xref: /freebsd/sys/netinet6/nd6_nbr.c (revision 19261079b74319502c6ffa1249920079f0f69a72)
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/route.h>
65 #include <net/vnet.h>
66 
67 #include <netinet/in.h>
68 #include <netinet/in_var.h>
69 #include <net/if_llatbl.h>
70 #include <netinet6/in6_var.h>
71 #include <netinet6/in6_ifattach.h>
72 #include <netinet/ip6.h>
73 #include <netinet6/ip6_var.h>
74 #include <netinet6/scope6_var.h>
75 #include <netinet6/nd6.h>
76 #include <netinet/icmp6.h>
77 #include <netinet/ip_carp.h>
78 #include <netinet6/send.h>
79 
80 #define SDL(s) ((struct sockaddr_dl *)s)
81 
82 struct dadq;
83 static struct dadq *nd6_dad_find(struct ifaddr *, struct nd_opt_nonce *);
84 static void nd6_dad_add(struct dadq *dp);
85 static void nd6_dad_del(struct dadq *dp);
86 static void nd6_dad_rele(struct dadq *);
87 static void nd6_dad_starttimer(struct dadq *, int);
88 static void nd6_dad_stoptimer(struct dadq *);
89 static void nd6_dad_timer(void *);
90 static void nd6_dad_duplicated(struct ifaddr *, struct dadq *);
91 static void nd6_dad_ns_output(struct dadq *);
92 static void nd6_dad_ns_input(struct ifaddr *, struct nd_opt_nonce *);
93 static void nd6_dad_na_input(struct ifaddr *);
94 static void nd6_na_output_fib(struct ifnet *, const struct in6_addr *,
95     const struct in6_addr *, u_long, int, struct sockaddr *, u_int);
96 static void nd6_ns_output_fib(struct ifnet *, const struct in6_addr *,
97     const struct in6_addr *, const struct in6_addr *, uint8_t *, u_int);
98 
99 VNET_DEFINE_STATIC(int, dad_enhanced) = 1;
100 #define	V_dad_enhanced			VNET(dad_enhanced)
101 
102 SYSCTL_DECL(_net_inet6_ip6);
103 SYSCTL_INT(_net_inet6_ip6, OID_AUTO, dad_enhanced, CTLFLAG_VNET | CTLFLAG_RW,
104     &VNET_NAME(dad_enhanced), 0,
105     "Enable Enhanced DAD, which adds a random nonce to NS messages for DAD.");
106 
107 VNET_DEFINE_STATIC(int, dad_maxtry) = 15;	/* max # of *tries* to
108 						   transmit DAD packet */
109 #define	V_dad_maxtry			VNET(dad_maxtry)
110 
111 /*
112  * Input a Neighbor Solicitation Message.
113  *
114  * Based on RFC 2461
115  * Based on RFC 2462 (duplicate address detection)
116  */
117 void
118 nd6_ns_input(struct mbuf *m, int off, int icmp6len)
119 {
120 	struct ifnet *ifp;
121 	struct ip6_hdr *ip6;
122 	struct nd_neighbor_solicit *nd_ns;
123 	struct in6_addr daddr6, myaddr6, saddr6, taddr6;
124 	struct ifaddr *ifa;
125 	struct sockaddr_dl proxydl;
126 	union nd_opts ndopts;
127 	char ip6bufs[INET6_ADDRSTRLEN], ip6bufd[INET6_ADDRSTRLEN];
128 	char *lladdr;
129 	int anycast, lladdrlen, proxy, rflag, tentative, tlladdr;
130 
131 	ifa = NULL;
132 
133 	/* RFC 6980: Nodes MUST silently ignore fragments */
134 	if(m->m_flags & M_FRAGMENTED)
135 		goto freeit;
136 
137 	ifp = m->m_pkthdr.rcvif;
138 	ip6 = mtod(m, struct ip6_hdr *);
139 	if (__predict_false(ip6->ip6_hlim != 255)) {
140 		ICMP6STAT_INC(icp6s_invlhlim);
141 		nd6log((LOG_ERR,
142 		    "nd6_ns_input: invalid hlim (%d) from %s to %s on %s\n",
143 		    ip6->ip6_hlim, ip6_sprintf(ip6bufs, &ip6->ip6_src),
144 		    ip6_sprintf(ip6bufd, &ip6->ip6_dst), if_name(ifp)));
145 		goto bads;
146 	}
147 
148 	if (m->m_len < off + icmp6len) {
149 		m = m_pullup(m, off + icmp6len);
150 		if (m == NULL) {
151 			IP6STAT_INC(ip6s_exthdrtoolong);
152 			return;
153 		}
154 	}
155 	ip6 = mtod(m, struct ip6_hdr *);
156 	nd_ns = (struct nd_neighbor_solicit *)((caddr_t)ip6 + off);
157 
158 	saddr6 = ip6->ip6_src;
159 	daddr6 = ip6->ip6_dst;
160 	taddr6 = nd_ns->nd_ns_target;
161 	if (in6_setscope(&taddr6, ifp, NULL) != 0)
162 		goto bad;
163 
164 	rflag = (V_ip6_forwarding) ? ND_NA_FLAG_ROUTER : 0;
165 	if (ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV && V_ip6_norbit_raif)
166 		rflag = 0;
167 
168 	if (IN6_IS_ADDR_UNSPECIFIED(&saddr6)) {
169 		/* dst has to be a solicited node multicast address. */
170 		if (daddr6.s6_addr16[0] == IPV6_ADDR_INT16_MLL &&
171 		    /* don't check ifindex portion */
172 		    daddr6.s6_addr32[1] == 0 &&
173 		    daddr6.s6_addr32[2] == IPV6_ADDR_INT32_ONE &&
174 		    daddr6.s6_addr8[12] == 0xff) {
175 			; /* good */
176 		} else {
177 			nd6log((LOG_INFO, "nd6_ns_input: bad DAD packet "
178 			    "(wrong ip6 dst)\n"));
179 			goto bad;
180 		}
181 	} else if (!V_nd6_onlink_ns_rfc4861) {
182 		struct sockaddr_in6 src_sa6;
183 
184 		/*
185 		 * According to recent IETF discussions, it is not a good idea
186 		 * to accept a NS from an address which would not be deemed
187 		 * to be a neighbor otherwise.  This point is expected to be
188 		 * clarified in future revisions of the specification.
189 		 */
190 		bzero(&src_sa6, sizeof(src_sa6));
191 		src_sa6.sin6_family = AF_INET6;
192 		src_sa6.sin6_len = sizeof(src_sa6);
193 		src_sa6.sin6_addr = saddr6;
194 		if (nd6_is_addr_neighbor(&src_sa6, ifp) == 0) {
195 			nd6log((LOG_INFO, "nd6_ns_input: "
196 				"NS packet from non-neighbor\n"));
197 			goto bad;
198 		}
199 	}
200 
201 	if (IN6_IS_ADDR_MULTICAST(&taddr6)) {
202 		nd6log((LOG_INFO, "nd6_ns_input: bad NS target (multicast)\n"));
203 		goto bad;
204 	}
205 
206 	icmp6len -= sizeof(*nd_ns);
207 	nd6_option_init(nd_ns + 1, icmp6len, &ndopts);
208 	if (nd6_options(&ndopts) < 0) {
209 		nd6log((LOG_INFO,
210 		    "nd6_ns_input: invalid ND option, ignored\n"));
211 		/* nd6_options have incremented stats */
212 		goto freeit;
213 	}
214 
215 	lladdr = NULL;
216 	lladdrlen = 0;
217 	if (ndopts.nd_opts_src_lladdr) {
218 		lladdr = (char *)(ndopts.nd_opts_src_lladdr + 1);
219 		lladdrlen = ndopts.nd_opts_src_lladdr->nd_opt_len << 3;
220 	}
221 
222 	if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src) && lladdr) {
223 		nd6log((LOG_INFO, "nd6_ns_input: bad DAD packet "
224 		    "(link-layer address option)\n"));
225 		goto bad;
226 	}
227 
228 	/*
229 	 * Attaching target link-layer address to the NA?
230 	 * (RFC 2461 7.2.4)
231 	 *
232 	 * NS IP dst is unicast/anycast			MUST NOT add
233 	 * NS IP dst is solicited-node multicast	MUST add
234 	 *
235 	 * In implementation, we add target link-layer address by default.
236 	 * We do not add one in MUST NOT cases.
237 	 */
238 	if (!IN6_IS_ADDR_MULTICAST(&daddr6))
239 		tlladdr = 0;
240 	else
241 		tlladdr = 1;
242 
243 	/*
244 	 * Target address (taddr6) must be either:
245 	 * (1) Valid unicast/anycast address for my receiving interface,
246 	 * (2) Unicast address for which I'm offering proxy service, or
247 	 * (3) "tentative" address on which DAD is being performed.
248 	 */
249 	/* (1) and (3) check. */
250 	if (ifp->if_carp)
251 		ifa = (*carp_iamatch6_p)(ifp, &taddr6);
252 	else
253 		ifa = (struct ifaddr *)in6ifa_ifpwithaddr(ifp, &taddr6);
254 
255 	/* (2) check. */
256 	proxy = 0;
257 	if (ifa == NULL) {
258 		struct sockaddr_dl rt_gateway;
259 		struct rt_addrinfo info;
260 		struct sockaddr_in6 dst6;
261 
262 		bzero(&dst6, sizeof(dst6));
263 		dst6.sin6_len = sizeof(struct sockaddr_in6);
264 		dst6.sin6_family = AF_INET6;
265 		dst6.sin6_addr = taddr6;
266 
267 		bzero(&rt_gateway, sizeof(rt_gateway));
268 		rt_gateway.sdl_len = sizeof(rt_gateway);
269 		bzero(&info, sizeof(info));
270 		info.rti_info[RTAX_GATEWAY] = (struct sockaddr *)&rt_gateway;
271 
272 		if (rib_lookup_info(ifp->if_fib, (struct sockaddr *)&dst6,
273 		    0, 0, &info) == 0) {
274 			if ((info.rti_flags & RTF_ANNOUNCE) != 0 &&
275 			    rt_gateway.sdl_family == AF_LINK) {
276 				/*
277 				 * proxy NDP for single entry
278 				 */
279 				proxydl = *SDL(&rt_gateway);
280 				ifa = (struct ifaddr *)in6ifa_ifpforlinklocal(
281 				    ifp, IN6_IFF_NOTREADY|IN6_IFF_ANYCAST);
282 				if (ifa)
283 					proxy = 1;
284 			}
285 		}
286 	}
287 	if (ifa == NULL) {
288 		/*
289 		 * We've got an NS packet, and we don't have that adddress
290 		 * assigned for us.  We MUST silently ignore it.
291 		 * See RFC2461 7.2.3.
292 		 */
293 		goto freeit;
294 	}
295 	myaddr6 = *IFA_IN6(ifa);
296 	anycast = ((struct in6_ifaddr *)ifa)->ia6_flags & IN6_IFF_ANYCAST;
297 	tentative = ((struct in6_ifaddr *)ifa)->ia6_flags & IN6_IFF_TENTATIVE;
298 	if (((struct in6_ifaddr *)ifa)->ia6_flags & IN6_IFF_DUPLICATED)
299 		goto freeit;
300 
301 	if (lladdr && ((ifp->if_addrlen + 2 + 7) & ~7) != lladdrlen) {
302 		nd6log((LOG_INFO, "nd6_ns_input: lladdrlen mismatch for %s "
303 		    "(if %d, NS packet %d)\n",
304 		    ip6_sprintf(ip6bufs, &taddr6),
305 		    ifp->if_addrlen, lladdrlen - 2));
306 		goto bad;
307 	}
308 
309 	if (IN6_ARE_ADDR_EQUAL(&myaddr6, &saddr6)) {
310 		nd6log((LOG_INFO, "nd6_ns_input: duplicate IP6 address %s\n",
311 		    ip6_sprintf(ip6bufs, &saddr6)));
312 		goto freeit;
313 	}
314 
315 	/*
316 	 * We have neighbor solicitation packet, with target address equals to
317 	 * one of my tentative address.
318 	 *
319 	 * src addr	how to process?
320 	 * ---		---
321 	 * multicast	of course, invalid (rejected in ip6_input)
322 	 * unicast	somebody is doing address resolution -> ignore
323 	 * unspec	dup address detection
324 	 *
325 	 * The processing is defined in RFC 2462.
326 	 */
327 	if (tentative) {
328 		/*
329 		 * If source address is unspecified address, it is for
330 		 * duplicate address detection.
331 		 *
332 		 * If not, the packet is for addess resolution;
333 		 * silently ignore it.
334 		 */
335 		if (IN6_IS_ADDR_UNSPECIFIED(&saddr6))
336 			nd6_dad_ns_input(ifa, ndopts.nd_opts_nonce);
337 
338 		goto freeit;
339 	}
340 
341 	/*
342 	 * If the source address is unspecified address, entries must not
343 	 * be created or updated.
344 	 * It looks that sender is performing DAD.  Output NA toward
345 	 * all-node multicast address, to tell the sender that I'm using
346 	 * the address.
347 	 * S bit ("solicited") must be zero.
348 	 */
349 	if (IN6_IS_ADDR_UNSPECIFIED(&saddr6)) {
350 		struct in6_addr in6_all;
351 
352 		in6_all = in6addr_linklocal_allnodes;
353 		if (in6_setscope(&in6_all, ifp, NULL) != 0)
354 			goto bad;
355 		nd6_na_output_fib(ifp, &in6_all, &taddr6,
356 		    ((anycast || proxy || !tlladdr) ? 0 : ND_NA_FLAG_OVERRIDE) |
357 		    rflag, tlladdr, proxy ? (struct sockaddr *)&proxydl : NULL,
358 		    M_GETFIB(m));
359 		goto freeit;
360 	}
361 
362 	nd6_cache_lladdr(ifp, &saddr6, lladdr, lladdrlen,
363 	    ND_NEIGHBOR_SOLICIT, 0);
364 
365 	nd6_na_output_fib(ifp, &saddr6, &taddr6,
366 	    ((anycast || proxy || !tlladdr) ? 0 : ND_NA_FLAG_OVERRIDE) |
367 	    rflag | ND_NA_FLAG_SOLICITED, tlladdr,
368 	    proxy ? (struct sockaddr *)&proxydl : NULL, M_GETFIB(m));
369  freeit:
370 	if (ifa != NULL)
371 		ifa_free(ifa);
372 	m_freem(m);
373 	return;
374 
375  bad:
376 	nd6log((LOG_ERR, "nd6_ns_input: src=%s\n",
377 		ip6_sprintf(ip6bufs, &saddr6)));
378 	nd6log((LOG_ERR, "nd6_ns_input: dst=%s\n",
379 		ip6_sprintf(ip6bufs, &daddr6)));
380 	nd6log((LOG_ERR, "nd6_ns_input: tgt=%s\n",
381 		ip6_sprintf(ip6bufs, &taddr6)));
382  bads:
383 	ICMP6STAT_INC(icp6s_badns);
384 	if (ifa != NULL)
385 		ifa_free(ifa);
386 	m_freem(m);
387 }
388 
389 /*
390  * Output a Neighbor Solicitation Message. Caller specifies:
391  *	- ICMP6 header source IP6 address
392  *	- ND6 header target IP6 address
393  *	- ND6 header source datalink address
394  *
395  * Based on RFC 2461
396  * Based on RFC 2462 (duplicate address detection)
397  *
398  *    ln - for source address determination
399  * nonce - If non-NULL, NS is used for duplicate address detection and
400  *         the value (length is ND_OPT_NONCE_LEN) is used as a random nonce.
401  */
402 static void
403 nd6_ns_output_fib(struct ifnet *ifp, const struct in6_addr *saddr6,
404     const struct in6_addr *daddr6, const struct in6_addr *taddr6,
405     uint8_t *nonce, u_int fibnum)
406 {
407 	struct mbuf *m;
408 	struct m_tag *mtag;
409 	struct ip6_hdr *ip6;
410 	struct nd_neighbor_solicit *nd_ns;
411 	struct ip6_moptions im6o;
412 	int icmp6len;
413 	int maxlen;
414 	caddr_t mac;
415 
416 	NET_EPOCH_ASSERT();
417 
418 	if (IN6_IS_ADDR_MULTICAST(taddr6))
419 		return;
420 
421 	/* estimate the size of message */
422 	maxlen = sizeof(*ip6) + sizeof(*nd_ns);
423 	maxlen += (sizeof(struct nd_opt_hdr) + ifp->if_addrlen + 7) & ~7;
424 	KASSERT(max_linkhdr + maxlen <= MCLBYTES, (
425 	    "%s: max_linkhdr + maxlen > MCLBYTES (%d + %d > %d)",
426 	    __func__, max_linkhdr, maxlen, MCLBYTES));
427 
428 	if (max_linkhdr + maxlen > MHLEN)
429 		m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR);
430 	else
431 		m = m_gethdr(M_NOWAIT, MT_DATA);
432 	if (m == NULL)
433 		return;
434 	M_SETFIB(m, fibnum);
435 
436 	if (daddr6 == NULL || IN6_IS_ADDR_MULTICAST(daddr6)) {
437 		m->m_flags |= M_MCAST;
438 		im6o.im6o_multicast_ifp = ifp;
439 		im6o.im6o_multicast_hlim = 255;
440 		im6o.im6o_multicast_loop = 0;
441 	}
442 
443 	icmp6len = sizeof(*nd_ns);
444 	m->m_pkthdr.len = m->m_len = sizeof(*ip6) + icmp6len;
445 	m->m_data += max_linkhdr;	/* or M_ALIGN() equivalent? */
446 
447 	/* fill neighbor solicitation packet */
448 	ip6 = mtod(m, struct ip6_hdr *);
449 	ip6->ip6_flow = 0;
450 	ip6->ip6_vfc &= ~IPV6_VERSION_MASK;
451 	ip6->ip6_vfc |= IPV6_VERSION;
452 	/* ip6->ip6_plen will be set later */
453 	ip6->ip6_nxt = IPPROTO_ICMPV6;
454 	ip6->ip6_hlim = 255;
455 	if (daddr6)
456 		ip6->ip6_dst = *daddr6;
457 	else {
458 		ip6->ip6_dst.s6_addr16[0] = IPV6_ADDR_INT16_MLL;
459 		ip6->ip6_dst.s6_addr16[1] = 0;
460 		ip6->ip6_dst.s6_addr32[1] = 0;
461 		ip6->ip6_dst.s6_addr32[2] = IPV6_ADDR_INT32_ONE;
462 		ip6->ip6_dst.s6_addr32[3] = taddr6->s6_addr32[3];
463 		ip6->ip6_dst.s6_addr8[12] = 0xff;
464 		if (in6_setscope(&ip6->ip6_dst, ifp, NULL) != 0)
465 			goto bad;
466 	}
467 	if (nonce == NULL) {
468 		struct ifaddr *ifa = NULL;
469 
470 		/*
471 		 * RFC2461 7.2.2:
472 		 * "If the source address of the packet prompting the
473 		 * solicitation is the same as one of the addresses assigned
474 		 * to the outgoing interface, that address SHOULD be placed
475 		 * in the IP Source Address of the outgoing solicitation.
476 		 * Otherwise, any one of the addresses assigned to the
477 		 * interface should be used."
478 		 *
479 		 * We use the source address for the prompting packet
480 		 * (saddr6), if saddr6 belongs to the outgoing interface.
481 		 * Otherwise, we perform the source address selection as usual.
482 		 */
483 
484 		if (saddr6 != NULL)
485 			ifa = (struct ifaddr *)in6ifa_ifpwithaddr(ifp, saddr6);
486 		if (ifa != NULL) {
487 			/* ip6_src set already. */
488 			ip6->ip6_src = *saddr6;
489 			ifa_free(ifa);
490 		} else {
491 			int error;
492 			struct in6_addr dst6, src6;
493 			uint32_t scopeid;
494 
495 			in6_splitscope(&ip6->ip6_dst, &dst6, &scopeid);
496 			error = in6_selectsrc_addr(fibnum, &dst6,
497 			    scopeid, ifp, &src6, NULL);
498 			if (error) {
499 				char ip6buf[INET6_ADDRSTRLEN];
500 				nd6log((LOG_DEBUG, "%s: source can't be "
501 				    "determined: dst=%s, error=%d\n", __func__,
502 				    ip6_sprintf(ip6buf, &dst6),
503 				    error));
504 				goto bad;
505 			}
506 			ip6->ip6_src = src6;
507 		}
508 	} else {
509 		/*
510 		 * Source address for DAD packet must always be IPv6
511 		 * unspecified address. (0::0)
512 		 * We actually don't have to 0-clear the address (we did it
513 		 * above), but we do so here explicitly to make the intention
514 		 * clearer.
515 		 */
516 		bzero(&ip6->ip6_src, sizeof(ip6->ip6_src));
517 	}
518 	nd_ns = (struct nd_neighbor_solicit *)(ip6 + 1);
519 	nd_ns->nd_ns_type = ND_NEIGHBOR_SOLICIT;
520 	nd_ns->nd_ns_code = 0;
521 	nd_ns->nd_ns_reserved = 0;
522 	nd_ns->nd_ns_target = *taddr6;
523 	in6_clearscope(&nd_ns->nd_ns_target); /* XXX */
524 
525 	/*
526 	 * Add source link-layer address option.
527 	 *
528 	 *				spec		implementation
529 	 *				---		---
530 	 * DAD packet			MUST NOT	do not add the option
531 	 * there's no link layer address:
532 	 *				impossible	do not add the option
533 	 * there's link layer address:
534 	 *	Multicast NS		MUST add one	add the option
535 	 *	Unicast NS		SHOULD add one	add the option
536 	 */
537 	if (nonce == NULL && (mac = nd6_ifptomac(ifp))) {
538 		int optlen = sizeof(struct nd_opt_hdr) + ifp->if_addrlen;
539 		struct nd_opt_hdr *nd_opt = (struct nd_opt_hdr *)(nd_ns + 1);
540 		/* 8 byte alignments... */
541 		optlen = (optlen + 7) & ~7;
542 
543 		m->m_pkthdr.len += optlen;
544 		m->m_len += optlen;
545 		icmp6len += optlen;
546 		bzero((caddr_t)nd_opt, optlen);
547 		nd_opt->nd_opt_type = ND_OPT_SOURCE_LINKADDR;
548 		nd_opt->nd_opt_len = optlen >> 3;
549 		bcopy(mac, (caddr_t)(nd_opt + 1), ifp->if_addrlen);
550 	}
551 	/*
552 	 * Add a Nonce option (RFC 3971) to detect looped back NS messages.
553 	 * This behavior is documented as Enhanced Duplicate Address
554 	 * Detection in RFC 7527.
555 	 * net.inet6.ip6.dad_enhanced=0 disables this.
556 	 */
557 	if (V_dad_enhanced != 0 && nonce != NULL) {
558 		int optlen = sizeof(struct nd_opt_hdr) + ND_OPT_NONCE_LEN;
559 		struct nd_opt_hdr *nd_opt = (struct nd_opt_hdr *)(nd_ns + 1);
560 		/* 8-byte alignment is required. */
561 		optlen = (optlen + 7) & ~7;
562 
563 		m->m_pkthdr.len += optlen;
564 		m->m_len += optlen;
565 		icmp6len += optlen;
566 		bzero((caddr_t)nd_opt, optlen);
567 		nd_opt->nd_opt_type = ND_OPT_NONCE;
568 		nd_opt->nd_opt_len = optlen >> 3;
569 		bcopy(nonce, (caddr_t)(nd_opt + 1), ND_OPT_NONCE_LEN);
570 	}
571 	ip6->ip6_plen = htons((u_short)icmp6len);
572 	nd_ns->nd_ns_cksum = 0;
573 	nd_ns->nd_ns_cksum =
574 	    in6_cksum(m, IPPROTO_ICMPV6, sizeof(*ip6), icmp6len);
575 
576 	if (send_sendso_input_hook != NULL) {
577 		mtag = m_tag_get(PACKET_TAG_ND_OUTGOING,
578 			sizeof(unsigned short), M_NOWAIT);
579 		if (mtag == NULL)
580 			goto bad;
581 		*(unsigned short *)(mtag + 1) = nd_ns->nd_ns_type;
582 		m_tag_prepend(m, mtag);
583 	}
584 
585 	ip6_output(m, NULL, NULL, (nonce != NULL) ? IPV6_UNSPECSRC : 0,
586 	    &im6o, NULL, NULL);
587 	icmp6_ifstat_inc(ifp, ifs6_out_msg);
588 	icmp6_ifstat_inc(ifp, ifs6_out_neighborsolicit);
589 	ICMP6STAT_INC(icp6s_outhist[ND_NEIGHBOR_SOLICIT]);
590 
591 	return;
592 
593   bad:
594 	m_freem(m);
595 }
596 
597 #ifndef BURN_BRIDGES
598 void
599 nd6_ns_output(struct ifnet *ifp, const struct in6_addr *saddr6,
600     const struct in6_addr *daddr6, const struct in6_addr *taddr6,uint8_t *nonce)
601 {
602 
603 	nd6_ns_output_fib(ifp, saddr6, daddr6, taddr6, nonce, RT_DEFAULT_FIB);
604 }
605 #endif
606 /*
607  * Neighbor advertisement input handling.
608  *
609  * Based on RFC 2461
610  * Based on RFC 2462 (duplicate address detection)
611  *
612  * the following items are not implemented yet:
613  * - proxy advertisement delay rule (RFC2461 7.2.8, last paragraph, SHOULD)
614  * - anycast advertisement delay rule (RFC2461 7.2.7, SHOULD)
615  */
616 void
617 nd6_na_input(struct mbuf *m, int off, int icmp6len)
618 {
619 	struct ifnet *ifp;
620 	struct ip6_hdr *ip6;
621 	struct ifaddr *ifa;
622 	struct llentry *ln;
623 	struct mbuf *chain;
624 	struct nd_neighbor_advert *nd_na;
625 	struct in6_addr daddr6, taddr6;
626 	union nd_opts ndopts;
627 	u_char linkhdr[LLE_MAX_LINKHDR];
628 	char ip6bufs[INET6_ADDRSTRLEN], ip6bufd[INET6_ADDRSTRLEN];
629 	char *lladdr;
630 	size_t linkhdrsize;
631 	int flags, is_override, is_router, is_solicited;
632 	int lladdr_off, lladdrlen, checklink;
633 	bool flush_holdchain = false;
634 
635 	NET_EPOCH_ASSERT();
636 
637 	chain = NULL;
638 	ln = NULL;
639 	checklink = 0;
640 
641 	/* RFC 6980: Nodes MUST silently ignore fragments */
642 	if(m->m_flags & M_FRAGMENTED)
643 		goto freeit;
644 
645 	ifp = m->m_pkthdr.rcvif;
646 	ip6 = mtod(m, struct ip6_hdr *);
647 	if (__predict_false(ip6->ip6_hlim != 255)) {
648 		ICMP6STAT_INC(icp6s_invlhlim);
649 		nd6log((LOG_ERR,
650 		    "nd6_na_input: invalid hlim (%d) from %s to %s on %s\n",
651 		    ip6->ip6_hlim, ip6_sprintf(ip6bufs, &ip6->ip6_src),
652 		    ip6_sprintf(ip6bufd, &ip6->ip6_dst), if_name(ifp)));
653 		goto bad;
654 	}
655 
656 	if (m->m_len < off + icmp6len) {
657 		m = m_pullup(m, off + icmp6len);
658 		if (m == NULL) {
659 			IP6STAT_INC(ip6s_exthdrtoolong);
660 			return;
661 		}
662 	}
663 	ip6 = mtod(m, struct ip6_hdr *);
664 	nd_na = (struct nd_neighbor_advert *)((caddr_t)ip6 + off);
665 
666 	flags = nd_na->nd_na_flags_reserved;
667 	is_router = ((flags & ND_NA_FLAG_ROUTER) != 0);
668 	is_solicited = ((flags & ND_NA_FLAG_SOLICITED) != 0);
669 	is_override = ((flags & ND_NA_FLAG_OVERRIDE) != 0);
670 
671 	taddr6 = nd_na->nd_na_target;
672 	if (in6_setscope(&taddr6, ifp, NULL))
673 		goto bad;	/* XXX: impossible */
674 
675 	if (IN6_IS_ADDR_MULTICAST(&taddr6)) {
676 		nd6log((LOG_ERR,
677 		    "nd6_na_input: invalid target address %s\n",
678 		    ip6_sprintf(ip6bufs, &taddr6)));
679 		goto bad;
680 	}
681 
682 	daddr6 = ip6->ip6_dst;
683 	if (IN6_IS_ADDR_MULTICAST(&daddr6))
684 		if (is_solicited) {
685 			nd6log((LOG_ERR,
686 			    "nd6_na_input: a solicited adv is multicasted\n"));
687 			goto bad;
688 		}
689 
690 	icmp6len -= sizeof(*nd_na);
691 	nd6_option_init(nd_na + 1, icmp6len, &ndopts);
692 	if (nd6_options(&ndopts) < 0) {
693 		nd6log((LOG_INFO,
694 		    "nd6_na_input: invalid ND option, ignored\n"));
695 		/* nd6_options have incremented stats */
696 		goto freeit;
697 	}
698 
699 	lladdr = NULL;
700 	lladdrlen = 0;
701 	if (ndopts.nd_opts_tgt_lladdr) {
702 		lladdr = (char *)(ndopts.nd_opts_tgt_lladdr + 1);
703 		lladdrlen = ndopts.nd_opts_tgt_lladdr->nd_opt_len << 3;
704 	}
705 
706 	/*
707 	 * This effectively disables the DAD check on a non-master CARP
708 	 * address.
709 	 */
710 	if (ifp->if_carp)
711 		ifa = (*carp_iamatch6_p)(ifp, &taddr6);
712 	else
713 		ifa = (struct ifaddr *)in6ifa_ifpwithaddr(ifp, &taddr6);
714 
715 	/*
716 	 * Target address matches one of my interface address.
717 	 *
718 	 * If my address is tentative, this means that there's somebody
719 	 * already using the same address as mine.  This indicates DAD failure.
720 	 * This is defined in RFC 2462.
721 	 *
722 	 * Otherwise, process as defined in RFC 2461.
723 	 */
724 	if (ifa
725 	 && (((struct in6_ifaddr *)ifa)->ia6_flags & IN6_IFF_TENTATIVE)) {
726 		nd6_dad_na_input(ifa);
727 		ifa_free(ifa);
728 		goto freeit;
729 	}
730 
731 	/* Just for safety, maybe unnecessary. */
732 	if (ifa) {
733 		ifa_free(ifa);
734 		log(LOG_ERR,
735 		    "nd6_na_input: duplicate IP6 address %s\n",
736 		    ip6_sprintf(ip6bufs, &taddr6));
737 		goto freeit;
738 	}
739 
740 	if (lladdr && ((ifp->if_addrlen + 2 + 7) & ~7) != lladdrlen) {
741 		nd6log((LOG_INFO, "nd6_na_input: lladdrlen mismatch for %s "
742 		    "(if %d, NA packet %d)\n", ip6_sprintf(ip6bufs, &taddr6),
743 		    ifp->if_addrlen, lladdrlen - 2));
744 		goto bad;
745 	}
746 
747 	/*
748 	 * If no neighbor cache entry is found, NA SHOULD silently be
749 	 * discarded.
750 	 */
751 	ln = nd6_lookup(&taddr6, LLE_SF(AF_INET6, LLE_EXCLUSIVE), ifp);
752 	if (ln == NULL) {
753 		goto freeit;
754 	}
755 
756 	/*
757 	 * Do not try to override static entry.
758 	 */
759 	if (ln->la_flags & LLE_STATIC)
760 		goto freeit;
761 
762 	if (ln->ln_state == ND6_LLINFO_INCOMPLETE) {
763 		/*
764 		 * If the link-layer has address, and no lladdr option came,
765 		 * discard the packet.
766 		 */
767 		if (ifp->if_addrlen && lladdr == NULL) {
768 			goto freeit;
769 		}
770 
771 		/*
772 		 * Record link-layer address, and update the state.
773 		 */
774 		if (!nd6_try_set_entry_addr(ifp, ln, lladdr))
775 			goto freeit;
776 
777 		flush_holdchain = true;
778 		EVENTHANDLER_INVOKE(lle_event, ln, LLENTRY_RESOLVED);
779 		if (is_solicited)
780 			nd6_llinfo_setstate(ln, ND6_LLINFO_REACHABLE);
781 		else
782 			nd6_llinfo_setstate(ln, ND6_LLINFO_STALE);
783 		if ((ln->ln_router = is_router) != 0) {
784 			/*
785 			 * This means a router's state has changed from
786 			 * non-reachable to probably reachable, and might
787 			 * affect the status of associated prefixes..
788 			 */
789 			checklink = 1;
790 		}
791 	} else {
792 		int llchange;
793 
794 		/*
795 		 * Check if the link-layer address has changed or not.
796 		 */
797 		if (lladdr == NULL)
798 			llchange = 0;
799 		else {
800 			if (ln->la_flags & LLE_VALID) {
801 				if (bcmp(lladdr, ln->ll_addr, ifp->if_addrlen))
802 					llchange = 1;
803 				else
804 					llchange = 0;
805 			} else
806 				llchange = 1;
807 		}
808 
809 		/*
810 		 * This is VERY complex.  Look at it with care.
811 		 *
812 		 * override solicit lladdr llchange	action
813 		 *					(L: record lladdr)
814 		 *
815 		 *	0	0	n	--	(2c)
816 		 *	0	0	y	n	(2b) L
817 		 *	0	0	y	y	(1)    REACHABLE->STALE
818 		 *	0	1	n	--	(2c)   *->REACHABLE
819 		 *	0	1	y	n	(2b) L *->REACHABLE
820 		 *	0	1	y	y	(1)    REACHABLE->STALE
821 		 *	1	0	n	--	(2a)
822 		 *	1	0	y	n	(2a) L
823 		 *	1	0	y	y	(2a) L *->STALE
824 		 *	1	1	n	--	(2a)   *->REACHABLE
825 		 *	1	1	y	n	(2a) L *->REACHABLE
826 		 *	1	1	y	y	(2a) L *->REACHABLE
827 		 */
828 		if (!is_override && (lladdr != NULL && llchange)) {  /* (1) */
829 			/*
830 			 * If state is REACHABLE, make it STALE.
831 			 * no other updates should be done.
832 			 */
833 			if (ln->ln_state == ND6_LLINFO_REACHABLE)
834 				nd6_llinfo_setstate(ln, ND6_LLINFO_STALE);
835 			goto freeit;
836 		} else if (is_override				   /* (2a) */
837 			|| (!is_override && (lladdr != NULL && !llchange)) /* (2b) */
838 			|| lladdr == NULL) {			   /* (2c) */
839 			/*
840 			 * Update link-local address, if any.
841 			 */
842 			if (lladdr != NULL) {
843 				linkhdrsize = sizeof(linkhdr);
844 				if (lltable_calc_llheader(ifp, AF_INET6, lladdr,
845 				    linkhdr, &linkhdrsize, &lladdr_off) != 0)
846 					goto freeit;
847 				if (lltable_try_set_entry_addr(ifp, ln, linkhdr,
848 				    linkhdrsize, lladdr_off) == 0) {
849 					ln = NULL;
850 					goto freeit;
851 				}
852 				EVENTHANDLER_INVOKE(lle_event, ln,
853 				    LLENTRY_RESOLVED);
854 			}
855 
856 			/*
857 			 * If solicited, make the state REACHABLE.
858 			 * If not solicited and the link-layer address was
859 			 * changed, make it STALE.
860 			 */
861 			if (is_solicited)
862 				nd6_llinfo_setstate(ln, ND6_LLINFO_REACHABLE);
863 			else {
864 				if (lladdr != NULL && llchange)
865 					nd6_llinfo_setstate(ln, ND6_LLINFO_STALE);
866 			}
867 		}
868 
869 		if (ln->ln_router && !is_router) {
870 			/*
871 			 * The peer dropped the router flag.
872 			 * Remove the sender from the Default Router List and
873 			 * update the Destination Cache entries.
874 			 */
875 			struct ifnet *nd6_ifp;
876 
877 			nd6_ifp = lltable_get_ifp(ln->lle_tbl);
878 			if (!defrouter_remove(&ln->r_l3addr.addr6, nd6_ifp) &&
879 			    (ND_IFINFO(nd6_ifp)->flags &
880 			     ND6_IFF_ACCEPT_RTADV) != 0)
881 				/*
882 				 * Even if the neighbor is not in the default
883 				 * router list, the neighbor may be used as a
884 				 * next hop for some destinations (e.g. redirect
885 				 * case). So we must call rt6_flush explicitly.
886 				 */
887 				rt6_flush(&ip6->ip6_src, ifp);
888 		}
889 		ln->ln_router = is_router;
890 	}
891         /* XXX - QL
892 	 *  Does this matter?
893 	 *  rt->rt_flags &= ~RTF_REJECT;
894 	 */
895 	ln->la_asked = 0;
896 	if (ln->la_hold != NULL)
897 		chain = nd6_grab_holdchain(ln);
898  freeit:
899 	if (ln != NULL)
900 		LLE_WUNLOCK(ln);
901 
902 	if (chain != NULL)
903 		nd6_flush_holdchain(ifp, ln, chain);
904 	if (flush_holdchain)
905 		nd6_flush_children_holdchain(ifp, ln);
906 
907 	if (checklink)
908 		pfxlist_onlink_check();
909 
910 	m_freem(m);
911 	return;
912 
913  bad:
914 	if (ln != NULL)
915 		LLE_WUNLOCK(ln);
916 
917 	ICMP6STAT_INC(icp6s_badna);
918 	m_freem(m);
919 }
920 
921 /*
922  * Neighbor advertisement output handling.
923  *
924  * Based on RFC 2461
925  *
926  * the following items are not implemented yet:
927  * - proxy advertisement delay rule (RFC2461 7.2.8, last paragraph, SHOULD)
928  * - anycast advertisement delay rule (RFC2461 7.2.7, SHOULD)
929  *
930  * tlladdr - 1 if include target link-layer address
931  * sdl0 - sockaddr_dl (= proxy NA) or NULL
932  */
933 static void
934 nd6_na_output_fib(struct ifnet *ifp, const struct in6_addr *daddr6_0,
935     const struct in6_addr *taddr6, u_long flags, int tlladdr,
936     struct sockaddr *sdl0, u_int fibnum)
937 {
938 	struct mbuf *m;
939 	struct m_tag *mtag;
940 	struct ip6_hdr *ip6;
941 	struct nd_neighbor_advert *nd_na;
942 	struct ip6_moptions im6o;
943 	struct in6_addr daddr6, dst6, src6;
944 	uint32_t scopeid;
945 
946 	NET_EPOCH_ASSERT();
947 
948 	int icmp6len, maxlen, error;
949 	caddr_t mac = NULL;
950 
951 	daddr6 = *daddr6_0;	/* make a local copy for modification */
952 
953 	/* estimate the size of message */
954 	maxlen = sizeof(*ip6) + sizeof(*nd_na);
955 	maxlen += (sizeof(struct nd_opt_hdr) + ifp->if_addrlen + 7) & ~7;
956 	KASSERT(max_linkhdr + maxlen <= MCLBYTES, (
957 	    "%s: max_linkhdr + maxlen > MCLBYTES (%d + %d > %d)",
958 	    __func__, max_linkhdr, maxlen, MCLBYTES));
959 
960 	if (max_linkhdr + maxlen > MHLEN)
961 		m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR);
962 	else
963 		m = m_gethdr(M_NOWAIT, MT_DATA);
964 	if (m == NULL)
965 		return;
966 	M_SETFIB(m, fibnum);
967 
968 	if (IN6_IS_ADDR_MULTICAST(&daddr6)) {
969 		m->m_flags |= M_MCAST;
970 		im6o.im6o_multicast_ifp = ifp;
971 		im6o.im6o_multicast_hlim = 255;
972 		im6o.im6o_multicast_loop = 0;
973 	}
974 
975 	icmp6len = sizeof(*nd_na);
976 	m->m_pkthdr.len = m->m_len = sizeof(struct ip6_hdr) + icmp6len;
977 	m->m_data += max_linkhdr;	/* or M_ALIGN() equivalent? */
978 
979 	/* fill neighbor advertisement packet */
980 	ip6 = mtod(m, struct ip6_hdr *);
981 	ip6->ip6_flow = 0;
982 	ip6->ip6_vfc &= ~IPV6_VERSION_MASK;
983 	ip6->ip6_vfc |= IPV6_VERSION;
984 	ip6->ip6_nxt = IPPROTO_ICMPV6;
985 	ip6->ip6_hlim = 255;
986 	if (IN6_IS_ADDR_UNSPECIFIED(&daddr6)) {
987 		/* reply to DAD */
988 		daddr6.s6_addr16[0] = IPV6_ADDR_INT16_MLL;
989 		daddr6.s6_addr16[1] = 0;
990 		daddr6.s6_addr32[1] = 0;
991 		daddr6.s6_addr32[2] = 0;
992 		daddr6.s6_addr32[3] = IPV6_ADDR_INT32_ONE;
993 		if (in6_setscope(&daddr6, ifp, NULL))
994 			goto bad;
995 
996 		flags &= ~ND_NA_FLAG_SOLICITED;
997 	}
998 	ip6->ip6_dst = daddr6;
999 
1000 	/*
1001 	 * Select a source whose scope is the same as that of the dest.
1002 	 */
1003 	in6_splitscope(&daddr6, &dst6, &scopeid);
1004 	error = in6_selectsrc_addr(fibnum, &dst6,
1005 	    scopeid, ifp, &src6, NULL);
1006 	if (error) {
1007 		char ip6buf[INET6_ADDRSTRLEN];
1008 		nd6log((LOG_DEBUG, "nd6_na_output: source can't be "
1009 		    "determined: dst=%s, error=%d\n",
1010 		    ip6_sprintf(ip6buf, &daddr6), error));
1011 		goto bad;
1012 	}
1013 	ip6->ip6_src = src6;
1014 	nd_na = (struct nd_neighbor_advert *)(ip6 + 1);
1015 	nd_na->nd_na_type = ND_NEIGHBOR_ADVERT;
1016 	nd_na->nd_na_code = 0;
1017 	nd_na->nd_na_target = *taddr6;
1018 	in6_clearscope(&nd_na->nd_na_target); /* XXX */
1019 
1020 	/*
1021 	 * "tlladdr" indicates NS's condition for adding tlladdr or not.
1022 	 * see nd6_ns_input() for details.
1023 	 * Basically, if NS packet is sent to unicast/anycast addr,
1024 	 * target lladdr option SHOULD NOT be included.
1025 	 */
1026 	if (tlladdr) {
1027 		/*
1028 		 * sdl0 != NULL indicates proxy NA.  If we do proxy, use
1029 		 * lladdr in sdl0.  If we are not proxying (sending NA for
1030 		 * my address) use lladdr configured for the interface.
1031 		 */
1032 		if (sdl0 == NULL) {
1033 			if (ifp->if_carp)
1034 				mac = (*carp_macmatch6_p)(ifp, m, taddr6);
1035 			if (mac == NULL)
1036 				mac = nd6_ifptomac(ifp);
1037 		} else if (sdl0->sa_family == AF_LINK) {
1038 			struct sockaddr_dl *sdl;
1039 			sdl = (struct sockaddr_dl *)sdl0;
1040 			if (sdl->sdl_alen == ifp->if_addrlen)
1041 				mac = LLADDR(sdl);
1042 		}
1043 	}
1044 	if (tlladdr && mac) {
1045 		int optlen = sizeof(struct nd_opt_hdr) + ifp->if_addrlen;
1046 		struct nd_opt_hdr *nd_opt = (struct nd_opt_hdr *)(nd_na + 1);
1047 
1048 		/* roundup to 8 bytes alignment! */
1049 		optlen = (optlen + 7) & ~7;
1050 
1051 		m->m_pkthdr.len += optlen;
1052 		m->m_len += optlen;
1053 		icmp6len += optlen;
1054 		bzero((caddr_t)nd_opt, optlen);
1055 		nd_opt->nd_opt_type = ND_OPT_TARGET_LINKADDR;
1056 		nd_opt->nd_opt_len = optlen >> 3;
1057 		bcopy(mac, (caddr_t)(nd_opt + 1), ifp->if_addrlen);
1058 	} else
1059 		flags &= ~ND_NA_FLAG_OVERRIDE;
1060 
1061 	ip6->ip6_plen = htons((u_short)icmp6len);
1062 	nd_na->nd_na_flags_reserved = flags;
1063 	nd_na->nd_na_cksum = 0;
1064 	nd_na->nd_na_cksum =
1065 	    in6_cksum(m, IPPROTO_ICMPV6, sizeof(struct ip6_hdr), icmp6len);
1066 
1067 	if (send_sendso_input_hook != NULL) {
1068 		mtag = m_tag_get(PACKET_TAG_ND_OUTGOING,
1069 		    sizeof(unsigned short), M_NOWAIT);
1070 		if (mtag == NULL)
1071 			goto bad;
1072 		*(unsigned short *)(mtag + 1) = nd_na->nd_na_type;
1073 		m_tag_prepend(m, mtag);
1074 	}
1075 
1076 	ip6_output(m, NULL, NULL, 0, &im6o, NULL, NULL);
1077 	icmp6_ifstat_inc(ifp, ifs6_out_msg);
1078 	icmp6_ifstat_inc(ifp, ifs6_out_neighboradvert);
1079 	ICMP6STAT_INC(icp6s_outhist[ND_NEIGHBOR_ADVERT]);
1080 
1081 	return;
1082 
1083   bad:
1084 	m_freem(m);
1085 }
1086 
1087 #ifndef BURN_BRIDGES
1088 void
1089 nd6_na_output(struct ifnet *ifp, const struct in6_addr *daddr6_0,
1090     const struct in6_addr *taddr6, u_long flags, int tlladdr,
1091     struct sockaddr *sdl0)
1092 {
1093 
1094 	nd6_na_output_fib(ifp, daddr6_0, taddr6, flags, tlladdr, sdl0,
1095 	    RT_DEFAULT_FIB);
1096 }
1097 #endif
1098 
1099 caddr_t
1100 nd6_ifptomac(struct ifnet *ifp)
1101 {
1102 	switch (ifp->if_type) {
1103 	case IFT_ETHER:
1104 	case IFT_IEEE1394:
1105 	case IFT_L2VLAN:
1106 	case IFT_INFINIBAND:
1107 	case IFT_BRIDGE:
1108 		return IF_LLADDR(ifp);
1109 	default:
1110 		return NULL;
1111 	}
1112 }
1113 
1114 struct dadq {
1115 	TAILQ_ENTRY(dadq) dad_list;
1116 	struct ifaddr *dad_ifa;
1117 	int dad_count;		/* max NS to send */
1118 	int dad_ns_tcount;	/* # of trials to send NS */
1119 	int dad_ns_ocount;	/* NS sent so far */
1120 	int dad_ns_icount;
1121 	int dad_na_icount;
1122 	int dad_ns_lcount;	/* looped back NS */
1123 	int dad_loopbackprobe;	/* probing state for loopback detection */
1124 	struct callout dad_timer_ch;
1125 	struct vnet *dad_vnet;
1126 	u_int dad_refcnt;
1127 #define	ND_OPT_NONCE_LEN32 \
1128 		((ND_OPT_NONCE_LEN + sizeof(uint32_t) - 1)/sizeof(uint32_t))
1129 	uint32_t dad_nonce[ND_OPT_NONCE_LEN32];
1130 	bool dad_ondadq;	/* on dadq? Protected by DADQ_WLOCK. */
1131 };
1132 
1133 VNET_DEFINE_STATIC(TAILQ_HEAD(, dadq), dadq);
1134 VNET_DEFINE_STATIC(struct rwlock, dad_rwlock);
1135 #define	V_dadq			VNET(dadq)
1136 #define	V_dad_rwlock		VNET(dad_rwlock)
1137 
1138 #define	DADQ_LOCKPTR()		(&V_dad_rwlock)
1139 #define	DADQ_LOCK_INIT()	rw_init(DADQ_LOCKPTR(), "nd6 DAD queue")
1140 #define	DADQ_RLOCK()		rw_rlock(DADQ_LOCKPTR())
1141 #define	DADQ_RUNLOCK()		rw_runlock(DADQ_LOCKPTR())
1142 #define	DADQ_WLOCK()		rw_wlock(DADQ_LOCKPTR())
1143 #define	DADQ_WUNLOCK()		rw_wunlock(DADQ_LOCKPTR())
1144 
1145 #define	DADQ_LOCK_ASSERT()	rw_assert(DADQ_LOCKPTR(), RA_LOCKED);
1146 #define	DADQ_RLOCK_ASSERT()	rw_assert(DADQ_LOCKPTR(), RA_RLOCKED);
1147 #define	DADQ_WLOCK_ASSERT()	rw_assert(DADQ_LOCKPTR(), RA_WLOCKED);
1148 
1149 static void
1150 nd6_dad_add(struct dadq *dp)
1151 {
1152 	DADQ_WLOCK_ASSERT();
1153 
1154 	TAILQ_INSERT_TAIL(&V_dadq, dp, dad_list);
1155 	dp->dad_ondadq = true;
1156 }
1157 
1158 static void
1159 nd6_dad_del(struct dadq *dp)
1160 {
1161 	DADQ_WLOCK_ASSERT();
1162 
1163 	if (dp->dad_ondadq) {
1164 		/*
1165 		 * Remove dp from the dadq and release the dadq's
1166 		 * reference.
1167 		 */
1168 		TAILQ_REMOVE(&V_dadq, dp, dad_list);
1169 		dp->dad_ondadq = false;
1170 		nd6_dad_rele(dp);
1171 	}
1172 }
1173 
1174 static struct dadq *
1175 nd6_dad_find(struct ifaddr *ifa, struct nd_opt_nonce *n)
1176 {
1177 	struct dadq *dp;
1178 
1179 	DADQ_LOCK_ASSERT();
1180 
1181 	TAILQ_FOREACH(dp, &V_dadq, dad_list) {
1182 		if (dp->dad_ifa != ifa)
1183 			continue;
1184 
1185 		/*
1186 		 * Skip if the nonce matches the received one.
1187 		 * +2 in the length is required because of type and
1188 		 * length fields are included in a header.
1189 		 */
1190 		if (n != NULL &&
1191 		    n->nd_opt_nonce_len == (ND_OPT_NONCE_LEN + 2) / 8 &&
1192 		    memcmp(&n->nd_opt_nonce[0], &dp->dad_nonce[0],
1193 		    ND_OPT_NONCE_LEN) == 0) {
1194 			dp->dad_ns_lcount++;
1195 			continue;
1196 		}
1197 		break;
1198 	}
1199 
1200 	return (dp);
1201 }
1202 
1203 static void
1204 nd6_dad_starttimer(struct dadq *dp, int ticks)
1205 {
1206 	DADQ_WLOCK_ASSERT();
1207 
1208 	callout_reset(&dp->dad_timer_ch, ticks, nd6_dad_timer, dp);
1209 }
1210 
1211 static void
1212 nd6_dad_stoptimer(struct dadq *dp)
1213 {
1214 	callout_drain(&dp->dad_timer_ch);
1215 }
1216 
1217 static void
1218 nd6_dad_rele(struct dadq *dp)
1219 {
1220 	if (refcount_release(&dp->dad_refcnt)) {
1221 		KASSERT(!dp->dad_ondadq, ("dp %p still on DAD queue", dp));
1222 		ifa_free(dp->dad_ifa);
1223 		free(dp, M_IP6NDP);
1224 	}
1225 }
1226 
1227 void
1228 nd6_dad_init(void)
1229 {
1230 	DADQ_LOCK_INIT();
1231 	TAILQ_INIT(&V_dadq);
1232 }
1233 
1234 /*
1235  * Start Duplicate Address Detection (DAD) for specified interface address.
1236  */
1237 void
1238 nd6_dad_start(struct ifaddr *ifa, int delay)
1239 {
1240 	struct in6_ifaddr *ia = (struct in6_ifaddr *)ifa;
1241 	struct dadq *dp;
1242 	char ip6buf[INET6_ADDRSTRLEN];
1243 
1244 	KASSERT((ia->ia6_flags & IN6_IFF_TENTATIVE) != 0,
1245 	    ("starting DAD on non-tentative address %p", ifa));
1246 
1247 	/*
1248 	 * If we don't need DAD, don't do it.
1249 	 * There are several cases:
1250 	 * - DAD is disabled globally or on the interface
1251 	 * - the interface address is anycast
1252 	 */
1253 	if ((ia->ia6_flags & IN6_IFF_ANYCAST) != 0 ||
1254 	    V_ip6_dad_count == 0 ||
1255 	    (ND_IFINFO(ifa->ifa_ifp)->flags & ND6_IFF_NO_DAD) != 0) {
1256 		ia->ia6_flags &= ~IN6_IFF_TENTATIVE;
1257 		return;
1258 	}
1259 	if ((ifa->ifa_ifp->if_flags & IFF_UP) == 0 ||
1260 	    (ifa->ifa_ifp->if_drv_flags & IFF_DRV_RUNNING) == 0 ||
1261 	    (ND_IFINFO(ifa->ifa_ifp)->flags & ND6_IFF_IFDISABLED) != 0)
1262 		return;
1263 
1264 	DADQ_WLOCK();
1265 	if ((dp = nd6_dad_find(ifa, NULL)) != NULL) {
1266 		/*
1267 		 * DAD is already in progress.  Let the existing entry
1268 		 * finish it.
1269 		 */
1270 		DADQ_WUNLOCK();
1271 		return;
1272 	}
1273 
1274 	dp = malloc(sizeof(*dp), M_IP6NDP, M_NOWAIT | M_ZERO);
1275 	if (dp == NULL) {
1276 		log(LOG_ERR, "nd6_dad_start: memory allocation failed for "
1277 			"%s(%s)\n",
1278 			ip6_sprintf(ip6buf, &ia->ia_addr.sin6_addr),
1279 			ifa->ifa_ifp ? if_name(ifa->ifa_ifp) : "???");
1280 		return;
1281 	}
1282 	callout_init_rw(&dp->dad_timer_ch, DADQ_LOCKPTR(),
1283 	    CALLOUT_RETURNUNLOCKED);
1284 #ifdef VIMAGE
1285 	dp->dad_vnet = curvnet;
1286 #endif
1287 	nd6log((LOG_DEBUG, "%s: starting DAD for %s\n", if_name(ifa->ifa_ifp),
1288 	    ip6_sprintf(ip6buf, &ia->ia_addr.sin6_addr)));
1289 
1290 	/*
1291 	 * Send NS packet for DAD, ip6_dad_count times.
1292 	 * Note that we must delay the first transmission, if this is the
1293 	 * first packet to be sent from the interface after interface
1294 	 * (re)initialization.
1295 	 */
1296 	dp->dad_ifa = ifa;
1297 	ifa_ref(dp->dad_ifa);
1298 	dp->dad_count = V_ip6_dad_count;
1299 	dp->dad_ns_icount = dp->dad_na_icount = 0;
1300 	dp->dad_ns_ocount = dp->dad_ns_tcount = 0;
1301 	dp->dad_ns_lcount = dp->dad_loopbackprobe = 0;
1302 
1303 	/* Add this to the dadq and add a reference for the dadq. */
1304 	refcount_init(&dp->dad_refcnt, 1);
1305 	nd6_dad_add(dp);
1306 	nd6_dad_starttimer(dp, delay);
1307 	DADQ_WUNLOCK();
1308 }
1309 
1310 /*
1311  * terminate DAD unconditionally.  used for address removals.
1312  */
1313 void
1314 nd6_dad_stop(struct ifaddr *ifa)
1315 {
1316 	struct dadq *dp;
1317 
1318 	DADQ_WLOCK();
1319 	dp = nd6_dad_find(ifa, NULL);
1320 	if (dp == NULL) {
1321 		DADQ_WUNLOCK();
1322 		/* DAD wasn't started yet */
1323 		return;
1324 	}
1325 
1326 	/*
1327 	 * Acquire a temporary reference so that we can safely stop the callout.
1328 	 */
1329 	(void)refcount_acquire(&dp->dad_refcnt);
1330 	nd6_dad_del(dp);
1331 	DADQ_WUNLOCK();
1332 
1333 	nd6_dad_stoptimer(dp);
1334 	nd6_dad_rele(dp);
1335 }
1336 
1337 static void
1338 nd6_dad_timer(void *arg)
1339 {
1340 	struct dadq *dp = arg;
1341 	struct ifaddr *ifa = dp->dad_ifa;
1342 	struct ifnet *ifp = dp->dad_ifa->ifa_ifp;
1343 	struct in6_ifaddr *ia = (struct in6_ifaddr *)ifa;
1344 	char ip6buf[INET6_ADDRSTRLEN];
1345 	struct epoch_tracker et;
1346 
1347 	CURVNET_SET(dp->dad_vnet);
1348 	KASSERT(ia != NULL, ("DAD entry %p with no address", dp));
1349 
1350 	NET_EPOCH_ENTER(et);
1351 	if (ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED) {
1352 		/* Do not need DAD for ifdisabled interface. */
1353 		log(LOG_ERR, "nd6_dad_timer: cancel DAD on %s because of "
1354 		    "ND6_IFF_IFDISABLED.\n", ifp->if_xname);
1355 		goto err;
1356 	}
1357 	if (ia->ia6_flags & IN6_IFF_DUPLICATED) {
1358 		log(LOG_ERR, "nd6_dad_timer: called with duplicated address "
1359 			"%s(%s)\n",
1360 			ip6_sprintf(ip6buf, &ia->ia_addr.sin6_addr),
1361 			ifa->ifa_ifp ? if_name(ifa->ifa_ifp) : "???");
1362 		goto err;
1363 	}
1364 	if ((ia->ia6_flags & IN6_IFF_TENTATIVE) == 0) {
1365 		log(LOG_ERR, "nd6_dad_timer: called with non-tentative address "
1366 			"%s(%s)\n",
1367 			ip6_sprintf(ip6buf, &ia->ia_addr.sin6_addr),
1368 			ifa->ifa_ifp ? if_name(ifa->ifa_ifp) : "???");
1369 		goto err;
1370 	}
1371 
1372 	/* Stop DAD if the interface is down even after dad_maxtry attempts. */
1373 	if ((dp->dad_ns_tcount > V_dad_maxtry) &&
1374 	    (((ifp->if_flags & IFF_UP) == 0) ||
1375 	     ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0))) {
1376 		nd6log((LOG_INFO, "%s: could not run DAD "
1377 		    "because the interface was down or not running.\n",
1378 		    if_name(ifa->ifa_ifp)));
1379 		goto err;
1380 	}
1381 
1382 	/* Need more checks? */
1383 	if (dp->dad_ns_ocount < dp->dad_count) {
1384 		/*
1385 		 * We have more NS to go.  Send NS packet for DAD.
1386 		 */
1387 		nd6_dad_starttimer(dp,
1388 		    (long)ND_IFINFO(ifa->ifa_ifp)->retrans * hz / 1000);
1389 		nd6_dad_ns_output(dp);
1390 		goto done;
1391 	} else {
1392 		/*
1393 		 * We have transmitted sufficient number of DAD packets.
1394 		 * See what we've got.
1395 		 */
1396 		if (dp->dad_ns_icount > 0 || dp->dad_na_icount > 0) {
1397 			/* We've seen NS or NA, means DAD has failed. */
1398 			nd6_dad_duplicated(ifa, dp);
1399 		} else if (V_dad_enhanced != 0 &&
1400 		    dp->dad_ns_lcount > 0 &&
1401 		    dp->dad_ns_lcount > dp->dad_loopbackprobe) {
1402 			/*
1403 			 * Sec. 4.1 in RFC 7527 requires transmission of
1404 			 * additional probes until the loopback condition
1405 			 * becomes clear when a looped back probe is detected.
1406 			 */
1407 			log(LOG_ERR, "%s: a looped back NS message is "
1408 			    "detected during DAD for %s.  "
1409 			    "Another DAD probes are being sent.\n",
1410 			    if_name(ifa->ifa_ifp),
1411 			    ip6_sprintf(ip6buf, IFA_IN6(ifa)));
1412 			dp->dad_loopbackprobe = dp->dad_ns_lcount;
1413 			/*
1414 			 * Send an NS immediately and increase dad_count by
1415 			 * V_nd6_mmaxtries - 1.
1416 			 */
1417 			dp->dad_count =
1418 			    dp->dad_ns_ocount + V_nd6_mmaxtries - 1;
1419 			nd6_dad_starttimer(dp,
1420 			    (long)ND_IFINFO(ifa->ifa_ifp)->retrans * hz / 1000);
1421 			nd6_dad_ns_output(dp);
1422 			goto done;
1423 		} else {
1424 			/*
1425 			 * We are done with DAD.  No NA came, no NS came.
1426 			 * No duplicate address found.  Check IFDISABLED flag
1427 			 * again in case that it is changed between the
1428 			 * beginning of this function and here.
1429 			 */
1430 			if ((ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED) == 0)
1431 				ia->ia6_flags &= ~IN6_IFF_TENTATIVE;
1432 
1433 			nd6log((LOG_DEBUG,
1434 			    "%s: DAD complete for %s - no duplicates found\n",
1435 			    if_name(ifa->ifa_ifp),
1436 			    ip6_sprintf(ip6buf, &ia->ia_addr.sin6_addr)));
1437 			if (dp->dad_ns_lcount > 0)
1438 				log(LOG_ERR, "%s: DAD completed while "
1439 				    "a looped back NS message is detected "
1440 				    "during DAD for %s.\n",
1441 				    if_name(ifa->ifa_ifp),
1442 				    ip6_sprintf(ip6buf, IFA_IN6(ifa)));
1443 		}
1444 	}
1445 err:
1446 	nd6_dad_del(dp);
1447 	DADQ_WUNLOCK();
1448 done:
1449 	NET_EPOCH_EXIT(et);
1450 	CURVNET_RESTORE();
1451 }
1452 
1453 static void
1454 nd6_dad_duplicated(struct ifaddr *ifa, struct dadq *dp)
1455 {
1456 	struct in6_ifaddr *ia = (struct in6_ifaddr *)ifa;
1457 	struct ifnet *ifp;
1458 	char ip6buf[INET6_ADDRSTRLEN];
1459 
1460 	log(LOG_ERR, "%s: DAD detected duplicate IPv6 address %s: "
1461 	    "NS in/out/loopback=%d/%d/%d, NA in=%d\n",
1462 	    if_name(ifa->ifa_ifp), ip6_sprintf(ip6buf, &ia->ia_addr.sin6_addr),
1463 	    dp->dad_ns_icount, dp->dad_ns_ocount, dp->dad_ns_lcount,
1464 	    dp->dad_na_icount);
1465 
1466 	ia->ia6_flags &= ~IN6_IFF_TENTATIVE;
1467 	ia->ia6_flags |= IN6_IFF_DUPLICATED;
1468 
1469 	ifp = ifa->ifa_ifp;
1470 	log(LOG_ERR, "%s: DAD complete for %s - duplicate found\n",
1471 	    if_name(ifp), ip6_sprintf(ip6buf, &ia->ia_addr.sin6_addr));
1472 	log(LOG_ERR, "%s: manual intervention required\n",
1473 	    if_name(ifp));
1474 
1475 	/*
1476 	 * If the address is a link-local address formed from an interface
1477 	 * identifier based on the hardware address which is supposed to be
1478 	 * uniquely assigned (e.g., EUI-64 for an Ethernet interface), IP
1479 	 * operation on the interface SHOULD be disabled.
1480 	 * [RFC 4862, Section 5.4.5]
1481 	 */
1482 	if (IN6_IS_ADDR_LINKLOCAL(&ia->ia_addr.sin6_addr)) {
1483 		struct in6_addr in6;
1484 
1485 		/*
1486 		 * To avoid over-reaction, we only apply this logic when we are
1487 		 * very sure that hardware addresses are supposed to be unique.
1488 		 */
1489 		switch (ifp->if_type) {
1490 		case IFT_ETHER:
1491 		case IFT_ATM:
1492 		case IFT_IEEE1394:
1493 		case IFT_INFINIBAND:
1494 			in6 = ia->ia_addr.sin6_addr;
1495 			if (in6_get_hw_ifid(ifp, &in6) == 0 &&
1496 			    IN6_ARE_ADDR_EQUAL(&ia->ia_addr.sin6_addr, &in6)) {
1497 				ND_IFINFO(ifp)->flags |= ND6_IFF_IFDISABLED;
1498 				log(LOG_ERR, "%s: possible hardware address "
1499 				    "duplication detected, disable IPv6\n",
1500 				    if_name(ifp));
1501 			}
1502 			break;
1503 		}
1504 	}
1505 }
1506 
1507 /*
1508  * Transmit a neighbour solicitation for the purpose of DAD.  Returns with the
1509  * DAD queue unlocked.
1510  */
1511 static void
1512 nd6_dad_ns_output(struct dadq *dp)
1513 {
1514 	struct in6_ifaddr *ia = (struct in6_ifaddr *)dp->dad_ifa;
1515 	struct ifnet *ifp = dp->dad_ifa->ifa_ifp;
1516 	int i;
1517 
1518 	DADQ_WLOCK_ASSERT();
1519 
1520 	dp->dad_ns_tcount++;
1521 	if ((ifp->if_flags & IFF_UP) == 0) {
1522 		DADQ_WUNLOCK();
1523 		return;
1524 	}
1525 	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) {
1526 		DADQ_WUNLOCK();
1527 		return;
1528 	}
1529 
1530 	dp->dad_ns_ocount++;
1531 	if (V_dad_enhanced != 0) {
1532 		for (i = 0; i < ND_OPT_NONCE_LEN32; i++)
1533 			dp->dad_nonce[i] = arc4random();
1534 		/*
1535 		 * XXXHRS: Note that in the case that
1536 		 * DupAddrDetectTransmits > 1, multiple NS messages with
1537 		 * different nonces can be looped back in an unexpected
1538 		 * order.  The current implementation recognizes only
1539 		 * the latest nonce on the sender side.  Practically it
1540 		 * should work well in almost all cases.
1541 		 */
1542 	}
1543 	DADQ_WUNLOCK();
1544 	nd6_ns_output(ifp, NULL, NULL, &ia->ia_addr.sin6_addr,
1545 	    (uint8_t *)&dp->dad_nonce[0]);
1546 }
1547 
1548 static void
1549 nd6_dad_ns_input(struct ifaddr *ifa, struct nd_opt_nonce *ndopt_nonce)
1550 {
1551 	struct dadq *dp;
1552 
1553 	if (ifa == NULL)
1554 		panic("ifa == NULL in nd6_dad_ns_input");
1555 
1556 	/* Ignore Nonce option when Enhanced DAD is disabled. */
1557 	if (V_dad_enhanced == 0)
1558 		ndopt_nonce = NULL;
1559 	DADQ_RLOCK();
1560 	dp = nd6_dad_find(ifa, ndopt_nonce);
1561 	if (dp != NULL)
1562 		dp->dad_ns_icount++;
1563 	DADQ_RUNLOCK();
1564 }
1565 
1566 static void
1567 nd6_dad_na_input(struct ifaddr *ifa)
1568 {
1569 	struct dadq *dp;
1570 
1571 	if (ifa == NULL)
1572 		panic("ifa == NULL in nd6_dad_na_input");
1573 
1574 	DADQ_RLOCK();
1575 	dp = nd6_dad_find(ifa, NULL);
1576 	if (dp != NULL)
1577 		dp->dad_na_icount++;
1578 	DADQ_RUNLOCK();
1579 }
1580