xref: /freebsd/sys/netinet6/in6_src.c (revision cfd6422a5217410fbd66f7a7a8a64d9d85e61229)
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: in6_src.c,v 1.132 2003/08/26 04:42:27 keiichi Exp $
32  */
33 
34 /*-
35  * Copyright (c) 1982, 1986, 1991, 1993
36  *	The Regents of the University of California.  All rights reserved.
37  *
38  * Redistribution and use in source and binary forms, with or without
39  * modification, are permitted provided that the following conditions
40  * are met:
41  * 1. Redistributions of source code must retain the above copyright
42  *    notice, this list of conditions and the following disclaimer.
43  * 2. Redistributions in binary form must reproduce the above copyright
44  *    notice, this list of conditions and the following disclaimer in the
45  *    documentation and/or other materials provided with the distribution.
46  * 3. Neither the name of the University nor the names of its contributors
47  *    may be used to endorse or promote products derived from this software
48  *    without specific prior written permission.
49  *
50  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
51  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
52  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
53  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
54  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
55  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
56  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
57  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
58  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
59  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
60  * SUCH DAMAGE.
61  *
62  *	@(#)in_pcb.c	8.2 (Berkeley) 1/4/94
63  */
64 
65 #include <sys/cdefs.h>
66 __FBSDID("$FreeBSD$");
67 
68 #include "opt_inet.h"
69 #include "opt_inet6.h"
70 #include <sys/param.h>
71 #include <sys/systm.h>
72 #include <sys/lock.h>
73 #include <sys/malloc.h>
74 #include <sys/mbuf.h>
75 #include <sys/priv.h>
76 #include <sys/protosw.h>
77 #include <sys/socket.h>
78 #include <sys/socketvar.h>
79 #include <sys/sockio.h>
80 #include <sys/sysctl.h>
81 #include <sys/errno.h>
82 #include <sys/time.h>
83 #include <sys/jail.h>
84 #include <sys/kernel.h>
85 #include <sys/rmlock.h>
86 #include <sys/sx.h>
87 
88 #include <net/if.h>
89 #include <net/if_var.h>
90 #include <net/if_dl.h>
91 #include <net/route.h>
92 #include <net/route/nhop.h>
93 #include <net/if_llatbl.h>
94 
95 #include <netinet/in.h>
96 #include <netinet/in_var.h>
97 #include <netinet/in_systm.h>
98 #include <netinet/ip.h>
99 #include <netinet/in_pcb.h>
100 #include <netinet/ip_var.h>
101 #include <netinet/udp.h>
102 #include <netinet/udp_var.h>
103 
104 #include <netinet6/in6_var.h>
105 #include <netinet/ip6.h>
106 #include <netinet6/in6_fib.h>
107 #include <netinet6/in6_pcb.h>
108 #include <netinet6/ip6_var.h>
109 #include <netinet6/scope6_var.h>
110 #include <netinet6/nd6.h>
111 
112 static struct mtx addrsel_lock;
113 #define	ADDRSEL_LOCK_INIT()	mtx_init(&addrsel_lock, "addrsel_lock", NULL, MTX_DEF)
114 #define	ADDRSEL_LOCK()		mtx_lock(&addrsel_lock)
115 #define	ADDRSEL_UNLOCK()	mtx_unlock(&addrsel_lock)
116 #define	ADDRSEL_LOCK_ASSERT()	mtx_assert(&addrsel_lock, MA_OWNED)
117 
118 static struct sx addrsel_sxlock;
119 #define	ADDRSEL_SXLOCK_INIT()	sx_init(&addrsel_sxlock, "addrsel_sxlock")
120 #define	ADDRSEL_SLOCK()		sx_slock(&addrsel_sxlock)
121 #define	ADDRSEL_SUNLOCK()	sx_sunlock(&addrsel_sxlock)
122 #define	ADDRSEL_XLOCK()		sx_xlock(&addrsel_sxlock)
123 #define	ADDRSEL_XUNLOCK()	sx_xunlock(&addrsel_sxlock)
124 
125 #define ADDR_LABEL_NOTAPP (-1)
126 VNET_DEFINE_STATIC(struct in6_addrpolicy, defaultaddrpolicy);
127 #define	V_defaultaddrpolicy		VNET(defaultaddrpolicy)
128 
129 VNET_DEFINE(int, ip6_prefer_tempaddr) = 0;
130 
131 static int selectroute(struct sockaddr_in6 *, struct ip6_pktopts *,
132 	struct ip6_moptions *, struct route_in6 *, struct ifnet **,
133 	struct nhop_object **, int, u_int, uint32_t);
134 static int in6_selectif(struct sockaddr_in6 *, struct ip6_pktopts *,
135 	struct ip6_moptions *, struct ifnet **,
136 	struct ifnet *, u_int);
137 static int in6_selectsrc(uint32_t, struct sockaddr_in6 *,
138 	struct ip6_pktopts *, struct inpcb *, struct ucred *,
139 	struct ifnet **, struct in6_addr *);
140 
141 static struct in6_addrpolicy *lookup_addrsel_policy(struct sockaddr_in6 *);
142 
143 static void init_policy_queue(void);
144 static int add_addrsel_policyent(struct in6_addrpolicy *);
145 static int delete_addrsel_policyent(struct in6_addrpolicy *);
146 static int walk_addrsel_policy(int (*)(struct in6_addrpolicy *, void *),
147 	void *);
148 static int dump_addrsel_policyent(struct in6_addrpolicy *, void *);
149 static struct in6_addrpolicy *match_addrsel_policy(struct sockaddr_in6 *);
150 
151 /*
152  * Return an IPv6 address, which is the most appropriate for a given
153  * destination and user specified options.
154  * If necessary, this function lookups the routing table and returns
155  * an entry to the caller for later use.
156  */
157 #define REPLACE(r) do {\
158 	IP6STAT_INC(ip6s_sources_rule[(r)]); \
159 	/* { \
160 	char ip6buf[INET6_ADDRSTRLEN], ip6b[INET6_ADDRSTRLEN]; \
161 	printf("in6_selectsrc: replace %s with %s by %d\n", ia_best ? ip6_sprintf(ip6buf, &ia_best->ia_addr.sin6_addr) : "none", ip6_sprintf(ip6b, &ia->ia_addr.sin6_addr), (r)); \
162 	} */ \
163 	goto replace; \
164 } while(0)
165 #define NEXT(r) do {\
166 	/* { \
167 	char ip6buf[INET6_ADDRSTRLEN], ip6b[INET6_ADDRSTRLEN]; \
168 	printf("in6_selectsrc: keep %s against %s by %d\n", ia_best ? ip6_sprintf(ip6buf, &ia_best->ia_addr.sin6_addr) : "none", ip6_sprintf(ip6b, &ia->ia_addr.sin6_addr), (r)); \
169 	} */ \
170 	goto next;		/* XXX: we can't use 'continue' here */ \
171 } while(0)
172 #define BREAK(r) do { \
173 	IP6STAT_INC(ip6s_sources_rule[(r)]); \
174 	goto out;		/* XXX: we can't use 'break' here */ \
175 } while(0)
176 
177 static int
178 in6_selectsrc(uint32_t fibnum, struct sockaddr_in6 *dstsock,
179     struct ip6_pktopts *opts, struct inpcb *inp, struct ucred *cred,
180     struct ifnet **ifpp, struct in6_addr *srcp)
181 {
182 	struct rm_priotracker in6_ifa_tracker;
183 	struct in6_addr dst, tmp;
184 	struct ifnet *ifp = NULL, *oifp = NULL;
185 	struct in6_ifaddr *ia = NULL, *ia_best = NULL;
186 	struct in6_pktinfo *pi = NULL;
187 	int dst_scope = -1, best_scope = -1, best_matchlen = -1;
188 	struct in6_addrpolicy *dst_policy = NULL, *best_policy = NULL;
189 	u_int32_t odstzone;
190 	int prefer_tempaddr;
191 	int error;
192 	struct ip6_moptions *mopts;
193 
194 	KASSERT(srcp != NULL, ("%s: srcp is NULL", __func__));
195 
196 	dst = dstsock->sin6_addr; /* make a copy for local operation */
197 	if (ifpp) {
198 		/*
199 		 * Save a possibly passed in ifp for in6_selectsrc. Only
200 		 * neighbor discovery code should use this feature, where
201 		 * we may know the interface but not the FIB number holding
202 		 * the connected subnet in case someone deleted it from the
203 		 * default FIB and we need to check the interface.
204 		 */
205 		if (*ifpp != NULL)
206 			oifp = *ifpp;
207 		*ifpp = NULL;
208 	}
209 
210 	if (inp != NULL) {
211 		INP_LOCK_ASSERT(inp);
212 		mopts = inp->in6p_moptions;
213 	} else {
214 		mopts = NULL;
215 	}
216 
217 	/*
218 	 * If the source address is explicitly specified by the caller,
219 	 * check if the requested source address is indeed a unicast address
220 	 * assigned to the node, and can be used as the packet's source
221 	 * address.  If everything is okay, use the address as source.
222 	 */
223 	if (opts && (pi = opts->ip6po_pktinfo) &&
224 	    !IN6_IS_ADDR_UNSPECIFIED(&pi->ipi6_addr)) {
225 		/* get the outgoing interface */
226 		if ((error = in6_selectif(dstsock, opts, mopts, &ifp, oifp,
227 		    fibnum))
228 		    != 0)
229 			return (error);
230 
231 		/*
232 		 * determine the appropriate zone id of the source based on
233 		 * the zone of the destination and the outgoing interface.
234 		 * If the specified address is ambiguous wrt the scope zone,
235 		 * the interface must be specified; otherwise, ifa_ifwithaddr()
236 		 * will fail matching the address.
237 		 */
238 		tmp = pi->ipi6_addr;
239 		if (ifp) {
240 			error = in6_setscope(&tmp, ifp, &odstzone);
241 			if (error)
242 				return (error);
243 		}
244 		if (cred != NULL && (error = prison_local_ip6(cred,
245 		    &tmp, (inp->inp_flags & IN6P_IPV6_V6ONLY) != 0)) != 0)
246 			return (error);
247 
248 		/*
249 		 * If IPV6_BINDANY socket option is set, we allow to specify
250 		 * non local addresses as source address in IPV6_PKTINFO
251 		 * ancillary data.
252 		 */
253 		if ((inp->inp_flags & INP_BINDANY) == 0) {
254 			ia = in6ifa_ifwithaddr(&tmp, 0 /* XXX */);
255 			if (ia == NULL || (ia->ia6_flags & (IN6_IFF_ANYCAST |
256 			    IN6_IFF_NOTREADY))) {
257 				if (ia != NULL)
258 					ifa_free(&ia->ia_ifa);
259 				return (EADDRNOTAVAIL);
260 			}
261 			bcopy(&ia->ia_addr.sin6_addr, srcp, sizeof(*srcp));
262 			ifa_free(&ia->ia_ifa);
263 		} else
264 			bcopy(&tmp, srcp, sizeof(*srcp));
265 		pi->ipi6_addr = tmp; /* XXX: this overrides pi */
266 		if (ifpp)
267 			*ifpp = ifp;
268 		return (0);
269 	}
270 
271 	/*
272 	 * Otherwise, if the socket has already bound the source, just use it.
273 	 */
274 	if (inp != NULL && !IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr)) {
275 		if (cred != NULL &&
276 		    (error = prison_local_ip6(cred, &inp->in6p_laddr,
277 		    ((inp->inp_flags & IN6P_IPV6_V6ONLY) != 0))) != 0)
278 			return (error);
279 		bcopy(&inp->in6p_laddr, srcp, sizeof(*srcp));
280 		return (0);
281 	}
282 
283 	/*
284 	 * Bypass source address selection and use the primary jail IP
285 	 * if requested.
286 	 */
287 	if (cred != NULL && !prison_saddrsel_ip6(cred, srcp))
288 		return (0);
289 
290 	/*
291 	 * If the address is not specified, choose the best one based on
292 	 * the outgoing interface and the destination address.
293 	 */
294 	/* get the outgoing interface */
295 	if ((error = in6_selectif(dstsock, opts, mopts, &ifp, oifp,
296 	    (inp != NULL) ? inp->inp_inc.inc_fibnum : fibnum)) != 0)
297 		return (error);
298 
299 #ifdef DIAGNOSTIC
300 	if (ifp == NULL)	/* this should not happen */
301 		panic("in6_selectsrc: NULL ifp");
302 #endif
303 	error = in6_setscope(&dst, ifp, &odstzone);
304 	if (error)
305 		return (error);
306 
307 	IN6_IFADDR_RLOCK(&in6_ifa_tracker);
308 	CK_STAILQ_FOREACH(ia, &V_in6_ifaddrhead, ia_link) {
309 		int new_scope = -1, new_matchlen = -1;
310 		struct in6_addrpolicy *new_policy = NULL;
311 		u_int32_t srczone, osrczone, dstzone;
312 		struct in6_addr src;
313 		struct ifnet *ifp1 = ia->ia_ifp;
314 
315 		/*
316 		 * We'll never take an address that breaks the scope zone
317 		 * of the destination.  We also skip an address if its zone
318 		 * does not contain the outgoing interface.
319 		 * XXX: we should probably use sin6_scope_id here.
320 		 */
321 		if (in6_setscope(&dst, ifp1, &dstzone) ||
322 		    odstzone != dstzone) {
323 			continue;
324 		}
325 		src = ia->ia_addr.sin6_addr;
326 		if (in6_setscope(&src, ifp, &osrczone) ||
327 		    in6_setscope(&src, ifp1, &srczone) ||
328 		    osrczone != srczone) {
329 			continue;
330 		}
331 
332 		/* avoid unusable addresses */
333 		if ((ia->ia6_flags &
334 		     (IN6_IFF_NOTREADY | IN6_IFF_ANYCAST | IN6_IFF_DETACHED))) {
335 				continue;
336 		}
337 		if (!V_ip6_use_deprecated && IFA6_IS_DEPRECATED(ia))
338 			continue;
339 
340 		/* If jailed only take addresses of the jail into account. */
341 		if (cred != NULL &&
342 		    prison_check_ip6(cred, &ia->ia_addr.sin6_addr) != 0)
343 			continue;
344 
345 		/* Rule 1: Prefer same address */
346 		if (IN6_ARE_ADDR_EQUAL(&dst, &ia->ia_addr.sin6_addr)) {
347 			ia_best = ia;
348 			BREAK(1); /* there should be no better candidate */
349 		}
350 
351 		if (ia_best == NULL)
352 			REPLACE(0);
353 
354 		/* Rule 2: Prefer appropriate scope */
355 		if (dst_scope < 0)
356 			dst_scope = in6_addrscope(&dst);
357 		new_scope = in6_addrscope(&ia->ia_addr.sin6_addr);
358 		if (IN6_ARE_SCOPE_CMP(best_scope, new_scope) < 0) {
359 			if (IN6_ARE_SCOPE_CMP(best_scope, dst_scope) < 0)
360 				REPLACE(2);
361 			NEXT(2);
362 		} else if (IN6_ARE_SCOPE_CMP(new_scope, best_scope) < 0) {
363 			if (IN6_ARE_SCOPE_CMP(new_scope, dst_scope) < 0)
364 				NEXT(2);
365 			REPLACE(2);
366 		}
367 
368 		/*
369 		 * Rule 3: Avoid deprecated addresses.  Note that the case of
370 		 * !ip6_use_deprecated is already rejected above.
371 		 */
372 		if (!IFA6_IS_DEPRECATED(ia_best) && IFA6_IS_DEPRECATED(ia))
373 			NEXT(3);
374 		if (IFA6_IS_DEPRECATED(ia_best) && !IFA6_IS_DEPRECATED(ia))
375 			REPLACE(3);
376 
377 		/* Rule 4: Prefer home addresses */
378 		/*
379 		 * XXX: This is a TODO.  We should probably merge the MIP6
380 		 * case above.
381 		 */
382 
383 		/* Rule 5: Prefer outgoing interface */
384 		if (!(ND_IFINFO(ifp)->flags & ND6_IFF_NO_PREFER_IFACE)) {
385 			if (ia_best->ia_ifp == ifp && ia->ia_ifp != ifp)
386 				NEXT(5);
387 			if (ia_best->ia_ifp != ifp && ia->ia_ifp == ifp)
388 				REPLACE(5);
389 		}
390 
391 		/*
392 		 * Rule 6: Prefer matching label
393 		 * Note that best_policy should be non-NULL here.
394 		 */
395 		if (dst_policy == NULL)
396 			dst_policy = lookup_addrsel_policy(dstsock);
397 		if (dst_policy->label != ADDR_LABEL_NOTAPP) {
398 			new_policy = lookup_addrsel_policy(&ia->ia_addr);
399 			if (dst_policy->label == best_policy->label &&
400 			    dst_policy->label != new_policy->label)
401 				NEXT(6);
402 			if (dst_policy->label != best_policy->label &&
403 			    dst_policy->label == new_policy->label)
404 				REPLACE(6);
405 		}
406 
407 		/*
408 		 * Rule 7: Prefer public addresses.
409 		 * We allow users to reverse the logic by configuring
410 		 * a sysctl variable, so that privacy conscious users can
411 		 * always prefer temporary addresses.
412 		 */
413 		if (opts == NULL ||
414 		    opts->ip6po_prefer_tempaddr == IP6PO_TEMPADDR_SYSTEM) {
415 			prefer_tempaddr = V_ip6_prefer_tempaddr;
416 		} else if (opts->ip6po_prefer_tempaddr ==
417 		    IP6PO_TEMPADDR_NOTPREFER) {
418 			prefer_tempaddr = 0;
419 		} else
420 			prefer_tempaddr = 1;
421 		if (!(ia_best->ia6_flags & IN6_IFF_TEMPORARY) &&
422 		    (ia->ia6_flags & IN6_IFF_TEMPORARY)) {
423 			if (prefer_tempaddr)
424 				REPLACE(7);
425 			else
426 				NEXT(7);
427 		}
428 		if ((ia_best->ia6_flags & IN6_IFF_TEMPORARY) &&
429 		    !(ia->ia6_flags & IN6_IFF_TEMPORARY)) {
430 			if (prefer_tempaddr)
431 				NEXT(7);
432 			else
433 				REPLACE(7);
434 		}
435 
436 		/*
437 		 * Rule 8: prefer addresses on alive interfaces.
438 		 * This is a KAME specific rule.
439 		 */
440 		if ((ia_best->ia_ifp->if_flags & IFF_UP) &&
441 		    !(ia->ia_ifp->if_flags & IFF_UP))
442 			NEXT(8);
443 		if (!(ia_best->ia_ifp->if_flags & IFF_UP) &&
444 		    (ia->ia_ifp->if_flags & IFF_UP))
445 			REPLACE(8);
446 
447 		/*
448 		 * Rule 9: prefer address with better virtual status.
449 		 */
450 		if (ifa_preferred(&ia_best->ia_ifa, &ia->ia_ifa))
451 			REPLACE(9);
452 		if (ifa_preferred(&ia->ia_ifa, &ia_best->ia_ifa))
453 			NEXT(9);
454 
455 		/*
456 		 * Rule 10: prefer address with `prefer_source' flag.
457 		 */
458 		if ((ia_best->ia6_flags & IN6_IFF_PREFER_SOURCE) == 0 &&
459 		    (ia->ia6_flags & IN6_IFF_PREFER_SOURCE) != 0)
460 			REPLACE(10);
461 		if ((ia_best->ia6_flags & IN6_IFF_PREFER_SOURCE) != 0 &&
462 		    (ia->ia6_flags & IN6_IFF_PREFER_SOURCE) == 0)
463 			NEXT(10);
464 
465 		/*
466 		 * Rule 14: Use longest matching prefix.
467 		 * Note: in the address selection draft, this rule is
468 		 * documented as "Rule 8".  However, since it is also
469 		 * documented that this rule can be overridden, we assign
470 		 * a large number so that it is easy to assign smaller numbers
471 		 * to more preferred rules.
472 		 */
473 		new_matchlen = in6_matchlen(&ia->ia_addr.sin6_addr, &dst);
474 		if (best_matchlen < new_matchlen)
475 			REPLACE(14);
476 		if (new_matchlen < best_matchlen)
477 			NEXT(14);
478 
479 		/* Rule 15 is reserved. */
480 
481 		/*
482 		 * Last resort: just keep the current candidate.
483 		 * Or, do we need more rules?
484 		 */
485 		continue;
486 
487 	  replace:
488 		ia_best = ia;
489 		best_scope = (new_scope >= 0 ? new_scope :
490 			      in6_addrscope(&ia_best->ia_addr.sin6_addr));
491 		best_policy = (new_policy ? new_policy :
492 			       lookup_addrsel_policy(&ia_best->ia_addr));
493 		best_matchlen = (new_matchlen >= 0 ? new_matchlen :
494 				 in6_matchlen(&ia_best->ia_addr.sin6_addr,
495 					      &dst));
496 
497 	  next:
498 		continue;
499 
500 	  out:
501 		break;
502 	}
503 
504 	if ((ia = ia_best) == NULL) {
505 		IN6_IFADDR_RUNLOCK(&in6_ifa_tracker);
506 		IP6STAT_INC(ip6s_sources_none);
507 		return (EADDRNOTAVAIL);
508 	}
509 
510 	/*
511 	 * At this point at least one of the addresses belonged to the jail
512 	 * but it could still be, that we want to further restrict it, e.g.
513 	 * theoratically IN6_IS_ADDR_LOOPBACK.
514 	 * It must not be IN6_IS_ADDR_UNSPECIFIED anymore.
515 	 * prison_local_ip6() will fix an IN6_IS_ADDR_LOOPBACK but should
516 	 * let all others previously selected pass.
517 	 * Use tmp to not change ::1 on lo0 to the primary jail address.
518 	 */
519 	tmp = ia->ia_addr.sin6_addr;
520 	if (cred != NULL && prison_local_ip6(cred, &tmp, (inp != NULL &&
521 	    (inp->inp_flags & IN6P_IPV6_V6ONLY) != 0)) != 0) {
522 		IN6_IFADDR_RUNLOCK(&in6_ifa_tracker);
523 		IP6STAT_INC(ip6s_sources_none);
524 		return (EADDRNOTAVAIL);
525 	}
526 
527 	if (ifpp)
528 		*ifpp = ifp;
529 
530 	bcopy(&tmp, srcp, sizeof(*srcp));
531 	if (ia->ia_ifp == ifp)
532 		IP6STAT_INC(ip6s_sources_sameif[best_scope]);
533 	else
534 		IP6STAT_INC(ip6s_sources_otherif[best_scope]);
535 	if (dst_scope == best_scope)
536 		IP6STAT_INC(ip6s_sources_samescope[best_scope]);
537 	else
538 		IP6STAT_INC(ip6s_sources_otherscope[best_scope]);
539 	if (IFA6_IS_DEPRECATED(ia))
540 		IP6STAT_INC(ip6s_sources_deprecated[best_scope]);
541 	IN6_IFADDR_RUNLOCK(&in6_ifa_tracker);
542 	return (0);
543 }
544 
545 /*
546  * Select source address based on @inp, @dstsock and @opts.
547  * Stores selected address to @srcp. If @scope_ambiguous is set,
548  * embed scope from selected outgoing interface. If @hlim pointer
549  * is provided, stores calculated hop limit there.
550  * Returns 0 on success.
551  */
552 int
553 in6_selectsrc_socket(struct sockaddr_in6 *dstsock, struct ip6_pktopts *opts,
554     struct inpcb *inp, struct ucred *cred, int scope_ambiguous,
555     struct in6_addr *srcp, int *hlim)
556 {
557 	struct ifnet *retifp;
558 	uint32_t fibnum;
559 	int error;
560 
561 	fibnum = inp->inp_inc.inc_fibnum;
562 	retifp = NULL;
563 
564 	error = in6_selectsrc(fibnum, dstsock, opts, inp, cred, &retifp, srcp);
565 	if (error != 0)
566 		return (error);
567 
568 	if (hlim != NULL)
569 		*hlim = in6_selecthlim(inp, retifp);
570 
571 	if (retifp == NULL || scope_ambiguous == 0)
572 		return (0);
573 
574 	/*
575 	 * Application should provide a proper zone ID or the use of
576 	 * default zone IDs should be enabled.  Unfortunately, some
577 	 * applications do not behave as it should, so we need a
578 	 * workaround.  Even if an appropriate ID is not determined
579 	 * (when it's required), if we can determine the outgoing
580 	 * interface. determine the zone ID based on the interface.
581 	 */
582 	error = in6_setscope(&dstsock->sin6_addr, retifp, NULL);
583 
584 	return (error);
585 }
586 
587 /*
588  * Select source address based on @fibnum, @dst and @scopeid.
589  * Stores selected address to @srcp.
590  * Returns 0 on success.
591  *
592  * Used by non-socket based consumers (ND code mostly)
593  */
594 int
595 in6_selectsrc_addr(uint32_t fibnum, const struct in6_addr *dst,
596     uint32_t scopeid, struct ifnet *ifp, struct in6_addr *srcp,
597     int *hlim)
598 {
599 	struct ifnet *retifp;
600 	struct sockaddr_in6 dst_sa;
601 	int error;
602 
603 	retifp = ifp;
604 	bzero(&dst_sa, sizeof(dst_sa));
605 	dst_sa.sin6_family = AF_INET6;
606 	dst_sa.sin6_len = sizeof(dst_sa);
607 	dst_sa.sin6_addr = *dst;
608 	dst_sa.sin6_scope_id = scopeid;
609 	sa6_embedscope(&dst_sa, 0);
610 
611 	error = in6_selectsrc(fibnum, &dst_sa, NULL, NULL, NULL, &retifp, srcp);
612 	if (hlim != NULL)
613 		*hlim = in6_selecthlim(NULL, retifp);
614 
615 	return (error);
616 }
617 
618 /*
619  * clone - meaningful only for bsdi and freebsd
620  */
621 static int
622 selectroute(struct sockaddr_in6 *dstsock, struct ip6_pktopts *opts,
623     struct ip6_moptions *mopts, struct route_in6 *ro,
624     struct ifnet **retifp, struct nhop_object **retnh, int norouteok,
625     u_int fibnum, uint32_t flowid)
626 {
627 	int error = 0;
628 	struct ifnet *ifp = NULL;
629 	struct nhop_object *nh = NULL;
630 	struct sockaddr_in6 *sin6_next;
631 	struct in6_pktinfo *pi = NULL;
632 	struct in6_addr *dst = &dstsock->sin6_addr;
633 	uint32_t zoneid;
634 #if 0
635 	char ip6buf[INET6_ADDRSTRLEN];
636 
637 	if (dstsock->sin6_addr.s6_addr32[0] == 0 &&
638 	    dstsock->sin6_addr.s6_addr32[1] == 0 &&
639 	    !IN6_IS_ADDR_LOOPBACK(&dstsock->sin6_addr)) {
640 		printf("%s: strange destination %s\n", __func__,
641 		       ip6_sprintf(ip6buf, &dstsock->sin6_addr));
642 	} else {
643 		printf("%s: destination = %s%%%d\n", __func__,
644 		       ip6_sprintf(ip6buf, &dstsock->sin6_addr),
645 		       dstsock->sin6_scope_id); /* for debug */
646 	}
647 #endif
648 
649 	/* If the caller specify the outgoing interface explicitly, use it. */
650 	if (opts && (pi = opts->ip6po_pktinfo) != NULL && pi->ipi6_ifindex) {
651 		/* XXX boundary check is assumed to be already done. */
652 		ifp = ifnet_byindex(pi->ipi6_ifindex);
653 		if (ifp != NULL &&
654 		    (norouteok || retnh == NULL ||
655 		    IN6_IS_ADDR_MULTICAST(dst))) {
656 			/*
657 			 * we do not have to check or get the route for
658 			 * multicast.
659 			 */
660 			goto done;
661 		} else
662 			goto getroute;
663 	}
664 	/*
665 	 * If the destination address is a multicast address and the outgoing
666 	 * interface for the address is specified by the caller, use it.
667 	 */
668 	if (IN6_IS_ADDR_MULTICAST(dst) &&
669 	    mopts != NULL && (ifp = mopts->im6o_multicast_ifp) != NULL) {
670 		goto done; /* we do not need a route for multicast. */
671 	}
672 	/*
673 	 * If destination address is LLA or link- or node-local multicast,
674 	 * use it's embedded scope zone id to determine outgoing interface.
675 	 */
676 	if (IN6_IS_ADDR_MC_LINKLOCAL(dst) ||
677 	    IN6_IS_ADDR_MC_NODELOCAL(dst)) {
678 		zoneid = ntohs(in6_getscope(dst));
679 		if (zoneid > 0) {
680 			ifp = in6_getlinkifnet(zoneid);
681 			goto done;
682 		}
683 	}
684 
685   getroute:
686 	/*
687 	 * If the next hop address for the packet is specified by the caller,
688 	 * use it as the gateway.
689 	 */
690 	if (opts && opts->ip6po_nexthop) {
691 		struct route_in6 *ron;
692 
693 		sin6_next = satosin6(opts->ip6po_nexthop);
694 		if (IN6_IS_ADDR_LINKLOCAL(&sin6_next->sin6_addr)) {
695 			/*
696 			 * Next hop is LLA, thus it should be neighbor.
697 			 * Determine outgoing interface by zone index.
698 			 */
699 			zoneid = ntohs(in6_getscope(&sin6_next->sin6_addr));
700 			if (zoneid > 0) {
701 				ifp = in6_getlinkifnet(zoneid);
702 				goto done;
703 			}
704 		}
705 		ron = &opts->ip6po_nextroute;
706 		/* Use a cached route if it exists and is valid. */
707 		if (ron->ro_nh != NULL && (
708 		    !NH_IS_VALID(ron->ro_nh) ||
709 		    ron->ro_dst.sin6_family != AF_INET6 ||
710 		    !IN6_ARE_ADDR_EQUAL(&ron->ro_dst.sin6_addr,
711 			&sin6_next->sin6_addr)))
712 			RO_NHFREE(ron);
713 		if (ron->ro_nh == NULL) {
714 			ron->ro_dst = *sin6_next;
715 			/*
716 			 * sin6_next is not link-local OR scopeid is 0,
717 			 * no need to clear scope
718 			 */
719 			ron->ro_nh = fib6_lookup(fibnum,
720 			    &sin6_next->sin6_addr, 0, NHR_REF, flowid);
721 		}
722 		/*
723 		 * The node identified by that address must be a
724 		 * neighbor of the sending host.
725 		 */
726 		if (ron->ro_nh == NULL ||
727 		    (ron->ro_nh->nh_flags & NHF_GATEWAY) != 0)
728 			error = EHOSTUNREACH;
729 		else {
730 			nh = ron->ro_nh;
731 			ifp = nh->nh_ifp;
732 		}
733 		goto done;
734 	}
735 
736 	/*
737 	 * Use a cached route if it exists and is valid, else try to allocate
738 	 * a new one.  Note that we should check the address family of the
739 	 * cached destination, in case of sharing the cache with IPv4.
740 	 */
741 	if (ro) {
742 		if (ro->ro_nh &&
743 		    (!NH_IS_VALID(ro->ro_nh) ||
744 		     ((struct sockaddr *)(&ro->ro_dst))->sa_family != AF_INET6 ||
745 		     !IN6_ARE_ADDR_EQUAL(&satosin6(&ro->ro_dst)->sin6_addr,
746 		     dst))) {
747 			RO_NHFREE(ro);
748 		}
749 		if (ro->ro_nh == (struct nhop_object *)NULL) {
750 			struct sockaddr_in6 *sa6;
751 
752 			/* No route yet, so try to acquire one */
753 			bzero(&ro->ro_dst, sizeof(struct sockaddr_in6));
754 			sa6 = (struct sockaddr_in6 *)&ro->ro_dst;
755 			*sa6 = *dstsock;
756 			sa6->sin6_scope_id = 0;
757 
758 			/*
759 			 * Currently dst has scopeid embedded iff it is LL.
760 			 * New routing API accepts scopeid as a separate argument.
761 			 * Convert dst before/after doing lookup
762 			 */
763 			uint32_t scopeid = 0;
764 			if (IN6_IS_SCOPE_LINKLOCAL(&sa6->sin6_addr)) {
765 				/* Unwrap in6_getscope() and in6_clearscope() */
766 				scopeid = ntohs(sa6->sin6_addr.s6_addr16[1]);
767 				sa6->sin6_addr.s6_addr16[1] = 0;
768 			}
769 
770 			ro->ro_nh = fib6_lookup(fibnum,
771 			    &sa6->sin6_addr, scopeid, NHR_REF, flowid);
772 
773 			if (IN6_IS_SCOPE_LINKLOCAL(&sa6->sin6_addr))
774 				sa6->sin6_addr.s6_addr16[1] = htons(scopeid);
775 		}
776 
777 		/*
778 		 * do not care about the result if we have the nexthop
779 		 * explicitly specified.
780 		 */
781 		if (opts && opts->ip6po_nexthop)
782 			goto done;
783 
784 		if (ro->ro_nh)
785 			ifp = ro->ro_nh->nh_ifp;
786 		else
787 			error = EHOSTUNREACH;
788 		nh = ro->ro_nh;
789 
790 		/*
791 		 * Check if the outgoing interface conflicts with
792 		 * the interface specified by ipi6_ifindex (if specified).
793 		 * Note that loopback interface is always okay.
794 		 * (this may happen when we are sending a packet to one of
795 		 *  our own addresses.)
796 		 */
797 		if (ifp && opts && opts->ip6po_pktinfo &&
798 		    opts->ip6po_pktinfo->ipi6_ifindex) {
799 			if (!(ifp->if_flags & IFF_LOOPBACK) &&
800 			    ifp->if_index !=
801 			    opts->ip6po_pktinfo->ipi6_ifindex) {
802 				error = EHOSTUNREACH;
803 				goto done;
804 			}
805 		}
806 	}
807 
808   done:
809 	if (ifp == NULL && nh == NULL) {
810 		/*
811 		 * This can happen if the caller did not pass a cached route
812 		 * nor any other hints.  We treat this case an error.
813 		 */
814 		error = EHOSTUNREACH;
815 	}
816 	if (error == EHOSTUNREACH)
817 		IP6STAT_INC(ip6s_noroute);
818 
819 	if (retifp != NULL) {
820 		if (nh != NULL)
821 			*retifp = nh->nh_aifp;
822 		else
823 			*retifp = ifp;
824 	}
825 
826 	if (retnh != NULL)
827 		*retnh = nh;	/* nh may be NULL */
828 
829 	return (error);
830 }
831 
832 static int
833 in6_selectif(struct sockaddr_in6 *dstsock, struct ip6_pktopts *opts,
834     struct ip6_moptions *mopts, struct ifnet **retifp,
835     struct ifnet *oifp, u_int fibnum)
836 {
837 	int error;
838 	struct route_in6 sro;
839 	struct nhop_object *nh = NULL;
840 	uint16_t nh_flags;
841 
842 	KASSERT(retifp != NULL, ("%s: retifp is NULL", __func__));
843 
844 	bzero(&sro, sizeof(sro));
845 	nh_flags = 0;
846 
847 	error = selectroute(dstsock, opts, mopts, &sro, retifp, &nh, 1, fibnum, 0);
848 
849 	if (nh != NULL)
850 		nh_flags = nh->nh_flags;
851 	if (nh != NULL && nh == sro.ro_nh)
852 		NH_FREE(nh);
853 
854 	if (error != 0) {
855 		/* Help ND. See oifp comment in in6_selectsrc(). */
856 		if (oifp != NULL && fibnum == RT_DEFAULT_FIB) {
857 			*retifp = oifp;
858 			error = 0;
859 		}
860 		return (error);
861 	}
862 
863 	/*
864 	 * do not use a rejected or black hole route.
865 	 * XXX: this check should be done in the L2 output routine.
866 	 * However, if we skipped this check here, we'd see the following
867 	 * scenario:
868 	 * - install a rejected route for a scoped address prefix
869 	 *   (like fe80::/10)
870 	 * - send a packet to a destination that matches the scoped prefix,
871 	 *   with ambiguity about the scope zone.
872 	 * - pick the outgoing interface from the route, and disambiguate the
873 	 *   scope zone with the interface.
874 	 * - ip6_output() would try to get another route with the "new"
875 	 *   destination, which may be valid.
876 	 * - we'd see no error on output.
877 	 * Although this may not be very harmful, it should still be confusing.
878 	 * We thus reject the case here.
879 	 */
880 
881 	if (nh_flags & (NHF_REJECT | NHF_BLACKHOLE)) {
882 		error = (nh_flags & NHF_HOST ? EHOSTUNREACH : ENETUNREACH);
883 		return (error);
884 	}
885 
886 	return (0);
887 }
888 
889 /* Public wrapper function to selectroute(). */
890 int
891 in6_selectroute(struct sockaddr_in6 *dstsock, struct ip6_pktopts *opts,
892     struct ip6_moptions *mopts, struct route_in6 *ro,
893     struct ifnet **retifp, struct nhop_object **retnh, u_int fibnum, uint32_t flowid)
894 {
895 
896 	return (selectroute(dstsock, opts, mopts, ro, retifp,
897 	    retnh, 0, fibnum, flowid));
898 }
899 
900 /*
901  * Default hop limit selection. The precedence is as follows:
902  * 1. Hoplimit value specified via ioctl.
903  * 2. (If the outgoing interface is detected) the current
904  *     hop limit of the interface specified by router advertisement.
905  * 3. The system default hoplimit.
906  */
907 int
908 in6_selecthlim(struct inpcb *inp, struct ifnet *ifp)
909 {
910 
911 	if (inp && inp->in6p_hops >= 0)
912 		return (inp->in6p_hops);
913 	else if (ifp)
914 		return (ND_IFINFO(ifp)->chlim);
915 	else if (inp && !IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_faddr)) {
916 		struct nhop_object *nh;
917 		struct in6_addr dst;
918 		uint32_t fibnum, scopeid;
919 		int hlim;
920 
921 		fibnum = inp->inp_inc.inc_fibnum;
922 		in6_splitscope(&inp->in6p_faddr, &dst, &scopeid);
923 		nh = fib6_lookup(fibnum, &dst, scopeid, 0, 0);
924 		if (nh != NULL) {
925 			hlim = ND_IFINFO(nh->nh_ifp)->chlim;
926 			return (hlim);
927 		}
928 	}
929 	return (V_ip6_defhlim);
930 }
931 
932 /*
933  * XXX: this is borrowed from in6_pcbbind(). If possible, we should
934  * share this function by all *bsd*...
935  */
936 int
937 in6_pcbsetport(struct in6_addr *laddr, struct inpcb *inp, struct ucred *cred)
938 {
939 	struct socket *so = inp->inp_socket;
940 	u_int16_t lport = 0;
941 	int error, lookupflags = 0;
942 #ifdef INVARIANTS
943 	struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
944 #endif
945 
946 	INP_WLOCK_ASSERT(inp);
947 	INP_HASH_WLOCK_ASSERT(pcbinfo);
948 
949 	error = prison_local_ip6(cred, laddr,
950 	    ((inp->inp_flags & IN6P_IPV6_V6ONLY) != 0));
951 	if (error)
952 		return(error);
953 
954 	/* XXX: this is redundant when called from in6_pcbbind */
955 	if ((so->so_options & (SO_REUSEADDR|SO_REUSEPORT|SO_REUSEPORT_LB)) == 0)
956 		lookupflags = INPLOOKUP_WILDCARD;
957 
958 	inp->inp_flags |= INP_ANONPORT;
959 
960 	error = in_pcb_lport(inp, NULL, &lport, cred, lookupflags);
961 	if (error != 0)
962 		return (error);
963 
964 	inp->inp_lport = lport;
965 	if (in_pcbinshash(inp) != 0) {
966 		inp->in6p_laddr = in6addr_any;
967 		inp->inp_lport = 0;
968 		return (EAGAIN);
969 	}
970 
971 	return (0);
972 }
973 
974 void
975 addrsel_policy_init(void)
976 {
977 
978 	init_policy_queue();
979 
980 	/* initialize the "last resort" policy */
981 	bzero(&V_defaultaddrpolicy, sizeof(V_defaultaddrpolicy));
982 	V_defaultaddrpolicy.label = ADDR_LABEL_NOTAPP;
983 
984 	if (!IS_DEFAULT_VNET(curvnet))
985 		return;
986 
987 	ADDRSEL_LOCK_INIT();
988 	ADDRSEL_SXLOCK_INIT();
989 }
990 
991 static struct in6_addrpolicy *
992 lookup_addrsel_policy(struct sockaddr_in6 *key)
993 {
994 	struct in6_addrpolicy *match = NULL;
995 
996 	ADDRSEL_LOCK();
997 	match = match_addrsel_policy(key);
998 
999 	if (match == NULL)
1000 		match = &V_defaultaddrpolicy;
1001 	else
1002 		match->use++;
1003 	ADDRSEL_UNLOCK();
1004 
1005 	return (match);
1006 }
1007 
1008 /*
1009  * Subroutines to manage the address selection policy table via sysctl.
1010  */
1011 struct walkarg {
1012 	struct sysctl_req *w_req;
1013 };
1014 
1015 static int in6_src_sysctl(SYSCTL_HANDLER_ARGS);
1016 SYSCTL_DECL(_net_inet6_ip6);
1017 static SYSCTL_NODE(_net_inet6_ip6, IPV6CTL_ADDRCTLPOLICY, addrctlpolicy,
1018     CTLFLAG_RD | CTLFLAG_MPSAFE, in6_src_sysctl,
1019     "");
1020 
1021 static int
1022 in6_src_sysctl(SYSCTL_HANDLER_ARGS)
1023 {
1024 	struct walkarg w;
1025 
1026 	if (req->newptr)
1027 		return EPERM;
1028 
1029 	bzero(&w, sizeof(w));
1030 	w.w_req = req;
1031 
1032 	return (walk_addrsel_policy(dump_addrsel_policyent, &w));
1033 }
1034 
1035 int
1036 in6_src_ioctl(u_long cmd, caddr_t data)
1037 {
1038 	struct in6_addrpolicy ent0;
1039 
1040 	if (cmd != SIOCAADDRCTL_POLICY && cmd != SIOCDADDRCTL_POLICY)
1041 		return (EOPNOTSUPP); /* check for safety */
1042 
1043 	ent0 = *(struct in6_addrpolicy *)data;
1044 
1045 	if (ent0.label == ADDR_LABEL_NOTAPP)
1046 		return (EINVAL);
1047 	/* check if the prefix mask is consecutive. */
1048 	if (in6_mask2len(&ent0.addrmask.sin6_addr, NULL) < 0)
1049 		return (EINVAL);
1050 	/* clear trailing garbages (if any) of the prefix address. */
1051 	IN6_MASK_ADDR(&ent0.addr.sin6_addr, &ent0.addrmask.sin6_addr);
1052 	ent0.use = 0;
1053 
1054 	switch (cmd) {
1055 	case SIOCAADDRCTL_POLICY:
1056 		return (add_addrsel_policyent(&ent0));
1057 	case SIOCDADDRCTL_POLICY:
1058 		return (delete_addrsel_policyent(&ent0));
1059 	}
1060 
1061 	return (0);		/* XXX: compromise compilers */
1062 }
1063 
1064 /*
1065  * The followings are implementation of the policy table using a
1066  * simple tail queue.
1067  * XXX such details should be hidden.
1068  * XXX implementation using binary tree should be more efficient.
1069  */
1070 struct addrsel_policyent {
1071 	TAILQ_ENTRY(addrsel_policyent) ape_entry;
1072 	struct in6_addrpolicy ape_policy;
1073 };
1074 
1075 TAILQ_HEAD(addrsel_policyhead, addrsel_policyent);
1076 
1077 VNET_DEFINE_STATIC(struct addrsel_policyhead, addrsel_policytab);
1078 #define	V_addrsel_policytab		VNET(addrsel_policytab)
1079 
1080 static void
1081 init_policy_queue(void)
1082 {
1083 
1084 	TAILQ_INIT(&V_addrsel_policytab);
1085 }
1086 
1087 static int
1088 add_addrsel_policyent(struct in6_addrpolicy *newpolicy)
1089 {
1090 	struct addrsel_policyent *new, *pol;
1091 
1092 	new = malloc(sizeof(*new), M_IFADDR,
1093 	       M_WAITOK);
1094 	ADDRSEL_XLOCK();
1095 	ADDRSEL_LOCK();
1096 
1097 	/* duplication check */
1098 	TAILQ_FOREACH(pol, &V_addrsel_policytab, ape_entry) {
1099 		if (IN6_ARE_ADDR_EQUAL(&newpolicy->addr.sin6_addr,
1100 				       &pol->ape_policy.addr.sin6_addr) &&
1101 		    IN6_ARE_ADDR_EQUAL(&newpolicy->addrmask.sin6_addr,
1102 				       &pol->ape_policy.addrmask.sin6_addr)) {
1103 			ADDRSEL_UNLOCK();
1104 			ADDRSEL_XUNLOCK();
1105 			free(new, M_IFADDR);
1106 			return (EEXIST);	/* or override it? */
1107 		}
1108 	}
1109 
1110 	bzero(new, sizeof(*new));
1111 
1112 	/* XXX: should validate entry */
1113 	new->ape_policy = *newpolicy;
1114 
1115 	TAILQ_INSERT_TAIL(&V_addrsel_policytab, new, ape_entry);
1116 	ADDRSEL_UNLOCK();
1117 	ADDRSEL_XUNLOCK();
1118 
1119 	return (0);
1120 }
1121 
1122 static int
1123 delete_addrsel_policyent(struct in6_addrpolicy *key)
1124 {
1125 	struct addrsel_policyent *pol;
1126 
1127 	ADDRSEL_XLOCK();
1128 	ADDRSEL_LOCK();
1129 
1130 	/* search for the entry in the table */
1131 	TAILQ_FOREACH(pol, &V_addrsel_policytab, ape_entry) {
1132 		if (IN6_ARE_ADDR_EQUAL(&key->addr.sin6_addr,
1133 		    &pol->ape_policy.addr.sin6_addr) &&
1134 		    IN6_ARE_ADDR_EQUAL(&key->addrmask.sin6_addr,
1135 		    &pol->ape_policy.addrmask.sin6_addr)) {
1136 			break;
1137 		}
1138 	}
1139 	if (pol == NULL) {
1140 		ADDRSEL_UNLOCK();
1141 		ADDRSEL_XUNLOCK();
1142 		return (ESRCH);
1143 	}
1144 
1145 	TAILQ_REMOVE(&V_addrsel_policytab, pol, ape_entry);
1146 	ADDRSEL_UNLOCK();
1147 	ADDRSEL_XUNLOCK();
1148 	free(pol, M_IFADDR);
1149 
1150 	return (0);
1151 }
1152 
1153 static int
1154 walk_addrsel_policy(int (*callback)(struct in6_addrpolicy *, void *), void *w)
1155 {
1156 	struct addrsel_policyent *pol;
1157 	int error = 0;
1158 
1159 	ADDRSEL_SLOCK();
1160 	TAILQ_FOREACH(pol, &V_addrsel_policytab, ape_entry) {
1161 		if ((error = (*callback)(&pol->ape_policy, w)) != 0) {
1162 			ADDRSEL_SUNLOCK();
1163 			return (error);
1164 		}
1165 	}
1166 	ADDRSEL_SUNLOCK();
1167 	return (error);
1168 }
1169 
1170 static int
1171 dump_addrsel_policyent(struct in6_addrpolicy *pol, void *arg)
1172 {
1173 	int error = 0;
1174 	struct walkarg *w = arg;
1175 
1176 	error = SYSCTL_OUT(w->w_req, pol, sizeof(*pol));
1177 
1178 	return (error);
1179 }
1180 
1181 static struct in6_addrpolicy *
1182 match_addrsel_policy(struct sockaddr_in6 *key)
1183 {
1184 	struct addrsel_policyent *pent;
1185 	struct in6_addrpolicy *bestpol = NULL, *pol;
1186 	int matchlen, bestmatchlen = -1;
1187 	u_char *mp, *ep, *k, *p, m;
1188 
1189 	TAILQ_FOREACH(pent, &V_addrsel_policytab, ape_entry) {
1190 		matchlen = 0;
1191 
1192 		pol = &pent->ape_policy;
1193 		mp = (u_char *)&pol->addrmask.sin6_addr;
1194 		ep = mp + 16;	/* XXX: scope field? */
1195 		k = (u_char *)&key->sin6_addr;
1196 		p = (u_char *)&pol->addr.sin6_addr;
1197 		for (; mp < ep && *mp; mp++, k++, p++) {
1198 			m = *mp;
1199 			if ((*k & m) != *p)
1200 				goto next; /* not match */
1201 			if (m == 0xff) /* short cut for a typical case */
1202 				matchlen += 8;
1203 			else {
1204 				while (m >= 0x80) {
1205 					matchlen++;
1206 					m <<= 1;
1207 				}
1208 			}
1209 		}
1210 
1211 		/* matched.  check if this is better than the current best. */
1212 		if (bestpol == NULL ||
1213 		    matchlen > bestmatchlen) {
1214 			bestpol = pol;
1215 			bestmatchlen = matchlen;
1216 		}
1217 
1218 	  next:
1219 		continue;
1220 	}
1221 
1222 	return (bestpol);
1223 }
1224