xref: /freebsd/sys/netinet6/in6_src.c (revision 0efd6615cd5f39b67cec82a7034e655f3b5801e3)
1 /*	$FreeBSD$	*/
2 /*	$KAME: in6_src.c,v 1.132 2003/08/26 04:42:27 keiichi Exp $	*/
3 
4 /*-
5  * Copyright (C) 1995, 1996, 1997, and 1998 WIDE Project.
6  * All rights reserved.
7  *
8  * Redistribution and use in source and binary forms, with or without
9  * modification, are permitted provided that the following conditions
10  * are met:
11  * 1. Redistributions of source code must retain the above copyright
12  *    notice, this list of conditions and the following disclaimer.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in the
15  *    documentation and/or other materials provided with the distribution.
16  * 3. Neither the name of the project nor the names of its contributors
17  *    may be used to endorse or promote products derived from this software
18  *    without specific prior written permission.
19  *
20  * THIS SOFTWARE IS PROVIDED BY THE PROJECT AND CONTRIBUTORS ``AS IS'' AND
21  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
22  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
23  * ARE DISCLAIMED.  IN NO EVENT SHALL THE PROJECT OR CONTRIBUTORS BE LIABLE
24  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
25  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
26  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
27  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
28  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
29  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
30  * SUCH DAMAGE.
31  */
32 
33 /*-
34  * Copyright (c) 1982, 1986, 1991, 1993
35  *	The Regents of the University of California.  All rights reserved.
36  *
37  * Redistribution and use in source and binary forms, with or without
38  * modification, are permitted provided that the following conditions
39  * are met:
40  * 1. Redistributions of source code must retain the above copyright
41  *    notice, this list of conditions and the following disclaimer.
42  * 2. Redistributions in binary form must reproduce the above copyright
43  *    notice, this list of conditions and the following disclaimer in the
44  *    documentation and/or other materials provided with the distribution.
45  * 4. Neither the name of the University nor the names of its contributors
46  *    may be used to endorse or promote products derived from this software
47  *    without specific prior written permission.
48  *
49  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
50  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
51  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
52  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
53  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
54  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
55  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
56  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
57  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
58  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
59  * SUCH DAMAGE.
60  *
61  *	@(#)in_pcb.c	8.2 (Berkeley) 1/4/94
62  */
63 
64 #include "opt_inet.h"
65 #include "opt_inet6.h"
66 
67 #include <sys/param.h>
68 #include <sys/systm.h>
69 #include <sys/malloc.h>
70 #include <sys/mbuf.h>
71 #include <sys/priv.h>
72 #include <sys/protosw.h>
73 #include <sys/socket.h>
74 #include <sys/socketvar.h>
75 #include <sys/sockio.h>
76 #include <sys/sysctl.h>
77 #include <sys/errno.h>
78 #include <sys/time.h>
79 #include <sys/kernel.h>
80 #include <sys/sx.h>
81 
82 #include <net/if.h>
83 #include <net/route.h>
84 
85 #include <netinet/in.h>
86 #include <netinet/in_var.h>
87 #include <netinet/in_systm.h>
88 #include <netinet/ip.h>
89 #include <netinet/in_pcb.h>
90 #include <netinet6/in6_var.h>
91 #include <netinet/ip6.h>
92 #include <netinet6/in6_pcb.h>
93 #include <netinet6/ip6_var.h>
94 #include <netinet6/scope6_var.h>
95 #include <netinet6/nd6.h>
96 
97 static struct mtx addrsel_lock;
98 #define	ADDRSEL_LOCK_INIT()	mtx_init(&addrsel_lock, "addrsel_lock", NULL, MTX_DEF)
99 #define	ADDRSEL_LOCK()		mtx_lock(&addrsel_lock)
100 #define	ADDRSEL_UNLOCK()	mtx_unlock(&addrsel_lock)
101 #define	ADDRSEL_LOCK_ASSERT()	mtx_assert(&addrsel_lock, MA_OWNED)
102 
103 static struct sx addrsel_sxlock;
104 #define	ADDRSEL_SXLOCK_INIT()	sx_init(&addrsel_sxlock, "addrsel_sxlock")
105 #define	ADDRSEL_SLOCK()		sx_slock(&addrsel_sxlock)
106 #define	ADDRSEL_SUNLOCK()	sx_sunlock(&addrsel_sxlock)
107 #define	ADDRSEL_XLOCK()		sx_xlock(&addrsel_sxlock)
108 #define	ADDRSEL_XUNLOCK()	sx_xunlock(&addrsel_sxlock)
109 
110 #define ADDR_LABEL_NOTAPP (-1)
111 struct in6_addrpolicy defaultaddrpolicy;
112 
113 int ip6_prefer_tempaddr = 0;
114 
115 static int selectroute __P((struct sockaddr_in6 *, struct ip6_pktopts *,
116 	struct ip6_moptions *, struct route_in6 *, struct ifnet **,
117 	struct rtentry **, int, int));
118 static int in6_selectif __P((struct sockaddr_in6 *, struct ip6_pktopts *,
119 	struct ip6_moptions *, struct route_in6 *ro, struct ifnet **));
120 
121 static struct in6_addrpolicy *lookup_addrsel_policy __P((struct sockaddr_in6 *));
122 
123 static void init_policy_queue __P((void));
124 static int add_addrsel_policyent __P((struct in6_addrpolicy *));
125 static int delete_addrsel_policyent __P((struct in6_addrpolicy *));
126 static int walk_addrsel_policy __P((int (*)(struct in6_addrpolicy *, void *),
127 				    void *));
128 static int dump_addrsel_policyent __P((struct in6_addrpolicy *, void *));
129 static struct in6_addrpolicy *match_addrsel_policy __P((struct sockaddr_in6 *));
130 
131 /*
132  * Return an IPv6 address, which is the most appropriate for a given
133  * destination and user specified options.
134  * If necessary, this function lookups the routing table and returns
135  * an entry to the caller for later use.
136  */
137 #define REPLACE(r) do {\
138 	if ((r) < sizeof(ip6stat.ip6s_sources_rule) / \
139 		sizeof(ip6stat.ip6s_sources_rule[0])) /* check for safety */ \
140 		ip6stat.ip6s_sources_rule[(r)]++; \
141 	/* 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)); */ \
142 	goto replace; \
143 } while(0)
144 #define NEXT(r) do {\
145 	if ((r) < sizeof(ip6stat.ip6s_sources_rule) / \
146 		sizeof(ip6stat.ip6s_sources_rule[0])) /* check for safety */ \
147 		ip6stat.ip6s_sources_rule[(r)]++; \
148 	/* 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)); */ \
149 	goto next; 		/* XXX: we can't use 'continue' here */ \
150 } while(0)
151 #define BREAK(r) do { \
152 	if ((r) < sizeof(ip6stat.ip6s_sources_rule) / \
153 		sizeof(ip6stat.ip6s_sources_rule[0])) /* check for safety */ \
154 		ip6stat.ip6s_sources_rule[(r)]++; \
155 	goto out; 		/* XXX: we can't use 'break' here */ \
156 } while(0)
157 
158 struct in6_addr *
159 in6_selectsrc(dstsock, opts, mopts, ro, laddr, ifpp, errorp)
160 	struct sockaddr_in6 *dstsock;
161 	struct ip6_pktopts *opts;
162 	struct ip6_moptions *mopts;
163 	struct route_in6 *ro;
164 	struct in6_addr *laddr;
165 	struct ifnet **ifpp;
166 	int *errorp;
167 {
168 	struct in6_addr dst;
169 	struct ifnet *ifp = NULL;
170 	struct in6_ifaddr *ia = NULL, *ia_best = NULL;
171 	struct in6_pktinfo *pi = NULL;
172 	int dst_scope = -1, best_scope = -1, best_matchlen = -1;
173 	struct in6_addrpolicy *dst_policy = NULL, *best_policy = NULL;
174 	u_int32_t odstzone;
175 	int prefer_tempaddr;
176 
177 	dst = dstsock->sin6_addr; /* make a copy for local operation */
178 	*errorp = 0;
179 	if (ifpp)
180 		*ifpp = NULL;
181 
182 	/*
183 	 * If the source address is explicitly specified by the caller,
184 	 * check if the requested source address is indeed a unicast address
185 	 * assigned to the node, and can be used as the packet's source
186 	 * address.  If everything is okay, use the address as source.
187 	 */
188 	if (opts && (pi = opts->ip6po_pktinfo) &&
189 	    !IN6_IS_ADDR_UNSPECIFIED(&pi->ipi6_addr)) {
190 		struct sockaddr_in6 srcsock;
191 		struct in6_ifaddr *ia6;
192 
193 		/* get the outgoing interface */
194 		if ((*errorp = in6_selectif(dstsock, opts, mopts, ro, &ifp))
195 		    != 0) {
196 			return (NULL);
197 		}
198 
199 		/*
200 		 * determine the appropriate zone id of the source based on
201 		 * the zone of the destination and the outgoing interface.
202 		 * If the specified address is ambiguous wrt the scope zone,
203 		 * the interface must be specified; otherwise, ifa_ifwithaddr()
204 		 * will fail matching the address.
205 		 */
206 		bzero(&srcsock, sizeof(srcsock));
207 		srcsock.sin6_family = AF_INET6;
208 		srcsock.sin6_len = sizeof(srcsock);
209 		srcsock.sin6_addr = pi->ipi6_addr;
210 		if (ifp) {
211 			*errorp = in6_setscope(&srcsock.sin6_addr, ifp, NULL);
212 			if (*errorp != 0)
213 				return (NULL);
214 		}
215 
216 		ia6 = (struct in6_ifaddr *)ifa_ifwithaddr((struct sockaddr *)(&srcsock));
217 		if (ia6 == NULL ||
218 		    (ia6->ia6_flags & (IN6_IFF_ANYCAST | IN6_IFF_NOTREADY))) {
219 			*errorp = EADDRNOTAVAIL;
220 			return (NULL);
221 		}
222 		pi->ipi6_addr = srcsock.sin6_addr; /* XXX: this overrides pi */
223 		if (ifpp)
224 			*ifpp = ifp;
225 		return (&ia6->ia_addr.sin6_addr);
226 	}
227 
228 	/*
229 	 * Otherwise, if the socket has already bound the source, just use it.
230 	 */
231 	if (laddr && !IN6_IS_ADDR_UNSPECIFIED(laddr))
232 		return (laddr);
233 
234 	/*
235 	 * If the address is not specified, choose the best one based on
236 	 * the outgoing interface and the destination address.
237 	 */
238 	/* get the outgoing interface */
239 	if ((*errorp = in6_selectif(dstsock, opts, mopts, ro, &ifp)) != 0)
240 		return (NULL);
241 
242 #ifdef DIAGNOSTIC
243 	if (ifp == NULL)	/* this should not happen */
244 		panic("in6_selectsrc: NULL ifp");
245 #endif
246 	*errorp = in6_setscope(&dst, ifp, &odstzone);
247 	if (*errorp != 0)
248 		return (NULL);
249 
250 	for (ia = in6_ifaddr; ia; ia = ia->ia_next) {
251 		int new_scope = -1, new_matchlen = -1;
252 		struct in6_addrpolicy *new_policy = NULL;
253 		u_int32_t srczone, osrczone, dstzone;
254 		struct in6_addr src;
255 		struct ifnet *ifp1 = ia->ia_ifp;
256 
257 		/*
258 		 * We'll never take an address that breaks the scope zone
259 		 * of the destination.  We also skip an address if its zone
260 		 * does not contain the outgoing interface.
261 		 * XXX: we should probably use sin6_scope_id here.
262 		 */
263 		if (in6_setscope(&dst, ifp1, &dstzone) ||
264 		    odstzone != dstzone) {
265 			continue;
266 		}
267 		src = ia->ia_addr.sin6_addr;
268 		if (in6_setscope(&src, ifp, &osrczone) ||
269 		    in6_setscope(&src, ifp1, &srczone) ||
270 		    osrczone != srczone) {
271 			continue;
272 		}
273 
274 		/* avoid unusable addresses */
275 		if ((ia->ia6_flags &
276 		     (IN6_IFF_NOTREADY | IN6_IFF_ANYCAST | IN6_IFF_DETACHED))) {
277 				continue;
278 		}
279 		if (!ip6_use_deprecated && IFA6_IS_DEPRECATED(ia))
280 			continue;
281 
282 		/* Rule 1: Prefer same address */
283 		if (IN6_ARE_ADDR_EQUAL(&dst, &ia->ia_addr.sin6_addr)) {
284 			ia_best = ia;
285 			BREAK(1); /* there should be no better candidate */
286 		}
287 
288 		if (ia_best == NULL)
289 			REPLACE(0);
290 
291 		/* Rule 2: Prefer appropriate scope */
292 		if (dst_scope < 0)
293 			dst_scope = in6_addrscope(&dst);
294 		new_scope = in6_addrscope(&ia->ia_addr.sin6_addr);
295 		if (IN6_ARE_SCOPE_CMP(best_scope, new_scope) < 0) {
296 			if (IN6_ARE_SCOPE_CMP(best_scope, dst_scope) < 0)
297 				REPLACE(2);
298 			NEXT(2);
299 		} else if (IN6_ARE_SCOPE_CMP(new_scope, best_scope) < 0) {
300 			if (IN6_ARE_SCOPE_CMP(new_scope, dst_scope) < 0)
301 				NEXT(2);
302 			REPLACE(2);
303 		}
304 
305 		/*
306 		 * Rule 3: Avoid deprecated addresses.  Note that the case of
307 		 * !ip6_use_deprecated is already rejected above.
308 		 */
309 		if (!IFA6_IS_DEPRECATED(ia_best) && IFA6_IS_DEPRECATED(ia))
310 			NEXT(3);
311 		if (IFA6_IS_DEPRECATED(ia_best) && !IFA6_IS_DEPRECATED(ia))
312 			REPLACE(3);
313 
314 		/* Rule 4: Prefer home addresses */
315 		/*
316 		 * XXX: This is a TODO.  We should probably merge the MIP6
317 		 * case above.
318 		 */
319 
320 		/* Rule 5: Prefer outgoing interface */
321 		if (ia_best->ia_ifp == ifp && ia->ia_ifp != ifp)
322 			NEXT(5);
323 		if (ia_best->ia_ifp != ifp && ia->ia_ifp == ifp)
324 			REPLACE(5);
325 
326 		/*
327 		 * Rule 6: Prefer matching label
328 		 * Note that best_policy should be non-NULL here.
329 		 */
330 		if (dst_policy == NULL)
331 			dst_policy = lookup_addrsel_policy(dstsock);
332 		if (dst_policy->label != ADDR_LABEL_NOTAPP) {
333 			new_policy = lookup_addrsel_policy(&ia->ia_addr);
334 			if (dst_policy->label == best_policy->label &&
335 			    dst_policy->label != new_policy->label)
336 				NEXT(6);
337 			if (dst_policy->label != best_policy->label &&
338 			    dst_policy->label == new_policy->label)
339 				REPLACE(6);
340 		}
341 
342 		/*
343 		 * Rule 7: Prefer public addresses.
344 		 * We allow users to reverse the logic by configuring
345 		 * a sysctl variable, so that privacy conscious users can
346 		 * always prefer temporary addresses.
347 		 */
348 		if (opts == NULL ||
349 		    opts->ip6po_prefer_tempaddr == IP6PO_TEMPADDR_SYSTEM) {
350 			prefer_tempaddr = ip6_prefer_tempaddr;
351 		} else if (opts->ip6po_prefer_tempaddr ==
352 		    IP6PO_TEMPADDR_NOTPREFER) {
353 			prefer_tempaddr = 0;
354 		} else
355 			prefer_tempaddr = 1;
356 		if (!(ia_best->ia6_flags & IN6_IFF_TEMPORARY) &&
357 		    (ia->ia6_flags & IN6_IFF_TEMPORARY)) {
358 			if (prefer_tempaddr)
359 				REPLACE(7);
360 			else
361 				NEXT(7);
362 		}
363 		if ((ia_best->ia6_flags & IN6_IFF_TEMPORARY) &&
364 		    !(ia->ia6_flags & IN6_IFF_TEMPORARY)) {
365 			if (prefer_tempaddr)
366 				NEXT(7);
367 			else
368 				REPLACE(7);
369 		}
370 
371 		/*
372 		 * Rule 8: prefer addresses on alive interfaces.
373 		 * This is a KAME specific rule.
374 		 */
375 		if ((ia_best->ia_ifp->if_flags & IFF_UP) &&
376 		    !(ia->ia_ifp->if_flags & IFF_UP))
377 			NEXT(8);
378 		if (!(ia_best->ia_ifp->if_flags & IFF_UP) &&
379 		    (ia->ia_ifp->if_flags & IFF_UP))
380 			REPLACE(8);
381 
382 		/*
383 		 * Rule 14: Use longest matching prefix.
384 		 * Note: in the address selection draft, this rule is
385 		 * documented as "Rule 8".  However, since it is also
386 		 * documented that this rule can be overridden, we assign
387 		 * a large number so that it is easy to assign smaller numbers
388 		 * to more preferred rules.
389 		 */
390 		new_matchlen = in6_matchlen(&ia->ia_addr.sin6_addr, &dst);
391 		if (best_matchlen < new_matchlen)
392 			REPLACE(14);
393 		if (new_matchlen < best_matchlen)
394 			NEXT(14);
395 
396 		/* Rule 15 is reserved. */
397 
398 		/*
399 		 * Last resort: just keep the current candidate.
400 		 * Or, do we need more rules?
401 		 */
402 		continue;
403 
404 	  replace:
405 		ia_best = ia;
406 		best_scope = (new_scope >= 0 ? new_scope :
407 			      in6_addrscope(&ia_best->ia_addr.sin6_addr));
408 		best_policy = (new_policy ? new_policy :
409 			       lookup_addrsel_policy(&ia_best->ia_addr));
410 		best_matchlen = (new_matchlen >= 0 ? new_matchlen :
411 				 in6_matchlen(&ia_best->ia_addr.sin6_addr,
412 					      &dst));
413 
414 	  next:
415 		continue;
416 
417 	  out:
418 		break;
419 	}
420 
421 	if ((ia = ia_best) == NULL) {
422 		*errorp = EADDRNOTAVAIL;
423 		return (NULL);
424 	}
425 
426 	if (ifpp)
427 		*ifpp = ifp;
428 
429 	return (&ia->ia_addr.sin6_addr);
430 }
431 
432 static int
433 selectroute(dstsock, opts, mopts, ro, retifp, retrt, clone, norouteok)
434 	struct sockaddr_in6 *dstsock;
435 	struct ip6_pktopts *opts;
436 	struct ip6_moptions *mopts;
437 	struct route_in6 *ro;
438 	struct ifnet **retifp;
439 	struct rtentry **retrt;
440 	int clone;		/* meaningful only for bsdi and freebsd. */
441 	int norouteok;
442 {
443 	int error = 0;
444 	struct ifnet *ifp = NULL;
445 	struct rtentry *rt = NULL;
446 	struct sockaddr_in6 *sin6_next;
447 	struct in6_pktinfo *pi = NULL;
448 	struct in6_addr *dst = &dstsock->sin6_addr;
449 #if 0
450 	char ip6buf[INET6_ADDRSTRLEN];
451 
452 	if (dstsock->sin6_addr.s6_addr32[0] == 0 &&
453 	    dstsock->sin6_addr.s6_addr32[1] == 0 &&
454 	    !IN6_IS_ADDR_LOOPBACK(&dstsock->sin6_addr)) {
455 		printf("in6_selectroute: strange destination %s\n",
456 		       ip6_sprintf(ip6buf, &dstsock->sin6_addr));
457 	} else {
458 		printf("in6_selectroute: destination = %s%%%d\n",
459 		       ip6_sprintf(ip6buf, &dstsock->sin6_addr),
460 		       dstsock->sin6_scope_id); /* for debug */
461 	}
462 #endif
463 
464 	/* If the caller specify the outgoing interface explicitly, use it. */
465 	if (opts && (pi = opts->ip6po_pktinfo) != NULL && pi->ipi6_ifindex) {
466 		/* XXX boundary check is assumed to be already done. */
467 		ifp = ifnet_byindex(pi->ipi6_ifindex);
468 		if (ifp != NULL &&
469 		    (norouteok || retrt == NULL ||
470 		    IN6_IS_ADDR_MULTICAST(dst))) {
471 			/*
472 			 * we do not have to check or get the route for
473 			 * multicast.
474 			 */
475 			goto done;
476 		} else
477 			goto getroute;
478 	}
479 
480 	/*
481 	 * If the destination address is a multicast address and the outgoing
482 	 * interface for the address is specified by the caller, use it.
483 	 */
484 	if (IN6_IS_ADDR_MULTICAST(dst) &&
485 	    mopts != NULL && (ifp = mopts->im6o_multicast_ifp) != NULL) {
486 		goto done; /* we do not need a route for multicast. */
487 	}
488 
489   getroute:
490 	/*
491 	 * If the next hop address for the packet is specified by the caller,
492 	 * use it as the gateway.
493 	 */
494 	if (opts && opts->ip6po_nexthop) {
495 		struct route_in6 *ron;
496 
497 		sin6_next = satosin6(opts->ip6po_nexthop);
498 
499 		/* at this moment, we only support AF_INET6 next hops */
500 		if (sin6_next->sin6_family != AF_INET6) {
501 			error = EAFNOSUPPORT; /* or should we proceed? */
502 			goto done;
503 		}
504 
505 		/*
506 		 * If the next hop is an IPv6 address, then the node identified
507 		 * by that address must be a neighbor of the sending host.
508 		 */
509 		ron = &opts->ip6po_nextroute;
510 		if ((ron->ro_rt &&
511 		     (ron->ro_rt->rt_flags & (RTF_UP | RTF_LLINFO)) !=
512 		     (RTF_UP | RTF_LLINFO)) ||
513 		    !IN6_ARE_ADDR_EQUAL(&satosin6(&ron->ro_dst)->sin6_addr,
514 		    &sin6_next->sin6_addr)) {
515 			if (ron->ro_rt) {
516 				RTFREE(ron->ro_rt);
517 				ron->ro_rt = NULL;
518 			}
519 			*satosin6(&ron->ro_dst) = *sin6_next;
520 		}
521 		if (ron->ro_rt == NULL) {
522 			rtalloc((struct route *)ron); /* multi path case? */
523 			if (ron->ro_rt == NULL ||
524 			    !(ron->ro_rt->rt_flags & RTF_LLINFO)) {
525 				if (ron->ro_rt) {
526 					RTFREE(ron->ro_rt);
527 					ron->ro_rt = NULL;
528 				}
529 				error = EHOSTUNREACH;
530 				goto done;
531 			}
532 		}
533 		rt = ron->ro_rt;
534 		ifp = rt->rt_ifp;
535 
536 		/*
537 		 * When cloning is required, try to allocate a route to the
538 		 * destination so that the caller can store path MTU
539 		 * information.
540 		 */
541 		if (!clone)
542 			goto done;
543 	}
544 
545 	/*
546 	 * Use a cached route if it exists and is valid, else try to allocate
547 	 * a new one.  Note that we should check the address family of the
548 	 * cached destination, in case of sharing the cache with IPv4.
549 	 */
550 	if (ro) {
551 		if (ro->ro_rt &&
552 		    (!(ro->ro_rt->rt_flags & RTF_UP) ||
553 		     ((struct sockaddr *)(&ro->ro_dst))->sa_family != AF_INET6 ||
554 		     !IN6_ARE_ADDR_EQUAL(&satosin6(&ro->ro_dst)->sin6_addr,
555 		     dst))) {
556 			RTFREE(ro->ro_rt);
557 			ro->ro_rt = (struct rtentry *)NULL;
558 		}
559 		if (ro->ro_rt == (struct rtentry *)NULL) {
560 			struct sockaddr_in6 *sa6;
561 
562 			/* No route yet, so try to acquire one */
563 			bzero(&ro->ro_dst, sizeof(struct sockaddr_in6));
564 			sa6 = (struct sockaddr_in6 *)&ro->ro_dst;
565 			*sa6 = *dstsock;
566 			sa6->sin6_scope_id = 0;
567 
568 			if (clone) {
569 				rtalloc((struct route *)ro);
570 			} else {
571 				ro->ro_rt = rtalloc1(&((struct route *)ro)
572 						     ->ro_dst, 0, 0UL);
573 				if (ro->ro_rt)
574 					RT_UNLOCK(ro->ro_rt);
575 			}
576 		}
577 
578 		/*
579 		 * do not care about the result if we have the nexthop
580 		 * explicitly specified.
581 		 */
582 		if (opts && opts->ip6po_nexthop)
583 			goto done;
584 
585 		if (ro->ro_rt) {
586 			ifp = ro->ro_rt->rt_ifp;
587 
588 			if (ifp == NULL) { /* can this really happen? */
589 				RTFREE(ro->ro_rt);
590 				ro->ro_rt = NULL;
591 			}
592 		}
593 		if (ro->ro_rt == NULL)
594 			error = EHOSTUNREACH;
595 		rt = ro->ro_rt;
596 
597 		/*
598 		 * Check if the outgoing interface conflicts with
599 		 * the interface specified by ipi6_ifindex (if specified).
600 		 * Note that loopback interface is always okay.
601 		 * (this may happen when we are sending a packet to one of
602 		 *  our own addresses.)
603 		 */
604 		if (ifp && opts && opts->ip6po_pktinfo &&
605 		    opts->ip6po_pktinfo->ipi6_ifindex) {
606 			if (!(ifp->if_flags & IFF_LOOPBACK) &&
607 			    ifp->if_index !=
608 			    opts->ip6po_pktinfo->ipi6_ifindex) {
609 				error = EHOSTUNREACH;
610 				goto done;
611 			}
612 		}
613 	}
614 
615   done:
616 	if (ifp == NULL && rt == NULL) {
617 		/*
618 		 * This can happen if the caller did not pass a cached route
619 		 * nor any other hints.  We treat this case an error.
620 		 */
621 		error = EHOSTUNREACH;
622 	}
623 	if (error == EHOSTUNREACH)
624 		ip6stat.ip6s_noroute++;
625 
626 	if (retifp != NULL)
627 		*retifp = ifp;
628 	if (retrt != NULL)
629 		*retrt = rt;	/* rt may be NULL */
630 
631 	return (error);
632 }
633 
634 static int
635 in6_selectif(dstsock, opts, mopts, ro, retifp)
636 	struct sockaddr_in6 *dstsock;
637 	struct ip6_pktopts *opts;
638 	struct ip6_moptions *mopts;
639 	struct route_in6 *ro;
640 	struct ifnet **retifp;
641 {
642 	int error;
643 	struct route_in6 sro;
644 	struct rtentry *rt = NULL;
645 
646 	if (ro == NULL) {
647 		bzero(&sro, sizeof(sro));
648 		ro = &sro;
649 	}
650 
651 	if ((error = selectroute(dstsock, opts, mopts, ro, retifp,
652 				     &rt, 0, 1)) != 0) {
653 		if (ro == &sro && rt && rt == sro.ro_rt)
654 			RTFREE(rt);
655 		return (error);
656 	}
657 
658 	/*
659 	 * do not use a rejected or black hole route.
660 	 * XXX: this check should be done in the L2 output routine.
661 	 * However, if we skipped this check here, we'd see the following
662 	 * scenario:
663 	 * - install a rejected route for a scoped address prefix
664 	 *   (like fe80::/10)
665 	 * - send a packet to a destination that matches the scoped prefix,
666 	 *   with ambiguity about the scope zone.
667 	 * - pick the outgoing interface from the route, and disambiguate the
668 	 *   scope zone with the interface.
669 	 * - ip6_output() would try to get another route with the "new"
670 	 *   destination, which may be valid.
671 	 * - we'd see no error on output.
672 	 * Although this may not be very harmful, it should still be confusing.
673 	 * We thus reject the case here.
674 	 */
675 	if (rt && (rt->rt_flags & (RTF_REJECT | RTF_BLACKHOLE))) {
676 		int flags = (rt->rt_flags & RTF_HOST ? EHOSTUNREACH : ENETUNREACH);
677 
678 		if (ro == &sro && rt && rt == sro.ro_rt)
679 			RTFREE(rt);
680 		return (flags);
681 	}
682 
683 	/*
684 	 * Adjust the "outgoing" interface.  If we're going to loop the packet
685 	 * back to ourselves, the ifp would be the loopback interface.
686 	 * However, we'd rather know the interface associated to the
687 	 * destination address (which should probably be one of our own
688 	 * addresses.)
689 	 */
690 	if (rt && rt->rt_ifa && rt->rt_ifa->ifa_ifp)
691 		*retifp = rt->rt_ifa->ifa_ifp;
692 
693 	if (ro == &sro && rt && rt == sro.ro_rt)
694 		RTFREE(rt);
695 	return (0);
696 }
697 
698 int
699 in6_selectroute(dstsock, opts, mopts, ro, retifp, retrt, clone)
700 	struct sockaddr_in6 *dstsock;
701 	struct ip6_pktopts *opts;
702 	struct ip6_moptions *mopts;
703 	struct route_in6 *ro;
704 	struct ifnet **retifp;
705 	struct rtentry **retrt;
706 	int clone;		/* meaningful only for bsdi and freebsd. */
707 {
708 	return (selectroute(dstsock, opts, mopts, ro, retifp,
709 	    retrt, clone, 0));
710 }
711 
712 /*
713  * Default hop limit selection. The precedence is as follows:
714  * 1. Hoplimit value specified via ioctl.
715  * 2. (If the outgoing interface is detected) the current
716  *     hop limit of the interface specified by router advertisement.
717  * 3. The system default hoplimit.
718  */
719 int
720 in6_selecthlim(in6p, ifp)
721 	struct in6pcb *in6p;
722 	struct ifnet *ifp;
723 {
724 	if (in6p && in6p->in6p_hops >= 0)
725 		return (in6p->in6p_hops);
726 	else if (ifp)
727 		return (ND_IFINFO(ifp)->chlim);
728 	else if (in6p && !IN6_IS_ADDR_UNSPECIFIED(&in6p->in6p_faddr)) {
729 		struct route_in6 ro6;
730 		struct ifnet *lifp;
731 
732 		bzero(&ro6, sizeof(ro6));
733 		ro6.ro_dst.sin6_family = AF_INET6;
734 		ro6.ro_dst.sin6_len = sizeof(struct sockaddr_in6);
735 		ro6.ro_dst.sin6_addr = in6p->in6p_faddr;
736 		rtalloc((struct route *)&ro6);
737 		if (ro6.ro_rt) {
738 			lifp = ro6.ro_rt->rt_ifp;
739 			RTFREE(ro6.ro_rt);
740 			if (lifp)
741 				return (ND_IFINFO(lifp)->chlim);
742 		} else
743 			return (ip6_defhlim);
744 	}
745 	return (ip6_defhlim);
746 }
747 
748 /*
749  * XXX: this is borrowed from in6_pcbbind(). If possible, we should
750  * share this function by all *bsd*...
751  */
752 int
753 in6_pcbsetport(laddr, inp, cred)
754 	struct in6_addr *laddr;
755 	struct inpcb *inp;
756 	struct ucred *cred;
757 {
758 	struct socket *so = inp->inp_socket;
759 	u_int16_t lport = 0, first, last, *lastport;
760 	int count, error = 0, wild = 0;
761 	struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
762 
763 	INP_INFO_WLOCK_ASSERT(pcbinfo);
764 	INP_LOCK_ASSERT(inp);
765 
766 	/* XXX: this is redundant when called from in6_pcbbind */
767 	if ((so->so_options & (SO_REUSEADDR|SO_REUSEPORT)) == 0)
768 		wild = INPLOOKUP_WILDCARD;
769 
770 	inp->inp_flags |= INP_ANONPORT;
771 
772 	if (inp->inp_flags & INP_HIGHPORT) {
773 		first = ipport_hifirstauto;	/* sysctl */
774 		last  = ipport_hilastauto;
775 		lastport = &pcbinfo->lasthi;
776 	} else if (inp->inp_flags & INP_LOWPORT) {
777 		error = priv_check_cred(cred, PRIV_NETINET_RESERVEDPORT,
778 		    SUSER_ALLOWJAIL);
779 		if (error)
780 			return error;
781 		first = ipport_lowfirstauto;	/* 1023 */
782 		last  = ipport_lowlastauto;	/* 600 */
783 		lastport = &pcbinfo->lastlow;
784 	} else {
785 		first = ipport_firstauto;	/* sysctl */
786 		last  = ipport_lastauto;
787 		lastport = &pcbinfo->lastport;
788 	}
789 	/*
790 	 * Simple check to ensure all ports are not used up causing
791 	 * a deadlock here.
792 	 *
793 	 * We split the two cases (up and down) so that the direction
794 	 * is not being tested on each round of the loop.
795 	 */
796 	if (first > last) {
797 		/*
798 		 * counting down
799 		 */
800 		count = first - last;
801 
802 		do {
803 			if (count-- < 0) {	/* completely used? */
804 				/*
805 				 * Undo any address bind that may have
806 				 * occurred above.
807 				 */
808 				inp->in6p_laddr = in6addr_any;
809 				return (EAGAIN);
810 			}
811 			--*lastport;
812 			if (*lastport > first || *lastport < last)
813 				*lastport = first;
814 			lport = htons(*lastport);
815 		} while (in6_pcblookup_local(pcbinfo, &inp->in6p_laddr,
816 					     lport, wild));
817 	} else {
818 		/*
819 			 * counting up
820 			 */
821 		count = last - first;
822 
823 		do {
824 			if (count-- < 0) {	/* completely used? */
825 				/*
826 				 * Undo any address bind that may have
827 				 * occurred above.
828 				 */
829 				inp->in6p_laddr = in6addr_any;
830 				return (EAGAIN);
831 			}
832 			++*lastport;
833 			if (*lastport < first || *lastport > last)
834 				*lastport = first;
835 			lport = htons(*lastport);
836 		} while (in6_pcblookup_local(pcbinfo,
837 					     &inp->in6p_laddr, lport, wild));
838 	}
839 
840 	inp->inp_lport = lport;
841 	if (in_pcbinshash(inp) != 0) {
842 		inp->in6p_laddr = in6addr_any;
843 		inp->inp_lport = 0;
844 		return (EAGAIN);
845 	}
846 
847 	return (0);
848 }
849 
850 void
851 addrsel_policy_init()
852 {
853 	ADDRSEL_LOCK_INIT();
854 	ADDRSEL_SXLOCK_INIT();
855 
856 	init_policy_queue();
857 
858 	/* initialize the "last resort" policy */
859 	bzero(&defaultaddrpolicy, sizeof(defaultaddrpolicy));
860 	defaultaddrpolicy.label = ADDR_LABEL_NOTAPP;
861 }
862 
863 static struct in6_addrpolicy *
864 lookup_addrsel_policy(key)
865 	struct sockaddr_in6 *key;
866 {
867 	struct in6_addrpolicy *match = NULL;
868 
869 	ADDRSEL_LOCK();
870 	match = match_addrsel_policy(key);
871 
872 	if (match == NULL)
873 		match = &defaultaddrpolicy;
874 	else
875 		match->use++;
876 	ADDRSEL_UNLOCK();
877 
878 	return (match);
879 }
880 
881 /*
882  * Subroutines to manage the address selection policy table via sysctl.
883  */
884 struct walkarg {
885 	struct sysctl_req *w_req;
886 };
887 
888 static int in6_src_sysctl(SYSCTL_HANDLER_ARGS);
889 SYSCTL_DECL(_net_inet6_ip6);
890 SYSCTL_NODE(_net_inet6_ip6, IPV6CTL_ADDRCTLPOLICY, addrctlpolicy,
891 	CTLFLAG_RD, in6_src_sysctl, "");
892 
893 static int
894 in6_src_sysctl(SYSCTL_HANDLER_ARGS)
895 {
896 	struct walkarg w;
897 
898 	if (req->newptr)
899 		return EPERM;
900 
901 	bzero(&w, sizeof(w));
902 	w.w_req = req;
903 
904 	return (walk_addrsel_policy(dump_addrsel_policyent, &w));
905 }
906 
907 int
908 in6_src_ioctl(cmd, data)
909 	u_long cmd;
910 	caddr_t data;
911 {
912 	int i;
913 	struct in6_addrpolicy ent0;
914 
915 	if (cmd != SIOCAADDRCTL_POLICY && cmd != SIOCDADDRCTL_POLICY)
916 		return (EOPNOTSUPP); /* check for safety */
917 
918 	ent0 = *(struct in6_addrpolicy *)data;
919 
920 	if (ent0.label == ADDR_LABEL_NOTAPP)
921 		return (EINVAL);
922 	/* check if the prefix mask is consecutive. */
923 	if (in6_mask2len(&ent0.addrmask.sin6_addr, NULL) < 0)
924 		return (EINVAL);
925 	/* clear trailing garbages (if any) of the prefix address. */
926 	for (i = 0; i < 4; i++) {
927 		ent0.addr.sin6_addr.s6_addr32[i] &=
928 			ent0.addrmask.sin6_addr.s6_addr32[i];
929 	}
930 	ent0.use = 0;
931 
932 	switch (cmd) {
933 	case SIOCAADDRCTL_POLICY:
934 		return (add_addrsel_policyent(&ent0));
935 	case SIOCDADDRCTL_POLICY:
936 		return (delete_addrsel_policyent(&ent0));
937 	}
938 
939 	return (0);		/* XXX: compromise compilers */
940 }
941 
942 /*
943  * The followings are implementation of the policy table using a
944  * simple tail queue.
945  * XXX such details should be hidden.
946  * XXX implementation using binary tree should be more efficient.
947  */
948 struct addrsel_policyent {
949 	TAILQ_ENTRY(addrsel_policyent) ape_entry;
950 	struct in6_addrpolicy ape_policy;
951 };
952 
953 TAILQ_HEAD(addrsel_policyhead, addrsel_policyent);
954 
955 struct addrsel_policyhead addrsel_policytab;
956 
957 static void
958 init_policy_queue()
959 {
960 	TAILQ_INIT(&addrsel_policytab);
961 }
962 
963 static int
964 add_addrsel_policyent(newpolicy)
965 	struct in6_addrpolicy *newpolicy;
966 {
967 	struct addrsel_policyent *new, *pol;
968 
969 	MALLOC(new, struct addrsel_policyent *, sizeof(*new), M_IFADDR,
970 	       M_WAITOK);
971 	ADDRSEL_XLOCK();
972 	ADDRSEL_LOCK();
973 
974 	/* duplication check */
975 	TAILQ_FOREACH(pol, &addrsel_policytab, ape_entry) {
976 		if (IN6_ARE_ADDR_EQUAL(&newpolicy->addr.sin6_addr,
977 				       &pol->ape_policy.addr.sin6_addr) &&
978 		    IN6_ARE_ADDR_EQUAL(&newpolicy->addrmask.sin6_addr,
979 				       &pol->ape_policy.addrmask.sin6_addr)) {
980 			ADDRSEL_UNLOCK();
981 			ADDRSEL_XUNLOCK();
982 			FREE(new, M_IFADDR);
983 			return (EEXIST);	/* or override it? */
984 		}
985 	}
986 
987 	bzero(new, sizeof(*new));
988 
989 	/* XXX: should validate entry */
990 	new->ape_policy = *newpolicy;
991 
992 	TAILQ_INSERT_TAIL(&addrsel_policytab, new, ape_entry);
993 	ADDRSEL_UNLOCK();
994 	ADDRSEL_XUNLOCK();
995 
996 	return (0);
997 }
998 
999 static int
1000 delete_addrsel_policyent(key)
1001 	struct in6_addrpolicy *key;
1002 {
1003 	struct addrsel_policyent *pol;
1004 
1005 	ADDRSEL_XLOCK();
1006 	ADDRSEL_LOCK();
1007 
1008 	/* search for the entry in the table */
1009 	TAILQ_FOREACH(pol, &addrsel_policytab, ape_entry) {
1010 		if (IN6_ARE_ADDR_EQUAL(&key->addr.sin6_addr,
1011 		    &pol->ape_policy.addr.sin6_addr) &&
1012 		    IN6_ARE_ADDR_EQUAL(&key->addrmask.sin6_addr,
1013 		    &pol->ape_policy.addrmask.sin6_addr)) {
1014 			break;
1015 		}
1016 	}
1017 	if (pol == NULL) {
1018 		ADDRSEL_UNLOCK();
1019 		ADDRSEL_XUNLOCK();
1020 		return (ESRCH);
1021 	}
1022 
1023 	TAILQ_REMOVE(&addrsel_policytab, pol, ape_entry);
1024 	ADDRSEL_UNLOCK();
1025 	ADDRSEL_XUNLOCK();
1026 
1027 	return (0);
1028 }
1029 
1030 static int
1031 walk_addrsel_policy(callback, w)
1032 	int (*callback) __P((struct in6_addrpolicy *, void *));
1033 	void *w;
1034 {
1035 	struct addrsel_policyent *pol;
1036 	int error = 0;
1037 
1038 	ADDRSEL_SLOCK();
1039 	TAILQ_FOREACH(pol, &addrsel_policytab, ape_entry) {
1040 		if ((error = (*callback)(&pol->ape_policy, w)) != 0) {
1041 			ADDRSEL_SUNLOCK();
1042 			return (error);
1043 		}
1044 	}
1045 	ADDRSEL_SUNLOCK();
1046 	return (error);
1047 }
1048 
1049 static int
1050 dump_addrsel_policyent(pol, arg)
1051 	struct in6_addrpolicy *pol;
1052 	void *arg;
1053 {
1054 	int error = 0;
1055 	struct walkarg *w = arg;
1056 
1057 	error = SYSCTL_OUT(w->w_req, pol, sizeof(*pol));
1058 
1059 	return (error);
1060 }
1061 
1062 static struct in6_addrpolicy *
1063 match_addrsel_policy(key)
1064 	struct sockaddr_in6 *key;
1065 {
1066 	struct addrsel_policyent *pent;
1067 	struct in6_addrpolicy *bestpol = NULL, *pol;
1068 	int matchlen, bestmatchlen = -1;
1069 	u_char *mp, *ep, *k, *p, m;
1070 
1071 	TAILQ_FOREACH(pent, &addrsel_policytab, ape_entry) {
1072 		matchlen = 0;
1073 
1074 		pol = &pent->ape_policy;
1075 		mp = (u_char *)&pol->addrmask.sin6_addr;
1076 		ep = mp + 16;	/* XXX: scope field? */
1077 		k = (u_char *)&key->sin6_addr;
1078 		p = (u_char *)&pol->addr.sin6_addr;
1079 		for (; mp < ep && *mp; mp++, k++, p++) {
1080 			m = *mp;
1081 			if ((*k & m) != *p)
1082 				goto next; /* not match */
1083 			if (m == 0xff) /* short cut for a typical case */
1084 				matchlen += 8;
1085 			else {
1086 				while (m >= 0x80) {
1087 					matchlen++;
1088 					m <<= 1;
1089 				}
1090 			}
1091 		}
1092 
1093 		/* matched.  check if this is better than the current best. */
1094 		if (bestpol == NULL ||
1095 		    matchlen > bestmatchlen) {
1096 			bestpol = pol;
1097 			bestmatchlen = matchlen;
1098 		}
1099 
1100 	  next:
1101 		continue;
1102 	}
1103 
1104 	return (bestpol);
1105 }
1106