xref: /freebsd/sys/netinet6/nd6_nbr.c (revision ba3c1f5972d7b90feb6e6da47905ff2757e0fe57)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (C) 1995, 1996, 1997, and 1998 WIDE Project.
5  * All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  * 3. Neither the name of the project nor the names of its contributors
16  *    may be used to endorse or promote products derived from this software
17  *    without specific prior written permission.
18  *
19  * THIS SOFTWARE IS PROVIDED BY THE PROJECT AND CONTRIBUTORS ``AS IS'' AND
20  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
21  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
22  * ARE DISCLAIMED.  IN NO EVENT SHALL THE PROJECT OR CONTRIBUTORS BE LIABLE
23  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
24  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
25  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
26  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
27  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
28  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
29  * SUCH DAMAGE.
30  *
31  *	$KAME: nd6_nbr.c,v 1.86 2002/01/21 02:33:04 jinmei Exp $
32  */
33 
34 #include <sys/cdefs.h>
35 __FBSDID("$FreeBSD$");
36 
37 #include "opt_inet.h"
38 #include "opt_inet6.h"
39 #include "opt_ipsec.h"
40 
41 #include <sys/param.h>
42 #include <sys/systm.h>
43 #include <sys/eventhandler.h>
44 #include <sys/malloc.h>
45 #include <sys/libkern.h>
46 #include <sys/lock.h>
47 #include <sys/rwlock.h>
48 #include <sys/mbuf.h>
49 #include <sys/socket.h>
50 #include <sys/sockio.h>
51 #include <sys/time.h>
52 #include <sys/kernel.h>
53 #include <sys/errno.h>
54 #include <sys/sysctl.h>
55 #include <sys/syslog.h>
56 #include <sys/queue.h>
57 #include <sys/callout.h>
58 #include <sys/refcount.h>
59 
60 #include <net/if.h>
61 #include <net/if_types.h>
62 #include <net/if_dl.h>
63 #include <net/if_var.h>
64 #include <net/if_private.h>
65 #include <net/route.h>
66 #include <net/vnet.h>
67 
68 #include <netinet/in.h>
69 #include <netinet/in_var.h>
70 #include <net/if_llatbl.h>
71 #include <netinet6/in6_var.h>
72 #include <netinet6/in6_ifattach.h>
73 #include <netinet/ip6.h>
74 #include <netinet6/ip6_var.h>
75 #include <netinet6/scope6_var.h>
76 #include <netinet6/nd6.h>
77 #include <netinet/icmp6.h>
78 #include <netinet/ip_carp.h>
79 #include <netinet6/send.h>
80 
81 #define SDL(s) ((struct sockaddr_dl *)s)
82 
83 struct dadq;
84 static struct dadq *nd6_dad_find(struct ifaddr *, struct nd_opt_nonce *);
85 static void nd6_dad_add(struct dadq *dp);
86 static void nd6_dad_del(struct dadq *dp);
87 static void nd6_dad_rele(struct dadq *);
88 static void nd6_dad_starttimer(struct dadq *, int);
89 static void nd6_dad_stoptimer(struct dadq *);
90 static void nd6_dad_timer(void *);
91 static void nd6_dad_duplicated(struct ifaddr *, struct dadq *);
92 static void nd6_dad_ns_output(struct dadq *);
93 static void nd6_dad_ns_input(struct ifaddr *, struct nd_opt_nonce *);
94 static void nd6_dad_na_input(struct ifaddr *);
95 static void nd6_na_output_fib(struct ifnet *, const struct in6_addr *,
96     const struct in6_addr *, u_long, int, struct sockaddr *, u_int);
97 static void nd6_ns_output_fib(struct ifnet *, const struct in6_addr *,
98     const struct in6_addr *, const struct in6_addr *, uint8_t *, u_int);
99 
100 static struct ifaddr *nd6_proxy_fill_sdl(struct ifnet *,
101     const struct in6_addr *, struct sockaddr_dl *);
102 
103 VNET_DEFINE_STATIC(int, dad_enhanced) = 1;
104 #define	V_dad_enhanced			VNET(dad_enhanced)
105 
106 SYSCTL_DECL(_net_inet6_ip6);
107 SYSCTL_INT(_net_inet6_ip6, OID_AUTO, dad_enhanced, CTLFLAG_VNET | CTLFLAG_RW,
108     &VNET_NAME(dad_enhanced), 0,
109     "Enable Enhanced DAD, which adds a random nonce to NS messages for DAD.");
110 
111 VNET_DEFINE_STATIC(int, dad_maxtry) = 15;	/* max # of *tries* to
112 						   transmit DAD packet */
113 #define	V_dad_maxtry			VNET(dad_maxtry)
114 
115 /*
116  * Input a Neighbor Solicitation Message.
117  *
118  * Based on RFC 2461
119  * Based on RFC 2462 (duplicate address detection)
120  */
121 void
122 nd6_ns_input(struct mbuf *m, int off, int icmp6len)
123 {
124 	struct ifnet *ifp;
125 	struct ip6_hdr *ip6;
126 	struct nd_neighbor_solicit *nd_ns;
127 	struct in6_addr daddr6, myaddr6, saddr6, taddr6;
128 	struct ifaddr *ifa;
129 	struct sockaddr_dl proxydl;
130 	union nd_opts ndopts;
131 	char ip6bufs[INET6_ADDRSTRLEN], ip6bufd[INET6_ADDRSTRLEN];
132 	char *lladdr;
133 	int anycast, lladdrlen, proxy, rflag, tentative, tlladdr;
134 
135 	ifa = NULL;
136 
137 	/* RFC 6980: Nodes MUST silently ignore fragments */
138 	if(m->m_flags & M_FRAGMENTED)
139 		goto freeit;
140 
141 	ifp = m->m_pkthdr.rcvif;
142 	ip6 = mtod(m, struct ip6_hdr *);
143 	if (__predict_false(ip6->ip6_hlim != 255)) {
144 		ICMP6STAT_INC(icp6s_invlhlim);
145 		nd6log((LOG_ERR,
146 		    "nd6_ns_input: invalid hlim (%d) from %s to %s on %s\n",
147 		    ip6->ip6_hlim, ip6_sprintf(ip6bufs, &ip6->ip6_src),
148 		    ip6_sprintf(ip6bufd, &ip6->ip6_dst), if_name(ifp)));
149 		goto bads;
150 	}
151 
152 	if (m->m_len < off + icmp6len) {
153 		m = m_pullup(m, off + icmp6len);
154 		if (m == NULL) {
155 			IP6STAT_INC(ip6s_exthdrtoolong);
156 			return;
157 		}
158 	}
159 	ip6 = mtod(m, struct ip6_hdr *);
160 	nd_ns = (struct nd_neighbor_solicit *)((caddr_t)ip6 + off);
161 
162 	saddr6 = ip6->ip6_src;
163 	daddr6 = ip6->ip6_dst;
164 	taddr6 = nd_ns->nd_ns_target;
165 	if (in6_setscope(&taddr6, ifp, NULL) != 0)
166 		goto bad;
167 
168 	rflag = (V_ip6_forwarding) ? ND_NA_FLAG_ROUTER : 0;
169 	if (ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV && V_ip6_norbit_raif)
170 		rflag = 0;
171 
172 	if (IN6_IS_ADDR_UNSPECIFIED(&saddr6)) {
173 		/* dst has to be a solicited node multicast address. */
174 		if (daddr6.s6_addr16[0] == IPV6_ADDR_INT16_MLL &&
175 		    /* don't check ifindex portion */
176 		    daddr6.s6_addr32[1] == 0 &&
177 		    daddr6.s6_addr32[2] == IPV6_ADDR_INT32_ONE &&
178 		    daddr6.s6_addr8[12] == 0xff) {
179 			; /* good */
180 		} else {
181 			nd6log((LOG_INFO, "nd6_ns_input: bad DAD packet "
182 			    "(wrong ip6 dst)\n"));
183 			goto bad;
184 		}
185 	} else if (!V_nd6_onlink_ns_rfc4861) {
186 		struct sockaddr_in6 src_sa6;
187 
188 		/*
189 		 * According to recent IETF discussions, it is not a good idea
190 		 * to accept a NS from an address which would not be deemed
191 		 * to be a neighbor otherwise.  This point is expected to be
192 		 * clarified in future revisions of the specification.
193 		 */
194 		bzero(&src_sa6, sizeof(src_sa6));
195 		src_sa6.sin6_family = AF_INET6;
196 		src_sa6.sin6_len = sizeof(src_sa6);
197 		src_sa6.sin6_addr = saddr6;
198 		if (nd6_is_addr_neighbor(&src_sa6, ifp) == 0) {
199 			nd6log((LOG_INFO, "nd6_ns_input: "
200 				"NS packet from non-neighbor\n"));
201 			goto bad;
202 		}
203 	}
204 
205 	if (IN6_IS_ADDR_MULTICAST(&taddr6)) {
206 		nd6log((LOG_INFO, "nd6_ns_input: bad NS target (multicast)\n"));
207 		goto bad;
208 	}
209 
210 	icmp6len -= sizeof(*nd_ns);
211 	nd6_option_init(nd_ns + 1, icmp6len, &ndopts);
212 	if (nd6_options(&ndopts) < 0) {
213 		nd6log((LOG_INFO,
214 		    "nd6_ns_input: invalid ND option, ignored\n"));
215 		/* nd6_options have incremented stats */
216 		goto freeit;
217 	}
218 
219 	lladdr = NULL;
220 	lladdrlen = 0;
221 	if (ndopts.nd_opts_src_lladdr) {
222 		lladdr = (char *)(ndopts.nd_opts_src_lladdr + 1);
223 		lladdrlen = ndopts.nd_opts_src_lladdr->nd_opt_len << 3;
224 	}
225 
226 	if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src) && lladdr) {
227 		nd6log((LOG_INFO, "nd6_ns_input: bad DAD packet "
228 		    "(link-layer address option)\n"));
229 		goto bad;
230 	}
231 
232 	/*
233 	 * Attaching target link-layer address to the NA?
234 	 * (RFC 2461 7.2.4)
235 	 *
236 	 * NS IP dst is unicast/anycast			MUST NOT add
237 	 * NS IP dst is solicited-node multicast	MUST add
238 	 *
239 	 * In implementation, we add target link-layer address by default.
240 	 * We do not add one in MUST NOT cases.
241 	 */
242 	if (!IN6_IS_ADDR_MULTICAST(&daddr6))
243 		tlladdr = 0;
244 	else
245 		tlladdr = 1;
246 
247 	/*
248 	 * Target address (taddr6) must be either:
249 	 * (1) Valid unicast/anycast address for my receiving interface,
250 	 * (2) Unicast address for which I'm offering proxy service, or
251 	 * (3) "tentative" address on which DAD is being performed.
252 	 */
253 	/* (1) and (3) check. */
254 	if (ifp->if_carp)
255 		ifa = (*carp_iamatch6_p)(ifp, &taddr6);
256 	else
257 		ifa = (struct ifaddr *)in6ifa_ifpwithaddr(ifp, &taddr6);
258 
259 	/* (2) check. */
260 	proxy = 0;
261 	if (ifa == NULL) {
262 		if ((ifa = nd6_proxy_fill_sdl(ifp, &taddr6, &proxydl)) != NULL)
263 			proxy = 1;
264 	}
265 	if (ifa == NULL) {
266 		/*
267 		 * We've got an NS packet, and we don't have that adddress
268 		 * assigned for us.  We MUST silently ignore it.
269 		 * See RFC2461 7.2.3.
270 		 */
271 		goto freeit;
272 	}
273 	myaddr6 = *IFA_IN6(ifa);
274 	anycast = ((struct in6_ifaddr *)ifa)->ia6_flags & IN6_IFF_ANYCAST;
275 	tentative = ((struct in6_ifaddr *)ifa)->ia6_flags & IN6_IFF_TENTATIVE;
276 	if (((struct in6_ifaddr *)ifa)->ia6_flags & IN6_IFF_DUPLICATED)
277 		goto freeit;
278 
279 	if (lladdr && ((ifp->if_addrlen + 2 + 7) & ~7) != lladdrlen) {
280 		nd6log((LOG_INFO, "nd6_ns_input: lladdrlen mismatch for %s "
281 		    "(if %d, NS packet %d)\n",
282 		    ip6_sprintf(ip6bufs, &taddr6),
283 		    ifp->if_addrlen, lladdrlen - 2));
284 		goto bad;
285 	}
286 
287 	if (IN6_ARE_ADDR_EQUAL(&myaddr6, &saddr6)) {
288 		nd6log((LOG_INFO, "nd6_ns_input: duplicate IP6 address %s\n",
289 		    ip6_sprintf(ip6bufs, &saddr6)));
290 		goto freeit;
291 	}
292 
293 	/*
294 	 * We have neighbor solicitation packet, with target address equals to
295 	 * one of my tentative address.
296 	 *
297 	 * src addr	how to process?
298 	 * ---		---
299 	 * multicast	of course, invalid (rejected in ip6_input)
300 	 * unicast	somebody is doing address resolution -> ignore
301 	 * unspec	dup address detection
302 	 *
303 	 * The processing is defined in RFC 2462.
304 	 */
305 	if (tentative) {
306 		/*
307 		 * If source address is unspecified address, it is for
308 		 * duplicate address detection.
309 		 *
310 		 * If not, the packet is for addess resolution;
311 		 * silently ignore it.
312 		 */
313 		if (IN6_IS_ADDR_UNSPECIFIED(&saddr6))
314 			nd6_dad_ns_input(ifa, ndopts.nd_opts_nonce);
315 
316 		goto freeit;
317 	}
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_fib(ifp, &in6_all, &taddr6,
334 		    ((anycast || proxy || !tlladdr) ? 0 : ND_NA_FLAG_OVERRIDE) |
335 		    rflag, tlladdr, proxy ? (struct sockaddr *)&proxydl : NULL,
336 		    M_GETFIB(m));
337 		goto freeit;
338 	}
339 
340 	nd6_cache_lladdr(ifp, &saddr6, lladdr, lladdrlen,
341 	    ND_NEIGHBOR_SOLICIT, 0);
342 
343 	nd6_na_output_fib(ifp, &saddr6, &taddr6,
344 	    ((anycast || proxy || !tlladdr) ? 0 : ND_NA_FLAG_OVERRIDE) |
345 	    rflag | ND_NA_FLAG_SOLICITED, tlladdr,
346 	    proxy ? (struct sockaddr *)&proxydl : NULL, M_GETFIB(m));
347  freeit:
348 	if (ifa != NULL)
349 		ifa_free(ifa);
350 	m_freem(m);
351 	return;
352 
353  bad:
354 	nd6log((LOG_ERR, "nd6_ns_input: src=%s\n",
355 		ip6_sprintf(ip6bufs, &saddr6)));
356 	nd6log((LOG_ERR, "nd6_ns_input: dst=%s\n",
357 		ip6_sprintf(ip6bufs, &daddr6)));
358 	nd6log((LOG_ERR, "nd6_ns_input: tgt=%s\n",
359 		ip6_sprintf(ip6bufs, &taddr6)));
360  bads:
361 	ICMP6STAT_INC(icp6s_badns);
362 	if (ifa != NULL)
363 		ifa_free(ifa);
364 	m_freem(m);
365 }
366 
367 static struct ifaddr *
368 nd6_proxy_fill_sdl(struct ifnet *ifp, const struct in6_addr *taddr6,
369     struct sockaddr_dl *sdl)
370 {
371 	struct ifaddr *ifa;
372 	struct llentry *ln;
373 
374 	ifa = NULL;
375 	ln = nd6_lookup(taddr6, LLE_SF(AF_INET6, 0), ifp);
376 	if (ln == NULL)
377 		return (ifa);
378 	if ((ln->la_flags & (LLE_PUB | LLE_VALID)) == (LLE_PUB | LLE_VALID)) {
379 		link_init_sdl(ifp, (struct sockaddr *)sdl, ifp->if_type);
380 		sdl->sdl_alen = ifp->if_addrlen;
381 		bcopy(ln->ll_addr, &sdl->sdl_data, ifp->if_addrlen);
382 		LLE_RUNLOCK(ln);
383 		ifa = (struct ifaddr *)in6ifa_ifpforlinklocal(ifp,
384 		    IN6_IFF_NOTREADY|IN6_IFF_ANYCAST);
385 	} else
386 		LLE_RUNLOCK(ln);
387 
388 	return (ifa);
389 }
390 
391 /*
392  * Output a Neighbor Solicitation Message. Caller specifies:
393  *	- ICMP6 header source IP6 address
394  *	- ND6 header target IP6 address
395  *	- ND6 header source datalink address
396  *
397  * Based on RFC 2461
398  * Based on RFC 2462 (duplicate address detection)
399  *
400  *    ln - for source address determination
401  * nonce - If non-NULL, NS is used for duplicate address detection and
402  *         the value (length is ND_OPT_NONCE_LEN) is used as a random nonce.
403  */
404 static void
405 nd6_ns_output_fib(struct ifnet *ifp, const struct in6_addr *saddr6,
406     const struct in6_addr *daddr6, const struct in6_addr *taddr6,
407     uint8_t *nonce, u_int fibnum)
408 {
409 	struct mbuf *m;
410 	struct m_tag *mtag;
411 	struct ip6_hdr *ip6;
412 	struct nd_neighbor_solicit *nd_ns;
413 	struct ip6_moptions im6o;
414 	int icmp6len;
415 	int maxlen;
416 
417 	NET_EPOCH_ASSERT();
418 
419 	if (IN6_IS_ADDR_MULTICAST(taddr6))
420 		return;
421 
422 	/* estimate the size of message */
423 	maxlen = sizeof(*ip6) + sizeof(*nd_ns);
424 	maxlen += (sizeof(struct nd_opt_hdr) + ifp->if_addrlen + 7) & ~7;
425 	KASSERT(max_linkhdr + maxlen <= MCLBYTES, (
426 	    "%s: max_linkhdr + maxlen > MCLBYTES (%d + %d > %d)",
427 	    __func__, max_linkhdr, maxlen, MCLBYTES));
428 
429 	if (max_linkhdr + maxlen > MHLEN)
430 		m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR);
431 	else
432 		m = m_gethdr(M_NOWAIT, MT_DATA);
433 	if (m == NULL)
434 		return;
435 	M_SETFIB(m, fibnum);
436 
437 	if (daddr6 == NULL || IN6_IS_ADDR_MULTICAST(daddr6)) {
438 		m->m_flags |= M_MCAST;
439 		im6o.im6o_multicast_ifp = ifp;
440 		im6o.im6o_multicast_hlim = 255;
441 		im6o.im6o_multicast_loop = 0;
442 	}
443 
444 	icmp6len = sizeof(*nd_ns);
445 	m->m_pkthdr.len = m->m_len = sizeof(*ip6) + icmp6len;
446 	m->m_data += max_linkhdr;	/* or M_ALIGN() equivalent? */
447 
448 	/* fill neighbor solicitation packet */
449 	ip6 = mtod(m, struct ip6_hdr *);
450 	ip6->ip6_flow = 0;
451 	ip6->ip6_vfc &= ~IPV6_VERSION_MASK;
452 	ip6->ip6_vfc |= IPV6_VERSION;
453 	/* ip6->ip6_plen will be set later */
454 	ip6->ip6_nxt = IPPROTO_ICMPV6;
455 	ip6->ip6_hlim = 255;
456 	if (daddr6)
457 		ip6->ip6_dst = *daddr6;
458 	else {
459 		ip6->ip6_dst.s6_addr16[0] = IPV6_ADDR_INT16_MLL;
460 		ip6->ip6_dst.s6_addr16[1] = 0;
461 		ip6->ip6_dst.s6_addr32[1] = 0;
462 		ip6->ip6_dst.s6_addr32[2] = IPV6_ADDR_INT32_ONE;
463 		ip6->ip6_dst.s6_addr32[3] = taddr6->s6_addr32[3];
464 		ip6->ip6_dst.s6_addr8[12] = 0xff;
465 		if (in6_setscope(&ip6->ip6_dst, ifp, NULL) != 0)
466 			goto bad;
467 	}
468 	if (nonce == NULL) {
469 		char ip6buf[INET6_ADDRSTRLEN];
470 		struct ifaddr *ifa = NULL;
471 
472 		/*
473 		 * RFC2461 7.2.2:
474 		 * "If the source address of the packet prompting the
475 		 * solicitation is the same as one of the addresses assigned
476 		 * to the outgoing interface, that address SHOULD be placed
477 		 * in the IP Source Address of the outgoing solicitation.
478 		 * Otherwise, any one of the addresses assigned to the
479 		 * interface should be used."
480 		 *
481 		 * We use the source address for the prompting packet
482 		 * (saddr6), if saddr6 belongs to the outgoing interface.
483 		 * Otherwise, we perform the source address selection as usual.
484 		 */
485 		if (saddr6 != NULL)
486 			ifa = (struct ifaddr *)in6ifa_ifpwithaddr(ifp, saddr6);
487 		if (ifa == NULL) {
488 			int error;
489 			struct in6_addr dst6, src6;
490 			uint32_t scopeid;
491 
492 			in6_splitscope(&ip6->ip6_dst, &dst6, &scopeid);
493 			error = in6_selectsrc_addr(fibnum, &dst6,
494 			    scopeid, ifp, &src6, NULL);
495 			if (error) {
496 				nd6log((LOG_DEBUG, "%s: source can't be "
497 				    "determined: dst=%s, error=%d\n", __func__,
498 				    ip6_sprintf(ip6buf, &dst6),
499 				    error));
500 				goto bad;
501 			}
502 			ip6->ip6_src = src6;
503 		} else
504 			ip6->ip6_src = *saddr6;
505 
506 		if (ifp->if_carp != NULL) {
507 			/*
508 			 * Check that selected source address belongs to
509 			 * CARP addresses.
510 			 */
511 			if (ifa == NULL)
512 				ifa = (struct ifaddr *)in6ifa_ifpwithaddr(ifp,
513 				    &ip6->ip6_src);
514 			/*
515 			 * Do not send NS for CARP address if we are not
516 			 * the CARP master.
517 			 */
518 			if (ifa != NULL && ifa->ifa_carp != NULL &&
519 			    !(*carp_master_p)(ifa)) {
520 				nd6log((LOG_DEBUG,
521 				    "nd6_ns_output: NS from BACKUP CARP address %s\n",
522 				    ip6_sprintf(ip6buf, &ip6->ip6_src)));
523 				ifa_free(ifa);
524 				goto bad;
525 			}
526 		}
527 		if (ifa != NULL)
528 			ifa_free(ifa);
529 	} else {
530 		/*
531 		 * Source address for DAD packet must always be IPv6
532 		 * unspecified address. (0::0)
533 		 * We actually don't have to 0-clear the address (we did it
534 		 * above), but we do so here explicitly to make the intention
535 		 * clearer.
536 		 */
537 		bzero(&ip6->ip6_src, sizeof(ip6->ip6_src));
538 	}
539 	nd_ns = (struct nd_neighbor_solicit *)(ip6 + 1);
540 	nd_ns->nd_ns_type = ND_NEIGHBOR_SOLICIT;
541 	nd_ns->nd_ns_code = 0;
542 	nd_ns->nd_ns_reserved = 0;
543 	nd_ns->nd_ns_target = *taddr6;
544 	in6_clearscope(&nd_ns->nd_ns_target); /* XXX */
545 
546 	/*
547 	 * Add source link-layer address option.
548 	 *
549 	 *				spec		implementation
550 	 *				---		---
551 	 * DAD packet			MUST NOT	do not add the option
552 	 * there's no link layer address:
553 	 *				impossible	do not add the option
554 	 * there's link layer address:
555 	 *	Multicast NS		MUST add one	add the option
556 	 *	Unicast NS		SHOULD add one	add the option
557 	 */
558 	if (nonce == NULL) {
559 		struct nd_opt_hdr *nd_opt;
560 		char *mac;
561 		int optlen;
562 
563 		mac = NULL;
564 		if (ifp->if_carp)
565 			mac = (*carp_macmatch6_p)(ifp, m, &ip6->ip6_src);
566 		if (mac == NULL)
567 			mac = nd6_ifptomac(ifp);
568 
569 		if (mac != NULL) {
570 			nd_opt = (struct nd_opt_hdr *)(nd_ns + 1);
571 			optlen = sizeof(struct nd_opt_hdr) + ifp->if_addrlen;
572 			/* 8 byte alignments... */
573 			optlen = (optlen + 7) & ~7;
574 			m->m_pkthdr.len += optlen;
575 			m->m_len += optlen;
576 			icmp6len += optlen;
577 			bzero(nd_opt, optlen);
578 			nd_opt->nd_opt_type = ND_OPT_SOURCE_LINKADDR;
579 			nd_opt->nd_opt_len = optlen >> 3;
580 			bcopy(mac, nd_opt + 1, ifp->if_addrlen);
581 		}
582 	}
583 	/*
584 	 * Add a Nonce option (RFC 3971) to detect looped back NS messages.
585 	 * This behavior is documented as Enhanced Duplicate Address
586 	 * Detection in RFC 7527.
587 	 * net.inet6.ip6.dad_enhanced=0 disables this.
588 	 */
589 	if (V_dad_enhanced != 0 && nonce != NULL) {
590 		int optlen = sizeof(struct nd_opt_hdr) + ND_OPT_NONCE_LEN;
591 		struct nd_opt_hdr *nd_opt = (struct nd_opt_hdr *)(nd_ns + 1);
592 		/* 8-byte alignment is required. */
593 		optlen = (optlen + 7) & ~7;
594 
595 		m->m_pkthdr.len += optlen;
596 		m->m_len += optlen;
597 		icmp6len += optlen;
598 		bzero((caddr_t)nd_opt, optlen);
599 		nd_opt->nd_opt_type = ND_OPT_NONCE;
600 		nd_opt->nd_opt_len = optlen >> 3;
601 		bcopy(nonce, (caddr_t)(nd_opt + 1), ND_OPT_NONCE_LEN);
602 	}
603 	ip6->ip6_plen = htons((u_short)icmp6len);
604 	nd_ns->nd_ns_cksum = 0;
605 	nd_ns->nd_ns_cksum =
606 	    in6_cksum(m, IPPROTO_ICMPV6, sizeof(*ip6), icmp6len);
607 
608 	if (send_sendso_input_hook != NULL) {
609 		mtag = m_tag_get(PACKET_TAG_ND_OUTGOING,
610 			sizeof(unsigned short), M_NOWAIT);
611 		if (mtag == NULL)
612 			goto bad;
613 		*(unsigned short *)(mtag + 1) = nd_ns->nd_ns_type;
614 		m_tag_prepend(m, mtag);
615 	}
616 
617 	ip6_output(m, NULL, NULL, (nonce != NULL) ? IPV6_UNSPECSRC : 0,
618 	    &im6o, NULL, NULL);
619 	icmp6_ifstat_inc(ifp, ifs6_out_msg);
620 	icmp6_ifstat_inc(ifp, ifs6_out_neighborsolicit);
621 	ICMP6STAT_INC(icp6s_outhist[ND_NEIGHBOR_SOLICIT]);
622 
623 	return;
624 
625   bad:
626 	m_freem(m);
627 }
628 
629 #ifndef BURN_BRIDGES
630 void
631 nd6_ns_output(struct ifnet *ifp, const struct in6_addr *saddr6,
632     const struct in6_addr *daddr6, const struct in6_addr *taddr6,uint8_t *nonce)
633 {
634 
635 	nd6_ns_output_fib(ifp, saddr6, daddr6, taddr6, nonce, RT_DEFAULT_FIB);
636 }
637 #endif
638 /*
639  * Neighbor advertisement input handling.
640  *
641  * Based on RFC 2461
642  * Based on RFC 2462 (duplicate address detection)
643  *
644  * the following items are not implemented yet:
645  * - proxy advertisement delay rule (RFC2461 7.2.8, last paragraph, SHOULD)
646  * - anycast advertisement delay rule (RFC2461 7.2.7, SHOULD)
647  */
648 void
649 nd6_na_input(struct mbuf *m, int off, int icmp6len)
650 {
651 	struct ifnet *ifp;
652 	struct ip6_hdr *ip6;
653 	struct ifaddr *ifa;
654 	struct llentry *ln;
655 	struct mbuf *chain;
656 	struct nd_neighbor_advert *nd_na;
657 	struct in6_addr daddr6, taddr6;
658 	union nd_opts ndopts;
659 	u_char linkhdr[LLE_MAX_LINKHDR];
660 	char ip6bufs[INET6_ADDRSTRLEN], ip6bufd[INET6_ADDRSTRLEN];
661 	char *lladdr;
662 	size_t linkhdrsize;
663 	int flags, is_override, is_router, is_solicited;
664 	int lladdr_off, lladdrlen, checklink;
665 	bool flush_holdchain = false;
666 
667 	NET_EPOCH_ASSERT();
668 
669 	chain = NULL;
670 	ln = NULL;
671 	checklink = 0;
672 
673 	/* RFC 6980: Nodes MUST silently ignore fragments */
674 	if(m->m_flags & M_FRAGMENTED)
675 		goto freeit;
676 
677 	ifp = m->m_pkthdr.rcvif;
678 	ip6 = mtod(m, struct ip6_hdr *);
679 	if (__predict_false(ip6->ip6_hlim != 255)) {
680 		ICMP6STAT_INC(icp6s_invlhlim);
681 		nd6log((LOG_ERR,
682 		    "nd6_na_input: invalid hlim (%d) from %s to %s on %s\n",
683 		    ip6->ip6_hlim, ip6_sprintf(ip6bufs, &ip6->ip6_src),
684 		    ip6_sprintf(ip6bufd, &ip6->ip6_dst), if_name(ifp)));
685 		goto bad;
686 	}
687 
688 	if (m->m_len < off + icmp6len) {
689 		m = m_pullup(m, off + icmp6len);
690 		if (m == NULL) {
691 			IP6STAT_INC(ip6s_exthdrtoolong);
692 			return;
693 		}
694 	}
695 	ip6 = mtod(m, struct ip6_hdr *);
696 	nd_na = (struct nd_neighbor_advert *)((caddr_t)ip6 + off);
697 
698 	flags = nd_na->nd_na_flags_reserved;
699 	is_router = ((flags & ND_NA_FLAG_ROUTER) != 0);
700 	is_solicited = ((flags & ND_NA_FLAG_SOLICITED) != 0);
701 	is_override = ((flags & ND_NA_FLAG_OVERRIDE) != 0);
702 
703 	taddr6 = nd_na->nd_na_target;
704 	if (in6_setscope(&taddr6, ifp, NULL))
705 		goto bad;	/* XXX: impossible */
706 
707 	if (IN6_IS_ADDR_MULTICAST(&taddr6)) {
708 		nd6log((LOG_ERR,
709 		    "nd6_na_input: invalid target address %s\n",
710 		    ip6_sprintf(ip6bufs, &taddr6)));
711 		goto bad;
712 	}
713 
714 	daddr6 = ip6->ip6_dst;
715 	if (IN6_IS_ADDR_MULTICAST(&daddr6))
716 		if (is_solicited) {
717 			nd6log((LOG_ERR,
718 			    "nd6_na_input: a solicited adv is multicasted\n"));
719 			goto bad;
720 		}
721 
722 	icmp6len -= sizeof(*nd_na);
723 	nd6_option_init(nd_na + 1, icmp6len, &ndopts);
724 	if (nd6_options(&ndopts) < 0) {
725 		nd6log((LOG_INFO,
726 		    "nd6_na_input: invalid ND option, ignored\n"));
727 		/* nd6_options have incremented stats */
728 		goto freeit;
729 	}
730 
731 	lladdr = NULL;
732 	lladdrlen = 0;
733 	if (ndopts.nd_opts_tgt_lladdr) {
734 		lladdr = (char *)(ndopts.nd_opts_tgt_lladdr + 1);
735 		lladdrlen = ndopts.nd_opts_tgt_lladdr->nd_opt_len << 3;
736 	}
737 
738 	ifa = (struct ifaddr *)in6ifa_ifpwithaddr(ifp, &taddr6);
739 	if (ifa != NULL && ifa->ifa_carp != NULL) {
740 		/*
741 		 * Silently ignore NAs for CARP addresses if we are not
742 		 * the CARP master.
743 		 */
744 		if (!(*carp_master_p)(ifa)) {
745 			nd6log((LOG_DEBUG,
746 			    "nd6_na_input: NA for BACKUP CARP address %s\n",
747 			    ip6_sprintf(ip6bufs, &taddr6)));
748 			ifa_free(ifa);
749 			goto freeit;
750 		}
751 	}
752 	/*
753 	 * Target address matches one of my interface address.
754 	 *
755 	 * If my address is tentative, this means that there's somebody
756 	 * already using the same address as mine.  This indicates DAD failure.
757 	 * This is defined in RFC 2462.
758 	 *
759 	 * Otherwise, process as defined in RFC 2461.
760 	 */
761 	if (ifa
762 	 && (((struct in6_ifaddr *)ifa)->ia6_flags & IN6_IFF_TENTATIVE)) {
763 		nd6_dad_na_input(ifa);
764 		ifa_free(ifa);
765 		goto freeit;
766 	}
767 
768 	/* Just for safety, maybe unnecessary. */
769 	if (ifa) {
770 		ifa_free(ifa);
771 		log(LOG_ERR,
772 		    "nd6_na_input: duplicate IP6 address %s\n",
773 		    ip6_sprintf(ip6bufs, &taddr6));
774 		goto freeit;
775 	}
776 
777 	if (lladdr && ((ifp->if_addrlen + 2 + 7) & ~7) != lladdrlen) {
778 		nd6log((LOG_INFO, "nd6_na_input: lladdrlen mismatch for %s "
779 		    "(if %d, NA packet %d)\n", ip6_sprintf(ip6bufs, &taddr6),
780 		    ifp->if_addrlen, lladdrlen - 2));
781 		goto bad;
782 	}
783 
784 	/*
785 	 * If no neighbor cache entry is found, NA SHOULD silently be
786 	 * discarded.
787 	 */
788 	ln = nd6_lookup(&taddr6, LLE_SF(AF_INET6, LLE_EXCLUSIVE), ifp);
789 	if (ln == NULL) {
790 		goto freeit;
791 	}
792 
793 	/*
794 	 * Do not try to override static entry.
795 	 */
796 	if (ln->la_flags & LLE_STATIC)
797 		goto freeit;
798 
799 	if (ln->ln_state == ND6_LLINFO_INCOMPLETE) {
800 		/*
801 		 * If the link-layer has address, and no lladdr option came,
802 		 * discard the packet.
803 		 */
804 		if (ifp->if_addrlen && lladdr == NULL) {
805 			goto freeit;
806 		}
807 
808 		/*
809 		 * Record link-layer address, and update the state.
810 		 */
811 		if (!nd6_try_set_entry_addr(ifp, ln, lladdr))
812 			goto freeit;
813 
814 		flush_holdchain = true;
815 		if (is_solicited)
816 			nd6_llinfo_setstate(ln, ND6_LLINFO_REACHABLE);
817 		else
818 			nd6_llinfo_setstate(ln, ND6_LLINFO_STALE);
819 		EVENTHANDLER_INVOKE(lle_event, ln, LLENTRY_RESOLVED);
820 		if ((ln->ln_router = is_router) != 0) {
821 			/*
822 			 * This means a router's state has changed from
823 			 * non-reachable to probably reachable, and might
824 			 * affect the status of associated prefixes..
825 			 */
826 			checklink = 1;
827 		}
828 	} else {
829 		int llchange;
830 
831 		/*
832 		 * Check if the link-layer address has changed or not.
833 		 */
834 		if (lladdr == NULL)
835 			llchange = 0;
836 		else {
837 			if (ln->la_flags & LLE_VALID) {
838 				if (bcmp(lladdr, ln->ll_addr, ifp->if_addrlen))
839 					llchange = 1;
840 				else
841 					llchange = 0;
842 			} else
843 				llchange = 1;
844 		}
845 
846 		/*
847 		 * This is VERY complex.  Look at it with care.
848 		 *
849 		 * override solicit lladdr llchange	action
850 		 *					(L: record lladdr)
851 		 *
852 		 *	0	0	n	--	(2c)
853 		 *	0	0	y	n	(2b) L
854 		 *	0	0	y	y	(1)    REACHABLE->STALE
855 		 *	0	1	n	--	(2c)   *->REACHABLE
856 		 *	0	1	y	n	(2b) L *->REACHABLE
857 		 *	0	1	y	y	(1)    REACHABLE->STALE
858 		 *	1	0	n	--	(2a)
859 		 *	1	0	y	n	(2a) L
860 		 *	1	0	y	y	(2a) L *->STALE
861 		 *	1	1	n	--	(2a)   *->REACHABLE
862 		 *	1	1	y	n	(2a) L *->REACHABLE
863 		 *	1	1	y	y	(2a) L *->REACHABLE
864 		 */
865 		if (!is_override && (lladdr != NULL && llchange)) {  /* (1) */
866 			/*
867 			 * If state is REACHABLE, make it STALE.
868 			 * no other updates should be done.
869 			 */
870 			if (ln->ln_state == ND6_LLINFO_REACHABLE)
871 				nd6_llinfo_setstate(ln, ND6_LLINFO_STALE);
872 			goto freeit;
873 		} else if (is_override				   /* (2a) */
874 			|| (!is_override && (lladdr != NULL && !llchange)) /* (2b) */
875 			|| lladdr == NULL) {			   /* (2c) */
876 			/*
877 			 * Update link-local address, if any.
878 			 */
879 			if (lladdr != NULL) {
880 				linkhdrsize = sizeof(linkhdr);
881 				if (lltable_calc_llheader(ifp, AF_INET6, lladdr,
882 				    linkhdr, &linkhdrsize, &lladdr_off) != 0)
883 					goto freeit;
884 				if (lltable_try_set_entry_addr(ifp, ln, linkhdr,
885 				    linkhdrsize, lladdr_off) == 0)
886 					goto freeit;
887 				EVENTHANDLER_INVOKE(lle_event, ln,
888 				    LLENTRY_RESOLVED);
889 			}
890 
891 			/*
892 			 * If solicited, make the state REACHABLE.
893 			 * If not solicited and the link-layer address was
894 			 * changed, make it STALE.
895 			 */
896 			if (is_solicited)
897 				nd6_llinfo_setstate(ln, ND6_LLINFO_REACHABLE);
898 			else {
899 				if (lladdr != NULL && llchange)
900 					nd6_llinfo_setstate(ln, ND6_LLINFO_STALE);
901 			}
902 		}
903 
904 		if (ln->ln_router && !is_router) {
905 			/*
906 			 * The peer dropped the router flag.
907 			 * Remove the sender from the Default Router List and
908 			 * update the Destination Cache entries.
909 			 */
910 			struct ifnet *nd6_ifp;
911 
912 			nd6_ifp = lltable_get_ifp(ln->lle_tbl);
913 			if (!defrouter_remove(&ln->r_l3addr.addr6, nd6_ifp) &&
914 			    (ND_IFINFO(nd6_ifp)->flags &
915 			     ND6_IFF_ACCEPT_RTADV) != 0)
916 				/*
917 				 * Even if the neighbor is not in the default
918 				 * router list, the neighbor may be used as a
919 				 * next hop for some destinations (e.g. redirect
920 				 * case). So we must call rt6_flush explicitly.
921 				 */
922 				rt6_flush(&ip6->ip6_src, ifp);
923 		}
924 		ln->ln_router = is_router;
925 	}
926         /* XXX - QL
927 	 *  Does this matter?
928 	 *  rt->rt_flags &= ~RTF_REJECT;
929 	 */
930 	ln->la_asked = 0;
931 	if (ln->la_hold != NULL)
932 		chain = nd6_grab_holdchain(ln);
933  freeit:
934 	if (ln != NULL)
935 		LLE_WUNLOCK(ln);
936 
937 	if (chain != NULL)
938 		nd6_flush_holdchain(ifp, ln, chain);
939 	if (flush_holdchain)
940 		nd6_flush_children_holdchain(ifp, ln);
941 
942 	if (checklink)
943 		pfxlist_onlink_check();
944 
945 	m_freem(m);
946 	return;
947 
948  bad:
949 	if (ln != NULL)
950 		LLE_WUNLOCK(ln);
951 
952 	ICMP6STAT_INC(icp6s_badna);
953 	m_freem(m);
954 }
955 
956 /*
957  * Neighbor advertisement output handling.
958  *
959  * Based on RFC 2461
960  *
961  * the following items are not implemented yet:
962  * - proxy advertisement delay rule (RFC2461 7.2.8, last paragraph, SHOULD)
963  * - anycast advertisement delay rule (RFC2461 7.2.7, SHOULD)
964  *
965  * tlladdr - 1 if include target link-layer address
966  * sdl0 - sockaddr_dl (= proxy NA) or NULL
967  */
968 static void
969 nd6_na_output_fib(struct ifnet *ifp, const struct in6_addr *daddr6_0,
970     const struct in6_addr *taddr6, u_long flags, int tlladdr,
971     struct sockaddr *sdl0, u_int fibnum)
972 {
973 	struct mbuf *m;
974 	struct m_tag *mtag;
975 	struct ip6_hdr *ip6;
976 	struct nd_neighbor_advert *nd_na;
977 	struct ip6_moptions im6o;
978 	struct in6_addr daddr6, dst6, src6;
979 	uint32_t scopeid;
980 
981 	NET_EPOCH_ASSERT();
982 
983 	int icmp6len, maxlen, error;
984 	caddr_t mac = NULL;
985 
986 	daddr6 = *daddr6_0;	/* make a local copy for modification */
987 
988 	/* estimate the size of message */
989 	maxlen = sizeof(*ip6) + sizeof(*nd_na);
990 	maxlen += (sizeof(struct nd_opt_hdr) + ifp->if_addrlen + 7) & ~7;
991 	KASSERT(max_linkhdr + maxlen <= MCLBYTES, (
992 	    "%s: max_linkhdr + maxlen > MCLBYTES (%d + %d > %d)",
993 	    __func__, max_linkhdr, maxlen, MCLBYTES));
994 
995 	if (max_linkhdr + maxlen > MHLEN)
996 		m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR);
997 	else
998 		m = m_gethdr(M_NOWAIT, MT_DATA);
999 	if (m == NULL)
1000 		return;
1001 	M_SETFIB(m, fibnum);
1002 
1003 	if (IN6_IS_ADDR_MULTICAST(&daddr6)) {
1004 		m->m_flags |= M_MCAST;
1005 		im6o.im6o_multicast_ifp = ifp;
1006 		im6o.im6o_multicast_hlim = 255;
1007 		im6o.im6o_multicast_loop = 0;
1008 	}
1009 
1010 	icmp6len = sizeof(*nd_na);
1011 	m->m_pkthdr.len = m->m_len = sizeof(struct ip6_hdr) + icmp6len;
1012 	m->m_data += max_linkhdr;	/* or M_ALIGN() equivalent? */
1013 
1014 	/* fill neighbor advertisement packet */
1015 	ip6 = mtod(m, struct ip6_hdr *);
1016 	ip6->ip6_flow = 0;
1017 	ip6->ip6_vfc &= ~IPV6_VERSION_MASK;
1018 	ip6->ip6_vfc |= IPV6_VERSION;
1019 	ip6->ip6_nxt = IPPROTO_ICMPV6;
1020 	ip6->ip6_hlim = 255;
1021 	if (IN6_IS_ADDR_UNSPECIFIED(&daddr6)) {
1022 		/* reply to DAD */
1023 		daddr6.s6_addr16[0] = IPV6_ADDR_INT16_MLL;
1024 		daddr6.s6_addr16[1] = 0;
1025 		daddr6.s6_addr32[1] = 0;
1026 		daddr6.s6_addr32[2] = 0;
1027 		daddr6.s6_addr32[3] = IPV6_ADDR_INT32_ONE;
1028 		if (in6_setscope(&daddr6, ifp, NULL))
1029 			goto bad;
1030 
1031 		flags &= ~ND_NA_FLAG_SOLICITED;
1032 	}
1033 	ip6->ip6_dst = daddr6;
1034 
1035 	/*
1036 	 * Select a source whose scope is the same as that of the dest.
1037 	 */
1038 	in6_splitscope(&daddr6, &dst6, &scopeid);
1039 	error = in6_selectsrc_addr(fibnum, &dst6,
1040 	    scopeid, ifp, &src6, NULL);
1041 	if (error) {
1042 		char ip6buf[INET6_ADDRSTRLEN];
1043 		nd6log((LOG_DEBUG, "nd6_na_output: source can't be "
1044 		    "determined: dst=%s, error=%d\n",
1045 		    ip6_sprintf(ip6buf, &daddr6), error));
1046 		goto bad;
1047 	}
1048 	ip6->ip6_src = src6;
1049 	nd_na = (struct nd_neighbor_advert *)(ip6 + 1);
1050 	nd_na->nd_na_type = ND_NEIGHBOR_ADVERT;
1051 	nd_na->nd_na_code = 0;
1052 	nd_na->nd_na_target = *taddr6;
1053 	in6_clearscope(&nd_na->nd_na_target); /* XXX */
1054 
1055 	/*
1056 	 * "tlladdr" indicates NS's condition for adding tlladdr or not.
1057 	 * see nd6_ns_input() for details.
1058 	 * Basically, if NS packet is sent to unicast/anycast addr,
1059 	 * target lladdr option SHOULD NOT be included.
1060 	 */
1061 	if (tlladdr) {
1062 		/*
1063 		 * sdl0 != NULL indicates proxy NA.  If we do proxy, use
1064 		 * lladdr in sdl0.  If we are not proxying (sending NA for
1065 		 * my address) use lladdr configured for the interface.
1066 		 */
1067 		if (sdl0 == NULL) {
1068 			if (ifp->if_carp)
1069 				mac = (*carp_macmatch6_p)(ifp, m, taddr6);
1070 			if (mac == NULL)
1071 				mac = nd6_ifptomac(ifp);
1072 		} else if (sdl0->sa_family == AF_LINK) {
1073 			struct sockaddr_dl *sdl;
1074 			sdl = (struct sockaddr_dl *)sdl0;
1075 			if (sdl->sdl_alen == ifp->if_addrlen)
1076 				mac = LLADDR(sdl);
1077 		}
1078 	}
1079 	if (tlladdr && mac) {
1080 		int optlen = sizeof(struct nd_opt_hdr) + ifp->if_addrlen;
1081 		struct nd_opt_hdr *nd_opt = (struct nd_opt_hdr *)(nd_na + 1);
1082 
1083 		/* roundup to 8 bytes alignment! */
1084 		optlen = (optlen + 7) & ~7;
1085 
1086 		m->m_pkthdr.len += optlen;
1087 		m->m_len += optlen;
1088 		icmp6len += optlen;
1089 		bzero((caddr_t)nd_opt, optlen);
1090 		nd_opt->nd_opt_type = ND_OPT_TARGET_LINKADDR;
1091 		nd_opt->nd_opt_len = optlen >> 3;
1092 		bcopy(mac, (caddr_t)(nd_opt + 1), ifp->if_addrlen);
1093 	} else
1094 		flags &= ~ND_NA_FLAG_OVERRIDE;
1095 
1096 	ip6->ip6_plen = htons((u_short)icmp6len);
1097 	nd_na->nd_na_flags_reserved = flags;
1098 	nd_na->nd_na_cksum = 0;
1099 	nd_na->nd_na_cksum =
1100 	    in6_cksum(m, IPPROTO_ICMPV6, sizeof(struct ip6_hdr), icmp6len);
1101 
1102 	if (send_sendso_input_hook != NULL) {
1103 		mtag = m_tag_get(PACKET_TAG_ND_OUTGOING,
1104 		    sizeof(unsigned short), M_NOWAIT);
1105 		if (mtag == NULL)
1106 			goto bad;
1107 		*(unsigned short *)(mtag + 1) = nd_na->nd_na_type;
1108 		m_tag_prepend(m, mtag);
1109 	}
1110 
1111 	ip6_output(m, NULL, NULL, 0, &im6o, NULL, NULL);
1112 	icmp6_ifstat_inc(ifp, ifs6_out_msg);
1113 	icmp6_ifstat_inc(ifp, ifs6_out_neighboradvert);
1114 	ICMP6STAT_INC(icp6s_outhist[ND_NEIGHBOR_ADVERT]);
1115 
1116 	return;
1117 
1118   bad:
1119 	m_freem(m);
1120 }
1121 
1122 #ifndef BURN_BRIDGES
1123 void
1124 nd6_na_output(struct ifnet *ifp, const struct in6_addr *daddr6_0,
1125     const struct in6_addr *taddr6, u_long flags, int tlladdr,
1126     struct sockaddr *sdl0)
1127 {
1128 
1129 	nd6_na_output_fib(ifp, daddr6_0, taddr6, flags, tlladdr, sdl0,
1130 	    RT_DEFAULT_FIB);
1131 }
1132 #endif
1133 
1134 caddr_t
1135 nd6_ifptomac(struct ifnet *ifp)
1136 {
1137 	switch (ifp->if_type) {
1138 	case IFT_ETHER:
1139 	case IFT_IEEE1394:
1140 	case IFT_L2VLAN:
1141 	case IFT_INFINIBAND:
1142 	case IFT_BRIDGE:
1143 		return IF_LLADDR(ifp);
1144 	default:
1145 		return NULL;
1146 	}
1147 }
1148 
1149 struct dadq {
1150 	TAILQ_ENTRY(dadq) dad_list;
1151 	struct ifaddr *dad_ifa;
1152 	int dad_count;		/* max NS to send */
1153 	int dad_ns_tcount;	/* # of trials to send NS */
1154 	int dad_ns_ocount;	/* NS sent so far */
1155 	int dad_ns_icount;
1156 	int dad_na_icount;
1157 	int dad_ns_lcount;	/* looped back NS */
1158 	int dad_loopbackprobe;	/* probing state for loopback detection */
1159 	struct callout dad_timer_ch;
1160 	struct vnet *dad_vnet;
1161 	u_int dad_refcnt;
1162 #define	ND_OPT_NONCE_LEN32 \
1163 		((ND_OPT_NONCE_LEN + sizeof(uint32_t) - 1)/sizeof(uint32_t))
1164 	uint32_t dad_nonce[ND_OPT_NONCE_LEN32];
1165 	bool dad_ondadq;	/* on dadq? Protected by DADQ_WLOCK. */
1166 };
1167 
1168 VNET_DEFINE_STATIC(TAILQ_HEAD(, dadq), dadq);
1169 VNET_DEFINE_STATIC(struct rwlock, dad_rwlock);
1170 #define	V_dadq			VNET(dadq)
1171 #define	V_dad_rwlock		VNET(dad_rwlock)
1172 
1173 #define	DADQ_LOCKPTR()		(&V_dad_rwlock)
1174 #define	DADQ_LOCK_INIT()	rw_init(DADQ_LOCKPTR(), "nd6 DAD queue")
1175 #define	DADQ_RLOCK()		rw_rlock(DADQ_LOCKPTR())
1176 #define	DADQ_RUNLOCK()		rw_runlock(DADQ_LOCKPTR())
1177 #define	DADQ_WLOCK()		rw_wlock(DADQ_LOCKPTR())
1178 #define	DADQ_WUNLOCK()		rw_wunlock(DADQ_LOCKPTR())
1179 
1180 #define	DADQ_LOCK_ASSERT()	rw_assert(DADQ_LOCKPTR(), RA_LOCKED);
1181 #define	DADQ_RLOCK_ASSERT()	rw_assert(DADQ_LOCKPTR(), RA_RLOCKED);
1182 #define	DADQ_WLOCK_ASSERT()	rw_assert(DADQ_LOCKPTR(), RA_WLOCKED);
1183 
1184 static void
1185 nd6_dad_add(struct dadq *dp)
1186 {
1187 	DADQ_WLOCK_ASSERT();
1188 
1189 	TAILQ_INSERT_TAIL(&V_dadq, dp, dad_list);
1190 	dp->dad_ondadq = true;
1191 }
1192 
1193 static void
1194 nd6_dad_del(struct dadq *dp)
1195 {
1196 	DADQ_WLOCK_ASSERT();
1197 
1198 	if (dp->dad_ondadq) {
1199 		/*
1200 		 * Remove dp from the dadq and release the dadq's
1201 		 * reference.
1202 		 */
1203 		TAILQ_REMOVE(&V_dadq, dp, dad_list);
1204 		dp->dad_ondadq = false;
1205 		nd6_dad_rele(dp);
1206 	}
1207 }
1208 
1209 static struct dadq *
1210 nd6_dad_find(struct ifaddr *ifa, struct nd_opt_nonce *n)
1211 {
1212 	struct dadq *dp;
1213 
1214 	DADQ_LOCK_ASSERT();
1215 
1216 	TAILQ_FOREACH(dp, &V_dadq, dad_list) {
1217 		if (dp->dad_ifa != ifa)
1218 			continue;
1219 
1220 		/*
1221 		 * Skip if the nonce matches the received one.
1222 		 * +2 in the length is required because of type and
1223 		 * length fields are included in a header.
1224 		 */
1225 		if (n != NULL &&
1226 		    n->nd_opt_nonce_len == (ND_OPT_NONCE_LEN + 2) / 8 &&
1227 		    memcmp(&n->nd_opt_nonce[0], &dp->dad_nonce[0],
1228 		    ND_OPT_NONCE_LEN) == 0) {
1229 			dp->dad_ns_lcount++;
1230 			continue;
1231 		}
1232 		break;
1233 	}
1234 
1235 	return (dp);
1236 }
1237 
1238 static void
1239 nd6_dad_starttimer(struct dadq *dp, int ticks)
1240 {
1241 	DADQ_WLOCK_ASSERT();
1242 
1243 	callout_reset(&dp->dad_timer_ch, ticks, nd6_dad_timer, dp);
1244 }
1245 
1246 static void
1247 nd6_dad_stoptimer(struct dadq *dp)
1248 {
1249 	callout_drain(&dp->dad_timer_ch);
1250 }
1251 
1252 static void
1253 nd6_dad_rele(struct dadq *dp)
1254 {
1255 	if (refcount_release(&dp->dad_refcnt)) {
1256 		KASSERT(!dp->dad_ondadq, ("dp %p still on DAD queue", dp));
1257 		ifa_free(dp->dad_ifa);
1258 		free(dp, M_IP6NDP);
1259 	}
1260 }
1261 
1262 void
1263 nd6_dad_init(void)
1264 {
1265 	DADQ_LOCK_INIT();
1266 	TAILQ_INIT(&V_dadq);
1267 }
1268 
1269 /*
1270  * Start Duplicate Address Detection (DAD) for specified interface address.
1271  */
1272 void
1273 nd6_dad_start(struct ifaddr *ifa, int delay)
1274 {
1275 	struct in6_ifaddr *ia = (struct in6_ifaddr *)ifa;
1276 	struct dadq *dp;
1277 	char ip6buf[INET6_ADDRSTRLEN];
1278 
1279 	KASSERT((ia->ia6_flags & IN6_IFF_TENTATIVE) != 0,
1280 	    ("starting DAD on non-tentative address %p", ifa));
1281 
1282 	/*
1283 	 * If we don't need DAD, don't do it.
1284 	 * There are several cases:
1285 	 * - DAD is disabled globally or on the interface
1286 	 * - the interface address is anycast
1287 	 */
1288 	if ((ia->ia6_flags & IN6_IFF_ANYCAST) != 0 ||
1289 	    V_ip6_dad_count == 0 ||
1290 	    (ND_IFINFO(ifa->ifa_ifp)->flags & ND6_IFF_NO_DAD) != 0) {
1291 		ia->ia6_flags &= ~IN6_IFF_TENTATIVE;
1292 		return;
1293 	}
1294 	if ((ifa->ifa_ifp->if_flags & IFF_UP) == 0 ||
1295 	    (ifa->ifa_ifp->if_drv_flags & IFF_DRV_RUNNING) == 0 ||
1296 	    (ND_IFINFO(ifa->ifa_ifp)->flags & ND6_IFF_IFDISABLED) != 0)
1297 		return;
1298 
1299 	DADQ_WLOCK();
1300 	if ((dp = nd6_dad_find(ifa, NULL)) != NULL) {
1301 		/*
1302 		 * DAD is already in progress.  Let the existing entry
1303 		 * finish it.
1304 		 */
1305 		DADQ_WUNLOCK();
1306 		return;
1307 	}
1308 
1309 	dp = malloc(sizeof(*dp), M_IP6NDP, M_NOWAIT | M_ZERO);
1310 	if (dp == NULL) {
1311 		log(LOG_ERR, "nd6_dad_start: memory allocation failed for "
1312 			"%s(%s)\n",
1313 			ip6_sprintf(ip6buf, &ia->ia_addr.sin6_addr),
1314 			ifa->ifa_ifp ? if_name(ifa->ifa_ifp) : "???");
1315 		return;
1316 	}
1317 	callout_init_rw(&dp->dad_timer_ch, DADQ_LOCKPTR(),
1318 	    CALLOUT_RETURNUNLOCKED);
1319 #ifdef VIMAGE
1320 	dp->dad_vnet = curvnet;
1321 #endif
1322 	nd6log((LOG_DEBUG, "%s: starting DAD for %s\n", if_name(ifa->ifa_ifp),
1323 	    ip6_sprintf(ip6buf, &ia->ia_addr.sin6_addr)));
1324 
1325 	/*
1326 	 * Send NS packet for DAD, ip6_dad_count times.
1327 	 * Note that we must delay the first transmission, if this is the
1328 	 * first packet to be sent from the interface after interface
1329 	 * (re)initialization.
1330 	 */
1331 	dp->dad_ifa = ifa;
1332 	ifa_ref(dp->dad_ifa);
1333 	dp->dad_count = V_ip6_dad_count;
1334 	dp->dad_ns_icount = dp->dad_na_icount = 0;
1335 	dp->dad_ns_ocount = dp->dad_ns_tcount = 0;
1336 	dp->dad_ns_lcount = dp->dad_loopbackprobe = 0;
1337 
1338 	/* Add this to the dadq and add a reference for the dadq. */
1339 	refcount_init(&dp->dad_refcnt, 1);
1340 	nd6_dad_add(dp);
1341 	nd6_dad_starttimer(dp, delay);
1342 	DADQ_WUNLOCK();
1343 }
1344 
1345 /*
1346  * terminate DAD unconditionally.  used for address removals.
1347  */
1348 void
1349 nd6_dad_stop(struct ifaddr *ifa)
1350 {
1351 	struct dadq *dp;
1352 
1353 	DADQ_WLOCK();
1354 	dp = nd6_dad_find(ifa, NULL);
1355 	if (dp == NULL) {
1356 		DADQ_WUNLOCK();
1357 		/* DAD wasn't started yet */
1358 		return;
1359 	}
1360 
1361 	/*
1362 	 * Acquire a temporary reference so that we can safely stop the callout.
1363 	 */
1364 	(void)refcount_acquire(&dp->dad_refcnt);
1365 	nd6_dad_del(dp);
1366 	DADQ_WUNLOCK();
1367 
1368 	nd6_dad_stoptimer(dp);
1369 	nd6_dad_rele(dp);
1370 }
1371 
1372 static void
1373 nd6_dad_timer(void *arg)
1374 {
1375 	struct dadq *dp = arg;
1376 	struct ifaddr *ifa = dp->dad_ifa;
1377 	struct ifnet *ifp = dp->dad_ifa->ifa_ifp;
1378 	struct in6_ifaddr *ia = (struct in6_ifaddr *)ifa;
1379 	char ip6buf[INET6_ADDRSTRLEN];
1380 	struct epoch_tracker et;
1381 
1382 	CURVNET_SET(dp->dad_vnet);
1383 	KASSERT(ia != NULL, ("DAD entry %p with no address", dp));
1384 
1385 	NET_EPOCH_ENTER(et);
1386 	if (ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED) {
1387 		/* Do not need DAD for ifdisabled interface. */
1388 		log(LOG_ERR, "nd6_dad_timer: cancel DAD on %s because of "
1389 		    "ND6_IFF_IFDISABLED.\n", ifp->if_xname);
1390 		goto err;
1391 	}
1392 	if (ia->ia6_flags & IN6_IFF_DUPLICATED) {
1393 		log(LOG_ERR, "nd6_dad_timer: called with duplicated address "
1394 			"%s(%s)\n",
1395 			ip6_sprintf(ip6buf, &ia->ia_addr.sin6_addr),
1396 			ifa->ifa_ifp ? if_name(ifa->ifa_ifp) : "???");
1397 		goto err;
1398 	}
1399 	if ((ia->ia6_flags & IN6_IFF_TENTATIVE) == 0) {
1400 		log(LOG_ERR, "nd6_dad_timer: called with non-tentative address "
1401 			"%s(%s)\n",
1402 			ip6_sprintf(ip6buf, &ia->ia_addr.sin6_addr),
1403 			ifa->ifa_ifp ? if_name(ifa->ifa_ifp) : "???");
1404 		goto err;
1405 	}
1406 
1407 	/* Stop DAD if the interface is down even after dad_maxtry attempts. */
1408 	if ((dp->dad_ns_tcount > V_dad_maxtry) &&
1409 	    (((ifp->if_flags & IFF_UP) == 0) ||
1410 	     ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0))) {
1411 		nd6log((LOG_INFO, "%s: could not run DAD "
1412 		    "because the interface was down or not running.\n",
1413 		    if_name(ifa->ifa_ifp)));
1414 		goto err;
1415 	}
1416 
1417 	/* Need more checks? */
1418 	if (dp->dad_ns_ocount < dp->dad_count) {
1419 		/*
1420 		 * We have more NS to go.  Send NS packet for DAD.
1421 		 */
1422 		nd6_dad_starttimer(dp,
1423 		    (long)ND_IFINFO(ifa->ifa_ifp)->retrans * hz / 1000);
1424 		nd6_dad_ns_output(dp);
1425 		goto done;
1426 	} else {
1427 		/*
1428 		 * We have transmitted sufficient number of DAD packets.
1429 		 * See what we've got.
1430 		 */
1431 		if (dp->dad_ns_icount > 0 || dp->dad_na_icount > 0) {
1432 			/* We've seen NS or NA, means DAD has failed. */
1433 			nd6_dad_duplicated(ifa, dp);
1434 		} else if (V_dad_enhanced != 0 &&
1435 		    dp->dad_ns_lcount > 0 &&
1436 		    dp->dad_ns_lcount > dp->dad_loopbackprobe) {
1437 			/*
1438 			 * Sec. 4.1 in RFC 7527 requires transmission of
1439 			 * additional probes until the loopback condition
1440 			 * becomes clear when a looped back probe is detected.
1441 			 */
1442 			log(LOG_ERR, "%s: a looped back NS message is "
1443 			    "detected during DAD for %s.  "
1444 			    "Another DAD probes are being sent.\n",
1445 			    if_name(ifa->ifa_ifp),
1446 			    ip6_sprintf(ip6buf, IFA_IN6(ifa)));
1447 			dp->dad_loopbackprobe = dp->dad_ns_lcount;
1448 			/*
1449 			 * Send an NS immediately and increase dad_count by
1450 			 * V_nd6_mmaxtries - 1.
1451 			 */
1452 			dp->dad_count =
1453 			    dp->dad_ns_ocount + V_nd6_mmaxtries - 1;
1454 			nd6_dad_starttimer(dp,
1455 			    (long)ND_IFINFO(ifa->ifa_ifp)->retrans * hz / 1000);
1456 			nd6_dad_ns_output(dp);
1457 			goto done;
1458 		} else {
1459 			/*
1460 			 * We are done with DAD.  No NA came, no NS came.
1461 			 * No duplicate address found.  Check IFDISABLED flag
1462 			 * again in case that it is changed between the
1463 			 * beginning of this function and here.
1464 			 */
1465 			if ((ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED) == 0)
1466 				ia->ia6_flags &= ~IN6_IFF_TENTATIVE;
1467 
1468 			nd6log((LOG_DEBUG,
1469 			    "%s: DAD complete for %s - no duplicates found\n",
1470 			    if_name(ifa->ifa_ifp),
1471 			    ip6_sprintf(ip6buf, &ia->ia_addr.sin6_addr)));
1472 			if (dp->dad_ns_lcount > 0)
1473 				log(LOG_ERR, "%s: DAD completed while "
1474 				    "a looped back NS message is detected "
1475 				    "during DAD for %s.\n",
1476 				    if_name(ifa->ifa_ifp),
1477 				    ip6_sprintf(ip6buf, IFA_IN6(ifa)));
1478 		}
1479 	}
1480 err:
1481 	nd6_dad_del(dp);
1482 	DADQ_WUNLOCK();
1483 done:
1484 	NET_EPOCH_EXIT(et);
1485 	CURVNET_RESTORE();
1486 }
1487 
1488 static void
1489 nd6_dad_duplicated(struct ifaddr *ifa, struct dadq *dp)
1490 {
1491 	struct in6_ifaddr *ia = (struct in6_ifaddr *)ifa;
1492 	struct ifnet *ifp;
1493 	char ip6buf[INET6_ADDRSTRLEN];
1494 
1495 	log(LOG_ERR, "%s: DAD detected duplicate IPv6 address %s: "
1496 	    "NS in/out/loopback=%d/%d/%d, NA in=%d\n",
1497 	    if_name(ifa->ifa_ifp), ip6_sprintf(ip6buf, &ia->ia_addr.sin6_addr),
1498 	    dp->dad_ns_icount, dp->dad_ns_ocount, dp->dad_ns_lcount,
1499 	    dp->dad_na_icount);
1500 
1501 	ia->ia6_flags &= ~IN6_IFF_TENTATIVE;
1502 	ia->ia6_flags |= IN6_IFF_DUPLICATED;
1503 
1504 	ifp = ifa->ifa_ifp;
1505 	log(LOG_ERR, "%s: DAD complete for %s - duplicate found\n",
1506 	    if_name(ifp), ip6_sprintf(ip6buf, &ia->ia_addr.sin6_addr));
1507 	log(LOG_ERR, "%s: manual intervention required\n",
1508 	    if_name(ifp));
1509 
1510 	/*
1511 	 * If the address is a link-local address formed from an interface
1512 	 * identifier based on the hardware address which is supposed to be
1513 	 * uniquely assigned (e.g., EUI-64 for an Ethernet interface), IP
1514 	 * operation on the interface SHOULD be disabled.
1515 	 * [RFC 4862, Section 5.4.5]
1516 	 */
1517 	if (IN6_IS_ADDR_LINKLOCAL(&ia->ia_addr.sin6_addr)) {
1518 		struct in6_addr in6;
1519 
1520 		/*
1521 		 * To avoid over-reaction, we only apply this logic when we are
1522 		 * very sure that hardware addresses are supposed to be unique.
1523 		 */
1524 		switch (ifp->if_type) {
1525 		case IFT_ETHER:
1526 		case IFT_ATM:
1527 		case IFT_IEEE1394:
1528 		case IFT_INFINIBAND:
1529 			in6 = ia->ia_addr.sin6_addr;
1530 			if (in6_get_hw_ifid(ifp, &in6) == 0 &&
1531 			    IN6_ARE_ADDR_EQUAL(&ia->ia_addr.sin6_addr, &in6)) {
1532 				ND_IFINFO(ifp)->flags |= ND6_IFF_IFDISABLED;
1533 				log(LOG_ERR, "%s: possible hardware address "
1534 				    "duplication detected, disable IPv6\n",
1535 				    if_name(ifp));
1536 			}
1537 			break;
1538 		}
1539 	}
1540 }
1541 
1542 /*
1543  * Transmit a neighbour solicitation for the purpose of DAD.  Returns with the
1544  * DAD queue unlocked.
1545  */
1546 static void
1547 nd6_dad_ns_output(struct dadq *dp)
1548 {
1549 	struct in6_ifaddr *ia = (struct in6_ifaddr *)dp->dad_ifa;
1550 	struct ifnet *ifp = dp->dad_ifa->ifa_ifp;
1551 	int i;
1552 
1553 	DADQ_WLOCK_ASSERT();
1554 
1555 	dp->dad_ns_tcount++;
1556 	if ((ifp->if_flags & IFF_UP) == 0) {
1557 		DADQ_WUNLOCK();
1558 		return;
1559 	}
1560 	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) {
1561 		DADQ_WUNLOCK();
1562 		return;
1563 	}
1564 
1565 	dp->dad_ns_ocount++;
1566 	if (V_dad_enhanced != 0) {
1567 		for (i = 0; i < ND_OPT_NONCE_LEN32; i++)
1568 			dp->dad_nonce[i] = arc4random();
1569 		/*
1570 		 * XXXHRS: Note that in the case that
1571 		 * DupAddrDetectTransmits > 1, multiple NS messages with
1572 		 * different nonces can be looped back in an unexpected
1573 		 * order.  The current implementation recognizes only
1574 		 * the latest nonce on the sender side.  Practically it
1575 		 * should work well in almost all cases.
1576 		 */
1577 	}
1578 	DADQ_WUNLOCK();
1579 	nd6_ns_output(ifp, NULL, NULL, &ia->ia_addr.sin6_addr,
1580 	    (uint8_t *)&dp->dad_nonce[0]);
1581 }
1582 
1583 static void
1584 nd6_dad_ns_input(struct ifaddr *ifa, struct nd_opt_nonce *ndopt_nonce)
1585 {
1586 	struct dadq *dp;
1587 
1588 	if (ifa == NULL)
1589 		panic("ifa == NULL in nd6_dad_ns_input");
1590 
1591 	/* Ignore Nonce option when Enhanced DAD is disabled. */
1592 	if (V_dad_enhanced == 0)
1593 		ndopt_nonce = NULL;
1594 	DADQ_RLOCK();
1595 	dp = nd6_dad_find(ifa, ndopt_nonce);
1596 	if (dp != NULL)
1597 		dp->dad_ns_icount++;
1598 	DADQ_RUNLOCK();
1599 }
1600 
1601 static void
1602 nd6_dad_na_input(struct ifaddr *ifa)
1603 {
1604 	struct dadq *dp;
1605 
1606 	if (ifa == NULL)
1607 		panic("ifa == NULL in nd6_dad_na_input");
1608 
1609 	DADQ_RLOCK();
1610 	dp = nd6_dad_find(ifa, NULL);
1611 	if (dp != NULL)
1612 		dp->dad_na_icount++;
1613 	DADQ_RUNLOCK();
1614 }
1615