xref: /dragonfly/sys/netinet6/nd6.c (revision d256b490)
1 /*	$FreeBSD: src/sys/netinet6/nd6.c,v 1.2.2.15 2003/05/06 06:46:58 suz Exp $	*/
2 /*	$KAME: nd6.c,v 1.144 2001/05/24 07:44:00 itojun Exp $	*/
3 
4 /*
5  * Copyright (C) 1995, 1996, 1997, and 1998 WIDE Project.
6  * All rights reserved.
7  *
8  * Redistribution and use in source and binary forms, with or without
9  * modification, are permitted provided that the following conditions
10  * are met:
11  * 1. Redistributions of source code must retain the above copyright
12  *    notice, this list of conditions and the following disclaimer.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in the
15  *    documentation and/or other materials provided with the distribution.
16  * 3. Neither the name of the project nor the names of its contributors
17  *    may be used to endorse or promote products derived from this software
18  *    without specific prior written permission.
19  *
20  * THIS SOFTWARE IS PROVIDED BY THE PROJECT AND CONTRIBUTORS ``AS IS'' AND
21  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
22  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
23  * ARE DISCLAIMED.  IN NO EVENT SHALL THE PROJECT OR CONTRIBUTORS BE LIABLE
24  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
25  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
26  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
27  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
28  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
29  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
30  * SUCH DAMAGE.
31  */
32 
33 #include "opt_inet.h"
34 #include "opt_inet6.h"
35 
36 #include <sys/param.h>
37 #include <sys/systm.h>
38 #include <sys/callout.h>
39 #include <sys/malloc.h>
40 #include <sys/mbuf.h>
41 #include <sys/socket.h>
42 #include <sys/sockio.h>
43 #include <sys/time.h>
44 #include <sys/kernel.h>
45 #include <sys/protosw.h>
46 #include <sys/errno.h>
47 #include <sys/syslog.h>
48 #include <sys/queue.h>
49 #include <sys/sysctl.h>
50 #include <sys/mutex.h>
51 
52 #include <sys/thread2.h>
53 #include <sys/mutex2.h>
54 
55 #include <net/if.h>
56 #include <net/if_dl.h>
57 #include <net/if_types.h>
58 #include <net/route.h>
59 #include <net/netisr2.h>
60 #include <net/netmsg2.h>
61 
62 #include <netinet/in.h>
63 #include <netinet/if_ether.h>
64 #include <netinet6/in6_var.h>
65 #include <netinet6/in6_ifattach.h>
66 #include <netinet/ip6.h>
67 #include <netinet6/ip6_var.h>
68 #include <netinet6/nd6.h>
69 #include <netinet/icmp6.h>
70 
71 #include <net/net_osdep.h>
72 
73 #define ND6_SLOWTIMER_INTERVAL (60 * 60) /* 1 hour */
74 #define ND6_RECALC_REACHTM_INTERVAL (60 * 120) /* 2 hours */
75 
76 #define SIN6(s) ((struct sockaddr_in6 *)s)
77 #define SDL(s) ((struct sockaddr_dl *)s)
78 
79 /*
80  * Determine if the route entry is a direct neighbor on the specified
81  * interface.  The interface test is not done if ifp is passed as NULL.
82  * The route entry is a neighbor if all of the following are true:
83  *
84  *	RTF_GATEWAY is FALSE
85  *	LLINFO is TRUE
86  *	rt_gateway family is AF_LINK
87  *	rt_llinfo is non-NULL
88  *	The interfaces matches (or ifp is passed as NULL)
89  *
90  *	NOTE: rt_llinfo can be NULL with LLINFO set, so both must be
91  *	      tested.
92  *
93  *	NOTE: We can't use rt->rt_ifp to check for the interface, since
94  *	      it may be the loopback interface if the entry is for our
95  *	      own address on a non-loopback interface.  Instead, we use
96  *	      rt->rt_ifa->ifa_ifp which should specify the REAL interface.
97  */
98 #define ND6_IFP_MATCHES(ifp, ifa_ifp)				\
99 	((ifp) == NULL ||					\
100 	 (ifa_ifp) == (ifp) ||					\
101 	 (((ifp)->if_flags & IFF_ISBRIDGE) &&			\
102 	  (ifa_ifp)->if_bridge == (ifp)->if_softc)		\
103 	)
104 
105 #define ND6_RTENTRY_IS_NEIGHBOR(rt, ifp)			\
106 	(((rt)->rt_flags & RTF_GATEWAY) == 0 &&			\
107 	 ((rt)->rt_flags & RTF_LLINFO) != 0 &&			\
108 	 (rt)->rt_gateway->sa_family == AF_LINK &&		\
109 	 (rt)->rt_llinfo &&					\
110 	 ND6_IFP_MATCHES((ifp), (rt)->rt_ifa->ifa_ifp)		\
111 	)
112 
113 #define ND6_RTENTRY_IS_LLCLONING(rt)				\
114 	(((rt)->rt_flags & (RTF_PRCLONING | RTF_LLINFO)) ==	\
115 	 (RTF_PRCLONING | RTF_LLINFO) ||			\
116 	 ((rt)->rt_flags & RTF_CLONING))
117 
118 /* timer values */
119 int	nd6_prune	= 1;	/* walk list every 1 seconds */
120 int	nd6_delay	= 5;	/* delay first probe time 5 second */
121 int	nd6_umaxtries	= 3;	/* maximum unicast query */
122 int	nd6_mmaxtries	= 3;	/* maximum multicast query */
123 int	nd6_useloopback = 1;	/* use loopback interface for local traffic */
124 int	nd6_gctimer	= (60 * 60 * 24); /* 1 day: garbage collection timer */
125 
126 /* preventing too many loops in ND option parsing */
127 int nd6_maxndopt = 10;	/* max # of ND options allowed */
128 
129 int nd6_maxnudhint = 0;	/* max # of subsequent upper layer hints */
130 
131 #ifdef ND6_DEBUG
132 int nd6_debug = 1;
133 #else
134 int nd6_debug = 0;
135 #endif
136 
137 /* for debugging? */
138 static int nd6_inuse, nd6_allocated;
139 
140 struct llinfo_nd6 llinfo_nd6 = {&llinfo_nd6, &llinfo_nd6};
141 struct nd_drhead nd_defrouter;
142 struct nd_prhead nd_prefix = { 0 };
143 struct mtx nd6_mtx = MTX_INITIALIZER("nd6");
144 
145 int nd6_recalc_reachtm_interval = ND6_RECALC_REACHTM_INTERVAL;
146 static struct sockaddr_in6 all1_sa;
147 
148 static void nd6_setmtu0 (struct ifnet *, struct nd_ifinfo *);
149 static int regen_tmpaddr (struct in6_ifaddr *);
150 static void nd6_slowtimo(void *);
151 static void nd6_slowtimo_dispatch(netmsg_t);
152 static void nd6_timer(void *);
153 static void nd6_timer_dispatch(netmsg_t);
154 
155 static struct callout nd6_slowtimo_ch;
156 static struct netmsg_base nd6_slowtimo_netmsg;
157 
158 static struct callout nd6_timer_ch;
159 static struct netmsg_base nd6_timer_netmsg;
160 
161 void
nd6_init(void)162 nd6_init(void)
163 {
164 	static int nd6_init_done = 0;
165 	int i;
166 
167 	if (nd6_init_done) {
168 		log(LOG_NOTICE, "nd6_init called more than once(ignored)\n");
169 		return;
170 	}
171 
172 	all1_sa.sin6_family = AF_INET6;
173 	all1_sa.sin6_len = sizeof(struct sockaddr_in6);
174 	for (i = 0; i < sizeof(all1_sa.sin6_addr); i++)
175 		all1_sa.sin6_addr.s6_addr[i] = 0xff;
176 
177 	/* initialization of the default router list */
178 	TAILQ_INIT(&nd_defrouter);
179 
180 	nd6_init_done = 1;
181 
182 	/* start timer */
183 	callout_init_mp(&nd6_slowtimo_ch);
184 	netmsg_init(&nd6_slowtimo_netmsg, NULL, &netisr_adone_rport,
185 	    MSGF_PRIORITY, nd6_slowtimo_dispatch);
186 	callout_reset_bycpu(&nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
187 	    nd6_slowtimo, NULL, 0);
188 }
189 
190 struct nd_ifinfo *
nd6_ifattach(struct ifnet * ifp)191 nd6_ifattach(struct ifnet *ifp)
192 {
193 	struct nd_ifinfo *nd;
194 
195 	nd = kmalloc(sizeof(*nd), M_IP6NDP, M_WAITOK | M_ZERO);
196 
197 	nd->initialized = 1;
198 
199 	nd->chlim = IPV6_DEFHLIM;
200 	nd->basereachable = REACHABLE_TIME;
201 	nd->reachable = ND_COMPUTE_RTIME(nd->basereachable);
202 	nd->retrans = RETRANS_TIMER;
203 
204 	nd->flags = ND6_IFF_PERFORMNUD;
205 	/* A loopback interface always has link-local address. */
206 	if (ip6_auto_linklocal || (ifp->if_flags & IFF_LOOPBACK))
207 		nd->flags |= ND6_IFF_AUTO_LINKLOCAL;
208 	/* A loopback interface does not need to accept RAs. */
209 	if (ip6_accept_rtadv && !(ifp->if_flags & IFF_LOOPBACK))
210 		nd->flags |= ND6_IFF_ACCEPT_RTADV;
211 
212 	/* XXX: we cannot call nd6_setmtu since ifp is not fully initialized */
213 	nd6_setmtu0(ifp, nd);
214 
215 	return nd;
216 }
217 
218 void
nd6_ifdetach(struct nd_ifinfo * nd)219 nd6_ifdetach(struct nd_ifinfo *nd)
220 {
221 	kfree(nd, M_IP6NDP);
222 }
223 
224 /*
225  * Reset ND level link MTU. This function is called when the physical MTU
226  * changes, which means we might have to adjust the ND level MTU.
227  */
228 void
nd6_setmtu(struct ifnet * ifp)229 nd6_setmtu(struct ifnet *ifp)
230 {
231 	nd6_setmtu0(ifp, ND_IFINFO(ifp));
232 }
233 
234 struct netmsg_nd6setmtu {
235 	struct netmsg_base	nmsg;
236 	struct ifnet		*ifp;
237 	struct nd_ifinfo	*ndi;
238 };
239 
240 /* XXX todo: do not maintain copy of ifp->if_mtu in ndi->maxmtu */
241 static void
nd6_setmtu0_dispatch(netmsg_t msg)242 nd6_setmtu0_dispatch(netmsg_t msg)
243 {
244 	struct netmsg_nd6setmtu *nmsg = (struct netmsg_nd6setmtu *)msg;
245 	struct ifnet *ifp = nmsg->ifp;
246 	struct nd_ifinfo *ndi = nmsg->ndi;
247 	uint32_t omaxmtu;
248 
249 	omaxmtu = ndi->maxmtu;
250 
251 	switch (ifp->if_type) {
252 	case IFT_ETHER:
253 		ndi->maxmtu = MIN(ETHERMTU, ifp->if_mtu);
254 		break;
255 	case IFT_IEEE1394:	/* XXX should be IEEE1394MTU(1500) */
256 		ndi->maxmtu = MIN(ETHERMTU, ifp->if_mtu);
257 		break;
258 #ifdef IFT_IEEE80211
259 	case IFT_IEEE80211:	/* XXX should be IEEE80211MTU(1500) */
260 		ndi->maxmtu = MIN(ETHERMTU, ifp->if_mtu);
261 		break;
262 #endif
263 	default:
264 		ndi->maxmtu = ifp->if_mtu;
265 		break;
266 	}
267 
268 	/*
269 	 * Decreasing the interface MTU under IPV6 minimum MTU may cause
270 	 * undesirable situation.  We thus notify the operator of the change
271 	 * explicitly.  The check for omaxmtu is necessary to restrict the
272 	 * log to the case of changing the MTU, not initializing it.
273 	 */
274 	if (omaxmtu >= IPV6_MMTU && ndi->maxmtu < IPV6_MMTU) {
275 		log(LOG_NOTICE, "nd6_setmtu0: "
276 		    "new link MTU on %s (%lu) is too small for IPv6\n",
277 		    if_name(ifp), (unsigned long)ndi->maxmtu);
278 	}
279 
280 	if (ndi->maxmtu > in6_maxmtu)
281 		in6_setmaxmtu(); /* check all interfaces just in case */
282 
283 	lwkt_replymsg(&nmsg->nmsg.lmsg, 0);
284 }
285 
286 void
nd6_setmtu0(struct ifnet * ifp,struct nd_ifinfo * ndi)287 nd6_setmtu0(struct ifnet *ifp, struct nd_ifinfo *ndi)
288 {
289 	struct netmsg_nd6setmtu nmsg;
290 
291 	netmsg_init(&nmsg.nmsg, NULL, &curthread->td_msgport, 0,
292 	    nd6_setmtu0_dispatch);
293 	nmsg.ifp = ifp;
294 	nmsg.ndi = ndi;
295 	lwkt_domsg(netisr_cpuport(0), &nmsg.nmsg.lmsg, 0);
296 }
297 
298 void
nd6_option_init(void * opt,int icmp6len,union nd_opts * ndopts)299 nd6_option_init(void *opt, int icmp6len, union nd_opts *ndopts)
300 {
301 	bzero(ndopts, sizeof(*ndopts));
302 	ndopts->nd_opts_search = (struct nd_opt_hdr *)opt;
303 	ndopts->nd_opts_last
304 		= (struct nd_opt_hdr *)(((u_char *)opt) + icmp6len);
305 
306 	if (icmp6len == 0) {
307 		ndopts->nd_opts_done = 1;
308 		ndopts->nd_opts_search = NULL;
309 	}
310 }
311 
312 /*
313  * Take one ND option.
314  */
315 struct nd_opt_hdr *
nd6_option(union nd_opts * ndopts)316 nd6_option(union nd_opts *ndopts)
317 {
318 	struct nd_opt_hdr *nd_opt;
319 	int olen;
320 
321 	if (!ndopts)
322 		panic("ndopts == NULL in nd6_option");
323 	if (!ndopts->nd_opts_last)
324 		panic("uninitialized ndopts in nd6_option");
325 	if (!ndopts->nd_opts_search)
326 		return NULL;
327 	if (ndopts->nd_opts_done)
328 		return NULL;
329 
330 	nd_opt = ndopts->nd_opts_search;
331 
332 	/* make sure nd_opt_len is inside the buffer */
333 	if ((caddr_t)&nd_opt->nd_opt_len >= (caddr_t)ndopts->nd_opts_last) {
334 		bzero(ndopts, sizeof(*ndopts));
335 		return NULL;
336 	}
337 
338 	olen = nd_opt->nd_opt_len << 3;
339 	if (olen == 0) {
340 		/*
341 		 * Message validation requires that all included
342 		 * options have a length that is greater than zero.
343 		 */
344 		bzero(ndopts, sizeof(*ndopts));
345 		return NULL;
346 	}
347 
348 	ndopts->nd_opts_search = (struct nd_opt_hdr *)((caddr_t)nd_opt + olen);
349 	if (ndopts->nd_opts_search > ndopts->nd_opts_last) {
350 		/* option overruns the end of buffer, invalid */
351 		bzero(ndopts, sizeof(*ndopts));
352 		return NULL;
353 	} else if (ndopts->nd_opts_search == ndopts->nd_opts_last) {
354 		/* reached the end of options chain */
355 		ndopts->nd_opts_done = 1;
356 		ndopts->nd_opts_search = NULL;
357 	}
358 	return nd_opt;
359 }
360 
361 /*
362  * Parse multiple ND options.
363  * This function is much easier to use, for ND routines that do not need
364  * multiple options of the same type.
365  */
366 int
nd6_options(union nd_opts * ndopts)367 nd6_options(union nd_opts *ndopts)
368 {
369 	struct nd_opt_hdr *nd_opt;
370 	int i = 0;
371 
372 	if (!ndopts)
373 		panic("ndopts == NULL in nd6_options");
374 	if (!ndopts->nd_opts_last)
375 		panic("uninitialized ndopts in nd6_options");
376 	if (!ndopts->nd_opts_search)
377 		return 0;
378 
379 	while (1) {
380 		nd_opt = nd6_option(ndopts);
381 		if (!nd_opt && !ndopts->nd_opts_last) {
382 			/*
383 			 * Message validation requires that all included
384 			 * options have a length that is greater than zero.
385 			 */
386 			icmp6stat.icp6s_nd_badopt++;
387 			bzero(ndopts, sizeof(*ndopts));
388 			return -1;
389 		}
390 
391 		if (!nd_opt)
392 			goto skip1;
393 
394 		switch (nd_opt->nd_opt_type) {
395 		case ND_OPT_SOURCE_LINKADDR:
396 		case ND_OPT_TARGET_LINKADDR:
397 		case ND_OPT_MTU:
398 		case ND_OPT_REDIRECTED_HEADER:
399 			if (ndopts->nd_opt_array[nd_opt->nd_opt_type]) {
400 				nd6log((LOG_INFO,
401 				    "duplicated ND6 option found (type=%d)\n",
402 				    nd_opt->nd_opt_type));
403 				/* XXX bark? */
404 			} else {
405 				ndopts->nd_opt_array[nd_opt->nd_opt_type]
406 					= nd_opt;
407 			}
408 			break;
409 		case ND_OPT_PREFIX_INFORMATION:
410 			if (ndopts->nd_opt_array[nd_opt->nd_opt_type] == 0) {
411 				ndopts->nd_opt_array[nd_opt->nd_opt_type]
412 					= nd_opt;
413 			}
414 			ndopts->nd_opts_pi_end =
415 				(struct nd_opt_prefix_info *)nd_opt;
416 			break;
417 		default:
418 			/*
419 			 * Unknown options must be silently ignored,
420 			 * to accomodate future extension to the protocol.
421 			 */
422 			nd6log((LOG_DEBUG,
423 			    "nd6_options: unsupported option %d - "
424 			    "option ignored\n", nd_opt->nd_opt_type));
425 		}
426 
427 skip1:
428 		i++;
429 		if (i > nd6_maxndopt) {
430 			icmp6stat.icp6s_nd_toomanyopt++;
431 			nd6log((LOG_INFO, "too many loop in nd opt\n"));
432 			break;
433 		}
434 
435 		if (ndopts->nd_opts_done)
436 			break;
437 	}
438 
439 	return 0;
440 }
441 
442 /*
443  * ND6 timer routine to expire default route list and prefix list
444  */
445 static void
nd6_timer_dispatch(netmsg_t nmsg)446 nd6_timer_dispatch(netmsg_t nmsg)
447 {
448 	struct llinfo_nd6 *ln;
449 	struct nd_defrouter *dr;
450 	struct nd_prefix *pr;
451 	struct ifnet *ifp;
452 	struct in6_ifaddr *ia6, *nia6;
453 
454 	ASSERT_NETISR0;
455 
456 	crit_enter();
457 	lwkt_replymsg(&nmsg->lmsg, 0);	/* reply ASAP */
458 	crit_exit();
459 
460 	mtx_lock(&nd6_mtx);
461 
462 	ln = llinfo_nd6.ln_next;
463 	while (ln && ln != &llinfo_nd6) {
464 		struct rtentry *rt;
465 		struct sockaddr_in6 *dst;
466 		struct llinfo_nd6 *next = ln->ln_next;
467 		/* XXX: used for the DELAY case only: */
468 		struct nd_ifinfo *ndi = NULL;
469 
470 		if ((rt = ln->ln_rt) == NULL) {
471 			ln = next;
472 			continue;
473 		}
474 		if ((ifp = rt->rt_ifp) == NULL) {
475 			ln = next;
476 			continue;
477 		}
478 		ndi = ND_IFINFO(ifp);
479 		dst = (struct sockaddr_in6 *)rt_key(rt);
480 
481 		if (ln->ln_expire > time_uptime) {
482 			ln = next;
483 			continue;
484 		}
485 
486 		/* sanity check */
487 		if (!rt)
488 			panic("rt=0 in nd6_timer(ln=%p)", ln);
489 		if (rt->rt_llinfo && (struct llinfo_nd6 *)rt->rt_llinfo != ln)
490 			panic("rt_llinfo(%p) is not equal to ln(%p)",
491 			      rt->rt_llinfo, ln);
492 		if (!dst)
493 			panic("dst=0 in nd6_timer(ln=%p)", ln);
494 
495 		switch (ln->ln_state) {
496 		case ND6_LLINFO_WAITDELETE:
497 			next = nd6_free(rt);
498 			break;
499 		case ND6_LLINFO_INCOMPLETE:
500 			if (ln->ln_asked++ >= nd6_mmaxtries) {
501 				struct mbuf *m = ln->ln_hold;
502 				if (m) {
503 					if (rt->rt_ifp) {
504 						/*
505 						 * Fake rcvif to make ICMP error
506 						 * more helpful in diagnosing
507 						 * for the receiver.
508 						 * XXX: should we consider
509 						 * older rcvif?
510 						 */
511 						m->m_pkthdr.rcvif = rt->rt_ifp;
512 					}
513 
514 					/*
515 					 * mbuf has empty MAC header, remove
516 					 * for icmp.  XXX layer violation.
517 					 */
518 					m_adj(m, ETHER_HDR_LEN);
519 					icmp6_error(m, ICMP6_DST_UNREACH,
520 						    ICMP6_DST_UNREACH_ADDR, 0);
521 					ln->ln_hold = NULL;
522 				}
523 				ln->ln_state = ND6_LLINFO_WAITDELETE;
524 				rt_rtmsg(RTM_MISS, rt, rt->rt_ifp, 0);
525 			}
526 			ln->ln_expire = time_uptime +
527 				ND_IFINFO(ifp)->retrans / 1000;
528 			nd6_ns_output(ifp, NULL, &dst->sin6_addr,
529 				ln, 0);
530 			break;
531 		case ND6_LLINFO_REACHABLE:
532 			if (ln->ln_expire) {
533 				ln->ln_state = ND6_LLINFO_STALE;
534 				ln->ln_expire = time_uptime + nd6_gctimer;
535 			}
536 			break;
537 
538 		case ND6_LLINFO_STALE:
539 			/* Garbage Collection(RFC 2461 5.3) */
540 			if (ln->ln_expire)
541 				next = nd6_free(rt);
542 			break;
543 
544 		case ND6_LLINFO_DELAY:
545 			if (ndi && (ndi->flags & ND6_IFF_PERFORMNUD)) {
546 				/* We need NUD */
547 				ln->ln_asked = 1;
548 				ln->ln_state = ND6_LLINFO_PROBE;
549 				ln->ln_expire = time_uptime +
550 					ndi->retrans / 1000;
551 				nd6_ns_output(ifp, &dst->sin6_addr,
552 					      &dst->sin6_addr,
553 					      ln, 0);
554 			} else {
555 				ln->ln_state = ND6_LLINFO_STALE; /* XXX */
556 				ln->ln_expire = time_uptime + nd6_gctimer;
557 			}
558 			break;
559 		case ND6_LLINFO_PROBE:
560 			if (ln->ln_asked < nd6_umaxtries) {
561 				ln->ln_asked++;
562 				ln->ln_expire = time_uptime +
563 					ND_IFINFO(ifp)->retrans / 1000;
564 				nd6_ns_output(ifp, &dst->sin6_addr,
565 					       &dst->sin6_addr, ln, 0);
566 			} else {
567 				rt_rtmsg(RTM_MISS, rt, rt->rt_ifp, 0);
568 				next = nd6_free(rt);
569 			}
570 			break;
571 		}
572 		ln = next;
573 	}
574 
575 	/* expire default router list */
576 	dr = TAILQ_FIRST(&nd_defrouter);
577 	while (dr) {
578 		if (dr->expire && dr->expire < time_uptime) {
579 			struct nd_defrouter *t;
580 			t = TAILQ_NEXT(dr, dr_entry);
581 			defrtrlist_del(dr);
582 			dr = t;
583 		} else {
584 			dr = TAILQ_NEXT(dr, dr_entry);
585 		}
586 	}
587 
588 	/*
589 	 * expire interface addresses.
590 	 * in the past the loop was inside prefix expiry processing.
591 	 * However, from a stricter speci-confrmance standpoint, we should
592 	 * rather separate address lifetimes and prefix lifetimes.
593 	 */
594 addrloop:
595 	for (ia6 = in6_ifaddr; ia6; ia6 = nia6) {
596 		nia6 = ia6->ia_next;
597 		/* check address lifetime */
598 		if (IFA6_IS_INVALID(ia6)) {
599 			int regen = 0;
600 
601 			/*
602 			 * If the expiring address is temporary, try
603 			 * regenerating a new one.  This would be useful when
604 			 * we suspended a laptop PC, then turned it on after a
605 			 * period that could invalidate all temporary
606 			 * addresses.  Although we may have to restart the
607 			 * loop (see below), it must be after purging the
608 			 * address.  Otherwise, we'd see an infinite loop of
609 			 * regeneration.
610 			 */
611 			if (ip6_use_tempaddr &&
612 			    (ia6->ia6_flags & IN6_IFF_TEMPORARY)) {
613 				if (regen_tmpaddr(ia6) == 0)
614 					regen = 1;
615 			}
616 
617 			in6_purgeaddr(&ia6->ia_ifa);
618 
619 			if (regen)
620 				goto addrloop; /* XXX: see below */
621 			/* ia6 is no longer good, continue on to next */
622 			continue;
623 		}
624 		if (IFA6_IS_DEPRECATED(ia6)) {
625 			int oldflags = ia6->ia6_flags;
626 
627 			if ((oldflags & IN6_IFF_DEPRECATED) == 0) {
628 				ia6->ia6_flags |= IN6_IFF_DEPRECATED;
629 				in6_newaddrmsg((struct ifaddr *)ia6);
630 			}
631 
632 			/*
633 			 * If a temporary address has just become deprecated,
634 			 * regenerate a new one if possible.
635 			 */
636 			if (ip6_use_tempaddr &&
637 			    (ia6->ia6_flags & IN6_IFF_TEMPORARY) &&
638 			    !(oldflags & IN6_IFF_DEPRECATED)) {
639 
640 				if (regen_tmpaddr(ia6) == 0) {
641 					/*
642 					 * A new temporary address is
643 					 * generated.
644 					 * XXX: this means the address chain
645 					 * has changed while we are still in
646 					 * the loop.  Although the change
647 					 * would not cause disaster (because
648 					 * it's not a deletion, but an
649 					 * addition,) we'd rather restart the
650 					 * loop just for safety.  Or does this
651 					 * significantly reduce performance??
652 					 */
653 					goto addrloop;
654 				}
655 			}
656 		} else {
657 			/*
658 			 * A new RA might have made a deprecated address
659 			 * preferred.
660 			 */
661 			if (ia6->ia6_flags & IN6_IFF_DEPRECATED) {
662 				ia6->ia6_flags &= ~IN6_IFF_DEPRECATED;
663 				in6_newaddrmsg((struct ifaddr *)ia6);
664 			}
665 		}
666 	}
667 
668 	/* expire prefix list */
669 	pr = nd_prefix.lh_first;
670 	while (pr) {
671 		/*
672 		 * check prefix lifetime.
673 		 * since pltime is just for autoconf, pltime processing for
674 		 * prefix is not necessary.
675 		 */
676 		if (pr->ndpr_expire && pr->ndpr_expire < time_uptime) {
677 			struct nd_prefix *t;
678 			t = pr->ndpr_next;
679 
680 			/*
681 			 * address expiration and prefix expiration are
682 			 * separate.  NEVER perform in6_purgeaddr here.
683 			 */
684 
685 			prelist_remove(pr);
686 			pr = t;
687 		} else
688 			pr = pr->ndpr_next;
689 	}
690 
691 	mtx_unlock(&nd6_mtx);
692 
693 	callout_reset(&nd6_timer_ch, nd6_prune * hz, nd6_timer, NULL);
694 }
695 
696 static void
nd6_timer(void * arg __unused)697 nd6_timer(void *arg __unused)
698 {
699 	struct lwkt_msg *lmsg = &nd6_timer_netmsg.lmsg;
700 
701 	KASSERT(mycpuid == 0, ("not on cpu0"));
702 	crit_enter();
703 	if (lmsg->ms_flags & MSGF_DONE)
704 		lwkt_sendmsg_oncpu(netisr_cpuport(0), lmsg);
705 	crit_exit();
706 }
707 
708 void
nd6_timer_init(void)709 nd6_timer_init(void)
710 {
711 	callout_init_mp(&nd6_timer_ch);
712 	netmsg_init(&nd6_timer_netmsg, NULL, &netisr_adone_rport,
713 	    MSGF_PRIORITY, nd6_timer_dispatch);
714 	callout_reset_bycpu(&nd6_timer_ch, hz, nd6_timer, NULL, 0);
715 }
716 
717 static int
regen_tmpaddr(struct in6_ifaddr * ia6)718 regen_tmpaddr(struct in6_ifaddr *ia6) /* deprecated/invalidated temporary
719 					 address */
720 {
721 	struct ifaddr_container *ifac;
722 	struct ifnet *ifp;
723 	struct in6_ifaddr *public_ifa6 = NULL;
724 
725 	ifp = ia6->ia_ifa.ifa_ifp;
726 	TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
727 		struct ifaddr *ifa = ifac->ifa;
728 		struct in6_ifaddr *it6;
729 
730 		if (ifa->ifa_addr->sa_family != AF_INET6)
731 			continue;
732 
733 		it6 = (struct in6_ifaddr *)ifa;
734 
735 		/* ignore no autoconf addresses. */
736 		if (!(it6->ia6_flags & IN6_IFF_AUTOCONF))
737 			continue;
738 
739 		/* ignore autoconf addresses with different prefixes. */
740 		if (it6->ia6_ndpr == NULL || it6->ia6_ndpr != ia6->ia6_ndpr)
741 			continue;
742 
743 		/*
744 		 * Now we are looking at an autoconf address with the same
745 		 * prefix as ours.  If the address is temporary and is still
746 		 * preferred, do not create another one.  It would be rare, but
747 		 * could happen, for example, when we resume a laptop PC after
748 		 * a long period.
749 		 */
750 		if ((it6->ia6_flags & IN6_IFF_TEMPORARY) &&
751 		    !IFA6_IS_DEPRECATED(it6)) {
752 			public_ifa6 = NULL;
753 			break;
754 		}
755 
756 		/*
757 		 * This is a public autoconf address that has the same prefix
758 		 * as ours.  If it is preferred, keep it.  We can't break the
759 		 * loop here, because there may be a still-preferred temporary
760 		 * address with the prefix.
761 		 */
762 		if (!IFA6_IS_DEPRECATED(it6))
763 		    public_ifa6 = it6;
764 	}
765 
766 	if (public_ifa6 != NULL) {
767 		int e;
768 
769 		if ((e = in6_tmpifadd(public_ifa6, 0)) != 0) {
770 			log(LOG_NOTICE, "regen_tmpaddr: failed to create a new"
771 			    " tmp addr,errno=%d\n", e);
772 			return (-1);
773 		}
774 		return (0);
775 	}
776 
777 	return (-1);
778 }
779 
780 /*
781  * Nuke neighbor cache/prefix/default router management table, right before
782  * ifp goes away.
783  */
784 void
nd6_purge(struct ifnet * ifp)785 nd6_purge(struct ifnet *ifp)
786 {
787 	struct llinfo_nd6 *ln, *nln;
788 	struct nd_defrouter *dr, *ndr, drany;
789 	struct nd_prefix *pr, *npr;
790 
791 	/* Nuke default router list entries toward ifp */
792 	if ((dr = TAILQ_FIRST(&nd_defrouter)) != NULL) {
793 		/*
794 		 * The first entry of the list may be stored in
795 		 * the routing table, so we'll delete it later.
796 		 */
797 		for (dr = TAILQ_NEXT(dr, dr_entry); dr; dr = ndr) {
798 			ndr = TAILQ_NEXT(dr, dr_entry);
799 			if (dr->ifp == ifp)
800 				defrtrlist_del(dr);
801 		}
802 		dr = TAILQ_FIRST(&nd_defrouter);
803 		if (dr->ifp == ifp)
804 			defrtrlist_del(dr);
805 	}
806 
807 	/* Nuke prefix list entries toward ifp */
808 	for (pr = nd_prefix.lh_first; pr; pr = npr) {
809 		npr = pr->ndpr_next;
810 		if (pr->ndpr_ifp == ifp) {
811 			/*
812 			 * Previously, pr->ndpr_addr is removed as well,
813 			 * but I strongly believe we don't have to do it.
814 			 * nd6_purge() is only called from in6_ifdetach(),
815 			 * which removes all the associated interface addresses
816 			 * by itself.
817 			 * (jinmei@kame.net 20010129)
818 			 */
819 			prelist_remove(pr);
820 		}
821 	}
822 
823 	/* cancel default outgoing interface setting */
824 	if (nd6_defifindex == ifp->if_index)
825 		nd6_setdefaultiface(0);
826 
827 	if (!ip6_forwarding &&
828 	    (ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV)) {
829 		/* refresh default router list */
830 		bzero(&drany, sizeof(drany));
831 		defrouter_delreq(&drany, 0);
832 		defrouter_select();
833 	}
834 
835 	/*
836 	 * Nuke neighbor cache entries for the ifp.
837 	 * Note that rt->rt_ifp may not be the same as ifp,
838 	 * due to KAME goto ours hack.  See RTM_RESOLVE case in
839 	 * nd6_rtrequest(), and ip6_input().
840 	 */
841 	ln = llinfo_nd6.ln_next;
842 	while (ln && ln != &llinfo_nd6) {
843 		struct rtentry *rt;
844 		struct sockaddr_dl *sdl;
845 
846 		nln = ln->ln_next;
847 		rt = ln->ln_rt;
848 		if (rt && rt->rt_gateway &&
849 		    rt->rt_gateway->sa_family == AF_LINK) {
850 			sdl = (struct sockaddr_dl *)rt->rt_gateway;
851 			if (sdl->sdl_index == ifp->if_index)
852 				nln = nd6_free(rt);
853 		}
854 		ln = nln;
855 	}
856 }
857 
858 struct rtentry *
nd6_lookup(struct in6_addr * addr6,int create,struct ifnet * ifp)859 nd6_lookup(struct in6_addr *addr6, int create, struct ifnet *ifp)
860 {
861 	struct rtentry *rt;
862 	struct sockaddr_in6 sin6;
863 
864 	bzero(&sin6, sizeof(sin6));
865 	sin6.sin6_len = sizeof(struct sockaddr_in6);
866 	sin6.sin6_family = AF_INET6;
867 	sin6.sin6_addr = *addr6;
868 
869 	if (create)
870 		rt = rtlookup((struct sockaddr *)&sin6);
871 	else
872 		rt = rtpurelookup((struct sockaddr *)&sin6);
873 	if (rt && !(rt->rt_flags & RTF_LLINFO)) {
874 		/*
875 		 * This is the case for the default route.
876 		 * If we want to create a neighbor cache for the address, we
877 		 * should free the route for the destination and allocate an
878 		 * interface route.
879 		 */
880 		if (create) {
881 			--rt->rt_refcnt;
882 			rt = NULL;
883 		}
884 	}
885 	if (!rt) {
886 		if (create && ifp) {
887 			int e;
888 
889 			/*
890 			 * If no route is available and create is set,
891 			 * we allocate a host route for the destination
892 			 * and treat it like an interface route.
893 			 * This hack is necessary for a neighbor which can't
894 			 * be covered by our own prefix.
895 			 */
896 			struct ifaddr *ifa;
897 
898 			ifa = ifaof_ifpforaddr((struct sockaddr *)&sin6, ifp);
899 			if (ifa == NULL)
900 				return (NULL);
901 
902 			/*
903 			 * Create a new route.  RTF_LLINFO is necessary
904 			 * to create a Neighbor Cache entry for the
905 			 * destination in nd6_rtrequest which will be
906 			 * called in rtrequest via ifa->ifa_rtrequest.
907 			 */
908 			if ((e = rtrequest(RTM_ADD, (struct sockaddr *)&sin6,
909 			     ifa->ifa_addr, (struct sockaddr *)&all1_sa,
910 			     (ifa->ifa_flags | RTF_HOST | RTF_LLINFO) &
911 			     ~RTF_CLONING, &rt)) != 0) {
912 				log(LOG_ERR,
913 				    "nd6_lookup: failed to add route for a "
914 				    "neighbor(%s), errno=%d\n",
915 				    ip6_sprintf(addr6), e);
916 			}
917 			if (rt == NULL)
918 				return (NULL);
919 			if (rt->rt_llinfo) {
920 				struct llinfo_nd6 *ln =
921 				    (struct llinfo_nd6 *)rt->rt_llinfo;
922 
923 				ln->ln_state = ND6_LLINFO_NOSTATE;
924 			}
925 		} else
926 			return (NULL);
927 	}
928 	rt->rt_refcnt--;
929 
930 	if (!ND6_RTENTRY_IS_NEIGHBOR(rt, ifp)) {
931 		if (create) {
932 			log(LOG_DEBUG,
933 			    "nd6_lookup: failed to lookup %s (if = %s)\n",
934 			    ip6_sprintf(addr6), ifp ? if_name(ifp) : "unspec");
935 			/* xxx more logs... kazu */
936 		}
937 		return (NULL);
938 	}
939 	return (rt);
940 }
941 
942 static struct rtentry *
nd6_neighbor_lookup(struct in6_addr * addr6,struct ifnet * ifp)943 nd6_neighbor_lookup(struct in6_addr *addr6, struct ifnet *ifp)
944 {
945 	struct rtentry *rt;
946 	struct sockaddr_in6 sin6;
947 
948 	bzero(&sin6, sizeof(sin6));
949 	sin6.sin6_len = sizeof(struct sockaddr_in6);
950 	sin6.sin6_family = AF_INET6;
951 	sin6.sin6_addr = *addr6;
952 
953 	rt = rtpurelookup((struct sockaddr *)&sin6);
954 	if (rt == NULL)
955 		return (NULL);
956 	rt->rt_refcnt--;
957 
958 	if (!ND6_RTENTRY_IS_NEIGHBOR(rt, ifp)) {
959 		if (nd6_onlink_ns_rfc4861 &&
960 		    (ND6_RTENTRY_IS_LLCLONING(rt) ||	/* not cloned yet */
961 		     (rt->rt_parent != NULL &&		/* cloning */
962 		      ND6_RTENTRY_IS_LLCLONING(rt->rt_parent)))) {
963 			/*
964 			 * If cloning ever happened or is happening,
965 			 * rtentry for addr6 would or will become a
966 			 * neighbor cache.
967 			 */
968 		} else {
969 			rt = NULL;
970 		}
971 	}
972 	return (rt);
973 }
974 
975 /*
976  * Detect if a given IPv6 address identifies a neighbor on a given link.
977  * XXX: should take care of the destination of a p2p link?
978  */
979 int
nd6_is_addr_neighbor(struct sockaddr_in6 * addr,struct ifnet * ifp)980 nd6_is_addr_neighbor(struct sockaddr_in6 *addr, struct ifnet *ifp)
981 {
982 	struct ifaddr_container *ifac;
983 
984 #define IFADDR6(a) ((((struct in6_ifaddr *)(a))->ia_addr).sin6_addr)
985 #define IFMASK6(a) ((((struct in6_ifaddr *)(a))->ia_prefixmask).sin6_addr)
986 
987 	/*
988 	 * A link-local address is always a neighbor.
989 	 * XXX: we should use the sin6_scope_id field rather than the embedded
990 	 * interface index.
991 	 */
992 	if (IN6_IS_ADDR_LINKLOCAL(&addr->sin6_addr) &&
993 	    ntohs(*(u_int16_t *)&addr->sin6_addr.s6_addr[2]) == ifp->if_index)
994 		return (1);
995 
996 	/*
997 	 * If the address matches one of our addresses,
998 	 * it should be a neighbor.
999 	 */
1000 	TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
1001 		struct ifaddr *ifa = ifac->ifa;
1002 
1003 		if (ifa->ifa_addr->sa_family != AF_INET6)
1004 			continue;
1005 
1006 		if (IN6_ARE_MASKED_ADDR_EQUAL(&IFADDR6(ifa), &addr->sin6_addr,
1007 		    &IFMASK6(ifa)))
1008 			return (1);
1009 	}
1010 
1011 	/*
1012 	 * Even if the address matches none of our addresses, it might be
1013 	 * in the neighbor cache.
1014 	 */
1015 	if (nd6_neighbor_lookup(&addr->sin6_addr, ifp) != NULL)
1016 		return (1);
1017 
1018 	return (0);
1019 #undef IFADDR6
1020 #undef IFMASK6
1021 }
1022 
1023 /*
1024  * Free an nd6 llinfo entry.
1025  */
1026 struct llinfo_nd6 *
nd6_free(struct rtentry * rt)1027 nd6_free(struct rtentry *rt)
1028 {
1029 	struct llinfo_nd6 *ln = (struct llinfo_nd6 *)rt->rt_llinfo, *next;
1030 	struct in6_addr in6 = ((struct sockaddr_in6 *)rt_key(rt))->sin6_addr;
1031 	struct nd_defrouter *dr;
1032 
1033 	/*
1034 	 * we used to have kpfctlinput(PRC_HOSTDEAD) here.
1035 	 * even though it is not harmful, it was not really necessary.
1036 	 */
1037 
1038 	/* XXX: this condition too restrictive? */
1039 	if (!ip6_forwarding &&
1040 	    (ND_IFINFO(rt->rt_ifp)->flags & ND6_IFF_ACCEPT_RTADV)) {
1041 		mtx_lock(&nd6_mtx);
1042 		dr = defrouter_lookup(
1043 		    &((struct sockaddr_in6 *)rt_key(rt))->sin6_addr,
1044 		    rt->rt_ifp);
1045 
1046 		if (ln->ln_router || dr) {
1047 			/*
1048 			 * rt6_flush must be called whether or not the neighbor
1049 			 * is in the Default Router List.
1050 			 * See a corresponding comment in nd6_na_input().
1051 			 */
1052 			rt6_flush(&in6, rt->rt_ifp);
1053 		}
1054 
1055 		if (dr) {
1056 			/*
1057 			 * Unreachablity of a router might affect the default
1058 			 * router selection and on-link detection of advertised
1059 			 * prefixes.
1060 			 */
1061 
1062 			/*
1063 			 * Temporarily fake the state to choose a new default
1064 			 * router and to perform on-link determination of
1065 			 * prefixes correctly.
1066 			 * Below the state will be set correctly,
1067 			 * or the entry itself will be deleted.
1068 			 */
1069 			ln->ln_state = ND6_LLINFO_INCOMPLETE;
1070 
1071 			/*
1072 			 * Since defrouter_select() does not affect the
1073 			 * on-link determination and MIP6 needs the check
1074 			 * before the default router selection, we perform
1075 			 * the check now.
1076 			 */
1077 			pfxlist_onlink_check();
1078 
1079 			if (dr == TAILQ_FIRST(&nd_defrouter)) {
1080 				/*
1081 				 * It is used as the current default router,
1082 				 * so we have to move it to the end of the
1083 				 * list and choose a new one.
1084 				 * XXX: it is not very efficient if this is
1085 				 *      the only router.
1086 				 */
1087 				TAILQ_REMOVE(&nd_defrouter, dr, dr_entry);
1088 				TAILQ_INSERT_TAIL(&nd_defrouter, dr, dr_entry);
1089 
1090 				defrouter_select();
1091 			}
1092 		}
1093 		mtx_unlock(&nd6_mtx);
1094 	}
1095 
1096 	/*
1097 	 * Before deleting the entry, remember the next entry as the
1098 	 * return value.  We need this because pfxlist_onlink_check() above
1099 	 * might have freed other entries (particularly the old next entry) as
1100 	 * a side effect (XXX).
1101 	 */
1102 	next = ln->ln_next;
1103 
1104 	/*
1105 	 * Detach the route from the routing tree and the list of neighbor
1106 	 * caches, and disable the route entry not to be used in already
1107 	 * cached routes.
1108 	 *
1109 	 * ND expiry happens under one big timer.
1110 	 * To avoid overflowing the route socket, don't report this.
1111 	 * Now that RTM_MISS is reported when an address is unresolvable
1112 	 * the benefit of reporting this deletion is questionable.
1113 	 */
1114 	rtrequest(RTM_DELETE, rt_key(rt), NULL, rt_mask(rt), 0, NULL);
1115 
1116 	return (next);
1117 }
1118 
1119 /*
1120  * Upper-layer reachability hint for Neighbor Unreachability Detection.
1121  *
1122  * XXX cost-effective metods?
1123  */
1124 void
nd6_nud_hint(struct rtentry * rt,struct in6_addr * dst6,int force)1125 nd6_nud_hint(struct rtentry *rt, struct in6_addr *dst6, int force)
1126 {
1127 	struct llinfo_nd6 *ln;
1128 
1129 	/*
1130 	 * If the caller specified "rt", use that.  Otherwise, resolve the
1131 	 * routing table by supplied "dst6".
1132 	 */
1133 	if (!rt) {
1134 		if (!dst6)
1135 			return;
1136 		if (!(rt = nd6_lookup(dst6, 0, NULL)))
1137 			return;
1138 	}
1139 
1140 	if ((rt->rt_flags & RTF_GATEWAY) ||
1141 	    !(rt->rt_flags & RTF_LLINFO) ||
1142 	    rt->rt_llinfo == NULL || rt->rt_gateway == NULL ||
1143 	    rt->rt_gateway->sa_family != AF_LINK) {
1144 		/* This is not a host route. */
1145 		return;
1146 	}
1147 
1148 	ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1149 	if (ln->ln_state < ND6_LLINFO_REACHABLE)
1150 		return;
1151 
1152 	/*
1153 	 * if we get upper-layer reachability confirmation many times,
1154 	 * it is possible we have false information.
1155 	 */
1156 	if (!force) {
1157 		ln->ln_byhint++;
1158 		if (ln->ln_byhint > nd6_maxnudhint)
1159 			return;
1160 	}
1161 
1162 	ln->ln_state = ND6_LLINFO_REACHABLE;
1163 	if (ln->ln_expire)
1164 		ln->ln_expire = time_uptime +
1165 			ND_IFINFO(rt->rt_ifp)->reachable;
1166 }
1167 
1168 void
nd6_rtrequest(int req,struct rtentry * rt)1169 nd6_rtrequest(int req, struct rtentry *rt)
1170 {
1171 	struct sockaddr *gate = rt->rt_gateway;
1172 	struct llinfo_nd6 *ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1173 	static struct sockaddr_dl null_sdl = {sizeof(null_sdl), AF_LINK};
1174 	struct ifnet *ifp = rt->rt_ifp;
1175 	struct ifaddr *ifa;
1176 
1177 	if ((rt->rt_flags & RTF_GATEWAY))
1178 		return;
1179 
1180 	if (nd6_need_cache(ifp) == 0 && !(rt->rt_flags & RTF_HOST)) {
1181 		/*
1182 		 * This is probably an interface direct route for a link
1183 		 * which does not need neighbor caches (e.g. fe80::%lo0/64).
1184 		 * We do not need special treatment below for such a route.
1185 		 * Moreover, the RTF_LLINFO flag which would be set below
1186 		 * would annoy the ndp(8) command.
1187 		 */
1188 		return;
1189 	}
1190 
1191 	if (req == RTM_RESOLVE &&
1192 	    (nd6_need_cache(ifp) == 0 || /* stf case */
1193 	     !nd6_is_addr_neighbor((struct sockaddr_in6 *)rt_key(rt), ifp))) {
1194 		/*
1195 		 * FreeBSD and BSD/OS often make a cloned host route based
1196 		 * on a less-specific route (e.g. the default route).
1197 		 * If the less specific route does not have a "gateway"
1198 		 * (this is the case when the route just goes to a p2p or an
1199 		 * stf interface), we'll mistakenly make a neighbor cache for
1200 		 * the host route, and will see strange neighbor solicitation
1201 		 * for the corresponding destination.  In order to avoid the
1202 		 * confusion, we check if the destination of the route is
1203 		 * a neighbor in terms of neighbor discovery, and stop the
1204 		 * process if not.  Additionally, we remove the LLINFO flag
1205 		 * so that ndp(8) will not try to get the neighbor information
1206 		 * of the destination.
1207 		 */
1208 		rt->rt_flags &= ~RTF_LLINFO;
1209 		return;
1210 	}
1211 
1212 	switch (req) {
1213 	case RTM_ADD:
1214 		/*
1215 		 * There is no backward compatibility :)
1216 		 *
1217 		 * if (!(rt->rt_flags & RTF_HOST) &&
1218 		 *     SIN(rt_mask(rt))->sin_addr.s_addr != 0xffffffff)
1219 		 *	   rt->rt_flags |= RTF_CLONING;
1220 		 */
1221 		if (rt->rt_flags & (RTF_CLONING | RTF_LLINFO)) {
1222 			/*
1223 			 * Case 1: This route should come from
1224 			 * a route to interface.  RTF_LLINFO flag is set
1225 			 * for a host route whose destination should be
1226 			 * treated as on-link.
1227 			 */
1228 			rt_setgate(rt, rt_key(rt),
1229 				   (struct sockaddr *)&null_sdl);
1230 			gate = rt->rt_gateway;
1231 			SDL(gate)->sdl_type = ifp->if_type;
1232 			SDL(gate)->sdl_index = ifp->if_index;
1233 			if (ln)
1234 				ln->ln_expire = time_uptime;
1235 			if (ln && ln->ln_expire == 0) {
1236 				/* kludge for desktops */
1237 				ln->ln_expire = 1;
1238 			}
1239 			if ((rt->rt_flags & RTF_CLONING))
1240 				break;
1241 		}
1242 		/*
1243 		 * In IPv4 code, we try to annonuce new RTF_ANNOUNCE entry here.
1244 		 * We don't do that here since llinfo is not ready yet.
1245 		 *
1246 		 * There are also couple of other things to be discussed:
1247 		 * - unsolicited NA code needs improvement beforehand
1248 		 * - RFC2461 says we MAY send multicast unsolicited NA
1249 		 *   (7.2.6 paragraph 4), however, it also says that we
1250 		 *   SHOULD provide a mechanism to prevent multicast NA storm.
1251 		 *   we don't have anything like it right now.
1252 		 *   note that the mechanism needs a mutual agreement
1253 		 *   between proxies, which means that we need to implement
1254 		 *   a new protocol, or a new kludge.
1255 		 * - from RFC2461 6.2.4, host MUST NOT send an unsolicited NA.
1256 		 *   we need to check ip6forwarding before sending it.
1257 		 *   (or should we allow proxy ND configuration only for
1258 		 *   routers?  there's no mention about proxy ND from hosts)
1259 		 */
1260 #if 0
1261 		/* XXX it does not work */
1262 		if ((rt->rt_flags & RTF_ANNOUNCE) && mycpuid == 0) {
1263 			nd6_na_output(ifp,
1264 			      &SIN6(rt_key(rt))->sin6_addr,
1265 			      &SIN6(rt_key(rt))->sin6_addr,
1266 			      ip6_forwarding ? ND_NA_FLAG_ROUTER : 0,
1267 			      1, NULL);
1268 		}
1269 #endif
1270 		/* FALLTHROUGH */
1271 	case RTM_RESOLVE:
1272 		if ((ifp->if_flags & (IFF_POINTOPOINT | IFF_LOOPBACK)) == 0) {
1273 			/*
1274 			 * Address resolution isn't necessary for a point to
1275 			 * point link, so we can skip this test for a p2p link.
1276 			 */
1277 			if (gate->sa_family != AF_LINK ||
1278 			    gate->sa_len < sizeof(null_sdl)) {
1279 				log(LOG_DEBUG,
1280 				    "nd6_rtrequest: bad gateway value: %s\n",
1281 				    if_name(ifp));
1282 				break;
1283 			}
1284 			SDL(gate)->sdl_type = ifp->if_type;
1285 			SDL(gate)->sdl_index = ifp->if_index;
1286 		}
1287 		if (ln != NULL)
1288 			break;	/* This happens on a route change */
1289 		/*
1290 		 * Case 2: This route may come from cloning, or a manual route
1291 		 * add with a LL address.
1292 		 */
1293 		R_Malloc(ln, struct llinfo_nd6 *, sizeof(*ln));
1294 		rt->rt_llinfo = (caddr_t)ln;
1295 		if (!ln) {
1296 			log(LOG_DEBUG, "nd6_rtrequest: malloc failed\n");
1297 			break;
1298 		}
1299 		nd6_inuse++;
1300 		nd6_allocated++;
1301 		bzero(ln, sizeof(*ln));
1302 		ln->ln_rt = rt;
1303 		/* this is required for "ndp" command. - shin */
1304 		if (req == RTM_ADD) {
1305 		        /*
1306 			 * gate should have some valid AF_LINK entry,
1307 			 * and ln->ln_expire should have some lifetime
1308 			 * which is specified by ndp command.
1309 			 */
1310 			ln->ln_state = ND6_LLINFO_REACHABLE;
1311 			ln->ln_byhint = 0;
1312 		} else {
1313 		        /*
1314 			 * When req == RTM_RESOLVE, rt is created and
1315 			 * initialized in rtrequest(), so rt_expire is 0.
1316 			 */
1317 			ln->ln_state = ND6_LLINFO_NOSTATE;
1318 			ln->ln_expire = time_uptime;
1319 		}
1320 		rt->rt_flags |= RTF_LLINFO;
1321 		ln->ln_next = llinfo_nd6.ln_next;
1322 		llinfo_nd6.ln_next = ln;
1323 		ln->ln_prev = &llinfo_nd6;
1324 		ln->ln_next->ln_prev = ln;
1325 
1326 		/*
1327 		 * check if rt_key(rt) is one of my address assigned
1328 		 * to the interface.
1329 		 */
1330 		ifa = (struct ifaddr *)in6ifa_ifpwithaddr(rt->rt_ifp,
1331 		    &SIN6(rt_key(rt))->sin6_addr);
1332 		if (ifa) {
1333 			caddr_t macp = nd6_ifptomac(ifp);
1334 			ln->ln_expire = 0;
1335 			ln->ln_state = ND6_LLINFO_REACHABLE;
1336 			ln->ln_byhint = 0;
1337 			if (macp) {
1338 				bcopy(macp, LLADDR(SDL(gate)), ifp->if_addrlen);
1339 				SDL(gate)->sdl_alen = ifp->if_addrlen;
1340 			}
1341 			if (nd6_useloopback) {
1342 				rt->rt_ifp = loif;	/* XXX */
1343 				/*
1344 				 * Make sure rt_ifa be equal to the ifaddr
1345 				 * corresponding to the address.
1346 				 * We need this because when we refer
1347 				 * rt_ifa->ia6_flags in ip6_input, we assume
1348 				 * that the rt_ifa points to the address instead
1349 				 * of the loopback address.
1350 				 */
1351 				if (ifa != rt->rt_ifa) {
1352 					IFAFREE(rt->rt_ifa);
1353 					IFAREF(ifa);
1354 					rt->rt_ifa = ifa;
1355 				}
1356 			}
1357 		} else if (rt->rt_flags & RTF_ANNOUNCE) {
1358 			ln->ln_expire = 0;
1359 			ln->ln_state = ND6_LLINFO_REACHABLE;
1360 			ln->ln_byhint = 0;
1361 
1362 			/*
1363 			 * Join solicited node multicast for proxy ND, and only
1364 			 * join it once on cpu0.
1365 			 */
1366 			if ((ifp->if_flags & IFF_MULTICAST) && mycpuid == 0) {
1367 				struct in6_addr llsol;
1368 				int error;
1369 
1370 				llsol = SIN6(rt_key(rt))->sin6_addr;
1371 				llsol.s6_addr16[0] = htons(0xff02);
1372 				llsol.s6_addr16[1] = htons(ifp->if_index);
1373 				llsol.s6_addr32[1] = 0;
1374 				llsol.s6_addr32[2] = htonl(1);
1375 				llsol.s6_addr8[12] = 0xff;
1376 
1377 				if (!in6_addmulti(&llsol, ifp, &error)) {
1378 					nd6log((LOG_ERR, "%s: failed to join "
1379 					    "%s (errno=%d)\n", if_name(ifp),
1380 					    ip6_sprintf(&llsol), error));
1381 				}
1382 			}
1383 		}
1384 		break;
1385 
1386 	case RTM_DELETE:
1387 		if (!ln)
1388 			break;
1389 		/*
1390 		 * Leave from solicited node multicast for proxy ND, and only
1391 		 * leave it once on cpu0 (since we joined it once on cpu0).
1392 		 */
1393 		if ((rt->rt_flags & RTF_ANNOUNCE) &&
1394 		    (ifp->if_flags & IFF_MULTICAST) && mycpuid == 0) {
1395 			struct in6_addr llsol;
1396 			struct in6_multi *in6m;
1397 
1398 			llsol = SIN6(rt_key(rt))->sin6_addr;
1399 			llsol.s6_addr16[0] = htons(0xff02);
1400 			llsol.s6_addr16[1] = htons(ifp->if_index);
1401 			llsol.s6_addr32[1] = 0;
1402 			llsol.s6_addr32[2] = htonl(1);
1403 			llsol.s6_addr8[12] = 0xff;
1404 
1405 			in6m = IN6_LOOKUP_MULTI(&llsol, ifp);
1406 			if (in6m)
1407 				in6_delmulti(in6m);
1408 		}
1409 		nd6_inuse--;
1410 		ln->ln_next->ln_prev = ln->ln_prev;
1411 		ln->ln_prev->ln_next = ln->ln_next;
1412 		ln->ln_prev = NULL;
1413 		rt->rt_llinfo = 0;
1414 		rt->rt_flags &= ~RTF_LLINFO;
1415 		if (ln->ln_hold)
1416 			m_freem(ln->ln_hold);
1417 		R_Free(ln);
1418 	}
1419 }
1420 
1421 int
nd6_ioctl(u_long cmd,caddr_t data,struct ifnet * ifp)1422 nd6_ioctl(u_long cmd, caddr_t data, struct ifnet *ifp)
1423 {
1424 	struct in6_drlist *drl = (struct in6_drlist *)data;
1425 	struct in6_prlist *prl = (struct in6_prlist *)data;
1426 	struct in6_ndireq *ndi = (struct in6_ndireq *)data;
1427 	struct in6_nbrinfo *nbi = (struct in6_nbrinfo *)data;
1428 	struct in6_ndifreq *ndif = (struct in6_ndifreq *)data;
1429 	struct nd_defrouter *dr, any;
1430 	struct nd_prefix *pr;
1431 	struct rtentry *rt;
1432 	int i = 0, error = 0;
1433 
1434 	switch (cmd) {
1435 	case SIOCGDRLST_IN6:
1436 		/*
1437 		 * obsolete API, use sysctl under net.inet6.icmp6
1438 		 */
1439 		bzero(drl, sizeof(*drl));
1440 		mtx_lock(&nd6_mtx);
1441 		dr = TAILQ_FIRST(&nd_defrouter);
1442 		while (dr && i < DRLSTSIZ) {
1443 			drl->defrouter[i].rtaddr = dr->rtaddr;
1444 			if (IN6_IS_ADDR_LINKLOCAL(&drl->defrouter[i].rtaddr)) {
1445 				/* XXX: need to this hack for KAME stack */
1446 				drl->defrouter[i].rtaddr.s6_addr16[1] = 0;
1447 			} else
1448 				log(LOG_ERR,
1449 				    "default router list contains a "
1450 				    "non-linklocal address(%s)\n",
1451 				    ip6_sprintf(&drl->defrouter[i].rtaddr));
1452 
1453 			drl->defrouter[i].flags = dr->flags;
1454 			drl->defrouter[i].rtlifetime = dr->rtlifetime;
1455 			drl->defrouter[i].expire = dr->expire;
1456 			drl->defrouter[i].if_index = dr->ifp->if_index;
1457 			i++;
1458 			dr = TAILQ_NEXT(dr, dr_entry);
1459 		}
1460 		mtx_unlock(&nd6_mtx);
1461 		break;
1462 	case SIOCGPRLST_IN6:
1463 		/*
1464 		 * obsolete API, use sysctl under net.inet6.icmp6
1465 		 */
1466 		/*
1467 		 * XXX meaning of fields, especialy "raflags", is very
1468 		 * differnet between RA prefix list and RR/static prefix list.
1469 		 * how about separating ioctls into two?
1470 		 */
1471 		bzero(prl, sizeof(*prl));
1472 		mtx_lock(&nd6_mtx);
1473 		pr = nd_prefix.lh_first;
1474 		while (pr && i < PRLSTSIZ) {
1475 			struct nd_pfxrouter *pfr;
1476 			int j;
1477 
1478 			in6_embedscope(&prl->prefix[i].prefix,
1479 			    &pr->ndpr_prefix, NULL, NULL);
1480 			prl->prefix[i].raflags = pr->ndpr_raf;
1481 			prl->prefix[i].prefixlen = pr->ndpr_plen;
1482 			prl->prefix[i].vltime = pr->ndpr_vltime;
1483 			prl->prefix[i].pltime = pr->ndpr_pltime;
1484 			prl->prefix[i].if_index = pr->ndpr_ifp->if_index;
1485 			prl->prefix[i].expire = pr->ndpr_expire;
1486 
1487 			pfr = pr->ndpr_advrtrs.lh_first;
1488 			j = 0;
1489 			while (pfr) {
1490 				if (j < DRLSTSIZ) {
1491 #define RTRADDR prl->prefix[i].advrtr[j]
1492 					RTRADDR = pfr->router->rtaddr;
1493 					if (IN6_IS_ADDR_LINKLOCAL(&RTRADDR)) {
1494 						/* XXX: hack for KAME */
1495 						RTRADDR.s6_addr16[1] = 0;
1496 					} else
1497 						log(LOG_ERR,
1498 						    "a router(%s) advertises "
1499 						    "a prefix with "
1500 						    "non-link local address\n",
1501 						    ip6_sprintf(&RTRADDR));
1502 #undef RTRADDR
1503 				}
1504 				j++;
1505 				pfr = pfr->pfr_next;
1506 			}
1507 			prl->prefix[i].advrtrs = j;
1508 			prl->prefix[i].origin = PR_ORIG_RA;
1509 
1510 			i++;
1511 			pr = pr->ndpr_next;
1512 		}
1513 		mtx_unlock(&nd6_mtx);
1514 
1515 		break;
1516 	case OSIOCGIFINFO_IN6:
1517 		/* XXX: old ndp(8) assumes a positive value for linkmtu. */
1518 		bzero(&ndi->ndi, sizeof(ndi->ndi));
1519 		ndi->ndi.linkmtu = IN6_LINKMTU(ifp);
1520 		ndi->ndi.maxmtu = ND_IFINFO(ifp)->maxmtu;
1521 		ndi->ndi.basereachable = ND_IFINFO(ifp)->basereachable;
1522 		ndi->ndi.reachable = ND_IFINFO(ifp)->reachable;
1523 		ndi->ndi.retrans = ND_IFINFO(ifp)->retrans;
1524 		ndi->ndi.flags = ND_IFINFO(ifp)->flags;
1525 		ndi->ndi.recalctm = ND_IFINFO(ifp)->recalctm;
1526 		ndi->ndi.chlim = ND_IFINFO(ifp)->chlim;
1527 		break;
1528 	case SIOCGIFINFO_IN6:
1529 		ndi->ndi = *ND_IFINFO(ifp);
1530 		ndi->ndi.linkmtu = IN6_LINKMTU(ifp);
1531 		break;
1532 	case SIOCSIFINFO_IN6:
1533 		/*
1534 		 * used to change host variables from userland.
1535 		 * intented for a use on router to reflect RA configurations.
1536 		 */
1537 		/* 0 means 'unspecified' */
1538 		if (ndi->ndi.linkmtu != 0) {
1539 			if (ndi->ndi.linkmtu < IPV6_MMTU ||
1540 			    ndi->ndi.linkmtu > IN6_LINKMTU(ifp)) {
1541 				error = EINVAL;
1542 				break;
1543 			}
1544 			ND_IFINFO(ifp)->linkmtu = ndi->ndi.linkmtu;
1545 		}
1546 
1547 		if (ndi->ndi.basereachable != 0) {
1548 			int obasereachable = ND_IFINFO(ifp)->basereachable;
1549 
1550 			ND_IFINFO(ifp)->basereachable = ndi->ndi.basereachable;
1551 			if (ndi->ndi.basereachable != obasereachable)
1552 				ND_IFINFO(ifp)->reachable =
1553 				    ND_COMPUTE_RTIME(ndi->ndi.basereachable);
1554 		}
1555 		if (ndi->ndi.retrans != 0)
1556 			ND_IFINFO(ifp)->retrans = ndi->ndi.retrans;
1557 		if (ndi->ndi.chlim != 0)
1558 			ND_IFINFO(ifp)->chlim = ndi->ndi.chlim;
1559 		/* FALLTHROUGH */
1560 	case SIOCSIFINFO_FLAGS:
1561 		if ((ndi->ndi.flags & ND6_IFF_AUTO_LINKLOCAL) &&
1562 		    !(ND_IFINFO(ifp)->flags & ND6_IFF_AUTO_LINKLOCAL)) {
1563 			/* auto_linklocal 0->1 transision */
1564 			ND_IFINFO(ifp)->flags |= ND6_IFF_AUTO_LINKLOCAL;
1565 			in6_ifattach(ifp, NULL);
1566 		}
1567 		ND_IFINFO(ifp)->flags = ndi->ndi.flags;
1568 		break;
1569 	case SIOCSNDFLUSH_IN6:	/* XXX: the ioctl name is confusing... */
1570 		/* flush default router list */
1571 		/*
1572 		 * xxx sumikawa: should not delete route if default
1573 		 * route equals to the top of default router list
1574 		 */
1575 		bzero(&any, sizeof(any));
1576 		defrouter_delreq(&any, 0);
1577 		defrouter_select();
1578 		/* xxx sumikawa: flush prefix list */
1579 		break;
1580 	case SIOCSPFXFLUSH_IN6:
1581 	{
1582 		/* flush all the prefix advertised by routers */
1583 		struct nd_prefix *pr, *next;
1584 
1585 		mtx_lock(&nd6_mtx);
1586 		for (pr = nd_prefix.lh_first; pr; pr = next) {
1587 			struct in6_ifaddr *ia, *ia_next;
1588 
1589 			next = pr->ndpr_next;
1590 
1591 			if (IN6_IS_ADDR_LINKLOCAL(&pr->ndpr_prefix.sin6_addr))
1592 				continue; /* XXX */
1593 
1594 			/* do we really have to remove addresses as well? */
1595 			for (ia = in6_ifaddr; ia; ia = ia_next) {
1596 				/* ia might be removed.  keep the next ptr. */
1597 				ia_next = ia->ia_next;
1598 
1599 				if (!(ia->ia6_flags & IN6_IFF_AUTOCONF))
1600 					continue;
1601 
1602 				if (ia->ia6_ndpr == pr)
1603 					in6_purgeaddr(&ia->ia_ifa);
1604 			}
1605 			prelist_remove(pr);
1606 		}
1607 		mtx_unlock(&nd6_mtx);
1608 		break;
1609 	}
1610 	case SIOCSRTRFLUSH_IN6:
1611 	{
1612 		/* flush all the default routers */
1613 		struct nd_defrouter *dr, *next;
1614 
1615 		mtx_lock(&nd6_mtx);
1616 		if ((dr = TAILQ_FIRST(&nd_defrouter)) != NULL) {
1617 			/*
1618 			 * The first entry of the list may be stored in
1619 			 * the routing table, so we'll delete it later.
1620 			 */
1621 			for (dr = TAILQ_NEXT(dr, dr_entry); dr; dr = next) {
1622 				next = TAILQ_NEXT(dr, dr_entry);
1623 				defrtrlist_del(dr);
1624 			}
1625 			defrtrlist_del(TAILQ_FIRST(&nd_defrouter));
1626 		}
1627 		mtx_unlock(&nd6_mtx);
1628 		break;
1629 	}
1630 	case SIOCGNBRINFO_IN6:
1631 	{
1632 		struct llinfo_nd6 *ln;
1633 		struct in6_addr nb_addr = nbi->addr; /* make local for safety */
1634 
1635 		/*
1636 		 * XXX: KAME specific hack for scoped addresses
1637 		 *      XXXX: for other scopes than link-local?
1638 		 */
1639 		if (IN6_IS_ADDR_LINKLOCAL(&nbi->addr) ||
1640 		    IN6_IS_ADDR_MC_LINKLOCAL(&nbi->addr)) {
1641 			u_int16_t *idp = (u_int16_t *)&nb_addr.s6_addr[2];
1642 
1643 			if (*idp == 0)
1644 				*idp = htons(ifp->if_index);
1645 		}
1646 
1647 		mtx_lock(&nd6_mtx);
1648 		if ((rt = nd6_lookup(&nb_addr, 0, ifp)) == NULL) {
1649 			error = EINVAL;
1650 			mtx_unlock(&nd6_mtx);
1651 			break;
1652 		}
1653 		ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1654 		nbi->state = ln->ln_state;
1655 		nbi->asked = ln->ln_asked;
1656 		nbi->isrouter = ln->ln_router;
1657 		nbi->expire = ln->ln_expire;
1658 		mtx_unlock(&nd6_mtx);
1659 
1660 		break;
1661 	}
1662 	case SIOCGDEFIFACE_IN6:	/* XXX: should be implemented as a sysctl? */
1663 		ndif->ifindex = nd6_defifindex;
1664 		break;
1665 	case SIOCSDEFIFACE_IN6:	/* XXX: should be implemented as a sysctl? */
1666 		return (nd6_setdefaultiface(ndif->ifindex));
1667 	}
1668 	return (error);
1669 }
1670 
1671 /*
1672  * Create neighbor cache entry and cache link-layer address,
1673  * on reception of inbound ND6 packets.  (RS/RA/NS/redirect)
1674  */
1675 struct rtentry *
nd6_cache_lladdr(struct ifnet * ifp,struct in6_addr * from,char * lladdr,int lladdrlen,int type,int code)1676 nd6_cache_lladdr(struct ifnet *ifp, struct in6_addr *from, char *lladdr,
1677 		 int lladdrlen,
1678 		 int type,	/* ICMP6 type */
1679 		 int code	/* type dependent information */)
1680 {
1681 	struct rtentry *rt = NULL;
1682 	struct llinfo_nd6 *ln = NULL;
1683 	int is_newentry;
1684 	struct sockaddr_dl *sdl = NULL;
1685 	int do_update;
1686 	int olladdr;
1687 	int llchange;
1688 	int newstate = 0;
1689 
1690 	if (!ifp)
1691 		panic("ifp == NULL in nd6_cache_lladdr");
1692 	if (!from)
1693 		panic("from == NULL in nd6_cache_lladdr");
1694 
1695 	/* nothing must be updated for unspecified address */
1696 	if (IN6_IS_ADDR_UNSPECIFIED(from))
1697 		return NULL;
1698 
1699 	/*
1700 	 * Validation about ifp->if_addrlen and lladdrlen must be done in
1701 	 * the caller.
1702 	 *
1703 	 * XXX If the link does not have link-layer adderss, what should
1704 	 * we do? (ifp->if_addrlen == 0)
1705 	 * Spec says nothing in sections for RA, RS and NA.  There's small
1706 	 * description on it in NS section (RFC 2461 7.2.3).
1707 	 */
1708 
1709 	rt = nd6_lookup(from, 0, ifp);
1710 	if (!rt) {
1711 #if 0
1712 		/* nothing must be done if there's no lladdr */
1713 		if (!lladdr || !lladdrlen)
1714 			return NULL;
1715 #endif
1716 
1717 		rt = nd6_lookup(from, 1, ifp);
1718 		is_newentry = 1;
1719 	} else {
1720 		/* do nothing if static ndp is set */
1721 		if (rt->rt_flags & RTF_STATIC)
1722 			return NULL;
1723 		is_newentry = 0;
1724 	}
1725 
1726 	if (!rt)
1727 		return NULL;
1728 	if ((rt->rt_flags & (RTF_GATEWAY | RTF_LLINFO)) != RTF_LLINFO) {
1729 fail:
1730 		nd6_free(rt);
1731 		return NULL;
1732 	}
1733 	ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1734 	if (!ln)
1735 		goto fail;
1736 	if (!rt->rt_gateway)
1737 		goto fail;
1738 	if (rt->rt_gateway->sa_family != AF_LINK)
1739 		goto fail;
1740 	sdl = SDL(rt->rt_gateway);
1741 
1742 	olladdr = (sdl->sdl_alen) ? 1 : 0;
1743 	if (olladdr && lladdr) {
1744 		if (bcmp(lladdr, LLADDR(sdl), ifp->if_addrlen))
1745 			llchange = 1;
1746 		else
1747 			llchange = 0;
1748 	} else
1749 		llchange = 0;
1750 
1751 	/*
1752 	 * newentry olladdr  lladdr  llchange	(*=record)
1753 	 *	0	n	n	--	(1)
1754 	 *	0	y	n	--	(2)
1755 	 *	0	n	y	--	(3) * STALE
1756 	 *	0	y	y	n	(4) *
1757 	 *	0	y	y	y	(5) * STALE
1758 	 *	1	--	n	--	(6)   NOSTATE(= PASSIVE)
1759 	 *	1	--	y	--	(7) * STALE
1760 	 */
1761 
1762 	if (lladdr) {		/* (3-5) and (7) */
1763 		/*
1764 		 * Record source link-layer address
1765 		 * XXX is it dependent to ifp->if_type?
1766 		 */
1767 		sdl->sdl_alen = ifp->if_addrlen;
1768 		bcopy(lladdr, LLADDR(sdl), ifp->if_addrlen);
1769 	}
1770 
1771 	if (!is_newentry) {
1772 		if ((!olladdr && lladdr) ||		/* (3) */
1773 		    (olladdr && lladdr && llchange)) {	/* (5) */
1774 			do_update = 1;
1775 			newstate = ND6_LLINFO_STALE;
1776 		} else {				/* (1-2,4) */
1777 			do_update = 0;
1778 		}
1779 	} else {
1780 		do_update = 1;
1781 		if (!lladdr)				/* (6) */
1782 			newstate = ND6_LLINFO_NOSTATE;
1783 		else					/* (7) */
1784 			newstate = ND6_LLINFO_STALE;
1785 	}
1786 
1787 	if (do_update) {
1788 		/*
1789 		 * Update the state of the neighbor cache.
1790 		 */
1791 		ln->ln_state = newstate;
1792 
1793 		if (ln->ln_state == ND6_LLINFO_STALE) {
1794 			/*
1795 			 * XXX: since nd6_output() below will cause
1796 			 * state tansition to DELAY and reset the timer,
1797 			 * we must set the timer now, although it is actually
1798 			 * meaningless.
1799 			 */
1800 			ln->ln_expire = time_uptime + nd6_gctimer;
1801 
1802 			if (ln->ln_hold) {
1803 				/*
1804 				 * we assume ifp is not a p2p here, so just
1805 				 * set the 2nd argument as the 1st one.
1806 				 */
1807 				nd6_output(ifp, ifp, ln->ln_hold,
1808 				    (struct sockaddr_in6 *)rt_key(rt), rt);
1809 				ln->ln_hold = NULL;
1810 			}
1811 		} else if (ln->ln_state == ND6_LLINFO_INCOMPLETE) {
1812 			/* probe right away */
1813 			ln->ln_expire = time_uptime;
1814 		}
1815 	}
1816 
1817 	/*
1818 	 * ICMP6 type dependent behavior.
1819 	 *
1820 	 * NS: clear IsRouter if new entry
1821 	 * RS: clear IsRouter
1822 	 * RA: set IsRouter if there's lladdr
1823 	 * redir: clear IsRouter if new entry
1824 	 *
1825 	 * RA case, (1):
1826 	 * The spec says that we must set IsRouter in the following cases:
1827 	 * - If lladdr exist, set IsRouter.  This means (1-5).
1828 	 * - If it is old entry (!newentry), set IsRouter.  This means (7).
1829 	 * So, based on the spec, in (1-5) and (7) cases we must set IsRouter.
1830 	 * A quetion arises for (1) case.  (1) case has no lladdr in the
1831 	 * neighbor cache, this is similar to (6).
1832 	 * This case is rare but we figured that we MUST NOT set IsRouter.
1833 	 *
1834 	 * newentry olladdr  lladdr  llchange	    NS  RS  RA	redir
1835 	 *							D R
1836 	 *	0	n	n	--	(1)	c   ?     s
1837 	 *	0	y	n	--	(2)	c   s     s
1838 	 *	0	n	y	--	(3)	c   s     s
1839 	 *	0	y	y	n	(4)	c   s     s
1840 	 *	0	y	y	y	(5)	c   s     s
1841 	 *	1	--	n	--	(6) c	c 	c s
1842 	 *	1	--	y	--	(7) c	c   s	c s
1843 	 *
1844 	 *					(c=clear s=set)
1845 	 */
1846 	switch (type & 0xff) {
1847 	case ND_NEIGHBOR_SOLICIT:
1848 		/*
1849 		 * New entry must have is_router flag cleared.
1850 		 */
1851 		if (is_newentry)	/* (6-7) */
1852 			ln->ln_router = 0;
1853 		break;
1854 	case ND_REDIRECT:
1855 		/*
1856 		 * If the icmp is a redirect to a better router, always set the
1857 		 * is_router flag.  Otherwise, if the entry is newly created,
1858 		 * clear the flag.  [RFC 2461, sec 8.3]
1859 		 */
1860 		if (code == ND_REDIRECT_ROUTER)
1861 			ln->ln_router = 1;
1862 		else if (is_newentry) /* (6-7) */
1863 			ln->ln_router = 0;
1864 		break;
1865 	case ND_ROUTER_SOLICIT:
1866 		/*
1867 		 * is_router flag must always be cleared.
1868 		 */
1869 		ln->ln_router = 0;
1870 		break;
1871 	case ND_ROUTER_ADVERT:
1872 		/*
1873 		 * Mark an entry with lladdr as a router.
1874 		 */
1875 		if ((!is_newentry && (olladdr || lladdr)) ||	/* (2-5) */
1876 		    (is_newentry && lladdr)) {			/* (7) */
1877 			ln->ln_router = 1;
1878 		}
1879 		break;
1880 	}
1881 
1882 	if (llchange || lladdr)
1883 		rt_rtmsg(llchange ? RTM_CHANGE : RTM_ADD, rt, rt->rt_ifp, 0);
1884 
1885 	/*
1886 	 * When the link-layer address of a router changes, select the
1887 	 * best router again.  In particular, when the neighbor entry is newly
1888 	 * created, it might affect the selection policy.
1889 	 * Question: can we restrict the first condition to the "is_newentry"
1890 	 * case?
1891 	 * XXX: when we hear an RA from a new router with the link-layer
1892 	 * address option, defrouter_select() is called twice, since
1893 	 * defrtrlist_update called the function as well.  However, I believe
1894 	 * we can compromise the overhead, since it only happens the first
1895 	 * time.
1896 	 * XXX: although defrouter_select() should not have a bad effect
1897 	 * for those are not autoconfigured hosts, we explicitly avoid such
1898 	 * cases for safety.
1899 	 */
1900 	if (do_update && ln->ln_router && !ip6_forwarding &&
1901 	    (ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV))
1902 		defrouter_select();
1903 
1904 	return rt;
1905 }
1906 
1907 static void
nd6_slowtimo(void * arg __unused)1908 nd6_slowtimo(void *arg __unused)
1909 {
1910 	struct lwkt_msg *lmsg = &nd6_slowtimo_netmsg.lmsg;
1911 
1912 	KASSERT(mycpuid == 0, ("not on cpu0"));
1913 	crit_enter();
1914 	if (lmsg->ms_flags & MSGF_DONE)
1915 		lwkt_sendmsg_oncpu(netisr_cpuport(0), lmsg);
1916 	crit_exit();
1917 }
1918 
1919 static void
nd6_slowtimo_dispatch(netmsg_t nmsg)1920 nd6_slowtimo_dispatch(netmsg_t nmsg)
1921 {
1922 	const struct ifnet_array *arr;
1923 	struct nd_ifinfo *nd6if;
1924 	int i;
1925 
1926 	ASSERT_NETISR0;
1927 
1928 	crit_enter();
1929 	lwkt_replymsg(&nmsg->lmsg, 0);	/* reply ASAP */
1930 	crit_exit();
1931 
1932 	arr = ifnet_array_get();
1933 
1934 	mtx_lock(&nd6_mtx);
1935 	for (i = 0; i < arr->ifnet_count; ++i) {
1936 		struct ifnet *ifp = arr->ifnet_arr[i];
1937 
1938 		if (ifp->if_afdata[AF_INET6] == NULL)
1939 			continue;
1940 		nd6if = ND_IFINFO(ifp);
1941 		if (nd6if->basereachable && /* already initialized */
1942 		    (nd6if->recalctm -= ND6_SLOWTIMER_INTERVAL) <= 0) {
1943 			/*
1944 			 * Since reachable time rarely changes by router
1945 			 * advertisements, we SHOULD insure that a new random
1946 			 * value gets recomputed at least once every few hours.
1947 			 * (RFC 2461, 6.3.4)
1948 			 */
1949 			nd6if->recalctm = nd6_recalc_reachtm_interval;
1950 			nd6if->reachable = ND_COMPUTE_RTIME(nd6if->basereachable);
1951 		}
1952 	}
1953 	mtx_unlock(&nd6_mtx);
1954 
1955 	callout_reset(&nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
1956 	    nd6_slowtimo, NULL);
1957 }
1958 
1959 int
nd6_output(struct ifnet * ifp,struct ifnet * origifp,struct mbuf * m,struct sockaddr_in6 * dst,struct rtentry * rt)1960 nd6_output(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *m,
1961 	   struct sockaddr_in6 *dst, struct rtentry *rt)
1962 {
1963 	int error;
1964 
1965 	if (ifp->if_flags & IFF_LOOPBACK)
1966 		error = ifp->if_output(origifp, m, (struct sockaddr *)dst, rt);
1967 	else
1968 		error = ifp->if_output(ifp, m, (struct sockaddr *)dst, rt);
1969 	return error;
1970 }
1971 
1972 int
nd6_resolve(struct ifnet * ifp,struct rtentry * rt0,struct mbuf * m,struct sockaddr * dst0,u_char * desten)1973 nd6_resolve(struct ifnet *ifp, struct rtentry *rt0, struct mbuf *m,
1974     struct sockaddr *dst0, u_char *desten)
1975 {
1976 	struct sockaddr_in6 *dst = SIN6(dst0);
1977 	struct rtentry *rt = NULL;
1978 	struct llinfo_nd6 *ln = NULL;
1979 	int error;
1980 
1981 	if (m->m_flags & M_MCAST) {
1982 		switch (ifp->if_type) {
1983 		case IFT_ETHER:
1984 #ifdef IFT_L2VLAN
1985 		case IFT_L2VLAN:
1986 #endif
1987 #ifdef IFT_IEEE80211
1988 		case IFT_IEEE80211:
1989 #endif
1990 			ETHER_MAP_IPV6_MULTICAST(&dst->sin6_addr,
1991 						 desten);
1992 			return 0;
1993 		case IFT_IEEE1394:
1994 			bcopy(ifp->if_broadcastaddr, desten, ifp->if_addrlen);
1995 			return 0;
1996 		default:
1997 			error = EAFNOSUPPORT;
1998 			goto bad;
1999 		}
2000 	}
2001 
2002 	if (rt0 != NULL) {
2003 		error = rt_llroute(dst0, rt0, &rt);
2004 		if (error != 0)
2005 			goto bad;
2006 		ln = rt->rt_llinfo;
2007 	}
2008 
2009 	/*
2010 	 * Since nd6_is_addr_neighbor() internally calls nd6_lookup(),
2011 	 * the condition below is not very efficient.  But we believe
2012 	 * it is tolerable, because this should be a rare case.
2013 	 */
2014 	if (ln == NULL && nd6_is_addr_neighbor(dst, ifp)) {
2015 		rt = nd6_lookup(&dst->sin6_addr, 1, ifp);
2016 		if (rt != NULL)
2017 			ln = rt->rt_llinfo;
2018 	}
2019 
2020 	if (ln == NULL || rt == NULL) {
2021 		if (!(ifp->if_flags & IFF_POINTOPOINT) &&
2022 		    !(ND_IFINFO(ifp)->flags & ND6_IFF_PERFORMNUD)) {
2023 			log(LOG_DEBUG,
2024 			    "nd6_output: can't allocate llinfo for %s "
2025 			    "(ln=%p, rt=%p)\n",
2026 			    ip6_sprintf(&dst->sin6_addr), ln, rt);
2027 			error = ENOBUFS;
2028 			goto bad;
2029 		}
2030 		return 0;
2031 	}
2032 
2033 	/* We don't have to do link-layer address resolution on a p2p link. */
2034 	if ((ifp->if_flags & IFF_POINTOPOINT) &&
2035 	    ln->ln_state < ND6_LLINFO_REACHABLE) {
2036 		ln->ln_state = ND6_LLINFO_STALE;
2037 		ln->ln_expire = time_uptime + nd6_gctimer;
2038 	}
2039 
2040 	/*
2041 	 * The first time we send a packet to a neighbor whose entry is
2042 	 * STALE, we have to change the state to DELAY and a sets a timer to
2043 	 * expire in DELAY_FIRST_PROBE_TIME seconds to ensure do
2044 	 * neighbor unreachability detection on expiration.
2045 	 * (RFC 2461 7.3.3)
2046 	 */
2047 	if (ln->ln_state == ND6_LLINFO_STALE) {
2048 		ln->ln_asked = 0;
2049 		ln->ln_state = ND6_LLINFO_DELAY;
2050 		ln->ln_expire = time_uptime + nd6_delay;
2051 	}
2052 
2053 	/*
2054 	 * If the neighbor cache entry has a state other than INCOMPLETE
2055 	 * (i.e. its link-layer address is already resolved), return it.
2056 	 */
2057 	if (ln->ln_state > ND6_LLINFO_INCOMPLETE) {
2058 		struct sockaddr_dl *sdl = SDL(rt->rt_gateway);
2059 
2060 		KKASSERT(sdl->sdl_family == AF_LINK && sdl->sdl_alen != 0);
2061 		bcopy(LLADDR(sdl), desten, sdl->sdl_alen);
2062 		return 0;
2063 	}
2064 
2065 	/*
2066 	 * There is a neighbor cache entry, but no ethernet address
2067 	 * response yet.  Replace the held mbuf (if any) with this
2068 	 * latest one.
2069 	 */
2070 	if (ln->ln_hold)
2071 		m_freem(ln->ln_hold);
2072 	ln->ln_hold = m;
2073 
2074 	/*
2075 	 * This code conforms to the rate-limiting rule described in Section
2076 	 * 7.2.2 of RFC 2461, because the timer is set correctly after sending
2077 	 * an NS below.
2078 	 */
2079 	if (ln->ln_state == ND6_LLINFO_NOSTATE ||
2080 	    ln->ln_state == ND6_LLINFO_WAITDELETE) {
2081 		/*
2082 		 * This neighbor cache entry was just created; change its
2083 		 * state to INCOMPLETE and start its life cycle.
2084 		 *
2085 		 * We force an NS output below by setting ln_expire to 1
2086 		 * (nd6_rtrequest() could set it to the current time_uptime)
2087 		 * and zeroing out ln_asked (XXX this may not be necessary).
2088 		 */
2089 		ln->ln_state = ND6_LLINFO_INCOMPLETE;
2090 		ln->ln_expire = 1;
2091 		ln->ln_asked = 0;
2092 	}
2093 	if (ln->ln_expire && ln->ln_expire < time_uptime && ln->ln_asked == 0) {
2094 		ln->ln_asked++;
2095 		ln->ln_expire = time_uptime + ND_IFINFO(ifp)->retrans / 1000;
2096 		nd6_ns_output(ifp, NULL, &dst->sin6_addr, ln, 0);
2097 	}
2098 
2099 	if (ln->ln_asked >= nd6_mmaxtries)
2100 		return (rt != NULL && rt->rt_flags & RTF_GATEWAY) ?
2101 		    EHOSTUNREACH : EHOSTDOWN;
2102 	return EWOULDBLOCK;
2103 
2104 bad:
2105 	m_freem(m);
2106 	return error;
2107 }
2108 
2109 int
nd6_need_cache(struct ifnet * ifp)2110 nd6_need_cache(struct ifnet *ifp)
2111 {
2112 	/*
2113 	 * XXX: we currently do not make neighbor cache on any interface
2114 	 * other than Ethernet and GIF.
2115 	 *
2116 	 * RFC2893 says:
2117 	 * - unidirectional tunnels needs no ND
2118 	 */
2119 	switch (ifp->if_type) {
2120 	case IFT_ETHER:
2121 	case IFT_IEEE1394:
2122 #ifdef IFT_L2VLAN
2123 	case IFT_L2VLAN:
2124 #endif
2125 #ifdef IFT_IEEE80211
2126 	case IFT_IEEE80211:
2127 #endif
2128 #ifdef IFT_CARP
2129 	case IFT_CARP:
2130 #endif
2131 	case IFT_GIF:		/* XXX need more cases? */
2132 		return (1);
2133 	default:
2134 		return (0);
2135 	}
2136 }
2137 
2138 static int nd6_sysctl_drlist(SYSCTL_HANDLER_ARGS);
2139 static int nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS);
2140 #ifdef SYSCTL_DECL
2141 SYSCTL_DECL(_net_inet6_icmp6);
2142 #endif
2143 SYSCTL_NODE(_net_inet6_icmp6, ICMPV6CTL_ND6_DRLIST, nd6_drlist,
2144 	CTLFLAG_RD, nd6_sysctl_drlist, "List default routers");
2145 SYSCTL_NODE(_net_inet6_icmp6, ICMPV6CTL_ND6_PRLIST, nd6_prlist,
2146 	CTLFLAG_RD, nd6_sysctl_prlist, "List prefixes");
2147 
2148 static int
nd6_sysctl_drlist(SYSCTL_HANDLER_ARGS)2149 nd6_sysctl_drlist(SYSCTL_HANDLER_ARGS)
2150 {
2151 	int error;
2152 	char buf[1024];
2153 	struct in6_defrouter *d, *de;
2154 	struct nd_defrouter *dr;
2155 
2156 	if (req->newptr)
2157 		return EPERM;
2158 	error = 0;
2159 
2160 	for (dr = TAILQ_FIRST(&nd_defrouter); dr;
2161 	     dr = TAILQ_NEXT(dr, dr_entry)) {
2162 		d = (struct in6_defrouter *)buf;
2163 		de = (struct in6_defrouter *)(buf + sizeof(buf));
2164 
2165 		if (d + 1 <= de) {
2166 			bzero(d, sizeof(*d));
2167 			d->rtaddr.sin6_family = AF_INET6;
2168 			d->rtaddr.sin6_len = sizeof(d->rtaddr);
2169 			if (in6_recoverscope(&d->rtaddr, &dr->rtaddr,
2170 			    dr->ifp) != 0)
2171 				log(LOG_ERR,
2172 				    "scope error in "
2173 				    "default router list (%s)\n",
2174 				    ip6_sprintf(&dr->rtaddr));
2175 			d->flags = dr->flags;
2176 			d->rtlifetime = dr->rtlifetime;
2177 			d->expire = dr->expire;
2178 			d->if_index = dr->ifp->if_index;
2179 		} else
2180 			panic("buffer too short");
2181 
2182 		error = SYSCTL_OUT(req, buf, sizeof(*d));
2183 		if (error)
2184 			break;
2185 	}
2186 	return error;
2187 }
2188 
2189 static int
nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS)2190 nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS)
2191 {
2192 	int error;
2193 	char buf[1024];
2194 	struct in6_prefix *p, *pe;
2195 	struct nd_prefix *pr;
2196 
2197 	if (req->newptr)
2198 		return EPERM;
2199 	error = 0;
2200 
2201 	for (pr = nd_prefix.lh_first; pr; pr = pr->ndpr_next) {
2202 		u_short advrtrs;
2203 		size_t advance;
2204 		struct sockaddr_in6 *sin6, *s6;
2205 		struct nd_pfxrouter *pfr;
2206 
2207 		p = (struct in6_prefix *)buf;
2208 		pe = (struct in6_prefix *)(buf + sizeof(buf));
2209 
2210 		if (p + 1 <= pe) {
2211 			bzero(p, sizeof(*p));
2212 			sin6 = (struct sockaddr_in6 *)(p + 1);
2213 
2214 			p->prefix = pr->ndpr_prefix;
2215 			if (in6_recoverscope(&p->prefix,
2216 			    &p->prefix.sin6_addr, pr->ndpr_ifp) != 0)
2217 				log(LOG_ERR,
2218 				    "scope error in prefix list (%s)\n",
2219 				    ip6_sprintf(&p->prefix.sin6_addr));
2220 			p->raflags = pr->ndpr_raf;
2221 			p->prefixlen = pr->ndpr_plen;
2222 			p->vltime = pr->ndpr_vltime;
2223 			p->pltime = pr->ndpr_pltime;
2224 			p->if_index = pr->ndpr_ifp->if_index;
2225 			p->expire = pr->ndpr_expire;
2226 			p->refcnt = pr->ndpr_refcnt;
2227 			p->flags = pr->ndpr_stateflags;
2228 			p->origin = PR_ORIG_RA;
2229 			advrtrs = 0;
2230 			for (pfr = pr->ndpr_advrtrs.lh_first; pfr;
2231 			     pfr = pfr->pfr_next) {
2232 				if ((void *)&sin6[advrtrs + 1] > (void *)pe) {
2233 					advrtrs++;
2234 					continue;
2235 				}
2236 				s6 = &sin6[advrtrs];
2237 				bzero(s6, sizeof(*s6));
2238 				s6->sin6_family = AF_INET6;
2239 				s6->sin6_len = sizeof(*sin6);
2240 				if (in6_recoverscope(s6, &pfr->router->rtaddr,
2241 				    pfr->router->ifp) != 0)
2242 					log(LOG_ERR,
2243 					    "scope error in "
2244 					    "prefix list (%s)\n",
2245 					    ip6_sprintf(&pfr->router->rtaddr));
2246 				advrtrs++;
2247 			}
2248 			p->advrtrs = advrtrs;
2249 		} else {
2250 			panic("buffer too short");
2251 		}
2252 
2253 		advance = sizeof(*p) + sizeof(*sin6) * advrtrs;
2254 		error = SYSCTL_OUT(req, buf, advance);
2255 		if (error)
2256 			break;
2257 	}
2258 	return error;
2259 }
2260