xref: /freebsd/sys/net/route.c (revision 59c7ad52aaa5b26e503871334672af0f58f9c2e8)
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_UINT(_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_UINT(_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 rtstat, rtstat);
92 #define	V_rtstat	VNET(rtstat)
93 
94 VNET_DEFINE(struct radix_node_head *, rt_tables);
95 #define	V_rt_tables	VNET(rt_tables)
96 
97 VNET_DEFINE(int, rttrash);		/* routes not in table but not freed */
98 #define	V_rttrash	VNET(rttrash)
99 
100 
101 /* compare two sockaddr structures */
102 #define	sa_equal(a1, a2) (bcmp((a1), (a2), (a1)->sa_len) == 0)
103 
104 /*
105  * Convert a 'struct radix_node *' to a 'struct rtentry *'.
106  * The operation can be done safely (in this code) because a
107  * 'struct rtentry' starts with two 'struct radix_node''s, the first
108  * one representing leaf nodes in the routing tree, which is
109  * what the code in radix.c passes us as a 'struct radix_node'.
110  *
111  * But because there are a lot of assumptions in this conversion,
112  * do not cast explicitly, but always use the macro below.
113  */
114 #define RNTORT(p)	((struct rtentry *)(p))
115 
116 static VNET_DEFINE(uma_zone_t, rtzone);		/* Routing table UMA zone. */
117 #define	V_rtzone	VNET(rtzone)
118 
119 #if 0
120 /* default fib for tunnels to use */
121 u_int tunnel_fib = 0;
122 SYSCTL_INT(_net, OID_AUTO, tunnelfib, CTLFLAG_RD, &tunnel_fib, 0, "");
123 #endif
124 
125 /*
126  * handler for net.my_fibnum
127  */
128 static int
129 sysctl_my_fibnum(SYSCTL_HANDLER_ARGS)
130 {
131         int fibnum;
132         int error;
133 
134         fibnum = curthread->td_proc->p_fibnum;
135         error = sysctl_handle_int(oidp, &fibnum, 0, req);
136         return (error);
137 }
138 
139 SYSCTL_PROC(_net, OID_AUTO, my_fibnum, CTLTYPE_INT|CTLFLAG_RD,
140             NULL, 0, &sysctl_my_fibnum, "I", "default FIB of caller");
141 
142 static __inline struct radix_node_head **
143 rt_tables_get_rnh_ptr(int table, int fam)
144 {
145 	struct radix_node_head **rnh;
146 
147 	KASSERT(table >= 0 && table < rt_numfibs, ("%s: table out of bounds.",
148 	    __func__));
149 	KASSERT(fam >= 0 && fam < (AF_MAX+1), ("%s: fam out of bounds.",
150 	    __func__));
151 
152 	/* rnh is [fib=0][af=0]. */
153 	rnh = (struct radix_node_head **)V_rt_tables;
154 	/* Get the offset to the requested table and fam. */
155 	rnh += table * (AF_MAX+1) + fam;
156 
157 	return (rnh);
158 }
159 
160 struct radix_node_head *
161 rt_tables_get_rnh(int table, int fam)
162 {
163 
164 	return (*rt_tables_get_rnh_ptr(table, fam));
165 }
166 
167 /*
168  * route initialization must occur before ip6_init2(), which happenas at
169  * SI_ORDER_MIDDLE.
170  */
171 static void
172 route_init(void)
173 {
174 	struct domain *dom;
175 	int max_keylen = 0;
176 
177 	/* whack the tunable ints into  line. */
178 	if (rt_numfibs > RT_MAXFIBS)
179 		rt_numfibs = RT_MAXFIBS;
180 	if (rt_numfibs == 0)
181 		rt_numfibs = 1;
182 
183 	for (dom = domains; dom; dom = dom->dom_next)
184 		if (dom->dom_maxrtkey > max_keylen)
185 			max_keylen = dom->dom_maxrtkey;
186 
187 	rn_init(max_keylen);	/* init all zeroes, all ones, mask table */
188 }
189 SYSINIT(route_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_THIRD, route_init, 0);
190 
191 static void
192 vnet_route_init(const void *unused __unused)
193 {
194 	struct domain *dom;
195 	struct radix_node_head **rnh;
196 	int table;
197 	int fam;
198 
199 	V_rt_tables = malloc(rt_numfibs * (AF_MAX+1) *
200 	    sizeof(struct radix_node_head *), M_RTABLE, M_WAITOK|M_ZERO);
201 
202 	V_rtzone = uma_zcreate("rtentry", sizeof(struct rtentry), NULL, NULL,
203 	    NULL, NULL, UMA_ALIGN_PTR, 0);
204 	for (dom = domains; dom; dom = dom->dom_next) {
205 		if (dom->dom_rtattach)  {
206 			for  (table = 0; table < rt_numfibs; table++) {
207 				if ( (fam = dom->dom_family) == AF_INET ||
208 				    table == 0) {
209  			        	/* for now only AF_INET has > 1 table */
210 					/* XXX MRT
211 					 * rtattach will be also called
212 					 * from vfs_export.c but the
213 					 * offset will be 0
214 					 * (only for AF_INET and AF_INET6
215 					 * which don't need it anyhow)
216 					 */
217 					rnh = rt_tables_get_rnh_ptr(table, fam);
218 					if (rnh == NULL)
219 						panic("%s: rnh NULL", __func__);
220 					dom->dom_rtattach((void **)rnh,
221 				    	    dom->dom_rtoffset);
222 				} else {
223 					break;
224 				}
225 			}
226 		}
227 	}
228 }
229 VNET_SYSINIT(vnet_route_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_FOURTH,
230     vnet_route_init, 0);
231 
232 #ifdef VIMAGE
233 static void
234 vnet_route_uninit(const void *unused __unused)
235 {
236 	int table;
237 	int fam;
238 	struct domain *dom;
239 	struct radix_node_head **rnh;
240 
241 	for (dom = domains; dom; dom = dom->dom_next) {
242 		if (dom->dom_rtdetach) {
243 			for (table = 0; table < rt_numfibs; table++) {
244 				if ( (fam = dom->dom_family) == AF_INET ||
245 				    table == 0) {
246 					/* For now only AF_INET has > 1 tbl. */
247 					rnh = rt_tables_get_rnh_ptr(table, fam);
248 					if (rnh == NULL)
249 						panic("%s: rnh NULL", __func__);
250 					dom->dom_rtdetach((void **)rnh,
251 					    dom->dom_rtoffset);
252 				} else {
253 					break;
254 				}
255 			}
256 		}
257 	}
258 }
259 VNET_SYSUNINIT(vnet_route_uninit, SI_SUB_PROTO_DOMAIN, SI_ORDER_THIRD,
260     vnet_route_uninit, 0);
261 #endif
262 
263 #ifndef _SYS_SYSPROTO_H_
264 struct setfib_args {
265 	int     fibnum;
266 };
267 #endif
268 int
269 setfib(struct thread *td, struct setfib_args *uap)
270 {
271 	if (uap->fibnum < 0 || uap->fibnum >= rt_numfibs)
272 		return EINVAL;
273 	td->td_proc->p_fibnum = uap->fibnum;
274 	return (0);
275 }
276 
277 /*
278  * Packet routing routines.
279  */
280 void
281 rtalloc(struct route *ro)
282 {
283 	rtalloc_ign_fib(ro, 0UL, 0);
284 }
285 
286 void
287 rtalloc_fib(struct route *ro, u_int fibnum)
288 {
289 	rtalloc_ign_fib(ro, 0UL, fibnum);
290 }
291 
292 void
293 rtalloc_ign(struct route *ro, u_long ignore)
294 {
295 	struct rtentry *rt;
296 
297 	if ((rt = ro->ro_rt) != NULL) {
298 		if (rt->rt_ifp != NULL && rt->rt_flags & RTF_UP)
299 			return;
300 		RTFREE(rt);
301 		ro->ro_rt = NULL;
302 	}
303 	ro->ro_rt = rtalloc1_fib(&ro->ro_dst, 1, ignore, 0);
304 	if (ro->ro_rt)
305 		RT_UNLOCK(ro->ro_rt);
306 }
307 
308 void
309 rtalloc_ign_fib(struct route *ro, u_long ignore, u_int fibnum)
310 {
311 	struct rtentry *rt;
312 
313 	if ((rt = ro->ro_rt) != NULL) {
314 		if (rt->rt_ifp != NULL && rt->rt_flags & RTF_UP)
315 			return;
316 		RTFREE(rt);
317 		ro->ro_rt = NULL;
318 	}
319 	ro->ro_rt = rtalloc1_fib(&ro->ro_dst, 1, ignore, fibnum);
320 	if (ro->ro_rt)
321 		RT_UNLOCK(ro->ro_rt);
322 }
323 
324 /*
325  * Look up the route that matches the address given
326  * Or, at least try.. Create a cloned route if needed.
327  *
328  * The returned route, if any, is locked.
329  */
330 struct rtentry *
331 rtalloc1(struct sockaddr *dst, int report, u_long ignflags)
332 {
333 	return (rtalloc1_fib(dst, report, ignflags, 0));
334 }
335 
336 struct rtentry *
337 rtalloc1_fib(struct sockaddr *dst, int report, u_long ignflags,
338 		    u_int fibnum)
339 {
340 	struct radix_node_head *rnh;
341 	struct radix_node *rn;
342 	struct rtentry *newrt;
343 	struct rt_addrinfo info;
344 	int err = 0, msgtype = RTM_MISS;
345 	int needlock;
346 
347 	KASSERT((fibnum < rt_numfibs), ("rtalloc1_fib: bad fibnum"));
348 	if (dst->sa_family != AF_INET)	/* Only INET supports > 1 fib now */
349 		fibnum = 0;
350 	rnh = rt_tables_get_rnh(fibnum, dst->sa_family);
351 	newrt = NULL;
352 	if (rnh == NULL)
353 		goto miss;
354 
355 	/*
356 	 * Look up the address in the table for that Address Family
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 = 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 miss:
383 	V_rtstat.rts_unreach++;
384 
385 	if (report) {
386 		/*
387 		 * If required, report the failure to the supervising
388 		 * Authorities.
389 		 * For a delete, this is not an error. (report == 0)
390 		 */
391 		bzero(&info, sizeof(info));
392 		info.rti_info[RTAX_DST] = dst;
393 		rt_missmsg(msgtype, &info, 0, err);
394 	}
395 done:
396 	if (newrt)
397 		RT_LOCK_ASSERT(newrt);
398 	return (newrt);
399 }
400 
401 /*
402  * Remove a reference count from an rtentry.
403  * If the count gets low enough, take it out of the routing table
404  */
405 void
406 rtfree(struct rtentry *rt)
407 {
408 	struct radix_node_head *rnh;
409 
410 	KASSERT(rt != NULL,("%s: NULL rt", __func__));
411 	rnh = rt_tables_get_rnh(rt->rt_fibnum, rt_key(rt)->sa_family);
412 	KASSERT(rnh != NULL,("%s: NULL rnh", __func__));
413 
414 	RT_LOCK_ASSERT(rt);
415 
416 	/*
417 	 * The callers should use RTFREE_LOCKED() or RTFREE(), so
418 	 * we should come here exactly with the last reference.
419 	 */
420 	RT_REMREF(rt);
421 	if (rt->rt_refcnt > 0) {
422 		log(LOG_DEBUG, "%s: %p has %d refs\n", __func__, rt, rt->rt_refcnt);
423 		goto done;
424 	}
425 
426 	/*
427 	 * On last reference give the "close method" a chance
428 	 * to cleanup private state.  This also permits (for
429 	 * IPv4 and IPv6) a chance to decide if the routing table
430 	 * entry should be purged immediately or at a later time.
431 	 * When an immediate purge is to happen the close routine
432 	 * typically calls rtexpunge which clears the RTF_UP flag
433 	 * on the entry so that the code below reclaims the storage.
434 	 */
435 	if (rt->rt_refcnt == 0 && rnh->rnh_close)
436 		rnh->rnh_close((struct radix_node *)rt, rnh);
437 
438 	/*
439 	 * If we are no longer "up" (and ref == 0)
440 	 * then we can free the resources associated
441 	 * with the route.
442 	 */
443 	if ((rt->rt_flags & RTF_UP) == 0) {
444 		if (rt->rt_nodes->rn_flags & (RNF_ACTIVE | RNF_ROOT))
445 			panic("rtfree 2");
446 		/*
447 		 * the rtentry must have been removed from the routing table
448 		 * so it is represented in rttrash.. remove that now.
449 		 */
450 		V_rttrash--;
451 #ifdef	DIAGNOSTIC
452 		if (rt->rt_refcnt < 0) {
453 			printf("rtfree: %p not freed (neg refs)\n", rt);
454 			goto done;
455 		}
456 #endif
457 		/*
458 		 * release references on items we hold them on..
459 		 * e.g other routes and ifaddrs.
460 		 */
461 		if (rt->rt_ifa)
462 			ifa_free(rt->rt_ifa);
463 		/*
464 		 * The key is separatly alloc'd so free it (see rt_setgate()).
465 		 * This also frees the gateway, as they are always malloc'd
466 		 * together.
467 		 */
468 		Free(rt_key(rt));
469 
470 		/*
471 		 * and the rtentry itself of course
472 		 */
473 		RT_LOCK_DESTROY(rt);
474 		uma_zfree(V_rtzone, rt);
475 		return;
476 	}
477 done:
478 	RT_UNLOCK(rt);
479 }
480 
481 
482 /*
483  * Force a routing table entry to the specified
484  * destination to go through the given gateway.
485  * Normally called as a result of a routing redirect
486  * message from the network layer.
487  */
488 void
489 rtredirect(struct sockaddr *dst,
490 	struct sockaddr *gateway,
491 	struct sockaddr *netmask,
492 	int flags,
493 	struct sockaddr *src)
494 {
495 	rtredirect_fib(dst, gateway, netmask, flags, src, 0);
496 }
497 
498 void
499 rtredirect_fib(struct sockaddr *dst,
500 	struct sockaddr *gateway,
501 	struct sockaddr *netmask,
502 	int flags,
503 	struct sockaddr *src,
504 	u_int fibnum)
505 {
506 	struct rtentry *rt, *rt0 = NULL;
507 	int error = 0;
508 	short *stat = NULL;
509 	struct rt_addrinfo info;
510 	struct ifaddr *ifa;
511 	struct radix_node_head *rnh;
512 
513 	ifa = NULL;
514 	rnh = rt_tables_get_rnh(fibnum, dst->sa_family);
515 	if (rnh == NULL) {
516 		error = EAFNOSUPPORT;
517 		goto out;
518 	}
519 
520 	/* verify the gateway is directly reachable */
521 	if ((ifa = ifa_ifwithnet(gateway, 0)) == NULL) {
522 		error = ENETUNREACH;
523 		goto out;
524 	}
525 	rt = rtalloc1_fib(dst, 0, 0UL, fibnum);	/* NB: rt is locked */
526 	/*
527 	 * If the redirect isn't from our current router for this dst,
528 	 * it's either old or wrong.  If it redirects us to ourselves,
529 	 * we have a routing loop, perhaps as a result of an interface
530 	 * going down recently.
531 	 */
532 	if (!(flags & RTF_DONE) && rt &&
533 	     (!sa_equal(src, rt->rt_gateway) || rt->rt_ifa != ifa))
534 		error = EINVAL;
535 	else if (ifa_ifwithaddr_check(gateway))
536 		error = EHOSTUNREACH;
537 	if (error)
538 		goto done;
539 	/*
540 	 * Create a new entry if we just got back a wildcard entry
541 	 * or the lookup failed.  This is necessary for hosts
542 	 * which use routing redirects generated by smart gateways
543 	 * to dynamically build the routing tables.
544 	 */
545 	if (rt == NULL || (rt_mask(rt) && rt_mask(rt)->sa_len < 2))
546 		goto create;
547 	/*
548 	 * Don't listen to the redirect if it's
549 	 * for a route to an interface.
550 	 */
551 	if (rt->rt_flags & RTF_GATEWAY) {
552 		if (((rt->rt_flags & RTF_HOST) == 0) && (flags & RTF_HOST)) {
553 			/*
554 			 * Changing from route to net => route to host.
555 			 * Create new route, rather than smashing route to net.
556 			 */
557 		create:
558 			rt0 = rt;
559 			rt = NULL;
560 
561 			flags |=  RTF_GATEWAY | RTF_DYNAMIC;
562 			bzero((caddr_t)&info, sizeof(info));
563 			info.rti_info[RTAX_DST] = dst;
564 			info.rti_info[RTAX_GATEWAY] = gateway;
565 			info.rti_info[RTAX_NETMASK] = netmask;
566 			info.rti_ifa = ifa;
567 			info.rti_flags = flags;
568 			if (rt0 != NULL)
569 				RT_UNLOCK(rt0);	/* drop lock to avoid LOR with RNH */
570 			error = rtrequest1_fib(RTM_ADD, &info, &rt, fibnum);
571 			if (rt != NULL) {
572 				RT_LOCK(rt);
573 				if (rt0 != NULL)
574 					EVENTHANDLER_INVOKE(route_redirect_event, rt0, rt, dst);
575 				flags = rt->rt_flags;
576 			}
577 			if (rt0 != NULL)
578 				RTFREE(rt0);
579 
580 			stat = &V_rtstat.rts_dynamic;
581 		} else {
582 			struct rtentry *gwrt;
583 
584 			/*
585 			 * Smash the current notion of the gateway to
586 			 * this destination.  Should check about netmask!!!
587 			 */
588 			rt->rt_flags |= RTF_MODIFIED;
589 			flags |= RTF_MODIFIED;
590 			stat = &V_rtstat.rts_newgateway;
591 			/*
592 			 * add the key and gateway (in one malloc'd chunk).
593 			 */
594 			RT_UNLOCK(rt);
595 			RADIX_NODE_HEAD_LOCK(rnh);
596 			RT_LOCK(rt);
597 			rt_setgate(rt, rt_key(rt), gateway);
598 			gwrt = rtalloc1(gateway, 1, RTF_RNH_LOCKED);
599 			RADIX_NODE_HEAD_UNLOCK(rnh);
600 			EVENTHANDLER_INVOKE(route_redirect_event, rt, gwrt, dst);
601 			RTFREE_LOCKED(gwrt);
602 		}
603 	} else
604 		error = EHOSTUNREACH;
605 done:
606 	if (rt)
607 		RTFREE_LOCKED(rt);
608 out:
609 	if (error)
610 		V_rtstat.rts_badredirect++;
611 	else if (stat != NULL)
612 		(*stat)++;
613 	bzero((caddr_t)&info, sizeof(info));
614 	info.rti_info[RTAX_DST] = dst;
615 	info.rti_info[RTAX_GATEWAY] = gateway;
616 	info.rti_info[RTAX_NETMASK] = netmask;
617 	info.rti_info[RTAX_AUTHOR] = src;
618 	rt_missmsg(RTM_REDIRECT, &info, flags, error);
619 	if (ifa != NULL)
620 		ifa_free(ifa);
621 }
622 
623 int
624 rtioctl(u_long req, caddr_t data)
625 {
626 	return (rtioctl_fib(req, data, 0));
627 }
628 
629 /*
630  * Routing table ioctl interface.
631  */
632 int
633 rtioctl_fib(u_long req, caddr_t data, u_int fibnum)
634 {
635 
636 	/*
637 	 * If more ioctl commands are added here, make sure the proper
638 	 * super-user checks are being performed because it is possible for
639 	 * prison-root to make it this far if raw sockets have been enabled
640 	 * in jails.
641 	 */
642 #ifdef INET
643 	/* Multicast goop, grrr... */
644 	return mrt_ioctl ? mrt_ioctl(req, data, fibnum) : EOPNOTSUPP;
645 #else /* INET */
646 	return ENXIO;
647 #endif /* INET */
648 }
649 
650 /*
651  * For both ifa_ifwithroute() routines, 'ifa' is returned referenced.
652  */
653 struct ifaddr *
654 ifa_ifwithroute(int flags, struct sockaddr *dst, struct sockaddr *gateway)
655 {
656 	return (ifa_ifwithroute_fib(flags, dst, gateway, 0));
657 }
658 
659 struct ifaddr *
660 ifa_ifwithroute_fib(int flags, struct sockaddr *dst, struct sockaddr *gateway,
661 				u_int fibnum)
662 {
663 	register struct ifaddr *ifa;
664 	int not_found = 0;
665 
666 	if ((flags & RTF_GATEWAY) == 0) {
667 		/*
668 		 * If we are adding a route to an interface,
669 		 * and the interface is a pt to pt link
670 		 * we should search for the destination
671 		 * as our clue to the interface.  Otherwise
672 		 * we can use the local address.
673 		 */
674 		ifa = NULL;
675 		if (flags & RTF_HOST)
676 			ifa = ifa_ifwithdstaddr(dst);
677 		if (ifa == NULL)
678 			ifa = ifa_ifwithaddr(gateway);
679 	} else {
680 		/*
681 		 * If we are adding a route to a remote net
682 		 * or host, the gateway may still be on the
683 		 * other end of a pt to pt link.
684 		 */
685 		ifa = ifa_ifwithdstaddr(gateway);
686 	}
687 	if (ifa == NULL)
688 		ifa = ifa_ifwithnet(gateway, 0);
689 	if (ifa == NULL) {
690 		struct rtentry *rt = rtalloc1_fib(gateway, 0, RTF_RNH_LOCKED, fibnum);
691 		if (rt == NULL)
692 			return (NULL);
693 		/*
694 		 * dismiss a gateway that is reachable only
695 		 * through the default router
696 		 */
697 		switch (gateway->sa_family) {
698 		case AF_INET:
699 			if (satosin(rt_key(rt))->sin_addr.s_addr == INADDR_ANY)
700 				not_found = 1;
701 			break;
702 		case AF_INET6:
703 			if (IN6_IS_ADDR_UNSPECIFIED(&satosin6(rt_key(rt))->sin6_addr))
704 				not_found = 1;
705 			break;
706 		default:
707 			break;
708 		}
709 		if (!not_found && rt->rt_ifa != NULL) {
710 			ifa = rt->rt_ifa;
711 			ifa_ref(ifa);
712 		}
713 		RT_REMREF(rt);
714 		RT_UNLOCK(rt);
715 		if (not_found || ifa == NULL)
716 			return (NULL);
717 	}
718 	if (ifa->ifa_addr->sa_family != dst->sa_family) {
719 		struct ifaddr *oifa = ifa;
720 		ifa = ifaof_ifpforaddr(dst, ifa->ifa_ifp);
721 		if (ifa == NULL)
722 			ifa = oifa;
723 		else
724 			ifa_free(oifa);
725 	}
726 	return (ifa);
727 }
728 
729 /*
730  * Do appropriate manipulations of a routing tree given
731  * all the bits of info needed
732  */
733 int
734 rtrequest(int req,
735 	struct sockaddr *dst,
736 	struct sockaddr *gateway,
737 	struct sockaddr *netmask,
738 	int flags,
739 	struct rtentry **ret_nrt)
740 {
741 	return (rtrequest_fib(req, dst, gateway, netmask, flags, ret_nrt, 0));
742 }
743 
744 int
745 rtrequest_fib(int req,
746 	struct sockaddr *dst,
747 	struct sockaddr *gateway,
748 	struct sockaddr *netmask,
749 	int flags,
750 	struct rtentry **ret_nrt,
751 	u_int fibnum)
752 {
753 	struct rt_addrinfo info;
754 
755 	if (dst->sa_len == 0)
756 		return(EINVAL);
757 
758 	bzero((caddr_t)&info, sizeof(info));
759 	info.rti_flags = flags;
760 	info.rti_info[RTAX_DST] = dst;
761 	info.rti_info[RTAX_GATEWAY] = gateway;
762 	info.rti_info[RTAX_NETMASK] = netmask;
763 	return rtrequest1_fib(req, &info, ret_nrt, fibnum);
764 }
765 
766 /*
767  * These (questionable) definitions of apparent local variables apply
768  * to the next two functions.  XXXXXX!!!
769  */
770 #define	dst	info->rti_info[RTAX_DST]
771 #define	gateway	info->rti_info[RTAX_GATEWAY]
772 #define	netmask	info->rti_info[RTAX_NETMASK]
773 #define	ifaaddr	info->rti_info[RTAX_IFA]
774 #define	ifpaddr	info->rti_info[RTAX_IFP]
775 #define	flags	info->rti_flags
776 
777 int
778 rt_getifa(struct rt_addrinfo *info)
779 {
780 	return (rt_getifa_fib(info, 0));
781 }
782 
783 /*
784  * Look up rt_addrinfo for a specific fib.  Note that if rti_ifa is defined,
785  * it will be referenced so the caller must free it.
786  */
787 int
788 rt_getifa_fib(struct rt_addrinfo *info, u_int fibnum)
789 {
790 	struct ifaddr *ifa;
791 	int error = 0;
792 
793 	/*
794 	 * ifp may be specified by sockaddr_dl
795 	 * when protocol address is ambiguous.
796 	 */
797 	if (info->rti_ifp == NULL && ifpaddr != NULL &&
798 	    ifpaddr->sa_family == AF_LINK &&
799 	    (ifa = ifa_ifwithnet(ifpaddr, 0)) != NULL) {
800 		info->rti_ifp = ifa->ifa_ifp;
801 		ifa_free(ifa);
802 	}
803 	if (info->rti_ifa == NULL && ifaaddr != NULL)
804 		info->rti_ifa = ifa_ifwithaddr(ifaaddr);
805 	if (info->rti_ifa == NULL) {
806 		struct sockaddr *sa;
807 
808 		sa = ifaaddr != NULL ? ifaaddr :
809 		    (gateway != NULL ? gateway : dst);
810 		if (sa != NULL && info->rti_ifp != NULL)
811 			info->rti_ifa = ifaof_ifpforaddr(sa, info->rti_ifp);
812 		else if (dst != NULL && gateway != NULL)
813 			info->rti_ifa = ifa_ifwithroute_fib(flags, dst, gateway,
814 							fibnum);
815 		else if (sa != NULL)
816 			info->rti_ifa = ifa_ifwithroute_fib(flags, sa, sa,
817 							fibnum);
818 	}
819 	if ((ifa = info->rti_ifa) != NULL) {
820 		if (info->rti_ifp == NULL)
821 			info->rti_ifp = ifa->ifa_ifp;
822 	} else
823 		error = ENETUNREACH;
824 	return (error);
825 }
826 
827 /*
828  * Expunges references to a route that's about to be reclaimed.
829  * The route must be locked.
830  */
831 int
832 rtexpunge(struct rtentry *rt)
833 {
834 #if !defined(RADIX_MPATH)
835 	struct radix_node *rn;
836 #else
837 	struct rt_addrinfo info;
838 	int fib;
839 	struct rtentry *rt0;
840 #endif
841 	struct radix_node_head *rnh;
842 	struct ifaddr *ifa;
843 	int error = 0;
844 
845 	/*
846 	 * Find the correct routing tree to use for this Address Family
847 	 */
848 	rnh = rt_tables_get_rnh(rt->rt_fibnum, rt_key(rt)->sa_family);
849 	RT_LOCK_ASSERT(rt);
850 	if (rnh == NULL)
851 		return (EAFNOSUPPORT);
852 	RADIX_NODE_HEAD_LOCK_ASSERT(rnh);
853 
854 #ifdef RADIX_MPATH
855 	fib = rt->rt_fibnum;
856 	bzero(&info, sizeof(info));
857 	info.rti_ifp = rt->rt_ifp;
858 	info.rti_flags = RTF_RNH_LOCKED;
859 	info.rti_info[RTAX_DST] = rt_key(rt);
860 	info.rti_info[RTAX_GATEWAY] = rt->rt_ifa->ifa_addr;
861 
862 	RT_UNLOCK(rt);
863 	error = rtrequest1_fib(RTM_DELETE, &info, &rt0, fib);
864 
865 	if (error == 0 && rt0 != NULL) {
866 		rt = rt0;
867 		RT_LOCK(rt);
868 	} else if (error != 0) {
869 		RT_LOCK(rt);
870 		return (error);
871 	}
872 #else
873 	/*
874 	 * Remove the item from the tree; it should be there,
875 	 * but when callers invoke us blindly it may not (sigh).
876 	 */
877 	rn = rnh->rnh_deladdr(rt_key(rt), rt_mask(rt), rnh);
878 	if (rn == NULL) {
879 		error = ESRCH;
880 		goto bad;
881 	}
882 	KASSERT((rn->rn_flags & (RNF_ACTIVE | RNF_ROOT)) == 0,
883 		("unexpected flags 0x%x", rn->rn_flags));
884 	KASSERT(rt == RNTORT(rn),
885 		("lookup mismatch, rt %p rn %p", rt, rn));
886 #endif /* RADIX_MPATH */
887 
888 	rt->rt_flags &= ~RTF_UP;
889 
890 	/*
891 	 * Give the protocol a chance to keep things in sync.
892 	 */
893 	if ((ifa = rt->rt_ifa) && ifa->ifa_rtrequest) {
894 		struct rt_addrinfo info;
895 
896 		bzero((caddr_t)&info, sizeof(info));
897 		info.rti_flags = rt->rt_flags;
898 		info.rti_info[RTAX_DST] = rt_key(rt);
899 		info.rti_info[RTAX_GATEWAY] = rt->rt_gateway;
900 		info.rti_info[RTAX_NETMASK] = rt_mask(rt);
901 		ifa->ifa_rtrequest(RTM_DELETE, rt, &info);
902 	}
903 
904 	/*
905 	 * one more rtentry floating around that is not
906 	 * linked to the routing table.
907 	 */
908 	V_rttrash++;
909 #if !defined(RADIX_MPATH)
910 bad:
911 #endif
912 	return (error);
913 }
914 
915 #ifdef RADIX_MPATH
916 static int
917 rn_mpath_update(int req, struct rt_addrinfo *info,
918     struct radix_node_head *rnh, struct rtentry **ret_nrt)
919 {
920 	/*
921 	 * if we got multipath routes, we require users to specify
922 	 * a matching RTAX_GATEWAY.
923 	 */
924 	struct rtentry *rt, *rto = NULL;
925 	register struct radix_node *rn;
926 	int error = 0;
927 
928 	rn = rnh->rnh_matchaddr(dst, rnh);
929 	if (rn == NULL)
930 		return (ESRCH);
931 	rto = rt = RNTORT(rn);
932 	rt = rt_mpath_matchgate(rt, gateway);
933 	if (rt == NULL)
934 		return (ESRCH);
935 	/*
936 	 * this is the first entry in the chain
937 	 */
938 	if (rto == rt) {
939 		rn = rn_mpath_next((struct radix_node *)rt);
940 		/*
941 		 * there is another entry, now it's active
942 		 */
943 		if (rn) {
944 			rto = RNTORT(rn);
945 			RT_LOCK(rto);
946 			rto->rt_flags |= RTF_UP;
947 			RT_UNLOCK(rto);
948 		} else if (rt->rt_flags & RTF_GATEWAY) {
949 			/*
950 			 * For gateway routes, we need to
951 			 * make sure that we we are deleting
952 			 * the correct gateway.
953 			 * rt_mpath_matchgate() does not
954 			 * check the case when there is only
955 			 * one route in the chain.
956 			 */
957 			if (gateway &&
958 			    (rt->rt_gateway->sa_len != gateway->sa_len ||
959 				memcmp(rt->rt_gateway, gateway, gateway->sa_len)))
960 				error = ESRCH;
961 			else {
962 				/*
963 				 * remove from tree before returning it
964 				 * to the caller
965 				 */
966 				rn = rnh->rnh_deladdr(dst, netmask, rnh);
967 				KASSERT(rt == RNTORT(rn), ("radix node disappeared"));
968 				goto gwdelete;
969 			}
970 
971 		}
972 		/*
973 		 * use the normal delete code to remove
974 		 * the first entry
975 		 */
976 		if (req != RTM_DELETE)
977 			goto nondelete;
978 
979 		error = ENOENT;
980 		goto done;
981 	}
982 
983 	/*
984 	 * if the entry is 2nd and on up
985 	 */
986 	if ((req == RTM_DELETE) && !rt_mpath_deldup(rto, rt))
987 		panic ("rtrequest1: rt_mpath_deldup");
988 gwdelete:
989 	RT_LOCK(rt);
990 	RT_ADDREF(rt);
991 	if (req == RTM_DELETE) {
992 		rt->rt_flags &= ~RTF_UP;
993 		/*
994 		 * One more rtentry floating around that is not
995 		 * linked to the routing table. rttrash will be decremented
996 		 * when RTFREE(rt) is eventually called.
997 		 */
998 		V_rttrash++;
999 	}
1000 
1001 nondelete:
1002 	if (req != RTM_DELETE)
1003 		panic("unrecognized request %d", req);
1004 
1005 
1006 	/*
1007 	 * If the caller wants it, then it can have it,
1008 	 * but it's up to it to free the rtentry as we won't be
1009 	 * doing it.
1010 	 */
1011 	if (ret_nrt) {
1012 		*ret_nrt = rt;
1013 		RT_UNLOCK(rt);
1014 	} else
1015 		RTFREE_LOCKED(rt);
1016 done:
1017 	return (error);
1018 }
1019 #endif
1020 
1021 int
1022 rtrequest1_fib(int req, struct rt_addrinfo *info, struct rtentry **ret_nrt,
1023 				u_int fibnum)
1024 {
1025 	int error = 0, needlock = 0;
1026 	register struct rtentry *rt;
1027 #ifdef FLOWTABLE
1028 	register struct rtentry *rt0;
1029 #endif
1030 	register struct radix_node *rn;
1031 	register struct radix_node_head *rnh;
1032 	struct ifaddr *ifa;
1033 	struct sockaddr *ndst;
1034 #define senderr(x) { error = x ; goto bad; }
1035 
1036 	KASSERT((fibnum < rt_numfibs), ("rtrequest1_fib: bad fibnum"));
1037 	if (dst->sa_family != AF_INET)	/* Only INET supports > 1 fib now */
1038 		fibnum = 0;
1039 	/*
1040 	 * Find the correct routing tree to use for this Address Family
1041 	 */
1042 	rnh = rt_tables_get_rnh(fibnum, dst->sa_family);
1043 	if (rnh == NULL)
1044 		return (EAFNOSUPPORT);
1045 	needlock = ((flags & RTF_RNH_LOCKED) == 0);
1046 	flags &= ~RTF_RNH_LOCKED;
1047 	if (needlock)
1048 		RADIX_NODE_HEAD_LOCK(rnh);
1049 	else
1050 		RADIX_NODE_HEAD_LOCK_ASSERT(rnh);
1051 	/*
1052 	 * If we are adding a host route then we don't want to put
1053 	 * a netmask in the tree, nor do we want to clone it.
1054 	 */
1055 	if (flags & RTF_HOST)
1056 		netmask = NULL;
1057 
1058 	switch (req) {
1059 	case RTM_DELETE:
1060 #ifdef RADIX_MPATH
1061 		if (rn_mpath_capable(rnh)) {
1062 			error = rn_mpath_update(req, info, rnh, ret_nrt);
1063 			/*
1064 			 * "bad" holds true for the success case
1065 			 * as well
1066 			 */
1067 			if (error != ENOENT)
1068 				goto bad;
1069 			error = 0;
1070 		}
1071 #endif
1072 		/*
1073 		 * Remove the item from the tree and return it.
1074 		 * Complain if it is not there and do no more processing.
1075 		 */
1076 		rn = rnh->rnh_deladdr(dst, netmask, rnh);
1077 		if (rn == NULL)
1078 			senderr(ESRCH);
1079 		if (rn->rn_flags & (RNF_ACTIVE | RNF_ROOT))
1080 			panic ("rtrequest delete");
1081 		rt = RNTORT(rn);
1082 		RT_LOCK(rt);
1083 		RT_ADDREF(rt);
1084 		rt->rt_flags &= ~RTF_UP;
1085 
1086 		/*
1087 		 * give the protocol a chance to keep things in sync.
1088 		 */
1089 		if ((ifa = rt->rt_ifa) && ifa->ifa_rtrequest)
1090 			ifa->ifa_rtrequest(RTM_DELETE, rt, info);
1091 
1092 		/*
1093 		 * One more rtentry floating around that is not
1094 		 * linked to the routing table. rttrash will be decremented
1095 		 * when RTFREE(rt) is eventually called.
1096 		 */
1097 		V_rttrash++;
1098 
1099 		/*
1100 		 * If the caller wants it, then it can have it,
1101 		 * but it's up to it to free the rtentry as we won't be
1102 		 * doing it.
1103 		 */
1104 		if (ret_nrt) {
1105 			*ret_nrt = rt;
1106 			RT_UNLOCK(rt);
1107 		} else
1108 			RTFREE_LOCKED(rt);
1109 		break;
1110 	case RTM_RESOLVE:
1111 		/*
1112 		 * resolve was only used for route cloning
1113 		 * here for compat
1114 		 */
1115 		break;
1116 	case RTM_ADD:
1117 		if ((flags & RTF_GATEWAY) && !gateway)
1118 			senderr(EINVAL);
1119 		if (dst && gateway && (dst->sa_family != gateway->sa_family) &&
1120 		    (gateway->sa_family != AF_UNSPEC) && (gateway->sa_family != AF_LINK))
1121 			senderr(EINVAL);
1122 
1123 		if (info->rti_ifa == NULL) {
1124 			error = rt_getifa_fib(info, fibnum);
1125 			if (error)
1126 				senderr(error);
1127 		} else
1128 			ifa_ref(info->rti_ifa);
1129 		ifa = info->rti_ifa;
1130 		rt = uma_zalloc(V_rtzone, M_NOWAIT | M_ZERO);
1131 		if (rt == NULL) {
1132 			if (ifa != NULL)
1133 				ifa_free(ifa);
1134 			senderr(ENOBUFS);
1135 		}
1136 		RT_LOCK_INIT(rt);
1137 		rt->rt_flags = RTF_UP | flags;
1138 		rt->rt_fibnum = fibnum;
1139 		/*
1140 		 * Add the gateway. Possibly re-malloc-ing the storage for it
1141 		 *
1142 		 */
1143 		RT_LOCK(rt);
1144 		if ((error = rt_setgate(rt, dst, gateway)) != 0) {
1145 			RT_LOCK_DESTROY(rt);
1146 			if (ifa != NULL)
1147 				ifa_free(ifa);
1148 			uma_zfree(V_rtzone, rt);
1149 			senderr(error);
1150 		}
1151 
1152 		/*
1153 		 * point to the (possibly newly malloc'd) dest address.
1154 		 */
1155 		ndst = (struct sockaddr *)rt_key(rt);
1156 
1157 		/*
1158 		 * make sure it contains the value we want (masked if needed).
1159 		 */
1160 		if (netmask) {
1161 			rt_maskedcopy(dst, ndst, netmask);
1162 		} else
1163 			bcopy(dst, ndst, dst->sa_len);
1164 
1165 		/*
1166 		 * We use the ifa reference returned by rt_getifa_fib().
1167 		 * This moved from below so that rnh->rnh_addaddr() can
1168 		 * examine the ifa and  ifa->ifa_ifp if it so desires.
1169 		 */
1170 		rt->rt_ifa = ifa;
1171 		rt->rt_ifp = ifa->ifa_ifp;
1172 		rt->rt_rmx.rmx_weight = 1;
1173 
1174 #ifdef RADIX_MPATH
1175 		/* do not permit exactly the same dst/mask/gw pair */
1176 		if (rn_mpath_capable(rnh) &&
1177 			rt_mpath_conflict(rnh, rt, netmask)) {
1178 			if (rt->rt_ifa) {
1179 				ifa_free(rt->rt_ifa);
1180 			}
1181 			Free(rt_key(rt));
1182 			RT_LOCK_DESTROY(rt);
1183 			uma_zfree(V_rtzone, rt);
1184 			senderr(EEXIST);
1185 		}
1186 #endif
1187 
1188 #ifdef FLOWTABLE
1189 		rt0 = NULL;
1190 		/* XXX
1191 		 * "flow-table" only support IPv4 at the moment.
1192 		 */
1193 #ifdef INET
1194 		if (dst->sa_family == AF_INET) {
1195 			rn = rnh->rnh_matchaddr(dst, rnh);
1196 			if (rn && ((rn->rn_flags & RNF_ROOT) == 0)) {
1197 				struct sockaddr *mask;
1198 				u_char *m, *n;
1199 				int len;
1200 
1201 				/*
1202 				 * compare mask to see if the new route is
1203 				 * more specific than the existing one
1204 				 */
1205 				rt0 = RNTORT(rn);
1206 				RT_LOCK(rt0);
1207 				RT_ADDREF(rt0);
1208 				RT_UNLOCK(rt0);
1209 				/*
1210 				 * A host route is already present, so
1211 				 * leave the flow-table entries as is.
1212 				 */
1213 				if (rt0->rt_flags & RTF_HOST) {
1214 					RTFREE(rt0);
1215 					rt0 = NULL;
1216 				} else if (!(flags & RTF_HOST) && netmask) {
1217 					mask = rt_mask(rt0);
1218 					len = mask->sa_len;
1219 					m = (u_char *)mask;
1220 					n = (u_char *)netmask;
1221 					while (len-- > 0) {
1222 						if (*n != *m)
1223 							break;
1224 						n++;
1225 						m++;
1226 					}
1227 					if (len == 0 || (*n < *m)) {
1228 						RTFREE(rt0);
1229 						rt0 = NULL;
1230 					}
1231 				}
1232 			}
1233 		}
1234 #endif
1235 #endif
1236 
1237 		/* XXX mtu manipulation will be done in rnh_addaddr -- itojun */
1238 		rn = rnh->rnh_addaddr(ndst, netmask, rnh, rt->rt_nodes);
1239 		/*
1240 		 * If it still failed to go into the tree,
1241 		 * then un-make it (this should be a function)
1242 		 */
1243 		if (rn == NULL) {
1244 			if (rt->rt_ifa)
1245 				ifa_free(rt->rt_ifa);
1246 			Free(rt_key(rt));
1247 			RT_LOCK_DESTROY(rt);
1248 			uma_zfree(V_rtzone, rt);
1249 #ifdef FLOWTABLE
1250 			if (rt0 != NULL)
1251 				RTFREE(rt0);
1252 #endif
1253 			senderr(EEXIST);
1254 		}
1255 #ifdef FLOWTABLE
1256 		else if (rt0 != NULL) {
1257 #ifdef INET
1258 			flowtable_route_flush(V_ip_ft, rt0);
1259 #endif
1260 			RTFREE(rt0);
1261 		}
1262 #endif
1263 
1264 		/*
1265 		 * If this protocol has something to add to this then
1266 		 * allow it to do that as well.
1267 		 */
1268 		if (ifa->ifa_rtrequest)
1269 			ifa->ifa_rtrequest(req, rt, info);
1270 
1271 		/*
1272 		 * actually return a resultant rtentry and
1273 		 * give the caller a single reference.
1274 		 */
1275 		if (ret_nrt) {
1276 			*ret_nrt = rt;
1277 			RT_ADDREF(rt);
1278 		}
1279 		RT_UNLOCK(rt);
1280 		break;
1281 	default:
1282 		error = EOPNOTSUPP;
1283 	}
1284 bad:
1285 	if (needlock)
1286 		RADIX_NODE_HEAD_UNLOCK(rnh);
1287 	return (error);
1288 #undef senderr
1289 }
1290 
1291 #undef dst
1292 #undef gateway
1293 #undef netmask
1294 #undef ifaaddr
1295 #undef ifpaddr
1296 #undef flags
1297 
1298 int
1299 rt_setgate(struct rtentry *rt, struct sockaddr *dst, struct sockaddr *gate)
1300 {
1301 	/* XXX dst may be overwritten, can we move this to below */
1302 	int dlen = SA_SIZE(dst), glen = SA_SIZE(gate);
1303 #ifdef INVARIANTS
1304 	struct radix_node_head *rnh;
1305 
1306 	rnh = rt_tables_get_rnh(rt->rt_fibnum, dst->sa_family);
1307 #endif
1308 
1309 	RT_LOCK_ASSERT(rt);
1310 	RADIX_NODE_HEAD_LOCK_ASSERT(rnh);
1311 
1312 	/*
1313 	 * Prepare to store the gateway in rt->rt_gateway.
1314 	 * Both dst and gateway are stored one after the other in the same
1315 	 * malloc'd chunk. If we have room, we can reuse the old buffer,
1316 	 * rt_gateway already points to the right place.
1317 	 * Otherwise, malloc a new block and update the 'dst' address.
1318 	 */
1319 	if (rt->rt_gateway == NULL || glen > SA_SIZE(rt->rt_gateway)) {
1320 		caddr_t new;
1321 
1322 		R_Malloc(new, caddr_t, dlen + glen);
1323 		if (new == NULL)
1324 			return ENOBUFS;
1325 		/*
1326 		 * XXX note, we copy from *dst and not *rt_key(rt) because
1327 		 * rt_setgate() can be called to initialize a newly
1328 		 * allocated route entry, in which case rt_key(rt) == NULL
1329 		 * (and also rt->rt_gateway == NULL).
1330 		 * Free()/free() handle a NULL argument just fine.
1331 		 */
1332 		bcopy(dst, new, dlen);
1333 		Free(rt_key(rt));	/* free old block, if any */
1334 		rt_key(rt) = (struct sockaddr *)new;
1335 		rt->rt_gateway = (struct sockaddr *)(new + dlen);
1336 	}
1337 
1338 	/*
1339 	 * Copy the new gateway value into the memory chunk.
1340 	 */
1341 	bcopy(gate, rt->rt_gateway, glen);
1342 
1343 	return (0);
1344 }
1345 
1346 void
1347 rt_maskedcopy(struct sockaddr *src, struct sockaddr *dst, struct sockaddr *netmask)
1348 {
1349 	register u_char *cp1 = (u_char *)src;
1350 	register u_char *cp2 = (u_char *)dst;
1351 	register u_char *cp3 = (u_char *)netmask;
1352 	u_char *cplim = cp2 + *cp3;
1353 	u_char *cplim2 = cp2 + *cp1;
1354 
1355 	*cp2++ = *cp1++; *cp2++ = *cp1++; /* copies sa_len & sa_family */
1356 	cp3 += 2;
1357 	if (cplim > cplim2)
1358 		cplim = cplim2;
1359 	while (cp2 < cplim)
1360 		*cp2++ = *cp1++ & *cp3++;
1361 	if (cp2 < cplim2)
1362 		bzero((caddr_t)cp2, (unsigned)(cplim2 - cp2));
1363 }
1364 
1365 /*
1366  * Set up a routing table entry, normally
1367  * for an interface.
1368  */
1369 #define _SOCKADDR_TMPSIZE 128 /* Not too big.. kernel stack size is limited */
1370 static inline  int
1371 rtinit1(struct ifaddr *ifa, int cmd, int flags, int fibnum)
1372 {
1373 	struct sockaddr *dst;
1374 	struct sockaddr *netmask;
1375 	struct rtentry *rt = NULL;
1376 	struct rt_addrinfo info;
1377 	int error = 0;
1378 	int startfib, endfib;
1379 	char tempbuf[_SOCKADDR_TMPSIZE];
1380 	int didwork = 0;
1381 	int a_failure = 0;
1382 	static struct sockaddr_dl null_sdl = {sizeof(null_sdl), AF_LINK};
1383 
1384 	if (flags & RTF_HOST) {
1385 		dst = ifa->ifa_dstaddr;
1386 		netmask = NULL;
1387 	} else {
1388 		dst = ifa->ifa_addr;
1389 		netmask = ifa->ifa_netmask;
1390 	}
1391 	if ( dst->sa_family != AF_INET)
1392 		fibnum = 0;
1393 	if (fibnum == -1) {
1394 		if (rt_add_addr_allfibs == 0 && cmd == (int)RTM_ADD) {
1395 			startfib = endfib = curthread->td_proc->p_fibnum;
1396 		} else {
1397 			startfib = 0;
1398 			endfib = rt_numfibs - 1;
1399 		}
1400 	} else {
1401 		KASSERT((fibnum < rt_numfibs), ("rtinit1: bad fibnum"));
1402 		startfib = fibnum;
1403 		endfib = fibnum;
1404 	}
1405 	if (dst->sa_len == 0)
1406 		return(EINVAL);
1407 
1408 	/*
1409 	 * If it's a delete, check that if it exists,
1410 	 * it's on the correct interface or we might scrub
1411 	 * a route to another ifa which would
1412 	 * be confusing at best and possibly worse.
1413 	 */
1414 	if (cmd == RTM_DELETE) {
1415 		/*
1416 		 * It's a delete, so it should already exist..
1417 		 * If it's a net, mask off the host bits
1418 		 * (Assuming we have a mask)
1419 		 * XXX this is kinda inet specific..
1420 		 */
1421 		if (netmask != NULL) {
1422 			rt_maskedcopy(dst, (struct sockaddr *)tempbuf, netmask);
1423 			dst = (struct sockaddr *)tempbuf;
1424 		}
1425 	}
1426 	/*
1427 	 * Now go through all the requested tables (fibs) and do the
1428 	 * requested action. Realistically, this will either be fib 0
1429 	 * for protocols that don't do multiple tables or all the
1430 	 * tables for those that do. XXX For this version only AF_INET.
1431 	 * When that changes code should be refactored to protocol
1432 	 * independent parts and protocol dependent parts.
1433 	 */
1434 	for ( fibnum = startfib; fibnum <= endfib; fibnum++) {
1435 		if (cmd == RTM_DELETE) {
1436 			struct radix_node_head *rnh;
1437 			struct radix_node *rn;
1438 			/*
1439 			 * Look up an rtentry that is in the routing tree and
1440 			 * contains the correct info.
1441 			 */
1442 			rnh = rt_tables_get_rnh(fibnum, dst->sa_family);
1443 			if (rnh == NULL)
1444 				/* this table doesn't exist but others might */
1445 				continue;
1446 			RADIX_NODE_HEAD_LOCK(rnh);
1447 #ifdef RADIX_MPATH
1448 			if (rn_mpath_capable(rnh)) {
1449 
1450 				rn = rnh->rnh_matchaddr(dst, rnh);
1451 				if (rn == NULL)
1452 					error = ESRCH;
1453 				else {
1454 					rt = RNTORT(rn);
1455 					/*
1456 					 * for interface route the
1457 					 * rt->rt_gateway is sockaddr_intf
1458 					 * for cloning ARP entries, so
1459 					 * rt_mpath_matchgate must use the
1460 					 * interface address
1461 					 */
1462 					rt = rt_mpath_matchgate(rt,
1463 					    ifa->ifa_addr);
1464 					if (!rt)
1465 						error = ESRCH;
1466 				}
1467 			}
1468 			else
1469 #endif
1470 			rn = rnh->rnh_lookup(dst, netmask, rnh);
1471 			error = (rn == NULL ||
1472 			    (rn->rn_flags & RNF_ROOT) ||
1473 			    RNTORT(rn)->rt_ifa != ifa ||
1474 			    !sa_equal((struct sockaddr *)rn->rn_key, dst));
1475 			RADIX_NODE_HEAD_UNLOCK(rnh);
1476 			if (error) {
1477 				/* this is only an error if bad on ALL tables */
1478 				continue;
1479 			}
1480 		}
1481 		/*
1482 		 * Do the actual request
1483 		 */
1484 		bzero((caddr_t)&info, sizeof(info));
1485 		info.rti_ifa = ifa;
1486 		info.rti_flags = flags | ifa->ifa_flags;
1487 		info.rti_info[RTAX_DST] = dst;
1488 		/*
1489 		 * doing this for compatibility reasons
1490 		 */
1491 		if (cmd == RTM_ADD)
1492 			info.rti_info[RTAX_GATEWAY] =
1493 			    (struct sockaddr *)&null_sdl;
1494 		else
1495 			info.rti_info[RTAX_GATEWAY] = ifa->ifa_addr;
1496 		info.rti_info[RTAX_NETMASK] = netmask;
1497 		error = rtrequest1_fib(cmd, &info, &rt, fibnum);
1498 		if (error == 0 && rt != NULL) {
1499 			/*
1500 			 * notify any listening routing agents of the change
1501 			 */
1502 			RT_LOCK(rt);
1503 #ifdef RADIX_MPATH
1504 			/*
1505 			 * in case address alias finds the first address
1506 			 * e.g. ifconfig bge0 192.103.54.246/24
1507 			 * e.g. ifconfig bge0 192.103.54.247/24
1508 			 * the address set in the route is 192.103.54.246
1509 			 * so we need to replace it with 192.103.54.247
1510 			 */
1511 			if (memcmp(rt->rt_ifa->ifa_addr,
1512 			    ifa->ifa_addr, ifa->ifa_addr->sa_len)) {
1513 				ifa_free(rt->rt_ifa);
1514 				ifa_ref(ifa);
1515 				rt->rt_ifp = ifa->ifa_ifp;
1516 				rt->rt_ifa = ifa;
1517 			}
1518 #endif
1519 			/*
1520 			 * doing this for compatibility reasons
1521 			 */
1522 			if (cmd == RTM_ADD) {
1523 			    ((struct sockaddr_dl *)rt->rt_gateway)->sdl_type  =
1524 				rt->rt_ifp->if_type;
1525 			    ((struct sockaddr_dl *)rt->rt_gateway)->sdl_index =
1526 				rt->rt_ifp->if_index;
1527 			}
1528 			RT_ADDREF(rt);
1529 			RT_UNLOCK(rt);
1530 			rt_newaddrmsg(cmd, ifa, error, rt);
1531 			RT_LOCK(rt);
1532 			RT_REMREF(rt);
1533 			if (cmd == RTM_DELETE) {
1534 				/*
1535 				 * If we are deleting, and we found an entry,
1536 				 * then it's been removed from the tree..
1537 				 * now throw it away.
1538 				 */
1539 				RTFREE_LOCKED(rt);
1540 			} else {
1541 				if (cmd == RTM_ADD) {
1542 					/*
1543 					 * We just wanted to add it..
1544 					 * we don't actually need a reference.
1545 					 */
1546 					RT_REMREF(rt);
1547 				}
1548 				RT_UNLOCK(rt);
1549 			}
1550 			didwork = 1;
1551 		}
1552 		if (error)
1553 			a_failure = error;
1554 	}
1555 	if (cmd == RTM_DELETE) {
1556 		if (didwork) {
1557 			error = 0;
1558 		} else {
1559 			/* we only give an error if it wasn't in any table */
1560 			error = ((flags & RTF_HOST) ?
1561 			    EHOSTUNREACH : ENETUNREACH);
1562 		}
1563 	} else {
1564 		if (a_failure) {
1565 			/* return an error if any of them failed */
1566 			error = a_failure;
1567 		}
1568 	}
1569 	return (error);
1570 }
1571 
1572 /* special one for inet internal use. may not use. */
1573 int
1574 rtinit_fib(struct ifaddr *ifa, int cmd, int flags)
1575 {
1576 	return (rtinit1(ifa, cmd, flags, -1));
1577 }
1578 
1579 /*
1580  * Set up a routing table entry, normally
1581  * for an interface.
1582  */
1583 int
1584 rtinit(struct ifaddr *ifa, int cmd, int flags)
1585 {
1586 	struct sockaddr *dst;
1587 	int fib = 0;
1588 
1589 	if (flags & RTF_HOST) {
1590 		dst = ifa->ifa_dstaddr;
1591 	} else {
1592 		dst = ifa->ifa_addr;
1593 	}
1594 
1595 	if (dst->sa_family == AF_INET)
1596 		fib = -1;
1597 	return (rtinit1(ifa, cmd, flags, fib));
1598 }
1599