xref: /dflybsd-src/sys/net/route.c (revision 0bb7d8c82a64940013681cf515d16f3e62eb7e3c)
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. All advertising materials mentioning features or use of this software
46  *    must display the following acknowledgement:
47  *	This product includes software developed by the University of
48  *	California, Berkeley and its contributors.
49  * 4. Neither the name of the University nor the names of its contributors
50  *    may be used to endorse or promote products derived from this software
51  *    without specific prior written permission.
52  *
53  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
54  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
55  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
56  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
57  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
58  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
59  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
60  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
61  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
62  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
63  * SUCH DAMAGE.
64  *
65  *	@(#)route.c	8.3 (Berkeley) 1/9/95
66  * $FreeBSD: src/sys/net/route.c,v 1.59.2.10 2003/01/17 08:04:00 ru Exp $
67  * $DragonFly: src/sys/net/route.c,v 1.41 2008/11/09 10:50:15 sephe Exp $
68  */
69 
70 #include "opt_inet.h"
71 #include "opt_mpls.h"
72 
73 #include <sys/param.h>
74 #include <sys/systm.h>
75 #include <sys/malloc.h>
76 #include <sys/mbuf.h>
77 #include <sys/socket.h>
78 #include <sys/domain.h>
79 #include <sys/kernel.h>
80 #include <sys/sysctl.h>
81 #include <sys/globaldata.h>
82 #include <sys/thread.h>
83 
84 #include <net/if.h>
85 #include <net/route.h>
86 #include <net/netisr.h>
87 
88 #include <netinet/in.h>
89 #include <net/ip_mroute/ip_mroute.h>
90 
91 #include <sys/thread2.h>
92 #include <sys/msgport2.h>
93 #include <sys/mplock2.h>
94 #include <net/netmsg2.h>
95 
96 #ifdef MPLS
97 #include <netproto/mpls/mpls.h>
98 #endif
99 
100 static struct rtstatistics rtstatistics_percpu[MAXCPU];
101 #ifdef SMP
102 #define rtstat	rtstatistics_percpu[mycpuid]
103 #else
104 #define rtstat	rtstatistics_percpu[0]
105 #endif
106 
107 struct radix_node_head *rt_tables[MAXCPU][AF_MAX+1];
108 struct lwkt_port *rt_ports[MAXCPU];
109 
110 static void	rt_maskedcopy (struct sockaddr *, struct sockaddr *,
111 			       struct sockaddr *);
112 static void rtable_init(void);
113 static void rtable_service_loop(void *dummy);
114 static void rtinit_rtrequest_callback(int, int, struct rt_addrinfo *,
115 				      struct rtentry *, void *);
116 
117 #ifdef SMP
118 static void rtredirect_msghandler(struct netmsg *netmsg);
119 static void rtrequest1_msghandler(struct netmsg *netmsg);
120 #endif
121 static void rtsearch_msghandler(struct netmsg *netmsg);
122 
123 static void rtmask_add_msghandler(struct netmsg *netmsg);
124 
125 static int rt_setshims(struct rtentry *, struct sockaddr **);
126 
127 SYSCTL_NODE(_net, OID_AUTO, route, CTLFLAG_RW, 0, "Routing");
128 
129 #ifdef ROUTE_DEBUG
130 static int route_debug = 1;
131 SYSCTL_INT(_net_route, OID_AUTO, route_debug, CTLFLAG_RW,
132            &route_debug, 0, "");
133 #endif
134 
135 int route_assert_owner_access = 0;
136 SYSCTL_INT(_net_route, OID_AUTO, assert_owner_access, CTLFLAG_RW,
137            &route_assert_owner_access, 0, "");
138 
139 /*
140  * Initialize the route table(s) for protocol domains and
141  * create a helper thread which will be responsible for updating
142  * route table entries on each cpu.
143  */
144 void
145 route_init(void)
146 {
147 	int cpu;
148 	thread_t rtd;
149 
150 	for (cpu = 0; cpu < ncpus; ++cpu)
151 		bzero(&rtstatistics_percpu[cpu], sizeof(struct rtstatistics));
152 	rn_init();      /* initialize all zeroes, all ones, mask table */
153 	rtable_init();	/* call dom_rtattach() on each cpu */
154 
155 	for (cpu = 0; cpu < ncpus; cpu++) {
156 		lwkt_create(rtable_service_loop, NULL, &rtd, NULL,
157 			    0, cpu, "rtable_cpu %d", cpu);
158 		rt_ports[cpu] = &rtd->td_msgport;
159 	}
160 }
161 
162 static void
163 rtable_init_oncpu(struct netmsg *nmsg)
164 {
165 	struct domain *dom;
166 	int cpu = mycpuid;
167 
168 	SLIST_FOREACH(dom, &domains, dom_next) {
169 		if (dom->dom_rtattach) {
170 			dom->dom_rtattach(
171 				(void **)&rt_tables[cpu][dom->dom_family],
172 			        dom->dom_rtoffset);
173 		}
174 	}
175 	ifnet_forwardmsg(&nmsg->nm_lmsg, cpu + 1);
176 }
177 
178 static void
179 rtable_init(void)
180 {
181 	struct netmsg nmsg;
182 
183 	netmsg_init(&nmsg, NULL, &curthread->td_msgport,
184 		    0, rtable_init_oncpu);
185 	ifnet_domsg(&nmsg.nm_lmsg, 0);
186 }
187 
188 /*
189  * Our per-cpu table management protocol thread.  All route table operations
190  * are sequentially chained through all cpus starting at cpu #0 in order to
191  * maintain duplicate route tables on each cpu.  Having a spearate route
192  * table management thread allows the protocol and interrupt threads to
193  * issue route table changes.
194  */
195 static void
196 rtable_service_loop(void *dummy __unused)
197 {
198 	struct netmsg *netmsg;
199 	thread_t td = curthread;
200 
201 	get_mplock();	/* XXX is this mpsafe yet? */
202 
203 	while ((netmsg = lwkt_waitport(&td->td_msgport, 0)) != NULL) {
204 		netmsg->nm_dispatch(netmsg);
205 	}
206 }
207 
208 /*
209  * Routing statistics.
210  */
211 #ifdef SMP
212 static int
213 sysctl_rtstatistics(SYSCTL_HANDLER_ARGS)
214 {
215 	int cpu, error = 0;
216 
217 	for (cpu = 0; cpu < ncpus; ++cpu) {
218 		if ((error = SYSCTL_OUT(req, &rtstatistics_percpu[cpu],
219 					sizeof(struct rtstatistics))))
220 				break;
221 		if ((error = SYSCTL_IN(req, &rtstatistics_percpu[cpu],
222 					sizeof(struct rtstatistics))))
223 				break;
224 	}
225 
226 	return (error);
227 }
228 SYSCTL_PROC(_net_route, OID_AUTO, stats, (CTLTYPE_OPAQUE|CTLFLAG_RW),
229 	0, 0, sysctl_rtstatistics, "S,rtstatistics", "Routing statistics");
230 #else
231 SYSCTL_STRUCT(_net_route, OID_AUTO, stats, CTLFLAG_RW, &rtstat, rtstatistics,
232 "Routing statistics");
233 #endif
234 
235 /*
236  * Packet routing routines.
237  */
238 
239 /*
240  * Look up and fill in the "ro_rt" rtentry field in a route structure given
241  * an address in the "ro_dst" field.  Always send a report on a miss and
242  * always clone routes.
243  */
244 void
245 rtalloc(struct route *ro)
246 {
247 	rtalloc_ign(ro, 0UL);
248 }
249 
250 /*
251  * Look up and fill in the "ro_rt" rtentry field in a route structure given
252  * an address in the "ro_dst" field.  Always send a report on a miss and
253  * optionally clone routes when RTF_CLONING or RTF_PRCLONING are not being
254  * ignored.
255  */
256 void
257 rtalloc_ign(struct route *ro, u_long ignoreflags)
258 {
259 	if (ro->ro_rt != NULL) {
260 		if (ro->ro_rt->rt_ifp != NULL && ro->ro_rt->rt_flags & RTF_UP)
261 			return;
262 		rtfree(ro->ro_rt);
263 		ro->ro_rt = NULL;
264 	}
265 	ro->ro_rt = _rtlookup(&ro->ro_dst, RTL_REPORTMSG, ignoreflags);
266 }
267 
268 /*
269  * Look up the route that matches the given "dst" address.
270  *
271  * Route lookup can have the side-effect of creating and returning
272  * a cloned route instead when "dst" matches a cloning route and the
273  * RTF_CLONING and RTF_PRCLONING flags are not being ignored.
274  *
275  * Any route returned has its reference count incremented.
276  */
277 struct rtentry *
278 _rtlookup(struct sockaddr *dst, boolean_t generate_report, u_long ignore)
279 {
280 	struct radix_node_head *rnh = rt_tables[mycpuid][dst->sa_family];
281 	struct rtentry *rt;
282 
283 	if (rnh == NULL)
284 		goto unreach;
285 
286 	/*
287 	 * Look up route in the radix tree.
288 	 */
289 	rt = (struct rtentry *) rnh->rnh_matchaddr((char *)dst, rnh);
290 	if (rt == NULL)
291 		goto unreach;
292 
293 	/*
294 	 * Handle cloning routes.
295 	 */
296 	if ((rt->rt_flags & ~ignore & (RTF_CLONING | RTF_PRCLONING)) != 0) {
297 		struct rtentry *clonedroute;
298 		int error;
299 
300 		clonedroute = rt;	/* copy in/copy out parameter */
301 		error = rtrequest(RTM_RESOLVE, dst, NULL, NULL, 0,
302 				  &clonedroute);	/* clone the route */
303 		if (error != 0) {	/* cloning failed */
304 			if (generate_report)
305 				rt_dstmsg(RTM_MISS, dst, error);
306 			rt->rt_refcnt++;
307 			return (rt);	/* return the uncloned route */
308 		}
309 		if (generate_report) {
310 			if (clonedroute->rt_flags & RTF_XRESOLVE)
311 				rt_dstmsg(RTM_RESOLVE, dst, 0);
312 			else
313 				rt_rtmsg(RTM_ADD, clonedroute,
314 					 clonedroute->rt_ifp, 0);
315 		}
316 		return (clonedroute);	/* return cloned route */
317 	}
318 
319 	/*
320 	 * Increment the reference count of the matched route and return.
321 	 */
322 	rt->rt_refcnt++;
323 	return (rt);
324 
325 unreach:
326 	rtstat.rts_unreach++;
327 	if (generate_report)
328 		rt_dstmsg(RTM_MISS, dst, 0);
329 	return (NULL);
330 }
331 
332 void
333 rtfree(struct rtentry *rt)
334 {
335 	if (rt->rt_cpuid == mycpuid)
336 		rtfree_oncpu(rt);
337 	else
338 		rtfree_remote(rt, 1);
339 }
340 
341 void
342 rtfree_oncpu(struct rtentry *rt)
343 {
344 	KKASSERT(rt->rt_cpuid == mycpuid);
345 	KASSERT(rt->rt_refcnt > 0, ("rtfree: rt_refcnt %ld", rt->rt_refcnt));
346 
347 	--rt->rt_refcnt;
348 	if (rt->rt_refcnt == 0) {
349 		struct radix_node_head *rnh =
350 		    rt_tables[mycpuid][rt_key(rt)->sa_family];
351 
352 		if (rnh->rnh_close)
353 			rnh->rnh_close((struct radix_node *)rt, rnh);
354 		if (!(rt->rt_flags & RTF_UP)) {
355 			/* deallocate route */
356 			if (rt->rt_ifa != NULL)
357 				IFAFREE(rt->rt_ifa);
358 			if (rt->rt_parent != NULL)
359 				RTFREE(rt->rt_parent);	/* recursive call! */
360 			Free(rt_key(rt));
361 			Free(rt);
362 		}
363 	}
364 }
365 
366 static void
367 rtfree_remote_dispatch(struct netmsg *nmsg)
368 {
369 	struct lwkt_msg *lmsg = &nmsg->nm_lmsg;
370 	struct rtentry *rt = lmsg->u.ms_resultp;
371 
372 	rtfree_oncpu(rt);
373 	lwkt_replymsg(lmsg, 0);
374 }
375 
376 void
377 rtfree_remote(struct rtentry *rt, int allow_panic)
378 {
379 	struct netmsg nmsg;
380 	struct lwkt_msg *lmsg;
381 
382 	KKASSERT(rt->rt_cpuid != mycpuid);
383 
384 	if (route_assert_owner_access && allow_panic) {
385 		panic("rt remote free rt_cpuid %d, mycpuid %d\n",
386 		      rt->rt_cpuid, mycpuid);
387 	} else {
388 		kprintf("rt remote free rt_cpuid %d, mycpuid %d\n",
389 			rt->rt_cpuid, mycpuid);
390 		print_backtrace(-1);
391 	}
392 
393 	netmsg_init(&nmsg, NULL, &curthread->td_msgport,
394 		    0, rtfree_remote_dispatch);
395 	lmsg = &nmsg.nm_lmsg;
396 	lmsg->u.ms_resultp = rt;
397 
398 	lwkt_domsg(rtable_portfn(rt->rt_cpuid), lmsg, 0);
399 }
400 
401 static int
402 rtredirect_oncpu(struct sockaddr *dst, struct sockaddr *gateway,
403 		 struct sockaddr *netmask, int flags, struct sockaddr *src)
404 {
405 	struct rtentry *rt = NULL;
406 	struct rt_addrinfo rtinfo;
407 	struct ifaddr *ifa;
408 	u_long *stat = NULL;
409 	int error;
410 
411 	/* verify the gateway is directly reachable */
412 	if ((ifa = ifa_ifwithnet(gateway)) == NULL) {
413 		error = ENETUNREACH;
414 		goto out;
415 	}
416 
417 	/*
418 	 * If the redirect isn't from our current router for this destination,
419 	 * it's either old or wrong.
420 	 */
421 	if (!(flags & RTF_DONE) &&		/* XXX JH */
422 	    (rt = rtpurelookup(dst)) != NULL &&
423 	    (!sa_equal(src, rt->rt_gateway) || rt->rt_ifa != ifa)) {
424 		error = EINVAL;
425 		goto done;
426 	}
427 
428 	/*
429 	 * If it redirects us to ourselves, we have a routing loop,
430 	 * perhaps as a result of an interface going down recently.
431 	 */
432 	if (ifa_ifwithaddr(gateway)) {
433 		error = EHOSTUNREACH;
434 		goto done;
435 	}
436 
437 	/*
438 	 * Create a new entry if the lookup failed or if we got back
439 	 * a wildcard entry for the default route.  This is necessary
440 	 * for hosts which use routing redirects generated by smart
441 	 * gateways to dynamically build the routing tables.
442 	 */
443 	if (rt == NULL)
444 		goto create;
445 	if ((rt_mask(rt) != NULL && rt_mask(rt)->sa_len < 2)) {
446 		rtfree(rt);
447 		goto create;
448 	}
449 
450 	/* Ignore redirects for directly connected hosts. */
451 	if (!(rt->rt_flags & RTF_GATEWAY)) {
452 		error = EHOSTUNREACH;
453 		goto done;
454 	}
455 
456 	if (!(rt->rt_flags & RTF_HOST) && (flags & RTF_HOST)) {
457 		/*
458 		 * Changing from a network route to a host route.
459 		 * Create a new host route rather than smashing the
460 		 * network route.
461 		 */
462 create:
463 		flags |=  RTF_GATEWAY | RTF_DYNAMIC;
464 		bzero(&rtinfo, sizeof(struct rt_addrinfo));
465 		rtinfo.rti_info[RTAX_DST] = dst;
466 		rtinfo.rti_info[RTAX_GATEWAY] = gateway;
467 		rtinfo.rti_info[RTAX_NETMASK] = netmask;
468 		rtinfo.rti_flags = flags;
469 		rtinfo.rti_ifa = ifa;
470 		rt = NULL;	/* copy-in/copy-out parameter */
471 		error = rtrequest1(RTM_ADD, &rtinfo, &rt);
472 		if (rt != NULL)
473 			flags = rt->rt_flags;
474 		stat = &rtstat.rts_dynamic;
475 	} else {
476 		/*
477 		 * Smash the current notion of the gateway to this destination.
478 		 * Should check about netmask!!!
479 		 */
480 		rt->rt_flags |= RTF_MODIFIED;
481 		flags |= RTF_MODIFIED;
482 
483 		/* We only need to report rtmsg on CPU0 */
484 		rt_setgate(rt, rt_key(rt), gateway,
485 			   mycpuid == 0 ? RTL_REPORTMSG : RTL_DONTREPORT);
486 		error = 0;
487 		stat = &rtstat.rts_newgateway;
488 	}
489 
490 done:
491 	if (rt != NULL)
492 		rtfree(rt);
493 out:
494 	if (error != 0)
495 		rtstat.rts_badredirect++;
496 	else if (stat != NULL)
497 		(*stat)++;
498 
499 	return error;
500 }
501 
502 #ifdef SMP
503 
504 struct netmsg_rtredirect {
505 	struct netmsg	netmsg;
506 	struct sockaddr *dst;
507 	struct sockaddr *gateway;
508 	struct sockaddr *netmask;
509 	int		flags;
510 	struct sockaddr *src;
511 };
512 
513 #endif
514 
515 /*
516  * Force a routing table entry to the specified
517  * destination to go through the given gateway.
518  * Normally called as a result of a routing redirect
519  * message from the network layer.
520  *
521  * N.B.: must be called at splnet
522  */
523 void
524 rtredirect(struct sockaddr *dst, struct sockaddr *gateway,
525 	   struct sockaddr *netmask, int flags, struct sockaddr *src)
526 {
527 	struct rt_addrinfo rtinfo;
528 	int error;
529 #ifdef SMP
530 	struct netmsg_rtredirect msg;
531 
532 	netmsg_init(&msg.netmsg, NULL, &curthread->td_msgport,
533 		    0, rtredirect_msghandler);
534 	msg.dst = dst;
535 	msg.gateway = gateway;
536 	msg.netmask = netmask;
537 	msg.flags = flags;
538 	msg.src = src;
539 	error = lwkt_domsg(rtable_portfn(0), &msg.netmsg.nm_lmsg, 0);
540 #else
541 	error = rtredirect_oncpu(dst, gateway, netmask, flags, src);
542 #endif
543 	bzero(&rtinfo, sizeof(struct rt_addrinfo));
544 	rtinfo.rti_info[RTAX_DST] = dst;
545 	rtinfo.rti_info[RTAX_GATEWAY] = gateway;
546 	rtinfo.rti_info[RTAX_NETMASK] = netmask;
547 	rtinfo.rti_info[RTAX_AUTHOR] = src;
548 	rt_missmsg(RTM_REDIRECT, &rtinfo, flags, error);
549 }
550 
551 #ifdef SMP
552 
553 static void
554 rtredirect_msghandler(struct netmsg *netmsg)
555 {
556 	struct netmsg_rtredirect *msg = (void *)netmsg;
557 	int nextcpu;
558 
559 	rtredirect_oncpu(msg->dst, msg->gateway, msg->netmask,
560 			 msg->flags, msg->src);
561 	nextcpu = mycpuid + 1;
562 	if (nextcpu < ncpus)
563 		lwkt_forwardmsg(rtable_portfn(nextcpu), &netmsg->nm_lmsg);
564 	else
565 		lwkt_replymsg(&netmsg->nm_lmsg, 0);
566 }
567 
568 #endif
569 
570 /*
571 * Routing table ioctl interface.
572 */
573 int
574 rtioctl(u_long req, caddr_t data, struct ucred *cred)
575 {
576 #ifdef INET
577 	/* Multicast goop, grrr... */
578 	return mrt_ioctl ? mrt_ioctl(req, data) : EOPNOTSUPP;
579 #else
580 	return ENXIO;
581 #endif
582 }
583 
584 struct ifaddr *
585 ifa_ifwithroute(int flags, struct sockaddr *dst, struct sockaddr *gateway)
586 {
587 	struct ifaddr *ifa;
588 
589 	if (!(flags & RTF_GATEWAY)) {
590 		/*
591 		 * If we are adding a route to an interface,
592 		 * and the interface is a point-to-point link,
593 		 * we should search for the destination
594 		 * as our clue to the interface.  Otherwise
595 		 * we can use the local address.
596 		 */
597 		ifa = NULL;
598 		if (flags & RTF_HOST) {
599 			ifa = ifa_ifwithdstaddr(dst);
600 		}
601 		if (ifa == NULL)
602 			ifa = ifa_ifwithaddr(gateway);
603 	} else {
604 		/*
605 		 * If we are adding a route to a remote net
606 		 * or host, the gateway may still be on the
607 		 * other end of a pt to pt link.
608 		 */
609 		ifa = ifa_ifwithdstaddr(gateway);
610 	}
611 	if (ifa == NULL)
612 		ifa = ifa_ifwithnet(gateway);
613 	if (ifa == NULL) {
614 		struct rtentry *rt;
615 
616 		rt = rtpurelookup(gateway);
617 		if (rt == NULL)
618 			return (NULL);
619 		rt->rt_refcnt--;
620 		if ((ifa = rt->rt_ifa) == NULL)
621 			return (NULL);
622 	}
623 	if (ifa->ifa_addr->sa_family != dst->sa_family) {
624 		struct ifaddr *oldifa = ifa;
625 
626 		ifa = ifaof_ifpforaddr(dst, ifa->ifa_ifp);
627 		if (ifa == NULL)
628 			ifa = oldifa;
629 	}
630 	return (ifa);
631 }
632 
633 static int rt_fixdelete (struct radix_node *, void *);
634 static int rt_fixchange (struct radix_node *, void *);
635 
636 struct rtfc_arg {
637 	struct rtentry *rt0;
638 	struct radix_node_head *rnh;
639 };
640 
641 /*
642  * Set rtinfo->rti_ifa and rtinfo->rti_ifp.
643  */
644 int
645 rt_getifa(struct rt_addrinfo *rtinfo)
646 {
647 	struct sockaddr *gateway = rtinfo->rti_info[RTAX_GATEWAY];
648 	struct sockaddr *dst = rtinfo->rti_info[RTAX_DST];
649 	struct sockaddr *ifaaddr = rtinfo->rti_info[RTAX_IFA];
650 	int flags = rtinfo->rti_flags;
651 
652 	/*
653 	 * ifp may be specified by sockaddr_dl
654 	 * when protocol address is ambiguous.
655 	 */
656 	if (rtinfo->rti_ifp == NULL) {
657 		struct sockaddr *ifpaddr;
658 
659 		ifpaddr = rtinfo->rti_info[RTAX_IFP];
660 		if (ifpaddr != NULL && ifpaddr->sa_family == AF_LINK) {
661 			struct ifaddr *ifa;
662 
663 			ifa = ifa_ifwithnet(ifpaddr);
664 			if (ifa != NULL)
665 				rtinfo->rti_ifp = ifa->ifa_ifp;
666 		}
667 	}
668 
669 	if (rtinfo->rti_ifa == NULL && ifaaddr != NULL)
670 		rtinfo->rti_ifa = ifa_ifwithaddr(ifaaddr);
671 	if (rtinfo->rti_ifa == NULL) {
672 		struct sockaddr *sa;
673 
674 		sa = ifaaddr != NULL ? ifaaddr :
675 		    (gateway != NULL ? gateway : dst);
676 		if (sa != NULL && rtinfo->rti_ifp != NULL)
677 			rtinfo->rti_ifa = ifaof_ifpforaddr(sa, rtinfo->rti_ifp);
678 		else if (dst != NULL && gateway != NULL)
679 			rtinfo->rti_ifa = ifa_ifwithroute(flags, dst, gateway);
680 		else if (sa != NULL)
681 			rtinfo->rti_ifa = ifa_ifwithroute(flags, sa, sa);
682 	}
683 	if (rtinfo->rti_ifa == NULL)
684 		return (ENETUNREACH);
685 
686 	if (rtinfo->rti_ifp == NULL)
687 		rtinfo->rti_ifp = rtinfo->rti_ifa->ifa_ifp;
688 	return (0);
689 }
690 
691 /*
692  * Do appropriate manipulations of a routing tree given
693  * all the bits of info needed
694  */
695 int
696 rtrequest(
697 	int req,
698 	struct sockaddr *dst,
699 	struct sockaddr *gateway,
700 	struct sockaddr *netmask,
701 	int flags,
702 	struct rtentry **ret_nrt)
703 {
704 	struct rt_addrinfo rtinfo;
705 
706 	bzero(&rtinfo, sizeof(struct rt_addrinfo));
707 	rtinfo.rti_info[RTAX_DST] = dst;
708 	rtinfo.rti_info[RTAX_GATEWAY] = gateway;
709 	rtinfo.rti_info[RTAX_NETMASK] = netmask;
710 	rtinfo.rti_flags = flags;
711 	return rtrequest1(req, &rtinfo, ret_nrt);
712 }
713 
714 int
715 rtrequest_global(
716 	int req,
717 	struct sockaddr *dst,
718 	struct sockaddr *gateway,
719 	struct sockaddr *netmask,
720 	int flags)
721 {
722 	struct rt_addrinfo rtinfo;
723 
724 	bzero(&rtinfo, sizeof(struct rt_addrinfo));
725 	rtinfo.rti_info[RTAX_DST] = dst;
726 	rtinfo.rti_info[RTAX_GATEWAY] = gateway;
727 	rtinfo.rti_info[RTAX_NETMASK] = netmask;
728 	rtinfo.rti_flags = flags;
729 	return rtrequest1_global(req, &rtinfo, NULL, NULL);
730 }
731 
732 #ifdef SMP
733 
734 struct netmsg_rtq {
735 	struct netmsg		netmsg;
736 	int			req;
737 	struct rt_addrinfo	*rtinfo;
738 	rtrequest1_callback_func_t callback;
739 	void			*arg;
740 };
741 
742 #endif
743 
744 int
745 rtrequest1_global(int req, struct rt_addrinfo *rtinfo,
746 		  rtrequest1_callback_func_t callback, void *arg)
747 {
748 	int error;
749 #ifdef SMP
750 	struct netmsg_rtq msg;
751 
752 	netmsg_init(&msg.netmsg, NULL, &curthread->td_msgport,
753 		    0, rtrequest1_msghandler);
754 	msg.netmsg.nm_lmsg.ms_error = -1;
755 	msg.req = req;
756 	msg.rtinfo = rtinfo;
757 	msg.callback = callback;
758 	msg.arg = arg;
759 	error = lwkt_domsg(rtable_portfn(0), &msg.netmsg.nm_lmsg, 0);
760 #else
761 	struct rtentry *rt = NULL;
762 
763 	error = rtrequest1(req, rtinfo, &rt);
764 	if (rt)
765 		--rt->rt_refcnt;
766 	if (callback)
767 		callback(req, error, rtinfo, rt, arg);
768 #endif
769 	return (error);
770 }
771 
772 /*
773  * Handle a route table request on the current cpu.  Since the route table's
774  * are supposed to be identical on each cpu, an error occuring later in the
775  * message chain is considered system-fatal.
776  */
777 #ifdef SMP
778 
779 static void
780 rtrequest1_msghandler(struct netmsg *netmsg)
781 {
782 	struct netmsg_rtq *msg = (void *)netmsg;
783 	struct rt_addrinfo rtinfo;
784 	struct rtentry *rt = NULL;
785 	int nextcpu;
786 	int error;
787 
788 	/*
789 	 * Copy the rtinfo.  We need to make sure that the original
790 	 * rtinfo, which is setup by the caller, in the netmsg will
791 	 * _not_ be changed; else the next CPU on the netmsg forwarding
792 	 * path will see a different rtinfo than what this CPU has seen.
793 	 */
794 	rtinfo = *msg->rtinfo;
795 
796 	error = rtrequest1(msg->req, &rtinfo, &rt);
797 	if (rt)
798 		--rt->rt_refcnt;
799 	if (msg->callback)
800 		msg->callback(msg->req, error, &rtinfo, rt, msg->arg);
801 
802 	/*
803 	 * RTM_DELETE's are propogated even if an error occurs, since a
804 	 * cloned route might be undergoing deletion and cloned routes
805 	 * are not necessarily replicated.  An overall error is returned
806 	 * only if no cpus have the route in question.
807 	 */
808 	if (msg->netmsg.nm_lmsg.ms_error < 0 || error == 0)
809 		msg->netmsg.nm_lmsg.ms_error = error;
810 
811 	nextcpu = mycpuid + 1;
812 	if (error && msg->req != RTM_DELETE) {
813 		if (mycpuid != 0) {
814 			panic("rtrequest1_msghandler: rtrequest table "
815 			      "error was not on cpu #0");
816 		}
817 		lwkt_replymsg(&msg->netmsg.nm_lmsg, error);
818 	} else if (nextcpu < ncpus) {
819 		lwkt_forwardmsg(rtable_portfn(nextcpu), &msg->netmsg.nm_lmsg);
820 	} else {
821 		lwkt_replymsg(&msg->netmsg.nm_lmsg,
822 			      msg->netmsg.nm_lmsg.ms_error);
823 	}
824 }
825 
826 #endif
827 
828 int
829 rtrequest1(int req, struct rt_addrinfo *rtinfo, struct rtentry **ret_nrt)
830 {
831 	struct sockaddr *dst = rtinfo->rti_info[RTAX_DST];
832 	struct rtentry *rt;
833 	struct radix_node *rn;
834 	struct radix_node_head *rnh;
835 	struct ifaddr *ifa;
836 	struct sockaddr *ndst;
837 	boolean_t reportmsg;
838 	int error = 0;
839 
840 #define gotoerr(x) { error = x ; goto bad; }
841 
842 #ifdef ROUTE_DEBUG
843 	if (route_debug)
844 		rt_addrinfo_print(req, rtinfo);
845 #endif
846 
847 	crit_enter();
848 	/*
849 	 * Find the correct routing tree to use for this Address Family
850 	 */
851 	if ((rnh = rt_tables[mycpuid][dst->sa_family]) == NULL)
852 		gotoerr(EAFNOSUPPORT);
853 
854 	/*
855 	 * If we are adding a host route then we don't want to put
856 	 * a netmask in the tree, nor do we want to clone it.
857 	 */
858 	if (rtinfo->rti_flags & RTF_HOST) {
859 		rtinfo->rti_info[RTAX_NETMASK] = NULL;
860 		rtinfo->rti_flags &= ~(RTF_CLONING | RTF_PRCLONING);
861 	}
862 
863 	switch (req) {
864 	case RTM_DELETE:
865 		/* Remove the item from the tree. */
866 		rn = rnh->rnh_deladdr((char *)rtinfo->rti_info[RTAX_DST],
867 				      (char *)rtinfo->rti_info[RTAX_NETMASK],
868 				      rnh);
869 		if (rn == NULL)
870 			gotoerr(ESRCH);
871 		KASSERT(!(rn->rn_flags & (RNF_ACTIVE | RNF_ROOT)),
872 			("rnh_deladdr returned flags 0x%x", rn->rn_flags));
873 		rt = (struct rtentry *)rn;
874 
875 		/* ref to prevent a deletion race */
876 		++rt->rt_refcnt;
877 
878 		/* Free any routes cloned from this one. */
879 		if ((rt->rt_flags & (RTF_CLONING | RTF_PRCLONING)) &&
880 		    rt_mask(rt) != NULL) {
881 			rnh->rnh_walktree_from(rnh, (char *)rt_key(rt),
882 					       (char *)rt_mask(rt),
883 					       rt_fixdelete, rt);
884 		}
885 
886 		if (rt->rt_gwroute != NULL) {
887 			RTFREE(rt->rt_gwroute);
888 			rt->rt_gwroute = NULL;
889 		}
890 
891 		/*
892 		 * NB: RTF_UP must be set during the search above,
893 		 * because we might delete the last ref, causing
894 		 * rt to get freed prematurely.
895 		 */
896 		rt->rt_flags &= ~RTF_UP;
897 
898 #ifdef ROUTE_DEBUG
899 		if (route_debug)
900 			rt_print(rtinfo, rt);
901 #endif
902 
903 		/* Give the protocol a chance to keep things in sync. */
904 		if ((ifa = rt->rt_ifa) && ifa->ifa_rtrequest)
905 			ifa->ifa_rtrequest(RTM_DELETE, rt, rtinfo);
906 
907 		/*
908 		 * If the caller wants it, then it can have it,
909 		 * but it's up to it to free the rtentry as we won't be
910 		 * doing it.
911 		 */
912 		KASSERT(rt->rt_refcnt >= 0,
913 			("rtrequest1(DELETE): refcnt %ld", rt->rt_refcnt));
914 		if (ret_nrt != NULL) {
915 			/* leave ref intact for return */
916 			*ret_nrt = rt;
917 		} else {
918 			/* deref / attempt to destroy */
919 			rtfree(rt);
920 		}
921 		break;
922 
923 	case RTM_RESOLVE:
924 		if (ret_nrt == NULL || (rt = *ret_nrt) == NULL)
925 			gotoerr(EINVAL);
926 		ifa = rt->rt_ifa;
927 		rtinfo->rti_flags =
928 		    rt->rt_flags & ~(RTF_CLONING | RTF_PRCLONING | RTF_STATIC);
929 		rtinfo->rti_flags |= RTF_WASCLONED;
930 		rtinfo->rti_info[RTAX_GATEWAY] = rt->rt_gateway;
931 		if ((rtinfo->rti_info[RTAX_NETMASK] = rt->rt_genmask) == NULL)
932 			rtinfo->rti_flags |= RTF_HOST;
933 		rtinfo->rti_info[RTAX_MPLS1] = rt->rt_shim[0];
934 		rtinfo->rti_info[RTAX_MPLS2] = rt->rt_shim[1];
935 		rtinfo->rti_info[RTAX_MPLS3] = rt->rt_shim[2];
936 		goto makeroute;
937 
938 	case RTM_ADD:
939 		KASSERT(!(rtinfo->rti_flags & RTF_GATEWAY) ||
940 			rtinfo->rti_info[RTAX_GATEWAY] != NULL,
941 		    ("rtrequest: GATEWAY but no gateway"));
942 
943 		if (rtinfo->rti_ifa == NULL && (error = rt_getifa(rtinfo)))
944 			gotoerr(error);
945 		ifa = rtinfo->rti_ifa;
946 makeroute:
947 		R_Malloc(rt, struct rtentry *, sizeof(struct rtentry));
948 		if (rt == NULL)
949 			gotoerr(ENOBUFS);
950 		bzero(rt, sizeof(struct rtentry));
951 		rt->rt_flags = RTF_UP | rtinfo->rti_flags;
952 		rt->rt_cpuid = mycpuid;
953 
954 		if (mycpuid != 0 && req == RTM_ADD) {
955 			/* For RTM_ADD, we have already sent rtmsg on CPU0. */
956 			reportmsg = RTL_DONTREPORT;
957 		} else {
958 			/*
959 			 * For RTM_ADD, we only send rtmsg on CPU0.
960 			 * For RTM_RESOLVE, we always send rtmsg. XXX
961 			 */
962 			reportmsg = RTL_REPORTMSG;
963 		}
964 		error = rt_setgate(rt, dst, rtinfo->rti_info[RTAX_GATEWAY],
965 				   reportmsg);
966 		if (error != 0) {
967 			Free(rt);
968 			gotoerr(error);
969 		}
970 
971 		ndst = rt_key(rt);
972 		if (rtinfo->rti_info[RTAX_NETMASK] != NULL)
973 			rt_maskedcopy(dst, ndst,
974 				      rtinfo->rti_info[RTAX_NETMASK]);
975 		else
976 			bcopy(dst, ndst, dst->sa_len);
977 
978 		if (rtinfo->rti_info[RTAX_MPLS1] != NULL)
979 			rt_setshims(rt, rtinfo->rti_info);
980 
981 		/*
982 		 * Note that we now have a reference to the ifa.
983 		 * This moved from below so that rnh->rnh_addaddr() can
984 		 * examine the ifa and  ifa->ifa_ifp if it so desires.
985 		 */
986 		IFAREF(ifa);
987 		rt->rt_ifa = ifa;
988 		rt->rt_ifp = ifa->ifa_ifp;
989 		/* XXX mtu manipulation will be done in rnh_addaddr -- itojun */
990 
991 		rn = rnh->rnh_addaddr((char *)ndst,
992 				      (char *)rtinfo->rti_info[RTAX_NETMASK],
993 				      rnh, rt->rt_nodes);
994 		if (rn == NULL) {
995 			struct rtentry *oldrt;
996 
997 			/*
998 			 * We already have one of these in the tree.
999 			 * We do a special hack: if the old route was
1000 			 * cloned, then we blow it away and try
1001 			 * re-inserting the new one.
1002 			 */
1003 			oldrt = rtpurelookup(ndst);
1004 			if (oldrt != NULL) {
1005 				--oldrt->rt_refcnt;
1006 				if (oldrt->rt_flags & RTF_WASCLONED) {
1007 					rtrequest(RTM_DELETE, rt_key(oldrt),
1008 						  oldrt->rt_gateway,
1009 						  rt_mask(oldrt),
1010 						  oldrt->rt_flags, NULL);
1011 					rn = rnh->rnh_addaddr((char *)ndst,
1012 					    (char *)
1013 						rtinfo->rti_info[RTAX_NETMASK],
1014 					    rnh, rt->rt_nodes);
1015 				}
1016 			}
1017 		}
1018 
1019 		/*
1020 		 * If it still failed to go into the tree,
1021 		 * then un-make it (this should be a function).
1022 		 */
1023 		if (rn == NULL) {
1024 			if (rt->rt_gwroute != NULL)
1025 				rtfree(rt->rt_gwroute);
1026 			IFAFREE(ifa);
1027 			Free(rt_key(rt));
1028 			Free(rt);
1029 			gotoerr(EEXIST);
1030 		}
1031 
1032 		/*
1033 		 * If we got here from RESOLVE, then we are cloning
1034 		 * so clone the rest, and note that we
1035 		 * are a clone (and increment the parent's references)
1036 		 */
1037 		if (req == RTM_RESOLVE) {
1038 			rt->rt_rmx = (*ret_nrt)->rt_rmx;    /* copy metrics */
1039 			rt->rt_rmx.rmx_pksent = 0;  /* reset packet counter */
1040 			if ((*ret_nrt)->rt_flags &
1041 				       (RTF_CLONING | RTF_PRCLONING)) {
1042 				rt->rt_parent = *ret_nrt;
1043 				(*ret_nrt)->rt_refcnt++;
1044 			}
1045 		}
1046 
1047 		/*
1048 		 * if this protocol has something to add to this then
1049 		 * allow it to do that as well.
1050 		 */
1051 		if (ifa->ifa_rtrequest != NULL)
1052 			ifa->ifa_rtrequest(req, rt, rtinfo);
1053 
1054 		/*
1055 		 * We repeat the same procedure from rt_setgate() here because
1056 		 * it doesn't fire when we call it there because the node
1057 		 * hasn't been added to the tree yet.
1058 		 */
1059 		if (req == RTM_ADD && !(rt->rt_flags & RTF_HOST) &&
1060 		    rt_mask(rt) != NULL) {
1061 			struct rtfc_arg arg = { rt, rnh };
1062 
1063 			rnh->rnh_walktree_from(rnh, (char *)rt_key(rt),
1064 					       (char *)rt_mask(rt),
1065 					       rt_fixchange, &arg);
1066 		}
1067 
1068 #ifdef ROUTE_DEBUG
1069 		if (route_debug)
1070 			rt_print(rtinfo, rt);
1071 #endif
1072 		/*
1073 		 * Return the resulting rtentry,
1074 		 * increasing the number of references by one.
1075 		 */
1076 		if (ret_nrt != NULL) {
1077 			rt->rt_refcnt++;
1078 			*ret_nrt = rt;
1079 		}
1080 		break;
1081 	default:
1082 		error = EOPNOTSUPP;
1083 	}
1084 bad:
1085 #ifdef ROUTE_DEBUG
1086 	if (route_debug) {
1087 		if (error)
1088 			kprintf("rti %p failed error %d\n", rtinfo, error);
1089 		else
1090 			kprintf("rti %p succeeded\n", rtinfo);
1091 	}
1092 #endif
1093 	crit_exit();
1094 	return (error);
1095 }
1096 
1097 /*
1098  * Called from rtrequest(RTM_DELETE, ...) to fix up the route's ``family''
1099  * (i.e., the routes related to it by the operation of cloning).  This
1100  * routine is iterated over all potential former-child-routes by way of
1101  * rnh->rnh_walktree_from() above, and those that actually are children of
1102  * the late parent (passed in as VP here) are themselves deleted.
1103  */
1104 static int
1105 rt_fixdelete(struct radix_node *rn, void *vp)
1106 {
1107 	struct rtentry *rt = (struct rtentry *)rn;
1108 	struct rtentry *rt0 = vp;
1109 
1110 	if (rt->rt_parent == rt0 &&
1111 	    !(rt->rt_flags & (RTF_PINNED | RTF_CLONING | RTF_PRCLONING))) {
1112 		return rtrequest(RTM_DELETE, rt_key(rt), NULL, rt_mask(rt),
1113 				 rt->rt_flags, NULL);
1114 	}
1115 	return 0;
1116 }
1117 
1118 /*
1119  * This routine is called from rt_setgate() to do the analogous thing for
1120  * adds and changes.  There is the added complication in this case of a
1121  * middle insert; i.e., insertion of a new network route between an older
1122  * network route and (cloned) host routes.  For this reason, a simple check
1123  * of rt->rt_parent is insufficient; each candidate route must be tested
1124  * against the (mask, value) of the new route (passed as before in vp)
1125  * to see if the new route matches it.
1126  *
1127  * XXX - it may be possible to do fixdelete() for changes and reserve this
1128  * routine just for adds.  I'm not sure why I thought it was necessary to do
1129  * changes this way.
1130  */
1131 #ifdef DEBUG
1132 static int rtfcdebug = 0;
1133 #endif
1134 
1135 static int
1136 rt_fixchange(struct radix_node *rn, void *vp)
1137 {
1138 	struct rtentry *rt = (struct rtentry *)rn;
1139 	struct rtfc_arg *ap = vp;
1140 	struct rtentry *rt0 = ap->rt0;
1141 	struct radix_node_head *rnh = ap->rnh;
1142 	u_char *xk1, *xm1, *xk2, *xmp;
1143 	int i, len, mlen;
1144 
1145 #ifdef DEBUG
1146 	if (rtfcdebug)
1147 		kprintf("rt_fixchange: rt %p, rt0 %p\n", rt, rt0);
1148 #endif
1149 
1150 	if (rt->rt_parent == NULL ||
1151 	    (rt->rt_flags & (RTF_PINNED | RTF_CLONING | RTF_PRCLONING))) {
1152 #ifdef DEBUG
1153 		if (rtfcdebug) kprintf("no parent, pinned or cloning\n");
1154 #endif
1155 		return 0;
1156 	}
1157 
1158 	if (rt->rt_parent == rt0) {
1159 #ifdef DEBUG
1160 		if (rtfcdebug) kprintf("parent match\n");
1161 #endif
1162 		return rtrequest(RTM_DELETE, rt_key(rt), NULL, rt_mask(rt),
1163 				 rt->rt_flags, NULL);
1164 	}
1165 
1166 	/*
1167 	 * There probably is a function somewhere which does this...
1168 	 * if not, there should be.
1169 	 */
1170 	len = imin(rt_key(rt0)->sa_len, rt_key(rt)->sa_len);
1171 
1172 	xk1 = (u_char *)rt_key(rt0);
1173 	xm1 = (u_char *)rt_mask(rt0);
1174 	xk2 = (u_char *)rt_key(rt);
1175 
1176 	/* avoid applying a less specific route */
1177 	xmp = (u_char *)rt_mask(rt->rt_parent);
1178 	mlen = rt_key(rt->rt_parent)->sa_len;
1179 	if (mlen > rt_key(rt0)->sa_len) {
1180 #ifdef DEBUG
1181 		if (rtfcdebug)
1182 			kprintf("rt_fixchange: inserting a less "
1183 			       "specific route\n");
1184 #endif
1185 		return 0;
1186 	}
1187 	for (i = rnh->rnh_treetop->rn_offset; i < mlen; i++) {
1188 		if ((xmp[i] & ~(xmp[i] ^ xm1[i])) != xmp[i]) {
1189 #ifdef DEBUG
1190 			if (rtfcdebug)
1191 				kprintf("rt_fixchange: inserting a less "
1192 				       "specific route\n");
1193 #endif
1194 			return 0;
1195 		}
1196 	}
1197 
1198 	for (i = rnh->rnh_treetop->rn_offset; i < len; i++) {
1199 		if ((xk2[i] & xm1[i]) != xk1[i]) {
1200 #ifdef DEBUG
1201 			if (rtfcdebug) kprintf("no match\n");
1202 #endif
1203 			return 0;
1204 		}
1205 	}
1206 
1207 	/*
1208 	 * OK, this node is a clone, and matches the node currently being
1209 	 * changed/added under the node's mask.  So, get rid of it.
1210 	 */
1211 #ifdef DEBUG
1212 	if (rtfcdebug) kprintf("deleting\n");
1213 #endif
1214 	return rtrequest(RTM_DELETE, rt_key(rt), NULL, rt_mask(rt),
1215 			 rt->rt_flags, NULL);
1216 }
1217 
1218 #define ROUNDUP(a) (a>0 ? (1 + (((a) - 1) | (sizeof(long) - 1))) : sizeof(long))
1219 
1220 int
1221 rt_setgate(struct rtentry *rt0, struct sockaddr *dst, struct sockaddr *gate,
1222 	   boolean_t generate_report)
1223 {
1224 	char *space, *oldspace;
1225 	int dlen = ROUNDUP(dst->sa_len), glen = ROUNDUP(gate->sa_len);
1226 	struct rtentry *rt = rt0;
1227 	struct radix_node_head *rnh = rt_tables[mycpuid][dst->sa_family];
1228 
1229 	/*
1230 	 * A host route with the destination equal to the gateway
1231 	 * will interfere with keeping LLINFO in the routing
1232 	 * table, so disallow it.
1233 	 */
1234 	if (((rt0->rt_flags & (RTF_HOST | RTF_GATEWAY | RTF_LLINFO)) ==
1235 			      (RTF_HOST | RTF_GATEWAY)) &&
1236 	    dst->sa_len == gate->sa_len &&
1237 	    sa_equal(dst, gate)) {
1238 		/*
1239 		 * The route might already exist if this is an RTM_CHANGE
1240 		 * or a routing redirect, so try to delete it.
1241 		 */
1242 		if (rt_key(rt0) != NULL)
1243 			rtrequest(RTM_DELETE, rt_key(rt0), rt0->rt_gateway,
1244 				  rt_mask(rt0), rt0->rt_flags, NULL);
1245 		return EADDRNOTAVAIL;
1246 	}
1247 
1248 	/*
1249 	 * Both dst and gateway are stored in the same malloc'ed chunk
1250 	 * (If I ever get my hands on....)
1251 	 * if we need to malloc a new chunk, then keep the old one around
1252 	 * till we don't need it any more.
1253 	 */
1254 	if (rt->rt_gateway == NULL || glen > ROUNDUP(rt->rt_gateway->sa_len)) {
1255 		oldspace = (char *)rt_key(rt);
1256 		R_Malloc(space, char *, dlen + glen);
1257 		if (space == NULL)
1258 			return ENOBUFS;
1259 		rt->rt_nodes->rn_key = space;
1260 	} else {
1261 		space = (char *)rt_key(rt);	/* Just use the old space. */
1262 		oldspace = NULL;
1263 	}
1264 
1265 	/* Set the gateway value. */
1266 	rt->rt_gateway = (struct sockaddr *)(space + dlen);
1267 	bcopy(gate, rt->rt_gateway, glen);
1268 
1269 	if (oldspace != NULL) {
1270 		/*
1271 		 * If we allocated a new chunk, preserve the original dst.
1272 		 * This way, rt_setgate() really just sets the gate
1273 		 * and leaves the dst field alone.
1274 		 */
1275 		bcopy(dst, space, dlen);
1276 		Free(oldspace);
1277 	}
1278 
1279 	/*
1280 	 * If there is already a gwroute, it's now almost definitely wrong
1281 	 * so drop it.
1282 	 */
1283 	if (rt->rt_gwroute != NULL) {
1284 		RTFREE(rt->rt_gwroute);
1285 		rt->rt_gwroute = NULL;
1286 	}
1287 	if (rt->rt_flags & RTF_GATEWAY) {
1288 		/*
1289 		 * Cloning loop avoidance: In the presence of
1290 		 * protocol-cloning and bad configuration, it is
1291 		 * possible to get stuck in bottomless mutual recursion
1292 		 * (rtrequest rt_setgate rtlookup).  We avoid this
1293 		 * by not allowing protocol-cloning to operate for
1294 		 * gateways (which is probably the correct choice
1295 		 * anyway), and avoid the resulting reference loops
1296 		 * by disallowing any route to run through itself as
1297 		 * a gateway.  This is obviously mandatory when we
1298 		 * get rt->rt_output().
1299 		 *
1300 		 * This breaks TTCP for hosts outside the gateway!  XXX JH
1301 		 */
1302 		rt->rt_gwroute = _rtlookup(gate, generate_report,
1303 					   RTF_PRCLONING);
1304 		if (rt->rt_gwroute == rt) {
1305 			rt->rt_gwroute = NULL;
1306 			--rt->rt_refcnt;
1307 			return EDQUOT; /* failure */
1308 		}
1309 	}
1310 
1311 	/*
1312 	 * This isn't going to do anything useful for host routes, so
1313 	 * don't bother.  Also make sure we have a reasonable mask
1314 	 * (we don't yet have one during adds).
1315 	 */
1316 	if (!(rt->rt_flags & RTF_HOST) && rt_mask(rt) != NULL) {
1317 		struct rtfc_arg arg = { rt, rnh };
1318 
1319 		rnh->rnh_walktree_from(rnh, (char *)rt_key(rt),
1320 				       (char *)rt_mask(rt),
1321 				       rt_fixchange, &arg);
1322 	}
1323 
1324 	return 0;
1325 }
1326 
1327 static void
1328 rt_maskedcopy(
1329 	struct sockaddr *src,
1330 	struct sockaddr *dst,
1331 	struct sockaddr *netmask)
1332 {
1333 	u_char *cp1 = (u_char *)src;
1334 	u_char *cp2 = (u_char *)dst;
1335 	u_char *cp3 = (u_char *)netmask;
1336 	u_char *cplim = cp2 + *cp3;
1337 	u_char *cplim2 = cp2 + *cp1;
1338 
1339 	*cp2++ = *cp1++; *cp2++ = *cp1++; /* copies sa_len & sa_family */
1340 	cp3 += 2;
1341 	if (cplim > cplim2)
1342 		cplim = cplim2;
1343 	while (cp2 < cplim)
1344 		*cp2++ = *cp1++ & *cp3++;
1345 	if (cp2 < cplim2)
1346 		bzero(cp2, cplim2 - cp2);
1347 }
1348 
1349 int
1350 rt_llroute(struct sockaddr *dst, struct rtentry *rt0, struct rtentry **drt)
1351 {
1352 	struct rtentry *up_rt, *rt;
1353 
1354 	if (!(rt0->rt_flags & RTF_UP)) {
1355 		up_rt = rtlookup(dst);
1356 		if (up_rt == NULL)
1357 			return (EHOSTUNREACH);
1358 		up_rt->rt_refcnt--;
1359 	} else
1360 		up_rt = rt0;
1361 	if (up_rt->rt_flags & RTF_GATEWAY) {
1362 		if (up_rt->rt_gwroute == NULL) {
1363 			up_rt->rt_gwroute = rtlookup(up_rt->rt_gateway);
1364 			if (up_rt->rt_gwroute == NULL)
1365 				return (EHOSTUNREACH);
1366 		} else if (!(up_rt->rt_gwroute->rt_flags & RTF_UP)) {
1367 			rtfree(up_rt->rt_gwroute);
1368 			up_rt->rt_gwroute = rtlookup(up_rt->rt_gateway);
1369 			if (up_rt->rt_gwroute == NULL)
1370 				return (EHOSTUNREACH);
1371 		}
1372 		rt = up_rt->rt_gwroute;
1373 	} else
1374 		rt = up_rt;
1375 	if (rt->rt_flags & RTF_REJECT &&
1376 	    (rt->rt_rmx.rmx_expire == 0 ||		/* rt doesn't expire */
1377 	     time_second < rt->rt_rmx.rmx_expire))	/* rt not expired */
1378 		return (rt->rt_flags & RTF_HOST ?  EHOSTDOWN : EHOSTUNREACH);
1379 	*drt = rt;
1380 	return 0;
1381 }
1382 
1383 static int
1384 rt_setshims(struct rtentry *rt, struct sockaddr **rt_shim){
1385 	int i;
1386 
1387 	for (i=0; i<3; i++) {
1388 		struct sockaddr *shim = rt_shim[RTAX_MPLS1 + i];
1389 		int shimlen;
1390 
1391 		if (shim == NULL)
1392 			break;
1393 
1394 		shimlen = ROUNDUP(shim->sa_len);
1395 		R_Malloc(rt->rt_shim[i], struct sockaddr *, shimlen);
1396 		bcopy(shim, rt->rt_shim[i], shimlen);
1397 	}
1398 
1399 	return 0;
1400 }
1401 
1402 #ifdef ROUTE_DEBUG
1403 
1404 /*
1405  * Print out a route table entry
1406  */
1407 void
1408 rt_print(struct rt_addrinfo *rtinfo, struct rtentry *rn)
1409 {
1410 	kprintf("rti %p cpu %d route %p flags %08lx: ",
1411 		rtinfo, mycpuid, rn, rn->rt_flags);
1412 	sockaddr_print(rt_key(rn));
1413 	kprintf(" mask ");
1414 	sockaddr_print(rt_mask(rn));
1415 	kprintf(" gw ");
1416 	sockaddr_print(rn->rt_gateway);
1417 	kprintf(" ifc \"%s\"", rn->rt_ifp ? rn->rt_ifp->if_dname : "?");
1418 	kprintf(" ifa %p\n", rn->rt_ifa);
1419 }
1420 
1421 void
1422 rt_addrinfo_print(int cmd, struct rt_addrinfo *rti)
1423 {
1424 	int didit = 0;
1425 	int i;
1426 
1427 #ifdef ROUTE_DEBUG
1428 	if (cmd == RTM_DELETE && route_debug > 1)
1429 		print_backtrace(-1);
1430 #endif
1431 
1432 	switch(cmd) {
1433 	case RTM_ADD:
1434 		kprintf("ADD ");
1435 		break;
1436 	case RTM_RESOLVE:
1437 		kprintf("RES ");
1438 		break;
1439 	case RTM_DELETE:
1440 		kprintf("DEL ");
1441 		break;
1442 	default:
1443 		kprintf("C%02d ", cmd);
1444 		break;
1445 	}
1446 	kprintf("rti %p cpu %d ", rti, mycpuid);
1447 	for (i = 0; i < rti->rti_addrs; ++i) {
1448 		if (rti->rti_info[i] == NULL)
1449 			continue;
1450 		if (didit)
1451 			kprintf(" ,");
1452 		switch(i) {
1453 		case RTAX_DST:
1454 			kprintf("(DST ");
1455 			break;
1456 		case RTAX_GATEWAY:
1457 			kprintf("(GWY ");
1458 			break;
1459 		case RTAX_NETMASK:
1460 			kprintf("(MSK ");
1461 			break;
1462 		case RTAX_GENMASK:
1463 			kprintf("(GEN ");
1464 			break;
1465 		case RTAX_IFP:
1466 			kprintf("(IFP ");
1467 			break;
1468 		case RTAX_IFA:
1469 			kprintf("(IFA ");
1470 			break;
1471 		case RTAX_AUTHOR:
1472 			kprintf("(AUT ");
1473 			break;
1474 		case RTAX_BRD:
1475 			kprintf("(BRD ");
1476 			break;
1477 		default:
1478 			kprintf("(?%02d ", i);
1479 			break;
1480 		}
1481 		sockaddr_print(rti->rti_info[i]);
1482 		kprintf(")");
1483 		didit = 1;
1484 	}
1485 	kprintf("\n");
1486 }
1487 
1488 void
1489 sockaddr_print(struct sockaddr *sa)
1490 {
1491 	struct sockaddr_in *sa4;
1492 	struct sockaddr_in6 *sa6;
1493 	int len;
1494 	int i;
1495 
1496 	if (sa == NULL) {
1497 		kprintf("NULL");
1498 		return;
1499 	}
1500 
1501 	len = sa->sa_len - offsetof(struct sockaddr, sa_data[0]);
1502 
1503 	switch(sa->sa_family) {
1504 	case AF_INET:
1505 	case AF_INET6:
1506 	default:
1507 		switch(sa->sa_family) {
1508 		case AF_INET:
1509 			sa4 = (struct sockaddr_in *)sa;
1510 			kprintf("INET %d %d.%d.%d.%d",
1511 				ntohs(sa4->sin_port),
1512 				(ntohl(sa4->sin_addr.s_addr) >> 24) & 255,
1513 				(ntohl(sa4->sin_addr.s_addr) >> 16) & 255,
1514 				(ntohl(sa4->sin_addr.s_addr) >> 8) & 255,
1515 				(ntohl(sa4->sin_addr.s_addr) >> 0) & 255
1516 			);
1517 			break;
1518 		case AF_INET6:
1519 			sa6 = (struct sockaddr_in6 *)sa;
1520 			kprintf("INET6 %d %04x:%04x%04x:%04x:%04x:%04x:%04x:%04x",
1521 				ntohs(sa6->sin6_port),
1522 				sa6->sin6_addr.s6_addr16[0],
1523 				sa6->sin6_addr.s6_addr16[1],
1524 				sa6->sin6_addr.s6_addr16[2],
1525 				sa6->sin6_addr.s6_addr16[3],
1526 				sa6->sin6_addr.s6_addr16[4],
1527 				sa6->sin6_addr.s6_addr16[5],
1528 				sa6->sin6_addr.s6_addr16[6],
1529 				sa6->sin6_addr.s6_addr16[7]
1530 			);
1531 			break;
1532 		default:
1533 			kprintf("AF%d ", sa->sa_family);
1534 			while (len > 0 && sa->sa_data[len-1] == 0)
1535 				--len;
1536 
1537 			for (i = 0; i < len; ++i) {
1538 				if (i)
1539 					kprintf(".");
1540 				kprintf("%d", (unsigned char)sa->sa_data[i]);
1541 			}
1542 			break;
1543 		}
1544 	}
1545 }
1546 
1547 #endif
1548 
1549 /*
1550  * Set up a routing table entry, normally for an interface.
1551  */
1552 int
1553 rtinit(struct ifaddr *ifa, int cmd, int flags)
1554 {
1555 	struct sockaddr *dst, *deldst, *netmask;
1556 	struct mbuf *m = NULL;
1557 	struct radix_node_head *rnh;
1558 	struct radix_node *rn;
1559 	struct rt_addrinfo rtinfo;
1560 	int error;
1561 
1562 	if (flags & RTF_HOST) {
1563 		dst = ifa->ifa_dstaddr;
1564 		netmask = NULL;
1565 	} else {
1566 		dst = ifa->ifa_addr;
1567 		netmask = ifa->ifa_netmask;
1568 	}
1569 	/*
1570 	 * If it's a delete, check that if it exists, it's on the correct
1571 	 * interface or we might scrub a route to another ifa which would
1572 	 * be confusing at best and possibly worse.
1573 	 */
1574 	if (cmd == RTM_DELETE) {
1575 		/*
1576 		 * It's a delete, so it should already exist..
1577 		 * If it's a net, mask off the host bits
1578 		 * (Assuming we have a mask)
1579 		 */
1580 		if (netmask != NULL) {
1581 			m = m_get(MB_DONTWAIT, MT_SONAME);
1582 			if (m == NULL)
1583 				return (ENOBUFS);
1584 			mbuftrackid(m, 34);
1585 			deldst = mtod(m, struct sockaddr *);
1586 			rt_maskedcopy(dst, deldst, netmask);
1587 			dst = deldst;
1588 		}
1589 		/*
1590 		 * Look up an rtentry that is in the routing tree and
1591 		 * contains the correct info.
1592 		 */
1593 		if ((rnh = rt_tables[mycpuid][dst->sa_family]) == NULL ||
1594 		    (rn = rnh->rnh_lookup((char *)dst,
1595 					  (char *)netmask, rnh)) == NULL ||
1596 		    ((struct rtentry *)rn)->rt_ifa != ifa ||
1597 		    !sa_equal((struct sockaddr *)rn->rn_key, dst)) {
1598 			if (m != NULL)
1599 				m_free(m);
1600 			return (flags & RTF_HOST ? EHOSTUNREACH : ENETUNREACH);
1601 		}
1602 		/* XXX */
1603 #if 0
1604 		else {
1605 			/*
1606 			 * One would think that as we are deleting, and we know
1607 			 * it doesn't exist, we could just return at this point
1608 			 * with an "ELSE" clause, but apparently not..
1609 			 */
1610 			return (flags & RTF_HOST ? EHOSTUNREACH : ENETUNREACH);
1611 		}
1612 #endif
1613 	}
1614 	/*
1615 	 * Do the actual request
1616 	 */
1617 	bzero(&rtinfo, sizeof(struct rt_addrinfo));
1618 	rtinfo.rti_info[RTAX_DST] = dst;
1619 	rtinfo.rti_info[RTAX_GATEWAY] = ifa->ifa_addr;
1620 	rtinfo.rti_info[RTAX_NETMASK] = netmask;
1621 	rtinfo.rti_flags = flags | ifa->ifa_flags;
1622 	rtinfo.rti_ifa = ifa;
1623 	error = rtrequest1_global(cmd, &rtinfo, rtinit_rtrequest_callback, ifa);
1624 	if (m != NULL)
1625 		m_free(m);
1626 	return (error);
1627 }
1628 
1629 static void
1630 rtinit_rtrequest_callback(int cmd, int error,
1631 			  struct rt_addrinfo *rtinfo, struct rtentry *rt,
1632 			  void *arg)
1633 {
1634 	struct ifaddr *ifa = arg;
1635 
1636 	if (error == 0 && rt) {
1637 		if (mycpuid == 0) {
1638 			++rt->rt_refcnt;
1639 			rt_newaddrmsg(cmd, ifa, error, rt);
1640 			--rt->rt_refcnt;
1641 		}
1642 		if (cmd == RTM_DELETE) {
1643 			if (rt->rt_refcnt == 0) {
1644 				++rt->rt_refcnt;
1645 				rtfree(rt);
1646 			}
1647 		}
1648 	}
1649 }
1650 
1651 struct netmsg_rts {
1652 	struct netmsg		netmsg;
1653 	int			req;
1654 	struct rt_addrinfo	*rtinfo;
1655 	rtsearch_callback_func_t callback;
1656 	void			*arg;
1657 	boolean_t		exact_match;
1658 	int			found_cnt;
1659 };
1660 
1661 int
1662 rtsearch_global(int req, struct rt_addrinfo *rtinfo,
1663 		rtsearch_callback_func_t callback, void *arg,
1664 		boolean_t exact_match)
1665 {
1666 	struct netmsg_rts msg;
1667 
1668 	netmsg_init(&msg.netmsg, NULL, &curthread->td_msgport,
1669 		    0, rtsearch_msghandler);
1670 	msg.req = req;
1671 	msg.rtinfo = rtinfo;
1672 	msg.callback = callback;
1673 	msg.arg = arg;
1674 	msg.exact_match = exact_match;
1675 	msg.found_cnt = 0;
1676 	return lwkt_domsg(rtable_portfn(0), &msg.netmsg.nm_lmsg, 0);
1677 }
1678 
1679 static void
1680 rtsearch_msghandler(struct netmsg *netmsg)
1681 {
1682 	struct netmsg_rts *msg = (void *)netmsg;
1683 	struct rt_addrinfo rtinfo;
1684 	struct radix_node_head *rnh;
1685 	struct rtentry *rt;
1686 	int nextcpu, error;
1687 
1688 	/*
1689 	 * Copy the rtinfo.  We need to make sure that the original
1690 	 * rtinfo, which is setup by the caller, in the netmsg will
1691 	 * _not_ be changed; else the next CPU on the netmsg forwarding
1692 	 * path will see a different rtinfo than what this CPU has seen.
1693 	 */
1694 	rtinfo = *msg->rtinfo;
1695 
1696 	/*
1697 	 * Find the correct routing tree to use for this Address Family
1698 	 */
1699 	if ((rnh = rt_tables[mycpuid][rtinfo.rti_dst->sa_family]) == NULL) {
1700 		if (mycpuid != 0)
1701 			panic("partially initialized routing tables\n");
1702 		lwkt_replymsg(&msg->netmsg.nm_lmsg, EAFNOSUPPORT);
1703 		return;
1704 	}
1705 
1706 	/*
1707 	 * Correct rtinfo for the host route searching.
1708 	 */
1709 	if (rtinfo.rti_flags & RTF_HOST) {
1710 		rtinfo.rti_netmask = NULL;
1711 		rtinfo.rti_flags &= ~(RTF_CLONING | RTF_PRCLONING);
1712 	}
1713 
1714 	rt = (struct rtentry *)
1715 	     rnh->rnh_lookup((char *)rtinfo.rti_dst,
1716 			     (char *)rtinfo.rti_netmask, rnh);
1717 
1718 	/*
1719 	 * If we are asked to do the "exact match", we need to make sure
1720 	 * that host route searching got a host route while a network
1721 	 * route searching got a network route.
1722 	 */
1723 	if (rt != NULL && msg->exact_match &&
1724 	    ((rt->rt_flags ^ rtinfo.rti_flags) & RTF_HOST))
1725 		rt = NULL;
1726 
1727 	if (rt == NULL) {
1728 		/*
1729 		 * No matching routes have been found, don't count this
1730 		 * as a critical error (here, we set 'error' to 0), just
1731 		 * keep moving on, since at least prcloned routes are not
1732 		 * duplicated onto each CPU.
1733 		 */
1734 		error = 0;
1735 	} else {
1736 		msg->found_cnt++;
1737 
1738 		rt->rt_refcnt++;
1739 		error = msg->callback(msg->req, &rtinfo, rt, msg->arg,
1740 				      msg->found_cnt);
1741 		rt->rt_refcnt--;
1742 
1743 		if (error == EJUSTRETURN) {
1744 			lwkt_replymsg(&msg->netmsg.nm_lmsg, 0);
1745 			return;
1746 		}
1747 	}
1748 
1749 	nextcpu = mycpuid + 1;
1750 	if (error) {
1751 		KKASSERT(msg->found_cnt > 0);
1752 
1753 		/*
1754 		 * Under following cases, unrecoverable error has
1755 		 * not occured:
1756 		 * o  Request is RTM_GET
1757 		 * o  The first time that we find the route, but the
1758 		 *    modification fails.
1759 		 */
1760 		if (msg->req != RTM_GET && msg->found_cnt > 1) {
1761 			panic("rtsearch_msghandler: unrecoverable error "
1762 			      "cpu %d", mycpuid);
1763 		}
1764 		lwkt_replymsg(&msg->netmsg.nm_lmsg, error);
1765 	} else if (nextcpu < ncpus) {
1766 		lwkt_forwardmsg(rtable_portfn(nextcpu), &msg->netmsg.nm_lmsg);
1767 	} else {
1768 		if (msg->found_cnt == 0) {
1769 			/* The requested route was never seen ... */
1770 			error = ESRCH;
1771 		}
1772 		lwkt_replymsg(&msg->netmsg.nm_lmsg, error);
1773 	}
1774 }
1775 
1776 int
1777 rtmask_add_global(struct sockaddr *mask)
1778 {
1779 	struct netmsg nmsg;
1780 
1781 	netmsg_init(&nmsg, NULL, &curthread->td_msgport,
1782 		    0, rtmask_add_msghandler);
1783 	nmsg.nm_lmsg.u.ms_resultp = mask;
1784 
1785 	return lwkt_domsg(rtable_portfn(0), &nmsg.nm_lmsg, 0);
1786 }
1787 
1788 struct sockaddr *
1789 _rtmask_lookup(struct sockaddr *mask, boolean_t search)
1790 {
1791 	struct radix_node *n;
1792 
1793 #define	clen(s)	(*(u_char *)(s))
1794 	n = rn_addmask((char *)mask, search, 1);
1795 	if (n != NULL &&
1796 	    mask->sa_len >= clen(n->rn_key) &&
1797 	    bcmp((char *)mask + 1,
1798 		 (char *)n->rn_key + 1, clen(n->rn_key) - 1) == 0) {
1799 		return (struct sockaddr *)n->rn_key;
1800 	} else {
1801 		return NULL;
1802 	}
1803 #undef clen
1804 }
1805 
1806 static void
1807 rtmask_add_msghandler(struct netmsg *nmsg)
1808 {
1809 	struct lwkt_msg *lmsg = &nmsg->nm_lmsg;
1810 	struct sockaddr *mask = lmsg->u.ms_resultp;
1811 	int error = 0, nextcpu;
1812 
1813 	if (rtmask_lookup(mask) == NULL)
1814 		error = ENOBUFS;
1815 
1816 	nextcpu = mycpuid + 1;
1817 	if (!error && nextcpu < ncpus)
1818 		lwkt_forwardmsg(rtable_portfn(nextcpu), lmsg);
1819 	else
1820 		lwkt_replymsg(lmsg, error);
1821 }
1822 
1823 /* This must be before ip6_init2(), which is now SI_ORDER_MIDDLE */
1824 SYSINIT(route, SI_SUB_PROTO_DOMAIN, SI_ORDER_THIRD, route_init, 0);
1825