1 /* Zebra Nexthop Group Code.
2 * Copyright (C) 2019 Cumulus Networks, Inc.
6 * This file is part of FRR.
8 * FRR is free software; you can redistribute it and/or modify it
9 * under the terms of the GNU General Public License as published by the
10 * Free Software Foundation; either version 2, or (at your option) any
13 * FRR is distributed in the hope that it will be useful, but
14 * WITHOUT ANY WARRANTY; without even the implied warranty of
15 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
16 * General Public License for more details.
18 * You should have received a copy of the GNU General Public License
19 * along with FRR; see the file COPYING. If not, write to the Free
20 * Software Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA
25 #include "lib/nexthop.h"
26 #include "lib/nexthop_group_private.h"
27 #include "lib/routemap.h"
29 #include "lib/jhash.h"
30 #include "lib/debug.h"
32 #include "zebra/connected.h"
33 #include "zebra/debug.h"
34 #include "zebra/zebra_router.h"
35 #include "zebra/zebra_nhg_private.h"
36 #include "zebra/zebra_rnh.h"
37 #include "zebra/zebra_routemap.h"
38 #include "zebra/zebra_memory.h"
39 #include "zebra/zserv.h"
41 #include "zebra_errors.h"
42 #include "zebra_dplane.h"
43 #include "zebra/interface.h"
45 DEFINE_MTYPE_STATIC(ZEBRA
, NHG
, "Nexthop Group Entry");
46 DEFINE_MTYPE_STATIC(ZEBRA
, NHG_CONNECTED
, "Nexthop Group Connected");
47 DEFINE_MTYPE_STATIC(ZEBRA
, NHG_CTX
, "Nexthop Group Context");
49 /* id counter to keep in sync with kernel */
52 static struct nhg_hash_entry
*depends_find(const struct nexthop
*nh
,
54 static void depends_add(struct nhg_connected_tree_head
*head
,
55 struct nhg_hash_entry
*depend
);
56 static struct nhg_hash_entry
*
57 depends_find_add(struct nhg_connected_tree_head
*head
, struct nexthop
*nh
,
59 static struct nhg_hash_entry
*
60 depends_find_id_add(struct nhg_connected_tree_head
*head
, uint32_t id
);
61 static void depends_decrement_free(struct nhg_connected_tree_head
*head
);
64 static void nhg_connected_free(struct nhg_connected
*dep
)
66 XFREE(MTYPE_NHG_CONNECTED
, dep
);
69 static struct nhg_connected
*nhg_connected_new(struct nhg_hash_entry
*nhe
)
71 struct nhg_connected
*new = NULL
;
73 new = XCALLOC(MTYPE_NHG_CONNECTED
, sizeof(struct nhg_connected
));
79 void nhg_connected_tree_free(struct nhg_connected_tree_head
*head
)
81 struct nhg_connected
*rb_node_dep
= NULL
;
83 if (!nhg_connected_tree_is_empty(head
)) {
84 frr_each_safe(nhg_connected_tree
, head
, rb_node_dep
) {
85 nhg_connected_tree_del(head
, rb_node_dep
);
86 nhg_connected_free(rb_node_dep
);
91 bool nhg_connected_tree_is_empty(const struct nhg_connected_tree_head
*head
)
93 return nhg_connected_tree_count(head
) ? false : true;
96 struct nhg_connected
*
97 nhg_connected_tree_root(struct nhg_connected_tree_head
*head
)
99 return nhg_connected_tree_first(head
);
102 struct nhg_hash_entry
*
103 nhg_connected_tree_del_nhe(struct nhg_connected_tree_head
*head
,
104 struct nhg_hash_entry
*depend
)
106 struct nhg_connected lookup
= {};
107 struct nhg_connected
*remove
= NULL
;
108 struct nhg_hash_entry
*removed_nhe
;
112 /* Lookup to find the element, then remove it */
113 remove
= nhg_connected_tree_find(head
, &lookup
);
115 /* Re-returning here just in case this API changes..
116 * the _del list api's are a bit undefined at the moment.
118 * So hopefully returning here will make it fail if the api
119 * changes to something different than currently expected.
121 remove
= nhg_connected_tree_del(head
, remove
);
123 /* If the entry was sucessfully removed, free the 'connected` struct */
125 removed_nhe
= remove
->nhe
;
126 nhg_connected_free(remove
);
133 /* Assuming UNIQUE RB tree. If this changes, assumptions here about
134 * insertion need to change.
136 struct nhg_hash_entry
*
137 nhg_connected_tree_add_nhe(struct nhg_connected_tree_head
*head
,
138 struct nhg_hash_entry
*depend
)
140 struct nhg_connected
*new = NULL
;
142 new = nhg_connected_new(depend
);
144 /* On success, NULL will be returned from the
147 if (new && (nhg_connected_tree_add(head
, new) == NULL
))
150 /* If it wasn't successful, it must be a duplicate. We enforce the
151 * unique property for the `nhg_connected` tree.
153 nhg_connected_free(new);
159 nhg_connected_tree_decrement_ref(struct nhg_connected_tree_head
*head
)
161 struct nhg_connected
*rb_node_dep
= NULL
;
163 frr_each_safe(nhg_connected_tree
, head
, rb_node_dep
) {
164 zebra_nhg_decrement_ref(rb_node_dep
->nhe
);
169 nhg_connected_tree_increment_ref(struct nhg_connected_tree_head
*head
)
171 struct nhg_connected
*rb_node_dep
= NULL
;
173 frr_each(nhg_connected_tree
, head
, rb_node_dep
) {
174 zebra_nhg_increment_ref(rb_node_dep
->nhe
);
178 struct nhg_hash_entry
*zebra_nhg_resolve(struct nhg_hash_entry
*nhe
)
180 if (CHECK_FLAG(nhe
->flags
, NEXTHOP_GROUP_RECURSIVE
)
181 && !zebra_nhg_depends_is_empty(nhe
)) {
182 nhe
= nhg_connected_tree_root(&nhe
->nhg_depends
)->nhe
;
183 return zebra_nhg_resolve(nhe
);
189 unsigned int zebra_nhg_depends_count(const struct nhg_hash_entry
*nhe
)
191 return nhg_connected_tree_count(&nhe
->nhg_depends
);
194 bool zebra_nhg_depends_is_empty(const struct nhg_hash_entry
*nhe
)
196 return nhg_connected_tree_is_empty(&nhe
->nhg_depends
);
199 static void zebra_nhg_depends_del(struct nhg_hash_entry
*from
,
200 struct nhg_hash_entry
*depend
)
202 nhg_connected_tree_del_nhe(&from
->nhg_depends
, depend
);
205 static void zebra_nhg_depends_init(struct nhg_hash_entry
*nhe
)
207 nhg_connected_tree_init(&nhe
->nhg_depends
);
210 unsigned int zebra_nhg_dependents_count(const struct nhg_hash_entry
*nhe
)
212 return nhg_connected_tree_count(&nhe
->nhg_dependents
);
216 bool zebra_nhg_dependents_is_empty(const struct nhg_hash_entry
*nhe
)
218 return nhg_connected_tree_is_empty(&nhe
->nhg_dependents
);
221 static void zebra_nhg_dependents_del(struct nhg_hash_entry
*from
,
222 struct nhg_hash_entry
*dependent
)
224 nhg_connected_tree_del_nhe(&from
->nhg_dependents
, dependent
);
227 static void zebra_nhg_dependents_add(struct nhg_hash_entry
*to
,
228 struct nhg_hash_entry
*dependent
)
230 nhg_connected_tree_add_nhe(&to
->nhg_dependents
, dependent
);
233 static void zebra_nhg_dependents_init(struct nhg_hash_entry
*nhe
)
235 nhg_connected_tree_init(&nhe
->nhg_dependents
);
238 /* Release this nhe from anything depending on it */
239 static void zebra_nhg_dependents_release(struct nhg_hash_entry
*nhe
)
241 struct nhg_connected
*rb_node_dep
= NULL
;
243 frr_each_safe(nhg_connected_tree
, &nhe
->nhg_dependents
, rb_node_dep
) {
244 zebra_nhg_depends_del(rb_node_dep
->nhe
, nhe
);
245 /* recheck validity of the dependent */
246 zebra_nhg_check_valid(rb_node_dep
->nhe
);
250 /* Release this nhe from anything that it depends on */
251 static void zebra_nhg_depends_release(struct nhg_hash_entry
*nhe
)
253 if (!zebra_nhg_depends_is_empty(nhe
)) {
254 struct nhg_connected
*rb_node_dep
= NULL
;
256 frr_each_safe(nhg_connected_tree
, &nhe
->nhg_depends
,
258 zebra_nhg_dependents_del(rb_node_dep
->nhe
, nhe
);
264 struct nhg_hash_entry
*zebra_nhg_lookup_id(uint32_t id
)
266 struct nhg_hash_entry lookup
= {};
269 return hash_lookup(zrouter
.nhgs_id
, &lookup
);
272 static int zebra_nhg_insert_id(struct nhg_hash_entry
*nhe
)
274 if (hash_lookup(zrouter
.nhgs_id
, nhe
)) {
276 EC_ZEBRA_NHG_TABLE_INSERT_FAILED
,
277 "Failed inserting NHG id=%u into the ID hash table, entry already exists",
282 hash_get(zrouter
.nhgs_id
, nhe
, hash_alloc_intern
);
287 static void zebra_nhg_set_if(struct nhg_hash_entry
*nhe
, struct interface
*ifp
)
290 if_nhg_dependents_add(ifp
, nhe
);
294 zebra_nhg_connect_depends(struct nhg_hash_entry
*nhe
,
295 struct nhg_connected_tree_head nhg_depends
)
297 struct nhg_connected
*rb_node_dep
= NULL
;
299 /* This has been allocated higher above in the stack. Could probably
300 * re-allocate and free the old stuff but just using the same memory
301 * for now. Otherwise, their might be a time trade-off for repeated
302 * alloc/frees as startup.
304 nhe
->nhg_depends
= nhg_depends
;
306 /* Attach backpointer to anything that it depends on */
307 zebra_nhg_dependents_init(nhe
);
308 if (!zebra_nhg_depends_is_empty(nhe
)) {
309 frr_each(nhg_connected_tree
, &nhe
->nhg_depends
, rb_node_dep
) {
310 zebra_nhg_dependents_add(rb_node_dep
->nhe
, nhe
);
314 /* Add the ifp now if its not a group or recursive and has ifindex */
315 if (zebra_nhg_depends_is_empty(nhe
) && nhe
->nhg
->nexthop
316 && nhe
->nhg
->nexthop
->ifindex
) {
317 struct interface
*ifp
= NULL
;
319 ifp
= if_lookup_by_index(nhe
->nhg
->nexthop
->ifindex
,
322 zebra_nhg_set_if(nhe
, ifp
);
325 EC_ZEBRA_IF_LOOKUP_FAILED
,
326 "Zebra failed to lookup an interface with ifindex=%d in vrf=%u for NHE id=%u",
327 nhe
->nhg
->nexthop
->ifindex
, nhe
->vrf_id
,
332 struct nhg_hash_entry
*zebra_nhg_alloc(void)
334 struct nhg_hash_entry
*nhe
;
336 nhe
= XCALLOC(MTYPE_NHG
, sizeof(struct nhg_hash_entry
));
341 static struct nhg_hash_entry
*zebra_nhg_copy(const struct nhg_hash_entry
*copy
,
344 struct nhg_hash_entry
*nhe
;
346 nhe
= zebra_nhg_alloc();
350 nhe
->nhg
= nexthop_group_new();
351 nexthop_group_copy(nhe
->nhg
, copy
->nhg
);
353 nhe
->vrf_id
= copy
->vrf_id
;
354 nhe
->afi
= copy
->afi
;
355 nhe
->type
= copy
->type
? copy
->type
: ZEBRA_ROUTE_NHG
;
357 nhe
->dplane_ref
= zebra_router_get_next_sequence();
362 /* Allocation via hash handler */
363 static void *zebra_nhg_hash_alloc(void *arg
)
365 struct nhg_hash_entry
*nhe
= NULL
;
366 struct nhg_hash_entry
*copy
= arg
;
368 nhe
= zebra_nhg_copy(copy
, copy
->id
);
370 /* Mark duplicate nexthops in a group at creation time. */
371 nexthop_group_mark_duplicates(nhe
->nhg
);
373 zebra_nhg_connect_depends(nhe
, copy
->nhg_depends
);
374 zebra_nhg_insert_id(nhe
);
379 uint32_t zebra_nhg_hash_key(const void *arg
)
381 const struct nhg_hash_entry
*nhe
= arg
;
383 uint32_t key
= 0x5a351234;
385 key
= jhash_3words(nhe
->vrf_id
, nhe
->afi
, nexthop_group_hash(nhe
->nhg
),
391 uint32_t zebra_nhg_id_key(const void *arg
)
393 const struct nhg_hash_entry
*nhe
= arg
;
398 bool zebra_nhg_hash_equal(const void *arg1
, const void *arg2
)
400 const struct nhg_hash_entry
*nhe1
= arg1
;
401 const struct nhg_hash_entry
*nhe2
= arg2
;
402 struct nexthop
*nexthop1
;
403 struct nexthop
*nexthop2
;
405 /* No matter what if they equal IDs, assume equal */
406 if (nhe1
->id
&& nhe2
->id
&& (nhe1
->id
== nhe2
->id
))
409 if (nhe1
->vrf_id
!= nhe2
->vrf_id
)
412 if (nhe1
->afi
!= nhe2
->afi
)
415 /* Nexthops should be sorted */
416 for (nexthop1
= nhe1
->nhg
->nexthop
, nexthop2
= nhe2
->nhg
->nexthop
;
417 nexthop1
|| nexthop2
;
418 nexthop1
= nexthop1
->next
, nexthop2
= nexthop2
->next
) {
419 if (nexthop1
&& !nexthop2
)
422 if (!nexthop1
&& nexthop2
)
426 * We have to check the active flag of each individual one,
427 * not just the overall active_num. This solves the special case
428 * issue of a route with a nexthop group with one nexthop
429 * resolving to itself and thus marking it inactive. If we
430 * have two different routes each wanting to mark a different
431 * nexthop inactive, they need to hash to two different groups.
433 * If we just hashed on num_active, they would hash the same
434 * which is incorrect.
438 * -> 1.1.1.1 dummy1 (inactive)
443 * -> 1.1.2.1 dummy2 (inactive)
445 * Without checking each individual one, they would hash to
446 * the same group and both have 1.1.1.1 dummy1 marked inactive.
449 if (CHECK_FLAG(nexthop1
->flags
, NEXTHOP_FLAG_ACTIVE
)
450 != CHECK_FLAG(nexthop2
->flags
, NEXTHOP_FLAG_ACTIVE
))
453 if (!nexthop_same(nexthop1
, nexthop2
))
460 bool zebra_nhg_hash_id_equal(const void *arg1
, const void *arg2
)
462 const struct nhg_hash_entry
*nhe1
= arg1
;
463 const struct nhg_hash_entry
*nhe2
= arg2
;
465 return nhe1
->id
== nhe2
->id
;
468 static int zebra_nhg_process_grp(struct nexthop_group
*nhg
,
469 struct nhg_connected_tree_head
*depends
,
470 struct nh_grp
*grp
, uint8_t count
)
472 nhg_connected_tree_init(depends
);
474 for (int i
= 0; i
< count
; i
++) {
475 struct nhg_hash_entry
*depend
= NULL
;
476 /* We do not care about nexthop_grp.weight at
477 * this time. But we should figure out
478 * how to adapt this to our code in
481 depend
= depends_find_id_add(depends
, grp
[i
].id
);
486 "Received Nexthop Group from the kernel with a dependent Nexthop ID (%u) which we do not have in our table",
492 * If this is a nexthop with its own group
493 * dependencies, add them as well. Not sure its
494 * even possible to have a group within a group
498 copy_nexthops(&nhg
->nexthop
, depend
->nhg
->nexthop
, NULL
);
504 static void handle_recursive_depend(struct nhg_connected_tree_head
*nhg_depends
,
505 struct nexthop
*nh
, afi_t afi
)
507 struct nhg_hash_entry
*depend
= NULL
;
508 struct nexthop_group resolved_ng
= {};
510 resolved_ng
.nexthop
= nh
;
512 depend
= zebra_nhg_rib_find(0, &resolved_ng
, afi
);
515 depends_add(nhg_depends
, depend
);
518 static bool zebra_nhg_find(struct nhg_hash_entry
**nhe
, uint32_t id
,
519 struct nexthop_group
*nhg
,
520 struct nhg_connected_tree_head
*nhg_depends
,
521 vrf_id_t vrf_id
, afi_t afi
, int type
)
523 struct nhg_hash_entry lookup
= {};
525 uint32_t old_id_counter
= id_counter
;
527 bool created
= false;
528 bool recursive
= false;
531 * If it has an id at this point, we must have gotten it from the kernel
533 lookup
.id
= id
? id
: ++id_counter
;
535 lookup
.type
= type
? type
: ZEBRA_ROUTE_NHG
;
538 if (lookup
.nhg
->nexthop
->next
) {
539 /* Groups can have all vrfs and AF's in them */
540 lookup
.afi
= AFI_UNSPEC
;
541 lookup
.vrf_id
= VRF_DEFAULT
;
543 switch (lookup
.nhg
->nexthop
->type
) {
544 case (NEXTHOP_TYPE_IFINDEX
):
545 case (NEXTHOP_TYPE_BLACKHOLE
):
547 * This switch case handles setting the afi different
548 * for ipv4/v6 routes. Ifindex/blackhole nexthop
549 * objects cannot be ambiguous, they must be Address
550 * Family specific. If we get here, we will either use
551 * the AF of the route, or the one we got passed from
552 * here from the kernel.
556 case (NEXTHOP_TYPE_IPV4_IFINDEX
):
557 case (NEXTHOP_TYPE_IPV4
):
560 case (NEXTHOP_TYPE_IPV6_IFINDEX
):
561 case (NEXTHOP_TYPE_IPV6
):
562 lookup
.afi
= AFI_IP6
;
566 lookup
.vrf_id
= vrf_id
;
570 (*nhe
) = zebra_nhg_lookup_id(id
);
572 (*nhe
) = hash_lookup(zrouter
.nhgs
, &lookup
);
574 /* If it found an nhe in our tables, this new ID is unused */
576 id_counter
= old_id_counter
;
579 /* Only hash/lookup the depends if the first lookup
580 * fails to find something. This should hopefully save a
581 * lot of cycles for larger ecmp sizes.
584 /* If you don't want to hash on each nexthop in the
585 * nexthop group struct you can pass the depends
586 * directly. Kernel-side we do this since it just looks
589 lookup
.nhg_depends
= *nhg_depends
;
591 if (nhg
->nexthop
->next
) {
592 zebra_nhg_depends_init(&lookup
);
594 /* If its a group, create a dependency tree */
595 struct nexthop
*nh
= NULL
;
597 for (nh
= nhg
->nexthop
; nh
; nh
= nh
->next
)
598 depends_find_add(&lookup
.nhg_depends
,
600 } else if (CHECK_FLAG(nhg
->nexthop
->flags
,
601 NEXTHOP_FLAG_RECURSIVE
)) {
602 zebra_nhg_depends_init(&lookup
);
603 handle_recursive_depend(&lookup
.nhg_depends
,
604 nhg
->nexthop
->resolved
,
610 (*nhe
) = hash_get(zrouter
.nhgs
, &lookup
, zebra_nhg_hash_alloc
);
614 SET_FLAG((*nhe
)->flags
, NEXTHOP_GROUP_RECURSIVE
);
619 /* Find/create a single nexthop */
620 static struct nhg_hash_entry
*
621 zebra_nhg_find_nexthop(uint32_t id
, struct nexthop
*nh
, afi_t afi
, int type
)
623 struct nhg_hash_entry
*nhe
= NULL
;
624 struct nexthop_group nhg
= {};
626 nexthop_group_add_sorted(&nhg
, nh
);
628 zebra_nhg_find(&nhe
, id
, &nhg
, NULL
, nh
->vrf_id
, afi
, type
);
633 static uint32_t nhg_ctx_get_id(const struct nhg_ctx
*ctx
)
638 static void nhg_ctx_set_status(struct nhg_ctx
*ctx
, enum nhg_ctx_status status
)
640 ctx
->status
= status
;
643 static enum nhg_ctx_status
nhg_ctx_get_status(const struct nhg_ctx
*ctx
)
648 static void nhg_ctx_set_op(struct nhg_ctx
*ctx
, enum nhg_ctx_op_e op
)
653 static enum nhg_ctx_op_e
nhg_ctx_get_op(const struct nhg_ctx
*ctx
)
658 static vrf_id_t
nhg_ctx_get_vrf_id(const struct nhg_ctx
*ctx
)
663 static int nhg_ctx_get_type(const struct nhg_ctx
*ctx
)
668 static int nhg_ctx_get_afi(const struct nhg_ctx
*ctx
)
673 static struct nexthop
*nhg_ctx_get_nh(struct nhg_ctx
*ctx
)
678 static uint8_t nhg_ctx_get_count(const struct nhg_ctx
*ctx
)
683 static struct nh_grp
*nhg_ctx_get_grp(struct nhg_ctx
*ctx
)
688 static struct nhg_ctx
*nhg_ctx_new()
690 struct nhg_ctx
*new = NULL
;
692 new = XCALLOC(MTYPE_NHG_CTX
, sizeof(struct nhg_ctx
));
697 static void nhg_ctx_free(struct nhg_ctx
**ctx
)
704 assert((*ctx
) != NULL
);
706 if (nhg_ctx_get_count(*ctx
))
709 nh
= nhg_ctx_get_nh(*ctx
);
711 nexthop_del_labels(nh
);
714 XFREE(MTYPE_NHG_CTX
, *ctx
);
718 static struct nhg_ctx
*nhg_ctx_init(uint32_t id
, struct nexthop
*nh
,
719 struct nh_grp
*grp
, vrf_id_t vrf_id
,
720 afi_t afi
, int type
, uint8_t count
)
722 struct nhg_ctx
*ctx
= NULL
;
727 ctx
->vrf_id
= vrf_id
;
733 /* Copy over the array */
734 memcpy(&ctx
->u
.grp
, grp
, count
* sizeof(struct nh_grp
));
741 static bool zebra_nhg_contains_unhashable(struct nhg_hash_entry
*nhe
)
743 struct nhg_connected
*rb_node_dep
= NULL
;
745 frr_each(nhg_connected_tree
, &nhe
->nhg_depends
, rb_node_dep
) {
746 if (CHECK_FLAG(rb_node_dep
->nhe
->flags
,
747 NEXTHOP_GROUP_UNHASHABLE
))
754 static void zebra_nhg_set_unhashable(struct nhg_hash_entry
*nhe
)
756 SET_FLAG(nhe
->flags
, NEXTHOP_GROUP_UNHASHABLE
);
757 SET_FLAG(nhe
->flags
, NEXTHOP_GROUP_INSTALLED
);
760 EC_ZEBRA_DUPLICATE_NHG_MESSAGE
,
761 "Nexthop Group with ID (%d) is a duplicate, therefore unhashable, ignoring",
765 static void zebra_nhg_set_valid(struct nhg_hash_entry
*nhe
)
767 struct nhg_connected
*rb_node_dep
;
769 SET_FLAG(nhe
->flags
, NEXTHOP_GROUP_VALID
);
771 frr_each(nhg_connected_tree
, &nhe
->nhg_dependents
, rb_node_dep
)
772 zebra_nhg_set_valid(rb_node_dep
->nhe
);
775 static void zebra_nhg_set_invalid(struct nhg_hash_entry
*nhe
)
777 struct nhg_connected
*rb_node_dep
;
779 UNSET_FLAG(nhe
->flags
, NEXTHOP_GROUP_VALID
);
781 /* Update validity of nexthops depending on it */
782 frr_each(nhg_connected_tree
, &nhe
->nhg_dependents
, rb_node_dep
)
783 zebra_nhg_check_valid(rb_node_dep
->nhe
);
786 void zebra_nhg_check_valid(struct nhg_hash_entry
*nhe
)
788 struct nhg_connected
*rb_node_dep
= NULL
;
791 /* If anthing else in the group is valid, the group is valid */
792 frr_each(nhg_connected_tree
, &nhe
->nhg_depends
, rb_node_dep
) {
793 if (CHECK_FLAG(rb_node_dep
->nhe
->flags
, NEXTHOP_GROUP_VALID
)) {
801 zebra_nhg_set_valid(nhe
);
803 zebra_nhg_set_invalid(nhe
);
807 static void zebra_nhg_release(struct nhg_hash_entry
*nhe
)
809 /* Remove it from any lists it may be on */
810 zebra_nhg_depends_release(nhe
);
811 zebra_nhg_dependents_release(nhe
);
813 if_nhg_dependents_del(nhe
->ifp
, nhe
);
816 * If its unhashable, we didn't store it here and have to be
817 * sure we don't clear one thats actually being used.
819 if (!CHECK_FLAG(nhe
->flags
, NEXTHOP_GROUP_UNHASHABLE
))
820 hash_release(zrouter
.nhgs
, nhe
);
822 hash_release(zrouter
.nhgs_id
, nhe
);
825 static void zebra_nhg_handle_uninstall(struct nhg_hash_entry
*nhe
)
827 zebra_nhg_release(nhe
);
831 static void zebra_nhg_handle_install(struct nhg_hash_entry
*nhe
)
833 /* Update validity of groups depending on it */
834 struct nhg_connected
*rb_node_dep
;
836 frr_each_safe(nhg_connected_tree
, &nhe
->nhg_dependents
, rb_node_dep
)
837 zebra_nhg_set_valid(rb_node_dep
->nhe
);
841 * The kernel/other program has changed the state of a nexthop object we are
844 static void zebra_nhg_handle_kernel_state_change(struct nhg_hash_entry
*nhe
,
850 "Kernel %s a nexthop group with ID (%u) that we are still using for a route, sending it back down",
851 (is_delete
? "deleted" : "updated"), nhe
->id
);
853 UNSET_FLAG(nhe
->flags
, NEXTHOP_GROUP_INSTALLED
);
854 zebra_nhg_install_kernel(nhe
);
856 zebra_nhg_handle_uninstall(nhe
);
859 static int nhg_ctx_process_new(struct nhg_ctx
*ctx
)
861 struct nexthop_group
*nhg
= NULL
;
862 struct nhg_connected_tree_head nhg_depends
= {};
863 struct nhg_hash_entry
*lookup
= NULL
;
864 struct nhg_hash_entry
*nhe
= NULL
;
866 uint32_t id
= nhg_ctx_get_id(ctx
);
867 uint8_t count
= nhg_ctx_get_count(ctx
);
868 vrf_id_t vrf_id
= nhg_ctx_get_vrf_id(ctx
);
869 int type
= nhg_ctx_get_type(ctx
);
870 afi_t afi
= nhg_ctx_get_afi(ctx
);
872 lookup
= zebra_nhg_lookup_id(id
);
875 /* This is already present in our table, hence an update
876 * that we did not initate.
878 zebra_nhg_handle_kernel_state_change(lookup
, false);
882 if (nhg_ctx_get_count(ctx
)) {
883 nhg
= nexthop_group_new();
884 if (zebra_nhg_process_grp(nhg
, &nhg_depends
,
885 nhg_ctx_get_grp(ctx
), count
)) {
886 depends_decrement_free(&nhg_depends
);
887 nexthop_group_delete(&nhg
);
891 if (!zebra_nhg_find(&nhe
, id
, nhg
, &nhg_depends
, vrf_id
, type
,
893 depends_decrement_free(&nhg_depends
);
895 /* These got copied over in zebra_nhg_alloc() */
896 nexthop_group_delete(&nhg
);
898 nhe
= zebra_nhg_find_nexthop(id
, nhg_ctx_get_nh(ctx
), afi
,
903 struct nhg_hash_entry
*kernel_nhe
= NULL
;
905 /* Duplicate but with different ID from
909 /* The kernel allows duplicate nexthops
910 * as long as they have different IDs.
911 * We are ignoring those to prevent
912 * syncing problems with the kernel
915 * We maintain them *ONLY* in the ID hash table to
916 * track them and set the flag to indicated
917 * their attributes are unhashable.
920 kernel_nhe
= zebra_nhg_copy(nhe
, id
);
921 zebra_nhg_insert_id(kernel_nhe
);
922 zebra_nhg_set_unhashable(kernel_nhe
);
923 } else if (zebra_nhg_contains_unhashable(nhe
)) {
924 /* The group we got contains an unhashable/duplicated
925 * depend, so lets mark this group as unhashable as well
926 * and release it from the non-ID hash.
928 hash_release(zrouter
.nhgs
, nhe
);
929 zebra_nhg_set_unhashable(nhe
);
931 /* It actually created a new nhe */
932 SET_FLAG(nhe
->flags
, NEXTHOP_GROUP_VALID
);
933 SET_FLAG(nhe
->flags
, NEXTHOP_GROUP_INSTALLED
);
937 EC_ZEBRA_TABLE_LOOKUP_FAILED
,
938 "Zebra failed to find or create a nexthop hash entry for ID (%u)",
946 static int nhg_ctx_process_del(struct nhg_ctx
*ctx
)
948 struct nhg_hash_entry
*nhe
= NULL
;
949 uint32_t id
= nhg_ctx_get_id(ctx
);
951 nhe
= zebra_nhg_lookup_id(id
);
955 EC_ZEBRA_BAD_NHG_MESSAGE
,
956 "Kernel delete message received for nexthop group ID (%u) that we do not have in our ID table",
961 zebra_nhg_handle_kernel_state_change(nhe
, true);
966 static void nhg_ctx_fini(struct nhg_ctx
**ctx
)
969 * Just freeing for now, maybe do something more in the future
976 static int queue_add(struct nhg_ctx
*ctx
)
978 /* If its queued or already processed do nothing */
979 if (nhg_ctx_get_status(ctx
) == NHG_CTX_QUEUED
)
982 if (rib_queue_nhg_add(ctx
)) {
983 nhg_ctx_set_status(ctx
, NHG_CTX_FAILURE
);
987 nhg_ctx_set_status(ctx
, NHG_CTX_QUEUED
);
992 int nhg_ctx_process(struct nhg_ctx
*ctx
)
996 switch (nhg_ctx_get_op(ctx
)) {
998 ret
= nhg_ctx_process_new(ctx
);
999 if (nhg_ctx_get_count(ctx
) && ret
== -ENOENT
1000 && nhg_ctx_get_status(ctx
) != NHG_CTX_REQUEUED
) {
1002 * We have entered a situation where we are
1003 * processing a group from the kernel
1004 * that has a contained nexthop which
1005 * we have not yet processed.
1007 * Re-enqueue this ctx to be handled exactly one
1008 * more time (indicated by the flag).
1010 * By the time we get back to it, we
1011 * should have processed its depends.
1013 nhg_ctx_set_status(ctx
, NHG_CTX_NONE
);
1014 if (queue_add(ctx
) == 0) {
1015 nhg_ctx_set_status(ctx
, NHG_CTX_REQUEUED
);
1020 case NHG_CTX_OP_DEL
:
1021 ret
= nhg_ctx_process_del(ctx
);
1022 case NHG_CTX_OP_NONE
:
1026 nhg_ctx_set_status(ctx
, (ret
? NHG_CTX_FAILURE
: NHG_CTX_SUCCESS
));
1033 /* Kernel-side, you either get a single new nexthop or a array of ID's */
1034 int zebra_nhg_kernel_find(uint32_t id
, struct nexthop
*nh
, struct nh_grp
*grp
,
1035 uint8_t count
, vrf_id_t vrf_id
, afi_t afi
, int type
,
1038 struct nhg_ctx
*ctx
= NULL
;
1040 if (id
> id_counter
)
1041 /* Increase our counter so we don't try to create
1042 * an ID that already exists
1046 ctx
= nhg_ctx_init(id
, nh
, grp
, vrf_id
, afi
, type
, count
);
1047 nhg_ctx_set_op(ctx
, NHG_CTX_OP_NEW
);
1049 /* Under statup conditions, we need to handle them immediately
1050 * like we do for routes. Otherwise, we are going to get a route
1051 * with a nhe_id that we have not handled.
1054 return nhg_ctx_process(ctx
);
1056 if (queue_add(ctx
)) {
1064 /* Kernel-side, received delete message */
1065 int zebra_nhg_kernel_del(uint32_t id
)
1067 struct nhg_ctx
*ctx
= NULL
;
1069 ctx
= nhg_ctx_init(id
, NULL
, NULL
, 0, 0, 0, 0);
1071 nhg_ctx_set_op(ctx
, NHG_CTX_OP_DEL
);
1073 if (queue_add(ctx
)) {
1081 /* Some dependency helper functions */
1082 static struct nhg_hash_entry
*depends_find_recursive(const struct nexthop
*nh
,
1085 struct nhg_hash_entry
*nhe
;
1086 struct nexthop
*lookup
= NULL
;
1088 lookup
= nexthop_dup(nh
, NULL
);
1090 nhe
= zebra_nhg_find_nexthop(0, lookup
, afi
, 0);
1092 nexthops_free(lookup
);
1097 static struct nhg_hash_entry
*depends_find_singleton(const struct nexthop
*nh
,
1100 struct nhg_hash_entry
*nhe
;
1101 struct nexthop lookup
= {};
1103 /* Capture a snapshot of this single nh; it might be part of a list,
1104 * so we need to make a standalone copy.
1106 nexthop_copy_no_recurse(&lookup
, nh
, NULL
);
1108 nhe
= zebra_nhg_find_nexthop(0, &lookup
, afi
, 0);
1110 /* The copy may have allocated labels; free them if necessary. */
1111 nexthop_del_labels(&lookup
);
1116 static struct nhg_hash_entry
*depends_find(const struct nexthop
*nh
, afi_t afi
)
1118 struct nhg_hash_entry
*nhe
= NULL
;
1123 /* We are separating these functions out to increase handling speed
1124 * in the non-recursive case (by not alloc/freeing)
1126 if (CHECK_FLAG(nh
->flags
, NEXTHOP_FLAG_RECURSIVE
))
1127 nhe
= depends_find_recursive(nh
, afi
);
1129 nhe
= depends_find_singleton(nh
, afi
);
1135 static void depends_add(struct nhg_connected_tree_head
*head
,
1136 struct nhg_hash_entry
*depend
)
1138 /* If NULL is returned, it was successfully added and
1139 * needs to have its refcnt incremented.
1141 * Else the NHE is already present in the tree and doesn't
1142 * need to increment the refcnt.
1144 if (nhg_connected_tree_add_nhe(head
, depend
) == NULL
)
1145 zebra_nhg_increment_ref(depend
);
1148 static struct nhg_hash_entry
*
1149 depends_find_add(struct nhg_connected_tree_head
*head
, struct nexthop
*nh
,
1152 struct nhg_hash_entry
*depend
= NULL
;
1154 depend
= depends_find(nh
, afi
);
1157 depends_add(head
, depend
);
1162 static struct nhg_hash_entry
*
1163 depends_find_id_add(struct nhg_connected_tree_head
*head
, uint32_t id
)
1165 struct nhg_hash_entry
*depend
= NULL
;
1167 depend
= zebra_nhg_lookup_id(id
);
1170 depends_add(head
, depend
);
1175 static void depends_decrement_free(struct nhg_connected_tree_head
*head
)
1177 nhg_connected_tree_decrement_ref(head
);
1178 nhg_connected_tree_free(head
);
1181 /* Rib-side, you get a nexthop group struct */
1182 struct nhg_hash_entry
*
1183 zebra_nhg_rib_find(uint32_t id
, struct nexthop_group
*nhg
, afi_t rt_afi
)
1185 struct nhg_hash_entry
*nhe
= NULL
;
1187 if (!(nhg
&& nhg
->nexthop
)) {
1188 flog_err(EC_ZEBRA_TABLE_LOOKUP_FAILED
,
1189 "No nexthop passed to %s", __func__
);
1193 zebra_nhg_find(&nhe
, id
, nhg
, NULL
, nhg
->nexthop
->vrf_id
, rt_afi
, 0);
1198 static void zebra_nhg_free_members(struct nhg_hash_entry
*nhe
)
1200 nexthop_group_delete(&nhe
->nhg
);
1201 /* Decrement to remove connection ref */
1202 nhg_connected_tree_decrement_ref(&nhe
->nhg_depends
);
1203 nhg_connected_tree_free(&nhe
->nhg_depends
);
1204 nhg_connected_tree_free(&nhe
->nhg_dependents
);
1207 void zebra_nhg_free(struct nhg_hash_entry
*nhe
)
1210 zlog_debug("nhe_id=%u hash refcnt=%d", nhe
->id
, nhe
->refcnt
);
1212 zebra_nhg_free_members(nhe
);
1214 XFREE(MTYPE_NHG
, nhe
);
1217 void zebra_nhg_hash_free(void *p
)
1219 zebra_nhg_free((struct nhg_hash_entry
*)p
);
1222 void zebra_nhg_decrement_ref(struct nhg_hash_entry
*nhe
)
1226 if (!zebra_nhg_depends_is_empty(nhe
))
1227 nhg_connected_tree_decrement_ref(&nhe
->nhg_depends
);
1229 if (ZEBRA_NHG_CREATED(nhe
) && nhe
->refcnt
<= 0)
1230 zebra_nhg_uninstall_kernel(nhe
);
1233 void zebra_nhg_increment_ref(struct nhg_hash_entry
*nhe
)
1237 if (!zebra_nhg_depends_is_empty(nhe
))
1238 nhg_connected_tree_increment_ref(&nhe
->nhg_depends
);
1241 static void nexthop_set_resolved(afi_t afi
, const struct nexthop
*newhop
,
1242 struct nexthop
*nexthop
)
1244 struct nexthop
*resolved_hop
;
1245 uint8_t num_labels
= 0;
1246 mpls_label_t labels
[MPLS_MAX_LABELS
];
1247 enum lsp_types_t label_type
= ZEBRA_LSP_NONE
;
1250 resolved_hop
= nexthop_new();
1251 SET_FLAG(resolved_hop
->flags
, NEXTHOP_FLAG_ACTIVE
);
1253 resolved_hop
->vrf_id
= nexthop
->vrf_id
;
1254 switch (newhop
->type
) {
1255 case NEXTHOP_TYPE_IPV4
:
1256 case NEXTHOP_TYPE_IPV4_IFINDEX
:
1257 /* If the resolving route specifies a gateway, use it */
1258 resolved_hop
->type
= newhop
->type
;
1259 resolved_hop
->gate
.ipv4
= newhop
->gate
.ipv4
;
1261 if (newhop
->ifindex
) {
1262 resolved_hop
->type
= NEXTHOP_TYPE_IPV4_IFINDEX
;
1263 resolved_hop
->ifindex
= newhop
->ifindex
;
1266 case NEXTHOP_TYPE_IPV6
:
1267 case NEXTHOP_TYPE_IPV6_IFINDEX
:
1268 resolved_hop
->type
= newhop
->type
;
1269 resolved_hop
->gate
.ipv6
= newhop
->gate
.ipv6
;
1271 if (newhop
->ifindex
) {
1272 resolved_hop
->type
= NEXTHOP_TYPE_IPV6_IFINDEX
;
1273 resolved_hop
->ifindex
= newhop
->ifindex
;
1276 case NEXTHOP_TYPE_IFINDEX
:
1277 /* If the resolving route is an interface route,
1278 * it means the gateway we are looking up is connected
1279 * to that interface. (The actual network is _not_ onlink).
1280 * Therefore, the resolved route should have the original
1281 * gateway as nexthop as it is directly connected.
1283 * On Linux, we have to set the onlink netlink flag because
1284 * otherwise, the kernel won't accept the route.
1286 resolved_hop
->flags
|= NEXTHOP_FLAG_ONLINK
;
1287 if (afi
== AFI_IP
) {
1288 resolved_hop
->type
= NEXTHOP_TYPE_IPV4_IFINDEX
;
1289 resolved_hop
->gate
.ipv4
= nexthop
->gate
.ipv4
;
1290 } else if (afi
== AFI_IP6
) {
1291 resolved_hop
->type
= NEXTHOP_TYPE_IPV6_IFINDEX
;
1292 resolved_hop
->gate
.ipv6
= nexthop
->gate
.ipv6
;
1294 resolved_hop
->ifindex
= newhop
->ifindex
;
1296 case NEXTHOP_TYPE_BLACKHOLE
:
1297 resolved_hop
->type
= NEXTHOP_TYPE_BLACKHOLE
;
1298 resolved_hop
->bh_type
= newhop
->bh_type
;
1302 if (newhop
->flags
& NEXTHOP_FLAG_ONLINK
)
1303 resolved_hop
->flags
|= NEXTHOP_FLAG_ONLINK
;
1305 /* Copy labels of the resolved route and the parent resolving to it */
1306 if (newhop
->nh_label
) {
1307 for (i
= 0; i
< newhop
->nh_label
->num_labels
; i
++)
1308 labels
[num_labels
++] = newhop
->nh_label
->label
[i
];
1309 label_type
= newhop
->nh_label_type
;
1312 if (nexthop
->nh_label
) {
1313 for (i
= 0; i
< nexthop
->nh_label
->num_labels
; i
++)
1314 labels
[num_labels
++] = nexthop
->nh_label
->label
[i
];
1316 /* If the parent has labels, use its type */
1317 label_type
= nexthop
->nh_label_type
;
1321 nexthop_add_labels(resolved_hop
, label_type
, num_labels
,
1324 resolved_hop
->rparent
= nexthop
;
1325 _nexthop_add(&nexthop
->resolved
, resolved_hop
);
1328 /* Checks if nexthop we are trying to resolve to is valid */
1329 static bool nexthop_valid_resolve(const struct nexthop
*nexthop
,
1330 const struct nexthop
*resolved
)
1332 /* Can't resolve to a recursive nexthop */
1333 if (CHECK_FLAG(resolved
->flags
, NEXTHOP_FLAG_RECURSIVE
))
1336 switch (nexthop
->type
) {
1337 case NEXTHOP_TYPE_IPV4_IFINDEX
:
1338 case NEXTHOP_TYPE_IPV6_IFINDEX
:
1339 /* If the nexthop we are resolving to does not match the
1340 * ifindex for the nexthop the route wanted, its not valid.
1342 if (nexthop
->ifindex
!= resolved
->ifindex
)
1345 case NEXTHOP_TYPE_IPV4
:
1346 case NEXTHOP_TYPE_IPV6
:
1347 case NEXTHOP_TYPE_IFINDEX
:
1348 case NEXTHOP_TYPE_BLACKHOLE
:
1356 * Given a nexthop we need to properly recursively resolve
1357 * the route. As such, do a table lookup to find and match
1358 * if at all possible. Set the nexthop->ifindex and resolved_id
1361 static int nexthop_active(afi_t afi
, struct route_entry
*re
,
1362 struct nexthop
*nexthop
, struct route_node
*top
)
1365 struct route_table
*table
;
1366 struct route_node
*rn
;
1367 struct route_entry
*match
= NULL
;
1369 struct nexthop
*newhop
;
1370 struct interface
*ifp
;
1372 struct zebra_vrf
*zvrf
;
1374 if ((nexthop
->type
== NEXTHOP_TYPE_IPV4
)
1375 || nexthop
->type
== NEXTHOP_TYPE_IPV6
)
1376 nexthop
->ifindex
= 0;
1379 UNSET_FLAG(nexthop
->flags
, NEXTHOP_FLAG_RECURSIVE
);
1380 nexthops_free(nexthop
->resolved
);
1381 nexthop
->resolved
= NULL
;
1382 re
->nexthop_mtu
= 0;
1385 * If the kernel has sent us a NEW route, then
1386 * by golly gee whiz it's a good route.
1388 * If its an already INSTALLED route we have already handled, then the
1389 * kernel route's nexthop might have became unreachable
1390 * and we have to handle that.
1392 if (!CHECK_FLAG(re
->status
, ROUTE_ENTRY_INSTALLED
)
1393 && (re
->type
== ZEBRA_ROUTE_KERNEL
1394 || re
->type
== ZEBRA_ROUTE_SYSTEM
))
1398 * Check to see if we should trust the passed in information
1399 * for UNNUMBERED interfaces as that we won't find the GW
1400 * address in the routing table.
1401 * This check should suffice to handle IPv4 or IPv6 routes
1402 * sourced from EVPN routes which are installed with the
1403 * next hop as the remote VTEP IP.
1405 if (CHECK_FLAG(nexthop
->flags
, NEXTHOP_FLAG_ONLINK
)) {
1406 ifp
= if_lookup_by_index(nexthop
->ifindex
, nexthop
->vrf_id
);
1408 if (IS_ZEBRA_DEBUG_RIB_DETAILED
)
1410 "\t%s: Onlink and interface: %u[%u] does not exist",
1411 __PRETTY_FUNCTION__
, nexthop
->ifindex
,
1415 if (connected_is_unnumbered(ifp
)) {
1416 if (if_is_operative(ifp
))
1419 if (IS_ZEBRA_DEBUG_RIB_DETAILED
)
1421 "\t%s: Onlink and interface %s is not operative",
1422 __PRETTY_FUNCTION__
, ifp
->name
);
1425 if (!if_is_operative(ifp
)) {
1426 if (IS_ZEBRA_DEBUG_RIB_DETAILED
)
1428 "\t%s: Interface %s is not unnumbered",
1429 __PRETTY_FUNCTION__
, ifp
->name
);
1434 if ((top
->p
.family
== AF_INET
&& top
->p
.prefixlen
== 32
1435 && nexthop
->gate
.ipv4
.s_addr
== top
->p
.u
.prefix4
.s_addr
)
1436 || (top
->p
.family
== AF_INET6
&& top
->p
.prefixlen
== 128
1437 && memcmp(&nexthop
->gate
.ipv6
, &top
->p
.u
.prefix6
, 16) == 0)) {
1438 if (IS_ZEBRA_DEBUG_RIB_DETAILED
)
1440 "\t:%s: Attempting to install a max prefixlength route through itself",
1441 __PRETTY_FUNCTION__
);
1445 /* Make lookup prefix. */
1446 memset(&p
, 0, sizeof(struct prefix
));
1450 p
.prefixlen
= IPV4_MAX_PREFIXLEN
;
1451 p
.u
.prefix4
= nexthop
->gate
.ipv4
;
1454 p
.family
= AF_INET6
;
1455 p
.prefixlen
= IPV6_MAX_PREFIXLEN
;
1456 p
.u
.prefix6
= nexthop
->gate
.ipv6
;
1459 assert(afi
!= AFI_IP
&& afi
!= AFI_IP6
);
1463 table
= zebra_vrf_table(afi
, SAFI_UNICAST
, nexthop
->vrf_id
);
1465 zvrf
= zebra_vrf_lookup_by_id(nexthop
->vrf_id
);
1466 if (!table
|| !zvrf
) {
1467 if (IS_ZEBRA_DEBUG_RIB_DETAILED
)
1468 zlog_debug("\t%s: Table not found",
1469 __PRETTY_FUNCTION__
);
1473 rn
= route_node_match(table
, (struct prefix
*)&p
);
1475 route_unlock_node(rn
);
1477 /* Lookup should halt if we've matched against ourselves ('top',
1478 * if specified) - i.e., we cannot have a nexthop NH1 is
1479 * resolved by a route NH1. The exception is if the route is a
1482 if (top
&& rn
== top
)
1483 if (((afi
== AFI_IP
) && (rn
->p
.prefixlen
!= 32))
1484 || ((afi
== AFI_IP6
) && (rn
->p
.prefixlen
!= 128))) {
1485 if (IS_ZEBRA_DEBUG_RIB_DETAILED
)
1487 "\t%s: Matched against ourself and prefix length is not max bit length",
1488 __PRETTY_FUNCTION__
);
1492 /* Pick up selected route. */
1493 /* However, do not resolve over default route unless explicitly
1496 if (is_default_prefix(&rn
->p
)
1497 && !rnh_resolve_via_default(zvrf
, p
.family
)) {
1498 if (IS_ZEBRA_DEBUG_RIB_DETAILED
)
1500 "\t:%s: Resolved against default route",
1501 __PRETTY_FUNCTION__
);
1505 dest
= rib_dest_from_rnode(rn
);
1506 if (dest
&& dest
->selected_fib
1507 && !CHECK_FLAG(dest
->selected_fib
->status
,
1508 ROUTE_ENTRY_REMOVED
)
1509 && dest
->selected_fib
->type
!= ZEBRA_ROUTE_TABLE
)
1510 match
= dest
->selected_fib
;
1512 /* If there is no selected route or matched route is EGP, go up
1518 } while (rn
&& rn
->info
== NULL
);
1520 route_lock_node(rn
);
1525 if (match
->type
== ZEBRA_ROUTE_CONNECT
) {
1526 /* Directly point connected route. */
1527 newhop
= match
->nhe
->nhg
->nexthop
;
1529 if (nexthop
->type
== NEXTHOP_TYPE_IPV4
1530 || nexthop
->type
== NEXTHOP_TYPE_IPV6
)
1531 nexthop
->ifindex
= newhop
->ifindex
;
1534 } else if (CHECK_FLAG(re
->flags
, ZEBRA_FLAG_ALLOW_RECURSION
)) {
1536 for (ALL_NEXTHOPS_PTR(match
->nhe
->nhg
, newhop
)) {
1537 if (!CHECK_FLAG(match
->status
,
1538 ROUTE_ENTRY_INSTALLED
))
1540 if (!nexthop_valid_resolve(nexthop
, newhop
))
1543 SET_FLAG(nexthop
->flags
,
1544 NEXTHOP_FLAG_RECURSIVE
);
1545 nexthop_set_resolved(afi
, newhop
, nexthop
);
1549 re
->nexthop_mtu
= match
->mtu
;
1551 if (!resolved
&& IS_ZEBRA_DEBUG_RIB_DETAILED
)
1552 zlog_debug("\t%s: Recursion failed to find",
1553 __PRETTY_FUNCTION__
);
1555 } else if (re
->type
== ZEBRA_ROUTE_STATIC
) {
1557 for (ALL_NEXTHOPS_PTR(match
->nhe
->nhg
, newhop
)) {
1558 if (!CHECK_FLAG(match
->status
,
1559 ROUTE_ENTRY_INSTALLED
))
1561 if (!nexthop_valid_resolve(nexthop
, newhop
))
1564 SET_FLAG(nexthop
->flags
,
1565 NEXTHOP_FLAG_RECURSIVE
);
1566 nexthop_set_resolved(afi
, newhop
, nexthop
);
1570 re
->nexthop_mtu
= match
->mtu
;
1572 if (!resolved
&& IS_ZEBRA_DEBUG_RIB_DETAILED
)
1574 "\t%s: Static route unable to resolve",
1575 __PRETTY_FUNCTION__
);
1578 if (IS_ZEBRA_DEBUG_RIB_DETAILED
) {
1580 "\t%s: Route Type %s has not turned on recursion",
1581 __PRETTY_FUNCTION__
,
1582 zebra_route_string(re
->type
));
1583 if (re
->type
== ZEBRA_ROUTE_BGP
1584 && !CHECK_FLAG(re
->flags
, ZEBRA_FLAG_IBGP
))
1586 "\tEBGP: see \"disable-ebgp-connected-route-check\" or \"disable-connected-check\"");
1591 if (IS_ZEBRA_DEBUG_RIB_DETAILED
)
1592 zlog_debug("\t%s: Nexthop did not lookup in table",
1593 __PRETTY_FUNCTION__
);
1597 /* This function verifies reachability of one given nexthop, which can be
1598 * numbered or unnumbered, IPv4 or IPv6. The result is unconditionally stored
1599 * in nexthop->flags field. The nexthop->ifindex will be updated
1600 * appropriately as well. An existing route map can turn
1601 * (otherwise active) nexthop into inactive, but not vice versa.
1603 * If it finds a nexthop recursivedly, set the resolved_id
1604 * to match that nexthop's nhg_hash_entry ID;
1606 * The return value is the final value of 'ACTIVE' flag.
1608 static unsigned nexthop_active_check(struct route_node
*rn
,
1609 struct route_entry
*re
,
1610 struct nexthop
*nexthop
)
1612 struct interface
*ifp
;
1613 route_map_result_t ret
= RMAP_PERMITMATCH
;
1615 char buf
[SRCDEST2STR_BUFFER
];
1616 const struct prefix
*p
, *src_p
;
1617 struct zebra_vrf
*zvrf
;
1619 srcdest_rnode_prefixes(rn
, &p
, &src_p
);
1621 if (rn
->p
.family
== AF_INET
)
1623 else if (rn
->p
.family
== AF_INET6
)
1627 switch (nexthop
->type
) {
1628 case NEXTHOP_TYPE_IFINDEX
:
1629 ifp
= if_lookup_by_index(nexthop
->ifindex
, nexthop
->vrf_id
);
1630 if (ifp
&& if_is_operative(ifp
))
1631 SET_FLAG(nexthop
->flags
, NEXTHOP_FLAG_ACTIVE
);
1633 UNSET_FLAG(nexthop
->flags
, NEXTHOP_FLAG_ACTIVE
);
1635 case NEXTHOP_TYPE_IPV4
:
1636 case NEXTHOP_TYPE_IPV4_IFINDEX
:
1638 if (nexthop_active(AFI_IP
, re
, nexthop
, rn
))
1639 SET_FLAG(nexthop
->flags
, NEXTHOP_FLAG_ACTIVE
);
1641 UNSET_FLAG(nexthop
->flags
, NEXTHOP_FLAG_ACTIVE
);
1643 case NEXTHOP_TYPE_IPV6
:
1645 if (nexthop_active(AFI_IP6
, re
, nexthop
, rn
))
1646 SET_FLAG(nexthop
->flags
, NEXTHOP_FLAG_ACTIVE
);
1648 UNSET_FLAG(nexthop
->flags
, NEXTHOP_FLAG_ACTIVE
);
1650 case NEXTHOP_TYPE_IPV6_IFINDEX
:
1651 /* RFC 5549, v4 prefix with v6 NH */
1652 if (rn
->p
.family
!= AF_INET
)
1654 if (IN6_IS_ADDR_LINKLOCAL(&nexthop
->gate
.ipv6
)) {
1655 ifp
= if_lookup_by_index(nexthop
->ifindex
,
1657 if (ifp
&& if_is_operative(ifp
))
1658 SET_FLAG(nexthop
->flags
, NEXTHOP_FLAG_ACTIVE
);
1660 UNSET_FLAG(nexthop
->flags
, NEXTHOP_FLAG_ACTIVE
);
1662 if (nexthop_active(AFI_IP6
, re
, nexthop
, rn
))
1663 SET_FLAG(nexthop
->flags
, NEXTHOP_FLAG_ACTIVE
);
1665 UNSET_FLAG(nexthop
->flags
, NEXTHOP_FLAG_ACTIVE
);
1668 case NEXTHOP_TYPE_BLACKHOLE
:
1669 SET_FLAG(nexthop
->flags
, NEXTHOP_FLAG_ACTIVE
);
1674 if (!CHECK_FLAG(nexthop
->flags
, NEXTHOP_FLAG_ACTIVE
)) {
1675 if (IS_ZEBRA_DEBUG_RIB_DETAILED
)
1676 zlog_debug("\t%s: Unable to find a active nexthop",
1677 __PRETTY_FUNCTION__
);
1681 /* XXX: What exactly do those checks do? Do we support
1682 * e.g. IPv4 routes with IPv6 nexthops or vice versa?
1684 if (RIB_SYSTEM_ROUTE(re
) || (family
== AFI_IP
&& p
->family
!= AF_INET
)
1685 || (family
== AFI_IP6
&& p
->family
!= AF_INET6
))
1686 return CHECK_FLAG(nexthop
->flags
, NEXTHOP_FLAG_ACTIVE
);
1688 /* The original code didn't determine the family correctly
1689 * e.g. for NEXTHOP_TYPE_IFINDEX. Retrieve the correct afi
1690 * from the rib_table_info in those cases.
1691 * Possibly it may be better to use only the rib_table_info
1695 rib_table_info_t
*info
;
1697 info
= srcdest_rnode_table_info(rn
);
1701 memset(&nexthop
->rmap_src
.ipv6
, 0, sizeof(union g_addr
));
1703 zvrf
= zebra_vrf_lookup_by_id(nexthop
->vrf_id
);
1705 if (IS_ZEBRA_DEBUG_RIB_DETAILED
)
1706 zlog_debug("\t%s: zvrf is NULL", __PRETTY_FUNCTION__
);
1707 return CHECK_FLAG(nexthop
->flags
, NEXTHOP_FLAG_ACTIVE
);
1710 /* It'll get set if required inside */
1711 ret
= zebra_route_map_check(family
, re
->type
, re
->instance
, p
, nexthop
,
1713 if (ret
== RMAP_DENYMATCH
) {
1714 if (IS_ZEBRA_DEBUG_RIB
) {
1715 srcdest_rnode2str(rn
, buf
, sizeof(buf
));
1717 "%u:%s: Filtering out with NH out %s due to route map",
1719 ifindex2ifname(nexthop
->ifindex
,
1722 UNSET_FLAG(nexthop
->flags
, NEXTHOP_FLAG_ACTIVE
);
1724 return CHECK_FLAG(nexthop
->flags
, NEXTHOP_FLAG_ACTIVE
);
1728 * Iterate over all nexthops of the given RIB entry and refresh their
1729 * ACTIVE flag. If any nexthop is found to toggle the ACTIVE flag,
1730 * the whole re structure is flagged with ROUTE_ENTRY_CHANGED.
1732 * Return value is the new number of active nexthops.
1734 int nexthop_active_update(struct route_node
*rn
, struct route_entry
*re
)
1736 struct nexthop_group new_grp
= {};
1737 struct nexthop
*nexthop
;
1738 union g_addr prev_src
;
1739 unsigned int prev_active
, new_active
;
1740 ifindex_t prev_index
;
1741 uint8_t curr_active
= 0;
1743 afi_t rt_afi
= family2afi(rn
->p
.family
);
1745 UNSET_FLAG(re
->status
, ROUTE_ENTRY_CHANGED
);
1747 /* Copy over the nexthops in current state */
1748 nexthop_group_copy(&new_grp
, re
->nhe
->nhg
);
1750 for (nexthop
= new_grp
.nexthop
; nexthop
; nexthop
= nexthop
->next
) {
1752 /* No protocol daemon provides src and so we're skipping
1754 prev_src
= nexthop
->rmap_src
;
1755 prev_active
= CHECK_FLAG(nexthop
->flags
, NEXTHOP_FLAG_ACTIVE
);
1756 prev_index
= nexthop
->ifindex
;
1758 * We need to respect the multipath_num here
1759 * as that what we should be able to install from
1760 * a multipath perpsective should not be a data plane
1764 nexthop_active_check(rn
, re
, nexthop
);
1766 if (new_active
&& curr_active
>= zrouter
.multipath_num
) {
1769 /* Set it and its resolved nexthop as inactive. */
1770 for (nh
= nexthop
; nh
; nh
= nh
->resolved
)
1771 UNSET_FLAG(nh
->flags
, NEXTHOP_FLAG_ACTIVE
);
1779 /* Don't allow src setting on IPv6 addr for now */
1780 if (prev_active
!= new_active
|| prev_index
!= nexthop
->ifindex
1781 || ((nexthop
->type
>= NEXTHOP_TYPE_IFINDEX
1782 && nexthop
->type
< NEXTHOP_TYPE_IPV6
)
1783 && prev_src
.ipv4
.s_addr
1784 != nexthop
->rmap_src
.ipv4
.s_addr
)
1785 || ((nexthop
->type
>= NEXTHOP_TYPE_IPV6
1786 && nexthop
->type
< NEXTHOP_TYPE_BLACKHOLE
)
1787 && !(IPV6_ADDR_SAME(&prev_src
.ipv6
,
1788 &nexthop
->rmap_src
.ipv6
)))
1789 || CHECK_FLAG(re
->status
, ROUTE_ENTRY_LABELS_CHANGED
))
1790 SET_FLAG(re
->status
, ROUTE_ENTRY_CHANGED
);
1793 if (CHECK_FLAG(re
->status
, ROUTE_ENTRY_CHANGED
)) {
1794 struct nhg_hash_entry
*new_nhe
= NULL
;
1796 new_nhe
= zebra_nhg_rib_find(0, &new_grp
, rt_afi
);
1798 route_entry_update_nhe(re
, new_nhe
);
1802 struct nhg_hash_entry
*nhe
= NULL
;
1804 nhe
= zebra_nhg_lookup_id(re
->nhe_id
);
1807 SET_FLAG(nhe
->flags
, NEXTHOP_GROUP_VALID
);
1810 EC_ZEBRA_TABLE_LOOKUP_FAILED
,
1811 "Active update on NHE id=%u that we do not have in our tables",
1816 * Do not need these nexthops anymore since they
1817 * were either copied over into an nhe or not
1820 nexthops_free(new_grp
.nexthop
);
1824 /* Convert a nhe into a group array */
1825 uint8_t zebra_nhg_nhe2grp(struct nh_grp
*grp
, struct nhg_hash_entry
*nhe
,
1828 struct nhg_connected
*rb_node_dep
= NULL
;
1829 struct nhg_hash_entry
*depend
= NULL
;
1832 frr_each(nhg_connected_tree
, &nhe
->nhg_depends
, rb_node_dep
) {
1833 bool duplicate
= false;
1835 depend
= rb_node_dep
->nhe
;
1838 * If its recursive, use its resolved nhe in the group
1840 if (CHECK_FLAG(depend
->flags
, NEXTHOP_GROUP_RECURSIVE
)) {
1841 depend
= zebra_nhg_resolve(depend
);
1844 EC_ZEBRA_NHG_FIB_UPDATE
,
1845 "Failed to recursively resolve Nexthop Hash Entry in the group id=%u",
1851 /* Check for duplicate IDs, kernel doesn't like that */
1852 for (int j
= 0; j
< i
; j
++) {
1853 if (depend
->id
== grp
[j
].id
)
1858 grp
[i
].id
= depend
->id
;
1859 /* We aren't using weights for anything right now */
1860 grp
[i
].weight
= depend
->nhg
->nexthop
->weight
;
1872 void zebra_nhg_install_kernel(struct nhg_hash_entry
*nhe
)
1874 struct nhg_connected
*rb_node_dep
= NULL
;
1876 /* Resolve it first */
1877 nhe
= zebra_nhg_resolve(nhe
);
1879 /* Make sure all depends are installed/queued */
1880 frr_each(nhg_connected_tree
, &nhe
->nhg_depends
, rb_node_dep
) {
1881 zebra_nhg_install_kernel(rb_node_dep
->nhe
);
1884 if (!CHECK_FLAG(nhe
->flags
, NEXTHOP_GROUP_INSTALLED
)
1885 && !CHECK_FLAG(nhe
->flags
, NEXTHOP_GROUP_QUEUED
)) {
1886 /* Change its type to us since we are installing it */
1887 nhe
->type
= ZEBRA_ROUTE_NHG
;
1889 int ret
= dplane_nexthop_add(nhe
);
1892 case ZEBRA_DPLANE_REQUEST_QUEUED
:
1893 SET_FLAG(nhe
->flags
, NEXTHOP_GROUP_QUEUED
);
1895 case ZEBRA_DPLANE_REQUEST_FAILURE
:
1897 EC_ZEBRA_DP_INSTALL_FAIL
,
1898 "Failed to install Nexthop ID (%u) into the kernel",
1901 case ZEBRA_DPLANE_REQUEST_SUCCESS
:
1902 SET_FLAG(nhe
->flags
, NEXTHOP_GROUP_INSTALLED
);
1903 zebra_nhg_handle_install(nhe
);
1909 void zebra_nhg_uninstall_kernel(struct nhg_hash_entry
*nhe
)
1911 if (CHECK_FLAG(nhe
->flags
, NEXTHOP_GROUP_INSTALLED
)) {
1912 int ret
= dplane_nexthop_delete(nhe
);
1915 case ZEBRA_DPLANE_REQUEST_QUEUED
:
1916 SET_FLAG(nhe
->flags
, NEXTHOP_GROUP_QUEUED
);
1918 case ZEBRA_DPLANE_REQUEST_FAILURE
:
1920 EC_ZEBRA_DP_DELETE_FAIL
,
1921 "Failed to uninstall Nexthop ID (%u) from the kernel",
1924 case ZEBRA_DPLANE_REQUEST_SUCCESS
:
1925 UNSET_FLAG(nhe
->flags
, NEXTHOP_GROUP_INSTALLED
);
1930 zebra_nhg_handle_uninstall(nhe
);
1933 void zebra_nhg_dplane_result(struct zebra_dplane_ctx
*ctx
)
1935 enum dplane_op_e op
;
1936 enum zebra_dplane_result status
;
1938 struct nhg_hash_entry
*nhe
= NULL
;
1940 op
= dplane_ctx_get_op(ctx
);
1941 status
= dplane_ctx_get_status(ctx
);
1943 id
= dplane_ctx_get_nhe_id(ctx
);
1945 if (IS_ZEBRA_DEBUG_DPLANE_DETAIL
)
1947 "Nexthop dplane ctx %p, op %s, nexthop ID (%u), result %s",
1948 ctx
, dplane_op2str(op
), id
, dplane_res2str(status
));
1951 case DPLANE_OP_NH_DELETE
:
1952 if (status
!= ZEBRA_DPLANE_REQUEST_SUCCESS
)
1954 EC_ZEBRA_DP_DELETE_FAIL
,
1955 "Failed to uninstall Nexthop ID (%u) from the kernel",
1957 /* We already free'd the data, nothing to do */
1959 case DPLANE_OP_NH_INSTALL
:
1960 case DPLANE_OP_NH_UPDATE
:
1961 nhe
= zebra_nhg_lookup_id(id
);
1966 "%s operation preformed on Nexthop ID (%u) in the kernel, that we no longer have in our table",
1967 dplane_op2str(op
), id
);
1971 UNSET_FLAG(nhe
->flags
, NEXTHOP_GROUP_QUEUED
);
1972 if (status
== ZEBRA_DPLANE_REQUEST_SUCCESS
) {
1973 SET_FLAG(nhe
->flags
, NEXTHOP_GROUP_VALID
);
1974 SET_FLAG(nhe
->flags
, NEXTHOP_GROUP_INSTALLED
);
1975 zebra_nhg_handle_install(nhe
);
1978 EC_ZEBRA_DP_INSTALL_FAIL
,
1979 "Failed to install Nexthop ID (%u) into the kernel",
1982 case DPLANE_OP_ROUTE_INSTALL
:
1983 case DPLANE_OP_ROUTE_UPDATE
:
1984 case DPLANE_OP_ROUTE_DELETE
:
1985 case DPLANE_OP_ROUTE_NOTIFY
:
1986 case DPLANE_OP_LSP_INSTALL
:
1987 case DPLANE_OP_LSP_UPDATE
:
1988 case DPLANE_OP_LSP_DELETE
:
1989 case DPLANE_OP_LSP_NOTIFY
:
1990 case DPLANE_OP_PW_INSTALL
:
1991 case DPLANE_OP_PW_UNINSTALL
:
1992 case DPLANE_OP_SYS_ROUTE_ADD
:
1993 case DPLANE_OP_SYS_ROUTE_DELETE
:
1994 case DPLANE_OP_ADDR_INSTALL
:
1995 case DPLANE_OP_ADDR_UNINSTALL
:
1996 case DPLANE_OP_MAC_INSTALL
:
1997 case DPLANE_OP_MAC_DELETE
:
1998 case DPLANE_OP_NEIGH_INSTALL
:
1999 case DPLANE_OP_NEIGH_UPDATE
:
2000 case DPLANE_OP_NEIGH_DELETE
:
2001 case DPLANE_OP_VTEP_ADD
:
2002 case DPLANE_OP_VTEP_DELETE
:
2003 case DPLANE_OP_NONE
:
2007 dplane_ctx_fini(&ctx
);
2010 static void zebra_nhg_sweep_entry(struct hash_bucket
*bucket
, void *arg
)
2012 struct nhg_hash_entry
*nhe
= NULL
;
2014 nhe
= (struct nhg_hash_entry
*)bucket
->data
;
2016 /* If its being ref'd, just let it be uninstalled via a route removal */
2017 if (ZEBRA_NHG_CREATED(nhe
) && nhe
->refcnt
<= 0)
2018 zebra_nhg_uninstall_kernel(nhe
);
2021 void zebra_nhg_sweep_table(struct hash
*hash
)
2023 hash_iterate(hash
, zebra_nhg_sweep_entry
, NULL
);