xref: /freebsd/sys/netinet6/nd6.c (revision 55620f43deef5c0eb5b4b0f675de18b30c8d1c2d)
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 #include <netinet/if_ether.h>
70 #include <netinet6/in6_var.h>
71 #include <netinet/ip6.h>
72 #include <netinet6/ip6_var.h>
73 #include <netinet6/scope6_var.h>
74 #include <netinet6/nd6.h>
75 #include <netinet6/in6_ifattach.h>
76 #include <netinet/icmp6.h>
77 #include <netinet6/send.h>
78 
79 #include <sys/limits.h>
80 
81 #include <security/mac/mac_framework.h>
82 
83 #define ND6_SLOWTIMER_INTERVAL (60 * 60) /* 1 hour */
84 #define ND6_RECALC_REACHTM_INTERVAL (60 * 120) /* 2 hours */
85 
86 #define SIN6(s) ((const struct sockaddr_in6 *)(s))
87 
88 MALLOC_DEFINE(M_IP6NDP, "ip6ndp", "IPv6 Neighbor Discovery");
89 
90 /* timer values */
91 VNET_DEFINE(int, nd6_prune)	= 1;	/* walk list every 1 seconds */
92 VNET_DEFINE(int, nd6_delay)	= 5;	/* delay first probe time 5 second */
93 VNET_DEFINE(int, nd6_umaxtries)	= 3;	/* maximum unicast query */
94 VNET_DEFINE(int, nd6_mmaxtries)	= 3;	/* maximum multicast query */
95 VNET_DEFINE(int, nd6_useloopback) = 1;	/* use loopback interface for
96 					 * local traffic */
97 VNET_DEFINE(int, nd6_gctimer)	= (60 * 60 * 24); /* 1 day: garbage
98 					 * collection timer */
99 
100 /* preventing too many loops in ND option parsing */
101 static VNET_DEFINE(int, nd6_maxndopt) = 10; /* max # of ND options allowed */
102 
103 VNET_DEFINE(int, nd6_maxnudhint) = 0;	/* max # of subsequent upper
104 					 * layer hints */
105 static VNET_DEFINE(int, nd6_maxqueuelen) = 1; /* max pkts cached in unresolved
106 					 * ND entries */
107 #define	V_nd6_maxndopt			VNET(nd6_maxndopt)
108 #define	V_nd6_maxqueuelen		VNET(nd6_maxqueuelen)
109 
110 #ifdef ND6_DEBUG
111 VNET_DEFINE(int, nd6_debug) = 1;
112 #else
113 VNET_DEFINE(int, nd6_debug) = 0;
114 #endif
115 
116 static eventhandler_tag lle_event_eh, iflladdr_event_eh;
117 
118 VNET_DEFINE(struct nd_drhead, nd_defrouter);
119 VNET_DEFINE(struct nd_prhead, nd_prefix);
120 VNET_DEFINE(struct rwlock, nd6_lock);
121 
122 VNET_DEFINE(int, nd6_recalc_reachtm_interval) = ND6_RECALC_REACHTM_INTERVAL;
123 #define	V_nd6_recalc_reachtm_interval	VNET(nd6_recalc_reachtm_interval)
124 
125 int	(*send_sendso_input_hook)(struct mbuf *, struct ifnet *, int, int);
126 
127 static int nd6_is_new_addr_neighbor(const struct sockaddr_in6 *,
128 	struct ifnet *);
129 static void nd6_setmtu0(struct ifnet *, struct nd_ifinfo *);
130 static void nd6_slowtimo(void *);
131 static int regen_tmpaddr(struct in6_ifaddr *);
132 static void nd6_free(struct llentry **, int);
133 static void nd6_free_redirect(const struct llentry *);
134 static void nd6_llinfo_timer(void *);
135 static void nd6_llinfo_settimer_locked(struct llentry *, long);
136 static void clear_llinfo_pqueue(struct llentry *);
137 static void nd6_rtrequest(int, struct rtentry *, struct rt_addrinfo *);
138 static int nd6_resolve_slow(struct ifnet *, int, struct mbuf *,
139     const struct sockaddr_in6 *, u_char *, uint32_t *, struct llentry **);
140 static int nd6_need_cache(struct ifnet *);
141 
142 
143 static VNET_DEFINE(struct callout, nd6_slowtimo_ch);
144 #define	V_nd6_slowtimo_ch		VNET(nd6_slowtimo_ch)
145 
146 VNET_DEFINE(struct callout, nd6_timer_ch);
147 #define	V_nd6_timer_ch			VNET(nd6_timer_ch)
148 
149 static void
150 nd6_lle_event(void *arg __unused, struct llentry *lle, int evt)
151 {
152 	struct rt_addrinfo rtinfo;
153 	struct sockaddr_in6 dst;
154 	struct sockaddr_dl gw;
155 	struct ifnet *ifp;
156 	int type;
157 
158 	LLE_WLOCK_ASSERT(lle);
159 
160 	if (lltable_get_af(lle->lle_tbl) != AF_INET6)
161 		return;
162 
163 	switch (evt) {
164 	case LLENTRY_RESOLVED:
165 		type = RTM_ADD;
166 		KASSERT(lle->la_flags & LLE_VALID,
167 		    ("%s: %p resolved but not valid?", __func__, lle));
168 		break;
169 	case LLENTRY_EXPIRED:
170 		type = RTM_DELETE;
171 		break;
172 	default:
173 		return;
174 	}
175 
176 	ifp = lltable_get_ifp(lle->lle_tbl);
177 
178 	bzero(&dst, sizeof(dst));
179 	bzero(&gw, sizeof(gw));
180 	bzero(&rtinfo, sizeof(rtinfo));
181 	lltable_fill_sa_entry(lle, (struct sockaddr *)&dst);
182 	dst.sin6_scope_id = in6_getscopezone(ifp,
183 	    in6_addrscope(&dst.sin6_addr));
184 	gw.sdl_len = sizeof(struct sockaddr_dl);
185 	gw.sdl_family = AF_LINK;
186 	gw.sdl_alen = ifp->if_addrlen;
187 	gw.sdl_index = ifp->if_index;
188 	gw.sdl_type = ifp->if_type;
189 	if (evt == LLENTRY_RESOLVED)
190 		bcopy(lle->ll_addr, gw.sdl_data, ifp->if_addrlen);
191 	rtinfo.rti_info[RTAX_DST] = (struct sockaddr *)&dst;
192 	rtinfo.rti_info[RTAX_GATEWAY] = (struct sockaddr *)&gw;
193 	rtinfo.rti_addrs = RTA_DST | RTA_GATEWAY;
194 	rt_missmsg_fib(type, &rtinfo, RTF_HOST | RTF_LLDATA | (
195 	    type == RTM_ADD ? RTF_UP: 0), 0, RT_DEFAULT_FIB);
196 }
197 
198 /*
199  * A handler for interface link layer address change event.
200  */
201 static void
202 nd6_iflladdr(void *arg __unused, struct ifnet *ifp)
203 {
204 
205 	lltable_update_ifaddr(LLTABLE6(ifp));
206 }
207 
208 void
209 nd6_init(void)
210 {
211 
212 	rw_init(&V_nd6_lock, "nd6");
213 
214 	LIST_INIT(&V_nd_prefix);
215 
216 	/* initialization of the default router list */
217 	TAILQ_INIT(&V_nd_defrouter);
218 
219 	/* Start timers. */
220 	callout_init(&V_nd6_slowtimo_ch, 0);
221 	callout_reset(&V_nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
222 	    nd6_slowtimo, curvnet);
223 
224 	callout_init(&V_nd6_timer_ch, 0);
225 	callout_reset(&V_nd6_timer_ch, hz, nd6_timer, curvnet);
226 
227 	nd6_dad_init();
228 	if (IS_DEFAULT_VNET(curvnet)) {
229 		lle_event_eh = EVENTHANDLER_REGISTER(lle_event, nd6_lle_event,
230 		    NULL, EVENTHANDLER_PRI_ANY);
231 		iflladdr_event_eh = EVENTHANDLER_REGISTER(iflladdr_event,
232 		    nd6_iflladdr, NULL, EVENTHANDLER_PRI_ANY);
233 	}
234 }
235 
236 #ifdef VIMAGE
237 void
238 nd6_destroy()
239 {
240 
241 	callout_drain(&V_nd6_slowtimo_ch);
242 	callout_drain(&V_nd6_timer_ch);
243 	if (IS_DEFAULT_VNET(curvnet)) {
244 		EVENTHANDLER_DEREGISTER(lle_event, lle_event_eh);
245 		EVENTHANDLER_DEREGISTER(iflladdr_event, iflladdr_event_eh);
246 	}
247 	rw_destroy(&V_nd6_lock);
248 }
249 #endif
250 
251 struct nd_ifinfo *
252 nd6_ifattach(struct ifnet *ifp)
253 {
254 	struct nd_ifinfo *nd;
255 
256 	nd = malloc(sizeof(*nd), M_IP6NDP, M_WAITOK | M_ZERO);
257 	nd->initialized = 1;
258 
259 	nd->chlim = IPV6_DEFHLIM;
260 	nd->basereachable = REACHABLE_TIME;
261 	nd->reachable = ND_COMPUTE_RTIME(nd->basereachable);
262 	nd->retrans = RETRANS_TIMER;
263 
264 	nd->flags = ND6_IFF_PERFORMNUD;
265 
266 	/* A loopback interface always has ND6_IFF_AUTO_LINKLOCAL.
267 	 * XXXHRS: Clear ND6_IFF_AUTO_LINKLOCAL on an IFT_BRIDGE interface by
268 	 * default regardless of the V_ip6_auto_linklocal configuration to
269 	 * give a reasonable default behavior.
270 	 */
271 	if ((V_ip6_auto_linklocal && ifp->if_type != IFT_BRIDGE) ||
272 	    (ifp->if_flags & IFF_LOOPBACK))
273 		nd->flags |= ND6_IFF_AUTO_LINKLOCAL;
274 	/*
275 	 * A loopback interface does not need to accept RTADV.
276 	 * XXXHRS: Clear ND6_IFF_ACCEPT_RTADV on an IFT_BRIDGE interface by
277 	 * default regardless of the V_ip6_accept_rtadv configuration to
278 	 * prevent the interface from accepting RA messages arrived
279 	 * on one of the member interfaces with ND6_IFF_ACCEPT_RTADV.
280 	 */
281 	if (V_ip6_accept_rtadv &&
282 	    !(ifp->if_flags & IFF_LOOPBACK) &&
283 	    (ifp->if_type != IFT_BRIDGE))
284 			nd->flags |= ND6_IFF_ACCEPT_RTADV;
285 	if (V_ip6_no_radr && !(ifp->if_flags & IFF_LOOPBACK))
286 		nd->flags |= ND6_IFF_NO_RADR;
287 
288 	/* XXX: we cannot call nd6_setmtu since ifp is not fully initialized */
289 	nd6_setmtu0(ifp, nd);
290 
291 	return nd;
292 }
293 
294 void
295 nd6_ifdetach(struct nd_ifinfo *nd)
296 {
297 
298 	free(nd, M_IP6NDP);
299 }
300 
301 /*
302  * Reset ND level link MTU. This function is called when the physical MTU
303  * changes, which means we might have to adjust the ND level MTU.
304  */
305 void
306 nd6_setmtu(struct ifnet *ifp)
307 {
308 	if (ifp->if_afdata[AF_INET6] == NULL)
309 		return;
310 
311 	nd6_setmtu0(ifp, ND_IFINFO(ifp));
312 }
313 
314 /* XXX todo: do not maintain copy of ifp->if_mtu in ndi->maxmtu */
315 void
316 nd6_setmtu0(struct ifnet *ifp, struct nd_ifinfo *ndi)
317 {
318 	u_int32_t omaxmtu;
319 
320 	omaxmtu = ndi->maxmtu;
321 
322 	switch (ifp->if_type) {
323 	case IFT_ARCNET:
324 		ndi->maxmtu = MIN(ARC_PHDS_MAXMTU, ifp->if_mtu); /* RFC2497 */
325 		break;
326 	case IFT_FDDI:
327 		ndi->maxmtu = MIN(FDDIIPMTU, ifp->if_mtu); /* RFC2467 */
328 		break;
329 	case IFT_ISO88025:
330 		 ndi->maxmtu = MIN(ISO88025_MAX_MTU, ifp->if_mtu);
331 		 break;
332 	default:
333 		ndi->maxmtu = ifp->if_mtu;
334 		break;
335 	}
336 
337 	/*
338 	 * Decreasing the interface MTU under IPV6 minimum MTU may cause
339 	 * undesirable situation.  We thus notify the operator of the change
340 	 * explicitly.  The check for omaxmtu is necessary to restrict the
341 	 * log to the case of changing the MTU, not initializing it.
342 	 */
343 	if (omaxmtu >= IPV6_MMTU && ndi->maxmtu < IPV6_MMTU) {
344 		log(LOG_NOTICE, "nd6_setmtu0: "
345 		    "new link MTU on %s (%lu) is too small for IPv6\n",
346 		    if_name(ifp), (unsigned long)ndi->maxmtu);
347 	}
348 
349 	if (ndi->maxmtu > V_in6_maxmtu)
350 		in6_setmaxmtu(); /* check all interfaces just in case */
351 
352 }
353 
354 void
355 nd6_option_init(void *opt, int icmp6len, union nd_opts *ndopts)
356 {
357 
358 	bzero(ndopts, sizeof(*ndopts));
359 	ndopts->nd_opts_search = (struct nd_opt_hdr *)opt;
360 	ndopts->nd_opts_last
361 		= (struct nd_opt_hdr *)(((u_char *)opt) + icmp6len);
362 
363 	if (icmp6len == 0) {
364 		ndopts->nd_opts_done = 1;
365 		ndopts->nd_opts_search = NULL;
366 	}
367 }
368 
369 /*
370  * Take one ND option.
371  */
372 struct nd_opt_hdr *
373 nd6_option(union nd_opts *ndopts)
374 {
375 	struct nd_opt_hdr *nd_opt;
376 	int olen;
377 
378 	KASSERT(ndopts != NULL, ("%s: ndopts == NULL", __func__));
379 	KASSERT(ndopts->nd_opts_last != NULL, ("%s: uninitialized ndopts",
380 	    __func__));
381 	if (ndopts->nd_opts_search == NULL)
382 		return NULL;
383 	if (ndopts->nd_opts_done)
384 		return NULL;
385 
386 	nd_opt = ndopts->nd_opts_search;
387 
388 	/* make sure nd_opt_len is inside the buffer */
389 	if ((caddr_t)&nd_opt->nd_opt_len >= (caddr_t)ndopts->nd_opts_last) {
390 		bzero(ndopts, sizeof(*ndopts));
391 		return NULL;
392 	}
393 
394 	olen = nd_opt->nd_opt_len << 3;
395 	if (olen == 0) {
396 		/*
397 		 * Message validation requires that all included
398 		 * options have a length that is greater than zero.
399 		 */
400 		bzero(ndopts, sizeof(*ndopts));
401 		return NULL;
402 	}
403 
404 	ndopts->nd_opts_search = (struct nd_opt_hdr *)((caddr_t)nd_opt + olen);
405 	if (ndopts->nd_opts_search > ndopts->nd_opts_last) {
406 		/* option overruns the end of buffer, invalid */
407 		bzero(ndopts, sizeof(*ndopts));
408 		return NULL;
409 	} else if (ndopts->nd_opts_search == ndopts->nd_opts_last) {
410 		/* reached the end of options chain */
411 		ndopts->nd_opts_done = 1;
412 		ndopts->nd_opts_search = NULL;
413 	}
414 	return nd_opt;
415 }
416 
417 /*
418  * Parse multiple ND options.
419  * This function is much easier to use, for ND routines that do not need
420  * multiple options of the same type.
421  */
422 int
423 nd6_options(union nd_opts *ndopts)
424 {
425 	struct nd_opt_hdr *nd_opt;
426 	int i = 0;
427 
428 	KASSERT(ndopts != NULL, ("%s: ndopts == NULL", __func__));
429 	KASSERT(ndopts->nd_opts_last != NULL, ("%s: uninitialized ndopts",
430 	    __func__));
431 	if (ndopts->nd_opts_search == NULL)
432 		return 0;
433 
434 	while (1) {
435 		nd_opt = nd6_option(ndopts);
436 		if (nd_opt == NULL && ndopts->nd_opts_last == NULL) {
437 			/*
438 			 * Message validation requires that all included
439 			 * options have a length that is greater than zero.
440 			 */
441 			ICMP6STAT_INC(icp6s_nd_badopt);
442 			bzero(ndopts, sizeof(*ndopts));
443 			return -1;
444 		}
445 
446 		if (nd_opt == NULL)
447 			goto skip1;
448 
449 		switch (nd_opt->nd_opt_type) {
450 		case ND_OPT_SOURCE_LINKADDR:
451 		case ND_OPT_TARGET_LINKADDR:
452 		case ND_OPT_MTU:
453 		case ND_OPT_REDIRECTED_HEADER:
454 		case ND_OPT_NONCE:
455 			if (ndopts->nd_opt_array[nd_opt->nd_opt_type]) {
456 				nd6log((LOG_INFO,
457 				    "duplicated ND6 option found (type=%d)\n",
458 				    nd_opt->nd_opt_type));
459 				/* XXX bark? */
460 			} else {
461 				ndopts->nd_opt_array[nd_opt->nd_opt_type]
462 					= nd_opt;
463 			}
464 			break;
465 		case ND_OPT_PREFIX_INFORMATION:
466 			if (ndopts->nd_opt_array[nd_opt->nd_opt_type] == 0) {
467 				ndopts->nd_opt_array[nd_opt->nd_opt_type]
468 					= nd_opt;
469 			}
470 			ndopts->nd_opts_pi_end =
471 				(struct nd_opt_prefix_info *)nd_opt;
472 			break;
473 		/* What about ND_OPT_ROUTE_INFO? RFC 4191 */
474 		case ND_OPT_RDNSS:	/* RFC 6106 */
475 		case ND_OPT_DNSSL:	/* RFC 6106 */
476 			/*
477 			 * Silently ignore options we know and do not care about
478 			 * in the kernel.
479 			 */
480 			break;
481 		default:
482 			/*
483 			 * Unknown options must be silently ignored,
484 			 * to accommodate future extension to the protocol.
485 			 */
486 			nd6log((LOG_DEBUG,
487 			    "nd6_options: unsupported option %d - "
488 			    "option ignored\n", nd_opt->nd_opt_type));
489 		}
490 
491 skip1:
492 		i++;
493 		if (i > V_nd6_maxndopt) {
494 			ICMP6STAT_INC(icp6s_nd_toomanyopt);
495 			nd6log((LOG_INFO, "too many loop in nd opt\n"));
496 			break;
497 		}
498 
499 		if (ndopts->nd_opts_done)
500 			break;
501 	}
502 
503 	return 0;
504 }
505 
506 /*
507  * ND6 timer routine to handle ND6 entries
508  */
509 static void
510 nd6_llinfo_settimer_locked(struct llentry *ln, long tick)
511 {
512 	int canceled;
513 
514 	LLE_WLOCK_ASSERT(ln);
515 
516 	if (tick < 0) {
517 		ln->la_expire = 0;
518 		ln->ln_ntick = 0;
519 		canceled = callout_stop(&ln->lle_timer);
520 	} else {
521 		ln->la_expire = time_uptime + tick / hz;
522 		LLE_ADDREF(ln);
523 		if (tick > INT_MAX) {
524 			ln->ln_ntick = tick - INT_MAX;
525 			canceled = callout_reset(&ln->lle_timer, INT_MAX,
526 			    nd6_llinfo_timer, ln);
527 		} else {
528 			ln->ln_ntick = 0;
529 			canceled = callout_reset(&ln->lle_timer, tick,
530 			    nd6_llinfo_timer, ln);
531 		}
532 	}
533 	if (canceled > 0)
534 		LLE_REMREF(ln);
535 }
536 
537 /*
538  * Gets source address of the first packet in hold queue
539  * and stores it in @src.
540  * Returns pointer to @src (if hold queue is not empty) or NULL.
541  *
542  * Set noinline to be dtrace-friendly
543  */
544 static __noinline struct in6_addr *
545 nd6_llinfo_get_holdsrc(struct llentry *ln, struct in6_addr *src)
546 {
547 	struct ip6_hdr hdr;
548 	struct mbuf *m;
549 
550 	if (ln->la_hold == NULL)
551 		return (NULL);
552 
553 	/*
554 	 * assume every packet in la_hold has the same IP header
555 	 */
556 	m = ln->la_hold;
557 	if (sizeof(hdr) > m->m_len)
558 		return (NULL);
559 
560 	m_copydata(m, 0, sizeof(hdr), (caddr_t)&hdr);
561 	*src = hdr.ip6_src;
562 
563 	return (src);
564 }
565 
566 /*
567  * Checks if we need to switch from STALE state.
568  *
569  * RFC 4861 requires switching from STALE to DELAY state
570  * on first packet matching entry, waiting V_nd6_delay and
571  * transition to PROBE state (if upper layer confirmation was
572  * not received).
573  *
574  * This code performs a bit differently:
575  * On packet hit we don't change state (but desired state
576  * can be guessed by control plane). However, after V_nd6_delay
577  * seconds code will transition to PROBE state (so DELAY state
578  * is kinda skipped in most situations).
579  *
580  * Typically, V_nd6_gctimer is bigger than V_nd6_delay, so
581  * we perform the following upon entering STALE state:
582  *
583  * 1) Arm timer to run each V_nd6_delay seconds to make sure that
584  * if packet was transmitted at the start of given interval, we
585  * would be able to switch to PROBE state in V_nd6_delay seconds
586  * as user expects.
587  *
588  * 2) Reschedule timer until original V_nd6_gctimer expires keeping
589  * lle in STALE state (remaining timer value stored in lle_remtime).
590  *
591  * 3) Reschedule timer if packet was transmitted less that V_nd6_delay
592  * seconds ago.
593  *
594  * Returns non-zero value if the entry is still STALE (storing
595  * the next timer interval in @pdelay).
596  *
597  * Returns zero value if original timer expired or we need to switch to
598  * PROBE (store that in @do_switch variable).
599  */
600 static int
601 nd6_is_stale(struct llentry *lle, long *pdelay, int *do_switch)
602 {
603 	int nd_delay, nd_gctimer, r_skip_req;
604 	time_t lle_hittime;
605 	long delay;
606 
607 	*do_switch = 0;
608 	nd_gctimer = V_nd6_gctimer;
609 	nd_delay = V_nd6_delay;
610 
611 	LLE_REQ_LOCK(lle);
612 	r_skip_req = lle->r_skip_req;
613 	lle_hittime = lle->lle_hittime;
614 	LLE_REQ_UNLOCK(lle);
615 
616 	if (r_skip_req > 0) {
617 
618 		/*
619 		 * Nonzero r_skip_req value was set upon entering
620 		 * STALE state. Since value was not changed, no
621 		 * packets were passed using this lle. Ask for
622 		 * timer reschedule and keep STALE state.
623 		 */
624 		delay = (long)(MIN(nd_gctimer, nd_delay));
625 		delay *= hz;
626 		if (lle->lle_remtime > delay)
627 			lle->lle_remtime -= delay;
628 		else {
629 			delay = lle->lle_remtime;
630 			lle->lle_remtime = 0;
631 		}
632 
633 		if (delay == 0) {
634 
635 			/*
636 			 * The original ng6_gctime timeout ended,
637 			 * no more rescheduling.
638 			 */
639 			return (0);
640 		}
641 
642 		*pdelay = delay;
643 		return (1);
644 	}
645 
646 	/*
647 	 * Packet received. Verify timestamp
648 	 */
649 	delay = (long)(time_uptime - lle_hittime);
650 	if (delay < nd_delay) {
651 
652 		/*
653 		 * V_nd6_delay still not passed since the first
654 		 * hit in STALE state.
655 		 * Reshedule timer and return.
656 		 */
657 		*pdelay = (long)(nd_delay - delay) * hz;
658 		return (1);
659 	}
660 
661 	/* Request switching to probe */
662 	*do_switch = 1;
663 	return (0);
664 }
665 
666 
667 /*
668  * Switch @lle state to new state optionally arming timers.
669  *
670  * Set noinline to be dtrace-friendly
671  */
672 __noinline void
673 nd6_llinfo_setstate(struct llentry *lle, int newstate)
674 {
675 	struct ifnet *ifp;
676 	int nd_gctimer, nd_delay;
677 	long delay, remtime;
678 
679 	delay = 0;
680 	remtime = 0;
681 
682 	switch (newstate) {
683 	case ND6_LLINFO_INCOMPLETE:
684 		ifp = lle->lle_tbl->llt_ifp;
685 		delay = (long)ND_IFINFO(ifp)->retrans * hz / 1000;
686 		break;
687 	case ND6_LLINFO_REACHABLE:
688 		if (!ND6_LLINFO_PERMANENT(lle)) {
689 			ifp = lle->lle_tbl->llt_ifp;
690 			delay = (long)ND_IFINFO(ifp)->reachable * hz;
691 		}
692 		break;
693 	case ND6_LLINFO_STALE:
694 
695 		/*
696 		 * Notify fast path that we want to know if any packet
697 		 * is transmitted by setting r_skip_req.
698 		 */
699 		LLE_REQ_LOCK(lle);
700 		lle->r_skip_req = 1;
701 		LLE_REQ_UNLOCK(lle);
702 		nd_delay = V_nd6_delay;
703 		nd_gctimer = V_nd6_gctimer;
704 
705 		delay = (long)(MIN(nd_gctimer, nd_delay)) * hz;
706 		remtime = (long)nd_gctimer * hz - delay;
707 		break;
708 	case ND6_LLINFO_DELAY:
709 		lle->la_asked = 0;
710 		delay = (long)V_nd6_delay * hz;
711 		break;
712 	}
713 
714 	if (delay > 0)
715 		nd6_llinfo_settimer_locked(lle, delay);
716 
717 	lle->lle_remtime = remtime;
718 	lle->ln_state = newstate;
719 }
720 
721 /*
722  * Timer-dependent part of nd state machine.
723  *
724  * Set noinline to be dtrace-friendly
725  */
726 static __noinline void
727 nd6_llinfo_timer(void *arg)
728 {
729 	struct llentry *ln;
730 	struct in6_addr *dst, *pdst, *psrc, src;
731 	struct ifnet *ifp;
732 	struct nd_ifinfo *ndi;
733 	int do_switch, send_ns;
734 	long delay;
735 
736 	KASSERT(arg != NULL, ("%s: arg NULL", __func__));
737 	ln = (struct llentry *)arg;
738 	ifp = lltable_get_ifp(ln->lle_tbl);
739 	CURVNET_SET(ifp->if_vnet);
740 
741 	ND6_RLOCK();
742 	LLE_WLOCK(ln);
743 	if (callout_pending(&ln->lle_timer)) {
744 		/*
745 		 * Here we are a bit odd here in the treatment of
746 		 * active/pending. If the pending bit is set, it got
747 		 * rescheduled before I ran. The active
748 		 * bit we ignore, since if it was stopped
749 		 * in ll_tablefree() and was currently running
750 		 * it would have return 0 so the code would
751 		 * not have deleted it since the callout could
752 		 * not be stopped so we want to go through
753 		 * with the delete here now. If the callout
754 		 * was restarted, the pending bit will be back on and
755 		 * we just want to bail since the callout_reset would
756 		 * return 1 and our reference would have been removed
757 		 * by nd6_llinfo_settimer_locked above since canceled
758 		 * would have been 1.
759 		 */
760 		LLE_WUNLOCK(ln);
761 		ND6_RUNLOCK();
762 		CURVNET_RESTORE();
763 		return;
764 	}
765 	ndi = ND_IFINFO(ifp);
766 	send_ns = 0;
767 	dst = &ln->r_l3addr.addr6;
768 	pdst = dst;
769 
770 	if (ln->ln_ntick > 0) {
771 		if (ln->ln_ntick > INT_MAX) {
772 			ln->ln_ntick -= INT_MAX;
773 			nd6_llinfo_settimer_locked(ln, INT_MAX);
774 		} else {
775 			ln->ln_ntick = 0;
776 			nd6_llinfo_settimer_locked(ln, ln->ln_ntick);
777 		}
778 		goto done;
779 	}
780 
781 	if (ln->la_flags & LLE_STATIC) {
782 		goto done;
783 	}
784 
785 	if (ln->la_flags & LLE_DELETED) {
786 		nd6_free(&ln, 0);
787 		goto done;
788 	}
789 
790 	switch (ln->ln_state) {
791 	case ND6_LLINFO_INCOMPLETE:
792 		if (ln->la_asked < V_nd6_mmaxtries) {
793 			ln->la_asked++;
794 			send_ns = 1;
795 			/* Send NS to multicast address */
796 			pdst = NULL;
797 		} else {
798 			struct mbuf *m = ln->la_hold;
799 			if (m) {
800 				struct mbuf *m0;
801 
802 				/*
803 				 * assuming every packet in la_hold has the
804 				 * same IP header.  Send error after unlock.
805 				 */
806 				m0 = m->m_nextpkt;
807 				m->m_nextpkt = NULL;
808 				ln->la_hold = m0;
809 				clear_llinfo_pqueue(ln);
810 			}
811 			nd6_free(&ln, 0);
812 			if (m != NULL)
813 				icmp6_error2(m, ICMP6_DST_UNREACH,
814 				    ICMP6_DST_UNREACH_ADDR, 0, ifp);
815 		}
816 		break;
817 	case ND6_LLINFO_REACHABLE:
818 		if (!ND6_LLINFO_PERMANENT(ln))
819 			nd6_llinfo_setstate(ln, ND6_LLINFO_STALE);
820 		break;
821 
822 	case ND6_LLINFO_STALE:
823 		if (nd6_is_stale(ln, &delay, &do_switch) != 0) {
824 
825 			/*
826 			 * No packet has used this entry and GC timeout
827 			 * has not been passed. Reshedule timer and
828 			 * return.
829 			 */
830 			nd6_llinfo_settimer_locked(ln, delay);
831 			break;
832 		}
833 
834 		if (do_switch == 0) {
835 
836 			/*
837 			 * GC timer has ended and entry hasn't been used.
838 			 * Run Garbage collector (RFC 4861, 5.3)
839 			 */
840 			if (!ND6_LLINFO_PERMANENT(ln))
841 				nd6_free(&ln, 1);
842 			break;
843 		}
844 
845 		/* Entry has been used AND delay timer has ended. */
846 
847 		/* FALLTHROUGH */
848 
849 	case ND6_LLINFO_DELAY:
850 		if (ndi && (ndi->flags & ND6_IFF_PERFORMNUD) != 0) {
851 			/* We need NUD */
852 			ln->la_asked = 1;
853 			nd6_llinfo_setstate(ln, ND6_LLINFO_PROBE);
854 			send_ns = 1;
855 		} else
856 			nd6_llinfo_setstate(ln, ND6_LLINFO_STALE); /* XXX */
857 		break;
858 	case ND6_LLINFO_PROBE:
859 		if (ln->la_asked < V_nd6_umaxtries) {
860 			ln->la_asked++;
861 			send_ns = 1;
862 		} else {
863 			nd6_free(&ln, 0);
864 		}
865 		break;
866 	default:
867 		panic("%s: paths in a dark night can be confusing: %d",
868 		    __func__, ln->ln_state);
869 	}
870 done:
871 	if (ln != NULL)
872 		ND6_RUNLOCK();
873 	if (send_ns != 0) {
874 		nd6_llinfo_settimer_locked(ln, (long)ndi->retrans * hz / 1000);
875 		psrc = nd6_llinfo_get_holdsrc(ln, &src);
876 		LLE_FREE_LOCKED(ln);
877 		ln = NULL;
878 		nd6_ns_output(ifp, psrc, pdst, dst, NULL);
879 	}
880 
881 	if (ln != NULL)
882 		LLE_FREE_LOCKED(ln);
883 	CURVNET_RESTORE();
884 }
885 
886 
887 /*
888  * ND6 timer routine to expire default route list and prefix list
889  */
890 void
891 nd6_timer(void *arg)
892 {
893 	CURVNET_SET((struct vnet *) arg);
894 	struct nd_drhead drq;
895 	struct nd_defrouter *dr, *ndr;
896 	struct nd_prefix *pr, *npr;
897 	struct in6_ifaddr *ia6, *nia6;
898 
899 	TAILQ_INIT(&drq);
900 
901 	/* expire default router list */
902 	ND6_WLOCK();
903 	TAILQ_FOREACH_SAFE(dr, &V_nd_defrouter, dr_entry, ndr)
904 		if (dr->expire && dr->expire < time_uptime)
905 			defrouter_unlink(dr, &drq);
906 	ND6_WUNLOCK();
907 
908 	while ((dr = TAILQ_FIRST(&drq)) != NULL) {
909 		TAILQ_REMOVE(&drq, dr, dr_entry);
910 		defrouter_del(dr);
911 	}
912 
913 	/*
914 	 * expire interface addresses.
915 	 * in the past the loop was inside prefix expiry processing.
916 	 * However, from a stricter speci-confrmance standpoint, we should
917 	 * rather separate address lifetimes and prefix lifetimes.
918 	 *
919 	 * XXXRW: in6_ifaddrhead locking.
920 	 */
921   addrloop:
922 	TAILQ_FOREACH_SAFE(ia6, &V_in6_ifaddrhead, ia_link, nia6) {
923 		/* check address lifetime */
924 		if (IFA6_IS_INVALID(ia6)) {
925 			int regen = 0;
926 
927 			/*
928 			 * If the expiring address is temporary, try
929 			 * regenerating a new one.  This would be useful when
930 			 * we suspended a laptop PC, then turned it on after a
931 			 * period that could invalidate all temporary
932 			 * addresses.  Although we may have to restart the
933 			 * loop (see below), it must be after purging the
934 			 * address.  Otherwise, we'd see an infinite loop of
935 			 * regeneration.
936 			 */
937 			if (V_ip6_use_tempaddr &&
938 			    (ia6->ia6_flags & IN6_IFF_TEMPORARY) != 0) {
939 				if (regen_tmpaddr(ia6) == 0)
940 					regen = 1;
941 			}
942 
943 			in6_purgeaddr(&ia6->ia_ifa);
944 
945 			if (regen)
946 				goto addrloop; /* XXX: see below */
947 		} else if (IFA6_IS_DEPRECATED(ia6)) {
948 			int oldflags = ia6->ia6_flags;
949 
950 			ia6->ia6_flags |= IN6_IFF_DEPRECATED;
951 
952 			/*
953 			 * If a temporary address has just become deprecated,
954 			 * regenerate a new one if possible.
955 			 */
956 			if (V_ip6_use_tempaddr &&
957 			    (ia6->ia6_flags & IN6_IFF_TEMPORARY) != 0 &&
958 			    (oldflags & IN6_IFF_DEPRECATED) == 0) {
959 
960 				if (regen_tmpaddr(ia6) == 0) {
961 					/*
962 					 * A new temporary address is
963 					 * generated.
964 					 * XXX: this means the address chain
965 					 * has changed while we are still in
966 					 * the loop.  Although the change
967 					 * would not cause disaster (because
968 					 * it's not a deletion, but an
969 					 * addition,) we'd rather restart the
970 					 * loop just for safety.  Or does this
971 					 * significantly reduce performance??
972 					 */
973 					goto addrloop;
974 				}
975 			}
976 		} else if ((ia6->ia6_flags & IN6_IFF_TENTATIVE) != 0) {
977 			/*
978 			 * Schedule DAD for a tentative address.  This happens
979 			 * if the interface was down or not running
980 			 * when the address was configured.
981 			 */
982 			int delay;
983 
984 			delay = arc4random() %
985 			    (MAX_RTR_SOLICITATION_DELAY * hz);
986 			nd6_dad_start((struct ifaddr *)ia6, delay);
987 		} else {
988 			/*
989 			 * Check status of the interface.  If it is down,
990 			 * mark the address as tentative for future DAD.
991 			 */
992 			if ((ia6->ia_ifp->if_flags & IFF_UP) == 0 ||
993 			    (ia6->ia_ifp->if_drv_flags & IFF_DRV_RUNNING)
994 				== 0 ||
995 			    (ND_IFINFO(ia6->ia_ifp)->flags &
996 				ND6_IFF_IFDISABLED) != 0) {
997 				ia6->ia6_flags &= ~IN6_IFF_DUPLICATED;
998 				ia6->ia6_flags |= IN6_IFF_TENTATIVE;
999 			}
1000 			/*
1001 			 * A new RA might have made a deprecated address
1002 			 * preferred.
1003 			 */
1004 			ia6->ia6_flags &= ~IN6_IFF_DEPRECATED;
1005 		}
1006 	}
1007 
1008 	/* expire prefix list */
1009 	LIST_FOREACH_SAFE(pr, &V_nd_prefix, ndpr_entry, npr) {
1010 		/*
1011 		 * check prefix lifetime.
1012 		 * since pltime is just for autoconf, pltime processing for
1013 		 * prefix is not necessary.
1014 		 */
1015 		if (pr->ndpr_vltime != ND6_INFINITE_LIFETIME &&
1016 		    time_uptime - pr->ndpr_lastupdate > pr->ndpr_vltime) {
1017 
1018 			/*
1019 			 * address expiration and prefix expiration are
1020 			 * separate.  NEVER perform in6_purgeaddr here.
1021 			 */
1022 			prelist_remove(pr);
1023 		}
1024 	}
1025 
1026 	callout_reset(&V_nd6_timer_ch, V_nd6_prune * hz,
1027 	    nd6_timer, curvnet);
1028 
1029 	CURVNET_RESTORE();
1030 }
1031 
1032 /*
1033  * ia6 - deprecated/invalidated temporary address
1034  */
1035 static int
1036 regen_tmpaddr(struct in6_ifaddr *ia6)
1037 {
1038 	struct ifaddr *ifa;
1039 	struct ifnet *ifp;
1040 	struct in6_ifaddr *public_ifa6 = NULL;
1041 
1042 	ifp = ia6->ia_ifa.ifa_ifp;
1043 	IF_ADDR_RLOCK(ifp);
1044 	TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1045 		struct in6_ifaddr *it6;
1046 
1047 		if (ifa->ifa_addr->sa_family != AF_INET6)
1048 			continue;
1049 
1050 		it6 = (struct in6_ifaddr *)ifa;
1051 
1052 		/* ignore no autoconf addresses. */
1053 		if ((it6->ia6_flags & IN6_IFF_AUTOCONF) == 0)
1054 			continue;
1055 
1056 		/* ignore autoconf addresses with different prefixes. */
1057 		if (it6->ia6_ndpr == NULL || it6->ia6_ndpr != ia6->ia6_ndpr)
1058 			continue;
1059 
1060 		/*
1061 		 * Now we are looking at an autoconf address with the same
1062 		 * prefix as ours.  If the address is temporary and is still
1063 		 * preferred, do not create another one.  It would be rare, but
1064 		 * could happen, for example, when we resume a laptop PC after
1065 		 * a long period.
1066 		 */
1067 		if ((it6->ia6_flags & IN6_IFF_TEMPORARY) != 0 &&
1068 		    !IFA6_IS_DEPRECATED(it6)) {
1069 			public_ifa6 = NULL;
1070 			break;
1071 		}
1072 
1073 		/*
1074 		 * This is a public autoconf address that has the same prefix
1075 		 * as ours.  If it is preferred, keep it.  We can't break the
1076 		 * loop here, because there may be a still-preferred temporary
1077 		 * address with the prefix.
1078 		 */
1079 		if (!IFA6_IS_DEPRECATED(it6))
1080 			public_ifa6 = it6;
1081 	}
1082 	if (public_ifa6 != NULL)
1083 		ifa_ref(&public_ifa6->ia_ifa);
1084 	IF_ADDR_RUNLOCK(ifp);
1085 
1086 	if (public_ifa6 != NULL) {
1087 		int e;
1088 
1089 		if ((e = in6_tmpifadd(public_ifa6, 0, 0)) != 0) {
1090 			ifa_free(&public_ifa6->ia_ifa);
1091 			log(LOG_NOTICE, "regen_tmpaddr: failed to create a new"
1092 			    " tmp addr,errno=%d\n", e);
1093 			return (-1);
1094 		}
1095 		ifa_free(&public_ifa6->ia_ifa);
1096 		return (0);
1097 	}
1098 
1099 	return (-1);
1100 }
1101 
1102 /*
1103  * Remove prefix and default router list entries corresponding to ifp. Neighbor
1104  * cache entries are freed in in6_domifdetach().
1105  */
1106 void
1107 nd6_purge(struct ifnet *ifp)
1108 {
1109 	struct nd_drhead drq;
1110 	struct nd_defrouter *dr, *ndr;
1111 	struct nd_prefix *pr, *npr;
1112 
1113 	TAILQ_INIT(&drq);
1114 
1115 	/*
1116 	 * Nuke default router list entries toward ifp.
1117 	 * We defer removal of default router list entries that is installed
1118 	 * in the routing table, in order to keep additional side effects as
1119 	 * small as possible.
1120 	 */
1121 	ND6_WLOCK();
1122 	TAILQ_FOREACH_SAFE(dr, &V_nd_defrouter, dr_entry, ndr) {
1123 		if (dr->installed)
1124 			continue;
1125 		if (dr->ifp == ifp)
1126 			defrouter_unlink(dr, &drq);
1127 	}
1128 
1129 	TAILQ_FOREACH_SAFE(dr, &V_nd_defrouter, dr_entry, ndr) {
1130 		if (!dr->installed)
1131 			continue;
1132 		if (dr->ifp == ifp)
1133 			defrouter_unlink(dr, &drq);
1134 	}
1135 	ND6_WUNLOCK();
1136 
1137 	while ((dr = TAILQ_FIRST(&drq)) != NULL) {
1138 		TAILQ_REMOVE(&drq, dr, dr_entry);
1139 		defrouter_del(dr);
1140 	}
1141 
1142 	/* Nuke prefix list entries toward ifp */
1143 	LIST_FOREACH_SAFE(pr, &V_nd_prefix, ndpr_entry, npr) {
1144 		if (pr->ndpr_ifp == ifp) {
1145 			/*
1146 			 * Because if_detach() does *not* release prefixes
1147 			 * while purging addresses the reference count will
1148 			 * still be above zero. We therefore reset it to
1149 			 * make sure that the prefix really gets purged.
1150 			 */
1151 			pr->ndpr_refcnt = 0;
1152 
1153 			prelist_remove(pr);
1154 		}
1155 	}
1156 
1157 	/* cancel default outgoing interface setting */
1158 	if (V_nd6_defifindex == ifp->if_index)
1159 		nd6_setdefaultiface(0);
1160 
1161 	if (ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) {
1162 		/* Refresh default router list. */
1163 		defrouter_select();
1164 	}
1165 }
1166 
1167 /*
1168  * the caller acquires and releases the lock on the lltbls
1169  * Returns the llentry locked
1170  */
1171 struct llentry *
1172 nd6_lookup(const struct in6_addr *addr6, int flags, struct ifnet *ifp)
1173 {
1174 	struct sockaddr_in6 sin6;
1175 	struct llentry *ln;
1176 
1177 	bzero(&sin6, sizeof(sin6));
1178 	sin6.sin6_len = sizeof(struct sockaddr_in6);
1179 	sin6.sin6_family = AF_INET6;
1180 	sin6.sin6_addr = *addr6;
1181 
1182 	IF_AFDATA_LOCK_ASSERT(ifp);
1183 
1184 	ln = lla_lookup(LLTABLE6(ifp), flags, (struct sockaddr *)&sin6);
1185 
1186 	return (ln);
1187 }
1188 
1189 struct llentry *
1190 nd6_alloc(const struct in6_addr *addr6, int flags, struct ifnet *ifp)
1191 {
1192 	struct sockaddr_in6 sin6;
1193 	struct llentry *ln;
1194 
1195 	bzero(&sin6, sizeof(sin6));
1196 	sin6.sin6_len = sizeof(struct sockaddr_in6);
1197 	sin6.sin6_family = AF_INET6;
1198 	sin6.sin6_addr = *addr6;
1199 
1200 	ln = lltable_alloc_entry(LLTABLE6(ifp), 0, (struct sockaddr *)&sin6);
1201 	if (ln != NULL)
1202 		ln->ln_state = ND6_LLINFO_NOSTATE;
1203 
1204 	return (ln);
1205 }
1206 
1207 /*
1208  * Test whether a given IPv6 address is a neighbor or not, ignoring
1209  * the actual neighbor cache.  The neighbor cache is ignored in order
1210  * to not reenter the routing code from within itself.
1211  */
1212 static int
1213 nd6_is_new_addr_neighbor(const struct sockaddr_in6 *addr, struct ifnet *ifp)
1214 {
1215 	struct nd_prefix *pr;
1216 	struct ifaddr *dstaddr;
1217 	struct rt_addrinfo info;
1218 	struct sockaddr_in6 rt_key;
1219 	struct sockaddr *dst6;
1220 	int fibnum;
1221 
1222 	/*
1223 	 * A link-local address is always a neighbor.
1224 	 * XXX: a link does not necessarily specify a single interface.
1225 	 */
1226 	if (IN6_IS_ADDR_LINKLOCAL(&addr->sin6_addr)) {
1227 		struct sockaddr_in6 sin6_copy;
1228 		u_int32_t zone;
1229 
1230 		/*
1231 		 * We need sin6_copy since sa6_recoverscope() may modify the
1232 		 * content (XXX).
1233 		 */
1234 		sin6_copy = *addr;
1235 		if (sa6_recoverscope(&sin6_copy))
1236 			return (0); /* XXX: should be impossible */
1237 		if (in6_setscope(&sin6_copy.sin6_addr, ifp, &zone))
1238 			return (0);
1239 		if (sin6_copy.sin6_scope_id == zone)
1240 			return (1);
1241 		else
1242 			return (0);
1243 	}
1244 
1245 	bzero(&rt_key, sizeof(rt_key));
1246 	bzero(&info, sizeof(info));
1247 	info.rti_info[RTAX_DST] = (struct sockaddr *)&rt_key;
1248 
1249 	/* Always use the default FIB here. XXME - why? */
1250 	fibnum = RT_DEFAULT_FIB;
1251 
1252 	/*
1253 	 * If the address matches one of our addresses,
1254 	 * it should be a neighbor.
1255 	 * If the address matches one of our on-link prefixes, it should be a
1256 	 * neighbor.
1257 	 */
1258 	LIST_FOREACH(pr, &V_nd_prefix, ndpr_entry) {
1259 		if (pr->ndpr_ifp != ifp)
1260 			continue;
1261 
1262 		if (!(pr->ndpr_stateflags & NDPRF_ONLINK)) {
1263 
1264 			/* Always use the default FIB here. */
1265 			dst6 = (struct sockaddr *)&pr->ndpr_prefix;
1266 
1267 			/* Restore length field before retrying lookup */
1268 			rt_key.sin6_len = sizeof(rt_key);
1269 			if (rib_lookup_info(fibnum, dst6, 0, 0, &info) != 0)
1270 				continue;
1271 			/*
1272 			 * This is the case where multiple interfaces
1273 			 * have the same prefix, but only one is installed
1274 			 * into the routing table and that prefix entry
1275 			 * is not the one being examined here. In the case
1276 			 * where RADIX_MPATH is enabled, multiple route
1277 			 * entries (of the same rt_key value) will be
1278 			 * installed because the interface addresses all
1279 			 * differ.
1280 			 */
1281 			if (!IN6_ARE_ADDR_EQUAL(&pr->ndpr_prefix.sin6_addr,
1282 			       &rt_key.sin6_addr))
1283 				continue;
1284 		}
1285 
1286 		if (IN6_ARE_MASKED_ADDR_EQUAL(&pr->ndpr_prefix.sin6_addr,
1287 		    &addr->sin6_addr, &pr->ndpr_mask))
1288 			return (1);
1289 	}
1290 
1291 	/*
1292 	 * If the address is assigned on the node of the other side of
1293 	 * a p2p interface, the address should be a neighbor.
1294 	 */
1295 	dstaddr = ifa_ifwithdstaddr((const struct sockaddr *)addr, RT_ALL_FIBS);
1296 	if (dstaddr != NULL) {
1297 		if (dstaddr->ifa_ifp == ifp) {
1298 			ifa_free(dstaddr);
1299 			return (1);
1300 		}
1301 		ifa_free(dstaddr);
1302 	}
1303 
1304 	/*
1305 	 * If the default router list is empty, all addresses are regarded
1306 	 * as on-link, and thus, as a neighbor.
1307 	 */
1308 	if (ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV &&
1309 	    TAILQ_EMPTY(&V_nd_defrouter) &&
1310 	    V_nd6_defifindex == ifp->if_index) {
1311 		return (1);
1312 	}
1313 
1314 	return (0);
1315 }
1316 
1317 
1318 /*
1319  * Detect if a given IPv6 address identifies a neighbor on a given link.
1320  * XXX: should take care of the destination of a p2p link?
1321  */
1322 int
1323 nd6_is_addr_neighbor(const struct sockaddr_in6 *addr, struct ifnet *ifp)
1324 {
1325 	struct llentry *lle;
1326 	int rc = 0;
1327 
1328 	IF_AFDATA_UNLOCK_ASSERT(ifp);
1329 	if (nd6_is_new_addr_neighbor(addr, ifp))
1330 		return (1);
1331 
1332 	/*
1333 	 * Even if the address matches none of our addresses, it might be
1334 	 * in the neighbor cache.
1335 	 */
1336 	IF_AFDATA_RLOCK(ifp);
1337 	if ((lle = nd6_lookup(&addr->sin6_addr, 0, ifp)) != NULL) {
1338 		LLE_RUNLOCK(lle);
1339 		rc = 1;
1340 	}
1341 	IF_AFDATA_RUNLOCK(ifp);
1342 	return (rc);
1343 }
1344 
1345 /*
1346  * Free an nd6 llinfo entry.
1347  * Since the function would cause significant changes in the kernel, DO NOT
1348  * make it global, unless you have a strong reason for the change, and are sure
1349  * that the change is safe.
1350  *
1351  * Set noinline to be dtrace-friendly
1352  */
1353 static __noinline void
1354 nd6_free(struct llentry **lnp, int gc)
1355 {
1356 	struct ifnet *ifp;
1357 	struct llentry *ln;
1358 	struct nd_defrouter *dr;
1359 
1360 	ln = *lnp;
1361 	*lnp = NULL;
1362 
1363 	LLE_WLOCK_ASSERT(ln);
1364 	ND6_RLOCK_ASSERT();
1365 
1366 	ifp = lltable_get_ifp(ln->lle_tbl);
1367 	if ((ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) != 0)
1368 		dr = defrouter_lookup_locked(&ln->r_l3addr.addr6, ifp);
1369 	else
1370 		dr = NULL;
1371 	ND6_RUNLOCK();
1372 
1373 	if ((ln->la_flags & LLE_DELETED) == 0)
1374 		EVENTHANDLER_INVOKE(lle_event, ln, LLENTRY_EXPIRED);
1375 
1376 	/*
1377 	 * we used to have pfctlinput(PRC_HOSTDEAD) here.
1378 	 * even though it is not harmful, it was not really necessary.
1379 	 */
1380 
1381 	/* cancel timer */
1382 	nd6_llinfo_settimer_locked(ln, -1);
1383 
1384 	if (ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) {
1385 		if (dr != NULL && dr->expire &&
1386 		    ln->ln_state == ND6_LLINFO_STALE && gc) {
1387 			/*
1388 			 * If the reason for the deletion is just garbage
1389 			 * collection, and the neighbor is an active default
1390 			 * router, do not delete it.  Instead, reset the GC
1391 			 * timer using the router's lifetime.
1392 			 * Simply deleting the entry would affect default
1393 			 * router selection, which is not necessarily a good
1394 			 * thing, especially when we're using router preference
1395 			 * values.
1396 			 * XXX: the check for ln_state would be redundant,
1397 			 *      but we intentionally keep it just in case.
1398 			 */
1399 			if (dr->expire > time_uptime)
1400 				nd6_llinfo_settimer_locked(ln,
1401 				    (dr->expire - time_uptime) * hz);
1402 			else
1403 				nd6_llinfo_settimer_locked(ln,
1404 				    (long)V_nd6_gctimer * hz);
1405 
1406 			LLE_REMREF(ln);
1407 			LLE_WUNLOCK(ln);
1408 			defrouter_rele(dr);
1409 			return;
1410 		}
1411 
1412 		if (dr) {
1413 			/*
1414 			 * Unreachablity of a router might affect the default
1415 			 * router selection and on-link detection of advertised
1416 			 * prefixes.
1417 			 */
1418 
1419 			/*
1420 			 * Temporarily fake the state to choose a new default
1421 			 * router and to perform on-link determination of
1422 			 * prefixes correctly.
1423 			 * Below the state will be set correctly,
1424 			 * or the entry itself will be deleted.
1425 			 */
1426 			ln->ln_state = ND6_LLINFO_INCOMPLETE;
1427 		}
1428 
1429 		if (ln->ln_router || dr) {
1430 
1431 			/*
1432 			 * We need to unlock to avoid a LOR with rt6_flush() with the
1433 			 * rnh and for the calls to pfxlist_onlink_check() and
1434 			 * defrouter_select() in the block further down for calls
1435 			 * into nd6_lookup().  We still hold a ref.
1436 			 */
1437 			LLE_WUNLOCK(ln);
1438 
1439 			/*
1440 			 * rt6_flush must be called whether or not the neighbor
1441 			 * is in the Default Router List.
1442 			 * See a corresponding comment in nd6_na_input().
1443 			 */
1444 			rt6_flush(&ln->r_l3addr.addr6, ifp);
1445 		}
1446 
1447 		if (dr) {
1448 			/*
1449 			 * Since defrouter_select() does not affect the
1450 			 * on-link determination and MIP6 needs the check
1451 			 * before the default router selection, we perform
1452 			 * the check now.
1453 			 */
1454 			pfxlist_onlink_check();
1455 
1456 			/*
1457 			 * Refresh default router list.
1458 			 */
1459 			defrouter_select();
1460 		}
1461 
1462 		/*
1463 		 * If this entry was added by an on-link redirect, remove the
1464 		 * corresponding host route.
1465 		 */
1466 		if (ln->la_flags & LLE_REDIRECT)
1467 			nd6_free_redirect(ln);
1468 
1469 		if (ln->ln_router || dr)
1470 			LLE_WLOCK(ln);
1471 	}
1472 
1473 	/*
1474 	 * Save to unlock. We still hold an extra reference and will not
1475 	 * free(9) in llentry_free() if someone else holds one as well.
1476 	 */
1477 	LLE_WUNLOCK(ln);
1478 	IF_AFDATA_LOCK(ifp);
1479 	LLE_WLOCK(ln);
1480 	/* Guard against race with other llentry_free(). */
1481 	if (ln->la_flags & LLE_LINKED) {
1482 		/* Remove callout reference */
1483 		LLE_REMREF(ln);
1484 		lltable_unlink_entry(ln->lle_tbl, ln);
1485 	}
1486 	IF_AFDATA_UNLOCK(ifp);
1487 
1488 	llentry_free(ln);
1489 	if (dr != NULL)
1490 		defrouter_rele(dr);
1491 }
1492 
1493 static int
1494 nd6_isdynrte(const struct rtentry *rt, void *xap)
1495 {
1496 
1497 	if (rt->rt_flags == (RTF_UP | RTF_HOST | RTF_DYNAMIC))
1498 		return (1);
1499 
1500 	return (0);
1501 }
1502 /*
1503  * Remove the rtentry for the given llentry,
1504  * both of which were installed by a redirect.
1505  */
1506 static void
1507 nd6_free_redirect(const struct llentry *ln)
1508 {
1509 	int fibnum;
1510 	struct sockaddr_in6 sin6;
1511 	struct rt_addrinfo info;
1512 
1513 	lltable_fill_sa_entry(ln, (struct sockaddr *)&sin6);
1514 	memset(&info, 0, sizeof(info));
1515 	info.rti_info[RTAX_DST] = (struct sockaddr *)&sin6;
1516 	info.rti_filter = nd6_isdynrte;
1517 
1518 	for (fibnum = 0; fibnum < rt_numfibs; fibnum++)
1519 		rtrequest1_fib(RTM_DELETE, &info, NULL, fibnum);
1520 }
1521 
1522 /*
1523  * Rejuvenate this function for routing operations related
1524  * processing.
1525  */
1526 void
1527 nd6_rtrequest(int req, struct rtentry *rt, struct rt_addrinfo *info)
1528 {
1529 	struct sockaddr_in6 *gateway;
1530 	struct nd_defrouter *dr;
1531 	struct ifnet *ifp;
1532 
1533 	gateway = (struct sockaddr_in6 *)rt->rt_gateway;
1534 	ifp = rt->rt_ifp;
1535 
1536 	switch (req) {
1537 	case RTM_ADD:
1538 		break;
1539 
1540 	case RTM_DELETE:
1541 		if (!ifp)
1542 			return;
1543 		/*
1544 		 * Only indirect routes are interesting.
1545 		 */
1546 		if ((rt->rt_flags & RTF_GATEWAY) == 0)
1547 			return;
1548 		/*
1549 		 * check for default route
1550 		 */
1551 		if (IN6_ARE_ADDR_EQUAL(&in6addr_any,
1552 		    &SIN6(rt_key(rt))->sin6_addr)) {
1553 			dr = defrouter_lookup(&gateway->sin6_addr, ifp);
1554 			if (dr != NULL) {
1555 				dr->installed = 0;
1556 				defrouter_rele(dr);
1557 			}
1558 		}
1559 		break;
1560 	}
1561 }
1562 
1563 
1564 int
1565 nd6_ioctl(u_long cmd, caddr_t data, struct ifnet *ifp)
1566 {
1567 	struct in6_ndireq *ndi = (struct in6_ndireq *)data;
1568 	struct in6_nbrinfo *nbi = (struct in6_nbrinfo *)data;
1569 	struct in6_ndifreq *ndif = (struct in6_ndifreq *)data;
1570 	int error = 0;
1571 
1572 	if (ifp->if_afdata[AF_INET6] == NULL)
1573 		return (EPFNOSUPPORT);
1574 	switch (cmd) {
1575 	case OSIOCGIFINFO_IN6:
1576 #define ND	ndi->ndi
1577 		/* XXX: old ndp(8) assumes a positive value for linkmtu. */
1578 		bzero(&ND, sizeof(ND));
1579 		ND.linkmtu = IN6_LINKMTU(ifp);
1580 		ND.maxmtu = ND_IFINFO(ifp)->maxmtu;
1581 		ND.basereachable = ND_IFINFO(ifp)->basereachable;
1582 		ND.reachable = ND_IFINFO(ifp)->reachable;
1583 		ND.retrans = ND_IFINFO(ifp)->retrans;
1584 		ND.flags = ND_IFINFO(ifp)->flags;
1585 		ND.recalctm = ND_IFINFO(ifp)->recalctm;
1586 		ND.chlim = ND_IFINFO(ifp)->chlim;
1587 		break;
1588 	case SIOCGIFINFO_IN6:
1589 		ND = *ND_IFINFO(ifp);
1590 		break;
1591 	case SIOCSIFINFO_IN6:
1592 		/*
1593 		 * used to change host variables from userland.
1594 		 * intended for a use on router to reflect RA configurations.
1595 		 */
1596 		/* 0 means 'unspecified' */
1597 		if (ND.linkmtu != 0) {
1598 			if (ND.linkmtu < IPV6_MMTU ||
1599 			    ND.linkmtu > IN6_LINKMTU(ifp)) {
1600 				error = EINVAL;
1601 				break;
1602 			}
1603 			ND_IFINFO(ifp)->linkmtu = ND.linkmtu;
1604 		}
1605 
1606 		if (ND.basereachable != 0) {
1607 			int obasereachable = ND_IFINFO(ifp)->basereachable;
1608 
1609 			ND_IFINFO(ifp)->basereachable = ND.basereachable;
1610 			if (ND.basereachable != obasereachable)
1611 				ND_IFINFO(ifp)->reachable =
1612 				    ND_COMPUTE_RTIME(ND.basereachable);
1613 		}
1614 		if (ND.retrans != 0)
1615 			ND_IFINFO(ifp)->retrans = ND.retrans;
1616 		if (ND.chlim != 0)
1617 			ND_IFINFO(ifp)->chlim = ND.chlim;
1618 		/* FALLTHROUGH */
1619 	case SIOCSIFINFO_FLAGS:
1620 	{
1621 		struct ifaddr *ifa;
1622 		struct in6_ifaddr *ia;
1623 
1624 		if ((ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED) &&
1625 		    !(ND.flags & ND6_IFF_IFDISABLED)) {
1626 			/* ifdisabled 1->0 transision */
1627 
1628 			/*
1629 			 * If the interface is marked as ND6_IFF_IFDISABLED and
1630 			 * has an link-local address with IN6_IFF_DUPLICATED,
1631 			 * do not clear ND6_IFF_IFDISABLED.
1632 			 * See RFC 4862, Section 5.4.5.
1633 			 */
1634 			IF_ADDR_RLOCK(ifp);
1635 			TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1636 				if (ifa->ifa_addr->sa_family != AF_INET6)
1637 					continue;
1638 				ia = (struct in6_ifaddr *)ifa;
1639 				if ((ia->ia6_flags & IN6_IFF_DUPLICATED) &&
1640 				    IN6_IS_ADDR_LINKLOCAL(IA6_IN6(ia)))
1641 					break;
1642 			}
1643 			IF_ADDR_RUNLOCK(ifp);
1644 
1645 			if (ifa != NULL) {
1646 				/* LLA is duplicated. */
1647 				ND.flags |= ND6_IFF_IFDISABLED;
1648 				log(LOG_ERR, "Cannot enable an interface"
1649 				    " with a link-local address marked"
1650 				    " duplicate.\n");
1651 			} else {
1652 				ND_IFINFO(ifp)->flags &= ~ND6_IFF_IFDISABLED;
1653 				if (ifp->if_flags & IFF_UP)
1654 					in6_if_up(ifp);
1655 			}
1656 		} else if (!(ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED) &&
1657 			    (ND.flags & ND6_IFF_IFDISABLED)) {
1658 			/* ifdisabled 0->1 transision */
1659 			/* Mark all IPv6 address as tentative. */
1660 
1661 			ND_IFINFO(ifp)->flags |= ND6_IFF_IFDISABLED;
1662 			if (V_ip6_dad_count > 0 &&
1663 			    (ND_IFINFO(ifp)->flags & ND6_IFF_NO_DAD) == 0) {
1664 				IF_ADDR_RLOCK(ifp);
1665 				TAILQ_FOREACH(ifa, &ifp->if_addrhead,
1666 				    ifa_link) {
1667 					if (ifa->ifa_addr->sa_family !=
1668 					    AF_INET6)
1669 						continue;
1670 					ia = (struct in6_ifaddr *)ifa;
1671 					ia->ia6_flags |= IN6_IFF_TENTATIVE;
1672 				}
1673 				IF_ADDR_RUNLOCK(ifp);
1674 			}
1675 		}
1676 
1677 		if (ND.flags & ND6_IFF_AUTO_LINKLOCAL) {
1678 			if (!(ND_IFINFO(ifp)->flags & ND6_IFF_AUTO_LINKLOCAL)) {
1679 				/* auto_linklocal 0->1 transision */
1680 
1681 				/* If no link-local address on ifp, configure */
1682 				ND_IFINFO(ifp)->flags |= ND6_IFF_AUTO_LINKLOCAL;
1683 				in6_ifattach(ifp, NULL);
1684 			} else if (!(ND.flags & ND6_IFF_IFDISABLED) &&
1685 			    ifp->if_flags & IFF_UP) {
1686 				/*
1687 				 * When the IF already has
1688 				 * ND6_IFF_AUTO_LINKLOCAL, no link-local
1689 				 * address is assigned, and IFF_UP, try to
1690 				 * assign one.
1691 				 */
1692 				IF_ADDR_RLOCK(ifp);
1693 				TAILQ_FOREACH(ifa, &ifp->if_addrhead,
1694 				    ifa_link) {
1695 					if (ifa->ifa_addr->sa_family !=
1696 					    AF_INET6)
1697 						continue;
1698 					ia = (struct in6_ifaddr *)ifa;
1699 					if (IN6_IS_ADDR_LINKLOCAL(IA6_IN6(ia)))
1700 						break;
1701 				}
1702 				IF_ADDR_RUNLOCK(ifp);
1703 				if (ifa != NULL)
1704 					/* No LLA is configured. */
1705 					in6_ifattach(ifp, NULL);
1706 			}
1707 		}
1708 	}
1709 		ND_IFINFO(ifp)->flags = ND.flags;
1710 		break;
1711 #undef ND
1712 	case SIOCSNDFLUSH_IN6:	/* XXX: the ioctl name is confusing... */
1713 		/* sync kernel routing table with the default router list */
1714 		defrouter_reset();
1715 		defrouter_select();
1716 		break;
1717 	case SIOCSPFXFLUSH_IN6:
1718 	{
1719 		/* flush all the prefix advertised by routers */
1720 		struct nd_prefix *pr, *next;
1721 
1722 		LIST_FOREACH_SAFE(pr, &V_nd_prefix, ndpr_entry, next) {
1723 			struct in6_ifaddr *ia, *ia_next;
1724 
1725 			if (IN6_IS_ADDR_LINKLOCAL(&pr->ndpr_prefix.sin6_addr))
1726 				continue; /* XXX */
1727 
1728 			/* do we really have to remove addresses as well? */
1729 			/* XXXRW: in6_ifaddrhead locking. */
1730 			TAILQ_FOREACH_SAFE(ia, &V_in6_ifaddrhead, ia_link,
1731 			    ia_next) {
1732 				if ((ia->ia6_flags & IN6_IFF_AUTOCONF) == 0)
1733 					continue;
1734 
1735 				if (ia->ia6_ndpr == pr)
1736 					in6_purgeaddr(&ia->ia_ifa);
1737 			}
1738 			prelist_remove(pr);
1739 		}
1740 		break;
1741 	}
1742 	case SIOCSRTRFLUSH_IN6:
1743 	{
1744 		/* flush all the default routers */
1745 		struct nd_drhead drq;
1746 		struct nd_defrouter *dr;
1747 
1748 		TAILQ_INIT(&drq);
1749 
1750 		defrouter_reset();
1751 
1752 		ND6_WLOCK();
1753 		while ((dr = TAILQ_FIRST(&V_nd_defrouter)) != NULL)
1754 			defrouter_unlink(dr, &drq);
1755 		ND6_WUNLOCK();
1756 		while ((dr = TAILQ_FIRST(&drq)) != NULL) {
1757 			TAILQ_REMOVE(&drq, dr, dr_entry);
1758 			defrouter_del(dr);
1759 		}
1760 
1761 		defrouter_select();
1762 		break;
1763 	}
1764 	case SIOCGNBRINFO_IN6:
1765 	{
1766 		struct llentry *ln;
1767 		struct in6_addr nb_addr = nbi->addr; /* make local for safety */
1768 
1769 		if ((error = in6_setscope(&nb_addr, ifp, NULL)) != 0)
1770 			return (error);
1771 
1772 		IF_AFDATA_RLOCK(ifp);
1773 		ln = nd6_lookup(&nb_addr, 0, ifp);
1774 		IF_AFDATA_RUNLOCK(ifp);
1775 
1776 		if (ln == NULL) {
1777 			error = EINVAL;
1778 			break;
1779 		}
1780 		nbi->state = ln->ln_state;
1781 		nbi->asked = ln->la_asked;
1782 		nbi->isrouter = ln->ln_router;
1783 		if (ln->la_expire == 0)
1784 			nbi->expire = 0;
1785 		else
1786 			nbi->expire = ln->la_expire + ln->lle_remtime / hz +
1787 			    (time_second - time_uptime);
1788 		LLE_RUNLOCK(ln);
1789 		break;
1790 	}
1791 	case SIOCGDEFIFACE_IN6:	/* XXX: should be implemented as a sysctl? */
1792 		ndif->ifindex = V_nd6_defifindex;
1793 		break;
1794 	case SIOCSDEFIFACE_IN6:	/* XXX: should be implemented as a sysctl? */
1795 		return (nd6_setdefaultiface(ndif->ifindex));
1796 	}
1797 	return (error);
1798 }
1799 
1800 /*
1801  * Calculates new isRouter value based on provided parameters and
1802  * returns it.
1803  */
1804 static int
1805 nd6_is_router(int type, int code, int is_new, int old_addr, int new_addr,
1806     int ln_router)
1807 {
1808 
1809 	/*
1810 	 * ICMP6 type dependent behavior.
1811 	 *
1812 	 * NS: clear IsRouter if new entry
1813 	 * RS: clear IsRouter
1814 	 * RA: set IsRouter if there's lladdr
1815 	 * redir: clear IsRouter if new entry
1816 	 *
1817 	 * RA case, (1):
1818 	 * The spec says that we must set IsRouter in the following cases:
1819 	 * - If lladdr exist, set IsRouter.  This means (1-5).
1820 	 * - If it is old entry (!newentry), set IsRouter.  This means (7).
1821 	 * So, based on the spec, in (1-5) and (7) cases we must set IsRouter.
1822 	 * A quetion arises for (1) case.  (1) case has no lladdr in the
1823 	 * neighbor cache, this is similar to (6).
1824 	 * This case is rare but we figured that we MUST NOT set IsRouter.
1825 	 *
1826 	 *   is_new  old_addr new_addr 	    NS  RS  RA	redir
1827 	 *							D R
1828 	 *	0	n	n	(1)	c   ?     s
1829 	 *	0	y	n	(2)	c   s     s
1830 	 *	0	n	y	(3)	c   s     s
1831 	 *	0	y	y	(4)	c   s     s
1832 	 *	0	y	y	(5)	c   s     s
1833 	 *	1	--	n	(6) c	c	c s
1834 	 *	1	--	y	(7) c	c   s	c s
1835 	 *
1836 	 *					(c=clear s=set)
1837 	 */
1838 	switch (type & 0xff) {
1839 	case ND_NEIGHBOR_SOLICIT:
1840 		/*
1841 		 * New entry must have is_router flag cleared.
1842 		 */
1843 		if (is_new)					/* (6-7) */
1844 			ln_router = 0;
1845 		break;
1846 	case ND_REDIRECT:
1847 		/*
1848 		 * If the icmp is a redirect to a better router, always set the
1849 		 * is_router flag.  Otherwise, if the entry is newly created,
1850 		 * clear the flag.  [RFC 2461, sec 8.3]
1851 		 */
1852 		if (code == ND_REDIRECT_ROUTER)
1853 			ln_router = 1;
1854 		else {
1855 			if (is_new)				/* (6-7) */
1856 				ln_router = 0;
1857 		}
1858 		break;
1859 	case ND_ROUTER_SOLICIT:
1860 		/*
1861 		 * is_router flag must always be cleared.
1862 		 */
1863 		ln_router = 0;
1864 		break;
1865 	case ND_ROUTER_ADVERT:
1866 		/*
1867 		 * Mark an entry with lladdr as a router.
1868 		 */
1869 		if ((!is_new && (old_addr || new_addr)) ||	/* (2-5) */
1870 		    (is_new && new_addr)) {			/* (7) */
1871 			ln_router = 1;
1872 		}
1873 		break;
1874 	}
1875 
1876 	return (ln_router);
1877 }
1878 
1879 /*
1880  * Create neighbor cache entry and cache link-layer address,
1881  * on reception of inbound ND6 packets.  (RS/RA/NS/redirect)
1882  *
1883  * type - ICMP6 type
1884  * code - type dependent information
1885  *
1886  */
1887 void
1888 nd6_cache_lladdr(struct ifnet *ifp, struct in6_addr *from, char *lladdr,
1889     int lladdrlen, int type, int code)
1890 {
1891 	struct llentry *ln = NULL, *ln_tmp;
1892 	int is_newentry;
1893 	int do_update;
1894 	int olladdr;
1895 	int llchange;
1896 	int flags;
1897 	uint16_t router = 0;
1898 	struct sockaddr_in6 sin6;
1899 	struct mbuf *chain = NULL;
1900 	u_char linkhdr[LLE_MAX_LINKHDR];
1901 	size_t linkhdrsize;
1902 	int lladdr_off;
1903 
1904 	IF_AFDATA_UNLOCK_ASSERT(ifp);
1905 
1906 	KASSERT(ifp != NULL, ("%s: ifp == NULL", __func__));
1907 	KASSERT(from != NULL, ("%s: from == NULL", __func__));
1908 
1909 	/* nothing must be updated for unspecified address */
1910 	if (IN6_IS_ADDR_UNSPECIFIED(from))
1911 		return;
1912 
1913 	/*
1914 	 * Validation about ifp->if_addrlen and lladdrlen must be done in
1915 	 * the caller.
1916 	 *
1917 	 * XXX If the link does not have link-layer adderss, what should
1918 	 * we do? (ifp->if_addrlen == 0)
1919 	 * Spec says nothing in sections for RA, RS and NA.  There's small
1920 	 * description on it in NS section (RFC 2461 7.2.3).
1921 	 */
1922 	flags = lladdr ? LLE_EXCLUSIVE : 0;
1923 	IF_AFDATA_RLOCK(ifp);
1924 	ln = nd6_lookup(from, flags, ifp);
1925 	IF_AFDATA_RUNLOCK(ifp);
1926 	is_newentry = 0;
1927 	if (ln == NULL) {
1928 		flags |= LLE_EXCLUSIVE;
1929 		ln = nd6_alloc(from, 0, ifp);
1930 		if (ln == NULL)
1931 			return;
1932 
1933 		/*
1934 		 * Since we already know all the data for the new entry,
1935 		 * fill it before insertion.
1936 		 */
1937 		if (lladdr != NULL) {
1938 			linkhdrsize = sizeof(linkhdr);
1939 			if (lltable_calc_llheader(ifp, AF_INET6, lladdr,
1940 			    linkhdr, &linkhdrsize, &lladdr_off) != 0)
1941 				return;
1942 			lltable_set_entry_addr(ifp, ln, linkhdr, linkhdrsize,
1943 			    lladdr_off);
1944 		}
1945 
1946 		IF_AFDATA_WLOCK(ifp);
1947 		LLE_WLOCK(ln);
1948 		/* Prefer any existing lle over newly-created one */
1949 		ln_tmp = nd6_lookup(from, LLE_EXCLUSIVE, ifp);
1950 		if (ln_tmp == NULL)
1951 			lltable_link_entry(LLTABLE6(ifp), ln);
1952 		IF_AFDATA_WUNLOCK(ifp);
1953 		if (ln_tmp == NULL) {
1954 			/* No existing lle, mark as new entry (6,7) */
1955 			is_newentry = 1;
1956 			nd6_llinfo_setstate(ln, ND6_LLINFO_STALE);
1957 			if (lladdr != NULL)	/* (7) */
1958 				EVENTHANDLER_INVOKE(lle_event, ln,
1959 				    LLENTRY_RESOLVED);
1960 		} else {
1961 			lltable_free_entry(LLTABLE6(ifp), ln);
1962 			ln = ln_tmp;
1963 			ln_tmp = NULL;
1964 		}
1965 	}
1966 	/* do nothing if static ndp is set */
1967 	if ((ln->la_flags & LLE_STATIC)) {
1968 		if (flags & LLE_EXCLUSIVE)
1969 			LLE_WUNLOCK(ln);
1970 		else
1971 			LLE_RUNLOCK(ln);
1972 		return;
1973 	}
1974 
1975 	olladdr = (ln->la_flags & LLE_VALID) ? 1 : 0;
1976 	if (olladdr && lladdr) {
1977 		llchange = bcmp(lladdr, ln->ll_addr,
1978 		    ifp->if_addrlen);
1979 	} else if (!olladdr && lladdr)
1980 		llchange = 1;
1981 	else
1982 		llchange = 0;
1983 
1984 	/*
1985 	 * newentry olladdr  lladdr  llchange	(*=record)
1986 	 *	0	n	n	--	(1)
1987 	 *	0	y	n	--	(2)
1988 	 *	0	n	y	y	(3) * STALE
1989 	 *	0	y	y	n	(4) *
1990 	 *	0	y	y	y	(5) * STALE
1991 	 *	1	--	n	--	(6)   NOSTATE(= PASSIVE)
1992 	 *	1	--	y	--	(7) * STALE
1993 	 */
1994 
1995 	do_update = 0;
1996 	if (is_newentry == 0 && llchange != 0) {
1997 		do_update = 1;	/* (3,5) */
1998 
1999 		/*
2000 		 * Record source link-layer address
2001 		 * XXX is it dependent to ifp->if_type?
2002 		 */
2003 		linkhdrsize = sizeof(linkhdr);
2004 		if (lltable_calc_llheader(ifp, AF_INET6, lladdr,
2005 		    linkhdr, &linkhdrsize, &lladdr_off) != 0)
2006 			return;
2007 
2008 		if (lltable_try_set_entry_addr(ifp, ln, linkhdr, linkhdrsize,
2009 		    lladdr_off) == 0) {
2010 			/* Entry was deleted */
2011 			return;
2012 		}
2013 
2014 		nd6_llinfo_setstate(ln, ND6_LLINFO_STALE);
2015 
2016 		EVENTHANDLER_INVOKE(lle_event, ln, LLENTRY_RESOLVED);
2017 
2018 		if (ln->la_hold != NULL)
2019 			nd6_grab_holdchain(ln, &chain, &sin6);
2020 	}
2021 
2022 	/* Calculates new router status */
2023 	router = nd6_is_router(type, code, is_newentry, olladdr,
2024 	    lladdr != NULL ? 1 : 0, ln->ln_router);
2025 
2026 	ln->ln_router = router;
2027 	/* Mark non-router redirects with special flag */
2028 	if ((type & 0xFF) == ND_REDIRECT && code != ND_REDIRECT_ROUTER)
2029 		ln->la_flags |= LLE_REDIRECT;
2030 
2031 	if (flags & LLE_EXCLUSIVE)
2032 		LLE_WUNLOCK(ln);
2033 	else
2034 		LLE_RUNLOCK(ln);
2035 
2036 	if (chain != NULL)
2037 		nd6_flush_holdchain(ifp, ifp, chain, &sin6);
2038 
2039 	/*
2040 	 * When the link-layer address of a router changes, select the
2041 	 * best router again.  In particular, when the neighbor entry is newly
2042 	 * created, it might affect the selection policy.
2043 	 * Question: can we restrict the first condition to the "is_newentry"
2044 	 * case?
2045 	 * XXX: when we hear an RA from a new router with the link-layer
2046 	 * address option, defrouter_select() is called twice, since
2047 	 * defrtrlist_update called the function as well.  However, I believe
2048 	 * we can compromise the overhead, since it only happens the first
2049 	 * time.
2050 	 * XXX: although defrouter_select() should not have a bad effect
2051 	 * for those are not autoconfigured hosts, we explicitly avoid such
2052 	 * cases for safety.
2053 	 */
2054 	if ((do_update || is_newentry) && router &&
2055 	    ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) {
2056 		/*
2057 		 * guaranteed recursion
2058 		 */
2059 		defrouter_select();
2060 	}
2061 }
2062 
2063 static void
2064 nd6_slowtimo(void *arg)
2065 {
2066 	CURVNET_SET((struct vnet *) arg);
2067 	struct nd_ifinfo *nd6if;
2068 	struct ifnet *ifp;
2069 
2070 	callout_reset(&V_nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
2071 	    nd6_slowtimo, curvnet);
2072 	IFNET_RLOCK_NOSLEEP();
2073 	TAILQ_FOREACH(ifp, &V_ifnet, if_link) {
2074 		if (ifp->if_afdata[AF_INET6] == NULL)
2075 			continue;
2076 		nd6if = ND_IFINFO(ifp);
2077 		if (nd6if->basereachable && /* already initialized */
2078 		    (nd6if->recalctm -= ND6_SLOWTIMER_INTERVAL) <= 0) {
2079 			/*
2080 			 * Since reachable time rarely changes by router
2081 			 * advertisements, we SHOULD insure that a new random
2082 			 * value gets recomputed at least once every few hours.
2083 			 * (RFC 2461, 6.3.4)
2084 			 */
2085 			nd6if->recalctm = V_nd6_recalc_reachtm_interval;
2086 			nd6if->reachable = ND_COMPUTE_RTIME(nd6if->basereachable);
2087 		}
2088 	}
2089 	IFNET_RUNLOCK_NOSLEEP();
2090 	CURVNET_RESTORE();
2091 }
2092 
2093 void
2094 nd6_grab_holdchain(struct llentry *ln, struct mbuf **chain,
2095     struct sockaddr_in6 *sin6)
2096 {
2097 
2098 	LLE_WLOCK_ASSERT(ln);
2099 
2100 	*chain = ln->la_hold;
2101 	ln->la_hold = NULL;
2102 	lltable_fill_sa_entry(ln, (struct sockaddr *)sin6);
2103 
2104 	if (ln->ln_state == ND6_LLINFO_STALE) {
2105 
2106 		/*
2107 		 * The first time we send a packet to a
2108 		 * neighbor whose entry is STALE, we have
2109 		 * to change the state to DELAY and a sets
2110 		 * a timer to expire in DELAY_FIRST_PROBE_TIME
2111 		 * seconds to ensure do neighbor unreachability
2112 		 * detection on expiration.
2113 		 * (RFC 2461 7.3.3)
2114 		 */
2115 		nd6_llinfo_setstate(ln, ND6_LLINFO_DELAY);
2116 	}
2117 }
2118 
2119 int
2120 nd6_output_ifp(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *m,
2121     struct sockaddr_in6 *dst, struct route *ro)
2122 {
2123 	int error;
2124 	int ip6len;
2125 	struct ip6_hdr *ip6;
2126 	struct m_tag *mtag;
2127 
2128 #ifdef MAC
2129 	mac_netinet6_nd6_send(ifp, m);
2130 #endif
2131 
2132 	/*
2133 	 * If called from nd6_ns_output() (NS), nd6_na_output() (NA),
2134 	 * icmp6_redirect_output() (REDIRECT) or from rip6_output() (RS, RA
2135 	 * as handled by rtsol and rtadvd), mbufs will be tagged for SeND
2136 	 * to be diverted to user space.  When re-injected into the kernel,
2137 	 * send_output() will directly dispatch them to the outgoing interface.
2138 	 */
2139 	if (send_sendso_input_hook != NULL) {
2140 		mtag = m_tag_find(m, PACKET_TAG_ND_OUTGOING, NULL);
2141 		if (mtag != NULL) {
2142 			ip6 = mtod(m, struct ip6_hdr *);
2143 			ip6len = sizeof(struct ip6_hdr) + ntohs(ip6->ip6_plen);
2144 			/* Use the SEND socket */
2145 			error = send_sendso_input_hook(m, ifp, SND_OUT,
2146 			    ip6len);
2147 			/* -1 == no app on SEND socket */
2148 			if (error == 0 || error != -1)
2149 			    return (error);
2150 		}
2151 	}
2152 
2153 	m_clrprotoflags(m);	/* Avoid confusing lower layers. */
2154 	IP_PROBE(send, NULL, NULL, mtod(m, struct ip6_hdr *), ifp, NULL,
2155 	    mtod(m, struct ip6_hdr *));
2156 
2157 	if ((ifp->if_flags & IFF_LOOPBACK) == 0)
2158 		origifp = ifp;
2159 
2160 	error = (*ifp->if_output)(origifp, m, (struct sockaddr *)dst, ro);
2161 	return (error);
2162 }
2163 
2164 /*
2165  * Lookup link headerfor @sa_dst address. Stores found
2166  * data in @desten buffer. Copy of lle ln_flags can be also
2167  * saved in @pflags if @pflags is non-NULL.
2168  *
2169  * If destination LLE does not exists or lle state modification
2170  * is required, call "slow" version.
2171  *
2172  * Return values:
2173  * - 0 on success (address copied to buffer).
2174  * - EWOULDBLOCK (no local error, but address is still unresolved)
2175  * - other errors (alloc failure, etc)
2176  */
2177 int
2178 nd6_resolve(struct ifnet *ifp, int is_gw, struct mbuf *m,
2179     const struct sockaddr *sa_dst, u_char *desten, uint32_t *pflags,
2180     struct llentry **plle)
2181 {
2182 	struct llentry *ln = NULL;
2183 	const struct sockaddr_in6 *dst6;
2184 
2185 	if (pflags != NULL)
2186 		*pflags = 0;
2187 
2188 	dst6 = (const struct sockaddr_in6 *)sa_dst;
2189 
2190 	/* discard the packet if IPv6 operation is disabled on the interface */
2191 	if ((ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED)) {
2192 		m_freem(m);
2193 		return (ENETDOWN); /* better error? */
2194 	}
2195 
2196 	if (m != NULL && m->m_flags & M_MCAST) {
2197 		switch (ifp->if_type) {
2198 		case IFT_ETHER:
2199 		case IFT_FDDI:
2200 		case IFT_L2VLAN:
2201 		case IFT_IEEE80211:
2202 		case IFT_BRIDGE:
2203 		case IFT_ISO88025:
2204 			ETHER_MAP_IPV6_MULTICAST(&dst6->sin6_addr,
2205 						 desten);
2206 			return (0);
2207 		default:
2208 			m_freem(m);
2209 			return (EAFNOSUPPORT);
2210 		}
2211 	}
2212 
2213 	IF_AFDATA_RLOCK(ifp);
2214 	ln = nd6_lookup(&dst6->sin6_addr, LLE_UNLOCKED, ifp);
2215 	if (ln != NULL && (ln->r_flags & RLLE_VALID) != 0) {
2216 		/* Entry found, let's copy lle info */
2217 		bcopy(ln->r_linkdata, desten, ln->r_hdrlen);
2218 		if (pflags != NULL)
2219 			*pflags = LLE_VALID | (ln->r_flags & RLLE_IFADDR);
2220 		/* Check if we have feedback request from nd6 timer */
2221 		if (ln->r_skip_req != 0) {
2222 			LLE_REQ_LOCK(ln);
2223 			ln->r_skip_req = 0; /* Notify that entry was used */
2224 			ln->lle_hittime = time_uptime;
2225 			LLE_REQ_UNLOCK(ln);
2226 		}
2227 		IF_AFDATA_RUNLOCK(ifp);
2228 		return (0);
2229 	}
2230 	IF_AFDATA_RUNLOCK(ifp);
2231 
2232 	return (nd6_resolve_slow(ifp, 0, m, dst6, desten, pflags, plle));
2233 }
2234 
2235 
2236 /*
2237  * Do L2 address resolution for @sa_dst address. Stores found
2238  * address in @desten buffer. Copy of lle ln_flags can be also
2239  * saved in @pflags if @pflags is non-NULL.
2240  *
2241  * Heavy version.
2242  * Function assume that destination LLE does not exist,
2243  * is invalid or stale, so LLE_EXCLUSIVE lock needs to be acquired.
2244  *
2245  * Set noinline to be dtrace-friendly
2246  */
2247 static __noinline int
2248 nd6_resolve_slow(struct ifnet *ifp, int flags, struct mbuf *m,
2249     const struct sockaddr_in6 *dst, u_char *desten, uint32_t *pflags,
2250     struct llentry **plle)
2251 {
2252 	struct llentry *lle = NULL, *lle_tmp;
2253 	struct in6_addr *psrc, src;
2254 	int send_ns, ll_len;
2255 	char *lladdr;
2256 
2257 	/*
2258 	 * Address resolution or Neighbor Unreachability Detection
2259 	 * for the next hop.
2260 	 * At this point, the destination of the packet must be a unicast
2261 	 * or an anycast address(i.e. not a multicast).
2262 	 */
2263 	if (lle == NULL) {
2264 		IF_AFDATA_RLOCK(ifp);
2265 		lle = nd6_lookup(&dst->sin6_addr, LLE_EXCLUSIVE, ifp);
2266 		IF_AFDATA_RUNLOCK(ifp);
2267 		if ((lle == NULL) && nd6_is_addr_neighbor(dst, ifp))  {
2268 			/*
2269 			 * Since nd6_is_addr_neighbor() internally calls nd6_lookup(),
2270 			 * the condition below is not very efficient.  But we believe
2271 			 * it is tolerable, because this should be a rare case.
2272 			 */
2273 			lle = nd6_alloc(&dst->sin6_addr, 0, ifp);
2274 			if (lle == NULL) {
2275 				char ip6buf[INET6_ADDRSTRLEN];
2276 				log(LOG_DEBUG,
2277 				    "nd6_output: can't allocate llinfo for %s "
2278 				    "(ln=%p)\n",
2279 				    ip6_sprintf(ip6buf, &dst->sin6_addr), lle);
2280 				m_freem(m);
2281 				return (ENOBUFS);
2282 			}
2283 
2284 			IF_AFDATA_WLOCK(ifp);
2285 			LLE_WLOCK(lle);
2286 			/* Prefer any existing entry over newly-created one */
2287 			lle_tmp = nd6_lookup(&dst->sin6_addr, LLE_EXCLUSIVE, ifp);
2288 			if (lle_tmp == NULL)
2289 				lltable_link_entry(LLTABLE6(ifp), lle);
2290 			IF_AFDATA_WUNLOCK(ifp);
2291 			if (lle_tmp != NULL) {
2292 				lltable_free_entry(LLTABLE6(ifp), lle);
2293 				lle = lle_tmp;
2294 				lle_tmp = NULL;
2295 			}
2296 		}
2297 	}
2298 	if (lle == NULL) {
2299 		if (!(ND_IFINFO(ifp)->flags & ND6_IFF_PERFORMNUD)) {
2300 			m_freem(m);
2301 			return (ENOBUFS);
2302 		}
2303 
2304 		if (m != NULL)
2305 			m_freem(m);
2306 		return (ENOBUFS);
2307 	}
2308 
2309 	LLE_WLOCK_ASSERT(lle);
2310 
2311 	/*
2312 	 * The first time we send a packet to a neighbor whose entry is
2313 	 * STALE, we have to change the state to DELAY and a sets a timer to
2314 	 * expire in DELAY_FIRST_PROBE_TIME seconds to ensure do
2315 	 * neighbor unreachability detection on expiration.
2316 	 * (RFC 2461 7.3.3)
2317 	 */
2318 	if (lle->ln_state == ND6_LLINFO_STALE)
2319 		nd6_llinfo_setstate(lle, ND6_LLINFO_DELAY);
2320 
2321 	/*
2322 	 * If the neighbor cache entry has a state other than INCOMPLETE
2323 	 * (i.e. its link-layer address is already resolved), just
2324 	 * send the packet.
2325 	 */
2326 	if (lle->ln_state > ND6_LLINFO_INCOMPLETE) {
2327 		if (flags & LLE_ADDRONLY) {
2328 			lladdr = lle->ll_addr;
2329 			ll_len = ifp->if_addrlen;
2330 		} else {
2331 			lladdr = lle->r_linkdata;
2332 			ll_len = lle->r_hdrlen;
2333 		}
2334 		bcopy(lladdr, desten, ll_len);
2335 		if (pflags != NULL)
2336 			*pflags = lle->la_flags;
2337 		if (plle) {
2338 			LLE_ADDREF(lle);
2339 			*plle = lle;
2340 		}
2341 		LLE_WUNLOCK(lle);
2342 		return (0);
2343 	}
2344 
2345 	/*
2346 	 * There is a neighbor cache entry, but no ethernet address
2347 	 * response yet.  Append this latest packet to the end of the
2348 	 * packet queue in the mbuf.  When it exceeds nd6_maxqueuelen,
2349 	 * the oldest packet in the queue will be removed.
2350 	 */
2351 
2352 	if (lle->la_hold != NULL) {
2353 		struct mbuf *m_hold;
2354 		int i;
2355 
2356 		i = 0;
2357 		for (m_hold = lle->la_hold; m_hold; m_hold = m_hold->m_nextpkt){
2358 			i++;
2359 			if (m_hold->m_nextpkt == NULL) {
2360 				m_hold->m_nextpkt = m;
2361 				break;
2362 			}
2363 		}
2364 		while (i >= V_nd6_maxqueuelen) {
2365 			m_hold = lle->la_hold;
2366 			lle->la_hold = lle->la_hold->m_nextpkt;
2367 			m_freem(m_hold);
2368 			i--;
2369 		}
2370 	} else {
2371 		lle->la_hold = m;
2372 	}
2373 
2374 	/*
2375 	 * If there has been no NS for the neighbor after entering the
2376 	 * INCOMPLETE state, send the first solicitation.
2377 	 * Note that for newly-created lle la_asked will be 0,
2378 	 * so we will transition from ND6_LLINFO_NOSTATE to
2379 	 * ND6_LLINFO_INCOMPLETE state here.
2380 	 */
2381 	psrc = NULL;
2382 	send_ns = 0;
2383 	if (lle->la_asked == 0) {
2384 		lle->la_asked++;
2385 		send_ns = 1;
2386 		psrc = nd6_llinfo_get_holdsrc(lle, &src);
2387 
2388 		nd6_llinfo_setstate(lle, ND6_LLINFO_INCOMPLETE);
2389 	}
2390 	LLE_WUNLOCK(lle);
2391 	if (send_ns != 0)
2392 		nd6_ns_output(ifp, psrc, NULL, &dst->sin6_addr, NULL);
2393 
2394 	return (EWOULDBLOCK);
2395 }
2396 
2397 /*
2398  * Do L2 address resolution for @sa_dst address. Stores found
2399  * address in @desten buffer. Copy of lle ln_flags can be also
2400  * saved in @pflags if @pflags is non-NULL.
2401  *
2402  * Return values:
2403  * - 0 on success (address copied to buffer).
2404  * - EWOULDBLOCK (no local error, but address is still unresolved)
2405  * - other errors (alloc failure, etc)
2406  */
2407 int
2408 nd6_resolve_addr(struct ifnet *ifp, int flags, const struct sockaddr *dst,
2409     char *desten, uint32_t *pflags)
2410 {
2411 	int error;
2412 
2413 	flags |= LLE_ADDRONLY;
2414 	error = nd6_resolve_slow(ifp, flags, NULL,
2415 	    (const struct sockaddr_in6 *)dst, desten, pflags, NULL);
2416 	return (error);
2417 }
2418 
2419 int
2420 nd6_flush_holdchain(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *chain,
2421     struct sockaddr_in6 *dst)
2422 {
2423 	struct mbuf *m, *m_head;
2424 	struct ifnet *outifp;
2425 	int error = 0;
2426 
2427 	m_head = chain;
2428 	if ((ifp->if_flags & IFF_LOOPBACK) != 0)
2429 		outifp = origifp;
2430 	else
2431 		outifp = ifp;
2432 
2433 	while (m_head) {
2434 		m = m_head;
2435 		m_head = m_head->m_nextpkt;
2436 		error = nd6_output_ifp(ifp, origifp, m, dst, NULL);
2437 	}
2438 
2439 	/*
2440 	 * XXX
2441 	 * note that intermediate errors are blindly ignored
2442 	 */
2443 	return (error);
2444 }
2445 
2446 static int
2447 nd6_need_cache(struct ifnet *ifp)
2448 {
2449 	/*
2450 	 * XXX: we currently do not make neighbor cache on any interface
2451 	 * other than ARCnet, Ethernet, FDDI and GIF.
2452 	 *
2453 	 * RFC2893 says:
2454 	 * - unidirectional tunnels needs no ND
2455 	 */
2456 	switch (ifp->if_type) {
2457 	case IFT_ARCNET:
2458 	case IFT_ETHER:
2459 	case IFT_FDDI:
2460 	case IFT_IEEE1394:
2461 	case IFT_L2VLAN:
2462 	case IFT_IEEE80211:
2463 	case IFT_INFINIBAND:
2464 	case IFT_BRIDGE:
2465 	case IFT_PROPVIRTUAL:
2466 		return (1);
2467 	default:
2468 		return (0);
2469 	}
2470 }
2471 
2472 /*
2473  * Add pernament ND6 link-layer record for given
2474  * interface address.
2475  *
2476  * Very similar to IPv4 arp_ifinit(), but:
2477  * 1) IPv6 DAD is performed in different place
2478  * 2) It is called by IPv6 protocol stack in contrast to
2479  * arp_ifinit() which is typically called in SIOCSIFADDR
2480  * driver ioctl handler.
2481  *
2482  */
2483 int
2484 nd6_add_ifa_lle(struct in6_ifaddr *ia)
2485 {
2486 	struct ifnet *ifp;
2487 	struct llentry *ln, *ln_tmp;
2488 	struct sockaddr *dst;
2489 
2490 	ifp = ia->ia_ifa.ifa_ifp;
2491 	if (nd6_need_cache(ifp) == 0)
2492 		return (0);
2493 
2494 	ia->ia_ifa.ifa_rtrequest = nd6_rtrequest;
2495 	dst = (struct sockaddr *)&ia->ia_addr;
2496 	ln = lltable_alloc_entry(LLTABLE6(ifp), LLE_IFADDR, dst);
2497 	if (ln == NULL)
2498 		return (ENOBUFS);
2499 
2500 	IF_AFDATA_WLOCK(ifp);
2501 	LLE_WLOCK(ln);
2502 	/* Unlink any entry if exists */
2503 	ln_tmp = lla_lookup(LLTABLE6(ifp), LLE_EXCLUSIVE, dst);
2504 	if (ln_tmp != NULL)
2505 		lltable_unlink_entry(LLTABLE6(ifp), ln_tmp);
2506 	lltable_link_entry(LLTABLE6(ifp), ln);
2507 	IF_AFDATA_WUNLOCK(ifp);
2508 
2509 	if (ln_tmp != NULL)
2510 		EVENTHANDLER_INVOKE(lle_event, ln_tmp, LLENTRY_EXPIRED);
2511 	EVENTHANDLER_INVOKE(lle_event, ln, LLENTRY_RESOLVED);
2512 
2513 	LLE_WUNLOCK(ln);
2514 	if (ln_tmp != NULL)
2515 		llentry_free(ln_tmp);
2516 
2517 	return (0);
2518 }
2519 
2520 /*
2521  * Removes either all lle entries for given @ia, or lle
2522  * corresponding to @ia address.
2523  */
2524 void
2525 nd6_rem_ifa_lle(struct in6_ifaddr *ia, int all)
2526 {
2527 	struct sockaddr_in6 mask, addr;
2528 	struct sockaddr *saddr, *smask;
2529 	struct ifnet *ifp;
2530 
2531 	ifp = ia->ia_ifa.ifa_ifp;
2532 	memcpy(&addr, &ia->ia_addr, sizeof(ia->ia_addr));
2533 	memcpy(&mask, &ia->ia_prefixmask, sizeof(ia->ia_prefixmask));
2534 	saddr = (struct sockaddr *)&addr;
2535 	smask = (struct sockaddr *)&mask;
2536 
2537 	if (all != 0)
2538 		lltable_prefix_free(AF_INET6, saddr, smask, LLE_STATIC);
2539 	else
2540 		lltable_delete_addr(LLTABLE6(ifp), LLE_IFADDR, saddr);
2541 }
2542 
2543 static void
2544 clear_llinfo_pqueue(struct llentry *ln)
2545 {
2546 	struct mbuf *m_hold, *m_hold_next;
2547 
2548 	for (m_hold = ln->la_hold; m_hold; m_hold = m_hold_next) {
2549 		m_hold_next = m_hold->m_nextpkt;
2550 		m_freem(m_hold);
2551 	}
2552 
2553 	ln->la_hold = NULL;
2554 }
2555 
2556 static int nd6_sysctl_drlist(SYSCTL_HANDLER_ARGS);
2557 static int nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS);
2558 
2559 SYSCTL_DECL(_net_inet6_icmp6);
2560 SYSCTL_PROC(_net_inet6_icmp6, ICMPV6CTL_ND6_DRLIST, nd6_drlist,
2561 	CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_MPSAFE,
2562 	NULL, 0, nd6_sysctl_drlist, "S,in6_defrouter",
2563 	"NDP default router list");
2564 SYSCTL_PROC(_net_inet6_icmp6, ICMPV6CTL_ND6_PRLIST, nd6_prlist,
2565 	CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_MPSAFE,
2566 	NULL, 0, nd6_sysctl_prlist, "S,in6_prefix",
2567 	"NDP prefix list");
2568 SYSCTL_INT(_net_inet6_icmp6, ICMPV6CTL_ND6_MAXQLEN, nd6_maxqueuelen,
2569 	CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(nd6_maxqueuelen), 1, "");
2570 SYSCTL_INT(_net_inet6_icmp6, OID_AUTO, nd6_gctimer,
2571 	CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(nd6_gctimer), (60 * 60 * 24), "");
2572 
2573 static int
2574 nd6_sysctl_drlist(SYSCTL_HANDLER_ARGS)
2575 {
2576 	struct in6_defrouter d;
2577 	struct nd_defrouter *dr;
2578 	int error;
2579 
2580 	if (req->newptr != NULL)
2581 		return (EPERM);
2582 
2583 	error = sysctl_wire_old_buffer(req, 0);
2584 	if (error != 0)
2585 		return (error);
2586 
2587 	bzero(&d, sizeof(d));
2588 	d.rtaddr.sin6_family = AF_INET6;
2589 	d.rtaddr.sin6_len = sizeof(d.rtaddr);
2590 
2591 	ND6_RLOCK();
2592 	TAILQ_FOREACH(dr, &V_nd_defrouter, dr_entry) {
2593 		d.rtaddr.sin6_addr = dr->rtaddr;
2594 		error = sa6_recoverscope(&d.rtaddr);
2595 		if (error != 0)
2596 			break;
2597 		d.flags = dr->raflags;
2598 		d.rtlifetime = dr->rtlifetime;
2599 		d.expire = dr->expire + (time_second - time_uptime);
2600 		d.if_index = dr->ifp->if_index;
2601 		error = SYSCTL_OUT(req, &d, sizeof(d));
2602 		if (error != 0)
2603 			break;
2604 	}
2605 	ND6_RUNLOCK();
2606 	return (error);
2607 }
2608 
2609 static int
2610 nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS)
2611 {
2612 	struct in6_prefix p;
2613 	struct sockaddr_in6 s6;
2614 	struct nd_prefix *pr;
2615 	struct nd_pfxrouter *pfr;
2616 	time_t maxexpire;
2617 	int error;
2618 	char ip6buf[INET6_ADDRSTRLEN];
2619 
2620 	if (req->newptr)
2621 		return (EPERM);
2622 
2623 	error = sysctl_wire_old_buffer(req, 0);
2624 	if (error != 0)
2625 		return (error);
2626 
2627 	bzero(&p, sizeof(p));
2628 	p.origin = PR_ORIG_RA;
2629 	bzero(&s6, sizeof(s6));
2630 	s6.sin6_family = AF_INET6;
2631 	s6.sin6_len = sizeof(s6);
2632 
2633 	ND6_RLOCK();
2634 	LIST_FOREACH(pr, &V_nd_prefix, ndpr_entry) {
2635 		p.prefix = pr->ndpr_prefix;
2636 		if (sa6_recoverscope(&p.prefix)) {
2637 			log(LOG_ERR, "scope error in prefix list (%s)\n",
2638 			    ip6_sprintf(ip6buf, &p.prefix.sin6_addr));
2639 			/* XXX: press on... */
2640 		}
2641 		p.raflags = pr->ndpr_raf;
2642 		p.prefixlen = pr->ndpr_plen;
2643 		p.vltime = pr->ndpr_vltime;
2644 		p.pltime = pr->ndpr_pltime;
2645 		p.if_index = pr->ndpr_ifp->if_index;
2646 		if (pr->ndpr_vltime == ND6_INFINITE_LIFETIME)
2647 			p.expire = 0;
2648 		else {
2649 			/* XXX: we assume time_t is signed. */
2650 			maxexpire = (-1) &
2651 			    ~((time_t)1 << ((sizeof(maxexpire) * 8) - 1));
2652 			if (pr->ndpr_vltime < maxexpire - pr->ndpr_lastupdate)
2653 				p.expire = pr->ndpr_lastupdate +
2654 				    pr->ndpr_vltime +
2655 				    (time_second - time_uptime);
2656 			else
2657 				p.expire = maxexpire;
2658 		}
2659 		p.refcnt = pr->ndpr_refcnt;
2660 		p.flags = pr->ndpr_stateflags;
2661 		p.advrtrs = 0;
2662 		LIST_FOREACH(pfr, &pr->ndpr_advrtrs, pfr_entry)
2663 			p.advrtrs++;
2664 		error = SYSCTL_OUT(req, &p, sizeof(p));
2665 		if (error != 0)
2666 			break;
2667 		LIST_FOREACH(pfr, &pr->ndpr_advrtrs, pfr_entry) {
2668 			s6.sin6_addr = pfr->router->rtaddr;
2669 			if (sa6_recoverscope(&s6))
2670 				log(LOG_ERR,
2671 				    "scope error in prefix list (%s)\n",
2672 				    ip6_sprintf(ip6buf, &pfr->router->rtaddr));
2673 			error = SYSCTL_OUT(req, &s6, sizeof(s6));
2674 			if (error != 0)
2675 				break;
2676 		}
2677 	}
2678 	ND6_RUNLOCK();
2679 	return (error);
2680 }
2681