xref: /freebsd/sys/net/route.c (revision 4ed925457ab06e83238a5db33e89ccc94b99a713)
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_route.h"
39 #include "opt_mrouting.h"
40 #include "opt_mpath.h"
41 
42 #include <sys/param.h>
43 #include <sys/systm.h>
44 #include <sys/syslog.h>
45 #include <sys/malloc.h>
46 #include <sys/mbuf.h>
47 #include <sys/socket.h>
48 #include <sys/sysctl.h>
49 #include <sys/syslog.h>
50 #include <sys/sysproto.h>
51 #include <sys/proc.h>
52 #include <sys/domain.h>
53 #include <sys/kernel.h>
54 
55 #include <net/if.h>
56 #include <net/if_dl.h>
57 #include <net/route.h>
58 #include <net/vnet.h>
59 #include <net/flowtable.h>
60 
61 #ifdef RADIX_MPATH
62 #include <net/radix_mpath.h>
63 #endif
64 
65 #include <netinet/in.h>
66 #include <netinet/ip_mroute.h>
67 
68 #include <vm/uma.h>
69 
70 u_int rt_numfibs = RT_NUMFIBS;
71 SYSCTL_INT(_net, OID_AUTO, fibs, CTLFLAG_RD, &rt_numfibs, 0, "");
72 /*
73  * Allow the boot code to allow LESS than RT_MAXFIBS to be used.
74  * We can't do more because storage is statically allocated for now.
75  * (for compatibility reasons.. this will change).
76  */
77 TUNABLE_INT("net.fibs", &rt_numfibs);
78 
79 /*
80  * By default add routes to all fibs for new interfaces.
81  * Once this is set to 0 then only allocate routes on interface
82  * changes for the FIB of the caller when adding a new set of addresses
83  * to an interface.  XXX this is a shotgun aproach to a problem that needs
84  * a more fine grained solution.. that will come.
85  */
86 u_int rt_add_addr_allfibs = 1;
87 SYSCTL_INT(_net, OID_AUTO, add_addr_allfibs, CTLFLAG_RW,
88     &rt_add_addr_allfibs, 0, "");
89 TUNABLE_INT("net.add_addr_allfibs", &rt_add_addr_allfibs);
90 
91 VNET_DEFINE(struct radix_node_head *, rt_tables);
92 static VNET_DEFINE(uma_zone_t, rtzone);		/* Routing table UMA zone. */
93 VNET_DEFINE(int, rttrash);		/* routes not in table but not freed */
94 VNET_DEFINE(struct rtstat, rtstat);
95 
96 #define	V_rt_tables	VNET(rt_tables)
97 #define	V_rtzone	VNET(rtzone)
98 #define	V_rttrash	VNET(rttrash)
99 #define	V_rtstat	VNET(rtstat)
100 
101 
102 /* compare two sockaddr structures */
103 #define	sa_equal(a1, a2) (bcmp((a1), (a2), (a1)->sa_len) == 0)
104 
105 /*
106  * Convert a 'struct radix_node *' to a 'struct rtentry *'.
107  * The operation can be done safely (in this code) because a
108  * 'struct rtentry' starts with two 'struct radix_node''s, the first
109  * one representing leaf nodes in the routing tree, which is
110  * what the code in radix.c passes us as a 'struct radix_node'.
111  *
112  * But because there are a lot of assumptions in this conversion,
113  * do not cast explicitly, but always use the macro below.
114  */
115 #define RNTORT(p)	((struct rtentry *)(p))
116 
117 #if 0
118 /* default fib for tunnels to use */
119 u_int tunnel_fib = 0;
120 SYSCTL_INT(_net, OID_AUTO, tunnelfib, CTLFLAG_RD, &tunnel_fib, 0, "");
121 #endif
122 
123 /*
124  * handler for net.my_fibnum
125  */
126 static int
127 sysctl_my_fibnum(SYSCTL_HANDLER_ARGS)
128 {
129         int fibnum;
130         int error;
131 
132         fibnum = curthread->td_proc->p_fibnum;
133         error = sysctl_handle_int(oidp, &fibnum, 0, req);
134         return (error);
135 }
136 
137 SYSCTL_PROC(_net, OID_AUTO, my_fibnum, CTLTYPE_INT|CTLFLAG_RD,
138             NULL, 0, &sysctl_my_fibnum, "I", "default FIB of caller");
139 
140 static __inline struct radix_node_head **
141 rt_tables_get_rnh_ptr(int table, int fam)
142 {
143 	struct radix_node_head **rnh;
144 
145 	KASSERT(table >= 0 && table < rt_numfibs, ("%s: table out of bounds.",
146 	    __func__));
147 	KASSERT(fam >= 0 && fam < (AF_MAX+1), ("%s: fam out of bounds.",
148 	    __func__));
149 
150 	/* rnh is [fib=0][af=0]. */
151 	rnh = (struct radix_node_head **)V_rt_tables;
152 	/* Get the offset to the requested table and fam. */
153 	rnh += table * (AF_MAX+1) + fam;
154 
155 	return (rnh);
156 }
157 
158 struct radix_node_head *
159 rt_tables_get_rnh(int table, int fam)
160 {
161 
162 	return (*rt_tables_get_rnh_ptr(table, fam));
163 }
164 
165 /*
166  * route initialization must occur before ip6_init2(), which happenas at
167  * SI_ORDER_MIDDLE.
168  */
169 static void
170 route_init(void)
171 {
172 	struct domain *dom;
173 	int max_keylen = 0;
174 
175 	/* whack the tunable ints into  line. */
176 	if (rt_numfibs > RT_MAXFIBS)
177 		rt_numfibs = RT_MAXFIBS;
178 	if (rt_numfibs == 0)
179 		rt_numfibs = 1;
180 
181 	for (dom = domains; dom; dom = dom->dom_next)
182 		if (dom->dom_maxrtkey > max_keylen)
183 			max_keylen = dom->dom_maxrtkey;
184 
185 	rn_init(max_keylen);	/* init all zeroes, all ones, mask table */
186 }
187 SYSINIT(route_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_THIRD, route_init, 0);
188 
189 static void
190 vnet_route_init(const void *unused __unused)
191 {
192 	struct domain *dom;
193 	struct radix_node_head **rnh;
194 	int table;
195 	int fam;
196 
197 	V_rt_tables = malloc(rt_numfibs * (AF_MAX+1) *
198 	    sizeof(struct radix_node_head *), M_RTABLE, M_WAITOK|M_ZERO);
199 
200 	V_rtzone = uma_zcreate("rtentry", sizeof(struct rtentry), NULL, NULL,
201 	    NULL, NULL, UMA_ALIGN_PTR, 0);
202 	for (dom = domains; dom; dom = dom->dom_next) {
203 		if (dom->dom_rtattach)  {
204 			for  (table = 0; table < rt_numfibs; table++) {
205 				if ( (fam = dom->dom_family) == AF_INET ||
206 				    table == 0) {
207  			        	/* for now only AF_INET has > 1 table */
208 					/* XXX MRT
209 					 * rtattach will be also called
210 					 * from vfs_export.c but the
211 					 * offset will be 0
212 					 * (only for AF_INET and AF_INET6
213 					 * which don't need it anyhow)
214 					 */
215 					rnh = rt_tables_get_rnh_ptr(table, fam);
216 					if (rnh == NULL)
217 						panic("%s: rnh NULL", __func__);
218 					dom->dom_rtattach((void **)rnh,
219 				    	    dom->dom_rtoffset);
220 				} else {
221 					break;
222 				}
223 			}
224 		}
225 	}
226 }
227 VNET_SYSINIT(vnet_route_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_FOURTH,
228     vnet_route_init, 0);
229 
230 #ifdef VIMAGE
231 static void
232 vnet_route_uninit(const void *unused __unused)
233 {
234 	int table;
235 	int fam;
236 	struct domain *dom;
237 	struct radix_node_head **rnh;
238 
239 	for (dom = domains; dom; dom = dom->dom_next) {
240 		if (dom->dom_rtdetach) {
241 			for (table = 0; table < rt_numfibs; table++) {
242 				if ( (fam = dom->dom_family) == AF_INET ||
243 				    table == 0) {
244 					/* For now only AF_INET has > 1 tbl. */
245 					rnh = rt_tables_get_rnh_ptr(table, fam);
246 					if (rnh == NULL)
247 						panic("%s: rnh NULL", __func__);
248 					dom->dom_rtdetach((void **)rnh,
249 					    dom->dom_rtoffset);
250 				} else {
251 					break;
252 				}
253 			}
254 		}
255 	}
256 }
257 VNET_SYSUNINIT(vnet_route_uninit, SI_SUB_PROTO_DOMAIN, SI_ORDER_THIRD,
258     vnet_route_uninit, 0);
259 #endif
260 
261 #ifndef _SYS_SYSPROTO_H_
262 struct setfib_args {
263 	int     fibnum;
264 };
265 #endif
266 int
267 setfib(struct thread *td, struct setfib_args *uap)
268 {
269 	if (uap->fibnum < 0 || uap->fibnum >= rt_numfibs)
270 		return EINVAL;
271 	td->td_proc->p_fibnum = uap->fibnum;
272 	return (0);
273 }
274 
275 /*
276  * Packet routing routines.
277  */
278 void
279 rtalloc(struct route *ro)
280 {
281 	rtalloc_ign_fib(ro, 0UL, 0);
282 }
283 
284 void
285 rtalloc_fib(struct route *ro, u_int fibnum)
286 {
287 	rtalloc_ign_fib(ro, 0UL, fibnum);
288 }
289 
290 void
291 rtalloc_ign(struct route *ro, u_long ignore)
292 {
293 	struct rtentry *rt;
294 
295 	if ((rt = ro->ro_rt) != NULL) {
296 		if (rt->rt_ifp != NULL && rt->rt_flags & RTF_UP)
297 			return;
298 		RTFREE(rt);
299 		ro->ro_rt = NULL;
300 	}
301 	ro->ro_rt = rtalloc1_fib(&ro->ro_dst, 1, ignore, 0);
302 	if (ro->ro_rt)
303 		RT_UNLOCK(ro->ro_rt);
304 }
305 
306 void
307 rtalloc_ign_fib(struct route *ro, u_long ignore, u_int fibnum)
308 {
309 	struct rtentry *rt;
310 
311 	if ((rt = ro->ro_rt) != NULL) {
312 		if (rt->rt_ifp != NULL && rt->rt_flags & RTF_UP)
313 			return;
314 		RTFREE(rt);
315 		ro->ro_rt = NULL;
316 	}
317 	ro->ro_rt = rtalloc1_fib(&ro->ro_dst, 1, ignore, fibnum);
318 	if (ro->ro_rt)
319 		RT_UNLOCK(ro->ro_rt);
320 }
321 
322 /*
323  * Look up the route that matches the address given
324  * Or, at least try.. Create a cloned route if needed.
325  *
326  * The returned route, if any, is locked.
327  */
328 struct rtentry *
329 rtalloc1(struct sockaddr *dst, int report, u_long ignflags)
330 {
331 	return (rtalloc1_fib(dst, report, ignflags, 0));
332 }
333 
334 struct rtentry *
335 rtalloc1_fib(struct sockaddr *dst, int report, u_long ignflags,
336 		    u_int fibnum)
337 {
338 	struct radix_node_head *rnh;
339 	struct rtentry *rt;
340 	struct radix_node *rn;
341 	struct rtentry *newrt;
342 	struct rt_addrinfo info;
343 	int err = 0, msgtype = RTM_MISS;
344 	int needlock;
345 
346 	KASSERT((fibnum < rt_numfibs), ("rtalloc1_fib: bad fibnum"));
347 	if (dst->sa_family != AF_INET)	/* Only INET supports > 1 fib now */
348 		fibnum = 0;
349 	rnh = rt_tables_get_rnh(fibnum, dst->sa_family);
350 	newrt = NULL;
351 	/*
352 	 * Look up the address in the table for that Address Family
353 	 */
354 	if (rnh == NULL) {
355 		V_rtstat.rts_unreach++;
356 		goto miss;
357 	}
358 	needlock = !(ignflags & RTF_RNH_LOCKED);
359 	if (needlock)
360 		RADIX_NODE_HEAD_RLOCK(rnh);
361 #ifdef INVARIANTS
362 	else
363 		RADIX_NODE_HEAD_LOCK_ASSERT(rnh);
364 #endif
365 	rn = rnh->rnh_matchaddr(dst, rnh);
366 	if (rn && ((rn->rn_flags & RNF_ROOT) == 0)) {
367 		newrt = rt = RNTORT(rn);
368 		RT_LOCK(newrt);
369 		RT_ADDREF(newrt);
370 		if (needlock)
371 			RADIX_NODE_HEAD_RUNLOCK(rnh);
372 		goto done;
373 
374 	} else if (needlock)
375 		RADIX_NODE_HEAD_RUNLOCK(rnh);
376 
377 	/*
378 	 * Either we hit the root or couldn't find any match,
379 	 * Which basically means
380 	 * "caint get there frm here"
381 	 */
382 	V_rtstat.rts_unreach++;
383 miss:
384 	if (report) {
385 		/*
386 		 * If required, report the failure to the supervising
387 		 * Authorities.
388 		 * For a delete, this is not an error. (report == 0)
389 		 */
390 		bzero(&info, sizeof(info));
391 		info.rti_info[RTAX_DST] = dst;
392 		rt_missmsg(msgtype, &info, 0, err);
393 	}
394 done:
395 	if (newrt)
396 		RT_LOCK_ASSERT(newrt);
397 	return (newrt);
398 }
399 
400 /*
401  * Remove a reference count from an rtentry.
402  * If the count gets low enough, take it out of the routing table
403  */
404 void
405 rtfree(struct rtentry *rt)
406 {
407 	struct radix_node_head *rnh;
408 
409 	KASSERT(rt != NULL,("%s: NULL rt", __func__));
410 	rnh = rt_tables_get_rnh(rt->rt_fibnum, rt_key(rt)->sa_family);
411 	KASSERT(rnh != NULL,("%s: NULL rnh", __func__));
412 
413 	RT_LOCK_ASSERT(rt);
414 
415 	/*
416 	 * The callers should use RTFREE_LOCKED() or RTFREE(), so
417 	 * we should come here exactly with the last reference.
418 	 */
419 	RT_REMREF(rt);
420 	if (rt->rt_refcnt > 0) {
421 		log(LOG_DEBUG, "%s: %p has %d refs\n", __func__, rt, rt->rt_refcnt);
422 		goto done;
423 	}
424 
425 	/*
426 	 * On last reference give the "close method" a chance
427 	 * to cleanup private state.  This also permits (for
428 	 * IPv4 and IPv6) a chance to decide if the routing table
429 	 * entry should be purged immediately or at a later time.
430 	 * When an immediate purge is to happen the close routine
431 	 * typically calls rtexpunge which clears the RTF_UP flag
432 	 * on the entry so that the code below reclaims the storage.
433 	 */
434 	if (rt->rt_refcnt == 0 && rnh->rnh_close)
435 		rnh->rnh_close((struct radix_node *)rt, rnh);
436 
437 	/*
438 	 * If we are no longer "up" (and ref == 0)
439 	 * then we can free the resources associated
440 	 * with the route.
441 	 */
442 	if ((rt->rt_flags & RTF_UP) == 0) {
443 		if (rt->rt_nodes->rn_flags & (RNF_ACTIVE | RNF_ROOT))
444 			panic("rtfree 2");
445 		/*
446 		 * the rtentry must have been removed from the routing table
447 		 * so it is represented in rttrash.. remove that now.
448 		 */
449 		V_rttrash--;
450 #ifdef	DIAGNOSTIC
451 		if (rt->rt_refcnt < 0) {
452 			printf("rtfree: %p not freed (neg refs)\n", rt);
453 			goto done;
454 		}
455 #endif
456 		/*
457 		 * release references on items we hold them on..
458 		 * e.g other routes and ifaddrs.
459 		 */
460 		if (rt->rt_ifa)
461 			ifa_free(rt->rt_ifa);
462 		/*
463 		 * The key is separatly alloc'd so free it (see rt_setgate()).
464 		 * This also frees the gateway, as they are always malloc'd
465 		 * together.
466 		 */
467 		Free(rt_key(rt));
468 
469 		/*
470 		 * and the rtentry itself of course
471 		 */
472 		RT_LOCK_DESTROY(rt);
473 		uma_zfree(V_rtzone, rt);
474 		return;
475 	}
476 done:
477 	RT_UNLOCK(rt);
478 }
479 
480 
481 /*
482  * Force a routing table entry to the specified
483  * destination to go through the given gateway.
484  * Normally called as a result of a routing redirect
485  * message from the network layer.
486  */
487 void
488 rtredirect(struct sockaddr *dst,
489 	struct sockaddr *gateway,
490 	struct sockaddr *netmask,
491 	int flags,
492 	struct sockaddr *src)
493 {
494 	rtredirect_fib(dst, gateway, netmask, flags, src, 0);
495 }
496 
497 void
498 rtredirect_fib(struct sockaddr *dst,
499 	struct sockaddr *gateway,
500 	struct sockaddr *netmask,
501 	int flags,
502 	struct sockaddr *src,
503 	u_int fibnum)
504 {
505 	struct rtentry *rt, *rt0 = NULL;
506 	int error = 0;
507 	short *stat = NULL;
508 	struct rt_addrinfo info;
509 	struct ifaddr *ifa;
510 	struct radix_node_head *rnh;
511 
512 	ifa = NULL;
513 	rnh = rt_tables_get_rnh(fibnum, dst->sa_family);
514 	if (rnh == NULL) {
515 		error = EAFNOSUPPORT;
516 		goto out;
517 	}
518 
519 	/* verify the gateway is directly reachable */
520 	if ((ifa = ifa_ifwithnet(gateway)) == NULL) {
521 		error = ENETUNREACH;
522 		goto out;
523 	}
524 	rt = rtalloc1_fib(dst, 0, 0UL, fibnum);	/* NB: rt is locked */
525 	/*
526 	 * If the redirect isn't from our current router for this dst,
527 	 * it's either old or wrong.  If it redirects us to ourselves,
528 	 * we have a routing loop, perhaps as a result of an interface
529 	 * going down recently.
530 	 */
531 	if (!(flags & RTF_DONE) && rt &&
532 	     (!sa_equal(src, rt->rt_gateway) || rt->rt_ifa != ifa))
533 		error = EINVAL;
534 	else if (ifa_ifwithaddr_check(gateway))
535 		error = EHOSTUNREACH;
536 	if (error)
537 		goto done;
538 	/*
539 	 * Create a new entry if we just got back a wildcard entry
540 	 * or the the lookup failed.  This is necessary for hosts
541 	 * which use routing redirects generated by smart gateways
542 	 * to dynamically build the routing tables.
543 	 */
544 	if (rt == NULL || (rt_mask(rt) && rt_mask(rt)->sa_len < 2))
545 		goto create;
546 	/*
547 	 * Don't listen to the redirect if it's
548 	 * for a route to an interface.
549 	 */
550 	if (rt->rt_flags & RTF_GATEWAY) {
551 		if (((rt->rt_flags & RTF_HOST) == 0) && (flags & RTF_HOST)) {
552 			/*
553 			 * Changing from route to net => route to host.
554 			 * Create new route, rather than smashing route to net.
555 			 */
556 		create:
557 			rt0 = rt;
558 			rt = NULL;
559 
560 			flags |=  RTF_GATEWAY | RTF_DYNAMIC;
561 			bzero((caddr_t)&info, sizeof(info));
562 			info.rti_info[RTAX_DST] = dst;
563 			info.rti_info[RTAX_GATEWAY] = gateway;
564 			info.rti_info[RTAX_NETMASK] = netmask;
565 			info.rti_ifa = ifa;
566 			info.rti_flags = flags;
567 			if (rt0 != NULL)
568 				RT_UNLOCK(rt0);	/* drop lock to avoid LOR with RNH */
569 			error = rtrequest1_fib(RTM_ADD, &info, &rt, fibnum);
570 			if (rt != NULL) {
571 				RT_LOCK(rt);
572 				if (rt0 != NULL)
573 					EVENTHANDLER_INVOKE(route_redirect_event, rt0, rt, dst);
574 				flags = rt->rt_flags;
575 			}
576 			if (rt0 != NULL)
577 				RTFREE(rt0);
578 
579 			stat = &V_rtstat.rts_dynamic;
580 		} else {
581 			struct rtentry *gwrt;
582 
583 			/*
584 			 * Smash the current notion of the gateway to
585 			 * this destination.  Should check about netmask!!!
586 			 */
587 			rt->rt_flags |= RTF_MODIFIED;
588 			flags |= RTF_MODIFIED;
589 			stat = &V_rtstat.rts_newgateway;
590 			/*
591 			 * add the key and gateway (in one malloc'd chunk).
592 			 */
593 			RT_UNLOCK(rt);
594 			RADIX_NODE_HEAD_LOCK(rnh);
595 			RT_LOCK(rt);
596 			rt_setgate(rt, rt_key(rt), gateway);
597 			gwrt = rtalloc1(gateway, 1, RTF_RNH_LOCKED);
598 			RADIX_NODE_HEAD_UNLOCK(rnh);
599 			EVENTHANDLER_INVOKE(route_redirect_event, rt, gwrt, dst);
600 			RTFREE_LOCKED(gwrt);
601 		}
602 	} else
603 		error = EHOSTUNREACH;
604 done:
605 	if (rt)
606 		RTFREE_LOCKED(rt);
607 out:
608 	if (error)
609 		V_rtstat.rts_badredirect++;
610 	else if (stat != NULL)
611 		(*stat)++;
612 	bzero((caddr_t)&info, sizeof(info));
613 	info.rti_info[RTAX_DST] = dst;
614 	info.rti_info[RTAX_GATEWAY] = gateway;
615 	info.rti_info[RTAX_NETMASK] = netmask;
616 	info.rti_info[RTAX_AUTHOR] = src;
617 	rt_missmsg(RTM_REDIRECT, &info, flags, error);
618 	if (ifa != NULL)
619 		ifa_free(ifa);
620 }
621 
622 int
623 rtioctl(u_long req, caddr_t data)
624 {
625 	return (rtioctl_fib(req, data, 0));
626 }
627 
628 /*
629  * Routing table ioctl interface.
630  */
631 int
632 rtioctl_fib(u_long req, caddr_t data, u_int fibnum)
633 {
634 
635 	/*
636 	 * If more ioctl commands are added here, make sure the proper
637 	 * super-user checks are being performed because it is possible for
638 	 * prison-root to make it this far if raw sockets have been enabled
639 	 * in jails.
640 	 */
641 #ifdef INET
642 	/* Multicast goop, grrr... */
643 	return mrt_ioctl ? mrt_ioctl(req, data, fibnum) : EOPNOTSUPP;
644 #else /* INET */
645 	return ENXIO;
646 #endif /* INET */
647 }
648 
649 /*
650  * For both ifa_ifwithroute() routines, 'ifa' is returned referenced.
651  */
652 struct ifaddr *
653 ifa_ifwithroute(int flags, struct sockaddr *dst, struct sockaddr *gateway)
654 {
655 	return (ifa_ifwithroute_fib(flags, dst, gateway, 0));
656 }
657 
658 struct ifaddr *
659 ifa_ifwithroute_fib(int flags, struct sockaddr *dst, struct sockaddr *gateway,
660 				u_int fibnum)
661 {
662 	register struct ifaddr *ifa;
663 	int not_found = 0;
664 
665 	if ((flags & RTF_GATEWAY) == 0) {
666 		/*
667 		 * If we are adding a route to an interface,
668 		 * and the interface is a pt to pt link
669 		 * we should search for the destination
670 		 * as our clue to the interface.  Otherwise
671 		 * we can use the local address.
672 		 */
673 		ifa = NULL;
674 		if (flags & RTF_HOST)
675 			ifa = ifa_ifwithdstaddr(dst);
676 		if (ifa == NULL)
677 			ifa = ifa_ifwithaddr(gateway);
678 	} else {
679 		/*
680 		 * If we are adding a route to a remote net
681 		 * or host, the gateway may still be on the
682 		 * other end of a pt to pt link.
683 		 */
684 		ifa = ifa_ifwithdstaddr(gateway);
685 	}
686 	if (ifa == NULL)
687 		ifa = ifa_ifwithnet(gateway);
688 	if (ifa == NULL) {
689 		struct rtentry *rt = rtalloc1_fib(gateway, 0, RTF_RNH_LOCKED, fibnum);
690 		if (rt == NULL)
691 			return (NULL);
692 		/*
693 		 * dismiss a gateway that is reachable only
694 		 * through the default router
695 		 */
696 		switch (gateway->sa_family) {
697 		case AF_INET:
698 			if (satosin(rt_key(rt))->sin_addr.s_addr == INADDR_ANY)
699 				not_found = 1;
700 			break;
701 		case AF_INET6:
702 			if (IN6_IS_ADDR_UNSPECIFIED(&satosin6(rt_key(rt))->sin6_addr))
703 				not_found = 1;
704 			break;
705 		default:
706 			break;
707 		}
708 		if (!not_found && rt->rt_ifa != NULL) {
709 			ifa = rt->rt_ifa;
710 			ifa_ref(ifa);
711 		}
712 		RT_REMREF(rt);
713 		RT_UNLOCK(rt);
714 		if (not_found || ifa == NULL)
715 			return (NULL);
716 	}
717 	if (ifa->ifa_addr->sa_family != dst->sa_family) {
718 		struct ifaddr *oifa = ifa;
719 		ifa = ifaof_ifpforaddr(dst, ifa->ifa_ifp);
720 		if (ifa == NULL)
721 			ifa = oifa;
722 		else
723 			ifa_free(oifa);
724 	}
725 	return (ifa);
726 }
727 
728 /*
729  * Do appropriate manipulations of a routing tree given
730  * all the bits of info needed
731  */
732 int
733 rtrequest(int req,
734 	struct sockaddr *dst,
735 	struct sockaddr *gateway,
736 	struct sockaddr *netmask,
737 	int flags,
738 	struct rtentry **ret_nrt)
739 {
740 	return (rtrequest_fib(req, dst, gateway, netmask, flags, ret_nrt, 0));
741 }
742 
743 int
744 rtrequest_fib(int req,
745 	struct sockaddr *dst,
746 	struct sockaddr *gateway,
747 	struct sockaddr *netmask,
748 	int flags,
749 	struct rtentry **ret_nrt,
750 	u_int fibnum)
751 {
752 	struct rt_addrinfo info;
753 
754 	if (dst->sa_len == 0)
755 		return(EINVAL);
756 
757 	bzero((caddr_t)&info, sizeof(info));
758 	info.rti_flags = flags;
759 	info.rti_info[RTAX_DST] = dst;
760 	info.rti_info[RTAX_GATEWAY] = gateway;
761 	info.rti_info[RTAX_NETMASK] = netmask;
762 	return rtrequest1_fib(req, &info, ret_nrt, fibnum);
763 }
764 
765 /*
766  * These (questionable) definitions of apparent local variables apply
767  * to the next two functions.  XXXXXX!!!
768  */
769 #define	dst	info->rti_info[RTAX_DST]
770 #define	gateway	info->rti_info[RTAX_GATEWAY]
771 #define	netmask	info->rti_info[RTAX_NETMASK]
772 #define	ifaaddr	info->rti_info[RTAX_IFA]
773 #define	ifpaddr	info->rti_info[RTAX_IFP]
774 #define	flags	info->rti_flags
775 
776 int
777 rt_getifa(struct rt_addrinfo *info)
778 {
779 	return (rt_getifa_fib(info, 0));
780 }
781 
782 /*
783  * Look up rt_addrinfo for a specific fib.  Note that if rti_ifa is defined,
784  * it will be referenced so the caller must free it.
785  */
786 int
787 rt_getifa_fib(struct rt_addrinfo *info, u_int fibnum)
788 {
789 	struct ifaddr *ifa;
790 	int error = 0;
791 
792 	/*
793 	 * ifp may be specified by sockaddr_dl
794 	 * when protocol address is ambiguous.
795 	 */
796 	if (info->rti_ifp == NULL && ifpaddr != NULL &&
797 	    ifpaddr->sa_family == AF_LINK &&
798 	    (ifa = ifa_ifwithnet(ifpaddr)) != NULL) {
799 		info->rti_ifp = ifa->ifa_ifp;
800 		ifa_free(ifa);
801 	}
802 	if (info->rti_ifa == NULL && ifaaddr != NULL)
803 		info->rti_ifa = ifa_ifwithaddr(ifaaddr);
804 	if (info->rti_ifa == NULL) {
805 		struct sockaddr *sa;
806 
807 		sa = ifaaddr != NULL ? ifaaddr :
808 		    (gateway != NULL ? gateway : dst);
809 		if (sa != NULL && info->rti_ifp != NULL)
810 			info->rti_ifa = ifaof_ifpforaddr(sa, info->rti_ifp);
811 		else if (dst != NULL && gateway != NULL)
812 			info->rti_ifa = ifa_ifwithroute_fib(flags, dst, gateway,
813 							fibnum);
814 		else if (sa != NULL)
815 			info->rti_ifa = ifa_ifwithroute_fib(flags, sa, sa,
816 							fibnum);
817 	}
818 	if ((ifa = info->rti_ifa) != NULL) {
819 		if (info->rti_ifp == NULL)
820 			info->rti_ifp = ifa->ifa_ifp;
821 	} else
822 		error = ENETUNREACH;
823 	return (error);
824 }
825 
826 /*
827  * Expunges references to a route that's about to be reclaimed.
828  * The route must be locked.
829  */
830 int
831 rtexpunge(struct rtentry *rt)
832 {
833 	struct radix_node *rn;
834 	struct radix_node_head *rnh;
835 	struct ifaddr *ifa;
836 	int error = 0;
837 
838 	/*
839 	 * Find the correct routing tree to use for this Address Family
840 	 */
841 	rnh = rt_tables_get_rnh(rt->rt_fibnum, rt_key(rt)->sa_family);
842 	RT_LOCK_ASSERT(rt);
843 	if (rnh == NULL)
844 		return (EAFNOSUPPORT);
845 	RADIX_NODE_HEAD_LOCK_ASSERT(rnh);
846 #if 0
847 	/*
848 	 * We cannot assume anything about the reference count
849 	 * because protocols call us in many situations; often
850 	 * before unwinding references to the table entry.
851 	 */
852 	KASSERT(rt->rt_refcnt <= 1, ("bogus refcnt %ld", rt->rt_refcnt));
853 #endif
854 	/*
855 	 * Remove the item from the tree; it should be there,
856 	 * but when callers invoke us blindly it may not (sigh).
857 	 */
858 	rn = rnh->rnh_deladdr(rt_key(rt), rt_mask(rt), rnh);
859 	if (rn == NULL) {
860 		error = ESRCH;
861 		goto bad;
862 	}
863 	KASSERT((rn->rn_flags & (RNF_ACTIVE | RNF_ROOT)) == 0,
864 		("unexpected flags 0x%x", rn->rn_flags));
865 	KASSERT(rt == RNTORT(rn),
866 		("lookup mismatch, rt %p rn %p", rt, rn));
867 
868 	rt->rt_flags &= ~RTF_UP;
869 
870 	/*
871 	 * Give the protocol a chance to keep things in sync.
872 	 */
873 	if ((ifa = rt->rt_ifa) && ifa->ifa_rtrequest) {
874 		struct rt_addrinfo info;
875 
876 		bzero((caddr_t)&info, sizeof(info));
877 		info.rti_flags = rt->rt_flags;
878 		info.rti_info[RTAX_DST] = rt_key(rt);
879 		info.rti_info[RTAX_GATEWAY] = rt->rt_gateway;
880 		info.rti_info[RTAX_NETMASK] = rt_mask(rt);
881 		ifa->ifa_rtrequest(RTM_DELETE, rt, &info);
882 	}
883 
884 	/*
885 	 * one more rtentry floating around that is not
886 	 * linked to the routing table.
887 	 */
888 	V_rttrash++;
889 bad:
890 	return (error);
891 }
892 
893 #ifdef RADIX_MPATH
894 static int
895 rn_mpath_update(int req, struct rt_addrinfo *info,
896     struct radix_node_head *rnh, struct rtentry **ret_nrt)
897 {
898 	/*
899 	 * if we got multipath routes, we require users to specify
900 	 * a matching RTAX_GATEWAY.
901 	 */
902 	struct rtentry *rt, *rto = NULL;
903 	register struct radix_node *rn;
904 	int error = 0;
905 
906 	rn = rnh->rnh_matchaddr(dst, rnh);
907 	if (rn == NULL)
908 		return (ESRCH);
909 	rto = rt = RNTORT(rn);
910 	rt = rt_mpath_matchgate(rt, gateway);
911 	if (rt == NULL)
912 		return (ESRCH);
913 	/*
914 	 * this is the first entry in the chain
915 	 */
916 	if (rto == rt) {
917 		rn = rn_mpath_next((struct radix_node *)rt);
918 		/*
919 		 * there is another entry, now it's active
920 		 */
921 		if (rn) {
922 			rto = RNTORT(rn);
923 			RT_LOCK(rto);
924 			rto->rt_flags |= RTF_UP;
925 			RT_UNLOCK(rto);
926 		} else if (rt->rt_flags & RTF_GATEWAY) {
927 			/*
928 			 * For gateway routes, we need to
929 			 * make sure that we we are deleting
930 			 * the correct gateway.
931 			 * rt_mpath_matchgate() does not
932 			 * check the case when there is only
933 			 * one route in the chain.
934 			 */
935 			if (gateway &&
936 			    (rt->rt_gateway->sa_len != gateway->sa_len ||
937 				memcmp(rt->rt_gateway, gateway, gateway->sa_len)))
938 				error = ESRCH;
939 			else {
940 				/*
941 				 * remove from tree before returning it
942 				 * to the caller
943 				 */
944 				rn = rnh->rnh_deladdr(dst, netmask, rnh);
945 				KASSERT(rt == RNTORT(rn), ("radix node disappeared"));
946 				goto gwdelete;
947 			}
948 
949 		}
950 		/*
951 		 * use the normal delete code to remove
952 		 * the first entry
953 		 */
954 		if (req != RTM_DELETE)
955 			goto nondelete;
956 
957 		error = ENOENT;
958 		goto done;
959 	}
960 
961 	/*
962 	 * if the entry is 2nd and on up
963 	 */
964 	if ((req == RTM_DELETE) && !rt_mpath_deldup(rto, rt))
965 		panic ("rtrequest1: rt_mpath_deldup");
966 gwdelete:
967 	RT_LOCK(rt);
968 	RT_ADDREF(rt);
969 	if (req == RTM_DELETE) {
970 		rt->rt_flags &= ~RTF_UP;
971 		/*
972 		 * One more rtentry floating around that is not
973 		 * linked to the routing table. rttrash will be decremented
974 		 * when RTFREE(rt) is eventually called.
975 		 */
976 		V_rttrash++;
977 	}
978 
979 nondelete:
980 	if (req != RTM_DELETE)
981 		panic("unrecognized request %d", req);
982 
983 
984 	/*
985 	 * If the caller wants it, then it can have it,
986 	 * but it's up to it to free the rtentry as we won't be
987 	 * doing it.
988 	 */
989 	if (ret_nrt) {
990 		*ret_nrt = rt;
991 		RT_UNLOCK(rt);
992 	} else
993 		RTFREE_LOCKED(rt);
994 done:
995 	return (error);
996 }
997 #endif
998 
999 int
1000 rtrequest1_fib(int req, struct rt_addrinfo *info, struct rtentry **ret_nrt,
1001 				u_int fibnum)
1002 {
1003 	int error = 0, needlock = 0;
1004 	register struct rtentry *rt;
1005 #ifdef FLOWTABLE
1006 	register struct rtentry *rt0;
1007 #endif
1008 	register struct radix_node *rn;
1009 	register struct radix_node_head *rnh;
1010 	struct ifaddr *ifa;
1011 	struct sockaddr *ndst;
1012 #define senderr(x) { error = x ; goto bad; }
1013 
1014 	KASSERT((fibnum < rt_numfibs), ("rtrequest1_fib: bad fibnum"));
1015 	if (dst->sa_family != AF_INET)	/* Only INET supports > 1 fib now */
1016 		fibnum = 0;
1017 	/*
1018 	 * Find the correct routing tree to use for this Address Family
1019 	 */
1020 	rnh = rt_tables_get_rnh(fibnum, dst->sa_family);
1021 	if (rnh == NULL)
1022 		return (EAFNOSUPPORT);
1023 	needlock = ((flags & RTF_RNH_LOCKED) == 0);
1024 	flags &= ~RTF_RNH_LOCKED;
1025 	if (needlock)
1026 		RADIX_NODE_HEAD_LOCK(rnh);
1027 	else
1028 		RADIX_NODE_HEAD_LOCK_ASSERT(rnh);
1029 	/*
1030 	 * If we are adding a host route then we don't want to put
1031 	 * a netmask in the tree, nor do we want to clone it.
1032 	 */
1033 	if (flags & RTF_HOST)
1034 		netmask = NULL;
1035 
1036 	switch (req) {
1037 	case RTM_DELETE:
1038 #ifdef RADIX_MPATH
1039 		if (rn_mpath_capable(rnh)) {
1040 			error = rn_mpath_update(req, info, rnh, ret_nrt);
1041 			/*
1042 			 * "bad" holds true for the success case
1043 			 * as well
1044 			 */
1045 			if (error != ENOENT)
1046 				goto bad;
1047 		}
1048 #endif
1049 		/*
1050 		 * Remove the item from the tree and return it.
1051 		 * Complain if it is not there and do no more processing.
1052 		 */
1053 		rn = rnh->rnh_deladdr(dst, netmask, rnh);
1054 		if (rn == NULL)
1055 			senderr(ESRCH);
1056 		if (rn->rn_flags & (RNF_ACTIVE | RNF_ROOT))
1057 			panic ("rtrequest delete");
1058 		rt = RNTORT(rn);
1059 		RT_LOCK(rt);
1060 		RT_ADDREF(rt);
1061 		rt->rt_flags &= ~RTF_UP;
1062 
1063 		/*
1064 		 * give the protocol a chance to keep things in sync.
1065 		 */
1066 		if ((ifa = rt->rt_ifa) && ifa->ifa_rtrequest)
1067 			ifa->ifa_rtrequest(RTM_DELETE, rt, info);
1068 
1069 		/*
1070 		 * One more rtentry floating around that is not
1071 		 * linked to the routing table. rttrash will be decremented
1072 		 * when RTFREE(rt) is eventually called.
1073 		 */
1074 		V_rttrash++;
1075 
1076 		/*
1077 		 * If the caller wants it, then it can have it,
1078 		 * but it's up to it to free the rtentry as we won't be
1079 		 * doing it.
1080 		 */
1081 		if (ret_nrt) {
1082 			*ret_nrt = rt;
1083 			RT_UNLOCK(rt);
1084 		} else
1085 			RTFREE_LOCKED(rt);
1086 		break;
1087 	case RTM_RESOLVE:
1088 		/*
1089 		 * resolve was only used for route cloning
1090 		 * here for compat
1091 		 */
1092 		break;
1093 	case RTM_ADD:
1094 		if ((flags & RTF_GATEWAY) && !gateway)
1095 			senderr(EINVAL);
1096 		if (dst && gateway && (dst->sa_family != gateway->sa_family) &&
1097 		    (gateway->sa_family != AF_UNSPEC) && (gateway->sa_family != AF_LINK))
1098 			senderr(EINVAL);
1099 
1100 		if (info->rti_ifa == NULL) {
1101 			error = rt_getifa_fib(info, fibnum);
1102 			if (error)
1103 				senderr(error);
1104 		} else
1105 			ifa_ref(info->rti_ifa);
1106 		ifa = info->rti_ifa;
1107 		rt = uma_zalloc(V_rtzone, M_NOWAIT | M_ZERO);
1108 		if (rt == NULL) {
1109 			if (ifa != NULL)
1110 				ifa_free(ifa);
1111 			senderr(ENOBUFS);
1112 		}
1113 		RT_LOCK_INIT(rt);
1114 		rt->rt_flags = RTF_UP | flags;
1115 		rt->rt_fibnum = fibnum;
1116 		/*
1117 		 * Add the gateway. Possibly re-malloc-ing the storage for it
1118 		 *
1119 		 */
1120 		RT_LOCK(rt);
1121 		if ((error = rt_setgate(rt, dst, gateway)) != 0) {
1122 			RT_LOCK_DESTROY(rt);
1123 			if (ifa != NULL)
1124 				ifa_free(ifa);
1125 			uma_zfree(V_rtzone, rt);
1126 			senderr(error);
1127 		}
1128 
1129 		/*
1130 		 * point to the (possibly newly malloc'd) dest address.
1131 		 */
1132 		ndst = (struct sockaddr *)rt_key(rt);
1133 
1134 		/*
1135 		 * make sure it contains the value we want (masked if needed).
1136 		 */
1137 		if (netmask) {
1138 			rt_maskedcopy(dst, ndst, netmask);
1139 		} else
1140 			bcopy(dst, ndst, dst->sa_len);
1141 
1142 		/*
1143 		 * We use the ifa reference returned by rt_getifa_fib().
1144 		 * This moved from below so that rnh->rnh_addaddr() can
1145 		 * examine the ifa and  ifa->ifa_ifp if it so desires.
1146 		 */
1147 		rt->rt_ifa = ifa;
1148 		rt->rt_ifp = ifa->ifa_ifp;
1149 		rt->rt_rmx.rmx_weight = 1;
1150 
1151 #ifdef RADIX_MPATH
1152 		/* do not permit exactly the same dst/mask/gw pair */
1153 		if (rn_mpath_capable(rnh) &&
1154 			rt_mpath_conflict(rnh, rt, netmask)) {
1155 			if (rt->rt_ifa) {
1156 				ifa_free(rt->rt_ifa);
1157 			}
1158 			Free(rt_key(rt));
1159 			RT_LOCK_DESTROY(rt);
1160 			uma_zfree(V_rtzone, rt);
1161 			senderr(EEXIST);
1162 		}
1163 #endif
1164 
1165 #ifdef FLOWTABLE
1166 		rt0 = NULL;
1167 		/* XXX
1168 		 * "flow-table" only support IPv4 at the moment.
1169 		 */
1170 #ifdef INET
1171 		if (dst->sa_family == AF_INET) {
1172 			rn = rnh->rnh_matchaddr(dst, rnh);
1173 			if (rn && ((rn->rn_flags & RNF_ROOT) == 0)) {
1174 				struct sockaddr *mask;
1175 				u_char *m, *n;
1176 				int len;
1177 
1178 				/*
1179 				 * compare mask to see if the new route is
1180 				 * more specific than the existing one
1181 				 */
1182 				rt0 = RNTORT(rn);
1183 				RT_LOCK(rt0);
1184 				RT_ADDREF(rt0);
1185 				RT_UNLOCK(rt0);
1186 				/*
1187 				 * A host route is already present, so
1188 				 * leave the flow-table entries as is.
1189 				 */
1190 				if (rt0->rt_flags & RTF_HOST) {
1191 					RTFREE(rt0);
1192 					rt0 = NULL;
1193 				} else if (!(flags & RTF_HOST) && netmask) {
1194 					mask = rt_mask(rt0);
1195 					len = mask->sa_len;
1196 					m = (u_char *)mask;
1197 					n = (u_char *)netmask;
1198 					while (len-- > 0) {
1199 						if (*n != *m)
1200 							break;
1201 						n++;
1202 						m++;
1203 					}
1204 					if (len == 0 || (*n < *m)) {
1205 						RTFREE(rt0);
1206 						rt0 = NULL;
1207 					}
1208 				}
1209 			}
1210 		}
1211 #endif
1212 #endif
1213 
1214 		/* XXX mtu manipulation will be done in rnh_addaddr -- itojun */
1215 		rn = rnh->rnh_addaddr(ndst, netmask, rnh, rt->rt_nodes);
1216 		/*
1217 		 * If it still failed to go into the tree,
1218 		 * then un-make it (this should be a function)
1219 		 */
1220 		if (rn == NULL) {
1221 			if (rt->rt_ifa)
1222 				ifa_free(rt->rt_ifa);
1223 			Free(rt_key(rt));
1224 			RT_LOCK_DESTROY(rt);
1225 			uma_zfree(V_rtzone, rt);
1226 #ifdef FLOWTABLE
1227 			if (rt0 != NULL)
1228 				RTFREE(rt0);
1229 #endif
1230 			senderr(EEXIST);
1231 		}
1232 #ifdef FLOWTABLE
1233 		else if (rt0 != NULL) {
1234 #ifdef INET
1235 			flowtable_route_flush(V_ip_ft, rt0);
1236 #endif
1237 			RTFREE(rt0);
1238 		}
1239 #endif
1240 
1241 		/*
1242 		 * If this protocol has something to add to this then
1243 		 * allow it to do that as well.
1244 		 */
1245 		if (ifa->ifa_rtrequest)
1246 			ifa->ifa_rtrequest(req, rt, info);
1247 
1248 		/*
1249 		 * actually return a resultant rtentry and
1250 		 * give the caller a single reference.
1251 		 */
1252 		if (ret_nrt) {
1253 			*ret_nrt = rt;
1254 			RT_ADDREF(rt);
1255 		}
1256 		RT_UNLOCK(rt);
1257 		break;
1258 	default:
1259 		error = EOPNOTSUPP;
1260 	}
1261 bad:
1262 	if (needlock)
1263 		RADIX_NODE_HEAD_UNLOCK(rnh);
1264 	return (error);
1265 #undef senderr
1266 }
1267 
1268 #undef dst
1269 #undef gateway
1270 #undef netmask
1271 #undef ifaaddr
1272 #undef ifpaddr
1273 #undef flags
1274 
1275 int
1276 rt_setgate(struct rtentry *rt, struct sockaddr *dst, struct sockaddr *gate)
1277 {
1278 	/* XXX dst may be overwritten, can we move this to below */
1279 	int dlen = SA_SIZE(dst), glen = SA_SIZE(gate);
1280 #ifdef INVARIANTS
1281 	struct radix_node_head *rnh;
1282 
1283 	rnh = rt_tables_get_rnh(rt->rt_fibnum, dst->sa_family);
1284 #endif
1285 
1286 	RT_LOCK_ASSERT(rt);
1287 	RADIX_NODE_HEAD_LOCK_ASSERT(rnh);
1288 
1289 	/*
1290 	 * Prepare to store the gateway in rt->rt_gateway.
1291 	 * Both dst and gateway are stored one after the other in the same
1292 	 * malloc'd chunk. If we have room, we can reuse the old buffer,
1293 	 * rt_gateway already points to the right place.
1294 	 * Otherwise, malloc a new block and update the 'dst' address.
1295 	 */
1296 	if (rt->rt_gateway == NULL || glen > SA_SIZE(rt->rt_gateway)) {
1297 		caddr_t new;
1298 
1299 		R_Malloc(new, caddr_t, dlen + glen);
1300 		if (new == NULL)
1301 			return ENOBUFS;
1302 		/*
1303 		 * XXX note, we copy from *dst and not *rt_key(rt) because
1304 		 * rt_setgate() can be called to initialize a newly
1305 		 * allocated route entry, in which case rt_key(rt) == NULL
1306 		 * (and also rt->rt_gateway == NULL).
1307 		 * Free()/free() handle a NULL argument just fine.
1308 		 */
1309 		bcopy(dst, new, dlen);
1310 		Free(rt_key(rt));	/* free old block, if any */
1311 		rt_key(rt) = (struct sockaddr *)new;
1312 		rt->rt_gateway = (struct sockaddr *)(new + dlen);
1313 	}
1314 
1315 	/*
1316 	 * Copy the new gateway value into the memory chunk.
1317 	 */
1318 	bcopy(gate, rt->rt_gateway, glen);
1319 
1320 	return (0);
1321 }
1322 
1323 void
1324 rt_maskedcopy(struct sockaddr *src, struct sockaddr *dst, struct sockaddr *netmask)
1325 {
1326 	register u_char *cp1 = (u_char *)src;
1327 	register u_char *cp2 = (u_char *)dst;
1328 	register u_char *cp3 = (u_char *)netmask;
1329 	u_char *cplim = cp2 + *cp3;
1330 	u_char *cplim2 = cp2 + *cp1;
1331 
1332 	*cp2++ = *cp1++; *cp2++ = *cp1++; /* copies sa_len & sa_family */
1333 	cp3 += 2;
1334 	if (cplim > cplim2)
1335 		cplim = cplim2;
1336 	while (cp2 < cplim)
1337 		*cp2++ = *cp1++ & *cp3++;
1338 	if (cp2 < cplim2)
1339 		bzero((caddr_t)cp2, (unsigned)(cplim2 - cp2));
1340 }
1341 
1342 /*
1343  * Set up a routing table entry, normally
1344  * for an interface.
1345  */
1346 #define _SOCKADDR_TMPSIZE 128 /* Not too big.. kernel stack size is limited */
1347 static inline  int
1348 rtinit1(struct ifaddr *ifa, int cmd, int flags, int fibnum)
1349 {
1350 	struct sockaddr *dst;
1351 	struct sockaddr *netmask;
1352 	struct rtentry *rt = NULL;
1353 	struct rt_addrinfo info;
1354 	int error = 0;
1355 	int startfib, endfib;
1356 	char tempbuf[_SOCKADDR_TMPSIZE];
1357 	int didwork = 0;
1358 	int a_failure = 0;
1359 	static struct sockaddr_dl null_sdl = {sizeof(null_sdl), AF_LINK};
1360 
1361 	if (flags & RTF_HOST) {
1362 		dst = ifa->ifa_dstaddr;
1363 		netmask = NULL;
1364 	} else {
1365 		dst = ifa->ifa_addr;
1366 		netmask = ifa->ifa_netmask;
1367 	}
1368 	if ( dst->sa_family != AF_INET)
1369 		fibnum = 0;
1370 	if (fibnum == -1) {
1371 		if (rt_add_addr_allfibs == 0 && cmd == (int)RTM_ADD) {
1372 			startfib = endfib = curthread->td_proc->p_fibnum;
1373 		} else {
1374 			startfib = 0;
1375 			endfib = rt_numfibs - 1;
1376 		}
1377 	} else {
1378 		KASSERT((fibnum < rt_numfibs), ("rtinit1: bad fibnum"));
1379 		startfib = fibnum;
1380 		endfib = fibnum;
1381 	}
1382 	if (dst->sa_len == 0)
1383 		return(EINVAL);
1384 
1385 	/*
1386 	 * If it's a delete, check that if it exists,
1387 	 * it's on the correct interface or we might scrub
1388 	 * a route to another ifa which would
1389 	 * be confusing at best and possibly worse.
1390 	 */
1391 	if (cmd == RTM_DELETE) {
1392 		/*
1393 		 * It's a delete, so it should already exist..
1394 		 * If it's a net, mask off the host bits
1395 		 * (Assuming we have a mask)
1396 		 * XXX this is kinda inet specific..
1397 		 */
1398 		if (netmask != NULL) {
1399 			rt_maskedcopy(dst, (struct sockaddr *)tempbuf, netmask);
1400 			dst = (struct sockaddr *)tempbuf;
1401 		}
1402 	}
1403 	/*
1404 	 * Now go through all the requested tables (fibs) and do the
1405 	 * requested action. Realistically, this will either be fib 0
1406 	 * for protocols that don't do multiple tables or all the
1407 	 * tables for those that do. XXX For this version only AF_INET.
1408 	 * When that changes code should be refactored to protocol
1409 	 * independent parts and protocol dependent parts.
1410 	 */
1411 	for ( fibnum = startfib; fibnum <= endfib; fibnum++) {
1412 		if (cmd == RTM_DELETE) {
1413 			struct radix_node_head *rnh;
1414 			struct radix_node *rn;
1415 			/*
1416 			 * Look up an rtentry that is in the routing tree and
1417 			 * contains the correct info.
1418 			 */
1419 			rnh = rt_tables_get_rnh(fibnum, dst->sa_family);
1420 			if (rnh == NULL)
1421 				/* this table doesn't exist but others might */
1422 				continue;
1423 			RADIX_NODE_HEAD_LOCK(rnh);
1424 #ifdef RADIX_MPATH
1425 			if (rn_mpath_capable(rnh)) {
1426 
1427 				rn = rnh->rnh_matchaddr(dst, rnh);
1428 				if (rn == NULL)
1429 					error = ESRCH;
1430 				else {
1431 					rt = RNTORT(rn);
1432 					/*
1433 					 * for interface route the
1434 					 * rt->rt_gateway is sockaddr_intf
1435 					 * for cloning ARP entries, so
1436 					 * rt_mpath_matchgate must use the
1437 					 * interface address
1438 					 */
1439 					rt = rt_mpath_matchgate(rt,
1440 					    ifa->ifa_addr);
1441 					if (!rt)
1442 						error = ESRCH;
1443 				}
1444 			}
1445 			else
1446 #endif
1447 			rn = rnh->rnh_lookup(dst, netmask, rnh);
1448 			error = (rn == NULL ||
1449 			    (rn->rn_flags & RNF_ROOT) ||
1450 			    RNTORT(rn)->rt_ifa != ifa ||
1451 			    !sa_equal((struct sockaddr *)rn->rn_key, dst));
1452 			RADIX_NODE_HEAD_UNLOCK(rnh);
1453 			if (error) {
1454 				/* this is only an error if bad on ALL tables */
1455 				continue;
1456 			}
1457 		}
1458 		/*
1459 		 * Do the actual request
1460 		 */
1461 		bzero((caddr_t)&info, sizeof(info));
1462 		info.rti_ifa = ifa;
1463 		info.rti_flags = flags | ifa->ifa_flags;
1464 		info.rti_info[RTAX_DST] = dst;
1465 		/*
1466 		 * doing this for compatibility reasons
1467 		 */
1468 		if (cmd == RTM_ADD)
1469 			info.rti_info[RTAX_GATEWAY] =
1470 			    (struct sockaddr *)&null_sdl;
1471 		else
1472 			info.rti_info[RTAX_GATEWAY] = ifa->ifa_addr;
1473 		info.rti_info[RTAX_NETMASK] = netmask;
1474 		error = rtrequest1_fib(cmd, &info, &rt, fibnum);
1475 		if (error == 0 && rt != NULL) {
1476 			/*
1477 			 * notify any listening routing agents of the change
1478 			 */
1479 			RT_LOCK(rt);
1480 #ifdef RADIX_MPATH
1481 			/*
1482 			 * in case address alias finds the first address
1483 			 * e.g. ifconfig bge0 192.103.54.246/24
1484 			 * e.g. ifconfig bge0 192.103.54.247/24
1485 			 * the address set in the route is 192.103.54.246
1486 			 * so we need to replace it with 192.103.54.247
1487 			 */
1488 			if (memcmp(rt->rt_ifa->ifa_addr,
1489 			    ifa->ifa_addr, ifa->ifa_addr->sa_len)) {
1490 				ifa_free(rt->rt_ifa);
1491 				ifa_ref(ifa);
1492 				rt->rt_ifp = ifa->ifa_ifp;
1493 				rt->rt_ifa = ifa;
1494 			}
1495 #endif
1496 			/*
1497 			 * doing this for compatibility reasons
1498 			 */
1499 			if (cmd == RTM_ADD) {
1500 			    ((struct sockaddr_dl *)rt->rt_gateway)->sdl_type  =
1501 				rt->rt_ifp->if_type;
1502 			    ((struct sockaddr_dl *)rt->rt_gateway)->sdl_index =
1503 				rt->rt_ifp->if_index;
1504 			}
1505 			RT_ADDREF(rt);
1506 			RT_UNLOCK(rt);
1507 			rt_newaddrmsg(cmd, ifa, error, rt);
1508 			RT_LOCK(rt);
1509 			RT_REMREF(rt);
1510 			if (cmd == RTM_DELETE) {
1511 				/*
1512 				 * If we are deleting, and we found an entry,
1513 				 * then it's been removed from the tree..
1514 				 * now throw it away.
1515 				 */
1516 				RTFREE_LOCKED(rt);
1517 			} else {
1518 				if (cmd == RTM_ADD) {
1519 					/*
1520 					 * We just wanted to add it..
1521 					 * we don't actually need a reference.
1522 					 */
1523 					RT_REMREF(rt);
1524 				}
1525 				RT_UNLOCK(rt);
1526 			}
1527 			didwork = 1;
1528 		}
1529 		if (error)
1530 			a_failure = error;
1531 	}
1532 	if (cmd == RTM_DELETE) {
1533 		if (didwork) {
1534 			error = 0;
1535 		} else {
1536 			/* we only give an error if it wasn't in any table */
1537 			error = ((flags & RTF_HOST) ?
1538 			    EHOSTUNREACH : ENETUNREACH);
1539 		}
1540 	} else {
1541 		if (a_failure) {
1542 			/* return an error if any of them failed */
1543 			error = a_failure;
1544 		}
1545 	}
1546 	return (error);
1547 }
1548 
1549 /* special one for inet internal use. may not use. */
1550 int
1551 rtinit_fib(struct ifaddr *ifa, int cmd, int flags)
1552 {
1553 	return (rtinit1(ifa, cmd, flags, -1));
1554 }
1555 
1556 /*
1557  * Set up a routing table entry, normally
1558  * for an interface.
1559  */
1560 int
1561 rtinit(struct ifaddr *ifa, int cmd, int flags)
1562 {
1563 	struct sockaddr *dst;
1564 	int fib = 0;
1565 
1566 	if (flags & RTF_HOST) {
1567 		dst = ifa->ifa_dstaddr;
1568 	} else {
1569 		dst = ifa->ifa_addr;
1570 	}
1571 
1572 	if (dst->sa_family == AF_INET)
1573 		fib = -1;
1574 	return (rtinit1(ifa, cmd, flags, fib));
1575 }
1576