Home | History | Annotate | Line # | Download | only in netinet6
nd6.c revision 1.48
      1 /*	$NetBSD: nd6.c,v 1.48 2001/06/27 17:36:14 itojun Exp $	*/
      2 /*	$KAME: nd6.c,v 1.151 2001/06/19 14:24:41 sumikawa 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 /*
     34  * XXX
     35  * KAME 970409 note:
     36  * BSD/OS version heavily modifies this code, related to llinfo.
     37  * Since we don't have BSD/OS version of net/route.c in our hand,
     38  * I left the code mostly as it was in 970310.  -- itojun
     39  */
     40 
     41 #include <sys/param.h>
     42 #include <sys/systm.h>
     43 #include <sys/callout.h>
     44 #include <sys/malloc.h>
     45 #include <sys/mbuf.h>
     46 #include <sys/socket.h>
     47 #include <sys/sockio.h>
     48 #include <sys/time.h>
     49 #include <sys/kernel.h>
     50 #include <sys/protosw.h>
     51 #include <sys/errno.h>
     52 #include <sys/ioctl.h>
     53 #include <sys/syslog.h>
     54 #include <sys/queue.h>
     55 
     56 #include <net/if.h>
     57 #include <net/if_dl.h>
     58 #include <net/if_types.h>
     59 #include <net/if_atm.h>
     60 #include <net/if_ieee1394.h>
     61 #include <net/route.h>
     62 
     63 #include <netinet/in.h>
     64 #include <net/if_ether.h>
     65 #include <netinet/if_inarp.h>
     66 #include <net/if_fddi.h>
     67 #include <netinet6/in6_var.h>
     68 #include <netinet/ip6.h>
     69 #include <netinet6/ip6_var.h>
     70 #include <netinet6/nd6.h>
     71 #include <netinet6/in6_prefix.h>
     72 #include <netinet/icmp6.h>
     73 
     74 #include "loop.h"
     75 extern struct ifnet loif[NLOOP];
     76 
     77 #include <net/net_osdep.h>
     78 
     79 #define ND6_SLOWTIMER_INTERVAL (60 * 60) /* 1 hour */
     80 #define ND6_RECALC_REACHTM_INTERVAL (60 * 120) /* 2 hours */
     81 
     82 #define SIN6(s) ((struct sockaddr_in6 *)s)
     83 #define SDL(s) ((struct sockaddr_dl *)s)
     84 
     85 /* timer values */
     86 int	nd6_prune	= 1;	/* walk list every 1 seconds */
     87 int	nd6_delay	= 5;	/* delay first probe time 5 second */
     88 int	nd6_umaxtries	= 3;	/* maximum unicast query */
     89 int	nd6_mmaxtries	= 3;	/* maximum multicast query */
     90 int	nd6_useloopback = 1;	/* use loopback interface for local traffic */
     91 int	nd6_gctimer	= (60 * 60 * 24); /* 1 day: garbage collection timer */
     92 
     93 /* preventing too many loops in ND option parsing */
     94 int nd6_maxndopt = 10;	/* max # of ND options allowed */
     95 
     96 int nd6_maxnudhint = 0;	/* max # of subsequent upper layer hints */
     97 
     98 #ifdef ND6_DEBUG
     99 int nd6_debug = 1;
    100 #else
    101 int nd6_debug = 0;
    102 #endif
    103 
    104 /* for debugging? */
    105 static int nd6_inuse, nd6_allocated;
    106 
    107 struct llinfo_nd6 llinfo_nd6 = {&llinfo_nd6, &llinfo_nd6};
    108 static size_t nd_ifinfo_indexlim = 8;
    109 struct nd_ifinfo *nd_ifinfo = NULL;
    110 struct nd_drhead nd_defrouter;
    111 struct nd_prhead nd_prefix = { 0 };
    112 
    113 int nd6_recalc_reachtm_interval = ND6_RECALC_REACHTM_INTERVAL;
    114 static struct sockaddr_in6 all1_sa;
    115 
    116 static void nd6_slowtimo __P((void *));
    117 
    118 struct callout nd6_slowtimo_ch;
    119 struct callout nd6_timer_ch;
    120 
    121 void
    122 nd6_init()
    123 {
    124 	static int nd6_init_done = 0;
    125 	int i;
    126 
    127 	if (nd6_init_done) {
    128 		log(LOG_NOTICE, "nd6_init called more than once(ignored)\n");
    129 		return;
    130 	}
    131 
    132 	all1_sa.sin6_family = AF_INET6;
    133 	all1_sa.sin6_len = sizeof(struct sockaddr_in6);
    134 	for (i = 0; i < sizeof(all1_sa.sin6_addr); i++)
    135 		all1_sa.sin6_addr.s6_addr[i] = 0xff;
    136 
    137 	/* initialization of the default router list */
    138 	TAILQ_INIT(&nd_defrouter);
    139 
    140 	nd6_init_done = 1;
    141 
    142 	/* start timer */
    143 	callout_reset(&nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
    144 	    nd6_slowtimo, NULL);
    145 }
    146 
    147 void
    148 nd6_ifattach(ifp)
    149 	struct ifnet *ifp;
    150 {
    151 
    152 	/*
    153 	 * We have some arrays that should be indexed by if_index.
    154 	 * since if_index will grow dynamically, they should grow too.
    155 	 */
    156 	if (nd_ifinfo == NULL || if_index >= nd_ifinfo_indexlim) {
    157 		size_t n;
    158 		caddr_t q;
    159 
    160 		while (if_index >= nd_ifinfo_indexlim)
    161 			nd_ifinfo_indexlim <<= 1;
    162 
    163 		/* grow nd_ifinfo */
    164 		n = nd_ifinfo_indexlim * sizeof(struct nd_ifinfo);
    165 		q = (caddr_t)malloc(n, M_IP6NDP, M_WAITOK);
    166 		bzero(q, n);
    167 		if (nd_ifinfo) {
    168 			bcopy((caddr_t)nd_ifinfo, q, n/2);
    169 			free((caddr_t)nd_ifinfo, M_IP6NDP);
    170 		}
    171 		nd_ifinfo = (struct nd_ifinfo *)q;
    172 	}
    173 
    174 #define ND nd_ifinfo[ifp->if_index]
    175 
    176 	/*
    177 	 * Don't initialize if called twice.
    178 	 * XXX: to detect this, we should choose a member that is never set
    179 	 * before initialization of the ND structure itself.  We formaly used
    180 	 * the linkmtu member, which was not suitable because it could be
    181 	 * initialized via "ifconfig mtu".
    182 	 */
    183 	if (ND.basereachable)
    184 		return;
    185 
    186 	ND.linkmtu = ifindex2ifnet[ifp->if_index]->if_mtu;
    187 	ND.chlim = IPV6_DEFHLIM;
    188 	ND.basereachable = REACHABLE_TIME;
    189 	ND.reachable = ND_COMPUTE_RTIME(ND.basereachable);
    190 	ND.retrans = RETRANS_TIMER;
    191 	ND.receivedra = 0;
    192 	ND.flags = ND6_IFF_PERFORMNUD;
    193 	nd6_setmtu(ifp);
    194 #undef ND
    195 }
    196 
    197 /*
    198  * Reset ND level link MTU. This function is called when the physical MTU
    199  * changes, which means we might have to adjust the ND level MTU.
    200  */
    201 void
    202 nd6_setmtu(ifp)
    203 	struct ifnet *ifp;
    204 {
    205 	struct nd_ifinfo *ndi = &nd_ifinfo[ifp->if_index];
    206 	u_long oldmaxmtu = ndi->maxmtu;
    207 	u_long oldlinkmtu = ndi->linkmtu;
    208 
    209 	switch (ifp->if_type) {
    210 	case IFT_ARCNET:	/* XXX MTU handling needs more work */
    211 		ndi->maxmtu = MIN(60480, ifp->if_mtu);
    212 		break;
    213 	case IFT_ETHER:
    214 		ndi->maxmtu = MIN(ETHERMTU, ifp->if_mtu);
    215 		break;
    216 	case IFT_ATM:
    217 		ndi->maxmtu = MIN(ATMMTU, ifp->if_mtu);
    218 		break;
    219 	case IFT_IEEE1394:
    220 		ndi->maxmtu = MIN(IEEE1394MTU, ifp->if_mtu);
    221 		break;
    222 	default:
    223 		ndi->maxmtu = ifp->if_mtu;
    224 		break;
    225 	}
    226 
    227 	if (oldmaxmtu != ndi->maxmtu) {
    228 		/*
    229 		 * If the ND level MTU is not set yet, or if the maxmtu
    230 		 * is reset to a smaller value than the ND level MTU,
    231 		 * also reset the ND level MTU.
    232 		 */
    233 		if (ndi->linkmtu == 0 ||
    234 		    ndi->maxmtu < ndi->linkmtu) {
    235 			ndi->linkmtu = ndi->maxmtu;
    236 			/* also adjust in6_maxmtu if necessary. */
    237 			if (oldlinkmtu == 0) {
    238 				/*
    239 				 * XXX: the case analysis is grotty, but
    240 				 * it is not efficient to call in6_setmaxmtu()
    241 				 * here when we are during the initialization
    242 				 * procedure.
    243 				 */
    244 				if (in6_maxmtu < ndi->linkmtu)
    245 					in6_maxmtu = ndi->linkmtu;
    246 			} else
    247 				in6_setmaxmtu();
    248 		}
    249 	}
    250 #undef MIN
    251 }
    252 
    253 void
    254 nd6_option_init(opt, icmp6len, ndopts)
    255 	void *opt;
    256 	int icmp6len;
    257 	union nd_opts *ndopts;
    258 {
    259 	bzero(ndopts, sizeof(*ndopts));
    260 	ndopts->nd_opts_search = (struct nd_opt_hdr *)opt;
    261 	ndopts->nd_opts_last
    262 		= (struct nd_opt_hdr *)(((u_char *)opt) + icmp6len);
    263 
    264 	if (icmp6len == 0) {
    265 		ndopts->nd_opts_done = 1;
    266 		ndopts->nd_opts_search = NULL;
    267 	}
    268 }
    269 
    270 /*
    271  * Take one ND option.
    272  */
    273 struct nd_opt_hdr *
    274 nd6_option(ndopts)
    275 	union nd_opts *ndopts;
    276 {
    277 	struct nd_opt_hdr *nd_opt;
    278 	int olen;
    279 
    280 	if (!ndopts)
    281 		panic("ndopts == NULL in nd6_option\n");
    282 	if (!ndopts->nd_opts_last)
    283 		panic("uninitialized ndopts in nd6_option\n");
    284 	if (!ndopts->nd_opts_search)
    285 		return NULL;
    286 	if (ndopts->nd_opts_done)
    287 		return NULL;
    288 
    289 	nd_opt = ndopts->nd_opts_search;
    290 
    291 	/* make sure nd_opt_len is inside the buffer */
    292 	if ((caddr_t)&nd_opt->nd_opt_len >= (caddr_t)ndopts->nd_opts_last) {
    293 		bzero(ndopts, sizeof(*ndopts));
    294 		return NULL;
    295 	}
    296 
    297 	olen = nd_opt->nd_opt_len << 3;
    298 	if (olen == 0) {
    299 		/*
    300 		 * Message validation requires that all included
    301 		 * options have a length that is greater than zero.
    302 		 */
    303 		bzero(ndopts, sizeof(*ndopts));
    304 		return NULL;
    305 	}
    306 
    307 	ndopts->nd_opts_search = (struct nd_opt_hdr *)((caddr_t)nd_opt + olen);
    308 	if (ndopts->nd_opts_search > ndopts->nd_opts_last) {
    309 		/* option overruns the end of buffer, invalid */
    310 		bzero(ndopts, sizeof(*ndopts));
    311 		return NULL;
    312 	} else if (ndopts->nd_opts_search == ndopts->nd_opts_last) {
    313 		/* reached the end of options chain */
    314 		ndopts->nd_opts_done = 1;
    315 		ndopts->nd_opts_search = NULL;
    316 	}
    317 	return nd_opt;
    318 }
    319 
    320 /*
    321  * Parse multiple ND options.
    322  * This function is much easier to use, for ND routines that do not need
    323  * multiple options of the same type.
    324  */
    325 int
    326 nd6_options(ndopts)
    327 	union nd_opts *ndopts;
    328 {
    329 	struct nd_opt_hdr *nd_opt;
    330 	int i = 0;
    331 
    332 	if (!ndopts)
    333 		panic("ndopts == NULL in nd6_options\n");
    334 	if (!ndopts->nd_opts_last)
    335 		panic("uninitialized ndopts in nd6_options\n");
    336 	if (!ndopts->nd_opts_search)
    337 		return 0;
    338 
    339 	while (1) {
    340 		nd_opt = nd6_option(ndopts);
    341 		if (!nd_opt && !ndopts->nd_opts_last) {
    342 			/*
    343 			 * Message validation requires that all included
    344 			 * options have a length that is greater than zero.
    345 			 */
    346 			icmp6stat.icp6s_nd_badopt++;
    347 			bzero(ndopts, sizeof(*ndopts));
    348 			return -1;
    349 		}
    350 
    351 		if (!nd_opt)
    352 			goto skip1;
    353 
    354 		switch (nd_opt->nd_opt_type) {
    355 		case ND_OPT_SOURCE_LINKADDR:
    356 		case ND_OPT_TARGET_LINKADDR:
    357 		case ND_OPT_MTU:
    358 		case ND_OPT_REDIRECTED_HEADER:
    359 			if (ndopts->nd_opt_array[nd_opt->nd_opt_type]) {
    360 				nd6log((LOG_INFO,
    361 				    "duplicated ND6 option found (type=%d)\n",
    362 				    nd_opt->nd_opt_type));
    363 				/* XXX bark? */
    364 			} else {
    365 				ndopts->nd_opt_array[nd_opt->nd_opt_type]
    366 					= nd_opt;
    367 			}
    368 			break;
    369 		case ND_OPT_PREFIX_INFORMATION:
    370 			if (ndopts->nd_opt_array[nd_opt->nd_opt_type] == 0) {
    371 				ndopts->nd_opt_array[nd_opt->nd_opt_type]
    372 					= nd_opt;
    373 			}
    374 			ndopts->nd_opts_pi_end =
    375 				(struct nd_opt_prefix_info *)nd_opt;
    376 			break;
    377 		default:
    378 			/*
    379 			 * Unknown options must be silently ignored,
    380 			 * to accomodate future extension to the protocol.
    381 			 */
    382 			nd6log((LOG_DEBUG,
    383 			    "nd6_options: unsupported option %d - "
    384 			    "option ignored\n", nd_opt->nd_opt_type));
    385 		}
    386 
    387 skip1:
    388 		i++;
    389 		if (i > nd6_maxndopt) {
    390 			icmp6stat.icp6s_nd_toomanyopt++;
    391 			nd6log((LOG_INFO, "too many loop in nd opt\n"));
    392 			break;
    393 		}
    394 
    395 		if (ndopts->nd_opts_done)
    396 			break;
    397 	}
    398 
    399 	return 0;
    400 }
    401 
    402 /*
    403  * ND6 timer routine to expire default route list and prefix list
    404  */
    405 void
    406 nd6_timer(ignored_arg)
    407 	void	*ignored_arg;
    408 {
    409 	int s;
    410 	struct llinfo_nd6 *ln;
    411 	struct nd_defrouter *dr;
    412 	struct nd_prefix *pr;
    413 	long time_second = time.tv_sec;
    414 
    415 	s = splsoftnet();
    416 	callout_reset(&nd6_timer_ch, nd6_prune * hz,
    417 	    nd6_timer, NULL);
    418 
    419 	ln = llinfo_nd6.ln_next;
    420 	/* XXX BSD/OS separates this code -- itojun */
    421 	while (ln && ln != &llinfo_nd6) {
    422 		struct rtentry *rt;
    423 		struct ifnet *ifp;
    424 		struct sockaddr_in6 *dst;
    425 		struct llinfo_nd6 *next = ln->ln_next;
    426 		/* XXX: used for the DELAY case only: */
    427 		struct nd_ifinfo *ndi = NULL;
    428 
    429 		if ((rt = ln->ln_rt) == NULL) {
    430 			ln = next;
    431 			continue;
    432 		}
    433 		if ((ifp = rt->rt_ifp) == NULL) {
    434 			ln = next;
    435 			continue;
    436 		}
    437 		ndi = &nd_ifinfo[ifp->if_index];
    438 		dst = (struct sockaddr_in6 *)rt_key(rt);
    439 
    440 		if (ln->ln_expire > time_second) {
    441 			ln = next;
    442 			continue;
    443 		}
    444 
    445 		/* sanity check */
    446 		if (!rt)
    447 			panic("rt=0 in nd6_timer(ln=%p)\n", ln);
    448 		if (rt->rt_llinfo && (struct llinfo_nd6 *)rt->rt_llinfo != ln)
    449 			panic("rt_llinfo(%p) is not equal to ln(%p)\n",
    450 			      rt->rt_llinfo, ln);
    451 		if (!dst)
    452 			panic("dst=0 in nd6_timer(ln=%p)\n", ln);
    453 
    454 		switch (ln->ln_state) {
    455 		case ND6_LLINFO_INCOMPLETE:
    456 			if (ln->ln_asked < nd6_mmaxtries) {
    457 				ln->ln_asked++;
    458 				ln->ln_expire = time_second +
    459 					nd_ifinfo[ifp->if_index].retrans / 1000;
    460 				nd6_ns_output(ifp, NULL, &dst->sin6_addr,
    461 					ln, 0);
    462 			} else {
    463 				struct mbuf *m = ln->ln_hold;
    464 				if (m) {
    465 					if (rt->rt_ifp) {
    466 						/*
    467 						 * Fake rcvif to make ICMP error
    468 						 * more helpful in diagnosing
    469 						 * for the receiver.
    470 						 * XXX: should we consider
    471 						 * older rcvif?
    472 						 */
    473 						m->m_pkthdr.rcvif = rt->rt_ifp;
    474 					}
    475 					icmp6_error(m, ICMP6_DST_UNREACH,
    476 						    ICMP6_DST_UNREACH_ADDR, 0);
    477 					ln->ln_hold = NULL;
    478 				}
    479 				next = nd6_free(rt);
    480 			}
    481 			break;
    482 		case ND6_LLINFO_REACHABLE:
    483 			if (ln->ln_expire) {
    484 				ln->ln_state = ND6_LLINFO_STALE;
    485 				ln->ln_expire = time_second + nd6_gctimer;
    486 			}
    487 			break;
    488 
    489 		case ND6_LLINFO_STALE:
    490 			/* Garbage Collection(RFC 2461 5.3) */
    491 			if (ln->ln_expire)
    492 				next = nd6_free(rt);
    493 			break;
    494 
    495 		case ND6_LLINFO_DELAY:
    496 			if (ndi && (ndi->flags & ND6_IFF_PERFORMNUD) != 0) {
    497 				/* We need NUD */
    498 				ln->ln_asked = 1;
    499 				ln->ln_state = ND6_LLINFO_PROBE;
    500 				ln->ln_expire = time_second +
    501 					ndi->retrans / 1000;
    502 				nd6_ns_output(ifp, &dst->sin6_addr,
    503 					      &dst->sin6_addr,
    504 					      ln, 0);
    505 			} else {
    506 				ln->ln_state = ND6_LLINFO_STALE; /* XXX */
    507 				ln->ln_expire = time_second + nd6_gctimer;
    508 			}
    509 			break;
    510 		case ND6_LLINFO_PROBE:
    511 			if (ln->ln_asked < nd6_umaxtries) {
    512 				ln->ln_asked++;
    513 				ln->ln_expire = time_second +
    514 					nd_ifinfo[ifp->if_index].retrans / 1000;
    515 				nd6_ns_output(ifp, &dst->sin6_addr,
    516 					       &dst->sin6_addr, ln, 0);
    517 			} else
    518 				next = nd6_free(rt);
    519 			break;
    520 		}
    521 		ln = next;
    522 	}
    523 
    524 	/* expire default router list */
    525 	dr = TAILQ_FIRST(&nd_defrouter);
    526 	while (dr) {
    527 		if (dr->expire && dr->expire < time_second) {
    528 			struct nd_defrouter *t;
    529 			t = TAILQ_NEXT(dr, dr_entry);
    530 			defrtrlist_del(dr);
    531 			dr = t;
    532 		} else {
    533 			dr = TAILQ_NEXT(dr, dr_entry);
    534 		}
    535 	}
    536 	pr = nd_prefix.lh_first;
    537 	while (pr) {
    538 		struct in6_ifaddr *ia6;
    539 		struct in6_addrlifetime *lt6;
    540 
    541 		if (IN6_IS_ADDR_UNSPECIFIED(&pr->ndpr_addr))
    542 			ia6 = NULL;
    543 		else
    544 			ia6 = in6ifa_ifpwithaddr(pr->ndpr_ifp, &pr->ndpr_addr);
    545 
    546 		if (ia6) {
    547 			/* check address lifetime */
    548 			lt6 = &ia6->ia6_lifetime;
    549 			if (lt6->ia6t_preferred && lt6->ia6t_preferred < time_second)
    550 				ia6->ia6_flags |= IN6_IFF_DEPRECATED;
    551 			if (lt6->ia6t_expire && lt6->ia6t_expire < time_second) {
    552 				if (!IN6_IS_ADDR_UNSPECIFIED(&pr->ndpr_addr))
    553 					in6_ifdel(pr->ndpr_ifp, &pr->ndpr_addr);
    554 				/* xxx ND_OPT_PI_FLAG_ONLINK processing */
    555 			}
    556 		}
    557 
    558 		/*
    559 		 * check prefix lifetime.
    560 		 * since pltime is just for autoconf, pltime processing for
    561 		 * prefix is not necessary.
    562 		 *
    563 		 * we offset expire time by NDPR_KEEP_EXPIRE, so that we
    564 		 * can use the old prefix information to validate the
    565 		 * next prefix information to come.  See prelist_update()
    566 		 * for actual validation.
    567 		 */
    568 		if (pr->ndpr_expire
    569 		 && pr->ndpr_expire + NDPR_KEEP_EXPIRED < time_second) {
    570 			struct nd_prefix *t;
    571 			t = pr->ndpr_next;
    572 
    573 			/*
    574 			 * address expiration and prefix expiration are
    575 			 * separate.  NEVER perform in6_ifdel here.
    576 			 */
    577 
    578 			prelist_remove(pr);
    579 			pr = t;
    580 		} else
    581 			pr = pr->ndpr_next;
    582 	}
    583 	splx(s);
    584 }
    585 
    586 /*
    587  * Nuke neighbor cache/prefix/default router management table, right before
    588  * ifp goes away.
    589  */
    590 void
    591 nd6_purge(ifp)
    592 	struct ifnet *ifp;
    593 {
    594 	struct llinfo_nd6 *ln, *nln;
    595 	struct nd_defrouter *dr, *ndr, drany;
    596 	struct nd_prefix *pr, *npr;
    597 
    598 	/* Nuke default router list entries toward ifp */
    599 	if ((dr = TAILQ_FIRST(&nd_defrouter)) != NULL) {
    600 		/*
    601 		 * The first entry of the list may be stored in
    602 		 * the routing table, so we'll delete it later.
    603 		 */
    604 		for (dr = TAILQ_NEXT(dr, dr_entry); dr; dr = ndr) {
    605 			ndr = TAILQ_NEXT(dr, dr_entry);
    606 			if (dr->ifp == ifp)
    607 				defrtrlist_del(dr);
    608 		}
    609 		dr = TAILQ_FIRST(&nd_defrouter);
    610 		if (dr->ifp == ifp)
    611 			defrtrlist_del(dr);
    612 	}
    613 
    614 	/* Nuke prefix list entries toward ifp */
    615 	for (pr = nd_prefix.lh_first; pr; pr = npr) {
    616 		npr = pr->ndpr_next;
    617 		if (pr->ndpr_ifp == ifp) {
    618 			if (!IN6_IS_ADDR_UNSPECIFIED(&pr->ndpr_addr))
    619 				in6_ifdel(pr->ndpr_ifp, &pr->ndpr_addr);
    620 			prelist_remove(pr);
    621 		}
    622 	}
    623 
    624 	/* cancel default outgoing interface setting */
    625 	if (nd6_defifindex == ifp->if_index)
    626 		nd6_setdefaultiface(0);
    627 
    628 	if (!ip6_forwarding && ip6_accept_rtadv) { /* XXX: too restrictive? */
    629 		/* refresh default router list */
    630 		bzero(&drany, sizeof(drany));
    631 		defrouter_delreq(&drany, 0);
    632 		defrouter_select();
    633 	}
    634 
    635 	/*
    636 	 * Nuke neighbor cache entries for the ifp.
    637 	 * Note that rt->rt_ifp may not be the same as ifp,
    638 	 * due to KAME goto ours hack.  See RTM_RESOLVE case in
    639 	 * nd6_rtrequest(), and ip6_input().
    640 	 */
    641 	ln = llinfo_nd6.ln_next;
    642 	while (ln && ln != &llinfo_nd6) {
    643 		struct rtentry *rt;
    644 		struct sockaddr_dl *sdl;
    645 
    646 		nln = ln->ln_next;
    647 		rt = ln->ln_rt;
    648 		if (rt && rt->rt_gateway &&
    649 		    rt->rt_gateway->sa_family == AF_LINK) {
    650 			sdl = (struct sockaddr_dl *)rt->rt_gateway;
    651 			if (sdl->sdl_index == ifp->if_index)
    652 				nln = nd6_free(rt);
    653 		}
    654 		ln = nln;
    655 	}
    656 }
    657 
    658 struct rtentry *
    659 nd6_lookup(addr6, create, ifp)
    660 	struct in6_addr *addr6;
    661 	int create;
    662 	struct ifnet *ifp;
    663 {
    664 	struct rtentry *rt;
    665 	struct sockaddr_in6 sin6;
    666 
    667 	bzero(&sin6, sizeof(sin6));
    668 	sin6.sin6_len = sizeof(struct sockaddr_in6);
    669 	sin6.sin6_family = AF_INET6;
    670 	sin6.sin6_addr = *addr6;
    671 	rt = rtalloc1((struct sockaddr *)&sin6, create);
    672 	if (rt && (rt->rt_flags & RTF_LLINFO) == 0) {
    673 		/*
    674 		 * This is the case for the default route.
    675 		 * If we want to create a neighbor cache for the address, we
    676 		 * should free the route for the destination and allocate an
    677 		 * interface route.
    678 		 */
    679 		if (create) {
    680 			RTFREE(rt);
    681 			rt = 0;
    682 		}
    683 	}
    684 	if (!rt) {
    685 		if (create && ifp) {
    686 			int e;
    687 
    688 			/*
    689 			 * If no route is available and create is set,
    690 			 * we allocate a host route for the destination
    691 			 * and treat it like an interface route.
    692 			 * This hack is necessary for a neighbor which can't
    693 			 * be covered by our own prefix.
    694 			 */
    695 			struct ifaddr *ifa =
    696 				ifaof_ifpforaddr((struct sockaddr *)&sin6, ifp);
    697 			if (ifa == NULL)
    698 				return(NULL);
    699 
    700 			/*
    701 			 * Create a new route. RTF_LLINFO is necessary
    702 			 * to create a Neighbor Cache entry for the
    703 			 * destination in nd6_rtrequest which will be
    704 			 * called in rtequest via ifa->ifa_rtrequest.
    705 			 */
    706 			if ((e = rtrequest(RTM_ADD, (struct sockaddr *)&sin6,
    707 					   ifa->ifa_addr,
    708 					   (struct sockaddr *)&all1_sa,
    709 					   (ifa->ifa_flags |
    710 					    RTF_HOST | RTF_LLINFO) &
    711 					   ~RTF_CLONING,
    712 					   &rt)) != 0)
    713 				log(LOG_ERR,
    714 				    "nd6_lookup: failed to add route for a "
    715 				    "neighbor(%s), errno=%d\n",
    716 				    ip6_sprintf(addr6), e);
    717 			if (rt == NULL)
    718 				return(NULL);
    719 			if (rt->rt_llinfo) {
    720 				struct llinfo_nd6 *ln =
    721 					(struct llinfo_nd6 *)rt->rt_llinfo;
    722 				ln->ln_state = ND6_LLINFO_NOSTATE;
    723 			}
    724 		} else
    725 			return(NULL);
    726 	}
    727 	rt->rt_refcnt--;
    728 	/*
    729 	 * Validation for the entry.
    730 	 * XXX: we can't use rt->rt_ifp to check for the interface, since
    731 	 *      it might be the loopback interface if the entry is for our
    732 	 *      own address on a non-loopback interface. Instead, we should
    733 	 *      use rt->rt_ifa->ifa_ifp, which would specify the REAL interface.
    734 	 */
    735 	if ((rt->rt_flags & RTF_GATEWAY) || (rt->rt_flags & RTF_LLINFO) == 0 ||
    736 	    rt->rt_gateway->sa_family != AF_LINK ||
    737 	    (ifp && rt->rt_ifa->ifa_ifp != ifp)) {
    738 		if (create) {
    739 			log(LOG_DEBUG, "nd6_lookup: failed to lookup %s (if = %s)\n",
    740 			    ip6_sprintf(addr6), ifp ? if_name(ifp) : "unspec");
    741 			/* xxx more logs... kazu */
    742 		}
    743 		return(0);
    744 	}
    745 	return(rt);
    746 }
    747 
    748 /*
    749  * Detect if a given IPv6 address identifies a neighbor on a given link.
    750  * XXX: should take care of the destination of a p2p link?
    751  */
    752 int
    753 nd6_is_addr_neighbor(addr, ifp)
    754 	struct sockaddr_in6 *addr;
    755 	struct ifnet *ifp;
    756 {
    757 	struct ifaddr *ifa;
    758 	int i;
    759 
    760 #define IFADDR6(a) ((((struct in6_ifaddr *)(a))->ia_addr).sin6_addr)
    761 #define IFMASK6(a) ((((struct in6_ifaddr *)(a))->ia_prefixmask).sin6_addr)
    762 
    763 	/*
    764 	 * A link-local address is always a neighbor.
    765 	 * XXX: we should use the sin6_scope_id field rather than the embedded
    766 	 * interface index.
    767 	 */
    768 	if (IN6_IS_ADDR_LINKLOCAL(&addr->sin6_addr) &&
    769 	    ntohs(*(u_int16_t *)&addr->sin6_addr.s6_addr[2]) == ifp->if_index)
    770 		return(1);
    771 
    772 	/*
    773 	 * If the address matches one of our addresses,
    774 	 * it should be a neighbor.
    775 	 */
    776 	for (ifa = ifp->if_addrlist.tqh_first;
    777 	     ifa;
    778 	     ifa = ifa->ifa_list.tqe_next)
    779 	{
    780 		if (ifa->ifa_addr->sa_family != AF_INET6)
    781 			next: continue;
    782 
    783 		for (i = 0; i < 4; i++) {
    784 			if ((IFADDR6(ifa).s6_addr32[i] ^
    785 			     addr->sin6_addr.s6_addr32[i]) &
    786 			    IFMASK6(ifa).s6_addr32[i])
    787 				goto next;
    788 		}
    789 		return(1);
    790 	}
    791 
    792 	/*
    793 	 * Even if the address matches none of our addresses, it might be
    794 	 * in the neighbor cache.
    795 	 */
    796 	if (nd6_lookup(&addr->sin6_addr, 0, ifp))
    797 		return(1);
    798 
    799 	return(0);
    800 #undef IFADDR6
    801 #undef IFMASK6
    802 }
    803 
    804 /*
    805  * Free an nd6 llinfo entry.
    806  */
    807 struct llinfo_nd6 *
    808 nd6_free(rt)
    809 	struct rtentry *rt;
    810 {
    811 	struct llinfo_nd6 *ln = (struct llinfo_nd6 *)rt->rt_llinfo, *next;
    812 	struct in6_addr in6 = ((struct sockaddr_in6 *)rt_key(rt))->sin6_addr;
    813 	struct nd_defrouter *dr;
    814 
    815 	/*
    816 	 * Clear all destination cache entries for the neighbor.
    817 	 * XXX: is it better to restrict this to hosts?
    818 	 */
    819 	pfctlinput(PRC_HOSTDEAD, rt_key(rt));
    820 
    821 	if (!ip6_forwarding && ip6_accept_rtadv) { /* XXX: too restrictive? */
    822 		int s;
    823 		s = splsoftnet();
    824 		dr = defrouter_lookup(&((struct sockaddr_in6 *)rt_key(rt))->sin6_addr,
    825 				      rt->rt_ifp);
    826 		if (ln->ln_router || dr) {
    827 			/*
    828 			 * rt6_flush must be called whether or not the neighbor
    829 			 * is in the Default Router List.
    830 			 * See a corresponding comment in nd6_na_input().
    831 			 */
    832 			rt6_flush(&in6, rt->rt_ifp);
    833 		}
    834 
    835 		if (dr) {
    836 			/*
    837 			 * Unreachablity of a router might affect the default
    838 			 * router selection and on-link detection of advertised
    839 			 * prefixes.
    840 			 */
    841 
    842 			/*
    843 			 * Temporarily fake the state to choose a new default
    844 			 * router and to perform on-link determination of
    845 			 * prefixes coreectly.
    846 			 * Below the state will be set correctly,
    847 			 * or the entry itself will be deleted.
    848 			 */
    849 			ln->ln_state = ND6_LLINFO_INCOMPLETE;
    850 
    851 			if (dr == TAILQ_FIRST(&nd_defrouter)) {
    852 				/*
    853 				 * It is used as the current default router,
    854 				 * so we have to move it to the end of the
    855 				 * list and choose a new one.
    856 				 * XXX: it is not very efficient if this is
    857 				 *      the only router.
    858 				 */
    859 				TAILQ_REMOVE(&nd_defrouter, dr, dr_entry);
    860 				TAILQ_INSERT_TAIL(&nd_defrouter, dr, dr_entry);
    861 
    862 				defrouter_select();
    863 			}
    864 			pfxlist_onlink_check();
    865 		}
    866 		splx(s);
    867 	}
    868 
    869 	/*
    870 	 * Before deleting the entry, remember the next entry as the
    871 	 * return value.  We need this because pfxlist_onlink_check() above
    872 	 * might have freed other entries (particularly the old next entry) as
    873 	 * a side effect (XXX).
    874 	 */
    875 	next = ln->ln_next;
    876 
    877 	/*
    878 	 * Detach the route from the routing tree and the list of neighbor
    879 	 * caches, and disable the route entry not to be used in already
    880 	 * cached routes.
    881 	 */
    882 	rtrequest(RTM_DELETE, rt_key(rt), (struct sockaddr *)0,
    883 		  rt_mask(rt), 0, (struct rtentry **)0);
    884 
    885 	return(next);
    886 }
    887 
    888 /*
    889  * Upper-layer reachability hint for Neighbor Unreachability Detection.
    890  *
    891  * XXX cost-effective metods?
    892  */
    893 void
    894 nd6_nud_hint(rt, dst6, force)
    895 	struct rtentry *rt;
    896 	struct in6_addr *dst6;
    897 	int force;
    898 {
    899 	struct llinfo_nd6 *ln;
    900 	long time_second = time.tv_sec;
    901 
    902 	/*
    903 	 * If the caller specified "rt", use that.  Otherwise, resolve the
    904 	 * routing table by supplied "dst6".
    905 	 */
    906 	if (!rt) {
    907 		if (!dst6)
    908 			return;
    909 		if (!(rt = nd6_lookup(dst6, 0, NULL)))
    910 			return;
    911 	}
    912 
    913 	if ((rt->rt_flags & RTF_GATEWAY) != 0 ||
    914 	    (rt->rt_flags & RTF_LLINFO) == 0 ||
    915 	    !rt->rt_llinfo || !rt->rt_gateway ||
    916 	    rt->rt_gateway->sa_family != AF_LINK) {
    917 		/* This is not a host route. */
    918 		return;
    919 	}
    920 
    921 	ln = (struct llinfo_nd6 *)rt->rt_llinfo;
    922 	if (ln->ln_state < ND6_LLINFO_REACHABLE)
    923 		return;
    924 
    925 	/*
    926 	 * if we get upper-layer reachability confirmation many times,
    927 	 * it is possible we have false information.
    928 	 */
    929 	if (!force) {
    930 		ln->ln_byhint++;
    931 		if (ln->ln_byhint > nd6_maxnudhint)
    932 			return;
    933 	}
    934 
    935 	ln->ln_state = ND6_LLINFO_REACHABLE;
    936 	if (ln->ln_expire)
    937 		ln->ln_expire = time_second +
    938 			nd_ifinfo[rt->rt_ifp->if_index].reachable;
    939 }
    940 
    941 #ifdef OLDIP6OUTPUT
    942 /*
    943  * Resolve an IP6 address into an ethernet address. If success,
    944  * desten is filled in. If there is no entry in ndptab,
    945  * set one up and multicast a solicitation for the IP6 address.
    946  * Hold onto this mbuf and resend it once the address
    947  * is finally resolved. A return value of 1 indicates
    948  * that desten has been filled in and the packet should be sent
    949  * normally; a 0 return indicates that the packet has been
    950  * taken over here, either now or for later transmission.
    951  */
    952 int
    953 nd6_resolve(ifp, rt, m, dst, desten)
    954 	struct ifnet *ifp;
    955 	struct rtentry *rt;
    956 	struct mbuf *m;
    957 	struct sockaddr *dst;
    958 	u_char *desten;
    959 {
    960 	struct llinfo_nd6 *ln = (struct llinfo_nd6 *)NULL;
    961 	struct sockaddr_dl *sdl;
    962 	long time_second = time.tv_sec;
    963 
    964 	if (m->m_flags & M_MCAST) {
    965 		switch (ifp->if_type) {
    966 		case IFT_ETHER:
    967 		case IFT_FDDI:
    968 			ETHER_MAP_IPV6_MULTICAST(&SIN6(dst)->sin6_addr,
    969 						 desten);
    970 			return(1);
    971 		case IFT_IEEE1394:
    972 			bcopy(ifp->if_broadcastaddr, desten, ifp->if_addrlen);
    973 			return(1);
    974 		case IFT_ARCNET:
    975 			*desten = 0;
    976 			return(1);
    977 		default:
    978 			m_freem(m);
    979 			return(0);
    980 		}
    981 	}
    982 	if (rt && (rt->rt_flags & RTF_LLINFO) != 0)
    983 		ln = (struct llinfo_nd6 *)rt->rt_llinfo;
    984 	else {
    985 		if ((rt = nd6_lookup(&(SIN6(dst)->sin6_addr), 1, ifp)) != NULL)
    986 			ln = (struct llinfo_nd6 *)rt->rt_llinfo;
    987 	}
    988 	if (!ln || !rt) {
    989 		log(LOG_DEBUG, "nd6_resolve: can't allocate llinfo for %s\n",
    990 			ip6_sprintf(&(SIN6(dst)->sin6_addr)));
    991 		m_freem(m);
    992 		return(0);
    993 	}
    994 	sdl = SDL(rt->rt_gateway);
    995 	/*
    996 	 * Ckeck the address family and length is valid, the address
    997 	 * is resolved; otherwise, try to resolve.
    998 	 */
    999 	if (ln->ln_state >= ND6_LLINFO_REACHABLE
   1000 	   && sdl->sdl_family == AF_LINK
   1001 	   && sdl->sdl_alen != 0) {
   1002 		bcopy(LLADDR(sdl), desten, sdl->sdl_alen);
   1003 		if (ln->ln_state == ND6_LLINFO_STALE) {
   1004 			ln->ln_asked = 0;
   1005 			ln->ln_state = ND6_LLINFO_DELAY;
   1006 			ln->ln_expire = time_second + nd6_delay;
   1007 		}
   1008 		return(1);
   1009 	}
   1010 	/*
   1011 	 * There is an ndp entry, but no ethernet address
   1012 	 * response yet. Replace the held mbuf with this
   1013 	 * latest one.
   1014 	 *
   1015 	 * XXX Does the code conform to rate-limiting rule?
   1016 	 * (RFC 2461 7.2.2)
   1017 	 */
   1018 	if (ln->ln_state == ND6_LLINFO_NOSTATE)
   1019 		ln->ln_state = ND6_LLINFO_INCOMPLETE;
   1020 	if (ln->ln_hold)
   1021 		m_freem(ln->ln_hold);
   1022 	ln->ln_hold = m;
   1023 	if (ln->ln_expire) {
   1024 		if (ln->ln_asked < nd6_mmaxtries &&
   1025 		    ln->ln_expire < time_second) {
   1026 			ln->ln_asked++;
   1027 			ln->ln_expire = time_second +
   1028 				nd_ifinfo[ifp->if_index].retrans / 1000;
   1029 			nd6_ns_output(ifp, NULL, &(SIN6(dst)->sin6_addr),
   1030 				ln, 0);
   1031 		}
   1032 	}
   1033 	/* do not free mbuf here, it is queued into llinfo_nd6 */
   1034 	return(0);
   1035 }
   1036 #endif /* OLDIP6OUTPUT */
   1037 
   1038 void
   1039 nd6_rtrequest(req, rt, info)
   1040 	int	req;
   1041 	struct rtentry *rt;
   1042 	struct rt_addrinfo *info; /* xxx unused */
   1043 {
   1044 	struct sockaddr *gate = rt->rt_gateway;
   1045 	struct llinfo_nd6 *ln = (struct llinfo_nd6 *)rt->rt_llinfo;
   1046 	static struct sockaddr_dl null_sdl = {sizeof(null_sdl), AF_LINK};
   1047 	struct ifnet *ifp = rt->rt_ifp;
   1048 	struct ifaddr *ifa;
   1049 	long time_second = time.tv_sec;
   1050 
   1051 	if (rt->rt_flags & RTF_GATEWAY)
   1052 		return;
   1053 
   1054 	switch (req) {
   1055 	case RTM_ADD:
   1056 		/*
   1057 		 * There is no backward compatibility :)
   1058 		 *
   1059 		 * if ((rt->rt_flags & RTF_HOST) == 0 &&
   1060 		 *     SIN(rt_mask(rt))->sin_addr.s_addr != 0xffffffff)
   1061 		 *	   rt->rt_flags |= RTF_CLONING;
   1062 		 */
   1063 		if (rt->rt_flags & (RTF_CLONING | RTF_LLINFO)) {
   1064 			/*
   1065 			 * Case 1: This route should come from
   1066 			 * a route to interface. RTF_LLINFO flag is set
   1067 			 * for a host route whose destination should be
   1068 			 * treated as on-link.
   1069 			 */
   1070 			rt_setgate(rt, rt_key(rt),
   1071 				   (struct sockaddr *)&null_sdl);
   1072 			gate = rt->rt_gateway;
   1073 			SDL(gate)->sdl_type = ifp->if_type;
   1074 			SDL(gate)->sdl_index = ifp->if_index;
   1075 			if (ln)
   1076 				ln->ln_expire = time_second;
   1077 #if 1
   1078 			if (ln && ln->ln_expire == 0) {
   1079 				/* kludge for desktops */
   1080 #if 0
   1081 				printf("nd6_request: time.tv_sec is zero; "
   1082 				       "treat it as 1\n");
   1083 #endif
   1084 				ln->ln_expire = 1;
   1085 			}
   1086 #endif
   1087 			if (rt->rt_flags & RTF_CLONING)
   1088 				break;
   1089 		}
   1090 		/*
   1091 		 * In IPv4 code, we try to annonuce new RTF_ANNOUNCE entry here.
   1092 		 * We don't do that here since llinfo is not ready yet.
   1093 		 *
   1094 		 * There are also couple of other things to be discussed:
   1095 		 * - unsolicited NA code needs improvement beforehand
   1096 		 * - RFC2461 says we MAY send multicast unsolicited NA
   1097 		 *   (7.2.6 paragraph 4), however, it also says that we
   1098 		 *   SHOULD provide a mechanism to prevent multicast NA storm.
   1099 		 *   we don't have anything like it right now.
   1100 		 *   note that the mechanism needs a mutual agreement
   1101 		 *   between proxies, which means that we need to implement
   1102 		 *   a new protocol, or a new kludge.
   1103 		 * - from RFC2461 6.2.4, host MUST NOT send an unsolicited NA.
   1104 		 *   we need to check ip6forwarding before sending it.
   1105 		 *   (or should we allow proxy ND configuration only for
   1106 		 *   routers?  there's no mention about proxy ND from hosts)
   1107 		 */
   1108 #if 0
   1109 		/* XXX it does not work */
   1110 		if (rt->rt_flags & RTF_ANNOUNCE)
   1111 			nd6_na_output(ifp,
   1112 			      &SIN6(rt_key(rt))->sin6_addr,
   1113 			      &SIN6(rt_key(rt))->sin6_addr,
   1114 			      ip6_forwarding ? ND_NA_FLAG_ROUTER : 0,
   1115 			      1, NULL);
   1116 #endif
   1117 		/* FALLTHROUGH */
   1118 	case RTM_RESOLVE:
   1119 		if ((ifp->if_flags & IFF_POINTOPOINT) == 0) {
   1120 			/*
   1121 			 * Address resolution isn't necessary for a point to
   1122 			 * point link, so we can skip this test for a p2p link.
   1123 			 */
   1124 			if (gate->sa_family != AF_LINK ||
   1125 			    gate->sa_len < sizeof(null_sdl)) {
   1126 				log(LOG_DEBUG,
   1127 				    "nd6_rtrequest: bad gateway value: %s\n",
   1128 				    if_name(ifp));
   1129 				break;
   1130 			}
   1131 			SDL(gate)->sdl_type = ifp->if_type;
   1132 			SDL(gate)->sdl_index = ifp->if_index;
   1133 		}
   1134 		if (ln != NULL)
   1135 			break;	/* This happens on a route change */
   1136 		/*
   1137 		 * Case 2: This route may come from cloning, or a manual route
   1138 		 * add with a LL address.
   1139 		 */
   1140 		R_Malloc(ln, struct llinfo_nd6 *, sizeof(*ln));
   1141 		rt->rt_llinfo = (caddr_t)ln;
   1142 		if (!ln) {
   1143 			log(LOG_DEBUG, "nd6_rtrequest: malloc failed\n");
   1144 			break;
   1145 		}
   1146 		nd6_inuse++;
   1147 		nd6_allocated++;
   1148 		Bzero(ln, sizeof(*ln));
   1149 		ln->ln_rt = rt;
   1150 		/* this is required for "ndp" command. - shin */
   1151 		if (req == RTM_ADD) {
   1152 		        /*
   1153 			 * gate should have some valid AF_LINK entry,
   1154 			 * and ln->ln_expire should have some lifetime
   1155 			 * which is specified by ndp command.
   1156 			 */
   1157 			ln->ln_state = ND6_LLINFO_REACHABLE;
   1158 			ln->ln_byhint = 0;
   1159 		} else {
   1160 		        /*
   1161 			 * When req == RTM_RESOLVE, rt is created and
   1162 			 * initialized in rtrequest(), so rt_expire is 0.
   1163 			 */
   1164 			ln->ln_state = ND6_LLINFO_NOSTATE;
   1165 			ln->ln_expire = time_second;
   1166 		}
   1167 		rt->rt_flags |= RTF_LLINFO;
   1168 		ln->ln_next = llinfo_nd6.ln_next;
   1169 		llinfo_nd6.ln_next = ln;
   1170 		ln->ln_prev = &llinfo_nd6;
   1171 		ln->ln_next->ln_prev = ln;
   1172 
   1173 		/*
   1174 		 * check if rt_key(rt) is one of my address assigned
   1175 		 * to the interface.
   1176 		 */
   1177 		ifa = (struct ifaddr *)in6ifa_ifpwithaddr(rt->rt_ifp,
   1178 					  &SIN6(rt_key(rt))->sin6_addr);
   1179 		if (ifa) {
   1180 			caddr_t macp = nd6_ifptomac(ifp);
   1181 			ln->ln_expire = 0;
   1182 			ln->ln_state = ND6_LLINFO_REACHABLE;
   1183 			ln->ln_byhint = 0;
   1184 			if (macp) {
   1185 				Bcopy(macp, LLADDR(SDL(gate)), ifp->if_addrlen);
   1186 				SDL(gate)->sdl_alen = ifp->if_addrlen;
   1187 			}
   1188 			if (nd6_useloopback) {
   1189 				rt->rt_ifp = &loif[0];	/*XXX*/
   1190 				/*
   1191 				 * Make sure rt_ifa be equal to the ifaddr
   1192 				 * corresponding to the address.
   1193 				 * We need this because when we refer
   1194 				 * rt_ifa->ia6_flags in ip6_input, we assume
   1195 				 * that the rt_ifa points to the address instead
   1196 				 * of the loopback address.
   1197 				 */
   1198 				if (ifa != rt->rt_ifa) {
   1199 					IFAFREE(rt->rt_ifa);
   1200 					IFAREF(ifa);
   1201 					rt->rt_ifa = ifa;
   1202 				}
   1203 			}
   1204 		} else if (rt->rt_flags & RTF_ANNOUNCE) {
   1205 			ln->ln_expire = 0;
   1206 			ln->ln_state = ND6_LLINFO_REACHABLE;
   1207 			ln->ln_byhint = 0;
   1208 
   1209 			/* join solicited node multicast for proxy ND */
   1210 			if (ifp->if_flags & IFF_MULTICAST) {
   1211 				struct in6_addr llsol;
   1212 				int error;
   1213 
   1214 				llsol = SIN6(rt_key(rt))->sin6_addr;
   1215 				llsol.s6_addr16[0] = htons(0xff02);
   1216 				llsol.s6_addr16[1] = htons(ifp->if_index);
   1217 				llsol.s6_addr32[1] = 0;
   1218 				llsol.s6_addr32[2] = htonl(1);
   1219 				llsol.s6_addr8[12] = 0xff;
   1220 
   1221 				if (!in6_addmulti(&llsol, ifp, &error)) {
   1222 					nd6log((LOG_ERR, "%s: "
   1223 					    "failed to join %s (errno=%d)\n",
   1224 					    if_name(ifp), ip6_sprintf(&llsol),
   1225 					    error));
   1226 				}
   1227 			}
   1228 		}
   1229 		break;
   1230 
   1231 	case RTM_DELETE:
   1232 		if (!ln)
   1233 			break;
   1234 		/* leave from solicited node multicast for proxy ND */
   1235 		if ((rt->rt_flags & RTF_ANNOUNCE) != 0 &&
   1236 		    (ifp->if_flags & IFF_MULTICAST) != 0) {
   1237 			struct in6_addr llsol;
   1238 			struct in6_multi *in6m;
   1239 
   1240 			llsol = SIN6(rt_key(rt))->sin6_addr;
   1241 			llsol.s6_addr16[0] = htons(0xff02);
   1242 			llsol.s6_addr16[1] = htons(ifp->if_index);
   1243 			llsol.s6_addr32[1] = 0;
   1244 			llsol.s6_addr32[2] = htonl(1);
   1245 			llsol.s6_addr8[12] = 0xff;
   1246 
   1247 			IN6_LOOKUP_MULTI(llsol, ifp, in6m);
   1248 			if (in6m)
   1249 				in6_delmulti(in6m);
   1250 		}
   1251 		nd6_inuse--;
   1252 		ln->ln_next->ln_prev = ln->ln_prev;
   1253 		ln->ln_prev->ln_next = ln->ln_next;
   1254 		ln->ln_prev = NULL;
   1255 		rt->rt_llinfo = 0;
   1256 		rt->rt_flags &= ~RTF_LLINFO;
   1257 		if (ln->ln_hold)
   1258 			m_freem(ln->ln_hold);
   1259 		Free((caddr_t)ln);
   1260 	}
   1261 }
   1262 
   1263 void
   1264 nd6_p2p_rtrequest(req, rt, info)
   1265 	int	req;
   1266 	struct rtentry *rt;
   1267 	struct rt_addrinfo *info; /* xxx unused */
   1268 {
   1269 	struct sockaddr *gate = rt->rt_gateway;
   1270 	static struct sockaddr_dl null_sdl = {sizeof(null_sdl), AF_LINK};
   1271 	struct ifnet *ifp = rt->rt_ifp;
   1272 	struct ifaddr *ifa;
   1273 
   1274 	if (rt->rt_flags & RTF_GATEWAY)
   1275 		return;
   1276 
   1277 	switch (req) {
   1278 	case RTM_ADD:
   1279 		/*
   1280 		 * There is no backward compatibility :)
   1281 		 *
   1282 		 * if ((rt->rt_flags & RTF_HOST) == 0 &&
   1283 		 *     SIN(rt_mask(rt))->sin_addr.s_addr != 0xffffffff)
   1284 		 *	   rt->rt_flags |= RTF_CLONING;
   1285 		 */
   1286 		if (rt->rt_flags & RTF_CLONING) {
   1287 			/*
   1288 			 * Case 1: This route should come from
   1289 			 * a route to interface.
   1290 			 */
   1291 			rt_setgate(rt, rt_key(rt),
   1292 				   (struct sockaddr *)&null_sdl);
   1293 			gate = rt->rt_gateway;
   1294 			SDL(gate)->sdl_type = ifp->if_type;
   1295 			SDL(gate)->sdl_index = ifp->if_index;
   1296 			break;
   1297 		}
   1298 		/* Announce a new entry if requested. */
   1299 		if (rt->rt_flags & RTF_ANNOUNCE)
   1300 			nd6_na_output(ifp,
   1301 				      &SIN6(rt_key(rt))->sin6_addr,
   1302 				      &SIN6(rt_key(rt))->sin6_addr,
   1303 				      ip6_forwarding ? ND_NA_FLAG_ROUTER : 0,
   1304 				      1, NULL);
   1305 		/* FALLTHROUGH */
   1306 	case RTM_RESOLVE:
   1307 		/*
   1308 		 * check if rt_key(rt) is one of my address assigned
   1309 		 * to the interface.
   1310 		 */
   1311  		ifa = (struct ifaddr *)in6ifa_ifpwithaddr(rt->rt_ifp,
   1312 					  &SIN6(rt_key(rt))->sin6_addr);
   1313 		if (ifa) {
   1314 			if (nd6_useloopback) {
   1315 				rt->rt_ifp = &loif[0];	/*XXX*/
   1316 			}
   1317 		}
   1318 		break;
   1319 	}
   1320 }
   1321 
   1322 int
   1323 nd6_ioctl(cmd, data, ifp)
   1324 	u_long cmd;
   1325 	caddr_t	data;
   1326 	struct ifnet *ifp;
   1327 {
   1328 	struct in6_drlist *drl = (struct in6_drlist *)data;
   1329 	struct in6_prlist *prl = (struct in6_prlist *)data;
   1330 	struct in6_ndireq *ndi = (struct in6_ndireq *)data;
   1331 	struct in6_nbrinfo *nbi = (struct in6_nbrinfo *)data;
   1332 	struct in6_ndifreq *ndif = (struct in6_ndifreq *)data;
   1333 	struct nd_defrouter *dr, any;
   1334 	struct nd_prefix *pr;
   1335 	struct rtentry *rt;
   1336 	int i = 0, error = 0;
   1337 	int s;
   1338 
   1339 	switch (cmd) {
   1340 	case SIOCGDRLST_IN6:
   1341 		bzero(drl, sizeof(*drl));
   1342 		s = splsoftnet();
   1343 		dr = TAILQ_FIRST(&nd_defrouter);
   1344 		while (dr && i < DRLSTSIZ) {
   1345 			drl->defrouter[i].rtaddr = dr->rtaddr;
   1346 			if (IN6_IS_ADDR_LINKLOCAL(&drl->defrouter[i].rtaddr)) {
   1347 				/* XXX: need to this hack for KAME stack */
   1348 				drl->defrouter[i].rtaddr.s6_addr16[1] = 0;
   1349 			} else
   1350 				log(LOG_ERR,
   1351 				    "default router list contains a "
   1352 				    "non-linklocal address(%s)\n",
   1353 				    ip6_sprintf(&drl->defrouter[i].rtaddr));
   1354 
   1355 			drl->defrouter[i].flags = dr->flags;
   1356 			drl->defrouter[i].rtlifetime = dr->rtlifetime;
   1357 			drl->defrouter[i].expire = dr->expire;
   1358 			drl->defrouter[i].if_index = dr->ifp->if_index;
   1359 			i++;
   1360 			dr = TAILQ_NEXT(dr, dr_entry);
   1361 		}
   1362 		splx(s);
   1363 		break;
   1364 	case SIOCGPRLST_IN6:
   1365 		/*
   1366 		 * XXX meaning of fields, especialy "raflags", is very
   1367 		 * differnet between RA prefix list and RR/static prefix list.
   1368 		 * how about separating ioctls into two?
   1369 		 */
   1370 		bzero(prl, sizeof(*prl));
   1371 		s = splsoftnet();
   1372 		pr = nd_prefix.lh_first;
   1373 		while (pr && i < PRLSTSIZ) {
   1374 			struct nd_pfxrouter *pfr;
   1375 			int j;
   1376 
   1377 			prl->prefix[i].prefix = pr->ndpr_prefix.sin6_addr;
   1378 			prl->prefix[i].raflags = pr->ndpr_raf;
   1379 			prl->prefix[i].prefixlen = pr->ndpr_plen;
   1380 			prl->prefix[i].vltime = pr->ndpr_vltime;
   1381 			prl->prefix[i].pltime = pr->ndpr_pltime;
   1382 			prl->prefix[i].if_index = pr->ndpr_ifp->if_index;
   1383 			prl->prefix[i].expire = pr->ndpr_expire;
   1384 
   1385 			pfr = pr->ndpr_advrtrs.lh_first;
   1386 			j = 0;
   1387 			while (pfr) {
   1388 				if (j < DRLSTSIZ) {
   1389 #define RTRADDR prl->prefix[i].advrtr[j]
   1390 					RTRADDR = pfr->router->rtaddr;
   1391 					if (IN6_IS_ADDR_LINKLOCAL(&RTRADDR)) {
   1392 						/* XXX: hack for KAME */
   1393 						RTRADDR.s6_addr16[1] = 0;
   1394 					} else
   1395 						log(LOG_ERR,
   1396 						    "a router(%s) advertises "
   1397 						    "a prefix with "
   1398 						    "non-link local address\n",
   1399 						    ip6_sprintf(&RTRADDR));
   1400 #undef RTRADDR
   1401 				}
   1402 				j++;
   1403 				pfr = pfr->pfr_next;
   1404 			}
   1405 			prl->prefix[i].advrtrs = j;
   1406 			prl->prefix[i].origin = PR_ORIG_RA;
   1407 
   1408 			i++;
   1409 			pr = pr->ndpr_next;
   1410 		}
   1411 	      {
   1412 		struct rr_prefix *rpp;
   1413 
   1414 		for (rpp = LIST_FIRST(&rr_prefix); rpp;
   1415 		     rpp = LIST_NEXT(rpp, rp_entry)) {
   1416 			if (i >= PRLSTSIZ)
   1417 				break;
   1418 			prl->prefix[i].prefix = rpp->rp_prefix.sin6_addr;
   1419 			prl->prefix[i].raflags = rpp->rp_raf;
   1420 			prl->prefix[i].prefixlen = rpp->rp_plen;
   1421 			prl->prefix[i].vltime = rpp->rp_vltime;
   1422 			prl->prefix[i].pltime = rpp->rp_pltime;
   1423 			prl->prefix[i].if_index = rpp->rp_ifp->if_index;
   1424 			prl->prefix[i].expire = rpp->rp_expire;
   1425 			prl->prefix[i].advrtrs = 0;
   1426 			prl->prefix[i].origin = rpp->rp_origin;
   1427 			i++;
   1428 		}
   1429 	      }
   1430 		splx(s);
   1431 
   1432 		break;
   1433 	case SIOCGIFINFO_IN6:
   1434 		if (!nd_ifinfo || i >= nd_ifinfo_indexlim) {
   1435 			error = EINVAL;
   1436 			break;
   1437 		}
   1438 		ndi->ndi = nd_ifinfo[ifp->if_index];
   1439 		break;
   1440 	case SIOCSIFINFO_FLAGS:
   1441 		/* XXX: almost all other fields of ndi->ndi is unused */
   1442 		if (!nd_ifinfo || i >= nd_ifinfo_indexlim) {
   1443 			error = EINVAL;
   1444 			break;
   1445 		}
   1446 		nd_ifinfo[ifp->if_index].flags = ndi->ndi.flags;
   1447 		break;
   1448 	case SIOCSNDFLUSH_IN6:	/* XXX: the ioctl name is confusing... */
   1449 		/* flush default router list */
   1450 		/*
   1451 		 * xxx sumikawa: should not delete route if default
   1452 		 * route equals to the top of default router list
   1453 		 */
   1454 		bzero(&any, sizeof(any));
   1455 		defrouter_delreq(&any, 0);
   1456 		defrouter_select();
   1457 		/* xxx sumikawa: flush prefix list */
   1458 		break;
   1459 	case SIOCSPFXFLUSH_IN6:
   1460 	    {
   1461 		/* flush all the prefix advertised by routers */
   1462 		struct nd_prefix *pr, *next;
   1463 
   1464 		s = splsoftnet();
   1465 		for (pr = nd_prefix.lh_first; pr; pr = next) {
   1466 			next = pr->ndpr_next;
   1467 			if (!IN6_IS_ADDR_UNSPECIFIED(&pr->ndpr_addr))
   1468 				in6_ifdel(pr->ndpr_ifp, &pr->ndpr_addr);
   1469 			prelist_remove(pr);
   1470 		}
   1471 		splx(s);
   1472 		break;
   1473 	    }
   1474 	case SIOCSRTRFLUSH_IN6:
   1475 	    {
   1476 		/* flush all the default routers */
   1477 		struct nd_defrouter *dr, *next;
   1478 
   1479 		s = splsoftnet();
   1480 		if ((dr = TAILQ_FIRST(&nd_defrouter)) != NULL) {
   1481 			/*
   1482 			 * The first entry of the list may be stored in
   1483 			 * the routing table, so we'll delete it later.
   1484 			 */
   1485 			for (dr = TAILQ_NEXT(dr, dr_entry); dr; dr = next) {
   1486 				next = TAILQ_NEXT(dr, dr_entry);
   1487 				defrtrlist_del(dr);
   1488 			}
   1489 			defrtrlist_del(TAILQ_FIRST(&nd_defrouter));
   1490 		}
   1491 		splx(s);
   1492 		break;
   1493 	    }
   1494 	case SIOCGNBRINFO_IN6:
   1495 	    {
   1496 		struct llinfo_nd6 *ln;
   1497 		struct in6_addr nb_addr = nbi->addr; /* make local for safety */
   1498 
   1499 		/*
   1500 		 * XXX: KAME specific hack for scoped addresses
   1501 		 *      XXXX: for other scopes than link-local?
   1502 		 */
   1503 		if (IN6_IS_ADDR_LINKLOCAL(&nbi->addr) ||
   1504 		    IN6_IS_ADDR_MC_LINKLOCAL(&nbi->addr)) {
   1505 			u_int16_t *idp = (u_int16_t *)&nb_addr.s6_addr[2];
   1506 
   1507 			if (*idp == 0)
   1508 				*idp = htons(ifp->if_index);
   1509 		}
   1510 
   1511 		s = splsoftnet();
   1512 		if ((rt = nd6_lookup(&nb_addr, 0, ifp)) == NULL) {
   1513 			error = EINVAL;
   1514 			splx(s);
   1515 			break;
   1516 		}
   1517 		ln = (struct llinfo_nd6 *)rt->rt_llinfo;
   1518 		nbi->state = ln->ln_state;
   1519 		nbi->asked = ln->ln_asked;
   1520 		nbi->isrouter = ln->ln_router;
   1521 		nbi->expire = ln->ln_expire;
   1522 		splx(s);
   1523 
   1524 		break;
   1525 	    }
   1526 	case SIOCGDEFIFACE_IN6:	/* XXX: should be implemented as a sysctl? */
   1527 		ndif->ifindex = nd6_defifindex;
   1528 		break;
   1529 	case SIOCSDEFIFACE_IN6:	/* XXX: should be implemented as a sysctl? */
   1530 		return(nd6_setdefaultiface(ndif->ifindex));
   1531 		break;
   1532 	}
   1533 	return(error);
   1534 }
   1535 
   1536 /*
   1537  * Create neighbor cache entry and cache link-layer address,
   1538  * on reception of inbound ND6 packets. (RS/RA/NS/redirect)
   1539  */
   1540 struct rtentry *
   1541 nd6_cache_lladdr(ifp, from, lladdr, lladdrlen, type, code)
   1542 	struct ifnet *ifp;
   1543 	struct in6_addr *from;
   1544 	char *lladdr;
   1545 	int lladdrlen;
   1546 	int type;	/* ICMP6 type */
   1547 	int code;	/* type dependent information */
   1548 {
   1549 	struct rtentry *rt = NULL;
   1550 	struct llinfo_nd6 *ln = NULL;
   1551 	int is_newentry;
   1552 	struct sockaddr_dl *sdl = NULL;
   1553 	int do_update;
   1554 	int olladdr;
   1555 	int llchange;
   1556 	int newstate = 0;
   1557 	long time_second = time.tv_sec;
   1558 
   1559 	if (!ifp)
   1560 		panic("ifp == NULL in nd6_cache_lladdr");
   1561 	if (!from)
   1562 		panic("from == NULL in nd6_cache_lladdr");
   1563 
   1564 	/* nothing must be updated for unspecified address */
   1565 	if (IN6_IS_ADDR_UNSPECIFIED(from))
   1566 		return NULL;
   1567 
   1568 	/*
   1569 	 * Validation about ifp->if_addrlen and lladdrlen must be done in
   1570 	 * the caller.
   1571 	 *
   1572 	 * XXX If the link does not have link-layer adderss, what should
   1573 	 * we do? (ifp->if_addrlen == 0)
   1574 	 * Spec says nothing in sections for RA, RS and NA.  There's small
   1575 	 * description on it in NS section (RFC 2461 7.2.3).
   1576 	 */
   1577 
   1578 	rt = nd6_lookup(from, 0, ifp);
   1579 	if (!rt) {
   1580 #if 0
   1581 		/* nothing must be done if there's no lladdr */
   1582 		if (!lladdr || !lladdrlen)
   1583 			return NULL;
   1584 #endif
   1585 
   1586 		rt = nd6_lookup(from, 1, ifp);
   1587 		is_newentry = 1;
   1588 	} else {
   1589 		/* do nothing if static ndp is set */
   1590 		if (rt->rt_flags & RTF_STATIC)
   1591 			return NULL;
   1592 		is_newentry = 0;
   1593 	}
   1594 
   1595 	if (!rt)
   1596 		return NULL;
   1597 	if ((rt->rt_flags & (RTF_GATEWAY | RTF_LLINFO)) != RTF_LLINFO) {
   1598 fail:
   1599 		(void)nd6_free(rt);
   1600 		return NULL;
   1601 	}
   1602 	ln = (struct llinfo_nd6 *)rt->rt_llinfo;
   1603 	if (!ln)
   1604 		goto fail;
   1605 	if (!rt->rt_gateway)
   1606 		goto fail;
   1607 	if (rt->rt_gateway->sa_family != AF_LINK)
   1608 		goto fail;
   1609 	sdl = SDL(rt->rt_gateway);
   1610 
   1611 	olladdr = (sdl->sdl_alen) ? 1 : 0;
   1612 	if (olladdr && lladdr) {
   1613 		if (bcmp(lladdr, LLADDR(sdl), ifp->if_addrlen))
   1614 			llchange = 1;
   1615 		else
   1616 			llchange = 0;
   1617 	} else
   1618 		llchange = 0;
   1619 
   1620 	/*
   1621 	 * newentry olladdr  lladdr  llchange	(*=record)
   1622 	 *	0	n	n	--	(1)
   1623 	 *	0	y	n	--	(2)
   1624 	 *	0	n	y	--	(3) * STALE
   1625 	 *	0	y	y	n	(4) *
   1626 	 *	0	y	y	y	(5) * STALE
   1627 	 *	1	--	n	--	(6)   NOSTATE(= PASSIVE)
   1628 	 *	1	--	y	--	(7) * STALE
   1629 	 */
   1630 
   1631 	if (lladdr) {		/*(3-5) and (7)*/
   1632 		/*
   1633 		 * Record source link-layer address
   1634 		 * XXX is it dependent to ifp->if_type?
   1635 		 */
   1636 		sdl->sdl_alen = ifp->if_addrlen;
   1637 		bcopy(lladdr, LLADDR(sdl), ifp->if_addrlen);
   1638 	}
   1639 
   1640 	if (!is_newentry) {
   1641 		if ((!olladdr && lladdr)		/*(3)*/
   1642 		 || (olladdr && lladdr && llchange)) {	/*(5)*/
   1643 			do_update = 1;
   1644 			newstate = ND6_LLINFO_STALE;
   1645 		} else					/*(1-2,4)*/
   1646 			do_update = 0;
   1647 	} else {
   1648 		do_update = 1;
   1649 		if (!lladdr)				/*(6)*/
   1650 			newstate = ND6_LLINFO_NOSTATE;
   1651 		else					/*(7)*/
   1652 			newstate = ND6_LLINFO_STALE;
   1653 	}
   1654 
   1655 	if (do_update) {
   1656 		/*
   1657 		 * Update the state of the neighbor cache.
   1658 		 */
   1659 		ln->ln_state = newstate;
   1660 
   1661 		if (ln->ln_state == ND6_LLINFO_STALE) {
   1662 			/*
   1663 			 * XXX: since nd6_output() below will cause
   1664 			 * state tansition to DELAY and reset the timer,
   1665 			 * we must set the timer now, although it is actually
   1666 			 * meaningless.
   1667 			 */
   1668 			ln->ln_expire = time_second + nd6_gctimer;
   1669 
   1670 			if (ln->ln_hold) {
   1671 #ifdef OLDIP6OUTPUT
   1672 				ln->ln_asked = 0;
   1673 				ln->ln_state = ND6_LLINFO_DELAY;
   1674 				ln->ln_expire = time_second + nd6_delay;
   1675 				(*ifp->if_output)(ifp, ln->ln_hold,
   1676 						  rt_key(rt), rt);
   1677 #else
   1678 				/*
   1679 				 * we assume ifp is not a p2p here, so just
   1680 				 * set the 2nd argument as the 1st one.
   1681 				 */
   1682 				nd6_output(ifp, ifp, ln->ln_hold,
   1683 					   (struct sockaddr_in6 *)rt_key(rt),
   1684 					   rt);
   1685 #endif
   1686 				ln->ln_hold = NULL;
   1687 			}
   1688 		} else if (ln->ln_state == ND6_LLINFO_INCOMPLETE) {
   1689 			/* probe right away */
   1690 			ln->ln_expire = time_second;
   1691 		}
   1692 	}
   1693 
   1694 	/*
   1695 	 * ICMP6 type dependent behavior.
   1696 	 *
   1697 	 * NS: clear IsRouter if new entry
   1698 	 * RS: clear IsRouter
   1699 	 * RA: set IsRouter if there's lladdr
   1700 	 * redir: clear IsRouter if new entry
   1701 	 *
   1702 	 * RA case, (1):
   1703 	 * The spec says that we must set IsRouter in the following cases:
   1704 	 * - If lladdr exist, set IsRouter.  This means (1-5).
   1705 	 * - If it is old entry (!newentry), set IsRouter.  This means (7).
   1706 	 * So, based on the spec, in (1-5) and (7) cases we must set IsRouter.
   1707 	 * A quetion arises for (1) case.  (1) case has no lladdr in the
   1708 	 * neighbor cache, this is similar to (6).
   1709 	 * This case is rare but we figured that we MUST NOT set IsRouter.
   1710 	 *
   1711 	 * newentry olladdr  lladdr  llchange	    NS  RS  RA	redir
   1712 	 *							D R
   1713 	 *	0	n	n	--	(1)	c   ?     s
   1714 	 *	0	y	n	--	(2)	c   s     s
   1715 	 *	0	n	y	--	(3)	c   s     s
   1716 	 *	0	y	y	n	(4)	c   s     s
   1717 	 *	0	y	y	y	(5)	c   s     s
   1718 	 *	1	--	n	--	(6) c	c 	c s
   1719 	 *	1	--	y	--	(7) c	c   s	c s
   1720 	 *
   1721 	 *					(c=clear s=set)
   1722 	 */
   1723 	switch (type & 0xff) {
   1724 	case ND_NEIGHBOR_SOLICIT:
   1725 		/*
   1726 		 * New entry must have is_router flag cleared.
   1727 		 */
   1728 		if (is_newentry)	/*(6-7)*/
   1729 			ln->ln_router = 0;
   1730 		break;
   1731 	case ND_REDIRECT:
   1732 		/*
   1733 		 * If the icmp is a redirect to a better router, always set the
   1734 		 * is_router flag. Otherwise, if the entry is newly created,
   1735 		 * clear the flag. [RFC 2461, sec 8.3]
   1736 		 */
   1737 		if (code == ND_REDIRECT_ROUTER)
   1738 			ln->ln_router = 1;
   1739 		else if (is_newentry) /*(6-7)*/
   1740 			ln->ln_router = 0;
   1741 		break;
   1742 	case ND_ROUTER_SOLICIT:
   1743 		/*
   1744 		 * is_router flag must always be cleared.
   1745 		 */
   1746 		ln->ln_router = 0;
   1747 		break;
   1748 	case ND_ROUTER_ADVERT:
   1749 		/*
   1750 		 * Mark an entry with lladdr as a router.
   1751 		 */
   1752 		if ((!is_newentry && (olladdr || lladdr))	/*(2-5)*/
   1753 		 || (is_newentry && lladdr)) {			/*(7)*/
   1754 			ln->ln_router = 1;
   1755 		}
   1756 		break;
   1757 	}
   1758 
   1759 	/*
   1760 	 * When the link-layer address of a router changes, select the
   1761 	 * best router again.  In particular, when the neighbor entry is newly
   1762 	 * created, it might affect the selection policy.
   1763 	 * Question: can we restrict the first condition to the "is_newentry"
   1764 	 * case?
   1765 	 * XXX: when we hear an RA from a new router with the link-layer
   1766 	 * address option, defrouter_select() is called twice, since
   1767 	 * defrtrlist_update called the function as well.  However, I believe
   1768 	 * we can compromise the overhead, since it only happens the first
   1769 	 * time.
   1770 	 */
   1771 	if (do_update && ln->ln_router)
   1772 		defrouter_select();
   1773 
   1774 	return rt;
   1775 }
   1776 
   1777 static void
   1778 nd6_slowtimo(ignored_arg)
   1779     void *ignored_arg;
   1780 {
   1781 	int s = splsoftnet();
   1782 	int i;
   1783 	struct nd_ifinfo *nd6if;
   1784 
   1785 	callout_reset(&nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
   1786 	    nd6_slowtimo, NULL);
   1787 	for (i = 1; i < if_index + 1; i++) {
   1788 		if (!nd_ifinfo || i >= nd_ifinfo_indexlim)
   1789 			continue;
   1790 		nd6if = &nd_ifinfo[i];
   1791 		if (nd6if->basereachable && /* already initialized */
   1792 		    (nd6if->recalctm -= ND6_SLOWTIMER_INTERVAL) <= 0) {
   1793 			/*
   1794 			 * Since reachable time rarely changes by router
   1795 			 * advertisements, we SHOULD insure that a new random
   1796 			 * value gets recomputed at least once every few hours.
   1797 			 * (RFC 2461, 6.3.4)
   1798 			 */
   1799 			nd6if->recalctm = nd6_recalc_reachtm_interval;
   1800 			nd6if->reachable = ND_COMPUTE_RTIME(nd6if->basereachable);
   1801 		}
   1802 	}
   1803 	splx(s);
   1804 }
   1805 
   1806 #define senderr(e) { error = (e); goto bad;}
   1807 int
   1808 nd6_output(ifp, origifp, m0, dst, rt0)
   1809 	struct ifnet *ifp;
   1810 	struct ifnet *origifp;
   1811 	struct mbuf *m0;
   1812 	struct sockaddr_in6 *dst;
   1813 	struct rtentry *rt0;
   1814 {
   1815 	struct mbuf *m = m0;
   1816 	struct rtentry *rt = rt0;
   1817 	struct sockaddr_in6 *gw6 = NULL;
   1818 	struct llinfo_nd6 *ln = NULL;
   1819 	int error = 0;
   1820 	long time_second = time.tv_sec;
   1821 
   1822 	if (IN6_IS_ADDR_MULTICAST(&dst->sin6_addr))
   1823 		goto sendpkt;
   1824 
   1825 	/*
   1826 	 * XXX: we currently do not make neighbor cache on any interface
   1827 	 * other than ARCnet, Ethernet, FDDI and GIF.
   1828 	 *
   1829 	 * RFC2893 says:
   1830 	 * - unidirectional tunnels needs no ND
   1831 	 */
   1832 	switch (ifp->if_type) {
   1833 	case IFT_ARCNET:
   1834 	case IFT_ETHER:
   1835 	case IFT_FDDI:
   1836 	case IFT_IEEE1394:
   1837 	case IFT_GIF:		/* XXX need more cases? */
   1838 		break;
   1839 	default:
   1840 		goto sendpkt;
   1841 	}
   1842 
   1843 	/*
   1844 	 * next hop determination. This routine is derived from ether_outpout.
   1845 	 */
   1846 	if (rt) {
   1847 		if ((rt->rt_flags & RTF_UP) == 0) {
   1848 			if ((rt0 = rt = rtalloc1((struct sockaddr *)dst, 1)) !=
   1849 				NULL)
   1850 			{
   1851 				rt->rt_refcnt--;
   1852 				if (rt->rt_ifp != ifp) {
   1853 					/* XXX: loop care? */
   1854 					return nd6_output(ifp, origifp, m0,
   1855 							  dst, rt);
   1856 				}
   1857 			} else
   1858 				senderr(EHOSTUNREACH);
   1859 		}
   1860 
   1861 		if (rt->rt_flags & RTF_GATEWAY) {
   1862 			gw6 = (struct sockaddr_in6 *)rt->rt_gateway;
   1863 
   1864 			/*
   1865 			 * We skip link-layer address resolution and NUD
   1866 			 * if the gateway is not a neighbor from ND point
   1867 			 * of view, regardless the value of the
   1868 			 * nd_ifinfo.flags.
   1869 			 * The second condition is a bit tricky: we skip
   1870 			 * if the gateway is our own address, which is
   1871 			 * sometimes used to install a route to a p2p link.
   1872 			 */
   1873 			if (!nd6_is_addr_neighbor(gw6, ifp) ||
   1874 			    in6ifa_ifpwithaddr(ifp, &gw6->sin6_addr)) {
   1875 				/*
   1876 				 * We allow this kind of tricky route only
   1877 				 * when the outgoing interface is p2p.
   1878 				 * XXX: we may need a more generic rule here.
   1879 				 */
   1880 				if ((ifp->if_flags & IFF_POINTOPOINT) == 0)
   1881 					senderr(EHOSTUNREACH);
   1882 
   1883 				goto sendpkt;
   1884 			}
   1885 
   1886 			if (rt->rt_gwroute == 0)
   1887 				goto lookup;
   1888 			if (((rt = rt->rt_gwroute)->rt_flags & RTF_UP) == 0) {
   1889 				rtfree(rt); rt = rt0;
   1890 			lookup: rt->rt_gwroute = rtalloc1(rt->rt_gateway, 1);
   1891 				if ((rt = rt->rt_gwroute) == 0)
   1892 					senderr(EHOSTUNREACH);
   1893 			}
   1894 		}
   1895 	}
   1896 
   1897 	/*
   1898 	 * Address resolution or Neighbor Unreachability Detection
   1899 	 * for the next hop.
   1900 	 * At this point, the destination of the packet must be a unicast
   1901 	 * or an anycast address(i.e. not a multicast).
   1902 	 */
   1903 
   1904 	/* Look up the neighbor cache for the nexthop */
   1905 	if (rt && (rt->rt_flags & RTF_LLINFO) != 0)
   1906 		ln = (struct llinfo_nd6 *)rt->rt_llinfo;
   1907 	else {
   1908 		/*
   1909 		 * Since nd6_is_addr_neighbor() internally calls nd6_lookup(),
   1910 		 * the condition below is not very efficient. But we believe
   1911 		 * it is tolerable, because this should be a rare case.
   1912 		 */
   1913 		if (nd6_is_addr_neighbor(dst, ifp) &&
   1914 		    (rt = nd6_lookup(&dst->sin6_addr, 1, ifp)) != NULL)
   1915 			ln = (struct llinfo_nd6 *)rt->rt_llinfo;
   1916 	}
   1917 	if (!ln || !rt) {
   1918 		if ((ifp->if_flags & IFF_POINTOPOINT) == 0 &&
   1919 		    !(nd_ifinfo[ifp->if_index].flags & ND6_IFF_PERFORMNUD)) {
   1920 			log(LOG_DEBUG,
   1921 			    "nd6_output: can't allocate llinfo for %s "
   1922 			    "(ln=%p, rt=%p)\n",
   1923 			    ip6_sprintf(&dst->sin6_addr), ln, rt);
   1924 			senderr(EIO);	/* XXX: good error? */
   1925 		}
   1926 
   1927 		goto sendpkt;	/* send anyway */
   1928 	}
   1929 
   1930 	/* We don't have to do link-layer address resolution on a p2p link. */
   1931 	if ((ifp->if_flags & IFF_POINTOPOINT) != 0 &&
   1932 	    ln->ln_state < ND6_LLINFO_REACHABLE) {
   1933 		ln->ln_state = ND6_LLINFO_STALE;
   1934 		ln->ln_expire = time_second + nd6_gctimer;
   1935 	}
   1936 
   1937 	/*
   1938 	 * The first time we send a packet to a neighbor whose entry is
   1939 	 * STALE, we have to change the state to DELAY and a sets a timer to
   1940 	 * expire in DELAY_FIRST_PROBE_TIME seconds to ensure do
   1941 	 * neighbor unreachability detection on expiration.
   1942 	 * (RFC 2461 7.3.3)
   1943 	 */
   1944 	if (ln->ln_state == ND6_LLINFO_STALE) {
   1945 		ln->ln_asked = 0;
   1946 		ln->ln_state = ND6_LLINFO_DELAY;
   1947 		ln->ln_expire = time_second + nd6_delay;
   1948 	}
   1949 
   1950 	/*
   1951 	 * If the neighbor cache entry has a state other than INCOMPLETE
   1952 	 * (i.e. its link-layer address is already reloved), just
   1953 	 * send the packet.
   1954 	 */
   1955 	if (ln->ln_state > ND6_LLINFO_INCOMPLETE)
   1956 		goto sendpkt;
   1957 
   1958 	/*
   1959 	 * There is a neighbor cache entry, but no ethernet address
   1960 	 * response yet. Replace the held mbuf (if any) with this
   1961 	 * latest one.
   1962 	 *
   1963 	 * XXX Does the code conform to rate-limiting rule?
   1964 	 * (RFC 2461 7.2.2)
   1965 	 */
   1966 	if (ln->ln_state == ND6_LLINFO_NOSTATE)
   1967 		ln->ln_state = ND6_LLINFO_INCOMPLETE;
   1968 	if (ln->ln_hold)
   1969 		m_freem(ln->ln_hold);
   1970 	ln->ln_hold = m;
   1971 	if (ln->ln_expire) {
   1972 		if (ln->ln_asked < nd6_mmaxtries &&
   1973 		    ln->ln_expire < time_second) {
   1974 			ln->ln_asked++;
   1975 			ln->ln_expire = time_second +
   1976 				nd_ifinfo[ifp->if_index].retrans / 1000;
   1977 			nd6_ns_output(ifp, NULL, &dst->sin6_addr, ln, 0);
   1978 		}
   1979 	}
   1980 	return(0);
   1981 
   1982   sendpkt:
   1983 
   1984 	if ((ifp->if_flags & IFF_LOOPBACK) != 0) {
   1985 		return((*ifp->if_output)(origifp, m, (struct sockaddr *)dst,
   1986 					 rt));
   1987 	}
   1988 	return((*ifp->if_output)(ifp, m, (struct sockaddr *)dst, rt));
   1989 
   1990   bad:
   1991 	if (m)
   1992 		m_freem(m);
   1993 	return (error);
   1994 }
   1995 #undef senderr
   1996 
   1997 int
   1998 nd6_storelladdr(ifp, rt, m, dst, desten)
   1999 	struct ifnet *ifp;
   2000 	struct rtentry *rt;
   2001 	struct mbuf *m;
   2002 	struct sockaddr *dst;
   2003 	u_char *desten;
   2004 {
   2005 	struct sockaddr_dl *sdl;
   2006 
   2007 	if (m->m_flags & M_MCAST) {
   2008 		switch (ifp->if_type) {
   2009 		case IFT_ETHER:
   2010 		case IFT_FDDI:
   2011 			ETHER_MAP_IPV6_MULTICAST(&SIN6(dst)->sin6_addr,
   2012 						 desten);
   2013 			return(1);
   2014 		case IFT_IEEE1394:
   2015 			bcopy(ifp->if_broadcastaddr, desten, ifp->if_addrlen);
   2016 			return(1);
   2017 		case IFT_ARCNET:
   2018 			*desten = 0;
   2019 			return(1);
   2020 		default:
   2021 			m_freem(m);
   2022 			return(0);
   2023 		}
   2024 	}
   2025 
   2026 	if (rt == NULL) {
   2027 		/* this could happen, if we could not allocate memory */
   2028 		m_freem(m);
   2029 		return(0);
   2030 	}
   2031 	if (rt->rt_gateway->sa_family != AF_LINK) {
   2032 		printf("nd6_storelladdr: something odd happens\n");
   2033 		m_freem(m);
   2034 		return(0);
   2035 	}
   2036 	sdl = SDL(rt->rt_gateway);
   2037 	if (sdl->sdl_alen == 0) {
   2038 		/* this should be impossible, but we bark here for debugging */
   2039 		printf("nd6_storelladdr: sdl_alen == 0\n");
   2040 		m_freem(m);
   2041 		return(0);
   2042 	}
   2043 
   2044 	bcopy(LLADDR(sdl), desten, sdl->sdl_alen);
   2045 	return(1);
   2046 }
   2047