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"
31 #include "lib/lib_errors.h"
33 #include "zebra/connected.h"
34 #include "zebra/debug.h"
35 #include "zebra/zebra_router.h"
36 #include "zebra/zebra_nhg_private.h"
37 #include "zebra/zebra_rnh.h"
38 #include "zebra/zebra_routemap.h"
39 #include "zebra/zebra_srte.h"
40 #include "zebra/zserv.h"
42 #include "zebra_errors.h"
43 #include "zebra_dplane.h"
44 #include "zebra/interface.h"
45 #include "zebra/zapi_msg.h"
46 #include "zebra/rib.h"
48 DEFINE_MTYPE_STATIC(ZEBRA
, NHG
, "Nexthop Group Entry");
49 DEFINE_MTYPE_STATIC(ZEBRA
, NHG_CONNECTED
, "Nexthop Group Connected");
50 DEFINE_MTYPE_STATIC(ZEBRA
, NHG_CTX
, "Nexthop Group Context");
52 /* Map backup nexthop indices between two nhes */
53 struct backup_nh_map_s
{
62 /* id counter to keep in sync with kernel */
65 /* Controlled through ui */
66 static bool g_nexthops_enabled
= true;
67 static bool proto_nexthops_only
;
68 static bool use_recursive_backups
= true;
70 static struct nhg_hash_entry
*depends_find(const struct nexthop
*nh
, afi_t afi
,
71 int type
, bool from_dplane
);
72 static void depends_add(struct nhg_connected_tree_head
*head
,
73 struct nhg_hash_entry
*depend
);
74 static struct nhg_hash_entry
*
75 depends_find_add(struct nhg_connected_tree_head
*head
, struct nexthop
*nh
,
76 afi_t afi
, int type
, bool from_dplane
);
77 static struct nhg_hash_entry
*
78 depends_find_id_add(struct nhg_connected_tree_head
*head
, uint32_t id
);
79 static void depends_decrement_free(struct nhg_connected_tree_head
*head
);
81 static struct nhg_backup_info
*
82 nhg_backup_copy(const struct nhg_backup_info
*orig
);
84 /* Helper function for getting the next allocatable ID */
85 static uint32_t nhg_get_next_id(void)
90 if (IS_ZEBRA_DEBUG_NHG_DETAIL
)
91 zlog_debug("%s: ID %u checking", __func__
, id_counter
);
93 if (id_counter
== ZEBRA_NHG_PROTO_LOWER
) {
94 if (IS_ZEBRA_DEBUG_NHG_DETAIL
)
95 zlog_debug("%s: ID counter wrapped", __func__
);
101 if (zebra_nhg_lookup_id(id_counter
)) {
102 if (IS_ZEBRA_DEBUG_NHG_DETAIL
)
103 zlog_debug("%s: ID already exists", __func__
);
114 static void nhg_connected_free(struct nhg_connected
*dep
)
116 XFREE(MTYPE_NHG_CONNECTED
, dep
);
119 static struct nhg_connected
*nhg_connected_new(struct nhg_hash_entry
*nhe
)
121 struct nhg_connected
*new = NULL
;
123 new = XCALLOC(MTYPE_NHG_CONNECTED
, sizeof(struct nhg_connected
));
129 void nhg_connected_tree_free(struct nhg_connected_tree_head
*head
)
131 struct nhg_connected
*rb_node_dep
= NULL
;
133 if (!nhg_connected_tree_is_empty(head
)) {
134 frr_each_safe(nhg_connected_tree
, head
, rb_node_dep
) {
135 nhg_connected_tree_del(head
, rb_node_dep
);
136 nhg_connected_free(rb_node_dep
);
141 bool nhg_connected_tree_is_empty(const struct nhg_connected_tree_head
*head
)
143 return nhg_connected_tree_count(head
) ? false : true;
146 struct nhg_connected
*
147 nhg_connected_tree_root(struct nhg_connected_tree_head
*head
)
149 return nhg_connected_tree_first(head
);
152 struct nhg_hash_entry
*
153 nhg_connected_tree_del_nhe(struct nhg_connected_tree_head
*head
,
154 struct nhg_hash_entry
*depend
)
156 struct nhg_connected lookup
= {};
157 struct nhg_connected
*remove
= NULL
;
158 struct nhg_hash_entry
*removed_nhe
;
162 /* Lookup to find the element, then remove it */
163 remove
= nhg_connected_tree_find(head
, &lookup
);
165 /* Re-returning here just in case this API changes..
166 * the _del list api's are a bit undefined at the moment.
168 * So hopefully returning here will make it fail if the api
169 * changes to something different than currently expected.
171 remove
= nhg_connected_tree_del(head
, remove
);
173 /* If the entry was sucessfully removed, free the 'connected` struct */
175 removed_nhe
= remove
->nhe
;
176 nhg_connected_free(remove
);
183 /* Assuming UNIQUE RB tree. If this changes, assumptions here about
184 * insertion need to change.
186 struct nhg_hash_entry
*
187 nhg_connected_tree_add_nhe(struct nhg_connected_tree_head
*head
,
188 struct nhg_hash_entry
*depend
)
190 struct nhg_connected
*new = NULL
;
192 new = nhg_connected_new(depend
);
194 /* On success, NULL will be returned from the
197 if (new && (nhg_connected_tree_add(head
, new) == NULL
))
200 /* If it wasn't successful, it must be a duplicate. We enforce the
201 * unique property for the `nhg_connected` tree.
203 nhg_connected_free(new);
209 nhg_connected_tree_decrement_ref(struct nhg_connected_tree_head
*head
)
211 struct nhg_connected
*rb_node_dep
= NULL
;
213 frr_each_safe(nhg_connected_tree
, head
, rb_node_dep
) {
214 zebra_nhg_decrement_ref(rb_node_dep
->nhe
);
219 nhg_connected_tree_increment_ref(struct nhg_connected_tree_head
*head
)
221 struct nhg_connected
*rb_node_dep
= NULL
;
223 frr_each(nhg_connected_tree
, head
, rb_node_dep
) {
224 zebra_nhg_increment_ref(rb_node_dep
->nhe
);
228 struct nhg_hash_entry
*zebra_nhg_resolve(struct nhg_hash_entry
*nhe
)
230 if (CHECK_FLAG(nhe
->flags
, NEXTHOP_GROUP_RECURSIVE
)
231 && !zebra_nhg_depends_is_empty(nhe
)) {
232 nhe
= nhg_connected_tree_root(&nhe
->nhg_depends
)->nhe
;
233 return zebra_nhg_resolve(nhe
);
239 unsigned int zebra_nhg_depends_count(const struct nhg_hash_entry
*nhe
)
241 return nhg_connected_tree_count(&nhe
->nhg_depends
);
244 bool zebra_nhg_depends_is_empty(const struct nhg_hash_entry
*nhe
)
246 return nhg_connected_tree_is_empty(&nhe
->nhg_depends
);
249 static void zebra_nhg_depends_del(struct nhg_hash_entry
*from
,
250 struct nhg_hash_entry
*depend
)
252 nhg_connected_tree_del_nhe(&from
->nhg_depends
, depend
);
255 static void zebra_nhg_depends_init(struct nhg_hash_entry
*nhe
)
257 nhg_connected_tree_init(&nhe
->nhg_depends
);
260 unsigned int zebra_nhg_dependents_count(const struct nhg_hash_entry
*nhe
)
262 return nhg_connected_tree_count(&nhe
->nhg_dependents
);
266 bool zebra_nhg_dependents_is_empty(const struct nhg_hash_entry
*nhe
)
268 return nhg_connected_tree_is_empty(&nhe
->nhg_dependents
);
271 static void zebra_nhg_dependents_del(struct nhg_hash_entry
*from
,
272 struct nhg_hash_entry
*dependent
)
274 nhg_connected_tree_del_nhe(&from
->nhg_dependents
, dependent
);
277 static void zebra_nhg_dependents_add(struct nhg_hash_entry
*to
,
278 struct nhg_hash_entry
*dependent
)
280 nhg_connected_tree_add_nhe(&to
->nhg_dependents
, dependent
);
283 static void zebra_nhg_dependents_init(struct nhg_hash_entry
*nhe
)
285 nhg_connected_tree_init(&nhe
->nhg_dependents
);
288 /* Release this nhe from anything depending on it */
289 static void zebra_nhg_dependents_release(struct nhg_hash_entry
*nhe
)
291 struct nhg_connected
*rb_node_dep
= NULL
;
293 frr_each_safe(nhg_connected_tree
, &nhe
->nhg_dependents
, rb_node_dep
) {
294 zebra_nhg_depends_del(rb_node_dep
->nhe
, nhe
);
295 /* recheck validity of the dependent */
296 zebra_nhg_check_valid(rb_node_dep
->nhe
);
300 /* Release this nhe from anything that it depends on */
301 static void zebra_nhg_depends_release(struct nhg_hash_entry
*nhe
)
303 if (!zebra_nhg_depends_is_empty(nhe
)) {
304 struct nhg_connected
*rb_node_dep
= NULL
;
306 frr_each_safe(nhg_connected_tree
, &nhe
->nhg_depends
,
308 zebra_nhg_dependents_del(rb_node_dep
->nhe
, nhe
);
314 struct nhg_hash_entry
*zebra_nhg_lookup_id(uint32_t id
)
316 struct nhg_hash_entry lookup
= {};
319 return hash_lookup(zrouter
.nhgs_id
, &lookup
);
322 static int zebra_nhg_insert_id(struct nhg_hash_entry
*nhe
)
324 if (hash_lookup(zrouter
.nhgs_id
, nhe
)) {
326 EC_ZEBRA_NHG_TABLE_INSERT_FAILED
,
327 "Failed inserting NHG %pNG into the ID hash table, entry already exists",
332 (void)hash_get(zrouter
.nhgs_id
, nhe
, hash_alloc_intern
);
337 static void zebra_nhg_set_if(struct nhg_hash_entry
*nhe
, struct interface
*ifp
)
340 if_nhg_dependents_add(ifp
, nhe
);
344 zebra_nhg_connect_depends(struct nhg_hash_entry
*nhe
,
345 struct nhg_connected_tree_head
*nhg_depends
)
347 struct nhg_connected
*rb_node_dep
= NULL
;
349 /* This has been allocated higher above in the stack. Could probably
350 * re-allocate and free the old stuff but just using the same memory
351 * for now. Otherwise, their might be a time trade-off for repeated
352 * alloc/frees as startup.
354 nhe
->nhg_depends
= *nhg_depends
;
356 /* Attach backpointer to anything that it depends on */
357 zebra_nhg_dependents_init(nhe
);
358 if (!zebra_nhg_depends_is_empty(nhe
)) {
359 frr_each(nhg_connected_tree
, &nhe
->nhg_depends
, rb_node_dep
) {
360 if (IS_ZEBRA_DEBUG_NHG_DETAIL
)
361 zlog_debug("%s: nhe %p (%pNG), dep %p (%pNG)",
362 __func__
, nhe
, nhe
, rb_node_dep
->nhe
,
365 zebra_nhg_dependents_add(rb_node_dep
->nhe
, nhe
);
370 /* Init an nhe, for use in a hash lookup for example */
371 void zebra_nhe_init(struct nhg_hash_entry
*nhe
, afi_t afi
,
372 const struct nexthop
*nh
)
374 memset(nhe
, 0, sizeof(struct nhg_hash_entry
));
375 nhe
->vrf_id
= VRF_DEFAULT
;
376 nhe
->type
= ZEBRA_ROUTE_NHG
;
377 nhe
->afi
= AFI_UNSPEC
;
379 /* There are some special rules that apply to groups representing
382 if (nh
&& (nh
->next
== NULL
)) {
384 case NEXTHOP_TYPE_IFINDEX
:
385 case NEXTHOP_TYPE_BLACKHOLE
:
387 * This switch case handles setting the afi different
388 * for ipv4/v6 routes. Ifindex/blackhole nexthop
389 * objects cannot be ambiguous, they must be Address
390 * Family specific. If we get here, we will either use
391 * the AF of the route, or the one we got passed from
392 * here from the kernel.
396 case NEXTHOP_TYPE_IPV4_IFINDEX
:
397 case NEXTHOP_TYPE_IPV4
:
400 case NEXTHOP_TYPE_IPV6_IFINDEX
:
401 case NEXTHOP_TYPE_IPV6
:
408 struct nhg_hash_entry
*zebra_nhg_alloc(void)
410 struct nhg_hash_entry
*nhe
;
412 nhe
= XCALLOC(MTYPE_NHG
, sizeof(struct nhg_hash_entry
));
418 * Allocate new nhe and make shallow copy of 'orig'; no
419 * recursive info is copied.
421 struct nhg_hash_entry
*zebra_nhe_copy(const struct nhg_hash_entry
*orig
,
424 struct nhg_hash_entry
*nhe
;
426 nhe
= zebra_nhg_alloc();
430 nexthop_group_copy(&(nhe
->nhg
), &(orig
->nhg
));
432 nhe
->vrf_id
= orig
->vrf_id
;
433 nhe
->afi
= orig
->afi
;
434 nhe
->type
= orig
->type
? orig
->type
: ZEBRA_ROUTE_NHG
;
436 nhe
->dplane_ref
= zebra_router_get_next_sequence();
438 /* Copy backup info also, if present */
439 if (orig
->backup_info
)
440 nhe
->backup_info
= nhg_backup_copy(orig
->backup_info
);
445 /* Allocation via hash handler */
446 static void *zebra_nhg_hash_alloc(void *arg
)
448 struct nhg_hash_entry
*nhe
= NULL
;
449 struct nhg_hash_entry
*copy
= arg
;
451 nhe
= zebra_nhe_copy(copy
, copy
->id
);
453 /* Mark duplicate nexthops in a group at creation time. */
454 nexthop_group_mark_duplicates(&(nhe
->nhg
));
457 * Add the ifp now if it's not a group or recursive and has ifindex.
459 * A proto-owned ID is always a group.
461 if (!PROTO_OWNED(nhe
) && nhe
->nhg
.nexthop
&& !nhe
->nhg
.nexthop
->next
462 && !nhe
->nhg
.nexthop
->resolved
&& nhe
->nhg
.nexthop
->ifindex
) {
463 struct interface
*ifp
= NULL
;
465 ifp
= if_lookup_by_index(nhe
->nhg
.nexthop
->ifindex
,
466 nhe
->nhg
.nexthop
->vrf_id
);
468 zebra_nhg_set_if(nhe
, ifp
);
470 if (IS_ZEBRA_DEBUG_NHG
)
472 "Failed to lookup an interface with ifindex=%d in vrf=%u for NHE %pNG",
473 nhe
->nhg
.nexthop
->ifindex
,
474 nhe
->nhg
.nexthop
->vrf_id
, nhe
);
481 uint32_t zebra_nhg_hash_key(const void *arg
)
483 const struct nhg_hash_entry
*nhe
= arg
;
484 uint32_t key
= 0x5a351234;
485 uint32_t primary
= 0;
488 primary
= nexthop_group_hash(&(nhe
->nhg
));
489 if (nhe
->backup_info
)
490 backup
= nexthop_group_hash(&(nhe
->backup_info
->nhe
->nhg
));
492 key
= jhash_3words(primary
, backup
, nhe
->type
, key
);
494 key
= jhash_2words(nhe
->vrf_id
, nhe
->afi
, key
);
499 uint32_t zebra_nhg_id_key(const void *arg
)
501 const struct nhg_hash_entry
*nhe
= arg
;
506 /* Helper with common nhg/nhe nexthop comparison logic */
507 static bool nhg_compare_nexthops(const struct nexthop
*nh1
,
508 const struct nexthop
*nh2
)
510 assert(nh1
!= NULL
&& nh2
!= NULL
);
513 * We have to check the active flag of each individual one,
514 * not just the overall active_num. This solves the special case
515 * issue of a route with a nexthop group with one nexthop
516 * resolving to itself and thus marking it inactive. If we
517 * have two different routes each wanting to mark a different
518 * nexthop inactive, they need to hash to two different groups.
520 * If we just hashed on num_active, they would hash the same
521 * which is incorrect.
525 * -> 1.1.1.1 dummy1 (inactive)
530 * -> 1.1.2.1 dummy2 (inactive)
532 * Without checking each individual one, they would hash to
533 * the same group and both have 1.1.1.1 dummy1 marked inactive.
536 if (CHECK_FLAG(nh1
->flags
, NEXTHOP_FLAG_ACTIVE
)
537 != CHECK_FLAG(nh2
->flags
, NEXTHOP_FLAG_ACTIVE
))
540 if (!nexthop_same(nh1
, nh2
))
546 bool zebra_nhg_hash_equal(const void *arg1
, const void *arg2
)
548 const struct nhg_hash_entry
*nhe1
= arg1
;
549 const struct nhg_hash_entry
*nhe2
= arg2
;
550 struct nexthop
*nexthop1
;
551 struct nexthop
*nexthop2
;
553 /* No matter what if they equal IDs, assume equal */
554 if (nhe1
->id
&& nhe2
->id
&& (nhe1
->id
== nhe2
->id
))
557 if (nhe1
->type
!= nhe2
->type
)
560 if (nhe1
->vrf_id
!= nhe2
->vrf_id
)
563 if (nhe1
->afi
!= nhe2
->afi
)
566 if (nhe1
->nhg
.nhgr
.buckets
!= nhe2
->nhg
.nhgr
.buckets
)
569 if (nhe1
->nhg
.nhgr
.idle_timer
!= nhe2
->nhg
.nhgr
.idle_timer
)
572 if (nhe1
->nhg
.nhgr
.unbalanced_timer
!= nhe2
->nhg
.nhgr
.unbalanced_timer
)
575 /* Nexthops should be in-order, so we simply compare them in-place */
576 for (nexthop1
= nhe1
->nhg
.nexthop
, nexthop2
= nhe2
->nhg
.nexthop
;
577 nexthop1
&& nexthop2
;
578 nexthop1
= nexthop1
->next
, nexthop2
= nexthop2
->next
) {
580 if (!nhg_compare_nexthops(nexthop1
, nexthop2
))
584 /* Check for unequal list lengths */
585 if (nexthop1
|| nexthop2
)
588 /* If there's no backup info, comparison is done. */
589 if ((nhe1
->backup_info
== NULL
) && (nhe2
->backup_info
== NULL
))
592 /* Compare backup info also - test the easy things first */
593 if (nhe1
->backup_info
&& (nhe2
->backup_info
== NULL
))
595 if (nhe2
->backup_info
&& (nhe1
->backup_info
== NULL
))
598 /* Compare number of backups before actually comparing any */
599 for (nexthop1
= nhe1
->backup_info
->nhe
->nhg
.nexthop
,
600 nexthop2
= nhe2
->backup_info
->nhe
->nhg
.nexthop
;
601 nexthop1
&& nexthop2
;
602 nexthop1
= nexthop1
->next
, nexthop2
= nexthop2
->next
) {
606 /* Did we find the end of one list before the other? */
607 if (nexthop1
|| nexthop2
)
610 /* Have to compare the backup nexthops */
611 for (nexthop1
= nhe1
->backup_info
->nhe
->nhg
.nexthop
,
612 nexthop2
= nhe2
->backup_info
->nhe
->nhg
.nexthop
;
613 nexthop1
&& nexthop2
;
614 nexthop1
= nexthop1
->next
, nexthop2
= nexthop2
->next
) {
616 if (!nhg_compare_nexthops(nexthop1
, nexthop2
))
623 bool zebra_nhg_hash_id_equal(const void *arg1
, const void *arg2
)
625 const struct nhg_hash_entry
*nhe1
= arg1
;
626 const struct nhg_hash_entry
*nhe2
= arg2
;
628 return nhe1
->id
== nhe2
->id
;
631 static int zebra_nhg_process_grp(struct nexthop_group
*nhg
,
632 struct nhg_connected_tree_head
*depends
,
633 struct nh_grp
*grp
, uint8_t count
,
634 struct nhg_resilience
*resilience
)
636 nhg_connected_tree_init(depends
);
638 for (int i
= 0; i
< count
; i
++) {
639 struct nhg_hash_entry
*depend
= NULL
;
640 /* We do not care about nexthop_grp.weight at
641 * this time. But we should figure out
642 * how to adapt this to our code in
645 depend
= depends_find_id_add(depends
, grp
[i
].id
);
650 "Received Nexthop Group from the kernel with a dependent Nexthop ID (%u) which we do not have in our table",
656 * If this is a nexthop with its own group
657 * dependencies, add them as well. Not sure its
658 * even possible to have a group within a group
662 copy_nexthops(&nhg
->nexthop
, depend
->nhg
.nexthop
, NULL
);
666 nhg
->nhgr
= *resilience
;
671 static void handle_recursive_depend(struct nhg_connected_tree_head
*nhg_depends
,
672 struct nexthop
*nh
, afi_t afi
, int type
)
674 struct nhg_hash_entry
*depend
= NULL
;
675 struct nexthop_group resolved_ng
= {};
677 resolved_ng
.nexthop
= nh
;
679 if (IS_ZEBRA_DEBUG_NHG_DETAIL
)
680 zlog_debug("%s: head %p, nh %pNHv",
681 __func__
, nhg_depends
, nh
);
683 depend
= zebra_nhg_rib_find(0, &resolved_ng
, afi
, type
);
685 if (IS_ZEBRA_DEBUG_NHG_DETAIL
)
686 zlog_debug("%s: nh %pNHv => %p (%u)",
687 __func__
, nh
, depend
,
688 depend
? depend
->id
: 0);
691 depends_add(nhg_depends
, depend
);
695 * Lookup an nhe in the global hash, using data from another nhe. If 'lookup'
696 * has an id value, that's used. Create a new global/shared nhe if not found.
698 static bool zebra_nhe_find(struct nhg_hash_entry
**nhe
, /* return value */
699 struct nhg_hash_entry
*lookup
,
700 struct nhg_connected_tree_head
*nhg_depends
,
701 afi_t afi
, bool from_dplane
)
703 bool created
= false;
704 bool recursive
= false;
705 struct nhg_hash_entry
*newnhe
, *backup_nhe
;
706 struct nexthop
*nh
= NULL
;
708 if (IS_ZEBRA_DEBUG_NHG_DETAIL
)
710 "%s: id %u, lookup %p, vrf %d, type %d, depends %p%s",
711 __func__
, lookup
->id
, lookup
, lookup
->vrf_id
,
712 lookup
->type
, nhg_depends
,
713 (from_dplane
? " (from dplane)" : ""));
716 (*nhe
) = zebra_nhg_lookup_id(lookup
->id
);
718 (*nhe
) = hash_lookup(zrouter
.nhgs
, lookup
);
720 if (IS_ZEBRA_DEBUG_NHG_DETAIL
)
721 zlog_debug("%s: lookup => %p (%pNG)", __func__
, *nhe
, *nhe
);
723 /* If we found an existing object, we're done */
727 /* We're going to create/insert a new nhe:
728 * assign the next global id value if necessary.
731 lookup
->id
= nhg_get_next_id();
733 if (!from_dplane
&& lookup
->id
< ZEBRA_NHG_PROTO_LOWER
) {
735 * This is a zebra hashed/owned NHG.
737 * It goes in HASH and ID table.
739 newnhe
= hash_get(zrouter
.nhgs
, lookup
, zebra_nhg_hash_alloc
);
740 zebra_nhg_insert_id(newnhe
);
743 * This is upperproto owned NHG or one we read in from dataplane
744 * and should not be hashed to.
746 * It goes in ID table.
749 hash_get(zrouter
.nhgs_id
, lookup
, zebra_nhg_hash_alloc
);
754 /* Mail back the new object */
757 if (IS_ZEBRA_DEBUG_NHG_DETAIL
)
758 zlog_debug("%s: => created %p (%pNG)", __func__
, newnhe
,
761 /* Only hash/lookup the depends if the first lookup
762 * fails to find something. This should hopefully save a
763 * lot of cycles for larger ecmp sizes.
766 /* If you don't want to hash on each nexthop in the
767 * nexthop group struct you can pass the depends
768 * directly. Kernel-side we do this since it just looks
771 zebra_nhg_connect_depends(newnhe
, nhg_depends
);
775 /* Prepare dependency relationships if this is not a
776 * singleton nexthop. There are two cases: a single
777 * recursive nexthop, where we need a relationship to the
778 * resolving nexthop; or a group of nexthops, where we need
779 * relationships with the corresponding singletons.
781 zebra_nhg_depends_init(newnhe
);
783 nh
= newnhe
->nhg
.nexthop
;
785 if (CHECK_FLAG(nh
->flags
, NEXTHOP_FLAG_ACTIVE
))
786 SET_FLAG(newnhe
->flags
, NEXTHOP_GROUP_VALID
);
788 if (nh
->next
== NULL
&& newnhe
->id
< ZEBRA_NHG_PROTO_LOWER
) {
789 if (CHECK_FLAG(nh
->flags
, NEXTHOP_FLAG_RECURSIVE
)) {
790 /* Single recursive nexthop */
791 handle_recursive_depend(&newnhe
->nhg_depends
,
797 /* Proto-owned are groups by default */
798 /* List of nexthops */
799 for (nh
= newnhe
->nhg
.nexthop
; nh
; nh
= nh
->next
) {
800 if (IS_ZEBRA_DEBUG_NHG_DETAIL
)
801 zlog_debug("%s: depends NH %pNHv %s",
803 CHECK_FLAG(nh
->flags
,
804 NEXTHOP_FLAG_RECURSIVE
) ?
807 depends_find_add(&newnhe
->nhg_depends
, nh
, afi
,
808 newnhe
->type
, from_dplane
);
813 SET_FLAG(newnhe
->flags
, NEXTHOP_GROUP_RECURSIVE
);
815 /* Attach dependent backpointers to singletons */
816 zebra_nhg_connect_depends(newnhe
, &newnhe
->nhg_depends
);
822 if (zebra_nhg_get_backup_nhg(newnhe
) == NULL
||
823 zebra_nhg_get_backup_nhg(newnhe
)->nexthop
== NULL
)
826 /* If there are backup nexthops, add them to the backup
827 * depends tree. The rules here are a little different.
830 backup_nhe
= newnhe
->backup_info
->nhe
;
832 nh
= backup_nhe
->nhg
.nexthop
;
834 /* Singleton recursive NH */
835 if (nh
->next
== NULL
&&
836 CHECK_FLAG(nh
->flags
, NEXTHOP_FLAG_RECURSIVE
)) {
837 if (IS_ZEBRA_DEBUG_NHG_DETAIL
)
838 zlog_debug("%s: backup depend NH %pNHv (R)",
841 /* Single recursive nexthop */
842 handle_recursive_depend(&backup_nhe
->nhg_depends
, nh
->resolved
,
843 afi
, backup_nhe
->type
);
846 /* One or more backup NHs */
847 for (; nh
; nh
= nh
->next
) {
848 if (IS_ZEBRA_DEBUG_NHG_DETAIL
)
849 zlog_debug("%s: backup depend NH %pNHv %s",
851 CHECK_FLAG(nh
->flags
,
852 NEXTHOP_FLAG_RECURSIVE
) ?
855 depends_find_add(&backup_nhe
->nhg_depends
, nh
, afi
,
856 backup_nhe
->type
, from_dplane
);
861 SET_FLAG(backup_nhe
->flags
, NEXTHOP_GROUP_RECURSIVE
);
864 /* Reset time since last update */
865 (*nhe
)->uptime
= monotime(NULL
);
871 * Lookup or create an nhe, based on an nhg or an nhe id.
873 static bool zebra_nhg_find(struct nhg_hash_entry
**nhe
, uint32_t id
,
874 struct nexthop_group
*nhg
,
875 struct nhg_connected_tree_head
*nhg_depends
,
876 vrf_id_t vrf_id
, afi_t afi
, int type
,
879 struct nhg_hash_entry lookup
= {};
880 bool created
= false;
882 if (IS_ZEBRA_DEBUG_NHG_DETAIL
)
883 zlog_debug("%s: id %u, nhg %p, vrf %d, type %d, depends %p",
884 __func__
, id
, nhg
, vrf_id
, type
,
887 /* Use a temporary nhe and call into the superset/common code */
889 lookup
.type
= type
? type
: ZEBRA_ROUTE_NHG
;
892 lookup
.vrf_id
= vrf_id
;
893 if (nhg_depends
|| lookup
.nhg
.nexthop
->next
) {
894 /* Groups can have all vrfs and AF's in them */
895 lookup
.afi
= AFI_UNSPEC
;
897 switch (lookup
.nhg
.nexthop
->type
) {
898 case (NEXTHOP_TYPE_IFINDEX
):
899 case (NEXTHOP_TYPE_BLACKHOLE
):
901 * This switch case handles setting the afi different
902 * for ipv4/v6 routes. Ifindex/blackhole nexthop
903 * objects cannot be ambiguous, they must be Address
904 * Family specific. If we get here, we will either use
905 * the AF of the route, or the one we got passed from
906 * here from the kernel.
910 case (NEXTHOP_TYPE_IPV4_IFINDEX
):
911 case (NEXTHOP_TYPE_IPV4
):
914 case (NEXTHOP_TYPE_IPV6_IFINDEX
):
915 case (NEXTHOP_TYPE_IPV6
):
916 lookup
.afi
= AFI_IP6
;
921 created
= zebra_nhe_find(nhe
, &lookup
, nhg_depends
, afi
, from_dplane
);
926 /* Find/create a single nexthop */
927 static struct nhg_hash_entry
*zebra_nhg_find_nexthop(uint32_t id
,
932 struct nhg_hash_entry
*nhe
= NULL
;
933 struct nexthop_group nhg
= {};
934 vrf_id_t vrf_id
= !vrf_is_backend_netns() ? VRF_DEFAULT
: nh
->vrf_id
;
936 nexthop_group_add_sorted(&nhg
, nh
);
938 zebra_nhg_find(&nhe
, id
, &nhg
, NULL
, vrf_id
, afi
, type
, from_dplane
);
940 if (IS_ZEBRA_DEBUG_NHG_DETAIL
)
941 zlog_debug("%s: nh %pNHv => %p (%pNG)", __func__
, nh
, nhe
, nhe
);
946 static uint32_t nhg_ctx_get_id(const struct nhg_ctx
*ctx
)
951 static void nhg_ctx_set_status(struct nhg_ctx
*ctx
, enum nhg_ctx_status status
)
953 ctx
->status
= status
;
956 static enum nhg_ctx_status
nhg_ctx_get_status(const struct nhg_ctx
*ctx
)
961 static void nhg_ctx_set_op(struct nhg_ctx
*ctx
, enum nhg_ctx_op_e op
)
966 static enum nhg_ctx_op_e
nhg_ctx_get_op(const struct nhg_ctx
*ctx
)
971 static vrf_id_t
nhg_ctx_get_vrf_id(const struct nhg_ctx
*ctx
)
976 static int nhg_ctx_get_type(const struct nhg_ctx
*ctx
)
981 static int nhg_ctx_get_afi(const struct nhg_ctx
*ctx
)
986 static struct nexthop
*nhg_ctx_get_nh(struct nhg_ctx
*ctx
)
991 static uint8_t nhg_ctx_get_count(const struct nhg_ctx
*ctx
)
996 static struct nh_grp
*nhg_ctx_get_grp(struct nhg_ctx
*ctx
)
1001 static struct nhg_resilience
*nhg_ctx_get_resilience(struct nhg_ctx
*ctx
)
1003 return &ctx
->resilience
;
1006 static struct nhg_ctx
*nhg_ctx_new(void)
1008 struct nhg_ctx
*new;
1010 new = XCALLOC(MTYPE_NHG_CTX
, sizeof(struct nhg_ctx
));
1015 void nhg_ctx_free(struct nhg_ctx
**ctx
)
1022 assert((*ctx
) != NULL
);
1024 if (nhg_ctx_get_count(*ctx
))
1027 nh
= nhg_ctx_get_nh(*ctx
);
1029 nexthop_del_labels(nh
);
1030 nexthop_del_srv6_seg6local(nh
);
1031 nexthop_del_srv6_seg6(nh
);
1034 XFREE(MTYPE_NHG_CTX
, *ctx
);
1037 static struct nhg_ctx
*nhg_ctx_init(uint32_t id
, struct nexthop
*nh
,
1038 struct nh_grp
*grp
, vrf_id_t vrf_id
,
1039 afi_t afi
, int type
, uint8_t count
,
1040 struct nhg_resilience
*resilience
)
1042 struct nhg_ctx
*ctx
= NULL
;
1044 ctx
= nhg_ctx_new();
1047 ctx
->vrf_id
= vrf_id
;
1053 ctx
->resilience
= *resilience
;
1056 /* Copy over the array */
1057 memcpy(&ctx
->u
.grp
, grp
, count
* sizeof(struct nh_grp
));
1064 static void zebra_nhg_set_valid(struct nhg_hash_entry
*nhe
)
1066 struct nhg_connected
*rb_node_dep
;
1068 SET_FLAG(nhe
->flags
, NEXTHOP_GROUP_VALID
);
1070 frr_each(nhg_connected_tree
, &nhe
->nhg_dependents
, rb_node_dep
)
1071 zebra_nhg_set_valid(rb_node_dep
->nhe
);
1074 static void zebra_nhg_set_invalid(struct nhg_hash_entry
*nhe
)
1076 struct nhg_connected
*rb_node_dep
;
1078 UNSET_FLAG(nhe
->flags
, NEXTHOP_GROUP_VALID
);
1080 /* If we're in shutdown, this interface event needs to clean
1081 * up installed NHGs, so don't clear that flag directly.
1083 if (!zebra_router_in_shutdown())
1084 UNSET_FLAG(nhe
->flags
, NEXTHOP_GROUP_INSTALLED
);
1086 /* Update validity of nexthops depending on it */
1087 frr_each(nhg_connected_tree
, &nhe
->nhg_dependents
, rb_node_dep
)
1088 zebra_nhg_check_valid(rb_node_dep
->nhe
);
1091 void zebra_nhg_check_valid(struct nhg_hash_entry
*nhe
)
1093 struct nhg_connected
*rb_node_dep
= NULL
;
1096 /* If anthing else in the group is valid, the group is valid */
1097 frr_each(nhg_connected_tree
, &nhe
->nhg_depends
, rb_node_dep
) {
1098 if (CHECK_FLAG(rb_node_dep
->nhe
->flags
, NEXTHOP_GROUP_VALID
)) {
1106 zebra_nhg_set_valid(nhe
);
1108 zebra_nhg_set_invalid(nhe
);
1111 static void zebra_nhg_release_all_deps(struct nhg_hash_entry
*nhe
)
1113 /* Remove it from any lists it may be on */
1114 zebra_nhg_depends_release(nhe
);
1115 zebra_nhg_dependents_release(nhe
);
1117 if_nhg_dependents_del(nhe
->ifp
, nhe
);
1120 static void zebra_nhg_release(struct nhg_hash_entry
*nhe
)
1122 if (IS_ZEBRA_DEBUG_NHG_DETAIL
)
1123 zlog_debug("%s: nhe %p (%pNG)", __func__
, nhe
, nhe
);
1125 zebra_nhg_release_all_deps(nhe
);
1128 * If its not zebra owned, we didn't store it here and have to be
1129 * sure we don't clear one thats actually being used.
1131 if (nhe
->id
< ZEBRA_NHG_PROTO_LOWER
)
1132 hash_release(zrouter
.nhgs
, nhe
);
1134 hash_release(zrouter
.nhgs_id
, nhe
);
1137 static void zebra_nhg_handle_uninstall(struct nhg_hash_entry
*nhe
)
1139 zebra_nhg_release(nhe
);
1140 zebra_nhg_free(nhe
);
1143 static void zebra_nhg_handle_install(struct nhg_hash_entry
*nhe
)
1145 /* Update validity of groups depending on it */
1146 struct nhg_connected
*rb_node_dep
;
1148 frr_each_safe(nhg_connected_tree
, &nhe
->nhg_dependents
, rb_node_dep
)
1149 zebra_nhg_set_valid(rb_node_dep
->nhe
);
1153 * The kernel/other program has changed the state of a nexthop object we are
1156 static void zebra_nhg_handle_kernel_state_change(struct nhg_hash_entry
*nhe
,
1162 "Kernel %s a nexthop group with ID (%pNG) that we are still using for a route, sending it back down",
1163 (is_delete
? "deleted" : "updated"), nhe
);
1165 UNSET_FLAG(nhe
->flags
, NEXTHOP_GROUP_INSTALLED
);
1166 zebra_nhg_install_kernel(nhe
);
1168 zebra_nhg_handle_uninstall(nhe
);
1171 static int nhg_ctx_process_new(struct nhg_ctx
*ctx
)
1173 struct nexthop_group
*nhg
= NULL
;
1174 struct nhg_connected_tree_head nhg_depends
= {};
1175 struct nhg_hash_entry
*lookup
= NULL
;
1176 struct nhg_hash_entry
*nhe
= NULL
;
1178 uint32_t id
= nhg_ctx_get_id(ctx
);
1179 uint8_t count
= nhg_ctx_get_count(ctx
);
1180 vrf_id_t vrf_id
= nhg_ctx_get_vrf_id(ctx
);
1181 int type
= nhg_ctx_get_type(ctx
);
1182 afi_t afi
= nhg_ctx_get_afi(ctx
);
1184 lookup
= zebra_nhg_lookup_id(id
);
1186 if (IS_ZEBRA_DEBUG_NHG_DETAIL
)
1187 zlog_debug("%s: id %u, count %d, lookup => %p",
1188 __func__
, id
, count
, lookup
);
1191 /* This is already present in our table, hence an update
1192 * that we did not initate.
1194 zebra_nhg_handle_kernel_state_change(lookup
, false);
1198 if (nhg_ctx_get_count(ctx
)) {
1199 nhg
= nexthop_group_new();
1200 if (zebra_nhg_process_grp(nhg
, &nhg_depends
,
1201 nhg_ctx_get_grp(ctx
), count
,
1202 nhg_ctx_get_resilience(ctx
))) {
1203 depends_decrement_free(&nhg_depends
);
1204 nexthop_group_delete(&nhg
);
1208 if (!zebra_nhg_find(&nhe
, id
, nhg
, &nhg_depends
, vrf_id
, afi
,
1210 depends_decrement_free(&nhg_depends
);
1212 /* These got copied over in zebra_nhg_alloc() */
1213 nexthop_group_delete(&nhg
);
1215 nhe
= zebra_nhg_find_nexthop(id
, nhg_ctx_get_nh(ctx
), afi
, type
,
1220 EC_ZEBRA_TABLE_LOOKUP_FAILED
,
1221 "Zebra failed to find or create a nexthop hash entry for ID (%u)",
1226 if (IS_ZEBRA_DEBUG_NHG_DETAIL
)
1227 zlog_debug("%s: nhe %p (%pNG) is new", __func__
, nhe
, nhe
);
1230 * If daemon nhg from the kernel, add a refcnt here to indicate the
1233 if (PROTO_OWNED(nhe
))
1234 zebra_nhg_increment_ref(nhe
);
1236 SET_FLAG(nhe
->flags
, NEXTHOP_GROUP_VALID
);
1237 SET_FLAG(nhe
->flags
, NEXTHOP_GROUP_INSTALLED
);
1242 static int nhg_ctx_process_del(struct nhg_ctx
*ctx
)
1244 struct nhg_hash_entry
*nhe
= NULL
;
1245 uint32_t id
= nhg_ctx_get_id(ctx
);
1247 nhe
= zebra_nhg_lookup_id(id
);
1251 EC_ZEBRA_BAD_NHG_MESSAGE
,
1252 "Kernel delete message received for nexthop group ID (%u) that we do not have in our ID table",
1257 zebra_nhg_handle_kernel_state_change(nhe
, true);
1262 static void nhg_ctx_fini(struct nhg_ctx
**ctx
)
1265 * Just freeing for now, maybe do something more in the future
1272 static int queue_add(struct nhg_ctx
*ctx
)
1274 /* If its queued or already processed do nothing */
1275 if (nhg_ctx_get_status(ctx
) == NHG_CTX_QUEUED
)
1278 if (rib_queue_nhg_ctx_add(ctx
)) {
1279 nhg_ctx_set_status(ctx
, NHG_CTX_FAILURE
);
1283 nhg_ctx_set_status(ctx
, NHG_CTX_QUEUED
);
1288 int nhg_ctx_process(struct nhg_ctx
*ctx
)
1292 switch (nhg_ctx_get_op(ctx
)) {
1293 case NHG_CTX_OP_NEW
:
1294 ret
= nhg_ctx_process_new(ctx
);
1295 if (nhg_ctx_get_count(ctx
) && ret
== -ENOENT
1296 && nhg_ctx_get_status(ctx
) != NHG_CTX_REQUEUED
) {
1298 * We have entered a situation where we are
1299 * processing a group from the kernel
1300 * that has a contained nexthop which
1301 * we have not yet processed.
1303 * Re-enqueue this ctx to be handled exactly one
1304 * more time (indicated by the flag).
1306 * By the time we get back to it, we
1307 * should have processed its depends.
1309 nhg_ctx_set_status(ctx
, NHG_CTX_NONE
);
1310 if (queue_add(ctx
) == 0) {
1311 nhg_ctx_set_status(ctx
, NHG_CTX_REQUEUED
);
1316 case NHG_CTX_OP_DEL
:
1317 ret
= nhg_ctx_process_del(ctx
);
1318 case NHG_CTX_OP_NONE
:
1322 nhg_ctx_set_status(ctx
, (ret
? NHG_CTX_FAILURE
: NHG_CTX_SUCCESS
));
1329 /* Kernel-side, you either get a single new nexthop or a array of ID's */
1330 int zebra_nhg_kernel_find(uint32_t id
, struct nexthop
*nh
, struct nh_grp
*grp
,
1331 uint8_t count
, vrf_id_t vrf_id
, afi_t afi
, int type
,
1332 int startup
, struct nhg_resilience
*nhgr
)
1334 struct nhg_ctx
*ctx
= NULL
;
1336 if (IS_ZEBRA_DEBUG_NHG_DETAIL
)
1337 zlog_debug("%s: nh %pNHv, id %u, count %d",
1338 __func__
, nh
, id
, (int)count
);
1340 if (id
> id_counter
&& id
< ZEBRA_NHG_PROTO_LOWER
)
1341 /* Increase our counter so we don't try to create
1342 * an ID that already exists
1346 ctx
= nhg_ctx_init(id
, nh
, grp
, vrf_id
, afi
, type
, count
, nhgr
);
1347 nhg_ctx_set_op(ctx
, NHG_CTX_OP_NEW
);
1349 /* Under statup conditions, we need to handle them immediately
1350 * like we do for routes. Otherwise, we are going to get a route
1351 * with a nhe_id that we have not handled.
1354 return nhg_ctx_process(ctx
);
1356 if (queue_add(ctx
)) {
1364 /* Kernel-side, received delete message */
1365 int zebra_nhg_kernel_del(uint32_t id
, vrf_id_t vrf_id
)
1367 struct nhg_ctx
*ctx
= NULL
;
1369 ctx
= nhg_ctx_init(id
, NULL
, NULL
, vrf_id
, 0, 0, 0, NULL
);
1371 nhg_ctx_set_op(ctx
, NHG_CTX_OP_DEL
);
1373 if (queue_add(ctx
)) {
1381 /* Some dependency helper functions */
1382 static struct nhg_hash_entry
*depends_find_recursive(const struct nexthop
*nh
,
1383 afi_t afi
, int type
)
1385 struct nhg_hash_entry
*nhe
;
1386 struct nexthop
*lookup
= NULL
;
1388 lookup
= nexthop_dup(nh
, NULL
);
1390 nhe
= zebra_nhg_find_nexthop(0, lookup
, afi
, type
, false);
1392 nexthops_free(lookup
);
1397 static struct nhg_hash_entry
*depends_find_singleton(const struct nexthop
*nh
,
1398 afi_t afi
, int type
,
1401 struct nhg_hash_entry
*nhe
;
1402 struct nexthop lookup
= {};
1404 /* Capture a snapshot of this single nh; it might be part of a list,
1405 * so we need to make a standalone copy.
1407 nexthop_copy_no_recurse(&lookup
, nh
, NULL
);
1409 nhe
= zebra_nhg_find_nexthop(0, &lookup
, afi
, type
, from_dplane
);
1411 /* The copy may have allocated labels; free them if necessary. */
1412 nexthop_del_labels(&lookup
);
1413 nexthop_del_srv6_seg6local(&lookup
);
1414 nexthop_del_srv6_seg6(&lookup
);
1416 if (IS_ZEBRA_DEBUG_NHG_DETAIL
)
1417 zlog_debug("%s: nh %pNHv => %p (%pNG)", __func__
, nh
, nhe
, nhe
);
1422 static struct nhg_hash_entry
*depends_find(const struct nexthop
*nh
, afi_t afi
,
1423 int type
, bool from_dplane
)
1425 struct nhg_hash_entry
*nhe
= NULL
;
1430 /* We are separating these functions out to increase handling speed
1431 * in the non-recursive case (by not alloc/freeing)
1433 if (CHECK_FLAG(nh
->flags
, NEXTHOP_FLAG_RECURSIVE
))
1434 nhe
= depends_find_recursive(nh
, afi
, type
);
1436 nhe
= depends_find_singleton(nh
, afi
, type
, from_dplane
);
1439 if (IS_ZEBRA_DEBUG_NHG_DETAIL
) {
1440 zlog_debug("%s: nh %pNHv %s => %p (%pNG)", __func__
, nh
,
1441 CHECK_FLAG(nh
->flags
, NEXTHOP_FLAG_RECURSIVE
) ? "(R)"
1450 static void depends_add(struct nhg_connected_tree_head
*head
,
1451 struct nhg_hash_entry
*depend
)
1453 if (IS_ZEBRA_DEBUG_NHG_DETAIL
)
1454 zlog_debug("%s: head %p nh %pNHv",
1455 __func__
, head
, depend
->nhg
.nexthop
);
1457 /* If NULL is returned, it was successfully added and
1458 * needs to have its refcnt incremented.
1460 * Else the NHE is already present in the tree and doesn't
1461 * need to increment the refcnt.
1463 if (nhg_connected_tree_add_nhe(head
, depend
) == NULL
)
1464 zebra_nhg_increment_ref(depend
);
1467 static struct nhg_hash_entry
*
1468 depends_find_add(struct nhg_connected_tree_head
*head
, struct nexthop
*nh
,
1469 afi_t afi
, int type
, bool from_dplane
)
1471 struct nhg_hash_entry
*depend
= NULL
;
1473 depend
= depends_find(nh
, afi
, type
, from_dplane
);
1475 if (IS_ZEBRA_DEBUG_NHG_DETAIL
)
1476 zlog_debug("%s: nh %pNHv => %p",
1477 __func__
, nh
, depend
);
1480 depends_add(head
, depend
);
1485 static struct nhg_hash_entry
*
1486 depends_find_id_add(struct nhg_connected_tree_head
*head
, uint32_t id
)
1488 struct nhg_hash_entry
*depend
= NULL
;
1490 depend
= zebra_nhg_lookup_id(id
);
1493 depends_add(head
, depend
);
1498 static void depends_decrement_free(struct nhg_connected_tree_head
*head
)
1500 nhg_connected_tree_decrement_ref(head
);
1501 nhg_connected_tree_free(head
);
1504 /* Find an nhe based on a list of nexthops */
1505 struct nhg_hash_entry
*zebra_nhg_rib_find(uint32_t id
,
1506 struct nexthop_group
*nhg
,
1507 afi_t rt_afi
, int type
)
1509 struct nhg_hash_entry
*nhe
= NULL
;
1513 * CLANG SA is complaining that nexthop may be NULL
1514 * Make it happy but this is ridonc
1516 assert(nhg
->nexthop
);
1517 vrf_id
= !vrf_is_backend_netns() ? VRF_DEFAULT
: nhg
->nexthop
->vrf_id
;
1519 zebra_nhg_find(&nhe
, id
, nhg
, NULL
, vrf_id
, rt_afi
, type
, false);
1521 if (IS_ZEBRA_DEBUG_NHG_DETAIL
)
1522 zlog_debug("%s: => nhe %p (%pNG)", __func__
, nhe
, nhe
);
1527 /* Find an nhe based on a route's nhe */
1528 struct nhg_hash_entry
*
1529 zebra_nhg_rib_find_nhe(struct nhg_hash_entry
*rt_nhe
, afi_t rt_afi
)
1531 struct nhg_hash_entry
*nhe
= NULL
;
1533 if (!(rt_nhe
&& rt_nhe
->nhg
.nexthop
)) {
1534 flog_err(EC_ZEBRA_TABLE_LOOKUP_FAILED
,
1535 "No nexthop passed to %s", __func__
);
1539 if (IS_ZEBRA_DEBUG_NHG_DETAIL
)
1540 zlog_debug("%s: rt_nhe %p (%pNG)", __func__
, rt_nhe
, rt_nhe
);
1542 zebra_nhe_find(&nhe
, rt_nhe
, NULL
, rt_afi
, false);
1544 if (IS_ZEBRA_DEBUG_NHG_DETAIL
)
1545 zlog_debug("%s: => nhe %p (%pNG)", __func__
, nhe
, nhe
);
1551 * Allocate backup nexthop info object. Typically these are embedded in
1552 * nhg_hash_entry objects.
1554 struct nhg_backup_info
*zebra_nhg_backup_alloc(void)
1556 struct nhg_backup_info
*p
;
1558 p
= XCALLOC(MTYPE_NHG
, sizeof(struct nhg_backup_info
));
1560 p
->nhe
= zebra_nhg_alloc();
1562 /* Identify the embedded group used to hold the list of backups */
1563 SET_FLAG(p
->nhe
->flags
, NEXTHOP_GROUP_BACKUP
);
1569 * Free backup nexthop info object, deal with any embedded allocations
1571 void zebra_nhg_backup_free(struct nhg_backup_info
**p
)
1575 zebra_nhg_free((*p
)->nhe
);
1577 XFREE(MTYPE_NHG
, (*p
));
1581 /* Accessor for backup nexthop group */
1582 struct nexthop_group
*zebra_nhg_get_backup_nhg(struct nhg_hash_entry
*nhe
)
1584 struct nexthop_group
*p
= NULL
;
1587 if (nhe
->backup_info
&& nhe
->backup_info
->nhe
)
1588 p
= &(nhe
->backup_info
->nhe
->nhg
);
1595 * Helper to return a copy of a backup_info - note that this is a shallow
1596 * copy, meant to be used when creating a new nhe from info passed in with
1599 static struct nhg_backup_info
*
1600 nhg_backup_copy(const struct nhg_backup_info
*orig
)
1602 struct nhg_backup_info
*b
;
1604 b
= zebra_nhg_backup_alloc();
1606 /* Copy list of nexthops */
1607 nexthop_group_copy(&(b
->nhe
->nhg
), &(orig
->nhe
->nhg
));
1612 static void zebra_nhg_free_members(struct nhg_hash_entry
*nhe
)
1614 nexthops_free(nhe
->nhg
.nexthop
);
1616 zebra_nhg_backup_free(&nhe
->backup_info
);
1618 /* Decrement to remove connection ref */
1619 nhg_connected_tree_decrement_ref(&nhe
->nhg_depends
);
1620 nhg_connected_tree_free(&nhe
->nhg_depends
);
1621 nhg_connected_tree_free(&nhe
->nhg_dependents
);
1624 void zebra_nhg_free(struct nhg_hash_entry
*nhe
)
1626 if (IS_ZEBRA_DEBUG_NHG_DETAIL
) {
1627 /* Group or singleton? */
1628 if (nhe
->nhg
.nexthop
&& nhe
->nhg
.nexthop
->next
)
1629 zlog_debug("%s: nhe %p (%pNG), refcnt %d", __func__
,
1630 nhe
, nhe
, nhe
->refcnt
);
1632 zlog_debug("%s: nhe %p (%pNG), refcnt %d, NH %pNHv",
1633 __func__
, nhe
, nhe
, nhe
->refcnt
,
1637 THREAD_OFF(nhe
->timer
);
1639 zebra_nhg_free_members(nhe
);
1641 XFREE(MTYPE_NHG
, nhe
);
1645 * Let's just drop the memory associated with each item
1647 void zebra_nhg_hash_free(void *p
)
1649 struct nhg_hash_entry
*nhe
= p
;
1651 if (IS_ZEBRA_DEBUG_NHG_DETAIL
) {
1652 /* Group or singleton? */
1653 if (nhe
->nhg
.nexthop
&& nhe
->nhg
.nexthop
->next
)
1654 zlog_debug("%s: nhe %p (%u), refcnt %d", __func__
, nhe
,
1655 nhe
->id
, nhe
->refcnt
);
1657 zlog_debug("%s: nhe %p (%pNG), refcnt %d, NH %pNHv",
1658 __func__
, nhe
, nhe
, nhe
->refcnt
,
1662 THREAD_OFF(nhe
->timer
);
1664 nexthops_free(nhe
->nhg
.nexthop
);
1666 XFREE(MTYPE_NHG
, nhe
);
1670 * On cleanup there are nexthop groups that have not
1671 * been resolved at all( a nhe->id of 0 ). As such
1672 * zebra needs to clean up the memory associated with
1675 void zebra_nhg_hash_free_zero_id(struct hash_bucket
*b
, void *arg
)
1677 struct nhg_hash_entry
*nhe
= b
->data
;
1678 struct nhg_connected
*dep
;
1680 while ((dep
= nhg_connected_tree_pop(&nhe
->nhg_depends
))) {
1681 if (dep
->nhe
->id
== 0)
1682 zebra_nhg_hash_free(dep
->nhe
);
1684 nhg_connected_free(dep
);
1687 while ((dep
= nhg_connected_tree_pop(&nhe
->nhg_dependents
)))
1688 nhg_connected_free(dep
);
1690 if (nhe
->backup_info
&& nhe
->backup_info
->nhe
->id
== 0) {
1691 while ((dep
= nhg_connected_tree_pop(
1692 &nhe
->backup_info
->nhe
->nhg_depends
)))
1693 nhg_connected_free(dep
);
1695 zebra_nhg_hash_free(nhe
->backup_info
->nhe
);
1697 XFREE(MTYPE_NHG
, nhe
->backup_info
);
1701 static void zebra_nhg_timer(struct thread
*thread
)
1703 struct nhg_hash_entry
*nhe
= THREAD_ARG(thread
);
1705 if (IS_ZEBRA_DEBUG_NHG_DETAIL
)
1706 zlog_debug("Nexthop Timer for nhe: %pNG", nhe
);
1708 if (nhe
->refcnt
== 1)
1709 zebra_nhg_decrement_ref(nhe
);
1712 void zebra_nhg_decrement_ref(struct nhg_hash_entry
*nhe
)
1714 if (IS_ZEBRA_DEBUG_NHG_DETAIL
)
1715 zlog_debug("%s: nhe %p (%pNG) %d => %d", __func__
, nhe
, nhe
,
1716 nhe
->refcnt
, nhe
->refcnt
- 1);
1720 if (!zebra_router_in_shutdown() && nhe
->refcnt
<= 0 &&
1721 CHECK_FLAG(nhe
->flags
, NEXTHOP_GROUP_INSTALLED
) &&
1722 !CHECK_FLAG(nhe
->flags
, NEXTHOP_GROUP_KEEP_AROUND
)) {
1724 SET_FLAG(nhe
->flags
, NEXTHOP_GROUP_KEEP_AROUND
);
1725 thread_add_timer(zrouter
.master
, zebra_nhg_timer
, nhe
,
1726 zrouter
.nhg_keep
, &nhe
->timer
);
1730 if (!zebra_nhg_depends_is_empty(nhe
))
1731 nhg_connected_tree_decrement_ref(&nhe
->nhg_depends
);
1733 if (ZEBRA_NHG_CREATED(nhe
) && nhe
->refcnt
<= 0)
1734 zebra_nhg_uninstall_kernel(nhe
);
1737 void zebra_nhg_increment_ref(struct nhg_hash_entry
*nhe
)
1739 if (IS_ZEBRA_DEBUG_NHG_DETAIL
)
1740 zlog_debug("%s: nhe %p (%pNG) %d => %d", __func__
, nhe
, nhe
,
1741 nhe
->refcnt
, nhe
->refcnt
+ 1);
1745 if (thread_is_scheduled(nhe
->timer
)) {
1746 THREAD_OFF(nhe
->timer
);
1748 UNSET_FLAG(nhe
->flags
, NEXTHOP_GROUP_KEEP_AROUND
);
1751 if (!zebra_nhg_depends_is_empty(nhe
))
1752 nhg_connected_tree_increment_ref(&nhe
->nhg_depends
);
1755 static struct nexthop
*nexthop_set_resolved(afi_t afi
,
1756 const struct nexthop
*newhop
,
1757 struct nexthop
*nexthop
,
1758 struct zebra_sr_policy
*policy
)
1760 struct nexthop
*resolved_hop
;
1761 uint8_t num_labels
= 0;
1762 mpls_label_t labels
[MPLS_MAX_LABELS
];
1763 enum lsp_types_t label_type
= ZEBRA_LSP_NONE
;
1766 resolved_hop
= nexthop_new();
1767 SET_FLAG(resolved_hop
->flags
, NEXTHOP_FLAG_ACTIVE
);
1769 resolved_hop
->vrf_id
= nexthop
->vrf_id
;
1770 switch (newhop
->type
) {
1771 case NEXTHOP_TYPE_IPV4
:
1772 case NEXTHOP_TYPE_IPV4_IFINDEX
:
1773 /* If the resolving route specifies a gateway, use it */
1774 resolved_hop
->type
= newhop
->type
;
1775 resolved_hop
->gate
.ipv4
= newhop
->gate
.ipv4
;
1777 if (newhop
->ifindex
) {
1778 resolved_hop
->type
= NEXTHOP_TYPE_IPV4_IFINDEX
;
1779 resolved_hop
->ifindex
= newhop
->ifindex
;
1782 case NEXTHOP_TYPE_IPV6
:
1783 case NEXTHOP_TYPE_IPV6_IFINDEX
:
1784 resolved_hop
->type
= newhop
->type
;
1785 resolved_hop
->gate
.ipv6
= newhop
->gate
.ipv6
;
1787 if (newhop
->ifindex
) {
1788 resolved_hop
->type
= NEXTHOP_TYPE_IPV6_IFINDEX
;
1789 resolved_hop
->ifindex
= newhop
->ifindex
;
1792 case NEXTHOP_TYPE_IFINDEX
:
1793 /* If the resolving route is an interface route,
1794 * it means the gateway we are looking up is connected
1795 * to that interface. (The actual network is _not_ onlink).
1796 * Therefore, the resolved route should have the original
1797 * gateway as nexthop as it is directly connected.
1799 * On Linux, we have to set the onlink netlink flag because
1800 * otherwise, the kernel won't accept the route.
1802 resolved_hop
->flags
|= NEXTHOP_FLAG_ONLINK
;
1803 if (afi
== AFI_IP
) {
1804 resolved_hop
->type
= NEXTHOP_TYPE_IPV4_IFINDEX
;
1805 resolved_hop
->gate
.ipv4
= nexthop
->gate
.ipv4
;
1806 } else if (afi
== AFI_IP6
) {
1807 resolved_hop
->type
= NEXTHOP_TYPE_IPV6_IFINDEX
;
1808 resolved_hop
->gate
.ipv6
= nexthop
->gate
.ipv6
;
1810 resolved_hop
->ifindex
= newhop
->ifindex
;
1812 case NEXTHOP_TYPE_BLACKHOLE
:
1813 resolved_hop
->type
= NEXTHOP_TYPE_BLACKHOLE
;
1814 resolved_hop
->bh_type
= newhop
->bh_type
;
1818 if (newhop
->flags
& NEXTHOP_FLAG_ONLINK
)
1819 resolved_hop
->flags
|= NEXTHOP_FLAG_ONLINK
;
1821 /* Copy labels of the resolved route and the parent resolving to it */
1826 * Don't push the first SID if the corresponding action in the
1829 if (!newhop
->nh_label
|| !newhop
->nh_label
->num_labels
1830 || newhop
->nh_label
->label
[0] == MPLS_LABEL_IMPLICIT_NULL
)
1833 for (; i
< policy
->segment_list
.label_num
; i
++)
1834 labels
[num_labels
++] = policy
->segment_list
.labels
[i
];
1835 label_type
= policy
->segment_list
.type
;
1836 } else if (newhop
->nh_label
) {
1837 for (i
= 0; i
< newhop
->nh_label
->num_labels
; i
++) {
1838 /* Be a bit picky about overrunning the local array */
1839 if (num_labels
>= MPLS_MAX_LABELS
) {
1840 if (IS_ZEBRA_DEBUG_NHG
|| IS_ZEBRA_DEBUG_RIB
)
1841 zlog_debug("%s: too many labels in newhop %pNHv",
1845 labels
[num_labels
++] = newhop
->nh_label
->label
[i
];
1847 /* Use the "outer" type */
1848 label_type
= newhop
->nh_label_type
;
1851 if (nexthop
->nh_label
) {
1852 for (i
= 0; i
< nexthop
->nh_label
->num_labels
; i
++) {
1853 /* Be a bit picky about overrunning the local array */
1854 if (num_labels
>= MPLS_MAX_LABELS
) {
1855 if (IS_ZEBRA_DEBUG_NHG
|| IS_ZEBRA_DEBUG_RIB
)
1856 zlog_debug("%s: too many labels in nexthop %pNHv",
1860 labels
[num_labels
++] = nexthop
->nh_label
->label
[i
];
1863 /* If the parent has labels, use its type if
1864 * we don't already have one.
1866 if (label_type
== ZEBRA_LSP_NONE
)
1867 label_type
= nexthop
->nh_label_type
;
1871 nexthop_add_labels(resolved_hop
, label_type
, num_labels
,
1874 if (nexthop
->nh_srv6
) {
1875 nexthop_add_srv6_seg6local(resolved_hop
,
1876 nexthop
->nh_srv6
->seg6local_action
,
1877 &nexthop
->nh_srv6
->seg6local_ctx
);
1878 nexthop_add_srv6_seg6(resolved_hop
,
1879 &nexthop
->nh_srv6
->seg6_segs
);
1882 resolved_hop
->rparent
= nexthop
;
1883 _nexthop_add(&nexthop
->resolved
, resolved_hop
);
1885 return resolved_hop
;
1888 /* Checks if nexthop we are trying to resolve to is valid */
1889 static bool nexthop_valid_resolve(const struct nexthop
*nexthop
,
1890 const struct nexthop
*resolved
)
1892 /* Can't resolve to a recursive nexthop */
1893 if (CHECK_FLAG(resolved
->flags
, NEXTHOP_FLAG_RECURSIVE
))
1896 /* Must be ACTIVE */
1897 if (!CHECK_FLAG(resolved
->flags
, NEXTHOP_FLAG_ACTIVE
))
1900 /* Must not be duplicate */
1901 if (CHECK_FLAG(resolved
->flags
, NEXTHOP_FLAG_DUPLICATE
))
1904 switch (nexthop
->type
) {
1905 case NEXTHOP_TYPE_IPV4_IFINDEX
:
1906 case NEXTHOP_TYPE_IPV6_IFINDEX
:
1907 /* If the nexthop we are resolving to does not match the
1908 * ifindex for the nexthop the route wanted, its not valid.
1910 if (nexthop
->ifindex
!= resolved
->ifindex
)
1913 case NEXTHOP_TYPE_IPV4
:
1914 case NEXTHOP_TYPE_IPV6
:
1915 case NEXTHOP_TYPE_IFINDEX
:
1916 case NEXTHOP_TYPE_BLACKHOLE
:
1924 * When resolving a recursive nexthop, capture backup nexthop(s) also
1925 * so they can be conveyed through the dataplane to the FIB. We'll look
1926 * at the backups in the resolving nh 'nexthop' and its nhe, and copy them
1927 * into the route's resolved nh 'resolved' and its nhe 'nhe'.
1929 static int resolve_backup_nexthops(const struct nexthop
*nexthop
,
1930 const struct nhg_hash_entry
*nhe
,
1931 struct nexthop
*resolved
,
1932 struct nhg_hash_entry
*resolve_nhe
,
1933 struct backup_nh_map_s
*map
)
1936 const struct nexthop
*bnh
;
1937 struct nexthop
*nh
, *newnh
;
1938 mpls_label_t labels
[MPLS_MAX_LABELS
];
1941 assert(nexthop
->backup_num
<= NEXTHOP_MAX_BACKUPS
);
1943 /* Locate backups from the original nexthop's backup index and nhe */
1944 for (i
= 0; i
< nexthop
->backup_num
; i
++) {
1945 idx
= nexthop
->backup_idx
[i
];
1947 /* Do we already know about this particular backup? */
1948 for (j
= 0; j
< map
->map_count
; j
++) {
1949 if (map
->map
[j
].orig_idx
== idx
)
1953 if (j
< map
->map_count
) {
1954 resolved
->backup_idx
[resolved
->backup_num
] =
1955 map
->map
[j
].new_idx
;
1956 resolved
->backup_num
++;
1958 SET_FLAG(resolved
->flags
, NEXTHOP_FLAG_HAS_BACKUP
);
1960 if (IS_ZEBRA_DEBUG_RIB_DETAILED
)
1961 zlog_debug("%s: found map idx orig %d, new %d",
1962 __func__
, map
->map
[j
].orig_idx
,
1963 map
->map
[j
].new_idx
);
1968 /* We can't handle any new map entries at this point. */
1969 if (map
->map_count
== MULTIPATH_NUM
)
1972 /* Need to create/copy a new backup */
1973 bnh
= nhe
->backup_info
->nhe
->nhg
.nexthop
;
1974 for (j
= 0; j
< idx
; j
++) {
1980 /* Whoops - bad index in the nexthop? */
1984 if (resolve_nhe
->backup_info
== NULL
)
1985 resolve_nhe
->backup_info
= zebra_nhg_backup_alloc();
1987 /* Update backup info in the resolving nexthop and its nhe */
1988 newnh
= nexthop_dup_no_recurse(bnh
, NULL
);
1990 /* We may need some special handling for mpls labels: the new
1991 * backup needs to carry the recursive nexthop's labels,
1992 * if any: they may be vrf labels e.g.
1993 * The original/inner labels are in the stack of 'resolve_nhe',
1994 * if that is longer than the stack in 'nexthop'.
1996 if (newnh
->nh_label
&& resolved
->nh_label
&&
1997 nexthop
->nh_label
) {
1998 if (resolved
->nh_label
->num_labels
>
1999 nexthop
->nh_label
->num_labels
) {
2000 /* Prepare new label stack */
2002 for (j
= 0; j
< newnh
->nh_label
->num_labels
;
2004 labels
[j
] = newnh
->nh_label
->label
[j
];
2008 /* Include inner labels */
2009 for (j
= nexthop
->nh_label
->num_labels
;
2010 j
< resolved
->nh_label
->num_labels
;
2012 labels
[num_labels
] =
2013 resolved
->nh_label
->label
[j
];
2017 /* Replace existing label stack in the backup */
2018 nexthop_del_labels(newnh
);
2019 nexthop_add_labels(newnh
, bnh
->nh_label_type
,
2020 num_labels
, labels
);
2024 /* Need to compute the new backup index in the new
2025 * backup list, and add to map struct.
2028 nh
= resolve_nhe
->backup_info
->nhe
->nhg
.nexthop
;
2038 } else /* First one */
2039 resolve_nhe
->backup_info
->nhe
->nhg
.nexthop
= newnh
;
2042 resolved
->backup_idx
[resolved
->backup_num
] = j
;
2043 resolved
->backup_num
++;
2045 SET_FLAG(resolved
->flags
, NEXTHOP_FLAG_HAS_BACKUP
);
2047 if (IS_ZEBRA_DEBUG_RIB_DETAILED
)
2048 zlog_debug("%s: added idx orig %d, new %d",
2051 /* Update map/cache */
2052 map
->map
[map
->map_count
].orig_idx
= idx
;
2053 map
->map
[map
->map_count
].new_idx
= j
;
2061 * So this nexthop resolution has decided that a connected route
2062 * is the correct choice. At this point in time if FRR has multiple
2063 * connected routes that all point to the same prefix one will be
2064 * selected, *but* the particular interface may not be the one
2065 * that the nexthop points at. Let's look at all the available
2066 * connected routes on this node and if any of them auto match
2067 * the routes nexthops ifindex that is good enough for a match
2069 * This code is depending on the fact that a nexthop->ifindex is 0
2070 * if it is not known, if this assumption changes, yummy!
2071 * Additionally a ifindx of 0 means figure it out for us.
2073 static struct route_entry
*
2074 zebra_nhg_connected_ifindex(struct route_node
*rn
, struct route_entry
*match
,
2075 int32_t curr_ifindex
)
2077 struct nexthop
*newhop
= match
->nhe
->nhg
.nexthop
;
2078 struct route_entry
*re
;
2080 assert(newhop
); /* What a kick in the patooey */
2082 if (curr_ifindex
== 0)
2085 if (curr_ifindex
== newhop
->ifindex
)
2089 * At this point we know that this route is matching a connected
2090 * but there are possibly a bunch of connected routes that are
2091 * alive that should be considered as well. So let's iterate over
2092 * all the re's and see if they are connected as well and maybe one
2093 * of those ifindexes match as well.
2095 RNODE_FOREACH_RE (rn
, re
) {
2096 if (re
->type
!= ZEBRA_ROUTE_CONNECT
)
2099 if (CHECK_FLAG(re
->status
, ROUTE_ENTRY_REMOVED
))
2103 * zebra has a connected route that is not removed
2104 * let's test if it is good
2106 newhop
= re
->nhe
->nhg
.nexthop
;
2108 if (curr_ifindex
== newhop
->ifindex
)
2116 * Given a nexthop we need to properly recursively resolve,
2117 * do a table lookup to find and match if at all possible.
2118 * Set the nexthop->ifindex and resolution info as appropriate.
2120 static int nexthop_active(struct nexthop
*nexthop
, struct nhg_hash_entry
*nhe
,
2121 const struct prefix
*top
, int type
, uint32_t flags
,
2122 uint32_t *pmtu
, vrf_id_t vrf_id
)
2125 struct route_table
*table
;
2126 struct route_node
*rn
;
2127 struct route_entry
*match
= NULL
;
2129 struct zebra_nhlfe
*nhlfe
;
2130 struct nexthop
*newhop
;
2131 struct interface
*ifp
;
2133 struct zebra_vrf
*zvrf
;
2134 struct in_addr local_ipv4
;
2135 struct in_addr
*ipv4
;
2138 /* Reset some nexthop attributes that we'll recompute if necessary */
2139 if ((nexthop
->type
== NEXTHOP_TYPE_IPV4
)
2140 || (nexthop
->type
== NEXTHOP_TYPE_IPV6
))
2141 nexthop
->ifindex
= 0;
2143 UNSET_FLAG(nexthop
->flags
, NEXTHOP_FLAG_RECURSIVE
);
2144 nexthops_free(nexthop
->resolved
);
2145 nexthop
->resolved
= NULL
;
2148 * Set afi based on nexthop type.
2149 * Some nexthop types get special handling, possibly skipping
2150 * the normal processing.
2152 switch (nexthop
->type
) {
2153 case NEXTHOP_TYPE_IFINDEX
:
2155 ifp
= if_lookup_by_index(nexthop
->ifindex
, nexthop
->vrf_id
);
2157 * If the interface exists and its operative or its a kernel
2158 * route and interface is up, its active. We trust kernel routes
2161 if (ifp
&& (if_is_operative(ifp
)))
2167 case NEXTHOP_TYPE_IPV6_IFINDEX
:
2170 if (IN6_IS_ADDR_LINKLOCAL(&nexthop
->gate
.ipv6
)) {
2171 ifp
= if_lookup_by_index(nexthop
->ifindex
,
2173 if (ifp
&& if_is_operative(ifp
))
2180 case NEXTHOP_TYPE_IPV4
:
2181 case NEXTHOP_TYPE_IPV4_IFINDEX
:
2184 case NEXTHOP_TYPE_IPV6
:
2188 case NEXTHOP_TYPE_BLACKHOLE
:
2193 * If the nexthop has been marked as 'onlink' we just need to make
2194 * sure the nexthop's interface is known and is operational.
2196 if (CHECK_FLAG(nexthop
->flags
, NEXTHOP_FLAG_ONLINK
)) {
2197 ifp
= if_lookup_by_index(nexthop
->ifindex
, nexthop
->vrf_id
);
2199 if (IS_ZEBRA_DEBUG_RIB_DETAILED
)
2200 zlog_debug("nexthop %pNHv marked onlink but nhif %u doesn't exist",
2201 nexthop
, nexthop
->ifindex
);
2204 if (!if_is_operative(ifp
)) {
2205 if (IS_ZEBRA_DEBUG_RIB_DETAILED
)
2206 zlog_debug("nexthop %pNHv marked onlink but nhif %s is not operational",
2207 nexthop
, ifp
->name
);
2214 ((top
->family
== AF_INET
&& top
->prefixlen
== IPV4_MAX_BITLEN
&&
2215 nexthop
->gate
.ipv4
.s_addr
== top
->u
.prefix4
.s_addr
) ||
2216 (top
->family
== AF_INET6
&& top
->prefixlen
== IPV6_MAX_BITLEN
&&
2217 memcmp(&nexthop
->gate
.ipv6
, &top
->u
.prefix6
, IPV6_MAX_BYTELEN
) ==
2219 nexthop
->vrf_id
== vrf_id
) {
2220 if (IS_ZEBRA_DEBUG_RIB_DETAILED
)
2222 " :%s: Attempting to install a max prefixlength route through itself",
2227 /* Validation for ipv4 mapped ipv6 nexthop. */
2228 if (IS_MAPPED_IPV6(&nexthop
->gate
.ipv6
)) {
2231 ipv4_mapped_ipv6_to_ipv4(&nexthop
->gate
.ipv6
, ipv4
);
2233 ipv4
= &nexthop
->gate
.ipv4
;
2236 /* Processing for nexthops with SR 'color' attribute, using
2237 * the corresponding SR policy object.
2239 if (nexthop
->srte_color
) {
2240 struct ipaddr endpoint
= {0};
2241 struct zebra_sr_policy
*policy
;
2245 endpoint
.ipa_type
= IPADDR_V4
;
2246 endpoint
.ipaddr_v4
= *ipv4
;
2249 endpoint
.ipa_type
= IPADDR_V6
;
2250 endpoint
.ipaddr_v6
= nexthop
->gate
.ipv6
;
2253 flog_err(EC_LIB_DEVELOPMENT
,
2254 "%s: unknown address-family: %u", __func__
,
2259 policy
= zebra_sr_policy_find(nexthop
->srte_color
, &endpoint
);
2260 if (policy
&& policy
->status
== ZEBRA_SR_POLICY_UP
) {
2262 frr_each_safe (nhlfe_list
, &policy
->lsp
->nhlfe_list
,
2264 if (!CHECK_FLAG(nhlfe
->flags
,
2265 NHLFE_FLAG_SELECTED
)
2266 || CHECK_FLAG(nhlfe
->flags
,
2267 NHLFE_FLAG_DELETED
))
2269 SET_FLAG(nexthop
->flags
,
2270 NEXTHOP_FLAG_RECURSIVE
);
2271 nexthop_set_resolved(afi
, nhlfe
->nexthop
,
2280 /* Make lookup prefix. */
2281 memset(&p
, 0, sizeof(struct prefix
));
2285 p
.prefixlen
= IPV4_MAX_BITLEN
;
2286 p
.u
.prefix4
= *ipv4
;
2289 p
.family
= AF_INET6
;
2290 p
.prefixlen
= IPV6_MAX_BITLEN
;
2291 p
.u
.prefix6
= nexthop
->gate
.ipv6
;
2294 assert(afi
!= AFI_IP
&& afi
!= AFI_IP6
);
2298 table
= zebra_vrf_table(afi
, SAFI_UNICAST
, nexthop
->vrf_id
);
2300 zvrf
= zebra_vrf_lookup_by_id(nexthop
->vrf_id
);
2301 if (!table
|| !zvrf
) {
2302 if (IS_ZEBRA_DEBUG_RIB_DETAILED
)
2303 zlog_debug(" %s: Table not found", __func__
);
2307 rn
= route_node_match(table
, (struct prefix
*)&p
);
2309 route_unlock_node(rn
);
2311 /* Lookup should halt if we've matched against ourselves ('top',
2312 * if specified) - i.e., we cannot have a nexthop NH1 is
2313 * resolved by a route NH1. The exception is if the route is a
2316 if (prefix_same(&rn
->p
, top
))
2317 if (((afi
== AFI_IP
)
2318 && (rn
->p
.prefixlen
!= IPV4_MAX_BITLEN
))
2319 || ((afi
== AFI_IP6
)
2320 && (rn
->p
.prefixlen
!= IPV6_MAX_BITLEN
))) {
2321 if (IS_ZEBRA_DEBUG_RIB_DETAILED
)
2323 " %s: Matched against ourself and prefix length is not max bit length",
2328 /* Pick up selected route. */
2329 /* However, do not resolve over default route unless explicitly
2332 if (is_default_prefix(&rn
->p
)
2333 && !rnh_resolve_via_default(zvrf
, p
.family
)) {
2334 if (IS_ZEBRA_DEBUG_RIB_DETAILED
)
2336 " :%s: Resolved against default route",
2341 dest
= rib_dest_from_rnode(rn
);
2342 if (dest
&& dest
->selected_fib
2343 && !CHECK_FLAG(dest
->selected_fib
->status
,
2344 ROUTE_ENTRY_REMOVED
)
2345 && dest
->selected_fib
->type
!= ZEBRA_ROUTE_TABLE
)
2346 match
= dest
->selected_fib
;
2348 /* If there is no selected route or matched route is EGP, go up
2354 } while (rn
&& rn
->info
== NULL
);
2356 route_lock_node(rn
);
2361 if ((match
->type
== ZEBRA_ROUTE_CONNECT
) ||
2362 (RIB_SYSTEM_ROUTE(match
) && RSYSTEM_ROUTE(type
))) {
2363 match
= zebra_nhg_connected_ifindex(rn
, match
,
2366 newhop
= match
->nhe
->nhg
.nexthop
;
2367 if (nexthop
->type
== NEXTHOP_TYPE_IPV4
||
2368 nexthop
->type
== NEXTHOP_TYPE_IPV6
)
2369 nexthop
->ifindex
= newhop
->ifindex
;
2370 else if (nexthop
->ifindex
!= newhop
->ifindex
) {
2371 if (IS_ZEBRA_DEBUG_RIB_DETAILED
)
2373 "%s: %pNHv given ifindex does not match nexthops ifindex found: %pNHv",
2374 __func__
, nexthop
, newhop
);
2376 * NEXTHOP_TYPE_*_IFINDEX but ifindex
2377 * doesn't match what we found.
2382 if (IS_ZEBRA_DEBUG_NHG_DETAIL
)
2384 "%s: CONNECT match %p (%pNG), newhop %pNHv",
2385 __func__
, match
, match
->nhe
, newhop
);
2388 } else if (CHECK_FLAG(flags
, ZEBRA_FLAG_ALLOW_RECURSION
)) {
2389 struct nexthop_group
*nhg
;
2390 struct nexthop
*resolver
;
2391 struct backup_nh_map_s map
= {};
2396 * Only useful if installed or being Route Replacing
2397 * Why Being Route Replaced as well?
2398 * Imagine a route A and route B( that depends on A )
2399 * for recursive resolution and A already exists in the
2400 * zebra rib. If zebra receives the routes
2401 * for resolution at aproximately the same time in the [
2402 * B, A ] order on the workQ. If this happens then
2403 * normal route resolution will happen and B will be
2404 * resolved successfully and then A will be resolved
2405 * successfully. Now imagine the reversed order [A, B].
2406 * A will be resolved and then scheduled for installed
2407 * (Thus not having the ROUTE_ENTRY_INSTALLED flag ). B
2408 * will then get resolved and fail to be installed
2409 * because the original below test. Let's `loosen` this
2410 * up a tiny bit and allow the
2411 * ROUTE_ENTRY_ROUTE_REPLACING flag ( that is set when a
2412 * Route Replace operation is being initiated on A now )
2413 * to now satisfy this situation. This will allow
2414 * either order in the workQ to work properly.
2416 if (!CHECK_FLAG(match
->status
, ROUTE_ENTRY_INSTALLED
) &&
2417 !CHECK_FLAG(match
->status
,
2418 ROUTE_ENTRY_ROUTE_REPLACING
)) {
2419 if (IS_ZEBRA_DEBUG_RIB_DETAILED
)
2421 "%s: match %p (%pNG) not installed or being Route Replaced",
2422 __func__
, match
, match
->nhe
);
2424 goto done_with_match
;
2427 /* Examine installed nexthops; note that there
2428 * may not be any installed primary nexthops if
2429 * only backups are installed.
2431 nhg
= rib_get_fib_nhg(match
);
2432 for (ALL_NEXTHOPS_PTR(nhg
, newhop
)) {
2433 if (!nexthop_valid_resolve(nexthop
, newhop
))
2436 if (IS_ZEBRA_DEBUG_NHG_DETAIL
)
2438 "%s: RECURSIVE match %p (%pNG), newhop %pNHv",
2439 __func__
, match
, match
->nhe
,
2442 SET_FLAG(nexthop
->flags
,
2443 NEXTHOP_FLAG_RECURSIVE
);
2444 resolver
= nexthop_set_resolved(afi
, newhop
,
2448 /* If there are backup nexthops, capture
2449 * that info with the resolving nexthop.
2451 if (resolver
&& newhop
->backup_num
> 0) {
2452 resolve_backup_nexthops(newhop
,
2459 /* Examine installed backup nexthops, if any. There
2460 * are only installed backups *if* there is a
2461 * dedicated fib list. The UI can also control use
2462 * of backups for resolution.
2464 nhg
= rib_get_fib_backup_nhg(match
);
2465 if (!use_recursive_backups
||
2466 nhg
== NULL
|| nhg
->nexthop
== NULL
)
2467 goto done_with_match
;
2469 for (ALL_NEXTHOPS_PTR(nhg
, newhop
)) {
2470 if (!nexthop_valid_resolve(nexthop
, newhop
))
2473 if (IS_ZEBRA_DEBUG_NHG_DETAIL
)
2475 "%s: RECURSIVE match backup %p (%pNG), newhop %pNHv",
2476 __func__
, match
, match
->nhe
,
2479 SET_FLAG(nexthop
->flags
,
2480 NEXTHOP_FLAG_RECURSIVE
);
2481 nexthop_set_resolved(afi
, newhop
, nexthop
,
2487 /* Capture resolving mtu */
2492 } else if (IS_ZEBRA_DEBUG_RIB_DETAILED
)
2494 " %s: Recursion failed to find",
2499 if (IS_ZEBRA_DEBUG_RIB_DETAILED
) {
2501 " %s: Route Type %s has not turned on recursion",
2502 __func__
, zebra_route_string(type
));
2503 if (type
== ZEBRA_ROUTE_BGP
2504 && !CHECK_FLAG(flags
, ZEBRA_FLAG_IBGP
))
2506 " EBGP: see \"disable-ebgp-connected-route-check\" or \"disable-connected-check\"");
2511 if (IS_ZEBRA_DEBUG_RIB_DETAILED
)
2512 zlog_debug(" %s: Nexthop did not lookup in table",
2517 /* This function verifies reachability of one given nexthop, which can be
2518 * numbered or unnumbered, IPv4 or IPv6. The result is unconditionally stored
2519 * in nexthop->flags field. The nexthop->ifindex will be updated
2520 * appropriately as well.
2522 * An existing route map can turn an otherwise active nexthop into inactive,
2523 * but not vice versa.
2525 * The return value is the final value of 'ACTIVE' flag.
2527 static unsigned nexthop_active_check(struct route_node
*rn
,
2528 struct route_entry
*re
,
2529 struct nexthop
*nexthop
,
2530 struct nhg_hash_entry
*nhe
)
2532 route_map_result_t ret
= RMAP_PERMITMATCH
;
2534 const struct prefix
*p
, *src_p
;
2535 struct zebra_vrf
*zvrf
;
2539 srcdest_rnode_prefixes(rn
, &p
, &src_p
);
2541 if (rn
->p
.family
== AF_INET
)
2543 else if (rn
->p
.family
== AF_INET6
)
2548 if (IS_ZEBRA_DEBUG_NHG_DETAIL
)
2549 zlog_debug("%s: re %p, nexthop %pNHv", __func__
, re
, nexthop
);
2552 * If this is a kernel route, then if the interface is *up* then
2553 * by golly gee whiz it's a good route.
2555 if (re
->type
== ZEBRA_ROUTE_KERNEL
|| re
->type
== ZEBRA_ROUTE_SYSTEM
) {
2556 struct interface
*ifp
;
2558 ifp
= if_lookup_by_index(nexthop
->ifindex
, nexthop
->vrf_id
);
2560 if (ifp
&& (if_is_operative(ifp
) || if_is_up(ifp
))) {
2561 SET_FLAG(nexthop
->flags
, NEXTHOP_FLAG_ACTIVE
);
2566 vrf_id
= zvrf_id(rib_dest_vrf(rib_dest_from_rnode(rn
)));
2567 switch (nexthop
->type
) {
2568 case NEXTHOP_TYPE_IFINDEX
:
2569 if (nexthop_active(nexthop
, nhe
, &rn
->p
, re
->type
, re
->flags
,
2571 SET_FLAG(nexthop
->flags
, NEXTHOP_FLAG_ACTIVE
);
2573 UNSET_FLAG(nexthop
->flags
, NEXTHOP_FLAG_ACTIVE
);
2575 case NEXTHOP_TYPE_IPV4
:
2576 case NEXTHOP_TYPE_IPV4_IFINDEX
:
2578 if (nexthop_active(nexthop
, nhe
, &rn
->p
, re
->type
, re
->flags
,
2580 SET_FLAG(nexthop
->flags
, NEXTHOP_FLAG_ACTIVE
);
2582 UNSET_FLAG(nexthop
->flags
, NEXTHOP_FLAG_ACTIVE
);
2584 case NEXTHOP_TYPE_IPV6
:
2586 if (nexthop_active(nexthop
, nhe
, &rn
->p
, re
->type
, re
->flags
,
2588 SET_FLAG(nexthop
->flags
, NEXTHOP_FLAG_ACTIVE
);
2590 UNSET_FLAG(nexthop
->flags
, NEXTHOP_FLAG_ACTIVE
);
2592 case NEXTHOP_TYPE_IPV6_IFINDEX
:
2593 /* RFC 5549, v4 prefix with v6 NH */
2594 if (rn
->p
.family
!= AF_INET
)
2597 if (nexthop_active(nexthop
, nhe
, &rn
->p
, re
->type
, re
->flags
,
2599 SET_FLAG(nexthop
->flags
, NEXTHOP_FLAG_ACTIVE
);
2601 UNSET_FLAG(nexthop
->flags
, NEXTHOP_FLAG_ACTIVE
);
2603 case NEXTHOP_TYPE_BLACKHOLE
:
2604 SET_FLAG(nexthop
->flags
, NEXTHOP_FLAG_ACTIVE
);
2612 if (!CHECK_FLAG(nexthop
->flags
, NEXTHOP_FLAG_ACTIVE
)) {
2613 if (IS_ZEBRA_DEBUG_RIB_DETAILED
)
2614 zlog_debug(" %s: Unable to find active nexthop",
2619 /* Capture recursive nexthop mtu.
2620 * TODO -- the code used to just reset the re's value to zero
2621 * for each nexthop, and then jam any resolving route's mtu value in,
2622 * whether or not that was zero, or lt/gt any existing value? The
2623 * way this is used appears to be as a floor value, so let's try
2624 * using it that way here.
2627 if (re
->nexthop_mtu
== 0 || re
->nexthop_mtu
> mtu
)
2628 re
->nexthop_mtu
= mtu
;
2631 /* XXX: What exactly do those checks do? Do we support
2632 * e.g. IPv4 routes with IPv6 nexthops or vice versa?
2634 if (RIB_SYSTEM_ROUTE(re
) || (family
== AFI_IP
&& p
->family
!= AF_INET
)
2635 || (family
== AFI_IP6
&& p
->family
!= AF_INET6
))
2636 return CHECK_FLAG(nexthop
->flags
, NEXTHOP_FLAG_ACTIVE
);
2638 /* The original code didn't determine the family correctly
2639 * e.g. for NEXTHOP_TYPE_IFINDEX. Retrieve the correct afi
2640 * from the rib_table_info in those cases.
2641 * Possibly it may be better to use only the rib_table_info
2645 struct rib_table_info
*info
;
2647 info
= srcdest_rnode_table_info(rn
);
2651 memset(&nexthop
->rmap_src
.ipv6
, 0, sizeof(union g_addr
));
2653 zvrf
= zebra_vrf_lookup_by_id(re
->vrf_id
);
2655 if (IS_ZEBRA_DEBUG_RIB_DETAILED
)
2656 zlog_debug(" %s: zvrf is NULL", __func__
);
2657 return CHECK_FLAG(nexthop
->flags
, NEXTHOP_FLAG_ACTIVE
);
2660 /* It'll get set if required inside */
2661 ret
= zebra_route_map_check(family
, re
->type
, re
->instance
, p
, nexthop
,
2663 if (ret
== RMAP_DENYMATCH
) {
2664 if (IS_ZEBRA_DEBUG_RIB
) {
2666 "%u:%pRN: Filtering out with NH %pNHv due to route map",
2667 re
->vrf_id
, rn
, nexthop
);
2669 UNSET_FLAG(nexthop
->flags
, NEXTHOP_FLAG_ACTIVE
);
2671 return CHECK_FLAG(nexthop
->flags
, NEXTHOP_FLAG_ACTIVE
);
2674 /* Helper function called after resolution to walk nhg rb trees
2675 * and toggle the NEXTHOP_GROUP_VALID flag if the nexthop
2676 * is active on singleton NHEs.
2678 static bool zebra_nhg_set_valid_if_active(struct nhg_hash_entry
*nhe
)
2680 struct nhg_connected
*rb_node_dep
= NULL
;
2683 if (!zebra_nhg_depends_is_empty(nhe
)) {
2684 /* Is at least one depend valid? */
2685 frr_each(nhg_connected_tree
, &nhe
->nhg_depends
, rb_node_dep
) {
2686 if (zebra_nhg_set_valid_if_active(rb_node_dep
->nhe
))
2693 /* should be fully resolved singleton at this point */
2694 if (CHECK_FLAG(nhe
->nhg
.nexthop
->flags
, NEXTHOP_FLAG_ACTIVE
))
2699 SET_FLAG(nhe
->flags
, NEXTHOP_GROUP_VALID
);
2705 * Process a list of nexthops, given an nhe, determining
2706 * whether each one is ACTIVE/installable at this time.
2708 static uint32_t nexthop_list_active_update(struct route_node
*rn
,
2709 struct route_entry
*re
,
2710 struct nhg_hash_entry
*nhe
,
2713 union g_addr prev_src
;
2714 unsigned int prev_active
, new_active
;
2715 ifindex_t prev_index
;
2716 uint32_t counter
= 0;
2717 struct nexthop
*nexthop
;
2718 struct nexthop_group
*nhg
= &nhe
->nhg
;
2720 nexthop
= nhg
->nexthop
;
2722 /* Init recursive nh mtu */
2723 re
->nexthop_mtu
= 0;
2725 /* Process nexthops one-by-one */
2726 for ( ; nexthop
; nexthop
= nexthop
->next
) {
2728 /* No protocol daemon provides src and so we're skipping
2731 prev_src
= nexthop
->rmap_src
;
2732 prev_active
= CHECK_FLAG(nexthop
->flags
, NEXTHOP_FLAG_ACTIVE
);
2733 prev_index
= nexthop
->ifindex
;
2735 /* Include the containing nhe for primary nexthops: if there's
2736 * recursive resolution, we capture the backup info also.
2739 nexthop_active_check(rn
, re
, nexthop
,
2740 (is_backup
? NULL
: nhe
));
2743 * We need to respect the multipath_num here
2744 * as that what we should be able to install from
2745 * a multipath perspective should not be a data plane
2748 if (new_active
&& counter
>= zrouter
.multipath_num
) {
2751 /* Set it and its resolved nexthop as inactive. */
2752 for (nh
= nexthop
; nh
; nh
= nh
->resolved
)
2753 UNSET_FLAG(nh
->flags
, NEXTHOP_FLAG_ACTIVE
);
2761 /* Check for changes to the nexthop - set ROUTE_ENTRY_CHANGED */
2762 if (prev_active
!= new_active
|| prev_index
!= nexthop
->ifindex
2763 || ((nexthop
->type
>= NEXTHOP_TYPE_IFINDEX
2764 && nexthop
->type
< NEXTHOP_TYPE_IPV6
)
2765 && prev_src
.ipv4
.s_addr
2766 != nexthop
->rmap_src
.ipv4
.s_addr
)
2767 || ((nexthop
->type
>= NEXTHOP_TYPE_IPV6
2768 && nexthop
->type
< NEXTHOP_TYPE_BLACKHOLE
)
2769 && !(IPV6_ADDR_SAME(&prev_src
.ipv6
,
2770 &nexthop
->rmap_src
.ipv6
)))
2771 || CHECK_FLAG(re
->status
, ROUTE_ENTRY_LABELS_CHANGED
))
2772 SET_FLAG(re
->status
, ROUTE_ENTRY_CHANGED
);
2779 static uint32_t proto_nhg_nexthop_active_update(struct nexthop_group
*nhg
)
2782 uint32_t curr_active
= 0;
2784 /* Assume all active for now */
2786 for (nh
= nhg
->nexthop
; nh
; nh
= nh
->next
) {
2787 SET_FLAG(nh
->flags
, NEXTHOP_FLAG_ACTIVE
);
2795 * Iterate over all nexthops of the given RIB entry and refresh their
2796 * ACTIVE flag. If any nexthop is found to toggle the ACTIVE flag,
2797 * the whole re structure is flagged with ROUTE_ENTRY_CHANGED.
2799 * Return value is the new number of active nexthops.
2801 int nexthop_active_update(struct route_node
*rn
, struct route_entry
*re
)
2803 struct nhg_hash_entry
*curr_nhe
;
2804 uint32_t curr_active
= 0, backup_active
= 0;
2806 if (PROTO_OWNED(re
->nhe
))
2807 return proto_nhg_nexthop_active_update(&re
->nhe
->nhg
);
2809 afi_t rt_afi
= family2afi(rn
->p
.family
);
2811 UNSET_FLAG(re
->status
, ROUTE_ENTRY_CHANGED
);
2813 /* Make a local copy of the existing nhe, so we don't work on/modify
2816 curr_nhe
= zebra_nhe_copy(re
->nhe
, re
->nhe
->id
);
2818 if (IS_ZEBRA_DEBUG_NHG_DETAIL
)
2819 zlog_debug("%s: re %p nhe %p (%pNG), curr_nhe %p", __func__
, re
,
2820 re
->nhe
, re
->nhe
, curr_nhe
);
2822 /* Clear the existing id, if any: this will avoid any confusion
2823 * if the id exists, and will also force the creation
2824 * of a new nhe reflecting the changes we may make in this local copy.
2828 /* Process nexthops */
2829 curr_active
= nexthop_list_active_update(rn
, re
, curr_nhe
, false);
2831 if (IS_ZEBRA_DEBUG_NHG_DETAIL
)
2832 zlog_debug("%s: re %p curr_active %u", __func__
, re
,
2835 /* If there are no backup nexthops, we are done */
2836 if (zebra_nhg_get_backup_nhg(curr_nhe
) == NULL
)
2839 backup_active
= nexthop_list_active_update(
2840 rn
, re
, curr_nhe
->backup_info
->nhe
, true /*is_backup*/);
2842 if (IS_ZEBRA_DEBUG_NHG_DETAIL
)
2843 zlog_debug("%s: re %p backup_active %u", __func__
, re
,
2849 * Ref or create an nhe that matches the current state of the
2852 if (CHECK_FLAG(re
->status
, ROUTE_ENTRY_CHANGED
)) {
2853 struct nhg_hash_entry
*new_nhe
= NULL
;
2855 new_nhe
= zebra_nhg_rib_find_nhe(curr_nhe
, rt_afi
);
2857 if (IS_ZEBRA_DEBUG_NHG_DETAIL
)
2859 "%s: re %p CHANGED: nhe %p (%pNG) => new_nhe %p (%pNG)",
2860 __func__
, re
, re
->nhe
, re
->nhe
, new_nhe
,
2863 route_entry_update_nhe(re
, new_nhe
);
2867 /* Walk the NHE depends tree and toggle NEXTHOP_GROUP_VALID
2868 * flag where appropriate.
2871 zebra_nhg_set_valid_if_active(re
->nhe
);
2874 * Do not need the old / copied nhe anymore since it
2875 * was either copied over into a new nhe or not
2878 zebra_nhg_free(curr_nhe
);
2882 /* Recursively construct a grp array of fully resolved IDs.
2884 * This function allows us to account for groups within groups,
2885 * by converting them into a flat array of IDs.
2887 * nh_grp is modified at every level of recursion to append
2888 * to it the next unique, fully resolved ID from the entire tree.
2892 * I'm pretty sure we only allow ONE level of group within group currently.
2893 * But making this recursive just in case that ever changes.
2895 static uint8_t zebra_nhg_nhe2grp_internal(struct nh_grp
*grp
,
2897 struct nhg_hash_entry
*nhe
,
2900 struct nhg_connected
*rb_node_dep
= NULL
;
2901 struct nhg_hash_entry
*depend
= NULL
;
2902 uint8_t i
= curr_index
;
2904 frr_each(nhg_connected_tree
, &nhe
->nhg_depends
, rb_node_dep
) {
2905 bool duplicate
= false;
2910 depend
= rb_node_dep
->nhe
;
2913 * If its recursive, use its resolved nhe in the group
2915 if (CHECK_FLAG(depend
->flags
, NEXTHOP_GROUP_RECURSIVE
)) {
2916 depend
= zebra_nhg_resolve(depend
);
2919 EC_ZEBRA_NHG_FIB_UPDATE
,
2920 "Failed to recursively resolve Nexthop Hash Entry in the group id=%pNG",
2926 if (!zebra_nhg_depends_is_empty(depend
)) {
2927 /* This is a group within a group */
2928 i
= zebra_nhg_nhe2grp_internal(grp
, i
, depend
, max_num
);
2930 if (!CHECK_FLAG(depend
->flags
, NEXTHOP_GROUP_VALID
)) {
2931 if (IS_ZEBRA_DEBUG_RIB_DETAILED
2932 || IS_ZEBRA_DEBUG_NHG
)
2934 "%s: Nexthop ID (%u) not valid, not appending to dataplane install group",
2935 __func__
, depend
->id
);
2939 /* If the nexthop not installed/queued for install don't
2940 * put in the ID array.
2942 if (!(CHECK_FLAG(depend
->flags
, NEXTHOP_GROUP_INSTALLED
)
2943 || CHECK_FLAG(depend
->flags
,
2944 NEXTHOP_GROUP_QUEUED
))) {
2945 if (IS_ZEBRA_DEBUG_RIB_DETAILED
2946 || IS_ZEBRA_DEBUG_NHG
)
2948 "%s: Nexthop ID (%u) not installed or queued for install, not appending to dataplane install group",
2949 __func__
, depend
->id
);
2953 /* Check for duplicate IDs, ignore if found. */
2954 for (int j
= 0; j
< i
; j
++) {
2955 if (depend
->id
== grp
[j
].id
) {
2962 if (IS_ZEBRA_DEBUG_RIB_DETAILED
2963 || IS_ZEBRA_DEBUG_NHG
)
2965 "%s: Nexthop ID (%u) is duplicate, not appending to dataplane install group",
2966 __func__
, depend
->id
);
2970 grp
[i
].id
= depend
->id
;
2971 grp
[i
].weight
= depend
->nhg
.nexthop
->weight
;
2976 if (nhe
->backup_info
== NULL
|| nhe
->backup_info
->nhe
== NULL
)
2979 /* TODO -- For now, we are not trying to use or install any
2980 * backup info in this nexthop-id path: we aren't prepared
2981 * to use the backups here yet. We're just debugging what we find.
2983 if (IS_ZEBRA_DEBUG_NHG_DETAIL
)
2984 zlog_debug("%s: skipping backup nhe", __func__
);
2990 /* Convert a nhe into a group array */
2991 uint8_t zebra_nhg_nhe2grp(struct nh_grp
*grp
, struct nhg_hash_entry
*nhe
,
2994 /* Call into the recursive function */
2995 return zebra_nhg_nhe2grp_internal(grp
, 0, nhe
, max_num
);
2998 void zebra_nhg_install_kernel(struct nhg_hash_entry
*nhe
)
3000 struct nhg_connected
*rb_node_dep
= NULL
;
3002 /* Resolve it first */
3003 nhe
= zebra_nhg_resolve(nhe
);
3005 /* Make sure all depends are installed/queued */
3006 frr_each(nhg_connected_tree
, &nhe
->nhg_depends
, rb_node_dep
) {
3007 zebra_nhg_install_kernel(rb_node_dep
->nhe
);
3010 if (CHECK_FLAG(nhe
->flags
, NEXTHOP_GROUP_VALID
)
3011 && !CHECK_FLAG(nhe
->flags
, NEXTHOP_GROUP_INSTALLED
)
3012 && !CHECK_FLAG(nhe
->flags
, NEXTHOP_GROUP_QUEUED
)) {
3013 /* Change its type to us since we are installing it */
3014 if (!ZEBRA_NHG_CREATED(nhe
))
3015 nhe
->type
= ZEBRA_ROUTE_NHG
;
3017 int ret
= dplane_nexthop_add(nhe
);
3020 case ZEBRA_DPLANE_REQUEST_QUEUED
:
3021 SET_FLAG(nhe
->flags
, NEXTHOP_GROUP_QUEUED
);
3023 case ZEBRA_DPLANE_REQUEST_FAILURE
:
3025 EC_ZEBRA_DP_INSTALL_FAIL
,
3026 "Failed to install Nexthop ID (%pNG) into the kernel",
3029 case ZEBRA_DPLANE_REQUEST_SUCCESS
:
3030 SET_FLAG(nhe
->flags
, NEXTHOP_GROUP_INSTALLED
);
3031 zebra_nhg_handle_install(nhe
);
3037 void zebra_nhg_uninstall_kernel(struct nhg_hash_entry
*nhe
)
3039 if (CHECK_FLAG(nhe
->flags
, NEXTHOP_GROUP_INSTALLED
)) {
3040 int ret
= dplane_nexthop_delete(nhe
);
3043 case ZEBRA_DPLANE_REQUEST_QUEUED
:
3044 SET_FLAG(nhe
->flags
, NEXTHOP_GROUP_QUEUED
);
3046 case ZEBRA_DPLANE_REQUEST_FAILURE
:
3048 EC_ZEBRA_DP_DELETE_FAIL
,
3049 "Failed to uninstall Nexthop ID (%pNG) from the kernel",
3052 case ZEBRA_DPLANE_REQUEST_SUCCESS
:
3053 UNSET_FLAG(nhe
->flags
, NEXTHOP_GROUP_INSTALLED
);
3058 zebra_nhg_handle_uninstall(nhe
);
3061 void zebra_nhg_dplane_result(struct zebra_dplane_ctx
*ctx
)
3063 enum dplane_op_e op
;
3064 enum zebra_dplane_result status
;
3066 struct nhg_hash_entry
*nhe
= NULL
;
3068 op
= dplane_ctx_get_op(ctx
);
3069 status
= dplane_ctx_get_status(ctx
);
3071 id
= dplane_ctx_get_nhe_id(ctx
);
3073 if (IS_ZEBRA_DEBUG_DPLANE_DETAIL
|| IS_ZEBRA_DEBUG_NHG_DETAIL
)
3075 "Nexthop dplane ctx %p, op %s, nexthop ID (%u), result %s",
3076 ctx
, dplane_op2str(op
), id
, dplane_res2str(status
));
3079 case DPLANE_OP_NH_DELETE
:
3080 if (status
!= ZEBRA_DPLANE_REQUEST_SUCCESS
)
3082 EC_ZEBRA_DP_DELETE_FAIL
,
3083 "Failed to uninstall Nexthop ID (%u) from the kernel",
3086 /* We already free'd the data, nothing to do */
3088 case DPLANE_OP_NH_INSTALL
:
3089 case DPLANE_OP_NH_UPDATE
:
3090 nhe
= zebra_nhg_lookup_id(id
);
3093 if (IS_ZEBRA_DEBUG_NHG
)
3095 "%s operation preformed on Nexthop ID (%u) in the kernel, that we no longer have in our table",
3096 dplane_op2str(op
), id
);
3101 UNSET_FLAG(nhe
->flags
, NEXTHOP_GROUP_QUEUED
);
3102 if (status
== ZEBRA_DPLANE_REQUEST_SUCCESS
) {
3103 SET_FLAG(nhe
->flags
, NEXTHOP_GROUP_VALID
);
3104 SET_FLAG(nhe
->flags
, NEXTHOP_GROUP_INSTALLED
);
3105 zebra_nhg_handle_install(nhe
);
3107 /* If daemon nhg, send it an update */
3108 if (PROTO_OWNED(nhe
))
3109 zsend_nhg_notify(nhe
->type
, nhe
->zapi_instance
,
3110 nhe
->zapi_session
, nhe
->id
,
3111 ZAPI_NHG_INSTALLED
);
3113 /* If daemon nhg, send it an update */
3114 if (PROTO_OWNED(nhe
))
3115 zsend_nhg_notify(nhe
->type
, nhe
->zapi_instance
,
3116 nhe
->zapi_session
, nhe
->id
,
3117 ZAPI_NHG_FAIL_INSTALL
);
3119 if (!(zebra_nhg_proto_nexthops_only() &&
3122 EC_ZEBRA_DP_INSTALL_FAIL
,
3123 "Failed to install Nexthop (%pNG) into the kernel",
3128 case DPLANE_OP_ROUTE_INSTALL
:
3129 case DPLANE_OP_ROUTE_UPDATE
:
3130 case DPLANE_OP_ROUTE_DELETE
:
3131 case DPLANE_OP_ROUTE_NOTIFY
:
3132 case DPLANE_OP_LSP_INSTALL
:
3133 case DPLANE_OP_LSP_UPDATE
:
3134 case DPLANE_OP_LSP_DELETE
:
3135 case DPLANE_OP_LSP_NOTIFY
:
3136 case DPLANE_OP_PW_INSTALL
:
3137 case DPLANE_OP_PW_UNINSTALL
:
3138 case DPLANE_OP_SYS_ROUTE_ADD
:
3139 case DPLANE_OP_SYS_ROUTE_DELETE
:
3140 case DPLANE_OP_ADDR_INSTALL
:
3141 case DPLANE_OP_ADDR_UNINSTALL
:
3142 case DPLANE_OP_MAC_INSTALL
:
3143 case DPLANE_OP_MAC_DELETE
:
3144 case DPLANE_OP_NEIGH_INSTALL
:
3145 case DPLANE_OP_NEIGH_UPDATE
:
3146 case DPLANE_OP_NEIGH_DELETE
:
3147 case DPLANE_OP_NEIGH_IP_INSTALL
:
3148 case DPLANE_OP_NEIGH_IP_DELETE
:
3149 case DPLANE_OP_VTEP_ADD
:
3150 case DPLANE_OP_VTEP_DELETE
:
3151 case DPLANE_OP_RULE_ADD
:
3152 case DPLANE_OP_RULE_DELETE
:
3153 case DPLANE_OP_RULE_UPDATE
:
3154 case DPLANE_OP_NEIGH_DISCOVER
:
3155 case DPLANE_OP_BR_PORT_UPDATE
:
3156 case DPLANE_OP_NONE
:
3157 case DPLANE_OP_IPTABLE_ADD
:
3158 case DPLANE_OP_IPTABLE_DELETE
:
3159 case DPLANE_OP_IPSET_ADD
:
3160 case DPLANE_OP_IPSET_DELETE
:
3161 case DPLANE_OP_IPSET_ENTRY_ADD
:
3162 case DPLANE_OP_IPSET_ENTRY_DELETE
:
3163 case DPLANE_OP_NEIGH_TABLE_UPDATE
:
3164 case DPLANE_OP_GRE_SET
:
3165 case DPLANE_OP_INTF_ADDR_ADD
:
3166 case DPLANE_OP_INTF_ADDR_DEL
:
3167 case DPLANE_OP_INTF_NETCONFIG
:
3168 case DPLANE_OP_INTF_INSTALL
:
3169 case DPLANE_OP_INTF_UPDATE
:
3170 case DPLANE_OP_INTF_DELETE
:
3171 case DPLANE_OP_TC_QDISC_INSTALL
:
3172 case DPLANE_OP_TC_QDISC_UNINSTALL
:
3173 case DPLANE_OP_TC_CLASS_ADD
:
3174 case DPLANE_OP_TC_CLASS_DELETE
:
3175 case DPLANE_OP_TC_CLASS_UPDATE
:
3176 case DPLANE_OP_TC_FILTER_ADD
:
3177 case DPLANE_OP_TC_FILTER_DELETE
:
3178 case DPLANE_OP_TC_FILTER_UPDATE
:
3183 static int zebra_nhg_sweep_entry(struct hash_bucket
*bucket
, void *arg
)
3185 struct nhg_hash_entry
*nhe
= NULL
;
3187 nhe
= (struct nhg_hash_entry
*)bucket
->data
;
3190 * same logic as with routes.
3192 * If older than startup time, we know we read them in from the
3193 * kernel and have not gotten and update for them since startup
3194 * from an upper level proto.
3196 if (zrouter
.startup_time
< nhe
->uptime
)
3197 return HASHWALK_CONTINUE
;
3200 * If it's proto-owned and not being used by a route, remove it since
3201 * we haven't gotten an update about it from the proto since startup.
3202 * This means that either the config for it was removed or the daemon
3203 * didn't get started. This handles graceful restart & retain scenario.
3205 if (PROTO_OWNED(nhe
) && nhe
->refcnt
== 1) {
3206 zebra_nhg_decrement_ref(nhe
);
3207 return HASHWALK_ABORT
;
3211 * If its being ref'd by routes, just let it be uninstalled via a route
3214 if (ZEBRA_NHG_CREATED(nhe
) && nhe
->refcnt
<= 0) {
3215 zebra_nhg_uninstall_kernel(nhe
);
3216 return HASHWALK_ABORT
;
3219 return HASHWALK_CONTINUE
;
3222 void zebra_nhg_sweep_table(struct hash
*hash
)
3227 * Yes this is extremely odd. Effectively nhg's have
3228 * other nexthop groups that depend on them and when you
3229 * remove them, you can have other entries blown up.
3230 * our hash code does not work with deleting multiple
3231 * entries at a time and will possibly cause crashes
3232 * So what to do? Whenever zebra_nhg_sweep_entry
3233 * deletes an entry it will return HASHWALK_ABORT,
3234 * cause that deletion might have triggered more.
3235 * then we can just keep sweeping this table
3236 * until nothing more is found to do.
3239 count
= hashcount(hash
);
3240 hash_walk(hash
, zebra_nhg_sweep_entry
, NULL
);
3241 } while (count
!= hashcount(hash
));
3244 static void zebra_nhg_mark_keep_entry(struct hash_bucket
*bucket
, void *arg
)
3246 struct nhg_hash_entry
*nhe
= bucket
->data
;
3248 UNSET_FLAG(nhe
->flags
, NEXTHOP_GROUP_INSTALLED
);
3252 * When we are shutting down and we have retain mode enabled
3253 * in zebra the process is to mark each vrf that it's
3254 * routes should not be deleted. The problem with that
3255 * is that shutdown actually free's up memory which
3256 * causes the nexthop group's ref counts to go to zero
3257 * we need a way to subtly tell the system to not remove
3258 * the nexthop groups from the kernel at the same time.
3259 * The easiest just looks like that we should not mark
3260 * the nhg's as installed any more and when the ref count
3261 * goes to zero we'll attempt to delete and do nothing
3263 void zebra_nhg_mark_keep(void)
3265 hash_iterate(zrouter
.nhgs_id
, zebra_nhg_mark_keep_entry
, NULL
);
3268 /* Global control to disable use of kernel nexthops, if available. We can't
3269 * force the kernel to support nexthop ids, of course, but we can disable
3270 * zebra's use of them, for testing e.g. By default, if the kernel supports
3271 * nexthop ids, zebra uses them.
3273 void zebra_nhg_enable_kernel_nexthops(bool set
)
3275 g_nexthops_enabled
= set
;
3278 bool zebra_nhg_kernel_nexthops_enabled(void)
3280 return g_nexthops_enabled
;
3283 /* Global control for use of activated backups for recursive resolution. */
3284 void zebra_nhg_set_recursive_use_backups(bool set
)
3286 use_recursive_backups
= set
;
3289 bool zebra_nhg_recursive_use_backups(void)
3291 return use_recursive_backups
;
3295 * Global control to only use kernel nexthops for protocol created NHGs.
3296 * There are some use cases where you may not want zebra to implicitly
3297 * create kernel nexthops for all routes and only create them for NHGs
3298 * passed down by upper level protos.
3302 void zebra_nhg_set_proto_nexthops_only(bool set
)
3304 proto_nexthops_only
= set
;
3307 bool zebra_nhg_proto_nexthops_only(void)
3309 return proto_nexthops_only
;
3312 /* Add NHE from upper level proto */
3313 struct nhg_hash_entry
*zebra_nhg_proto_add(uint32_t id
, int type
,
3314 uint16_t instance
, uint32_t session
,
3315 struct nexthop_group
*nhg
, afi_t afi
)
3317 struct nhg_hash_entry lookup
;
3318 struct nhg_hash_entry
*new, *old
;
3319 struct nhg_connected
*rb_node_dep
= NULL
;
3320 struct nexthop
*newhop
;
3321 bool replace
= false;
3323 if (!nhg
->nexthop
) {
3324 if (IS_ZEBRA_DEBUG_NHG
)
3325 zlog_debug("%s: id %u, no nexthops passed to add",
3331 /* Set nexthop list as active, since they wont go through rib
3334 * Assuming valid/onlink for now.
3336 * Once resolution is figured out, we won't need this!
3338 for (ALL_NEXTHOPS_PTR(nhg
, newhop
)) {
3339 if (CHECK_FLAG(newhop
->flags
, NEXTHOP_FLAG_HAS_BACKUP
)) {
3340 if (IS_ZEBRA_DEBUG_NHG
)
3342 "%s: id %u, backup nexthops not supported",
3347 if (newhop
->type
== NEXTHOP_TYPE_BLACKHOLE
) {
3348 if (IS_ZEBRA_DEBUG_NHG
)
3350 "%s: id %u, blackhole nexthop not supported",
3355 if (newhop
->type
== NEXTHOP_TYPE_IFINDEX
) {
3356 if (IS_ZEBRA_DEBUG_NHG
)
3358 "%s: id %u, nexthop without gateway not supported",
3363 if (!newhop
->ifindex
) {
3364 if (IS_ZEBRA_DEBUG_NHG
)
3366 "%s: id %u, nexthop without ifindex is not supported",
3370 SET_FLAG(newhop
->flags
, NEXTHOP_FLAG_ACTIVE
);
3373 zebra_nhe_init(&lookup
, afi
, nhg
->nexthop
);
3374 lookup
.nhg
.nexthop
= nhg
->nexthop
;
3375 lookup
.nhg
.nhgr
= nhg
->nhgr
;
3379 old
= zebra_nhg_lookup_id(id
);
3383 * This is a replace, just release NHE from ID for now, The
3384 * depends/dependents may still be used in the replacement so
3385 * we don't touch them other than to remove their refs to their
3389 hash_release(zrouter
.nhgs_id
, old
);
3391 /* Free all the things */
3392 zebra_nhg_release_all_deps(old
);
3395 new = zebra_nhg_rib_find_nhe(&lookup
, afi
);
3397 zebra_nhg_increment_ref(new);
3399 /* Capture zapi client info */
3400 new->zapi_instance
= instance
;
3401 new->zapi_session
= session
;
3403 zebra_nhg_set_valid_if_active(new);
3405 zebra_nhg_install_kernel(new);
3409 * Check to handle recving DEL while routes still in use then
3412 * In this case we would have decremented the refcnt already
3413 * but set the FLAG here. Go ahead and increment once to fix
3414 * the misordering we have been sent.
3416 if (CHECK_FLAG(old
->flags
, NEXTHOP_GROUP_PROTO_RELEASED
))
3417 zebra_nhg_increment_ref(old
);
3419 rib_handle_nhg_replace(old
, new);
3421 /* We have to decrement its singletons
3422 * because some might not exist in NEW.
3424 if (!zebra_nhg_depends_is_empty(old
)) {
3425 frr_each (nhg_connected_tree
, &old
->nhg_depends
,
3427 zebra_nhg_decrement_ref(rb_node_dep
->nhe
);
3430 /* Dont call the dec API, we dont want to uninstall the ID */
3432 THREAD_OFF(old
->timer
);
3433 zebra_nhg_free(old
);
3437 if (IS_ZEBRA_DEBUG_NHG_DETAIL
)
3438 zlog_debug("%s: %s nhe %p (%u), vrf %d, type %s", __func__
,
3439 (replace
? "replaced" : "added"), new, new->id
,
3440 new->vrf_id
, zebra_route_string(new->type
));
3445 /* Delete NHE from upper level proto, caller must decrement ref */
3446 struct nhg_hash_entry
*zebra_nhg_proto_del(uint32_t id
, int type
)
3448 struct nhg_hash_entry
*nhe
;
3450 nhe
= zebra_nhg_lookup_id(id
);
3453 if (IS_ZEBRA_DEBUG_NHG
)
3454 zlog_debug("%s: id %u, lookup failed", __func__
, id
);
3459 if (type
!= nhe
->type
) {
3460 if (IS_ZEBRA_DEBUG_NHG
)
3462 "%s: id %u, type %s mismatch, sent by %s, ignoring",
3463 __func__
, id
, zebra_route_string(nhe
->type
),
3464 zebra_route_string(type
));
3468 if (CHECK_FLAG(nhe
->flags
, NEXTHOP_GROUP_PROTO_RELEASED
)) {
3469 if (IS_ZEBRA_DEBUG_NHG
)
3470 zlog_debug("%s: id %u, already released", __func__
, id
);
3475 SET_FLAG(nhe
->flags
, NEXTHOP_GROUP_PROTO_RELEASED
);
3477 if (nhe
->refcnt
> 1) {
3478 if (IS_ZEBRA_DEBUG_NHG
)
3480 "%s: %pNG, still being used by routes refcnt %u",
3481 __func__
, nhe
, nhe
->refcnt
);
3485 if (IS_ZEBRA_DEBUG_NHG_DETAIL
)
3486 zlog_debug("%s: deleted nhe %p (%pNG), vrf %d, type %s",
3487 __func__
, nhe
, nhe
, nhe
->vrf_id
,
3488 zebra_route_string(nhe
->type
));
3493 struct nhg_score_proto_iter
{
3498 static void zebra_nhg_score_proto_entry(struct hash_bucket
*bucket
, void *arg
)
3500 struct nhg_hash_entry
*nhe
;
3501 struct nhg_score_proto_iter
*iter
;
3503 nhe
= (struct nhg_hash_entry
*)bucket
->data
;
3506 /* Needs to match type and outside zebra ID space */
3507 if (nhe
->type
== iter
->type
&& PROTO_OWNED(nhe
)) {
3508 if (IS_ZEBRA_DEBUG_NHG_DETAIL
)
3510 "%s: found nhe %p (%pNG), vrf %d, type %s after client disconnect",
3511 __func__
, nhe
, nhe
, nhe
->vrf_id
,
3512 zebra_route_string(nhe
->type
));
3514 /* Add to removal list */
3515 listnode_add(iter
->found
, nhe
);
3519 /* Remove specific by proto NHGs */
3520 unsigned long zebra_nhg_score_proto(int type
)
3522 struct nhg_hash_entry
*nhe
;
3523 struct nhg_score_proto_iter iter
= {};
3524 struct listnode
*ln
;
3525 unsigned long count
;
3528 iter
.found
= list_new();
3530 /* Find matching entries to remove */
3531 hash_iterate(zrouter
.nhgs_id
, zebra_nhg_score_proto_entry
, &iter
);
3533 /* Now remove them */
3534 for (ALL_LIST_ELEMENTS_RO(iter
.found
, ln
, nhe
)) {
3536 * This should be the last ref if we remove client routes too,
3537 * and thus should remove and free them.
3539 zebra_nhg_decrement_ref(nhe
);
3542 count
= iter
.found
->count
;
3543 list_delete(&iter
.found
);
3548 printfrr_ext_autoreg_p("NG", printfrr_nhghe
);
3549 static ssize_t
printfrr_nhghe(struct fbuf
*buf
, struct printfrr_eargs
*ea
,
3552 const struct nhg_hash_entry
*nhe
= ptr
;
3553 const struct nhg_connected
*dep
;
3557 return bputs(buf
, "[NULL]");
3559 ret
+= bprintfrr(buf
, "%u[", nhe
->id
);
3561 ret
+= printfrr_nhs(buf
, nhe
->nhg
.nexthop
);
3563 int count
= zebra_nhg_depends_count(nhe
);
3565 frr_each (nhg_connected_tree_const
, &nhe
->nhg_depends
, dep
) {
3566 ret
+= bprintfrr(buf
, "%u", dep
->nhe
->id
);
3568 ret
+= bputs(buf
, "/");
3573 ret
+= bputs(buf
, "]");