xref: /freebsd/sys/net/route.c (revision f7167e0ea0bf5aaabff9490453b3b71b3f1b4d51)
1 /*-
2  * Copyright (c) 1980, 1986, 1991, 1993
3  *	The Regents of the University of California.  All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice, this list of conditions and the following disclaimer.
10  * 2. Redistributions in binary form must reproduce the above copyright
11  *    notice, this list of conditions and the following disclaimer in the
12  *    documentation and/or other materials provided with the distribution.
13  * 4. Neither the name of the University nor the names of its contributors
14  *    may be used to endorse or promote products derived from this software
15  *    without specific prior written permission.
16  *
17  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
18  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
21  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27  * SUCH DAMAGE.
28  *
29  *	@(#)route.c	8.3.1.1 (Berkeley) 2/23/95
30  * $FreeBSD$
31  */
32 /************************************************************************
33  * Note: In this file a 'fib' is a "forwarding information base"	*
34  * Which is the new name for an in kernel routing (next hop) table.	*
35  ***********************************************************************/
36 
37 #include "opt_inet.h"
38 #include "opt_inet6.h"
39 #include "opt_route.h"
40 #include "opt_sctp.h"
41 #include "opt_mrouting.h"
42 #include "opt_mpath.h"
43 
44 #include <sys/param.h>
45 #include <sys/systm.h>
46 #include <sys/syslog.h>
47 #include <sys/malloc.h>
48 #include <sys/mbuf.h>
49 #include <sys/socket.h>
50 #include <sys/sysctl.h>
51 #include <sys/syslog.h>
52 #include <sys/sysproto.h>
53 #include <sys/proc.h>
54 #include <sys/domain.h>
55 #include <sys/kernel.h>
56 
57 #include <net/if.h>
58 #include <net/if_var.h>
59 #include <net/if_dl.h>
60 #include <net/route.h>
61 #include <net/vnet.h>
62 #include <net/flowtable.h>
63 
64 #ifdef RADIX_MPATH
65 #include <net/radix_mpath.h>
66 #endif
67 
68 #include <netinet/in.h>
69 #include <netinet/ip_mroute.h>
70 
71 #include <vm/uma.h>
72 
73 #define	RT_MAXFIBS	UINT16_MAX
74 
75 /* Kernel config default option. */
76 #ifdef ROUTETABLES
77 #if ROUTETABLES <= 0
78 #error "ROUTETABLES defined too low"
79 #endif
80 #if ROUTETABLES > RT_MAXFIBS
81 #error "ROUTETABLES defined too big"
82 #endif
83 #define	RT_NUMFIBS	ROUTETABLES
84 #endif /* ROUTETABLES */
85 /* Initialize to default if not otherwise set. */
86 #ifndef	RT_NUMFIBS
87 #define	RT_NUMFIBS	1
88 #endif
89 
90 #if defined(INET) || defined(INET6)
91 #ifdef SCTP
92 extern void sctp_addr_change(struct ifaddr *ifa, int cmd);
93 #endif /* SCTP */
94 #endif
95 
96 
97 /* This is read-only.. */
98 u_int rt_numfibs = RT_NUMFIBS;
99 SYSCTL_UINT(_net, OID_AUTO, fibs, CTLFLAG_RD, &rt_numfibs, 0, "");
100 /* and this can be set too big but will be fixed before it is used */
101 TUNABLE_INT("net.fibs", &rt_numfibs);
102 
103 /*
104  * By default add routes to all fibs for new interfaces.
105  * Once this is set to 0 then only allocate routes on interface
106  * changes for the FIB of the caller when adding a new set of addresses
107  * to an interface.  XXX this is a shotgun aproach to a problem that needs
108  * a more fine grained solution.. that will come.
109  * XXX also has the problems getting the FIB from curthread which will not
110  * always work given the fib can be overridden and prefixes can be added
111  * from the network stack context.
112  */
113 u_int rt_add_addr_allfibs = 1;
114 SYSCTL_UINT(_net, OID_AUTO, add_addr_allfibs, CTLFLAG_RW,
115     &rt_add_addr_allfibs, 0, "");
116 TUNABLE_INT("net.add_addr_allfibs", &rt_add_addr_allfibs);
117 
118 VNET_DEFINE(struct rtstat, rtstat);
119 #define	V_rtstat	VNET(rtstat)
120 
121 VNET_DEFINE(struct radix_node_head *, rt_tables);
122 #define	V_rt_tables	VNET(rt_tables)
123 
124 VNET_DEFINE(int, rttrash);		/* routes not in table but not freed */
125 #define	V_rttrash	VNET(rttrash)
126 
127 
128 /* compare two sockaddr structures */
129 #define	sa_equal(a1, a2) (((a1)->sa_len == (a2)->sa_len) && \
130     (bcmp((a1), (a2), (a1)->sa_len) == 0))
131 
132 /*
133  * Convert a 'struct radix_node *' to a 'struct rtentry *'.
134  * The operation can be done safely (in this code) because a
135  * 'struct rtentry' starts with two 'struct radix_node''s, the first
136  * one representing leaf nodes in the routing tree, which is
137  * what the code in radix.c passes us as a 'struct radix_node'.
138  *
139  * But because there are a lot of assumptions in this conversion,
140  * do not cast explicitly, but always use the macro below.
141  */
142 #define RNTORT(p)	((struct rtentry *)(p))
143 
144 static VNET_DEFINE(uma_zone_t, rtzone);		/* Routing table UMA zone. */
145 #define	V_rtzone	VNET(rtzone)
146 
147 /*
148  * handler for net.my_fibnum
149  */
150 static int
151 sysctl_my_fibnum(SYSCTL_HANDLER_ARGS)
152 {
153         int fibnum;
154         int error;
155 
156         fibnum = curthread->td_proc->p_fibnum;
157         error = sysctl_handle_int(oidp, &fibnum, 0, req);
158         return (error);
159 }
160 
161 SYSCTL_PROC(_net, OID_AUTO, my_fibnum, CTLTYPE_INT|CTLFLAG_RD,
162             NULL, 0, &sysctl_my_fibnum, "I", "default FIB of caller");
163 
164 static __inline struct radix_node_head **
165 rt_tables_get_rnh_ptr(int table, int fam)
166 {
167 	struct radix_node_head **rnh;
168 
169 	KASSERT(table >= 0 && table < rt_numfibs, ("%s: table out of bounds.",
170 	    __func__));
171 	KASSERT(fam >= 0 && fam < (AF_MAX+1), ("%s: fam out of bounds.",
172 	    __func__));
173 
174 	/* rnh is [fib=0][af=0]. */
175 	rnh = (struct radix_node_head **)V_rt_tables;
176 	/* Get the offset to the requested table and fam. */
177 	rnh += table * (AF_MAX+1) + fam;
178 
179 	return (rnh);
180 }
181 
182 struct radix_node_head *
183 rt_tables_get_rnh(int table, int fam)
184 {
185 
186 	return (*rt_tables_get_rnh_ptr(table, fam));
187 }
188 
189 /*
190  * route initialization must occur before ip6_init2(), which happenas at
191  * SI_ORDER_MIDDLE.
192  */
193 static void
194 route_init(void)
195 {
196 
197 	/* whack the tunable ints into  line. */
198 	if (rt_numfibs > RT_MAXFIBS)
199 		rt_numfibs = RT_MAXFIBS;
200 	if (rt_numfibs == 0)
201 		rt_numfibs = 1;
202 }
203 SYSINIT(route_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_THIRD, route_init, 0);
204 
205 static void
206 vnet_route_init(const void *unused __unused)
207 {
208 	struct domain *dom;
209 	struct radix_node_head **rnh;
210 	int table;
211 	int fam;
212 
213 	V_rt_tables = malloc(rt_numfibs * (AF_MAX+1) *
214 	    sizeof(struct radix_node_head *), M_RTABLE, M_WAITOK|M_ZERO);
215 
216 	V_rtzone = uma_zcreate("rtentry", sizeof(struct rtentry), NULL, NULL,
217 	    NULL, NULL, UMA_ALIGN_PTR, 0);
218 	for (dom = domains; dom; dom = dom->dom_next) {
219 		if (dom->dom_rtattach == NULL)
220 			continue;
221 
222 		for  (table = 0; table < rt_numfibs; table++) {
223 			fam = dom->dom_family;
224 			if (table != 0 && fam != AF_INET6 && fam != AF_INET)
225 				break;
226 
227 			/*
228 			 * XXX MRT rtattach will be also called from
229 			 * vfs_export.c but the offset will be 0 (only for
230 			 * AF_INET and AF_INET6 which don't need it anyhow).
231 			 */
232 			rnh = rt_tables_get_rnh_ptr(table, fam);
233 			if (rnh == NULL)
234 				panic("%s: rnh NULL", __func__);
235 			dom->dom_rtattach((void **)rnh, dom->dom_rtoffset);
236 		}
237 	}
238 }
239 VNET_SYSINIT(vnet_route_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_FOURTH,
240     vnet_route_init, 0);
241 
242 #ifdef VIMAGE
243 static void
244 vnet_route_uninit(const void *unused __unused)
245 {
246 	int table;
247 	int fam;
248 	struct domain *dom;
249 	struct radix_node_head **rnh;
250 
251 	for (dom = domains; dom; dom = dom->dom_next) {
252 		if (dom->dom_rtdetach == NULL)
253 			continue;
254 
255 		for (table = 0; table < rt_numfibs; table++) {
256 			fam = dom->dom_family;
257 
258 			if (table != 0 && fam != AF_INET6 && fam != AF_INET)
259 				break;
260 
261 			rnh = rt_tables_get_rnh_ptr(table, fam);
262 			if (rnh == NULL)
263 				panic("%s: rnh NULL", __func__);
264 			dom->dom_rtdetach((void **)rnh, dom->dom_rtoffset);
265 		}
266 	}
267 
268 	free(V_rt_tables, M_RTABLE);
269 	uma_zdestroy(V_rtzone);
270 }
271 VNET_SYSUNINIT(vnet_route_uninit, SI_SUB_PROTO_DOMAIN, SI_ORDER_THIRD,
272     vnet_route_uninit, 0);
273 #endif
274 
275 #ifndef _SYS_SYSPROTO_H_
276 struct setfib_args {
277 	int     fibnum;
278 };
279 #endif
280 int
281 sys_setfib(struct thread *td, struct setfib_args *uap)
282 {
283 	if (uap->fibnum < 0 || uap->fibnum >= rt_numfibs)
284 		return EINVAL;
285 	td->td_proc->p_fibnum = uap->fibnum;
286 	return (0);
287 }
288 
289 /*
290  * Packet routing routines.
291  */
292 void
293 rtalloc(struct route *ro)
294 {
295 
296 	rtalloc_ign_fib(ro, 0UL, RT_DEFAULT_FIB);
297 }
298 
299 void
300 rtalloc_fib(struct route *ro, u_int fibnum)
301 {
302 	rtalloc_ign_fib(ro, 0UL, fibnum);
303 }
304 
305 void
306 rtalloc_ign(struct route *ro, u_long ignore)
307 {
308 	struct rtentry *rt;
309 
310 	if ((rt = ro->ro_rt) != NULL) {
311 		if (rt->rt_ifp != NULL && rt->rt_flags & RTF_UP)
312 			return;
313 		RTFREE(rt);
314 		ro->ro_rt = NULL;
315 	}
316 	ro->ro_rt = rtalloc1_fib(&ro->ro_dst, 1, ignore, RT_DEFAULT_FIB);
317 	if (ro->ro_rt)
318 		RT_UNLOCK(ro->ro_rt);
319 }
320 
321 void
322 rtalloc_ign_fib(struct route *ro, u_long ignore, u_int fibnum)
323 {
324 	struct rtentry *rt;
325 
326 	if ((rt = ro->ro_rt) != NULL) {
327 		if (rt->rt_ifp != NULL && rt->rt_flags & RTF_UP)
328 			return;
329 		RTFREE(rt);
330 		ro->ro_rt = NULL;
331 	}
332 	ro->ro_rt = rtalloc1_fib(&ro->ro_dst, 1, ignore, fibnum);
333 	if (ro->ro_rt)
334 		RT_UNLOCK(ro->ro_rt);
335 }
336 
337 /*
338  * Look up the route that matches the address given
339  * Or, at least try.. Create a cloned route if needed.
340  *
341  * The returned route, if any, is locked.
342  */
343 struct rtentry *
344 rtalloc1(struct sockaddr *dst, int report, u_long ignflags)
345 {
346 
347 	return (rtalloc1_fib(dst, report, ignflags, RT_DEFAULT_FIB));
348 }
349 
350 struct rtentry *
351 rtalloc1_fib(struct sockaddr *dst, int report, u_long ignflags,
352 		    u_int fibnum)
353 {
354 	struct radix_node_head *rnh;
355 	struct radix_node *rn;
356 	struct rtentry *newrt;
357 	struct rt_addrinfo info;
358 	int err = 0, msgtype = RTM_MISS;
359 	int needlock;
360 
361 	KASSERT((fibnum < rt_numfibs), ("rtalloc1_fib: bad fibnum"));
362 	switch (dst->sa_family) {
363 	case AF_INET6:
364 	case AF_INET:
365 		/* We support multiple FIBs. */
366 		break;
367 	default:
368 		fibnum = RT_DEFAULT_FIB;
369 		break;
370 	}
371 	rnh = rt_tables_get_rnh(fibnum, dst->sa_family);
372 	newrt = NULL;
373 	if (rnh == NULL)
374 		goto miss;
375 
376 	/*
377 	 * Look up the address in the table for that Address Family
378 	 */
379 	needlock = !(ignflags & RTF_RNH_LOCKED);
380 	if (needlock)
381 		RADIX_NODE_HEAD_RLOCK(rnh);
382 #ifdef INVARIANTS
383 	else
384 		RADIX_NODE_HEAD_LOCK_ASSERT(rnh);
385 #endif
386 	rn = rnh->rnh_matchaddr(dst, rnh);
387 	if (rn && ((rn->rn_flags & RNF_ROOT) == 0)) {
388 		newrt = RNTORT(rn);
389 		RT_LOCK(newrt);
390 		RT_ADDREF(newrt);
391 		if (needlock)
392 			RADIX_NODE_HEAD_RUNLOCK(rnh);
393 		goto done;
394 
395 	} else if (needlock)
396 		RADIX_NODE_HEAD_RUNLOCK(rnh);
397 
398 	/*
399 	 * Either we hit the root or couldn't find any match,
400 	 * Which basically means
401 	 * "caint get there frm here"
402 	 */
403 miss:
404 	V_rtstat.rts_unreach++;
405 
406 	if (report) {
407 		/*
408 		 * If required, report the failure to the supervising
409 		 * Authorities.
410 		 * For a delete, this is not an error. (report == 0)
411 		 */
412 		bzero(&info, sizeof(info));
413 		info.rti_info[RTAX_DST] = dst;
414 		rt_missmsg_fib(msgtype, &info, 0, err, fibnum);
415 	}
416 done:
417 	if (newrt)
418 		RT_LOCK_ASSERT(newrt);
419 	return (newrt);
420 }
421 
422 /*
423  * Remove a reference count from an rtentry.
424  * If the count gets low enough, take it out of the routing table
425  */
426 void
427 rtfree(struct rtentry *rt)
428 {
429 	struct radix_node_head *rnh;
430 
431 	KASSERT(rt != NULL,("%s: NULL rt", __func__));
432 	rnh = rt_tables_get_rnh(rt->rt_fibnum, rt_key(rt)->sa_family);
433 	KASSERT(rnh != NULL,("%s: NULL rnh", __func__));
434 
435 	RT_LOCK_ASSERT(rt);
436 
437 	/*
438 	 * The callers should use RTFREE_LOCKED() or RTFREE(), so
439 	 * we should come here exactly with the last reference.
440 	 */
441 	RT_REMREF(rt);
442 	if (rt->rt_refcnt > 0) {
443 		log(LOG_DEBUG, "%s: %p has %d refs\n", __func__, rt, rt->rt_refcnt);
444 		goto done;
445 	}
446 
447 	/*
448 	 * On last reference give the "close method" a chance
449 	 * to cleanup private state.  This also permits (for
450 	 * IPv4 and IPv6) a chance to decide if the routing table
451 	 * entry should be purged immediately or at a later time.
452 	 * When an immediate purge is to happen the close routine
453 	 * typically calls rtexpunge which clears the RTF_UP flag
454 	 * on the entry so that the code below reclaims the storage.
455 	 */
456 	if (rt->rt_refcnt == 0 && rnh->rnh_close)
457 		rnh->rnh_close((struct radix_node *)rt, rnh);
458 
459 	/*
460 	 * If we are no longer "up" (and ref == 0)
461 	 * then we can free the resources associated
462 	 * with the route.
463 	 */
464 	if ((rt->rt_flags & RTF_UP) == 0) {
465 		if (rt->rt_nodes->rn_flags & (RNF_ACTIVE | RNF_ROOT))
466 			panic("rtfree 2");
467 		/*
468 		 * the rtentry must have been removed from the routing table
469 		 * so it is represented in rttrash.. remove that now.
470 		 */
471 		V_rttrash--;
472 #ifdef	DIAGNOSTIC
473 		if (rt->rt_refcnt < 0) {
474 			printf("rtfree: %p not freed (neg refs)\n", rt);
475 			goto done;
476 		}
477 #endif
478 		/*
479 		 * release references on items we hold them on..
480 		 * e.g other routes and ifaddrs.
481 		 */
482 		if (rt->rt_ifa)
483 			ifa_free(rt->rt_ifa);
484 		/*
485 		 * The key is separatly alloc'd so free it (see rt_setgate()).
486 		 * This also frees the gateway, as they are always malloc'd
487 		 * together.
488 		 */
489 		Free(rt_key(rt));
490 
491 		/*
492 		 * and the rtentry itself of course
493 		 */
494 		RT_LOCK_DESTROY(rt);
495 		uma_zfree(V_rtzone, rt);
496 		return;
497 	}
498 done:
499 	RT_UNLOCK(rt);
500 }
501 
502 
503 /*
504  * Force a routing table entry to the specified
505  * destination to go through the given gateway.
506  * Normally called as a result of a routing redirect
507  * message from the network layer.
508  */
509 void
510 rtredirect(struct sockaddr *dst,
511 	struct sockaddr *gateway,
512 	struct sockaddr *netmask,
513 	int flags,
514 	struct sockaddr *src)
515 {
516 
517 	rtredirect_fib(dst, gateway, netmask, flags, src, RT_DEFAULT_FIB);
518 }
519 
520 void
521 rtredirect_fib(struct sockaddr *dst,
522 	struct sockaddr *gateway,
523 	struct sockaddr *netmask,
524 	int flags,
525 	struct sockaddr *src,
526 	u_int fibnum)
527 {
528 	struct rtentry *rt, *rt0 = NULL;
529 	int error = 0;
530 	short *stat = NULL;
531 	struct rt_addrinfo info;
532 	struct ifaddr *ifa;
533 	struct radix_node_head *rnh;
534 
535 	ifa = NULL;
536 	rnh = rt_tables_get_rnh(fibnum, dst->sa_family);
537 	if (rnh == NULL) {
538 		error = EAFNOSUPPORT;
539 		goto out;
540 	}
541 
542 	/* verify the gateway is directly reachable */
543 	if ((ifa = ifa_ifwithnet(gateway, 0)) == NULL) {
544 		error = ENETUNREACH;
545 		goto out;
546 	}
547 	rt = rtalloc1_fib(dst, 0, 0UL, fibnum);	/* NB: rt is locked */
548 	/*
549 	 * If the redirect isn't from our current router for this dst,
550 	 * it's either old or wrong.  If it redirects us to ourselves,
551 	 * we have a routing loop, perhaps as a result of an interface
552 	 * going down recently.
553 	 */
554 	if (!(flags & RTF_DONE) && rt &&
555 	     (!sa_equal(src, rt->rt_gateway) || rt->rt_ifa != ifa))
556 		error = EINVAL;
557 	else if (ifa_ifwithaddr_check(gateway))
558 		error = EHOSTUNREACH;
559 	if (error)
560 		goto done;
561 	/*
562 	 * Create a new entry if we just got back a wildcard entry
563 	 * or the lookup failed.  This is necessary for hosts
564 	 * which use routing redirects generated by smart gateways
565 	 * to dynamically build the routing tables.
566 	 */
567 	if (rt == NULL || (rt_mask(rt) && rt_mask(rt)->sa_len < 2))
568 		goto create;
569 	/*
570 	 * Don't listen to the redirect if it's
571 	 * for a route to an interface.
572 	 */
573 	if (rt->rt_flags & RTF_GATEWAY) {
574 		if (((rt->rt_flags & RTF_HOST) == 0) && (flags & RTF_HOST)) {
575 			/*
576 			 * Changing from route to net => route to host.
577 			 * Create new route, rather than smashing route to net.
578 			 */
579 		create:
580 			rt0 = rt;
581 			rt = NULL;
582 
583 			flags |=  RTF_GATEWAY | RTF_DYNAMIC;
584 			bzero((caddr_t)&info, sizeof(info));
585 			info.rti_info[RTAX_DST] = dst;
586 			info.rti_info[RTAX_GATEWAY] = gateway;
587 			info.rti_info[RTAX_NETMASK] = netmask;
588 			info.rti_ifa = ifa;
589 			info.rti_flags = flags;
590 			if (rt0 != NULL)
591 				RT_UNLOCK(rt0);	/* drop lock to avoid LOR with RNH */
592 			error = rtrequest1_fib(RTM_ADD, &info, &rt, fibnum);
593 			if (rt != NULL) {
594 				RT_LOCK(rt);
595 				if (rt0 != NULL)
596 					EVENTHANDLER_INVOKE(route_redirect_event, rt0, rt, dst);
597 				flags = rt->rt_flags;
598 			}
599 			if (rt0 != NULL)
600 				RTFREE(rt0);
601 
602 			stat = &V_rtstat.rts_dynamic;
603 		} else {
604 			struct rtentry *gwrt;
605 
606 			/*
607 			 * Smash the current notion of the gateway to
608 			 * this destination.  Should check about netmask!!!
609 			 */
610 			rt->rt_flags |= RTF_MODIFIED;
611 			flags |= RTF_MODIFIED;
612 			stat = &V_rtstat.rts_newgateway;
613 			/*
614 			 * add the key and gateway (in one malloc'd chunk).
615 			 */
616 			RT_UNLOCK(rt);
617 			RADIX_NODE_HEAD_LOCK(rnh);
618 			RT_LOCK(rt);
619 			rt_setgate(rt, rt_key(rt), gateway);
620 			gwrt = rtalloc1(gateway, 1, RTF_RNH_LOCKED);
621 			RADIX_NODE_HEAD_UNLOCK(rnh);
622 			EVENTHANDLER_INVOKE(route_redirect_event, rt, gwrt, dst);
623 			RTFREE_LOCKED(gwrt);
624 		}
625 	} else
626 		error = EHOSTUNREACH;
627 done:
628 	if (rt)
629 		RTFREE_LOCKED(rt);
630 out:
631 	if (error)
632 		V_rtstat.rts_badredirect++;
633 	else if (stat != NULL)
634 		(*stat)++;
635 	bzero((caddr_t)&info, sizeof(info));
636 	info.rti_info[RTAX_DST] = dst;
637 	info.rti_info[RTAX_GATEWAY] = gateway;
638 	info.rti_info[RTAX_NETMASK] = netmask;
639 	info.rti_info[RTAX_AUTHOR] = src;
640 	rt_missmsg_fib(RTM_REDIRECT, &info, flags, error, fibnum);
641 	if (ifa != NULL)
642 		ifa_free(ifa);
643 }
644 
645 int
646 rtioctl(u_long req, caddr_t data)
647 {
648 
649 	return (rtioctl_fib(req, data, RT_DEFAULT_FIB));
650 }
651 
652 /*
653  * Routing table ioctl interface.
654  */
655 int
656 rtioctl_fib(u_long req, caddr_t data, u_int fibnum)
657 {
658 
659 	/*
660 	 * If more ioctl commands are added here, make sure the proper
661 	 * super-user checks are being performed because it is possible for
662 	 * prison-root to make it this far if raw sockets have been enabled
663 	 * in jails.
664 	 */
665 #ifdef INET
666 	/* Multicast goop, grrr... */
667 	return mrt_ioctl ? mrt_ioctl(req, data, fibnum) : EOPNOTSUPP;
668 #else /* INET */
669 	return ENXIO;
670 #endif /* INET */
671 }
672 
673 /*
674  * For both ifa_ifwithroute() routines, 'ifa' is returned referenced.
675  */
676 struct ifaddr *
677 ifa_ifwithroute(int flags, struct sockaddr *dst, struct sockaddr *gateway)
678 {
679 
680 	return (ifa_ifwithroute_fib(flags, dst, gateway, RT_DEFAULT_FIB));
681 }
682 
683 struct ifaddr *
684 ifa_ifwithroute_fib(int flags, struct sockaddr *dst, struct sockaddr *gateway,
685 				u_int fibnum)
686 {
687 	register struct ifaddr *ifa;
688 	int not_found = 0;
689 
690 	if ((flags & RTF_GATEWAY) == 0) {
691 		/*
692 		 * If we are adding a route to an interface,
693 		 * and the interface is a pt to pt link
694 		 * we should search for the destination
695 		 * as our clue to the interface.  Otherwise
696 		 * we can use the local address.
697 		 */
698 		ifa = NULL;
699 		if (flags & RTF_HOST)
700 			ifa = ifa_ifwithdstaddr(dst);
701 		if (ifa == NULL)
702 			ifa = ifa_ifwithaddr(gateway);
703 	} else {
704 		/*
705 		 * If we are adding a route to a remote net
706 		 * or host, the gateway may still be on the
707 		 * other end of a pt to pt link.
708 		 */
709 		ifa = ifa_ifwithdstaddr(gateway);
710 	}
711 	if (ifa == NULL)
712 		ifa = ifa_ifwithnet(gateway, 0);
713 	if (ifa == NULL) {
714 		struct rtentry *rt = rtalloc1_fib(gateway, 0, RTF_RNH_LOCKED, fibnum);
715 		if (rt == NULL)
716 			return (NULL);
717 		/*
718 		 * dismiss a gateway that is reachable only
719 		 * through the default router
720 		 */
721 		switch (gateway->sa_family) {
722 		case AF_INET:
723 			if (satosin(rt_key(rt))->sin_addr.s_addr == INADDR_ANY)
724 				not_found = 1;
725 			break;
726 		case AF_INET6:
727 			if (IN6_IS_ADDR_UNSPECIFIED(&satosin6(rt_key(rt))->sin6_addr))
728 				not_found = 1;
729 			break;
730 		default:
731 			break;
732 		}
733 		if (!not_found && rt->rt_ifa != NULL) {
734 			ifa = rt->rt_ifa;
735 			ifa_ref(ifa);
736 		}
737 		RT_REMREF(rt);
738 		RT_UNLOCK(rt);
739 		if (not_found || ifa == NULL)
740 			return (NULL);
741 	}
742 	if (ifa->ifa_addr->sa_family != dst->sa_family) {
743 		struct ifaddr *oifa = ifa;
744 		ifa = ifaof_ifpforaddr(dst, ifa->ifa_ifp);
745 		if (ifa == NULL)
746 			ifa = oifa;
747 		else
748 			ifa_free(oifa);
749 	}
750 	return (ifa);
751 }
752 
753 /*
754  * Do appropriate manipulations of a routing tree given
755  * all the bits of info needed
756  */
757 int
758 rtrequest(int req,
759 	struct sockaddr *dst,
760 	struct sockaddr *gateway,
761 	struct sockaddr *netmask,
762 	int flags,
763 	struct rtentry **ret_nrt)
764 {
765 
766 	return (rtrequest_fib(req, dst, gateway, netmask, flags, ret_nrt,
767 	    RT_DEFAULT_FIB));
768 }
769 
770 int
771 rtrequest_fib(int req,
772 	struct sockaddr *dst,
773 	struct sockaddr *gateway,
774 	struct sockaddr *netmask,
775 	int flags,
776 	struct rtentry **ret_nrt,
777 	u_int fibnum)
778 {
779 	struct rt_addrinfo info;
780 
781 	if (dst->sa_len == 0)
782 		return(EINVAL);
783 
784 	bzero((caddr_t)&info, sizeof(info));
785 	info.rti_flags = flags;
786 	info.rti_info[RTAX_DST] = dst;
787 	info.rti_info[RTAX_GATEWAY] = gateway;
788 	info.rti_info[RTAX_NETMASK] = netmask;
789 	return rtrequest1_fib(req, &info, ret_nrt, fibnum);
790 }
791 
792 /*
793  * These (questionable) definitions of apparent local variables apply
794  * to the next two functions.  XXXXXX!!!
795  */
796 #define	dst	info->rti_info[RTAX_DST]
797 #define	gateway	info->rti_info[RTAX_GATEWAY]
798 #define	netmask	info->rti_info[RTAX_NETMASK]
799 #define	ifaaddr	info->rti_info[RTAX_IFA]
800 #define	ifpaddr	info->rti_info[RTAX_IFP]
801 #define	flags	info->rti_flags
802 
803 int
804 rt_getifa(struct rt_addrinfo *info)
805 {
806 
807 	return (rt_getifa_fib(info, RT_DEFAULT_FIB));
808 }
809 
810 /*
811  * Look up rt_addrinfo for a specific fib.  Note that if rti_ifa is defined,
812  * it will be referenced so the caller must free it.
813  */
814 int
815 rt_getifa_fib(struct rt_addrinfo *info, u_int fibnum)
816 {
817 	struct ifaddr *ifa;
818 	int error = 0;
819 
820 	/*
821 	 * ifp may be specified by sockaddr_dl
822 	 * when protocol address is ambiguous.
823 	 */
824 	if (info->rti_ifp == NULL && ifpaddr != NULL &&
825 	    ifpaddr->sa_family == AF_LINK &&
826 	    (ifa = ifa_ifwithnet(ifpaddr, 0)) != NULL) {
827 		info->rti_ifp = ifa->ifa_ifp;
828 		ifa_free(ifa);
829 	}
830 	if (info->rti_ifa == NULL && ifaaddr != NULL)
831 		info->rti_ifa = ifa_ifwithaddr(ifaaddr);
832 	if (info->rti_ifa == NULL) {
833 		struct sockaddr *sa;
834 
835 		sa = ifaaddr != NULL ? ifaaddr :
836 		    (gateway != NULL ? gateway : dst);
837 		if (sa != NULL && info->rti_ifp != NULL)
838 			info->rti_ifa = ifaof_ifpforaddr(sa, info->rti_ifp);
839 		else if (dst != NULL && gateway != NULL)
840 			info->rti_ifa = ifa_ifwithroute_fib(flags, dst, gateway,
841 							fibnum);
842 		else if (sa != NULL)
843 			info->rti_ifa = ifa_ifwithroute_fib(flags, sa, sa,
844 							fibnum);
845 	}
846 	if ((ifa = info->rti_ifa) != NULL) {
847 		if (info->rti_ifp == NULL)
848 			info->rti_ifp = ifa->ifa_ifp;
849 	} else
850 		error = ENETUNREACH;
851 	return (error);
852 }
853 
854 /*
855  * Expunges references to a route that's about to be reclaimed.
856  * The route must be locked.
857  */
858 int
859 rtexpunge(struct rtentry *rt)
860 {
861 #if !defined(RADIX_MPATH)
862 	struct radix_node *rn;
863 #else
864 	struct rt_addrinfo info;
865 	int fib;
866 	struct rtentry *rt0;
867 #endif
868 	struct radix_node_head *rnh;
869 	struct ifaddr *ifa;
870 	int error = 0;
871 
872 	/*
873 	 * Find the correct routing tree to use for this Address Family
874 	 */
875 	rnh = rt_tables_get_rnh(rt->rt_fibnum, rt_key(rt)->sa_family);
876 	RT_LOCK_ASSERT(rt);
877 	if (rnh == NULL)
878 		return (EAFNOSUPPORT);
879 	RADIX_NODE_HEAD_LOCK_ASSERT(rnh);
880 
881 #ifdef RADIX_MPATH
882 	fib = rt->rt_fibnum;
883 	bzero(&info, sizeof(info));
884 	info.rti_ifp = rt->rt_ifp;
885 	info.rti_flags = RTF_RNH_LOCKED;
886 	info.rti_info[RTAX_DST] = rt_key(rt);
887 	info.rti_info[RTAX_GATEWAY] = rt->rt_ifa->ifa_addr;
888 
889 	RT_UNLOCK(rt);
890 	error = rtrequest1_fib(RTM_DELETE, &info, &rt0, fib);
891 
892 	if (error == 0 && rt0 != NULL) {
893 		rt = rt0;
894 		RT_LOCK(rt);
895 	} else if (error != 0) {
896 		RT_LOCK(rt);
897 		return (error);
898 	}
899 #else
900 	/*
901 	 * Remove the item from the tree; it should be there,
902 	 * but when callers invoke us blindly it may not (sigh).
903 	 */
904 	rn = rnh->rnh_deladdr(rt_key(rt), rt_mask(rt), rnh);
905 	if (rn == NULL) {
906 		error = ESRCH;
907 		goto bad;
908 	}
909 	KASSERT((rn->rn_flags & (RNF_ACTIVE | RNF_ROOT)) == 0,
910 		("unexpected flags 0x%x", rn->rn_flags));
911 	KASSERT(rt == RNTORT(rn),
912 		("lookup mismatch, rt %p rn %p", rt, rn));
913 #endif /* RADIX_MPATH */
914 
915 	rt->rt_flags &= ~RTF_UP;
916 
917 	/*
918 	 * Give the protocol a chance to keep things in sync.
919 	 */
920 	if ((ifa = rt->rt_ifa) && ifa->ifa_rtrequest) {
921 		struct rt_addrinfo info;
922 
923 		bzero((caddr_t)&info, sizeof(info));
924 		info.rti_flags = rt->rt_flags;
925 		info.rti_info[RTAX_DST] = rt_key(rt);
926 		info.rti_info[RTAX_GATEWAY] = rt->rt_gateway;
927 		info.rti_info[RTAX_NETMASK] = rt_mask(rt);
928 		ifa->ifa_rtrequest(RTM_DELETE, rt, &info);
929 	}
930 
931 	/*
932 	 * one more rtentry floating around that is not
933 	 * linked to the routing table.
934 	 */
935 	V_rttrash++;
936 #if !defined(RADIX_MPATH)
937 bad:
938 #endif
939 	return (error);
940 }
941 
942 #if 0
943 int p_sockaddr(char *buf, int buflen, struct sockaddr *s);
944 int rt_print(char *buf, int buflen, struct rtentry *rt);
945 
946 int
947 p_sockaddr(char *buf, int buflen, struct sockaddr *s)
948 {
949 	void *paddr = NULL;
950 
951 	switch (s->sa_family) {
952 	case AF_INET:
953 		paddr = &((struct sockaddr_in *)s)->sin_addr;
954 		break;
955 	case AF_INET6:
956 		paddr = &((struct sockaddr_in6 *)s)->sin6_addr;
957 		break;
958 	}
959 
960 	if (paddr == NULL)
961 		return (0);
962 
963 	if (inet_ntop(s->sa_family, paddr, buf, buflen) == NULL)
964 		return (0);
965 
966 	return (strlen(buf));
967 }
968 
969 int
970 rt_print(char *buf, int buflen, struct rtentry *rt)
971 {
972 	struct sockaddr *addr, *mask;
973 	int i = 0;
974 
975 	addr = rt_key(rt);
976 	mask = rt_mask(rt);
977 
978 	i = p_sockaddr(buf, buflen, addr);
979 	if (!(rt->rt_flags & RTF_HOST)) {
980 		buf[i++] = '/';
981 		i += p_sockaddr(buf + i, buflen - i, mask);
982 	}
983 
984 	if (rt->rt_flags & RTF_GATEWAY) {
985 		buf[i++] = '>';
986 		i += p_sockaddr(buf + i, buflen - i, rt->rt_gateway);
987 	}
988 
989 	return (i);
990 }
991 #endif
992 
993 #ifdef RADIX_MPATH
994 static int
995 rn_mpath_update(int req, struct rt_addrinfo *info,
996     struct radix_node_head *rnh, struct rtentry **ret_nrt)
997 {
998 	/*
999 	 * if we got multipath routes, we require users to specify
1000 	 * a matching RTAX_GATEWAY.
1001 	 */
1002 	struct rtentry *rt, *rto = NULL;
1003 	register struct radix_node *rn;
1004 	int error = 0;
1005 
1006 	rn = rnh->rnh_lookup(dst, netmask, rnh);
1007 	if (rn == NULL)
1008 		return (ESRCH);
1009 	rto = rt = RNTORT(rn);
1010 
1011 	rt = rt_mpath_matchgate(rt, gateway);
1012 	if (rt == NULL)
1013 		return (ESRCH);
1014 	/*
1015 	 * this is the first entry in the chain
1016 	 */
1017 	if (rto == rt) {
1018 		rn = rn_mpath_next((struct radix_node *)rt);
1019 		/*
1020 		 * there is another entry, now it's active
1021 		 */
1022 		if (rn) {
1023 			rto = RNTORT(rn);
1024 			RT_LOCK(rto);
1025 			rto->rt_flags |= RTF_UP;
1026 			RT_UNLOCK(rto);
1027 		} else if (rt->rt_flags & RTF_GATEWAY) {
1028 			/*
1029 			 * For gateway routes, we need to
1030 			 * make sure that we we are deleting
1031 			 * the correct gateway.
1032 			 * rt_mpath_matchgate() does not
1033 			 * check the case when there is only
1034 			 * one route in the chain.
1035 			 */
1036 			if (gateway &&
1037 			    (rt->rt_gateway->sa_len != gateway->sa_len ||
1038 				memcmp(rt->rt_gateway, gateway, gateway->sa_len)))
1039 				error = ESRCH;
1040 			else {
1041 				/*
1042 				 * remove from tree before returning it
1043 				 * to the caller
1044 				 */
1045 				rn = rnh->rnh_deladdr(dst, netmask, rnh);
1046 				KASSERT(rt == RNTORT(rn), ("radix node disappeared"));
1047 				goto gwdelete;
1048 			}
1049 
1050 		}
1051 		/*
1052 		 * use the normal delete code to remove
1053 		 * the first entry
1054 		 */
1055 		if (req != RTM_DELETE)
1056 			goto nondelete;
1057 
1058 		error = ENOENT;
1059 		goto done;
1060 	}
1061 
1062 	/*
1063 	 * if the entry is 2nd and on up
1064 	 */
1065 	if ((req == RTM_DELETE) && !rt_mpath_deldup(rto, rt))
1066 		panic ("rtrequest1: rt_mpath_deldup");
1067 gwdelete:
1068 	RT_LOCK(rt);
1069 	RT_ADDREF(rt);
1070 	if (req == RTM_DELETE) {
1071 		rt->rt_flags &= ~RTF_UP;
1072 		/*
1073 		 * One more rtentry floating around that is not
1074 		 * linked to the routing table. rttrash will be decremented
1075 		 * when RTFREE(rt) is eventually called.
1076 		 */
1077 		V_rttrash++;
1078 	}
1079 
1080 nondelete:
1081 	if (req != RTM_DELETE)
1082 		panic("unrecognized request %d", req);
1083 
1084 
1085 	/*
1086 	 * If the caller wants it, then it can have it,
1087 	 * but it's up to it to free the rtentry as we won't be
1088 	 * doing it.
1089 	 */
1090 	if (ret_nrt) {
1091 		*ret_nrt = rt;
1092 		RT_UNLOCK(rt);
1093 	} else
1094 		RTFREE_LOCKED(rt);
1095 done:
1096 	return (error);
1097 }
1098 #endif
1099 
1100 int
1101 rtrequest1_fib(int req, struct rt_addrinfo *info, struct rtentry **ret_nrt,
1102 				u_int fibnum)
1103 {
1104 	int error = 0, needlock = 0;
1105 	register struct rtentry *rt;
1106 #ifdef FLOWTABLE
1107 	register struct rtentry *rt0;
1108 #endif
1109 	register struct radix_node *rn;
1110 	register struct radix_node_head *rnh;
1111 	struct ifaddr *ifa;
1112 	struct sockaddr *ndst;
1113 	struct sockaddr_storage mdst;
1114 #define senderr(x) { error = x ; goto bad; }
1115 
1116 	KASSERT((fibnum < rt_numfibs), ("rtrequest1_fib: bad fibnum"));
1117 	switch (dst->sa_family) {
1118 	case AF_INET6:
1119 	case AF_INET:
1120 		/* We support multiple FIBs. */
1121 		break;
1122 	default:
1123 		fibnum = RT_DEFAULT_FIB;
1124 		break;
1125 	}
1126 
1127 	/*
1128 	 * Find the correct routing tree to use for this Address Family
1129 	 */
1130 	rnh = rt_tables_get_rnh(fibnum, dst->sa_family);
1131 	if (rnh == NULL)
1132 		return (EAFNOSUPPORT);
1133 	needlock = ((flags & RTF_RNH_LOCKED) == 0);
1134 	flags &= ~RTF_RNH_LOCKED;
1135 	if (needlock)
1136 		RADIX_NODE_HEAD_LOCK(rnh);
1137 	else
1138 		RADIX_NODE_HEAD_LOCK_ASSERT(rnh);
1139 	/*
1140 	 * If we are adding a host route then we don't want to put
1141 	 * a netmask in the tree, nor do we want to clone it.
1142 	 */
1143 	if (flags & RTF_HOST)
1144 		netmask = NULL;
1145 
1146 	switch (req) {
1147 	case RTM_DELETE:
1148 		if (netmask) {
1149 			rt_maskedcopy(dst, (struct sockaddr *)&mdst, netmask);
1150 			dst = (struct sockaddr *)&mdst;
1151 		}
1152 #ifdef RADIX_MPATH
1153 		if (rn_mpath_capable(rnh)) {
1154 			error = rn_mpath_update(req, info, rnh, ret_nrt);
1155 			/*
1156 			 * "bad" holds true for the success case
1157 			 * as well
1158 			 */
1159 			if (error != ENOENT)
1160 				goto bad;
1161 			error = 0;
1162 		}
1163 #endif
1164 		if ((flags & RTF_PINNED) == 0) {
1165 			/* Check if target route can be deleted */
1166 			rt = (struct rtentry *)rnh->rnh_lookup(dst,
1167 			    netmask, rnh);
1168 			if ((rt != NULL) && (rt->rt_flags & RTF_PINNED))
1169 				senderr(EADDRINUSE);
1170 		}
1171 
1172 		/*
1173 		 * Remove the item from the tree and return it.
1174 		 * Complain if it is not there and do no more processing.
1175 		 */
1176 		rn = rnh->rnh_deladdr(dst, netmask, rnh);
1177 		if (rn == NULL)
1178 			senderr(ESRCH);
1179 		if (rn->rn_flags & (RNF_ACTIVE | RNF_ROOT))
1180 			panic ("rtrequest delete");
1181 		rt = RNTORT(rn);
1182 		RT_LOCK(rt);
1183 		RT_ADDREF(rt);
1184 		rt->rt_flags &= ~RTF_UP;
1185 
1186 		/*
1187 		 * give the protocol a chance to keep things in sync.
1188 		 */
1189 		if ((ifa = rt->rt_ifa) && ifa->ifa_rtrequest)
1190 			ifa->ifa_rtrequest(RTM_DELETE, rt, info);
1191 
1192 		/*
1193 		 * One more rtentry floating around that is not
1194 		 * linked to the routing table. rttrash will be decremented
1195 		 * when RTFREE(rt) is eventually called.
1196 		 */
1197 		V_rttrash++;
1198 
1199 		/*
1200 		 * If the caller wants it, then it can have it,
1201 		 * but it's up to it to free the rtentry as we won't be
1202 		 * doing it.
1203 		 */
1204 		if (ret_nrt) {
1205 			*ret_nrt = rt;
1206 			RT_UNLOCK(rt);
1207 		} else
1208 			RTFREE_LOCKED(rt);
1209 		break;
1210 	case RTM_RESOLVE:
1211 		/*
1212 		 * resolve was only used for route cloning
1213 		 * here for compat
1214 		 */
1215 		break;
1216 	case RTM_ADD:
1217 		if ((flags & RTF_GATEWAY) && !gateway)
1218 			senderr(EINVAL);
1219 		if (dst && gateway && (dst->sa_family != gateway->sa_family) &&
1220 		    (gateway->sa_family != AF_UNSPEC) && (gateway->sa_family != AF_LINK))
1221 			senderr(EINVAL);
1222 
1223 		if (info->rti_ifa == NULL) {
1224 			error = rt_getifa_fib(info, fibnum);
1225 			if (error)
1226 				senderr(error);
1227 		} else
1228 			ifa_ref(info->rti_ifa);
1229 		ifa = info->rti_ifa;
1230 		rt = uma_zalloc(V_rtzone, M_NOWAIT | M_ZERO);
1231 		if (rt == NULL) {
1232 			ifa_free(ifa);
1233 			senderr(ENOBUFS);
1234 		}
1235 		RT_LOCK_INIT(rt);
1236 		rt->rt_flags = RTF_UP | flags;
1237 		rt->rt_fibnum = fibnum;
1238 		/*
1239 		 * Add the gateway. Possibly re-malloc-ing the storage for it.
1240 		 */
1241 		RT_LOCK(rt);
1242 		if ((error = rt_setgate(rt, dst, gateway)) != 0) {
1243 			RT_LOCK_DESTROY(rt);
1244 			ifa_free(ifa);
1245 			uma_zfree(V_rtzone, rt);
1246 			senderr(error);
1247 		}
1248 
1249 		/*
1250 		 * point to the (possibly newly malloc'd) dest address.
1251 		 */
1252 		ndst = (struct sockaddr *)rt_key(rt);
1253 
1254 		/*
1255 		 * make sure it contains the value we want (masked if needed).
1256 		 */
1257 		if (netmask) {
1258 			rt_maskedcopy(dst, ndst, netmask);
1259 		} else
1260 			bcopy(dst, ndst, dst->sa_len);
1261 
1262 		/*
1263 		 * We use the ifa reference returned by rt_getifa_fib().
1264 		 * This moved from below so that rnh->rnh_addaddr() can
1265 		 * examine the ifa and  ifa->ifa_ifp if it so desires.
1266 		 */
1267 		rt->rt_ifa = ifa;
1268 		rt->rt_ifp = ifa->ifa_ifp;
1269 		rt->rt_rmx.rmx_weight = 1;
1270 
1271 #ifdef RADIX_MPATH
1272 		/* do not permit exactly the same dst/mask/gw pair */
1273 		if (rn_mpath_capable(rnh) &&
1274 			rt_mpath_conflict(rnh, rt, netmask)) {
1275 			ifa_free(rt->rt_ifa);
1276 			Free(rt_key(rt));
1277 			RT_LOCK_DESTROY(rt);
1278 			uma_zfree(V_rtzone, rt);
1279 			senderr(EEXIST);
1280 		}
1281 #endif
1282 
1283 #ifdef FLOWTABLE
1284 		rt0 = NULL;
1285 		/* "flow-table" only supports IPv6 and IPv4 at the moment. */
1286 		switch (dst->sa_family) {
1287 #ifdef INET6
1288 		case AF_INET6:
1289 #endif
1290 #ifdef INET
1291 		case AF_INET:
1292 #endif
1293 #if defined(INET6) || defined(INET)
1294 			rn = rnh->rnh_matchaddr(dst, rnh);
1295 			if (rn && ((rn->rn_flags & RNF_ROOT) == 0)) {
1296 				struct sockaddr *mask;
1297 				u_char *m, *n;
1298 				int len;
1299 
1300 				/*
1301 				 * compare mask to see if the new route is
1302 				 * more specific than the existing one
1303 				 */
1304 				rt0 = RNTORT(rn);
1305 				RT_LOCK(rt0);
1306 				RT_ADDREF(rt0);
1307 				RT_UNLOCK(rt0);
1308 				/*
1309 				 * A host route is already present, so
1310 				 * leave the flow-table entries as is.
1311 				 */
1312 				if (rt0->rt_flags & RTF_HOST) {
1313 					RTFREE(rt0);
1314 					rt0 = NULL;
1315 				} else if (!(flags & RTF_HOST) && netmask) {
1316 					mask = rt_mask(rt0);
1317 					len = mask->sa_len;
1318 					m = (u_char *)mask;
1319 					n = (u_char *)netmask;
1320 					while (len-- > 0) {
1321 						if (*n != *m)
1322 							break;
1323 						n++;
1324 						m++;
1325 					}
1326 					if (len == 0 || (*n < *m)) {
1327 						RTFREE(rt0);
1328 						rt0 = NULL;
1329 					}
1330 				}
1331 			}
1332 #endif/* INET6 || INET */
1333 		}
1334 #endif /* FLOWTABLE */
1335 
1336 		/* XXX mtu manipulation will be done in rnh_addaddr -- itojun */
1337 		rn = rnh->rnh_addaddr(ndst, netmask, rnh, rt->rt_nodes);
1338 		/*
1339 		 * If it still failed to go into the tree,
1340 		 * then un-make it (this should be a function)
1341 		 */
1342 		if (rn == NULL) {
1343 			ifa_free(rt->rt_ifa);
1344 			Free(rt_key(rt));
1345 			RT_LOCK_DESTROY(rt);
1346 			uma_zfree(V_rtzone, rt);
1347 #ifdef FLOWTABLE
1348 			if (rt0 != NULL)
1349 				RTFREE(rt0);
1350 #endif
1351 			senderr(EEXIST);
1352 		}
1353 #ifdef FLOWTABLE
1354 		else if (rt0 != NULL) {
1355 			switch (dst->sa_family) {
1356 #ifdef INET6
1357 			case AF_INET6:
1358 				flowtable_route_flush(V_ip6_ft, rt0);
1359 				break;
1360 #endif
1361 #ifdef INET
1362 			case AF_INET:
1363 				flowtable_route_flush(V_ip_ft, rt0);
1364 				break;
1365 #endif
1366 			}
1367 			RTFREE(rt0);
1368 		}
1369 #endif
1370 
1371 		/*
1372 		 * If this protocol has something to add to this then
1373 		 * allow it to do that as well.
1374 		 */
1375 		if (ifa->ifa_rtrequest)
1376 			ifa->ifa_rtrequest(req, rt, info);
1377 
1378 		/*
1379 		 * actually return a resultant rtentry and
1380 		 * give the caller a single reference.
1381 		 */
1382 		if (ret_nrt) {
1383 			*ret_nrt = rt;
1384 			RT_ADDREF(rt);
1385 		}
1386 		RT_UNLOCK(rt);
1387 		break;
1388 	default:
1389 		error = EOPNOTSUPP;
1390 	}
1391 bad:
1392 	if (needlock)
1393 		RADIX_NODE_HEAD_UNLOCK(rnh);
1394 	return (error);
1395 #undef senderr
1396 }
1397 
1398 #undef dst
1399 #undef gateway
1400 #undef netmask
1401 #undef ifaaddr
1402 #undef ifpaddr
1403 #undef flags
1404 
1405 int
1406 rt_setgate(struct rtentry *rt, struct sockaddr *dst, struct sockaddr *gate)
1407 {
1408 	/* XXX dst may be overwritten, can we move this to below */
1409 	int dlen = SA_SIZE(dst), glen = SA_SIZE(gate);
1410 #ifdef INVARIANTS
1411 	struct radix_node_head *rnh;
1412 
1413 	rnh = rt_tables_get_rnh(rt->rt_fibnum, dst->sa_family);
1414 #endif
1415 
1416 	RT_LOCK_ASSERT(rt);
1417 	RADIX_NODE_HEAD_LOCK_ASSERT(rnh);
1418 
1419 	/*
1420 	 * Prepare to store the gateway in rt->rt_gateway.
1421 	 * Both dst and gateway are stored one after the other in the same
1422 	 * malloc'd chunk. If we have room, we can reuse the old buffer,
1423 	 * rt_gateway already points to the right place.
1424 	 * Otherwise, malloc a new block and update the 'dst' address.
1425 	 */
1426 	if (rt->rt_gateway == NULL || glen > SA_SIZE(rt->rt_gateway)) {
1427 		caddr_t new;
1428 
1429 		R_Malloc(new, caddr_t, dlen + glen);
1430 		if (new == NULL)
1431 			return ENOBUFS;
1432 		/*
1433 		 * XXX note, we copy from *dst and not *rt_key(rt) because
1434 		 * rt_setgate() can be called to initialize a newly
1435 		 * allocated route entry, in which case rt_key(rt) == NULL
1436 		 * (and also rt->rt_gateway == NULL).
1437 		 * Free()/free() handle a NULL argument just fine.
1438 		 */
1439 		bcopy(dst, new, dlen);
1440 		Free(rt_key(rt));	/* free old block, if any */
1441 		rt_key(rt) = (struct sockaddr *)new;
1442 		rt->rt_gateway = (struct sockaddr *)(new + dlen);
1443 	}
1444 
1445 	/*
1446 	 * Copy the new gateway value into the memory chunk.
1447 	 */
1448 	bcopy(gate, rt->rt_gateway, glen);
1449 
1450 	return (0);
1451 }
1452 
1453 void
1454 rt_maskedcopy(struct sockaddr *src, struct sockaddr *dst, struct sockaddr *netmask)
1455 {
1456 	register u_char *cp1 = (u_char *)src;
1457 	register u_char *cp2 = (u_char *)dst;
1458 	register u_char *cp3 = (u_char *)netmask;
1459 	u_char *cplim = cp2 + *cp3;
1460 	u_char *cplim2 = cp2 + *cp1;
1461 
1462 	*cp2++ = *cp1++; *cp2++ = *cp1++; /* copies sa_len & sa_family */
1463 	cp3 += 2;
1464 	if (cplim > cplim2)
1465 		cplim = cplim2;
1466 	while (cp2 < cplim)
1467 		*cp2++ = *cp1++ & *cp3++;
1468 	if (cp2 < cplim2)
1469 		bzero((caddr_t)cp2, (unsigned)(cplim2 - cp2));
1470 }
1471 
1472 /*
1473  * Set up a routing table entry, normally
1474  * for an interface.
1475  */
1476 #define _SOCKADDR_TMPSIZE 128 /* Not too big.. kernel stack size is limited */
1477 static inline  int
1478 rtinit1(struct ifaddr *ifa, int cmd, int flags, int fibnum)
1479 {
1480 	struct sockaddr *dst;
1481 	struct sockaddr *netmask;
1482 	struct rtentry *rt = NULL;
1483 	struct rt_addrinfo info;
1484 	int error = 0;
1485 	int startfib, endfib;
1486 	char tempbuf[_SOCKADDR_TMPSIZE];
1487 	int didwork = 0;
1488 	int a_failure = 0;
1489 	static struct sockaddr_dl null_sdl = {sizeof(null_sdl), AF_LINK};
1490 	struct radix_node_head *rnh;
1491 
1492 	if (flags & RTF_HOST) {
1493 		dst = ifa->ifa_dstaddr;
1494 		netmask = NULL;
1495 	} else {
1496 		dst = ifa->ifa_addr;
1497 		netmask = ifa->ifa_netmask;
1498 	}
1499 	if (dst->sa_len == 0)
1500 		return(EINVAL);
1501 	switch (dst->sa_family) {
1502 	case AF_INET6:
1503 	case AF_INET:
1504 		/* We support multiple FIBs. */
1505 		break;
1506 	default:
1507 		fibnum = RT_DEFAULT_FIB;
1508 		break;
1509 	}
1510 	if (fibnum == RT_ALL_FIBS) {
1511 		if (rt_add_addr_allfibs == 0 && cmd == (int)RTM_ADD) {
1512 			startfib = endfib = curthread->td_proc->p_fibnum;
1513 		} else {
1514 			startfib = 0;
1515 			endfib = rt_numfibs - 1;
1516 		}
1517 	} else {
1518 		KASSERT((fibnum < rt_numfibs), ("rtinit1: bad fibnum"));
1519 		startfib = fibnum;
1520 		endfib = fibnum;
1521 	}
1522 
1523 	/*
1524 	 * If it's a delete, check that if it exists,
1525 	 * it's on the correct interface or we might scrub
1526 	 * a route to another ifa which would
1527 	 * be confusing at best and possibly worse.
1528 	 */
1529 	if (cmd == RTM_DELETE) {
1530 		/*
1531 		 * It's a delete, so it should already exist..
1532 		 * If it's a net, mask off the host bits
1533 		 * (Assuming we have a mask)
1534 		 * XXX this is kinda inet specific..
1535 		 */
1536 		if (netmask != NULL) {
1537 			rt_maskedcopy(dst, (struct sockaddr *)tempbuf, netmask);
1538 			dst = (struct sockaddr *)tempbuf;
1539 		}
1540 	}
1541 	/*
1542 	 * Now go through all the requested tables (fibs) and do the
1543 	 * requested action. Realistically, this will either be fib 0
1544 	 * for protocols that don't do multiple tables or all the
1545 	 * tables for those that do.
1546 	 */
1547 	for ( fibnum = startfib; fibnum <= endfib; fibnum++) {
1548 		if (cmd == RTM_DELETE) {
1549 			struct radix_node *rn;
1550 			/*
1551 			 * Look up an rtentry that is in the routing tree and
1552 			 * contains the correct info.
1553 			 */
1554 			rnh = rt_tables_get_rnh(fibnum, dst->sa_family);
1555 			if (rnh == NULL)
1556 				/* this table doesn't exist but others might */
1557 				continue;
1558 			RADIX_NODE_HEAD_RLOCK(rnh);
1559 			rn = rnh->rnh_lookup(dst, netmask, rnh);
1560 #ifdef RADIX_MPATH
1561 			if (rn_mpath_capable(rnh)) {
1562 
1563 				if (rn == NULL)
1564 					error = ESRCH;
1565 				else {
1566 					rt = RNTORT(rn);
1567 					/*
1568 					 * for interface route the
1569 					 * rt->rt_gateway is sockaddr_intf
1570 					 * for cloning ARP entries, so
1571 					 * rt_mpath_matchgate must use the
1572 					 * interface address
1573 					 */
1574 					rt = rt_mpath_matchgate(rt,
1575 					    ifa->ifa_addr);
1576 					if (rt == NULL)
1577 						error = ESRCH;
1578 				}
1579 			}
1580 #endif
1581 			error = (rn == NULL ||
1582 			    (rn->rn_flags & RNF_ROOT) ||
1583 			    RNTORT(rn)->rt_ifa != ifa);
1584 			RADIX_NODE_HEAD_RUNLOCK(rnh);
1585 			if (error) {
1586 				/* this is only an error if bad on ALL tables */
1587 				continue;
1588 			}
1589 		}
1590 		/*
1591 		 * Do the actual request
1592 		 */
1593 		bzero((caddr_t)&info, sizeof(info));
1594 		info.rti_ifa = ifa;
1595 		info.rti_flags = flags |
1596 		    (ifa->ifa_flags & ~IFA_RTSELF) | RTF_PINNED;
1597 		info.rti_info[RTAX_DST] = dst;
1598 		/*
1599 		 * doing this for compatibility reasons
1600 		 */
1601 		if (cmd == RTM_ADD)
1602 			info.rti_info[RTAX_GATEWAY] =
1603 			    (struct sockaddr *)&null_sdl;
1604 		else
1605 			info.rti_info[RTAX_GATEWAY] = ifa->ifa_addr;
1606 		info.rti_info[RTAX_NETMASK] = netmask;
1607 		error = rtrequest1_fib(cmd, &info, &rt, fibnum);
1608 
1609 		if ((error == EEXIST) && (cmd == RTM_ADD)) {
1610 			/*
1611 			 * Interface route addition failed.
1612 			 * Atomically delete current prefix generating
1613 			 * RTM_DELETE message, and retry adding
1614 			 * interface prefix.
1615 			 */
1616 			rnh = rt_tables_get_rnh(fibnum, dst->sa_family);
1617 			RADIX_NODE_HEAD_LOCK(rnh);
1618 
1619 			/* Delete old prefix */
1620 			info.rti_ifa = NULL;
1621 			info.rti_flags = RTF_RNH_LOCKED;
1622 
1623 			error = rtrequest1_fib(RTM_DELETE, &info, NULL, fibnum);
1624 			if (error == 0) {
1625 				info.rti_ifa = ifa;
1626 				info.rti_flags = flags | RTF_RNH_LOCKED |
1627 				    (ifa->ifa_flags & ~IFA_RTSELF) | RTF_PINNED;
1628 				error = rtrequest1_fib(cmd, &info, &rt, fibnum);
1629 			}
1630 
1631 			RADIX_NODE_HEAD_UNLOCK(rnh);
1632 		}
1633 
1634 
1635 		if (error == 0 && rt != NULL) {
1636 			/*
1637 			 * notify any listening routing agents of the change
1638 			 */
1639 			RT_LOCK(rt);
1640 #ifdef RADIX_MPATH
1641 			/*
1642 			 * in case address alias finds the first address
1643 			 * e.g. ifconfig bge0 192.0.2.246/24
1644 			 * e.g. ifconfig bge0 192.0.2.247/24
1645 			 * the address set in the route is 192.0.2.246
1646 			 * so we need to replace it with 192.0.2.247
1647 			 */
1648 			if (memcmp(rt->rt_ifa->ifa_addr,
1649 			    ifa->ifa_addr, ifa->ifa_addr->sa_len)) {
1650 				ifa_free(rt->rt_ifa);
1651 				ifa_ref(ifa);
1652 				rt->rt_ifp = ifa->ifa_ifp;
1653 				rt->rt_ifa = ifa;
1654 			}
1655 #endif
1656 			/*
1657 			 * doing this for compatibility reasons
1658 			 */
1659 			if (cmd == RTM_ADD) {
1660 			    ((struct sockaddr_dl *)rt->rt_gateway)->sdl_type  =
1661 				rt->rt_ifp->if_type;
1662 			    ((struct sockaddr_dl *)rt->rt_gateway)->sdl_index =
1663 				rt->rt_ifp->if_index;
1664 			}
1665 			RT_ADDREF(rt);
1666 			RT_UNLOCK(rt);
1667 			rt_newaddrmsg_fib(cmd, ifa, error, rt, fibnum);
1668 			RT_LOCK(rt);
1669 			RT_REMREF(rt);
1670 			if (cmd == RTM_DELETE) {
1671 				/*
1672 				 * If we are deleting, and we found an entry,
1673 				 * then it's been removed from the tree..
1674 				 * now throw it away.
1675 				 */
1676 				RTFREE_LOCKED(rt);
1677 			} else {
1678 				if (cmd == RTM_ADD) {
1679 					/*
1680 					 * We just wanted to add it..
1681 					 * we don't actually need a reference.
1682 					 */
1683 					RT_REMREF(rt);
1684 				}
1685 				RT_UNLOCK(rt);
1686 			}
1687 			didwork = 1;
1688 		}
1689 		if (error)
1690 			a_failure = error;
1691 	}
1692 	if (cmd == RTM_DELETE) {
1693 		if (didwork) {
1694 			error = 0;
1695 		} else {
1696 			/* we only give an error if it wasn't in any table */
1697 			error = ((flags & RTF_HOST) ?
1698 			    EHOSTUNREACH : ENETUNREACH);
1699 		}
1700 	} else {
1701 		if (a_failure) {
1702 			/* return an error if any of them failed */
1703 			error = a_failure;
1704 		}
1705 	}
1706 	return (error);
1707 }
1708 
1709 #ifndef BURN_BRIDGES
1710 /* special one for inet internal use. may not use. */
1711 int
1712 rtinit_fib(struct ifaddr *ifa, int cmd, int flags)
1713 {
1714 	return (rtinit1(ifa, cmd, flags, RT_ALL_FIBS));
1715 }
1716 #endif
1717 
1718 /*
1719  * Set up a routing table entry, normally
1720  * for an interface.
1721  */
1722 int
1723 rtinit(struct ifaddr *ifa, int cmd, int flags)
1724 {
1725 	struct sockaddr *dst;
1726 	int fib = RT_DEFAULT_FIB;
1727 
1728 	if (flags & RTF_HOST) {
1729 		dst = ifa->ifa_dstaddr;
1730 	} else {
1731 		dst = ifa->ifa_addr;
1732 	}
1733 
1734 	switch (dst->sa_family) {
1735 	case AF_INET6:
1736 	case AF_INET:
1737 		/* We do support multiple FIBs. */
1738 		fib = RT_ALL_FIBS;
1739 		break;
1740 	}
1741 	return (rtinit1(ifa, cmd, flags, fib));
1742 }
1743 
1744 /*
1745  * Announce interface address arrival/withdraw
1746  * Returns 0 on success.
1747  */
1748 int
1749 rt_addrmsg(int cmd, struct ifaddr *ifa, int fibnum)
1750 {
1751 
1752 	KASSERT(cmd == RTM_ADD || cmd == RTM_DELETE,
1753 	    ("unexpected cmd %d", cmd));
1754 
1755 	KASSERT(fibnum == RT_ALL_FIBS || (fibnum >= 0 && fibnum < rt_numfibs),
1756 	    ("%s: fib out of range 0 <=%d<%d", __func__, fibnum, rt_numfibs));
1757 
1758 	return (rtsock_addrmsg(cmd, ifa, fibnum));
1759 }
1760 
1761 /*
1762  * Announce route addition/removal.
1763  * Users of this function MUST validate input data BEFORE calling.
1764  * However we have to be able to handle invalid data:
1765  * if some userland app sends us "invalid" route message (invalid mask,
1766  * no dst, wrong address families, etc...) we need to pass it back
1767  * to app (and any other rtsock consumers) with rtm_errno field set to
1768  * non-zero value.
1769  * Returns 0 on success.
1770  */
1771 int
1772 rt_routemsg(int cmd, struct ifnet *ifp, int error, struct rtentry *rt,
1773     int fibnum)
1774 {
1775 
1776 	KASSERT(cmd == RTM_ADD || cmd == RTM_DELETE,
1777 	    ("unexpected cmd %d", cmd));
1778 
1779 	KASSERT(fibnum == RT_ALL_FIBS || (fibnum >= 0 && fibnum < rt_numfibs),
1780 	    ("%s: fib out of range 0 <=%d<%d", __func__, fibnum, rt_numfibs));
1781 
1782 	KASSERT(rt_key(rt) != NULL, (":%s: rt_key must be supplied", __func__));
1783 
1784 	return (rtsock_routemsg(cmd, ifp, error, rt, fibnum));
1785 }
1786 
1787 void
1788 rt_newaddrmsg(int cmd, struct ifaddr *ifa, int error, struct rtentry *rt)
1789 {
1790 
1791 	rt_newaddrmsg_fib(cmd, ifa, error, rt, RT_ALL_FIBS);
1792 }
1793 
1794 /*
1795  * This is called to generate messages from the routing socket
1796  * indicating a network interface has had addresses associated with it.
1797  */
1798 void
1799 rt_newaddrmsg_fib(int cmd, struct ifaddr *ifa, int error, struct rtentry *rt,
1800     int fibnum)
1801 {
1802 
1803 	KASSERT(cmd == RTM_ADD || cmd == RTM_DELETE,
1804 		("unexpected cmd %u", cmd));
1805 	KASSERT(fibnum == RT_ALL_FIBS || (fibnum >= 0 && fibnum < rt_numfibs),
1806 	    ("%s: fib out of range 0 <=%d<%d", __func__, fibnum, rt_numfibs));
1807 
1808 #if defined(INET) || defined(INET6)
1809 #ifdef SCTP
1810 	/*
1811 	 * notify the SCTP stack
1812 	 * this will only get called when an address is added/deleted
1813 	 * XXX pass the ifaddr struct instead if ifa->ifa_addr...
1814 	 */
1815 	sctp_addr_change(ifa, cmd);
1816 #endif /* SCTP */
1817 #endif
1818 	if (cmd == RTM_ADD) {
1819 		rt_addrmsg(cmd, ifa, fibnum);
1820 		if (rt != NULL)
1821 			rt_routemsg(cmd, ifa->ifa_ifp, error, rt, fibnum);
1822 	} else {
1823 		if (rt != NULL)
1824 			rt_routemsg(cmd, ifa->ifa_ifp, error, rt, fibnum);
1825 		rt_addrmsg(cmd, ifa, fibnum);
1826 	}
1827 }
1828 
1829