xref: /freebsd/sys/netinet6/nd6.c (revision 1f4bcc459a76b7aa664f3fd557684cd0ba6da352)
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 /* for debugging? */
117 #if 0
118 static int nd6_inuse, nd6_allocated;
119 #endif
120 
121 VNET_DEFINE(struct nd_drhead, nd_defrouter);
122 VNET_DEFINE(struct nd_prhead, nd_prefix);
123 
124 VNET_DEFINE(int, nd6_recalc_reachtm_interval) = ND6_RECALC_REACHTM_INTERVAL;
125 #define	V_nd6_recalc_reachtm_interval	VNET(nd6_recalc_reachtm_interval)
126 
127 int	(*send_sendso_input_hook)(struct mbuf *, struct ifnet *, int, int);
128 
129 static int nd6_is_new_addr_neighbor(const struct sockaddr_in6 *,
130 	struct ifnet *);
131 static void nd6_setmtu0(struct ifnet *, struct nd_ifinfo *);
132 static void nd6_slowtimo(void *);
133 static int regen_tmpaddr(struct in6_ifaddr *);
134 static void nd6_free(struct llentry *, int);
135 static void nd6_free_redirect(const struct llentry *);
136 static void nd6_llinfo_timer(void *);
137 static void nd6_llinfo_settimer_locked(struct llentry *, long);
138 static void clear_llinfo_pqueue(struct llentry *);
139 static void nd6_rtrequest(int, struct rtentry *, struct rt_addrinfo *);
140 static int nd6_resolve_slow(struct ifnet *, int, struct mbuf *,
141     const struct sockaddr_in6 *, u_char *, uint32_t *);
142 static int nd6_need_cache(struct ifnet *);
143 
144 
145 static VNET_DEFINE(struct callout, nd6_slowtimo_ch);
146 #define	V_nd6_slowtimo_ch		VNET(nd6_slowtimo_ch)
147 
148 VNET_DEFINE(struct callout, nd6_timer_ch);
149 
150 static void
151 nd6_lle_event(void *arg __unused, struct llentry *lle, int evt)
152 {
153 	struct rt_addrinfo rtinfo;
154 	struct sockaddr_in6 dst;
155 	struct sockaddr_dl gw;
156 	struct ifnet *ifp;
157 	int type;
158 
159 	LLE_WLOCK_ASSERT(lle);
160 
161 	if (lltable_get_af(lle->lle_tbl) != AF_INET6)
162 		return;
163 
164 	switch (evt) {
165 	case LLENTRY_RESOLVED:
166 		type = RTM_ADD;
167 		KASSERT(lle->la_flags & LLE_VALID,
168 		    ("%s: %p resolved but not valid?", __func__, lle));
169 		break;
170 	case LLENTRY_EXPIRED:
171 		type = RTM_DELETE;
172 		break;
173 	default:
174 		return;
175 	}
176 
177 	ifp = lltable_get_ifp(lle->lle_tbl);
178 
179 	bzero(&dst, sizeof(dst));
180 	bzero(&gw, sizeof(gw));
181 	bzero(&rtinfo, sizeof(rtinfo));
182 	lltable_fill_sa_entry(lle, (struct sockaddr *)&dst);
183 	dst.sin6_scope_id = in6_getscopezone(ifp,
184 	    in6_addrscope(&dst.sin6_addr));
185 	gw.sdl_len = sizeof(struct sockaddr_dl);
186 	gw.sdl_family = AF_LINK;
187 	gw.sdl_alen = ifp->if_addrlen;
188 	gw.sdl_index = ifp->if_index;
189 	gw.sdl_type = ifp->if_type;
190 	if (evt == LLENTRY_RESOLVED)
191 		bcopy(lle->ll_addr, gw.sdl_data, ifp->if_addrlen);
192 	rtinfo.rti_info[RTAX_DST] = (struct sockaddr *)&dst;
193 	rtinfo.rti_info[RTAX_GATEWAY] = (struct sockaddr *)&gw;
194 	rtinfo.rti_addrs = RTA_DST | RTA_GATEWAY;
195 	rt_missmsg_fib(type, &rtinfo, RTF_HOST | RTF_LLDATA | (
196 	    type == RTM_ADD ? RTF_UP: 0), 0, RT_DEFAULT_FIB);
197 }
198 
199 /*
200  * A handler for interface link layer address change event.
201  */
202 static void
203 nd6_iflladdr(void *arg __unused, struct ifnet *ifp)
204 {
205 
206 	lltable_update_ifaddr(LLTABLE6(ifp));
207 }
208 
209 void
210 nd6_init(void)
211 {
212 
213 	LIST_INIT(&V_nd_prefix);
214 
215 	/* initialization of the default router list */
216 	TAILQ_INIT(&V_nd_defrouter);
217 
218 	/* start timer */
219 	callout_init(&V_nd6_slowtimo_ch, 0);
220 	callout_reset(&V_nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
221 	    nd6_slowtimo, curvnet);
222 
223 	nd6_dad_init();
224 	if (IS_DEFAULT_VNET(curvnet)) {
225 		lle_event_eh = EVENTHANDLER_REGISTER(lle_event, nd6_lle_event,
226 		    NULL, EVENTHANDLER_PRI_ANY);
227 		iflladdr_event_eh = EVENTHANDLER_REGISTER(iflladdr_event,
228 		    nd6_iflladdr, NULL, EVENTHANDLER_PRI_ANY);
229 	}
230 }
231 
232 #ifdef VIMAGE
233 void
234 nd6_destroy()
235 {
236 
237 	callout_drain(&V_nd6_slowtimo_ch);
238 	callout_drain(&V_nd6_timer_ch);
239 	if (IS_DEFAULT_VNET(curvnet)) {
240 		EVENTHANDLER_DEREGISTER(lle_event, lle_event_eh);
241 		EVENTHANDLER_DEREGISTER(iflladdr_event, iflladdr_event_eh);
242 	}
243 }
244 #endif
245 
246 struct nd_ifinfo *
247 nd6_ifattach(struct ifnet *ifp)
248 {
249 	struct nd_ifinfo *nd;
250 
251 	nd = (struct nd_ifinfo *)malloc(sizeof(*nd), M_IP6NDP, M_WAITOK|M_ZERO);
252 	nd->initialized = 1;
253 
254 	nd->chlim = IPV6_DEFHLIM;
255 	nd->basereachable = REACHABLE_TIME;
256 	nd->reachable = ND_COMPUTE_RTIME(nd->basereachable);
257 	nd->retrans = RETRANS_TIMER;
258 
259 	nd->flags = ND6_IFF_PERFORMNUD;
260 
261 	/* A loopback interface always has ND6_IFF_AUTO_LINKLOCAL.
262 	 * XXXHRS: Clear ND6_IFF_AUTO_LINKLOCAL on an IFT_BRIDGE interface by
263 	 * default regardless of the V_ip6_auto_linklocal configuration to
264 	 * give a reasonable default behavior.
265 	 */
266 	if ((V_ip6_auto_linklocal && ifp->if_type != IFT_BRIDGE) ||
267 	    (ifp->if_flags & IFF_LOOPBACK))
268 		nd->flags |= ND6_IFF_AUTO_LINKLOCAL;
269 	/*
270 	 * A loopback interface does not need to accept RTADV.
271 	 * XXXHRS: Clear ND6_IFF_ACCEPT_RTADV on an IFT_BRIDGE interface by
272 	 * default regardless of the V_ip6_accept_rtadv configuration to
273 	 * prevent the interface from accepting RA messages arrived
274 	 * on one of the member interfaces with ND6_IFF_ACCEPT_RTADV.
275 	 */
276 	if (V_ip6_accept_rtadv &&
277 	    !(ifp->if_flags & IFF_LOOPBACK) &&
278 	    (ifp->if_type != IFT_BRIDGE))
279 			nd->flags |= ND6_IFF_ACCEPT_RTADV;
280 	if (V_ip6_no_radr && !(ifp->if_flags & IFF_LOOPBACK))
281 		nd->flags |= ND6_IFF_NO_RADR;
282 
283 	/* XXX: we cannot call nd6_setmtu since ifp is not fully initialized */
284 	nd6_setmtu0(ifp, nd);
285 
286 	return nd;
287 }
288 
289 void
290 nd6_ifdetach(struct nd_ifinfo *nd)
291 {
292 
293 	free(nd, M_IP6NDP);
294 }
295 
296 /*
297  * Reset ND level link MTU. This function is called when the physical MTU
298  * changes, which means we might have to adjust the ND level MTU.
299  */
300 void
301 nd6_setmtu(struct ifnet *ifp)
302 {
303 	if (ifp->if_afdata[AF_INET6] == NULL)
304 		return;
305 
306 	nd6_setmtu0(ifp, ND_IFINFO(ifp));
307 }
308 
309 /* XXX todo: do not maintain copy of ifp->if_mtu in ndi->maxmtu */
310 void
311 nd6_setmtu0(struct ifnet *ifp, struct nd_ifinfo *ndi)
312 {
313 	u_int32_t omaxmtu;
314 
315 	omaxmtu = ndi->maxmtu;
316 
317 	switch (ifp->if_type) {
318 	case IFT_ARCNET:
319 		ndi->maxmtu = MIN(ARC_PHDS_MAXMTU, ifp->if_mtu); /* RFC2497 */
320 		break;
321 	case IFT_FDDI:
322 		ndi->maxmtu = MIN(FDDIIPMTU, ifp->if_mtu); /* RFC2467 */
323 		break;
324 	case IFT_ISO88025:
325 		 ndi->maxmtu = MIN(ISO88025_MAX_MTU, ifp->if_mtu);
326 		 break;
327 	default:
328 		ndi->maxmtu = ifp->if_mtu;
329 		break;
330 	}
331 
332 	/*
333 	 * Decreasing the interface MTU under IPV6 minimum MTU may cause
334 	 * undesirable situation.  We thus notify the operator of the change
335 	 * explicitly.  The check for omaxmtu is necessary to restrict the
336 	 * log to the case of changing the MTU, not initializing it.
337 	 */
338 	if (omaxmtu >= IPV6_MMTU && ndi->maxmtu < IPV6_MMTU) {
339 		log(LOG_NOTICE, "nd6_setmtu0: "
340 		    "new link MTU on %s (%lu) is too small for IPv6\n",
341 		    if_name(ifp), (unsigned long)ndi->maxmtu);
342 	}
343 
344 	if (ndi->maxmtu > V_in6_maxmtu)
345 		in6_setmaxmtu(); /* check all interfaces just in case */
346 
347 }
348 
349 void
350 nd6_option_init(void *opt, int icmp6len, union nd_opts *ndopts)
351 {
352 
353 	bzero(ndopts, sizeof(*ndopts));
354 	ndopts->nd_opts_search = (struct nd_opt_hdr *)opt;
355 	ndopts->nd_opts_last
356 		= (struct nd_opt_hdr *)(((u_char *)opt) + icmp6len);
357 
358 	if (icmp6len == 0) {
359 		ndopts->nd_opts_done = 1;
360 		ndopts->nd_opts_search = NULL;
361 	}
362 }
363 
364 /*
365  * Take one ND option.
366  */
367 struct nd_opt_hdr *
368 nd6_option(union nd_opts *ndopts)
369 {
370 	struct nd_opt_hdr *nd_opt;
371 	int olen;
372 
373 	KASSERT(ndopts != NULL, ("%s: ndopts == NULL", __func__));
374 	KASSERT(ndopts->nd_opts_last != NULL, ("%s: uninitialized ndopts",
375 	    __func__));
376 	if (ndopts->nd_opts_search == NULL)
377 		return NULL;
378 	if (ndopts->nd_opts_done)
379 		return NULL;
380 
381 	nd_opt = ndopts->nd_opts_search;
382 
383 	/* make sure nd_opt_len is inside the buffer */
384 	if ((caddr_t)&nd_opt->nd_opt_len >= (caddr_t)ndopts->nd_opts_last) {
385 		bzero(ndopts, sizeof(*ndopts));
386 		return NULL;
387 	}
388 
389 	olen = nd_opt->nd_opt_len << 3;
390 	if (olen == 0) {
391 		/*
392 		 * Message validation requires that all included
393 		 * options have a length that is greater than zero.
394 		 */
395 		bzero(ndopts, sizeof(*ndopts));
396 		return NULL;
397 	}
398 
399 	ndopts->nd_opts_search = (struct nd_opt_hdr *)((caddr_t)nd_opt + olen);
400 	if (ndopts->nd_opts_search > ndopts->nd_opts_last) {
401 		/* option overruns the end of buffer, invalid */
402 		bzero(ndopts, sizeof(*ndopts));
403 		return NULL;
404 	} else if (ndopts->nd_opts_search == ndopts->nd_opts_last) {
405 		/* reached the end of options chain */
406 		ndopts->nd_opts_done = 1;
407 		ndopts->nd_opts_search = NULL;
408 	}
409 	return nd_opt;
410 }
411 
412 /*
413  * Parse multiple ND options.
414  * This function is much easier to use, for ND routines that do not need
415  * multiple options of the same type.
416  */
417 int
418 nd6_options(union nd_opts *ndopts)
419 {
420 	struct nd_opt_hdr *nd_opt;
421 	int i = 0;
422 
423 	KASSERT(ndopts != NULL, ("%s: ndopts == NULL", __func__));
424 	KASSERT(ndopts->nd_opts_last != NULL, ("%s: uninitialized ndopts",
425 	    __func__));
426 	if (ndopts->nd_opts_search == NULL)
427 		return 0;
428 
429 	while (1) {
430 		nd_opt = nd6_option(ndopts);
431 		if (nd_opt == NULL && ndopts->nd_opts_last == NULL) {
432 			/*
433 			 * Message validation requires that all included
434 			 * options have a length that is greater than zero.
435 			 */
436 			ICMP6STAT_INC(icp6s_nd_badopt);
437 			bzero(ndopts, sizeof(*ndopts));
438 			return -1;
439 		}
440 
441 		if (nd_opt == NULL)
442 			goto skip1;
443 
444 		switch (nd_opt->nd_opt_type) {
445 		case ND_OPT_SOURCE_LINKADDR:
446 		case ND_OPT_TARGET_LINKADDR:
447 		case ND_OPT_MTU:
448 		case ND_OPT_REDIRECTED_HEADER:
449 		case ND_OPT_NONCE:
450 			if (ndopts->nd_opt_array[nd_opt->nd_opt_type]) {
451 				nd6log((LOG_INFO,
452 				    "duplicated ND6 option found (type=%d)\n",
453 				    nd_opt->nd_opt_type));
454 				/* XXX bark? */
455 			} else {
456 				ndopts->nd_opt_array[nd_opt->nd_opt_type]
457 					= nd_opt;
458 			}
459 			break;
460 		case ND_OPT_PREFIX_INFORMATION:
461 			if (ndopts->nd_opt_array[nd_opt->nd_opt_type] == 0) {
462 				ndopts->nd_opt_array[nd_opt->nd_opt_type]
463 					= nd_opt;
464 			}
465 			ndopts->nd_opts_pi_end =
466 				(struct nd_opt_prefix_info *)nd_opt;
467 			break;
468 		/* What about ND_OPT_ROUTE_INFO? RFC 4191 */
469 		case ND_OPT_RDNSS:	/* RFC 6106 */
470 		case ND_OPT_DNSSL:	/* RFC 6106 */
471 			/*
472 			 * Silently ignore options we know and do not care about
473 			 * in the kernel.
474 			 */
475 			break;
476 		default:
477 			/*
478 			 * Unknown options must be silently ignored,
479 			 * to accomodate future extension to the protocol.
480 			 */
481 			nd6log((LOG_DEBUG,
482 			    "nd6_options: unsupported option %d - "
483 			    "option ignored\n", nd_opt->nd_opt_type));
484 		}
485 
486 skip1:
487 		i++;
488 		if (i > V_nd6_maxndopt) {
489 			ICMP6STAT_INC(icp6s_nd_toomanyopt);
490 			nd6log((LOG_INFO, "too many loop in nd opt\n"));
491 			break;
492 		}
493 
494 		if (ndopts->nd_opts_done)
495 			break;
496 	}
497 
498 	return 0;
499 }
500 
501 /*
502  * ND6 timer routine to handle ND6 entries
503  */
504 static void
505 nd6_llinfo_settimer_locked(struct llentry *ln, long tick)
506 {
507 	int canceled;
508 
509 	LLE_WLOCK_ASSERT(ln);
510 
511 	if (tick < 0) {
512 		ln->la_expire = 0;
513 		ln->ln_ntick = 0;
514 		canceled = callout_stop(&ln->lle_timer);
515 	} else {
516 		ln->la_expire = time_uptime + tick / hz;
517 		LLE_ADDREF(ln);
518 		if (tick > INT_MAX) {
519 			ln->ln_ntick = tick - INT_MAX;
520 			canceled = callout_reset(&ln->lle_timer, INT_MAX,
521 			    nd6_llinfo_timer, ln);
522 		} else {
523 			ln->ln_ntick = 0;
524 			canceled = callout_reset(&ln->lle_timer, tick,
525 			    nd6_llinfo_timer, ln);
526 		}
527 	}
528 	if (canceled > 0)
529 		LLE_REMREF(ln);
530 }
531 
532 /*
533  * Gets source address of the first packet in hold queue
534  * and stores it in @src.
535  * Returns pointer to @src (if hold queue is not empty) or NULL.
536  *
537  * Set noinline to be dtrace-friendly
538  */
539 static __noinline struct in6_addr *
540 nd6_llinfo_get_holdsrc(struct llentry *ln, struct in6_addr *src)
541 {
542 	struct ip6_hdr hdr;
543 	struct mbuf *m;
544 
545 	if (ln->la_hold == NULL)
546 		return (NULL);
547 
548 	/*
549 	 * assume every packet in la_hold has the same IP header
550 	 */
551 	m = ln->la_hold;
552 	if (sizeof(hdr) > m->m_len)
553 		return (NULL);
554 
555 	m_copydata(m, 0, sizeof(hdr), (caddr_t)&hdr);
556 	*src = hdr.ip6_src;
557 
558 	return (src);
559 }
560 
561 /*
562  * Checks if we need to switch from STALE state.
563  *
564  * RFC 4861 requires switching from STALE to DELAY state
565  * on first packet matching entry, waiting V_nd6_delay and
566  * transition to PROBE state (if upper layer confirmation was
567  * not received).
568  *
569  * This code performs a bit differently:
570  * On packet hit we don't change state (but desired state
571  * can be guessed by control plane). However, after V_nd6_delay
572  * seconds code will transition to PROBE state (so DELAY state
573  * is kinda skipped in most situations).
574  *
575  * Typically, V_nd6_gctimer is bigger than V_nd6_delay, so
576  * we perform the following upon entering STALE state:
577  *
578  * 1) Arm timer to run each V_nd6_delay seconds to make sure that
579  * if packet was transmitted at the start of given interval, we
580  * would be able to switch to PROBE state in V_nd6_delay seconds
581  * as user expects.
582  *
583  * 2) Reschedule timer until original V_nd6_gctimer expires keeping
584  * lle in STALE state (remaining timer value stored in lle_remtime).
585  *
586  * 3) Reschedule timer if packet was transmitted less that V_nd6_delay
587  * seconds ago.
588  *
589  * Returns non-zero value if the entry is still STALE (storing
590  * the next timer interval in @pdelay).
591  *
592  * Returns zero value if original timer expired or we need to switch to
593  * PROBE (store that in @do_switch variable).
594  */
595 static int
596 nd6_is_stale(struct llentry *lle, long *pdelay, int *do_switch)
597 {
598 	int nd_delay, nd_gctimer, r_skip_req;
599 	time_t lle_hittime;
600 	long delay;
601 
602 	*do_switch = 0;
603 	nd_gctimer = V_nd6_gctimer;
604 	nd_delay = V_nd6_delay;
605 
606 	LLE_REQ_LOCK(lle);
607 	r_skip_req = lle->r_skip_req;
608 	lle_hittime = lle->lle_hittime;
609 	LLE_REQ_UNLOCK(lle);
610 
611 	if (r_skip_req > 0) {
612 
613 		/*
614 		 * Nonzero r_skip_req value was set upon entering
615 		 * STALE state. Since value was not changed, no
616 		 * packets were passed using this lle. Ask for
617 		 * timer reschedule and keep STALE state.
618 		 */
619 		delay = (long)(MIN(nd_gctimer, nd_delay));
620 		delay *= hz;
621 		if (lle->lle_remtime > delay)
622 			lle->lle_remtime -= delay;
623 		else {
624 			delay = lle->lle_remtime;
625 			lle->lle_remtime = 0;
626 		}
627 
628 		if (delay == 0) {
629 
630 			/*
631 			 * The original ng6_gctime timeout ended,
632 			 * no more rescheduling.
633 			 */
634 			return (0);
635 		}
636 
637 		*pdelay = delay;
638 		return (1);
639 	}
640 
641 	/*
642 	 * Packet received. Verify timestamp
643 	 */
644 	delay = (long)(time_uptime - lle_hittime);
645 	if (delay < nd_delay) {
646 
647 		/*
648 		 * V_nd6_delay still not passed since the first
649 		 * hit in STALE state.
650 		 * Reshedule timer and return.
651 		 */
652 		*pdelay = (long)(nd_delay - delay) * hz;
653 		return (1);
654 	}
655 
656 	/* Request switching to probe */
657 	*do_switch = 1;
658 	return (0);
659 }
660 
661 
662 /*
663  * Switch @lle state to new state optionally arming timers.
664  *
665  * Set noinline to be dtrace-friendly
666  */
667 __noinline void
668 nd6_llinfo_setstate(struct llentry *lle, int newstate)
669 {
670 	struct ifnet *ifp;
671 	int nd_gctimer, nd_delay;
672 	long delay, remtime;
673 
674 	delay = 0;
675 	remtime = 0;
676 
677 	switch (newstate) {
678 	case ND6_LLINFO_INCOMPLETE:
679 		ifp = lle->lle_tbl->llt_ifp;
680 		delay = (long)ND_IFINFO(ifp)->retrans * hz / 1000;
681 		break;
682 	case ND6_LLINFO_REACHABLE:
683 		if (!ND6_LLINFO_PERMANENT(lle)) {
684 			ifp = lle->lle_tbl->llt_ifp;
685 			delay = (long)ND_IFINFO(ifp)->reachable * hz;
686 		}
687 		break;
688 	case ND6_LLINFO_STALE:
689 
690 		/*
691 		 * Notify fast path that we want to know if any packet
692 		 * is transmitted by setting r_skip_req.
693 		 */
694 		LLE_REQ_LOCK(lle);
695 		lle->r_skip_req = 1;
696 		LLE_REQ_UNLOCK(lle);
697 		nd_delay = V_nd6_delay;
698 		nd_gctimer = V_nd6_gctimer;
699 
700 		delay = (long)(MIN(nd_gctimer, nd_delay)) * hz;
701 		remtime = (long)nd_gctimer * hz - delay;
702 		break;
703 	case ND6_LLINFO_DELAY:
704 		lle->la_asked = 0;
705 		delay = (long)V_nd6_delay * hz;
706 		break;
707 	}
708 
709 	if (delay > 0)
710 		nd6_llinfo_settimer_locked(lle, delay);
711 
712 	lle->lle_remtime = remtime;
713 	lle->ln_state = newstate;
714 }
715 
716 /*
717  * Timer-dependent part of nd state machine.
718  *
719  * Set noinline to be dtrace-friendly
720  */
721 static __noinline void
722 nd6_llinfo_timer(void *arg)
723 {
724 	struct llentry *ln;
725 	struct in6_addr *dst, *pdst, *psrc, src;
726 	struct ifnet *ifp;
727 	struct nd_ifinfo *ndi = NULL;
728 	int do_switch, send_ns;
729 	long delay;
730 
731 	KASSERT(arg != NULL, ("%s: arg NULL", __func__));
732 	ln = (struct llentry *)arg;
733 	LLE_WLOCK(ln);
734 	if (callout_pending(&ln->lle_timer)) {
735 		/*
736 		 * Here we are a bit odd here in the treatment of
737 		 * active/pending. If the pending bit is set, it got
738 		 * rescheduled before I ran. The active
739 		 * bit we ignore, since if it was stopped
740 		 * in ll_tablefree() and was currently running
741 		 * it would have return 0 so the code would
742 		 * not have deleted it since the callout could
743 		 * not be stopped so we want to go through
744 		 * with the delete here now. If the callout
745 		 * was restarted, the pending bit will be back on and
746 		 * we just want to bail since the callout_reset would
747 		 * return 1 and our reference would have been removed
748 		 * by nd6_llinfo_settimer_locked above since canceled
749 		 * would have been 1.
750 		 */
751 		LLE_WUNLOCK(ln);
752 		return;
753 	}
754 	ifp = ln->lle_tbl->llt_ifp;
755 	CURVNET_SET(ifp->if_vnet);
756 	ndi = ND_IFINFO(ifp);
757 	send_ns = 0;
758 	dst = &ln->r_l3addr.addr6;
759 	pdst = dst;
760 
761 	if (ln->ln_ntick > 0) {
762 		if (ln->ln_ntick > INT_MAX) {
763 			ln->ln_ntick -= INT_MAX;
764 			nd6_llinfo_settimer_locked(ln, INT_MAX);
765 		} else {
766 			ln->ln_ntick = 0;
767 			nd6_llinfo_settimer_locked(ln, ln->ln_ntick);
768 		}
769 		goto done;
770 	}
771 
772 	if (ln->la_flags & LLE_STATIC) {
773 		goto done;
774 	}
775 
776 	if (ln->la_flags & LLE_DELETED) {
777 		nd6_free(ln, 0);
778 		ln = NULL;
779 		goto done;
780 	}
781 
782 	switch (ln->ln_state) {
783 	case ND6_LLINFO_INCOMPLETE:
784 		if (ln->la_asked < V_nd6_mmaxtries) {
785 			ln->la_asked++;
786 			send_ns = 1;
787 			/* Send NS to multicast address */
788 			pdst = NULL;
789 		} else {
790 			struct mbuf *m = ln->la_hold;
791 			if (m) {
792 				struct mbuf *m0;
793 
794 				/*
795 				 * assuming every packet in la_hold has the
796 				 * same IP header.  Send error after unlock.
797 				 */
798 				m0 = m->m_nextpkt;
799 				m->m_nextpkt = NULL;
800 				ln->la_hold = m0;
801 				clear_llinfo_pqueue(ln);
802 			}
803 			EVENTHANDLER_INVOKE(lle_event, ln, LLENTRY_TIMEDOUT);
804 			nd6_free(ln, 0);
805 			ln = NULL;
806 			if (m != NULL)
807 				icmp6_error2(m, ICMP6_DST_UNREACH,
808 				    ICMP6_DST_UNREACH_ADDR, 0, ifp);
809 		}
810 		break;
811 	case ND6_LLINFO_REACHABLE:
812 		if (!ND6_LLINFO_PERMANENT(ln))
813 			nd6_llinfo_setstate(ln, ND6_LLINFO_STALE);
814 		break;
815 
816 	case ND6_LLINFO_STALE:
817 		if (nd6_is_stale(ln, &delay, &do_switch) != 0) {
818 
819 			/*
820 			 * No packet has used this entry and GC timeout
821 			 * has not been passed. Reshedule timer and
822 			 * return.
823 			 */
824 			nd6_llinfo_settimer_locked(ln, delay);
825 			break;
826 		}
827 
828 		if (do_switch == 0) {
829 
830 			/*
831 			 * GC timer has ended and entry hasn't been used.
832 			 * Run Garbage collector (RFC 4861, 5.3)
833 			 */
834 			if (!ND6_LLINFO_PERMANENT(ln)) {
835 				EVENTHANDLER_INVOKE(lle_event, ln,
836 				    LLENTRY_EXPIRED);
837 				nd6_free(ln, 1);
838 				ln = NULL;
839 			}
840 			break;
841 		}
842 
843 		/* Entry has been used AND delay timer has ended. */
844 
845 		/* FALLTHROUGH */
846 
847 	case ND6_LLINFO_DELAY:
848 		if (ndi && (ndi->flags & ND6_IFF_PERFORMNUD) != 0) {
849 			/* We need NUD */
850 			ln->la_asked = 1;
851 			nd6_llinfo_setstate(ln, ND6_LLINFO_PROBE);
852 			send_ns = 1;
853 		} else
854 			nd6_llinfo_setstate(ln, ND6_LLINFO_STALE); /* XXX */
855 		break;
856 	case ND6_LLINFO_PROBE:
857 		if (ln->la_asked < V_nd6_umaxtries) {
858 			ln->la_asked++;
859 			send_ns = 1;
860 		} else {
861 			EVENTHANDLER_INVOKE(lle_event, ln, LLENTRY_EXPIRED);
862 			nd6_free(ln, 0);
863 			ln = NULL;
864 		}
865 		break;
866 	default:
867 		panic("%s: paths in a dark night can be confusing: %d",
868 		    __func__, ln->ln_state);
869 	}
870 done:
871 	if (send_ns != 0) {
872 		nd6_llinfo_settimer_locked(ln, (long)ndi->retrans * hz / 1000);
873 		psrc = nd6_llinfo_get_holdsrc(ln, &src);
874 		LLE_FREE_LOCKED(ln);
875 		ln = NULL;
876 		nd6_ns_output(ifp, psrc, pdst, dst, NULL);
877 	}
878 
879 	if (ln != NULL)
880 		LLE_FREE_LOCKED(ln);
881 	CURVNET_RESTORE();
882 }
883 
884 
885 /*
886  * ND6 timer routine to expire default route list and prefix list
887  */
888 void
889 nd6_timer(void *arg)
890 {
891 	CURVNET_SET((struct vnet *) arg);
892 	struct nd_defrouter *dr, *ndr;
893 	struct nd_prefix *pr, *npr;
894 	struct in6_ifaddr *ia6, *nia6;
895 
896 	callout_reset(&V_nd6_timer_ch, V_nd6_prune * hz,
897 	    nd6_timer, curvnet);
898 
899 	/* expire default router list */
900 	TAILQ_FOREACH_SAFE(dr, &V_nd_defrouter, dr_entry, ndr) {
901 		if (dr->expire && dr->expire < time_uptime)
902 			defrtrlist_del(dr);
903 	}
904 
905 	/*
906 	 * expire interface addresses.
907 	 * in the past the loop was inside prefix expiry processing.
908 	 * However, from a stricter speci-confrmance standpoint, we should
909 	 * rather separate address lifetimes and prefix lifetimes.
910 	 *
911 	 * XXXRW: in6_ifaddrhead locking.
912 	 */
913   addrloop:
914 	TAILQ_FOREACH_SAFE(ia6, &V_in6_ifaddrhead, ia_link, nia6) {
915 		/* check address lifetime */
916 		if (IFA6_IS_INVALID(ia6)) {
917 			int regen = 0;
918 
919 			/*
920 			 * If the expiring address is temporary, try
921 			 * regenerating a new one.  This would be useful when
922 			 * we suspended a laptop PC, then turned it on after a
923 			 * period that could invalidate all temporary
924 			 * addresses.  Although we may have to restart the
925 			 * loop (see below), it must be after purging the
926 			 * address.  Otherwise, we'd see an infinite loop of
927 			 * regeneration.
928 			 */
929 			if (V_ip6_use_tempaddr &&
930 			    (ia6->ia6_flags & IN6_IFF_TEMPORARY) != 0) {
931 				if (regen_tmpaddr(ia6) == 0)
932 					regen = 1;
933 			}
934 
935 			in6_purgeaddr(&ia6->ia_ifa);
936 
937 			if (regen)
938 				goto addrloop; /* XXX: see below */
939 		} else if (IFA6_IS_DEPRECATED(ia6)) {
940 			int oldflags = ia6->ia6_flags;
941 
942 			ia6->ia6_flags |= IN6_IFF_DEPRECATED;
943 
944 			/*
945 			 * If a temporary address has just become deprecated,
946 			 * regenerate a new one if possible.
947 			 */
948 			if (V_ip6_use_tempaddr &&
949 			    (ia6->ia6_flags & IN6_IFF_TEMPORARY) != 0 &&
950 			    (oldflags & IN6_IFF_DEPRECATED) == 0) {
951 
952 				if (regen_tmpaddr(ia6) == 0) {
953 					/*
954 					 * A new temporary address is
955 					 * generated.
956 					 * XXX: this means the address chain
957 					 * has changed while we are still in
958 					 * the loop.  Although the change
959 					 * would not cause disaster (because
960 					 * it's not a deletion, but an
961 					 * addition,) we'd rather restart the
962 					 * loop just for safety.  Or does this
963 					 * significantly reduce performance??
964 					 */
965 					goto addrloop;
966 				}
967 			}
968 		} else if ((ia6->ia6_flags & IN6_IFF_TENTATIVE) != 0) {
969 			/*
970 			 * Schedule DAD for a tentative address.  This happens
971 			 * if the interface was down or not running
972 			 * when the address was configured.
973 			 */
974 			int delay;
975 
976 			delay = arc4random() %
977 			    (MAX_RTR_SOLICITATION_DELAY * hz);
978 			nd6_dad_start((struct ifaddr *)ia6, delay);
979 		} else {
980 			/*
981 			 * Check status of the interface.  If it is down,
982 			 * mark the address as tentative for future DAD.
983 			 */
984 			if ((ia6->ia_ifp->if_flags & IFF_UP) == 0 ||
985 			    (ia6->ia_ifp->if_drv_flags & IFF_DRV_RUNNING)
986 				== 0 ||
987 			    (ND_IFINFO(ia6->ia_ifp)->flags &
988 				ND6_IFF_IFDISABLED) != 0) {
989 				ia6->ia6_flags &= ~IN6_IFF_DUPLICATED;
990 				ia6->ia6_flags |= IN6_IFF_TENTATIVE;
991 			}
992 			/*
993 			 * A new RA might have made a deprecated address
994 			 * preferred.
995 			 */
996 			ia6->ia6_flags &= ~IN6_IFF_DEPRECATED;
997 		}
998 	}
999 
1000 	/* expire prefix list */
1001 	LIST_FOREACH_SAFE(pr, &V_nd_prefix, ndpr_entry, npr) {
1002 		/*
1003 		 * check prefix lifetime.
1004 		 * since pltime is just for autoconf, pltime processing for
1005 		 * prefix is not necessary.
1006 		 */
1007 		if (pr->ndpr_vltime != ND6_INFINITE_LIFETIME &&
1008 		    time_uptime - pr->ndpr_lastupdate > pr->ndpr_vltime) {
1009 
1010 			/*
1011 			 * address expiration and prefix expiration are
1012 			 * separate.  NEVER perform in6_purgeaddr here.
1013 			 */
1014 			prelist_remove(pr);
1015 		}
1016 	}
1017 	CURVNET_RESTORE();
1018 }
1019 
1020 /*
1021  * ia6 - deprecated/invalidated temporary address
1022  */
1023 static int
1024 regen_tmpaddr(struct in6_ifaddr *ia6)
1025 {
1026 	struct ifaddr *ifa;
1027 	struct ifnet *ifp;
1028 	struct in6_ifaddr *public_ifa6 = NULL;
1029 
1030 	ifp = ia6->ia_ifa.ifa_ifp;
1031 	IF_ADDR_RLOCK(ifp);
1032 	TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1033 		struct in6_ifaddr *it6;
1034 
1035 		if (ifa->ifa_addr->sa_family != AF_INET6)
1036 			continue;
1037 
1038 		it6 = (struct in6_ifaddr *)ifa;
1039 
1040 		/* ignore no autoconf addresses. */
1041 		if ((it6->ia6_flags & IN6_IFF_AUTOCONF) == 0)
1042 			continue;
1043 
1044 		/* ignore autoconf addresses with different prefixes. */
1045 		if (it6->ia6_ndpr == NULL || it6->ia6_ndpr != ia6->ia6_ndpr)
1046 			continue;
1047 
1048 		/*
1049 		 * Now we are looking at an autoconf address with the same
1050 		 * prefix as ours.  If the address is temporary and is still
1051 		 * preferred, do not create another one.  It would be rare, but
1052 		 * could happen, for example, when we resume a laptop PC after
1053 		 * a long period.
1054 		 */
1055 		if ((it6->ia6_flags & IN6_IFF_TEMPORARY) != 0 &&
1056 		    !IFA6_IS_DEPRECATED(it6)) {
1057 			public_ifa6 = NULL;
1058 			break;
1059 		}
1060 
1061 		/*
1062 		 * This is a public autoconf address that has the same prefix
1063 		 * as ours.  If it is preferred, keep it.  We can't break the
1064 		 * loop here, because there may be a still-preferred temporary
1065 		 * address with the prefix.
1066 		 */
1067 		if (!IFA6_IS_DEPRECATED(it6))
1068 			public_ifa6 = it6;
1069 	}
1070 	if (public_ifa6 != NULL)
1071 		ifa_ref(&public_ifa6->ia_ifa);
1072 	IF_ADDR_RUNLOCK(ifp);
1073 
1074 	if (public_ifa6 != NULL) {
1075 		int e;
1076 
1077 		if ((e = in6_tmpifadd(public_ifa6, 0, 0)) != 0) {
1078 			ifa_free(&public_ifa6->ia_ifa);
1079 			log(LOG_NOTICE, "regen_tmpaddr: failed to create a new"
1080 			    " tmp addr,errno=%d\n", e);
1081 			return (-1);
1082 		}
1083 		ifa_free(&public_ifa6->ia_ifa);
1084 		return (0);
1085 	}
1086 
1087 	return (-1);
1088 }
1089 
1090 /*
1091  * Nuke neighbor cache/prefix/default router management table, right before
1092  * ifp goes away.
1093  */
1094 void
1095 nd6_purge(struct ifnet *ifp)
1096 {
1097 	struct nd_defrouter *dr, *ndr;
1098 	struct nd_prefix *pr, *npr;
1099 
1100 	/*
1101 	 * Nuke default router list entries toward ifp.
1102 	 * We defer removal of default router list entries that is installed
1103 	 * in the routing table, in order to keep additional side effects as
1104 	 * small as possible.
1105 	 */
1106 	TAILQ_FOREACH_SAFE(dr, &V_nd_defrouter, dr_entry, ndr) {
1107 		if (dr->installed)
1108 			continue;
1109 
1110 		if (dr->ifp == ifp)
1111 			defrtrlist_del(dr);
1112 	}
1113 
1114 	TAILQ_FOREACH_SAFE(dr, &V_nd_defrouter, dr_entry, ndr) {
1115 		if (!dr->installed)
1116 			continue;
1117 
1118 		if (dr->ifp == ifp)
1119 			defrtrlist_del(dr);
1120 	}
1121 
1122 	/* Nuke prefix list entries toward ifp */
1123 	LIST_FOREACH_SAFE(pr, &V_nd_prefix, ndpr_entry, npr) {
1124 		if (pr->ndpr_ifp == ifp) {
1125 			/*
1126 			 * Because if_detach() does *not* release prefixes
1127 			 * while purging addresses the reference count will
1128 			 * still be above zero. We therefore reset it to
1129 			 * make sure that the prefix really gets purged.
1130 			 */
1131 			pr->ndpr_refcnt = 0;
1132 
1133 			/*
1134 			 * Previously, pr->ndpr_addr is removed as well,
1135 			 * but I strongly believe we don't have to do it.
1136 			 * nd6_purge() is only called from in6_ifdetach(),
1137 			 * which removes all the associated interface addresses
1138 			 * by itself.
1139 			 * (jinmei@kame.net 20010129)
1140 			 */
1141 			prelist_remove(pr);
1142 		}
1143 	}
1144 
1145 	/* cancel default outgoing interface setting */
1146 	if (V_nd6_defifindex == ifp->if_index)
1147 		nd6_setdefaultiface(0);
1148 
1149 	if (ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) {
1150 		/* Refresh default router list. */
1151 		defrouter_select();
1152 	}
1153 
1154 	/* XXXXX
1155 	 * We do not nuke the neighbor cache entries here any more
1156 	 * because the neighbor cache is kept in if_afdata[AF_INET6].
1157 	 * nd6_purge() is invoked by in6_ifdetach() which is called
1158 	 * from if_detach() where everything gets purged. So let
1159 	 * in6_domifdetach() do the actual L2 table purging work.
1160 	 */
1161 }
1162 
1163 /*
1164  * the caller acquires and releases the lock on the lltbls
1165  * Returns the llentry locked
1166  */
1167 struct llentry *
1168 nd6_lookup(const struct in6_addr *addr6, int flags, struct ifnet *ifp)
1169 {
1170 	struct sockaddr_in6 sin6;
1171 	struct llentry *ln;
1172 
1173 	bzero(&sin6, sizeof(sin6));
1174 	sin6.sin6_len = sizeof(struct sockaddr_in6);
1175 	sin6.sin6_family = AF_INET6;
1176 	sin6.sin6_addr = *addr6;
1177 
1178 	IF_AFDATA_LOCK_ASSERT(ifp);
1179 
1180 	ln = lla_lookup(LLTABLE6(ifp), flags, (struct sockaddr *)&sin6);
1181 
1182 	return (ln);
1183 }
1184 
1185 struct llentry *
1186 nd6_alloc(const struct in6_addr *addr6, int flags, struct ifnet *ifp)
1187 {
1188 	struct sockaddr_in6 sin6;
1189 	struct llentry *ln;
1190 
1191 	bzero(&sin6, sizeof(sin6));
1192 	sin6.sin6_len = sizeof(struct sockaddr_in6);
1193 	sin6.sin6_family = AF_INET6;
1194 	sin6.sin6_addr = *addr6;
1195 
1196 	ln = lltable_alloc_entry(LLTABLE6(ifp), 0, (struct sockaddr *)&sin6);
1197 	if (ln != NULL)
1198 		ln->ln_state = ND6_LLINFO_NOSTATE;
1199 
1200 	return (ln);
1201 }
1202 
1203 /*
1204  * Test whether a given IPv6 address is a neighbor or not, ignoring
1205  * the actual neighbor cache.  The neighbor cache is ignored in order
1206  * to not reenter the routing code from within itself.
1207  */
1208 static int
1209 nd6_is_new_addr_neighbor(const struct sockaddr_in6 *addr, struct ifnet *ifp)
1210 {
1211 	struct nd_prefix *pr;
1212 	struct ifaddr *dstaddr;
1213 	struct rt_addrinfo info;
1214 	struct sockaddr_in6 rt_key;
1215 	struct sockaddr *dst6;
1216 	int fibnum;
1217 
1218 	/*
1219 	 * A link-local address is always a neighbor.
1220 	 * XXX: a link does not necessarily specify a single interface.
1221 	 */
1222 	if (IN6_IS_ADDR_LINKLOCAL(&addr->sin6_addr)) {
1223 		struct sockaddr_in6 sin6_copy;
1224 		u_int32_t zone;
1225 
1226 		/*
1227 		 * We need sin6_copy since sa6_recoverscope() may modify the
1228 		 * content (XXX).
1229 		 */
1230 		sin6_copy = *addr;
1231 		if (sa6_recoverscope(&sin6_copy))
1232 			return (0); /* XXX: should be impossible */
1233 		if (in6_setscope(&sin6_copy.sin6_addr, ifp, &zone))
1234 			return (0);
1235 		if (sin6_copy.sin6_scope_id == zone)
1236 			return (1);
1237 		else
1238 			return (0);
1239 	}
1240 
1241 	bzero(&rt_key, sizeof(rt_key));
1242 	bzero(&info, sizeof(info));
1243 	info.rti_info[RTAX_DST] = (struct sockaddr *)&rt_key;
1244 
1245 	/* Always use the default FIB here. XXME - why? */
1246 	fibnum = RT_DEFAULT_FIB;
1247 
1248 	/*
1249 	 * If the address matches one of our addresses,
1250 	 * it should be a neighbor.
1251 	 * If the address matches one of our on-link prefixes, it should be a
1252 	 * neighbor.
1253 	 */
1254 	LIST_FOREACH(pr, &V_nd_prefix, ndpr_entry) {
1255 		if (pr->ndpr_ifp != ifp)
1256 			continue;
1257 
1258 		if (!(pr->ndpr_stateflags & NDPRF_ONLINK)) {
1259 
1260 			/* Always use the default FIB here. */
1261 			dst6 = (struct sockaddr *)&pr->ndpr_prefix;
1262 
1263 			/* Restore length field before retrying lookup */
1264 			rt_key.sin6_len = sizeof(rt_key);
1265 			if (rib_lookup_info(fibnum, dst6, 0, 0, &info) != 0)
1266 				continue;
1267 			/*
1268 			 * This is the case where multiple interfaces
1269 			 * have the same prefix, but only one is installed
1270 			 * into the routing table and that prefix entry
1271 			 * is not the one being examined here. In the case
1272 			 * where RADIX_MPATH is enabled, multiple route
1273 			 * entries (of the same rt_key value) will be
1274 			 * installed because the interface addresses all
1275 			 * differ.
1276 			 */
1277 			if (!IN6_ARE_ADDR_EQUAL(&pr->ndpr_prefix.sin6_addr,
1278 			       &rt_key.sin6_addr))
1279 				continue;
1280 		}
1281 
1282 		if (IN6_ARE_MASKED_ADDR_EQUAL(&pr->ndpr_prefix.sin6_addr,
1283 		    &addr->sin6_addr, &pr->ndpr_mask))
1284 			return (1);
1285 	}
1286 
1287 	/*
1288 	 * If the address is assigned on the node of the other side of
1289 	 * a p2p interface, the address should be a neighbor.
1290 	 */
1291 	dstaddr = ifa_ifwithdstaddr((const struct sockaddr *)addr, RT_ALL_FIBS);
1292 	if (dstaddr != NULL) {
1293 		if (dstaddr->ifa_ifp == ifp) {
1294 			ifa_free(dstaddr);
1295 			return (1);
1296 		}
1297 		ifa_free(dstaddr);
1298 	}
1299 
1300 	/*
1301 	 * If the default router list is empty, all addresses are regarded
1302 	 * as on-link, and thus, as a neighbor.
1303 	 */
1304 	if (ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV &&
1305 	    TAILQ_EMPTY(&V_nd_defrouter) &&
1306 	    V_nd6_defifindex == ifp->if_index) {
1307 		return (1);
1308 	}
1309 
1310 	return (0);
1311 }
1312 
1313 
1314 /*
1315  * Detect if a given IPv6 address identifies a neighbor on a given link.
1316  * XXX: should take care of the destination of a p2p link?
1317  */
1318 int
1319 nd6_is_addr_neighbor(const struct sockaddr_in6 *addr, struct ifnet *ifp)
1320 {
1321 	struct llentry *lle;
1322 	int rc = 0;
1323 
1324 	IF_AFDATA_UNLOCK_ASSERT(ifp);
1325 	if (nd6_is_new_addr_neighbor(addr, ifp))
1326 		return (1);
1327 
1328 	/*
1329 	 * Even if the address matches none of our addresses, it might be
1330 	 * in the neighbor cache.
1331 	 */
1332 	IF_AFDATA_RLOCK(ifp);
1333 	if ((lle = nd6_lookup(&addr->sin6_addr, 0, ifp)) != NULL) {
1334 		LLE_RUNLOCK(lle);
1335 		rc = 1;
1336 	}
1337 	IF_AFDATA_RUNLOCK(ifp);
1338 	return (rc);
1339 }
1340 
1341 /*
1342  * Free an nd6 llinfo entry.
1343  * Since the function would cause significant changes in the kernel, DO NOT
1344  * make it global, unless you have a strong reason for the change, and are sure
1345  * that the change is safe.
1346  *
1347  * Set noinline to be dtrace-friendly
1348  */
1349 static __noinline void
1350 nd6_free(struct llentry *ln, int gc)
1351 {
1352 	struct nd_defrouter *dr;
1353 	struct ifnet *ifp;
1354 
1355 	LLE_WLOCK_ASSERT(ln);
1356 
1357 	/*
1358 	 * we used to have pfctlinput(PRC_HOSTDEAD) here.
1359 	 * even though it is not harmful, it was not really necessary.
1360 	 */
1361 
1362 	/* cancel timer */
1363 	nd6_llinfo_settimer_locked(ln, -1);
1364 
1365 	ifp = ln->lle_tbl->llt_ifp;
1366 
1367 	if (ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) {
1368 		dr = defrouter_lookup(&ln->r_l3addr.addr6, ifp);
1369 
1370 		if (dr != NULL && dr->expire &&
1371 		    ln->ln_state == ND6_LLINFO_STALE && gc) {
1372 			/*
1373 			 * If the reason for the deletion is just garbage
1374 			 * collection, and the neighbor is an active default
1375 			 * router, do not delete it.  Instead, reset the GC
1376 			 * timer using the router's lifetime.
1377 			 * Simply deleting the entry would affect default
1378 			 * router selection, which is not necessarily a good
1379 			 * thing, especially when we're using router preference
1380 			 * values.
1381 			 * XXX: the check for ln_state would be redundant,
1382 			 *      but we intentionally keep it just in case.
1383 			 */
1384 			if (dr->expire > time_uptime)
1385 				nd6_llinfo_settimer_locked(ln,
1386 				    (dr->expire - time_uptime) * hz);
1387 			else
1388 				nd6_llinfo_settimer_locked(ln,
1389 				    (long)V_nd6_gctimer * hz);
1390 
1391 			LLE_REMREF(ln);
1392 			LLE_WUNLOCK(ln);
1393 			return;
1394 		}
1395 
1396 		if (dr) {
1397 			/*
1398 			 * Unreachablity of a router might affect the default
1399 			 * router selection and on-link detection of advertised
1400 			 * prefixes.
1401 			 */
1402 
1403 			/*
1404 			 * Temporarily fake the state to choose a new default
1405 			 * router and to perform on-link determination of
1406 			 * prefixes correctly.
1407 			 * Below the state will be set correctly,
1408 			 * or the entry itself will be deleted.
1409 			 */
1410 			ln->ln_state = ND6_LLINFO_INCOMPLETE;
1411 		}
1412 
1413 		if (ln->ln_router || dr) {
1414 
1415 			/*
1416 			 * We need to unlock to avoid a LOR with rt6_flush() with the
1417 			 * rnh and for the calls to pfxlist_onlink_check() and
1418 			 * defrouter_select() in the block further down for calls
1419 			 * into nd6_lookup().  We still hold a ref.
1420 			 */
1421 			LLE_WUNLOCK(ln);
1422 
1423 			/*
1424 			 * rt6_flush must be called whether or not the neighbor
1425 			 * is in the Default Router List.
1426 			 * See a corresponding comment in nd6_na_input().
1427 			 */
1428 			rt6_flush(&ln->r_l3addr.addr6, ifp);
1429 		}
1430 
1431 		if (dr) {
1432 			/*
1433 			 * Since defrouter_select() does not affect the
1434 			 * on-link determination and MIP6 needs the check
1435 			 * before the default router selection, we perform
1436 			 * the check now.
1437 			 */
1438 			pfxlist_onlink_check();
1439 
1440 			/*
1441 			 * Refresh default router list.
1442 			 */
1443 			defrouter_select();
1444 		}
1445 
1446 		/*
1447 		 * If this entry was added by an on-link redirect, remove the
1448 		 * corresponding host route.
1449 		 */
1450 		if (ln->la_flags & LLE_REDIRECT)
1451 			nd6_free_redirect(ln);
1452 
1453 		if (ln->ln_router || dr)
1454 			LLE_WLOCK(ln);
1455 	}
1456 
1457 	/*
1458 	 * Save to unlock. We still hold an extra reference and will not
1459 	 * free(9) in llentry_free() if someone else holds one as well.
1460 	 */
1461 	LLE_WUNLOCK(ln);
1462 	IF_AFDATA_LOCK(ifp);
1463 	LLE_WLOCK(ln);
1464 	/* Guard against race with other llentry_free(). */
1465 	if (ln->la_flags & LLE_LINKED) {
1466 		/* Remove callout reference */
1467 		LLE_REMREF(ln);
1468 		lltable_unlink_entry(ln->lle_tbl, ln);
1469 	}
1470 	IF_AFDATA_UNLOCK(ifp);
1471 
1472 	llentry_free(ln);
1473 }
1474 
1475 static int
1476 nd6_isdynrte(const struct rtentry *rt, void *xap)
1477 {
1478 
1479 	if (rt->rt_flags == (RTF_UP | RTF_HOST | RTF_DYNAMIC))
1480 		return (1);
1481 
1482 	return (0);
1483 }
1484 /*
1485  * Remove the rtentry for the given llentry,
1486  * both of which were installed by a redirect.
1487  */
1488 static void
1489 nd6_free_redirect(const struct llentry *ln)
1490 {
1491 	int fibnum;
1492 	struct sockaddr_in6 sin6;
1493 	struct rt_addrinfo info;
1494 
1495 	lltable_fill_sa_entry(ln, (struct sockaddr *)&sin6);
1496 	memset(&info, 0, sizeof(info));
1497 	info.rti_info[RTAX_DST] = (struct sockaddr *)&sin6;
1498 	info.rti_filter = nd6_isdynrte;
1499 
1500 	for (fibnum = 0; fibnum < rt_numfibs; fibnum++)
1501 		rtrequest1_fib(RTM_DELETE, &info, NULL, fibnum);
1502 }
1503 
1504 /*
1505  * Rejuvenate this function for routing operations related
1506  * processing.
1507  */
1508 void
1509 nd6_rtrequest(int req, struct rtentry *rt, struct rt_addrinfo *info)
1510 {
1511 	struct sockaddr_in6 *gateway;
1512 	struct nd_defrouter *dr;
1513 	struct ifnet *ifp;
1514 
1515 	gateway = (struct sockaddr_in6 *)rt->rt_gateway;
1516 	ifp = rt->rt_ifp;
1517 
1518 	switch (req) {
1519 	case RTM_ADD:
1520 		break;
1521 
1522 	case RTM_DELETE:
1523 		if (!ifp)
1524 			return;
1525 		/*
1526 		 * Only indirect routes are interesting.
1527 		 */
1528 		if ((rt->rt_flags & RTF_GATEWAY) == 0)
1529 			return;
1530 		/*
1531 		 * check for default route
1532 		 */
1533 		if (IN6_ARE_ADDR_EQUAL(&in6addr_any,
1534 				       &SIN6(rt_key(rt))->sin6_addr)) {
1535 
1536 			dr = defrouter_lookup(&gateway->sin6_addr, ifp);
1537 			if (dr != NULL)
1538 				dr->installed = 0;
1539 		}
1540 		break;
1541 	}
1542 }
1543 
1544 
1545 int
1546 nd6_ioctl(u_long cmd, caddr_t data, struct ifnet *ifp)
1547 {
1548 	struct in6_ndireq *ndi = (struct in6_ndireq *)data;
1549 	struct in6_nbrinfo *nbi = (struct in6_nbrinfo *)data;
1550 	struct in6_ndifreq *ndif = (struct in6_ndifreq *)data;
1551 	int error = 0;
1552 
1553 	if (ifp->if_afdata[AF_INET6] == NULL)
1554 		return (EPFNOSUPPORT);
1555 	switch (cmd) {
1556 	case OSIOCGIFINFO_IN6:
1557 #define ND	ndi->ndi
1558 		/* XXX: old ndp(8) assumes a positive value for linkmtu. */
1559 		bzero(&ND, sizeof(ND));
1560 		ND.linkmtu = IN6_LINKMTU(ifp);
1561 		ND.maxmtu = ND_IFINFO(ifp)->maxmtu;
1562 		ND.basereachable = ND_IFINFO(ifp)->basereachable;
1563 		ND.reachable = ND_IFINFO(ifp)->reachable;
1564 		ND.retrans = ND_IFINFO(ifp)->retrans;
1565 		ND.flags = ND_IFINFO(ifp)->flags;
1566 		ND.recalctm = ND_IFINFO(ifp)->recalctm;
1567 		ND.chlim = ND_IFINFO(ifp)->chlim;
1568 		break;
1569 	case SIOCGIFINFO_IN6:
1570 		ND = *ND_IFINFO(ifp);
1571 		break;
1572 	case SIOCSIFINFO_IN6:
1573 		/*
1574 		 * used to change host variables from userland.
1575 		 * intented for a use on router to reflect RA configurations.
1576 		 */
1577 		/* 0 means 'unspecified' */
1578 		if (ND.linkmtu != 0) {
1579 			if (ND.linkmtu < IPV6_MMTU ||
1580 			    ND.linkmtu > IN6_LINKMTU(ifp)) {
1581 				error = EINVAL;
1582 				break;
1583 			}
1584 			ND_IFINFO(ifp)->linkmtu = ND.linkmtu;
1585 		}
1586 
1587 		if (ND.basereachable != 0) {
1588 			int obasereachable = ND_IFINFO(ifp)->basereachable;
1589 
1590 			ND_IFINFO(ifp)->basereachable = ND.basereachable;
1591 			if (ND.basereachable != obasereachable)
1592 				ND_IFINFO(ifp)->reachable =
1593 				    ND_COMPUTE_RTIME(ND.basereachable);
1594 		}
1595 		if (ND.retrans != 0)
1596 			ND_IFINFO(ifp)->retrans = ND.retrans;
1597 		if (ND.chlim != 0)
1598 			ND_IFINFO(ifp)->chlim = ND.chlim;
1599 		/* FALLTHROUGH */
1600 	case SIOCSIFINFO_FLAGS:
1601 	{
1602 		struct ifaddr *ifa;
1603 		struct in6_ifaddr *ia;
1604 
1605 		if ((ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED) &&
1606 		    !(ND.flags & ND6_IFF_IFDISABLED)) {
1607 			/* ifdisabled 1->0 transision */
1608 
1609 			/*
1610 			 * If the interface is marked as ND6_IFF_IFDISABLED and
1611 			 * has an link-local address with IN6_IFF_DUPLICATED,
1612 			 * do not clear ND6_IFF_IFDISABLED.
1613 			 * See RFC 4862, Section 5.4.5.
1614 			 */
1615 			IF_ADDR_RLOCK(ifp);
1616 			TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1617 				if (ifa->ifa_addr->sa_family != AF_INET6)
1618 					continue;
1619 				ia = (struct in6_ifaddr *)ifa;
1620 				if ((ia->ia6_flags & IN6_IFF_DUPLICATED) &&
1621 				    IN6_IS_ADDR_LINKLOCAL(IA6_IN6(ia)))
1622 					break;
1623 			}
1624 			IF_ADDR_RUNLOCK(ifp);
1625 
1626 			if (ifa != NULL) {
1627 				/* LLA is duplicated. */
1628 				ND.flags |= ND6_IFF_IFDISABLED;
1629 				log(LOG_ERR, "Cannot enable an interface"
1630 				    " with a link-local address marked"
1631 				    " duplicate.\n");
1632 			} else {
1633 				ND_IFINFO(ifp)->flags &= ~ND6_IFF_IFDISABLED;
1634 				if (ifp->if_flags & IFF_UP)
1635 					in6_if_up(ifp);
1636 			}
1637 		} else if (!(ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED) &&
1638 			    (ND.flags & ND6_IFF_IFDISABLED)) {
1639 			/* ifdisabled 0->1 transision */
1640 			/* Mark all IPv6 address as tentative. */
1641 
1642 			ND_IFINFO(ifp)->flags |= ND6_IFF_IFDISABLED;
1643 			if (V_ip6_dad_count > 0 &&
1644 			    (ND_IFINFO(ifp)->flags & ND6_IFF_NO_DAD) == 0) {
1645 				IF_ADDR_RLOCK(ifp);
1646 				TAILQ_FOREACH(ifa, &ifp->if_addrhead,
1647 				    ifa_link) {
1648 					if (ifa->ifa_addr->sa_family !=
1649 					    AF_INET6)
1650 						continue;
1651 					ia = (struct in6_ifaddr *)ifa;
1652 					ia->ia6_flags |= IN6_IFF_TENTATIVE;
1653 				}
1654 				IF_ADDR_RUNLOCK(ifp);
1655 			}
1656 		}
1657 
1658 		if (ND.flags & ND6_IFF_AUTO_LINKLOCAL) {
1659 			if (!(ND_IFINFO(ifp)->flags & ND6_IFF_AUTO_LINKLOCAL)) {
1660 				/* auto_linklocal 0->1 transision */
1661 
1662 				/* If no link-local address on ifp, configure */
1663 				ND_IFINFO(ifp)->flags |= ND6_IFF_AUTO_LINKLOCAL;
1664 				in6_ifattach(ifp, NULL);
1665 			} else if (!(ND.flags & ND6_IFF_IFDISABLED) &&
1666 			    ifp->if_flags & IFF_UP) {
1667 				/*
1668 				 * When the IF already has
1669 				 * ND6_IFF_AUTO_LINKLOCAL, no link-local
1670 				 * address is assigned, and IFF_UP, try to
1671 				 * assign one.
1672 				 */
1673 				IF_ADDR_RLOCK(ifp);
1674 				TAILQ_FOREACH(ifa, &ifp->if_addrhead,
1675 				    ifa_link) {
1676 					if (ifa->ifa_addr->sa_family !=
1677 					    AF_INET6)
1678 						continue;
1679 					ia = (struct in6_ifaddr *)ifa;
1680 					if (IN6_IS_ADDR_LINKLOCAL(IA6_IN6(ia)))
1681 						break;
1682 				}
1683 				IF_ADDR_RUNLOCK(ifp);
1684 				if (ifa != NULL)
1685 					/* No LLA is configured. */
1686 					in6_ifattach(ifp, NULL);
1687 			}
1688 		}
1689 	}
1690 		ND_IFINFO(ifp)->flags = ND.flags;
1691 		break;
1692 #undef ND
1693 	case SIOCSNDFLUSH_IN6:	/* XXX: the ioctl name is confusing... */
1694 		/* sync kernel routing table with the default router list */
1695 		defrouter_reset();
1696 		defrouter_select();
1697 		break;
1698 	case SIOCSPFXFLUSH_IN6:
1699 	{
1700 		/* flush all the prefix advertised by routers */
1701 		struct nd_prefix *pr, *next;
1702 
1703 		LIST_FOREACH_SAFE(pr, &V_nd_prefix, ndpr_entry, next) {
1704 			struct in6_ifaddr *ia, *ia_next;
1705 
1706 			if (IN6_IS_ADDR_LINKLOCAL(&pr->ndpr_prefix.sin6_addr))
1707 				continue; /* XXX */
1708 
1709 			/* do we really have to remove addresses as well? */
1710 			/* XXXRW: in6_ifaddrhead locking. */
1711 			TAILQ_FOREACH_SAFE(ia, &V_in6_ifaddrhead, ia_link,
1712 			    ia_next) {
1713 				if ((ia->ia6_flags & IN6_IFF_AUTOCONF) == 0)
1714 					continue;
1715 
1716 				if (ia->ia6_ndpr == pr)
1717 					in6_purgeaddr(&ia->ia_ifa);
1718 			}
1719 			prelist_remove(pr);
1720 		}
1721 		break;
1722 	}
1723 	case SIOCSRTRFLUSH_IN6:
1724 	{
1725 		/* flush all the default routers */
1726 		struct nd_defrouter *dr, *next;
1727 
1728 		defrouter_reset();
1729 		TAILQ_FOREACH_SAFE(dr, &V_nd_defrouter, dr_entry, next) {
1730 			defrtrlist_del(dr);
1731 		}
1732 		defrouter_select();
1733 		break;
1734 	}
1735 	case SIOCGNBRINFO_IN6:
1736 	{
1737 		struct llentry *ln;
1738 		struct in6_addr nb_addr = nbi->addr; /* make local for safety */
1739 
1740 		if ((error = in6_setscope(&nb_addr, ifp, NULL)) != 0)
1741 			return (error);
1742 
1743 		IF_AFDATA_RLOCK(ifp);
1744 		ln = nd6_lookup(&nb_addr, 0, ifp);
1745 		IF_AFDATA_RUNLOCK(ifp);
1746 
1747 		if (ln == NULL) {
1748 			error = EINVAL;
1749 			break;
1750 		}
1751 		nbi->state = ln->ln_state;
1752 		nbi->asked = ln->la_asked;
1753 		nbi->isrouter = ln->ln_router;
1754 		if (ln->la_expire == 0)
1755 			nbi->expire = 0;
1756 		else
1757 			nbi->expire = ln->la_expire + ln->lle_remtime / hz +
1758 			    (time_second - time_uptime);
1759 		LLE_RUNLOCK(ln);
1760 		break;
1761 	}
1762 	case SIOCGDEFIFACE_IN6:	/* XXX: should be implemented as a sysctl? */
1763 		ndif->ifindex = V_nd6_defifindex;
1764 		break;
1765 	case SIOCSDEFIFACE_IN6:	/* XXX: should be implemented as a sysctl? */
1766 		return (nd6_setdefaultiface(ndif->ifindex));
1767 	}
1768 	return (error);
1769 }
1770 
1771 /*
1772  * Calculates new isRouter value based on provided parameters and
1773  * returns it.
1774  */
1775 static int
1776 nd6_is_router(int type, int code, int is_new, int old_addr, int new_addr,
1777     int ln_router)
1778 {
1779 
1780 	/*
1781 	 * ICMP6 type dependent behavior.
1782 	 *
1783 	 * NS: clear IsRouter if new entry
1784 	 * RS: clear IsRouter
1785 	 * RA: set IsRouter if there's lladdr
1786 	 * redir: clear IsRouter if new entry
1787 	 *
1788 	 * RA case, (1):
1789 	 * The spec says that we must set IsRouter in the following cases:
1790 	 * - If lladdr exist, set IsRouter.  This means (1-5).
1791 	 * - If it is old entry (!newentry), set IsRouter.  This means (7).
1792 	 * So, based on the spec, in (1-5) and (7) cases we must set IsRouter.
1793 	 * A quetion arises for (1) case.  (1) case has no lladdr in the
1794 	 * neighbor cache, this is similar to (6).
1795 	 * This case is rare but we figured that we MUST NOT set IsRouter.
1796 	 *
1797 	 *   is_new  old_addr new_addr 	    NS  RS  RA	redir
1798 	 *							D R
1799 	 *	0	n	n	(1)	c   ?     s
1800 	 *	0	y	n	(2)	c   s     s
1801 	 *	0	n	y	(3)	c   s     s
1802 	 *	0	y	y	(4)	c   s     s
1803 	 *	0	y	y	(5)	c   s     s
1804 	 *	1	--	n	(6) c	c	c s
1805 	 *	1	--	y	(7) c	c   s	c s
1806 	 *
1807 	 *					(c=clear s=set)
1808 	 */
1809 	switch (type & 0xff) {
1810 	case ND_NEIGHBOR_SOLICIT:
1811 		/*
1812 		 * New entry must have is_router flag cleared.
1813 		 */
1814 		if (is_new)					/* (6-7) */
1815 			ln_router = 0;
1816 		break;
1817 	case ND_REDIRECT:
1818 		/*
1819 		 * If the icmp is a redirect to a better router, always set the
1820 		 * is_router flag.  Otherwise, if the entry is newly created,
1821 		 * clear the flag.  [RFC 2461, sec 8.3]
1822 		 */
1823 		if (code == ND_REDIRECT_ROUTER)
1824 			ln_router = 1;
1825 		else {
1826 			if (is_new)				/* (6-7) */
1827 				ln_router = 0;
1828 		}
1829 		break;
1830 	case ND_ROUTER_SOLICIT:
1831 		/*
1832 		 * is_router flag must always be cleared.
1833 		 */
1834 		ln_router = 0;
1835 		break;
1836 	case ND_ROUTER_ADVERT:
1837 		/*
1838 		 * Mark an entry with lladdr as a router.
1839 		 */
1840 		if ((!is_new && (old_addr || new_addr)) ||	/* (2-5) */
1841 		    (is_new && new_addr)) {			/* (7) */
1842 			ln_router = 1;
1843 		}
1844 		break;
1845 	}
1846 
1847 	return (ln_router);
1848 }
1849 
1850 /*
1851  * Create neighbor cache entry and cache link-layer address,
1852  * on reception of inbound ND6 packets.  (RS/RA/NS/redirect)
1853  *
1854  * type - ICMP6 type
1855  * code - type dependent information
1856  *
1857  */
1858 void
1859 nd6_cache_lladdr(struct ifnet *ifp, struct in6_addr *from, char *lladdr,
1860     int lladdrlen, int type, int code)
1861 {
1862 	struct llentry *ln = NULL, *ln_tmp;
1863 	int is_newentry;
1864 	int do_update;
1865 	int olladdr;
1866 	int llchange;
1867 	int flags;
1868 	uint16_t router = 0;
1869 	struct sockaddr_in6 sin6;
1870 	struct mbuf *chain = NULL;
1871 	u_char linkhdr[LLE_MAX_LINKHDR];
1872 	size_t linkhdrsize;
1873 	int lladdr_off;
1874 
1875 	IF_AFDATA_UNLOCK_ASSERT(ifp);
1876 
1877 	KASSERT(ifp != NULL, ("%s: ifp == NULL", __func__));
1878 	KASSERT(from != NULL, ("%s: from == NULL", __func__));
1879 
1880 	/* nothing must be updated for unspecified address */
1881 	if (IN6_IS_ADDR_UNSPECIFIED(from))
1882 		return;
1883 
1884 	/*
1885 	 * Validation about ifp->if_addrlen and lladdrlen must be done in
1886 	 * the caller.
1887 	 *
1888 	 * XXX If the link does not have link-layer adderss, what should
1889 	 * we do? (ifp->if_addrlen == 0)
1890 	 * Spec says nothing in sections for RA, RS and NA.  There's small
1891 	 * description on it in NS section (RFC 2461 7.2.3).
1892 	 */
1893 	flags = lladdr ? LLE_EXCLUSIVE : 0;
1894 	IF_AFDATA_RLOCK(ifp);
1895 	ln = nd6_lookup(from, flags, ifp);
1896 	IF_AFDATA_RUNLOCK(ifp);
1897 	is_newentry = 0;
1898 	if (ln == NULL) {
1899 		flags |= LLE_EXCLUSIVE;
1900 		ln = nd6_alloc(from, 0, ifp);
1901 		if (ln == NULL)
1902 			return;
1903 
1904 		/*
1905 		 * Since we already know all the data for the new entry,
1906 		 * fill it before insertion.
1907 		 */
1908 		if (lladdr != NULL) {
1909 			linkhdrsize = sizeof(linkhdr);
1910 			if (lltable_calc_llheader(ifp, AF_INET6, lladdr,
1911 			    linkhdr, &linkhdrsize, &lladdr_off) != 0)
1912 				return;
1913 			lltable_set_entry_addr(ifp, ln, linkhdr, linkhdrsize,
1914 			    lladdr_off);
1915 		}
1916 
1917 		IF_AFDATA_WLOCK(ifp);
1918 		LLE_WLOCK(ln);
1919 		/* Prefer any existing lle over newly-created one */
1920 		ln_tmp = nd6_lookup(from, LLE_EXCLUSIVE, ifp);
1921 		if (ln_tmp == NULL)
1922 			lltable_link_entry(LLTABLE6(ifp), ln);
1923 		IF_AFDATA_WUNLOCK(ifp);
1924 		if (ln_tmp == NULL) {
1925 			/* No existing lle, mark as new entry (6,7) */
1926 			is_newentry = 1;
1927 			nd6_llinfo_setstate(ln, ND6_LLINFO_STALE);
1928 			if (lladdr != NULL)	/* (7) */
1929 				EVENTHANDLER_INVOKE(lle_event, ln,
1930 				    LLENTRY_RESOLVED);
1931 		} else {
1932 			lltable_free_entry(LLTABLE6(ifp), ln);
1933 			ln = ln_tmp;
1934 			ln_tmp = NULL;
1935 		}
1936 	}
1937 	/* do nothing if static ndp is set */
1938 	if ((ln->la_flags & LLE_STATIC)) {
1939 		if (flags & LLE_EXCLUSIVE)
1940 			LLE_WUNLOCK(ln);
1941 		else
1942 			LLE_RUNLOCK(ln);
1943 		return;
1944 	}
1945 
1946 	olladdr = (ln->la_flags & LLE_VALID) ? 1 : 0;
1947 	if (olladdr && lladdr) {
1948 		llchange = bcmp(lladdr, ln->ll_addr,
1949 		    ifp->if_addrlen);
1950 	} else if (!olladdr && lladdr)
1951 		llchange = 1;
1952 	else
1953 		llchange = 0;
1954 
1955 	/*
1956 	 * newentry olladdr  lladdr  llchange	(*=record)
1957 	 *	0	n	n	--	(1)
1958 	 *	0	y	n	--	(2)
1959 	 *	0	n	y	y	(3) * STALE
1960 	 *	0	y	y	n	(4) *
1961 	 *	0	y	y	y	(5) * STALE
1962 	 *	1	--	n	--	(6)   NOSTATE(= PASSIVE)
1963 	 *	1	--	y	--	(7) * STALE
1964 	 */
1965 
1966 	do_update = 0;
1967 	if (is_newentry == 0 && llchange != 0) {
1968 		do_update = 1;	/* (3,5) */
1969 
1970 		/*
1971 		 * Record source link-layer address
1972 		 * XXX is it dependent to ifp->if_type?
1973 		 */
1974 		linkhdrsize = sizeof(linkhdr);
1975 		if (lltable_calc_llheader(ifp, AF_INET6, lladdr,
1976 		    linkhdr, &linkhdrsize, &lladdr_off) != 0)
1977 			return;
1978 
1979 		if (lltable_try_set_entry_addr(ifp, ln, linkhdr, linkhdrsize,
1980 		    lladdr_off) == 0) {
1981 			/* Entry was deleted */
1982 			return;
1983 		}
1984 
1985 		nd6_llinfo_setstate(ln, ND6_LLINFO_STALE);
1986 
1987 		EVENTHANDLER_INVOKE(lle_event, ln, LLENTRY_RESOLVED);
1988 
1989 		if (ln->la_hold != NULL)
1990 			nd6_grab_holdchain(ln, &chain, &sin6);
1991 	}
1992 
1993 	/* Calculates new router status */
1994 	router = nd6_is_router(type, code, is_newentry, olladdr,
1995 	    lladdr != NULL ? 1 : 0, ln->ln_router);
1996 
1997 	ln->ln_router = router;
1998 	/* Mark non-router redirects with special flag */
1999 	if ((type & 0xFF) == ND_REDIRECT && code != ND_REDIRECT_ROUTER)
2000 		ln->la_flags |= LLE_REDIRECT;
2001 
2002 	if (flags & LLE_EXCLUSIVE)
2003 		LLE_WUNLOCK(ln);
2004 	else
2005 		LLE_RUNLOCK(ln);
2006 
2007 	if (chain != NULL)
2008 		nd6_flush_holdchain(ifp, ifp, chain, &sin6);
2009 
2010 	/*
2011 	 * When the link-layer address of a router changes, select the
2012 	 * best router again.  In particular, when the neighbor entry is newly
2013 	 * created, it might affect the selection policy.
2014 	 * Question: can we restrict the first condition to the "is_newentry"
2015 	 * case?
2016 	 * XXX: when we hear an RA from a new router with the link-layer
2017 	 * address option, defrouter_select() is called twice, since
2018 	 * defrtrlist_update called the function as well.  However, I believe
2019 	 * we can compromise the overhead, since it only happens the first
2020 	 * time.
2021 	 * XXX: although defrouter_select() should not have a bad effect
2022 	 * for those are not autoconfigured hosts, we explicitly avoid such
2023 	 * cases for safety.
2024 	 */
2025 	if ((do_update || is_newentry) && router &&
2026 	    ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) {
2027 		/*
2028 		 * guaranteed recursion
2029 		 */
2030 		defrouter_select();
2031 	}
2032 }
2033 
2034 static void
2035 nd6_slowtimo(void *arg)
2036 {
2037 	CURVNET_SET((struct vnet *) arg);
2038 	struct nd_ifinfo *nd6if;
2039 	struct ifnet *ifp;
2040 
2041 	callout_reset(&V_nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
2042 	    nd6_slowtimo, curvnet);
2043 	IFNET_RLOCK_NOSLEEP();
2044 	TAILQ_FOREACH(ifp, &V_ifnet, if_link) {
2045 		if (ifp->if_afdata[AF_INET6] == NULL)
2046 			continue;
2047 		nd6if = ND_IFINFO(ifp);
2048 		if (nd6if->basereachable && /* already initialized */
2049 		    (nd6if->recalctm -= ND6_SLOWTIMER_INTERVAL) <= 0) {
2050 			/*
2051 			 * Since reachable time rarely changes by router
2052 			 * advertisements, we SHOULD insure that a new random
2053 			 * value gets recomputed at least once every few hours.
2054 			 * (RFC 2461, 6.3.4)
2055 			 */
2056 			nd6if->recalctm = V_nd6_recalc_reachtm_interval;
2057 			nd6if->reachable = ND_COMPUTE_RTIME(nd6if->basereachable);
2058 		}
2059 	}
2060 	IFNET_RUNLOCK_NOSLEEP();
2061 	CURVNET_RESTORE();
2062 }
2063 
2064 void
2065 nd6_grab_holdchain(struct llentry *ln, struct mbuf **chain,
2066     struct sockaddr_in6 *sin6)
2067 {
2068 
2069 	LLE_WLOCK_ASSERT(ln);
2070 
2071 	*chain = ln->la_hold;
2072 	ln->la_hold = NULL;
2073 	lltable_fill_sa_entry(ln, (struct sockaddr *)sin6);
2074 
2075 	if (ln->ln_state == ND6_LLINFO_STALE) {
2076 
2077 		/*
2078 		 * The first time we send a packet to a
2079 		 * neighbor whose entry is STALE, we have
2080 		 * to change the state to DELAY and a sets
2081 		 * a timer to expire in DELAY_FIRST_PROBE_TIME
2082 		 * seconds to ensure do neighbor unreachability
2083 		 * detection on expiration.
2084 		 * (RFC 2461 7.3.3)
2085 		 */
2086 		nd6_llinfo_setstate(ln, ND6_LLINFO_DELAY);
2087 	}
2088 }
2089 
2090 int
2091 nd6_output_ifp(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *m,
2092     struct sockaddr_in6 *dst, struct route *ro)
2093 {
2094 	int error;
2095 	int ip6len;
2096 	struct ip6_hdr *ip6;
2097 	struct m_tag *mtag;
2098 
2099 #ifdef MAC
2100 	mac_netinet6_nd6_send(ifp, m);
2101 #endif
2102 
2103 	/*
2104 	 * If called from nd6_ns_output() (NS), nd6_na_output() (NA),
2105 	 * icmp6_redirect_output() (REDIRECT) or from rip6_output() (RS, RA
2106 	 * as handled by rtsol and rtadvd), mbufs will be tagged for SeND
2107 	 * to be diverted to user space.  When re-injected into the kernel,
2108 	 * send_output() will directly dispatch them to the outgoing interface.
2109 	 */
2110 	if (send_sendso_input_hook != NULL) {
2111 		mtag = m_tag_find(m, PACKET_TAG_ND_OUTGOING, NULL);
2112 		if (mtag != NULL) {
2113 			ip6 = mtod(m, struct ip6_hdr *);
2114 			ip6len = sizeof(struct ip6_hdr) + ntohs(ip6->ip6_plen);
2115 			/* Use the SEND socket */
2116 			error = send_sendso_input_hook(m, ifp, SND_OUT,
2117 			    ip6len);
2118 			/* -1 == no app on SEND socket */
2119 			if (error == 0 || error != -1)
2120 			    return (error);
2121 		}
2122 	}
2123 
2124 	m_clrprotoflags(m);	/* Avoid confusing lower layers. */
2125 	IP_PROBE(send, NULL, NULL, mtod(m, struct ip6_hdr *), ifp, NULL,
2126 	    mtod(m, struct ip6_hdr *));
2127 
2128 	if ((ifp->if_flags & IFF_LOOPBACK) == 0)
2129 		origifp = ifp;
2130 
2131 	error = (*ifp->if_output)(origifp, m, (struct sockaddr *)dst, ro);
2132 	return (error);
2133 }
2134 
2135 /*
2136  * Lookup link headerfor @sa_dst address. Stores found
2137  * data in @desten buffer. Copy of lle ln_flags can be also
2138  * saved in @pflags if @pflags is non-NULL.
2139  *
2140  * If destination LLE does not exists or lle state modification
2141  * is required, call "slow" version.
2142  *
2143  * Return values:
2144  * - 0 on success (address copied to buffer).
2145  * - EWOULDBLOCK (no local error, but address is still unresolved)
2146  * - other errors (alloc failure, etc)
2147  */
2148 int
2149 nd6_resolve(struct ifnet *ifp, int is_gw, struct mbuf *m,
2150     const struct sockaddr *sa_dst, u_char *desten, uint32_t *pflags)
2151 {
2152 	struct llentry *ln = NULL;
2153 	const struct sockaddr_in6 *dst6;
2154 
2155 	if (pflags != NULL)
2156 		*pflags = 0;
2157 
2158 	dst6 = (const struct sockaddr_in6 *)sa_dst;
2159 
2160 	/* discard the packet if IPv6 operation is disabled on the interface */
2161 	if ((ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED)) {
2162 		m_freem(m);
2163 		return (ENETDOWN); /* better error? */
2164 	}
2165 
2166 	if (m != NULL && m->m_flags & M_MCAST) {
2167 		switch (ifp->if_type) {
2168 		case IFT_ETHER:
2169 		case IFT_FDDI:
2170 		case IFT_L2VLAN:
2171 		case IFT_IEEE80211:
2172 		case IFT_BRIDGE:
2173 		case IFT_ISO88025:
2174 			ETHER_MAP_IPV6_MULTICAST(&dst6->sin6_addr,
2175 						 desten);
2176 			return (0);
2177 		default:
2178 			m_freem(m);
2179 			return (EAFNOSUPPORT);
2180 		}
2181 	}
2182 
2183 	IF_AFDATA_RLOCK(ifp);
2184 	ln = nd6_lookup(&dst6->sin6_addr, LLE_UNLOCKED, ifp);
2185 	if (ln != NULL && (ln->r_flags & RLLE_VALID) != 0) {
2186 		/* Entry found, let's copy lle info */
2187 		bcopy(ln->r_linkdata, desten, ln->r_hdrlen);
2188 		if (pflags != NULL)
2189 			*pflags = LLE_VALID | (ln->r_flags & RLLE_IFADDR);
2190 		/* Check if we have feedback request from nd6 timer */
2191 		if (ln->r_skip_req != 0) {
2192 			LLE_REQ_LOCK(ln);
2193 			ln->r_skip_req = 0; /* Notify that entry was used */
2194 			ln->lle_hittime = time_uptime;
2195 			LLE_REQ_UNLOCK(ln);
2196 		}
2197 		IF_AFDATA_RUNLOCK(ifp);
2198 		return (0);
2199 	}
2200 	IF_AFDATA_RUNLOCK(ifp);
2201 
2202 	return (nd6_resolve_slow(ifp, 0, m, dst6, desten, pflags));
2203 }
2204 
2205 
2206 /*
2207  * Do L2 address resolution for @sa_dst address. Stores found
2208  * address in @desten buffer. Copy of lle ln_flags can be also
2209  * saved in @pflags if @pflags is non-NULL.
2210  *
2211  * Heavy version.
2212  * Function assume that destination LLE does not exist,
2213  * is invalid or stale, so LLE_EXCLUSIVE lock needs to be acquired.
2214  *
2215  * Set noinline to be dtrace-friendly
2216  */
2217 static __noinline int
2218 nd6_resolve_slow(struct ifnet *ifp, int flags, struct mbuf *m,
2219     const struct sockaddr_in6 *dst, u_char *desten, uint32_t *pflags)
2220 {
2221 	struct llentry *lle = NULL, *lle_tmp;
2222 	struct in6_addr *psrc, src;
2223 	int send_ns, ll_len;
2224 	char *lladdr;
2225 
2226 	/*
2227 	 * Address resolution or Neighbor Unreachability Detection
2228 	 * for the next hop.
2229 	 * At this point, the destination of the packet must be a unicast
2230 	 * or an anycast address(i.e. not a multicast).
2231 	 */
2232 	if (lle == NULL) {
2233 		IF_AFDATA_RLOCK(ifp);
2234 		lle = nd6_lookup(&dst->sin6_addr, LLE_EXCLUSIVE, ifp);
2235 		IF_AFDATA_RUNLOCK(ifp);
2236 		if ((lle == NULL) && nd6_is_addr_neighbor(dst, ifp))  {
2237 			/*
2238 			 * Since nd6_is_addr_neighbor() internally calls nd6_lookup(),
2239 			 * the condition below is not very efficient.  But we believe
2240 			 * it is tolerable, because this should be a rare case.
2241 			 */
2242 			lle = nd6_alloc(&dst->sin6_addr, 0, ifp);
2243 			if (lle == NULL) {
2244 				char ip6buf[INET6_ADDRSTRLEN];
2245 				log(LOG_DEBUG,
2246 				    "nd6_output: can't allocate llinfo for %s "
2247 				    "(ln=%p)\n",
2248 				    ip6_sprintf(ip6buf, &dst->sin6_addr), lle);
2249 				m_freem(m);
2250 				return (ENOBUFS);
2251 			}
2252 
2253 			IF_AFDATA_WLOCK(ifp);
2254 			LLE_WLOCK(lle);
2255 			/* Prefer any existing entry over newly-created one */
2256 			lle_tmp = nd6_lookup(&dst->sin6_addr, LLE_EXCLUSIVE, ifp);
2257 			if (lle_tmp == NULL)
2258 				lltable_link_entry(LLTABLE6(ifp), lle);
2259 			IF_AFDATA_WUNLOCK(ifp);
2260 			if (lle_tmp != NULL) {
2261 				lltable_free_entry(LLTABLE6(ifp), lle);
2262 				lle = lle_tmp;
2263 				lle_tmp = NULL;
2264 			}
2265 		}
2266 	}
2267 	if (lle == NULL) {
2268 		if (!(ND_IFINFO(ifp)->flags & ND6_IFF_PERFORMNUD)) {
2269 			m_freem(m);
2270 			return (ENOBUFS);
2271 		}
2272 
2273 		if (m != NULL)
2274 			m_freem(m);
2275 		return (ENOBUFS);
2276 	}
2277 
2278 	LLE_WLOCK_ASSERT(lle);
2279 
2280 	/*
2281 	 * The first time we send a packet to a neighbor whose entry is
2282 	 * STALE, we have to change the state to DELAY and a sets a timer to
2283 	 * expire in DELAY_FIRST_PROBE_TIME seconds to ensure do
2284 	 * neighbor unreachability detection on expiration.
2285 	 * (RFC 2461 7.3.3)
2286 	 */
2287 	if (lle->ln_state == ND6_LLINFO_STALE)
2288 		nd6_llinfo_setstate(lle, ND6_LLINFO_DELAY);
2289 
2290 	/*
2291 	 * If the neighbor cache entry has a state other than INCOMPLETE
2292 	 * (i.e. its link-layer address is already resolved), just
2293 	 * send the packet.
2294 	 */
2295 	if (lle->ln_state > ND6_LLINFO_INCOMPLETE) {
2296 		if (flags & LLE_ADDRONLY) {
2297 			lladdr = lle->ll_addr;
2298 			ll_len = ifp->if_addrlen;
2299 		} else {
2300 			lladdr = lle->r_linkdata;
2301 			ll_len = lle->r_hdrlen;
2302 		}
2303 		bcopy(lladdr, desten, ll_len);
2304 		if (pflags != NULL)
2305 			*pflags = lle->la_flags;
2306 		LLE_WUNLOCK(lle);
2307 		return (0);
2308 	}
2309 
2310 	/*
2311 	 * There is a neighbor cache entry, but no ethernet address
2312 	 * response yet.  Append this latest packet to the end of the
2313 	 * packet queue in the mbuf, unless the number of the packet
2314 	 * does not exceed nd6_maxqueuelen.  When it exceeds nd6_maxqueuelen,
2315 	 * the oldest packet in the queue will be removed.
2316 	 */
2317 
2318 	if (lle->la_hold != NULL) {
2319 		struct mbuf *m_hold;
2320 		int i;
2321 
2322 		i = 0;
2323 		for (m_hold = lle->la_hold; m_hold; m_hold = m_hold->m_nextpkt){
2324 			i++;
2325 			if (m_hold->m_nextpkt == NULL) {
2326 				m_hold->m_nextpkt = m;
2327 				break;
2328 			}
2329 		}
2330 		while (i >= V_nd6_maxqueuelen) {
2331 			m_hold = lle->la_hold;
2332 			lle->la_hold = lle->la_hold->m_nextpkt;
2333 			m_freem(m_hold);
2334 			i--;
2335 		}
2336 	} else {
2337 		lle->la_hold = m;
2338 	}
2339 
2340 	/*
2341 	 * If there has been no NS for the neighbor after entering the
2342 	 * INCOMPLETE state, send the first solicitation.
2343 	 * Note that for newly-created lle la_asked will be 0,
2344 	 * so we will transition from ND6_LLINFO_NOSTATE to
2345 	 * ND6_LLINFO_INCOMPLETE state here.
2346 	 */
2347 	psrc = NULL;
2348 	send_ns = 0;
2349 	if (lle->la_asked == 0) {
2350 		lle->la_asked++;
2351 		send_ns = 1;
2352 		psrc = nd6_llinfo_get_holdsrc(lle, &src);
2353 
2354 		nd6_llinfo_setstate(lle, ND6_LLINFO_INCOMPLETE);
2355 	}
2356 	LLE_WUNLOCK(lle);
2357 	if (send_ns != 0)
2358 		nd6_ns_output(ifp, psrc, NULL, &dst->sin6_addr, NULL);
2359 
2360 	return (EWOULDBLOCK);
2361 }
2362 
2363 /*
2364  * Do L2 address resolution for @sa_dst address. Stores found
2365  * address in @desten buffer. Copy of lle ln_flags can be also
2366  * saved in @pflags if @pflags is non-NULL.
2367  *
2368  * Return values:
2369  * - 0 on success (address copied to buffer).
2370  * - EWOULDBLOCK (no local error, but address is still unresolved)
2371  * - other errors (alloc failure, etc)
2372  */
2373 int
2374 nd6_resolve_addr(struct ifnet *ifp, int flags, const struct sockaddr *dst,
2375     char *desten, uint32_t *pflags)
2376 {
2377 	int error;
2378 
2379 	flags |= LLE_ADDRONLY;
2380 	error = nd6_resolve_slow(ifp, flags, NULL,
2381 	    (const struct sockaddr_in6 *)dst, desten, pflags);
2382 	return (error);
2383 }
2384 
2385 int
2386 nd6_flush_holdchain(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *chain,
2387     struct sockaddr_in6 *dst)
2388 {
2389 	struct mbuf *m, *m_head;
2390 	struct ifnet *outifp;
2391 	int error = 0;
2392 
2393 	m_head = chain;
2394 	if ((ifp->if_flags & IFF_LOOPBACK) != 0)
2395 		outifp = origifp;
2396 	else
2397 		outifp = ifp;
2398 
2399 	while (m_head) {
2400 		m = m_head;
2401 		m_head = m_head->m_nextpkt;
2402 		error = nd6_output_ifp(ifp, origifp, m, dst, NULL);
2403 	}
2404 
2405 	/*
2406 	 * XXX
2407 	 * note that intermediate errors are blindly ignored
2408 	 */
2409 	return (error);
2410 }
2411 
2412 static int
2413 nd6_need_cache(struct ifnet *ifp)
2414 {
2415 	/*
2416 	 * XXX: we currently do not make neighbor cache on any interface
2417 	 * other than ARCnet, Ethernet, FDDI and GIF.
2418 	 *
2419 	 * RFC2893 says:
2420 	 * - unidirectional tunnels needs no ND
2421 	 */
2422 	switch (ifp->if_type) {
2423 	case IFT_ARCNET:
2424 	case IFT_ETHER:
2425 	case IFT_FDDI:
2426 	case IFT_IEEE1394:
2427 	case IFT_L2VLAN:
2428 	case IFT_IEEE80211:
2429 	case IFT_INFINIBAND:
2430 	case IFT_BRIDGE:
2431 	case IFT_PROPVIRTUAL:
2432 		return (1);
2433 	default:
2434 		return (0);
2435 	}
2436 }
2437 
2438 /*
2439  * Add pernament ND6 link-layer record for given
2440  * interface address.
2441  *
2442  * Very similar to IPv4 arp_ifinit(), but:
2443  * 1) IPv6 DAD is performed in different place
2444  * 2) It is called by IPv6 protocol stack in contrast to
2445  * arp_ifinit() which is typically called in SIOCSIFADDR
2446  * driver ioctl handler.
2447  *
2448  */
2449 int
2450 nd6_add_ifa_lle(struct in6_ifaddr *ia)
2451 {
2452 	struct ifnet *ifp;
2453 	struct llentry *ln, *ln_tmp;
2454 	struct sockaddr *dst;
2455 
2456 	ifp = ia->ia_ifa.ifa_ifp;
2457 	if (nd6_need_cache(ifp) == 0)
2458 		return (0);
2459 
2460 	ia->ia_ifa.ifa_rtrequest = nd6_rtrequest;
2461 	dst = (struct sockaddr *)&ia->ia_addr;
2462 	ln = lltable_alloc_entry(LLTABLE6(ifp), LLE_IFADDR, dst);
2463 	if (ln == NULL)
2464 		return (ENOBUFS);
2465 
2466 	IF_AFDATA_WLOCK(ifp);
2467 	LLE_WLOCK(ln);
2468 	/* Unlink any entry if exists */
2469 	ln_tmp = lla_lookup(LLTABLE6(ifp), LLE_EXCLUSIVE, dst);
2470 	if (ln_tmp != NULL)
2471 		lltable_unlink_entry(LLTABLE6(ifp), ln_tmp);
2472 	lltable_link_entry(LLTABLE6(ifp), ln);
2473 	IF_AFDATA_WUNLOCK(ifp);
2474 
2475 	if (ln_tmp != NULL)
2476 		EVENTHANDLER_INVOKE(lle_event, ln_tmp, LLENTRY_EXPIRED);
2477 	EVENTHANDLER_INVOKE(lle_event, ln, LLENTRY_RESOLVED);
2478 
2479 	LLE_WUNLOCK(ln);
2480 	if (ln_tmp != NULL)
2481 		llentry_free(ln_tmp);
2482 
2483 	return (0);
2484 }
2485 
2486 /*
2487  * Removes either all lle entries for given @ia, or lle
2488  * corresponding to @ia address.
2489  */
2490 void
2491 nd6_rem_ifa_lle(struct in6_ifaddr *ia, int all)
2492 {
2493 	struct sockaddr_in6 mask, addr;
2494 	struct sockaddr *saddr, *smask;
2495 	struct ifnet *ifp;
2496 
2497 	ifp = ia->ia_ifa.ifa_ifp;
2498 	memcpy(&addr, &ia->ia_addr, sizeof(ia->ia_addr));
2499 	memcpy(&mask, &ia->ia_prefixmask, sizeof(ia->ia_prefixmask));
2500 	saddr = (struct sockaddr *)&addr;
2501 	smask = (struct sockaddr *)&mask;
2502 
2503 	if (all != 0)
2504 		lltable_prefix_free(AF_INET6, saddr, smask, LLE_STATIC);
2505 	else
2506 		lltable_delete_addr(LLTABLE6(ifp), LLE_IFADDR, saddr);
2507 }
2508 
2509 static void
2510 clear_llinfo_pqueue(struct llentry *ln)
2511 {
2512 	struct mbuf *m_hold, *m_hold_next;
2513 
2514 	for (m_hold = ln->la_hold; m_hold; m_hold = m_hold_next) {
2515 		m_hold_next = m_hold->m_nextpkt;
2516 		m_freem(m_hold);
2517 	}
2518 
2519 	ln->la_hold = NULL;
2520 	return;
2521 }
2522 
2523 static int nd6_sysctl_drlist(SYSCTL_HANDLER_ARGS);
2524 static int nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS);
2525 #ifdef SYSCTL_DECL
2526 SYSCTL_DECL(_net_inet6_icmp6);
2527 #endif
2528 SYSCTL_NODE(_net_inet6_icmp6, ICMPV6CTL_ND6_DRLIST, nd6_drlist,
2529 	CTLFLAG_RD, nd6_sysctl_drlist, "");
2530 SYSCTL_NODE(_net_inet6_icmp6, ICMPV6CTL_ND6_PRLIST, nd6_prlist,
2531 	CTLFLAG_RD, nd6_sysctl_prlist, "");
2532 SYSCTL_INT(_net_inet6_icmp6, ICMPV6CTL_ND6_MAXQLEN, nd6_maxqueuelen,
2533 	CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(nd6_maxqueuelen), 1, "");
2534 SYSCTL_INT(_net_inet6_icmp6, OID_AUTO, nd6_gctimer,
2535 	CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(nd6_gctimer), (60 * 60 * 24), "");
2536 
2537 static int
2538 nd6_sysctl_drlist(SYSCTL_HANDLER_ARGS)
2539 {
2540 	struct in6_defrouter d;
2541 	struct nd_defrouter *dr;
2542 	int error;
2543 
2544 	if (req->newptr)
2545 		return (EPERM);
2546 
2547 	bzero(&d, sizeof(d));
2548 	d.rtaddr.sin6_family = AF_INET6;
2549 	d.rtaddr.sin6_len = sizeof(d.rtaddr);
2550 
2551 	/*
2552 	 * XXX locking
2553 	 */
2554 	TAILQ_FOREACH(dr, &V_nd_defrouter, dr_entry) {
2555 		d.rtaddr.sin6_addr = dr->rtaddr;
2556 		error = sa6_recoverscope(&d.rtaddr);
2557 		if (error != 0)
2558 			return (error);
2559 		d.flags = dr->flags;
2560 		d.rtlifetime = dr->rtlifetime;
2561 		d.expire = dr->expire + (time_second - time_uptime);
2562 		d.if_index = dr->ifp->if_index;
2563 		error = SYSCTL_OUT(req, &d, sizeof(d));
2564 		if (error != 0)
2565 			return (error);
2566 	}
2567 	return (0);
2568 }
2569 
2570 static int
2571 nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS)
2572 {
2573 	struct in6_prefix p;
2574 	struct sockaddr_in6 s6;
2575 	struct nd_prefix *pr;
2576 	struct nd_pfxrouter *pfr;
2577 	time_t maxexpire;
2578 	int error;
2579 	char ip6buf[INET6_ADDRSTRLEN];
2580 
2581 	if (req->newptr)
2582 		return (EPERM);
2583 
2584 	bzero(&p, sizeof(p));
2585 	p.origin = PR_ORIG_RA;
2586 	bzero(&s6, sizeof(s6));
2587 	s6.sin6_family = AF_INET6;
2588 	s6.sin6_len = sizeof(s6);
2589 
2590 	/*
2591 	 * XXX locking
2592 	 */
2593 	LIST_FOREACH(pr, &V_nd_prefix, ndpr_entry) {
2594 		p.prefix = pr->ndpr_prefix;
2595 		if (sa6_recoverscope(&p.prefix)) {
2596 			log(LOG_ERR, "scope error in prefix list (%s)\n",
2597 			    ip6_sprintf(ip6buf, &p.prefix.sin6_addr));
2598 			/* XXX: press on... */
2599 		}
2600 		p.raflags = pr->ndpr_raf;
2601 		p.prefixlen = pr->ndpr_plen;
2602 		p.vltime = pr->ndpr_vltime;
2603 		p.pltime = pr->ndpr_pltime;
2604 		p.if_index = pr->ndpr_ifp->if_index;
2605 		if (pr->ndpr_vltime == ND6_INFINITE_LIFETIME)
2606 			p.expire = 0;
2607 		else {
2608 			/* XXX: we assume time_t is signed. */
2609 			maxexpire = (-1) &
2610 			    ~((time_t)1 << ((sizeof(maxexpire) * 8) - 1));
2611 			if (pr->ndpr_vltime < maxexpire - pr->ndpr_lastupdate)
2612 				p.expire = pr->ndpr_lastupdate +
2613 				    pr->ndpr_vltime +
2614 				    (time_second - time_uptime);
2615 			else
2616 				p.expire = maxexpire;
2617 		}
2618 		p.refcnt = pr->ndpr_refcnt;
2619 		p.flags = pr->ndpr_stateflags;
2620 		p.advrtrs = 0;
2621 		LIST_FOREACH(pfr, &pr->ndpr_advrtrs, pfr_entry)
2622 			p.advrtrs++;
2623 		error = SYSCTL_OUT(req, &p, sizeof(p));
2624 		if (error != 0)
2625 			return (error);
2626 		LIST_FOREACH(pfr, &pr->ndpr_advrtrs, pfr_entry) {
2627 			s6.sin6_addr = pfr->router->rtaddr;
2628 			if (sa6_recoverscope(&s6))
2629 				log(LOG_ERR,
2630 				    "scope error in prefix list (%s)\n",
2631 				    ip6_sprintf(ip6buf, &pfr->router->rtaddr));
2632 			error = SYSCTL_OUT(req, &s6, sizeof(s6));
2633 			if (error != 0)
2634 				return (error);
2635 		}
2636 	}
2637 	return (0);
2638 }
2639