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_var.h>
58 #include <net/if_dl.h>
59 #include <net/route.h>
61 #include <net/flowtable.h>
64 #include <net/radix_mpath.h>
67 #include <netinet/in.h>
68 #include <netinet/ip_mroute.h>
72 #define RT_MAXFIBS UINT16_MAX
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 /* This is read-only.. */
90 u_int rt_numfibs = RT_NUMFIBS;
91 SYSCTL_UINT(_net, OID_AUTO, fibs, CTLFLAG_RD, &rt_numfibs, 0, "");
92 /* and this can be set too big but will be fixed before it is used */
93 TUNABLE_INT("net.fibs", &rt_numfibs);
96 * By default add routes to all fibs for new interfaces.
97 * Once this is set to 0 then only allocate routes on interface
98 * changes for the FIB of the caller when adding a new set of addresses
99 * to an interface. XXX this is a shotgun aproach to a problem that needs
100 * a more fine grained solution.. that will come.
101 * XXX also has the problems getting the FIB from curthread which will not
102 * always work given the fib can be overridden and prefixes can be added
103 * from the network stack context.
105 u_int rt_add_addr_allfibs = 1;
106 SYSCTL_UINT(_net, OID_AUTO, add_addr_allfibs, CTLFLAG_RW,
107 &rt_add_addr_allfibs, 0, "");
108 TUNABLE_INT("net.add_addr_allfibs", &rt_add_addr_allfibs);
110 VNET_DEFINE(struct rtstat, rtstat);
111 #define V_rtstat VNET(rtstat)
113 VNET_DEFINE(struct radix_node_head *, rt_tables);
114 #define V_rt_tables VNET(rt_tables)
116 VNET_DEFINE(int, rttrash); /* routes not in table but not freed */
117 #define V_rttrash VNET(rttrash)
120 /* compare two sockaddr structures */
121 #define sa_equal(a1, a2) (bcmp((a1), (a2), (a1)->sa_len) == 0)
124 * Convert a 'struct radix_node *' to a 'struct rtentry *'.
125 * The operation can be done safely (in this code) because a
126 * 'struct rtentry' starts with two 'struct radix_node''s, the first
127 * one representing leaf nodes in the routing tree, which is
128 * what the code in radix.c passes us as a 'struct radix_node'.
130 * But because there are a lot of assumptions in this conversion,
131 * do not cast explicitly, but always use the macro below.
133 #define RNTORT(p) ((struct rtentry *)(p))
135 static VNET_DEFINE(uma_zone_t, rtzone); /* Routing table UMA zone. */
136 #define V_rtzone VNET(rtzone)
139 * handler for net.my_fibnum
142 sysctl_my_fibnum(SYSCTL_HANDLER_ARGS)
147 fibnum = curthread->td_proc->p_fibnum;
148 error = sysctl_handle_int(oidp, &fibnum, 0, req);
152 SYSCTL_PROC(_net, OID_AUTO, my_fibnum, CTLTYPE_INT|CTLFLAG_RD,
153 NULL, 0, &sysctl_my_fibnum, "I", "default FIB of caller");
155 static __inline struct radix_node_head **
156 rt_tables_get_rnh_ptr(int table, int fam)
158 struct radix_node_head **rnh;
160 KASSERT(table >= 0 && table < rt_numfibs, ("%s: table out of bounds.",
162 KASSERT(fam >= 0 && fam < (AF_MAX+1), ("%s: fam out of bounds.",
165 /* rnh is [fib=0][af=0]. */
166 rnh = (struct radix_node_head **)V_rt_tables;
167 /* Get the offset to the requested table and fam. */
168 rnh += table * (AF_MAX+1) + fam;
173 struct radix_node_head *
174 rt_tables_get_rnh(int table, int fam)
177 return (*rt_tables_get_rnh_ptr(table, fam));
181 * route initialization must occur before ip6_init2(), which happenas at
188 /* whack the tunable ints into line. */
189 if (rt_numfibs > RT_MAXFIBS)
190 rt_numfibs = RT_MAXFIBS;
194 SYSINIT(route_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_THIRD, route_init, 0);
197 vnet_route_init(const void *unused __unused)
200 struct radix_node_head **rnh;
204 V_rt_tables = malloc(rt_numfibs * (AF_MAX+1) *
205 sizeof(struct radix_node_head *), M_RTABLE, M_WAITOK|M_ZERO);
207 V_rtzone = uma_zcreate("rtentry", sizeof(struct rtentry), NULL, NULL,
208 NULL, NULL, UMA_ALIGN_PTR, 0);
209 for (dom = domains; dom; dom = dom->dom_next) {
210 if (dom->dom_rtattach == NULL)
213 for (table = 0; table < rt_numfibs; table++) {
214 fam = dom->dom_family;
215 if (table != 0 && fam != AF_INET6 && fam != AF_INET)
219 * XXX MRT rtattach will be also called from
220 * vfs_export.c but the offset will be 0 (only for
221 * AF_INET and AF_INET6 which don't need it anyhow).
223 rnh = rt_tables_get_rnh_ptr(table, fam);
225 panic("%s: rnh NULL", __func__);
226 dom->dom_rtattach((void **)rnh, dom->dom_rtoffset);
230 VNET_SYSINIT(vnet_route_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_FOURTH,
235 vnet_route_uninit(const void *unused __unused)
240 struct radix_node_head **rnh;
242 for (dom = domains; dom; dom = dom->dom_next) {
243 if (dom->dom_rtdetach == NULL)
246 for (table = 0; table < rt_numfibs; table++) {
247 fam = dom->dom_family;
249 if (table != 0 && fam != AF_INET6 && fam != AF_INET)
252 rnh = rt_tables_get_rnh_ptr(table, fam);
254 panic("%s: rnh NULL", __func__);
255 dom->dom_rtdetach((void **)rnh, dom->dom_rtoffset);
259 free(V_rt_tables, M_RTABLE);
260 uma_zdestroy(V_rtzone);
262 VNET_SYSUNINIT(vnet_route_uninit, SI_SUB_PROTO_DOMAIN, SI_ORDER_THIRD,
263 vnet_route_uninit, 0);
266 #ifndef _SYS_SYSPROTO_H_
272 sys_setfib(struct thread *td, struct setfib_args *uap)
274 if (uap->fibnum < 0 || uap->fibnum >= rt_numfibs)
276 td->td_proc->p_fibnum = uap->fibnum;
281 * Packet routing routines.
284 rtalloc(struct route *ro)
287 rtalloc_ign_fib(ro, 0UL, RT_DEFAULT_FIB);
291 rtalloc_fib(struct route *ro, u_int fibnum)
293 rtalloc_ign_fib(ro, 0UL, fibnum);
297 rtalloc_ign(struct route *ro, u_long ignore)
301 if ((rt = ro->ro_rt) != NULL) {
302 if (rt->rt_ifp != NULL && rt->rt_flags & RTF_UP)
307 ro->ro_rt = rtalloc1_fib(&ro->ro_dst, 1, ignore, RT_DEFAULT_FIB);
309 RT_UNLOCK(ro->ro_rt);
313 rtalloc_ign_fib(struct route *ro, u_long ignore, u_int fibnum)
317 if ((rt = ro->ro_rt) != NULL) {
318 if (rt->rt_ifp != NULL && rt->rt_flags & RTF_UP)
323 ro->ro_rt = rtalloc1_fib(&ro->ro_dst, 1, ignore, fibnum);
325 RT_UNLOCK(ro->ro_rt);
329 * Look up the route that matches the address given
330 * Or, at least try.. Create a cloned route if needed.
332 * The returned route, if any, is locked.
335 rtalloc1(struct sockaddr *dst, int report, u_long ignflags)
338 return (rtalloc1_fib(dst, report, ignflags, RT_DEFAULT_FIB));
342 rtalloc1_fib(struct sockaddr *dst, int report, u_long ignflags,
345 struct radix_node_head *rnh;
346 struct radix_node *rn;
347 struct rtentry *newrt;
348 struct rt_addrinfo info;
349 int err = 0, msgtype = RTM_MISS;
352 KASSERT((fibnum < rt_numfibs), ("rtalloc1_fib: bad fibnum"));
353 switch (dst->sa_family) {
356 /* We support multiple FIBs. */
359 fibnum = RT_DEFAULT_FIB;
362 rnh = rt_tables_get_rnh(fibnum, dst->sa_family);
368 * Look up the address in the table for that Address Family
370 needlock = !(ignflags & RTF_RNH_LOCKED);
372 RADIX_NODE_HEAD_RLOCK(rnh);
375 RADIX_NODE_HEAD_LOCK_ASSERT(rnh);
377 rn = rnh->rnh_matchaddr(dst, rnh);
378 if (rn && ((rn->rn_flags & RNF_ROOT) == 0)) {
383 RADIX_NODE_HEAD_RUNLOCK(rnh);
387 RADIX_NODE_HEAD_RUNLOCK(rnh);
390 * Either we hit the root or couldn't find any match,
391 * Which basically means
392 * "caint get there frm here"
395 V_rtstat.rts_unreach++;
399 * If required, report the failure to the supervising
401 * For a delete, this is not an error. (report == 0)
403 bzero(&info, sizeof(info));
404 info.rti_info[RTAX_DST] = dst;
405 rt_missmsg_fib(msgtype, &info, 0, err, fibnum);
409 RT_LOCK_ASSERT(newrt);
414 * Remove a reference count from an rtentry.
415 * If the count gets low enough, take it out of the routing table
418 rtfree(struct rtentry *rt)
420 struct radix_node_head *rnh;
422 KASSERT(rt != NULL,("%s: NULL rt", __func__));
423 rnh = rt_tables_get_rnh(rt->rt_fibnum, rt_key(rt)->sa_family);
424 KASSERT(rnh != NULL,("%s: NULL rnh", __func__));
429 * The callers should use RTFREE_LOCKED() or RTFREE(), so
430 * we should come here exactly with the last reference.
433 if (rt->rt_refcnt > 0) {
434 log(LOG_DEBUG, "%s: %p has %d refs\n", __func__, rt, rt->rt_refcnt);
439 * On last reference give the "close method" a chance
440 * to cleanup private state. This also permits (for
441 * IPv4 and IPv6) a chance to decide if the routing table
442 * entry should be purged immediately or at a later time.
443 * When an immediate purge is to happen the close routine
444 * typically calls rtexpunge which clears the RTF_UP flag
445 * on the entry so that the code below reclaims the storage.
447 if (rt->rt_refcnt == 0 && rnh->rnh_close)
448 rnh->rnh_close((struct radix_node *)rt, rnh);
451 * If we are no longer "up" (and ref == 0)
452 * then we can free the resources associated
455 if ((rt->rt_flags & RTF_UP) == 0) {
456 if (rt->rt_nodes->rn_flags & (RNF_ACTIVE | RNF_ROOT))
459 * the rtentry must have been removed from the routing table
460 * so it is represented in rttrash.. remove that now.
464 if (rt->rt_refcnt < 0) {
465 printf("rtfree: %p not freed (neg refs)\n", rt);
470 * release references on items we hold them on..
471 * e.g other routes and ifaddrs.
474 ifa_free(rt->rt_ifa);
476 * The key is separatly alloc'd so free it (see rt_setgate()).
477 * This also frees the gateway, as they are always malloc'd
483 * and the rtentry itself of course
486 uma_zfree(V_rtzone, rt);
495 * Force a routing table entry to the specified
496 * destination to go through the given gateway.
497 * Normally called as a result of a routing redirect
498 * message from the network layer.
501 rtredirect(struct sockaddr *dst,
502 struct sockaddr *gateway,
503 struct sockaddr *netmask,
505 struct sockaddr *src)
508 rtredirect_fib(dst, gateway, netmask, flags, src, RT_DEFAULT_FIB);
512 rtredirect_fib(struct sockaddr *dst,
513 struct sockaddr *gateway,
514 struct sockaddr *netmask,
516 struct sockaddr *src,
519 struct rtentry *rt, *rt0 = NULL;
522 struct rt_addrinfo info;
524 struct radix_node_head *rnh;
527 rnh = rt_tables_get_rnh(fibnum, dst->sa_family);
529 error = EAFNOSUPPORT;
533 /* verify the gateway is directly reachable */
534 if ((ifa = ifa_ifwithnet(gateway, 0)) == NULL) {
538 rt = rtalloc1_fib(dst, 0, 0UL, fibnum); /* NB: rt is locked */
540 * If the redirect isn't from our current router for this dst,
541 * it's either old or wrong. If it redirects us to ourselves,
542 * we have a routing loop, perhaps as a result of an interface
543 * going down recently.
545 if (!(flags & RTF_DONE) && rt &&
546 (!sa_equal(src, rt->rt_gateway) || rt->rt_ifa != ifa))
548 else if (ifa_ifwithaddr_check(gateway))
549 error = EHOSTUNREACH;
553 * Create a new entry if we just got back a wildcard entry
554 * or the lookup failed. This is necessary for hosts
555 * which use routing redirects generated by smart gateways
556 * to dynamically build the routing tables.
558 if (rt == NULL || (rt_mask(rt) && rt_mask(rt)->sa_len < 2))
561 * Don't listen to the redirect if it's
562 * for a route to an interface.
564 if (rt->rt_flags & RTF_GATEWAY) {
565 if (((rt->rt_flags & RTF_HOST) == 0) && (flags & RTF_HOST)) {
567 * Changing from route to net => route to host.
568 * Create new route, rather than smashing route to net.
574 flags |= RTF_GATEWAY | RTF_DYNAMIC;
575 bzero((caddr_t)&info, sizeof(info));
576 info.rti_info[RTAX_DST] = dst;
577 info.rti_info[RTAX_GATEWAY] = gateway;
578 info.rti_info[RTAX_NETMASK] = netmask;
580 info.rti_flags = flags;
582 RT_UNLOCK(rt0); /* drop lock to avoid LOR with RNH */
583 error = rtrequest1_fib(RTM_ADD, &info, &rt, fibnum);
587 EVENTHANDLER_INVOKE(route_redirect_event, rt0, rt, dst);
588 flags = rt->rt_flags;
593 stat = &V_rtstat.rts_dynamic;
595 struct rtentry *gwrt;
598 * Smash the current notion of the gateway to
599 * this destination. Should check about netmask!!!
601 rt->rt_flags |= RTF_MODIFIED;
602 flags |= RTF_MODIFIED;
603 stat = &V_rtstat.rts_newgateway;
605 * add the key and gateway (in one malloc'd chunk).
608 RADIX_NODE_HEAD_LOCK(rnh);
610 rt_setgate(rt, rt_key(rt), gateway);
611 gwrt = rtalloc1(gateway, 1, RTF_RNH_LOCKED);
612 RADIX_NODE_HEAD_UNLOCK(rnh);
613 EVENTHANDLER_INVOKE(route_redirect_event, rt, gwrt, dst);
617 error = EHOSTUNREACH;
623 V_rtstat.rts_badredirect++;
624 else if (stat != NULL)
626 bzero((caddr_t)&info, sizeof(info));
627 info.rti_info[RTAX_DST] = dst;
628 info.rti_info[RTAX_GATEWAY] = gateway;
629 info.rti_info[RTAX_NETMASK] = netmask;
630 info.rti_info[RTAX_AUTHOR] = src;
631 rt_missmsg_fib(RTM_REDIRECT, &info, flags, error, fibnum);
637 rtioctl(u_long req, caddr_t data)
640 return (rtioctl_fib(req, data, RT_DEFAULT_FIB));
644 * Routing table ioctl interface.
647 rtioctl_fib(u_long req, caddr_t data, u_int fibnum)
651 * If more ioctl commands are added here, make sure the proper
652 * super-user checks are being performed because it is possible for
653 * prison-root to make it this far if raw sockets have been enabled
657 /* Multicast goop, grrr... */
658 return mrt_ioctl ? mrt_ioctl(req, data, fibnum) : EOPNOTSUPP;
665 * For both ifa_ifwithroute() routines, 'ifa' is returned referenced.
668 ifa_ifwithroute(int flags, struct sockaddr *dst, struct sockaddr *gateway)
671 return (ifa_ifwithroute_fib(flags, dst, gateway, RT_DEFAULT_FIB));
675 ifa_ifwithroute_fib(int flags, struct sockaddr *dst, struct sockaddr *gateway,
678 register struct ifaddr *ifa;
681 if ((flags & RTF_GATEWAY) == 0) {
683 * If we are adding a route to an interface,
684 * and the interface is a pt to pt link
685 * we should search for the destination
686 * as our clue to the interface. Otherwise
687 * we can use the local address.
690 if (flags & RTF_HOST)
691 ifa = ifa_ifwithdstaddr(dst);
693 ifa = ifa_ifwithaddr(gateway);
696 * If we are adding a route to a remote net
697 * or host, the gateway may still be on the
698 * other end of a pt to pt link.
700 ifa = ifa_ifwithdstaddr(gateway);
703 ifa = ifa_ifwithnet(gateway, 0);
705 struct rtentry *rt = rtalloc1_fib(gateway, 0, RTF_RNH_LOCKED, fibnum);
709 * dismiss a gateway that is reachable only
710 * through the default router
712 switch (gateway->sa_family) {
714 if (satosin(rt_key(rt))->sin_addr.s_addr == INADDR_ANY)
718 if (IN6_IS_ADDR_UNSPECIFIED(&satosin6(rt_key(rt))->sin6_addr))
724 if (!not_found && rt->rt_ifa != NULL) {
730 if (not_found || ifa == NULL)
733 if (ifa->ifa_addr->sa_family != dst->sa_family) {
734 struct ifaddr *oifa = ifa;
735 ifa = ifaof_ifpforaddr(dst, ifa->ifa_ifp);
745 * Do appropriate manipulations of a routing tree given
746 * all the bits of info needed
750 struct sockaddr *dst,
751 struct sockaddr *gateway,
752 struct sockaddr *netmask,
754 struct rtentry **ret_nrt)
757 return (rtrequest_fib(req, dst, gateway, netmask, flags, ret_nrt,
762 rtrequest_fib(int req,
763 struct sockaddr *dst,
764 struct sockaddr *gateway,
765 struct sockaddr *netmask,
767 struct rtentry **ret_nrt,
770 struct rt_addrinfo info;
772 if (dst->sa_len == 0)
775 bzero((caddr_t)&info, sizeof(info));
776 info.rti_flags = flags;
777 info.rti_info[RTAX_DST] = dst;
778 info.rti_info[RTAX_GATEWAY] = gateway;
779 info.rti_info[RTAX_NETMASK] = netmask;
780 return rtrequest1_fib(req, &info, ret_nrt, fibnum);
784 * These (questionable) definitions of apparent local variables apply
785 * to the next two functions. XXXXXX!!!
787 #define dst info->rti_info[RTAX_DST]
788 #define gateway info->rti_info[RTAX_GATEWAY]
789 #define netmask info->rti_info[RTAX_NETMASK]
790 #define ifaaddr info->rti_info[RTAX_IFA]
791 #define ifpaddr info->rti_info[RTAX_IFP]
792 #define flags info->rti_flags
795 rt_getifa(struct rt_addrinfo *info)
798 return (rt_getifa_fib(info, RT_DEFAULT_FIB));
802 * Look up rt_addrinfo for a specific fib. Note that if rti_ifa is defined,
803 * it will be referenced so the caller must free it.
806 rt_getifa_fib(struct rt_addrinfo *info, u_int fibnum)
812 * ifp may be specified by sockaddr_dl
813 * when protocol address is ambiguous.
815 if (info->rti_ifp == NULL && ifpaddr != NULL &&
816 ifpaddr->sa_family == AF_LINK &&
817 (ifa = ifa_ifwithnet(ifpaddr, 0)) != NULL) {
818 info->rti_ifp = ifa->ifa_ifp;
821 if (info->rti_ifa == NULL && ifaaddr != NULL)
822 info->rti_ifa = ifa_ifwithaddr(ifaaddr);
823 if (info->rti_ifa == NULL) {
826 sa = ifaaddr != NULL ? ifaaddr :
827 (gateway != NULL ? gateway : dst);
828 if (sa != NULL && info->rti_ifp != NULL)
829 info->rti_ifa = ifaof_ifpforaddr(sa, info->rti_ifp);
830 else if (dst != NULL && gateway != NULL)
831 info->rti_ifa = ifa_ifwithroute_fib(flags, dst, gateway,
834 info->rti_ifa = ifa_ifwithroute_fib(flags, sa, sa,
837 if ((ifa = info->rti_ifa) != NULL) {
838 if (info->rti_ifp == NULL)
839 info->rti_ifp = ifa->ifa_ifp;
846 * Expunges references to a route that's about to be reclaimed.
847 * The route must be locked.
850 rtexpunge(struct rtentry *rt)
852 #if !defined(RADIX_MPATH)
853 struct radix_node *rn;
855 struct rt_addrinfo info;
859 struct radix_node_head *rnh;
864 * Find the correct routing tree to use for this Address Family
866 rnh = rt_tables_get_rnh(rt->rt_fibnum, rt_key(rt)->sa_family);
869 return (EAFNOSUPPORT);
870 RADIX_NODE_HEAD_LOCK_ASSERT(rnh);
874 bzero(&info, sizeof(info));
875 info.rti_ifp = rt->rt_ifp;
876 info.rti_flags = RTF_RNH_LOCKED;
877 info.rti_info[RTAX_DST] = rt_key(rt);
878 info.rti_info[RTAX_GATEWAY] = rt->rt_ifa->ifa_addr;
881 error = rtrequest1_fib(RTM_DELETE, &info, &rt0, fib);
883 if (error == 0 && rt0 != NULL) {
886 } else if (error != 0) {
892 * Remove the item from the tree; it should be there,
893 * but when callers invoke us blindly it may not (sigh).
895 rn = rnh->rnh_deladdr(rt_key(rt), rt_mask(rt), rnh);
900 KASSERT((rn->rn_flags & (RNF_ACTIVE | RNF_ROOT)) == 0,
901 ("unexpected flags 0x%x", rn->rn_flags));
902 KASSERT(rt == RNTORT(rn),
903 ("lookup mismatch, rt %p rn %p", rt, rn));
904 #endif /* RADIX_MPATH */
906 rt->rt_flags &= ~RTF_UP;
909 * Give the protocol a chance to keep things in sync.
911 if ((ifa = rt->rt_ifa) && ifa->ifa_rtrequest) {
912 struct rt_addrinfo info;
914 bzero((caddr_t)&info, sizeof(info));
915 info.rti_flags = rt->rt_flags;
916 info.rti_info[RTAX_DST] = rt_key(rt);
917 info.rti_info[RTAX_GATEWAY] = rt->rt_gateway;
918 info.rti_info[RTAX_NETMASK] = rt_mask(rt);
919 ifa->ifa_rtrequest(RTM_DELETE, rt, &info);
923 * one more rtentry floating around that is not
924 * linked to the routing table.
927 #if !defined(RADIX_MPATH)
935 rn_mpath_update(int req, struct rt_addrinfo *info,
936 struct radix_node_head *rnh, struct rtentry **ret_nrt)
939 * if we got multipath routes, we require users to specify
940 * a matching RTAX_GATEWAY.
942 struct rtentry *rt, *rto = NULL;
943 register struct radix_node *rn;
946 rn = rnh->rnh_matchaddr(dst, rnh);
949 rto = rt = RNTORT(rn);
950 rt = rt_mpath_matchgate(rt, gateway);
954 * this is the first entry in the chain
957 rn = rn_mpath_next((struct radix_node *)rt);
959 * there is another entry, now it's active
964 rto->rt_flags |= RTF_UP;
966 } else if (rt->rt_flags & RTF_GATEWAY) {
968 * For gateway routes, we need to
969 * make sure that we we are deleting
970 * the correct gateway.
971 * rt_mpath_matchgate() does not
972 * check the case when there is only
973 * one route in the chain.
976 (rt->rt_gateway->sa_len != gateway->sa_len ||
977 memcmp(rt->rt_gateway, gateway, gateway->sa_len)))
981 * remove from tree before returning it
984 rn = rnh->rnh_deladdr(dst, netmask, rnh);
985 KASSERT(rt == RNTORT(rn), ("radix node disappeared"));
991 * use the normal delete code to remove
994 if (req != RTM_DELETE)
1002 * if the entry is 2nd and on up
1004 if ((req == RTM_DELETE) && !rt_mpath_deldup(rto, rt))
1005 panic ("rtrequest1: rt_mpath_deldup");
1009 if (req == RTM_DELETE) {
1010 rt->rt_flags &= ~RTF_UP;
1012 * One more rtentry floating around that is not
1013 * linked to the routing table. rttrash will be decremented
1014 * when RTFREE(rt) is eventually called.
1020 if (req != RTM_DELETE)
1021 panic("unrecognized request %d", req);
1025 * If the caller wants it, then it can have it,
1026 * but it's up to it to free the rtentry as we won't be
1040 rtrequest1_fib(int req, struct rt_addrinfo *info, struct rtentry **ret_nrt,
1043 int error = 0, needlock = 0;
1044 register struct rtentry *rt;
1046 register struct rtentry *rt0;
1048 register struct radix_node *rn;
1049 register struct radix_node_head *rnh;
1051 struct sockaddr *ndst;
1052 struct sockaddr_storage mdst;
1053 #define senderr(x) { error = x ; goto bad; }
1055 KASSERT((fibnum < rt_numfibs), ("rtrequest1_fib: bad fibnum"));
1056 switch (dst->sa_family) {
1059 /* We support multiple FIBs. */
1062 fibnum = RT_DEFAULT_FIB;
1067 * Find the correct routing tree to use for this Address Family
1069 rnh = rt_tables_get_rnh(fibnum, dst->sa_family);
1071 return (EAFNOSUPPORT);
1072 needlock = ((flags & RTF_RNH_LOCKED) == 0);
1073 flags &= ~RTF_RNH_LOCKED;
1075 RADIX_NODE_HEAD_LOCK(rnh);
1077 RADIX_NODE_HEAD_LOCK_ASSERT(rnh);
1079 * If we are adding a host route then we don't want to put
1080 * a netmask in the tree, nor do we want to clone it.
1082 if (flags & RTF_HOST)
1088 rt_maskedcopy(dst, (struct sockaddr *)&mdst, netmask);
1089 dst = (struct sockaddr *)&mdst;
1092 if (rn_mpath_capable(rnh)) {
1093 error = rn_mpath_update(req, info, rnh, ret_nrt);
1095 * "bad" holds true for the success case
1098 if (error != ENOENT)
1103 if ((flags & RTF_PINNED) == 0) {
1104 /* Check if target route can be deleted */
1105 rt = (struct rtentry *)rnh->rnh_lookup(dst,
1107 if ((rt != NULL) && (rt->rt_flags & RTF_PINNED))
1108 senderr(EADDRINUSE);
1112 * Remove the item from the tree and return it.
1113 * Complain if it is not there and do no more processing.
1115 rn = rnh->rnh_deladdr(dst, netmask, rnh);
1118 if (rn->rn_flags & (RNF_ACTIVE | RNF_ROOT))
1119 panic ("rtrequest delete");
1123 rt->rt_flags &= ~RTF_UP;
1126 * give the protocol a chance to keep things in sync.
1128 if ((ifa = rt->rt_ifa) && ifa->ifa_rtrequest)
1129 ifa->ifa_rtrequest(RTM_DELETE, rt, info);
1132 * One more rtentry floating around that is not
1133 * linked to the routing table. rttrash will be decremented
1134 * when RTFREE(rt) is eventually called.
1139 * If the caller wants it, then it can have it,
1140 * but it's up to it to free the rtentry as we won't be
1151 * resolve was only used for route cloning
1156 if ((flags & RTF_GATEWAY) && !gateway)
1158 if (dst && gateway && (dst->sa_family != gateway->sa_family) &&
1159 (gateway->sa_family != AF_UNSPEC) && (gateway->sa_family != AF_LINK))
1162 if (info->rti_ifa == NULL) {
1163 error = rt_getifa_fib(info, fibnum);
1167 ifa_ref(info->rti_ifa);
1168 ifa = info->rti_ifa;
1169 rt = uma_zalloc(V_rtzone, M_NOWAIT | M_ZERO);
1175 rt->rt_flags = RTF_UP | flags;
1176 rt->rt_fibnum = fibnum;
1178 * Add the gateway. Possibly re-malloc-ing the storage for it.
1181 if ((error = rt_setgate(rt, dst, gateway)) != 0) {
1182 RT_LOCK_DESTROY(rt);
1184 uma_zfree(V_rtzone, rt);
1189 * point to the (possibly newly malloc'd) dest address.
1191 ndst = (struct sockaddr *)rt_key(rt);
1194 * make sure it contains the value we want (masked if needed).
1197 rt_maskedcopy(dst, ndst, netmask);
1199 bcopy(dst, ndst, dst->sa_len);
1202 * We use the ifa reference returned by rt_getifa_fib().
1203 * This moved from below so that rnh->rnh_addaddr() can
1204 * examine the ifa and ifa->ifa_ifp if it so desires.
1207 rt->rt_ifp = ifa->ifa_ifp;
1208 rt->rt_rmx.rmx_weight = 1;
1211 /* do not permit exactly the same dst/mask/gw pair */
1212 if (rn_mpath_capable(rnh) &&
1213 rt_mpath_conflict(rnh, rt, netmask)) {
1214 ifa_free(rt->rt_ifa);
1216 RT_LOCK_DESTROY(rt);
1217 uma_zfree(V_rtzone, rt);
1224 /* "flow-table" only supports IPv6 and IPv4 at the moment. */
1225 switch (dst->sa_family) {
1232 #if defined(INET6) || defined(INET)
1233 rn = rnh->rnh_matchaddr(dst, rnh);
1234 if (rn && ((rn->rn_flags & RNF_ROOT) == 0)) {
1235 struct sockaddr *mask;
1240 * compare mask to see if the new route is
1241 * more specific than the existing one
1248 * A host route is already present, so
1249 * leave the flow-table entries as is.
1251 if (rt0->rt_flags & RTF_HOST) {
1254 } else if (!(flags & RTF_HOST) && netmask) {
1255 mask = rt_mask(rt0);
1258 n = (u_char *)netmask;
1265 if (len == 0 || (*n < *m)) {
1271 #endif/* INET6 || INET */
1273 #endif /* FLOWTABLE */
1275 /* XXX mtu manipulation will be done in rnh_addaddr -- itojun */
1276 rn = rnh->rnh_addaddr(ndst, netmask, rnh, rt->rt_nodes);
1278 * If it still failed to go into the tree,
1279 * then un-make it (this should be a function)
1282 ifa_free(rt->rt_ifa);
1284 RT_LOCK_DESTROY(rt);
1285 uma_zfree(V_rtzone, rt);
1293 else if (rt0 != NULL) {
1294 switch (dst->sa_family) {
1297 flowtable_route_flush(V_ip6_ft, rt0);
1302 flowtable_route_flush(V_ip_ft, rt0);
1311 * If this protocol has something to add to this then
1312 * allow it to do that as well.
1314 if (ifa->ifa_rtrequest)
1315 ifa->ifa_rtrequest(req, rt, info);
1318 * actually return a resultant rtentry and
1319 * give the caller a single reference.
1332 RADIX_NODE_HEAD_UNLOCK(rnh);
1345 rt_setgate(struct rtentry *rt, struct sockaddr *dst, struct sockaddr *gate)
1347 /* XXX dst may be overwritten, can we move this to below */
1348 int dlen = SA_SIZE(dst), glen = SA_SIZE(gate);
1350 struct radix_node_head *rnh;
1352 rnh = rt_tables_get_rnh(rt->rt_fibnum, dst->sa_family);
1356 RADIX_NODE_HEAD_LOCK_ASSERT(rnh);
1359 * Prepare to store the gateway in rt->rt_gateway.
1360 * Both dst and gateway are stored one after the other in the same
1361 * malloc'd chunk. If we have room, we can reuse the old buffer,
1362 * rt_gateway already points to the right place.
1363 * Otherwise, malloc a new block and update the 'dst' address.
1365 if (rt->rt_gateway == NULL || glen > SA_SIZE(rt->rt_gateway)) {
1368 R_Malloc(new, caddr_t, dlen + glen);
1372 * XXX note, we copy from *dst and not *rt_key(rt) because
1373 * rt_setgate() can be called to initialize a newly
1374 * allocated route entry, in which case rt_key(rt) == NULL
1375 * (and also rt->rt_gateway == NULL).
1376 * Free()/free() handle a NULL argument just fine.
1378 bcopy(dst, new, dlen);
1379 Free(rt_key(rt)); /* free old block, if any */
1380 rt_key(rt) = (struct sockaddr *)new;
1381 rt->rt_gateway = (struct sockaddr *)(new + dlen);
1385 * Copy the new gateway value into the memory chunk.
1387 bcopy(gate, rt->rt_gateway, glen);
1393 rt_maskedcopy(struct sockaddr *src, struct sockaddr *dst, struct sockaddr *netmask)
1395 register u_char *cp1 = (u_char *)src;
1396 register u_char *cp2 = (u_char *)dst;
1397 register u_char *cp3 = (u_char *)netmask;
1398 u_char *cplim = cp2 + *cp3;
1399 u_char *cplim2 = cp2 + *cp1;
1401 *cp2++ = *cp1++; *cp2++ = *cp1++; /* copies sa_len & sa_family */
1406 *cp2++ = *cp1++ & *cp3++;
1408 bzero((caddr_t)cp2, (unsigned)(cplim2 - cp2));
1412 * Set up a routing table entry, normally
1415 #define _SOCKADDR_TMPSIZE 128 /* Not too big.. kernel stack size is limited */
1417 rtinit1(struct ifaddr *ifa, int cmd, int flags, int fibnum)
1419 struct sockaddr *dst;
1420 struct sockaddr *netmask;
1421 struct rtentry *rt = NULL;
1422 struct rt_addrinfo info;
1424 int startfib, endfib;
1425 char tempbuf[_SOCKADDR_TMPSIZE];
1428 static struct sockaddr_dl null_sdl = {sizeof(null_sdl), AF_LINK};
1429 struct radix_node_head *rnh;
1431 if (flags & RTF_HOST) {
1432 dst = ifa->ifa_dstaddr;
1435 dst = ifa->ifa_addr;
1436 netmask = ifa->ifa_netmask;
1438 if (dst->sa_len == 0)
1440 switch (dst->sa_family) {
1443 /* We support multiple FIBs. */
1446 fibnum = RT_DEFAULT_FIB;
1450 if (rt_add_addr_allfibs == 0 && cmd == (int)RTM_ADD) {
1451 startfib = endfib = curthread->td_proc->p_fibnum;
1454 endfib = rt_numfibs - 1;
1457 KASSERT((fibnum < rt_numfibs), ("rtinit1: bad fibnum"));
1463 * If it's a delete, check that if it exists,
1464 * it's on the correct interface or we might scrub
1465 * a route to another ifa which would
1466 * be confusing at best and possibly worse.
1468 if (cmd == RTM_DELETE) {
1470 * It's a delete, so it should already exist..
1471 * If it's a net, mask off the host bits
1472 * (Assuming we have a mask)
1473 * XXX this is kinda inet specific..
1475 if (netmask != NULL) {
1476 rt_maskedcopy(dst, (struct sockaddr *)tempbuf, netmask);
1477 dst = (struct sockaddr *)tempbuf;
1481 * Now go through all the requested tables (fibs) and do the
1482 * requested action. Realistically, this will either be fib 0
1483 * for protocols that don't do multiple tables or all the
1484 * tables for those that do.
1486 for ( fibnum = startfib; fibnum <= endfib; fibnum++) {
1487 if (cmd == RTM_DELETE) {
1488 struct radix_node *rn;
1490 * Look up an rtentry that is in the routing tree and
1491 * contains the correct info.
1493 rnh = rt_tables_get_rnh(fibnum, dst->sa_family);
1495 /* this table doesn't exist but others might */
1497 RADIX_NODE_HEAD_RLOCK(rnh);
1499 if (rn_mpath_capable(rnh)) {
1501 rn = rnh->rnh_matchaddr(dst, rnh);
1507 * for interface route the
1508 * rt->rt_gateway is sockaddr_intf
1509 * for cloning ARP entries, so
1510 * rt_mpath_matchgate must use the
1513 rt = rt_mpath_matchgate(rt,
1521 rn = rnh->rnh_lookup(dst, netmask, rnh);
1522 error = (rn == NULL ||
1523 (rn->rn_flags & RNF_ROOT) ||
1524 RNTORT(rn)->rt_ifa != ifa ||
1525 !sa_equal((struct sockaddr *)rn->rn_key, dst));
1526 RADIX_NODE_HEAD_RUNLOCK(rnh);
1528 /* this is only an error if bad on ALL tables */
1533 * Do the actual request
1535 bzero((caddr_t)&info, sizeof(info));
1537 info.rti_flags = flags |
1538 (ifa->ifa_flags & ~IFA_RTSELF) | RTF_PINNED;
1539 info.rti_info[RTAX_DST] = dst;
1541 * doing this for compatibility reasons
1544 info.rti_info[RTAX_GATEWAY] =
1545 (struct sockaddr *)&null_sdl;
1547 info.rti_info[RTAX_GATEWAY] = ifa->ifa_addr;
1548 info.rti_info[RTAX_NETMASK] = netmask;
1549 error = rtrequest1_fib(cmd, &info, &rt, fibnum);
1551 if ((error == EEXIST) && (cmd == RTM_ADD)) {
1553 * Interface route addition failed.
1554 * Atomically delete current prefix generating
1555 * RTM_DELETE message, and retry adding
1558 rnh = rt_tables_get_rnh(fibnum, dst->sa_family);
1559 RADIX_NODE_HEAD_LOCK(rnh);
1561 /* Delete old prefix */
1562 info.rti_ifa = NULL;
1563 info.rti_flags = RTF_RNH_LOCKED;
1565 error = rtrequest1_fib(RTM_DELETE, &info, NULL, fibnum);
1568 info.rti_flags = flags | RTF_RNH_LOCKED |
1569 (ifa->ifa_flags & ~IFA_RTSELF) | RTF_PINNED;
1570 error = rtrequest1_fib(cmd, &info, &rt, fibnum);
1573 RADIX_NODE_HEAD_UNLOCK(rnh);
1577 if (error == 0 && rt != NULL) {
1579 * notify any listening routing agents of the change
1584 * in case address alias finds the first address
1585 * e.g. ifconfig bge0 192.0.2.246/24
1586 * e.g. ifconfig bge0 192.0.2.247/24
1587 * the address set in the route is 192.0.2.246
1588 * so we need to replace it with 192.0.2.247
1590 if (memcmp(rt->rt_ifa->ifa_addr,
1591 ifa->ifa_addr, ifa->ifa_addr->sa_len)) {
1592 ifa_free(rt->rt_ifa);
1594 rt->rt_ifp = ifa->ifa_ifp;
1599 * doing this for compatibility reasons
1601 if (cmd == RTM_ADD) {
1602 ((struct sockaddr_dl *)rt->rt_gateway)->sdl_type =
1603 rt->rt_ifp->if_type;
1604 ((struct sockaddr_dl *)rt->rt_gateway)->sdl_index =
1605 rt->rt_ifp->if_index;
1609 rt_newaddrmsg_fib(cmd, ifa, error, rt, fibnum);
1612 if (cmd == RTM_DELETE) {
1614 * If we are deleting, and we found an entry,
1615 * then it's been removed from the tree..
1616 * now throw it away.
1620 if (cmd == RTM_ADD) {
1622 * We just wanted to add it..
1623 * we don't actually need a reference.
1634 if (cmd == RTM_DELETE) {
1638 /* we only give an error if it wasn't in any table */
1639 error = ((flags & RTF_HOST) ?
1640 EHOSTUNREACH : ENETUNREACH);
1644 /* return an error if any of them failed */
1651 #ifndef BURN_BRIDGES
1652 /* special one for inet internal use. may not use. */
1654 rtinit_fib(struct ifaddr *ifa, int cmd, int flags)
1656 return (rtinit1(ifa, cmd, flags, -1));
1661 * Set up a routing table entry, normally
1665 rtinit(struct ifaddr *ifa, int cmd, int flags)
1667 struct sockaddr *dst;
1668 int fib = RT_DEFAULT_FIB;
1670 if (flags & RTF_HOST) {
1671 dst = ifa->ifa_dstaddr;
1673 dst = ifa->ifa_addr;
1676 switch (dst->sa_family) {
1679 /* We do support multiple FIBs. */
1683 return (rtinit1(ifa, cmd, flags, fib));