FreeBSD/Linux Kernel Cross Reference
sys/net/route.c
1 /*-
2 * Copyright (c) 1980, 1986, 1991, 1993
3 * The Regents of the University of California. All rights reserved.
4 *
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
7 * are met:
8 * 1. Redistributions of source code must retain the above copyright
9 * notice, this list of conditions and the following disclaimer.
10 * 2. Redistributions in binary form must reproduce the above copyright
11 * notice, this list of conditions and the following disclaimer in the
12 * documentation and/or other materials provided with the distribution.
13 * 4. Neither the name of the University nor the names of its contributors
14 * may be used to endorse or promote products derived from this software
15 * without specific prior written permission.
16 *
17 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
18 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
21 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27 * SUCH DAMAGE.
28 *
29 * @(#)route.c 8.3.1.1 (Berkeley) 2/23/95
30 * $FreeBSD: releng/8.0/sys/net/route.c 197869 2009-10-08 20:58:09Z bz $
31 */
32 /************************************************************************
33 * Note: In this file a 'fib' is a "forwarding information base" *
34 * Which is the new name for an in kernel routing (next hop) table. *
35 ***********************************************************************/
36
37 #include "opt_inet.h"
38 #include "opt_route.h"
39 #include "opt_mrouting.h"
40 #include "opt_mpath.h"
41
42 #include <sys/param.h>
43 #include <sys/systm.h>
44 #include <sys/syslog.h>
45 #include <sys/malloc.h>
46 #include <sys/mbuf.h>
47 #include <sys/socket.h>
48 #include <sys/sysctl.h>
49 #include <sys/syslog.h>
50 #include <sys/sysproto.h>
51 #include <sys/proc.h>
52 #include <sys/domain.h>
53 #include <sys/kernel.h>
54
55 #include <net/if.h>
56 #include <net/if_dl.h>
57 #include <net/route.h>
58 #include <net/vnet.h>
59 #include <net/flowtable.h>
60
61 #ifdef RADIX_MPATH
62 #include <net/radix_mpath.h>
63 #endif
64
65 #include <netinet/in.h>
66 #include <netinet/ip_mroute.h>
67
68 #include <vm/uma.h>
69
70 u_int rt_numfibs = RT_NUMFIBS;
71 SYSCTL_INT(_net, OID_AUTO, fibs, CTLFLAG_RD, &rt_numfibs, 0, "");
72 /*
73 * Allow the boot code to allow LESS than RT_MAXFIBS to be used.
74 * We can't do more because storage is statically allocated for now.
75 * (for compatibility reasons.. this will change).
76 */
77 TUNABLE_INT("net.fibs", &rt_numfibs);
78
79 /*
80 * By default add routes to all fibs for new interfaces.
81 * Once this is set to 0 then only allocate routes on interface
82 * changes for the FIB of the caller when adding a new set of addresses
83 * to an interface. XXX this is a shotgun aproach to a problem that needs
84 * a more fine grained solution.. that will come.
85 */
86 u_int rt_add_addr_allfibs = 1;
87 SYSCTL_INT(_net, OID_AUTO, add_addr_allfibs, CTLFLAG_RW,
88 &rt_add_addr_allfibs, 0, "");
89 TUNABLE_INT("net.add_addr_allfibs", &rt_add_addr_allfibs);
90
91 VNET_DEFINE(struct radix_node_head *, rt_tables);
92 static VNET_DEFINE(uma_zone_t, rtzone); /* Routing table UMA zone. */
93 VNET_DEFINE(int, rttrash); /* routes not in table but not freed */
94 VNET_DEFINE(struct rtstat, rtstat);
95
96 #define V_rt_tables VNET(rt_tables)
97 #define V_rtzone VNET(rtzone)
98 #define V_rttrash VNET(rttrash)
99 #define V_rtstat VNET(rtstat)
100
101 static void rt_maskedcopy(struct sockaddr *,
102 struct sockaddr *, struct sockaddr *);
103
104 /* compare two sockaddr structures */
105 #define sa_equal(a1, a2) (bcmp((a1), (a2), (a1)->sa_len) == 0)
106
107 /*
108 * Convert a 'struct radix_node *' to a 'struct rtentry *'.
109 * The operation can be done safely (in this code) because a
110 * 'struct rtentry' starts with two 'struct radix_node''s, the first
111 * one representing leaf nodes in the routing tree, which is
112 * what the code in radix.c passes us as a 'struct radix_node'.
113 *
114 * But because there are a lot of assumptions in this conversion,
115 * do not cast explicitly, but always use the macro below.
116 */
117 #define RNTORT(p) ((struct rtentry *)(p))
118
119 #if 0
120 /* default fib for tunnels to use */
121 u_int tunnel_fib = 0;
122 SYSCTL_INT(_net, OID_AUTO, tunnelfib, CTLFLAG_RD, &tunnel_fib, 0, "");
123 #endif
124
125 /*
126 * handler for net.my_fibnum
127 */
128 static int
129 sysctl_my_fibnum(SYSCTL_HANDLER_ARGS)
130 {
131 int fibnum;
132 int error;
133
134 fibnum = curthread->td_proc->p_fibnum;
135 error = sysctl_handle_int(oidp, &fibnum, 0, req);
136 return (error);
137 }
138
139 SYSCTL_PROC(_net, OID_AUTO, my_fibnum, CTLTYPE_INT|CTLFLAG_RD,
140 NULL, 0, &sysctl_my_fibnum, "I", "default FIB of caller");
141
142 static __inline struct radix_node_head **
143 rt_tables_get_rnh_ptr(int table, int fam)
144 {
145 struct radix_node_head **rnh;
146
147 KASSERT(table >= 0 && table < rt_numfibs, ("%s: table out of bounds.",
148 __func__));
149 KASSERT(fam >= 0 && fam < (AF_MAX+1), ("%s: fam out of bounds.",
150 __func__));
151
152 /* rnh is [fib=0][af=0]. */
153 rnh = (struct radix_node_head **)V_rt_tables;
154 /* Get the offset to the requested table and fam. */
155 rnh += table * (AF_MAX+1) + fam;
156
157 return (rnh);
158 }
159
160 struct radix_node_head *
161 rt_tables_get_rnh(int table, int fam)
162 {
163
164 return (*rt_tables_get_rnh_ptr(table, fam));
165 }
166
167 /*
168 * route initialization must occur before ip6_init2(), which happenas at
169 * SI_ORDER_MIDDLE.
170 */
171 static void
172 route_init(void)
173 {
174
175 /* whack the tunable ints into line. */
176 if (rt_numfibs > RT_MAXFIBS)
177 rt_numfibs = RT_MAXFIBS;
178 if (rt_numfibs == 0)
179 rt_numfibs = 1;
180 rn_init(); /* initialize all zeroes, all ones, mask table */
181 }
182 SYSINIT(route_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_THIRD, route_init, 0);
183
184 static void
185 vnet_route_init(const void *unused __unused)
186 {
187 struct domain *dom;
188 struct radix_node_head **rnh;
189 int table;
190 int fam;
191
192 V_rt_tables = malloc(rt_numfibs * (AF_MAX+1) *
193 sizeof(struct radix_node_head *), M_RTABLE, M_WAITOK|M_ZERO);
194
195 V_rtzone = uma_zcreate("rtentry", sizeof(struct rtentry), NULL, NULL,
196 NULL, NULL, UMA_ALIGN_PTR, 0);
197 for (dom = domains; dom; dom = dom->dom_next) {
198 if (dom->dom_rtattach) {
199 for (table = 0; table < rt_numfibs; table++) {
200 if ( (fam = dom->dom_family) == AF_INET ||
201 table == 0) {
202 /* for now only AF_INET has > 1 table */
203 /* XXX MRT
204 * rtattach will be also called
205 * from vfs_export.c but the
206 * offset will be 0
207 * (only for AF_INET and AF_INET6
208 * which don't need it anyhow)
209 */
210 rnh = rt_tables_get_rnh_ptr(table, fam);
211 if (rnh == NULL)
212 panic("%s: rnh NULL", __func__);
213 dom->dom_rtattach((void **)rnh,
214 dom->dom_rtoffset);
215 } else {
216 break;
217 }
218 }
219 }
220 }
221 }
222 VNET_SYSINIT(vnet_route_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_FOURTH,
223 vnet_route_init, 0);
224
225 #ifdef VIMAGE
226 static void
227 vnet_route_uninit(const void *unused __unused)
228 {
229 int table;
230 int fam;
231 struct domain *dom;
232 struct radix_node_head **rnh;
233
234 for (dom = domains; dom; dom = dom->dom_next) {
235 if (dom->dom_rtdetach) {
236 for (table = 0; table < rt_numfibs; table++) {
237 if ( (fam = dom->dom_family) == AF_INET ||
238 table == 0) {
239 /* For now only AF_INET has > 1 tbl. */
240 rnh = rt_tables_get_rnh_ptr(table, fam);
241 if (rnh == NULL)
242 panic("%s: rnh NULL", __func__);
243 dom->dom_rtdetach((void **)rnh,
244 dom->dom_rtoffset);
245 } else {
246 break;
247 }
248 }
249 }
250 }
251 }
252 VNET_SYSUNINIT(vnet_route_uninit, SI_SUB_PROTO_DOMAIN, SI_ORDER_THIRD,
253 vnet_route_uninit, 0);
254 #endif
255
256 #ifndef _SYS_SYSPROTO_H_
257 struct setfib_args {
258 int fibnum;
259 };
260 #endif
261 int
262 setfib(struct thread *td, struct setfib_args *uap)
263 {
264 if (uap->fibnum < 0 || uap->fibnum >= rt_numfibs)
265 return EINVAL;
266 td->td_proc->p_fibnum = uap->fibnum;
267 return (0);
268 }
269
270 /*
271 * Packet routing routines.
272 */
273 void
274 rtalloc(struct route *ro)
275 {
276 rtalloc_ign_fib(ro, 0UL, 0);
277 }
278
279 void
280 rtalloc_fib(struct route *ro, u_int fibnum)
281 {
282 rtalloc_ign_fib(ro, 0UL, fibnum);
283 }
284
285 void
286 rtalloc_ign(struct route *ro, u_long ignore)
287 {
288 struct rtentry *rt;
289
290 if ((rt = ro->ro_rt) != NULL) {
291 if (rt->rt_ifp != NULL && rt->rt_flags & RTF_UP)
292 return;
293 RTFREE(rt);
294 ro->ro_rt = NULL;
295 }
296 ro->ro_rt = rtalloc1_fib(&ro->ro_dst, 1, ignore, 0);
297 if (ro->ro_rt)
298 RT_UNLOCK(ro->ro_rt);
299 }
300
301 void
302 rtalloc_ign_fib(struct route *ro, u_long ignore, u_int fibnum)
303 {
304 struct rtentry *rt;
305
306 if ((rt = ro->ro_rt) != NULL) {
307 if (rt->rt_ifp != NULL && rt->rt_flags & RTF_UP)
308 return;
309 RTFREE(rt);
310 ro->ro_rt = NULL;
311 }
312 ro->ro_rt = rtalloc1_fib(&ro->ro_dst, 1, ignore, fibnum);
313 if (ro->ro_rt)
314 RT_UNLOCK(ro->ro_rt);
315 }
316
317 /*
318 * Look up the route that matches the address given
319 * Or, at least try.. Create a cloned route if needed.
320 *
321 * The returned route, if any, is locked.
322 */
323 struct rtentry *
324 rtalloc1(struct sockaddr *dst, int report, u_long ignflags)
325 {
326 return (rtalloc1_fib(dst, report, ignflags, 0));
327 }
328
329 struct rtentry *
330 rtalloc1_fib(struct sockaddr *dst, int report, u_long ignflags,
331 u_int fibnum)
332 {
333 struct radix_node_head *rnh;
334 struct rtentry *rt;
335 struct radix_node *rn;
336 struct rtentry *newrt;
337 struct rt_addrinfo info;
338 int err = 0, msgtype = RTM_MISS;
339 int needlock;
340
341 KASSERT((fibnum < rt_numfibs), ("rtalloc1_fib: bad fibnum"));
342 if (dst->sa_family != AF_INET) /* Only INET supports > 1 fib now */
343 fibnum = 0;
344 rnh = rt_tables_get_rnh(fibnum, dst->sa_family);
345 newrt = NULL;
346 /*
347 * Look up the address in the table for that Address Family
348 */
349 if (rnh == NULL) {
350 V_rtstat.rts_unreach++;
351 goto miss;
352 }
353 needlock = !(ignflags & RTF_RNH_LOCKED);
354 if (needlock)
355 RADIX_NODE_HEAD_RLOCK(rnh);
356 #ifdef INVARIANTS
357 else
358 RADIX_NODE_HEAD_LOCK_ASSERT(rnh);
359 #endif
360 rn = rnh->rnh_matchaddr(dst, rnh);
361 if (rn && ((rn->rn_flags & RNF_ROOT) == 0)) {
362 newrt = rt = RNTORT(rn);
363 RT_LOCK(newrt);
364 RT_ADDREF(newrt);
365 if (needlock)
366 RADIX_NODE_HEAD_RUNLOCK(rnh);
367 goto done;
368
369 } else if (needlock)
370 RADIX_NODE_HEAD_RUNLOCK(rnh);
371
372 /*
373 * Either we hit the root or couldn't find any match,
374 * Which basically means
375 * "caint get there frm here"
376 */
377 V_rtstat.rts_unreach++;
378 miss:
379 if (report) {
380 /*
381 * If required, report the failure to the supervising
382 * Authorities.
383 * For a delete, this is not an error. (report == 0)
384 */
385 bzero(&info, sizeof(info));
386 info.rti_info[RTAX_DST] = dst;
387 rt_missmsg(msgtype, &info, 0, err);
388 }
389 done:
390 if (newrt)
391 RT_LOCK_ASSERT(newrt);
392 return (newrt);
393 }
394
395 /*
396 * Remove a reference count from an rtentry.
397 * If the count gets low enough, take it out of the routing table
398 */
399 void
400 rtfree(struct rtentry *rt)
401 {
402 struct radix_node_head *rnh;
403
404 KASSERT(rt != NULL,("%s: NULL rt", __func__));
405 rnh = rt_tables_get_rnh(rt->rt_fibnum, rt_key(rt)->sa_family);
406 KASSERT(rnh != NULL,("%s: NULL rnh", __func__));
407
408 RT_LOCK_ASSERT(rt);
409
410 /*
411 * The callers should use RTFREE_LOCKED() or RTFREE(), so
412 * we should come here exactly with the last reference.
413 */
414 RT_REMREF(rt);
415 if (rt->rt_refcnt > 0) {
416 log(LOG_DEBUG, "%s: %p has %d refs\n", __func__, rt, rt->rt_refcnt);
417 goto done;
418 }
419
420 /*
421 * On last reference give the "close method" a chance
422 * to cleanup private state. This also permits (for
423 * IPv4 and IPv6) a chance to decide if the routing table
424 * entry should be purged immediately or at a later time.
425 * When an immediate purge is to happen the close routine
426 * typically calls rtexpunge which clears the RTF_UP flag
427 * on the entry so that the code below reclaims the storage.
428 */
429 if (rt->rt_refcnt == 0 && rnh->rnh_close)
430 rnh->rnh_close((struct radix_node *)rt, rnh);
431
432 /*
433 * If we are no longer "up" (and ref == 0)
434 * then we can free the resources associated
435 * with the route.
436 */
437 if ((rt->rt_flags & RTF_UP) == 0) {
438 if (rt->rt_nodes->rn_flags & (RNF_ACTIVE | RNF_ROOT))
439 panic("rtfree 2");
440 /*
441 * the rtentry must have been removed from the routing table
442 * so it is represented in rttrash.. remove that now.
443 */
444 V_rttrash--;
445 #ifdef DIAGNOSTIC
446 if (rt->rt_refcnt < 0) {
447 printf("rtfree: %p not freed (neg refs)\n", rt);
448 goto done;
449 }
450 #endif
451 /*
452 * release references on items we hold them on..
453 * e.g other routes and ifaddrs.
454 */
455 if (rt->rt_ifa)
456 ifa_free(rt->rt_ifa);
457 /*
458 * The key is separatly alloc'd so free it (see rt_setgate()).
459 * This also frees the gateway, as they are always malloc'd
460 * together.
461 */
462 Free(rt_key(rt));
463
464 /*
465 * and the rtentry itself of course
466 */
467 RT_LOCK_DESTROY(rt);
468 uma_zfree(V_rtzone, rt);
469 return;
470 }
471 done:
472 RT_UNLOCK(rt);
473 }
474
475
476 /*
477 * Force a routing table entry to the specified
478 * destination to go through the given gateway.
479 * Normally called as a result of a routing redirect
480 * message from the network layer.
481 */
482 void
483 rtredirect(struct sockaddr *dst,
484 struct sockaddr *gateway,
485 struct sockaddr *netmask,
486 int flags,
487 struct sockaddr *src)
488 {
489 rtredirect_fib(dst, gateway, netmask, flags, src, 0);
490 }
491
492 void
493 rtredirect_fib(struct sockaddr *dst,
494 struct sockaddr *gateway,
495 struct sockaddr *netmask,
496 int flags,
497 struct sockaddr *src,
498 u_int fibnum)
499 {
500 struct rtentry *rt, *rt0 = NULL;
501 int error = 0;
502 short *stat = NULL;
503 struct rt_addrinfo info;
504 struct ifaddr *ifa;
505 struct radix_node_head *rnh;
506
507 ifa = NULL;
508 rnh = rt_tables_get_rnh(fibnum, dst->sa_family);
509 if (rnh == NULL) {
510 error = EAFNOSUPPORT;
511 goto out;
512 }
513
514 /* verify the gateway is directly reachable */
515 if ((ifa = ifa_ifwithnet(gateway)) == NULL) {
516 error = ENETUNREACH;
517 goto out;
518 }
519 rt = rtalloc1_fib(dst, 0, 0UL, fibnum); /* NB: rt is locked */
520 /*
521 * If the redirect isn't from our current router for this dst,
522 * it's either old or wrong. If it redirects us to ourselves,
523 * we have a routing loop, perhaps as a result of an interface
524 * going down recently.
525 */
526 if (!(flags & RTF_DONE) && rt &&
527 (!sa_equal(src, rt->rt_gateway) || rt->rt_ifa != ifa))
528 error = EINVAL;
529 else if (ifa_ifwithaddr_check(gateway))
530 error = EHOSTUNREACH;
531 if (error)
532 goto done;
533 /*
534 * Create a new entry if we just got back a wildcard entry
535 * or the the lookup failed. This is necessary for hosts
536 * which use routing redirects generated by smart gateways
537 * to dynamically build the routing tables.
538 */
539 if (rt == NULL || (rt_mask(rt) && rt_mask(rt)->sa_len < 2))
540 goto create;
541 /*
542 * Don't listen to the redirect if it's
543 * for a route to an interface.
544 */
545 if (rt->rt_flags & RTF_GATEWAY) {
546 if (((rt->rt_flags & RTF_HOST) == 0) && (flags & RTF_HOST)) {
547 /*
548 * Changing from route to net => route to host.
549 * Create new route, rather than smashing route to net.
550 */
551 create:
552 rt0 = rt;
553 rt = NULL;
554
555 flags |= RTF_GATEWAY | RTF_DYNAMIC;
556 bzero((caddr_t)&info, sizeof(info));
557 info.rti_info[RTAX_DST] = dst;
558 info.rti_info[RTAX_GATEWAY] = gateway;
559 info.rti_info[RTAX_NETMASK] = netmask;
560 info.rti_ifa = ifa;
561 info.rti_flags = flags;
562 if (rt0 != NULL)
563 RT_UNLOCK(rt0); /* drop lock to avoid LOR with RNH */
564 error = rtrequest1_fib(RTM_ADD, &info, &rt, fibnum);
565 if (rt != NULL) {
566 RT_LOCK(rt);
567 if (rt0 != NULL)
568 EVENTHANDLER_INVOKE(route_redirect_event, rt0, rt, dst);
569 flags = rt->rt_flags;
570 }
571 if (rt0 != NULL)
572 RTFREE(rt0);
573
574 stat = &V_rtstat.rts_dynamic;
575 } else {
576 struct rtentry *gwrt;
577
578 /*
579 * Smash the current notion of the gateway to
580 * this destination. Should check about netmask!!!
581 */
582 rt->rt_flags |= RTF_MODIFIED;
583 flags |= RTF_MODIFIED;
584 stat = &V_rtstat.rts_newgateway;
585 /*
586 * add the key and gateway (in one malloc'd chunk).
587 */
588 RT_UNLOCK(rt);
589 RADIX_NODE_HEAD_LOCK(rnh);
590 RT_LOCK(rt);
591 rt_setgate(rt, rt_key(rt), gateway);
592 gwrt = rtalloc1(gateway, 1, RTF_RNH_LOCKED);
593 RADIX_NODE_HEAD_UNLOCK(rnh);
594 EVENTHANDLER_INVOKE(route_redirect_event, rt, gwrt, dst);
595 RTFREE_LOCKED(gwrt);
596 }
597 } else
598 error = EHOSTUNREACH;
599 done:
600 if (rt)
601 RTFREE_LOCKED(rt);
602 out:
603 if (error)
604 V_rtstat.rts_badredirect++;
605 else if (stat != NULL)
606 (*stat)++;
607 bzero((caddr_t)&info, sizeof(info));
608 info.rti_info[RTAX_DST] = dst;
609 info.rti_info[RTAX_GATEWAY] = gateway;
610 info.rti_info[RTAX_NETMASK] = netmask;
611 info.rti_info[RTAX_AUTHOR] = src;
612 rt_missmsg(RTM_REDIRECT, &info, flags, error);
613 if (ifa != NULL)
614 ifa_free(ifa);
615 }
616
617 int
618 rtioctl(u_long req, caddr_t data)
619 {
620 return (rtioctl_fib(req, data, 0));
621 }
622
623 /*
624 * Routing table ioctl interface.
625 */
626 int
627 rtioctl_fib(u_long req, caddr_t data, u_int fibnum)
628 {
629
630 /*
631 * If more ioctl commands are added here, make sure the proper
632 * super-user checks are being performed because it is possible for
633 * prison-root to make it this far if raw sockets have been enabled
634 * in jails.
635 */
636 #ifdef INET
637 /* Multicast goop, grrr... */
638 return mrt_ioctl ? mrt_ioctl(req, data, fibnum) : EOPNOTSUPP;
639 #else /* INET */
640 return ENXIO;
641 #endif /* INET */
642 }
643
644 /*
645 * For both ifa_ifwithroute() routines, 'ifa' is returned referenced.
646 */
647 struct ifaddr *
648 ifa_ifwithroute(int flags, struct sockaddr *dst, struct sockaddr *gateway)
649 {
650 return (ifa_ifwithroute_fib(flags, dst, gateway, 0));
651 }
652
653 struct ifaddr *
654 ifa_ifwithroute_fib(int flags, struct sockaddr *dst, struct sockaddr *gateway,
655 u_int fibnum)
656 {
657 register struct ifaddr *ifa;
658 int not_found = 0;
659
660 if ((flags & RTF_GATEWAY) == 0) {
661 /*
662 * If we are adding a route to an interface,
663 * and the interface is a pt to pt link
664 * we should search for the destination
665 * as our clue to the interface. Otherwise
666 * we can use the local address.
667 */
668 ifa = NULL;
669 if (flags & RTF_HOST)
670 ifa = ifa_ifwithdstaddr(dst);
671 if (ifa == NULL)
672 ifa = ifa_ifwithaddr(gateway);
673 } else {
674 /*
675 * If we are adding a route to a remote net
676 * or host, the gateway may still be on the
677 * other end of a pt to pt link.
678 */
679 ifa = ifa_ifwithdstaddr(gateway);
680 }
681 if (ifa == NULL)
682 ifa = ifa_ifwithnet(gateway);
683 if (ifa == NULL) {
684 struct rtentry *rt = rtalloc1_fib(gateway, 0, RTF_RNH_LOCKED, fibnum);
685 if (rt == NULL)
686 return (NULL);
687 /*
688 * dismiss a gateway that is reachable only
689 * through the default router
690 */
691 switch (gateway->sa_family) {
692 case AF_INET:
693 if (satosin(rt_key(rt))->sin_addr.s_addr == INADDR_ANY)
694 not_found = 1;
695 break;
696 case AF_INET6:
697 if (IN6_IS_ADDR_UNSPECIFIED(&satosin6(rt_key(rt))->sin6_addr))
698 not_found = 1;
699 break;
700 default:
701 break;
702 }
703 if (!not_found && rt->rt_ifa != NULL) {
704 ifa = rt->rt_ifa;
705 ifa_ref(ifa);
706 }
707 RT_REMREF(rt);
708 RT_UNLOCK(rt);
709 if (not_found || ifa == NULL)
710 return (NULL);
711 }
712 if (ifa->ifa_addr->sa_family != dst->sa_family) {
713 struct ifaddr *oifa = ifa;
714 ifa = ifaof_ifpforaddr(dst, ifa->ifa_ifp);
715 if (ifa == NULL)
716 ifa = oifa;
717 else
718 ifa_free(oifa);
719 }
720 return (ifa);
721 }
722
723 /*
724 * Do appropriate manipulations of a routing tree given
725 * all the bits of info needed
726 */
727 int
728 rtrequest(int req,
729 struct sockaddr *dst,
730 struct sockaddr *gateway,
731 struct sockaddr *netmask,
732 int flags,
733 struct rtentry **ret_nrt)
734 {
735 return (rtrequest_fib(req, dst, gateway, netmask, flags, ret_nrt, 0));
736 }
737
738 int
739 rtrequest_fib(int req,
740 struct sockaddr *dst,
741 struct sockaddr *gateway,
742 struct sockaddr *netmask,
743 int flags,
744 struct rtentry **ret_nrt,
745 u_int fibnum)
746 {
747 struct rt_addrinfo info;
748
749 if (dst->sa_len == 0)
750 return(EINVAL);
751
752 bzero((caddr_t)&info, sizeof(info));
753 info.rti_flags = flags;
754 info.rti_info[RTAX_DST] = dst;
755 info.rti_info[RTAX_GATEWAY] = gateway;
756 info.rti_info[RTAX_NETMASK] = netmask;
757 return rtrequest1_fib(req, &info, ret_nrt, fibnum);
758 }
759
760 /*
761 * These (questionable) definitions of apparent local variables apply
762 * to the next two functions. XXXXXX!!!
763 */
764 #define dst info->rti_info[RTAX_DST]
765 #define gateway info->rti_info[RTAX_GATEWAY]
766 #define netmask info->rti_info[RTAX_NETMASK]
767 #define ifaaddr info->rti_info[RTAX_IFA]
768 #define ifpaddr info->rti_info[RTAX_IFP]
769 #define flags info->rti_flags
770
771 int
772 rt_getifa(struct rt_addrinfo *info)
773 {
774 return (rt_getifa_fib(info, 0));
775 }
776
777 /*
778 * Look up rt_addrinfo for a specific fib. Note that if rti_ifa is defined,
779 * it will be referenced so the caller must free it.
780 */
781 int
782 rt_getifa_fib(struct rt_addrinfo *info, u_int fibnum)
783 {
784 struct ifaddr *ifa;
785 int error = 0;
786
787 /*
788 * ifp may be specified by sockaddr_dl
789 * when protocol address is ambiguous.
790 */
791 if (info->rti_ifp == NULL && ifpaddr != NULL &&
792 ifpaddr->sa_family == AF_LINK &&
793 (ifa = ifa_ifwithnet(ifpaddr)) != NULL) {
794 info->rti_ifp = ifa->ifa_ifp;
795 ifa_free(ifa);
796 }
797 if (info->rti_ifa == NULL && ifaaddr != NULL)
798 info->rti_ifa = ifa_ifwithaddr(ifaaddr);
799 if (info->rti_ifa == NULL) {
800 struct sockaddr *sa;
801
802 sa = ifaaddr != NULL ? ifaaddr :
803 (gateway != NULL ? gateway : dst);
804 if (sa != NULL && info->rti_ifp != NULL)
805 info->rti_ifa = ifaof_ifpforaddr(sa, info->rti_ifp);
806 else if (dst != NULL && gateway != NULL)
807 info->rti_ifa = ifa_ifwithroute_fib(flags, dst, gateway,
808 fibnum);
809 else if (sa != NULL)
810 info->rti_ifa = ifa_ifwithroute_fib(flags, sa, sa,
811 fibnum);
812 }
813 if ((ifa = info->rti_ifa) != NULL) {
814 if (info->rti_ifp == NULL)
815 info->rti_ifp = ifa->ifa_ifp;
816 } else
817 error = ENETUNREACH;
818 return (error);
819 }
820
821 /*
822 * Expunges references to a route that's about to be reclaimed.
823 * The route must be locked.
824 */
825 int
826 rtexpunge(struct rtentry *rt)
827 {
828 struct radix_node *rn;
829 struct radix_node_head *rnh;
830 struct ifaddr *ifa;
831 int error = 0;
832
833 /*
834 * Find the correct routing tree to use for this Address Family
835 */
836 rnh = rt_tables_get_rnh(rt->rt_fibnum, rt_key(rt)->sa_family);
837 RT_LOCK_ASSERT(rt);
838 if (rnh == NULL)
839 return (EAFNOSUPPORT);
840 RADIX_NODE_HEAD_LOCK_ASSERT(rnh);
841 #if 0
842 /*
843 * We cannot assume anything about the reference count
844 * because protocols call us in many situations; often
845 * before unwinding references to the table entry.
846 */
847 KASSERT(rt->rt_refcnt <= 1, ("bogus refcnt %ld", rt->rt_refcnt));
848 #endif
849 /*
850 * Remove the item from the tree; it should be there,
851 * but when callers invoke us blindly it may not (sigh).
852 */
853 rn = rnh->rnh_deladdr(rt_key(rt), rt_mask(rt), rnh);
854 if (rn == NULL) {
855 error = ESRCH;
856 goto bad;
857 }
858 KASSERT((rn->rn_flags & (RNF_ACTIVE | RNF_ROOT)) == 0,
859 ("unexpected flags 0x%x", rn->rn_flags));
860 KASSERT(rt == RNTORT(rn),
861 ("lookup mismatch, rt %p rn %p", rt, rn));
862
863 rt->rt_flags &= ~RTF_UP;
864
865 /*
866 * Give the protocol a chance to keep things in sync.
867 */
868 if ((ifa = rt->rt_ifa) && ifa->ifa_rtrequest) {
869 struct rt_addrinfo info;
870
871 bzero((caddr_t)&info, sizeof(info));
872 info.rti_flags = rt->rt_flags;
873 info.rti_info[RTAX_DST] = rt_key(rt);
874 info.rti_info[RTAX_GATEWAY] = rt->rt_gateway;
875 info.rti_info[RTAX_NETMASK] = rt_mask(rt);
876 ifa->ifa_rtrequest(RTM_DELETE, rt, &info);
877 }
878
879 /*
880 * one more rtentry floating around that is not
881 * linked to the routing table.
882 */
883 V_rttrash++;
884 bad:
885 return (error);
886 }
887
888 #ifdef RADIX_MPATH
889 static int
890 rn_mpath_update(int req, struct rt_addrinfo *info,
891 struct radix_node_head *rnh, struct rtentry **ret_nrt)
892 {
893 /*
894 * if we got multipath routes, we require users to specify
895 * a matching RTAX_GATEWAY.
896 */
897 struct rtentry *rt, *rto = NULL;
898 register struct radix_node *rn;
899 int error = 0;
900
901 rn = rnh->rnh_matchaddr(dst, rnh);
902 if (rn == NULL)
903 return (ESRCH);
904 rto = rt = RNTORT(rn);
905 rt = rt_mpath_matchgate(rt, gateway);
906 if (rt == NULL)
907 return (ESRCH);
908 /*
909 * this is the first entry in the chain
910 */
911 if (rto == rt) {
912 rn = rn_mpath_next((struct radix_node *)rt);
913 /*
914 * there is another entry, now it's active
915 */
916 if (rn) {
917 rto = RNTORT(rn);
918 RT_LOCK(rto);
919 rto->rt_flags |= RTF_UP;
920 RT_UNLOCK(rto);
921 } else if (rt->rt_flags & RTF_GATEWAY) {
922 /*
923 * For gateway routes, we need to
924 * make sure that we we are deleting
925 * the correct gateway.
926 * rt_mpath_matchgate() does not
927 * check the case when there is only
928 * one route in the chain.
929 */
930 if (gateway &&
931 (rt->rt_gateway->sa_len != gateway->sa_len ||
932 memcmp(rt->rt_gateway, gateway, gateway->sa_len)))
933 error = ESRCH;
934 else {
935 /*
936 * remove from tree before returning it
937 * to the caller
938 */
939 rn = rnh->rnh_deladdr(dst, netmask, rnh);
940 KASSERT(rt == RNTORT(rn), ("radix node disappeared"));
941 goto gwdelete;
942 }
943
944 }
945 /*
946 * use the normal delete code to remove
947 * the first entry
948 */
949 if (req != RTM_DELETE)
950 goto nondelete;
951
952 error = ENOENT;
953 goto done;
954 }
955
956 /*
957 * if the entry is 2nd and on up
958 */
959 if ((req == RTM_DELETE) && !rt_mpath_deldup(rto, rt))
960 panic ("rtrequest1: rt_mpath_deldup");
961 gwdelete:
962 RT_LOCK(rt);
963 RT_ADDREF(rt);
964 if (req == RTM_DELETE) {
965 rt->rt_flags &= ~RTF_UP;
966 /*
967 * One more rtentry floating around that is not
968 * linked to the routing table. rttrash will be decremented
969 * when RTFREE(rt) is eventually called.
970 */
971 V_rttrash++;
972 }
973
974 nondelete:
975 if (req != RTM_DELETE)
976 panic("unrecognized request %d", req);
977
978
979 /*
980 * If the caller wants it, then it can have it,
981 * but it's up to it to free the rtentry as we won't be
982 * doing it.
983 */
984 if (ret_nrt) {
985 *ret_nrt = rt;
986 RT_UNLOCK(rt);
987 } else
988 RTFREE_LOCKED(rt);
989 done:
990 return (error);
991 }
992 #endif
993
994 int
995 rtrequest1_fib(int req, struct rt_addrinfo *info, struct rtentry **ret_nrt,
996 u_int fibnum)
997 {
998 int error = 0, needlock = 0;
999 register struct rtentry *rt;
1000 #ifdef FLOWTABLE
1001 register struct rtentry *rt0;
1002 #endif
1003 register struct radix_node *rn;
1004 register struct radix_node_head *rnh;
1005 struct ifaddr *ifa;
1006 struct sockaddr *ndst;
1007 #define senderr(x) { error = x ; goto bad; }
1008
1009 KASSERT((fibnum < rt_numfibs), ("rtrequest1_fib: bad fibnum"));
1010 if (dst->sa_family != AF_INET) /* Only INET supports > 1 fib now */
1011 fibnum = 0;
1012 /*
1013 * Find the correct routing tree to use for this Address Family
1014 */
1015 rnh = rt_tables_get_rnh(fibnum, dst->sa_family);
1016 if (rnh == NULL)
1017 return (EAFNOSUPPORT);
1018 needlock = ((flags & RTF_RNH_LOCKED) == 0);
1019 flags &= ~RTF_RNH_LOCKED;
1020 if (needlock)
1021 RADIX_NODE_HEAD_LOCK(rnh);
1022 else
1023 RADIX_NODE_HEAD_LOCK_ASSERT(rnh);
1024 /*
1025 * If we are adding a host route then we don't want to put
1026 * a netmask in the tree, nor do we want to clone it.
1027 */
1028 if (flags & RTF_HOST)
1029 netmask = NULL;
1030
1031 switch (req) {
1032 case RTM_DELETE:
1033 #ifdef RADIX_MPATH
1034 if (rn_mpath_capable(rnh)) {
1035 error = rn_mpath_update(req, info, rnh, ret_nrt);
1036 /*
1037 * "bad" holds true for the success case
1038 * as well
1039 */
1040 if (error != ENOENT)
1041 goto bad;
1042 }
1043 #endif
1044 /*
1045 * Remove the item from the tree and return it.
1046 * Complain if it is not there and do no more processing.
1047 */
1048 rn = rnh->rnh_deladdr(dst, netmask, rnh);
1049 if (rn == NULL)
1050 senderr(ESRCH);
1051 if (rn->rn_flags & (RNF_ACTIVE | RNF_ROOT))
1052 panic ("rtrequest delete");
1053 rt = RNTORT(rn);
1054 RT_LOCK(rt);
1055 RT_ADDREF(rt);
1056 rt->rt_flags &= ~RTF_UP;
1057
1058 /*
1059 * give the protocol a chance to keep things in sync.
1060 */
1061 if ((ifa = rt->rt_ifa) && ifa->ifa_rtrequest)
1062 ifa->ifa_rtrequest(RTM_DELETE, rt, info);
1063
1064 /*
1065 * One more rtentry floating around that is not
1066 * linked to the routing table. rttrash will be decremented
1067 * when RTFREE(rt) is eventually called.
1068 */
1069 V_rttrash++;
1070
1071 /*
1072 * If the caller wants it, then it can have it,
1073 * but it's up to it to free the rtentry as we won't be
1074 * doing it.
1075 */
1076 if (ret_nrt) {
1077 *ret_nrt = rt;
1078 RT_UNLOCK(rt);
1079 } else
1080 RTFREE_LOCKED(rt);
1081 break;
1082 case RTM_RESOLVE:
1083 /*
1084 * resolve was only used for route cloning
1085 * here for compat
1086 */
1087 break;
1088 case RTM_ADD:
1089 if ((flags & RTF_GATEWAY) && !gateway)
1090 senderr(EINVAL);
1091 if (dst && gateway && (dst->sa_family != gateway->sa_family) &&
1092 (gateway->sa_family != AF_UNSPEC) && (gateway->sa_family != AF_LINK))
1093 senderr(EINVAL);
1094
1095 if (info->rti_ifa == NULL) {
1096 error = rt_getifa_fib(info, fibnum);
1097 if (error)
1098 senderr(error);
1099 } else
1100 ifa_ref(info->rti_ifa);
1101 ifa = info->rti_ifa;
1102 rt = uma_zalloc(V_rtzone, M_NOWAIT | M_ZERO);
1103 if (rt == NULL) {
1104 if (ifa != NULL)
1105 ifa_free(ifa);
1106 senderr(ENOBUFS);
1107 }
1108 RT_LOCK_INIT(rt);
1109 rt->rt_flags = RTF_UP | flags;
1110 rt->rt_fibnum = fibnum;
1111 /*
1112 * Add the gateway. Possibly re-malloc-ing the storage for it
1113 *
1114 */
1115 RT_LOCK(rt);
1116 if ((error = rt_setgate(rt, dst, gateway)) != 0) {
1117 RT_LOCK_DESTROY(rt);
1118 if (ifa != NULL)
1119 ifa_free(ifa);
1120 uma_zfree(V_rtzone, rt);
1121 senderr(error);
1122 }
1123
1124 /*
1125 * point to the (possibly newly malloc'd) dest address.
1126 */
1127 ndst = (struct sockaddr *)rt_key(rt);
1128
1129 /*
1130 * make sure it contains the value we want (masked if needed).
1131 */
1132 if (netmask) {
1133 rt_maskedcopy(dst, ndst, netmask);
1134 } else
1135 bcopy(dst, ndst, dst->sa_len);
1136
1137 /*
1138 * We use the ifa reference returned by rt_getifa_fib().
1139 * This moved from below so that rnh->rnh_addaddr() can
1140 * examine the ifa and ifa->ifa_ifp if it so desires.
1141 */
1142 rt->rt_ifa = ifa;
1143 rt->rt_ifp = ifa->ifa_ifp;
1144 rt->rt_rmx.rmx_weight = 1;
1145
1146 #ifdef RADIX_MPATH
1147 /* do not permit exactly the same dst/mask/gw pair */
1148 if (rn_mpath_capable(rnh) &&
1149 rt_mpath_conflict(rnh, rt, netmask)) {
1150 if (rt->rt_ifa) {
1151 ifa_free(rt->rt_ifa);
1152 }
1153 Free(rt_key(rt));
1154 RT_LOCK_DESTROY(rt);
1155 uma_zfree(V_rtzone, rt);
1156 senderr(EEXIST);
1157 }
1158 #endif
1159
1160 #ifdef FLOWTABLE
1161 rt0 = NULL;
1162 /* XXX
1163 * "flow-table" only support IPv4 at the moment.
1164 */
1165 #ifdef INET
1166 if (dst->sa_family == AF_INET) {
1167 rn = rnh->rnh_matchaddr(dst, rnh);
1168 if (rn && ((rn->rn_flags & RNF_ROOT) == 0)) {
1169 struct sockaddr *mask;
1170 u_char *m, *n;
1171 int len;
1172
1173 /*
1174 * compare mask to see if the new route is
1175 * more specific than the existing one
1176 */
1177 rt0 = RNTORT(rn);
1178 RT_LOCK(rt0);
1179 RT_ADDREF(rt0);
1180 RT_UNLOCK(rt0);
1181 /*
1182 * A host route is already present, so
1183 * leave the flow-table entries as is.
1184 */
1185 if (rt0->rt_flags & RTF_HOST) {
1186 RTFREE(rt0);
1187 rt0 = NULL;
1188 } else if (!(flags & RTF_HOST) && netmask) {
1189 mask = rt_mask(rt0);
1190 len = mask->sa_len;
1191 m = (u_char *)mask;
1192 n = (u_char *)netmask;
1193 while (len-- > 0) {
1194 if (*n != *m)
1195 break;
1196 n++;
1197 m++;
1198 }
1199 if (len == 0 || (*n < *m)) {
1200 RTFREE(rt0);
1201 rt0 = NULL;
1202 }
1203 }
1204 }
1205 }
1206 #endif
1207 #endif
1208
1209 /* XXX mtu manipulation will be done in rnh_addaddr -- itojun */
1210 rn = rnh->rnh_addaddr(ndst, netmask, rnh, rt->rt_nodes);
1211 /*
1212 * If it still failed to go into the tree,
1213 * then un-make it (this should be a function)
1214 */
1215 if (rn == NULL) {
1216 if (rt->rt_ifa)
1217 ifa_free(rt->rt_ifa);
1218 Free(rt_key(rt));
1219 RT_LOCK_DESTROY(rt);
1220 uma_zfree(V_rtzone, rt);
1221 #ifdef FLOWTABLE
1222 if (rt0 != NULL)
1223 RTFREE(rt0);
1224 #endif
1225 senderr(EEXIST);
1226 }
1227 #ifdef FLOWTABLE
1228 else if (rt0 != NULL) {
1229 #ifdef INET
1230 flowtable_route_flush(V_ip_ft, rt0);
1231 #endif
1232 RTFREE(rt0);
1233 }
1234 #endif
1235
1236 /*
1237 * If this protocol has something to add to this then
1238 * allow it to do that as well.
1239 */
1240 if (ifa->ifa_rtrequest)
1241 ifa->ifa_rtrequest(req, rt, info);
1242
1243 /*
1244 * actually return a resultant rtentry and
1245 * give the caller a single reference.
1246 */
1247 if (ret_nrt) {
1248 *ret_nrt = rt;
1249 RT_ADDREF(rt);
1250 }
1251 RT_UNLOCK(rt);
1252 break;
1253 default:
1254 error = EOPNOTSUPP;
1255 }
1256 bad:
1257 if (needlock)
1258 RADIX_NODE_HEAD_UNLOCK(rnh);
1259 return (error);
1260 #undef senderr
1261 }
1262
1263 #undef dst
1264 #undef gateway
1265 #undef netmask
1266 #undef ifaaddr
1267 #undef ifpaddr
1268 #undef flags
1269
1270 int
1271 rt_setgate(struct rtentry *rt, struct sockaddr *dst, struct sockaddr *gate)
1272 {
1273 /* XXX dst may be overwritten, can we move this to below */
1274 int dlen = SA_SIZE(dst), glen = SA_SIZE(gate);
1275 #ifdef INVARIANTS
1276 struct radix_node_head *rnh;
1277
1278 rnh = rt_tables_get_rnh(rt->rt_fibnum, dst->sa_family);
1279 #endif
1280
1281 RT_LOCK_ASSERT(rt);
1282 RADIX_NODE_HEAD_LOCK_ASSERT(rnh);
1283
1284 /*
1285 * Prepare to store the gateway in rt->rt_gateway.
1286 * Both dst and gateway are stored one after the other in the same
1287 * malloc'd chunk. If we have room, we can reuse the old buffer,
1288 * rt_gateway already points to the right place.
1289 * Otherwise, malloc a new block and update the 'dst' address.
1290 */
1291 if (rt->rt_gateway == NULL || glen > SA_SIZE(rt->rt_gateway)) {
1292 caddr_t new;
1293
1294 R_Malloc(new, caddr_t, dlen + glen);
1295 if (new == NULL)
1296 return ENOBUFS;
1297 /*
1298 * XXX note, we copy from *dst and not *rt_key(rt) because
1299 * rt_setgate() can be called to initialize a newly
1300 * allocated route entry, in which case rt_key(rt) == NULL
1301 * (and also rt->rt_gateway == NULL).
1302 * Free()/free() handle a NULL argument just fine.
1303 */
1304 bcopy(dst, new, dlen);
1305 Free(rt_key(rt)); /* free old block, if any */
1306 rt_key(rt) = (struct sockaddr *)new;
1307 rt->rt_gateway = (struct sockaddr *)(new + dlen);
1308 }
1309
1310 /*
1311 * Copy the new gateway value into the memory chunk.
1312 */
1313 bcopy(gate, rt->rt_gateway, glen);
1314
1315 return (0);
1316 }
1317
1318 static void
1319 rt_maskedcopy(struct sockaddr *src, struct sockaddr *dst, struct sockaddr *netmask)
1320 {
1321 register u_char *cp1 = (u_char *)src;
1322 register u_char *cp2 = (u_char *)dst;
1323 register u_char *cp3 = (u_char *)netmask;
1324 u_char *cplim = cp2 + *cp3;
1325 u_char *cplim2 = cp2 + *cp1;
1326
1327 *cp2++ = *cp1++; *cp2++ = *cp1++; /* copies sa_len & sa_family */
1328 cp3 += 2;
1329 if (cplim > cplim2)
1330 cplim = cplim2;
1331 while (cp2 < cplim)
1332 *cp2++ = *cp1++ & *cp3++;
1333 if (cp2 < cplim2)
1334 bzero((caddr_t)cp2, (unsigned)(cplim2 - cp2));
1335 }
1336
1337 /*
1338 * Set up a routing table entry, normally
1339 * for an interface.
1340 */
1341 #define _SOCKADDR_TMPSIZE 128 /* Not too big.. kernel stack size is limited */
1342 static inline int
1343 rtinit1(struct ifaddr *ifa, int cmd, int flags, int fibnum)
1344 {
1345 struct sockaddr *dst;
1346 struct sockaddr *netmask;
1347 struct rtentry *rt = NULL;
1348 struct rt_addrinfo info;
1349 int error = 0;
1350 int startfib, endfib;
1351 char tempbuf[_SOCKADDR_TMPSIZE];
1352 int didwork = 0;
1353 int a_failure = 0;
1354 static struct sockaddr_dl null_sdl = {sizeof(null_sdl), AF_LINK};
1355
1356 if (flags & RTF_HOST) {
1357 dst = ifa->ifa_dstaddr;
1358 netmask = NULL;
1359 } else {
1360 dst = ifa->ifa_addr;
1361 netmask = ifa->ifa_netmask;
1362 }
1363 if ( dst->sa_family != AF_INET)
1364 fibnum = 0;
1365 if (fibnum == -1) {
1366 if (rt_add_addr_allfibs == 0 && cmd == (int)RTM_ADD) {
1367 startfib = endfib = curthread->td_proc->p_fibnum;
1368 } else {
1369 startfib = 0;
1370 endfib = rt_numfibs - 1;
1371 }
1372 } else {
1373 KASSERT((fibnum < rt_numfibs), ("rtinit1: bad fibnum"));
1374 startfib = fibnum;
1375 endfib = fibnum;
1376 }
1377 if (dst->sa_len == 0)
1378 return(EINVAL);
1379
1380 /*
1381 * If it's a delete, check that if it exists,
1382 * it's on the correct interface or we might scrub
1383 * a route to another ifa which would
1384 * be confusing at best and possibly worse.
1385 */
1386 if (cmd == RTM_DELETE) {
1387 /*
1388 * It's a delete, so it should already exist..
1389 * If it's a net, mask off the host bits
1390 * (Assuming we have a mask)
1391 * XXX this is kinda inet specific..
1392 */
1393 if (netmask != NULL) {
1394 rt_maskedcopy(dst, (struct sockaddr *)tempbuf, netmask);
1395 dst = (struct sockaddr *)tempbuf;
1396 }
1397 }
1398 /*
1399 * Now go through all the requested tables (fibs) and do the
1400 * requested action. Realistically, this will either be fib 0
1401 * for protocols that don't do multiple tables or all the
1402 * tables for those that do. XXX For this version only AF_INET.
1403 * When that changes code should be refactored to protocol
1404 * independent parts and protocol dependent parts.
1405 */
1406 for ( fibnum = startfib; fibnum <= endfib; fibnum++) {
1407 if (cmd == RTM_DELETE) {
1408 struct radix_node_head *rnh;
1409 struct radix_node *rn;
1410 /*
1411 * Look up an rtentry that is in the routing tree and
1412 * contains the correct info.
1413 */
1414 rnh = rt_tables_get_rnh(fibnum, dst->sa_family);
1415 if (rnh == NULL)
1416 /* this table doesn't exist but others might */
1417 continue;
1418 RADIX_NODE_HEAD_LOCK(rnh);
1419 #ifdef RADIX_MPATH
1420 if (rn_mpath_capable(rnh)) {
1421
1422 rn = rnh->rnh_matchaddr(dst, rnh);
1423 if (rn == NULL)
1424 error = ESRCH;
1425 else {
1426 rt = RNTORT(rn);
1427 /*
1428 * for interface route the
1429 * rt->rt_gateway is sockaddr_intf
1430 * for cloning ARP entries, so
1431 * rt_mpath_matchgate must use the
1432 * interface address
1433 */
1434 rt = rt_mpath_matchgate(rt,
1435 ifa->ifa_addr);
1436 if (!rt)
1437 error = ESRCH;
1438 }
1439 }
1440 else
1441 #endif
1442 rn = rnh->rnh_lookup(dst, netmask, rnh);
1443 error = (rn == NULL ||
1444 (rn->rn_flags & RNF_ROOT) ||
1445 RNTORT(rn)->rt_ifa != ifa ||
1446 !sa_equal((struct sockaddr *)rn->rn_key, dst));
1447 RADIX_NODE_HEAD_UNLOCK(rnh);
1448 if (error) {
1449 /* this is only an error if bad on ALL tables */
1450 continue;
1451 }
1452 }
1453 /*
1454 * Do the actual request
1455 */
1456 bzero((caddr_t)&info, sizeof(info));
1457 info.rti_ifa = ifa;
1458 info.rti_flags = flags | ifa->ifa_flags;
1459 info.rti_info[RTAX_DST] = dst;
1460 /*
1461 * doing this for compatibility reasons
1462 */
1463 if (cmd == RTM_ADD)
1464 info.rti_info[RTAX_GATEWAY] =
1465 (struct sockaddr *)&null_sdl;
1466 else
1467 info.rti_info[RTAX_GATEWAY] = ifa->ifa_addr;
1468 info.rti_info[RTAX_NETMASK] = netmask;
1469 error = rtrequest1_fib(cmd, &info, &rt, fibnum);
1470 if (error == 0 && rt != NULL) {
1471 /*
1472 * notify any listening routing agents of the change
1473 */
1474 RT_LOCK(rt);
1475 #ifdef RADIX_MPATH
1476 /*
1477 * in case address alias finds the first address
1478 * e.g. ifconfig bge0 192.103.54.246/24
1479 * e.g. ifconfig bge0 192.103.54.247/24
1480 * the address set in the route is 192.103.54.246
1481 * so we need to replace it with 192.103.54.247
1482 */
1483 if (memcmp(rt->rt_ifa->ifa_addr,
1484 ifa->ifa_addr, ifa->ifa_addr->sa_len)) {
1485 ifa_free(rt->rt_ifa);
1486 ifa_ref(ifa);
1487 rt->rt_ifp = ifa->ifa_ifp;
1488 rt->rt_ifa = ifa;
1489 }
1490 #endif
1491 /*
1492 * doing this for compatibility reasons
1493 */
1494 if (cmd == RTM_ADD) {
1495 ((struct sockaddr_dl *)rt->rt_gateway)->sdl_type =
1496 rt->rt_ifp->if_type;
1497 ((struct sockaddr_dl *)rt->rt_gateway)->sdl_index =
1498 rt->rt_ifp->if_index;
1499 }
1500 rt_newaddrmsg(cmd, ifa, error, rt);
1501 if (cmd == RTM_DELETE) {
1502 /*
1503 * If we are deleting, and we found an entry,
1504 * then it's been removed from the tree..
1505 * now throw it away.
1506 */
1507 RTFREE_LOCKED(rt);
1508 } else {
1509 if (cmd == RTM_ADD) {
1510 /*
1511 * We just wanted to add it..
1512 * we don't actually need a reference.
1513 */
1514 RT_REMREF(rt);
1515 }
1516 RT_UNLOCK(rt);
1517 }
1518 didwork = 1;
1519 }
1520 if (error)
1521 a_failure = error;
1522 }
1523 if (cmd == RTM_DELETE) {
1524 if (didwork) {
1525 error = 0;
1526 } else {
1527 /* we only give an error if it wasn't in any table */
1528 error = ((flags & RTF_HOST) ?
1529 EHOSTUNREACH : ENETUNREACH);
1530 }
1531 } else {
1532 if (a_failure) {
1533 /* return an error if any of them failed */
1534 error = a_failure;
1535 }
1536 }
1537 return (error);
1538 }
1539
1540 /* special one for inet internal use. may not use. */
1541 int
1542 rtinit_fib(struct ifaddr *ifa, int cmd, int flags)
1543 {
1544 return (rtinit1(ifa, cmd, flags, -1));
1545 }
1546
1547 /*
1548 * Set up a routing table entry, normally
1549 * for an interface.
1550 */
1551 int
1552 rtinit(struct ifaddr *ifa, int cmd, int flags)
1553 {
1554 struct sockaddr *dst;
1555 int fib = 0;
1556
1557 if (flags & RTF_HOST) {
1558 dst = ifa->ifa_dstaddr;
1559 } else {
1560 dst = ifa->ifa_addr;
1561 }
1562
1563 if (dst->sa_family == AF_INET)
1564 fib = -1;
1565 return (rtinit1(ifa, cmd, flags, fib));
1566 }
Cache object: f75de9d4963b21200d3d135907ffe2e1
|