xref: /freebsd/sys/netinet6/nd6.c (revision e94f204a324d7dc60476c32be5af474a736c50d4)
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 /* timer values */
89 VNET_DEFINE(int, nd6_prune)	= 1;	/* walk list every 1 seconds */
90 VNET_DEFINE(int, nd6_delay)	= 5;	/* delay first probe time 5 second */
91 VNET_DEFINE(int, nd6_umaxtries)	= 3;	/* maximum unicast query */
92 VNET_DEFINE(int, nd6_mmaxtries)	= 3;	/* maximum multicast query */
93 VNET_DEFINE(int, nd6_useloopback) = 1;	/* use loopback interface for
94 					 * local traffic */
95 VNET_DEFINE(int, nd6_gctimer)	= (60 * 60 * 24); /* 1 day: garbage
96 					 * collection timer */
97 
98 /* preventing too many loops in ND option parsing */
99 static VNET_DEFINE(int, nd6_maxndopt) = 10; /* max # of ND options allowed */
100 
101 VNET_DEFINE(int, nd6_maxnudhint) = 0;	/* max # of subsequent upper
102 					 * layer hints */
103 static VNET_DEFINE(int, nd6_maxqueuelen) = 1; /* max pkts cached in unresolved
104 					 * ND entries */
105 #define	V_nd6_maxndopt			VNET(nd6_maxndopt)
106 #define	V_nd6_maxqueuelen		VNET(nd6_maxqueuelen)
107 
108 #ifdef ND6_DEBUG
109 VNET_DEFINE(int, nd6_debug) = 1;
110 #else
111 VNET_DEFINE(int, nd6_debug) = 0;
112 #endif
113 
114 static eventhandler_tag lle_event_eh, iflladdr_event_eh;
115 
116 VNET_DEFINE(struct nd_drhead, nd_defrouter);
117 VNET_DEFINE(struct nd_prhead, nd_prefix);
118 VNET_DEFINE(struct rwlock, nd6_lock);
119 
120 VNET_DEFINE(int, nd6_recalc_reachtm_interval) = ND6_RECALC_REACHTM_INTERVAL;
121 #define	V_nd6_recalc_reachtm_interval	VNET(nd6_recalc_reachtm_interval)
122 
123 int	(*send_sendso_input_hook)(struct mbuf *, struct ifnet *, int, int);
124 
125 static int nd6_is_new_addr_neighbor(const struct sockaddr_in6 *,
126 	struct ifnet *);
127 static void nd6_setmtu0(struct ifnet *, struct nd_ifinfo *);
128 static void nd6_slowtimo(void *);
129 static int regen_tmpaddr(struct in6_ifaddr *);
130 static void nd6_free(struct llentry **, int);
131 static void nd6_free_redirect(const struct llentry *);
132 static void nd6_llinfo_timer(void *);
133 static void nd6_llinfo_settimer_locked(struct llentry *, long);
134 static void clear_llinfo_pqueue(struct llentry *);
135 static void nd6_rtrequest(int, struct rtentry *, struct rt_addrinfo *);
136 static int nd6_resolve_slow(struct ifnet *, int, struct mbuf *,
137     const struct sockaddr_in6 *, u_char *, uint32_t *);
138 static int nd6_need_cache(struct ifnet *);
139 
140 
141 static VNET_DEFINE(struct callout, nd6_slowtimo_ch);
142 #define	V_nd6_slowtimo_ch		VNET(nd6_slowtimo_ch)
143 
144 VNET_DEFINE(struct callout, nd6_timer_ch);
145 #define	V_nd6_timer_ch			VNET(nd6_timer_ch)
146 
147 static void
148 nd6_lle_event(void *arg __unused, struct llentry *lle, int evt)
149 {
150 	struct rt_addrinfo rtinfo;
151 	struct sockaddr_in6 dst;
152 	struct sockaddr_dl gw;
153 	struct ifnet *ifp;
154 	int type;
155 
156 	LLE_WLOCK_ASSERT(lle);
157 
158 	if (lltable_get_af(lle->lle_tbl) != AF_INET6)
159 		return;
160 
161 	switch (evt) {
162 	case LLENTRY_RESOLVED:
163 		type = RTM_ADD;
164 		KASSERT(lle->la_flags & LLE_VALID,
165 		    ("%s: %p resolved but not valid?", __func__, lle));
166 		break;
167 	case LLENTRY_EXPIRED:
168 		type = RTM_DELETE;
169 		break;
170 	default:
171 		return;
172 	}
173 
174 	ifp = lltable_get_ifp(lle->lle_tbl);
175 
176 	bzero(&dst, sizeof(dst));
177 	bzero(&gw, sizeof(gw));
178 	bzero(&rtinfo, sizeof(rtinfo));
179 	lltable_fill_sa_entry(lle, (struct sockaddr *)&dst);
180 	dst.sin6_scope_id = in6_getscopezone(ifp,
181 	    in6_addrscope(&dst.sin6_addr));
182 	gw.sdl_len = sizeof(struct sockaddr_dl);
183 	gw.sdl_family = AF_LINK;
184 	gw.sdl_alen = ifp->if_addrlen;
185 	gw.sdl_index = ifp->if_index;
186 	gw.sdl_type = ifp->if_type;
187 	if (evt == LLENTRY_RESOLVED)
188 		bcopy(lle->ll_addr, gw.sdl_data, ifp->if_addrlen);
189 	rtinfo.rti_info[RTAX_DST] = (struct sockaddr *)&dst;
190 	rtinfo.rti_info[RTAX_GATEWAY] = (struct sockaddr *)&gw;
191 	rtinfo.rti_addrs = RTA_DST | RTA_GATEWAY;
192 	rt_missmsg_fib(type, &rtinfo, RTF_HOST | RTF_LLDATA | (
193 	    type == RTM_ADD ? RTF_UP: 0), 0, RT_DEFAULT_FIB);
194 }
195 
196 /*
197  * A handler for interface link layer address change event.
198  */
199 static void
200 nd6_iflladdr(void *arg __unused, struct ifnet *ifp)
201 {
202 
203 	lltable_update_ifaddr(LLTABLE6(ifp));
204 }
205 
206 void
207 nd6_init(void)
208 {
209 
210 	rw_init(&V_nd6_lock, "nd6");
211 
212 	LIST_INIT(&V_nd_prefix);
213 
214 	/* initialization of the default router list */
215 	TAILQ_INIT(&V_nd_defrouter);
216 
217 	/* Start timers. */
218 	callout_init(&V_nd6_slowtimo_ch, 0);
219 	callout_reset(&V_nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
220 	    nd6_slowtimo, curvnet);
221 
222 	callout_init(&V_nd6_timer_ch, 0);
223 	callout_reset(&V_nd6_timer_ch, hz, nd6_timer, curvnet);
224 
225 	nd6_dad_init();
226 	if (IS_DEFAULT_VNET(curvnet)) {
227 		lle_event_eh = EVENTHANDLER_REGISTER(lle_event, nd6_lle_event,
228 		    NULL, EVENTHANDLER_PRI_ANY);
229 		iflladdr_event_eh = EVENTHANDLER_REGISTER(iflladdr_event,
230 		    nd6_iflladdr, NULL, EVENTHANDLER_PRI_ANY);
231 	}
232 }
233 
234 #ifdef VIMAGE
235 void
236 nd6_destroy()
237 {
238 
239 	callout_drain(&V_nd6_slowtimo_ch);
240 	callout_drain(&V_nd6_timer_ch);
241 	if (IS_DEFAULT_VNET(curvnet)) {
242 		EVENTHANDLER_DEREGISTER(lle_event, lle_event_eh);
243 		EVENTHANDLER_DEREGISTER(iflladdr_event, iflladdr_event_eh);
244 	}
245 	rw_destroy(&V_nd6_lock);
246 }
247 #endif
248 
249 struct nd_ifinfo *
250 nd6_ifattach(struct ifnet *ifp)
251 {
252 	struct nd_ifinfo *nd;
253 
254 	nd = malloc(sizeof(*nd), M_IP6NDP, M_WAITOK | M_ZERO);
255 	nd->initialized = 1;
256 
257 	nd->chlim = IPV6_DEFHLIM;
258 	nd->basereachable = REACHABLE_TIME;
259 	nd->reachable = ND_COMPUTE_RTIME(nd->basereachable);
260 	nd->retrans = RETRANS_TIMER;
261 
262 	nd->flags = ND6_IFF_PERFORMNUD;
263 
264 	/* A loopback interface always has ND6_IFF_AUTO_LINKLOCAL.
265 	 * XXXHRS: Clear ND6_IFF_AUTO_LINKLOCAL on an IFT_BRIDGE interface by
266 	 * default regardless of the V_ip6_auto_linklocal configuration to
267 	 * give a reasonable default behavior.
268 	 */
269 	if ((V_ip6_auto_linklocal && ifp->if_type != IFT_BRIDGE) ||
270 	    (ifp->if_flags & IFF_LOOPBACK))
271 		nd->flags |= ND6_IFF_AUTO_LINKLOCAL;
272 	/*
273 	 * A loopback interface does not need to accept RTADV.
274 	 * XXXHRS: Clear ND6_IFF_ACCEPT_RTADV on an IFT_BRIDGE interface by
275 	 * default regardless of the V_ip6_accept_rtadv configuration to
276 	 * prevent the interface from accepting RA messages arrived
277 	 * on one of the member interfaces with ND6_IFF_ACCEPT_RTADV.
278 	 */
279 	if (V_ip6_accept_rtadv &&
280 	    !(ifp->if_flags & IFF_LOOPBACK) &&
281 	    (ifp->if_type != IFT_BRIDGE))
282 			nd->flags |= ND6_IFF_ACCEPT_RTADV;
283 	if (V_ip6_no_radr && !(ifp->if_flags & IFF_LOOPBACK))
284 		nd->flags |= ND6_IFF_NO_RADR;
285 
286 	/* XXX: we cannot call nd6_setmtu since ifp is not fully initialized */
287 	nd6_setmtu0(ifp, nd);
288 
289 	return nd;
290 }
291 
292 void
293 nd6_ifdetach(struct nd_ifinfo *nd)
294 {
295 
296 	free(nd, M_IP6NDP);
297 }
298 
299 /*
300  * Reset ND level link MTU. This function is called when the physical MTU
301  * changes, which means we might have to adjust the ND level MTU.
302  */
303 void
304 nd6_setmtu(struct ifnet *ifp)
305 {
306 	if (ifp->if_afdata[AF_INET6] == NULL)
307 		return;
308 
309 	nd6_setmtu0(ifp, ND_IFINFO(ifp));
310 }
311 
312 /* XXX todo: do not maintain copy of ifp->if_mtu in ndi->maxmtu */
313 void
314 nd6_setmtu0(struct ifnet *ifp, struct nd_ifinfo *ndi)
315 {
316 	u_int32_t omaxmtu;
317 
318 	omaxmtu = ndi->maxmtu;
319 
320 	switch (ifp->if_type) {
321 	case IFT_ARCNET:
322 		ndi->maxmtu = MIN(ARC_PHDS_MAXMTU, ifp->if_mtu); /* RFC2497 */
323 		break;
324 	case IFT_FDDI:
325 		ndi->maxmtu = MIN(FDDIIPMTU, ifp->if_mtu); /* RFC2467 */
326 		break;
327 	case IFT_ISO88025:
328 		 ndi->maxmtu = MIN(ISO88025_MAX_MTU, ifp->if_mtu);
329 		 break;
330 	default:
331 		ndi->maxmtu = ifp->if_mtu;
332 		break;
333 	}
334 
335 	/*
336 	 * Decreasing the interface MTU under IPV6 minimum MTU may cause
337 	 * undesirable situation.  We thus notify the operator of the change
338 	 * explicitly.  The check for omaxmtu is necessary to restrict the
339 	 * log to the case of changing the MTU, not initializing it.
340 	 */
341 	if (omaxmtu >= IPV6_MMTU && ndi->maxmtu < IPV6_MMTU) {
342 		log(LOG_NOTICE, "nd6_setmtu0: "
343 		    "new link MTU on %s (%lu) is too small for IPv6\n",
344 		    if_name(ifp), (unsigned long)ndi->maxmtu);
345 	}
346 
347 	if (ndi->maxmtu > V_in6_maxmtu)
348 		in6_setmaxmtu(); /* check all interfaces just in case */
349 
350 }
351 
352 void
353 nd6_option_init(void *opt, int icmp6len, union nd_opts *ndopts)
354 {
355 
356 	bzero(ndopts, sizeof(*ndopts));
357 	ndopts->nd_opts_search = (struct nd_opt_hdr *)opt;
358 	ndopts->nd_opts_last
359 		= (struct nd_opt_hdr *)(((u_char *)opt) + icmp6len);
360 
361 	if (icmp6len == 0) {
362 		ndopts->nd_opts_done = 1;
363 		ndopts->nd_opts_search = NULL;
364 	}
365 }
366 
367 /*
368  * Take one ND option.
369  */
370 struct nd_opt_hdr *
371 nd6_option(union nd_opts *ndopts)
372 {
373 	struct nd_opt_hdr *nd_opt;
374 	int olen;
375 
376 	KASSERT(ndopts != NULL, ("%s: ndopts == NULL", __func__));
377 	KASSERT(ndopts->nd_opts_last != NULL, ("%s: uninitialized ndopts",
378 	    __func__));
379 	if (ndopts->nd_opts_search == NULL)
380 		return NULL;
381 	if (ndopts->nd_opts_done)
382 		return NULL;
383 
384 	nd_opt = ndopts->nd_opts_search;
385 
386 	/* make sure nd_opt_len is inside the buffer */
387 	if ((caddr_t)&nd_opt->nd_opt_len >= (caddr_t)ndopts->nd_opts_last) {
388 		bzero(ndopts, sizeof(*ndopts));
389 		return NULL;
390 	}
391 
392 	olen = nd_opt->nd_opt_len << 3;
393 	if (olen == 0) {
394 		/*
395 		 * Message validation requires that all included
396 		 * options have a length that is greater than zero.
397 		 */
398 		bzero(ndopts, sizeof(*ndopts));
399 		return NULL;
400 	}
401 
402 	ndopts->nd_opts_search = (struct nd_opt_hdr *)((caddr_t)nd_opt + olen);
403 	if (ndopts->nd_opts_search > ndopts->nd_opts_last) {
404 		/* option overruns the end of buffer, invalid */
405 		bzero(ndopts, sizeof(*ndopts));
406 		return NULL;
407 	} else if (ndopts->nd_opts_search == ndopts->nd_opts_last) {
408 		/* reached the end of options chain */
409 		ndopts->nd_opts_done = 1;
410 		ndopts->nd_opts_search = NULL;
411 	}
412 	return nd_opt;
413 }
414 
415 /*
416  * Parse multiple ND options.
417  * This function is much easier to use, for ND routines that do not need
418  * multiple options of the same type.
419  */
420 int
421 nd6_options(union nd_opts *ndopts)
422 {
423 	struct nd_opt_hdr *nd_opt;
424 	int i = 0;
425 
426 	KASSERT(ndopts != NULL, ("%s: ndopts == NULL", __func__));
427 	KASSERT(ndopts->nd_opts_last != NULL, ("%s: uninitialized ndopts",
428 	    __func__));
429 	if (ndopts->nd_opts_search == NULL)
430 		return 0;
431 
432 	while (1) {
433 		nd_opt = nd6_option(ndopts);
434 		if (nd_opt == NULL && ndopts->nd_opts_last == NULL) {
435 			/*
436 			 * Message validation requires that all included
437 			 * options have a length that is greater than zero.
438 			 */
439 			ICMP6STAT_INC(icp6s_nd_badopt);
440 			bzero(ndopts, sizeof(*ndopts));
441 			return -1;
442 		}
443 
444 		if (nd_opt == NULL)
445 			goto skip1;
446 
447 		switch (nd_opt->nd_opt_type) {
448 		case ND_OPT_SOURCE_LINKADDR:
449 		case ND_OPT_TARGET_LINKADDR:
450 		case ND_OPT_MTU:
451 		case ND_OPT_REDIRECTED_HEADER:
452 		case ND_OPT_NONCE:
453 			if (ndopts->nd_opt_array[nd_opt->nd_opt_type]) {
454 				nd6log((LOG_INFO,
455 				    "duplicated ND6 option found (type=%d)\n",
456 				    nd_opt->nd_opt_type));
457 				/* XXX bark? */
458 			} else {
459 				ndopts->nd_opt_array[nd_opt->nd_opt_type]
460 					= nd_opt;
461 			}
462 			break;
463 		case ND_OPT_PREFIX_INFORMATION:
464 			if (ndopts->nd_opt_array[nd_opt->nd_opt_type] == 0) {
465 				ndopts->nd_opt_array[nd_opt->nd_opt_type]
466 					= nd_opt;
467 			}
468 			ndopts->nd_opts_pi_end =
469 				(struct nd_opt_prefix_info *)nd_opt;
470 			break;
471 		/* What about ND_OPT_ROUTE_INFO? RFC 4191 */
472 		case ND_OPT_RDNSS:	/* RFC 6106 */
473 		case ND_OPT_DNSSL:	/* RFC 6106 */
474 			/*
475 			 * Silently ignore options we know and do not care about
476 			 * in the kernel.
477 			 */
478 			break;
479 		default:
480 			/*
481 			 * Unknown options must be silently ignored,
482 			 * to accommodate future extension to the protocol.
483 			 */
484 			nd6log((LOG_DEBUG,
485 			    "nd6_options: unsupported option %d - "
486 			    "option ignored\n", nd_opt->nd_opt_type));
487 		}
488 
489 skip1:
490 		i++;
491 		if (i > V_nd6_maxndopt) {
492 			ICMP6STAT_INC(icp6s_nd_toomanyopt);
493 			nd6log((LOG_INFO, "too many loop in nd opt\n"));
494 			break;
495 		}
496 
497 		if (ndopts->nd_opts_done)
498 			break;
499 	}
500 
501 	return 0;
502 }
503 
504 /*
505  * ND6 timer routine to handle ND6 entries
506  */
507 static void
508 nd6_llinfo_settimer_locked(struct llentry *ln, long tick)
509 {
510 	int canceled;
511 
512 	LLE_WLOCK_ASSERT(ln);
513 
514 	if (tick < 0) {
515 		ln->la_expire = 0;
516 		ln->ln_ntick = 0;
517 		canceled = callout_stop(&ln->lle_timer);
518 	} else {
519 		ln->la_expire = time_uptime + tick / hz;
520 		LLE_ADDREF(ln);
521 		if (tick > INT_MAX) {
522 			ln->ln_ntick = tick - INT_MAX;
523 			canceled = callout_reset(&ln->lle_timer, INT_MAX,
524 			    nd6_llinfo_timer, ln);
525 		} else {
526 			ln->ln_ntick = 0;
527 			canceled = callout_reset(&ln->lle_timer, tick,
528 			    nd6_llinfo_timer, ln);
529 		}
530 	}
531 	if (canceled > 0)
532 		LLE_REMREF(ln);
533 }
534 
535 /*
536  * Gets source address of the first packet in hold queue
537  * and stores it in @src.
538  * Returns pointer to @src (if hold queue is not empty) or NULL.
539  *
540  * Set noinline to be dtrace-friendly
541  */
542 static __noinline struct in6_addr *
543 nd6_llinfo_get_holdsrc(struct llentry *ln, struct in6_addr *src)
544 {
545 	struct ip6_hdr hdr;
546 	struct mbuf *m;
547 
548 	if (ln->la_hold == NULL)
549 		return (NULL);
550 
551 	/*
552 	 * assume every packet in la_hold has the same IP header
553 	 */
554 	m = ln->la_hold;
555 	if (sizeof(hdr) > m->m_len)
556 		return (NULL);
557 
558 	m_copydata(m, 0, sizeof(hdr), (caddr_t)&hdr);
559 	*src = hdr.ip6_src;
560 
561 	return (src);
562 }
563 
564 /*
565  * Checks if we need to switch from STALE state.
566  *
567  * RFC 4861 requires switching from STALE to DELAY state
568  * on first packet matching entry, waiting V_nd6_delay and
569  * transition to PROBE state (if upper layer confirmation was
570  * not received).
571  *
572  * This code performs a bit differently:
573  * On packet hit we don't change state (but desired state
574  * can be guessed by control plane). However, after V_nd6_delay
575  * seconds code will transition to PROBE state (so DELAY state
576  * is kinda skipped in most situations).
577  *
578  * Typically, V_nd6_gctimer is bigger than V_nd6_delay, so
579  * we perform the following upon entering STALE state:
580  *
581  * 1) Arm timer to run each V_nd6_delay seconds to make sure that
582  * if packet was transmitted at the start of given interval, we
583  * would be able to switch to PROBE state in V_nd6_delay seconds
584  * as user expects.
585  *
586  * 2) Reschedule timer until original V_nd6_gctimer expires keeping
587  * lle in STALE state (remaining timer value stored in lle_remtime).
588  *
589  * 3) Reschedule timer if packet was transmitted less that V_nd6_delay
590  * seconds ago.
591  *
592  * Returns non-zero value if the entry is still STALE (storing
593  * the next timer interval in @pdelay).
594  *
595  * Returns zero value if original timer expired or we need to switch to
596  * PROBE (store that in @do_switch variable).
597  */
598 static int
599 nd6_is_stale(struct llentry *lle, long *pdelay, int *do_switch)
600 {
601 	int nd_delay, nd_gctimer, r_skip_req;
602 	time_t lle_hittime;
603 	long delay;
604 
605 	*do_switch = 0;
606 	nd_gctimer = V_nd6_gctimer;
607 	nd_delay = V_nd6_delay;
608 
609 	LLE_REQ_LOCK(lle);
610 	r_skip_req = lle->r_skip_req;
611 	lle_hittime = lle->lle_hittime;
612 	LLE_REQ_UNLOCK(lle);
613 
614 	if (r_skip_req > 0) {
615 
616 		/*
617 		 * Nonzero r_skip_req value was set upon entering
618 		 * STALE state. Since value was not changed, no
619 		 * packets were passed using this lle. Ask for
620 		 * timer reschedule and keep STALE state.
621 		 */
622 		delay = (long)(MIN(nd_gctimer, nd_delay));
623 		delay *= hz;
624 		if (lle->lle_remtime > delay)
625 			lle->lle_remtime -= delay;
626 		else {
627 			delay = lle->lle_remtime;
628 			lle->lle_remtime = 0;
629 		}
630 
631 		if (delay == 0) {
632 
633 			/*
634 			 * The original ng6_gctime timeout ended,
635 			 * no more rescheduling.
636 			 */
637 			return (0);
638 		}
639 
640 		*pdelay = delay;
641 		return (1);
642 	}
643 
644 	/*
645 	 * Packet received. Verify timestamp
646 	 */
647 	delay = (long)(time_uptime - lle_hittime);
648 	if (delay < nd_delay) {
649 
650 		/*
651 		 * V_nd6_delay still not passed since the first
652 		 * hit in STALE state.
653 		 * Reshedule timer and return.
654 		 */
655 		*pdelay = (long)(nd_delay - delay) * hz;
656 		return (1);
657 	}
658 
659 	/* Request switching to probe */
660 	*do_switch = 1;
661 	return (0);
662 }
663 
664 
665 /*
666  * Switch @lle state to new state optionally arming timers.
667  *
668  * Set noinline to be dtrace-friendly
669  */
670 __noinline void
671 nd6_llinfo_setstate(struct llentry *lle, int newstate)
672 {
673 	struct ifnet *ifp;
674 	int nd_gctimer, nd_delay;
675 	long delay, remtime;
676 
677 	delay = 0;
678 	remtime = 0;
679 
680 	switch (newstate) {
681 	case ND6_LLINFO_INCOMPLETE:
682 		ifp = lle->lle_tbl->llt_ifp;
683 		delay = (long)ND_IFINFO(ifp)->retrans * hz / 1000;
684 		break;
685 	case ND6_LLINFO_REACHABLE:
686 		if (!ND6_LLINFO_PERMANENT(lle)) {
687 			ifp = lle->lle_tbl->llt_ifp;
688 			delay = (long)ND_IFINFO(ifp)->reachable * hz;
689 		}
690 		break;
691 	case ND6_LLINFO_STALE:
692 
693 		/*
694 		 * Notify fast path that we want to know if any packet
695 		 * is transmitted by setting r_skip_req.
696 		 */
697 		LLE_REQ_LOCK(lle);
698 		lle->r_skip_req = 1;
699 		LLE_REQ_UNLOCK(lle);
700 		nd_delay = V_nd6_delay;
701 		nd_gctimer = V_nd6_gctimer;
702 
703 		delay = (long)(MIN(nd_gctimer, nd_delay)) * hz;
704 		remtime = (long)nd_gctimer * hz - delay;
705 		break;
706 	case ND6_LLINFO_DELAY:
707 		lle->la_asked = 0;
708 		delay = (long)V_nd6_delay * hz;
709 		break;
710 	}
711 
712 	if (delay > 0)
713 		nd6_llinfo_settimer_locked(lle, delay);
714 
715 	lle->lle_remtime = remtime;
716 	lle->ln_state = newstate;
717 }
718 
719 /*
720  * Timer-dependent part of nd state machine.
721  *
722  * Set noinline to be dtrace-friendly
723  */
724 static __noinline void
725 nd6_llinfo_timer(void *arg)
726 {
727 	struct llentry *ln;
728 	struct in6_addr *dst, *pdst, *psrc, src;
729 	struct ifnet *ifp;
730 	struct nd_ifinfo *ndi;
731 	int do_switch, send_ns;
732 	long delay;
733 
734 	KASSERT(arg != NULL, ("%s: arg NULL", __func__));
735 	ln = (struct llentry *)arg;
736 	ifp = lltable_get_ifp(ln->lle_tbl);
737 	CURVNET_SET(ifp->if_vnet);
738 
739 	ND6_RLOCK();
740 	LLE_WLOCK(ln);
741 	if (callout_pending(&ln->lle_timer)) {
742 		/*
743 		 * Here we are a bit odd here in the treatment of
744 		 * active/pending. If the pending bit is set, it got
745 		 * rescheduled before I ran. The active
746 		 * bit we ignore, since if it was stopped
747 		 * in ll_tablefree() and was currently running
748 		 * it would have return 0 so the code would
749 		 * not have deleted it since the callout could
750 		 * not be stopped so we want to go through
751 		 * with the delete here now. If the callout
752 		 * was restarted, the pending bit will be back on and
753 		 * we just want to bail since the callout_reset would
754 		 * return 1 and our reference would have been removed
755 		 * by nd6_llinfo_settimer_locked above since canceled
756 		 * would have been 1.
757 		 */
758 		LLE_WUNLOCK(ln);
759 		ND6_RUNLOCK();
760 		CURVNET_RESTORE();
761 		return;
762 	}
763 	ndi = ND_IFINFO(ifp);
764 	send_ns = 0;
765 	dst = &ln->r_l3addr.addr6;
766 	pdst = dst;
767 
768 	if (ln->ln_ntick > 0) {
769 		if (ln->ln_ntick > INT_MAX) {
770 			ln->ln_ntick -= INT_MAX;
771 			nd6_llinfo_settimer_locked(ln, INT_MAX);
772 		} else {
773 			ln->ln_ntick = 0;
774 			nd6_llinfo_settimer_locked(ln, ln->ln_ntick);
775 		}
776 		goto done;
777 	}
778 
779 	if (ln->la_flags & LLE_STATIC) {
780 		goto done;
781 	}
782 
783 	if (ln->la_flags & LLE_DELETED) {
784 		nd6_free(&ln, 0);
785 		goto done;
786 	}
787 
788 	switch (ln->ln_state) {
789 	case ND6_LLINFO_INCOMPLETE:
790 		if (ln->la_asked < V_nd6_mmaxtries) {
791 			ln->la_asked++;
792 			send_ns = 1;
793 			/* Send NS to multicast address */
794 			pdst = NULL;
795 		} else {
796 			struct mbuf *m = ln->la_hold;
797 			if (m) {
798 				struct mbuf *m0;
799 
800 				/*
801 				 * assuming every packet in la_hold has the
802 				 * same IP header.  Send error after unlock.
803 				 */
804 				m0 = m->m_nextpkt;
805 				m->m_nextpkt = NULL;
806 				ln->la_hold = m0;
807 				clear_llinfo_pqueue(ln);
808 			}
809 			nd6_free(&ln, 0);
810 			if (m != NULL)
811 				icmp6_error2(m, ICMP6_DST_UNREACH,
812 				    ICMP6_DST_UNREACH_ADDR, 0, ifp);
813 		}
814 		break;
815 	case ND6_LLINFO_REACHABLE:
816 		if (!ND6_LLINFO_PERMANENT(ln))
817 			nd6_llinfo_setstate(ln, ND6_LLINFO_STALE);
818 		break;
819 
820 	case ND6_LLINFO_STALE:
821 		if (nd6_is_stale(ln, &delay, &do_switch) != 0) {
822 
823 			/*
824 			 * No packet has used this entry and GC timeout
825 			 * has not been passed. Reshedule timer and
826 			 * return.
827 			 */
828 			nd6_llinfo_settimer_locked(ln, delay);
829 			break;
830 		}
831 
832 		if (do_switch == 0) {
833 
834 			/*
835 			 * GC timer has ended and entry hasn't been used.
836 			 * Run Garbage collector (RFC 4861, 5.3)
837 			 */
838 			if (!ND6_LLINFO_PERMANENT(ln))
839 				nd6_free(&ln, 1);
840 			break;
841 		}
842 
843 		/* Entry has been used AND delay timer has ended. */
844 
845 		/* FALLTHROUGH */
846 
847 	case ND6_LLINFO_DELAY:
848 		if (ndi && (ndi->flags & ND6_IFF_PERFORMNUD) != 0) {
849 			/* We need NUD */
850 			ln->la_asked = 1;
851 			nd6_llinfo_setstate(ln, ND6_LLINFO_PROBE);
852 			send_ns = 1;
853 		} else
854 			nd6_llinfo_setstate(ln, ND6_LLINFO_STALE); /* XXX */
855 		break;
856 	case ND6_LLINFO_PROBE:
857 		if (ln->la_asked < V_nd6_umaxtries) {
858 			ln->la_asked++;
859 			send_ns = 1;
860 		} else {
861 			nd6_free(&ln, 0);
862 		}
863 		break;
864 	default:
865 		panic("%s: paths in a dark night can be confusing: %d",
866 		    __func__, ln->ln_state);
867 	}
868 done:
869 	if (ln != NULL)
870 		ND6_RUNLOCK();
871 	if (send_ns != 0) {
872 		nd6_llinfo_settimer_locked(ln, (long)ndi->retrans * hz / 1000);
873 		psrc = nd6_llinfo_get_holdsrc(ln, &src);
874 		LLE_FREE_LOCKED(ln);
875 		ln = NULL;
876 		nd6_ns_output(ifp, psrc, pdst, dst, NULL);
877 	}
878 
879 	if (ln != NULL)
880 		LLE_FREE_LOCKED(ln);
881 	CURVNET_RESTORE();
882 }
883 
884 
885 /*
886  * ND6 timer routine to expire default route list and prefix list
887  */
888 void
889 nd6_timer(void *arg)
890 {
891 	CURVNET_SET((struct vnet *) arg);
892 	struct nd_drhead drq;
893 	struct nd_defrouter *dr, *ndr;
894 	struct nd_prefix *pr, *npr;
895 	struct in6_ifaddr *ia6, *nia6;
896 
897 	callout_reset(&V_nd6_timer_ch, V_nd6_prune * hz,
898 	    nd6_timer, curvnet);
899 
900 	TAILQ_INIT(&drq);
901 
902 	/* expire default router list */
903 	ND6_WLOCK();
904 	TAILQ_FOREACH_SAFE(dr, &V_nd_defrouter, dr_entry, ndr)
905 		if (dr->expire && dr->expire < time_uptime)
906 			defrouter_unlink(dr, &drq);
907 	ND6_WUNLOCK();
908 
909 	while ((dr = TAILQ_FIRST(&drq)) != NULL) {
910 		TAILQ_REMOVE(&drq, dr, dr_entry);
911 		defrouter_del(dr);
912 	}
913 
914 	/*
915 	 * expire interface addresses.
916 	 * in the past the loop was inside prefix expiry processing.
917 	 * However, from a stricter speci-confrmance standpoint, we should
918 	 * rather separate address lifetimes and prefix lifetimes.
919 	 *
920 	 * XXXRW: in6_ifaddrhead locking.
921 	 */
922   addrloop:
923 	TAILQ_FOREACH_SAFE(ia6, &V_in6_ifaddrhead, ia_link, nia6) {
924 		/* check address lifetime */
925 		if (IFA6_IS_INVALID(ia6)) {
926 			int regen = 0;
927 
928 			/*
929 			 * If the expiring address is temporary, try
930 			 * regenerating a new one.  This would be useful when
931 			 * we suspended a laptop PC, then turned it on after a
932 			 * period that could invalidate all temporary
933 			 * addresses.  Although we may have to restart the
934 			 * loop (see below), it must be after purging the
935 			 * address.  Otherwise, we'd see an infinite loop of
936 			 * regeneration.
937 			 */
938 			if (V_ip6_use_tempaddr &&
939 			    (ia6->ia6_flags & IN6_IFF_TEMPORARY) != 0) {
940 				if (regen_tmpaddr(ia6) == 0)
941 					regen = 1;
942 			}
943 
944 			in6_purgeaddr(&ia6->ia_ifa);
945 
946 			if (regen)
947 				goto addrloop; /* XXX: see below */
948 		} else if (IFA6_IS_DEPRECATED(ia6)) {
949 			int oldflags = ia6->ia6_flags;
950 
951 			ia6->ia6_flags |= IN6_IFF_DEPRECATED;
952 
953 			/*
954 			 * If a temporary address has just become deprecated,
955 			 * regenerate a new one if possible.
956 			 */
957 			if (V_ip6_use_tempaddr &&
958 			    (ia6->ia6_flags & IN6_IFF_TEMPORARY) != 0 &&
959 			    (oldflags & IN6_IFF_DEPRECATED) == 0) {
960 
961 				if (regen_tmpaddr(ia6) == 0) {
962 					/*
963 					 * A new temporary address is
964 					 * generated.
965 					 * XXX: this means the address chain
966 					 * has changed while we are still in
967 					 * the loop.  Although the change
968 					 * would not cause disaster (because
969 					 * it's not a deletion, but an
970 					 * addition,) we'd rather restart the
971 					 * loop just for safety.  Or does this
972 					 * significantly reduce performance??
973 					 */
974 					goto addrloop;
975 				}
976 			}
977 		} else if ((ia6->ia6_flags & IN6_IFF_TENTATIVE) != 0) {
978 			/*
979 			 * Schedule DAD for a tentative address.  This happens
980 			 * if the interface was down or not running
981 			 * when the address was configured.
982 			 */
983 			int delay;
984 
985 			delay = arc4random() %
986 			    (MAX_RTR_SOLICITATION_DELAY * hz);
987 			nd6_dad_start((struct ifaddr *)ia6, delay);
988 		} else {
989 			/*
990 			 * Check status of the interface.  If it is down,
991 			 * mark the address as tentative for future DAD.
992 			 */
993 			if ((ia6->ia_ifp->if_flags & IFF_UP) == 0 ||
994 			    (ia6->ia_ifp->if_drv_flags & IFF_DRV_RUNNING)
995 				== 0 ||
996 			    (ND_IFINFO(ia6->ia_ifp)->flags &
997 				ND6_IFF_IFDISABLED) != 0) {
998 				ia6->ia6_flags &= ~IN6_IFF_DUPLICATED;
999 				ia6->ia6_flags |= IN6_IFF_TENTATIVE;
1000 			}
1001 			/*
1002 			 * A new RA might have made a deprecated address
1003 			 * preferred.
1004 			 */
1005 			ia6->ia6_flags &= ~IN6_IFF_DEPRECATED;
1006 		}
1007 	}
1008 
1009 	/* expire prefix list */
1010 	LIST_FOREACH_SAFE(pr, &V_nd_prefix, ndpr_entry, npr) {
1011 		/*
1012 		 * check prefix lifetime.
1013 		 * since pltime is just for autoconf, pltime processing for
1014 		 * prefix is not necessary.
1015 		 */
1016 		if (pr->ndpr_vltime != ND6_INFINITE_LIFETIME &&
1017 		    time_uptime - pr->ndpr_lastupdate > pr->ndpr_vltime) {
1018 
1019 			/*
1020 			 * address expiration and prefix expiration are
1021 			 * separate.  NEVER perform in6_purgeaddr here.
1022 			 */
1023 			prelist_remove(pr);
1024 		}
1025 	}
1026 	CURVNET_RESTORE();
1027 }
1028 
1029 /*
1030  * ia6 - deprecated/invalidated temporary address
1031  */
1032 static int
1033 regen_tmpaddr(struct in6_ifaddr *ia6)
1034 {
1035 	struct ifaddr *ifa;
1036 	struct ifnet *ifp;
1037 	struct in6_ifaddr *public_ifa6 = NULL;
1038 
1039 	ifp = ia6->ia_ifa.ifa_ifp;
1040 	IF_ADDR_RLOCK(ifp);
1041 	TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1042 		struct in6_ifaddr *it6;
1043 
1044 		if (ifa->ifa_addr->sa_family != AF_INET6)
1045 			continue;
1046 
1047 		it6 = (struct in6_ifaddr *)ifa;
1048 
1049 		/* ignore no autoconf addresses. */
1050 		if ((it6->ia6_flags & IN6_IFF_AUTOCONF) == 0)
1051 			continue;
1052 
1053 		/* ignore autoconf addresses with different prefixes. */
1054 		if (it6->ia6_ndpr == NULL || it6->ia6_ndpr != ia6->ia6_ndpr)
1055 			continue;
1056 
1057 		/*
1058 		 * Now we are looking at an autoconf address with the same
1059 		 * prefix as ours.  If the address is temporary and is still
1060 		 * preferred, do not create another one.  It would be rare, but
1061 		 * could happen, for example, when we resume a laptop PC after
1062 		 * a long period.
1063 		 */
1064 		if ((it6->ia6_flags & IN6_IFF_TEMPORARY) != 0 &&
1065 		    !IFA6_IS_DEPRECATED(it6)) {
1066 			public_ifa6 = NULL;
1067 			break;
1068 		}
1069 
1070 		/*
1071 		 * This is a public autoconf address that has the same prefix
1072 		 * as ours.  If it is preferred, keep it.  We can't break the
1073 		 * loop here, because there may be a still-preferred temporary
1074 		 * address with the prefix.
1075 		 */
1076 		if (!IFA6_IS_DEPRECATED(it6))
1077 			public_ifa6 = it6;
1078 	}
1079 	if (public_ifa6 != NULL)
1080 		ifa_ref(&public_ifa6->ia_ifa);
1081 	IF_ADDR_RUNLOCK(ifp);
1082 
1083 	if (public_ifa6 != NULL) {
1084 		int e;
1085 
1086 		if ((e = in6_tmpifadd(public_ifa6, 0, 0)) != 0) {
1087 			ifa_free(&public_ifa6->ia_ifa);
1088 			log(LOG_NOTICE, "regen_tmpaddr: failed to create a new"
1089 			    " tmp addr,errno=%d\n", e);
1090 			return (-1);
1091 		}
1092 		ifa_free(&public_ifa6->ia_ifa);
1093 		return (0);
1094 	}
1095 
1096 	return (-1);
1097 }
1098 
1099 /*
1100  * Nuke neighbor cache/prefix/default router management table, right before
1101  * ifp goes away.
1102  */
1103 void
1104 nd6_purge(struct ifnet *ifp)
1105 {
1106 	struct nd_drhead drq;
1107 	struct nd_defrouter *dr, *ndr;
1108 	struct nd_prefix *pr, *npr;
1109 
1110 	TAILQ_INIT(&drq);
1111 
1112 	/*
1113 	 * Nuke default router list entries toward ifp.
1114 	 * We defer removal of default router list entries that is installed
1115 	 * in the routing table, in order to keep additional side effects as
1116 	 * small as possible.
1117 	 */
1118 	ND6_WLOCK();
1119 	TAILQ_FOREACH_SAFE(dr, &V_nd_defrouter, dr_entry, ndr) {
1120 		if (dr->installed)
1121 			continue;
1122 		if (dr->ifp == ifp)
1123 			defrouter_unlink(dr, &drq);
1124 	}
1125 
1126 	TAILQ_FOREACH_SAFE(dr, &V_nd_defrouter, dr_entry, ndr) {
1127 		if (!dr->installed)
1128 			continue;
1129 		if (dr->ifp == ifp)
1130 			defrouter_unlink(dr, &drq);
1131 	}
1132 	ND6_WUNLOCK();
1133 
1134 	while ((dr = TAILQ_FIRST(&drq)) != NULL) {
1135 		TAILQ_REMOVE(&drq, dr, dr_entry);
1136 		defrouter_del(dr);
1137 	}
1138 
1139 	/* Nuke prefix list entries toward ifp */
1140 	LIST_FOREACH_SAFE(pr, &V_nd_prefix, ndpr_entry, npr) {
1141 		if (pr->ndpr_ifp == ifp) {
1142 			/*
1143 			 * Because if_detach() does *not* release prefixes
1144 			 * while purging addresses the reference count will
1145 			 * still be above zero. We therefore reset it to
1146 			 * make sure that the prefix really gets purged.
1147 			 */
1148 			pr->ndpr_refcnt = 0;
1149 
1150 			/*
1151 			 * Previously, pr->ndpr_addr is removed as well,
1152 			 * but I strongly believe we don't have to do it.
1153 			 * nd6_purge() is only called from in6_ifdetach(),
1154 			 * which removes all the associated interface addresses
1155 			 * by itself.
1156 			 * (jinmei@kame.net 20010129)
1157 			 */
1158 			prelist_remove(pr);
1159 		}
1160 	}
1161 
1162 	/* cancel default outgoing interface setting */
1163 	if (V_nd6_defifindex == ifp->if_index)
1164 		nd6_setdefaultiface(0);
1165 
1166 	if (ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) {
1167 		/* Refresh default router list. */
1168 		defrouter_select();
1169 	}
1170 
1171 	/* XXXXX
1172 	 * We do not nuke the neighbor cache entries here any more
1173 	 * because the neighbor cache is kept in if_afdata[AF_INET6].
1174 	 * nd6_purge() is invoked by in6_ifdetach() which is called
1175 	 * from if_detach() where everything gets purged. So let
1176 	 * in6_domifdetach() do the actual L2 table purging work.
1177 	 */
1178 }
1179 
1180 /*
1181  * the caller acquires and releases the lock on the lltbls
1182  * Returns the llentry locked
1183  */
1184 struct llentry *
1185 nd6_lookup(const struct in6_addr *addr6, int flags, struct ifnet *ifp)
1186 {
1187 	struct sockaddr_in6 sin6;
1188 	struct llentry *ln;
1189 
1190 	bzero(&sin6, sizeof(sin6));
1191 	sin6.sin6_len = sizeof(struct sockaddr_in6);
1192 	sin6.sin6_family = AF_INET6;
1193 	sin6.sin6_addr = *addr6;
1194 
1195 	IF_AFDATA_LOCK_ASSERT(ifp);
1196 
1197 	ln = lla_lookup(LLTABLE6(ifp), flags, (struct sockaddr *)&sin6);
1198 
1199 	return (ln);
1200 }
1201 
1202 struct llentry *
1203 nd6_alloc(const struct in6_addr *addr6, int flags, struct ifnet *ifp)
1204 {
1205 	struct sockaddr_in6 sin6;
1206 	struct llentry *ln;
1207 
1208 	bzero(&sin6, sizeof(sin6));
1209 	sin6.sin6_len = sizeof(struct sockaddr_in6);
1210 	sin6.sin6_family = AF_INET6;
1211 	sin6.sin6_addr = *addr6;
1212 
1213 	ln = lltable_alloc_entry(LLTABLE6(ifp), 0, (struct sockaddr *)&sin6);
1214 	if (ln != NULL)
1215 		ln->ln_state = ND6_LLINFO_NOSTATE;
1216 
1217 	return (ln);
1218 }
1219 
1220 /*
1221  * Test whether a given IPv6 address is a neighbor or not, ignoring
1222  * the actual neighbor cache.  The neighbor cache is ignored in order
1223  * to not reenter the routing code from within itself.
1224  */
1225 static int
1226 nd6_is_new_addr_neighbor(const struct sockaddr_in6 *addr, struct ifnet *ifp)
1227 {
1228 	struct nd_prefix *pr;
1229 	struct ifaddr *dstaddr;
1230 	struct rt_addrinfo info;
1231 	struct sockaddr_in6 rt_key;
1232 	struct sockaddr *dst6;
1233 	int fibnum;
1234 
1235 	/*
1236 	 * A link-local address is always a neighbor.
1237 	 * XXX: a link does not necessarily specify a single interface.
1238 	 */
1239 	if (IN6_IS_ADDR_LINKLOCAL(&addr->sin6_addr)) {
1240 		struct sockaddr_in6 sin6_copy;
1241 		u_int32_t zone;
1242 
1243 		/*
1244 		 * We need sin6_copy since sa6_recoverscope() may modify the
1245 		 * content (XXX).
1246 		 */
1247 		sin6_copy = *addr;
1248 		if (sa6_recoverscope(&sin6_copy))
1249 			return (0); /* XXX: should be impossible */
1250 		if (in6_setscope(&sin6_copy.sin6_addr, ifp, &zone))
1251 			return (0);
1252 		if (sin6_copy.sin6_scope_id == zone)
1253 			return (1);
1254 		else
1255 			return (0);
1256 	}
1257 
1258 	bzero(&rt_key, sizeof(rt_key));
1259 	bzero(&info, sizeof(info));
1260 	info.rti_info[RTAX_DST] = (struct sockaddr *)&rt_key;
1261 
1262 	/* Always use the default FIB here. XXME - why? */
1263 	fibnum = RT_DEFAULT_FIB;
1264 
1265 	/*
1266 	 * If the address matches one of our addresses,
1267 	 * it should be a neighbor.
1268 	 * If the address matches one of our on-link prefixes, it should be a
1269 	 * neighbor.
1270 	 */
1271 	LIST_FOREACH(pr, &V_nd_prefix, ndpr_entry) {
1272 		if (pr->ndpr_ifp != ifp)
1273 			continue;
1274 
1275 		if (!(pr->ndpr_stateflags & NDPRF_ONLINK)) {
1276 
1277 			/* Always use the default FIB here. */
1278 			dst6 = (struct sockaddr *)&pr->ndpr_prefix;
1279 
1280 			/* Restore length field before retrying lookup */
1281 			rt_key.sin6_len = sizeof(rt_key);
1282 			if (rib_lookup_info(fibnum, dst6, 0, 0, &info) != 0)
1283 				continue;
1284 			/*
1285 			 * This is the case where multiple interfaces
1286 			 * have the same prefix, but only one is installed
1287 			 * into the routing table and that prefix entry
1288 			 * is not the one being examined here. In the case
1289 			 * where RADIX_MPATH is enabled, multiple route
1290 			 * entries (of the same rt_key value) will be
1291 			 * installed because the interface addresses all
1292 			 * differ.
1293 			 */
1294 			if (!IN6_ARE_ADDR_EQUAL(&pr->ndpr_prefix.sin6_addr,
1295 			       &rt_key.sin6_addr))
1296 				continue;
1297 		}
1298 
1299 		if (IN6_ARE_MASKED_ADDR_EQUAL(&pr->ndpr_prefix.sin6_addr,
1300 		    &addr->sin6_addr, &pr->ndpr_mask))
1301 			return (1);
1302 	}
1303 
1304 	/*
1305 	 * If the address is assigned on the node of the other side of
1306 	 * a p2p interface, the address should be a neighbor.
1307 	 */
1308 	dstaddr = ifa_ifwithdstaddr((const struct sockaddr *)addr, RT_ALL_FIBS);
1309 	if (dstaddr != NULL) {
1310 		if (dstaddr->ifa_ifp == ifp) {
1311 			ifa_free(dstaddr);
1312 			return (1);
1313 		}
1314 		ifa_free(dstaddr);
1315 	}
1316 
1317 	/*
1318 	 * If the default router list is empty, all addresses are regarded
1319 	 * as on-link, and thus, as a neighbor.
1320 	 */
1321 	if (ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV &&
1322 	    TAILQ_EMPTY(&V_nd_defrouter) &&
1323 	    V_nd6_defifindex == ifp->if_index) {
1324 		return (1);
1325 	}
1326 
1327 	return (0);
1328 }
1329 
1330 
1331 /*
1332  * Detect if a given IPv6 address identifies a neighbor on a given link.
1333  * XXX: should take care of the destination of a p2p link?
1334  */
1335 int
1336 nd6_is_addr_neighbor(const struct sockaddr_in6 *addr, struct ifnet *ifp)
1337 {
1338 	struct llentry *lle;
1339 	int rc = 0;
1340 
1341 	IF_AFDATA_UNLOCK_ASSERT(ifp);
1342 	if (nd6_is_new_addr_neighbor(addr, ifp))
1343 		return (1);
1344 
1345 	/*
1346 	 * Even if the address matches none of our addresses, it might be
1347 	 * in the neighbor cache.
1348 	 */
1349 	IF_AFDATA_RLOCK(ifp);
1350 	if ((lle = nd6_lookup(&addr->sin6_addr, 0, ifp)) != NULL) {
1351 		LLE_RUNLOCK(lle);
1352 		rc = 1;
1353 	}
1354 	IF_AFDATA_RUNLOCK(ifp);
1355 	return (rc);
1356 }
1357 
1358 /*
1359  * Free an nd6 llinfo entry.
1360  * Since the function would cause significant changes in the kernel, DO NOT
1361  * make it global, unless you have a strong reason for the change, and are sure
1362  * that the change is safe.
1363  *
1364  * Set noinline to be dtrace-friendly
1365  */
1366 static __noinline void
1367 nd6_free(struct llentry **lnp, int gc)
1368 {
1369 	struct ifnet *ifp;
1370 	struct llentry *ln;
1371 	struct nd_defrouter *dr;
1372 
1373 	ln = *lnp;
1374 	*lnp = NULL;
1375 
1376 	LLE_WLOCK_ASSERT(ln);
1377 	ND6_RLOCK_ASSERT();
1378 
1379 	ifp = lltable_get_ifp(ln->lle_tbl);
1380 	if ((ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) != 0)
1381 		dr = defrouter_lookup_locked(&ln->r_l3addr.addr6, ifp);
1382 	else
1383 		dr = NULL;
1384 	ND6_RUNLOCK();
1385 
1386 	if ((ln->la_flags & LLE_DELETED) == 0)
1387 		EVENTHANDLER_INVOKE(lle_event, ln, LLENTRY_EXPIRED);
1388 
1389 	/*
1390 	 * we used to have pfctlinput(PRC_HOSTDEAD) here.
1391 	 * even though it is not harmful, it was not really necessary.
1392 	 */
1393 
1394 	/* cancel timer */
1395 	nd6_llinfo_settimer_locked(ln, -1);
1396 
1397 	if (ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) {
1398 		if (dr != NULL && dr->expire &&
1399 		    ln->ln_state == ND6_LLINFO_STALE && gc) {
1400 			/*
1401 			 * If the reason for the deletion is just garbage
1402 			 * collection, and the neighbor is an active default
1403 			 * router, do not delete it.  Instead, reset the GC
1404 			 * timer using the router's lifetime.
1405 			 * Simply deleting the entry would affect default
1406 			 * router selection, which is not necessarily a good
1407 			 * thing, especially when we're using router preference
1408 			 * values.
1409 			 * XXX: the check for ln_state would be redundant,
1410 			 *      but we intentionally keep it just in case.
1411 			 */
1412 			if (dr->expire > time_uptime)
1413 				nd6_llinfo_settimer_locked(ln,
1414 				    (dr->expire - time_uptime) * hz);
1415 			else
1416 				nd6_llinfo_settimer_locked(ln,
1417 				    (long)V_nd6_gctimer * hz);
1418 
1419 			LLE_REMREF(ln);
1420 			LLE_WUNLOCK(ln);
1421 			defrouter_rele(dr);
1422 			return;
1423 		}
1424 
1425 		if (dr) {
1426 			/*
1427 			 * Unreachablity of a router might affect the default
1428 			 * router selection and on-link detection of advertised
1429 			 * prefixes.
1430 			 */
1431 
1432 			/*
1433 			 * Temporarily fake the state to choose a new default
1434 			 * router and to perform on-link determination of
1435 			 * prefixes correctly.
1436 			 * Below the state will be set correctly,
1437 			 * or the entry itself will be deleted.
1438 			 */
1439 			ln->ln_state = ND6_LLINFO_INCOMPLETE;
1440 		}
1441 
1442 		if (ln->ln_router || dr) {
1443 
1444 			/*
1445 			 * We need to unlock to avoid a LOR with rt6_flush() with the
1446 			 * rnh and for the calls to pfxlist_onlink_check() and
1447 			 * defrouter_select() in the block further down for calls
1448 			 * into nd6_lookup().  We still hold a ref.
1449 			 */
1450 			LLE_WUNLOCK(ln);
1451 
1452 			/*
1453 			 * rt6_flush must be called whether or not the neighbor
1454 			 * is in the Default Router List.
1455 			 * See a corresponding comment in nd6_na_input().
1456 			 */
1457 			rt6_flush(&ln->r_l3addr.addr6, ifp);
1458 		}
1459 
1460 		if (dr) {
1461 			/*
1462 			 * Since defrouter_select() does not affect the
1463 			 * on-link determination and MIP6 needs the check
1464 			 * before the default router selection, we perform
1465 			 * the check now.
1466 			 */
1467 			pfxlist_onlink_check();
1468 
1469 			/*
1470 			 * Refresh default router list.
1471 			 */
1472 			defrouter_select();
1473 		}
1474 
1475 		/*
1476 		 * If this entry was added by an on-link redirect, remove the
1477 		 * corresponding host route.
1478 		 */
1479 		if (ln->la_flags & LLE_REDIRECT)
1480 			nd6_free_redirect(ln);
1481 
1482 		if (ln->ln_router || dr)
1483 			LLE_WLOCK(ln);
1484 	}
1485 
1486 	/*
1487 	 * Save to unlock. We still hold an extra reference and will not
1488 	 * free(9) in llentry_free() if someone else holds one as well.
1489 	 */
1490 	LLE_WUNLOCK(ln);
1491 	IF_AFDATA_LOCK(ifp);
1492 	LLE_WLOCK(ln);
1493 	/* Guard against race with other llentry_free(). */
1494 	if (ln->la_flags & LLE_LINKED) {
1495 		/* Remove callout reference */
1496 		LLE_REMREF(ln);
1497 		lltable_unlink_entry(ln->lle_tbl, ln);
1498 	}
1499 	IF_AFDATA_UNLOCK(ifp);
1500 
1501 	llentry_free(ln);
1502 	if (dr != NULL)
1503 		defrouter_rele(dr);
1504 }
1505 
1506 static int
1507 nd6_isdynrte(const struct rtentry *rt, void *xap)
1508 {
1509 
1510 	if (rt->rt_flags == (RTF_UP | RTF_HOST | RTF_DYNAMIC))
1511 		return (1);
1512 
1513 	return (0);
1514 }
1515 /*
1516  * Remove the rtentry for the given llentry,
1517  * both of which were installed by a redirect.
1518  */
1519 static void
1520 nd6_free_redirect(const struct llentry *ln)
1521 {
1522 	int fibnum;
1523 	struct sockaddr_in6 sin6;
1524 	struct rt_addrinfo info;
1525 
1526 	lltable_fill_sa_entry(ln, (struct sockaddr *)&sin6);
1527 	memset(&info, 0, sizeof(info));
1528 	info.rti_info[RTAX_DST] = (struct sockaddr *)&sin6;
1529 	info.rti_filter = nd6_isdynrte;
1530 
1531 	for (fibnum = 0; fibnum < rt_numfibs; fibnum++)
1532 		rtrequest1_fib(RTM_DELETE, &info, NULL, fibnum);
1533 }
1534 
1535 /*
1536  * Rejuvenate this function for routing operations related
1537  * processing.
1538  */
1539 void
1540 nd6_rtrequest(int req, struct rtentry *rt, struct rt_addrinfo *info)
1541 {
1542 	struct sockaddr_in6 *gateway;
1543 	struct nd_defrouter *dr;
1544 	struct ifnet *ifp;
1545 
1546 	gateway = (struct sockaddr_in6 *)rt->rt_gateway;
1547 	ifp = rt->rt_ifp;
1548 
1549 	switch (req) {
1550 	case RTM_ADD:
1551 		break;
1552 
1553 	case RTM_DELETE:
1554 		if (!ifp)
1555 			return;
1556 		/*
1557 		 * Only indirect routes are interesting.
1558 		 */
1559 		if ((rt->rt_flags & RTF_GATEWAY) == 0)
1560 			return;
1561 		/*
1562 		 * check for default route
1563 		 */
1564 		if (IN6_ARE_ADDR_EQUAL(&in6addr_any,
1565 		    &SIN6(rt_key(rt))->sin6_addr)) {
1566 			dr = defrouter_lookup(&gateway->sin6_addr, ifp);
1567 			if (dr != NULL) {
1568 				dr->installed = 0;
1569 				defrouter_rele(dr);
1570 			}
1571 		}
1572 		break;
1573 	}
1574 }
1575 
1576 
1577 int
1578 nd6_ioctl(u_long cmd, caddr_t data, struct ifnet *ifp)
1579 {
1580 	struct in6_ndireq *ndi = (struct in6_ndireq *)data;
1581 	struct in6_nbrinfo *nbi = (struct in6_nbrinfo *)data;
1582 	struct in6_ndifreq *ndif = (struct in6_ndifreq *)data;
1583 	int error = 0;
1584 
1585 	if (ifp->if_afdata[AF_INET6] == NULL)
1586 		return (EPFNOSUPPORT);
1587 	switch (cmd) {
1588 	case OSIOCGIFINFO_IN6:
1589 #define ND	ndi->ndi
1590 		/* XXX: old ndp(8) assumes a positive value for linkmtu. */
1591 		bzero(&ND, sizeof(ND));
1592 		ND.linkmtu = IN6_LINKMTU(ifp);
1593 		ND.maxmtu = ND_IFINFO(ifp)->maxmtu;
1594 		ND.basereachable = ND_IFINFO(ifp)->basereachable;
1595 		ND.reachable = ND_IFINFO(ifp)->reachable;
1596 		ND.retrans = ND_IFINFO(ifp)->retrans;
1597 		ND.flags = ND_IFINFO(ifp)->flags;
1598 		ND.recalctm = ND_IFINFO(ifp)->recalctm;
1599 		ND.chlim = ND_IFINFO(ifp)->chlim;
1600 		break;
1601 	case SIOCGIFINFO_IN6:
1602 		ND = *ND_IFINFO(ifp);
1603 		break;
1604 	case SIOCSIFINFO_IN6:
1605 		/*
1606 		 * used to change host variables from userland.
1607 		 * intended for a use on router to reflect RA configurations.
1608 		 */
1609 		/* 0 means 'unspecified' */
1610 		if (ND.linkmtu != 0) {
1611 			if (ND.linkmtu < IPV6_MMTU ||
1612 			    ND.linkmtu > IN6_LINKMTU(ifp)) {
1613 				error = EINVAL;
1614 				break;
1615 			}
1616 			ND_IFINFO(ifp)->linkmtu = ND.linkmtu;
1617 		}
1618 
1619 		if (ND.basereachable != 0) {
1620 			int obasereachable = ND_IFINFO(ifp)->basereachable;
1621 
1622 			ND_IFINFO(ifp)->basereachable = ND.basereachable;
1623 			if (ND.basereachable != obasereachable)
1624 				ND_IFINFO(ifp)->reachable =
1625 				    ND_COMPUTE_RTIME(ND.basereachable);
1626 		}
1627 		if (ND.retrans != 0)
1628 			ND_IFINFO(ifp)->retrans = ND.retrans;
1629 		if (ND.chlim != 0)
1630 			ND_IFINFO(ifp)->chlim = ND.chlim;
1631 		/* FALLTHROUGH */
1632 	case SIOCSIFINFO_FLAGS:
1633 	{
1634 		struct ifaddr *ifa;
1635 		struct in6_ifaddr *ia;
1636 
1637 		if ((ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED) &&
1638 		    !(ND.flags & ND6_IFF_IFDISABLED)) {
1639 			/* ifdisabled 1->0 transision */
1640 
1641 			/*
1642 			 * If the interface is marked as ND6_IFF_IFDISABLED and
1643 			 * has an link-local address with IN6_IFF_DUPLICATED,
1644 			 * do not clear ND6_IFF_IFDISABLED.
1645 			 * See RFC 4862, Section 5.4.5.
1646 			 */
1647 			IF_ADDR_RLOCK(ifp);
1648 			TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1649 				if (ifa->ifa_addr->sa_family != AF_INET6)
1650 					continue;
1651 				ia = (struct in6_ifaddr *)ifa;
1652 				if ((ia->ia6_flags & IN6_IFF_DUPLICATED) &&
1653 				    IN6_IS_ADDR_LINKLOCAL(IA6_IN6(ia)))
1654 					break;
1655 			}
1656 			IF_ADDR_RUNLOCK(ifp);
1657 
1658 			if (ifa != NULL) {
1659 				/* LLA is duplicated. */
1660 				ND.flags |= ND6_IFF_IFDISABLED;
1661 				log(LOG_ERR, "Cannot enable an interface"
1662 				    " with a link-local address marked"
1663 				    " duplicate.\n");
1664 			} else {
1665 				ND_IFINFO(ifp)->flags &= ~ND6_IFF_IFDISABLED;
1666 				if (ifp->if_flags & IFF_UP)
1667 					in6_if_up(ifp);
1668 			}
1669 		} else if (!(ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED) &&
1670 			    (ND.flags & ND6_IFF_IFDISABLED)) {
1671 			/* ifdisabled 0->1 transision */
1672 			/* Mark all IPv6 address as tentative. */
1673 
1674 			ND_IFINFO(ifp)->flags |= ND6_IFF_IFDISABLED;
1675 			if (V_ip6_dad_count > 0 &&
1676 			    (ND_IFINFO(ifp)->flags & ND6_IFF_NO_DAD) == 0) {
1677 				IF_ADDR_RLOCK(ifp);
1678 				TAILQ_FOREACH(ifa, &ifp->if_addrhead,
1679 				    ifa_link) {
1680 					if (ifa->ifa_addr->sa_family !=
1681 					    AF_INET6)
1682 						continue;
1683 					ia = (struct in6_ifaddr *)ifa;
1684 					ia->ia6_flags |= IN6_IFF_TENTATIVE;
1685 				}
1686 				IF_ADDR_RUNLOCK(ifp);
1687 			}
1688 		}
1689 
1690 		if (ND.flags & ND6_IFF_AUTO_LINKLOCAL) {
1691 			if (!(ND_IFINFO(ifp)->flags & ND6_IFF_AUTO_LINKLOCAL)) {
1692 				/* auto_linklocal 0->1 transision */
1693 
1694 				/* If no link-local address on ifp, configure */
1695 				ND_IFINFO(ifp)->flags |= ND6_IFF_AUTO_LINKLOCAL;
1696 				in6_ifattach(ifp, NULL);
1697 			} else if (!(ND.flags & ND6_IFF_IFDISABLED) &&
1698 			    ifp->if_flags & IFF_UP) {
1699 				/*
1700 				 * When the IF already has
1701 				 * ND6_IFF_AUTO_LINKLOCAL, no link-local
1702 				 * address is assigned, and IFF_UP, try to
1703 				 * assign one.
1704 				 */
1705 				IF_ADDR_RLOCK(ifp);
1706 				TAILQ_FOREACH(ifa, &ifp->if_addrhead,
1707 				    ifa_link) {
1708 					if (ifa->ifa_addr->sa_family !=
1709 					    AF_INET6)
1710 						continue;
1711 					ia = (struct in6_ifaddr *)ifa;
1712 					if (IN6_IS_ADDR_LINKLOCAL(IA6_IN6(ia)))
1713 						break;
1714 				}
1715 				IF_ADDR_RUNLOCK(ifp);
1716 				if (ifa != NULL)
1717 					/* No LLA is configured. */
1718 					in6_ifattach(ifp, NULL);
1719 			}
1720 		}
1721 	}
1722 		ND_IFINFO(ifp)->flags = ND.flags;
1723 		break;
1724 #undef ND
1725 	case SIOCSNDFLUSH_IN6:	/* XXX: the ioctl name is confusing... */
1726 		/* sync kernel routing table with the default router list */
1727 		defrouter_reset();
1728 		defrouter_select();
1729 		break;
1730 	case SIOCSPFXFLUSH_IN6:
1731 	{
1732 		/* flush all the prefix advertised by routers */
1733 		struct nd_prefix *pr, *next;
1734 
1735 		LIST_FOREACH_SAFE(pr, &V_nd_prefix, ndpr_entry, next) {
1736 			struct in6_ifaddr *ia, *ia_next;
1737 
1738 			if (IN6_IS_ADDR_LINKLOCAL(&pr->ndpr_prefix.sin6_addr))
1739 				continue; /* XXX */
1740 
1741 			/* do we really have to remove addresses as well? */
1742 			/* XXXRW: in6_ifaddrhead locking. */
1743 			TAILQ_FOREACH_SAFE(ia, &V_in6_ifaddrhead, ia_link,
1744 			    ia_next) {
1745 				if ((ia->ia6_flags & IN6_IFF_AUTOCONF) == 0)
1746 					continue;
1747 
1748 				if (ia->ia6_ndpr == pr)
1749 					in6_purgeaddr(&ia->ia_ifa);
1750 			}
1751 			prelist_remove(pr);
1752 		}
1753 		break;
1754 	}
1755 	case SIOCSRTRFLUSH_IN6:
1756 	{
1757 		/* flush all the default routers */
1758 		struct nd_drhead drq;
1759 		struct nd_defrouter *dr;
1760 
1761 		TAILQ_INIT(&drq);
1762 
1763 		defrouter_reset();
1764 
1765 		ND6_WLOCK();
1766 		while ((dr = TAILQ_FIRST(&V_nd_defrouter)) != NULL)
1767 			defrouter_unlink(dr, &drq);
1768 		ND6_WUNLOCK();
1769 		while ((dr = TAILQ_FIRST(&drq)) != NULL) {
1770 			TAILQ_REMOVE(&drq, dr, dr_entry);
1771 			defrouter_del(dr);
1772 		}
1773 
1774 		defrouter_select();
1775 		break;
1776 	}
1777 	case SIOCGNBRINFO_IN6:
1778 	{
1779 		struct llentry *ln;
1780 		struct in6_addr nb_addr = nbi->addr; /* make local for safety */
1781 
1782 		if ((error = in6_setscope(&nb_addr, ifp, NULL)) != 0)
1783 			return (error);
1784 
1785 		IF_AFDATA_RLOCK(ifp);
1786 		ln = nd6_lookup(&nb_addr, 0, ifp);
1787 		IF_AFDATA_RUNLOCK(ifp);
1788 
1789 		if (ln == NULL) {
1790 			error = EINVAL;
1791 			break;
1792 		}
1793 		nbi->state = ln->ln_state;
1794 		nbi->asked = ln->la_asked;
1795 		nbi->isrouter = ln->ln_router;
1796 		if (ln->la_expire == 0)
1797 			nbi->expire = 0;
1798 		else
1799 			nbi->expire = ln->la_expire + ln->lle_remtime / hz +
1800 			    (time_second - time_uptime);
1801 		LLE_RUNLOCK(ln);
1802 		break;
1803 	}
1804 	case SIOCGDEFIFACE_IN6:	/* XXX: should be implemented as a sysctl? */
1805 		ndif->ifindex = V_nd6_defifindex;
1806 		break;
1807 	case SIOCSDEFIFACE_IN6:	/* XXX: should be implemented as a sysctl? */
1808 		return (nd6_setdefaultiface(ndif->ifindex));
1809 	}
1810 	return (error);
1811 }
1812 
1813 /*
1814  * Calculates new isRouter value based on provided parameters and
1815  * returns it.
1816  */
1817 static int
1818 nd6_is_router(int type, int code, int is_new, int old_addr, int new_addr,
1819     int ln_router)
1820 {
1821 
1822 	/*
1823 	 * ICMP6 type dependent behavior.
1824 	 *
1825 	 * NS: clear IsRouter if new entry
1826 	 * RS: clear IsRouter
1827 	 * RA: set IsRouter if there's lladdr
1828 	 * redir: clear IsRouter if new entry
1829 	 *
1830 	 * RA case, (1):
1831 	 * The spec says that we must set IsRouter in the following cases:
1832 	 * - If lladdr exist, set IsRouter.  This means (1-5).
1833 	 * - If it is old entry (!newentry), set IsRouter.  This means (7).
1834 	 * So, based on the spec, in (1-5) and (7) cases we must set IsRouter.
1835 	 * A quetion arises for (1) case.  (1) case has no lladdr in the
1836 	 * neighbor cache, this is similar to (6).
1837 	 * This case is rare but we figured that we MUST NOT set IsRouter.
1838 	 *
1839 	 *   is_new  old_addr new_addr 	    NS  RS  RA	redir
1840 	 *							D R
1841 	 *	0	n	n	(1)	c   ?     s
1842 	 *	0	y	n	(2)	c   s     s
1843 	 *	0	n	y	(3)	c   s     s
1844 	 *	0	y	y	(4)	c   s     s
1845 	 *	0	y	y	(5)	c   s     s
1846 	 *	1	--	n	(6) c	c	c s
1847 	 *	1	--	y	(7) c	c   s	c s
1848 	 *
1849 	 *					(c=clear s=set)
1850 	 */
1851 	switch (type & 0xff) {
1852 	case ND_NEIGHBOR_SOLICIT:
1853 		/*
1854 		 * New entry must have is_router flag cleared.
1855 		 */
1856 		if (is_new)					/* (6-7) */
1857 			ln_router = 0;
1858 		break;
1859 	case ND_REDIRECT:
1860 		/*
1861 		 * If the icmp is a redirect to a better router, always set the
1862 		 * is_router flag.  Otherwise, if the entry is newly created,
1863 		 * clear the flag.  [RFC 2461, sec 8.3]
1864 		 */
1865 		if (code == ND_REDIRECT_ROUTER)
1866 			ln_router = 1;
1867 		else {
1868 			if (is_new)				/* (6-7) */
1869 				ln_router = 0;
1870 		}
1871 		break;
1872 	case ND_ROUTER_SOLICIT:
1873 		/*
1874 		 * is_router flag must always be cleared.
1875 		 */
1876 		ln_router = 0;
1877 		break;
1878 	case ND_ROUTER_ADVERT:
1879 		/*
1880 		 * Mark an entry with lladdr as a router.
1881 		 */
1882 		if ((!is_new && (old_addr || new_addr)) ||	/* (2-5) */
1883 		    (is_new && new_addr)) {			/* (7) */
1884 			ln_router = 1;
1885 		}
1886 		break;
1887 	}
1888 
1889 	return (ln_router);
1890 }
1891 
1892 /*
1893  * Create neighbor cache entry and cache link-layer address,
1894  * on reception of inbound ND6 packets.  (RS/RA/NS/redirect)
1895  *
1896  * type - ICMP6 type
1897  * code - type dependent information
1898  *
1899  */
1900 void
1901 nd6_cache_lladdr(struct ifnet *ifp, struct in6_addr *from, char *lladdr,
1902     int lladdrlen, int type, int code)
1903 {
1904 	struct llentry *ln = NULL, *ln_tmp;
1905 	int is_newentry;
1906 	int do_update;
1907 	int olladdr;
1908 	int llchange;
1909 	int flags;
1910 	uint16_t router = 0;
1911 	struct sockaddr_in6 sin6;
1912 	struct mbuf *chain = NULL;
1913 	u_char linkhdr[LLE_MAX_LINKHDR];
1914 	size_t linkhdrsize;
1915 	int lladdr_off;
1916 
1917 	IF_AFDATA_UNLOCK_ASSERT(ifp);
1918 
1919 	KASSERT(ifp != NULL, ("%s: ifp == NULL", __func__));
1920 	KASSERT(from != NULL, ("%s: from == NULL", __func__));
1921 
1922 	/* nothing must be updated for unspecified address */
1923 	if (IN6_IS_ADDR_UNSPECIFIED(from))
1924 		return;
1925 
1926 	/*
1927 	 * Validation about ifp->if_addrlen and lladdrlen must be done in
1928 	 * the caller.
1929 	 *
1930 	 * XXX If the link does not have link-layer adderss, what should
1931 	 * we do? (ifp->if_addrlen == 0)
1932 	 * Spec says nothing in sections for RA, RS and NA.  There's small
1933 	 * description on it in NS section (RFC 2461 7.2.3).
1934 	 */
1935 	flags = lladdr ? LLE_EXCLUSIVE : 0;
1936 	IF_AFDATA_RLOCK(ifp);
1937 	ln = nd6_lookup(from, flags, ifp);
1938 	IF_AFDATA_RUNLOCK(ifp);
1939 	is_newentry = 0;
1940 	if (ln == NULL) {
1941 		flags |= LLE_EXCLUSIVE;
1942 		ln = nd6_alloc(from, 0, ifp);
1943 		if (ln == NULL)
1944 			return;
1945 
1946 		/*
1947 		 * Since we already know all the data for the new entry,
1948 		 * fill it before insertion.
1949 		 */
1950 		if (lladdr != NULL) {
1951 			linkhdrsize = sizeof(linkhdr);
1952 			if (lltable_calc_llheader(ifp, AF_INET6, lladdr,
1953 			    linkhdr, &linkhdrsize, &lladdr_off) != 0)
1954 				return;
1955 			lltable_set_entry_addr(ifp, ln, linkhdr, linkhdrsize,
1956 			    lladdr_off);
1957 		}
1958 
1959 		IF_AFDATA_WLOCK(ifp);
1960 		LLE_WLOCK(ln);
1961 		/* Prefer any existing lle over newly-created one */
1962 		ln_tmp = nd6_lookup(from, LLE_EXCLUSIVE, ifp);
1963 		if (ln_tmp == NULL)
1964 			lltable_link_entry(LLTABLE6(ifp), ln);
1965 		IF_AFDATA_WUNLOCK(ifp);
1966 		if (ln_tmp == NULL) {
1967 			/* No existing lle, mark as new entry (6,7) */
1968 			is_newentry = 1;
1969 			nd6_llinfo_setstate(ln, ND6_LLINFO_STALE);
1970 			if (lladdr != NULL)	/* (7) */
1971 				EVENTHANDLER_INVOKE(lle_event, ln,
1972 				    LLENTRY_RESOLVED);
1973 		} else {
1974 			lltable_free_entry(LLTABLE6(ifp), ln);
1975 			ln = ln_tmp;
1976 			ln_tmp = NULL;
1977 		}
1978 	}
1979 	/* do nothing if static ndp is set */
1980 	if ((ln->la_flags & LLE_STATIC)) {
1981 		if (flags & LLE_EXCLUSIVE)
1982 			LLE_WUNLOCK(ln);
1983 		else
1984 			LLE_RUNLOCK(ln);
1985 		return;
1986 	}
1987 
1988 	olladdr = (ln->la_flags & LLE_VALID) ? 1 : 0;
1989 	if (olladdr && lladdr) {
1990 		llchange = bcmp(lladdr, ln->ll_addr,
1991 		    ifp->if_addrlen);
1992 	} else if (!olladdr && lladdr)
1993 		llchange = 1;
1994 	else
1995 		llchange = 0;
1996 
1997 	/*
1998 	 * newentry olladdr  lladdr  llchange	(*=record)
1999 	 *	0	n	n	--	(1)
2000 	 *	0	y	n	--	(2)
2001 	 *	0	n	y	y	(3) * STALE
2002 	 *	0	y	y	n	(4) *
2003 	 *	0	y	y	y	(5) * STALE
2004 	 *	1	--	n	--	(6)   NOSTATE(= PASSIVE)
2005 	 *	1	--	y	--	(7) * STALE
2006 	 */
2007 
2008 	do_update = 0;
2009 	if (is_newentry == 0 && llchange != 0) {
2010 		do_update = 1;	/* (3,5) */
2011 
2012 		/*
2013 		 * Record source link-layer address
2014 		 * XXX is it dependent to ifp->if_type?
2015 		 */
2016 		linkhdrsize = sizeof(linkhdr);
2017 		if (lltable_calc_llheader(ifp, AF_INET6, lladdr,
2018 		    linkhdr, &linkhdrsize, &lladdr_off) != 0)
2019 			return;
2020 
2021 		if (lltable_try_set_entry_addr(ifp, ln, linkhdr, linkhdrsize,
2022 		    lladdr_off) == 0) {
2023 			/* Entry was deleted */
2024 			return;
2025 		}
2026 
2027 		nd6_llinfo_setstate(ln, ND6_LLINFO_STALE);
2028 
2029 		EVENTHANDLER_INVOKE(lle_event, ln, LLENTRY_RESOLVED);
2030 
2031 		if (ln->la_hold != NULL)
2032 			nd6_grab_holdchain(ln, &chain, &sin6);
2033 	}
2034 
2035 	/* Calculates new router status */
2036 	router = nd6_is_router(type, code, is_newentry, olladdr,
2037 	    lladdr != NULL ? 1 : 0, ln->ln_router);
2038 
2039 	ln->ln_router = router;
2040 	/* Mark non-router redirects with special flag */
2041 	if ((type & 0xFF) == ND_REDIRECT && code != ND_REDIRECT_ROUTER)
2042 		ln->la_flags |= LLE_REDIRECT;
2043 
2044 	if (flags & LLE_EXCLUSIVE)
2045 		LLE_WUNLOCK(ln);
2046 	else
2047 		LLE_RUNLOCK(ln);
2048 
2049 	if (chain != NULL)
2050 		nd6_flush_holdchain(ifp, ifp, chain, &sin6);
2051 
2052 	/*
2053 	 * When the link-layer address of a router changes, select the
2054 	 * best router again.  In particular, when the neighbor entry is newly
2055 	 * created, it might affect the selection policy.
2056 	 * Question: can we restrict the first condition to the "is_newentry"
2057 	 * case?
2058 	 * XXX: when we hear an RA from a new router with the link-layer
2059 	 * address option, defrouter_select() is called twice, since
2060 	 * defrtrlist_update called the function as well.  However, I believe
2061 	 * we can compromise the overhead, since it only happens the first
2062 	 * time.
2063 	 * XXX: although defrouter_select() should not have a bad effect
2064 	 * for those are not autoconfigured hosts, we explicitly avoid such
2065 	 * cases for safety.
2066 	 */
2067 	if ((do_update || is_newentry) && router &&
2068 	    ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) {
2069 		/*
2070 		 * guaranteed recursion
2071 		 */
2072 		defrouter_select();
2073 	}
2074 }
2075 
2076 static void
2077 nd6_slowtimo(void *arg)
2078 {
2079 	CURVNET_SET((struct vnet *) arg);
2080 	struct nd_ifinfo *nd6if;
2081 	struct ifnet *ifp;
2082 
2083 	callout_reset(&V_nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
2084 	    nd6_slowtimo, curvnet);
2085 	IFNET_RLOCK_NOSLEEP();
2086 	TAILQ_FOREACH(ifp, &V_ifnet, if_link) {
2087 		if (ifp->if_afdata[AF_INET6] == NULL)
2088 			continue;
2089 		nd6if = ND_IFINFO(ifp);
2090 		if (nd6if->basereachable && /* already initialized */
2091 		    (nd6if->recalctm -= ND6_SLOWTIMER_INTERVAL) <= 0) {
2092 			/*
2093 			 * Since reachable time rarely changes by router
2094 			 * advertisements, we SHOULD insure that a new random
2095 			 * value gets recomputed at least once every few hours.
2096 			 * (RFC 2461, 6.3.4)
2097 			 */
2098 			nd6if->recalctm = V_nd6_recalc_reachtm_interval;
2099 			nd6if->reachable = ND_COMPUTE_RTIME(nd6if->basereachable);
2100 		}
2101 	}
2102 	IFNET_RUNLOCK_NOSLEEP();
2103 	CURVNET_RESTORE();
2104 }
2105 
2106 void
2107 nd6_grab_holdchain(struct llentry *ln, struct mbuf **chain,
2108     struct sockaddr_in6 *sin6)
2109 {
2110 
2111 	LLE_WLOCK_ASSERT(ln);
2112 
2113 	*chain = ln->la_hold;
2114 	ln->la_hold = NULL;
2115 	lltable_fill_sa_entry(ln, (struct sockaddr *)sin6);
2116 
2117 	if (ln->ln_state == ND6_LLINFO_STALE) {
2118 
2119 		/*
2120 		 * The first time we send a packet to a
2121 		 * neighbor whose entry is STALE, we have
2122 		 * to change the state to DELAY and a sets
2123 		 * a timer to expire in DELAY_FIRST_PROBE_TIME
2124 		 * seconds to ensure do neighbor unreachability
2125 		 * detection on expiration.
2126 		 * (RFC 2461 7.3.3)
2127 		 */
2128 		nd6_llinfo_setstate(ln, ND6_LLINFO_DELAY);
2129 	}
2130 }
2131 
2132 int
2133 nd6_output_ifp(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *m,
2134     struct sockaddr_in6 *dst, struct route *ro)
2135 {
2136 	int error;
2137 	int ip6len;
2138 	struct ip6_hdr *ip6;
2139 	struct m_tag *mtag;
2140 
2141 #ifdef MAC
2142 	mac_netinet6_nd6_send(ifp, m);
2143 #endif
2144 
2145 	/*
2146 	 * If called from nd6_ns_output() (NS), nd6_na_output() (NA),
2147 	 * icmp6_redirect_output() (REDIRECT) or from rip6_output() (RS, RA
2148 	 * as handled by rtsol and rtadvd), mbufs will be tagged for SeND
2149 	 * to be diverted to user space.  When re-injected into the kernel,
2150 	 * send_output() will directly dispatch them to the outgoing interface.
2151 	 */
2152 	if (send_sendso_input_hook != NULL) {
2153 		mtag = m_tag_find(m, PACKET_TAG_ND_OUTGOING, NULL);
2154 		if (mtag != NULL) {
2155 			ip6 = mtod(m, struct ip6_hdr *);
2156 			ip6len = sizeof(struct ip6_hdr) + ntohs(ip6->ip6_plen);
2157 			/* Use the SEND socket */
2158 			error = send_sendso_input_hook(m, ifp, SND_OUT,
2159 			    ip6len);
2160 			/* -1 == no app on SEND socket */
2161 			if (error == 0 || error != -1)
2162 			    return (error);
2163 		}
2164 	}
2165 
2166 	m_clrprotoflags(m);	/* Avoid confusing lower layers. */
2167 	IP_PROBE(send, NULL, NULL, mtod(m, struct ip6_hdr *), ifp, NULL,
2168 	    mtod(m, struct ip6_hdr *));
2169 
2170 	if ((ifp->if_flags & IFF_LOOPBACK) == 0)
2171 		origifp = ifp;
2172 
2173 	error = (*ifp->if_output)(origifp, m, (struct sockaddr *)dst, ro);
2174 	return (error);
2175 }
2176 
2177 /*
2178  * Lookup link headerfor @sa_dst address. Stores found
2179  * data in @desten buffer. Copy of lle ln_flags can be also
2180  * saved in @pflags if @pflags is non-NULL.
2181  *
2182  * If destination LLE does not exists or lle state modification
2183  * is required, call "slow" version.
2184  *
2185  * Return values:
2186  * - 0 on success (address copied to buffer).
2187  * - EWOULDBLOCK (no local error, but address is still unresolved)
2188  * - other errors (alloc failure, etc)
2189  */
2190 int
2191 nd6_resolve(struct ifnet *ifp, int is_gw, struct mbuf *m,
2192     const struct sockaddr *sa_dst, u_char *desten, uint32_t *pflags)
2193 {
2194 	struct llentry *ln = NULL;
2195 	const struct sockaddr_in6 *dst6;
2196 
2197 	if (pflags != NULL)
2198 		*pflags = 0;
2199 
2200 	dst6 = (const struct sockaddr_in6 *)sa_dst;
2201 
2202 	/* discard the packet if IPv6 operation is disabled on the interface */
2203 	if ((ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED)) {
2204 		m_freem(m);
2205 		return (ENETDOWN); /* better error? */
2206 	}
2207 
2208 	if (m != NULL && m->m_flags & M_MCAST) {
2209 		switch (ifp->if_type) {
2210 		case IFT_ETHER:
2211 		case IFT_FDDI:
2212 		case IFT_L2VLAN:
2213 		case IFT_IEEE80211:
2214 		case IFT_BRIDGE:
2215 		case IFT_ISO88025:
2216 			ETHER_MAP_IPV6_MULTICAST(&dst6->sin6_addr,
2217 						 desten);
2218 			return (0);
2219 		default:
2220 			m_freem(m);
2221 			return (EAFNOSUPPORT);
2222 		}
2223 	}
2224 
2225 	IF_AFDATA_RLOCK(ifp);
2226 	ln = nd6_lookup(&dst6->sin6_addr, LLE_UNLOCKED, ifp);
2227 	if (ln != NULL && (ln->r_flags & RLLE_VALID) != 0) {
2228 		/* Entry found, let's copy lle info */
2229 		bcopy(ln->r_linkdata, desten, ln->r_hdrlen);
2230 		if (pflags != NULL)
2231 			*pflags = LLE_VALID | (ln->r_flags & RLLE_IFADDR);
2232 		/* Check if we have feedback request from nd6 timer */
2233 		if (ln->r_skip_req != 0) {
2234 			LLE_REQ_LOCK(ln);
2235 			ln->r_skip_req = 0; /* Notify that entry was used */
2236 			ln->lle_hittime = time_uptime;
2237 			LLE_REQ_UNLOCK(ln);
2238 		}
2239 		IF_AFDATA_RUNLOCK(ifp);
2240 		return (0);
2241 	}
2242 	IF_AFDATA_RUNLOCK(ifp);
2243 
2244 	return (nd6_resolve_slow(ifp, 0, m, dst6, desten, pflags));
2245 }
2246 
2247 
2248 /*
2249  * Do L2 address resolution for @sa_dst address. Stores found
2250  * address in @desten buffer. Copy of lle ln_flags can be also
2251  * saved in @pflags if @pflags is non-NULL.
2252  *
2253  * Heavy version.
2254  * Function assume that destination LLE does not exist,
2255  * is invalid or stale, so LLE_EXCLUSIVE lock needs to be acquired.
2256  *
2257  * Set noinline to be dtrace-friendly
2258  */
2259 static __noinline int
2260 nd6_resolve_slow(struct ifnet *ifp, int flags, struct mbuf *m,
2261     const struct sockaddr_in6 *dst, u_char *desten, uint32_t *pflags)
2262 {
2263 	struct llentry *lle = NULL, *lle_tmp;
2264 	struct in6_addr *psrc, src;
2265 	int send_ns, ll_len;
2266 	char *lladdr;
2267 
2268 	/*
2269 	 * Address resolution or Neighbor Unreachability Detection
2270 	 * for the next hop.
2271 	 * At this point, the destination of the packet must be a unicast
2272 	 * or an anycast address(i.e. not a multicast).
2273 	 */
2274 	if (lle == NULL) {
2275 		IF_AFDATA_RLOCK(ifp);
2276 		lle = nd6_lookup(&dst->sin6_addr, LLE_EXCLUSIVE, ifp);
2277 		IF_AFDATA_RUNLOCK(ifp);
2278 		if ((lle == NULL) && nd6_is_addr_neighbor(dst, ifp))  {
2279 			/*
2280 			 * Since nd6_is_addr_neighbor() internally calls nd6_lookup(),
2281 			 * the condition below is not very efficient.  But we believe
2282 			 * it is tolerable, because this should be a rare case.
2283 			 */
2284 			lle = nd6_alloc(&dst->sin6_addr, 0, ifp);
2285 			if (lle == NULL) {
2286 				char ip6buf[INET6_ADDRSTRLEN];
2287 				log(LOG_DEBUG,
2288 				    "nd6_output: can't allocate llinfo for %s "
2289 				    "(ln=%p)\n",
2290 				    ip6_sprintf(ip6buf, &dst->sin6_addr), lle);
2291 				m_freem(m);
2292 				return (ENOBUFS);
2293 			}
2294 
2295 			IF_AFDATA_WLOCK(ifp);
2296 			LLE_WLOCK(lle);
2297 			/* Prefer any existing entry over newly-created one */
2298 			lle_tmp = nd6_lookup(&dst->sin6_addr, LLE_EXCLUSIVE, ifp);
2299 			if (lle_tmp == NULL)
2300 				lltable_link_entry(LLTABLE6(ifp), lle);
2301 			IF_AFDATA_WUNLOCK(ifp);
2302 			if (lle_tmp != NULL) {
2303 				lltable_free_entry(LLTABLE6(ifp), lle);
2304 				lle = lle_tmp;
2305 				lle_tmp = NULL;
2306 			}
2307 		}
2308 	}
2309 	if (lle == NULL) {
2310 		if (!(ND_IFINFO(ifp)->flags & ND6_IFF_PERFORMNUD)) {
2311 			m_freem(m);
2312 			return (ENOBUFS);
2313 		}
2314 
2315 		if (m != NULL)
2316 			m_freem(m);
2317 		return (ENOBUFS);
2318 	}
2319 
2320 	LLE_WLOCK_ASSERT(lle);
2321 
2322 	/*
2323 	 * The first time we send a packet to a neighbor whose entry is
2324 	 * STALE, we have to change the state to DELAY and a sets a timer to
2325 	 * expire in DELAY_FIRST_PROBE_TIME seconds to ensure do
2326 	 * neighbor unreachability detection on expiration.
2327 	 * (RFC 2461 7.3.3)
2328 	 */
2329 	if (lle->ln_state == ND6_LLINFO_STALE)
2330 		nd6_llinfo_setstate(lle, ND6_LLINFO_DELAY);
2331 
2332 	/*
2333 	 * If the neighbor cache entry has a state other than INCOMPLETE
2334 	 * (i.e. its link-layer address is already resolved), just
2335 	 * send the packet.
2336 	 */
2337 	if (lle->ln_state > ND6_LLINFO_INCOMPLETE) {
2338 		if (flags & LLE_ADDRONLY) {
2339 			lladdr = lle->ll_addr;
2340 			ll_len = ifp->if_addrlen;
2341 		} else {
2342 			lladdr = lle->r_linkdata;
2343 			ll_len = lle->r_hdrlen;
2344 		}
2345 		bcopy(lladdr, desten, ll_len);
2346 		if (pflags != NULL)
2347 			*pflags = lle->la_flags;
2348 		LLE_WUNLOCK(lle);
2349 		return (0);
2350 	}
2351 
2352 	/*
2353 	 * There is a neighbor cache entry, but no ethernet address
2354 	 * response yet.  Append this latest packet to the end of the
2355 	 * packet queue in the mbuf.  When it exceeds nd6_maxqueuelen,
2356 	 * the oldest packet in the queue will be removed.
2357 	 */
2358 
2359 	if (lle->la_hold != NULL) {
2360 		struct mbuf *m_hold;
2361 		int i;
2362 
2363 		i = 0;
2364 		for (m_hold = lle->la_hold; m_hold; m_hold = m_hold->m_nextpkt){
2365 			i++;
2366 			if (m_hold->m_nextpkt == NULL) {
2367 				m_hold->m_nextpkt = m;
2368 				break;
2369 			}
2370 		}
2371 		while (i >= V_nd6_maxqueuelen) {
2372 			m_hold = lle->la_hold;
2373 			lle->la_hold = lle->la_hold->m_nextpkt;
2374 			m_freem(m_hold);
2375 			i--;
2376 		}
2377 	} else {
2378 		lle->la_hold = m;
2379 	}
2380 
2381 	/*
2382 	 * If there has been no NS for the neighbor after entering the
2383 	 * INCOMPLETE state, send the first solicitation.
2384 	 * Note that for newly-created lle la_asked will be 0,
2385 	 * so we will transition from ND6_LLINFO_NOSTATE to
2386 	 * ND6_LLINFO_INCOMPLETE state here.
2387 	 */
2388 	psrc = NULL;
2389 	send_ns = 0;
2390 	if (lle->la_asked == 0) {
2391 		lle->la_asked++;
2392 		send_ns = 1;
2393 		psrc = nd6_llinfo_get_holdsrc(lle, &src);
2394 
2395 		nd6_llinfo_setstate(lle, ND6_LLINFO_INCOMPLETE);
2396 	}
2397 	LLE_WUNLOCK(lle);
2398 	if (send_ns != 0)
2399 		nd6_ns_output(ifp, psrc, NULL, &dst->sin6_addr, NULL);
2400 
2401 	return (EWOULDBLOCK);
2402 }
2403 
2404 /*
2405  * Do L2 address resolution for @sa_dst address. Stores found
2406  * address in @desten buffer. Copy of lle ln_flags can be also
2407  * saved in @pflags if @pflags is non-NULL.
2408  *
2409  * Return values:
2410  * - 0 on success (address copied to buffer).
2411  * - EWOULDBLOCK (no local error, but address is still unresolved)
2412  * - other errors (alloc failure, etc)
2413  */
2414 int
2415 nd6_resolve_addr(struct ifnet *ifp, int flags, const struct sockaddr *dst,
2416     char *desten, uint32_t *pflags)
2417 {
2418 	int error;
2419 
2420 	flags |= LLE_ADDRONLY;
2421 	error = nd6_resolve_slow(ifp, flags, NULL,
2422 	    (const struct sockaddr_in6 *)dst, desten, pflags);
2423 	return (error);
2424 }
2425 
2426 int
2427 nd6_flush_holdchain(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *chain,
2428     struct sockaddr_in6 *dst)
2429 {
2430 	struct mbuf *m, *m_head;
2431 	struct ifnet *outifp;
2432 	int error = 0;
2433 
2434 	m_head = chain;
2435 	if ((ifp->if_flags & IFF_LOOPBACK) != 0)
2436 		outifp = origifp;
2437 	else
2438 		outifp = ifp;
2439 
2440 	while (m_head) {
2441 		m = m_head;
2442 		m_head = m_head->m_nextpkt;
2443 		error = nd6_output_ifp(ifp, origifp, m, dst, NULL);
2444 	}
2445 
2446 	/*
2447 	 * XXX
2448 	 * note that intermediate errors are blindly ignored
2449 	 */
2450 	return (error);
2451 }
2452 
2453 static int
2454 nd6_need_cache(struct ifnet *ifp)
2455 {
2456 	/*
2457 	 * XXX: we currently do not make neighbor cache on any interface
2458 	 * other than ARCnet, Ethernet, FDDI and GIF.
2459 	 *
2460 	 * RFC2893 says:
2461 	 * - unidirectional tunnels needs no ND
2462 	 */
2463 	switch (ifp->if_type) {
2464 	case IFT_ARCNET:
2465 	case IFT_ETHER:
2466 	case IFT_FDDI:
2467 	case IFT_IEEE1394:
2468 	case IFT_L2VLAN:
2469 	case IFT_IEEE80211:
2470 	case IFT_INFINIBAND:
2471 	case IFT_BRIDGE:
2472 	case IFT_PROPVIRTUAL:
2473 		return (1);
2474 	default:
2475 		return (0);
2476 	}
2477 }
2478 
2479 /*
2480  * Add pernament ND6 link-layer record for given
2481  * interface address.
2482  *
2483  * Very similar to IPv4 arp_ifinit(), but:
2484  * 1) IPv6 DAD is performed in different place
2485  * 2) It is called by IPv6 protocol stack in contrast to
2486  * arp_ifinit() which is typically called in SIOCSIFADDR
2487  * driver ioctl handler.
2488  *
2489  */
2490 int
2491 nd6_add_ifa_lle(struct in6_ifaddr *ia)
2492 {
2493 	struct ifnet *ifp;
2494 	struct llentry *ln, *ln_tmp;
2495 	struct sockaddr *dst;
2496 
2497 	ifp = ia->ia_ifa.ifa_ifp;
2498 	if (nd6_need_cache(ifp) == 0)
2499 		return (0);
2500 
2501 	ia->ia_ifa.ifa_rtrequest = nd6_rtrequest;
2502 	dst = (struct sockaddr *)&ia->ia_addr;
2503 	ln = lltable_alloc_entry(LLTABLE6(ifp), LLE_IFADDR, dst);
2504 	if (ln == NULL)
2505 		return (ENOBUFS);
2506 
2507 	IF_AFDATA_WLOCK(ifp);
2508 	LLE_WLOCK(ln);
2509 	/* Unlink any entry if exists */
2510 	ln_tmp = lla_lookup(LLTABLE6(ifp), LLE_EXCLUSIVE, dst);
2511 	if (ln_tmp != NULL)
2512 		lltable_unlink_entry(LLTABLE6(ifp), ln_tmp);
2513 	lltable_link_entry(LLTABLE6(ifp), ln);
2514 	IF_AFDATA_WUNLOCK(ifp);
2515 
2516 	if (ln_tmp != NULL)
2517 		EVENTHANDLER_INVOKE(lle_event, ln_tmp, LLENTRY_EXPIRED);
2518 	EVENTHANDLER_INVOKE(lle_event, ln, LLENTRY_RESOLVED);
2519 
2520 	LLE_WUNLOCK(ln);
2521 	if (ln_tmp != NULL)
2522 		llentry_free(ln_tmp);
2523 
2524 	return (0);
2525 }
2526 
2527 /*
2528  * Removes either all lle entries for given @ia, or lle
2529  * corresponding to @ia address.
2530  */
2531 void
2532 nd6_rem_ifa_lle(struct in6_ifaddr *ia, int all)
2533 {
2534 	struct sockaddr_in6 mask, addr;
2535 	struct sockaddr *saddr, *smask;
2536 	struct ifnet *ifp;
2537 
2538 	ifp = ia->ia_ifa.ifa_ifp;
2539 	memcpy(&addr, &ia->ia_addr, sizeof(ia->ia_addr));
2540 	memcpy(&mask, &ia->ia_prefixmask, sizeof(ia->ia_prefixmask));
2541 	saddr = (struct sockaddr *)&addr;
2542 	smask = (struct sockaddr *)&mask;
2543 
2544 	if (all != 0)
2545 		lltable_prefix_free(AF_INET6, saddr, smask, LLE_STATIC);
2546 	else
2547 		lltable_delete_addr(LLTABLE6(ifp), LLE_IFADDR, saddr);
2548 }
2549 
2550 static void
2551 clear_llinfo_pqueue(struct llentry *ln)
2552 {
2553 	struct mbuf *m_hold, *m_hold_next;
2554 
2555 	for (m_hold = ln->la_hold; m_hold; m_hold = m_hold_next) {
2556 		m_hold_next = m_hold->m_nextpkt;
2557 		m_freem(m_hold);
2558 	}
2559 
2560 	ln->la_hold = NULL;
2561 }
2562 
2563 static int nd6_sysctl_drlist(SYSCTL_HANDLER_ARGS);
2564 static int nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS);
2565 #ifdef SYSCTL_DECL
2566 SYSCTL_DECL(_net_inet6_icmp6);
2567 #endif
2568 SYSCTL_NODE(_net_inet6_icmp6, ICMPV6CTL_ND6_DRLIST, nd6_drlist,
2569 	CTLFLAG_RD, nd6_sysctl_drlist, "");
2570 SYSCTL_NODE(_net_inet6_icmp6, ICMPV6CTL_ND6_PRLIST, nd6_prlist,
2571 	CTLFLAG_RD, nd6_sysctl_prlist, "");
2572 SYSCTL_INT(_net_inet6_icmp6, ICMPV6CTL_ND6_MAXQLEN, nd6_maxqueuelen,
2573 	CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(nd6_maxqueuelen), 1, "");
2574 SYSCTL_INT(_net_inet6_icmp6, OID_AUTO, nd6_gctimer,
2575 	CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(nd6_gctimer), (60 * 60 * 24), "");
2576 
2577 static int
2578 nd6_sysctl_drlist(SYSCTL_HANDLER_ARGS)
2579 {
2580 	struct in6_defrouter d;
2581 	struct nd_defrouter *dr;
2582 	int error;
2583 
2584 	if (req->newptr != NULL)
2585 		return (EPERM);
2586 
2587 	error = sysctl_wire_old_buffer(req, 0);
2588 	if (error != 0)
2589 		return (error);
2590 
2591 	bzero(&d, sizeof(d));
2592 	d.rtaddr.sin6_family = AF_INET6;
2593 	d.rtaddr.sin6_len = sizeof(d.rtaddr);
2594 
2595 	ND6_RLOCK();
2596 	TAILQ_FOREACH(dr, &V_nd_defrouter, dr_entry) {
2597 		d.rtaddr.sin6_addr = dr->rtaddr;
2598 		error = sa6_recoverscope(&d.rtaddr);
2599 		if (error != 0)
2600 			break;
2601 		d.flags = dr->raflags;
2602 		d.rtlifetime = dr->rtlifetime;
2603 		d.expire = dr->expire + (time_second - time_uptime);
2604 		d.if_index = dr->ifp->if_index;
2605 		error = SYSCTL_OUT(req, &d, sizeof(d));
2606 		if (error != 0)
2607 			break;
2608 	}
2609 	ND6_RUNLOCK();
2610 	return (error);
2611 }
2612 
2613 static int
2614 nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS)
2615 {
2616 	struct in6_prefix p;
2617 	struct sockaddr_in6 s6;
2618 	struct nd_prefix *pr;
2619 	struct nd_pfxrouter *pfr;
2620 	time_t maxexpire;
2621 	int error;
2622 	char ip6buf[INET6_ADDRSTRLEN];
2623 
2624 	if (req->newptr)
2625 		return (EPERM);
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 	/*
2634 	 * XXX locking
2635 	 */
2636 	LIST_FOREACH(pr, &V_nd_prefix, ndpr_entry) {
2637 		p.prefix = pr->ndpr_prefix;
2638 		if (sa6_recoverscope(&p.prefix)) {
2639 			log(LOG_ERR, "scope error in prefix list (%s)\n",
2640 			    ip6_sprintf(ip6buf, &p.prefix.sin6_addr));
2641 			/* XXX: press on... */
2642 		}
2643 		p.raflags = pr->ndpr_raf;
2644 		p.prefixlen = pr->ndpr_plen;
2645 		p.vltime = pr->ndpr_vltime;
2646 		p.pltime = pr->ndpr_pltime;
2647 		p.if_index = pr->ndpr_ifp->if_index;
2648 		if (pr->ndpr_vltime == ND6_INFINITE_LIFETIME)
2649 			p.expire = 0;
2650 		else {
2651 			/* XXX: we assume time_t is signed. */
2652 			maxexpire = (-1) &
2653 			    ~((time_t)1 << ((sizeof(maxexpire) * 8) - 1));
2654 			if (pr->ndpr_vltime < maxexpire - pr->ndpr_lastupdate)
2655 				p.expire = pr->ndpr_lastupdate +
2656 				    pr->ndpr_vltime +
2657 				    (time_second - time_uptime);
2658 			else
2659 				p.expire = maxexpire;
2660 		}
2661 		p.refcnt = pr->ndpr_refcnt;
2662 		p.flags = pr->ndpr_stateflags;
2663 		p.advrtrs = 0;
2664 		LIST_FOREACH(pfr, &pr->ndpr_advrtrs, pfr_entry)
2665 			p.advrtrs++;
2666 		error = SYSCTL_OUT(req, &p, sizeof(p));
2667 		if (error != 0)
2668 			return (error);
2669 		LIST_FOREACH(pfr, &pr->ndpr_advrtrs, pfr_entry) {
2670 			s6.sin6_addr = pfr->router->rtaddr;
2671 			if (sa6_recoverscope(&s6))
2672 				log(LOG_ERR,
2673 				    "scope error in prefix list (%s)\n",
2674 				    ip6_sprintf(ip6buf, &pfr->router->rtaddr));
2675 			error = SYSCTL_OUT(req, &s6, sizeof(s6));
2676 			if (error != 0)
2677 				return (error);
2678 		}
2679 	}
2680 	return (0);
2681 }
2682