xref: /freebsd/sys/net/route.c (revision 4ac4a7f80b2be339371421b9972535ae5fc473d4)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (c) 1980, 1986, 1991, 1993
5  *	The Regents of the University of California.  All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  * 3. Neither the name of the University nor the names of its contributors
16  *    may be used to endorse or promote products derived from this software
17  *    without specific prior written permission.
18  *
19  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
20  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
21  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
22  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
23  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
24  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
25  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
26  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
27  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
28  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
29  * SUCH DAMAGE.
30  *
31  *	@(#)route.c	8.3.1.1 (Berkeley) 2/23/95
32  * $FreeBSD$
33  */
34 /************************************************************************
35  * Note: In this file a 'fib' is a "forwarding information base"	*
36  * Which is the new name for an in kernel routing (next hop) table.	*
37  ***********************************************************************/
38 
39 #include "opt_inet.h"
40 #include "opt_inet6.h"
41 #include "opt_mrouting.h"
42 #include "opt_mpath.h"
43 #include "opt_route.h"
44 
45 #include <sys/param.h>
46 #include <sys/systm.h>
47 #include <sys/malloc.h>
48 #include <sys/mbuf.h>
49 #include <sys/socket.h>
50 #include <sys/sysctl.h>
51 #include <sys/syslog.h>
52 #include <sys/sysproto.h>
53 #include <sys/proc.h>
54 #include <sys/domain.h>
55 #include <sys/eventhandler.h>
56 #include <sys/kernel.h>
57 #include <sys/lock.h>
58 #include <sys/rmlock.h>
59 
60 #include <net/if.h>
61 #include <net/if_var.h>
62 #include <net/if_dl.h>
63 #include <net/route.h>
64 #include <net/route/route_var.h>
65 #include <net/route/nhop.h>
66 #include <net/route/shared.h>
67 #include <net/vnet.h>
68 
69 #ifdef RADIX_MPATH
70 #include <net/radix_mpath.h>
71 #endif
72 
73 #include <netinet/in.h>
74 #include <netinet/ip_mroute.h>
75 
76 #include <vm/uma.h>
77 
78 #define	RT_MAXFIBS	UINT16_MAX
79 
80 /* Kernel config default option. */
81 #ifdef ROUTETABLES
82 #if ROUTETABLES <= 0
83 #error "ROUTETABLES defined too low"
84 #endif
85 #if ROUTETABLES > RT_MAXFIBS
86 #error "ROUTETABLES defined too big"
87 #endif
88 #define	RT_NUMFIBS	ROUTETABLES
89 #endif /* ROUTETABLES */
90 /* Initialize to default if not otherwise set. */
91 #ifndef	RT_NUMFIBS
92 #define	RT_NUMFIBS	1
93 #endif
94 
95 /* This is read-only.. */
96 u_int rt_numfibs = RT_NUMFIBS;
97 SYSCTL_UINT(_net, OID_AUTO, fibs, CTLFLAG_RDTUN, &rt_numfibs, 0, "");
98 
99 /*
100  * By default add routes to all fibs for new interfaces.
101  * Once this is set to 0 then only allocate routes on interface
102  * changes for the FIB of the caller when adding a new set of addresses
103  * to an interface.  XXX this is a shotgun aproach to a problem that needs
104  * a more fine grained solution.. that will come.
105  * XXX also has the problems getting the FIB from curthread which will not
106  * always work given the fib can be overridden and prefixes can be added
107  * from the network stack context.
108  */
109 VNET_DEFINE(u_int, rt_add_addr_allfibs) = 1;
110 SYSCTL_UINT(_net, OID_AUTO, add_addr_allfibs, CTLFLAG_RWTUN | CTLFLAG_VNET,
111     &VNET_NAME(rt_add_addr_allfibs), 0, "");
112 
113 VNET_PCPUSTAT_DEFINE(struct rtstat, rtstat);
114 
115 VNET_PCPUSTAT_SYSINIT(rtstat);
116 #ifdef VIMAGE
117 VNET_PCPUSTAT_SYSUNINIT(rtstat);
118 #endif
119 
120 VNET_DEFINE(struct rib_head *, rt_tables);
121 #define	V_rt_tables	VNET(rt_tables)
122 
123 VNET_DEFINE(int, rttrash);		/* routes not in table but not freed */
124 #define	V_rttrash	VNET(rttrash)
125 
126 
127 /*
128  * Convert a 'struct radix_node *' to a 'struct rtentry *'.
129  * The operation can be done safely (in this code) because a
130  * 'struct rtentry' starts with two 'struct radix_node''s, the first
131  * one representing leaf nodes in the routing tree, which is
132  * what the code in radix.c passes us as a 'struct radix_node'.
133  *
134  * But because there are a lot of assumptions in this conversion,
135  * do not cast explicitly, but always use the macro below.
136  */
137 #define RNTORT(p)	((struct rtentry *)(p))
138 
139 VNET_DEFINE_STATIC(uma_zone_t, rtzone);		/* Routing table UMA zone. */
140 #define	V_rtzone	VNET(rtzone)
141 
142 EVENTHANDLER_LIST_DEFINE(rt_addrmsg);
143 
144 static int rt_getifa_fib(struct rt_addrinfo *, u_int);
145 static void rt_setmetrics(const struct rt_addrinfo *, struct rtentry *);
146 static int rt_ifdelroute(const struct rtentry *rt, const struct nhop_object *,
147     void *arg);
148 static struct rtentry *rt_unlinkrte(struct rib_head *rnh,
149     struct rt_addrinfo *info, int *perror);
150 static void rt_notifydelete(struct rtentry *rt, struct rt_addrinfo *info);
151 #ifdef RADIX_MPATH
152 static struct radix_node *rt_mpath_unlink(struct rib_head *rnh,
153     struct rt_addrinfo *info, struct rtentry *rto, int *perror);
154 #endif
155 static int rt_exportinfo(struct rtentry *rt, struct rt_addrinfo *info,
156     int flags);
157 
158 static int add_route(struct rib_head *rnh, struct rt_addrinfo *info,
159     struct rtentry **ret_nrt);
160 static int del_route(struct rib_head *rnh, struct rt_addrinfo *info,
161     struct rtentry **ret_nrt);
162 static int change_route(struct rib_head *, struct rt_addrinfo *,
163     struct rtentry **);
164 
165 /*
166  * handler for net.my_fibnum
167  */
168 static int
169 sysctl_my_fibnum(SYSCTL_HANDLER_ARGS)
170 {
171         int fibnum;
172         int error;
173 
174         fibnum = curthread->td_proc->p_fibnum;
175         error = sysctl_handle_int(oidp, &fibnum, 0, req);
176         return (error);
177 }
178 
179 SYSCTL_PROC(_net, OID_AUTO, my_fibnum,
180     CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0,
181     &sysctl_my_fibnum, "I",
182     "default FIB of caller");
183 
184 static __inline struct rib_head **
185 rt_tables_get_rnh_ptr(int table, int fam)
186 {
187 	struct rib_head **rnh;
188 
189 	KASSERT(table >= 0 && table < rt_numfibs,
190 	    ("%s: table out of bounds (0 <= %d < %d)", __func__, table,
191 	     rt_numfibs));
192 	KASSERT(fam >= 0 && fam < (AF_MAX + 1),
193 	    ("%s: fam out of bounds (0 <= %d < %d)", __func__, fam, AF_MAX+1));
194 
195 	/* rnh is [fib=0][af=0]. */
196 	rnh = (struct rib_head **)V_rt_tables;
197 	/* Get the offset to the requested table and fam. */
198 	rnh += table * (AF_MAX+1) + fam;
199 
200 	return (rnh);
201 }
202 
203 struct rib_head *
204 rt_tables_get_rnh(int table, int fam)
205 {
206 
207 	return (*rt_tables_get_rnh_ptr(table, fam));
208 }
209 
210 u_int
211 rt_tables_get_gen(int table, int fam)
212 {
213 	struct rib_head *rnh;
214 
215 	rnh = *rt_tables_get_rnh_ptr(table, fam);
216 	KASSERT(rnh != NULL, ("%s: NULL rib_head pointer table %d fam %d",
217 	    __func__, table, fam));
218 	return (rnh->rnh_gen);
219 }
220 
221 
222 /*
223  * route initialization must occur before ip6_init2(), which happenas at
224  * SI_ORDER_MIDDLE.
225  */
226 static void
227 route_init(void)
228 {
229 
230 	/* whack the tunable ints into  line. */
231 	if (rt_numfibs > RT_MAXFIBS)
232 		rt_numfibs = RT_MAXFIBS;
233 	if (rt_numfibs == 0)
234 		rt_numfibs = 1;
235 	nhops_init();
236 }
237 SYSINIT(route_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_THIRD, route_init, NULL);
238 
239 static int
240 rtentry_zinit(void *mem, int size, int how)
241 {
242 	struct rtentry *rt = mem;
243 
244 	rt->rt_pksent = counter_u64_alloc(how);
245 	if (rt->rt_pksent == NULL)
246 		return (ENOMEM);
247 
248 	RT_LOCK_INIT(rt);
249 
250 	return (0);
251 }
252 
253 static void
254 rtentry_zfini(void *mem, int size)
255 {
256 	struct rtentry *rt = mem;
257 
258 	RT_LOCK_DESTROY(rt);
259 	counter_u64_free(rt->rt_pksent);
260 }
261 
262 static int
263 rtentry_ctor(void *mem, int size, void *arg, int how)
264 {
265 	struct rtentry *rt = mem;
266 
267 	bzero(rt, offsetof(struct rtentry, rt_endzero));
268 	counter_u64_zero(rt->rt_pksent);
269 	rt->rt_chain = NULL;
270 
271 	return (0);
272 }
273 
274 static void
275 rtentry_dtor(void *mem, int size, void *arg)
276 {
277 	struct rtentry *rt = mem;
278 
279 	RT_UNLOCK_COND(rt);
280 }
281 
282 static void
283 vnet_route_init(const void *unused __unused)
284 {
285 	struct domain *dom;
286 	struct rib_head **rnh;
287 	int table;
288 	int fam;
289 
290 	V_rt_tables = malloc(rt_numfibs * (AF_MAX+1) *
291 	    sizeof(struct rib_head *), M_RTABLE, M_WAITOK|M_ZERO);
292 
293 	V_rtzone = uma_zcreate("rtentry", sizeof(struct rtentry),
294 	    rtentry_ctor, rtentry_dtor,
295 	    rtentry_zinit, rtentry_zfini, UMA_ALIGN_PTR, 0);
296 	for (dom = domains; dom; dom = dom->dom_next) {
297 		if (dom->dom_rtattach == NULL)
298 			continue;
299 
300 		for  (table = 0; table < rt_numfibs; table++) {
301 			fam = dom->dom_family;
302 			if (table != 0 && fam != AF_INET6 && fam != AF_INET)
303 				break;
304 
305 			rnh = rt_tables_get_rnh_ptr(table, fam);
306 			if (rnh == NULL)
307 				panic("%s: rnh NULL", __func__);
308 			dom->dom_rtattach((void **)rnh, 0, table);
309 		}
310 	}
311 }
312 VNET_SYSINIT(vnet_route_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_FOURTH,
313     vnet_route_init, 0);
314 
315 #ifdef VIMAGE
316 static void
317 vnet_route_uninit(const void *unused __unused)
318 {
319 	int table;
320 	int fam;
321 	struct domain *dom;
322 	struct rib_head **rnh;
323 
324 	for (dom = domains; dom; dom = dom->dom_next) {
325 		if (dom->dom_rtdetach == NULL)
326 			continue;
327 
328 		for (table = 0; table < rt_numfibs; table++) {
329 			fam = dom->dom_family;
330 
331 			if (table != 0 && fam != AF_INET6 && fam != AF_INET)
332 				break;
333 
334 			rnh = rt_tables_get_rnh_ptr(table, fam);
335 			if (rnh == NULL)
336 				panic("%s: rnh NULL", __func__);
337 			dom->dom_rtdetach((void **)rnh, 0);
338 		}
339 	}
340 
341 	free(V_rt_tables, M_RTABLE);
342 	uma_zdestroy(V_rtzone);
343 }
344 VNET_SYSUNINIT(vnet_route_uninit, SI_SUB_PROTO_DOMAIN, SI_ORDER_FIRST,
345     vnet_route_uninit, 0);
346 #endif
347 
348 struct rib_head *
349 rt_table_init(int offset, int family, u_int fibnum)
350 {
351 	struct rib_head *rh;
352 
353 	rh = malloc(sizeof(struct rib_head), M_RTABLE, M_WAITOK | M_ZERO);
354 
355 	/* TODO: These details should be hidded inside radix.c */
356 	/* Init masks tree */
357 	rn_inithead_internal(&rh->head, rh->rnh_nodes, offset);
358 	rn_inithead_internal(&rh->rmhead.head, rh->rmhead.mask_nodes, 0);
359 	rh->head.rnh_masks = &rh->rmhead;
360 
361 	/* Save metadata associated with this routing table. */
362 	rh->rib_family = family;
363 	rh->rib_fibnum = fibnum;
364 #ifdef VIMAGE
365 	rh->rib_vnet = curvnet;
366 #endif
367 
368 	tmproutes_init(rh);
369 
370 	/* Init locks */
371 	RIB_LOCK_INIT(rh);
372 
373 	nhops_init_rib(rh);
374 
375 	/* Finally, set base callbacks */
376 	rh->rnh_addaddr = rn_addroute;
377 	rh->rnh_deladdr = rn_delete;
378 	rh->rnh_matchaddr = rn_match;
379 	rh->rnh_lookup = rn_lookup;
380 	rh->rnh_walktree = rn_walktree;
381 	rh->rnh_walktree_from = rn_walktree_from;
382 
383 	return (rh);
384 }
385 
386 static int
387 rt_freeentry(struct radix_node *rn, void *arg)
388 {
389 	struct radix_head * const rnh = arg;
390 	struct radix_node *x;
391 
392 	x = (struct radix_node *)rn_delete(rn + 2, NULL, rnh);
393 	if (x != NULL)
394 		R_Free(x);
395 	return (0);
396 }
397 
398 void
399 rt_table_destroy(struct rib_head *rh)
400 {
401 
402 	tmproutes_destroy(rh);
403 
404 	rn_walktree(&rh->rmhead.head, rt_freeentry, &rh->rmhead.head);
405 
406 	nhops_destroy_rib(rh);
407 
408 	/* Assume table is already empty */
409 	RIB_LOCK_DESTROY(rh);
410 	free(rh, M_RTABLE);
411 }
412 
413 
414 #ifndef _SYS_SYSPROTO_H_
415 struct setfib_args {
416 	int     fibnum;
417 };
418 #endif
419 int
420 sys_setfib(struct thread *td, struct setfib_args *uap)
421 {
422 	if (uap->fibnum < 0 || uap->fibnum >= rt_numfibs)
423 		return EINVAL;
424 	td->td_proc->p_fibnum = uap->fibnum;
425 	return (0);
426 }
427 
428 /*
429  * Look up the route that matches the address given
430  * Or, at least try.. Create a cloned route if needed.
431  *
432  * The returned route, if any, is locked.
433  */
434 struct rtentry *
435 rtalloc1(struct sockaddr *dst, int report, u_long ignflags)
436 {
437 
438 	return (rtalloc1_fib(dst, report, ignflags, RT_DEFAULT_FIB));
439 }
440 
441 struct rtentry *
442 rtalloc1_fib(struct sockaddr *dst, int report, u_long ignflags,
443 		    u_int fibnum)
444 {
445 	RIB_RLOCK_TRACKER;
446 	struct rib_head *rh;
447 	struct radix_node *rn;
448 	struct rtentry *newrt;
449 	struct rt_addrinfo info;
450 	int err = 0, msgtype = RTM_MISS;
451 
452 	KASSERT((fibnum < rt_numfibs), ("rtalloc1_fib: bad fibnum"));
453 	rh = rt_tables_get_rnh(fibnum, dst->sa_family);
454 	newrt = NULL;
455 	if (rh == NULL)
456 		goto miss;
457 
458 	/*
459 	 * Look up the address in the table for that Address Family
460 	 */
461 	if ((ignflags & RTF_RNH_LOCKED) == 0)
462 		RIB_RLOCK(rh);
463 #ifdef INVARIANTS
464 	else
465 		RIB_LOCK_ASSERT(rh);
466 #endif
467 	rn = rh->rnh_matchaddr(dst, &rh->head);
468 	if (rn && ((rn->rn_flags & RNF_ROOT) == 0)) {
469 		newrt = RNTORT(rn);
470 		RT_LOCK(newrt);
471 		RT_ADDREF(newrt);
472 		if ((ignflags & RTF_RNH_LOCKED) == 0)
473 			RIB_RUNLOCK(rh);
474 		return (newrt);
475 
476 	} else if ((ignflags & RTF_RNH_LOCKED) == 0)
477 		RIB_RUNLOCK(rh);
478 	/*
479 	 * Either we hit the root or could not find any match,
480 	 * which basically means: "cannot get there from here".
481 	 */
482 miss:
483 	RTSTAT_INC(rts_unreach);
484 
485 	if (report) {
486 		/*
487 		 * If required, report the failure to the supervising
488 		 * Authorities.
489 		 * For a delete, this is not an error. (report == 0)
490 		 */
491 		bzero(&info, sizeof(info));
492 		info.rti_info[RTAX_DST] = dst;
493 		rt_missmsg_fib(msgtype, &info, 0, err, fibnum);
494 	}
495 	return (newrt);
496 }
497 
498 /*
499  * Remove a reference count from an rtentry.
500  * If the count gets low enough, take it out of the routing table
501  */
502 void
503 rtfree(struct rtentry *rt)
504 {
505 	struct rib_head *rnh;
506 
507 	KASSERT(rt != NULL,("%s: NULL rt", __func__));
508 	rnh = rt_tables_get_rnh(rt->rt_fibnum, rt_key(rt)->sa_family);
509 	KASSERT(rnh != NULL,("%s: NULL rnh", __func__));
510 
511 	RT_LOCK_ASSERT(rt);
512 
513 	/*
514 	 * The callers should use RTFREE_LOCKED() or RTFREE(), so
515 	 * we should come here exactly with the last reference.
516 	 */
517 	RT_REMREF(rt);
518 	if (rt->rt_refcnt > 0) {
519 		log(LOG_DEBUG, "%s: %p has %d refs\n", __func__, rt, rt->rt_refcnt);
520 		goto done;
521 	}
522 
523 	/*
524 	 * On last reference give the "close method" a chance
525 	 * to cleanup private state.  This also permits (for
526 	 * IPv4 and IPv6) a chance to decide if the routing table
527 	 * entry should be purged immediately or at a later time.
528 	 * When an immediate purge is to happen the close routine
529 	 * typically calls rtexpunge which clears the RTF_UP flag
530 	 * on the entry so that the code below reclaims the storage.
531 	 */
532 	if (rt->rt_refcnt == 0 && rnh->rnh_close)
533 		rnh->rnh_close((struct radix_node *)rt, &rnh->head);
534 
535 	/*
536 	 * If we are no longer "up" (and ref == 0)
537 	 * then we can free the resources associated
538 	 * with the route.
539 	 */
540 	if ((rt->rt_flags & RTF_UP) == 0) {
541 		if (rt->rt_nodes->rn_flags & (RNF_ACTIVE | RNF_ROOT))
542 			panic("rtfree 2");
543 		/*
544 		 * the rtentry must have been removed from the routing table
545 		 * so it is represented in rttrash.. remove that now.
546 		 */
547 		V_rttrash--;
548 #ifdef	DIAGNOSTIC
549 		if (rt->rt_refcnt < 0) {
550 			printf("rtfree: %p not freed (neg refs)\n", rt);
551 			goto done;
552 		}
553 #endif
554 		/*
555 		 * The key is separatly alloc'd so free it (see rt_setgate()).
556 		 * This also frees the gateway, as they are always malloc'd
557 		 * together.
558 		 */
559 		R_Free(rt_key(rt));
560 
561 		/* Unreference nexthop */
562 		nhop_free(rt->rt_nhop);
563 
564 		/*
565 		 * and the rtentry itself of course
566 		 */
567 		uma_zfree(V_rtzone, rt);
568 		return;
569 	}
570 done:
571 	RT_UNLOCK(rt);
572 }
573 
574 /*
575  * Temporary RTFREE() function wrapper.
576  *  Intended to use in control plane code to
577  *  avoid exposing internal layout of 'struct rtentry'.
578  */
579 void
580 rtfree_func(struct rtentry *rt)
581 {
582 
583 	RTFREE(rt);
584 }
585 
586 /*
587  * Adds a temporal redirect entry to the routing table.
588  * @fibnum: fib number
589  * @dst: destination to install redirect to
590  * @gateway: gateway to go via
591  * @author: sockaddr of originating router, can be NULL
592  * @ifp: interface to use for the redirected route
593  * @flags: set of flags to add. Allowed: RTF_GATEWAY
594  * @lifetime_sec: time in seconds to expire this redirect.
595  *
596  * Retuns 0 on success, errno otherwise.
597  */
598 int
599 rib_add_redirect(u_int fibnum, struct sockaddr *dst, struct sockaddr *gateway,
600     struct sockaddr *author, struct ifnet *ifp, int flags, int lifetime_sec)
601 {
602 	struct rtentry *rt;
603 	int error;
604 	struct rt_addrinfo info;
605 	struct rt_metrics rti_rmx;
606 	struct ifaddr *ifa;
607 
608 	NET_EPOCH_ASSERT();
609 
610 	if (rt_tables_get_rnh(fibnum, dst->sa_family) == NULL)
611 		return (EAFNOSUPPORT);
612 
613 	/* Verify the allowed flag mask. */
614 	KASSERT(((flags & ~(RTF_GATEWAY)) == 0),
615 	    ("invalid redirect flags: %x", flags));
616 
617 	/* Get the best ifa for the given interface and gateway. */
618 	if ((ifa = ifaof_ifpforaddr(gateway, ifp)) == NULL)
619 		return (ENETUNREACH);
620 	ifa_ref(ifa);
621 
622 	bzero(&info, sizeof(info));
623 	info.rti_info[RTAX_DST] = dst;
624 	info.rti_info[RTAX_GATEWAY] = gateway;
625 	info.rti_ifa = ifa;
626 	info.rti_ifp = ifp;
627 	info.rti_flags = flags | RTF_HOST | RTF_DYNAMIC;
628 
629 	/* Setup route metrics to define expire time. */
630 	bzero(&rti_rmx, sizeof(rti_rmx));
631 	/* Set expire time as absolute. */
632 	rti_rmx.rmx_expire = lifetime_sec + time_second;
633 	info.rti_mflags |= RTV_EXPIRE;
634 	info.rti_rmx = &rti_rmx;
635 
636 	error = rtrequest1_fib(RTM_ADD, &info, &rt, fibnum);
637 	ifa_free(ifa);
638 
639 	if (error != 0) {
640 		/* TODO: add per-fib redirect stats. */
641 		return (error);
642 	}
643 
644 	RT_LOCK(rt);
645 	flags = rt->rt_flags;
646 	RTFREE_LOCKED(rt);
647 
648 	RTSTAT_INC(rts_dynamic);
649 
650 	/* Send notification of a route addition to userland. */
651 	bzero(&info, sizeof(info));
652 	info.rti_info[RTAX_DST] = dst;
653 	info.rti_info[RTAX_GATEWAY] = gateway;
654 	info.rti_info[RTAX_AUTHOR] = author;
655 	rt_missmsg_fib(RTM_REDIRECT, &info, flags, error, fibnum);
656 
657 	return (0);
658 }
659 
660 /*
661  * Routing table ioctl interface.
662  */
663 int
664 rtioctl_fib(u_long req, caddr_t data, u_int fibnum)
665 {
666 
667 	/*
668 	 * If more ioctl commands are added here, make sure the proper
669 	 * super-user checks are being performed because it is possible for
670 	 * prison-root to make it this far if raw sockets have been enabled
671 	 * in jails.
672 	 */
673 #ifdef INET
674 	/* Multicast goop, grrr... */
675 	return mrt_ioctl ? mrt_ioctl(req, data, fibnum) : EOPNOTSUPP;
676 #else /* INET */
677 	return ENXIO;
678 #endif /* INET */
679 }
680 
681 struct ifaddr *
682 ifa_ifwithroute(int flags, const struct sockaddr *dst, struct sockaddr *gateway,
683 				u_int fibnum)
684 {
685 	struct ifaddr *ifa;
686 	int not_found = 0;
687 
688 	NET_EPOCH_ASSERT();
689 	if ((flags & RTF_GATEWAY) == 0) {
690 		/*
691 		 * If we are adding a route to an interface,
692 		 * and the interface is a pt to pt link
693 		 * we should search for the destination
694 		 * as our clue to the interface.  Otherwise
695 		 * we can use the local address.
696 		 */
697 		ifa = NULL;
698 		if (flags & RTF_HOST)
699 			ifa = ifa_ifwithdstaddr(dst, fibnum);
700 		if (ifa == NULL)
701 			ifa = ifa_ifwithaddr(gateway);
702 	} else {
703 		/*
704 		 * If we are adding a route to a remote net
705 		 * or host, the gateway may still be on the
706 		 * other end of a pt to pt link.
707 		 */
708 		ifa = ifa_ifwithdstaddr(gateway, fibnum);
709 	}
710 	if (ifa == NULL)
711 		ifa = ifa_ifwithnet(gateway, 0, fibnum);
712 	if (ifa == NULL) {
713 		struct rtentry *rt;
714 
715 		rt = rtalloc1_fib(gateway, 0, flags, fibnum);
716 		if (rt == NULL)
717 			goto out;
718 		/*
719 		 * dismiss a gateway that is reachable only
720 		 * through the default router
721 		 */
722 		switch (gateway->sa_family) {
723 		case AF_INET:
724 			if (satosin(rt_key(rt))->sin_addr.s_addr == INADDR_ANY)
725 				not_found = 1;
726 			break;
727 		case AF_INET6:
728 			if (IN6_IS_ADDR_UNSPECIFIED(&satosin6(rt_key(rt))->sin6_addr))
729 				not_found = 1;
730 			break;
731 		default:
732 			break;
733 		}
734 		if (!not_found && rt->rt_nhop->nh_ifa != NULL) {
735 			ifa = rt->rt_nhop->nh_ifa;
736 		}
737 		RT_REMREF(rt);
738 		RT_UNLOCK(rt);
739 		if (not_found || ifa == NULL)
740 			goto out;
741 	}
742 	if (ifa->ifa_addr->sa_family != dst->sa_family) {
743 		struct ifaddr *oifa = ifa;
744 		ifa = ifaof_ifpforaddr(dst, ifa->ifa_ifp);
745 		if (ifa == NULL)
746 			ifa = oifa;
747 	}
748  out:
749 	return (ifa);
750 }
751 
752 /*
753  * Do appropriate manipulations of a routing tree given
754  * all the bits of info needed
755  */
756 int
757 rtrequest_fib(int req,
758 	struct sockaddr *dst,
759 	struct sockaddr *gateway,
760 	struct sockaddr *netmask,
761 	int flags,
762 	struct rtentry **ret_nrt,
763 	u_int fibnum)
764 {
765 	struct rt_addrinfo info;
766 
767 	if (dst->sa_len == 0)
768 		return(EINVAL);
769 
770 	bzero((caddr_t)&info, sizeof(info));
771 	info.rti_flags = flags;
772 	info.rti_info[RTAX_DST] = dst;
773 	info.rti_info[RTAX_GATEWAY] = gateway;
774 	info.rti_info[RTAX_NETMASK] = netmask;
775 	return rtrequest1_fib(req, &info, ret_nrt, fibnum);
776 }
777 
778 
779 /*
780  * Copy most of @rt data into @info.
781  *
782  * If @flags contains NHR_COPY, copies dst,netmask and gw to the
783  * pointers specified by @info structure. Assume such pointers
784  * are zeroed sockaddr-like structures with sa_len field initialized
785  * to reflect size of the provided buffer. if no NHR_COPY is specified,
786  * point dst,netmask and gw @info fields to appropriate @rt values.
787  *
788  * if @flags contains NHR_REF, do refcouting on rt_ifp and rt_ifa.
789  *
790  * Returns 0 on success.
791  */
792 int
793 rt_exportinfo(struct rtentry *rt, struct rt_addrinfo *info, int flags)
794 {
795 	struct rt_metrics *rmx;
796 	struct sockaddr *src, *dst;
797 	struct nhop_object *nh;
798 	int sa_len;
799 
800 	if (flags & NHR_COPY) {
801 		/* Copy destination if dst is non-zero */
802 		src = rt_key(rt);
803 		dst = info->rti_info[RTAX_DST];
804 		sa_len = src->sa_len;
805 		if (dst != NULL) {
806 			if (src->sa_len > dst->sa_len)
807 				return (ENOMEM);
808 			memcpy(dst, src, src->sa_len);
809 			info->rti_addrs |= RTA_DST;
810 		}
811 
812 		/* Copy mask if set && dst is non-zero */
813 		src = rt_mask(rt);
814 		dst = info->rti_info[RTAX_NETMASK];
815 		if (src != NULL && dst != NULL) {
816 
817 			/*
818 			 * Radix stores different value in sa_len,
819 			 * assume rt_mask() to have the same length
820 			 * as rt_key()
821 			 */
822 			if (sa_len > dst->sa_len)
823 				return (ENOMEM);
824 			memcpy(dst, src, src->sa_len);
825 			info->rti_addrs |= RTA_NETMASK;
826 		}
827 
828 		/* Copy gateway is set && dst is non-zero */
829 		src = &rt->rt_nhop->gw_sa;
830 		dst = info->rti_info[RTAX_GATEWAY];
831 		if ((rt->rt_flags & RTF_GATEWAY) && src != NULL && dst != NULL){
832 			if (src->sa_len > dst->sa_len)
833 				return (ENOMEM);
834 			memcpy(dst, src, src->sa_len);
835 			info->rti_addrs |= RTA_GATEWAY;
836 		}
837 	} else {
838 		info->rti_info[RTAX_DST] = rt_key(rt);
839 		info->rti_addrs |= RTA_DST;
840 		if (rt_mask(rt) != NULL) {
841 			info->rti_info[RTAX_NETMASK] = rt_mask(rt);
842 			info->rti_addrs |= RTA_NETMASK;
843 		}
844 		if (rt->rt_flags & RTF_GATEWAY) {
845 			info->rti_info[RTAX_GATEWAY] = &rt->rt_nhop->gw_sa;
846 			info->rti_addrs |= RTA_GATEWAY;
847 		}
848 	}
849 
850 	nh = rt->rt_nhop;
851 	rmx = info->rti_rmx;
852 	if (rmx != NULL) {
853 		info->rti_mflags |= RTV_MTU;
854 		rmx->rmx_mtu = nh->nh_mtu;
855 	}
856 
857 	info->rti_flags = rt->rt_flags | nhop_get_rtflags(nh);
858 	info->rti_ifp = nh->nh_ifp;
859 	info->rti_ifa = nh->nh_ifa;
860 	if (flags & NHR_REF) {
861 		if_ref(info->rti_ifp);
862 		ifa_ref(info->rti_ifa);
863 	}
864 
865 	return (0);
866 }
867 
868 /*
869  * Lookups up route entry for @dst in RIB database for fib @fibnum.
870  * Exports entry data to @info using rt_exportinfo().
871  *
872  * If @flags contains NHR_REF, refcouting is performed on rt_ifp and rt_ifa.
873  * All references can be released later by calling rib_free_info().
874  *
875  * Returns 0 on success.
876  * Returns ENOENT for lookup failure, ENOMEM for export failure.
877  */
878 int
879 rib_lookup_info(uint32_t fibnum, const struct sockaddr *dst, uint32_t flags,
880     uint32_t flowid, struct rt_addrinfo *info)
881 {
882 	RIB_RLOCK_TRACKER;
883 	struct rib_head *rh;
884 	struct radix_node *rn;
885 	struct rtentry *rt;
886 	int error;
887 
888 	KASSERT((fibnum < rt_numfibs), ("rib_lookup_rte: bad fibnum"));
889 	rh = rt_tables_get_rnh(fibnum, dst->sa_family);
890 	if (rh == NULL)
891 		return (ENOENT);
892 
893 	RIB_RLOCK(rh);
894 	rn = rh->rnh_matchaddr(__DECONST(void *, dst), &rh->head);
895 	if (rn != NULL && ((rn->rn_flags & RNF_ROOT) == 0)) {
896 		rt = RNTORT(rn);
897 		/* Ensure route & ifp is UP */
898 		if (RT_LINK_IS_UP(rt->rt_nhop->nh_ifp)) {
899 			flags = (flags & NHR_REF) | NHR_COPY;
900 			error = rt_exportinfo(rt, info, flags);
901 			RIB_RUNLOCK(rh);
902 
903 			return (error);
904 		}
905 	}
906 	RIB_RUNLOCK(rh);
907 
908 	return (ENOENT);
909 }
910 
911 /*
912  * Releases all references acquired by rib_lookup_info() when
913  * called with NHR_REF flags.
914  */
915 void
916 rib_free_info(struct rt_addrinfo *info)
917 {
918 
919 	ifa_free(info->rti_ifa);
920 	if_rele(info->rti_ifp);
921 }
922 
923 /*
924  * Iterates over all existing fibs in system calling
925  *  @setwa_f function prior to traversing each fib.
926  *  Calls @wa_f function for each element in current fib.
927  * If af is not AF_UNSPEC, iterates over fibs in particular
928  * address family.
929  */
930 void
931 rt_foreach_fib_walk(int af, rt_setwarg_t *setwa_f, rt_walktree_f_t *wa_f,
932     void *arg)
933 {
934 	struct rib_head *rnh;
935 	uint32_t fibnum;
936 	int i;
937 
938 	for (fibnum = 0; fibnum < rt_numfibs; fibnum++) {
939 		/* Do we want some specific family? */
940 		if (af != AF_UNSPEC) {
941 			rnh = rt_tables_get_rnh(fibnum, af);
942 			if (rnh == NULL)
943 				continue;
944 			if (setwa_f != NULL)
945 				setwa_f(rnh, fibnum, af, arg);
946 
947 			RIB_WLOCK(rnh);
948 			rnh->rnh_walktree(&rnh->head, (walktree_f_t *)wa_f,arg);
949 			RIB_WUNLOCK(rnh);
950 			continue;
951 		}
952 
953 		for (i = 1; i <= AF_MAX; i++) {
954 			rnh = rt_tables_get_rnh(fibnum, i);
955 			if (rnh == NULL)
956 				continue;
957 			if (setwa_f != NULL)
958 				setwa_f(rnh, fibnum, i, arg);
959 
960 			RIB_WLOCK(rnh);
961 			rnh->rnh_walktree(&rnh->head, (walktree_f_t *)wa_f,arg);
962 			RIB_WUNLOCK(rnh);
963 		}
964 	}
965 }
966 
967 struct rt_delinfo
968 {
969 	struct rt_addrinfo info;
970 	struct rib_head *rnh;
971 	struct rtentry *head;
972 };
973 
974 /*
975  * Conditionally unlinks @rn from radix tree based
976  * on info data passed in @arg.
977  */
978 static int
979 rt_checkdelroute(struct radix_node *rn, void *arg)
980 {
981 	struct rt_delinfo *di;
982 	struct rt_addrinfo *info;
983 	struct rtentry *rt;
984 	int error;
985 
986 	di = (struct rt_delinfo *)arg;
987 	rt = (struct rtentry *)rn;
988 	info = &di->info;
989 	error = 0;
990 
991 	info->rti_info[RTAX_DST] = rt_key(rt);
992 	info->rti_info[RTAX_NETMASK] = rt_mask(rt);
993 	info->rti_info[RTAX_GATEWAY] = &rt->rt_nhop->gw_sa;
994 
995 	rt = rt_unlinkrte(di->rnh, info, &error);
996 	if (rt == NULL) {
997 		/* Either not allowed or not matched. Skip entry */
998 		return (0);
999 	}
1000 
1001 	/* Entry was unlinked. Add to the list and return */
1002 	rt->rt_chain = di->head;
1003 	di->head = rt;
1004 
1005 	return (0);
1006 }
1007 
1008 /*
1009  * Iterates over a routing table specified by @fibnum and @family and
1010  *  deletes elements marked by @filter_f.
1011  * @fibnum: rtable id
1012  * @family: AF_ address family
1013  * @filter_f: function returning non-zero value for items to delete
1014  * @arg: data to pass to the @filter_f function
1015  * @report: true if rtsock notification is needed.
1016  */
1017 void
1018 rib_walk_del(u_int fibnum, int family, rt_filter_f_t *filter_f, void *arg, bool report)
1019 {
1020 	struct rib_head *rnh;
1021 	struct rt_delinfo di;
1022 	struct rtentry *rt;
1023 
1024 	rnh = rt_tables_get_rnh(fibnum, family);
1025 	if (rnh == NULL)
1026 		return;
1027 
1028 	bzero(&di, sizeof(di));
1029 	di.info.rti_filter = filter_f;
1030 	di.info.rti_filterdata = arg;
1031 	di.rnh = rnh;
1032 
1033 	RIB_WLOCK(rnh);
1034 	rnh->rnh_walktree(&rnh->head, rt_checkdelroute, &di);
1035 	RIB_WUNLOCK(rnh);
1036 
1037 	if (di.head == NULL)
1038 		return;
1039 
1040 	/* We might have something to reclaim. */
1041 	while (di.head != NULL) {
1042 		rt = di.head;
1043 		di.head = rt->rt_chain;
1044 		rt->rt_chain = NULL;
1045 
1046 		/* TODO std rt -> rt_addrinfo export */
1047 		di.info.rti_info[RTAX_DST] = rt_key(rt);
1048 		di.info.rti_info[RTAX_NETMASK] = rt_mask(rt);
1049 
1050 		rt_notifydelete(rt, &di.info);
1051 
1052 		if (report)
1053 			rt_routemsg(RTM_DELETE, rt, rt->rt_nhop->nh_ifp, 0,
1054 			    fibnum);
1055 		RTFREE_LOCKED(rt);
1056 	}
1057 }
1058 
1059 /*
1060  * Iterates over all existing fibs in system and deletes each element
1061  *  for which @filter_f function returns non-zero value.
1062  * If @family is not AF_UNSPEC, iterates over fibs in particular
1063  * address family.
1064  */
1065 void
1066 rt_foreach_fib_walk_del(int family, rt_filter_f_t *filter_f, void *arg)
1067 {
1068 	u_int fibnum;
1069 	int i, start, end;
1070 
1071 	for (fibnum = 0; fibnum < rt_numfibs; fibnum++) {
1072 		/* Do we want some specific family? */
1073 		if (family != AF_UNSPEC) {
1074 			start = family;
1075 			end = family;
1076 		} else {
1077 			start = 1;
1078 			end = AF_MAX;
1079 		}
1080 
1081 		for (i = start; i <= end; i++) {
1082 			if (rt_tables_get_rnh(fibnum, i) == NULL)
1083 				continue;
1084 
1085 			rib_walk_del(fibnum, i, filter_f, arg, 0);
1086 		}
1087 	}
1088 }
1089 
1090 /*
1091  * Delete Routes for a Network Interface
1092  *
1093  * Called for each routing entry via the rnh->rnh_walktree() call above
1094  * to delete all route entries referencing a detaching network interface.
1095  *
1096  * Arguments:
1097  *	rt	pointer to rtentry
1098  *	nh	pointer to nhop
1099  *	arg	argument passed to rnh->rnh_walktree() - detaching interface
1100  *
1101  * Returns:
1102  *	0	successful
1103  *	errno	failed - reason indicated
1104  */
1105 static int
1106 rt_ifdelroute(const struct rtentry *rt, const struct nhop_object *nh, void *arg)
1107 {
1108 	struct ifnet	*ifp = arg;
1109 
1110 	if (nh->nh_ifp != ifp)
1111 		return (0);
1112 
1113 	/*
1114 	 * Protect (sorta) against walktree recursion problems
1115 	 * with cloned routes
1116 	 */
1117 	if ((rt->rt_flags & RTF_UP) == 0)
1118 		return (0);
1119 
1120 	return (1);
1121 }
1122 
1123 /*
1124  * Delete all remaining routes using this interface
1125  * Unfortuneatly the only way to do this is to slog through
1126  * the entire routing table looking for routes which point
1127  * to this interface...oh well...
1128  */
1129 void
1130 rt_flushifroutes_af(struct ifnet *ifp, int af)
1131 {
1132 	KASSERT((af >= 1 && af <= AF_MAX), ("%s: af %d not >= 1 and <= %d",
1133 	    __func__, af, AF_MAX));
1134 
1135 	rt_foreach_fib_walk_del(af, rt_ifdelroute, ifp);
1136 }
1137 
1138 void
1139 rt_flushifroutes(struct ifnet *ifp)
1140 {
1141 
1142 	rt_foreach_fib_walk_del(AF_UNSPEC, rt_ifdelroute, ifp);
1143 }
1144 
1145 /*
1146  * Conditionally unlinks rtentry matching data inside @info from @rnh.
1147  * Returns unlinked, locked and referenced @rtentry on success,
1148  * Returns NULL and sets @perror to:
1149  * ESRCH - if prefix was not found,
1150  * EADDRINUSE - if trying to delete PINNED route without appropriate flag.
1151  * ENOENT - if supplied filter function returned 0 (not matched).
1152  */
1153 static struct rtentry *
1154 rt_unlinkrte(struct rib_head *rnh, struct rt_addrinfo *info, int *perror)
1155 {
1156 	struct sockaddr *dst, *netmask;
1157 	struct rtentry *rt;
1158 	struct radix_node *rn;
1159 
1160 	dst = info->rti_info[RTAX_DST];
1161 	netmask = info->rti_info[RTAX_NETMASK];
1162 
1163 	rt = (struct rtentry *)rnh->rnh_lookup(dst, netmask, &rnh->head);
1164 	if (rt == NULL) {
1165 		*perror = ESRCH;
1166 		return (NULL);
1167 	}
1168 
1169 	if ((info->rti_flags & RTF_PINNED) == 0) {
1170 		/* Check if target route can be deleted */
1171 		if (rt->rt_flags & RTF_PINNED) {
1172 			*perror = EADDRINUSE;
1173 			return (NULL);
1174 		}
1175 	}
1176 
1177 	if (info->rti_filter != NULL) {
1178 		if (info->rti_filter(rt, rt->rt_nhop, info->rti_filterdata)==0){
1179 			/* Not matched */
1180 			*perror = ENOENT;
1181 			return (NULL);
1182 		}
1183 
1184 		/*
1185 		 * Filter function requested rte deletion.
1186 		 * Ease the caller work by filling in remaining info
1187 		 * from that particular entry.
1188 		 */
1189 		info->rti_info[RTAX_GATEWAY] = &rt->rt_nhop->gw_sa;
1190 	}
1191 
1192 	/*
1193 	 * Remove the item from the tree and return it.
1194 	 * Complain if it is not there and do no more processing.
1195 	 */
1196 	*perror = ESRCH;
1197 #ifdef RADIX_MPATH
1198 	if (rt_mpath_capable(rnh))
1199 		rn = rt_mpath_unlink(rnh, info, rt, perror);
1200 	else
1201 #endif
1202 	rn = rnh->rnh_deladdr(dst, netmask, &rnh->head);
1203 	if (rn == NULL)
1204 		return (NULL);
1205 
1206 	if (rn->rn_flags & (RNF_ACTIVE | RNF_ROOT))
1207 		panic ("rtrequest delete");
1208 
1209 	rt = RNTORT(rn);
1210 	RT_LOCK(rt);
1211 	RT_ADDREF(rt);
1212 	rt->rt_flags &= ~RTF_UP;
1213 
1214 	*perror = 0;
1215 
1216 	return (rt);
1217 }
1218 
1219 static void
1220 rt_notifydelete(struct rtentry *rt, struct rt_addrinfo *info)
1221 {
1222 	struct ifaddr *ifa;
1223 
1224 	/*
1225 	 * give the protocol a chance to keep things in sync.
1226 	 */
1227 	ifa = rt->rt_nhop->nh_ifa;
1228 	if (ifa != NULL && ifa->ifa_rtrequest != NULL)
1229 		ifa->ifa_rtrequest(RTM_DELETE, rt, rt->rt_nhop, info);
1230 
1231 	/*
1232 	 * One more rtentry floating around that is not
1233 	 * linked to the routing table. rttrash will be decremented
1234 	 * when RTFREE(rt) is eventually called.
1235 	 */
1236 	V_rttrash++;
1237 }
1238 
1239 
1240 /*
1241  * These (questionable) definitions of apparent local variables apply
1242  * to the next two functions.  XXXXXX!!!
1243  */
1244 #define	dst	info->rti_info[RTAX_DST]
1245 #define	gateway	info->rti_info[RTAX_GATEWAY]
1246 #define	netmask	info->rti_info[RTAX_NETMASK]
1247 #define	ifaaddr	info->rti_info[RTAX_IFA]
1248 #define	ifpaddr	info->rti_info[RTAX_IFP]
1249 #define	flags	info->rti_flags
1250 
1251 /*
1252  * Look up rt_addrinfo for a specific fib.  Note that if rti_ifa is defined,
1253  * it will be referenced so the caller must free it.
1254  *
1255  * Assume basic consistency checks are executed by callers:
1256  * RTAX_DST exists, if RTF_GATEWAY is set, RTAX_GATEWAY exists as well.
1257  */
1258 int
1259 rt_getifa_fib(struct rt_addrinfo *info, u_int fibnum)
1260 {
1261 	struct epoch_tracker et;
1262 	int needref, error;
1263 
1264 	/*
1265 	 * ifp may be specified by sockaddr_dl
1266 	 * when protocol address is ambiguous.
1267 	 */
1268 	error = 0;
1269 	needref = (info->rti_ifa == NULL);
1270 	NET_EPOCH_ENTER(et);
1271 
1272 	/* If we have interface specified by the ifindex in the address, use it */
1273 	if (info->rti_ifp == NULL && ifpaddr != NULL &&
1274 	    ifpaddr->sa_family == AF_LINK) {
1275 	    const struct sockaddr_dl *sdl = (const struct sockaddr_dl *)ifpaddr;
1276 	    if (sdl->sdl_index != 0)
1277 		    info->rti_ifp = ifnet_byindex(sdl->sdl_index);
1278 	}
1279 	/*
1280 	 * If we have source address specified, try to find it
1281 	 * TODO: avoid enumerating all ifas on all interfaces.
1282 	 */
1283 	if (info->rti_ifa == NULL && ifaaddr != NULL)
1284 		info->rti_ifa = ifa_ifwithaddr(ifaaddr);
1285 	if (info->rti_ifa == NULL) {
1286 		struct sockaddr *sa;
1287 
1288 		/*
1289 		 * Most common use case for the userland-supplied routes.
1290 		 *
1291 		 * Choose sockaddr to select ifa.
1292 		 * -- if ifp is set --
1293 		 * Order of preference:
1294 		 * 1) IFA address
1295 		 * 2) gateway address
1296 		 *   Note: for interface routes link-level gateway address
1297 		 *     is specified to indicate the interface index without
1298 		 *     specifying RTF_GATEWAY. In this case, ignore gateway
1299 		 *   Note: gateway AF may be different from dst AF. In this case,
1300 		 *   ignore gateway
1301 		 * 3) final destination.
1302 		 * 4) if all of these fails, try to get at least link-level ifa.
1303 		 * -- else --
1304 		 * try to lookup gateway or dst in the routing table to get ifa
1305 		 */
1306 		if (info->rti_info[RTAX_IFA] != NULL)
1307 			sa = info->rti_info[RTAX_IFA];
1308 		else if ((info->rti_flags & RTF_GATEWAY) != 0 &&
1309 		    gateway->sa_family == dst->sa_family)
1310 			sa = gateway;
1311 		else
1312 			sa = dst;
1313 		if (info->rti_ifp != NULL) {
1314 			info->rti_ifa = ifaof_ifpforaddr(sa, info->rti_ifp);
1315 			/* Case 4 */
1316 			if (info->rti_ifa == NULL && gateway != NULL)
1317 				info->rti_ifa = ifaof_ifpforaddr(gateway, info->rti_ifp);
1318 		} else if (dst != NULL && gateway != NULL)
1319 			info->rti_ifa = ifa_ifwithroute(flags, dst, gateway,
1320 							fibnum);
1321 		else if (sa != NULL)
1322 			info->rti_ifa = ifa_ifwithroute(flags, sa, sa,
1323 							fibnum);
1324 	}
1325 	if (needref && info->rti_ifa != NULL) {
1326 		if (info->rti_ifp == NULL)
1327 			info->rti_ifp = info->rti_ifa->ifa_ifp;
1328 		ifa_ref(info->rti_ifa);
1329 	} else
1330 		error = ENETUNREACH;
1331 	NET_EPOCH_EXIT(et);
1332 	return (error);
1333 }
1334 
1335 void
1336 rt_updatemtu(struct ifnet *ifp)
1337 {
1338 	struct rib_head *rnh;
1339 	int mtu;
1340 	int i, j;
1341 
1342 	/*
1343 	 * Try to update rt_mtu for all routes using this interface
1344 	 * Unfortunately the only way to do this is to traverse all
1345 	 * routing tables in all fibs/domains.
1346 	 */
1347 	for (i = 1; i <= AF_MAX; i++) {
1348 		mtu = if_getmtu_family(ifp, i);
1349 		for (j = 0; j < rt_numfibs; j++) {
1350 			rnh = rt_tables_get_rnh(j, i);
1351 			if (rnh == NULL)
1352 				continue;
1353 			nhops_update_ifmtu(rnh, ifp, mtu);
1354 		}
1355 	}
1356 }
1357 
1358 
1359 #if 0
1360 int p_sockaddr(char *buf, int buflen, struct sockaddr *s);
1361 int rt_print(char *buf, int buflen, struct rtentry *rt);
1362 
1363 int
1364 p_sockaddr(char *buf, int buflen, struct sockaddr *s)
1365 {
1366 	void *paddr = NULL;
1367 
1368 	switch (s->sa_family) {
1369 	case AF_INET:
1370 		paddr = &((struct sockaddr_in *)s)->sin_addr;
1371 		break;
1372 	case AF_INET6:
1373 		paddr = &((struct sockaddr_in6 *)s)->sin6_addr;
1374 		break;
1375 	}
1376 
1377 	if (paddr == NULL)
1378 		return (0);
1379 
1380 	if (inet_ntop(s->sa_family, paddr, buf, buflen) == NULL)
1381 		return (0);
1382 
1383 	return (strlen(buf));
1384 }
1385 
1386 int
1387 rt_print(char *buf, int buflen, struct rtentry *rt)
1388 {
1389 	struct sockaddr *addr, *mask;
1390 	int i = 0;
1391 
1392 	addr = rt_key(rt);
1393 	mask = rt_mask(rt);
1394 
1395 	i = p_sockaddr(buf, buflen, addr);
1396 	if (!(rt->rt_flags & RTF_HOST)) {
1397 		buf[i++] = '/';
1398 		i += p_sockaddr(buf + i, buflen - i, mask);
1399 	}
1400 
1401 	if (rt->rt_flags & RTF_GATEWAY) {
1402 		buf[i++] = '>';
1403 		i += p_sockaddr(buf + i, buflen - i, &rt->rt_nhop->gw_sa);
1404 	}
1405 
1406 	return (i);
1407 }
1408 #endif
1409 
1410 #ifdef RADIX_MPATH
1411 /*
1412  * Deletes key for single-path routes, unlinks rtentry with
1413  * gateway specified in @info from multi-path routes.
1414  *
1415  * Returnes unlinked entry. In case of failure, returns NULL
1416  * and sets @perror to ESRCH.
1417  */
1418 static struct radix_node *
1419 rt_mpath_unlink(struct rib_head *rnh, struct rt_addrinfo *info,
1420     struct rtentry *rto, int *perror)
1421 {
1422 	/*
1423 	 * if we got multipath routes, we require users to specify
1424 	 * a matching RTAX_GATEWAY.
1425 	 */
1426 	struct rtentry *rt; // *rto = NULL;
1427 	struct radix_node *rn;
1428 	struct sockaddr *gw;
1429 
1430 	gw = info->rti_info[RTAX_GATEWAY];
1431 	rt = rt_mpath_matchgate(rto, gw);
1432 	if (rt == NULL) {
1433 		*perror = ESRCH;
1434 		return (NULL);
1435 	}
1436 
1437 	/*
1438 	 * this is the first entry in the chain
1439 	 */
1440 	if (rto == rt) {
1441 		rn = rn_mpath_next((struct radix_node *)rt);
1442 		/*
1443 		 * there is another entry, now it's active
1444 		 */
1445 		if (rn) {
1446 			rto = RNTORT(rn);
1447 			RT_LOCK(rto);
1448 			rto->rt_flags |= RTF_UP;
1449 			RT_UNLOCK(rto);
1450 		} else if (rt->rt_flags & RTF_GATEWAY) {
1451 			/*
1452 			 * For gateway routes, we need to
1453 			 * make sure that we we are deleting
1454 			 * the correct gateway.
1455 			 * rt_mpath_matchgate() does not
1456 			 * check the case when there is only
1457 			 * one route in the chain.
1458 			 */
1459 			if (gw &&
1460 			    (rt->rt_nhop->gw_sa.sa_len != gw->sa_len ||
1461 				memcmp(&rt->rt_nhop->gw_sa, gw, gw->sa_len))) {
1462 				*perror = ESRCH;
1463 				return (NULL);
1464 			}
1465 		}
1466 
1467 		/*
1468 		 * use the normal delete code to remove
1469 		 * the first entry
1470 		 */
1471 		rn = rnh->rnh_deladdr(dst, netmask, &rnh->head);
1472 		*perror = 0;
1473 		return (rn);
1474 	}
1475 
1476 	/*
1477 	 * if the entry is 2nd and on up
1478 	 */
1479 	if (rt_mpath_deldup(rto, rt) == 0)
1480 		panic ("rtrequest1: rt_mpath_deldup");
1481 	*perror = 0;
1482 	rn = (struct radix_node *)rt;
1483 	return (rn);
1484 }
1485 #endif
1486 
1487 #undef dst
1488 #undef gateway
1489 #undef netmask
1490 #undef ifaaddr
1491 #undef ifpaddr
1492 #undef flags
1493 
1494 int
1495 rtrequest1_fib(int req, struct rt_addrinfo *info, struct rtentry **ret_nrt,
1496 				u_int fibnum)
1497 {
1498 	const struct sockaddr *dst;
1499 	struct rib_head *rnh;
1500 	int error;
1501 
1502 	KASSERT((fibnum < rt_numfibs), ("rtrequest1_fib: bad fibnum"));
1503 	KASSERT((info->rti_flags & RTF_RNH_LOCKED) == 0, ("rtrequest1_fib: locked"));
1504 
1505 	dst = info->rti_info[RTAX_DST];
1506 
1507 	switch (dst->sa_family) {
1508 	case AF_INET6:
1509 	case AF_INET:
1510 		/* We support multiple FIBs. */
1511 		break;
1512 	default:
1513 		fibnum = RT_DEFAULT_FIB;
1514 		break;
1515 	}
1516 
1517 	/*
1518 	 * Find the correct routing tree to use for this Address Family
1519 	 */
1520 	rnh = rt_tables_get_rnh(fibnum, dst->sa_family);
1521 	if (rnh == NULL)
1522 		return (EAFNOSUPPORT);
1523 
1524 	/*
1525 	 * If we are adding a host route then we don't want to put
1526 	 * a netmask in the tree, nor do we want to clone it.
1527 	 */
1528 	if (info->rti_flags & RTF_HOST)
1529 		info->rti_info[RTAX_NETMASK] = NULL;
1530 
1531 	error = 0;
1532 	switch (req) {
1533 	case RTM_DELETE:
1534 		error = del_route(rnh, info, ret_nrt);
1535 		break;
1536 	case RTM_RESOLVE:
1537 		/*
1538 		 * resolve was only used for route cloning
1539 		 * here for compat
1540 		 */
1541 		break;
1542 	case RTM_ADD:
1543 		error = add_route(rnh, info, ret_nrt);
1544 		break;
1545 	case RTM_CHANGE:
1546 		error = change_route(rnh, info, ret_nrt);
1547 		break;
1548 	default:
1549 		error = EOPNOTSUPP;
1550 	}
1551 
1552 	return (error);
1553 }
1554 
1555 static int
1556 add_route(struct rib_head *rnh, struct rt_addrinfo *info,
1557     struct rtentry **ret_nrt)
1558 {
1559 	struct sockaddr *dst, *ndst, *gateway, *netmask;
1560 	struct rtentry *rt, *rt_old;
1561 	struct nhop_object *nh;
1562 	struct radix_node *rn;
1563 	struct ifaddr *ifa;
1564 	int error, flags;
1565 	struct epoch_tracker et;
1566 
1567 	dst = info->rti_info[RTAX_DST];
1568 	gateway = info->rti_info[RTAX_GATEWAY];
1569 	netmask = info->rti_info[RTAX_NETMASK];
1570 	flags = info->rti_flags;
1571 
1572 	if ((flags & RTF_GATEWAY) && !gateway)
1573 		return (EINVAL);
1574 	if (dst && gateway && (dst->sa_family != gateway->sa_family) &&
1575 	    (gateway->sa_family != AF_UNSPEC) && (gateway->sa_family != AF_LINK))
1576 		return (EINVAL);
1577 
1578 	if (info->rti_ifa == NULL) {
1579 		error = rt_getifa_fib(info, rnh->rib_fibnum);
1580 		if (error)
1581 			return (error);
1582 	} else {
1583 		ifa_ref(info->rti_ifa);
1584 	}
1585 
1586 	NET_EPOCH_ENTER(et);
1587 	error = nhop_create_from_info(rnh, info, &nh);
1588 	NET_EPOCH_EXIT(et);
1589 	if (error != 0) {
1590 		ifa_free(info->rti_ifa);
1591 		return (error);
1592 	}
1593 
1594 	rt = uma_zalloc(V_rtzone, M_NOWAIT);
1595 	if (rt == NULL) {
1596 		ifa_free(info->rti_ifa);
1597 		nhop_free(nh);
1598 		return (ENOBUFS);
1599 	}
1600 	rt->rt_flags = RTF_UP | flags;
1601 	rt->rt_fibnum = rnh->rib_fibnum;
1602 	rt->rt_nhop = nh;
1603 	/*
1604 	 * Add the gateway. Possibly re-malloc-ing the storage for it.
1605 	 */
1606 	if ((error = rt_setgate(rt, dst, gateway)) != 0) {
1607 		ifa_free(info->rti_ifa);
1608 		nhop_free(nh);
1609 		uma_zfree(V_rtzone, rt);
1610 		return (error);
1611 	}
1612 
1613 	/*
1614 	 * point to the (possibly newly malloc'd) dest address.
1615 	 */
1616 	ndst = (struct sockaddr *)rt_key(rt);
1617 
1618 	/*
1619 	 * make sure it contains the value we want (masked if needed).
1620 	 */
1621 	if (netmask) {
1622 		rt_maskedcopy(dst, ndst, netmask);
1623 	} else
1624 		bcopy(dst, ndst, dst->sa_len);
1625 
1626 	/*
1627 	 * We use the ifa reference returned by rt_getifa_fib().
1628 	 * This moved from below so that rnh->rnh_addaddr() can
1629 	 * examine the ifa and  ifa->ifa_ifp if it so desires.
1630 	 */
1631 	ifa = info->rti_ifa;
1632 	rt->rt_weight = 1;
1633 
1634 	rt_setmetrics(info, rt);
1635 
1636 	RIB_WLOCK(rnh);
1637 	RT_LOCK(rt);
1638 #ifdef RADIX_MPATH
1639 	/* do not permit exactly the same dst/mask/gw pair */
1640 	if (rt_mpath_capable(rnh) &&
1641 		rt_mpath_conflict(rnh, rt, netmask)) {
1642 		RIB_WUNLOCK(rnh);
1643 
1644 		R_Free(rt_key(rt));
1645 		nhop_free(nh);
1646 		uma_zfree(V_rtzone, rt);
1647 		return (EEXIST);
1648 	}
1649 #endif
1650 
1651 	rn = rnh->rnh_addaddr(ndst, netmask, &rnh->head, rt->rt_nodes);
1652 
1653 	if (rn != NULL && rt->rt_expire > 0)
1654 		tmproutes_update(rnh, rt);
1655 
1656 	rt_old = NULL;
1657 	if (rn == NULL && (info->rti_flags & RTF_PINNED) != 0) {
1658 
1659 		/*
1660 		 * Force removal and re-try addition
1661 		 * TODO: better multipath&pinned support
1662 		 */
1663 		struct sockaddr *info_dst = info->rti_info[RTAX_DST];
1664 		info->rti_info[RTAX_DST] = ndst;
1665 		/* Do not delete existing PINNED(interface) routes */
1666 		info->rti_flags &= ~RTF_PINNED;
1667 		rt_old = rt_unlinkrte(rnh, info, &error);
1668 		info->rti_flags |= RTF_PINNED;
1669 		info->rti_info[RTAX_DST] = info_dst;
1670 		if (rt_old != NULL)
1671 			rn = rnh->rnh_addaddr(ndst, netmask, &rnh->head,
1672 			    rt->rt_nodes);
1673 	}
1674 	RIB_WUNLOCK(rnh);
1675 
1676 	if (rt_old != NULL)
1677 		RT_UNLOCK(rt_old);
1678 
1679 	/*
1680 	 * If it still failed to go into the tree,
1681 	 * then un-make it (this should be a function)
1682 	 */
1683 	if (rn == NULL) {
1684 		R_Free(rt_key(rt));
1685 		nhop_free(nh);
1686 		uma_zfree(V_rtzone, rt);
1687 		return (EEXIST);
1688 	}
1689 
1690 	if (rt_old != NULL) {
1691 		rt_notifydelete(rt_old, info);
1692 		RTFREE(rt_old);
1693 	}
1694 
1695 	/*
1696 	 * If this protocol has something to add to this then
1697 	 * allow it to do that as well.
1698 	 */
1699 	if (ifa->ifa_rtrequest)
1700 		ifa->ifa_rtrequest(RTM_ADD, rt, rt->rt_nhop, info);
1701 
1702 	/*
1703 	 * actually return a resultant rtentry and
1704 	 * give the caller a single reference.
1705 	 */
1706 	if (ret_nrt) {
1707 		*ret_nrt = rt;
1708 		RT_ADDREF(rt);
1709 	}
1710 	rnh->rnh_gen++;		/* Routing table updated */
1711 	RT_UNLOCK(rt);
1712 
1713 	return (0);
1714 }
1715 
1716 static int
1717 del_route(struct rib_head *rnh, struct rt_addrinfo *info,
1718     struct rtentry **ret_nrt)
1719 {
1720 	struct sockaddr *dst, *netmask;
1721 	struct sockaddr_storage mdst;
1722 	struct rtentry *rt;
1723 	int error;
1724 
1725 	dst = info->rti_info[RTAX_DST];
1726 	netmask = info->rti_info[RTAX_NETMASK];
1727 
1728 	if (netmask) {
1729 		if (dst->sa_len > sizeof(mdst))
1730 			return (EINVAL);
1731 		rt_maskedcopy(dst, (struct sockaddr *)&mdst, netmask);
1732 		dst = (struct sockaddr *)&mdst;
1733 	}
1734 
1735 	RIB_WLOCK(rnh);
1736 	rt = rt_unlinkrte(rnh, info, &error);
1737 	RIB_WUNLOCK(rnh);
1738 	if (error != 0)
1739 		return (error);
1740 
1741 	rt_notifydelete(rt, info);
1742 
1743 	/*
1744 	 * If the caller wants it, then it can have it,
1745 	 * but it's up to it to free the rtentry as we won't be
1746 	 * doing it.
1747 	 */
1748 	if (ret_nrt) {
1749 		*ret_nrt = rt;
1750 		RT_UNLOCK(rt);
1751 	} else
1752 		RTFREE_LOCKED(rt);
1753 
1754 	return (0);
1755 }
1756 
1757 static int
1758 change_route_one(struct rib_head *rnh, struct rt_addrinfo *info,
1759     struct rtentry **ret_nrt)
1760 {
1761 	RIB_RLOCK_TRACKER;
1762 	struct rtentry *rt = NULL;
1763 	int error = 0;
1764 	int free_ifa = 0;
1765 	struct nhop_object *nh, *nh_orig;
1766 
1767 	RIB_RLOCK(rnh);
1768 	rt = (struct rtentry *)rnh->rnh_lookup(info->rti_info[RTAX_DST],
1769 	    info->rti_info[RTAX_NETMASK], &rnh->head);
1770 
1771 	if (rt == NULL) {
1772 		RIB_RUNLOCK(rnh);
1773 		return (ESRCH);
1774 	}
1775 
1776 #ifdef RADIX_MPATH
1777 	/*
1778 	 * If we got multipath routes,
1779 	 * we require users to specify a matching RTAX_GATEWAY.
1780 	 */
1781 	if (rt_mpath_capable(rnh)) {
1782 		rt = rt_mpath_matchgate(rt, info->rti_info[RTAX_GATEWAY]);
1783 		if (rt == NULL) {
1784 			RIB_RUNLOCK(rnh);
1785 			return (ESRCH);
1786 		}
1787 	}
1788 #endif
1789 	nh_orig = rt->rt_nhop;
1790 
1791 	RIB_RUNLOCK(rnh);
1792 
1793 	rt = NULL;
1794 	nh = NULL;
1795 
1796 	/*
1797 	 * New gateway could require new ifaddr, ifp;
1798 	 * flags may also be different; ifp may be specified
1799 	 * by ll sockaddr when protocol address is ambiguous
1800 	 */
1801 	if (((nh_orig->nh_flags & NHF_GATEWAY) &&
1802 	    info->rti_info[RTAX_GATEWAY] != NULL) ||
1803 	    info->rti_info[RTAX_IFP] != NULL ||
1804 	    (info->rti_info[RTAX_IFA] != NULL &&
1805 	     !sa_equal(info->rti_info[RTAX_IFA], nh_orig->nh_ifa->ifa_addr))) {
1806 		error = rt_getifa_fib(info, rnh->rib_fibnum);
1807 		if (info->rti_ifa != NULL)
1808 			free_ifa = 1;
1809 
1810 		if (error != 0) {
1811 			if (free_ifa) {
1812 				ifa_free(info->rti_ifa);
1813 				info->rti_ifa = NULL;
1814 			}
1815 
1816 			return (error);
1817 		}
1818 	}
1819 
1820 	error = nhop_create_from_nhop(rnh, nh_orig, info, &nh);
1821 	if (free_ifa) {
1822 		ifa_free(info->rti_ifa);
1823 		info->rti_ifa = NULL;
1824 	}
1825 	if (error != 0)
1826 		return (error);
1827 
1828 	RIB_WLOCK(rnh);
1829 
1830 	/* Lookup rtentry once again and check if nexthop is still the same */
1831 	rt = (struct rtentry *)rnh->rnh_lookup(info->rti_info[RTAX_DST],
1832 	    info->rti_info[RTAX_NETMASK], &rnh->head);
1833 
1834 	if (rt == NULL) {
1835 		RIB_WUNLOCK(rnh);
1836 		nhop_free(nh);
1837 		return (ESRCH);
1838 	}
1839 
1840 	if (rt->rt_nhop != nh_orig) {
1841 		RIB_WUNLOCK(rnh);
1842 		nhop_free(nh);
1843 		return (EAGAIN);
1844 	}
1845 
1846 	/* Proceed with the update */
1847 	RT_LOCK(rt);
1848 
1849 	/* Provide notification to the protocols.*/
1850 	if ((nh_orig->nh_ifa != nh->nh_ifa) && nh_orig->nh_ifa->ifa_rtrequest)
1851 		nh_orig->nh_ifa->ifa_rtrequest(RTM_DELETE, rt, nh_orig, info);
1852 
1853 	rt->rt_nhop = nh;
1854 	rt_setmetrics(info, rt);
1855 
1856 	if ((nh_orig->nh_ifa != nh->nh_ifa) && nh_orig->nh_ifa->ifa_rtrequest)
1857 		nh_orig->nh_ifa->ifa_rtrequest(RTM_DELETE, rt, nh_orig, info);
1858 
1859 	if (ret_nrt != NULL) {
1860 		*ret_nrt = rt;
1861 		RT_ADDREF(rt);
1862 	}
1863 
1864 	RT_UNLOCK(rt);
1865 
1866 	/* Update generation id to reflect rtable change */
1867 	rnh->rnh_gen++;
1868 
1869 	RIB_WUNLOCK(rnh);
1870 
1871 	nhop_free(nh_orig);
1872 
1873 	return (0);
1874 }
1875 
1876 static int
1877 change_route(struct rib_head *rnh, struct rt_addrinfo *info,
1878     struct rtentry **ret_nrt)
1879 {
1880 	struct epoch_tracker et;
1881 	int error;
1882 
1883 	/* Check if updated gateway exists */
1884 	if ((info->rti_flags & RTF_GATEWAY) &&
1885 	    (info->rti_info[RTAX_GATEWAY] == NULL))
1886 		return (EINVAL);
1887 
1888 	NET_EPOCH_ENTER(et);
1889 
1890 	/*
1891 	 * route change is done in multiple steps, with dropping and
1892 	 * reacquiring lock. In the situations with multiple processes
1893 	 * changes the same route in can lead to the case when route
1894 	 * is changed between the steps. Address it by retrying the operation
1895 	 * multiple times before failing.
1896 	 */
1897 	for (int i = 0; i < RIB_MAX_RETRIES; i++) {
1898 		error = change_route_one(rnh, info, ret_nrt);
1899 		if (error != EAGAIN)
1900 			break;
1901 	}
1902 	NET_EPOCH_EXIT(et);
1903 
1904 	return (error);
1905 }
1906 
1907 
1908 static void
1909 rt_setmetrics(const struct rt_addrinfo *info, struct rtentry *rt)
1910 {
1911 
1912 	if (info->rti_mflags & RTV_WEIGHT)
1913 		rt->rt_weight = info->rti_rmx->rmx_weight;
1914 	/* Kernel -> userland timebase conversion. */
1915 	if (info->rti_mflags & RTV_EXPIRE)
1916 		rt->rt_expire = info->rti_rmx->rmx_expire ?
1917 		    info->rti_rmx->rmx_expire - time_second + time_uptime : 0;
1918 }
1919 
1920 int
1921 rt_setgate(struct rtentry *rt, struct sockaddr *dst, struct sockaddr *gate)
1922 {
1923 	/* XXX dst may be overwritten, can we move this to below */
1924 	int dlen = SA_SIZE(dst), glen = SA_SIZE(gate);
1925 
1926 	/*
1927 	 * Prepare to store the gateway in rt->rt_gateway.
1928 	 * Both dst and gateway are stored one after the other in the same
1929 	 * malloc'd chunk. If we have room, we can reuse the old buffer,
1930 	 * rt_gateway already points to the right place.
1931 	 * Otherwise, malloc a new block and update the 'dst' address.
1932 	 */
1933 	if (rt_key(rt) == NULL) {
1934 		caddr_t new;
1935 
1936 		R_Malloc(new, caddr_t, dlen + glen);
1937 		if (new == NULL)
1938 			return ENOBUFS;
1939 		/*
1940 		 * XXX note, we copy from *dst and not *rt_key(rt) because
1941 		 * rt_setgate() can be called to initialize a newly
1942 		 * allocated route entry, in which case rt_key(rt) == NULL
1943 		 * (and also rt->rt_gateway == NULL).
1944 		 * Free()/free() handle a NULL argument just fine.
1945 		 */
1946 		bcopy(dst, new, dlen);
1947 		R_Free(rt_key(rt));	/* free old block, if any */
1948 		rt_key(rt) = (struct sockaddr *)new;
1949 	}
1950 
1951 	return (0);
1952 }
1953 
1954 void
1955 rt_maskedcopy(struct sockaddr *src, struct sockaddr *dst, struct sockaddr *netmask)
1956 {
1957 	u_char *cp1 = (u_char *)src;
1958 	u_char *cp2 = (u_char *)dst;
1959 	u_char *cp3 = (u_char *)netmask;
1960 	u_char *cplim = cp2 + *cp3;
1961 	u_char *cplim2 = cp2 + *cp1;
1962 
1963 	*cp2++ = *cp1++; *cp2++ = *cp1++; /* copies sa_len & sa_family */
1964 	cp3 += 2;
1965 	if (cplim > cplim2)
1966 		cplim = cplim2;
1967 	while (cp2 < cplim)
1968 		*cp2++ = *cp1++ & *cp3++;
1969 	if (cp2 < cplim2)
1970 		bzero((caddr_t)cp2, (unsigned)(cplim2 - cp2));
1971 }
1972 
1973 /*
1974  * Set up a routing table entry, normally
1975  * for an interface.
1976  */
1977 #define _SOCKADDR_TMPSIZE 128 /* Not too big.. kernel stack size is limited */
1978 static inline  int
1979 rtinit1(struct ifaddr *ifa, int cmd, int flags, int fibnum)
1980 {
1981 	RIB_RLOCK_TRACKER;
1982 	struct sockaddr *dst;
1983 	struct sockaddr *netmask;
1984 	struct rtentry *rt = NULL;
1985 	struct rt_addrinfo info;
1986 	int error = 0;
1987 	int startfib, endfib;
1988 	char tempbuf[_SOCKADDR_TMPSIZE];
1989 	int didwork = 0;
1990 	int a_failure = 0;
1991 	struct sockaddr_dl_short *sdl = NULL;
1992 	struct rib_head *rnh;
1993 
1994 	if (flags & RTF_HOST) {
1995 		dst = ifa->ifa_dstaddr;
1996 		netmask = NULL;
1997 	} else {
1998 		dst = ifa->ifa_addr;
1999 		netmask = ifa->ifa_netmask;
2000 	}
2001 	if (dst->sa_len == 0)
2002 		return(EINVAL);
2003 	switch (dst->sa_family) {
2004 	case AF_INET6:
2005 	case AF_INET:
2006 		/* We support multiple FIBs. */
2007 		break;
2008 	default:
2009 		fibnum = RT_DEFAULT_FIB;
2010 		break;
2011 	}
2012 	if (fibnum == RT_ALL_FIBS) {
2013 		if (V_rt_add_addr_allfibs == 0 && cmd == (int)RTM_ADD)
2014 			startfib = endfib = ifa->ifa_ifp->if_fib;
2015 		else {
2016 			startfib = 0;
2017 			endfib = rt_numfibs - 1;
2018 		}
2019 	} else {
2020 		KASSERT((fibnum < rt_numfibs), ("rtinit1: bad fibnum"));
2021 		startfib = fibnum;
2022 		endfib = fibnum;
2023 	}
2024 
2025 	/*
2026 	 * If it's a delete, check that if it exists,
2027 	 * it's on the correct interface or we might scrub
2028 	 * a route to another ifa which would
2029 	 * be confusing at best and possibly worse.
2030 	 */
2031 	if (cmd == RTM_DELETE) {
2032 		/*
2033 		 * It's a delete, so it should already exist..
2034 		 * If it's a net, mask off the host bits
2035 		 * (Assuming we have a mask)
2036 		 * XXX this is kinda inet specific..
2037 		 */
2038 		if (netmask != NULL) {
2039 			rt_maskedcopy(dst, (struct sockaddr *)tempbuf, netmask);
2040 			dst = (struct sockaddr *)tempbuf;
2041 		}
2042 	} else if (cmd == RTM_ADD) {
2043 		sdl = (struct sockaddr_dl_short *)tempbuf;
2044 		bzero(sdl, sizeof(struct sockaddr_dl_short));
2045 		sdl->sdl_family = AF_LINK;
2046 		sdl->sdl_len = sizeof(struct sockaddr_dl_short);
2047 		sdl->sdl_type = ifa->ifa_ifp->if_type;
2048 		sdl->sdl_index = ifa->ifa_ifp->if_index;
2049         }
2050 	/*
2051 	 * Now go through all the requested tables (fibs) and do the
2052 	 * requested action. Realistically, this will either be fib 0
2053 	 * for protocols that don't do multiple tables or all the
2054 	 * tables for those that do.
2055 	 */
2056 	for ( fibnum = startfib; fibnum <= endfib; fibnum++) {
2057 		if (cmd == RTM_DELETE) {
2058 			struct radix_node *rn;
2059 			/*
2060 			 * Look up an rtentry that is in the routing tree and
2061 			 * contains the correct info.
2062 			 */
2063 			rnh = rt_tables_get_rnh(fibnum, dst->sa_family);
2064 			if (rnh == NULL)
2065 				/* this table doesn't exist but others might */
2066 				continue;
2067 			RIB_RLOCK(rnh);
2068 			rn = rnh->rnh_lookup(dst, netmask, &rnh->head);
2069 #ifdef RADIX_MPATH
2070 			if (rt_mpath_capable(rnh)) {
2071 
2072 				if (rn == NULL)
2073 					error = ESRCH;
2074 				else {
2075 					rt = RNTORT(rn);
2076 					/*
2077 					 * for interface route the gateway
2078 					 * gateway is sockaddr_dl, so
2079 					 * rt_mpath_matchgate must use the
2080 					 * interface address
2081 					 */
2082 					rt = rt_mpath_matchgate(rt,
2083 					    ifa->ifa_addr);
2084 					if (rt == NULL)
2085 						error = ESRCH;
2086 				}
2087 			}
2088 #endif
2089 			error = (rn == NULL ||
2090 			    (rn->rn_flags & RNF_ROOT) ||
2091 			    RNTORT(rn)->rt_nhop->nh_ifa != ifa);
2092 			RIB_RUNLOCK(rnh);
2093 			if (error) {
2094 				/* this is only an error if bad on ALL tables */
2095 				continue;
2096 			}
2097 		}
2098 		/*
2099 		 * Do the actual request
2100 		 */
2101 		bzero((caddr_t)&info, sizeof(info));
2102 		info.rti_ifa = ifa;
2103 		info.rti_flags = flags |
2104 		    (ifa->ifa_flags & ~IFA_RTSELF) | RTF_PINNED;
2105 		info.rti_info[RTAX_DST] = dst;
2106 		/*
2107 		 * doing this for compatibility reasons
2108 		 */
2109 		if (cmd == RTM_ADD)
2110 			info.rti_info[RTAX_GATEWAY] = (struct sockaddr *)sdl;
2111 		else
2112 			info.rti_info[RTAX_GATEWAY] = ifa->ifa_addr;
2113 		info.rti_info[RTAX_NETMASK] = netmask;
2114 		error = rtrequest1_fib(cmd, &info, &rt, fibnum);
2115 		if (error == 0 && rt != NULL) {
2116 			/*
2117 			 * notify any listening routing agents of the change
2118 			 */
2119 			RT_LOCK(rt);
2120 
2121 			/* TODO: interface routes/aliases */
2122 			RT_ADDREF(rt);
2123 			RT_UNLOCK(rt);
2124 			rt_newaddrmsg_fib(cmd, ifa, rt, fibnum);
2125 			RT_LOCK(rt);
2126 			RT_REMREF(rt);
2127 			if (cmd == RTM_DELETE) {
2128 				/*
2129 				 * If we are deleting, and we found an entry,
2130 				 * then it's been removed from the tree..
2131 				 * now throw it away.
2132 				 */
2133 				RTFREE_LOCKED(rt);
2134 			} else {
2135 				if (cmd == RTM_ADD) {
2136 					/*
2137 					 * We just wanted to add it..
2138 					 * we don't actually need a reference.
2139 					 */
2140 					RT_REMREF(rt);
2141 				}
2142 				RT_UNLOCK(rt);
2143 			}
2144 			didwork = 1;
2145 		}
2146 		if (error)
2147 			a_failure = error;
2148 	}
2149 	if (cmd == RTM_DELETE) {
2150 		if (didwork) {
2151 			error = 0;
2152 		} else {
2153 			/* we only give an error if it wasn't in any table */
2154 			error = ((flags & RTF_HOST) ?
2155 			    EHOSTUNREACH : ENETUNREACH);
2156 		}
2157 	} else {
2158 		if (a_failure) {
2159 			/* return an error if any of them failed */
2160 			error = a_failure;
2161 		}
2162 	}
2163 	return (error);
2164 }
2165 
2166 /*
2167  * Set up a routing table entry, normally
2168  * for an interface.
2169  */
2170 int
2171 rtinit(struct ifaddr *ifa, int cmd, int flags)
2172 {
2173 	struct sockaddr *dst;
2174 	int fib = RT_DEFAULT_FIB;
2175 
2176 	if (flags & RTF_HOST) {
2177 		dst = ifa->ifa_dstaddr;
2178 	} else {
2179 		dst = ifa->ifa_addr;
2180 	}
2181 
2182 	switch (dst->sa_family) {
2183 	case AF_INET6:
2184 	case AF_INET:
2185 		/* We do support multiple FIBs. */
2186 		fib = RT_ALL_FIBS;
2187 		break;
2188 	}
2189 	return (rtinit1(ifa, cmd, flags, fib));
2190 }
2191 
2192 /*
2193  * Announce interface address arrival/withdraw
2194  * Returns 0 on success.
2195  */
2196 int
2197 rt_addrmsg(int cmd, struct ifaddr *ifa, int fibnum)
2198 {
2199 
2200 	KASSERT(cmd == RTM_ADD || cmd == RTM_DELETE,
2201 	    ("unexpected cmd %d", cmd));
2202 	KASSERT(fibnum == RT_ALL_FIBS || (fibnum >= 0 && fibnum < rt_numfibs),
2203 	    ("%s: fib out of range 0 <=%d<%d", __func__, fibnum, rt_numfibs));
2204 
2205 	EVENTHANDLER_DIRECT_INVOKE(rt_addrmsg, ifa, cmd);
2206 	return (rtsock_addrmsg(cmd, ifa, fibnum));
2207 }
2208 
2209 /*
2210  * Announce kernel-originated route addition/removal to rtsock based on @rt data.
2211  * cmd: RTM_ cmd
2212  * @rt: valid rtentry
2213  * @ifp: target route interface
2214  * @fibnum: fib id or RT_ALL_FIBS
2215  *
2216  * Returns 0 on success.
2217  */
2218 int
2219 rt_routemsg(int cmd, struct rtentry *rt, struct ifnet *ifp, int rti_addrs,
2220     int fibnum)
2221 {
2222 
2223 	KASSERT(cmd == RTM_ADD || cmd == RTM_DELETE,
2224 	    ("unexpected cmd %d", cmd));
2225 
2226 	KASSERT(fibnum == RT_ALL_FIBS || (fibnum >= 0 && fibnum < rt_numfibs),
2227 	    ("%s: fib out of range 0 <=%d<%d", __func__, fibnum, rt_numfibs));
2228 
2229 	KASSERT(rt_key(rt) != NULL, (":%s: rt_key must be supplied", __func__));
2230 
2231 	return (rtsock_routemsg(cmd, rt, ifp, 0, fibnum));
2232 }
2233 
2234 /*
2235  * Announce kernel-originated route addition/removal to rtsock based on @rt data.
2236  * cmd: RTM_ cmd
2237  * @info: addrinfo structure with valid data.
2238  * @fibnum: fib id or RT_ALL_FIBS
2239  *
2240  * Returns 0 on success.
2241  */
2242 int
2243 rt_routemsg_info(int cmd, struct rt_addrinfo *info, int fibnum)
2244 {
2245 
2246 	KASSERT(cmd == RTM_ADD || cmd == RTM_DELETE || cmd == RTM_CHANGE,
2247 	    ("unexpected cmd %d", cmd));
2248 
2249 	KASSERT(fibnum == RT_ALL_FIBS || (fibnum >= 0 && fibnum < rt_numfibs),
2250 	    ("%s: fib out of range 0 <=%d<%d", __func__, fibnum, rt_numfibs));
2251 
2252 	KASSERT(info->rti_info[RTAX_DST] != NULL, (":%s: RTAX_DST must be supplied", __func__));
2253 
2254 	return (rtsock_routemsg_info(cmd, info, fibnum));
2255 }
2256 
2257 
2258 /*
2259  * This is called to generate messages from the routing socket
2260  * indicating a network interface has had addresses associated with it.
2261  */
2262 void
2263 rt_newaddrmsg_fib(int cmd, struct ifaddr *ifa, struct rtentry *rt, int fibnum)
2264 {
2265 
2266 	KASSERT(cmd == RTM_ADD || cmd == RTM_DELETE,
2267 		("unexpected cmd %u", cmd));
2268 	KASSERT(fibnum == RT_ALL_FIBS || (fibnum >= 0 && fibnum < rt_numfibs),
2269 	    ("%s: fib out of range 0 <=%d<%d", __func__, fibnum, rt_numfibs));
2270 
2271 	if (cmd == RTM_ADD) {
2272 		rt_addrmsg(cmd, ifa, fibnum);
2273 		if (rt != NULL)
2274 			rt_routemsg(cmd, rt, ifa->ifa_ifp, 0, fibnum);
2275 	} else {
2276 		if (rt != NULL)
2277 			rt_routemsg(cmd, rt, ifa->ifa_ifp, 0, fibnum);
2278 		rt_addrmsg(cmd, ifa, fibnum);
2279 	}
2280 }
2281 
2282