xref: /freebsd/sys/net/route.c (revision 3b8f08459569bf0faa21473e5cec2491e95c9349)
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 /*
129  * Convert a 'struct radix_node *' to a 'struct rtentry *'.
130  * The operation can be done safely (in this code) because a
131  * 'struct rtentry' starts with two 'struct radix_node''s, the first
132  * one representing leaf nodes in the routing tree, which is
133  * what the code in radix.c passes us as a 'struct radix_node'.
134  *
135  * But because there are a lot of assumptions in this conversion,
136  * do not cast explicitly, but always use the macro below.
137  */
138 #define RNTORT(p)	((struct rtentry *)(p))
139 
140 static VNET_DEFINE(uma_zone_t, rtzone);		/* Routing table UMA zone. */
141 #define	V_rtzone	VNET(rtzone)
142 
143 static int rtrequest1_fib_change(struct radix_node_head *, struct rt_addrinfo *,
144     struct rtentry **, u_int);
145 
146 /*
147  * handler for net.my_fibnum
148  */
149 static int
150 sysctl_my_fibnum(SYSCTL_HANDLER_ARGS)
151 {
152         int fibnum;
153         int error;
154 
155         fibnum = curthread->td_proc->p_fibnum;
156         error = sysctl_handle_int(oidp, &fibnum, 0, req);
157         return (error);
158 }
159 
160 SYSCTL_PROC(_net, OID_AUTO, my_fibnum, CTLTYPE_INT|CTLFLAG_RD,
161             NULL, 0, &sysctl_my_fibnum, "I", "default FIB of caller");
162 
163 static __inline struct radix_node_head **
164 rt_tables_get_rnh_ptr(int table, int fam)
165 {
166 	struct radix_node_head **rnh;
167 
168 	KASSERT(table >= 0 && table < rt_numfibs, ("%s: table out of bounds.",
169 	    __func__));
170 	KASSERT(fam >= 0 && fam < (AF_MAX+1), ("%s: fam out of bounds.",
171 	    __func__));
172 
173 	/* rnh is [fib=0][af=0]. */
174 	rnh = (struct radix_node_head **)V_rt_tables;
175 	/* Get the offset to the requested table and fam. */
176 	rnh += table * (AF_MAX+1) + fam;
177 
178 	return (rnh);
179 }
180 
181 struct radix_node_head *
182 rt_tables_get_rnh(int table, int fam)
183 {
184 
185 	return (*rt_tables_get_rnh_ptr(table, fam));
186 }
187 
188 /*
189  * route initialization must occur before ip6_init2(), which happenas at
190  * SI_ORDER_MIDDLE.
191  */
192 static void
193 route_init(void)
194 {
195 
196 	/* whack the tunable ints into  line. */
197 	if (rt_numfibs > RT_MAXFIBS)
198 		rt_numfibs = RT_MAXFIBS;
199 	if (rt_numfibs == 0)
200 		rt_numfibs = 1;
201 }
202 SYSINIT(route_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_THIRD, route_init, 0);
203 
204 static int
205 rtentry_zinit(void *mem, int size, int how)
206 {
207 	struct rtentry *rt = mem;
208 
209 	rt->rt_pksent = counter_u64_alloc(how);
210 	if (rt->rt_pksent == NULL)
211 		return (ENOMEM);
212 
213 	RT_LOCK_INIT(rt);
214 
215 	return (0);
216 }
217 
218 static void
219 rtentry_zfini(void *mem, int size)
220 {
221 	struct rtentry *rt = mem;
222 
223 	RT_LOCK_DESTROY(rt);
224 	counter_u64_free(rt->rt_pksent);
225 }
226 
227 static int
228 rtentry_ctor(void *mem, int size, void *arg, int how)
229 {
230 	struct rtentry *rt = mem;
231 
232 	bzero(rt, offsetof(struct rtentry, rt_endzero));
233 	counter_u64_zero(rt->rt_pksent);
234 
235 	return (0);
236 }
237 
238 static void
239 rtentry_dtor(void *mem, int size, void *arg)
240 {
241 	struct rtentry *rt = mem;
242 
243 	RT_UNLOCK_COND(rt);
244 }
245 
246 static void
247 vnet_route_init(const void *unused __unused)
248 {
249 	struct domain *dom;
250 	struct radix_node_head **rnh;
251 	int table;
252 	int fam;
253 
254 	V_rt_tables = malloc(rt_numfibs * (AF_MAX+1) *
255 	    sizeof(struct radix_node_head *), M_RTABLE, M_WAITOK|M_ZERO);
256 
257 	V_rtzone = uma_zcreate("rtentry", sizeof(struct rtentry),
258 	    rtentry_ctor, rtentry_dtor,
259 	    rtentry_zinit, rtentry_zfini, UMA_ALIGN_PTR, 0);
260 	for (dom = domains; dom; dom = dom->dom_next) {
261 		if (dom->dom_rtattach == NULL)
262 			continue;
263 
264 		for  (table = 0; table < rt_numfibs; table++) {
265 			fam = dom->dom_family;
266 			if (table != 0 && fam != AF_INET6 && fam != AF_INET)
267 				break;
268 
269 			/*
270 			 * XXX MRT rtattach will be also called from
271 			 * vfs_export.c but the offset will be 0 (only for
272 			 * AF_INET and AF_INET6 which don't need it anyhow).
273 			 */
274 			rnh = rt_tables_get_rnh_ptr(table, fam);
275 			if (rnh == NULL)
276 				panic("%s: rnh NULL", __func__);
277 			dom->dom_rtattach((void **)rnh, dom->dom_rtoffset);
278 		}
279 	}
280 }
281 VNET_SYSINIT(vnet_route_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_FOURTH,
282     vnet_route_init, 0);
283 
284 #ifdef VIMAGE
285 static void
286 vnet_route_uninit(const void *unused __unused)
287 {
288 	int table;
289 	int fam;
290 	struct domain *dom;
291 	struct radix_node_head **rnh;
292 
293 	for (dom = domains; dom; dom = dom->dom_next) {
294 		if (dom->dom_rtdetach == NULL)
295 			continue;
296 
297 		for (table = 0; table < rt_numfibs; table++) {
298 			fam = dom->dom_family;
299 
300 			if (table != 0 && fam != AF_INET6 && fam != AF_INET)
301 				break;
302 
303 			rnh = rt_tables_get_rnh_ptr(table, fam);
304 			if (rnh == NULL)
305 				panic("%s: rnh NULL", __func__);
306 			dom->dom_rtdetach((void **)rnh, dom->dom_rtoffset);
307 		}
308 	}
309 
310 	free(V_rt_tables, M_RTABLE);
311 	uma_zdestroy(V_rtzone);
312 }
313 VNET_SYSUNINIT(vnet_route_uninit, SI_SUB_PROTO_DOMAIN, SI_ORDER_THIRD,
314     vnet_route_uninit, 0);
315 #endif
316 
317 #ifndef _SYS_SYSPROTO_H_
318 struct setfib_args {
319 	int     fibnum;
320 };
321 #endif
322 int
323 sys_setfib(struct thread *td, struct setfib_args *uap)
324 {
325 	if (uap->fibnum < 0 || uap->fibnum >= rt_numfibs)
326 		return EINVAL;
327 	td->td_proc->p_fibnum = uap->fibnum;
328 	return (0);
329 }
330 
331 /*
332  * Packet routing routines.
333  */
334 void
335 rtalloc(struct route *ro)
336 {
337 
338 	rtalloc_ign_fib(ro, 0UL, RT_DEFAULT_FIB);
339 }
340 
341 void
342 rtalloc_fib(struct route *ro, u_int fibnum)
343 {
344 	rtalloc_ign_fib(ro, 0UL, fibnum);
345 }
346 
347 void
348 rtalloc_ign(struct route *ro, u_long ignore)
349 {
350 	struct rtentry *rt;
351 
352 	if ((rt = ro->ro_rt) != NULL) {
353 		if (rt->rt_ifp != NULL && rt->rt_flags & RTF_UP)
354 			return;
355 		RTFREE(rt);
356 		ro->ro_rt = NULL;
357 	}
358 	ro->ro_rt = rtalloc1_fib(&ro->ro_dst, 1, ignore, RT_DEFAULT_FIB);
359 	if (ro->ro_rt)
360 		RT_UNLOCK(ro->ro_rt);
361 }
362 
363 void
364 rtalloc_ign_fib(struct route *ro, u_long ignore, u_int fibnum)
365 {
366 	struct rtentry *rt;
367 
368 	if ((rt = ro->ro_rt) != NULL) {
369 		if (rt->rt_ifp != NULL && rt->rt_flags & RTF_UP)
370 			return;
371 		RTFREE(rt);
372 		ro->ro_rt = NULL;
373 	}
374 	ro->ro_rt = rtalloc1_fib(&ro->ro_dst, 1, ignore, fibnum);
375 	if (ro->ro_rt)
376 		RT_UNLOCK(ro->ro_rt);
377 }
378 
379 /*
380  * Look up the route that matches the address given
381  * Or, at least try.. Create a cloned route if needed.
382  *
383  * The returned route, if any, is locked.
384  */
385 struct rtentry *
386 rtalloc1(struct sockaddr *dst, int report, u_long ignflags)
387 {
388 
389 	return (rtalloc1_fib(dst, report, ignflags, RT_DEFAULT_FIB));
390 }
391 
392 struct rtentry *
393 rtalloc1_fib(struct sockaddr *dst, int report, u_long ignflags,
394 		    u_int fibnum)
395 {
396 	struct radix_node_head *rnh;
397 	struct radix_node *rn;
398 	struct rtentry *newrt;
399 	struct rt_addrinfo info;
400 	int err = 0, msgtype = RTM_MISS;
401 	int needlock;
402 
403 	KASSERT((fibnum < rt_numfibs), ("rtalloc1_fib: bad fibnum"));
404 	switch (dst->sa_family) {
405 	case AF_INET6:
406 	case AF_INET:
407 		/* We support multiple FIBs. */
408 		break;
409 	default:
410 		fibnum = RT_DEFAULT_FIB;
411 		break;
412 	}
413 	rnh = rt_tables_get_rnh(fibnum, dst->sa_family);
414 	newrt = NULL;
415 	if (rnh == NULL)
416 		goto miss;
417 
418 	/*
419 	 * Look up the address in the table for that Address Family
420 	 */
421 	needlock = !(ignflags & RTF_RNH_LOCKED);
422 	if (needlock)
423 		RADIX_NODE_HEAD_RLOCK(rnh);
424 #ifdef INVARIANTS
425 	else
426 		RADIX_NODE_HEAD_LOCK_ASSERT(rnh);
427 #endif
428 	rn = rnh->rnh_matchaddr(dst, rnh);
429 	if (rn && ((rn->rn_flags & RNF_ROOT) == 0)) {
430 		newrt = RNTORT(rn);
431 		RT_LOCK(newrt);
432 		RT_ADDREF(newrt);
433 		if (needlock)
434 			RADIX_NODE_HEAD_RUNLOCK(rnh);
435 		goto done;
436 
437 	} else if (needlock)
438 		RADIX_NODE_HEAD_RUNLOCK(rnh);
439 
440 	/*
441 	 * Either we hit the root or couldn't find any match,
442 	 * Which basically means
443 	 * "caint get there frm here"
444 	 */
445 miss:
446 	V_rtstat.rts_unreach++;
447 
448 	if (report) {
449 		/*
450 		 * If required, report the failure to the supervising
451 		 * Authorities.
452 		 * For a delete, this is not an error. (report == 0)
453 		 */
454 		bzero(&info, sizeof(info));
455 		info.rti_info[RTAX_DST] = dst;
456 		rt_missmsg_fib(msgtype, &info, 0, err, fibnum);
457 	}
458 done:
459 	if (newrt)
460 		RT_LOCK_ASSERT(newrt);
461 	return (newrt);
462 }
463 
464 /*
465  * Remove a reference count from an rtentry.
466  * If the count gets low enough, take it out of the routing table
467  */
468 void
469 rtfree(struct rtentry *rt)
470 {
471 	struct radix_node_head *rnh;
472 
473 	KASSERT(rt != NULL,("%s: NULL rt", __func__));
474 	rnh = rt_tables_get_rnh(rt->rt_fibnum, rt_key(rt)->sa_family);
475 	KASSERT(rnh != NULL,("%s: NULL rnh", __func__));
476 
477 	RT_LOCK_ASSERT(rt);
478 
479 	/*
480 	 * The callers should use RTFREE_LOCKED() or RTFREE(), so
481 	 * we should come here exactly with the last reference.
482 	 */
483 	RT_REMREF(rt);
484 	if (rt->rt_refcnt > 0) {
485 		log(LOG_DEBUG, "%s: %p has %d refs\n", __func__, rt, rt->rt_refcnt);
486 		goto done;
487 	}
488 
489 	/*
490 	 * On last reference give the "close method" a chance
491 	 * to cleanup private state.  This also permits (for
492 	 * IPv4 and IPv6) a chance to decide if the routing table
493 	 * entry should be purged immediately or at a later time.
494 	 * When an immediate purge is to happen the close routine
495 	 * typically calls rtexpunge which clears the RTF_UP flag
496 	 * on the entry so that the code below reclaims the storage.
497 	 */
498 	if (rt->rt_refcnt == 0 && rnh->rnh_close)
499 		rnh->rnh_close((struct radix_node *)rt, rnh);
500 
501 	/*
502 	 * If we are no longer "up" (and ref == 0)
503 	 * then we can free the resources associated
504 	 * with the route.
505 	 */
506 	if ((rt->rt_flags & RTF_UP) == 0) {
507 		if (rt->rt_nodes->rn_flags & (RNF_ACTIVE | RNF_ROOT))
508 			panic("rtfree 2");
509 		/*
510 		 * the rtentry must have been removed from the routing table
511 		 * so it is represented in rttrash.. remove that now.
512 		 */
513 		V_rttrash--;
514 #ifdef	DIAGNOSTIC
515 		if (rt->rt_refcnt < 0) {
516 			printf("rtfree: %p not freed (neg refs)\n", rt);
517 			goto done;
518 		}
519 #endif
520 		/*
521 		 * release references on items we hold them on..
522 		 * e.g other routes and ifaddrs.
523 		 */
524 		if (rt->rt_ifa)
525 			ifa_free(rt->rt_ifa);
526 		/*
527 		 * The key is separatly alloc'd so free it (see rt_setgate()).
528 		 * This also frees the gateway, as they are always malloc'd
529 		 * together.
530 		 */
531 		Free(rt_key(rt));
532 
533 		/*
534 		 * and the rtentry itself of course
535 		 */
536 		uma_zfree(V_rtzone, rt);
537 		return;
538 	}
539 done:
540 	RT_UNLOCK(rt);
541 }
542 
543 
544 /*
545  * Force a routing table entry to the specified
546  * destination to go through the given gateway.
547  * Normally called as a result of a routing redirect
548  * message from the network layer.
549  */
550 void
551 rtredirect(struct sockaddr *dst,
552 	struct sockaddr *gateway,
553 	struct sockaddr *netmask,
554 	int flags,
555 	struct sockaddr *src)
556 {
557 
558 	rtredirect_fib(dst, gateway, netmask, flags, src, RT_DEFAULT_FIB);
559 }
560 
561 void
562 rtredirect_fib(struct sockaddr *dst,
563 	struct sockaddr *gateway,
564 	struct sockaddr *netmask,
565 	int flags,
566 	struct sockaddr *src,
567 	u_int fibnum)
568 {
569 	struct rtentry *rt, *rt0 = NULL;
570 	int error = 0;
571 	short *stat = NULL;
572 	struct rt_addrinfo info;
573 	struct ifaddr *ifa;
574 	struct radix_node_head *rnh;
575 
576 	ifa = NULL;
577 	rnh = rt_tables_get_rnh(fibnum, dst->sa_family);
578 	if (rnh == NULL) {
579 		error = EAFNOSUPPORT;
580 		goto out;
581 	}
582 
583 	/* verify the gateway is directly reachable */
584 	if ((ifa = ifa_ifwithnet(gateway, 0, fibnum)) == NULL) {
585 		error = ENETUNREACH;
586 		goto out;
587 	}
588 	rt = rtalloc1_fib(dst, 0, 0UL, fibnum);	/* NB: rt is locked */
589 	/*
590 	 * If the redirect isn't from our current router for this dst,
591 	 * it's either old or wrong.  If it redirects us to ourselves,
592 	 * we have a routing loop, perhaps as a result of an interface
593 	 * going down recently.
594 	 */
595 	if (!(flags & RTF_DONE) && rt &&
596 	     (!sa_equal(src, rt->rt_gateway) || rt->rt_ifa != ifa))
597 		error = EINVAL;
598 	else if (ifa_ifwithaddr_check(gateway))
599 		error = EHOSTUNREACH;
600 	if (error)
601 		goto done;
602 	/*
603 	 * Create a new entry if we just got back a wildcard entry
604 	 * or the lookup failed.  This is necessary for hosts
605 	 * which use routing redirects generated by smart gateways
606 	 * to dynamically build the routing tables.
607 	 */
608 	if (rt == NULL || (rt_mask(rt) && rt_mask(rt)->sa_len < 2))
609 		goto create;
610 	/*
611 	 * Don't listen to the redirect if it's
612 	 * for a route to an interface.
613 	 */
614 	if (rt->rt_flags & RTF_GATEWAY) {
615 		if (((rt->rt_flags & RTF_HOST) == 0) && (flags & RTF_HOST)) {
616 			/*
617 			 * Changing from route to net => route to host.
618 			 * Create new route, rather than smashing route to net.
619 			 */
620 		create:
621 			rt0 = rt;
622 			rt = NULL;
623 
624 			flags |=  RTF_GATEWAY | RTF_DYNAMIC;
625 			bzero((caddr_t)&info, sizeof(info));
626 			info.rti_info[RTAX_DST] = dst;
627 			info.rti_info[RTAX_GATEWAY] = gateway;
628 			info.rti_info[RTAX_NETMASK] = netmask;
629 			info.rti_ifa = ifa;
630 			info.rti_flags = flags;
631 			if (rt0 != NULL)
632 				RT_UNLOCK(rt0);	/* drop lock to avoid LOR with RNH */
633 			error = rtrequest1_fib(RTM_ADD, &info, &rt, fibnum);
634 			if (rt != NULL) {
635 				RT_LOCK(rt);
636 				if (rt0 != NULL)
637 					EVENTHANDLER_INVOKE(route_redirect_event, rt0, rt, dst);
638 				flags = rt->rt_flags;
639 			}
640 			if (rt0 != NULL)
641 				RTFREE(rt0);
642 
643 			stat = &V_rtstat.rts_dynamic;
644 		} else {
645 			struct rtentry *gwrt;
646 
647 			/*
648 			 * Smash the current notion of the gateway to
649 			 * this destination.  Should check about netmask!!!
650 			 */
651 			rt->rt_flags |= RTF_MODIFIED;
652 			flags |= RTF_MODIFIED;
653 			stat = &V_rtstat.rts_newgateway;
654 			/*
655 			 * add the key and gateway (in one malloc'd chunk).
656 			 */
657 			RT_UNLOCK(rt);
658 			RADIX_NODE_HEAD_LOCK(rnh);
659 			RT_LOCK(rt);
660 			rt_setgate(rt, rt_key(rt), gateway);
661 			gwrt = rtalloc1(gateway, 1, RTF_RNH_LOCKED);
662 			RADIX_NODE_HEAD_UNLOCK(rnh);
663 			EVENTHANDLER_INVOKE(route_redirect_event, rt, gwrt, dst);
664 			RTFREE_LOCKED(gwrt);
665 		}
666 	} else
667 		error = EHOSTUNREACH;
668 done:
669 	if (rt)
670 		RTFREE_LOCKED(rt);
671 out:
672 	if (error)
673 		V_rtstat.rts_badredirect++;
674 	else if (stat != NULL)
675 		(*stat)++;
676 	bzero((caddr_t)&info, sizeof(info));
677 	info.rti_info[RTAX_DST] = dst;
678 	info.rti_info[RTAX_GATEWAY] = gateway;
679 	info.rti_info[RTAX_NETMASK] = netmask;
680 	info.rti_info[RTAX_AUTHOR] = src;
681 	rt_missmsg_fib(RTM_REDIRECT, &info, flags, error, fibnum);
682 	if (ifa != NULL)
683 		ifa_free(ifa);
684 }
685 
686 int
687 rtioctl(u_long req, caddr_t data)
688 {
689 
690 	return (rtioctl_fib(req, data, RT_DEFAULT_FIB));
691 }
692 
693 /*
694  * Routing table ioctl interface.
695  */
696 int
697 rtioctl_fib(u_long req, caddr_t data, u_int fibnum)
698 {
699 
700 	/*
701 	 * If more ioctl commands are added here, make sure the proper
702 	 * super-user checks are being performed because it is possible for
703 	 * prison-root to make it this far if raw sockets have been enabled
704 	 * in jails.
705 	 */
706 #ifdef INET
707 	/* Multicast goop, grrr... */
708 	return mrt_ioctl ? mrt_ioctl(req, data, fibnum) : EOPNOTSUPP;
709 #else /* INET */
710 	return ENXIO;
711 #endif /* INET */
712 }
713 
714 /*
715  * For both ifa_ifwithroute() routines, 'ifa' is returned referenced.
716  */
717 struct ifaddr *
718 ifa_ifwithroute(int flags, struct sockaddr *dst, struct sockaddr *gateway)
719 {
720 
721 	return (ifa_ifwithroute_fib(flags, dst, gateway, RT_DEFAULT_FIB));
722 }
723 
724 struct ifaddr *
725 ifa_ifwithroute_fib(int flags, struct sockaddr *dst, struct sockaddr *gateway,
726 				u_int fibnum)
727 {
728 	struct ifaddr *ifa;
729 	int not_found = 0;
730 
731 	if ((flags & RTF_GATEWAY) == 0) {
732 		/*
733 		 * If we are adding a route to an interface,
734 		 * and the interface is a pt to pt link
735 		 * we should search for the destination
736 		 * as our clue to the interface.  Otherwise
737 		 * we can use the local address.
738 		 */
739 		ifa = NULL;
740 		if (flags & RTF_HOST)
741 			ifa = ifa_ifwithdstaddr(dst, fibnum);
742 		if (ifa == NULL)
743 			ifa = ifa_ifwithaddr(gateway);
744 	} else {
745 		/*
746 		 * If we are adding a route to a remote net
747 		 * or host, the gateway may still be on the
748 		 * other end of a pt to pt link.
749 		 */
750 		ifa = ifa_ifwithdstaddr(gateway, fibnum);
751 	}
752 	if (ifa == NULL)
753 		ifa = ifa_ifwithnet(gateway, 0, fibnum);
754 	if (ifa == NULL) {
755 		struct rtentry *rt = rtalloc1_fib(gateway, 0, RTF_RNH_LOCKED, fibnum);
756 		if (rt == NULL)
757 			return (NULL);
758 		/*
759 		 * dismiss a gateway that is reachable only
760 		 * through the default router
761 		 */
762 		switch (gateway->sa_family) {
763 		case AF_INET:
764 			if (satosin(rt_key(rt))->sin_addr.s_addr == INADDR_ANY)
765 				not_found = 1;
766 			break;
767 		case AF_INET6:
768 			if (IN6_IS_ADDR_UNSPECIFIED(&satosin6(rt_key(rt))->sin6_addr))
769 				not_found = 1;
770 			break;
771 		default:
772 			break;
773 		}
774 		if (!not_found && rt->rt_ifa != NULL) {
775 			ifa = rt->rt_ifa;
776 			ifa_ref(ifa);
777 		}
778 		RT_REMREF(rt);
779 		RT_UNLOCK(rt);
780 		if (not_found || ifa == NULL)
781 			return (NULL);
782 	}
783 	if (ifa->ifa_addr->sa_family != dst->sa_family) {
784 		struct ifaddr *oifa = ifa;
785 		ifa = ifaof_ifpforaddr(dst, ifa->ifa_ifp);
786 		if (ifa == NULL)
787 			ifa = oifa;
788 		else
789 			ifa_free(oifa);
790 	}
791 	return (ifa);
792 }
793 
794 /*
795  * Do appropriate manipulations of a routing tree given
796  * all the bits of info needed
797  */
798 int
799 rtrequest(int req,
800 	struct sockaddr *dst,
801 	struct sockaddr *gateway,
802 	struct sockaddr *netmask,
803 	int flags,
804 	struct rtentry **ret_nrt)
805 {
806 
807 	return (rtrequest_fib(req, dst, gateway, netmask, flags, ret_nrt,
808 	    RT_DEFAULT_FIB));
809 }
810 
811 int
812 rtrequest_fib(int req,
813 	struct sockaddr *dst,
814 	struct sockaddr *gateway,
815 	struct sockaddr *netmask,
816 	int flags,
817 	struct rtentry **ret_nrt,
818 	u_int fibnum)
819 {
820 	struct rt_addrinfo info;
821 
822 	if (dst->sa_len == 0)
823 		return(EINVAL);
824 
825 	bzero((caddr_t)&info, sizeof(info));
826 	info.rti_flags = flags;
827 	info.rti_info[RTAX_DST] = dst;
828 	info.rti_info[RTAX_GATEWAY] = gateway;
829 	info.rti_info[RTAX_NETMASK] = netmask;
830 	return rtrequest1_fib(req, &info, ret_nrt, fibnum);
831 }
832 
833 /*
834  * These (questionable) definitions of apparent local variables apply
835  * to the next two functions.  XXXXXX!!!
836  */
837 #define	dst	info->rti_info[RTAX_DST]
838 #define	gateway	info->rti_info[RTAX_GATEWAY]
839 #define	netmask	info->rti_info[RTAX_NETMASK]
840 #define	ifaaddr	info->rti_info[RTAX_IFA]
841 #define	ifpaddr	info->rti_info[RTAX_IFP]
842 #define	flags	info->rti_flags
843 
844 int
845 rt_getifa(struct rt_addrinfo *info)
846 {
847 
848 	return (rt_getifa_fib(info, RT_DEFAULT_FIB));
849 }
850 
851 /*
852  * Look up rt_addrinfo for a specific fib.  Note that if rti_ifa is defined,
853  * it will be referenced so the caller must free it.
854  */
855 int
856 rt_getifa_fib(struct rt_addrinfo *info, u_int fibnum)
857 {
858 	struct ifaddr *ifa;
859 	int error = 0;
860 
861 	/*
862 	 * ifp may be specified by sockaddr_dl
863 	 * when protocol address is ambiguous.
864 	 */
865 	if (info->rti_ifp == NULL && ifpaddr != NULL &&
866 	    ifpaddr->sa_family == AF_LINK &&
867 	    (ifa = ifa_ifwithnet(ifpaddr, 0, fibnum)) != NULL) {
868 		info->rti_ifp = ifa->ifa_ifp;
869 		ifa_free(ifa);
870 	}
871 	if (info->rti_ifa == NULL && ifaaddr != NULL)
872 		info->rti_ifa = ifa_ifwithaddr(ifaaddr);
873 	if (info->rti_ifa == NULL) {
874 		struct sockaddr *sa;
875 
876 		sa = ifaaddr != NULL ? ifaaddr :
877 		    (gateway != NULL ? gateway : dst);
878 		if (sa != NULL && info->rti_ifp != NULL)
879 			info->rti_ifa = ifaof_ifpforaddr(sa, info->rti_ifp);
880 		else if (dst != NULL && gateway != NULL)
881 			info->rti_ifa = ifa_ifwithroute_fib(flags, dst, gateway,
882 							fibnum);
883 		else if (sa != NULL)
884 			info->rti_ifa = ifa_ifwithroute_fib(flags, sa, sa,
885 							fibnum);
886 	}
887 	if ((ifa = info->rti_ifa) != NULL) {
888 		if (info->rti_ifp == NULL)
889 			info->rti_ifp = ifa->ifa_ifp;
890 	} else
891 		error = ENETUNREACH;
892 	return (error);
893 }
894 
895 /*
896  * Expunges references to a route that's about to be reclaimed.
897  * The route must be locked.
898  */
899 int
900 rtexpunge(struct rtentry *rt)
901 {
902 #if !defined(RADIX_MPATH)
903 	struct radix_node *rn;
904 #else
905 	struct rt_addrinfo info;
906 	int fib;
907 	struct rtentry *rt0;
908 #endif
909 	struct radix_node_head *rnh;
910 	struct ifaddr *ifa;
911 	int error = 0;
912 
913 	/*
914 	 * Find the correct routing tree to use for this Address Family
915 	 */
916 	rnh = rt_tables_get_rnh(rt->rt_fibnum, rt_key(rt)->sa_family);
917 	RT_LOCK_ASSERT(rt);
918 	if (rnh == NULL)
919 		return (EAFNOSUPPORT);
920 	RADIX_NODE_HEAD_LOCK_ASSERT(rnh);
921 
922 #ifdef RADIX_MPATH
923 	fib = rt->rt_fibnum;
924 	bzero(&info, sizeof(info));
925 	info.rti_ifp = rt->rt_ifp;
926 	info.rti_flags = RTF_RNH_LOCKED;
927 	info.rti_info[RTAX_DST] = rt_key(rt);
928 	info.rti_info[RTAX_GATEWAY] = rt->rt_ifa->ifa_addr;
929 
930 	RT_UNLOCK(rt);
931 	error = rtrequest1_fib(RTM_DELETE, &info, &rt0, fib);
932 
933 	if (error == 0 && rt0 != NULL) {
934 		rt = rt0;
935 		RT_LOCK(rt);
936 	} else if (error != 0) {
937 		RT_LOCK(rt);
938 		return (error);
939 	}
940 #else
941 	/*
942 	 * Remove the item from the tree; it should be there,
943 	 * but when callers invoke us blindly it may not (sigh).
944 	 */
945 	rn = rnh->rnh_deladdr(rt_key(rt), rt_mask(rt), rnh);
946 	if (rn == NULL) {
947 		error = ESRCH;
948 		goto bad;
949 	}
950 	KASSERT((rn->rn_flags & (RNF_ACTIVE | RNF_ROOT)) == 0,
951 		("unexpected flags 0x%x", rn->rn_flags));
952 	KASSERT(rt == RNTORT(rn),
953 		("lookup mismatch, rt %p rn %p", rt, rn));
954 #endif /* RADIX_MPATH */
955 
956 	rt->rt_flags &= ~RTF_UP;
957 
958 	/*
959 	 * Give the protocol a chance to keep things in sync.
960 	 */
961 	if ((ifa = rt->rt_ifa) && ifa->ifa_rtrequest) {
962 		struct rt_addrinfo info;
963 
964 		bzero((caddr_t)&info, sizeof(info));
965 		info.rti_flags = rt->rt_flags;
966 		info.rti_info[RTAX_DST] = rt_key(rt);
967 		info.rti_info[RTAX_GATEWAY] = rt->rt_gateway;
968 		info.rti_info[RTAX_NETMASK] = rt_mask(rt);
969 		ifa->ifa_rtrequest(RTM_DELETE, rt, &info);
970 	}
971 
972 	/*
973 	 * one more rtentry floating around that is not
974 	 * linked to the routing table.
975 	 */
976 	V_rttrash++;
977 #if !defined(RADIX_MPATH)
978 bad:
979 #endif
980 	return (error);
981 }
982 
983 #if 0
984 int p_sockaddr(char *buf, int buflen, struct sockaddr *s);
985 int rt_print(char *buf, int buflen, struct rtentry *rt);
986 
987 int
988 p_sockaddr(char *buf, int buflen, struct sockaddr *s)
989 {
990 	void *paddr = NULL;
991 
992 	switch (s->sa_family) {
993 	case AF_INET:
994 		paddr = &((struct sockaddr_in *)s)->sin_addr;
995 		break;
996 	case AF_INET6:
997 		paddr = &((struct sockaddr_in6 *)s)->sin6_addr;
998 		break;
999 	}
1000 
1001 	if (paddr == NULL)
1002 		return (0);
1003 
1004 	if (inet_ntop(s->sa_family, paddr, buf, buflen) == NULL)
1005 		return (0);
1006 
1007 	return (strlen(buf));
1008 }
1009 
1010 int
1011 rt_print(char *buf, int buflen, struct rtentry *rt)
1012 {
1013 	struct sockaddr *addr, *mask;
1014 	int i = 0;
1015 
1016 	addr = rt_key(rt);
1017 	mask = rt_mask(rt);
1018 
1019 	i = p_sockaddr(buf, buflen, addr);
1020 	if (!(rt->rt_flags & RTF_HOST)) {
1021 		buf[i++] = '/';
1022 		i += p_sockaddr(buf + i, buflen - i, mask);
1023 	}
1024 
1025 	if (rt->rt_flags & RTF_GATEWAY) {
1026 		buf[i++] = '>';
1027 		i += p_sockaddr(buf + i, buflen - i, rt->rt_gateway);
1028 	}
1029 
1030 	return (i);
1031 }
1032 #endif
1033 
1034 #ifdef RADIX_MPATH
1035 static int
1036 rn_mpath_update(int req, struct rt_addrinfo *info,
1037     struct radix_node_head *rnh, struct rtentry **ret_nrt)
1038 {
1039 	/*
1040 	 * if we got multipath routes, we require users to specify
1041 	 * a matching RTAX_GATEWAY.
1042 	 */
1043 	struct rtentry *rt, *rto = NULL;
1044 	struct radix_node *rn;
1045 	int error = 0;
1046 
1047 	rn = rnh->rnh_lookup(dst, netmask, rnh);
1048 	if (rn == NULL)
1049 		return (ESRCH);
1050 	rto = rt = RNTORT(rn);
1051 
1052 	rt = rt_mpath_matchgate(rt, gateway);
1053 	if (rt == NULL)
1054 		return (ESRCH);
1055 	/*
1056 	 * this is the first entry in the chain
1057 	 */
1058 	if (rto == rt) {
1059 		rn = rn_mpath_next((struct radix_node *)rt);
1060 		/*
1061 		 * there is another entry, now it's active
1062 		 */
1063 		if (rn) {
1064 			rto = RNTORT(rn);
1065 			RT_LOCK(rto);
1066 			rto->rt_flags |= RTF_UP;
1067 			RT_UNLOCK(rto);
1068 		} else if (rt->rt_flags & RTF_GATEWAY) {
1069 			/*
1070 			 * For gateway routes, we need to
1071 			 * make sure that we we are deleting
1072 			 * the correct gateway.
1073 			 * rt_mpath_matchgate() does not
1074 			 * check the case when there is only
1075 			 * one route in the chain.
1076 			 */
1077 			if (gateway &&
1078 			    (rt->rt_gateway->sa_len != gateway->sa_len ||
1079 				memcmp(rt->rt_gateway, gateway, gateway->sa_len)))
1080 				error = ESRCH;
1081 			else {
1082 				/*
1083 				 * remove from tree before returning it
1084 				 * to the caller
1085 				 */
1086 				rn = rnh->rnh_deladdr(dst, netmask, rnh);
1087 				KASSERT(rt == RNTORT(rn), ("radix node disappeared"));
1088 				goto gwdelete;
1089 			}
1090 
1091 		}
1092 		/*
1093 		 * use the normal delete code to remove
1094 		 * the first entry
1095 		 */
1096 		if (req != RTM_DELETE)
1097 			goto nondelete;
1098 
1099 		error = ENOENT;
1100 		goto done;
1101 	}
1102 
1103 	/*
1104 	 * if the entry is 2nd and on up
1105 	 */
1106 	if ((req == RTM_DELETE) && !rt_mpath_deldup(rto, rt))
1107 		panic ("rtrequest1: rt_mpath_deldup");
1108 gwdelete:
1109 	RT_LOCK(rt);
1110 	RT_ADDREF(rt);
1111 	if (req == RTM_DELETE) {
1112 		rt->rt_flags &= ~RTF_UP;
1113 		/*
1114 		 * One more rtentry floating around that is not
1115 		 * linked to the routing table. rttrash will be decremented
1116 		 * when RTFREE(rt) is eventually called.
1117 		 */
1118 		V_rttrash++;
1119 	}
1120 
1121 nondelete:
1122 	if (req != RTM_DELETE)
1123 		panic("unrecognized request %d", req);
1124 
1125 
1126 	/*
1127 	 * If the caller wants it, then it can have it,
1128 	 * but it's up to it to free the rtentry as we won't be
1129 	 * doing it.
1130 	 */
1131 	if (ret_nrt) {
1132 		*ret_nrt = rt;
1133 		RT_UNLOCK(rt);
1134 	} else
1135 		RTFREE_LOCKED(rt);
1136 done:
1137 	return (error);
1138 }
1139 #endif
1140 
1141 int
1142 rtrequest1_fib(int req, struct rt_addrinfo *info, struct rtentry **ret_nrt,
1143 				u_int fibnum)
1144 {
1145 	int error = 0, needlock = 0;
1146 	struct rtentry *rt;
1147 #ifdef FLOWTABLE
1148 	struct rtentry *rt0;
1149 #endif
1150 	struct radix_node *rn;
1151 	struct radix_node_head *rnh;
1152 	struct ifaddr *ifa;
1153 	struct sockaddr *ndst;
1154 	struct sockaddr_storage mdst;
1155 #define senderr(x) { error = x ; goto bad; }
1156 
1157 	KASSERT((fibnum < rt_numfibs), ("rtrequest1_fib: bad fibnum"));
1158 	switch (dst->sa_family) {
1159 	case AF_INET6:
1160 	case AF_INET:
1161 		/* We support multiple FIBs. */
1162 		break;
1163 	default:
1164 		fibnum = RT_DEFAULT_FIB;
1165 		break;
1166 	}
1167 
1168 	/*
1169 	 * Find the correct routing tree to use for this Address Family
1170 	 */
1171 	rnh = rt_tables_get_rnh(fibnum, dst->sa_family);
1172 	if (rnh == NULL)
1173 		return (EAFNOSUPPORT);
1174 	needlock = ((flags & RTF_RNH_LOCKED) == 0);
1175 	flags &= ~RTF_RNH_LOCKED;
1176 	if (needlock)
1177 		RADIX_NODE_HEAD_LOCK(rnh);
1178 	else
1179 		RADIX_NODE_HEAD_LOCK_ASSERT(rnh);
1180 	/*
1181 	 * If we are adding a host route then we don't want to put
1182 	 * a netmask in the tree, nor do we want to clone it.
1183 	 */
1184 	if (flags & RTF_HOST)
1185 		netmask = NULL;
1186 
1187 	switch (req) {
1188 	case RTM_DELETE:
1189 		if (netmask) {
1190 			rt_maskedcopy(dst, (struct sockaddr *)&mdst, netmask);
1191 			dst = (struct sockaddr *)&mdst;
1192 		}
1193 #ifdef RADIX_MPATH
1194 		if (rn_mpath_capable(rnh)) {
1195 			error = rn_mpath_update(req, info, rnh, ret_nrt);
1196 			/*
1197 			 * "bad" holds true for the success case
1198 			 * as well
1199 			 */
1200 			if (error != ENOENT)
1201 				goto bad;
1202 			error = 0;
1203 		}
1204 #endif
1205 		if ((flags & RTF_PINNED) == 0) {
1206 			/* Check if target route can be deleted */
1207 			rt = (struct rtentry *)rnh->rnh_lookup(dst,
1208 			    netmask, rnh);
1209 			if ((rt != NULL) && (rt->rt_flags & RTF_PINNED))
1210 				senderr(EADDRINUSE);
1211 		}
1212 
1213 		/*
1214 		 * Remove the item from the tree and return it.
1215 		 * Complain if it is not there and do no more processing.
1216 		 */
1217 		rn = rnh->rnh_deladdr(dst, netmask, rnh);
1218 		if (rn == NULL)
1219 			senderr(ESRCH);
1220 		if (rn->rn_flags & (RNF_ACTIVE | RNF_ROOT))
1221 			panic ("rtrequest delete");
1222 		rt = RNTORT(rn);
1223 		RT_LOCK(rt);
1224 		RT_ADDREF(rt);
1225 		rt->rt_flags &= ~RTF_UP;
1226 
1227 		/*
1228 		 * give the protocol a chance to keep things in sync.
1229 		 */
1230 		if ((ifa = rt->rt_ifa) && ifa->ifa_rtrequest)
1231 			ifa->ifa_rtrequest(RTM_DELETE, rt, info);
1232 
1233 		/*
1234 		 * One more rtentry floating around that is not
1235 		 * linked to the routing table. rttrash will be decremented
1236 		 * when RTFREE(rt) is eventually called.
1237 		 */
1238 		V_rttrash++;
1239 
1240 		/*
1241 		 * If the caller wants it, then it can have it,
1242 		 * but it's up to it to free the rtentry as we won't be
1243 		 * doing it.
1244 		 */
1245 		if (ret_nrt) {
1246 			*ret_nrt = rt;
1247 			RT_UNLOCK(rt);
1248 		} else
1249 			RTFREE_LOCKED(rt);
1250 		break;
1251 	case RTM_RESOLVE:
1252 		/*
1253 		 * resolve was only used for route cloning
1254 		 * here for compat
1255 		 */
1256 		break;
1257 	case RTM_ADD:
1258 		if ((flags & RTF_GATEWAY) && !gateway)
1259 			senderr(EINVAL);
1260 		if (dst && gateway && (dst->sa_family != gateway->sa_family) &&
1261 		    (gateway->sa_family != AF_UNSPEC) && (gateway->sa_family != AF_LINK))
1262 			senderr(EINVAL);
1263 
1264 		if (info->rti_ifa == NULL) {
1265 			error = rt_getifa_fib(info, fibnum);
1266 			if (error)
1267 				senderr(error);
1268 		} else
1269 			ifa_ref(info->rti_ifa);
1270 		ifa = info->rti_ifa;
1271 		rt = uma_zalloc(V_rtzone, M_NOWAIT);
1272 		if (rt == NULL) {
1273 			ifa_free(ifa);
1274 			senderr(ENOBUFS);
1275 		}
1276 		rt->rt_flags = RTF_UP | flags;
1277 		rt->rt_fibnum = fibnum;
1278 		/*
1279 		 * Add the gateway. Possibly re-malloc-ing the storage for it.
1280 		 */
1281 		RT_LOCK(rt);
1282 		if ((error = rt_setgate(rt, dst, gateway)) != 0) {
1283 			ifa_free(ifa);
1284 			uma_zfree(V_rtzone, rt);
1285 			senderr(error);
1286 		}
1287 
1288 		/*
1289 		 * point to the (possibly newly malloc'd) dest address.
1290 		 */
1291 		ndst = (struct sockaddr *)rt_key(rt);
1292 
1293 		/*
1294 		 * make sure it contains the value we want (masked if needed).
1295 		 */
1296 		if (netmask) {
1297 			rt_maskedcopy(dst, ndst, netmask);
1298 		} else
1299 			bcopy(dst, ndst, dst->sa_len);
1300 
1301 		/*
1302 		 * We use the ifa reference returned by rt_getifa_fib().
1303 		 * This moved from below so that rnh->rnh_addaddr() can
1304 		 * examine the ifa and  ifa->ifa_ifp if it so desires.
1305 		 */
1306 		rt->rt_ifa = ifa;
1307 		rt->rt_ifp = ifa->ifa_ifp;
1308 		rt->rt_weight = 1;
1309 
1310 #ifdef RADIX_MPATH
1311 		/* do not permit exactly the same dst/mask/gw pair */
1312 		if (rn_mpath_capable(rnh) &&
1313 			rt_mpath_conflict(rnh, rt, netmask)) {
1314 			ifa_free(rt->rt_ifa);
1315 			Free(rt_key(rt));
1316 			uma_zfree(V_rtzone, rt);
1317 			senderr(EEXIST);
1318 		}
1319 #endif
1320 
1321 #ifdef FLOWTABLE
1322 		rt0 = NULL;
1323 		/* "flow-table" only supports IPv6 and IPv4 at the moment. */
1324 		switch (dst->sa_family) {
1325 #ifdef INET6
1326 		case AF_INET6:
1327 #endif
1328 #ifdef INET
1329 		case AF_INET:
1330 #endif
1331 #if defined(INET6) || defined(INET)
1332 			rn = rnh->rnh_matchaddr(dst, rnh);
1333 			if (rn && ((rn->rn_flags & RNF_ROOT) == 0)) {
1334 				struct sockaddr *mask;
1335 				u_char *m, *n;
1336 				int len;
1337 
1338 				/*
1339 				 * compare mask to see if the new route is
1340 				 * more specific than the existing one
1341 				 */
1342 				rt0 = RNTORT(rn);
1343 				RT_LOCK(rt0);
1344 				RT_ADDREF(rt0);
1345 				RT_UNLOCK(rt0);
1346 				/*
1347 				 * A host route is already present, so
1348 				 * leave the flow-table entries as is.
1349 				 */
1350 				if (rt0->rt_flags & RTF_HOST) {
1351 					RTFREE(rt0);
1352 					rt0 = NULL;
1353 				} else if (!(flags & RTF_HOST) && netmask) {
1354 					mask = rt_mask(rt0);
1355 					len = mask->sa_len;
1356 					m = (u_char *)mask;
1357 					n = (u_char *)netmask;
1358 					while (len-- > 0) {
1359 						if (*n != *m)
1360 							break;
1361 						n++;
1362 						m++;
1363 					}
1364 					if (len == 0 || (*n < *m)) {
1365 						RTFREE(rt0);
1366 						rt0 = NULL;
1367 					}
1368 				}
1369 			}
1370 #endif/* INET6 || INET */
1371 		}
1372 #endif /* FLOWTABLE */
1373 
1374 		/* XXX mtu manipulation will be done in rnh_addaddr -- itojun */
1375 		rn = rnh->rnh_addaddr(ndst, netmask, rnh, rt->rt_nodes);
1376 		/*
1377 		 * If it still failed to go into the tree,
1378 		 * then un-make it (this should be a function)
1379 		 */
1380 		if (rn == NULL) {
1381 			ifa_free(rt->rt_ifa);
1382 			Free(rt_key(rt));
1383 			uma_zfree(V_rtzone, rt);
1384 #ifdef FLOWTABLE
1385 			if (rt0 != NULL)
1386 				RTFREE(rt0);
1387 #endif
1388 			senderr(EEXIST);
1389 		}
1390 #ifdef FLOWTABLE
1391 		else if (rt0 != NULL) {
1392 			flowtable_route_flush(dst->sa_family, rt0);
1393 			RTFREE(rt0);
1394 		}
1395 #endif
1396 
1397 		/*
1398 		 * If this protocol has something to add to this then
1399 		 * allow it to do that as well.
1400 		 */
1401 		if (ifa->ifa_rtrequest)
1402 			ifa->ifa_rtrequest(req, rt, info);
1403 
1404 		/*
1405 		 * actually return a resultant rtentry and
1406 		 * give the caller a single reference.
1407 		 */
1408 		if (ret_nrt) {
1409 			*ret_nrt = rt;
1410 			RT_ADDREF(rt);
1411 		}
1412 		RT_UNLOCK(rt);
1413 		break;
1414 	case RTM_CHANGE:
1415 		error = rtrequest1_fib_change(rnh, info, ret_nrt, fibnum);
1416 		break;
1417 	default:
1418 		error = EOPNOTSUPP;
1419 	}
1420 bad:
1421 	if (needlock)
1422 		RADIX_NODE_HEAD_UNLOCK(rnh);
1423 	return (error);
1424 #undef senderr
1425 }
1426 
1427 #undef dst
1428 #undef gateway
1429 #undef netmask
1430 #undef ifaaddr
1431 #undef ifpaddr
1432 #undef flags
1433 
1434 #define	senderr(e) { error = e; goto bad; }
1435 static int
1436 rtrequest1_fib_change(struct radix_node_head *rnh, struct rt_addrinfo *info,
1437     struct rtentry **ret_nrt, u_int fibnum)
1438 {
1439 	struct rtentry *rt = NULL;
1440 	int error = 0;
1441 	int free_ifa = 0;
1442 
1443 	rt = (struct rtentry *)rnh->rnh_lookup(info->rti_info[RTAX_DST],
1444 	    info->rti_info[RTAX_NETMASK], rnh);
1445 
1446 	if (rt == NULL)
1447 		return (ESRCH);
1448 
1449 #ifdef RADIX_MPATH
1450 	/*
1451 	 * If we got multipath routes,
1452 	 * we require users to specify a matching RTAX_GATEWAY.
1453 	 */
1454 	if (rn_mpath_capable(rnh)) {
1455 		rt = rt_mpath_matchgate(rt, info->rti_info[RTAX_GATEWAY]);
1456 		if (rt == NULL)
1457 			return (ESRCH);
1458 	}
1459 #endif
1460 
1461 	RT_LOCK(rt);
1462 
1463 	/*
1464 	 * New gateway could require new ifaddr, ifp;
1465 	 * flags may also be different; ifp may be specified
1466 	 * by ll sockaddr when protocol address is ambiguous
1467 	 */
1468 	if (((rt->rt_flags & RTF_GATEWAY) &&
1469 	    info->rti_info[RTAX_GATEWAY] != NULL) ||
1470 	    info->rti_info[RTAX_IFP] != NULL ||
1471 	    (info->rti_info[RTAX_IFA] != NULL &&
1472 	     !sa_equal(info->rti_info[RTAX_IFA], rt->rt_ifa->ifa_addr))) {
1473 
1474 		error = rt_getifa_fib(info, fibnum);
1475 		if (info->rti_ifa != NULL)
1476 			free_ifa = 1;
1477 
1478 		if (error != 0)
1479 			senderr(error);
1480 	}
1481 
1482 	/* Check if outgoing interface has changed */
1483 	if (info->rti_ifa != NULL && info->rti_ifa != rt->rt_ifa &&
1484 	    rt->rt_ifa != NULL && rt->rt_ifa->ifa_rtrequest != NULL) {
1485 		rt->rt_ifa->ifa_rtrequest(RTM_DELETE, rt, info);
1486 		ifa_free(rt->rt_ifa);
1487 	}
1488 	/* Update gateway address */
1489 	if (info->rti_info[RTAX_GATEWAY] != NULL) {
1490 		error = rt_setgate(rt, rt_key(rt), info->rti_info[RTAX_GATEWAY]);
1491 		if (error != 0)
1492 			senderr(error);
1493 
1494 		rt->rt_flags &= ~RTF_GATEWAY;
1495 		rt->rt_flags |= (RTF_GATEWAY & info->rti_flags);
1496 	}
1497 
1498 	if (info->rti_ifa != NULL && info->rti_ifa != rt->rt_ifa) {
1499 		ifa_ref(info->rti_ifa);
1500 		rt->rt_ifa = info->rti_ifa;
1501 		rt->rt_ifp = info->rti_ifp;
1502 	}
1503 	/* Allow some flags to be toggled on change. */
1504 	rt->rt_flags &= ~RTF_FMASK;
1505 	rt->rt_flags |= info->rti_flags & RTF_FMASK;
1506 
1507 	if (rt->rt_ifa && rt->rt_ifa->ifa_rtrequest != NULL)
1508 	       rt->rt_ifa->ifa_rtrequest(RTM_ADD, rt, info);
1509 
1510 	if (ret_nrt) {
1511 		*ret_nrt = rt;
1512 		RT_ADDREF(rt);
1513 	}
1514 bad:
1515 	RT_UNLOCK(rt);
1516 	if (free_ifa != 0)
1517 		ifa_free(info->rti_ifa);
1518 	return (error);
1519 }
1520 #undef senderr
1521 
1522 
1523 int
1524 rt_setgate(struct rtentry *rt, struct sockaddr *dst, struct sockaddr *gate)
1525 {
1526 	/* XXX dst may be overwritten, can we move this to below */
1527 	int dlen = SA_SIZE(dst), glen = SA_SIZE(gate);
1528 #ifdef INVARIANTS
1529 	struct radix_node_head *rnh;
1530 
1531 	rnh = rt_tables_get_rnh(rt->rt_fibnum, dst->sa_family);
1532 #endif
1533 
1534 	RT_LOCK_ASSERT(rt);
1535 	RADIX_NODE_HEAD_LOCK_ASSERT(rnh);
1536 
1537 	/*
1538 	 * Prepare to store the gateway in rt->rt_gateway.
1539 	 * Both dst and gateway are stored one after the other in the same
1540 	 * malloc'd chunk. If we have room, we can reuse the old buffer,
1541 	 * rt_gateway already points to the right place.
1542 	 * Otherwise, malloc a new block and update the 'dst' address.
1543 	 */
1544 	if (rt->rt_gateway == NULL || glen > SA_SIZE(rt->rt_gateway)) {
1545 		caddr_t new;
1546 
1547 		R_Malloc(new, caddr_t, dlen + glen);
1548 		if (new == NULL)
1549 			return ENOBUFS;
1550 		/*
1551 		 * XXX note, we copy from *dst and not *rt_key(rt) because
1552 		 * rt_setgate() can be called to initialize a newly
1553 		 * allocated route entry, in which case rt_key(rt) == NULL
1554 		 * (and also rt->rt_gateway == NULL).
1555 		 * Free()/free() handle a NULL argument just fine.
1556 		 */
1557 		bcopy(dst, new, dlen);
1558 		Free(rt_key(rt));	/* free old block, if any */
1559 		rt_key(rt) = (struct sockaddr *)new;
1560 		rt->rt_gateway = (struct sockaddr *)(new + dlen);
1561 	}
1562 
1563 	/*
1564 	 * Copy the new gateway value into the memory chunk.
1565 	 */
1566 	bcopy(gate, rt->rt_gateway, glen);
1567 
1568 	return (0);
1569 }
1570 
1571 void
1572 rt_maskedcopy(struct sockaddr *src, struct sockaddr *dst, struct sockaddr *netmask)
1573 {
1574 	u_char *cp1 = (u_char *)src;
1575 	u_char *cp2 = (u_char *)dst;
1576 	u_char *cp3 = (u_char *)netmask;
1577 	u_char *cplim = cp2 + *cp3;
1578 	u_char *cplim2 = cp2 + *cp1;
1579 
1580 	*cp2++ = *cp1++; *cp2++ = *cp1++; /* copies sa_len & sa_family */
1581 	cp3 += 2;
1582 	if (cplim > cplim2)
1583 		cplim = cplim2;
1584 	while (cp2 < cplim)
1585 		*cp2++ = *cp1++ & *cp3++;
1586 	if (cp2 < cplim2)
1587 		bzero((caddr_t)cp2, (unsigned)(cplim2 - cp2));
1588 }
1589 
1590 /*
1591  * Set up a routing table entry, normally
1592  * for an interface.
1593  */
1594 #define _SOCKADDR_TMPSIZE 128 /* Not too big.. kernel stack size is limited */
1595 static inline  int
1596 rtinit1(struct ifaddr *ifa, int cmd, int flags, int fibnum)
1597 {
1598 	struct sockaddr *dst;
1599 	struct sockaddr *netmask;
1600 	struct rtentry *rt = NULL;
1601 	struct rt_addrinfo info;
1602 	int error = 0;
1603 	int startfib, endfib;
1604 	char tempbuf[_SOCKADDR_TMPSIZE];
1605 	int didwork = 0;
1606 	int a_failure = 0;
1607 	static struct sockaddr_dl null_sdl = {sizeof(null_sdl), AF_LINK};
1608 	struct radix_node_head *rnh;
1609 
1610 	if (flags & RTF_HOST) {
1611 		dst = ifa->ifa_dstaddr;
1612 		netmask = NULL;
1613 	} else {
1614 		dst = ifa->ifa_addr;
1615 		netmask = ifa->ifa_netmask;
1616 	}
1617 	if (dst->sa_len == 0)
1618 		return(EINVAL);
1619 	switch (dst->sa_family) {
1620 	case AF_INET6:
1621 	case AF_INET:
1622 		/* We support multiple FIBs. */
1623 		break;
1624 	default:
1625 		fibnum = RT_DEFAULT_FIB;
1626 		break;
1627 	}
1628 	if (fibnum == RT_ALL_FIBS) {
1629 		if (rt_add_addr_allfibs == 0 && cmd == (int)RTM_ADD) {
1630 			startfib = endfib = ifa->ifa_ifp->if_fib;
1631 		} else {
1632 			startfib = 0;
1633 			endfib = rt_numfibs - 1;
1634 		}
1635 	} else {
1636 		KASSERT((fibnum < rt_numfibs), ("rtinit1: bad fibnum"));
1637 		startfib = fibnum;
1638 		endfib = fibnum;
1639 	}
1640 
1641 	/*
1642 	 * If it's a delete, check that if it exists,
1643 	 * it's on the correct interface or we might scrub
1644 	 * a route to another ifa which would
1645 	 * be confusing at best and possibly worse.
1646 	 */
1647 	if (cmd == RTM_DELETE) {
1648 		/*
1649 		 * It's a delete, so it should already exist..
1650 		 * If it's a net, mask off the host bits
1651 		 * (Assuming we have a mask)
1652 		 * XXX this is kinda inet specific..
1653 		 */
1654 		if (netmask != NULL) {
1655 			rt_maskedcopy(dst, (struct sockaddr *)tempbuf, netmask);
1656 			dst = (struct sockaddr *)tempbuf;
1657 		}
1658 	}
1659 	/*
1660 	 * Now go through all the requested tables (fibs) and do the
1661 	 * requested action. Realistically, this will either be fib 0
1662 	 * for protocols that don't do multiple tables or all the
1663 	 * tables for those that do.
1664 	 */
1665 	for ( fibnum = startfib; fibnum <= endfib; fibnum++) {
1666 		if (cmd == RTM_DELETE) {
1667 			struct radix_node *rn;
1668 			/*
1669 			 * Look up an rtentry that is in the routing tree and
1670 			 * contains the correct info.
1671 			 */
1672 			rnh = rt_tables_get_rnh(fibnum, dst->sa_family);
1673 			if (rnh == NULL)
1674 				/* this table doesn't exist but others might */
1675 				continue;
1676 			RADIX_NODE_HEAD_RLOCK(rnh);
1677 			rn = rnh->rnh_lookup(dst, netmask, rnh);
1678 #ifdef RADIX_MPATH
1679 			if (rn_mpath_capable(rnh)) {
1680 
1681 				if (rn == NULL)
1682 					error = ESRCH;
1683 				else {
1684 					rt = RNTORT(rn);
1685 					/*
1686 					 * for interface route the
1687 					 * rt->rt_gateway is sockaddr_intf
1688 					 * for cloning ARP entries, so
1689 					 * rt_mpath_matchgate must use the
1690 					 * interface address
1691 					 */
1692 					rt = rt_mpath_matchgate(rt,
1693 					    ifa->ifa_addr);
1694 					if (rt == NULL)
1695 						error = ESRCH;
1696 				}
1697 			}
1698 #endif
1699 			error = (rn == NULL ||
1700 			    (rn->rn_flags & RNF_ROOT) ||
1701 			    RNTORT(rn)->rt_ifa != ifa);
1702 			RADIX_NODE_HEAD_RUNLOCK(rnh);
1703 			if (error) {
1704 				/* this is only an error if bad on ALL tables */
1705 				continue;
1706 			}
1707 		}
1708 		/*
1709 		 * Do the actual request
1710 		 */
1711 		bzero((caddr_t)&info, sizeof(info));
1712 		info.rti_ifa = ifa;
1713 		info.rti_flags = flags |
1714 		    (ifa->ifa_flags & ~IFA_RTSELF) | RTF_PINNED;
1715 		info.rti_info[RTAX_DST] = dst;
1716 		/*
1717 		 * doing this for compatibility reasons
1718 		 */
1719 		if (cmd == RTM_ADD)
1720 			info.rti_info[RTAX_GATEWAY] =
1721 			    (struct sockaddr *)&null_sdl;
1722 		else
1723 			info.rti_info[RTAX_GATEWAY] = ifa->ifa_addr;
1724 		info.rti_info[RTAX_NETMASK] = netmask;
1725 		error = rtrequest1_fib(cmd, &info, &rt, fibnum);
1726 
1727 		if ((error == EEXIST) && (cmd == RTM_ADD)) {
1728 			/*
1729 			 * Interface route addition failed.
1730 			 * Atomically delete current prefix generating
1731 			 * RTM_DELETE message, and retry adding
1732 			 * interface prefix.
1733 			 */
1734 			rnh = rt_tables_get_rnh(fibnum, dst->sa_family);
1735 			RADIX_NODE_HEAD_LOCK(rnh);
1736 
1737 			/* Delete old prefix */
1738 			info.rti_ifa = NULL;
1739 			info.rti_flags = RTF_RNH_LOCKED;
1740 
1741 			error = rtrequest1_fib(RTM_DELETE, &info, NULL, fibnum);
1742 			if (error == 0) {
1743 				info.rti_ifa = ifa;
1744 				info.rti_flags = flags | RTF_RNH_LOCKED |
1745 				    (ifa->ifa_flags & ~IFA_RTSELF) | RTF_PINNED;
1746 				error = rtrequest1_fib(cmd, &info, &rt, fibnum);
1747 			}
1748 
1749 			RADIX_NODE_HEAD_UNLOCK(rnh);
1750 		}
1751 
1752 
1753 		if (error == 0 && rt != NULL) {
1754 			/*
1755 			 * notify any listening routing agents of the change
1756 			 */
1757 			RT_LOCK(rt);
1758 #ifdef RADIX_MPATH
1759 			/*
1760 			 * in case address alias finds the first address
1761 			 * e.g. ifconfig bge0 192.0.2.246/24
1762 			 * e.g. ifconfig bge0 192.0.2.247/24
1763 			 * the address set in the route is 192.0.2.246
1764 			 * so we need to replace it with 192.0.2.247
1765 			 */
1766 			if (memcmp(rt->rt_ifa->ifa_addr,
1767 			    ifa->ifa_addr, ifa->ifa_addr->sa_len)) {
1768 				ifa_free(rt->rt_ifa);
1769 				ifa_ref(ifa);
1770 				rt->rt_ifp = ifa->ifa_ifp;
1771 				rt->rt_ifa = ifa;
1772 			}
1773 #endif
1774 			/*
1775 			 * doing this for compatibility reasons
1776 			 */
1777 			if (cmd == RTM_ADD) {
1778 			    ((struct sockaddr_dl *)rt->rt_gateway)->sdl_type  =
1779 				rt->rt_ifp->if_type;
1780 			    ((struct sockaddr_dl *)rt->rt_gateway)->sdl_index =
1781 				rt->rt_ifp->if_index;
1782 			}
1783 			RT_ADDREF(rt);
1784 			RT_UNLOCK(rt);
1785 			rt_newaddrmsg_fib(cmd, ifa, error, rt, fibnum);
1786 			RT_LOCK(rt);
1787 			RT_REMREF(rt);
1788 			if (cmd == RTM_DELETE) {
1789 				/*
1790 				 * If we are deleting, and we found an entry,
1791 				 * then it's been removed from the tree..
1792 				 * now throw it away.
1793 				 */
1794 				RTFREE_LOCKED(rt);
1795 			} else {
1796 				if (cmd == RTM_ADD) {
1797 					/*
1798 					 * We just wanted to add it..
1799 					 * we don't actually need a reference.
1800 					 */
1801 					RT_REMREF(rt);
1802 				}
1803 				RT_UNLOCK(rt);
1804 			}
1805 			didwork = 1;
1806 		}
1807 		if (error)
1808 			a_failure = error;
1809 	}
1810 	if (cmd == RTM_DELETE) {
1811 		if (didwork) {
1812 			error = 0;
1813 		} else {
1814 			/* we only give an error if it wasn't in any table */
1815 			error = ((flags & RTF_HOST) ?
1816 			    EHOSTUNREACH : ENETUNREACH);
1817 		}
1818 	} else {
1819 		if (a_failure) {
1820 			/* return an error if any of them failed */
1821 			error = a_failure;
1822 		}
1823 	}
1824 	return (error);
1825 }
1826 
1827 /*
1828  * Set up a routing table entry, normally
1829  * for an interface.
1830  */
1831 int
1832 rtinit(struct ifaddr *ifa, int cmd, int flags)
1833 {
1834 	struct sockaddr *dst;
1835 	int fib = RT_DEFAULT_FIB;
1836 
1837 	if (flags & RTF_HOST) {
1838 		dst = ifa->ifa_dstaddr;
1839 	} else {
1840 		dst = ifa->ifa_addr;
1841 	}
1842 
1843 	switch (dst->sa_family) {
1844 	case AF_INET6:
1845 	case AF_INET:
1846 		/* We do support multiple FIBs. */
1847 		fib = RT_ALL_FIBS;
1848 		break;
1849 	}
1850 	return (rtinit1(ifa, cmd, flags, fib));
1851 }
1852 
1853 /*
1854  * Announce interface address arrival/withdraw
1855  * Returns 0 on success.
1856  */
1857 int
1858 rt_addrmsg(int cmd, struct ifaddr *ifa, int fibnum)
1859 {
1860 
1861 	KASSERT(cmd == RTM_ADD || cmd == RTM_DELETE,
1862 	    ("unexpected cmd %d", cmd));
1863 
1864 	KASSERT(fibnum == RT_ALL_FIBS || (fibnum >= 0 && fibnum < rt_numfibs),
1865 	    ("%s: fib out of range 0 <=%d<%d", __func__, fibnum, rt_numfibs));
1866 
1867 #if defined(INET) || defined(INET6)
1868 #ifdef SCTP
1869 	/*
1870 	 * notify the SCTP stack
1871 	 * this will only get called when an address is added/deleted
1872 	 * XXX pass the ifaddr struct instead if ifa->ifa_addr...
1873 	 */
1874 	sctp_addr_change(ifa, cmd);
1875 #endif /* SCTP */
1876 #endif
1877 	return (rtsock_addrmsg(cmd, ifa, fibnum));
1878 }
1879 
1880 /*
1881  * Announce route addition/removal.
1882  * Users of this function MUST validate input data BEFORE calling.
1883  * However we have to be able to handle invalid data:
1884  * if some userland app sends us "invalid" route message (invalid mask,
1885  * no dst, wrong address families, etc...) we need to pass it back
1886  * to app (and any other rtsock consumers) with rtm_errno field set to
1887  * non-zero value.
1888  * Returns 0 on success.
1889  */
1890 int
1891 rt_routemsg(int cmd, struct ifnet *ifp, int error, struct rtentry *rt,
1892     int fibnum)
1893 {
1894 
1895 	KASSERT(cmd == RTM_ADD || cmd == RTM_DELETE,
1896 	    ("unexpected cmd %d", cmd));
1897 
1898 	KASSERT(fibnum == RT_ALL_FIBS || (fibnum >= 0 && fibnum < rt_numfibs),
1899 	    ("%s: fib out of range 0 <=%d<%d", __func__, fibnum, rt_numfibs));
1900 
1901 	KASSERT(rt_key(rt) != NULL, (":%s: rt_key must be supplied", __func__));
1902 
1903 	return (rtsock_routemsg(cmd, ifp, error, rt, fibnum));
1904 }
1905 
1906 void
1907 rt_newaddrmsg(int cmd, struct ifaddr *ifa, int error, struct rtentry *rt)
1908 {
1909 
1910 	rt_newaddrmsg_fib(cmd, ifa, error, rt, RT_ALL_FIBS);
1911 }
1912 
1913 /*
1914  * This is called to generate messages from the routing socket
1915  * indicating a network interface has had addresses associated with it.
1916  */
1917 void
1918 rt_newaddrmsg_fib(int cmd, struct ifaddr *ifa, int error, struct rtentry *rt,
1919     int fibnum)
1920 {
1921 
1922 	KASSERT(cmd == RTM_ADD || cmd == RTM_DELETE,
1923 		("unexpected cmd %u", cmd));
1924 	KASSERT(fibnum == RT_ALL_FIBS || (fibnum >= 0 && fibnum < rt_numfibs),
1925 	    ("%s: fib out of range 0 <=%d<%d", __func__, fibnum, rt_numfibs));
1926 
1927 	if (cmd == RTM_ADD) {
1928 		rt_addrmsg(cmd, ifa, fibnum);
1929 		if (rt != NULL)
1930 			rt_routemsg(cmd, ifa->ifa_ifp, error, rt, fibnum);
1931 	} else {
1932 		if (rt != NULL)
1933 			rt_routemsg(cmd, ifa->ifa_ifp, error, rt, fibnum);
1934 		rt_addrmsg(cmd, ifa, fibnum);
1935 	}
1936 }
1937 
1938