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 struct nhg_ctx
*nhg_ctx_new()
594 struct nhg_ctx
*new = NULL
;
596 new = XCALLOC(MTYPE_NHG_CTX
, sizeof(struct nhg_ctx
));
601 static void nhg_ctx_free(struct nhg_ctx
*ctx
)
603 XFREE(MTYPE_NHG_CTX
, ctx
);
606 static uint32_t nhg_ctx_get_id(const struct nhg_ctx
*ctx
)
611 static void nhg_ctx_set_status(struct nhg_ctx
*ctx
, enum nhg_ctx_status status
)
613 ctx
->status
= status
;
616 static enum nhg_ctx_status
nhg_ctx_get_status(const struct nhg_ctx
*ctx
)
621 static void nhg_ctx_set_op(struct nhg_ctx
*ctx
, enum nhg_ctx_op_e op
)
626 static enum nhg_ctx_op_e
nhg_ctx_get_op(const struct nhg_ctx
*ctx
)
631 static vrf_id_t
nhg_ctx_get_vrf_id(const struct nhg_ctx
*ctx
)
636 static int nhg_ctx_get_type(const struct nhg_ctx
*ctx
)
641 static int nhg_ctx_get_afi(const struct nhg_ctx
*ctx
)
646 static struct nexthop
*nhg_ctx_get_nh(struct nhg_ctx
*ctx
)
651 static uint8_t nhg_ctx_get_count(const struct nhg_ctx
*ctx
)
656 static struct nh_grp
*nhg_ctx_get_grp(struct nhg_ctx
*ctx
)
661 static struct nhg_ctx
*nhg_ctx_init(uint32_t id
, struct nexthop
*nh
,
662 struct nh_grp
*grp
, vrf_id_t vrf_id
,
663 afi_t afi
, int type
, uint8_t count
)
665 struct nhg_ctx
*ctx
= NULL
;
670 ctx
->vrf_id
= vrf_id
;
676 /* Copy over the array */
677 memcpy(&ctx
->u
.grp
, grp
, count
* sizeof(struct nh_grp
));
684 static bool zebra_nhg_contains_unhashable(struct nhg_hash_entry
*nhe
)
686 struct nhg_connected
*rb_node_dep
= NULL
;
688 frr_each(nhg_connected_tree
, &nhe
->nhg_depends
, rb_node_dep
) {
689 if (CHECK_FLAG(rb_node_dep
->nhe
->flags
,
690 NEXTHOP_GROUP_UNHASHABLE
))
697 static void zebra_nhg_set_unhashable(struct nhg_hash_entry
*nhe
)
699 SET_FLAG(nhe
->flags
, NEXTHOP_GROUP_UNHASHABLE
);
700 SET_FLAG(nhe
->flags
, NEXTHOP_GROUP_INSTALLED
);
703 EC_ZEBRA_DUPLICATE_NHG_MESSAGE
,
704 "Nexthop Group with ID (%d) is a duplicate, therefore unhashable, ignoring",
708 static void zebra_nhg_set_valid(struct nhg_hash_entry
*nhe
)
710 struct nhg_connected
*rb_node_dep
;
712 SET_FLAG(nhe
->flags
, NEXTHOP_GROUP_VALID
);
714 frr_each(nhg_connected_tree
, &nhe
->nhg_dependents
, rb_node_dep
)
715 zebra_nhg_set_valid(rb_node_dep
->nhe
);
718 static void zebra_nhg_set_invalid(struct nhg_hash_entry
*nhe
)
720 struct nhg_connected
*rb_node_dep
;
722 UNSET_FLAG(nhe
->flags
, NEXTHOP_GROUP_VALID
);
724 /* Update validity of nexthops depending on it */
725 frr_each(nhg_connected_tree
, &nhe
->nhg_dependents
, rb_node_dep
)
726 zebra_nhg_check_valid(rb_node_dep
->nhe
);
729 void zebra_nhg_check_valid(struct nhg_hash_entry
*nhe
)
731 struct nhg_connected
*rb_node_dep
= NULL
;
734 /* If anthing else in the group is valid, the group is valid */
735 frr_each(nhg_connected_tree
, &nhe
->nhg_depends
, rb_node_dep
) {
736 if (CHECK_FLAG(rb_node_dep
->nhe
->flags
, NEXTHOP_GROUP_VALID
)) {
744 zebra_nhg_set_valid(nhe
);
746 zebra_nhg_set_invalid(nhe
);
750 static void zebra_nhg_release(struct nhg_hash_entry
*nhe
)
752 /* Remove it from any lists it may be on */
753 zebra_nhg_depends_release(nhe
);
754 zebra_nhg_dependents_release(nhe
);
756 if_nhg_dependents_del(nhe
->ifp
, nhe
);
759 * If its unhashable, we didn't store it here and have to be
760 * sure we don't clear one thats actually being used.
762 if (!CHECK_FLAG(nhe
->flags
, NEXTHOP_GROUP_UNHASHABLE
))
763 hash_release(zrouter
.nhgs
, nhe
);
765 hash_release(zrouter
.nhgs_id
, nhe
);
768 static void zebra_nhg_handle_uninstall(struct nhg_hash_entry
*nhe
)
770 zebra_nhg_release(nhe
);
774 static void zebra_nhg_handle_install(struct nhg_hash_entry
*nhe
)
776 /* Update validity of groups depending on it */
777 struct nhg_connected
*rb_node_dep
;
779 frr_each_safe(nhg_connected_tree
, &nhe
->nhg_dependents
, rb_node_dep
)
780 zebra_nhg_set_valid(rb_node_dep
->nhe
);
784 * The kernel/other program has changed the state of a nexthop object we are
787 static void zebra_nhg_handle_kernel_state_change(struct nhg_hash_entry
*nhe
,
793 "Kernel %s a nexthop group with ID (%u) that we are still using for a route, sending it back down",
794 (is_delete
? "deleted" : "updated"), nhe
->id
);
796 UNSET_FLAG(nhe
->flags
, NEXTHOP_GROUP_INSTALLED
);
797 zebra_nhg_install_kernel(nhe
);
799 zebra_nhg_handle_uninstall(nhe
);
802 static int nhg_ctx_process_new(struct nhg_ctx
*ctx
)
804 struct nexthop_group
*nhg
= NULL
;
805 struct nhg_connected_tree_head nhg_depends
= {};
806 struct nhg_hash_entry
*lookup
= NULL
;
807 struct nhg_hash_entry
*nhe
= NULL
;
809 uint32_t id
= nhg_ctx_get_id(ctx
);
810 uint8_t count
= nhg_ctx_get_count(ctx
);
811 vrf_id_t vrf_id
= nhg_ctx_get_vrf_id(ctx
);
812 int type
= nhg_ctx_get_type(ctx
);
813 afi_t afi
= nhg_ctx_get_afi(ctx
);
815 lookup
= zebra_nhg_lookup_id(id
);
818 /* This is already present in our table, hence an update
819 * that we did not initate.
821 zebra_nhg_handle_kernel_state_change(lookup
, false);
825 if (nhg_ctx_get_count(ctx
)) {
826 nhg
= nexthop_group_new();
827 if (zebra_nhg_process_grp(nhg
, &nhg_depends
,
828 nhg_ctx_get_grp(ctx
), count
)) {
829 depends_decrement_free(&nhg_depends
);
830 nexthop_group_delete(&nhg
);
834 if (!zebra_nhg_find(&nhe
, id
, nhg
, &nhg_depends
, vrf_id
, type
,
836 depends_decrement_free(&nhg_depends
);
838 /* These got copied over in zebra_nhg_alloc() */
839 nexthop_group_delete(&nhg
);
841 nhe
= zebra_nhg_find_nexthop(id
, nhg_ctx_get_nh(ctx
), afi
,
846 struct nhg_hash_entry
*kernel_nhe
= NULL
;
848 /* Duplicate but with different ID from
852 /* The kernel allows duplicate nexthops
853 * as long as they have different IDs.
854 * We are ignoring those to prevent
855 * syncing problems with the kernel
858 * We maintain them *ONLY* in the ID hash table to
859 * track them and set the flag to indicated
860 * their attributes are unhashable.
863 kernel_nhe
= zebra_nhg_copy(nhe
, id
);
864 zebra_nhg_insert_id(kernel_nhe
);
865 zebra_nhg_set_unhashable(kernel_nhe
);
866 } else if (zebra_nhg_contains_unhashable(nhe
)) {
867 /* The group we got contains an unhashable/duplicated
868 * depend, so lets mark this group as unhashable as well
869 * and release it from the non-ID hash.
871 hash_release(zrouter
.nhgs
, nhe
);
872 zebra_nhg_set_unhashable(nhe
);
874 /* It actually created a new nhe */
875 SET_FLAG(nhe
->flags
, NEXTHOP_GROUP_VALID
);
876 SET_FLAG(nhe
->flags
, NEXTHOP_GROUP_INSTALLED
);
880 EC_ZEBRA_TABLE_LOOKUP_FAILED
,
881 "Zebra failed to find or create a nexthop hash entry for ID (%u)",
889 static int nhg_ctx_process_del(struct nhg_ctx
*ctx
)
891 struct nhg_hash_entry
*nhe
= NULL
;
892 uint32_t id
= nhg_ctx_get_id(ctx
);
894 nhe
= zebra_nhg_lookup_id(id
);
898 EC_ZEBRA_BAD_NHG_MESSAGE
,
899 "Kernel delete message received for nexthop group ID (%u) that we do not have in our ID table",
904 zebra_nhg_handle_kernel_state_change(nhe
, true);
909 static void nhg_ctx_process_finish(struct nhg_ctx
*ctx
)
914 * Just freeing for now, maybe do something more in the future
918 if (nhg_ctx_get_count(ctx
))
921 nh
= nhg_ctx_get_nh(ctx
);
923 nexthop_del_labels(nh
);
929 static int queue_add(struct nhg_ctx
*ctx
)
931 /* If its queued or already processed do nothing */
932 if (nhg_ctx_get_status(ctx
) == NHG_CTX_QUEUED
)
935 if (rib_queue_nhg_add(ctx
)) {
936 nhg_ctx_set_status(ctx
, NHG_CTX_FAILURE
);
940 nhg_ctx_set_status(ctx
, NHG_CTX_QUEUED
);
945 int nhg_ctx_process(struct nhg_ctx
*ctx
)
949 switch (nhg_ctx_get_op(ctx
)) {
951 ret
= nhg_ctx_process_new(ctx
);
952 if (nhg_ctx_get_count(ctx
) && ret
== -ENOENT
953 && nhg_ctx_get_status(ctx
) != NHG_CTX_REQUEUED
) {
955 * We have entered a situation where we are
956 * processing a group from the kernel
957 * that has a contained nexthop which
958 * we have not yet processed.
960 * Re-enqueue this ctx to be handled exactly one
961 * more time (indicated by the flag).
963 * By the time we get back to it, we
964 * should have processed its depends.
966 nhg_ctx_set_status(ctx
, NHG_CTX_NONE
);
967 if (queue_add(ctx
) == 0) {
968 nhg_ctx_set_status(ctx
, NHG_CTX_REQUEUED
);
974 ret
= nhg_ctx_process_del(ctx
);
975 case NHG_CTX_OP_NONE
:
979 nhg_ctx_set_status(ctx
, (ret
? NHG_CTX_FAILURE
: NHG_CTX_SUCCESS
));
981 nhg_ctx_process_finish(ctx
);
986 /* Kernel-side, you either get a single new nexthop or a array of ID's */
987 int zebra_nhg_kernel_find(uint32_t id
, struct nexthop
*nh
, struct nh_grp
*grp
,
988 uint8_t count
, vrf_id_t vrf_id
, afi_t afi
, int type
,
991 struct nhg_ctx
*ctx
= NULL
;
994 /* Increase our counter so we don't try to create
995 * an ID that already exists
999 ctx
= nhg_ctx_init(id
, nh
, grp
, vrf_id
, afi
, type
, count
);
1000 nhg_ctx_set_op(ctx
, NHG_CTX_OP_NEW
);
1002 /* Under statup conditions, we need to handle them immediately
1003 * like we do for routes. Otherwise, we are going to get a route
1004 * with a nhe_id that we have not handled.
1007 return nhg_ctx_process(ctx
);
1009 if (queue_add(ctx
)) {
1010 nhg_ctx_process_finish(ctx
);
1017 /* Kernel-side, received delete message */
1018 int zebra_nhg_kernel_del(uint32_t id
)
1020 struct nhg_ctx
*ctx
= NULL
;
1022 ctx
= nhg_ctx_init(id
, NULL
, NULL
, 0, 0, 0, 0);
1024 nhg_ctx_set_op(ctx
, NHG_CTX_OP_DEL
);
1026 if (queue_add(ctx
)) {
1027 nhg_ctx_process_finish(ctx
);
1034 /* Some dependency helper functions */
1035 static struct nhg_hash_entry
*depends_find(struct nexthop
*nh
, afi_t afi
)
1037 struct nexthop
*lookup
= NULL
;
1038 struct nhg_hash_entry
*nhe
= NULL
;
1040 copy_nexthops(&lookup
, nh
, NULL
);
1042 /* Clear it, in case its a group */
1043 nexthops_free(lookup
->next
);
1044 nexthops_free(lookup
->prev
);
1045 lookup
->next
= NULL
;
1046 lookup
->prev
= NULL
;
1048 nhe
= zebra_nhg_find_nexthop(0, lookup
, afi
, 0);
1050 nexthops_free(lookup
);
1055 static void depends_add(struct nhg_connected_tree_head
*head
,
1056 struct nhg_hash_entry
*depend
)
1058 nhg_connected_tree_add_nhe(head
, depend
);
1059 zebra_nhg_increment_ref(depend
);
1062 static struct nhg_hash_entry
*
1063 depends_find_add(struct nhg_connected_tree_head
*head
, struct nexthop
*nh
,
1066 struct nhg_hash_entry
*depend
= NULL
;
1068 depend
= depends_find(nh
, afi
);
1071 depends_add(head
, depend
);
1076 static struct nhg_hash_entry
*
1077 depends_find_id_add(struct nhg_connected_tree_head
*head
, uint32_t id
)
1079 struct nhg_hash_entry
*depend
= NULL
;
1081 depend
= zebra_nhg_lookup_id(id
);
1084 depends_add(head
, depend
);
1089 static void depends_decrement_free(struct nhg_connected_tree_head
*head
)
1091 nhg_connected_tree_decrement_ref(head
);
1092 nhg_connected_tree_free(head
);
1095 /* Rib-side, you get a nexthop group struct */
1096 struct nhg_hash_entry
*
1097 zebra_nhg_rib_find(uint32_t id
, struct nexthop_group
*nhg
, afi_t rt_afi
)
1099 struct nhg_hash_entry
*nhe
= NULL
;
1101 if (!(nhg
&& nhg
->nexthop
)) {
1102 flog_err(EC_ZEBRA_TABLE_LOOKUP_FAILED
,
1103 "No nexthop passed to %s", __func__
);
1107 zebra_nhg_find(&nhe
, id
, nhg
, NULL
, nhg
->nexthop
->vrf_id
, rt_afi
, 0);
1112 static void zebra_nhg_free_members(struct nhg_hash_entry
*nhe
)
1114 nexthop_group_delete(&nhe
->nhg
);
1115 /* Decrement to remove connection ref */
1116 nhg_connected_tree_decrement_ref(&nhe
->nhg_depends
);
1117 nhg_connected_tree_free(&nhe
->nhg_depends
);
1118 nhg_connected_tree_free(&nhe
->nhg_dependents
);
1121 void zebra_nhg_free(void *arg
)
1123 struct nhg_hash_entry
*nhe
= NULL
;
1125 nhe
= (struct nhg_hash_entry
*)arg
;
1128 zlog_debug("nhe_id=%u hash refcnt=%d", nhe
->id
, nhe
->refcnt
);
1130 zebra_nhg_free_members(nhe
);
1132 XFREE(MTYPE_NHG
, nhe
);
1135 void zebra_nhg_decrement_ref(struct nhg_hash_entry
*nhe
)
1139 if (!zebra_nhg_depends_is_empty(nhe
))
1140 nhg_connected_tree_decrement_ref(&nhe
->nhg_depends
);
1142 if (ZEBRA_NHG_CREATED(nhe
) && nhe
->refcnt
<= 0)
1143 zebra_nhg_uninstall_kernel(nhe
);
1146 void zebra_nhg_increment_ref(struct nhg_hash_entry
*nhe
)
1150 if (!zebra_nhg_depends_is_empty(nhe
))
1151 nhg_connected_tree_increment_ref(&nhe
->nhg_depends
);
1154 static void nexthop_set_resolved(afi_t afi
, const struct nexthop
*newhop
,
1155 struct nexthop
*nexthop
)
1157 struct nexthop
*resolved_hop
;
1158 uint8_t num_labels
= 0;
1159 mpls_label_t labels
[MPLS_MAX_LABELS
];
1160 enum lsp_types_t label_type
= ZEBRA_LSP_NONE
;
1163 resolved_hop
= nexthop_new();
1164 SET_FLAG(resolved_hop
->flags
, NEXTHOP_FLAG_ACTIVE
);
1166 resolved_hop
->vrf_id
= nexthop
->vrf_id
;
1167 switch (newhop
->type
) {
1168 case NEXTHOP_TYPE_IPV4
:
1169 case NEXTHOP_TYPE_IPV4_IFINDEX
:
1170 /* If the resolving route specifies a gateway, use it */
1171 resolved_hop
->type
= newhop
->type
;
1172 resolved_hop
->gate
.ipv4
= newhop
->gate
.ipv4
;
1174 if (newhop
->ifindex
) {
1175 resolved_hop
->type
= NEXTHOP_TYPE_IPV4_IFINDEX
;
1176 resolved_hop
->ifindex
= newhop
->ifindex
;
1179 case NEXTHOP_TYPE_IPV6
:
1180 case NEXTHOP_TYPE_IPV6_IFINDEX
:
1181 resolved_hop
->type
= newhop
->type
;
1182 resolved_hop
->gate
.ipv6
= newhop
->gate
.ipv6
;
1184 if (newhop
->ifindex
) {
1185 resolved_hop
->type
= NEXTHOP_TYPE_IPV6_IFINDEX
;
1186 resolved_hop
->ifindex
= newhop
->ifindex
;
1189 case NEXTHOP_TYPE_IFINDEX
:
1190 /* If the resolving route is an interface route,
1191 * it means the gateway we are looking up is connected
1192 * to that interface. (The actual network is _not_ onlink).
1193 * Therefore, the resolved route should have the original
1194 * gateway as nexthop as it is directly connected.
1196 * On Linux, we have to set the onlink netlink flag because
1197 * otherwise, the kernel won't accept the route.
1199 resolved_hop
->flags
|= NEXTHOP_FLAG_ONLINK
;
1200 if (afi
== AFI_IP
) {
1201 resolved_hop
->type
= NEXTHOP_TYPE_IPV4_IFINDEX
;
1202 resolved_hop
->gate
.ipv4
= nexthop
->gate
.ipv4
;
1203 } else if (afi
== AFI_IP6
) {
1204 resolved_hop
->type
= NEXTHOP_TYPE_IPV6_IFINDEX
;
1205 resolved_hop
->gate
.ipv6
= nexthop
->gate
.ipv6
;
1207 resolved_hop
->ifindex
= newhop
->ifindex
;
1209 case NEXTHOP_TYPE_BLACKHOLE
:
1210 resolved_hop
->type
= NEXTHOP_TYPE_BLACKHOLE
;
1211 resolved_hop
->bh_type
= newhop
->bh_type
;
1215 if (newhop
->flags
& NEXTHOP_FLAG_ONLINK
)
1216 resolved_hop
->flags
|= NEXTHOP_FLAG_ONLINK
;
1218 /* Copy labels of the resolved route and the parent resolving to it */
1219 if (newhop
->nh_label
) {
1220 for (i
= 0; i
< newhop
->nh_label
->num_labels
; i
++)
1221 labels
[num_labels
++] = newhop
->nh_label
->label
[i
];
1222 label_type
= newhop
->nh_label_type
;
1225 if (nexthop
->nh_label
) {
1226 for (i
= 0; i
< nexthop
->nh_label
->num_labels
; i
++)
1227 labels
[num_labels
++] = nexthop
->nh_label
->label
[i
];
1229 /* If the parent has labels, use its type */
1230 label_type
= nexthop
->nh_label_type
;
1234 nexthop_add_labels(resolved_hop
, label_type
, num_labels
,
1237 resolved_hop
->rparent
= nexthop
;
1238 _nexthop_add(&nexthop
->resolved
, resolved_hop
);
1241 /* Checks if nexthop we are trying to resolve to is valid */
1242 static bool nexthop_valid_resolve(const struct nexthop
*nexthop
,
1243 const struct nexthop
*resolved
)
1245 /* Can't resolve to a recursive nexthop */
1246 if (CHECK_FLAG(resolved
->flags
, NEXTHOP_FLAG_RECURSIVE
))
1249 switch (nexthop
->type
) {
1250 case NEXTHOP_TYPE_IPV4_IFINDEX
:
1251 case NEXTHOP_TYPE_IPV6_IFINDEX
:
1252 /* If the nexthop we are resolving to does not match the
1253 * ifindex for the nexthop the route wanted, its not valid.
1255 if (nexthop
->ifindex
!= resolved
->ifindex
)
1258 case NEXTHOP_TYPE_IPV4
:
1259 case NEXTHOP_TYPE_IPV6
:
1260 case NEXTHOP_TYPE_IFINDEX
:
1261 case NEXTHOP_TYPE_BLACKHOLE
:
1269 * Given a nexthop we need to properly recursively resolve
1270 * the route. As such, do a table lookup to find and match
1271 * if at all possible. Set the nexthop->ifindex and resolved_id
1274 static int nexthop_active(afi_t afi
, struct route_entry
*re
,
1275 struct nexthop
*nexthop
, struct route_node
*top
)
1278 struct route_table
*table
;
1279 struct route_node
*rn
;
1280 struct route_entry
*match
= NULL
;
1282 struct nexthop
*newhop
;
1283 struct interface
*ifp
;
1285 struct zebra_vrf
*zvrf
;
1287 if ((nexthop
->type
== NEXTHOP_TYPE_IPV4
)
1288 || nexthop
->type
== NEXTHOP_TYPE_IPV6
)
1289 nexthop
->ifindex
= 0;
1292 UNSET_FLAG(nexthop
->flags
, NEXTHOP_FLAG_RECURSIVE
);
1293 nexthops_free(nexthop
->resolved
);
1294 nexthop
->resolved
= NULL
;
1295 re
->nexthop_mtu
= 0;
1298 * If the kernel has sent us a NEW route, then
1299 * by golly gee whiz it's a good route.
1301 * If its an already INSTALLED route we have already handled, then the
1302 * kernel route's nexthop might have became unreachable
1303 * and we have to handle that.
1305 if (!CHECK_FLAG(re
->status
, ROUTE_ENTRY_INSTALLED
)
1306 && (re
->type
== ZEBRA_ROUTE_KERNEL
1307 || re
->type
== ZEBRA_ROUTE_SYSTEM
))
1311 * Check to see if we should trust the passed in information
1312 * for UNNUMBERED interfaces as that we won't find the GW
1313 * address in the routing table.
1314 * This check should suffice to handle IPv4 or IPv6 routes
1315 * sourced from EVPN routes which are installed with the
1316 * next hop as the remote VTEP IP.
1318 if (CHECK_FLAG(nexthop
->flags
, NEXTHOP_FLAG_ONLINK
)) {
1319 ifp
= if_lookup_by_index(nexthop
->ifindex
, nexthop
->vrf_id
);
1321 if (IS_ZEBRA_DEBUG_RIB_DETAILED
)
1323 "\t%s: Onlink and interface: %u[%u] does not exist",
1324 __PRETTY_FUNCTION__
, nexthop
->ifindex
,
1328 if (connected_is_unnumbered(ifp
)) {
1329 if (if_is_operative(ifp
))
1332 if (IS_ZEBRA_DEBUG_RIB_DETAILED
)
1334 "\t%s: Onlink and interface %s is not operative",
1335 __PRETTY_FUNCTION__
, ifp
->name
);
1338 if (!if_is_operative(ifp
)) {
1339 if (IS_ZEBRA_DEBUG_RIB_DETAILED
)
1341 "\t%s: Interface %s is not unnumbered",
1342 __PRETTY_FUNCTION__
, ifp
->name
);
1347 /* Make lookup prefix. */
1348 memset(&p
, 0, sizeof(struct prefix
));
1352 p
.prefixlen
= IPV4_MAX_PREFIXLEN
;
1353 p
.u
.prefix4
= nexthop
->gate
.ipv4
;
1356 p
.family
= AF_INET6
;
1357 p
.prefixlen
= IPV6_MAX_PREFIXLEN
;
1358 p
.u
.prefix6
= nexthop
->gate
.ipv6
;
1361 assert(afi
!= AFI_IP
&& afi
!= AFI_IP6
);
1365 table
= zebra_vrf_table(afi
, SAFI_UNICAST
, nexthop
->vrf_id
);
1367 zvrf
= zebra_vrf_lookup_by_id(nexthop
->vrf_id
);
1368 if (!table
|| !zvrf
) {
1369 if (IS_ZEBRA_DEBUG_RIB_DETAILED
)
1370 zlog_debug("\t%s: Table not found",
1371 __PRETTY_FUNCTION__
);
1375 rn
= route_node_match(table
, (struct prefix
*)&p
);
1377 route_unlock_node(rn
);
1379 /* Lookup should halt if we've matched against ourselves ('top',
1380 * if specified) - i.e., we cannot have a nexthop NH1 is
1381 * resolved by a route NH1. The exception is if the route is a
1384 if (top
&& rn
== top
)
1385 if (((afi
== AFI_IP
) && (rn
->p
.prefixlen
!= 32))
1386 || ((afi
== AFI_IP6
) && (rn
->p
.prefixlen
!= 128))) {
1387 if (IS_ZEBRA_DEBUG_RIB_DETAILED
)
1389 "\t%s: Matched against ourself and prefix length is not max bit length",
1390 __PRETTY_FUNCTION__
);
1394 /* Pick up selected route. */
1395 /* However, do not resolve over default route unless explicitly
1398 if (is_default_prefix(&rn
->p
)
1399 && !rnh_resolve_via_default(zvrf
, p
.family
)) {
1400 if (IS_ZEBRA_DEBUG_RIB_DETAILED
)
1402 "\t:%s: Resolved against default route",
1403 __PRETTY_FUNCTION__
);
1407 dest
= rib_dest_from_rnode(rn
);
1408 if (dest
&& dest
->selected_fib
1409 && !CHECK_FLAG(dest
->selected_fib
->status
,
1410 ROUTE_ENTRY_REMOVED
)
1411 && dest
->selected_fib
->type
!= ZEBRA_ROUTE_TABLE
)
1412 match
= dest
->selected_fib
;
1414 /* If there is no selected route or matched route is EGP, go up
1420 } while (rn
&& rn
->info
== NULL
);
1422 route_lock_node(rn
);
1427 if (match
->type
== ZEBRA_ROUTE_CONNECT
) {
1428 /* Directly point connected route. */
1429 newhop
= match
->ng
->nexthop
;
1431 if (nexthop
->type
== NEXTHOP_TYPE_IPV4
1432 || nexthop
->type
== NEXTHOP_TYPE_IPV6
)
1433 nexthop
->ifindex
= newhop
->ifindex
;
1436 } else if (CHECK_FLAG(re
->flags
, ZEBRA_FLAG_ALLOW_RECURSION
)) {
1438 for (ALL_NEXTHOPS_PTR(match
->ng
, newhop
)) {
1439 if (!CHECK_FLAG(match
->status
,
1440 ROUTE_ENTRY_INSTALLED
))
1442 if (!nexthop_valid_resolve(nexthop
, newhop
))
1445 SET_FLAG(nexthop
->flags
,
1446 NEXTHOP_FLAG_RECURSIVE
);
1447 nexthop_set_resolved(afi
, newhop
, nexthop
);
1451 re
->nexthop_mtu
= match
->mtu
;
1453 if (!resolved
&& IS_ZEBRA_DEBUG_RIB_DETAILED
)
1454 zlog_debug("\t%s: Recursion failed to find",
1455 __PRETTY_FUNCTION__
);
1457 } else if (re
->type
== ZEBRA_ROUTE_STATIC
) {
1459 for (ALL_NEXTHOPS_PTR(match
->ng
, newhop
)) {
1460 if (!CHECK_FLAG(match
->status
,
1461 ROUTE_ENTRY_INSTALLED
))
1463 if (!nexthop_valid_resolve(nexthop
, newhop
))
1466 SET_FLAG(nexthop
->flags
,
1467 NEXTHOP_FLAG_RECURSIVE
);
1468 nexthop_set_resolved(afi
, newhop
, nexthop
);
1472 re
->nexthop_mtu
= match
->mtu
;
1474 if (!resolved
&& IS_ZEBRA_DEBUG_RIB_DETAILED
)
1476 "\t%s: Static route unable to resolve",
1477 __PRETTY_FUNCTION__
);
1480 if (IS_ZEBRA_DEBUG_RIB_DETAILED
) {
1482 "\t%s: Route Type %s has not turned on recursion",
1483 __PRETTY_FUNCTION__
,
1484 zebra_route_string(re
->type
));
1485 if (re
->type
== ZEBRA_ROUTE_BGP
1486 && !CHECK_FLAG(re
->flags
, ZEBRA_FLAG_IBGP
))
1488 "\tEBGP: see \"disable-ebgp-connected-route-check\" or \"disable-connected-check\"");
1493 if (IS_ZEBRA_DEBUG_RIB_DETAILED
)
1494 zlog_debug("\t%s: Nexthop did not lookup in table",
1495 __PRETTY_FUNCTION__
);
1499 /* This function verifies reachability of one given nexthop, which can be
1500 * numbered or unnumbered, IPv4 or IPv6. The result is unconditionally stored
1501 * in nexthop->flags field. The nexthop->ifindex will be updated
1502 * appropriately as well. An existing route map can turn
1503 * (otherwise active) nexthop into inactive, but not vice versa.
1505 * If it finds a nexthop recursivedly, set the resolved_id
1506 * to match that nexthop's nhg_hash_entry ID;
1508 * The return value is the final value of 'ACTIVE' flag.
1510 static unsigned nexthop_active_check(struct route_node
*rn
,
1511 struct route_entry
*re
,
1512 struct nexthop
*nexthop
)
1514 struct interface
*ifp
;
1515 route_map_result_t ret
= RMAP_PERMITMATCH
;
1517 char buf
[SRCDEST2STR_BUFFER
];
1518 const struct prefix
*p
, *src_p
;
1519 struct zebra_vrf
*zvrf
;
1521 srcdest_rnode_prefixes(rn
, &p
, &src_p
);
1523 if (rn
->p
.family
== AF_INET
)
1525 else if (rn
->p
.family
== AF_INET6
)
1529 switch (nexthop
->type
) {
1530 case NEXTHOP_TYPE_IFINDEX
:
1531 ifp
= if_lookup_by_index(nexthop
->ifindex
, nexthop
->vrf_id
);
1532 if (ifp
&& if_is_operative(ifp
))
1533 SET_FLAG(nexthop
->flags
, NEXTHOP_FLAG_ACTIVE
);
1535 UNSET_FLAG(nexthop
->flags
, NEXTHOP_FLAG_ACTIVE
);
1537 case NEXTHOP_TYPE_IPV4
:
1538 case NEXTHOP_TYPE_IPV4_IFINDEX
:
1540 if (nexthop_active(AFI_IP
, re
, nexthop
, rn
))
1541 SET_FLAG(nexthop
->flags
, NEXTHOP_FLAG_ACTIVE
);
1543 UNSET_FLAG(nexthop
->flags
, NEXTHOP_FLAG_ACTIVE
);
1545 case NEXTHOP_TYPE_IPV6
:
1547 if (nexthop_active(AFI_IP6
, re
, nexthop
, rn
))
1548 SET_FLAG(nexthop
->flags
, NEXTHOP_FLAG_ACTIVE
);
1550 UNSET_FLAG(nexthop
->flags
, NEXTHOP_FLAG_ACTIVE
);
1552 case NEXTHOP_TYPE_IPV6_IFINDEX
:
1553 /* RFC 5549, v4 prefix with v6 NH */
1554 if (rn
->p
.family
!= AF_INET
)
1556 if (IN6_IS_ADDR_LINKLOCAL(&nexthop
->gate
.ipv6
)) {
1557 ifp
= if_lookup_by_index(nexthop
->ifindex
,
1559 if (ifp
&& if_is_operative(ifp
))
1560 SET_FLAG(nexthop
->flags
, NEXTHOP_FLAG_ACTIVE
);
1562 UNSET_FLAG(nexthop
->flags
, NEXTHOP_FLAG_ACTIVE
);
1564 if (nexthop_active(AFI_IP6
, re
, nexthop
, rn
))
1565 SET_FLAG(nexthop
->flags
, NEXTHOP_FLAG_ACTIVE
);
1567 UNSET_FLAG(nexthop
->flags
, NEXTHOP_FLAG_ACTIVE
);
1570 case NEXTHOP_TYPE_BLACKHOLE
:
1571 SET_FLAG(nexthop
->flags
, NEXTHOP_FLAG_ACTIVE
);
1576 if (!CHECK_FLAG(nexthop
->flags
, NEXTHOP_FLAG_ACTIVE
)) {
1577 if (IS_ZEBRA_DEBUG_RIB_DETAILED
)
1578 zlog_debug("\t%s: Unable to find a active nexthop",
1579 __PRETTY_FUNCTION__
);
1583 /* XXX: What exactly do those checks do? Do we support
1584 * e.g. IPv4 routes with IPv6 nexthops or vice versa?
1586 if (RIB_SYSTEM_ROUTE(re
) || (family
== AFI_IP
&& p
->family
!= AF_INET
)
1587 || (family
== AFI_IP6
&& p
->family
!= AF_INET6
))
1588 return CHECK_FLAG(nexthop
->flags
, NEXTHOP_FLAG_ACTIVE
);
1590 /* The original code didn't determine the family correctly
1591 * e.g. for NEXTHOP_TYPE_IFINDEX. Retrieve the correct afi
1592 * from the rib_table_info in those cases.
1593 * Possibly it may be better to use only the rib_table_info
1597 rib_table_info_t
*info
;
1599 info
= srcdest_rnode_table_info(rn
);
1603 memset(&nexthop
->rmap_src
.ipv6
, 0, sizeof(union g_addr
));
1605 zvrf
= zebra_vrf_lookup_by_id(nexthop
->vrf_id
);
1607 if (IS_ZEBRA_DEBUG_RIB_DETAILED
)
1608 zlog_debug("\t%s: zvrf is NULL", __PRETTY_FUNCTION__
);
1609 return CHECK_FLAG(nexthop
->flags
, NEXTHOP_FLAG_ACTIVE
);
1612 /* It'll get set if required inside */
1613 ret
= zebra_route_map_check(family
, re
->type
, re
->instance
, p
, nexthop
,
1615 if (ret
== RMAP_DENYMATCH
) {
1616 if (IS_ZEBRA_DEBUG_RIB
) {
1617 srcdest_rnode2str(rn
, buf
, sizeof(buf
));
1619 "%u:%s: Filtering out with NH out %s due to route map",
1621 ifindex2ifname(nexthop
->ifindex
,
1624 UNSET_FLAG(nexthop
->flags
, NEXTHOP_FLAG_ACTIVE
);
1626 return CHECK_FLAG(nexthop
->flags
, NEXTHOP_FLAG_ACTIVE
);
1630 * Iterate over all nexthops of the given RIB entry and refresh their
1631 * ACTIVE flag. If any nexthop is found to toggle the ACTIVE flag,
1632 * the whole re structure is flagged with ROUTE_ENTRY_CHANGED.
1634 * Return value is the new number of active nexthops.
1636 int nexthop_active_update(struct route_node
*rn
, struct route_entry
*re
)
1638 struct nexthop_group new_grp
= {};
1639 struct nexthop
*nexthop
;
1640 union g_addr prev_src
;
1641 unsigned int prev_active
, new_active
;
1642 ifindex_t prev_index
;
1643 uint8_t curr_active
= 0;
1645 afi_t rt_afi
= family2afi(rn
->p
.family
);
1647 UNSET_FLAG(re
->status
, ROUTE_ENTRY_CHANGED
);
1649 /* Copy over the nexthops in current state */
1650 nexthop_group_copy(&new_grp
, re
->ng
);
1652 for (nexthop
= new_grp
.nexthop
; nexthop
; nexthop
= nexthop
->next
) {
1654 /* No protocol daemon provides src and so we're skipping
1656 prev_src
= nexthop
->rmap_src
;
1657 prev_active
= CHECK_FLAG(nexthop
->flags
, NEXTHOP_FLAG_ACTIVE
);
1658 prev_index
= nexthop
->ifindex
;
1660 * We need to respect the multipath_num here
1661 * as that what we should be able to install from
1662 * a multipath perpsective should not be a data plane
1666 nexthop_active_check(rn
, re
, nexthop
);
1669 && nexthop_group_active_nexthop_num(&new_grp
)
1670 >= zrouter
.multipath_num
) {
1671 UNSET_FLAG(nexthop
->flags
, NEXTHOP_FLAG_ACTIVE
);
1678 /* Don't allow src setting on IPv6 addr for now */
1679 if (prev_active
!= new_active
|| prev_index
!= nexthop
->ifindex
1680 || ((nexthop
->type
>= NEXTHOP_TYPE_IFINDEX
1681 && nexthop
->type
< NEXTHOP_TYPE_IPV6
)
1682 && prev_src
.ipv4
.s_addr
1683 != nexthop
->rmap_src
.ipv4
.s_addr
)
1684 || ((nexthop
->type
>= NEXTHOP_TYPE_IPV6
1685 && nexthop
->type
< NEXTHOP_TYPE_BLACKHOLE
)
1686 && !(IPV6_ADDR_SAME(&prev_src
.ipv6
,
1687 &nexthop
->rmap_src
.ipv6
)))
1688 || CHECK_FLAG(re
->status
, ROUTE_ENTRY_LABELS_CHANGED
))
1689 SET_FLAG(re
->status
, ROUTE_ENTRY_CHANGED
);
1692 if (CHECK_FLAG(re
->status
, ROUTE_ENTRY_CHANGED
)) {
1693 struct nhg_hash_entry
*new_nhe
= NULL
;
1695 new_nhe
= zebra_nhg_rib_find(0, &new_grp
, rt_afi
);
1697 zebra_nhg_re_update_ref(re
, new_nhe
);
1701 struct nhg_hash_entry
*nhe
= NULL
;
1703 nhe
= zebra_nhg_lookup_id(re
->nhe_id
);
1706 SET_FLAG(nhe
->flags
, NEXTHOP_GROUP_VALID
);
1709 EC_ZEBRA_TABLE_LOOKUP_FAILED
,
1710 "Active update on NHE id=%u that we do not have in our tables",
1715 * Do not need these nexthops anymore since they
1716 * were either copied over into an nhe or not
1719 nexthops_free(new_grp
.nexthop
);
1723 static void zebra_nhg_re_attach_ref(struct route_entry
*re
,
1724 struct nhg_hash_entry
*new)
1727 re
->nhe_id
= new->id
;
1729 zebra_nhg_increment_ref(new);
1732 int zebra_nhg_re_update_ref(struct route_entry
*re
, struct nhg_hash_entry
*new)
1734 struct nhg_hash_entry
*old
= NULL
;
1742 if (re
->nhe_id
!= new->id
) {
1743 old
= zebra_nhg_lookup_id(re
->nhe_id
);
1745 zebra_nhg_re_attach_ref(re
, new);
1748 zebra_nhg_decrement_ref(old
);
1750 /* This is the first time it's being attached */
1751 zebra_nhg_re_attach_ref(re
, new);
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 */
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
);