xref: /freebsd/sys/net/route.c (revision 0c54932d50a0cbffdd083bf6b2e8d587902f90c9)
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_ctl.h>
65 #include <net/route/route_var.h>
66 #include <net/route/nhop.h>
67 #include <net/route/shared.h>
68 #include <net/vnet.h>
69 
70 #ifdef RADIX_MPATH
71 #include <net/radix_mpath.h>
72 #endif
73 
74 #include <netinet/in.h>
75 #include <netinet/ip_mroute.h>
76 
77 /*
78  * By default add routes to all fibs for new interfaces.
79  * Once this is set to 0 then only allocate routes on interface
80  * changes for the FIB of the caller when adding a new set of addresses
81  * to an interface.  XXX this is a shotgun aproach to a problem that needs
82  * a more fine grained solution.. that will come.
83  * XXX also has the problems getting the FIB from curthread which will not
84  * always work given the fib can be overridden and prefixes can be added
85  * from the network stack context.
86  */
87 VNET_DEFINE(u_int, rt_add_addr_allfibs) = 1;
88 SYSCTL_UINT(_net, OID_AUTO, add_addr_allfibs, CTLFLAG_RWTUN | CTLFLAG_VNET,
89     &VNET_NAME(rt_add_addr_allfibs), 0, "");
90 
91 VNET_PCPUSTAT_DEFINE(struct rtstat, rtstat);
92 
93 VNET_PCPUSTAT_SYSINIT(rtstat);
94 #ifdef VIMAGE
95 VNET_PCPUSTAT_SYSUNINIT(rtstat);
96 #endif
97 
98 EVENTHANDLER_LIST_DEFINE(rt_addrmsg);
99 
100 static int rt_ifdelroute(const struct rtentry *rt, const struct nhop_object *,
101     void *arg);
102 static int rt_exportinfo(struct rtentry *rt, struct rt_addrinfo *info,
103     int flags);
104 
105 /*
106  * route initialization must occur before ip6_init2(), which happenas at
107  * SI_ORDER_MIDDLE.
108  */
109 static void
110 route_init(void)
111 {
112 
113 	nhops_init();
114 }
115 SYSINIT(route_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_THIRD, route_init, NULL);
116 
117 struct rib_head *
118 rt_table_init(int offset, int family, u_int fibnum)
119 {
120 	struct rib_head *rh;
121 
122 	rh = malloc(sizeof(struct rib_head), M_RTABLE, M_WAITOK | M_ZERO);
123 
124 	/* TODO: These details should be hidded inside radix.c */
125 	/* Init masks tree */
126 	rn_inithead_internal(&rh->head, rh->rnh_nodes, offset);
127 	rn_inithead_internal(&rh->rmhead.head, rh->rmhead.mask_nodes, 0);
128 	rh->head.rnh_masks = &rh->rmhead;
129 
130 	/* Save metadata associated with this routing table. */
131 	rh->rib_family = family;
132 	rh->rib_fibnum = fibnum;
133 #ifdef VIMAGE
134 	rh->rib_vnet = curvnet;
135 #endif
136 
137 	tmproutes_init(rh);
138 
139 	/* Init locks */
140 	RIB_LOCK_INIT(rh);
141 
142 	nhops_init_rib(rh);
143 
144 	/* Init subscription system */
145 	rib_init_subscriptions(rh);
146 
147 	/* Finally, set base callbacks */
148 	rh->rnh_addaddr = rn_addroute;
149 	rh->rnh_deladdr = rn_delete;
150 	rh->rnh_matchaddr = rn_match;
151 	rh->rnh_lookup = rn_lookup;
152 	rh->rnh_walktree = rn_walktree;
153 	rh->rnh_walktree_from = rn_walktree_from;
154 
155 	return (rh);
156 }
157 
158 static int
159 rt_freeentry(struct radix_node *rn, void *arg)
160 {
161 	struct radix_head * const rnh = arg;
162 	struct radix_node *x;
163 
164 	x = (struct radix_node *)rn_delete(rn + 2, NULL, rnh);
165 	if (x != NULL)
166 		R_Free(x);
167 	return (0);
168 }
169 
170 void
171 rt_table_destroy(struct rib_head *rh)
172 {
173 
174 	tmproutes_destroy(rh);
175 
176 	rn_walktree(&rh->rmhead.head, rt_freeentry, &rh->rmhead.head);
177 
178 	nhops_destroy_rib(rh);
179 
180 	rib_destroy_subscriptions(rh);
181 
182 	/* Assume table is already empty */
183 	RIB_LOCK_DESTROY(rh);
184 	free(rh, M_RTABLE);
185 }
186 
187 /*
188  * Adds a temporal redirect entry to the routing table.
189  * @fibnum: fib number
190  * @dst: destination to install redirect to
191  * @gateway: gateway to go via
192  * @author: sockaddr of originating router, can be NULL
193  * @ifp: interface to use for the redirected route
194  * @flags: set of flags to add. Allowed: RTF_GATEWAY
195  * @lifetime_sec: time in seconds to expire this redirect.
196  *
197  * Retuns 0 on success, errno otherwise.
198  */
199 int
200 rib_add_redirect(u_int fibnum, struct sockaddr *dst, struct sockaddr *gateway,
201     struct sockaddr *author, struct ifnet *ifp, int flags, int lifetime_sec)
202 {
203 	struct rib_cmd_info rc;
204 	int error;
205 	struct rt_addrinfo info;
206 	struct rt_metrics rti_rmx;
207 	struct ifaddr *ifa;
208 
209 	NET_EPOCH_ASSERT();
210 
211 	if (rt_tables_get_rnh(fibnum, dst->sa_family) == NULL)
212 		return (EAFNOSUPPORT);
213 
214 	/* Verify the allowed flag mask. */
215 	KASSERT(((flags & ~(RTF_GATEWAY)) == 0),
216 	    ("invalid redirect flags: %x", flags));
217 	flags |= RTF_HOST | RTF_DYNAMIC;
218 
219 	/* Get the best ifa for the given interface and gateway. */
220 	if ((ifa = ifaof_ifpforaddr(gateway, ifp)) == NULL)
221 		return (ENETUNREACH);
222 	ifa_ref(ifa);
223 
224 	bzero(&info, sizeof(info));
225 	info.rti_info[RTAX_DST] = dst;
226 	info.rti_info[RTAX_GATEWAY] = gateway;
227 	info.rti_ifa = ifa;
228 	info.rti_ifp = ifp;
229 	info.rti_flags = flags;
230 
231 	/* Setup route metrics to define expire time. */
232 	bzero(&rti_rmx, sizeof(rti_rmx));
233 	/* Set expire time as absolute. */
234 	rti_rmx.rmx_expire = lifetime_sec + time_second;
235 	info.rti_mflags |= RTV_EXPIRE;
236 	info.rti_rmx = &rti_rmx;
237 
238 	error = rib_action(fibnum, RTM_ADD, &info, &rc);
239 	ifa_free(ifa);
240 
241 	if (error != 0) {
242 		/* TODO: add per-fib redirect stats. */
243 		return (error);
244 	}
245 
246 	RTSTAT_INC(rts_dynamic);
247 
248 	/* Send notification of a route addition to userland. */
249 	bzero(&info, sizeof(info));
250 	info.rti_info[RTAX_DST] = dst;
251 	info.rti_info[RTAX_GATEWAY] = gateway;
252 	info.rti_info[RTAX_AUTHOR] = author;
253 	rt_missmsg_fib(RTM_REDIRECT, &info, flags | RTF_UP, error, fibnum);
254 
255 	return (0);
256 }
257 
258 /*
259  * Routing table ioctl interface.
260  */
261 int
262 rtioctl_fib(u_long req, caddr_t data, u_int fibnum)
263 {
264 
265 	/*
266 	 * If more ioctl commands are added here, make sure the proper
267 	 * super-user checks are being performed because it is possible for
268 	 * prison-root to make it this far if raw sockets have been enabled
269 	 * in jails.
270 	 */
271 #ifdef INET
272 	/* Multicast goop, grrr... */
273 	return mrt_ioctl ? mrt_ioctl(req, data, fibnum) : EOPNOTSUPP;
274 #else /* INET */
275 	return ENXIO;
276 #endif /* INET */
277 }
278 
279 struct ifaddr *
280 ifa_ifwithroute(int flags, const struct sockaddr *dst,
281     const struct sockaddr *gateway, u_int fibnum)
282 {
283 	struct ifaddr *ifa;
284 
285 	NET_EPOCH_ASSERT();
286 	if ((flags & RTF_GATEWAY) == 0) {
287 		/*
288 		 * If we are adding a route to an interface,
289 		 * and the interface is a pt to pt link
290 		 * we should search for the destination
291 		 * as our clue to the interface.  Otherwise
292 		 * we can use the local address.
293 		 */
294 		ifa = NULL;
295 		if (flags & RTF_HOST)
296 			ifa = ifa_ifwithdstaddr(dst, fibnum);
297 		if (ifa == NULL)
298 			ifa = ifa_ifwithaddr(gateway);
299 	} else {
300 		/*
301 		 * If we are adding a route to a remote net
302 		 * or host, the gateway may still be on the
303 		 * other end of a pt to pt link.
304 		 */
305 		ifa = ifa_ifwithdstaddr(gateway, fibnum);
306 	}
307 	if (ifa == NULL)
308 		ifa = ifa_ifwithnet(gateway, 0, fibnum);
309 	if (ifa == NULL) {
310 		struct nhop_object *nh;
311 
312 		nh = rib_lookup(fibnum, gateway, NHR_NONE, 0);
313 
314 		/*
315 		 * dismiss a gateway that is reachable only
316 		 * through the default router
317 		 */
318 		if ((nh == NULL) || (nh->nh_flags & NHF_DEFAULT))
319 			return (NULL);
320 		ifa = nh->nh_ifa;
321 	}
322 	if (ifa->ifa_addr->sa_family != dst->sa_family) {
323 		struct ifaddr *oifa = ifa;
324 		ifa = ifaof_ifpforaddr(dst, ifa->ifa_ifp);
325 		if (ifa == NULL)
326 			ifa = oifa;
327 	}
328 
329 	return (ifa);
330 }
331 
332 
333 /*
334  * Copy most of @rt data into @info.
335  *
336  * If @flags contains NHR_COPY, copies dst,netmask and gw to the
337  * pointers specified by @info structure. Assume such pointers
338  * are zeroed sockaddr-like structures with sa_len field initialized
339  * to reflect size of the provided buffer. if no NHR_COPY is specified,
340  * point dst,netmask and gw @info fields to appropriate @rt values.
341  *
342  * if @flags contains NHR_REF, do refcouting on rt_ifp and rt_ifa.
343  *
344  * Returns 0 on success.
345  */
346 int
347 rt_exportinfo(struct rtentry *rt, struct rt_addrinfo *info, int flags)
348 {
349 	struct rt_metrics *rmx;
350 	struct sockaddr *src, *dst;
351 	struct nhop_object *nh;
352 	int sa_len;
353 
354 	nh = rt->rt_nhop;
355 	if (flags & NHR_COPY) {
356 		/* Copy destination if dst is non-zero */
357 		src = rt_key(rt);
358 		dst = info->rti_info[RTAX_DST];
359 		sa_len = src->sa_len;
360 		if (dst != NULL) {
361 			if (src->sa_len > dst->sa_len)
362 				return (ENOMEM);
363 			memcpy(dst, src, src->sa_len);
364 			info->rti_addrs |= RTA_DST;
365 		}
366 
367 		/* Copy mask if set && dst is non-zero */
368 		src = rt_mask(rt);
369 		dst = info->rti_info[RTAX_NETMASK];
370 		if (src != NULL && dst != NULL) {
371 
372 			/*
373 			 * Radix stores different value in sa_len,
374 			 * assume rt_mask() to have the same length
375 			 * as rt_key()
376 			 */
377 			if (sa_len > dst->sa_len)
378 				return (ENOMEM);
379 			memcpy(dst, src, src->sa_len);
380 			info->rti_addrs |= RTA_NETMASK;
381 		}
382 
383 		/* Copy gateway is set && dst is non-zero */
384 		src = &nh->gw_sa;
385 		dst = info->rti_info[RTAX_GATEWAY];
386 		if ((nhop_get_rtflags(nh) & RTF_GATEWAY) &&
387 		    src != NULL && dst != NULL) {
388 			if (src->sa_len > dst->sa_len)
389 				return (ENOMEM);
390 			memcpy(dst, src, src->sa_len);
391 			info->rti_addrs |= RTA_GATEWAY;
392 		}
393 	} else {
394 		info->rti_info[RTAX_DST] = rt_key(rt);
395 		info->rti_addrs |= RTA_DST;
396 		if (rt_mask(rt) != NULL) {
397 			info->rti_info[RTAX_NETMASK] = rt_mask(rt);
398 			info->rti_addrs |= RTA_NETMASK;
399 		}
400 		if (nhop_get_rtflags(nh) & RTF_GATEWAY) {
401 			info->rti_info[RTAX_GATEWAY] = &nh->gw_sa;
402 			info->rti_addrs |= RTA_GATEWAY;
403 		}
404 	}
405 
406 	rmx = info->rti_rmx;
407 	if (rmx != NULL) {
408 		info->rti_mflags |= RTV_MTU;
409 		rmx->rmx_mtu = nh->nh_mtu;
410 	}
411 
412 	info->rti_flags = rt->rte_flags | nhop_get_rtflags(nh);
413 	info->rti_ifp = nh->nh_ifp;
414 	info->rti_ifa = nh->nh_ifa;
415 	if (flags & NHR_REF) {
416 		if_ref(info->rti_ifp);
417 		ifa_ref(info->rti_ifa);
418 	}
419 
420 	return (0);
421 }
422 
423 /*
424  * Lookups up route entry for @dst in RIB database for fib @fibnum.
425  * Exports entry data to @info using rt_exportinfo().
426  *
427  * If @flags contains NHR_REF, refcouting is performed on rt_ifp and rt_ifa.
428  * All references can be released later by calling rib_free_info().
429  *
430  * Returns 0 on success.
431  * Returns ENOENT for lookup failure, ENOMEM for export failure.
432  */
433 int
434 rib_lookup_info(uint32_t fibnum, const struct sockaddr *dst, uint32_t flags,
435     uint32_t flowid, struct rt_addrinfo *info)
436 {
437 	RIB_RLOCK_TRACKER;
438 	struct rib_head *rh;
439 	struct radix_node *rn;
440 	struct rtentry *rt;
441 	int error;
442 
443 	KASSERT((fibnum < rt_numfibs), ("rib_lookup_rte: bad fibnum"));
444 	rh = rt_tables_get_rnh(fibnum, dst->sa_family);
445 	if (rh == NULL)
446 		return (ENOENT);
447 
448 	RIB_RLOCK(rh);
449 	rn = rh->rnh_matchaddr(__DECONST(void *, dst), &rh->head);
450 	if (rn != NULL && ((rn->rn_flags & RNF_ROOT) == 0)) {
451 		rt = RNTORT(rn);
452 		/* Ensure route & ifp is UP */
453 		if (RT_LINK_IS_UP(rt->rt_nhop->nh_ifp)) {
454 			flags = (flags & NHR_REF) | NHR_COPY;
455 			error = rt_exportinfo(rt, info, flags);
456 			RIB_RUNLOCK(rh);
457 
458 			return (error);
459 		}
460 	}
461 	RIB_RUNLOCK(rh);
462 
463 	return (ENOENT);
464 }
465 
466 /*
467  * Releases all references acquired by rib_lookup_info() when
468  * called with NHR_REF flags.
469  */
470 void
471 rib_free_info(struct rt_addrinfo *info)
472 {
473 
474 	ifa_free(info->rti_ifa);
475 	if_rele(info->rti_ifp);
476 }
477 
478 /*
479  * Iterates over all existing fibs in system calling
480  *  @setwa_f function prior to traversing each fib.
481  *  Calls @wa_f function for each element in current fib.
482  * If af is not AF_UNSPEC, iterates over fibs in particular
483  * address family.
484  */
485 void
486 rt_foreach_fib_walk(int af, rt_setwarg_t *setwa_f, rt_walktree_f_t *wa_f,
487     void *arg)
488 {
489 	struct rib_head *rnh;
490 	uint32_t fibnum;
491 	int i;
492 
493 	for (fibnum = 0; fibnum < rt_numfibs; fibnum++) {
494 		/* Do we want some specific family? */
495 		if (af != AF_UNSPEC) {
496 			rnh = rt_tables_get_rnh(fibnum, af);
497 			if (rnh == NULL)
498 				continue;
499 			if (setwa_f != NULL)
500 				setwa_f(rnh, fibnum, af, arg);
501 
502 			RIB_WLOCK(rnh);
503 			rnh->rnh_walktree(&rnh->head, (walktree_f_t *)wa_f,arg);
504 			RIB_WUNLOCK(rnh);
505 			continue;
506 		}
507 
508 		for (i = 1; i <= AF_MAX; i++) {
509 			rnh = rt_tables_get_rnh(fibnum, i);
510 			if (rnh == NULL)
511 				continue;
512 			if (setwa_f != NULL)
513 				setwa_f(rnh, fibnum, i, arg);
514 
515 			RIB_WLOCK(rnh);
516 			rnh->rnh_walktree(&rnh->head, (walktree_f_t *)wa_f,arg);
517 			RIB_WUNLOCK(rnh);
518 		}
519 	}
520 }
521 
522 /*
523  * Iterates over all existing fibs in system and deletes each element
524  *  for which @filter_f function returns non-zero value.
525  * If @family is not AF_UNSPEC, iterates over fibs in particular
526  * address family.
527  */
528 void
529 rt_foreach_fib_walk_del(int family, rt_filter_f_t *filter_f, void *arg)
530 {
531 	u_int fibnum;
532 	int i, start, end;
533 
534 	for (fibnum = 0; fibnum < rt_numfibs; fibnum++) {
535 		/* Do we want some specific family? */
536 		if (family != AF_UNSPEC) {
537 			start = family;
538 			end = family;
539 		} else {
540 			start = 1;
541 			end = AF_MAX;
542 		}
543 
544 		for (i = start; i <= end; i++) {
545 			if (rt_tables_get_rnh(fibnum, i) == NULL)
546 				continue;
547 
548 			rib_walk_del(fibnum, i, filter_f, arg, 0);
549 		}
550 	}
551 }
552 
553 /*
554  * Delete Routes for a Network Interface
555  *
556  * Called for each routing entry via the rnh->rnh_walktree() call above
557  * to delete all route entries referencing a detaching network interface.
558  *
559  * Arguments:
560  *	rt	pointer to rtentry
561  *	nh	pointer to nhop
562  *	arg	argument passed to rnh->rnh_walktree() - detaching interface
563  *
564  * Returns:
565  *	0	successful
566  *	errno	failed - reason indicated
567  */
568 static int
569 rt_ifdelroute(const struct rtentry *rt, const struct nhop_object *nh, void *arg)
570 {
571 	struct ifnet	*ifp = arg;
572 
573 	if (nh->nh_ifp != ifp)
574 		return (0);
575 
576 	/*
577 	 * Protect (sorta) against walktree recursion problems
578 	 * with cloned routes
579 	 */
580 	if ((rt->rte_flags & RTF_UP) == 0)
581 		return (0);
582 
583 	return (1);
584 }
585 
586 /*
587  * Delete all remaining routes using this interface
588  * Unfortuneatly the only way to do this is to slog through
589  * the entire routing table looking for routes which point
590  * to this interface...oh well...
591  */
592 void
593 rt_flushifroutes_af(struct ifnet *ifp, int af)
594 {
595 	KASSERT((af >= 1 && af <= AF_MAX), ("%s: af %d not >= 1 and <= %d",
596 	    __func__, af, AF_MAX));
597 
598 	rt_foreach_fib_walk_del(af, rt_ifdelroute, ifp);
599 }
600 
601 void
602 rt_flushifroutes(struct ifnet *ifp)
603 {
604 
605 	rt_foreach_fib_walk_del(AF_UNSPEC, rt_ifdelroute, ifp);
606 }
607 
608 /*
609  * Look up rt_addrinfo for a specific fib.  Note that if rti_ifa is defined,
610  * it will be referenced so the caller must free it.
611  *
612  * Assume basic consistency checks are executed by callers:
613  * RTAX_DST exists, if RTF_GATEWAY is set, RTAX_GATEWAY exists as well.
614  */
615 int
616 rt_getifa_fib(struct rt_addrinfo *info, u_int fibnum)
617 {
618 	const struct sockaddr *dst, *gateway, *ifpaddr, *ifaaddr;
619 	struct epoch_tracker et;
620 	int needref, error, flags;
621 
622 	dst = info->rti_info[RTAX_DST];
623 	gateway = info->rti_info[RTAX_GATEWAY];
624 	ifpaddr = info->rti_info[RTAX_IFP];
625 	ifaaddr = info->rti_info[RTAX_IFA];
626 	flags = info->rti_flags;
627 
628 	/*
629 	 * ifp may be specified by sockaddr_dl
630 	 * when protocol address is ambiguous.
631 	 */
632 	error = 0;
633 	needref = (info->rti_ifa == NULL);
634 	NET_EPOCH_ENTER(et);
635 
636 	/* If we have interface specified by the ifindex in the address, use it */
637 	if (info->rti_ifp == NULL && ifpaddr != NULL &&
638 	    ifpaddr->sa_family == AF_LINK) {
639 	    const struct sockaddr_dl *sdl = (const struct sockaddr_dl *)ifpaddr;
640 	    if (sdl->sdl_index != 0)
641 		    info->rti_ifp = ifnet_byindex(sdl->sdl_index);
642 	}
643 	/*
644 	 * If we have source address specified, try to find it
645 	 * TODO: avoid enumerating all ifas on all interfaces.
646 	 */
647 	if (info->rti_ifa == NULL && ifaaddr != NULL)
648 		info->rti_ifa = ifa_ifwithaddr(ifaaddr);
649 	if (info->rti_ifa == NULL) {
650 		const struct sockaddr *sa;
651 
652 		/*
653 		 * Most common use case for the userland-supplied routes.
654 		 *
655 		 * Choose sockaddr to select ifa.
656 		 * -- if ifp is set --
657 		 * Order of preference:
658 		 * 1) IFA address
659 		 * 2) gateway address
660 		 *   Note: for interface routes link-level gateway address
661 		 *     is specified to indicate the interface index without
662 		 *     specifying RTF_GATEWAY. In this case, ignore gateway
663 		 *   Note: gateway AF may be different from dst AF. In this case,
664 		 *   ignore gateway
665 		 * 3) final destination.
666 		 * 4) if all of these fails, try to get at least link-level ifa.
667 		 * -- else --
668 		 * try to lookup gateway or dst in the routing table to get ifa
669 		 */
670 		if (info->rti_info[RTAX_IFA] != NULL)
671 			sa = info->rti_info[RTAX_IFA];
672 		else if ((info->rti_flags & RTF_GATEWAY) != 0 &&
673 		    gateway->sa_family == dst->sa_family)
674 			sa = gateway;
675 		else
676 			sa = dst;
677 		if (info->rti_ifp != NULL) {
678 			info->rti_ifa = ifaof_ifpforaddr(sa, info->rti_ifp);
679 			/* Case 4 */
680 			if (info->rti_ifa == NULL && gateway != NULL)
681 				info->rti_ifa = ifaof_ifpforaddr(gateway, info->rti_ifp);
682 		} else if (dst != NULL && gateway != NULL)
683 			info->rti_ifa = ifa_ifwithroute(flags, dst, gateway,
684 							fibnum);
685 		else if (sa != NULL)
686 			info->rti_ifa = ifa_ifwithroute(flags, sa, sa,
687 							fibnum);
688 	}
689 	if (needref && info->rti_ifa != NULL) {
690 		if (info->rti_ifp == NULL)
691 			info->rti_ifp = info->rti_ifa->ifa_ifp;
692 		ifa_ref(info->rti_ifa);
693 	} else
694 		error = ENETUNREACH;
695 	NET_EPOCH_EXIT(et);
696 	return (error);
697 }
698 
699 void
700 rt_updatemtu(struct ifnet *ifp)
701 {
702 	struct rib_head *rnh;
703 	int mtu;
704 	int i, j;
705 
706 	/*
707 	 * Try to update rt_mtu for all routes using this interface
708 	 * Unfortunately the only way to do this is to traverse all
709 	 * routing tables in all fibs/domains.
710 	 */
711 	for (i = 1; i <= AF_MAX; i++) {
712 		mtu = if_getmtu_family(ifp, i);
713 		for (j = 0; j < rt_numfibs; j++) {
714 			rnh = rt_tables_get_rnh(j, i);
715 			if (rnh == NULL)
716 				continue;
717 			nhops_update_ifmtu(rnh, ifp, mtu);
718 		}
719 	}
720 }
721 
722 
723 #if 0
724 int p_sockaddr(char *buf, int buflen, struct sockaddr *s);
725 int rt_print(char *buf, int buflen, struct rtentry *rt);
726 
727 int
728 p_sockaddr(char *buf, int buflen, struct sockaddr *s)
729 {
730 	void *paddr = NULL;
731 
732 	switch (s->sa_family) {
733 	case AF_INET:
734 		paddr = &((struct sockaddr_in *)s)->sin_addr;
735 		break;
736 	case AF_INET6:
737 		paddr = &((struct sockaddr_in6 *)s)->sin6_addr;
738 		break;
739 	}
740 
741 	if (paddr == NULL)
742 		return (0);
743 
744 	if (inet_ntop(s->sa_family, paddr, buf, buflen) == NULL)
745 		return (0);
746 
747 	return (strlen(buf));
748 }
749 
750 int
751 rt_print(char *buf, int buflen, struct rtentry *rt)
752 {
753 	struct sockaddr *addr, *mask;
754 	int i = 0;
755 
756 	addr = rt_key(rt);
757 	mask = rt_mask(rt);
758 
759 	i = p_sockaddr(buf, buflen, addr);
760 	if (!(rt->rt_flags & RTF_HOST)) {
761 		buf[i++] = '/';
762 		i += p_sockaddr(buf + i, buflen - i, mask);
763 	}
764 
765 	if (rt->rt_flags & RTF_GATEWAY) {
766 		buf[i++] = '>';
767 		i += p_sockaddr(buf + i, buflen - i, &rt->rt_nhop->gw_sa);
768 	}
769 
770 	return (i);
771 }
772 #endif
773 
774 #ifdef RADIX_MPATH
775 /*
776  * Deletes key for single-path routes, unlinks rtentry with
777  * gateway specified in @info from multi-path routes.
778  *
779  * Returnes unlinked entry. In case of failure, returns NULL
780  * and sets @perror to ESRCH.
781  */
782 struct radix_node *
783 rt_mpath_unlink(struct rib_head *rnh, struct rt_addrinfo *info,
784     struct rtentry *rto, int *perror)
785 {
786 	/*
787 	 * if we got multipath routes, we require users to specify
788 	 * a matching RTAX_GATEWAY.
789 	 */
790 	struct rtentry *rt; // *rto = NULL;
791 	struct radix_node *rn;
792 	struct sockaddr *gw;
793 
794 	gw = info->rti_info[RTAX_GATEWAY];
795 	rt = rt_mpath_matchgate(rto, gw);
796 	if (rt == NULL) {
797 		*perror = ESRCH;
798 		return (NULL);
799 	}
800 
801 	/*
802 	 * this is the first entry in the chain
803 	 */
804 	if (rto == rt) {
805 		rn = rn_mpath_next((struct radix_node *)rt);
806 		/*
807 		 * there is another entry, now it's active
808 		 */
809 		if (rn) {
810 			rto = RNTORT(rn);
811 			rto->rte_flags |= RTF_UP;
812 		} else if (rt->rte_flags & RTF_GATEWAY) {
813 			/*
814 			 * For gateway routes, we need to
815 			 * make sure that we we are deleting
816 			 * the correct gateway.
817 			 * rt_mpath_matchgate() does not
818 			 * check the case when there is only
819 			 * one route in the chain.
820 			 */
821 			if (gw &&
822 			    (rt->rt_nhop->gw_sa.sa_len != gw->sa_len ||
823 				memcmp(&rt->rt_nhop->gw_sa, gw, gw->sa_len))) {
824 				*perror = ESRCH;
825 				return (NULL);
826 			}
827 		}
828 
829 		/*
830 		 * use the normal delete code to remove
831 		 * the first entry
832 		 */
833 		rn = rnh->rnh_deladdr(info->rti_info[RTAX_DST],
834 					info->rti_info[RTAX_NETMASK],
835 					&rnh->head);
836 		if (rn != NULL) {
837 			*perror = 0;
838 		} else {
839 			*perror = ESRCH;
840 		}
841 		return (rn);
842 	}
843 
844 	/*
845 	 * if the entry is 2nd and on up
846 	 */
847 	if (rt_mpath_deldup(rto, rt) == 0)
848 		panic ("rtrequest1: rt_mpath_deldup");
849 	*perror = 0;
850 	rn = (struct radix_node *)rt;
851 	return (rn);
852 }
853 #endif
854 
855 void
856 rt_setmetrics(const struct rt_addrinfo *info, struct rtentry *rt)
857 {
858 
859 	if (info->rti_mflags & RTV_WEIGHT)
860 		rt->rt_weight = info->rti_rmx->rmx_weight;
861 	/* Kernel -> userland timebase conversion. */
862 	if (info->rti_mflags & RTV_EXPIRE)
863 		rt->rt_expire = info->rti_rmx->rmx_expire ?
864 		    info->rti_rmx->rmx_expire - time_second + time_uptime : 0;
865 }
866 
867 void
868 rt_maskedcopy(struct sockaddr *src, struct sockaddr *dst, struct sockaddr *netmask)
869 {
870 	u_char *cp1 = (u_char *)src;
871 	u_char *cp2 = (u_char *)dst;
872 	u_char *cp3 = (u_char *)netmask;
873 	u_char *cplim = cp2 + *cp3;
874 	u_char *cplim2 = cp2 + *cp1;
875 
876 	*cp2++ = *cp1++; *cp2++ = *cp1++; /* copies sa_len & sa_family */
877 	cp3 += 2;
878 	if (cplim > cplim2)
879 		cplim = cplim2;
880 	while (cp2 < cplim)
881 		*cp2++ = *cp1++ & *cp3++;
882 	if (cp2 < cplim2)
883 		bzero((caddr_t)cp2, (unsigned)(cplim2 - cp2));
884 }
885 
886 /*
887  * Set up a routing table entry, normally
888  * for an interface.
889  */
890 #define _SOCKADDR_TMPSIZE 128 /* Not too big.. kernel stack size is limited */
891 static inline  int
892 rtinit1(struct ifaddr *ifa, int cmd, int flags, int fibnum)
893 {
894 	RIB_RLOCK_TRACKER;
895 	struct epoch_tracker et;
896 	struct sockaddr *dst;
897 	struct sockaddr *netmask;
898 	struct rib_cmd_info rc;
899 	struct rt_addrinfo info;
900 	int error = 0;
901 	int startfib, endfib;
902 	char tempbuf[_SOCKADDR_TMPSIZE];
903 	int didwork = 0;
904 	int a_failure = 0;
905 	struct sockaddr_dl_short *sdl = NULL;
906 	struct rib_head *rnh;
907 
908 	if (flags & RTF_HOST) {
909 		dst = ifa->ifa_dstaddr;
910 		netmask = NULL;
911 	} else {
912 		dst = ifa->ifa_addr;
913 		netmask = ifa->ifa_netmask;
914 	}
915 	if (dst->sa_len == 0)
916 		return(EINVAL);
917 	switch (dst->sa_family) {
918 	case AF_INET6:
919 	case AF_INET:
920 		/* We support multiple FIBs. */
921 		break;
922 	default:
923 		fibnum = RT_DEFAULT_FIB;
924 		break;
925 	}
926 	if (fibnum == RT_ALL_FIBS) {
927 		if (V_rt_add_addr_allfibs == 0 && cmd == (int)RTM_ADD)
928 			startfib = endfib = ifa->ifa_ifp->if_fib;
929 		else {
930 			startfib = 0;
931 			endfib = rt_numfibs - 1;
932 		}
933 	} else {
934 		KASSERT((fibnum < rt_numfibs), ("rtinit1: bad fibnum"));
935 		startfib = fibnum;
936 		endfib = fibnum;
937 	}
938 
939 	/*
940 	 * If it's a delete, check that if it exists,
941 	 * it's on the correct interface or we might scrub
942 	 * a route to another ifa which would
943 	 * be confusing at best and possibly worse.
944 	 */
945 	if (cmd == RTM_DELETE) {
946 		/*
947 		 * It's a delete, so it should already exist..
948 		 * If it's a net, mask off the host bits
949 		 * (Assuming we have a mask)
950 		 * XXX this is kinda inet specific..
951 		 */
952 		if (netmask != NULL) {
953 			rt_maskedcopy(dst, (struct sockaddr *)tempbuf, netmask);
954 			dst = (struct sockaddr *)tempbuf;
955 		}
956 	} else if (cmd == RTM_ADD) {
957 		sdl = (struct sockaddr_dl_short *)tempbuf;
958 		bzero(sdl, sizeof(struct sockaddr_dl_short));
959 		sdl->sdl_family = AF_LINK;
960 		sdl->sdl_len = sizeof(struct sockaddr_dl_short);
961 		sdl->sdl_type = ifa->ifa_ifp->if_type;
962 		sdl->sdl_index = ifa->ifa_ifp->if_index;
963         }
964 	/*
965 	 * Now go through all the requested tables (fibs) and do the
966 	 * requested action. Realistically, this will either be fib 0
967 	 * for protocols that don't do multiple tables or all the
968 	 * tables for those that do.
969 	 */
970 	for ( fibnum = startfib; fibnum <= endfib; fibnum++) {
971 		if (cmd == RTM_DELETE) {
972 			struct radix_node *rn;
973 			/*
974 			 * Look up an rtentry that is in the routing tree and
975 			 * contains the correct info.
976 			 */
977 			rnh = rt_tables_get_rnh(fibnum, dst->sa_family);
978 			if (rnh == NULL)
979 				/* this table doesn't exist but others might */
980 				continue;
981 			RIB_RLOCK(rnh);
982 			rn = rnh->rnh_lookup(dst, netmask, &rnh->head);
983 #ifdef RADIX_MPATH
984 			if (rt_mpath_capable(rnh)) {
985 
986 				if (rn == NULL)
987 					error = ESRCH;
988 				else {
989 					struct rtentry *rt = RNTORT(rn);
990 					/*
991 					 * for interface route the gateway
992 					 * gateway is sockaddr_dl, so
993 					 * rt_mpath_matchgate must use the
994 					 * interface address
995 					 */
996 					rt = rt_mpath_matchgate(rt,
997 					    ifa->ifa_addr);
998 					if (rt == NULL)
999 						error = ESRCH;
1000 				}
1001 			}
1002 #endif
1003 			error = (rn == NULL ||
1004 			    (rn->rn_flags & RNF_ROOT) ||
1005 			    RNTORT(rn)->rt_nhop->nh_ifa != ifa);
1006 			RIB_RUNLOCK(rnh);
1007 			if (error) {
1008 				/* this is only an error if bad on ALL tables */
1009 				continue;
1010 			}
1011 		}
1012 		/*
1013 		 * Do the actual request
1014 		 */
1015 		bzero((caddr_t)&info, sizeof(info));
1016 		info.rti_ifa = ifa;
1017 		info.rti_flags = flags |
1018 		    (ifa->ifa_flags & ~IFA_RTSELF) | RTF_PINNED;
1019 		info.rti_info[RTAX_DST] = dst;
1020 		/*
1021 		 * doing this for compatibility reasons
1022 		 */
1023 		if (cmd == RTM_ADD)
1024 			info.rti_info[RTAX_GATEWAY] = (struct sockaddr *)sdl;
1025 		else
1026 			info.rti_info[RTAX_GATEWAY] = ifa->ifa_addr;
1027 		info.rti_info[RTAX_NETMASK] = netmask;
1028 		NET_EPOCH_ENTER(et);
1029 		error = rib_action(fibnum, cmd, &info, &rc);
1030 		if (error == 0 && rc.rc_rt != NULL) {
1031 			/*
1032 			 * notify any listening routing agents of the change
1033 			 */
1034 
1035 			/* TODO: interface routes/aliases */
1036 			rt_newaddrmsg_fib(cmd, ifa, rc.rc_rt, fibnum);
1037 			didwork = 1;
1038 		}
1039 		NET_EPOCH_EXIT(et);
1040 		if (error)
1041 			a_failure = error;
1042 	}
1043 	if (cmd == RTM_DELETE) {
1044 		if (didwork) {
1045 			error = 0;
1046 		} else {
1047 			/* we only give an error if it wasn't in any table */
1048 			error = ((flags & RTF_HOST) ?
1049 			    EHOSTUNREACH : ENETUNREACH);
1050 		}
1051 	} else {
1052 		if (a_failure) {
1053 			/* return an error if any of them failed */
1054 			error = a_failure;
1055 		}
1056 	}
1057 	return (error);
1058 }
1059 
1060 /*
1061  * Set up a routing table entry, normally
1062  * for an interface.
1063  */
1064 int
1065 rtinit(struct ifaddr *ifa, int cmd, int flags)
1066 {
1067 	struct sockaddr *dst;
1068 	int fib = RT_DEFAULT_FIB;
1069 
1070 	if (flags & RTF_HOST) {
1071 		dst = ifa->ifa_dstaddr;
1072 	} else {
1073 		dst = ifa->ifa_addr;
1074 	}
1075 
1076 	switch (dst->sa_family) {
1077 	case AF_INET6:
1078 	case AF_INET:
1079 		/* We do support multiple FIBs. */
1080 		fib = RT_ALL_FIBS;
1081 		break;
1082 	}
1083 	return (rtinit1(ifa, cmd, flags, fib));
1084 }
1085 
1086 /*
1087  * Announce interface address arrival/withdraw
1088  * Returns 0 on success.
1089  */
1090 int
1091 rt_addrmsg(int cmd, struct ifaddr *ifa, int fibnum)
1092 {
1093 
1094 	KASSERT(cmd == RTM_ADD || cmd == RTM_DELETE,
1095 	    ("unexpected cmd %d", cmd));
1096 	KASSERT(fibnum == RT_ALL_FIBS || (fibnum >= 0 && fibnum < rt_numfibs),
1097 	    ("%s: fib out of range 0 <=%d<%d", __func__, fibnum, rt_numfibs));
1098 
1099 	EVENTHANDLER_DIRECT_INVOKE(rt_addrmsg, ifa, cmd);
1100 	return (rtsock_addrmsg(cmd, ifa, fibnum));
1101 }
1102 
1103 /*
1104  * Announce kernel-originated route addition/removal to rtsock based on @rt data.
1105  * cmd: RTM_ cmd
1106  * @rt: valid rtentry
1107  * @ifp: target route interface
1108  * @fibnum: fib id or RT_ALL_FIBS
1109  *
1110  * Returns 0 on success.
1111  */
1112 int
1113 rt_routemsg(int cmd, struct rtentry *rt, struct ifnet *ifp, int rti_addrs,
1114     int fibnum)
1115 {
1116 
1117 	KASSERT(cmd == RTM_ADD || cmd == RTM_DELETE,
1118 	    ("unexpected cmd %d", cmd));
1119 
1120 	KASSERT(fibnum == RT_ALL_FIBS || (fibnum >= 0 && fibnum < rt_numfibs),
1121 	    ("%s: fib out of range 0 <=%d<%d", __func__, fibnum, rt_numfibs));
1122 
1123 	KASSERT(rt_key(rt) != NULL, (":%s: rt_key must be supplied", __func__));
1124 
1125 	return (rtsock_routemsg(cmd, rt, ifp, 0, fibnum));
1126 }
1127 
1128 /*
1129  * Announce kernel-originated route addition/removal to rtsock based on @rt data.
1130  * cmd: RTM_ cmd
1131  * @info: addrinfo structure with valid data.
1132  * @fibnum: fib id or RT_ALL_FIBS
1133  *
1134  * Returns 0 on success.
1135  */
1136 int
1137 rt_routemsg_info(int cmd, struct rt_addrinfo *info, int fibnum)
1138 {
1139 
1140 	KASSERT(cmd == RTM_ADD || cmd == RTM_DELETE || cmd == RTM_CHANGE,
1141 	    ("unexpected cmd %d", cmd));
1142 
1143 	KASSERT(fibnum == RT_ALL_FIBS || (fibnum >= 0 && fibnum < rt_numfibs),
1144 	    ("%s: fib out of range 0 <=%d<%d", __func__, fibnum, rt_numfibs));
1145 
1146 	KASSERT(info->rti_info[RTAX_DST] != NULL, (":%s: RTAX_DST must be supplied", __func__));
1147 
1148 	return (rtsock_routemsg_info(cmd, info, fibnum));
1149 }
1150 
1151 
1152 /*
1153  * This is called to generate messages from the routing socket
1154  * indicating a network interface has had addresses associated with it.
1155  */
1156 void
1157 rt_newaddrmsg_fib(int cmd, struct ifaddr *ifa, struct rtentry *rt, int fibnum)
1158 {
1159 
1160 	KASSERT(cmd == RTM_ADD || cmd == RTM_DELETE,
1161 		("unexpected cmd %u", cmd));
1162 	KASSERT(fibnum == RT_ALL_FIBS || (fibnum >= 0 && fibnum < rt_numfibs),
1163 	    ("%s: fib out of range 0 <=%d<%d", __func__, fibnum, rt_numfibs));
1164 
1165 	if (cmd == RTM_ADD) {
1166 		rt_addrmsg(cmd, ifa, fibnum);
1167 		if (rt != NULL)
1168 			rt_routemsg(cmd, rt, ifa->ifa_ifp, 0, fibnum);
1169 	} else {
1170 		if (rt != NULL)
1171 			rt_routemsg(cmd, rt, ifa->ifa_ifp, 0, fibnum);
1172 		rt_addrmsg(cmd, ifa, fibnum);
1173 	}
1174 }
1175 
1176