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 void nhg_connected_tree_del_nhe(struct nhg_connected_tree_head
*head
,
103 struct nhg_hash_entry
*depend
)
105 struct nhg_connected lookup
= {};
106 struct nhg_connected
*remove
= NULL
;
110 /* Lookup to find the element, then remove it */
111 remove
= nhg_connected_tree_find(head
, &lookup
);
112 remove
= nhg_connected_tree_del(head
, remove
);
115 nhg_connected_free(remove
);
118 void nhg_connected_tree_add_nhe(struct nhg_connected_tree_head
*head
,
119 struct nhg_hash_entry
*depend
)
121 struct nhg_connected
*new = NULL
;
123 new = nhg_connected_new(depend
);
126 nhg_connected_tree_add(head
, new);
130 nhg_connected_tree_decrement_ref(struct nhg_connected_tree_head
*head
)
132 struct nhg_connected
*rb_node_dep
= NULL
;
134 frr_each_safe(nhg_connected_tree
, head
, rb_node_dep
) {
135 zebra_nhg_decrement_ref(rb_node_dep
->nhe
);
140 nhg_connected_tree_increment_ref(struct nhg_connected_tree_head
*head
)
142 struct nhg_connected
*rb_node_dep
= NULL
;
144 frr_each(nhg_connected_tree
, head
, rb_node_dep
) {
145 zebra_nhg_increment_ref(rb_node_dep
->nhe
);
149 struct nhg_hash_entry
*zebra_nhg_resolve(struct nhg_hash_entry
*nhe
)
151 if (CHECK_FLAG(nhe
->flags
, NEXTHOP_GROUP_RECURSIVE
)
152 && !zebra_nhg_depends_is_empty(nhe
)) {
153 nhe
= nhg_connected_tree_root(&nhe
->nhg_depends
)->nhe
;
154 return zebra_nhg_resolve(nhe
);
160 unsigned int zebra_nhg_depends_count(const struct nhg_hash_entry
*nhe
)
162 return nhg_connected_tree_count(&nhe
->nhg_depends
);
165 bool zebra_nhg_depends_is_empty(const struct nhg_hash_entry
*nhe
)
167 return nhg_connected_tree_is_empty(&nhe
->nhg_depends
);
170 static void zebra_nhg_depends_del(struct nhg_hash_entry
*from
,
171 struct nhg_hash_entry
*depend
)
173 nhg_connected_tree_del_nhe(&from
->nhg_depends
, depend
);
176 static void zebra_nhg_depends_init(struct nhg_hash_entry
*nhe
)
178 nhg_connected_tree_init(&nhe
->nhg_depends
);
181 unsigned int zebra_nhg_dependents_count(const struct nhg_hash_entry
*nhe
)
183 return nhg_connected_tree_count(&nhe
->nhg_dependents
);
187 bool zebra_nhg_dependents_is_empty(const struct nhg_hash_entry
*nhe
)
189 return nhg_connected_tree_is_empty(&nhe
->nhg_dependents
);
192 static void zebra_nhg_dependents_del(struct nhg_hash_entry
*from
,
193 struct nhg_hash_entry
*dependent
)
195 nhg_connected_tree_del_nhe(&from
->nhg_dependents
, dependent
);
198 static void zebra_nhg_dependents_add(struct nhg_hash_entry
*to
,
199 struct nhg_hash_entry
*dependent
)
201 nhg_connected_tree_add_nhe(&to
->nhg_dependents
, dependent
);
204 static void zebra_nhg_dependents_init(struct nhg_hash_entry
*nhe
)
206 nhg_connected_tree_init(&nhe
->nhg_dependents
);
209 /* Release this nhe from anything depending on it */
210 static void zebra_nhg_dependents_release(struct nhg_hash_entry
*nhe
)
212 struct nhg_connected
*rb_node_dep
= NULL
;
214 frr_each_safe(nhg_connected_tree
, &nhe
->nhg_dependents
, rb_node_dep
) {
215 zebra_nhg_depends_del(rb_node_dep
->nhe
, nhe
);
216 /* recheck validity of the dependent */
217 zebra_nhg_check_valid(rb_node_dep
->nhe
);
221 /* Release this nhe from anything that it depends on */
222 static void zebra_nhg_depends_release(struct nhg_hash_entry
*nhe
)
224 if (!zebra_nhg_depends_is_empty(nhe
)) {
225 struct nhg_connected
*rb_node_dep
= NULL
;
227 frr_each_safe(nhg_connected_tree
, &nhe
->nhg_depends
,
229 zebra_nhg_dependents_del(rb_node_dep
->nhe
, nhe
);
235 struct nhg_hash_entry
*zebra_nhg_lookup_id(uint32_t id
)
237 struct nhg_hash_entry lookup
= {};
240 return hash_lookup(zrouter
.nhgs_id
, &lookup
);
243 static int zebra_nhg_insert_id(struct nhg_hash_entry
*nhe
)
245 if (hash_lookup(zrouter
.nhgs_id
, nhe
)) {
247 EC_ZEBRA_NHG_TABLE_INSERT_FAILED
,
248 "Failed inserting NHG id=%u into the ID hash table, entry already exists",
253 hash_get(zrouter
.nhgs_id
, nhe
, hash_alloc_intern
);
258 static void zebra_nhg_set_if(struct nhg_hash_entry
*nhe
, struct interface
*ifp
)
261 if_nhg_dependents_add(ifp
, nhe
);
265 zebra_nhg_connect_depends(struct nhg_hash_entry
*nhe
,
266 struct nhg_connected_tree_head nhg_depends
)
268 struct nhg_connected
*rb_node_dep
= NULL
;
270 /* This has been allocated higher above in the stack. Could probably
271 * re-allocate and free the old stuff but just using the same memory
272 * for now. Otherwise, their might be a time trade-off for repeated
273 * alloc/frees as startup.
275 nhe
->nhg_depends
= nhg_depends
;
277 /* Attach backpointer to anything that it depends on */
278 zebra_nhg_dependents_init(nhe
);
279 if (!zebra_nhg_depends_is_empty(nhe
)) {
280 frr_each(nhg_connected_tree
, &nhe
->nhg_depends
, rb_node_dep
) {
281 zebra_nhg_dependents_add(rb_node_dep
->nhe
, nhe
);
285 /* Add the ifp now if its not a group or recursive and has ifindex */
286 if (zebra_nhg_depends_is_empty(nhe
) && nhe
->nhg
->nexthop
287 && nhe
->nhg
->nexthop
->ifindex
) {
288 struct interface
*ifp
= NULL
;
290 ifp
= if_lookup_by_index(nhe
->nhg
->nexthop
->ifindex
,
293 zebra_nhg_set_if(nhe
, ifp
);
296 EC_ZEBRA_IF_LOOKUP_FAILED
,
297 "Zebra failed to lookup an interface with ifindex=%d in vrf=%u for NHE id=%u",
298 nhe
->nhg
->nexthop
->ifindex
, nhe
->vrf_id
,
303 struct nhg_hash_entry
*zebra_nhg_alloc(void)
305 struct nhg_hash_entry
*nhe
;
307 nhe
= XCALLOC(MTYPE_NHG
, sizeof(struct nhg_hash_entry
));
312 static struct nhg_hash_entry
*zebra_nhg_copy(const struct nhg_hash_entry
*copy
,
315 struct nhg_hash_entry
*nhe
;
317 nhe
= zebra_nhg_alloc();
321 nhe
->nhg
= nexthop_group_new();
322 nexthop_group_copy(nhe
->nhg
, copy
->nhg
);
324 nhe
->vrf_id
= copy
->vrf_id
;
325 nhe
->afi
= copy
->afi
;
326 nhe
->type
= copy
->type
? copy
->type
: ZEBRA_ROUTE_NHG
;
328 nhe
->dplane_ref
= zebra_router_get_next_sequence();
333 /* Allocation via hash handler */
334 static void *zebra_nhg_hash_alloc(void *arg
)
336 struct nhg_hash_entry
*nhe
= NULL
;
337 struct nhg_hash_entry
*copy
= arg
;
339 nhe
= zebra_nhg_copy(copy
, copy
->id
);
341 /* Mark duplicate nexthops in a group at creation time. */
342 nexthop_group_mark_duplicates(nhe
->nhg
);
344 zebra_nhg_connect_depends(nhe
, copy
->nhg_depends
);
345 zebra_nhg_insert_id(nhe
);
350 uint32_t zebra_nhg_hash_key(const void *arg
)
352 const struct nhg_hash_entry
*nhe
= arg
;
354 uint32_t key
= 0x5a351234;
356 key
= jhash_3words(nhe
->vrf_id
, nhe
->afi
, nexthop_group_hash(nhe
->nhg
),
362 uint32_t zebra_nhg_id_key(const void *arg
)
364 const struct nhg_hash_entry
*nhe
= arg
;
369 bool zebra_nhg_hash_equal(const void *arg1
, const void *arg2
)
371 const struct nhg_hash_entry
*nhe1
= arg1
;
372 const struct nhg_hash_entry
*nhe2
= arg2
;
373 struct nexthop
*nexthop1
;
374 struct nexthop
*nexthop2
;
376 /* No matter what if they equal IDs, assume equal */
377 if (nhe1
->id
&& nhe2
->id
&& (nhe1
->id
== nhe2
->id
))
380 if (nhe1
->vrf_id
!= nhe2
->vrf_id
)
383 if (nhe1
->afi
!= nhe2
->afi
)
386 /* Nexthops should be sorted */
387 for (nexthop1
= nhe1
->nhg
->nexthop
, nexthop2
= nhe2
->nhg
->nexthop
;
388 nexthop1
|| nexthop2
;
389 nexthop1
= nexthop1
->next
, nexthop2
= nexthop2
->next
) {
390 if (nexthop1
&& !nexthop2
)
393 if (!nexthop1
&& nexthop2
)
397 * We have to check the active flag of each individual one,
398 * not just the overall active_num. This solves the special case
399 * issue of a route with a nexthop group with one nexthop
400 * resolving to itself and thus marking it inactive. If we
401 * have two different routes each wanting to mark a different
402 * nexthop inactive, they need to hash to two different groups.
404 * If we just hashed on num_active, they would hash the same
405 * which is incorrect.
409 * -> 1.1.1.1 dummy1 (inactive)
414 * -> 1.1.2.1 dummy2 (inactive)
416 * Without checking each individual one, they would hash to
417 * the same group and both have 1.1.1.1 dummy1 marked inactive.
420 if (CHECK_FLAG(nexthop1
->flags
, NEXTHOP_FLAG_ACTIVE
)
421 != CHECK_FLAG(nexthop2
->flags
, NEXTHOP_FLAG_ACTIVE
))
424 if (!nexthop_same(nexthop1
, nexthop2
))
431 bool zebra_nhg_hash_id_equal(const void *arg1
, const void *arg2
)
433 const struct nhg_hash_entry
*nhe1
= arg1
;
434 const struct nhg_hash_entry
*nhe2
= arg2
;
436 return nhe1
->id
== nhe2
->id
;
439 static int zebra_nhg_process_grp(struct nexthop_group
*nhg
,
440 struct nhg_connected_tree_head
*depends
,
441 struct nh_grp
*grp
, uint8_t count
)
443 nhg_connected_tree_init(depends
);
445 for (int i
= 0; i
< count
; i
++) {
446 struct nhg_hash_entry
*depend
= NULL
;
447 /* We do not care about nexthop_grp.weight at
448 * this time. But we should figure out
449 * how to adapt this to our code in
452 depend
= depends_find_id_add(depends
, grp
[i
].id
);
457 "Received Nexthop Group from the kernel with a dependent Nexthop ID (%u) which we do not have in our table",
463 * If this is a nexthop with its own group
464 * dependencies, add them as well. Not sure its
465 * even possible to have a group within a group
469 copy_nexthops(&nhg
->nexthop
, depend
->nhg
->nexthop
, NULL
);
475 static void handle_recursive_depend(struct nhg_connected_tree_head
*nhg_depends
,
476 struct nexthop
*nh
, afi_t afi
)
478 struct nhg_hash_entry
*depend
= NULL
;
479 struct nexthop_group resolved_ng
= {};
481 nexthop_group_add_sorted(&resolved_ng
, nh
);
483 depend
= zebra_nhg_rib_find(0, &resolved_ng
, afi
);
484 depends_add(nhg_depends
, depend
);
487 static bool zebra_nhg_find(struct nhg_hash_entry
**nhe
, uint32_t id
,
488 struct nexthop_group
*nhg
,
489 struct nhg_connected_tree_head
*nhg_depends
,
490 vrf_id_t vrf_id
, afi_t afi
, int type
)
492 struct nhg_hash_entry lookup
= {};
494 uint32_t old_id_counter
= id_counter
;
496 bool created
= false;
497 bool recursive
= false;
500 * If it has an id at this point, we must have gotten it from the kernel
502 lookup
.id
= id
? id
: ++id_counter
;
504 lookup
.type
= type
? type
: ZEBRA_ROUTE_NHG
;
507 if (lookup
.nhg
->nexthop
->next
) {
508 /* Groups can have all vrfs and AF's in them */
509 lookup
.afi
= AFI_UNSPEC
;
512 switch (lookup
.nhg
->nexthop
->type
) {
513 case (NEXTHOP_TYPE_IFINDEX
):
514 case (NEXTHOP_TYPE_BLACKHOLE
):
516 * This switch case handles setting the afi different
517 * for ipv4/v6 routes. Ifindex/blackhole nexthop
518 * objects cannot be ambiguous, they must be Address
519 * Family specific. If we get here, we will either use
520 * the AF of the route, or the one we got passed from
521 * here from the kernel.
525 case (NEXTHOP_TYPE_IPV4_IFINDEX
):
526 case (NEXTHOP_TYPE_IPV4
):
529 case (NEXTHOP_TYPE_IPV6_IFINDEX
):
530 case (NEXTHOP_TYPE_IPV6
):
531 lookup
.afi
= AFI_IP6
;
535 lookup
.vrf_id
= vrf_id
;
539 (*nhe
) = zebra_nhg_lookup_id(id
);
541 (*nhe
) = hash_lookup(zrouter
.nhgs
, &lookup
);
543 /* If it found an nhe in our tables, this new ID is unused */
545 id_counter
= old_id_counter
;
548 /* Only hash/lookup the depends if the first lookup
549 * fails to find something. This should hopefully save a
550 * lot of cycles for larger ecmp sizes.
553 /* If you don't want to hash on each nexthop in the
554 * nexthop group struct you can pass the depends
555 * directly. Kernel-side we do this since it just looks
558 lookup
.nhg_depends
= *nhg_depends
;
560 if (nhg
->nexthop
->next
) {
561 zebra_nhg_depends_init(&lookup
);
563 /* If its a group, create a dependency tree */
564 struct nexthop
*nh
= NULL
;
566 for (nh
= nhg
->nexthop
; nh
; nh
= nh
->next
)
567 depends_find_add(&lookup
.nhg_depends
,
569 } else if (CHECK_FLAG(nhg
->nexthop
->flags
,
570 NEXTHOP_FLAG_RECURSIVE
)) {
571 zebra_nhg_depends_init(&lookup
);
572 handle_recursive_depend(&lookup
.nhg_depends
,
573 nhg
->nexthop
->resolved
,
579 (*nhe
) = hash_get(zrouter
.nhgs
, &lookup
, zebra_nhg_hash_alloc
);
583 SET_FLAG((*nhe
)->flags
, NEXTHOP_GROUP_RECURSIVE
);
588 /* Find/create a single nexthop */
589 static struct nhg_hash_entry
*
590 zebra_nhg_find_nexthop(uint32_t id
, struct nexthop
*nh
, afi_t afi
, int type
)
592 struct nhg_hash_entry
*nhe
= NULL
;
593 struct nexthop_group nhg
= {};
595 nexthop_group_add_sorted(&nhg
, nh
);
597 zebra_nhg_find(&nhe
, id
, &nhg
, NULL
, nh
->vrf_id
, afi
, type
);
602 static uint32_t nhg_ctx_get_id(const struct nhg_ctx
*ctx
)
607 static void nhg_ctx_set_status(struct nhg_ctx
*ctx
, enum nhg_ctx_status status
)
609 ctx
->status
= status
;
612 static enum nhg_ctx_status
nhg_ctx_get_status(const struct nhg_ctx
*ctx
)
617 static void nhg_ctx_set_op(struct nhg_ctx
*ctx
, enum nhg_ctx_op_e op
)
622 static enum nhg_ctx_op_e
nhg_ctx_get_op(const struct nhg_ctx
*ctx
)
627 static vrf_id_t
nhg_ctx_get_vrf_id(const struct nhg_ctx
*ctx
)
632 static int nhg_ctx_get_type(const struct nhg_ctx
*ctx
)
637 static int nhg_ctx_get_afi(const struct nhg_ctx
*ctx
)
642 static struct nexthop
*nhg_ctx_get_nh(struct nhg_ctx
*ctx
)
647 static uint8_t nhg_ctx_get_count(const struct nhg_ctx
*ctx
)
652 static struct nh_grp
*nhg_ctx_get_grp(struct nhg_ctx
*ctx
)
657 static struct nhg_ctx
*nhg_ctx_new()
659 struct nhg_ctx
*new = NULL
;
661 new = XCALLOC(MTYPE_NHG_CTX
, sizeof(struct nhg_ctx
));
666 static void nhg_ctx_free(struct nhg_ctx
**ctx
)
673 assert((*ctx
) != NULL
);
675 if (nhg_ctx_get_count(*ctx
))
678 nh
= nhg_ctx_get_nh(*ctx
);
680 nexthop_del_labels(nh
);
683 XFREE(MTYPE_NHG_CTX
, *ctx
);
687 static struct nhg_ctx
*nhg_ctx_init(uint32_t id
, struct nexthop
*nh
,
688 struct nh_grp
*grp
, vrf_id_t vrf_id
,
689 afi_t afi
, int type
, uint8_t count
)
691 struct nhg_ctx
*ctx
= NULL
;
696 ctx
->vrf_id
= vrf_id
;
702 /* Copy over the array */
703 memcpy(&ctx
->u
.grp
, grp
, count
* sizeof(struct nh_grp
));
710 static bool zebra_nhg_contains_unhashable(struct nhg_hash_entry
*nhe
)
712 struct nhg_connected
*rb_node_dep
= NULL
;
714 frr_each(nhg_connected_tree
, &nhe
->nhg_depends
, rb_node_dep
) {
715 if (CHECK_FLAG(rb_node_dep
->nhe
->flags
,
716 NEXTHOP_GROUP_UNHASHABLE
))
723 static void zebra_nhg_set_unhashable(struct nhg_hash_entry
*nhe
)
725 SET_FLAG(nhe
->flags
, NEXTHOP_GROUP_UNHASHABLE
);
726 SET_FLAG(nhe
->flags
, NEXTHOP_GROUP_INSTALLED
);
729 EC_ZEBRA_DUPLICATE_NHG_MESSAGE
,
730 "Nexthop Group with ID (%d) is a duplicate, therefore unhashable, ignoring",
734 static void zebra_nhg_set_valid(struct nhg_hash_entry
*nhe
)
736 struct nhg_connected
*rb_node_dep
;
738 SET_FLAG(nhe
->flags
, NEXTHOP_GROUP_VALID
);
740 frr_each(nhg_connected_tree
, &nhe
->nhg_dependents
, rb_node_dep
)
741 zebra_nhg_set_valid(rb_node_dep
->nhe
);
744 static void zebra_nhg_set_invalid(struct nhg_hash_entry
*nhe
)
746 struct nhg_connected
*rb_node_dep
;
748 UNSET_FLAG(nhe
->flags
, NEXTHOP_GROUP_VALID
);
750 /* Update validity of nexthops depending on it */
751 frr_each(nhg_connected_tree
, &nhe
->nhg_dependents
, rb_node_dep
)
752 zebra_nhg_check_valid(rb_node_dep
->nhe
);
755 void zebra_nhg_check_valid(struct nhg_hash_entry
*nhe
)
757 struct nhg_connected
*rb_node_dep
= NULL
;
760 /* If anthing else in the group is valid, the group is valid */
761 frr_each(nhg_connected_tree
, &nhe
->nhg_depends
, rb_node_dep
) {
762 if (CHECK_FLAG(rb_node_dep
->nhe
->flags
, NEXTHOP_GROUP_VALID
)) {
770 zebra_nhg_set_valid(nhe
);
772 zebra_nhg_set_invalid(nhe
);
776 static void zebra_nhg_release(struct nhg_hash_entry
*nhe
)
778 /* Remove it from any lists it may be on */
779 zebra_nhg_depends_release(nhe
);
780 zebra_nhg_dependents_release(nhe
);
782 if_nhg_dependents_del(nhe
->ifp
, nhe
);
785 * If its unhashable, we didn't store it here and have to be
786 * sure we don't clear one thats actually being used.
788 if (!CHECK_FLAG(nhe
->flags
, NEXTHOP_GROUP_UNHASHABLE
))
789 hash_release(zrouter
.nhgs
, nhe
);
791 hash_release(zrouter
.nhgs_id
, nhe
);
794 static void zebra_nhg_handle_uninstall(struct nhg_hash_entry
*nhe
)
796 zebra_nhg_release(nhe
);
800 static void zebra_nhg_handle_install(struct nhg_hash_entry
*nhe
)
802 /* Update validity of groups depending on it */
803 struct nhg_connected
*rb_node_dep
;
805 frr_each_safe(nhg_connected_tree
, &nhe
->nhg_dependents
, rb_node_dep
)
806 zebra_nhg_set_valid(rb_node_dep
->nhe
);
810 * The kernel/other program has changed the state of a nexthop object we are
813 static void zebra_nhg_handle_kernel_state_change(struct nhg_hash_entry
*nhe
,
819 "Kernel %s a nexthop group with ID (%u) that we are still using for a route, sending it back down",
820 (is_delete
? "deleted" : "updated"), nhe
->id
);
822 UNSET_FLAG(nhe
->flags
, NEXTHOP_GROUP_INSTALLED
);
823 zebra_nhg_install_kernel(nhe
);
825 zebra_nhg_handle_uninstall(nhe
);
828 static int nhg_ctx_process_new(struct nhg_ctx
*ctx
)
830 struct nexthop_group
*nhg
= NULL
;
831 struct nhg_connected_tree_head nhg_depends
= {};
832 struct nhg_hash_entry
*lookup
= NULL
;
833 struct nhg_hash_entry
*nhe
= NULL
;
835 uint32_t id
= nhg_ctx_get_id(ctx
);
836 uint8_t count
= nhg_ctx_get_count(ctx
);
837 vrf_id_t vrf_id
= nhg_ctx_get_vrf_id(ctx
);
838 int type
= nhg_ctx_get_type(ctx
);
839 afi_t afi
= nhg_ctx_get_afi(ctx
);
841 lookup
= zebra_nhg_lookup_id(id
);
844 /* This is already present in our table, hence an update
845 * that we did not initate.
847 zebra_nhg_handle_kernel_state_change(lookup
, false);
851 if (nhg_ctx_get_count(ctx
)) {
852 nhg
= nexthop_group_new();
853 if (zebra_nhg_process_grp(nhg
, &nhg_depends
,
854 nhg_ctx_get_grp(ctx
), count
)) {
855 depends_decrement_free(&nhg_depends
);
856 nexthop_group_delete(&nhg
);
860 if (!zebra_nhg_find(&nhe
, id
, nhg
, &nhg_depends
, vrf_id
, type
,
862 depends_decrement_free(&nhg_depends
);
864 /* These got copied over in zebra_nhg_alloc() */
865 nexthop_group_delete(&nhg
);
867 nhe
= zebra_nhg_find_nexthop(id
, nhg_ctx_get_nh(ctx
), afi
,
872 struct nhg_hash_entry
*kernel_nhe
= NULL
;
874 /* Duplicate but with different ID from
878 /* The kernel allows duplicate nexthops
879 * as long as they have different IDs.
880 * We are ignoring those to prevent
881 * syncing problems with the kernel
884 * We maintain them *ONLY* in the ID hash table to
885 * track them and set the flag to indicated
886 * their attributes are unhashable.
889 kernel_nhe
= zebra_nhg_copy(nhe
, id
);
890 zebra_nhg_insert_id(kernel_nhe
);
891 zebra_nhg_set_unhashable(kernel_nhe
);
892 } else if (zebra_nhg_contains_unhashable(nhe
)) {
893 /* The group we got contains an unhashable/duplicated
894 * depend, so lets mark this group as unhashable as well
895 * and release it from the non-ID hash.
897 hash_release(zrouter
.nhgs
, nhe
);
898 zebra_nhg_set_unhashable(nhe
);
900 /* It actually created a new nhe */
901 SET_FLAG(nhe
->flags
, NEXTHOP_GROUP_VALID
);
902 SET_FLAG(nhe
->flags
, NEXTHOP_GROUP_INSTALLED
);
906 EC_ZEBRA_TABLE_LOOKUP_FAILED
,
907 "Zebra failed to find or create a nexthop hash entry for ID (%u)",
915 static int nhg_ctx_process_del(struct nhg_ctx
*ctx
)
917 struct nhg_hash_entry
*nhe
= NULL
;
918 uint32_t id
= nhg_ctx_get_id(ctx
);
920 nhe
= zebra_nhg_lookup_id(id
);
924 EC_ZEBRA_BAD_NHG_MESSAGE
,
925 "Kernel delete message received for nexthop group ID (%u) that we do not have in our ID table",
930 zebra_nhg_handle_kernel_state_change(nhe
, true);
935 static void nhg_ctx_fini(struct nhg_ctx
**ctx
)
938 * Just freeing for now, maybe do something more in the future
945 static int queue_add(struct nhg_ctx
*ctx
)
947 /* If its queued or already processed do nothing */
948 if (nhg_ctx_get_status(ctx
) == NHG_CTX_QUEUED
)
951 if (rib_queue_nhg_add(ctx
)) {
952 nhg_ctx_set_status(ctx
, NHG_CTX_FAILURE
);
956 nhg_ctx_set_status(ctx
, NHG_CTX_QUEUED
);
961 int nhg_ctx_process(struct nhg_ctx
*ctx
)
965 switch (nhg_ctx_get_op(ctx
)) {
967 ret
= nhg_ctx_process_new(ctx
);
968 if (nhg_ctx_get_count(ctx
) && ret
== -ENOENT
969 && nhg_ctx_get_status(ctx
) != NHG_CTX_REQUEUED
) {
971 * We have entered a situation where we are
972 * processing a group from the kernel
973 * that has a contained nexthop which
974 * we have not yet processed.
976 * Re-enqueue this ctx to be handled exactly one
977 * more time (indicated by the flag).
979 * By the time we get back to it, we
980 * should have processed its depends.
982 nhg_ctx_set_status(ctx
, NHG_CTX_NONE
);
983 if (queue_add(ctx
) == 0) {
984 nhg_ctx_set_status(ctx
, NHG_CTX_REQUEUED
);
990 ret
= nhg_ctx_process_del(ctx
);
991 case NHG_CTX_OP_NONE
:
995 nhg_ctx_set_status(ctx
, (ret
? NHG_CTX_FAILURE
: NHG_CTX_SUCCESS
));
1002 /* Kernel-side, you either get a single new nexthop or a array of ID's */
1003 int zebra_nhg_kernel_find(uint32_t id
, struct nexthop
*nh
, struct nh_grp
*grp
,
1004 uint8_t count
, vrf_id_t vrf_id
, afi_t afi
, int type
,
1007 struct nhg_ctx
*ctx
= NULL
;
1009 if (id
> id_counter
)
1010 /* Increase our counter so we don't try to create
1011 * an ID that already exists
1015 ctx
= nhg_ctx_init(id
, nh
, grp
, vrf_id
, afi
, type
, count
);
1016 nhg_ctx_set_op(ctx
, NHG_CTX_OP_NEW
);
1018 /* Under statup conditions, we need to handle them immediately
1019 * like we do for routes. Otherwise, we are going to get a route
1020 * with a nhe_id that we have not handled.
1023 return nhg_ctx_process(ctx
);
1025 if (queue_add(ctx
)) {
1033 /* Kernel-side, received delete message */
1034 int zebra_nhg_kernel_del(uint32_t id
)
1036 struct nhg_ctx
*ctx
= NULL
;
1038 ctx
= nhg_ctx_init(id
, NULL
, NULL
, 0, 0, 0, 0);
1040 nhg_ctx_set_op(ctx
, NHG_CTX_OP_DEL
);
1042 if (queue_add(ctx
)) {
1050 /* Some dependency helper functions */
1051 static struct nhg_hash_entry
*depends_find(const struct nexthop
*nh
, afi_t afi
)
1053 struct nexthop lookup
;
1054 struct nhg_hash_entry
*nhe
= NULL
;
1059 /* Capture a snapshot of this single nh; it might be part of a list,
1060 * so we need to make a standalone copy.
1062 memset(&lookup
, 0, sizeof(lookup
));
1063 nexthop_copy(&lookup
, nh
, NULL
);
1065 nhe
= zebra_nhg_find_nexthop(0, &lookup
, afi
, 0);
1067 /* The copy may have allocated labels; free them if necessary. */
1068 nexthop_del_labels(&lookup
);
1074 static void depends_add(struct nhg_connected_tree_head
*head
,
1075 struct nhg_hash_entry
*depend
)
1077 nhg_connected_tree_add_nhe(head
, depend
);
1078 zebra_nhg_increment_ref(depend
);
1081 static struct nhg_hash_entry
*
1082 depends_find_add(struct nhg_connected_tree_head
*head
, struct nexthop
*nh
,
1085 struct nhg_hash_entry
*depend
= NULL
;
1087 depend
= depends_find(nh
, afi
);
1090 depends_add(head
, depend
);
1095 static struct nhg_hash_entry
*
1096 depends_find_id_add(struct nhg_connected_tree_head
*head
, uint32_t id
)
1098 struct nhg_hash_entry
*depend
= NULL
;
1100 depend
= zebra_nhg_lookup_id(id
);
1103 depends_add(head
, depend
);
1108 static void depends_decrement_free(struct nhg_connected_tree_head
*head
)
1110 nhg_connected_tree_decrement_ref(head
);
1111 nhg_connected_tree_free(head
);
1114 /* Rib-side, you get a nexthop group struct */
1115 struct nhg_hash_entry
*
1116 zebra_nhg_rib_find(uint32_t id
, struct nexthop_group
*nhg
, afi_t rt_afi
)
1118 struct nhg_hash_entry
*nhe
= NULL
;
1120 if (!(nhg
&& nhg
->nexthop
)) {
1121 flog_err(EC_ZEBRA_TABLE_LOOKUP_FAILED
,
1122 "No nexthop passed to %s", __func__
);
1126 zebra_nhg_find(&nhe
, id
, nhg
, NULL
, nhg
->nexthop
->vrf_id
, rt_afi
, 0);
1131 static void zebra_nhg_free_members(struct nhg_hash_entry
*nhe
)
1133 nexthop_group_delete(&nhe
->nhg
);
1134 /* Decrement to remove connection ref */
1135 nhg_connected_tree_decrement_ref(&nhe
->nhg_depends
);
1136 nhg_connected_tree_free(&nhe
->nhg_depends
);
1137 nhg_connected_tree_free(&nhe
->nhg_dependents
);
1140 void zebra_nhg_free(struct nhg_hash_entry
*nhe
)
1143 zlog_debug("nhe_id=%u hash refcnt=%d", nhe
->id
, nhe
->refcnt
);
1145 zebra_nhg_free_members(nhe
);
1147 XFREE(MTYPE_NHG
, nhe
);
1150 void zebra_nhg_hash_free(void *p
)
1152 zebra_nhg_free((struct nhg_hash_entry
*)p
);
1155 void zebra_nhg_decrement_ref(struct nhg_hash_entry
*nhe
)
1159 if (!zebra_nhg_depends_is_empty(nhe
))
1160 nhg_connected_tree_decrement_ref(&nhe
->nhg_depends
);
1162 if (ZEBRA_NHG_CREATED(nhe
) && nhe
->refcnt
<= 0)
1163 zebra_nhg_uninstall_kernel(nhe
);
1166 void zebra_nhg_increment_ref(struct nhg_hash_entry
*nhe
)
1170 if (!zebra_nhg_depends_is_empty(nhe
))
1171 nhg_connected_tree_increment_ref(&nhe
->nhg_depends
);
1174 static void nexthop_set_resolved(afi_t afi
, const struct nexthop
*newhop
,
1175 struct nexthop
*nexthop
)
1177 struct nexthop
*resolved_hop
;
1178 uint8_t num_labels
= 0;
1179 mpls_label_t labels
[MPLS_MAX_LABELS
];
1180 enum lsp_types_t label_type
= ZEBRA_LSP_NONE
;
1183 resolved_hop
= nexthop_new();
1184 SET_FLAG(resolved_hop
->flags
, NEXTHOP_FLAG_ACTIVE
);
1186 resolved_hop
->vrf_id
= nexthop
->vrf_id
;
1187 switch (newhop
->type
) {
1188 case NEXTHOP_TYPE_IPV4
:
1189 case NEXTHOP_TYPE_IPV4_IFINDEX
:
1190 /* If the resolving route specifies a gateway, use it */
1191 resolved_hop
->type
= newhop
->type
;
1192 resolved_hop
->gate
.ipv4
= newhop
->gate
.ipv4
;
1194 if (newhop
->ifindex
) {
1195 resolved_hop
->type
= NEXTHOP_TYPE_IPV4_IFINDEX
;
1196 resolved_hop
->ifindex
= newhop
->ifindex
;
1199 case NEXTHOP_TYPE_IPV6
:
1200 case NEXTHOP_TYPE_IPV6_IFINDEX
:
1201 resolved_hop
->type
= newhop
->type
;
1202 resolved_hop
->gate
.ipv6
= newhop
->gate
.ipv6
;
1204 if (newhop
->ifindex
) {
1205 resolved_hop
->type
= NEXTHOP_TYPE_IPV6_IFINDEX
;
1206 resolved_hop
->ifindex
= newhop
->ifindex
;
1209 case NEXTHOP_TYPE_IFINDEX
:
1210 /* If the resolving route is an interface route,
1211 * it means the gateway we are looking up is connected
1212 * to that interface. (The actual network is _not_ onlink).
1213 * Therefore, the resolved route should have the original
1214 * gateway as nexthop as it is directly connected.
1216 * On Linux, we have to set the onlink netlink flag because
1217 * otherwise, the kernel won't accept the route.
1219 resolved_hop
->flags
|= NEXTHOP_FLAG_ONLINK
;
1220 if (afi
== AFI_IP
) {
1221 resolved_hop
->type
= NEXTHOP_TYPE_IPV4_IFINDEX
;
1222 resolved_hop
->gate
.ipv4
= nexthop
->gate
.ipv4
;
1223 } else if (afi
== AFI_IP6
) {
1224 resolved_hop
->type
= NEXTHOP_TYPE_IPV6_IFINDEX
;
1225 resolved_hop
->gate
.ipv6
= nexthop
->gate
.ipv6
;
1227 resolved_hop
->ifindex
= newhop
->ifindex
;
1229 case NEXTHOP_TYPE_BLACKHOLE
:
1230 resolved_hop
->type
= NEXTHOP_TYPE_BLACKHOLE
;
1231 resolved_hop
->bh_type
= newhop
->bh_type
;
1235 if (newhop
->flags
& NEXTHOP_FLAG_ONLINK
)
1236 resolved_hop
->flags
|= NEXTHOP_FLAG_ONLINK
;
1238 /* Copy labels of the resolved route and the parent resolving to it */
1239 if (newhop
->nh_label
) {
1240 for (i
= 0; i
< newhop
->nh_label
->num_labels
; i
++)
1241 labels
[num_labels
++] = newhop
->nh_label
->label
[i
];
1242 label_type
= newhop
->nh_label_type
;
1245 if (nexthop
->nh_label
) {
1246 for (i
= 0; i
< nexthop
->nh_label
->num_labels
; i
++)
1247 labels
[num_labels
++] = nexthop
->nh_label
->label
[i
];
1249 /* If the parent has labels, use its type */
1250 label_type
= nexthop
->nh_label_type
;
1254 nexthop_add_labels(resolved_hop
, label_type
, num_labels
,
1257 resolved_hop
->rparent
= nexthop
;
1258 _nexthop_add(&nexthop
->resolved
, resolved_hop
);
1261 /* Checks if nexthop we are trying to resolve to is valid */
1262 static bool nexthop_valid_resolve(const struct nexthop
*nexthop
,
1263 const struct nexthop
*resolved
)
1265 /* Can't resolve to a recursive nexthop */
1266 if (CHECK_FLAG(resolved
->flags
, NEXTHOP_FLAG_RECURSIVE
))
1269 switch (nexthop
->type
) {
1270 case NEXTHOP_TYPE_IPV4_IFINDEX
:
1271 case NEXTHOP_TYPE_IPV6_IFINDEX
:
1272 /* If the nexthop we are resolving to does not match the
1273 * ifindex for the nexthop the route wanted, its not valid.
1275 if (nexthop
->ifindex
!= resolved
->ifindex
)
1278 case NEXTHOP_TYPE_IPV4
:
1279 case NEXTHOP_TYPE_IPV6
:
1280 case NEXTHOP_TYPE_IFINDEX
:
1281 case NEXTHOP_TYPE_BLACKHOLE
:
1289 * Given a nexthop we need to properly recursively resolve
1290 * the route. As such, do a table lookup to find and match
1291 * if at all possible. Set the nexthop->ifindex and resolved_id
1294 static int nexthop_active(afi_t afi
, struct route_entry
*re
,
1295 struct nexthop
*nexthop
, struct route_node
*top
)
1298 struct route_table
*table
;
1299 struct route_node
*rn
;
1300 struct route_entry
*match
= NULL
;
1302 struct nexthop
*newhop
;
1303 struct interface
*ifp
;
1305 struct zebra_vrf
*zvrf
;
1307 if ((nexthop
->type
== NEXTHOP_TYPE_IPV4
)
1308 || nexthop
->type
== NEXTHOP_TYPE_IPV6
)
1309 nexthop
->ifindex
= 0;
1312 UNSET_FLAG(nexthop
->flags
, NEXTHOP_FLAG_RECURSIVE
);
1313 nexthops_free(nexthop
->resolved
);
1314 nexthop
->resolved
= NULL
;
1315 re
->nexthop_mtu
= 0;
1318 * If the kernel has sent us a NEW route, then
1319 * by golly gee whiz it's a good route.
1321 * If its an already INSTALLED route we have already handled, then the
1322 * kernel route's nexthop might have became unreachable
1323 * and we have to handle that.
1325 if (!CHECK_FLAG(re
->status
, ROUTE_ENTRY_INSTALLED
)
1326 && (re
->type
== ZEBRA_ROUTE_KERNEL
1327 || re
->type
== ZEBRA_ROUTE_SYSTEM
))
1331 * Check to see if we should trust the passed in information
1332 * for UNNUMBERED interfaces as that we won't find the GW
1333 * address in the routing table.
1334 * This check should suffice to handle IPv4 or IPv6 routes
1335 * sourced from EVPN routes which are installed with the
1336 * next hop as the remote VTEP IP.
1338 if (CHECK_FLAG(nexthop
->flags
, NEXTHOP_FLAG_ONLINK
)) {
1339 ifp
= if_lookup_by_index(nexthop
->ifindex
, nexthop
->vrf_id
);
1341 if (IS_ZEBRA_DEBUG_RIB_DETAILED
)
1343 "\t%s: Onlink and interface: %u[%u] does not exist",
1344 __PRETTY_FUNCTION__
, nexthop
->ifindex
,
1348 if (connected_is_unnumbered(ifp
)) {
1349 if (if_is_operative(ifp
))
1352 if (IS_ZEBRA_DEBUG_RIB_DETAILED
)
1354 "\t%s: Onlink and interface %s is not operative",
1355 __PRETTY_FUNCTION__
, ifp
->name
);
1358 if (!if_is_operative(ifp
)) {
1359 if (IS_ZEBRA_DEBUG_RIB_DETAILED
)
1361 "\t%s: Interface %s is not unnumbered",
1362 __PRETTY_FUNCTION__
, ifp
->name
);
1367 if ((top
->p
.family
== AF_INET
&& top
->p
.prefixlen
== 32
1368 && nexthop
->gate
.ipv4
.s_addr
== top
->p
.u
.prefix4
.s_addr
)
1369 || (top
->p
.family
== AF_INET6
&& top
->p
.prefixlen
== 128
1370 && memcmp(&nexthop
->gate
.ipv6
, &top
->p
.u
.prefix6
, 16) == 0)) {
1371 if (IS_ZEBRA_DEBUG_RIB_DETAILED
)
1373 "\t:%s: Attempting to install a max prefixlength route through itself",
1374 __PRETTY_FUNCTION__
);
1378 /* Make lookup prefix. */
1379 memset(&p
, 0, sizeof(struct prefix
));
1383 p
.prefixlen
= IPV4_MAX_PREFIXLEN
;
1384 p
.u
.prefix4
= nexthop
->gate
.ipv4
;
1387 p
.family
= AF_INET6
;
1388 p
.prefixlen
= IPV6_MAX_PREFIXLEN
;
1389 p
.u
.prefix6
= nexthop
->gate
.ipv6
;
1392 assert(afi
!= AFI_IP
&& afi
!= AFI_IP6
);
1396 table
= zebra_vrf_table(afi
, SAFI_UNICAST
, nexthop
->vrf_id
);
1398 zvrf
= zebra_vrf_lookup_by_id(nexthop
->vrf_id
);
1399 if (!table
|| !zvrf
) {
1400 if (IS_ZEBRA_DEBUG_RIB_DETAILED
)
1401 zlog_debug("\t%s: Table not found",
1402 __PRETTY_FUNCTION__
);
1406 rn
= route_node_match(table
, (struct prefix
*)&p
);
1408 route_unlock_node(rn
);
1410 /* Lookup should halt if we've matched against ourselves ('top',
1411 * if specified) - i.e., we cannot have a nexthop NH1 is
1412 * resolved by a route NH1. The exception is if the route is a
1415 if (top
&& rn
== top
)
1416 if (((afi
== AFI_IP
) && (rn
->p
.prefixlen
!= 32))
1417 || ((afi
== AFI_IP6
) && (rn
->p
.prefixlen
!= 128))) {
1418 if (IS_ZEBRA_DEBUG_RIB_DETAILED
)
1420 "\t%s: Matched against ourself and prefix length is not max bit length",
1421 __PRETTY_FUNCTION__
);
1425 /* Pick up selected route. */
1426 /* However, do not resolve over default route unless explicitly
1429 if (is_default_prefix(&rn
->p
)
1430 && !rnh_resolve_via_default(zvrf
, p
.family
)) {
1431 if (IS_ZEBRA_DEBUG_RIB_DETAILED
)
1433 "\t:%s: Resolved against default route",
1434 __PRETTY_FUNCTION__
);
1438 dest
= rib_dest_from_rnode(rn
);
1439 if (dest
&& dest
->selected_fib
1440 && !CHECK_FLAG(dest
->selected_fib
->status
,
1441 ROUTE_ENTRY_REMOVED
)
1442 && dest
->selected_fib
->type
!= ZEBRA_ROUTE_TABLE
)
1443 match
= dest
->selected_fib
;
1445 /* If there is no selected route or matched route is EGP, go up
1451 } while (rn
&& rn
->info
== NULL
);
1453 route_lock_node(rn
);
1458 if (match
->type
== ZEBRA_ROUTE_CONNECT
) {
1459 /* Directly point connected route. */
1460 newhop
= match
->nhe
->nhg
->nexthop
;
1462 if (nexthop
->type
== NEXTHOP_TYPE_IPV4
1463 || nexthop
->type
== NEXTHOP_TYPE_IPV6
)
1464 nexthop
->ifindex
= newhop
->ifindex
;
1467 } else if (CHECK_FLAG(re
->flags
, ZEBRA_FLAG_ALLOW_RECURSION
)) {
1469 for (ALL_NEXTHOPS_PTR(match
->nhe
->nhg
, newhop
)) {
1470 if (!CHECK_FLAG(match
->status
,
1471 ROUTE_ENTRY_INSTALLED
))
1473 if (!nexthop_valid_resolve(nexthop
, newhop
))
1476 SET_FLAG(nexthop
->flags
,
1477 NEXTHOP_FLAG_RECURSIVE
);
1478 nexthop_set_resolved(afi
, newhop
, nexthop
);
1482 re
->nexthop_mtu
= match
->mtu
;
1484 if (!resolved
&& IS_ZEBRA_DEBUG_RIB_DETAILED
)
1485 zlog_debug("\t%s: Recursion failed to find",
1486 __PRETTY_FUNCTION__
);
1488 } else if (re
->type
== ZEBRA_ROUTE_STATIC
) {
1490 for (ALL_NEXTHOPS_PTR(match
->nhe
->nhg
, newhop
)) {
1491 if (!CHECK_FLAG(match
->status
,
1492 ROUTE_ENTRY_INSTALLED
))
1494 if (!nexthop_valid_resolve(nexthop
, newhop
))
1497 SET_FLAG(nexthop
->flags
,
1498 NEXTHOP_FLAG_RECURSIVE
);
1499 nexthop_set_resolved(afi
, newhop
, nexthop
);
1503 re
->nexthop_mtu
= match
->mtu
;
1505 if (!resolved
&& IS_ZEBRA_DEBUG_RIB_DETAILED
)
1507 "\t%s: Static route unable to resolve",
1508 __PRETTY_FUNCTION__
);
1511 if (IS_ZEBRA_DEBUG_RIB_DETAILED
) {
1513 "\t%s: Route Type %s has not turned on recursion",
1514 __PRETTY_FUNCTION__
,
1515 zebra_route_string(re
->type
));
1516 if (re
->type
== ZEBRA_ROUTE_BGP
1517 && !CHECK_FLAG(re
->flags
, ZEBRA_FLAG_IBGP
))
1519 "\tEBGP: see \"disable-ebgp-connected-route-check\" or \"disable-connected-check\"");
1524 if (IS_ZEBRA_DEBUG_RIB_DETAILED
)
1525 zlog_debug("\t%s: Nexthop did not lookup in table",
1526 __PRETTY_FUNCTION__
);
1530 /* This function verifies reachability of one given nexthop, which can be
1531 * numbered or unnumbered, IPv4 or IPv6. The result is unconditionally stored
1532 * in nexthop->flags field. The nexthop->ifindex will be updated
1533 * appropriately as well. An existing route map can turn
1534 * (otherwise active) nexthop into inactive, but not vice versa.
1536 * If it finds a nexthop recursivedly, set the resolved_id
1537 * to match that nexthop's nhg_hash_entry ID;
1539 * The return value is the final value of 'ACTIVE' flag.
1541 static unsigned nexthop_active_check(struct route_node
*rn
,
1542 struct route_entry
*re
,
1543 struct nexthop
*nexthop
)
1545 struct interface
*ifp
;
1546 route_map_result_t ret
= RMAP_PERMITMATCH
;
1548 char buf
[SRCDEST2STR_BUFFER
];
1549 const struct prefix
*p
, *src_p
;
1550 struct zebra_vrf
*zvrf
;
1552 srcdest_rnode_prefixes(rn
, &p
, &src_p
);
1554 if (rn
->p
.family
== AF_INET
)
1556 else if (rn
->p
.family
== AF_INET6
)
1560 switch (nexthop
->type
) {
1561 case NEXTHOP_TYPE_IFINDEX
:
1562 ifp
= if_lookup_by_index(nexthop
->ifindex
, nexthop
->vrf_id
);
1563 if (ifp
&& if_is_operative(ifp
))
1564 SET_FLAG(nexthop
->flags
, NEXTHOP_FLAG_ACTIVE
);
1566 UNSET_FLAG(nexthop
->flags
, NEXTHOP_FLAG_ACTIVE
);
1568 case NEXTHOP_TYPE_IPV4
:
1569 case NEXTHOP_TYPE_IPV4_IFINDEX
:
1571 if (nexthop_active(AFI_IP
, re
, nexthop
, rn
))
1572 SET_FLAG(nexthop
->flags
, NEXTHOP_FLAG_ACTIVE
);
1574 UNSET_FLAG(nexthop
->flags
, NEXTHOP_FLAG_ACTIVE
);
1576 case NEXTHOP_TYPE_IPV6
:
1578 if (nexthop_active(AFI_IP6
, re
, nexthop
, rn
))
1579 SET_FLAG(nexthop
->flags
, NEXTHOP_FLAG_ACTIVE
);
1581 UNSET_FLAG(nexthop
->flags
, NEXTHOP_FLAG_ACTIVE
);
1583 case NEXTHOP_TYPE_IPV6_IFINDEX
:
1584 /* RFC 5549, v4 prefix with v6 NH */
1585 if (rn
->p
.family
!= AF_INET
)
1587 if (IN6_IS_ADDR_LINKLOCAL(&nexthop
->gate
.ipv6
)) {
1588 ifp
= if_lookup_by_index(nexthop
->ifindex
,
1590 if (ifp
&& if_is_operative(ifp
))
1591 SET_FLAG(nexthop
->flags
, NEXTHOP_FLAG_ACTIVE
);
1593 UNSET_FLAG(nexthop
->flags
, NEXTHOP_FLAG_ACTIVE
);
1595 if (nexthop_active(AFI_IP6
, re
, nexthop
, rn
))
1596 SET_FLAG(nexthop
->flags
, NEXTHOP_FLAG_ACTIVE
);
1598 UNSET_FLAG(nexthop
->flags
, NEXTHOP_FLAG_ACTIVE
);
1601 case NEXTHOP_TYPE_BLACKHOLE
:
1602 SET_FLAG(nexthop
->flags
, NEXTHOP_FLAG_ACTIVE
);
1607 if (!CHECK_FLAG(nexthop
->flags
, NEXTHOP_FLAG_ACTIVE
)) {
1608 if (IS_ZEBRA_DEBUG_RIB_DETAILED
)
1609 zlog_debug("\t%s: Unable to find a active nexthop",
1610 __PRETTY_FUNCTION__
);
1614 /* XXX: What exactly do those checks do? Do we support
1615 * e.g. IPv4 routes with IPv6 nexthops or vice versa?
1617 if (RIB_SYSTEM_ROUTE(re
) || (family
== AFI_IP
&& p
->family
!= AF_INET
)
1618 || (family
== AFI_IP6
&& p
->family
!= AF_INET6
))
1619 return CHECK_FLAG(nexthop
->flags
, NEXTHOP_FLAG_ACTIVE
);
1621 /* The original code didn't determine the family correctly
1622 * e.g. for NEXTHOP_TYPE_IFINDEX. Retrieve the correct afi
1623 * from the rib_table_info in those cases.
1624 * Possibly it may be better to use only the rib_table_info
1628 rib_table_info_t
*info
;
1630 info
= srcdest_rnode_table_info(rn
);
1634 memset(&nexthop
->rmap_src
.ipv6
, 0, sizeof(union g_addr
));
1636 zvrf
= zebra_vrf_lookup_by_id(nexthop
->vrf_id
);
1638 if (IS_ZEBRA_DEBUG_RIB_DETAILED
)
1639 zlog_debug("\t%s: zvrf is NULL", __PRETTY_FUNCTION__
);
1640 return CHECK_FLAG(nexthop
->flags
, NEXTHOP_FLAG_ACTIVE
);
1643 /* It'll get set if required inside */
1644 ret
= zebra_route_map_check(family
, re
->type
, re
->instance
, p
, nexthop
,
1646 if (ret
== RMAP_DENYMATCH
) {
1647 if (IS_ZEBRA_DEBUG_RIB
) {
1648 srcdest_rnode2str(rn
, buf
, sizeof(buf
));
1650 "%u:%s: Filtering out with NH out %s due to route map",
1652 ifindex2ifname(nexthop
->ifindex
,
1655 UNSET_FLAG(nexthop
->flags
, NEXTHOP_FLAG_ACTIVE
);
1657 return CHECK_FLAG(nexthop
->flags
, NEXTHOP_FLAG_ACTIVE
);
1661 * Iterate over all nexthops of the given RIB entry and refresh their
1662 * ACTIVE flag. If any nexthop is found to toggle the ACTIVE flag,
1663 * the whole re structure is flagged with ROUTE_ENTRY_CHANGED.
1665 * Return value is the new number of active nexthops.
1667 int nexthop_active_update(struct route_node
*rn
, struct route_entry
*re
)
1669 struct nexthop_group new_grp
= {};
1670 struct nexthop
*nexthop
;
1671 union g_addr prev_src
;
1672 unsigned int prev_active
, new_active
;
1673 ifindex_t prev_index
;
1674 uint8_t curr_active
= 0;
1676 afi_t rt_afi
= family2afi(rn
->p
.family
);
1678 UNSET_FLAG(re
->status
, ROUTE_ENTRY_CHANGED
);
1680 /* Copy over the nexthops in current state */
1681 nexthop_group_copy(&new_grp
, re
->nhe
->nhg
);
1683 for (nexthop
= new_grp
.nexthop
; nexthop
; nexthop
= nexthop
->next
) {
1685 /* No protocol daemon provides src and so we're skipping
1687 prev_src
= nexthop
->rmap_src
;
1688 prev_active
= CHECK_FLAG(nexthop
->flags
, NEXTHOP_FLAG_ACTIVE
);
1689 prev_index
= nexthop
->ifindex
;
1691 * We need to respect the multipath_num here
1692 * as that what we should be able to install from
1693 * a multipath perpsective should not be a data plane
1697 nexthop_active_check(rn
, re
, nexthop
);
1699 if (new_active
&& curr_active
>= zrouter
.multipath_num
) {
1702 /* Set it and its resolved nexthop as inactive. */
1703 for (nh
= nexthop
; nh
; nh
= nh
->resolved
)
1704 UNSET_FLAG(nh
->flags
, NEXTHOP_FLAG_ACTIVE
);
1712 /* Don't allow src setting on IPv6 addr for now */
1713 if (prev_active
!= new_active
|| prev_index
!= nexthop
->ifindex
1714 || ((nexthop
->type
>= NEXTHOP_TYPE_IFINDEX
1715 && nexthop
->type
< NEXTHOP_TYPE_IPV6
)
1716 && prev_src
.ipv4
.s_addr
1717 != nexthop
->rmap_src
.ipv4
.s_addr
)
1718 || ((nexthop
->type
>= NEXTHOP_TYPE_IPV6
1719 && nexthop
->type
< NEXTHOP_TYPE_BLACKHOLE
)
1720 && !(IPV6_ADDR_SAME(&prev_src
.ipv6
,
1721 &nexthop
->rmap_src
.ipv6
)))
1722 || CHECK_FLAG(re
->status
, ROUTE_ENTRY_LABELS_CHANGED
))
1723 SET_FLAG(re
->status
, ROUTE_ENTRY_CHANGED
);
1726 if (CHECK_FLAG(re
->status
, ROUTE_ENTRY_CHANGED
)) {
1727 struct nhg_hash_entry
*new_nhe
= NULL
;
1729 new_nhe
= zebra_nhg_rib_find(0, &new_grp
, rt_afi
);
1731 route_entry_update_nhe(re
, new_nhe
);
1735 struct nhg_hash_entry
*nhe
= NULL
;
1737 nhe
= zebra_nhg_lookup_id(re
->nhe_id
);
1740 SET_FLAG(nhe
->flags
, NEXTHOP_GROUP_VALID
);
1743 EC_ZEBRA_TABLE_LOOKUP_FAILED
,
1744 "Active update on NHE id=%u that we do not have in our tables",
1749 * Do not need these nexthops anymore since they
1750 * were either copied over into an nhe or not
1753 nexthops_free(new_grp
.nexthop
);
1757 /* Convert a nhe into a group array */
1758 uint8_t zebra_nhg_nhe2grp(struct nh_grp
*grp
, struct nhg_hash_entry
*nhe
,
1761 struct nhg_connected
*rb_node_dep
= NULL
;
1762 struct nhg_hash_entry
*depend
= NULL
;
1765 frr_each(nhg_connected_tree
, &nhe
->nhg_depends
, rb_node_dep
) {
1766 bool duplicate
= false;
1768 depend
= rb_node_dep
->nhe
;
1771 * If its recursive, use its resolved nhe in the group
1773 if (CHECK_FLAG(depend
->flags
, NEXTHOP_GROUP_RECURSIVE
)) {
1774 depend
= zebra_nhg_resolve(depend
);
1777 EC_ZEBRA_NHG_FIB_UPDATE
,
1778 "Failed to recursively resolve Nexthop Hash Entry in the group id=%u",
1784 /* Check for duplicate IDs, kernel doesn't like that */
1785 for (int j
= 0; j
< i
; j
++) {
1786 if (depend
->id
== grp
[j
].id
)
1791 grp
[i
].id
= depend
->id
;
1792 /* We aren't using weights for anything right now */
1793 grp
[i
].weight
= depend
->nhg
->nexthop
->weight
;
1805 void zebra_nhg_install_kernel(struct nhg_hash_entry
*nhe
)
1807 struct nhg_connected
*rb_node_dep
= NULL
;
1809 /* Resolve it first */
1810 nhe
= zebra_nhg_resolve(nhe
);
1812 /* Make sure all depends are installed/queued */
1813 frr_each(nhg_connected_tree
, &nhe
->nhg_depends
, rb_node_dep
) {
1814 zebra_nhg_install_kernel(rb_node_dep
->nhe
);
1817 if (!CHECK_FLAG(nhe
->flags
, NEXTHOP_GROUP_INSTALLED
)
1818 && !CHECK_FLAG(nhe
->flags
, NEXTHOP_GROUP_QUEUED
)) {
1819 /* Change its type to us since we are installing it */
1820 nhe
->type
= ZEBRA_ROUTE_NHG
;
1822 int ret
= dplane_nexthop_add(nhe
);
1825 case ZEBRA_DPLANE_REQUEST_QUEUED
:
1826 SET_FLAG(nhe
->flags
, NEXTHOP_GROUP_QUEUED
);
1828 case ZEBRA_DPLANE_REQUEST_FAILURE
:
1830 EC_ZEBRA_DP_INSTALL_FAIL
,
1831 "Failed to install Nexthop ID (%u) into the kernel",
1834 case ZEBRA_DPLANE_REQUEST_SUCCESS
:
1835 SET_FLAG(nhe
->flags
, NEXTHOP_GROUP_INSTALLED
);
1836 zebra_nhg_handle_install(nhe
);
1842 void zebra_nhg_uninstall_kernel(struct nhg_hash_entry
*nhe
)
1844 if (CHECK_FLAG(nhe
->flags
, NEXTHOP_GROUP_INSTALLED
)) {
1845 int ret
= dplane_nexthop_delete(nhe
);
1848 case ZEBRA_DPLANE_REQUEST_QUEUED
:
1849 SET_FLAG(nhe
->flags
, NEXTHOP_GROUP_QUEUED
);
1851 case ZEBRA_DPLANE_REQUEST_FAILURE
:
1853 EC_ZEBRA_DP_DELETE_FAIL
,
1854 "Failed to uninstall Nexthop ID (%u) from the kernel",
1857 case ZEBRA_DPLANE_REQUEST_SUCCESS
:
1858 UNSET_FLAG(nhe
->flags
, NEXTHOP_GROUP_INSTALLED
);
1863 zebra_nhg_handle_uninstall(nhe
);
1866 void zebra_nhg_dplane_result(struct zebra_dplane_ctx
*ctx
)
1868 enum dplane_op_e op
;
1869 enum zebra_dplane_result status
;
1871 struct nhg_hash_entry
*nhe
= NULL
;
1873 op
= dplane_ctx_get_op(ctx
);
1874 status
= dplane_ctx_get_status(ctx
);
1876 id
= dplane_ctx_get_nhe_id(ctx
);
1878 if (IS_ZEBRA_DEBUG_DPLANE_DETAIL
)
1880 "Nexthop dplane ctx %p, op %s, nexthop ID (%u), result %s",
1881 ctx
, dplane_op2str(op
), id
, dplane_res2str(status
));
1884 case DPLANE_OP_NH_DELETE
:
1885 if (status
!= ZEBRA_DPLANE_REQUEST_SUCCESS
)
1887 EC_ZEBRA_DP_DELETE_FAIL
,
1888 "Failed to uninstall Nexthop ID (%u) from the kernel",
1890 /* We already free'd the data, nothing to do */
1892 case DPLANE_OP_NH_INSTALL
:
1893 case DPLANE_OP_NH_UPDATE
:
1894 nhe
= zebra_nhg_lookup_id(id
);
1899 "%s operation preformed on Nexthop ID (%u) in the kernel, that we no longer have in our table",
1900 dplane_op2str(op
), id
);
1904 UNSET_FLAG(nhe
->flags
, NEXTHOP_GROUP_QUEUED
);
1905 if (status
== ZEBRA_DPLANE_REQUEST_SUCCESS
) {
1906 SET_FLAG(nhe
->flags
, NEXTHOP_GROUP_VALID
);
1907 SET_FLAG(nhe
->flags
, NEXTHOP_GROUP_INSTALLED
);
1908 zebra_nhg_handle_install(nhe
);
1911 EC_ZEBRA_DP_INSTALL_FAIL
,
1912 "Failed to install Nexthop ID (%u) into the kernel",
1915 case DPLANE_OP_ROUTE_INSTALL
:
1916 case DPLANE_OP_ROUTE_UPDATE
:
1917 case DPLANE_OP_ROUTE_DELETE
:
1918 case DPLANE_OP_ROUTE_NOTIFY
:
1919 case DPLANE_OP_LSP_INSTALL
:
1920 case DPLANE_OP_LSP_UPDATE
:
1921 case DPLANE_OP_LSP_DELETE
:
1922 case DPLANE_OP_LSP_NOTIFY
:
1923 case DPLANE_OP_PW_INSTALL
:
1924 case DPLANE_OP_PW_UNINSTALL
:
1925 case DPLANE_OP_SYS_ROUTE_ADD
:
1926 case DPLANE_OP_SYS_ROUTE_DELETE
:
1927 case DPLANE_OP_ADDR_INSTALL
:
1928 case DPLANE_OP_ADDR_UNINSTALL
:
1929 case DPLANE_OP_MAC_INSTALL
:
1930 case DPLANE_OP_MAC_DELETE
:
1931 case DPLANE_OP_NEIGH_INSTALL
:
1932 case DPLANE_OP_NEIGH_UPDATE
:
1933 case DPLANE_OP_NEIGH_DELETE
:
1934 case DPLANE_OP_VTEP_ADD
:
1935 case DPLANE_OP_VTEP_DELETE
:
1936 case DPLANE_OP_NONE
:
1940 dplane_ctx_fini(&ctx
);
1943 static void zebra_nhg_sweep_entry(struct hash_bucket
*bucket
, void *arg
)
1945 struct nhg_hash_entry
*nhe
= NULL
;
1947 nhe
= (struct nhg_hash_entry
*)bucket
->data
;
1949 /* If its being ref'd, just let it be uninstalled via a route removal */
1950 if (ZEBRA_NHG_CREATED(nhe
) && nhe
->refcnt
<= 0)
1951 zebra_nhg_uninstall_kernel(nhe
);
1954 void zebra_nhg_sweep_table(struct hash
*hash
)
1956 hash_iterate(hash
, zebra_nhg_sweep_entry
, NULL
);