Home | History | Annotate | Line # | Download | only in netinet6
nd6_nbr.c revision 1.134
      1 /*	$NetBSD: nd6_nbr.c,v 1.134 2016/12/19 07:51:34 ozaki-r Exp $	*/
      2 /*	$KAME: nd6_nbr.c,v 1.61 2001/02/10 16:06:14 jinmei Exp $	*/
      3 
      4 /*
      5  * Copyright (C) 1995, 1996, 1997, and 1998 WIDE Project.
      6  * All rights reserved.
      7  *
      8  * Redistribution and use in source and binary forms, with or without
      9  * modification, are permitted provided that the following conditions
     10  * are met:
     11  * 1. Redistributions of source code must retain the above copyright
     12  *    notice, this list of conditions and the following disclaimer.
     13  * 2. Redistributions in binary form must reproduce the above copyright
     14  *    notice, this list of conditions and the following disclaimer in the
     15  *    documentation and/or other materials provided with the distribution.
     16  * 3. Neither the name of the project nor the names of its contributors
     17  *    may be used to endorse or promote products derived from this software
     18  *    without specific prior written permission.
     19  *
     20  * THIS SOFTWARE IS PROVIDED BY THE PROJECT AND CONTRIBUTORS ``AS IS'' AND
     21  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
     22  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
     23  * ARE DISCLAIMED.  IN NO EVENT SHALL THE PROJECT OR CONTRIBUTORS BE LIABLE
     24  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
     25  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
     26  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
     27  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
     28  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
     29  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
     30  * SUCH DAMAGE.
     31  */
     32 
     33 #include <sys/cdefs.h>
     34 __KERNEL_RCSID(0, "$NetBSD: nd6_nbr.c,v 1.134 2016/12/19 07:51:34 ozaki-r Exp $");
     35 
     36 #ifdef _KERNEL_OPT
     37 #include "opt_inet.h"
     38 #include "opt_net_mpsafe.h"
     39 #endif
     40 
     41 #include <sys/param.h>
     42 #include <sys/systm.h>
     43 #include <sys/malloc.h>
     44 #include <sys/mbuf.h>
     45 #include <sys/socket.h>
     46 #include <sys/socketvar.h>
     47 #include <sys/sockio.h>
     48 #include <sys/time.h>
     49 #include <sys/kernel.h>
     50 #include <sys/errno.h>
     51 #include <sys/ioctl.h>
     52 #include <sys/syslog.h>
     53 #include <sys/queue.h>
     54 #include <sys/callout.h>
     55 
     56 #include <net/if.h>
     57 #include <net/if_types.h>
     58 #include <net/if_dl.h>
     59 #include <net/route.h>
     60 
     61 #include <netinet/in.h>
     62 #include <netinet/in_var.h>
     63 #include <netinet6/in6_var.h>
     64 #include <netinet6/in6_ifattach.h>
     65 #include <netinet/ip6.h>
     66 #include <netinet6/ip6_var.h>
     67 #include <netinet6/scope6_var.h>
     68 #include <netinet6/nd6.h>
     69 #include <netinet/icmp6.h>
     70 #include <netinet6/icmp6_private.h>
     71 
     72 #include "carp.h"
     73 #if NCARP > 0
     74 #include <netinet/ip_carp.h>
     75 #endif
     76 
     77 #include <net/net_osdep.h>
     78 
     79 struct dadq;
     80 static struct dadq *nd6_dad_find(struct ifaddr *);
     81 static void nd6_dad_starttimer(struct dadq *, int);
     82 static void nd6_dad_stoptimer(struct dadq *);
     83 static void nd6_dad_timer(struct ifaddr *);
     84 static void nd6_dad_ns_output(struct dadq *, struct ifaddr *);
     85 static void nd6_dad_ns_input(struct ifaddr *);
     86 static void nd6_dad_na_input(struct ifaddr *);
     87 static void nd6_dad_duplicated(struct ifaddr *);
     88 
     89 static int dad_ignore_ns = 0;	/* ignore NS in DAD - specwise incorrect*/
     90 static int dad_maxtry = 15;	/* max # of *tries* to transmit DAD packet */
     91 
     92 /*
     93  * Input a Neighbor Solicitation Message.
     94  *
     95  * Based on RFC 2461
     96  * Based on RFC 2462 (duplicate address detection)
     97  */
     98 void
     99 nd6_ns_input(struct mbuf *m, int off, int icmp6len)
    100 {
    101 	struct ifnet *ifp;
    102 	struct ip6_hdr *ip6 = mtod(m, struct ip6_hdr *);
    103 	struct nd_neighbor_solicit *nd_ns;
    104 	struct in6_addr saddr6 = ip6->ip6_src;
    105 	struct in6_addr daddr6 = ip6->ip6_dst;
    106 	struct in6_addr taddr6;
    107 	struct in6_addr myaddr6;
    108 	char *lladdr = NULL;
    109 	struct ifaddr *ifa = NULL;
    110 	int lladdrlen = 0;
    111 	int anycast = 0, proxy = 0, tentative = 0;
    112 	int router = ip6_forwarding;
    113 	int tlladdr;
    114 	union nd_opts ndopts;
    115 	const struct sockaddr_dl *proxydl = NULL;
    116 	struct psref psref;
    117 	struct psref psref_ia;
    118 
    119 	ifp = m_get_rcvif_psref(m, &psref);
    120 	if (ifp == NULL)
    121 		goto freeit;
    122 
    123 	IP6_EXTHDR_GET(nd_ns, struct nd_neighbor_solicit *, m, off, icmp6len);
    124 	if (nd_ns == NULL) {
    125 		ICMP6_STATINC(ICMP6_STAT_TOOSHORT);
    126 		m_put_rcvif_psref(ifp, &psref);
    127 		return;
    128 	}
    129 	ip6 = mtod(m, struct ip6_hdr *); /* adjust pointer for safety */
    130 	taddr6 = nd_ns->nd_ns_target;
    131 	if (in6_setscope(&taddr6, ifp, NULL) != 0)
    132 		goto bad;
    133 
    134 	if (ip6->ip6_hlim != 255) {
    135 		nd6log(LOG_ERR, "invalid hlim (%d) from %s to %s on %s\n",
    136 		    ip6->ip6_hlim, ip6_sprintf(&ip6->ip6_src),
    137 		    ip6_sprintf(&ip6->ip6_dst), if_name(ifp));
    138 		goto bad;
    139 	}
    140 
    141 	if (IN6_IS_ADDR_UNSPECIFIED(&saddr6)) {
    142 		/* dst has to be a solicited node multicast address. */
    143 		/* don't check ifindex portion */
    144 		if (daddr6.s6_addr16[0] == IPV6_ADDR_INT16_MLL &&
    145 		    daddr6.s6_addr32[1] == 0 &&
    146 		    daddr6.s6_addr32[2] == IPV6_ADDR_INT32_ONE &&
    147 		    daddr6.s6_addr8[12] == 0xff) {
    148 			; /* good */
    149 		} else {
    150 			nd6log(LOG_INFO, "bad DAD packet (wrong ip6 dst)\n");
    151 			goto bad;
    152 		}
    153 	} else {
    154 		struct sockaddr_in6 ssin6;
    155 
    156 		/*
    157 		 * Make sure the source address is from a neighbor's address.
    158 		 */
    159 		sockaddr_in6_init(&ssin6, &saddr6, 0, 0, 0);
    160 		if (nd6_is_addr_neighbor(&ssin6, ifp) == 0) {
    161 			nd6log(LOG_INFO,
    162 			    "NS packet from non-neighbor %s on %s\n",
    163 			    ip6_sprintf(&saddr6), if_name(ifp));
    164 			goto bad;
    165 		}
    166 	}
    167 
    168 
    169 	if (IN6_IS_ADDR_MULTICAST(&taddr6)) {
    170 		nd6log(LOG_INFO, "bad NS target (multicast)\n");
    171 		goto bad;
    172 	}
    173 
    174 	icmp6len -= sizeof(*nd_ns);
    175 	nd6_option_init(nd_ns + 1, icmp6len, &ndopts);
    176 	if (nd6_options(&ndopts) < 0) {
    177 		nd6log(LOG_INFO, "invalid ND option, ignored\n");
    178 		/* nd6_options have incremented stats */
    179 		goto freeit;
    180 	}
    181 
    182 	if (ndopts.nd_opts_src_lladdr) {
    183 		lladdr = (char *)(ndopts.nd_opts_src_lladdr + 1);
    184 		lladdrlen = ndopts.nd_opts_src_lladdr->nd_opt_len << 3;
    185 	}
    186 
    187 	if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src) && lladdr) {
    188 		nd6log(LOG_INFO,
    189 		    "bad DAD packet (link-layer address option)\n");
    190 		goto bad;
    191 	}
    192 
    193 	/*
    194 	 * Attaching target link-layer address to the NA?
    195 	 * (RFC 2461 7.2.4)
    196 	 *
    197 	 * NS IP dst is multicast			MUST add
    198 	 * Otherwise					MAY be omitted
    199 	 *
    200 	 * In this implementation, we omit the target link-layer address
    201 	 * in the "MAY" case.
    202 	 */
    203 #if 0 /* too much! */
    204 	ifa = (struct ifaddr *)in6ifa_ifpwithaddr(ifp, &daddr6);
    205 	if (ifa && (((struct in6_ifaddr *)ifa)->ia6_flags & IN6_IFF_ANYCAST))
    206 		tlladdr = 0;
    207 	else
    208 #endif
    209 	if (!IN6_IS_ADDR_MULTICAST(&daddr6))
    210 		tlladdr = 0;
    211 	else
    212 		tlladdr = 1;
    213 
    214 	/*
    215 	 * Target address (taddr6) must be either:
    216 	 * (1) Valid unicast/anycast address for my receiving interface,
    217 	 * (2) Unicast address for which I'm offering proxy service, or
    218 	 * (3) "tentative" address on which DAD is being performed.
    219 	 */
    220 	/* (1) and (3) check. */
    221 #if NCARP > 0
    222 	if (ifp->if_carp && ifp->if_type != IFT_CARP) {
    223 		int s = pserialize_read_enter();
    224 		ifa = carp_iamatch6(ifp->if_carp, &taddr6);
    225 		if (ifa != NULL)
    226 			ifa_acquire(ifa, &psref_ia);
    227 		pserialize_read_exit(s);
    228 	} else
    229 		ifa = NULL;
    230 	if (!ifa)
    231 		ifa = (struct ifaddr *)in6ifa_ifpwithaddr_psref(ifp, &taddr6,
    232 		    &psref_ia);
    233 #else
    234 	ifa = (struct ifaddr *)in6ifa_ifpwithaddr_psref(ifp, &taddr6,
    235 	    &psref_ia);
    236 #endif
    237 
    238 	/* (2) check. */
    239 	if (ifa == NULL) {
    240 		struct rtentry *rt;
    241 		struct sockaddr_in6 tsin6;
    242 
    243 		sockaddr_in6_init(&tsin6, &taddr6, 0, 0, 0);
    244 
    245 		rt = rtalloc1(sin6tosa(&tsin6), 0);
    246 		if (rt && (rt->rt_flags & RTF_ANNOUNCE) != 0 &&
    247 		    rt->rt_gateway->sa_family == AF_LINK) {
    248 			/*
    249 			 * proxy NDP for single entry
    250 			 */
    251 			ifa = (struct ifaddr *)in6ifa_ifpforlinklocal_psref(ifp,
    252 				IN6_IFF_NOTREADY|IN6_IFF_ANYCAST, &psref_ia);
    253 			if (ifa) {
    254 				proxy = 1;
    255 				proxydl = satocsdl(rt->rt_gateway);
    256 				router = 0;	/* XXX */
    257 			}
    258 		}
    259 		if (rt)
    260 			rt_unref(rt);
    261 	}
    262 	if (ifa == NULL) {
    263 		/*
    264 		 * We've got an NS packet, and we don't have that address
    265 		 * assigned for us.  We MUST silently ignore it.
    266 		 * See RFC2461 7.2.3.
    267 		 */
    268 		goto freeit;
    269 	}
    270 	myaddr6 = *IFA_IN6(ifa);
    271 	anycast = ((struct in6_ifaddr *)ifa)->ia6_flags & IN6_IFF_ANYCAST;
    272 	tentative = ((struct in6_ifaddr *)ifa)->ia6_flags & IN6_IFF_TENTATIVE;
    273 	if (((struct in6_ifaddr *)ifa)->ia6_flags & IN6_IFF_DUPLICATED)
    274 		goto freeit;
    275 
    276 	if (lladdr && ((ifp->if_addrlen + 2 + 7) & ~7) != lladdrlen) {
    277 		nd6log(LOG_INFO, "lladdrlen mismatch for %s "
    278 		    "(if %d, NS packet %d)\n",
    279 		    ip6_sprintf(&taddr6), ifp->if_addrlen, lladdrlen - 2);
    280 		goto bad;
    281 	}
    282 
    283 	if (IN6_ARE_ADDR_EQUAL(&myaddr6, &saddr6)) {
    284 		nd6log(LOG_INFO, "duplicate IP6 address %s\n",
    285 		    ip6_sprintf(&saddr6));
    286 		goto freeit;
    287 	}
    288 
    289 	/*
    290 	 * We have neighbor solicitation packet, with target address equals to
    291 	 * one of my tentative address.
    292 	 *
    293 	 * src addr	how to process?
    294 	 * ---		---
    295 	 * multicast	of course, invalid (rejected in ip6_input)
    296 	 * unicast	somebody is doing address resolution -> ignore
    297 	 * unspec	dup address detection
    298 	 *
    299 	 * The processing is defined in RFC 2462.
    300 	 */
    301 	if (tentative) {
    302 		/*
    303 		 * If source address is unspecified address, it is for
    304 		 * duplicate address detection.
    305 		 *
    306 		 * If not, the packet is for addess resolution;
    307 		 * silently ignore it.
    308 		 */
    309 		if (IN6_IS_ADDR_UNSPECIFIED(&saddr6))
    310 			nd6_dad_ns_input(ifa);
    311 		ifa_release(ifa, &psref_ia);
    312 		ifa = NULL;
    313 
    314 		goto freeit;
    315 	}
    316 	ifa_release(ifa, &psref_ia);
    317 	ifa = NULL;
    318 
    319 	/*
    320 	 * If the source address is unspecified address, entries must not
    321 	 * be created or updated.
    322 	 * It looks that sender is performing DAD.  Output NA toward
    323 	 * all-node multicast address, to tell the sender that I'm using
    324 	 * the address.
    325 	 * S bit ("solicited") must be zero.
    326 	 */
    327 	if (IN6_IS_ADDR_UNSPECIFIED(&saddr6)) {
    328 		struct in6_addr in6_all;
    329 
    330 		in6_all = in6addr_linklocal_allnodes;
    331 		if (in6_setscope(&in6_all, ifp, NULL) != 0)
    332 			goto bad;
    333 		nd6_na_output(ifp, &in6_all, &taddr6,
    334 		    ((anycast || proxy || !tlladdr) ? 0 : ND_NA_FLAG_OVERRIDE) |
    335 		    (ip6_forwarding ? ND_NA_FLAG_ROUTER : 0),
    336 		    tlladdr, (const struct sockaddr *)proxydl);
    337 		goto freeit;
    338 	}
    339 
    340 	nd6_cache_lladdr(ifp, &saddr6, lladdr, lladdrlen, ND_NEIGHBOR_SOLICIT, 0);
    341 
    342 	nd6_na_output(ifp, &saddr6, &taddr6,
    343 	    ((anycast || proxy || !tlladdr) ? 0 : ND_NA_FLAG_OVERRIDE) |
    344 	    (router ? ND_NA_FLAG_ROUTER : 0) | ND_NA_FLAG_SOLICITED,
    345 	    tlladdr, (const struct sockaddr *)proxydl);
    346  freeit:
    347 	ifa_release(ifa, &psref_ia);
    348 	m_put_rcvif_psref(ifp, &psref);
    349 	m_freem(m);
    350 	return;
    351 
    352  bad:
    353 	nd6log(LOG_ERR, "src=%s\n", ip6_sprintf(&saddr6));
    354 	nd6log(LOG_ERR, "dst=%s\n", ip6_sprintf(&daddr6));
    355 	nd6log(LOG_ERR, "tgt=%s\n", ip6_sprintf(&taddr6));
    356 	ICMP6_STATINC(ICMP6_STAT_BADNS);
    357 	ifa_release(ifa, &psref_ia);
    358 	m_put_rcvif_psref(ifp, &psref);
    359 	m_freem(m);
    360 }
    361 
    362 /*
    363  * Output a Neighbor Solicitation Message. Caller specifies:
    364  *	- ICMP6 header source IP6 address
    365  *	- ND6 header target IP6 address
    366  *	- ND6 header source datalink address
    367  *
    368  * Based on RFC 2461
    369  * Based on RFC 2462 (duplicate address detection)
    370  */
    371 void
    372 nd6_ns_output(struct ifnet *ifp, const struct in6_addr *daddr6,
    373     const struct in6_addr *taddr6,
    374     struct in6_addr *hsrc,
    375     int dad			/* duplicate address detection */)
    376 {
    377 	struct mbuf *m;
    378 	struct ip6_hdr *ip6;
    379 	struct nd_neighbor_solicit *nd_ns;
    380 	struct in6_addr *src, src_in;
    381 	struct ip6_moptions im6o;
    382 	int icmp6len;
    383 	int maxlen;
    384 	const void *mac;
    385 	struct route ro;
    386 
    387 	if (IN6_IS_ADDR_MULTICAST(taddr6))
    388 		return;
    389 
    390 	memset(&ro, 0, sizeof(ro));
    391 
    392 	/* estimate the size of message */
    393 	maxlen = sizeof(*ip6) + sizeof(*nd_ns);
    394 	maxlen += (sizeof(struct nd_opt_hdr) + ifp->if_addrlen + 7) & ~7;
    395 #ifdef DIAGNOSTIC
    396 	if (max_linkhdr + maxlen >= MCLBYTES) {
    397 		printf("nd6_ns_output: max_linkhdr + maxlen >= MCLBYTES "
    398 		    "(%d + %d > %d)\n", max_linkhdr, maxlen, MCLBYTES);
    399 		panic("nd6_ns_output: insufficient MCLBYTES");
    400 		/* NOTREACHED */
    401 	}
    402 #endif
    403 
    404 	MGETHDR(m, M_DONTWAIT, MT_DATA);
    405 	if (m && max_linkhdr + maxlen >= MHLEN) {
    406 		MCLGET(m, M_DONTWAIT);
    407 		if ((m->m_flags & M_EXT) == 0) {
    408 			m_free(m);
    409 			m = NULL;
    410 		}
    411 	}
    412 	if (m == NULL)
    413 		return;
    414 	m_reset_rcvif(m);
    415 
    416 	if (daddr6 == NULL || IN6_IS_ADDR_MULTICAST(daddr6)) {
    417 		m->m_flags |= M_MCAST;
    418 		im6o.im6o_multicast_if_index = if_get_index(ifp);
    419 		im6o.im6o_multicast_hlim = 255;
    420 		im6o.im6o_multicast_loop = 0;
    421 	}
    422 
    423 	icmp6len = sizeof(*nd_ns);
    424 	m->m_pkthdr.len = m->m_len = sizeof(*ip6) + icmp6len;
    425 	m->m_data += max_linkhdr;	/* or MH_ALIGN() equivalent? */
    426 
    427 	/* fill neighbor solicitation packet */
    428 	ip6 = mtod(m, struct ip6_hdr *);
    429 	ip6->ip6_flow = 0;
    430 	ip6->ip6_vfc &= ~IPV6_VERSION_MASK;
    431 	ip6->ip6_vfc |= IPV6_VERSION;
    432 	/* ip6->ip6_plen will be set later */
    433 	ip6->ip6_nxt = IPPROTO_ICMPV6;
    434 	ip6->ip6_hlim = 255;
    435 	if (daddr6)
    436 		ip6->ip6_dst = *daddr6;
    437 	else {
    438 		ip6->ip6_dst.s6_addr16[0] = IPV6_ADDR_INT16_MLL;
    439 		ip6->ip6_dst.s6_addr16[1] = 0;
    440 		ip6->ip6_dst.s6_addr32[1] = 0;
    441 		ip6->ip6_dst.s6_addr32[2] = IPV6_ADDR_INT32_ONE;
    442 		ip6->ip6_dst.s6_addr32[3] = taddr6->s6_addr32[3];
    443 		ip6->ip6_dst.s6_addr8[12] = 0xff;
    444 		if (in6_setscope(&ip6->ip6_dst, ifp, NULL) != 0)
    445 			goto bad;
    446 	}
    447 	if (!dad) {
    448 		int s;
    449 		/*
    450 		 * RFC2461 7.2.2:
    451 		 * "If the source address of the packet prompting the
    452 		 * solicitation is the same as one of the addresses assigned
    453 		 * to the outgoing interface, that address SHOULD be placed
    454 		 * in the IP Source Address of the outgoing solicitation.
    455 		 * Otherwise, any one of the addresses assigned to the
    456 		 * interface should be used."
    457 		 *
    458 		 * We use the source address for the prompting packet
    459 		 * (hsrc), if:
    460 		 * - hsrc is given from the caller (by giving "ln"), and
    461 		 * - hsrc belongs to the outgoing interface.
    462 		 * Otherwise, we perform the source address selection as usual.
    463 		 */
    464 		s = pserialize_read_enter();
    465 		if (hsrc && in6ifa_ifpwithaddr(ifp, hsrc))
    466 			src = hsrc;
    467 		else {
    468 			int error;
    469 			struct sockaddr_in6 dst_sa;
    470 
    471 			sockaddr_in6_init(&dst_sa, &ip6->ip6_dst, 0, 0, 0);
    472 
    473 			error = in6_selectsrc(&dst_sa, NULL,
    474 			    NULL, &ro, NULL, NULL, NULL, &src_in);
    475 			if (error != 0) {
    476 				nd6log(LOG_DEBUG, "source can't be "
    477 				    "determined: dst=%s, error=%d\n",
    478 				    ip6_sprintf(&dst_sa.sin6_addr), error);
    479 				pserialize_read_exit(s);
    480 				goto bad;
    481 			}
    482 			src = &src_in;
    483 		}
    484 		pserialize_read_exit(s);
    485 	} else {
    486 		/*
    487 		 * Source address for DAD packet must always be IPv6
    488 		 * unspecified address. (0::0)
    489 		 * We actually don't have to 0-clear the address (we did it
    490 		 * above), but we do so here explicitly to make the intention
    491 		 * clearer.
    492 		 */
    493 		memset(&src_in, 0, sizeof(src_in));
    494 		src = &src_in;
    495 	}
    496 	ip6->ip6_src = *src;
    497 	nd_ns = (struct nd_neighbor_solicit *)(ip6 + 1);
    498 	nd_ns->nd_ns_type = ND_NEIGHBOR_SOLICIT;
    499 	nd_ns->nd_ns_code = 0;
    500 	nd_ns->nd_ns_reserved = 0;
    501 	nd_ns->nd_ns_target = *taddr6;
    502 	in6_clearscope(&nd_ns->nd_ns_target); /* XXX */
    503 
    504 	/*
    505 	 * Add source link-layer address option.
    506 	 *
    507 	 *				spec		implementation
    508 	 *				---		---
    509 	 * DAD packet			MUST NOT	do not add the option
    510 	 * there's no link layer address:
    511 	 *				impossible	do not add the option
    512 	 * there's link layer address:
    513 	 *	Multicast NS		MUST add one	add the option
    514 	 *	Unicast NS		SHOULD add one	add the option
    515 	 */
    516 	if (!dad && (mac = nd6_ifptomac(ifp))) {
    517 		int optlen = sizeof(struct nd_opt_hdr) + ifp->if_addrlen;
    518 		struct nd_opt_hdr *nd_opt = (struct nd_opt_hdr *)(nd_ns + 1);
    519 		/* 8 byte alignments... */
    520 		optlen = (optlen + 7) & ~7;
    521 
    522 		m->m_pkthdr.len += optlen;
    523 		m->m_len += optlen;
    524 		icmp6len += optlen;
    525 		memset((void *)nd_opt, 0, optlen);
    526 		nd_opt->nd_opt_type = ND_OPT_SOURCE_LINKADDR;
    527 		nd_opt->nd_opt_len = optlen >> 3;
    528 		memcpy((void *)(nd_opt + 1), mac, ifp->if_addrlen);
    529 	}
    530 
    531 	ip6->ip6_plen = htons((u_int16_t)icmp6len);
    532 	nd_ns->nd_ns_cksum = 0;
    533 	nd_ns->nd_ns_cksum =
    534 	    in6_cksum(m, IPPROTO_ICMPV6, sizeof(*ip6), icmp6len);
    535 
    536 	ip6_output(m, NULL, &ro, dad ? IPV6_UNSPECSRC : 0, &im6o, NULL, NULL);
    537 	icmp6_ifstat_inc(ifp, ifs6_out_msg);
    538 	icmp6_ifstat_inc(ifp, ifs6_out_neighborsolicit);
    539 	ICMP6_STATINC(ICMP6_STAT_OUTHIST + ND_NEIGHBOR_SOLICIT);
    540 
    541 	rtcache_free(&ro);
    542 	return;
    543 
    544   bad:
    545 	rtcache_free(&ro);
    546 	m_freem(m);
    547 	return;
    548 }
    549 
    550 /*
    551  * Neighbor advertisement input handling.
    552  *
    553  * Based on RFC 2461
    554  * Based on RFC 2462 (duplicate address detection)
    555  *
    556  * the following items are not implemented yet:
    557  * - proxy advertisement delay rule (RFC2461 7.2.8, last paragraph, SHOULD)
    558  * - anycast advertisement delay rule (RFC2461 7.2.7, SHOULD)
    559  */
    560 void
    561 nd6_na_input(struct mbuf *m, int off, int icmp6len)
    562 {
    563 	struct ifnet *ifp;
    564 	struct ip6_hdr *ip6 = mtod(m, struct ip6_hdr *);
    565 	struct nd_neighbor_advert *nd_na;
    566 	struct in6_addr saddr6 = ip6->ip6_src;
    567 	struct in6_addr daddr6 = ip6->ip6_dst;
    568 	struct in6_addr taddr6;
    569 	int flags;
    570 	int is_router;
    571 	int is_solicited;
    572 	int is_override;
    573 	char *lladdr = NULL;
    574 	int lladdrlen = 0;
    575 	struct ifaddr *ifa;
    576 	struct llentry *ln = NULL;
    577 	union nd_opts ndopts;
    578 	struct sockaddr_in6 ssin6;
    579 	int rt_announce;
    580 	bool checklink = false;
    581 	struct psref psref;
    582 	struct psref psref_ia;
    583 
    584 	ifp = m_get_rcvif_psref(m, &psref);
    585 	if (ifp == NULL)
    586 		goto freeit;
    587 
    588 	if (ip6->ip6_hlim != 255) {
    589 		nd6log(LOG_ERR,
    590 		    "invalid hlim (%d) from %s to %s on %s\n",
    591 		    ip6->ip6_hlim, ip6_sprintf(&ip6->ip6_src),
    592 		    ip6_sprintf(&ip6->ip6_dst), if_name(ifp));
    593 		goto bad;
    594 	}
    595 
    596 	IP6_EXTHDR_GET(nd_na, struct nd_neighbor_advert *, m, off, icmp6len);
    597 	if (nd_na == NULL) {
    598 		m_put_rcvif_psref(ifp, &psref);
    599 		ICMP6_STATINC(ICMP6_STAT_TOOSHORT);
    600 		return;
    601 	}
    602 
    603 	flags = nd_na->nd_na_flags_reserved;
    604 	is_router = ((flags & ND_NA_FLAG_ROUTER) != 0);
    605 	is_solicited = ((flags & ND_NA_FLAG_SOLICITED) != 0);
    606 	is_override = ((flags & ND_NA_FLAG_OVERRIDE) != 0);
    607 
    608 	taddr6 = nd_na->nd_na_target;
    609 	if (in6_setscope(&taddr6, ifp, NULL)) {
    610 		m_put_rcvif_psref(ifp, &psref);
    611 		return;		/* XXX: impossible */
    612 	}
    613 
    614 	if (IN6_IS_ADDR_MULTICAST(&taddr6)) {
    615 		nd6log(LOG_ERR, "invalid target address %s\n",
    616 		    ip6_sprintf(&taddr6));
    617 		goto bad;
    618 	}
    619 	if (is_solicited && IN6_IS_ADDR_MULTICAST(&daddr6)) {
    620 		nd6log(LOG_ERR, "a solicited adv is multicasted\n");
    621 		goto bad;
    622 	}
    623 
    624 	icmp6len -= sizeof(*nd_na);
    625 	nd6_option_init(nd_na + 1, icmp6len, &ndopts);
    626 	if (nd6_options(&ndopts) < 0) {
    627 		nd6log(LOG_INFO, "invalid ND option, ignored\n");
    628 		/* nd6_options have incremented stats */
    629 		goto freeit;
    630 	}
    631 
    632 	if (ndopts.nd_opts_tgt_lladdr) {
    633 		lladdr = (char *)(ndopts.nd_opts_tgt_lladdr + 1);
    634 		lladdrlen = ndopts.nd_opts_tgt_lladdr->nd_opt_len << 3;
    635 	}
    636 
    637 	ifa = (struct ifaddr *)in6ifa_ifpwithaddr_psref(ifp, &taddr6, &psref_ia);
    638 
    639 	/*
    640 	 * Target address matches one of my interface address.
    641 	 *
    642 	 * If my address is tentative, this means that there's somebody
    643 	 * already using the same address as mine.  This indicates DAD failure.
    644 	 * This is defined in RFC 2462.
    645 	 *
    646 	 * Otherwise, process as defined in RFC 2461.
    647 	 */
    648 	if (ifa
    649 	 && (((struct in6_ifaddr *)ifa)->ia6_flags & IN6_IFF_TENTATIVE)) {
    650 		nd6_dad_na_input(ifa);
    651 		ifa_release(ifa, &psref_ia);
    652 		ifa = NULL;
    653 		goto freeit;
    654 	}
    655 
    656 	/* Just for safety, maybe unnecessary. */
    657 	if (ifa) {
    658 		log(LOG_ERR,
    659 		    "nd6_na_input: duplicate IP6 address %s\n",
    660 		    ip6_sprintf(&taddr6));
    661 		ifa_release(ifa, &psref_ia);
    662 		ifa = NULL;
    663 		goto freeit;
    664 	}
    665 
    666 	/*
    667 	 * Make sure the source address is from a neighbor's address.
    668 	 */
    669 	sockaddr_in6_init(&ssin6, &saddr6, 0, 0, 0);
    670 	if (nd6_is_addr_neighbor(&ssin6, ifp) == 0) {
    671 		nd6log(LOG_INFO, "ND packet from non-neighbor %s on %s\n",
    672 		    ip6_sprintf(&saddr6), if_name(ifp));
    673 		goto bad;
    674 	}
    675 
    676 	if (lladdr && ((ifp->if_addrlen + 2 + 7) & ~7) != lladdrlen) {
    677 		nd6log(LOG_INFO, "lladdrlen mismatch for %s "
    678 		    "(if %d, NA packet %d)\n", ip6_sprintf(&taddr6),
    679 		    ifp->if_addrlen, lladdrlen - 2);
    680 		goto bad;
    681 	}
    682 
    683 	/*
    684 	 * If no neighbor cache entry is found, NA SHOULD silently be
    685 	 * discarded.
    686 	 */
    687 	ln = nd6_lookup(&taddr6, ifp, true);
    688 	if (ln == NULL)
    689 		goto freeit;
    690 
    691 	rt_announce = 0;
    692 	if (ln->ln_state == ND6_LLINFO_INCOMPLETE) {
    693 		/*
    694 		 * If the link-layer has address, and no lladdr option came,
    695 		 * discard the packet.
    696 		 */
    697 		if (ifp->if_addrlen && !lladdr)
    698 			goto freeit;
    699 
    700 		/*
    701 		 * Record link-layer address, and update the state.
    702 		 */
    703 		memcpy(&ln->ll_addr, lladdr, ifp->if_addrlen);
    704 		ln->la_flags |= LLE_VALID;
    705 		rt_announce = 1;
    706 		if (is_solicited) {
    707 			ln->ln_state = ND6_LLINFO_REACHABLE;
    708 			ln->ln_byhint = 0;
    709 			if (!ND6_LLINFO_PERMANENT(ln)) {
    710 				nd6_llinfo_settimer(ln,
    711 				    ND_IFINFO(ln->lle_tbl->llt_ifp)->reachable * hz);
    712 			}
    713 		} else {
    714 			ln->ln_state = ND6_LLINFO_STALE;
    715 			nd6_llinfo_settimer(ln, nd6_gctimer * hz);
    716 		}
    717 		if ((ln->ln_router = is_router) != 0) {
    718 			/*
    719 			 * This means a router's state has changed from
    720 			 * non-reachable to probably reachable, and might
    721 			 * affect the status of associated prefixes..
    722 			 */
    723 			checklink = true;
    724 		}
    725 	} else {
    726 		int llchange;
    727 
    728 		/*
    729 		 * Check if the link-layer address has changed or not.
    730 		 */
    731 		if (lladdr == NULL)
    732 			llchange = 0;
    733 		else {
    734 			if (ln->la_flags & LLE_VALID) {
    735 				if (memcmp(lladdr, &ln->ll_addr, ifp->if_addrlen))
    736 					llchange = rt_announce = 1;
    737 				else
    738 					llchange = 0;
    739 			} else
    740 				llchange = rt_announce = 1;
    741 		}
    742 
    743 		/*
    744 		 * This is VERY complex.  Look at it with care.
    745 		 *
    746 		 * override solicit lladdr llchange	action
    747 		 *					(L: record lladdr)
    748 		 *
    749 		 *	0	0	n	--	(2c)
    750 		 *	0	0	y	n	(2b) L
    751 		 *	0	0	y	y	(1)    REACHABLE->STALE
    752 		 *	0	1	n	--	(2c)   *->REACHABLE
    753 		 *	0	1	y	n	(2b) L *->REACHABLE
    754 		 *	0	1	y	y	(1)    REACHABLE->STALE
    755 		 *	1	0	n	--	(2a)
    756 		 *	1	0	y	n	(2a) L
    757 		 *	1	0	y	y	(2a) L *->STALE
    758 		 *	1	1	n	--	(2a)   *->REACHABLE
    759 		 *	1	1	y	n	(2a) L *->REACHABLE
    760 		 *	1	1	y	y	(2a) L *->REACHABLE
    761 		 */
    762 		if (!is_override && lladdr != NULL && llchange) { /* (1) */
    763 			/*
    764 			 * If state is REACHABLE, make it STALE.
    765 			 * no other updates should be done.
    766 			 */
    767 			if (ln->ln_state == ND6_LLINFO_REACHABLE) {
    768 				ln->ln_state = ND6_LLINFO_STALE;
    769 				nd6_llinfo_settimer(ln, nd6_gctimer * hz);
    770 			}
    771 			goto freeit;
    772 		} else if (is_override				   /* (2a) */
    773 		    || (!is_override && lladdr != NULL && !llchange) /* (2b) */
    774 		    || lladdr == NULL) {			   /* (2c) */
    775 			/*
    776 			 * Update link-local address, if any.
    777 			 */
    778 			if (lladdr != NULL) {
    779 				memcpy(&ln->ll_addr, lladdr, ifp->if_addrlen);
    780 				ln->la_flags |= LLE_VALID;
    781 			}
    782 
    783 			/*
    784 			 * If solicited, make the state REACHABLE.
    785 			 * If not solicited and the link-layer address was
    786 			 * changed, make it STALE.
    787 			 */
    788 			if (is_solicited) {
    789 				ln->ln_state = ND6_LLINFO_REACHABLE;
    790 				ln->ln_byhint = 0;
    791 				if (!ND6_LLINFO_PERMANENT(ln)) {
    792 					nd6_llinfo_settimer(ln,
    793 					    ND_IFINFO(ifp)->reachable * hz);
    794 				}
    795 			} else {
    796 				if (lladdr && llchange) {
    797 					ln->ln_state = ND6_LLINFO_STALE;
    798 					nd6_llinfo_settimer(ln,
    799 					    nd6_gctimer * hz);
    800 				}
    801 			}
    802 		}
    803 
    804 		if (ln->ln_router && !is_router) {
    805 			/*
    806 			 * The peer dropped the router flag.
    807 			 * Remove the sender from the Default Router List and
    808 			 * update the Destination Cache entries.
    809 			 */
    810 			struct nd_defrouter *dr;
    811 			const struct in6_addr *in6;
    812 
    813 			in6 = &ln->r_l3addr.addr6;
    814 
    815 			ND6_WLOCK();
    816 			dr = nd6_defrouter_lookup(in6, ln->lle_tbl->llt_ifp);
    817 			if (dr)
    818 				nd6_defrtrlist_del(dr, NULL);
    819 			else if (!ip6_forwarding) {
    820 				/*
    821 				 * Even if the neighbor is not in the default
    822 				 * router list, the neighbor may be used
    823 				 * as a next hop for some destinations
    824 				 * (e.g. redirect case). So we must
    825 				 * call nd6_rt_flush explicitly.
    826 				 */
    827 				nd6_rt_flush(&ip6->ip6_src, ln->lle_tbl->llt_ifp);
    828 			}
    829 			ND6_UNLOCK();
    830 		}
    831 		ln->ln_router = is_router;
    832 	}
    833         /*
    834 	 * XXX: does this matter?
    835 	 * rt->rt_flags &= ~RTF_REJECT;
    836 	 */
    837 	ln->ln_asked = 0;
    838 	nd6_llinfo_release_pkts(ln, ifp);
    839 	/* FIXME */
    840 #if 0
    841 	if (rt_announce) /* tell user process about any new lladdr */
    842 		rt_newmsg(RTM_CHANGE, rt);
    843 #endif
    844 
    845  freeit:
    846 	if (ln != NULL)
    847 		LLE_WUNLOCK(ln);
    848 
    849 	if (checklink) {
    850 		ND6_WLOCK();
    851 		nd6_pfxlist_onlink_check();
    852 		ND6_UNLOCK();
    853 	}
    854 
    855 	m_put_rcvif_psref(ifp, &psref);
    856 	m_freem(m);
    857 	return;
    858 
    859  bad:
    860 	if (ln != NULL)
    861 		LLE_WUNLOCK(ln);
    862 
    863 	ICMP6_STATINC(ICMP6_STAT_BADNA);
    864 	m_put_rcvif_psref(ifp, &psref);
    865 	m_freem(m);
    866 }
    867 
    868 /*
    869  * Neighbor advertisement output handling.
    870  *
    871  * Based on RFC 2461
    872  *
    873  * the following items are not implemented yet:
    874  * - proxy advertisement delay rule (RFC2461 7.2.8, last paragraph, SHOULD)
    875  * - anycast advertisement delay rule (RFC2461 7.2.7, SHOULD)
    876  */
    877 void
    878 nd6_na_output(
    879 	struct ifnet *ifp,
    880 	const struct in6_addr *daddr6_0,
    881 	const struct in6_addr *taddr6,
    882 	u_long flags,
    883 	int tlladdr,		/* 1 if include target link-layer address */
    884 	const struct sockaddr *sdl0)	/* sockaddr_dl (= proxy NA) or NULL */
    885 {
    886 	struct mbuf *m;
    887 	struct ip6_hdr *ip6;
    888 	struct nd_neighbor_advert *nd_na;
    889 	struct ip6_moptions im6o;
    890 	struct sockaddr *dst;
    891 	union {
    892 		struct sockaddr		dst;
    893 		struct sockaddr_in6	dst6;
    894 	} u;
    895 	struct in6_addr daddr6;
    896 	int icmp6len, maxlen, error;
    897 	const void *mac;
    898 	struct route ro;
    899 
    900 	mac = NULL;
    901 	memset(&ro, 0, sizeof(ro));
    902 
    903 	daddr6 = *daddr6_0;	/* make a local copy for modification */
    904 
    905 	/* estimate the size of message */
    906 	maxlen = sizeof(*ip6) + sizeof(*nd_na);
    907 	maxlen += (sizeof(struct nd_opt_hdr) + ifp->if_addrlen + 7) & ~7;
    908 #ifdef DIAGNOSTIC
    909 	if (max_linkhdr + maxlen >= MCLBYTES) {
    910 		printf("nd6_na_output: max_linkhdr + maxlen >= MCLBYTES "
    911 		    "(%d + %d > %d)\n", max_linkhdr, maxlen, MCLBYTES);
    912 		panic("nd6_na_output: insufficient MCLBYTES");
    913 		/* NOTREACHED */
    914 	}
    915 #endif
    916 
    917 	MGETHDR(m, M_DONTWAIT, MT_DATA);
    918 	if (m && max_linkhdr + maxlen >= MHLEN) {
    919 		MCLGET(m, M_DONTWAIT);
    920 		if ((m->m_flags & M_EXT) == 0) {
    921 			m_free(m);
    922 			m = NULL;
    923 		}
    924 	}
    925 	if (m == NULL)
    926 		return;
    927 	m_reset_rcvif(m);
    928 
    929 	if (IN6_IS_ADDR_MULTICAST(&daddr6)) {
    930 		m->m_flags |= M_MCAST;
    931 		im6o.im6o_multicast_if_index = if_get_index(ifp);
    932 		im6o.im6o_multicast_hlim = 255;
    933 		im6o.im6o_multicast_loop = 0;
    934 	}
    935 
    936 	icmp6len = sizeof(*nd_na);
    937 	m->m_pkthdr.len = m->m_len = sizeof(struct ip6_hdr) + icmp6len;
    938 	m->m_data += max_linkhdr;	/* or MH_ALIGN() equivalent? */
    939 
    940 	/* fill neighbor advertisement packet */
    941 	ip6 = mtod(m, struct ip6_hdr *);
    942 	ip6->ip6_flow = 0;
    943 	ip6->ip6_vfc &= ~IPV6_VERSION_MASK;
    944 	ip6->ip6_vfc |= IPV6_VERSION;
    945 	ip6->ip6_nxt = IPPROTO_ICMPV6;
    946 	ip6->ip6_hlim = 255;
    947 	if (IN6_IS_ADDR_UNSPECIFIED(&daddr6)) {
    948 		/* reply to DAD */
    949 		daddr6.s6_addr16[0] = IPV6_ADDR_INT16_MLL;
    950 		daddr6.s6_addr16[1] = 0;
    951 		daddr6.s6_addr32[1] = 0;
    952 		daddr6.s6_addr32[2] = 0;
    953 		daddr6.s6_addr32[3] = IPV6_ADDR_INT32_ONE;
    954 		if (in6_setscope(&daddr6, ifp, NULL))
    955 			goto bad;
    956 
    957 		flags &= ~ND_NA_FLAG_SOLICITED;
    958 	}
    959 	ip6->ip6_dst = daddr6;
    960 	sockaddr_in6_init(&u.dst6, &daddr6, 0, 0, 0);
    961 	dst = &u.dst;
    962 	if (rtcache_setdst(&ro, dst) != 0)
    963 		goto bad;
    964 
    965 	/*
    966 	 * Select a source whose scope is the same as that of the dest.
    967 	 */
    968 	error = in6_selectsrc(satosin6(dst), NULL, NULL, &ro, NULL, NULL, NULL,
    969 	    &ip6->ip6_src);
    970 	if (error != 0) {
    971 		nd6log(LOG_DEBUG, "source can't be "
    972 		    "determined: dst=%s, error=%d\n",
    973 		    ip6_sprintf(&satocsin6(dst)->sin6_addr), error);
    974 		goto bad;
    975 	}
    976 	nd_na = (struct nd_neighbor_advert *)(ip6 + 1);
    977 	nd_na->nd_na_type = ND_NEIGHBOR_ADVERT;
    978 	nd_na->nd_na_code = 0;
    979 	nd_na->nd_na_target = *taddr6;
    980 	in6_clearscope(&nd_na->nd_na_target); /* XXX */
    981 
    982 	/*
    983 	 * "tlladdr" indicates NS's condition for adding tlladdr or not.
    984 	 * see nd6_ns_input() for details.
    985 	 * Basically, if NS packet is sent to unicast/anycast addr,
    986 	 * target lladdr option SHOULD NOT be included.
    987 	 */
    988 	if (tlladdr) {
    989 		/*
    990 		 * sdl0 != NULL indicates proxy NA.  If we do proxy, use
    991 		 * lladdr in sdl0.  If we are not proxying (sending NA for
    992 		 * my address) use lladdr configured for the interface.
    993 		 */
    994 		if (sdl0 == NULL)
    995 			mac = nd6_ifptomac(ifp);
    996 		else if (sdl0->sa_family == AF_LINK) {
    997 			const struct sockaddr_dl *sdl;
    998 			sdl = satocsdl(sdl0);
    999 			if (sdl->sdl_alen == ifp->if_addrlen)
   1000 				mac = CLLADDR(sdl);
   1001 		}
   1002 	}
   1003 	if (tlladdr && mac) {
   1004 		int optlen = sizeof(struct nd_opt_hdr) + ifp->if_addrlen;
   1005 		struct nd_opt_hdr *nd_opt = (struct nd_opt_hdr *)(nd_na + 1);
   1006 
   1007 		/* roundup to 8 bytes alignment! */
   1008 		optlen = (optlen + 7) & ~7;
   1009 
   1010 		m->m_pkthdr.len += optlen;
   1011 		m->m_len += optlen;
   1012 		icmp6len += optlen;
   1013 		memset((void *)nd_opt, 0, optlen);
   1014 		nd_opt->nd_opt_type = ND_OPT_TARGET_LINKADDR;
   1015 		nd_opt->nd_opt_len = optlen >> 3;
   1016 		memcpy((void *)(nd_opt + 1), mac, ifp->if_addrlen);
   1017 	} else
   1018 		flags &= ~ND_NA_FLAG_OVERRIDE;
   1019 
   1020 	ip6->ip6_plen = htons((u_int16_t)icmp6len);
   1021 	nd_na->nd_na_flags_reserved = flags;
   1022 	nd_na->nd_na_cksum = 0;
   1023 	nd_na->nd_na_cksum =
   1024 	    in6_cksum(m, IPPROTO_ICMPV6, sizeof(struct ip6_hdr), icmp6len);
   1025 
   1026 	ip6_output(m, NULL, NULL, 0, &im6o, NULL, NULL);
   1027 
   1028 	icmp6_ifstat_inc(ifp, ifs6_out_msg);
   1029 	icmp6_ifstat_inc(ifp, ifs6_out_neighboradvert);
   1030 	ICMP6_STATINC(ICMP6_STAT_OUTHIST + ND_NEIGHBOR_ADVERT);
   1031 
   1032 	rtcache_free(&ro);
   1033 	return;
   1034 
   1035   bad:
   1036 	rtcache_free(&ro);
   1037 	m_freem(m);
   1038 	return;
   1039 }
   1040 
   1041 const void *
   1042 nd6_ifptomac(const struct ifnet *ifp)
   1043 {
   1044 	switch (ifp->if_type) {
   1045 	case IFT_ARCNET:
   1046 	case IFT_ETHER:
   1047 	case IFT_FDDI:
   1048 	case IFT_IEEE1394:
   1049 	case IFT_PROPVIRTUAL:
   1050 	case IFT_CARP:
   1051 	case IFT_L2VLAN:
   1052 	case IFT_IEEE80211:
   1053 		return CLLADDR(ifp->if_sadl);
   1054 	default:
   1055 		return NULL;
   1056 	}
   1057 }
   1058 
   1059 TAILQ_HEAD(dadq_head, dadq);
   1060 struct dadq {
   1061 	TAILQ_ENTRY(dadq) dad_list;
   1062 	struct ifaddr *dad_ifa;
   1063 	int dad_count;		/* max NS to send */
   1064 	int dad_ns_tcount;	/* # of trials to send NS */
   1065 	int dad_ns_ocount;	/* NS sent so far */
   1066 	int dad_ns_icount;
   1067 	int dad_na_icount;
   1068 	struct callout dad_timer_ch;
   1069 };
   1070 
   1071 static struct dadq_head dadq;
   1072 static int dad_init = 0;
   1073 static kmutex_t nd6_dad_lock;
   1074 
   1075 static struct dadq *
   1076 nd6_dad_find(struct ifaddr *ifa)
   1077 {
   1078 	struct dadq *dp;
   1079 
   1080 	KASSERT(mutex_owned(&nd6_dad_lock));
   1081 
   1082 	TAILQ_FOREACH(dp, &dadq, dad_list) {
   1083 		if (dp->dad_ifa == ifa)
   1084 			return dp;
   1085 	}
   1086 	return NULL;
   1087 }
   1088 
   1089 static void
   1090 nd6_dad_starttimer(struct dadq *dp, int ticks)
   1091 {
   1092 
   1093 	callout_reset(&dp->dad_timer_ch, ticks,
   1094 	    (void (*)(void *))nd6_dad_timer, (void *)dp->dad_ifa);
   1095 }
   1096 
   1097 static void
   1098 nd6_dad_stoptimer(struct dadq *dp)
   1099 {
   1100 
   1101 #ifdef NET_MPSAFE
   1102 	callout_halt(&dp->dad_timer_ch, NULL);
   1103 #else
   1104 	callout_halt(&dp->dad_timer_ch, softnet_lock);
   1105 #endif
   1106 }
   1107 
   1108 /*
   1109  * Start Duplicate Address Detection (DAD) for specified interface address.
   1110  *
   1111  * Note that callout is used when xtick > 0 and not when xtick == 0.
   1112  *
   1113  * xtick: minimum delay ticks for IFF_UP event
   1114  */
   1115 void
   1116 nd6_dad_start(struct ifaddr *ifa, int xtick)
   1117 {
   1118 	struct in6_ifaddr *ia = (struct in6_ifaddr *)ifa;
   1119 	struct dadq *dp;
   1120 
   1121 	if (!dad_init) {
   1122 		TAILQ_INIT(&dadq);
   1123 		mutex_init(&nd6_dad_lock, MUTEX_DEFAULT, IPL_NONE);
   1124 		dad_init++;
   1125 	}
   1126 
   1127 	/*
   1128 	 * If we don't need DAD, don't do it.
   1129 	 * There are several cases:
   1130 	 * - DAD is disabled (ip6_dad_count == 0)
   1131 	 * - the interface address is anycast
   1132 	 */
   1133 	if (!(ia->ia6_flags & IN6_IFF_TENTATIVE)) {
   1134 		log(LOG_DEBUG,
   1135 			"nd6_dad_start: called with non-tentative address "
   1136 			"%s(%s)\n",
   1137 			ip6_sprintf(&ia->ia_addr.sin6_addr),
   1138 			ifa->ifa_ifp ? if_name(ifa->ifa_ifp) : "???");
   1139 		return;
   1140 	}
   1141 	if (ia->ia6_flags & IN6_IFF_ANYCAST || !ip6_dad_count) {
   1142 		ia->ia6_flags &= ~IN6_IFF_TENTATIVE;
   1143 		rt_newaddrmsg(RTM_NEWADDR, ifa, 0, NULL);
   1144 		return;
   1145 	}
   1146 	KASSERT(ifa->ifa_ifp != NULL);
   1147 	if (!(ifa->ifa_ifp->if_flags & IFF_UP))
   1148 		return;
   1149 
   1150 	mutex_enter(&nd6_dad_lock);
   1151 	if (nd6_dad_find(ifa) != NULL) {
   1152 		mutex_exit(&nd6_dad_lock);
   1153 		/* DAD already in progress */
   1154 		return;
   1155 	}
   1156 
   1157 	dp = malloc(sizeof(*dp), M_IP6NDP, M_NOWAIT);
   1158 	if (dp == NULL) {
   1159 		mutex_exit(&nd6_dad_lock);
   1160 		log(LOG_ERR, "nd6_dad_start: memory allocation failed for "
   1161 			"%s(%s)\n",
   1162 			ip6_sprintf(&ia->ia_addr.sin6_addr),
   1163 			ifa->ifa_ifp ? if_name(ifa->ifa_ifp) : "???");
   1164 		return;
   1165 	}
   1166 	memset(dp, 0, sizeof(*dp));
   1167 	callout_init(&dp->dad_timer_ch, CALLOUT_MPSAFE);
   1168 
   1169 	/*
   1170 	 * Send NS packet for DAD, ip6_dad_count times.
   1171 	 * Note that we must delay the first transmission, if this is the
   1172 	 * first packet to be sent from the interface after interface
   1173 	 * (re)initialization.
   1174 	 */
   1175 	dp->dad_ifa = ifa;
   1176 	ifaref(ifa);	/* just for safety */
   1177 	dp->dad_count = ip6_dad_count;
   1178 	dp->dad_ns_icount = dp->dad_na_icount = 0;
   1179 	dp->dad_ns_ocount = dp->dad_ns_tcount = 0;
   1180 	TAILQ_INSERT_TAIL(&dadq, (struct dadq *)dp, dad_list);
   1181 
   1182 	nd6log(LOG_DEBUG, "%s: starting DAD for %s\n", if_name(ifa->ifa_ifp),
   1183 	    ip6_sprintf(&ia->ia_addr.sin6_addr));
   1184 
   1185 	if (xtick == 0) {
   1186 		nd6_dad_ns_output(dp, ifa);
   1187 		nd6_dad_starttimer(dp,
   1188 		    (long)ND_IFINFO(ifa->ifa_ifp)->retrans * hz / 1000);
   1189 	} else
   1190 		nd6_dad_starttimer(dp, xtick);
   1191 	mutex_exit(&nd6_dad_lock);
   1192 }
   1193 
   1194 /*
   1195  * terminate DAD unconditionally.  used for address removals.
   1196  */
   1197 void
   1198 nd6_dad_stop(struct ifaddr *ifa)
   1199 {
   1200 	struct dadq *dp;
   1201 
   1202 	if (!dad_init)
   1203 		return;
   1204 
   1205 	mutex_enter(&nd6_dad_lock);
   1206 	dp = nd6_dad_find(ifa);
   1207 	if (dp == NULL) {
   1208 		mutex_exit(&nd6_dad_lock);
   1209 		/* DAD wasn't started yet */
   1210 		return;
   1211 	}
   1212 
   1213 	/* Prevent the timer from running anymore. */
   1214 	TAILQ_REMOVE(&dadq, dp, dad_list);
   1215 	mutex_exit(&nd6_dad_lock);
   1216 
   1217 	nd6_dad_stoptimer(dp);
   1218 
   1219 	free(dp, M_IP6NDP);
   1220 	dp = NULL;
   1221 	ifafree(ifa);
   1222 }
   1223 
   1224 static void
   1225 nd6_dad_timer(struct ifaddr *ifa)
   1226 {
   1227 	struct in6_ifaddr *ia = (struct in6_ifaddr *)ifa;
   1228 	struct dadq *dp;
   1229 	int duplicate = 0;
   1230 
   1231 #ifndef NET_MPSAFE
   1232 	mutex_enter(softnet_lock);
   1233 	KERNEL_LOCK(1, NULL);
   1234 #endif
   1235 	mutex_enter(&nd6_dad_lock);
   1236 
   1237 	/* Sanity check */
   1238 	if (ia == NULL) {
   1239 		log(LOG_ERR, "nd6_dad_timer: called with null parameter\n");
   1240 		goto done;
   1241 	}
   1242 	dp = nd6_dad_find(ifa);
   1243 	if (dp == NULL) {
   1244 		/* DAD seems to be stopping, so do nothing. */
   1245 		goto done;
   1246 	}
   1247 	if (ia->ia6_flags & IN6_IFF_DUPLICATED) {
   1248 		log(LOG_ERR, "nd6_dad_timer: called with duplicate address "
   1249 			"%s(%s)\n",
   1250 			ip6_sprintf(&ia->ia_addr.sin6_addr),
   1251 			ifa->ifa_ifp ? if_name(ifa->ifa_ifp) : "???");
   1252 		goto done;
   1253 	}
   1254 	if ((ia->ia6_flags & IN6_IFF_TENTATIVE) == 0) {
   1255 		log(LOG_ERR, "nd6_dad_timer: called with non-tentative address "
   1256 			"%s(%s)\n",
   1257 			ip6_sprintf(&ia->ia_addr.sin6_addr),
   1258 			ifa->ifa_ifp ? if_name(ifa->ifa_ifp) : "???");
   1259 		goto done;
   1260 	}
   1261 
   1262 	/* timeouted with IFF_{RUNNING,UP} check */
   1263 	if (dp->dad_ns_tcount > dad_maxtry) {
   1264 		nd6log(LOG_INFO, "%s: could not run DAD, driver problem?\n",
   1265 			if_name(ifa->ifa_ifp));
   1266 
   1267 		TAILQ_REMOVE(&dadq, dp, dad_list);
   1268 		free(dp, M_IP6NDP);
   1269 		dp = NULL;
   1270 		ifafree(ifa);
   1271 		goto done;
   1272 	}
   1273 
   1274 	/* Need more checks? */
   1275 	if (dp->dad_ns_ocount < dp->dad_count) {
   1276 		/*
   1277 		 * We have more NS to go.  Send NS packet for DAD.
   1278 		 */
   1279 		nd6_dad_ns_output(dp, ifa);
   1280 		nd6_dad_starttimer(dp,
   1281 		    (long)ND_IFINFO(ifa->ifa_ifp)->retrans * hz / 1000);
   1282 	} else {
   1283 		/*
   1284 		 * We have transmitted sufficient number of DAD packets.
   1285 		 * See what we've got.
   1286 		 */
   1287 		if (dp->dad_na_icount) {
   1288 			/*
   1289 			 * the check is in nd6_dad_na_input(),
   1290 			 * but just in case
   1291 			 */
   1292 			duplicate++;
   1293 		}
   1294 
   1295 		if (dp->dad_ns_icount) {
   1296 			/* We've seen NS, means DAD has failed. */
   1297 			duplicate++;
   1298 		}
   1299 
   1300 		if (duplicate) {
   1301 			/* (*dp) will be freed in nd6_dad_duplicated() */
   1302 			dp = NULL;
   1303 		} else {
   1304 			/*
   1305 			 * We are done with DAD.  No NA came, no NS came.
   1306 			 * No duplicate address found.
   1307 			 */
   1308 			ia->ia6_flags &= ~IN6_IFF_TENTATIVE;
   1309 			rt_newaddrmsg(RTM_NEWADDR, ifa, 0, NULL);
   1310 
   1311 			nd6log(LOG_DEBUG,
   1312 			    "%s: DAD complete for %s - no duplicates found\n",
   1313 			    if_name(ifa->ifa_ifp),
   1314 			    ip6_sprintf(&ia->ia_addr.sin6_addr));
   1315 
   1316 			TAILQ_REMOVE(&dadq, dp, dad_list);
   1317 			free(dp, M_IP6NDP);
   1318 			dp = NULL;
   1319 			ifafree(ifa);
   1320 		}
   1321 	}
   1322 
   1323 done:
   1324 	mutex_exit(&nd6_dad_lock);
   1325 
   1326 	if (duplicate)
   1327 		nd6_dad_duplicated(ifa);
   1328 
   1329 #ifndef NET_MPSAFE
   1330 	KERNEL_UNLOCK_ONE(NULL);
   1331 	mutex_exit(softnet_lock);
   1332 #endif
   1333 }
   1334 
   1335 static void
   1336 nd6_dad_duplicated(struct ifaddr *ifa)
   1337 {
   1338 	struct in6_ifaddr *ia = (struct in6_ifaddr *)ifa;
   1339 	struct ifnet *ifp;
   1340 	struct dadq *dp;
   1341 
   1342 	mutex_enter(&nd6_dad_lock);
   1343 	dp = nd6_dad_find(ifa);
   1344 	if (dp == NULL) {
   1345 		mutex_exit(&nd6_dad_lock);
   1346 		/* DAD seems to be stopping, so do nothing. */
   1347 		return;
   1348 	}
   1349 
   1350 	ifp = ifa->ifa_ifp;
   1351 	log(LOG_ERR, "%s: DAD detected duplicate IPv6 address %s: "
   1352 	    "NS in/out=%d/%d, NA in=%d\n",
   1353 	    if_name(ifp), ip6_sprintf(&ia->ia_addr.sin6_addr),
   1354 	    dp->dad_ns_icount, dp->dad_ns_ocount, dp->dad_na_icount);
   1355 
   1356 	ia->ia6_flags &= ~IN6_IFF_TENTATIVE;
   1357 	ia->ia6_flags |= IN6_IFF_DUPLICATED;
   1358 
   1359 	/* We are done with DAD, with duplicated address found. (failure) */
   1360 	nd6_dad_stoptimer(dp);
   1361 
   1362 	log(LOG_ERR, "%s: DAD complete for %s - duplicate found\n",
   1363 	    if_name(ifp), ip6_sprintf(&ia->ia_addr.sin6_addr));
   1364 	log(LOG_ERR, "%s: manual intervention required\n",
   1365 	    if_name(ifp));
   1366 
   1367 	/* Inform the routing socket that DAD has completed */
   1368 	rt_newaddrmsg(RTM_NEWADDR, ifa, 0, NULL);
   1369 
   1370 	/*
   1371 	 * If the address is a link-local address formed from an interface
   1372 	 * identifier based on the hardware address which is supposed to be
   1373 	 * uniquely assigned (e.g., EUI-64 for an Ethernet interface), IP
   1374 	 * operation on the interface SHOULD be disabled.
   1375 	 * [rfc2462bis-03 Section 5.4.5]
   1376 	 */
   1377 	if (IN6_IS_ADDR_LINKLOCAL(&ia->ia_addr.sin6_addr)) {
   1378 		struct in6_addr in6;
   1379 
   1380 		/*
   1381 		 * To avoid over-reaction, we only apply this logic when we are
   1382 		 * very sure that hardware addresses are supposed to be unique.
   1383 		 */
   1384 		switch (ifp->if_type) {
   1385 		case IFT_ETHER:
   1386 		case IFT_FDDI:
   1387 		case IFT_ATM:
   1388 		case IFT_IEEE1394:
   1389 #ifdef IFT_IEEE80211
   1390 		case IFT_IEEE80211:
   1391 #endif
   1392 			in6 = ia->ia_addr.sin6_addr;
   1393 			if (in6_get_hw_ifid(ifp, &in6) == 0 &&
   1394 			    IN6_ARE_ADDR_EQUAL(&ia->ia_addr.sin6_addr, &in6)) {
   1395 				ND_IFINFO(ifp)->flags |= ND6_IFF_IFDISABLED;
   1396 				log(LOG_ERR, "%s: possible hardware address "
   1397 				    "duplication detected, disable IPv6\n",
   1398 				    if_name(ifp));
   1399 			}
   1400 			break;
   1401 		}
   1402 	}
   1403 
   1404 	TAILQ_REMOVE(&dadq, dp, dad_list);
   1405 	mutex_exit(&nd6_dad_lock);
   1406 
   1407 	free(dp, M_IP6NDP);
   1408 	dp = NULL;
   1409 	ifafree(ifa);
   1410 }
   1411 
   1412 static void
   1413 nd6_dad_ns_output(struct dadq *dp, struct ifaddr *ifa)
   1414 {
   1415 	struct in6_ifaddr *ia = (struct in6_ifaddr *)ifa;
   1416 	struct ifnet *ifp = ifa->ifa_ifp;
   1417 
   1418 	dp->dad_ns_tcount++;
   1419 	if ((ifp->if_flags & IFF_UP) == 0) {
   1420 #if 0
   1421 		printf("%s: interface down?\n", if_name(ifp));
   1422 #endif
   1423 		return;
   1424 	}
   1425 	if ((ifp->if_flags & IFF_RUNNING) == 0) {
   1426 #if 0
   1427 		printf("%s: interface not running?\n", if_name(ifp));
   1428 #endif
   1429 		return;
   1430 	}
   1431 
   1432 	dp->dad_ns_tcount = 0;
   1433 	dp->dad_ns_ocount++;
   1434 	nd6_ns_output(ifp, NULL, &ia->ia_addr.sin6_addr, NULL, 1);
   1435 }
   1436 
   1437 static void
   1438 nd6_dad_ns_input(struct ifaddr *ifa)
   1439 {
   1440 	struct in6_ifaddr *ia;
   1441 	const struct in6_addr *taddr6;
   1442 	struct dadq *dp;
   1443 	int duplicate;
   1444 
   1445 	if (ifa == NULL)
   1446 		panic("ifa == NULL in nd6_dad_ns_input");
   1447 
   1448 	ia = (struct in6_ifaddr *)ifa;
   1449 	taddr6 = &ia->ia_addr.sin6_addr;
   1450 	duplicate = 0;
   1451 
   1452 	mutex_enter(&nd6_dad_lock);
   1453 	dp = nd6_dad_find(ifa);
   1454 
   1455 	/* Quickhack - completely ignore DAD NS packets */
   1456 	if (dad_ignore_ns) {
   1457 		nd6log(LOG_INFO, "ignoring DAD NS packet for "
   1458 		    "address %s(%s)\n", ip6_sprintf(taddr6),
   1459 		    if_name(ifa->ifa_ifp));
   1460 		return;
   1461 	}
   1462 
   1463 	/*
   1464 	 * if I'm yet to start DAD, someone else started using this address
   1465 	 * first.  I have a duplicate and you win.
   1466 	 */
   1467 	if (dp == NULL || dp->dad_ns_ocount == 0)
   1468 		duplicate++;
   1469 
   1470 	/* XXX more checks for loopback situation - see nd6_dad_timer too */
   1471 
   1472 	if (duplicate) {
   1473 		dp = NULL;	/* will be freed in nd6_dad_duplicated() */
   1474 		mutex_exit(&nd6_dad_lock);
   1475 		nd6_dad_duplicated(ifa);
   1476 	} else {
   1477 		/*
   1478 		 * not sure if I got a duplicate.
   1479 		 * increment ns count and see what happens.
   1480 		 */
   1481 		if (dp)
   1482 			dp->dad_ns_icount++;
   1483 		mutex_exit(&nd6_dad_lock);
   1484 	}
   1485 }
   1486 
   1487 static void
   1488 nd6_dad_na_input(struct ifaddr *ifa)
   1489 {
   1490 	struct dadq *dp;
   1491 
   1492 	if (ifa == NULL)
   1493 		panic("ifa == NULL in nd6_dad_na_input");
   1494 
   1495 	mutex_enter(&nd6_dad_lock);
   1496 	dp = nd6_dad_find(ifa);
   1497 	if (dp)
   1498 		dp->dad_na_icount++;
   1499 	mutex_exit(&nd6_dad_lock);
   1500 
   1501 	/* remove the address. */
   1502 	nd6_dad_duplicated(ifa);
   1503 }
   1504