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 #define RT_MAXFIBS UINT16_MAX
73 /* Kernel config default option. */
76 #error "ROUTETABLES defined too low"
78 #if ROUTETABLES > RT_MAXFIBS
79 #error "ROUTETABLES defined too big"
81 #define RT_NUMFIBS ROUTETABLES
82 #endif /* ROUTETABLES */
83 /* Initialize to default if not otherwise set. */
88 /* This is read-only.. */
89 u_int rt_numfibs = RT_NUMFIBS;
90 SYSCTL_UINT(_net, OID_AUTO, fibs, CTLFLAG_RD, &rt_numfibs, 0, "");
91 /* and this can be set too big but will be fixed before it is used */
92 TUNABLE_INT("net.fibs", &rt_numfibs);
95 * By default add routes to all fibs for new interfaces.
96 * Once this is set to 0 then only allocate routes on interface
97 * changes for the FIB of the caller when adding a new set of addresses
98 * to an interface. XXX this is a shotgun aproach to a problem that needs
99 * a more fine grained solution.. that will come.
100 * XXX also has the problems getting the FIB from curthread which will not
101 * always work given the fib can be overridden and prefixes can be added
102 * from the network stack context.
104 u_int rt_add_addr_allfibs = 1;
105 SYSCTL_UINT(_net, OID_AUTO, add_addr_allfibs, CTLFLAG_RW,
106 &rt_add_addr_allfibs, 0, "");
107 TUNABLE_INT("net.add_addr_allfibs", &rt_add_addr_allfibs);
109 VNET_DEFINE(struct rtstat, rtstat);
110 #define V_rtstat VNET(rtstat)
112 VNET_DEFINE(struct radix_node_head *, rt_tables);
113 #define V_rt_tables VNET(rt_tables)
115 VNET_DEFINE(int, rttrash); /* routes not in table but not freed */
116 #define V_rttrash VNET(rttrash)
119 /* compare two sockaddr structures */
120 #define sa_equal(a1, a2) (bcmp((a1), (a2), (a1)->sa_len) == 0)
123 * Convert a 'struct radix_node *' to a 'struct rtentry *'.
124 * The operation can be done safely (in this code) because a
125 * 'struct rtentry' starts with two 'struct radix_node''s, the first
126 * one representing leaf nodes in the routing tree, which is
127 * what the code in radix.c passes us as a 'struct radix_node'.
129 * But because there are a lot of assumptions in this conversion,
130 * do not cast explicitly, but always use the macro below.
132 #define RNTORT(p) ((struct rtentry *)(p))
134 static VNET_DEFINE(uma_zone_t, rtzone); /* Routing table UMA zone. */
135 #define V_rtzone VNET(rtzone)
138 * handler for net.my_fibnum
141 sysctl_my_fibnum(SYSCTL_HANDLER_ARGS)
146 fibnum = curthread->td_proc->p_fibnum;
147 error = sysctl_handle_int(oidp, &fibnum, 0, req);
151 SYSCTL_PROC(_net, OID_AUTO, my_fibnum, CTLTYPE_INT|CTLFLAG_RD,
152 NULL, 0, &sysctl_my_fibnum, "I", "default FIB of caller");
154 static __inline struct radix_node_head **
155 rt_tables_get_rnh_ptr(int table, int fam)
157 struct radix_node_head **rnh;
159 KASSERT(table >= 0 && table < rt_numfibs, ("%s: table out of bounds.",
161 KASSERT(fam >= 0 && fam < (AF_MAX+1), ("%s: fam out of bounds.",
164 /* rnh is [fib=0][af=0]. */
165 rnh = (struct radix_node_head **)V_rt_tables;
166 /* Get the offset to the requested table and fam. */
167 rnh += table * (AF_MAX+1) + fam;
172 struct radix_node_head *
173 rt_tables_get_rnh(int table, int fam)
176 return (*rt_tables_get_rnh_ptr(table, fam));
180 * route initialization must occur before ip6_init2(), which happenas at
189 /* whack the tunable ints into line. */
190 if (rt_numfibs > RT_MAXFIBS)
191 rt_numfibs = RT_MAXFIBS;
195 for (dom = domains; dom; dom = dom->dom_next)
196 if (dom->dom_maxrtkey > max_keylen)
197 max_keylen = dom->dom_maxrtkey;
199 rn_init(max_keylen); /* init all zeroes, all ones, mask table */
201 SYSINIT(route_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_THIRD, route_init, 0);
204 vnet_route_init(const void *unused __unused)
207 struct radix_node_head **rnh;
211 V_rt_tables = malloc(rt_numfibs * (AF_MAX+1) *
212 sizeof(struct radix_node_head *), M_RTABLE, M_WAITOK|M_ZERO);
214 V_rtzone = uma_zcreate("rtentry", sizeof(struct rtentry), NULL, NULL,
215 NULL, NULL, UMA_ALIGN_PTR, 0);
216 for (dom = domains; dom; dom = dom->dom_next) {
217 if (dom->dom_rtattach == NULL)
220 for (table = 0; table < rt_numfibs; table++) {
221 fam = dom->dom_family;
222 if (table != 0 && fam != AF_INET6 && fam != AF_INET)
226 * XXX MRT rtattach will be also called from
227 * vfs_export.c but the offset will be 0 (only for
228 * AF_INET and AF_INET6 which don't need it anyhow).
230 rnh = rt_tables_get_rnh_ptr(table, fam);
232 panic("%s: rnh NULL", __func__);
233 dom->dom_rtattach((void **)rnh, dom->dom_rtoffset);
237 VNET_SYSINIT(vnet_route_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_FOURTH,
242 vnet_route_uninit(const void *unused __unused)
247 struct radix_node_head **rnh;
249 for (dom = domains; dom; dom = dom->dom_next) {
250 if (dom->dom_rtdetach == NULL)
253 for (table = 0; table < rt_numfibs; table++) {
254 fam = dom->dom_family;
256 if (table != 0 && fam != AF_INET6 && fam != AF_INET)
259 rnh = rt_tables_get_rnh_ptr(table, fam);
261 panic("%s: rnh NULL", __func__);
262 dom->dom_rtdetach((void **)rnh, dom->dom_rtoffset);
266 VNET_SYSUNINIT(vnet_route_uninit, SI_SUB_PROTO_DOMAIN, SI_ORDER_THIRD,
267 vnet_route_uninit, 0);
270 #ifndef _SYS_SYSPROTO_H_
276 sys_setfib(struct thread *td, struct setfib_args *uap)
278 if (uap->fibnum < 0 || uap->fibnum >= rt_numfibs)
280 td->td_proc->p_fibnum = uap->fibnum;
285 * Packet routing routines.
288 rtalloc(struct route *ro)
291 rtalloc_ign_fib(ro, 0UL, RT_DEFAULT_FIB);
295 rtalloc_fib(struct route *ro, u_int fibnum)
297 rtalloc_ign_fib(ro, 0UL, fibnum);
301 rtalloc_ign(struct route *ro, u_long ignore)
305 if ((rt = ro->ro_rt) != NULL) {
306 if (rt->rt_ifp != NULL && rt->rt_flags & RTF_UP)
311 ro->ro_rt = rtalloc1_fib(&ro->ro_dst, 1, ignore, RT_DEFAULT_FIB);
313 RT_UNLOCK(ro->ro_rt);
317 rtalloc_ign_fib(struct route *ro, u_long ignore, u_int fibnum)
321 if ((rt = ro->ro_rt) != NULL) {
322 if (rt->rt_ifp != NULL && rt->rt_flags & RTF_UP)
327 ro->ro_rt = rtalloc1_fib(&ro->ro_dst, 1, ignore, fibnum);
329 RT_UNLOCK(ro->ro_rt);
333 * Look up the route that matches the address given
334 * Or, at least try.. Create a cloned route if needed.
336 * The returned route, if any, is locked.
339 rtalloc1(struct sockaddr *dst, int report, u_long ignflags)
342 return (rtalloc1_fib(dst, report, ignflags, RT_DEFAULT_FIB));
346 rtalloc1_fib(struct sockaddr *dst, int report, u_long ignflags,
349 struct radix_node_head *rnh;
350 struct radix_node *rn;
351 struct rtentry *newrt;
352 struct rt_addrinfo info;
353 int err = 0, msgtype = RTM_MISS;
356 KASSERT((fibnum < rt_numfibs), ("rtalloc1_fib: bad fibnum"));
357 switch (dst->sa_family) {
360 /* We support multiple FIBs. */
363 fibnum = RT_DEFAULT_FIB;
366 rnh = rt_tables_get_rnh(fibnum, dst->sa_family);
372 * Look up the address in the table for that Address Family
374 needlock = !(ignflags & RTF_RNH_LOCKED);
376 RADIX_NODE_HEAD_RLOCK(rnh);
379 RADIX_NODE_HEAD_LOCK_ASSERT(rnh);
381 rn = rnh->rnh_matchaddr(dst, rnh);
382 if (rn && ((rn->rn_flags & RNF_ROOT) == 0)) {
387 RADIX_NODE_HEAD_RUNLOCK(rnh);
391 RADIX_NODE_HEAD_RUNLOCK(rnh);
394 * Either we hit the root or couldn't find any match,
395 * Which basically means
396 * "caint get there frm here"
399 V_rtstat.rts_unreach++;
403 * If required, report the failure to the supervising
405 * For a delete, this is not an error. (report == 0)
407 bzero(&info, sizeof(info));
408 info.rti_info[RTAX_DST] = dst;
409 rt_missmsg_fib(msgtype, &info, 0, err, fibnum);
413 RT_LOCK_ASSERT(newrt);
418 * Remove a reference count from an rtentry.
419 * If the count gets low enough, take it out of the routing table
422 rtfree(struct rtentry *rt)
424 struct radix_node_head *rnh;
426 KASSERT(rt != NULL,("%s: NULL rt", __func__));
427 rnh = rt_tables_get_rnh(rt->rt_fibnum, rt_key(rt)->sa_family);
428 KASSERT(rnh != NULL,("%s: NULL rnh", __func__));
433 * The callers should use RTFREE_LOCKED() or RTFREE(), so
434 * we should come here exactly with the last reference.
437 if (rt->rt_refcnt > 0) {
438 log(LOG_DEBUG, "%s: %p has %d refs\n", __func__, rt, rt->rt_refcnt);
443 * On last reference give the "close method" a chance
444 * to cleanup private state. This also permits (for
445 * IPv4 and IPv6) a chance to decide if the routing table
446 * entry should be purged immediately or at a later time.
447 * When an immediate purge is to happen the close routine
448 * typically calls rtexpunge which clears the RTF_UP flag
449 * on the entry so that the code below reclaims the storage.
451 if (rt->rt_refcnt == 0 && rnh->rnh_close)
452 rnh->rnh_close((struct radix_node *)rt, rnh);
455 * If we are no longer "up" (and ref == 0)
456 * then we can free the resources associated
459 if ((rt->rt_flags & RTF_UP) == 0) {
460 if (rt->rt_nodes->rn_flags & (RNF_ACTIVE | RNF_ROOT))
463 * the rtentry must have been removed from the routing table
464 * so it is represented in rttrash.. remove that now.
468 if (rt->rt_refcnt < 0) {
469 printf("rtfree: %p not freed (neg refs)\n", rt);
474 * release references on items we hold them on..
475 * e.g other routes and ifaddrs.
478 ifa_free(rt->rt_ifa);
480 * The key is separatly alloc'd so free it (see rt_setgate()).
481 * This also frees the gateway, as they are always malloc'd
487 * and the rtentry itself of course
490 uma_zfree(V_rtzone, rt);
499 * Force a routing table entry to the specified
500 * destination to go through the given gateway.
501 * Normally called as a result of a routing redirect
502 * message from the network layer.
505 rtredirect(struct sockaddr *dst,
506 struct sockaddr *gateway,
507 struct sockaddr *netmask,
509 struct sockaddr *src)
512 rtredirect_fib(dst, gateway, netmask, flags, src, RT_DEFAULT_FIB);
516 rtredirect_fib(struct sockaddr *dst,
517 struct sockaddr *gateway,
518 struct sockaddr *netmask,
520 struct sockaddr *src,
523 struct rtentry *rt, *rt0 = NULL;
526 struct rt_addrinfo info;
528 struct radix_node_head *rnh;
531 rnh = rt_tables_get_rnh(fibnum, dst->sa_family);
533 error = EAFNOSUPPORT;
537 /* verify the gateway is directly reachable */
538 if ((ifa = ifa_ifwithnet(gateway, 0)) == NULL) {
542 rt = rtalloc1_fib(dst, 0, 0UL, fibnum); /* NB: rt is locked */
544 * If the redirect isn't from our current router for this dst,
545 * it's either old or wrong. If it redirects us to ourselves,
546 * we have a routing loop, perhaps as a result of an interface
547 * going down recently.
549 if (!(flags & RTF_DONE) && rt &&
550 (!sa_equal(src, rt->rt_gateway) || rt->rt_ifa != ifa))
552 else if (ifa_ifwithaddr_check(gateway))
553 error = EHOSTUNREACH;
557 * Create a new entry if we just got back a wildcard entry
558 * or the lookup failed. This is necessary for hosts
559 * which use routing redirects generated by smart gateways
560 * to dynamically build the routing tables.
562 if (rt == NULL || (rt_mask(rt) && rt_mask(rt)->sa_len < 2))
565 * Don't listen to the redirect if it's
566 * for a route to an interface.
568 if (rt->rt_flags & RTF_GATEWAY) {
569 if (((rt->rt_flags & RTF_HOST) == 0) && (flags & RTF_HOST)) {
571 * Changing from route to net => route to host.
572 * Create new route, rather than smashing route to net.
578 flags |= RTF_GATEWAY | RTF_DYNAMIC;
579 bzero((caddr_t)&info, sizeof(info));
580 info.rti_info[RTAX_DST] = dst;
581 info.rti_info[RTAX_GATEWAY] = gateway;
582 info.rti_info[RTAX_NETMASK] = netmask;
584 info.rti_flags = flags;
586 RT_UNLOCK(rt0); /* drop lock to avoid LOR with RNH */
587 error = rtrequest1_fib(RTM_ADD, &info, &rt, fibnum);
591 EVENTHANDLER_INVOKE(route_redirect_event, rt0, rt, dst);
592 flags = rt->rt_flags;
597 stat = &V_rtstat.rts_dynamic;
599 struct rtentry *gwrt;
602 * Smash the current notion of the gateway to
603 * this destination. Should check about netmask!!!
605 rt->rt_flags |= RTF_MODIFIED;
606 flags |= RTF_MODIFIED;
607 stat = &V_rtstat.rts_newgateway;
609 * add the key and gateway (in one malloc'd chunk).
612 RADIX_NODE_HEAD_LOCK(rnh);
614 rt_setgate(rt, rt_key(rt), gateway);
615 gwrt = rtalloc1(gateway, 1, RTF_RNH_LOCKED);
616 RADIX_NODE_HEAD_UNLOCK(rnh);
617 EVENTHANDLER_INVOKE(route_redirect_event, rt, gwrt, dst);
621 error = EHOSTUNREACH;
627 V_rtstat.rts_badredirect++;
628 else if (stat != NULL)
630 bzero((caddr_t)&info, sizeof(info));
631 info.rti_info[RTAX_DST] = dst;
632 info.rti_info[RTAX_GATEWAY] = gateway;
633 info.rti_info[RTAX_NETMASK] = netmask;
634 info.rti_info[RTAX_AUTHOR] = src;
635 rt_missmsg_fib(RTM_REDIRECT, &info, flags, error, fibnum);
641 rtioctl(u_long req, caddr_t data)
644 return (rtioctl_fib(req, data, RT_DEFAULT_FIB));
648 * Routing table ioctl interface.
651 rtioctl_fib(u_long req, caddr_t data, u_int fibnum)
655 * If more ioctl commands are added here, make sure the proper
656 * super-user checks are being performed because it is possible for
657 * prison-root to make it this far if raw sockets have been enabled
661 /* Multicast goop, grrr... */
662 return mrt_ioctl ? mrt_ioctl(req, data, fibnum) : EOPNOTSUPP;
669 * For both ifa_ifwithroute() routines, 'ifa' is returned referenced.
672 ifa_ifwithroute(int flags, struct sockaddr *dst, struct sockaddr *gateway)
675 return (ifa_ifwithroute_fib(flags, dst, gateway, RT_DEFAULT_FIB));
679 ifa_ifwithroute_fib(int flags, struct sockaddr *dst, struct sockaddr *gateway,
682 register struct ifaddr *ifa;
685 if ((flags & RTF_GATEWAY) == 0) {
687 * If we are adding a route to an interface,
688 * and the interface is a pt to pt link
689 * we should search for the destination
690 * as our clue to the interface. Otherwise
691 * we can use the local address.
694 if (flags & RTF_HOST)
695 ifa = ifa_ifwithdstaddr(dst);
697 ifa = ifa_ifwithaddr(gateway);
700 * If we are adding a route to a remote net
701 * or host, the gateway may still be on the
702 * other end of a pt to pt link.
704 ifa = ifa_ifwithdstaddr(gateway);
707 ifa = ifa_ifwithnet(gateway, 0);
709 struct rtentry *rt = rtalloc1_fib(gateway, 0, RTF_RNH_LOCKED, fibnum);
713 * dismiss a gateway that is reachable only
714 * through the default router
716 switch (gateway->sa_family) {
718 if (satosin(rt_key(rt))->sin_addr.s_addr == INADDR_ANY)
722 if (IN6_IS_ADDR_UNSPECIFIED(&satosin6(rt_key(rt))->sin6_addr))
728 if (!not_found && rt->rt_ifa != NULL) {
734 if (not_found || ifa == NULL)
737 if (ifa->ifa_addr->sa_family != dst->sa_family) {
738 struct ifaddr *oifa = ifa;
739 ifa = ifaof_ifpforaddr(dst, ifa->ifa_ifp);
749 * Do appropriate manipulations of a routing tree given
750 * all the bits of info needed
754 struct sockaddr *dst,
755 struct sockaddr *gateway,
756 struct sockaddr *netmask,
758 struct rtentry **ret_nrt)
761 return (rtrequest_fib(req, dst, gateway, netmask, flags, ret_nrt,
766 rtrequest_fib(int req,
767 struct sockaddr *dst,
768 struct sockaddr *gateway,
769 struct sockaddr *netmask,
771 struct rtentry **ret_nrt,
774 struct rt_addrinfo info;
776 if (dst->sa_len == 0)
779 bzero((caddr_t)&info, sizeof(info));
780 info.rti_flags = flags;
781 info.rti_info[RTAX_DST] = dst;
782 info.rti_info[RTAX_GATEWAY] = gateway;
783 info.rti_info[RTAX_NETMASK] = netmask;
784 return rtrequest1_fib(req, &info, ret_nrt, fibnum);
788 * These (questionable) definitions of apparent local variables apply
789 * to the next two functions. XXXXXX!!!
791 #define dst info->rti_info[RTAX_DST]
792 #define gateway info->rti_info[RTAX_GATEWAY]
793 #define netmask info->rti_info[RTAX_NETMASK]
794 #define ifaaddr info->rti_info[RTAX_IFA]
795 #define ifpaddr info->rti_info[RTAX_IFP]
796 #define flags info->rti_flags
799 rt_getifa(struct rt_addrinfo *info)
802 return (rt_getifa_fib(info, RT_DEFAULT_FIB));
806 * Look up rt_addrinfo for a specific fib. Note that if rti_ifa is defined,
807 * it will be referenced so the caller must free it.
810 rt_getifa_fib(struct rt_addrinfo *info, u_int fibnum)
816 * ifp may be specified by sockaddr_dl
817 * when protocol address is ambiguous.
819 if (info->rti_ifp == NULL && ifpaddr != NULL &&
820 ifpaddr->sa_family == AF_LINK &&
821 (ifa = ifa_ifwithnet(ifpaddr, 0)) != NULL) {
822 info->rti_ifp = ifa->ifa_ifp;
825 if (info->rti_ifa == NULL && ifaaddr != NULL)
826 info->rti_ifa = ifa_ifwithaddr(ifaaddr);
827 if (info->rti_ifa == NULL) {
830 sa = ifaaddr != NULL ? ifaaddr :
831 (gateway != NULL ? gateway : dst);
832 if (sa != NULL && info->rti_ifp != NULL)
833 info->rti_ifa = ifaof_ifpforaddr(sa, info->rti_ifp);
834 else if (dst != NULL && gateway != NULL)
835 info->rti_ifa = ifa_ifwithroute_fib(flags, dst, gateway,
838 info->rti_ifa = ifa_ifwithroute_fib(flags, sa, sa,
841 if ((ifa = info->rti_ifa) != NULL) {
842 if (info->rti_ifp == NULL)
843 info->rti_ifp = ifa->ifa_ifp;
850 * Expunges references to a route that's about to be reclaimed.
851 * The route must be locked.
854 rtexpunge(struct rtentry *rt)
856 #if !defined(RADIX_MPATH)
857 struct radix_node *rn;
859 struct rt_addrinfo info;
863 struct radix_node_head *rnh;
868 * Find the correct routing tree to use for this Address Family
870 rnh = rt_tables_get_rnh(rt->rt_fibnum, rt_key(rt)->sa_family);
873 return (EAFNOSUPPORT);
874 RADIX_NODE_HEAD_LOCK_ASSERT(rnh);
878 bzero(&info, sizeof(info));
879 info.rti_ifp = rt->rt_ifp;
880 info.rti_flags = RTF_RNH_LOCKED;
881 info.rti_info[RTAX_DST] = rt_key(rt);
882 info.rti_info[RTAX_GATEWAY] = rt->rt_ifa->ifa_addr;
885 error = rtrequest1_fib(RTM_DELETE, &info, &rt0, fib);
887 if (error == 0 && rt0 != NULL) {
890 } else if (error != 0) {
896 * Remove the item from the tree; it should be there,
897 * but when callers invoke us blindly it may not (sigh).
899 rn = rnh->rnh_deladdr(rt_key(rt), rt_mask(rt), rnh);
904 KASSERT((rn->rn_flags & (RNF_ACTIVE | RNF_ROOT)) == 0,
905 ("unexpected flags 0x%x", rn->rn_flags));
906 KASSERT(rt == RNTORT(rn),
907 ("lookup mismatch, rt %p rn %p", rt, rn));
908 #endif /* RADIX_MPATH */
910 rt->rt_flags &= ~RTF_UP;
913 * Give the protocol a chance to keep things in sync.
915 if ((ifa = rt->rt_ifa) && ifa->ifa_rtrequest) {
916 struct rt_addrinfo info;
918 bzero((caddr_t)&info, sizeof(info));
919 info.rti_flags = rt->rt_flags;
920 info.rti_info[RTAX_DST] = rt_key(rt);
921 info.rti_info[RTAX_GATEWAY] = rt->rt_gateway;
922 info.rti_info[RTAX_NETMASK] = rt_mask(rt);
923 ifa->ifa_rtrequest(RTM_DELETE, rt, &info);
927 * one more rtentry floating around that is not
928 * linked to the routing table.
931 #if !defined(RADIX_MPATH)
939 rn_mpath_update(int req, struct rt_addrinfo *info,
940 struct radix_node_head *rnh, struct rtentry **ret_nrt)
943 * if we got multipath routes, we require users to specify
944 * a matching RTAX_GATEWAY.
946 struct rtentry *rt, *rto = NULL;
947 register struct radix_node *rn;
950 rn = rnh->rnh_matchaddr(dst, rnh);
953 rto = rt = RNTORT(rn);
954 rt = rt_mpath_matchgate(rt, gateway);
958 * this is the first entry in the chain
961 rn = rn_mpath_next((struct radix_node *)rt);
963 * there is another entry, now it's active
968 rto->rt_flags |= RTF_UP;
970 } else if (rt->rt_flags & RTF_GATEWAY) {
972 * For gateway routes, we need to
973 * make sure that we we are deleting
974 * the correct gateway.
975 * rt_mpath_matchgate() does not
976 * check the case when there is only
977 * one route in the chain.
980 (rt->rt_gateway->sa_len != gateway->sa_len ||
981 memcmp(rt->rt_gateway, gateway, gateway->sa_len)))
985 * remove from tree before returning it
988 rn = rnh->rnh_deladdr(dst, netmask, rnh);
989 KASSERT(rt == RNTORT(rn), ("radix node disappeared"));
995 * use the normal delete code to remove
998 if (req != RTM_DELETE)
1006 * if the entry is 2nd and on up
1008 if ((req == RTM_DELETE) && !rt_mpath_deldup(rto, rt))
1009 panic ("rtrequest1: rt_mpath_deldup");
1013 if (req == RTM_DELETE) {
1014 rt->rt_flags &= ~RTF_UP;
1016 * One more rtentry floating around that is not
1017 * linked to the routing table. rttrash will be decremented
1018 * when RTFREE(rt) is eventually called.
1024 if (req != RTM_DELETE)
1025 panic("unrecognized request %d", req);
1029 * If the caller wants it, then it can have it,
1030 * but it's up to it to free the rtentry as we won't be
1044 rtrequest1_fib(int req, struct rt_addrinfo *info, struct rtentry **ret_nrt,
1047 int error = 0, needlock = 0;
1048 register struct rtentry *rt;
1050 register struct rtentry *rt0;
1052 register struct radix_node *rn;
1053 register struct radix_node_head *rnh;
1055 struct sockaddr *ndst;
1056 struct sockaddr_storage mdst;
1057 #define senderr(x) { error = x ; goto bad; }
1059 KASSERT((fibnum < rt_numfibs), ("rtrequest1_fib: bad fibnum"));
1060 switch (dst->sa_family) {
1063 /* We support multiple FIBs. */
1066 fibnum = RT_DEFAULT_FIB;
1071 * Find the correct routing tree to use for this Address Family
1073 rnh = rt_tables_get_rnh(fibnum, dst->sa_family);
1075 return (EAFNOSUPPORT);
1076 needlock = ((flags & RTF_RNH_LOCKED) == 0);
1077 flags &= ~RTF_RNH_LOCKED;
1079 RADIX_NODE_HEAD_LOCK(rnh);
1081 RADIX_NODE_HEAD_LOCK_ASSERT(rnh);
1083 * If we are adding a host route then we don't want to put
1084 * a netmask in the tree, nor do we want to clone it.
1086 if (flags & RTF_HOST)
1092 rt_maskedcopy(dst, (struct sockaddr *)&mdst, netmask);
1093 dst = (struct sockaddr *)&mdst;
1096 if (rn_mpath_capable(rnh)) {
1097 error = rn_mpath_update(req, info, rnh, ret_nrt);
1099 * "bad" holds true for the success case
1102 if (error != ENOENT)
1107 if ((flags & RTF_PINNED) == 0) {
1108 /* Check if target route can be deleted */
1109 rt = (struct rtentry *)rnh->rnh_lookup(dst,
1111 if ((rt != NULL) && (rt->rt_flags & RTF_PINNED))
1112 senderr(EADDRINUSE);
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);
1179 rt->rt_flags = RTF_UP | flags;
1180 rt->rt_fibnum = fibnum;
1182 * Add the gateway. Possibly re-malloc-ing the storage for it.
1185 if ((error = rt_setgate(rt, dst, gateway)) != 0) {
1186 RT_LOCK_DESTROY(rt);
1188 uma_zfree(V_rtzone, rt);
1193 * point to the (possibly newly malloc'd) dest address.
1195 ndst = (struct sockaddr *)rt_key(rt);
1198 * make sure it contains the value we want (masked if needed).
1201 rt_maskedcopy(dst, ndst, netmask);
1203 bcopy(dst, ndst, dst->sa_len);
1206 * We use the ifa reference returned by rt_getifa_fib().
1207 * This moved from below so that rnh->rnh_addaddr() can
1208 * examine the ifa and ifa->ifa_ifp if it so desires.
1211 rt->rt_ifp = ifa->ifa_ifp;
1212 rt->rt_rmx.rmx_weight = 1;
1215 /* do not permit exactly the same dst/mask/gw pair */
1216 if (rn_mpath_capable(rnh) &&
1217 rt_mpath_conflict(rnh, rt, netmask)) {
1218 ifa_free(rt->rt_ifa);
1220 RT_LOCK_DESTROY(rt);
1221 uma_zfree(V_rtzone, rt);
1228 /* "flow-table" only supports IPv6 and IPv4 at the moment. */
1229 switch (dst->sa_family) {
1236 #if defined(INET6) || defined(INET)
1237 rn = rnh->rnh_matchaddr(dst, rnh);
1238 if (rn && ((rn->rn_flags & RNF_ROOT) == 0)) {
1239 struct sockaddr *mask;
1244 * compare mask to see if the new route is
1245 * more specific than the existing one
1252 * A host route is already present, so
1253 * leave the flow-table entries as is.
1255 if (rt0->rt_flags & RTF_HOST) {
1258 } else if (!(flags & RTF_HOST) && netmask) {
1259 mask = rt_mask(rt0);
1262 n = (u_char *)netmask;
1269 if (len == 0 || (*n < *m)) {
1275 #endif/* INET6 || INET */
1277 #endif /* FLOWTABLE */
1279 /* XXX mtu manipulation will be done in rnh_addaddr -- itojun */
1280 rn = rnh->rnh_addaddr(ndst, netmask, rnh, rt->rt_nodes);
1282 * If it still failed to go into the tree,
1283 * then un-make it (this should be a function)
1286 ifa_free(rt->rt_ifa);
1288 RT_LOCK_DESTROY(rt);
1289 uma_zfree(V_rtzone, rt);
1297 else if (rt0 != NULL) {
1298 switch (dst->sa_family) {
1301 flowtable_route_flush(V_ip6_ft, rt0);
1306 flowtable_route_flush(V_ip_ft, rt0);
1315 * If this protocol has something to add to this then
1316 * allow it to do that as well.
1318 if (ifa->ifa_rtrequest)
1319 ifa->ifa_rtrequest(req, rt, info);
1322 * actually return a resultant rtentry and
1323 * give the caller a single reference.
1336 RADIX_NODE_HEAD_UNLOCK(rnh);
1349 rt_setgate(struct rtentry *rt, struct sockaddr *dst, struct sockaddr *gate)
1351 /* XXX dst may be overwritten, can we move this to below */
1352 int dlen = SA_SIZE(dst), glen = SA_SIZE(gate);
1354 struct radix_node_head *rnh;
1356 rnh = rt_tables_get_rnh(rt->rt_fibnum, dst->sa_family);
1360 RADIX_NODE_HEAD_LOCK_ASSERT(rnh);
1363 * Prepare to store the gateway in rt->rt_gateway.
1364 * Both dst and gateway are stored one after the other in the same
1365 * malloc'd chunk. If we have room, we can reuse the old buffer,
1366 * rt_gateway already points to the right place.
1367 * Otherwise, malloc a new block and update the 'dst' address.
1369 if (rt->rt_gateway == NULL || glen > SA_SIZE(rt->rt_gateway)) {
1372 R_Malloc(new, caddr_t, dlen + glen);
1376 * XXX note, we copy from *dst and not *rt_key(rt) because
1377 * rt_setgate() can be called to initialize a newly
1378 * allocated route entry, in which case rt_key(rt) == NULL
1379 * (and also rt->rt_gateway == NULL).
1380 * Free()/free() handle a NULL argument just fine.
1382 bcopy(dst, new, dlen);
1383 Free(rt_key(rt)); /* free old block, if any */
1384 rt_key(rt) = (struct sockaddr *)new;
1385 rt->rt_gateway = (struct sockaddr *)(new + dlen);
1389 * Copy the new gateway value into the memory chunk.
1391 bcopy(gate, rt->rt_gateway, glen);
1397 rt_maskedcopy(struct sockaddr *src, struct sockaddr *dst, struct sockaddr *netmask)
1399 register u_char *cp1 = (u_char *)src;
1400 register u_char *cp2 = (u_char *)dst;
1401 register u_char *cp3 = (u_char *)netmask;
1402 u_char *cplim = cp2 + *cp3;
1403 u_char *cplim2 = cp2 + *cp1;
1405 *cp2++ = *cp1++; *cp2++ = *cp1++; /* copies sa_len & sa_family */
1410 *cp2++ = *cp1++ & *cp3++;
1412 bzero((caddr_t)cp2, (unsigned)(cplim2 - cp2));
1416 * Set up a routing table entry, normally
1419 #define _SOCKADDR_TMPSIZE 128 /* Not too big.. kernel stack size is limited */
1421 rtinit1(struct ifaddr *ifa, int cmd, int flags, int fibnum)
1423 struct sockaddr *dst;
1424 struct sockaddr *netmask;
1425 struct rtentry *rt = NULL;
1426 struct rt_addrinfo info;
1428 int startfib, endfib;
1429 char tempbuf[_SOCKADDR_TMPSIZE];
1432 static struct sockaddr_dl null_sdl = {sizeof(null_sdl), AF_LINK};
1433 struct radix_node_head *rnh;
1435 if (flags & RTF_HOST) {
1436 dst = ifa->ifa_dstaddr;
1439 dst = ifa->ifa_addr;
1440 netmask = ifa->ifa_netmask;
1442 if (dst->sa_len == 0)
1444 switch (dst->sa_family) {
1447 /* We support multiple FIBs. */
1450 fibnum = RT_DEFAULT_FIB;
1454 if (rt_add_addr_allfibs == 0 && cmd == (int)RTM_ADD) {
1455 startfib = endfib = curthread->td_proc->p_fibnum;
1458 endfib = rt_numfibs - 1;
1461 KASSERT((fibnum < rt_numfibs), ("rtinit1: bad fibnum"));
1467 * If it's a delete, check that if it exists,
1468 * it's on the correct interface or we might scrub
1469 * a route to another ifa which would
1470 * be confusing at best and possibly worse.
1472 if (cmd == RTM_DELETE) {
1474 * It's a delete, so it should already exist..
1475 * If it's a net, mask off the host bits
1476 * (Assuming we have a mask)
1477 * XXX this is kinda inet specific..
1479 if (netmask != NULL) {
1480 rt_maskedcopy(dst, (struct sockaddr *)tempbuf, netmask);
1481 dst = (struct sockaddr *)tempbuf;
1485 * Now go through all the requested tables (fibs) and do the
1486 * requested action. Realistically, this will either be fib 0
1487 * for protocols that don't do multiple tables or all the
1488 * tables for those that do.
1490 for ( fibnum = startfib; fibnum <= endfib; fibnum++) {
1491 if (cmd == RTM_DELETE) {
1492 struct radix_node *rn;
1494 * Look up an rtentry that is in the routing tree and
1495 * contains the correct info.
1497 rnh = rt_tables_get_rnh(fibnum, dst->sa_family);
1499 /* this table doesn't exist but others might */
1501 RADIX_NODE_HEAD_RLOCK(rnh);
1503 if (rn_mpath_capable(rnh)) {
1505 rn = rnh->rnh_matchaddr(dst, rnh);
1511 * for interface route the
1512 * rt->rt_gateway is sockaddr_intf
1513 * for cloning ARP entries, so
1514 * rt_mpath_matchgate must use the
1517 rt = rt_mpath_matchgate(rt,
1525 rn = rnh->rnh_lookup(dst, netmask, rnh);
1526 error = (rn == NULL ||
1527 (rn->rn_flags & RNF_ROOT) ||
1528 RNTORT(rn)->rt_ifa != ifa ||
1529 !sa_equal((struct sockaddr *)rn->rn_key, dst));
1530 RADIX_NODE_HEAD_RUNLOCK(rnh);
1532 /* this is only an error if bad on ALL tables */
1537 * Do the actual request
1539 bzero((caddr_t)&info, sizeof(info));
1541 info.rti_flags = flags |
1542 (ifa->ifa_flags & ~IFA_RTSELF) | RTF_PINNED;
1543 info.rti_info[RTAX_DST] = dst;
1545 * doing this for compatibility reasons
1548 info.rti_info[RTAX_GATEWAY] =
1549 (struct sockaddr *)&null_sdl;
1551 info.rti_info[RTAX_GATEWAY] = ifa->ifa_addr;
1552 info.rti_info[RTAX_NETMASK] = netmask;
1553 error = rtrequest1_fib(cmd, &info, &rt, fibnum);
1555 if ((error == EEXIST) && (cmd == RTM_ADD)) {
1557 * Interface route addition failed.
1558 * Atomically delete current prefix generating
1559 * RTM_DELETE message, and retry adding
1562 rnh = rt_tables_get_rnh(fibnum, dst->sa_family);
1563 RADIX_NODE_HEAD_LOCK(rnh);
1565 /* Delete old prefix */
1566 info.rti_ifa = NULL;
1567 info.rti_flags = RTF_RNH_LOCKED;
1569 error = rtrequest1_fib(RTM_DELETE, &info, NULL, fibnum);
1572 info.rti_flags = flags | RTF_RNH_LOCKED |
1573 (ifa->ifa_flags & ~IFA_RTSELF) | RTF_PINNED;
1574 error = rtrequest1_fib(cmd, &info, &rt, fibnum);
1577 RADIX_NODE_HEAD_UNLOCK(rnh);
1581 if (error == 0 && rt != NULL) {
1583 * notify any listening routing agents of the change
1588 * in case address alias finds the first address
1589 * e.g. ifconfig bge0 192.0.2.246/24
1590 * e.g. ifconfig bge0 192.0.2.247/24
1591 * the address set in the route is 192.0.2.246
1592 * so we need to replace it with 192.0.2.247
1594 if (memcmp(rt->rt_ifa->ifa_addr,
1595 ifa->ifa_addr, ifa->ifa_addr->sa_len)) {
1596 ifa_free(rt->rt_ifa);
1598 rt->rt_ifp = ifa->ifa_ifp;
1603 * doing this for compatibility reasons
1605 if (cmd == RTM_ADD) {
1606 ((struct sockaddr_dl *)rt->rt_gateway)->sdl_type =
1607 rt->rt_ifp->if_type;
1608 ((struct sockaddr_dl *)rt->rt_gateway)->sdl_index =
1609 rt->rt_ifp->if_index;
1613 rt_newaddrmsg_fib(cmd, ifa, error, rt, fibnum);
1616 if (cmd == RTM_DELETE) {
1618 * If we are deleting, and we found an entry,
1619 * then it's been removed from the tree..
1620 * now throw it away.
1624 if (cmd == RTM_ADD) {
1626 * We just wanted to add it..
1627 * we don't actually need a reference.
1638 if (cmd == RTM_DELETE) {
1642 /* we only give an error if it wasn't in any table */
1643 error = ((flags & RTF_HOST) ?
1644 EHOSTUNREACH : ENETUNREACH);
1648 /* return an error if any of them failed */
1655 #ifndef BURN_BRIDGES
1656 /* special one for inet internal use. may not use. */
1658 rtinit_fib(struct ifaddr *ifa, int cmd, int flags)
1660 return (rtinit1(ifa, cmd, flags, -1));
1665 * Set up a routing table entry, normally
1669 rtinit(struct ifaddr *ifa, int cmd, int flags)
1671 struct sockaddr *dst;
1672 int fib = RT_DEFAULT_FIB;
1674 if (flags & RTF_HOST) {
1675 dst = ifa->ifa_dstaddr;
1677 dst = ifa->ifa_addr;
1680 switch (dst->sa_family) {
1683 /* We do support multiple FIBs. */
1687 return (rtinit1(ifa, cmd, flags, fib));