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(struct nexthop
*nh
, afi_t afi
);
53 static void depends_add(struct nhg_connected_tree_head
*head
,
54 struct nhg_hash_entry
*depend
);
55 static struct nhg_hash_entry
*
56 depends_find_add(struct nhg_connected_tree_head
*head
, struct nexthop
*nh
,
58 static struct nhg_hash_entry
*
59 depends_find_id_add(struct nhg_connected_tree_head
*head
, uint32_t id
);
60 static void depends_decrement_free(struct nhg_connected_tree_head
*head
);
63 static void nhg_connected_free(struct nhg_connected
*dep
)
65 XFREE(MTYPE_NHG_CONNECTED
, dep
);
68 static struct nhg_connected
*nhg_connected_new(struct nhg_hash_entry
*nhe
)
70 struct nhg_connected
*new = NULL
;
72 new = XCALLOC(MTYPE_NHG_CONNECTED
, sizeof(struct nhg_connected
));
78 void nhg_connected_tree_free(struct nhg_connected_tree_head
*head
)
80 struct nhg_connected
*rb_node_dep
= NULL
;
82 if (!nhg_connected_tree_is_empty(head
)) {
83 frr_each_safe(nhg_connected_tree
, head
, rb_node_dep
) {
84 nhg_connected_tree_del(head
, rb_node_dep
);
85 nhg_connected_free(rb_node_dep
);
90 bool nhg_connected_tree_is_empty(const struct nhg_connected_tree_head
*head
)
92 return nhg_connected_tree_count(head
) ? false : true;
95 struct nhg_connected
*
96 nhg_connected_tree_root(struct nhg_connected_tree_head
*head
)
98 return nhg_connected_tree_first(head
);
101 void nhg_connected_tree_del_nhe(struct nhg_connected_tree_head
*head
,
102 struct nhg_hash_entry
*depend
)
104 struct nhg_connected lookup
= {};
105 struct nhg_connected
*remove
= NULL
;
109 /* Lookup to find the element, then remove it */
110 remove
= nhg_connected_tree_find(head
, &lookup
);
111 remove
= nhg_connected_tree_del(head
, remove
);
114 nhg_connected_free(remove
);
117 void nhg_connected_tree_add_nhe(struct nhg_connected_tree_head
*head
,
118 struct nhg_hash_entry
*depend
)
120 struct nhg_connected
*new = NULL
;
122 new = nhg_connected_new(depend
);
125 nhg_connected_tree_add(head
, new);
129 nhg_connected_tree_decrement_ref(struct nhg_connected_tree_head
*head
)
131 struct nhg_connected
*rb_node_dep
= NULL
;
133 frr_each_safe(nhg_connected_tree
, head
, rb_node_dep
) {
134 zebra_nhg_decrement_ref(rb_node_dep
->nhe
);
139 nhg_connected_tree_increment_ref(struct nhg_connected_tree_head
*head
)
141 struct nhg_connected
*rb_node_dep
= NULL
;
143 frr_each(nhg_connected_tree
, head
, rb_node_dep
) {
144 zebra_nhg_increment_ref(rb_node_dep
->nhe
);
148 struct nhg_hash_entry
*zebra_nhg_resolve(struct nhg_hash_entry
*nhe
)
150 if (CHECK_FLAG(nhe
->flags
, NEXTHOP_GROUP_RECURSIVE
)
151 && !zebra_nhg_depends_is_empty(nhe
)) {
152 nhe
= nhg_connected_tree_root(&nhe
->nhg_depends
)->nhe
;
153 return zebra_nhg_resolve(nhe
);
159 unsigned int zebra_nhg_depends_count(const struct nhg_hash_entry
*nhe
)
161 return nhg_connected_tree_count(&nhe
->nhg_depends
);
164 bool zebra_nhg_depends_is_empty(const struct nhg_hash_entry
*nhe
)
166 return nhg_connected_tree_is_empty(&nhe
->nhg_depends
);
169 static void zebra_nhg_depends_del(struct nhg_hash_entry
*from
,
170 struct nhg_hash_entry
*depend
)
172 nhg_connected_tree_del_nhe(&from
->nhg_depends
, depend
);
175 static void zebra_nhg_depends_init(struct nhg_hash_entry
*nhe
)
177 nhg_connected_tree_init(&nhe
->nhg_depends
);
180 unsigned int zebra_nhg_dependents_count(const struct nhg_hash_entry
*nhe
)
182 return nhg_connected_tree_count(&nhe
->nhg_dependents
);
186 bool zebra_nhg_dependents_is_empty(const struct nhg_hash_entry
*nhe
)
188 return nhg_connected_tree_is_empty(&nhe
->nhg_dependents
);
191 static void zebra_nhg_dependents_del(struct nhg_hash_entry
*from
,
192 struct nhg_hash_entry
*dependent
)
194 nhg_connected_tree_del_nhe(&from
->nhg_dependents
, dependent
);
197 static void zebra_nhg_dependents_add(struct nhg_hash_entry
*to
,
198 struct nhg_hash_entry
*dependent
)
200 nhg_connected_tree_add_nhe(&to
->nhg_dependents
, dependent
);
203 static void zebra_nhg_dependents_init(struct nhg_hash_entry
*nhe
)
205 nhg_connected_tree_init(&nhe
->nhg_dependents
);
208 /* Release this nhe from anything depending on it */
209 static void zebra_nhg_dependents_release(struct nhg_hash_entry
*nhe
)
211 struct nhg_connected
*rb_node_dep
= NULL
;
213 frr_each_safe(nhg_connected_tree
, &nhe
->nhg_dependents
, rb_node_dep
) {
214 zebra_nhg_depends_del(rb_node_dep
->nhe
, nhe
);
215 /* recheck validity of the dependent */
216 zebra_nhg_check_valid(rb_node_dep
->nhe
);
220 /* Release this nhe from anything that it depends on */
221 static void zebra_nhg_depends_release(struct nhg_hash_entry
*nhe
)
223 if (!zebra_nhg_depends_is_empty(nhe
)) {
224 struct nhg_connected
*rb_node_dep
= NULL
;
226 frr_each_safe(nhg_connected_tree
, &nhe
->nhg_depends
,
228 zebra_nhg_dependents_del(rb_node_dep
->nhe
, nhe
);
234 struct nhg_hash_entry
*zebra_nhg_lookup_id(uint32_t id
)
236 struct nhg_hash_entry lookup
= {};
239 return hash_lookup(zrouter
.nhgs_id
, &lookup
);
242 static int zebra_nhg_insert_id(struct nhg_hash_entry
*nhe
)
244 if (hash_lookup(zrouter
.nhgs_id
, nhe
)) {
246 EC_ZEBRA_NHG_TABLE_INSERT_FAILED
,
247 "Failed inserting NHG id=%u into the ID hash table, entry already exists",
252 hash_get(zrouter
.nhgs_id
, nhe
, hash_alloc_intern
);
257 static void zebra_nhg_set_if(struct nhg_hash_entry
*nhe
, struct interface
*ifp
)
260 if_nhg_dependents_add(ifp
, nhe
);
264 zebra_nhg_connect_depends(struct nhg_hash_entry
*nhe
,
265 struct nhg_connected_tree_head nhg_depends
)
267 struct nhg_connected
*rb_node_dep
= NULL
;
269 /* This has been allocated higher above in the stack. Could probably
270 * re-allocate and free the old stuff but just using the same memory
271 * for now. Otherwise, their might be a time trade-off for repeated
272 * alloc/frees as startup.
274 nhe
->nhg_depends
= nhg_depends
;
276 /* Attach backpointer to anything that it depends on */
277 zebra_nhg_dependents_init(nhe
);
278 if (!zebra_nhg_depends_is_empty(nhe
)) {
279 frr_each(nhg_connected_tree
, &nhe
->nhg_depends
, rb_node_dep
) {
280 zebra_nhg_dependents_add(rb_node_dep
->nhe
, nhe
);
284 /* Add the ifp now if its not a group or recursive and has ifindex */
285 if (zebra_nhg_depends_is_empty(nhe
) && nhe
->nhg
->nexthop
286 && nhe
->nhg
->nexthop
->ifindex
) {
287 struct interface
*ifp
= NULL
;
289 ifp
= if_lookup_by_index(nhe
->nhg
->nexthop
->ifindex
,
292 zebra_nhg_set_if(nhe
, ifp
);
295 EC_ZEBRA_IF_LOOKUP_FAILED
,
296 "Zebra failed to lookup an interface with ifindex=%d in vrf=%u for NHE id=%u",
297 nhe
->nhg
->nexthop
->ifindex
, nhe
->vrf_id
,
302 static struct nhg_hash_entry
*zebra_nhg_copy(struct nhg_hash_entry
*copy
,
305 struct nhg_hash_entry
*nhe
;
307 nhe
= XCALLOC(MTYPE_NHG
, sizeof(struct nhg_hash_entry
));
311 nhe
->nhg
= nexthop_group_new();
312 nexthop_group_copy(nhe
->nhg
, copy
->nhg
);
314 nhe
->vrf_id
= copy
->vrf_id
;
315 nhe
->afi
= copy
->afi
;
316 nhe
->type
= copy
->type
? copy
->type
: ZEBRA_ROUTE_NHG
;
318 nhe
->dplane_ref
= zebra_router_get_next_sequence();
323 /* Allocation via hash handler */
324 static void *zebra_nhg_hash_alloc(void *arg
)
326 struct nhg_hash_entry
*nhe
= NULL
;
327 struct nhg_hash_entry
*copy
= arg
;
329 nhe
= zebra_nhg_copy(copy
, copy
->id
);
331 /* Mark duplicate nexthops in a group at creation time. */
332 nexthop_group_mark_duplicates(nhe
->nhg
);
334 zebra_nhg_connect_depends(nhe
, copy
->nhg_depends
);
335 zebra_nhg_insert_id(nhe
);
340 uint32_t zebra_nhg_hash_key(const void *arg
)
342 const struct nhg_hash_entry
*nhe
= arg
;
344 uint32_t key
= 0x5a351234;
346 key
= jhash_3words(nhe
->vrf_id
, nhe
->afi
, nexthop_group_hash(nhe
->nhg
),
352 uint32_t zebra_nhg_id_key(const void *arg
)
354 const struct nhg_hash_entry
*nhe
= arg
;
359 bool zebra_nhg_hash_equal(const void *arg1
, const void *arg2
)
361 const struct nhg_hash_entry
*nhe1
= arg1
;
362 const struct nhg_hash_entry
*nhe2
= arg2
;
363 struct nexthop
*nexthop1
;
364 struct nexthop
*nexthop2
;
366 /* No matter what if they equal IDs, assume equal */
367 if (nhe1
->id
&& nhe2
->id
&& (nhe1
->id
== nhe2
->id
))
370 if (nhe1
->vrf_id
!= nhe2
->vrf_id
)
373 if (nhe1
->afi
!= nhe2
->afi
)
376 /* Nexthops should be sorted */
377 for (nexthop1
= nhe1
->nhg
->nexthop
, nexthop2
= nhe2
->nhg
->nexthop
;
378 nexthop1
|| nexthop2
;
379 nexthop1
= nexthop1
->next
, nexthop2
= nexthop2
->next
) {
380 if (nexthop1
&& !nexthop2
)
383 if (!nexthop1
&& nexthop2
)
387 * We have to check the active flag of each individual one,
388 * not just the overall active_num. This solves the special case
389 * issue of a route with a nexthop group with one nexthop
390 * resolving to itself and thus marking it inactive. If we
391 * have two different routes each wanting to mark a different
392 * nexthop inactive, they need to hash to two different groups.
394 * If we just hashed on num_active, they would hash the same
395 * which is incorrect.
399 * -> 1.1.1.1 dummy1 (inactive)
404 * -> 1.1.2.1 dummy2 (inactive)
406 * Without checking each individual one, they would hash to
407 * the same group and both have 1.1.1.1 dummy1 marked inactive.
410 if (CHECK_FLAG(nexthop1
->flags
, NEXTHOP_FLAG_ACTIVE
)
411 != CHECK_FLAG(nexthop2
->flags
, NEXTHOP_FLAG_ACTIVE
))
414 if (!nexthop_same(nexthop1
, nexthop2
))
421 bool zebra_nhg_hash_id_equal(const void *arg1
, const void *arg2
)
423 const struct nhg_hash_entry
*nhe1
= arg1
;
424 const struct nhg_hash_entry
*nhe2
= arg2
;
426 return nhe1
->id
== nhe2
->id
;
429 static int zebra_nhg_process_grp(struct nexthop_group
*nhg
,
430 struct nhg_connected_tree_head
*depends
,
431 struct nh_grp
*grp
, uint8_t count
)
433 nhg_connected_tree_init(depends
);
435 for (int i
= 0; i
< count
; i
++) {
436 struct nhg_hash_entry
*depend
= NULL
;
437 /* We do not care about nexthop_grp.weight at
438 * this time. But we should figure out
439 * how to adapt this to our code in
442 depend
= depends_find_id_add(depends
, grp
[i
].id
);
447 "Received Nexthop Group from the kernel with a dependent Nexthop ID (%u) which we do not have in our table",
453 * If this is a nexthop with its own group
454 * dependencies, add them as well. Not sure its
455 * even possible to have a group within a group
459 copy_nexthops(&nhg
->nexthop
, depend
->nhg
->nexthop
, NULL
);
465 static void handle_recursive_depend(struct nhg_connected_tree_head
*nhg_depends
,
466 struct nexthop
*nh
, afi_t afi
)
468 struct nhg_hash_entry
*depend
= NULL
;
469 struct nexthop_group resolved_ng
= {};
471 _nexthop_group_add_sorted(&resolved_ng
, nh
);
473 depend
= zebra_nhg_rib_find(0, &resolved_ng
, afi
);
474 depends_add(nhg_depends
, depend
);
477 static bool zebra_nhg_find(struct nhg_hash_entry
**nhe
, uint32_t id
,
478 struct nexthop_group
*nhg
,
479 struct nhg_connected_tree_head
*nhg_depends
,
480 vrf_id_t vrf_id
, afi_t afi
, int type
)
482 struct nhg_hash_entry lookup
= {};
484 uint32_t old_id_counter
= id_counter
;
486 bool created
= false;
487 bool recursive
= false;
490 * If it has an id at this point, we must have gotten it from the kernel
492 lookup
.id
= id
? id
: ++id_counter
;
494 lookup
.type
= type
? type
: ZEBRA_ROUTE_NHG
;
497 if (lookup
.nhg
->nexthop
->next
) {
498 /* Groups can have all vrfs and AF's in them */
499 lookup
.afi
= AFI_UNSPEC
;
502 switch (lookup
.nhg
->nexthop
->type
) {
503 case (NEXTHOP_TYPE_IFINDEX
):
504 case (NEXTHOP_TYPE_BLACKHOLE
):
506 * This switch case handles setting the afi different
507 * for ipv4/v6 routes. Ifindex/blackhole nexthop
508 * objects cannot be ambiguous, they must be Address
509 * Family specific. If we get here, we will either use
510 * the AF of the route, or the one we got passed from
511 * here from the kernel.
515 case (NEXTHOP_TYPE_IPV4_IFINDEX
):
516 case (NEXTHOP_TYPE_IPV4
):
519 case (NEXTHOP_TYPE_IPV6_IFINDEX
):
520 case (NEXTHOP_TYPE_IPV6
):
521 lookup
.afi
= AFI_IP6
;
525 lookup
.vrf_id
= vrf_id
;
529 (*nhe
) = zebra_nhg_lookup_id(id
);
531 (*nhe
) = hash_lookup(zrouter
.nhgs
, &lookup
);
533 /* If it found an nhe in our tables, this new ID is unused */
535 id_counter
= old_id_counter
;
538 /* Only hash/lookup the depends if the first lookup
539 * fails to find something. This should hopefully save a
540 * lot of cycles for larger ecmp sizes.
543 /* If you don't want to hash on each nexthop in the
544 * nexthop group struct you can pass the depends
545 * directly. Kernel-side we do this since it just looks
548 lookup
.nhg_depends
= *nhg_depends
;
550 if (nhg
->nexthop
->next
) {
551 zebra_nhg_depends_init(&lookup
);
553 /* If its a group, create a dependency tree */
554 struct nexthop
*nh
= NULL
;
556 for (nh
= nhg
->nexthop
; nh
; nh
= nh
->next
)
557 depends_find_add(&lookup
.nhg_depends
,
559 } else if (CHECK_FLAG(nhg
->nexthop
->flags
,
560 NEXTHOP_FLAG_RECURSIVE
)) {
561 zebra_nhg_depends_init(&lookup
);
562 handle_recursive_depend(&lookup
.nhg_depends
,
563 nhg
->nexthop
->resolved
,
569 (*nhe
) = hash_get(zrouter
.nhgs
, &lookup
, zebra_nhg_hash_alloc
);
573 SET_FLAG((*nhe
)->flags
, NEXTHOP_GROUP_RECURSIVE
);
578 /* Find/create a single nexthop */
579 static struct nhg_hash_entry
*
580 zebra_nhg_find_nexthop(uint32_t id
, struct nexthop
*nh
, afi_t afi
, int type
)
582 struct nhg_hash_entry
*nhe
= NULL
;
583 struct nexthop_group nhg
= {};
585 _nexthop_group_add_sorted(&nhg
, nh
);
587 zebra_nhg_find(&nhe
, id
, &nhg
, NULL
, nh
->vrf_id
, afi
, 0);
592 static uint32_t nhg_ctx_get_id(const struct nhg_ctx
*ctx
)
597 static void nhg_ctx_set_status(struct nhg_ctx
*ctx
, enum nhg_ctx_status status
)
599 ctx
->status
= status
;
602 static enum nhg_ctx_status
nhg_ctx_get_status(const struct nhg_ctx
*ctx
)
607 static void nhg_ctx_set_op(struct nhg_ctx
*ctx
, enum nhg_ctx_op_e op
)
612 static enum nhg_ctx_op_e
nhg_ctx_get_op(const struct nhg_ctx
*ctx
)
617 static vrf_id_t
nhg_ctx_get_vrf_id(const struct nhg_ctx
*ctx
)
622 static int nhg_ctx_get_type(const struct nhg_ctx
*ctx
)
627 static int nhg_ctx_get_afi(const struct nhg_ctx
*ctx
)
632 static struct nexthop
*nhg_ctx_get_nh(struct nhg_ctx
*ctx
)
637 static uint8_t nhg_ctx_get_count(const struct nhg_ctx
*ctx
)
642 static struct nh_grp
*nhg_ctx_get_grp(struct nhg_ctx
*ctx
)
647 static struct nhg_ctx
*nhg_ctx_new()
649 struct nhg_ctx
*new = NULL
;
651 new = XCALLOC(MTYPE_NHG_CTX
, sizeof(struct nhg_ctx
));
656 static void nhg_ctx_free(struct nhg_ctx
**ctx
)
663 assert((*ctx
) != NULL
);
665 if (nhg_ctx_get_count(*ctx
))
668 nh
= nhg_ctx_get_nh(*ctx
);
670 nexthop_del_labels(nh
);
673 XFREE(MTYPE_NHG_CTX
, *ctx
);
677 static struct nhg_ctx
*nhg_ctx_init(uint32_t id
, struct nexthop
*nh
,
678 struct nh_grp
*grp
, vrf_id_t vrf_id
,
679 afi_t afi
, int type
, uint8_t count
)
681 struct nhg_ctx
*ctx
= NULL
;
686 ctx
->vrf_id
= vrf_id
;
692 /* Copy over the array */
693 memcpy(&ctx
->u
.grp
, grp
, count
* sizeof(struct nh_grp
));
700 static bool zebra_nhg_contains_unhashable(struct nhg_hash_entry
*nhe
)
702 struct nhg_connected
*rb_node_dep
= NULL
;
704 frr_each(nhg_connected_tree
, &nhe
->nhg_depends
, rb_node_dep
) {
705 if (CHECK_FLAG(rb_node_dep
->nhe
->flags
,
706 NEXTHOP_GROUP_UNHASHABLE
))
713 static void zebra_nhg_set_unhashable(struct nhg_hash_entry
*nhe
)
715 SET_FLAG(nhe
->flags
, NEXTHOP_GROUP_UNHASHABLE
);
716 SET_FLAG(nhe
->flags
, NEXTHOP_GROUP_INSTALLED
);
719 EC_ZEBRA_DUPLICATE_NHG_MESSAGE
,
720 "Nexthop Group with ID (%d) is a duplicate, therefore unhashable, ignoring",
724 static void zebra_nhg_set_valid(struct nhg_hash_entry
*nhe
)
726 struct nhg_connected
*rb_node_dep
;
728 SET_FLAG(nhe
->flags
, NEXTHOP_GROUP_VALID
);
730 frr_each(nhg_connected_tree
, &nhe
->nhg_dependents
, rb_node_dep
)
731 zebra_nhg_set_valid(rb_node_dep
->nhe
);
734 static void zebra_nhg_set_invalid(struct nhg_hash_entry
*nhe
)
736 struct nhg_connected
*rb_node_dep
;
738 UNSET_FLAG(nhe
->flags
, NEXTHOP_GROUP_VALID
);
740 /* Update validity of nexthops depending on it */
741 frr_each(nhg_connected_tree
, &nhe
->nhg_dependents
, rb_node_dep
)
742 zebra_nhg_check_valid(rb_node_dep
->nhe
);
745 void zebra_nhg_check_valid(struct nhg_hash_entry
*nhe
)
747 struct nhg_connected
*rb_node_dep
= NULL
;
750 /* If anthing else in the group is valid, the group is valid */
751 frr_each(nhg_connected_tree
, &nhe
->nhg_depends
, rb_node_dep
) {
752 if (CHECK_FLAG(rb_node_dep
->nhe
->flags
, NEXTHOP_GROUP_VALID
)) {
760 zebra_nhg_set_valid(nhe
);
762 zebra_nhg_set_invalid(nhe
);
766 static void zebra_nhg_release(struct nhg_hash_entry
*nhe
)
768 /* Remove it from any lists it may be on */
769 zebra_nhg_depends_release(nhe
);
770 zebra_nhg_dependents_release(nhe
);
772 if_nhg_dependents_del(nhe
->ifp
, nhe
);
775 * If its unhashable, we didn't store it here and have to be
776 * sure we don't clear one thats actually being used.
778 if (!CHECK_FLAG(nhe
->flags
, NEXTHOP_GROUP_UNHASHABLE
))
779 hash_release(zrouter
.nhgs
, nhe
);
781 hash_release(zrouter
.nhgs_id
, nhe
);
784 static void zebra_nhg_handle_uninstall(struct nhg_hash_entry
*nhe
)
786 zebra_nhg_release(nhe
);
790 static void zebra_nhg_handle_install(struct nhg_hash_entry
*nhe
)
792 /* Update validity of groups depending on it */
793 struct nhg_connected
*rb_node_dep
;
795 frr_each_safe(nhg_connected_tree
, &nhe
->nhg_dependents
, rb_node_dep
)
796 zebra_nhg_set_valid(rb_node_dep
->nhe
);
800 * The kernel/other program has changed the state of a nexthop object we are
803 static void zebra_nhg_handle_kernel_state_change(struct nhg_hash_entry
*nhe
,
809 "Kernel %s a nexthop group with ID (%u) that we are still using for a route, sending it back down",
810 (is_delete
? "deleted" : "updated"), nhe
->id
);
812 UNSET_FLAG(nhe
->flags
, NEXTHOP_GROUP_INSTALLED
);
813 zebra_nhg_install_kernel(nhe
);
815 zebra_nhg_handle_uninstall(nhe
);
818 static int nhg_ctx_process_new(struct nhg_ctx
*ctx
)
820 struct nexthop_group
*nhg
= NULL
;
821 struct nhg_connected_tree_head nhg_depends
= {};
822 struct nhg_hash_entry
*lookup
= NULL
;
823 struct nhg_hash_entry
*nhe
= NULL
;
825 uint32_t id
= nhg_ctx_get_id(ctx
);
826 uint8_t count
= nhg_ctx_get_count(ctx
);
827 vrf_id_t vrf_id
= nhg_ctx_get_vrf_id(ctx
);
828 int type
= nhg_ctx_get_type(ctx
);
829 afi_t afi
= nhg_ctx_get_afi(ctx
);
831 lookup
= zebra_nhg_lookup_id(id
);
834 /* This is already present in our table, hence an update
835 * that we did not initate.
837 zebra_nhg_handle_kernel_state_change(lookup
, false);
841 if (nhg_ctx_get_count(ctx
)) {
842 nhg
= nexthop_group_new();
843 if (zebra_nhg_process_grp(nhg
, &nhg_depends
,
844 nhg_ctx_get_grp(ctx
), count
)) {
845 depends_decrement_free(&nhg_depends
);
846 nexthop_group_delete(&nhg
);
850 if (!zebra_nhg_find(&nhe
, id
, nhg
, &nhg_depends
, vrf_id
, type
,
852 depends_decrement_free(&nhg_depends
);
854 /* These got copied over in zebra_nhg_alloc() */
855 nexthop_group_delete(&nhg
);
857 nhe
= zebra_nhg_find_nexthop(id
, nhg_ctx_get_nh(ctx
), afi
,
862 struct nhg_hash_entry
*kernel_nhe
= NULL
;
864 /* Duplicate but with different ID from
868 /* The kernel allows duplicate nexthops
869 * as long as they have different IDs.
870 * We are ignoring those to prevent
871 * syncing problems with the kernel
874 * We maintain them *ONLY* in the ID hash table to
875 * track them and set the flag to indicated
876 * their attributes are unhashable.
879 kernel_nhe
= zebra_nhg_copy(nhe
, id
);
880 zebra_nhg_insert_id(kernel_nhe
);
881 zebra_nhg_set_unhashable(kernel_nhe
);
882 } else if (zebra_nhg_contains_unhashable(nhe
)) {
883 /* The group we got contains an unhashable/duplicated
884 * depend, so lets mark this group as unhashable as well
885 * and release it from the non-ID hash.
887 hash_release(zrouter
.nhgs
, nhe
);
888 zebra_nhg_set_unhashable(nhe
);
890 /* It actually created a new nhe */
891 SET_FLAG(nhe
->flags
, NEXTHOP_GROUP_VALID
);
892 SET_FLAG(nhe
->flags
, NEXTHOP_GROUP_INSTALLED
);
896 EC_ZEBRA_TABLE_LOOKUP_FAILED
,
897 "Zebra failed to find or create a nexthop hash entry for ID (%u)",
905 static int nhg_ctx_process_del(struct nhg_ctx
*ctx
)
907 struct nhg_hash_entry
*nhe
= NULL
;
908 uint32_t id
= nhg_ctx_get_id(ctx
);
910 nhe
= zebra_nhg_lookup_id(id
);
914 EC_ZEBRA_BAD_NHG_MESSAGE
,
915 "Kernel delete message received for nexthop group ID (%u) that we do not have in our ID table",
920 zebra_nhg_handle_kernel_state_change(nhe
, true);
925 static void nhg_ctx_fini(struct nhg_ctx
**ctx
)
928 * Just freeing for now, maybe do something more in the future
935 static int queue_add(struct nhg_ctx
*ctx
)
937 /* If its queued or already processed do nothing */
938 if (nhg_ctx_get_status(ctx
) == NHG_CTX_QUEUED
)
941 if (rib_queue_nhg_add(ctx
)) {
942 nhg_ctx_set_status(ctx
, NHG_CTX_FAILURE
);
946 nhg_ctx_set_status(ctx
, NHG_CTX_QUEUED
);
951 int nhg_ctx_process(struct nhg_ctx
*ctx
)
955 switch (nhg_ctx_get_op(ctx
)) {
957 ret
= nhg_ctx_process_new(ctx
);
958 if (nhg_ctx_get_count(ctx
) && ret
== -ENOENT
959 && nhg_ctx_get_status(ctx
) != NHG_CTX_REQUEUED
) {
961 * We have entered a situation where we are
962 * processing a group from the kernel
963 * that has a contained nexthop which
964 * we have not yet processed.
966 * Re-enqueue this ctx to be handled exactly one
967 * more time (indicated by the flag).
969 * By the time we get back to it, we
970 * should have processed its depends.
972 nhg_ctx_set_status(ctx
, NHG_CTX_NONE
);
973 if (queue_add(ctx
) == 0) {
974 nhg_ctx_set_status(ctx
, NHG_CTX_REQUEUED
);
980 ret
= nhg_ctx_process_del(ctx
);
981 case NHG_CTX_OP_NONE
:
985 nhg_ctx_set_status(ctx
, (ret
? NHG_CTX_FAILURE
: NHG_CTX_SUCCESS
));
992 /* Kernel-side, you either get a single new nexthop or a array of ID's */
993 int zebra_nhg_kernel_find(uint32_t id
, struct nexthop
*nh
, struct nh_grp
*grp
,
994 uint8_t count
, vrf_id_t vrf_id
, afi_t afi
, int type
,
997 struct nhg_ctx
*ctx
= NULL
;
1000 /* Increase our counter so we don't try to create
1001 * an ID that already exists
1005 ctx
= nhg_ctx_init(id
, nh
, grp
, vrf_id
, afi
, type
, count
);
1006 nhg_ctx_set_op(ctx
, NHG_CTX_OP_NEW
);
1008 /* Under statup conditions, we need to handle them immediately
1009 * like we do for routes. Otherwise, we are going to get a route
1010 * with a nhe_id that we have not handled.
1013 return nhg_ctx_process(ctx
);
1015 if (queue_add(ctx
)) {
1023 /* Kernel-side, received delete message */
1024 int zebra_nhg_kernel_del(uint32_t id
)
1026 struct nhg_ctx
*ctx
= NULL
;
1028 ctx
= nhg_ctx_init(id
, NULL
, NULL
, 0, 0, 0, 0);
1030 nhg_ctx_set_op(ctx
, NHG_CTX_OP_DEL
);
1032 if (queue_add(ctx
)) {
1040 /* Some dependency helper functions */
1041 static struct nhg_hash_entry
*depends_find(struct nexthop
*nh
, afi_t afi
)
1043 struct nexthop
*lookup
= NULL
;
1044 struct nhg_hash_entry
*nhe
= NULL
;
1049 copy_nexthops(&lookup
, nh
, NULL
);
1051 /* Clear it, in case its a group */
1052 nexthops_free(lookup
->next
);
1053 nexthops_free(lookup
->prev
);
1054 lookup
->next
= NULL
;
1055 lookup
->prev
= NULL
;
1057 nhe
= zebra_nhg_find_nexthop(0, lookup
, afi
, 0);
1059 nexthops_free(lookup
);
1065 static void depends_add(struct nhg_connected_tree_head
*head
,
1066 struct nhg_hash_entry
*depend
)
1068 nhg_connected_tree_add_nhe(head
, depend
);
1069 zebra_nhg_increment_ref(depend
);
1072 static struct nhg_hash_entry
*
1073 depends_find_add(struct nhg_connected_tree_head
*head
, struct nexthop
*nh
,
1076 struct nhg_hash_entry
*depend
= NULL
;
1078 depend
= depends_find(nh
, afi
);
1081 depends_add(head
, depend
);
1086 static struct nhg_hash_entry
*
1087 depends_find_id_add(struct nhg_connected_tree_head
*head
, uint32_t id
)
1089 struct nhg_hash_entry
*depend
= NULL
;
1091 depend
= zebra_nhg_lookup_id(id
);
1094 depends_add(head
, depend
);
1099 static void depends_decrement_free(struct nhg_connected_tree_head
*head
)
1101 nhg_connected_tree_decrement_ref(head
);
1102 nhg_connected_tree_free(head
);
1105 /* Rib-side, you get a nexthop group struct */
1106 struct nhg_hash_entry
*
1107 zebra_nhg_rib_find(uint32_t id
, struct nexthop_group
*nhg
, afi_t rt_afi
)
1109 struct nhg_hash_entry
*nhe
= NULL
;
1111 if (!(nhg
&& nhg
->nexthop
)) {
1112 flog_err(EC_ZEBRA_TABLE_LOOKUP_FAILED
,
1113 "No nexthop passed to %s", __func__
);
1117 zebra_nhg_find(&nhe
, id
, nhg
, NULL
, nhg
->nexthop
->vrf_id
, rt_afi
, 0);
1122 static void zebra_nhg_free_members(struct nhg_hash_entry
*nhe
)
1124 nexthop_group_delete(&nhe
->nhg
);
1125 /* Decrement to remove connection ref */
1126 nhg_connected_tree_decrement_ref(&nhe
->nhg_depends
);
1127 nhg_connected_tree_free(&nhe
->nhg_depends
);
1128 nhg_connected_tree_free(&nhe
->nhg_dependents
);
1131 void zebra_nhg_free(void *arg
)
1133 struct nhg_hash_entry
*nhe
= NULL
;
1135 nhe
= (struct nhg_hash_entry
*)arg
;
1138 zlog_debug("nhe_id=%u hash refcnt=%d", nhe
->id
, nhe
->refcnt
);
1140 zebra_nhg_free_members(nhe
);
1142 XFREE(MTYPE_NHG
, nhe
);
1145 void zebra_nhg_decrement_ref(struct nhg_hash_entry
*nhe
)
1149 if (!zebra_nhg_depends_is_empty(nhe
))
1150 nhg_connected_tree_decrement_ref(&nhe
->nhg_depends
);
1152 if (ZEBRA_NHG_CREATED(nhe
) && nhe
->refcnt
<= 0)
1153 zebra_nhg_uninstall_kernel(nhe
);
1156 void zebra_nhg_increment_ref(struct nhg_hash_entry
*nhe
)
1160 if (!zebra_nhg_depends_is_empty(nhe
))
1161 nhg_connected_tree_increment_ref(&nhe
->nhg_depends
);
1164 static void nexthop_set_resolved(afi_t afi
, const struct nexthop
*newhop
,
1165 struct nexthop
*nexthop
)
1167 struct nexthop
*resolved_hop
;
1168 uint8_t num_labels
= 0;
1169 mpls_label_t labels
[MPLS_MAX_LABELS
];
1170 enum lsp_types_t label_type
= ZEBRA_LSP_NONE
;
1173 resolved_hop
= nexthop_new();
1174 SET_FLAG(resolved_hop
->flags
, NEXTHOP_FLAG_ACTIVE
);
1176 resolved_hop
->vrf_id
= nexthop
->vrf_id
;
1177 switch (newhop
->type
) {
1178 case NEXTHOP_TYPE_IPV4
:
1179 case NEXTHOP_TYPE_IPV4_IFINDEX
:
1180 /* If the resolving route specifies a gateway, use it */
1181 resolved_hop
->type
= newhop
->type
;
1182 resolved_hop
->gate
.ipv4
= newhop
->gate
.ipv4
;
1184 if (newhop
->ifindex
) {
1185 resolved_hop
->type
= NEXTHOP_TYPE_IPV4_IFINDEX
;
1186 resolved_hop
->ifindex
= newhop
->ifindex
;
1189 case NEXTHOP_TYPE_IPV6
:
1190 case NEXTHOP_TYPE_IPV6_IFINDEX
:
1191 resolved_hop
->type
= newhop
->type
;
1192 resolved_hop
->gate
.ipv6
= newhop
->gate
.ipv6
;
1194 if (newhop
->ifindex
) {
1195 resolved_hop
->type
= NEXTHOP_TYPE_IPV6_IFINDEX
;
1196 resolved_hop
->ifindex
= newhop
->ifindex
;
1199 case NEXTHOP_TYPE_IFINDEX
:
1200 /* If the resolving route is an interface route,
1201 * it means the gateway we are looking up is connected
1202 * to that interface. (The actual network is _not_ onlink).
1203 * Therefore, the resolved route should have the original
1204 * gateway as nexthop as it is directly connected.
1206 * On Linux, we have to set the onlink netlink flag because
1207 * otherwise, the kernel won't accept the route.
1209 resolved_hop
->flags
|= NEXTHOP_FLAG_ONLINK
;
1210 if (afi
== AFI_IP
) {
1211 resolved_hop
->type
= NEXTHOP_TYPE_IPV4_IFINDEX
;
1212 resolved_hop
->gate
.ipv4
= nexthop
->gate
.ipv4
;
1213 } else if (afi
== AFI_IP6
) {
1214 resolved_hop
->type
= NEXTHOP_TYPE_IPV6_IFINDEX
;
1215 resolved_hop
->gate
.ipv6
= nexthop
->gate
.ipv6
;
1217 resolved_hop
->ifindex
= newhop
->ifindex
;
1219 case NEXTHOP_TYPE_BLACKHOLE
:
1220 resolved_hop
->type
= NEXTHOP_TYPE_BLACKHOLE
;
1221 resolved_hop
->bh_type
= newhop
->bh_type
;
1225 if (newhop
->flags
& NEXTHOP_FLAG_ONLINK
)
1226 resolved_hop
->flags
|= NEXTHOP_FLAG_ONLINK
;
1228 /* Copy labels of the resolved route and the parent resolving to it */
1229 if (newhop
->nh_label
) {
1230 for (i
= 0; i
< newhop
->nh_label
->num_labels
; i
++)
1231 labels
[num_labels
++] = newhop
->nh_label
->label
[i
];
1232 label_type
= newhop
->nh_label_type
;
1235 if (nexthop
->nh_label
) {
1236 for (i
= 0; i
< nexthop
->nh_label
->num_labels
; i
++)
1237 labels
[num_labels
++] = nexthop
->nh_label
->label
[i
];
1239 /* If the parent has labels, use its type */
1240 label_type
= nexthop
->nh_label_type
;
1244 nexthop_add_labels(resolved_hop
, label_type
, num_labels
,
1247 resolved_hop
->rparent
= nexthop
;
1248 _nexthop_add(&nexthop
->resolved
, resolved_hop
);
1251 /* Checks if nexthop we are trying to resolve to is valid */
1252 static bool nexthop_valid_resolve(const struct nexthop
*nexthop
,
1253 const struct nexthop
*resolved
)
1255 /* Can't resolve to a recursive nexthop */
1256 if (CHECK_FLAG(resolved
->flags
, NEXTHOP_FLAG_RECURSIVE
))
1259 switch (nexthop
->type
) {
1260 case NEXTHOP_TYPE_IPV4_IFINDEX
:
1261 case NEXTHOP_TYPE_IPV6_IFINDEX
:
1262 /* If the nexthop we are resolving to does not match the
1263 * ifindex for the nexthop the route wanted, its not valid.
1265 if (nexthop
->ifindex
!= resolved
->ifindex
)
1268 case NEXTHOP_TYPE_IPV4
:
1269 case NEXTHOP_TYPE_IPV6
:
1270 case NEXTHOP_TYPE_IFINDEX
:
1271 case NEXTHOP_TYPE_BLACKHOLE
:
1279 * Given a nexthop we need to properly recursively resolve
1280 * the route. As such, do a table lookup to find and match
1281 * if at all possible. Set the nexthop->ifindex and resolved_id
1284 static int nexthop_active(afi_t afi
, struct route_entry
*re
,
1285 struct nexthop
*nexthop
, struct route_node
*top
)
1288 struct route_table
*table
;
1289 struct route_node
*rn
;
1290 struct route_entry
*match
= NULL
;
1292 struct nexthop
*newhop
;
1293 struct interface
*ifp
;
1295 struct zebra_vrf
*zvrf
;
1297 if ((nexthop
->type
== NEXTHOP_TYPE_IPV4
)
1298 || nexthop
->type
== NEXTHOP_TYPE_IPV6
)
1299 nexthop
->ifindex
= 0;
1302 UNSET_FLAG(nexthop
->flags
, NEXTHOP_FLAG_RECURSIVE
);
1303 nexthops_free(nexthop
->resolved
);
1304 nexthop
->resolved
= NULL
;
1305 re
->nexthop_mtu
= 0;
1308 * If the kernel has sent us a NEW route, then
1309 * by golly gee whiz it's a good route.
1311 * If its an already INSTALLED route we have already handled, then the
1312 * kernel route's nexthop might have became unreachable
1313 * and we have to handle that.
1315 if (!CHECK_FLAG(re
->status
, ROUTE_ENTRY_INSTALLED
)
1316 && (re
->type
== ZEBRA_ROUTE_KERNEL
1317 || re
->type
== ZEBRA_ROUTE_SYSTEM
))
1321 * Check to see if we should trust the passed in information
1322 * for UNNUMBERED interfaces as that we won't find the GW
1323 * address in the routing table.
1324 * This check should suffice to handle IPv4 or IPv6 routes
1325 * sourced from EVPN routes which are installed with the
1326 * next hop as the remote VTEP IP.
1328 if (CHECK_FLAG(nexthop
->flags
, NEXTHOP_FLAG_ONLINK
)) {
1329 ifp
= if_lookup_by_index(nexthop
->ifindex
, nexthop
->vrf_id
);
1331 if (IS_ZEBRA_DEBUG_RIB_DETAILED
)
1333 "\t%s: Onlink and interface: %u[%u] does not exist",
1334 __PRETTY_FUNCTION__
, nexthop
->ifindex
,
1338 if (connected_is_unnumbered(ifp
)) {
1339 if (if_is_operative(ifp
))
1342 if (IS_ZEBRA_DEBUG_RIB_DETAILED
)
1344 "\t%s: Onlink and interface %s is not operative",
1345 __PRETTY_FUNCTION__
, ifp
->name
);
1348 if (!if_is_operative(ifp
)) {
1349 if (IS_ZEBRA_DEBUG_RIB_DETAILED
)
1351 "\t%s: Interface %s is not unnumbered",
1352 __PRETTY_FUNCTION__
, ifp
->name
);
1357 /* Make lookup prefix. */
1358 memset(&p
, 0, sizeof(struct prefix
));
1362 p
.prefixlen
= IPV4_MAX_PREFIXLEN
;
1363 p
.u
.prefix4
= nexthop
->gate
.ipv4
;
1366 p
.family
= AF_INET6
;
1367 p
.prefixlen
= IPV6_MAX_PREFIXLEN
;
1368 p
.u
.prefix6
= nexthop
->gate
.ipv6
;
1371 assert(afi
!= AFI_IP
&& afi
!= AFI_IP6
);
1375 table
= zebra_vrf_table(afi
, SAFI_UNICAST
, nexthop
->vrf_id
);
1377 zvrf
= zebra_vrf_lookup_by_id(nexthop
->vrf_id
);
1378 if (!table
|| !zvrf
) {
1379 if (IS_ZEBRA_DEBUG_RIB_DETAILED
)
1380 zlog_debug("\t%s: Table not found",
1381 __PRETTY_FUNCTION__
);
1385 rn
= route_node_match(table
, (struct prefix
*)&p
);
1387 route_unlock_node(rn
);
1389 /* Lookup should halt if we've matched against ourselves ('top',
1390 * if specified) - i.e., we cannot have a nexthop NH1 is
1391 * resolved by a route NH1. The exception is if the route is a
1394 if (top
&& rn
== top
)
1395 if (((afi
== AFI_IP
) && (rn
->p
.prefixlen
!= 32))
1396 || ((afi
== AFI_IP6
) && (rn
->p
.prefixlen
!= 128))) {
1397 if (IS_ZEBRA_DEBUG_RIB_DETAILED
)
1399 "\t%s: Matched against ourself and prefix length is not max bit length",
1400 __PRETTY_FUNCTION__
);
1404 /* Pick up selected route. */
1405 /* However, do not resolve over default route unless explicitly
1408 if (is_default_prefix(&rn
->p
)
1409 && !rnh_resolve_via_default(zvrf
, p
.family
)) {
1410 if (IS_ZEBRA_DEBUG_RIB_DETAILED
)
1412 "\t:%s: Resolved against default route",
1413 __PRETTY_FUNCTION__
);
1417 dest
= rib_dest_from_rnode(rn
);
1418 if (dest
&& dest
->selected_fib
1419 && !CHECK_FLAG(dest
->selected_fib
->status
,
1420 ROUTE_ENTRY_REMOVED
)
1421 && dest
->selected_fib
->type
!= ZEBRA_ROUTE_TABLE
)
1422 match
= dest
->selected_fib
;
1424 /* If there is no selected route or matched route is EGP, go up
1430 } while (rn
&& rn
->info
== NULL
);
1432 route_lock_node(rn
);
1437 if (match
->type
== ZEBRA_ROUTE_CONNECT
) {
1438 /* Directly point connected route. */
1439 newhop
= match
->ng
->nexthop
;
1441 if (nexthop
->type
== NEXTHOP_TYPE_IPV4
1442 || nexthop
->type
== NEXTHOP_TYPE_IPV6
)
1443 nexthop
->ifindex
= newhop
->ifindex
;
1446 } else if (CHECK_FLAG(re
->flags
, ZEBRA_FLAG_ALLOW_RECURSION
)) {
1448 for (ALL_NEXTHOPS_PTR(match
->ng
, newhop
)) {
1449 if (!CHECK_FLAG(match
->status
,
1450 ROUTE_ENTRY_INSTALLED
))
1452 if (!nexthop_valid_resolve(nexthop
, newhop
))
1455 SET_FLAG(nexthop
->flags
,
1456 NEXTHOP_FLAG_RECURSIVE
);
1457 nexthop_set_resolved(afi
, newhop
, nexthop
);
1461 re
->nexthop_mtu
= match
->mtu
;
1463 if (!resolved
&& IS_ZEBRA_DEBUG_RIB_DETAILED
)
1464 zlog_debug("\t%s: Recursion failed to find",
1465 __PRETTY_FUNCTION__
);
1467 } else if (re
->type
== ZEBRA_ROUTE_STATIC
) {
1469 for (ALL_NEXTHOPS_PTR(match
->ng
, 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
)
1486 "\t%s: Static route unable to resolve",
1487 __PRETTY_FUNCTION__
);
1490 if (IS_ZEBRA_DEBUG_RIB_DETAILED
) {
1492 "\t%s: Route Type %s has not turned on recursion",
1493 __PRETTY_FUNCTION__
,
1494 zebra_route_string(re
->type
));
1495 if (re
->type
== ZEBRA_ROUTE_BGP
1496 && !CHECK_FLAG(re
->flags
, ZEBRA_FLAG_IBGP
))
1498 "\tEBGP: see \"disable-ebgp-connected-route-check\" or \"disable-connected-check\"");
1503 if (IS_ZEBRA_DEBUG_RIB_DETAILED
)
1504 zlog_debug("\t%s: Nexthop did not lookup in table",
1505 __PRETTY_FUNCTION__
);
1509 /* This function verifies reachability of one given nexthop, which can be
1510 * numbered or unnumbered, IPv4 or IPv6. The result is unconditionally stored
1511 * in nexthop->flags field. The nexthop->ifindex will be updated
1512 * appropriately as well. An existing route map can turn
1513 * (otherwise active) nexthop into inactive, but not vice versa.
1515 * If it finds a nexthop recursivedly, set the resolved_id
1516 * to match that nexthop's nhg_hash_entry ID;
1518 * The return value is the final value of 'ACTIVE' flag.
1520 static unsigned nexthop_active_check(struct route_node
*rn
,
1521 struct route_entry
*re
,
1522 struct nexthop
*nexthop
)
1524 struct interface
*ifp
;
1525 route_map_result_t ret
= RMAP_PERMITMATCH
;
1527 char buf
[SRCDEST2STR_BUFFER
];
1528 const struct prefix
*p
, *src_p
;
1529 struct zebra_vrf
*zvrf
;
1531 srcdest_rnode_prefixes(rn
, &p
, &src_p
);
1533 if (rn
->p
.family
== AF_INET
)
1535 else if (rn
->p
.family
== AF_INET6
)
1539 switch (nexthop
->type
) {
1540 case NEXTHOP_TYPE_IFINDEX
:
1541 ifp
= if_lookup_by_index(nexthop
->ifindex
, nexthop
->vrf_id
);
1542 if (ifp
&& if_is_operative(ifp
))
1543 SET_FLAG(nexthop
->flags
, NEXTHOP_FLAG_ACTIVE
);
1545 UNSET_FLAG(nexthop
->flags
, NEXTHOP_FLAG_ACTIVE
);
1547 case NEXTHOP_TYPE_IPV4
:
1548 case NEXTHOP_TYPE_IPV4_IFINDEX
:
1550 if (nexthop_active(AFI_IP
, re
, nexthop
, rn
))
1551 SET_FLAG(nexthop
->flags
, NEXTHOP_FLAG_ACTIVE
);
1553 UNSET_FLAG(nexthop
->flags
, NEXTHOP_FLAG_ACTIVE
);
1555 case NEXTHOP_TYPE_IPV6
:
1557 if (nexthop_active(AFI_IP6
, re
, nexthop
, rn
))
1558 SET_FLAG(nexthop
->flags
, NEXTHOP_FLAG_ACTIVE
);
1560 UNSET_FLAG(nexthop
->flags
, NEXTHOP_FLAG_ACTIVE
);
1562 case NEXTHOP_TYPE_IPV6_IFINDEX
:
1563 /* RFC 5549, v4 prefix with v6 NH */
1564 if (rn
->p
.family
!= AF_INET
)
1566 if (IN6_IS_ADDR_LINKLOCAL(&nexthop
->gate
.ipv6
)) {
1567 ifp
= if_lookup_by_index(nexthop
->ifindex
,
1569 if (ifp
&& if_is_operative(ifp
))
1570 SET_FLAG(nexthop
->flags
, NEXTHOP_FLAG_ACTIVE
);
1572 UNSET_FLAG(nexthop
->flags
, NEXTHOP_FLAG_ACTIVE
);
1574 if (nexthop_active(AFI_IP6
, re
, nexthop
, rn
))
1575 SET_FLAG(nexthop
->flags
, NEXTHOP_FLAG_ACTIVE
);
1577 UNSET_FLAG(nexthop
->flags
, NEXTHOP_FLAG_ACTIVE
);
1580 case NEXTHOP_TYPE_BLACKHOLE
:
1581 SET_FLAG(nexthop
->flags
, NEXTHOP_FLAG_ACTIVE
);
1586 if (!CHECK_FLAG(nexthop
->flags
, NEXTHOP_FLAG_ACTIVE
)) {
1587 if (IS_ZEBRA_DEBUG_RIB_DETAILED
)
1588 zlog_debug("\t%s: Unable to find a active nexthop",
1589 __PRETTY_FUNCTION__
);
1593 /* XXX: What exactly do those checks do? Do we support
1594 * e.g. IPv4 routes with IPv6 nexthops or vice versa?
1596 if (RIB_SYSTEM_ROUTE(re
) || (family
== AFI_IP
&& p
->family
!= AF_INET
)
1597 || (family
== AFI_IP6
&& p
->family
!= AF_INET6
))
1598 return CHECK_FLAG(nexthop
->flags
, NEXTHOP_FLAG_ACTIVE
);
1600 /* The original code didn't determine the family correctly
1601 * e.g. for NEXTHOP_TYPE_IFINDEX. Retrieve the correct afi
1602 * from the rib_table_info in those cases.
1603 * Possibly it may be better to use only the rib_table_info
1607 rib_table_info_t
*info
;
1609 info
= srcdest_rnode_table_info(rn
);
1613 memset(&nexthop
->rmap_src
.ipv6
, 0, sizeof(union g_addr
));
1615 zvrf
= zebra_vrf_lookup_by_id(nexthop
->vrf_id
);
1617 if (IS_ZEBRA_DEBUG_RIB_DETAILED
)
1618 zlog_debug("\t%s: zvrf is NULL", __PRETTY_FUNCTION__
);
1619 return CHECK_FLAG(nexthop
->flags
, NEXTHOP_FLAG_ACTIVE
);
1622 /* It'll get set if required inside */
1623 ret
= zebra_route_map_check(family
, re
->type
, re
->instance
, p
, nexthop
,
1625 if (ret
== RMAP_DENYMATCH
) {
1626 if (IS_ZEBRA_DEBUG_RIB
) {
1627 srcdest_rnode2str(rn
, buf
, sizeof(buf
));
1629 "%u:%s: Filtering out with NH out %s due to route map",
1631 ifindex2ifname(nexthop
->ifindex
,
1634 UNSET_FLAG(nexthop
->flags
, NEXTHOP_FLAG_ACTIVE
);
1636 return CHECK_FLAG(nexthop
->flags
, NEXTHOP_FLAG_ACTIVE
);
1640 * Iterate over all nexthops of the given RIB entry and refresh their
1641 * ACTIVE flag. If any nexthop is found to toggle the ACTIVE flag,
1642 * the whole re structure is flagged with ROUTE_ENTRY_CHANGED.
1644 * Return value is the new number of active nexthops.
1646 int nexthop_active_update(struct route_node
*rn
, struct route_entry
*re
)
1648 struct nexthop_group new_grp
= {};
1649 struct nexthop
*nexthop
;
1650 union g_addr prev_src
;
1651 unsigned int prev_active
, new_active
;
1652 ifindex_t prev_index
;
1653 uint8_t curr_active
= 0;
1655 afi_t rt_afi
= family2afi(rn
->p
.family
);
1657 UNSET_FLAG(re
->status
, ROUTE_ENTRY_CHANGED
);
1659 /* Copy over the nexthops in current state */
1660 nexthop_group_copy(&new_grp
, re
->ng
);
1662 for (nexthop
= new_grp
.nexthop
; nexthop
; nexthop
= nexthop
->next
) {
1664 /* No protocol daemon provides src and so we're skipping
1666 prev_src
= nexthop
->rmap_src
;
1667 prev_active
= CHECK_FLAG(nexthop
->flags
, NEXTHOP_FLAG_ACTIVE
);
1668 prev_index
= nexthop
->ifindex
;
1670 * We need to respect the multipath_num here
1671 * as that what we should be able to install from
1672 * a multipath perpsective should not be a data plane
1676 nexthop_active_check(rn
, re
, nexthop
);
1679 && nexthop_group_active_nexthop_num(&new_grp
)
1680 >= zrouter
.multipath_num
) {
1681 UNSET_FLAG(nexthop
->flags
, NEXTHOP_FLAG_ACTIVE
);
1688 /* Don't allow src setting on IPv6 addr for now */
1689 if (prev_active
!= new_active
|| prev_index
!= nexthop
->ifindex
1690 || ((nexthop
->type
>= NEXTHOP_TYPE_IFINDEX
1691 && nexthop
->type
< NEXTHOP_TYPE_IPV6
)
1692 && prev_src
.ipv4
.s_addr
1693 != nexthop
->rmap_src
.ipv4
.s_addr
)
1694 || ((nexthop
->type
>= NEXTHOP_TYPE_IPV6
1695 && nexthop
->type
< NEXTHOP_TYPE_BLACKHOLE
)
1696 && !(IPV6_ADDR_SAME(&prev_src
.ipv6
,
1697 &nexthop
->rmap_src
.ipv6
)))
1698 || CHECK_FLAG(re
->status
, ROUTE_ENTRY_LABELS_CHANGED
))
1699 SET_FLAG(re
->status
, ROUTE_ENTRY_CHANGED
);
1702 if (CHECK_FLAG(re
->status
, ROUTE_ENTRY_CHANGED
)) {
1703 struct nhg_hash_entry
*new_nhe
= NULL
;
1705 new_nhe
= zebra_nhg_rib_find(0, &new_grp
, rt_afi
);
1707 zebra_nhg_re_update_ref(re
, new_nhe
);
1711 struct nhg_hash_entry
*nhe
= NULL
;
1713 nhe
= zebra_nhg_lookup_id(re
->nhe_id
);
1716 SET_FLAG(nhe
->flags
, NEXTHOP_GROUP_VALID
);
1719 EC_ZEBRA_TABLE_LOOKUP_FAILED
,
1720 "Active update on NHE id=%u that we do not have in our tables",
1725 * Do not need these nexthops anymore since they
1726 * were either copied over into an nhe or not
1729 nexthops_free(new_grp
.nexthop
);
1733 static void zebra_nhg_re_attach_ref(struct route_entry
*re
,
1734 struct nhg_hash_entry
*new)
1737 re
->nhe_id
= new->id
;
1739 zebra_nhg_increment_ref(new);
1742 int zebra_nhg_re_update_ref(struct route_entry
*re
, struct nhg_hash_entry
*new)
1744 struct nhg_hash_entry
*old
= NULL
;
1752 if (re
->nhe_id
!= new->id
) {
1753 old
= zebra_nhg_lookup_id(re
->nhe_id
);
1755 zebra_nhg_re_attach_ref(re
, new);
1758 zebra_nhg_decrement_ref(old
);
1760 /* This is the first time it's being attached */
1761 zebra_nhg_re_attach_ref(re
, new);
1767 /* Convert a nhe into a group array */
1768 uint8_t zebra_nhg_nhe2grp(struct nh_grp
*grp
, struct nhg_hash_entry
*nhe
,
1771 struct nhg_connected
*rb_node_dep
= NULL
;
1772 struct nhg_hash_entry
*depend
= NULL
;
1775 frr_each(nhg_connected_tree
, &nhe
->nhg_depends
, rb_node_dep
) {
1776 bool duplicate
= false;
1778 depend
= rb_node_dep
->nhe
;
1781 * If its recursive, use its resolved nhe in the group
1783 if (CHECK_FLAG(depend
->flags
, NEXTHOP_GROUP_RECURSIVE
)) {
1784 depend
= zebra_nhg_resolve(depend
);
1787 EC_ZEBRA_NHG_FIB_UPDATE
,
1788 "Failed to recursively resolve Nexthop Hash Entry in the group id=%u",
1794 /* Check for duplicate IDs, kernel doesn't like that */
1795 for (int j
= 0; j
< i
; j
++) {
1796 if (depend
->id
== grp
[j
].id
)
1801 grp
[i
].id
= depend
->id
;
1802 /* We aren't using weights for anything right now */
1815 void zebra_nhg_install_kernel(struct nhg_hash_entry
*nhe
)
1817 struct nhg_connected
*rb_node_dep
= NULL
;
1819 /* Resolve it first */
1820 nhe
= zebra_nhg_resolve(nhe
);
1822 /* Make sure all depends are installed/queued */
1823 frr_each(nhg_connected_tree
, &nhe
->nhg_depends
, rb_node_dep
) {
1824 zebra_nhg_install_kernel(rb_node_dep
->nhe
);
1827 if (!CHECK_FLAG(nhe
->flags
, NEXTHOP_GROUP_INSTALLED
)
1828 && !CHECK_FLAG(nhe
->flags
, NEXTHOP_GROUP_QUEUED
)) {
1829 /* Change its type to us since we are installing it */
1830 nhe
->type
= ZEBRA_ROUTE_NHG
;
1832 int ret
= dplane_nexthop_add(nhe
);
1835 case ZEBRA_DPLANE_REQUEST_QUEUED
:
1836 SET_FLAG(nhe
->flags
, NEXTHOP_GROUP_QUEUED
);
1838 case ZEBRA_DPLANE_REQUEST_FAILURE
:
1840 EC_ZEBRA_DP_INSTALL_FAIL
,
1841 "Failed to install Nexthop ID (%u) into the kernel",
1844 case ZEBRA_DPLANE_REQUEST_SUCCESS
:
1845 SET_FLAG(nhe
->flags
, NEXTHOP_GROUP_INSTALLED
);
1846 zebra_nhg_handle_install(nhe
);
1852 void zebra_nhg_uninstall_kernel(struct nhg_hash_entry
*nhe
)
1854 if (CHECK_FLAG(nhe
->flags
, NEXTHOP_GROUP_INSTALLED
)) {
1855 int ret
= dplane_nexthop_delete(nhe
);
1858 case ZEBRA_DPLANE_REQUEST_QUEUED
:
1859 SET_FLAG(nhe
->flags
, NEXTHOP_GROUP_QUEUED
);
1861 case ZEBRA_DPLANE_REQUEST_FAILURE
:
1863 EC_ZEBRA_DP_DELETE_FAIL
,
1864 "Failed to uninstall Nexthop ID (%u) from the kernel",
1867 case ZEBRA_DPLANE_REQUEST_SUCCESS
:
1868 UNSET_FLAG(nhe
->flags
, NEXTHOP_GROUP_INSTALLED
);
1873 zebra_nhg_handle_uninstall(nhe
);
1876 void zebra_nhg_dplane_result(struct zebra_dplane_ctx
*ctx
)
1878 enum dplane_op_e op
;
1879 enum zebra_dplane_result status
;
1881 struct nhg_hash_entry
*nhe
= NULL
;
1883 op
= dplane_ctx_get_op(ctx
);
1884 status
= dplane_ctx_get_status(ctx
);
1886 id
= dplane_ctx_get_nhe_id(ctx
);
1888 if (IS_ZEBRA_DEBUG_DPLANE_DETAIL
)
1890 "Nexthop dplane ctx %p, op %s, nexthop ID (%u), result %s",
1891 ctx
, dplane_op2str(op
), id
, dplane_res2str(status
));
1894 case DPLANE_OP_NH_DELETE
:
1895 if (status
!= ZEBRA_DPLANE_REQUEST_SUCCESS
)
1897 EC_ZEBRA_DP_DELETE_FAIL
,
1898 "Failed to uninstall Nexthop ID (%u) from the kernel",
1900 /* We already free'd the data, nothing to do */
1902 case DPLANE_OP_NH_INSTALL
:
1903 case DPLANE_OP_NH_UPDATE
:
1904 nhe
= zebra_nhg_lookup_id(id
);
1909 "%s operation preformed on Nexthop ID (%u) in the kernel, that we no longer have in our table",
1910 dplane_op2str(op
), id
);
1914 UNSET_FLAG(nhe
->flags
, NEXTHOP_GROUP_QUEUED
);
1915 if (status
== ZEBRA_DPLANE_REQUEST_SUCCESS
) {
1916 SET_FLAG(nhe
->flags
, NEXTHOP_GROUP_VALID
);
1917 SET_FLAG(nhe
->flags
, NEXTHOP_GROUP_INSTALLED
);
1918 zebra_nhg_handle_install(nhe
);
1921 EC_ZEBRA_DP_INSTALL_FAIL
,
1922 "Failed to install Nexthop ID (%u) into the kernel",
1925 case DPLANE_OP_ROUTE_INSTALL
:
1926 case DPLANE_OP_ROUTE_UPDATE
:
1927 case DPLANE_OP_ROUTE_DELETE
:
1928 case DPLANE_OP_ROUTE_NOTIFY
:
1929 case DPLANE_OP_LSP_INSTALL
:
1930 case DPLANE_OP_LSP_UPDATE
:
1931 case DPLANE_OP_LSP_DELETE
:
1932 case DPLANE_OP_LSP_NOTIFY
:
1933 case DPLANE_OP_PW_INSTALL
:
1934 case DPLANE_OP_PW_UNINSTALL
:
1935 case DPLANE_OP_SYS_ROUTE_ADD
:
1936 case DPLANE_OP_SYS_ROUTE_DELETE
:
1937 case DPLANE_OP_ADDR_INSTALL
:
1938 case DPLANE_OP_ADDR_UNINSTALL
:
1939 case DPLANE_OP_MAC_INSTALL
:
1940 case DPLANE_OP_MAC_DELETE
:
1941 case DPLANE_OP_NEIGH_INSTALL
:
1942 case DPLANE_OP_NEIGH_UPDATE
:
1943 case DPLANE_OP_NEIGH_DELETE
:
1944 case DPLANE_OP_VTEP_ADD
:
1945 case DPLANE_OP_VTEP_DELETE
:
1946 case DPLANE_OP_NONE
:
1950 dplane_ctx_fini(&ctx
);
1953 static void zebra_nhg_sweep_entry(struct hash_bucket
*bucket
, void *arg
)
1955 struct nhg_hash_entry
*nhe
= NULL
;
1957 nhe
= (struct nhg_hash_entry
*)bucket
->data
;
1959 /* If its being ref'd, just let it be uninstalled via a route removal */
1960 if (ZEBRA_NHG_CREATED(nhe
) && nhe
->refcnt
<= 0)
1961 zebra_nhg_uninstall_kernel(nhe
);
1964 void zebra_nhg_sweep_table(struct hash
*hash
)
1966 hash_iterate(hash
, zebra_nhg_sweep_entry
, NULL
);