2 * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
4 * Copyright (c) 2020 Alexander V. Chernikov
6 * Redistribution and use in source and binary forms, with or without
7 * modification, are permitted provided that the following conditions
9 * 1. Redistributions of source code must retain the above copyright
10 * notice, this list of conditions and the following disclaimer.
11 * 2. Redistributions in binary form must reproduce the above copyright
12 * notice, this list of conditions and the following disclaimer in the
13 * documentation and/or other materials provided with the distribution.
15 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
16 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
17 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
18 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
19 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
20 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
21 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
22 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
23 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
24 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
28 #include <sys/cdefs.h>
29 __FBSDID("$FreeBSD$");
31 #include "opt_inet6.h"
32 #include "opt_route.h"
34 #include <sys/param.h>
35 #include <sys/systm.h>
36 #include <sys/malloc.h>
38 #include <sys/socket.h>
39 #include <sys/sysctl.h>
40 #include <sys/syslog.h>
41 #include <sys/kernel.h>
43 #include <sys/rmlock.h>
46 #include <net/if_var.h>
47 #include <net/if_dl.h>
49 #include <net/route.h>
50 #include <net/route/route_ctl.h>
51 #include <net/route/route_var.h>
52 #include <net/route/nhop_utils.h>
53 #include <net/route/nhop.h>
54 #include <net/route/nhop_var.h>
55 #include <netinet/in.h>
56 #include <netinet6/scope6_var.h>
61 * This file contains control plane routing tables functions.
63 * All functions assumes they are called in net epoch.
66 struct rib_subscription {
67 CK_STAILQ_ENTRY(rib_subscription) next;
68 rib_subscription_cb_t *func;
71 enum rib_subscription_type type;
72 struct epoch_context epoch_ctx;
75 static int add_route(struct rib_head *rnh, struct rt_addrinfo *info,
76 struct rib_cmd_info *rc);
77 static int add_route_nhop(struct rib_head *rnh, struct rtentry *rt,
78 struct rt_addrinfo *info, struct route_nhop_data *rnd,
79 struct rib_cmd_info *rc);
80 static int del_route(struct rib_head *rnh, struct rt_addrinfo *info,
81 struct rib_cmd_info *rc);
82 static int change_route(struct rib_head *rnh, struct rt_addrinfo *info,
83 struct route_nhop_data *nhd_orig, struct rib_cmd_info *rc);
85 static int rt_unlinkrte(struct rib_head *rnh, struct rt_addrinfo *info,
86 struct rib_cmd_info *rc);
88 static void rib_notify(struct rib_head *rnh, enum rib_subscription_type type,
89 struct rib_cmd_info *rc);
91 static void destroy_subscription_epoch(epoch_context_t ctx);
93 static bool rib_can_multipath(struct rib_head *rh);
96 /* Per-vnet multipath routing configuration */
97 SYSCTL_DECL(_net_route);
98 #define V_rib_route_multipath VNET(rib_route_multipath)
100 #define _MP_FLAGS CTLFLAG_RW
102 #define _MP_FLAGS CTLFLAG_RD
104 VNET_DEFINE(u_int, rib_route_multipath) = 1;
105 SYSCTL_UINT(_net_route, OID_AUTO, multipath, _MP_FLAGS | CTLFLAG_VNET,
106 &VNET_NAME(rib_route_multipath), 0, "Enable route multipath");
109 /* Routing table UMA zone */
110 VNET_DEFINE_STATIC(uma_zone_t, rtzone);
111 #define V_rtzone VNET(rtzone)
117 V_rtzone = uma_zcreate("rtentry", sizeof(struct rtentry),
118 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
123 vnet_rtzone_destroy()
126 uma_zdestroy(V_rtzone);
131 destroy_rtentry(struct rtentry *rt)
133 struct nhop_object *nh = rt->rt_nhop;
136 * At this moment rnh, nh_control may be already freed.
137 * nhop interface may have been migrated to a different vnet.
138 * Use vnet stored in the nexthop to delete the entry.
141 if (NH_IS_NHGRP(nh)) {
142 struct weightened_nhop *wn;
144 wn = nhgrp_get_nhops((struct nhgrp_object *)nh, &num_nhops);
148 CURVNET_SET(nhop_get_vnet(nh));
150 /* Unreference nexthop */
151 nhop_free_any(rt->rt_nhop);
153 uma_zfree(V_rtzone, rt);
159 * Epoch callback indicating rtentry is safe to destroy
162 destroy_rtentry_epoch(epoch_context_t ctx)
166 rt = __containerof(ctx, struct rtentry, rt_epoch_ctx);
172 * Schedule rtentry deletion
175 rtfree(struct rtentry *rt)
178 KASSERT(rt != NULL, ("%s: NULL rt", __func__));
180 epoch_call(net_epoch_preempt, destroy_rtentry_epoch,
184 static struct rib_head *
185 get_rnh(uint32_t fibnum, const struct rt_addrinfo *info)
187 struct rib_head *rnh;
188 struct sockaddr *dst;
190 KASSERT((fibnum < rt_numfibs), ("rib_add_route: bad fibnum"));
192 dst = info->rti_info[RTAX_DST];
193 rnh = rt_tables_get_rnh(fibnum, dst->sa_family);
200 rib_can_multipath(struct rib_head *rh)
204 CURVNET_SET(rh->rib_vnet);
205 result = !!V_rib_route_multipath;
212 * Check is nhop is multipath-eligible.
213 * Avoid nhops without gateways and redirects.
215 * Returns 1 for multipath-eligible nexthop,
219 nhop_can_multipath(const struct nhop_object *nh)
222 if ((nh->nh_flags & NHF_MULTIPATH) != 0)
224 if ((nh->nh_flags & NHF_GATEWAY) == 0)
226 if ((nh->nh_flags & NHF_REDIRECT) != 0)
234 get_info_weight(const struct rt_addrinfo *info, uint32_t default_weight)
238 if (info->rti_mflags & RTV_WEIGHT)
239 weight = info->rti_rmx->rmx_weight;
241 weight = default_weight;
242 /* Keep upper 1 byte for adm distance purposes */
243 if (weight > RT_MAX_WEIGHT)
244 weight = RT_MAX_WEIGHT;
250 rt_is_host(const struct rtentry *rt)
253 return (rt->rte_flags & RTF_HOST);
257 rt_get_family(const struct rtentry *rt)
259 const struct sockaddr *dst;
261 dst = (const struct sockaddr *)rt_key_const(rt);
263 return (dst->sa_family);
267 * Returns pointer to nexthop or nexthop group
268 * associated with @rt
271 rt_get_raw_nhop(const struct rtentry *rt)
274 return (rt->rt_nhop);
279 * Stores IPv4 address and prefix length of @rt inside
281 * @pscopeid is currently always set to 0.
284 rt_get_inet_prefix_plen(const struct rtentry *rt, struct in_addr *paddr,
285 int *plen, uint32_t *pscopeid)
287 const struct sockaddr_in *dst;
289 dst = (const struct sockaddr_in *)rt_key_const(rt);
290 KASSERT((dst->sin_family == AF_INET),
291 ("rt family is %d, not inet", dst->sin_family));
292 *paddr = dst->sin_addr;
293 dst = (const struct sockaddr_in *)rt_mask_const(rt);
297 *plen = bitcount32(dst->sin_addr.s_addr);
302 * Stores IPv4 address and prefix mask of @rt inside
303 * @paddr and @pmask. Sets mask to INADDR_ANY for host routes.
304 * @pscopeid is currently always set to 0.
307 rt_get_inet_prefix_pmask(const struct rtentry *rt, struct in_addr *paddr,
308 struct in_addr *pmask, uint32_t *pscopeid)
310 const struct sockaddr_in *dst;
312 dst = (const struct sockaddr_in *)rt_key_const(rt);
313 KASSERT((dst->sin_family == AF_INET),
314 ("rt family is %d, not inet", dst->sin_family));
315 *paddr = dst->sin_addr;
316 dst = (const struct sockaddr_in *)rt_mask_const(rt);
318 pmask->s_addr = INADDR_BROADCAST;
320 *pmask = dst->sin_addr;
327 inet6_get_plen(const struct in6_addr *addr)
330 return (bitcount32(addr->s6_addr32[0]) + bitcount32(addr->s6_addr32[1]) +
331 bitcount32(addr->s6_addr32[2]) + bitcount32(addr->s6_addr32[3]));
335 * Stores IPv6 address and prefix length of @rt inside
336 * @paddr and @plen. Addresses are returned in de-embedded form.
337 * Scopeid is set to 0 for non-LL addresses.
340 rt_get_inet6_prefix_plen(const struct rtentry *rt, struct in6_addr *paddr,
341 int *plen, uint32_t *pscopeid)
343 const struct sockaddr_in6 *dst;
345 dst = (const struct sockaddr_in6 *)rt_key_const(rt);
346 KASSERT((dst->sin6_family == AF_INET6),
347 ("rt family is %d, not inet6", dst->sin6_family));
348 if (IN6_IS_SCOPE_LINKLOCAL(&dst->sin6_addr))
349 in6_splitscope(&dst->sin6_addr, paddr, pscopeid);
351 *paddr = dst->sin6_addr;
352 dst = (const struct sockaddr_in6 *)rt_mask_const(rt);
356 *plen = inet6_get_plen(&dst->sin6_addr);
360 * Stores IPv6 address and prefix mask of @rt inside
361 * @paddr and @pmask. Addresses are returned in de-embedded form.
362 * Scopeid is set to 0 for non-LL addresses.
365 rt_get_inet6_prefix_pmask(const struct rtentry *rt, struct in6_addr *paddr,
366 struct in6_addr *pmask, uint32_t *pscopeid)
368 const struct sockaddr_in6 *dst;
370 dst = (const struct sockaddr_in6 *)rt_key_const(rt);
371 KASSERT((dst->sin6_family == AF_INET6),
372 ("rt family is %d, not inet", dst->sin6_family));
373 if (IN6_IS_SCOPE_LINKLOCAL(&dst->sin6_addr))
374 in6_splitscope(&dst->sin6_addr, paddr, pscopeid);
376 *paddr = dst->sin6_addr;
377 dst = (const struct sockaddr_in6 *)rt_mask_const(rt);
379 memset(pmask, 0xFF, sizeof(struct in6_addr));
381 *pmask = dst->sin6_addr;
386 rt_set_expire_info(struct rtentry *rt, const struct rt_addrinfo *info)
389 /* Kernel -> userland timebase conversion. */
390 if (info->rti_mflags & RTV_EXPIRE)
391 rt->rt_expire = info->rti_rmx->rmx_expire ?
392 info->rti_rmx->rmx_expire - time_second + time_uptime : 0;
396 * Check if specified @gw matches gw data in the nexthop @nh.
398 * Returns true if matches, false otherwise.
401 match_nhop_gw(const struct nhop_object *nh, const struct sockaddr *gw)
404 if (nh->gw_sa.sa_family != gw->sa_family)
407 switch (gw->sa_family) {
409 return (nh->gw4_sa.sin_addr.s_addr ==
410 ((const struct sockaddr_in *)gw)->sin_addr.s_addr);
413 const struct sockaddr_in6 *gw6;
414 gw6 = (const struct sockaddr_in6 *)gw;
417 * Currently (2020-09) IPv6 gws in kernel have their
418 * scope embedded. Once this becomes false, this code
419 * has to be revisited.
421 if (IN6_ARE_ADDR_EQUAL(&nh->gw6_sa.sin6_addr,
428 const struct sockaddr_dl *sdl;
429 sdl = (const struct sockaddr_dl *)gw;
430 return (nh->gwl_sa.sdl_index == sdl->sdl_index);
433 return (memcmp(&nh->gw_sa, gw, nh->gw_sa.sa_len) == 0);
441 * Checks if data in @info matches nexhop @nh.
443 * Returns 0 on success,
444 * ESRCH if not matched,
445 * ENOENT if filter function returned false
448 check_info_match_nhop(const struct rt_addrinfo *info, const struct rtentry *rt,
449 const struct nhop_object *nh)
451 const struct sockaddr *gw = info->rti_info[RTAX_GATEWAY];
453 if (info->rti_filter != NULL) {
454 if (info->rti_filter(rt, nh, info->rti_filterdata) == 0)
459 if ((gw != NULL) && !match_nhop_gw(nh, gw))
466 * Checks if nexhop @nh can be rewritten by data in @info because
467 * of higher "priority". Currently the only case for such scenario
468 * is kernel installing interface routes, marked by RTF_PINNED flag.
471 * 1 if @info data has higher priority
472 * 0 if priority is the same
473 * -1 if priority is lower
476 can_override_nhop(const struct rt_addrinfo *info, const struct nhop_object *nh)
479 if (info->rti_flags & RTF_PINNED) {
480 return (NH_IS_PINNED(nh)) ? 0 : 1;
482 return (NH_IS_PINNED(nh)) ? -1 : 0;
487 * Runs exact prefix match based on @dst and @netmask.
488 * Returns matched @rtentry if found or NULL.
489 * If rtentry was found, saves nexthop / weight value into @rnd.
491 static struct rtentry *
492 lookup_prefix_bysa(struct rib_head *rnh, const struct sockaddr *dst,
493 const struct sockaddr *netmask, struct route_nhop_data *rnd)
497 RIB_LOCK_ASSERT(rnh);
499 rt = (struct rtentry *)rnh->rnh_lookup(__DECONST(void *, dst),
500 __DECONST(void *, netmask), &rnh->head);
502 rnd->rnd_nhop = rt->rt_nhop;
503 rnd->rnd_weight = rt->rt_weight;
505 rnd->rnd_nhop = NULL;
513 * Runs exact prefix match based on dst/netmask from @info.
514 * Assumes RIB lock is held.
515 * Returns matched @rtentry if found or NULL.
516 * If rtentry was found, saves nexthop / weight value into @rnd.
519 lookup_prefix(struct rib_head *rnh, const struct rt_addrinfo *info,
520 struct route_nhop_data *rnd)
524 rt = lookup_prefix_bysa(rnh, info->rti_info[RTAX_DST],
525 info->rti_info[RTAX_NETMASK], rnd);
531 * Adds route defined by @info into the kernel table specified by @fibnum and
532 * sa_family in @info->rti_info[RTAX_DST].
534 * Returns 0 on success and fills in operation metadata into @rc.
537 rib_add_route(uint32_t fibnum, struct rt_addrinfo *info,
538 struct rib_cmd_info *rc)
540 struct rib_head *rnh;
545 rnh = get_rnh(fibnum, info);
547 return (EAFNOSUPPORT);
550 * Check consistency between RTF_HOST flag and netmask
553 if (info->rti_flags & RTF_HOST)
554 info->rti_info[RTAX_NETMASK] = NULL;
555 else if (info->rti_info[RTAX_NETMASK] == NULL)
558 bzero(rc, sizeof(struct rib_cmd_info));
559 rc->rc_cmd = RTM_ADD;
561 error = add_route(rnh, info, rc);
563 rib_notify(rnh, RIB_NOTIFY_DELAYED, rc);
569 * Creates rtentry and nexthop based on @info data.
570 * Return 0 and fills in rtentry into @prt on success,
571 * return errno otherwise.
574 create_rtentry(struct rib_head *rnh, struct rt_addrinfo *info,
575 struct rtentry **prt)
577 struct sockaddr *dst, *ndst, *gateway, *netmask;
579 struct nhop_object *nh;
583 dst = info->rti_info[RTAX_DST];
584 gateway = info->rti_info[RTAX_GATEWAY];
585 netmask = info->rti_info[RTAX_NETMASK];
586 flags = info->rti_flags;
588 if ((flags & RTF_GATEWAY) && !gateway)
590 if (dst && gateway && (dst->sa_family != gateway->sa_family) &&
591 (gateway->sa_family != AF_UNSPEC) && (gateway->sa_family != AF_LINK))
594 if (dst->sa_len > sizeof(((struct rtentry *)NULL)->rt_dstb))
597 if (info->rti_ifa == NULL) {
598 error = rt_getifa_fib(info, rnh->rib_fibnum);
603 error = nhop_create_from_info(rnh, info, &nh);
607 rt = uma_zalloc(V_rtzone, M_NOWAIT | M_ZERO);
612 rt->rte_flags = (RTF_UP | flags) & RTE_RT_FLAG_MASK;
616 memcpy(&rt->rt_dst, dst, dst->sa_len);
617 rt_key(rt) = &rt->rt_dst;
620 * point to the (possibly newly malloc'd) dest address.
622 ndst = (struct sockaddr *)rt_key(rt);
625 * make sure it contains the value we want (masked if needed).
628 rt_maskedcopy(dst, ndst, netmask);
630 bcopy(dst, ndst, dst->sa_len);
633 * We use the ifa reference returned by rt_getifa_fib().
634 * This moved from below so that rnh->rnh_addaddr() can
635 * examine the ifa and ifa->ifa_ifp if it so desires.
638 rt->rt_weight = get_info_weight(info, RT_DEFAULT_WEIGHT);
639 rt_set_expire_info(rt, info);
646 add_route(struct rib_head *rnh, struct rt_addrinfo *info,
647 struct rib_cmd_info *rc)
649 struct nhop_object *nh_orig;
650 struct route_nhop_data rnd_orig, rnd_add;
651 struct nhop_object *nh;
652 struct rtentry *rt, *rt_orig;
655 error = create_rtentry(rnh, info, &rt);
659 rnd_add.rnd_nhop = rt->rt_nhop;
660 rnd_add.rnd_weight = rt->rt_weight;
664 error = add_route_nhop(rnh, rt, info, &rnd_add, rc);
670 /* addition failed. Lookup prefix in the rib to determine the cause */
671 rt_orig = lookup_prefix(rnh, info, &rnd_orig);
672 if (rt_orig == NULL) {
673 /* No prefix -> rnh_addaddr() failed to allocate memory */
676 uma_zfree(V_rtzone, rt);
680 /* We have existing route in the RIB. */
681 nh_orig = rnd_orig.rnd_nhop;
682 /* Check if new route has higher preference */
683 if (can_override_nhop(info, nh_orig) > 0) {
684 /* Update nexthop to the new route */
685 change_route_nhop(rnh, rt_orig, info, &rnd_add, rc);
687 uma_zfree(V_rtzone, rt);
695 if (rib_can_multipath(rnh) && nhop_can_multipath(rnd_add.rnd_nhop) &&
696 nhop_can_multipath(rnd_orig.rnd_nhop))
697 error = add_route_mpath(rnh, info, rt, &rnd_add, &rnd_orig, rc);
700 /* Unable to add - another route with the same preference exists */
704 * ROUTE_MPATH disabled: failed to add route, free both nhop and rt.
705 * ROUTE_MPATH enabled: original nhop reference is unused in any case,
706 * free rt only if not _adding_ new route to rib (e.g. the case
707 * when initial lookup returned existing route, but then it got
708 * deleted prior to multipath group insertion, leading to a simple
709 * non-multipath add as a result).
712 if ((error != 0) || rc->rc_cmd != RTM_ADD)
713 uma_zfree(V_rtzone, rt);
719 * Removes route defined by @info from the kernel table specified by @fibnum and
720 * sa_family in @info->rti_info[RTAX_DST].
722 * Returns 0 on success and fills in operation metadata into @rc.
725 rib_del_route(uint32_t fibnum, struct rt_addrinfo *info, struct rib_cmd_info *rc)
727 struct rib_head *rnh;
728 struct sockaddr *dst_orig, *netmask;
729 struct sockaddr_storage mdst;
734 rnh = get_rnh(fibnum, info);
736 return (EAFNOSUPPORT);
738 bzero(rc, sizeof(struct rib_cmd_info));
739 rc->rc_cmd = RTM_DELETE;
741 dst_orig = info->rti_info[RTAX_DST];
742 netmask = info->rti_info[RTAX_NETMASK];
744 if (netmask != NULL) {
745 /* Ensure @dst is always properly masked */
746 if (dst_orig->sa_len > sizeof(mdst))
748 rt_maskedcopy(dst_orig, (struct sockaddr *)&mdst, netmask);
749 info->rti_info[RTAX_DST] = (struct sockaddr *)&mdst;
751 error = del_route(rnh, info, rc);
752 info->rti_info[RTAX_DST] = dst_orig;
758 * Conditionally unlinks rtentry matching data inside @info from @rnh.
759 * Returns 0 on success with operation result stored in @rc.
761 * ESRCH - if prefix was not found,
762 * EADDRINUSE - if trying to delete higher priority route.
763 * ENOENT - if supplied filter function returned 0 (not matched).
766 rt_unlinkrte(struct rib_head *rnh, struct rt_addrinfo *info, struct rib_cmd_info *rc)
769 struct nhop_object *nh;
770 struct radix_node *rn;
771 struct route_nhop_data rnd;
774 rt = lookup_prefix(rnh, info, &rnd);
780 if (NH_IS_NHGRP(nh)) {
781 error = del_route_mpath(rnh, info, rt,
782 (struct nhgrp_object *)nh, rc);
786 error = check_info_match_nhop(info, rt, nh);
790 if (can_override_nhop(info, nh) < 0)
794 * Remove the item from the tree and return it.
795 * Complain if it is not there and do no more processing.
797 rn = rnh->rnh_deladdr(info->rti_info[RTAX_DST],
798 info->rti_info[RTAX_NETMASK], &rnh->head);
802 if (rn->rn_flags & (RNF_ACTIVE | RNF_ROOT))
803 panic ("rtrequest delete");
806 rt->rte_flags &= ~RTF_UP;
808 /* Finalize notification */
812 rc->rc_cmd = RTM_DELETE;
814 rc->rc_nh_old = rt->rt_nhop;
815 rc->rc_nh_weight = rt->rt_weight;
816 rib_notify(rnh, RIB_NOTIFY_IMMEDIATE, rc);
822 del_route(struct rib_head *rnh, struct rt_addrinfo *info,
823 struct rib_cmd_info *rc)
828 error = rt_unlinkrte(rnh, info, rc);
833 rib_notify(rnh, RIB_NOTIFY_DELAYED, rc);
836 * If the caller wants it, then it can have it,
837 * the entry will be deleted after the end of the current epoch.
839 if (rc->rc_cmd == RTM_DELETE)
844 * Deleting 1 path may result in RTM_CHANGE to
845 * a different mpath group/nhop.
846 * Free old mpath group.
848 nhop_free_any(rc->rc_nh_old);
856 rib_change_route(uint32_t fibnum, struct rt_addrinfo *info,
857 struct rib_cmd_info *rc)
860 struct route_nhop_data rnd_orig;
861 struct rib_head *rnh;
867 rnh = get_rnh(fibnum, info);
869 return (EAFNOSUPPORT);
871 bzero(rc, sizeof(struct rib_cmd_info));
872 rc->rc_cmd = RTM_CHANGE;
874 /* Check if updated gateway exists */
875 if ((info->rti_flags & RTF_GATEWAY) &&
876 (info->rti_info[RTAX_GATEWAY] == NULL)) {
879 * route(8) adds RTF_GATEWAY flag if -interface is not set.
880 * Remove RTF_GATEWAY to enforce consistency and maintain
883 info->rti_flags &= ~RTF_GATEWAY;
887 * route change is done in multiple steps, with dropping and
888 * reacquiring lock. In the situations with multiple processes
889 * changes the same route in can lead to the case when route
890 * is changed between the steps. Address it by retrying the operation
891 * multiple times before failing.
895 rt = (struct rtentry *)rnh->rnh_lookup(info->rti_info[RTAX_DST],
896 info->rti_info[RTAX_NETMASK], &rnh->head);
903 rnd_orig.rnd_nhop = rt->rt_nhop;
904 rnd_orig.rnd_weight = rt->rt_weight;
908 for (int i = 0; i < RIB_MAX_RETRIES; i++) {
909 error = change_route(rnh, info, &rnd_orig, rc);
918 change_nhop(struct rib_head *rnh, struct rt_addrinfo *info,
919 struct nhop_object *nh_orig, struct nhop_object **nh_new)
924 * New gateway could require new ifaddr, ifp;
925 * flags may also be different; ifp may be specified
926 * by ll sockaddr when protocol address is ambiguous
928 if (((nh_orig->nh_flags & NHF_GATEWAY) &&
929 info->rti_info[RTAX_GATEWAY] != NULL) ||
930 info->rti_info[RTAX_IFP] != NULL ||
931 (info->rti_info[RTAX_IFA] != NULL &&
932 !sa_equal(info->rti_info[RTAX_IFA], nh_orig->nh_ifa->ifa_addr))) {
933 error = rt_getifa_fib(info, rnh->rib_fibnum);
936 info->rti_ifa = NULL;
941 error = nhop_create_from_nhop(rnh, nh_orig, info, nh_new);
942 info->rti_ifa = NULL;
949 change_mpath_route(struct rib_head *rnh, struct rt_addrinfo *info,
950 struct route_nhop_data *rnd_orig, struct rib_cmd_info *rc)
953 struct nhop_object *nh, *nh_orig, *nh_new;
954 struct route_nhop_data rnd_new;
957 nh_orig = rnd_orig->rnd_nhop;
959 struct weightened_nhop *wn = NULL, *wn_new;
962 wn = nhgrp_get_nhops((struct nhgrp_object *)nh_orig, &num_nhops);
964 for (int i = 0; i < num_nhops; i++) {
965 if (check_info_match_nhop(info, NULL, wn[i].nh)) {
974 error = change_nhop(rnh, info, nh_orig, &nh_new);
978 wn_new = mallocarray(num_nhops, sizeof(struct weightened_nhop),
979 M_TEMP, M_NOWAIT | M_ZERO);
980 if (wn_new == NULL) {
985 memcpy(wn_new, wn, num_nhops * sizeof(struct weightened_nhop));
986 for (int i = 0; i < num_nhops; i++) {
987 if (wn[i].nh == nh_orig) {
989 wn[i].weight = get_info_weight(info, rnd_orig->rnd_weight);
994 error = nhgrp_get_group(rnh, wn_new, num_nhops, &rnd_new);
996 free(wn_new, M_TEMP);
1001 error = change_route_conditional(rnh, NULL, info, rnd_orig, &rnd_new, rc);
1008 change_route(struct rib_head *rnh, struct rt_addrinfo *info,
1009 struct route_nhop_data *rnd_orig, struct rib_cmd_info *rc)
1012 struct nhop_object *nh, *nh_orig;
1013 struct route_nhop_data rnd_new;
1016 nh_orig = rnd_orig->rnd_nhop;
1017 if (nh_orig == NULL)
1021 if (NH_IS_NHGRP(nh_orig))
1022 return (change_mpath_route(rnh, info, rnd_orig, rc));
1025 rnd_new.rnd_weight = get_info_weight(info, rnd_orig->rnd_weight);
1026 error = change_nhop(rnh, info, nh_orig, &rnd_new.rnd_nhop);
1029 error = change_route_conditional(rnh, NULL, info, rnd_orig, &rnd_new, rc);
1035 * Insert @rt with nhop data from @rnd_new to @rnh.
1036 * Returns 0 on success and stores operation results in @rc.
1039 add_route_nhop(struct rib_head *rnh, struct rtentry *rt,
1040 struct rt_addrinfo *info, struct route_nhop_data *rnd,
1041 struct rib_cmd_info *rc)
1043 struct sockaddr *ndst, *netmask;
1044 struct radix_node *rn;
1047 RIB_WLOCK_ASSERT(rnh);
1049 ndst = (struct sockaddr *)rt_key(rt);
1050 netmask = info->rti_info[RTAX_NETMASK];
1052 rt->rt_nhop = rnd->rnd_nhop;
1053 rt->rt_weight = rnd->rnd_weight;
1054 rn = rnh->rnh_addaddr(ndst, netmask, &rnh->head, rt->rt_nodes);
1057 if (rt->rt_expire > 0)
1058 tmproutes_update(rnh, rt);
1060 /* Finalize notification */
1062 rnh->rnh_prefixes++;
1064 rc->rc_cmd = RTM_ADD;
1066 rc->rc_nh_old = NULL;
1067 rc->rc_nh_new = rnd->rnd_nhop;
1068 rc->rc_nh_weight = rnd->rnd_weight;
1070 rib_notify(rnh, RIB_NOTIFY_IMMEDIATE, rc);
1072 /* Existing route or memory allocation failure */
1080 * Switch @rt nhop/weigh to the ones specified in @rnd.
1081 * Conditionally set rt_expire if set in @info.
1082 * Returns 0 on success.
1085 change_route_nhop(struct rib_head *rnh, struct rtentry *rt,
1086 struct rt_addrinfo *info, struct route_nhop_data *rnd,
1087 struct rib_cmd_info *rc)
1089 struct nhop_object *nh_orig;
1091 RIB_WLOCK_ASSERT(rnh);
1093 nh_orig = rt->rt_nhop;
1095 if (rnd->rnd_nhop != NULL) {
1096 /* Changing expiration & nexthop & weight to a new one */
1097 rt_set_expire_info(rt, info);
1098 rt->rt_nhop = rnd->rnd_nhop;
1099 rt->rt_weight = rnd->rnd_weight;
1100 if (rt->rt_expire > 0)
1101 tmproutes_update(rnh, rt);
1103 /* Route deletion requested. */
1104 struct sockaddr *ndst, *netmask;
1105 struct radix_node *rn;
1107 ndst = (struct sockaddr *)rt_key(rt);
1108 netmask = info->rti_info[RTAX_NETMASK];
1109 rn = rnh->rnh_deladdr(ndst, netmask, &rnh->head);
1113 rt->rte_flags &= ~RTF_UP;
1116 /* Finalize notification */
1118 if (rnd->rnd_nhop == NULL)
1119 rnh->rnh_prefixes--;
1121 rc->rc_cmd = (rnd->rnd_nhop != NULL) ? RTM_CHANGE : RTM_DELETE;
1123 rc->rc_nh_old = nh_orig;
1124 rc->rc_nh_new = rnd->rnd_nhop;
1125 rc->rc_nh_weight = rnd->rnd_weight;
1127 rib_notify(rnh, RIB_NOTIFY_IMMEDIATE, rc);
1133 * Conditionally update route nhop/weight IFF data in @nhd_orig is
1134 * consistent with the current route data.
1135 * Nexthop in @nhd_new is consumed.
1138 change_route_conditional(struct rib_head *rnh, struct rtentry *rt,
1139 struct rt_addrinfo *info, struct route_nhop_data *rnd_orig,
1140 struct route_nhop_data *rnd_new, struct rib_cmd_info *rc)
1142 struct rtentry *rt_new;
1147 rt_new = (struct rtentry *)rnh->rnh_lookup(info->rti_info[RTAX_DST],
1148 info->rti_info[RTAX_NETMASK], &rnh->head);
1150 if (rt_new == NULL) {
1151 if (rnd_orig->rnd_nhop == NULL)
1152 error = add_route_nhop(rnh, rt, info, rnd_new, rc);
1155 * Prefix does not exist, which was not our assumption.
1156 * Update @rnd_orig with the new data and return
1158 rnd_orig->rnd_nhop = NULL;
1159 rnd_orig->rnd_weight = 0;
1163 /* Prefix exists, try to update */
1164 if (rnd_orig->rnd_nhop == rt_new->rt_nhop) {
1166 * Nhop/mpath group hasn't changed. Flip
1167 * to the new precalculated one and return
1169 error = change_route_nhop(rnh, rt_new, info, rnd_new, rc);
1171 /* Update and retry */
1172 rnd_orig->rnd_nhop = rt_new->rt_nhop;
1173 rnd_orig->rnd_weight = rt_new->rt_weight;
1181 rib_notify(rnh, RIB_NOTIFY_DELAYED, rc);
1183 if (rnd_orig->rnd_nhop != NULL)
1184 nhop_free_any(rnd_orig->rnd_nhop);
1187 if (rnd_new->rnd_nhop != NULL)
1188 nhop_free_any(rnd_new->rnd_nhop);
1195 * Performs modification of routing table specificed by @action.
1196 * Table is specified by @fibnum and sa_family in @info->rti_info[RTAX_DST].
1197 * Needs to be run in network epoch.
1199 * Returns 0 on success and fills in @rc with action result.
1202 rib_action(uint32_t fibnum, int action, struct rt_addrinfo *info,
1203 struct rib_cmd_info *rc)
1209 error = rib_add_route(fibnum, info, rc);
1212 error = rib_del_route(fibnum, info, rc);
1215 error = rib_change_route(fibnum, info, rc);
1226 struct rt_addrinfo info;
1227 struct rib_head *rnh;
1228 struct rtentry *head;
1229 struct rib_cmd_info rc;
1233 * Conditionally unlinks @rn from radix tree based
1234 * on info data passed in @arg.
1237 rt_checkdelroute(struct radix_node *rn, void *arg)
1239 struct rt_delinfo *di;
1240 struct rt_addrinfo *info;
1244 di = (struct rt_delinfo *)arg;
1245 rt = (struct rtentry *)rn;
1248 info->rti_info[RTAX_DST] = rt_key(rt);
1249 info->rti_info[RTAX_NETMASK] = rt_mask(rt);
1251 error = rt_unlinkrte(di->rnh, info, &di->rc);
1254 * Add deleted rtentries to the list to GC them
1255 * after dropping the lock.
1257 * XXX: Delayed notifications not implemented
1258 * for nexthop updates.
1260 if ((error == 0) && (di->rc.rc_cmd == RTM_DELETE)) {
1261 /* Add to the list and return */
1262 rt->rt_chain = di->head;
1270 * Iterates over a routing table specified by @fibnum and @family and
1271 * deletes elements marked by @filter_f.
1272 * @fibnum: rtable id
1273 * @family: AF_ address family
1274 * @filter_f: function returning non-zero value for items to delete
1275 * @arg: data to pass to the @filter_f function
1276 * @report: true if rtsock notification is needed.
1279 rib_walk_del(u_int fibnum, int family, rib_filter_f_t *filter_f, void *arg, bool report)
1281 struct rib_head *rnh;
1282 struct rt_delinfo di;
1284 struct nhop_object *nh;
1285 struct epoch_tracker et;
1287 rnh = rt_tables_get_rnh(fibnum, family);
1291 bzero(&di, sizeof(di));
1292 di.info.rti_filter = filter_f;
1293 di.info.rti_filterdata = arg;
1295 di.rc.rc_cmd = RTM_DELETE;
1297 NET_EPOCH_ENTER(et);
1300 rnh->rnh_walktree(&rnh->head, rt_checkdelroute, &di);
1303 /* We might have something to reclaim. */
1304 bzero(&di.rc, sizeof(di.rc));
1305 di.rc.rc_cmd = RTM_DELETE;
1306 while (di.head != NULL) {
1308 di.head = rt->rt_chain;
1309 rt->rt_chain = NULL;
1313 di.rc.rc_nh_old = nh;
1314 rib_notify(rnh, RIB_NOTIFY_DELAYED, &di.rc);
1316 /* TODO std rt -> rt_addrinfo export */
1317 di.info.rti_info[RTAX_DST] = rt_key(rt);
1318 di.info.rti_info[RTAX_NETMASK] = rt_mask(rt);
1322 struct nhgrp_object *nhg;
1323 struct weightened_nhop *wn;
1325 if (NH_IS_NHGRP(nh)) {
1326 nhg = (struct nhgrp_object *)nh;
1327 wn = nhgrp_get_nhops(nhg, &num_nhops);
1328 for (int i = 0; i < num_nhops; i++)
1329 rt_routemsg(RTM_DELETE, rt, wn[i].nh, fibnum);
1332 rt_routemsg(RTM_DELETE, rt, nh, fibnum);
1341 rt_delete_unconditional(struct radix_node *rn, void *arg)
1343 struct rtentry *rt = RNTORT(rn);
1344 struct rib_head *rnh = (struct rib_head *)arg;
1346 rn = rnh->rnh_deladdr(rt_key(rt), rt_mask(rt), &rnh->head);
1347 if (RNTORT(rn) == rt)
1354 * Removes all routes from the routing table without executing notifications.
1355 * rtentres will be removed after the end of a current epoch.
1358 rib_flush_routes(struct rib_head *rnh)
1361 rnh->rnh_walktree(&rnh->head, rt_delete_unconditional, rnh);
1366 rib_flush_routes_family(int family)
1368 struct rib_head *rnh;
1370 for (uint32_t fibnum = 0; fibnum < rt_numfibs; fibnum++) {
1371 if ((rnh = rt_tables_get_rnh(fibnum, family)) != NULL)
1372 rib_flush_routes(rnh);
1377 rib_notify(struct rib_head *rnh, enum rib_subscription_type type,
1378 struct rib_cmd_info *rc)
1380 struct rib_subscription *rs;
1382 CK_STAILQ_FOREACH(rs, &rnh->rnh_subscribers, next) {
1383 if (rs->type == type)
1384 rs->func(rnh, rc, rs->arg);
1388 static struct rib_subscription *
1389 allocate_subscription(rib_subscription_cb_t *f, void *arg,
1390 enum rib_subscription_type type, bool waitok)
1392 struct rib_subscription *rs;
1393 int flags = M_ZERO | (waitok ? M_WAITOK : M_NOWAIT);
1395 rs = malloc(sizeof(struct rib_subscription), M_RTABLE, flags);
1407 * Subscribe for the changes in the routing table specified by @fibnum and
1410 * Returns pointer to the subscription structure on success.
1412 struct rib_subscription *
1413 rib_subscribe(uint32_t fibnum, int family, rib_subscription_cb_t *f, void *arg,
1414 enum rib_subscription_type type, bool waitok)
1416 struct rib_head *rnh;
1417 struct epoch_tracker et;
1419 NET_EPOCH_ENTER(et);
1420 KASSERT((fibnum < rt_numfibs), ("%s: bad fibnum", __func__));
1421 rnh = rt_tables_get_rnh(fibnum, family);
1424 return (rib_subscribe_internal(rnh, f, arg, type, waitok));
1427 struct rib_subscription *
1428 rib_subscribe_internal(struct rib_head *rnh, rib_subscription_cb_t *f, void *arg,
1429 enum rib_subscription_type type, bool waitok)
1431 struct rib_subscription *rs;
1432 struct epoch_tracker et;
1434 if ((rs = allocate_subscription(f, arg, type, waitok)) == NULL)
1438 NET_EPOCH_ENTER(et);
1440 CK_STAILQ_INSERT_HEAD(&rnh->rnh_subscribers, rs, next);
1447 struct rib_subscription *
1448 rib_subscribe_locked(struct rib_head *rnh, rib_subscription_cb_t *f, void *arg,
1449 enum rib_subscription_type type)
1451 struct rib_subscription *rs;
1454 RIB_WLOCK_ASSERT(rnh);
1456 if ((rs = allocate_subscription(f, arg, type, false)) == NULL)
1460 CK_STAILQ_INSERT_HEAD(&rnh->rnh_subscribers, rs, next);
1466 * Remove rtable subscription @rs from the routing table.
1467 * Needs to be run in network epoch.
1470 rib_unsibscribe(struct rib_subscription *rs)
1472 struct rib_head *rnh = rs->rnh;
1477 CK_STAILQ_REMOVE(&rnh->rnh_subscribers, rs, rib_subscription, next);
1480 epoch_call(net_epoch_preempt, destroy_subscription_epoch,
1485 rib_unsibscribe_locked(struct rib_subscription *rs)
1487 struct rib_head *rnh = rs->rnh;
1490 RIB_WLOCK_ASSERT(rnh);
1492 CK_STAILQ_REMOVE(&rnh->rnh_subscribers, rs, rib_subscription, next);
1494 epoch_call(net_epoch_preempt, destroy_subscription_epoch,
1499 * Epoch callback indicating subscription is safe to destroy
1502 destroy_subscription_epoch(epoch_context_t ctx)
1504 struct rib_subscription *rs;
1506 rs = __containerof(ctx, struct rib_subscription, epoch_ctx);
1512 rib_init_subscriptions(struct rib_head *rnh)
1515 CK_STAILQ_INIT(&rnh->rnh_subscribers);
1519 rib_destroy_subscriptions(struct rib_head *rnh)
1521 struct rib_subscription *rs;
1522 struct epoch_tracker et;
1524 NET_EPOCH_ENTER(et);
1526 while ((rs = CK_STAILQ_FIRST(&rnh->rnh_subscribers)) != NULL) {
1527 CK_STAILQ_REMOVE_HEAD(&rnh->rnh_subscribers, next);
1528 epoch_call(net_epoch_preempt, destroy_subscription_epoch,