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"
41 #include "opt_mrouting.h"
42 #include "opt_mpath.h"
44 #include <sys/param.h>
45 #include <sys/systm.h>
46 #include <sys/syslog.h>
47 #include <sys/malloc.h>
49 #include <sys/socket.h>
50 #include <sys/sysctl.h>
51 #include <sys/syslog.h>
52 #include <sys/sysproto.h>
54 #include <sys/domain.h>
55 #include <sys/kernel.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 /* We use 4 bits in the mbuf flags, thus we are limited to 16 FIBS. */
75 /* Kernel config default option. */
78 #error "ROUTETABLES defined too low"
80 #if ROUTETABLES > RT_MAXFIBS
81 #error "ROUTETABLES defined too big"
83 #define RT_NUMFIBS ROUTETABLES
84 #endif /* ROUTETABLES */
85 /* Initialize to default if not otherwise set. */
90 #if defined(INET) || defined(INET6)
92 extern void sctp_addr_change(struct ifaddr *ifa, int cmd);
97 /* This is read-only.. */
98 u_int rt_numfibs = RT_NUMFIBS;
99 SYSCTL_UINT(_net, OID_AUTO, fibs, CTLFLAG_RD, &rt_numfibs, 0, "");
101 * Allow the boot code to allow LESS than RT_MAXFIBS to be used.
102 * We can't do more because storage is statically allocated for now.
103 * (for compatibility reasons.. this will change. When this changes, code should
104 * be refactored to protocol independent parts and protocol dependent parts,
105 * probably hanging of domain(9) specific storage to not need the full
106 * fib * af RNH allocation etc. but allow tuning the number of tables per
109 TUNABLE_INT("net.fibs", &rt_numfibs);
112 * By default add routes to all fibs for new interfaces.
113 * Once this is set to 0 then only allocate routes on interface
114 * changes for the FIB of the caller when adding a new set of addresses
115 * to an interface. XXX this is a shotgun aproach to a problem that needs
116 * a more fine grained solution.. that will come.
117 * XXX also has the problems getting the FIB from curthread which will not
118 * always work given the fib can be overridden and prefixes can be added
119 * from the network stack context.
121 u_int rt_add_addr_allfibs = 1;
122 SYSCTL_UINT(_net, OID_AUTO, add_addr_allfibs, CTLFLAG_RW,
123 &rt_add_addr_allfibs, 0, "");
124 TUNABLE_INT("net.add_addr_allfibs", &rt_add_addr_allfibs);
126 VNET_DEFINE(struct rtstat, rtstat);
127 #define V_rtstat VNET(rtstat)
129 VNET_DEFINE(struct radix_node_head *, rt_tables);
130 #define V_rt_tables VNET(rt_tables)
132 VNET_DEFINE(int, rttrash); /* routes not in table but not freed */
133 #define V_rttrash VNET(rttrash)
136 /* compare two sockaddr structures */
137 #define sa_equal(a1, a2) (((a1)->sa_len == (a2)->sa_len) && \
138 (bcmp((a1), (a2), (a1)->sa_len) == 0))
141 * Convert a 'struct radix_node *' to a 'struct rtentry *'.
142 * The operation can be done safely (in this code) because a
143 * 'struct rtentry' starts with two 'struct radix_node''s, the first
144 * one representing leaf nodes in the routing tree, which is
145 * what the code in radix.c passes us as a 'struct radix_node'.
147 * But because there are a lot of assumptions in this conversion,
148 * do not cast explicitly, but always use the macro below.
150 #define RNTORT(p) ((struct rtentry *)(p))
152 static VNET_DEFINE(uma_zone_t, rtzone); /* Routing table UMA zone. */
153 #define V_rtzone VNET(rtzone)
156 * handler for net.my_fibnum
159 sysctl_my_fibnum(SYSCTL_HANDLER_ARGS)
164 fibnum = curthread->td_proc->p_fibnum;
165 error = sysctl_handle_int(oidp, &fibnum, 0, req);
169 SYSCTL_PROC(_net, OID_AUTO, my_fibnum, CTLTYPE_INT|CTLFLAG_RD,
170 NULL, 0, &sysctl_my_fibnum, "I", "default FIB of caller");
172 static __inline struct radix_node_head **
173 rt_tables_get_rnh_ptr(int table, int fam)
175 struct radix_node_head **rnh;
177 KASSERT(table >= 0 && table < rt_numfibs, ("%s: table out of bounds.",
179 KASSERT(fam >= 0 && fam < (AF_MAX+1), ("%s: fam out of bounds.",
182 /* rnh is [fib=0][af=0]. */
183 rnh = (struct radix_node_head **)V_rt_tables;
184 /* Get the offset to the requested table and fam. */
185 rnh += table * (AF_MAX+1) + fam;
190 struct radix_node_head *
191 rt_tables_get_rnh(int table, int fam)
194 return (*rt_tables_get_rnh_ptr(table, fam));
198 * route initialization must occur before ip6_init2(), which happenas at
207 /* whack the tunable ints into line. */
208 if (rt_numfibs > RT_MAXFIBS)
209 rt_numfibs = RT_MAXFIBS;
213 for (dom = domains; dom; dom = dom->dom_next)
214 if (dom->dom_maxrtkey > max_keylen)
215 max_keylen = dom->dom_maxrtkey;
217 rn_init(max_keylen); /* init all zeroes, all ones, mask table */
219 SYSINIT(route_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_THIRD, route_init, 0);
222 vnet_route_init(const void *unused __unused)
225 struct radix_node_head **rnh;
229 V_rt_tables = malloc(rt_numfibs * (AF_MAX+1) *
230 sizeof(struct radix_node_head *), M_RTABLE, M_WAITOK|M_ZERO);
232 V_rtzone = uma_zcreate("rtentry", sizeof(struct rtentry), NULL, NULL,
233 NULL, NULL, UMA_ALIGN_PTR, 0);
234 for (dom = domains; dom; dom = dom->dom_next) {
235 if (dom->dom_rtattach == NULL)
238 for (table = 0; table < rt_numfibs; table++) {
239 fam = dom->dom_family;
240 if (table != 0 && fam != AF_INET6 && fam != AF_INET)
244 * XXX MRT rtattach will be also called from
245 * vfs_export.c but the offset will be 0 (only for
246 * AF_INET and AF_INET6 which don't need it anyhow).
248 rnh = rt_tables_get_rnh_ptr(table, fam);
250 panic("%s: rnh NULL", __func__);
251 dom->dom_rtattach((void **)rnh, dom->dom_rtoffset);
255 VNET_SYSINIT(vnet_route_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_FOURTH,
260 vnet_route_uninit(const void *unused __unused)
265 struct radix_node_head **rnh;
267 for (dom = domains; dom; dom = dom->dom_next) {
268 if (dom->dom_rtdetach == NULL)
271 for (table = 0; table < rt_numfibs; table++) {
272 fam = dom->dom_family;
274 if (table != 0 && fam != AF_INET6 && fam != AF_INET)
277 rnh = rt_tables_get_rnh_ptr(table, fam);
279 panic("%s: rnh NULL", __func__);
280 dom->dom_rtdetach((void **)rnh, dom->dom_rtoffset);
284 VNET_SYSUNINIT(vnet_route_uninit, SI_SUB_PROTO_DOMAIN, SI_ORDER_THIRD,
285 vnet_route_uninit, 0);
288 #ifndef _SYS_SYSPROTO_H_
294 sys_setfib(struct thread *td, struct setfib_args *uap)
296 if (uap->fibnum < 0 || uap->fibnum >= rt_numfibs)
298 td->td_proc->p_fibnum = uap->fibnum;
303 * Packet routing routines.
306 rtalloc(struct route *ro)
309 rtalloc_ign_fib(ro, 0UL, RT_DEFAULT_FIB);
313 rtalloc_fib(struct route *ro, u_int fibnum)
315 rtalloc_ign_fib(ro, 0UL, fibnum);
319 rtalloc_ign(struct route *ro, u_long ignore)
323 if ((rt = ro->ro_rt) != NULL) {
324 if (rt->rt_ifp != NULL && rt->rt_flags & RTF_UP)
329 ro->ro_rt = rtalloc1_fib(&ro->ro_dst, 1, ignore, RT_DEFAULT_FIB);
331 RT_UNLOCK(ro->ro_rt);
335 rtalloc_ign_fib(struct route *ro, u_long ignore, u_int fibnum)
339 if ((rt = ro->ro_rt) != NULL) {
340 if (rt->rt_ifp != NULL && rt->rt_flags & RTF_UP)
345 ro->ro_rt = rtalloc1_fib(&ro->ro_dst, 1, ignore, fibnum);
347 RT_UNLOCK(ro->ro_rt);
351 * Look up the route that matches the address given
352 * Or, at least try.. Create a cloned route if needed.
354 * The returned route, if any, is locked.
357 rtalloc1(struct sockaddr *dst, int report, u_long ignflags)
360 return (rtalloc1_fib(dst, report, ignflags, RT_DEFAULT_FIB));
364 rtalloc1_fib(struct sockaddr *dst, int report, u_long ignflags,
367 struct radix_node_head *rnh;
368 struct radix_node *rn;
369 struct rtentry *newrt;
370 struct rt_addrinfo info;
371 int err = 0, msgtype = RTM_MISS;
374 KASSERT((fibnum < rt_numfibs), ("rtalloc1_fib: bad fibnum"));
375 switch (dst->sa_family) {
378 /* We support multiple FIBs. */
381 fibnum = RT_DEFAULT_FIB;
384 rnh = rt_tables_get_rnh(fibnum, dst->sa_family);
390 * Look up the address in the table for that Address Family
392 needlock = !(ignflags & RTF_RNH_LOCKED);
394 RADIX_NODE_HEAD_RLOCK(rnh);
397 RADIX_NODE_HEAD_LOCK_ASSERT(rnh);
399 rn = rnh->rnh_matchaddr(dst, rnh);
400 if (rn && ((rn->rn_flags & RNF_ROOT) == 0)) {
405 RADIX_NODE_HEAD_RUNLOCK(rnh);
409 RADIX_NODE_HEAD_RUNLOCK(rnh);
412 * Either we hit the root or couldn't find any match,
413 * Which basically means
414 * "caint get there frm here"
417 V_rtstat.rts_unreach++;
421 * If required, report the failure to the supervising
423 * For a delete, this is not an error. (report == 0)
425 bzero(&info, sizeof(info));
426 info.rti_info[RTAX_DST] = dst;
427 rt_missmsg_fib(msgtype, &info, 0, err, fibnum);
431 RT_LOCK_ASSERT(newrt);
436 * Remove a reference count from an rtentry.
437 * If the count gets low enough, take it out of the routing table
440 rtfree(struct rtentry *rt)
442 struct radix_node_head *rnh;
444 KASSERT(rt != NULL,("%s: NULL rt", __func__));
445 rnh = rt_tables_get_rnh(rt->rt_fibnum, rt_key(rt)->sa_family);
446 KASSERT(rnh != NULL,("%s: NULL rnh", __func__));
451 * The callers should use RTFREE_LOCKED() or RTFREE(), so
452 * we should come here exactly with the last reference.
455 if (rt->rt_refcnt > 0) {
456 log(LOG_DEBUG, "%s: %p has %d refs\n", __func__, rt, rt->rt_refcnt);
461 * On last reference give the "close method" a chance
462 * to cleanup private state. This also permits (for
463 * IPv4 and IPv6) a chance to decide if the routing table
464 * entry should be purged immediately or at a later time.
465 * When an immediate purge is to happen the close routine
466 * typically calls rtexpunge which clears the RTF_UP flag
467 * on the entry so that the code below reclaims the storage.
469 if (rt->rt_refcnt == 0 && rnh->rnh_close)
470 rnh->rnh_close((struct radix_node *)rt, rnh);
473 * If we are no longer "up" (and ref == 0)
474 * then we can free the resources associated
477 if ((rt->rt_flags & RTF_UP) == 0) {
478 if (rt->rt_nodes->rn_flags & (RNF_ACTIVE | RNF_ROOT))
481 * the rtentry must have been removed from the routing table
482 * so it is represented in rttrash.. remove that now.
486 if (rt->rt_refcnt < 0) {
487 printf("rtfree: %p not freed (neg refs)\n", rt);
492 * release references on items we hold them on..
493 * e.g other routes and ifaddrs.
496 ifa_free(rt->rt_ifa);
498 * The key is separatly alloc'd so free it (see rt_setgate()).
499 * This also frees the gateway, as they are always malloc'd
505 * and the rtentry itself of course
508 uma_zfree(V_rtzone, rt);
517 * Force a routing table entry to the specified
518 * destination to go through the given gateway.
519 * Normally called as a result of a routing redirect
520 * message from the network layer.
523 rtredirect(struct sockaddr *dst,
524 struct sockaddr *gateway,
525 struct sockaddr *netmask,
527 struct sockaddr *src)
530 rtredirect_fib(dst, gateway, netmask, flags, src, RT_DEFAULT_FIB);
534 rtredirect_fib(struct sockaddr *dst,
535 struct sockaddr *gateway,
536 struct sockaddr *netmask,
538 struct sockaddr *src,
541 struct rtentry *rt, *rt0 = NULL;
544 struct rt_addrinfo info;
546 struct radix_node_head *rnh;
549 rnh = rt_tables_get_rnh(fibnum, dst->sa_family);
551 error = EAFNOSUPPORT;
555 /* verify the gateway is directly reachable */
556 if ((ifa = ifa_ifwithnet_fib(gateway, 0, fibnum)) == NULL) {
560 rt = rtalloc1_fib(dst, 0, 0UL, fibnum); /* NB: rt is locked */
562 * If the redirect isn't from our current router for this dst,
563 * it's either old or wrong. If it redirects us to ourselves,
564 * we have a routing loop, perhaps as a result of an interface
565 * going down recently.
567 if (!(flags & RTF_DONE) && rt &&
568 (!sa_equal(src, rt->rt_gateway) || rt->rt_ifa != ifa))
570 else if (ifa_ifwithaddr_check(gateway))
571 error = EHOSTUNREACH;
575 * Create a new entry if we just got back a wildcard entry
576 * or the lookup failed. This is necessary for hosts
577 * which use routing redirects generated by smart gateways
578 * to dynamically build the routing tables.
580 if (rt == NULL || (rt_mask(rt) && rt_mask(rt)->sa_len < 2))
583 * Don't listen to the redirect if it's
584 * for a route to an interface.
586 if (rt->rt_flags & RTF_GATEWAY) {
587 if (((rt->rt_flags & RTF_HOST) == 0) && (flags & RTF_HOST)) {
589 * Changing from route to net => route to host.
590 * Create new route, rather than smashing route to net.
596 flags |= RTF_GATEWAY | RTF_DYNAMIC;
597 bzero((caddr_t)&info, sizeof(info));
598 info.rti_info[RTAX_DST] = dst;
599 info.rti_info[RTAX_GATEWAY] = gateway;
600 info.rti_info[RTAX_NETMASK] = netmask;
602 info.rti_flags = flags;
604 RT_UNLOCK(rt0); /* drop lock to avoid LOR with RNH */
605 error = rtrequest1_fib(RTM_ADD, &info, &rt, fibnum);
609 EVENTHANDLER_INVOKE(route_redirect_event, rt0, rt, dst);
610 flags = rt->rt_flags;
615 stat = &V_rtstat.rts_dynamic;
617 struct rtentry *gwrt;
620 * Smash the current notion of the gateway to
621 * this destination. Should check about netmask!!!
623 rt->rt_flags |= RTF_MODIFIED;
624 flags |= RTF_MODIFIED;
625 stat = &V_rtstat.rts_newgateway;
627 * add the key and gateway (in one malloc'd chunk).
630 RADIX_NODE_HEAD_LOCK(rnh);
632 rt_setgate(rt, rt_key(rt), gateway);
633 gwrt = rtalloc1(gateway, 1, RTF_RNH_LOCKED);
634 RADIX_NODE_HEAD_UNLOCK(rnh);
635 EVENTHANDLER_INVOKE(route_redirect_event, rt, gwrt, dst);
639 error = EHOSTUNREACH;
645 V_rtstat.rts_badredirect++;
646 else if (stat != NULL)
648 bzero((caddr_t)&info, sizeof(info));
649 info.rti_info[RTAX_DST] = dst;
650 info.rti_info[RTAX_GATEWAY] = gateway;
651 info.rti_info[RTAX_NETMASK] = netmask;
652 info.rti_info[RTAX_AUTHOR] = src;
653 rt_missmsg_fib(RTM_REDIRECT, &info, flags, error, fibnum);
659 rtioctl(u_long req, caddr_t data)
662 return (rtioctl_fib(req, data, RT_DEFAULT_FIB));
666 * Routing table ioctl interface.
669 rtioctl_fib(u_long req, caddr_t data, u_int fibnum)
673 * If more ioctl commands are added here, make sure the proper
674 * super-user checks are being performed because it is possible for
675 * prison-root to make it this far if raw sockets have been enabled
679 /* Multicast goop, grrr... */
680 return mrt_ioctl ? mrt_ioctl(req, data, fibnum) : EOPNOTSUPP;
687 * For both ifa_ifwithroute() routines, 'ifa' is returned referenced.
690 ifa_ifwithroute(int flags, struct sockaddr *dst, struct sockaddr *gateway)
693 return (ifa_ifwithroute_fib(flags, dst, gateway, RT_DEFAULT_FIB));
697 ifa_ifwithroute_fib(int flags, struct sockaddr *dst, struct sockaddr *gateway,
700 register struct ifaddr *ifa;
703 if ((flags & RTF_GATEWAY) == 0) {
705 * If we are adding a route to an interface,
706 * and the interface is a pt to pt link
707 * we should search for the destination
708 * as our clue to the interface. Otherwise
709 * we can use the local address.
712 if (flags & RTF_HOST)
713 ifa = ifa_ifwithdstaddr_fib(dst, fibnum);
715 ifa = ifa_ifwithaddr(gateway);
718 * If we are adding a route to a remote net
719 * or host, the gateway may still be on the
720 * other end of a pt to pt link.
722 ifa = ifa_ifwithdstaddr_fib(gateway, fibnum);
725 ifa = ifa_ifwithnet_fib(gateway, 0, fibnum);
727 struct rtentry *rt = rtalloc1_fib(gateway, 0, RTF_RNH_LOCKED, fibnum);
731 * dismiss a gateway that is reachable only
732 * through the default router
734 switch (gateway->sa_family) {
736 if (satosin(rt_key(rt))->sin_addr.s_addr == INADDR_ANY)
740 if (IN6_IS_ADDR_UNSPECIFIED(&satosin6(rt_key(rt))->sin6_addr))
746 if (!not_found && rt->rt_ifa != NULL) {
752 if (not_found || ifa == NULL)
755 if (ifa->ifa_addr->sa_family != dst->sa_family) {
756 struct ifaddr *oifa = ifa;
757 ifa = ifaof_ifpforaddr(dst, ifa->ifa_ifp);
767 * Do appropriate manipulations of a routing tree given
768 * all the bits of info needed
772 struct sockaddr *dst,
773 struct sockaddr *gateway,
774 struct sockaddr *netmask,
776 struct rtentry **ret_nrt)
779 return (rtrequest_fib(req, dst, gateway, netmask, flags, ret_nrt,
784 rtrequest_fib(int req,
785 struct sockaddr *dst,
786 struct sockaddr *gateway,
787 struct sockaddr *netmask,
789 struct rtentry **ret_nrt,
792 struct rt_addrinfo info;
794 if (dst->sa_len == 0)
797 bzero((caddr_t)&info, sizeof(info));
798 info.rti_flags = flags;
799 info.rti_info[RTAX_DST] = dst;
800 info.rti_info[RTAX_GATEWAY] = gateway;
801 info.rti_info[RTAX_NETMASK] = netmask;
802 return rtrequest1_fib(req, &info, ret_nrt, fibnum);
806 * These (questionable) definitions of apparent local variables apply
807 * to the next two functions. XXXXXX!!!
809 #define dst info->rti_info[RTAX_DST]
810 #define gateway info->rti_info[RTAX_GATEWAY]
811 #define netmask info->rti_info[RTAX_NETMASK]
812 #define ifaaddr info->rti_info[RTAX_IFA]
813 #define ifpaddr info->rti_info[RTAX_IFP]
814 #define flags info->rti_flags
817 rt_getifa(struct rt_addrinfo *info)
820 return (rt_getifa_fib(info, RT_DEFAULT_FIB));
824 * Look up rt_addrinfo for a specific fib. Note that if rti_ifa is defined,
825 * it will be referenced so the caller must free it.
828 rt_getifa_fib(struct rt_addrinfo *info, u_int fibnum)
834 * ifp may be specified by sockaddr_dl
835 * when protocol address is ambiguous.
837 if (info->rti_ifp == NULL && ifpaddr != NULL &&
838 ifpaddr->sa_family == AF_LINK &&
839 (ifa = ifa_ifwithnet_fib(ifpaddr, 0, fibnum)) != NULL) {
840 info->rti_ifp = ifa->ifa_ifp;
843 if (info->rti_ifa == NULL && ifaaddr != NULL)
844 info->rti_ifa = ifa_ifwithaddr(ifaaddr);
845 if (info->rti_ifa == NULL) {
848 sa = ifaaddr != NULL ? ifaaddr :
849 (gateway != NULL ? gateway : dst);
850 if (sa != NULL && info->rti_ifp != NULL)
851 info->rti_ifa = ifaof_ifpforaddr(sa, info->rti_ifp);
852 else if (dst != NULL && gateway != NULL)
853 info->rti_ifa = ifa_ifwithroute_fib(flags, dst, gateway,
856 info->rti_ifa = ifa_ifwithroute_fib(flags, sa, sa,
859 if ((ifa = info->rti_ifa) != NULL) {
860 if (info->rti_ifp == NULL)
861 info->rti_ifp = ifa->ifa_ifp;
868 * Expunges references to a route that's about to be reclaimed.
869 * The route must be locked.
872 rtexpunge(struct rtentry *rt)
874 #if !defined(RADIX_MPATH)
875 struct radix_node *rn;
877 struct rt_addrinfo info;
881 struct radix_node_head *rnh;
886 * Find the correct routing tree to use for this Address Family
888 rnh = rt_tables_get_rnh(rt->rt_fibnum, rt_key(rt)->sa_family);
891 return (EAFNOSUPPORT);
892 RADIX_NODE_HEAD_LOCK_ASSERT(rnh);
896 bzero(&info, sizeof(info));
897 info.rti_ifp = rt->rt_ifp;
898 info.rti_flags = RTF_RNH_LOCKED;
899 info.rti_info[RTAX_DST] = rt_key(rt);
900 info.rti_info[RTAX_GATEWAY] = rt->rt_ifa->ifa_addr;
903 error = rtrequest1_fib(RTM_DELETE, &info, &rt0, fib);
905 if (error == 0 && rt0 != NULL) {
908 } else if (error != 0) {
914 * Remove the item from the tree; it should be there,
915 * but when callers invoke us blindly it may not (sigh).
917 rn = rnh->rnh_deladdr(rt_key(rt), rt_mask(rt), rnh);
922 KASSERT((rn->rn_flags & (RNF_ACTIVE | RNF_ROOT)) == 0,
923 ("unexpected flags 0x%x", rn->rn_flags));
924 KASSERT(rt == RNTORT(rn),
925 ("lookup mismatch, rt %p rn %p", rt, rn));
926 #endif /* RADIX_MPATH */
928 rt->rt_flags &= ~RTF_UP;
931 * Give the protocol a chance to keep things in sync.
933 if ((ifa = rt->rt_ifa) && ifa->ifa_rtrequest) {
934 struct rt_addrinfo info;
936 bzero((caddr_t)&info, sizeof(info));
937 info.rti_flags = rt->rt_flags;
938 info.rti_info[RTAX_DST] = rt_key(rt);
939 info.rti_info[RTAX_GATEWAY] = rt->rt_gateway;
940 info.rti_info[RTAX_NETMASK] = rt_mask(rt);
941 ifa->ifa_rtrequest(RTM_DELETE, rt, &info);
945 * one more rtentry floating around that is not
946 * linked to the routing table.
949 #if !defined(RADIX_MPATH)
956 int p_sockaddr(char *buf, int buflen, struct sockaddr *s);
957 int rt_print(char *buf, int buflen, struct rtentry *rt);
960 p_sockaddr(char *buf, int buflen, struct sockaddr *s)
964 switch (s->sa_family) {
966 paddr = &((struct sockaddr_in *)s)->sin_addr;
969 paddr = &((struct sockaddr_in6 *)s)->sin6_addr;
976 if (inet_ntop(s->sa_family, paddr, buf, buflen) == NULL)
979 return (strlen(buf));
983 rt_print(char *buf, int buflen, struct rtentry *rt)
985 struct sockaddr *addr, *mask;
991 i = p_sockaddr(buf, buflen, addr);
992 if (!(rt->rt_flags & RTF_HOST)) {
994 i += p_sockaddr(buf + i, buflen - i, mask);
997 if (rt->rt_flags & RTF_GATEWAY) {
999 i += p_sockaddr(buf + i, buflen - i, rt->rt_gateway);
1008 rn_mpath_update(int req, struct rt_addrinfo *info,
1009 struct radix_node_head *rnh, struct rtentry **ret_nrt)
1012 * if we got multipath routes, we require users to specify
1013 * a matching RTAX_GATEWAY.
1015 struct rtentry *rt, *rto = NULL;
1016 register struct radix_node *rn;
1019 rn = rnh->rnh_lookup(dst, netmask, rnh);
1022 rto = rt = RNTORT(rn);
1024 rt = rt_mpath_matchgate(rt, gateway);
1028 * this is the first entry in the chain
1031 rn = rn_mpath_next((struct radix_node *)rt);
1033 * there is another entry, now it's active
1038 rto->rt_flags |= RTF_UP;
1040 } else if (rt->rt_flags & RTF_GATEWAY) {
1042 * For gateway routes, we need to
1043 * make sure that we we are deleting
1044 * the correct gateway.
1045 * rt_mpath_matchgate() does not
1046 * check the case when there is only
1047 * one route in the chain.
1050 (rt->rt_gateway->sa_len != gateway->sa_len ||
1051 memcmp(rt->rt_gateway, gateway, gateway->sa_len)))
1055 * remove from tree before returning it
1058 rn = rnh->rnh_deladdr(dst, netmask, rnh);
1059 KASSERT(rt == RNTORT(rn), ("radix node disappeared"));
1065 * use the normal delete code to remove
1068 if (req != RTM_DELETE)
1076 * if the entry is 2nd and on up
1078 if ((req == RTM_DELETE) && !rt_mpath_deldup(rto, rt))
1079 panic ("rtrequest1: rt_mpath_deldup");
1083 if (req == RTM_DELETE) {
1084 rt->rt_flags &= ~RTF_UP;
1086 * One more rtentry floating around that is not
1087 * linked to the routing table. rttrash will be decremented
1088 * when RTFREE(rt) is eventually called.
1094 if (req != RTM_DELETE)
1095 panic("unrecognized request %d", req);
1099 * If the caller wants it, then it can have it,
1100 * but it's up to it to free the rtentry as we won't be
1114 rtrequest1_fib(int req, struct rt_addrinfo *info, struct rtentry **ret_nrt,
1117 int error = 0, needlock = 0;
1118 register struct rtentry *rt;
1120 register struct rtentry *rt0;
1122 register struct radix_node *rn;
1123 register struct radix_node_head *rnh;
1125 struct sockaddr *ndst;
1126 struct sockaddr_storage mdst;
1127 #define senderr(x) { error = x ; goto bad; }
1129 KASSERT((fibnum < rt_numfibs), ("rtrequest1_fib: bad fibnum"));
1130 switch (dst->sa_family) {
1133 /* We support multiple FIBs. */
1136 fibnum = RT_DEFAULT_FIB;
1141 * Find the correct routing tree to use for this Address Family
1143 rnh = rt_tables_get_rnh(fibnum, dst->sa_family);
1145 return (EAFNOSUPPORT);
1146 needlock = ((flags & RTF_RNH_LOCKED) == 0);
1147 flags &= ~RTF_RNH_LOCKED;
1149 RADIX_NODE_HEAD_LOCK(rnh);
1151 RADIX_NODE_HEAD_LOCK_ASSERT(rnh);
1153 * If we are adding a host route then we don't want to put
1154 * a netmask in the tree, nor do we want to clone it.
1156 if (flags & RTF_HOST)
1162 rt_maskedcopy(dst, (struct sockaddr *)&mdst, netmask);
1163 dst = (struct sockaddr *)&mdst;
1166 if (rn_mpath_capable(rnh)) {
1167 error = rn_mpath_update(req, info, rnh, ret_nrt);
1169 * "bad" holds true for the success case
1172 if (error != ENOENT)
1177 if ((flags & RTF_PINNED) == 0) {
1178 /* Check if target route can be deleted */
1179 rt = (struct rtentry *)rnh->rnh_lookup(dst,
1181 if ((rt != NULL) && (rt->rt_flags & RTF_PINNED))
1182 senderr(EADDRINUSE);
1186 * Remove the item from the tree and return it.
1187 * Complain if it is not there and do no more processing.
1189 rn = rnh->rnh_deladdr(dst, netmask, rnh);
1192 if (rn->rn_flags & (RNF_ACTIVE | RNF_ROOT))
1193 panic ("rtrequest delete");
1197 rt->rt_flags &= ~RTF_UP;
1200 * give the protocol a chance to keep things in sync.
1202 if ((ifa = rt->rt_ifa) && ifa->ifa_rtrequest)
1203 ifa->ifa_rtrequest(RTM_DELETE, rt, info);
1206 * One more rtentry floating around that is not
1207 * linked to the routing table. rttrash will be decremented
1208 * when RTFREE(rt) is eventually called.
1213 * If the caller wants it, then it can have it,
1214 * but it's up to it to free the rtentry as we won't be
1225 * resolve was only used for route cloning
1230 if ((flags & RTF_GATEWAY) && !gateway)
1232 if (dst && gateway && (dst->sa_family != gateway->sa_family) &&
1233 (gateway->sa_family != AF_UNSPEC) && (gateway->sa_family != AF_LINK))
1236 if (info->rti_ifa == NULL) {
1237 error = rt_getifa_fib(info, fibnum);
1241 ifa_ref(info->rti_ifa);
1242 ifa = info->rti_ifa;
1243 rt = uma_zalloc(V_rtzone, M_NOWAIT | M_ZERO);
1250 rt->rt_flags = RTF_UP | flags;
1251 rt->rt_fibnum = fibnum;
1253 * Add the gateway. Possibly re-malloc-ing the storage for it.
1256 if ((error = rt_setgate(rt, dst, gateway)) != 0) {
1257 RT_LOCK_DESTROY(rt);
1260 uma_zfree(V_rtzone, rt);
1265 * point to the (possibly newly malloc'd) dest address.
1267 ndst = (struct sockaddr *)rt_key(rt);
1270 * make sure it contains the value we want (masked if needed).
1273 rt_maskedcopy(dst, ndst, netmask);
1275 bcopy(dst, ndst, dst->sa_len);
1278 * We use the ifa reference returned by rt_getifa_fib().
1279 * This moved from below so that rnh->rnh_addaddr() can
1280 * examine the ifa and ifa->ifa_ifp if it so desires.
1283 rt->rt_ifp = ifa->ifa_ifp;
1284 rt->rt_rmx.rmx_weight = 1;
1287 /* do not permit exactly the same dst/mask/gw pair */
1288 if (rn_mpath_capable(rnh) &&
1289 rt_mpath_conflict(rnh, rt, netmask)) {
1291 ifa_free(rt->rt_ifa);
1294 RT_LOCK_DESTROY(rt);
1295 uma_zfree(V_rtzone, rt);
1302 /* "flow-table" only supports IPv6 and IPv4 at the moment. */
1303 switch (dst->sa_family) {
1310 #if defined(INET6) || defined(INET)
1311 rn = rnh->rnh_matchaddr(dst, rnh);
1312 if (rn && ((rn->rn_flags & RNF_ROOT) == 0)) {
1313 struct sockaddr *mask;
1318 * compare mask to see if the new route is
1319 * more specific than the existing one
1326 * A host route is already present, so
1327 * leave the flow-table entries as is.
1329 if (rt0->rt_flags & RTF_HOST) {
1332 } else if (!(flags & RTF_HOST) && netmask) {
1333 mask = rt_mask(rt0);
1336 n = (u_char *)netmask;
1343 if (len == 0 || (*n < *m)) {
1349 #endif/* INET6 || INET */
1351 #endif /* FLOWTABLE */
1353 /* XXX mtu manipulation will be done in rnh_addaddr -- itojun */
1354 rn = rnh->rnh_addaddr(ndst, netmask, rnh, rt->rt_nodes);
1356 * If it still failed to go into the tree,
1357 * then un-make it (this should be a function)
1361 ifa_free(rt->rt_ifa);
1363 RT_LOCK_DESTROY(rt);
1364 uma_zfree(V_rtzone, rt);
1372 else if (rt0 != NULL) {
1373 switch (dst->sa_family) {
1376 flowtable_route_flush(V_ip6_ft, rt0);
1381 flowtable_route_flush(V_ip_ft, rt0);
1390 * If this protocol has something to add to this then
1391 * allow it to do that as well.
1393 if (ifa->ifa_rtrequest)
1394 ifa->ifa_rtrequest(req, rt, info);
1397 * actually return a resultant rtentry and
1398 * give the caller a single reference.
1411 RADIX_NODE_HEAD_UNLOCK(rnh);
1424 rt_setgate(struct rtentry *rt, struct sockaddr *dst, struct sockaddr *gate)
1426 /* XXX dst may be overwritten, can we move this to below */
1427 int dlen = SA_SIZE(dst), glen = SA_SIZE(gate);
1429 struct radix_node_head *rnh;
1431 rnh = rt_tables_get_rnh(rt->rt_fibnum, dst->sa_family);
1435 RADIX_NODE_HEAD_LOCK_ASSERT(rnh);
1438 * Prepare to store the gateway in rt->rt_gateway.
1439 * Both dst and gateway are stored one after the other in the same
1440 * malloc'd chunk. If we have room, we can reuse the old buffer,
1441 * rt_gateway already points to the right place.
1442 * Otherwise, malloc a new block and update the 'dst' address.
1444 if (rt->rt_gateway == NULL || glen > SA_SIZE(rt->rt_gateway)) {
1447 R_Malloc(new, caddr_t, dlen + glen);
1451 * XXX note, we copy from *dst and not *rt_key(rt) because
1452 * rt_setgate() can be called to initialize a newly
1453 * allocated route entry, in which case rt_key(rt) == NULL
1454 * (and also rt->rt_gateway == NULL).
1455 * Free()/free() handle a NULL argument just fine.
1457 bcopy(dst, new, dlen);
1458 Free(rt_key(rt)); /* free old block, if any */
1459 rt_key(rt) = (struct sockaddr *)new;
1460 rt->rt_gateway = (struct sockaddr *)(new + dlen);
1464 * Copy the new gateway value into the memory chunk.
1466 bcopy(gate, rt->rt_gateway, glen);
1472 rt_maskedcopy(struct sockaddr *src, struct sockaddr *dst, struct sockaddr *netmask)
1474 register u_char *cp1 = (u_char *)src;
1475 register u_char *cp2 = (u_char *)dst;
1476 register u_char *cp3 = (u_char *)netmask;
1477 u_char *cplim = cp2 + *cp3;
1478 u_char *cplim2 = cp2 + *cp1;
1480 *cp2++ = *cp1++; *cp2++ = *cp1++; /* copies sa_len & sa_family */
1485 *cp2++ = *cp1++ & *cp3++;
1487 bzero((caddr_t)cp2, (unsigned)(cplim2 - cp2));
1491 * Set up a routing table entry, normally
1494 #define _SOCKADDR_TMPSIZE 128 /* Not too big.. kernel stack size is limited */
1496 rtinit1(struct ifaddr *ifa, int cmd, int flags, int fibnum)
1498 struct sockaddr *dst;
1499 struct sockaddr *netmask;
1500 struct rtentry *rt = NULL;
1501 struct rt_addrinfo info;
1503 int startfib, endfib;
1504 char tempbuf[_SOCKADDR_TMPSIZE];
1507 static struct sockaddr_dl null_sdl = {sizeof(null_sdl), AF_LINK};
1508 struct radix_node_head *rnh;
1510 if (flags & RTF_HOST) {
1511 dst = ifa->ifa_dstaddr;
1514 dst = ifa->ifa_addr;
1515 netmask = ifa->ifa_netmask;
1517 if (dst->sa_len == 0)
1519 switch (dst->sa_family) {
1522 /* We support multiple FIBs. */
1525 fibnum = RT_DEFAULT_FIB;
1528 if (fibnum == RT_ALL_FIBS) {
1529 if (rt_add_addr_allfibs == 0 && cmd == (int)RTM_ADD) {
1530 startfib = endfib = ifa->ifa_ifp->if_fib;
1533 endfib = rt_numfibs - 1;
1536 KASSERT((fibnum < rt_numfibs), ("rtinit1: bad fibnum"));
1542 * If it's a delete, check that if it exists,
1543 * it's on the correct interface or we might scrub
1544 * a route to another ifa which would
1545 * be confusing at best and possibly worse.
1547 if (cmd == RTM_DELETE) {
1549 * It's a delete, so it should already exist..
1550 * If it's a net, mask off the host bits
1551 * (Assuming we have a mask)
1552 * XXX this is kinda inet specific..
1554 if (netmask != NULL) {
1555 rt_maskedcopy(dst, (struct sockaddr *)tempbuf, netmask);
1556 dst = (struct sockaddr *)tempbuf;
1560 * Now go through all the requested tables (fibs) and do the
1561 * requested action. Realistically, this will either be fib 0
1562 * for protocols that don't do multiple tables or all the
1563 * tables for those that do.
1565 for ( fibnum = startfib; fibnum <= endfib; fibnum++) {
1566 if (cmd == RTM_DELETE) {
1567 struct radix_node *rn;
1569 * Look up an rtentry that is in the routing tree and
1570 * contains the correct info.
1572 rnh = rt_tables_get_rnh(fibnum, dst->sa_family);
1574 /* this table doesn't exist but others might */
1576 RADIX_NODE_HEAD_RLOCK(rnh);
1577 rn = rnh->rnh_lookup(dst, netmask, rnh);
1579 if (rn_mpath_capable(rnh)) {
1586 * for interface route the
1587 * rt->rt_gateway is sockaddr_intf
1588 * for cloning ARP entries, so
1589 * rt_mpath_matchgate must use the
1592 rt = rt_mpath_matchgate(rt,
1599 error = (rn == NULL ||
1600 (rn->rn_flags & RNF_ROOT) ||
1601 RNTORT(rn)->rt_ifa != ifa);
1602 RADIX_NODE_HEAD_RUNLOCK(rnh);
1604 /* this is only an error if bad on ALL tables */
1609 * Do the actual request
1611 bzero((caddr_t)&info, sizeof(info));
1613 info.rti_flags = flags |
1614 (ifa->ifa_flags & ~IFA_RTSELF) | RTF_PINNED;
1615 info.rti_info[RTAX_DST] = dst;
1617 * doing this for compatibility reasons
1620 info.rti_info[RTAX_GATEWAY] =
1621 (struct sockaddr *)&null_sdl;
1623 info.rti_info[RTAX_GATEWAY] = ifa->ifa_addr;
1624 info.rti_info[RTAX_NETMASK] = netmask;
1625 error = rtrequest1_fib(cmd, &info, &rt, fibnum);
1627 if ((error == EEXIST) && (cmd == RTM_ADD)) {
1629 * Interface route addition failed.
1630 * Atomically delete current prefix generating
1631 * RTM_DELETE message, and retry adding
1634 rnh = rt_tables_get_rnh(fibnum, dst->sa_family);
1635 RADIX_NODE_HEAD_LOCK(rnh);
1637 /* Delete old prefix */
1638 info.rti_ifa = NULL;
1639 info.rti_flags = RTF_RNH_LOCKED;
1641 error = rtrequest1_fib(RTM_DELETE, &info, NULL, fibnum);
1644 info.rti_flags = flags | RTF_RNH_LOCKED |
1645 (ifa->ifa_flags & ~IFA_RTSELF) | RTF_PINNED;
1646 error = rtrequest1_fib(cmd, &info, &rt, fibnum);
1649 RADIX_NODE_HEAD_UNLOCK(rnh);
1653 if (error == 0 && rt != NULL) {
1655 * notify any listening routing agents of the change
1660 * in case address alias finds the first address
1661 * e.g. ifconfig bge0 192.0.2.246/24
1662 * e.g. ifconfig bge0 192.0.2.247/24
1663 * the address set in the route is 192.0.2.246
1664 * so we need to replace it with 192.0.2.247
1666 if (memcmp(rt->rt_ifa->ifa_addr,
1667 ifa->ifa_addr, ifa->ifa_addr->sa_len)) {
1668 ifa_free(rt->rt_ifa);
1670 rt->rt_ifp = ifa->ifa_ifp;
1675 * doing this for compatibility reasons
1677 if (cmd == RTM_ADD) {
1678 ((struct sockaddr_dl *)rt->rt_gateway)->sdl_type =
1679 rt->rt_ifp->if_type;
1680 ((struct sockaddr_dl *)rt->rt_gateway)->sdl_index =
1681 rt->rt_ifp->if_index;
1685 rt_newaddrmsg_fib(cmd, ifa, error, rt, fibnum);
1688 if (cmd == RTM_DELETE) {
1690 * If we are deleting, and we found an entry,
1691 * then it's been removed from the tree..
1692 * now throw it away.
1696 if (cmd == RTM_ADD) {
1698 * We just wanted to add it..
1699 * we don't actually need a reference.
1710 if (cmd == RTM_DELETE) {
1714 /* we only give an error if it wasn't in any table */
1715 error = ((flags & RTF_HOST) ?
1716 EHOSTUNREACH : ENETUNREACH);
1720 /* return an error if any of them failed */
1727 #ifndef BURN_BRIDGES
1728 /* special one for inet internal use. may not use. */
1730 rtinit_fib(struct ifaddr *ifa, int cmd, int flags)
1732 return (rtinit1(ifa, cmd, flags, RT_ALL_FIBS));
1737 * Set up a routing table entry, normally
1741 rtinit(struct ifaddr *ifa, int cmd, int flags)
1743 struct sockaddr *dst;
1744 int fib = RT_DEFAULT_FIB;
1746 if (flags & RTF_HOST) {
1747 dst = ifa->ifa_dstaddr;
1749 dst = ifa->ifa_addr;
1752 switch (dst->sa_family) {
1755 /* We do support multiple FIBs. */
1759 return (rtinit1(ifa, cmd, flags, fib));
1763 * Announce interface address arrival/withdraw
1764 * Returns 0 on success.
1767 rt_addrmsg(int cmd, struct ifaddr *ifa, int fibnum)
1770 KASSERT(cmd == RTM_ADD || cmd == RTM_DELETE,
1771 ("unexpected cmd %d", cmd));
1773 KASSERT(fibnum == RT_ALL_FIBS || (fibnum >= 0 && fibnum < rt_numfibs),
1774 ("%s: fib out of range 0 <=%d<%d", __func__, fibnum, rt_numfibs));
1776 #if defined(INET) || defined(INET6)
1779 * notify the SCTP stack
1780 * this will only get called when an address is added/deleted
1781 * XXX pass the ifaddr struct instead if ifa->ifa_addr...
1783 sctp_addr_change(ifa, cmd);
1786 return (rtsock_addrmsg(cmd, ifa, fibnum));
1790 * Announce route addition/removal.
1791 * Users of this function MUST validate input data BEFORE calling.
1792 * However we have to be able to handle invalid data:
1793 * if some userland app sends us "invalid" route message (invalid mask,
1794 * no dst, wrong address families, etc...) we need to pass it back
1795 * to app (and any other rtsock consumers) with rtm_errno field set to
1797 * Returns 0 on success.
1800 rt_routemsg(int cmd, struct ifnet *ifp, int error, struct rtentry *rt,
1804 KASSERT(cmd == RTM_ADD || cmd == RTM_DELETE,
1805 ("unexpected cmd %d", cmd));
1807 KASSERT(fibnum == RT_ALL_FIBS || (fibnum >= 0 && fibnum < rt_numfibs),
1808 ("%s: fib out of range 0 <=%d<%d", __func__, fibnum, rt_numfibs));
1810 KASSERT(rt_key(rt) != NULL, (":%s: rt_key must be supplied", __func__));
1812 return (rtsock_routemsg(cmd, ifp, error, rt, fibnum));
1816 rt_newaddrmsg(int cmd, struct ifaddr *ifa, int error, struct rtentry *rt)
1819 rt_newaddrmsg_fib(cmd, ifa, error, rt, RT_ALL_FIBS);
1823 * This is called to generate messages from the routing socket
1824 * indicating a network interface has had addresses associated with it.
1827 rt_newaddrmsg_fib(int cmd, struct ifaddr *ifa, int error, struct rtentry *rt,
1831 KASSERT(cmd == RTM_ADD || cmd == RTM_DELETE,
1832 ("unexpected cmd %u", cmd));
1833 KASSERT(fibnum == RT_ALL_FIBS || (fibnum >= 0 && fibnum < rt_numfibs),
1834 ("%s: fib out of range 0 <=%d<%d", __func__, fibnum, rt_numfibs));
1836 if (cmd == RTM_ADD) {
1837 rt_addrmsg(cmd, ifa, fibnum);
1839 rt_routemsg(cmd, ifa->ifa_ifp, error, rt, fibnum);
1842 rt_routemsg(cmd, ifa->ifa_ifp, error, rt, fibnum);
1843 rt_addrmsg(cmd, ifa, fibnum);