2 * Copyright (c) 2004, 2005 The DragonFly Project. All rights reserved.
4 * This code is derived from software contributed to The DragonFly Project
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
10 * 1. Redistributions of source code must retain the above copyright
11 * notice, this list of conditions and the following disclaimer.
12 * 2. Redistributions in binary form must reproduce the above copyright
13 * notice, this list of conditions and the following disclaimer in the
14 * documentation and/or other materials provided with the distribution.
15 * 3. Neither the name of The DragonFly Project nor the names of its
16 * contributors may be used to endorse or promote products derived
17 * from this software without specific, prior written permission.
19 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
20 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
21 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
22 * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
23 * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
24 * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
25 * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
26 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
27 * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
28 * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
29 * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34 * Copyright (c) 1980, 1986, 1991, 1993
35 * The Regents of the University of California. All rights reserved.
37 * Redistribution and use in source and binary forms, with or without
38 * modification, are permitted provided that the following conditions
40 * 1. Redistributions of source code must retain the above copyright
41 * notice, this list of conditions and the following disclaimer.
42 * 2. Redistributions in binary form must reproduce the above copyright
43 * notice, this list of conditions and the following disclaimer in the
44 * documentation and/or other materials provided with the distribution.
45 * 3. Neither the name of the University nor the names of its contributors
46 * may be used to endorse or promote products derived from this software
47 * without specific prior written permission.
49 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
50 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
51 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
52 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
53 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
54 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
55 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
56 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
57 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
58 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
61 * @(#)route.c 8.3 (Berkeley) 1/9/95
62 * $FreeBSD: src/sys/net/route.c,v 1.59.2.10 2003/01/17 08:04:00 ru Exp $
68 #include <sys/param.h>
69 #include <sys/systm.h>
70 #include <sys/malloc.h>
72 #include <sys/socket.h>
73 #include <sys/domain.h>
74 #include <sys/kernel.h>
75 #include <sys/sysctl.h>
76 #include <sys/globaldata.h>
77 #include <sys/thread.h>
80 #include <net/route.h>
81 #include <net/netisr.h>
83 #include <netinet/in.h>
84 #include <net/ip_mroute/ip_mroute.h>
86 #include <sys/thread2.h>
87 #include <sys/msgport2.h>
88 #include <net/netmsg2.h>
91 #include <netproto/mpls/mpls.h>
94 static struct rtstatistics rtstatistics_percpu[MAXCPU];
95 #define rtstat rtstatistics_percpu[mycpuid]
97 struct radix_node_head *rt_tables[MAXCPU][AF_MAX+1];
98 struct lwkt_port *rt_ports[MAXCPU];
100 static void rt_maskedcopy (struct sockaddr *, struct sockaddr *,
102 static void rtable_init(void);
103 static void rtable_service_loop(void *dummy);
104 static void rtinit_rtrequest_callback(int, int, struct rt_addrinfo *,
105 struct rtentry *, void *);
107 static void rtredirect_msghandler(netmsg_t msg);
108 static void rtrequest1_msghandler(netmsg_t msg);
109 static void rtsearch_msghandler(netmsg_t msg);
110 static void rtmask_add_msghandler(netmsg_t msg);
112 static int rt_setshims(struct rtentry *, struct sockaddr **);
114 SYSCTL_NODE(_net, OID_AUTO, route, CTLFLAG_RW, 0, "Routing");
117 static int route_debug = 1;
118 SYSCTL_INT(_net_route, OID_AUTO, route_debug, CTLFLAG_RW,
119 &route_debug, 0, "");
122 int route_assert_owner_access = 1;
123 SYSCTL_INT(_net_route, OID_AUTO, assert_owner_access, CTLFLAG_RW,
124 &route_assert_owner_access, 0, "");
126 u_long route_kmalloc_limit = 0;
127 TUNABLE_ULONG("net.route.kmalloc_limit", &route_kmalloc_limit);
130 * Initialize the route table(s) for protocol domains and
131 * create a helper thread which will be responsible for updating
132 * route table entries on each cpu.
140 for (cpu = 0; cpu < ncpus; ++cpu)
141 bzero(&rtstatistics_percpu[cpu], sizeof(struct rtstatistics));
142 rn_init(); /* initialize all zeroes, all ones, mask table */
143 rtable_init(); /* call dom_rtattach() on each cpu */
145 for (cpu = 0; cpu < ncpus; cpu++) {
146 lwkt_create(rtable_service_loop, NULL, &rtd, NULL,
147 0, cpu, "rtable_cpu %d", cpu);
148 rt_ports[cpu] = &rtd->td_msgport;
151 if (route_kmalloc_limit)
152 kmalloc_raise_limit(M_RTABLE, route_kmalloc_limit);
156 rtable_init_oncpu(netmsg_t msg)
161 SLIST_FOREACH(dom, &domains, dom_next) {
162 if (dom->dom_rtattach) {
164 (void **)&rt_tables[cpu][dom->dom_family],
168 ifnet_forwardmsg(&msg->lmsg, cpu + 1);
174 struct netmsg_base msg;
176 netmsg_init(&msg, NULL, &curthread->td_msgport, 0, rtable_init_oncpu);
177 ifnet_domsg(&msg.lmsg, 0);
181 * Our per-cpu table management protocol thread. All route table operations
182 * are sequentially chained through all cpus starting at cpu #0 in order to
183 * maintain duplicate route tables on each cpu. Having a spearate route
184 * table management thread allows the protocol and interrupt threads to
185 * issue route table changes.
188 rtable_service_loop(void *dummy __unused)
191 thread_t td = curthread;
193 while ((msg = lwkt_waitport(&td->td_msgport, 0)) != NULL) {
194 msg->nm_dispatch((netmsg_t)msg);
199 * Routing statistics.
202 sysctl_rtstatistics(SYSCTL_HANDLER_ARGS)
206 for (cpu = 0; cpu < ncpus; ++cpu) {
207 if ((error = SYSCTL_OUT(req, &rtstatistics_percpu[cpu],
208 sizeof(struct rtstatistics))))
210 if ((error = SYSCTL_IN(req, &rtstatistics_percpu[cpu],
211 sizeof(struct rtstatistics))))
217 SYSCTL_PROC(_net_route, OID_AUTO, stats, (CTLTYPE_OPAQUE|CTLFLAG_RW),
218 0, 0, sysctl_rtstatistics, "S,rtstatistics", "Routing statistics");
221 * Packet routing routines.
225 * Look up and fill in the "ro_rt" rtentry field in a route structure given
226 * an address in the "ro_dst" field. Always send a report on a miss and
227 * always clone routes.
230 rtalloc(struct route *ro)
232 rtalloc_ign(ro, 0UL);
236 * Look up and fill in the "ro_rt" rtentry field in a route structure given
237 * an address in the "ro_dst" field. Always send a report on a miss and
238 * optionally clone routes when RTF_CLONING or RTF_PRCLONING are not being
242 rtalloc_ign(struct route *ro, u_long ignoreflags)
244 if (ro->ro_rt != NULL) {
245 if (ro->ro_rt->rt_ifp != NULL && ro->ro_rt->rt_flags & RTF_UP)
250 ro->ro_rt = _rtlookup(&ro->ro_dst, RTL_REPORTMSG, ignoreflags);
254 * Look up the route that matches the given "dst" address.
256 * Route lookup can have the side-effect of creating and returning
257 * a cloned route instead when "dst" matches a cloning route and the
258 * RTF_CLONING and RTF_PRCLONING flags are not being ignored.
260 * Any route returned has its reference count incremented.
263 _rtlookup(struct sockaddr *dst, boolean_t generate_report, u_long ignore)
265 struct radix_node_head *rnh = rt_tables[mycpuid][dst->sa_family];
272 * Look up route in the radix tree.
274 rt = (struct rtentry *) rnh->rnh_matchaddr((char *)dst, rnh);
279 * Handle cloning routes.
281 if ((rt->rt_flags & ~ignore & (RTF_CLONING | RTF_PRCLONING)) != 0) {
282 struct rtentry *clonedroute;
285 clonedroute = rt; /* copy in/copy out parameter */
286 error = rtrequest(RTM_RESOLVE, dst, NULL, NULL, 0,
287 &clonedroute); /* clone the route */
288 if (error != 0) { /* cloning failed */
290 rt_dstmsg(RTM_MISS, dst, error);
292 return (rt); /* return the uncloned route */
294 if (generate_report) {
295 if (clonedroute->rt_flags & RTF_XRESOLVE)
296 rt_dstmsg(RTM_RESOLVE, dst, 0);
298 rt_rtmsg(RTM_ADD, clonedroute,
299 clonedroute->rt_ifp, 0);
301 return (clonedroute); /* return cloned route */
305 * Increment the reference count of the matched route and return.
311 rtstat.rts_unreach++;
313 rt_dstmsg(RTM_MISS, dst, 0);
318 rtfree(struct rtentry *rt)
320 if (rt->rt_cpuid == mycpuid)
327 rtfree_oncpu(struct rtentry *rt)
329 KKASSERT(rt->rt_cpuid == mycpuid);
330 KASSERT(rt->rt_refcnt > 0, ("rtfree: rt_refcnt %ld", rt->rt_refcnt));
333 if (rt->rt_refcnt == 0) {
334 struct radix_node_head *rnh =
335 rt_tables[mycpuid][rt_key(rt)->sa_family];
338 rnh->rnh_close((struct radix_node *)rt, rnh);
339 if (!(rt->rt_flags & RTF_UP)) {
340 /* deallocate route */
341 if (rt->rt_ifa != NULL)
343 if (rt->rt_parent != NULL)
344 RTFREE(rt->rt_parent); /* recursive call! */
352 rtfree_remote_dispatch(netmsg_t msg)
354 struct lwkt_msg *lmsg = &msg->lmsg;
355 struct rtentry *rt = lmsg->u.ms_resultp;
358 lwkt_replymsg(lmsg, 0);
362 rtfree_remote(struct rtentry *rt)
364 struct netmsg_base msg;
365 struct lwkt_msg *lmsg;
367 KKASSERT(rt->rt_cpuid != mycpuid);
369 if (route_assert_owner_access) {
370 panic("rt remote free rt_cpuid %d, mycpuid %d",
371 rt->rt_cpuid, mycpuid);
373 kprintf("rt remote free rt_cpuid %d, mycpuid %d\n",
374 rt->rt_cpuid, mycpuid);
378 netmsg_init(&msg, NULL, &curthread->td_msgport,
379 0, rtfree_remote_dispatch);
381 lmsg->u.ms_resultp = rt;
383 lwkt_domsg(rtable_portfn(rt->rt_cpuid), lmsg, 0);
387 rtredirect_oncpu(struct sockaddr *dst, struct sockaddr *gateway,
388 struct sockaddr *netmask, int flags, struct sockaddr *src)
390 struct rtentry *rt = NULL;
391 struct rt_addrinfo rtinfo;
396 /* verify the gateway is directly reachable */
397 if ((ifa = ifa_ifwithnet(gateway)) == NULL) {
403 * If the redirect isn't from our current router for this destination,
404 * it's either old or wrong.
406 if (!(flags & RTF_DONE) && /* XXX JH */
407 (rt = rtpurelookup(dst)) != NULL &&
408 (!sa_equal(src, rt->rt_gateway) || rt->rt_ifa != ifa)) {
414 * If it redirects us to ourselves, we have a routing loop,
415 * perhaps as a result of an interface going down recently.
417 if (ifa_ifwithaddr(gateway)) {
418 error = EHOSTUNREACH;
423 * Create a new entry if the lookup failed or if we got back
424 * a wildcard entry for the default route. This is necessary
425 * for hosts which use routing redirects generated by smart
426 * gateways to dynamically build the routing tables.
430 if ((rt_mask(rt) != NULL && rt_mask(rt)->sa_len < 2)) {
435 /* Ignore redirects for directly connected hosts. */
436 if (!(rt->rt_flags & RTF_GATEWAY)) {
437 error = EHOSTUNREACH;
441 if (!(rt->rt_flags & RTF_HOST) && (flags & RTF_HOST)) {
443 * Changing from a network route to a host route.
444 * Create a new host route rather than smashing the
448 flags |= RTF_GATEWAY | RTF_DYNAMIC;
449 bzero(&rtinfo, sizeof(struct rt_addrinfo));
450 rtinfo.rti_info[RTAX_DST] = dst;
451 rtinfo.rti_info[RTAX_GATEWAY] = gateway;
452 rtinfo.rti_info[RTAX_NETMASK] = netmask;
453 rtinfo.rti_flags = flags;
454 rtinfo.rti_ifa = ifa;
455 rt = NULL; /* copy-in/copy-out parameter */
456 error = rtrequest1(RTM_ADD, &rtinfo, &rt);
458 flags = rt->rt_flags;
459 stat = &rtstat.rts_dynamic;
462 * Smash the current notion of the gateway to this destination.
463 * Should check about netmask!!!
465 rt->rt_flags |= RTF_MODIFIED;
466 flags |= RTF_MODIFIED;
468 /* We only need to report rtmsg on CPU0 */
469 rt_setgate(rt, rt_key(rt), gateway,
470 mycpuid == 0 ? RTL_REPORTMSG : RTL_DONTREPORT);
472 stat = &rtstat.rts_newgateway;
480 rtstat.rts_badredirect++;
481 else if (stat != NULL)
487 struct netmsg_rtredirect {
488 struct netmsg_base base;
489 struct sockaddr *dst;
490 struct sockaddr *gateway;
491 struct sockaddr *netmask;
493 struct sockaddr *src;
497 * Force a routing table entry to the specified
498 * destination to go through the given gateway.
499 * Normally called as a result of a routing redirect
500 * message from the network layer.
502 * N.B.: must be called at splnet
505 rtredirect(struct sockaddr *dst, struct sockaddr *gateway,
506 struct sockaddr *netmask, int flags, struct sockaddr *src)
508 struct rt_addrinfo rtinfo;
510 struct netmsg_rtredirect msg;
512 netmsg_init(&msg.base, NULL, &curthread->td_msgport,
513 0, rtredirect_msghandler);
515 msg.gateway = gateway;
516 msg.netmask = netmask;
519 error = lwkt_domsg(rtable_portfn(0), &msg.base.lmsg, 0);
520 bzero(&rtinfo, sizeof(struct rt_addrinfo));
521 rtinfo.rti_info[RTAX_DST] = dst;
522 rtinfo.rti_info[RTAX_GATEWAY] = gateway;
523 rtinfo.rti_info[RTAX_NETMASK] = netmask;
524 rtinfo.rti_info[RTAX_AUTHOR] = src;
525 rt_missmsg(RTM_REDIRECT, &rtinfo, flags, error);
529 rtredirect_msghandler(netmsg_t msg)
531 struct netmsg_rtredirect *rmsg = (void *)msg;
534 rtredirect_oncpu(rmsg->dst, rmsg->gateway, rmsg->netmask,
535 rmsg->flags, rmsg->src);
536 nextcpu = mycpuid + 1;
538 lwkt_forwardmsg(rtable_portfn(nextcpu), &msg->lmsg);
540 lwkt_replymsg(&msg->lmsg, 0);
544 * Routing table ioctl interface.
547 rtioctl(u_long req, caddr_t data, struct ucred *cred)
550 /* Multicast goop, grrr... */
551 return mrt_ioctl ? mrt_ioctl(req, data) : EOPNOTSUPP;
558 ifa_ifwithroute(int flags, struct sockaddr *dst, struct sockaddr *gateway)
562 if (!(flags & RTF_GATEWAY)) {
564 * If we are adding a route to an interface,
565 * and the interface is a point-to-point link,
566 * we should search for the destination
567 * as our clue to the interface. Otherwise
568 * we can use the local address.
571 if (flags & RTF_HOST) {
572 ifa = ifa_ifwithdstaddr(dst);
575 ifa = ifa_ifwithaddr(gateway);
578 * If we are adding a route to a remote net
579 * or host, the gateway may still be on the
580 * other end of a pt to pt link.
582 ifa = ifa_ifwithdstaddr(gateway);
585 ifa = ifa_ifwithnet(gateway);
589 rt = rtpurelookup(gateway);
593 if ((ifa = rt->rt_ifa) == NULL)
596 if (ifa->ifa_addr->sa_family != dst->sa_family) {
597 struct ifaddr *oldifa = ifa;
599 ifa = ifaof_ifpforaddr(dst, ifa->ifa_ifp);
606 static int rt_fixdelete (struct radix_node *, void *);
607 static int rt_fixchange (struct radix_node *, void *);
611 struct radix_node_head *rnh;
615 * Set rtinfo->rti_ifa and rtinfo->rti_ifp.
618 rt_getifa(struct rt_addrinfo *rtinfo)
620 struct sockaddr *gateway = rtinfo->rti_info[RTAX_GATEWAY];
621 struct sockaddr *dst = rtinfo->rti_info[RTAX_DST];
622 struct sockaddr *ifaaddr = rtinfo->rti_info[RTAX_IFA];
623 int flags = rtinfo->rti_flags;
626 * ifp may be specified by sockaddr_dl
627 * when protocol address is ambiguous.
629 if (rtinfo->rti_ifp == NULL) {
630 struct sockaddr *ifpaddr;
632 ifpaddr = rtinfo->rti_info[RTAX_IFP];
633 if (ifpaddr != NULL && ifpaddr->sa_family == AF_LINK) {
636 ifa = ifa_ifwithnet(ifpaddr);
638 rtinfo->rti_ifp = ifa->ifa_ifp;
642 if (rtinfo->rti_ifa == NULL && ifaaddr != NULL)
643 rtinfo->rti_ifa = ifa_ifwithaddr(ifaaddr);
644 if (rtinfo->rti_ifa == NULL) {
647 sa = ifaaddr != NULL ? ifaaddr :
648 (gateway != NULL ? gateway : dst);
649 if (sa != NULL && rtinfo->rti_ifp != NULL)
650 rtinfo->rti_ifa = ifaof_ifpforaddr(sa, rtinfo->rti_ifp);
651 else if (dst != NULL && gateway != NULL)
652 rtinfo->rti_ifa = ifa_ifwithroute(flags, dst, gateway);
654 rtinfo->rti_ifa = ifa_ifwithroute(flags, sa, sa);
656 if (rtinfo->rti_ifa == NULL)
657 return (ENETUNREACH);
659 if (rtinfo->rti_ifp == NULL)
660 rtinfo->rti_ifp = rtinfo->rti_ifa->ifa_ifp;
665 * Do appropriate manipulations of a routing tree given
666 * all the bits of info needed
671 struct sockaddr *dst,
672 struct sockaddr *gateway,
673 struct sockaddr *netmask,
675 struct rtentry **ret_nrt)
677 struct rt_addrinfo rtinfo;
679 bzero(&rtinfo, sizeof(struct rt_addrinfo));
680 rtinfo.rti_info[RTAX_DST] = dst;
681 rtinfo.rti_info[RTAX_GATEWAY] = gateway;
682 rtinfo.rti_info[RTAX_NETMASK] = netmask;
683 rtinfo.rti_flags = flags;
684 return rtrequest1(req, &rtinfo, ret_nrt);
690 struct sockaddr *dst,
691 struct sockaddr *gateway,
692 struct sockaddr *netmask,
695 struct rt_addrinfo rtinfo;
697 bzero(&rtinfo, sizeof(struct rt_addrinfo));
698 rtinfo.rti_info[RTAX_DST] = dst;
699 rtinfo.rti_info[RTAX_GATEWAY] = gateway;
700 rtinfo.rti_info[RTAX_NETMASK] = netmask;
701 rtinfo.rti_flags = flags;
702 return rtrequest1_global(req, &rtinfo, NULL, NULL);
706 struct netmsg_base base;
708 struct rt_addrinfo *rtinfo;
709 rtrequest1_callback_func_t callback;
714 rtrequest1_global(int req, struct rt_addrinfo *rtinfo,
715 rtrequest1_callback_func_t callback, void *arg)
718 struct netmsg_rtq msg;
720 netmsg_init(&msg.base, NULL, &curthread->td_msgport,
721 0, rtrequest1_msghandler);
722 msg.base.lmsg.ms_error = -1;
725 msg.callback = callback;
727 error = lwkt_domsg(rtable_portfn(0), &msg.base.lmsg, 0);
732 * Handle a route table request on the current cpu. Since the route table's
733 * are supposed to be identical on each cpu, an error occuring later in the
734 * message chain is considered system-fatal.
737 rtrequest1_msghandler(netmsg_t msg)
739 struct netmsg_rtq *rmsg = (void *)msg;
740 struct rt_addrinfo rtinfo;
741 struct rtentry *rt = NULL;
746 * Copy the rtinfo. We need to make sure that the original
747 * rtinfo, which is setup by the caller, in the netmsg will
748 * _not_ be changed; else the next CPU on the netmsg forwarding
749 * path will see a different rtinfo than what this CPU has seen.
751 rtinfo = *rmsg->rtinfo;
753 error = rtrequest1(rmsg->req, &rtinfo, &rt);
757 rmsg->callback(rmsg->req, error, &rtinfo, rt, rmsg->arg);
760 * RTM_DELETE's are propogated even if an error occurs, since a
761 * cloned route might be undergoing deletion and cloned routes
762 * are not necessarily replicated. An overall error is returned
763 * only if no cpus have the route in question.
765 if (rmsg->base.lmsg.ms_error < 0 || error == 0)
766 rmsg->base.lmsg.ms_error = error;
768 nextcpu = mycpuid + 1;
769 if (error && rmsg->req != RTM_DELETE) {
771 panic("rtrequest1_msghandler: rtrequest table "
772 "error was cpu%d, err %d\n", mycpuid, error);
774 lwkt_replymsg(&rmsg->base.lmsg, error);
775 } else if (nextcpu < ncpus) {
776 lwkt_forwardmsg(rtable_portfn(nextcpu), &rmsg->base.lmsg);
778 lwkt_replymsg(&rmsg->base.lmsg, rmsg->base.lmsg.ms_error);
783 rtrequest1(int req, struct rt_addrinfo *rtinfo, struct rtentry **ret_nrt)
785 struct sockaddr *dst = rtinfo->rti_info[RTAX_DST];
787 struct radix_node *rn;
788 struct radix_node_head *rnh;
790 struct sockaddr *ndst;
794 #define gotoerr(x) { error = x ; goto bad; }
798 rt_addrinfo_print(req, rtinfo);
803 * Find the correct routing tree to use for this Address Family
805 if ((rnh = rt_tables[mycpuid][dst->sa_family]) == NULL)
806 gotoerr(EAFNOSUPPORT);
809 * If we are adding a host route then we don't want to put
810 * a netmask in the tree, nor do we want to clone it.
812 if (rtinfo->rti_flags & RTF_HOST) {
813 rtinfo->rti_info[RTAX_NETMASK] = NULL;
814 rtinfo->rti_flags &= ~(RTF_CLONING | RTF_PRCLONING);
819 /* Remove the item from the tree. */
820 rn = rnh->rnh_deladdr((char *)rtinfo->rti_info[RTAX_DST],
821 (char *)rtinfo->rti_info[RTAX_NETMASK],
825 KASSERT(!(rn->rn_flags & (RNF_ACTIVE | RNF_ROOT)),
826 ("rnh_deladdr returned flags 0x%x", rn->rn_flags));
827 rt = (struct rtentry *)rn;
829 /* ref to prevent a deletion race */
832 /* Free any routes cloned from this one. */
833 if ((rt->rt_flags & (RTF_CLONING | RTF_PRCLONING)) &&
834 rt_mask(rt) != NULL) {
835 rnh->rnh_walktree_from(rnh, (char *)rt_key(rt),
840 if (rt->rt_gwroute != NULL) {
841 RTFREE(rt->rt_gwroute);
842 rt->rt_gwroute = NULL;
846 * NB: RTF_UP must be set during the search above,
847 * because we might delete the last ref, causing
848 * rt to get freed prematurely.
850 rt->rt_flags &= ~RTF_UP;
854 rt_print(rtinfo, rt);
857 /* Give the protocol a chance to keep things in sync. */
858 if ((ifa = rt->rt_ifa) && ifa->ifa_rtrequest)
859 ifa->ifa_rtrequest(RTM_DELETE, rt);
862 * If the caller wants it, then it can have it,
863 * but it's up to it to free the rtentry as we won't be
866 KASSERT(rt->rt_refcnt >= 0,
867 ("rtrequest1(DELETE): refcnt %ld", rt->rt_refcnt));
868 if (ret_nrt != NULL) {
869 /* leave ref intact for return */
872 /* deref / attempt to destroy */
878 if (ret_nrt == NULL || (rt = *ret_nrt) == NULL)
882 rt->rt_flags & ~(RTF_CLONING | RTF_PRCLONING | RTF_STATIC);
883 rtinfo->rti_flags |= RTF_WASCLONED;
884 rtinfo->rti_info[RTAX_GATEWAY] = rt->rt_gateway;
885 if ((rtinfo->rti_info[RTAX_NETMASK] = rt->rt_genmask) == NULL)
886 rtinfo->rti_flags |= RTF_HOST;
887 rtinfo->rti_info[RTAX_MPLS1] = rt->rt_shim[0];
888 rtinfo->rti_info[RTAX_MPLS2] = rt->rt_shim[1];
889 rtinfo->rti_info[RTAX_MPLS3] = rt->rt_shim[2];
893 KASSERT(!(rtinfo->rti_flags & RTF_GATEWAY) ||
894 rtinfo->rti_info[RTAX_GATEWAY] != NULL,
895 ("rtrequest: GATEWAY but no gateway"));
897 if (rtinfo->rti_ifa == NULL && (error = rt_getifa(rtinfo)))
899 ifa = rtinfo->rti_ifa;
901 R_Malloc(rt, struct rtentry *, sizeof(struct rtentry));
903 if (req == RTM_ADD) {
904 kprintf("rtrequest1: alloc rtentry failed on "
909 bzero(rt, sizeof(struct rtentry));
910 rt->rt_flags = RTF_UP | rtinfo->rti_flags;
911 rt->rt_cpuid = mycpuid;
913 if (mycpuid != 0 && req == RTM_ADD) {
914 /* For RTM_ADD, we have already sent rtmsg on CPU0. */
915 reportmsg = RTL_DONTREPORT;
918 * For RTM_ADD, we only send rtmsg on CPU0.
919 * For RTM_RESOLVE, we always send rtmsg. XXX
921 reportmsg = RTL_REPORTMSG;
923 error = rt_setgate(rt, dst, rtinfo->rti_info[RTAX_GATEWAY],
931 if (rtinfo->rti_info[RTAX_NETMASK] != NULL)
932 rt_maskedcopy(dst, ndst,
933 rtinfo->rti_info[RTAX_NETMASK]);
935 bcopy(dst, ndst, dst->sa_len);
937 if (rtinfo->rti_info[RTAX_MPLS1] != NULL)
938 rt_setshims(rt, rtinfo->rti_info);
941 * Note that we now have a reference to the ifa.
942 * This moved from below so that rnh->rnh_addaddr() can
943 * examine the ifa and ifa->ifa_ifp if it so desires.
947 rt->rt_ifp = ifa->ifa_ifp;
948 /* XXX mtu manipulation will be done in rnh_addaddr -- itojun */
950 rn = rnh->rnh_addaddr((char *)ndst,
951 (char *)rtinfo->rti_info[RTAX_NETMASK],
954 struct rtentry *oldrt;
957 * We already have one of these in the tree.
958 * We do a special hack: if the old route was
959 * cloned, then we blow it away and try
960 * re-inserting the new one.
962 oldrt = rtpurelookup(ndst);
965 if (oldrt->rt_flags & RTF_WASCLONED) {
966 rtrequest(RTM_DELETE, rt_key(oldrt),
969 oldrt->rt_flags, NULL);
970 rn = rnh->rnh_addaddr((char *)ndst,
972 rtinfo->rti_info[RTAX_NETMASK],
979 * If it still failed to go into the tree,
980 * then un-make it (this should be a function).
983 if (rt->rt_gwroute != NULL)
984 rtfree(rt->rt_gwroute);
992 * If we got here from RESOLVE, then we are cloning
993 * so clone the rest, and note that we
994 * are a clone (and increment the parent's references)
996 if (req == RTM_RESOLVE) {
997 rt->rt_rmx = (*ret_nrt)->rt_rmx; /* copy metrics */
998 rt->rt_rmx.rmx_pksent = 0; /* reset packet counter */
999 if ((*ret_nrt)->rt_flags &
1000 (RTF_CLONING | RTF_PRCLONING)) {
1001 rt->rt_parent = *ret_nrt;
1002 (*ret_nrt)->rt_refcnt++;
1007 * if this protocol has something to add to this then
1008 * allow it to do that as well.
1010 if (ifa->ifa_rtrequest != NULL)
1011 ifa->ifa_rtrequest(req, rt);
1014 * We repeat the same procedure from rt_setgate() here because
1015 * it doesn't fire when we call it there because the node
1016 * hasn't been added to the tree yet.
1018 if (req == RTM_ADD && !(rt->rt_flags & RTF_HOST) &&
1019 rt_mask(rt) != NULL) {
1020 struct rtfc_arg arg = { rt, rnh };
1022 rnh->rnh_walktree_from(rnh, (char *)rt_key(rt),
1023 (char *)rt_mask(rt),
1024 rt_fixchange, &arg);
1029 rt_print(rtinfo, rt);
1032 * Return the resulting rtentry,
1033 * increasing the number of references by one.
1035 if (ret_nrt != NULL) {
1047 kprintf("rti %p failed error %d\n", rtinfo, error);
1049 kprintf("rti %p succeeded\n", rtinfo);
1057 * Called from rtrequest(RTM_DELETE, ...) to fix up the route's ``family''
1058 * (i.e., the routes related to it by the operation of cloning). This
1059 * routine is iterated over all potential former-child-routes by way of
1060 * rnh->rnh_walktree_from() above, and those that actually are children of
1061 * the late parent (passed in as VP here) are themselves deleted.
1064 rt_fixdelete(struct radix_node *rn, void *vp)
1066 struct rtentry *rt = (struct rtentry *)rn;
1067 struct rtentry *rt0 = vp;
1069 if (rt->rt_parent == rt0 &&
1070 !(rt->rt_flags & (RTF_PINNED | RTF_CLONING | RTF_PRCLONING))) {
1071 return rtrequest(RTM_DELETE, rt_key(rt), NULL, rt_mask(rt),
1072 rt->rt_flags, NULL);
1078 * This routine is called from rt_setgate() to do the analogous thing for
1079 * adds and changes. There is the added complication in this case of a
1080 * middle insert; i.e., insertion of a new network route between an older
1081 * network route and (cloned) host routes. For this reason, a simple check
1082 * of rt->rt_parent is insufficient; each candidate route must be tested
1083 * against the (mask, value) of the new route (passed as before in vp)
1084 * to see if the new route matches it.
1086 * XXX - it may be possible to do fixdelete() for changes and reserve this
1087 * routine just for adds. I'm not sure why I thought it was necessary to do
1091 static int rtfcdebug = 0;
1095 rt_fixchange(struct radix_node *rn, void *vp)
1097 struct rtentry *rt = (struct rtentry *)rn;
1098 struct rtfc_arg *ap = vp;
1099 struct rtentry *rt0 = ap->rt0;
1100 struct radix_node_head *rnh = ap->rnh;
1101 u_char *xk1, *xm1, *xk2, *xmp;
1106 kprintf("rt_fixchange: rt %p, rt0 %p\n", rt, rt0);
1109 if (rt->rt_parent == NULL ||
1110 (rt->rt_flags & (RTF_PINNED | RTF_CLONING | RTF_PRCLONING))) {
1112 if (rtfcdebug) kprintf("no parent, pinned or cloning\n");
1117 if (rt->rt_parent == rt0) {
1119 if (rtfcdebug) kprintf("parent match\n");
1121 return rtrequest(RTM_DELETE, rt_key(rt), NULL, rt_mask(rt),
1122 rt->rt_flags, NULL);
1126 * There probably is a function somewhere which does this...
1127 * if not, there should be.
1129 len = imin(rt_key(rt0)->sa_len, rt_key(rt)->sa_len);
1131 xk1 = (u_char *)rt_key(rt0);
1132 xm1 = (u_char *)rt_mask(rt0);
1133 xk2 = (u_char *)rt_key(rt);
1135 /* avoid applying a less specific route */
1136 xmp = (u_char *)rt_mask(rt->rt_parent);
1137 mlen = rt_key(rt->rt_parent)->sa_len;
1138 if (mlen > rt_key(rt0)->sa_len) {
1141 kprintf("rt_fixchange: inserting a less "
1142 "specific route\n");
1146 for (i = rnh->rnh_treetop->rn_offset; i < mlen; i++) {
1147 if ((xmp[i] & ~(xmp[i] ^ xm1[i])) != xmp[i]) {
1150 kprintf("rt_fixchange: inserting a less "
1151 "specific route\n");
1157 for (i = rnh->rnh_treetop->rn_offset; i < len; i++) {
1158 if ((xk2[i] & xm1[i]) != xk1[i]) {
1160 if (rtfcdebug) kprintf("no match\n");
1167 * OK, this node is a clone, and matches the node currently being
1168 * changed/added under the node's mask. So, get rid of it.
1171 if (rtfcdebug) kprintf("deleting\n");
1173 return rtrequest(RTM_DELETE, rt_key(rt), NULL, rt_mask(rt),
1174 rt->rt_flags, NULL);
1177 #define ROUNDUP(a) (a>0 ? (1 + (((a) - 1) | (sizeof(long) - 1))) : sizeof(long))
1180 rt_setgate(struct rtentry *rt0, struct sockaddr *dst, struct sockaddr *gate,
1181 boolean_t generate_report)
1183 char *space, *oldspace;
1184 int dlen = ROUNDUP(dst->sa_len), glen = ROUNDUP(gate->sa_len);
1185 struct rtentry *rt = rt0;
1186 struct radix_node_head *rnh = rt_tables[mycpuid][dst->sa_family];
1189 * A host route with the destination equal to the gateway
1190 * will interfere with keeping LLINFO in the routing
1191 * table, so disallow it.
1193 if (((rt0->rt_flags & (RTF_HOST | RTF_GATEWAY | RTF_LLINFO)) ==
1194 (RTF_HOST | RTF_GATEWAY)) &&
1195 dst->sa_len == gate->sa_len &&
1196 sa_equal(dst, gate)) {
1198 * The route might already exist if this is an RTM_CHANGE
1199 * or a routing redirect, so try to delete it.
1201 if (rt_key(rt0) != NULL)
1202 rtrequest(RTM_DELETE, rt_key(rt0), rt0->rt_gateway,
1203 rt_mask(rt0), rt0->rt_flags, NULL);
1204 return EADDRNOTAVAIL;
1208 * Both dst and gateway are stored in the same malloc'ed chunk
1209 * (If I ever get my hands on....)
1210 * if we need to malloc a new chunk, then keep the old one around
1211 * till we don't need it any more.
1213 if (rt->rt_gateway == NULL || glen > ROUNDUP(rt->rt_gateway->sa_len)) {
1214 oldspace = (char *)rt_key(rt);
1215 R_Malloc(space, char *, dlen + glen);
1218 rt->rt_nodes->rn_key = space;
1220 space = (char *)rt_key(rt); /* Just use the old space. */
1224 /* Set the gateway value. */
1225 rt->rt_gateway = (struct sockaddr *)(space + dlen);
1226 bcopy(gate, rt->rt_gateway, glen);
1228 if (oldspace != NULL) {
1230 * If we allocated a new chunk, preserve the original dst.
1231 * This way, rt_setgate() really just sets the gate
1232 * and leaves the dst field alone.
1234 bcopy(dst, space, dlen);
1239 * If there is already a gwroute, it's now almost definitely wrong
1242 if (rt->rt_gwroute != NULL) {
1243 RTFREE(rt->rt_gwroute);
1244 rt->rt_gwroute = NULL;
1246 if (rt->rt_flags & RTF_GATEWAY) {
1248 * Cloning loop avoidance: In the presence of
1249 * protocol-cloning and bad configuration, it is
1250 * possible to get stuck in bottomless mutual recursion
1251 * (rtrequest rt_setgate rtlookup). We avoid this
1252 * by not allowing protocol-cloning to operate for
1253 * gateways (which is probably the correct choice
1254 * anyway), and avoid the resulting reference loops
1255 * by disallowing any route to run through itself as
1256 * a gateway. This is obviously mandatory when we
1257 * get rt->rt_output().
1259 * This breaks TTCP for hosts outside the gateway! XXX JH
1261 rt->rt_gwroute = _rtlookup(gate, generate_report,
1263 if (rt->rt_gwroute == rt) {
1264 rt->rt_gwroute = NULL;
1266 return EDQUOT; /* failure */
1271 * This isn't going to do anything useful for host routes, so
1272 * don't bother. Also make sure we have a reasonable mask
1273 * (we don't yet have one during adds).
1275 if (!(rt->rt_flags & RTF_HOST) && rt_mask(rt) != NULL) {
1276 struct rtfc_arg arg = { rt, rnh };
1278 rnh->rnh_walktree_from(rnh, (char *)rt_key(rt),
1279 (char *)rt_mask(rt),
1280 rt_fixchange, &arg);
1288 struct sockaddr *src,
1289 struct sockaddr *dst,
1290 struct sockaddr *netmask)
1292 u_char *cp1 = (u_char *)src;
1293 u_char *cp2 = (u_char *)dst;
1294 u_char *cp3 = (u_char *)netmask;
1295 u_char *cplim = cp2 + *cp3;
1296 u_char *cplim2 = cp2 + *cp1;
1298 *cp2++ = *cp1++; *cp2++ = *cp1++; /* copies sa_len & sa_family */
1303 *cp2++ = *cp1++ & *cp3++;
1305 bzero(cp2, cplim2 - cp2);
1309 rt_llroute(struct sockaddr *dst, struct rtentry *rt0, struct rtentry **drt)
1311 struct rtentry *up_rt, *rt;
1313 if (!(rt0->rt_flags & RTF_UP)) {
1314 up_rt = rtlookup(dst);
1316 return (EHOSTUNREACH);
1320 if (up_rt->rt_flags & RTF_GATEWAY) {
1321 if (up_rt->rt_gwroute == NULL) {
1322 up_rt->rt_gwroute = rtlookup(up_rt->rt_gateway);
1323 if (up_rt->rt_gwroute == NULL)
1324 return (EHOSTUNREACH);
1325 } else if (!(up_rt->rt_gwroute->rt_flags & RTF_UP)) {
1326 rtfree(up_rt->rt_gwroute);
1327 up_rt->rt_gwroute = rtlookup(up_rt->rt_gateway);
1328 if (up_rt->rt_gwroute == NULL)
1329 return (EHOSTUNREACH);
1331 rt = up_rt->rt_gwroute;
1334 if (rt->rt_flags & RTF_REJECT &&
1335 (rt->rt_rmx.rmx_expire == 0 || /* rt doesn't expire */
1336 time_second < rt->rt_rmx.rmx_expire)) /* rt not expired */
1337 return (rt->rt_flags & RTF_HOST ? EHOSTDOWN : EHOSTUNREACH);
1343 rt_setshims(struct rtentry *rt, struct sockaddr **rt_shim){
1346 for (i=0; i<3; i++) {
1347 struct sockaddr *shim = rt_shim[RTAX_MPLS1 + i];
1353 shimlen = ROUNDUP(shim->sa_len);
1354 R_Malloc(rt->rt_shim[i], struct sockaddr *, shimlen);
1355 bcopy(shim, rt->rt_shim[i], shimlen);
1364 * Print out a route table entry
1367 rt_print(struct rt_addrinfo *rtinfo, struct rtentry *rn)
1369 kprintf("rti %p cpu %d route %p flags %08lx: ",
1370 rtinfo, mycpuid, rn, rn->rt_flags);
1371 sockaddr_print(rt_key(rn));
1373 sockaddr_print(rt_mask(rn));
1375 sockaddr_print(rn->rt_gateway);
1376 kprintf(" ifc \"%s\"", rn->rt_ifp ? rn->rt_ifp->if_dname : "?");
1377 kprintf(" ifa %p\n", rn->rt_ifa);
1381 rt_addrinfo_print(int cmd, struct rt_addrinfo *rti)
1387 if (cmd == RTM_DELETE && route_debug > 1)
1388 print_backtrace(-1);
1402 kprintf("C%02d ", cmd);
1405 kprintf("rti %p cpu %d ", rti, mycpuid);
1406 for (i = 0; i < rti->rti_addrs; ++i) {
1407 if (rti->rti_info[i] == NULL)
1437 kprintf("(?%02d ", i);
1440 sockaddr_print(rti->rti_info[i]);
1448 sockaddr_print(struct sockaddr *sa)
1450 struct sockaddr_in *sa4;
1451 struct sockaddr_in6 *sa6;
1460 len = sa->sa_len - offsetof(struct sockaddr, sa_data[0]);
1462 switch(sa->sa_family) {
1466 switch(sa->sa_family) {
1468 sa4 = (struct sockaddr_in *)sa;
1469 kprintf("INET %d %d.%d.%d.%d",
1470 ntohs(sa4->sin_port),
1471 (ntohl(sa4->sin_addr.s_addr) >> 24) & 255,
1472 (ntohl(sa4->sin_addr.s_addr) >> 16) & 255,
1473 (ntohl(sa4->sin_addr.s_addr) >> 8) & 255,
1474 (ntohl(sa4->sin_addr.s_addr) >> 0) & 255
1478 sa6 = (struct sockaddr_in6 *)sa;
1479 kprintf("INET6 %d %04x:%04x%04x:%04x:%04x:%04x:%04x:%04x",
1480 ntohs(sa6->sin6_port),
1481 sa6->sin6_addr.s6_addr16[0],
1482 sa6->sin6_addr.s6_addr16[1],
1483 sa6->sin6_addr.s6_addr16[2],
1484 sa6->sin6_addr.s6_addr16[3],
1485 sa6->sin6_addr.s6_addr16[4],
1486 sa6->sin6_addr.s6_addr16[5],
1487 sa6->sin6_addr.s6_addr16[6],
1488 sa6->sin6_addr.s6_addr16[7]
1492 kprintf("AF%d ", sa->sa_family);
1493 while (len > 0 && sa->sa_data[len-1] == 0)
1496 for (i = 0; i < len; ++i) {
1499 kprintf("%d", (unsigned char)sa->sa_data[i]);
1509 * Set up a routing table entry, normally for an interface.
1512 rtinit(struct ifaddr *ifa, int cmd, int flags)
1514 struct sockaddr *dst, *deldst, *netmask;
1515 struct mbuf *m = NULL;
1516 struct radix_node_head *rnh;
1517 struct radix_node *rn;
1518 struct rt_addrinfo rtinfo;
1521 if (flags & RTF_HOST) {
1522 dst = ifa->ifa_dstaddr;
1525 dst = ifa->ifa_addr;
1526 netmask = ifa->ifa_netmask;
1529 * If it's a delete, check that if it exists, it's on the correct
1530 * interface or we might scrub a route to another ifa which would
1531 * be confusing at best and possibly worse.
1533 if (cmd == RTM_DELETE) {
1535 * It's a delete, so it should already exist..
1536 * If it's a net, mask off the host bits
1537 * (Assuming we have a mask)
1539 if (netmask != NULL) {
1540 m = m_get(MB_DONTWAIT, MT_SONAME);
1544 deldst = mtod(m, struct sockaddr *);
1545 rt_maskedcopy(dst, deldst, netmask);
1549 * Look up an rtentry that is in the routing tree and
1550 * contains the correct info.
1552 if ((rnh = rt_tables[mycpuid][dst->sa_family]) == NULL ||
1553 (rn = rnh->rnh_lookup((char *)dst,
1554 (char *)netmask, rnh)) == NULL ||
1555 ((struct rtentry *)rn)->rt_ifa != ifa ||
1556 !sa_equal((struct sockaddr *)rn->rn_key, dst)) {
1559 return (flags & RTF_HOST ? EHOSTUNREACH : ENETUNREACH);
1565 * One would think that as we are deleting, and we know
1566 * it doesn't exist, we could just return at this point
1567 * with an "ELSE" clause, but apparently not..
1569 return (flags & RTF_HOST ? EHOSTUNREACH : ENETUNREACH);
1574 * Do the actual request
1576 bzero(&rtinfo, sizeof(struct rt_addrinfo));
1577 rtinfo.rti_info[RTAX_DST] = dst;
1578 rtinfo.rti_info[RTAX_GATEWAY] = ifa->ifa_addr;
1579 rtinfo.rti_info[RTAX_NETMASK] = netmask;
1580 rtinfo.rti_flags = flags | ifa->ifa_flags;
1581 rtinfo.rti_ifa = ifa;
1582 error = rtrequest1_global(cmd, &rtinfo, rtinit_rtrequest_callback, ifa);
1589 rtinit_rtrequest_callback(int cmd, int error,
1590 struct rt_addrinfo *rtinfo, struct rtentry *rt,
1593 struct ifaddr *ifa = arg;
1595 if (error == 0 && rt) {
1598 rt_newaddrmsg(cmd, ifa, error, rt);
1601 if (cmd == RTM_DELETE) {
1602 if (rt->rt_refcnt == 0) {
1611 struct netmsg_base base;
1613 struct rt_addrinfo *rtinfo;
1614 rtsearch_callback_func_t callback;
1616 boolean_t exact_match;
1621 rtsearch_global(int req, struct rt_addrinfo *rtinfo,
1622 rtsearch_callback_func_t callback, void *arg,
1623 boolean_t exact_match)
1625 struct netmsg_rts msg;
1627 netmsg_init(&msg.base, NULL, &curthread->td_msgport,
1628 0, rtsearch_msghandler);
1630 msg.rtinfo = rtinfo;
1631 msg.callback = callback;
1633 msg.exact_match = exact_match;
1635 return lwkt_domsg(rtable_portfn(0), &msg.base.lmsg, 0);
1639 rtsearch_msghandler(netmsg_t msg)
1641 struct netmsg_rts *rmsg = (void *)msg;
1642 struct rt_addrinfo rtinfo;
1643 struct radix_node_head *rnh;
1648 * Copy the rtinfo. We need to make sure that the original
1649 * rtinfo, which is setup by the caller, in the netmsg will
1650 * _not_ be changed; else the next CPU on the netmsg forwarding
1651 * path will see a different rtinfo than what this CPU has seen.
1653 rtinfo = *rmsg->rtinfo;
1656 * Find the correct routing tree to use for this Address Family
1658 if ((rnh = rt_tables[mycpuid][rtinfo.rti_dst->sa_family]) == NULL) {
1660 panic("partially initialized routing tables");
1661 lwkt_replymsg(&rmsg->base.lmsg, EAFNOSUPPORT);
1666 * Correct rtinfo for the host route searching.
1668 if (rtinfo.rti_flags & RTF_HOST) {
1669 rtinfo.rti_netmask = NULL;
1670 rtinfo.rti_flags &= ~(RTF_CLONING | RTF_PRCLONING);
1673 rt = (struct rtentry *)
1674 rnh->rnh_lookup((char *)rtinfo.rti_dst,
1675 (char *)rtinfo.rti_netmask, rnh);
1678 * If we are asked to do the "exact match", we need to make sure
1679 * that host route searching got a host route while a network
1680 * route searching got a network route.
1682 if (rt != NULL && rmsg->exact_match &&
1683 ((rt->rt_flags ^ rtinfo.rti_flags) & RTF_HOST))
1688 * No matching routes have been found, don't count this
1689 * as a critical error (here, we set 'error' to 0), just
1690 * keep moving on, since at least prcloned routes are not
1691 * duplicated onto each CPU.
1698 error = rmsg->callback(rmsg->req, &rtinfo, rt, rmsg->arg,
1702 if (error == EJUSTRETURN) {
1703 lwkt_replymsg(&rmsg->base.lmsg, 0);
1708 nextcpu = mycpuid + 1;
1710 KKASSERT(rmsg->found_cnt > 0);
1713 * Under following cases, unrecoverable error has
1715 * o Request is RTM_GET
1716 * o The first time that we find the route, but the
1717 * modification fails.
1719 if (rmsg->req != RTM_GET && rmsg->found_cnt > 1) {
1720 panic("rtsearch_msghandler: unrecoverable error "
1723 lwkt_replymsg(&rmsg->base.lmsg, error);
1724 } else if (nextcpu < ncpus) {
1725 lwkt_forwardmsg(rtable_portfn(nextcpu), &rmsg->base.lmsg);
1727 if (rmsg->found_cnt == 0) {
1728 /* The requested route was never seen ... */
1731 lwkt_replymsg(&rmsg->base.lmsg, error);
1736 rtmask_add_global(struct sockaddr *mask)
1738 struct netmsg_base msg;
1740 netmsg_init(&msg, NULL, &curthread->td_msgport,
1741 0, rtmask_add_msghandler);
1742 msg.lmsg.u.ms_resultp = mask;
1744 return lwkt_domsg(rtable_portfn(0), &msg.lmsg, 0);
1748 _rtmask_lookup(struct sockaddr *mask, boolean_t search)
1750 struct radix_node *n;
1752 #define clen(s) (*(u_char *)(s))
1753 n = rn_addmask((char *)mask, search, 1, rn_cpumaskhead(mycpuid));
1755 mask->sa_len >= clen(n->rn_key) &&
1756 bcmp((char *)mask + 1,
1757 (char *)n->rn_key + 1, clen(n->rn_key) - 1) == 0) {
1758 return (struct sockaddr *)n->rn_key;
1766 rtmask_add_msghandler(netmsg_t msg)
1768 struct lwkt_msg *lmsg = &msg->lmsg;
1769 struct sockaddr *mask = lmsg->u.ms_resultp;
1770 int error = 0, nextcpu;
1772 if (rtmask_lookup(mask) == NULL)
1775 nextcpu = mycpuid + 1;
1776 if (!error && nextcpu < ncpus)
1777 lwkt_forwardmsg(rtable_portfn(nextcpu), lmsg);
1779 lwkt_replymsg(lmsg, error);
1782 /* This must be before ip6_init2(), which is now SI_ORDER_MIDDLE */
1783 SYSINIT(route, SI_SUB_PROTO_DOMAIN, SI_ORDER_THIRD, route_init, 0);
1785 struct rtchange_arg {
1786 struct ifaddr *old_ifa;
1787 struct ifaddr *new_ifa;
1793 rtchange_ifa(struct rtentry *rt, struct rtchange_arg *ap)
1795 if (rt->rt_ifa->ifa_rtrequest != NULL)
1796 rt->rt_ifa->ifa_rtrequest(RTM_DELETE, rt);
1797 IFAFREE(rt->rt_ifa);
1799 IFAREF(ap->new_ifa);
1800 rt->rt_ifa = ap->new_ifa;
1801 rt->rt_ifp = ap->new_ifa->ifa_ifp;
1802 if (rt->rt_ifa->ifa_rtrequest != NULL)
1803 rt->rt_ifa->ifa_rtrequest(RTM_ADD, rt);
1809 rtchange_callback(struct radix_node *rn, void *xap)
1811 struct rtchange_arg *ap = xap;
1812 struct rtentry *rt = (struct rtentry *)rn;
1814 if (rt->rt_ifa == ap->old_ifa) {
1815 if (rt->rt_flags & (RTF_CLONING | RTF_PRCLONING)) {
1817 * We could saw the branch off when we are
1818 * still sitting on it, if the ifa_rtrequest
1819 * DEL/ADD are called directly from here.
1824 rtchange_ifa(rt, ap);
1829 struct netmsg_rtchange {
1830 struct netmsg_base base;
1831 struct ifaddr *old_ifa;
1832 struct ifaddr *new_ifa;
1837 rtchange_dispatch(netmsg_t msg)
1839 struct netmsg_rtchange *rmsg = (void *)msg;
1840 struct radix_node_head *rnh;
1841 struct rtchange_arg arg;
1846 memset(&arg, 0, sizeof(arg));
1847 arg.old_ifa = rmsg->old_ifa;
1848 arg.new_ifa = rmsg->new_ifa;
1850 rnh = rt_tables[cpu][AF_INET];
1854 KKASSERT(arg.rt == NULL);
1855 error = rnh->rnh_walktree(rnh, rtchange_callback, &arg);
1856 if (arg.rt != NULL) {
1861 rtchange_ifa(rt, &arg);
1870 if (nextcpu < ncpus)
1871 lwkt_forwardmsg(rtable_portfn(nextcpu), &rmsg->base.lmsg);
1873 lwkt_replymsg(&rmsg->base.lmsg, 0);
1877 rtchange(struct ifaddr *old_ifa, struct ifaddr *new_ifa)
1879 struct netmsg_rtchange msg;
1882 * XXX individual requests are not independantly chained,
1883 * which means that the per-cpu route tables will not be
1884 * consistent in the middle of the operation. If routes
1885 * related to the interface are manipulated while we are
1886 * doing this the inconsistancy could trigger a panic.
1888 netmsg_init(&msg.base, NULL, &curthread->td_msgport, 0,
1890 msg.old_ifa = old_ifa;
1891 msg.new_ifa = new_ifa;
1893 KASSERT(&curthread->td_msgport != rtable_portfn(0),
1894 ("rtchange in rtable thread"));
1895 lwkt_domsg(rtable_portfn(0), &msg.base.lmsg, 0);
1898 old_ifa->ifa_flags &= ~IFA_ROUTE;
1899 new_ifa->ifa_flags |= IFA_ROUTE;