Home | History | Annotate | Line # | Download | only in netinet6
nd6.c revision 1.51
      1 /*	$NetBSD: nd6.c,v 1.51 2001/07/25 06:59:52 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 #ifdef DIAGNOSTIC
    187 	if (!ifindex2ifnet[ifp->if_index])
    188 		panic("nd6_ifattach: ifindex2ifnet is NULL");
    189 #endif
    190 	ND.linkmtu = ifindex2ifnet[ifp->if_index]->if_mtu;
    191 	ND.chlim = IPV6_DEFHLIM;
    192 	ND.basereachable = REACHABLE_TIME;
    193 	ND.reachable = ND_COMPUTE_RTIME(ND.basereachable);
    194 	ND.retrans = RETRANS_TIMER;
    195 	ND.receivedra = 0;
    196 	ND.flags = ND6_IFF_PERFORMNUD;
    197 	nd6_setmtu(ifp);
    198 #undef ND
    199 }
    200 
    201 /*
    202  * Reset ND level link MTU. This function is called when the physical MTU
    203  * changes, which means we might have to adjust the ND level MTU.
    204  */
    205 void
    206 nd6_setmtu(ifp)
    207 	struct ifnet *ifp;
    208 {
    209 	struct nd_ifinfo *ndi = &nd_ifinfo[ifp->if_index];
    210 	u_long oldmaxmtu = ndi->maxmtu;
    211 	u_long oldlinkmtu = ndi->linkmtu;
    212 
    213 	switch (ifp->if_type) {
    214 	case IFT_ARCNET:	/* XXX MTU handling needs more work */
    215 		ndi->maxmtu = MIN(60480, ifp->if_mtu);
    216 		break;
    217 	case IFT_ETHER:
    218 		ndi->maxmtu = MIN(ETHERMTU, ifp->if_mtu);
    219 		break;
    220 	case IFT_ATM:
    221 		ndi->maxmtu = MIN(ATMMTU, ifp->if_mtu);
    222 		break;
    223 	case IFT_IEEE1394:
    224 		ndi->maxmtu = MIN(IEEE1394MTU, ifp->if_mtu);
    225 		break;
    226 	default:
    227 		ndi->maxmtu = ifp->if_mtu;
    228 		break;
    229 	}
    230 
    231 	if (oldmaxmtu != ndi->maxmtu) {
    232 		/*
    233 		 * If the ND level MTU is not set yet, or if the maxmtu
    234 		 * is reset to a smaller value than the ND level MTU,
    235 		 * also reset the ND level MTU.
    236 		 */
    237 		if (ndi->linkmtu == 0 ||
    238 		    ndi->maxmtu < ndi->linkmtu) {
    239 			ndi->linkmtu = ndi->maxmtu;
    240 			/* also adjust in6_maxmtu if necessary. */
    241 			if (oldlinkmtu == 0) {
    242 				/*
    243 				 * XXX: the case analysis is grotty, but
    244 				 * it is not efficient to call in6_setmaxmtu()
    245 				 * here when we are during the initialization
    246 				 * procedure.
    247 				 */
    248 				if (in6_maxmtu < ndi->linkmtu)
    249 					in6_maxmtu = ndi->linkmtu;
    250 			} else
    251 				in6_setmaxmtu();
    252 		}
    253 	}
    254 #undef MIN
    255 }
    256 
    257 void
    258 nd6_option_init(opt, icmp6len, ndopts)
    259 	void *opt;
    260 	int icmp6len;
    261 	union nd_opts *ndopts;
    262 {
    263 	bzero(ndopts, sizeof(*ndopts));
    264 	ndopts->nd_opts_search = (struct nd_opt_hdr *)opt;
    265 	ndopts->nd_opts_last
    266 		= (struct nd_opt_hdr *)(((u_char *)opt) + icmp6len);
    267 
    268 	if (icmp6len == 0) {
    269 		ndopts->nd_opts_done = 1;
    270 		ndopts->nd_opts_search = NULL;
    271 	}
    272 }
    273 
    274 /*
    275  * Take one ND option.
    276  */
    277 struct nd_opt_hdr *
    278 nd6_option(ndopts)
    279 	union nd_opts *ndopts;
    280 {
    281 	struct nd_opt_hdr *nd_opt;
    282 	int olen;
    283 
    284 	if (!ndopts)
    285 		panic("ndopts == NULL in nd6_option\n");
    286 	if (!ndopts->nd_opts_last)
    287 		panic("uninitialized ndopts in nd6_option\n");
    288 	if (!ndopts->nd_opts_search)
    289 		return NULL;
    290 	if (ndopts->nd_opts_done)
    291 		return NULL;
    292 
    293 	nd_opt = ndopts->nd_opts_search;
    294 
    295 	/* make sure nd_opt_len is inside the buffer */
    296 	if ((caddr_t)&nd_opt->nd_opt_len >= (caddr_t)ndopts->nd_opts_last) {
    297 		bzero(ndopts, sizeof(*ndopts));
    298 		return NULL;
    299 	}
    300 
    301 	olen = nd_opt->nd_opt_len << 3;
    302 	if (olen == 0) {
    303 		/*
    304 		 * Message validation requires that all included
    305 		 * options have a length that is greater than zero.
    306 		 */
    307 		bzero(ndopts, sizeof(*ndopts));
    308 		return NULL;
    309 	}
    310 
    311 	ndopts->nd_opts_search = (struct nd_opt_hdr *)((caddr_t)nd_opt + olen);
    312 	if (ndopts->nd_opts_search > ndopts->nd_opts_last) {
    313 		/* option overruns the end of buffer, invalid */
    314 		bzero(ndopts, sizeof(*ndopts));
    315 		return NULL;
    316 	} else if (ndopts->nd_opts_search == ndopts->nd_opts_last) {
    317 		/* reached the end of options chain */
    318 		ndopts->nd_opts_done = 1;
    319 		ndopts->nd_opts_search = NULL;
    320 	}
    321 	return nd_opt;
    322 }
    323 
    324 /*
    325  * Parse multiple ND options.
    326  * This function is much easier to use, for ND routines that do not need
    327  * multiple options of the same type.
    328  */
    329 int
    330 nd6_options(ndopts)
    331 	union nd_opts *ndopts;
    332 {
    333 	struct nd_opt_hdr *nd_opt;
    334 	int i = 0;
    335 
    336 	if (!ndopts)
    337 		panic("ndopts == NULL in nd6_options\n");
    338 	if (!ndopts->nd_opts_last)
    339 		panic("uninitialized ndopts in nd6_options\n");
    340 	if (!ndopts->nd_opts_search)
    341 		return 0;
    342 
    343 	while (1) {
    344 		nd_opt = nd6_option(ndopts);
    345 		if (!nd_opt && !ndopts->nd_opts_last) {
    346 			/*
    347 			 * Message validation requires that all included
    348 			 * options have a length that is greater than zero.
    349 			 */
    350 			icmp6stat.icp6s_nd_badopt++;
    351 			bzero(ndopts, sizeof(*ndopts));
    352 			return -1;
    353 		}
    354 
    355 		if (!nd_opt)
    356 			goto skip1;
    357 
    358 		switch (nd_opt->nd_opt_type) {
    359 		case ND_OPT_SOURCE_LINKADDR:
    360 		case ND_OPT_TARGET_LINKADDR:
    361 		case ND_OPT_MTU:
    362 		case ND_OPT_REDIRECTED_HEADER:
    363 			if (ndopts->nd_opt_array[nd_opt->nd_opt_type]) {
    364 				nd6log((LOG_INFO,
    365 				    "duplicated ND6 option found (type=%d)\n",
    366 				    nd_opt->nd_opt_type));
    367 				/* XXX bark? */
    368 			} else {
    369 				ndopts->nd_opt_array[nd_opt->nd_opt_type]
    370 					= nd_opt;
    371 			}
    372 			break;
    373 		case ND_OPT_PREFIX_INFORMATION:
    374 			if (ndopts->nd_opt_array[nd_opt->nd_opt_type] == 0) {
    375 				ndopts->nd_opt_array[nd_opt->nd_opt_type]
    376 					= nd_opt;
    377 			}
    378 			ndopts->nd_opts_pi_end =
    379 				(struct nd_opt_prefix_info *)nd_opt;
    380 			break;
    381 		default:
    382 			/*
    383 			 * Unknown options must be silently ignored,
    384 			 * to accomodate future extension to the protocol.
    385 			 */
    386 			nd6log((LOG_DEBUG,
    387 			    "nd6_options: unsupported option %d - "
    388 			    "option ignored\n", nd_opt->nd_opt_type));
    389 		}
    390 
    391 skip1:
    392 		i++;
    393 		if (i > nd6_maxndopt) {
    394 			icmp6stat.icp6s_nd_toomanyopt++;
    395 			nd6log((LOG_INFO, "too many loop in nd opt\n"));
    396 			break;
    397 		}
    398 
    399 		if (ndopts->nd_opts_done)
    400 			break;
    401 	}
    402 
    403 	return 0;
    404 }
    405 
    406 /*
    407  * ND6 timer routine to expire default route list and prefix list
    408  */
    409 void
    410 nd6_timer(ignored_arg)
    411 	void	*ignored_arg;
    412 {
    413 	int s;
    414 	struct llinfo_nd6 *ln;
    415 	struct nd_defrouter *dr;
    416 	struct nd_prefix *pr;
    417 	long time_second = time.tv_sec;
    418 
    419 	s = splsoftnet();
    420 	callout_reset(&nd6_timer_ch, nd6_prune * hz,
    421 	    nd6_timer, NULL);
    422 
    423 	ln = llinfo_nd6.ln_next;
    424 	/* XXX BSD/OS separates this code -- itojun */
    425 	while (ln && ln != &llinfo_nd6) {
    426 		struct rtentry *rt;
    427 		struct ifnet *ifp;
    428 		struct sockaddr_in6 *dst;
    429 		struct llinfo_nd6 *next = ln->ln_next;
    430 		/* XXX: used for the DELAY case only: */
    431 		struct nd_ifinfo *ndi = NULL;
    432 
    433 		if ((rt = ln->ln_rt) == NULL) {
    434 			ln = next;
    435 			continue;
    436 		}
    437 		if ((ifp = rt->rt_ifp) == NULL) {
    438 			ln = next;
    439 			continue;
    440 		}
    441 		ndi = &nd_ifinfo[ifp->if_index];
    442 		dst = (struct sockaddr_in6 *)rt_key(rt);
    443 
    444 		if (ln->ln_expire > time_second) {
    445 			ln = next;
    446 			continue;
    447 		}
    448 
    449 		/* sanity check */
    450 		if (!rt)
    451 			panic("rt=0 in nd6_timer(ln=%p)\n", ln);
    452 		if (rt->rt_llinfo && (struct llinfo_nd6 *)rt->rt_llinfo != ln)
    453 			panic("rt_llinfo(%p) is not equal to ln(%p)\n",
    454 			      rt->rt_llinfo, ln);
    455 		if (!dst)
    456 			panic("dst=0 in nd6_timer(ln=%p)\n", ln);
    457 
    458 		switch (ln->ln_state) {
    459 		case ND6_LLINFO_INCOMPLETE:
    460 			if (ln->ln_asked < nd6_mmaxtries) {
    461 				ln->ln_asked++;
    462 				ln->ln_expire = time_second +
    463 					nd_ifinfo[ifp->if_index].retrans / 1000;
    464 				nd6_ns_output(ifp, NULL, &dst->sin6_addr,
    465 					ln, 0);
    466 			} else {
    467 				struct mbuf *m = ln->ln_hold;
    468 				if (m) {
    469 					if (rt->rt_ifp) {
    470 						/*
    471 						 * Fake rcvif to make ICMP error
    472 						 * more helpful in diagnosing
    473 						 * for the receiver.
    474 						 * XXX: should we consider
    475 						 * older rcvif?
    476 						 */
    477 						m->m_pkthdr.rcvif = rt->rt_ifp;
    478 					}
    479 					icmp6_error(m, ICMP6_DST_UNREACH,
    480 						    ICMP6_DST_UNREACH_ADDR, 0);
    481 					ln->ln_hold = NULL;
    482 				}
    483 				next = nd6_free(rt);
    484 			}
    485 			break;
    486 		case ND6_LLINFO_REACHABLE:
    487 			if (ln->ln_expire) {
    488 				ln->ln_state = ND6_LLINFO_STALE;
    489 				ln->ln_expire = time_second + nd6_gctimer;
    490 			}
    491 			break;
    492 
    493 		case ND6_LLINFO_STALE:
    494 			/* Garbage Collection(RFC 2461 5.3) */
    495 			if (ln->ln_expire)
    496 				next = nd6_free(rt);
    497 			break;
    498 
    499 		case ND6_LLINFO_DELAY:
    500 			if (ndi && (ndi->flags & ND6_IFF_PERFORMNUD) != 0) {
    501 				/* We need NUD */
    502 				ln->ln_asked = 1;
    503 				ln->ln_state = ND6_LLINFO_PROBE;
    504 				ln->ln_expire = time_second +
    505 					ndi->retrans / 1000;
    506 				nd6_ns_output(ifp, &dst->sin6_addr,
    507 					      &dst->sin6_addr,
    508 					      ln, 0);
    509 			} else {
    510 				ln->ln_state = ND6_LLINFO_STALE; /* XXX */
    511 				ln->ln_expire = time_second + nd6_gctimer;
    512 			}
    513 			break;
    514 		case ND6_LLINFO_PROBE:
    515 			if (ln->ln_asked < nd6_umaxtries) {
    516 				ln->ln_asked++;
    517 				ln->ln_expire = time_second +
    518 					nd_ifinfo[ifp->if_index].retrans / 1000;
    519 				nd6_ns_output(ifp, &dst->sin6_addr,
    520 					       &dst->sin6_addr, ln, 0);
    521 			} else
    522 				next = nd6_free(rt);
    523 			break;
    524 		}
    525 		ln = next;
    526 	}
    527 
    528 	/* expire default router list */
    529 	dr = TAILQ_FIRST(&nd_defrouter);
    530 	while (dr) {
    531 		if (dr->expire && dr->expire < time_second) {
    532 			struct nd_defrouter *t;
    533 			t = TAILQ_NEXT(dr, dr_entry);
    534 			defrtrlist_del(dr);
    535 			dr = t;
    536 		} else {
    537 			dr = TAILQ_NEXT(dr, dr_entry);
    538 		}
    539 	}
    540 	pr = nd_prefix.lh_first;
    541 	while (pr) {
    542 		struct in6_ifaddr *ia6;
    543 		struct in6_addrlifetime *lt6;
    544 
    545 		if (IN6_IS_ADDR_UNSPECIFIED(&pr->ndpr_addr))
    546 			ia6 = NULL;
    547 		else
    548 			ia6 = in6ifa_ifpwithaddr(pr->ndpr_ifp, &pr->ndpr_addr);
    549 
    550 		if (ia6) {
    551 			/* check address lifetime */
    552 			lt6 = &ia6->ia6_lifetime;
    553 			if (lt6->ia6t_preferred && lt6->ia6t_preferred < time_second)
    554 				ia6->ia6_flags |= IN6_IFF_DEPRECATED;
    555 			if (lt6->ia6t_expire && lt6->ia6t_expire < time_second) {
    556 				if (!IN6_IS_ADDR_UNSPECIFIED(&pr->ndpr_addr))
    557 					in6_ifdel(pr->ndpr_ifp, &pr->ndpr_addr);
    558 				/* xxx ND_OPT_PI_FLAG_ONLINK processing */
    559 			}
    560 		}
    561 
    562 		/*
    563 		 * check prefix lifetime.
    564 		 * since pltime is just for autoconf, pltime processing for
    565 		 * prefix is not necessary.
    566 		 *
    567 		 * we offset expire time by NDPR_KEEP_EXPIRE, so that we
    568 		 * can use the old prefix information to validate the
    569 		 * next prefix information to come.  See prelist_update()
    570 		 * for actual validation.
    571 		 */
    572 		if (pr->ndpr_expire
    573 		 && pr->ndpr_expire + NDPR_KEEP_EXPIRED < time_second) {
    574 			struct nd_prefix *t;
    575 			t = pr->ndpr_next;
    576 
    577 			/*
    578 			 * address expiration and prefix expiration are
    579 			 * separate.  NEVER perform in6_ifdel here.
    580 			 */
    581 
    582 			prelist_remove(pr);
    583 			pr = t;
    584 		} else
    585 			pr = pr->ndpr_next;
    586 	}
    587 	splx(s);
    588 }
    589 
    590 /*
    591  * Nuke neighbor cache/prefix/default router management table, right before
    592  * ifp goes away.
    593  */
    594 void
    595 nd6_purge(ifp)
    596 	struct ifnet *ifp;
    597 {
    598 	struct llinfo_nd6 *ln, *nln;
    599 	struct nd_defrouter *dr, *ndr, drany;
    600 	struct nd_prefix *pr, *npr;
    601 
    602 	/* Nuke default router list entries toward ifp */
    603 	if ((dr = TAILQ_FIRST(&nd_defrouter)) != NULL) {
    604 		/*
    605 		 * The first entry of the list may be stored in
    606 		 * the routing table, so we'll delete it later.
    607 		 */
    608 		for (dr = TAILQ_NEXT(dr, dr_entry); dr; dr = ndr) {
    609 			ndr = TAILQ_NEXT(dr, dr_entry);
    610 			if (dr->ifp == ifp)
    611 				defrtrlist_del(dr);
    612 		}
    613 		dr = TAILQ_FIRST(&nd_defrouter);
    614 		if (dr->ifp == ifp)
    615 			defrtrlist_del(dr);
    616 	}
    617 
    618 	/* Nuke prefix list entries toward ifp */
    619 	for (pr = nd_prefix.lh_first; pr; pr = npr) {
    620 		npr = pr->ndpr_next;
    621 		if (pr->ndpr_ifp == ifp) {
    622 			if (!IN6_IS_ADDR_UNSPECIFIED(&pr->ndpr_addr))
    623 				in6_ifdel(pr->ndpr_ifp, &pr->ndpr_addr);
    624 			prelist_remove(pr);
    625 		}
    626 	}
    627 
    628 	/* cancel default outgoing interface setting */
    629 	if (nd6_defifindex == ifp->if_index)
    630 		nd6_setdefaultiface(0);
    631 
    632 	if (!ip6_forwarding && ip6_accept_rtadv) { /* XXX: too restrictive? */
    633 		/* refresh default router list */
    634 		bzero(&drany, sizeof(drany));
    635 		defrouter_delreq(&drany, 0);
    636 		defrouter_select();
    637 	}
    638 
    639 	/*
    640 	 * Nuke neighbor cache entries for the ifp.
    641 	 * Note that rt->rt_ifp may not be the same as ifp,
    642 	 * due to KAME goto ours hack.  See RTM_RESOLVE case in
    643 	 * nd6_rtrequest(), and ip6_input().
    644 	 */
    645 	ln = llinfo_nd6.ln_next;
    646 	while (ln && ln != &llinfo_nd6) {
    647 		struct rtentry *rt;
    648 		struct sockaddr_dl *sdl;
    649 
    650 		nln = ln->ln_next;
    651 		rt = ln->ln_rt;
    652 		if (rt && rt->rt_gateway &&
    653 		    rt->rt_gateway->sa_family == AF_LINK) {
    654 			sdl = (struct sockaddr_dl *)rt->rt_gateway;
    655 			if (sdl->sdl_index == ifp->if_index)
    656 				nln = nd6_free(rt);
    657 		}
    658 		ln = nln;
    659 	}
    660 }
    661 
    662 struct rtentry *
    663 nd6_lookup(addr6, create, ifp)
    664 	struct in6_addr *addr6;
    665 	int create;
    666 	struct ifnet *ifp;
    667 {
    668 	struct rtentry *rt;
    669 	struct sockaddr_in6 sin6;
    670 
    671 	bzero(&sin6, sizeof(sin6));
    672 	sin6.sin6_len = sizeof(struct sockaddr_in6);
    673 	sin6.sin6_family = AF_INET6;
    674 	sin6.sin6_addr = *addr6;
    675 	rt = rtalloc1((struct sockaddr *)&sin6, create);
    676 	if (rt && (rt->rt_flags & RTF_LLINFO) == 0) {
    677 		/*
    678 		 * This is the case for the default route.
    679 		 * If we want to create a neighbor cache for the address, we
    680 		 * should free the route for the destination and allocate an
    681 		 * interface route.
    682 		 */
    683 		if (create) {
    684 			RTFREE(rt);
    685 			rt = 0;
    686 		}
    687 	}
    688 	if (!rt) {
    689 		if (create && ifp) {
    690 			int e;
    691 
    692 			/*
    693 			 * If no route is available and create is set,
    694 			 * we allocate a host route for the destination
    695 			 * and treat it like an interface route.
    696 			 * This hack is necessary for a neighbor which can't
    697 			 * be covered by our own prefix.
    698 			 */
    699 			struct ifaddr *ifa =
    700 				ifaof_ifpforaddr((struct sockaddr *)&sin6, ifp);
    701 			if (ifa == NULL)
    702 				return(NULL);
    703 
    704 			/*
    705 			 * Create a new route. RTF_LLINFO is necessary
    706 			 * to create a Neighbor Cache entry for the
    707 			 * destination in nd6_rtrequest which will be
    708 			 * called in rtequest via ifa->ifa_rtrequest.
    709 			 */
    710 			if ((e = rtrequest(RTM_ADD, (struct sockaddr *)&sin6,
    711 					   ifa->ifa_addr,
    712 					   (struct sockaddr *)&all1_sa,
    713 					   (ifa->ifa_flags |
    714 					    RTF_HOST | RTF_LLINFO) &
    715 					   ~RTF_CLONING,
    716 					   &rt)) != 0)
    717 				log(LOG_ERR,
    718 				    "nd6_lookup: failed to add route for a "
    719 				    "neighbor(%s), errno=%d\n",
    720 				    ip6_sprintf(addr6), e);
    721 			if (rt == NULL)
    722 				return(NULL);
    723 			if (rt->rt_llinfo) {
    724 				struct llinfo_nd6 *ln =
    725 					(struct llinfo_nd6 *)rt->rt_llinfo;
    726 				ln->ln_state = ND6_LLINFO_NOSTATE;
    727 			}
    728 		} else
    729 			return(NULL);
    730 	}
    731 	rt->rt_refcnt--;
    732 	/*
    733 	 * Validation for the entry.
    734 	 * XXX: we can't use rt->rt_ifp to check for the interface, since
    735 	 *      it might be the loopback interface if the entry is for our
    736 	 *      own address on a non-loopback interface. Instead, we should
    737 	 *      use rt->rt_ifa->ifa_ifp, which would specify the REAL interface.
    738 	 */
    739 	if ((rt->rt_flags & RTF_GATEWAY) || (rt->rt_flags & RTF_LLINFO) == 0 ||
    740 	    rt->rt_gateway->sa_family != AF_LINK ||
    741 	    (ifp && rt->rt_ifa->ifa_ifp != ifp)) {
    742 		if (create) {
    743 			log(LOG_DEBUG, "nd6_lookup: failed to lookup %s (if = %s)\n",
    744 			    ip6_sprintf(addr6), ifp ? if_name(ifp) : "unspec");
    745 			/* xxx more logs... kazu */
    746 		}
    747 		return(0);
    748 	}
    749 	return(rt);
    750 }
    751 
    752 /*
    753  * Detect if a given IPv6 address identifies a neighbor on a given link.
    754  * XXX: should take care of the destination of a p2p link?
    755  */
    756 int
    757 nd6_is_addr_neighbor(addr, ifp)
    758 	struct sockaddr_in6 *addr;
    759 	struct ifnet *ifp;
    760 {
    761 	struct ifaddr *ifa;
    762 	int i;
    763 
    764 #define IFADDR6(a) ((((struct in6_ifaddr *)(a))->ia_addr).sin6_addr)
    765 #define IFMASK6(a) ((((struct in6_ifaddr *)(a))->ia_prefixmask).sin6_addr)
    766 
    767 	/*
    768 	 * A link-local address is always a neighbor.
    769 	 * XXX: we should use the sin6_scope_id field rather than the embedded
    770 	 * interface index.
    771 	 */
    772 	if (IN6_IS_ADDR_LINKLOCAL(&addr->sin6_addr) &&
    773 	    ntohs(*(u_int16_t *)&addr->sin6_addr.s6_addr[2]) == ifp->if_index)
    774 		return(1);
    775 
    776 	/*
    777 	 * If the address matches one of our addresses,
    778 	 * it should be a neighbor.
    779 	 */
    780 	for (ifa = ifp->if_addrlist.tqh_first;
    781 	     ifa;
    782 	     ifa = ifa->ifa_list.tqe_next)
    783 	{
    784 		if (ifa->ifa_addr->sa_family != AF_INET6)
    785 			next: continue;
    786 
    787 		for (i = 0; i < 4; i++) {
    788 			if ((IFADDR6(ifa).s6_addr32[i] ^
    789 			     addr->sin6_addr.s6_addr32[i]) &
    790 			    IFMASK6(ifa).s6_addr32[i])
    791 				goto next;
    792 		}
    793 		return(1);
    794 	}
    795 
    796 	/*
    797 	 * Even if the address matches none of our addresses, it might be
    798 	 * in the neighbor cache.
    799 	 */
    800 	if (nd6_lookup(&addr->sin6_addr, 0, ifp))
    801 		return(1);
    802 
    803 	return(0);
    804 #undef IFADDR6
    805 #undef IFMASK6
    806 }
    807 
    808 /*
    809  * Free an nd6 llinfo entry.
    810  */
    811 struct llinfo_nd6 *
    812 nd6_free(rt)
    813 	struct rtentry *rt;
    814 {
    815 	struct llinfo_nd6 *ln = (struct llinfo_nd6 *)rt->rt_llinfo, *next;
    816 	struct in6_addr in6 = ((struct sockaddr_in6 *)rt_key(rt))->sin6_addr;
    817 	struct nd_defrouter *dr;
    818 
    819 	/*
    820 	 * Clear all destination cache entries for the neighbor.
    821 	 * XXX: is it better to restrict this to hosts?
    822 	 */
    823 	pfctlinput(PRC_HOSTDEAD, rt_key(rt));
    824 
    825 	if (!ip6_forwarding && ip6_accept_rtadv) { /* XXX: too restrictive? */
    826 		int s;
    827 		s = splsoftnet();
    828 		dr = defrouter_lookup(&((struct sockaddr_in6 *)rt_key(rt))->sin6_addr,
    829 				      rt->rt_ifp);
    830 		if (ln->ln_router || dr) {
    831 			/*
    832 			 * rt6_flush must be called whether or not the neighbor
    833 			 * is in the Default Router List.
    834 			 * See a corresponding comment in nd6_na_input().
    835 			 */
    836 			rt6_flush(&in6, rt->rt_ifp);
    837 		}
    838 
    839 		if (dr) {
    840 			/*
    841 			 * Unreachablity of a router might affect the default
    842 			 * router selection and on-link detection of advertised
    843 			 * prefixes.
    844 			 */
    845 
    846 			/*
    847 			 * Temporarily fake the state to choose a new default
    848 			 * router and to perform on-link determination of
    849 			 * prefixes coreectly.
    850 			 * Below the state will be set correctly,
    851 			 * or the entry itself will be deleted.
    852 			 */
    853 			ln->ln_state = ND6_LLINFO_INCOMPLETE;
    854 
    855 			if (dr == TAILQ_FIRST(&nd_defrouter)) {
    856 				/*
    857 				 * It is used as the current default router,
    858 				 * so we have to move it to the end of the
    859 				 * list and choose a new one.
    860 				 * XXX: it is not very efficient if this is
    861 				 *      the only router.
    862 				 */
    863 				TAILQ_REMOVE(&nd_defrouter, dr, dr_entry);
    864 				TAILQ_INSERT_TAIL(&nd_defrouter, dr, dr_entry);
    865 
    866 				defrouter_select();
    867 			}
    868 			pfxlist_onlink_check();
    869 		}
    870 		splx(s);
    871 	}
    872 
    873 	/*
    874 	 * Before deleting the entry, remember the next entry as the
    875 	 * return value.  We need this because pfxlist_onlink_check() above
    876 	 * might have freed other entries (particularly the old next entry) as
    877 	 * a side effect (XXX).
    878 	 */
    879 	next = ln->ln_next;
    880 
    881 	/*
    882 	 * Detach the route from the routing tree and the list of neighbor
    883 	 * caches, and disable the route entry not to be used in already
    884 	 * cached routes.
    885 	 */
    886 	rtrequest(RTM_DELETE, rt_key(rt), (struct sockaddr *)0,
    887 		  rt_mask(rt), 0, (struct rtentry **)0);
    888 
    889 	return(next);
    890 }
    891 
    892 /*
    893  * Upper-layer reachability hint for Neighbor Unreachability Detection.
    894  *
    895  * XXX cost-effective metods?
    896  */
    897 void
    898 nd6_nud_hint(rt, dst6, force)
    899 	struct rtentry *rt;
    900 	struct in6_addr *dst6;
    901 	int force;
    902 {
    903 	struct llinfo_nd6 *ln;
    904 	long time_second = time.tv_sec;
    905 
    906 	/*
    907 	 * If the caller specified "rt", use that.  Otherwise, resolve the
    908 	 * routing table by supplied "dst6".
    909 	 */
    910 	if (!rt) {
    911 		if (!dst6)
    912 			return;
    913 		if (!(rt = nd6_lookup(dst6, 0, NULL)))
    914 			return;
    915 	}
    916 
    917 	if ((rt->rt_flags & RTF_GATEWAY) != 0 ||
    918 	    (rt->rt_flags & RTF_LLINFO) == 0 ||
    919 	    !rt->rt_llinfo || !rt->rt_gateway ||
    920 	    rt->rt_gateway->sa_family != AF_LINK) {
    921 		/* This is not a host route. */
    922 		return;
    923 	}
    924 
    925 	ln = (struct llinfo_nd6 *)rt->rt_llinfo;
    926 	if (ln->ln_state < ND6_LLINFO_REACHABLE)
    927 		return;
    928 
    929 	/*
    930 	 * if we get upper-layer reachability confirmation many times,
    931 	 * it is possible we have false information.
    932 	 */
    933 	if (!force) {
    934 		ln->ln_byhint++;
    935 		if (ln->ln_byhint > nd6_maxnudhint)
    936 			return;
    937 	}
    938 
    939 	ln->ln_state = ND6_LLINFO_REACHABLE;
    940 	if (ln->ln_expire)
    941 		ln->ln_expire = time_second +
    942 			nd_ifinfo[rt->rt_ifp->if_index].reachable;
    943 }
    944 
    945 #ifdef OLDIP6OUTPUT
    946 /*
    947  * Resolve an IP6 address into an ethernet address. If success,
    948  * desten is filled in. If there is no entry in ndptab,
    949  * set one up and multicast a solicitation for the IP6 address.
    950  * Hold onto this mbuf and resend it once the address
    951  * is finally resolved. A return value of 1 indicates
    952  * that desten has been filled in and the packet should be sent
    953  * normally; a 0 return indicates that the packet has been
    954  * taken over here, either now or for later transmission.
    955  */
    956 int
    957 nd6_resolve(ifp, rt, m, dst, desten)
    958 	struct ifnet *ifp;
    959 	struct rtentry *rt;
    960 	struct mbuf *m;
    961 	struct sockaddr *dst;
    962 	u_char *desten;
    963 {
    964 	struct llinfo_nd6 *ln = (struct llinfo_nd6 *)NULL;
    965 	struct sockaddr_dl *sdl;
    966 	long time_second = time.tv_sec;
    967 
    968 	if (m->m_flags & M_MCAST) {
    969 		switch (ifp->if_type) {
    970 		case IFT_ETHER:
    971 		case IFT_FDDI:
    972 			ETHER_MAP_IPV6_MULTICAST(&SIN6(dst)->sin6_addr,
    973 						 desten);
    974 			return(1);
    975 		case IFT_IEEE1394:
    976 			bcopy(ifp->if_broadcastaddr, desten, ifp->if_addrlen);
    977 			return(1);
    978 		case IFT_ARCNET:
    979 			*desten = 0;
    980 			return(1);
    981 		default:
    982 			m_freem(m);
    983 			return(0);
    984 		}
    985 	}
    986 	if (rt && (rt->rt_flags & RTF_LLINFO) != 0)
    987 		ln = (struct llinfo_nd6 *)rt->rt_llinfo;
    988 	else {
    989 		if ((rt = nd6_lookup(&(SIN6(dst)->sin6_addr), 1, ifp)) != NULL)
    990 			ln = (struct llinfo_nd6 *)rt->rt_llinfo;
    991 	}
    992 	if (!ln || !rt) {
    993 		log(LOG_DEBUG, "nd6_resolve: can't allocate llinfo for %s\n",
    994 			ip6_sprintf(&(SIN6(dst)->sin6_addr)));
    995 		m_freem(m);
    996 		return(0);
    997 	}
    998 	sdl = SDL(rt->rt_gateway);
    999 	/*
   1000 	 * Ckeck the address family and length is valid, the address
   1001 	 * is resolved; otherwise, try to resolve.
   1002 	 */
   1003 	if (ln->ln_state >= ND6_LLINFO_REACHABLE
   1004 	   && sdl->sdl_family == AF_LINK
   1005 	   && sdl->sdl_alen != 0) {
   1006 		bcopy(LLADDR(sdl), desten, sdl->sdl_alen);
   1007 		if (ln->ln_state == ND6_LLINFO_STALE) {
   1008 			ln->ln_asked = 0;
   1009 			ln->ln_state = ND6_LLINFO_DELAY;
   1010 			ln->ln_expire = time_second + nd6_delay;
   1011 		}
   1012 		return(1);
   1013 	}
   1014 	/*
   1015 	 * There is an ndp entry, but no ethernet address
   1016 	 * response yet. Replace the held mbuf with this
   1017 	 * latest one.
   1018 	 *
   1019 	 * XXX Does the code conform to rate-limiting rule?
   1020 	 * (RFC 2461 7.2.2)
   1021 	 */
   1022 	if (ln->ln_state == ND6_LLINFO_NOSTATE)
   1023 		ln->ln_state = ND6_LLINFO_INCOMPLETE;
   1024 	if (ln->ln_hold)
   1025 		m_freem(ln->ln_hold);
   1026 	ln->ln_hold = m;
   1027 	if (ln->ln_expire) {
   1028 		if (ln->ln_asked < nd6_mmaxtries &&
   1029 		    ln->ln_expire < time_second) {
   1030 			ln->ln_asked++;
   1031 			ln->ln_expire = time_second +
   1032 				nd_ifinfo[ifp->if_index].retrans / 1000;
   1033 			nd6_ns_output(ifp, NULL, &(SIN6(dst)->sin6_addr),
   1034 				ln, 0);
   1035 		}
   1036 	}
   1037 	/* do not free mbuf here, it is queued into llinfo_nd6 */
   1038 	return(0);
   1039 }
   1040 #endif /* OLDIP6OUTPUT */
   1041 
   1042 void
   1043 nd6_rtrequest(req, rt, info)
   1044 	int	req;
   1045 	struct rtentry *rt;
   1046 	struct rt_addrinfo *info; /* xxx unused */
   1047 {
   1048 	struct sockaddr *gate = rt->rt_gateway;
   1049 	struct llinfo_nd6 *ln = (struct llinfo_nd6 *)rt->rt_llinfo;
   1050 	static struct sockaddr_dl null_sdl = {sizeof(null_sdl), AF_LINK};
   1051 	struct ifnet *ifp = rt->rt_ifp;
   1052 	struct ifaddr *ifa;
   1053 	long time_second = time.tv_sec;
   1054 
   1055 	if (rt->rt_flags & RTF_GATEWAY)
   1056 		return;
   1057 
   1058 	switch (req) {
   1059 	case RTM_ADD:
   1060 		/*
   1061 		 * There is no backward compatibility :)
   1062 		 *
   1063 		 * if ((rt->rt_flags & RTF_HOST) == 0 &&
   1064 		 *     SIN(rt_mask(rt))->sin_addr.s_addr != 0xffffffff)
   1065 		 *	   rt->rt_flags |= RTF_CLONING;
   1066 		 */
   1067 		if (rt->rt_flags & (RTF_CLONING | RTF_LLINFO)) {
   1068 			/*
   1069 			 * Case 1: This route should come from
   1070 			 * a route to interface. RTF_LLINFO flag is set
   1071 			 * for a host route whose destination should be
   1072 			 * treated as on-link.
   1073 			 */
   1074 			rt_setgate(rt, rt_key(rt),
   1075 				   (struct sockaddr *)&null_sdl);
   1076 			gate = rt->rt_gateway;
   1077 			SDL(gate)->sdl_type = ifp->if_type;
   1078 			SDL(gate)->sdl_index = ifp->if_index;
   1079 			if (ln)
   1080 				ln->ln_expire = time_second;
   1081 #if 1
   1082 			if (ln && ln->ln_expire == 0) {
   1083 				/* kludge for desktops */
   1084 #if 0
   1085 				printf("nd6_request: time.tv_sec is zero; "
   1086 				       "treat it as 1\n");
   1087 #endif
   1088 				ln->ln_expire = 1;
   1089 			}
   1090 #endif
   1091 			if (rt->rt_flags & RTF_CLONING)
   1092 				break;
   1093 		}
   1094 		/*
   1095 		 * In IPv4 code, we try to annonuce new RTF_ANNOUNCE entry here.
   1096 		 * We don't do that here since llinfo is not ready yet.
   1097 		 *
   1098 		 * There are also couple of other things to be discussed:
   1099 		 * - unsolicited NA code needs improvement beforehand
   1100 		 * - RFC2461 says we MAY send multicast unsolicited NA
   1101 		 *   (7.2.6 paragraph 4), however, it also says that we
   1102 		 *   SHOULD provide a mechanism to prevent multicast NA storm.
   1103 		 *   we don't have anything like it right now.
   1104 		 *   note that the mechanism needs a mutual agreement
   1105 		 *   between proxies, which means that we need to implement
   1106 		 *   a new protocol, or a new kludge.
   1107 		 * - from RFC2461 6.2.4, host MUST NOT send an unsolicited NA.
   1108 		 *   we need to check ip6forwarding before sending it.
   1109 		 *   (or should we allow proxy ND configuration only for
   1110 		 *   routers?  there's no mention about proxy ND from hosts)
   1111 		 */
   1112 #if 0
   1113 		/* XXX it does not work */
   1114 		if (rt->rt_flags & RTF_ANNOUNCE)
   1115 			nd6_na_output(ifp,
   1116 			      &SIN6(rt_key(rt))->sin6_addr,
   1117 			      &SIN6(rt_key(rt))->sin6_addr,
   1118 			      ip6_forwarding ? ND_NA_FLAG_ROUTER : 0,
   1119 			      1, NULL);
   1120 #endif
   1121 		/* FALLTHROUGH */
   1122 	case RTM_RESOLVE:
   1123 		if ((ifp->if_flags & IFF_POINTOPOINT) == 0) {
   1124 			/*
   1125 			 * Address resolution isn't necessary for a point to
   1126 			 * point link, so we can skip this test for a p2p link.
   1127 			 */
   1128 			if (gate->sa_family != AF_LINK ||
   1129 			    gate->sa_len < sizeof(null_sdl)) {
   1130 				log(LOG_DEBUG,
   1131 				    "nd6_rtrequest: bad gateway value: %s\n",
   1132 				    if_name(ifp));
   1133 				break;
   1134 			}
   1135 			SDL(gate)->sdl_type = ifp->if_type;
   1136 			SDL(gate)->sdl_index = ifp->if_index;
   1137 		}
   1138 		if (ln != NULL)
   1139 			break;	/* This happens on a route change */
   1140 		/*
   1141 		 * Case 2: This route may come from cloning, or a manual route
   1142 		 * add with a LL address.
   1143 		 */
   1144 		R_Malloc(ln, struct llinfo_nd6 *, sizeof(*ln));
   1145 		rt->rt_llinfo = (caddr_t)ln;
   1146 		if (!ln) {
   1147 			log(LOG_DEBUG, "nd6_rtrequest: malloc failed\n");
   1148 			break;
   1149 		}
   1150 		nd6_inuse++;
   1151 		nd6_allocated++;
   1152 		Bzero(ln, sizeof(*ln));
   1153 		ln->ln_rt = rt;
   1154 		/* this is required for "ndp" command. - shin */
   1155 		if (req == RTM_ADD) {
   1156 		        /*
   1157 			 * gate should have some valid AF_LINK entry,
   1158 			 * and ln->ln_expire should have some lifetime
   1159 			 * which is specified by ndp command.
   1160 			 */
   1161 			ln->ln_state = ND6_LLINFO_REACHABLE;
   1162 			ln->ln_byhint = 0;
   1163 		} else {
   1164 		        /*
   1165 			 * When req == RTM_RESOLVE, rt is created and
   1166 			 * initialized in rtrequest(), so rt_expire is 0.
   1167 			 */
   1168 			ln->ln_state = ND6_LLINFO_NOSTATE;
   1169 			ln->ln_expire = time_second;
   1170 		}
   1171 		rt->rt_flags |= RTF_LLINFO;
   1172 		ln->ln_next = llinfo_nd6.ln_next;
   1173 		llinfo_nd6.ln_next = ln;
   1174 		ln->ln_prev = &llinfo_nd6;
   1175 		ln->ln_next->ln_prev = ln;
   1176 
   1177 		/*
   1178 		 * check if rt_key(rt) is one of my address assigned
   1179 		 * to the interface.
   1180 		 */
   1181 		ifa = (struct ifaddr *)in6ifa_ifpwithaddr(rt->rt_ifp,
   1182 					  &SIN6(rt_key(rt))->sin6_addr);
   1183 		if (ifa) {
   1184 			caddr_t macp = nd6_ifptomac(ifp);
   1185 			ln->ln_expire = 0;
   1186 			ln->ln_state = ND6_LLINFO_REACHABLE;
   1187 			ln->ln_byhint = 0;
   1188 			if (macp) {
   1189 				Bcopy(macp, LLADDR(SDL(gate)), ifp->if_addrlen);
   1190 				SDL(gate)->sdl_alen = ifp->if_addrlen;
   1191 			}
   1192 			if (nd6_useloopback) {
   1193 				rt->rt_ifp = &loif[0];	/*XXX*/
   1194 				/*
   1195 				 * Make sure rt_ifa be equal to the ifaddr
   1196 				 * corresponding to the address.
   1197 				 * We need this because when we refer
   1198 				 * rt_ifa->ia6_flags in ip6_input, we assume
   1199 				 * that the rt_ifa points to the address instead
   1200 				 * of the loopback address.
   1201 				 */
   1202 				if (ifa != rt->rt_ifa) {
   1203 					IFAFREE(rt->rt_ifa);
   1204 					IFAREF(ifa);
   1205 					rt->rt_ifa = ifa;
   1206 				}
   1207 			}
   1208 		} else if (rt->rt_flags & RTF_ANNOUNCE) {
   1209 			ln->ln_expire = 0;
   1210 			ln->ln_state = ND6_LLINFO_REACHABLE;
   1211 			ln->ln_byhint = 0;
   1212 
   1213 			/* join solicited node multicast for proxy ND */
   1214 			if (ifp->if_flags & IFF_MULTICAST) {
   1215 				struct in6_addr llsol;
   1216 				int error;
   1217 
   1218 				llsol = SIN6(rt_key(rt))->sin6_addr;
   1219 				llsol.s6_addr16[0] = htons(0xff02);
   1220 				llsol.s6_addr16[1] = htons(ifp->if_index);
   1221 				llsol.s6_addr32[1] = 0;
   1222 				llsol.s6_addr32[2] = htonl(1);
   1223 				llsol.s6_addr8[12] = 0xff;
   1224 
   1225 				if (!in6_addmulti(&llsol, ifp, &error)) {
   1226 					nd6log((LOG_ERR, "%s: "
   1227 					    "failed to join %s (errno=%d)\n",
   1228 					    if_name(ifp), ip6_sprintf(&llsol),
   1229 					    error));
   1230 				}
   1231 			}
   1232 		}
   1233 		break;
   1234 
   1235 	case RTM_DELETE:
   1236 		if (!ln)
   1237 			break;
   1238 		/* leave from solicited node multicast for proxy ND */
   1239 		if ((rt->rt_flags & RTF_ANNOUNCE) != 0 &&
   1240 		    (ifp->if_flags & IFF_MULTICAST) != 0) {
   1241 			struct in6_addr llsol;
   1242 			struct in6_multi *in6m;
   1243 
   1244 			llsol = SIN6(rt_key(rt))->sin6_addr;
   1245 			llsol.s6_addr16[0] = htons(0xff02);
   1246 			llsol.s6_addr16[1] = htons(ifp->if_index);
   1247 			llsol.s6_addr32[1] = 0;
   1248 			llsol.s6_addr32[2] = htonl(1);
   1249 			llsol.s6_addr8[12] = 0xff;
   1250 
   1251 			IN6_LOOKUP_MULTI(llsol, ifp, in6m);
   1252 			if (in6m)
   1253 				in6_delmulti(in6m);
   1254 		}
   1255 		nd6_inuse--;
   1256 		ln->ln_next->ln_prev = ln->ln_prev;
   1257 		ln->ln_prev->ln_next = ln->ln_next;
   1258 		ln->ln_prev = NULL;
   1259 		rt->rt_llinfo = 0;
   1260 		rt->rt_flags &= ~RTF_LLINFO;
   1261 		if (ln->ln_hold)
   1262 			m_freem(ln->ln_hold);
   1263 		Free((caddr_t)ln);
   1264 	}
   1265 }
   1266 
   1267 void
   1268 nd6_p2p_rtrequest(req, rt, info)
   1269 	int	req;
   1270 	struct rtentry *rt;
   1271 	struct rt_addrinfo *info; /* xxx unused */
   1272 {
   1273 	struct sockaddr *gate = rt->rt_gateway;
   1274 	static struct sockaddr_dl null_sdl = {sizeof(null_sdl), AF_LINK};
   1275 	struct ifnet *ifp = rt->rt_ifp;
   1276 	struct ifaddr *ifa;
   1277 
   1278 	if (rt->rt_flags & RTF_GATEWAY)
   1279 		return;
   1280 
   1281 	switch (req) {
   1282 	case RTM_ADD:
   1283 		/*
   1284 		 * There is no backward compatibility :)
   1285 		 *
   1286 		 * if ((rt->rt_flags & RTF_HOST) == 0 &&
   1287 		 *     SIN(rt_mask(rt))->sin_addr.s_addr != 0xffffffff)
   1288 		 *	   rt->rt_flags |= RTF_CLONING;
   1289 		 */
   1290 		if (rt->rt_flags & RTF_CLONING) {
   1291 			/*
   1292 			 * Case 1: This route should come from
   1293 			 * a route to interface.
   1294 			 */
   1295 			rt_setgate(rt, rt_key(rt),
   1296 				   (struct sockaddr *)&null_sdl);
   1297 			gate = rt->rt_gateway;
   1298 			SDL(gate)->sdl_type = ifp->if_type;
   1299 			SDL(gate)->sdl_index = ifp->if_index;
   1300 			break;
   1301 		}
   1302 		/* Announce a new entry if requested. */
   1303 		if (rt->rt_flags & RTF_ANNOUNCE)
   1304 			nd6_na_output(ifp,
   1305 				      &SIN6(rt_key(rt))->sin6_addr,
   1306 				      &SIN6(rt_key(rt))->sin6_addr,
   1307 				      ip6_forwarding ? ND_NA_FLAG_ROUTER : 0,
   1308 				      1, NULL);
   1309 		/* FALLTHROUGH */
   1310 	case RTM_RESOLVE:
   1311 		/*
   1312 		 * check if rt_key(rt) is one of my address assigned
   1313 		 * to the interface.
   1314 		 */
   1315  		ifa = (struct ifaddr *)in6ifa_ifpwithaddr(rt->rt_ifp,
   1316 					  &SIN6(rt_key(rt))->sin6_addr);
   1317 		if (ifa) {
   1318 			if (nd6_useloopback) {
   1319 				rt->rt_ifp = &loif[0];	/*XXX*/
   1320 			}
   1321 		}
   1322 		break;
   1323 	}
   1324 }
   1325 
   1326 int
   1327 nd6_ioctl(cmd, data, ifp)
   1328 	u_long cmd;
   1329 	caddr_t	data;
   1330 	struct ifnet *ifp;
   1331 {
   1332 	struct in6_drlist *drl = (struct in6_drlist *)data;
   1333 	struct in6_prlist *prl = (struct in6_prlist *)data;
   1334 	struct in6_ndireq *ndi = (struct in6_ndireq *)data;
   1335 	struct in6_nbrinfo *nbi = (struct in6_nbrinfo *)data;
   1336 	struct in6_ndifreq *ndif = (struct in6_ndifreq *)data;
   1337 	struct nd_defrouter *dr, any;
   1338 	struct nd_prefix *pr;
   1339 	struct rtentry *rt;
   1340 	int i = 0, error = 0;
   1341 	int s;
   1342 
   1343 	switch (cmd) {
   1344 	case SIOCGDRLST_IN6:
   1345 		bzero(drl, sizeof(*drl));
   1346 		s = splsoftnet();
   1347 		dr = TAILQ_FIRST(&nd_defrouter);
   1348 		while (dr && i < DRLSTSIZ) {
   1349 			drl->defrouter[i].rtaddr = dr->rtaddr;
   1350 			if (IN6_IS_ADDR_LINKLOCAL(&drl->defrouter[i].rtaddr)) {
   1351 				/* XXX: need to this hack for KAME stack */
   1352 				drl->defrouter[i].rtaddr.s6_addr16[1] = 0;
   1353 			} else
   1354 				log(LOG_ERR,
   1355 				    "default router list contains a "
   1356 				    "non-linklocal address(%s)\n",
   1357 				    ip6_sprintf(&drl->defrouter[i].rtaddr));
   1358 
   1359 			drl->defrouter[i].flags = dr->flags;
   1360 			drl->defrouter[i].rtlifetime = dr->rtlifetime;
   1361 			drl->defrouter[i].expire = dr->expire;
   1362 			drl->defrouter[i].if_index = dr->ifp->if_index;
   1363 			i++;
   1364 			dr = TAILQ_NEXT(dr, dr_entry);
   1365 		}
   1366 		splx(s);
   1367 		break;
   1368 	case SIOCGPRLST_IN6:
   1369 		/*
   1370 		 * XXX meaning of fields, especialy "raflags", is very
   1371 		 * differnet between RA prefix list and RR/static prefix list.
   1372 		 * how about separating ioctls into two?
   1373 		 */
   1374 		bzero(prl, sizeof(*prl));
   1375 		s = splsoftnet();
   1376 		pr = nd_prefix.lh_first;
   1377 		while (pr && i < PRLSTSIZ) {
   1378 			struct nd_pfxrouter *pfr;
   1379 			int j;
   1380 
   1381 			prl->prefix[i].prefix = pr->ndpr_prefix.sin6_addr;
   1382 			prl->prefix[i].raflags = pr->ndpr_raf;
   1383 			prl->prefix[i].prefixlen = pr->ndpr_plen;
   1384 			prl->prefix[i].vltime = pr->ndpr_vltime;
   1385 			prl->prefix[i].pltime = pr->ndpr_pltime;
   1386 			prl->prefix[i].if_index = pr->ndpr_ifp->if_index;
   1387 			prl->prefix[i].expire = pr->ndpr_expire;
   1388 
   1389 			pfr = pr->ndpr_advrtrs.lh_first;
   1390 			j = 0;
   1391 			while (pfr) {
   1392 				if (j < DRLSTSIZ) {
   1393 #define RTRADDR prl->prefix[i].advrtr[j]
   1394 					RTRADDR = pfr->router->rtaddr;
   1395 					if (IN6_IS_ADDR_LINKLOCAL(&RTRADDR)) {
   1396 						/* XXX: hack for KAME */
   1397 						RTRADDR.s6_addr16[1] = 0;
   1398 					} else
   1399 						log(LOG_ERR,
   1400 						    "a router(%s) advertises "
   1401 						    "a prefix with "
   1402 						    "non-link local address\n",
   1403 						    ip6_sprintf(&RTRADDR));
   1404 #undef RTRADDR
   1405 				}
   1406 				j++;
   1407 				pfr = pfr->pfr_next;
   1408 			}
   1409 			prl->prefix[i].advrtrs = j;
   1410 			prl->prefix[i].origin = PR_ORIG_RA;
   1411 
   1412 			i++;
   1413 			pr = pr->ndpr_next;
   1414 		}
   1415 	      {
   1416 		struct rr_prefix *rpp;
   1417 
   1418 		for (rpp = LIST_FIRST(&rr_prefix); rpp;
   1419 		     rpp = LIST_NEXT(rpp, rp_entry)) {
   1420 			if (i >= PRLSTSIZ)
   1421 				break;
   1422 			prl->prefix[i].prefix = rpp->rp_prefix.sin6_addr;
   1423 			prl->prefix[i].raflags = rpp->rp_raf;
   1424 			prl->prefix[i].prefixlen = rpp->rp_plen;
   1425 			prl->prefix[i].vltime = rpp->rp_vltime;
   1426 			prl->prefix[i].pltime = rpp->rp_pltime;
   1427 			prl->prefix[i].if_index = rpp->rp_ifp->if_index;
   1428 			prl->prefix[i].expire = rpp->rp_expire;
   1429 			prl->prefix[i].advrtrs = 0;
   1430 			prl->prefix[i].origin = rpp->rp_origin;
   1431 			i++;
   1432 		}
   1433 	      }
   1434 		splx(s);
   1435 
   1436 		break;
   1437 	case SIOCGIFINFO_IN6:
   1438 		if (!nd_ifinfo || i >= nd_ifinfo_indexlim) {
   1439 			error = EINVAL;
   1440 			break;
   1441 		}
   1442 		ndi->ndi = nd_ifinfo[ifp->if_index];
   1443 		break;
   1444 	case SIOCSIFINFO_FLAGS:
   1445 		/* XXX: almost all other fields of ndi->ndi is unused */
   1446 		if (!nd_ifinfo || i >= nd_ifinfo_indexlim) {
   1447 			error = EINVAL;
   1448 			break;
   1449 		}
   1450 		nd_ifinfo[ifp->if_index].flags = ndi->ndi.flags;
   1451 		break;
   1452 	case SIOCSNDFLUSH_IN6:	/* XXX: the ioctl name is confusing... */
   1453 		/* flush default router list */
   1454 		/*
   1455 		 * xxx sumikawa: should not delete route if default
   1456 		 * route equals to the top of default router list
   1457 		 */
   1458 		bzero(&any, sizeof(any));
   1459 		defrouter_delreq(&any, 0);
   1460 		defrouter_select();
   1461 		/* xxx sumikawa: flush prefix list */
   1462 		break;
   1463 	case SIOCSPFXFLUSH_IN6:
   1464 	    {
   1465 		/* flush all the prefix advertised by routers */
   1466 		struct nd_prefix *pr, *next;
   1467 
   1468 		s = splsoftnet();
   1469 		for (pr = nd_prefix.lh_first; pr; pr = next) {
   1470 			next = pr->ndpr_next;
   1471 			if (!IN6_IS_ADDR_UNSPECIFIED(&pr->ndpr_addr))
   1472 				in6_ifdel(pr->ndpr_ifp, &pr->ndpr_addr);
   1473 			prelist_remove(pr);
   1474 		}
   1475 		splx(s);
   1476 		break;
   1477 	    }
   1478 	case SIOCSRTRFLUSH_IN6:
   1479 	    {
   1480 		/* flush all the default routers */
   1481 		struct nd_defrouter *dr, *next;
   1482 
   1483 		s = splsoftnet();
   1484 		if ((dr = TAILQ_FIRST(&nd_defrouter)) != NULL) {
   1485 			/*
   1486 			 * The first entry of the list may be stored in
   1487 			 * the routing table, so we'll delete it later.
   1488 			 */
   1489 			for (dr = TAILQ_NEXT(dr, dr_entry); dr; dr = next) {
   1490 				next = TAILQ_NEXT(dr, dr_entry);
   1491 				defrtrlist_del(dr);
   1492 			}
   1493 			defrtrlist_del(TAILQ_FIRST(&nd_defrouter));
   1494 		}
   1495 		splx(s);
   1496 		break;
   1497 	    }
   1498 	case SIOCGNBRINFO_IN6:
   1499 	    {
   1500 		struct llinfo_nd6 *ln;
   1501 		struct in6_addr nb_addr = nbi->addr; /* make local for safety */
   1502 
   1503 		/*
   1504 		 * XXX: KAME specific hack for scoped addresses
   1505 		 *      XXXX: for other scopes than link-local?
   1506 		 */
   1507 		if (IN6_IS_ADDR_LINKLOCAL(&nbi->addr) ||
   1508 		    IN6_IS_ADDR_MC_LINKLOCAL(&nbi->addr)) {
   1509 			u_int16_t *idp = (u_int16_t *)&nb_addr.s6_addr[2];
   1510 
   1511 			if (*idp == 0)
   1512 				*idp = htons(ifp->if_index);
   1513 		}
   1514 
   1515 		s = splsoftnet();
   1516 		if ((rt = nd6_lookup(&nb_addr, 0, ifp)) == NULL) {
   1517 			error = EINVAL;
   1518 			splx(s);
   1519 			break;
   1520 		}
   1521 		ln = (struct llinfo_nd6 *)rt->rt_llinfo;
   1522 		nbi->state = ln->ln_state;
   1523 		nbi->asked = ln->ln_asked;
   1524 		nbi->isrouter = ln->ln_router;
   1525 		nbi->expire = ln->ln_expire;
   1526 		splx(s);
   1527 
   1528 		break;
   1529 	    }
   1530 	case SIOCGDEFIFACE_IN6:	/* XXX: should be implemented as a sysctl? */
   1531 		ndif->ifindex = nd6_defifindex;
   1532 		break;
   1533 	case SIOCSDEFIFACE_IN6:	/* XXX: should be implemented as a sysctl? */
   1534 		return(nd6_setdefaultiface(ndif->ifindex));
   1535 		break;
   1536 	}
   1537 	return(error);
   1538 }
   1539 
   1540 /*
   1541  * Create neighbor cache entry and cache link-layer address,
   1542  * on reception of inbound ND6 packets. (RS/RA/NS/redirect)
   1543  */
   1544 struct rtentry *
   1545 nd6_cache_lladdr(ifp, from, lladdr, lladdrlen, type, code)
   1546 	struct ifnet *ifp;
   1547 	struct in6_addr *from;
   1548 	char *lladdr;
   1549 	int lladdrlen;
   1550 	int type;	/* ICMP6 type */
   1551 	int code;	/* type dependent information */
   1552 {
   1553 	struct rtentry *rt = NULL;
   1554 	struct llinfo_nd6 *ln = NULL;
   1555 	int is_newentry;
   1556 	struct sockaddr_dl *sdl = NULL;
   1557 	int do_update;
   1558 	int olladdr;
   1559 	int llchange;
   1560 	int newstate = 0;
   1561 	long time_second = time.tv_sec;
   1562 
   1563 	if (!ifp)
   1564 		panic("ifp == NULL in nd6_cache_lladdr");
   1565 	if (!from)
   1566 		panic("from == NULL in nd6_cache_lladdr");
   1567 
   1568 	/* nothing must be updated for unspecified address */
   1569 	if (IN6_IS_ADDR_UNSPECIFIED(from))
   1570 		return NULL;
   1571 
   1572 	/*
   1573 	 * Validation about ifp->if_addrlen and lladdrlen must be done in
   1574 	 * the caller.
   1575 	 *
   1576 	 * XXX If the link does not have link-layer adderss, what should
   1577 	 * we do? (ifp->if_addrlen == 0)
   1578 	 * Spec says nothing in sections for RA, RS and NA.  There's small
   1579 	 * description on it in NS section (RFC 2461 7.2.3).
   1580 	 */
   1581 
   1582 	rt = nd6_lookup(from, 0, ifp);
   1583 	if (!rt) {
   1584 #if 0
   1585 		/* nothing must be done if there's no lladdr */
   1586 		if (!lladdr || !lladdrlen)
   1587 			return NULL;
   1588 #endif
   1589 
   1590 		rt = nd6_lookup(from, 1, ifp);
   1591 		is_newentry = 1;
   1592 	} else {
   1593 		/* do nothing if static ndp is set */
   1594 		if (rt->rt_flags & RTF_STATIC)
   1595 			return NULL;
   1596 		is_newentry = 0;
   1597 	}
   1598 
   1599 	if (!rt)
   1600 		return NULL;
   1601 	if ((rt->rt_flags & (RTF_GATEWAY | RTF_LLINFO)) != RTF_LLINFO) {
   1602 fail:
   1603 		(void)nd6_free(rt);
   1604 		return NULL;
   1605 	}
   1606 	ln = (struct llinfo_nd6 *)rt->rt_llinfo;
   1607 	if (!ln)
   1608 		goto fail;
   1609 	if (!rt->rt_gateway)
   1610 		goto fail;
   1611 	if (rt->rt_gateway->sa_family != AF_LINK)
   1612 		goto fail;
   1613 	sdl = SDL(rt->rt_gateway);
   1614 
   1615 	olladdr = (sdl->sdl_alen) ? 1 : 0;
   1616 	if (olladdr && lladdr) {
   1617 		if (bcmp(lladdr, LLADDR(sdl), ifp->if_addrlen))
   1618 			llchange = 1;
   1619 		else
   1620 			llchange = 0;
   1621 	} else
   1622 		llchange = 0;
   1623 
   1624 	/*
   1625 	 * newentry olladdr  lladdr  llchange	(*=record)
   1626 	 *	0	n	n	--	(1)
   1627 	 *	0	y	n	--	(2)
   1628 	 *	0	n	y	--	(3) * STALE
   1629 	 *	0	y	y	n	(4) *
   1630 	 *	0	y	y	y	(5) * STALE
   1631 	 *	1	--	n	--	(6)   NOSTATE(= PASSIVE)
   1632 	 *	1	--	y	--	(7) * STALE
   1633 	 */
   1634 
   1635 	if (lladdr) {		/*(3-5) and (7)*/
   1636 		/*
   1637 		 * Record source link-layer address
   1638 		 * XXX is it dependent to ifp->if_type?
   1639 		 */
   1640 		sdl->sdl_alen = ifp->if_addrlen;
   1641 		bcopy(lladdr, LLADDR(sdl), ifp->if_addrlen);
   1642 	}
   1643 
   1644 	if (!is_newentry) {
   1645 		if ((!olladdr && lladdr)		/*(3)*/
   1646 		 || (olladdr && lladdr && llchange)) {	/*(5)*/
   1647 			do_update = 1;
   1648 			newstate = ND6_LLINFO_STALE;
   1649 		} else					/*(1-2,4)*/
   1650 			do_update = 0;
   1651 	} else {
   1652 		do_update = 1;
   1653 		if (!lladdr)				/*(6)*/
   1654 			newstate = ND6_LLINFO_NOSTATE;
   1655 		else					/*(7)*/
   1656 			newstate = ND6_LLINFO_STALE;
   1657 	}
   1658 
   1659 	if (do_update) {
   1660 		/*
   1661 		 * Update the state of the neighbor cache.
   1662 		 */
   1663 		ln->ln_state = newstate;
   1664 
   1665 		if (ln->ln_state == ND6_LLINFO_STALE) {
   1666 			/*
   1667 			 * XXX: since nd6_output() below will cause
   1668 			 * state tansition to DELAY and reset the timer,
   1669 			 * we must set the timer now, although it is actually
   1670 			 * meaningless.
   1671 			 */
   1672 			ln->ln_expire = time_second + nd6_gctimer;
   1673 
   1674 			if (ln->ln_hold) {
   1675 #ifdef OLDIP6OUTPUT
   1676 				ln->ln_asked = 0;
   1677 				ln->ln_state = ND6_LLINFO_DELAY;
   1678 				ln->ln_expire = time_second + nd6_delay;
   1679 				(*ifp->if_output)(ifp, ln->ln_hold,
   1680 						  rt_key(rt), rt);
   1681 #else
   1682 				/*
   1683 				 * we assume ifp is not a p2p here, so just
   1684 				 * set the 2nd argument as the 1st one.
   1685 				 */
   1686 				nd6_output(ifp, ifp, ln->ln_hold,
   1687 					   (struct sockaddr_in6 *)rt_key(rt),
   1688 					   rt);
   1689 #endif
   1690 				ln->ln_hold = NULL;
   1691 			}
   1692 		} else if (ln->ln_state == ND6_LLINFO_INCOMPLETE) {
   1693 			/* probe right away */
   1694 			ln->ln_expire = time_second;
   1695 		}
   1696 	}
   1697 
   1698 	/*
   1699 	 * ICMP6 type dependent behavior.
   1700 	 *
   1701 	 * NS: clear IsRouter if new entry
   1702 	 * RS: clear IsRouter
   1703 	 * RA: set IsRouter if there's lladdr
   1704 	 * redir: clear IsRouter if new entry
   1705 	 *
   1706 	 * RA case, (1):
   1707 	 * The spec says that we must set IsRouter in the following cases:
   1708 	 * - If lladdr exist, set IsRouter.  This means (1-5).
   1709 	 * - If it is old entry (!newentry), set IsRouter.  This means (7).
   1710 	 * So, based on the spec, in (1-5) and (7) cases we must set IsRouter.
   1711 	 * A quetion arises for (1) case.  (1) case has no lladdr in the
   1712 	 * neighbor cache, this is similar to (6).
   1713 	 * This case is rare but we figured that we MUST NOT set IsRouter.
   1714 	 *
   1715 	 * newentry olladdr  lladdr  llchange	    NS  RS  RA	redir
   1716 	 *							D R
   1717 	 *	0	n	n	--	(1)	c   ?     s
   1718 	 *	0	y	n	--	(2)	c   s     s
   1719 	 *	0	n	y	--	(3)	c   s     s
   1720 	 *	0	y	y	n	(4)	c   s     s
   1721 	 *	0	y	y	y	(5)	c   s     s
   1722 	 *	1	--	n	--	(6) c	c 	c s
   1723 	 *	1	--	y	--	(7) c	c   s	c s
   1724 	 *
   1725 	 *					(c=clear s=set)
   1726 	 */
   1727 	switch (type & 0xff) {
   1728 	case ND_NEIGHBOR_SOLICIT:
   1729 		/*
   1730 		 * New entry must have is_router flag cleared.
   1731 		 */
   1732 		if (is_newentry)	/*(6-7)*/
   1733 			ln->ln_router = 0;
   1734 		break;
   1735 	case ND_REDIRECT:
   1736 		/*
   1737 		 * If the icmp is a redirect to a better router, always set the
   1738 		 * is_router flag. Otherwise, if the entry is newly created,
   1739 		 * clear the flag. [RFC 2461, sec 8.3]
   1740 		 */
   1741 		if (code == ND_REDIRECT_ROUTER)
   1742 			ln->ln_router = 1;
   1743 		else if (is_newentry) /*(6-7)*/
   1744 			ln->ln_router = 0;
   1745 		break;
   1746 	case ND_ROUTER_SOLICIT:
   1747 		/*
   1748 		 * is_router flag must always be cleared.
   1749 		 */
   1750 		ln->ln_router = 0;
   1751 		break;
   1752 	case ND_ROUTER_ADVERT:
   1753 		/*
   1754 		 * Mark an entry with lladdr as a router.
   1755 		 */
   1756 		if ((!is_newentry && (olladdr || lladdr))	/*(2-5)*/
   1757 		 || (is_newentry && lladdr)) {			/*(7)*/
   1758 			ln->ln_router = 1;
   1759 		}
   1760 		break;
   1761 	}
   1762 
   1763 	/*
   1764 	 * When the link-layer address of a router changes, select the
   1765 	 * best router again.  In particular, when the neighbor entry is newly
   1766 	 * created, it might affect the selection policy.
   1767 	 * Question: can we restrict the first condition to the "is_newentry"
   1768 	 * case?
   1769 	 * XXX: when we hear an RA from a new router with the link-layer
   1770 	 * address option, defrouter_select() is called twice, since
   1771 	 * defrtrlist_update called the function as well.  However, I believe
   1772 	 * we can compromise the overhead, since it only happens the first
   1773 	 * time.
   1774 	 */
   1775 	if (do_update && ln->ln_router && !ip6_forwarding && ip6_accept_rtadv)
   1776 		defrouter_select();
   1777 
   1778 	return rt;
   1779 }
   1780 
   1781 static void
   1782 nd6_slowtimo(ignored_arg)
   1783     void *ignored_arg;
   1784 {
   1785 	int s = splsoftnet();
   1786 	int i;
   1787 	struct nd_ifinfo *nd6if;
   1788 
   1789 	callout_reset(&nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
   1790 	    nd6_slowtimo, NULL);
   1791 	for (i = 1; i < if_index + 1; i++) {
   1792 		if (!nd_ifinfo || i >= nd_ifinfo_indexlim)
   1793 			continue;
   1794 		nd6if = &nd_ifinfo[i];
   1795 		if (nd6if->basereachable && /* already initialized */
   1796 		    (nd6if->recalctm -= ND6_SLOWTIMER_INTERVAL) <= 0) {
   1797 			/*
   1798 			 * Since reachable time rarely changes by router
   1799 			 * advertisements, we SHOULD insure that a new random
   1800 			 * value gets recomputed at least once every few hours.
   1801 			 * (RFC 2461, 6.3.4)
   1802 			 */
   1803 			nd6if->recalctm = nd6_recalc_reachtm_interval;
   1804 			nd6if->reachable = ND_COMPUTE_RTIME(nd6if->basereachable);
   1805 		}
   1806 	}
   1807 	splx(s);
   1808 }
   1809 
   1810 #define senderr(e) { error = (e); goto bad;}
   1811 int
   1812 nd6_output(ifp, origifp, m0, dst, rt0)
   1813 	struct ifnet *ifp;
   1814 	struct ifnet *origifp;
   1815 	struct mbuf *m0;
   1816 	struct sockaddr_in6 *dst;
   1817 	struct rtentry *rt0;
   1818 {
   1819 	struct mbuf *m = m0;
   1820 	struct rtentry *rt = rt0;
   1821 	struct sockaddr_in6 *gw6 = NULL;
   1822 	struct llinfo_nd6 *ln = NULL;
   1823 	int error = 0;
   1824 	long time_second = time.tv_sec;
   1825 
   1826 	if (IN6_IS_ADDR_MULTICAST(&dst->sin6_addr))
   1827 		goto sendpkt;
   1828 
   1829 	/*
   1830 	 * XXX: we currently do not make neighbor cache on any interface
   1831 	 * other than ARCnet, Ethernet, FDDI and GIF.
   1832 	 *
   1833 	 * RFC2893 says:
   1834 	 * - unidirectional tunnels needs no ND
   1835 	 */
   1836 	switch (ifp->if_type) {
   1837 	case IFT_ARCNET:
   1838 	case IFT_ETHER:
   1839 	case IFT_FDDI:
   1840 	case IFT_IEEE1394:
   1841 	case IFT_GIF:		/* XXX need more cases? */
   1842 		break;
   1843 	default:
   1844 		goto sendpkt;
   1845 	}
   1846 
   1847 	/*
   1848 	 * next hop determination. This routine is derived from ether_outpout.
   1849 	 */
   1850 	if (rt) {
   1851 		if ((rt->rt_flags & RTF_UP) == 0) {
   1852 			if ((rt0 = rt = rtalloc1((struct sockaddr *)dst, 1)) !=
   1853 				NULL)
   1854 			{
   1855 				rt->rt_refcnt--;
   1856 				if (rt->rt_ifp != ifp) {
   1857 					/* XXX: loop care? */
   1858 					return nd6_output(ifp, origifp, m0,
   1859 							  dst, rt);
   1860 				}
   1861 			} else
   1862 				senderr(EHOSTUNREACH);
   1863 		}
   1864 
   1865 		if (rt->rt_flags & RTF_GATEWAY) {
   1866 			gw6 = (struct sockaddr_in6 *)rt->rt_gateway;
   1867 
   1868 			/*
   1869 			 * We skip link-layer address resolution and NUD
   1870 			 * if the gateway is not a neighbor from ND point
   1871 			 * of view, regardless the value of the
   1872 			 * nd_ifinfo.flags.
   1873 			 * The second condition is a bit tricky: we skip
   1874 			 * if the gateway is our own address, which is
   1875 			 * sometimes used to install a route to a p2p link.
   1876 			 */
   1877 			if (!nd6_is_addr_neighbor(gw6, ifp) ||
   1878 			    in6ifa_ifpwithaddr(ifp, &gw6->sin6_addr)) {
   1879 				/*
   1880 				 * We allow this kind of tricky route only
   1881 				 * when the outgoing interface is p2p.
   1882 				 * XXX: we may need a more generic rule here.
   1883 				 */
   1884 				if ((ifp->if_flags & IFF_POINTOPOINT) == 0)
   1885 					senderr(EHOSTUNREACH);
   1886 
   1887 				goto sendpkt;
   1888 			}
   1889 
   1890 			if (rt->rt_gwroute == 0)
   1891 				goto lookup;
   1892 			if (((rt = rt->rt_gwroute)->rt_flags & RTF_UP) == 0) {
   1893 				rtfree(rt); rt = rt0;
   1894 			lookup:
   1895 				rt->rt_gwroute = rtalloc1(rt->rt_gateway, 1);
   1896 				if ((rt = rt->rt_gwroute) == 0)
   1897 					senderr(EHOSTUNREACH);
   1898 				/* the "G" test below also prevents rt == rt0 */
   1899 				if ((rt->rt_flags & RTF_GATEWAY) ||
   1900 				    (rt->rt_ifp != ifp)) {
   1901 					rt->rt_refcnt--;
   1902 					rt0->rt_gwroute = 0;
   1903 					senderr(EHOSTUNREACH);
   1904 				}
   1905 			}
   1906 		}
   1907 	}
   1908 
   1909 	/*
   1910 	 * Address resolution or Neighbor Unreachability Detection
   1911 	 * for the next hop.
   1912 	 * At this point, the destination of the packet must be a unicast
   1913 	 * or an anycast address(i.e. not a multicast).
   1914 	 */
   1915 
   1916 	/* Look up the neighbor cache for the nexthop */
   1917 	if (rt && (rt->rt_flags & RTF_LLINFO) != 0)
   1918 		ln = (struct llinfo_nd6 *)rt->rt_llinfo;
   1919 	else {
   1920 		/*
   1921 		 * Since nd6_is_addr_neighbor() internally calls nd6_lookup(),
   1922 		 * the condition below is not very efficient. But we believe
   1923 		 * it is tolerable, because this should be a rare case.
   1924 		 */
   1925 		if (nd6_is_addr_neighbor(dst, ifp) &&
   1926 		    (rt = nd6_lookup(&dst->sin6_addr, 1, ifp)) != NULL)
   1927 			ln = (struct llinfo_nd6 *)rt->rt_llinfo;
   1928 	}
   1929 	if (!ln || !rt) {
   1930 		if ((ifp->if_flags & IFF_POINTOPOINT) == 0 &&
   1931 		    !(nd_ifinfo[ifp->if_index].flags & ND6_IFF_PERFORMNUD)) {
   1932 			log(LOG_DEBUG,
   1933 			    "nd6_output: can't allocate llinfo for %s "
   1934 			    "(ln=%p, rt=%p)\n",
   1935 			    ip6_sprintf(&dst->sin6_addr), ln, rt);
   1936 			senderr(EIO);	/* XXX: good error? */
   1937 		}
   1938 
   1939 		goto sendpkt;	/* send anyway */
   1940 	}
   1941 
   1942 	/* We don't have to do link-layer address resolution on a p2p link. */
   1943 	if ((ifp->if_flags & IFF_POINTOPOINT) != 0 &&
   1944 	    ln->ln_state < ND6_LLINFO_REACHABLE) {
   1945 		ln->ln_state = ND6_LLINFO_STALE;
   1946 		ln->ln_expire = time_second + nd6_gctimer;
   1947 	}
   1948 
   1949 	/*
   1950 	 * The first time we send a packet to a neighbor whose entry is
   1951 	 * STALE, we have to change the state to DELAY and a sets a timer to
   1952 	 * expire in DELAY_FIRST_PROBE_TIME seconds to ensure do
   1953 	 * neighbor unreachability detection on expiration.
   1954 	 * (RFC 2461 7.3.3)
   1955 	 */
   1956 	if (ln->ln_state == ND6_LLINFO_STALE) {
   1957 		ln->ln_asked = 0;
   1958 		ln->ln_state = ND6_LLINFO_DELAY;
   1959 		ln->ln_expire = time_second + nd6_delay;
   1960 	}
   1961 
   1962 	/*
   1963 	 * If the neighbor cache entry has a state other than INCOMPLETE
   1964 	 * (i.e. its link-layer address is already reloved), just
   1965 	 * send the packet.
   1966 	 */
   1967 	if (ln->ln_state > ND6_LLINFO_INCOMPLETE)
   1968 		goto sendpkt;
   1969 
   1970 	/*
   1971 	 * There is a neighbor cache entry, but no ethernet address
   1972 	 * response yet. Replace the held mbuf (if any) with this
   1973 	 * latest one.
   1974 	 *
   1975 	 * XXX Does the code conform to rate-limiting rule?
   1976 	 * (RFC 2461 7.2.2)
   1977 	 */
   1978 	if (ln->ln_state == ND6_LLINFO_NOSTATE)
   1979 		ln->ln_state = ND6_LLINFO_INCOMPLETE;
   1980 	if (ln->ln_hold)
   1981 		m_freem(ln->ln_hold);
   1982 	ln->ln_hold = m;
   1983 	if (ln->ln_expire) {
   1984 		if (ln->ln_asked < nd6_mmaxtries &&
   1985 		    ln->ln_expire < time_second) {
   1986 			ln->ln_asked++;
   1987 			ln->ln_expire = time_second +
   1988 				nd_ifinfo[ifp->if_index].retrans / 1000;
   1989 			nd6_ns_output(ifp, NULL, &dst->sin6_addr, ln, 0);
   1990 		}
   1991 	}
   1992 	return(0);
   1993 
   1994   sendpkt:
   1995 
   1996 	if ((ifp->if_flags & IFF_LOOPBACK) != 0) {
   1997 		return((*ifp->if_output)(origifp, m, (struct sockaddr *)dst,
   1998 					 rt));
   1999 	}
   2000 	return((*ifp->if_output)(ifp, m, (struct sockaddr *)dst, rt));
   2001 
   2002   bad:
   2003 	if (m)
   2004 		m_freem(m);
   2005 	return (error);
   2006 }
   2007 #undef senderr
   2008 
   2009 int
   2010 nd6_storelladdr(ifp, rt, m, dst, desten)
   2011 	struct ifnet *ifp;
   2012 	struct rtentry *rt;
   2013 	struct mbuf *m;
   2014 	struct sockaddr *dst;
   2015 	u_char *desten;
   2016 {
   2017 	struct sockaddr_dl *sdl;
   2018 
   2019 	if (m->m_flags & M_MCAST) {
   2020 		switch (ifp->if_type) {
   2021 		case IFT_ETHER:
   2022 		case IFT_FDDI:
   2023 			ETHER_MAP_IPV6_MULTICAST(&SIN6(dst)->sin6_addr,
   2024 						 desten);
   2025 			return(1);
   2026 		case IFT_IEEE1394:
   2027 			bcopy(ifp->if_broadcastaddr, desten, ifp->if_addrlen);
   2028 			return(1);
   2029 		case IFT_ARCNET:
   2030 			*desten = 0;
   2031 			return(1);
   2032 		default:
   2033 			m_freem(m);
   2034 			return(0);
   2035 		}
   2036 	}
   2037 
   2038 	if (rt == NULL) {
   2039 		/* this could happen, if we could not allocate memory */
   2040 		m_freem(m);
   2041 		return(0);
   2042 	}
   2043 	if (rt->rt_gateway->sa_family != AF_LINK) {
   2044 		printf("nd6_storelladdr: something odd happens\n");
   2045 		m_freem(m);
   2046 		return(0);
   2047 	}
   2048 	sdl = SDL(rt->rt_gateway);
   2049 	if (sdl->sdl_alen == 0) {
   2050 		/* this should be impossible, but we bark here for debugging */
   2051 		printf("nd6_storelladdr: sdl_alen == 0\n");
   2052 		m_freem(m);
   2053 		return(0);
   2054 	}
   2055 
   2056 	bcopy(LLADDR(sdl), desten, sdl->sdl_alen);
   2057 	return(1);
   2058 }
   2059