xref: /freebsd/sys/netinet6/nd6.c (revision b78ee15e9f04ae15c3e1200df974473167524d17)
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/sdt.h>
54 #include <sys/sysctl.h>
55 
56 #include <net/if.h>
57 #include <net/if_var.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 static void nd6_rtrequest(int, struct rtentry *, struct rt_addrinfo *);
137 static int nd6_output_lle(struct ifnet *, struct ifnet *, struct mbuf *,
138 	struct sockaddr_in6 *);
139 static int nd6_output_ifp(struct ifnet *, struct ifnet *, struct mbuf *,
140     struct sockaddr_in6 *);
141 
142 static VNET_DEFINE(struct callout, nd6_slowtimo_ch);
143 #define	V_nd6_slowtimo_ch		VNET(nd6_slowtimo_ch)
144 
145 VNET_DEFINE(struct callout, nd6_timer_ch);
146 
147 void
148 nd6_init(void)
149 {
150 
151 	LIST_INIT(&V_nd_prefix);
152 
153 	/* initialization of the default router list */
154 	TAILQ_INIT(&V_nd_defrouter);
155 
156 	/* start timer */
157 	callout_init(&V_nd6_slowtimo_ch, 0);
158 	callout_reset(&V_nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
159 	    nd6_slowtimo, curvnet);
160 
161 	nd6_dad_init();
162 }
163 
164 #ifdef VIMAGE
165 void
166 nd6_destroy()
167 {
168 
169 	callout_drain(&V_nd6_slowtimo_ch);
170 	callout_drain(&V_nd6_timer_ch);
171 }
172 #endif
173 
174 struct nd_ifinfo *
175 nd6_ifattach(struct ifnet *ifp)
176 {
177 	struct nd_ifinfo *nd;
178 
179 	nd = (struct nd_ifinfo *)malloc(sizeof(*nd), M_IP6NDP, M_WAITOK|M_ZERO);
180 	nd->initialized = 1;
181 
182 	nd->chlim = IPV6_DEFHLIM;
183 	nd->basereachable = REACHABLE_TIME;
184 	nd->reachable = ND_COMPUTE_RTIME(nd->basereachable);
185 	nd->retrans = RETRANS_TIMER;
186 
187 	nd->flags = ND6_IFF_PERFORMNUD;
188 
189 	/* A loopback interface always has ND6_IFF_AUTO_LINKLOCAL.
190 	 * XXXHRS: Clear ND6_IFF_AUTO_LINKLOCAL on an IFT_BRIDGE interface by
191 	 * default regardless of the V_ip6_auto_linklocal configuration to
192 	 * give a reasonable default behavior.
193 	 */
194 	if ((V_ip6_auto_linklocal && ifp->if_type != IFT_BRIDGE) ||
195 	    (ifp->if_flags & IFF_LOOPBACK))
196 		nd->flags |= ND6_IFF_AUTO_LINKLOCAL;
197 	/*
198 	 * A loopback interface does not need to accept RTADV.
199 	 * XXXHRS: Clear ND6_IFF_ACCEPT_RTADV on an IFT_BRIDGE interface by
200 	 * default regardless of the V_ip6_accept_rtadv configuration to
201 	 * prevent the interface from accepting RA messages arrived
202 	 * on one of the member interfaces with ND6_IFF_ACCEPT_RTADV.
203 	 */
204 	if (V_ip6_accept_rtadv &&
205 	    !(ifp->if_flags & IFF_LOOPBACK) &&
206 	    (ifp->if_type != IFT_BRIDGE))
207 			nd->flags |= ND6_IFF_ACCEPT_RTADV;
208 	if (V_ip6_no_radr && !(ifp->if_flags & IFF_LOOPBACK))
209 		nd->flags |= ND6_IFF_NO_RADR;
210 
211 	/* XXX: we cannot call nd6_setmtu since ifp is not fully initialized */
212 	nd6_setmtu0(ifp, nd);
213 
214 	return nd;
215 }
216 
217 void
218 nd6_ifdetach(struct nd_ifinfo *nd)
219 {
220 
221 	free(nd, M_IP6NDP);
222 }
223 
224 /*
225  * Reset ND level link MTU. This function is called when the physical MTU
226  * changes, which means we might have to adjust the ND level MTU.
227  */
228 void
229 nd6_setmtu(struct ifnet *ifp)
230 {
231 
232 	nd6_setmtu0(ifp, ND_IFINFO(ifp));
233 }
234 
235 /* XXX todo: do not maintain copy of ifp->if_mtu in ndi->maxmtu */
236 void
237 nd6_setmtu0(struct ifnet *ifp, struct nd_ifinfo *ndi)
238 {
239 	u_int32_t omaxmtu;
240 
241 	omaxmtu = ndi->maxmtu;
242 
243 	switch (ifp->if_type) {
244 	case IFT_ARCNET:
245 		ndi->maxmtu = MIN(ARC_PHDS_MAXMTU, ifp->if_mtu); /* RFC2497 */
246 		break;
247 	case IFT_FDDI:
248 		ndi->maxmtu = MIN(FDDIIPMTU, ifp->if_mtu); /* RFC2467 */
249 		break;
250 	case IFT_ISO88025:
251 		 ndi->maxmtu = MIN(ISO88025_MAX_MTU, ifp->if_mtu);
252 		 break;
253 	default:
254 		ndi->maxmtu = ifp->if_mtu;
255 		break;
256 	}
257 
258 	/*
259 	 * Decreasing the interface MTU under IPV6 minimum MTU may cause
260 	 * undesirable situation.  We thus notify the operator of the change
261 	 * explicitly.  The check for omaxmtu is necessary to restrict the
262 	 * log to the case of changing the MTU, not initializing it.
263 	 */
264 	if (omaxmtu >= IPV6_MMTU && ndi->maxmtu < IPV6_MMTU) {
265 		log(LOG_NOTICE, "nd6_setmtu0: "
266 		    "new link MTU on %s (%lu) is too small for IPv6\n",
267 		    if_name(ifp), (unsigned long)ndi->maxmtu);
268 	}
269 
270 	if (ndi->maxmtu > V_in6_maxmtu)
271 		in6_setmaxmtu(); /* check all interfaces just in case */
272 
273 }
274 
275 void
276 nd6_option_init(void *opt, int icmp6len, union nd_opts *ndopts)
277 {
278 
279 	bzero(ndopts, sizeof(*ndopts));
280 	ndopts->nd_opts_search = (struct nd_opt_hdr *)opt;
281 	ndopts->nd_opts_last
282 		= (struct nd_opt_hdr *)(((u_char *)opt) + icmp6len);
283 
284 	if (icmp6len == 0) {
285 		ndopts->nd_opts_done = 1;
286 		ndopts->nd_opts_search = NULL;
287 	}
288 }
289 
290 /*
291  * Take one ND option.
292  */
293 struct nd_opt_hdr *
294 nd6_option(union nd_opts *ndopts)
295 {
296 	struct nd_opt_hdr *nd_opt;
297 	int olen;
298 
299 	KASSERT(ndopts != NULL, ("%s: ndopts == NULL", __func__));
300 	KASSERT(ndopts->nd_opts_last != NULL, ("%s: uninitialized ndopts",
301 	    __func__));
302 	if (ndopts->nd_opts_search == NULL)
303 		return NULL;
304 	if (ndopts->nd_opts_done)
305 		return NULL;
306 
307 	nd_opt = ndopts->nd_opts_search;
308 
309 	/* make sure nd_opt_len is inside the buffer */
310 	if ((caddr_t)&nd_opt->nd_opt_len >= (caddr_t)ndopts->nd_opts_last) {
311 		bzero(ndopts, sizeof(*ndopts));
312 		return NULL;
313 	}
314 
315 	olen = nd_opt->nd_opt_len << 3;
316 	if (olen == 0) {
317 		/*
318 		 * Message validation requires that all included
319 		 * options have a length that is greater than zero.
320 		 */
321 		bzero(ndopts, sizeof(*ndopts));
322 		return NULL;
323 	}
324 
325 	ndopts->nd_opts_search = (struct nd_opt_hdr *)((caddr_t)nd_opt + olen);
326 	if (ndopts->nd_opts_search > ndopts->nd_opts_last) {
327 		/* option overruns the end of buffer, invalid */
328 		bzero(ndopts, sizeof(*ndopts));
329 		return NULL;
330 	} else if (ndopts->nd_opts_search == ndopts->nd_opts_last) {
331 		/* reached the end of options chain */
332 		ndopts->nd_opts_done = 1;
333 		ndopts->nd_opts_search = NULL;
334 	}
335 	return nd_opt;
336 }
337 
338 /*
339  * Parse multiple ND options.
340  * This function is much easier to use, for ND routines that do not need
341  * multiple options of the same type.
342  */
343 int
344 nd6_options(union nd_opts *ndopts)
345 {
346 	struct nd_opt_hdr *nd_opt;
347 	int i = 0;
348 
349 	KASSERT(ndopts != NULL, ("%s: ndopts == NULL", __func__));
350 	KASSERT(ndopts->nd_opts_last != NULL, ("%s: uninitialized ndopts",
351 	    __func__));
352 	if (ndopts->nd_opts_search == NULL)
353 		return 0;
354 
355 	while (1) {
356 		nd_opt = nd6_option(ndopts);
357 		if (nd_opt == NULL && ndopts->nd_opts_last == NULL) {
358 			/*
359 			 * Message validation requires that all included
360 			 * options have a length that is greater than zero.
361 			 */
362 			ICMP6STAT_INC(icp6s_nd_badopt);
363 			bzero(ndopts, sizeof(*ndopts));
364 			return -1;
365 		}
366 
367 		if (nd_opt == NULL)
368 			goto skip1;
369 
370 		switch (nd_opt->nd_opt_type) {
371 		case ND_OPT_SOURCE_LINKADDR:
372 		case ND_OPT_TARGET_LINKADDR:
373 		case ND_OPT_MTU:
374 		case ND_OPT_REDIRECTED_HEADER:
375 		case ND_OPT_NONCE:
376 			if (ndopts->nd_opt_array[nd_opt->nd_opt_type]) {
377 				nd6log((LOG_INFO,
378 				    "duplicated ND6 option found (type=%d)\n",
379 				    nd_opt->nd_opt_type));
380 				/* XXX bark? */
381 			} else {
382 				ndopts->nd_opt_array[nd_opt->nd_opt_type]
383 					= nd_opt;
384 			}
385 			break;
386 		case ND_OPT_PREFIX_INFORMATION:
387 			if (ndopts->nd_opt_array[nd_opt->nd_opt_type] == 0) {
388 				ndopts->nd_opt_array[nd_opt->nd_opt_type]
389 					= nd_opt;
390 			}
391 			ndopts->nd_opts_pi_end =
392 				(struct nd_opt_prefix_info *)nd_opt;
393 			break;
394 		/* What about ND_OPT_ROUTE_INFO? RFC 4191 */
395 		case ND_OPT_RDNSS:	/* RFC 6106 */
396 		case ND_OPT_DNSSL:	/* RFC 6106 */
397 			/*
398 			 * Silently ignore options we know and do not care about
399 			 * in the kernel.
400 			 */
401 			break;
402 		default:
403 			/*
404 			 * Unknown options must be silently ignored,
405 			 * to accomodate future extension to the protocol.
406 			 */
407 			nd6log((LOG_DEBUG,
408 			    "nd6_options: unsupported option %d - "
409 			    "option ignored\n", nd_opt->nd_opt_type));
410 		}
411 
412 skip1:
413 		i++;
414 		if (i > V_nd6_maxndopt) {
415 			ICMP6STAT_INC(icp6s_nd_toomanyopt);
416 			nd6log((LOG_INFO, "too many loop in nd opt\n"));
417 			break;
418 		}
419 
420 		if (ndopts->nd_opts_done)
421 			break;
422 	}
423 
424 	return 0;
425 }
426 
427 /*
428  * ND6 timer routine to handle ND6 entries
429  */
430 void
431 nd6_llinfo_settimer_locked(struct llentry *ln, long tick)
432 {
433 	int canceled;
434 
435 	LLE_WLOCK_ASSERT(ln);
436 
437 	if (tick < 0) {
438 		ln->la_expire = 0;
439 		ln->ln_ntick = 0;
440 		canceled = callout_stop(&ln->ln_timer_ch);
441 	} else {
442 		ln->la_expire = time_uptime + tick / hz;
443 		LLE_ADDREF(ln);
444 		if (tick > INT_MAX) {
445 			ln->ln_ntick = tick - INT_MAX;
446 			canceled = callout_reset(&ln->ln_timer_ch, INT_MAX,
447 			    nd6_llinfo_timer, ln);
448 		} else {
449 			ln->ln_ntick = 0;
450 			canceled = callout_reset(&ln->ln_timer_ch, tick,
451 			    nd6_llinfo_timer, ln);
452 		}
453 	}
454 	if (canceled)
455 		LLE_REMREF(ln);
456 }
457 
458 void
459 nd6_llinfo_settimer(struct llentry *ln, long tick)
460 {
461 
462 	LLE_WLOCK(ln);
463 	nd6_llinfo_settimer_locked(ln, tick);
464 	LLE_WUNLOCK(ln);
465 }
466 
467 static void
468 nd6_llinfo_timer(void *arg)
469 {
470 	struct llentry *ln;
471 	struct in6_addr *dst;
472 	struct ifnet *ifp;
473 	struct nd_ifinfo *ndi = NULL;
474 
475 	KASSERT(arg != NULL, ("%s: arg NULL", __func__));
476 	ln = (struct llentry *)arg;
477 	LLE_WLOCK(ln);
478 	if (callout_pending(&ln->la_timer)) {
479 		/*
480 		 * Here we are a bit odd here in the treatment of
481 		 * active/pending. If the pending bit is set, it got
482 		 * rescheduled before I ran. The active
483 		 * bit we ignore, since if it was stopped
484 		 * in ll_tablefree() and was currently running
485 		 * it would have return 0 so the code would
486 		 * not have deleted it since the callout could
487 		 * not be stopped so we want to go through
488 		 * with the delete here now. If the callout
489 		 * was restarted, the pending bit will be back on and
490 		 * we just want to bail since the callout_reset would
491 		 * return 1 and our reference would have been removed
492 		 * by nd6_llinfo_settimer_locked above since canceled
493 		 * would have been 1.
494 		 */
495 		LLE_WUNLOCK(ln);
496 		return;
497 	}
498 	ifp = ln->lle_tbl->llt_ifp;
499 	CURVNET_SET(ifp->if_vnet);
500 
501 	if (ln->ln_ntick > 0) {
502 		if (ln->ln_ntick > INT_MAX) {
503 			ln->ln_ntick -= INT_MAX;
504 			nd6_llinfo_settimer_locked(ln, INT_MAX);
505 		} else {
506 			ln->ln_ntick = 0;
507 			nd6_llinfo_settimer_locked(ln, ln->ln_ntick);
508 		}
509 		goto done;
510 	}
511 
512 	ndi = ND_IFINFO(ifp);
513 	dst = &L3_ADDR_SIN6(ln)->sin6_addr;
514 	if (ln->la_flags & LLE_STATIC) {
515 		goto done;
516 	}
517 
518 	if (ln->la_flags & LLE_DELETED) {
519 		(void)nd6_free(ln, 0);
520 		ln = NULL;
521 		goto done;
522 	}
523 
524 	switch (ln->ln_state) {
525 	case ND6_LLINFO_INCOMPLETE:
526 		if (ln->la_asked < V_nd6_mmaxtries) {
527 			ln->la_asked++;
528 			nd6_llinfo_settimer_locked(ln, (long)ndi->retrans * hz / 1000);
529 			LLE_WUNLOCK(ln);
530 			nd6_ns_output(ifp, NULL, dst, ln, NULL);
531 			LLE_WLOCK(ln);
532 		} else {
533 			struct mbuf *m = ln->la_hold;
534 			if (m) {
535 				struct mbuf *m0;
536 
537 				/*
538 				 * assuming every packet in la_hold has the
539 				 * same IP header.  Send error after unlock.
540 				 */
541 				m0 = m->m_nextpkt;
542 				m->m_nextpkt = NULL;
543 				ln->la_hold = m0;
544 				clear_llinfo_pqueue(ln);
545 			}
546 			EVENTHANDLER_INVOKE(lle_event, ln, LLENTRY_TIMEDOUT);
547 			(void)nd6_free(ln, 0);
548 			ln = NULL;
549 			if (m != NULL)
550 				icmp6_error2(m, ICMP6_DST_UNREACH,
551 				    ICMP6_DST_UNREACH_ADDR, 0, ifp);
552 		}
553 		break;
554 	case ND6_LLINFO_REACHABLE:
555 		if (!ND6_LLINFO_PERMANENT(ln)) {
556 			ln->ln_state = ND6_LLINFO_STALE;
557 			nd6_llinfo_settimer_locked(ln, (long)V_nd6_gctimer * hz);
558 		}
559 		break;
560 
561 	case ND6_LLINFO_STALE:
562 		/* Garbage Collection(RFC 2461 5.3) */
563 		if (!ND6_LLINFO_PERMANENT(ln)) {
564 			EVENTHANDLER_INVOKE(lle_event, ln, LLENTRY_EXPIRED);
565 			(void)nd6_free(ln, 1);
566 			ln = NULL;
567 		}
568 		break;
569 
570 	case ND6_LLINFO_DELAY:
571 		if (ndi && (ndi->flags & ND6_IFF_PERFORMNUD) != 0) {
572 			/* We need NUD */
573 			ln->la_asked = 1;
574 			ln->ln_state = ND6_LLINFO_PROBE;
575 			nd6_llinfo_settimer_locked(ln, (long)ndi->retrans * hz / 1000);
576 			LLE_WUNLOCK(ln);
577 			nd6_ns_output(ifp, dst, dst, ln, NULL);
578 			LLE_WLOCK(ln);
579 		} else {
580 			ln->ln_state = ND6_LLINFO_STALE; /* XXX */
581 			nd6_llinfo_settimer_locked(ln, (long)V_nd6_gctimer * hz);
582 		}
583 		break;
584 	case ND6_LLINFO_PROBE:
585 		if (ln->la_asked < V_nd6_umaxtries) {
586 			ln->la_asked++;
587 			nd6_llinfo_settimer_locked(ln, (long)ndi->retrans * hz / 1000);
588 			LLE_WUNLOCK(ln);
589 			nd6_ns_output(ifp, dst, dst, ln, NULL);
590 			LLE_WLOCK(ln);
591 		} else {
592 			EVENTHANDLER_INVOKE(lle_event, ln, LLENTRY_EXPIRED);
593 			(void)nd6_free(ln, 0);
594 			ln = NULL;
595 		}
596 		break;
597 	default:
598 		panic("%s: paths in a dark night can be confusing: %d",
599 		    __func__, ln->ln_state);
600 	}
601 done:
602 	if (ln != NULL)
603 		LLE_FREE_LOCKED(ln);
604 	CURVNET_RESTORE();
605 }
606 
607 
608 /*
609  * ND6 timer routine to expire default route list and prefix list
610  */
611 void
612 nd6_timer(void *arg)
613 {
614 	CURVNET_SET((struct vnet *) arg);
615 	struct nd_defrouter *dr, *ndr;
616 	struct nd_prefix *pr, *npr;
617 	struct in6_ifaddr *ia6, *nia6;
618 
619 	callout_reset(&V_nd6_timer_ch, V_nd6_prune * hz,
620 	    nd6_timer, curvnet);
621 
622 	/* expire default router list */
623 	TAILQ_FOREACH_SAFE(dr, &V_nd_defrouter, dr_entry, ndr) {
624 		if (dr->expire && dr->expire < time_uptime)
625 			defrtrlist_del(dr);
626 	}
627 
628 	/*
629 	 * expire interface addresses.
630 	 * in the past the loop was inside prefix expiry processing.
631 	 * However, from a stricter speci-confrmance standpoint, we should
632 	 * rather separate address lifetimes and prefix lifetimes.
633 	 *
634 	 * XXXRW: in6_ifaddrhead locking.
635 	 */
636   addrloop:
637 	TAILQ_FOREACH_SAFE(ia6, &V_in6_ifaddrhead, ia_link, nia6) {
638 		/* check address lifetime */
639 		if (IFA6_IS_INVALID(ia6)) {
640 			int regen = 0;
641 
642 			/*
643 			 * If the expiring address is temporary, try
644 			 * regenerating a new one.  This would be useful when
645 			 * we suspended a laptop PC, then turned it on after a
646 			 * period that could invalidate all temporary
647 			 * addresses.  Although we may have to restart the
648 			 * loop (see below), it must be after purging the
649 			 * address.  Otherwise, we'd see an infinite loop of
650 			 * regeneration.
651 			 */
652 			if (V_ip6_use_tempaddr &&
653 			    (ia6->ia6_flags & IN6_IFF_TEMPORARY) != 0) {
654 				if (regen_tmpaddr(ia6) == 0)
655 					regen = 1;
656 			}
657 
658 			in6_purgeaddr(&ia6->ia_ifa);
659 
660 			if (regen)
661 				goto addrloop; /* XXX: see below */
662 		} else if (IFA6_IS_DEPRECATED(ia6)) {
663 			int oldflags = ia6->ia6_flags;
664 
665 			ia6->ia6_flags |= IN6_IFF_DEPRECATED;
666 
667 			/*
668 			 * If a temporary address has just become deprecated,
669 			 * regenerate a new one if possible.
670 			 */
671 			if (V_ip6_use_tempaddr &&
672 			    (ia6->ia6_flags & IN6_IFF_TEMPORARY) != 0 &&
673 			    (oldflags & IN6_IFF_DEPRECATED) == 0) {
674 
675 				if (regen_tmpaddr(ia6) == 0) {
676 					/*
677 					 * A new temporary address is
678 					 * generated.
679 					 * XXX: this means the address chain
680 					 * has changed while we are still in
681 					 * the loop.  Although the change
682 					 * would not cause disaster (because
683 					 * it's not a deletion, but an
684 					 * addition,) we'd rather restart the
685 					 * loop just for safety.  Or does this
686 					 * significantly reduce performance??
687 					 */
688 					goto addrloop;
689 				}
690 			}
691 		} else {
692 			/*
693 			 * A new RA might have made a deprecated address
694 			 * preferred.
695 			 */
696 			ia6->ia6_flags &= ~IN6_IFF_DEPRECATED;
697 		}
698 	}
699 
700 	/* expire prefix list */
701 	LIST_FOREACH_SAFE(pr, &V_nd_prefix, ndpr_entry, npr) {
702 		/*
703 		 * check prefix lifetime.
704 		 * since pltime is just for autoconf, pltime processing for
705 		 * prefix is not necessary.
706 		 */
707 		if (pr->ndpr_vltime != ND6_INFINITE_LIFETIME &&
708 		    time_uptime - pr->ndpr_lastupdate > pr->ndpr_vltime) {
709 
710 			/*
711 			 * address expiration and prefix expiration are
712 			 * separate.  NEVER perform in6_purgeaddr here.
713 			 */
714 			prelist_remove(pr);
715 		}
716 	}
717 	CURVNET_RESTORE();
718 }
719 
720 /*
721  * ia6 - deprecated/invalidated temporary address
722  */
723 static int
724 regen_tmpaddr(struct in6_ifaddr *ia6)
725 {
726 	struct ifaddr *ifa;
727 	struct ifnet *ifp;
728 	struct in6_ifaddr *public_ifa6 = NULL;
729 
730 	ifp = ia6->ia_ifa.ifa_ifp;
731 	IF_ADDR_RLOCK(ifp);
732 	TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
733 		struct in6_ifaddr *it6;
734 
735 		if (ifa->ifa_addr->sa_family != AF_INET6)
736 			continue;
737 
738 		it6 = (struct in6_ifaddr *)ifa;
739 
740 		/* ignore no autoconf addresses. */
741 		if ((it6->ia6_flags & IN6_IFF_AUTOCONF) == 0)
742 			continue;
743 
744 		/* ignore autoconf addresses with different prefixes. */
745 		if (it6->ia6_ndpr == NULL || it6->ia6_ndpr != ia6->ia6_ndpr)
746 			continue;
747 
748 		/*
749 		 * Now we are looking at an autoconf address with the same
750 		 * prefix as ours.  If the address is temporary and is still
751 		 * preferred, do not create another one.  It would be rare, but
752 		 * could happen, for example, when we resume a laptop PC after
753 		 * a long period.
754 		 */
755 		if ((it6->ia6_flags & IN6_IFF_TEMPORARY) != 0 &&
756 		    !IFA6_IS_DEPRECATED(it6)) {
757 			public_ifa6 = NULL;
758 			break;
759 		}
760 
761 		/*
762 		 * This is a public autoconf address that has the same prefix
763 		 * as ours.  If it is preferred, keep it.  We can't break the
764 		 * loop here, because there may be a still-preferred temporary
765 		 * address with the prefix.
766 		 */
767 		if (!IFA6_IS_DEPRECATED(it6))
768 			public_ifa6 = it6;
769 	}
770 	if (public_ifa6 != NULL)
771 		ifa_ref(&public_ifa6->ia_ifa);
772 	IF_ADDR_RUNLOCK(ifp);
773 
774 	if (public_ifa6 != NULL) {
775 		int e;
776 
777 		if ((e = in6_tmpifadd(public_ifa6, 0, 0)) != 0) {
778 			ifa_free(&public_ifa6->ia_ifa);
779 			log(LOG_NOTICE, "regen_tmpaddr: failed to create a new"
780 			    " tmp addr,errno=%d\n", e);
781 			return (-1);
782 		}
783 		ifa_free(&public_ifa6->ia_ifa);
784 		return (0);
785 	}
786 
787 	return (-1);
788 }
789 
790 /*
791  * Nuke neighbor cache/prefix/default router management table, right before
792  * ifp goes away.
793  */
794 void
795 nd6_purge(struct ifnet *ifp)
796 {
797 	struct nd_defrouter *dr, *ndr;
798 	struct nd_prefix *pr, *npr;
799 
800 	/*
801 	 * Nuke default router list entries toward ifp.
802 	 * We defer removal of default router list entries that is installed
803 	 * in the routing table, in order to keep additional side effects as
804 	 * small as possible.
805 	 */
806 	TAILQ_FOREACH_SAFE(dr, &V_nd_defrouter, dr_entry, ndr) {
807 		if (dr->installed)
808 			continue;
809 
810 		if (dr->ifp == ifp)
811 			defrtrlist_del(dr);
812 	}
813 
814 	TAILQ_FOREACH_SAFE(dr, &V_nd_defrouter, dr_entry, ndr) {
815 		if (!dr->installed)
816 			continue;
817 
818 		if (dr->ifp == ifp)
819 			defrtrlist_del(dr);
820 	}
821 
822 	/* Nuke prefix list entries toward ifp */
823 	LIST_FOREACH_SAFE(pr, &V_nd_prefix, ndpr_entry, npr) {
824 		if (pr->ndpr_ifp == ifp) {
825 			/*
826 			 * Because if_detach() does *not* release prefixes
827 			 * while purging addresses the reference count will
828 			 * still be above zero. We therefore reset it to
829 			 * make sure that the prefix really gets purged.
830 			 */
831 			pr->ndpr_refcnt = 0;
832 
833 			/*
834 			 * Previously, pr->ndpr_addr is removed as well,
835 			 * but I strongly believe we don't have to do it.
836 			 * nd6_purge() is only called from in6_ifdetach(),
837 			 * which removes all the associated interface addresses
838 			 * by itself.
839 			 * (jinmei@kame.net 20010129)
840 			 */
841 			prelist_remove(pr);
842 		}
843 	}
844 
845 	/* cancel default outgoing interface setting */
846 	if (V_nd6_defifindex == ifp->if_index)
847 		nd6_setdefaultiface(0);
848 
849 	if (ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) {
850 		/* Refresh default router list. */
851 		defrouter_select();
852 	}
853 
854 	/* XXXXX
855 	 * We do not nuke the neighbor cache entries here any more
856 	 * because the neighbor cache is kept in if_afdata[AF_INET6].
857 	 * nd6_purge() is invoked by in6_ifdetach() which is called
858 	 * from if_detach() where everything gets purged. So let
859 	 * in6_domifdetach() do the actual L2 table purging work.
860 	 */
861 }
862 
863 /*
864  * the caller acquires and releases the lock on the lltbls
865  * Returns the llentry locked
866  */
867 struct llentry *
868 nd6_lookup(struct in6_addr *addr6, int flags, struct ifnet *ifp)
869 {
870 	struct sockaddr_in6 sin6;
871 	struct llentry *ln;
872 	int llflags;
873 
874 	bzero(&sin6, sizeof(sin6));
875 	sin6.sin6_len = sizeof(struct sockaddr_in6);
876 	sin6.sin6_family = AF_INET6;
877 	sin6.sin6_addr = *addr6;
878 
879 	IF_AFDATA_LOCK_ASSERT(ifp);
880 
881 	llflags = 0;
882 	if (flags & ND6_CREATE)
883 	    llflags |= LLE_CREATE;
884 	if (flags & ND6_EXCLUSIVE)
885 	    llflags |= LLE_EXCLUSIVE;
886 
887 	ln = lla_lookup(LLTABLE6(ifp), llflags, (struct sockaddr *)&sin6);
888 	if ((ln != NULL) && (llflags & LLE_CREATE))
889 		ln->ln_state = ND6_LLINFO_NOSTATE;
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 	LIST_FOREACH(pr, &V_nd_prefix, ndpr_entry) {
935 		if (pr->ndpr_ifp != ifp)
936 			continue;
937 
938 		if (!(pr->ndpr_stateflags & NDPRF_ONLINK)) {
939 			struct rtentry *rt;
940 
941 			/* Always use the default FIB here. */
942 			rt = in6_rtalloc1((struct sockaddr *)&pr->ndpr_prefix,
943 			    0, 0, RT_DEFAULT_FIB);
944 			if (rt == NULL)
945 				continue;
946 			/*
947 			 * This is the case where multiple interfaces
948 			 * have the same prefix, but only one is installed
949 			 * into the routing table and that prefix entry
950 			 * is not the one being examined here. In the case
951 			 * where RADIX_MPATH is enabled, multiple route
952 			 * entries (of the same rt_key value) will be
953 			 * installed because the interface addresses all
954 			 * differ.
955 			 */
956 			if (!IN6_ARE_ADDR_EQUAL(&pr->ndpr_prefix.sin6_addr,
957 			       &((struct sockaddr_in6 *)rt_key(rt))->sin6_addr)) {
958 				RTFREE_LOCKED(rt);
959 				continue;
960 			}
961 			RTFREE_LOCKED(rt);
962 		}
963 
964 		if (IN6_ARE_MASKED_ADDR_EQUAL(&pr->ndpr_prefix.sin6_addr,
965 		    &addr->sin6_addr, &pr->ndpr_mask))
966 			return (1);
967 	}
968 
969 	/*
970 	 * If the address is assigned on the node of the other side of
971 	 * a p2p interface, the address should be a neighbor.
972 	 */
973 	dstaddr = ifa_ifwithdstaddr((struct sockaddr *)addr, RT_ALL_FIBS);
974 	if (dstaddr != NULL) {
975 		if (dstaddr->ifa_ifp == ifp) {
976 			ifa_free(dstaddr);
977 			return (1);
978 		}
979 		ifa_free(dstaddr);
980 	}
981 
982 	/*
983 	 * If the default router list is empty, all addresses are regarded
984 	 * as on-link, and thus, as a neighbor.
985 	 */
986 	if (ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV &&
987 	    TAILQ_EMPTY(&V_nd_defrouter) &&
988 	    V_nd6_defifindex == ifp->if_index) {
989 		return (1);
990 	}
991 
992 	return (0);
993 }
994 
995 
996 /*
997  * Detect if a given IPv6 address identifies a neighbor on a given link.
998  * XXX: should take care of the destination of a p2p link?
999  */
1000 int
1001 nd6_is_addr_neighbor(struct sockaddr_in6 *addr, struct ifnet *ifp)
1002 {
1003 	struct llentry *lle;
1004 	int rc = 0;
1005 
1006 	IF_AFDATA_UNLOCK_ASSERT(ifp);
1007 	if (nd6_is_new_addr_neighbor(addr, ifp))
1008 		return (1);
1009 
1010 	/*
1011 	 * Even if the address matches none of our addresses, it might be
1012 	 * in the neighbor cache.
1013 	 */
1014 	IF_AFDATA_RLOCK(ifp);
1015 	if ((lle = nd6_lookup(&addr->sin6_addr, 0, ifp)) != NULL) {
1016 		LLE_RUNLOCK(lle);
1017 		rc = 1;
1018 	}
1019 	IF_AFDATA_RUNLOCK(ifp);
1020 	return (rc);
1021 }
1022 
1023 /*
1024  * Free an nd6 llinfo entry.
1025  * Since the function would cause significant changes in the kernel, DO NOT
1026  * make it global, unless you have a strong reason for the change, and are sure
1027  * that the change is safe.
1028  */
1029 static struct llentry *
1030 nd6_free(struct llentry *ln, int gc)
1031 {
1032         struct llentry *next;
1033 	struct nd_defrouter *dr;
1034 	struct ifnet *ifp;
1035 
1036 	LLE_WLOCK_ASSERT(ln);
1037 
1038 	/*
1039 	 * we used to have pfctlinput(PRC_HOSTDEAD) here.
1040 	 * even though it is not harmful, it was not really necessary.
1041 	 */
1042 
1043 	/* cancel timer */
1044 	nd6_llinfo_settimer_locked(ln, -1);
1045 
1046 	ifp = ln->lle_tbl->llt_ifp;
1047 
1048 	if (ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) {
1049 		dr = defrouter_lookup(&L3_ADDR_SIN6(ln)->sin6_addr, ifp);
1050 
1051 		if (dr != NULL && dr->expire &&
1052 		    ln->ln_state == ND6_LLINFO_STALE && gc) {
1053 			/*
1054 			 * If the reason for the deletion is just garbage
1055 			 * collection, and the neighbor is an active default
1056 			 * router, do not delete it.  Instead, reset the GC
1057 			 * timer using the router's lifetime.
1058 			 * Simply deleting the entry would affect default
1059 			 * router selection, which is not necessarily a good
1060 			 * thing, especially when we're using router preference
1061 			 * values.
1062 			 * XXX: the check for ln_state would be redundant,
1063 			 *      but we intentionally keep it just in case.
1064 			 */
1065 			if (dr->expire > time_uptime)
1066 				nd6_llinfo_settimer_locked(ln,
1067 				    (dr->expire - time_uptime) * hz);
1068 			else
1069 				nd6_llinfo_settimer_locked(ln,
1070 				    (long)V_nd6_gctimer * hz);
1071 
1072 			next = LIST_NEXT(ln, lle_next);
1073 			LLE_REMREF(ln);
1074 			LLE_WUNLOCK(ln);
1075 			return (next);
1076 		}
1077 
1078 		if (dr) {
1079 			/*
1080 			 * Unreachablity of a router might affect the default
1081 			 * router selection and on-link detection of advertised
1082 			 * prefixes.
1083 			 */
1084 
1085 			/*
1086 			 * Temporarily fake the state to choose a new default
1087 			 * router and to perform on-link determination of
1088 			 * prefixes correctly.
1089 			 * Below the state will be set correctly,
1090 			 * or the entry itself will be deleted.
1091 			 */
1092 			ln->ln_state = ND6_LLINFO_INCOMPLETE;
1093 		}
1094 
1095 		if (ln->ln_router || dr) {
1096 
1097 			/*
1098 			 * We need to unlock to avoid a LOR with rt6_flush() with the
1099 			 * rnh and for the calls to pfxlist_onlink_check() and
1100 			 * defrouter_select() in the block further down for calls
1101 			 * into nd6_lookup().  We still hold a ref.
1102 			 */
1103 			LLE_WUNLOCK(ln);
1104 
1105 			/*
1106 			 * rt6_flush must be called whether or not the neighbor
1107 			 * is in the Default Router List.
1108 			 * See a corresponding comment in nd6_na_input().
1109 			 */
1110 			rt6_flush(&L3_ADDR_SIN6(ln)->sin6_addr, ifp);
1111 		}
1112 
1113 		if (dr) {
1114 			/*
1115 			 * Since defrouter_select() does not affect the
1116 			 * on-link determination and MIP6 needs the check
1117 			 * before the default router selection, we perform
1118 			 * the check now.
1119 			 */
1120 			pfxlist_onlink_check();
1121 
1122 			/*
1123 			 * Refresh default router list.
1124 			 */
1125 			defrouter_select();
1126 		}
1127 
1128 		if (ln->ln_router || dr)
1129 			LLE_WLOCK(ln);
1130 	}
1131 
1132 	/*
1133 	 * Before deleting the entry, remember the next entry as the
1134 	 * return value.  We need this because pfxlist_onlink_check() above
1135 	 * might have freed other entries (particularly the old next entry) as
1136 	 * a side effect (XXX).
1137 	 */
1138 	next = LIST_NEXT(ln, lle_next);
1139 
1140 	/*
1141 	 * Save to unlock. We still hold an extra reference and will not
1142 	 * free(9) in llentry_free() if someone else holds one as well.
1143 	 */
1144 	LLE_WUNLOCK(ln);
1145 	IF_AFDATA_LOCK(ifp);
1146 	LLE_WLOCK(ln);
1147 
1148 	/* Guard against race with other llentry_free(). */
1149 	if (ln->la_flags & LLE_LINKED) {
1150 		LLE_REMREF(ln);
1151 		llentry_free(ln);
1152 	} else
1153 		LLE_FREE_LOCKED(ln);
1154 
1155 	IF_AFDATA_UNLOCK(ifp);
1156 
1157 	return (next);
1158 }
1159 
1160 /*
1161  * Upper-layer reachability hint for Neighbor Unreachability Detection.
1162  *
1163  * XXX cost-effective methods?
1164  */
1165 void
1166 nd6_nud_hint(struct rtentry *rt, struct in6_addr *dst6, int force)
1167 {
1168 	struct llentry *ln;
1169 	struct ifnet *ifp;
1170 
1171 	if ((dst6 == NULL) || (rt == NULL))
1172 		return;
1173 
1174 	ifp = rt->rt_ifp;
1175 	IF_AFDATA_RLOCK(ifp);
1176 	ln = nd6_lookup(dst6, ND6_EXCLUSIVE, NULL);
1177 	IF_AFDATA_RUNLOCK(ifp);
1178 	if (ln == NULL)
1179 		return;
1180 
1181 	if (ln->ln_state < ND6_LLINFO_REACHABLE)
1182 		goto done;
1183 
1184 	/*
1185 	 * if we get upper-layer reachability confirmation many times,
1186 	 * it is possible we have false information.
1187 	 */
1188 	if (!force) {
1189 		ln->ln_byhint++;
1190 		if (ln->ln_byhint > V_nd6_maxnudhint) {
1191 			goto done;
1192 		}
1193 	}
1194 
1195  	ln->ln_state = ND6_LLINFO_REACHABLE;
1196 	if (!ND6_LLINFO_PERMANENT(ln)) {
1197 		nd6_llinfo_settimer_locked(ln,
1198 		    (long)ND_IFINFO(rt->rt_ifp)->reachable * hz);
1199 	}
1200 done:
1201 	LLE_WUNLOCK(ln);
1202 }
1203 
1204 
1205 /*
1206  * Rejuvenate this function for routing operations related
1207  * processing.
1208  */
1209 void
1210 nd6_rtrequest(int req, struct rtentry *rt, struct rt_addrinfo *info)
1211 {
1212 	struct sockaddr_in6 *gateway;
1213 	struct nd_defrouter *dr;
1214 	struct ifnet *ifp;
1215 
1216 	gateway = (struct sockaddr_in6 *)rt->rt_gateway;
1217 	ifp = rt->rt_ifp;
1218 
1219 	switch (req) {
1220 	case RTM_ADD:
1221 		break;
1222 
1223 	case RTM_DELETE:
1224 		if (!ifp)
1225 			return;
1226 		/*
1227 		 * Only indirect routes are interesting.
1228 		 */
1229 		if ((rt->rt_flags & RTF_GATEWAY) == 0)
1230 			return;
1231 		/*
1232 		 * check for default route
1233 		 */
1234 		if (IN6_ARE_ADDR_EQUAL(&in6addr_any,
1235 				       &SIN6(rt_key(rt))->sin6_addr)) {
1236 
1237 			dr = defrouter_lookup(&gateway->sin6_addr, ifp);
1238 			if (dr != NULL)
1239 				dr->installed = 0;
1240 		}
1241 		break;
1242 	}
1243 }
1244 
1245 
1246 int
1247 nd6_ioctl(u_long cmd, caddr_t data, struct ifnet *ifp)
1248 {
1249 	struct in6_drlist *drl = (struct in6_drlist *)data;
1250 	struct in6_oprlist *oprl = (struct in6_oprlist *)data;
1251 	struct in6_ndireq *ndi = (struct in6_ndireq *)data;
1252 	struct in6_nbrinfo *nbi = (struct in6_nbrinfo *)data;
1253 	struct in6_ndifreq *ndif = (struct in6_ndifreq *)data;
1254 	struct nd_defrouter *dr;
1255 	struct nd_prefix *pr;
1256 	int i = 0, error = 0;
1257 
1258 	if (ifp->if_afdata[AF_INET6] == NULL)
1259 		return (EPFNOSUPPORT);
1260 	switch (cmd) {
1261 	case SIOCGDRLST_IN6:
1262 		/*
1263 		 * obsolete API, use sysctl under net.inet6.icmp6
1264 		 */
1265 		bzero(drl, sizeof(*drl));
1266 		TAILQ_FOREACH(dr, &V_nd_defrouter, dr_entry) {
1267 			if (i >= DRLSTSIZ)
1268 				break;
1269 			drl->defrouter[i].rtaddr = dr->rtaddr;
1270 			in6_clearscope(&drl->defrouter[i].rtaddr);
1271 
1272 			drl->defrouter[i].flags = dr->flags;
1273 			drl->defrouter[i].rtlifetime = dr->rtlifetime;
1274 			drl->defrouter[i].expire = dr->expire +
1275 			    (time_second - time_uptime);
1276 			drl->defrouter[i].if_index = dr->ifp->if_index;
1277 			i++;
1278 		}
1279 		break;
1280 	case SIOCGPRLST_IN6:
1281 		/*
1282 		 * obsolete API, use sysctl under net.inet6.icmp6
1283 		 *
1284 		 * XXX the structure in6_prlist was changed in backward-
1285 		 * incompatible manner.  in6_oprlist is used for SIOCGPRLST_IN6,
1286 		 * in6_prlist is used for nd6_sysctl() - fill_prlist().
1287 		 */
1288 		/*
1289 		 * XXX meaning of fields, especialy "raflags", is very
1290 		 * differnet between RA prefix list and RR/static prefix list.
1291 		 * how about separating ioctls into two?
1292 		 */
1293 		bzero(oprl, sizeof(*oprl));
1294 		LIST_FOREACH(pr, &V_nd_prefix, ndpr_entry) {
1295 			struct nd_pfxrouter *pfr;
1296 			int j;
1297 
1298 			if (i >= PRLSTSIZ)
1299 				break;
1300 			oprl->prefix[i].prefix = pr->ndpr_prefix.sin6_addr;
1301 			oprl->prefix[i].raflags = pr->ndpr_raf;
1302 			oprl->prefix[i].prefixlen = pr->ndpr_plen;
1303 			oprl->prefix[i].vltime = pr->ndpr_vltime;
1304 			oprl->prefix[i].pltime = pr->ndpr_pltime;
1305 			oprl->prefix[i].if_index = pr->ndpr_ifp->if_index;
1306 			if (pr->ndpr_vltime == ND6_INFINITE_LIFETIME)
1307 				oprl->prefix[i].expire = 0;
1308 			else {
1309 				time_t maxexpire;
1310 
1311 				/* XXX: we assume time_t is signed. */
1312 				maxexpire = (-1) &
1313 				    ~((time_t)1 <<
1314 				    ((sizeof(maxexpire) * 8) - 1));
1315 				if (pr->ndpr_vltime <
1316 				    maxexpire - pr->ndpr_lastupdate) {
1317 					oprl->prefix[i].expire =
1318 					    pr->ndpr_lastupdate +
1319 					    pr->ndpr_vltime +
1320 					    (time_second - time_uptime);
1321 				} else
1322 					oprl->prefix[i].expire = maxexpire;
1323 			}
1324 
1325 			j = 0;
1326 			LIST_FOREACH(pfr, &pr->ndpr_advrtrs, pfr_entry) {
1327 				if (j < DRLSTSIZ) {
1328 #define RTRADDR oprl->prefix[i].advrtr[j]
1329 					RTRADDR = pfr->router->rtaddr;
1330 					in6_clearscope(&RTRADDR);
1331 #undef RTRADDR
1332 				}
1333 				j++;
1334 			}
1335 			oprl->prefix[i].advrtrs = j;
1336 			oprl->prefix[i].origin = PR_ORIG_RA;
1337 
1338 			i++;
1339 		}
1340 
1341 		break;
1342 	case OSIOCGIFINFO_IN6:
1343 #define ND	ndi->ndi
1344 		/* XXX: old ndp(8) assumes a positive value for linkmtu. */
1345 		bzero(&ND, sizeof(ND));
1346 		ND.linkmtu = IN6_LINKMTU(ifp);
1347 		ND.maxmtu = ND_IFINFO(ifp)->maxmtu;
1348 		ND.basereachable = ND_IFINFO(ifp)->basereachable;
1349 		ND.reachable = ND_IFINFO(ifp)->reachable;
1350 		ND.retrans = ND_IFINFO(ifp)->retrans;
1351 		ND.flags = ND_IFINFO(ifp)->flags;
1352 		ND.recalctm = ND_IFINFO(ifp)->recalctm;
1353 		ND.chlim = ND_IFINFO(ifp)->chlim;
1354 		break;
1355 	case SIOCGIFINFO_IN6:
1356 		ND = *ND_IFINFO(ifp);
1357 		break;
1358 	case SIOCSIFINFO_IN6:
1359 		/*
1360 		 * used to change host variables from userland.
1361 		 * intented for a use on router to reflect RA configurations.
1362 		 */
1363 		/* 0 means 'unspecified' */
1364 		if (ND.linkmtu != 0) {
1365 			if (ND.linkmtu < IPV6_MMTU ||
1366 			    ND.linkmtu > IN6_LINKMTU(ifp)) {
1367 				error = EINVAL;
1368 				break;
1369 			}
1370 			ND_IFINFO(ifp)->linkmtu = ND.linkmtu;
1371 		}
1372 
1373 		if (ND.basereachable != 0) {
1374 			int obasereachable = ND_IFINFO(ifp)->basereachable;
1375 
1376 			ND_IFINFO(ifp)->basereachable = ND.basereachable;
1377 			if (ND.basereachable != obasereachable)
1378 				ND_IFINFO(ifp)->reachable =
1379 				    ND_COMPUTE_RTIME(ND.basereachable);
1380 		}
1381 		if (ND.retrans != 0)
1382 			ND_IFINFO(ifp)->retrans = ND.retrans;
1383 		if (ND.chlim != 0)
1384 			ND_IFINFO(ifp)->chlim = ND.chlim;
1385 		/* FALLTHROUGH */
1386 	case SIOCSIFINFO_FLAGS:
1387 	{
1388 		struct ifaddr *ifa;
1389 		struct in6_ifaddr *ia;
1390 
1391 		if ((ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED) &&
1392 		    !(ND.flags & ND6_IFF_IFDISABLED)) {
1393 			/* ifdisabled 1->0 transision */
1394 
1395 			/*
1396 			 * If the interface is marked as ND6_IFF_IFDISABLED and
1397 			 * has an link-local address with IN6_IFF_DUPLICATED,
1398 			 * do not clear ND6_IFF_IFDISABLED.
1399 			 * See RFC 4862, Section 5.4.5.
1400 			 */
1401 			int duplicated_linklocal = 0;
1402 
1403 			IF_ADDR_RLOCK(ifp);
1404 			TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1405 				if (ifa->ifa_addr->sa_family != AF_INET6)
1406 					continue;
1407 				ia = (struct in6_ifaddr *)ifa;
1408 				if ((ia->ia6_flags & IN6_IFF_DUPLICATED) &&
1409 				    IN6_IS_ADDR_LINKLOCAL(IA6_IN6(ia))) {
1410 					duplicated_linklocal = 1;
1411 					break;
1412 				}
1413 			}
1414 			IF_ADDR_RUNLOCK(ifp);
1415 
1416 			if (duplicated_linklocal) {
1417 				ND.flags |= ND6_IFF_IFDISABLED;
1418 				log(LOG_ERR, "Cannot enable an interface"
1419 				    " with a link-local address marked"
1420 				    " duplicate.\n");
1421 			} else {
1422 				ND_IFINFO(ifp)->flags &= ~ND6_IFF_IFDISABLED;
1423 				if (ifp->if_flags & IFF_UP)
1424 					in6_if_up(ifp);
1425 			}
1426 		} else if (!(ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED) &&
1427 			    (ND.flags & ND6_IFF_IFDISABLED)) {
1428 			/* ifdisabled 0->1 transision */
1429 			/* Mark all IPv6 address as tentative. */
1430 
1431 			ND_IFINFO(ifp)->flags |= ND6_IFF_IFDISABLED;
1432 			IF_ADDR_RLOCK(ifp);
1433 			TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1434 				if (ifa->ifa_addr->sa_family != AF_INET6)
1435 					continue;
1436 				ia = (struct in6_ifaddr *)ifa;
1437 				ia->ia6_flags |= IN6_IFF_TENTATIVE;
1438 			}
1439 			IF_ADDR_RUNLOCK(ifp);
1440 		}
1441 
1442 		if (ND.flags & ND6_IFF_AUTO_LINKLOCAL) {
1443 			if (!(ND_IFINFO(ifp)->flags & ND6_IFF_AUTO_LINKLOCAL)) {
1444 				/* auto_linklocal 0->1 transision */
1445 
1446 				/* If no link-local address on ifp, configure */
1447 				ND_IFINFO(ifp)->flags |= ND6_IFF_AUTO_LINKLOCAL;
1448 				in6_ifattach(ifp, NULL);
1449 			} else if (!(ND.flags & ND6_IFF_IFDISABLED) &&
1450 			    ifp->if_flags & IFF_UP) {
1451 				/*
1452 				 * When the IF already has
1453 				 * ND6_IFF_AUTO_LINKLOCAL, no link-local
1454 				 * address is assigned, and IFF_UP, try to
1455 				 * assign one.
1456 				 */
1457 				int haslinklocal = 0;
1458 
1459 				IF_ADDR_RLOCK(ifp);
1460 				TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1461 					if (ifa->ifa_addr->sa_family != AF_INET6)
1462 						continue;
1463 					ia = (struct in6_ifaddr *)ifa;
1464 					if (IN6_IS_ADDR_LINKLOCAL(IA6_IN6(ia))) {
1465 						haslinklocal = 1;
1466 						break;
1467 					}
1468 				}
1469 				IF_ADDR_RUNLOCK(ifp);
1470 				if (!haslinklocal)
1471 					in6_ifattach(ifp, NULL);
1472 			}
1473 		}
1474 	}
1475 		ND_IFINFO(ifp)->flags = ND.flags;
1476 		break;
1477 #undef ND
1478 	case SIOCSNDFLUSH_IN6:	/* XXX: the ioctl name is confusing... */
1479 		/* sync kernel routing table with the default router list */
1480 		defrouter_reset();
1481 		defrouter_select();
1482 		break;
1483 	case SIOCSPFXFLUSH_IN6:
1484 	{
1485 		/* flush all the prefix advertised by routers */
1486 		struct nd_prefix *pr, *next;
1487 
1488 		LIST_FOREACH_SAFE(pr, &V_nd_prefix, ndpr_entry, next) {
1489 			struct in6_ifaddr *ia, *ia_next;
1490 
1491 			if (IN6_IS_ADDR_LINKLOCAL(&pr->ndpr_prefix.sin6_addr))
1492 				continue; /* XXX */
1493 
1494 			/* do we really have to remove addresses as well? */
1495 			/* XXXRW: in6_ifaddrhead locking. */
1496 			TAILQ_FOREACH_SAFE(ia, &V_in6_ifaddrhead, ia_link,
1497 			    ia_next) {
1498 				if ((ia->ia6_flags & IN6_IFF_AUTOCONF) == 0)
1499 					continue;
1500 
1501 				if (ia->ia6_ndpr == pr)
1502 					in6_purgeaddr(&ia->ia_ifa);
1503 			}
1504 			prelist_remove(pr);
1505 		}
1506 		break;
1507 	}
1508 	case SIOCSRTRFLUSH_IN6:
1509 	{
1510 		/* flush all the default routers */
1511 		struct nd_defrouter *dr, *next;
1512 
1513 		defrouter_reset();
1514 		TAILQ_FOREACH_SAFE(dr, &V_nd_defrouter, dr_entry, next) {
1515 			defrtrlist_del(dr);
1516 		}
1517 		defrouter_select();
1518 		break;
1519 	}
1520 	case SIOCGNBRINFO_IN6:
1521 	{
1522 		struct llentry *ln;
1523 		struct in6_addr nb_addr = nbi->addr; /* make local for safety */
1524 
1525 		if ((error = in6_setscope(&nb_addr, ifp, NULL)) != 0)
1526 			return (error);
1527 
1528 		IF_AFDATA_RLOCK(ifp);
1529 		ln = nd6_lookup(&nb_addr, 0, ifp);
1530 		IF_AFDATA_RUNLOCK(ifp);
1531 
1532 		if (ln == NULL) {
1533 			error = EINVAL;
1534 			break;
1535 		}
1536 		nbi->state = ln->ln_state;
1537 		nbi->asked = ln->la_asked;
1538 		nbi->isrouter = ln->ln_router;
1539 		if (ln->la_expire == 0)
1540 			nbi->expire = 0;
1541 		else
1542 			nbi->expire = ln->la_expire +
1543 			    (time_second - time_uptime);
1544 		LLE_RUNLOCK(ln);
1545 		break;
1546 	}
1547 	case SIOCGDEFIFACE_IN6:	/* XXX: should be implemented as a sysctl? */
1548 		ndif->ifindex = V_nd6_defifindex;
1549 		break;
1550 	case SIOCSDEFIFACE_IN6:	/* XXX: should be implemented as a sysctl? */
1551 		return (nd6_setdefaultiface(ndif->ifindex));
1552 	}
1553 	return (error);
1554 }
1555 
1556 /*
1557  * Create neighbor cache entry and cache link-layer address,
1558  * on reception of inbound ND6 packets.  (RS/RA/NS/redirect)
1559  *
1560  * type - ICMP6 type
1561  * code - type dependent information
1562  *
1563  * XXXXX
1564  *  The caller of this function already acquired the ndp
1565  *  cache table lock because the cache entry is returned.
1566  */
1567 struct llentry *
1568 nd6_cache_lladdr(struct ifnet *ifp, struct in6_addr *from, char *lladdr,
1569     int lladdrlen, int type, int code)
1570 {
1571 	struct llentry *ln = NULL;
1572 	int is_newentry;
1573 	int do_update;
1574 	int olladdr;
1575 	int llchange;
1576 	int flags;
1577 	int newstate = 0;
1578 	uint16_t router = 0;
1579 	struct sockaddr_in6 sin6;
1580 	struct mbuf *chain = NULL;
1581 	int static_route = 0;
1582 
1583 	IF_AFDATA_UNLOCK_ASSERT(ifp);
1584 
1585 	KASSERT(ifp != NULL, ("%s: ifp == NULL", __func__));
1586 	KASSERT(from != NULL, ("%s: from == NULL", __func__));
1587 
1588 	/* nothing must be updated for unspecified address */
1589 	if (IN6_IS_ADDR_UNSPECIFIED(from))
1590 		return NULL;
1591 
1592 	/*
1593 	 * Validation about ifp->if_addrlen and lladdrlen must be done in
1594 	 * the caller.
1595 	 *
1596 	 * XXX If the link does not have link-layer adderss, what should
1597 	 * we do? (ifp->if_addrlen == 0)
1598 	 * Spec says nothing in sections for RA, RS and NA.  There's small
1599 	 * description on it in NS section (RFC 2461 7.2.3).
1600 	 */
1601 	flags = lladdr ? ND6_EXCLUSIVE : 0;
1602 	IF_AFDATA_RLOCK(ifp);
1603 	ln = nd6_lookup(from, flags, ifp);
1604 	IF_AFDATA_RUNLOCK(ifp);
1605 	if (ln == NULL) {
1606 		flags |= ND6_EXCLUSIVE;
1607 		IF_AFDATA_LOCK(ifp);
1608 		ln = nd6_lookup(from, flags | ND6_CREATE, ifp);
1609 		IF_AFDATA_UNLOCK(ifp);
1610 		is_newentry = 1;
1611 	} else {
1612 		/* do nothing if static ndp is set */
1613 		if (ln->la_flags & LLE_STATIC) {
1614 			static_route = 1;
1615 			goto done;
1616 		}
1617 		is_newentry = 0;
1618 	}
1619 	if (ln == NULL)
1620 		return (NULL);
1621 
1622 	olladdr = (ln->la_flags & LLE_VALID) ? 1 : 0;
1623 	if (olladdr && lladdr) {
1624 		llchange = bcmp(lladdr, &ln->ll_addr,
1625 		    ifp->if_addrlen);
1626 	} else
1627 		llchange = 0;
1628 
1629 	/*
1630 	 * newentry olladdr  lladdr  llchange	(*=record)
1631 	 *	0	n	n	--	(1)
1632 	 *	0	y	n	--	(2)
1633 	 *	0	n	y	--	(3) * STALE
1634 	 *	0	y	y	n	(4) *
1635 	 *	0	y	y	y	(5) * STALE
1636 	 *	1	--	n	--	(6)   NOSTATE(= PASSIVE)
1637 	 *	1	--	y	--	(7) * STALE
1638 	 */
1639 
1640 	if (lladdr) {		/* (3-5) and (7) */
1641 		/*
1642 		 * Record source link-layer address
1643 		 * XXX is it dependent to ifp->if_type?
1644 		 */
1645 		bcopy(lladdr, &ln->ll_addr, ifp->if_addrlen);
1646 		ln->la_flags |= LLE_VALID;
1647 		EVENTHANDLER_INVOKE(lle_event, ln, LLENTRY_RESOLVED);
1648 	}
1649 
1650 	if (!is_newentry) {
1651 		if ((!olladdr && lladdr != NULL) ||	/* (3) */
1652 		    (olladdr && lladdr != NULL && llchange)) {	/* (5) */
1653 			do_update = 1;
1654 			newstate = ND6_LLINFO_STALE;
1655 		} else					/* (1-2,4) */
1656 			do_update = 0;
1657 	} else {
1658 		do_update = 1;
1659 		if (lladdr == NULL)			/* (6) */
1660 			newstate = ND6_LLINFO_NOSTATE;
1661 		else					/* (7) */
1662 			newstate = ND6_LLINFO_STALE;
1663 	}
1664 
1665 	if (do_update) {
1666 		/*
1667 		 * Update the state of the neighbor cache.
1668 		 */
1669 		ln->ln_state = newstate;
1670 
1671 		if (ln->ln_state == ND6_LLINFO_STALE) {
1672 			if (ln->la_hold != NULL)
1673 				nd6_grab_holdchain(ln, &chain, &sin6);
1674 		} else if (ln->ln_state == ND6_LLINFO_INCOMPLETE) {
1675 			/* probe right away */
1676 			nd6_llinfo_settimer_locked((void *)ln, 0);
1677 		}
1678 	}
1679 
1680 	/*
1681 	 * ICMP6 type dependent behavior.
1682 	 *
1683 	 * NS: clear IsRouter if new entry
1684 	 * RS: clear IsRouter
1685 	 * RA: set IsRouter if there's lladdr
1686 	 * redir: clear IsRouter if new entry
1687 	 *
1688 	 * RA case, (1):
1689 	 * The spec says that we must set IsRouter in the following cases:
1690 	 * - If lladdr exist, set IsRouter.  This means (1-5).
1691 	 * - If it is old entry (!newentry), set IsRouter.  This means (7).
1692 	 * So, based on the spec, in (1-5) and (7) cases we must set IsRouter.
1693 	 * A quetion arises for (1) case.  (1) case has no lladdr in the
1694 	 * neighbor cache, this is similar to (6).
1695 	 * This case is rare but we figured that we MUST NOT set IsRouter.
1696 	 *
1697 	 * newentry olladdr  lladdr  llchange	    NS  RS  RA	redir
1698 	 *							D R
1699 	 *	0	n	n	--	(1)	c   ?     s
1700 	 *	0	y	n	--	(2)	c   s     s
1701 	 *	0	n	y	--	(3)	c   s     s
1702 	 *	0	y	y	n	(4)	c   s     s
1703 	 *	0	y	y	y	(5)	c   s     s
1704 	 *	1	--	n	--	(6) c	c	c s
1705 	 *	1	--	y	--	(7) c	c   s	c s
1706 	 *
1707 	 *					(c=clear s=set)
1708 	 */
1709 	switch (type & 0xff) {
1710 	case ND_NEIGHBOR_SOLICIT:
1711 		/*
1712 		 * New entry must have is_router flag cleared.
1713 		 */
1714 		if (is_newentry)	/* (6-7) */
1715 			ln->ln_router = 0;
1716 		break;
1717 	case ND_REDIRECT:
1718 		/*
1719 		 * If the icmp is a redirect to a better router, always set the
1720 		 * is_router flag.  Otherwise, if the entry is newly created,
1721 		 * clear the flag.  [RFC 2461, sec 8.3]
1722 		 */
1723 		if (code == ND_REDIRECT_ROUTER)
1724 			ln->ln_router = 1;
1725 		else if (is_newentry) /* (6-7) */
1726 			ln->ln_router = 0;
1727 		break;
1728 	case ND_ROUTER_SOLICIT:
1729 		/*
1730 		 * is_router flag must always be cleared.
1731 		 */
1732 		ln->ln_router = 0;
1733 		break;
1734 	case ND_ROUTER_ADVERT:
1735 		/*
1736 		 * Mark an entry with lladdr as a router.
1737 		 */
1738 		if ((!is_newentry && (olladdr || lladdr)) ||	/* (2-5) */
1739 		    (is_newentry && lladdr)) {			/* (7) */
1740 			ln->ln_router = 1;
1741 		}
1742 		break;
1743 	}
1744 
1745 	if (ln != NULL) {
1746 		static_route = (ln->la_flags & LLE_STATIC);
1747 		router = ln->ln_router;
1748 
1749 		if (flags & ND6_EXCLUSIVE)
1750 			LLE_WUNLOCK(ln);
1751 		else
1752 			LLE_RUNLOCK(ln);
1753 		if (static_route)
1754 			ln = NULL;
1755 	}
1756 	if (chain != NULL)
1757 		nd6_flush_holdchain(ifp, ifp, chain, &sin6);
1758 
1759 	/*
1760 	 * When the link-layer address of a router changes, select the
1761 	 * best router again.  In particular, when the neighbor entry is newly
1762 	 * created, it might affect the selection policy.
1763 	 * Question: can we restrict the first condition to the "is_newentry"
1764 	 * case?
1765 	 * XXX: when we hear an RA from a new router with the link-layer
1766 	 * address option, defrouter_select() is called twice, since
1767 	 * defrtrlist_update called the function as well.  However, I believe
1768 	 * we can compromise the overhead, since it only happens the first
1769 	 * time.
1770 	 * XXX: although defrouter_select() should not have a bad effect
1771 	 * for those are not autoconfigured hosts, we explicitly avoid such
1772 	 * cases for safety.
1773 	 */
1774 	if (do_update && router &&
1775 	    ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) {
1776 		/*
1777 		 * guaranteed recursion
1778 		 */
1779 		defrouter_select();
1780 	}
1781 
1782 	return (ln);
1783 done:
1784 	if (ln != NULL) {
1785 		if (flags & ND6_EXCLUSIVE)
1786 			LLE_WUNLOCK(ln);
1787 		else
1788 			LLE_RUNLOCK(ln);
1789 		if (static_route)
1790 			ln = NULL;
1791 	}
1792 	return (ln);
1793 }
1794 
1795 static void
1796 nd6_slowtimo(void *arg)
1797 {
1798 	CURVNET_SET((struct vnet *) arg);
1799 	struct nd_ifinfo *nd6if;
1800 	struct ifnet *ifp;
1801 
1802 	callout_reset(&V_nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
1803 	    nd6_slowtimo, curvnet);
1804 	IFNET_RLOCK_NOSLEEP();
1805 	TAILQ_FOREACH(ifp, &V_ifnet, if_link) {
1806 		if (ifp->if_afdata[AF_INET6] == NULL)
1807 			continue;
1808 		nd6if = ND_IFINFO(ifp);
1809 		if (nd6if->basereachable && /* already initialized */
1810 		    (nd6if->recalctm -= ND6_SLOWTIMER_INTERVAL) <= 0) {
1811 			/*
1812 			 * Since reachable time rarely changes by router
1813 			 * advertisements, we SHOULD insure that a new random
1814 			 * value gets recomputed at least once every few hours.
1815 			 * (RFC 2461, 6.3.4)
1816 			 */
1817 			nd6if->recalctm = V_nd6_recalc_reachtm_interval;
1818 			nd6if->reachable = ND_COMPUTE_RTIME(nd6if->basereachable);
1819 		}
1820 	}
1821 	IFNET_RUNLOCK_NOSLEEP();
1822 	CURVNET_RESTORE();
1823 }
1824 
1825 void
1826 nd6_grab_holdchain(struct llentry *ln, struct mbuf **chain,
1827     struct sockaddr_in6 *sin6)
1828 {
1829 
1830 	LLE_WLOCK_ASSERT(ln);
1831 
1832 	*chain = ln->la_hold;
1833 	ln->la_hold = NULL;
1834 	memcpy(sin6, L3_ADDR_SIN6(ln), sizeof(*sin6));
1835 
1836 	if (ln->ln_state == ND6_LLINFO_STALE) {
1837 
1838 		/*
1839 		 * The first time we send a packet to a
1840 		 * neighbor whose entry is STALE, we have
1841 		 * to change the state to DELAY and a sets
1842 		 * a timer to expire in DELAY_FIRST_PROBE_TIME
1843 		 * seconds to ensure do neighbor unreachability
1844 		 * detection on expiration.
1845 		 * (RFC 2461 7.3.3)
1846 		 */
1847 		ln->la_asked = 0;
1848 		ln->ln_state = ND6_LLINFO_DELAY;
1849 		nd6_llinfo_settimer_locked(ln, (long)V_nd6_delay * hz);
1850 	}
1851 }
1852 
1853 static int
1854 nd6_output_ifp(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *m,
1855     struct sockaddr_in6 *dst)
1856 {
1857 	int error;
1858 	int ip6len;
1859 	struct ip6_hdr *ip6;
1860 	struct m_tag *mtag;
1861 
1862 #ifdef MAC
1863 	mac_netinet6_nd6_send(ifp, m);
1864 #endif
1865 
1866 	/*
1867 	 * If called from nd6_ns_output() (NS), nd6_na_output() (NA),
1868 	 * icmp6_redirect_output() (REDIRECT) or from rip6_output() (RS, RA
1869 	 * as handled by rtsol and rtadvd), mbufs will be tagged for SeND
1870 	 * to be diverted to user space.  When re-injected into the kernel,
1871 	 * send_output() will directly dispatch them to the outgoing interface.
1872 	 */
1873 	if (send_sendso_input_hook != NULL) {
1874 		mtag = m_tag_find(m, PACKET_TAG_ND_OUTGOING, NULL);
1875 		if (mtag != NULL) {
1876 			ip6 = mtod(m, struct ip6_hdr *);
1877 			ip6len = sizeof(struct ip6_hdr) + ntohs(ip6->ip6_plen);
1878 			/* Use the SEND socket */
1879 			error = send_sendso_input_hook(m, ifp, SND_OUT,
1880 			    ip6len);
1881 			/* -1 == no app on SEND socket */
1882 			if (error == 0 || error != -1)
1883 			    return (error);
1884 		}
1885 	}
1886 
1887 	m_clrprotoflags(m);	/* Avoid confusing lower layers. */
1888 	IP_PROBE(send, NULL, NULL, mtod(m, struct ip6_hdr *), ifp, NULL,
1889 	    mtod(m, struct ip6_hdr *));
1890 
1891 	if ((ifp->if_flags & IFF_LOOPBACK) == 0)
1892 		origifp = ifp;
1893 
1894 	error = (*ifp->if_output)(origifp, m, (struct sockaddr *)dst, NULL);
1895 	return (error);
1896 }
1897 
1898 /*
1899  * IPv6 packet output - light version.
1900  * Checks if destination LLE exists and is in proper state
1901  * (e.g no modification required). If not true, fall back to
1902  * "heavy" version.
1903  */
1904 int
1905 nd6_output(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *m,
1906     struct sockaddr_in6 *dst, struct rtentry *rt0)
1907 {
1908 	struct llentry *ln = NULL;
1909 
1910 	/* discard the packet if IPv6 operation is disabled on the interface */
1911 	if ((ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED)) {
1912 		m_freem(m);
1913 		return (ENETDOWN); /* better error? */
1914 	}
1915 
1916 	if (IN6_IS_ADDR_MULTICAST(&dst->sin6_addr))
1917 		goto sendpkt;
1918 
1919 	if (nd6_need_cache(ifp) == 0)
1920 		goto sendpkt;
1921 
1922 	IF_AFDATA_RLOCK(ifp);
1923 	ln = nd6_lookup(&dst->sin6_addr, 0, ifp);
1924 	IF_AFDATA_RUNLOCK(ifp);
1925 
1926 	/*
1927 	 * Perform fast path for the following cases:
1928 	 * 1) lle state is REACHABLE
1929 	 * 2) lle state is DELAY (NS message sentNS message sent)
1930 	 *
1931 	 * Every other case involves lle modification, so we handle
1932 	 * them separately.
1933 	 */
1934 	if (ln == NULL || (ln->ln_state != ND6_LLINFO_REACHABLE &&
1935 	    ln->ln_state != ND6_LLINFO_DELAY)) {
1936 		/* Fall back to slow processing path */
1937 		if (ln != NULL)
1938 			LLE_RUNLOCK(ln);
1939 		return (nd6_output_lle(ifp, origifp, m, dst));
1940 	}
1941 
1942 sendpkt:
1943 	if (ln != NULL)
1944 		LLE_RUNLOCK(ln);
1945 
1946 	return (nd6_output_ifp(ifp, origifp, m, dst));
1947 }
1948 
1949 
1950 /*
1951  * Output IPv6 packet - heavy version.
1952  * Function assume that either
1953  * 1) destination LLE does not exist, is invalid or stale, so
1954  *   ND6_EXCLUSIVE lock needs to be acquired
1955  * 2) destination lle is provided (with ND6_EXCLUSIVE lock),
1956  *   in that case packets are queued in &chain.
1957  *
1958  */
1959 static int
1960 nd6_output_lle(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *m,
1961     struct sockaddr_in6 *dst)
1962 {
1963 	struct llentry *lle = NULL;
1964 	int flags = 0;
1965 
1966 	KASSERT(m != NULL, ("NULL mbuf, nothing to send"));
1967 	/* discard the packet if IPv6 operation is disabled on the interface */
1968 	if ((ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED)) {
1969 		m_freem(m);
1970 		return (ENETDOWN); /* better error? */
1971 	}
1972 
1973 	if (IN6_IS_ADDR_MULTICAST(&dst->sin6_addr))
1974 		goto sendpkt;
1975 
1976 	if (nd6_need_cache(ifp) == 0)
1977 		goto sendpkt;
1978 
1979 	/*
1980 	 * Address resolution or Neighbor Unreachability Detection
1981 	 * for the next hop.
1982 	 * At this point, the destination of the packet must be a unicast
1983 	 * or an anycast address(i.e. not a multicast).
1984 	 */
1985 	if (lle == NULL) {
1986 		IF_AFDATA_RLOCK(ifp);
1987 		lle = nd6_lookup(&dst->sin6_addr, ND6_EXCLUSIVE, ifp);
1988 		IF_AFDATA_RUNLOCK(ifp);
1989 		if ((lle == NULL) && nd6_is_addr_neighbor(dst, ifp))  {
1990 			/*
1991 			 * Since nd6_is_addr_neighbor() internally calls nd6_lookup(),
1992 			 * the condition below is not very efficient.  But we believe
1993 			 * it is tolerable, because this should be a rare case.
1994 			 */
1995 			flags = ND6_CREATE | ND6_EXCLUSIVE;
1996 			IF_AFDATA_LOCK(ifp);
1997 			lle = nd6_lookup(&dst->sin6_addr, flags, ifp);
1998 			IF_AFDATA_UNLOCK(ifp);
1999 		}
2000 	}
2001 	if (lle == NULL) {
2002 		if ((ifp->if_flags & IFF_POINTOPOINT) == 0 &&
2003 		    !(ND_IFINFO(ifp)->flags & ND6_IFF_PERFORMNUD)) {
2004 			char ip6buf[INET6_ADDRSTRLEN];
2005 			log(LOG_DEBUG,
2006 			    "nd6_output: can't allocate llinfo for %s "
2007 			    "(ln=%p)\n",
2008 			    ip6_sprintf(ip6buf, &dst->sin6_addr), lle);
2009 			m_freem(m);
2010 			return (ENOBUFS);
2011 		}
2012 		goto sendpkt;	/* send anyway */
2013 	}
2014 
2015 	LLE_WLOCK_ASSERT(lle);
2016 
2017 	/* We don't have to do link-layer address resolution on a p2p link. */
2018 	if ((ifp->if_flags & IFF_POINTOPOINT) != 0 &&
2019 	    lle->ln_state < ND6_LLINFO_REACHABLE) {
2020 		lle->ln_state = ND6_LLINFO_STALE;
2021 		nd6_llinfo_settimer_locked(lle, (long)V_nd6_gctimer * hz);
2022 	}
2023 
2024 	/*
2025 	 * The first time we send a packet to a neighbor whose entry is
2026 	 * STALE, we have to change the state to DELAY and a sets a timer to
2027 	 * expire in DELAY_FIRST_PROBE_TIME seconds to ensure do
2028 	 * neighbor unreachability detection on expiration.
2029 	 * (RFC 2461 7.3.3)
2030 	 */
2031 	if (lle->ln_state == ND6_LLINFO_STALE) {
2032 		lle->la_asked = 0;
2033 		lle->ln_state = ND6_LLINFO_DELAY;
2034 		nd6_llinfo_settimer_locked(lle, (long)V_nd6_delay * hz);
2035 	}
2036 
2037 	/*
2038 	 * If the neighbor cache entry has a state other than INCOMPLETE
2039 	 * (i.e. its link-layer address is already resolved), just
2040 	 * send the packet.
2041 	 */
2042 	if (lle->ln_state > ND6_LLINFO_INCOMPLETE)
2043 		goto sendpkt;
2044 
2045 	/*
2046 	 * There is a neighbor cache entry, but no ethernet address
2047 	 * response yet.  Append this latest packet to the end of the
2048 	 * packet queue in the mbuf, unless the number of the packet
2049 	 * does not exceed nd6_maxqueuelen.  When it exceeds nd6_maxqueuelen,
2050 	 * the oldest packet in the queue will be removed.
2051 	 */
2052 	if (lle->ln_state == ND6_LLINFO_NOSTATE)
2053 		lle->ln_state = ND6_LLINFO_INCOMPLETE;
2054 
2055 	if (lle->la_hold != NULL) {
2056 		struct mbuf *m_hold;
2057 		int i;
2058 
2059 		i = 0;
2060 		for (m_hold = lle->la_hold; m_hold; m_hold = m_hold->m_nextpkt){
2061 			i++;
2062 			if (m_hold->m_nextpkt == NULL) {
2063 				m_hold->m_nextpkt = m;
2064 				break;
2065 			}
2066 		}
2067 		while (i >= V_nd6_maxqueuelen) {
2068 			m_hold = lle->la_hold;
2069 			lle->la_hold = lle->la_hold->m_nextpkt;
2070 			m_freem(m_hold);
2071 			i--;
2072 		}
2073 	} else {
2074 		lle->la_hold = m;
2075 	}
2076 
2077 	/*
2078 	 * If there has been no NS for the neighbor after entering the
2079 	 * INCOMPLETE state, send the first solicitation.
2080 	 */
2081 	if (!ND6_LLINFO_PERMANENT(lle) && lle->la_asked == 0) {
2082 		lle->la_asked++;
2083 
2084 		nd6_llinfo_settimer_locked(lle,
2085 		    (long)ND_IFINFO(ifp)->retrans * hz / 1000);
2086 		LLE_WUNLOCK(lle);
2087 		nd6_ns_output(ifp, NULL, &dst->sin6_addr, lle, NULL);
2088 	} else {
2089 		/* We did the lookup so we need to do the unlock here. */
2090 		LLE_WUNLOCK(lle);
2091 	}
2092 
2093 	return (0);
2094 
2095   sendpkt:
2096 	if (lle != NULL)
2097 		LLE_WUNLOCK(lle);
2098 
2099 	return (nd6_output_ifp(ifp, origifp, m, dst));
2100 }
2101 
2102 
2103 int
2104 nd6_flush_holdchain(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *chain,
2105     struct sockaddr_in6 *dst)
2106 {
2107 	struct mbuf *m, *m_head;
2108 	struct ifnet *outifp;
2109 	int error = 0;
2110 
2111 	m_head = chain;
2112 	if ((ifp->if_flags & IFF_LOOPBACK) != 0)
2113 		outifp = origifp;
2114 	else
2115 		outifp = ifp;
2116 
2117 	while (m_head) {
2118 		m = m_head;
2119 		m_head = m_head->m_nextpkt;
2120 		error = nd6_output_ifp(ifp, origifp, m, dst);
2121 	}
2122 
2123 	/*
2124 	 * XXX
2125 	 * note that intermediate errors are blindly ignored - but this is
2126 	 * the same convention as used with nd6_output when called by
2127 	 * nd6_cache_lladdr
2128 	 */
2129 	return (error);
2130 }
2131 
2132 
2133 int
2134 nd6_need_cache(struct ifnet *ifp)
2135 {
2136 	/*
2137 	 * XXX: we currently do not make neighbor cache on any interface
2138 	 * other than ARCnet, Ethernet, FDDI and GIF.
2139 	 *
2140 	 * RFC2893 says:
2141 	 * - unidirectional tunnels needs no ND
2142 	 */
2143 	switch (ifp->if_type) {
2144 	case IFT_ARCNET:
2145 	case IFT_ETHER:
2146 	case IFT_FDDI:
2147 	case IFT_IEEE1394:
2148 	case IFT_L2VLAN:
2149 	case IFT_IEEE80211:
2150 	case IFT_INFINIBAND:
2151 	case IFT_BRIDGE:
2152 	case IFT_PROPVIRTUAL:
2153 		return (1);
2154 	default:
2155 		return (0);
2156 	}
2157 }
2158 
2159 /*
2160  * Add pernament ND6 link-layer record for given
2161  * interface address.
2162  *
2163  * Very similar to IPv4 arp_ifinit(), but:
2164  * 1) IPv6 DAD is performed in different place
2165  * 2) It is called by IPv6 protocol stack in contrast to
2166  * arp_ifinit() which is typically called in SIOCSIFADDR
2167  * driver ioctl handler.
2168  *
2169  */
2170 int
2171 nd6_add_ifa_lle(struct in6_ifaddr *ia)
2172 {
2173 	struct ifnet *ifp;
2174 	struct llentry *ln;
2175 
2176 	ifp = ia->ia_ifa.ifa_ifp;
2177 	if (nd6_need_cache(ifp) == 0)
2178 		return (0);
2179 	IF_AFDATA_LOCK(ifp);
2180 	ia->ia_ifa.ifa_rtrequest = nd6_rtrequest;
2181 	ln = lla_lookup(LLTABLE6(ifp), (LLE_CREATE | LLE_IFADDR |
2182 	    LLE_EXCLUSIVE), (struct sockaddr *)&ia->ia_addr);
2183 	IF_AFDATA_UNLOCK(ifp);
2184 	if (ln != NULL) {
2185 		ln->la_expire = 0;  /* for IPv6 this means permanent */
2186 		ln->ln_state = ND6_LLINFO_REACHABLE;
2187 		LLE_WUNLOCK(ln);
2188 		return (0);
2189 	}
2190 
2191 	return (ENOBUFS);
2192 }
2193 
2194 /*
2195  * Removes ALL lle records for interface address prefix.
2196  * XXXME: That's probably not we really want to do, we need
2197  * to remove address record only and keep other records
2198  * until we determine if given prefix is really going
2199  * to be removed.
2200  */
2201 void
2202 nd6_rem_ifa_lle(struct in6_ifaddr *ia)
2203 {
2204 	struct sockaddr_in6 mask, addr;
2205 	struct ifnet *ifp;
2206 
2207 	ifp = ia->ia_ifa.ifa_ifp;
2208 	memcpy(&addr, &ia->ia_addr, sizeof(ia->ia_addr));
2209 	memcpy(&mask, &ia->ia_prefixmask, sizeof(ia->ia_prefixmask));
2210 	lltable_prefix_free(AF_INET6, (struct sockaddr *)&addr,
2211 	            (struct sockaddr *)&mask, LLE_STATIC);
2212 }
2213 
2214 /*
2215  * the callers of this function need to be re-worked to drop
2216  * the lle lock, drop here for now
2217  */
2218 int
2219 nd6_storelladdr(struct ifnet *ifp, struct mbuf *m,
2220     const struct sockaddr *dst, u_char *desten, uint32_t *pflags)
2221 {
2222 	struct llentry *ln;
2223 
2224 	if (pflags != NULL)
2225 		*pflags = 0;
2226 	IF_AFDATA_UNLOCK_ASSERT(ifp);
2227 	if (m != NULL && m->m_flags & M_MCAST) {
2228 		switch (ifp->if_type) {
2229 		case IFT_ETHER:
2230 		case IFT_FDDI:
2231 		case IFT_L2VLAN:
2232 		case IFT_IEEE80211:
2233 		case IFT_BRIDGE:
2234 		case IFT_ISO88025:
2235 			ETHER_MAP_IPV6_MULTICAST(&SIN6(dst)->sin6_addr,
2236 						 desten);
2237 			return (0);
2238 		default:
2239 			m_freem(m);
2240 			return (EAFNOSUPPORT);
2241 		}
2242 	}
2243 
2244 
2245 	/*
2246 	 * the entry should have been created in nd6_store_lladdr
2247 	 */
2248 	IF_AFDATA_RLOCK(ifp);
2249 	ln = lla_lookup(LLTABLE6(ifp), 0, dst);
2250 	IF_AFDATA_RUNLOCK(ifp);
2251 	if ((ln == NULL) || !(ln->la_flags & LLE_VALID)) {
2252 		if (ln != NULL)
2253 			LLE_RUNLOCK(ln);
2254 		/* this could happen, if we could not allocate memory */
2255 		m_freem(m);
2256 		return (1);
2257 	}
2258 
2259 	bcopy(&ln->ll_addr, desten, ifp->if_addrlen);
2260 	if (pflags != NULL)
2261 		*pflags = ln->la_flags;
2262 	LLE_RUNLOCK(ln);
2263 	/*
2264 	 * A *small* use after free race exists here
2265 	 */
2266 	return (0);
2267 }
2268 
2269 static void
2270 clear_llinfo_pqueue(struct llentry *ln)
2271 {
2272 	struct mbuf *m_hold, *m_hold_next;
2273 
2274 	for (m_hold = ln->la_hold; m_hold; m_hold = m_hold_next) {
2275 		m_hold_next = m_hold->m_nextpkt;
2276 		m_freem(m_hold);
2277 	}
2278 
2279 	ln->la_hold = NULL;
2280 	return;
2281 }
2282 
2283 static int nd6_sysctl_drlist(SYSCTL_HANDLER_ARGS);
2284 static int nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS);
2285 #ifdef SYSCTL_DECL
2286 SYSCTL_DECL(_net_inet6_icmp6);
2287 #endif
2288 SYSCTL_NODE(_net_inet6_icmp6, ICMPV6CTL_ND6_DRLIST, nd6_drlist,
2289 	CTLFLAG_RD, nd6_sysctl_drlist, "");
2290 SYSCTL_NODE(_net_inet6_icmp6, ICMPV6CTL_ND6_PRLIST, nd6_prlist,
2291 	CTLFLAG_RD, nd6_sysctl_prlist, "");
2292 SYSCTL_INT(_net_inet6_icmp6, ICMPV6CTL_ND6_MAXQLEN, nd6_maxqueuelen,
2293 	CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(nd6_maxqueuelen), 1, "");
2294 SYSCTL_INT(_net_inet6_icmp6, OID_AUTO, nd6_gctimer,
2295 	CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(nd6_gctimer), (60 * 60 * 24), "");
2296 
2297 static int
2298 nd6_sysctl_drlist(SYSCTL_HANDLER_ARGS)
2299 {
2300 	struct in6_defrouter d;
2301 	struct nd_defrouter *dr;
2302 	int error;
2303 
2304 	if (req->newptr)
2305 		return (EPERM);
2306 
2307 	bzero(&d, sizeof(d));
2308 	d.rtaddr.sin6_family = AF_INET6;
2309 	d.rtaddr.sin6_len = sizeof(d.rtaddr);
2310 
2311 	/*
2312 	 * XXX locking
2313 	 */
2314 	TAILQ_FOREACH(dr, &V_nd_defrouter, dr_entry) {
2315 		d.rtaddr.sin6_addr = dr->rtaddr;
2316 		error = sa6_recoverscope(&d.rtaddr);
2317 		if (error != 0)
2318 			return (error);
2319 		d.flags = dr->flags;
2320 		d.rtlifetime = dr->rtlifetime;
2321 		d.expire = dr->expire + (time_second - time_uptime);
2322 		d.if_index = dr->ifp->if_index;
2323 		error = SYSCTL_OUT(req, &d, sizeof(d));
2324 		if (error != 0)
2325 			return (error);
2326 	}
2327 	return (0);
2328 }
2329 
2330 static int
2331 nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS)
2332 {
2333 	struct in6_prefix p;
2334 	struct sockaddr_in6 s6;
2335 	struct nd_prefix *pr;
2336 	struct nd_pfxrouter *pfr;
2337 	time_t maxexpire;
2338 	int error;
2339 	char ip6buf[INET6_ADDRSTRLEN];
2340 
2341 	if (req->newptr)
2342 		return (EPERM);
2343 
2344 	bzero(&p, sizeof(p));
2345 	p.origin = PR_ORIG_RA;
2346 	bzero(&s6, sizeof(s6));
2347 	s6.sin6_family = AF_INET6;
2348 	s6.sin6_len = sizeof(s6);
2349 
2350 	/*
2351 	 * XXX locking
2352 	 */
2353 	LIST_FOREACH(pr, &V_nd_prefix, ndpr_entry) {
2354 		p.prefix = pr->ndpr_prefix;
2355 		if (sa6_recoverscope(&p.prefix)) {
2356 			log(LOG_ERR, "scope error in prefix list (%s)\n",
2357 			    ip6_sprintf(ip6buf, &p.prefix.sin6_addr));
2358 			/* XXX: press on... */
2359 		}
2360 		p.raflags = pr->ndpr_raf;
2361 		p.prefixlen = pr->ndpr_plen;
2362 		p.vltime = pr->ndpr_vltime;
2363 		p.pltime = pr->ndpr_pltime;
2364 		p.if_index = pr->ndpr_ifp->if_index;
2365 		if (pr->ndpr_vltime == ND6_INFINITE_LIFETIME)
2366 			p.expire = 0;
2367 		else {
2368 			/* XXX: we assume time_t is signed. */
2369 			maxexpire = (-1) &
2370 			    ~((time_t)1 << ((sizeof(maxexpire) * 8) - 1));
2371 			if (pr->ndpr_vltime < maxexpire - pr->ndpr_lastupdate)
2372 				p.expire = pr->ndpr_lastupdate +
2373 				    pr->ndpr_vltime +
2374 				    (time_second - time_uptime);
2375 			else
2376 				p.expire = maxexpire;
2377 		}
2378 		p.refcnt = pr->ndpr_refcnt;
2379 		p.flags = pr->ndpr_stateflags;
2380 		p.advrtrs = 0;
2381 		LIST_FOREACH(pfr, &pr->ndpr_advrtrs, pfr_entry)
2382 			p.advrtrs++;
2383 		error = SYSCTL_OUT(req, &p, sizeof(p));
2384 		if (error != 0)
2385 			return (error);
2386 		LIST_FOREACH(pfr, &pr->ndpr_advrtrs, pfr_entry) {
2387 			s6.sin6_addr = pfr->router->rtaddr;
2388 			if (sa6_recoverscope(&s6))
2389 				log(LOG_ERR,
2390 				    "scope error in prefix list (%s)\n",
2391 				    ip6_sprintf(ip6buf, &pfr->router->rtaddr));
2392 			error = SYSCTL_OUT(req, &s6, sizeof(s6));
2393 			if (error != 0)
2394 				return (error);
2395 		}
2396 	}
2397 	return (0);
2398 }
2399