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/10.2/sys/net/route.c 267728 2014-06-22 16:36:14Z tuexen $
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_inet6.h"
39 #include "opt_route.h"
40 #include "opt_sctp.h"
41 #include "opt_mrouting.h"
42 #include "opt_mpath.h"
43
44 #include <sys/param.h>
45 #include <sys/systm.h>
46 #include <sys/syslog.h>
47 #include <sys/malloc.h>
48 #include <sys/mbuf.h>
49 #include <sys/socket.h>
50 #include <sys/sysctl.h>
51 #include <sys/syslog.h>
52 #include <sys/sysproto.h>
53 #include <sys/proc.h>
54 #include <sys/domain.h>
55 #include <sys/kernel.h>
56
57 #include <net/if.h>
58 #include <net/if_dl.h>
59 #include <net/route.h>
60 #include <net/vnet.h>
61 #include <net/flowtable.h>
62
63 #ifdef RADIX_MPATH
64 #include <net/radix_mpath.h>
65 #endif
66
67 #include <netinet/in.h>
68 #include <netinet/ip_mroute.h>
69
70 #include <vm/uma.h>
71
72 #define RT_MAXFIBS UINT16_MAX
73
74 /* Kernel config default option. */
75 #ifdef ROUTETABLES
76 #if ROUTETABLES <= 0
77 #error "ROUTETABLES defined too low"
78 #endif
79 #if ROUTETABLES > RT_MAXFIBS
80 #error "ROUTETABLES defined too big"
81 #endif
82 #define RT_NUMFIBS ROUTETABLES
83 #endif /* ROUTETABLES */
84 /* Initialize to default if not otherwise set. */
85 #ifndef RT_NUMFIBS
86 #define RT_NUMFIBS 1
87 #endif
88
89 #if defined(INET) || defined(INET6)
90 #ifdef SCTP
91 extern void sctp_addr_change(struct ifaddr *ifa, int cmd);
92 #endif /* SCTP */
93 #endif
94
95
96 /* This is read-only.. */
97 u_int rt_numfibs = RT_NUMFIBS;
98 SYSCTL_UINT(_net, OID_AUTO, fibs, CTLFLAG_RD, &rt_numfibs, 0, "");
99 /* and this can be set too big but will be fixed before it is used */
100 TUNABLE_INT("net.fibs", &rt_numfibs);
101
102 /*
103 * By default add routes to all fibs for new interfaces.
104 * Once this is set to 0 then only allocate routes on interface
105 * changes for the FIB of the caller when adding a new set of addresses
106 * to an interface. XXX this is a shotgun aproach to a problem that needs
107 * a more fine grained solution.. that will come.
108 * XXX also has the problems getting the FIB from curthread which will not
109 * always work given the fib can be overridden and prefixes can be added
110 * from the network stack context.
111 */
112 u_int rt_add_addr_allfibs = 1;
113 SYSCTL_UINT(_net, OID_AUTO, add_addr_allfibs, CTLFLAG_RW,
114 &rt_add_addr_allfibs, 0, "");
115 TUNABLE_INT("net.add_addr_allfibs", &rt_add_addr_allfibs);
116
117 VNET_DEFINE(struct rtstat, rtstat);
118 #define V_rtstat VNET(rtstat)
119
120 VNET_DEFINE(struct radix_node_head *, rt_tables);
121 #define V_rt_tables VNET(rt_tables)
122
123 VNET_DEFINE(int, rttrash); /* routes not in table but not freed */
124 #define V_rttrash VNET(rttrash)
125
126
127 /* compare two sockaddr structures */
128 #define sa_equal(a1, a2) (((a1)->sa_len == (a2)->sa_len) && \
129 (bcmp((a1), (a2), (a1)->sa_len) == 0))
130
131 /*
132 * Convert a 'struct radix_node *' to a 'struct rtentry *'.
133 * The operation can be done safely (in this code) because a
134 * 'struct rtentry' starts with two 'struct radix_node''s, the first
135 * one representing leaf nodes in the routing tree, which is
136 * what the code in radix.c passes us as a 'struct radix_node'.
137 *
138 * But because there are a lot of assumptions in this conversion,
139 * do not cast explicitly, but always use the macro below.
140 */
141 #define RNTORT(p) ((struct rtentry *)(p))
142
143 static VNET_DEFINE(uma_zone_t, rtzone); /* Routing table UMA zone. */
144 #define V_rtzone VNET(rtzone)
145
146 /*
147 * handler for net.my_fibnum
148 */
149 static int
150 sysctl_my_fibnum(SYSCTL_HANDLER_ARGS)
151 {
152 int fibnum;
153 int error;
154
155 fibnum = curthread->td_proc->p_fibnum;
156 error = sysctl_handle_int(oidp, &fibnum, 0, req);
157 return (error);
158 }
159
160 SYSCTL_PROC(_net, OID_AUTO, my_fibnum, CTLTYPE_INT|CTLFLAG_RD,
161 NULL, 0, &sysctl_my_fibnum, "I", "default FIB of caller");
162
163 static __inline struct radix_node_head **
164 rt_tables_get_rnh_ptr(int table, int fam)
165 {
166 struct radix_node_head **rnh;
167
168 KASSERT(table >= 0 && table < rt_numfibs, ("%s: table out of bounds.",
169 __func__));
170 KASSERT(fam >= 0 && fam < (AF_MAX+1), ("%s: fam out of bounds.",
171 __func__));
172
173 /* rnh is [fib=0][af=0]. */
174 rnh = (struct radix_node_head **)V_rt_tables;
175 /* Get the offset to the requested table and fam. */
176 rnh += table * (AF_MAX+1) + fam;
177
178 return (rnh);
179 }
180
181 struct radix_node_head *
182 rt_tables_get_rnh(int table, int fam)
183 {
184
185 return (*rt_tables_get_rnh_ptr(table, fam));
186 }
187
188 /*
189 * route initialization must occur before ip6_init2(), which happenas at
190 * SI_ORDER_MIDDLE.
191 */
192 static void
193 route_init(void)
194 {
195 struct domain *dom;
196 int max_keylen = 0;
197
198 /* whack the tunable ints into line. */
199 if (rt_numfibs > RT_MAXFIBS)
200 rt_numfibs = RT_MAXFIBS;
201 if (rt_numfibs == 0)
202 rt_numfibs = 1;
203
204 for (dom = domains; dom; dom = dom->dom_next)
205 if (dom->dom_maxrtkey > max_keylen)
206 max_keylen = dom->dom_maxrtkey;
207
208 rn_init(max_keylen); /* init all zeroes, all ones, mask table */
209 }
210 SYSINIT(route_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_THIRD, route_init, 0);
211
212 static int
213 rtentry_zinit(void *mem, int size, int how)
214 {
215 struct rtentry *rt = mem;
216
217 rt->rt_pksent = counter_u64_alloc(how);
218 if (rt->rt_pksent == NULL)
219 return (ENOMEM);
220
221 RT_LOCK_INIT(rt);
222
223 return (0);
224 }
225
226 static void
227 rtentry_zfini(void *mem, int size)
228 {
229 struct rtentry *rt = mem;
230
231 RT_LOCK_DESTROY(rt);
232 counter_u64_free(rt->rt_pksent);
233 }
234
235 static int
236 rtentry_ctor(void *mem, int size, void *arg, int how)
237 {
238 struct rtentry *rt = mem;
239
240 bzero(rt, offsetof(struct rtentry, rt_endzero));
241 counter_u64_zero(rt->rt_pksent);
242
243 return (0);
244 }
245
246 static void
247 rtentry_dtor(void *mem, int size, void *arg)
248 {
249 struct rtentry *rt = mem;
250
251 RT_UNLOCK_COND(rt);
252 }
253
254 static void
255 vnet_route_init(const void *unused __unused)
256 {
257 struct domain *dom;
258 struct radix_node_head **rnh;
259 int table;
260 int fam;
261
262 V_rt_tables = malloc(rt_numfibs * (AF_MAX+1) *
263 sizeof(struct radix_node_head *), M_RTABLE, M_WAITOK|M_ZERO);
264
265 V_rtzone = uma_zcreate("rtentry", sizeof(struct rtentry),
266 rtentry_ctor, rtentry_dtor,
267 rtentry_zinit, rtentry_zfini, UMA_ALIGN_PTR, 0);
268 for (dom = domains; dom; dom = dom->dom_next) {
269 if (dom->dom_rtattach == NULL)
270 continue;
271
272 for (table = 0; table < rt_numfibs; table++) {
273 fam = dom->dom_family;
274 if (table != 0 && fam != AF_INET6 && fam != AF_INET)
275 break;
276
277 /*
278 * XXX MRT rtattach will be also called from
279 * vfs_export.c but the offset will be 0 (only for
280 * AF_INET and AF_INET6 which don't need it anyhow).
281 */
282 rnh = rt_tables_get_rnh_ptr(table, fam);
283 if (rnh == NULL)
284 panic("%s: rnh NULL", __func__);
285 dom->dom_rtattach((void **)rnh, dom->dom_rtoffset);
286 }
287 }
288 }
289 VNET_SYSINIT(vnet_route_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_FOURTH,
290 vnet_route_init, 0);
291
292 #ifdef VIMAGE
293 static void
294 vnet_route_uninit(const void *unused __unused)
295 {
296 int table;
297 int fam;
298 struct domain *dom;
299 struct radix_node_head **rnh;
300
301 for (dom = domains; dom; dom = dom->dom_next) {
302 if (dom->dom_rtdetach == NULL)
303 continue;
304
305 for (table = 0; table < rt_numfibs; table++) {
306 fam = dom->dom_family;
307
308 if (table != 0 && fam != AF_INET6 && fam != AF_INET)
309 break;
310
311 rnh = rt_tables_get_rnh_ptr(table, fam);
312 if (rnh == NULL)
313 panic("%s: rnh NULL", __func__);
314 dom->dom_rtdetach((void **)rnh, dom->dom_rtoffset);
315 }
316 }
317
318 free(V_rt_tables, M_RTABLE);
319 uma_zdestroy(V_rtzone);
320 }
321 VNET_SYSUNINIT(vnet_route_uninit, SI_SUB_PROTO_DOMAIN, SI_ORDER_THIRD,
322 vnet_route_uninit, 0);
323 #endif
324
325 #ifndef _SYS_SYSPROTO_H_
326 struct setfib_args {
327 int fibnum;
328 };
329 #endif
330 int
331 sys_setfib(struct thread *td, struct setfib_args *uap)
332 {
333 if (uap->fibnum < 0 || uap->fibnum >= rt_numfibs)
334 return EINVAL;
335 td->td_proc->p_fibnum = uap->fibnum;
336 return (0);
337 }
338
339 /*
340 * Packet routing routines.
341 */
342 void
343 rtalloc(struct route *ro)
344 {
345
346 rtalloc_ign_fib(ro, 0UL, RT_DEFAULT_FIB);
347 }
348
349 void
350 rtalloc_fib(struct route *ro, u_int fibnum)
351 {
352 rtalloc_ign_fib(ro, 0UL, fibnum);
353 }
354
355 void
356 rtalloc_ign(struct route *ro, u_long ignore)
357 {
358 struct rtentry *rt;
359
360 if ((rt = ro->ro_rt) != NULL) {
361 if (rt->rt_ifp != NULL && rt->rt_flags & RTF_UP)
362 return;
363 RTFREE(rt);
364 ro->ro_rt = NULL;
365 }
366 ro->ro_rt = rtalloc1_fib(&ro->ro_dst, 1, ignore, RT_DEFAULT_FIB);
367 if (ro->ro_rt)
368 RT_UNLOCK(ro->ro_rt);
369 }
370
371 void
372 rtalloc_ign_fib(struct route *ro, u_long ignore, u_int fibnum)
373 {
374 struct rtentry *rt;
375
376 if ((rt = ro->ro_rt) != NULL) {
377 if (rt->rt_ifp != NULL && rt->rt_flags & RTF_UP)
378 return;
379 RTFREE(rt);
380 ro->ro_rt = NULL;
381 }
382 ro->ro_rt = rtalloc1_fib(&ro->ro_dst, 1, ignore, fibnum);
383 if (ro->ro_rt)
384 RT_UNLOCK(ro->ro_rt);
385 }
386
387 /*
388 * Look up the route that matches the address given
389 * Or, at least try.. Create a cloned route if needed.
390 *
391 * The returned route, if any, is locked.
392 */
393 struct rtentry *
394 rtalloc1(struct sockaddr *dst, int report, u_long ignflags)
395 {
396
397 return (rtalloc1_fib(dst, report, ignflags, RT_DEFAULT_FIB));
398 }
399
400 struct rtentry *
401 rtalloc1_fib(struct sockaddr *dst, int report, u_long ignflags,
402 u_int fibnum)
403 {
404 struct radix_node_head *rnh;
405 struct radix_node *rn;
406 struct rtentry *newrt;
407 struct rt_addrinfo info;
408 int err = 0, msgtype = RTM_MISS;
409 int needlock;
410
411 KASSERT((fibnum < rt_numfibs), ("rtalloc1_fib: bad fibnum"));
412 switch (dst->sa_family) {
413 case AF_INET6:
414 case AF_INET:
415 /* We support multiple FIBs. */
416 break;
417 default:
418 fibnum = RT_DEFAULT_FIB;
419 break;
420 }
421 rnh = rt_tables_get_rnh(fibnum, dst->sa_family);
422 newrt = NULL;
423 if (rnh == NULL)
424 goto miss;
425
426 /*
427 * Look up the address in the table for that Address Family
428 */
429 needlock = !(ignflags & RTF_RNH_LOCKED);
430 if (needlock)
431 RADIX_NODE_HEAD_RLOCK(rnh);
432 #ifdef INVARIANTS
433 else
434 RADIX_NODE_HEAD_LOCK_ASSERT(rnh);
435 #endif
436 rn = rnh->rnh_matchaddr(dst, rnh);
437 if (rn && ((rn->rn_flags & RNF_ROOT) == 0)) {
438 newrt = RNTORT(rn);
439 RT_LOCK(newrt);
440 RT_ADDREF(newrt);
441 if (needlock)
442 RADIX_NODE_HEAD_RUNLOCK(rnh);
443 goto done;
444
445 } else if (needlock)
446 RADIX_NODE_HEAD_RUNLOCK(rnh);
447
448 /*
449 * Either we hit the root or couldn't find any match,
450 * Which basically means
451 * "caint get there frm here"
452 */
453 miss:
454 V_rtstat.rts_unreach++;
455
456 if (report) {
457 /*
458 * If required, report the failure to the supervising
459 * Authorities.
460 * For a delete, this is not an error. (report == 0)
461 */
462 bzero(&info, sizeof(info));
463 info.rti_info[RTAX_DST] = dst;
464 rt_missmsg_fib(msgtype, &info, 0, err, fibnum);
465 }
466 done:
467 if (newrt)
468 RT_LOCK_ASSERT(newrt);
469 return (newrt);
470 }
471
472 /*
473 * Remove a reference count from an rtentry.
474 * If the count gets low enough, take it out of the routing table
475 */
476 void
477 rtfree(struct rtentry *rt)
478 {
479 struct radix_node_head *rnh;
480
481 KASSERT(rt != NULL,("%s: NULL rt", __func__));
482 rnh = rt_tables_get_rnh(rt->rt_fibnum, rt_key(rt)->sa_family);
483 KASSERT(rnh != NULL,("%s: NULL rnh", __func__));
484
485 RT_LOCK_ASSERT(rt);
486
487 /*
488 * The callers should use RTFREE_LOCKED() or RTFREE(), so
489 * we should come here exactly with the last reference.
490 */
491 RT_REMREF(rt);
492 if (rt->rt_refcnt > 0) {
493 log(LOG_DEBUG, "%s: %p has %d refs\n", __func__, rt, rt->rt_refcnt);
494 goto done;
495 }
496
497 /*
498 * On last reference give the "close method" a chance
499 * to cleanup private state. This also permits (for
500 * IPv4 and IPv6) a chance to decide if the routing table
501 * entry should be purged immediately or at a later time.
502 * When an immediate purge is to happen the close routine
503 * typically calls rtexpunge which clears the RTF_UP flag
504 * on the entry so that the code below reclaims the storage.
505 */
506 if (rt->rt_refcnt == 0 && rnh->rnh_close)
507 rnh->rnh_close((struct radix_node *)rt, rnh);
508
509 /*
510 * If we are no longer "up" (and ref == 0)
511 * then we can free the resources associated
512 * with the route.
513 */
514 if ((rt->rt_flags & RTF_UP) == 0) {
515 if (rt->rt_nodes->rn_flags & (RNF_ACTIVE | RNF_ROOT))
516 panic("rtfree 2");
517 /*
518 * the rtentry must have been removed from the routing table
519 * so it is represented in rttrash.. remove that now.
520 */
521 V_rttrash--;
522 #ifdef DIAGNOSTIC
523 if (rt->rt_refcnt < 0) {
524 printf("rtfree: %p not freed (neg refs)\n", rt);
525 goto done;
526 }
527 #endif
528 /*
529 * release references on items we hold them on..
530 * e.g other routes and ifaddrs.
531 */
532 if (rt->rt_ifa)
533 ifa_free(rt->rt_ifa);
534 /*
535 * The key is separatly alloc'd so free it (see rt_setgate()).
536 * This also frees the gateway, as they are always malloc'd
537 * together.
538 */
539 Free(rt_key(rt));
540
541 /*
542 * and the rtentry itself of course
543 */
544 uma_zfree(V_rtzone, rt);
545 return;
546 }
547 done:
548 RT_UNLOCK(rt);
549 }
550
551
552 /*
553 * Force a routing table entry to the specified
554 * destination to go through the given gateway.
555 * Normally called as a result of a routing redirect
556 * message from the network layer.
557 */
558 void
559 rtredirect(struct sockaddr *dst,
560 struct sockaddr *gateway,
561 struct sockaddr *netmask,
562 int flags,
563 struct sockaddr *src)
564 {
565
566 rtredirect_fib(dst, gateway, netmask, flags, src, RT_DEFAULT_FIB);
567 }
568
569 void
570 rtredirect_fib(struct sockaddr *dst,
571 struct sockaddr *gateway,
572 struct sockaddr *netmask,
573 int flags,
574 struct sockaddr *src,
575 u_int fibnum)
576 {
577 struct rtentry *rt, *rt0 = NULL;
578 int error = 0;
579 short *stat = NULL;
580 struct rt_addrinfo info;
581 struct ifaddr *ifa;
582 struct radix_node_head *rnh;
583
584 ifa = NULL;
585 rnh = rt_tables_get_rnh(fibnum, dst->sa_family);
586 if (rnh == NULL) {
587 error = EAFNOSUPPORT;
588 goto out;
589 }
590
591 /* verify the gateway is directly reachable */
592 if ((ifa = ifa_ifwithnet_fib(gateway, 0, fibnum)) == NULL) {
593 error = ENETUNREACH;
594 goto out;
595 }
596 rt = rtalloc1_fib(dst, 0, 0UL, fibnum); /* NB: rt is locked */
597 /*
598 * If the redirect isn't from our current router for this dst,
599 * it's either old or wrong. If it redirects us to ourselves,
600 * we have a routing loop, perhaps as a result of an interface
601 * going down recently.
602 */
603 if (!(flags & RTF_DONE) && rt &&
604 (!sa_equal(src, rt->rt_gateway) || rt->rt_ifa != ifa))
605 error = EINVAL;
606 else if (ifa_ifwithaddr_check(gateway))
607 error = EHOSTUNREACH;
608 if (error)
609 goto done;
610 /*
611 * Create a new entry if we just got back a wildcard entry
612 * or the lookup failed. This is necessary for hosts
613 * which use routing redirects generated by smart gateways
614 * to dynamically build the routing tables.
615 */
616 if (rt == NULL || (rt_mask(rt) && rt_mask(rt)->sa_len < 2))
617 goto create;
618 /*
619 * Don't listen to the redirect if it's
620 * for a route to an interface.
621 */
622 if (rt->rt_flags & RTF_GATEWAY) {
623 if (((rt->rt_flags & RTF_HOST) == 0) && (flags & RTF_HOST)) {
624 /*
625 * Changing from route to net => route to host.
626 * Create new route, rather than smashing route to net.
627 */
628 create:
629 rt0 = rt;
630 rt = NULL;
631
632 flags |= RTF_GATEWAY | RTF_DYNAMIC;
633 bzero((caddr_t)&info, sizeof(info));
634 info.rti_info[RTAX_DST] = dst;
635 info.rti_info[RTAX_GATEWAY] = gateway;
636 info.rti_info[RTAX_NETMASK] = netmask;
637 info.rti_ifa = ifa;
638 info.rti_flags = flags;
639 if (rt0 != NULL)
640 RT_UNLOCK(rt0); /* drop lock to avoid LOR with RNH */
641 error = rtrequest1_fib(RTM_ADD, &info, &rt, fibnum);
642 if (rt != NULL) {
643 RT_LOCK(rt);
644 if (rt0 != NULL)
645 EVENTHANDLER_INVOKE(route_redirect_event, rt0, rt, dst);
646 flags = rt->rt_flags;
647 }
648 if (rt0 != NULL)
649 RTFREE(rt0);
650
651 stat = &V_rtstat.rts_dynamic;
652 } else {
653 struct rtentry *gwrt;
654
655 /*
656 * Smash the current notion of the gateway to
657 * this destination. Should check about netmask!!!
658 */
659 rt->rt_flags |= RTF_MODIFIED;
660 flags |= RTF_MODIFIED;
661 stat = &V_rtstat.rts_newgateway;
662 /*
663 * add the key and gateway (in one malloc'd chunk).
664 */
665 RT_UNLOCK(rt);
666 RADIX_NODE_HEAD_LOCK(rnh);
667 RT_LOCK(rt);
668 rt_setgate(rt, rt_key(rt), gateway);
669 gwrt = rtalloc1(gateway, 1, RTF_RNH_LOCKED);
670 RADIX_NODE_HEAD_UNLOCK(rnh);
671 EVENTHANDLER_INVOKE(route_redirect_event, rt, gwrt, dst);
672 RTFREE_LOCKED(gwrt);
673 }
674 } else
675 error = EHOSTUNREACH;
676 done:
677 if (rt)
678 RTFREE_LOCKED(rt);
679 out:
680 if (error)
681 V_rtstat.rts_badredirect++;
682 else if (stat != NULL)
683 (*stat)++;
684 bzero((caddr_t)&info, sizeof(info));
685 info.rti_info[RTAX_DST] = dst;
686 info.rti_info[RTAX_GATEWAY] = gateway;
687 info.rti_info[RTAX_NETMASK] = netmask;
688 info.rti_info[RTAX_AUTHOR] = src;
689 rt_missmsg_fib(RTM_REDIRECT, &info, flags, error, fibnum);
690 if (ifa != NULL)
691 ifa_free(ifa);
692 }
693
694 int
695 rtioctl(u_long req, caddr_t data)
696 {
697
698 return (rtioctl_fib(req, data, RT_DEFAULT_FIB));
699 }
700
701 /*
702 * Routing table ioctl interface.
703 */
704 int
705 rtioctl_fib(u_long req, caddr_t data, u_int fibnum)
706 {
707
708 /*
709 * If more ioctl commands are added here, make sure the proper
710 * super-user checks are being performed because it is possible for
711 * prison-root to make it this far if raw sockets have been enabled
712 * in jails.
713 */
714 #ifdef INET
715 /* Multicast goop, grrr... */
716 return mrt_ioctl ? mrt_ioctl(req, data, fibnum) : EOPNOTSUPP;
717 #else /* INET */
718 return ENXIO;
719 #endif /* INET */
720 }
721
722 /*
723 * For both ifa_ifwithroute() routines, 'ifa' is returned referenced.
724 */
725 struct ifaddr *
726 ifa_ifwithroute(int flags, struct sockaddr *dst, struct sockaddr *gateway)
727 {
728
729 return (ifa_ifwithroute_fib(flags, dst, gateway, RT_DEFAULT_FIB));
730 }
731
732 struct ifaddr *
733 ifa_ifwithroute_fib(int flags, struct sockaddr *dst, struct sockaddr *gateway,
734 u_int fibnum)
735 {
736 register struct ifaddr *ifa;
737 int not_found = 0;
738
739 if ((flags & RTF_GATEWAY) == 0) {
740 /*
741 * If we are adding a route to an interface,
742 * and the interface is a pt to pt link
743 * we should search for the destination
744 * as our clue to the interface. Otherwise
745 * we can use the local address.
746 */
747 ifa = NULL;
748 if (flags & RTF_HOST)
749 ifa = ifa_ifwithdstaddr_fib(dst, fibnum);
750 if (ifa == NULL)
751 ifa = ifa_ifwithaddr(gateway);
752 } else {
753 /*
754 * If we are adding a route to a remote net
755 * or host, the gateway may still be on the
756 * other end of a pt to pt link.
757 */
758 ifa = ifa_ifwithdstaddr_fib(gateway, fibnum);
759 }
760 if (ifa == NULL)
761 ifa = ifa_ifwithnet_fib(gateway, 0, fibnum);
762 if (ifa == NULL) {
763 struct rtentry *rt = rtalloc1_fib(gateway, 0, RTF_RNH_LOCKED, fibnum);
764 if (rt == NULL)
765 return (NULL);
766 /*
767 * dismiss a gateway that is reachable only
768 * through the default router
769 */
770 switch (gateway->sa_family) {
771 case AF_INET:
772 if (satosin(rt_key(rt))->sin_addr.s_addr == INADDR_ANY)
773 not_found = 1;
774 break;
775 case AF_INET6:
776 if (IN6_IS_ADDR_UNSPECIFIED(&satosin6(rt_key(rt))->sin6_addr))
777 not_found = 1;
778 break;
779 default:
780 break;
781 }
782 if (!not_found && rt->rt_ifa != NULL) {
783 ifa = rt->rt_ifa;
784 ifa_ref(ifa);
785 }
786 RT_REMREF(rt);
787 RT_UNLOCK(rt);
788 if (not_found || ifa == NULL)
789 return (NULL);
790 }
791 if (ifa->ifa_addr->sa_family != dst->sa_family) {
792 struct ifaddr *oifa = ifa;
793 ifa = ifaof_ifpforaddr(dst, ifa->ifa_ifp);
794 if (ifa == NULL)
795 ifa = oifa;
796 else
797 ifa_free(oifa);
798 }
799 return (ifa);
800 }
801
802 /*
803 * Do appropriate manipulations of a routing tree given
804 * all the bits of info needed
805 */
806 int
807 rtrequest(int req,
808 struct sockaddr *dst,
809 struct sockaddr *gateway,
810 struct sockaddr *netmask,
811 int flags,
812 struct rtentry **ret_nrt)
813 {
814
815 return (rtrequest_fib(req, dst, gateway, netmask, flags, ret_nrt,
816 RT_DEFAULT_FIB));
817 }
818
819 int
820 rtrequest_fib(int req,
821 struct sockaddr *dst,
822 struct sockaddr *gateway,
823 struct sockaddr *netmask,
824 int flags,
825 struct rtentry **ret_nrt,
826 u_int fibnum)
827 {
828 struct rt_addrinfo info;
829
830 if (dst->sa_len == 0)
831 return(EINVAL);
832
833 bzero((caddr_t)&info, sizeof(info));
834 info.rti_flags = flags;
835 info.rti_info[RTAX_DST] = dst;
836 info.rti_info[RTAX_GATEWAY] = gateway;
837 info.rti_info[RTAX_NETMASK] = netmask;
838 return rtrequest1_fib(req, &info, ret_nrt, fibnum);
839 }
840
841 /*
842 * These (questionable) definitions of apparent local variables apply
843 * to the next two functions. XXXXXX!!!
844 */
845 #define dst info->rti_info[RTAX_DST]
846 #define gateway info->rti_info[RTAX_GATEWAY]
847 #define netmask info->rti_info[RTAX_NETMASK]
848 #define ifaaddr info->rti_info[RTAX_IFA]
849 #define ifpaddr info->rti_info[RTAX_IFP]
850 #define flags info->rti_flags
851
852 int
853 rt_getifa(struct rt_addrinfo *info)
854 {
855
856 return (rt_getifa_fib(info, RT_DEFAULT_FIB));
857 }
858
859 /*
860 * Look up rt_addrinfo for a specific fib. Note that if rti_ifa is defined,
861 * it will be referenced so the caller must free it.
862 */
863 int
864 rt_getifa_fib(struct rt_addrinfo *info, u_int fibnum)
865 {
866 struct ifaddr *ifa;
867 int error = 0;
868
869 /*
870 * ifp may be specified by sockaddr_dl
871 * when protocol address is ambiguous.
872 */
873 if (info->rti_ifp == NULL && ifpaddr != NULL &&
874 ifpaddr->sa_family == AF_LINK &&
875 (ifa = ifa_ifwithnet_fib(ifpaddr, 0, fibnum)) != NULL) {
876 info->rti_ifp = ifa->ifa_ifp;
877 ifa_free(ifa);
878 }
879 if (info->rti_ifa == NULL && ifaaddr != NULL)
880 info->rti_ifa = ifa_ifwithaddr(ifaaddr);
881 if (info->rti_ifa == NULL) {
882 struct sockaddr *sa;
883
884 sa = ifaaddr != NULL ? ifaaddr :
885 (gateway != NULL ? gateway : dst);
886 if (sa != NULL && info->rti_ifp != NULL)
887 info->rti_ifa = ifaof_ifpforaddr(sa, info->rti_ifp);
888 else if (dst != NULL && gateway != NULL)
889 info->rti_ifa = ifa_ifwithroute_fib(flags, dst, gateway,
890 fibnum);
891 else if (sa != NULL)
892 info->rti_ifa = ifa_ifwithroute_fib(flags, sa, sa,
893 fibnum);
894 }
895 if ((ifa = info->rti_ifa) != NULL) {
896 if (info->rti_ifp == NULL)
897 info->rti_ifp = ifa->ifa_ifp;
898 } else
899 error = ENETUNREACH;
900 return (error);
901 }
902
903 /*
904 * Expunges references to a route that's about to be reclaimed.
905 * The route must be locked.
906 */
907 int
908 rtexpunge(struct rtentry *rt)
909 {
910 #if !defined(RADIX_MPATH)
911 struct radix_node *rn;
912 #else
913 struct rt_addrinfo info;
914 int fib;
915 struct rtentry *rt0;
916 #endif
917 struct radix_node_head *rnh;
918 struct ifaddr *ifa;
919 int error = 0;
920
921 /*
922 * Find the correct routing tree to use for this Address Family
923 */
924 rnh = rt_tables_get_rnh(rt->rt_fibnum, rt_key(rt)->sa_family);
925 RT_LOCK_ASSERT(rt);
926 if (rnh == NULL)
927 return (EAFNOSUPPORT);
928 RADIX_NODE_HEAD_LOCK_ASSERT(rnh);
929
930 #ifdef RADIX_MPATH
931 fib = rt->rt_fibnum;
932 bzero(&info, sizeof(info));
933 info.rti_ifp = rt->rt_ifp;
934 info.rti_flags = RTF_RNH_LOCKED;
935 info.rti_info[RTAX_DST] = rt_key(rt);
936 info.rti_info[RTAX_GATEWAY] = rt->rt_ifa->ifa_addr;
937
938 RT_UNLOCK(rt);
939 error = rtrequest1_fib(RTM_DELETE, &info, &rt0, fib);
940
941 if (error == 0 && rt0 != NULL) {
942 rt = rt0;
943 RT_LOCK(rt);
944 } else if (error != 0) {
945 RT_LOCK(rt);
946 return (error);
947 }
948 #else
949 /*
950 * Remove the item from the tree; it should be there,
951 * but when callers invoke us blindly it may not (sigh).
952 */
953 rn = rnh->rnh_deladdr(rt_key(rt), rt_mask(rt), rnh);
954 if (rn == NULL) {
955 error = ESRCH;
956 goto bad;
957 }
958 KASSERT((rn->rn_flags & (RNF_ACTIVE | RNF_ROOT)) == 0,
959 ("unexpected flags 0x%x", rn->rn_flags));
960 KASSERT(rt == RNTORT(rn),
961 ("lookup mismatch, rt %p rn %p", rt, rn));
962 #endif /* RADIX_MPATH */
963
964 rt->rt_flags &= ~RTF_UP;
965
966 /*
967 * Give the protocol a chance to keep things in sync.
968 */
969 if ((ifa = rt->rt_ifa) && ifa->ifa_rtrequest) {
970 struct rt_addrinfo info;
971
972 bzero((caddr_t)&info, sizeof(info));
973 info.rti_flags = rt->rt_flags;
974 info.rti_info[RTAX_DST] = rt_key(rt);
975 info.rti_info[RTAX_GATEWAY] = rt->rt_gateway;
976 info.rti_info[RTAX_NETMASK] = rt_mask(rt);
977 ifa->ifa_rtrequest(RTM_DELETE, rt, &info);
978 }
979
980 /*
981 * one more rtentry floating around that is not
982 * linked to the routing table.
983 */
984 V_rttrash++;
985 #if !defined(RADIX_MPATH)
986 bad:
987 #endif
988 return (error);
989 }
990
991 #if 0
992 int p_sockaddr(char *buf, int buflen, struct sockaddr *s);
993 int rt_print(char *buf, int buflen, struct rtentry *rt);
994
995 int
996 p_sockaddr(char *buf, int buflen, struct sockaddr *s)
997 {
998 void *paddr = NULL;
999
1000 switch (s->sa_family) {
1001 case AF_INET:
1002 paddr = &((struct sockaddr_in *)s)->sin_addr;
1003 break;
1004 case AF_INET6:
1005 paddr = &((struct sockaddr_in6 *)s)->sin6_addr;
1006 break;
1007 }
1008
1009 if (paddr == NULL)
1010 return (0);
1011
1012 if (inet_ntop(s->sa_family, paddr, buf, buflen) == NULL)
1013 return (0);
1014
1015 return (strlen(buf));
1016 }
1017
1018 int
1019 rt_print(char *buf, int buflen, struct rtentry *rt)
1020 {
1021 struct sockaddr *addr, *mask;
1022 int i = 0;
1023
1024 addr = rt_key(rt);
1025 mask = rt_mask(rt);
1026
1027 i = p_sockaddr(buf, buflen, addr);
1028 if (!(rt->rt_flags & RTF_HOST)) {
1029 buf[i++] = '/';
1030 i += p_sockaddr(buf + i, buflen - i, mask);
1031 }
1032
1033 if (rt->rt_flags & RTF_GATEWAY) {
1034 buf[i++] = '>';
1035 i += p_sockaddr(buf + i, buflen - i, rt->rt_gateway);
1036 }
1037
1038 return (i);
1039 }
1040 #endif
1041
1042 #ifdef RADIX_MPATH
1043 static int
1044 rn_mpath_update(int req, struct rt_addrinfo *info,
1045 struct radix_node_head *rnh, struct rtentry **ret_nrt)
1046 {
1047 /*
1048 * if we got multipath routes, we require users to specify
1049 * a matching RTAX_GATEWAY.
1050 */
1051 struct rtentry *rt, *rto = NULL;
1052 register struct radix_node *rn;
1053 int error = 0;
1054
1055 rn = rnh->rnh_lookup(dst, netmask, rnh);
1056 if (rn == NULL)
1057 return (ESRCH);
1058 rto = rt = RNTORT(rn);
1059
1060 rt = rt_mpath_matchgate(rt, gateway);
1061 if (rt == NULL)
1062 return (ESRCH);
1063 /*
1064 * this is the first entry in the chain
1065 */
1066 if (rto == rt) {
1067 rn = rn_mpath_next((struct radix_node *)rt);
1068 /*
1069 * there is another entry, now it's active
1070 */
1071 if (rn) {
1072 rto = RNTORT(rn);
1073 RT_LOCK(rto);
1074 rto->rt_flags |= RTF_UP;
1075 RT_UNLOCK(rto);
1076 } else if (rt->rt_flags & RTF_GATEWAY) {
1077 /*
1078 * For gateway routes, we need to
1079 * make sure that we we are deleting
1080 * the correct gateway.
1081 * rt_mpath_matchgate() does not
1082 * check the case when there is only
1083 * one route in the chain.
1084 */
1085 if (gateway &&
1086 (rt->rt_gateway->sa_len != gateway->sa_len ||
1087 memcmp(rt->rt_gateway, gateway, gateway->sa_len)))
1088 error = ESRCH;
1089 else {
1090 /*
1091 * remove from tree before returning it
1092 * to the caller
1093 */
1094 rn = rnh->rnh_deladdr(dst, netmask, rnh);
1095 KASSERT(rt == RNTORT(rn), ("radix node disappeared"));
1096 goto gwdelete;
1097 }
1098
1099 }
1100 /*
1101 * use the normal delete code to remove
1102 * the first entry
1103 */
1104 if (req != RTM_DELETE)
1105 goto nondelete;
1106
1107 error = ENOENT;
1108 goto done;
1109 }
1110
1111 /*
1112 * if the entry is 2nd and on up
1113 */
1114 if ((req == RTM_DELETE) && !rt_mpath_deldup(rto, rt))
1115 panic ("rtrequest1: rt_mpath_deldup");
1116 gwdelete:
1117 RT_LOCK(rt);
1118 RT_ADDREF(rt);
1119 if (req == RTM_DELETE) {
1120 rt->rt_flags &= ~RTF_UP;
1121 /*
1122 * One more rtentry floating around that is not
1123 * linked to the routing table. rttrash will be decremented
1124 * when RTFREE(rt) is eventually called.
1125 */
1126 V_rttrash++;
1127 }
1128
1129 nondelete:
1130 if (req != RTM_DELETE)
1131 panic("unrecognized request %d", req);
1132
1133
1134 /*
1135 * If the caller wants it, then it can have it,
1136 * but it's up to it to free the rtentry as we won't be
1137 * doing it.
1138 */
1139 if (ret_nrt) {
1140 *ret_nrt = rt;
1141 RT_UNLOCK(rt);
1142 } else
1143 RTFREE_LOCKED(rt);
1144 done:
1145 return (error);
1146 }
1147 #endif
1148
1149 int
1150 rtrequest1_fib(int req, struct rt_addrinfo *info, struct rtentry **ret_nrt,
1151 u_int fibnum)
1152 {
1153 int error = 0, needlock = 0;
1154 register struct rtentry *rt;
1155 #ifdef FLOWTABLE
1156 register struct rtentry *rt0;
1157 #endif
1158 register struct radix_node *rn;
1159 register struct radix_node_head *rnh;
1160 struct ifaddr *ifa;
1161 struct sockaddr *ndst;
1162 struct sockaddr_storage mdst;
1163 #define senderr(x) { error = x ; goto bad; }
1164
1165 KASSERT((fibnum < rt_numfibs), ("rtrequest1_fib: bad fibnum"));
1166 switch (dst->sa_family) {
1167 case AF_INET6:
1168 case AF_INET:
1169 /* We support multiple FIBs. */
1170 break;
1171 default:
1172 fibnum = RT_DEFAULT_FIB;
1173 break;
1174 }
1175
1176 /*
1177 * Find the correct routing tree to use for this Address Family
1178 */
1179 rnh = rt_tables_get_rnh(fibnum, dst->sa_family);
1180 if (rnh == NULL)
1181 return (EAFNOSUPPORT);
1182 needlock = ((flags & RTF_RNH_LOCKED) == 0);
1183 flags &= ~RTF_RNH_LOCKED;
1184 if (needlock)
1185 RADIX_NODE_HEAD_LOCK(rnh);
1186 else
1187 RADIX_NODE_HEAD_LOCK_ASSERT(rnh);
1188 /*
1189 * If we are adding a host route then we don't want to put
1190 * a netmask in the tree, nor do we want to clone it.
1191 */
1192 if (flags & RTF_HOST)
1193 netmask = NULL;
1194
1195 switch (req) {
1196 case RTM_DELETE:
1197 if (netmask) {
1198 rt_maskedcopy(dst, (struct sockaddr *)&mdst, netmask);
1199 dst = (struct sockaddr *)&mdst;
1200 }
1201 #ifdef RADIX_MPATH
1202 if (rn_mpath_capable(rnh)) {
1203 error = rn_mpath_update(req, info, rnh, ret_nrt);
1204 /*
1205 * "bad" holds true for the success case
1206 * as well
1207 */
1208 if (error != ENOENT)
1209 goto bad;
1210 error = 0;
1211 }
1212 #endif
1213 if ((flags & RTF_PINNED) == 0) {
1214 /* Check if target route can be deleted */
1215 rt = (struct rtentry *)rnh->rnh_lookup(dst,
1216 netmask, rnh);
1217 if ((rt != NULL) && (rt->rt_flags & RTF_PINNED))
1218 senderr(EADDRINUSE);
1219 }
1220
1221 /*
1222 * Remove the item from the tree and return it.
1223 * Complain if it is not there and do no more processing.
1224 */
1225 rn = rnh->rnh_deladdr(dst, netmask, rnh);
1226 if (rn == NULL)
1227 senderr(ESRCH);
1228 if (rn->rn_flags & (RNF_ACTIVE | RNF_ROOT))
1229 panic ("rtrequest delete");
1230 rt = RNTORT(rn);
1231 RT_LOCK(rt);
1232 RT_ADDREF(rt);
1233 rt->rt_flags &= ~RTF_UP;
1234
1235 /*
1236 * give the protocol a chance to keep things in sync.
1237 */
1238 if ((ifa = rt->rt_ifa) && ifa->ifa_rtrequest)
1239 ifa->ifa_rtrequest(RTM_DELETE, rt, info);
1240
1241 /*
1242 * One more rtentry floating around that is not
1243 * linked to the routing table. rttrash will be decremented
1244 * when RTFREE(rt) is eventually called.
1245 */
1246 V_rttrash++;
1247
1248 /*
1249 * If the caller wants it, then it can have it,
1250 * but it's up to it to free the rtentry as we won't be
1251 * doing it.
1252 */
1253 if (ret_nrt) {
1254 *ret_nrt = rt;
1255 RT_UNLOCK(rt);
1256 } else
1257 RTFREE_LOCKED(rt);
1258 break;
1259 case RTM_RESOLVE:
1260 /*
1261 * resolve was only used for route cloning
1262 * here for compat
1263 */
1264 break;
1265 case RTM_ADD:
1266 if ((flags & RTF_GATEWAY) && !gateway)
1267 senderr(EINVAL);
1268 if (dst && gateway && (dst->sa_family != gateway->sa_family) &&
1269 (gateway->sa_family != AF_UNSPEC) && (gateway->sa_family != AF_LINK))
1270 senderr(EINVAL);
1271
1272 if (info->rti_ifa == NULL) {
1273 error = rt_getifa_fib(info, fibnum);
1274 if (error)
1275 senderr(error);
1276 } else
1277 ifa_ref(info->rti_ifa);
1278 ifa = info->rti_ifa;
1279 rt = uma_zalloc(V_rtzone, M_NOWAIT);
1280 if (rt == NULL) {
1281 ifa_free(ifa);
1282 senderr(ENOBUFS);
1283 }
1284 rt->rt_flags = RTF_UP | flags;
1285 rt->rt_fibnum = fibnum;
1286 /*
1287 * Add the gateway. Possibly re-malloc-ing the storage for it.
1288 */
1289 RT_LOCK(rt);
1290 if ((error = rt_setgate(rt, dst, gateway)) != 0) {
1291 ifa_free(ifa);
1292 uma_zfree(V_rtzone, rt);
1293 senderr(error);
1294 }
1295
1296 /*
1297 * point to the (possibly newly malloc'd) dest address.
1298 */
1299 ndst = (struct sockaddr *)rt_key(rt);
1300
1301 /*
1302 * make sure it contains the value we want (masked if needed).
1303 */
1304 if (netmask) {
1305 rt_maskedcopy(dst, ndst, netmask);
1306 } else
1307 bcopy(dst, ndst, dst->sa_len);
1308
1309 /*
1310 * We use the ifa reference returned by rt_getifa_fib().
1311 * This moved from below so that rnh->rnh_addaddr() can
1312 * examine the ifa and ifa->ifa_ifp if it so desires.
1313 */
1314 rt->rt_ifa = ifa;
1315 rt->rt_ifp = ifa->ifa_ifp;
1316 rt->rt_weight = 1;
1317
1318 #ifdef RADIX_MPATH
1319 /* do not permit exactly the same dst/mask/gw pair */
1320 if (rn_mpath_capable(rnh) &&
1321 rt_mpath_conflict(rnh, rt, netmask)) {
1322 ifa_free(rt->rt_ifa);
1323 Free(rt_key(rt));
1324 uma_zfree(V_rtzone, rt);
1325 senderr(EEXIST);
1326 }
1327 #endif
1328
1329 #ifdef FLOWTABLE
1330 rt0 = NULL;
1331 /* "flow-table" only supports IPv6 and IPv4 at the moment. */
1332 switch (dst->sa_family) {
1333 #ifdef INET6
1334 case AF_INET6:
1335 #endif
1336 #ifdef INET
1337 case AF_INET:
1338 #endif
1339 #if defined(INET6) || defined(INET)
1340 rn = rnh->rnh_matchaddr(dst, rnh);
1341 if (rn && ((rn->rn_flags & RNF_ROOT) == 0)) {
1342 struct sockaddr *mask;
1343 u_char *m, *n;
1344 int len;
1345
1346 /*
1347 * compare mask to see if the new route is
1348 * more specific than the existing one
1349 */
1350 rt0 = RNTORT(rn);
1351 RT_LOCK(rt0);
1352 RT_ADDREF(rt0);
1353 RT_UNLOCK(rt0);
1354 /*
1355 * A host route is already present, so
1356 * leave the flow-table entries as is.
1357 */
1358 if (rt0->rt_flags & RTF_HOST) {
1359 RTFREE(rt0);
1360 rt0 = NULL;
1361 } else if (!(flags & RTF_HOST) && netmask) {
1362 mask = rt_mask(rt0);
1363 len = mask->sa_len;
1364 m = (u_char *)mask;
1365 n = (u_char *)netmask;
1366 while (len-- > 0) {
1367 if (*n != *m)
1368 break;
1369 n++;
1370 m++;
1371 }
1372 if (len == 0 || (*n < *m)) {
1373 RTFREE(rt0);
1374 rt0 = NULL;
1375 }
1376 }
1377 }
1378 #endif/* INET6 || INET */
1379 }
1380 #endif /* FLOWTABLE */
1381
1382 /* XXX mtu manipulation will be done in rnh_addaddr -- itojun */
1383 rn = rnh->rnh_addaddr(ndst, netmask, rnh, rt->rt_nodes);
1384 /*
1385 * If it still failed to go into the tree,
1386 * then un-make it (this should be a function)
1387 */
1388 if (rn == NULL) {
1389 ifa_free(rt->rt_ifa);
1390 Free(rt_key(rt));
1391 uma_zfree(V_rtzone, rt);
1392 #ifdef FLOWTABLE
1393 if (rt0 != NULL)
1394 RTFREE(rt0);
1395 #endif
1396 senderr(EEXIST);
1397 }
1398 #ifdef FLOWTABLE
1399 else if (rt0 != NULL) {
1400 flowtable_route_flush(dst->sa_family, rt0);
1401 RTFREE(rt0);
1402 }
1403 #endif
1404
1405 /*
1406 * If this protocol has something to add to this then
1407 * allow it to do that as well.
1408 */
1409 if (ifa->ifa_rtrequest)
1410 ifa->ifa_rtrequest(req, rt, info);
1411
1412 /*
1413 * actually return a resultant rtentry and
1414 * give the caller a single reference.
1415 */
1416 if (ret_nrt) {
1417 *ret_nrt = rt;
1418 RT_ADDREF(rt);
1419 }
1420 RT_UNLOCK(rt);
1421 break;
1422 default:
1423 error = EOPNOTSUPP;
1424 }
1425 bad:
1426 if (needlock)
1427 RADIX_NODE_HEAD_UNLOCK(rnh);
1428 return (error);
1429 #undef senderr
1430 }
1431
1432 #undef dst
1433 #undef gateway
1434 #undef netmask
1435 #undef ifaaddr
1436 #undef ifpaddr
1437 #undef flags
1438
1439 int
1440 rt_setgate(struct rtentry *rt, struct sockaddr *dst, struct sockaddr *gate)
1441 {
1442 /* XXX dst may be overwritten, can we move this to below */
1443 int dlen = SA_SIZE(dst), glen = SA_SIZE(gate);
1444 #ifdef INVARIANTS
1445 struct radix_node_head *rnh;
1446
1447 rnh = rt_tables_get_rnh(rt->rt_fibnum, dst->sa_family);
1448 #endif
1449
1450 RT_LOCK_ASSERT(rt);
1451 RADIX_NODE_HEAD_LOCK_ASSERT(rnh);
1452
1453 /*
1454 * Prepare to store the gateway in rt->rt_gateway.
1455 * Both dst and gateway are stored one after the other in the same
1456 * malloc'd chunk. If we have room, we can reuse the old buffer,
1457 * rt_gateway already points to the right place.
1458 * Otherwise, malloc a new block and update the 'dst' address.
1459 */
1460 if (rt->rt_gateway == NULL || glen > SA_SIZE(rt->rt_gateway)) {
1461 caddr_t new;
1462
1463 R_Malloc(new, caddr_t, dlen + glen);
1464 if (new == NULL)
1465 return ENOBUFS;
1466 /*
1467 * XXX note, we copy from *dst and not *rt_key(rt) because
1468 * rt_setgate() can be called to initialize a newly
1469 * allocated route entry, in which case rt_key(rt) == NULL
1470 * (and also rt->rt_gateway == NULL).
1471 * Free()/free() handle a NULL argument just fine.
1472 */
1473 bcopy(dst, new, dlen);
1474 Free(rt_key(rt)); /* free old block, if any */
1475 rt_key(rt) = (struct sockaddr *)new;
1476 rt->rt_gateway = (struct sockaddr *)(new + dlen);
1477 }
1478
1479 /*
1480 * Copy the new gateway value into the memory chunk.
1481 */
1482 bcopy(gate, rt->rt_gateway, glen);
1483
1484 return (0);
1485 }
1486
1487 void
1488 rt_maskedcopy(struct sockaddr *src, struct sockaddr *dst, struct sockaddr *netmask)
1489 {
1490 register u_char *cp1 = (u_char *)src;
1491 register u_char *cp2 = (u_char *)dst;
1492 register u_char *cp3 = (u_char *)netmask;
1493 u_char *cplim = cp2 + *cp3;
1494 u_char *cplim2 = cp2 + *cp1;
1495
1496 *cp2++ = *cp1++; *cp2++ = *cp1++; /* copies sa_len & sa_family */
1497 cp3 += 2;
1498 if (cplim > cplim2)
1499 cplim = cplim2;
1500 while (cp2 < cplim)
1501 *cp2++ = *cp1++ & *cp3++;
1502 if (cp2 < cplim2)
1503 bzero((caddr_t)cp2, (unsigned)(cplim2 - cp2));
1504 }
1505
1506 /*
1507 * Set up a routing table entry, normally
1508 * for an interface.
1509 */
1510 #define _SOCKADDR_TMPSIZE 128 /* Not too big.. kernel stack size is limited */
1511 static inline int
1512 rtinit1(struct ifaddr *ifa, int cmd, int flags, int fibnum)
1513 {
1514 struct sockaddr *dst;
1515 struct sockaddr *netmask;
1516 struct rtentry *rt = NULL;
1517 struct rt_addrinfo info;
1518 int error = 0;
1519 int startfib, endfib;
1520 char tempbuf[_SOCKADDR_TMPSIZE];
1521 int didwork = 0;
1522 int a_failure = 0;
1523 static struct sockaddr_dl null_sdl = {sizeof(null_sdl), AF_LINK};
1524 struct radix_node_head *rnh;
1525
1526 if (flags & RTF_HOST) {
1527 dst = ifa->ifa_dstaddr;
1528 netmask = NULL;
1529 } else {
1530 dst = ifa->ifa_addr;
1531 netmask = ifa->ifa_netmask;
1532 }
1533 if (dst->sa_len == 0)
1534 return(EINVAL);
1535 switch (dst->sa_family) {
1536 case AF_INET6:
1537 case AF_INET:
1538 /* We support multiple FIBs. */
1539 break;
1540 default:
1541 fibnum = RT_DEFAULT_FIB;
1542 break;
1543 }
1544 if (fibnum == RT_ALL_FIBS) {
1545 if (rt_add_addr_allfibs == 0 && cmd == (int)RTM_ADD) {
1546 startfib = endfib = ifa->ifa_ifp->if_fib;
1547 } else {
1548 startfib = 0;
1549 endfib = rt_numfibs - 1;
1550 }
1551 } else {
1552 KASSERT((fibnum < rt_numfibs), ("rtinit1: bad fibnum"));
1553 startfib = fibnum;
1554 endfib = fibnum;
1555 }
1556
1557 /*
1558 * If it's a delete, check that if it exists,
1559 * it's on the correct interface or we might scrub
1560 * a route to another ifa which would
1561 * be confusing at best and possibly worse.
1562 */
1563 if (cmd == RTM_DELETE) {
1564 /*
1565 * It's a delete, so it should already exist..
1566 * If it's a net, mask off the host bits
1567 * (Assuming we have a mask)
1568 * XXX this is kinda inet specific..
1569 */
1570 if (netmask != NULL) {
1571 rt_maskedcopy(dst, (struct sockaddr *)tempbuf, netmask);
1572 dst = (struct sockaddr *)tempbuf;
1573 }
1574 }
1575 /*
1576 * Now go through all the requested tables (fibs) and do the
1577 * requested action. Realistically, this will either be fib 0
1578 * for protocols that don't do multiple tables or all the
1579 * tables for those that do.
1580 */
1581 for ( fibnum = startfib; fibnum <= endfib; fibnum++) {
1582 if (cmd == RTM_DELETE) {
1583 struct radix_node *rn;
1584 /*
1585 * Look up an rtentry that is in the routing tree and
1586 * contains the correct info.
1587 */
1588 rnh = rt_tables_get_rnh(fibnum, dst->sa_family);
1589 if (rnh == NULL)
1590 /* this table doesn't exist but others might */
1591 continue;
1592 RADIX_NODE_HEAD_RLOCK(rnh);
1593 rn = rnh->rnh_lookup(dst, netmask, rnh);
1594 #ifdef RADIX_MPATH
1595 if (rn_mpath_capable(rnh)) {
1596
1597 if (rn == NULL)
1598 error = ESRCH;
1599 else {
1600 rt = RNTORT(rn);
1601 /*
1602 * for interface route the
1603 * rt->rt_gateway is sockaddr_intf
1604 * for cloning ARP entries, so
1605 * rt_mpath_matchgate must use the
1606 * interface address
1607 */
1608 rt = rt_mpath_matchgate(rt,
1609 ifa->ifa_addr);
1610 if (rt == NULL)
1611 error = ESRCH;
1612 }
1613 }
1614 #endif
1615 error = (rn == NULL ||
1616 (rn->rn_flags & RNF_ROOT) ||
1617 RNTORT(rn)->rt_ifa != ifa);
1618 RADIX_NODE_HEAD_RUNLOCK(rnh);
1619 if (error) {
1620 /* this is only an error if bad on ALL tables */
1621 continue;
1622 }
1623 }
1624 /*
1625 * Do the actual request
1626 */
1627 bzero((caddr_t)&info, sizeof(info));
1628 info.rti_ifa = ifa;
1629 info.rti_flags = flags |
1630 (ifa->ifa_flags & ~IFA_RTSELF) | RTF_PINNED;
1631 info.rti_info[RTAX_DST] = dst;
1632 /*
1633 * doing this for compatibility reasons
1634 */
1635 if (cmd == RTM_ADD)
1636 info.rti_info[RTAX_GATEWAY] =
1637 (struct sockaddr *)&null_sdl;
1638 else
1639 info.rti_info[RTAX_GATEWAY] = ifa->ifa_addr;
1640 info.rti_info[RTAX_NETMASK] = netmask;
1641 error = rtrequest1_fib(cmd, &info, &rt, fibnum);
1642
1643 if ((error == EEXIST) && (cmd == RTM_ADD)) {
1644 /*
1645 * Interface route addition failed.
1646 * Atomically delete current prefix generating
1647 * RTM_DELETE message, and retry adding
1648 * interface prefix.
1649 */
1650 rnh = rt_tables_get_rnh(fibnum, dst->sa_family);
1651 RADIX_NODE_HEAD_LOCK(rnh);
1652
1653 /* Delete old prefix */
1654 info.rti_ifa = NULL;
1655 info.rti_flags = RTF_RNH_LOCKED;
1656
1657 error = rtrequest1_fib(RTM_DELETE, &info, NULL, fibnum);
1658 if (error == 0) {
1659 info.rti_ifa = ifa;
1660 info.rti_flags = flags | RTF_RNH_LOCKED |
1661 (ifa->ifa_flags & ~IFA_RTSELF) | RTF_PINNED;
1662 error = rtrequest1_fib(cmd, &info, &rt, fibnum);
1663 }
1664
1665 RADIX_NODE_HEAD_UNLOCK(rnh);
1666 }
1667
1668
1669 if (error == 0 && rt != NULL) {
1670 /*
1671 * notify any listening routing agents of the change
1672 */
1673 RT_LOCK(rt);
1674 #ifdef RADIX_MPATH
1675 /*
1676 * in case address alias finds the first address
1677 * e.g. ifconfig bge0 192.0.2.246/24
1678 * e.g. ifconfig bge0 192.0.2.247/24
1679 * the address set in the route is 192.0.2.246
1680 * so we need to replace it with 192.0.2.247
1681 */
1682 if (memcmp(rt->rt_ifa->ifa_addr,
1683 ifa->ifa_addr, ifa->ifa_addr->sa_len)) {
1684 ifa_free(rt->rt_ifa);
1685 ifa_ref(ifa);
1686 rt->rt_ifp = ifa->ifa_ifp;
1687 rt->rt_ifa = ifa;
1688 }
1689 #endif
1690 /*
1691 * doing this for compatibility reasons
1692 */
1693 if (cmd == RTM_ADD) {
1694 ((struct sockaddr_dl *)rt->rt_gateway)->sdl_type =
1695 rt->rt_ifp->if_type;
1696 ((struct sockaddr_dl *)rt->rt_gateway)->sdl_index =
1697 rt->rt_ifp->if_index;
1698 }
1699 RT_ADDREF(rt);
1700 RT_UNLOCK(rt);
1701 rt_newaddrmsg_fib(cmd, ifa, error, rt, fibnum);
1702 RT_LOCK(rt);
1703 RT_REMREF(rt);
1704 if (cmd == RTM_DELETE) {
1705 /*
1706 * If we are deleting, and we found an entry,
1707 * then it's been removed from the tree..
1708 * now throw it away.
1709 */
1710 RTFREE_LOCKED(rt);
1711 } else {
1712 if (cmd == RTM_ADD) {
1713 /*
1714 * We just wanted to add it..
1715 * we don't actually need a reference.
1716 */
1717 RT_REMREF(rt);
1718 }
1719 RT_UNLOCK(rt);
1720 }
1721 didwork = 1;
1722 }
1723 if (error)
1724 a_failure = error;
1725 }
1726 if (cmd == RTM_DELETE) {
1727 if (didwork) {
1728 error = 0;
1729 } else {
1730 /* we only give an error if it wasn't in any table */
1731 error = ((flags & RTF_HOST) ?
1732 EHOSTUNREACH : ENETUNREACH);
1733 }
1734 } else {
1735 if (a_failure) {
1736 /* return an error if any of them failed */
1737 error = a_failure;
1738 }
1739 }
1740 return (error);
1741 }
1742
1743 /*
1744 * Set up a routing table entry, normally
1745 * for an interface.
1746 */
1747 int
1748 rtinit(struct ifaddr *ifa, int cmd, int flags)
1749 {
1750 struct sockaddr *dst;
1751 int fib = RT_DEFAULT_FIB;
1752
1753 if (flags & RTF_HOST) {
1754 dst = ifa->ifa_dstaddr;
1755 } else {
1756 dst = ifa->ifa_addr;
1757 }
1758
1759 switch (dst->sa_family) {
1760 case AF_INET6:
1761 case AF_INET:
1762 /* We do support multiple FIBs. */
1763 fib = RT_ALL_FIBS;
1764 break;
1765 }
1766 return (rtinit1(ifa, cmd, flags, fib));
1767 }
1768
1769 /*
1770 * Announce interface address arrival/withdraw
1771 * Returns 0 on success.
1772 */
1773 int
1774 rt_addrmsg(int cmd, struct ifaddr *ifa, int fibnum)
1775 {
1776
1777 KASSERT(cmd == RTM_ADD || cmd == RTM_DELETE,
1778 ("unexpected cmd %d", cmd));
1779
1780 KASSERT(fibnum == RT_ALL_FIBS || (fibnum >= 0 && fibnum < rt_numfibs),
1781 ("%s: fib out of range 0 <=%d<%d", __func__, fibnum, rt_numfibs));
1782
1783 #if defined(INET) || defined(INET6)
1784 #ifdef SCTP
1785 /*
1786 * notify the SCTP stack
1787 * this will only get called when an address is added/deleted
1788 * XXX pass the ifaddr struct instead if ifa->ifa_addr...
1789 */
1790 sctp_addr_change(ifa, cmd);
1791 #endif /* SCTP */
1792 #endif
1793 return (rtsock_addrmsg(cmd, ifa, fibnum));
1794 }
1795
1796 /*
1797 * Announce route addition/removal.
1798 * Users of this function MUST validate input data BEFORE calling.
1799 * However we have to be able to handle invalid data:
1800 * if some userland app sends us "invalid" route message (invalid mask,
1801 * no dst, wrong address families, etc...) we need to pass it back
1802 * to app (and any other rtsock consumers) with rtm_errno field set to
1803 * non-zero value.
1804 * Returns 0 on success.
1805 */
1806 int
1807 rt_routemsg(int cmd, struct ifnet *ifp, int error, struct rtentry *rt,
1808 int fibnum)
1809 {
1810
1811 KASSERT(cmd == RTM_ADD || cmd == RTM_DELETE,
1812 ("unexpected cmd %d", cmd));
1813
1814 KASSERT(fibnum == RT_ALL_FIBS || (fibnum >= 0 && fibnum < rt_numfibs),
1815 ("%s: fib out of range 0 <=%d<%d", __func__, fibnum, rt_numfibs));
1816
1817 KASSERT(rt_key(rt) != NULL, (":%s: rt_key must be supplied", __func__));
1818
1819 return (rtsock_routemsg(cmd, ifp, error, rt, fibnum));
1820 }
1821
1822 void
1823 rt_newaddrmsg(int cmd, struct ifaddr *ifa, int error, struct rtentry *rt)
1824 {
1825
1826 rt_newaddrmsg_fib(cmd, ifa, error, rt, RT_ALL_FIBS);
1827 }
1828
1829 /*
1830 * This is called to generate messages from the routing socket
1831 * indicating a network interface has had addresses associated with it.
1832 */
1833 void
1834 rt_newaddrmsg_fib(int cmd, struct ifaddr *ifa, int error, struct rtentry *rt,
1835 int fibnum)
1836 {
1837
1838 KASSERT(cmd == RTM_ADD || cmd == RTM_DELETE,
1839 ("unexpected cmd %u", cmd));
1840 KASSERT(fibnum == RT_ALL_FIBS || (fibnum >= 0 && fibnum < rt_numfibs),
1841 ("%s: fib out of range 0 <=%d<%d", __func__, fibnum, rt_numfibs));
1842
1843 if (cmd == RTM_ADD) {
1844 rt_addrmsg(cmd, ifa, fibnum);
1845 if (rt != NULL)
1846 rt_routemsg(cmd, ifa->ifa_ifp, error, rt, fibnum);
1847 } else {
1848 if (rt != NULL)
1849 rt_routemsg(cmd, ifa->ifa_ifp, error, rt, fibnum);
1850 rt_addrmsg(cmd, ifa, fibnum);
1851 }
1852 }
1853
Cache object: 5fada434a4dbf63ee1eb4530571fafd8
|