xref: /freebsd/sys/netinet6/nd6.c (revision f677a9e2672665f4eb3dd4111c07ee8f1f954262)
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 #include "opt_kdtrace.h"
38 
39 #include <sys/param.h>
40 #include <sys/systm.h>
41 #include <sys/callout.h>
42 #include <sys/malloc.h>
43 #include <sys/mbuf.h>
44 #include <sys/socket.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/syslog.h>
51 #include <sys/lock.h>
52 #include <sys/rwlock.h>
53 #include <sys/queue.h>
54 #include <sys/sdt.h>
55 #include <sys/sysctl.h>
56 
57 #include <net/if.h>
58 #include <net/if_arc.h>
59 #include <net/if_dl.h>
60 #include <net/if_types.h>
61 #include <net/iso88025.h>
62 #include <net/fddi.h>
63 #include <net/route.h>
64 #include <net/vnet.h>
65 
66 #include <netinet/in.h>
67 #include <netinet/in_kdtrace.h>
68 #include <net/if_llatbl.h>
69 #define	L3_ADDR_SIN6(le)	((struct sockaddr_in6 *) L3_ADDR(le))
70 #include <netinet/if_ether.h>
71 #include <netinet6/in6_var.h>
72 #include <netinet/ip6.h>
73 #include <netinet6/ip6_var.h>
74 #include <netinet6/scope6_var.h>
75 #include <netinet6/nd6.h>
76 #include <netinet6/in6_ifattach.h>
77 #include <netinet/icmp6.h>
78 #include <netinet6/send.h>
79 
80 #include <sys/limits.h>
81 
82 #include <security/mac/mac_framework.h>
83 
84 #define ND6_SLOWTIMER_INTERVAL (60 * 60) /* 1 hour */
85 #define ND6_RECALC_REACHTM_INTERVAL (60 * 120) /* 2 hours */
86 
87 #define SIN6(s) ((const struct sockaddr_in6 *)(s))
88 
89 /* timer values */
90 VNET_DEFINE(int, nd6_prune)	= 1;	/* walk list every 1 seconds */
91 VNET_DEFINE(int, nd6_delay)	= 5;	/* delay first probe time 5 second */
92 VNET_DEFINE(int, nd6_umaxtries)	= 3;	/* maximum unicast query */
93 VNET_DEFINE(int, nd6_mmaxtries)	= 3;	/* maximum multicast query */
94 VNET_DEFINE(int, nd6_useloopback) = 1;	/* use loopback interface for
95 					 * local traffic */
96 VNET_DEFINE(int, nd6_gctimer)	= (60 * 60 * 24); /* 1 day: garbage
97 					 * collection timer */
98 
99 /* preventing too many loops in ND option parsing */
100 static VNET_DEFINE(int, nd6_maxndopt) = 10; /* max # of ND options allowed */
101 
102 VNET_DEFINE(int, nd6_maxnudhint) = 0;	/* max # of subsequent upper
103 					 * layer hints */
104 static VNET_DEFINE(int, nd6_maxqueuelen) = 1; /* max pkts cached in unresolved
105 					 * ND entries */
106 #define	V_nd6_maxndopt			VNET(nd6_maxndopt)
107 #define	V_nd6_maxqueuelen		VNET(nd6_maxqueuelen)
108 
109 #ifdef ND6_DEBUG
110 VNET_DEFINE(int, nd6_debug) = 1;
111 #else
112 VNET_DEFINE(int, nd6_debug) = 0;
113 #endif
114 
115 /* for debugging? */
116 #if 0
117 static int nd6_inuse, nd6_allocated;
118 #endif
119 
120 VNET_DEFINE(struct nd_drhead, nd_defrouter);
121 VNET_DEFINE(struct nd_prhead, nd_prefix);
122 
123 VNET_DEFINE(int, nd6_recalc_reachtm_interval) = ND6_RECALC_REACHTM_INTERVAL;
124 #define	V_nd6_recalc_reachtm_interval	VNET(nd6_recalc_reachtm_interval)
125 
126 int	(*send_sendso_input_hook)(struct mbuf *, struct ifnet *, int, int);
127 
128 static int nd6_is_new_addr_neighbor(struct sockaddr_in6 *,
129 	struct ifnet *);
130 static void nd6_setmtu0(struct ifnet *, struct nd_ifinfo *);
131 static void nd6_slowtimo(void *);
132 static int regen_tmpaddr(struct in6_ifaddr *);
133 static struct llentry *nd6_free(struct llentry *, int);
134 static void nd6_llinfo_timer(void *);
135 static void clear_llinfo_pqueue(struct llentry *);
136 
137 static VNET_DEFINE(struct callout, nd6_slowtimo_ch);
138 #define	V_nd6_slowtimo_ch		VNET(nd6_slowtimo_ch)
139 
140 VNET_DEFINE(struct callout, nd6_timer_ch);
141 
142 void
143 nd6_init(void)
144 {
145 
146 	LIST_INIT(&V_nd_prefix);
147 
148 	/* initialization of the default router list */
149 	TAILQ_INIT(&V_nd_defrouter);
150 
151 	/* start timer */
152 	callout_init(&V_nd6_slowtimo_ch, 0);
153 	callout_reset(&V_nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
154 	    nd6_slowtimo, curvnet);
155 }
156 
157 #ifdef VIMAGE
158 void
159 nd6_destroy()
160 {
161 
162 	callout_drain(&V_nd6_slowtimo_ch);
163 	callout_drain(&V_nd6_timer_ch);
164 }
165 #endif
166 
167 struct nd_ifinfo *
168 nd6_ifattach(struct ifnet *ifp)
169 {
170 	struct nd_ifinfo *nd;
171 
172 	nd = (struct nd_ifinfo *)malloc(sizeof(*nd), M_IP6NDP, M_WAITOK|M_ZERO);
173 	nd->initialized = 1;
174 
175 	nd->chlim = IPV6_DEFHLIM;
176 	nd->basereachable = REACHABLE_TIME;
177 	nd->reachable = ND_COMPUTE_RTIME(nd->basereachable);
178 	nd->retrans = RETRANS_TIMER;
179 
180 	nd->flags = ND6_IFF_PERFORMNUD;
181 
182 	/* A loopback interface always has ND6_IFF_AUTO_LINKLOCAL.
183 	 * XXXHRS: Clear ND6_IFF_AUTO_LINKLOCAL on an IFT_BRIDGE interface by
184 	 * default regardless of the V_ip6_auto_linklocal configuration to
185 	 * give a reasonable default behavior.
186 	 */
187 	if ((V_ip6_auto_linklocal && ifp->if_type != IFT_BRIDGE) ||
188 	    (ifp->if_flags & IFF_LOOPBACK))
189 		nd->flags |= ND6_IFF_AUTO_LINKLOCAL;
190 	/*
191 	 * A loopback interface does not need to accept RTADV.
192 	 * XXXHRS: Clear ND6_IFF_ACCEPT_RTADV on an IFT_BRIDGE interface by
193 	 * default regardless of the V_ip6_accept_rtadv configuration to
194 	 * prevent the interface from accepting RA messages arrived
195 	 * on one of the member interfaces with ND6_IFF_ACCEPT_RTADV.
196 	 */
197 	if (V_ip6_accept_rtadv &&
198 	    !(ifp->if_flags & IFF_LOOPBACK) &&
199 	    (ifp->if_type != IFT_BRIDGE))
200 			nd->flags |= ND6_IFF_ACCEPT_RTADV;
201 	if (V_ip6_no_radr && !(ifp->if_flags & IFF_LOOPBACK))
202 		nd->flags |= ND6_IFF_NO_RADR;
203 
204 	/* XXX: we cannot call nd6_setmtu since ifp is not fully initialized */
205 	nd6_setmtu0(ifp, nd);
206 
207 	return nd;
208 }
209 
210 void
211 nd6_ifdetach(struct nd_ifinfo *nd)
212 {
213 
214 	free(nd, M_IP6NDP);
215 }
216 
217 /*
218  * Reset ND level link MTU. This function is called when the physical MTU
219  * changes, which means we might have to adjust the ND level MTU.
220  */
221 void
222 nd6_setmtu(struct ifnet *ifp)
223 {
224 
225 	nd6_setmtu0(ifp, ND_IFINFO(ifp));
226 }
227 
228 /* XXX todo: do not maintain copy of ifp->if_mtu in ndi->maxmtu */
229 void
230 nd6_setmtu0(struct ifnet *ifp, struct nd_ifinfo *ndi)
231 {
232 	u_int32_t omaxmtu;
233 
234 	omaxmtu = ndi->maxmtu;
235 
236 	switch (ifp->if_type) {
237 	case IFT_ARCNET:
238 		ndi->maxmtu = MIN(ARC_PHDS_MAXMTU, ifp->if_mtu); /* RFC2497 */
239 		break;
240 	case IFT_FDDI:
241 		ndi->maxmtu = MIN(FDDIIPMTU, ifp->if_mtu); /* RFC2467 */
242 		break;
243 	case IFT_ISO88025:
244 		 ndi->maxmtu = MIN(ISO88025_MAX_MTU, ifp->if_mtu);
245 		 break;
246 	default:
247 		ndi->maxmtu = ifp->if_mtu;
248 		break;
249 	}
250 
251 	/*
252 	 * Decreasing the interface MTU under IPV6 minimum MTU may cause
253 	 * undesirable situation.  We thus notify the operator of the change
254 	 * explicitly.  The check for omaxmtu is necessary to restrict the
255 	 * log to the case of changing the MTU, not initializing it.
256 	 */
257 	if (omaxmtu >= IPV6_MMTU && ndi->maxmtu < IPV6_MMTU) {
258 		log(LOG_NOTICE, "nd6_setmtu0: "
259 		    "new link MTU on %s (%lu) is too small for IPv6\n",
260 		    if_name(ifp), (unsigned long)ndi->maxmtu);
261 	}
262 
263 	if (ndi->maxmtu > V_in6_maxmtu)
264 		in6_setmaxmtu(); /* check all interfaces just in case */
265 
266 }
267 
268 void
269 nd6_option_init(void *opt, int icmp6len, union nd_opts *ndopts)
270 {
271 
272 	bzero(ndopts, sizeof(*ndopts));
273 	ndopts->nd_opts_search = (struct nd_opt_hdr *)opt;
274 	ndopts->nd_opts_last
275 		= (struct nd_opt_hdr *)(((u_char *)opt) + icmp6len);
276 
277 	if (icmp6len == 0) {
278 		ndopts->nd_opts_done = 1;
279 		ndopts->nd_opts_search = NULL;
280 	}
281 }
282 
283 /*
284  * Take one ND option.
285  */
286 struct nd_opt_hdr *
287 nd6_option(union nd_opts *ndopts)
288 {
289 	struct nd_opt_hdr *nd_opt;
290 	int olen;
291 
292 	KASSERT(ndopts != NULL, ("%s: ndopts == NULL", __func__));
293 	KASSERT(ndopts->nd_opts_last != NULL, ("%s: uninitialized ndopts",
294 	    __func__));
295 	if (ndopts->nd_opts_search == NULL)
296 		return NULL;
297 	if (ndopts->nd_opts_done)
298 		return NULL;
299 
300 	nd_opt = ndopts->nd_opts_search;
301 
302 	/* make sure nd_opt_len is inside the buffer */
303 	if ((caddr_t)&nd_opt->nd_opt_len >= (caddr_t)ndopts->nd_opts_last) {
304 		bzero(ndopts, sizeof(*ndopts));
305 		return NULL;
306 	}
307 
308 	olen = nd_opt->nd_opt_len << 3;
309 	if (olen == 0) {
310 		/*
311 		 * Message validation requires that all included
312 		 * options have a length that is greater than zero.
313 		 */
314 		bzero(ndopts, sizeof(*ndopts));
315 		return NULL;
316 	}
317 
318 	ndopts->nd_opts_search = (struct nd_opt_hdr *)((caddr_t)nd_opt + olen);
319 	if (ndopts->nd_opts_search > ndopts->nd_opts_last) {
320 		/* option overruns the end of buffer, invalid */
321 		bzero(ndopts, sizeof(*ndopts));
322 		return NULL;
323 	} else if (ndopts->nd_opts_search == ndopts->nd_opts_last) {
324 		/* reached the end of options chain */
325 		ndopts->nd_opts_done = 1;
326 		ndopts->nd_opts_search = NULL;
327 	}
328 	return nd_opt;
329 }
330 
331 /*
332  * Parse multiple ND options.
333  * This function is much easier to use, for ND routines that do not need
334  * multiple options of the same type.
335  */
336 int
337 nd6_options(union nd_opts *ndopts)
338 {
339 	struct nd_opt_hdr *nd_opt;
340 	int i = 0;
341 
342 	KASSERT(ndopts != NULL, ("%s: ndopts == NULL", __func__));
343 	KASSERT(ndopts->nd_opts_last != NULL, ("%s: uninitialized ndopts",
344 	    __func__));
345 	if (ndopts->nd_opts_search == NULL)
346 		return 0;
347 
348 	while (1) {
349 		nd_opt = nd6_option(ndopts);
350 		if (nd_opt == NULL && ndopts->nd_opts_last == NULL) {
351 			/*
352 			 * Message validation requires that all included
353 			 * options have a length that is greater than zero.
354 			 */
355 			ICMP6STAT_INC(icp6s_nd_badopt);
356 			bzero(ndopts, sizeof(*ndopts));
357 			return -1;
358 		}
359 
360 		if (nd_opt == NULL)
361 			goto skip1;
362 
363 		switch (nd_opt->nd_opt_type) {
364 		case ND_OPT_SOURCE_LINKADDR:
365 		case ND_OPT_TARGET_LINKADDR:
366 		case ND_OPT_MTU:
367 		case ND_OPT_REDIRECTED_HEADER:
368 			if (ndopts->nd_opt_array[nd_opt->nd_opt_type]) {
369 				nd6log((LOG_INFO,
370 				    "duplicated ND6 option found (type=%d)\n",
371 				    nd_opt->nd_opt_type));
372 				/* XXX bark? */
373 			} else {
374 				ndopts->nd_opt_array[nd_opt->nd_opt_type]
375 					= nd_opt;
376 			}
377 			break;
378 		case ND_OPT_PREFIX_INFORMATION:
379 			if (ndopts->nd_opt_array[nd_opt->nd_opt_type] == 0) {
380 				ndopts->nd_opt_array[nd_opt->nd_opt_type]
381 					= nd_opt;
382 			}
383 			ndopts->nd_opts_pi_end =
384 				(struct nd_opt_prefix_info *)nd_opt;
385 			break;
386 		/* What about ND_OPT_ROUTE_INFO? RFC 4191 */
387 		case ND_OPT_RDNSS:	/* RFC 6106 */
388 		case ND_OPT_DNSSL:	/* RFC 6106 */
389 			/*
390 			 * Silently ignore options we know and do not care about
391 			 * in the kernel.
392 			 */
393 			break;
394 		default:
395 			/*
396 			 * Unknown options must be silently ignored,
397 			 * to accomodate future extension to the protocol.
398 			 */
399 			nd6log((LOG_DEBUG,
400 			    "nd6_options: unsupported option %d - "
401 			    "option ignored\n", nd_opt->nd_opt_type));
402 		}
403 
404 skip1:
405 		i++;
406 		if (i > V_nd6_maxndopt) {
407 			ICMP6STAT_INC(icp6s_nd_toomanyopt);
408 			nd6log((LOG_INFO, "too many loop in nd opt\n"));
409 			break;
410 		}
411 
412 		if (ndopts->nd_opts_done)
413 			break;
414 	}
415 
416 	return 0;
417 }
418 
419 /*
420  * ND6 timer routine to handle ND6 entries
421  */
422 void
423 nd6_llinfo_settimer_locked(struct llentry *ln, long tick)
424 {
425 	int canceled;
426 
427 	LLE_WLOCK_ASSERT(ln);
428 
429 	if (tick < 0) {
430 		ln->la_expire = 0;
431 		ln->ln_ntick = 0;
432 		canceled = callout_stop(&ln->ln_timer_ch);
433 	} else {
434 		ln->la_expire = time_uptime + tick / hz;
435 		LLE_ADDREF(ln);
436 		if (tick > INT_MAX) {
437 			ln->ln_ntick = tick - INT_MAX;
438 			canceled = callout_reset(&ln->ln_timer_ch, INT_MAX,
439 			    nd6_llinfo_timer, ln);
440 		} else {
441 			ln->ln_ntick = 0;
442 			canceled = callout_reset(&ln->ln_timer_ch, tick,
443 			    nd6_llinfo_timer, ln);
444 		}
445 	}
446 	if (canceled)
447 		LLE_REMREF(ln);
448 }
449 
450 void
451 nd6_llinfo_settimer(struct llentry *ln, long tick)
452 {
453 
454 	LLE_WLOCK(ln);
455 	nd6_llinfo_settimer_locked(ln, tick);
456 	LLE_WUNLOCK(ln);
457 }
458 
459 static void
460 nd6_llinfo_timer(void *arg)
461 {
462 	struct llentry *ln;
463 	struct in6_addr *dst;
464 	struct ifnet *ifp;
465 	struct nd_ifinfo *ndi = NULL;
466 
467 	KASSERT(arg != NULL, ("%s: arg NULL", __func__));
468 	ln = (struct llentry *)arg;
469 	LLE_WLOCK_ASSERT(ln);
470 	ifp = ln->lle_tbl->llt_ifp;
471 
472 	CURVNET_SET(ifp->if_vnet);
473 
474 	if (ln->ln_ntick > 0) {
475 		if (ln->ln_ntick > INT_MAX) {
476 			ln->ln_ntick -= INT_MAX;
477 			nd6_llinfo_settimer_locked(ln, INT_MAX);
478 		} else {
479 			ln->ln_ntick = 0;
480 			nd6_llinfo_settimer_locked(ln, ln->ln_ntick);
481 		}
482 		goto done;
483 	}
484 
485 	ndi = ND_IFINFO(ifp);
486 	dst = &L3_ADDR_SIN6(ln)->sin6_addr;
487 	if (ln->la_flags & LLE_STATIC) {
488 		goto done;
489 	}
490 
491 	if (ln->la_flags & LLE_DELETED) {
492 		(void)nd6_free(ln, 0);
493 		ln = NULL;
494 		goto done;
495 	}
496 
497 	switch (ln->ln_state) {
498 	case ND6_LLINFO_INCOMPLETE:
499 		if (ln->la_asked < V_nd6_mmaxtries) {
500 			ln->la_asked++;
501 			nd6_llinfo_settimer_locked(ln, (long)ndi->retrans * hz / 1000);
502 			LLE_WUNLOCK(ln);
503 			nd6_ns_output(ifp, NULL, dst, ln, 0);
504 			LLE_WLOCK(ln);
505 		} else {
506 			struct mbuf *m = ln->la_hold;
507 			if (m) {
508 				struct mbuf *m0;
509 
510 				/*
511 				 * assuming every packet in la_hold has the
512 				 * same IP header.  Send error after unlock.
513 				 */
514 				m0 = m->m_nextpkt;
515 				m->m_nextpkt = NULL;
516 				ln->la_hold = m0;
517 				clear_llinfo_pqueue(ln);
518 			}
519 			EVENTHANDLER_INVOKE(lle_event, ln, LLENTRY_TIMEDOUT);
520 			(void)nd6_free(ln, 0);
521 			ln = NULL;
522 			if (m != NULL)
523 				icmp6_error2(m, ICMP6_DST_UNREACH,
524 				    ICMP6_DST_UNREACH_ADDR, 0, ifp);
525 		}
526 		break;
527 	case ND6_LLINFO_REACHABLE:
528 		if (!ND6_LLINFO_PERMANENT(ln)) {
529 			ln->ln_state = ND6_LLINFO_STALE;
530 			nd6_llinfo_settimer_locked(ln, (long)V_nd6_gctimer * hz);
531 		}
532 		break;
533 
534 	case ND6_LLINFO_STALE:
535 		/* Garbage Collection(RFC 2461 5.3) */
536 		if (!ND6_LLINFO_PERMANENT(ln)) {
537 			EVENTHANDLER_INVOKE(lle_event, ln, LLENTRY_EXPIRED);
538 			(void)nd6_free(ln, 1);
539 			ln = NULL;
540 		}
541 		break;
542 
543 	case ND6_LLINFO_DELAY:
544 		if (ndi && (ndi->flags & ND6_IFF_PERFORMNUD) != 0) {
545 			/* We need NUD */
546 			ln->la_asked = 1;
547 			ln->ln_state = ND6_LLINFO_PROBE;
548 			nd6_llinfo_settimer_locked(ln, (long)ndi->retrans * hz / 1000);
549 			LLE_WUNLOCK(ln);
550 			nd6_ns_output(ifp, dst, dst, ln, 0);
551 			LLE_WLOCK(ln);
552 		} else {
553 			ln->ln_state = ND6_LLINFO_STALE; /* XXX */
554 			nd6_llinfo_settimer_locked(ln, (long)V_nd6_gctimer * hz);
555 		}
556 		break;
557 	case ND6_LLINFO_PROBE:
558 		if (ln->la_asked < V_nd6_umaxtries) {
559 			ln->la_asked++;
560 			nd6_llinfo_settimer_locked(ln, (long)ndi->retrans * hz / 1000);
561 			LLE_WUNLOCK(ln);
562 			nd6_ns_output(ifp, dst, dst, ln, 0);
563 			LLE_WLOCK(ln);
564 		} else {
565 			EVENTHANDLER_INVOKE(lle_event, ln, LLENTRY_EXPIRED);
566 			(void)nd6_free(ln, 0);
567 			ln = NULL;
568 		}
569 		break;
570 	default:
571 		panic("%s: paths in a dark night can be confusing: %d",
572 		    __func__, ln->ln_state);
573 	}
574 done:
575 	if (ln != NULL)
576 		LLE_FREE_LOCKED(ln);
577 	CURVNET_RESTORE();
578 }
579 
580 
581 /*
582  * ND6 timer routine to expire default route list and prefix list
583  */
584 void
585 nd6_timer(void *arg)
586 {
587 	CURVNET_SET((struct vnet *) arg);
588 	struct nd_defrouter *dr, *ndr;
589 	struct nd_prefix *pr, *npr;
590 	struct in6_ifaddr *ia6, *nia6;
591 
592 	callout_reset(&V_nd6_timer_ch, V_nd6_prune * hz,
593 	    nd6_timer, curvnet);
594 
595 	/* expire default router list */
596 	TAILQ_FOREACH_SAFE(dr, &V_nd_defrouter, dr_entry, ndr) {
597 		if (dr->expire && dr->expire < time_uptime)
598 			defrtrlist_del(dr);
599 	}
600 
601 	/*
602 	 * expire interface addresses.
603 	 * in the past the loop was inside prefix expiry processing.
604 	 * However, from a stricter speci-confrmance standpoint, we should
605 	 * rather separate address lifetimes and prefix lifetimes.
606 	 *
607 	 * XXXRW: in6_ifaddrhead locking.
608 	 */
609   addrloop:
610 	TAILQ_FOREACH_SAFE(ia6, &V_in6_ifaddrhead, ia_link, nia6) {
611 		/* check address lifetime */
612 		if (IFA6_IS_INVALID(ia6)) {
613 			int regen = 0;
614 
615 			/*
616 			 * If the expiring address is temporary, try
617 			 * regenerating a new one.  This would be useful when
618 			 * we suspended a laptop PC, then turned it on after a
619 			 * period that could invalidate all temporary
620 			 * addresses.  Although we may have to restart the
621 			 * loop (see below), it must be after purging the
622 			 * address.  Otherwise, we'd see an infinite loop of
623 			 * regeneration.
624 			 */
625 			if (V_ip6_use_tempaddr &&
626 			    (ia6->ia6_flags & IN6_IFF_TEMPORARY) != 0) {
627 				if (regen_tmpaddr(ia6) == 0)
628 					regen = 1;
629 			}
630 
631 			in6_purgeaddr(&ia6->ia_ifa);
632 
633 			if (regen)
634 				goto addrloop; /* XXX: see below */
635 		} else if (IFA6_IS_DEPRECATED(ia6)) {
636 			int oldflags = ia6->ia6_flags;
637 
638 			ia6->ia6_flags |= IN6_IFF_DEPRECATED;
639 
640 			/*
641 			 * If a temporary address has just become deprecated,
642 			 * regenerate a new one if possible.
643 			 */
644 			if (V_ip6_use_tempaddr &&
645 			    (ia6->ia6_flags & IN6_IFF_TEMPORARY) != 0 &&
646 			    (oldflags & IN6_IFF_DEPRECATED) == 0) {
647 
648 				if (regen_tmpaddr(ia6) == 0) {
649 					/*
650 					 * A new temporary address is
651 					 * generated.
652 					 * XXX: this means the address chain
653 					 * has changed while we are still in
654 					 * the loop.  Although the change
655 					 * would not cause disaster (because
656 					 * it's not a deletion, but an
657 					 * addition,) we'd rather restart the
658 					 * loop just for safety.  Or does this
659 					 * significantly reduce performance??
660 					 */
661 					goto addrloop;
662 				}
663 			}
664 		} else {
665 			/*
666 			 * A new RA might have made a deprecated address
667 			 * preferred.
668 			 */
669 			ia6->ia6_flags &= ~IN6_IFF_DEPRECATED;
670 		}
671 	}
672 
673 	/* expire prefix list */
674 	LIST_FOREACH_SAFE(pr, &V_nd_prefix, ndpr_entry, npr) {
675 		/*
676 		 * check prefix lifetime.
677 		 * since pltime is just for autoconf, pltime processing for
678 		 * prefix is not necessary.
679 		 */
680 		if (pr->ndpr_vltime != ND6_INFINITE_LIFETIME &&
681 		    time_uptime - pr->ndpr_lastupdate > pr->ndpr_vltime) {
682 
683 			/*
684 			 * address expiration and prefix expiration are
685 			 * separate.  NEVER perform in6_purgeaddr here.
686 			 */
687 			prelist_remove(pr);
688 		}
689 	}
690 	CURVNET_RESTORE();
691 }
692 
693 /*
694  * ia6 - deprecated/invalidated temporary address
695  */
696 static int
697 regen_tmpaddr(struct in6_ifaddr *ia6)
698 {
699 	struct ifaddr *ifa;
700 	struct ifnet *ifp;
701 	struct in6_ifaddr *public_ifa6 = NULL;
702 
703 	ifp = ia6->ia_ifa.ifa_ifp;
704 	IF_ADDR_RLOCK(ifp);
705 	TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
706 		struct in6_ifaddr *it6;
707 
708 		if (ifa->ifa_addr->sa_family != AF_INET6)
709 			continue;
710 
711 		it6 = (struct in6_ifaddr *)ifa;
712 
713 		/* ignore no autoconf addresses. */
714 		if ((it6->ia6_flags & IN6_IFF_AUTOCONF) == 0)
715 			continue;
716 
717 		/* ignore autoconf addresses with different prefixes. */
718 		if (it6->ia6_ndpr == NULL || it6->ia6_ndpr != ia6->ia6_ndpr)
719 			continue;
720 
721 		/*
722 		 * Now we are looking at an autoconf address with the same
723 		 * prefix as ours.  If the address is temporary and is still
724 		 * preferred, do not create another one.  It would be rare, but
725 		 * could happen, for example, when we resume a laptop PC after
726 		 * a long period.
727 		 */
728 		if ((it6->ia6_flags & IN6_IFF_TEMPORARY) != 0 &&
729 		    !IFA6_IS_DEPRECATED(it6)) {
730 			public_ifa6 = NULL;
731 			break;
732 		}
733 
734 		/*
735 		 * This is a public autoconf address that has the same prefix
736 		 * as ours.  If it is preferred, keep it.  We can't break the
737 		 * loop here, because there may be a still-preferred temporary
738 		 * address with the prefix.
739 		 */
740 		if (!IFA6_IS_DEPRECATED(it6))
741 		    public_ifa6 = it6;
742 
743 		if (public_ifa6 != NULL)
744 			ifa_ref(&public_ifa6->ia_ifa);
745 	}
746 	IF_ADDR_RUNLOCK(ifp);
747 
748 	if (public_ifa6 != NULL) {
749 		int e;
750 
751 		if ((e = in6_tmpifadd(public_ifa6, 0, 0)) != 0) {
752 			ifa_free(&public_ifa6->ia_ifa);
753 			log(LOG_NOTICE, "regen_tmpaddr: failed to create a new"
754 			    " tmp addr,errno=%d\n", e);
755 			return (-1);
756 		}
757 		ifa_free(&public_ifa6->ia_ifa);
758 		return (0);
759 	}
760 
761 	return (-1);
762 }
763 
764 /*
765  * Nuke neighbor cache/prefix/default router management table, right before
766  * ifp goes away.
767  */
768 void
769 nd6_purge(struct ifnet *ifp)
770 {
771 	struct nd_defrouter *dr, *ndr;
772 	struct nd_prefix *pr, *npr;
773 
774 	/*
775 	 * Nuke default router list entries toward ifp.
776 	 * We defer removal of default router list entries that is installed
777 	 * in the routing table, in order to keep additional side effects as
778 	 * small as possible.
779 	 */
780 	TAILQ_FOREACH_SAFE(dr, &V_nd_defrouter, dr_entry, ndr) {
781 		if (dr->installed)
782 			continue;
783 
784 		if (dr->ifp == ifp)
785 			defrtrlist_del(dr);
786 	}
787 
788 	TAILQ_FOREACH_SAFE(dr, &V_nd_defrouter, dr_entry, ndr) {
789 		if (!dr->installed)
790 			continue;
791 
792 		if (dr->ifp == ifp)
793 			defrtrlist_del(dr);
794 	}
795 
796 	/* Nuke prefix list entries toward ifp */
797 	LIST_FOREACH_SAFE(pr, &V_nd_prefix, ndpr_entry, npr) {
798 		if (pr->ndpr_ifp == ifp) {
799 			/*
800 			 * Because if_detach() does *not* release prefixes
801 			 * while purging addresses the reference count will
802 			 * still be above zero. We therefore reset it to
803 			 * make sure that the prefix really gets purged.
804 			 */
805 			pr->ndpr_refcnt = 0;
806 
807 			/*
808 			 * Previously, pr->ndpr_addr is removed as well,
809 			 * but I strongly believe we don't have to do it.
810 			 * nd6_purge() is only called from in6_ifdetach(),
811 			 * which removes all the associated interface addresses
812 			 * by itself.
813 			 * (jinmei@kame.net 20010129)
814 			 */
815 			prelist_remove(pr);
816 		}
817 	}
818 
819 	/* cancel default outgoing interface setting */
820 	if (V_nd6_defifindex == ifp->if_index)
821 		nd6_setdefaultiface(0);
822 
823 	if (ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) {
824 		/* Refresh default router list. */
825 		defrouter_select();
826 	}
827 
828 	/* XXXXX
829 	 * We do not nuke the neighbor cache entries here any more
830 	 * because the neighbor cache is kept in if_afdata[AF_INET6].
831 	 * nd6_purge() is invoked by in6_ifdetach() which is called
832 	 * from if_detach() where everything gets purged. So let
833 	 * in6_domifdetach() do the actual L2 table purging work.
834 	 */
835 }
836 
837 /*
838  * the caller acquires and releases the lock on the lltbls
839  * Returns the llentry locked
840  */
841 struct llentry *
842 nd6_lookup(struct in6_addr *addr6, int flags, struct ifnet *ifp)
843 {
844 	struct sockaddr_in6 sin6;
845 	struct llentry *ln;
846 	int llflags;
847 
848 	bzero(&sin6, sizeof(sin6));
849 	sin6.sin6_len = sizeof(struct sockaddr_in6);
850 	sin6.sin6_family = AF_INET6;
851 	sin6.sin6_addr = *addr6;
852 
853 	IF_AFDATA_LOCK_ASSERT(ifp);
854 
855 	llflags = 0;
856 	if (flags & ND6_CREATE)
857 	    llflags |= LLE_CREATE;
858 	if (flags & ND6_EXCLUSIVE)
859 	    llflags |= LLE_EXCLUSIVE;
860 
861 	ln = lla_lookup(LLTABLE6(ifp), llflags, (struct sockaddr *)&sin6);
862 	if ((ln != NULL) && (llflags & LLE_CREATE))
863 		ln->ln_state = ND6_LLINFO_NOSTATE;
864 
865 	return (ln);
866 }
867 
868 /*
869  * Test whether a given IPv6 address is a neighbor or not, ignoring
870  * the actual neighbor cache.  The neighbor cache is ignored in order
871  * to not reenter the routing code from within itself.
872  */
873 static int
874 nd6_is_new_addr_neighbor(struct sockaddr_in6 *addr, struct ifnet *ifp)
875 {
876 	struct nd_prefix *pr;
877 	struct ifaddr *dstaddr;
878 
879 	/*
880 	 * A link-local address is always a neighbor.
881 	 * XXX: a link does not necessarily specify a single interface.
882 	 */
883 	if (IN6_IS_ADDR_LINKLOCAL(&addr->sin6_addr)) {
884 		struct sockaddr_in6 sin6_copy;
885 		u_int32_t zone;
886 
887 		/*
888 		 * We need sin6_copy since sa6_recoverscope() may modify the
889 		 * content (XXX).
890 		 */
891 		sin6_copy = *addr;
892 		if (sa6_recoverscope(&sin6_copy))
893 			return (0); /* XXX: should be impossible */
894 		if (in6_setscope(&sin6_copy.sin6_addr, ifp, &zone))
895 			return (0);
896 		if (sin6_copy.sin6_scope_id == zone)
897 			return (1);
898 		else
899 			return (0);
900 	}
901 
902 	/*
903 	 * If the address matches one of our addresses,
904 	 * it should be a neighbor.
905 	 * If the address matches one of our on-link prefixes, it should be a
906 	 * neighbor.
907 	 */
908 	LIST_FOREACH(pr, &V_nd_prefix, ndpr_entry) {
909 		if (pr->ndpr_ifp != ifp)
910 			continue;
911 
912 		if (!(pr->ndpr_stateflags & NDPRF_ONLINK)) {
913 			struct rtentry *rt;
914 
915 			/* Always use the default FIB here. */
916 			rt = in6_rtalloc1((struct sockaddr *)&pr->ndpr_prefix,
917 			    0, 0, RT_DEFAULT_FIB);
918 			if (rt == NULL)
919 				continue;
920 			/*
921 			 * This is the case where multiple interfaces
922 			 * have the same prefix, but only one is installed
923 			 * into the routing table and that prefix entry
924 			 * is not the one being examined here. In the case
925 			 * where RADIX_MPATH is enabled, multiple route
926 			 * entries (of the same rt_key value) will be
927 			 * installed because the interface addresses all
928 			 * differ.
929 			 */
930 			if (!IN6_ARE_ADDR_EQUAL(&pr->ndpr_prefix.sin6_addr,
931 			       &((struct sockaddr_in6 *)rt_key(rt))->sin6_addr)) {
932 				RTFREE_LOCKED(rt);
933 				continue;
934 			}
935 			RTFREE_LOCKED(rt);
936 		}
937 
938 		if (IN6_ARE_MASKED_ADDR_EQUAL(&pr->ndpr_prefix.sin6_addr,
939 		    &addr->sin6_addr, &pr->ndpr_mask))
940 			return (1);
941 	}
942 
943 	/*
944 	 * If the address is assigned on the node of the other side of
945 	 * a p2p interface, the address should be a neighbor.
946 	 */
947 	dstaddr = ifa_ifwithdstaddr((struct sockaddr *)addr);
948 	if (dstaddr != NULL) {
949 		if (dstaddr->ifa_ifp == ifp) {
950 			ifa_free(dstaddr);
951 			return (1);
952 		}
953 		ifa_free(dstaddr);
954 	}
955 
956 	/*
957 	 * If the default router list is empty, all addresses are regarded
958 	 * as on-link, and thus, as a neighbor.
959 	 */
960 	if (ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV &&
961 	    TAILQ_EMPTY(&V_nd_defrouter) &&
962 	    V_nd6_defifindex == ifp->if_index) {
963 		return (1);
964 	}
965 
966 	return (0);
967 }
968 
969 
970 /*
971  * Detect if a given IPv6 address identifies a neighbor on a given link.
972  * XXX: should take care of the destination of a p2p link?
973  */
974 int
975 nd6_is_addr_neighbor(struct sockaddr_in6 *addr, struct ifnet *ifp)
976 {
977 	struct llentry *lle;
978 	int rc = 0;
979 
980 	IF_AFDATA_UNLOCK_ASSERT(ifp);
981 	if (nd6_is_new_addr_neighbor(addr, ifp))
982 		return (1);
983 
984 	/*
985 	 * Even if the address matches none of our addresses, it might be
986 	 * in the neighbor cache.
987 	 */
988 	IF_AFDATA_RLOCK(ifp);
989 	if ((lle = nd6_lookup(&addr->sin6_addr, 0, ifp)) != NULL) {
990 		LLE_RUNLOCK(lle);
991 		rc = 1;
992 	}
993 	IF_AFDATA_RUNLOCK(ifp);
994 	return (rc);
995 }
996 
997 /*
998  * Free an nd6 llinfo entry.
999  * Since the function would cause significant changes in the kernel, DO NOT
1000  * make it global, unless you have a strong reason for the change, and are sure
1001  * that the change is safe.
1002  */
1003 static struct llentry *
1004 nd6_free(struct llentry *ln, int gc)
1005 {
1006         struct llentry *next;
1007 	struct nd_defrouter *dr;
1008 	struct ifnet *ifp;
1009 
1010 	LLE_WLOCK_ASSERT(ln);
1011 
1012 	/*
1013 	 * we used to have pfctlinput(PRC_HOSTDEAD) here.
1014 	 * even though it is not harmful, it was not really necessary.
1015 	 */
1016 
1017 	/* cancel timer */
1018 	nd6_llinfo_settimer_locked(ln, -1);
1019 
1020 	ifp = ln->lle_tbl->llt_ifp;
1021 
1022 	if (ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) {
1023 		dr = defrouter_lookup(&L3_ADDR_SIN6(ln)->sin6_addr, ifp);
1024 
1025 		if (dr != NULL && dr->expire &&
1026 		    ln->ln_state == ND6_LLINFO_STALE && gc) {
1027 			/*
1028 			 * If the reason for the deletion is just garbage
1029 			 * collection, and the neighbor is an active default
1030 			 * router, do not delete it.  Instead, reset the GC
1031 			 * timer using the router's lifetime.
1032 			 * Simply deleting the entry would affect default
1033 			 * router selection, which is not necessarily a good
1034 			 * thing, especially when we're using router preference
1035 			 * values.
1036 			 * XXX: the check for ln_state would be redundant,
1037 			 *      but we intentionally keep it just in case.
1038 			 */
1039 			if (dr->expire > time_uptime)
1040 				nd6_llinfo_settimer_locked(ln,
1041 				    (dr->expire - time_uptime) * hz);
1042 			else
1043 				nd6_llinfo_settimer_locked(ln,
1044 				    (long)V_nd6_gctimer * hz);
1045 
1046 			next = LIST_NEXT(ln, lle_next);
1047 			LLE_REMREF(ln);
1048 			LLE_WUNLOCK(ln);
1049 			return (next);
1050 		}
1051 
1052 		if (dr) {
1053 			/*
1054 			 * Unreachablity of a router might affect the default
1055 			 * router selection and on-link detection of advertised
1056 			 * prefixes.
1057 			 */
1058 
1059 			/*
1060 			 * Temporarily fake the state to choose a new default
1061 			 * router and to perform on-link determination of
1062 			 * prefixes correctly.
1063 			 * Below the state will be set correctly,
1064 			 * or the entry itself will be deleted.
1065 			 */
1066 			ln->ln_state = ND6_LLINFO_INCOMPLETE;
1067 		}
1068 
1069 		if (ln->ln_router || dr) {
1070 
1071 			/*
1072 			 * We need to unlock to avoid a LOR with rt6_flush() with the
1073 			 * rnh and for the calls to pfxlist_onlink_check() and
1074 			 * defrouter_select() in the block further down for calls
1075 			 * into nd6_lookup().  We still hold a ref.
1076 			 */
1077 			LLE_WUNLOCK(ln);
1078 
1079 			/*
1080 			 * rt6_flush must be called whether or not the neighbor
1081 			 * is in the Default Router List.
1082 			 * See a corresponding comment in nd6_na_input().
1083 			 */
1084 			rt6_flush(&L3_ADDR_SIN6(ln)->sin6_addr, ifp);
1085 		}
1086 
1087 		if (dr) {
1088 			/*
1089 			 * Since defrouter_select() does not affect the
1090 			 * on-link determination and MIP6 needs the check
1091 			 * before the default router selection, we perform
1092 			 * the check now.
1093 			 */
1094 			pfxlist_onlink_check();
1095 
1096 			/*
1097 			 * Refresh default router list.
1098 			 */
1099 			defrouter_select();
1100 		}
1101 
1102 		if (ln->ln_router || dr)
1103 			LLE_WLOCK(ln);
1104 	}
1105 
1106 	/*
1107 	 * Before deleting the entry, remember the next entry as the
1108 	 * return value.  We need this because pfxlist_onlink_check() above
1109 	 * might have freed other entries (particularly the old next entry) as
1110 	 * a side effect (XXX).
1111 	 */
1112 	next = LIST_NEXT(ln, lle_next);
1113 
1114 	/*
1115 	 * Save to unlock. We still hold an extra reference and will not
1116 	 * free(9) in llentry_free() if someone else holds one as well.
1117 	 */
1118 	LLE_WUNLOCK(ln);
1119 	IF_AFDATA_LOCK(ifp);
1120 	LLE_WLOCK(ln);
1121 
1122 	/* Guard against race with other llentry_free(). */
1123 	if (ln->la_flags & LLE_LINKED) {
1124 		LLE_REMREF(ln);
1125 		llentry_free(ln);
1126 	} else
1127 		LLE_FREE_LOCKED(ln);
1128 
1129 	IF_AFDATA_UNLOCK(ifp);
1130 
1131 	return (next);
1132 }
1133 
1134 /*
1135  * Upper-layer reachability hint for Neighbor Unreachability Detection.
1136  *
1137  * XXX cost-effective methods?
1138  */
1139 void
1140 nd6_nud_hint(struct rtentry *rt, struct in6_addr *dst6, int force)
1141 {
1142 	struct llentry *ln;
1143 	struct ifnet *ifp;
1144 
1145 	if ((dst6 == NULL) || (rt == NULL))
1146 		return;
1147 
1148 	ifp = rt->rt_ifp;
1149 	IF_AFDATA_LOCK(ifp);
1150 	ln = nd6_lookup(dst6, ND6_EXCLUSIVE, NULL);
1151 	IF_AFDATA_UNLOCK(ifp);
1152 	if (ln == NULL)
1153 		return;
1154 
1155 	if (ln->ln_state < ND6_LLINFO_REACHABLE)
1156 		goto done;
1157 
1158 	/*
1159 	 * if we get upper-layer reachability confirmation many times,
1160 	 * it is possible we have false information.
1161 	 */
1162 	if (!force) {
1163 		ln->ln_byhint++;
1164 		if (ln->ln_byhint > V_nd6_maxnudhint) {
1165 			goto done;
1166 		}
1167 	}
1168 
1169  	ln->ln_state = ND6_LLINFO_REACHABLE;
1170 	if (!ND6_LLINFO_PERMANENT(ln)) {
1171 		nd6_llinfo_settimer_locked(ln,
1172 		    (long)ND_IFINFO(rt->rt_ifp)->reachable * hz);
1173 	}
1174 done:
1175 	LLE_WUNLOCK(ln);
1176 }
1177 
1178 
1179 /*
1180  * Rejuvenate this function for routing operations related
1181  * processing.
1182  */
1183 void
1184 nd6_rtrequest(int req, struct rtentry *rt, struct rt_addrinfo *info)
1185 {
1186 	struct sockaddr_in6 *gateway;
1187 	struct nd_defrouter *dr;
1188 	struct ifnet *ifp;
1189 
1190 	RT_LOCK_ASSERT(rt);
1191 	gateway = (struct sockaddr_in6 *)rt->rt_gateway;
1192 	ifp = rt->rt_ifp;
1193 
1194 	switch (req) {
1195 	case RTM_ADD:
1196 		break;
1197 
1198 	case RTM_DELETE:
1199 		if (!ifp)
1200 			return;
1201 		/*
1202 		 * Only indirect routes are interesting.
1203 		 */
1204 		if ((rt->rt_flags & RTF_GATEWAY) == 0)
1205 			return;
1206 		/*
1207 		 * check for default route
1208 		 */
1209 		if (IN6_ARE_ADDR_EQUAL(&in6addr_any,
1210 				       &SIN6(rt_key(rt))->sin6_addr)) {
1211 
1212 			dr = defrouter_lookup(&gateway->sin6_addr, ifp);
1213 			if (dr != NULL)
1214 				dr->installed = 0;
1215 		}
1216 		break;
1217 	}
1218 }
1219 
1220 
1221 int
1222 nd6_ioctl(u_long cmd, caddr_t data, struct ifnet *ifp)
1223 {
1224 	struct in6_drlist *drl = (struct in6_drlist *)data;
1225 	struct in6_oprlist *oprl = (struct in6_oprlist *)data;
1226 	struct in6_ndireq *ndi = (struct in6_ndireq *)data;
1227 	struct in6_nbrinfo *nbi = (struct in6_nbrinfo *)data;
1228 	struct in6_ndifreq *ndif = (struct in6_ndifreq *)data;
1229 	struct nd_defrouter *dr;
1230 	struct nd_prefix *pr;
1231 	int i = 0, error = 0;
1232 
1233 	if (ifp->if_afdata[AF_INET6] == NULL)
1234 		return (EPFNOSUPPORT);
1235 	switch (cmd) {
1236 	case SIOCGDRLST_IN6:
1237 		/*
1238 		 * obsolete API, use sysctl under net.inet6.icmp6
1239 		 */
1240 		bzero(drl, sizeof(*drl));
1241 		TAILQ_FOREACH(dr, &V_nd_defrouter, dr_entry) {
1242 			if (i >= DRLSTSIZ)
1243 				break;
1244 			drl->defrouter[i].rtaddr = dr->rtaddr;
1245 			in6_clearscope(&drl->defrouter[i].rtaddr);
1246 
1247 			drl->defrouter[i].flags = dr->flags;
1248 			drl->defrouter[i].rtlifetime = dr->rtlifetime;
1249 			drl->defrouter[i].expire = dr->expire +
1250 			    (time_second - time_uptime);
1251 			drl->defrouter[i].if_index = dr->ifp->if_index;
1252 			i++;
1253 		}
1254 		break;
1255 	case SIOCGPRLST_IN6:
1256 		/*
1257 		 * obsolete API, use sysctl under net.inet6.icmp6
1258 		 *
1259 		 * XXX the structure in6_prlist was changed in backward-
1260 		 * incompatible manner.  in6_oprlist is used for SIOCGPRLST_IN6,
1261 		 * in6_prlist is used for nd6_sysctl() - fill_prlist().
1262 		 */
1263 		/*
1264 		 * XXX meaning of fields, especialy "raflags", is very
1265 		 * differnet between RA prefix list and RR/static prefix list.
1266 		 * how about separating ioctls into two?
1267 		 */
1268 		bzero(oprl, sizeof(*oprl));
1269 		LIST_FOREACH(pr, &V_nd_prefix, ndpr_entry) {
1270 			struct nd_pfxrouter *pfr;
1271 			int j;
1272 
1273 			if (i >= PRLSTSIZ)
1274 				break;
1275 			oprl->prefix[i].prefix = pr->ndpr_prefix.sin6_addr;
1276 			oprl->prefix[i].raflags = pr->ndpr_raf;
1277 			oprl->prefix[i].prefixlen = pr->ndpr_plen;
1278 			oprl->prefix[i].vltime = pr->ndpr_vltime;
1279 			oprl->prefix[i].pltime = pr->ndpr_pltime;
1280 			oprl->prefix[i].if_index = pr->ndpr_ifp->if_index;
1281 			if (pr->ndpr_vltime == ND6_INFINITE_LIFETIME)
1282 				oprl->prefix[i].expire = 0;
1283 			else {
1284 				time_t maxexpire;
1285 
1286 				/* XXX: we assume time_t is signed. */
1287 				maxexpire = (-1) &
1288 				    ~((time_t)1 <<
1289 				    ((sizeof(maxexpire) * 8) - 1));
1290 				if (pr->ndpr_vltime <
1291 				    maxexpire - pr->ndpr_lastupdate) {
1292 					oprl->prefix[i].expire =
1293 					    pr->ndpr_lastupdate +
1294 					    pr->ndpr_vltime +
1295 					    (time_second - time_uptime);
1296 				} else
1297 					oprl->prefix[i].expire = maxexpire;
1298 			}
1299 
1300 			j = 0;
1301 			LIST_FOREACH(pfr, &pr->ndpr_advrtrs, pfr_entry) {
1302 				if (j < DRLSTSIZ) {
1303 #define RTRADDR oprl->prefix[i].advrtr[j]
1304 					RTRADDR = pfr->router->rtaddr;
1305 					in6_clearscope(&RTRADDR);
1306 #undef RTRADDR
1307 				}
1308 				j++;
1309 			}
1310 			oprl->prefix[i].advrtrs = j;
1311 			oprl->prefix[i].origin = PR_ORIG_RA;
1312 
1313 			i++;
1314 		}
1315 
1316 		break;
1317 	case OSIOCGIFINFO_IN6:
1318 #define ND	ndi->ndi
1319 		/* XXX: old ndp(8) assumes a positive value for linkmtu. */
1320 		bzero(&ND, sizeof(ND));
1321 		ND.linkmtu = IN6_LINKMTU(ifp);
1322 		ND.maxmtu = ND_IFINFO(ifp)->maxmtu;
1323 		ND.basereachable = ND_IFINFO(ifp)->basereachable;
1324 		ND.reachable = ND_IFINFO(ifp)->reachable;
1325 		ND.retrans = ND_IFINFO(ifp)->retrans;
1326 		ND.flags = ND_IFINFO(ifp)->flags;
1327 		ND.recalctm = ND_IFINFO(ifp)->recalctm;
1328 		ND.chlim = ND_IFINFO(ifp)->chlim;
1329 		break;
1330 	case SIOCGIFINFO_IN6:
1331 		ND = *ND_IFINFO(ifp);
1332 		break;
1333 	case SIOCSIFINFO_IN6:
1334 		/*
1335 		 * used to change host variables from userland.
1336 		 * intented for a use on router to reflect RA configurations.
1337 		 */
1338 		/* 0 means 'unspecified' */
1339 		if (ND.linkmtu != 0) {
1340 			if (ND.linkmtu < IPV6_MMTU ||
1341 			    ND.linkmtu > IN6_LINKMTU(ifp)) {
1342 				error = EINVAL;
1343 				break;
1344 			}
1345 			ND_IFINFO(ifp)->linkmtu = ND.linkmtu;
1346 		}
1347 
1348 		if (ND.basereachable != 0) {
1349 			int obasereachable = ND_IFINFO(ifp)->basereachable;
1350 
1351 			ND_IFINFO(ifp)->basereachable = ND.basereachable;
1352 			if (ND.basereachable != obasereachable)
1353 				ND_IFINFO(ifp)->reachable =
1354 				    ND_COMPUTE_RTIME(ND.basereachable);
1355 		}
1356 		if (ND.retrans != 0)
1357 			ND_IFINFO(ifp)->retrans = ND.retrans;
1358 		if (ND.chlim != 0)
1359 			ND_IFINFO(ifp)->chlim = ND.chlim;
1360 		/* FALLTHROUGH */
1361 	case SIOCSIFINFO_FLAGS:
1362 	{
1363 		struct ifaddr *ifa;
1364 		struct in6_ifaddr *ia;
1365 
1366 		if ((ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED) &&
1367 		    !(ND.flags & ND6_IFF_IFDISABLED)) {
1368 			/* ifdisabled 1->0 transision */
1369 
1370 			/*
1371 			 * If the interface is marked as ND6_IFF_IFDISABLED and
1372 			 * has an link-local address with IN6_IFF_DUPLICATED,
1373 			 * do not clear ND6_IFF_IFDISABLED.
1374 			 * See RFC 4862, Section 5.4.5.
1375 			 */
1376 			int duplicated_linklocal = 0;
1377 
1378 			IF_ADDR_RLOCK(ifp);
1379 			TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1380 				if (ifa->ifa_addr->sa_family != AF_INET6)
1381 					continue;
1382 				ia = (struct in6_ifaddr *)ifa;
1383 				if ((ia->ia6_flags & IN6_IFF_DUPLICATED) &&
1384 				    IN6_IS_ADDR_LINKLOCAL(IA6_IN6(ia))) {
1385 					duplicated_linklocal = 1;
1386 					break;
1387 				}
1388 			}
1389 			IF_ADDR_RUNLOCK(ifp);
1390 
1391 			if (duplicated_linklocal) {
1392 				ND.flags |= ND6_IFF_IFDISABLED;
1393 				log(LOG_ERR, "Cannot enable an interface"
1394 				    " with a link-local address marked"
1395 				    " duplicate.\n");
1396 			} else {
1397 				ND_IFINFO(ifp)->flags &= ~ND6_IFF_IFDISABLED;
1398 				if (ifp->if_flags & IFF_UP)
1399 					in6_if_up(ifp);
1400 			}
1401 		} else if (!(ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED) &&
1402 			    (ND.flags & ND6_IFF_IFDISABLED)) {
1403 			/* ifdisabled 0->1 transision */
1404 			/* Mark all IPv6 address as tentative. */
1405 
1406 			ND_IFINFO(ifp)->flags |= ND6_IFF_IFDISABLED;
1407 			IF_ADDR_RLOCK(ifp);
1408 			TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1409 				if (ifa->ifa_addr->sa_family != AF_INET6)
1410 					continue;
1411 				ia = (struct in6_ifaddr *)ifa;
1412 				ia->ia6_flags |= IN6_IFF_TENTATIVE;
1413 			}
1414 			IF_ADDR_RUNLOCK(ifp);
1415 		}
1416 
1417 		if (ND.flags & ND6_IFF_AUTO_LINKLOCAL) {
1418 			if (!(ND_IFINFO(ifp)->flags & ND6_IFF_AUTO_LINKLOCAL)) {
1419 				/* auto_linklocal 0->1 transision */
1420 
1421 				/* If no link-local address on ifp, configure */
1422 				ND_IFINFO(ifp)->flags |= ND6_IFF_AUTO_LINKLOCAL;
1423 				in6_ifattach(ifp, NULL);
1424 			} else if (!(ND.flags & ND6_IFF_IFDISABLED) &&
1425 			    ifp->if_flags & IFF_UP) {
1426 				/*
1427 				 * When the IF already has
1428 				 * ND6_IFF_AUTO_LINKLOCAL, no link-local
1429 				 * address is assigned, and IFF_UP, try to
1430 				 * assign one.
1431 				 */
1432 				int haslinklocal = 0;
1433 
1434 				IF_ADDR_RLOCK(ifp);
1435 				TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1436 					if (ifa->ifa_addr->sa_family != AF_INET6)
1437 						continue;
1438 					ia = (struct in6_ifaddr *)ifa;
1439 					if (IN6_IS_ADDR_LINKLOCAL(IA6_IN6(ia))) {
1440 						haslinklocal = 1;
1441 						break;
1442 					}
1443 				}
1444 				IF_ADDR_RUNLOCK(ifp);
1445 				if (!haslinklocal)
1446 					in6_ifattach(ifp, NULL);
1447 			}
1448 		}
1449 	}
1450 		ND_IFINFO(ifp)->flags = ND.flags;
1451 		break;
1452 #undef ND
1453 	case SIOCSNDFLUSH_IN6:	/* XXX: the ioctl name is confusing... */
1454 		/* sync kernel routing table with the default router list */
1455 		defrouter_reset();
1456 		defrouter_select();
1457 		break;
1458 	case SIOCSPFXFLUSH_IN6:
1459 	{
1460 		/* flush all the prefix advertised by routers */
1461 		struct nd_prefix *pr, *next;
1462 
1463 		LIST_FOREACH_SAFE(pr, &V_nd_prefix, ndpr_entry, next) {
1464 			struct in6_ifaddr *ia, *ia_next;
1465 
1466 			if (IN6_IS_ADDR_LINKLOCAL(&pr->ndpr_prefix.sin6_addr))
1467 				continue; /* XXX */
1468 
1469 			/* do we really have to remove addresses as well? */
1470 			/* XXXRW: in6_ifaddrhead locking. */
1471 			TAILQ_FOREACH_SAFE(ia, &V_in6_ifaddrhead, ia_link,
1472 			    ia_next) {
1473 				if ((ia->ia6_flags & IN6_IFF_AUTOCONF) == 0)
1474 					continue;
1475 
1476 				if (ia->ia6_ndpr == pr)
1477 					in6_purgeaddr(&ia->ia_ifa);
1478 			}
1479 			prelist_remove(pr);
1480 		}
1481 		break;
1482 	}
1483 	case SIOCSRTRFLUSH_IN6:
1484 	{
1485 		/* flush all the default routers */
1486 		struct nd_defrouter *dr, *next;
1487 
1488 		defrouter_reset();
1489 		TAILQ_FOREACH_SAFE(dr, &V_nd_defrouter, dr_entry, next) {
1490 			defrtrlist_del(dr);
1491 		}
1492 		defrouter_select();
1493 		break;
1494 	}
1495 	case SIOCGNBRINFO_IN6:
1496 	{
1497 		struct llentry *ln;
1498 		struct in6_addr nb_addr = nbi->addr; /* make local for safety */
1499 
1500 		if ((error = in6_setscope(&nb_addr, ifp, NULL)) != 0)
1501 			return (error);
1502 
1503 		IF_AFDATA_RLOCK(ifp);
1504 		ln = nd6_lookup(&nb_addr, 0, ifp);
1505 		IF_AFDATA_RUNLOCK(ifp);
1506 
1507 		if (ln == NULL) {
1508 			error = EINVAL;
1509 			break;
1510 		}
1511 		nbi->state = ln->ln_state;
1512 		nbi->asked = ln->la_asked;
1513 		nbi->isrouter = ln->ln_router;
1514 		if (ln->la_expire == 0)
1515 			nbi->expire = 0;
1516 		else
1517 			nbi->expire = ln->la_expire +
1518 			    (time_second - time_uptime);
1519 		LLE_RUNLOCK(ln);
1520 		break;
1521 	}
1522 	case SIOCGDEFIFACE_IN6:	/* XXX: should be implemented as a sysctl? */
1523 		ndif->ifindex = V_nd6_defifindex;
1524 		break;
1525 	case SIOCSDEFIFACE_IN6:	/* XXX: should be implemented as a sysctl? */
1526 		return (nd6_setdefaultiface(ndif->ifindex));
1527 	}
1528 	return (error);
1529 }
1530 
1531 /*
1532  * Create neighbor cache entry and cache link-layer address,
1533  * on reception of inbound ND6 packets.  (RS/RA/NS/redirect)
1534  *
1535  * type - ICMP6 type
1536  * code - type dependent information
1537  *
1538  * XXXXX
1539  *  The caller of this function already acquired the ndp
1540  *  cache table lock because the cache entry is returned.
1541  */
1542 struct llentry *
1543 nd6_cache_lladdr(struct ifnet *ifp, struct in6_addr *from, char *lladdr,
1544     int lladdrlen, int type, int code)
1545 {
1546 	struct llentry *ln = NULL;
1547 	int is_newentry;
1548 	int do_update;
1549 	int olladdr;
1550 	int llchange;
1551 	int flags;
1552 	int newstate = 0;
1553 	uint16_t router = 0;
1554 	struct sockaddr_in6 sin6;
1555 	struct mbuf *chain = NULL;
1556 	int static_route = 0;
1557 
1558 	IF_AFDATA_UNLOCK_ASSERT(ifp);
1559 
1560 	KASSERT(ifp != NULL, ("%s: ifp == NULL", __func__));
1561 	KASSERT(from != NULL, ("%s: from == NULL", __func__));
1562 
1563 	/* nothing must be updated for unspecified address */
1564 	if (IN6_IS_ADDR_UNSPECIFIED(from))
1565 		return NULL;
1566 
1567 	/*
1568 	 * Validation about ifp->if_addrlen and lladdrlen must be done in
1569 	 * the caller.
1570 	 *
1571 	 * XXX If the link does not have link-layer adderss, what should
1572 	 * we do? (ifp->if_addrlen == 0)
1573 	 * Spec says nothing in sections for RA, RS and NA.  There's small
1574 	 * description on it in NS section (RFC 2461 7.2.3).
1575 	 */
1576 	flags = lladdr ? ND6_EXCLUSIVE : 0;
1577 	IF_AFDATA_LOCK(ifp);
1578 	ln = nd6_lookup(from, flags, ifp);
1579 
1580 	if (ln == NULL) {
1581 		flags |= ND6_EXCLUSIVE;
1582 		ln = nd6_lookup(from, flags | ND6_CREATE, ifp);
1583 		IF_AFDATA_UNLOCK(ifp);
1584 		is_newentry = 1;
1585 	} else {
1586 		IF_AFDATA_UNLOCK(ifp);
1587 		/* do nothing if static ndp is set */
1588 		if (ln->la_flags & LLE_STATIC) {
1589 			static_route = 1;
1590 			goto done;
1591 		}
1592 		is_newentry = 0;
1593 	}
1594 	if (ln == NULL)
1595 		return (NULL);
1596 
1597 	olladdr = (ln->la_flags & LLE_VALID) ? 1 : 0;
1598 	if (olladdr && lladdr) {
1599 		llchange = bcmp(lladdr, &ln->ll_addr,
1600 		    ifp->if_addrlen);
1601 	} else
1602 		llchange = 0;
1603 
1604 	/*
1605 	 * newentry olladdr  lladdr  llchange	(*=record)
1606 	 *	0	n	n	--	(1)
1607 	 *	0	y	n	--	(2)
1608 	 *	0	n	y	--	(3) * STALE
1609 	 *	0	y	y	n	(4) *
1610 	 *	0	y	y	y	(5) * STALE
1611 	 *	1	--	n	--	(6)   NOSTATE(= PASSIVE)
1612 	 *	1	--	y	--	(7) * STALE
1613 	 */
1614 
1615 	if (lladdr) {		/* (3-5) and (7) */
1616 		/*
1617 		 * Record source link-layer address
1618 		 * XXX is it dependent to ifp->if_type?
1619 		 */
1620 		bcopy(lladdr, &ln->ll_addr, ifp->if_addrlen);
1621 		ln->la_flags |= LLE_VALID;
1622 		EVENTHANDLER_INVOKE(lle_event, ln, LLENTRY_RESOLVED);
1623 	}
1624 
1625 	if (!is_newentry) {
1626 		if ((!olladdr && lladdr != NULL) ||	/* (3) */
1627 		    (olladdr && lladdr != NULL && llchange)) {	/* (5) */
1628 			do_update = 1;
1629 			newstate = ND6_LLINFO_STALE;
1630 		} else					/* (1-2,4) */
1631 			do_update = 0;
1632 	} else {
1633 		do_update = 1;
1634 		if (lladdr == NULL)			/* (6) */
1635 			newstate = ND6_LLINFO_NOSTATE;
1636 		else					/* (7) */
1637 			newstate = ND6_LLINFO_STALE;
1638 	}
1639 
1640 	if (do_update) {
1641 		/*
1642 		 * Update the state of the neighbor cache.
1643 		 */
1644 		ln->ln_state = newstate;
1645 
1646 		if (ln->ln_state == ND6_LLINFO_STALE) {
1647 			/*
1648 			 * XXX: since nd6_output() below will cause
1649 			 * state tansition to DELAY and reset the timer,
1650 			 * we must set the timer now, although it is actually
1651 			 * meaningless.
1652 			 */
1653 			nd6_llinfo_settimer_locked(ln, (long)V_nd6_gctimer * hz);
1654 
1655 			if (ln->la_hold) {
1656 				struct mbuf *m_hold, *m_hold_next;
1657 
1658 				/*
1659 				 * reset the la_hold in advance, to explicitly
1660 				 * prevent a la_hold lookup in nd6_output()
1661 				 * (wouldn't happen, though...)
1662 				 */
1663 				for (m_hold = ln->la_hold, ln->la_hold = NULL;
1664 				    m_hold; m_hold = m_hold_next) {
1665 					m_hold_next = m_hold->m_nextpkt;
1666 					m_hold->m_nextpkt = NULL;
1667 
1668 					/*
1669 					 * we assume ifp is not a p2p here, so
1670 					 * just set the 2nd argument as the
1671 					 * 1st one.
1672 					 */
1673 					nd6_output_lle(ifp, ifp, m_hold, L3_ADDR_SIN6(ln), NULL, ln, &chain);
1674 				}
1675 				/*
1676 				 * If we have mbufs in the chain we need to do
1677 				 * deferred transmit. Copy the address from the
1678 				 * llentry before dropping the lock down below.
1679 				 */
1680 				if (chain != NULL)
1681 					memcpy(&sin6, L3_ADDR_SIN6(ln), sizeof(sin6));
1682 			}
1683 		} else if (ln->ln_state == ND6_LLINFO_INCOMPLETE) {
1684 			/* probe right away */
1685 			nd6_llinfo_settimer_locked((void *)ln, 0);
1686 		}
1687 	}
1688 
1689 	/*
1690 	 * ICMP6 type dependent behavior.
1691 	 *
1692 	 * NS: clear IsRouter if new entry
1693 	 * RS: clear IsRouter
1694 	 * RA: set IsRouter if there's lladdr
1695 	 * redir: clear IsRouter if new entry
1696 	 *
1697 	 * RA case, (1):
1698 	 * The spec says that we must set IsRouter in the following cases:
1699 	 * - If lladdr exist, set IsRouter.  This means (1-5).
1700 	 * - If it is old entry (!newentry), set IsRouter.  This means (7).
1701 	 * So, based on the spec, in (1-5) and (7) cases we must set IsRouter.
1702 	 * A quetion arises for (1) case.  (1) case has no lladdr in the
1703 	 * neighbor cache, this is similar to (6).
1704 	 * This case is rare but we figured that we MUST NOT set IsRouter.
1705 	 *
1706 	 * newentry olladdr  lladdr  llchange	    NS  RS  RA	redir
1707 	 *							D R
1708 	 *	0	n	n	--	(1)	c   ?     s
1709 	 *	0	y	n	--	(2)	c   s     s
1710 	 *	0	n	y	--	(3)	c   s     s
1711 	 *	0	y	y	n	(4)	c   s     s
1712 	 *	0	y	y	y	(5)	c   s     s
1713 	 *	1	--	n	--	(6) c	c	c s
1714 	 *	1	--	y	--	(7) c	c   s	c s
1715 	 *
1716 	 *					(c=clear s=set)
1717 	 */
1718 	switch (type & 0xff) {
1719 	case ND_NEIGHBOR_SOLICIT:
1720 		/*
1721 		 * New entry must have is_router flag cleared.
1722 		 */
1723 		if (is_newentry)	/* (6-7) */
1724 			ln->ln_router = 0;
1725 		break;
1726 	case ND_REDIRECT:
1727 		/*
1728 		 * If the icmp is a redirect to a better router, always set the
1729 		 * is_router flag.  Otherwise, if the entry is newly created,
1730 		 * clear the flag.  [RFC 2461, sec 8.3]
1731 		 */
1732 		if (code == ND_REDIRECT_ROUTER)
1733 			ln->ln_router = 1;
1734 		else if (is_newentry) /* (6-7) */
1735 			ln->ln_router = 0;
1736 		break;
1737 	case ND_ROUTER_SOLICIT:
1738 		/*
1739 		 * is_router flag must always be cleared.
1740 		 */
1741 		ln->ln_router = 0;
1742 		break;
1743 	case ND_ROUTER_ADVERT:
1744 		/*
1745 		 * Mark an entry with lladdr as a router.
1746 		 */
1747 		if ((!is_newentry && (olladdr || lladdr)) ||	/* (2-5) */
1748 		    (is_newentry && lladdr)) {			/* (7) */
1749 			ln->ln_router = 1;
1750 		}
1751 		break;
1752 	}
1753 
1754 	if (ln != NULL) {
1755 		static_route = (ln->la_flags & LLE_STATIC);
1756 		router = ln->ln_router;
1757 
1758 		if (flags & ND6_EXCLUSIVE)
1759 			LLE_WUNLOCK(ln);
1760 		else
1761 			LLE_RUNLOCK(ln);
1762 		if (static_route)
1763 			ln = NULL;
1764 	}
1765 	if (chain)
1766 		nd6_output_flush(ifp, ifp, chain, &sin6, NULL);
1767 
1768 	/*
1769 	 * When the link-layer address of a router changes, select the
1770 	 * best router again.  In particular, when the neighbor entry is newly
1771 	 * created, it might affect the selection policy.
1772 	 * Question: can we restrict the first condition to the "is_newentry"
1773 	 * case?
1774 	 * XXX: when we hear an RA from a new router with the link-layer
1775 	 * address option, defrouter_select() is called twice, since
1776 	 * defrtrlist_update called the function as well.  However, I believe
1777 	 * we can compromise the overhead, since it only happens the first
1778 	 * time.
1779 	 * XXX: although defrouter_select() should not have a bad effect
1780 	 * for those are not autoconfigured hosts, we explicitly avoid such
1781 	 * cases for safety.
1782 	 */
1783 	if (do_update && router &&
1784 	    ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) {
1785 		/*
1786 		 * guaranteed recursion
1787 		 */
1788 		defrouter_select();
1789 	}
1790 
1791 	return (ln);
1792 done:
1793 	if (ln != NULL) {
1794 		if (flags & ND6_EXCLUSIVE)
1795 			LLE_WUNLOCK(ln);
1796 		else
1797 			LLE_RUNLOCK(ln);
1798 		if (static_route)
1799 			ln = NULL;
1800 	}
1801 	return (ln);
1802 }
1803 
1804 static void
1805 nd6_slowtimo(void *arg)
1806 {
1807 	CURVNET_SET((struct vnet *) arg);
1808 	struct nd_ifinfo *nd6if;
1809 	struct ifnet *ifp;
1810 
1811 	callout_reset(&V_nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
1812 	    nd6_slowtimo, curvnet);
1813 	IFNET_RLOCK_NOSLEEP();
1814 	TAILQ_FOREACH(ifp, &V_ifnet, if_list) {
1815 		if (ifp->if_afdata[AF_INET6] == NULL)
1816 			continue;
1817 		nd6if = ND_IFINFO(ifp);
1818 		if (nd6if->basereachable && /* already initialized */
1819 		    (nd6if->recalctm -= ND6_SLOWTIMER_INTERVAL) <= 0) {
1820 			/*
1821 			 * Since reachable time rarely changes by router
1822 			 * advertisements, we SHOULD insure that a new random
1823 			 * value gets recomputed at least once every few hours.
1824 			 * (RFC 2461, 6.3.4)
1825 			 */
1826 			nd6if->recalctm = V_nd6_recalc_reachtm_interval;
1827 			nd6if->reachable = ND_COMPUTE_RTIME(nd6if->basereachable);
1828 		}
1829 	}
1830 	IFNET_RUNLOCK_NOSLEEP();
1831 	CURVNET_RESTORE();
1832 }
1833 
1834 int
1835 nd6_output(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *m0,
1836     struct sockaddr_in6 *dst, struct rtentry *rt0)
1837 {
1838 
1839 	return (nd6_output_lle(ifp, origifp, m0, dst, rt0, NULL, NULL));
1840 }
1841 
1842 
1843 /*
1844  * Note that I'm not enforcing any global serialization
1845  * lle state or asked changes here as the logic is too
1846  * complicated to avoid having to always acquire an exclusive
1847  * lock
1848  * KMM
1849  *
1850  */
1851 #define senderr(e) { error = (e); goto bad;}
1852 
1853 int
1854 nd6_output_lle(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *m0,
1855     struct sockaddr_in6 *dst, struct rtentry *rt0, struct llentry *lle,
1856 	struct mbuf **chain)
1857 {
1858 	struct mbuf *m = m0;
1859 	struct m_tag *mtag;
1860 	struct llentry *ln = lle;
1861 	struct ip6_hdr *ip6;
1862 	int error = 0;
1863 	int flags = 0;
1864 	int ip6len;
1865 
1866 #ifdef INVARIANTS
1867 	if (lle != NULL) {
1868 
1869 		LLE_WLOCK_ASSERT(lle);
1870 
1871 		KASSERT(chain != NULL, (" lle locked but no mbuf chain pointer passed"));
1872 	}
1873 #endif
1874 	if (IN6_IS_ADDR_MULTICAST(&dst->sin6_addr))
1875 		goto sendpkt;
1876 
1877 	if (nd6_need_cache(ifp) == 0)
1878 		goto sendpkt;
1879 
1880 	/*
1881 	 * next hop determination.  This routine is derived from ether_output.
1882 	 */
1883 
1884 	/*
1885 	 * Address resolution or Neighbor Unreachability Detection
1886 	 * for the next hop.
1887 	 * At this point, the destination of the packet must be a unicast
1888 	 * or an anycast address(i.e. not a multicast).
1889 	 */
1890 
1891 	flags = ((m != NULL) || (lle != NULL)) ? LLE_EXCLUSIVE : 0;
1892 	if (ln == NULL) {
1893 	retry:
1894 		IF_AFDATA_LOCK(ifp);
1895 		ln = lla_lookup(LLTABLE6(ifp), flags, (struct sockaddr *)dst);
1896 		IF_AFDATA_UNLOCK(ifp);
1897 		if ((ln == NULL) && nd6_is_addr_neighbor(dst, ifp))  {
1898 			/*
1899 			 * Since nd6_is_addr_neighbor() internally calls nd6_lookup(),
1900 			 * the condition below is not very efficient.  But we believe
1901 			 * it is tolerable, because this should be a rare case.
1902 			 */
1903 			flags = ND6_CREATE | (m ? ND6_EXCLUSIVE : 0);
1904 			IF_AFDATA_LOCK(ifp);
1905 			ln = nd6_lookup(&dst->sin6_addr, flags, ifp);
1906 			IF_AFDATA_UNLOCK(ifp);
1907 		}
1908 	}
1909 	if (ln == NULL) {
1910 		if ((ifp->if_flags & IFF_POINTOPOINT) == 0 &&
1911 		    !(ND_IFINFO(ifp)->flags & ND6_IFF_PERFORMNUD)) {
1912 			char ip6buf[INET6_ADDRSTRLEN];
1913 			log(LOG_DEBUG,
1914 			    "nd6_output: can't allocate llinfo for %s "
1915 			    "(ln=%p)\n",
1916 			    ip6_sprintf(ip6buf, &dst->sin6_addr), ln);
1917 			senderr(EIO);	/* XXX: good error? */
1918 		}
1919 		goto sendpkt;	/* send anyway */
1920 	}
1921 
1922 	/* We don't have to do link-layer address resolution on a p2p link. */
1923 	if ((ifp->if_flags & IFF_POINTOPOINT) != 0 &&
1924 	    ln->ln_state < ND6_LLINFO_REACHABLE) {
1925 		if ((flags & LLE_EXCLUSIVE) == 0) {
1926 			flags |= LLE_EXCLUSIVE;
1927 			goto retry;
1928 		}
1929 		ln->ln_state = ND6_LLINFO_STALE;
1930 		nd6_llinfo_settimer_locked(ln, (long)V_nd6_gctimer * hz);
1931 	}
1932 
1933 	/*
1934 	 * The first time we send a packet to a neighbor whose entry is
1935 	 * STALE, we have to change the state to DELAY and a sets a timer to
1936 	 * expire in DELAY_FIRST_PROBE_TIME seconds to ensure do
1937 	 * neighbor unreachability detection on expiration.
1938 	 * (RFC 2461 7.3.3)
1939 	 */
1940 	if (ln->ln_state == ND6_LLINFO_STALE) {
1941 		if ((flags & LLE_EXCLUSIVE) == 0) {
1942 			flags |= LLE_EXCLUSIVE;
1943 			LLE_RUNLOCK(ln);
1944 			goto retry;
1945 		}
1946 		ln->la_asked = 0;
1947 		ln->ln_state = ND6_LLINFO_DELAY;
1948 		nd6_llinfo_settimer_locked(ln, (long)V_nd6_delay * hz);
1949 	}
1950 
1951 	/*
1952 	 * If the neighbor cache entry has a state other than INCOMPLETE
1953 	 * (i.e. its link-layer address is already resolved), just
1954 	 * send the packet.
1955 	 */
1956 	if (ln->ln_state > ND6_LLINFO_INCOMPLETE)
1957 		goto sendpkt;
1958 
1959 	/*
1960 	 * There is a neighbor cache entry, but no ethernet address
1961 	 * response yet.  Append this latest packet to the end of the
1962 	 * packet queue in the mbuf, unless the number of the packet
1963 	 * does not exceed nd6_maxqueuelen.  When it exceeds nd6_maxqueuelen,
1964 	 * the oldest packet in the queue will be removed.
1965 	 */
1966 	if (ln->ln_state == ND6_LLINFO_NOSTATE)
1967 		ln->ln_state = ND6_LLINFO_INCOMPLETE;
1968 
1969 	if ((flags & LLE_EXCLUSIVE) == 0) {
1970 		flags |= LLE_EXCLUSIVE;
1971 		LLE_RUNLOCK(ln);
1972 		goto retry;
1973 	}
1974 
1975 	LLE_WLOCK_ASSERT(ln);
1976 
1977 	if (ln->la_hold) {
1978 		struct mbuf *m_hold;
1979 		int i;
1980 
1981 		i = 0;
1982 		for (m_hold = ln->la_hold; m_hold; m_hold = m_hold->m_nextpkt) {
1983 			i++;
1984 			if (m_hold->m_nextpkt == NULL) {
1985 				m_hold->m_nextpkt = m;
1986 				break;
1987 			}
1988 		}
1989 		while (i >= V_nd6_maxqueuelen) {
1990 			m_hold = ln->la_hold;
1991 			ln->la_hold = ln->la_hold->m_nextpkt;
1992 			m_freem(m_hold);
1993 			i--;
1994 		}
1995 	} else {
1996 		ln->la_hold = m;
1997 	}
1998 
1999 	/*
2000 	 * If there has been no NS for the neighbor after entering the
2001 	 * INCOMPLETE state, send the first solicitation.
2002 	 */
2003 	if (!ND6_LLINFO_PERMANENT(ln) && ln->la_asked == 0) {
2004 		ln->la_asked++;
2005 
2006 		nd6_llinfo_settimer_locked(ln,
2007 		    (long)ND_IFINFO(ifp)->retrans * hz / 1000);
2008 		LLE_WUNLOCK(ln);
2009 		nd6_ns_output(ifp, NULL, &dst->sin6_addr, ln, 0);
2010 		if (lle != NULL && ln == lle)
2011 			LLE_WLOCK(lle);
2012 
2013 	} else if (lle == NULL || ln != lle) {
2014 		/*
2015 		 * We did the lookup (no lle arg) so we
2016 		 * need to do the unlock here.
2017 		 */
2018 		LLE_WUNLOCK(ln);
2019 	}
2020 
2021 	return (0);
2022 
2023   sendpkt:
2024 	/* discard the packet if IPv6 operation is disabled on the interface */
2025 	if ((ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED)) {
2026 		error = ENETDOWN; /* better error? */
2027 		goto bad;
2028 	}
2029 	/*
2030 	 * ln is valid and the caller did not pass in
2031 	 * an llentry
2032 	 */
2033 	if ((ln != NULL) && (lle == NULL)) {
2034 		if (flags & LLE_EXCLUSIVE)
2035 			LLE_WUNLOCK(ln);
2036 		else
2037 			LLE_RUNLOCK(ln);
2038 	}
2039 
2040 #ifdef MAC
2041 	mac_netinet6_nd6_send(ifp, m);
2042 #endif
2043 
2044 	/*
2045 	 * If called from nd6_ns_output() (NS), nd6_na_output() (NA),
2046 	 * icmp6_redirect_output() (REDIRECT) or from rip6_output() (RS, RA
2047 	 * as handled by rtsol and rtadvd), mbufs will be tagged for SeND
2048 	 * to be diverted to user space.  When re-injected into the kernel,
2049 	 * send_output() will directly dispatch them to the outgoing interface.
2050 	 */
2051 	if (send_sendso_input_hook != NULL) {
2052 		mtag = m_tag_find(m, PACKET_TAG_ND_OUTGOING, NULL);
2053 		if (mtag != NULL) {
2054 			ip6 = mtod(m, struct ip6_hdr *);
2055 			ip6len = sizeof(struct ip6_hdr) + ntohs(ip6->ip6_plen);
2056 			/* Use the SEND socket */
2057 			error = send_sendso_input_hook(m, ifp, SND_OUT,
2058 			    ip6len);
2059 			/* -1 == no app on SEND socket */
2060 			if (error == 0 || error != -1)
2061 			    return (error);
2062 		}
2063 	}
2064 
2065 	/*
2066 	 * We were passed in a pointer to an lle with the lock held
2067 	 * this means that we can't call if_output as we will
2068 	 * recurse on the lle lock - so what we do is we create
2069 	 * a list of mbufs to send and transmit them in the caller
2070 	 * after the lock is dropped
2071 	 */
2072 	if (lle != NULL) {
2073 		if (*chain == NULL)
2074 			*chain = m;
2075 		else {
2076 			struct mbuf *mb;
2077 
2078 			/*
2079 			 * append mbuf to end of deferred chain
2080 			 */
2081 			mb = *chain;
2082 			while (mb->m_nextpkt != NULL)
2083 				mb = mb->m_nextpkt;
2084 			mb->m_nextpkt = m;
2085 		}
2086 		return (error);
2087 	}
2088 	m_clrprotoflags(m);	/* Avoid confusing lower layers. */
2089 	IP_PROBE(send, NULL, NULL, mtod(m, struct ip6_hdr *), ifp, NULL,
2090 	    mtod(m, struct ip6_hdr *));
2091 	if ((ifp->if_flags & IFF_LOOPBACK) != 0) {
2092 		return ((*ifp->if_output)(origifp, m, (struct sockaddr *)dst,
2093 		    NULL));
2094 	}
2095 	error = (*ifp->if_output)(ifp, m, (struct sockaddr *)dst, NULL);
2096 	return (error);
2097 
2098   bad:
2099 	/*
2100 	 * ln is valid and the caller did not pass in
2101 	 * an llentry
2102 	 */
2103 	if ((ln != NULL) && (lle == NULL)) {
2104 		if (flags & LLE_EXCLUSIVE)
2105 			LLE_WUNLOCK(ln);
2106 		else
2107 			LLE_RUNLOCK(ln);
2108 	}
2109 	if (m)
2110 		m_freem(m);
2111 	return (error);
2112 }
2113 #undef senderr
2114 
2115 
2116 int
2117 nd6_output_flush(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *chain,
2118     struct sockaddr_in6 *dst, struct route *ro)
2119 {
2120 	struct mbuf *m, *m_head;
2121 	struct ifnet *outifp;
2122 	int error = 0;
2123 
2124 	m_head = chain;
2125 	if ((ifp->if_flags & IFF_LOOPBACK) != 0)
2126 		outifp = origifp;
2127 	else
2128 		outifp = ifp;
2129 
2130 	while (m_head) {
2131 		m = m_head;
2132 		m_head = m_head->m_nextpkt;
2133 		error = (*ifp->if_output)(ifp, m, (struct sockaddr *)dst, ro);
2134 	}
2135 
2136 	/*
2137 	 * XXX
2138 	 * note that intermediate errors are blindly ignored - but this is
2139 	 * the same convention as used with nd6_output when called by
2140 	 * nd6_cache_lladdr
2141 	 */
2142 	return (error);
2143 }
2144 
2145 
2146 int
2147 nd6_need_cache(struct ifnet *ifp)
2148 {
2149 	/*
2150 	 * XXX: we currently do not make neighbor cache on any interface
2151 	 * other than ARCnet, Ethernet, FDDI and GIF.
2152 	 *
2153 	 * RFC2893 says:
2154 	 * - unidirectional tunnels needs no ND
2155 	 */
2156 	switch (ifp->if_type) {
2157 	case IFT_ARCNET:
2158 	case IFT_ETHER:
2159 	case IFT_FDDI:
2160 	case IFT_IEEE1394:
2161 #ifdef IFT_L2VLAN
2162 	case IFT_L2VLAN:
2163 #endif
2164 #ifdef IFT_IEEE80211
2165 	case IFT_IEEE80211:
2166 #endif
2167 	case IFT_INFINIBAND:
2168 	case IFT_GIF:		/* XXX need more cases? */
2169 	case IFT_PPP:
2170 	case IFT_TUNNEL:
2171 	case IFT_BRIDGE:
2172 	case IFT_PROPVIRTUAL:
2173 		return (1);
2174 	default:
2175 		return (0);
2176 	}
2177 }
2178 
2179 /*
2180  * the callers of this function need to be re-worked to drop
2181  * the lle lock, drop here for now
2182  */
2183 int
2184 nd6_storelladdr(struct ifnet *ifp, struct mbuf *m,
2185     const struct sockaddr *dst, u_char *desten, struct llentry **lle)
2186 {
2187 	struct llentry *ln;
2188 
2189 	*lle = NULL;
2190 	IF_AFDATA_UNLOCK_ASSERT(ifp);
2191 	if (m != NULL && m->m_flags & M_MCAST) {
2192 		int i;
2193 
2194 		switch (ifp->if_type) {
2195 		case IFT_ETHER:
2196 		case IFT_FDDI:
2197 #ifdef IFT_L2VLAN
2198 		case IFT_L2VLAN:
2199 #endif
2200 #ifdef IFT_IEEE80211
2201 		case IFT_IEEE80211:
2202 #endif
2203 		case IFT_BRIDGE:
2204 		case IFT_ISO88025:
2205 			ETHER_MAP_IPV6_MULTICAST(&SIN6(dst)->sin6_addr,
2206 						 desten);
2207 			return (0);
2208 		case IFT_IEEE1394:
2209 			/*
2210 			 * netbsd can use if_broadcastaddr, but we don't do so
2211 			 * to reduce # of ifdef.
2212 			 */
2213 			for (i = 0; i < ifp->if_addrlen; i++)
2214 				desten[i] = ~0;
2215 			return (0);
2216 		case IFT_ARCNET:
2217 			*desten = 0;
2218 			return (0);
2219 		default:
2220 			m_freem(m);
2221 			return (EAFNOSUPPORT);
2222 		}
2223 	}
2224 
2225 
2226 	/*
2227 	 * the entry should have been created in nd6_store_lladdr
2228 	 */
2229 	IF_AFDATA_RLOCK(ifp);
2230 	ln = lla_lookup(LLTABLE6(ifp), 0, dst);
2231 	IF_AFDATA_RUNLOCK(ifp);
2232 	if ((ln == NULL) || !(ln->la_flags & LLE_VALID)) {
2233 		if (ln != NULL)
2234 			LLE_RUNLOCK(ln);
2235 		/* this could happen, if we could not allocate memory */
2236 		m_freem(m);
2237 		return (1);
2238 	}
2239 
2240 	bcopy(&ln->ll_addr, desten, ifp->if_addrlen);
2241 	*lle = ln;
2242 	LLE_RUNLOCK(ln);
2243 	/*
2244 	 * A *small* use after free race exists here
2245 	 */
2246 	return (0);
2247 }
2248 
2249 static void
2250 clear_llinfo_pqueue(struct llentry *ln)
2251 {
2252 	struct mbuf *m_hold, *m_hold_next;
2253 
2254 	for (m_hold = ln->la_hold; m_hold; m_hold = m_hold_next) {
2255 		m_hold_next = m_hold->m_nextpkt;
2256 		m_freem(m_hold);
2257 	}
2258 
2259 	ln->la_hold = NULL;
2260 	return;
2261 }
2262 
2263 static int nd6_sysctl_drlist(SYSCTL_HANDLER_ARGS);
2264 static int nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS);
2265 #ifdef SYSCTL_DECL
2266 SYSCTL_DECL(_net_inet6_icmp6);
2267 #endif
2268 SYSCTL_NODE(_net_inet6_icmp6, ICMPV6CTL_ND6_DRLIST, nd6_drlist,
2269 	CTLFLAG_RD, nd6_sysctl_drlist, "");
2270 SYSCTL_NODE(_net_inet6_icmp6, ICMPV6CTL_ND6_PRLIST, nd6_prlist,
2271 	CTLFLAG_RD, nd6_sysctl_prlist, "");
2272 SYSCTL_VNET_INT(_net_inet6_icmp6, ICMPV6CTL_ND6_MAXQLEN, nd6_maxqueuelen,
2273 	CTLFLAG_RW, &VNET_NAME(nd6_maxqueuelen), 1, "");
2274 
2275 static int
2276 nd6_sysctl_drlist(SYSCTL_HANDLER_ARGS)
2277 {
2278 	struct in6_defrouter d;
2279 	struct nd_defrouter *dr;
2280 	int error;
2281 
2282 	if (req->newptr)
2283 		return (EPERM);
2284 
2285 	bzero(&d, sizeof(d));
2286 	d.rtaddr.sin6_family = AF_INET6;
2287 	d.rtaddr.sin6_len = sizeof(d.rtaddr);
2288 
2289 	/*
2290 	 * XXX locking
2291 	 */
2292 	TAILQ_FOREACH(dr, &V_nd_defrouter, dr_entry) {
2293 		d.rtaddr.sin6_addr = dr->rtaddr;
2294 		error = sa6_recoverscope(&d.rtaddr);
2295 		if (error != 0)
2296 			return (error);
2297 		d.flags = dr->flags;
2298 		d.rtlifetime = dr->rtlifetime;
2299 		d.expire = dr->expire + (time_second - time_uptime);
2300 		d.if_index = dr->ifp->if_index;
2301 		error = SYSCTL_OUT(req, &d, sizeof(d));
2302 		if (error != 0)
2303 			return (error);
2304 	}
2305 	return (0);
2306 }
2307 
2308 static int
2309 nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS)
2310 {
2311 	struct in6_prefix p;
2312 	struct sockaddr_in6 s6;
2313 	struct nd_prefix *pr;
2314 	struct nd_pfxrouter *pfr;
2315 	time_t maxexpire;
2316 	int error;
2317 	char ip6buf[INET6_ADDRSTRLEN];
2318 
2319 	if (req->newptr)
2320 		return (EPERM);
2321 
2322 	bzero(&p, sizeof(p));
2323 	p.origin = PR_ORIG_RA;
2324 	bzero(&s6, sizeof(s6));
2325 	s6.sin6_family = AF_INET6;
2326 	s6.sin6_len = sizeof(s6);
2327 
2328 	/*
2329 	 * XXX locking
2330 	 */
2331 	LIST_FOREACH(pr, &V_nd_prefix, ndpr_entry) {
2332 		p.prefix = pr->ndpr_prefix;
2333 		if (sa6_recoverscope(&p.prefix)) {
2334 			log(LOG_ERR, "scope error in prefix list (%s)\n",
2335 			    ip6_sprintf(ip6buf, &p.prefix.sin6_addr));
2336 			/* XXX: press on... */
2337 		}
2338 		p.raflags = pr->ndpr_raf;
2339 		p.prefixlen = pr->ndpr_plen;
2340 		p.vltime = pr->ndpr_vltime;
2341 		p.pltime = pr->ndpr_pltime;
2342 		p.if_index = pr->ndpr_ifp->if_index;
2343 		if (pr->ndpr_vltime == ND6_INFINITE_LIFETIME)
2344 			p.expire = 0;
2345 		else {
2346 			/* XXX: we assume time_t is signed. */
2347 			maxexpire = (-1) &
2348 			    ~((time_t)1 << ((sizeof(maxexpire) * 8) - 1));
2349 			if (pr->ndpr_vltime < maxexpire - pr->ndpr_lastupdate)
2350 				p.expire = pr->ndpr_lastupdate +
2351 				    pr->ndpr_vltime +
2352 				    (time_second - time_uptime);
2353 			else
2354 				p.expire = maxexpire;
2355 		}
2356 		p.refcnt = pr->ndpr_refcnt;
2357 		p.flags = pr->ndpr_stateflags;
2358 		p.advrtrs = 0;
2359 		LIST_FOREACH(pfr, &pr->ndpr_advrtrs, pfr_entry)
2360 			p.advrtrs++;
2361 		error = SYSCTL_OUT(req, &p, sizeof(p));
2362 		if (error != 0)
2363 			return (error);
2364 		LIST_FOREACH(pfr, &pr->ndpr_advrtrs, pfr_entry) {
2365 			s6.sin6_addr = pfr->router->rtaddr;
2366 			if (sa6_recoverscope(&s6))
2367 				log(LOG_ERR,
2368 				    "scope error in prefix list (%s)\n",
2369 				    ip6_sprintf(ip6buf, &pfr->router->rtaddr));
2370 			error = SYSCTL_OUT(req, &s6, sizeof(s6));
2371 			if (error != 0)
2372 				return (error);
2373 		}
2374 	}
2375 	return (0);
2376 }
2377