xref: /freebsd/sys/netinet6/nd6.c (revision 3884d6f8bd7e13a2b152fedc2d0d4f6de5ad17b8)
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 
119 VNET_DEFINE(int, nd6_recalc_reachtm_interval) = ND6_RECALC_REACHTM_INTERVAL;
120 #define	V_nd6_recalc_reachtm_interval	VNET(nd6_recalc_reachtm_interval)
121 
122 int	(*send_sendso_input_hook)(struct mbuf *, struct ifnet *, int, int);
123 
124 static int nd6_is_new_addr_neighbor(const struct sockaddr_in6 *,
125 	struct ifnet *);
126 static void nd6_setmtu0(struct ifnet *, struct nd_ifinfo *);
127 static void nd6_slowtimo(void *);
128 static int regen_tmpaddr(struct in6_ifaddr *);
129 static void nd6_free(struct llentry *, int);
130 static void nd6_free_redirect(const struct llentry *);
131 static void nd6_llinfo_timer(void *);
132 static void nd6_llinfo_settimer_locked(struct llentry *, long);
133 static void clear_llinfo_pqueue(struct llentry *);
134 static void nd6_rtrequest(int, struct rtentry *, struct rt_addrinfo *);
135 static int nd6_resolve_slow(struct ifnet *, int, struct mbuf *,
136     const struct sockaddr_in6 *, u_char *, uint32_t *);
137 static int nd6_need_cache(struct ifnet *);
138 
139 
140 static VNET_DEFINE(struct callout, nd6_slowtimo_ch);
141 #define	V_nd6_slowtimo_ch		VNET(nd6_slowtimo_ch)
142 
143 VNET_DEFINE(struct callout, nd6_timer_ch);
144 
145 static void
146 nd6_lle_event(void *arg __unused, struct llentry *lle, int evt)
147 {
148 	struct rt_addrinfo rtinfo;
149 	struct sockaddr_in6 dst;
150 	struct sockaddr_dl gw;
151 	struct ifnet *ifp;
152 	int type;
153 
154 	LLE_WLOCK_ASSERT(lle);
155 
156 	if (lltable_get_af(lle->lle_tbl) != AF_INET6)
157 		return;
158 
159 	switch (evt) {
160 	case LLENTRY_RESOLVED:
161 		type = RTM_ADD;
162 		KASSERT(lle->la_flags & LLE_VALID,
163 		    ("%s: %p resolved but not valid?", __func__, lle));
164 		break;
165 	case LLENTRY_EXPIRED:
166 		type = RTM_DELETE;
167 		break;
168 	default:
169 		return;
170 	}
171 
172 	ifp = lltable_get_ifp(lle->lle_tbl);
173 
174 	bzero(&dst, sizeof(dst));
175 	bzero(&gw, sizeof(gw));
176 	bzero(&rtinfo, sizeof(rtinfo));
177 	lltable_fill_sa_entry(lle, (struct sockaddr *)&dst);
178 	dst.sin6_scope_id = in6_getscopezone(ifp,
179 	    in6_addrscope(&dst.sin6_addr));
180 	gw.sdl_len = sizeof(struct sockaddr_dl);
181 	gw.sdl_family = AF_LINK;
182 	gw.sdl_alen = ifp->if_addrlen;
183 	gw.sdl_index = ifp->if_index;
184 	gw.sdl_type = ifp->if_type;
185 	if (evt == LLENTRY_RESOLVED)
186 		bcopy(lle->ll_addr, gw.sdl_data, ifp->if_addrlen);
187 	rtinfo.rti_info[RTAX_DST] = (struct sockaddr *)&dst;
188 	rtinfo.rti_info[RTAX_GATEWAY] = (struct sockaddr *)&gw;
189 	rtinfo.rti_addrs = RTA_DST | RTA_GATEWAY;
190 	rt_missmsg_fib(type, &rtinfo, RTF_HOST | RTF_LLDATA | (
191 	    type == RTM_ADD ? RTF_UP: 0), 0, RT_DEFAULT_FIB);
192 }
193 
194 /*
195  * A handler for interface link layer address change event.
196  */
197 static void
198 nd6_iflladdr(void *arg __unused, struct ifnet *ifp)
199 {
200 
201 	lltable_update_ifaddr(LLTABLE6(ifp));
202 }
203 
204 void
205 nd6_init(void)
206 {
207 
208 	LIST_INIT(&V_nd_prefix);
209 
210 	/* initialization of the default router list */
211 	TAILQ_INIT(&V_nd_defrouter);
212 
213 	/* start timer */
214 	callout_init(&V_nd6_slowtimo_ch, 0);
215 	callout_reset(&V_nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
216 	    nd6_slowtimo, curvnet);
217 
218 	nd6_dad_init();
219 	if (IS_DEFAULT_VNET(curvnet)) {
220 		lle_event_eh = EVENTHANDLER_REGISTER(lle_event, nd6_lle_event,
221 		    NULL, EVENTHANDLER_PRI_ANY);
222 		iflladdr_event_eh = EVENTHANDLER_REGISTER(iflladdr_event,
223 		    nd6_iflladdr, NULL, EVENTHANDLER_PRI_ANY);
224 	}
225 }
226 
227 #ifdef VIMAGE
228 void
229 nd6_destroy()
230 {
231 
232 	callout_drain(&V_nd6_slowtimo_ch);
233 	callout_drain(&V_nd6_timer_ch);
234 	if (IS_DEFAULT_VNET(curvnet)) {
235 		EVENTHANDLER_DEREGISTER(lle_event, lle_event_eh);
236 		EVENTHANDLER_DEREGISTER(iflladdr_event, iflladdr_event_eh);
237 	}
238 }
239 #endif
240 
241 struct nd_ifinfo *
242 nd6_ifattach(struct ifnet *ifp)
243 {
244 	struct nd_ifinfo *nd;
245 
246 	nd = malloc(sizeof(*nd), M_IP6NDP, M_WAITOK | M_ZERO);
247 	nd->initialized = 1;
248 
249 	nd->chlim = IPV6_DEFHLIM;
250 	nd->basereachable = REACHABLE_TIME;
251 	nd->reachable = ND_COMPUTE_RTIME(nd->basereachable);
252 	nd->retrans = RETRANS_TIMER;
253 
254 	nd->flags = ND6_IFF_PERFORMNUD;
255 
256 	/* A loopback interface always has ND6_IFF_AUTO_LINKLOCAL.
257 	 * XXXHRS: Clear ND6_IFF_AUTO_LINKLOCAL on an IFT_BRIDGE interface by
258 	 * default regardless of the V_ip6_auto_linklocal configuration to
259 	 * give a reasonable default behavior.
260 	 */
261 	if ((V_ip6_auto_linklocal && ifp->if_type != IFT_BRIDGE) ||
262 	    (ifp->if_flags & IFF_LOOPBACK))
263 		nd->flags |= ND6_IFF_AUTO_LINKLOCAL;
264 	/*
265 	 * A loopback interface does not need to accept RTADV.
266 	 * XXXHRS: Clear ND6_IFF_ACCEPT_RTADV on an IFT_BRIDGE interface by
267 	 * default regardless of the V_ip6_accept_rtadv configuration to
268 	 * prevent the interface from accepting RA messages arrived
269 	 * on one of the member interfaces with ND6_IFF_ACCEPT_RTADV.
270 	 */
271 	if (V_ip6_accept_rtadv &&
272 	    !(ifp->if_flags & IFF_LOOPBACK) &&
273 	    (ifp->if_type != IFT_BRIDGE))
274 			nd->flags |= ND6_IFF_ACCEPT_RTADV;
275 	if (V_ip6_no_radr && !(ifp->if_flags & IFF_LOOPBACK))
276 		nd->flags |= ND6_IFF_NO_RADR;
277 
278 	/* XXX: we cannot call nd6_setmtu since ifp is not fully initialized */
279 	nd6_setmtu0(ifp, nd);
280 
281 	return nd;
282 }
283 
284 void
285 nd6_ifdetach(struct nd_ifinfo *nd)
286 {
287 
288 	free(nd, M_IP6NDP);
289 }
290 
291 /*
292  * Reset ND level link MTU. This function is called when the physical MTU
293  * changes, which means we might have to adjust the ND level MTU.
294  */
295 void
296 nd6_setmtu(struct ifnet *ifp)
297 {
298 	if (ifp->if_afdata[AF_INET6] == NULL)
299 		return;
300 
301 	nd6_setmtu0(ifp, ND_IFINFO(ifp));
302 }
303 
304 /* XXX todo: do not maintain copy of ifp->if_mtu in ndi->maxmtu */
305 void
306 nd6_setmtu0(struct ifnet *ifp, struct nd_ifinfo *ndi)
307 {
308 	u_int32_t omaxmtu;
309 
310 	omaxmtu = ndi->maxmtu;
311 
312 	switch (ifp->if_type) {
313 	case IFT_ARCNET:
314 		ndi->maxmtu = MIN(ARC_PHDS_MAXMTU, ifp->if_mtu); /* RFC2497 */
315 		break;
316 	case IFT_FDDI:
317 		ndi->maxmtu = MIN(FDDIIPMTU, ifp->if_mtu); /* RFC2467 */
318 		break;
319 	case IFT_ISO88025:
320 		 ndi->maxmtu = MIN(ISO88025_MAX_MTU, ifp->if_mtu);
321 		 break;
322 	default:
323 		ndi->maxmtu = ifp->if_mtu;
324 		break;
325 	}
326 
327 	/*
328 	 * Decreasing the interface MTU under IPV6 minimum MTU may cause
329 	 * undesirable situation.  We thus notify the operator of the change
330 	 * explicitly.  The check for omaxmtu is necessary to restrict the
331 	 * log to the case of changing the MTU, not initializing it.
332 	 */
333 	if (omaxmtu >= IPV6_MMTU && ndi->maxmtu < IPV6_MMTU) {
334 		log(LOG_NOTICE, "nd6_setmtu0: "
335 		    "new link MTU on %s (%lu) is too small for IPv6\n",
336 		    if_name(ifp), (unsigned long)ndi->maxmtu);
337 	}
338 
339 	if (ndi->maxmtu > V_in6_maxmtu)
340 		in6_setmaxmtu(); /* check all interfaces just in case */
341 
342 }
343 
344 void
345 nd6_option_init(void *opt, int icmp6len, union nd_opts *ndopts)
346 {
347 
348 	bzero(ndopts, sizeof(*ndopts));
349 	ndopts->nd_opts_search = (struct nd_opt_hdr *)opt;
350 	ndopts->nd_opts_last
351 		= (struct nd_opt_hdr *)(((u_char *)opt) + icmp6len);
352 
353 	if (icmp6len == 0) {
354 		ndopts->nd_opts_done = 1;
355 		ndopts->nd_opts_search = NULL;
356 	}
357 }
358 
359 /*
360  * Take one ND option.
361  */
362 struct nd_opt_hdr *
363 nd6_option(union nd_opts *ndopts)
364 {
365 	struct nd_opt_hdr *nd_opt;
366 	int olen;
367 
368 	KASSERT(ndopts != NULL, ("%s: ndopts == NULL", __func__));
369 	KASSERT(ndopts->nd_opts_last != NULL, ("%s: uninitialized ndopts",
370 	    __func__));
371 	if (ndopts->nd_opts_search == NULL)
372 		return NULL;
373 	if (ndopts->nd_opts_done)
374 		return NULL;
375 
376 	nd_opt = ndopts->nd_opts_search;
377 
378 	/* make sure nd_opt_len is inside the buffer */
379 	if ((caddr_t)&nd_opt->nd_opt_len >= (caddr_t)ndopts->nd_opts_last) {
380 		bzero(ndopts, sizeof(*ndopts));
381 		return NULL;
382 	}
383 
384 	olen = nd_opt->nd_opt_len << 3;
385 	if (olen == 0) {
386 		/*
387 		 * Message validation requires that all included
388 		 * options have a length that is greater than zero.
389 		 */
390 		bzero(ndopts, sizeof(*ndopts));
391 		return NULL;
392 	}
393 
394 	ndopts->nd_opts_search = (struct nd_opt_hdr *)((caddr_t)nd_opt + olen);
395 	if (ndopts->nd_opts_search > ndopts->nd_opts_last) {
396 		/* option overruns the end of buffer, invalid */
397 		bzero(ndopts, sizeof(*ndopts));
398 		return NULL;
399 	} else if (ndopts->nd_opts_search == ndopts->nd_opts_last) {
400 		/* reached the end of options chain */
401 		ndopts->nd_opts_done = 1;
402 		ndopts->nd_opts_search = NULL;
403 	}
404 	return nd_opt;
405 }
406 
407 /*
408  * Parse multiple ND options.
409  * This function is much easier to use, for ND routines that do not need
410  * multiple options of the same type.
411  */
412 int
413 nd6_options(union nd_opts *ndopts)
414 {
415 	struct nd_opt_hdr *nd_opt;
416 	int i = 0;
417 
418 	KASSERT(ndopts != NULL, ("%s: ndopts == NULL", __func__));
419 	KASSERT(ndopts->nd_opts_last != NULL, ("%s: uninitialized ndopts",
420 	    __func__));
421 	if (ndopts->nd_opts_search == NULL)
422 		return 0;
423 
424 	while (1) {
425 		nd_opt = nd6_option(ndopts);
426 		if (nd_opt == NULL && ndopts->nd_opts_last == NULL) {
427 			/*
428 			 * Message validation requires that all included
429 			 * options have a length that is greater than zero.
430 			 */
431 			ICMP6STAT_INC(icp6s_nd_badopt);
432 			bzero(ndopts, sizeof(*ndopts));
433 			return -1;
434 		}
435 
436 		if (nd_opt == NULL)
437 			goto skip1;
438 
439 		switch (nd_opt->nd_opt_type) {
440 		case ND_OPT_SOURCE_LINKADDR:
441 		case ND_OPT_TARGET_LINKADDR:
442 		case ND_OPT_MTU:
443 		case ND_OPT_REDIRECTED_HEADER:
444 		case ND_OPT_NONCE:
445 			if (ndopts->nd_opt_array[nd_opt->nd_opt_type]) {
446 				nd6log((LOG_INFO,
447 				    "duplicated ND6 option found (type=%d)\n",
448 				    nd_opt->nd_opt_type));
449 				/* XXX bark? */
450 			} else {
451 				ndopts->nd_opt_array[nd_opt->nd_opt_type]
452 					= nd_opt;
453 			}
454 			break;
455 		case ND_OPT_PREFIX_INFORMATION:
456 			if (ndopts->nd_opt_array[nd_opt->nd_opt_type] == 0) {
457 				ndopts->nd_opt_array[nd_opt->nd_opt_type]
458 					= nd_opt;
459 			}
460 			ndopts->nd_opts_pi_end =
461 				(struct nd_opt_prefix_info *)nd_opt;
462 			break;
463 		/* What about ND_OPT_ROUTE_INFO? RFC 4191 */
464 		case ND_OPT_RDNSS:	/* RFC 6106 */
465 		case ND_OPT_DNSSL:	/* RFC 6106 */
466 			/*
467 			 * Silently ignore options we know and do not care about
468 			 * in the kernel.
469 			 */
470 			break;
471 		default:
472 			/*
473 			 * Unknown options must be silently ignored,
474 			 * to accomodate future extension to the protocol.
475 			 */
476 			nd6log((LOG_DEBUG,
477 			    "nd6_options: unsupported option %d - "
478 			    "option ignored\n", nd_opt->nd_opt_type));
479 		}
480 
481 skip1:
482 		i++;
483 		if (i > V_nd6_maxndopt) {
484 			ICMP6STAT_INC(icp6s_nd_toomanyopt);
485 			nd6log((LOG_INFO, "too many loop in nd opt\n"));
486 			break;
487 		}
488 
489 		if (ndopts->nd_opts_done)
490 			break;
491 	}
492 
493 	return 0;
494 }
495 
496 /*
497  * ND6 timer routine to handle ND6 entries
498  */
499 static void
500 nd6_llinfo_settimer_locked(struct llentry *ln, long tick)
501 {
502 	int canceled;
503 
504 	LLE_WLOCK_ASSERT(ln);
505 
506 	if (tick < 0) {
507 		ln->la_expire = 0;
508 		ln->ln_ntick = 0;
509 		canceled = callout_stop(&ln->lle_timer);
510 	} else {
511 		ln->la_expire = time_uptime + tick / hz;
512 		LLE_ADDREF(ln);
513 		if (tick > INT_MAX) {
514 			ln->ln_ntick = tick - INT_MAX;
515 			canceled = callout_reset(&ln->lle_timer, INT_MAX,
516 			    nd6_llinfo_timer, ln);
517 		} else {
518 			ln->ln_ntick = 0;
519 			canceled = callout_reset(&ln->lle_timer, tick,
520 			    nd6_llinfo_timer, ln);
521 		}
522 	}
523 	if (canceled > 0)
524 		LLE_REMREF(ln);
525 }
526 
527 /*
528  * Gets source address of the first packet in hold queue
529  * and stores it in @src.
530  * Returns pointer to @src (if hold queue is not empty) or NULL.
531  *
532  * Set noinline to be dtrace-friendly
533  */
534 static __noinline struct in6_addr *
535 nd6_llinfo_get_holdsrc(struct llentry *ln, struct in6_addr *src)
536 {
537 	struct ip6_hdr hdr;
538 	struct mbuf *m;
539 
540 	if (ln->la_hold == NULL)
541 		return (NULL);
542 
543 	/*
544 	 * assume every packet in la_hold has the same IP header
545 	 */
546 	m = ln->la_hold;
547 	if (sizeof(hdr) > m->m_len)
548 		return (NULL);
549 
550 	m_copydata(m, 0, sizeof(hdr), (caddr_t)&hdr);
551 	*src = hdr.ip6_src;
552 
553 	return (src);
554 }
555 
556 /*
557  * Checks if we need to switch from STALE state.
558  *
559  * RFC 4861 requires switching from STALE to DELAY state
560  * on first packet matching entry, waiting V_nd6_delay and
561  * transition to PROBE state (if upper layer confirmation was
562  * not received).
563  *
564  * This code performs a bit differently:
565  * On packet hit we don't change state (but desired state
566  * can be guessed by control plane). However, after V_nd6_delay
567  * seconds code will transition to PROBE state (so DELAY state
568  * is kinda skipped in most situations).
569  *
570  * Typically, V_nd6_gctimer is bigger than V_nd6_delay, so
571  * we perform the following upon entering STALE state:
572  *
573  * 1) Arm timer to run each V_nd6_delay seconds to make sure that
574  * if packet was transmitted at the start of given interval, we
575  * would be able to switch to PROBE state in V_nd6_delay seconds
576  * as user expects.
577  *
578  * 2) Reschedule timer until original V_nd6_gctimer expires keeping
579  * lle in STALE state (remaining timer value stored in lle_remtime).
580  *
581  * 3) Reschedule timer if packet was transmitted less that V_nd6_delay
582  * seconds ago.
583  *
584  * Returns non-zero value if the entry is still STALE (storing
585  * the next timer interval in @pdelay).
586  *
587  * Returns zero value if original timer expired or we need to switch to
588  * PROBE (store that in @do_switch variable).
589  */
590 static int
591 nd6_is_stale(struct llentry *lle, long *pdelay, int *do_switch)
592 {
593 	int nd_delay, nd_gctimer, r_skip_req;
594 	time_t lle_hittime;
595 	long delay;
596 
597 	*do_switch = 0;
598 	nd_gctimer = V_nd6_gctimer;
599 	nd_delay = V_nd6_delay;
600 
601 	LLE_REQ_LOCK(lle);
602 	r_skip_req = lle->r_skip_req;
603 	lle_hittime = lle->lle_hittime;
604 	LLE_REQ_UNLOCK(lle);
605 
606 	if (r_skip_req > 0) {
607 
608 		/*
609 		 * Nonzero r_skip_req value was set upon entering
610 		 * STALE state. Since value was not changed, no
611 		 * packets were passed using this lle. Ask for
612 		 * timer reschedule and keep STALE state.
613 		 */
614 		delay = (long)(MIN(nd_gctimer, nd_delay));
615 		delay *= hz;
616 		if (lle->lle_remtime > delay)
617 			lle->lle_remtime -= delay;
618 		else {
619 			delay = lle->lle_remtime;
620 			lle->lle_remtime = 0;
621 		}
622 
623 		if (delay == 0) {
624 
625 			/*
626 			 * The original ng6_gctime timeout ended,
627 			 * no more rescheduling.
628 			 */
629 			return (0);
630 		}
631 
632 		*pdelay = delay;
633 		return (1);
634 	}
635 
636 	/*
637 	 * Packet received. Verify timestamp
638 	 */
639 	delay = (long)(time_uptime - lle_hittime);
640 	if (delay < nd_delay) {
641 
642 		/*
643 		 * V_nd6_delay still not passed since the first
644 		 * hit in STALE state.
645 		 * Reshedule timer and return.
646 		 */
647 		*pdelay = (long)(nd_delay - delay) * hz;
648 		return (1);
649 	}
650 
651 	/* Request switching to probe */
652 	*do_switch = 1;
653 	return (0);
654 }
655 
656 
657 /*
658  * Switch @lle state to new state optionally arming timers.
659  *
660  * Set noinline to be dtrace-friendly
661  */
662 __noinline void
663 nd6_llinfo_setstate(struct llentry *lle, int newstate)
664 {
665 	struct ifnet *ifp;
666 	int nd_gctimer, nd_delay;
667 	long delay, remtime;
668 
669 	delay = 0;
670 	remtime = 0;
671 
672 	switch (newstate) {
673 	case ND6_LLINFO_INCOMPLETE:
674 		ifp = lle->lle_tbl->llt_ifp;
675 		delay = (long)ND_IFINFO(ifp)->retrans * hz / 1000;
676 		break;
677 	case ND6_LLINFO_REACHABLE:
678 		if (!ND6_LLINFO_PERMANENT(lle)) {
679 			ifp = lle->lle_tbl->llt_ifp;
680 			delay = (long)ND_IFINFO(ifp)->reachable * hz;
681 		}
682 		break;
683 	case ND6_LLINFO_STALE:
684 
685 		/*
686 		 * Notify fast path that we want to know if any packet
687 		 * is transmitted by setting r_skip_req.
688 		 */
689 		LLE_REQ_LOCK(lle);
690 		lle->r_skip_req = 1;
691 		LLE_REQ_UNLOCK(lle);
692 		nd_delay = V_nd6_delay;
693 		nd_gctimer = V_nd6_gctimer;
694 
695 		delay = (long)(MIN(nd_gctimer, nd_delay)) * hz;
696 		remtime = (long)nd_gctimer * hz - delay;
697 		break;
698 	case ND6_LLINFO_DELAY:
699 		lle->la_asked = 0;
700 		delay = (long)V_nd6_delay * hz;
701 		break;
702 	}
703 
704 	if (delay > 0)
705 		nd6_llinfo_settimer_locked(lle, delay);
706 
707 	lle->lle_remtime = remtime;
708 	lle->ln_state = newstate;
709 }
710 
711 /*
712  * Timer-dependent part of nd state machine.
713  *
714  * Set noinline to be dtrace-friendly
715  */
716 static __noinline void
717 nd6_llinfo_timer(void *arg)
718 {
719 	struct llentry *ln;
720 	struct in6_addr *dst, *pdst, *psrc, src;
721 	struct ifnet *ifp;
722 	struct nd_ifinfo *ndi = NULL;
723 	int do_switch, send_ns;
724 	long delay;
725 
726 	KASSERT(arg != NULL, ("%s: arg NULL", __func__));
727 	ln = (struct llentry *)arg;
728 	LLE_WLOCK(ln);
729 	if (callout_pending(&ln->lle_timer)) {
730 		/*
731 		 * Here we are a bit odd here in the treatment of
732 		 * active/pending. If the pending bit is set, it got
733 		 * rescheduled before I ran. The active
734 		 * bit we ignore, since if it was stopped
735 		 * in ll_tablefree() and was currently running
736 		 * it would have return 0 so the code would
737 		 * not have deleted it since the callout could
738 		 * not be stopped so we want to go through
739 		 * with the delete here now. If the callout
740 		 * was restarted, the pending bit will be back on and
741 		 * we just want to bail since the callout_reset would
742 		 * return 1 and our reference would have been removed
743 		 * by nd6_llinfo_settimer_locked above since canceled
744 		 * would have been 1.
745 		 */
746 		LLE_WUNLOCK(ln);
747 		return;
748 	}
749 	ifp = ln->lle_tbl->llt_ifp;
750 	CURVNET_SET(ifp->if_vnet);
751 	ndi = ND_IFINFO(ifp);
752 	send_ns = 0;
753 	dst = &ln->r_l3addr.addr6;
754 	pdst = dst;
755 
756 	if (ln->ln_ntick > 0) {
757 		if (ln->ln_ntick > INT_MAX) {
758 			ln->ln_ntick -= INT_MAX;
759 			nd6_llinfo_settimer_locked(ln, INT_MAX);
760 		} else {
761 			ln->ln_ntick = 0;
762 			nd6_llinfo_settimer_locked(ln, ln->ln_ntick);
763 		}
764 		goto done;
765 	}
766 
767 	if (ln->la_flags & LLE_STATIC) {
768 		goto done;
769 	}
770 
771 	if (ln->la_flags & LLE_DELETED) {
772 		nd6_free(ln, 0);
773 		ln = NULL;
774 		goto done;
775 	}
776 
777 	switch (ln->ln_state) {
778 	case ND6_LLINFO_INCOMPLETE:
779 		if (ln->la_asked < V_nd6_mmaxtries) {
780 			ln->la_asked++;
781 			send_ns = 1;
782 			/* Send NS to multicast address */
783 			pdst = NULL;
784 		} else {
785 			struct mbuf *m = ln->la_hold;
786 			if (m) {
787 				struct mbuf *m0;
788 
789 				/*
790 				 * assuming every packet in la_hold has the
791 				 * same IP header.  Send error after unlock.
792 				 */
793 				m0 = m->m_nextpkt;
794 				m->m_nextpkt = NULL;
795 				ln->la_hold = m0;
796 				clear_llinfo_pqueue(ln);
797 			}
798 			EVENTHANDLER_INVOKE(lle_event, ln, LLENTRY_TIMEDOUT);
799 			nd6_free(ln, 0);
800 			ln = NULL;
801 			if (m != NULL)
802 				icmp6_error2(m, ICMP6_DST_UNREACH,
803 				    ICMP6_DST_UNREACH_ADDR, 0, ifp);
804 		}
805 		break;
806 	case ND6_LLINFO_REACHABLE:
807 		if (!ND6_LLINFO_PERMANENT(ln))
808 			nd6_llinfo_setstate(ln, ND6_LLINFO_STALE);
809 		break;
810 
811 	case ND6_LLINFO_STALE:
812 		if (nd6_is_stale(ln, &delay, &do_switch) != 0) {
813 
814 			/*
815 			 * No packet has used this entry and GC timeout
816 			 * has not been passed. Reshedule timer and
817 			 * return.
818 			 */
819 			nd6_llinfo_settimer_locked(ln, delay);
820 			break;
821 		}
822 
823 		if (do_switch == 0) {
824 
825 			/*
826 			 * GC timer has ended and entry hasn't been used.
827 			 * Run Garbage collector (RFC 4861, 5.3)
828 			 */
829 			if (!ND6_LLINFO_PERMANENT(ln)) {
830 				EVENTHANDLER_INVOKE(lle_event, ln,
831 				    LLENTRY_EXPIRED);
832 				nd6_free(ln, 1);
833 				ln = NULL;
834 			}
835 			break;
836 		}
837 
838 		/* Entry has been used AND delay timer has ended. */
839 
840 		/* FALLTHROUGH */
841 
842 	case ND6_LLINFO_DELAY:
843 		if (ndi && (ndi->flags & ND6_IFF_PERFORMNUD) != 0) {
844 			/* We need NUD */
845 			ln->la_asked = 1;
846 			nd6_llinfo_setstate(ln, ND6_LLINFO_PROBE);
847 			send_ns = 1;
848 		} else
849 			nd6_llinfo_setstate(ln, ND6_LLINFO_STALE); /* XXX */
850 		break;
851 	case ND6_LLINFO_PROBE:
852 		if (ln->la_asked < V_nd6_umaxtries) {
853 			ln->la_asked++;
854 			send_ns = 1;
855 		} else {
856 			EVENTHANDLER_INVOKE(lle_event, ln, LLENTRY_EXPIRED);
857 			nd6_free(ln, 0);
858 			ln = NULL;
859 		}
860 		break;
861 	default:
862 		panic("%s: paths in a dark night can be confusing: %d",
863 		    __func__, ln->ln_state);
864 	}
865 done:
866 	if (send_ns != 0) {
867 		nd6_llinfo_settimer_locked(ln, (long)ndi->retrans * hz / 1000);
868 		psrc = nd6_llinfo_get_holdsrc(ln, &src);
869 		LLE_FREE_LOCKED(ln);
870 		ln = NULL;
871 		nd6_ns_output(ifp, psrc, pdst, dst, NULL);
872 	}
873 
874 	if (ln != NULL)
875 		LLE_FREE_LOCKED(ln);
876 	CURVNET_RESTORE();
877 }
878 
879 
880 /*
881  * ND6 timer routine to expire default route list and prefix list
882  */
883 void
884 nd6_timer(void *arg)
885 {
886 	CURVNET_SET((struct vnet *) arg);
887 	struct nd_defrouter *dr, *ndr;
888 	struct nd_prefix *pr, *npr;
889 	struct in6_ifaddr *ia6, *nia6;
890 
891 	callout_reset(&V_nd6_timer_ch, V_nd6_prune * hz,
892 	    nd6_timer, curvnet);
893 
894 	/* expire default router list */
895 	TAILQ_FOREACH_SAFE(dr, &V_nd_defrouter, dr_entry, ndr) {
896 		if (dr->expire && dr->expire < time_uptime)
897 			defrtrlist_del(dr);
898 	}
899 
900 	/*
901 	 * expire interface addresses.
902 	 * in the past the loop was inside prefix expiry processing.
903 	 * However, from a stricter speci-confrmance standpoint, we should
904 	 * rather separate address lifetimes and prefix lifetimes.
905 	 *
906 	 * XXXRW: in6_ifaddrhead locking.
907 	 */
908   addrloop:
909 	TAILQ_FOREACH_SAFE(ia6, &V_in6_ifaddrhead, ia_link, nia6) {
910 		/* check address lifetime */
911 		if (IFA6_IS_INVALID(ia6)) {
912 			int regen = 0;
913 
914 			/*
915 			 * If the expiring address is temporary, try
916 			 * regenerating a new one.  This would be useful when
917 			 * we suspended a laptop PC, then turned it on after a
918 			 * period that could invalidate all temporary
919 			 * addresses.  Although we may have to restart the
920 			 * loop (see below), it must be after purging the
921 			 * address.  Otherwise, we'd see an infinite loop of
922 			 * regeneration.
923 			 */
924 			if (V_ip6_use_tempaddr &&
925 			    (ia6->ia6_flags & IN6_IFF_TEMPORARY) != 0) {
926 				if (regen_tmpaddr(ia6) == 0)
927 					regen = 1;
928 			}
929 
930 			in6_purgeaddr(&ia6->ia_ifa);
931 
932 			if (regen)
933 				goto addrloop; /* XXX: see below */
934 		} else if (IFA6_IS_DEPRECATED(ia6)) {
935 			int oldflags = ia6->ia6_flags;
936 
937 			ia6->ia6_flags |= IN6_IFF_DEPRECATED;
938 
939 			/*
940 			 * If a temporary address has just become deprecated,
941 			 * regenerate a new one if possible.
942 			 */
943 			if (V_ip6_use_tempaddr &&
944 			    (ia6->ia6_flags & IN6_IFF_TEMPORARY) != 0 &&
945 			    (oldflags & IN6_IFF_DEPRECATED) == 0) {
946 
947 				if (regen_tmpaddr(ia6) == 0) {
948 					/*
949 					 * A new temporary address is
950 					 * generated.
951 					 * XXX: this means the address chain
952 					 * has changed while we are still in
953 					 * the loop.  Although the change
954 					 * would not cause disaster (because
955 					 * it's not a deletion, but an
956 					 * addition,) we'd rather restart the
957 					 * loop just for safety.  Or does this
958 					 * significantly reduce performance??
959 					 */
960 					goto addrloop;
961 				}
962 			}
963 		} else if ((ia6->ia6_flags & IN6_IFF_TENTATIVE) != 0) {
964 			/*
965 			 * Schedule DAD for a tentative address.  This happens
966 			 * if the interface was down or not running
967 			 * when the address was configured.
968 			 */
969 			int delay;
970 
971 			delay = arc4random() %
972 			    (MAX_RTR_SOLICITATION_DELAY * hz);
973 			nd6_dad_start((struct ifaddr *)ia6, delay);
974 		} else {
975 			/*
976 			 * Check status of the interface.  If it is down,
977 			 * mark the address as tentative for future DAD.
978 			 */
979 			if ((ia6->ia_ifp->if_flags & IFF_UP) == 0 ||
980 			    (ia6->ia_ifp->if_drv_flags & IFF_DRV_RUNNING)
981 				== 0 ||
982 			    (ND_IFINFO(ia6->ia_ifp)->flags &
983 				ND6_IFF_IFDISABLED) != 0) {
984 				ia6->ia6_flags &= ~IN6_IFF_DUPLICATED;
985 				ia6->ia6_flags |= IN6_IFF_TENTATIVE;
986 			}
987 			/*
988 			 * A new RA might have made a deprecated address
989 			 * preferred.
990 			 */
991 			ia6->ia6_flags &= ~IN6_IFF_DEPRECATED;
992 		}
993 	}
994 
995 	/* expire prefix list */
996 	LIST_FOREACH_SAFE(pr, &V_nd_prefix, ndpr_entry, npr) {
997 		/*
998 		 * check prefix lifetime.
999 		 * since pltime is just for autoconf, pltime processing for
1000 		 * prefix is not necessary.
1001 		 */
1002 		if (pr->ndpr_vltime != ND6_INFINITE_LIFETIME &&
1003 		    time_uptime - pr->ndpr_lastupdate > pr->ndpr_vltime) {
1004 
1005 			/*
1006 			 * address expiration and prefix expiration are
1007 			 * separate.  NEVER perform in6_purgeaddr here.
1008 			 */
1009 			prelist_remove(pr);
1010 		}
1011 	}
1012 	CURVNET_RESTORE();
1013 }
1014 
1015 /*
1016  * ia6 - deprecated/invalidated temporary address
1017  */
1018 static int
1019 regen_tmpaddr(struct in6_ifaddr *ia6)
1020 {
1021 	struct ifaddr *ifa;
1022 	struct ifnet *ifp;
1023 	struct in6_ifaddr *public_ifa6 = NULL;
1024 
1025 	ifp = ia6->ia_ifa.ifa_ifp;
1026 	IF_ADDR_RLOCK(ifp);
1027 	TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1028 		struct in6_ifaddr *it6;
1029 
1030 		if (ifa->ifa_addr->sa_family != AF_INET6)
1031 			continue;
1032 
1033 		it6 = (struct in6_ifaddr *)ifa;
1034 
1035 		/* ignore no autoconf addresses. */
1036 		if ((it6->ia6_flags & IN6_IFF_AUTOCONF) == 0)
1037 			continue;
1038 
1039 		/* ignore autoconf addresses with different prefixes. */
1040 		if (it6->ia6_ndpr == NULL || it6->ia6_ndpr != ia6->ia6_ndpr)
1041 			continue;
1042 
1043 		/*
1044 		 * Now we are looking at an autoconf address with the same
1045 		 * prefix as ours.  If the address is temporary and is still
1046 		 * preferred, do not create another one.  It would be rare, but
1047 		 * could happen, for example, when we resume a laptop PC after
1048 		 * a long period.
1049 		 */
1050 		if ((it6->ia6_flags & IN6_IFF_TEMPORARY) != 0 &&
1051 		    !IFA6_IS_DEPRECATED(it6)) {
1052 			public_ifa6 = NULL;
1053 			break;
1054 		}
1055 
1056 		/*
1057 		 * This is a public autoconf address that has the same prefix
1058 		 * as ours.  If it is preferred, keep it.  We can't break the
1059 		 * loop here, because there may be a still-preferred temporary
1060 		 * address with the prefix.
1061 		 */
1062 		if (!IFA6_IS_DEPRECATED(it6))
1063 			public_ifa6 = it6;
1064 	}
1065 	if (public_ifa6 != NULL)
1066 		ifa_ref(&public_ifa6->ia_ifa);
1067 	IF_ADDR_RUNLOCK(ifp);
1068 
1069 	if (public_ifa6 != NULL) {
1070 		int e;
1071 
1072 		if ((e = in6_tmpifadd(public_ifa6, 0, 0)) != 0) {
1073 			ifa_free(&public_ifa6->ia_ifa);
1074 			log(LOG_NOTICE, "regen_tmpaddr: failed to create a new"
1075 			    " tmp addr,errno=%d\n", e);
1076 			return (-1);
1077 		}
1078 		ifa_free(&public_ifa6->ia_ifa);
1079 		return (0);
1080 	}
1081 
1082 	return (-1);
1083 }
1084 
1085 /*
1086  * Nuke neighbor cache/prefix/default router management table, right before
1087  * ifp goes away.
1088  */
1089 void
1090 nd6_purge(struct ifnet *ifp)
1091 {
1092 	struct nd_defrouter *dr, *ndr;
1093 	struct nd_prefix *pr, *npr;
1094 
1095 	/*
1096 	 * Nuke default router list entries toward ifp.
1097 	 * We defer removal of default router list entries that is installed
1098 	 * in the routing table, in order to keep additional side effects as
1099 	 * small as possible.
1100 	 */
1101 	TAILQ_FOREACH_SAFE(dr, &V_nd_defrouter, dr_entry, ndr) {
1102 		if (dr->installed)
1103 			continue;
1104 
1105 		if (dr->ifp == ifp)
1106 			defrtrlist_del(dr);
1107 	}
1108 
1109 	TAILQ_FOREACH_SAFE(dr, &V_nd_defrouter, dr_entry, ndr) {
1110 		if (!dr->installed)
1111 			continue;
1112 
1113 		if (dr->ifp == ifp)
1114 			defrtrlist_del(dr);
1115 	}
1116 
1117 	/* Nuke prefix list entries toward ifp */
1118 	LIST_FOREACH_SAFE(pr, &V_nd_prefix, ndpr_entry, npr) {
1119 		if (pr->ndpr_ifp == ifp) {
1120 			/*
1121 			 * Because if_detach() does *not* release prefixes
1122 			 * while purging addresses the reference count will
1123 			 * still be above zero. We therefore reset it to
1124 			 * make sure that the prefix really gets purged.
1125 			 */
1126 			pr->ndpr_refcnt = 0;
1127 
1128 			/*
1129 			 * Previously, pr->ndpr_addr is removed as well,
1130 			 * but I strongly believe we don't have to do it.
1131 			 * nd6_purge() is only called from in6_ifdetach(),
1132 			 * which removes all the associated interface addresses
1133 			 * by itself.
1134 			 * (jinmei@kame.net 20010129)
1135 			 */
1136 			prelist_remove(pr);
1137 		}
1138 	}
1139 
1140 	/* cancel default outgoing interface setting */
1141 	if (V_nd6_defifindex == ifp->if_index)
1142 		nd6_setdefaultiface(0);
1143 
1144 	if (ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) {
1145 		/* Refresh default router list. */
1146 		defrouter_select();
1147 	}
1148 
1149 	/* XXXXX
1150 	 * We do not nuke the neighbor cache entries here any more
1151 	 * because the neighbor cache is kept in if_afdata[AF_INET6].
1152 	 * nd6_purge() is invoked by in6_ifdetach() which is called
1153 	 * from if_detach() where everything gets purged. So let
1154 	 * in6_domifdetach() do the actual L2 table purging work.
1155 	 */
1156 }
1157 
1158 /*
1159  * the caller acquires and releases the lock on the lltbls
1160  * Returns the llentry locked
1161  */
1162 struct llentry *
1163 nd6_lookup(const struct in6_addr *addr6, int flags, struct ifnet *ifp)
1164 {
1165 	struct sockaddr_in6 sin6;
1166 	struct llentry *ln;
1167 
1168 	bzero(&sin6, sizeof(sin6));
1169 	sin6.sin6_len = sizeof(struct sockaddr_in6);
1170 	sin6.sin6_family = AF_INET6;
1171 	sin6.sin6_addr = *addr6;
1172 
1173 	IF_AFDATA_LOCK_ASSERT(ifp);
1174 
1175 	ln = lla_lookup(LLTABLE6(ifp), flags, (struct sockaddr *)&sin6);
1176 
1177 	return (ln);
1178 }
1179 
1180 struct llentry *
1181 nd6_alloc(const struct in6_addr *addr6, int flags, struct ifnet *ifp)
1182 {
1183 	struct sockaddr_in6 sin6;
1184 	struct llentry *ln;
1185 
1186 	bzero(&sin6, sizeof(sin6));
1187 	sin6.sin6_len = sizeof(struct sockaddr_in6);
1188 	sin6.sin6_family = AF_INET6;
1189 	sin6.sin6_addr = *addr6;
1190 
1191 	ln = lltable_alloc_entry(LLTABLE6(ifp), 0, (struct sockaddr *)&sin6);
1192 	if (ln != NULL)
1193 		ln->ln_state = ND6_LLINFO_NOSTATE;
1194 
1195 	return (ln);
1196 }
1197 
1198 /*
1199  * Test whether a given IPv6 address is a neighbor or not, ignoring
1200  * the actual neighbor cache.  The neighbor cache is ignored in order
1201  * to not reenter the routing code from within itself.
1202  */
1203 static int
1204 nd6_is_new_addr_neighbor(const struct sockaddr_in6 *addr, struct ifnet *ifp)
1205 {
1206 	struct nd_prefix *pr;
1207 	struct ifaddr *dstaddr;
1208 	struct rt_addrinfo info;
1209 	struct sockaddr_in6 rt_key;
1210 	struct sockaddr *dst6;
1211 	int fibnum;
1212 
1213 	/*
1214 	 * A link-local address is always a neighbor.
1215 	 * XXX: a link does not necessarily specify a single interface.
1216 	 */
1217 	if (IN6_IS_ADDR_LINKLOCAL(&addr->sin6_addr)) {
1218 		struct sockaddr_in6 sin6_copy;
1219 		u_int32_t zone;
1220 
1221 		/*
1222 		 * We need sin6_copy since sa6_recoverscope() may modify the
1223 		 * content (XXX).
1224 		 */
1225 		sin6_copy = *addr;
1226 		if (sa6_recoverscope(&sin6_copy))
1227 			return (0); /* XXX: should be impossible */
1228 		if (in6_setscope(&sin6_copy.sin6_addr, ifp, &zone))
1229 			return (0);
1230 		if (sin6_copy.sin6_scope_id == zone)
1231 			return (1);
1232 		else
1233 			return (0);
1234 	}
1235 
1236 	bzero(&rt_key, sizeof(rt_key));
1237 	bzero(&info, sizeof(info));
1238 	info.rti_info[RTAX_DST] = (struct sockaddr *)&rt_key;
1239 
1240 	/* Always use the default FIB here. XXME - why? */
1241 	fibnum = RT_DEFAULT_FIB;
1242 
1243 	/*
1244 	 * If the address matches one of our addresses,
1245 	 * it should be a neighbor.
1246 	 * If the address matches one of our on-link prefixes, it should be a
1247 	 * neighbor.
1248 	 */
1249 	LIST_FOREACH(pr, &V_nd_prefix, ndpr_entry) {
1250 		if (pr->ndpr_ifp != ifp)
1251 			continue;
1252 
1253 		if (!(pr->ndpr_stateflags & NDPRF_ONLINK)) {
1254 
1255 			/* Always use the default FIB here. */
1256 			dst6 = (struct sockaddr *)&pr->ndpr_prefix;
1257 
1258 			/* Restore length field before retrying lookup */
1259 			rt_key.sin6_len = sizeof(rt_key);
1260 			if (rib_lookup_info(fibnum, dst6, 0, 0, &info) != 0)
1261 				continue;
1262 			/*
1263 			 * This is the case where multiple interfaces
1264 			 * have the same prefix, but only one is installed
1265 			 * into the routing table and that prefix entry
1266 			 * is not the one being examined here. In the case
1267 			 * where RADIX_MPATH is enabled, multiple route
1268 			 * entries (of the same rt_key value) will be
1269 			 * installed because the interface addresses all
1270 			 * differ.
1271 			 */
1272 			if (!IN6_ARE_ADDR_EQUAL(&pr->ndpr_prefix.sin6_addr,
1273 			       &rt_key.sin6_addr))
1274 				continue;
1275 		}
1276 
1277 		if (IN6_ARE_MASKED_ADDR_EQUAL(&pr->ndpr_prefix.sin6_addr,
1278 		    &addr->sin6_addr, &pr->ndpr_mask))
1279 			return (1);
1280 	}
1281 
1282 	/*
1283 	 * If the address is assigned on the node of the other side of
1284 	 * a p2p interface, the address should be a neighbor.
1285 	 */
1286 	dstaddr = ifa_ifwithdstaddr((const struct sockaddr *)addr, RT_ALL_FIBS);
1287 	if (dstaddr != NULL) {
1288 		if (dstaddr->ifa_ifp == ifp) {
1289 			ifa_free(dstaddr);
1290 			return (1);
1291 		}
1292 		ifa_free(dstaddr);
1293 	}
1294 
1295 	/*
1296 	 * If the default router list is empty, all addresses are regarded
1297 	 * as on-link, and thus, as a neighbor.
1298 	 */
1299 	if (ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV &&
1300 	    TAILQ_EMPTY(&V_nd_defrouter) &&
1301 	    V_nd6_defifindex == ifp->if_index) {
1302 		return (1);
1303 	}
1304 
1305 	return (0);
1306 }
1307 
1308 
1309 /*
1310  * Detect if a given IPv6 address identifies a neighbor on a given link.
1311  * XXX: should take care of the destination of a p2p link?
1312  */
1313 int
1314 nd6_is_addr_neighbor(const struct sockaddr_in6 *addr, struct ifnet *ifp)
1315 {
1316 	struct llentry *lle;
1317 	int rc = 0;
1318 
1319 	IF_AFDATA_UNLOCK_ASSERT(ifp);
1320 	if (nd6_is_new_addr_neighbor(addr, ifp))
1321 		return (1);
1322 
1323 	/*
1324 	 * Even if the address matches none of our addresses, it might be
1325 	 * in the neighbor cache.
1326 	 */
1327 	IF_AFDATA_RLOCK(ifp);
1328 	if ((lle = nd6_lookup(&addr->sin6_addr, 0, ifp)) != NULL) {
1329 		LLE_RUNLOCK(lle);
1330 		rc = 1;
1331 	}
1332 	IF_AFDATA_RUNLOCK(ifp);
1333 	return (rc);
1334 }
1335 
1336 /*
1337  * Free an nd6 llinfo entry.
1338  * Since the function would cause significant changes in the kernel, DO NOT
1339  * make it global, unless you have a strong reason for the change, and are sure
1340  * that the change is safe.
1341  *
1342  * Set noinline to be dtrace-friendly
1343  */
1344 static __noinline void
1345 nd6_free(struct llentry *ln, int gc)
1346 {
1347 	struct nd_defrouter *dr;
1348 	struct ifnet *ifp;
1349 
1350 	LLE_WLOCK_ASSERT(ln);
1351 
1352 	/*
1353 	 * we used to have pfctlinput(PRC_HOSTDEAD) here.
1354 	 * even though it is not harmful, it was not really necessary.
1355 	 */
1356 
1357 	/* cancel timer */
1358 	nd6_llinfo_settimer_locked(ln, -1);
1359 
1360 	ifp = ln->lle_tbl->llt_ifp;
1361 
1362 	if (ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) {
1363 		dr = defrouter_lookup(&ln->r_l3addr.addr6, ifp);
1364 
1365 		if (dr != NULL && dr->expire &&
1366 		    ln->ln_state == ND6_LLINFO_STALE && gc) {
1367 			/*
1368 			 * If the reason for the deletion is just garbage
1369 			 * collection, and the neighbor is an active default
1370 			 * router, do not delete it.  Instead, reset the GC
1371 			 * timer using the router's lifetime.
1372 			 * Simply deleting the entry would affect default
1373 			 * router selection, which is not necessarily a good
1374 			 * thing, especially when we're using router preference
1375 			 * values.
1376 			 * XXX: the check for ln_state would be redundant,
1377 			 *      but we intentionally keep it just in case.
1378 			 */
1379 			if (dr->expire > time_uptime)
1380 				nd6_llinfo_settimer_locked(ln,
1381 				    (dr->expire - time_uptime) * hz);
1382 			else
1383 				nd6_llinfo_settimer_locked(ln,
1384 				    (long)V_nd6_gctimer * hz);
1385 
1386 			LLE_REMREF(ln);
1387 			LLE_WUNLOCK(ln);
1388 			return;
1389 		}
1390 
1391 		if (dr) {
1392 			/*
1393 			 * Unreachablity of a router might affect the default
1394 			 * router selection and on-link detection of advertised
1395 			 * prefixes.
1396 			 */
1397 
1398 			/*
1399 			 * Temporarily fake the state to choose a new default
1400 			 * router and to perform on-link determination of
1401 			 * prefixes correctly.
1402 			 * Below the state will be set correctly,
1403 			 * or the entry itself will be deleted.
1404 			 */
1405 			ln->ln_state = ND6_LLINFO_INCOMPLETE;
1406 		}
1407 
1408 		if (ln->ln_router || dr) {
1409 
1410 			/*
1411 			 * We need to unlock to avoid a LOR with rt6_flush() with the
1412 			 * rnh and for the calls to pfxlist_onlink_check() and
1413 			 * defrouter_select() in the block further down for calls
1414 			 * into nd6_lookup().  We still hold a ref.
1415 			 */
1416 			LLE_WUNLOCK(ln);
1417 
1418 			/*
1419 			 * rt6_flush must be called whether or not the neighbor
1420 			 * is in the Default Router List.
1421 			 * See a corresponding comment in nd6_na_input().
1422 			 */
1423 			rt6_flush(&ln->r_l3addr.addr6, ifp);
1424 		}
1425 
1426 		if (dr) {
1427 			/*
1428 			 * Since defrouter_select() does not affect the
1429 			 * on-link determination and MIP6 needs the check
1430 			 * before the default router selection, we perform
1431 			 * the check now.
1432 			 */
1433 			pfxlist_onlink_check();
1434 
1435 			/*
1436 			 * Refresh default router list.
1437 			 */
1438 			defrouter_select();
1439 		}
1440 
1441 		/*
1442 		 * If this entry was added by an on-link redirect, remove the
1443 		 * corresponding host route.
1444 		 */
1445 		if (ln->la_flags & LLE_REDIRECT)
1446 			nd6_free_redirect(ln);
1447 
1448 		if (ln->ln_router || dr)
1449 			LLE_WLOCK(ln);
1450 	}
1451 
1452 	/*
1453 	 * Save to unlock. We still hold an extra reference and will not
1454 	 * free(9) in llentry_free() if someone else holds one as well.
1455 	 */
1456 	LLE_WUNLOCK(ln);
1457 	IF_AFDATA_LOCK(ifp);
1458 	LLE_WLOCK(ln);
1459 	/* Guard against race with other llentry_free(). */
1460 	if (ln->la_flags & LLE_LINKED) {
1461 		/* Remove callout reference */
1462 		LLE_REMREF(ln);
1463 		lltable_unlink_entry(ln->lle_tbl, ln);
1464 	}
1465 	IF_AFDATA_UNLOCK(ifp);
1466 
1467 	llentry_free(ln);
1468 }
1469 
1470 static int
1471 nd6_isdynrte(const struct rtentry *rt, void *xap)
1472 {
1473 
1474 	if (rt->rt_flags == (RTF_UP | RTF_HOST | RTF_DYNAMIC))
1475 		return (1);
1476 
1477 	return (0);
1478 }
1479 /*
1480  * Remove the rtentry for the given llentry,
1481  * both of which were installed by a redirect.
1482  */
1483 static void
1484 nd6_free_redirect(const struct llentry *ln)
1485 {
1486 	int fibnum;
1487 	struct sockaddr_in6 sin6;
1488 	struct rt_addrinfo info;
1489 
1490 	lltable_fill_sa_entry(ln, (struct sockaddr *)&sin6);
1491 	memset(&info, 0, sizeof(info));
1492 	info.rti_info[RTAX_DST] = (struct sockaddr *)&sin6;
1493 	info.rti_filter = nd6_isdynrte;
1494 
1495 	for (fibnum = 0; fibnum < rt_numfibs; fibnum++)
1496 		rtrequest1_fib(RTM_DELETE, &info, NULL, fibnum);
1497 }
1498 
1499 /*
1500  * Rejuvenate this function for routing operations related
1501  * processing.
1502  */
1503 void
1504 nd6_rtrequest(int req, struct rtentry *rt, struct rt_addrinfo *info)
1505 {
1506 	struct sockaddr_in6 *gateway;
1507 	struct nd_defrouter *dr;
1508 	struct ifnet *ifp;
1509 
1510 	gateway = (struct sockaddr_in6 *)rt->rt_gateway;
1511 	ifp = rt->rt_ifp;
1512 
1513 	switch (req) {
1514 	case RTM_ADD:
1515 		break;
1516 
1517 	case RTM_DELETE:
1518 		if (!ifp)
1519 			return;
1520 		/*
1521 		 * Only indirect routes are interesting.
1522 		 */
1523 		if ((rt->rt_flags & RTF_GATEWAY) == 0)
1524 			return;
1525 		/*
1526 		 * check for default route
1527 		 */
1528 		if (IN6_ARE_ADDR_EQUAL(&in6addr_any,
1529 				       &SIN6(rt_key(rt))->sin6_addr)) {
1530 
1531 			dr = defrouter_lookup(&gateway->sin6_addr, ifp);
1532 			if (dr != NULL)
1533 				dr->installed = 0;
1534 		}
1535 		break;
1536 	}
1537 }
1538 
1539 
1540 int
1541 nd6_ioctl(u_long cmd, caddr_t data, struct ifnet *ifp)
1542 {
1543 	struct in6_ndireq *ndi = (struct in6_ndireq *)data;
1544 	struct in6_nbrinfo *nbi = (struct in6_nbrinfo *)data;
1545 	struct in6_ndifreq *ndif = (struct in6_ndifreq *)data;
1546 	int error = 0;
1547 
1548 	if (ifp->if_afdata[AF_INET6] == NULL)
1549 		return (EPFNOSUPPORT);
1550 	switch (cmd) {
1551 	case OSIOCGIFINFO_IN6:
1552 #define ND	ndi->ndi
1553 		/* XXX: old ndp(8) assumes a positive value for linkmtu. */
1554 		bzero(&ND, sizeof(ND));
1555 		ND.linkmtu = IN6_LINKMTU(ifp);
1556 		ND.maxmtu = ND_IFINFO(ifp)->maxmtu;
1557 		ND.basereachable = ND_IFINFO(ifp)->basereachable;
1558 		ND.reachable = ND_IFINFO(ifp)->reachable;
1559 		ND.retrans = ND_IFINFO(ifp)->retrans;
1560 		ND.flags = ND_IFINFO(ifp)->flags;
1561 		ND.recalctm = ND_IFINFO(ifp)->recalctm;
1562 		ND.chlim = ND_IFINFO(ifp)->chlim;
1563 		break;
1564 	case SIOCGIFINFO_IN6:
1565 		ND = *ND_IFINFO(ifp);
1566 		break;
1567 	case SIOCSIFINFO_IN6:
1568 		/*
1569 		 * used to change host variables from userland.
1570 		 * intented for a use on router to reflect RA configurations.
1571 		 */
1572 		/* 0 means 'unspecified' */
1573 		if (ND.linkmtu != 0) {
1574 			if (ND.linkmtu < IPV6_MMTU ||
1575 			    ND.linkmtu > IN6_LINKMTU(ifp)) {
1576 				error = EINVAL;
1577 				break;
1578 			}
1579 			ND_IFINFO(ifp)->linkmtu = ND.linkmtu;
1580 		}
1581 
1582 		if (ND.basereachable != 0) {
1583 			int obasereachable = ND_IFINFO(ifp)->basereachable;
1584 
1585 			ND_IFINFO(ifp)->basereachable = ND.basereachable;
1586 			if (ND.basereachable != obasereachable)
1587 				ND_IFINFO(ifp)->reachable =
1588 				    ND_COMPUTE_RTIME(ND.basereachable);
1589 		}
1590 		if (ND.retrans != 0)
1591 			ND_IFINFO(ifp)->retrans = ND.retrans;
1592 		if (ND.chlim != 0)
1593 			ND_IFINFO(ifp)->chlim = ND.chlim;
1594 		/* FALLTHROUGH */
1595 	case SIOCSIFINFO_FLAGS:
1596 	{
1597 		struct ifaddr *ifa;
1598 		struct in6_ifaddr *ia;
1599 
1600 		if ((ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED) &&
1601 		    !(ND.flags & ND6_IFF_IFDISABLED)) {
1602 			/* ifdisabled 1->0 transision */
1603 
1604 			/*
1605 			 * If the interface is marked as ND6_IFF_IFDISABLED and
1606 			 * has an link-local address with IN6_IFF_DUPLICATED,
1607 			 * do not clear ND6_IFF_IFDISABLED.
1608 			 * See RFC 4862, Section 5.4.5.
1609 			 */
1610 			IF_ADDR_RLOCK(ifp);
1611 			TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1612 				if (ifa->ifa_addr->sa_family != AF_INET6)
1613 					continue;
1614 				ia = (struct in6_ifaddr *)ifa;
1615 				if ((ia->ia6_flags & IN6_IFF_DUPLICATED) &&
1616 				    IN6_IS_ADDR_LINKLOCAL(IA6_IN6(ia)))
1617 					break;
1618 			}
1619 			IF_ADDR_RUNLOCK(ifp);
1620 
1621 			if (ifa != NULL) {
1622 				/* LLA is duplicated. */
1623 				ND.flags |= ND6_IFF_IFDISABLED;
1624 				log(LOG_ERR, "Cannot enable an interface"
1625 				    " with a link-local address marked"
1626 				    " duplicate.\n");
1627 			} else {
1628 				ND_IFINFO(ifp)->flags &= ~ND6_IFF_IFDISABLED;
1629 				if (ifp->if_flags & IFF_UP)
1630 					in6_if_up(ifp);
1631 			}
1632 		} else if (!(ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED) &&
1633 			    (ND.flags & ND6_IFF_IFDISABLED)) {
1634 			/* ifdisabled 0->1 transision */
1635 			/* Mark all IPv6 address as tentative. */
1636 
1637 			ND_IFINFO(ifp)->flags |= ND6_IFF_IFDISABLED;
1638 			if (V_ip6_dad_count > 0 &&
1639 			    (ND_IFINFO(ifp)->flags & ND6_IFF_NO_DAD) == 0) {
1640 				IF_ADDR_RLOCK(ifp);
1641 				TAILQ_FOREACH(ifa, &ifp->if_addrhead,
1642 				    ifa_link) {
1643 					if (ifa->ifa_addr->sa_family !=
1644 					    AF_INET6)
1645 						continue;
1646 					ia = (struct in6_ifaddr *)ifa;
1647 					ia->ia6_flags |= IN6_IFF_TENTATIVE;
1648 				}
1649 				IF_ADDR_RUNLOCK(ifp);
1650 			}
1651 		}
1652 
1653 		if (ND.flags & ND6_IFF_AUTO_LINKLOCAL) {
1654 			if (!(ND_IFINFO(ifp)->flags & ND6_IFF_AUTO_LINKLOCAL)) {
1655 				/* auto_linklocal 0->1 transision */
1656 
1657 				/* If no link-local address on ifp, configure */
1658 				ND_IFINFO(ifp)->flags |= ND6_IFF_AUTO_LINKLOCAL;
1659 				in6_ifattach(ifp, NULL);
1660 			} else if (!(ND.flags & ND6_IFF_IFDISABLED) &&
1661 			    ifp->if_flags & IFF_UP) {
1662 				/*
1663 				 * When the IF already has
1664 				 * ND6_IFF_AUTO_LINKLOCAL, no link-local
1665 				 * address is assigned, and IFF_UP, try to
1666 				 * assign one.
1667 				 */
1668 				IF_ADDR_RLOCK(ifp);
1669 				TAILQ_FOREACH(ifa, &ifp->if_addrhead,
1670 				    ifa_link) {
1671 					if (ifa->ifa_addr->sa_family !=
1672 					    AF_INET6)
1673 						continue;
1674 					ia = (struct in6_ifaddr *)ifa;
1675 					if (IN6_IS_ADDR_LINKLOCAL(IA6_IN6(ia)))
1676 						break;
1677 				}
1678 				IF_ADDR_RUNLOCK(ifp);
1679 				if (ifa != NULL)
1680 					/* No LLA is configured. */
1681 					in6_ifattach(ifp, NULL);
1682 			}
1683 		}
1684 	}
1685 		ND_IFINFO(ifp)->flags = ND.flags;
1686 		break;
1687 #undef ND
1688 	case SIOCSNDFLUSH_IN6:	/* XXX: the ioctl name is confusing... */
1689 		/* sync kernel routing table with the default router list */
1690 		defrouter_reset();
1691 		defrouter_select();
1692 		break;
1693 	case SIOCSPFXFLUSH_IN6:
1694 	{
1695 		/* flush all the prefix advertised by routers */
1696 		struct nd_prefix *pr, *next;
1697 
1698 		LIST_FOREACH_SAFE(pr, &V_nd_prefix, ndpr_entry, next) {
1699 			struct in6_ifaddr *ia, *ia_next;
1700 
1701 			if (IN6_IS_ADDR_LINKLOCAL(&pr->ndpr_prefix.sin6_addr))
1702 				continue; /* XXX */
1703 
1704 			/* do we really have to remove addresses as well? */
1705 			/* XXXRW: in6_ifaddrhead locking. */
1706 			TAILQ_FOREACH_SAFE(ia, &V_in6_ifaddrhead, ia_link,
1707 			    ia_next) {
1708 				if ((ia->ia6_flags & IN6_IFF_AUTOCONF) == 0)
1709 					continue;
1710 
1711 				if (ia->ia6_ndpr == pr)
1712 					in6_purgeaddr(&ia->ia_ifa);
1713 			}
1714 			prelist_remove(pr);
1715 		}
1716 		break;
1717 	}
1718 	case SIOCSRTRFLUSH_IN6:
1719 	{
1720 		/* flush all the default routers */
1721 		struct nd_defrouter *dr, *next;
1722 
1723 		defrouter_reset();
1724 		TAILQ_FOREACH_SAFE(dr, &V_nd_defrouter, dr_entry, next) {
1725 			defrtrlist_del(dr);
1726 		}
1727 		defrouter_select();
1728 		break;
1729 	}
1730 	case SIOCGNBRINFO_IN6:
1731 	{
1732 		struct llentry *ln;
1733 		struct in6_addr nb_addr = nbi->addr; /* make local for safety */
1734 
1735 		if ((error = in6_setscope(&nb_addr, ifp, NULL)) != 0)
1736 			return (error);
1737 
1738 		IF_AFDATA_RLOCK(ifp);
1739 		ln = nd6_lookup(&nb_addr, 0, ifp);
1740 		IF_AFDATA_RUNLOCK(ifp);
1741 
1742 		if (ln == NULL) {
1743 			error = EINVAL;
1744 			break;
1745 		}
1746 		nbi->state = ln->ln_state;
1747 		nbi->asked = ln->la_asked;
1748 		nbi->isrouter = ln->ln_router;
1749 		if (ln->la_expire == 0)
1750 			nbi->expire = 0;
1751 		else
1752 			nbi->expire = ln->la_expire + ln->lle_remtime / hz +
1753 			    (time_second - time_uptime);
1754 		LLE_RUNLOCK(ln);
1755 		break;
1756 	}
1757 	case SIOCGDEFIFACE_IN6:	/* XXX: should be implemented as a sysctl? */
1758 		ndif->ifindex = V_nd6_defifindex;
1759 		break;
1760 	case SIOCSDEFIFACE_IN6:	/* XXX: should be implemented as a sysctl? */
1761 		return (nd6_setdefaultiface(ndif->ifindex));
1762 	}
1763 	return (error);
1764 }
1765 
1766 /*
1767  * Calculates new isRouter value based on provided parameters and
1768  * returns it.
1769  */
1770 static int
1771 nd6_is_router(int type, int code, int is_new, int old_addr, int new_addr,
1772     int ln_router)
1773 {
1774 
1775 	/*
1776 	 * ICMP6 type dependent behavior.
1777 	 *
1778 	 * NS: clear IsRouter if new entry
1779 	 * RS: clear IsRouter
1780 	 * RA: set IsRouter if there's lladdr
1781 	 * redir: clear IsRouter if new entry
1782 	 *
1783 	 * RA case, (1):
1784 	 * The spec says that we must set IsRouter in the following cases:
1785 	 * - If lladdr exist, set IsRouter.  This means (1-5).
1786 	 * - If it is old entry (!newentry), set IsRouter.  This means (7).
1787 	 * So, based on the spec, in (1-5) and (7) cases we must set IsRouter.
1788 	 * A quetion arises for (1) case.  (1) case has no lladdr in the
1789 	 * neighbor cache, this is similar to (6).
1790 	 * This case is rare but we figured that we MUST NOT set IsRouter.
1791 	 *
1792 	 *   is_new  old_addr new_addr 	    NS  RS  RA	redir
1793 	 *							D R
1794 	 *	0	n	n	(1)	c   ?     s
1795 	 *	0	y	n	(2)	c   s     s
1796 	 *	0	n	y	(3)	c   s     s
1797 	 *	0	y	y	(4)	c   s     s
1798 	 *	0	y	y	(5)	c   s     s
1799 	 *	1	--	n	(6) c	c	c s
1800 	 *	1	--	y	(7) c	c   s	c s
1801 	 *
1802 	 *					(c=clear s=set)
1803 	 */
1804 	switch (type & 0xff) {
1805 	case ND_NEIGHBOR_SOLICIT:
1806 		/*
1807 		 * New entry must have is_router flag cleared.
1808 		 */
1809 		if (is_new)					/* (6-7) */
1810 			ln_router = 0;
1811 		break;
1812 	case ND_REDIRECT:
1813 		/*
1814 		 * If the icmp is a redirect to a better router, always set the
1815 		 * is_router flag.  Otherwise, if the entry is newly created,
1816 		 * clear the flag.  [RFC 2461, sec 8.3]
1817 		 */
1818 		if (code == ND_REDIRECT_ROUTER)
1819 			ln_router = 1;
1820 		else {
1821 			if (is_new)				/* (6-7) */
1822 				ln_router = 0;
1823 		}
1824 		break;
1825 	case ND_ROUTER_SOLICIT:
1826 		/*
1827 		 * is_router flag must always be cleared.
1828 		 */
1829 		ln_router = 0;
1830 		break;
1831 	case ND_ROUTER_ADVERT:
1832 		/*
1833 		 * Mark an entry with lladdr as a router.
1834 		 */
1835 		if ((!is_new && (old_addr || new_addr)) ||	/* (2-5) */
1836 		    (is_new && new_addr)) {			/* (7) */
1837 			ln_router = 1;
1838 		}
1839 		break;
1840 	}
1841 
1842 	return (ln_router);
1843 }
1844 
1845 /*
1846  * Create neighbor cache entry and cache link-layer address,
1847  * on reception of inbound ND6 packets.  (RS/RA/NS/redirect)
1848  *
1849  * type - ICMP6 type
1850  * code - type dependent information
1851  *
1852  */
1853 void
1854 nd6_cache_lladdr(struct ifnet *ifp, struct in6_addr *from, char *lladdr,
1855     int lladdrlen, int type, int code)
1856 {
1857 	struct llentry *ln = NULL, *ln_tmp;
1858 	int is_newentry;
1859 	int do_update;
1860 	int olladdr;
1861 	int llchange;
1862 	int flags;
1863 	uint16_t router = 0;
1864 	struct sockaddr_in6 sin6;
1865 	struct mbuf *chain = NULL;
1866 	u_char linkhdr[LLE_MAX_LINKHDR];
1867 	size_t linkhdrsize;
1868 	int lladdr_off;
1869 
1870 	IF_AFDATA_UNLOCK_ASSERT(ifp);
1871 
1872 	KASSERT(ifp != NULL, ("%s: ifp == NULL", __func__));
1873 	KASSERT(from != NULL, ("%s: from == NULL", __func__));
1874 
1875 	/* nothing must be updated for unspecified address */
1876 	if (IN6_IS_ADDR_UNSPECIFIED(from))
1877 		return;
1878 
1879 	/*
1880 	 * Validation about ifp->if_addrlen and lladdrlen must be done in
1881 	 * the caller.
1882 	 *
1883 	 * XXX If the link does not have link-layer adderss, what should
1884 	 * we do? (ifp->if_addrlen == 0)
1885 	 * Spec says nothing in sections for RA, RS and NA.  There's small
1886 	 * description on it in NS section (RFC 2461 7.2.3).
1887 	 */
1888 	flags = lladdr ? LLE_EXCLUSIVE : 0;
1889 	IF_AFDATA_RLOCK(ifp);
1890 	ln = nd6_lookup(from, flags, ifp);
1891 	IF_AFDATA_RUNLOCK(ifp);
1892 	is_newentry = 0;
1893 	if (ln == NULL) {
1894 		flags |= LLE_EXCLUSIVE;
1895 		ln = nd6_alloc(from, 0, ifp);
1896 		if (ln == NULL)
1897 			return;
1898 
1899 		/*
1900 		 * Since we already know all the data for the new entry,
1901 		 * fill it before insertion.
1902 		 */
1903 		if (lladdr != NULL) {
1904 			linkhdrsize = sizeof(linkhdr);
1905 			if (lltable_calc_llheader(ifp, AF_INET6, lladdr,
1906 			    linkhdr, &linkhdrsize, &lladdr_off) != 0)
1907 				return;
1908 			lltable_set_entry_addr(ifp, ln, linkhdr, linkhdrsize,
1909 			    lladdr_off);
1910 		}
1911 
1912 		IF_AFDATA_WLOCK(ifp);
1913 		LLE_WLOCK(ln);
1914 		/* Prefer any existing lle over newly-created one */
1915 		ln_tmp = nd6_lookup(from, LLE_EXCLUSIVE, ifp);
1916 		if (ln_tmp == NULL)
1917 			lltable_link_entry(LLTABLE6(ifp), ln);
1918 		IF_AFDATA_WUNLOCK(ifp);
1919 		if (ln_tmp == NULL) {
1920 			/* No existing lle, mark as new entry (6,7) */
1921 			is_newentry = 1;
1922 			nd6_llinfo_setstate(ln, ND6_LLINFO_STALE);
1923 			if (lladdr != NULL)	/* (7) */
1924 				EVENTHANDLER_INVOKE(lle_event, ln,
1925 				    LLENTRY_RESOLVED);
1926 		} else {
1927 			lltable_free_entry(LLTABLE6(ifp), ln);
1928 			ln = ln_tmp;
1929 			ln_tmp = NULL;
1930 		}
1931 	}
1932 	/* do nothing if static ndp is set */
1933 	if ((ln->la_flags & LLE_STATIC)) {
1934 		if (flags & LLE_EXCLUSIVE)
1935 			LLE_WUNLOCK(ln);
1936 		else
1937 			LLE_RUNLOCK(ln);
1938 		return;
1939 	}
1940 
1941 	olladdr = (ln->la_flags & LLE_VALID) ? 1 : 0;
1942 	if (olladdr && lladdr) {
1943 		llchange = bcmp(lladdr, ln->ll_addr,
1944 		    ifp->if_addrlen);
1945 	} else if (!olladdr && lladdr)
1946 		llchange = 1;
1947 	else
1948 		llchange = 0;
1949 
1950 	/*
1951 	 * newentry olladdr  lladdr  llchange	(*=record)
1952 	 *	0	n	n	--	(1)
1953 	 *	0	y	n	--	(2)
1954 	 *	0	n	y	y	(3) * STALE
1955 	 *	0	y	y	n	(4) *
1956 	 *	0	y	y	y	(5) * STALE
1957 	 *	1	--	n	--	(6)   NOSTATE(= PASSIVE)
1958 	 *	1	--	y	--	(7) * STALE
1959 	 */
1960 
1961 	do_update = 0;
1962 	if (is_newentry == 0 && llchange != 0) {
1963 		do_update = 1;	/* (3,5) */
1964 
1965 		/*
1966 		 * Record source link-layer address
1967 		 * XXX is it dependent to ifp->if_type?
1968 		 */
1969 		linkhdrsize = sizeof(linkhdr);
1970 		if (lltable_calc_llheader(ifp, AF_INET6, lladdr,
1971 		    linkhdr, &linkhdrsize, &lladdr_off) != 0)
1972 			return;
1973 
1974 		if (lltable_try_set_entry_addr(ifp, ln, linkhdr, linkhdrsize,
1975 		    lladdr_off) == 0) {
1976 			/* Entry was deleted */
1977 			return;
1978 		}
1979 
1980 		nd6_llinfo_setstate(ln, ND6_LLINFO_STALE);
1981 
1982 		EVENTHANDLER_INVOKE(lle_event, ln, LLENTRY_RESOLVED);
1983 
1984 		if (ln->la_hold != NULL)
1985 			nd6_grab_holdchain(ln, &chain, &sin6);
1986 	}
1987 
1988 	/* Calculates new router status */
1989 	router = nd6_is_router(type, code, is_newentry, olladdr,
1990 	    lladdr != NULL ? 1 : 0, ln->ln_router);
1991 
1992 	ln->ln_router = router;
1993 	/* Mark non-router redirects with special flag */
1994 	if ((type & 0xFF) == ND_REDIRECT && code != ND_REDIRECT_ROUTER)
1995 		ln->la_flags |= LLE_REDIRECT;
1996 
1997 	if (flags & LLE_EXCLUSIVE)
1998 		LLE_WUNLOCK(ln);
1999 	else
2000 		LLE_RUNLOCK(ln);
2001 
2002 	if (chain != NULL)
2003 		nd6_flush_holdchain(ifp, ifp, chain, &sin6);
2004 
2005 	/*
2006 	 * When the link-layer address of a router changes, select the
2007 	 * best router again.  In particular, when the neighbor entry is newly
2008 	 * created, it might affect the selection policy.
2009 	 * Question: can we restrict the first condition to the "is_newentry"
2010 	 * case?
2011 	 * XXX: when we hear an RA from a new router with the link-layer
2012 	 * address option, defrouter_select() is called twice, since
2013 	 * defrtrlist_update called the function as well.  However, I believe
2014 	 * we can compromise the overhead, since it only happens the first
2015 	 * time.
2016 	 * XXX: although defrouter_select() should not have a bad effect
2017 	 * for those are not autoconfigured hosts, we explicitly avoid such
2018 	 * cases for safety.
2019 	 */
2020 	if ((do_update || is_newentry) && router &&
2021 	    ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) {
2022 		/*
2023 		 * guaranteed recursion
2024 		 */
2025 		defrouter_select();
2026 	}
2027 }
2028 
2029 static void
2030 nd6_slowtimo(void *arg)
2031 {
2032 	CURVNET_SET((struct vnet *) arg);
2033 	struct nd_ifinfo *nd6if;
2034 	struct ifnet *ifp;
2035 
2036 	callout_reset(&V_nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
2037 	    nd6_slowtimo, curvnet);
2038 	IFNET_RLOCK_NOSLEEP();
2039 	TAILQ_FOREACH(ifp, &V_ifnet, if_link) {
2040 		if (ifp->if_afdata[AF_INET6] == NULL)
2041 			continue;
2042 		nd6if = ND_IFINFO(ifp);
2043 		if (nd6if->basereachable && /* already initialized */
2044 		    (nd6if->recalctm -= ND6_SLOWTIMER_INTERVAL) <= 0) {
2045 			/*
2046 			 * Since reachable time rarely changes by router
2047 			 * advertisements, we SHOULD insure that a new random
2048 			 * value gets recomputed at least once every few hours.
2049 			 * (RFC 2461, 6.3.4)
2050 			 */
2051 			nd6if->recalctm = V_nd6_recalc_reachtm_interval;
2052 			nd6if->reachable = ND_COMPUTE_RTIME(nd6if->basereachable);
2053 		}
2054 	}
2055 	IFNET_RUNLOCK_NOSLEEP();
2056 	CURVNET_RESTORE();
2057 }
2058 
2059 void
2060 nd6_grab_holdchain(struct llentry *ln, struct mbuf **chain,
2061     struct sockaddr_in6 *sin6)
2062 {
2063 
2064 	LLE_WLOCK_ASSERT(ln);
2065 
2066 	*chain = ln->la_hold;
2067 	ln->la_hold = NULL;
2068 	lltable_fill_sa_entry(ln, (struct sockaddr *)sin6);
2069 
2070 	if (ln->ln_state == ND6_LLINFO_STALE) {
2071 
2072 		/*
2073 		 * The first time we send a packet to a
2074 		 * neighbor whose entry is STALE, we have
2075 		 * to change the state to DELAY and a sets
2076 		 * a timer to expire in DELAY_FIRST_PROBE_TIME
2077 		 * seconds to ensure do neighbor unreachability
2078 		 * detection on expiration.
2079 		 * (RFC 2461 7.3.3)
2080 		 */
2081 		nd6_llinfo_setstate(ln, ND6_LLINFO_DELAY);
2082 	}
2083 }
2084 
2085 int
2086 nd6_output_ifp(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *m,
2087     struct sockaddr_in6 *dst, struct route *ro)
2088 {
2089 	int error;
2090 	int ip6len;
2091 	struct ip6_hdr *ip6;
2092 	struct m_tag *mtag;
2093 
2094 #ifdef MAC
2095 	mac_netinet6_nd6_send(ifp, m);
2096 #endif
2097 
2098 	/*
2099 	 * If called from nd6_ns_output() (NS), nd6_na_output() (NA),
2100 	 * icmp6_redirect_output() (REDIRECT) or from rip6_output() (RS, RA
2101 	 * as handled by rtsol and rtadvd), mbufs will be tagged for SeND
2102 	 * to be diverted to user space.  When re-injected into the kernel,
2103 	 * send_output() will directly dispatch them to the outgoing interface.
2104 	 */
2105 	if (send_sendso_input_hook != NULL) {
2106 		mtag = m_tag_find(m, PACKET_TAG_ND_OUTGOING, NULL);
2107 		if (mtag != NULL) {
2108 			ip6 = mtod(m, struct ip6_hdr *);
2109 			ip6len = sizeof(struct ip6_hdr) + ntohs(ip6->ip6_plen);
2110 			/* Use the SEND socket */
2111 			error = send_sendso_input_hook(m, ifp, SND_OUT,
2112 			    ip6len);
2113 			/* -1 == no app on SEND socket */
2114 			if (error == 0 || error != -1)
2115 			    return (error);
2116 		}
2117 	}
2118 
2119 	m_clrprotoflags(m);	/* Avoid confusing lower layers. */
2120 	IP_PROBE(send, NULL, NULL, mtod(m, struct ip6_hdr *), ifp, NULL,
2121 	    mtod(m, struct ip6_hdr *));
2122 
2123 	if ((ifp->if_flags & IFF_LOOPBACK) == 0)
2124 		origifp = ifp;
2125 
2126 	error = (*ifp->if_output)(origifp, m, (struct sockaddr *)dst, ro);
2127 	return (error);
2128 }
2129 
2130 /*
2131  * Lookup link headerfor @sa_dst address. Stores found
2132  * data in @desten buffer. Copy of lle ln_flags can be also
2133  * saved in @pflags if @pflags is non-NULL.
2134  *
2135  * If destination LLE does not exists or lle state modification
2136  * is required, call "slow" version.
2137  *
2138  * Return values:
2139  * - 0 on success (address copied to buffer).
2140  * - EWOULDBLOCK (no local error, but address is still unresolved)
2141  * - other errors (alloc failure, etc)
2142  */
2143 int
2144 nd6_resolve(struct ifnet *ifp, int is_gw, struct mbuf *m,
2145     const struct sockaddr *sa_dst, u_char *desten, uint32_t *pflags)
2146 {
2147 	struct llentry *ln = NULL;
2148 	const struct sockaddr_in6 *dst6;
2149 
2150 	if (pflags != NULL)
2151 		*pflags = 0;
2152 
2153 	dst6 = (const struct sockaddr_in6 *)sa_dst;
2154 
2155 	/* discard the packet if IPv6 operation is disabled on the interface */
2156 	if ((ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED)) {
2157 		m_freem(m);
2158 		return (ENETDOWN); /* better error? */
2159 	}
2160 
2161 	if (m != NULL && m->m_flags & M_MCAST) {
2162 		switch (ifp->if_type) {
2163 		case IFT_ETHER:
2164 		case IFT_FDDI:
2165 		case IFT_L2VLAN:
2166 		case IFT_IEEE80211:
2167 		case IFT_BRIDGE:
2168 		case IFT_ISO88025:
2169 			ETHER_MAP_IPV6_MULTICAST(&dst6->sin6_addr,
2170 						 desten);
2171 			return (0);
2172 		default:
2173 			m_freem(m);
2174 			return (EAFNOSUPPORT);
2175 		}
2176 	}
2177 
2178 	IF_AFDATA_RLOCK(ifp);
2179 	ln = nd6_lookup(&dst6->sin6_addr, LLE_UNLOCKED, ifp);
2180 	if (ln != NULL && (ln->r_flags & RLLE_VALID) != 0) {
2181 		/* Entry found, let's copy lle info */
2182 		bcopy(ln->r_linkdata, desten, ln->r_hdrlen);
2183 		if (pflags != NULL)
2184 			*pflags = LLE_VALID | (ln->r_flags & RLLE_IFADDR);
2185 		/* Check if we have feedback request from nd6 timer */
2186 		if (ln->r_skip_req != 0) {
2187 			LLE_REQ_LOCK(ln);
2188 			ln->r_skip_req = 0; /* Notify that entry was used */
2189 			ln->lle_hittime = time_uptime;
2190 			LLE_REQ_UNLOCK(ln);
2191 		}
2192 		IF_AFDATA_RUNLOCK(ifp);
2193 		return (0);
2194 	}
2195 	IF_AFDATA_RUNLOCK(ifp);
2196 
2197 	return (nd6_resolve_slow(ifp, 0, m, dst6, desten, pflags));
2198 }
2199 
2200 
2201 /*
2202  * Do L2 address resolution for @sa_dst address. Stores found
2203  * address in @desten buffer. Copy of lle ln_flags can be also
2204  * saved in @pflags if @pflags is non-NULL.
2205  *
2206  * Heavy version.
2207  * Function assume that destination LLE does not exist,
2208  * is invalid or stale, so LLE_EXCLUSIVE lock needs to be acquired.
2209  *
2210  * Set noinline to be dtrace-friendly
2211  */
2212 static __noinline int
2213 nd6_resolve_slow(struct ifnet *ifp, int flags, struct mbuf *m,
2214     const struct sockaddr_in6 *dst, u_char *desten, uint32_t *pflags)
2215 {
2216 	struct llentry *lle = NULL, *lle_tmp;
2217 	struct in6_addr *psrc, src;
2218 	int send_ns, ll_len;
2219 	char *lladdr;
2220 
2221 	/*
2222 	 * Address resolution or Neighbor Unreachability Detection
2223 	 * for the next hop.
2224 	 * At this point, the destination of the packet must be a unicast
2225 	 * or an anycast address(i.e. not a multicast).
2226 	 */
2227 	if (lle == NULL) {
2228 		IF_AFDATA_RLOCK(ifp);
2229 		lle = nd6_lookup(&dst->sin6_addr, LLE_EXCLUSIVE, ifp);
2230 		IF_AFDATA_RUNLOCK(ifp);
2231 		if ((lle == NULL) && nd6_is_addr_neighbor(dst, ifp))  {
2232 			/*
2233 			 * Since nd6_is_addr_neighbor() internally calls nd6_lookup(),
2234 			 * the condition below is not very efficient.  But we believe
2235 			 * it is tolerable, because this should be a rare case.
2236 			 */
2237 			lle = nd6_alloc(&dst->sin6_addr, 0, ifp);
2238 			if (lle == NULL) {
2239 				char ip6buf[INET6_ADDRSTRLEN];
2240 				log(LOG_DEBUG,
2241 				    "nd6_output: can't allocate llinfo for %s "
2242 				    "(ln=%p)\n",
2243 				    ip6_sprintf(ip6buf, &dst->sin6_addr), lle);
2244 				m_freem(m);
2245 				return (ENOBUFS);
2246 			}
2247 
2248 			IF_AFDATA_WLOCK(ifp);
2249 			LLE_WLOCK(lle);
2250 			/* Prefer any existing entry over newly-created one */
2251 			lle_tmp = nd6_lookup(&dst->sin6_addr, LLE_EXCLUSIVE, ifp);
2252 			if (lle_tmp == NULL)
2253 				lltable_link_entry(LLTABLE6(ifp), lle);
2254 			IF_AFDATA_WUNLOCK(ifp);
2255 			if (lle_tmp != NULL) {
2256 				lltable_free_entry(LLTABLE6(ifp), lle);
2257 				lle = lle_tmp;
2258 				lle_tmp = NULL;
2259 			}
2260 		}
2261 	}
2262 	if (lle == NULL) {
2263 		if (!(ND_IFINFO(ifp)->flags & ND6_IFF_PERFORMNUD)) {
2264 			m_freem(m);
2265 			return (ENOBUFS);
2266 		}
2267 
2268 		if (m != NULL)
2269 			m_freem(m);
2270 		return (ENOBUFS);
2271 	}
2272 
2273 	LLE_WLOCK_ASSERT(lle);
2274 
2275 	/*
2276 	 * The first time we send a packet to a neighbor whose entry is
2277 	 * STALE, we have to change the state to DELAY and a sets a timer to
2278 	 * expire in DELAY_FIRST_PROBE_TIME seconds to ensure do
2279 	 * neighbor unreachability detection on expiration.
2280 	 * (RFC 2461 7.3.3)
2281 	 */
2282 	if (lle->ln_state == ND6_LLINFO_STALE)
2283 		nd6_llinfo_setstate(lle, ND6_LLINFO_DELAY);
2284 
2285 	/*
2286 	 * If the neighbor cache entry has a state other than INCOMPLETE
2287 	 * (i.e. its link-layer address is already resolved), just
2288 	 * send the packet.
2289 	 */
2290 	if (lle->ln_state > ND6_LLINFO_INCOMPLETE) {
2291 		if (flags & LLE_ADDRONLY) {
2292 			lladdr = lle->ll_addr;
2293 			ll_len = ifp->if_addrlen;
2294 		} else {
2295 			lladdr = lle->r_linkdata;
2296 			ll_len = lle->r_hdrlen;
2297 		}
2298 		bcopy(lladdr, desten, ll_len);
2299 		if (pflags != NULL)
2300 			*pflags = lle->la_flags;
2301 		LLE_WUNLOCK(lle);
2302 		return (0);
2303 	}
2304 
2305 	/*
2306 	 * There is a neighbor cache entry, but no ethernet address
2307 	 * response yet.  Append this latest packet to the end of the
2308 	 * packet queue in the mbuf, unless the number of the packet
2309 	 * does not exceed nd6_maxqueuelen.  When it exceeds nd6_maxqueuelen,
2310 	 * the oldest packet in the queue will be removed.
2311 	 */
2312 
2313 	if (lle->la_hold != NULL) {
2314 		struct mbuf *m_hold;
2315 		int i;
2316 
2317 		i = 0;
2318 		for (m_hold = lle->la_hold; m_hold; m_hold = m_hold->m_nextpkt){
2319 			i++;
2320 			if (m_hold->m_nextpkt == NULL) {
2321 				m_hold->m_nextpkt = m;
2322 				break;
2323 			}
2324 		}
2325 		while (i >= V_nd6_maxqueuelen) {
2326 			m_hold = lle->la_hold;
2327 			lle->la_hold = lle->la_hold->m_nextpkt;
2328 			m_freem(m_hold);
2329 			i--;
2330 		}
2331 	} else {
2332 		lle->la_hold = m;
2333 	}
2334 
2335 	/*
2336 	 * If there has been no NS for the neighbor after entering the
2337 	 * INCOMPLETE state, send the first solicitation.
2338 	 * Note that for newly-created lle la_asked will be 0,
2339 	 * so we will transition from ND6_LLINFO_NOSTATE to
2340 	 * ND6_LLINFO_INCOMPLETE state here.
2341 	 */
2342 	psrc = NULL;
2343 	send_ns = 0;
2344 	if (lle->la_asked == 0) {
2345 		lle->la_asked++;
2346 		send_ns = 1;
2347 		psrc = nd6_llinfo_get_holdsrc(lle, &src);
2348 
2349 		nd6_llinfo_setstate(lle, ND6_LLINFO_INCOMPLETE);
2350 	}
2351 	LLE_WUNLOCK(lle);
2352 	if (send_ns != 0)
2353 		nd6_ns_output(ifp, psrc, NULL, &dst->sin6_addr, NULL);
2354 
2355 	return (EWOULDBLOCK);
2356 }
2357 
2358 /*
2359  * Do L2 address resolution for @sa_dst address. Stores found
2360  * address in @desten buffer. Copy of lle ln_flags can be also
2361  * saved in @pflags if @pflags is non-NULL.
2362  *
2363  * Return values:
2364  * - 0 on success (address copied to buffer).
2365  * - EWOULDBLOCK (no local error, but address is still unresolved)
2366  * - other errors (alloc failure, etc)
2367  */
2368 int
2369 nd6_resolve_addr(struct ifnet *ifp, int flags, const struct sockaddr *dst,
2370     char *desten, uint32_t *pflags)
2371 {
2372 	int error;
2373 
2374 	flags |= LLE_ADDRONLY;
2375 	error = nd6_resolve_slow(ifp, flags, NULL,
2376 	    (const struct sockaddr_in6 *)dst, desten, pflags);
2377 	return (error);
2378 }
2379 
2380 int
2381 nd6_flush_holdchain(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *chain,
2382     struct sockaddr_in6 *dst)
2383 {
2384 	struct mbuf *m, *m_head;
2385 	struct ifnet *outifp;
2386 	int error = 0;
2387 
2388 	m_head = chain;
2389 	if ((ifp->if_flags & IFF_LOOPBACK) != 0)
2390 		outifp = origifp;
2391 	else
2392 		outifp = ifp;
2393 
2394 	while (m_head) {
2395 		m = m_head;
2396 		m_head = m_head->m_nextpkt;
2397 		error = nd6_output_ifp(ifp, origifp, m, dst, NULL);
2398 	}
2399 
2400 	/*
2401 	 * XXX
2402 	 * note that intermediate errors are blindly ignored
2403 	 */
2404 	return (error);
2405 }
2406 
2407 static int
2408 nd6_need_cache(struct ifnet *ifp)
2409 {
2410 	/*
2411 	 * XXX: we currently do not make neighbor cache on any interface
2412 	 * other than ARCnet, Ethernet, FDDI and GIF.
2413 	 *
2414 	 * RFC2893 says:
2415 	 * - unidirectional tunnels needs no ND
2416 	 */
2417 	switch (ifp->if_type) {
2418 	case IFT_ARCNET:
2419 	case IFT_ETHER:
2420 	case IFT_FDDI:
2421 	case IFT_IEEE1394:
2422 	case IFT_L2VLAN:
2423 	case IFT_IEEE80211:
2424 	case IFT_INFINIBAND:
2425 	case IFT_BRIDGE:
2426 	case IFT_PROPVIRTUAL:
2427 		return (1);
2428 	default:
2429 		return (0);
2430 	}
2431 }
2432 
2433 /*
2434  * Add pernament ND6 link-layer record for given
2435  * interface address.
2436  *
2437  * Very similar to IPv4 arp_ifinit(), but:
2438  * 1) IPv6 DAD is performed in different place
2439  * 2) It is called by IPv6 protocol stack in contrast to
2440  * arp_ifinit() which is typically called in SIOCSIFADDR
2441  * driver ioctl handler.
2442  *
2443  */
2444 int
2445 nd6_add_ifa_lle(struct in6_ifaddr *ia)
2446 {
2447 	struct ifnet *ifp;
2448 	struct llentry *ln, *ln_tmp;
2449 	struct sockaddr *dst;
2450 
2451 	ifp = ia->ia_ifa.ifa_ifp;
2452 	if (nd6_need_cache(ifp) == 0)
2453 		return (0);
2454 
2455 	ia->ia_ifa.ifa_rtrequest = nd6_rtrequest;
2456 	dst = (struct sockaddr *)&ia->ia_addr;
2457 	ln = lltable_alloc_entry(LLTABLE6(ifp), LLE_IFADDR, dst);
2458 	if (ln == NULL)
2459 		return (ENOBUFS);
2460 
2461 	IF_AFDATA_WLOCK(ifp);
2462 	LLE_WLOCK(ln);
2463 	/* Unlink any entry if exists */
2464 	ln_tmp = lla_lookup(LLTABLE6(ifp), LLE_EXCLUSIVE, dst);
2465 	if (ln_tmp != NULL)
2466 		lltable_unlink_entry(LLTABLE6(ifp), ln_tmp);
2467 	lltable_link_entry(LLTABLE6(ifp), ln);
2468 	IF_AFDATA_WUNLOCK(ifp);
2469 
2470 	if (ln_tmp != NULL)
2471 		EVENTHANDLER_INVOKE(lle_event, ln_tmp, LLENTRY_EXPIRED);
2472 	EVENTHANDLER_INVOKE(lle_event, ln, LLENTRY_RESOLVED);
2473 
2474 	LLE_WUNLOCK(ln);
2475 	if (ln_tmp != NULL)
2476 		llentry_free(ln_tmp);
2477 
2478 	return (0);
2479 }
2480 
2481 /*
2482  * Removes either all lle entries for given @ia, or lle
2483  * corresponding to @ia address.
2484  */
2485 void
2486 nd6_rem_ifa_lle(struct in6_ifaddr *ia, int all)
2487 {
2488 	struct sockaddr_in6 mask, addr;
2489 	struct sockaddr *saddr, *smask;
2490 	struct ifnet *ifp;
2491 
2492 	ifp = ia->ia_ifa.ifa_ifp;
2493 	memcpy(&addr, &ia->ia_addr, sizeof(ia->ia_addr));
2494 	memcpy(&mask, &ia->ia_prefixmask, sizeof(ia->ia_prefixmask));
2495 	saddr = (struct sockaddr *)&addr;
2496 	smask = (struct sockaddr *)&mask;
2497 
2498 	if (all != 0)
2499 		lltable_prefix_free(AF_INET6, saddr, smask, LLE_STATIC);
2500 	else
2501 		lltable_delete_addr(LLTABLE6(ifp), LLE_IFADDR, saddr);
2502 }
2503 
2504 static void
2505 clear_llinfo_pqueue(struct llentry *ln)
2506 {
2507 	struct mbuf *m_hold, *m_hold_next;
2508 
2509 	for (m_hold = ln->la_hold; m_hold; m_hold = m_hold_next) {
2510 		m_hold_next = m_hold->m_nextpkt;
2511 		m_freem(m_hold);
2512 	}
2513 
2514 	ln->la_hold = NULL;
2515 }
2516 
2517 static int nd6_sysctl_drlist(SYSCTL_HANDLER_ARGS);
2518 static int nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS);
2519 #ifdef SYSCTL_DECL
2520 SYSCTL_DECL(_net_inet6_icmp6);
2521 #endif
2522 SYSCTL_NODE(_net_inet6_icmp6, ICMPV6CTL_ND6_DRLIST, nd6_drlist,
2523 	CTLFLAG_RD, nd6_sysctl_drlist, "");
2524 SYSCTL_NODE(_net_inet6_icmp6, ICMPV6CTL_ND6_PRLIST, nd6_prlist,
2525 	CTLFLAG_RD, nd6_sysctl_prlist, "");
2526 SYSCTL_INT(_net_inet6_icmp6, ICMPV6CTL_ND6_MAXQLEN, nd6_maxqueuelen,
2527 	CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(nd6_maxqueuelen), 1, "");
2528 SYSCTL_INT(_net_inet6_icmp6, OID_AUTO, nd6_gctimer,
2529 	CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(nd6_gctimer), (60 * 60 * 24), "");
2530 
2531 static int
2532 nd6_sysctl_drlist(SYSCTL_HANDLER_ARGS)
2533 {
2534 	struct in6_defrouter d;
2535 	struct nd_defrouter *dr;
2536 	int error;
2537 
2538 	if (req->newptr)
2539 		return (EPERM);
2540 
2541 	bzero(&d, sizeof(d));
2542 	d.rtaddr.sin6_family = AF_INET6;
2543 	d.rtaddr.sin6_len = sizeof(d.rtaddr);
2544 
2545 	/*
2546 	 * XXX locking
2547 	 */
2548 	TAILQ_FOREACH(dr, &V_nd_defrouter, dr_entry) {
2549 		d.rtaddr.sin6_addr = dr->rtaddr;
2550 		error = sa6_recoverscope(&d.rtaddr);
2551 		if (error != 0)
2552 			return (error);
2553 		d.flags = dr->raflags;
2554 		d.rtlifetime = dr->rtlifetime;
2555 		d.expire = dr->expire + (time_second - time_uptime);
2556 		d.if_index = dr->ifp->if_index;
2557 		error = SYSCTL_OUT(req, &d, sizeof(d));
2558 		if (error != 0)
2559 			return (error);
2560 	}
2561 	return (0);
2562 }
2563 
2564 static int
2565 nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS)
2566 {
2567 	struct in6_prefix p;
2568 	struct sockaddr_in6 s6;
2569 	struct nd_prefix *pr;
2570 	struct nd_pfxrouter *pfr;
2571 	time_t maxexpire;
2572 	int error;
2573 	char ip6buf[INET6_ADDRSTRLEN];
2574 
2575 	if (req->newptr)
2576 		return (EPERM);
2577 
2578 	bzero(&p, sizeof(p));
2579 	p.origin = PR_ORIG_RA;
2580 	bzero(&s6, sizeof(s6));
2581 	s6.sin6_family = AF_INET6;
2582 	s6.sin6_len = sizeof(s6);
2583 
2584 	/*
2585 	 * XXX locking
2586 	 */
2587 	LIST_FOREACH(pr, &V_nd_prefix, ndpr_entry) {
2588 		p.prefix = pr->ndpr_prefix;
2589 		if (sa6_recoverscope(&p.prefix)) {
2590 			log(LOG_ERR, "scope error in prefix list (%s)\n",
2591 			    ip6_sprintf(ip6buf, &p.prefix.sin6_addr));
2592 			/* XXX: press on... */
2593 		}
2594 		p.raflags = pr->ndpr_raf;
2595 		p.prefixlen = pr->ndpr_plen;
2596 		p.vltime = pr->ndpr_vltime;
2597 		p.pltime = pr->ndpr_pltime;
2598 		p.if_index = pr->ndpr_ifp->if_index;
2599 		if (pr->ndpr_vltime == ND6_INFINITE_LIFETIME)
2600 			p.expire = 0;
2601 		else {
2602 			/* XXX: we assume time_t is signed. */
2603 			maxexpire = (-1) &
2604 			    ~((time_t)1 << ((sizeof(maxexpire) * 8) - 1));
2605 			if (pr->ndpr_vltime < maxexpire - pr->ndpr_lastupdate)
2606 				p.expire = pr->ndpr_lastupdate +
2607 				    pr->ndpr_vltime +
2608 				    (time_second - time_uptime);
2609 			else
2610 				p.expire = maxexpire;
2611 		}
2612 		p.refcnt = pr->ndpr_refcnt;
2613 		p.flags = pr->ndpr_stateflags;
2614 		p.advrtrs = 0;
2615 		LIST_FOREACH(pfr, &pr->ndpr_advrtrs, pfr_entry)
2616 			p.advrtrs++;
2617 		error = SYSCTL_OUT(req, &p, sizeof(p));
2618 		if (error != 0)
2619 			return (error);
2620 		LIST_FOREACH(pfr, &pr->ndpr_advrtrs, pfr_entry) {
2621 			s6.sin6_addr = pfr->router->rtaddr;
2622 			if (sa6_recoverscope(&s6))
2623 				log(LOG_ERR,
2624 				    "scope error in prefix list (%s)\n",
2625 				    ip6_sprintf(ip6buf, &pfr->router->rtaddr));
2626 			error = SYSCTL_OUT(req, &s6, sizeof(s6));
2627 			if (error != 0)
2628 				return (error);
2629 		}
2630 	}
2631 	return (0);
2632 }
2633