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_var.h>
59 #include <net/if_dl.h>
60 #include <net/route.h>
62 #include <net/flowtable.h>
65 #include <net/radix_mpath.h>
68 #include <netinet/in.h>
69 #include <netinet/ip_mroute.h>
73 #define RT_MAXFIBS UINT16_MAX
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, "");
100 /* and this can be set too big but will be fixed before it is used */
101 TUNABLE_INT("net.fibs", &rt_numfibs);
104 * By default add routes to all fibs for new interfaces.
105 * Once this is set to 0 then only allocate routes on interface
106 * changes for the FIB of the caller when adding a new set of addresses
107 * to an interface. XXX this is a shotgun aproach to a problem that needs
108 * a more fine grained solution.. that will come.
109 * XXX also has the problems getting the FIB from curthread which will not
110 * always work given the fib can be overridden and prefixes can be added
111 * from the network stack context.
113 u_int rt_add_addr_allfibs = 1;
114 SYSCTL_UINT(_net, OID_AUTO, add_addr_allfibs, CTLFLAG_RW,
115 &rt_add_addr_allfibs, 0, "");
116 TUNABLE_INT("net.add_addr_allfibs", &rt_add_addr_allfibs);
118 VNET_DEFINE(struct rtstat, rtstat);
119 #define V_rtstat VNET(rtstat)
121 VNET_DEFINE(struct radix_node_head *, rt_tables);
122 #define V_rt_tables VNET(rt_tables)
124 VNET_DEFINE(int, rttrash); /* routes not in table but not freed */
125 #define V_rttrash VNET(rttrash)
128 /* compare two sockaddr structures */
129 #define sa_equal(a1, a2) (((a1)->sa_len == (a2)->sa_len) && \
130 (bcmp((a1), (a2), (a1)->sa_len) == 0))
133 * Convert a 'struct radix_node *' to a 'struct rtentry *'.
134 * The operation can be done safely (in this code) because a
135 * 'struct rtentry' starts with two 'struct radix_node''s, the first
136 * one representing leaf nodes in the routing tree, which is
137 * what the code in radix.c passes us as a 'struct radix_node'.
139 * But because there are a lot of assumptions in this conversion,
140 * do not cast explicitly, but always use the macro below.
142 #define RNTORT(p) ((struct rtentry *)(p))
144 static VNET_DEFINE(uma_zone_t, rtzone); /* Routing table UMA zone. */
145 #define V_rtzone VNET(rtzone)
148 * handler for net.my_fibnum
151 sysctl_my_fibnum(SYSCTL_HANDLER_ARGS)
156 fibnum = curthread->td_proc->p_fibnum;
157 error = sysctl_handle_int(oidp, &fibnum, 0, req);
161 SYSCTL_PROC(_net, OID_AUTO, my_fibnum, CTLTYPE_INT|CTLFLAG_RD,
162 NULL, 0, &sysctl_my_fibnum, "I", "default FIB of caller");
164 static __inline struct radix_node_head **
165 rt_tables_get_rnh_ptr(int table, int fam)
167 struct radix_node_head **rnh;
169 KASSERT(table >= 0 && table < rt_numfibs, ("%s: table out of bounds.",
171 KASSERT(fam >= 0 && fam < (AF_MAX+1), ("%s: fam out of bounds.",
174 /* rnh is [fib=0][af=0]. */
175 rnh = (struct radix_node_head **)V_rt_tables;
176 /* Get the offset to the requested table and fam. */
177 rnh += table * (AF_MAX+1) + fam;
182 struct radix_node_head *
183 rt_tables_get_rnh(int table, int fam)
186 return (*rt_tables_get_rnh_ptr(table, fam));
190 * route initialization must occur before ip6_init2(), which happenas at
197 /* whack the tunable ints into line. */
198 if (rt_numfibs > RT_MAXFIBS)
199 rt_numfibs = RT_MAXFIBS;
203 SYSINIT(route_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_THIRD, route_init, 0);
206 vnet_route_init(const void *unused __unused)
209 struct radix_node_head **rnh;
213 V_rt_tables = malloc(rt_numfibs * (AF_MAX+1) *
214 sizeof(struct radix_node_head *), M_RTABLE, M_WAITOK|M_ZERO);
216 V_rtzone = uma_zcreate("rtentry", sizeof(struct rtentry), NULL, NULL,
217 NULL, NULL, UMA_ALIGN_PTR, 0);
218 for (dom = domains; dom; dom = dom->dom_next) {
219 if (dom->dom_rtattach == NULL)
222 for (table = 0; table < rt_numfibs; table++) {
223 fam = dom->dom_family;
224 if (table != 0 && fam != AF_INET6 && fam != AF_INET)
228 * XXX MRT rtattach will be also called from
229 * vfs_export.c but the offset will be 0 (only for
230 * AF_INET and AF_INET6 which don't need it anyhow).
232 rnh = rt_tables_get_rnh_ptr(table, fam);
234 panic("%s: rnh NULL", __func__);
235 dom->dom_rtattach((void **)rnh, dom->dom_rtoffset);
239 VNET_SYSINIT(vnet_route_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_FOURTH,
244 vnet_route_uninit(const void *unused __unused)
249 struct radix_node_head **rnh;
251 for (dom = domains; dom; dom = dom->dom_next) {
252 if (dom->dom_rtdetach == NULL)
255 for (table = 0; table < rt_numfibs; table++) {
256 fam = dom->dom_family;
258 if (table != 0 && fam != AF_INET6 && fam != AF_INET)
261 rnh = rt_tables_get_rnh_ptr(table, fam);
263 panic("%s: rnh NULL", __func__);
264 dom->dom_rtdetach((void **)rnh, dom->dom_rtoffset);
268 free(V_rt_tables, M_RTABLE);
269 uma_zdestroy(V_rtzone);
271 VNET_SYSUNINIT(vnet_route_uninit, SI_SUB_PROTO_DOMAIN, SI_ORDER_THIRD,
272 vnet_route_uninit, 0);
275 #ifndef _SYS_SYSPROTO_H_
281 sys_setfib(struct thread *td, struct setfib_args *uap)
283 if (uap->fibnum < 0 || uap->fibnum >= rt_numfibs)
285 td->td_proc->p_fibnum = uap->fibnum;
290 * Packet routing routines.
293 rtalloc(struct route *ro)
296 rtalloc_ign_fib(ro, 0UL, RT_DEFAULT_FIB);
300 rtalloc_fib(struct route *ro, u_int fibnum)
302 rtalloc_ign_fib(ro, 0UL, fibnum);
306 rtalloc_ign(struct route *ro, u_long ignore)
310 if ((rt = ro->ro_rt) != NULL) {
311 if (rt->rt_ifp != NULL && rt->rt_flags & RTF_UP)
316 ro->ro_rt = rtalloc1_fib(&ro->ro_dst, 1, ignore, RT_DEFAULT_FIB);
318 RT_UNLOCK(ro->ro_rt);
322 rtalloc_ign_fib(struct route *ro, u_long ignore, u_int fibnum)
326 if ((rt = ro->ro_rt) != NULL) {
327 if (rt->rt_ifp != NULL && rt->rt_flags & RTF_UP)
332 ro->ro_rt = rtalloc1_fib(&ro->ro_dst, 1, ignore, fibnum);
334 RT_UNLOCK(ro->ro_rt);
338 * Look up the route that matches the address given
339 * Or, at least try.. Create a cloned route if needed.
341 * The returned route, if any, is locked.
344 rtalloc1(struct sockaddr *dst, int report, u_long ignflags)
347 return (rtalloc1_fib(dst, report, ignflags, RT_DEFAULT_FIB));
351 rtalloc1_fib(struct sockaddr *dst, int report, u_long ignflags,
354 struct radix_node_head *rnh;
355 struct radix_node *rn;
356 struct rtentry *newrt;
357 struct rt_addrinfo info;
358 int err = 0, msgtype = RTM_MISS;
361 KASSERT((fibnum < rt_numfibs), ("rtalloc1_fib: bad fibnum"));
362 switch (dst->sa_family) {
365 /* We support multiple FIBs. */
368 fibnum = RT_DEFAULT_FIB;
371 rnh = rt_tables_get_rnh(fibnum, dst->sa_family);
377 * Look up the address in the table for that Address Family
379 needlock = !(ignflags & RTF_RNH_LOCKED);
381 RADIX_NODE_HEAD_RLOCK(rnh);
384 RADIX_NODE_HEAD_LOCK_ASSERT(rnh);
386 rn = rnh->rnh_matchaddr(dst, rnh);
387 if (rn && ((rn->rn_flags & RNF_ROOT) == 0)) {
392 RADIX_NODE_HEAD_RUNLOCK(rnh);
396 RADIX_NODE_HEAD_RUNLOCK(rnh);
399 * Either we hit the root or couldn't find any match,
400 * Which basically means
401 * "caint get there frm here"
404 V_rtstat.rts_unreach++;
408 * If required, report the failure to the supervising
410 * For a delete, this is not an error. (report == 0)
412 bzero(&info, sizeof(info));
413 info.rti_info[RTAX_DST] = dst;
414 rt_missmsg_fib(msgtype, &info, 0, err, fibnum);
418 RT_LOCK_ASSERT(newrt);
423 * Remove a reference count from an rtentry.
424 * If the count gets low enough, take it out of the routing table
427 rtfree(struct rtentry *rt)
429 struct radix_node_head *rnh;
431 KASSERT(rt != NULL,("%s: NULL rt", __func__));
432 rnh = rt_tables_get_rnh(rt->rt_fibnum, rt_key(rt)->sa_family);
433 KASSERT(rnh != NULL,("%s: NULL rnh", __func__));
438 * The callers should use RTFREE_LOCKED() or RTFREE(), so
439 * we should come here exactly with the last reference.
442 if (rt->rt_refcnt > 0) {
443 log(LOG_DEBUG, "%s: %p has %d refs\n", __func__, rt, rt->rt_refcnt);
448 * On last reference give the "close method" a chance
449 * to cleanup private state. This also permits (for
450 * IPv4 and IPv6) a chance to decide if the routing table
451 * entry should be purged immediately or at a later time.
452 * When an immediate purge is to happen the close routine
453 * typically calls rtexpunge which clears the RTF_UP flag
454 * on the entry so that the code below reclaims the storage.
456 if (rt->rt_refcnt == 0 && rnh->rnh_close)
457 rnh->rnh_close((struct radix_node *)rt, rnh);
460 * If we are no longer "up" (and ref == 0)
461 * then we can free the resources associated
464 if ((rt->rt_flags & RTF_UP) == 0) {
465 if (rt->rt_nodes->rn_flags & (RNF_ACTIVE | RNF_ROOT))
468 * the rtentry must have been removed from the routing table
469 * so it is represented in rttrash.. remove that now.
473 if (rt->rt_refcnt < 0) {
474 printf("rtfree: %p not freed (neg refs)\n", rt);
479 * release references on items we hold them on..
480 * e.g other routes and ifaddrs.
483 ifa_free(rt->rt_ifa);
485 * The key is separatly alloc'd so free it (see rt_setgate()).
486 * This also frees the gateway, as they are always malloc'd
492 * and the rtentry itself of course
495 uma_zfree(V_rtzone, rt);
504 * Force a routing table entry to the specified
505 * destination to go through the given gateway.
506 * Normally called as a result of a routing redirect
507 * message from the network layer.
510 rtredirect(struct sockaddr *dst,
511 struct sockaddr *gateway,
512 struct sockaddr *netmask,
514 struct sockaddr *src)
517 rtredirect_fib(dst, gateway, netmask, flags, src, RT_DEFAULT_FIB);
521 rtredirect_fib(struct sockaddr *dst,
522 struct sockaddr *gateway,
523 struct sockaddr *netmask,
525 struct sockaddr *src,
528 struct rtentry *rt, *rt0 = NULL;
531 struct rt_addrinfo info;
533 struct radix_node_head *rnh;
536 rnh = rt_tables_get_rnh(fibnum, dst->sa_family);
538 error = EAFNOSUPPORT;
542 /* verify the gateway is directly reachable */
543 if ((ifa = ifa_ifwithnet(gateway, 0)) == NULL) {
547 rt = rtalloc1_fib(dst, 0, 0UL, fibnum); /* NB: rt is locked */
549 * If the redirect isn't from our current router for this dst,
550 * it's either old or wrong. If it redirects us to ourselves,
551 * we have a routing loop, perhaps as a result of an interface
552 * going down recently.
554 if (!(flags & RTF_DONE) && rt &&
555 (!sa_equal(src, rt->rt_gateway) || rt->rt_ifa != ifa))
557 else if (ifa_ifwithaddr_check(gateway))
558 error = EHOSTUNREACH;
562 * Create a new entry if we just got back a wildcard entry
563 * or the lookup failed. This is necessary for hosts
564 * which use routing redirects generated by smart gateways
565 * to dynamically build the routing tables.
567 if (rt == NULL || (rt_mask(rt) && rt_mask(rt)->sa_len < 2))
570 * Don't listen to the redirect if it's
571 * for a route to an interface.
573 if (rt->rt_flags & RTF_GATEWAY) {
574 if (((rt->rt_flags & RTF_HOST) == 0) && (flags & RTF_HOST)) {
576 * Changing from route to net => route to host.
577 * Create new route, rather than smashing route to net.
583 flags |= RTF_GATEWAY | RTF_DYNAMIC;
584 bzero((caddr_t)&info, sizeof(info));
585 info.rti_info[RTAX_DST] = dst;
586 info.rti_info[RTAX_GATEWAY] = gateway;
587 info.rti_info[RTAX_NETMASK] = netmask;
589 info.rti_flags = flags;
591 RT_UNLOCK(rt0); /* drop lock to avoid LOR with RNH */
592 error = rtrequest1_fib(RTM_ADD, &info, &rt, fibnum);
596 EVENTHANDLER_INVOKE(route_redirect_event, rt0, rt, dst);
597 flags = rt->rt_flags;
602 stat = &V_rtstat.rts_dynamic;
604 struct rtentry *gwrt;
607 * Smash the current notion of the gateway to
608 * this destination. Should check about netmask!!!
610 rt->rt_flags |= RTF_MODIFIED;
611 flags |= RTF_MODIFIED;
612 stat = &V_rtstat.rts_newgateway;
614 * add the key and gateway (in one malloc'd chunk).
617 RADIX_NODE_HEAD_LOCK(rnh);
619 rt_setgate(rt, rt_key(rt), gateway);
620 gwrt = rtalloc1(gateway, 1, RTF_RNH_LOCKED);
621 RADIX_NODE_HEAD_UNLOCK(rnh);
622 EVENTHANDLER_INVOKE(route_redirect_event, rt, gwrt, dst);
626 error = EHOSTUNREACH;
632 V_rtstat.rts_badredirect++;
633 else if (stat != NULL)
635 bzero((caddr_t)&info, sizeof(info));
636 info.rti_info[RTAX_DST] = dst;
637 info.rti_info[RTAX_GATEWAY] = gateway;
638 info.rti_info[RTAX_NETMASK] = netmask;
639 info.rti_info[RTAX_AUTHOR] = src;
640 rt_missmsg_fib(RTM_REDIRECT, &info, flags, error, fibnum);
646 rtioctl(u_long req, caddr_t data)
649 return (rtioctl_fib(req, data, RT_DEFAULT_FIB));
653 * Routing table ioctl interface.
656 rtioctl_fib(u_long req, caddr_t data, u_int fibnum)
660 * If more ioctl commands are added here, make sure the proper
661 * super-user checks are being performed because it is possible for
662 * prison-root to make it this far if raw sockets have been enabled
666 /* Multicast goop, grrr... */
667 return mrt_ioctl ? mrt_ioctl(req, data, fibnum) : EOPNOTSUPP;
674 * For both ifa_ifwithroute() routines, 'ifa' is returned referenced.
677 ifa_ifwithroute(int flags, struct sockaddr *dst, struct sockaddr *gateway)
680 return (ifa_ifwithroute_fib(flags, dst, gateway, RT_DEFAULT_FIB));
684 ifa_ifwithroute_fib(int flags, struct sockaddr *dst, struct sockaddr *gateway,
687 register struct ifaddr *ifa;
690 if ((flags & RTF_GATEWAY) == 0) {
692 * If we are adding a route to an interface,
693 * and the interface is a pt to pt link
694 * we should search for the destination
695 * as our clue to the interface. Otherwise
696 * we can use the local address.
699 if (flags & RTF_HOST)
700 ifa = ifa_ifwithdstaddr(dst);
702 ifa = ifa_ifwithaddr(gateway);
705 * If we are adding a route to a remote net
706 * or host, the gateway may still be on the
707 * other end of a pt to pt link.
709 ifa = ifa_ifwithdstaddr(gateway);
712 ifa = ifa_ifwithnet(gateway, 0);
714 struct rtentry *rt = rtalloc1_fib(gateway, 0, RTF_RNH_LOCKED, fibnum);
718 * dismiss a gateway that is reachable only
719 * through the default router
721 switch (gateway->sa_family) {
723 if (satosin(rt_key(rt))->sin_addr.s_addr == INADDR_ANY)
727 if (IN6_IS_ADDR_UNSPECIFIED(&satosin6(rt_key(rt))->sin6_addr))
733 if (!not_found && rt->rt_ifa != NULL) {
739 if (not_found || ifa == NULL)
742 if (ifa->ifa_addr->sa_family != dst->sa_family) {
743 struct ifaddr *oifa = ifa;
744 ifa = ifaof_ifpforaddr(dst, ifa->ifa_ifp);
754 * Do appropriate manipulations of a routing tree given
755 * all the bits of info needed
759 struct sockaddr *dst,
760 struct sockaddr *gateway,
761 struct sockaddr *netmask,
763 struct rtentry **ret_nrt)
766 return (rtrequest_fib(req, dst, gateway, netmask, flags, ret_nrt,
771 rtrequest_fib(int req,
772 struct sockaddr *dst,
773 struct sockaddr *gateway,
774 struct sockaddr *netmask,
776 struct rtentry **ret_nrt,
779 struct rt_addrinfo info;
781 if (dst->sa_len == 0)
784 bzero((caddr_t)&info, sizeof(info));
785 info.rti_flags = flags;
786 info.rti_info[RTAX_DST] = dst;
787 info.rti_info[RTAX_GATEWAY] = gateway;
788 info.rti_info[RTAX_NETMASK] = netmask;
789 return rtrequest1_fib(req, &info, ret_nrt, fibnum);
793 * These (questionable) definitions of apparent local variables apply
794 * to the next two functions. XXXXXX!!!
796 #define dst info->rti_info[RTAX_DST]
797 #define gateway info->rti_info[RTAX_GATEWAY]
798 #define netmask info->rti_info[RTAX_NETMASK]
799 #define ifaaddr info->rti_info[RTAX_IFA]
800 #define ifpaddr info->rti_info[RTAX_IFP]
801 #define flags info->rti_flags
804 rt_getifa(struct rt_addrinfo *info)
807 return (rt_getifa_fib(info, RT_DEFAULT_FIB));
811 * Look up rt_addrinfo for a specific fib. Note that if rti_ifa is defined,
812 * it will be referenced so the caller must free it.
815 rt_getifa_fib(struct rt_addrinfo *info, u_int fibnum)
821 * ifp may be specified by sockaddr_dl
822 * when protocol address is ambiguous.
824 if (info->rti_ifp == NULL && ifpaddr != NULL &&
825 ifpaddr->sa_family == AF_LINK &&
826 (ifa = ifa_ifwithnet(ifpaddr, 0)) != NULL) {
827 info->rti_ifp = ifa->ifa_ifp;
830 if (info->rti_ifa == NULL && ifaaddr != NULL)
831 info->rti_ifa = ifa_ifwithaddr(ifaaddr);
832 if (info->rti_ifa == NULL) {
835 sa = ifaaddr != NULL ? ifaaddr :
836 (gateway != NULL ? gateway : dst);
837 if (sa != NULL && info->rti_ifp != NULL)
838 info->rti_ifa = ifaof_ifpforaddr(sa, info->rti_ifp);
839 else if (dst != NULL && gateway != NULL)
840 info->rti_ifa = ifa_ifwithroute_fib(flags, dst, gateway,
843 info->rti_ifa = ifa_ifwithroute_fib(flags, sa, sa,
846 if ((ifa = info->rti_ifa) != NULL) {
847 if (info->rti_ifp == NULL)
848 info->rti_ifp = ifa->ifa_ifp;
855 * Expunges references to a route that's about to be reclaimed.
856 * The route must be locked.
859 rtexpunge(struct rtentry *rt)
861 #if !defined(RADIX_MPATH)
862 struct radix_node *rn;
864 struct rt_addrinfo info;
868 struct radix_node_head *rnh;
873 * Find the correct routing tree to use for this Address Family
875 rnh = rt_tables_get_rnh(rt->rt_fibnum, rt_key(rt)->sa_family);
878 return (EAFNOSUPPORT);
879 RADIX_NODE_HEAD_LOCK_ASSERT(rnh);
883 bzero(&info, sizeof(info));
884 info.rti_ifp = rt->rt_ifp;
885 info.rti_flags = RTF_RNH_LOCKED;
886 info.rti_info[RTAX_DST] = rt_key(rt);
887 info.rti_info[RTAX_GATEWAY] = rt->rt_ifa->ifa_addr;
890 error = rtrequest1_fib(RTM_DELETE, &info, &rt0, fib);
892 if (error == 0 && rt0 != NULL) {
895 } else if (error != 0) {
901 * Remove the item from the tree; it should be there,
902 * but when callers invoke us blindly it may not (sigh).
904 rn = rnh->rnh_deladdr(rt_key(rt), rt_mask(rt), rnh);
909 KASSERT((rn->rn_flags & (RNF_ACTIVE | RNF_ROOT)) == 0,
910 ("unexpected flags 0x%x", rn->rn_flags));
911 KASSERT(rt == RNTORT(rn),
912 ("lookup mismatch, rt %p rn %p", rt, rn));
913 #endif /* RADIX_MPATH */
915 rt->rt_flags &= ~RTF_UP;
918 * Give the protocol a chance to keep things in sync.
920 if ((ifa = rt->rt_ifa) && ifa->ifa_rtrequest) {
921 struct rt_addrinfo info;
923 bzero((caddr_t)&info, sizeof(info));
924 info.rti_flags = rt->rt_flags;
925 info.rti_info[RTAX_DST] = rt_key(rt);
926 info.rti_info[RTAX_GATEWAY] = rt->rt_gateway;
927 info.rti_info[RTAX_NETMASK] = rt_mask(rt);
928 ifa->ifa_rtrequest(RTM_DELETE, rt, &info);
932 * one more rtentry floating around that is not
933 * linked to the routing table.
936 #if !defined(RADIX_MPATH)
943 int p_sockaddr(char *buf, int buflen, struct sockaddr *s);
944 int rt_print(char *buf, int buflen, struct rtentry *rt);
947 p_sockaddr(char *buf, int buflen, struct sockaddr *s)
951 switch (s->sa_family) {
953 paddr = &((struct sockaddr_in *)s)->sin_addr;
956 paddr = &((struct sockaddr_in6 *)s)->sin6_addr;
963 if (inet_ntop(s->sa_family, paddr, buf, buflen) == NULL)
966 return (strlen(buf));
970 rt_print(char *buf, int buflen, struct rtentry *rt)
972 struct sockaddr *addr, *mask;
978 i = p_sockaddr(buf, buflen, addr);
979 if (!(rt->rt_flags & RTF_HOST)) {
981 i += p_sockaddr(buf + i, buflen - i, mask);
984 if (rt->rt_flags & RTF_GATEWAY) {
986 i += p_sockaddr(buf + i, buflen - i, rt->rt_gateway);
995 rn_mpath_update(int req, struct rt_addrinfo *info,
996 struct radix_node_head *rnh, struct rtentry **ret_nrt)
999 * if we got multipath routes, we require users to specify
1000 * a matching RTAX_GATEWAY.
1002 struct rtentry *rt, *rto = NULL;
1003 register struct radix_node *rn;
1006 rn = rnh->rnh_lookup(dst, netmask, rnh);
1009 rto = rt = RNTORT(rn);
1011 rt = rt_mpath_matchgate(rt, gateway);
1015 * this is the first entry in the chain
1018 rn = rn_mpath_next((struct radix_node *)rt);
1020 * there is another entry, now it's active
1025 rto->rt_flags |= RTF_UP;
1027 } else if (rt->rt_flags & RTF_GATEWAY) {
1029 * For gateway routes, we need to
1030 * make sure that we we are deleting
1031 * the correct gateway.
1032 * rt_mpath_matchgate() does not
1033 * check the case when there is only
1034 * one route in the chain.
1037 (rt->rt_gateway->sa_len != gateway->sa_len ||
1038 memcmp(rt->rt_gateway, gateway, gateway->sa_len)))
1042 * remove from tree before returning it
1045 rn = rnh->rnh_deladdr(dst, netmask, rnh);
1046 KASSERT(rt == RNTORT(rn), ("radix node disappeared"));
1052 * use the normal delete code to remove
1055 if (req != RTM_DELETE)
1063 * if the entry is 2nd and on up
1065 if ((req == RTM_DELETE) && !rt_mpath_deldup(rto, rt))
1066 panic ("rtrequest1: rt_mpath_deldup");
1070 if (req == RTM_DELETE) {
1071 rt->rt_flags &= ~RTF_UP;
1073 * One more rtentry floating around that is not
1074 * linked to the routing table. rttrash will be decremented
1075 * when RTFREE(rt) is eventually called.
1081 if (req != RTM_DELETE)
1082 panic("unrecognized request %d", req);
1086 * If the caller wants it, then it can have it,
1087 * but it's up to it to free the rtentry as we won't be
1101 rtrequest1_fib(int req, struct rt_addrinfo *info, struct rtentry **ret_nrt,
1104 int error = 0, needlock = 0;
1105 register struct rtentry *rt;
1107 register struct rtentry *rt0;
1109 register struct radix_node *rn;
1110 register struct radix_node_head *rnh;
1112 struct sockaddr *ndst;
1113 struct sockaddr_storage mdst;
1114 #define senderr(x) { error = x ; goto bad; }
1116 KASSERT((fibnum < rt_numfibs), ("rtrequest1_fib: bad fibnum"));
1117 switch (dst->sa_family) {
1120 /* We support multiple FIBs. */
1123 fibnum = RT_DEFAULT_FIB;
1128 * Find the correct routing tree to use for this Address Family
1130 rnh = rt_tables_get_rnh(fibnum, dst->sa_family);
1132 return (EAFNOSUPPORT);
1133 needlock = ((flags & RTF_RNH_LOCKED) == 0);
1134 flags &= ~RTF_RNH_LOCKED;
1136 RADIX_NODE_HEAD_LOCK(rnh);
1138 RADIX_NODE_HEAD_LOCK_ASSERT(rnh);
1140 * If we are adding a host route then we don't want to put
1141 * a netmask in the tree, nor do we want to clone it.
1143 if (flags & RTF_HOST)
1149 rt_maskedcopy(dst, (struct sockaddr *)&mdst, netmask);
1150 dst = (struct sockaddr *)&mdst;
1153 if (rn_mpath_capable(rnh)) {
1154 error = rn_mpath_update(req, info, rnh, ret_nrt);
1156 * "bad" holds true for the success case
1159 if (error != ENOENT)
1164 if ((flags & RTF_PINNED) == 0) {
1165 /* Check if target route can be deleted */
1166 rt = (struct rtentry *)rnh->rnh_lookup(dst,
1168 if ((rt != NULL) && (rt->rt_flags & RTF_PINNED))
1169 senderr(EADDRINUSE);
1173 * Remove the item from the tree and return it.
1174 * Complain if it is not there and do no more processing.
1176 rn = rnh->rnh_deladdr(dst, netmask, rnh);
1179 if (rn->rn_flags & (RNF_ACTIVE | RNF_ROOT))
1180 panic ("rtrequest delete");
1184 rt->rt_flags &= ~RTF_UP;
1187 * give the protocol a chance to keep things in sync.
1189 if ((ifa = rt->rt_ifa) && ifa->ifa_rtrequest)
1190 ifa->ifa_rtrequest(RTM_DELETE, rt, info);
1193 * One more rtentry floating around that is not
1194 * linked to the routing table. rttrash will be decremented
1195 * when RTFREE(rt) is eventually called.
1200 * If the caller wants it, then it can have it,
1201 * but it's up to it to free the rtentry as we won't be
1212 * resolve was only used for route cloning
1217 if ((flags & RTF_GATEWAY) && !gateway)
1219 if (dst && gateway && (dst->sa_family != gateway->sa_family) &&
1220 (gateway->sa_family != AF_UNSPEC) && (gateway->sa_family != AF_LINK))
1223 if (info->rti_ifa == NULL) {
1224 error = rt_getifa_fib(info, fibnum);
1228 ifa_ref(info->rti_ifa);
1229 ifa = info->rti_ifa;
1230 rt = uma_zalloc(V_rtzone, M_NOWAIT | M_ZERO);
1236 rt->rt_flags = RTF_UP | flags;
1237 rt->rt_fibnum = fibnum;
1239 * Add the gateway. Possibly re-malloc-ing the storage for it.
1242 if ((error = rt_setgate(rt, dst, gateway)) != 0) {
1243 RT_LOCK_DESTROY(rt);
1245 uma_zfree(V_rtzone, rt);
1250 * point to the (possibly newly malloc'd) dest address.
1252 ndst = (struct sockaddr *)rt_key(rt);
1255 * make sure it contains the value we want (masked if needed).
1258 rt_maskedcopy(dst, ndst, netmask);
1260 bcopy(dst, ndst, dst->sa_len);
1263 * We use the ifa reference returned by rt_getifa_fib().
1264 * This moved from below so that rnh->rnh_addaddr() can
1265 * examine the ifa and ifa->ifa_ifp if it so desires.
1268 rt->rt_ifp = ifa->ifa_ifp;
1269 rt->rt_rmx.rmx_weight = 1;
1272 /* do not permit exactly the same dst/mask/gw pair */
1273 if (rn_mpath_capable(rnh) &&
1274 rt_mpath_conflict(rnh, rt, netmask)) {
1275 ifa_free(rt->rt_ifa);
1277 RT_LOCK_DESTROY(rt);
1278 uma_zfree(V_rtzone, rt);
1285 /* "flow-table" only supports IPv6 and IPv4 at the moment. */
1286 switch (dst->sa_family) {
1293 #if defined(INET6) || defined(INET)
1294 rn = rnh->rnh_matchaddr(dst, rnh);
1295 if (rn && ((rn->rn_flags & RNF_ROOT) == 0)) {
1296 struct sockaddr *mask;
1301 * compare mask to see if the new route is
1302 * more specific than the existing one
1309 * A host route is already present, so
1310 * leave the flow-table entries as is.
1312 if (rt0->rt_flags & RTF_HOST) {
1315 } else if (!(flags & RTF_HOST) && netmask) {
1316 mask = rt_mask(rt0);
1319 n = (u_char *)netmask;
1326 if (len == 0 || (*n < *m)) {
1332 #endif/* INET6 || INET */
1334 #endif /* FLOWTABLE */
1336 /* XXX mtu manipulation will be done in rnh_addaddr -- itojun */
1337 rn = rnh->rnh_addaddr(ndst, netmask, rnh, rt->rt_nodes);
1339 * If it still failed to go into the tree,
1340 * then un-make it (this should be a function)
1343 ifa_free(rt->rt_ifa);
1345 RT_LOCK_DESTROY(rt);
1346 uma_zfree(V_rtzone, rt);
1354 else if (rt0 != NULL) {
1355 switch (dst->sa_family) {
1358 flowtable_route_flush(V_ip6_ft, rt0);
1363 flowtable_route_flush(V_ip_ft, rt0);
1372 * If this protocol has something to add to this then
1373 * allow it to do that as well.
1375 if (ifa->ifa_rtrequest)
1376 ifa->ifa_rtrequest(req, rt, info);
1379 * actually return a resultant rtentry and
1380 * give the caller a single reference.
1393 RADIX_NODE_HEAD_UNLOCK(rnh);
1406 rt_setgate(struct rtentry *rt, struct sockaddr *dst, struct sockaddr *gate)
1408 /* XXX dst may be overwritten, can we move this to below */
1409 int dlen = SA_SIZE(dst), glen = SA_SIZE(gate);
1411 struct radix_node_head *rnh;
1413 rnh = rt_tables_get_rnh(rt->rt_fibnum, dst->sa_family);
1417 RADIX_NODE_HEAD_LOCK_ASSERT(rnh);
1420 * Prepare to store the gateway in rt->rt_gateway.
1421 * Both dst and gateway are stored one after the other in the same
1422 * malloc'd chunk. If we have room, we can reuse the old buffer,
1423 * rt_gateway already points to the right place.
1424 * Otherwise, malloc a new block and update the 'dst' address.
1426 if (rt->rt_gateway == NULL || glen > SA_SIZE(rt->rt_gateway)) {
1429 R_Malloc(new, caddr_t, dlen + glen);
1433 * XXX note, we copy from *dst and not *rt_key(rt) because
1434 * rt_setgate() can be called to initialize a newly
1435 * allocated route entry, in which case rt_key(rt) == NULL
1436 * (and also rt->rt_gateway == NULL).
1437 * Free()/free() handle a NULL argument just fine.
1439 bcopy(dst, new, dlen);
1440 Free(rt_key(rt)); /* free old block, if any */
1441 rt_key(rt) = (struct sockaddr *)new;
1442 rt->rt_gateway = (struct sockaddr *)(new + dlen);
1446 * Copy the new gateway value into the memory chunk.
1448 bcopy(gate, rt->rt_gateway, glen);
1454 rt_maskedcopy(struct sockaddr *src, struct sockaddr *dst, struct sockaddr *netmask)
1456 register u_char *cp1 = (u_char *)src;
1457 register u_char *cp2 = (u_char *)dst;
1458 register u_char *cp3 = (u_char *)netmask;
1459 u_char *cplim = cp2 + *cp3;
1460 u_char *cplim2 = cp2 + *cp1;
1462 *cp2++ = *cp1++; *cp2++ = *cp1++; /* copies sa_len & sa_family */
1467 *cp2++ = *cp1++ & *cp3++;
1469 bzero((caddr_t)cp2, (unsigned)(cplim2 - cp2));
1473 * Set up a routing table entry, normally
1476 #define _SOCKADDR_TMPSIZE 128 /* Not too big.. kernel stack size is limited */
1478 rtinit1(struct ifaddr *ifa, int cmd, int flags, int fibnum)
1480 struct sockaddr *dst;
1481 struct sockaddr *netmask;
1482 struct rtentry *rt = NULL;
1483 struct rt_addrinfo info;
1485 int startfib, endfib;
1486 char tempbuf[_SOCKADDR_TMPSIZE];
1489 static struct sockaddr_dl null_sdl = {sizeof(null_sdl), AF_LINK};
1490 struct radix_node_head *rnh;
1492 if (flags & RTF_HOST) {
1493 dst = ifa->ifa_dstaddr;
1496 dst = ifa->ifa_addr;
1497 netmask = ifa->ifa_netmask;
1499 if (dst->sa_len == 0)
1501 switch (dst->sa_family) {
1504 /* We support multiple FIBs. */
1507 fibnum = RT_DEFAULT_FIB;
1510 if (fibnum == RT_ALL_FIBS) {
1511 if (rt_add_addr_allfibs == 0 && cmd == (int)RTM_ADD) {
1512 startfib = endfib = curthread->td_proc->p_fibnum;
1515 endfib = rt_numfibs - 1;
1518 KASSERT((fibnum < rt_numfibs), ("rtinit1: bad fibnum"));
1524 * If it's a delete, check that if it exists,
1525 * it's on the correct interface or we might scrub
1526 * a route to another ifa which would
1527 * be confusing at best and possibly worse.
1529 if (cmd == RTM_DELETE) {
1531 * It's a delete, so it should already exist..
1532 * If it's a net, mask off the host bits
1533 * (Assuming we have a mask)
1534 * XXX this is kinda inet specific..
1536 if (netmask != NULL) {
1537 rt_maskedcopy(dst, (struct sockaddr *)tempbuf, netmask);
1538 dst = (struct sockaddr *)tempbuf;
1542 * Now go through all the requested tables (fibs) and do the
1543 * requested action. Realistically, this will either be fib 0
1544 * for protocols that don't do multiple tables or all the
1545 * tables for those that do.
1547 for ( fibnum = startfib; fibnum <= endfib; fibnum++) {
1548 if (cmd == RTM_DELETE) {
1549 struct radix_node *rn;
1551 * Look up an rtentry that is in the routing tree and
1552 * contains the correct info.
1554 rnh = rt_tables_get_rnh(fibnum, dst->sa_family);
1556 /* this table doesn't exist but others might */
1558 RADIX_NODE_HEAD_RLOCK(rnh);
1559 rn = rnh->rnh_lookup(dst, netmask, rnh);
1561 if (rn_mpath_capable(rnh)) {
1568 * for interface route the
1569 * rt->rt_gateway is sockaddr_intf
1570 * for cloning ARP entries, so
1571 * rt_mpath_matchgate must use the
1574 rt = rt_mpath_matchgate(rt,
1581 error = (rn == NULL ||
1582 (rn->rn_flags & RNF_ROOT) ||
1583 RNTORT(rn)->rt_ifa != ifa);
1584 RADIX_NODE_HEAD_RUNLOCK(rnh);
1586 /* this is only an error if bad on ALL tables */
1591 * Do the actual request
1593 bzero((caddr_t)&info, sizeof(info));
1595 info.rti_flags = flags |
1596 (ifa->ifa_flags & ~IFA_RTSELF) | RTF_PINNED;
1597 info.rti_info[RTAX_DST] = dst;
1599 * doing this for compatibility reasons
1602 info.rti_info[RTAX_GATEWAY] =
1603 (struct sockaddr *)&null_sdl;
1605 info.rti_info[RTAX_GATEWAY] = ifa->ifa_addr;
1606 info.rti_info[RTAX_NETMASK] = netmask;
1607 error = rtrequest1_fib(cmd, &info, &rt, fibnum);
1609 if ((error == EEXIST) && (cmd == RTM_ADD)) {
1611 * Interface route addition failed.
1612 * Atomically delete current prefix generating
1613 * RTM_DELETE message, and retry adding
1616 rnh = rt_tables_get_rnh(fibnum, dst->sa_family);
1617 RADIX_NODE_HEAD_LOCK(rnh);
1619 /* Delete old prefix */
1620 info.rti_ifa = NULL;
1621 info.rti_flags = RTF_RNH_LOCKED;
1623 error = rtrequest1_fib(RTM_DELETE, &info, NULL, fibnum);
1626 info.rti_flags = flags | RTF_RNH_LOCKED |
1627 (ifa->ifa_flags & ~IFA_RTSELF) | RTF_PINNED;
1628 error = rtrequest1_fib(cmd, &info, &rt, fibnum);
1631 RADIX_NODE_HEAD_UNLOCK(rnh);
1635 if (error == 0 && rt != NULL) {
1637 * notify any listening routing agents of the change
1642 * in case address alias finds the first address
1643 * e.g. ifconfig bge0 192.0.2.246/24
1644 * e.g. ifconfig bge0 192.0.2.247/24
1645 * the address set in the route is 192.0.2.246
1646 * so we need to replace it with 192.0.2.247
1648 if (memcmp(rt->rt_ifa->ifa_addr,
1649 ifa->ifa_addr, ifa->ifa_addr->sa_len)) {
1650 ifa_free(rt->rt_ifa);
1652 rt->rt_ifp = ifa->ifa_ifp;
1657 * doing this for compatibility reasons
1659 if (cmd == RTM_ADD) {
1660 ((struct sockaddr_dl *)rt->rt_gateway)->sdl_type =
1661 rt->rt_ifp->if_type;
1662 ((struct sockaddr_dl *)rt->rt_gateway)->sdl_index =
1663 rt->rt_ifp->if_index;
1667 rt_newaddrmsg_fib(cmd, ifa, error, rt, fibnum);
1670 if (cmd == RTM_DELETE) {
1672 * If we are deleting, and we found an entry,
1673 * then it's been removed from the tree..
1674 * now throw it away.
1678 if (cmd == RTM_ADD) {
1680 * We just wanted to add it..
1681 * we don't actually need a reference.
1692 if (cmd == RTM_DELETE) {
1696 /* we only give an error if it wasn't in any table */
1697 error = ((flags & RTF_HOST) ?
1698 EHOSTUNREACH : ENETUNREACH);
1702 /* return an error if any of them failed */
1709 #ifndef BURN_BRIDGES
1710 /* special one for inet internal use. may not use. */
1712 rtinit_fib(struct ifaddr *ifa, int cmd, int flags)
1714 return (rtinit1(ifa, cmd, flags, RT_ALL_FIBS));
1719 * Set up a routing table entry, normally
1723 rtinit(struct ifaddr *ifa, int cmd, int flags)
1725 struct sockaddr *dst;
1726 int fib = RT_DEFAULT_FIB;
1728 if (flags & RTF_HOST) {
1729 dst = ifa->ifa_dstaddr;
1731 dst = ifa->ifa_addr;
1734 switch (dst->sa_family) {
1737 /* We do support multiple FIBs. */
1741 return (rtinit1(ifa, cmd, flags, fib));
1745 * Announce interface address arrival/withdraw
1746 * Returns 0 on success.
1749 rt_addrmsg(int cmd, struct ifaddr *ifa, int fibnum)
1752 KASSERT(cmd == RTM_ADD || cmd == RTM_DELETE,
1753 ("unexpected cmd %d", cmd));
1755 KASSERT(fibnum == RT_ALL_FIBS || (fibnum >= 0 && fibnum < rt_numfibs),
1756 ("%s: fib out of range 0 <=%d<%d", __func__, fibnum, rt_numfibs));
1758 return (rtsock_addrmsg(cmd, ifa, fibnum));
1762 * Announce route addition/removal.
1763 * Users of this function MUST validate input data BEFORE calling.
1764 * However we have to be able to handle invalid data:
1765 * if some userland app sends us "invalid" route message (invalid mask,
1766 * no dst, wrong address families, etc...) we need to pass it back
1767 * to app (and any other rtsock consumers) with rtm_errno field set to
1769 * Returns 0 on success.
1772 rt_routemsg(int cmd, struct ifnet *ifp, int error, struct rtentry *rt,
1776 KASSERT(cmd == RTM_ADD || cmd == RTM_DELETE,
1777 ("unexpected cmd %d", cmd));
1779 KASSERT(fibnum == RT_ALL_FIBS || (fibnum >= 0 && fibnum < rt_numfibs),
1780 ("%s: fib out of range 0 <=%d<%d", __func__, fibnum, rt_numfibs));
1782 KASSERT(rt_key(rt) != NULL, (":%s: rt_key must be supplied", __func__));
1784 return (rtsock_routemsg(cmd, ifp, error, rt, fibnum));
1788 rt_newaddrmsg(int cmd, struct ifaddr *ifa, int error, struct rtentry *rt)
1791 rt_newaddrmsg_fib(cmd, ifa, error, rt, RT_ALL_FIBS);
1795 * This is called to generate messages from the routing socket
1796 * indicating a network interface has had addresses associated with it.
1799 rt_newaddrmsg_fib(int cmd, struct ifaddr *ifa, int error, struct rtentry *rt,
1803 KASSERT(cmd == RTM_ADD || cmd == RTM_DELETE,
1804 ("unexpected cmd %u", cmd));
1805 KASSERT(fibnum == RT_ALL_FIBS || (fibnum >= 0 && fibnum < rt_numfibs),
1806 ("%s: fib out of range 0 <=%d<%d", __func__, fibnum, rt_numfibs));
1808 #if defined(INET) || defined(INET6)
1811 * notify the SCTP stack
1812 * this will only get called when an address is added/deleted
1813 * XXX pass the ifaddr struct instead if ifa->ifa_addr...
1815 sctp_addr_change(ifa, cmd);
1818 if (cmd == RTM_ADD) {
1819 rt_addrmsg(cmd, ifa, fibnum);
1821 rt_routemsg(cmd, ifa->ifa_ifp, error, rt, fibnum);
1824 rt_routemsg(cmd, ifa->ifa_ifp, error, rt, fibnum);
1825 rt_addrmsg(cmd, ifa, fibnum);