xref: /freebsd/sys/netinet6/nd6.c (revision 3fc36ee018bb836bd1796067cf4ef8683f166ebc)
1 /*-
2  * Copyright (C) 1995, 1996, 1997, and 1998 WIDE Project.
3  * All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice, this list of conditions and the following disclaimer.
10  * 2. Redistributions in binary form must reproduce the above copyright
11  *    notice, this list of conditions and the following disclaimer in the
12  *    documentation and/or other materials provided with the distribution.
13  * 3. Neither the name of the project nor the names of its contributors
14  *    may be used to endorse or promote products derived from this software
15  *    without specific prior written permission.
16  *
17  * THIS SOFTWARE IS PROVIDED BY THE PROJECT AND CONTRIBUTORS ``AS IS'' AND
18  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20  * ARE DISCLAIMED.  IN NO EVENT SHALL THE PROJECT OR CONTRIBUTORS BE LIABLE
21  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27  * SUCH DAMAGE.
28  *
29  *	$KAME: nd6.c,v 1.144 2001/05/24 07:44:00 itojun Exp $
30  */
31 
32 #include <sys/cdefs.h>
33 __FBSDID("$FreeBSD$");
34 
35 #include "opt_inet.h"
36 #include "opt_inet6.h"
37 
38 #include <sys/param.h>
39 #include <sys/systm.h>
40 #include <sys/callout.h>
41 #include <sys/malloc.h>
42 #include <sys/mbuf.h>
43 #include <sys/socket.h>
44 #include <sys/sockio.h>
45 #include <sys/time.h>
46 #include <sys/kernel.h>
47 #include <sys/protosw.h>
48 #include <sys/errno.h>
49 #include <sys/syslog.h>
50 #include <sys/lock.h>
51 #include <sys/rwlock.h>
52 #include <sys/queue.h>
53 #include <sys/sdt.h>
54 #include <sys/sysctl.h>
55 
56 #include <net/if.h>
57 #include <net/if_var.h>
58 #include <net/if_arc.h>
59 #include <net/if_dl.h>
60 #include <net/if_types.h>
61 #include <net/iso88025.h>
62 #include <net/fddi.h>
63 #include <net/route.h>
64 #include <net/vnet.h>
65 
66 #include <netinet/in.h>
67 #include <netinet/in_kdtrace.h>
68 #include <net/if_llatbl.h>
69 #include <netinet/if_ether.h>
70 #include <netinet6/in6_var.h>
71 #include <netinet/ip6.h>
72 #include <netinet6/ip6_var.h>
73 #include <netinet6/scope6_var.h>
74 #include <netinet6/nd6.h>
75 #include <netinet6/in6_ifattach.h>
76 #include <netinet/icmp6.h>
77 #include <netinet6/send.h>
78 
79 #include <sys/limits.h>
80 
81 #include <security/mac/mac_framework.h>
82 
83 #define ND6_SLOWTIMER_INTERVAL (60 * 60) /* 1 hour */
84 #define ND6_RECALC_REACHTM_INTERVAL (60 * 120) /* 2 hours */
85 
86 #define SIN6(s) ((const struct sockaddr_in6 *)(s))
87 
88 MALLOC_DEFINE(M_IP6NDP, "ip6ndp", "IPv6 Neighbor Discovery");
89 
90 /* timer values */
91 VNET_DEFINE(int, nd6_prune)	= 1;	/* walk list every 1 seconds */
92 VNET_DEFINE(int, nd6_delay)	= 5;	/* delay first probe time 5 second */
93 VNET_DEFINE(int, nd6_umaxtries)	= 3;	/* maximum unicast query */
94 VNET_DEFINE(int, nd6_mmaxtries)	= 3;	/* maximum multicast query */
95 VNET_DEFINE(int, nd6_useloopback) = 1;	/* use loopback interface for
96 					 * local traffic */
97 VNET_DEFINE(int, nd6_gctimer)	= (60 * 60 * 24); /* 1 day: garbage
98 					 * collection timer */
99 
100 /* preventing too many loops in ND option parsing */
101 static VNET_DEFINE(int, nd6_maxndopt) = 10; /* max # of ND options allowed */
102 
103 VNET_DEFINE(int, nd6_maxnudhint) = 0;	/* max # of subsequent upper
104 					 * layer hints */
105 static VNET_DEFINE(int, nd6_maxqueuelen) = 1; /* max pkts cached in unresolved
106 					 * ND entries */
107 #define	V_nd6_maxndopt			VNET(nd6_maxndopt)
108 #define	V_nd6_maxqueuelen		VNET(nd6_maxqueuelen)
109 
110 #ifdef ND6_DEBUG
111 VNET_DEFINE(int, nd6_debug) = 1;
112 #else
113 VNET_DEFINE(int, nd6_debug) = 0;
114 #endif
115 
116 static eventhandler_tag lle_event_eh, iflladdr_event_eh;
117 
118 VNET_DEFINE(struct nd_drhead, nd_defrouter);
119 VNET_DEFINE(struct nd_prhead, nd_prefix);
120 VNET_DEFINE(struct rwlock, nd6_lock);
121 
122 VNET_DEFINE(int, nd6_recalc_reachtm_interval) = ND6_RECALC_REACHTM_INTERVAL;
123 #define	V_nd6_recalc_reachtm_interval	VNET(nd6_recalc_reachtm_interval)
124 
125 int	(*send_sendso_input_hook)(struct mbuf *, struct ifnet *, int, int);
126 
127 static int nd6_is_new_addr_neighbor(const struct sockaddr_in6 *,
128 	struct ifnet *);
129 static void nd6_setmtu0(struct ifnet *, struct nd_ifinfo *);
130 static void nd6_slowtimo(void *);
131 static int regen_tmpaddr(struct in6_ifaddr *);
132 static void nd6_free(struct llentry **, int);
133 static void nd6_free_redirect(const struct llentry *);
134 static void nd6_llinfo_timer(void *);
135 static void nd6_llinfo_settimer_locked(struct llentry *, long);
136 static void clear_llinfo_pqueue(struct llentry *);
137 static void nd6_rtrequest(int, struct rtentry *, struct rt_addrinfo *);
138 static int nd6_resolve_slow(struct ifnet *, int, struct mbuf *,
139     const struct sockaddr_in6 *, u_char *, uint32_t *, struct llentry **);
140 static int nd6_need_cache(struct ifnet *);
141 
142 
143 static VNET_DEFINE(struct callout, nd6_slowtimo_ch);
144 #define	V_nd6_slowtimo_ch		VNET(nd6_slowtimo_ch)
145 
146 VNET_DEFINE(struct callout, nd6_timer_ch);
147 #define	V_nd6_timer_ch			VNET(nd6_timer_ch)
148 
149 static void
150 nd6_lle_event(void *arg __unused, struct llentry *lle, int evt)
151 {
152 	struct rt_addrinfo rtinfo;
153 	struct sockaddr_in6 dst;
154 	struct sockaddr_dl gw;
155 	struct ifnet *ifp;
156 	int type;
157 
158 	LLE_WLOCK_ASSERT(lle);
159 
160 	if (lltable_get_af(lle->lle_tbl) != AF_INET6)
161 		return;
162 
163 	switch (evt) {
164 	case LLENTRY_RESOLVED:
165 		type = RTM_ADD;
166 		KASSERT(lle->la_flags & LLE_VALID,
167 		    ("%s: %p resolved but not valid?", __func__, lle));
168 		break;
169 	case LLENTRY_EXPIRED:
170 		type = RTM_DELETE;
171 		break;
172 	default:
173 		return;
174 	}
175 
176 	ifp = lltable_get_ifp(lle->lle_tbl);
177 
178 	bzero(&dst, sizeof(dst));
179 	bzero(&gw, sizeof(gw));
180 	bzero(&rtinfo, sizeof(rtinfo));
181 	lltable_fill_sa_entry(lle, (struct sockaddr *)&dst);
182 	dst.sin6_scope_id = in6_getscopezone(ifp,
183 	    in6_addrscope(&dst.sin6_addr));
184 	gw.sdl_len = sizeof(struct sockaddr_dl);
185 	gw.sdl_family = AF_LINK;
186 	gw.sdl_alen = ifp->if_addrlen;
187 	gw.sdl_index = ifp->if_index;
188 	gw.sdl_type = ifp->if_type;
189 	if (evt == LLENTRY_RESOLVED)
190 		bcopy(lle->ll_addr, gw.sdl_data, ifp->if_addrlen);
191 	rtinfo.rti_info[RTAX_DST] = (struct sockaddr *)&dst;
192 	rtinfo.rti_info[RTAX_GATEWAY] = (struct sockaddr *)&gw;
193 	rtinfo.rti_addrs = RTA_DST | RTA_GATEWAY;
194 	rt_missmsg_fib(type, &rtinfo, RTF_HOST | RTF_LLDATA | (
195 	    type == RTM_ADD ? RTF_UP: 0), 0, RT_DEFAULT_FIB);
196 }
197 
198 /*
199  * A handler for interface link layer address change event.
200  */
201 static void
202 nd6_iflladdr(void *arg __unused, struct ifnet *ifp)
203 {
204 
205 	lltable_update_ifaddr(LLTABLE6(ifp));
206 }
207 
208 void
209 nd6_init(void)
210 {
211 
212 	rw_init(&V_nd6_lock, "nd6");
213 
214 	LIST_INIT(&V_nd_prefix);
215 
216 	/* initialization of the default router list */
217 	TAILQ_INIT(&V_nd_defrouter);
218 
219 	/* Start timers. */
220 	callout_init(&V_nd6_slowtimo_ch, 0);
221 	callout_reset(&V_nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
222 	    nd6_slowtimo, curvnet);
223 
224 	callout_init(&V_nd6_timer_ch, 0);
225 	callout_reset(&V_nd6_timer_ch, hz, nd6_timer, curvnet);
226 
227 	nd6_dad_init();
228 	if (IS_DEFAULT_VNET(curvnet)) {
229 		lle_event_eh = EVENTHANDLER_REGISTER(lle_event, nd6_lle_event,
230 		    NULL, EVENTHANDLER_PRI_ANY);
231 		iflladdr_event_eh = EVENTHANDLER_REGISTER(iflladdr_event,
232 		    nd6_iflladdr, NULL, EVENTHANDLER_PRI_ANY);
233 	}
234 }
235 
236 #ifdef VIMAGE
237 void
238 nd6_destroy()
239 {
240 
241 	callout_drain(&V_nd6_slowtimo_ch);
242 	callout_drain(&V_nd6_timer_ch);
243 	if (IS_DEFAULT_VNET(curvnet)) {
244 		EVENTHANDLER_DEREGISTER(lle_event, lle_event_eh);
245 		EVENTHANDLER_DEREGISTER(iflladdr_event, iflladdr_event_eh);
246 	}
247 	rw_destroy(&V_nd6_lock);
248 }
249 #endif
250 
251 struct nd_ifinfo *
252 nd6_ifattach(struct ifnet *ifp)
253 {
254 	struct nd_ifinfo *nd;
255 
256 	nd = malloc(sizeof(*nd), M_IP6NDP, M_WAITOK | M_ZERO);
257 	nd->initialized = 1;
258 
259 	nd->chlim = IPV6_DEFHLIM;
260 	nd->basereachable = REACHABLE_TIME;
261 	nd->reachable = ND_COMPUTE_RTIME(nd->basereachable);
262 	nd->retrans = RETRANS_TIMER;
263 
264 	nd->flags = ND6_IFF_PERFORMNUD;
265 
266 	/* A loopback interface always has ND6_IFF_AUTO_LINKLOCAL.
267 	 * XXXHRS: Clear ND6_IFF_AUTO_LINKLOCAL on an IFT_BRIDGE interface by
268 	 * default regardless of the V_ip6_auto_linklocal configuration to
269 	 * give a reasonable default behavior.
270 	 */
271 	if ((V_ip6_auto_linklocal && ifp->if_type != IFT_BRIDGE) ||
272 	    (ifp->if_flags & IFF_LOOPBACK))
273 		nd->flags |= ND6_IFF_AUTO_LINKLOCAL;
274 	/*
275 	 * A loopback interface does not need to accept RTADV.
276 	 * XXXHRS: Clear ND6_IFF_ACCEPT_RTADV on an IFT_BRIDGE interface by
277 	 * default regardless of the V_ip6_accept_rtadv configuration to
278 	 * prevent the interface from accepting RA messages arrived
279 	 * on one of the member interfaces with ND6_IFF_ACCEPT_RTADV.
280 	 */
281 	if (V_ip6_accept_rtadv &&
282 	    !(ifp->if_flags & IFF_LOOPBACK) &&
283 	    (ifp->if_type != IFT_BRIDGE))
284 			nd->flags |= ND6_IFF_ACCEPT_RTADV;
285 	if (V_ip6_no_radr && !(ifp->if_flags & IFF_LOOPBACK))
286 		nd->flags |= ND6_IFF_NO_RADR;
287 
288 	/* XXX: we cannot call nd6_setmtu since ifp is not fully initialized */
289 	nd6_setmtu0(ifp, nd);
290 
291 	return nd;
292 }
293 
294 void
295 nd6_ifdetach(struct ifnet *ifp, struct nd_ifinfo *nd)
296 {
297 	struct ifaddr *ifa, *next;
298 
299 	IF_ADDR_RLOCK(ifp);
300 	TAILQ_FOREACH_SAFE(ifa, &ifp->if_addrhead, ifa_link, next) {
301 		if (ifa->ifa_addr->sa_family != AF_INET6)
302 			continue;
303 
304 		/* stop DAD processing */
305 		nd6_dad_stop(ifa);
306 	}
307 	IF_ADDR_RUNLOCK(ifp);
308 
309 	free(nd, M_IP6NDP);
310 }
311 
312 /*
313  * Reset ND level link MTU. This function is called when the physical MTU
314  * changes, which means we might have to adjust the ND level MTU.
315  */
316 void
317 nd6_setmtu(struct ifnet *ifp)
318 {
319 	if (ifp->if_afdata[AF_INET6] == NULL)
320 		return;
321 
322 	nd6_setmtu0(ifp, ND_IFINFO(ifp));
323 }
324 
325 /* XXX todo: do not maintain copy of ifp->if_mtu in ndi->maxmtu */
326 void
327 nd6_setmtu0(struct ifnet *ifp, struct nd_ifinfo *ndi)
328 {
329 	u_int32_t omaxmtu;
330 
331 	omaxmtu = ndi->maxmtu;
332 
333 	switch (ifp->if_type) {
334 	case IFT_ARCNET:
335 		ndi->maxmtu = MIN(ARC_PHDS_MAXMTU, ifp->if_mtu); /* RFC2497 */
336 		break;
337 	case IFT_FDDI:
338 		ndi->maxmtu = MIN(FDDIIPMTU, ifp->if_mtu); /* RFC2467 */
339 		break;
340 	case IFT_ISO88025:
341 		 ndi->maxmtu = MIN(ISO88025_MAX_MTU, ifp->if_mtu);
342 		 break;
343 	default:
344 		ndi->maxmtu = ifp->if_mtu;
345 		break;
346 	}
347 
348 	/*
349 	 * Decreasing the interface MTU under IPV6 minimum MTU may cause
350 	 * undesirable situation.  We thus notify the operator of the change
351 	 * explicitly.  The check for omaxmtu is necessary to restrict the
352 	 * log to the case of changing the MTU, not initializing it.
353 	 */
354 	if (omaxmtu >= IPV6_MMTU && ndi->maxmtu < IPV6_MMTU) {
355 		log(LOG_NOTICE, "nd6_setmtu0: "
356 		    "new link MTU on %s (%lu) is too small for IPv6\n",
357 		    if_name(ifp), (unsigned long)ndi->maxmtu);
358 	}
359 
360 	if (ndi->maxmtu > V_in6_maxmtu)
361 		in6_setmaxmtu(); /* check all interfaces just in case */
362 
363 }
364 
365 void
366 nd6_option_init(void *opt, int icmp6len, union nd_opts *ndopts)
367 {
368 
369 	bzero(ndopts, sizeof(*ndopts));
370 	ndopts->nd_opts_search = (struct nd_opt_hdr *)opt;
371 	ndopts->nd_opts_last
372 		= (struct nd_opt_hdr *)(((u_char *)opt) + icmp6len);
373 
374 	if (icmp6len == 0) {
375 		ndopts->nd_opts_done = 1;
376 		ndopts->nd_opts_search = NULL;
377 	}
378 }
379 
380 /*
381  * Take one ND option.
382  */
383 struct nd_opt_hdr *
384 nd6_option(union nd_opts *ndopts)
385 {
386 	struct nd_opt_hdr *nd_opt;
387 	int olen;
388 
389 	KASSERT(ndopts != NULL, ("%s: ndopts == NULL", __func__));
390 	KASSERT(ndopts->nd_opts_last != NULL, ("%s: uninitialized ndopts",
391 	    __func__));
392 	if (ndopts->nd_opts_search == NULL)
393 		return NULL;
394 	if (ndopts->nd_opts_done)
395 		return NULL;
396 
397 	nd_opt = ndopts->nd_opts_search;
398 
399 	/* make sure nd_opt_len is inside the buffer */
400 	if ((caddr_t)&nd_opt->nd_opt_len >= (caddr_t)ndopts->nd_opts_last) {
401 		bzero(ndopts, sizeof(*ndopts));
402 		return NULL;
403 	}
404 
405 	olen = nd_opt->nd_opt_len << 3;
406 	if (olen == 0) {
407 		/*
408 		 * Message validation requires that all included
409 		 * options have a length that is greater than zero.
410 		 */
411 		bzero(ndopts, sizeof(*ndopts));
412 		return NULL;
413 	}
414 
415 	ndopts->nd_opts_search = (struct nd_opt_hdr *)((caddr_t)nd_opt + olen);
416 	if (ndopts->nd_opts_search > ndopts->nd_opts_last) {
417 		/* option overruns the end of buffer, invalid */
418 		bzero(ndopts, sizeof(*ndopts));
419 		return NULL;
420 	} else if (ndopts->nd_opts_search == ndopts->nd_opts_last) {
421 		/* reached the end of options chain */
422 		ndopts->nd_opts_done = 1;
423 		ndopts->nd_opts_search = NULL;
424 	}
425 	return nd_opt;
426 }
427 
428 /*
429  * Parse multiple ND options.
430  * This function is much easier to use, for ND routines that do not need
431  * multiple options of the same type.
432  */
433 int
434 nd6_options(union nd_opts *ndopts)
435 {
436 	struct nd_opt_hdr *nd_opt;
437 	int i = 0;
438 
439 	KASSERT(ndopts != NULL, ("%s: ndopts == NULL", __func__));
440 	KASSERT(ndopts->nd_opts_last != NULL, ("%s: uninitialized ndopts",
441 	    __func__));
442 	if (ndopts->nd_opts_search == NULL)
443 		return 0;
444 
445 	while (1) {
446 		nd_opt = nd6_option(ndopts);
447 		if (nd_opt == NULL && ndopts->nd_opts_last == NULL) {
448 			/*
449 			 * Message validation requires that all included
450 			 * options have a length that is greater than zero.
451 			 */
452 			ICMP6STAT_INC(icp6s_nd_badopt);
453 			bzero(ndopts, sizeof(*ndopts));
454 			return -1;
455 		}
456 
457 		if (nd_opt == NULL)
458 			goto skip1;
459 
460 		switch (nd_opt->nd_opt_type) {
461 		case ND_OPT_SOURCE_LINKADDR:
462 		case ND_OPT_TARGET_LINKADDR:
463 		case ND_OPT_MTU:
464 		case ND_OPT_REDIRECTED_HEADER:
465 		case ND_OPT_NONCE:
466 			if (ndopts->nd_opt_array[nd_opt->nd_opt_type]) {
467 				nd6log((LOG_INFO,
468 				    "duplicated ND6 option found (type=%d)\n",
469 				    nd_opt->nd_opt_type));
470 				/* XXX bark? */
471 			} else {
472 				ndopts->nd_opt_array[nd_opt->nd_opt_type]
473 					= nd_opt;
474 			}
475 			break;
476 		case ND_OPT_PREFIX_INFORMATION:
477 			if (ndopts->nd_opt_array[nd_opt->nd_opt_type] == 0) {
478 				ndopts->nd_opt_array[nd_opt->nd_opt_type]
479 					= nd_opt;
480 			}
481 			ndopts->nd_opts_pi_end =
482 				(struct nd_opt_prefix_info *)nd_opt;
483 			break;
484 		/* What about ND_OPT_ROUTE_INFO? RFC 4191 */
485 		case ND_OPT_RDNSS:	/* RFC 6106 */
486 		case ND_OPT_DNSSL:	/* RFC 6106 */
487 			/*
488 			 * Silently ignore options we know and do not care about
489 			 * in the kernel.
490 			 */
491 			break;
492 		default:
493 			/*
494 			 * Unknown options must be silently ignored,
495 			 * to accommodate future extension to the protocol.
496 			 */
497 			nd6log((LOG_DEBUG,
498 			    "nd6_options: unsupported option %d - "
499 			    "option ignored\n", nd_opt->nd_opt_type));
500 		}
501 
502 skip1:
503 		i++;
504 		if (i > V_nd6_maxndopt) {
505 			ICMP6STAT_INC(icp6s_nd_toomanyopt);
506 			nd6log((LOG_INFO, "too many loop in nd opt\n"));
507 			break;
508 		}
509 
510 		if (ndopts->nd_opts_done)
511 			break;
512 	}
513 
514 	return 0;
515 }
516 
517 /*
518  * ND6 timer routine to handle ND6 entries
519  */
520 static void
521 nd6_llinfo_settimer_locked(struct llentry *ln, long tick)
522 {
523 	int canceled;
524 
525 	LLE_WLOCK_ASSERT(ln);
526 
527 	if (tick < 0) {
528 		ln->la_expire = 0;
529 		ln->ln_ntick = 0;
530 		canceled = callout_stop(&ln->lle_timer);
531 	} else {
532 		ln->la_expire = time_uptime + tick / hz;
533 		LLE_ADDREF(ln);
534 		if (tick > INT_MAX) {
535 			ln->ln_ntick = tick - INT_MAX;
536 			canceled = callout_reset(&ln->lle_timer, INT_MAX,
537 			    nd6_llinfo_timer, ln);
538 		} else {
539 			ln->ln_ntick = 0;
540 			canceled = callout_reset(&ln->lle_timer, tick,
541 			    nd6_llinfo_timer, ln);
542 		}
543 	}
544 	if (canceled > 0)
545 		LLE_REMREF(ln);
546 }
547 
548 /*
549  * Gets source address of the first packet in hold queue
550  * and stores it in @src.
551  * Returns pointer to @src (if hold queue is not empty) or NULL.
552  *
553  * Set noinline to be dtrace-friendly
554  */
555 static __noinline struct in6_addr *
556 nd6_llinfo_get_holdsrc(struct llentry *ln, struct in6_addr *src)
557 {
558 	struct ip6_hdr hdr;
559 	struct mbuf *m;
560 
561 	if (ln->la_hold == NULL)
562 		return (NULL);
563 
564 	/*
565 	 * assume every packet in la_hold has the same IP header
566 	 */
567 	m = ln->la_hold;
568 	if (sizeof(hdr) > m->m_len)
569 		return (NULL);
570 
571 	m_copydata(m, 0, sizeof(hdr), (caddr_t)&hdr);
572 	*src = hdr.ip6_src;
573 
574 	return (src);
575 }
576 
577 /*
578  * Checks if we need to switch from STALE state.
579  *
580  * RFC 4861 requires switching from STALE to DELAY state
581  * on first packet matching entry, waiting V_nd6_delay and
582  * transition to PROBE state (if upper layer confirmation was
583  * not received).
584  *
585  * This code performs a bit differently:
586  * On packet hit we don't change state (but desired state
587  * can be guessed by control plane). However, after V_nd6_delay
588  * seconds code will transition to PROBE state (so DELAY state
589  * is kinda skipped in most situations).
590  *
591  * Typically, V_nd6_gctimer is bigger than V_nd6_delay, so
592  * we perform the following upon entering STALE state:
593  *
594  * 1) Arm timer to run each V_nd6_delay seconds to make sure that
595  * if packet was transmitted at the start of given interval, we
596  * would be able to switch to PROBE state in V_nd6_delay seconds
597  * as user expects.
598  *
599  * 2) Reschedule timer until original V_nd6_gctimer expires keeping
600  * lle in STALE state (remaining timer value stored in lle_remtime).
601  *
602  * 3) Reschedule timer if packet was transmitted less that V_nd6_delay
603  * seconds ago.
604  *
605  * Returns non-zero value if the entry is still STALE (storing
606  * the next timer interval in @pdelay).
607  *
608  * Returns zero value if original timer expired or we need to switch to
609  * PROBE (store that in @do_switch variable).
610  */
611 static int
612 nd6_is_stale(struct llentry *lle, long *pdelay, int *do_switch)
613 {
614 	int nd_delay, nd_gctimer, r_skip_req;
615 	time_t lle_hittime;
616 	long delay;
617 
618 	*do_switch = 0;
619 	nd_gctimer = V_nd6_gctimer;
620 	nd_delay = V_nd6_delay;
621 
622 	LLE_REQ_LOCK(lle);
623 	r_skip_req = lle->r_skip_req;
624 	lle_hittime = lle->lle_hittime;
625 	LLE_REQ_UNLOCK(lle);
626 
627 	if (r_skip_req > 0) {
628 
629 		/*
630 		 * Nonzero r_skip_req value was set upon entering
631 		 * STALE state. Since value was not changed, no
632 		 * packets were passed using this lle. Ask for
633 		 * timer reschedule and keep STALE state.
634 		 */
635 		delay = (long)(MIN(nd_gctimer, nd_delay));
636 		delay *= hz;
637 		if (lle->lle_remtime > delay)
638 			lle->lle_remtime -= delay;
639 		else {
640 			delay = lle->lle_remtime;
641 			lle->lle_remtime = 0;
642 		}
643 
644 		if (delay == 0) {
645 
646 			/*
647 			 * The original ng6_gctime timeout ended,
648 			 * no more rescheduling.
649 			 */
650 			return (0);
651 		}
652 
653 		*pdelay = delay;
654 		return (1);
655 	}
656 
657 	/*
658 	 * Packet received. Verify timestamp
659 	 */
660 	delay = (long)(time_uptime - lle_hittime);
661 	if (delay < nd_delay) {
662 
663 		/*
664 		 * V_nd6_delay still not passed since the first
665 		 * hit in STALE state.
666 		 * Reshedule timer and return.
667 		 */
668 		*pdelay = (long)(nd_delay - delay) * hz;
669 		return (1);
670 	}
671 
672 	/* Request switching to probe */
673 	*do_switch = 1;
674 	return (0);
675 }
676 
677 
678 /*
679  * Switch @lle state to new state optionally arming timers.
680  *
681  * Set noinline to be dtrace-friendly
682  */
683 __noinline void
684 nd6_llinfo_setstate(struct llentry *lle, int newstate)
685 {
686 	struct ifnet *ifp;
687 	int nd_gctimer, nd_delay;
688 	long delay, remtime;
689 
690 	delay = 0;
691 	remtime = 0;
692 
693 	switch (newstate) {
694 	case ND6_LLINFO_INCOMPLETE:
695 		ifp = lle->lle_tbl->llt_ifp;
696 		delay = (long)ND_IFINFO(ifp)->retrans * hz / 1000;
697 		break;
698 	case ND6_LLINFO_REACHABLE:
699 		if (!ND6_LLINFO_PERMANENT(lle)) {
700 			ifp = lle->lle_tbl->llt_ifp;
701 			delay = (long)ND_IFINFO(ifp)->reachable * hz;
702 		}
703 		break;
704 	case ND6_LLINFO_STALE:
705 
706 		/*
707 		 * Notify fast path that we want to know if any packet
708 		 * is transmitted by setting r_skip_req.
709 		 */
710 		LLE_REQ_LOCK(lle);
711 		lle->r_skip_req = 1;
712 		LLE_REQ_UNLOCK(lle);
713 		nd_delay = V_nd6_delay;
714 		nd_gctimer = V_nd6_gctimer;
715 
716 		delay = (long)(MIN(nd_gctimer, nd_delay)) * hz;
717 		remtime = (long)nd_gctimer * hz - delay;
718 		break;
719 	case ND6_LLINFO_DELAY:
720 		lle->la_asked = 0;
721 		delay = (long)V_nd6_delay * hz;
722 		break;
723 	}
724 
725 	if (delay > 0)
726 		nd6_llinfo_settimer_locked(lle, delay);
727 
728 	lle->lle_remtime = remtime;
729 	lle->ln_state = newstate;
730 }
731 
732 /*
733  * Timer-dependent part of nd state machine.
734  *
735  * Set noinline to be dtrace-friendly
736  */
737 static __noinline void
738 nd6_llinfo_timer(void *arg)
739 {
740 	struct llentry *ln;
741 	struct in6_addr *dst, *pdst, *psrc, src;
742 	struct ifnet *ifp;
743 	struct nd_ifinfo *ndi;
744 	int do_switch, send_ns;
745 	long delay;
746 
747 	KASSERT(arg != NULL, ("%s: arg NULL", __func__));
748 	ln = (struct llentry *)arg;
749 	ifp = lltable_get_ifp(ln->lle_tbl);
750 	CURVNET_SET(ifp->if_vnet);
751 
752 	ND6_RLOCK();
753 	LLE_WLOCK(ln);
754 	if (callout_pending(&ln->lle_timer)) {
755 		/*
756 		 * Here we are a bit odd here in the treatment of
757 		 * active/pending. If the pending bit is set, it got
758 		 * rescheduled before I ran. The active
759 		 * bit we ignore, since if it was stopped
760 		 * in ll_tablefree() and was currently running
761 		 * it would have return 0 so the code would
762 		 * not have deleted it since the callout could
763 		 * not be stopped so we want to go through
764 		 * with the delete here now. If the callout
765 		 * was restarted, the pending bit will be back on and
766 		 * we just want to bail since the callout_reset would
767 		 * return 1 and our reference would have been removed
768 		 * by nd6_llinfo_settimer_locked above since canceled
769 		 * would have been 1.
770 		 */
771 		LLE_WUNLOCK(ln);
772 		ND6_RUNLOCK();
773 		CURVNET_RESTORE();
774 		return;
775 	}
776 	ndi = ND_IFINFO(ifp);
777 	send_ns = 0;
778 	dst = &ln->r_l3addr.addr6;
779 	pdst = dst;
780 
781 	if (ln->ln_ntick > 0) {
782 		if (ln->ln_ntick > INT_MAX) {
783 			ln->ln_ntick -= INT_MAX;
784 			nd6_llinfo_settimer_locked(ln, INT_MAX);
785 		} else {
786 			ln->ln_ntick = 0;
787 			nd6_llinfo_settimer_locked(ln, ln->ln_ntick);
788 		}
789 		goto done;
790 	}
791 
792 	if (ln->la_flags & LLE_STATIC) {
793 		goto done;
794 	}
795 
796 	if (ln->la_flags & LLE_DELETED) {
797 		nd6_free(&ln, 0);
798 		goto done;
799 	}
800 
801 	switch (ln->ln_state) {
802 	case ND6_LLINFO_INCOMPLETE:
803 		if (ln->la_asked < V_nd6_mmaxtries) {
804 			ln->la_asked++;
805 			send_ns = 1;
806 			/* Send NS to multicast address */
807 			pdst = NULL;
808 		} else {
809 			struct mbuf *m = ln->la_hold;
810 			if (m) {
811 				struct mbuf *m0;
812 
813 				/*
814 				 * assuming every packet in la_hold has the
815 				 * same IP header.  Send error after unlock.
816 				 */
817 				m0 = m->m_nextpkt;
818 				m->m_nextpkt = NULL;
819 				ln->la_hold = m0;
820 				clear_llinfo_pqueue(ln);
821 			}
822 			nd6_free(&ln, 0);
823 			if (m != NULL)
824 				icmp6_error2(m, ICMP6_DST_UNREACH,
825 				    ICMP6_DST_UNREACH_ADDR, 0, ifp);
826 		}
827 		break;
828 	case ND6_LLINFO_REACHABLE:
829 		if (!ND6_LLINFO_PERMANENT(ln))
830 			nd6_llinfo_setstate(ln, ND6_LLINFO_STALE);
831 		break;
832 
833 	case ND6_LLINFO_STALE:
834 		if (nd6_is_stale(ln, &delay, &do_switch) != 0) {
835 
836 			/*
837 			 * No packet has used this entry and GC timeout
838 			 * has not been passed. Reshedule timer and
839 			 * return.
840 			 */
841 			nd6_llinfo_settimer_locked(ln, delay);
842 			break;
843 		}
844 
845 		if (do_switch == 0) {
846 
847 			/*
848 			 * GC timer has ended and entry hasn't been used.
849 			 * Run Garbage collector (RFC 4861, 5.3)
850 			 */
851 			if (!ND6_LLINFO_PERMANENT(ln))
852 				nd6_free(&ln, 1);
853 			break;
854 		}
855 
856 		/* Entry has been used AND delay timer has ended. */
857 
858 		/* FALLTHROUGH */
859 
860 	case ND6_LLINFO_DELAY:
861 		if (ndi && (ndi->flags & ND6_IFF_PERFORMNUD) != 0) {
862 			/* We need NUD */
863 			ln->la_asked = 1;
864 			nd6_llinfo_setstate(ln, ND6_LLINFO_PROBE);
865 			send_ns = 1;
866 		} else
867 			nd6_llinfo_setstate(ln, ND6_LLINFO_STALE); /* XXX */
868 		break;
869 	case ND6_LLINFO_PROBE:
870 		if (ln->la_asked < V_nd6_umaxtries) {
871 			ln->la_asked++;
872 			send_ns = 1;
873 		} else {
874 			nd6_free(&ln, 0);
875 		}
876 		break;
877 	default:
878 		panic("%s: paths in a dark night can be confusing: %d",
879 		    __func__, ln->ln_state);
880 	}
881 done:
882 	if (ln != NULL)
883 		ND6_RUNLOCK();
884 	if (send_ns != 0) {
885 		nd6_llinfo_settimer_locked(ln, (long)ndi->retrans * hz / 1000);
886 		psrc = nd6_llinfo_get_holdsrc(ln, &src);
887 		LLE_FREE_LOCKED(ln);
888 		ln = NULL;
889 		nd6_ns_output(ifp, psrc, pdst, dst, NULL);
890 	}
891 
892 	if (ln != NULL)
893 		LLE_FREE_LOCKED(ln);
894 	CURVNET_RESTORE();
895 }
896 
897 
898 /*
899  * ND6 timer routine to expire default route list and prefix list
900  */
901 void
902 nd6_timer(void *arg)
903 {
904 	CURVNET_SET((struct vnet *) arg);
905 	struct nd_drhead drq;
906 	struct nd_defrouter *dr, *ndr;
907 	struct nd_prefix *pr, *npr;
908 	struct in6_ifaddr *ia6, *nia6;
909 
910 	TAILQ_INIT(&drq);
911 
912 	/* expire default router list */
913 	ND6_WLOCK();
914 	TAILQ_FOREACH_SAFE(dr, &V_nd_defrouter, dr_entry, ndr)
915 		if (dr->expire && dr->expire < time_uptime)
916 			defrouter_unlink(dr, &drq);
917 	ND6_WUNLOCK();
918 
919 	while ((dr = TAILQ_FIRST(&drq)) != NULL) {
920 		TAILQ_REMOVE(&drq, dr, dr_entry);
921 		defrouter_del(dr);
922 	}
923 
924 	/*
925 	 * expire interface addresses.
926 	 * in the past the loop was inside prefix expiry processing.
927 	 * However, from a stricter speci-confrmance standpoint, we should
928 	 * rather separate address lifetimes and prefix lifetimes.
929 	 *
930 	 * XXXRW: in6_ifaddrhead locking.
931 	 */
932   addrloop:
933 	TAILQ_FOREACH_SAFE(ia6, &V_in6_ifaddrhead, ia_link, nia6) {
934 		/* check address lifetime */
935 		if (IFA6_IS_INVALID(ia6)) {
936 			int regen = 0;
937 
938 			/*
939 			 * If the expiring address is temporary, try
940 			 * regenerating a new one.  This would be useful when
941 			 * we suspended a laptop PC, then turned it on after a
942 			 * period that could invalidate all temporary
943 			 * addresses.  Although we may have to restart the
944 			 * loop (see below), it must be after purging the
945 			 * address.  Otherwise, we'd see an infinite loop of
946 			 * regeneration.
947 			 */
948 			if (V_ip6_use_tempaddr &&
949 			    (ia6->ia6_flags & IN6_IFF_TEMPORARY) != 0) {
950 				if (regen_tmpaddr(ia6) == 0)
951 					regen = 1;
952 			}
953 
954 			in6_purgeaddr(&ia6->ia_ifa);
955 
956 			if (regen)
957 				goto addrloop; /* XXX: see below */
958 		} else if (IFA6_IS_DEPRECATED(ia6)) {
959 			int oldflags = ia6->ia6_flags;
960 
961 			ia6->ia6_flags |= IN6_IFF_DEPRECATED;
962 
963 			/*
964 			 * If a temporary address has just become deprecated,
965 			 * regenerate a new one if possible.
966 			 */
967 			if (V_ip6_use_tempaddr &&
968 			    (ia6->ia6_flags & IN6_IFF_TEMPORARY) != 0 &&
969 			    (oldflags & IN6_IFF_DEPRECATED) == 0) {
970 
971 				if (regen_tmpaddr(ia6) == 0) {
972 					/*
973 					 * A new temporary address is
974 					 * generated.
975 					 * XXX: this means the address chain
976 					 * has changed while we are still in
977 					 * the loop.  Although the change
978 					 * would not cause disaster (because
979 					 * it's not a deletion, but an
980 					 * addition,) we'd rather restart the
981 					 * loop just for safety.  Or does this
982 					 * significantly reduce performance??
983 					 */
984 					goto addrloop;
985 				}
986 			}
987 		} else if ((ia6->ia6_flags & IN6_IFF_TENTATIVE) != 0) {
988 			/*
989 			 * Schedule DAD for a tentative address.  This happens
990 			 * if the interface was down or not running
991 			 * when the address was configured.
992 			 */
993 			int delay;
994 
995 			delay = arc4random() %
996 			    (MAX_RTR_SOLICITATION_DELAY * hz);
997 			nd6_dad_start((struct ifaddr *)ia6, delay);
998 		} else {
999 			/*
1000 			 * Check status of the interface.  If it is down,
1001 			 * mark the address as tentative for future DAD.
1002 			 */
1003 			if ((ia6->ia_ifp->if_flags & IFF_UP) == 0 ||
1004 			    (ia6->ia_ifp->if_drv_flags & IFF_DRV_RUNNING)
1005 				== 0 ||
1006 			    (ND_IFINFO(ia6->ia_ifp)->flags &
1007 				ND6_IFF_IFDISABLED) != 0) {
1008 				ia6->ia6_flags &= ~IN6_IFF_DUPLICATED;
1009 				ia6->ia6_flags |= IN6_IFF_TENTATIVE;
1010 			}
1011 			/*
1012 			 * A new RA might have made a deprecated address
1013 			 * preferred.
1014 			 */
1015 			ia6->ia6_flags &= ~IN6_IFF_DEPRECATED;
1016 		}
1017 	}
1018 
1019 	/* expire prefix list */
1020 	LIST_FOREACH_SAFE(pr, &V_nd_prefix, ndpr_entry, npr) {
1021 		/*
1022 		 * check prefix lifetime.
1023 		 * since pltime is just for autoconf, pltime processing for
1024 		 * prefix is not necessary.
1025 		 */
1026 		if (pr->ndpr_vltime != ND6_INFINITE_LIFETIME &&
1027 		    time_uptime - pr->ndpr_lastupdate > pr->ndpr_vltime) {
1028 
1029 			/*
1030 			 * address expiration and prefix expiration are
1031 			 * separate.  NEVER perform in6_purgeaddr here.
1032 			 */
1033 			prelist_remove(pr);
1034 		}
1035 	}
1036 
1037 	callout_reset(&V_nd6_timer_ch, V_nd6_prune * hz,
1038 	    nd6_timer, curvnet);
1039 
1040 	CURVNET_RESTORE();
1041 }
1042 
1043 /*
1044  * ia6 - deprecated/invalidated temporary address
1045  */
1046 static int
1047 regen_tmpaddr(struct in6_ifaddr *ia6)
1048 {
1049 	struct ifaddr *ifa;
1050 	struct ifnet *ifp;
1051 	struct in6_ifaddr *public_ifa6 = NULL;
1052 
1053 	ifp = ia6->ia_ifa.ifa_ifp;
1054 	IF_ADDR_RLOCK(ifp);
1055 	TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1056 		struct in6_ifaddr *it6;
1057 
1058 		if (ifa->ifa_addr->sa_family != AF_INET6)
1059 			continue;
1060 
1061 		it6 = (struct in6_ifaddr *)ifa;
1062 
1063 		/* ignore no autoconf addresses. */
1064 		if ((it6->ia6_flags & IN6_IFF_AUTOCONF) == 0)
1065 			continue;
1066 
1067 		/* ignore autoconf addresses with different prefixes. */
1068 		if (it6->ia6_ndpr == NULL || it6->ia6_ndpr != ia6->ia6_ndpr)
1069 			continue;
1070 
1071 		/*
1072 		 * Now we are looking at an autoconf address with the same
1073 		 * prefix as ours.  If the address is temporary and is still
1074 		 * preferred, do not create another one.  It would be rare, but
1075 		 * could happen, for example, when we resume a laptop PC after
1076 		 * a long period.
1077 		 */
1078 		if ((it6->ia6_flags & IN6_IFF_TEMPORARY) != 0 &&
1079 		    !IFA6_IS_DEPRECATED(it6)) {
1080 			public_ifa6 = NULL;
1081 			break;
1082 		}
1083 
1084 		/*
1085 		 * This is a public autoconf address that has the same prefix
1086 		 * as ours.  If it is preferred, keep it.  We can't break the
1087 		 * loop here, because there may be a still-preferred temporary
1088 		 * address with the prefix.
1089 		 */
1090 		if (!IFA6_IS_DEPRECATED(it6))
1091 			public_ifa6 = it6;
1092 	}
1093 	if (public_ifa6 != NULL)
1094 		ifa_ref(&public_ifa6->ia_ifa);
1095 	IF_ADDR_RUNLOCK(ifp);
1096 
1097 	if (public_ifa6 != NULL) {
1098 		int e;
1099 
1100 		if ((e = in6_tmpifadd(public_ifa6, 0, 0)) != 0) {
1101 			ifa_free(&public_ifa6->ia_ifa);
1102 			log(LOG_NOTICE, "regen_tmpaddr: failed to create a new"
1103 			    " tmp addr,errno=%d\n", e);
1104 			return (-1);
1105 		}
1106 		ifa_free(&public_ifa6->ia_ifa);
1107 		return (0);
1108 	}
1109 
1110 	return (-1);
1111 }
1112 
1113 /*
1114  * Remove prefix and default router list entries corresponding to ifp. Neighbor
1115  * cache entries are freed in in6_domifdetach().
1116  */
1117 void
1118 nd6_purge(struct ifnet *ifp)
1119 {
1120 	struct nd_drhead drq;
1121 	struct nd_defrouter *dr, *ndr;
1122 	struct nd_prefix *pr, *npr;
1123 
1124 	TAILQ_INIT(&drq);
1125 
1126 	/*
1127 	 * Nuke default router list entries toward ifp.
1128 	 * We defer removal of default router list entries that is installed
1129 	 * in the routing table, in order to keep additional side effects as
1130 	 * small as possible.
1131 	 */
1132 	ND6_WLOCK();
1133 	TAILQ_FOREACH_SAFE(dr, &V_nd_defrouter, dr_entry, ndr) {
1134 		if (dr->installed)
1135 			continue;
1136 		if (dr->ifp == ifp)
1137 			defrouter_unlink(dr, &drq);
1138 	}
1139 
1140 	TAILQ_FOREACH_SAFE(dr, &V_nd_defrouter, dr_entry, ndr) {
1141 		if (!dr->installed)
1142 			continue;
1143 		if (dr->ifp == ifp)
1144 			defrouter_unlink(dr, &drq);
1145 	}
1146 	ND6_WUNLOCK();
1147 
1148 	while ((dr = TAILQ_FIRST(&drq)) != NULL) {
1149 		TAILQ_REMOVE(&drq, dr, dr_entry);
1150 		defrouter_del(dr);
1151 	}
1152 
1153 	/* Nuke prefix list entries toward ifp */
1154 	LIST_FOREACH_SAFE(pr, &V_nd_prefix, ndpr_entry, npr) {
1155 		if (pr->ndpr_ifp == ifp) {
1156 			/*
1157 			 * Because if_detach() does *not* release prefixes
1158 			 * while purging addresses the reference count will
1159 			 * still be above zero. We therefore reset it to
1160 			 * make sure that the prefix really gets purged.
1161 			 */
1162 			pr->ndpr_refcnt = 0;
1163 
1164 			prelist_remove(pr);
1165 		}
1166 	}
1167 
1168 	/* cancel default outgoing interface setting */
1169 	if (V_nd6_defifindex == ifp->if_index)
1170 		nd6_setdefaultiface(0);
1171 
1172 	if (ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) {
1173 		/* Refresh default router list. */
1174 		defrouter_select();
1175 	}
1176 }
1177 
1178 /*
1179  * the caller acquires and releases the lock on the lltbls
1180  * Returns the llentry locked
1181  */
1182 struct llentry *
1183 nd6_lookup(const struct in6_addr *addr6, int flags, struct ifnet *ifp)
1184 {
1185 	struct sockaddr_in6 sin6;
1186 	struct llentry *ln;
1187 
1188 	bzero(&sin6, sizeof(sin6));
1189 	sin6.sin6_len = sizeof(struct sockaddr_in6);
1190 	sin6.sin6_family = AF_INET6;
1191 	sin6.sin6_addr = *addr6;
1192 
1193 	IF_AFDATA_LOCK_ASSERT(ifp);
1194 
1195 	ln = lla_lookup(LLTABLE6(ifp), flags, (struct sockaddr *)&sin6);
1196 
1197 	return (ln);
1198 }
1199 
1200 struct llentry *
1201 nd6_alloc(const struct in6_addr *addr6, int flags, struct ifnet *ifp)
1202 {
1203 	struct sockaddr_in6 sin6;
1204 	struct llentry *ln;
1205 
1206 	bzero(&sin6, sizeof(sin6));
1207 	sin6.sin6_len = sizeof(struct sockaddr_in6);
1208 	sin6.sin6_family = AF_INET6;
1209 	sin6.sin6_addr = *addr6;
1210 
1211 	ln = lltable_alloc_entry(LLTABLE6(ifp), 0, (struct sockaddr *)&sin6);
1212 	if (ln != NULL)
1213 		ln->ln_state = ND6_LLINFO_NOSTATE;
1214 
1215 	return (ln);
1216 }
1217 
1218 /*
1219  * Test whether a given IPv6 address is a neighbor or not, ignoring
1220  * the actual neighbor cache.  The neighbor cache is ignored in order
1221  * to not reenter the routing code from within itself.
1222  */
1223 static int
1224 nd6_is_new_addr_neighbor(const struct sockaddr_in6 *addr, struct ifnet *ifp)
1225 {
1226 	struct nd_prefix *pr;
1227 	struct ifaddr *dstaddr;
1228 	struct rt_addrinfo info;
1229 	struct sockaddr_in6 rt_key;
1230 	struct sockaddr *dst6;
1231 	int fibnum;
1232 
1233 	/*
1234 	 * A link-local address is always a neighbor.
1235 	 * XXX: a link does not necessarily specify a single interface.
1236 	 */
1237 	if (IN6_IS_ADDR_LINKLOCAL(&addr->sin6_addr)) {
1238 		struct sockaddr_in6 sin6_copy;
1239 		u_int32_t zone;
1240 
1241 		/*
1242 		 * We need sin6_copy since sa6_recoverscope() may modify the
1243 		 * content (XXX).
1244 		 */
1245 		sin6_copy = *addr;
1246 		if (sa6_recoverscope(&sin6_copy))
1247 			return (0); /* XXX: should be impossible */
1248 		if (in6_setscope(&sin6_copy.sin6_addr, ifp, &zone))
1249 			return (0);
1250 		if (sin6_copy.sin6_scope_id == zone)
1251 			return (1);
1252 		else
1253 			return (0);
1254 	}
1255 
1256 	bzero(&rt_key, sizeof(rt_key));
1257 	bzero(&info, sizeof(info));
1258 	info.rti_info[RTAX_DST] = (struct sockaddr *)&rt_key;
1259 
1260 	/* Always use the default FIB here. XXME - why? */
1261 	fibnum = RT_DEFAULT_FIB;
1262 
1263 	/*
1264 	 * If the address matches one of our addresses,
1265 	 * it should be a neighbor.
1266 	 * If the address matches one of our on-link prefixes, it should be a
1267 	 * neighbor.
1268 	 */
1269 	LIST_FOREACH(pr, &V_nd_prefix, ndpr_entry) {
1270 		if (pr->ndpr_ifp != ifp)
1271 			continue;
1272 
1273 		if (!(pr->ndpr_stateflags & NDPRF_ONLINK)) {
1274 
1275 			/* Always use the default FIB here. */
1276 			dst6 = (struct sockaddr *)&pr->ndpr_prefix;
1277 
1278 			/* Restore length field before retrying lookup */
1279 			rt_key.sin6_len = sizeof(rt_key);
1280 			if (rib_lookup_info(fibnum, dst6, 0, 0, &info) != 0)
1281 				continue;
1282 			/*
1283 			 * This is the case where multiple interfaces
1284 			 * have the same prefix, but only one is installed
1285 			 * into the routing table and that prefix entry
1286 			 * is not the one being examined here. In the case
1287 			 * where RADIX_MPATH is enabled, multiple route
1288 			 * entries (of the same rt_key value) will be
1289 			 * installed because the interface addresses all
1290 			 * differ.
1291 			 */
1292 			if (!IN6_ARE_ADDR_EQUAL(&pr->ndpr_prefix.sin6_addr,
1293 			       &rt_key.sin6_addr))
1294 				continue;
1295 		}
1296 
1297 		if (IN6_ARE_MASKED_ADDR_EQUAL(&pr->ndpr_prefix.sin6_addr,
1298 		    &addr->sin6_addr, &pr->ndpr_mask))
1299 			return (1);
1300 	}
1301 
1302 	/*
1303 	 * If the address is assigned on the node of the other side of
1304 	 * a p2p interface, the address should be a neighbor.
1305 	 */
1306 	dstaddr = ifa_ifwithdstaddr((const struct sockaddr *)addr, RT_ALL_FIBS);
1307 	if (dstaddr != NULL) {
1308 		if (dstaddr->ifa_ifp == ifp) {
1309 			ifa_free(dstaddr);
1310 			return (1);
1311 		}
1312 		ifa_free(dstaddr);
1313 	}
1314 
1315 	/*
1316 	 * If the default router list is empty, all addresses are regarded
1317 	 * as on-link, and thus, as a neighbor.
1318 	 */
1319 	if (ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV &&
1320 	    TAILQ_EMPTY(&V_nd_defrouter) &&
1321 	    V_nd6_defifindex == ifp->if_index) {
1322 		return (1);
1323 	}
1324 
1325 	return (0);
1326 }
1327 
1328 
1329 /*
1330  * Detect if a given IPv6 address identifies a neighbor on a given link.
1331  * XXX: should take care of the destination of a p2p link?
1332  */
1333 int
1334 nd6_is_addr_neighbor(const struct sockaddr_in6 *addr, struct ifnet *ifp)
1335 {
1336 	struct llentry *lle;
1337 	int rc = 0;
1338 
1339 	IF_AFDATA_UNLOCK_ASSERT(ifp);
1340 	if (nd6_is_new_addr_neighbor(addr, ifp))
1341 		return (1);
1342 
1343 	/*
1344 	 * Even if the address matches none of our addresses, it might be
1345 	 * in the neighbor cache.
1346 	 */
1347 	IF_AFDATA_RLOCK(ifp);
1348 	if ((lle = nd6_lookup(&addr->sin6_addr, 0, ifp)) != NULL) {
1349 		LLE_RUNLOCK(lle);
1350 		rc = 1;
1351 	}
1352 	IF_AFDATA_RUNLOCK(ifp);
1353 	return (rc);
1354 }
1355 
1356 /*
1357  * Free an nd6 llinfo entry.
1358  * Since the function would cause significant changes in the kernel, DO NOT
1359  * make it global, unless you have a strong reason for the change, and are sure
1360  * that the change is safe.
1361  *
1362  * Set noinline to be dtrace-friendly
1363  */
1364 static __noinline void
1365 nd6_free(struct llentry **lnp, int gc)
1366 {
1367 	struct ifnet *ifp;
1368 	struct llentry *ln;
1369 	struct nd_defrouter *dr;
1370 
1371 	ln = *lnp;
1372 	*lnp = NULL;
1373 
1374 	LLE_WLOCK_ASSERT(ln);
1375 	ND6_RLOCK_ASSERT();
1376 
1377 	ifp = lltable_get_ifp(ln->lle_tbl);
1378 	if ((ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) != 0)
1379 		dr = defrouter_lookup_locked(&ln->r_l3addr.addr6, ifp);
1380 	else
1381 		dr = NULL;
1382 	ND6_RUNLOCK();
1383 
1384 	if ((ln->la_flags & LLE_DELETED) == 0)
1385 		EVENTHANDLER_INVOKE(lle_event, ln, LLENTRY_EXPIRED);
1386 
1387 	/*
1388 	 * we used to have pfctlinput(PRC_HOSTDEAD) here.
1389 	 * even though it is not harmful, it was not really necessary.
1390 	 */
1391 
1392 	/* cancel timer */
1393 	nd6_llinfo_settimer_locked(ln, -1);
1394 
1395 	if (ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) {
1396 		if (dr != NULL && dr->expire &&
1397 		    ln->ln_state == ND6_LLINFO_STALE && gc) {
1398 			/*
1399 			 * If the reason for the deletion is just garbage
1400 			 * collection, and the neighbor is an active default
1401 			 * router, do not delete it.  Instead, reset the GC
1402 			 * timer using the router's lifetime.
1403 			 * Simply deleting the entry would affect default
1404 			 * router selection, which is not necessarily a good
1405 			 * thing, especially when we're using router preference
1406 			 * values.
1407 			 * XXX: the check for ln_state would be redundant,
1408 			 *      but we intentionally keep it just in case.
1409 			 */
1410 			if (dr->expire > time_uptime)
1411 				nd6_llinfo_settimer_locked(ln,
1412 				    (dr->expire - time_uptime) * hz);
1413 			else
1414 				nd6_llinfo_settimer_locked(ln,
1415 				    (long)V_nd6_gctimer * hz);
1416 
1417 			LLE_REMREF(ln);
1418 			LLE_WUNLOCK(ln);
1419 			defrouter_rele(dr);
1420 			return;
1421 		}
1422 
1423 		if (dr) {
1424 			/*
1425 			 * Unreachablity of a router might affect the default
1426 			 * router selection and on-link detection of advertised
1427 			 * prefixes.
1428 			 */
1429 
1430 			/*
1431 			 * Temporarily fake the state to choose a new default
1432 			 * router and to perform on-link determination of
1433 			 * prefixes correctly.
1434 			 * Below the state will be set correctly,
1435 			 * or the entry itself will be deleted.
1436 			 */
1437 			ln->ln_state = ND6_LLINFO_INCOMPLETE;
1438 		}
1439 
1440 		if (ln->ln_router || dr) {
1441 
1442 			/*
1443 			 * We need to unlock to avoid a LOR with rt6_flush() with the
1444 			 * rnh and for the calls to pfxlist_onlink_check() and
1445 			 * defrouter_select() in the block further down for calls
1446 			 * into nd6_lookup().  We still hold a ref.
1447 			 */
1448 			LLE_WUNLOCK(ln);
1449 
1450 			/*
1451 			 * rt6_flush must be called whether or not the neighbor
1452 			 * is in the Default Router List.
1453 			 * See a corresponding comment in nd6_na_input().
1454 			 */
1455 			rt6_flush(&ln->r_l3addr.addr6, ifp);
1456 		}
1457 
1458 		if (dr) {
1459 			/*
1460 			 * Since defrouter_select() does not affect the
1461 			 * on-link determination and MIP6 needs the check
1462 			 * before the default router selection, we perform
1463 			 * the check now.
1464 			 */
1465 			pfxlist_onlink_check();
1466 
1467 			/*
1468 			 * Refresh default router list.
1469 			 */
1470 			defrouter_select();
1471 		}
1472 
1473 		/*
1474 		 * If this entry was added by an on-link redirect, remove the
1475 		 * corresponding host route.
1476 		 */
1477 		if (ln->la_flags & LLE_REDIRECT)
1478 			nd6_free_redirect(ln);
1479 
1480 		if (ln->ln_router || dr)
1481 			LLE_WLOCK(ln);
1482 	}
1483 
1484 	/*
1485 	 * Save to unlock. We still hold an extra reference and will not
1486 	 * free(9) in llentry_free() if someone else holds one as well.
1487 	 */
1488 	LLE_WUNLOCK(ln);
1489 	IF_AFDATA_LOCK(ifp);
1490 	LLE_WLOCK(ln);
1491 	/* Guard against race with other llentry_free(). */
1492 	if (ln->la_flags & LLE_LINKED) {
1493 		/* Remove callout reference */
1494 		LLE_REMREF(ln);
1495 		lltable_unlink_entry(ln->lle_tbl, ln);
1496 	}
1497 	IF_AFDATA_UNLOCK(ifp);
1498 
1499 	llentry_free(ln);
1500 	if (dr != NULL)
1501 		defrouter_rele(dr);
1502 }
1503 
1504 static int
1505 nd6_isdynrte(const struct rtentry *rt, void *xap)
1506 {
1507 
1508 	if (rt->rt_flags == (RTF_UP | RTF_HOST | RTF_DYNAMIC))
1509 		return (1);
1510 
1511 	return (0);
1512 }
1513 /*
1514  * Remove the rtentry for the given llentry,
1515  * both of which were installed by a redirect.
1516  */
1517 static void
1518 nd6_free_redirect(const struct llentry *ln)
1519 {
1520 	int fibnum;
1521 	struct sockaddr_in6 sin6;
1522 	struct rt_addrinfo info;
1523 
1524 	lltable_fill_sa_entry(ln, (struct sockaddr *)&sin6);
1525 	memset(&info, 0, sizeof(info));
1526 	info.rti_info[RTAX_DST] = (struct sockaddr *)&sin6;
1527 	info.rti_filter = nd6_isdynrte;
1528 
1529 	for (fibnum = 0; fibnum < rt_numfibs; fibnum++)
1530 		rtrequest1_fib(RTM_DELETE, &info, NULL, fibnum);
1531 }
1532 
1533 /*
1534  * Rejuvenate this function for routing operations related
1535  * processing.
1536  */
1537 void
1538 nd6_rtrequest(int req, struct rtentry *rt, struct rt_addrinfo *info)
1539 {
1540 	struct sockaddr_in6 *gateway;
1541 	struct nd_defrouter *dr;
1542 	struct ifnet *ifp;
1543 
1544 	gateway = (struct sockaddr_in6 *)rt->rt_gateway;
1545 	ifp = rt->rt_ifp;
1546 
1547 	switch (req) {
1548 	case RTM_ADD:
1549 		break;
1550 
1551 	case RTM_DELETE:
1552 		if (!ifp)
1553 			return;
1554 		/*
1555 		 * Only indirect routes are interesting.
1556 		 */
1557 		if ((rt->rt_flags & RTF_GATEWAY) == 0)
1558 			return;
1559 		/*
1560 		 * check for default route
1561 		 */
1562 		if (IN6_ARE_ADDR_EQUAL(&in6addr_any,
1563 		    &SIN6(rt_key(rt))->sin6_addr)) {
1564 			dr = defrouter_lookup(&gateway->sin6_addr, ifp);
1565 			if (dr != NULL) {
1566 				dr->installed = 0;
1567 				defrouter_rele(dr);
1568 			}
1569 		}
1570 		break;
1571 	}
1572 }
1573 
1574 
1575 int
1576 nd6_ioctl(u_long cmd, caddr_t data, struct ifnet *ifp)
1577 {
1578 	struct in6_ndireq *ndi = (struct in6_ndireq *)data;
1579 	struct in6_nbrinfo *nbi = (struct in6_nbrinfo *)data;
1580 	struct in6_ndifreq *ndif = (struct in6_ndifreq *)data;
1581 	int error = 0;
1582 
1583 	if (ifp->if_afdata[AF_INET6] == NULL)
1584 		return (EPFNOSUPPORT);
1585 	switch (cmd) {
1586 	case OSIOCGIFINFO_IN6:
1587 #define ND	ndi->ndi
1588 		/* XXX: old ndp(8) assumes a positive value for linkmtu. */
1589 		bzero(&ND, sizeof(ND));
1590 		ND.linkmtu = IN6_LINKMTU(ifp);
1591 		ND.maxmtu = ND_IFINFO(ifp)->maxmtu;
1592 		ND.basereachable = ND_IFINFO(ifp)->basereachable;
1593 		ND.reachable = ND_IFINFO(ifp)->reachable;
1594 		ND.retrans = ND_IFINFO(ifp)->retrans;
1595 		ND.flags = ND_IFINFO(ifp)->flags;
1596 		ND.recalctm = ND_IFINFO(ifp)->recalctm;
1597 		ND.chlim = ND_IFINFO(ifp)->chlim;
1598 		break;
1599 	case SIOCGIFINFO_IN6:
1600 		ND = *ND_IFINFO(ifp);
1601 		break;
1602 	case SIOCSIFINFO_IN6:
1603 		/*
1604 		 * used to change host variables from userland.
1605 		 * intended for a use on router to reflect RA configurations.
1606 		 */
1607 		/* 0 means 'unspecified' */
1608 		if (ND.linkmtu != 0) {
1609 			if (ND.linkmtu < IPV6_MMTU ||
1610 			    ND.linkmtu > IN6_LINKMTU(ifp)) {
1611 				error = EINVAL;
1612 				break;
1613 			}
1614 			ND_IFINFO(ifp)->linkmtu = ND.linkmtu;
1615 		}
1616 
1617 		if (ND.basereachable != 0) {
1618 			int obasereachable = ND_IFINFO(ifp)->basereachable;
1619 
1620 			ND_IFINFO(ifp)->basereachable = ND.basereachable;
1621 			if (ND.basereachable != obasereachable)
1622 				ND_IFINFO(ifp)->reachable =
1623 				    ND_COMPUTE_RTIME(ND.basereachable);
1624 		}
1625 		if (ND.retrans != 0)
1626 			ND_IFINFO(ifp)->retrans = ND.retrans;
1627 		if (ND.chlim != 0)
1628 			ND_IFINFO(ifp)->chlim = ND.chlim;
1629 		/* FALLTHROUGH */
1630 	case SIOCSIFINFO_FLAGS:
1631 	{
1632 		struct ifaddr *ifa;
1633 		struct in6_ifaddr *ia;
1634 
1635 		if ((ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED) &&
1636 		    !(ND.flags & ND6_IFF_IFDISABLED)) {
1637 			/* ifdisabled 1->0 transision */
1638 
1639 			/*
1640 			 * If the interface is marked as ND6_IFF_IFDISABLED and
1641 			 * has an link-local address with IN6_IFF_DUPLICATED,
1642 			 * do not clear ND6_IFF_IFDISABLED.
1643 			 * See RFC 4862, Section 5.4.5.
1644 			 */
1645 			IF_ADDR_RLOCK(ifp);
1646 			TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1647 				if (ifa->ifa_addr->sa_family != AF_INET6)
1648 					continue;
1649 				ia = (struct in6_ifaddr *)ifa;
1650 				if ((ia->ia6_flags & IN6_IFF_DUPLICATED) &&
1651 				    IN6_IS_ADDR_LINKLOCAL(IA6_IN6(ia)))
1652 					break;
1653 			}
1654 			IF_ADDR_RUNLOCK(ifp);
1655 
1656 			if (ifa != NULL) {
1657 				/* LLA is duplicated. */
1658 				ND.flags |= ND6_IFF_IFDISABLED;
1659 				log(LOG_ERR, "Cannot enable an interface"
1660 				    " with a link-local address marked"
1661 				    " duplicate.\n");
1662 			} else {
1663 				ND_IFINFO(ifp)->flags &= ~ND6_IFF_IFDISABLED;
1664 				if (ifp->if_flags & IFF_UP)
1665 					in6_if_up(ifp);
1666 			}
1667 		} else if (!(ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED) &&
1668 			    (ND.flags & ND6_IFF_IFDISABLED)) {
1669 			/* ifdisabled 0->1 transision */
1670 			/* Mark all IPv6 address as tentative. */
1671 
1672 			ND_IFINFO(ifp)->flags |= ND6_IFF_IFDISABLED;
1673 			if (V_ip6_dad_count > 0 &&
1674 			    (ND_IFINFO(ifp)->flags & ND6_IFF_NO_DAD) == 0) {
1675 				IF_ADDR_RLOCK(ifp);
1676 				TAILQ_FOREACH(ifa, &ifp->if_addrhead,
1677 				    ifa_link) {
1678 					if (ifa->ifa_addr->sa_family !=
1679 					    AF_INET6)
1680 						continue;
1681 					ia = (struct in6_ifaddr *)ifa;
1682 					ia->ia6_flags |= IN6_IFF_TENTATIVE;
1683 				}
1684 				IF_ADDR_RUNLOCK(ifp);
1685 			}
1686 		}
1687 
1688 		if (ND.flags & ND6_IFF_AUTO_LINKLOCAL) {
1689 			if (!(ND_IFINFO(ifp)->flags & ND6_IFF_AUTO_LINKLOCAL)) {
1690 				/* auto_linklocal 0->1 transision */
1691 
1692 				/* If no link-local address on ifp, configure */
1693 				ND_IFINFO(ifp)->flags |= ND6_IFF_AUTO_LINKLOCAL;
1694 				in6_ifattach(ifp, NULL);
1695 			} else if (!(ND.flags & ND6_IFF_IFDISABLED) &&
1696 			    ifp->if_flags & IFF_UP) {
1697 				/*
1698 				 * When the IF already has
1699 				 * ND6_IFF_AUTO_LINKLOCAL, no link-local
1700 				 * address is assigned, and IFF_UP, try to
1701 				 * assign one.
1702 				 */
1703 				IF_ADDR_RLOCK(ifp);
1704 				TAILQ_FOREACH(ifa, &ifp->if_addrhead,
1705 				    ifa_link) {
1706 					if (ifa->ifa_addr->sa_family !=
1707 					    AF_INET6)
1708 						continue;
1709 					ia = (struct in6_ifaddr *)ifa;
1710 					if (IN6_IS_ADDR_LINKLOCAL(IA6_IN6(ia)))
1711 						break;
1712 				}
1713 				IF_ADDR_RUNLOCK(ifp);
1714 				if (ifa != NULL)
1715 					/* No LLA is configured. */
1716 					in6_ifattach(ifp, NULL);
1717 			}
1718 		}
1719 	}
1720 		ND_IFINFO(ifp)->flags = ND.flags;
1721 		break;
1722 #undef ND
1723 	case SIOCSNDFLUSH_IN6:	/* XXX: the ioctl name is confusing... */
1724 		/* sync kernel routing table with the default router list */
1725 		defrouter_reset();
1726 		defrouter_select();
1727 		break;
1728 	case SIOCSPFXFLUSH_IN6:
1729 	{
1730 		/* flush all the prefix advertised by routers */
1731 		struct nd_prefix *pr, *next;
1732 
1733 		LIST_FOREACH_SAFE(pr, &V_nd_prefix, ndpr_entry, next) {
1734 			struct in6_ifaddr *ia, *ia_next;
1735 
1736 			if (IN6_IS_ADDR_LINKLOCAL(&pr->ndpr_prefix.sin6_addr))
1737 				continue; /* XXX */
1738 
1739 			/* do we really have to remove addresses as well? */
1740 			/* XXXRW: in6_ifaddrhead locking. */
1741 			TAILQ_FOREACH_SAFE(ia, &V_in6_ifaddrhead, ia_link,
1742 			    ia_next) {
1743 				if ((ia->ia6_flags & IN6_IFF_AUTOCONF) == 0)
1744 					continue;
1745 
1746 				if (ia->ia6_ndpr == pr)
1747 					in6_purgeaddr(&ia->ia_ifa);
1748 			}
1749 			prelist_remove(pr);
1750 		}
1751 		break;
1752 	}
1753 	case SIOCSRTRFLUSH_IN6:
1754 	{
1755 		/* flush all the default routers */
1756 		struct nd_drhead drq;
1757 		struct nd_defrouter *dr;
1758 
1759 		TAILQ_INIT(&drq);
1760 
1761 		defrouter_reset();
1762 
1763 		ND6_WLOCK();
1764 		while ((dr = TAILQ_FIRST(&V_nd_defrouter)) != NULL)
1765 			defrouter_unlink(dr, &drq);
1766 		ND6_WUNLOCK();
1767 		while ((dr = TAILQ_FIRST(&drq)) != NULL) {
1768 			TAILQ_REMOVE(&drq, dr, dr_entry);
1769 			defrouter_del(dr);
1770 		}
1771 
1772 		defrouter_select();
1773 		break;
1774 	}
1775 	case SIOCGNBRINFO_IN6:
1776 	{
1777 		struct llentry *ln;
1778 		struct in6_addr nb_addr = nbi->addr; /* make local for safety */
1779 
1780 		if ((error = in6_setscope(&nb_addr, ifp, NULL)) != 0)
1781 			return (error);
1782 
1783 		IF_AFDATA_RLOCK(ifp);
1784 		ln = nd6_lookup(&nb_addr, 0, ifp);
1785 		IF_AFDATA_RUNLOCK(ifp);
1786 
1787 		if (ln == NULL) {
1788 			error = EINVAL;
1789 			break;
1790 		}
1791 		nbi->state = ln->ln_state;
1792 		nbi->asked = ln->la_asked;
1793 		nbi->isrouter = ln->ln_router;
1794 		if (ln->la_expire == 0)
1795 			nbi->expire = 0;
1796 		else
1797 			nbi->expire = ln->la_expire + ln->lle_remtime / hz +
1798 			    (time_second - time_uptime);
1799 		LLE_RUNLOCK(ln);
1800 		break;
1801 	}
1802 	case SIOCGDEFIFACE_IN6:	/* XXX: should be implemented as a sysctl? */
1803 		ndif->ifindex = V_nd6_defifindex;
1804 		break;
1805 	case SIOCSDEFIFACE_IN6:	/* XXX: should be implemented as a sysctl? */
1806 		return (nd6_setdefaultiface(ndif->ifindex));
1807 	}
1808 	return (error);
1809 }
1810 
1811 /*
1812  * Calculates new isRouter value based on provided parameters and
1813  * returns it.
1814  */
1815 static int
1816 nd6_is_router(int type, int code, int is_new, int old_addr, int new_addr,
1817     int ln_router)
1818 {
1819 
1820 	/*
1821 	 * ICMP6 type dependent behavior.
1822 	 *
1823 	 * NS: clear IsRouter if new entry
1824 	 * RS: clear IsRouter
1825 	 * RA: set IsRouter if there's lladdr
1826 	 * redir: clear IsRouter if new entry
1827 	 *
1828 	 * RA case, (1):
1829 	 * The spec says that we must set IsRouter in the following cases:
1830 	 * - If lladdr exist, set IsRouter.  This means (1-5).
1831 	 * - If it is old entry (!newentry), set IsRouter.  This means (7).
1832 	 * So, based on the spec, in (1-5) and (7) cases we must set IsRouter.
1833 	 * A quetion arises for (1) case.  (1) case has no lladdr in the
1834 	 * neighbor cache, this is similar to (6).
1835 	 * This case is rare but we figured that we MUST NOT set IsRouter.
1836 	 *
1837 	 *   is_new  old_addr new_addr 	    NS  RS  RA	redir
1838 	 *							D R
1839 	 *	0	n	n	(1)	c   ?     s
1840 	 *	0	y	n	(2)	c   s     s
1841 	 *	0	n	y	(3)	c   s     s
1842 	 *	0	y	y	(4)	c   s     s
1843 	 *	0	y	y	(5)	c   s     s
1844 	 *	1	--	n	(6) c	c	c s
1845 	 *	1	--	y	(7) c	c   s	c s
1846 	 *
1847 	 *					(c=clear s=set)
1848 	 */
1849 	switch (type & 0xff) {
1850 	case ND_NEIGHBOR_SOLICIT:
1851 		/*
1852 		 * New entry must have is_router flag cleared.
1853 		 */
1854 		if (is_new)					/* (6-7) */
1855 			ln_router = 0;
1856 		break;
1857 	case ND_REDIRECT:
1858 		/*
1859 		 * If the icmp is a redirect to a better router, always set the
1860 		 * is_router flag.  Otherwise, if the entry is newly created,
1861 		 * clear the flag.  [RFC 2461, sec 8.3]
1862 		 */
1863 		if (code == ND_REDIRECT_ROUTER)
1864 			ln_router = 1;
1865 		else {
1866 			if (is_new)				/* (6-7) */
1867 				ln_router = 0;
1868 		}
1869 		break;
1870 	case ND_ROUTER_SOLICIT:
1871 		/*
1872 		 * is_router flag must always be cleared.
1873 		 */
1874 		ln_router = 0;
1875 		break;
1876 	case ND_ROUTER_ADVERT:
1877 		/*
1878 		 * Mark an entry with lladdr as a router.
1879 		 */
1880 		if ((!is_new && (old_addr || new_addr)) ||	/* (2-5) */
1881 		    (is_new && new_addr)) {			/* (7) */
1882 			ln_router = 1;
1883 		}
1884 		break;
1885 	}
1886 
1887 	return (ln_router);
1888 }
1889 
1890 /*
1891  * Create neighbor cache entry and cache link-layer address,
1892  * on reception of inbound ND6 packets.  (RS/RA/NS/redirect)
1893  *
1894  * type - ICMP6 type
1895  * code - type dependent information
1896  *
1897  */
1898 void
1899 nd6_cache_lladdr(struct ifnet *ifp, struct in6_addr *from, char *lladdr,
1900     int lladdrlen, int type, int code)
1901 {
1902 	struct llentry *ln = NULL, *ln_tmp;
1903 	int is_newentry;
1904 	int do_update;
1905 	int olladdr;
1906 	int llchange;
1907 	int flags;
1908 	uint16_t router = 0;
1909 	struct sockaddr_in6 sin6;
1910 	struct mbuf *chain = NULL;
1911 	u_char linkhdr[LLE_MAX_LINKHDR];
1912 	size_t linkhdrsize;
1913 	int lladdr_off;
1914 
1915 	IF_AFDATA_UNLOCK_ASSERT(ifp);
1916 
1917 	KASSERT(ifp != NULL, ("%s: ifp == NULL", __func__));
1918 	KASSERT(from != NULL, ("%s: from == NULL", __func__));
1919 
1920 	/* nothing must be updated for unspecified address */
1921 	if (IN6_IS_ADDR_UNSPECIFIED(from))
1922 		return;
1923 
1924 	/*
1925 	 * Validation about ifp->if_addrlen and lladdrlen must be done in
1926 	 * the caller.
1927 	 *
1928 	 * XXX If the link does not have link-layer adderss, what should
1929 	 * we do? (ifp->if_addrlen == 0)
1930 	 * Spec says nothing in sections for RA, RS and NA.  There's small
1931 	 * description on it in NS section (RFC 2461 7.2.3).
1932 	 */
1933 	flags = lladdr ? LLE_EXCLUSIVE : 0;
1934 	IF_AFDATA_RLOCK(ifp);
1935 	ln = nd6_lookup(from, flags, ifp);
1936 	IF_AFDATA_RUNLOCK(ifp);
1937 	is_newentry = 0;
1938 	if (ln == NULL) {
1939 		flags |= LLE_EXCLUSIVE;
1940 		ln = nd6_alloc(from, 0, ifp);
1941 		if (ln == NULL)
1942 			return;
1943 
1944 		/*
1945 		 * Since we already know all the data for the new entry,
1946 		 * fill it before insertion.
1947 		 */
1948 		if (lladdr != NULL) {
1949 			linkhdrsize = sizeof(linkhdr);
1950 			if (lltable_calc_llheader(ifp, AF_INET6, lladdr,
1951 			    linkhdr, &linkhdrsize, &lladdr_off) != 0)
1952 				return;
1953 			lltable_set_entry_addr(ifp, ln, linkhdr, linkhdrsize,
1954 			    lladdr_off);
1955 		}
1956 
1957 		IF_AFDATA_WLOCK(ifp);
1958 		LLE_WLOCK(ln);
1959 		/* Prefer any existing lle over newly-created one */
1960 		ln_tmp = nd6_lookup(from, LLE_EXCLUSIVE, ifp);
1961 		if (ln_tmp == NULL)
1962 			lltable_link_entry(LLTABLE6(ifp), ln);
1963 		IF_AFDATA_WUNLOCK(ifp);
1964 		if (ln_tmp == NULL) {
1965 			/* No existing lle, mark as new entry (6,7) */
1966 			is_newentry = 1;
1967 			nd6_llinfo_setstate(ln, ND6_LLINFO_STALE);
1968 			if (lladdr != NULL)	/* (7) */
1969 				EVENTHANDLER_INVOKE(lle_event, ln,
1970 				    LLENTRY_RESOLVED);
1971 		} else {
1972 			lltable_free_entry(LLTABLE6(ifp), ln);
1973 			ln = ln_tmp;
1974 			ln_tmp = NULL;
1975 		}
1976 	}
1977 	/* do nothing if static ndp is set */
1978 	if ((ln->la_flags & LLE_STATIC)) {
1979 		if (flags & LLE_EXCLUSIVE)
1980 			LLE_WUNLOCK(ln);
1981 		else
1982 			LLE_RUNLOCK(ln);
1983 		return;
1984 	}
1985 
1986 	olladdr = (ln->la_flags & LLE_VALID) ? 1 : 0;
1987 	if (olladdr && lladdr) {
1988 		llchange = bcmp(lladdr, ln->ll_addr,
1989 		    ifp->if_addrlen);
1990 	} else if (!olladdr && lladdr)
1991 		llchange = 1;
1992 	else
1993 		llchange = 0;
1994 
1995 	/*
1996 	 * newentry olladdr  lladdr  llchange	(*=record)
1997 	 *	0	n	n	--	(1)
1998 	 *	0	y	n	--	(2)
1999 	 *	0	n	y	y	(3) * STALE
2000 	 *	0	y	y	n	(4) *
2001 	 *	0	y	y	y	(5) * STALE
2002 	 *	1	--	n	--	(6)   NOSTATE(= PASSIVE)
2003 	 *	1	--	y	--	(7) * STALE
2004 	 */
2005 
2006 	do_update = 0;
2007 	if (is_newentry == 0 && llchange != 0) {
2008 		do_update = 1;	/* (3,5) */
2009 
2010 		/*
2011 		 * Record source link-layer address
2012 		 * XXX is it dependent to ifp->if_type?
2013 		 */
2014 		linkhdrsize = sizeof(linkhdr);
2015 		if (lltable_calc_llheader(ifp, AF_INET6, lladdr,
2016 		    linkhdr, &linkhdrsize, &lladdr_off) != 0)
2017 			return;
2018 
2019 		if (lltable_try_set_entry_addr(ifp, ln, linkhdr, linkhdrsize,
2020 		    lladdr_off) == 0) {
2021 			/* Entry was deleted */
2022 			return;
2023 		}
2024 
2025 		nd6_llinfo_setstate(ln, ND6_LLINFO_STALE);
2026 
2027 		EVENTHANDLER_INVOKE(lle_event, ln, LLENTRY_RESOLVED);
2028 
2029 		if (ln->la_hold != NULL)
2030 			nd6_grab_holdchain(ln, &chain, &sin6);
2031 	}
2032 
2033 	/* Calculates new router status */
2034 	router = nd6_is_router(type, code, is_newentry, olladdr,
2035 	    lladdr != NULL ? 1 : 0, ln->ln_router);
2036 
2037 	ln->ln_router = router;
2038 	/* Mark non-router redirects with special flag */
2039 	if ((type & 0xFF) == ND_REDIRECT && code != ND_REDIRECT_ROUTER)
2040 		ln->la_flags |= LLE_REDIRECT;
2041 
2042 	if (flags & LLE_EXCLUSIVE)
2043 		LLE_WUNLOCK(ln);
2044 	else
2045 		LLE_RUNLOCK(ln);
2046 
2047 	if (chain != NULL)
2048 		nd6_flush_holdchain(ifp, ifp, chain, &sin6);
2049 
2050 	/*
2051 	 * When the link-layer address of a router changes, select the
2052 	 * best router again.  In particular, when the neighbor entry is newly
2053 	 * created, it might affect the selection policy.
2054 	 * Question: can we restrict the first condition to the "is_newentry"
2055 	 * case?
2056 	 * XXX: when we hear an RA from a new router with the link-layer
2057 	 * address option, defrouter_select() is called twice, since
2058 	 * defrtrlist_update called the function as well.  However, I believe
2059 	 * we can compromise the overhead, since it only happens the first
2060 	 * time.
2061 	 * XXX: although defrouter_select() should not have a bad effect
2062 	 * for those are not autoconfigured hosts, we explicitly avoid such
2063 	 * cases for safety.
2064 	 */
2065 	if ((do_update || is_newentry) && router &&
2066 	    ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) {
2067 		/*
2068 		 * guaranteed recursion
2069 		 */
2070 		defrouter_select();
2071 	}
2072 }
2073 
2074 static void
2075 nd6_slowtimo(void *arg)
2076 {
2077 	CURVNET_SET((struct vnet *) arg);
2078 	struct nd_ifinfo *nd6if;
2079 	struct ifnet *ifp;
2080 
2081 	callout_reset(&V_nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
2082 	    nd6_slowtimo, curvnet);
2083 	IFNET_RLOCK_NOSLEEP();
2084 	TAILQ_FOREACH(ifp, &V_ifnet, if_link) {
2085 		if (ifp->if_afdata[AF_INET6] == NULL)
2086 			continue;
2087 		nd6if = ND_IFINFO(ifp);
2088 		if (nd6if->basereachable && /* already initialized */
2089 		    (nd6if->recalctm -= ND6_SLOWTIMER_INTERVAL) <= 0) {
2090 			/*
2091 			 * Since reachable time rarely changes by router
2092 			 * advertisements, we SHOULD insure that a new random
2093 			 * value gets recomputed at least once every few hours.
2094 			 * (RFC 2461, 6.3.4)
2095 			 */
2096 			nd6if->recalctm = V_nd6_recalc_reachtm_interval;
2097 			nd6if->reachable = ND_COMPUTE_RTIME(nd6if->basereachable);
2098 		}
2099 	}
2100 	IFNET_RUNLOCK_NOSLEEP();
2101 	CURVNET_RESTORE();
2102 }
2103 
2104 void
2105 nd6_grab_holdchain(struct llentry *ln, struct mbuf **chain,
2106     struct sockaddr_in6 *sin6)
2107 {
2108 
2109 	LLE_WLOCK_ASSERT(ln);
2110 
2111 	*chain = ln->la_hold;
2112 	ln->la_hold = NULL;
2113 	lltable_fill_sa_entry(ln, (struct sockaddr *)sin6);
2114 
2115 	if (ln->ln_state == ND6_LLINFO_STALE) {
2116 
2117 		/*
2118 		 * The first time we send a packet to a
2119 		 * neighbor whose entry is STALE, we have
2120 		 * to change the state to DELAY and a sets
2121 		 * a timer to expire in DELAY_FIRST_PROBE_TIME
2122 		 * seconds to ensure do neighbor unreachability
2123 		 * detection on expiration.
2124 		 * (RFC 2461 7.3.3)
2125 		 */
2126 		nd6_llinfo_setstate(ln, ND6_LLINFO_DELAY);
2127 	}
2128 }
2129 
2130 int
2131 nd6_output_ifp(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *m,
2132     struct sockaddr_in6 *dst, struct route *ro)
2133 {
2134 	int error;
2135 	int ip6len;
2136 	struct ip6_hdr *ip6;
2137 	struct m_tag *mtag;
2138 
2139 #ifdef MAC
2140 	mac_netinet6_nd6_send(ifp, m);
2141 #endif
2142 
2143 	/*
2144 	 * If called from nd6_ns_output() (NS), nd6_na_output() (NA),
2145 	 * icmp6_redirect_output() (REDIRECT) or from rip6_output() (RS, RA
2146 	 * as handled by rtsol and rtadvd), mbufs will be tagged for SeND
2147 	 * to be diverted to user space.  When re-injected into the kernel,
2148 	 * send_output() will directly dispatch them to the outgoing interface.
2149 	 */
2150 	if (send_sendso_input_hook != NULL) {
2151 		mtag = m_tag_find(m, PACKET_TAG_ND_OUTGOING, NULL);
2152 		if (mtag != NULL) {
2153 			ip6 = mtod(m, struct ip6_hdr *);
2154 			ip6len = sizeof(struct ip6_hdr) + ntohs(ip6->ip6_plen);
2155 			/* Use the SEND socket */
2156 			error = send_sendso_input_hook(m, ifp, SND_OUT,
2157 			    ip6len);
2158 			/* -1 == no app on SEND socket */
2159 			if (error == 0 || error != -1)
2160 			    return (error);
2161 		}
2162 	}
2163 
2164 	m_clrprotoflags(m);	/* Avoid confusing lower layers. */
2165 	IP_PROBE(send, NULL, NULL, mtod(m, struct ip6_hdr *), ifp, NULL,
2166 	    mtod(m, struct ip6_hdr *));
2167 
2168 	if ((ifp->if_flags & IFF_LOOPBACK) == 0)
2169 		origifp = ifp;
2170 
2171 	error = (*ifp->if_output)(origifp, m, (struct sockaddr *)dst, ro);
2172 	return (error);
2173 }
2174 
2175 /*
2176  * Lookup link headerfor @sa_dst address. Stores found
2177  * data in @desten buffer. Copy of lle ln_flags can be also
2178  * saved in @pflags if @pflags is non-NULL.
2179  *
2180  * If destination LLE does not exists or lle state modification
2181  * is required, call "slow" version.
2182  *
2183  * Return values:
2184  * - 0 on success (address copied to buffer).
2185  * - EWOULDBLOCK (no local error, but address is still unresolved)
2186  * - other errors (alloc failure, etc)
2187  */
2188 int
2189 nd6_resolve(struct ifnet *ifp, int is_gw, struct mbuf *m,
2190     const struct sockaddr *sa_dst, u_char *desten, uint32_t *pflags,
2191     struct llentry **plle)
2192 {
2193 	struct llentry *ln = NULL;
2194 	const struct sockaddr_in6 *dst6;
2195 
2196 	if (pflags != NULL)
2197 		*pflags = 0;
2198 
2199 	dst6 = (const struct sockaddr_in6 *)sa_dst;
2200 
2201 	/* discard the packet if IPv6 operation is disabled on the interface */
2202 	if ((ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED)) {
2203 		m_freem(m);
2204 		return (ENETDOWN); /* better error? */
2205 	}
2206 
2207 	if (m != NULL && m->m_flags & M_MCAST) {
2208 		switch (ifp->if_type) {
2209 		case IFT_ETHER:
2210 		case IFT_FDDI:
2211 		case IFT_L2VLAN:
2212 		case IFT_IEEE80211:
2213 		case IFT_BRIDGE:
2214 		case IFT_ISO88025:
2215 			ETHER_MAP_IPV6_MULTICAST(&dst6->sin6_addr,
2216 						 desten);
2217 			return (0);
2218 		default:
2219 			m_freem(m);
2220 			return (EAFNOSUPPORT);
2221 		}
2222 	}
2223 
2224 	IF_AFDATA_RLOCK(ifp);
2225 	ln = nd6_lookup(&dst6->sin6_addr, plle ? LLE_EXCLUSIVE : LLE_UNLOCKED,
2226 	    ifp);
2227 	if (ln != NULL && (ln->r_flags & RLLE_VALID) != 0) {
2228 		/* Entry found, let's copy lle info */
2229 		bcopy(ln->r_linkdata, desten, ln->r_hdrlen);
2230 		if (pflags != NULL)
2231 			*pflags = LLE_VALID | (ln->r_flags & RLLE_IFADDR);
2232 		/* Check if we have feedback request from nd6 timer */
2233 		if (ln->r_skip_req != 0) {
2234 			LLE_REQ_LOCK(ln);
2235 			ln->r_skip_req = 0; /* Notify that entry was used */
2236 			ln->lle_hittime = time_uptime;
2237 			LLE_REQ_UNLOCK(ln);
2238 		}
2239 		if (plle) {
2240 			LLE_ADDREF(ln);
2241 			*plle = ln;
2242 			LLE_WUNLOCK(ln);
2243 		}
2244 		IF_AFDATA_RUNLOCK(ifp);
2245 		return (0);
2246 	} else if (plle && ln)
2247 		LLE_WUNLOCK(ln);
2248 	IF_AFDATA_RUNLOCK(ifp);
2249 
2250 	return (nd6_resolve_slow(ifp, 0, m, dst6, desten, pflags, plle));
2251 }
2252 
2253 
2254 /*
2255  * Do L2 address resolution for @sa_dst address. Stores found
2256  * address in @desten buffer. Copy of lle ln_flags can be also
2257  * saved in @pflags if @pflags is non-NULL.
2258  *
2259  * Heavy version.
2260  * Function assume that destination LLE does not exist,
2261  * is invalid or stale, so LLE_EXCLUSIVE lock needs to be acquired.
2262  *
2263  * Set noinline to be dtrace-friendly
2264  */
2265 static __noinline int
2266 nd6_resolve_slow(struct ifnet *ifp, int flags, struct mbuf *m,
2267     const struct sockaddr_in6 *dst, u_char *desten, uint32_t *pflags,
2268     struct llentry **plle)
2269 {
2270 	struct llentry *lle = NULL, *lle_tmp;
2271 	struct in6_addr *psrc, src;
2272 	int send_ns, ll_len;
2273 	char *lladdr;
2274 
2275 	/*
2276 	 * Address resolution or Neighbor Unreachability Detection
2277 	 * for the next hop.
2278 	 * At this point, the destination of the packet must be a unicast
2279 	 * or an anycast address(i.e. not a multicast).
2280 	 */
2281 	if (lle == NULL) {
2282 		IF_AFDATA_RLOCK(ifp);
2283 		lle = nd6_lookup(&dst->sin6_addr, LLE_EXCLUSIVE, ifp);
2284 		IF_AFDATA_RUNLOCK(ifp);
2285 		if ((lle == NULL) && nd6_is_addr_neighbor(dst, ifp))  {
2286 			/*
2287 			 * Since nd6_is_addr_neighbor() internally calls nd6_lookup(),
2288 			 * the condition below is not very efficient.  But we believe
2289 			 * it is tolerable, because this should be a rare case.
2290 			 */
2291 			lle = nd6_alloc(&dst->sin6_addr, 0, ifp);
2292 			if (lle == NULL) {
2293 				char ip6buf[INET6_ADDRSTRLEN];
2294 				log(LOG_DEBUG,
2295 				    "nd6_output: can't allocate llinfo for %s "
2296 				    "(ln=%p)\n",
2297 				    ip6_sprintf(ip6buf, &dst->sin6_addr), lle);
2298 				m_freem(m);
2299 				return (ENOBUFS);
2300 			}
2301 
2302 			IF_AFDATA_WLOCK(ifp);
2303 			LLE_WLOCK(lle);
2304 			/* Prefer any existing entry over newly-created one */
2305 			lle_tmp = nd6_lookup(&dst->sin6_addr, LLE_EXCLUSIVE, ifp);
2306 			if (lle_tmp == NULL)
2307 				lltable_link_entry(LLTABLE6(ifp), lle);
2308 			IF_AFDATA_WUNLOCK(ifp);
2309 			if (lle_tmp != NULL) {
2310 				lltable_free_entry(LLTABLE6(ifp), lle);
2311 				lle = lle_tmp;
2312 				lle_tmp = NULL;
2313 			}
2314 		}
2315 	}
2316 	if (lle == NULL) {
2317 		if (!(ND_IFINFO(ifp)->flags & ND6_IFF_PERFORMNUD)) {
2318 			m_freem(m);
2319 			return (ENOBUFS);
2320 		}
2321 
2322 		if (m != NULL)
2323 			m_freem(m);
2324 		return (ENOBUFS);
2325 	}
2326 
2327 	LLE_WLOCK_ASSERT(lle);
2328 
2329 	/*
2330 	 * The first time we send a packet to a neighbor whose entry is
2331 	 * STALE, we have to change the state to DELAY and a sets a timer to
2332 	 * expire in DELAY_FIRST_PROBE_TIME seconds to ensure do
2333 	 * neighbor unreachability detection on expiration.
2334 	 * (RFC 2461 7.3.3)
2335 	 */
2336 	if (lle->ln_state == ND6_LLINFO_STALE)
2337 		nd6_llinfo_setstate(lle, ND6_LLINFO_DELAY);
2338 
2339 	/*
2340 	 * If the neighbor cache entry has a state other than INCOMPLETE
2341 	 * (i.e. its link-layer address is already resolved), just
2342 	 * send the packet.
2343 	 */
2344 	if (lle->ln_state > ND6_LLINFO_INCOMPLETE) {
2345 		if (flags & LLE_ADDRONLY) {
2346 			lladdr = lle->ll_addr;
2347 			ll_len = ifp->if_addrlen;
2348 		} else {
2349 			lladdr = lle->r_linkdata;
2350 			ll_len = lle->r_hdrlen;
2351 		}
2352 		bcopy(lladdr, desten, ll_len);
2353 		if (pflags != NULL)
2354 			*pflags = lle->la_flags;
2355 		if (plle) {
2356 			LLE_ADDREF(lle);
2357 			*plle = lle;
2358 		}
2359 		LLE_WUNLOCK(lle);
2360 		return (0);
2361 	}
2362 
2363 	/*
2364 	 * There is a neighbor cache entry, but no ethernet address
2365 	 * response yet.  Append this latest packet to the end of the
2366 	 * packet queue in the mbuf.  When it exceeds nd6_maxqueuelen,
2367 	 * the oldest packet in the queue will be removed.
2368 	 */
2369 
2370 	if (lle->la_hold != NULL) {
2371 		struct mbuf *m_hold;
2372 		int i;
2373 
2374 		i = 0;
2375 		for (m_hold = lle->la_hold; m_hold; m_hold = m_hold->m_nextpkt){
2376 			i++;
2377 			if (m_hold->m_nextpkt == NULL) {
2378 				m_hold->m_nextpkt = m;
2379 				break;
2380 			}
2381 		}
2382 		while (i >= V_nd6_maxqueuelen) {
2383 			m_hold = lle->la_hold;
2384 			lle->la_hold = lle->la_hold->m_nextpkt;
2385 			m_freem(m_hold);
2386 			i--;
2387 		}
2388 	} else {
2389 		lle->la_hold = m;
2390 	}
2391 
2392 	/*
2393 	 * If there has been no NS for the neighbor after entering the
2394 	 * INCOMPLETE state, send the first solicitation.
2395 	 * Note that for newly-created lle la_asked will be 0,
2396 	 * so we will transition from ND6_LLINFO_NOSTATE to
2397 	 * ND6_LLINFO_INCOMPLETE state here.
2398 	 */
2399 	psrc = NULL;
2400 	send_ns = 0;
2401 	if (lle->la_asked == 0) {
2402 		lle->la_asked++;
2403 		send_ns = 1;
2404 		psrc = nd6_llinfo_get_holdsrc(lle, &src);
2405 
2406 		nd6_llinfo_setstate(lle, ND6_LLINFO_INCOMPLETE);
2407 	}
2408 	LLE_WUNLOCK(lle);
2409 	if (send_ns != 0)
2410 		nd6_ns_output(ifp, psrc, NULL, &dst->sin6_addr, NULL);
2411 
2412 	return (EWOULDBLOCK);
2413 }
2414 
2415 /*
2416  * Do L2 address resolution for @sa_dst address. Stores found
2417  * address in @desten buffer. Copy of lle ln_flags can be also
2418  * saved in @pflags if @pflags is non-NULL.
2419  *
2420  * Return values:
2421  * - 0 on success (address copied to buffer).
2422  * - EWOULDBLOCK (no local error, but address is still unresolved)
2423  * - other errors (alloc failure, etc)
2424  */
2425 int
2426 nd6_resolve_addr(struct ifnet *ifp, int flags, const struct sockaddr *dst,
2427     char *desten, uint32_t *pflags)
2428 {
2429 	int error;
2430 
2431 	flags |= LLE_ADDRONLY;
2432 	error = nd6_resolve_slow(ifp, flags, NULL,
2433 	    (const struct sockaddr_in6 *)dst, desten, pflags, NULL);
2434 	return (error);
2435 }
2436 
2437 int
2438 nd6_flush_holdchain(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *chain,
2439     struct sockaddr_in6 *dst)
2440 {
2441 	struct mbuf *m, *m_head;
2442 	struct ifnet *outifp;
2443 	int error = 0;
2444 
2445 	m_head = chain;
2446 	if ((ifp->if_flags & IFF_LOOPBACK) != 0)
2447 		outifp = origifp;
2448 	else
2449 		outifp = ifp;
2450 
2451 	while (m_head) {
2452 		m = m_head;
2453 		m_head = m_head->m_nextpkt;
2454 		error = nd6_output_ifp(ifp, origifp, m, dst, NULL);
2455 	}
2456 
2457 	/*
2458 	 * XXX
2459 	 * note that intermediate errors are blindly ignored
2460 	 */
2461 	return (error);
2462 }
2463 
2464 static int
2465 nd6_need_cache(struct ifnet *ifp)
2466 {
2467 	/*
2468 	 * XXX: we currently do not make neighbor cache on any interface
2469 	 * other than ARCnet, Ethernet, FDDI and GIF.
2470 	 *
2471 	 * RFC2893 says:
2472 	 * - unidirectional tunnels needs no ND
2473 	 */
2474 	switch (ifp->if_type) {
2475 	case IFT_ARCNET:
2476 	case IFT_ETHER:
2477 	case IFT_FDDI:
2478 	case IFT_IEEE1394:
2479 	case IFT_L2VLAN:
2480 	case IFT_IEEE80211:
2481 	case IFT_INFINIBAND:
2482 	case IFT_BRIDGE:
2483 	case IFT_PROPVIRTUAL:
2484 		return (1);
2485 	default:
2486 		return (0);
2487 	}
2488 }
2489 
2490 /*
2491  * Add pernament ND6 link-layer record for given
2492  * interface address.
2493  *
2494  * Very similar to IPv4 arp_ifinit(), but:
2495  * 1) IPv6 DAD is performed in different place
2496  * 2) It is called by IPv6 protocol stack in contrast to
2497  * arp_ifinit() which is typically called in SIOCSIFADDR
2498  * driver ioctl handler.
2499  *
2500  */
2501 int
2502 nd6_add_ifa_lle(struct in6_ifaddr *ia)
2503 {
2504 	struct ifnet *ifp;
2505 	struct llentry *ln, *ln_tmp;
2506 	struct sockaddr *dst;
2507 
2508 	ifp = ia->ia_ifa.ifa_ifp;
2509 	if (nd6_need_cache(ifp) == 0)
2510 		return (0);
2511 
2512 	ia->ia_ifa.ifa_rtrequest = nd6_rtrequest;
2513 	dst = (struct sockaddr *)&ia->ia_addr;
2514 	ln = lltable_alloc_entry(LLTABLE6(ifp), LLE_IFADDR, dst);
2515 	if (ln == NULL)
2516 		return (ENOBUFS);
2517 
2518 	IF_AFDATA_WLOCK(ifp);
2519 	LLE_WLOCK(ln);
2520 	/* Unlink any entry if exists */
2521 	ln_tmp = lla_lookup(LLTABLE6(ifp), LLE_EXCLUSIVE, dst);
2522 	if (ln_tmp != NULL)
2523 		lltable_unlink_entry(LLTABLE6(ifp), ln_tmp);
2524 	lltable_link_entry(LLTABLE6(ifp), ln);
2525 	IF_AFDATA_WUNLOCK(ifp);
2526 
2527 	if (ln_tmp != NULL)
2528 		EVENTHANDLER_INVOKE(lle_event, ln_tmp, LLENTRY_EXPIRED);
2529 	EVENTHANDLER_INVOKE(lle_event, ln, LLENTRY_RESOLVED);
2530 
2531 	LLE_WUNLOCK(ln);
2532 	if (ln_tmp != NULL)
2533 		llentry_free(ln_tmp);
2534 
2535 	return (0);
2536 }
2537 
2538 /*
2539  * Removes either all lle entries for given @ia, or lle
2540  * corresponding to @ia address.
2541  */
2542 void
2543 nd6_rem_ifa_lle(struct in6_ifaddr *ia, int all)
2544 {
2545 	struct sockaddr_in6 mask, addr;
2546 	struct sockaddr *saddr, *smask;
2547 	struct ifnet *ifp;
2548 
2549 	ifp = ia->ia_ifa.ifa_ifp;
2550 	memcpy(&addr, &ia->ia_addr, sizeof(ia->ia_addr));
2551 	memcpy(&mask, &ia->ia_prefixmask, sizeof(ia->ia_prefixmask));
2552 	saddr = (struct sockaddr *)&addr;
2553 	smask = (struct sockaddr *)&mask;
2554 
2555 	if (all != 0)
2556 		lltable_prefix_free(AF_INET6, saddr, smask, LLE_STATIC);
2557 	else
2558 		lltable_delete_addr(LLTABLE6(ifp), LLE_IFADDR, saddr);
2559 }
2560 
2561 static void
2562 clear_llinfo_pqueue(struct llentry *ln)
2563 {
2564 	struct mbuf *m_hold, *m_hold_next;
2565 
2566 	for (m_hold = ln->la_hold; m_hold; m_hold = m_hold_next) {
2567 		m_hold_next = m_hold->m_nextpkt;
2568 		m_freem(m_hold);
2569 	}
2570 
2571 	ln->la_hold = NULL;
2572 }
2573 
2574 static int nd6_sysctl_drlist(SYSCTL_HANDLER_ARGS);
2575 static int nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS);
2576 
2577 SYSCTL_DECL(_net_inet6_icmp6);
2578 SYSCTL_PROC(_net_inet6_icmp6, ICMPV6CTL_ND6_DRLIST, nd6_drlist,
2579 	CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_MPSAFE,
2580 	NULL, 0, nd6_sysctl_drlist, "S,in6_defrouter",
2581 	"NDP default router list");
2582 SYSCTL_PROC(_net_inet6_icmp6, ICMPV6CTL_ND6_PRLIST, nd6_prlist,
2583 	CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_MPSAFE,
2584 	NULL, 0, nd6_sysctl_prlist, "S,in6_prefix",
2585 	"NDP prefix list");
2586 SYSCTL_INT(_net_inet6_icmp6, ICMPV6CTL_ND6_MAXQLEN, nd6_maxqueuelen,
2587 	CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(nd6_maxqueuelen), 1, "");
2588 SYSCTL_INT(_net_inet6_icmp6, OID_AUTO, nd6_gctimer,
2589 	CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(nd6_gctimer), (60 * 60 * 24), "");
2590 
2591 static int
2592 nd6_sysctl_drlist(SYSCTL_HANDLER_ARGS)
2593 {
2594 	struct in6_defrouter d;
2595 	struct nd_defrouter *dr;
2596 	int error;
2597 
2598 	if (req->newptr != NULL)
2599 		return (EPERM);
2600 
2601 	error = sysctl_wire_old_buffer(req, 0);
2602 	if (error != 0)
2603 		return (error);
2604 
2605 	bzero(&d, sizeof(d));
2606 	d.rtaddr.sin6_family = AF_INET6;
2607 	d.rtaddr.sin6_len = sizeof(d.rtaddr);
2608 
2609 	ND6_RLOCK();
2610 	TAILQ_FOREACH(dr, &V_nd_defrouter, dr_entry) {
2611 		d.rtaddr.sin6_addr = dr->rtaddr;
2612 		error = sa6_recoverscope(&d.rtaddr);
2613 		if (error != 0)
2614 			break;
2615 		d.flags = dr->raflags;
2616 		d.rtlifetime = dr->rtlifetime;
2617 		d.expire = dr->expire + (time_second - time_uptime);
2618 		d.if_index = dr->ifp->if_index;
2619 		error = SYSCTL_OUT(req, &d, sizeof(d));
2620 		if (error != 0)
2621 			break;
2622 	}
2623 	ND6_RUNLOCK();
2624 	return (error);
2625 }
2626 
2627 static int
2628 nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS)
2629 {
2630 	struct in6_prefix p;
2631 	struct sockaddr_in6 s6;
2632 	struct nd_prefix *pr;
2633 	struct nd_pfxrouter *pfr;
2634 	time_t maxexpire;
2635 	int error;
2636 	char ip6buf[INET6_ADDRSTRLEN];
2637 
2638 	if (req->newptr)
2639 		return (EPERM);
2640 
2641 	error = sysctl_wire_old_buffer(req, 0);
2642 	if (error != 0)
2643 		return (error);
2644 
2645 	bzero(&p, sizeof(p));
2646 	p.origin = PR_ORIG_RA;
2647 	bzero(&s6, sizeof(s6));
2648 	s6.sin6_family = AF_INET6;
2649 	s6.sin6_len = sizeof(s6);
2650 
2651 	ND6_RLOCK();
2652 	LIST_FOREACH(pr, &V_nd_prefix, ndpr_entry) {
2653 		p.prefix = pr->ndpr_prefix;
2654 		if (sa6_recoverscope(&p.prefix)) {
2655 			log(LOG_ERR, "scope error in prefix list (%s)\n",
2656 			    ip6_sprintf(ip6buf, &p.prefix.sin6_addr));
2657 			/* XXX: press on... */
2658 		}
2659 		p.raflags = pr->ndpr_raf;
2660 		p.prefixlen = pr->ndpr_plen;
2661 		p.vltime = pr->ndpr_vltime;
2662 		p.pltime = pr->ndpr_pltime;
2663 		p.if_index = pr->ndpr_ifp->if_index;
2664 		if (pr->ndpr_vltime == ND6_INFINITE_LIFETIME)
2665 			p.expire = 0;
2666 		else {
2667 			/* XXX: we assume time_t is signed. */
2668 			maxexpire = (-1) &
2669 			    ~((time_t)1 << ((sizeof(maxexpire) * 8) - 1));
2670 			if (pr->ndpr_vltime < maxexpire - pr->ndpr_lastupdate)
2671 				p.expire = pr->ndpr_lastupdate +
2672 				    pr->ndpr_vltime +
2673 				    (time_second - time_uptime);
2674 			else
2675 				p.expire = maxexpire;
2676 		}
2677 		p.refcnt = pr->ndpr_refcnt;
2678 		p.flags = pr->ndpr_stateflags;
2679 		p.advrtrs = 0;
2680 		LIST_FOREACH(pfr, &pr->ndpr_advrtrs, pfr_entry)
2681 			p.advrtrs++;
2682 		error = SYSCTL_OUT(req, &p, sizeof(p));
2683 		if (error != 0)
2684 			break;
2685 		LIST_FOREACH(pfr, &pr->ndpr_advrtrs, pfr_entry) {
2686 			s6.sin6_addr = pfr->router->rtaddr;
2687 			if (sa6_recoverscope(&s6))
2688 				log(LOG_ERR,
2689 				    "scope error in prefix list (%s)\n",
2690 				    ip6_sprintf(ip6buf, &pfr->router->rtaddr));
2691 			error = SYSCTL_OUT(req, &s6, sizeof(s6));
2692 			if (error != 0)
2693 				break;
2694 		}
2695 	}
2696 	ND6_RUNLOCK();
2697 	return (error);
2698 }
2699