xref: /freebsd/sys/netinet6/in6_src.c (revision 1e413cf93298b5b97441a21d9a50fdcd0ee9945e)
1 /*-
2  * Copyright (C) 1995, 1996, 1997, and 1998 WIDE Project.
3  * All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice, this list of conditions and the following disclaimer.
10  * 2. Redistributions in binary form must reproduce the above copyright
11  *    notice, this list of conditions and the following disclaimer in the
12  *    documentation and/or other materials provided with the distribution.
13  * 3. Neither the name of the project nor the names of its contributors
14  *    may be used to endorse or promote products derived from this software
15  *    without specific prior written permission.
16  *
17  * THIS SOFTWARE IS PROVIDED BY THE PROJECT AND CONTRIBUTORS ``AS IS'' AND
18  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20  * ARE DISCLAIMED.  IN NO EVENT SHALL THE PROJECT OR CONTRIBUTORS BE LIABLE
21  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27  * SUCH DAMAGE.
28  *
29  *	$KAME: in6_src.c,v 1.132 2003/08/26 04:42:27 keiichi Exp $
30  */
31 
32 /*-
33  * Copyright (c) 1982, 1986, 1991, 1993
34  *	The Regents of the University of California.  All rights reserved.
35  *
36  * Redistribution and use in source and binary forms, with or without
37  * modification, are permitted provided that the following conditions
38  * are met:
39  * 1. Redistributions of source code must retain the above copyright
40  *    notice, this list of conditions and the following disclaimer.
41  * 2. Redistributions in binary form must reproduce the above copyright
42  *    notice, this list of conditions and the following disclaimer in the
43  *    documentation and/or other materials provided with the distribution.
44  * 4. Neither the name of the University nor the names of its contributors
45  *    may be used to endorse or promote products derived from this software
46  *    without specific prior written permission.
47  *
48  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
49  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
50  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
51  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
52  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
53  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
54  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
55  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
56  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
57  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
58  * SUCH DAMAGE.
59  *
60  *	@(#)in_pcb.c	8.2 (Berkeley) 1/4/94
61  */
62 
63 #include <sys/cdefs.h>
64 __FBSDID("$FreeBSD$");
65 
66 #include "opt_inet.h"
67 #include "opt_inet6.h"
68 
69 #include <sys/param.h>
70 #include <sys/systm.h>
71 #include <sys/lock.h>
72 #include <sys/malloc.h>
73 #include <sys/mbuf.h>
74 #include <sys/priv.h>
75 #include <sys/protosw.h>
76 #include <sys/socket.h>
77 #include <sys/socketvar.h>
78 #include <sys/sockio.h>
79 #include <sys/sysctl.h>
80 #include <sys/errno.h>
81 #include <sys/time.h>
82 #include <sys/kernel.h>
83 #include <sys/sx.h>
84 
85 #include <net/if.h>
86 #include <net/route.h>
87 
88 #include <netinet/in.h>
89 #include <netinet/in_var.h>
90 #include <netinet/in_systm.h>
91 #include <netinet/ip.h>
92 #include <netinet/in_pcb.h>
93 #include <netinet6/in6_var.h>
94 #include <netinet/ip6.h>
95 #include <netinet6/in6_pcb.h>
96 #include <netinet6/ip6_var.h>
97 #include <netinet6/scope6_var.h>
98 #include <netinet6/nd6.h>
99 
100 static struct mtx addrsel_lock;
101 #define	ADDRSEL_LOCK_INIT()	mtx_init(&addrsel_lock, "addrsel_lock", NULL, MTX_DEF)
102 #define	ADDRSEL_LOCK()		mtx_lock(&addrsel_lock)
103 #define	ADDRSEL_UNLOCK()	mtx_unlock(&addrsel_lock)
104 #define	ADDRSEL_LOCK_ASSERT()	mtx_assert(&addrsel_lock, MA_OWNED)
105 
106 static struct sx addrsel_sxlock;
107 #define	ADDRSEL_SXLOCK_INIT()	sx_init(&addrsel_sxlock, "addrsel_sxlock")
108 #define	ADDRSEL_SLOCK()		sx_slock(&addrsel_sxlock)
109 #define	ADDRSEL_SUNLOCK()	sx_sunlock(&addrsel_sxlock)
110 #define	ADDRSEL_XLOCK()		sx_xlock(&addrsel_sxlock)
111 #define	ADDRSEL_XUNLOCK()	sx_xunlock(&addrsel_sxlock)
112 
113 #define ADDR_LABEL_NOTAPP (-1)
114 struct in6_addrpolicy defaultaddrpolicy;
115 
116 int ip6_prefer_tempaddr = 0;
117 
118 static int selectroute __P((struct sockaddr_in6 *, struct ip6_pktopts *,
119 	struct ip6_moptions *, struct route_in6 *, struct ifnet **,
120 	struct rtentry **, int, int));
121 static int in6_selectif __P((struct sockaddr_in6 *, struct ip6_pktopts *,
122 	struct ip6_moptions *, struct route_in6 *ro, struct ifnet **));
123 
124 static struct in6_addrpolicy *lookup_addrsel_policy(struct sockaddr_in6 *);
125 
126 static void init_policy_queue(void);
127 static int add_addrsel_policyent(struct in6_addrpolicy *);
128 static int delete_addrsel_policyent(struct in6_addrpolicy *);
129 static int walk_addrsel_policy __P((int (*)(struct in6_addrpolicy *, void *),
130 				    void *));
131 static int dump_addrsel_policyent(struct in6_addrpolicy *, void *);
132 static struct in6_addrpolicy *match_addrsel_policy(struct sockaddr_in6 *);
133 
134 /*
135  * Return an IPv6 address, which is the most appropriate for a given
136  * destination and user specified options.
137  * If necessary, this function lookups the routing table and returns
138  * an entry to the caller for later use.
139  */
140 #define REPLACE(r) do {\
141 	if ((r) < sizeof(ip6stat.ip6s_sources_rule) / \
142 		sizeof(ip6stat.ip6s_sources_rule[0])) /* check for safety */ \
143 		ip6stat.ip6s_sources_rule[(r)]++; \
144 	/* printf("in6_selectsrc: replace %s with %s by %d\n", ia_best ? ip6_sprintf(&ia_best->ia_addr.sin6_addr) : "none", ip6_sprintf(&ia->ia_addr.sin6_addr), (r)); */ \
145 	goto replace; \
146 } while(0)
147 #define NEXT(r) do {\
148 	if ((r) < sizeof(ip6stat.ip6s_sources_rule) / \
149 		sizeof(ip6stat.ip6s_sources_rule[0])) /* check for safety */ \
150 		ip6stat.ip6s_sources_rule[(r)]++; \
151 	/* printf("in6_selectsrc: keep %s against %s by %d\n", ia_best ? ip6_sprintf(&ia_best->ia_addr.sin6_addr) : "none", ip6_sprintf(&ia->ia_addr.sin6_addr), (r)); */ \
152 	goto next;		/* XXX: we can't use 'continue' here */ \
153 } while(0)
154 #define BREAK(r) do { \
155 	if ((r) < sizeof(ip6stat.ip6s_sources_rule) / \
156 		sizeof(ip6stat.ip6s_sources_rule[0])) /* check for safety */ \
157 		ip6stat.ip6s_sources_rule[(r)]++; \
158 	goto out;		/* XXX: we can't use 'break' here */ \
159 } while(0)
160 
161 struct in6_addr *
162 in6_selectsrc(struct sockaddr_in6 *dstsock, struct ip6_pktopts *opts,
163     struct ip6_moptions *mopts, struct route_in6 *ro,
164     struct in6_addr *laddr, struct ifnet **ifpp, int *errorp)
165 {
166 	struct in6_addr dst;
167 	struct ifnet *ifp = NULL;
168 	struct in6_ifaddr *ia = NULL, *ia_best = NULL;
169 	struct in6_pktinfo *pi = NULL;
170 	int dst_scope = -1, best_scope = -1, best_matchlen = -1;
171 	struct in6_addrpolicy *dst_policy = NULL, *best_policy = NULL;
172 	u_int32_t odstzone;
173 	int prefer_tempaddr;
174 
175 	dst = dstsock->sin6_addr; /* make a copy for local operation */
176 	*errorp = 0;
177 	if (ifpp)
178 		*ifpp = NULL;
179 
180 	/*
181 	 * If the source address is explicitly specified by the caller,
182 	 * check if the requested source address is indeed a unicast address
183 	 * assigned to the node, and can be used as the packet's source
184 	 * address.  If everything is okay, use the address as source.
185 	 */
186 	if (opts && (pi = opts->ip6po_pktinfo) &&
187 	    !IN6_IS_ADDR_UNSPECIFIED(&pi->ipi6_addr)) {
188 		struct sockaddr_in6 srcsock;
189 		struct in6_ifaddr *ia6;
190 
191 		/* get the outgoing interface */
192 		if ((*errorp = in6_selectif(dstsock, opts, mopts, ro, &ifp))
193 		    != 0) {
194 			return (NULL);
195 		}
196 
197 		/*
198 		 * determine the appropriate zone id of the source based on
199 		 * the zone of the destination and the outgoing interface.
200 		 * If the specified address is ambiguous wrt the scope zone,
201 		 * the interface must be specified; otherwise, ifa_ifwithaddr()
202 		 * will fail matching the address.
203 		 */
204 		bzero(&srcsock, sizeof(srcsock));
205 		srcsock.sin6_family = AF_INET6;
206 		srcsock.sin6_len = sizeof(srcsock);
207 		srcsock.sin6_addr = pi->ipi6_addr;
208 		if (ifp) {
209 			*errorp = in6_setscope(&srcsock.sin6_addr, ifp, NULL);
210 			if (*errorp != 0)
211 				return (NULL);
212 		}
213 
214 		ia6 = (struct in6_ifaddr *)ifa_ifwithaddr((struct sockaddr *)(&srcsock));
215 		if (ia6 == NULL ||
216 		    (ia6->ia6_flags & (IN6_IFF_ANYCAST | IN6_IFF_NOTREADY))) {
217 			*errorp = EADDRNOTAVAIL;
218 			return (NULL);
219 		}
220 		pi->ipi6_addr = srcsock.sin6_addr; /* XXX: this overrides pi */
221 		if (ifpp)
222 			*ifpp = ifp;
223 		return (&ia6->ia_addr.sin6_addr);
224 	}
225 
226 	/*
227 	 * Otherwise, if the socket has already bound the source, just use it.
228 	 */
229 	if (laddr && !IN6_IS_ADDR_UNSPECIFIED(laddr))
230 		return (laddr);
231 
232 	/*
233 	 * If the address is not specified, choose the best one based on
234 	 * the outgoing interface and the destination address.
235 	 */
236 	/* get the outgoing interface */
237 	if ((*errorp = in6_selectif(dstsock, opts, mopts, ro, &ifp)) != 0)
238 		return (NULL);
239 
240 #ifdef DIAGNOSTIC
241 	if (ifp == NULL)	/* this should not happen */
242 		panic("in6_selectsrc: NULL ifp");
243 #endif
244 	*errorp = in6_setscope(&dst, ifp, &odstzone);
245 	if (*errorp != 0)
246 		return (NULL);
247 
248 	for (ia = in6_ifaddr; ia; ia = ia->ia_next) {
249 		int new_scope = -1, new_matchlen = -1;
250 		struct in6_addrpolicy *new_policy = NULL;
251 		u_int32_t srczone, osrczone, dstzone;
252 		struct in6_addr src;
253 		struct ifnet *ifp1 = ia->ia_ifp;
254 
255 		/*
256 		 * We'll never take an address that breaks the scope zone
257 		 * of the destination.  We also skip an address if its zone
258 		 * does not contain the outgoing interface.
259 		 * XXX: we should probably use sin6_scope_id here.
260 		 */
261 		if (in6_setscope(&dst, ifp1, &dstzone) ||
262 		    odstzone != dstzone) {
263 			continue;
264 		}
265 		src = ia->ia_addr.sin6_addr;
266 		if (in6_setscope(&src, ifp, &osrczone) ||
267 		    in6_setscope(&src, ifp1, &srczone) ||
268 		    osrczone != srczone) {
269 			continue;
270 		}
271 
272 		/* avoid unusable addresses */
273 		if ((ia->ia6_flags &
274 		     (IN6_IFF_NOTREADY | IN6_IFF_ANYCAST | IN6_IFF_DETACHED))) {
275 				continue;
276 		}
277 		if (!ip6_use_deprecated && IFA6_IS_DEPRECATED(ia))
278 			continue;
279 
280 		/* Rule 1: Prefer same address */
281 		if (IN6_ARE_ADDR_EQUAL(&dst, &ia->ia_addr.sin6_addr)) {
282 			ia_best = ia;
283 			BREAK(1); /* there should be no better candidate */
284 		}
285 
286 		if (ia_best == NULL)
287 			REPLACE(0);
288 
289 		/* Rule 2: Prefer appropriate scope */
290 		if (dst_scope < 0)
291 			dst_scope = in6_addrscope(&dst);
292 		new_scope = in6_addrscope(&ia->ia_addr.sin6_addr);
293 		if (IN6_ARE_SCOPE_CMP(best_scope, new_scope) < 0) {
294 			if (IN6_ARE_SCOPE_CMP(best_scope, dst_scope) < 0)
295 				REPLACE(2);
296 			NEXT(2);
297 		} else if (IN6_ARE_SCOPE_CMP(new_scope, best_scope) < 0) {
298 			if (IN6_ARE_SCOPE_CMP(new_scope, dst_scope) < 0)
299 				NEXT(2);
300 			REPLACE(2);
301 		}
302 
303 		/*
304 		 * Rule 3: Avoid deprecated addresses.  Note that the case of
305 		 * !ip6_use_deprecated is already rejected above.
306 		 */
307 		if (!IFA6_IS_DEPRECATED(ia_best) && IFA6_IS_DEPRECATED(ia))
308 			NEXT(3);
309 		if (IFA6_IS_DEPRECATED(ia_best) && !IFA6_IS_DEPRECATED(ia))
310 			REPLACE(3);
311 
312 		/* Rule 4: Prefer home addresses */
313 		/*
314 		 * XXX: This is a TODO.  We should probably merge the MIP6
315 		 * case above.
316 		 */
317 
318 		/* Rule 5: Prefer outgoing interface */
319 		if (ia_best->ia_ifp == ifp && ia->ia_ifp != ifp)
320 			NEXT(5);
321 		if (ia_best->ia_ifp != ifp && ia->ia_ifp == ifp)
322 			REPLACE(5);
323 
324 		/*
325 		 * Rule 6: Prefer matching label
326 		 * Note that best_policy should be non-NULL here.
327 		 */
328 		if (dst_policy == NULL)
329 			dst_policy = lookup_addrsel_policy(dstsock);
330 		if (dst_policy->label != ADDR_LABEL_NOTAPP) {
331 			new_policy = lookup_addrsel_policy(&ia->ia_addr);
332 			if (dst_policy->label == best_policy->label &&
333 			    dst_policy->label != new_policy->label)
334 				NEXT(6);
335 			if (dst_policy->label != best_policy->label &&
336 			    dst_policy->label == new_policy->label)
337 				REPLACE(6);
338 		}
339 
340 		/*
341 		 * Rule 7: Prefer public addresses.
342 		 * We allow users to reverse the logic by configuring
343 		 * a sysctl variable, so that privacy conscious users can
344 		 * always prefer temporary addresses.
345 		 */
346 		if (opts == NULL ||
347 		    opts->ip6po_prefer_tempaddr == IP6PO_TEMPADDR_SYSTEM) {
348 			prefer_tempaddr = ip6_prefer_tempaddr;
349 		} else if (opts->ip6po_prefer_tempaddr ==
350 		    IP6PO_TEMPADDR_NOTPREFER) {
351 			prefer_tempaddr = 0;
352 		} else
353 			prefer_tempaddr = 1;
354 		if (!(ia_best->ia6_flags & IN6_IFF_TEMPORARY) &&
355 		    (ia->ia6_flags & IN6_IFF_TEMPORARY)) {
356 			if (prefer_tempaddr)
357 				REPLACE(7);
358 			else
359 				NEXT(7);
360 		}
361 		if ((ia_best->ia6_flags & IN6_IFF_TEMPORARY) &&
362 		    !(ia->ia6_flags & IN6_IFF_TEMPORARY)) {
363 			if (prefer_tempaddr)
364 				NEXT(7);
365 			else
366 				REPLACE(7);
367 		}
368 
369 		/*
370 		 * Rule 8: prefer addresses on alive interfaces.
371 		 * This is a KAME specific rule.
372 		 */
373 		if ((ia_best->ia_ifp->if_flags & IFF_UP) &&
374 		    !(ia->ia_ifp->if_flags & IFF_UP))
375 			NEXT(8);
376 		if (!(ia_best->ia_ifp->if_flags & IFF_UP) &&
377 		    (ia->ia_ifp->if_flags & IFF_UP))
378 			REPLACE(8);
379 
380 		/*
381 		 * Rule 14: Use longest matching prefix.
382 		 * Note: in the address selection draft, this rule is
383 		 * documented as "Rule 8".  However, since it is also
384 		 * documented that this rule can be overridden, we assign
385 		 * a large number so that it is easy to assign smaller numbers
386 		 * to more preferred rules.
387 		 */
388 		new_matchlen = in6_matchlen(&ia->ia_addr.sin6_addr, &dst);
389 		if (best_matchlen < new_matchlen)
390 			REPLACE(14);
391 		if (new_matchlen < best_matchlen)
392 			NEXT(14);
393 
394 		/* Rule 15 is reserved. */
395 
396 		/*
397 		 * Last resort: just keep the current candidate.
398 		 * Or, do we need more rules?
399 		 */
400 		continue;
401 
402 	  replace:
403 		ia_best = ia;
404 		best_scope = (new_scope >= 0 ? new_scope :
405 			      in6_addrscope(&ia_best->ia_addr.sin6_addr));
406 		best_policy = (new_policy ? new_policy :
407 			       lookup_addrsel_policy(&ia_best->ia_addr));
408 		best_matchlen = (new_matchlen >= 0 ? new_matchlen :
409 				 in6_matchlen(&ia_best->ia_addr.sin6_addr,
410 					      &dst));
411 
412 	  next:
413 		continue;
414 
415 	  out:
416 		break;
417 	}
418 
419 	if ((ia = ia_best) == NULL) {
420 		*errorp = EADDRNOTAVAIL;
421 		return (NULL);
422 	}
423 
424 	if (ifpp)
425 		*ifpp = ifp;
426 
427 	return (&ia->ia_addr.sin6_addr);
428 }
429 
430 /*
431  * clone - meaningful only for bsdi and freebsd
432  */
433 static int
434 selectroute(struct sockaddr_in6 *dstsock, struct ip6_pktopts *opts,
435     struct ip6_moptions *mopts, struct route_in6 *ro,
436     struct ifnet **retifp, struct rtentry **retrt, int clone,
437     int norouteok)
438 {
439 	int error = 0;
440 	struct ifnet *ifp = NULL;
441 	struct rtentry *rt = NULL;
442 	struct sockaddr_in6 *sin6_next;
443 	struct in6_pktinfo *pi = NULL;
444 	struct in6_addr *dst = &dstsock->sin6_addr;
445 #if 0
446 	char ip6buf[INET6_ADDRSTRLEN];
447 
448 	if (dstsock->sin6_addr.s6_addr32[0] == 0 &&
449 	    dstsock->sin6_addr.s6_addr32[1] == 0 &&
450 	    !IN6_IS_ADDR_LOOPBACK(&dstsock->sin6_addr)) {
451 		printf("in6_selectroute: strange destination %s\n",
452 		       ip6_sprintf(ip6buf, &dstsock->sin6_addr));
453 	} else {
454 		printf("in6_selectroute: destination = %s%%%d\n",
455 		       ip6_sprintf(ip6buf, &dstsock->sin6_addr),
456 		       dstsock->sin6_scope_id); /* for debug */
457 	}
458 #endif
459 
460 	/* If the caller specify the outgoing interface explicitly, use it. */
461 	if (opts && (pi = opts->ip6po_pktinfo) != NULL && pi->ipi6_ifindex) {
462 		/* XXX boundary check is assumed to be already done. */
463 		ifp = ifnet_byindex(pi->ipi6_ifindex);
464 		if (ifp != NULL &&
465 		    (norouteok || retrt == NULL ||
466 		    IN6_IS_ADDR_MULTICAST(dst))) {
467 			/*
468 			 * we do not have to check or get the route for
469 			 * multicast.
470 			 */
471 			goto done;
472 		} else
473 			goto getroute;
474 	}
475 
476 	/*
477 	 * If the destination address is a multicast address and the outgoing
478 	 * interface for the address is specified by the caller, use it.
479 	 */
480 	if (IN6_IS_ADDR_MULTICAST(dst) &&
481 	    mopts != NULL && (ifp = mopts->im6o_multicast_ifp) != NULL) {
482 		goto done; /* we do not need a route for multicast. */
483 	}
484 
485   getroute:
486 	/*
487 	 * If the next hop address for the packet is specified by the caller,
488 	 * use it as the gateway.
489 	 */
490 	if (opts && opts->ip6po_nexthop) {
491 		struct route_in6 *ron;
492 
493 		sin6_next = satosin6(opts->ip6po_nexthop);
494 
495 		/* at this moment, we only support AF_INET6 next hops */
496 		if (sin6_next->sin6_family != AF_INET6) {
497 			error = EAFNOSUPPORT; /* or should we proceed? */
498 			goto done;
499 		}
500 
501 		/*
502 		 * If the next hop is an IPv6 address, then the node identified
503 		 * by that address must be a neighbor of the sending host.
504 		 */
505 		ron = &opts->ip6po_nextroute;
506 		if ((ron->ro_rt &&
507 		     (ron->ro_rt->rt_flags & (RTF_UP | RTF_LLINFO)) !=
508 		     (RTF_UP | RTF_LLINFO)) ||
509 		    !IN6_ARE_ADDR_EQUAL(&satosin6(&ron->ro_dst)->sin6_addr,
510 		    &sin6_next->sin6_addr)) {
511 			if (ron->ro_rt) {
512 				RTFREE(ron->ro_rt);
513 				ron->ro_rt = NULL;
514 			}
515 			*satosin6(&ron->ro_dst) = *sin6_next;
516 		}
517 		if (ron->ro_rt == NULL) {
518 			rtalloc((struct route *)ron); /* multi path case? */
519 			if (ron->ro_rt == NULL ||
520 			    !(ron->ro_rt->rt_flags & RTF_LLINFO)) {
521 				if (ron->ro_rt) {
522 					RTFREE(ron->ro_rt);
523 					ron->ro_rt = NULL;
524 				}
525 				error = EHOSTUNREACH;
526 				goto done;
527 			}
528 		}
529 		rt = ron->ro_rt;
530 		ifp = rt->rt_ifp;
531 
532 		/*
533 		 * When cloning is required, try to allocate a route to the
534 		 * destination so that the caller can store path MTU
535 		 * information.
536 		 */
537 		if (!clone)
538 			goto done;
539 	}
540 
541 	/*
542 	 * Use a cached route if it exists and is valid, else try to allocate
543 	 * a new one.  Note that we should check the address family of the
544 	 * cached destination, in case of sharing the cache with IPv4.
545 	 */
546 	if (ro) {
547 		if (ro->ro_rt &&
548 		    (!(ro->ro_rt->rt_flags & RTF_UP) ||
549 		     ((struct sockaddr *)(&ro->ro_dst))->sa_family != AF_INET6 ||
550 		     !IN6_ARE_ADDR_EQUAL(&satosin6(&ro->ro_dst)->sin6_addr,
551 		     dst))) {
552 			RTFREE(ro->ro_rt);
553 			ro->ro_rt = (struct rtentry *)NULL;
554 		}
555 		if (ro->ro_rt == (struct rtentry *)NULL) {
556 			struct sockaddr_in6 *sa6;
557 
558 			/* No route yet, so try to acquire one */
559 			bzero(&ro->ro_dst, sizeof(struct sockaddr_in6));
560 			sa6 = (struct sockaddr_in6 *)&ro->ro_dst;
561 			*sa6 = *dstsock;
562 			sa6->sin6_scope_id = 0;
563 
564 			if (clone) {
565 				rtalloc((struct route *)ro);
566 			} else {
567 				ro->ro_rt = rtalloc1(&((struct route *)ro)
568 						     ->ro_dst, 0, 0UL);
569 				if (ro->ro_rt)
570 					RT_UNLOCK(ro->ro_rt);
571 			}
572 		}
573 
574 		/*
575 		 * do not care about the result if we have the nexthop
576 		 * explicitly specified.
577 		 */
578 		if (opts && opts->ip6po_nexthop)
579 			goto done;
580 
581 		if (ro->ro_rt) {
582 			ifp = ro->ro_rt->rt_ifp;
583 
584 			if (ifp == NULL) { /* can this really happen? */
585 				RTFREE(ro->ro_rt);
586 				ro->ro_rt = NULL;
587 			}
588 		}
589 		if (ro->ro_rt == NULL)
590 			error = EHOSTUNREACH;
591 		rt = ro->ro_rt;
592 
593 		/*
594 		 * Check if the outgoing interface conflicts with
595 		 * the interface specified by ipi6_ifindex (if specified).
596 		 * Note that loopback interface is always okay.
597 		 * (this may happen when we are sending a packet to one of
598 		 *  our own addresses.)
599 		 */
600 		if (ifp && opts && opts->ip6po_pktinfo &&
601 		    opts->ip6po_pktinfo->ipi6_ifindex) {
602 			if (!(ifp->if_flags & IFF_LOOPBACK) &&
603 			    ifp->if_index !=
604 			    opts->ip6po_pktinfo->ipi6_ifindex) {
605 				error = EHOSTUNREACH;
606 				goto done;
607 			}
608 		}
609 	}
610 
611   done:
612 	if (ifp == NULL && rt == NULL) {
613 		/*
614 		 * This can happen if the caller did not pass a cached route
615 		 * nor any other hints.  We treat this case an error.
616 		 */
617 		error = EHOSTUNREACH;
618 	}
619 	if (error == EHOSTUNREACH)
620 		ip6stat.ip6s_noroute++;
621 
622 	if (retifp != NULL)
623 		*retifp = ifp;
624 	if (retrt != NULL)
625 		*retrt = rt;	/* rt may be NULL */
626 
627 	return (error);
628 }
629 
630 static int
631 in6_selectif(struct sockaddr_in6 *dstsock, struct ip6_pktopts *opts,
632     struct ip6_moptions *mopts, struct route_in6 *ro, struct ifnet **retifp)
633 {
634 	int error;
635 	struct route_in6 sro;
636 	struct rtentry *rt = NULL;
637 
638 	if (ro == NULL) {
639 		bzero(&sro, sizeof(sro));
640 		ro = &sro;
641 	}
642 
643 	if ((error = selectroute(dstsock, opts, mopts, ro, retifp,
644 				     &rt, 0, 1)) != 0) {
645 		if (ro == &sro && rt && rt == sro.ro_rt)
646 			RTFREE(rt);
647 		return (error);
648 	}
649 
650 	/*
651 	 * do not use a rejected or black hole route.
652 	 * XXX: this check should be done in the L2 output routine.
653 	 * However, if we skipped this check here, we'd see the following
654 	 * scenario:
655 	 * - install a rejected route for a scoped address prefix
656 	 *   (like fe80::/10)
657 	 * - send a packet to a destination that matches the scoped prefix,
658 	 *   with ambiguity about the scope zone.
659 	 * - pick the outgoing interface from the route, and disambiguate the
660 	 *   scope zone with the interface.
661 	 * - ip6_output() would try to get another route with the "new"
662 	 *   destination, which may be valid.
663 	 * - we'd see no error on output.
664 	 * Although this may not be very harmful, it should still be confusing.
665 	 * We thus reject the case here.
666 	 */
667 	if (rt && (rt->rt_flags & (RTF_REJECT | RTF_BLACKHOLE))) {
668 		int flags = (rt->rt_flags & RTF_HOST ? EHOSTUNREACH : ENETUNREACH);
669 
670 		if (ro == &sro && rt && rt == sro.ro_rt)
671 			RTFREE(rt);
672 		return (flags);
673 	}
674 
675 	/*
676 	 * Adjust the "outgoing" interface.  If we're going to loop the packet
677 	 * back to ourselves, the ifp would be the loopback interface.
678 	 * However, we'd rather know the interface associated to the
679 	 * destination address (which should probably be one of our own
680 	 * addresses.)
681 	 */
682 	if (rt && rt->rt_ifa && rt->rt_ifa->ifa_ifp)
683 		*retifp = rt->rt_ifa->ifa_ifp;
684 
685 	if (ro == &sro && rt && rt == sro.ro_rt)
686 		RTFREE(rt);
687 	return (0);
688 }
689 
690 /*
691  * clone - meaningful only for bsdi and freebsd
692  */
693 int
694 in6_selectroute(struct sockaddr_in6 *dstsock, struct ip6_pktopts *opts,
695     struct ip6_moptions *mopts, struct route_in6 *ro,
696     struct ifnet **retifp, struct rtentry **retrt, int clone)
697 {
698 
699 	return (selectroute(dstsock, opts, mopts, ro, retifp,
700 	    retrt, clone, 0));
701 }
702 
703 /*
704  * Default hop limit selection. The precedence is as follows:
705  * 1. Hoplimit value specified via ioctl.
706  * 2. (If the outgoing interface is detected) the current
707  *     hop limit of the interface specified by router advertisement.
708  * 3. The system default hoplimit.
709  */
710 int
711 in6_selecthlim(struct in6pcb *in6p, struct ifnet *ifp)
712 {
713 
714 	if (in6p && in6p->in6p_hops >= 0)
715 		return (in6p->in6p_hops);
716 	else if (ifp)
717 		return (ND_IFINFO(ifp)->chlim);
718 	else if (in6p && !IN6_IS_ADDR_UNSPECIFIED(&in6p->in6p_faddr)) {
719 		struct route_in6 ro6;
720 		struct ifnet *lifp;
721 
722 		bzero(&ro6, sizeof(ro6));
723 		ro6.ro_dst.sin6_family = AF_INET6;
724 		ro6.ro_dst.sin6_len = sizeof(struct sockaddr_in6);
725 		ro6.ro_dst.sin6_addr = in6p->in6p_faddr;
726 		rtalloc((struct route *)&ro6);
727 		if (ro6.ro_rt) {
728 			lifp = ro6.ro_rt->rt_ifp;
729 			RTFREE(ro6.ro_rt);
730 			if (lifp)
731 				return (ND_IFINFO(lifp)->chlim);
732 		} else
733 			return (ip6_defhlim);
734 	}
735 	return (ip6_defhlim);
736 }
737 
738 /*
739  * XXX: this is borrowed from in6_pcbbind(). If possible, we should
740  * share this function by all *bsd*...
741  */
742 int
743 in6_pcbsetport(struct in6_addr *laddr, struct inpcb *inp, struct ucred *cred)
744 {
745 	struct socket *so = inp->inp_socket;
746 	u_int16_t lport = 0, first, last, *lastport;
747 	int count, error = 0, wild = 0;
748 	struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
749 
750 	INP_INFO_WLOCK_ASSERT(pcbinfo);
751 	INP_LOCK_ASSERT(inp);
752 
753 	/* XXX: this is redundant when called from in6_pcbbind */
754 	if ((so->so_options & (SO_REUSEADDR|SO_REUSEPORT)) == 0)
755 		wild = INPLOOKUP_WILDCARD;
756 
757 	inp->inp_flags |= INP_ANONPORT;
758 
759 	if (inp->inp_flags & INP_HIGHPORT) {
760 		first = ipport_hifirstauto;	/* sysctl */
761 		last  = ipport_hilastauto;
762 		lastport = &pcbinfo->ipi_lasthi;
763 	} else if (inp->inp_flags & INP_LOWPORT) {
764 		error = priv_check_cred(cred, PRIV_NETINET_RESERVEDPORT, 0);
765 		if (error)
766 			return error;
767 		first = ipport_lowfirstauto;	/* 1023 */
768 		last  = ipport_lowlastauto;	/* 600 */
769 		lastport = &pcbinfo->ipi_lastlow;
770 	} else {
771 		first = ipport_firstauto;	/* sysctl */
772 		last  = ipport_lastauto;
773 		lastport = &pcbinfo->ipi_lastport;
774 	}
775 	/*
776 	 * Simple check to ensure all ports are not used up causing
777 	 * a deadlock here.
778 	 *
779 	 * We split the two cases (up and down) so that the direction
780 	 * is not being tested on each round of the loop.
781 	 */
782 	if (first > last) {
783 		/*
784 		 * counting down
785 		 */
786 		count = first - last;
787 
788 		do {
789 			if (count-- < 0) {	/* completely used? */
790 				/*
791 				 * Undo any address bind that may have
792 				 * occurred above.
793 				 */
794 				inp->in6p_laddr = in6addr_any;
795 				return (EAGAIN);
796 			}
797 			--*lastport;
798 			if (*lastport > first || *lastport < last)
799 				*lastport = first;
800 			lport = htons(*lastport);
801 		} while (in6_pcblookup_local(pcbinfo, &inp->in6p_laddr,
802 					     lport, wild));
803 	} else {
804 		/*
805 			 * counting up
806 			 */
807 		count = last - first;
808 
809 		do {
810 			if (count-- < 0) {	/* completely used? */
811 				/*
812 				 * Undo any address bind that may have
813 				 * occurred above.
814 				 */
815 				inp->in6p_laddr = in6addr_any;
816 				return (EAGAIN);
817 			}
818 			++*lastport;
819 			if (*lastport < first || *lastport > last)
820 				*lastport = first;
821 			lport = htons(*lastport);
822 		} while (in6_pcblookup_local(pcbinfo,
823 					     &inp->in6p_laddr, lport, wild));
824 	}
825 
826 	inp->inp_lport = lport;
827 	if (in_pcbinshash(inp) != 0) {
828 		inp->in6p_laddr = in6addr_any;
829 		inp->inp_lport = 0;
830 		return (EAGAIN);
831 	}
832 
833 	return (0);
834 }
835 
836 void
837 addrsel_policy_init(void)
838 {
839 	ADDRSEL_LOCK_INIT();
840 	ADDRSEL_SXLOCK_INIT();
841 
842 	init_policy_queue();
843 
844 	/* initialize the "last resort" policy */
845 	bzero(&defaultaddrpolicy, sizeof(defaultaddrpolicy));
846 	defaultaddrpolicy.label = ADDR_LABEL_NOTAPP;
847 }
848 
849 static struct in6_addrpolicy *
850 lookup_addrsel_policy(struct sockaddr_in6 *key)
851 {
852 	struct in6_addrpolicy *match = NULL;
853 
854 	ADDRSEL_LOCK();
855 	match = match_addrsel_policy(key);
856 
857 	if (match == NULL)
858 		match = &defaultaddrpolicy;
859 	else
860 		match->use++;
861 	ADDRSEL_UNLOCK();
862 
863 	return (match);
864 }
865 
866 /*
867  * Subroutines to manage the address selection policy table via sysctl.
868  */
869 struct walkarg {
870 	struct sysctl_req *w_req;
871 };
872 
873 static int in6_src_sysctl(SYSCTL_HANDLER_ARGS);
874 SYSCTL_DECL(_net_inet6_ip6);
875 SYSCTL_NODE(_net_inet6_ip6, IPV6CTL_ADDRCTLPOLICY, addrctlpolicy,
876 	CTLFLAG_RD, in6_src_sysctl, "");
877 
878 static int
879 in6_src_sysctl(SYSCTL_HANDLER_ARGS)
880 {
881 	struct walkarg w;
882 
883 	if (req->newptr)
884 		return EPERM;
885 
886 	bzero(&w, sizeof(w));
887 	w.w_req = req;
888 
889 	return (walk_addrsel_policy(dump_addrsel_policyent, &w));
890 }
891 
892 int
893 in6_src_ioctl(u_long cmd, caddr_t data)
894 {
895 	int i;
896 	struct in6_addrpolicy ent0;
897 
898 	if (cmd != SIOCAADDRCTL_POLICY && cmd != SIOCDADDRCTL_POLICY)
899 		return (EOPNOTSUPP); /* check for safety */
900 
901 	ent0 = *(struct in6_addrpolicy *)data;
902 
903 	if (ent0.label == ADDR_LABEL_NOTAPP)
904 		return (EINVAL);
905 	/* check if the prefix mask is consecutive. */
906 	if (in6_mask2len(&ent0.addrmask.sin6_addr, NULL) < 0)
907 		return (EINVAL);
908 	/* clear trailing garbages (if any) of the prefix address. */
909 	for (i = 0; i < 4; i++) {
910 		ent0.addr.sin6_addr.s6_addr32[i] &=
911 			ent0.addrmask.sin6_addr.s6_addr32[i];
912 	}
913 	ent0.use = 0;
914 
915 	switch (cmd) {
916 	case SIOCAADDRCTL_POLICY:
917 		return (add_addrsel_policyent(&ent0));
918 	case SIOCDADDRCTL_POLICY:
919 		return (delete_addrsel_policyent(&ent0));
920 	}
921 
922 	return (0);		/* XXX: compromise compilers */
923 }
924 
925 /*
926  * The followings are implementation of the policy table using a
927  * simple tail queue.
928  * XXX such details should be hidden.
929  * XXX implementation using binary tree should be more efficient.
930  */
931 struct addrsel_policyent {
932 	TAILQ_ENTRY(addrsel_policyent) ape_entry;
933 	struct in6_addrpolicy ape_policy;
934 };
935 
936 TAILQ_HEAD(addrsel_policyhead, addrsel_policyent);
937 
938 struct addrsel_policyhead addrsel_policytab;
939 
940 static void
941 init_policy_queue(void)
942 {
943 
944 	TAILQ_INIT(&addrsel_policytab);
945 }
946 
947 static int
948 add_addrsel_policyent(struct in6_addrpolicy *newpolicy)
949 {
950 	struct addrsel_policyent *new, *pol;
951 
952 	MALLOC(new, struct addrsel_policyent *, sizeof(*new), M_IFADDR,
953 	       M_WAITOK);
954 	ADDRSEL_XLOCK();
955 	ADDRSEL_LOCK();
956 
957 	/* duplication check */
958 	TAILQ_FOREACH(pol, &addrsel_policytab, ape_entry) {
959 		if (IN6_ARE_ADDR_EQUAL(&newpolicy->addr.sin6_addr,
960 				       &pol->ape_policy.addr.sin6_addr) &&
961 		    IN6_ARE_ADDR_EQUAL(&newpolicy->addrmask.sin6_addr,
962 				       &pol->ape_policy.addrmask.sin6_addr)) {
963 			ADDRSEL_UNLOCK();
964 			ADDRSEL_XUNLOCK();
965 			FREE(new, M_IFADDR);
966 			return (EEXIST);	/* or override it? */
967 		}
968 	}
969 
970 	bzero(new, sizeof(*new));
971 
972 	/* XXX: should validate entry */
973 	new->ape_policy = *newpolicy;
974 
975 	TAILQ_INSERT_TAIL(&addrsel_policytab, new, ape_entry);
976 	ADDRSEL_UNLOCK();
977 	ADDRSEL_XUNLOCK();
978 
979 	return (0);
980 }
981 
982 static int
983 delete_addrsel_policyent(struct in6_addrpolicy *key)
984 {
985 	struct addrsel_policyent *pol;
986 
987 	ADDRSEL_XLOCK();
988 	ADDRSEL_LOCK();
989 
990 	/* search for the entry in the table */
991 	TAILQ_FOREACH(pol, &addrsel_policytab, ape_entry) {
992 		if (IN6_ARE_ADDR_EQUAL(&key->addr.sin6_addr,
993 		    &pol->ape_policy.addr.sin6_addr) &&
994 		    IN6_ARE_ADDR_EQUAL(&key->addrmask.sin6_addr,
995 		    &pol->ape_policy.addrmask.sin6_addr)) {
996 			break;
997 		}
998 	}
999 	if (pol == NULL) {
1000 		ADDRSEL_UNLOCK();
1001 		ADDRSEL_XUNLOCK();
1002 		return (ESRCH);
1003 	}
1004 
1005 	TAILQ_REMOVE(&addrsel_policytab, pol, ape_entry);
1006 	ADDRSEL_UNLOCK();
1007 	ADDRSEL_XUNLOCK();
1008 
1009 	return (0);
1010 }
1011 
1012 static int
1013 walk_addrsel_policy(int (*callback)(struct in6_addrpolicy *, void *),
1014     void *w)
1015 {
1016 	struct addrsel_policyent *pol;
1017 	int error = 0;
1018 
1019 	ADDRSEL_SLOCK();
1020 	TAILQ_FOREACH(pol, &addrsel_policytab, ape_entry) {
1021 		if ((error = (*callback)(&pol->ape_policy, w)) != 0) {
1022 			ADDRSEL_SUNLOCK();
1023 			return (error);
1024 		}
1025 	}
1026 	ADDRSEL_SUNLOCK();
1027 	return (error);
1028 }
1029 
1030 static int
1031 dump_addrsel_policyent(struct in6_addrpolicy *pol, void *arg)
1032 {
1033 	int error = 0;
1034 	struct walkarg *w = arg;
1035 
1036 	error = SYSCTL_OUT(w->w_req, pol, sizeof(*pol));
1037 
1038 	return (error);
1039 }
1040 
1041 static struct in6_addrpolicy *
1042 match_addrsel_policy(struct sockaddr_in6 *key)
1043 {
1044 	struct addrsel_policyent *pent;
1045 	struct in6_addrpolicy *bestpol = NULL, *pol;
1046 	int matchlen, bestmatchlen = -1;
1047 	u_char *mp, *ep, *k, *p, m;
1048 
1049 	TAILQ_FOREACH(pent, &addrsel_policytab, ape_entry) {
1050 		matchlen = 0;
1051 
1052 		pol = &pent->ape_policy;
1053 		mp = (u_char *)&pol->addrmask.sin6_addr;
1054 		ep = mp + 16;	/* XXX: scope field? */
1055 		k = (u_char *)&key->sin6_addr;
1056 		p = (u_char *)&pol->addr.sin6_addr;
1057 		for (; mp < ep && *mp; mp++, k++, p++) {
1058 			m = *mp;
1059 			if ((*k & m) != *p)
1060 				goto next; /* not match */
1061 			if (m == 0xff) /* short cut for a typical case */
1062 				matchlen += 8;
1063 			else {
1064 				while (m >= 0x80) {
1065 					matchlen++;
1066 					m <<= 1;
1067 				}
1068 			}
1069 		}
1070 
1071 		/* matched.  check if this is better than the current best. */
1072 		if (bestpol == NULL ||
1073 		    matchlen > bestmatchlen) {
1074 			bestpol = pol;
1075 			bestmatchlen = matchlen;
1076 		}
1077 
1078 	  next:
1079 		continue;
1080 	}
1081 
1082 	return (bestpol);
1083 }
1084