xref: /freebsd/sys/netinet6/nd6.c (revision acc1a9ef8333c798c210fa94be6af4d5fe2dd794)
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 
146 static void
147 nd6_lle_event(void *arg __unused, struct llentry *lle, int evt)
148 {
149 	struct rt_addrinfo rtinfo;
150 	struct sockaddr_in6 dst;
151 	struct sockaddr_dl gw;
152 	struct ifnet *ifp;
153 	int type;
154 
155 	LLE_WLOCK_ASSERT(lle);
156 
157 	if (lltable_get_af(lle->lle_tbl) != AF_INET6)
158 		return;
159 
160 	switch (evt) {
161 	case LLENTRY_RESOLVED:
162 		type = RTM_ADD;
163 		KASSERT(lle->la_flags & LLE_VALID,
164 		    ("%s: %p resolved but not valid?", __func__, lle));
165 		break;
166 	case LLENTRY_EXPIRED:
167 		type = RTM_DELETE;
168 		break;
169 	default:
170 		return;
171 	}
172 
173 	ifp = lltable_get_ifp(lle->lle_tbl);
174 
175 	bzero(&dst, sizeof(dst));
176 	bzero(&gw, sizeof(gw));
177 	bzero(&rtinfo, sizeof(rtinfo));
178 	lltable_fill_sa_entry(lle, (struct sockaddr *)&dst);
179 	dst.sin6_scope_id = in6_getscopezone(ifp,
180 	    in6_addrscope(&dst.sin6_addr));
181 	gw.sdl_len = sizeof(struct sockaddr_dl);
182 	gw.sdl_family = AF_LINK;
183 	gw.sdl_alen = ifp->if_addrlen;
184 	gw.sdl_index = ifp->if_index;
185 	gw.sdl_type = ifp->if_type;
186 	if (evt == LLENTRY_RESOLVED)
187 		bcopy(lle->ll_addr, gw.sdl_data, ifp->if_addrlen);
188 	rtinfo.rti_info[RTAX_DST] = (struct sockaddr *)&dst;
189 	rtinfo.rti_info[RTAX_GATEWAY] = (struct sockaddr *)&gw;
190 	rtinfo.rti_addrs = RTA_DST | RTA_GATEWAY;
191 	rt_missmsg_fib(type, &rtinfo, RTF_HOST | RTF_LLDATA | (
192 	    type == RTM_ADD ? RTF_UP: 0), 0, RT_DEFAULT_FIB);
193 }
194 
195 /*
196  * A handler for interface link layer address change event.
197  */
198 static void
199 nd6_iflladdr(void *arg __unused, struct ifnet *ifp)
200 {
201 
202 	lltable_update_ifaddr(LLTABLE6(ifp));
203 }
204 
205 void
206 nd6_init(void)
207 {
208 
209 	rw_init(&V_nd6_lock, "nd6");
210 
211 	LIST_INIT(&V_nd_prefix);
212 
213 	/* initialization of the default router list */
214 	TAILQ_INIT(&V_nd_defrouter);
215 
216 	/* start timer */
217 	callout_init(&V_nd6_slowtimo_ch, 0);
218 	callout_reset(&V_nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
219 	    nd6_slowtimo, curvnet);
220 
221 	nd6_dad_init();
222 	if (IS_DEFAULT_VNET(curvnet)) {
223 		lle_event_eh = EVENTHANDLER_REGISTER(lle_event, nd6_lle_event,
224 		    NULL, EVENTHANDLER_PRI_ANY);
225 		iflladdr_event_eh = EVENTHANDLER_REGISTER(iflladdr_event,
226 		    nd6_iflladdr, NULL, EVENTHANDLER_PRI_ANY);
227 	}
228 }
229 
230 #ifdef VIMAGE
231 void
232 nd6_destroy()
233 {
234 
235 	callout_drain(&V_nd6_slowtimo_ch);
236 	callout_drain(&V_nd6_timer_ch);
237 	if (IS_DEFAULT_VNET(curvnet)) {
238 		EVENTHANDLER_DEREGISTER(lle_event, lle_event_eh);
239 		EVENTHANDLER_DEREGISTER(iflladdr_event, iflladdr_event_eh);
240 	}
241 	rw_destroy(&V_nd6_lock);
242 }
243 #endif
244 
245 struct nd_ifinfo *
246 nd6_ifattach(struct ifnet *ifp)
247 {
248 	struct nd_ifinfo *nd;
249 
250 	nd = malloc(sizeof(*nd), M_IP6NDP, M_WAITOK | M_ZERO);
251 	nd->initialized = 1;
252 
253 	nd->chlim = IPV6_DEFHLIM;
254 	nd->basereachable = REACHABLE_TIME;
255 	nd->reachable = ND_COMPUTE_RTIME(nd->basereachable);
256 	nd->retrans = RETRANS_TIMER;
257 
258 	nd->flags = ND6_IFF_PERFORMNUD;
259 
260 	/* A loopback interface always has ND6_IFF_AUTO_LINKLOCAL.
261 	 * XXXHRS: Clear ND6_IFF_AUTO_LINKLOCAL on an IFT_BRIDGE interface by
262 	 * default regardless of the V_ip6_auto_linklocal configuration to
263 	 * give a reasonable default behavior.
264 	 */
265 	if ((V_ip6_auto_linklocal && ifp->if_type != IFT_BRIDGE) ||
266 	    (ifp->if_flags & IFF_LOOPBACK))
267 		nd->flags |= ND6_IFF_AUTO_LINKLOCAL;
268 	/*
269 	 * A loopback interface does not need to accept RTADV.
270 	 * XXXHRS: Clear ND6_IFF_ACCEPT_RTADV on an IFT_BRIDGE interface by
271 	 * default regardless of the V_ip6_accept_rtadv configuration to
272 	 * prevent the interface from accepting RA messages arrived
273 	 * on one of the member interfaces with ND6_IFF_ACCEPT_RTADV.
274 	 */
275 	if (V_ip6_accept_rtadv &&
276 	    !(ifp->if_flags & IFF_LOOPBACK) &&
277 	    (ifp->if_type != IFT_BRIDGE))
278 			nd->flags |= ND6_IFF_ACCEPT_RTADV;
279 	if (V_ip6_no_radr && !(ifp->if_flags & IFF_LOOPBACK))
280 		nd->flags |= ND6_IFF_NO_RADR;
281 
282 	/* XXX: we cannot call nd6_setmtu since ifp is not fully initialized */
283 	nd6_setmtu0(ifp, nd);
284 
285 	return nd;
286 }
287 
288 void
289 nd6_ifdetach(struct nd_ifinfo *nd)
290 {
291 
292 	free(nd, M_IP6NDP);
293 }
294 
295 /*
296  * Reset ND level link MTU. This function is called when the physical MTU
297  * changes, which means we might have to adjust the ND level MTU.
298  */
299 void
300 nd6_setmtu(struct ifnet *ifp)
301 {
302 	if (ifp->if_afdata[AF_INET6] == NULL)
303 		return;
304 
305 	nd6_setmtu0(ifp, ND_IFINFO(ifp));
306 }
307 
308 /* XXX todo: do not maintain copy of ifp->if_mtu in ndi->maxmtu */
309 void
310 nd6_setmtu0(struct ifnet *ifp, struct nd_ifinfo *ndi)
311 {
312 	u_int32_t omaxmtu;
313 
314 	omaxmtu = ndi->maxmtu;
315 
316 	switch (ifp->if_type) {
317 	case IFT_ARCNET:
318 		ndi->maxmtu = MIN(ARC_PHDS_MAXMTU, ifp->if_mtu); /* RFC2497 */
319 		break;
320 	case IFT_FDDI:
321 		ndi->maxmtu = MIN(FDDIIPMTU, ifp->if_mtu); /* RFC2467 */
322 		break;
323 	case IFT_ISO88025:
324 		 ndi->maxmtu = MIN(ISO88025_MAX_MTU, ifp->if_mtu);
325 		 break;
326 	default:
327 		ndi->maxmtu = ifp->if_mtu;
328 		break;
329 	}
330 
331 	/*
332 	 * Decreasing the interface MTU under IPV6 minimum MTU may cause
333 	 * undesirable situation.  We thus notify the operator of the change
334 	 * explicitly.  The check for omaxmtu is necessary to restrict the
335 	 * log to the case of changing the MTU, not initializing it.
336 	 */
337 	if (omaxmtu >= IPV6_MMTU && ndi->maxmtu < IPV6_MMTU) {
338 		log(LOG_NOTICE, "nd6_setmtu0: "
339 		    "new link MTU on %s (%lu) is too small for IPv6\n",
340 		    if_name(ifp), (unsigned long)ndi->maxmtu);
341 	}
342 
343 	if (ndi->maxmtu > V_in6_maxmtu)
344 		in6_setmaxmtu(); /* check all interfaces just in case */
345 
346 }
347 
348 void
349 nd6_option_init(void *opt, int icmp6len, union nd_opts *ndopts)
350 {
351 
352 	bzero(ndopts, sizeof(*ndopts));
353 	ndopts->nd_opts_search = (struct nd_opt_hdr *)opt;
354 	ndopts->nd_opts_last
355 		= (struct nd_opt_hdr *)(((u_char *)opt) + icmp6len);
356 
357 	if (icmp6len == 0) {
358 		ndopts->nd_opts_done = 1;
359 		ndopts->nd_opts_search = NULL;
360 	}
361 }
362 
363 /*
364  * Take one ND option.
365  */
366 struct nd_opt_hdr *
367 nd6_option(union nd_opts *ndopts)
368 {
369 	struct nd_opt_hdr *nd_opt;
370 	int olen;
371 
372 	KASSERT(ndopts != NULL, ("%s: ndopts == NULL", __func__));
373 	KASSERT(ndopts->nd_opts_last != NULL, ("%s: uninitialized ndopts",
374 	    __func__));
375 	if (ndopts->nd_opts_search == NULL)
376 		return NULL;
377 	if (ndopts->nd_opts_done)
378 		return NULL;
379 
380 	nd_opt = ndopts->nd_opts_search;
381 
382 	/* make sure nd_opt_len is inside the buffer */
383 	if ((caddr_t)&nd_opt->nd_opt_len >= (caddr_t)ndopts->nd_opts_last) {
384 		bzero(ndopts, sizeof(*ndopts));
385 		return NULL;
386 	}
387 
388 	olen = nd_opt->nd_opt_len << 3;
389 	if (olen == 0) {
390 		/*
391 		 * Message validation requires that all included
392 		 * options have a length that is greater than zero.
393 		 */
394 		bzero(ndopts, sizeof(*ndopts));
395 		return NULL;
396 	}
397 
398 	ndopts->nd_opts_search = (struct nd_opt_hdr *)((caddr_t)nd_opt + olen);
399 	if (ndopts->nd_opts_search > ndopts->nd_opts_last) {
400 		/* option overruns the end of buffer, invalid */
401 		bzero(ndopts, sizeof(*ndopts));
402 		return NULL;
403 	} else if (ndopts->nd_opts_search == ndopts->nd_opts_last) {
404 		/* reached the end of options chain */
405 		ndopts->nd_opts_done = 1;
406 		ndopts->nd_opts_search = NULL;
407 	}
408 	return nd_opt;
409 }
410 
411 /*
412  * Parse multiple ND options.
413  * This function is much easier to use, for ND routines that do not need
414  * multiple options of the same type.
415  */
416 int
417 nd6_options(union nd_opts *ndopts)
418 {
419 	struct nd_opt_hdr *nd_opt;
420 	int i = 0;
421 
422 	KASSERT(ndopts != NULL, ("%s: ndopts == NULL", __func__));
423 	KASSERT(ndopts->nd_opts_last != NULL, ("%s: uninitialized ndopts",
424 	    __func__));
425 	if (ndopts->nd_opts_search == NULL)
426 		return 0;
427 
428 	while (1) {
429 		nd_opt = nd6_option(ndopts);
430 		if (nd_opt == NULL && ndopts->nd_opts_last == NULL) {
431 			/*
432 			 * Message validation requires that all included
433 			 * options have a length that is greater than zero.
434 			 */
435 			ICMP6STAT_INC(icp6s_nd_badopt);
436 			bzero(ndopts, sizeof(*ndopts));
437 			return -1;
438 		}
439 
440 		if (nd_opt == NULL)
441 			goto skip1;
442 
443 		switch (nd_opt->nd_opt_type) {
444 		case ND_OPT_SOURCE_LINKADDR:
445 		case ND_OPT_TARGET_LINKADDR:
446 		case ND_OPT_MTU:
447 		case ND_OPT_REDIRECTED_HEADER:
448 		case ND_OPT_NONCE:
449 			if (ndopts->nd_opt_array[nd_opt->nd_opt_type]) {
450 				nd6log((LOG_INFO,
451 				    "duplicated ND6 option found (type=%d)\n",
452 				    nd_opt->nd_opt_type));
453 				/* XXX bark? */
454 			} else {
455 				ndopts->nd_opt_array[nd_opt->nd_opt_type]
456 					= nd_opt;
457 			}
458 			break;
459 		case ND_OPT_PREFIX_INFORMATION:
460 			if (ndopts->nd_opt_array[nd_opt->nd_opt_type] == 0) {
461 				ndopts->nd_opt_array[nd_opt->nd_opt_type]
462 					= nd_opt;
463 			}
464 			ndopts->nd_opts_pi_end =
465 				(struct nd_opt_prefix_info *)nd_opt;
466 			break;
467 		/* What about ND_OPT_ROUTE_INFO? RFC 4191 */
468 		case ND_OPT_RDNSS:	/* RFC 6106 */
469 		case ND_OPT_DNSSL:	/* RFC 6106 */
470 			/*
471 			 * Silently ignore options we know and do not care about
472 			 * in the kernel.
473 			 */
474 			break;
475 		default:
476 			/*
477 			 * Unknown options must be silently ignored,
478 			 * to accomodate future extension to the protocol.
479 			 */
480 			nd6log((LOG_DEBUG,
481 			    "nd6_options: unsupported option %d - "
482 			    "option ignored\n", nd_opt->nd_opt_type));
483 		}
484 
485 skip1:
486 		i++;
487 		if (i > V_nd6_maxndopt) {
488 			ICMP6STAT_INC(icp6s_nd_toomanyopt);
489 			nd6log((LOG_INFO, "too many loop in nd opt\n"));
490 			break;
491 		}
492 
493 		if (ndopts->nd_opts_done)
494 			break;
495 	}
496 
497 	return 0;
498 }
499 
500 /*
501  * ND6 timer routine to handle ND6 entries
502  */
503 static void
504 nd6_llinfo_settimer_locked(struct llentry *ln, long tick)
505 {
506 	int canceled;
507 
508 	LLE_WLOCK_ASSERT(ln);
509 
510 	if (tick < 0) {
511 		ln->la_expire = 0;
512 		ln->ln_ntick = 0;
513 		canceled = callout_stop(&ln->lle_timer);
514 	} else {
515 		ln->la_expire = time_uptime + tick / hz;
516 		LLE_ADDREF(ln);
517 		if (tick > INT_MAX) {
518 			ln->ln_ntick = tick - INT_MAX;
519 			canceled = callout_reset(&ln->lle_timer, INT_MAX,
520 			    nd6_llinfo_timer, ln);
521 		} else {
522 			ln->ln_ntick = 0;
523 			canceled = callout_reset(&ln->lle_timer, tick,
524 			    nd6_llinfo_timer, ln);
525 		}
526 	}
527 	if (canceled > 0)
528 		LLE_REMREF(ln);
529 }
530 
531 /*
532  * Gets source address of the first packet in hold queue
533  * and stores it in @src.
534  * Returns pointer to @src (if hold queue is not empty) or NULL.
535  *
536  * Set noinline to be dtrace-friendly
537  */
538 static __noinline struct in6_addr *
539 nd6_llinfo_get_holdsrc(struct llentry *ln, struct in6_addr *src)
540 {
541 	struct ip6_hdr hdr;
542 	struct mbuf *m;
543 
544 	if (ln->la_hold == NULL)
545 		return (NULL);
546 
547 	/*
548 	 * assume every packet in la_hold has the same IP header
549 	 */
550 	m = ln->la_hold;
551 	if (sizeof(hdr) > m->m_len)
552 		return (NULL);
553 
554 	m_copydata(m, 0, sizeof(hdr), (caddr_t)&hdr);
555 	*src = hdr.ip6_src;
556 
557 	return (src);
558 }
559 
560 /*
561  * Checks if we need to switch from STALE state.
562  *
563  * RFC 4861 requires switching from STALE to DELAY state
564  * on first packet matching entry, waiting V_nd6_delay and
565  * transition to PROBE state (if upper layer confirmation was
566  * not received).
567  *
568  * This code performs a bit differently:
569  * On packet hit we don't change state (but desired state
570  * can be guessed by control plane). However, after V_nd6_delay
571  * seconds code will transition to PROBE state (so DELAY state
572  * is kinda skipped in most situations).
573  *
574  * Typically, V_nd6_gctimer is bigger than V_nd6_delay, so
575  * we perform the following upon entering STALE state:
576  *
577  * 1) Arm timer to run each V_nd6_delay seconds to make sure that
578  * if packet was transmitted at the start of given interval, we
579  * would be able to switch to PROBE state in V_nd6_delay seconds
580  * as user expects.
581  *
582  * 2) Reschedule timer until original V_nd6_gctimer expires keeping
583  * lle in STALE state (remaining timer value stored in lle_remtime).
584  *
585  * 3) Reschedule timer if packet was transmitted less that V_nd6_delay
586  * seconds ago.
587  *
588  * Returns non-zero value if the entry is still STALE (storing
589  * the next timer interval in @pdelay).
590  *
591  * Returns zero value if original timer expired or we need to switch to
592  * PROBE (store that in @do_switch variable).
593  */
594 static int
595 nd6_is_stale(struct llentry *lle, long *pdelay, int *do_switch)
596 {
597 	int nd_delay, nd_gctimer, r_skip_req;
598 	time_t lle_hittime;
599 	long delay;
600 
601 	*do_switch = 0;
602 	nd_gctimer = V_nd6_gctimer;
603 	nd_delay = V_nd6_delay;
604 
605 	LLE_REQ_LOCK(lle);
606 	r_skip_req = lle->r_skip_req;
607 	lle_hittime = lle->lle_hittime;
608 	LLE_REQ_UNLOCK(lle);
609 
610 	if (r_skip_req > 0) {
611 
612 		/*
613 		 * Nonzero r_skip_req value was set upon entering
614 		 * STALE state. Since value was not changed, no
615 		 * packets were passed using this lle. Ask for
616 		 * timer reschedule and keep STALE state.
617 		 */
618 		delay = (long)(MIN(nd_gctimer, nd_delay));
619 		delay *= hz;
620 		if (lle->lle_remtime > delay)
621 			lle->lle_remtime -= delay;
622 		else {
623 			delay = lle->lle_remtime;
624 			lle->lle_remtime = 0;
625 		}
626 
627 		if (delay == 0) {
628 
629 			/*
630 			 * The original ng6_gctime timeout ended,
631 			 * no more rescheduling.
632 			 */
633 			return (0);
634 		}
635 
636 		*pdelay = delay;
637 		return (1);
638 	}
639 
640 	/*
641 	 * Packet received. Verify timestamp
642 	 */
643 	delay = (long)(time_uptime - lle_hittime);
644 	if (delay < nd_delay) {
645 
646 		/*
647 		 * V_nd6_delay still not passed since the first
648 		 * hit in STALE state.
649 		 * Reshedule timer and return.
650 		 */
651 		*pdelay = (long)(nd_delay - delay) * hz;
652 		return (1);
653 	}
654 
655 	/* Request switching to probe */
656 	*do_switch = 1;
657 	return (0);
658 }
659 
660 
661 /*
662  * Switch @lle state to new state optionally arming timers.
663  *
664  * Set noinline to be dtrace-friendly
665  */
666 __noinline void
667 nd6_llinfo_setstate(struct llentry *lle, int newstate)
668 {
669 	struct ifnet *ifp;
670 	int nd_gctimer, nd_delay;
671 	long delay, remtime;
672 
673 	delay = 0;
674 	remtime = 0;
675 
676 	switch (newstate) {
677 	case ND6_LLINFO_INCOMPLETE:
678 		ifp = lle->lle_tbl->llt_ifp;
679 		delay = (long)ND_IFINFO(ifp)->retrans * hz / 1000;
680 		break;
681 	case ND6_LLINFO_REACHABLE:
682 		if (!ND6_LLINFO_PERMANENT(lle)) {
683 			ifp = lle->lle_tbl->llt_ifp;
684 			delay = (long)ND_IFINFO(ifp)->reachable * hz;
685 		}
686 		break;
687 	case ND6_LLINFO_STALE:
688 
689 		/*
690 		 * Notify fast path that we want to know if any packet
691 		 * is transmitted by setting r_skip_req.
692 		 */
693 		LLE_REQ_LOCK(lle);
694 		lle->r_skip_req = 1;
695 		LLE_REQ_UNLOCK(lle);
696 		nd_delay = V_nd6_delay;
697 		nd_gctimer = V_nd6_gctimer;
698 
699 		delay = (long)(MIN(nd_gctimer, nd_delay)) * hz;
700 		remtime = (long)nd_gctimer * hz - delay;
701 		break;
702 	case ND6_LLINFO_DELAY:
703 		lle->la_asked = 0;
704 		delay = (long)V_nd6_delay * hz;
705 		break;
706 	}
707 
708 	if (delay > 0)
709 		nd6_llinfo_settimer_locked(lle, delay);
710 
711 	lle->lle_remtime = remtime;
712 	lle->ln_state = newstate;
713 }
714 
715 /*
716  * Timer-dependent part of nd state machine.
717  *
718  * Set noinline to be dtrace-friendly
719  */
720 static __noinline void
721 nd6_llinfo_timer(void *arg)
722 {
723 	struct llentry *ln;
724 	struct in6_addr *dst, *pdst, *psrc, src;
725 	struct ifnet *ifp;
726 	struct nd_ifinfo *ndi = NULL;
727 	int do_switch, send_ns;
728 	long delay;
729 
730 	KASSERT(arg != NULL, ("%s: arg NULL", __func__));
731 	ln = (struct llentry *)arg;
732 	LLE_WLOCK(ln);
733 	if (callout_pending(&ln->lle_timer)) {
734 		/*
735 		 * Here we are a bit odd here in the treatment of
736 		 * active/pending. If the pending bit is set, it got
737 		 * rescheduled before I ran. The active
738 		 * bit we ignore, since if it was stopped
739 		 * in ll_tablefree() and was currently running
740 		 * it would have return 0 so the code would
741 		 * not have deleted it since the callout could
742 		 * not be stopped so we want to go through
743 		 * with the delete here now. If the callout
744 		 * was restarted, the pending bit will be back on and
745 		 * we just want to bail since the callout_reset would
746 		 * return 1 and our reference would have been removed
747 		 * by nd6_llinfo_settimer_locked above since canceled
748 		 * would have been 1.
749 		 */
750 		LLE_WUNLOCK(ln);
751 		return;
752 	}
753 	ifp = ln->lle_tbl->llt_ifp;
754 	CURVNET_SET(ifp->if_vnet);
755 	ndi = ND_IFINFO(ifp);
756 	send_ns = 0;
757 	dst = &ln->r_l3addr.addr6;
758 	pdst = dst;
759 
760 	if (ln->ln_ntick > 0) {
761 		if (ln->ln_ntick > INT_MAX) {
762 			ln->ln_ntick -= INT_MAX;
763 			nd6_llinfo_settimer_locked(ln, INT_MAX);
764 		} else {
765 			ln->ln_ntick = 0;
766 			nd6_llinfo_settimer_locked(ln, ln->ln_ntick);
767 		}
768 		goto done;
769 	}
770 
771 	if (ln->la_flags & LLE_STATIC) {
772 		goto done;
773 	}
774 
775 	if (ln->la_flags & LLE_DELETED) {
776 		nd6_free(ln, 0);
777 		ln = NULL;
778 		goto done;
779 	}
780 
781 	switch (ln->ln_state) {
782 	case ND6_LLINFO_INCOMPLETE:
783 		if (ln->la_asked < V_nd6_mmaxtries) {
784 			ln->la_asked++;
785 			send_ns = 1;
786 			/* Send NS to multicast address */
787 			pdst = NULL;
788 		} else {
789 			struct mbuf *m = ln->la_hold;
790 			if (m) {
791 				struct mbuf *m0;
792 
793 				/*
794 				 * assuming every packet in la_hold has the
795 				 * same IP header.  Send error after unlock.
796 				 */
797 				m0 = m->m_nextpkt;
798 				m->m_nextpkt = NULL;
799 				ln->la_hold = m0;
800 				clear_llinfo_pqueue(ln);
801 			}
802 			EVENTHANDLER_INVOKE(lle_event, ln, LLENTRY_TIMEDOUT);
803 			nd6_free(ln, 0);
804 			ln = NULL;
805 			if (m != NULL)
806 				icmp6_error2(m, ICMP6_DST_UNREACH,
807 				    ICMP6_DST_UNREACH_ADDR, 0, ifp);
808 		}
809 		break;
810 	case ND6_LLINFO_REACHABLE:
811 		if (!ND6_LLINFO_PERMANENT(ln))
812 			nd6_llinfo_setstate(ln, ND6_LLINFO_STALE);
813 		break;
814 
815 	case ND6_LLINFO_STALE:
816 		if (nd6_is_stale(ln, &delay, &do_switch) != 0) {
817 
818 			/*
819 			 * No packet has used this entry and GC timeout
820 			 * has not been passed. Reshedule timer and
821 			 * return.
822 			 */
823 			nd6_llinfo_settimer_locked(ln, delay);
824 			break;
825 		}
826 
827 		if (do_switch == 0) {
828 
829 			/*
830 			 * GC timer has ended and entry hasn't been used.
831 			 * Run Garbage collector (RFC 4861, 5.3)
832 			 */
833 			if (!ND6_LLINFO_PERMANENT(ln)) {
834 				EVENTHANDLER_INVOKE(lle_event, ln,
835 				    LLENTRY_EXPIRED);
836 				nd6_free(ln, 1);
837 				ln = NULL;
838 			}
839 			break;
840 		}
841 
842 		/* Entry has been used AND delay timer has ended. */
843 
844 		/* FALLTHROUGH */
845 
846 	case ND6_LLINFO_DELAY:
847 		if (ndi && (ndi->flags & ND6_IFF_PERFORMNUD) != 0) {
848 			/* We need NUD */
849 			ln->la_asked = 1;
850 			nd6_llinfo_setstate(ln, ND6_LLINFO_PROBE);
851 			send_ns = 1;
852 		} else
853 			nd6_llinfo_setstate(ln, ND6_LLINFO_STALE); /* XXX */
854 		break;
855 	case ND6_LLINFO_PROBE:
856 		if (ln->la_asked < V_nd6_umaxtries) {
857 			ln->la_asked++;
858 			send_ns = 1;
859 		} else {
860 			EVENTHANDLER_INVOKE(lle_event, ln, LLENTRY_EXPIRED);
861 			nd6_free(ln, 0);
862 			ln = NULL;
863 		}
864 		break;
865 	default:
866 		panic("%s: paths in a dark night can be confusing: %d",
867 		    __func__, ln->ln_state);
868 	}
869 done:
870 	if (send_ns != 0) {
871 		nd6_llinfo_settimer_locked(ln, (long)ndi->retrans * hz / 1000);
872 		psrc = nd6_llinfo_get_holdsrc(ln, &src);
873 		LLE_FREE_LOCKED(ln);
874 		ln = NULL;
875 		nd6_ns_output(ifp, psrc, pdst, dst, NULL);
876 	}
877 
878 	if (ln != NULL)
879 		LLE_FREE_LOCKED(ln);
880 	CURVNET_RESTORE();
881 }
882 
883 
884 /*
885  * ND6 timer routine to expire default route list and prefix list
886  */
887 void
888 nd6_timer(void *arg)
889 {
890 	CURVNET_SET((struct vnet *) arg);
891 	struct nd_drhead drq;
892 	struct nd_defrouter *dr, *ndr;
893 	struct nd_prefix *pr, *npr;
894 	struct in6_ifaddr *ia6, *nia6;
895 
896 	callout_reset(&V_nd6_timer_ch, V_nd6_prune * hz,
897 	    nd6_timer, curvnet);
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 	CURVNET_RESTORE();
1026 }
1027 
1028 /*
1029  * ia6 - deprecated/invalidated temporary address
1030  */
1031 static int
1032 regen_tmpaddr(struct in6_ifaddr *ia6)
1033 {
1034 	struct ifaddr *ifa;
1035 	struct ifnet *ifp;
1036 	struct in6_ifaddr *public_ifa6 = NULL;
1037 
1038 	ifp = ia6->ia_ifa.ifa_ifp;
1039 	IF_ADDR_RLOCK(ifp);
1040 	TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1041 		struct in6_ifaddr *it6;
1042 
1043 		if (ifa->ifa_addr->sa_family != AF_INET6)
1044 			continue;
1045 
1046 		it6 = (struct in6_ifaddr *)ifa;
1047 
1048 		/* ignore no autoconf addresses. */
1049 		if ((it6->ia6_flags & IN6_IFF_AUTOCONF) == 0)
1050 			continue;
1051 
1052 		/* ignore autoconf addresses with different prefixes. */
1053 		if (it6->ia6_ndpr == NULL || it6->ia6_ndpr != ia6->ia6_ndpr)
1054 			continue;
1055 
1056 		/*
1057 		 * Now we are looking at an autoconf address with the same
1058 		 * prefix as ours.  If the address is temporary and is still
1059 		 * preferred, do not create another one.  It would be rare, but
1060 		 * could happen, for example, when we resume a laptop PC after
1061 		 * a long period.
1062 		 */
1063 		if ((it6->ia6_flags & IN6_IFF_TEMPORARY) != 0 &&
1064 		    !IFA6_IS_DEPRECATED(it6)) {
1065 			public_ifa6 = NULL;
1066 			break;
1067 		}
1068 
1069 		/*
1070 		 * This is a public autoconf address that has the same prefix
1071 		 * as ours.  If it is preferred, keep it.  We can't break the
1072 		 * loop here, because there may be a still-preferred temporary
1073 		 * address with the prefix.
1074 		 */
1075 		if (!IFA6_IS_DEPRECATED(it6))
1076 			public_ifa6 = it6;
1077 	}
1078 	if (public_ifa6 != NULL)
1079 		ifa_ref(&public_ifa6->ia_ifa);
1080 	IF_ADDR_RUNLOCK(ifp);
1081 
1082 	if (public_ifa6 != NULL) {
1083 		int e;
1084 
1085 		if ((e = in6_tmpifadd(public_ifa6, 0, 0)) != 0) {
1086 			ifa_free(&public_ifa6->ia_ifa);
1087 			log(LOG_NOTICE, "regen_tmpaddr: failed to create a new"
1088 			    " tmp addr,errno=%d\n", e);
1089 			return (-1);
1090 		}
1091 		ifa_free(&public_ifa6->ia_ifa);
1092 		return (0);
1093 	}
1094 
1095 	return (-1);
1096 }
1097 
1098 /*
1099  * Nuke neighbor cache/prefix/default router management table, right before
1100  * ifp goes away.
1101  */
1102 void
1103 nd6_purge(struct ifnet *ifp)
1104 {
1105 	struct nd_drhead drq;
1106 	struct nd_defrouter *dr, *ndr;
1107 	struct nd_prefix *pr, *npr;
1108 
1109 	TAILQ_INIT(&drq);
1110 
1111 	/*
1112 	 * Nuke default router list entries toward ifp.
1113 	 * We defer removal of default router list entries that is installed
1114 	 * in the routing table, in order to keep additional side effects as
1115 	 * small as possible.
1116 	 */
1117 	ND6_WLOCK();
1118 	TAILQ_FOREACH_SAFE(dr, &V_nd_defrouter, dr_entry, ndr) {
1119 		if (dr->installed)
1120 			continue;
1121 		if (dr->ifp == ifp)
1122 			defrouter_unlink(dr, &drq);
1123 	}
1124 
1125 	TAILQ_FOREACH_SAFE(dr, &V_nd_defrouter, dr_entry, ndr) {
1126 		if (!dr->installed)
1127 			continue;
1128 		if (dr->ifp == ifp)
1129 			defrouter_unlink(dr, &drq);
1130 	}
1131 	ND6_WUNLOCK();
1132 
1133 	while ((dr = TAILQ_FIRST(&drq)) != NULL) {
1134 		TAILQ_REMOVE(&drq, dr, dr_entry);
1135 		defrouter_del(dr);
1136 	}
1137 
1138 	/* Nuke prefix list entries toward ifp */
1139 	LIST_FOREACH_SAFE(pr, &V_nd_prefix, ndpr_entry, npr) {
1140 		if (pr->ndpr_ifp == ifp) {
1141 			/*
1142 			 * Because if_detach() does *not* release prefixes
1143 			 * while purging addresses the reference count will
1144 			 * still be above zero. We therefore reset it to
1145 			 * make sure that the prefix really gets purged.
1146 			 */
1147 			pr->ndpr_refcnt = 0;
1148 
1149 			/*
1150 			 * Previously, pr->ndpr_addr is removed as well,
1151 			 * but I strongly believe we don't have to do it.
1152 			 * nd6_purge() is only called from in6_ifdetach(),
1153 			 * which removes all the associated interface addresses
1154 			 * by itself.
1155 			 * (jinmei@kame.net 20010129)
1156 			 */
1157 			prelist_remove(pr);
1158 		}
1159 	}
1160 
1161 	/* cancel default outgoing interface setting */
1162 	if (V_nd6_defifindex == ifp->if_index)
1163 		nd6_setdefaultiface(0);
1164 
1165 	if (ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) {
1166 		/* Refresh default router list. */
1167 		defrouter_select();
1168 	}
1169 
1170 	/* XXXXX
1171 	 * We do not nuke the neighbor cache entries here any more
1172 	 * because the neighbor cache is kept in if_afdata[AF_INET6].
1173 	 * nd6_purge() is invoked by in6_ifdetach() which is called
1174 	 * from if_detach() where everything gets purged. So let
1175 	 * in6_domifdetach() do the actual L2 table purging work.
1176 	 */
1177 }
1178 
1179 /*
1180  * the caller acquires and releases the lock on the lltbls
1181  * Returns the llentry locked
1182  */
1183 struct llentry *
1184 nd6_lookup(const struct in6_addr *addr6, int flags, struct ifnet *ifp)
1185 {
1186 	struct sockaddr_in6 sin6;
1187 	struct llentry *ln;
1188 
1189 	bzero(&sin6, sizeof(sin6));
1190 	sin6.sin6_len = sizeof(struct sockaddr_in6);
1191 	sin6.sin6_family = AF_INET6;
1192 	sin6.sin6_addr = *addr6;
1193 
1194 	IF_AFDATA_LOCK_ASSERT(ifp);
1195 
1196 	ln = lla_lookup(LLTABLE6(ifp), flags, (struct sockaddr *)&sin6);
1197 
1198 	return (ln);
1199 }
1200 
1201 struct llentry *
1202 nd6_alloc(const struct in6_addr *addr6, int flags, struct ifnet *ifp)
1203 {
1204 	struct sockaddr_in6 sin6;
1205 	struct llentry *ln;
1206 
1207 	bzero(&sin6, sizeof(sin6));
1208 	sin6.sin6_len = sizeof(struct sockaddr_in6);
1209 	sin6.sin6_family = AF_INET6;
1210 	sin6.sin6_addr = *addr6;
1211 
1212 	ln = lltable_alloc_entry(LLTABLE6(ifp), 0, (struct sockaddr *)&sin6);
1213 	if (ln != NULL)
1214 		ln->ln_state = ND6_LLINFO_NOSTATE;
1215 
1216 	return (ln);
1217 }
1218 
1219 /*
1220  * Test whether a given IPv6 address is a neighbor or not, ignoring
1221  * the actual neighbor cache.  The neighbor cache is ignored in order
1222  * to not reenter the routing code from within itself.
1223  */
1224 static int
1225 nd6_is_new_addr_neighbor(const struct sockaddr_in6 *addr, struct ifnet *ifp)
1226 {
1227 	struct nd_prefix *pr;
1228 	struct ifaddr *dstaddr;
1229 	struct rt_addrinfo info;
1230 	struct sockaddr_in6 rt_key;
1231 	struct sockaddr *dst6;
1232 	int fibnum;
1233 
1234 	/*
1235 	 * A link-local address is always a neighbor.
1236 	 * XXX: a link does not necessarily specify a single interface.
1237 	 */
1238 	if (IN6_IS_ADDR_LINKLOCAL(&addr->sin6_addr)) {
1239 		struct sockaddr_in6 sin6_copy;
1240 		u_int32_t zone;
1241 
1242 		/*
1243 		 * We need sin6_copy since sa6_recoverscope() may modify the
1244 		 * content (XXX).
1245 		 */
1246 		sin6_copy = *addr;
1247 		if (sa6_recoverscope(&sin6_copy))
1248 			return (0); /* XXX: should be impossible */
1249 		if (in6_setscope(&sin6_copy.sin6_addr, ifp, &zone))
1250 			return (0);
1251 		if (sin6_copy.sin6_scope_id == zone)
1252 			return (1);
1253 		else
1254 			return (0);
1255 	}
1256 
1257 	bzero(&rt_key, sizeof(rt_key));
1258 	bzero(&info, sizeof(info));
1259 	info.rti_info[RTAX_DST] = (struct sockaddr *)&rt_key;
1260 
1261 	/* Always use the default FIB here. XXME - why? */
1262 	fibnum = RT_DEFAULT_FIB;
1263 
1264 	/*
1265 	 * If the address matches one of our addresses,
1266 	 * it should be a neighbor.
1267 	 * If the address matches one of our on-link prefixes, it should be a
1268 	 * neighbor.
1269 	 */
1270 	LIST_FOREACH(pr, &V_nd_prefix, ndpr_entry) {
1271 		if (pr->ndpr_ifp != ifp)
1272 			continue;
1273 
1274 		if (!(pr->ndpr_stateflags & NDPRF_ONLINK)) {
1275 
1276 			/* Always use the default FIB here. */
1277 			dst6 = (struct sockaddr *)&pr->ndpr_prefix;
1278 
1279 			/* Restore length field before retrying lookup */
1280 			rt_key.sin6_len = sizeof(rt_key);
1281 			if (rib_lookup_info(fibnum, dst6, 0, 0, &info) != 0)
1282 				continue;
1283 			/*
1284 			 * This is the case where multiple interfaces
1285 			 * have the same prefix, but only one is installed
1286 			 * into the routing table and that prefix entry
1287 			 * is not the one being examined here. In the case
1288 			 * where RADIX_MPATH is enabled, multiple route
1289 			 * entries (of the same rt_key value) will be
1290 			 * installed because the interface addresses all
1291 			 * differ.
1292 			 */
1293 			if (!IN6_ARE_ADDR_EQUAL(&pr->ndpr_prefix.sin6_addr,
1294 			       &rt_key.sin6_addr))
1295 				continue;
1296 		}
1297 
1298 		if (IN6_ARE_MASKED_ADDR_EQUAL(&pr->ndpr_prefix.sin6_addr,
1299 		    &addr->sin6_addr, &pr->ndpr_mask))
1300 			return (1);
1301 	}
1302 
1303 	/*
1304 	 * If the address is assigned on the node of the other side of
1305 	 * a p2p interface, the address should be a neighbor.
1306 	 */
1307 	dstaddr = ifa_ifwithdstaddr((const struct sockaddr *)addr, RT_ALL_FIBS);
1308 	if (dstaddr != NULL) {
1309 		if (dstaddr->ifa_ifp == ifp) {
1310 			ifa_free(dstaddr);
1311 			return (1);
1312 		}
1313 		ifa_free(dstaddr);
1314 	}
1315 
1316 	/*
1317 	 * If the default router list is empty, all addresses are regarded
1318 	 * as on-link, and thus, as a neighbor.
1319 	 */
1320 	if (ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV &&
1321 	    TAILQ_EMPTY(&V_nd_defrouter) &&
1322 	    V_nd6_defifindex == ifp->if_index) {
1323 		return (1);
1324 	}
1325 
1326 	return (0);
1327 }
1328 
1329 
1330 /*
1331  * Detect if a given IPv6 address identifies a neighbor on a given link.
1332  * XXX: should take care of the destination of a p2p link?
1333  */
1334 int
1335 nd6_is_addr_neighbor(const struct sockaddr_in6 *addr, struct ifnet *ifp)
1336 {
1337 	struct llentry *lle;
1338 	int rc = 0;
1339 
1340 	IF_AFDATA_UNLOCK_ASSERT(ifp);
1341 	if (nd6_is_new_addr_neighbor(addr, ifp))
1342 		return (1);
1343 
1344 	/*
1345 	 * Even if the address matches none of our addresses, it might be
1346 	 * in the neighbor cache.
1347 	 */
1348 	IF_AFDATA_RLOCK(ifp);
1349 	if ((lle = nd6_lookup(&addr->sin6_addr, 0, ifp)) != NULL) {
1350 		LLE_RUNLOCK(lle);
1351 		rc = 1;
1352 	}
1353 	IF_AFDATA_RUNLOCK(ifp);
1354 	return (rc);
1355 }
1356 
1357 /*
1358  * Free an nd6 llinfo entry.
1359  * Since the function would cause significant changes in the kernel, DO NOT
1360  * make it global, unless you have a strong reason for the change, and are sure
1361  * that the change is safe.
1362  *
1363  * Set noinline to be dtrace-friendly
1364  */
1365 static __noinline void
1366 nd6_free(struct llentry *ln, int gc)
1367 {
1368 	struct nd_defrouter *dr;
1369 	struct ifnet *ifp;
1370 
1371 	LLE_WLOCK_ASSERT(ln);
1372 
1373 	/*
1374 	 * we used to have pfctlinput(PRC_HOSTDEAD) here.
1375 	 * even though it is not harmful, it was not really necessary.
1376 	 */
1377 
1378 	/* cancel timer */
1379 	nd6_llinfo_settimer_locked(ln, -1);
1380 
1381 	dr = NULL;
1382 	ifp = ln->lle_tbl->llt_ifp;
1383 	if (ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) {
1384 		dr = defrouter_lookup(&ln->r_l3addr.addr6, ifp);
1385 
1386 		if (dr != NULL && dr->expire &&
1387 		    ln->ln_state == ND6_LLINFO_STALE && gc) {
1388 			/*
1389 			 * If the reason for the deletion is just garbage
1390 			 * collection, and the neighbor is an active default
1391 			 * router, do not delete it.  Instead, reset the GC
1392 			 * timer using the router's lifetime.
1393 			 * Simply deleting the entry would affect default
1394 			 * router selection, which is not necessarily a good
1395 			 * thing, especially when we're using router preference
1396 			 * values.
1397 			 * XXX: the check for ln_state would be redundant,
1398 			 *      but we intentionally keep it just in case.
1399 			 */
1400 			if (dr->expire > time_uptime)
1401 				nd6_llinfo_settimer_locked(ln,
1402 				    (dr->expire - time_uptime) * hz);
1403 			else
1404 				nd6_llinfo_settimer_locked(ln,
1405 				    (long)V_nd6_gctimer * hz);
1406 
1407 			LLE_REMREF(ln);
1408 			LLE_WUNLOCK(ln);
1409 			defrouter_rele(dr);
1410 			return;
1411 		}
1412 
1413 		if (dr) {
1414 			/*
1415 			 * Unreachablity of a router might affect the default
1416 			 * router selection and on-link detection of advertised
1417 			 * prefixes.
1418 			 */
1419 
1420 			/*
1421 			 * Temporarily fake the state to choose a new default
1422 			 * router and to perform on-link determination of
1423 			 * prefixes correctly.
1424 			 * Below the state will be set correctly,
1425 			 * or the entry itself will be deleted.
1426 			 */
1427 			ln->ln_state = ND6_LLINFO_INCOMPLETE;
1428 		}
1429 
1430 		if (ln->ln_router || dr) {
1431 
1432 			/*
1433 			 * We need to unlock to avoid a LOR with rt6_flush() with the
1434 			 * rnh and for the calls to pfxlist_onlink_check() and
1435 			 * defrouter_select() in the block further down for calls
1436 			 * into nd6_lookup().  We still hold a ref.
1437 			 */
1438 			LLE_WUNLOCK(ln);
1439 
1440 			/*
1441 			 * rt6_flush must be called whether or not the neighbor
1442 			 * is in the Default Router List.
1443 			 * See a corresponding comment in nd6_na_input().
1444 			 */
1445 			rt6_flush(&ln->r_l3addr.addr6, ifp);
1446 		}
1447 
1448 		if (dr) {
1449 			/*
1450 			 * Since defrouter_select() does not affect the
1451 			 * on-link determination and MIP6 needs the check
1452 			 * before the default router selection, we perform
1453 			 * the check now.
1454 			 */
1455 			pfxlist_onlink_check();
1456 
1457 			/*
1458 			 * Refresh default router list.
1459 			 */
1460 			defrouter_select();
1461 		}
1462 
1463 		/*
1464 		 * If this entry was added by an on-link redirect, remove the
1465 		 * corresponding host route.
1466 		 */
1467 		if (ln->la_flags & LLE_REDIRECT)
1468 			nd6_free_redirect(ln);
1469 
1470 		if (ln->ln_router || dr)
1471 			LLE_WLOCK(ln);
1472 	}
1473 
1474 	/*
1475 	 * Save to unlock. We still hold an extra reference and will not
1476 	 * free(9) in llentry_free() if someone else holds one as well.
1477 	 */
1478 	LLE_WUNLOCK(ln);
1479 	IF_AFDATA_LOCK(ifp);
1480 	LLE_WLOCK(ln);
1481 	/* Guard against race with other llentry_free(). */
1482 	if (ln->la_flags & LLE_LINKED) {
1483 		/* Remove callout reference */
1484 		LLE_REMREF(ln);
1485 		lltable_unlink_entry(ln->lle_tbl, ln);
1486 	}
1487 	IF_AFDATA_UNLOCK(ifp);
1488 
1489 	llentry_free(ln);
1490 	if (dr != NULL)
1491 		defrouter_rele(dr);
1492 }
1493 
1494 static int
1495 nd6_isdynrte(const struct rtentry *rt, void *xap)
1496 {
1497 
1498 	if (rt->rt_flags == (RTF_UP | RTF_HOST | RTF_DYNAMIC))
1499 		return (1);
1500 
1501 	return (0);
1502 }
1503 /*
1504  * Remove the rtentry for the given llentry,
1505  * both of which were installed by a redirect.
1506  */
1507 static void
1508 nd6_free_redirect(const struct llentry *ln)
1509 {
1510 	int fibnum;
1511 	struct sockaddr_in6 sin6;
1512 	struct rt_addrinfo info;
1513 
1514 	lltable_fill_sa_entry(ln, (struct sockaddr *)&sin6);
1515 	memset(&info, 0, sizeof(info));
1516 	info.rti_info[RTAX_DST] = (struct sockaddr *)&sin6;
1517 	info.rti_filter = nd6_isdynrte;
1518 
1519 	for (fibnum = 0; fibnum < rt_numfibs; fibnum++)
1520 		rtrequest1_fib(RTM_DELETE, &info, NULL, fibnum);
1521 }
1522 
1523 /*
1524  * Rejuvenate this function for routing operations related
1525  * processing.
1526  */
1527 void
1528 nd6_rtrequest(int req, struct rtentry *rt, struct rt_addrinfo *info)
1529 {
1530 	struct sockaddr_in6 *gateway;
1531 	struct nd_defrouter *dr;
1532 	struct ifnet *ifp;
1533 
1534 	gateway = (struct sockaddr_in6 *)rt->rt_gateway;
1535 	ifp = rt->rt_ifp;
1536 
1537 	switch (req) {
1538 	case RTM_ADD:
1539 		break;
1540 
1541 	case RTM_DELETE:
1542 		if (!ifp)
1543 			return;
1544 		/*
1545 		 * Only indirect routes are interesting.
1546 		 */
1547 		if ((rt->rt_flags & RTF_GATEWAY) == 0)
1548 			return;
1549 		/*
1550 		 * check for default route
1551 		 */
1552 		if (IN6_ARE_ADDR_EQUAL(&in6addr_any,
1553 		    &SIN6(rt_key(rt))->sin6_addr)) {
1554 			dr = defrouter_lookup(&gateway->sin6_addr, ifp);
1555 			if (dr != NULL) {
1556 				dr->installed = 0;
1557 				defrouter_rele(dr);
1558 			}
1559 		}
1560 		break;
1561 	}
1562 }
1563 
1564 
1565 int
1566 nd6_ioctl(u_long cmd, caddr_t data, struct ifnet *ifp)
1567 {
1568 	struct in6_ndireq *ndi = (struct in6_ndireq *)data;
1569 	struct in6_nbrinfo *nbi = (struct in6_nbrinfo *)data;
1570 	struct in6_ndifreq *ndif = (struct in6_ndifreq *)data;
1571 	int error = 0;
1572 
1573 	if (ifp->if_afdata[AF_INET6] == NULL)
1574 		return (EPFNOSUPPORT);
1575 	switch (cmd) {
1576 	case OSIOCGIFINFO_IN6:
1577 #define ND	ndi->ndi
1578 		/* XXX: old ndp(8) assumes a positive value for linkmtu. */
1579 		bzero(&ND, sizeof(ND));
1580 		ND.linkmtu = IN6_LINKMTU(ifp);
1581 		ND.maxmtu = ND_IFINFO(ifp)->maxmtu;
1582 		ND.basereachable = ND_IFINFO(ifp)->basereachable;
1583 		ND.reachable = ND_IFINFO(ifp)->reachable;
1584 		ND.retrans = ND_IFINFO(ifp)->retrans;
1585 		ND.flags = ND_IFINFO(ifp)->flags;
1586 		ND.recalctm = ND_IFINFO(ifp)->recalctm;
1587 		ND.chlim = ND_IFINFO(ifp)->chlim;
1588 		break;
1589 	case SIOCGIFINFO_IN6:
1590 		ND = *ND_IFINFO(ifp);
1591 		break;
1592 	case SIOCSIFINFO_IN6:
1593 		/*
1594 		 * used to change host variables from userland.
1595 		 * intented for a use on router to reflect RA configurations.
1596 		 */
1597 		/* 0 means 'unspecified' */
1598 		if (ND.linkmtu != 0) {
1599 			if (ND.linkmtu < IPV6_MMTU ||
1600 			    ND.linkmtu > IN6_LINKMTU(ifp)) {
1601 				error = EINVAL;
1602 				break;
1603 			}
1604 			ND_IFINFO(ifp)->linkmtu = ND.linkmtu;
1605 		}
1606 
1607 		if (ND.basereachable != 0) {
1608 			int obasereachable = ND_IFINFO(ifp)->basereachable;
1609 
1610 			ND_IFINFO(ifp)->basereachable = ND.basereachable;
1611 			if (ND.basereachable != obasereachable)
1612 				ND_IFINFO(ifp)->reachable =
1613 				    ND_COMPUTE_RTIME(ND.basereachable);
1614 		}
1615 		if (ND.retrans != 0)
1616 			ND_IFINFO(ifp)->retrans = ND.retrans;
1617 		if (ND.chlim != 0)
1618 			ND_IFINFO(ifp)->chlim = ND.chlim;
1619 		/* FALLTHROUGH */
1620 	case SIOCSIFINFO_FLAGS:
1621 	{
1622 		struct ifaddr *ifa;
1623 		struct in6_ifaddr *ia;
1624 
1625 		if ((ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED) &&
1626 		    !(ND.flags & ND6_IFF_IFDISABLED)) {
1627 			/* ifdisabled 1->0 transision */
1628 
1629 			/*
1630 			 * If the interface is marked as ND6_IFF_IFDISABLED and
1631 			 * has an link-local address with IN6_IFF_DUPLICATED,
1632 			 * do not clear ND6_IFF_IFDISABLED.
1633 			 * See RFC 4862, Section 5.4.5.
1634 			 */
1635 			IF_ADDR_RLOCK(ifp);
1636 			TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1637 				if (ifa->ifa_addr->sa_family != AF_INET6)
1638 					continue;
1639 				ia = (struct in6_ifaddr *)ifa;
1640 				if ((ia->ia6_flags & IN6_IFF_DUPLICATED) &&
1641 				    IN6_IS_ADDR_LINKLOCAL(IA6_IN6(ia)))
1642 					break;
1643 			}
1644 			IF_ADDR_RUNLOCK(ifp);
1645 
1646 			if (ifa != NULL) {
1647 				/* LLA is duplicated. */
1648 				ND.flags |= ND6_IFF_IFDISABLED;
1649 				log(LOG_ERR, "Cannot enable an interface"
1650 				    " with a link-local address marked"
1651 				    " duplicate.\n");
1652 			} else {
1653 				ND_IFINFO(ifp)->flags &= ~ND6_IFF_IFDISABLED;
1654 				if (ifp->if_flags & IFF_UP)
1655 					in6_if_up(ifp);
1656 			}
1657 		} else if (!(ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED) &&
1658 			    (ND.flags & ND6_IFF_IFDISABLED)) {
1659 			/* ifdisabled 0->1 transision */
1660 			/* Mark all IPv6 address as tentative. */
1661 
1662 			ND_IFINFO(ifp)->flags |= ND6_IFF_IFDISABLED;
1663 			if (V_ip6_dad_count > 0 &&
1664 			    (ND_IFINFO(ifp)->flags & ND6_IFF_NO_DAD) == 0) {
1665 				IF_ADDR_RLOCK(ifp);
1666 				TAILQ_FOREACH(ifa, &ifp->if_addrhead,
1667 				    ifa_link) {
1668 					if (ifa->ifa_addr->sa_family !=
1669 					    AF_INET6)
1670 						continue;
1671 					ia = (struct in6_ifaddr *)ifa;
1672 					ia->ia6_flags |= IN6_IFF_TENTATIVE;
1673 				}
1674 				IF_ADDR_RUNLOCK(ifp);
1675 			}
1676 		}
1677 
1678 		if (ND.flags & ND6_IFF_AUTO_LINKLOCAL) {
1679 			if (!(ND_IFINFO(ifp)->flags & ND6_IFF_AUTO_LINKLOCAL)) {
1680 				/* auto_linklocal 0->1 transision */
1681 
1682 				/* If no link-local address on ifp, configure */
1683 				ND_IFINFO(ifp)->flags |= ND6_IFF_AUTO_LINKLOCAL;
1684 				in6_ifattach(ifp, NULL);
1685 			} else if (!(ND.flags & ND6_IFF_IFDISABLED) &&
1686 			    ifp->if_flags & IFF_UP) {
1687 				/*
1688 				 * When the IF already has
1689 				 * ND6_IFF_AUTO_LINKLOCAL, no link-local
1690 				 * address is assigned, and IFF_UP, try to
1691 				 * assign one.
1692 				 */
1693 				IF_ADDR_RLOCK(ifp);
1694 				TAILQ_FOREACH(ifa, &ifp->if_addrhead,
1695 				    ifa_link) {
1696 					if (ifa->ifa_addr->sa_family !=
1697 					    AF_INET6)
1698 						continue;
1699 					ia = (struct in6_ifaddr *)ifa;
1700 					if (IN6_IS_ADDR_LINKLOCAL(IA6_IN6(ia)))
1701 						break;
1702 				}
1703 				IF_ADDR_RUNLOCK(ifp);
1704 				if (ifa != NULL)
1705 					/* No LLA is configured. */
1706 					in6_ifattach(ifp, NULL);
1707 			}
1708 		}
1709 	}
1710 		ND_IFINFO(ifp)->flags = ND.flags;
1711 		break;
1712 #undef ND
1713 	case SIOCSNDFLUSH_IN6:	/* XXX: the ioctl name is confusing... */
1714 		/* sync kernel routing table with the default router list */
1715 		defrouter_reset();
1716 		defrouter_select();
1717 		break;
1718 	case SIOCSPFXFLUSH_IN6:
1719 	{
1720 		/* flush all the prefix advertised by routers */
1721 		struct nd_prefix *pr, *next;
1722 
1723 		LIST_FOREACH_SAFE(pr, &V_nd_prefix, ndpr_entry, next) {
1724 			struct in6_ifaddr *ia, *ia_next;
1725 
1726 			if (IN6_IS_ADDR_LINKLOCAL(&pr->ndpr_prefix.sin6_addr))
1727 				continue; /* XXX */
1728 
1729 			/* do we really have to remove addresses as well? */
1730 			/* XXXRW: in6_ifaddrhead locking. */
1731 			TAILQ_FOREACH_SAFE(ia, &V_in6_ifaddrhead, ia_link,
1732 			    ia_next) {
1733 				if ((ia->ia6_flags & IN6_IFF_AUTOCONF) == 0)
1734 					continue;
1735 
1736 				if (ia->ia6_ndpr == pr)
1737 					in6_purgeaddr(&ia->ia_ifa);
1738 			}
1739 			prelist_remove(pr);
1740 		}
1741 		break;
1742 	}
1743 	case SIOCSRTRFLUSH_IN6:
1744 	{
1745 		/* flush all the default routers */
1746 		struct nd_drhead drq;
1747 		struct nd_defrouter *dr;
1748 
1749 		TAILQ_INIT(&drq);
1750 
1751 		defrouter_reset();
1752 
1753 		ND6_WLOCK();
1754 		while ((dr = TAILQ_FIRST(&V_nd_defrouter)) != NULL)
1755 			defrouter_unlink(dr, &drq);
1756 		ND6_WUNLOCK();
1757 		while ((dr = TAILQ_FIRST(&drq)) != NULL) {
1758 			TAILQ_REMOVE(&drq, dr, dr_entry);
1759 			defrouter_del(dr);
1760 		}
1761 
1762 		defrouter_select();
1763 		break;
1764 	}
1765 	case SIOCGNBRINFO_IN6:
1766 	{
1767 		struct llentry *ln;
1768 		struct in6_addr nb_addr = nbi->addr; /* make local for safety */
1769 
1770 		if ((error = in6_setscope(&nb_addr, ifp, NULL)) != 0)
1771 			return (error);
1772 
1773 		IF_AFDATA_RLOCK(ifp);
1774 		ln = nd6_lookup(&nb_addr, 0, ifp);
1775 		IF_AFDATA_RUNLOCK(ifp);
1776 
1777 		if (ln == NULL) {
1778 			error = EINVAL;
1779 			break;
1780 		}
1781 		nbi->state = ln->ln_state;
1782 		nbi->asked = ln->la_asked;
1783 		nbi->isrouter = ln->ln_router;
1784 		if (ln->la_expire == 0)
1785 			nbi->expire = 0;
1786 		else
1787 			nbi->expire = ln->la_expire + ln->lle_remtime / hz +
1788 			    (time_second - time_uptime);
1789 		LLE_RUNLOCK(ln);
1790 		break;
1791 	}
1792 	case SIOCGDEFIFACE_IN6:	/* XXX: should be implemented as a sysctl? */
1793 		ndif->ifindex = V_nd6_defifindex;
1794 		break;
1795 	case SIOCSDEFIFACE_IN6:	/* XXX: should be implemented as a sysctl? */
1796 		return (nd6_setdefaultiface(ndif->ifindex));
1797 	}
1798 	return (error);
1799 }
1800 
1801 /*
1802  * Calculates new isRouter value based on provided parameters and
1803  * returns it.
1804  */
1805 static int
1806 nd6_is_router(int type, int code, int is_new, int old_addr, int new_addr,
1807     int ln_router)
1808 {
1809 
1810 	/*
1811 	 * ICMP6 type dependent behavior.
1812 	 *
1813 	 * NS: clear IsRouter if new entry
1814 	 * RS: clear IsRouter
1815 	 * RA: set IsRouter if there's lladdr
1816 	 * redir: clear IsRouter if new entry
1817 	 *
1818 	 * RA case, (1):
1819 	 * The spec says that we must set IsRouter in the following cases:
1820 	 * - If lladdr exist, set IsRouter.  This means (1-5).
1821 	 * - If it is old entry (!newentry), set IsRouter.  This means (7).
1822 	 * So, based on the spec, in (1-5) and (7) cases we must set IsRouter.
1823 	 * A quetion arises for (1) case.  (1) case has no lladdr in the
1824 	 * neighbor cache, this is similar to (6).
1825 	 * This case is rare but we figured that we MUST NOT set IsRouter.
1826 	 *
1827 	 *   is_new  old_addr new_addr 	    NS  RS  RA	redir
1828 	 *							D R
1829 	 *	0	n	n	(1)	c   ?     s
1830 	 *	0	y	n	(2)	c   s     s
1831 	 *	0	n	y	(3)	c   s     s
1832 	 *	0	y	y	(4)	c   s     s
1833 	 *	0	y	y	(5)	c   s     s
1834 	 *	1	--	n	(6) c	c	c s
1835 	 *	1	--	y	(7) c	c   s	c s
1836 	 *
1837 	 *					(c=clear s=set)
1838 	 */
1839 	switch (type & 0xff) {
1840 	case ND_NEIGHBOR_SOLICIT:
1841 		/*
1842 		 * New entry must have is_router flag cleared.
1843 		 */
1844 		if (is_new)					/* (6-7) */
1845 			ln_router = 0;
1846 		break;
1847 	case ND_REDIRECT:
1848 		/*
1849 		 * If the icmp is a redirect to a better router, always set the
1850 		 * is_router flag.  Otherwise, if the entry is newly created,
1851 		 * clear the flag.  [RFC 2461, sec 8.3]
1852 		 */
1853 		if (code == ND_REDIRECT_ROUTER)
1854 			ln_router = 1;
1855 		else {
1856 			if (is_new)				/* (6-7) */
1857 				ln_router = 0;
1858 		}
1859 		break;
1860 	case ND_ROUTER_SOLICIT:
1861 		/*
1862 		 * is_router flag must always be cleared.
1863 		 */
1864 		ln_router = 0;
1865 		break;
1866 	case ND_ROUTER_ADVERT:
1867 		/*
1868 		 * Mark an entry with lladdr as a router.
1869 		 */
1870 		if ((!is_new && (old_addr || new_addr)) ||	/* (2-5) */
1871 		    (is_new && new_addr)) {			/* (7) */
1872 			ln_router = 1;
1873 		}
1874 		break;
1875 	}
1876 
1877 	return (ln_router);
1878 }
1879 
1880 /*
1881  * Create neighbor cache entry and cache link-layer address,
1882  * on reception of inbound ND6 packets.  (RS/RA/NS/redirect)
1883  *
1884  * type - ICMP6 type
1885  * code - type dependent information
1886  *
1887  */
1888 void
1889 nd6_cache_lladdr(struct ifnet *ifp, struct in6_addr *from, char *lladdr,
1890     int lladdrlen, int type, int code)
1891 {
1892 	struct llentry *ln = NULL, *ln_tmp;
1893 	int is_newentry;
1894 	int do_update;
1895 	int olladdr;
1896 	int llchange;
1897 	int flags;
1898 	uint16_t router = 0;
1899 	struct sockaddr_in6 sin6;
1900 	struct mbuf *chain = NULL;
1901 	u_char linkhdr[LLE_MAX_LINKHDR];
1902 	size_t linkhdrsize;
1903 	int lladdr_off;
1904 
1905 	IF_AFDATA_UNLOCK_ASSERT(ifp);
1906 
1907 	KASSERT(ifp != NULL, ("%s: ifp == NULL", __func__));
1908 	KASSERT(from != NULL, ("%s: from == NULL", __func__));
1909 
1910 	/* nothing must be updated for unspecified address */
1911 	if (IN6_IS_ADDR_UNSPECIFIED(from))
1912 		return;
1913 
1914 	/*
1915 	 * Validation about ifp->if_addrlen and lladdrlen must be done in
1916 	 * the caller.
1917 	 *
1918 	 * XXX If the link does not have link-layer adderss, what should
1919 	 * we do? (ifp->if_addrlen == 0)
1920 	 * Spec says nothing in sections for RA, RS and NA.  There's small
1921 	 * description on it in NS section (RFC 2461 7.2.3).
1922 	 */
1923 	flags = lladdr ? LLE_EXCLUSIVE : 0;
1924 	IF_AFDATA_RLOCK(ifp);
1925 	ln = nd6_lookup(from, flags, ifp);
1926 	IF_AFDATA_RUNLOCK(ifp);
1927 	is_newentry = 0;
1928 	if (ln == NULL) {
1929 		flags |= LLE_EXCLUSIVE;
1930 		ln = nd6_alloc(from, 0, ifp);
1931 		if (ln == NULL)
1932 			return;
1933 
1934 		/*
1935 		 * Since we already know all the data for the new entry,
1936 		 * fill it before insertion.
1937 		 */
1938 		if (lladdr != NULL) {
1939 			linkhdrsize = sizeof(linkhdr);
1940 			if (lltable_calc_llheader(ifp, AF_INET6, lladdr,
1941 			    linkhdr, &linkhdrsize, &lladdr_off) != 0)
1942 				return;
1943 			lltable_set_entry_addr(ifp, ln, linkhdr, linkhdrsize,
1944 			    lladdr_off);
1945 		}
1946 
1947 		IF_AFDATA_WLOCK(ifp);
1948 		LLE_WLOCK(ln);
1949 		/* Prefer any existing lle over newly-created one */
1950 		ln_tmp = nd6_lookup(from, LLE_EXCLUSIVE, ifp);
1951 		if (ln_tmp == NULL)
1952 			lltable_link_entry(LLTABLE6(ifp), ln);
1953 		IF_AFDATA_WUNLOCK(ifp);
1954 		if (ln_tmp == NULL) {
1955 			/* No existing lle, mark as new entry (6,7) */
1956 			is_newentry = 1;
1957 			nd6_llinfo_setstate(ln, ND6_LLINFO_STALE);
1958 			if (lladdr != NULL)	/* (7) */
1959 				EVENTHANDLER_INVOKE(lle_event, ln,
1960 				    LLENTRY_RESOLVED);
1961 		} else {
1962 			lltable_free_entry(LLTABLE6(ifp), ln);
1963 			ln = ln_tmp;
1964 			ln_tmp = NULL;
1965 		}
1966 	}
1967 	/* do nothing if static ndp is set */
1968 	if ((ln->la_flags & LLE_STATIC)) {
1969 		if (flags & LLE_EXCLUSIVE)
1970 			LLE_WUNLOCK(ln);
1971 		else
1972 			LLE_RUNLOCK(ln);
1973 		return;
1974 	}
1975 
1976 	olladdr = (ln->la_flags & LLE_VALID) ? 1 : 0;
1977 	if (olladdr && lladdr) {
1978 		llchange = bcmp(lladdr, ln->ll_addr,
1979 		    ifp->if_addrlen);
1980 	} else if (!olladdr && lladdr)
1981 		llchange = 1;
1982 	else
1983 		llchange = 0;
1984 
1985 	/*
1986 	 * newentry olladdr  lladdr  llchange	(*=record)
1987 	 *	0	n	n	--	(1)
1988 	 *	0	y	n	--	(2)
1989 	 *	0	n	y	y	(3) * STALE
1990 	 *	0	y	y	n	(4) *
1991 	 *	0	y	y	y	(5) * STALE
1992 	 *	1	--	n	--	(6)   NOSTATE(= PASSIVE)
1993 	 *	1	--	y	--	(7) * STALE
1994 	 */
1995 
1996 	do_update = 0;
1997 	if (is_newentry == 0 && llchange != 0) {
1998 		do_update = 1;	/* (3,5) */
1999 
2000 		/*
2001 		 * Record source link-layer address
2002 		 * XXX is it dependent to ifp->if_type?
2003 		 */
2004 		linkhdrsize = sizeof(linkhdr);
2005 		if (lltable_calc_llheader(ifp, AF_INET6, lladdr,
2006 		    linkhdr, &linkhdrsize, &lladdr_off) != 0)
2007 			return;
2008 
2009 		if (lltable_try_set_entry_addr(ifp, ln, linkhdr, linkhdrsize,
2010 		    lladdr_off) == 0) {
2011 			/* Entry was deleted */
2012 			return;
2013 		}
2014 
2015 		nd6_llinfo_setstate(ln, ND6_LLINFO_STALE);
2016 
2017 		EVENTHANDLER_INVOKE(lle_event, ln, LLENTRY_RESOLVED);
2018 
2019 		if (ln->la_hold != NULL)
2020 			nd6_grab_holdchain(ln, &chain, &sin6);
2021 	}
2022 
2023 	/* Calculates new router status */
2024 	router = nd6_is_router(type, code, is_newentry, olladdr,
2025 	    lladdr != NULL ? 1 : 0, ln->ln_router);
2026 
2027 	ln->ln_router = router;
2028 	/* Mark non-router redirects with special flag */
2029 	if ((type & 0xFF) == ND_REDIRECT && code != ND_REDIRECT_ROUTER)
2030 		ln->la_flags |= LLE_REDIRECT;
2031 
2032 	if (flags & LLE_EXCLUSIVE)
2033 		LLE_WUNLOCK(ln);
2034 	else
2035 		LLE_RUNLOCK(ln);
2036 
2037 	if (chain != NULL)
2038 		nd6_flush_holdchain(ifp, ifp, chain, &sin6);
2039 
2040 	/*
2041 	 * When the link-layer address of a router changes, select the
2042 	 * best router again.  In particular, when the neighbor entry is newly
2043 	 * created, it might affect the selection policy.
2044 	 * Question: can we restrict the first condition to the "is_newentry"
2045 	 * case?
2046 	 * XXX: when we hear an RA from a new router with the link-layer
2047 	 * address option, defrouter_select() is called twice, since
2048 	 * defrtrlist_update called the function as well.  However, I believe
2049 	 * we can compromise the overhead, since it only happens the first
2050 	 * time.
2051 	 * XXX: although defrouter_select() should not have a bad effect
2052 	 * for those are not autoconfigured hosts, we explicitly avoid such
2053 	 * cases for safety.
2054 	 */
2055 	if ((do_update || is_newentry) && router &&
2056 	    ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) {
2057 		/*
2058 		 * guaranteed recursion
2059 		 */
2060 		defrouter_select();
2061 	}
2062 }
2063 
2064 static void
2065 nd6_slowtimo(void *arg)
2066 {
2067 	CURVNET_SET((struct vnet *) arg);
2068 	struct nd_ifinfo *nd6if;
2069 	struct ifnet *ifp;
2070 
2071 	callout_reset(&V_nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
2072 	    nd6_slowtimo, curvnet);
2073 	IFNET_RLOCK_NOSLEEP();
2074 	TAILQ_FOREACH(ifp, &V_ifnet, if_link) {
2075 		if (ifp->if_afdata[AF_INET6] == NULL)
2076 			continue;
2077 		nd6if = ND_IFINFO(ifp);
2078 		if (nd6if->basereachable && /* already initialized */
2079 		    (nd6if->recalctm -= ND6_SLOWTIMER_INTERVAL) <= 0) {
2080 			/*
2081 			 * Since reachable time rarely changes by router
2082 			 * advertisements, we SHOULD insure that a new random
2083 			 * value gets recomputed at least once every few hours.
2084 			 * (RFC 2461, 6.3.4)
2085 			 */
2086 			nd6if->recalctm = V_nd6_recalc_reachtm_interval;
2087 			nd6if->reachable = ND_COMPUTE_RTIME(nd6if->basereachable);
2088 		}
2089 	}
2090 	IFNET_RUNLOCK_NOSLEEP();
2091 	CURVNET_RESTORE();
2092 }
2093 
2094 void
2095 nd6_grab_holdchain(struct llentry *ln, struct mbuf **chain,
2096     struct sockaddr_in6 *sin6)
2097 {
2098 
2099 	LLE_WLOCK_ASSERT(ln);
2100 
2101 	*chain = ln->la_hold;
2102 	ln->la_hold = NULL;
2103 	lltable_fill_sa_entry(ln, (struct sockaddr *)sin6);
2104 
2105 	if (ln->ln_state == ND6_LLINFO_STALE) {
2106 
2107 		/*
2108 		 * The first time we send a packet to a
2109 		 * neighbor whose entry is STALE, we have
2110 		 * to change the state to DELAY and a sets
2111 		 * a timer to expire in DELAY_FIRST_PROBE_TIME
2112 		 * seconds to ensure do neighbor unreachability
2113 		 * detection on expiration.
2114 		 * (RFC 2461 7.3.3)
2115 		 */
2116 		nd6_llinfo_setstate(ln, ND6_LLINFO_DELAY);
2117 	}
2118 }
2119 
2120 int
2121 nd6_output_ifp(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *m,
2122     struct sockaddr_in6 *dst, struct route *ro)
2123 {
2124 	int error;
2125 	int ip6len;
2126 	struct ip6_hdr *ip6;
2127 	struct m_tag *mtag;
2128 
2129 #ifdef MAC
2130 	mac_netinet6_nd6_send(ifp, m);
2131 #endif
2132 
2133 	/*
2134 	 * If called from nd6_ns_output() (NS), nd6_na_output() (NA),
2135 	 * icmp6_redirect_output() (REDIRECT) or from rip6_output() (RS, RA
2136 	 * as handled by rtsol and rtadvd), mbufs will be tagged for SeND
2137 	 * to be diverted to user space.  When re-injected into the kernel,
2138 	 * send_output() will directly dispatch them to the outgoing interface.
2139 	 */
2140 	if (send_sendso_input_hook != NULL) {
2141 		mtag = m_tag_find(m, PACKET_TAG_ND_OUTGOING, NULL);
2142 		if (mtag != NULL) {
2143 			ip6 = mtod(m, struct ip6_hdr *);
2144 			ip6len = sizeof(struct ip6_hdr) + ntohs(ip6->ip6_plen);
2145 			/* Use the SEND socket */
2146 			error = send_sendso_input_hook(m, ifp, SND_OUT,
2147 			    ip6len);
2148 			/* -1 == no app on SEND socket */
2149 			if (error == 0 || error != -1)
2150 			    return (error);
2151 		}
2152 	}
2153 
2154 	m_clrprotoflags(m);	/* Avoid confusing lower layers. */
2155 	IP_PROBE(send, NULL, NULL, mtod(m, struct ip6_hdr *), ifp, NULL,
2156 	    mtod(m, struct ip6_hdr *));
2157 
2158 	if ((ifp->if_flags & IFF_LOOPBACK) == 0)
2159 		origifp = ifp;
2160 
2161 	error = (*ifp->if_output)(origifp, m, (struct sockaddr *)dst, ro);
2162 	return (error);
2163 }
2164 
2165 /*
2166  * Lookup link headerfor @sa_dst address. Stores found
2167  * data in @desten buffer. Copy of lle ln_flags can be also
2168  * saved in @pflags if @pflags is non-NULL.
2169  *
2170  * If destination LLE does not exists or lle state modification
2171  * is required, call "slow" version.
2172  *
2173  * Return values:
2174  * - 0 on success (address copied to buffer).
2175  * - EWOULDBLOCK (no local error, but address is still unresolved)
2176  * - other errors (alloc failure, etc)
2177  */
2178 int
2179 nd6_resolve(struct ifnet *ifp, int is_gw, struct mbuf *m,
2180     const struct sockaddr *sa_dst, u_char *desten, uint32_t *pflags)
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));
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 {
2251 	struct llentry *lle = NULL, *lle_tmp;
2252 	struct in6_addr *psrc, src;
2253 	int send_ns, ll_len;
2254 	char *lladdr;
2255 
2256 	/*
2257 	 * Address resolution or Neighbor Unreachability Detection
2258 	 * for the next hop.
2259 	 * At this point, the destination of the packet must be a unicast
2260 	 * or an anycast address(i.e. not a multicast).
2261 	 */
2262 	if (lle == NULL) {
2263 		IF_AFDATA_RLOCK(ifp);
2264 		lle = nd6_lookup(&dst->sin6_addr, LLE_EXCLUSIVE, ifp);
2265 		IF_AFDATA_RUNLOCK(ifp);
2266 		if ((lle == NULL) && nd6_is_addr_neighbor(dst, ifp))  {
2267 			/*
2268 			 * Since nd6_is_addr_neighbor() internally calls nd6_lookup(),
2269 			 * the condition below is not very efficient.  But we believe
2270 			 * it is tolerable, because this should be a rare case.
2271 			 */
2272 			lle = nd6_alloc(&dst->sin6_addr, 0, ifp);
2273 			if (lle == NULL) {
2274 				char ip6buf[INET6_ADDRSTRLEN];
2275 				log(LOG_DEBUG,
2276 				    "nd6_output: can't allocate llinfo for %s "
2277 				    "(ln=%p)\n",
2278 				    ip6_sprintf(ip6buf, &dst->sin6_addr), lle);
2279 				m_freem(m);
2280 				return (ENOBUFS);
2281 			}
2282 
2283 			IF_AFDATA_WLOCK(ifp);
2284 			LLE_WLOCK(lle);
2285 			/* Prefer any existing entry over newly-created one */
2286 			lle_tmp = nd6_lookup(&dst->sin6_addr, LLE_EXCLUSIVE, ifp);
2287 			if (lle_tmp == NULL)
2288 				lltable_link_entry(LLTABLE6(ifp), lle);
2289 			IF_AFDATA_WUNLOCK(ifp);
2290 			if (lle_tmp != NULL) {
2291 				lltable_free_entry(LLTABLE6(ifp), lle);
2292 				lle = lle_tmp;
2293 				lle_tmp = NULL;
2294 			}
2295 		}
2296 	}
2297 	if (lle == NULL) {
2298 		if (!(ND_IFINFO(ifp)->flags & ND6_IFF_PERFORMNUD)) {
2299 			m_freem(m);
2300 			return (ENOBUFS);
2301 		}
2302 
2303 		if (m != NULL)
2304 			m_freem(m);
2305 		return (ENOBUFS);
2306 	}
2307 
2308 	LLE_WLOCK_ASSERT(lle);
2309 
2310 	/*
2311 	 * The first time we send a packet to a neighbor whose entry is
2312 	 * STALE, we have to change the state to DELAY and a sets a timer to
2313 	 * expire in DELAY_FIRST_PROBE_TIME seconds to ensure do
2314 	 * neighbor unreachability detection on expiration.
2315 	 * (RFC 2461 7.3.3)
2316 	 */
2317 	if (lle->ln_state == ND6_LLINFO_STALE)
2318 		nd6_llinfo_setstate(lle, ND6_LLINFO_DELAY);
2319 
2320 	/*
2321 	 * If the neighbor cache entry has a state other than INCOMPLETE
2322 	 * (i.e. its link-layer address is already resolved), just
2323 	 * send the packet.
2324 	 */
2325 	if (lle->ln_state > ND6_LLINFO_INCOMPLETE) {
2326 		if (flags & LLE_ADDRONLY) {
2327 			lladdr = lle->ll_addr;
2328 			ll_len = ifp->if_addrlen;
2329 		} else {
2330 			lladdr = lle->r_linkdata;
2331 			ll_len = lle->r_hdrlen;
2332 		}
2333 		bcopy(lladdr, desten, ll_len);
2334 		if (pflags != NULL)
2335 			*pflags = lle->la_flags;
2336 		LLE_WUNLOCK(lle);
2337 		return (0);
2338 	}
2339 
2340 	/*
2341 	 * There is a neighbor cache entry, but no ethernet address
2342 	 * response yet.  Append this latest packet to the end of the
2343 	 * packet queue in the mbuf, unless the number of the packet
2344 	 * does not exceed nd6_maxqueuelen.  When it exceeds nd6_maxqueuelen,
2345 	 * the oldest packet in the queue will be removed.
2346 	 */
2347 
2348 	if (lle->la_hold != NULL) {
2349 		struct mbuf *m_hold;
2350 		int i;
2351 
2352 		i = 0;
2353 		for (m_hold = lle->la_hold; m_hold; m_hold = m_hold->m_nextpkt){
2354 			i++;
2355 			if (m_hold->m_nextpkt == NULL) {
2356 				m_hold->m_nextpkt = m;
2357 				break;
2358 			}
2359 		}
2360 		while (i >= V_nd6_maxqueuelen) {
2361 			m_hold = lle->la_hold;
2362 			lle->la_hold = lle->la_hold->m_nextpkt;
2363 			m_freem(m_hold);
2364 			i--;
2365 		}
2366 	} else {
2367 		lle->la_hold = m;
2368 	}
2369 
2370 	/*
2371 	 * If there has been no NS for the neighbor after entering the
2372 	 * INCOMPLETE state, send the first solicitation.
2373 	 * Note that for newly-created lle la_asked will be 0,
2374 	 * so we will transition from ND6_LLINFO_NOSTATE to
2375 	 * ND6_LLINFO_INCOMPLETE state here.
2376 	 */
2377 	psrc = NULL;
2378 	send_ns = 0;
2379 	if (lle->la_asked == 0) {
2380 		lle->la_asked++;
2381 		send_ns = 1;
2382 		psrc = nd6_llinfo_get_holdsrc(lle, &src);
2383 
2384 		nd6_llinfo_setstate(lle, ND6_LLINFO_INCOMPLETE);
2385 	}
2386 	LLE_WUNLOCK(lle);
2387 	if (send_ns != 0)
2388 		nd6_ns_output(ifp, psrc, NULL, &dst->sin6_addr, NULL);
2389 
2390 	return (EWOULDBLOCK);
2391 }
2392 
2393 /*
2394  * Do L2 address resolution for @sa_dst address. Stores found
2395  * address in @desten buffer. Copy of lle ln_flags can be also
2396  * saved in @pflags if @pflags is non-NULL.
2397  *
2398  * Return values:
2399  * - 0 on success (address copied to buffer).
2400  * - EWOULDBLOCK (no local error, but address is still unresolved)
2401  * - other errors (alloc failure, etc)
2402  */
2403 int
2404 nd6_resolve_addr(struct ifnet *ifp, int flags, const struct sockaddr *dst,
2405     char *desten, uint32_t *pflags)
2406 {
2407 	int error;
2408 
2409 	flags |= LLE_ADDRONLY;
2410 	error = nd6_resolve_slow(ifp, flags, NULL,
2411 	    (const struct sockaddr_in6 *)dst, desten, pflags);
2412 	return (error);
2413 }
2414 
2415 int
2416 nd6_flush_holdchain(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *chain,
2417     struct sockaddr_in6 *dst)
2418 {
2419 	struct mbuf *m, *m_head;
2420 	struct ifnet *outifp;
2421 	int error = 0;
2422 
2423 	m_head = chain;
2424 	if ((ifp->if_flags & IFF_LOOPBACK) != 0)
2425 		outifp = origifp;
2426 	else
2427 		outifp = ifp;
2428 
2429 	while (m_head) {
2430 		m = m_head;
2431 		m_head = m_head->m_nextpkt;
2432 		error = nd6_output_ifp(ifp, origifp, m, dst, NULL);
2433 	}
2434 
2435 	/*
2436 	 * XXX
2437 	 * note that intermediate errors are blindly ignored
2438 	 */
2439 	return (error);
2440 }
2441 
2442 static int
2443 nd6_need_cache(struct ifnet *ifp)
2444 {
2445 	/*
2446 	 * XXX: we currently do not make neighbor cache on any interface
2447 	 * other than ARCnet, Ethernet, FDDI and GIF.
2448 	 *
2449 	 * RFC2893 says:
2450 	 * - unidirectional tunnels needs no ND
2451 	 */
2452 	switch (ifp->if_type) {
2453 	case IFT_ARCNET:
2454 	case IFT_ETHER:
2455 	case IFT_FDDI:
2456 	case IFT_IEEE1394:
2457 	case IFT_L2VLAN:
2458 	case IFT_IEEE80211:
2459 	case IFT_INFINIBAND:
2460 	case IFT_BRIDGE:
2461 	case IFT_PROPVIRTUAL:
2462 		return (1);
2463 	default:
2464 		return (0);
2465 	}
2466 }
2467 
2468 /*
2469  * Add pernament ND6 link-layer record for given
2470  * interface address.
2471  *
2472  * Very similar to IPv4 arp_ifinit(), but:
2473  * 1) IPv6 DAD is performed in different place
2474  * 2) It is called by IPv6 protocol stack in contrast to
2475  * arp_ifinit() which is typically called in SIOCSIFADDR
2476  * driver ioctl handler.
2477  *
2478  */
2479 int
2480 nd6_add_ifa_lle(struct in6_ifaddr *ia)
2481 {
2482 	struct ifnet *ifp;
2483 	struct llentry *ln, *ln_tmp;
2484 	struct sockaddr *dst;
2485 
2486 	ifp = ia->ia_ifa.ifa_ifp;
2487 	if (nd6_need_cache(ifp) == 0)
2488 		return (0);
2489 
2490 	ia->ia_ifa.ifa_rtrequest = nd6_rtrequest;
2491 	dst = (struct sockaddr *)&ia->ia_addr;
2492 	ln = lltable_alloc_entry(LLTABLE6(ifp), LLE_IFADDR, dst);
2493 	if (ln == NULL)
2494 		return (ENOBUFS);
2495 
2496 	IF_AFDATA_WLOCK(ifp);
2497 	LLE_WLOCK(ln);
2498 	/* Unlink any entry if exists */
2499 	ln_tmp = lla_lookup(LLTABLE6(ifp), LLE_EXCLUSIVE, dst);
2500 	if (ln_tmp != NULL)
2501 		lltable_unlink_entry(LLTABLE6(ifp), ln_tmp);
2502 	lltable_link_entry(LLTABLE6(ifp), ln);
2503 	IF_AFDATA_WUNLOCK(ifp);
2504 
2505 	if (ln_tmp != NULL)
2506 		EVENTHANDLER_INVOKE(lle_event, ln_tmp, LLENTRY_EXPIRED);
2507 	EVENTHANDLER_INVOKE(lle_event, ln, LLENTRY_RESOLVED);
2508 
2509 	LLE_WUNLOCK(ln);
2510 	if (ln_tmp != NULL)
2511 		llentry_free(ln_tmp);
2512 
2513 	return (0);
2514 }
2515 
2516 /*
2517  * Removes either all lle entries for given @ia, or lle
2518  * corresponding to @ia address.
2519  */
2520 void
2521 nd6_rem_ifa_lle(struct in6_ifaddr *ia, int all)
2522 {
2523 	struct sockaddr_in6 mask, addr;
2524 	struct sockaddr *saddr, *smask;
2525 	struct ifnet *ifp;
2526 
2527 	ifp = ia->ia_ifa.ifa_ifp;
2528 	memcpy(&addr, &ia->ia_addr, sizeof(ia->ia_addr));
2529 	memcpy(&mask, &ia->ia_prefixmask, sizeof(ia->ia_prefixmask));
2530 	saddr = (struct sockaddr *)&addr;
2531 	smask = (struct sockaddr *)&mask;
2532 
2533 	if (all != 0)
2534 		lltable_prefix_free(AF_INET6, saddr, smask, LLE_STATIC);
2535 	else
2536 		lltable_delete_addr(LLTABLE6(ifp), LLE_IFADDR, saddr);
2537 }
2538 
2539 static void
2540 clear_llinfo_pqueue(struct llentry *ln)
2541 {
2542 	struct mbuf *m_hold, *m_hold_next;
2543 
2544 	for (m_hold = ln->la_hold; m_hold; m_hold = m_hold_next) {
2545 		m_hold_next = m_hold->m_nextpkt;
2546 		m_freem(m_hold);
2547 	}
2548 
2549 	ln->la_hold = NULL;
2550 }
2551 
2552 static int nd6_sysctl_drlist(SYSCTL_HANDLER_ARGS);
2553 static int nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS);
2554 #ifdef SYSCTL_DECL
2555 SYSCTL_DECL(_net_inet6_icmp6);
2556 #endif
2557 SYSCTL_NODE(_net_inet6_icmp6, ICMPV6CTL_ND6_DRLIST, nd6_drlist,
2558 	CTLFLAG_RD, nd6_sysctl_drlist, "");
2559 SYSCTL_NODE(_net_inet6_icmp6, ICMPV6CTL_ND6_PRLIST, nd6_prlist,
2560 	CTLFLAG_RD, nd6_sysctl_prlist, "");
2561 SYSCTL_INT(_net_inet6_icmp6, ICMPV6CTL_ND6_MAXQLEN, nd6_maxqueuelen,
2562 	CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(nd6_maxqueuelen), 1, "");
2563 SYSCTL_INT(_net_inet6_icmp6, OID_AUTO, nd6_gctimer,
2564 	CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(nd6_gctimer), (60 * 60 * 24), "");
2565 
2566 static int
2567 nd6_sysctl_drlist(SYSCTL_HANDLER_ARGS)
2568 {
2569 	struct in6_defrouter d;
2570 	struct nd_defrouter *dr;
2571 	int error;
2572 
2573 	if (req->newptr != NULL)
2574 		return (EPERM);
2575 
2576 	error = sysctl_wire_old_buffer(req, 0);
2577 	if (error != 0)
2578 		return (error);
2579 
2580 	bzero(&d, sizeof(d));
2581 	d.rtaddr.sin6_family = AF_INET6;
2582 	d.rtaddr.sin6_len = sizeof(d.rtaddr);
2583 
2584 	ND6_RLOCK();
2585 	TAILQ_FOREACH(dr, &V_nd_defrouter, dr_entry) {
2586 		d.rtaddr.sin6_addr = dr->rtaddr;
2587 		error = sa6_recoverscope(&d.rtaddr);
2588 		if (error != 0)
2589 			break;
2590 		d.flags = dr->raflags;
2591 		d.rtlifetime = dr->rtlifetime;
2592 		d.expire = dr->expire + (time_second - time_uptime);
2593 		d.if_index = dr->ifp->if_index;
2594 		error = SYSCTL_OUT(req, &d, sizeof(d));
2595 		if (error != 0)
2596 			break;
2597 	}
2598 	ND6_RUNLOCK();
2599 	return (error);
2600 }
2601 
2602 static int
2603 nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS)
2604 {
2605 	struct in6_prefix p;
2606 	struct sockaddr_in6 s6;
2607 	struct nd_prefix *pr;
2608 	struct nd_pfxrouter *pfr;
2609 	time_t maxexpire;
2610 	int error;
2611 	char ip6buf[INET6_ADDRSTRLEN];
2612 
2613 	if (req->newptr)
2614 		return (EPERM);
2615 
2616 	bzero(&p, sizeof(p));
2617 	p.origin = PR_ORIG_RA;
2618 	bzero(&s6, sizeof(s6));
2619 	s6.sin6_family = AF_INET6;
2620 	s6.sin6_len = sizeof(s6);
2621 
2622 	/*
2623 	 * XXX locking
2624 	 */
2625 	LIST_FOREACH(pr, &V_nd_prefix, ndpr_entry) {
2626 		p.prefix = pr->ndpr_prefix;
2627 		if (sa6_recoverscope(&p.prefix)) {
2628 			log(LOG_ERR, "scope error in prefix list (%s)\n",
2629 			    ip6_sprintf(ip6buf, &p.prefix.sin6_addr));
2630 			/* XXX: press on... */
2631 		}
2632 		p.raflags = pr->ndpr_raf;
2633 		p.prefixlen = pr->ndpr_plen;
2634 		p.vltime = pr->ndpr_vltime;
2635 		p.pltime = pr->ndpr_pltime;
2636 		p.if_index = pr->ndpr_ifp->if_index;
2637 		if (pr->ndpr_vltime == ND6_INFINITE_LIFETIME)
2638 			p.expire = 0;
2639 		else {
2640 			/* XXX: we assume time_t is signed. */
2641 			maxexpire = (-1) &
2642 			    ~((time_t)1 << ((sizeof(maxexpire) * 8) - 1));
2643 			if (pr->ndpr_vltime < maxexpire - pr->ndpr_lastupdate)
2644 				p.expire = pr->ndpr_lastupdate +
2645 				    pr->ndpr_vltime +
2646 				    (time_second - time_uptime);
2647 			else
2648 				p.expire = maxexpire;
2649 		}
2650 		p.refcnt = pr->ndpr_refcnt;
2651 		p.flags = pr->ndpr_stateflags;
2652 		p.advrtrs = 0;
2653 		LIST_FOREACH(pfr, &pr->ndpr_advrtrs, pfr_entry)
2654 			p.advrtrs++;
2655 		error = SYSCTL_OUT(req, &p, sizeof(p));
2656 		if (error != 0)
2657 			return (error);
2658 		LIST_FOREACH(pfr, &pr->ndpr_advrtrs, pfr_entry) {
2659 			s6.sin6_addr = pfr->router->rtaddr;
2660 			if (sa6_recoverscope(&s6))
2661 				log(LOG_ERR,
2662 				    "scope error in prefix list (%s)\n",
2663 				    ip6_sprintf(ip6buf, &pfr->router->rtaddr));
2664 			error = SYSCTL_OUT(req, &s6, sizeof(s6));
2665 			if (error != 0)
2666 				return (error);
2667 		}
2668 	}
2669 	return (0);
2670 }
2671