xref: /dflybsd-src/sys/net/route.c (revision 086b156cdf18f6fbb398ffa6f62a25f4f158853e)
1 /*
2  * Copyright (c) 2004, 2005 The DragonFly Project.  All rights reserved.
3  *
4  * This code is derived from software contributed to The DragonFly Project
5  * by Jeffrey M. Hsu.
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 DragonFly Project nor the names of its
16  *    contributors may be used to endorse or promote products derived
17  *    from this software without specific, prior written permission.
18  *
19  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
20  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
21  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
22  * FOR A PARTICULAR PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE
23  * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
24  * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
25  * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
26  * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
27  * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
28  * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
29  * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
30  * SUCH DAMAGE.
31  */
32 
33 /*
34  * Copyright (c) 1980, 1986, 1991, 1993
35  *	The Regents of the University of California.  All rights reserved.
36  *
37  * Redistribution and use in source and binary forms, with or without
38  * modification, are permitted provided that the following conditions
39  * are met:
40  * 1. Redistributions of source code must retain the above copyright
41  *    notice, this list of conditions and the following disclaimer.
42  * 2. Redistributions in binary form must reproduce the above copyright
43  *    notice, this list of conditions and the following disclaimer in the
44  *    documentation and/or other materials provided with the distribution.
45  * 3. Neither the name of the University nor the names of its contributors
46  *    may be used to endorse or promote products derived from this software
47  *    without specific prior written permission.
48  *
49  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
50  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
51  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
52  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
53  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
54  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
55  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
56  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
57  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
58  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
59  * SUCH DAMAGE.
60  *
61  *	@(#)route.c	8.3 (Berkeley) 1/9/95
62  * $FreeBSD: src/sys/net/route.c,v 1.59.2.10 2003/01/17 08:04:00 ru Exp $
63  */
64 
65 #include "opt_inet.h"
66 #include "opt_mpls.h"
67 
68 #include <sys/param.h>
69 #include <sys/systm.h>
70 #include <sys/malloc.h>
71 #include <sys/mbuf.h>
72 #include <sys/socket.h>
73 #include <sys/domain.h>
74 #include <sys/kernel.h>
75 #include <sys/sysctl.h>
76 #include <sys/globaldata.h>
77 #include <sys/thread.h>
78 
79 #include <net/if.h>
80 #include <net/if_var.h>
81 #include <net/route.h>
82 #include <net/netisr.h>
83 
84 #include <netinet/in.h>
85 #include <net/ip_mroute/ip_mroute.h>
86 
87 #include <sys/thread2.h>
88 #include <sys/msgport2.h>
89 #include <net/netmsg2.h>
90 #include <net/netisr2.h>
91 
92 #ifdef MPLS
93 #include <netproto/mpls/mpls.h>
94 #endif
95 
96 static struct rtstatistics rtstatistics_percpu[MAXCPU] __cachealign;
97 #define rtstat	rtstatistics_percpu[mycpuid]
98 
99 struct radix_node_head *rt_tables[MAXCPU][AF_MAX+1];
100 
101 static void	rt_maskedcopy (struct sockaddr *, struct sockaddr *,
102 			       struct sockaddr *);
103 static void rtable_init(void);
104 static void rtinit_rtrequest_callback(int, int, struct rt_addrinfo *,
105 				      struct rtentry *, void *);
106 
107 static void rtredirect_msghandler(netmsg_t msg);
108 static void rtrequest1_msghandler(netmsg_t msg);
109 static void rtsearch_msghandler(netmsg_t msg);
110 static void rtmask_add_msghandler(netmsg_t msg);
111 
112 static int rt_setshims(struct rtentry *, struct sockaddr **);
113 
114 SYSCTL_NODE(_net, OID_AUTO, route, CTLFLAG_RW, 0, "Routing");
115 
116 #ifdef ROUTE_DEBUG
117 static int route_debug = 1;
118 SYSCTL_INT(_net_route, OID_AUTO, route_debug, CTLFLAG_RW,
119 	   &route_debug, 0, "");
120 #endif
121 
122 u_long route_kmalloc_limit = 0;
123 TUNABLE_ULONG("net.route.kmalloc_limit", &route_kmalloc_limit);
124 
125 /*
126  * Initialize the route table(s) for protocol domains and
127  * create a helper thread which will be responsible for updating
128  * route table entries on each cpu.
129  */
130 void
131 route_init(void)
132 {
133 	int cpu;
134 
135 	if (route_kmalloc_limit)
136 		kmalloc_raise_limit(M_RTABLE, route_kmalloc_limit);
137 
138 	for (cpu = 0; cpu < netisr_ncpus; ++cpu)
139 		bzero(&rtstatistics_percpu[cpu], sizeof(struct rtstatistics));
140 	rn_init();      /* initialize all zeroes, all ones, mask table */
141 	rtable_init();	/* call dom_rtattach() on each cpu */
142 }
143 
144 static void
145 rtable_init_oncpu(netmsg_t msg)
146 {
147 	struct domain *dom;
148 	int cpu = mycpuid;
149 
150 	ASSERT_NETISR_NCPUS(cpu);
151 
152 	SLIST_FOREACH(dom, &domains, dom_next) {
153 		if (dom->dom_rtattach) {
154 			dom->dom_rtattach(
155 				(void **)&rt_tables[cpu][dom->dom_family],
156 				dom->dom_rtoffset);
157 		}
158 	}
159 	netisr_forwardmsg(&msg->base, cpu + 1);
160 }
161 
162 static void
163 rtable_init(void)
164 {
165 	struct netmsg_base msg;
166 
167 	netmsg_init(&msg, NULL, &curthread->td_msgport, 0, rtable_init_oncpu);
168 	netisr_domsg_global(&msg);
169 }
170 
171 /*
172  * Routing statistics.
173  */
174 static int
175 sysctl_rtstatistics(SYSCTL_HANDLER_ARGS)
176 {
177 	int cpu, error = 0;
178 
179 	for (cpu = 0; cpu < netisr_ncpus; ++cpu) {
180 		if ((error = SYSCTL_OUT(req, &rtstatistics_percpu[cpu],
181 					sizeof(struct rtstatistics))))
182 				break;
183 		if ((error = SYSCTL_IN(req, &rtstatistics_percpu[cpu],
184 					sizeof(struct rtstatistics))))
185 				break;
186 	}
187 
188 	return (error);
189 }
190 SYSCTL_PROC(_net_route, OID_AUTO, stats, (CTLTYPE_OPAQUE|CTLFLAG_RW),
191 	0, 0, sysctl_rtstatistics, "S,rtstatistics", "Routing statistics");
192 
193 /*
194  * Packet routing routines.
195  */
196 
197 /*
198  * Look up and fill in the "ro_rt" rtentry field in a route structure given
199  * an address in the "ro_dst" field.  Always send a report on a miss and
200  * always clone routes.
201  */
202 void
203 rtalloc(struct route *ro)
204 {
205 	rtalloc_ign(ro, 0UL);
206 }
207 
208 /*
209  * Look up and fill in the "ro_rt" rtentry field in a route structure given
210  * an address in the "ro_dst" field.  Always send a report on a miss and
211  * optionally clone routes when RTF_CLONING or RTF_PRCLONING are not being
212  * ignored.
213  */
214 void
215 rtalloc_ign(struct route *ro, u_long ignoreflags)
216 {
217 	if (ro->ro_rt != NULL) {
218 		if (ro->ro_rt->rt_ifp != NULL && ro->ro_rt->rt_flags & RTF_UP)
219 			return;
220 		rtfree(ro->ro_rt);
221 		ro->ro_rt = NULL;
222 	}
223 	ro->ro_rt = _rtlookup(&ro->ro_dst, ignoreflags);
224 }
225 
226 /*
227  * Look up the route that matches the given "dst" address.
228  *
229  * Route lookup can have the side-effect of creating and returning
230  * a cloned route instead when "dst" matches a cloning route and the
231  * RTF_CLONING and RTF_PRCLONING flags are not being ignored.
232  *
233  * Any route returned has its reference count incremented.
234  */
235 struct rtentry *
236 _rtlookup(struct sockaddr *dst, u_long ignore)
237 {
238 	struct radix_node_head *rnh = rt_tables[mycpuid][dst->sa_family];
239 	struct rtentry *rt;
240 
241 	ASSERT_NETISR_NCPUS(mycpuid);
242 
243 	if (rnh == NULL)
244 		goto unreach;
245 
246 	/*
247 	 * Look up route in the radix tree.
248 	 */
249 	rt = (struct rtentry *) rnh->rnh_matchaddr((char *)dst, rnh);
250 	if (rt == NULL)
251 		goto unreach;
252 
253 	/*
254 	 * Handle cloning routes.
255 	 */
256 	if ((rt->rt_flags & ~ignore & (RTF_CLONING | RTF_PRCLONING)) != 0) {
257 		struct rtentry *clonedroute;
258 		int error;
259 
260 		clonedroute = rt;	/* copy in/copy out parameter */
261 		error = rtrequest(RTM_RESOLVE, dst, NULL, NULL, 0,
262 				  &clonedroute);	/* clone the route */
263 		if (error != 0) {	/* cloning failed */
264 			rt_dstmsg(RTM_MISS, dst, error);
265 			rt->rt_refcnt++;
266 			return (rt);	/* return the uncloned route */
267 		}
268 		if (clonedroute->rt_flags & RTF_XRESOLVE)
269 			rt_dstmsg(RTM_RESOLVE, dst, 0);
270 		return (clonedroute);	/* return cloned route */
271 	}
272 
273 	/*
274 	 * Increment the reference count of the matched route and return.
275 	 */
276 	rt->rt_refcnt++;
277 	return (rt);
278 
279 unreach:
280 	rtstat.rts_unreach++;
281 	rt_dstmsg(RTM_MISS, dst, 0);
282 	return (NULL);
283 }
284 
285 void
286 rtfree(struct rtentry *rt)
287 {
288 
289 	ASSERT_NETISR_NCPUS(rt->rt_cpuid);
290 	KASSERT(rt->rt_refcnt > 0, ("rtfree: rt_refcnt %ld", rt->rt_refcnt));
291 
292 	--rt->rt_refcnt;
293 	if (rt->rt_refcnt == 0) {
294 		struct radix_node_head *rnh =
295 		    rt_tables[mycpuid][rt_key(rt)->sa_family];
296 
297 		if (rnh->rnh_close)
298 			rnh->rnh_close((struct radix_node *)rt, rnh);
299 		if (!(rt->rt_flags & RTF_UP)) {
300 			/* deallocate route */
301 			if (rt->rt_ifa != NULL)
302 				IFAFREE(rt->rt_ifa);
303 			if (rt->rt_parent != NULL)
304 				RTFREE(rt->rt_parent);	/* recursive call! */
305 			Free(rt_key(rt));
306 			Free(rt);
307 		}
308 	}
309 }
310 
311 static void
312 rtfree_async_dispatch(netmsg_t msg)
313 {
314 	struct rtentry *rt = msg->lmsg.u.ms_resultp;
315 
316 	rtfree(rt);
317 	netisr_replymsg(&msg->base, 0);
318 }
319 
320 void
321 rtfree_async(struct rtentry *rt)
322 {
323 	struct netmsg_base *msg;
324 
325 	if (IN_NETISR_NCPUS(rt->rt_cpuid)) {
326 		rtfree(rt);
327 		return;
328 	}
329 
330 	KASSERT(rt->rt_refcnt > 0,
331 	    ("rtfree_async: rt_refcnt %ld", rt->rt_refcnt));
332 
333 	msg = kmalloc(sizeof(*msg), M_LWKTMSG, M_INTWAIT);
334 	netmsg_init(msg, NULL, &netisr_afree_rport, 0, rtfree_async_dispatch);
335 	msg->lmsg.u.ms_resultp = rt;
336 
337 	netisr_sendmsg(msg, rt->rt_cpuid);
338 }
339 
340 int
341 rtredirect_oncpu(struct sockaddr *dst, struct sockaddr *gateway,
342 		 struct sockaddr *netmask, int flags, struct sockaddr *src)
343 {
344 	struct rtentry *rt = NULL;
345 	struct rt_addrinfo rtinfo;
346 	struct ifaddr *ifa;
347 	u_long *stat = NULL;
348 	int error;
349 
350 	ASSERT_NETISR_NCPUS(mycpuid);
351 
352 	/* verify the gateway is directly reachable */
353 	if ((ifa = ifa_ifwithnet(gateway)) == NULL) {
354 		error = ENETUNREACH;
355 		goto out;
356 	}
357 
358 	/*
359 	 * If the redirect isn't from our current router for this destination,
360 	 * it's either old or wrong.
361 	 */
362 	if (!(flags & RTF_DONE) &&		/* XXX JH */
363 	    (rt = rtpurelookup(dst)) != NULL &&
364 	    (!sa_equal(src, rt->rt_gateway) || rt->rt_ifa != ifa)) {
365 		error = EINVAL;
366 		goto done;
367 	}
368 
369 	/*
370 	 * If it redirects us to ourselves, we have a routing loop,
371 	 * perhaps as a result of an interface going down recently.
372 	 */
373 	if (ifa_ifwithaddr(gateway)) {
374 		error = EHOSTUNREACH;
375 		goto done;
376 	}
377 
378 	/*
379 	 * Create a new entry if the lookup failed or if we got back
380 	 * a wildcard entry for the default route.  This is necessary
381 	 * for hosts which use routing redirects generated by smart
382 	 * gateways to dynamically build the routing tables.
383 	 */
384 	if (rt == NULL)
385 		goto create;
386 	if ((rt_mask(rt) != NULL && rt_mask(rt)->sa_len < 2)) {
387 		rtfree(rt);
388 		goto create;
389 	}
390 
391 	/* Ignore redirects for directly connected hosts. */
392 	if (!(rt->rt_flags & RTF_GATEWAY)) {
393 		error = EHOSTUNREACH;
394 		goto done;
395 	}
396 
397 	if (!(rt->rt_flags & RTF_HOST) && (flags & RTF_HOST)) {
398 		/*
399 		 * Changing from a network route to a host route.
400 		 * Create a new host route rather than smashing the
401 		 * network route.
402 		 */
403 create:
404 		flags |=  RTF_GATEWAY | RTF_DYNAMIC;
405 		bzero(&rtinfo, sizeof(struct rt_addrinfo));
406 		rtinfo.rti_info[RTAX_DST] = dst;
407 		rtinfo.rti_info[RTAX_GATEWAY] = gateway;
408 		rtinfo.rti_info[RTAX_NETMASK] = netmask;
409 		rtinfo.rti_flags = flags;
410 		rtinfo.rti_ifa = ifa;
411 		rt = NULL;	/* copy-in/copy-out parameter */
412 		error = rtrequest1(RTM_ADD, &rtinfo, &rt);
413 		if (rt != NULL)
414 			flags = rt->rt_flags;
415 		stat = &rtstat.rts_dynamic;
416 	} else {
417 		/*
418 		 * Smash the current notion of the gateway to this destination.
419 		 * Should check about netmask!!!
420 		 */
421 		rt->rt_flags |= RTF_MODIFIED;
422 		flags |= RTF_MODIFIED;
423 
424 		/* We only need to report rtmsg on CPU0 */
425 		rt_setgate(rt, rt_key(rt), gateway);
426 		if (mycpuid == 0)
427 			rt_rtmsg(RTM_CHANGE, rt, rt->rt_ifp, 0);
428 		error = 0;
429 		stat = &rtstat.rts_newgateway;
430 	}
431 
432 done:
433 	if (rt != NULL)
434 		rtfree(rt);
435 out:
436 	if (error != 0)
437 		rtstat.rts_badredirect++;
438 	else if (stat != NULL)
439 		(*stat)++;
440 
441 	return error;
442 }
443 
444 struct netmsg_rtredirect {
445 	struct netmsg_base base;
446 	struct sockaddr *dst;
447 	struct sockaddr *gateway;
448 	struct sockaddr *netmask;
449 	int		flags;
450 	struct sockaddr *src;
451 };
452 
453 /*
454  * Force a routing table entry to the specified
455  * destination to go through the given gateway.
456  * Normally called as a result of a routing redirect
457  * message from the network layer.
458  */
459 void
460 rtredirect(struct sockaddr *dst, struct sockaddr *gateway,
461 	   struct sockaddr *netmask, int flags, struct sockaddr *src)
462 {
463 	struct rt_addrinfo rtinfo;
464 	int error;
465 	struct netmsg_rtredirect msg;
466 
467 	netmsg_init(&msg.base, NULL, &curthread->td_msgport,
468 		    0, rtredirect_msghandler);
469 	msg.dst = dst;
470 	msg.gateway = gateway;
471 	msg.netmask = netmask;
472 	msg.flags = flags;
473 	msg.src = src;
474 	error = netisr_domsg_global(&msg.base);
475 
476 	bzero(&rtinfo, sizeof(struct rt_addrinfo));
477 	rtinfo.rti_info[RTAX_DST] = dst;
478 	rtinfo.rti_info[RTAX_GATEWAY] = gateway;
479 	rtinfo.rti_info[RTAX_NETMASK] = netmask;
480 	rtinfo.rti_info[RTAX_AUTHOR] = src;
481 	rt_missmsg(RTM_REDIRECT, &rtinfo, flags, error);
482 }
483 
484 static void
485 rtredirect_msghandler(netmsg_t msg)
486 {
487 	struct netmsg_rtredirect *rmsg = (void *)msg;
488 
489 	rtredirect_oncpu(rmsg->dst, rmsg->gateway, rmsg->netmask,
490 			 rmsg->flags, rmsg->src);
491 	netisr_forwardmsg(&msg->base, mycpuid + 1);
492 }
493 
494 /*
495 * Routing table ioctl interface.
496 */
497 int
498 rtioctl(u_long req, caddr_t data, struct ucred *cred)
499 {
500 #ifdef INET
501 	/* Multicast goop, grrr... */
502 	return mrt_ioctl ? mrt_ioctl(req, data) : EOPNOTSUPP;
503 #else
504 	return ENXIO;
505 #endif
506 }
507 
508 struct ifaddr *
509 ifa_ifwithroute(int flags, struct sockaddr *dst, struct sockaddr *gateway)
510 {
511 	struct ifaddr *ifa;
512 
513 	if (!(flags & RTF_GATEWAY)) {
514 		/*
515 		 * If we are adding a route to an interface,
516 		 * and the interface is a point-to-point link,
517 		 * we should search for the destination
518 		 * as our clue to the interface.  Otherwise
519 		 * we can use the local address.
520 		 */
521 		ifa = NULL;
522 		if (flags & RTF_HOST) {
523 			ifa = ifa_ifwithdstaddr(dst);
524 		}
525 		if (ifa == NULL)
526 			ifa = ifa_ifwithaddr(gateway);
527 	} else {
528 		/*
529 		 * If we are adding a route to a remote net
530 		 * or host, the gateway may still be on the
531 		 * other end of a pt to pt link.
532 		 */
533 		ifa = ifa_ifwithdstaddr(gateway);
534 	}
535 	if (ifa == NULL)
536 		ifa = ifa_ifwithnet(gateway);
537 	if (ifa == NULL) {
538 		struct rtentry *rt;
539 
540 		rt = rtpurelookup(gateway);
541 		if (rt == NULL)
542 			return (NULL);
543 		rt->rt_refcnt--;
544 		if ((ifa = rt->rt_ifa) == NULL)
545 			return (NULL);
546 	}
547 	if (ifa->ifa_addr->sa_family != dst->sa_family) {
548 		struct ifaddr *oldifa = ifa;
549 
550 		ifa = ifaof_ifpforaddr(dst, ifa->ifa_ifp);
551 		if (ifa == NULL)
552 			ifa = oldifa;
553 	}
554 	return (ifa);
555 }
556 
557 static int rt_fixdelete (struct radix_node *, void *);
558 static int rt_fixchange (struct radix_node *, void *);
559 
560 struct rtfc_arg {
561 	struct rtentry *rt0;
562 	struct radix_node_head *rnh;
563 };
564 
565 /*
566  * Set rtinfo->rti_ifa and rtinfo->rti_ifp.
567  */
568 int
569 rt_getifa(struct rt_addrinfo *rtinfo)
570 {
571 	struct sockaddr *gateway = rtinfo->rti_info[RTAX_GATEWAY];
572 	struct sockaddr *dst = rtinfo->rti_info[RTAX_DST];
573 	struct sockaddr *ifaaddr = rtinfo->rti_info[RTAX_IFA];
574 	int flags = rtinfo->rti_flags;
575 
576 	/*
577 	 * ifp may be specified by sockaddr_dl
578 	 * when protocol address is ambiguous.
579 	 */
580 	if (rtinfo->rti_ifp == NULL) {
581 		struct sockaddr *ifpaddr;
582 
583 		ifpaddr = rtinfo->rti_info[RTAX_IFP];
584 		if (ifpaddr != NULL && ifpaddr->sa_family == AF_LINK) {
585 			struct ifaddr *ifa;
586 
587 			ifa = ifa_ifwithnet(ifpaddr);
588 			if (ifa != NULL)
589 				rtinfo->rti_ifp = ifa->ifa_ifp;
590 		}
591 	}
592 
593 	if (rtinfo->rti_ifa == NULL && ifaaddr != NULL)
594 		rtinfo->rti_ifa = ifa_ifwithaddr(ifaaddr);
595 	if (rtinfo->rti_ifa == NULL) {
596 		struct sockaddr *sa;
597 
598 		sa = ifaaddr != NULL ? ifaaddr :
599 		    (gateway != NULL ? gateway : dst);
600 		if (sa != NULL && rtinfo->rti_ifp != NULL)
601 			rtinfo->rti_ifa = ifaof_ifpforaddr(sa, rtinfo->rti_ifp);
602 		else if (dst != NULL && gateway != NULL)
603 			rtinfo->rti_ifa = ifa_ifwithroute(flags, dst, gateway);
604 		else if (sa != NULL)
605 			rtinfo->rti_ifa = ifa_ifwithroute(flags, sa, sa);
606 	}
607 	if (rtinfo->rti_ifa == NULL)
608 		return (ENETUNREACH);
609 
610 	if (rtinfo->rti_ifp == NULL)
611 		rtinfo->rti_ifp = rtinfo->rti_ifa->ifa_ifp;
612 	return (0);
613 }
614 
615 /*
616  * Do appropriate manipulations of a routing tree given
617  * all the bits of info needed
618  */
619 int
620 rtrequest(
621 	int req,
622 	struct sockaddr *dst,
623 	struct sockaddr *gateway,
624 	struct sockaddr *netmask,
625 	int flags,
626 	struct rtentry **ret_nrt)
627 {
628 	struct rt_addrinfo rtinfo;
629 
630 	bzero(&rtinfo, sizeof(struct rt_addrinfo));
631 	rtinfo.rti_info[RTAX_DST] = dst;
632 	rtinfo.rti_info[RTAX_GATEWAY] = gateway;
633 	rtinfo.rti_info[RTAX_NETMASK] = netmask;
634 	rtinfo.rti_flags = flags;
635 	return rtrequest1(req, &rtinfo, ret_nrt);
636 }
637 
638 int
639 rtrequest_global(
640 	int req,
641 	struct sockaddr *dst,
642 	struct sockaddr *gateway,
643 	struct sockaddr *netmask,
644 	int flags)
645 {
646 	struct rt_addrinfo rtinfo;
647 
648 	bzero(&rtinfo, sizeof(struct rt_addrinfo));
649 	rtinfo.rti_info[RTAX_DST] = dst;
650 	rtinfo.rti_info[RTAX_GATEWAY] = gateway;
651 	rtinfo.rti_info[RTAX_NETMASK] = netmask;
652 	rtinfo.rti_flags = flags;
653 	return rtrequest1_global(req, &rtinfo, NULL, NULL, RTREQ_PRIO_NORM);
654 }
655 
656 struct netmsg_rtq {
657 	struct netmsg_base	base;
658 	int			req;
659 	struct rt_addrinfo	*rtinfo;
660 	rtrequest1_callback_func_t callback;
661 	void			*arg;
662 };
663 
664 int
665 rtrequest1_global(int req, struct rt_addrinfo *rtinfo,
666     rtrequest1_callback_func_t callback, void *arg, boolean_t req_prio)
667 {
668 	struct netmsg_rtq msg;
669 	int flags = 0;
670 
671 	if (req_prio)
672 		flags = MSGF_PRIORITY;
673 	netmsg_init(&msg.base, NULL, &curthread->td_msgport, flags,
674 	    rtrequest1_msghandler);
675 	msg.base.lmsg.ms_error = -1;
676 	msg.req = req;
677 	msg.rtinfo = rtinfo;
678 	msg.callback = callback;
679 	msg.arg = arg;
680 	return (netisr_domsg_global(&msg.base));
681 }
682 
683 /*
684  * Handle a route table request on the current cpu.  Since the route table's
685  * are supposed to be identical on each cpu, an error occuring later in the
686  * message chain is considered system-fatal.
687  */
688 static void
689 rtrequest1_msghandler(netmsg_t msg)
690 {
691 	struct netmsg_rtq *rmsg = (void *)msg;
692 	struct rt_addrinfo rtinfo;
693 	struct rtentry *rt = NULL;
694 	int error;
695 
696 	/*
697 	 * Copy the rtinfo.  We need to make sure that the original
698 	 * rtinfo, which is setup by the caller, in the netmsg will
699 	 * _not_ be changed; else the next CPU on the netmsg forwarding
700 	 * path will see a different rtinfo than what this CPU has seen.
701 	 */
702 	rtinfo = *rmsg->rtinfo;
703 
704 	error = rtrequest1(rmsg->req, &rtinfo, &rt);
705 	if (rt)
706 		--rt->rt_refcnt;
707 	if (rmsg->callback)
708 		rmsg->callback(rmsg->req, error, &rtinfo, rt, rmsg->arg);
709 
710 	/*
711 	 * RTM_DELETE's are propogated even if an error occurs, since a
712 	 * cloned route might be undergoing deletion and cloned routes
713 	 * are not necessarily replicated.  An overall error is returned
714 	 * only if no cpus have the route in question.
715 	 */
716 	if (rmsg->base.lmsg.ms_error < 0 || error == 0)
717 		rmsg->base.lmsg.ms_error = error;
718 
719 	if (error && rmsg->req != RTM_DELETE) {
720 		if (mycpuid != 0) {
721 			panic("rtrequest1_msghandler: rtrequest table req %d, "
722 			    "failed on cpu%d, error %d\n",
723 			    rmsg->req, mycpuid, error);
724 		}
725 		netisr_replymsg(&rmsg->base, error);
726 	} else {
727 		netisr_forwardmsg_error(&rmsg->base, mycpuid + 1,
728 		    rmsg->base.lmsg.ms_error);
729 	}
730 }
731 
732 int
733 rtrequest1(int req, struct rt_addrinfo *rtinfo, struct rtentry **ret_nrt)
734 {
735 	struct sockaddr *dst = rtinfo->rti_info[RTAX_DST];
736 	struct rtentry *rt;
737 	struct radix_node *rn;
738 	struct radix_node_head *rnh;
739 	struct ifaddr *ifa;
740 	struct sockaddr *ndst;
741 	int error = 0;
742 
743 	ASSERT_NETISR_NCPUS(mycpuid);
744 
745 #define gotoerr(x) { error = x ; goto bad; }
746 
747 #ifdef ROUTE_DEBUG
748 	if (route_debug)
749 		rt_addrinfo_print(req, rtinfo);
750 #endif
751 
752 	crit_enter();
753 	/*
754 	 * Find the correct routing tree to use for this Address Family
755 	 */
756 	if ((rnh = rt_tables[mycpuid][dst->sa_family]) == NULL)
757 		gotoerr(EAFNOSUPPORT);
758 
759 	/*
760 	 * If we are adding a host route then we don't want to put
761 	 * a netmask in the tree, nor do we want to clone it.
762 	 */
763 	if (rtinfo->rti_flags & RTF_HOST) {
764 		rtinfo->rti_info[RTAX_NETMASK] = NULL;
765 		rtinfo->rti_flags &= ~(RTF_CLONING | RTF_PRCLONING);
766 	}
767 
768 	switch (req) {
769 	case RTM_DELETE:
770 		/* Remove the item from the tree. */
771 		rn = rnh->rnh_deladdr((char *)rtinfo->rti_info[RTAX_DST],
772 				      (char *)rtinfo->rti_info[RTAX_NETMASK],
773 				      rnh);
774 		if (rn == NULL)
775 			gotoerr(ESRCH);
776 		KASSERT(!(rn->rn_flags & (RNF_ACTIVE | RNF_ROOT)),
777 			("rnh_deladdr returned flags 0x%x", rn->rn_flags));
778 		rt = (struct rtentry *)rn;
779 
780 		/* ref to prevent a deletion race */
781 		++rt->rt_refcnt;
782 
783 		/* Free any routes cloned from this one. */
784 		if ((rt->rt_flags & (RTF_CLONING | RTF_PRCLONING)) &&
785 		    rt_mask(rt) != NULL) {
786 			rnh->rnh_walktree_from(rnh, (char *)rt_key(rt),
787 					       (char *)rt_mask(rt),
788 					       rt_fixdelete, rt);
789 		}
790 
791 		if (rt->rt_gwroute != NULL) {
792 			RTFREE(rt->rt_gwroute);
793 			rt->rt_gwroute = NULL;
794 		}
795 
796 		/*
797 		 * NB: RTF_UP must be set during the search above,
798 		 * because we might delete the last ref, causing
799 		 * rt to get freed prematurely.
800 		 */
801 		rt->rt_flags &= ~RTF_UP;
802 
803 #ifdef ROUTE_DEBUG
804 		if (route_debug)
805 			rt_print(rtinfo, rt);
806 #endif
807 
808 		/* Give the protocol a chance to keep things in sync. */
809 		if ((ifa = rt->rt_ifa) && ifa->ifa_rtrequest)
810 			ifa->ifa_rtrequest(RTM_DELETE, rt);
811 
812 		/*
813 		 * If the caller wants it, then it can have it,
814 		 * but it's up to it to free the rtentry as we won't be
815 		 * doing it.
816 		 */
817 		KASSERT(rt->rt_refcnt >= 0,
818 			("rtrequest1(DELETE): refcnt %ld", rt->rt_refcnt));
819 		if (ret_nrt != NULL) {
820 			/* leave ref intact for return */
821 			*ret_nrt = rt;
822 		} else {
823 			/* deref / attempt to destroy */
824 			rtfree(rt);
825 		}
826 		break;
827 
828 	case RTM_RESOLVE:
829 		if (ret_nrt == NULL || (rt = *ret_nrt) == NULL)
830 			gotoerr(EINVAL);
831 
832 		KASSERT(rt->rt_cpuid == mycpuid,
833 		    ("rt resolve rt_cpuid %d, mycpuid %d",
834 		     rt->rt_cpuid, mycpuid));
835 
836 		ifa = rt->rt_ifa;
837 		rtinfo->rti_flags =
838 		    rt->rt_flags & ~(RTF_CLONING | RTF_PRCLONING | RTF_STATIC);
839 		rtinfo->rti_flags |= RTF_WASCLONED;
840 		rtinfo->rti_info[RTAX_GATEWAY] = rt->rt_gateway;
841 		if ((rtinfo->rti_info[RTAX_NETMASK] = rt->rt_genmask) == NULL)
842 			rtinfo->rti_flags |= RTF_HOST;
843 		rtinfo->rti_info[RTAX_MPLS1] = rt->rt_shim[0];
844 		rtinfo->rti_info[RTAX_MPLS2] = rt->rt_shim[1];
845 		rtinfo->rti_info[RTAX_MPLS3] = rt->rt_shim[2];
846 		goto makeroute;
847 
848 	case RTM_ADD:
849 		KASSERT(!(rtinfo->rti_flags & RTF_GATEWAY) ||
850 			rtinfo->rti_info[RTAX_GATEWAY] != NULL,
851 		    ("rtrequest: GATEWAY but no gateway"));
852 
853 		if (rtinfo->rti_ifa == NULL && (error = rt_getifa(rtinfo)))
854 			gotoerr(error);
855 		ifa = rtinfo->rti_ifa;
856 makeroute:
857 		R_Malloc(rt, struct rtentry *, sizeof(struct rtentry));
858 		if (rt == NULL) {
859 			if (req == RTM_ADD) {
860 				kprintf("rtrequest1: alloc rtentry failed on "
861 				    "cpu%d\n", mycpuid);
862 			}
863 			gotoerr(ENOBUFS);
864 		}
865 		bzero(rt, sizeof(struct rtentry));
866 		rt->rt_flags = RTF_UP | rtinfo->rti_flags;
867 		rt->rt_cpuid = mycpuid;
868 
869 		error = rt_setgate(rt, dst, rtinfo->rti_info[RTAX_GATEWAY]);
870 		if (error != 0) {
871 			Free(rt);
872 			gotoerr(error);
873 		}
874 
875 		ndst = rt_key(rt);
876 		if (rtinfo->rti_info[RTAX_NETMASK] != NULL)
877 			rt_maskedcopy(dst, ndst,
878 				      rtinfo->rti_info[RTAX_NETMASK]);
879 		else
880 			bcopy(dst, ndst, dst->sa_len);
881 
882 		if (rtinfo->rti_info[RTAX_MPLS1] != NULL)
883 			rt_setshims(rt, rtinfo->rti_info);
884 
885 		/*
886 		 * Note that we now have a reference to the ifa.
887 		 * This moved from below so that rnh->rnh_addaddr() can
888 		 * examine the ifa and  ifa->ifa_ifp if it so desires.
889 		 */
890 		IFAREF(ifa);
891 		rt->rt_ifa = ifa;
892 		rt->rt_ifp = ifa->ifa_ifp;
893 		/* XXX mtu manipulation will be done in rnh_addaddr -- itojun */
894 
895 		rn = rnh->rnh_addaddr((char *)ndst,
896 				      (char *)rtinfo->rti_info[RTAX_NETMASK],
897 				      rnh, rt->rt_nodes);
898 		if (rn == NULL) {
899 			struct rtentry *oldrt;
900 
901 			/*
902 			 * We already have one of these in the tree.
903 			 * We do a special hack: if the old route was
904 			 * cloned, then we blow it away and try
905 			 * re-inserting the new one.
906 			 */
907 			oldrt = rtpurelookup(ndst);
908 			if (oldrt != NULL) {
909 				--oldrt->rt_refcnt;
910 				if (oldrt->rt_flags & RTF_WASCLONED) {
911 					rtrequest(RTM_DELETE, rt_key(oldrt),
912 						  oldrt->rt_gateway,
913 						  rt_mask(oldrt),
914 						  oldrt->rt_flags, NULL);
915 					rn = rnh->rnh_addaddr((char *)ndst,
916 					    (char *)
917 						rtinfo->rti_info[RTAX_NETMASK],
918 					    rnh, rt->rt_nodes);
919 				}
920 			}
921 		}
922 		/* NOTE: rt_ifa may have been changed */
923 		ifa = rt->rt_ifa;
924 
925 		/*
926 		 * If it still failed to go into the tree,
927 		 * then un-make it (this should be a function).
928 		 */
929 		if (rn == NULL) {
930 			if (rt->rt_gwroute != NULL)
931 				rtfree(rt->rt_gwroute);
932 			IFAFREE(ifa);
933 			Free(rt_key(rt));
934 			Free(rt);
935 			gotoerr(EEXIST);
936 		}
937 
938 		/*
939 		 * If we got here from RESOLVE, then we are cloning
940 		 * so clone the rest, and note that we
941 		 * are a clone (and increment the parent's references)
942 		 */
943 		if (req == RTM_RESOLVE) {
944 			rt->rt_rmx = (*ret_nrt)->rt_rmx;    /* copy metrics */
945 			rt->rt_rmx.rmx_pksent = 0;  /* reset packet counter */
946 			if ((*ret_nrt)->rt_flags &
947 				       (RTF_CLONING | RTF_PRCLONING)) {
948 				rt->rt_parent = *ret_nrt;
949 				(*ret_nrt)->rt_refcnt++;
950 			}
951 		}
952 
953 		/*
954 		 * if this protocol has something to add to this then
955 		 * allow it to do that as well.
956 		 */
957 		if (ifa->ifa_rtrequest != NULL)
958 			ifa->ifa_rtrequest(req, rt);
959 
960 		/*
961 		 * We repeat the same procedure from rt_setgate() here because
962 		 * it doesn't fire when we call it there because the node
963 		 * hasn't been added to the tree yet.
964 		 */
965 		if (req == RTM_ADD && !(rt->rt_flags & RTF_HOST) &&
966 		    rt_mask(rt) != NULL) {
967 			struct rtfc_arg arg = { rt, rnh };
968 
969 			rnh->rnh_walktree_from(rnh, (char *)rt_key(rt),
970 					       (char *)rt_mask(rt),
971 					       rt_fixchange, &arg);
972 		}
973 
974 #ifdef ROUTE_DEBUG
975 		if (route_debug)
976 			rt_print(rtinfo, rt);
977 #endif
978 		/*
979 		 * Return the resulting rtentry,
980 		 * increasing the number of references by one.
981 		 */
982 		if (ret_nrt != NULL) {
983 			rt->rt_refcnt++;
984 			*ret_nrt = rt;
985 		}
986 		break;
987 	case RTM_GET:
988 		/* Get the item from the tree. */
989 		rn = rnh->rnh_lookup((char *)rtinfo->rti_info[RTAX_DST],
990 				     (char *)rtinfo->rti_info[RTAX_NETMASK],
991 				      rnh);
992 		if (rn == NULL)
993 			gotoerr(ESRCH);
994 		if (ret_nrt != NULL) {
995 			rt = (struct rtentry *)rn;
996 			rt->rt_refcnt++;
997 			*ret_nrt = rt;
998 		}
999 		break;
1000 	default:
1001 		error = EOPNOTSUPP;
1002 	}
1003 bad:
1004 #ifdef ROUTE_DEBUG
1005 	if (route_debug) {
1006 		if (error)
1007 			kprintf("rti %p failed error %d\n", rtinfo, error);
1008 		else
1009 			kprintf("rti %p succeeded\n", rtinfo);
1010 	}
1011 #endif
1012 	crit_exit();
1013 	return (error);
1014 }
1015 
1016 /*
1017  * Called from rtrequest(RTM_DELETE, ...) to fix up the route's ``family''
1018  * (i.e., the routes related to it by the operation of cloning).  This
1019  * routine is iterated over all potential former-child-routes by way of
1020  * rnh->rnh_walktree_from() above, and those that actually are children of
1021  * the late parent (passed in as VP here) are themselves deleted.
1022  */
1023 static int
1024 rt_fixdelete(struct radix_node *rn, void *vp)
1025 {
1026 	struct rtentry *rt = (struct rtentry *)rn;
1027 	struct rtentry *rt0 = vp;
1028 
1029 	if (rt->rt_parent == rt0 &&
1030 	    !(rt->rt_flags & (RTF_PINNED | RTF_CLONING | RTF_PRCLONING))) {
1031 		return rtrequest(RTM_DELETE, rt_key(rt), NULL, rt_mask(rt),
1032 				 rt->rt_flags, NULL);
1033 	}
1034 	return 0;
1035 }
1036 
1037 /*
1038  * This routine is called from rt_setgate() to do the analogous thing for
1039  * adds and changes.  There is the added complication in this case of a
1040  * middle insert; i.e., insertion of a new network route between an older
1041  * network route and (cloned) host routes.  For this reason, a simple check
1042  * of rt->rt_parent is insufficient; each candidate route must be tested
1043  * against the (mask, value) of the new route (passed as before in vp)
1044  * to see if the new route matches it.
1045  *
1046  * XXX - it may be possible to do fixdelete() for changes and reserve this
1047  * routine just for adds.  I'm not sure why I thought it was necessary to do
1048  * changes this way.
1049  */
1050 #ifdef DEBUG
1051 static int rtfcdebug = 0;
1052 #endif
1053 
1054 static int
1055 rt_fixchange(struct radix_node *rn, void *vp)
1056 {
1057 	struct rtentry *rt = (struct rtentry *)rn;
1058 	struct rtfc_arg *ap = vp;
1059 	struct rtentry *rt0 = ap->rt0;
1060 	struct radix_node_head *rnh = ap->rnh;
1061 	u_char *xk1, *xm1, *xk2, *xmp;
1062 	int i, len, mlen;
1063 
1064 #ifdef DEBUG
1065 	if (rtfcdebug)
1066 		kprintf("rt_fixchange: rt %p, rt0 %p\n", rt, rt0);
1067 #endif
1068 
1069 	if (rt->rt_parent == NULL ||
1070 	    (rt->rt_flags & (RTF_PINNED | RTF_CLONING | RTF_PRCLONING))) {
1071 #ifdef DEBUG
1072 		if (rtfcdebug) kprintf("no parent, pinned or cloning\n");
1073 #endif
1074 		return 0;
1075 	}
1076 
1077 	if (rt->rt_parent == rt0) {
1078 #ifdef DEBUG
1079 		if (rtfcdebug) kprintf("parent match\n");
1080 #endif
1081 		return rtrequest(RTM_DELETE, rt_key(rt), NULL, rt_mask(rt),
1082 				 rt->rt_flags, NULL);
1083 	}
1084 
1085 	/*
1086 	 * There probably is a function somewhere which does this...
1087 	 * if not, there should be.
1088 	 */
1089 	len = imin(rt_key(rt0)->sa_len, rt_key(rt)->sa_len);
1090 
1091 	xk1 = (u_char *)rt_key(rt0);
1092 	xm1 = (u_char *)rt_mask(rt0);
1093 	xk2 = (u_char *)rt_key(rt);
1094 
1095 	/* avoid applying a less specific route */
1096 	xmp = (u_char *)rt_mask(rt->rt_parent);
1097 	mlen = rt_key(rt->rt_parent)->sa_len;
1098 	if (mlen > rt_key(rt0)->sa_len) {
1099 #ifdef DEBUG
1100 		if (rtfcdebug)
1101 			kprintf("rt_fixchange: inserting a less "
1102 			       "specific route\n");
1103 #endif
1104 		return 0;
1105 	}
1106 	for (i = rnh->rnh_treetop->rn_offset; i < mlen; i++) {
1107 		if ((xmp[i] & ~(xmp[i] ^ xm1[i])) != xmp[i]) {
1108 #ifdef DEBUG
1109 			if (rtfcdebug)
1110 				kprintf("rt_fixchange: inserting a less "
1111 				       "specific route\n");
1112 #endif
1113 			return 0;
1114 		}
1115 	}
1116 
1117 	for (i = rnh->rnh_treetop->rn_offset; i < len; i++) {
1118 		if ((xk2[i] & xm1[i]) != xk1[i]) {
1119 #ifdef DEBUG
1120 			if (rtfcdebug) kprintf("no match\n");
1121 #endif
1122 			return 0;
1123 		}
1124 	}
1125 
1126 	/*
1127 	 * OK, this node is a clone, and matches the node currently being
1128 	 * changed/added under the node's mask.  So, get rid of it.
1129 	 */
1130 #ifdef DEBUG
1131 	if (rtfcdebug) kprintf("deleting\n");
1132 #endif
1133 	return rtrequest(RTM_DELETE, rt_key(rt), NULL, rt_mask(rt),
1134 			 rt->rt_flags, NULL);
1135 }
1136 
1137 int
1138 rt_setgate(struct rtentry *rt0, struct sockaddr *dst, struct sockaddr *gate)
1139 {
1140 	char *space, *oldspace;
1141 	int dlen = RT_ROUNDUP(dst->sa_len), glen = RT_ROUNDUP(gate->sa_len);
1142 	struct rtentry *rt = rt0;
1143 	struct radix_node_head *rnh = rt_tables[mycpuid][dst->sa_family];
1144 
1145 	ASSERT_NETISR_NCPUS(mycpuid);
1146 
1147 	/*
1148 	 * A host route with the destination equal to the gateway
1149 	 * will interfere with keeping LLINFO in the routing
1150 	 * table, so disallow it.
1151 	 */
1152 	if (((rt0->rt_flags & (RTF_HOST | RTF_GATEWAY | RTF_LLINFO)) ==
1153 			      (RTF_HOST | RTF_GATEWAY)) &&
1154 	    dst->sa_len == gate->sa_len &&
1155 	    sa_equal(dst, gate)) {
1156 		/*
1157 		 * The route might already exist if this is an RTM_CHANGE
1158 		 * or a routing redirect, so try to delete it.
1159 		 */
1160 		if (rt_key(rt0) != NULL)
1161 			rtrequest(RTM_DELETE, rt_key(rt0), rt0->rt_gateway,
1162 				  rt_mask(rt0), rt0->rt_flags, NULL);
1163 		return EADDRNOTAVAIL;
1164 	}
1165 
1166 	/*
1167 	 * Both dst and gateway are stored in the same malloc'ed chunk
1168 	 * (If I ever get my hands on....)
1169 	 * if we need to malloc a new chunk, then keep the old one around
1170 	 * till we don't need it any more.
1171 	 */
1172 	if (rt->rt_gateway == NULL ||
1173 	    glen > RT_ROUNDUP(rt->rt_gateway->sa_len)) {
1174 		oldspace = (char *)rt_key(rt);
1175 		R_Malloc(space, char *, dlen + glen);
1176 		if (space == NULL)
1177 			return ENOBUFS;
1178 		rt->rt_nodes->rn_key = space;
1179 	} else {
1180 		space = (char *)rt_key(rt);	/* Just use the old space. */
1181 		oldspace = NULL;
1182 	}
1183 
1184 	/* Set the gateway value. */
1185 	rt->rt_gateway = (struct sockaddr *)(space + dlen);
1186 	bcopy(gate, rt->rt_gateway, glen);
1187 
1188 	if (oldspace != NULL) {
1189 		/*
1190 		 * If we allocated a new chunk, preserve the original dst.
1191 		 * This way, rt_setgate() really just sets the gate
1192 		 * and leaves the dst field alone.
1193 		 */
1194 		bcopy(dst, space, dlen);
1195 		Free(oldspace);
1196 	}
1197 
1198 	/*
1199 	 * If there is already a gwroute, it's now almost definitely wrong
1200 	 * so drop it.
1201 	 */
1202 	if (rt->rt_gwroute != NULL) {
1203 		RTFREE(rt->rt_gwroute);
1204 		rt->rt_gwroute = NULL;
1205 	}
1206 	if (rt->rt_flags & RTF_GATEWAY) {
1207 		/*
1208 		 * Cloning loop avoidance: In the presence of
1209 		 * protocol-cloning and bad configuration, it is
1210 		 * possible to get stuck in bottomless mutual recursion
1211 		 * (rtrequest rt_setgate rtlookup).  We avoid this
1212 		 * by not allowing protocol-cloning to operate for
1213 		 * gateways (which is probably the correct choice
1214 		 * anyway), and avoid the resulting reference loops
1215 		 * by disallowing any route to run through itself as
1216 		 * a gateway.  This is obviously mandatory when we
1217 		 * get rt->rt_output().
1218 		 *
1219 		 * This breaks TTCP for hosts outside the gateway!  XXX JH
1220 		 */
1221 		rt->rt_gwroute = _rtlookup(gate, RTF_PRCLONING);
1222 		if (rt->rt_gwroute == rt) {
1223 			rt->rt_gwroute = NULL;
1224 			--rt->rt_refcnt;
1225 			return EDQUOT; /* failure */
1226 		}
1227 	}
1228 
1229 	/*
1230 	 * This isn't going to do anything useful for host routes, so
1231 	 * don't bother.  Also make sure we have a reasonable mask
1232 	 * (we don't yet have one during adds).
1233 	 */
1234 	if (!(rt->rt_flags & RTF_HOST) && rt_mask(rt) != NULL) {
1235 		struct rtfc_arg arg = { rt, rnh };
1236 
1237 		rnh->rnh_walktree_from(rnh, (char *)rt_key(rt),
1238 				       (char *)rt_mask(rt),
1239 				       rt_fixchange, &arg);
1240 	}
1241 
1242 	return 0;
1243 }
1244 
1245 static void
1246 rt_maskedcopy(
1247 	struct sockaddr *src,
1248 	struct sockaddr *dst,
1249 	struct sockaddr *netmask)
1250 {
1251 	u_char *cp1 = (u_char *)src;
1252 	u_char *cp2 = (u_char *)dst;
1253 	u_char *cp3 = (u_char *)netmask;
1254 	u_char *cplim = cp2 + *cp3;
1255 	u_char *cplim2 = cp2 + *cp1;
1256 
1257 	*cp2++ = *cp1++; *cp2++ = *cp1++; /* copies sa_len & sa_family */
1258 	cp3 += 2;
1259 	if (cplim > cplim2)
1260 		cplim = cplim2;
1261 	while (cp2 < cplim)
1262 		*cp2++ = *cp1++ & *cp3++;
1263 	if (cp2 < cplim2)
1264 		bzero(cp2, cplim2 - cp2);
1265 }
1266 
1267 int
1268 rt_llroute(struct sockaddr *dst, struct rtentry *rt0, struct rtentry **drt)
1269 {
1270 	struct rtentry *up_rt, *rt;
1271 
1272 	ASSERT_NETISR_NCPUS(mycpuid);
1273 
1274 	if (!(rt0->rt_flags & RTF_UP)) {
1275 		up_rt = rtlookup(dst);
1276 		if (up_rt == NULL)
1277 			return (EHOSTUNREACH);
1278 		up_rt->rt_refcnt--;
1279 	} else
1280 		up_rt = rt0;
1281 	if (up_rt->rt_flags & RTF_GATEWAY) {
1282 		if (up_rt->rt_gwroute == NULL) {
1283 			up_rt->rt_gwroute = rtlookup(up_rt->rt_gateway);
1284 			if (up_rt->rt_gwroute == NULL)
1285 				return (EHOSTUNREACH);
1286 		} else if (!(up_rt->rt_gwroute->rt_flags & RTF_UP)) {
1287 			rtfree(up_rt->rt_gwroute);
1288 			up_rt->rt_gwroute = rtlookup(up_rt->rt_gateway);
1289 			if (up_rt->rt_gwroute == NULL)
1290 				return (EHOSTUNREACH);
1291 		}
1292 		rt = up_rt->rt_gwroute;
1293 	} else
1294 		rt = up_rt;
1295 	if (rt->rt_flags & RTF_REJECT &&
1296 	    (rt->rt_rmx.rmx_expire == 0 ||		/* rt doesn't expire */
1297 	     time_uptime < rt->rt_rmx.rmx_expire))	/* rt not expired */
1298 		return (rt->rt_flags & RTF_HOST ?  EHOSTDOWN : EHOSTUNREACH);
1299 	*drt = rt;
1300 	return 0;
1301 }
1302 
1303 static int
1304 rt_setshims(struct rtentry *rt, struct sockaddr **rt_shim){
1305 	int i;
1306 
1307 	for (i=0; i<3; i++) {
1308 		struct sockaddr *shim = rt_shim[RTAX_MPLS1 + i];
1309 		int shimlen;
1310 
1311 		if (shim == NULL)
1312 			break;
1313 
1314 		shimlen = RT_ROUNDUP(shim->sa_len);
1315 		R_Malloc(rt->rt_shim[i], struct sockaddr *, shimlen);
1316 		bcopy(shim, rt->rt_shim[i], shimlen);
1317 	}
1318 
1319 	return 0;
1320 }
1321 
1322 #ifdef ROUTE_DEBUG
1323 
1324 /*
1325  * Print out a route table entry
1326  */
1327 void
1328 rt_print(struct rt_addrinfo *rtinfo, struct rtentry *rn)
1329 {
1330 	kprintf("rti %p cpu %d route %p flags %08lx: ",
1331 		rtinfo, mycpuid, rn, rn->rt_flags);
1332 	sockaddr_print(rt_key(rn));
1333 	kprintf(" mask ");
1334 	sockaddr_print(rt_mask(rn));
1335 	kprintf(" gw ");
1336 	sockaddr_print(rn->rt_gateway);
1337 	kprintf(" ifc \"%s\"", rn->rt_ifp ? rn->rt_ifp->if_dname : "?");
1338 	kprintf(" ifa %p\n", rn->rt_ifa);
1339 }
1340 
1341 void
1342 rt_addrinfo_print(int cmd, struct rt_addrinfo *rti)
1343 {
1344 	int didit = 0;
1345 	int i;
1346 
1347 #ifdef ROUTE_DEBUG
1348 	if (cmd == RTM_DELETE && route_debug > 1)
1349 		print_backtrace(-1);
1350 #endif
1351 
1352 	switch(cmd) {
1353 	case RTM_ADD:
1354 		kprintf("ADD ");
1355 		break;
1356 	case RTM_RESOLVE:
1357 		kprintf("RES ");
1358 		break;
1359 	case RTM_DELETE:
1360 		kprintf("DEL ");
1361 		break;
1362 	default:
1363 		kprintf("C%02d ", cmd);
1364 		break;
1365 	}
1366 	kprintf("rti %p cpu %d ", rti, mycpuid);
1367 	for (i = 0; i < rti->rti_addrs; ++i) {
1368 		if (rti->rti_info[i] == NULL)
1369 			continue;
1370 		if (didit)
1371 			kprintf(" ,");
1372 		switch(i) {
1373 		case RTAX_DST:
1374 			kprintf("(DST ");
1375 			break;
1376 		case RTAX_GATEWAY:
1377 			kprintf("(GWY ");
1378 			break;
1379 		case RTAX_NETMASK:
1380 			kprintf("(MSK ");
1381 			break;
1382 		case RTAX_GENMASK:
1383 			kprintf("(GEN ");
1384 			break;
1385 		case RTAX_IFP:
1386 			kprintf("(IFP ");
1387 			break;
1388 		case RTAX_IFA:
1389 			kprintf("(IFA ");
1390 			break;
1391 		case RTAX_AUTHOR:
1392 			kprintf("(AUT ");
1393 			break;
1394 		case RTAX_BRD:
1395 			kprintf("(BRD ");
1396 			break;
1397 		default:
1398 			kprintf("(?%02d ", i);
1399 			break;
1400 		}
1401 		sockaddr_print(rti->rti_info[i]);
1402 		kprintf(")");
1403 		didit = 1;
1404 	}
1405 	kprintf("\n");
1406 }
1407 
1408 void
1409 sockaddr_print(struct sockaddr *sa)
1410 {
1411 	struct sockaddr_in *sa4;
1412 	struct sockaddr_in6 *sa6;
1413 	int len;
1414 	int i;
1415 
1416 	if (sa == NULL) {
1417 		kprintf("NULL");
1418 		return;
1419 	}
1420 
1421 	len = sa->sa_len - offsetof(struct sockaddr, sa_data[0]);
1422 
1423 	switch(sa->sa_family) {
1424 	case AF_INET:
1425 	case AF_INET6:
1426 	default:
1427 		switch(sa->sa_family) {
1428 		case AF_INET:
1429 			sa4 = (struct sockaddr_in *)sa;
1430 			kprintf("INET %d %d.%d.%d.%d",
1431 				ntohs(sa4->sin_port),
1432 				(ntohl(sa4->sin_addr.s_addr) >> 24) & 255,
1433 				(ntohl(sa4->sin_addr.s_addr) >> 16) & 255,
1434 				(ntohl(sa4->sin_addr.s_addr) >> 8) & 255,
1435 				(ntohl(sa4->sin_addr.s_addr) >> 0) & 255
1436 			);
1437 			break;
1438 		case AF_INET6:
1439 			sa6 = (struct sockaddr_in6 *)sa;
1440 			kprintf("INET6 %d %04x:%04x%04x:%04x:%04x:%04x:%04x:%04x",
1441 				ntohs(sa6->sin6_port),
1442 				sa6->sin6_addr.s6_addr16[0],
1443 				sa6->sin6_addr.s6_addr16[1],
1444 				sa6->sin6_addr.s6_addr16[2],
1445 				sa6->sin6_addr.s6_addr16[3],
1446 				sa6->sin6_addr.s6_addr16[4],
1447 				sa6->sin6_addr.s6_addr16[5],
1448 				sa6->sin6_addr.s6_addr16[6],
1449 				sa6->sin6_addr.s6_addr16[7]
1450 			);
1451 			break;
1452 		default:
1453 			kprintf("AF%d ", sa->sa_family);
1454 			while (len > 0 && sa->sa_data[len-1] == 0)
1455 				--len;
1456 
1457 			for (i = 0; i < len; ++i) {
1458 				if (i)
1459 					kprintf(".");
1460 				kprintf("%d", (unsigned char)sa->sa_data[i]);
1461 			}
1462 			break;
1463 		}
1464 	}
1465 }
1466 
1467 #endif
1468 
1469 /*
1470  * Set up a routing table entry, normally for an interface.
1471  */
1472 int
1473 rtinit(struct ifaddr *ifa, int cmd, int flags)
1474 {
1475 	struct sockaddr *dst, *deldst, *netmask;
1476 	struct mbuf *m = NULL;
1477 	struct radix_node_head *rnh;
1478 	struct radix_node *rn;
1479 	struct rt_addrinfo rtinfo;
1480 	int error;
1481 
1482 	ASSERT_NETISR0;
1483 
1484 	if (flags & RTF_HOST) {
1485 		dst = ifa->ifa_dstaddr;
1486 		netmask = NULL;
1487 	} else {
1488 		dst = ifa->ifa_addr;
1489 		netmask = ifa->ifa_netmask;
1490 	}
1491 	/*
1492 	 * If it's a delete, check that if it exists, it's on the correct
1493 	 * interface or we might scrub a route to another ifa which would
1494 	 * be confusing at best and possibly worse.
1495 	 */
1496 	if (cmd == RTM_DELETE) {
1497 		/*
1498 		 * It's a delete, so it should already exist..
1499 		 * If it's a net, mask off the host bits
1500 		 * (Assuming we have a mask)
1501 		 */
1502 		if (netmask != NULL) {
1503 			m = m_get(M_NOWAIT, MT_SONAME);
1504 			if (m == NULL)
1505 				return (ENOBUFS);
1506 			mbuftrackid(m, 34);
1507 			deldst = mtod(m, struct sockaddr *);
1508 			rt_maskedcopy(dst, deldst, netmask);
1509 			dst = deldst;
1510 		}
1511 		/*
1512 		 * Look up an rtentry that is in the routing tree and
1513 		 * contains the correct info.
1514 		 */
1515 		if ((rnh = rt_tables[mycpuid][dst->sa_family]) == NULL ||
1516 		    (rn = rnh->rnh_lookup((char *)dst,
1517 					  (char *)netmask, rnh)) == NULL ||
1518 		    ((struct rtentry *)rn)->rt_ifa != ifa ||
1519 		    !sa_equal((struct sockaddr *)rn->rn_key, dst)) {
1520 			if (m != NULL)
1521 				m_free(m);
1522 			return (flags & RTF_HOST ? EHOSTUNREACH : ENETUNREACH);
1523 		}
1524 		/* XXX */
1525 #if 0
1526 		else {
1527 			/*
1528 			 * One would think that as we are deleting, and we know
1529 			 * it doesn't exist, we could just return at this point
1530 			 * with an "ELSE" clause, but apparently not..
1531 			 */
1532 			return (flags & RTF_HOST ? EHOSTUNREACH : ENETUNREACH);
1533 		}
1534 #endif
1535 	}
1536 	/*
1537 	 * Do the actual request
1538 	 */
1539 	bzero(&rtinfo, sizeof(struct rt_addrinfo));
1540 	rtinfo.rti_info[RTAX_DST] = dst;
1541 	rtinfo.rti_info[RTAX_GATEWAY] = ifa->ifa_addr;
1542 	rtinfo.rti_info[RTAX_NETMASK] = netmask;
1543 	rtinfo.rti_flags = flags | ifa->ifa_flags;
1544 	rtinfo.rti_ifa = ifa;
1545 	error = rtrequest1_global(cmd, &rtinfo, rtinit_rtrequest_callback, ifa,
1546 	    RTREQ_PRIO_HIGH);
1547 	if (m != NULL)
1548 		m_free(m);
1549 	return (error);
1550 }
1551 
1552 static void
1553 rtinit_rtrequest_callback(int cmd, int error,
1554 			  struct rt_addrinfo *rtinfo, struct rtentry *rt,
1555 			  void *arg)
1556 {
1557 	struct ifaddr *ifa = arg;
1558 
1559 	if (error == 0 && rt) {
1560 		if (mycpuid == 0)
1561 			rt_newaddrmsg(cmd, ifa, error, rt);
1562 		if (cmd == RTM_DELETE) {
1563 			if (rt->rt_refcnt == 0) {
1564 				++rt->rt_refcnt;
1565 				rtfree(rt);
1566 			}
1567 		}
1568 	}
1569 }
1570 
1571 struct netmsg_rts {
1572 	struct netmsg_base	base;
1573 	int			req;
1574 	struct rt_addrinfo	*rtinfo;
1575 	rtsearch_callback_func_t callback;
1576 	void			*arg;
1577 	boolean_t		exact_match;
1578 	int			found_cnt;
1579 };
1580 
1581 int
1582 rtsearch_global(int req, struct rt_addrinfo *rtinfo,
1583     rtsearch_callback_func_t callback, void *arg, boolean_t exact_match,
1584     boolean_t req_prio)
1585 {
1586 	struct netmsg_rts msg;
1587 	int flags = 0;
1588 
1589 	if (req_prio)
1590 		flags = MSGF_PRIORITY;
1591 	netmsg_init(&msg.base, NULL, &curthread->td_msgport, flags,
1592 	    rtsearch_msghandler);
1593 	msg.req = req;
1594 	msg.rtinfo = rtinfo;
1595 	msg.callback = callback;
1596 	msg.arg = arg;
1597 	msg.exact_match = exact_match;
1598 	msg.found_cnt = 0;
1599 	return (netisr_domsg_global(&msg.base));
1600 }
1601 
1602 static void
1603 rtsearch_msghandler(netmsg_t msg)
1604 {
1605 	struct netmsg_rts *rmsg = (void *)msg;
1606 	struct rt_addrinfo rtinfo;
1607 	struct radix_node_head *rnh;
1608 	struct rtentry *rt;
1609 	int error;
1610 
1611 	ASSERT_NETISR_NCPUS(mycpuid);
1612 
1613 	/*
1614 	 * Copy the rtinfo.  We need to make sure that the original
1615 	 * rtinfo, which is setup by the caller, in the netmsg will
1616 	 * _not_ be changed; else the next CPU on the netmsg forwarding
1617 	 * path will see a different rtinfo than what this CPU has seen.
1618 	 */
1619 	rtinfo = *rmsg->rtinfo;
1620 
1621 	/*
1622 	 * Find the correct routing tree to use for this Address Family
1623 	 */
1624 	if ((rnh = rt_tables[mycpuid][rtinfo.rti_dst->sa_family]) == NULL) {
1625 		if (mycpuid != 0)
1626 			panic("partially initialized routing tables");
1627 		netisr_replymsg(&rmsg->base, EAFNOSUPPORT);
1628 		return;
1629 	}
1630 
1631 	/*
1632 	 * Correct rtinfo for the host route searching.
1633 	 */
1634 	if (rtinfo.rti_flags & RTF_HOST) {
1635 		rtinfo.rti_netmask = NULL;
1636 		rtinfo.rti_flags &= ~(RTF_CLONING | RTF_PRCLONING);
1637 	}
1638 
1639 	rt = (struct rtentry *)
1640 	     rnh->rnh_lookup((char *)rtinfo.rti_dst,
1641 			     (char *)rtinfo.rti_netmask, rnh);
1642 
1643 	/*
1644 	 * If we are asked to do the "exact match", we need to make sure
1645 	 * that host route searching got a host route while a network
1646 	 * route searching got a network route.
1647 	 */
1648 	if (rt != NULL && rmsg->exact_match &&
1649 	    ((rt->rt_flags ^ rtinfo.rti_flags) & RTF_HOST))
1650 		rt = NULL;
1651 
1652 	if (rt == NULL) {
1653 		/*
1654 		 * No matching routes have been found, don't count this
1655 		 * as a critical error (here, we set 'error' to 0), just
1656 		 * keep moving on, since at least prcloned routes are not
1657 		 * duplicated onto each CPU.
1658 		 */
1659 		error = 0;
1660 	} else {
1661 		rmsg->found_cnt++;
1662 
1663 		rt->rt_refcnt++;
1664 		error = rmsg->callback(rmsg->req, &rtinfo, rt, rmsg->arg,
1665 				      rmsg->found_cnt);
1666 		rt->rt_refcnt--;
1667 
1668 		if (error == EJUSTRETURN) {
1669 			netisr_replymsg(&rmsg->base, 0);
1670 			return;
1671 		}
1672 	}
1673 
1674 	if (error) {
1675 		KKASSERT(rmsg->found_cnt > 0);
1676 
1677 		/*
1678 		 * Under following cases, unrecoverable error has
1679 		 * not occured:
1680 		 * o  Request is RTM_GET
1681 		 * o  The first time that we find the route, but the
1682 		 *    modification fails.
1683 		 */
1684 		if (rmsg->req != RTM_GET && rmsg->found_cnt > 1) {
1685 			panic("rtsearch_msghandler: unrecoverable error "
1686 			      "cpu %d", mycpuid);
1687 		}
1688 		netisr_replymsg(&rmsg->base, error);
1689 	} else {
1690 		if (rmsg->found_cnt == 0) {
1691 			/* The requested route has not been seen ... */
1692 			error = ESRCH;
1693 		}
1694 		netisr_forwardmsg_error(&rmsg->base, mycpuid + 1, error);
1695 	}
1696 }
1697 
1698 int
1699 rtmask_add_global(struct sockaddr *mask, boolean_t req_prio)
1700 {
1701 	struct netmsg_base msg;
1702 	int flags = 0;
1703 
1704 	if (req_prio)
1705 		flags = MSGF_PRIORITY;
1706 	netmsg_init(&msg, NULL, &curthread->td_msgport, flags,
1707 	    rtmask_add_msghandler);
1708 	msg.lmsg.u.ms_resultp = mask;
1709 
1710 	return (netisr_domsg_global(&msg));
1711 }
1712 
1713 struct sockaddr *
1714 _rtmask_lookup(struct sockaddr *mask, boolean_t search)
1715 {
1716 	struct radix_node *n;
1717 
1718 #define	clen(s)	(*(u_char *)(s))
1719 	n = rn_addmask((char *)mask, search, 1, rn_cpumaskhead(mycpuid));
1720 	if (n != NULL &&
1721 	    mask->sa_len >= clen(n->rn_key) &&
1722 	    bcmp((char *)mask + 1,
1723 		 (char *)n->rn_key + 1, clen(n->rn_key) - 1) == 0) {
1724 		return (struct sockaddr *)n->rn_key;
1725 	} else {
1726 		return NULL;
1727 	}
1728 #undef clen
1729 }
1730 
1731 static void
1732 rtmask_add_msghandler(netmsg_t msg)
1733 {
1734 	struct sockaddr *mask = msg->lmsg.u.ms_resultp;
1735 
1736 	ASSERT_NETISR_NCPUS(mycpuid);
1737 
1738 	if (rtmask_lookup(mask) == NULL) {
1739 		netisr_replymsg(&msg->base, ENOBUFS);
1740 		return;
1741 	}
1742 	netisr_forwardmsg(&msg->base, mycpuid + 1);
1743 }
1744 
1745 /* This must be before ip6_init2(), which is now SI_ORDER_MIDDLE */
1746 SYSINIT(route, SI_SUB_PROTO_DOMAIN, SI_ORDER_THIRD, route_init, 0);
1747 
1748 struct rtchange_arg {
1749 	struct ifaddr	*old_ifa;
1750 	struct ifaddr	*new_ifa;
1751 	struct rtentry	*rt;
1752 	int		changed;
1753 };
1754 
1755 static void
1756 rtchange_ifa(struct rtentry *rt, struct rtchange_arg *ap)
1757 {
1758 	if (rt->rt_ifa->ifa_rtrequest != NULL)
1759 		rt->rt_ifa->ifa_rtrequest(RTM_DELETE, rt);
1760 	IFAFREE(rt->rt_ifa);
1761 
1762 	IFAREF(ap->new_ifa);
1763 	rt->rt_ifa = ap->new_ifa;
1764 	rt->rt_ifp = ap->new_ifa->ifa_ifp;
1765 	if (rt->rt_ifa->ifa_rtrequest != NULL)
1766 		rt->rt_ifa->ifa_rtrequest(RTM_ADD, rt);
1767 
1768 	ap->changed = 1;
1769 }
1770 
1771 static int
1772 rtchange_callback(struct radix_node *rn, void *xap)
1773 {
1774 	struct rtchange_arg *ap = xap;
1775 	struct rtentry *rt = (struct rtentry *)rn;
1776 
1777 	if (rt->rt_ifa == ap->old_ifa) {
1778 		if (rt->rt_flags & (RTF_CLONING | RTF_PRCLONING)) {
1779 			/*
1780 			 * We could saw the branch off when we are
1781 			 * still sitting on it, if the ifa_rtrequest
1782 			 * DEL/ADD are called directly from here.
1783 			 */
1784 			ap->rt = rt;
1785 			return EJUSTRETURN;
1786 		}
1787 		rtchange_ifa(rt, ap);
1788 	}
1789 	return 0;
1790 }
1791 
1792 struct netmsg_rtchange {
1793 	struct netmsg_base	base;
1794 	struct ifaddr		*old_ifa;
1795 	struct ifaddr		*new_ifa;
1796 	int			changed;
1797 };
1798 
1799 static void
1800 rtchange_dispatch(netmsg_t msg)
1801 {
1802 	struct netmsg_rtchange *rmsg = (void *)msg;
1803 	struct radix_node_head *rnh;
1804 	struct rtchange_arg arg;
1805 	int cpu;
1806 
1807 	cpu = mycpuid;
1808 	ASSERT_NETISR_NCPUS(cpu);
1809 
1810 	memset(&arg, 0, sizeof(arg));
1811 	arg.old_ifa = rmsg->old_ifa;
1812 	arg.new_ifa = rmsg->new_ifa;
1813 
1814 	rnh = rt_tables[cpu][AF_INET];
1815 	for (;;) {
1816 		int error;
1817 
1818 		KKASSERT(arg.rt == NULL);
1819 		error = rnh->rnh_walktree(rnh, rtchange_callback, &arg);
1820 		if (arg.rt != NULL) {
1821 			struct rtentry *rt;
1822 
1823 			rt = arg.rt;
1824 			arg.rt = NULL;
1825 			rtchange_ifa(rt, &arg);
1826 		} else {
1827 			break;
1828 		}
1829 	}
1830 	if (arg.changed)
1831 		rmsg->changed = 1;
1832 
1833 	netisr_forwardmsg(&rmsg->base, cpu + 1);
1834 }
1835 
1836 int
1837 rtchange(struct ifaddr *old_ifa, struct ifaddr *new_ifa)
1838 {
1839 	struct netmsg_rtchange msg;
1840 
1841 	/*
1842 	 * XXX individual requests are not independantly chained,
1843 	 * which means that the per-cpu route tables will not be
1844 	 * consistent in the middle of the operation.  If routes
1845 	 * related to the interface are manipulated while we are
1846 	 * doing this the inconsistancy could trigger a panic.
1847 	 */
1848 	netmsg_init(&msg.base, NULL, &curthread->td_msgport, MSGF_PRIORITY,
1849 	    rtchange_dispatch);
1850 	msg.old_ifa = old_ifa;
1851 	msg.new_ifa = new_ifa;
1852 	msg.changed = 0;
1853 	netisr_domsg_global(&msg.base);
1854 
1855 	if (msg.changed) {
1856 		old_ifa->ifa_flags &= ~IFA_ROUTE;
1857 		new_ifa->ifa_flags |= IFA_ROUTE;
1858 		return 0;
1859 	} else {
1860 		return ENOENT;
1861 	}
1862 }
1863