xref: /freebsd/sys/netinet6/nd6_nbr.c (revision bac4760524a2a15ce75e35251f0c9cdf31732f3f)
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 "opt_inet.h"
35 #include "opt_inet6.h"
36 #include "opt_ipsec.h"
37 
38 #include <sys/param.h>
39 #include <sys/systm.h>
40 #include <sys/counter.h>
41 #include <sys/eventhandler.h>
42 #include <sys/malloc.h>
43 #include <sys/libkern.h>
44 #include <sys/lock.h>
45 #include <sys/rwlock.h>
46 #include <sys/mbuf.h>
47 #include <sys/socket.h>
48 #include <sys/sockio.h>
49 #include <sys/time.h>
50 #include <sys/kernel.h>
51 #include <sys/errno.h>
52 #include <sys/sysctl.h>
53 #include <sys/syslog.h>
54 #include <sys/queue.h>
55 #include <sys/callout.h>
56 #include <sys/refcount.h>
57 
58 #include <net/if.h>
59 #include <net/if_types.h>
60 #include <net/if_dl.h>
61 #include <net/if_var.h>
62 #include <net/if_private.h>
63 #include <net/route.h>
64 #include <net/vnet.h>
65 
66 #include <netinet/in.h>
67 #include <netinet/in_var.h>
68 #include <net/if_llatbl.h>
69 #include <netinet6/in6_var.h>
70 #include <netinet6/in6_ifattach.h>
71 #include <netinet/ip6.h>
72 #include <netinet6/ip6_var.h>
73 #include <netinet6/scope6_var.h>
74 #include <netinet6/nd6.h>
75 #include <netinet/icmp6.h>
76 #include <netinet/ip_carp.h>
77 #include <netinet6/send.h>
78 
79 #include <machine/atomic.h>
80 
81 #define SDL(s) ((struct sockaddr_dl *)s)
82 
83 MALLOC_DECLARE(M_IP6NDP);
84 
85 struct dadq;
86 static struct dadq *nd6_dad_find(struct ifaddr *, struct nd_opt_nonce *);
87 static void nd6_dad_add(struct dadq *dp);
88 static void nd6_dad_del(struct dadq *dp);
89 static void nd6_dad_rele(struct dadq *);
90 static void nd6_dad_starttimer(struct dadq *, int);
91 static void nd6_dad_stoptimer(struct dadq *);
92 static void nd6_dad_timer(void *);
93 static void nd6_dad_duplicated(struct ifaddr *, struct dadq *);
94 static void nd6_dad_ns_output(struct dadq *);
95 static void nd6_dad_ns_input(struct ifaddr *, struct nd_opt_nonce *);
96 static void nd6_dad_na_input(struct ifaddr *);
97 static void nd6_na_output_fib(struct ifnet *, const struct in6_addr *,
98     const struct in6_addr *, u_long, int, struct sockaddr *, u_int);
99 static void nd6_ns_output_fib(struct ifnet *, const struct in6_addr *,
100     const struct in6_addr *, const struct in6_addr *, uint8_t *, u_int);
101 static void nd6_queue_add(struct ifaddr *, struct in6_addr *,
102     struct in6_addr *, struct sockaddr_dl *, int, uint32_t);
103 
104 static struct ifaddr *nd6_proxy_fill_sdl(struct ifnet *,
105     const struct in6_addr *, struct sockaddr_dl *);
106 
107 VNET_DEFINE_STATIC(int, dad_enhanced) = 1;
108 #define	V_dad_enhanced			VNET(dad_enhanced)
109 
110 SYSCTL_DECL(_net_inet6_ip6);
111 SYSCTL_INT(_net_inet6_ip6, OID_AUTO, dad_enhanced, CTLFLAG_VNET | CTLFLAG_RW,
112     &VNET_NAME(dad_enhanced), 0,
113     "Enable Enhanced DAD, which adds a random nonce to NS messages for DAD.");
114 
115 VNET_DEFINE_STATIC(int, dad_maxtry) = 15;	/* max # of *tries* to
116 						   transmit DAD packet */
117 #define	V_dad_maxtry			VNET(dad_maxtry)
118 
119 VNET_DEFINE_STATIC(int, nd6_onlink_ns_rfc4861) = 0;
120 #define	V_nd6_onlink_ns_rfc4861		VNET(nd6_onlink_ns_rfc4861)
121 SYSCTL_INT(_net_inet6_icmp6, ICMPV6CTL_ND6_ONLINKNSRFC4861,
122     nd6_onlink_ns_rfc4861, CTLFLAG_VNET | CTLFLAG_RW,
123     &VNET_NAME(nd6_onlink_ns_rfc4861), 0,
124     "Accept 'on-link' ICMPv6 NS messages in compliance with RFC 4861");
125 
126 struct nd_queue {
127 	TAILQ_ENTRY(nd_queue) ndq_list;
128 	struct ifaddr		*ndq_ifa;
129 	struct in6_addr		ndq_daddr;
130 	struct in6_addr		ndq_taddr;
131 	struct sockaddr_dl	ndq_sdl;
132 	uint32_t		ndq_flags;
133 	struct callout		ndq_callout;
134 };
135 
136 /*
137  * Input a Neighbor Solicitation Message.
138  *
139  * Based on RFC 2461
140  * Based on RFC 2462 (duplicate address detection)
141  */
142 void
nd6_ns_input(struct mbuf * m,int off,int icmp6len)143 nd6_ns_input(struct mbuf *m, int off, int icmp6len)
144 {
145 	struct ifnet *ifp;
146 	struct ip6_hdr *ip6;
147 	struct nd_neighbor_solicit *nd_ns;
148 	struct in6_addr daddr6, myaddr6, saddr6, taddr6;
149 	struct ifaddr *ifa;
150 	struct sockaddr_dl proxydl;
151 	union nd_opts ndopts;
152 	char ip6bufs[INET6_ADDRSTRLEN], ip6bufd[INET6_ADDRSTRLEN];
153 	char *lladdr;
154 	int rflag, tentative, tlladdr;
155 	u_int lladdr_pad, lladdrlen;
156 	uint32_t dflags;
157 
158 	ifa = NULL;
159 
160 	/* RFC 6980: Nodes MUST silently ignore fragments */
161 	if(m->m_flags & M_FRAGMENTED)
162 		goto freeit;
163 
164 	ifp = m->m_pkthdr.rcvif;
165 	ip6 = mtod(m, struct ip6_hdr *);
166 	if (__predict_false(ip6->ip6_hlim != 255)) {
167 		ICMP6STAT_INC(icp6s_invlhlim);
168 		nd6log((LOG_ERR,
169 		    "nd6_ns_input: invalid hlim (%d) from %s to %s on %s\n",
170 		    ip6->ip6_hlim, ip6_sprintf(ip6bufs, &ip6->ip6_src),
171 		    ip6_sprintf(ip6bufd, &ip6->ip6_dst), if_name(ifp)));
172 		goto bads;
173 	}
174 
175 	if (m->m_len < off + icmp6len) {
176 		m = m_pullup(m, off + icmp6len);
177 		if (m == NULL) {
178 			IP6STAT_INC(ip6s_exthdrtoolong);
179 			return;
180 		}
181 	}
182 	ip6 = mtod(m, struct ip6_hdr *);
183 	nd_ns = (struct nd_neighbor_solicit *)((caddr_t)ip6 + off);
184 
185 	saddr6 = ip6->ip6_src;
186 	daddr6 = ip6->ip6_dst;
187 	taddr6 = nd_ns->nd_ns_target;
188 	if (in6_setscope(&taddr6, ifp, NULL) != 0)
189 		goto bad;
190 
191 	rflag = (V_ip6_forwarding) ? ND_NA_FLAG_ROUTER : 0;
192 	if (ifp->if_inet6->nd_flags & ND6_IFF_ACCEPT_RTADV && V_ip6_norbit_raif)
193 		rflag = 0;
194 
195 	if (IN6_IS_ADDR_UNSPECIFIED(&saddr6)) {
196 		/* dst has to be a solicited node multicast address. */
197 		if (daddr6.s6_addr16[0] == IPV6_ADDR_INT16_MLL &&
198 		    /* don't check ifindex portion */
199 		    daddr6.s6_addr32[1] == 0 &&
200 		    daddr6.s6_addr32[2] == IPV6_ADDR_INT32_ONE &&
201 		    daddr6.s6_addr8[12] == 0xff) {
202 			; /* good */
203 		} else {
204 			nd6log((LOG_INFO, "nd6_ns_input: bad DAD packet "
205 			    "(wrong ip6 dst)\n"));
206 			goto bad;
207 		}
208 	} else if (!V_nd6_onlink_ns_rfc4861) {
209 		struct sockaddr_in6 src_sa6;
210 
211 		/*
212 		 * According to recent IETF discussions, it is not a good idea
213 		 * to accept a NS from an address which would not be deemed
214 		 * to be a neighbor otherwise.  This point is expected to be
215 		 * clarified in future revisions of the specification.
216 		 */
217 		bzero(&src_sa6, sizeof(src_sa6));
218 		src_sa6.sin6_family = AF_INET6;
219 		src_sa6.sin6_len = sizeof(src_sa6);
220 		src_sa6.sin6_addr = saddr6;
221 		if (nd6_is_addr_neighbor(&src_sa6, ifp) == 0) {
222 			nd6log((LOG_INFO, "nd6_ns_input: "
223 				"NS packet from non-neighbor\n"));
224 			goto bad;
225 		}
226 	}
227 
228 	if (IN6_IS_ADDR_MULTICAST(&taddr6)) {
229 		nd6log((LOG_INFO, "nd6_ns_input: bad NS target (multicast)\n"));
230 		goto bad;
231 	}
232 
233 	icmp6len -= sizeof(*nd_ns);
234 	nd6_option_init(nd_ns + 1, icmp6len, &ndopts);
235 	if (nd6_options(&ndopts) < 0) {
236 		nd6log((LOG_INFO,
237 		    "nd6_ns_input: invalid ND option, ignored\n"));
238 		/* nd6_options have incremented stats */
239 		goto freeit;
240 	}
241 
242 	lladdr = NULL;
243 	lladdrlen = 0;
244 	lladdr_pad = nd6_lladdr_opt_pad(ifp);
245 	if (ndopts.nd_opts_src_lladdr) {
246 		lladdr = (char *)(ndopts.nd_opts_src_lladdr + 1) + lladdr_pad;
247 		lladdrlen = ndopts.nd_opts_src_lladdr->nd_opt_len << 3;
248 	}
249 
250 	if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src) && lladdr) {
251 		nd6log((LOG_INFO, "nd6_ns_input: bad DAD packet "
252 		    "(link-layer address option)\n"));
253 		goto bad;
254 	}
255 
256 	/*
257 	 * Attaching target link-layer address to the NA?
258 	 * (RFC 2461 7.2.4)
259 	 *
260 	 * NS IP dst is unicast/anycast			MUST NOT add
261 	 * NS IP dst is solicited-node multicast	MUST add
262 	 *
263 	 * In implementation, we add target link-layer address by default.
264 	 * We do not add one in MUST NOT cases.
265 	 */
266 	tlladdr = 0;
267 	if (IN6_IS_ADDR_MULTICAST(&daddr6))
268 		tlladdr |= ND6_NA_OPT_LLA;
269 
270 	/*
271 	 * Target address (taddr6) must be either:
272 	 * (1) Valid unicast/anycast address for my receiving interface,
273 	 * (2) Unicast address for which I'm offering proxy service, or
274 	 * (3) "tentative" address on which DAD is being performed.
275 	 */
276 	/* (1) and (3) check. */
277 	if (ifp->if_carp) {
278 		ifa = (*carp_iamatch6_p)(ifp, &taddr6);
279 		if (ifa != NULL)
280 			tlladdr |= ND6_NA_CARP_MASTER;
281 	} else
282 		ifa = (struct ifaddr *)in6ifa_ifpwithaddr(ifp, &taddr6);
283 
284 	/* (2) check. */
285 	dflags = 0;
286 	if (ifa == NULL) {
287 		if ((ifa = nd6_proxy_fill_sdl(ifp, &taddr6, &proxydl)) != NULL)
288 			dflags |= ND6_QUEUE_FLAG_PROXY;
289 	}
290 	if (ifa == NULL) {
291 		/*
292 		 * We've got an NS packet, and we don't have that address
293 		 * assigned for us.  We MUST silently ignore it.
294 		 * See RFC2461 7.2.3.
295 		 */
296 		goto freeit;
297 	}
298 	if ((((struct in6_ifaddr *)ifa)->ia6_flags & IN6_IFF_ANYCAST) != 0)
299 		dflags |= ND6_QUEUE_FLAG_ANYCAST;
300 	myaddr6 = *IFA_IN6(ifa);
301 	tentative = ((struct in6_ifaddr *)ifa)->ia6_flags & IN6_IFF_TENTATIVE;
302 	if (((struct in6_ifaddr *)ifa)->ia6_flags & IN6_IFF_DUPLICATED)
303 		goto freeit;
304 
305 	if (lladdr && ((ifp->if_addrlen + 2 + 7) & ~7) != lladdrlen) {
306 		nd6log((LOG_INFO, "nd6_ns_input: lladdrlen mismatch for %s "
307 		    "(if %d, NS packet %u)\n",
308 		    ip6_sprintf(ip6bufs, &taddr6),
309 		    ifp->if_addrlen, lladdrlen - 2 - lladdr_pad));
310 		goto bad;
311 	}
312 
313 	if (IN6_ARE_ADDR_EQUAL(&myaddr6, &saddr6)) {
314 		nd6log((LOG_INFO, "nd6_ns_input: duplicate IP6 address %s\n",
315 		    ip6_sprintf(ip6bufs, &saddr6)));
316 		goto freeit;
317 	}
318 
319 	/*
320 	 * We have neighbor solicitation packet, with target address equals to
321 	 * one of my tentative address.
322 	 *
323 	 * src addr	how to process?
324 	 * ---		---
325 	 * multicast	of course, invalid (rejected in ip6_input)
326 	 * unicast	somebody is doing address resolution -> ignore
327 	 * unspec	dup address detection
328 	 *
329 	 * The processing is defined in RFC 2462.
330 	 */
331 	if (tentative) {
332 		/*
333 		 * If source address is unspecified address, it is for
334 		 * duplicate address detection.
335 		 *
336 		 * If not, the packet is for addess resolution;
337 		 * silently ignore it.
338 		 */
339 		if (IN6_IS_ADDR_UNSPECIFIED(&saddr6))
340 			nd6_dad_ns_input(ifa, ndopts.nd_opts_nonce);
341 
342 		goto freeit;
343 	}
344 
345 	/*
346 	 * If the Target Address is either an anycast address or a unicast
347 	 * address for which the node is providing proxy service, or the Target
348 	 * Link-Layer Address option is not included, the Override flag SHOULD
349 	 * be set to zero.  Otherwise, the Override flag SHOULD be set to one.
350 	 */
351 	if (dflags == 0 && (tlladdr & ND6_NA_OPT_LLA) != 0)
352 		rflag |= ND_NA_FLAG_OVERRIDE;
353 	/*
354 	 * If the source address is unspecified address, entries must not
355 	 * be created or updated.
356 	 * It looks that sender is performing DAD. nd6_na_output() will
357 	 * send NA toward all-node multicast address, to tell the sender
358 	 * that I'm using the address.
359 	 * S bit ("solicited") must be zero.
360 	 */
361 	if (!IN6_IS_ADDR_UNSPECIFIED(&saddr6)) {
362 		nd6_cache_lladdr(ifp, &saddr6, lladdr, lladdrlen,
363 		    ND_NEIGHBOR_SOLICIT, 0);
364 		rflag |= ND_NA_FLAG_SOLICITED;
365 	}
366 
367 	/*
368 	 * RFC 4861, anycast or proxy NA sent in response to a NS SHOULD
369 	 * be delayed by a random time between 0 and MAX_ANYCAST_DELAY_TIME
370 	 * to reduce the probability of network congestion.
371 	 */
372 	if (dflags == 0)
373 		nd6_na_output_fib(ifp, &saddr6, &taddr6, rflag, tlladdr, NULL, M_GETFIB(m));
374 	else
375 		nd6_queue_add(ifa, &saddr6, &taddr6, &proxydl, arc4random() %
376 		    (MAX_ANYCAST_DELAY_TIME * hz), dflags);
377  freeit:
378 	if (ifa != NULL)
379 		ifa_free(ifa);
380 	m_freem(m);
381 	return;
382 
383  bad:
384 	nd6log((LOG_ERR, "nd6_ns_input: src=%s\n",
385 		ip6_sprintf(ip6bufs, &saddr6)));
386 	nd6log((LOG_ERR, "nd6_ns_input: dst=%s\n",
387 		ip6_sprintf(ip6bufs, &daddr6)));
388 	nd6log((LOG_ERR, "nd6_ns_input: tgt=%s\n",
389 		ip6_sprintf(ip6bufs, &taddr6)));
390  bads:
391 	ICMP6STAT_INC(icp6s_badns);
392 	if (ifa != NULL)
393 		ifa_free(ifa);
394 	m_freem(m);
395 }
396 
397 static struct ifaddr *
nd6_proxy_fill_sdl(struct ifnet * ifp,const struct in6_addr * taddr6,struct sockaddr_dl * sdl)398 nd6_proxy_fill_sdl(struct ifnet *ifp, const struct in6_addr *taddr6,
399     struct sockaddr_dl *sdl)
400 {
401 	struct ifaddr *ifa;
402 	struct llentry *ln;
403 
404 	ifa = NULL;
405 	ln = nd6_lookup(taddr6, LLE_SF(AF_INET6, 0), ifp);
406 	if (ln == NULL)
407 		return (ifa);
408 	if ((ln->la_flags & (LLE_PUB | LLE_VALID)) == (LLE_PUB | LLE_VALID)) {
409 		link_init_sdl(ifp, (struct sockaddr *)sdl, ifp->if_type);
410 		sdl->sdl_alen = ifp->if_addrlen;
411 		bcopy(ln->ll_addr, &sdl->sdl_data, ifp->if_addrlen);
412 		LLE_RUNLOCK(ln);
413 		ifa = (struct ifaddr *)in6ifa_ifpforlinklocal(ifp,
414 		    IN6_IFF_NOTREADY|IN6_IFF_ANYCAST);
415 	} else
416 		LLE_RUNLOCK(ln);
417 
418 	return (ifa);
419 }
420 
421 /*
422  * Output a Neighbor Solicitation Message. Caller specifies:
423  *	- ICMP6 header source IP6 address
424  *	- ND6 header target IP6 address
425  *	- ND6 header source datalink address
426  *
427  * Based on RFC 2461
428  * Based on RFC 2462 (duplicate address detection)
429  *
430  *    ln - for source address determination
431  * nonce - If non-NULL, NS is used for duplicate address detection and
432  *         the value (length is ND_OPT_NONCE_LEN) is used as a random nonce.
433  */
434 static void
nd6_ns_output_fib(struct ifnet * ifp,const struct in6_addr * saddr6,const struct in6_addr * daddr6,const struct in6_addr * taddr6,uint8_t * nonce,u_int fibnum)435 nd6_ns_output_fib(struct ifnet *ifp, const struct in6_addr *saddr6,
436     const struct in6_addr *daddr6, const struct in6_addr *taddr6,
437     uint8_t *nonce, u_int fibnum)
438 {
439 	struct mbuf *m;
440 	struct m_tag *mtag;
441 	struct ip6_hdr *ip6;
442 	struct nd_neighbor_solicit *nd_ns;
443 	struct ip6_moptions im6o;
444 	int icmp6len;
445 	int maxlen;
446 
447 	NET_EPOCH_ASSERT();
448 
449 	if (IN6_IS_ADDR_MULTICAST(taddr6))
450 		return;
451 
452 	/* estimate the size of message */
453 	maxlen = sizeof(*ip6) + sizeof(*nd_ns);
454 	maxlen += (sizeof(struct nd_opt_hdr) + ifp->if_addrlen + 7) & ~7;
455 	KASSERT(max_linkhdr + maxlen <= MCLBYTES, (
456 	    "%s: max_linkhdr + maxlen > MCLBYTES (%d + %d > %d)",
457 	    __func__, max_linkhdr, maxlen, MCLBYTES));
458 
459 	if (max_linkhdr + maxlen > MHLEN)
460 		m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR);
461 	else
462 		m = m_gethdr(M_NOWAIT, MT_DATA);
463 	if (m == NULL)
464 		return;
465 	M_SETFIB(m, fibnum);
466 
467 	icmp6len = sizeof(*nd_ns);
468 	m->m_pkthdr.len = m->m_len = sizeof(*ip6) + icmp6len;
469 	m->m_data += max_linkhdr;	/* or M_ALIGN() equivalent? */
470 
471 	/* fill neighbor solicitation packet */
472 	ip6 = mtod(m, struct ip6_hdr *);
473 	ip6->ip6_flow = 0;
474 	ip6->ip6_vfc &= ~IPV6_VERSION_MASK;
475 	ip6->ip6_vfc |= IPV6_VERSION;
476 	/* ip6->ip6_plen will be set later */
477 	ip6->ip6_nxt = IPPROTO_ICMPV6;
478 	ip6->ip6_hlim = 255;
479 	if (daddr6)
480 		ip6->ip6_dst = *daddr6;
481 	else {
482 		ip6->ip6_dst.s6_addr16[0] = IPV6_ADDR_INT16_MLL;
483 		ip6->ip6_dst.s6_addr16[1] = 0;
484 		ip6->ip6_dst.s6_addr32[1] = 0;
485 		ip6->ip6_dst.s6_addr32[2] = IPV6_ADDR_INT32_ONE;
486 		ip6->ip6_dst.s6_addr32[3] = taddr6->s6_addr32[3];
487 		ip6->ip6_dst.s6_addr8[12] = 0xff;
488 		if (in6_setscope(&ip6->ip6_dst, ifp, NULL) != 0)
489 			goto bad;
490 	}
491 	if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst)) {
492 		m->m_flags |= M_MCAST;
493 		im6o.im6o_multicast_ifp = ifp;
494 		im6o.im6o_multicast_hlim = 255;
495 		im6o.im6o_multicast_loop = 0;
496 	}
497 	if (nonce == NULL) {
498 		char ip6buf[INET6_ADDRSTRLEN];
499 		struct ifaddr *ifa = NULL;
500 
501 		/*
502 		 * RFC2461 7.2.2:
503 		 * "If the source address of the packet prompting the
504 		 * solicitation is the same as one of the addresses assigned
505 		 * to the outgoing interface, that address SHOULD be placed
506 		 * in the IP Source Address of the outgoing solicitation.
507 		 * Otherwise, any one of the addresses assigned to the
508 		 * interface should be used."
509 		 *
510 		 * We use the source address for the prompting packet
511 		 * (saddr6), if saddr6 belongs to the outgoing interface.
512 		 * Otherwise, we perform the source address selection as usual.
513 		 */
514 		if (saddr6 != NULL)
515 			ifa = (struct ifaddr *)in6ifa_ifpwithaddr(ifp, saddr6);
516 		if (ifa == NULL) {
517 			int error;
518 
519 			error = in6_selectsrc_nbr(fibnum, &ip6->ip6_dst, &im6o,
520 			    ifp, &ip6->ip6_src);
521 			if (error) {
522 				nd6log((LOG_DEBUG, "%s: source can't be "
523 				    "determined: dst=%s, error=%d\n", __func__,
524 				    ip6_sprintf(ip6buf, &ip6->ip6_dst),
525 				    error));
526 				goto bad;
527 			}
528 		} else
529 			ip6->ip6_src = *saddr6;
530 
531 		if (ifp->if_carp != NULL) {
532 			/*
533 			 * Check that selected source address belongs to
534 			 * CARP addresses.
535 			 */
536 			if (ifa == NULL)
537 				ifa = (struct ifaddr *)in6ifa_ifpwithaddr(ifp,
538 				    &ip6->ip6_src);
539 			/*
540 			 * Do not send NS for CARP address if we are not
541 			 * the CARP master.
542 			 */
543 			if (ifa != NULL && ifa->ifa_carp != NULL &&
544 			    !(*carp_master_p)(ifa)) {
545 				nd6log((LOG_DEBUG,
546 				    "nd6_ns_output: NS from BACKUP CARP address %s\n",
547 				    ip6_sprintf(ip6buf, &ip6->ip6_src)));
548 				ifa_free(ifa);
549 				goto bad;
550 			}
551 		}
552 		if (ifa != NULL)
553 			ifa_free(ifa);
554 	} else {
555 		/*
556 		 * Source address for DAD packet must always be IPv6
557 		 * unspecified address. (0::0)
558 		 * We actually don't have to 0-clear the address (we did it
559 		 * above), but we do so here explicitly to make the intention
560 		 * clearer.
561 		 */
562 		bzero(&ip6->ip6_src, sizeof(ip6->ip6_src));
563 	}
564 	nd_ns = (struct nd_neighbor_solicit *)(ip6 + 1);
565 	nd_ns->nd_ns_type = ND_NEIGHBOR_SOLICIT;
566 	nd_ns->nd_ns_code = 0;
567 	nd_ns->nd_ns_reserved = 0;
568 	nd_ns->nd_ns_target = *taddr6;
569 	in6_clearscope(&nd_ns->nd_ns_target); /* XXX */
570 
571 	/*
572 	 * Add source link-layer address option.
573 	 *
574 	 *				spec		implementation
575 	 *				---		---
576 	 * DAD packet			MUST NOT	do not add the option
577 	 * there's no link layer address:
578 	 *				impossible	do not add the option
579 	 * there's link layer address:
580 	 *	Multicast NS		MUST add one	add the option
581 	 *	Unicast NS		SHOULD add one	add the option
582 	 */
583 	if (nonce == NULL) {
584 		struct nd_opt_hdr *nd_opt;
585 		char *mac;
586 		int optlen;
587 		u_int pad;
588 
589 		mac = NULL;
590 		if (ifp->if_carp)
591 			mac = (*carp_macmatch6_p)(ifp, m, &ip6->ip6_src);
592 		if (mac == NULL)
593 			mac = nd6_ifptomac(ifp);
594 
595 		if (mac != NULL) {
596 			nd_opt = (struct nd_opt_hdr *)(nd_ns + 1);
597 			pad = nd6_lladdr_opt_pad(ifp);
598 			optlen = sizeof(struct nd_opt_hdr) + ifp->if_addrlen;
599 			/* 8 byte alignments... */
600 			optlen = (optlen + 7) & ~7;
601 			m->m_pkthdr.len += optlen;
602 			m->m_len += optlen;
603 			icmp6len += optlen;
604 			bzero(nd_opt, optlen);
605 			nd_opt->nd_opt_type = ND_OPT_SOURCE_LINKADDR;
606 			nd_opt->nd_opt_len = optlen >> 3;
607 			memcpy((char *)(nd_opt + 1) + pad, mac,
608 			    ifp->if_addrlen);
609 		}
610 	}
611 	/*
612 	 * Add a Nonce option (RFC 3971) to detect looped back NS messages.
613 	 * This behavior is documented as Enhanced Duplicate Address
614 	 * Detection in RFC 7527.
615 	 * net.inet6.ip6.dad_enhanced=0 disables this.
616 	 */
617 	if (V_dad_enhanced != 0 && nonce != NULL) {
618 		int optlen = sizeof(struct nd_opt_hdr) + ND_OPT_NONCE_LEN;
619 		struct nd_opt_hdr *nd_opt = (struct nd_opt_hdr *)(nd_ns + 1);
620 		/* 8-byte alignment is required. */
621 		optlen = (optlen + 7) & ~7;
622 
623 		m->m_pkthdr.len += optlen;
624 		m->m_len += optlen;
625 		icmp6len += optlen;
626 		bzero((caddr_t)nd_opt, optlen);
627 		nd_opt->nd_opt_type = ND_OPT_NONCE;
628 		nd_opt->nd_opt_len = optlen >> 3;
629 		bcopy(nonce, (caddr_t)(nd_opt + 1), ND_OPT_NONCE_LEN);
630 	}
631 	ip6->ip6_plen = htons((u_short)icmp6len);
632 	nd_ns->nd_ns_cksum = 0;
633 	nd_ns->nd_ns_cksum =
634 	    in6_cksum(m, IPPROTO_ICMPV6, sizeof(*ip6), icmp6len);
635 
636 	if (send_sendso_input_hook != NULL) {
637 		mtag = m_tag_get(PACKET_TAG_ND_OUTGOING,
638 			sizeof(unsigned short), M_NOWAIT);
639 		if (mtag == NULL)
640 			goto bad;
641 		*(unsigned short *)(mtag + 1) = nd_ns->nd_ns_type;
642 		m_tag_prepend(m, mtag);
643 	}
644 
645 	ip6_output(m, NULL, NULL, (nonce != NULL) ? IPV6_UNSPECSRC : 0,
646 	    &im6o, NULL, NULL);
647 	icmp6_ifstat_inc(ifp, ifs6_out_msg);
648 	icmp6_ifstat_inc(ifp, ifs6_out_neighborsolicit);
649 	ICMP6STAT_INC2(icp6s_outhist, ND_NEIGHBOR_SOLICIT);
650 
651 	return;
652 
653   bad:
654 	m_freem(m);
655 }
656 
657 #ifndef BURN_BRIDGES
658 void
nd6_ns_output(struct ifnet * ifp,const struct in6_addr * saddr6,const struct in6_addr * daddr6,const struct in6_addr * taddr6,uint8_t * nonce)659 nd6_ns_output(struct ifnet *ifp, const struct in6_addr *saddr6,
660     const struct in6_addr *daddr6, const struct in6_addr *taddr6,uint8_t *nonce)
661 {
662 
663 	nd6_ns_output_fib(ifp, saddr6, daddr6, taddr6, nonce, RT_DEFAULT_FIB);
664 }
665 #endif
666 /*
667  * Neighbor advertisement input handling.
668  *
669  * Based on RFC 2461
670  * Based on RFC 2462 (duplicate address detection)
671  */
672 void
nd6_na_input(struct mbuf * m,int off,int icmp6len)673 nd6_na_input(struct mbuf *m, int off, int icmp6len)
674 {
675 	struct ifnet *ifp;
676 	struct ip6_hdr *ip6;
677 	struct ifaddr *ifa;
678 	struct llentry *ln;
679 	struct mbuf *chain;
680 	struct nd_neighbor_advert *nd_na;
681 	struct in6_addr daddr6, taddr6;
682 	union nd_opts ndopts;
683 	u_char linkhdr[LLE_MAX_LINKHDR];
684 	char ip6bufs[INET6_ADDRSTRLEN], ip6bufd[INET6_ADDRSTRLEN];
685 	char *lladdr;
686 	size_t linkhdrsize;
687 	int flags, is_override, is_router, is_solicited;
688 	int lladdr_off, checklink;
689 	u_int lladdr_pad, lladdrlen;
690 	bool flush_holdchain = false;
691 
692 	NET_EPOCH_ASSERT();
693 
694 	chain = NULL;
695 	ln = NULL;
696 	checklink = 0;
697 
698 	/* RFC 6980: Nodes MUST silently ignore fragments */
699 	if(m->m_flags & M_FRAGMENTED)
700 		goto freeit;
701 
702 	ifp = m->m_pkthdr.rcvif;
703 	ip6 = mtod(m, struct ip6_hdr *);
704 	if (__predict_false(ip6->ip6_hlim != 255)) {
705 		ICMP6STAT_INC(icp6s_invlhlim);
706 		nd6log((LOG_ERR,
707 		    "nd6_na_input: invalid hlim (%d) from %s to %s on %s\n",
708 		    ip6->ip6_hlim, ip6_sprintf(ip6bufs, &ip6->ip6_src),
709 		    ip6_sprintf(ip6bufd, &ip6->ip6_dst), if_name(ifp)));
710 		goto bad;
711 	}
712 
713 	if (m->m_len < off + icmp6len) {
714 		m = m_pullup(m, off + icmp6len);
715 		if (m == NULL) {
716 			IP6STAT_INC(ip6s_exthdrtoolong);
717 			return;
718 		}
719 	}
720 	ip6 = mtod(m, struct ip6_hdr *);
721 	nd_na = (struct nd_neighbor_advert *)((caddr_t)ip6 + off);
722 
723 	flags = nd_na->nd_na_flags_reserved;
724 	is_router = ((flags & ND_NA_FLAG_ROUTER) != 0);
725 	is_solicited = ((flags & ND_NA_FLAG_SOLICITED) != 0);
726 	is_override = ((flags & ND_NA_FLAG_OVERRIDE) != 0);
727 
728 	taddr6 = nd_na->nd_na_target;
729 	if (in6_setscope(&taddr6, ifp, NULL))
730 		goto bad;	/* XXX: impossible */
731 
732 	if (IN6_IS_ADDR_MULTICAST(&taddr6)) {
733 		nd6log((LOG_ERR,
734 		    "nd6_na_input: invalid target address %s\n",
735 		    ip6_sprintf(ip6bufs, &taddr6)));
736 		goto bad;
737 	}
738 
739 	daddr6 = ip6->ip6_dst;
740 	if (IN6_IS_ADDR_MULTICAST(&daddr6))
741 		if (is_solicited) {
742 			nd6log((LOG_ERR,
743 			    "nd6_na_input: a solicited adv is multicasted\n"));
744 			goto bad;
745 		}
746 
747 	icmp6len -= sizeof(*nd_na);
748 	nd6_option_init(nd_na + 1, icmp6len, &ndopts);
749 	if (nd6_options(&ndopts) < 0) {
750 		nd6log((LOG_INFO,
751 		    "nd6_na_input: invalid ND option, ignored\n"));
752 		/* nd6_options have incremented stats */
753 		goto freeit;
754 	}
755 
756 	lladdr = NULL;
757 	lladdrlen = 0;
758 	lladdr_pad = nd6_lladdr_opt_pad(ifp);
759 	if (ndopts.nd_opts_tgt_lladdr) {
760 		lladdr = (char *)(ndopts.nd_opts_tgt_lladdr + 1) + lladdr_pad;
761 		lladdrlen = ndopts.nd_opts_tgt_lladdr->nd_opt_len << 3;
762 	}
763 
764 	ifa = (struct ifaddr *)in6ifa_ifpwithaddr(ifp, &taddr6);
765 	if (ifa != NULL && ifa->ifa_carp != NULL) {
766 		/*
767 		 * Silently ignore NAs for CARP addresses if we are not
768 		 * the CARP master.
769 		 */
770 		if (!(*carp_master_p)(ifa)) {
771 			nd6log((LOG_DEBUG,
772 			    "nd6_na_input: NA for BACKUP CARP address %s\n",
773 			    ip6_sprintf(ip6bufs, &taddr6)));
774 			ifa_free(ifa);
775 			goto freeit;
776 		}
777 	}
778 	/*
779 	 * Target address matches one of my interface address.
780 	 *
781 	 * If my address is tentative, this means that there's somebody
782 	 * already using the same address as mine.  This indicates DAD failure.
783 	 * This is defined in RFC 2462.
784 	 *
785 	 * Otherwise, process as defined in RFC 2461.
786 	 */
787 	if (ifa
788 	 && (((struct in6_ifaddr *)ifa)->ia6_flags & IN6_IFF_TENTATIVE)) {
789 		nd6_dad_na_input(ifa);
790 		ifa_free(ifa);
791 		goto freeit;
792 	}
793 
794 	/* Just for safety, maybe unnecessary. */
795 	if (ifa) {
796 		ifa_free(ifa);
797 		log(LOG_ERR,
798 		    "nd6_na_input: duplicate IP6 address %s\n",
799 		    ip6_sprintf(ip6bufs, &taddr6));
800 		goto freeit;
801 	}
802 
803 	if (lladdr && ((ifp->if_addrlen + 2 + 7) & ~7) != lladdrlen) {
804 		nd6log((LOG_INFO, "nd6_na_input: lladdrlen mismatch for %s "
805 		    "(if %d, NA packet %u)\n", ip6_sprintf(ip6bufs, &taddr6),
806 		    ifp->if_addrlen, lladdrlen - 2 - lladdr_pad));
807 		goto bad;
808 	}
809 
810 	/*
811 	 * If no neighbor cache entry is found, NA SHOULD silently be
812 	 * discarded.
813 	 */
814 	ln = nd6_lookup(&taddr6, LLE_SF(AF_INET6, LLE_EXCLUSIVE), ifp);
815 	if (ln == NULL) {
816 		goto freeit;
817 	}
818 
819 	/*
820 	 * Do not try to override static entry.
821 	 */
822 	if (ln->la_flags & LLE_STATIC)
823 		goto freeit;
824 
825 	if (ln->ln_state == ND6_LLINFO_INCOMPLETE) {
826 		/*
827 		 * If the link-layer has address, and no lladdr option came,
828 		 * discard the packet.
829 		 */
830 		if (ifp->if_addrlen && lladdr == NULL) {
831 			goto freeit;
832 		}
833 
834 		/*
835 		 * Record link-layer address, and update the state.
836 		 */
837 		if (!nd6_try_set_entry_addr(ifp, ln, lladdr))
838 			goto freeit;
839 
840 		flush_holdchain = true;
841 		if (is_solicited)
842 			nd6_llinfo_setstate(ln, ND6_LLINFO_REACHABLE);
843 		else
844 			nd6_llinfo_setstate(ln, ND6_LLINFO_STALE);
845 		EVENTHANDLER_INVOKE(lle_event, ln, LLENTRY_RESOLVED);
846 		if ((ln->ln_router = is_router) != 0) {
847 			/*
848 			 * This means a router's state has changed from
849 			 * non-reachable to probably reachable, and might
850 			 * affect the status of associated prefixes..
851 			 */
852 			checklink = 1;
853 		}
854 	} else {
855 		int llchange;
856 
857 		/*
858 		 * Check if the link-layer address has changed or not.
859 		 */
860 		if (lladdr == NULL)
861 			llchange = 0;
862 		else {
863 			if (ln->la_flags & LLE_VALID) {
864 				if (bcmp(lladdr, ln->ll_addr, ifp->if_addrlen))
865 					llchange = 1;
866 				else
867 					llchange = 0;
868 			} else
869 				llchange = 1;
870 		}
871 
872 		/*
873 		 * This is VERY complex.  Look at it with care.
874 		 *
875 		 * override solicit lladdr llchange	action
876 		 *					(L: record lladdr)
877 		 *
878 		 *	0	0	n	--	(2c)
879 		 *	0	0	y	n	(2b) L
880 		 *	0	0	y	y	(1)    REACHABLE->STALE
881 		 *	0	1	n	--	(2c)   *->REACHABLE
882 		 *	0	1	y	n	(2b) L *->REACHABLE
883 		 *	0	1	y	y	(1)    REACHABLE->STALE
884 		 *	1	0	n	--	(2a)
885 		 *	1	0	y	n	(2a) L
886 		 *	1	0	y	y	(2a) L *->STALE
887 		 *	1	1	n	--	(2a)   *->REACHABLE
888 		 *	1	1	y	n	(2a) L *->REACHABLE
889 		 *	1	1	y	y	(2a) L *->REACHABLE
890 		 */
891 		if (!is_override && (lladdr != NULL && llchange)) {  /* (1) */
892 			/*
893 			 * If state is REACHABLE, make it STALE.
894 			 * no other updates should be done.
895 			 */
896 			if (ln->ln_state == ND6_LLINFO_REACHABLE)
897 				nd6_llinfo_setstate(ln, ND6_LLINFO_STALE);
898 			goto freeit;
899 		} else if (is_override				   /* (2a) */
900 			|| (!is_override && (lladdr != NULL && !llchange)) /* (2b) */
901 			|| lladdr == NULL) {			   /* (2c) */
902 			/*
903 			 * Update link-local address, if any.
904 			 */
905 			if (lladdr != NULL) {
906 				linkhdrsize = sizeof(linkhdr);
907 				if (lltable_calc_llheader(ifp, AF_INET6, lladdr,
908 				    linkhdr, &linkhdrsize, &lladdr_off) != 0)
909 					goto freeit;
910 				if (lltable_try_set_entry_addr(ifp, ln, linkhdr,
911 				    linkhdrsize, lladdr_off) == 0)
912 					goto freeit;
913 				EVENTHANDLER_INVOKE(lle_event, ln,
914 				    LLENTRY_RESOLVED);
915 			}
916 
917 			/*
918 			 * If solicited, make the state REACHABLE.
919 			 * If not solicited and the link-layer address was
920 			 * changed, make it STALE.
921 			 */
922 			if (is_solicited)
923 				nd6_llinfo_setstate(ln, ND6_LLINFO_REACHABLE);
924 			else {
925 				if (lladdr != NULL && llchange)
926 					nd6_llinfo_setstate(ln, ND6_LLINFO_STALE);
927 			}
928 		}
929 
930 		if (ln->ln_router && !is_router) {
931 			/*
932 			 * The peer dropped the router flag.
933 			 * Remove the sender from the Default Router List and
934 			 * update the Destination Cache entries.
935 			 */
936 			struct ifnet *nd6_ifp;
937 
938 			nd6_ifp = lltable_get_ifp(ln->lle_tbl);
939 			if (!defrouter_remove(&ln->r_l3addr.addr6, nd6_ifp) &&
940 			    (nd6_ifp->if_inet6->nd_flags &
941 			     ND6_IFF_ACCEPT_RTADV) != 0)
942 				/*
943 				 * Even if the neighbor is not in the default
944 				 * router list, the neighbor may be used as a
945 				 * next hop for some destinations (e.g. redirect
946 				 * case). So we must call rt6_flush explicitly.
947 				 */
948 				rt6_flush(&ip6->ip6_src, ifp);
949 		}
950 		ln->ln_router = is_router;
951 	}
952         /* XXX - QL
953 	 *  Does this matter?
954 	 *  rt->rt_flags &= ~RTF_REJECT;
955 	 */
956 	ln->la_asked = 0;
957 	if (ln->la_hold != NULL)
958 		chain = nd6_grab_holdchain(ln);
959  freeit:
960 	if (ln != NULL)
961 		LLE_WUNLOCK(ln);
962 
963 	if (chain != NULL)
964 		nd6_flush_holdchain(ifp, ln, chain);
965 	if (flush_holdchain)
966 		nd6_flush_children_holdchain(ifp, ln);
967 
968 	if (checklink)
969 		pfxlist_onlink_check();
970 
971 	m_freem(m);
972 	return;
973 
974  bad:
975 	if (ln != NULL)
976 		LLE_WUNLOCK(ln);
977 
978 	ICMP6STAT_INC(icp6s_badna);
979 	m_freem(m);
980 }
981 
982 /*
983  * Neighbor advertisement output handling.
984  *
985  * Based on RFC 2461
986  *
987  * tlladdr:
988  * - 0x01 if include target link-layer address
989  * - 0x02 if target address is CARP MASTER
990  * sdl0 - sockaddr_dl (= proxy NA) or NULL
991  */
992 static void
nd6_na_output_fib(struct ifnet * ifp,const struct in6_addr * daddr6_0,const struct in6_addr * taddr6,u_long flags,int tlladdr,struct sockaddr * sdl0,u_int fibnum)993 nd6_na_output_fib(struct ifnet *ifp, const struct in6_addr *daddr6_0,
994     const struct in6_addr *taddr6, u_long flags, int tlladdr,
995     struct sockaddr *sdl0, u_int fibnum)
996 {
997 	struct mbuf *m;
998 	struct m_tag *mtag;
999 	struct ip6_hdr *ip6;
1000 	struct nd_neighbor_advert *nd_na;
1001 	struct ip6_moptions im6o;
1002 	struct in6_addr daddr6;
1003 
1004 	NET_EPOCH_ASSERT();
1005 
1006 	int icmp6len, maxlen, error;
1007 	caddr_t mac = NULL;
1008 
1009 	daddr6 = *daddr6_0;	/* make a local copy for modification */
1010 
1011 	/* estimate the size of message */
1012 	maxlen = sizeof(*ip6) + sizeof(*nd_na);
1013 	maxlen += (sizeof(struct nd_opt_hdr) + ifp->if_addrlen + 7) & ~7;
1014 	KASSERT(max_linkhdr + maxlen <= MCLBYTES, (
1015 	    "%s: max_linkhdr + maxlen > MCLBYTES (%d + %d > %d)",
1016 	    __func__, max_linkhdr, maxlen, MCLBYTES));
1017 
1018 	if (max_linkhdr + maxlen > MHLEN)
1019 		m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR);
1020 	else
1021 		m = m_gethdr(M_NOWAIT, MT_DATA);
1022 	if (m == NULL)
1023 		return;
1024 	M_SETFIB(m, fibnum);
1025 
1026 	icmp6len = sizeof(*nd_na);
1027 	m->m_pkthdr.len = m->m_len = sizeof(struct ip6_hdr) + icmp6len;
1028 	m->m_data += max_linkhdr;	/* or M_ALIGN() equivalent? */
1029 
1030 	/* fill neighbor advertisement packet */
1031 	ip6 = mtod(m, struct ip6_hdr *);
1032 	ip6->ip6_flow = 0;
1033 	ip6->ip6_vfc &= ~IPV6_VERSION_MASK;
1034 	ip6->ip6_vfc |= IPV6_VERSION;
1035 	ip6->ip6_nxt = IPPROTO_ICMPV6;
1036 	ip6->ip6_hlim = 255;
1037 
1038 	if (IN6_IS_ADDR_UNSPECIFIED(&daddr6)) {
1039 		/* reply to DAD */
1040 		daddr6 = in6addr_linklocal_allnodes;
1041 		if (in6_setscope(&daddr6, ifp, NULL))
1042 			goto bad;
1043 
1044 		flags &= ~ND_NA_FLAG_SOLICITED;
1045 	}
1046 	if (IN6_IS_ADDR_MULTICAST(&daddr6)) {
1047 		m->m_flags |= M_MCAST;
1048 		im6o.im6o_multicast_ifp = ifp;
1049 		im6o.im6o_multicast_hlim = 255;
1050 		im6o.im6o_multicast_loop = 0;
1051 	}
1052 
1053 	ip6->ip6_dst = daddr6;
1054 	error = in6_selectsrc_nbr(fibnum, &daddr6, &im6o, ifp, &ip6->ip6_src);
1055 	if (error) {
1056 		char ip6buf[INET6_ADDRSTRLEN];
1057 		nd6log((LOG_DEBUG, "nd6_na_output: source can't be "
1058 		    "determined: dst=%s, error=%d\n",
1059 		    ip6_sprintf(ip6buf, &daddr6), error));
1060 		goto bad;
1061 	}
1062 	nd_na = (struct nd_neighbor_advert *)(ip6 + 1);
1063 	nd_na->nd_na_type = ND_NEIGHBOR_ADVERT;
1064 	nd_na->nd_na_code = 0;
1065 	nd_na->nd_na_target = *taddr6;
1066 	in6_clearscope(&nd_na->nd_na_target); /* XXX */
1067 
1068 	/*
1069 	 * If we respond from CARP address, we need to prepare mac address
1070 	 * for carp_output().
1071 	 */
1072 	if (ifp->if_carp && (tlladdr & ND6_NA_CARP_MASTER))
1073 		mac = (*carp_macmatch6_p)(ifp, m, taddr6);
1074 	/*
1075 	 * "tlladdr" indicates NS's condition for adding tlladdr or not.
1076 	 * see nd6_ns_input() for details.
1077 	 * Basically, if NS packet is sent to unicast/anycast addr,
1078 	 * target lladdr option SHOULD NOT be included.
1079 	 */
1080 	if (tlladdr & ND6_NA_OPT_LLA) {
1081 		/*
1082 		 * sdl0 != NULL indicates proxy NA.  If we do proxy, use
1083 		 * lladdr in sdl0.  If we are not proxying (sending NA for
1084 		 * my address) use lladdr configured for the interface.
1085 		 */
1086 		if (sdl0 == NULL) {
1087 			if (mac == NULL)
1088 				mac = nd6_ifptomac(ifp);
1089 		} else if (sdl0->sa_family == AF_LINK) {
1090 			struct sockaddr_dl *sdl;
1091 			sdl = (struct sockaddr_dl *)sdl0;
1092 			if (sdl->sdl_alen == ifp->if_addrlen)
1093 				mac = LLADDR(sdl);
1094 		}
1095 	}
1096 	if ((tlladdr & ND6_NA_OPT_LLA) && mac != NULL) {
1097 		u_int pad = nd6_lladdr_opt_pad(ifp);
1098 		int optlen = sizeof(struct nd_opt_hdr) + ifp->if_addrlen;
1099 		struct nd_opt_hdr *nd_opt = (struct nd_opt_hdr *)(nd_na + 1);
1100 
1101 		/* roundup to 8 bytes alignment! */
1102 		optlen = (optlen + 7) & ~7;
1103 
1104 		m->m_pkthdr.len += optlen;
1105 		m->m_len += optlen;
1106 		icmp6len += optlen;
1107 		bzero((caddr_t)nd_opt, optlen);
1108 		nd_opt->nd_opt_type = ND_OPT_TARGET_LINKADDR;
1109 		nd_opt->nd_opt_len = optlen >> 3;
1110 		memcpy((char *)(nd_opt + 1) + pad, mac, ifp->if_addrlen);
1111 	} else
1112 		flags &= ~ND_NA_FLAG_OVERRIDE;
1113 
1114 	ip6->ip6_plen = htons((u_short)icmp6len);
1115 	nd_na->nd_na_flags_reserved = flags;
1116 	nd_na->nd_na_cksum = 0;
1117 	nd_na->nd_na_cksum =
1118 	    in6_cksum(m, IPPROTO_ICMPV6, sizeof(struct ip6_hdr), icmp6len);
1119 
1120 	if (send_sendso_input_hook != NULL) {
1121 		mtag = m_tag_get(PACKET_TAG_ND_OUTGOING,
1122 		    sizeof(unsigned short), M_NOWAIT);
1123 		if (mtag == NULL)
1124 			goto bad;
1125 		*(unsigned short *)(mtag + 1) = nd_na->nd_na_type;
1126 		m_tag_prepend(m, mtag);
1127 	}
1128 
1129 	ip6_output(m, NULL, NULL, 0, &im6o, NULL, NULL);
1130 	icmp6_ifstat_inc(ifp, ifs6_out_msg);
1131 	icmp6_ifstat_inc(ifp, ifs6_out_neighboradvert);
1132 	ICMP6STAT_INC2(icp6s_outhist, ND_NEIGHBOR_ADVERT);
1133 
1134 	return;
1135 
1136   bad:
1137 	m_freem(m);
1138 }
1139 
1140 #ifndef BURN_BRIDGES
1141 void
nd6_na_output(struct ifnet * ifp,const struct in6_addr * daddr6_0,const struct in6_addr * taddr6,u_long flags,int tlladdr,struct sockaddr * sdl0)1142 nd6_na_output(struct ifnet *ifp, const struct in6_addr *daddr6_0,
1143     const struct in6_addr *taddr6, u_long flags, int tlladdr,
1144     struct sockaddr *sdl0)
1145 {
1146 
1147 	nd6_na_output_fib(ifp, daddr6_0, taddr6, flags, tlladdr, sdl0,
1148 	    RT_DEFAULT_FIB);
1149 }
1150 #endif
1151 
1152 caddr_t
nd6_ifptomac(struct ifnet * ifp)1153 nd6_ifptomac(struct ifnet *ifp)
1154 {
1155 	switch (ifp->if_type) {
1156 	case IFT_ETHER:
1157 	case IFT_IEEE1394:
1158 	case IFT_L2VLAN:
1159 	case IFT_INFINIBAND:
1160 	case IFT_BRIDGE:
1161 		return IF_LLADDR(ifp);
1162 	default:
1163 		return NULL;
1164 	}
1165 }
1166 
1167 /*
1168  * Number of reserved octets between the header of an ND link-layer
1169  * address option and the link-layer address itself.  On IPoIB links
1170  * two zero octets are prepended to the 20-octet link-layer address
1171  * to fill the 24-octet option (RFC 4391, section 9.3); on all other
1172  * link types the address immediately follows the option header.
1173  */
1174 u_int
nd6_lladdr_opt_pad(struct ifnet * ifp)1175 nd6_lladdr_opt_pad(struct ifnet *ifp)
1176 {
1177 	switch (ifp->if_type) {
1178 	case IFT_INFINIBAND:
1179 	case IFT_INFINIBANDLAG:
1180 		return (2);
1181 	default:
1182 		return (0);
1183 	}
1184 }
1185 
1186 struct dadq {
1187 	TAILQ_ENTRY(dadq) dad_list;
1188 	struct ifaddr *dad_ifa;
1189 	int dad_count;		/* max NS to send */
1190 	int dad_ns_tcount;	/* # of trials to send NS */
1191 	int dad_ns_ocount;	/* NS sent so far */
1192 	int dad_ns_icount;
1193 	int dad_na_icount;
1194 	int dad_ns_lcount;	/* looped back NS */
1195 	int dad_loopbackprobe;	/* probing state for loopback detection */
1196 	struct callout dad_timer_ch;
1197 	struct vnet *dad_vnet;
1198 	u_int dad_refcnt;
1199 #define	ND_OPT_NONCE_LEN32 \
1200 		((ND_OPT_NONCE_LEN + sizeof(uint32_t) - 1)/sizeof(uint32_t))
1201 	uint32_t dad_nonce[ND_OPT_NONCE_LEN32];
1202 	bool dad_ondadq;	/* on dadq? Protected by DADQ_WLOCK. */
1203 };
1204 
1205 VNET_DEFINE_STATIC(TAILQ_HEAD(, dadq), dadq);
1206 VNET_DEFINE_STATIC(struct rwlock, dad_rwlock);
1207 #define	V_dadq			VNET(dadq)
1208 #define	V_dad_rwlock		VNET(dad_rwlock)
1209 
1210 #define	DADQ_LOCKPTR()		(&V_dad_rwlock)
1211 #define	DADQ_LOCK_INIT()	rw_init(DADQ_LOCKPTR(), "nd6 DAD queue")
1212 #define	DADQ_RLOCK()		rw_rlock(DADQ_LOCKPTR())
1213 #define	DADQ_RUNLOCK()		rw_runlock(DADQ_LOCKPTR())
1214 #define	DADQ_WLOCK()		rw_wlock(DADQ_LOCKPTR())
1215 #define	DADQ_WUNLOCK()		rw_wunlock(DADQ_LOCKPTR())
1216 
1217 #define	DADQ_LOCK_ASSERT()	rw_assert(DADQ_LOCKPTR(), RA_LOCKED);
1218 #define	DADQ_RLOCK_ASSERT()	rw_assert(DADQ_LOCKPTR(), RA_RLOCKED);
1219 #define	DADQ_WLOCK_ASSERT()	rw_assert(DADQ_LOCKPTR(), RA_WLOCKED);
1220 
1221 static void
nd6_dad_add(struct dadq * dp)1222 nd6_dad_add(struct dadq *dp)
1223 {
1224 	DADQ_WLOCK_ASSERT();
1225 
1226 	TAILQ_INSERT_TAIL(&V_dadq, dp, dad_list);
1227 	dp->dad_ondadq = true;
1228 }
1229 
1230 static void
nd6_dad_del(struct dadq * dp)1231 nd6_dad_del(struct dadq *dp)
1232 {
1233 	DADQ_WLOCK_ASSERT();
1234 
1235 	if (dp->dad_ondadq) {
1236 		/*
1237 		 * Remove dp from the dadq and release the dadq's
1238 		 * reference.
1239 		 */
1240 		TAILQ_REMOVE(&V_dadq, dp, dad_list);
1241 		dp->dad_ondadq = false;
1242 		nd6_dad_rele(dp);
1243 	}
1244 }
1245 
1246 static struct dadq *
nd6_dad_find(struct ifaddr * ifa,struct nd_opt_nonce * n)1247 nd6_dad_find(struct ifaddr *ifa, struct nd_opt_nonce *n)
1248 {
1249 	struct dadq *dp;
1250 
1251 	DADQ_LOCK_ASSERT();
1252 
1253 	TAILQ_FOREACH(dp, &V_dadq, dad_list) {
1254 		if (dp->dad_ifa != ifa)
1255 			continue;
1256 
1257 		/*
1258 		 * Skip if the nonce matches the received one.
1259 		 * +2 in the length is required because of type and
1260 		 * length fields are included in a header.
1261 		 */
1262 		if (n != NULL &&
1263 		    n->nd_opt_nonce_len == (ND_OPT_NONCE_LEN + 2) / 8 &&
1264 		    memcmp(&n->nd_opt_nonce[0], &dp->dad_nonce[0],
1265 		    ND_OPT_NONCE_LEN) == 0) {
1266 			dp->dad_ns_lcount++;
1267 			continue;
1268 		}
1269 		break;
1270 	}
1271 
1272 	return (dp);
1273 }
1274 
1275 static void
nd6_dad_starttimer(struct dadq * dp,int ticks)1276 nd6_dad_starttimer(struct dadq *dp, int ticks)
1277 {
1278 	DADQ_WLOCK_ASSERT();
1279 
1280 	callout_reset(&dp->dad_timer_ch, ticks, nd6_dad_timer, dp);
1281 }
1282 
1283 static void
nd6_dad_stoptimer(struct dadq * dp)1284 nd6_dad_stoptimer(struct dadq *dp)
1285 {
1286 	callout_drain(&dp->dad_timer_ch);
1287 }
1288 
1289 static void
nd6_dad_rele(struct dadq * dp)1290 nd6_dad_rele(struct dadq *dp)
1291 {
1292 	if (refcount_release(&dp->dad_refcnt)) {
1293 		KASSERT(!dp->dad_ondadq, ("dp %p still on DAD queue", dp));
1294 		ifa_free(dp->dad_ifa);
1295 		free(dp, M_IP6NDP);
1296 	}
1297 }
1298 
1299 void
nd6_dad_init(void)1300 nd6_dad_init(void)
1301 {
1302 	DADQ_LOCK_INIT();
1303 	TAILQ_INIT(&V_dadq);
1304 }
1305 
1306 /*
1307  * Start Duplicate Address Detection (DAD) for specified interface address.
1308  */
1309 void
nd6_dad_start(struct ifaddr * ifa,int delay)1310 nd6_dad_start(struct ifaddr *ifa, int delay)
1311 {
1312 	struct in6_ifaddr *ia = (struct in6_ifaddr *)ifa;
1313 	struct dadq *dp;
1314 	char ip6buf[INET6_ADDRSTRLEN];
1315 
1316 	KASSERT((ia->ia6_flags & IN6_IFF_TENTATIVE) != 0,
1317 	    ("starting DAD on non-tentative address %p", ifa));
1318 
1319 	/*
1320 	 * If we don't need DAD, don't do it.
1321 	 * There are several cases:
1322 	 * - DAD is disabled globally or on the interface
1323 	 * - the interface address is anycast
1324 	 */
1325 	if ((ia->ia6_flags & IN6_IFF_ANYCAST) != 0 ||
1326 	    V_ip6_dad_count == 0 ||
1327 	    (ifa->ifa_ifp->if_inet6->nd_flags & ND6_IFF_NO_DAD) != 0) {
1328 		ia->ia6_flags &= ~IN6_IFF_TENTATIVE;
1329 		return;
1330 	}
1331 	if ((ifa->ifa_ifp->if_flags & IFF_UP) == 0 ||
1332 	    (ifa->ifa_ifp->if_drv_flags & IFF_DRV_RUNNING) == 0 ||
1333 	    (ifa->ifa_ifp->if_inet6->nd_flags & ND6_IFF_IFDISABLED) != 0)
1334 		return;
1335 
1336 	DADQ_WLOCK();
1337 	if ((dp = nd6_dad_find(ifa, NULL)) != NULL) {
1338 		/*
1339 		 * DAD is already in progress.  Let the existing entry
1340 		 * finish it.
1341 		 */
1342 		DADQ_WUNLOCK();
1343 		return;
1344 	}
1345 
1346 	dp = malloc(sizeof(*dp), M_IP6NDP, M_NOWAIT | M_ZERO);
1347 	if (dp == NULL) {
1348 		log(LOG_ERR, "nd6_dad_start: memory allocation failed for "
1349 			"%s(%s)\n",
1350 			ip6_sprintf(ip6buf, &ia->ia_addr.sin6_addr),
1351 			ifa->ifa_ifp ? if_name(ifa->ifa_ifp) : "???");
1352 		return;
1353 	}
1354 	callout_init_rw(&dp->dad_timer_ch, DADQ_LOCKPTR(),
1355 	    CALLOUT_RETURNUNLOCKED);
1356 #ifdef VIMAGE
1357 	dp->dad_vnet = curvnet;
1358 #endif
1359 	nd6log((LOG_DEBUG, "%s: starting DAD for %s\n", if_name(ifa->ifa_ifp),
1360 	    ip6_sprintf(ip6buf, &ia->ia_addr.sin6_addr)));
1361 
1362 	/*
1363 	 * Send NS packet for DAD, ip6_dad_count times.
1364 	 * Note that we must delay the first transmission, if this is the
1365 	 * first packet to be sent from the interface after interface
1366 	 * (re)initialization.
1367 	 */
1368 	dp->dad_ifa = ifa;
1369 	ifa_ref(dp->dad_ifa);
1370 	dp->dad_count = V_ip6_dad_count;
1371 	dp->dad_ns_icount = dp->dad_na_icount = 0;
1372 	dp->dad_ns_ocount = dp->dad_ns_tcount = 0;
1373 	dp->dad_ns_lcount = dp->dad_loopbackprobe = 0;
1374 
1375 	/* Add this to the dadq and add a reference for the dadq. */
1376 	refcount_init(&dp->dad_refcnt, 1);
1377 	nd6_dad_add(dp);
1378 	nd6_dad_starttimer(dp, delay);
1379 	DADQ_WUNLOCK();
1380 }
1381 
1382 /*
1383  * terminate DAD unconditionally.  used for address removals.
1384  */
1385 void
nd6_dad_stop(struct ifaddr * ifa)1386 nd6_dad_stop(struct ifaddr *ifa)
1387 {
1388 	struct dadq *dp;
1389 
1390 	DADQ_WLOCK();
1391 	dp = nd6_dad_find(ifa, NULL);
1392 	if (dp == NULL) {
1393 		DADQ_WUNLOCK();
1394 		/* DAD wasn't started yet */
1395 		return;
1396 	}
1397 
1398 	/*
1399 	 * Acquire a temporary reference so that we can safely stop the callout.
1400 	 */
1401 	(void)refcount_acquire(&dp->dad_refcnt);
1402 	nd6_dad_del(dp);
1403 	DADQ_WUNLOCK();
1404 
1405 	nd6_dad_stoptimer(dp);
1406 	nd6_dad_rele(dp);
1407 }
1408 
1409 static void
nd6_dad_timer(void * arg)1410 nd6_dad_timer(void *arg)
1411 {
1412 	struct dadq *dp = arg;
1413 	struct ifaddr *ifa = dp->dad_ifa;
1414 	struct ifnet *ifp = dp->dad_ifa->ifa_ifp;
1415 	struct in6_ifaddr *ia = (struct in6_ifaddr *)ifa;
1416 	char ip6buf[INET6_ADDRSTRLEN];
1417 	struct epoch_tracker et;
1418 
1419 	CURVNET_SET(dp->dad_vnet);
1420 	KASSERT(ia != NULL, ("DAD entry %p with no address", dp));
1421 
1422 	NET_EPOCH_ENTER(et);
1423 	if (ifp->if_inet6->nd_flags & ND6_IFF_IFDISABLED) {
1424 		/* Do not need DAD for ifdisabled interface. */
1425 		log(LOG_ERR, "nd6_dad_timer: cancel DAD on %s because of "
1426 		    "ND6_IFF_IFDISABLED.\n", ifp->if_xname);
1427 		goto err;
1428 	}
1429 	if (ia->ia6_flags & IN6_IFF_DUPLICATED) {
1430 		log(LOG_ERR, "nd6_dad_timer: called with duplicated address "
1431 			"%s(%s)\n",
1432 			ip6_sprintf(ip6buf, &ia->ia_addr.sin6_addr),
1433 			ifa->ifa_ifp ? if_name(ifa->ifa_ifp) : "???");
1434 		goto err;
1435 	}
1436 	if ((ia->ia6_flags & IN6_IFF_TENTATIVE) == 0) {
1437 		log(LOG_ERR, "nd6_dad_timer: called with non-tentative address "
1438 			"%s(%s)\n",
1439 			ip6_sprintf(ip6buf, &ia->ia_addr.sin6_addr),
1440 			ifa->ifa_ifp ? if_name(ifa->ifa_ifp) : "???");
1441 		goto err;
1442 	}
1443 
1444 	/* Stop DAD if the interface is down even after dad_maxtry attempts. */
1445 	if ((dp->dad_ns_tcount > V_dad_maxtry) &&
1446 	    (((ifp->if_flags & IFF_UP) == 0) ||
1447 	     ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0))) {
1448 		nd6log((LOG_INFO, "%s: could not run DAD "
1449 		    "because the interface was down or not running.\n",
1450 		    if_name(ifa->ifa_ifp)));
1451 		goto err;
1452 	}
1453 
1454 	/* Need more checks? */
1455 	if (dp->dad_ns_ocount < dp->dad_count) {
1456 		/*
1457 		 * We have more NS to go.  Send NS packet for DAD.
1458 		 */
1459 		nd6_dad_starttimer(dp,
1460 		    (long)ifa->ifa_ifp->if_inet6->nd_retrans * hz / 1000);
1461 		nd6_dad_ns_output(dp);
1462 		goto done;
1463 	} else {
1464 		/*
1465 		 * We have transmitted sufficient number of DAD packets.
1466 		 * See what we've got.
1467 		 */
1468 		if (dp->dad_ns_icount > 0 || dp->dad_na_icount > 0) {
1469 			/* We've seen NS or NA, means DAD has failed. */
1470 			nd6_dad_duplicated(ifa, dp);
1471 		} else if (V_dad_enhanced != 0 &&
1472 		    dp->dad_ns_lcount > 0 &&
1473 		    dp->dad_ns_lcount > dp->dad_loopbackprobe) {
1474 			/*
1475 			 * Sec. 4.1 in RFC 7527 requires transmission of
1476 			 * additional probes until the loopback condition
1477 			 * becomes clear when a looped back probe is detected.
1478 			 */
1479 			log(LOG_ERR, "%s: a looped back NS message is "
1480 			    "detected during DAD for %s.  "
1481 			    "Another DAD probes are being sent.\n",
1482 			    if_name(ifa->ifa_ifp),
1483 			    ip6_sprintf(ip6buf, IFA_IN6(ifa)));
1484 			dp->dad_loopbackprobe = dp->dad_ns_lcount;
1485 			/*
1486 			 * Send an NS immediately and increase dad_count by
1487 			 * V_nd6_mmaxtries - 1.
1488 			 */
1489 			dp->dad_count =
1490 			    dp->dad_ns_ocount + V_nd6_mmaxtries - 1;
1491 			nd6_dad_starttimer(dp,
1492 			    (long)ifa->ifa_ifp->if_inet6->nd_retrans * hz / 1000);
1493 			nd6_dad_ns_output(dp);
1494 			goto done;
1495 		} else {
1496 			/*
1497 			 * We are done with DAD.  No NA came, no NS came.
1498 			 * No duplicate address found.  Check IFDISABLED flag
1499 			 * again in case that it is changed between the
1500 			 * beginning of this function and here.
1501 			 *
1502 			 * Reset DAD failures counter if using stable addresses.
1503 			 */
1504 			if ((ifp->if_inet6->nd_flags & ND6_IFF_IFDISABLED) == 0) {
1505 				ia->ia6_flags &= ~IN6_IFF_TENTATIVE;
1506 				if ((ifp->if_inet6->nd_flags & ND6_IFF_STABLEADDR) && !(ia->ia6_flags & IN6_IFF_TEMPORARY))
1507 					atomic_store_int(&DAD_FAILURES(ifp), 0);
1508 				/*
1509 				 * RFC 9131 Section 6.1.2: The first advertisement
1510 				 * SHOULD be sent as soon as an address changes the
1511 				 * state from tentative to preferred.
1512 				 */
1513 				nd6_grand_start(ifa, ND6_QUEUE_FLAG_NEWGUA);
1514 			}
1515 
1516 			nd6log((LOG_DEBUG,
1517 			    "%s: DAD complete for %s - no duplicates found\n",
1518 			    if_name(ifa->ifa_ifp),
1519 			    ip6_sprintf(ip6buf, &ia->ia_addr.sin6_addr)));
1520 			if (dp->dad_ns_lcount > 0)
1521 				log(LOG_ERR, "%s: DAD completed while "
1522 				    "a looped back NS message is detected "
1523 				    "during DAD for %s.\n",
1524 				    if_name(ifa->ifa_ifp),
1525 				    ip6_sprintf(ip6buf, IFA_IN6(ifa)));
1526 		}
1527 	}
1528 err:
1529 	nd6_dad_del(dp);
1530 	DADQ_WUNLOCK();
1531 done:
1532 	NET_EPOCH_EXIT(et);
1533 	CURVNET_RESTORE();
1534 }
1535 
1536 static void
nd6_dad_duplicated(struct ifaddr * ifa,struct dadq * dp)1537 nd6_dad_duplicated(struct ifaddr *ifa, struct dadq *dp)
1538 {
1539 	struct in6_ifaddr *ia = (struct in6_ifaddr *)ifa;
1540 	struct ifnet *ifp;
1541 	char ip6buf[INET6_ADDRSTRLEN];
1542 
1543 	ifp = ifa->ifa_ifp;
1544 
1545 	log(LOG_ERR, "%s: DAD detected duplicate IPv6 address %s: "
1546 	    "NS in/out/loopback=%d/%d/%d, NA in=%d\n",
1547 	    if_name(ifp), ip6_sprintf(ip6buf, &ia->ia_addr.sin6_addr),
1548 	    dp->dad_ns_icount, dp->dad_ns_ocount, dp->dad_ns_lcount,
1549 	    dp->dad_na_icount);
1550 
1551 	ia->ia6_flags &= ~IN6_IFF_TENTATIVE;
1552 	ia->ia6_flags |= IN6_IFF_DUPLICATED;
1553 
1554 	log(LOG_ERR, "%s: DAD complete for %s - duplicate found\n",
1555 	    if_name(ifp), ip6_sprintf(ip6buf, &ia->ia_addr.sin6_addr));
1556 
1557 	/*
1558 	 * For RFC 7217 stable addresses, increment failure counter here if we still have retries.
1559 	 * More addresses will be generated as long as retries are not exhausted.
1560 	 */
1561 	if ((ifp->if_inet6->nd_flags & ND6_IFF_STABLEADDR) && !(ia->ia6_flags & IN6_IFF_TEMPORARY)) {
1562 		u_int dad_failures = atomic_load_int(&DAD_FAILURES(ifp));
1563 
1564 		if (dad_failures <= V_ip6_stableaddr_maxretries) {
1565 			atomic_add_int(&DAD_FAILURES(ifp), 1);
1566 			/* if retries exhausted, output an informative error message */
1567 			if (dad_failures == V_ip6_stableaddr_maxretries)
1568 				log(LOG_ERR, "%s: manual intervention required, consider disabling \"stableaddr\" on the interface"
1569 				    " or checking hostuuid for uniqueness\n",
1570 				    if_name(ifp));
1571 		}
1572 	} else {
1573 		log(LOG_ERR, "%s: manual intervention required\n",
1574 		    if_name(ifp));
1575 	}
1576 
1577 	/*
1578 	 * If the address is a link-local address formed from an interface
1579 	 * identifier based on the hardware address which is supposed to be
1580 	 * uniquely assigned (e.g., EUI-64 for an Ethernet interface), IP
1581 	 * operation on the interface SHOULD be disabled.
1582 	 * [RFC 4862, Section 5.4.5]
1583 	 */
1584 	if (IN6_IS_ADDR_LINKLOCAL(&ia->ia_addr.sin6_addr)) {
1585 		struct in6_addr in6;
1586 
1587 		/*
1588 		 * To avoid over-reaction, we only apply this logic when we are
1589 		 * very sure that hardware addresses are supposed to be unique.
1590 		 */
1591 		switch (ifp->if_type) {
1592 		case IFT_ETHER:
1593 		case IFT_ATM:
1594 		case IFT_IEEE1394:
1595 		case IFT_INFINIBAND:
1596 			in6 = ia->ia_addr.sin6_addr;
1597 			if (in6_get_hw_ifid(ifp, &in6) == 0 &&
1598 			    IN6_ARE_ADDR_EQUAL(&ia->ia_addr.sin6_addr, &in6)) {
1599 				ifp->if_inet6->nd_flags |= ND6_IFF_IFDISABLED;
1600 				log(LOG_ERR, "%s: possible hardware address "
1601 				    "duplication detected, disable IPv6\n",
1602 				    if_name(ifp));
1603 			}
1604 			break;
1605 		}
1606 	}
1607 }
1608 
1609 /*
1610  * Transmit a neighbour solicitation for the purpose of DAD.  Returns with the
1611  * DAD queue unlocked.
1612  */
1613 static void
nd6_dad_ns_output(struct dadq * dp)1614 nd6_dad_ns_output(struct dadq *dp)
1615 {
1616 	struct in6_ifaddr *ia = (struct in6_ifaddr *)dp->dad_ifa;
1617 	struct ifnet *ifp = dp->dad_ifa->ifa_ifp;
1618 	int i;
1619 
1620 	DADQ_WLOCK_ASSERT();
1621 
1622 	dp->dad_ns_tcount++;
1623 	if ((ifp->if_flags & IFF_UP) == 0) {
1624 		DADQ_WUNLOCK();
1625 		return;
1626 	}
1627 	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) {
1628 		DADQ_WUNLOCK();
1629 		return;
1630 	}
1631 
1632 	dp->dad_ns_ocount++;
1633 	if (V_dad_enhanced != 0) {
1634 		for (i = 0; i < ND_OPT_NONCE_LEN32; i++)
1635 			dp->dad_nonce[i] = arc4random();
1636 		/*
1637 		 * XXXHRS: Note that in the case that
1638 		 * DupAddrDetectTransmits > 1, multiple NS messages with
1639 		 * different nonces can be looped back in an unexpected
1640 		 * order.  The current implementation recognizes only
1641 		 * the latest nonce on the sender side.  Practically it
1642 		 * should work well in almost all cases.
1643 		 */
1644 	}
1645 	DADQ_WUNLOCK();
1646 	nd6_ns_output(ifp, NULL, NULL, &ia->ia_addr.sin6_addr,
1647 	    (uint8_t *)&dp->dad_nonce[0]);
1648 }
1649 
1650 static void
nd6_dad_ns_input(struct ifaddr * ifa,struct nd_opt_nonce * ndopt_nonce)1651 nd6_dad_ns_input(struct ifaddr *ifa, struct nd_opt_nonce *ndopt_nonce)
1652 {
1653 	struct dadq *dp;
1654 
1655 	if (ifa == NULL)
1656 		panic("ifa == NULL in nd6_dad_ns_input");
1657 
1658 	/* Ignore Nonce option when Enhanced DAD is disabled. */
1659 	if (V_dad_enhanced == 0)
1660 		ndopt_nonce = NULL;
1661 	DADQ_RLOCK();
1662 	dp = nd6_dad_find(ifa, ndopt_nonce);
1663 	if (dp != NULL)
1664 		dp->dad_ns_icount++;
1665 	DADQ_RUNLOCK();
1666 }
1667 
1668 static void
nd6_dad_na_input(struct ifaddr * ifa)1669 nd6_dad_na_input(struct ifaddr *ifa)
1670 {
1671 	struct dadq *dp;
1672 
1673 	if (ifa == NULL)
1674 		panic("ifa == NULL in nd6_dad_na_input");
1675 
1676 	DADQ_RLOCK();
1677 	dp = nd6_dad_find(ifa, NULL);
1678 	if (dp != NULL)
1679 		dp->dad_na_icount++;
1680 	DADQ_RUNLOCK();
1681 }
1682 
1683 static void
nd6_queue_rel(void * arg)1684 nd6_queue_rel(void *arg)
1685 {
1686 	struct nd_queue *ndq = arg;
1687 	struct ifaddr *ifa;
1688 
1689 	ifa = ndq->ndq_ifa;
1690 	IF_ADDR_WLOCK_ASSERT(ifa->ifa_ifp);
1691 
1692 	/* Remove ndq from the nd_queue list and free it */
1693 	TAILQ_REMOVE(&ifa->ifa_ifp->if_inet6->nd_queue, ndq, ndq_list);
1694 	IF_ADDR_WUNLOCK(ifa->ifa_ifp);
1695 
1696 	free(ndq, M_IP6NDP);
1697 	ifa_free(ifa);
1698 }
1699 
1700 static void
nd6_queue_timer(void * arg)1701 nd6_queue_timer(void *arg)
1702 {
1703 	struct nd_queue *ndq = arg;
1704 	struct ifaddr *ifa = ndq->ndq_ifa;
1705 	struct ifnet *ifp;
1706 	struct in6_addr daddr, taddr;
1707 	struct sockaddr_dl sdl;
1708 	struct epoch_tracker et;
1709 	int delay, tlladdr;
1710 	u_long flags;
1711 	bool proxy;
1712 
1713 	KASSERT(ifa != NULL, ("ND6 queue entry %p with no address", ndq));
1714 
1715 	ifp = ifa->ifa_ifp;
1716 	CURVNET_SET(ifp->if_vnet);
1717 	NET_EPOCH_ENTER(et);
1718 
1719 	daddr = ndq->ndq_daddr;
1720 	taddr = ndq->ndq_taddr;
1721 	tlladdr = ND6_NA_OPT_LLA;
1722 	flags = (V_ip6_forwarding) ? ND_NA_FLAG_ROUTER : 0;
1723 	if ((ifp->if_inet6->nd_flags & ND6_IFF_ACCEPT_RTADV) != 0 && V_ip6_norbit_raif)
1724 		flags &= ~ND_NA_FLAG_ROUTER;
1725 
1726 	/*
1727 	 * RFC 9131 Section 6.1.2: If the address is preferred,
1728 	 * then the Override flag SHOULD NOT be set.
1729 	 */
1730 	if ((ndq->ndq_flags & ND6_QUEUE_FLAG_NEWGUA) != 0) {
1731 		/*
1732 		 * XXX: If the address is in the Optimistic state,
1733 		 * then the Override flag MUST NOT be set.
1734 		 * We don't support RFC 4429 yet.
1735 		 */
1736 		if ((ifp->if_inet6->nd_flags & ND6_IFF_PREFER_SOURCE) == 0)
1737 			flags |= ND_NA_FLAG_OVERRIDE;
1738 	}
1739 	/*
1740 	 * RFC 4861 Section 7.2.6: if link-layer address changed,
1741 	 * The Override flag MAY be set to either zero or one.
1742 	 */
1743 	if ((ndq->ndq_flags & ND6_QUEUE_FLAG_LLADDR) != 0)
1744 		flags |= ND_NA_FLAG_OVERRIDE;
1745 	/* anycast advertisement delay rule (RFC 4861 7.2.7, SHOULD) */
1746 	if ((ndq->ndq_flags & ND6_QUEUE_FLAG_ANYCAST) != 0)
1747 		flags |= ND_NA_FLAG_SOLICITED;
1748 	/* proxy advertisement delay rule (RFC 4861 7.2.8, SHOULD) */
1749 	proxy = false;
1750 	if ((ndq->ndq_flags & ND6_QUEUE_FLAG_PROXY) != 0) {
1751 		flags |= ND_NA_FLAG_SOLICITED;
1752 		sdl = ndq->ndq_sdl;
1753 		proxy = true;
1754 	}
1755 
1756 	/*
1757 	 * if it was GRAND, wait at least a RetransTimer
1758 	 * before removing from queue.
1759 	 */
1760 	if ((ndq->ndq_flags & ND6_QUEUE_GRAND_MASK) != 0) {
1761 		delay = ifp->if_inet6->nd_retrans * hz / 1000;
1762 		callout_reset(&ndq->ndq_callout, delay, nd6_queue_rel, ndq);
1763 		IF_ADDR_WUNLOCK(ifp);
1764 	} else
1765 		nd6_queue_rel(ndq);
1766 
1767 	if (__predict_true(in6_setscope(&daddr, ifp, NULL) == 0))
1768 		nd6_na_output_fib(ifp, &daddr, &taddr, flags, tlladdr,
1769 		    proxy ? (struct sockaddr *)&sdl : NULL, ifp->if_fib);
1770 
1771 	NET_EPOCH_EXIT(et);
1772 	CURVNET_RESTORE();
1773 }
1774 
1775 /*
1776  * Queue a delayed IPv6 Neighbor Advertisement.
1777  *
1778  * daddr: destination address (who the NA is sent to)
1779  * taddr: target address being advertised (used for proxy NAs)
1780  * sdl: link-layer address (used for proxy NAs)
1781  */
1782 static void
nd6_queue_add(struct ifaddr * ifa,struct in6_addr * daddr,struct in6_addr * taddr,struct sockaddr_dl * sdl,int delay,uint32_t flags)1783 nd6_queue_add(struct ifaddr *ifa, struct in6_addr *daddr,
1784     struct in6_addr *taddr, struct sockaddr_dl *sdl, int delay, uint32_t flags)
1785 {
1786 	struct nd_queue *ndq = NULL;
1787 	struct ifnet *ifp;
1788 	struct in6_ifextra *ext;
1789 	char ip6buf[INET6_ADDRSTRLEN];
1790 
1791 	NET_EPOCH_ASSERT();
1792 
1793 	ifp = ifa->ifa_ifp;
1794 	ext = ifp->if_inet6;
1795 	IF_ADDR_WLOCK(ifp);
1796 	/*
1797 	 * if request comes from GRAND, check whether another delayed
1798 	 * GRAND NA exists in the queue.
1799 	 * If it exists, cancel previous one and reuse its ndq.
1800 	 */
1801 	if ((flags & ND6_QUEUE_GRAND_MASK) != 0) {
1802 		TAILQ_FOREACH(ndq, &ext->nd_queue, ndq_list) {
1803 			if (ndq->ndq_ifa == ifa &&
1804 			    (flags & ND6_QUEUE_GRAND_MASK) != 0)
1805 				break;
1806 		}
1807 	}
1808 	if (ndq == NULL) {
1809 		ndq = malloc(sizeof(*ndq), M_IP6NDP, M_NOWAIT | M_ZERO);
1810 		if (ndq == NULL) {
1811 			log(LOG_ERR, "%s: memory allocation failed for %s(%s)\n",
1812 			    __func__, ip6_sprintf(ip6buf, IFA_IN6(ifa)),
1813 			    ifp ? if_name(ifp) : "???");
1814 			IF_ADDR_WUNLOCK(ifp);
1815 			return;
1816 		}
1817 
1818 		callout_init_mtx(&ndq->ndq_callout, &ifp->if_addr_lock,
1819 		    CALLOUT_TRYLOCK | CALLOUT_RETURNUNLOCKED);
1820 		ifa_ref(ifa);
1821 		ndq->ndq_ifa = ifa;
1822 		TAILQ_INSERT_TAIL(&ext->nd_queue, ndq, ndq_list);
1823 	}
1824 
1825 	memcpy(&ndq->ndq_daddr, daddr, sizeof(struct in6_addr));
1826 	/*
1827 	 * For proxy NAs, the target address (taddr) being advertised differs from
1828 	 * the interface address (ifa), so we must explicitly store both the proxy
1829 	 * target address and its link-layer address (sdl).
1830 	 * For non-proxy NAs, use the interface address (ifa) itself as the target.
1831 	 */
1832 	if ((flags & ND6_QUEUE_FLAG_PROXY) != 0) {
1833 		memcpy(&ndq->ndq_taddr, taddr, sizeof(struct in6_addr));
1834 		memcpy(&ndq->ndq_sdl, sdl, sizeof(struct sockaddr_dl));
1835 	} else
1836 		memcpy(&ndq->ndq_taddr, IFA_IN6(ifa), sizeof(struct in6_addr));
1837 	ndq->ndq_flags = flags;
1838 
1839 	nd6log((LOG_DEBUG, "%s: delay IPv6 NA for %s\n", if_name(ifp),
1840 	    ip6_sprintf(ip6buf, IFA_IN6(ifa))));
1841 	callout_reset(&ndq->ndq_callout, delay, nd6_queue_timer, ndq);
1842 	IF_ADDR_WUNLOCK(ifp);
1843 }
1844 
1845 /*
1846  * Start Gratuitous Neighbor Discovery (GRAND) for specified address.
1847  * Called after DAD completes and by interface link layer change event.
1848  */
1849 void
nd6_grand_start(struct ifaddr * ifa,uint32_t flags)1850 nd6_grand_start(struct ifaddr *ifa, uint32_t flags)
1851 {
1852 	struct nd_queue *ndq;
1853 	struct in6_ifextra *ext = ifa->ifa_ifp->if_inet6;
1854 	struct in6_addr daddr = IN6ADDR_ANY_INIT;
1855 	int delay, count = 0;
1856 
1857 	NET_EPOCH_ASSERT();
1858 	/* If we don't need GRAND, don't do it. */
1859 	if (V_ip6_grand_count == 0 ||
1860 	    ifa->ifa_carp != NULL)
1861 		return;
1862 
1863 	/* Check if new address is global */
1864 	if ((flags & ND6_QUEUE_FLAG_NEWGUA) != 0 &&
1865 	    in6_addrscope(IFA_IN6(ifa)) != IPV6_ADDR_SCOPE_GLOBAL)
1866 		return;
1867 
1868 	/*
1869 	 * RFC 9131 Section 6.1.2: These advertisements MUST be
1870 	 * separated by at least RetransTimer seconds.
1871 	 */
1872 	TAILQ_FOREACH(ndq, &ext->nd_queue, ndq_list) {
1873 		/*
1874 		 * RFC 9131 Section 6.1.2: a node SHOULD send
1875 		 * up to MAX_NEIGHBOR_ADVERTISEMENT Neighbor Advertisement messages.
1876 		 * Make sure we don't queue GRAND more than V_ip6_grand_count
1877 		 * per interface.
1878 		 * Since this limitation only applies to GRAND, don't
1879 		 * count non-GRAND ndq.
1880 		 */
1881 		if ((ndq->ndq_flags & ND6_QUEUE_GRAND_MASK) == 0)
1882 			continue;
1883 
1884 		count++;
1885 		if (count >= V_ip6_grand_count)
1886 			return;
1887 	}
1888 
1889 	/*
1890 	 * RFC 9131 Section 6.1.2: if new global address added,
1891 	 * use the all-routers multicast address.
1892 	 */
1893 	if ((flags & ND6_QUEUE_FLAG_NEWGUA) != 0)
1894 		daddr = in6addr_linklocal_allrouters;
1895 
1896 	/*
1897 	 * RFC 4861 Section 7.2.6: if link-layer address changed,
1898 	 * use the all-nodes multicast address.
1899 	 */
1900 	if ((flags & ND6_QUEUE_FLAG_LLADDR) != 0)
1901 		daddr = in6addr_linklocal_allnodes;
1902 
1903 	delay = ext->nd_retrans * hz / 1000;
1904 	nd6_queue_add(ifa, &daddr, NULL, NULL, count * delay, flags);
1905 }
1906 
1907 /*
1908  * drain nd6 queue. used for address removals.
1909  */
1910 void
nd6_queue_stop(struct ifaddr * ifa)1911 nd6_queue_stop(struct ifaddr *ifa)
1912 {
1913 	struct nd_queue *ndq, *dndq;
1914 	struct ifnet *ifp;
1915 
1916 	ifp = ifa->ifa_ifp;
1917 	IF_ADDR_WLOCK(ifp);
1918 	TAILQ_FOREACH_SAFE(ndq, &ifp->if_inet6->nd_queue, ndq_list, dndq) {
1919 		if (ndq->ndq_ifa != ifa)
1920 			continue;
1921 
1922 		callout_stop(&ndq->ndq_callout);
1923 
1924 		/* Remove ndq from the nd_queue list and free it */
1925 		TAILQ_REMOVE(&ifa->ifa_ifp->if_inet6->nd_queue, ndq, ndq_list);
1926 		free(ndq, M_IP6NDP);
1927 		ifa_free(ifa);
1928 	}
1929 	IF_ADDR_WUNLOCK(ifp);
1930 }
1931