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 #define RT_MAXFIBS UINT16_MAX
74 /* Kernel config default option. */
77 #error "ROUTETABLES defined too low"
79 #if ROUTETABLES > RT_MAXFIBS
80 #error "ROUTETABLES defined too big"
82 #define RT_NUMFIBS ROUTETABLES
83 #endif /* ROUTETABLES */
84 /* Initialize to default if not otherwise set. */
89 #if defined(INET) || defined(INET6)
91 extern void sctp_addr_change(struct ifaddr *ifa, int cmd);
96 /* This is read-only.. */
97 u_int rt_numfibs = RT_NUMFIBS;
98 SYSCTL_UINT(_net, OID_AUTO, fibs, CTLFLAG_RD, &rt_numfibs, 0, "");
99 /* and this can be set too big but will be fixed before it is used */
100 TUNABLE_INT("net.fibs", &rt_numfibs);
103 * By default add routes to all fibs for new interfaces.
104 * Once this is set to 0 then only allocate routes on interface
105 * changes for the FIB of the caller when adding a new set of addresses
106 * to an interface. XXX this is a shotgun aproach to a problem that needs
107 * a more fine grained solution.. that will come.
108 * XXX also has the problems getting the FIB from curthread which will not
109 * always work given the fib can be overridden and prefixes can be added
110 * from the network stack context.
112 u_int rt_add_addr_allfibs = 1;
113 SYSCTL_UINT(_net, OID_AUTO, add_addr_allfibs, CTLFLAG_RW,
114 &rt_add_addr_allfibs, 0, "");
115 TUNABLE_INT("net.add_addr_allfibs", &rt_add_addr_allfibs);
117 VNET_DEFINE(struct rtstat, rtstat);
118 #define V_rtstat VNET(rtstat)
120 VNET_DEFINE(struct radix_node_head *, rt_tables);
121 #define V_rt_tables VNET(rt_tables)
123 VNET_DEFINE(int, rttrash); /* routes not in table but not freed */
124 #define V_rttrash VNET(rttrash)
127 /* compare two sockaddr structures */
128 #define sa_equal(a1, a2) (((a1)->sa_len == (a2)->sa_len) && \
129 (bcmp((a1), (a2), (a1)->sa_len) == 0))
132 * Convert a 'struct radix_node *' to a 'struct rtentry *'.
133 * The operation can be done safely (in this code) because a
134 * 'struct rtentry' starts with two 'struct radix_node''s, the first
135 * one representing leaf nodes in the routing tree, which is
136 * what the code in radix.c passes us as a 'struct radix_node'.
138 * But because there are a lot of assumptions in this conversion,
139 * do not cast explicitly, but always use the macro below.
141 #define RNTORT(p) ((struct rtentry *)(p))
143 static VNET_DEFINE(uma_zone_t, rtzone); /* Routing table UMA zone. */
144 #define V_rtzone VNET(rtzone)
147 * handler for net.my_fibnum
150 sysctl_my_fibnum(SYSCTL_HANDLER_ARGS)
155 fibnum = curthread->td_proc->p_fibnum;
156 error = sysctl_handle_int(oidp, &fibnum, 0, req);
160 SYSCTL_PROC(_net, OID_AUTO, my_fibnum, CTLTYPE_INT|CTLFLAG_RD,
161 NULL, 0, &sysctl_my_fibnum, "I", "default FIB of caller");
163 static __inline struct radix_node_head **
164 rt_tables_get_rnh_ptr(int table, int fam)
166 struct radix_node_head **rnh;
168 KASSERT(table >= 0 && table < rt_numfibs, ("%s: table out of bounds.",
170 KASSERT(fam >= 0 && fam < (AF_MAX+1), ("%s: fam out of bounds.",
173 /* rnh is [fib=0][af=0]. */
174 rnh = (struct radix_node_head **)V_rt_tables;
175 /* Get the offset to the requested table and fam. */
176 rnh += table * (AF_MAX+1) + fam;
181 struct radix_node_head *
182 rt_tables_get_rnh(int table, int fam)
185 return (*rt_tables_get_rnh_ptr(table, fam));
189 * route initialization must occur before ip6_init2(), which happenas at
198 /* whack the tunable ints into line. */
199 if (rt_numfibs > RT_MAXFIBS)
200 rt_numfibs = RT_MAXFIBS;
204 for (dom = domains; dom; dom = dom->dom_next)
205 if (dom->dom_maxrtkey > max_keylen)
206 max_keylen = dom->dom_maxrtkey;
208 rn_init(max_keylen); /* init all zeroes, all ones, mask table */
210 SYSINIT(route_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_THIRD, route_init, 0);
213 rtentry_zinit(void *mem, int size, int how)
215 struct rtentry *rt = mem;
217 rt->rt_pksent = counter_u64_alloc(how);
218 if (rt->rt_pksent == NULL)
227 rtentry_zfini(void *mem, int size)
229 struct rtentry *rt = mem;
232 counter_u64_free(rt->rt_pksent);
236 rtentry_ctor(void *mem, int size, void *arg, int how)
238 struct rtentry *rt = mem;
240 bzero(rt, offsetof(struct rtentry, rt_endzero));
241 counter_u64_zero(rt->rt_pksent);
247 rtentry_dtor(void *mem, int size, void *arg)
249 struct rtentry *rt = mem;
255 vnet_route_init(const void *unused __unused)
258 struct radix_node_head **rnh;
262 V_rt_tables = malloc(rt_numfibs * (AF_MAX+1) *
263 sizeof(struct radix_node_head *), M_RTABLE, M_WAITOK|M_ZERO);
265 V_rtzone = uma_zcreate("rtentry", sizeof(struct rtentry),
266 rtentry_ctor, rtentry_dtor,
267 rtentry_zinit, rtentry_zfini, UMA_ALIGN_PTR, 0);
268 for (dom = domains; dom; dom = dom->dom_next) {
269 if (dom->dom_rtattach == NULL)
272 for (table = 0; table < rt_numfibs; table++) {
273 fam = dom->dom_family;
274 if (table != 0 && fam != AF_INET6 && fam != AF_INET)
278 * XXX MRT rtattach will be also called from
279 * vfs_export.c but the offset will be 0 (only for
280 * AF_INET and AF_INET6 which don't need it anyhow).
282 rnh = rt_tables_get_rnh_ptr(table, fam);
284 panic("%s: rnh NULL", __func__);
285 dom->dom_rtattach((void **)rnh, dom->dom_rtoffset);
289 VNET_SYSINIT(vnet_route_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_FOURTH,
294 vnet_route_uninit(const void *unused __unused)
299 struct radix_node_head **rnh;
301 for (dom = domains; dom; dom = dom->dom_next) {
302 if (dom->dom_rtdetach == NULL)
305 for (table = 0; table < rt_numfibs; table++) {
306 fam = dom->dom_family;
308 if (table != 0 && fam != AF_INET6 && fam != AF_INET)
311 rnh = rt_tables_get_rnh_ptr(table, fam);
313 panic("%s: rnh NULL", __func__);
314 dom->dom_rtdetach((void **)rnh, dom->dom_rtoffset);
318 free(V_rt_tables, M_RTABLE);
319 uma_zdestroy(V_rtzone);
321 VNET_SYSUNINIT(vnet_route_uninit, SI_SUB_PROTO_DOMAIN, SI_ORDER_THIRD,
322 vnet_route_uninit, 0);
325 #ifndef _SYS_SYSPROTO_H_
331 sys_setfib(struct thread *td, struct setfib_args *uap)
333 if (uap->fibnum < 0 || uap->fibnum >= rt_numfibs)
335 td->td_proc->p_fibnum = uap->fibnum;
340 * Packet routing routines.
343 rtalloc(struct route *ro)
346 rtalloc_ign_fib(ro, 0UL, RT_DEFAULT_FIB);
350 rtalloc_fib(struct route *ro, u_int fibnum)
352 rtalloc_ign_fib(ro, 0UL, fibnum);
356 rtalloc_ign(struct route *ro, u_long ignore)
360 if ((rt = ro->ro_rt) != NULL) {
361 if (rt->rt_ifp != NULL && rt->rt_flags & RTF_UP)
366 ro->ro_rt = rtalloc1_fib(&ro->ro_dst, 1, ignore, RT_DEFAULT_FIB);
368 RT_UNLOCK(ro->ro_rt);
372 rtalloc_ign_fib(struct route *ro, u_long ignore, u_int fibnum)
376 if ((rt = ro->ro_rt) != NULL) {
377 if (rt->rt_ifp != NULL && rt->rt_flags & RTF_UP)
382 ro->ro_rt = rtalloc1_fib(&ro->ro_dst, 1, ignore, fibnum);
384 RT_UNLOCK(ro->ro_rt);
388 * Look up the route that matches the address given
389 * Or, at least try.. Create a cloned route if needed.
391 * The returned route, if any, is locked.
394 rtalloc1(struct sockaddr *dst, int report, u_long ignflags)
397 return (rtalloc1_fib(dst, report, ignflags, RT_DEFAULT_FIB));
401 rtalloc1_fib(struct sockaddr *dst, int report, u_long ignflags,
404 struct radix_node_head *rnh;
405 struct radix_node *rn;
406 struct rtentry *newrt;
407 struct rt_addrinfo info;
408 int err = 0, msgtype = RTM_MISS;
411 KASSERT((fibnum < rt_numfibs), ("rtalloc1_fib: bad fibnum"));
412 switch (dst->sa_family) {
415 /* We support multiple FIBs. */
418 fibnum = RT_DEFAULT_FIB;
421 rnh = rt_tables_get_rnh(fibnum, dst->sa_family);
427 * Look up the address in the table for that Address Family
429 needlock = !(ignflags & RTF_RNH_LOCKED);
431 RADIX_NODE_HEAD_RLOCK(rnh);
434 RADIX_NODE_HEAD_LOCK_ASSERT(rnh);
436 rn = rnh->rnh_matchaddr(dst, rnh);
437 if (rn && ((rn->rn_flags & RNF_ROOT) == 0)) {
442 RADIX_NODE_HEAD_RUNLOCK(rnh);
446 RADIX_NODE_HEAD_RUNLOCK(rnh);
449 * Either we hit the root or couldn't find any match,
450 * Which basically means
451 * "caint get there frm here"
454 V_rtstat.rts_unreach++;
458 * If required, report the failure to the supervising
460 * For a delete, this is not an error. (report == 0)
462 bzero(&info, sizeof(info));
463 info.rti_info[RTAX_DST] = dst;
464 rt_missmsg_fib(msgtype, &info, 0, err, fibnum);
468 RT_LOCK_ASSERT(newrt);
473 * Remove a reference count from an rtentry.
474 * If the count gets low enough, take it out of the routing table
477 rtfree(struct rtentry *rt)
479 struct radix_node_head *rnh;
481 KASSERT(rt != NULL,("%s: NULL rt", __func__));
482 rnh = rt_tables_get_rnh(rt->rt_fibnum, rt_key(rt)->sa_family);
483 KASSERT(rnh != NULL,("%s: NULL rnh", __func__));
488 * The callers should use RTFREE_LOCKED() or RTFREE(), so
489 * we should come here exactly with the last reference.
492 if (rt->rt_refcnt > 0) {
493 log(LOG_DEBUG, "%s: %p has %d refs\n", __func__, rt, rt->rt_refcnt);
498 * On last reference give the "close method" a chance
499 * to cleanup private state. This also permits (for
500 * IPv4 and IPv6) a chance to decide if the routing table
501 * entry should be purged immediately or at a later time.
502 * When an immediate purge is to happen the close routine
503 * typically calls rtexpunge which clears the RTF_UP flag
504 * on the entry so that the code below reclaims the storage.
506 if (rt->rt_refcnt == 0 && rnh->rnh_close)
507 rnh->rnh_close((struct radix_node *)rt, rnh);
510 * If we are no longer "up" (and ref == 0)
511 * then we can free the resources associated
514 if ((rt->rt_flags & RTF_UP) == 0) {
515 if (rt->rt_nodes->rn_flags & (RNF_ACTIVE | RNF_ROOT))
518 * the rtentry must have been removed from the routing table
519 * so it is represented in rttrash.. remove that now.
523 if (rt->rt_refcnt < 0) {
524 printf("rtfree: %p not freed (neg refs)\n", rt);
529 * release references on items we hold them on..
530 * e.g other routes and ifaddrs.
533 ifa_free(rt->rt_ifa);
535 * The key is separatly alloc'd so free it (see rt_setgate()).
536 * This also frees the gateway, as they are always malloc'd
542 * and the rtentry itself of course
544 uma_zfree(V_rtzone, rt);
553 * Force a routing table entry to the specified
554 * destination to go through the given gateway.
555 * Normally called as a result of a routing redirect
556 * message from the network layer.
559 rtredirect(struct sockaddr *dst,
560 struct sockaddr *gateway,
561 struct sockaddr *netmask,
563 struct sockaddr *src)
566 rtredirect_fib(dst, gateway, netmask, flags, src, RT_DEFAULT_FIB);
570 rtredirect_fib(struct sockaddr *dst,
571 struct sockaddr *gateway,
572 struct sockaddr *netmask,
574 struct sockaddr *src,
577 struct rtentry *rt, *rt0 = NULL;
580 struct rt_addrinfo info;
582 struct radix_node_head *rnh;
585 rnh = rt_tables_get_rnh(fibnum, dst->sa_family);
587 error = EAFNOSUPPORT;
591 /* verify the gateway is directly reachable */
592 if ((ifa = ifa_ifwithnet_fib(gateway, 0, fibnum)) == NULL) {
596 rt = rtalloc1_fib(dst, 0, 0UL, fibnum); /* NB: rt is locked */
598 * If the redirect isn't from our current router for this dst,
599 * it's either old or wrong. If it redirects us to ourselves,
600 * we have a routing loop, perhaps as a result of an interface
601 * going down recently.
603 if (!(flags & RTF_DONE) && rt &&
604 (!sa_equal(src, rt->rt_gateway) || rt->rt_ifa != ifa))
606 else if (ifa_ifwithaddr_check(gateway))
607 error = EHOSTUNREACH;
611 * Create a new entry if we just got back a wildcard entry
612 * or the lookup failed. This is necessary for hosts
613 * which use routing redirects generated by smart gateways
614 * to dynamically build the routing tables.
616 if (rt == NULL || (rt_mask(rt) && rt_mask(rt)->sa_len < 2))
619 * Don't listen to the redirect if it's
620 * for a route to an interface.
622 if (rt->rt_flags & RTF_GATEWAY) {
623 if (((rt->rt_flags & RTF_HOST) == 0) && (flags & RTF_HOST)) {
625 * Changing from route to net => route to host.
626 * Create new route, rather than smashing route to net.
632 flags |= RTF_GATEWAY | RTF_DYNAMIC;
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;
638 info.rti_flags = flags;
640 RT_UNLOCK(rt0); /* drop lock to avoid LOR with RNH */
641 error = rtrequest1_fib(RTM_ADD, &info, &rt, fibnum);
645 EVENTHANDLER_INVOKE(route_redirect_event, rt0, rt, dst);
646 flags = rt->rt_flags;
651 stat = &V_rtstat.rts_dynamic;
653 struct rtentry *gwrt;
656 * Smash the current notion of the gateway to
657 * this destination. Should check about netmask!!!
659 rt->rt_flags |= RTF_MODIFIED;
660 flags |= RTF_MODIFIED;
661 stat = &V_rtstat.rts_newgateway;
663 * add the key and gateway (in one malloc'd chunk).
666 RADIX_NODE_HEAD_LOCK(rnh);
668 rt_setgate(rt, rt_key(rt), gateway);
669 gwrt = rtalloc1(gateway, 1, RTF_RNH_LOCKED);
670 RADIX_NODE_HEAD_UNLOCK(rnh);
671 EVENTHANDLER_INVOKE(route_redirect_event, rt, gwrt, dst);
675 error = EHOSTUNREACH;
681 V_rtstat.rts_badredirect++;
682 else if (stat != NULL)
684 bzero((caddr_t)&info, sizeof(info));
685 info.rti_info[RTAX_DST] = dst;
686 info.rti_info[RTAX_GATEWAY] = gateway;
687 info.rti_info[RTAX_NETMASK] = netmask;
688 info.rti_info[RTAX_AUTHOR] = src;
689 rt_missmsg_fib(RTM_REDIRECT, &info, flags, error, fibnum);
695 rtioctl(u_long req, caddr_t data)
698 return (rtioctl_fib(req, data, RT_DEFAULT_FIB));
702 * Routing table ioctl interface.
705 rtioctl_fib(u_long req, caddr_t data, u_int fibnum)
709 * If more ioctl commands are added here, make sure the proper
710 * super-user checks are being performed because it is possible for
711 * prison-root to make it this far if raw sockets have been enabled
715 /* Multicast goop, grrr... */
716 return mrt_ioctl ? mrt_ioctl(req, data, fibnum) : EOPNOTSUPP;
723 * For both ifa_ifwithroute() routines, 'ifa' is returned referenced.
726 ifa_ifwithroute(int flags, struct sockaddr *dst, struct sockaddr *gateway)
729 return (ifa_ifwithroute_fib(flags, dst, gateway, RT_DEFAULT_FIB));
733 ifa_ifwithroute_fib(int flags, struct sockaddr *dst, struct sockaddr *gateway,
736 register struct ifaddr *ifa;
739 if ((flags & RTF_GATEWAY) == 0) {
741 * If we are adding a route to an interface,
742 * and the interface is a pt to pt link
743 * we should search for the destination
744 * as our clue to the interface. Otherwise
745 * we can use the local address.
748 if (flags & RTF_HOST)
749 ifa = ifa_ifwithdstaddr_fib(dst, fibnum);
751 ifa = ifa_ifwithaddr(gateway);
754 * If we are adding a route to a remote net
755 * or host, the gateway may still be on the
756 * other end of a pt to pt link.
758 ifa = ifa_ifwithdstaddr_fib(gateway, fibnum);
761 ifa = ifa_ifwithnet_fib(gateway, 0, fibnum);
763 struct rtentry *rt = rtalloc1_fib(gateway, 0, RTF_RNH_LOCKED, fibnum);
767 * dismiss a gateway that is reachable only
768 * through the default router
770 switch (gateway->sa_family) {
772 if (satosin(rt_key(rt))->sin_addr.s_addr == INADDR_ANY)
776 if (IN6_IS_ADDR_UNSPECIFIED(&satosin6(rt_key(rt))->sin6_addr))
782 if (!not_found && rt->rt_ifa != NULL) {
788 if (not_found || ifa == NULL)
791 if (ifa->ifa_addr->sa_family != dst->sa_family) {
792 struct ifaddr *oifa = ifa;
793 ifa = ifaof_ifpforaddr(dst, ifa->ifa_ifp);
803 * Do appropriate manipulations of a routing tree given
804 * all the bits of info needed
808 struct sockaddr *dst,
809 struct sockaddr *gateway,
810 struct sockaddr *netmask,
812 struct rtentry **ret_nrt)
815 return (rtrequest_fib(req, dst, gateway, netmask, flags, ret_nrt,
820 rtrequest_fib(int req,
821 struct sockaddr *dst,
822 struct sockaddr *gateway,
823 struct sockaddr *netmask,
825 struct rtentry **ret_nrt,
828 struct rt_addrinfo info;
830 if (dst->sa_len == 0)
833 bzero((caddr_t)&info, sizeof(info));
834 info.rti_flags = flags;
835 info.rti_info[RTAX_DST] = dst;
836 info.rti_info[RTAX_GATEWAY] = gateway;
837 info.rti_info[RTAX_NETMASK] = netmask;
838 return rtrequest1_fib(req, &info, ret_nrt, fibnum);
842 * These (questionable) definitions of apparent local variables apply
843 * to the next two functions. XXXXXX!!!
845 #define dst info->rti_info[RTAX_DST]
846 #define gateway info->rti_info[RTAX_GATEWAY]
847 #define netmask info->rti_info[RTAX_NETMASK]
848 #define ifaaddr info->rti_info[RTAX_IFA]
849 #define ifpaddr info->rti_info[RTAX_IFP]
850 #define flags info->rti_flags
853 rt_getifa(struct rt_addrinfo *info)
856 return (rt_getifa_fib(info, RT_DEFAULT_FIB));
860 * Look up rt_addrinfo for a specific fib. Note that if rti_ifa is defined,
861 * it will be referenced so the caller must free it.
864 rt_getifa_fib(struct rt_addrinfo *info, u_int fibnum)
870 * ifp may be specified by sockaddr_dl
871 * when protocol address is ambiguous.
873 if (info->rti_ifp == NULL && ifpaddr != NULL &&
874 ifpaddr->sa_family == AF_LINK &&
875 (ifa = ifa_ifwithnet_fib(ifpaddr, 0, fibnum)) != NULL) {
876 info->rti_ifp = ifa->ifa_ifp;
879 if (info->rti_ifa == NULL && ifaaddr != NULL)
880 info->rti_ifa = ifa_ifwithaddr(ifaaddr);
881 if (info->rti_ifa == NULL) {
884 sa = ifaaddr != NULL ? ifaaddr :
885 (gateway != NULL ? gateway : dst);
886 if (sa != NULL && info->rti_ifp != NULL)
887 info->rti_ifa = ifaof_ifpforaddr(sa, info->rti_ifp);
888 else if (dst != NULL && gateway != NULL)
889 info->rti_ifa = ifa_ifwithroute_fib(flags, dst, gateway,
892 info->rti_ifa = ifa_ifwithroute_fib(flags, sa, sa,
895 if ((ifa = info->rti_ifa) != NULL) {
896 if (info->rti_ifp == NULL)
897 info->rti_ifp = ifa->ifa_ifp;
904 * Expunges references to a route that's about to be reclaimed.
905 * The route must be locked.
908 rtexpunge(struct rtentry *rt)
910 #if !defined(RADIX_MPATH)
911 struct radix_node *rn;
913 struct rt_addrinfo info;
917 struct radix_node_head *rnh;
922 * Find the correct routing tree to use for this Address Family
924 rnh = rt_tables_get_rnh(rt->rt_fibnum, rt_key(rt)->sa_family);
927 return (EAFNOSUPPORT);
928 RADIX_NODE_HEAD_LOCK_ASSERT(rnh);
932 bzero(&info, sizeof(info));
933 info.rti_ifp = rt->rt_ifp;
934 info.rti_flags = RTF_RNH_LOCKED;
935 info.rti_info[RTAX_DST] = rt_key(rt);
936 info.rti_info[RTAX_GATEWAY] = rt->rt_ifa->ifa_addr;
939 error = rtrequest1_fib(RTM_DELETE, &info, &rt0, fib);
941 if (error == 0 && rt0 != NULL) {
944 } else if (error != 0) {
950 * Remove the item from the tree; it should be there,
951 * but when callers invoke us blindly it may not (sigh).
953 rn = rnh->rnh_deladdr(rt_key(rt), rt_mask(rt), rnh);
958 KASSERT((rn->rn_flags & (RNF_ACTIVE | RNF_ROOT)) == 0,
959 ("unexpected flags 0x%x", rn->rn_flags));
960 KASSERT(rt == RNTORT(rn),
961 ("lookup mismatch, rt %p rn %p", rt, rn));
962 #endif /* RADIX_MPATH */
964 rt->rt_flags &= ~RTF_UP;
967 * Give the protocol a chance to keep things in sync.
969 if ((ifa = rt->rt_ifa) && ifa->ifa_rtrequest) {
970 struct rt_addrinfo info;
972 bzero((caddr_t)&info, sizeof(info));
973 info.rti_flags = rt->rt_flags;
974 info.rti_info[RTAX_DST] = rt_key(rt);
975 info.rti_info[RTAX_GATEWAY] = rt->rt_gateway;
976 info.rti_info[RTAX_NETMASK] = rt_mask(rt);
977 ifa->ifa_rtrequest(RTM_DELETE, rt, &info);
981 * one more rtentry floating around that is not
982 * linked to the routing table.
985 #if !defined(RADIX_MPATH)
992 int p_sockaddr(char *buf, int buflen, struct sockaddr *s);
993 int rt_print(char *buf, int buflen, struct rtentry *rt);
996 p_sockaddr(char *buf, int buflen, struct sockaddr *s)
1000 switch (s->sa_family) {
1002 paddr = &((struct sockaddr_in *)s)->sin_addr;
1005 paddr = &((struct sockaddr_in6 *)s)->sin6_addr;
1012 if (inet_ntop(s->sa_family, paddr, buf, buflen) == NULL)
1015 return (strlen(buf));
1019 rt_print(char *buf, int buflen, struct rtentry *rt)
1021 struct sockaddr *addr, *mask;
1027 i = p_sockaddr(buf, buflen, addr);
1028 if (!(rt->rt_flags & RTF_HOST)) {
1030 i += p_sockaddr(buf + i, buflen - i, mask);
1033 if (rt->rt_flags & RTF_GATEWAY) {
1035 i += p_sockaddr(buf + i, buflen - i, rt->rt_gateway);
1044 rn_mpath_update(int req, struct rt_addrinfo *info,
1045 struct radix_node_head *rnh, struct rtentry **ret_nrt)
1048 * if we got multipath routes, we require users to specify
1049 * a matching RTAX_GATEWAY.
1051 struct rtentry *rt, *rto = NULL;
1052 register struct radix_node *rn;
1055 rn = rnh->rnh_lookup(dst, netmask, rnh);
1058 rto = rt = RNTORT(rn);
1060 rt = rt_mpath_matchgate(rt, gateway);
1064 * this is the first entry in the chain
1067 rn = rn_mpath_next((struct radix_node *)rt);
1069 * there is another entry, now it's active
1074 rto->rt_flags |= RTF_UP;
1076 } else if (rt->rt_flags & RTF_GATEWAY) {
1078 * For gateway routes, we need to
1079 * make sure that we we are deleting
1080 * the correct gateway.
1081 * rt_mpath_matchgate() does not
1082 * check the case when there is only
1083 * one route in the chain.
1086 (rt->rt_gateway->sa_len != gateway->sa_len ||
1087 memcmp(rt->rt_gateway, gateway, gateway->sa_len)))
1091 * remove from tree before returning it
1094 rn = rnh->rnh_deladdr(dst, netmask, rnh);
1095 KASSERT(rt == RNTORT(rn), ("radix node disappeared"));
1101 * use the normal delete code to remove
1104 if (req != RTM_DELETE)
1112 * if the entry is 2nd and on up
1114 if ((req == RTM_DELETE) && !rt_mpath_deldup(rto, rt))
1115 panic ("rtrequest1: rt_mpath_deldup");
1119 if (req == RTM_DELETE) {
1120 rt->rt_flags &= ~RTF_UP;
1122 * One more rtentry floating around that is not
1123 * linked to the routing table. rttrash will be decremented
1124 * when RTFREE(rt) is eventually called.
1130 if (req != RTM_DELETE)
1131 panic("unrecognized request %d", req);
1135 * If the caller wants it, then it can have it,
1136 * but it's up to it to free the rtentry as we won't be
1150 rtrequest1_fib(int req, struct rt_addrinfo *info, struct rtentry **ret_nrt,
1153 int error = 0, needlock = 0;
1154 register struct rtentry *rt;
1156 register struct rtentry *rt0;
1158 register struct radix_node *rn;
1159 register struct radix_node_head *rnh;
1161 struct sockaddr *ndst;
1162 struct sockaddr_storage mdst;
1163 #define senderr(x) { error = x ; goto bad; }
1165 KASSERT((fibnum < rt_numfibs), ("rtrequest1_fib: bad fibnum"));
1166 switch (dst->sa_family) {
1169 /* We support multiple FIBs. */
1172 fibnum = RT_DEFAULT_FIB;
1177 * Find the correct routing tree to use for this Address Family
1179 rnh = rt_tables_get_rnh(fibnum, dst->sa_family);
1181 return (EAFNOSUPPORT);
1182 needlock = ((flags & RTF_RNH_LOCKED) == 0);
1183 flags &= ~RTF_RNH_LOCKED;
1185 RADIX_NODE_HEAD_LOCK(rnh);
1187 RADIX_NODE_HEAD_LOCK_ASSERT(rnh);
1189 * If we are adding a host route then we don't want to put
1190 * a netmask in the tree, nor do we want to clone it.
1192 if (flags & RTF_HOST)
1198 rt_maskedcopy(dst, (struct sockaddr *)&mdst, netmask);
1199 dst = (struct sockaddr *)&mdst;
1202 if (rn_mpath_capable(rnh)) {
1203 error = rn_mpath_update(req, info, rnh, ret_nrt);
1205 * "bad" holds true for the success case
1208 if (error != ENOENT)
1213 if ((flags & RTF_PINNED) == 0) {
1214 /* Check if target route can be deleted */
1215 rt = (struct rtentry *)rnh->rnh_lookup(dst,
1217 if ((rt != NULL) && (rt->rt_flags & RTF_PINNED))
1218 senderr(EADDRINUSE);
1222 * Remove the item from the tree and return it.
1223 * Complain if it is not there and do no more processing.
1225 rn = rnh->rnh_deladdr(dst, netmask, rnh);
1228 if (rn->rn_flags & (RNF_ACTIVE | RNF_ROOT))
1229 panic ("rtrequest delete");
1233 rt->rt_flags &= ~RTF_UP;
1236 * give the protocol a chance to keep things in sync.
1238 if ((ifa = rt->rt_ifa) && ifa->ifa_rtrequest)
1239 ifa->ifa_rtrequest(RTM_DELETE, rt, info);
1242 * One more rtentry floating around that is not
1243 * linked to the routing table. rttrash will be decremented
1244 * when RTFREE(rt) is eventually called.
1249 * If the caller wants it, then it can have it,
1250 * but it's up to it to free the rtentry as we won't be
1261 * resolve was only used for route cloning
1266 if ((flags & RTF_GATEWAY) && !gateway)
1268 if (dst && gateway && (dst->sa_family != gateway->sa_family) &&
1269 (gateway->sa_family != AF_UNSPEC) && (gateway->sa_family != AF_LINK))
1272 if (info->rti_ifa == NULL) {
1273 error = rt_getifa_fib(info, fibnum);
1277 ifa_ref(info->rti_ifa);
1278 ifa = info->rti_ifa;
1279 rt = uma_zalloc(V_rtzone, M_NOWAIT);
1284 rt->rt_flags = RTF_UP | flags;
1285 rt->rt_fibnum = fibnum;
1287 * Add the gateway. Possibly re-malloc-ing the storage for it.
1290 if ((error = rt_setgate(rt, dst, gateway)) != 0) {
1292 uma_zfree(V_rtzone, rt);
1297 * point to the (possibly newly malloc'd) dest address.
1299 ndst = (struct sockaddr *)rt_key(rt);
1302 * make sure it contains the value we want (masked if needed).
1305 rt_maskedcopy(dst, ndst, netmask);
1307 bcopy(dst, ndst, dst->sa_len);
1310 * We use the ifa reference returned by rt_getifa_fib().
1311 * This moved from below so that rnh->rnh_addaddr() can
1312 * examine the ifa and ifa->ifa_ifp if it so desires.
1315 rt->rt_ifp = ifa->ifa_ifp;
1319 /* do not permit exactly the same dst/mask/gw pair */
1320 if (rn_mpath_capable(rnh) &&
1321 rt_mpath_conflict(rnh, rt, netmask)) {
1322 ifa_free(rt->rt_ifa);
1324 uma_zfree(V_rtzone, rt);
1331 /* "flow-table" only supports IPv6 and IPv4 at the moment. */
1332 switch (dst->sa_family) {
1339 #if defined(INET6) || defined(INET)
1340 rn = rnh->rnh_matchaddr(dst, rnh);
1341 if (rn && ((rn->rn_flags & RNF_ROOT) == 0)) {
1342 struct sockaddr *mask;
1347 * compare mask to see if the new route is
1348 * more specific than the existing one
1355 * A host route is already present, so
1356 * leave the flow-table entries as is.
1358 if (rt0->rt_flags & RTF_HOST) {
1361 } else if (!(flags & RTF_HOST) && netmask) {
1362 mask = rt_mask(rt0);
1365 n = (u_char *)netmask;
1372 if (len == 0 || (*n < *m)) {
1378 #endif/* INET6 || INET */
1380 #endif /* FLOWTABLE */
1382 /* XXX mtu manipulation will be done in rnh_addaddr -- itojun */
1383 rn = rnh->rnh_addaddr(ndst, netmask, rnh, rt->rt_nodes);
1385 * If it still failed to go into the tree,
1386 * then un-make it (this should be a function)
1389 ifa_free(rt->rt_ifa);
1391 uma_zfree(V_rtzone, rt);
1399 else if (rt0 != NULL) {
1400 flowtable_route_flush(dst->sa_family, rt0);
1406 * If this protocol has something to add to this then
1407 * allow it to do that as well.
1409 if (ifa->ifa_rtrequest)
1410 ifa->ifa_rtrequest(req, rt, info);
1413 * actually return a resultant rtentry and
1414 * give the caller a single reference.
1427 RADIX_NODE_HEAD_UNLOCK(rnh);
1440 rt_setgate(struct rtentry *rt, struct sockaddr *dst, struct sockaddr *gate)
1442 /* XXX dst may be overwritten, can we move this to below */
1443 int dlen = SA_SIZE(dst), glen = SA_SIZE(gate);
1445 struct radix_node_head *rnh;
1447 rnh = rt_tables_get_rnh(rt->rt_fibnum, dst->sa_family);
1451 RADIX_NODE_HEAD_LOCK_ASSERT(rnh);
1454 * Prepare to store the gateway in rt->rt_gateway.
1455 * Both dst and gateway are stored one after the other in the same
1456 * malloc'd chunk. If we have room, we can reuse the old buffer,
1457 * rt_gateway already points to the right place.
1458 * Otherwise, malloc a new block and update the 'dst' address.
1460 if (rt->rt_gateway == NULL || glen > SA_SIZE(rt->rt_gateway)) {
1463 R_Malloc(new, caddr_t, dlen + glen);
1467 * XXX note, we copy from *dst and not *rt_key(rt) because
1468 * rt_setgate() can be called to initialize a newly
1469 * allocated route entry, in which case rt_key(rt) == NULL
1470 * (and also rt->rt_gateway == NULL).
1471 * Free()/free() handle a NULL argument just fine.
1473 bcopy(dst, new, dlen);
1474 Free(rt_key(rt)); /* free old block, if any */
1475 rt_key(rt) = (struct sockaddr *)new;
1476 rt->rt_gateway = (struct sockaddr *)(new + dlen);
1480 * Copy the new gateway value into the memory chunk.
1482 bcopy(gate, rt->rt_gateway, glen);
1488 rt_maskedcopy(struct sockaddr *src, struct sockaddr *dst, struct sockaddr *netmask)
1490 register u_char *cp1 = (u_char *)src;
1491 register u_char *cp2 = (u_char *)dst;
1492 register u_char *cp3 = (u_char *)netmask;
1493 u_char *cplim = cp2 + *cp3;
1494 u_char *cplim2 = cp2 + *cp1;
1496 *cp2++ = *cp1++; *cp2++ = *cp1++; /* copies sa_len & sa_family */
1501 *cp2++ = *cp1++ & *cp3++;
1503 bzero((caddr_t)cp2, (unsigned)(cplim2 - cp2));
1507 * Set up a routing table entry, normally
1510 #define _SOCKADDR_TMPSIZE 128 /* Not too big.. kernel stack size is limited */
1512 rtinit1(struct ifaddr *ifa, int cmd, int flags, int fibnum)
1514 struct sockaddr *dst;
1515 struct sockaddr *netmask;
1516 struct rtentry *rt = NULL;
1517 struct rt_addrinfo info;
1519 int startfib, endfib;
1520 char tempbuf[_SOCKADDR_TMPSIZE];
1523 static struct sockaddr_dl null_sdl = {sizeof(null_sdl), AF_LINK};
1524 struct radix_node_head *rnh;
1526 if (flags & RTF_HOST) {
1527 dst = ifa->ifa_dstaddr;
1530 dst = ifa->ifa_addr;
1531 netmask = ifa->ifa_netmask;
1533 if (dst->sa_len == 0)
1535 switch (dst->sa_family) {
1538 /* We support multiple FIBs. */
1541 fibnum = RT_DEFAULT_FIB;
1544 if (fibnum == RT_ALL_FIBS) {
1545 if (rt_add_addr_allfibs == 0 && cmd == (int)RTM_ADD) {
1546 startfib = endfib = ifa->ifa_ifp->if_fib;
1549 endfib = rt_numfibs - 1;
1552 KASSERT((fibnum < rt_numfibs), ("rtinit1: bad fibnum"));
1558 * If it's a delete, check that if it exists,
1559 * it's on the correct interface or we might scrub
1560 * a route to another ifa which would
1561 * be confusing at best and possibly worse.
1563 if (cmd == RTM_DELETE) {
1565 * It's a delete, so it should already exist..
1566 * If it's a net, mask off the host bits
1567 * (Assuming we have a mask)
1568 * XXX this is kinda inet specific..
1570 if (netmask != NULL) {
1571 rt_maskedcopy(dst, (struct sockaddr *)tempbuf, netmask);
1572 dst = (struct sockaddr *)tempbuf;
1576 * Now go through all the requested tables (fibs) and do the
1577 * requested action. Realistically, this will either be fib 0
1578 * for protocols that don't do multiple tables or all the
1579 * tables for those that do.
1581 for ( fibnum = startfib; fibnum <= endfib; fibnum++) {
1582 if (cmd == RTM_DELETE) {
1583 struct radix_node *rn;
1585 * Look up an rtentry that is in the routing tree and
1586 * contains the correct info.
1588 rnh = rt_tables_get_rnh(fibnum, dst->sa_family);
1590 /* this table doesn't exist but others might */
1592 RADIX_NODE_HEAD_RLOCK(rnh);
1593 rn = rnh->rnh_lookup(dst, netmask, rnh);
1595 if (rn_mpath_capable(rnh)) {
1602 * for interface route the
1603 * rt->rt_gateway is sockaddr_intf
1604 * for cloning ARP entries, so
1605 * rt_mpath_matchgate must use the
1608 rt = rt_mpath_matchgate(rt,
1615 error = (rn == NULL ||
1616 (rn->rn_flags & RNF_ROOT) ||
1617 RNTORT(rn)->rt_ifa != ifa);
1618 RADIX_NODE_HEAD_RUNLOCK(rnh);
1620 /* this is only an error if bad on ALL tables */
1625 * Do the actual request
1627 bzero((caddr_t)&info, sizeof(info));
1629 info.rti_flags = flags |
1630 (ifa->ifa_flags & ~IFA_RTSELF) | RTF_PINNED;
1631 info.rti_info[RTAX_DST] = dst;
1633 * doing this for compatibility reasons
1636 info.rti_info[RTAX_GATEWAY] =
1637 (struct sockaddr *)&null_sdl;
1639 info.rti_info[RTAX_GATEWAY] = ifa->ifa_addr;
1640 info.rti_info[RTAX_NETMASK] = netmask;
1641 error = rtrequest1_fib(cmd, &info, &rt, fibnum);
1643 if ((error == EEXIST) && (cmd == RTM_ADD)) {
1645 * Interface route addition failed.
1646 * Atomically delete current prefix generating
1647 * RTM_DELETE message, and retry adding
1650 rnh = rt_tables_get_rnh(fibnum, dst->sa_family);
1651 RADIX_NODE_HEAD_LOCK(rnh);
1653 /* Delete old prefix */
1654 info.rti_ifa = NULL;
1655 info.rti_flags = RTF_RNH_LOCKED;
1657 error = rtrequest1_fib(RTM_DELETE, &info, NULL, fibnum);
1660 info.rti_flags = flags | RTF_RNH_LOCKED |
1661 (ifa->ifa_flags & ~IFA_RTSELF) | RTF_PINNED;
1662 error = rtrequest1_fib(cmd, &info, &rt, fibnum);
1665 RADIX_NODE_HEAD_UNLOCK(rnh);
1669 if (error == 0 && rt != NULL) {
1671 * notify any listening routing agents of the change
1676 * in case address alias finds the first address
1677 * e.g. ifconfig bge0 192.0.2.246/24
1678 * e.g. ifconfig bge0 192.0.2.247/24
1679 * the address set in the route is 192.0.2.246
1680 * so we need to replace it with 192.0.2.247
1682 if (memcmp(rt->rt_ifa->ifa_addr,
1683 ifa->ifa_addr, ifa->ifa_addr->sa_len)) {
1684 ifa_free(rt->rt_ifa);
1686 rt->rt_ifp = ifa->ifa_ifp;
1691 * doing this for compatibility reasons
1693 if (cmd == RTM_ADD) {
1694 ((struct sockaddr_dl *)rt->rt_gateway)->sdl_type =
1695 rt->rt_ifp->if_type;
1696 ((struct sockaddr_dl *)rt->rt_gateway)->sdl_index =
1697 rt->rt_ifp->if_index;
1701 rt_newaddrmsg_fib(cmd, ifa, error, rt, fibnum);
1704 if (cmd == RTM_DELETE) {
1706 * If we are deleting, and we found an entry,
1707 * then it's been removed from the tree..
1708 * now throw it away.
1712 if (cmd == RTM_ADD) {
1714 * We just wanted to add it..
1715 * we don't actually need a reference.
1726 if (cmd == RTM_DELETE) {
1730 /* we only give an error if it wasn't in any table */
1731 error = ((flags & RTF_HOST) ?
1732 EHOSTUNREACH : ENETUNREACH);
1736 /* return an error if any of them failed */
1744 * Set up a routing table entry, normally
1748 rtinit(struct ifaddr *ifa, int cmd, int flags)
1750 struct sockaddr *dst;
1751 int fib = RT_DEFAULT_FIB;
1753 if (flags & RTF_HOST) {
1754 dst = ifa->ifa_dstaddr;
1756 dst = ifa->ifa_addr;
1759 switch (dst->sa_family) {
1762 /* We do support multiple FIBs. */
1766 return (rtinit1(ifa, cmd, flags, fib));
1770 * Announce interface address arrival/withdraw
1771 * Returns 0 on success.
1774 rt_addrmsg(int cmd, struct ifaddr *ifa, int fibnum)
1777 KASSERT(cmd == RTM_ADD || cmd == RTM_DELETE,
1778 ("unexpected cmd %d", cmd));
1780 KASSERT(fibnum == RT_ALL_FIBS || (fibnum >= 0 && fibnum < rt_numfibs),
1781 ("%s: fib out of range 0 <=%d<%d", __func__, fibnum, rt_numfibs));
1783 #if defined(INET) || defined(INET6)
1786 * notify the SCTP stack
1787 * this will only get called when an address is added/deleted
1788 * XXX pass the ifaddr struct instead if ifa->ifa_addr...
1790 sctp_addr_change(ifa, cmd);
1793 return (rtsock_addrmsg(cmd, ifa, fibnum));
1797 * Announce route addition/removal.
1798 * Users of this function MUST validate input data BEFORE calling.
1799 * However we have to be able to handle invalid data:
1800 * if some userland app sends us "invalid" route message (invalid mask,
1801 * no dst, wrong address families, etc...) we need to pass it back
1802 * to app (and any other rtsock consumers) with rtm_errno field set to
1804 * Returns 0 on success.
1807 rt_routemsg(int cmd, struct ifnet *ifp, int error, struct rtentry *rt,
1811 KASSERT(cmd == RTM_ADD || cmd == RTM_DELETE,
1812 ("unexpected cmd %d", cmd));
1814 KASSERT(fibnum == RT_ALL_FIBS || (fibnum >= 0 && fibnum < rt_numfibs),
1815 ("%s: fib out of range 0 <=%d<%d", __func__, fibnum, rt_numfibs));
1817 KASSERT(rt_key(rt) != NULL, (":%s: rt_key must be supplied", __func__));
1819 return (rtsock_routemsg(cmd, ifp, error, rt, fibnum));
1823 rt_newaddrmsg(int cmd, struct ifaddr *ifa, int error, struct rtentry *rt)
1826 rt_newaddrmsg_fib(cmd, ifa, error, rt, RT_ALL_FIBS);
1830 * This is called to generate messages from the routing socket
1831 * indicating a network interface has had addresses associated with it.
1834 rt_newaddrmsg_fib(int cmd, struct ifaddr *ifa, int error, struct rtentry *rt,
1838 KASSERT(cmd == RTM_ADD || cmd == RTM_DELETE,
1839 ("unexpected cmd %u", cmd));
1840 KASSERT(fibnum == RT_ALL_FIBS || (fibnum >= 0 && fibnum < rt_numfibs),
1841 ("%s: fib out of range 0 <=%d<%d", __func__, fibnum, rt_numfibs));
1843 if (cmd == RTM_ADD) {
1844 rt_addrmsg(cmd, ifa, fibnum);
1846 rt_routemsg(cmd, ifa->ifa_ifp, error, rt, fibnum);
1849 rt_routemsg(cmd, ifa->ifa_ifp, error, rt, fibnum);
1850 rt_addrmsg(cmd, ifa, fibnum);