xref: /freebsd/sys/netinet6/in6_src.c (revision d5b0e70f7e04d971691517ce1304d86a1e367e2e)
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 	NET_EPOCH_ASSERT();
195 	KASSERT(srcp != NULL, ("%s: srcp is NULL", __func__));
196 
197 	dst = dstsock->sin6_addr; /* make a copy for local operation */
198 	if (ifpp) {
199 		/*
200 		 * Save a possibly passed in ifp for in6_selectsrc. Only
201 		 * neighbor discovery code should use this feature, where
202 		 * we may know the interface but not the FIB number holding
203 		 * the connected subnet in case someone deleted it from the
204 		 * default FIB and we need to check the interface.
205 		 */
206 		if (*ifpp != NULL)
207 			oifp = *ifpp;
208 		*ifpp = NULL;
209 	}
210 
211 	if (inp != NULL) {
212 		INP_LOCK_ASSERT(inp);
213 		mopts = inp->in6p_moptions;
214 	} else {
215 		mopts = NULL;
216 	}
217 
218 	/*
219 	 * If the source address is explicitly specified by the caller,
220 	 * check if the requested source address is indeed a unicast address
221 	 * assigned to the node, and can be used as the packet's source
222 	 * address.  If everything is okay, use the address as source.
223 	 */
224 	if (opts && (pi = opts->ip6po_pktinfo) &&
225 	    !IN6_IS_ADDR_UNSPECIFIED(&pi->ipi6_addr)) {
226 		/* get the outgoing interface */
227 		if ((error = in6_selectif(dstsock, opts, mopts, &ifp, oifp,
228 		    fibnum))
229 		    != 0)
230 			return (error);
231 
232 		/*
233 		 * determine the appropriate zone id of the source based on
234 		 * the zone of the destination and the outgoing interface.
235 		 * If the specified address is ambiguous wrt the scope zone,
236 		 * the interface must be specified; otherwise, ifa_ifwithaddr()
237 		 * will fail matching the address.
238 		 */
239 		tmp = pi->ipi6_addr;
240 		if (ifp) {
241 			error = in6_setscope(&tmp, ifp, &odstzone);
242 			if (error)
243 				return (error);
244 		}
245 		if (cred != NULL && (error = prison_local_ip6(cred,
246 		    &tmp, (inp->inp_flags & IN6P_IPV6_V6ONLY) != 0)) != 0)
247 			return (error);
248 
249 		/*
250 		 * If IPV6_BINDANY socket option is set, we allow to specify
251 		 * non local addresses as source address in IPV6_PKTINFO
252 		 * ancillary data.
253 		 */
254 		if ((inp->inp_flags & INP_BINDANY) == 0) {
255 			ia = in6ifa_ifwithaddr(&tmp, 0 /* XXX */, false);
256 			if (ia == NULL || (ia->ia6_flags & (IN6_IFF_ANYCAST |
257 			    IN6_IFF_NOTREADY)))
258 				return (EADDRNOTAVAIL);
259 			bcopy(&ia->ia_addr.sin6_addr, srcp, sizeof(*srcp));
260 		} else
261 			bcopy(&tmp, srcp, sizeof(*srcp));
262 		pi->ipi6_addr = tmp; /* XXX: this overrides pi */
263 		if (ifpp)
264 			*ifpp = ifp;
265 		return (0);
266 	}
267 
268 	/*
269 	 * Otherwise, if the socket has already bound the source, just use it.
270 	 */
271 	if (inp != NULL && !IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr)) {
272 		if (cred != NULL &&
273 		    (error = prison_local_ip6(cred, &inp->in6p_laddr,
274 		    ((inp->inp_flags & IN6P_IPV6_V6ONLY) != 0))) != 0)
275 			return (error);
276 		bcopy(&inp->in6p_laddr, srcp, sizeof(*srcp));
277 		return (0);
278 	}
279 
280 	/*
281 	 * Bypass source address selection and use the primary jail IP
282 	 * if requested.
283 	 */
284 	if (cred != NULL && !prison_saddrsel_ip6(cred, srcp))
285 		return (0);
286 
287 	/*
288 	 * If the address is not specified, choose the best one based on
289 	 * the outgoing interface and the destination address.
290 	 */
291 	/* get the outgoing interface */
292 	if ((error = in6_selectif(dstsock, opts, mopts, &ifp, oifp,
293 	    (inp != NULL) ? inp->inp_inc.inc_fibnum : fibnum)) != 0)
294 		return (error);
295 
296 #ifdef DIAGNOSTIC
297 	if (ifp == NULL)	/* this should not happen */
298 		panic("in6_selectsrc: NULL ifp");
299 #endif
300 	error = in6_setscope(&dst, ifp, &odstzone);
301 	if (error)
302 		return (error);
303 
304 	IN6_IFADDR_RLOCK(&in6_ifa_tracker);
305 	CK_STAILQ_FOREACH(ia, &V_in6_ifaddrhead, ia_link) {
306 		int new_scope = -1, new_matchlen = -1;
307 		struct in6_addrpolicy *new_policy = NULL;
308 		u_int32_t srczone, osrczone, dstzone;
309 		struct in6_addr src;
310 		struct ifnet *ifp1 = ia->ia_ifp;
311 
312 		/*
313 		 * We'll never take an address that breaks the scope zone
314 		 * of the destination.  We also skip an address if its zone
315 		 * does not contain the outgoing interface.
316 		 * XXX: we should probably use sin6_scope_id here.
317 		 */
318 		if (in6_setscope(&dst, ifp1, &dstzone) ||
319 		    odstzone != dstzone) {
320 			continue;
321 		}
322 		src = ia->ia_addr.sin6_addr;
323 		if (in6_setscope(&src, ifp, &osrczone) ||
324 		    in6_setscope(&src, ifp1, &srczone) ||
325 		    osrczone != srczone) {
326 			continue;
327 		}
328 
329 		/* avoid unusable addresses */
330 		if ((ia->ia6_flags &
331 		     (IN6_IFF_NOTREADY | IN6_IFF_ANYCAST | IN6_IFF_DETACHED))) {
332 				continue;
333 		}
334 		if (!V_ip6_use_deprecated && IFA6_IS_DEPRECATED(ia))
335 			continue;
336 
337 		/* If jailed only take addresses of the jail into account. */
338 		if (cred != NULL &&
339 		    prison_check_ip6(cred, &ia->ia_addr.sin6_addr) != 0)
340 			continue;
341 
342 		/* Rule 1: Prefer same address */
343 		if (IN6_ARE_ADDR_EQUAL(&dst, &ia->ia_addr.sin6_addr)) {
344 			ia_best = ia;
345 			BREAK(1); /* there should be no better candidate */
346 		}
347 
348 		if (ia_best == NULL)
349 			REPLACE(0);
350 
351 		/* Rule 2: Prefer appropriate scope */
352 		if (dst_scope < 0)
353 			dst_scope = in6_addrscope(&dst);
354 		new_scope = in6_addrscope(&ia->ia_addr.sin6_addr);
355 		if (IN6_ARE_SCOPE_CMP(best_scope, new_scope) < 0) {
356 			if (IN6_ARE_SCOPE_CMP(best_scope, dst_scope) < 0)
357 				REPLACE(2);
358 			NEXT(2);
359 		} else if (IN6_ARE_SCOPE_CMP(new_scope, best_scope) < 0) {
360 			if (IN6_ARE_SCOPE_CMP(new_scope, dst_scope) < 0)
361 				NEXT(2);
362 			REPLACE(2);
363 		}
364 
365 		/*
366 		 * Rule 3: Avoid deprecated addresses.  Note that the case of
367 		 * !ip6_use_deprecated is already rejected above.
368 		 */
369 		if (!IFA6_IS_DEPRECATED(ia_best) && IFA6_IS_DEPRECATED(ia))
370 			NEXT(3);
371 		if (IFA6_IS_DEPRECATED(ia_best) && !IFA6_IS_DEPRECATED(ia))
372 			REPLACE(3);
373 
374 		/* Rule 4: Prefer home addresses */
375 		/*
376 		 * XXX: This is a TODO.  We should probably merge the MIP6
377 		 * case above.
378 		 */
379 
380 		/* Rule 5: Prefer outgoing interface */
381 		if (!(ND_IFINFO(ifp)->flags & ND6_IFF_NO_PREFER_IFACE)) {
382 			if (ia_best->ia_ifp == ifp && ia->ia_ifp != ifp)
383 				NEXT(5);
384 			if (ia_best->ia_ifp != ifp && ia->ia_ifp == ifp)
385 				REPLACE(5);
386 		}
387 
388 		/*
389 		 * Rule 6: Prefer matching label
390 		 * Note that best_policy should be non-NULL here.
391 		 */
392 		if (dst_policy == NULL)
393 			dst_policy = lookup_addrsel_policy(dstsock);
394 		if (dst_policy->label != ADDR_LABEL_NOTAPP) {
395 			new_policy = lookup_addrsel_policy(&ia->ia_addr);
396 			if (dst_policy->label == best_policy->label &&
397 			    dst_policy->label != new_policy->label)
398 				NEXT(6);
399 			if (dst_policy->label != best_policy->label &&
400 			    dst_policy->label == new_policy->label)
401 				REPLACE(6);
402 		}
403 
404 		/*
405 		 * Rule 7: Prefer public addresses.
406 		 * We allow users to reverse the logic by configuring
407 		 * a sysctl variable, so that privacy conscious users can
408 		 * always prefer temporary addresses.
409 		 */
410 		if (opts == NULL ||
411 		    opts->ip6po_prefer_tempaddr == IP6PO_TEMPADDR_SYSTEM) {
412 			prefer_tempaddr = V_ip6_prefer_tempaddr;
413 		} else if (opts->ip6po_prefer_tempaddr ==
414 		    IP6PO_TEMPADDR_NOTPREFER) {
415 			prefer_tempaddr = 0;
416 		} else
417 			prefer_tempaddr = 1;
418 		if (!(ia_best->ia6_flags & IN6_IFF_TEMPORARY) &&
419 		    (ia->ia6_flags & IN6_IFF_TEMPORARY)) {
420 			if (prefer_tempaddr)
421 				REPLACE(7);
422 			else
423 				NEXT(7);
424 		}
425 		if ((ia_best->ia6_flags & IN6_IFF_TEMPORARY) &&
426 		    !(ia->ia6_flags & IN6_IFF_TEMPORARY)) {
427 			if (prefer_tempaddr)
428 				NEXT(7);
429 			else
430 				REPLACE(7);
431 		}
432 
433 		/*
434 		 * Rule 8: prefer addresses on alive interfaces.
435 		 * This is a KAME specific rule.
436 		 */
437 		if ((ia_best->ia_ifp->if_flags & IFF_UP) &&
438 		    !(ia->ia_ifp->if_flags & IFF_UP))
439 			NEXT(8);
440 		if (!(ia_best->ia_ifp->if_flags & IFF_UP) &&
441 		    (ia->ia_ifp->if_flags & IFF_UP))
442 			REPLACE(8);
443 
444 		/*
445 		 * Rule 9: prefer address with better virtual status.
446 		 */
447 		if (ifa_preferred(&ia_best->ia_ifa, &ia->ia_ifa))
448 			REPLACE(9);
449 		if (ifa_preferred(&ia->ia_ifa, &ia_best->ia_ifa))
450 			NEXT(9);
451 
452 		/*
453 		 * Rule 10: prefer address with `prefer_source' flag.
454 		 */
455 		if ((ia_best->ia6_flags & IN6_IFF_PREFER_SOURCE) == 0 &&
456 		    (ia->ia6_flags & IN6_IFF_PREFER_SOURCE) != 0)
457 			REPLACE(10);
458 		if ((ia_best->ia6_flags & IN6_IFF_PREFER_SOURCE) != 0 &&
459 		    (ia->ia6_flags & IN6_IFF_PREFER_SOURCE) == 0)
460 			NEXT(10);
461 
462 		/*
463 		 * Rule 14: Use longest matching prefix.
464 		 * Note: in the address selection draft, this rule is
465 		 * documented as "Rule 8".  However, since it is also
466 		 * documented that this rule can be overridden, we assign
467 		 * a large number so that it is easy to assign smaller numbers
468 		 * to more preferred rules.
469 		 */
470 		new_matchlen = in6_matchlen(&ia->ia_addr.sin6_addr, &dst);
471 		if (best_matchlen < new_matchlen)
472 			REPLACE(14);
473 		if (new_matchlen < best_matchlen)
474 			NEXT(14);
475 
476 		/* Rule 15 is reserved. */
477 
478 		/*
479 		 * Last resort: just keep the current candidate.
480 		 * Or, do we need more rules?
481 		 */
482 		continue;
483 
484 	  replace:
485 		ia_best = ia;
486 		best_scope = (new_scope >= 0 ? new_scope :
487 			      in6_addrscope(&ia_best->ia_addr.sin6_addr));
488 		best_policy = (new_policy ? new_policy :
489 			       lookup_addrsel_policy(&ia_best->ia_addr));
490 		best_matchlen = (new_matchlen >= 0 ? new_matchlen :
491 				 in6_matchlen(&ia_best->ia_addr.sin6_addr,
492 					      &dst));
493 
494 	  next:
495 		continue;
496 
497 	  out:
498 		break;
499 	}
500 
501 	if ((ia = ia_best) == NULL) {
502 		IN6_IFADDR_RUNLOCK(&in6_ifa_tracker);
503 		IP6STAT_INC(ip6s_sources_none);
504 		return (EADDRNOTAVAIL);
505 	}
506 
507 	/*
508 	 * At this point at least one of the addresses belonged to the jail
509 	 * but it could still be, that we want to further restrict it, e.g.
510 	 * theoratically IN6_IS_ADDR_LOOPBACK.
511 	 * It must not be IN6_IS_ADDR_UNSPECIFIED anymore.
512 	 * prison_local_ip6() will fix an IN6_IS_ADDR_LOOPBACK but should
513 	 * let all others previously selected pass.
514 	 * Use tmp to not change ::1 on lo0 to the primary jail address.
515 	 */
516 	tmp = ia->ia_addr.sin6_addr;
517 	if (cred != NULL && prison_local_ip6(cred, &tmp, (inp != NULL &&
518 	    (inp->inp_flags & IN6P_IPV6_V6ONLY) != 0)) != 0) {
519 		IN6_IFADDR_RUNLOCK(&in6_ifa_tracker);
520 		IP6STAT_INC(ip6s_sources_none);
521 		return (EADDRNOTAVAIL);
522 	}
523 
524 	if (ifpp)
525 		*ifpp = ifp;
526 
527 	bcopy(&tmp, srcp, sizeof(*srcp));
528 	if (ia->ia_ifp == ifp)
529 		IP6STAT_INC(ip6s_sources_sameif[best_scope]);
530 	else
531 		IP6STAT_INC(ip6s_sources_otherif[best_scope]);
532 	if (dst_scope == best_scope)
533 		IP6STAT_INC(ip6s_sources_samescope[best_scope]);
534 	else
535 		IP6STAT_INC(ip6s_sources_otherscope[best_scope]);
536 	if (IFA6_IS_DEPRECATED(ia))
537 		IP6STAT_INC(ip6s_sources_deprecated[best_scope]);
538 	IN6_IFADDR_RUNLOCK(&in6_ifa_tracker);
539 	return (0);
540 }
541 
542 /*
543  * Select source address based on @inp, @dstsock and @opts.
544  * Stores selected address to @srcp. If @scope_ambiguous is set,
545  * embed scope from selected outgoing interface. If @hlim pointer
546  * is provided, stores calculated hop limit there.
547  * Returns 0 on success.
548  */
549 int
550 in6_selectsrc_socket(struct sockaddr_in6 *dstsock, struct ip6_pktopts *opts,
551     struct inpcb *inp, struct ucred *cred, int scope_ambiguous,
552     struct in6_addr *srcp, int *hlim)
553 {
554 	struct ifnet *retifp;
555 	uint32_t fibnum;
556 	int error;
557 
558 	fibnum = inp->inp_inc.inc_fibnum;
559 	retifp = NULL;
560 
561 	error = in6_selectsrc(fibnum, dstsock, opts, inp, cred, &retifp, srcp);
562 	if (error != 0)
563 		return (error);
564 
565 	if (hlim != NULL)
566 		*hlim = in6_selecthlim(inp, retifp);
567 
568 	if (retifp == NULL || scope_ambiguous == 0)
569 		return (0);
570 
571 	/*
572 	 * Application should provide a proper zone ID or the use of
573 	 * default zone IDs should be enabled.  Unfortunately, some
574 	 * applications do not behave as it should, so we need a
575 	 * workaround.  Even if an appropriate ID is not determined
576 	 * (when it's required), if we can determine the outgoing
577 	 * interface. determine the zone ID based on the interface.
578 	 */
579 	error = in6_setscope(&dstsock->sin6_addr, retifp, NULL);
580 
581 	return (error);
582 }
583 
584 /*
585  * Select source address based on @fibnum, @dst and @scopeid.
586  * Stores selected address to @srcp.
587  * Returns 0 on success.
588  *
589  * Used by non-socket based consumers (ND code mostly)
590  */
591 int
592 in6_selectsrc_addr(uint32_t fibnum, const struct in6_addr *dst,
593     uint32_t scopeid, struct ifnet *ifp, struct in6_addr *srcp,
594     int *hlim)
595 {
596 	struct ifnet *retifp;
597 	struct sockaddr_in6 dst_sa;
598 	int error;
599 
600 	retifp = ifp;
601 	bzero(&dst_sa, sizeof(dst_sa));
602 	dst_sa.sin6_family = AF_INET6;
603 	dst_sa.sin6_len = sizeof(dst_sa);
604 	dst_sa.sin6_addr = *dst;
605 	dst_sa.sin6_scope_id = scopeid;
606 	sa6_embedscope(&dst_sa, 0);
607 
608 	error = in6_selectsrc(fibnum, &dst_sa, NULL, NULL, NULL, &retifp, srcp);
609 	if (hlim != NULL)
610 		*hlim = in6_selecthlim(NULL, retifp);
611 
612 	return (error);
613 }
614 
615 static struct nhop_object *
616 cache_route(uint32_t fibnum, const struct sockaddr_in6 *dst, struct route_in6 *ro,
617     uint32_t flowid)
618 {
619 	/*
620 	 * Use a cached route if it exists and is valid, else try to allocate
621 	 * a new one. Note that we should check the address family of the
622 	 * cached destination, in case of sharing the cache with IPv4.
623 	 * Assumes that 'struct route_in6' is exclusively locked.
624 	 */
625 	if (ro->ro_nh != NULL && (
626 	    !NH_IS_VALID(ro->ro_nh) || ro->ro_dst.sin6_family != AF_INET6 ||
627 	    !IN6_ARE_ADDR_EQUAL(&ro->ro_dst.sin6_addr, &dst->sin6_addr)))
628 		RO_NHFREE(ro);
629 
630 	if (ro->ro_nh == NULL) {
631 		ro->ro_dst = *dst;
632 
633 		const struct in6_addr *paddr;
634 		struct in6_addr unscoped_addr;
635 		uint32_t scopeid = 0;
636 		if (IN6_IS_SCOPE_LINKLOCAL(&dst->sin6_addr)) {
637 			in6_splitscope(&dst->sin6_addr, &unscoped_addr, &scopeid);
638 			paddr = &unscoped_addr;
639 		} else
640 			paddr = &dst->sin6_addr;
641 		ro->ro_nh = fib6_lookup(fibnum, paddr, scopeid, NHR_REF, flowid);
642 	}
643 	return (ro->ro_nh);
644 }
645 
646 static struct nhop_object *
647 lookup_route(uint32_t fibnum, struct sockaddr_in6 *dst, struct route_in6 *ro,
648     struct ip6_pktopts *opts, uint32_t flowid)
649 {
650 	struct nhop_object *nh = NULL;
651 
652 	/*
653 	 * If the next hop address for the packet is specified by the caller,
654 	 * use it as the gateway.
655 	 */
656 	if (opts && opts->ip6po_nexthop) {
657 		struct route_in6 *ron = &opts->ip6po_nextroute;
658 		struct sockaddr_in6 *sin6_next = satosin6(opts->ip6po_nexthop);
659 
660 		nh = cache_route(fibnum, sin6_next, ron, flowid);
661 
662 		/*
663 		 * The node identified by that address must be a
664 		 * neighbor of the sending host.
665 		 */
666 		if (nh != NULL && (nh->nh_flags & NHF_GATEWAY) != 0)
667 			nh = NULL;
668 	} else if (ro != NULL) {
669 		nh = cache_route(fibnum, dst, ro, flowid);
670 		if (nh == NULL)
671 			return (NULL);
672 
673 		/*
674 		 * Check if the outgoing interface conflicts with
675 		 * the interface specified by ipi6_ifindex (if specified).
676 		 */
677 		struct in6_pktinfo *pi;
678 		if (opts && (pi = opts->ip6po_pktinfo) != NULL && pi->ipi6_ifindex) {
679 			if (nh->nh_aifp->if_index != pi->ipi6_ifindex)
680 				nh = NULL;
681 		}
682 	}
683 
684 	return (nh);
685 }
686 
687 /*
688  * Finds outgoing nexthop or the outgoing interface for the
689  * @dstsock.
690  * Return 0 on success and stores the lookup result in @retnh and @retifp
691  */
692 static int
693 selectroute(struct sockaddr_in6 *dstsock, struct ip6_pktopts *opts,
694     struct ip6_moptions *mopts, struct route_in6 *ro,
695     struct ifnet **retifp, struct nhop_object **retnh, int norouteok,
696     u_int fibnum, uint32_t flowid)
697 {
698 	int error = 0;
699 	struct ifnet *ifp = NULL;
700 	struct in6_pktinfo *pi = NULL;
701 	struct in6_addr *dst = &dstsock->sin6_addr;
702 
703 	/* If the caller specify the outgoing interface explicitly, use it. */
704 	if (opts && (pi = opts->ip6po_pktinfo) != NULL && pi->ipi6_ifindex) {
705 		/* XXX boundary check is assumed to be already done. */
706 		ifp = ifnet_byindex(pi->ipi6_ifindex);
707 		if (ifp != NULL && (norouteok || IN6_IS_ADDR_MULTICAST(dst))) {
708 			/*
709 			 * we do not have to check or get the route for
710 			 * multicast.
711 			 */
712 			goto done;
713 		} else
714 			goto getroute;
715 	}
716 	/*
717 	 * If the destination address is a multicast address and the outgoing
718 	 * interface for the address is specified by the caller, use it.
719 	 */
720 	if (IN6_IS_ADDR_MULTICAST(dst) &&
721 	    mopts != NULL && (ifp = mopts->im6o_multicast_ifp) != NULL) {
722 		goto done; /* we do not need a route for multicast. */
723 	}
724 	/*
725 	 * If destination address is LLA or link- or node-local multicast,
726 	 * use it's embedded scope zone id to determine outgoing interface.
727 	 */
728 	if (IN6_IS_ADDR_MC_LINKLOCAL(dst) ||
729 	    IN6_IS_ADDR_MC_NODELOCAL(dst)) {
730 		uint32_t zoneid = ntohs(in6_getscope(dst));
731 		if (zoneid > 0) {
732 			ifp = in6_getlinkifnet(zoneid);
733 			goto done;
734 		}
735 	}
736 
737   getroute:;
738 	struct nhop_object *nh = lookup_route(fibnum, dstsock, ro, opts, flowid);
739 	if (nh != NULL) {
740 		*retifp = nh->nh_aifp;
741 		error = 0;
742 	} else {
743 		*retifp = NULL;
744 		IP6STAT_INC(ip6s_noroute);
745 		error = EHOSTUNREACH;
746 	}
747 	*retnh = nh;
748 	return (error);
749 
750   done:
751 	if (ifp == NULL) {
752 		/*
753 		 * This can happen if the caller did not pass a cached route
754 		 * nor any other hints.  We treat this case an error.
755 		 */
756 		error = EHOSTUNREACH;
757 	}
758 	if (error == EHOSTUNREACH)
759 		IP6STAT_INC(ip6s_noroute);
760 
761 	*retifp = ifp;
762 	*retnh = NULL;
763 
764 	return (error);
765 }
766 
767 static int
768 in6_selectif(struct sockaddr_in6 *dstsock, struct ip6_pktopts *opts,
769     struct ip6_moptions *mopts, struct ifnet **retifp,
770     struct ifnet *oifp, u_int fibnum)
771 {
772 	int error;
773 	struct route_in6 sro;
774 	struct nhop_object *nh = NULL;
775 	uint16_t nh_flags;
776 
777 	KASSERT(retifp != NULL, ("%s: retifp is NULL", __func__));
778 
779 	bzero(&sro, sizeof(sro));
780 	nh_flags = 0;
781 
782 	error = selectroute(dstsock, opts, mopts, &sro, retifp, &nh, 1, fibnum, 0);
783 
784 	if (nh != NULL)
785 		nh_flags = nh->nh_flags;
786 	if (nh != NULL && nh == sro.ro_nh)
787 		NH_FREE(nh);
788 
789 	if (error != 0) {
790 		/* Help ND. See oifp comment in in6_selectsrc(). */
791 		if (oifp != NULL && fibnum == RT_DEFAULT_FIB) {
792 			*retifp = oifp;
793 			error = 0;
794 		}
795 		return (error);
796 	}
797 
798 	/*
799 	 * do not use a rejected or black hole route.
800 	 * XXX: this check should be done in the L2 output routine.
801 	 * However, if we skipped this check here, we'd see the following
802 	 * scenario:
803 	 * - install a rejected route for a scoped address prefix
804 	 *   (like fe80::/10)
805 	 * - send a packet to a destination that matches the scoped prefix,
806 	 *   with ambiguity about the scope zone.
807 	 * - pick the outgoing interface from the route, and disambiguate the
808 	 *   scope zone with the interface.
809 	 * - ip6_output() would try to get another route with the "new"
810 	 *   destination, which may be valid.
811 	 * - we'd see no error on output.
812 	 * Although this may not be very harmful, it should still be confusing.
813 	 * We thus reject the case here.
814 	 */
815 
816 	if (nh_flags & (NHF_REJECT | NHF_BLACKHOLE)) {
817 		error = (nh_flags & NHF_HOST ? EHOSTUNREACH : ENETUNREACH);
818 		return (error);
819 	}
820 
821 	return (0);
822 }
823 
824 /* Public wrapper function to selectroute(). */
825 int
826 in6_selectroute(struct sockaddr_in6 *dstsock, struct ip6_pktopts *opts,
827     struct ip6_moptions *mopts, struct route_in6 *ro,
828     struct ifnet **retifp, struct nhop_object **retnh, u_int fibnum, uint32_t flowid)
829 {
830 	MPASS(retifp != NULL);
831 	MPASS(retnh != NULL);
832 
833 	return (selectroute(dstsock, opts, mopts, ro, retifp,
834 	    retnh, 0, fibnum, flowid));
835 }
836 
837 /*
838  * Default hop limit selection. The precedence is as follows:
839  * 1. Hoplimit value specified via ioctl.
840  * 2. (If the outgoing interface is detected) the current
841  *     hop limit of the interface specified by router advertisement.
842  * 3. The system default hoplimit.
843  */
844 int
845 in6_selecthlim(struct inpcb *inp, struct ifnet *ifp)
846 {
847 
848 	if (inp && inp->in6p_hops >= 0)
849 		return (inp->in6p_hops);
850 	else if (ifp)
851 		return (ND_IFINFO(ifp)->chlim);
852 	else if (inp && !IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_faddr)) {
853 		struct nhop_object *nh;
854 		struct in6_addr dst;
855 		uint32_t fibnum, scopeid;
856 		int hlim;
857 
858 		fibnum = inp->inp_inc.inc_fibnum;
859 		in6_splitscope(&inp->in6p_faddr, &dst, &scopeid);
860 		nh = fib6_lookup(fibnum, &dst, scopeid, 0, 0);
861 		if (nh != NULL) {
862 			hlim = ND_IFINFO(nh->nh_ifp)->chlim;
863 			return (hlim);
864 		}
865 	}
866 	return (V_ip6_defhlim);
867 }
868 
869 void
870 addrsel_policy_init(void)
871 {
872 
873 	init_policy_queue();
874 
875 	/* initialize the "last resort" policy */
876 	bzero(&V_defaultaddrpolicy, sizeof(V_defaultaddrpolicy));
877 	V_defaultaddrpolicy.label = ADDR_LABEL_NOTAPP;
878 
879 	if (!IS_DEFAULT_VNET(curvnet))
880 		return;
881 
882 	ADDRSEL_LOCK_INIT();
883 	ADDRSEL_SXLOCK_INIT();
884 }
885 
886 static struct in6_addrpolicy *
887 lookup_addrsel_policy(struct sockaddr_in6 *key)
888 {
889 	struct in6_addrpolicy *match = NULL;
890 
891 	ADDRSEL_LOCK();
892 	match = match_addrsel_policy(key);
893 
894 	if (match == NULL)
895 		match = &V_defaultaddrpolicy;
896 	else
897 		match->use++;
898 	ADDRSEL_UNLOCK();
899 
900 	return (match);
901 }
902 
903 /*
904  * Subroutines to manage the address selection policy table via sysctl.
905  */
906 struct walkarg {
907 	struct sysctl_req *w_req;
908 };
909 
910 static int in6_src_sysctl(SYSCTL_HANDLER_ARGS);
911 SYSCTL_DECL(_net_inet6_ip6);
912 static SYSCTL_NODE(_net_inet6_ip6, IPV6CTL_ADDRCTLPOLICY, addrctlpolicy,
913     CTLFLAG_RD | CTLFLAG_MPSAFE, in6_src_sysctl,
914     "");
915 
916 static int
917 in6_src_sysctl(SYSCTL_HANDLER_ARGS)
918 {
919 	struct walkarg w;
920 
921 	if (req->newptr)
922 		return EPERM;
923 
924 	bzero(&w, sizeof(w));
925 	w.w_req = req;
926 
927 	return (walk_addrsel_policy(dump_addrsel_policyent, &w));
928 }
929 
930 int
931 in6_src_ioctl(u_long cmd, caddr_t data)
932 {
933 	struct in6_addrpolicy ent0;
934 
935 	if (cmd != SIOCAADDRCTL_POLICY && cmd != SIOCDADDRCTL_POLICY)
936 		return (EOPNOTSUPP); /* check for safety */
937 
938 	ent0 = *(struct in6_addrpolicy *)data;
939 
940 	if (ent0.label == ADDR_LABEL_NOTAPP)
941 		return (EINVAL);
942 	/* check if the prefix mask is consecutive. */
943 	if (in6_mask2len(&ent0.addrmask.sin6_addr, NULL) < 0)
944 		return (EINVAL);
945 	/* clear trailing garbages (if any) of the prefix address. */
946 	IN6_MASK_ADDR(&ent0.addr.sin6_addr, &ent0.addrmask.sin6_addr);
947 	ent0.use = 0;
948 
949 	switch (cmd) {
950 	case SIOCAADDRCTL_POLICY:
951 		return (add_addrsel_policyent(&ent0));
952 	case SIOCDADDRCTL_POLICY:
953 		return (delete_addrsel_policyent(&ent0));
954 	}
955 
956 	return (0);		/* XXX: compromise compilers */
957 }
958 
959 /*
960  * The followings are implementation of the policy table using a
961  * simple tail queue.
962  * XXX such details should be hidden.
963  * XXX implementation using binary tree should be more efficient.
964  */
965 struct addrsel_policyent {
966 	TAILQ_ENTRY(addrsel_policyent) ape_entry;
967 	struct in6_addrpolicy ape_policy;
968 };
969 
970 TAILQ_HEAD(addrsel_policyhead, addrsel_policyent);
971 
972 VNET_DEFINE_STATIC(struct addrsel_policyhead, addrsel_policytab);
973 #define	V_addrsel_policytab		VNET(addrsel_policytab)
974 
975 static void
976 init_policy_queue(void)
977 {
978 
979 	TAILQ_INIT(&V_addrsel_policytab);
980 }
981 
982 static int
983 add_addrsel_policyent(struct in6_addrpolicy *newpolicy)
984 {
985 	struct addrsel_policyent *new, *pol;
986 
987 	new = malloc(sizeof(*new), M_IFADDR,
988 	       M_WAITOK);
989 	ADDRSEL_XLOCK();
990 	ADDRSEL_LOCK();
991 
992 	/* duplication check */
993 	TAILQ_FOREACH(pol, &V_addrsel_policytab, ape_entry) {
994 		if (IN6_ARE_ADDR_EQUAL(&newpolicy->addr.sin6_addr,
995 				       &pol->ape_policy.addr.sin6_addr) &&
996 		    IN6_ARE_ADDR_EQUAL(&newpolicy->addrmask.sin6_addr,
997 				       &pol->ape_policy.addrmask.sin6_addr)) {
998 			ADDRSEL_UNLOCK();
999 			ADDRSEL_XUNLOCK();
1000 			free(new, M_IFADDR);
1001 			return (EEXIST);	/* or override it? */
1002 		}
1003 	}
1004 
1005 	bzero(new, sizeof(*new));
1006 
1007 	/* XXX: should validate entry */
1008 	new->ape_policy = *newpolicy;
1009 
1010 	TAILQ_INSERT_TAIL(&V_addrsel_policytab, new, ape_entry);
1011 	ADDRSEL_UNLOCK();
1012 	ADDRSEL_XUNLOCK();
1013 
1014 	return (0);
1015 }
1016 
1017 static int
1018 delete_addrsel_policyent(struct in6_addrpolicy *key)
1019 {
1020 	struct addrsel_policyent *pol;
1021 
1022 	ADDRSEL_XLOCK();
1023 	ADDRSEL_LOCK();
1024 
1025 	/* search for the entry in the table */
1026 	TAILQ_FOREACH(pol, &V_addrsel_policytab, ape_entry) {
1027 		if (IN6_ARE_ADDR_EQUAL(&key->addr.sin6_addr,
1028 		    &pol->ape_policy.addr.sin6_addr) &&
1029 		    IN6_ARE_ADDR_EQUAL(&key->addrmask.sin6_addr,
1030 		    &pol->ape_policy.addrmask.sin6_addr)) {
1031 			break;
1032 		}
1033 	}
1034 	if (pol == NULL) {
1035 		ADDRSEL_UNLOCK();
1036 		ADDRSEL_XUNLOCK();
1037 		return (ESRCH);
1038 	}
1039 
1040 	TAILQ_REMOVE(&V_addrsel_policytab, pol, ape_entry);
1041 	ADDRSEL_UNLOCK();
1042 	ADDRSEL_XUNLOCK();
1043 	free(pol, M_IFADDR);
1044 
1045 	return (0);
1046 }
1047 
1048 static int
1049 walk_addrsel_policy(int (*callback)(struct in6_addrpolicy *, void *), void *w)
1050 {
1051 	struct addrsel_policyent *pol;
1052 	int error = 0;
1053 
1054 	ADDRSEL_SLOCK();
1055 	TAILQ_FOREACH(pol, &V_addrsel_policytab, ape_entry) {
1056 		if ((error = (*callback)(&pol->ape_policy, w)) != 0) {
1057 			ADDRSEL_SUNLOCK();
1058 			return (error);
1059 		}
1060 	}
1061 	ADDRSEL_SUNLOCK();
1062 	return (error);
1063 }
1064 
1065 static int
1066 dump_addrsel_policyent(struct in6_addrpolicy *pol, void *arg)
1067 {
1068 	int error = 0;
1069 	struct walkarg *w = arg;
1070 
1071 	error = SYSCTL_OUT(w->w_req, pol, sizeof(*pol));
1072 
1073 	return (error);
1074 }
1075 
1076 static struct in6_addrpolicy *
1077 match_addrsel_policy(struct sockaddr_in6 *key)
1078 {
1079 	struct addrsel_policyent *pent;
1080 	struct in6_addrpolicy *bestpol = NULL, *pol;
1081 	int matchlen, bestmatchlen = -1;
1082 	u_char *mp, *ep, *k, *p, m;
1083 
1084 	TAILQ_FOREACH(pent, &V_addrsel_policytab, ape_entry) {
1085 		matchlen = 0;
1086 
1087 		pol = &pent->ape_policy;
1088 		mp = (u_char *)&pol->addrmask.sin6_addr;
1089 		ep = mp + 16;	/* XXX: scope field? */
1090 		k = (u_char *)&key->sin6_addr;
1091 		p = (u_char *)&pol->addr.sin6_addr;
1092 		for (; mp < ep && *mp; mp++, k++, p++) {
1093 			m = *mp;
1094 			if ((*k & m) != *p)
1095 				goto next; /* not match */
1096 			if (m == 0xff) /* short cut for a typical case */
1097 				matchlen += 8;
1098 			else {
1099 				while (m >= 0x80) {
1100 					matchlen++;
1101 					m <<= 1;
1102 				}
1103 			}
1104 		}
1105 
1106 		/* matched.  check if this is better than the current best. */
1107 		if (bestpol == NULL ||
1108 		    matchlen > bestmatchlen) {
1109 			bestpol = pol;
1110 			bestmatchlen = matchlen;
1111 		}
1112 
1113 	  next:
1114 		continue;
1115 	}
1116 
1117 	return (bestpol);
1118 }
1119