2 * Copyright (c) 1980, 1986, 1991, 1993
3 * The Regents of the University of California. All rights reserved.
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
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.
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
29 * @(#)route.c 8.3.1.1 (Berkeley) 2/23/95
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 ***********************************************************************/
38 #include "opt_inet6.h"
39 #include "opt_route.h"
40 #include "opt_mrouting.h"
41 #include "opt_mpath.h"
43 #include <sys/param.h>
44 #include <sys/systm.h>
45 #include <sys/syslog.h>
46 #include <sys/malloc.h>
48 #include <sys/socket.h>
49 #include <sys/sysctl.h>
50 #include <sys/syslog.h>
51 #include <sys/sysproto.h>
53 #include <sys/domain.h>
54 #include <sys/kernel.h>
57 #include <net/if_dl.h>
58 #include <net/route.h>
60 #include <net/flowtable.h>
63 #include <net/radix_mpath.h>
66 #include <netinet/in.h>
67 #include <netinet/ip_mroute.h>
71 /* We use 4 bits in the mbuf flags, thus we are limited to 16 FIBS. */
74 /* Kernel config default option. */
77 #error "ROUTETABLES defined too low"
79 #if ROUTETABLES > RT_MAXFIBS
80 #error "ROUTETABLES defined too big"
82 #define RT_NUMFIBS ROUTETABLES
83 #endif /* ROUTETABLES */
84 /* Initialize to default if not otherwise set. */
89 u_int rt_numfibs = RT_NUMFIBS;
90 SYSCTL_UINT(_net, OID_AUTO, fibs, CTLFLAG_RD, &rt_numfibs, 0, "");
92 * Allow the boot code to allow LESS than RT_MAXFIBS to be used.
93 * We can't do more because storage is statically allocated for now.
94 * (for compatibility reasons.. this will change. When this changes, code should
95 * be refactored to protocol independent parts and protocol dependent parts,
96 * probably hanging of domain(9) specific storage to not need the full
97 * fib * af RNH allocation etc. but allow tuning the number of tables per
100 TUNABLE_INT("net.fibs", &rt_numfibs);
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.
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);
117 VNET_DEFINE(struct rtstat, rtstat);
118 #define V_rtstat VNET(rtstat)
120 VNET_DEFINE(struct radix_node_head *, rt_tables);
121 #define V_rt_tables VNET(rt_tables)
123 VNET_DEFINE(int, rttrash); /* routes not in table but not freed */
124 #define V_rttrash VNET(rttrash)
127 /* compare two sockaddr structures */
128 #define sa_equal(a1, a2) (bcmp((a1), (a2), (a1)->sa_len) == 0)
131 * Convert a 'struct radix_node *' to a 'struct rtentry *'.
132 * The operation can be done safely (in this code) because a
133 * 'struct rtentry' starts with two 'struct radix_node''s, the first
134 * one representing leaf nodes in the routing tree, which is
135 * what the code in radix.c passes us as a 'struct radix_node'.
137 * But because there are a lot of assumptions in this conversion,
138 * do not cast explicitly, but always use the macro below.
140 #define RNTORT(p) ((struct rtentry *)(p))
142 static VNET_DEFINE(uma_zone_t, rtzone); /* Routing table UMA zone. */
143 #define V_rtzone VNET(rtzone)
146 * handler for net.my_fibnum
149 sysctl_my_fibnum(SYSCTL_HANDLER_ARGS)
154 fibnum = curthread->td_proc->p_fibnum;
155 error = sysctl_handle_int(oidp, &fibnum, 0, req);
159 SYSCTL_PROC(_net, OID_AUTO, my_fibnum, CTLTYPE_INT|CTLFLAG_RD,
160 NULL, 0, &sysctl_my_fibnum, "I", "default FIB of caller");
162 static __inline struct radix_node_head **
163 rt_tables_get_rnh_ptr(int table, int fam)
165 struct radix_node_head **rnh;
167 KASSERT(table >= 0 && table < rt_numfibs, ("%s: table out of bounds.",
169 KASSERT(fam >= 0 && fam < (AF_MAX+1), ("%s: fam out of bounds.",
172 /* rnh is [fib=0][af=0]. */
173 rnh = (struct radix_node_head **)V_rt_tables;
174 /* Get the offset to the requested table and fam. */
175 rnh += table * (AF_MAX+1) + fam;
180 struct radix_node_head *
181 rt_tables_get_rnh(int table, int fam)
184 return (*rt_tables_get_rnh_ptr(table, fam));
188 * route initialization must occur before ip6_init2(), which happenas at
197 /* whack the tunable ints into line. */
198 if (rt_numfibs > RT_MAXFIBS)
199 rt_numfibs = RT_MAXFIBS;
203 for (dom = domains; dom; dom = dom->dom_next)
204 if (dom->dom_maxrtkey > max_keylen)
205 max_keylen = dom->dom_maxrtkey;
207 rn_init(max_keylen); /* init all zeroes, all ones, mask table */
209 SYSINIT(route_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_THIRD, route_init, 0);
212 vnet_route_init(const void *unused __unused)
215 struct radix_node_head **rnh;
219 V_rt_tables = malloc(rt_numfibs * (AF_MAX+1) *
220 sizeof(struct radix_node_head *), M_RTABLE, M_WAITOK|M_ZERO);
222 V_rtzone = uma_zcreate("rtentry", sizeof(struct rtentry), NULL, NULL,
223 NULL, NULL, UMA_ALIGN_PTR, 0);
224 for (dom = domains; dom; dom = dom->dom_next) {
225 if (dom->dom_rtattach == NULL)
228 for (table = 0; table < rt_numfibs; table++) {
229 fam = dom->dom_family;
230 if (table != 0 && fam != AF_INET6 && fam != AF_INET)
234 * XXX MRT rtattach will be also called from
235 * vfs_export.c but the offset will be 0 (only for
236 * AF_INET and AF_INET6 which don't need it anyhow).
238 rnh = rt_tables_get_rnh_ptr(table, fam);
240 panic("%s: rnh NULL", __func__);
241 dom->dom_rtattach((void **)rnh, dom->dom_rtoffset);
245 VNET_SYSINIT(vnet_route_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_FOURTH,
250 vnet_route_uninit(const void *unused __unused)
255 struct radix_node_head **rnh;
257 for (dom = domains; dom; dom = dom->dom_next) {
258 if (dom->dom_rtdetach == NULL)
261 for (table = 0; table < rt_numfibs; table++) {
262 fam = dom->dom_family;
264 if (table != 0 && fam != AF_INET6 && fam != AF_INET)
267 rnh = rt_tables_get_rnh_ptr(table, fam);
269 panic("%s: rnh NULL", __func__);
270 dom->dom_rtdetach((void **)rnh, dom->dom_rtoffset);
274 VNET_SYSUNINIT(vnet_route_uninit, SI_SUB_PROTO_DOMAIN, SI_ORDER_THIRD,
275 vnet_route_uninit, 0);
278 #ifndef _SYS_SYSPROTO_H_
284 sys_setfib(struct thread *td, struct setfib_args *uap)
286 if (uap->fibnum < 0 || uap->fibnum >= rt_numfibs)
288 td->td_proc->p_fibnum = uap->fibnum;
293 * Packet routing routines.
296 rtalloc(struct route *ro)
299 rtalloc_ign_fib(ro, 0UL, RT_DEFAULT_FIB);
303 rtalloc_fib(struct route *ro, u_int fibnum)
305 rtalloc_ign_fib(ro, 0UL, fibnum);
309 rtalloc_ign(struct route *ro, u_long ignore)
313 if ((rt = ro->ro_rt) != NULL) {
314 if (rt->rt_ifp != NULL && rt->rt_flags & RTF_UP)
319 ro->ro_rt = rtalloc1_fib(&ro->ro_dst, 1, ignore, RT_DEFAULT_FIB);
321 RT_UNLOCK(ro->ro_rt);
325 rtalloc_ign_fib(struct route *ro, u_long ignore, u_int fibnum)
329 if ((rt = ro->ro_rt) != NULL) {
330 if (rt->rt_ifp != NULL && rt->rt_flags & RTF_UP)
335 ro->ro_rt = rtalloc1_fib(&ro->ro_dst, 1, ignore, fibnum);
337 RT_UNLOCK(ro->ro_rt);
341 * Look up the route that matches the address given
342 * Or, at least try.. Create a cloned route if needed.
344 * The returned route, if any, is locked.
347 rtalloc1(struct sockaddr *dst, int report, u_long ignflags)
350 return (rtalloc1_fib(dst, report, ignflags, RT_DEFAULT_FIB));
354 rtalloc1_fib(struct sockaddr *dst, int report, u_long ignflags,
357 struct radix_node_head *rnh;
358 struct radix_node *rn;
359 struct rtentry *newrt;
360 struct rt_addrinfo info;
361 int err = 0, msgtype = RTM_MISS;
364 KASSERT((fibnum < rt_numfibs), ("rtalloc1_fib: bad fibnum"));
365 switch (dst->sa_family) {
368 /* We support multiple FIBs. */
371 fibnum = RT_DEFAULT_FIB;
374 rnh = rt_tables_get_rnh(fibnum, dst->sa_family);
380 * Look up the address in the table for that Address Family
382 needlock = !(ignflags & RTF_RNH_LOCKED);
384 RADIX_NODE_HEAD_RLOCK(rnh);
387 RADIX_NODE_HEAD_LOCK_ASSERT(rnh);
389 rn = rnh->rnh_matchaddr(dst, rnh);
390 if (rn && ((rn->rn_flags & RNF_ROOT) == 0)) {
395 RADIX_NODE_HEAD_RUNLOCK(rnh);
399 RADIX_NODE_HEAD_RUNLOCK(rnh);
402 * Either we hit the root or couldn't find any match,
403 * Which basically means
404 * "caint get there frm here"
407 V_rtstat.rts_unreach++;
411 * If required, report the failure to the supervising
413 * For a delete, this is not an error. (report == 0)
415 bzero(&info, sizeof(info));
416 info.rti_info[RTAX_DST] = dst;
417 rt_missmsg_fib(msgtype, &info, 0, err, fibnum);
421 RT_LOCK_ASSERT(newrt);
426 * Remove a reference count from an rtentry.
427 * If the count gets low enough, take it out of the routing table
430 rtfree(struct rtentry *rt)
432 struct radix_node_head *rnh;
434 KASSERT(rt != NULL,("%s: NULL rt", __func__));
435 rnh = rt_tables_get_rnh(rt->rt_fibnum, rt_key(rt)->sa_family);
436 KASSERT(rnh != NULL,("%s: NULL rnh", __func__));
441 * The callers should use RTFREE_LOCKED() or RTFREE(), so
442 * we should come here exactly with the last reference.
445 if (rt->rt_refcnt > 0) {
446 log(LOG_DEBUG, "%s: %p has %d refs\n", __func__, rt, rt->rt_refcnt);
451 * On last reference give the "close method" a chance
452 * to cleanup private state. This also permits (for
453 * IPv4 and IPv6) a chance to decide if the routing table
454 * entry should be purged immediately or at a later time.
455 * When an immediate purge is to happen the close routine
456 * typically calls rtexpunge which clears the RTF_UP flag
457 * on the entry so that the code below reclaims the storage.
459 if (rt->rt_refcnt == 0 && rnh->rnh_close)
460 rnh->rnh_close((struct radix_node *)rt, rnh);
463 * If we are no longer "up" (and ref == 0)
464 * then we can free the resources associated
467 if ((rt->rt_flags & RTF_UP) == 0) {
468 if (rt->rt_nodes->rn_flags & (RNF_ACTIVE | RNF_ROOT))
471 * the rtentry must have been removed from the routing table
472 * so it is represented in rttrash.. remove that now.
476 if (rt->rt_refcnt < 0) {
477 printf("rtfree: %p not freed (neg refs)\n", rt);
482 * release references on items we hold them on..
483 * e.g other routes and ifaddrs.
486 ifa_free(rt->rt_ifa);
488 * The key is separatly alloc'd so free it (see rt_setgate()).
489 * This also frees the gateway, as they are always malloc'd
495 * and the rtentry itself of course
498 uma_zfree(V_rtzone, rt);
507 * Force a routing table entry to the specified
508 * destination to go through the given gateway.
509 * Normally called as a result of a routing redirect
510 * message from the network layer.
513 rtredirect(struct sockaddr *dst,
514 struct sockaddr *gateway,
515 struct sockaddr *netmask,
517 struct sockaddr *src)
520 rtredirect_fib(dst, gateway, netmask, flags, src, RT_DEFAULT_FIB);
524 rtredirect_fib(struct sockaddr *dst,
525 struct sockaddr *gateway,
526 struct sockaddr *netmask,
528 struct sockaddr *src,
531 struct rtentry *rt, *rt0 = NULL;
534 struct rt_addrinfo info;
536 struct radix_node_head *rnh;
539 rnh = rt_tables_get_rnh(fibnum, dst->sa_family);
541 error = EAFNOSUPPORT;
545 /* verify the gateway is directly reachable */
546 if ((ifa = ifa_ifwithnet(gateway, 0)) == NULL) {
550 rt = rtalloc1_fib(dst, 0, 0UL, fibnum); /* NB: rt is locked */
552 * If the redirect isn't from our current router for this dst,
553 * it's either old or wrong. If it redirects us to ourselves,
554 * we have a routing loop, perhaps as a result of an interface
555 * going down recently.
557 if (!(flags & RTF_DONE) && rt &&
558 (!sa_equal(src, rt->rt_gateway) || rt->rt_ifa != ifa))
560 else if (ifa_ifwithaddr_check(gateway))
561 error = EHOSTUNREACH;
565 * Create a new entry if we just got back a wildcard entry
566 * or the lookup failed. This is necessary for hosts
567 * which use routing redirects generated by smart gateways
568 * to dynamically build the routing tables.
570 if (rt == NULL || (rt_mask(rt) && rt_mask(rt)->sa_len < 2))
573 * Don't listen to the redirect if it's
574 * for a route to an interface.
576 if (rt->rt_flags & RTF_GATEWAY) {
577 if (((rt->rt_flags & RTF_HOST) == 0) && (flags & RTF_HOST)) {
579 * Changing from route to net => route to host.
580 * Create new route, rather than smashing route to net.
586 flags |= RTF_GATEWAY | RTF_DYNAMIC;
587 bzero((caddr_t)&info, sizeof(info));
588 info.rti_info[RTAX_DST] = dst;
589 info.rti_info[RTAX_GATEWAY] = gateway;
590 info.rti_info[RTAX_NETMASK] = netmask;
592 info.rti_flags = flags;
594 RT_UNLOCK(rt0); /* drop lock to avoid LOR with RNH */
595 error = rtrequest1_fib(RTM_ADD, &info, &rt, fibnum);
599 EVENTHANDLER_INVOKE(route_redirect_event, rt0, rt, dst);
600 flags = rt->rt_flags;
605 stat = &V_rtstat.rts_dynamic;
607 struct rtentry *gwrt;
610 * Smash the current notion of the gateway to
611 * this destination. Should check about netmask!!!
613 rt->rt_flags |= RTF_MODIFIED;
614 flags |= RTF_MODIFIED;
615 stat = &V_rtstat.rts_newgateway;
617 * add the key and gateway (in one malloc'd chunk).
620 RADIX_NODE_HEAD_LOCK(rnh);
622 rt_setgate(rt, rt_key(rt), gateway);
623 gwrt = rtalloc1(gateway, 1, RTF_RNH_LOCKED);
624 RADIX_NODE_HEAD_UNLOCK(rnh);
625 EVENTHANDLER_INVOKE(route_redirect_event, rt, gwrt, dst);
629 error = EHOSTUNREACH;
635 V_rtstat.rts_badredirect++;
636 else if (stat != NULL)
638 bzero((caddr_t)&info, sizeof(info));
639 info.rti_info[RTAX_DST] = dst;
640 info.rti_info[RTAX_GATEWAY] = gateway;
641 info.rti_info[RTAX_NETMASK] = netmask;
642 info.rti_info[RTAX_AUTHOR] = src;
643 rt_missmsg_fib(RTM_REDIRECT, &info, flags, error, fibnum);
649 rtioctl(u_long req, caddr_t data)
652 return (rtioctl_fib(req, data, RT_DEFAULT_FIB));
656 * Routing table ioctl interface.
659 rtioctl_fib(u_long req, caddr_t data, u_int fibnum)
663 * If more ioctl commands are added here, make sure the proper
664 * super-user checks are being performed because it is possible for
665 * prison-root to make it this far if raw sockets have been enabled
669 /* Multicast goop, grrr... */
670 return mrt_ioctl ? mrt_ioctl(req, data, fibnum) : EOPNOTSUPP;
677 * For both ifa_ifwithroute() routines, 'ifa' is returned referenced.
680 ifa_ifwithroute(int flags, struct sockaddr *dst, struct sockaddr *gateway)
683 return (ifa_ifwithroute_fib(flags, dst, gateway, RT_DEFAULT_FIB));
687 ifa_ifwithroute_fib(int flags, struct sockaddr *dst, struct sockaddr *gateway,
690 register struct ifaddr *ifa;
693 if ((flags & RTF_GATEWAY) == 0) {
695 * If we are adding a route to an interface,
696 * and the interface is a pt to pt link
697 * we should search for the destination
698 * as our clue to the interface. Otherwise
699 * we can use the local address.
702 if (flags & RTF_HOST)
703 ifa = ifa_ifwithdstaddr(dst);
705 ifa = ifa_ifwithaddr(gateway);
708 * If we are adding a route to a remote net
709 * or host, the gateway may still be on the
710 * other end of a pt to pt link.
712 ifa = ifa_ifwithdstaddr(gateway);
715 ifa = ifa_ifwithnet(gateway, 0);
717 struct rtentry *rt = rtalloc1_fib(gateway, 0, RTF_RNH_LOCKED, fibnum);
721 * dismiss a gateway that is reachable only
722 * through the default router
724 switch (gateway->sa_family) {
726 if (satosin(rt_key(rt))->sin_addr.s_addr == INADDR_ANY)
730 if (IN6_IS_ADDR_UNSPECIFIED(&satosin6(rt_key(rt))->sin6_addr))
736 if (!not_found && rt->rt_ifa != NULL) {
742 if (not_found || ifa == NULL)
745 if (ifa->ifa_addr->sa_family != dst->sa_family) {
746 struct ifaddr *oifa = ifa;
747 ifa = ifaof_ifpforaddr(dst, ifa->ifa_ifp);
757 * Do appropriate manipulations of a routing tree given
758 * all the bits of info needed
762 struct sockaddr *dst,
763 struct sockaddr *gateway,
764 struct sockaddr *netmask,
766 struct rtentry **ret_nrt)
769 return (rtrequest_fib(req, dst, gateway, netmask, flags, ret_nrt,
774 rtrequest_fib(int req,
775 struct sockaddr *dst,
776 struct sockaddr *gateway,
777 struct sockaddr *netmask,
779 struct rtentry **ret_nrt,
782 struct rt_addrinfo info;
784 if (dst->sa_len == 0)
787 bzero((caddr_t)&info, sizeof(info));
788 info.rti_flags = flags;
789 info.rti_info[RTAX_DST] = dst;
790 info.rti_info[RTAX_GATEWAY] = gateway;
791 info.rti_info[RTAX_NETMASK] = netmask;
792 return rtrequest1_fib(req, &info, ret_nrt, fibnum);
796 * These (questionable) definitions of apparent local variables apply
797 * to the next two functions. XXXXXX!!!
799 #define dst info->rti_info[RTAX_DST]
800 #define gateway info->rti_info[RTAX_GATEWAY]
801 #define netmask info->rti_info[RTAX_NETMASK]
802 #define ifaaddr info->rti_info[RTAX_IFA]
803 #define ifpaddr info->rti_info[RTAX_IFP]
804 #define flags info->rti_flags
807 rt_getifa(struct rt_addrinfo *info)
810 return (rt_getifa_fib(info, RT_DEFAULT_FIB));
814 * Look up rt_addrinfo for a specific fib. Note that if rti_ifa is defined,
815 * it will be referenced so the caller must free it.
818 rt_getifa_fib(struct rt_addrinfo *info, u_int fibnum)
824 * ifp may be specified by sockaddr_dl
825 * when protocol address is ambiguous.
827 if (info->rti_ifp == NULL && ifpaddr != NULL &&
828 ifpaddr->sa_family == AF_LINK &&
829 (ifa = ifa_ifwithnet(ifpaddr, 0)) != NULL) {
830 info->rti_ifp = ifa->ifa_ifp;
833 if (info->rti_ifa == NULL && ifaaddr != NULL)
834 info->rti_ifa = ifa_ifwithaddr(ifaaddr);
835 if (info->rti_ifa == NULL) {
838 sa = ifaaddr != NULL ? ifaaddr :
839 (gateway != NULL ? gateway : dst);
840 if (sa != NULL && info->rti_ifp != NULL)
841 info->rti_ifa = ifaof_ifpforaddr(sa, info->rti_ifp);
842 else if (dst != NULL && gateway != NULL)
843 info->rti_ifa = ifa_ifwithroute_fib(flags, dst, gateway,
846 info->rti_ifa = ifa_ifwithroute_fib(flags, sa, sa,
849 if ((ifa = info->rti_ifa) != NULL) {
850 if (info->rti_ifp == NULL)
851 info->rti_ifp = ifa->ifa_ifp;
858 * Expunges references to a route that's about to be reclaimed.
859 * The route must be locked.
862 rtexpunge(struct rtentry *rt)
864 #if !defined(RADIX_MPATH)
865 struct radix_node *rn;
867 struct rt_addrinfo info;
871 struct radix_node_head *rnh;
876 * Find the correct routing tree to use for this Address Family
878 rnh = rt_tables_get_rnh(rt->rt_fibnum, rt_key(rt)->sa_family);
881 return (EAFNOSUPPORT);
882 RADIX_NODE_HEAD_LOCK_ASSERT(rnh);
886 bzero(&info, sizeof(info));
887 info.rti_ifp = rt->rt_ifp;
888 info.rti_flags = RTF_RNH_LOCKED;
889 info.rti_info[RTAX_DST] = rt_key(rt);
890 info.rti_info[RTAX_GATEWAY] = rt->rt_ifa->ifa_addr;
893 error = rtrequest1_fib(RTM_DELETE, &info, &rt0, fib);
895 if (error == 0 && rt0 != NULL) {
898 } else if (error != 0) {
904 * Remove the item from the tree; it should be there,
905 * but when callers invoke us blindly it may not (sigh).
907 rn = rnh->rnh_deladdr(rt_key(rt), rt_mask(rt), rnh);
912 KASSERT((rn->rn_flags & (RNF_ACTIVE | RNF_ROOT)) == 0,
913 ("unexpected flags 0x%x", rn->rn_flags));
914 KASSERT(rt == RNTORT(rn),
915 ("lookup mismatch, rt %p rn %p", rt, rn));
916 #endif /* RADIX_MPATH */
918 rt->rt_flags &= ~RTF_UP;
921 * Give the protocol a chance to keep things in sync.
923 if ((ifa = rt->rt_ifa) && ifa->ifa_rtrequest) {
924 struct rt_addrinfo info;
926 bzero((caddr_t)&info, sizeof(info));
927 info.rti_flags = rt->rt_flags;
928 info.rti_info[RTAX_DST] = rt_key(rt);
929 info.rti_info[RTAX_GATEWAY] = rt->rt_gateway;
930 info.rti_info[RTAX_NETMASK] = rt_mask(rt);
931 ifa->ifa_rtrequest(RTM_DELETE, rt, &info);
935 * one more rtentry floating around that is not
936 * linked to the routing table.
939 #if !defined(RADIX_MPATH)
947 rn_mpath_update(int req, struct rt_addrinfo *info,
948 struct radix_node_head *rnh, struct rtentry **ret_nrt)
951 * if we got multipath routes, we require users to specify
952 * a matching RTAX_GATEWAY.
954 struct rtentry *rt, *rto = NULL;
955 register struct radix_node *rn;
958 rn = rnh->rnh_matchaddr(dst, rnh);
961 rto = rt = RNTORT(rn);
962 rt = rt_mpath_matchgate(rt, gateway);
966 * this is the first entry in the chain
969 rn = rn_mpath_next((struct radix_node *)rt);
971 * there is another entry, now it's active
976 rto->rt_flags |= RTF_UP;
978 } else if (rt->rt_flags & RTF_GATEWAY) {
980 * For gateway routes, we need to
981 * make sure that we we are deleting
982 * the correct gateway.
983 * rt_mpath_matchgate() does not
984 * check the case when there is only
985 * one route in the chain.
988 (rt->rt_gateway->sa_len != gateway->sa_len ||
989 memcmp(rt->rt_gateway, gateway, gateway->sa_len)))
993 * remove from tree before returning it
996 rn = rnh->rnh_deladdr(dst, netmask, rnh);
997 KASSERT(rt == RNTORT(rn), ("radix node disappeared"));
1003 * use the normal delete code to remove
1006 if (req != RTM_DELETE)
1014 * if the entry is 2nd and on up
1016 if ((req == RTM_DELETE) && !rt_mpath_deldup(rto, rt))
1017 panic ("rtrequest1: rt_mpath_deldup");
1021 if (req == RTM_DELETE) {
1022 rt->rt_flags &= ~RTF_UP;
1024 * One more rtentry floating around that is not
1025 * linked to the routing table. rttrash will be decremented
1026 * when RTFREE(rt) is eventually called.
1032 if (req != RTM_DELETE)
1033 panic("unrecognized request %d", req);
1037 * If the caller wants it, then it can have it,
1038 * but it's up to it to free the rtentry as we won't be
1052 rtrequest1_fib(int req, struct rt_addrinfo *info, struct rtentry **ret_nrt,
1055 int error = 0, needlock = 0;
1056 register struct rtentry *rt;
1058 register struct rtentry *rt0;
1060 register struct radix_node *rn;
1061 register struct radix_node_head *rnh;
1063 struct sockaddr *ndst;
1064 struct sockaddr_storage mdst;
1065 #define senderr(x) { error = x ; goto bad; }
1067 KASSERT((fibnum < rt_numfibs), ("rtrequest1_fib: bad fibnum"));
1068 switch (dst->sa_family) {
1071 /* We support multiple FIBs. */
1074 fibnum = RT_DEFAULT_FIB;
1079 * Find the correct routing tree to use for this Address Family
1081 rnh = rt_tables_get_rnh(fibnum, dst->sa_family);
1083 return (EAFNOSUPPORT);
1084 needlock = ((flags & RTF_RNH_LOCKED) == 0);
1085 flags &= ~RTF_RNH_LOCKED;
1087 RADIX_NODE_HEAD_LOCK(rnh);
1089 RADIX_NODE_HEAD_LOCK_ASSERT(rnh);
1091 * If we are adding a host route then we don't want to put
1092 * a netmask in the tree, nor do we want to clone it.
1094 if (flags & RTF_HOST)
1100 rt_maskedcopy(dst, (struct sockaddr *)&mdst, netmask);
1101 dst = (struct sockaddr *)&mdst;
1104 if (rn_mpath_capable(rnh)) {
1105 error = rn_mpath_update(req, info, rnh, ret_nrt);
1107 * "bad" holds true for the success case
1110 if (error != ENOENT)
1116 * Remove the item from the tree and return it.
1117 * Complain if it is not there and do no more processing.
1119 rn = rnh->rnh_deladdr(dst, netmask, rnh);
1122 if (rn->rn_flags & (RNF_ACTIVE | RNF_ROOT))
1123 panic ("rtrequest delete");
1127 rt->rt_flags &= ~RTF_UP;
1130 * give the protocol a chance to keep things in sync.
1132 if ((ifa = rt->rt_ifa) && ifa->ifa_rtrequest)
1133 ifa->ifa_rtrequest(RTM_DELETE, rt, info);
1136 * One more rtentry floating around that is not
1137 * linked to the routing table. rttrash will be decremented
1138 * when RTFREE(rt) is eventually called.
1143 * If the caller wants it, then it can have it,
1144 * but it's up to it to free the rtentry as we won't be
1155 * resolve was only used for route cloning
1160 if ((flags & RTF_GATEWAY) && !gateway)
1162 if (dst && gateway && (dst->sa_family != gateway->sa_family) &&
1163 (gateway->sa_family != AF_UNSPEC) && (gateway->sa_family != AF_LINK))
1166 if (info->rti_ifa == NULL) {
1167 error = rt_getifa_fib(info, fibnum);
1171 ifa_ref(info->rti_ifa);
1172 ifa = info->rti_ifa;
1173 rt = uma_zalloc(V_rtzone, M_NOWAIT | M_ZERO);
1180 rt->rt_flags = RTF_UP | flags;
1181 rt->rt_fibnum = fibnum;
1183 * Add the gateway. Possibly re-malloc-ing the storage for it.
1186 if ((error = rt_setgate(rt, dst, gateway)) != 0) {
1187 RT_LOCK_DESTROY(rt);
1190 uma_zfree(V_rtzone, rt);
1195 * point to the (possibly newly malloc'd) dest address.
1197 ndst = (struct sockaddr *)rt_key(rt);
1200 * make sure it contains the value we want (masked if needed).
1203 rt_maskedcopy(dst, ndst, netmask);
1205 bcopy(dst, ndst, dst->sa_len);
1208 * We use the ifa reference returned by rt_getifa_fib().
1209 * This moved from below so that rnh->rnh_addaddr() can
1210 * examine the ifa and ifa->ifa_ifp if it so desires.
1213 rt->rt_ifp = ifa->ifa_ifp;
1214 rt->rt_rmx.rmx_weight = 1;
1217 /* do not permit exactly the same dst/mask/gw pair */
1218 if (rn_mpath_capable(rnh) &&
1219 rt_mpath_conflict(rnh, rt, netmask)) {
1221 ifa_free(rt->rt_ifa);
1224 RT_LOCK_DESTROY(rt);
1225 uma_zfree(V_rtzone, rt);
1232 /* "flow-table" only supports IPv6 and IPv4 at the moment. */
1233 switch (dst->sa_family) {
1240 #if defined(INET6) || defined(INET)
1241 rn = rnh->rnh_matchaddr(dst, rnh);
1242 if (rn && ((rn->rn_flags & RNF_ROOT) == 0)) {
1243 struct sockaddr *mask;
1248 * compare mask to see if the new route is
1249 * more specific than the existing one
1256 * A host route is already present, so
1257 * leave the flow-table entries as is.
1259 if (rt0->rt_flags & RTF_HOST) {
1262 } else if (!(flags & RTF_HOST) && netmask) {
1263 mask = rt_mask(rt0);
1266 n = (u_char *)netmask;
1273 if (len == 0 || (*n < *m)) {
1279 #endif/* INET6 || INET */
1281 #endif /* FLOWTABLE */
1283 /* XXX mtu manipulation will be done in rnh_addaddr -- itojun */
1284 rn = rnh->rnh_addaddr(ndst, netmask, rnh, rt->rt_nodes);
1286 * If it still failed to go into the tree,
1287 * then un-make it (this should be a function)
1291 ifa_free(rt->rt_ifa);
1293 RT_LOCK_DESTROY(rt);
1294 uma_zfree(V_rtzone, rt);
1302 else if (rt0 != NULL) {
1303 switch (dst->sa_family) {
1306 flowtable_route_flush(V_ip6_ft, rt0);
1311 flowtable_route_flush(V_ip_ft, rt0);
1320 * If this protocol has something to add to this then
1321 * allow it to do that as well.
1323 if (ifa->ifa_rtrequest)
1324 ifa->ifa_rtrequest(req, rt, info);
1327 * actually return a resultant rtentry and
1328 * give the caller a single reference.
1341 RADIX_NODE_HEAD_UNLOCK(rnh);
1354 rt_setgate(struct rtentry *rt, struct sockaddr *dst, struct sockaddr *gate)
1356 /* XXX dst may be overwritten, can we move this to below */
1357 int dlen = SA_SIZE(dst), glen = SA_SIZE(gate);
1359 struct radix_node_head *rnh;
1361 rnh = rt_tables_get_rnh(rt->rt_fibnum, dst->sa_family);
1365 RADIX_NODE_HEAD_LOCK_ASSERT(rnh);
1368 * Prepare to store the gateway in rt->rt_gateway.
1369 * Both dst and gateway are stored one after the other in the same
1370 * malloc'd chunk. If we have room, we can reuse the old buffer,
1371 * rt_gateway already points to the right place.
1372 * Otherwise, malloc a new block and update the 'dst' address.
1374 if (rt->rt_gateway == NULL || glen > SA_SIZE(rt->rt_gateway)) {
1377 R_Malloc(new, caddr_t, dlen + glen);
1381 * XXX note, we copy from *dst and not *rt_key(rt) because
1382 * rt_setgate() can be called to initialize a newly
1383 * allocated route entry, in which case rt_key(rt) == NULL
1384 * (and also rt->rt_gateway == NULL).
1385 * Free()/free() handle a NULL argument just fine.
1387 bcopy(dst, new, dlen);
1388 Free(rt_key(rt)); /* free old block, if any */
1389 rt_key(rt) = (struct sockaddr *)new;
1390 rt->rt_gateway = (struct sockaddr *)(new + dlen);
1394 * Copy the new gateway value into the memory chunk.
1396 bcopy(gate, rt->rt_gateway, glen);
1402 rt_maskedcopy(struct sockaddr *src, struct sockaddr *dst, struct sockaddr *netmask)
1404 register u_char *cp1 = (u_char *)src;
1405 register u_char *cp2 = (u_char *)dst;
1406 register u_char *cp3 = (u_char *)netmask;
1407 u_char *cplim = cp2 + *cp3;
1408 u_char *cplim2 = cp2 + *cp1;
1410 *cp2++ = *cp1++; *cp2++ = *cp1++; /* copies sa_len & sa_family */
1415 *cp2++ = *cp1++ & *cp3++;
1417 bzero((caddr_t)cp2, (unsigned)(cplim2 - cp2));
1421 * Set up a routing table entry, normally
1424 #define _SOCKADDR_TMPSIZE 128 /* Not too big.. kernel stack size is limited */
1426 rtinit1(struct ifaddr *ifa, int cmd, int flags, int fibnum)
1428 struct sockaddr *dst;
1429 struct sockaddr *netmask;
1430 struct rtentry *rt = NULL;
1431 struct rt_addrinfo info;
1433 int startfib, endfib;
1434 char tempbuf[_SOCKADDR_TMPSIZE];
1437 static struct sockaddr_dl null_sdl = {sizeof(null_sdl), AF_LINK};
1439 if (flags & RTF_HOST) {
1440 dst = ifa->ifa_dstaddr;
1443 dst = ifa->ifa_addr;
1444 netmask = ifa->ifa_netmask;
1446 if (dst->sa_len == 0)
1448 switch (dst->sa_family) {
1451 /* We support multiple FIBs. */
1454 fibnum = RT_DEFAULT_FIB;
1458 if (rt_add_addr_allfibs == 0 && cmd == (int)RTM_ADD) {
1459 startfib = endfib = curthread->td_proc->p_fibnum;
1462 endfib = rt_numfibs - 1;
1465 KASSERT((fibnum < rt_numfibs), ("rtinit1: bad fibnum"));
1471 * If it's a delete, check that if it exists,
1472 * it's on the correct interface or we might scrub
1473 * a route to another ifa which would
1474 * be confusing at best and possibly worse.
1476 if (cmd == RTM_DELETE) {
1478 * It's a delete, so it should already exist..
1479 * If it's a net, mask off the host bits
1480 * (Assuming we have a mask)
1481 * XXX this is kinda inet specific..
1483 if (netmask != NULL) {
1484 rt_maskedcopy(dst, (struct sockaddr *)tempbuf, netmask);
1485 dst = (struct sockaddr *)tempbuf;
1489 * Now go through all the requested tables (fibs) and do the
1490 * requested action. Realistically, this will either be fib 0
1491 * for protocols that don't do multiple tables or all the
1492 * tables for those that do.
1494 for ( fibnum = startfib; fibnum <= endfib; fibnum++) {
1495 if (cmd == RTM_DELETE) {
1496 struct radix_node_head *rnh;
1497 struct radix_node *rn;
1499 * Look up an rtentry that is in the routing tree and
1500 * contains the correct info.
1502 rnh = rt_tables_get_rnh(fibnum, dst->sa_family);
1504 /* this table doesn't exist but others might */
1506 RADIX_NODE_HEAD_LOCK(rnh);
1508 if (rn_mpath_capable(rnh)) {
1510 rn = rnh->rnh_matchaddr(dst, rnh);
1516 * for interface route the
1517 * rt->rt_gateway is sockaddr_intf
1518 * for cloning ARP entries, so
1519 * rt_mpath_matchgate must use the
1522 rt = rt_mpath_matchgate(rt,
1530 rn = rnh->rnh_lookup(dst, netmask, rnh);
1531 error = (rn == NULL ||
1532 (rn->rn_flags & RNF_ROOT) ||
1533 RNTORT(rn)->rt_ifa != ifa ||
1534 !sa_equal((struct sockaddr *)rn->rn_key, dst));
1535 RADIX_NODE_HEAD_UNLOCK(rnh);
1537 /* this is only an error if bad on ALL tables */
1542 * Do the actual request
1544 bzero((caddr_t)&info, sizeof(info));
1546 info.rti_flags = flags | (ifa->ifa_flags & ~IFA_RTSELF);
1547 info.rti_info[RTAX_DST] = dst;
1549 * doing this for compatibility reasons
1552 info.rti_info[RTAX_GATEWAY] =
1553 (struct sockaddr *)&null_sdl;
1555 info.rti_info[RTAX_GATEWAY] = ifa->ifa_addr;
1556 info.rti_info[RTAX_NETMASK] = netmask;
1557 error = rtrequest1_fib(cmd, &info, &rt, fibnum);
1558 if (error == 0 && rt != NULL) {
1560 * notify any listening routing agents of the change
1565 * in case address alias finds the first address
1566 * e.g. ifconfig bge0 192.0.2.246/24
1567 * e.g. ifconfig bge0 192.0.2.247/24
1568 * the address set in the route is 192.0.2.246
1569 * so we need to replace it with 192.0.2.247
1571 if (memcmp(rt->rt_ifa->ifa_addr,
1572 ifa->ifa_addr, ifa->ifa_addr->sa_len)) {
1573 ifa_free(rt->rt_ifa);
1575 rt->rt_ifp = ifa->ifa_ifp;
1580 * doing this for compatibility reasons
1582 if (cmd == RTM_ADD) {
1583 ((struct sockaddr_dl *)rt->rt_gateway)->sdl_type =
1584 rt->rt_ifp->if_type;
1585 ((struct sockaddr_dl *)rt->rt_gateway)->sdl_index =
1586 rt->rt_ifp->if_index;
1590 rt_newaddrmsg_fib(cmd, ifa, error, rt, fibnum);
1593 if (cmd == RTM_DELETE) {
1595 * If we are deleting, and we found an entry,
1596 * then it's been removed from the tree..
1597 * now throw it away.
1601 if (cmd == RTM_ADD) {
1603 * We just wanted to add it..
1604 * we don't actually need a reference.
1615 if (cmd == RTM_DELETE) {
1619 /* we only give an error if it wasn't in any table */
1620 error = ((flags & RTF_HOST) ?
1621 EHOSTUNREACH : ENETUNREACH);
1625 /* return an error if any of them failed */
1632 #ifndef BURN_BRIDGES
1633 /* special one for inet internal use. may not use. */
1635 rtinit_fib(struct ifaddr *ifa, int cmd, int flags)
1637 return (rtinit1(ifa, cmd, flags, -1));
1642 * Set up a routing table entry, normally
1646 rtinit(struct ifaddr *ifa, int cmd, int flags)
1648 struct sockaddr *dst;
1649 int fib = RT_DEFAULT_FIB;
1651 if (flags & RTF_HOST) {
1652 dst = ifa->ifa_dstaddr;
1654 dst = ifa->ifa_addr;
1657 switch (dst->sa_family) {
1660 /* We do support multiple FIBs. */
1664 return (rtinit1(ifa, cmd, flags, fib));