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 free(V_rt_tables, M_RTABLE);
267 uma_zdestroy(V_rtzone);
269 VNET_SYSUNINIT(vnet_route_uninit, SI_SUB_PROTO_DOMAIN, SI_ORDER_THIRD,
270 vnet_route_uninit, 0);
273 #ifndef _SYS_SYSPROTO_H_
279 sys_setfib(struct thread *td, struct setfib_args *uap)
281 if (uap->fibnum < 0 || uap->fibnum >= rt_numfibs)
283 td->td_proc->p_fibnum = uap->fibnum;
288 * Packet routing routines.
291 rtalloc(struct route *ro)
294 rtalloc_ign_fib(ro, 0UL, RT_DEFAULT_FIB);
298 rtalloc_fib(struct route *ro, u_int fibnum)
300 rtalloc_ign_fib(ro, 0UL, fibnum);
304 rtalloc_ign(struct route *ro, u_long ignore)
308 if ((rt = ro->ro_rt) != NULL) {
309 if (rt->rt_ifp != NULL && rt->rt_flags & RTF_UP)
314 ro->ro_rt = rtalloc1_fib(&ro->ro_dst, 1, ignore, RT_DEFAULT_FIB);
316 RT_UNLOCK(ro->ro_rt);
320 rtalloc_ign_fib(struct route *ro, u_long ignore, u_int fibnum)
324 if ((rt = ro->ro_rt) != NULL) {
325 if (rt->rt_ifp != NULL && rt->rt_flags & RTF_UP)
330 ro->ro_rt = rtalloc1_fib(&ro->ro_dst, 1, ignore, fibnum);
332 RT_UNLOCK(ro->ro_rt);
336 * Look up the route that matches the address given
337 * Or, at least try.. Create a cloned route if needed.
339 * The returned route, if any, is locked.
342 rtalloc1(struct sockaddr *dst, int report, u_long ignflags)
345 return (rtalloc1_fib(dst, report, ignflags, RT_DEFAULT_FIB));
349 rtalloc1_fib(struct sockaddr *dst, int report, u_long ignflags,
352 struct radix_node_head *rnh;
353 struct radix_node *rn;
354 struct rtentry *newrt;
355 struct rt_addrinfo info;
356 int err = 0, msgtype = RTM_MISS;
359 KASSERT((fibnum < rt_numfibs), ("rtalloc1_fib: bad fibnum"));
360 switch (dst->sa_family) {
363 /* We support multiple FIBs. */
366 fibnum = RT_DEFAULT_FIB;
369 rnh = rt_tables_get_rnh(fibnum, dst->sa_family);
375 * Look up the address in the table for that Address Family
377 needlock = !(ignflags & RTF_RNH_LOCKED);
379 RADIX_NODE_HEAD_RLOCK(rnh);
382 RADIX_NODE_HEAD_LOCK_ASSERT(rnh);
384 rn = rnh->rnh_matchaddr(dst, rnh);
385 if (rn && ((rn->rn_flags & RNF_ROOT) == 0)) {
390 RADIX_NODE_HEAD_RUNLOCK(rnh);
394 RADIX_NODE_HEAD_RUNLOCK(rnh);
397 * Either we hit the root or couldn't find any match,
398 * Which basically means
399 * "caint get there frm here"
402 V_rtstat.rts_unreach++;
406 * If required, report the failure to the supervising
408 * For a delete, this is not an error. (report == 0)
410 bzero(&info, sizeof(info));
411 info.rti_info[RTAX_DST] = dst;
412 rt_missmsg_fib(msgtype, &info, 0, err, fibnum);
416 RT_LOCK_ASSERT(newrt);
421 * Remove a reference count from an rtentry.
422 * If the count gets low enough, take it out of the routing table
425 rtfree(struct rtentry *rt)
427 struct radix_node_head *rnh;
429 KASSERT(rt != NULL,("%s: NULL rt", __func__));
430 rnh = rt_tables_get_rnh(rt->rt_fibnum, rt_key(rt)->sa_family);
431 KASSERT(rnh != NULL,("%s: NULL rnh", __func__));
436 * The callers should use RTFREE_LOCKED() or RTFREE(), so
437 * we should come here exactly with the last reference.
440 if (rt->rt_refcnt > 0) {
441 log(LOG_DEBUG, "%s: %p has %d refs\n", __func__, rt, rt->rt_refcnt);
446 * On last reference give the "close method" a chance
447 * to cleanup private state. This also permits (for
448 * IPv4 and IPv6) a chance to decide if the routing table
449 * entry should be purged immediately or at a later time.
450 * When an immediate purge is to happen the close routine
451 * typically calls rtexpunge which clears the RTF_UP flag
452 * on the entry so that the code below reclaims the storage.
454 if (rt->rt_refcnt == 0 && rnh->rnh_close)
455 rnh->rnh_close((struct radix_node *)rt, rnh);
458 * If we are no longer "up" (and ref == 0)
459 * then we can free the resources associated
462 if ((rt->rt_flags & RTF_UP) == 0) {
463 if (rt->rt_nodes->rn_flags & (RNF_ACTIVE | RNF_ROOT))
466 * the rtentry must have been removed from the routing table
467 * so it is represented in rttrash.. remove that now.
471 if (rt->rt_refcnt < 0) {
472 printf("rtfree: %p not freed (neg refs)\n", rt);
477 * release references on items we hold them on..
478 * e.g other routes and ifaddrs.
481 ifa_free(rt->rt_ifa);
483 * The key is separatly alloc'd so free it (see rt_setgate()).
484 * This also frees the gateway, as they are always malloc'd
490 * and the rtentry itself of course
493 uma_zfree(V_rtzone, rt);
502 * Force a routing table entry to the specified
503 * destination to go through the given gateway.
504 * Normally called as a result of a routing redirect
505 * message from the network layer.
508 rtredirect(struct sockaddr *dst,
509 struct sockaddr *gateway,
510 struct sockaddr *netmask,
512 struct sockaddr *src)
515 rtredirect_fib(dst, gateway, netmask, flags, src, RT_DEFAULT_FIB);
519 rtredirect_fib(struct sockaddr *dst,
520 struct sockaddr *gateway,
521 struct sockaddr *netmask,
523 struct sockaddr *src,
526 struct rtentry *rt, *rt0 = NULL;
529 struct rt_addrinfo info;
531 struct radix_node_head *rnh;
534 rnh = rt_tables_get_rnh(fibnum, dst->sa_family);
536 error = EAFNOSUPPORT;
540 /* verify the gateway is directly reachable */
541 if ((ifa = ifa_ifwithnet(gateway, 0)) == NULL) {
545 rt = rtalloc1_fib(dst, 0, 0UL, fibnum); /* NB: rt is locked */
547 * If the redirect isn't from our current router for this dst,
548 * it's either old or wrong. If it redirects us to ourselves,
549 * we have a routing loop, perhaps as a result of an interface
550 * going down recently.
552 if (!(flags & RTF_DONE) && rt &&
553 (!sa_equal(src, rt->rt_gateway) || rt->rt_ifa != ifa))
555 else if (ifa_ifwithaddr_check(gateway))
556 error = EHOSTUNREACH;
560 * Create a new entry if we just got back a wildcard entry
561 * or the lookup failed. This is necessary for hosts
562 * which use routing redirects generated by smart gateways
563 * to dynamically build the routing tables.
565 if (rt == NULL || (rt_mask(rt) && rt_mask(rt)->sa_len < 2))
568 * Don't listen to the redirect if it's
569 * for a route to an interface.
571 if (rt->rt_flags & RTF_GATEWAY) {
572 if (((rt->rt_flags & RTF_HOST) == 0) && (flags & RTF_HOST)) {
574 * Changing from route to net => route to host.
575 * Create new route, rather than smashing route to net.
581 flags |= RTF_GATEWAY | RTF_DYNAMIC;
582 bzero((caddr_t)&info, sizeof(info));
583 info.rti_info[RTAX_DST] = dst;
584 info.rti_info[RTAX_GATEWAY] = gateway;
585 info.rti_info[RTAX_NETMASK] = netmask;
587 info.rti_flags = flags;
589 RT_UNLOCK(rt0); /* drop lock to avoid LOR with RNH */
590 error = rtrequest1_fib(RTM_ADD, &info, &rt, fibnum);
594 EVENTHANDLER_INVOKE(route_redirect_event, rt0, rt, dst);
595 flags = rt->rt_flags;
600 stat = &V_rtstat.rts_dynamic;
602 struct rtentry *gwrt;
605 * Smash the current notion of the gateway to
606 * this destination. Should check about netmask!!!
608 rt->rt_flags |= RTF_MODIFIED;
609 flags |= RTF_MODIFIED;
610 stat = &V_rtstat.rts_newgateway;
612 * add the key and gateway (in one malloc'd chunk).
615 RADIX_NODE_HEAD_LOCK(rnh);
617 rt_setgate(rt, rt_key(rt), gateway);
618 gwrt = rtalloc1(gateway, 1, RTF_RNH_LOCKED);
619 RADIX_NODE_HEAD_UNLOCK(rnh);
620 EVENTHANDLER_INVOKE(route_redirect_event, rt, gwrt, dst);
624 error = EHOSTUNREACH;
630 V_rtstat.rts_badredirect++;
631 else if (stat != NULL)
633 bzero((caddr_t)&info, sizeof(info));
634 info.rti_info[RTAX_DST] = dst;
635 info.rti_info[RTAX_GATEWAY] = gateway;
636 info.rti_info[RTAX_NETMASK] = netmask;
637 info.rti_info[RTAX_AUTHOR] = src;
638 rt_missmsg_fib(RTM_REDIRECT, &info, flags, error, fibnum);
644 rtioctl(u_long req, caddr_t data)
647 return (rtioctl_fib(req, data, RT_DEFAULT_FIB));
651 * Routing table ioctl interface.
654 rtioctl_fib(u_long req, caddr_t data, u_int fibnum)
658 * If more ioctl commands are added here, make sure the proper
659 * super-user checks are being performed because it is possible for
660 * prison-root to make it this far if raw sockets have been enabled
664 /* Multicast goop, grrr... */
665 return mrt_ioctl ? mrt_ioctl(req, data, fibnum) : EOPNOTSUPP;
672 * For both ifa_ifwithroute() routines, 'ifa' is returned referenced.
675 ifa_ifwithroute(int flags, struct sockaddr *dst, struct sockaddr *gateway)
678 return (ifa_ifwithroute_fib(flags, dst, gateway, RT_DEFAULT_FIB));
682 ifa_ifwithroute_fib(int flags, struct sockaddr *dst, struct sockaddr *gateway,
685 register struct ifaddr *ifa;
688 if ((flags & RTF_GATEWAY) == 0) {
690 * If we are adding a route to an interface,
691 * and the interface is a pt to pt link
692 * we should search for the destination
693 * as our clue to the interface. Otherwise
694 * we can use the local address.
697 if (flags & RTF_HOST)
698 ifa = ifa_ifwithdstaddr(dst);
700 ifa = ifa_ifwithaddr(gateway);
703 * If we are adding a route to a remote net
704 * or host, the gateway may still be on the
705 * other end of a pt to pt link.
707 ifa = ifa_ifwithdstaddr(gateway);
710 ifa = ifa_ifwithnet(gateway, 0);
712 struct rtentry *rt = rtalloc1_fib(gateway, 0, RTF_RNH_LOCKED, fibnum);
716 * dismiss a gateway that is reachable only
717 * through the default router
719 switch (gateway->sa_family) {
721 if (satosin(rt_key(rt))->sin_addr.s_addr == INADDR_ANY)
725 if (IN6_IS_ADDR_UNSPECIFIED(&satosin6(rt_key(rt))->sin6_addr))
731 if (!not_found && rt->rt_ifa != NULL) {
737 if (not_found || ifa == NULL)
740 if (ifa->ifa_addr->sa_family != dst->sa_family) {
741 struct ifaddr *oifa = ifa;
742 ifa = ifaof_ifpforaddr(dst, ifa->ifa_ifp);
752 * Do appropriate manipulations of a routing tree given
753 * all the bits of info needed
757 struct sockaddr *dst,
758 struct sockaddr *gateway,
759 struct sockaddr *netmask,
761 struct rtentry **ret_nrt)
764 return (rtrequest_fib(req, dst, gateway, netmask, flags, ret_nrt,
769 rtrequest_fib(int req,
770 struct sockaddr *dst,
771 struct sockaddr *gateway,
772 struct sockaddr *netmask,
774 struct rtentry **ret_nrt,
777 struct rt_addrinfo info;
779 if (dst->sa_len == 0)
782 bzero((caddr_t)&info, sizeof(info));
783 info.rti_flags = flags;
784 info.rti_info[RTAX_DST] = dst;
785 info.rti_info[RTAX_GATEWAY] = gateway;
786 info.rti_info[RTAX_NETMASK] = netmask;
787 return rtrequest1_fib(req, &info, ret_nrt, fibnum);
791 * These (questionable) definitions of apparent local variables apply
792 * to the next two functions. XXXXXX!!!
794 #define dst info->rti_info[RTAX_DST]
795 #define gateway info->rti_info[RTAX_GATEWAY]
796 #define netmask info->rti_info[RTAX_NETMASK]
797 #define ifaaddr info->rti_info[RTAX_IFA]
798 #define ifpaddr info->rti_info[RTAX_IFP]
799 #define flags info->rti_flags
802 rt_getifa(struct rt_addrinfo *info)
805 return (rt_getifa_fib(info, RT_DEFAULT_FIB));
809 * Look up rt_addrinfo for a specific fib. Note that if rti_ifa is defined,
810 * it will be referenced so the caller must free it.
813 rt_getifa_fib(struct rt_addrinfo *info, u_int fibnum)
819 * ifp may be specified by sockaddr_dl
820 * when protocol address is ambiguous.
822 if (info->rti_ifp == NULL && ifpaddr != NULL &&
823 ifpaddr->sa_family == AF_LINK &&
824 (ifa = ifa_ifwithnet(ifpaddr, 0)) != NULL) {
825 info->rti_ifp = ifa->ifa_ifp;
828 if (info->rti_ifa == NULL && ifaaddr != NULL)
829 info->rti_ifa = ifa_ifwithaddr(ifaaddr);
830 if (info->rti_ifa == NULL) {
833 sa = ifaaddr != NULL ? ifaaddr :
834 (gateway != NULL ? gateway : dst);
835 if (sa != NULL && info->rti_ifp != NULL)
836 info->rti_ifa = ifaof_ifpforaddr(sa, info->rti_ifp);
837 else if (dst != NULL && gateway != NULL)
838 info->rti_ifa = ifa_ifwithroute_fib(flags, dst, gateway,
841 info->rti_ifa = ifa_ifwithroute_fib(flags, sa, sa,
844 if ((ifa = info->rti_ifa) != NULL) {
845 if (info->rti_ifp == NULL)
846 info->rti_ifp = ifa->ifa_ifp;
853 * Expunges references to a route that's about to be reclaimed.
854 * The route must be locked.
857 rtexpunge(struct rtentry *rt)
859 #if !defined(RADIX_MPATH)
860 struct radix_node *rn;
862 struct rt_addrinfo info;
866 struct radix_node_head *rnh;
871 * Find the correct routing tree to use for this Address Family
873 rnh = rt_tables_get_rnh(rt->rt_fibnum, rt_key(rt)->sa_family);
876 return (EAFNOSUPPORT);
877 RADIX_NODE_HEAD_LOCK_ASSERT(rnh);
881 bzero(&info, sizeof(info));
882 info.rti_ifp = rt->rt_ifp;
883 info.rti_flags = RTF_RNH_LOCKED;
884 info.rti_info[RTAX_DST] = rt_key(rt);
885 info.rti_info[RTAX_GATEWAY] = rt->rt_ifa->ifa_addr;
888 error = rtrequest1_fib(RTM_DELETE, &info, &rt0, fib);
890 if (error == 0 && rt0 != NULL) {
893 } else if (error != 0) {
899 * Remove the item from the tree; it should be there,
900 * but when callers invoke us blindly it may not (sigh).
902 rn = rnh->rnh_deladdr(rt_key(rt), rt_mask(rt), rnh);
907 KASSERT((rn->rn_flags & (RNF_ACTIVE | RNF_ROOT)) == 0,
908 ("unexpected flags 0x%x", rn->rn_flags));
909 KASSERT(rt == RNTORT(rn),
910 ("lookup mismatch, rt %p rn %p", rt, rn));
911 #endif /* RADIX_MPATH */
913 rt->rt_flags &= ~RTF_UP;
916 * Give the protocol a chance to keep things in sync.
918 if ((ifa = rt->rt_ifa) && ifa->ifa_rtrequest) {
919 struct rt_addrinfo info;
921 bzero((caddr_t)&info, sizeof(info));
922 info.rti_flags = rt->rt_flags;
923 info.rti_info[RTAX_DST] = rt_key(rt);
924 info.rti_info[RTAX_GATEWAY] = rt->rt_gateway;
925 info.rti_info[RTAX_NETMASK] = rt_mask(rt);
926 ifa->ifa_rtrequest(RTM_DELETE, rt, &info);
930 * one more rtentry floating around that is not
931 * linked to the routing table.
934 #if !defined(RADIX_MPATH)
942 rn_mpath_update(int req, struct rt_addrinfo *info,
943 struct radix_node_head *rnh, struct rtentry **ret_nrt)
946 * if we got multipath routes, we require users to specify
947 * a matching RTAX_GATEWAY.
949 struct rtentry *rt, *rto = NULL;
950 register struct radix_node *rn;
953 rn = rnh->rnh_matchaddr(dst, rnh);
956 rto = rt = RNTORT(rn);
957 rt = rt_mpath_matchgate(rt, gateway);
961 * this is the first entry in the chain
964 rn = rn_mpath_next((struct radix_node *)rt);
966 * there is another entry, now it's active
971 rto->rt_flags |= RTF_UP;
973 } else if (rt->rt_flags & RTF_GATEWAY) {
975 * For gateway routes, we need to
976 * make sure that we we are deleting
977 * the correct gateway.
978 * rt_mpath_matchgate() does not
979 * check the case when there is only
980 * one route in the chain.
983 (rt->rt_gateway->sa_len != gateway->sa_len ||
984 memcmp(rt->rt_gateway, gateway, gateway->sa_len)))
988 * remove from tree before returning it
991 rn = rnh->rnh_deladdr(dst, netmask, rnh);
992 KASSERT(rt == RNTORT(rn), ("radix node disappeared"));
998 * use the normal delete code to remove
1001 if (req != RTM_DELETE)
1009 * if the entry is 2nd and on up
1011 if ((req == RTM_DELETE) && !rt_mpath_deldup(rto, rt))
1012 panic ("rtrequest1: rt_mpath_deldup");
1016 if (req == RTM_DELETE) {
1017 rt->rt_flags &= ~RTF_UP;
1019 * One more rtentry floating around that is not
1020 * linked to the routing table. rttrash will be decremented
1021 * when RTFREE(rt) is eventually called.
1027 if (req != RTM_DELETE)
1028 panic("unrecognized request %d", req);
1032 * If the caller wants it, then it can have it,
1033 * but it's up to it to free the rtentry as we won't be
1047 rtrequest1_fib(int req, struct rt_addrinfo *info, struct rtentry **ret_nrt,
1050 int error = 0, needlock = 0;
1051 register struct rtentry *rt;
1053 register struct rtentry *rt0;
1055 register struct radix_node *rn;
1056 register struct radix_node_head *rnh;
1058 struct sockaddr *ndst;
1059 struct sockaddr_storage mdst;
1060 #define senderr(x) { error = x ; goto bad; }
1062 KASSERT((fibnum < rt_numfibs), ("rtrequest1_fib: bad fibnum"));
1063 switch (dst->sa_family) {
1066 /* We support multiple FIBs. */
1069 fibnum = RT_DEFAULT_FIB;
1074 * Find the correct routing tree to use for this Address Family
1076 rnh = rt_tables_get_rnh(fibnum, dst->sa_family);
1078 return (EAFNOSUPPORT);
1079 needlock = ((flags & RTF_RNH_LOCKED) == 0);
1080 flags &= ~RTF_RNH_LOCKED;
1082 RADIX_NODE_HEAD_LOCK(rnh);
1084 RADIX_NODE_HEAD_LOCK_ASSERT(rnh);
1086 * If we are adding a host route then we don't want to put
1087 * a netmask in the tree, nor do we want to clone it.
1089 if (flags & RTF_HOST)
1095 rt_maskedcopy(dst, (struct sockaddr *)&mdst, netmask);
1096 dst = (struct sockaddr *)&mdst;
1099 if (rn_mpath_capable(rnh)) {
1100 error = rn_mpath_update(req, info, rnh, ret_nrt);
1102 * "bad" holds true for the success case
1105 if (error != ENOENT)
1110 if ((flags & RTF_PINNED) == 0) {
1111 /* Check if target route can be deleted */
1112 rt = (struct rtentry *)rnh->rnh_lookup(dst,
1114 if ((rt != NULL) && (rt->rt_flags & RTF_PINNED))
1115 senderr(EADDRINUSE);
1119 * Remove the item from the tree and return it.
1120 * Complain if it is not there and do no more processing.
1122 rn = rnh->rnh_deladdr(dst, netmask, rnh);
1125 if (rn->rn_flags & (RNF_ACTIVE | RNF_ROOT))
1126 panic ("rtrequest delete");
1130 rt->rt_flags &= ~RTF_UP;
1133 * give the protocol a chance to keep things in sync.
1135 if ((ifa = rt->rt_ifa) && ifa->ifa_rtrequest)
1136 ifa->ifa_rtrequest(RTM_DELETE, rt, info);
1139 * One more rtentry floating around that is not
1140 * linked to the routing table. rttrash will be decremented
1141 * when RTFREE(rt) is eventually called.
1146 * If the caller wants it, then it can have it,
1147 * but it's up to it to free the rtentry as we won't be
1158 * resolve was only used for route cloning
1163 if ((flags & RTF_GATEWAY) && !gateway)
1165 if (dst && gateway && (dst->sa_family != gateway->sa_family) &&
1166 (gateway->sa_family != AF_UNSPEC) && (gateway->sa_family != AF_LINK))
1169 if (info->rti_ifa == NULL) {
1170 error = rt_getifa_fib(info, fibnum);
1174 ifa_ref(info->rti_ifa);
1175 ifa = info->rti_ifa;
1176 rt = uma_zalloc(V_rtzone, M_NOWAIT | M_ZERO);
1182 rt->rt_flags = RTF_UP | flags;
1183 rt->rt_fibnum = fibnum;
1185 * Add the gateway. Possibly re-malloc-ing the storage for it.
1188 if ((error = rt_setgate(rt, dst, gateway)) != 0) {
1189 RT_LOCK_DESTROY(rt);
1191 uma_zfree(V_rtzone, rt);
1196 * point to the (possibly newly malloc'd) dest address.
1198 ndst = (struct sockaddr *)rt_key(rt);
1201 * make sure it contains the value we want (masked if needed).
1204 rt_maskedcopy(dst, ndst, netmask);
1206 bcopy(dst, ndst, dst->sa_len);
1209 * We use the ifa reference returned by rt_getifa_fib().
1210 * This moved from below so that rnh->rnh_addaddr() can
1211 * examine the ifa and ifa->ifa_ifp if it so desires.
1214 rt->rt_ifp = ifa->ifa_ifp;
1215 rt->rt_rmx.rmx_weight = 1;
1218 /* do not permit exactly the same dst/mask/gw pair */
1219 if (rn_mpath_capable(rnh) &&
1220 rt_mpath_conflict(rnh, rt, netmask)) {
1221 ifa_free(rt->rt_ifa);
1223 RT_LOCK_DESTROY(rt);
1224 uma_zfree(V_rtzone, rt);
1231 /* "flow-table" only supports IPv6 and IPv4 at the moment. */
1232 switch (dst->sa_family) {
1239 #if defined(INET6) || defined(INET)
1240 rn = rnh->rnh_matchaddr(dst, rnh);
1241 if (rn && ((rn->rn_flags & RNF_ROOT) == 0)) {
1242 struct sockaddr *mask;
1247 * compare mask to see if the new route is
1248 * more specific than the existing one
1255 * A host route is already present, so
1256 * leave the flow-table entries as is.
1258 if (rt0->rt_flags & RTF_HOST) {
1261 } else if (!(flags & RTF_HOST) && netmask) {
1262 mask = rt_mask(rt0);
1265 n = (u_char *)netmask;
1272 if (len == 0 || (*n < *m)) {
1278 #endif/* INET6 || INET */
1280 #endif /* FLOWTABLE */
1282 /* XXX mtu manipulation will be done in rnh_addaddr -- itojun */
1283 rn = rnh->rnh_addaddr(ndst, netmask, rnh, rt->rt_nodes);
1285 * If it still failed to go into the tree,
1286 * then un-make it (this should be a function)
1289 ifa_free(rt->rt_ifa);
1291 RT_LOCK_DESTROY(rt);
1292 uma_zfree(V_rtzone, rt);
1300 else if (rt0 != NULL) {
1301 switch (dst->sa_family) {
1304 flowtable_route_flush(V_ip6_ft, rt0);
1309 flowtable_route_flush(V_ip_ft, rt0);
1318 * If this protocol has something to add to this then
1319 * allow it to do that as well.
1321 if (ifa->ifa_rtrequest)
1322 ifa->ifa_rtrequest(req, rt, info);
1325 * actually return a resultant rtentry and
1326 * give the caller a single reference.
1339 RADIX_NODE_HEAD_UNLOCK(rnh);
1352 rt_setgate(struct rtentry *rt, struct sockaddr *dst, struct sockaddr *gate)
1354 /* XXX dst may be overwritten, can we move this to below */
1355 int dlen = SA_SIZE(dst), glen = SA_SIZE(gate);
1357 struct radix_node_head *rnh;
1359 rnh = rt_tables_get_rnh(rt->rt_fibnum, dst->sa_family);
1363 RADIX_NODE_HEAD_LOCK_ASSERT(rnh);
1366 * Prepare to store the gateway in rt->rt_gateway.
1367 * Both dst and gateway are stored one after the other in the same
1368 * malloc'd chunk. If we have room, we can reuse the old buffer,
1369 * rt_gateway already points to the right place.
1370 * Otherwise, malloc a new block and update the 'dst' address.
1372 if (rt->rt_gateway == NULL || glen > SA_SIZE(rt->rt_gateway)) {
1375 R_Malloc(new, caddr_t, dlen + glen);
1379 * XXX note, we copy from *dst and not *rt_key(rt) because
1380 * rt_setgate() can be called to initialize a newly
1381 * allocated route entry, in which case rt_key(rt) == NULL
1382 * (and also rt->rt_gateway == NULL).
1383 * Free()/free() handle a NULL argument just fine.
1385 bcopy(dst, new, dlen);
1386 Free(rt_key(rt)); /* free old block, if any */
1387 rt_key(rt) = (struct sockaddr *)new;
1388 rt->rt_gateway = (struct sockaddr *)(new + dlen);
1392 * Copy the new gateway value into the memory chunk.
1394 bcopy(gate, rt->rt_gateway, glen);
1400 rt_maskedcopy(struct sockaddr *src, struct sockaddr *dst, struct sockaddr *netmask)
1402 register u_char *cp1 = (u_char *)src;
1403 register u_char *cp2 = (u_char *)dst;
1404 register u_char *cp3 = (u_char *)netmask;
1405 u_char *cplim = cp2 + *cp3;
1406 u_char *cplim2 = cp2 + *cp1;
1408 *cp2++ = *cp1++; *cp2++ = *cp1++; /* copies sa_len & sa_family */
1413 *cp2++ = *cp1++ & *cp3++;
1415 bzero((caddr_t)cp2, (unsigned)(cplim2 - cp2));
1419 * Set up a routing table entry, normally
1422 #define _SOCKADDR_TMPSIZE 128 /* Not too big.. kernel stack size is limited */
1424 rtinit1(struct ifaddr *ifa, int cmd, int flags, int fibnum)
1426 struct sockaddr *dst;
1427 struct sockaddr *netmask;
1428 struct rtentry *rt = NULL;
1429 struct rt_addrinfo info;
1431 int startfib, endfib;
1432 char tempbuf[_SOCKADDR_TMPSIZE];
1435 static struct sockaddr_dl null_sdl = {sizeof(null_sdl), AF_LINK};
1436 struct radix_node_head *rnh;
1438 if (flags & RTF_HOST) {
1439 dst = ifa->ifa_dstaddr;
1442 dst = ifa->ifa_addr;
1443 netmask = ifa->ifa_netmask;
1445 if (dst->sa_len == 0)
1447 switch (dst->sa_family) {
1450 /* We support multiple FIBs. */
1453 fibnum = RT_DEFAULT_FIB;
1457 if (rt_add_addr_allfibs == 0 && cmd == (int)RTM_ADD) {
1458 startfib = endfib = curthread->td_proc->p_fibnum;
1461 endfib = rt_numfibs - 1;
1464 KASSERT((fibnum < rt_numfibs), ("rtinit1: bad fibnum"));
1470 * If it's a delete, check that if it exists,
1471 * it's on the correct interface or we might scrub
1472 * a route to another ifa which would
1473 * be confusing at best and possibly worse.
1475 if (cmd == RTM_DELETE) {
1477 * It's a delete, so it should already exist..
1478 * If it's a net, mask off the host bits
1479 * (Assuming we have a mask)
1480 * XXX this is kinda inet specific..
1482 if (netmask != NULL) {
1483 rt_maskedcopy(dst, (struct sockaddr *)tempbuf, netmask);
1484 dst = (struct sockaddr *)tempbuf;
1488 * Now go through all the requested tables (fibs) and do the
1489 * requested action. Realistically, this will either be fib 0
1490 * for protocols that don't do multiple tables or all the
1491 * tables for those that do.
1493 for ( fibnum = startfib; fibnum <= endfib; fibnum++) {
1494 if (cmd == RTM_DELETE) {
1495 struct radix_node *rn;
1497 * Look up an rtentry that is in the routing tree and
1498 * contains the correct info.
1500 rnh = rt_tables_get_rnh(fibnum, dst->sa_family);
1502 /* this table doesn't exist but others might */
1504 RADIX_NODE_HEAD_RLOCK(rnh);
1506 if (rn_mpath_capable(rnh)) {
1508 rn = rnh->rnh_matchaddr(dst, rnh);
1514 * for interface route the
1515 * rt->rt_gateway is sockaddr_intf
1516 * for cloning ARP entries, so
1517 * rt_mpath_matchgate must use the
1520 rt = rt_mpath_matchgate(rt,
1528 rn = rnh->rnh_lookup(dst, netmask, rnh);
1529 error = (rn == NULL ||
1530 (rn->rn_flags & RNF_ROOT) ||
1531 RNTORT(rn)->rt_ifa != ifa ||
1532 !sa_equal((struct sockaddr *)rn->rn_key, dst));
1533 RADIX_NODE_HEAD_RUNLOCK(rnh);
1535 /* this is only an error if bad on ALL tables */
1540 * Do the actual request
1542 bzero((caddr_t)&info, sizeof(info));
1544 info.rti_flags = flags |
1545 (ifa->ifa_flags & ~IFA_RTSELF) | RTF_PINNED;
1546 info.rti_info[RTAX_DST] = dst;
1548 * doing this for compatibility reasons
1551 info.rti_info[RTAX_GATEWAY] =
1552 (struct sockaddr *)&null_sdl;
1554 info.rti_info[RTAX_GATEWAY] = ifa->ifa_addr;
1555 info.rti_info[RTAX_NETMASK] = netmask;
1556 error = rtrequest1_fib(cmd, &info, &rt, fibnum);
1558 if ((error == EEXIST) && (cmd == RTM_ADD)) {
1560 * Interface route addition failed.
1561 * Atomically delete current prefix generating
1562 * RTM_DELETE message, and retry adding
1565 rnh = rt_tables_get_rnh(fibnum, dst->sa_family);
1566 RADIX_NODE_HEAD_LOCK(rnh);
1568 /* Delete old prefix */
1569 info.rti_ifa = NULL;
1570 info.rti_flags = RTF_RNH_LOCKED;
1572 error = rtrequest1_fib(RTM_DELETE, &info, NULL, fibnum);
1575 info.rti_flags = flags | RTF_RNH_LOCKED |
1576 (ifa->ifa_flags & ~IFA_RTSELF) | RTF_PINNED;
1577 error = rtrequest1_fib(cmd, &info, &rt, fibnum);
1580 RADIX_NODE_HEAD_UNLOCK(rnh);
1584 if (error == 0 && rt != NULL) {
1586 * notify any listening routing agents of the change
1591 * in case address alias finds the first address
1592 * e.g. ifconfig bge0 192.0.2.246/24
1593 * e.g. ifconfig bge0 192.0.2.247/24
1594 * the address set in the route is 192.0.2.246
1595 * so we need to replace it with 192.0.2.247
1597 if (memcmp(rt->rt_ifa->ifa_addr,
1598 ifa->ifa_addr, ifa->ifa_addr->sa_len)) {
1599 ifa_free(rt->rt_ifa);
1601 rt->rt_ifp = ifa->ifa_ifp;
1606 * doing this for compatibility reasons
1608 if (cmd == RTM_ADD) {
1609 ((struct sockaddr_dl *)rt->rt_gateway)->sdl_type =
1610 rt->rt_ifp->if_type;
1611 ((struct sockaddr_dl *)rt->rt_gateway)->sdl_index =
1612 rt->rt_ifp->if_index;
1616 rt_newaddrmsg_fib(cmd, ifa, error, rt, fibnum);
1619 if (cmd == RTM_DELETE) {
1621 * If we are deleting, and we found an entry,
1622 * then it's been removed from the tree..
1623 * now throw it away.
1627 if (cmd == RTM_ADD) {
1629 * We just wanted to add it..
1630 * we don't actually need a reference.
1641 if (cmd == RTM_DELETE) {
1645 /* we only give an error if it wasn't in any table */
1646 error = ((flags & RTF_HOST) ?
1647 EHOSTUNREACH : ENETUNREACH);
1651 /* return an error if any of them failed */
1658 #ifndef BURN_BRIDGES
1659 /* special one for inet internal use. may not use. */
1661 rtinit_fib(struct ifaddr *ifa, int cmd, int flags)
1663 return (rtinit1(ifa, cmd, flags, -1));
1668 * Set up a routing table entry, normally
1672 rtinit(struct ifaddr *ifa, int cmd, int flags)
1674 struct sockaddr *dst;
1675 int fib = RT_DEFAULT_FIB;
1677 if (flags & RTF_HOST) {
1678 dst = ifa->ifa_dstaddr;
1680 dst = ifa->ifa_addr;
1683 switch (dst->sa_family) {
1686 /* We do support multiple FIBs. */
1690 return (rtinit1(ifa, cmd, flags, fib));