr->sa_family != AF_INET6)
734 			continue;
735 
736 		it6 = (struct in6_ifaddr *)ifa;
737 
738 		/* ignore no autoconf addresses. */
739 		if ((it6->ia6_flags & IN6_IFF_AUTOCONF) == 0)
740 			continue;
741 
742 		/* ignore autoconf addresses with different prefixes. */
743 		if (it6->ia6_ndpr == NULL || it6->ia6_ndpr != ia6->ia6_ndpr)
744 			continue;
745 
746 		/*
747 		 * Now we are looking at an autoconf address with the same
748 		 * prefix as ours.  If the address is temporary and is still
749 		 * preferred, do not create another one.  It would be rare, but
750 		 * could happen, for example, when we resume a laptop PC after
751 		 * a long period.
752 		 */
753 		if ((it6->ia6_flags & IN6_IFF_TEMPORARY) != 0 &&
754 		    !IFA6_IS_DEPRECATED(it6)) {
755 			public_ifa6 = NULL;
756 			break;
757 		}
758 
759 		/*
760 		 * This is a public autoconf address that has the same prefix
761 		 * as ours.  If it is preferred, keep it.  We can't break the
762 		 * loop here, because there may be a still-preferred temporary
763 		 * address with the prefix.
764 		 */
765 		if (!IFA6_IS_DEPRECATED(it6))
766 		    public_ifa6 = it6;
767 	}
768 
769 	if (public_ifa6 != NULL) {
770 		int e;
771 
772 		if ((e = in6_tmpifadd(public_ifa6, 0, 0)) != 0) {
773 			IF_ADDR_UNLOCK(ifp);
774 			log(LOG_NOTICE, "regen_tmpaddr: failed to create a new"
775 			    " tmp addr,errno=%d\n", e);
776 			return (-1);
777 		}
778 		IF_ADDR_UNLOCK(ifp);
779 		return (0);
780 	}
781 
782 	IF_ADDR_UNLOCK(ifp);
783 	return (-1);
784 }
785 
786 /*
787  * Nuke neighbor cache/prefix/default router management table, right before
788  * ifp goes away.
789  */
790 void
791 nd6_purge(struct ifnet *ifp)
792 {
793 	struct nd_defrouter *dr, *ndr;
794 	struct nd_prefix *pr, *npr;
795 
796 	/*
797 	 * Nuke default router list entries toward ifp.
798 	 * We defer removal of default router list entries that is installed
799 	 * in the routing table, in order to keep additional side effects as
800 	 * small as possible.
801 	 */
802 	for (dr = TAILQ_FIRST(&V_nd_defrouter); dr; dr = ndr) {
803 		ndr = TAILQ_NEXT(dr, dr_entry);
804 		if (dr->installed)
805 			continue;
806 
807 		if (dr->ifp == ifp)
808 			defrtrlist_del(dr);
809 	}
810 
811 	for (dr = TAILQ_FIRST(&V_nd_defrouter); dr; dr = ndr) {
812 		ndr = TAILQ_NEXT(dr, dr_entry);
813 		if (!dr->installed)
814 			continue;
815 
816 		if (dr->ifp == ifp)
817 			defrtrlist_del(dr);
818 	}
819 
820 	/* Nuke prefix list entries toward ifp */
821 	for (pr = V_nd_prefix.lh_first; pr; pr = npr) {
822 		npr = pr->ndpr_next;
823 		if (pr->ndpr_ifp == ifp) {
824 			/*
825 			 * Because if_detach() does *not* release prefixes
826 			 * while purging addresses the reference count will
827 			 * still be above zero. We therefore reset it to
828 			 * make sure that the prefix really gets purged.
829 			 */
830 			pr->ndpr_refcnt = 0;
831 
832 			/*
833 			 * Previously, pr->ndpr_addr is removed as well,
834 			 * but I strongly believe we don't have to do it.
835 			 * nd6_purge() is only called from in6_ifdetach(),
836 			 * which removes all the associated interface addresses
837 			 * by itself.
838 			 * (jinmei@kame.net 20010129)
839 			 */
840 			prelist_remove(pr);
841 		}
842 	}
843 
844 	/* cancel default outgoing interface setting */
845 	if (V_nd6_defifindex == ifp->if_index)
846 		nd6_setdefaultiface(0);
847 
848 	if (!V_ip6_forwarding && ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) {
849 		/* Refresh default router list. */
850 		defrouter_select();
851 	}
852 
853 	/* XXXXX
854 	 * We do not nuke the neighbor cache entries here any more
855 	 * because the neighbor cache is kept in if_afdata[AF_INET6].
856 	 * nd6_purge() is invoked by in6_ifdetach() which is called
857 	 * from if_detach() where everything gets purged. So let
858 	 * in6_domifdetach() do the actual L2 table purging work.
859 	 */
860 }
861 
862 /*
863  * the caller acquires and releases the lock on the lltbls
864  * Returns the llentry locked
865  */
866 struct llentry *
867 nd6_lookup(struct in6_addr *addr6, int flags, struct ifnet *ifp)
868 {
869 	struct sockaddr_in6 sin6;
870 	struct llentry *ln;
871 	int llflags = 0;
872 
873 	bzero(&sin6, sizeof(sin6));
874 	sin6.sin6_len = sizeof(struct sockaddr_in6);
875 	sin6.sin6_family = AF_INET6;
876 	sin6.sin6_addr = *addr6;
877 
878 	IF_AFDATA_LOCK_ASSERT(ifp);
879 
880 	if (flags & ND6_CREATE)
881 	    llflags |= LLE_CREATE;
882 	if (flags & ND6_EXCLUSIVE)
883 	    llflags |= LLE_EXCLUSIVE;
884 
885 	ln = lla_lookup(LLTABLE6(ifp), llflags, (struct sockaddr *)&sin6);
886 	if ((ln != NULL) && (flags & LLE_CREATE)) {
887 		ln->ln_state = ND6_LLINFO_NOSTATE;
888 		callout_init(&ln->ln_timer_ch, 0);
889 	}
890 
891 	return (ln);
892 }
893 
894 /*
895  * Test whether a given IPv6 address is a neighbor or not, ignoring
896  * the actual neighbor cache.  The neighbor cache is ignored in order
897  * to not reenter the routing code from within itself.
898  */
899 static int
900 nd6_is_new_addr_neighbor(struct sockaddr_in6 *addr, struct ifnet *ifp)
901 {
902 	struct nd_prefix *pr;
903 	struct ifaddr *dstaddr;
904 
905 	/*
906 	 * A link-local address is always a neighbor.
907 	 * XXX: a link does not necessarily specify a single interface.
908 	 */
909 	if (IN6_IS_ADDR_LINKLOCAL(&addr->sin6_addr)) {
910 		struct sockaddr_in6 sin6_copy;
911 		u_int32_t zone;
912 
913 		/*
914 		 * We need sin6_copy since sa6_recoverscope() may modify the
915 		 * content (XXX).
916 		 */
917 		sin6_copy = *addr;
918 		if (sa6_recoverscope(&sin6_copy))
919 			return (0); /* XXX: should be impossible */
920 		if (in6_setscope(&sin6_copy.sin6_addr, ifp, &zone))
921 			return (0);
922 		if (sin6_copy.sin6_scope_id == zone)
923 			return (1);
924 		else
925 			return (0);
926 	}
927 
928 	/*
929 	 * If the address matches one of our addresses,
930 	 * it should be a neighbor.
931 	 * If the address matches one of our on-link prefixes, it should be a
932 	 * neighbor.
933 	 */
934 	for (pr = V_nd_prefix.lh_first; pr; pr = pr->ndpr_next) {
935 		if (pr->ndpr_ifp != ifp)
936 			continue;
937 
938 		if (!(pr->ndpr_stateflags & NDPRF_ONLINK)) {
939 			struct rtentry *rt;
940 			rt = rtalloc1((struct sockaddr *)&pr->ndpr_prefix, 0, 0);
941 			if (rt == NULL)
942 				continue;
943 			/*
944 			 * This is the case where multiple interfaces
945 			 * have the same prefix, but only one is installed
946 			 * into the routing table and that prefix entry
947 			 * is not the one being examined here. In the case
948 			 * where RADIX_MPATH is enabled, multiple route
949 			 * entries (of the same rt_key value) will be
950 			 * installed because the interface addresses all
951 			 * differ.
952 			 */
953 			if (!IN6_ARE_ADDR_EQUAL(&pr->ndpr_prefix.sin6_addr,
954 			       &((struct sockaddr_in6 *)rt_key(rt))->sin6_addr)) {
955 				RTFREE_LOCKED(rt);
956 				continue;
957 			}
958 			RTFREE_LOCKED(rt);
959 		}
960 
961 		if (IN6_ARE_MASKED_ADDR_EQUAL(&pr->ndpr_prefix.sin6_addr,
962 		    &addr->sin6_addr, &pr->ndpr_mask))
963 			return (1);
964 	}
965 
966 	/*
967 	 * If the address is assigned on the node of the other side of
968 	 * a p2p interface, the address should be a neighbor.
969 	 */
970 	dstaddr = ifa_ifwithdstaddr((struct sockaddr *)addr);
971 	if (dstaddr != NULL) {
972 		if (dstaddr->ifa_ifp == ifp) {
973 			ifa_free(dstaddr);
974 			return (1);
975 		}
976 		ifa_free(dstaddr);
977 	}
978 
979 	/*
980 	 * If the default router list is empty, all addresses are regarded
981 	 * as on-link, and thus, as a neighbor.
982 	 * XXX: we restrict the condition to hosts, because routers usually do
983 	 * not have the "default router list".
984 	 */
985 	if (!V_ip6_forwarding && TAILQ_FIRST(&V_nd_defrouter) == NULL &&
986 	    V_nd6_defifindex == ifp->if_index) {
987 		return (1);
988 	}
989 
990 	return (0);
991 }
992 
993 
994 /*
995  * Detect if a given IPv6 address identifies a neighbor on a given link.
996  * XXX: should take care of the destination of a p2p link?
997  */
998 int
999 nd6_is_addr_neighbor(struct sockaddr_in6 *addr, struct ifnet *ifp)
1000 {
1001 	struct llentry *lle;
1002 	int rc = 0;
1003 
1004 	IF_AFDATA_UNLOCK_ASSERT(ifp);
1005 	if (nd6_is_new_addr_neighbor(addr, ifp))
1006 		return (1);
1007 
1008 	/*
1009 	 * Even if the address matches none of our addresses, it might be
1010 	 * in the neighbor cache.
1011 	 */
1012 	IF_AFDATA_LOCK(ifp);
1013 	if ((lle = nd6_lookup(&addr->sin6_addr, 0, ifp)) != NULL) {
1014 		LLE_RUNLOCK(lle);
1015 		rc = 1;
1016 	}
1017 	IF_AFDATA_UNLOCK(ifp);
1018 	return (rc);
1019 }
1020 
1021 /*
1022  * Free an nd6 llinfo entry.
1023  * Since the function would cause significant changes in the kernel, DO NOT
1024  * make it global, unless you have a strong reason for the change, and are sure
1025  * that the change is safe.
1026  */
1027 static struct llentry *
1028 nd6_free(struct llentry *ln, int gc)
1029 {
1030         struct llentry *next;
1031 	struct nd_defrouter *dr;
1032 	struct ifnet *ifp=NULL;
1033 
1034 	/*
1035 	 * we used to have pfctlinput(PRC_HOSTDEAD) here.
1036 	 * even though it is not harmful, it was not really necessary.
1037 	 */
1038 
1039 	/* cancel timer */
1040 	nd6_llinfo_settimer(ln, -1);
1041 
1042 	if (!V_ip6_forwarding) {
1043 		int s;
1044 		s = splnet();
1045 		dr = defrouter_lookup(&L3_ADDR_SIN6(ln)->sin6_addr, ln->lle_tbl->llt_ifp);
1046 
1047 		if (dr != NULL && dr->expire &&
1048 		    ln->ln_state == ND6_LLINFO_STALE && gc) {
1049 			/*
1050 			 * If the reason for the deletion is just garbage
1051 			 * collection, and the neighbor is an active default
1052 			 * router, do not delete it.  Instead, reset the GC
1053 			 * timer using the router's lifetime.
1054 			 * Simply deleting the entry would affect default
1055 			 * router selection, which is not necessarily a good
1056 			 * thing, especially when we're using router preference
1057 			 * values.
1058 			 * XXX: the check for ln_state would be redundant,
1059 			 *      but we intentionally keep it just in case.
1060 			 */
1061 			if (dr->expire > time_second)
1062 				nd6_llinfo_settimer(ln,
1063 				    (dr->expire - time_second) * hz);
1064 			else
1065 				nd6_llinfo_settimer(ln, (long)V_nd6_gctimer * hz);
1066 			splx(s);
1067 			LLE_WLOCK(ln);
1068 			LLE_REMREF(ln);
1069 			LLE_WUNLOCK(ln);
1070 			return (LIST_NEXT(ln, lle_next));
1071 		}
1072 
1073 		if (ln->ln_router || dr) {
1074 			/*
1075 			 * rt6_flush must be called whether or not the neighbor
1076 			 * is in the Default Router List.
1077 			 * See a corresponding comment in nd6_na_input().
1078 			 */
1079 			rt6_flush(&L3_ADDR_SIN6(ln)->sin6_addr, ln->lle_tbl->llt_ifp);
1080 		}
1081 
1082 		if (dr) {
1083 			/*
1084 			 * Unreachablity of a router might affect the default
1085 			 * router selection and on-link detection of advertised
1086 			 * prefixes.
1087 			 */
1088 
1089 			/*
1090 			 * Temporarily fake the state to choose a new default
1091 			 * router and to perform on-link determination of
1092 			 * prefixes correctly.
1093 			 * Below the state will be set correctly,
1094 			 * or the entry itself will be deleted.
1095 			 */
1096 			ln->ln_state = ND6_LLINFO_INCOMPLETE;
1097 
1098 			/*
1099 			 * Since defrouter_select() does not affect the
1100 			 * on-link determination and MIP6 needs the check
1101 			 * before the default router selection, we perform
1102 			 * the check now.
1103 			 */
1104 			pfxlist_onlink_check();
1105 
1106 			/*
1107 			 * refresh default router list
1108 			 */
1109 			defrouter_select();
1110 		}
1111 		splx(s);
1112 	}
1113 
1114 	/*
1115 	 * Before deleting the entry, remember the next entry as the
1116 	 * return value.  We need this because pfxlist_onlink_check() above
1117 	 * might have freed other entries (particularly the old next entry) as
1118 	 * a side effect (XXX).
1119 	 */
1120 	next = LIST_NEXT(ln, lle_next);
1121 
1122 	ifp = ln->lle_tbl->llt_ifp;
1123 	IF_AFDATA_LOCK(ifp);
1124 	LLE_WLOCK(ln);
1125 	llentry_free(ln);
1126 	IF_AFDATA_UNLOCK(ifp);
1127 
1128 	return (next);
1129 }
1130 
1131 /*
1132  * Upper-layer reachability hint for Neighbor Unreachability Detection.
1133  *
1134  * XXX cost-effective methods?
1135  */
1136 void
1137 nd6_nud_hint(struct rtentry *rt, struct in6_addr *dst6, int force)
1138 {
1139 	struct llentry *ln;
1140 	struct ifnet *ifp;
1141 
1142 	if ((dst6 == NULL) || (rt == NULL))
1143 		return;
1144 
1145 	ifp = rt->rt_ifp;
1146 	IF_AFDATA_LOCK(ifp);
1147 	ln = nd6_lookup(dst6, ND6_EXCLUSIVE, NULL);
1148 	IF_AFDATA_UNLOCK(ifp);
1149 	if (ln == NULL)
1150 		return;
1151 
1152 	if (ln->ln_state < ND6_LLINFO_REACHABLE)
1153 		goto done;
1154 
1155 	/*
1156 	 * if we get upper-layer reachability confirmation many times,
1157 	 * it is possible we have false information.
1158 	 */
1159 	if (!force) {
1160 		ln->ln_byhint++;
1161 		if (ln->ln_byhint > V_nd6_maxnudhint) {
1162 			goto done;
1163 		}
1164 	}
1165 
1166  	ln->ln_state = ND6_LLINFO_REACHABLE;
1167 	if (!ND6_LLINFO_PERMANENT(ln)) {
1168 		nd6_llinfo_settimer(ln,
1169 		    (long)ND_IFINFO(rt->rt_ifp)->reachable * hz);
1170 	}
1171 done:
1172 	LLE_WUNLOCK(ln);
1173 }
1174 
1175 
1176 int
1177 nd6_ioctl(u_long cmd, caddr_t data, struct ifnet *ifp)
1178 {
1179 	struct in6_drlist *drl = (struct in6_drlist *)data;
1180 	struct in6_oprlist *oprl = (struct in6_oprlist *)data;
1181 	struct in6_ndireq *ndi = (struct in6_ndireq *)data;
1182 	struct in6_nbrinfo *nbi = (struct in6_nbrinfo *)data;
1183 	struct in6_ndifreq *ndif = (struct in6_ndifreq *)data;
1184 	struct nd_defrouter *dr;
1185 	struct nd_prefix *pr;
1186 	int i = 0, error = 0;
1187 	int s;
1188 
1189 	switch (cmd) {
1190 	case SIOCGDRLST_IN6:
1191 		/*
1192 		 * obsolete API, use sysctl under net.inet6.icmp6
1193 		 */
1194 		bzero(drl, sizeof(*drl));
1195 		s = splnet();
1196 		dr = TAILQ_FIRST(&V_nd_defrouter);
1197 		while (dr && i < DRLSTSIZ) {
1198 			drl->defrouter[i].rtaddr = dr->rtaddr;
1199 			in6_clearscope(&drl->defrouter[i].rtaddr);
1200 
1201 			drl->defrouter[i].flags = dr->flags;
1202 			drl->defrouter[i].rtlifetime = dr->rtlifetime;
1203 			drl->defrouter[i].expire = dr->expire;
1204 			drl->defrouter[i].if_index = dr->ifp->if_index;
1205 			i++;
1206 			dr = TAILQ_NEXT(dr, dr_entry);
1207 		}
1208 		splx(s);
1209 		break;
1210 	case SIOCGPRLST_IN6:
1211 		/*
1212 		 * obsolete API, use sysctl under net.inet6.icmp6
1213 		 *
1214 		 * XXX the structure in6_prlist was changed in backward-
1215 		 * incompatible manner.  in6_oprlist is used for SIOCGPRLST_IN6,
1216 		 * in6_prlist is used for nd6_sysctl() - fill_prlist().
1217 		 */
1218 		/*
1219 		 * XXX meaning of fields, especialy "raflags", is very
1220 		 * differnet between RA prefix list and RR/static prefix list.
1221 		 * how about separating ioctls into two?
1222 		 */
1223 		bzero(oprl, sizeof(*oprl));
1224 		s = splnet();
1225 		pr = V_nd_prefix.lh_first;
1226 		while (pr && i < PRLSTSIZ) {
1227 			struct nd_pfxrouter *pfr;
1228 			int j;
1229 
1230 			oprl->prefix[i].prefix = pr->ndpr_prefix.sin6_addr;
1231 			oprl->prefix[i].raflags = pr->ndpr_raf;
1232 			oprl->prefix[i].prefixlen = pr->ndpr_plen;
1233 			oprl->prefix[i].vltime = pr->ndpr_vltime;
1234 			oprl->prefix[i].pltime = pr->ndpr_pltime;
1235 			oprl->prefix[i].if_index = pr->ndpr_ifp->if_index;
1236 			if (pr->ndpr_vltime == ND6_INFINITE_LIFETIME)
1237 				oprl->prefix[i].expire = 0;
1238 			else {
1239 				time_t maxexpire;
1240 
1241 				/* XXX: we assume time_t is signed. */
1242 				maxexpire = (-1) &
1243 				    ~((time_t)1 <<
1244 				    ((sizeof(maxexpire) * 8) - 1));
1245 				if (pr->ndpr_vltime <
1246 				    maxexpire - pr->ndpr_lastupdate) {
1247 					oprl->prefix[i].expire =
1248 					    pr->ndpr_lastupdate +
1249 					    pr->ndpr_vltime;
1250 				} else
1251 					oprl->prefix[i].expire = maxexpire;
1252 			}
1253 
1254 			pfr = pr->ndpr_advrtrs.lh_first;
1255 			j = 0;
1256 			while (pfr) {
1257 				if (j < DRLSTSIZ) {
1258 #define RTRADDR oprl->prefix[i].advrtr[j]
1259 					RTRADDR = pfr->router->rtaddr;
1260 					in6_clearscope(&RTRADDR);
1261 #undef RTRADDR
1262 				}
1263 				j++;
1264 				pfr = pfr->pfr_next;
1265 			}
1266 			oprl->prefix[i].advrtrs = j;
1267 			oprl->prefix[i].origin = PR_ORIG_RA;
1268 
1269 			i++;
1270 			pr = pr->ndpr_next;
1271 		}
1272 		splx(s);
1273 
1274 		break;
1275 	case OSIOCGIFINFO_IN6:
1276 #define ND	ndi->ndi
1277 		/* XXX: old ndp(8) assumes a positive value for linkmtu. */
1278 		bzero(&ND, sizeof(ND));
1279 		ND.linkmtu = IN6_LINKMTU(ifp);
1280 		ND.maxmtu = ND_IFINFO(ifp)->maxmtu;
1281 		ND.basereachable = ND_IFINFO(ifp)->basereachable;
1282 		ND.reachable = ND_IFINFO(ifp)->reachable;
1283 		ND.retrans = ND_IFINFO(ifp)->retrans;
1284 		ND.flags = ND_IFINFO(ifp)->flags;
1285 		ND.recalctm = ND_IFINFO(ifp)->recalctm;
1286 		ND.chlim = ND_IFINFO(ifp)->chlim;
1287 		break;
1288 	case SIOCGIFINFO_IN6:
1289 		ND = *ND_IFINFO(ifp);
1290 		break;
1291 	case SIOCSIFINFO_IN6:
1292 		/*
1293 		 * used to change host variables from userland.
1294 		 * intented for a use on router to reflect RA configurations.
1295 		 */
1296 		/* 0 means 'unspecified' */
1297 		if (ND.linkmtu != 0) {
1298 			if (ND.linkmtu < IPV6_MMTU ||
1299 			    ND.linkmtu > IN6_LINKMTU(ifp)) {
1300 				error = EINVAL;
1301 				break;
1302 			}
1303 			ND_IFINFO(ifp)->linkmtu = ND.linkmtu;
1304 		}
1305 
1306 		if (ND.basereachable != 0) {
1307 			int obasereachable = ND_IFINFO(ifp)->basereachable;
1308 
1309 			ND_IFINFO(ifp)->basereachable = ND.basereachable;
1310 			if (ND.basereachable != obasereachable)
1311 				ND_IFINFO(ifp)->reachable =
1312 				    ND_COMPUTE_RTIME(ND.basereachable);
1313 		}
1314 		if (ND.retrans != 0)
1315 			ND_IFINFO(ifp)->retrans = ND.retrans;
1316 		if (ND.chlim != 0)
1317 			ND_IFINFO(ifp)->chlim = ND.chlim;
1318 		/* FALLTHROUGH */
1319 	case SIOCSIFINFO_FLAGS:
1320 	{
1321 		struct ifaddr *ifa;
1322 		struct in6_ifaddr *ia;
1323 
1324 		if ((ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED) &&
1325 		    !(ND.flags & ND6_IFF_IFDISABLED)) {
1326 			/* ifdisabled 1->0 transision */
1327 
1328 			/*
1329 			 * If the interface is marked as ND6_IFF_IFDISABLED and
1330 			 * has an link-local address with IN6_IFF_DUPLICATED,
1331 			 * do not clear ND6_IFF_IFDISABLED.
1332 			 * See RFC 4862, Section 5.4.5.
1333 			 */
1334 			int duplicated_linklocal = 0;
1335 
1336 			IF_ADDR_LOCK(ifp);
1337 			TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1338 				if (ifa->ifa_addr->sa_family != AF_INET6)
1339 					continue;
1340 				ia = (struct in6_ifaddr *)ifa;
1341 				if ((ia->ia6_flags & IN6_IFF_DUPLICATED) &&
1342 				    IN6_IS_ADDR_LINKLOCAL(&ia->ia_addr.sin6_addr)) {
1343 					duplicated_linklocal = 1;
1344 					break;
1345 				}
1346 			}
1347 			IF_ADDR_UNLOCK(ifp);
1348 
1349 			if (duplicated_linklocal) {
1350 				ND.flags |= ND6_IFF_IFDISABLED;
1351 				log(LOG_ERR, "Cannot enable an interface"
1352 				    " with a link-local address marked"
1353 				    " duplicate.\n");
1354 			} else {
1355 				ND_IFINFO(ifp)->flags &= ~ND6_IFF_IFDISABLED;
1356 				in6_if_up(ifp);
1357 			}
1358 		} else if (!(ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED) &&
1359 			    (ND.flags & ND6_IFF_IFDISABLED)) {
1360 			/* ifdisabled 0->1 transision */
1361 			/* Mark all IPv6 address as tentative. */
1362 
1363 			ND_IFINFO(ifp)->flags |= ND6_IFF_IFDISABLED;
1364 			IF_ADDR_LOCK(ifp);
1365 			TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1366 				if (ifa->ifa_addr->sa_family != AF_INET6)
1367 					continue;
1368 				ia = (struct in6_ifaddr *)ifa;
1369 				ia->ia6_flags |= IN6_IFF_TENTATIVE;
1370 			}
1371 			IF_ADDR_UNLOCK(ifp);
1372 		}
1373 
1374 		if (!(ND_IFINFO(ifp)->flags & ND6_IFF_AUTO_LINKLOCAL) &&
1375 		    (ND.flags & ND6_IFF_AUTO_LINKLOCAL)) {
1376 			/* auto_linklocal 0->1 transision */
1377 
1378 			/* If no link-local address on ifp, configure */
1379 			ND_IFINFO(ifp)->flags |= ND6_IFF_AUTO_LINKLOCAL;
1380 			in6_ifattach(ifp, NULL);
1381 		}
1382 	}
1383 		ND_IFINFO(ifp)->flags = ND.flags;
1384 		break;
1385 #undef ND
1386 	case SIOCSNDFLUSH_IN6:	/* XXX: the ioctl name is confusing... */
1387 		/* sync kernel routing table with the default router list */
1388 		defrouter_reset();
1389 		defrouter_select();
1390 		break;
1391 	case SIOCSPFXFLUSH_IN6:
1392 	{
1393 		/* flush all the prefix advertised by routers */
1394 		struct nd_prefix *pr, *next;
1395 
1396 		s = splnet();
1397 		for (pr = V_nd_prefix.lh_first; pr; pr = next) {
1398 			struct in6_ifaddr *ia, *ia_next;
1399 
1400 			next = pr->ndpr_next;
1401 
1402 			if (IN6_IS_ADDR_LINKLOCAL(&pr->ndpr_prefix.sin6_addr))
1403 				continue; /* XXX */
1404 
1405 			/* do we really have to remove addresses as well? */
1406 			/* XXXRW: in6_ifaddrhead locking. */
1407 			TAILQ_FOREACH_SAFE(ia, &V_in6_ifaddrhead, ia_link,
1408 			    ia_next) {
1409 				if ((ia->ia6_flags & IN6_IFF_AUTOCONF) == 0)
1410 					continue;
1411 
1412 				if (ia->ia6_ndpr == pr)
1413 					in6_purgeaddr(&ia->ia_ifa);
1414 			}
1415 			prelist_remove(pr);
1416 		}
1417 		splx(s);
1418 		break;
1419 	}
1420 	case SIOCSRTRFLUSH_IN6:
1421 	{
1422 		/* flush all the default routers */
1423 		struct nd_defrouter *dr, *next;
1424 
1425 		s = splnet();
1426 		defrouter_reset();
1427 		for (dr = TAILQ_FIRST(&V_nd_defrouter); dr; dr = next) {
1428 			next = TAILQ_NEXT(dr, dr_entry);
1429 			defrtrlist_del(dr);
1430 		}
1431 		defrouter_select();
1432 		splx(s);
1433 		break;
1434 	}
1435 	case SIOCGNBRINFO_IN6:
1436 	{
1437 		struct llentry *ln;
1438 		struct in6_addr nb_addr = nbi->addr; /* make local for safety */
1439 
1440 		if ((error = in6_setscope(&nb_addr, ifp, NULL)) != 0)
1441 			return (error);
1442 
1443 		IF_AFDATA_LOCK(ifp);
1444 		ln = nd6_lookup(&nb_addr, 0, ifp);
1445 		IF_AFDATA_UNLOCK(ifp);
1446 
1447 		if (ln == NULL) {
1448 			error = EINVAL;
1449 			break;
1450 		}
1451 		nbi->state = ln->ln_state;
1452 		nbi->asked = ln->la_asked;
1453 		nbi->isrouter = ln->ln_router;
1454 		nbi->expire = ln->la_expire;
1455 		LLE_RUNLOCK(ln);
1456 		break;
1457 	}
1458 	case SIOCGDEFIFACE_IN6:	/* XXX: should be implemented as a sysctl? */
1459 		ndif->ifindex = V_nd6_defifindex;
1460 		break;
1461 	case SIOCSDEFIFACE_IN6:	/* XXX: should be implemented as a sysctl? */
1462 		return (nd6_setdefaultiface(ndif->ifindex));
1463 	}
1464 	return (error);
1465 }
1466 
1467 /*
1468  * Create neighbor cache entry and cache link-layer address,
1469  * on reception of inbound ND6 packets.  (RS/RA/NS/redirect)
1470  *
1471  * type - ICMP6 type
1472  * code - type dependent information
1473  *
1474  * XXXXX
1475  *  The caller of this function already acquired the ndp
1476  *  cache table lock because the cache entry is returned.
1477  */
1478 struct llentry *
1479 nd6_cache_lladdr(struct ifnet *ifp, struct in6_addr *from, char *lladdr,
1480     int lladdrlen, int type, int code)
1481 {
1482 	struct llentry *ln = NULL;
1483 	int is_newentry;
1484 	int do_update;
1485 	int olladdr;
1486 	int llchange;
1487 	int flags = 0;
1488 	int newstate = 0;
1489 	uint16_t router = 0;
1490 	struct sockaddr_in6 sin6;
1491 	struct mbuf *chain = NULL;
1492 	int static_route = 0;
1493 
1494 	IF_AFDATA_UNLOCK_ASSERT(ifp);
1495 
1496 	if (ifp == NULL)
1497 		panic("ifp == NULL in nd6_cache_lladdr");
1498 	if (from == NULL)
1499 		panic("from == NULL in nd6_cache_lladdr");
1500 
1501 	/* nothing must be updated for unspecified address */
1502 	if (IN6_IS_ADDR_UNSPECIFIED(from))
1503 		return NULL;
1504 
1505 	/*
1506 	 * Validation about ifp->if_addrlen and lladdrlen must be done in
1507 	 * the caller.
1508 	 *
1509 	 * XXX If the link does not have link-layer adderss, what should
1510 	 * we do? (ifp->if_addrlen == 0)
1511 	 * Spec says nothing in sections for RA, RS and NA.  There's small
1512 	 * description on it in NS section (RFC 2461 7.2.3).
1513 	 */
1514 	flags |= lladdr ? ND6_EXCLUSIVE : 0;
1515 	IF_AFDATA_LOCK(ifp);
1516 	ln = nd6_lookup(from, flags, ifp);
1517 
1518 	if (ln == NULL) {
1519 		flags |= LLE_EXCLUSIVE;
1520 		ln = nd6_lookup(from, flags |ND6_CREATE, ifp);
1521 		IF_AFDATA_UNLOCK(ifp);
1522 		is_newentry = 1;
1523 	} else {
1524 		IF_AFDATA_UNLOCK(ifp);
1525 		/* do nothing if static ndp is set */
1526 		if (ln->la_flags & LLE_STATIC) {
1527 			static_route = 1;
1528 			goto done;
1529 		}
1530 		is_newentry = 0;
1531 	}
1532 	if (ln == NULL)
1533 		return (NULL);
1534 
1535 	olladdr = (ln->la_flags & LLE_VALID) ? 1 : 0;
1536 	if (olladdr && lladdr) {
1537 		llchange = bcmp(lladdr, &ln->ll_addr,
1538 		    ifp->if_addrlen);
1539 	} else
1540 		llchange = 0;
1541 
1542 	/*
1543 	 * newentry olladdr  lladdr  llchange	(*=record)
1544 	 *	0	n	n	--	(1)
1545 	 *	0	y	n	--	(2)
1546 	 *	0	n	y	--	(3) * STALE
1547 	 *	0	y	y	n	(4) *
1548 	 *	0	y	y	y	(5) * STALE
1549 	 *	1	--	n	--	(6)   NOSTATE(= PASSIVE)
1550 	 *	1	--	y	--	(7) * STALE
1551 	 */
1552 
1553 	if (lladdr) {		/* (3-5) and (7) */
1554 		/*
1555 		 * Record source link-layer address
1556 		 * XXX is it dependent to ifp->if_type?
1557 		 */
1558 		bcopy(lladdr, &ln->ll_addr, ifp->if_addrlen);
1559 		ln->la_flags |= LLE_VALID;
1560 	}
1561 
1562 	if (!is_newentry) {
1563 		if ((!olladdr && lladdr != NULL) ||	/* (3) */
1564 		    (olladdr && lladdr != NULL && llchange)) {	/* (5) */
1565 			do_update = 1;
1566 			newstate = ND6_LLINFO_STALE;
1567 		} else					/* (1-2,4) */
1568 			do_update = 0;
1569 	} else {
1570 		do_update = 1;
1571 		if (lladdr == NULL)			/* (6) */
1572 			newstate = ND6_LLINFO_NOSTATE;
1573 		else					/* (7) */
1574 			newstate = ND6_LLINFO_STALE;
1575 	}
1576 
1577 	if (do_update) {
1578 		/*
1579 		 * Update the state of the neighbor cache.
1580 		 */
1581 		ln->ln_state = newstate;
1582 
1583 		if (ln->ln_state == ND6_LLINFO_STALE) {
1584 			/*
1585 			 * XXX: since nd6_output() below will cause
1586 			 * state tansition to DELAY and reset the timer,
1587 			 * we must set the timer now, although it is actually
1588 			 * meaningless.
1589 			 */
1590 			nd6_llinfo_settimer_locked(ln, (long)V_nd6_gctimer * hz);
1591 
1592 			if (ln->la_hold) {
1593 				struct mbuf *m_hold, *m_hold_next;
1594 
1595 				/*
1596 				 * reset the la_hold in advance, to explicitly
1597 				 * prevent a la_hold lookup in nd6_output()
1598 				 * (wouldn't happen, though...)
1599 				 */
1600 				for (m_hold = ln->la_hold, ln->la_hold = NULL;
1601 				    m_hold; m_hold = m_hold_next) {
1602 					m_hold_next = m_hold->m_nextpkt;
1603 					m_hold->m_nextpkt = NULL;
1604 
1605 					/*
1606 					 * we assume ifp is not a p2p here, so
1607 					 * just set the 2nd argument as the
1608 					 * 1st one.
1609 					 */
1610 					nd6_output_lle(ifp, ifp, m_hold, L3_ADDR_SIN6(ln), NULL, ln, &chain);
1611 				}
1612 				/*
1613 				 * If we have mbufs in the chain we need to do
1614 				 * deferred transmit. Copy the address from the
1615 				 * llentry before dropping the lock down below.
1616 				 */
1617 				if (chain != NULL)
1618 					memcpy(&sin6, L3_ADDR_SIN6(ln), sizeof(sin6));
1619 			}
1620 		} else if (ln->ln_state == ND6_LLINFO_INCOMPLETE) {
1621 			/* probe right away */
1622 			nd6_llinfo_settimer_locked((void *)ln, 0);
1623 		}
1624 	}
1625 
1626 	/*
1627 	 * ICMP6 type dependent behavior.
1628 	 *
1629 	 * NS: clear IsRouter if new entry
1630 	 * RS: clear IsRouter
1631 	 * RA: set IsRouter if there's lladdr
1632 	 * redir: clear IsRouter if new entry
1633 	 *
1634 	 * RA case, (1):
1635 	 * The spec says that we must set IsRouter in the following cases:
1636 	 * - If lladdr exist, set IsRouter.  This means (1-5).
1637 	 * - If it is old entry (!newentry), set IsRouter.  This means (7).
1638 	 * So, based on the spec, in (1-5) and (7) cases we must set IsRouter.
1639 	 * A quetion arises for (1) case.  (1) case has no lladdr in the
1640 	 * neighbor cache, this is similar to (6).
1641 	 * This case is rare but we figured that we MUST NOT set IsRouter.
1642 	 *
1643 	 * newentry olladdr  lladdr  llchange	    NS  RS  RA	redir
1644 	 *							D R
1645 	 *	0	n	n	--	(1)	c   ?     s
1646 	 *	0	y	n	--	(2)	c   s     s
1647 	 *	0	n	y	--	(3)	c   s     s
1648 	 *	0	y	y	n	(4)	c   s     s
1649 	 *	0	y	y	y	(5)	c   s     s
1650 	 *	1	--	n	--	(6) c	c	c s
1651 	 *	1	--	y	--	(7) c	c   s	c s
1652 	 *
1653 	 *					(c=clear s=set)
1654 	 */
1655 	switch (type & 0xff) {
1656 	case ND_NEIGHBOR_SOLICIT:
1657 		/*
1658 		 * New entry must have is_router flag cleared.
1659 		 */
1660 		if (is_newentry)	/* (6-7) */
1661 			ln->ln_router = 0;
1662 		break;
1663 	case ND_REDIRECT:
1664 		/*
1665 		 * If the icmp is a redirect to a better router, always set the
1666 		 * is_router flag.  Otherwise, if the entry is newly created,
1667 		 * clear the flag.  [RFC 2461, sec 8.3]
1668 		 */
1669 		if (code == ND_REDIRECT_ROUTER)
1670 			ln->ln_router = 1;
1671 		else if (is_newentry) /* (6-7) */
1672 			ln->ln_router = 0;
1673 		break;
1674 	case ND_ROUTER_SOLICIT:
1675 		/*
1676 		 * is_router flag must always be cleared.
1677 		 */
1678 		ln->ln_router = 0;
1679 		break;
1680 	case ND_ROUTER_ADVERT:
1681 		/*
1682 		 * Mark an entry with lladdr as a router.
1683 		 */
1684 		if ((!is_newentry && (olladdr || lladdr)) ||	/* (2-5) */
1685 		    (is_newentry && lladdr)) {			/* (7) */
1686 			ln->ln_router = 1;
1687 		}
1688 		break;
1689 	}
1690 
1691 	if (ln != NULL) {
1692 		static_route = (ln->la_flags & LLE_STATIC);
1693 		router = ln->ln_router;
1694 
1695 		if (flags & ND6_EXCLUSIVE)
1696 			LLE_WUNLOCK(ln);
1697 		else
1698 			LLE_RUNLOCK(ln);
1699 		if (static_route)
1700 			ln = NULL;
1701 	}
1702 	if (chain)
1703 		nd6_output_flush(ifp, ifp, chain, &sin6, NULL);
1704 
1705 	/*
1706 	 * When the link-layer address of a router changes, select the
1707 	 * best router again.  In particular, when the neighbor entry is newly
1708 	 * created, it might affect the selection policy.
1709 	 * Question: can we restrict the first condition to the "is_newentry"
1710 	 * case?
1711 	 * XXX: when we hear an RA from a new router with the link-layer
1712 	 * address option, defrouter_select() is called twice, since
1713 	 * defrtrlist_update called the function as well.  However, I believe
1714 	 * we can compromise the overhead, since it only happens the first
1715 	 * time.
1716 	 * XXX: although defrouter_select() should not have a bad effect
1717 	 * for those are not autoconfigured hosts, we explicitly avoid such
1718 	 * cases for safety.
1719 	 */
1720 	if (do_update && router && !V_ip6_forwarding &&
1721 	    ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) {
1722 		/*
1723 		 * guaranteed recursion
1724 		 */
1725 		defrouter_select();
1726 	}
1727 
1728 	return (ln);
1729 done:
1730 	if (ln != NULL) {
1731 		if (flags & ND6_EXCLUSIVE)
1732 			LLE_WUNLOCK(ln);
1733 		else
1734 			LLE_RUNLOCK(ln);
1735 		if (static_route)
1736 			ln = NULL;
1737 	}
1738 	return (ln);
1739 }
1740 
1741 static void
1742 nd6_slowtimo(void *arg)
1743 {
1744 	CURVNET_SET((struct vnet *) arg);
1745 	struct nd_ifinfo *nd6if;
1746 	struct ifnet *ifp;
1747 
1748 	callout_reset(&V_nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
1749 	    nd6_slowtimo, curvnet);
1750 	IFNET_RLOCK_NOSLEEP();
1751 	for (ifp = TAILQ_FIRST(&V_ifnet); ifp;
1752 	    ifp = TAILQ_NEXT(ifp, if_list)) {
1753 		nd6if = ND_IFINFO(ifp);
1754 		if (nd6if->basereachable && /* already initialized */
1755 		    (nd6if->recalctm -= ND6_SLOWTIMER_INTERVAL) <= 0) {
1756 			/*
1757 			 * Since reachable time rarely changes by router
1758 			 * advertisements, we SHOULD insure that a new random
1759 			 * value gets recomputed at least once every few hours.
1760 			 * (RFC 2461, 6.3.4)
1761 			 */
1762 			nd6if->recalctm = V_nd6_recalc_reachtm_interval;
1763 			nd6if->reachable = ND_COMPUTE_RTIME(nd6if->basereachable);
1764 		}
1765 	}
1766 	IFNET_RUNLOCK_NOSLEEP();
1767 	CURVNET_RESTORE();
1768 }
1769 
1770 int
1771 nd6_output(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *m0,
1772     struct sockaddr_in6 *dst, struct rtentry *rt0)
1773 {
1774 
1775 	return (nd6_output_lle(ifp, origifp, m0, dst, rt0, NULL, NULL));
1776 }
1777 
1778 
1779 /*
1780  * Note that I'm not enforcing any global serialization
1781  * lle state or asked changes here as the logic is too
1782  * complicated to avoid having to always acquire an exclusive
1783  * lock
1784  * KMM
1785  *
1786  */
1787 #define senderr(e) { error = (e); goto bad;}
1788 
1789 int
1790 nd6_output_lle(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *m0,
1791     struct sockaddr_in6 *dst, struct rtentry *rt0, struct llentry *lle,
1792 	struct mbuf **chain)
1793 {
1794 	struct mbuf *m = m0;
1795 	struct llentry *ln = lle;
1796 	int error = 0;
1797 	int flags = 0;
1798 
1799 #ifdef INVARIANTS
1800 	if (lle != NULL) {
1801 
1802 		LLE_WLOCK_ASSERT(lle);
1803 
1804 		KASSERT(chain != NULL, (" lle locked but no mbuf chain pointer passed"));
1805 	}
1806 #endif
1807 	if (IN6_IS_ADDR_MULTICAST(&dst->sin6_addr))
1808 		goto sendpkt;
1809 
1810 	if (nd6_need_cache(ifp) == 0)
1811 		goto sendpkt;
1812 
1813 	/*
1814 	 * next hop determination.  This routine is derived from ether_output.
1815 	 */
1816 
1817 	/*
1818 	 * Address resolution or Neighbor Unreachability Detection
1819 	 * for the next hop.
1820 	 * At this point, the destination of the packet must be a unicast
1821 	 * or an anycast address(i.e. not a multicast).
1822 	 */
1823 
1824 	flags = ((m != NULL) || (lle != NULL)) ? LLE_EXCLUSIVE : 0;
1825 	if (ln == NULL) {
1826 	retry:
1827 		IF_AFDATA_LOCK(ifp);
1828 		ln = lla_lookup(LLTABLE6(ifp), flags, (struct sockaddr *)dst);
1829 		IF_AFDATA_UNLOCK(ifp);
1830 		if ((ln == NULL) && nd6_is_addr_neighbor(dst, ifp))  {
1831 			/*
1832 			 * Since nd6_is_addr_neighbor() internally calls nd6_lookup(),
1833 			 * the condition below is not very efficient.  But we believe
1834 			 * it is tolerable, because this should be a rare case.
1835 			 */
1836 			flags = ND6_CREATE | (m ? ND6_EXCLUSIVE : 0);
1837 			IF_AFDATA_LOCK(ifp);
1838 			ln = nd6_lookup(&dst->sin6_addr, flags, ifp);
1839 			IF_AFDATA_UNLOCK(ifp);
1840 		}
1841 	}
1842 	if (ln == NULL) {
1843 		if ((ifp->if_flags & IFF_POINTOPOINT) == 0 &&
1844 		    !(ND_IFINFO(ifp)->flags & ND6_IFF_PERFORMNUD)) {
1845 			char ip6buf[INET6_ADDRSTRLEN];
1846 			log(LOG_DEBUG,
1847 			    "nd6_output: can't allocate llinfo for %s "
1848 			    "(ln=%p)\n",
1849 			    ip6_sprintf(ip6buf, &dst->sin6_addr), ln);
1850 			senderr(EIO);	/* XXX: good error? */
1851 		}
1852 		goto sendpkt;	/* send anyway */
1853 	}
1854 
1855 	/* We don't have to do link-layer address resolution on a p2p link. */
1856 	if ((ifp->if_flags & IFF_POINTOPOINT) != 0 &&
1857 	    ln->ln_state < ND6_LLINFO_REACHABLE) {
1858 		if ((flags & LLE_EXCLUSIVE) == 0) {
1859 			flags |= LLE_EXCLUSIVE;
1860 			goto retry;
1861 		}
1862 		ln->ln_state = ND6_LLINFO_STALE;
1863 		nd6_llinfo_settimer_locked(ln, (long)V_nd6_gctimer * hz);
1864 	}
1865 
1866 	/*
1867 	 * The first time we send a packet to a neighbor whose entry is
1868 	 * STALE, we have to change the state to DELAY and a sets a timer to
1869 	 * expire in DELAY_FIRST_PROBE_TIME seconds to ensure do
1870 	 * neighbor unreachability detection on expiration.
1871 	 * (RFC 2461 7.3.3)
1872 	 */
1873 	if (ln->ln_state == ND6_LLINFO_STALE) {
1874 		if ((flags & LLE_EXCLUSIVE) == 0) {
1875 			flags |= LLE_EXCLUSIVE;
1876 			LLE_RUNLOCK(ln);
1877 			goto retry;
1878 		}
1879 		ln->la_asked = 0;
1880 		ln->ln_state = ND6_LLINFO_DELAY;
1881 		nd6_llinfo_settimer_locked(ln, (long)V_nd6_delay * hz);
1882 	}
1883 
1884 	/*
1885 	 * If the neighbor cache entry has a state other than INCOMPLETE
1886 	 * (i.e. its link-layer address is already resolved), just
1887 	 * send the packet.
1888 	 */
1889 	if (ln->ln_state > ND6_LLINFO_INCOMPLETE)
1890 		goto sendpkt;
1891 
1892 	/*
1893 	 * There is a neighbor cache entry, but no ethernet address
1894 	 * response yet.  Append this latest packet to the end of the
1895 	 * packet queue in the mbuf, unless the number of the packet
1896 	 * does not exceed nd6_maxqueuelen.  When it exceeds nd6_maxqueuelen,
1897 	 * the oldest packet in the queue will be removed.
1898 	 */
1899 	if (ln->ln_state == ND6_LLINFO_NOSTATE)
1900 		ln->ln_state = ND6_LLINFO_INCOMPLETE;
1901 
1902 	if ((flags & LLE_EXCLUSIVE) == 0) {
1903 		flags |= LLE_EXCLUSIVE;
1904 		LLE_RUNLOCK(ln);
1905 		goto retry;
1906 	}
1907 	if (ln->la_hold) {
1908 		struct mbuf *m_hold;
1909 		int i;
1910 
1911 		i = 0;
1912 		for (m_hold = ln->la_hold; m_hold; m_hold = m_hold->m_nextpkt) {
1913 			i++;
1914 			if (m_hold->m_nextpkt == NULL) {
1915 				m_hold->m_nextpkt = m;
1916 				break;
1917 			}
1918 		}
1919 		while (i >= V_nd6_maxqueuelen) {
1920 			m_hold = ln->la_hold;
1921 			ln->la_hold = ln->la_hold->m_nextpkt;
1922 			m_freem(m_hold);
1923 			i--;
1924 		}
1925 	} else {
1926 		ln->la_hold = m;
1927 	}
1928 	/*
1929 	 * We did the lookup (no lle arg) so we
1930 	 * need to do the unlock here
1931 	 */
1932 	if (lle == NULL) {
1933 		if (flags & LLE_EXCLUSIVE)
1934 			LLE_WUNLOCK(ln);
1935 		else
1936 			LLE_RUNLOCK(ln);
1937 	}
1938 
1939 	/*
1940 	 * If there has been no NS for the neighbor after entering the
1941 	 * INCOMPLETE state, send the first solicitation.
1942 	 */
1943 	if (!ND6_LLINFO_PERMANENT(ln) && ln->la_asked == 0) {
1944 		ln->la_asked++;
1945 
1946 		nd6_llinfo_settimer(ln,
1947 		    (long)ND_IFINFO(ifp)->retrans * hz / 1000);
1948 		nd6_ns_output(ifp, NULL, &dst->sin6_addr, ln, 0);
1949 	}
1950 	return (0);
1951 
1952   sendpkt:
1953 	/* discard the packet if IPv6 operation is disabled on the interface */
1954 	if ((ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED)) {
1955 		error = ENETDOWN; /* better error? */
1956 		goto bad;
1957 	}
1958 	/*
1959 	 * ln is valid and the caller did not pass in
1960 	 * an llentry
1961 	 */
1962 	if ((ln != NULL) && (lle == NULL)) {
1963 		if (flags & LLE_EXCLUSIVE)
1964 			LLE_WUNLOCK(ln);
1965 		else
1966 			LLE_RUNLOCK(ln);
1967 	}
1968 
1969 #ifdef MAC
1970 	mac_netinet6_nd6_send(ifp, m);
1971 #endif
1972 	/*
1973 	 * We were passed in a pointer to an lle with the lock held
1974 	 * this means that we can't call if_output as we will
1975 	 * recurse on the lle lock - so what we do is we create
1976 	 * a list of mbufs to send and transmit them in the caller
1977 	 * after the lock is dropped
1978 	 */
1979 	if (lle != NULL) {
1980 		if (*chain == NULL)
1981 			*chain = m;
1982 		else {
1983 			struct mbuf *m = *chain;
1984 
1985 			/*
1986 			 * append mbuf to end of deferred chain
1987 			 */
1988 			while (m->m_nextpkt != NULL)
1989 				m = m->m_nextpkt;
1990 			m->m_nextpkt = m;
1991 		}
1992 		return (error);
1993 	}
1994 	if ((ifp->if_flags & IFF_LOOPBACK) != 0) {
1995 		return ((*ifp->if_output)(origifp, m, (struct sockaddr *)dst,
1996 		    NULL));
1997 	}
1998 	error = (*ifp->if_output)(ifp, m, (struct sockaddr *)dst, NULL);
1999 	return (error);
2000 
2001   bad:
2002 	/*
2003 	 * ln is valid and the caller did not pass in
2004 	 * an llentry
2005 	 */
2006 	if ((ln != NULL) && (lle == NULL)) {
2007 		if (flags & LLE_EXCLUSIVE)
2008 			LLE_WUNLOCK(ln);
2009 		else
2010 			LLE_RUNLOCK(ln);
2011 	}
2012 	if (m)
2013 		m_freem(m);
2014 	return (error);
2015 }
2016 #undef senderr
2017 
2018 
2019 int
2020 nd6_output_flush(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *chain,
2021     struct sockaddr_in6 *dst, struct route *ro)
2022 {
2023 	struct mbuf *m, *m_head;
2024 	struct ifnet *outifp;
2025 	int error = 0;
2026 
2027 	m_head = chain;
2028 	if ((ifp->if_flags & IFF_LOOPBACK) != 0)
2029 		outifp = origifp;
2030 	else
2031 		outifp = ifp;
2032 
2033 	while (m_head) {
2034 		m = m_head;
2035 		m_head = m_head->m_nextpkt;
2036 		error = (*ifp->if_output)(ifp, m, (struct sockaddr *)dst, ro);
2037 	}
2038 
2039 	/*
2040 	 * XXX
2041 	 * note that intermediate errors are blindly ignored - but this is
2042 	 * the same convention as used with nd6_output when called by
2043 	 * nd6_cache_lladdr
2044 	 */
2045 	return (error);
2046 }
2047 
2048 
2049 int
2050 nd6_need_cache(struct ifnet *ifp)
2051 {
2052 	/*
2053 	 * XXX: we currently do not make neighbor cache on any interface
2054 	 * other than ARCnet, Ethernet, FDDI and GIF.
2055 	 *
2056 	 * RFC2893 says:
2057 	 * - unidirectional tunnels needs no ND
2058 	 */
2059 	switch (ifp->if_type) {
2060 	case IFT_ARCNET:
2061 	case IFT_ETHER:
2062 	case IFT_FDDI:
2063 	case IFT_IEEE1394:
2064 #ifdef IFT_L2VLAN
2065 	case IFT_L2VLAN:
2066 #endif
2067 #ifdef IFT_IEEE80211
2068 	case IFT_IEEE80211:
2069 #endif
2070 #ifdef IFT_CARP
2071 	case IFT_CARP:
2072 #endif
2073 	case IFT_GIF:		/* XXX need more cases? */
2074 	case IFT_PPP:
2075 	case IFT_TUNNEL:
2076 	case IFT_BRIDGE:
2077 	case IFT_PROPVIRTUAL:
2078 		return (1);
2079 	default:
2080 		return (0);
2081 	}
2082 }
2083 
2084 /*
2085  * the callers of this function need to be re-worked to drop
2086  * the lle lock, drop here for now
2087  */
2088 int
2089 nd6_storelladdr(struct ifnet *ifp, struct mbuf *m,
2090     struct sockaddr *dst, u_char *desten, struct llentry **lle)
2091 {
2092 	struct llentry *ln;
2093 
2094 	*lle = NULL;
2095 	IF_AFDATA_UNLOCK_ASSERT(ifp);
2096 	if (m->m_flags & M_MCAST) {
2097 		int i;
2098 
2099 		switch (ifp->if_type) {
2100 		case IFT_ETHER:
2101 		case IFT_FDDI:
2102 #ifdef IFT_L2VLAN
2103 		case IFT_L2VLAN:
2104 #endif
2105 #ifdef IFT_IEEE80211
2106 		case IFT_IEEE80211:
2107 #endif
2108 		case IFT_BRIDGE:
2109 		case IFT_ISO88025:
2110 			ETHER_MAP_IPV6_MULTICAST(&SIN6(dst)->sin6_addr,
2111 						 desten);
2112 			return (0);
2113 		case IFT_IEEE1394:
2114 			/*
2115 			 * netbsd can use if_broadcastaddr, but we don't do so
2116 			 * to reduce # of ifdef.
2117 			 */
2118 			for (i = 0; i < ifp->if_addrlen; i++)
2119 				desten[i] = ~0;
2120 			return (0);
2121 		case IFT_ARCNET:
2122 			*desten = 0;
2123 			return (0);
2124 		default:
2125 			m_freem(m);
2126 			return (EAFNOSUPPORT);
2127 		}
2128 	}
2129 
2130 
2131 	/*
2132 	 * the entry should have been created in nd6_store_lladdr
2133 	 */
2134 	IF_AFDATA_LOCK(ifp);
2135 	ln = lla_lookup(LLTABLE6(ifp), 0, dst);
2136 	IF_AFDATA_UNLOCK(ifp);
2137 	if ((ln == NULL) || !(ln->la_flags & LLE_VALID)) {
2138 		if (ln != NULL)
2139 			LLE_RUNLOCK(ln);
2140 		/* this could happen, if we could not allocate memory */
2141 		m_freem(m);
2142 		return (1);
2143 	}
2144 
2145 	bcopy(&ln->ll_addr, desten, ifp->if_addrlen);
2146 	*lle = ln;
2147 	LLE_RUNLOCK(ln);
2148 	/*
2149 	 * A *small* use after free race exists here
2150 	 */
2151 	return (0);
2152 }
2153 
2154 static void
2155 clear_llinfo_pqueue(struct llentry *ln)
2156 {
2157 	struct mbuf *m_hold, *m_hold_next;
2158 
2159 	for (m_hold = ln->la_hold; m_hold; m_hold = m_hold_next) {
2160 		m_hold_next = m_hold->m_nextpkt;
2161 		m_hold->m_nextpkt = NULL;
2162 		m_freem(m_hold);
2163 	}
2164 
2165 	ln->la_hold = NULL;
2166 	return;
2167 }
2168 
2169 static int nd6_sysctl_drlist(SYSCTL_HANDLER_ARGS);
2170 static int nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS);
2171 #ifdef SYSCTL_DECL
2172 SYSCTL_DECL(_net_inet6_icmp6);
2173 #endif
2174 SYSCTL_NODE(_net_inet6_icmp6, ICMPV6CTL_ND6_DRLIST, nd6_drlist,
2175 	CTLFLAG_RD, nd6_sysctl_drlist, "");
2176 SYSCTL_NODE(_net_inet6_icmp6, ICMPV6CTL_ND6_PRLIST, nd6_prlist,
2177 	CTLFLAG_RD, nd6_sysctl_prlist, "");
2178 SYSCTL_VNET_INT(_net_inet6_icmp6, ICMPV6CTL_ND6_MAXQLEN, nd6_maxqueuelen,
2179 	CTLFLAG_RW, &VNET_NAME(nd6_maxqueuelen), 1, "");
2180 
2181 static int
2182 nd6_sysctl_drlist(SYSCTL_HANDLER_ARGS)
2183 {
2184 	int error;
2185 	char buf[1024] __aligned(4);
2186 	struct in6_defrouter *d, *de;
2187 	struct nd_defrouter *dr;
2188 
2189 	if (req->newptr)
2190 		return EPERM;
2191 	error = 0;
2192 
2193 	for (dr = TAILQ_FIRST(&V_nd_defrouter); dr;
2194 	     dr = TAILQ_NEXT(dr, dr_entry)) {
2195 		d = (struct in6_defrouter *)buf;
2196 		de = (struct in6_defrouter *)(buf + sizeof(buf));
2197 
2198 		if (d + 1 <= de) {
2199 			bzero(d, sizeof(*d));
2200 			d->rtaddr.sin6_family = AF_INET6;
2201 			d->rtaddr.sin6_len = sizeof(d->rtaddr);
2202 			d->rtaddr.sin6_addr = dr->rtaddr;
2203 			error = sa6_recoverscope(&d->rtaddr);
2204 			if (error != 0)
2205 				return (error);
2206 			d->flags = dr->flags;
2207 			d->rtlifetime = dr->rtlifetime;
2208 			d->expire = dr->expire;
2209 			d->if_index = dr->ifp->if_index;
2210 		} else
2211 			panic("buffer too short");
2212 
2213 		error = SYSCTL_OUT(req, buf, sizeof(*d));
2214 		if (error)
2215 			break;
2216 	}
2217 
2218 	return (error);
2219 }
2220 
2221 static int
2222 nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS)
2223 {
2224 	int error;
2225 	char buf[1024] __aligned(4);
2226 	struct in6_prefix *p, *pe;
2227 	struct nd_prefix *pr;
2228 	char ip6buf[INET6_ADDRSTRLEN];
2229 
2230 	if (req->newptr)
2231 		return EPERM;
2232 	error = 0;
2233 
2234 	for (pr = V_nd_prefix.lh_first; pr; pr = pr->ndpr_next) {
2235 		u_short advrtrs;
2236 		size_t advance;
2237 		struct sockaddr_in6 *sin6, *s6;
2238 		struct nd_pfxrouter *pfr;
2239 
2240 		p = (struct in6_prefix *)buf;
2241 		pe = (struct in6_prefix *)(buf + sizeof(buf));
2242 
2243 		if (p + 1 <= pe) {
2244 			bzero(p, sizeof(*p));
2245 			sin6 = (struct sockaddr_in6 *)(p + 1);
2246 
2247 			p->prefix = pr->ndpr_prefix;
2248 			if (sa6_recoverscope(&p->prefix)) {
2249 				log(LOG_ERR,
2250 				    "scope error in prefix list (%s)\n",
2251 				    ip6_sprintf(ip6buf, &p->prefix.sin6_addr));
2252 				/* XXX: press on... */
2253 			}
2254 			p->raflags = pr->ndpr_raf;
2255 			p->prefixlen = pr->ndpr_plen;
2256 			p->vltime = pr->ndpr_vltime;
2257 			p->pltime = pr->ndpr_pltime;
2258 			p->if_index = pr->ndpr_ifp->if_index;
2259 			if (pr->ndpr_vltime == ND6_INFINITE_LIFETIME)
2260 				p->expire = 0;
2261 			else {
2262 				time_t maxexpire;
2263 
2264 				/* XXX: we assume time_t is signed. */
2265 				maxexpire = (-1) &
2266 				    ~((time_t)1 <<
2267 				    ((sizeof(maxexpire) * 8) - 1));
2268 				if (pr->ndpr_vltime <
2269 				    maxexpire - pr->ndpr_lastupdate) {
2270 				    p->expire = pr->ndpr_lastupdate +
2271 				        pr->ndpr_vltime;
2272 				} else
2273 					p->expire = maxexpire;
2274 			}
2275 			p->refcnt = pr->ndpr_refcnt;
2276 			p->flags = pr->ndpr_stateflags;
2277 			p->origin = PR_ORIG_RA;
2278 			advrtrs = 0;
2279 			for (pfr = pr->ndpr_advrtrs.lh_first; pfr;
2280 			     pfr = pfr->pfr_next) {
2281 				if ((void *)&sin6[advrtrs + 1] > (void *)pe) {
2282 					advrtrs++;
2283 					continue;
2284 				}
2285 				s6 = &sin6[advrtrs];
2286 				bzero(s6, sizeof(*s6));
2287 				s6->sin6_family = AF_INET6;
2288 				s6->sin6_len = sizeof(*sin6);
2289 				s6->sin6_addr = pfr->router->rtaddr;
2290 				if (sa6_recoverscope(s6)) {
2291 					log(LOG_ERR,
2292 					    "scope error in "
2293 					    "prefix list (%s)\n",
2294 					    ip6_sprintf(ip6buf,
2295 						    &pfr->router->rtaddr));
2296 				}
2297 				advrtrs++;
2298 			}
2299 			p->advrtrs = advrtrs;
2300 		} else
2301 			panic("buffer too short");
2302 
2303 		advance = sizeof(*p) + sizeof(*sin6) * advrtrs;
2304 		error = SYSCTL_OUT(req, buf, advance);
2305 		if (error)
2306 			break;
2307 	}
2308 
2309 	return (error);
2310 }
2311