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"
47 DEFINE_MTYPE_STATIC(ZEBRA
, NHG
, "Nexthop Group Entry");
48 DEFINE_MTYPE_STATIC(ZEBRA
, NHG_CONNECTED
, "Nexthop Group Connected");
49 DEFINE_MTYPE_STATIC(ZEBRA
, NHG_CTX
, "Nexthop Group Context");
51 /* Map backup nexthop indices between two nhes */
52 struct backup_nh_map_s
{
61 /* id counter to keep in sync with kernel */
64 /* Controlled through ui */
65 static bool g_nexthops_enabled
= true;
66 static bool proto_nexthops_only
;
67 static bool use_recursive_backups
= true;
69 static struct nhg_hash_entry
*depends_find(const struct nexthop
*nh
, afi_t afi
,
70 int type
, bool from_dplane
);
71 static void depends_add(struct nhg_connected_tree_head
*head
,
72 struct nhg_hash_entry
*depend
);
73 static struct nhg_hash_entry
*
74 depends_find_add(struct nhg_connected_tree_head
*head
, struct nexthop
*nh
,
75 afi_t afi
, int type
, bool from_dplane
);
76 static struct nhg_hash_entry
*
77 depends_find_id_add(struct nhg_connected_tree_head
*head
, uint32_t id
);
78 static void depends_decrement_free(struct nhg_connected_tree_head
*head
);
80 static struct nhg_backup_info
*
81 nhg_backup_copy(const struct nhg_backup_info
*orig
);
83 /* Helper function for getting the next allocatable ID */
84 static uint32_t nhg_get_next_id(void)
89 if (IS_ZEBRA_DEBUG_NHG_DETAIL
)
90 zlog_debug("%s: ID %u checking", __func__
, id_counter
);
92 if (id_counter
== ZEBRA_NHG_PROTO_LOWER
) {
93 if (IS_ZEBRA_DEBUG_NHG_DETAIL
)
94 zlog_debug("%s: ID counter wrapped", __func__
);
100 if (zebra_nhg_lookup_id(id_counter
)) {
101 if (IS_ZEBRA_DEBUG_NHG_DETAIL
)
102 zlog_debug("%s: ID already exists", __func__
);
113 static void nhg_connected_free(struct nhg_connected
*dep
)
115 XFREE(MTYPE_NHG_CONNECTED
, dep
);
118 static struct nhg_connected
*nhg_connected_new(struct nhg_hash_entry
*nhe
)
120 struct nhg_connected
*new = NULL
;
122 new = XCALLOC(MTYPE_NHG_CONNECTED
, sizeof(struct nhg_connected
));
128 void nhg_connected_tree_free(struct nhg_connected_tree_head
*head
)
130 struct nhg_connected
*rb_node_dep
= NULL
;
132 if (!nhg_connected_tree_is_empty(head
)) {
133 frr_each_safe(nhg_connected_tree
, head
, rb_node_dep
) {
134 nhg_connected_tree_del(head
, rb_node_dep
);
135 nhg_connected_free(rb_node_dep
);
140 bool nhg_connected_tree_is_empty(const struct nhg_connected_tree_head
*head
)
142 return nhg_connected_tree_count(head
) ? false : true;
145 struct nhg_connected
*
146 nhg_connected_tree_root(struct nhg_connected_tree_head
*head
)
148 return nhg_connected_tree_first(head
);
151 struct nhg_hash_entry
*
152 nhg_connected_tree_del_nhe(struct nhg_connected_tree_head
*head
,
153 struct nhg_hash_entry
*depend
)
155 struct nhg_connected lookup
= {};
156 struct nhg_connected
*remove
= NULL
;
157 struct nhg_hash_entry
*removed_nhe
;
161 /* Lookup to find the element, then remove it */
162 remove
= nhg_connected_tree_find(head
, &lookup
);
164 /* Re-returning here just in case this API changes..
165 * the _del list api's are a bit undefined at the moment.
167 * So hopefully returning here will make it fail if the api
168 * changes to something different than currently expected.
170 remove
= nhg_connected_tree_del(head
, remove
);
172 /* If the entry was sucessfully removed, free the 'connected` struct */
174 removed_nhe
= remove
->nhe
;
175 nhg_connected_free(remove
);
182 /* Assuming UNIQUE RB tree. If this changes, assumptions here about
183 * insertion need to change.
185 struct nhg_hash_entry
*
186 nhg_connected_tree_add_nhe(struct nhg_connected_tree_head
*head
,
187 struct nhg_hash_entry
*depend
)
189 struct nhg_connected
*new = NULL
;
191 new = nhg_connected_new(depend
);
193 /* On success, NULL will be returned from the
196 if (new && (nhg_connected_tree_add(head
, new) == NULL
))
199 /* If it wasn't successful, it must be a duplicate. We enforce the
200 * unique property for the `nhg_connected` tree.
202 nhg_connected_free(new);
208 nhg_connected_tree_decrement_ref(struct nhg_connected_tree_head
*head
)
210 struct nhg_connected
*rb_node_dep
= NULL
;
212 frr_each_safe(nhg_connected_tree
, head
, rb_node_dep
) {
213 zebra_nhg_decrement_ref(rb_node_dep
->nhe
);
218 nhg_connected_tree_increment_ref(struct nhg_connected_tree_head
*head
)
220 struct nhg_connected
*rb_node_dep
= NULL
;
222 frr_each(nhg_connected_tree
, head
, rb_node_dep
) {
223 zebra_nhg_increment_ref(rb_node_dep
->nhe
);
227 struct nhg_hash_entry
*zebra_nhg_resolve(struct nhg_hash_entry
*nhe
)
229 if (CHECK_FLAG(nhe
->flags
, NEXTHOP_GROUP_RECURSIVE
)
230 && !zebra_nhg_depends_is_empty(nhe
)) {
231 nhe
= nhg_connected_tree_root(&nhe
->nhg_depends
)->nhe
;
232 return zebra_nhg_resolve(nhe
);
238 unsigned int zebra_nhg_depends_count(const struct nhg_hash_entry
*nhe
)
240 return nhg_connected_tree_count(&nhe
->nhg_depends
);
243 bool zebra_nhg_depends_is_empty(const struct nhg_hash_entry
*nhe
)
245 return nhg_connected_tree_is_empty(&nhe
->nhg_depends
);
248 static void zebra_nhg_depends_del(struct nhg_hash_entry
*from
,
249 struct nhg_hash_entry
*depend
)
251 nhg_connected_tree_del_nhe(&from
->nhg_depends
, depend
);
254 static void zebra_nhg_depends_init(struct nhg_hash_entry
*nhe
)
256 nhg_connected_tree_init(&nhe
->nhg_depends
);
259 unsigned int zebra_nhg_dependents_count(const struct nhg_hash_entry
*nhe
)
261 return nhg_connected_tree_count(&nhe
->nhg_dependents
);
265 bool zebra_nhg_dependents_is_empty(const struct nhg_hash_entry
*nhe
)
267 return nhg_connected_tree_is_empty(&nhe
->nhg_dependents
);
270 static void zebra_nhg_dependents_del(struct nhg_hash_entry
*from
,
271 struct nhg_hash_entry
*dependent
)
273 nhg_connected_tree_del_nhe(&from
->nhg_dependents
, dependent
);
276 static void zebra_nhg_dependents_add(struct nhg_hash_entry
*to
,
277 struct nhg_hash_entry
*dependent
)
279 nhg_connected_tree_add_nhe(&to
->nhg_dependents
, dependent
);
282 static void zebra_nhg_dependents_init(struct nhg_hash_entry
*nhe
)
284 nhg_connected_tree_init(&nhe
->nhg_dependents
);
287 /* Release this nhe from anything depending on it */
288 static void zebra_nhg_dependents_release(struct nhg_hash_entry
*nhe
)
290 struct nhg_connected
*rb_node_dep
= NULL
;
292 frr_each_safe(nhg_connected_tree
, &nhe
->nhg_dependents
, rb_node_dep
) {
293 zebra_nhg_depends_del(rb_node_dep
->nhe
, nhe
);
294 /* recheck validity of the dependent */
295 zebra_nhg_check_valid(rb_node_dep
->nhe
);
299 /* Release this nhe from anything that it depends on */
300 static void zebra_nhg_depends_release(struct nhg_hash_entry
*nhe
)
302 if (!zebra_nhg_depends_is_empty(nhe
)) {
303 struct nhg_connected
*rb_node_dep
= NULL
;
305 frr_each_safe(nhg_connected_tree
, &nhe
->nhg_depends
,
307 zebra_nhg_dependents_del(rb_node_dep
->nhe
, nhe
);
313 struct nhg_hash_entry
*zebra_nhg_lookup_id(uint32_t id
)
315 struct nhg_hash_entry lookup
= {};
318 return hash_lookup(zrouter
.nhgs_id
, &lookup
);
321 static int zebra_nhg_insert_id(struct nhg_hash_entry
*nhe
)
323 if (hash_lookup(zrouter
.nhgs_id
, nhe
)) {
325 EC_ZEBRA_NHG_TABLE_INSERT_FAILED
,
326 "Failed inserting NHG id=%u into the ID hash table, entry already exists",
331 hash_get(zrouter
.nhgs_id
, nhe
, hash_alloc_intern
);
336 static void zebra_nhg_set_if(struct nhg_hash_entry
*nhe
, struct interface
*ifp
)
339 if_nhg_dependents_add(ifp
, nhe
);
343 zebra_nhg_connect_depends(struct nhg_hash_entry
*nhe
,
344 struct nhg_connected_tree_head
*nhg_depends
)
346 struct nhg_connected
*rb_node_dep
= NULL
;
348 /* This has been allocated higher above in the stack. Could probably
349 * re-allocate and free the old stuff but just using the same memory
350 * for now. Otherwise, their might be a time trade-off for repeated
351 * alloc/frees as startup.
353 nhe
->nhg_depends
= *nhg_depends
;
355 /* Attach backpointer to anything that it depends on */
356 zebra_nhg_dependents_init(nhe
);
357 if (!zebra_nhg_depends_is_empty(nhe
)) {
358 frr_each(nhg_connected_tree
, &nhe
->nhg_depends
, rb_node_dep
) {
359 if (IS_ZEBRA_DEBUG_NHG_DETAIL
)
360 zlog_debug("%s: nhe %p (%u), dep %p (%u)",
361 __func__
, nhe
, nhe
->id
,
363 rb_node_dep
->nhe
->id
);
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 id=%u",
473 nhe
->nhg
.nexthop
->ifindex
,
474 nhe
->nhg
.nexthop
->vrf_id
, nhe
->id
);
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 /* Nexthops should be in-order, so we simply compare them in-place */
567 for (nexthop1
= nhe1
->nhg
.nexthop
, nexthop2
= nhe2
->nhg
.nexthop
;
568 nexthop1
&& nexthop2
;
569 nexthop1
= nexthop1
->next
, nexthop2
= nexthop2
->next
) {
571 if (!nhg_compare_nexthops(nexthop1
, nexthop2
))
575 /* Check for unequal list lengths */
576 if (nexthop1
|| nexthop2
)
579 /* If there's no backup info, comparison is done. */
580 if ((nhe1
->backup_info
== NULL
) && (nhe2
->backup_info
== NULL
))
583 /* Compare backup info also - test the easy things first */
584 if (nhe1
->backup_info
&& (nhe2
->backup_info
== NULL
))
586 if (nhe2
->backup_info
&& (nhe1
->backup_info
== NULL
))
589 /* Compare number of backups before actually comparing any */
590 for (nexthop1
= nhe1
->backup_info
->nhe
->nhg
.nexthop
,
591 nexthop2
= nhe2
->backup_info
->nhe
->nhg
.nexthop
;
592 nexthop1
&& nexthop2
;
593 nexthop1
= nexthop1
->next
, nexthop2
= nexthop2
->next
) {
597 /* Did we find the end of one list before the other? */
598 if (nexthop1
|| nexthop2
)
601 /* Have to compare the backup nexthops */
602 for (nexthop1
= nhe1
->backup_info
->nhe
->nhg
.nexthop
,
603 nexthop2
= nhe2
->backup_info
->nhe
->nhg
.nexthop
;
604 nexthop1
&& nexthop2
;
605 nexthop1
= nexthop1
->next
, nexthop2
= nexthop2
->next
) {
607 if (!nhg_compare_nexthops(nexthop1
, nexthop2
))
614 bool zebra_nhg_hash_id_equal(const void *arg1
, const void *arg2
)
616 const struct nhg_hash_entry
*nhe1
= arg1
;
617 const struct nhg_hash_entry
*nhe2
= arg2
;
619 return nhe1
->id
== nhe2
->id
;
622 static int zebra_nhg_process_grp(struct nexthop_group
*nhg
,
623 struct nhg_connected_tree_head
*depends
,
624 struct nh_grp
*grp
, uint8_t count
)
626 nhg_connected_tree_init(depends
);
628 for (int i
= 0; i
< count
; i
++) {
629 struct nhg_hash_entry
*depend
= NULL
;
630 /* We do not care about nexthop_grp.weight at
631 * this time. But we should figure out
632 * how to adapt this to our code in
635 depend
= depends_find_id_add(depends
, grp
[i
].id
);
640 "Received Nexthop Group from the kernel with a dependent Nexthop ID (%u) which we do not have in our table",
646 * If this is a nexthop with its own group
647 * dependencies, add them as well. Not sure its
648 * even possible to have a group within a group
652 copy_nexthops(&nhg
->nexthop
, depend
->nhg
.nexthop
, NULL
);
658 static void handle_recursive_depend(struct nhg_connected_tree_head
*nhg_depends
,
659 struct nexthop
*nh
, afi_t afi
, int type
)
661 struct nhg_hash_entry
*depend
= NULL
;
662 struct nexthop_group resolved_ng
= {};
664 resolved_ng
.nexthop
= nh
;
666 if (IS_ZEBRA_DEBUG_NHG_DETAIL
)
667 zlog_debug("%s: head %p, nh %pNHv",
668 __func__
, nhg_depends
, nh
);
670 depend
= zebra_nhg_rib_find(0, &resolved_ng
, afi
, type
);
672 if (IS_ZEBRA_DEBUG_NHG_DETAIL
)
673 zlog_debug("%s: nh %pNHv => %p (%u)",
674 __func__
, nh
, depend
,
675 depend
? depend
->id
: 0);
678 depends_add(nhg_depends
, depend
);
682 * Lookup an nhe in the global hash, using data from another nhe. If 'lookup'
683 * has an id value, that's used. Create a new global/shared nhe if not found.
685 static bool zebra_nhe_find(struct nhg_hash_entry
**nhe
, /* return value */
686 struct nhg_hash_entry
*lookup
,
687 struct nhg_connected_tree_head
*nhg_depends
,
688 afi_t afi
, bool from_dplane
)
690 bool created
= false;
691 bool recursive
= false;
692 struct nhg_hash_entry
*newnhe
, *backup_nhe
;
693 struct nexthop
*nh
= NULL
;
695 if (IS_ZEBRA_DEBUG_NHG_DETAIL
)
697 "%s: id %u, lookup %p, vrf %d, type %d, depends %p%s",
698 __func__
, lookup
->id
, lookup
, lookup
->vrf_id
,
699 lookup
->type
, nhg_depends
,
700 (from_dplane
? " (from dplane)" : ""));
703 (*nhe
) = zebra_nhg_lookup_id(lookup
->id
);
705 (*nhe
) = hash_lookup(zrouter
.nhgs
, lookup
);
707 if (IS_ZEBRA_DEBUG_NHG_DETAIL
)
708 zlog_debug("%s: lookup => %p (%u)",
710 (*nhe
) ? (*nhe
)->id
: 0);
712 /* If we found an existing object, we're done */
716 /* We're going to create/insert a new nhe:
717 * assign the next global id value if necessary.
720 lookup
->id
= nhg_get_next_id();
722 if (!from_dplane
&& lookup
->id
< ZEBRA_NHG_PROTO_LOWER
) {
724 * This is a zebra hashed/owned NHG.
726 * It goes in HASH and ID table.
728 newnhe
= hash_get(zrouter
.nhgs
, lookup
, zebra_nhg_hash_alloc
);
729 zebra_nhg_insert_id(newnhe
);
732 * This is upperproto owned NHG or one we read in from dataplane
733 * and should not be hashed to.
735 * It goes in ID table.
738 hash_get(zrouter
.nhgs_id
, lookup
, zebra_nhg_hash_alloc
);
743 /* Mail back the new object */
746 if (IS_ZEBRA_DEBUG_NHG_DETAIL
)
747 zlog_debug("%s: => created %p (%u)", __func__
, newnhe
,
750 /* Only hash/lookup the depends if the first lookup
751 * fails to find something. This should hopefully save a
752 * lot of cycles for larger ecmp sizes.
755 /* If you don't want to hash on each nexthop in the
756 * nexthop group struct you can pass the depends
757 * directly. Kernel-side we do this since it just looks
760 zebra_nhg_connect_depends(newnhe
, nhg_depends
);
764 /* Prepare dependency relationships if this is not a
765 * singleton nexthop. There are two cases: a single
766 * recursive nexthop, where we need a relationship to the
767 * resolving nexthop; or a group of nexthops, where we need
768 * relationships with the corresponding singletons.
770 zebra_nhg_depends_init(newnhe
);
772 nh
= newnhe
->nhg
.nexthop
;
774 if (CHECK_FLAG(nh
->flags
, NEXTHOP_FLAG_ACTIVE
))
775 SET_FLAG(newnhe
->flags
, NEXTHOP_GROUP_VALID
);
777 if (nh
->next
== NULL
&& newnhe
->id
< ZEBRA_NHG_PROTO_LOWER
) {
778 if (CHECK_FLAG(nh
->flags
, NEXTHOP_FLAG_RECURSIVE
)) {
779 /* Single recursive nexthop */
780 handle_recursive_depend(&newnhe
->nhg_depends
,
786 /* Proto-owned are groups by default */
787 /* List of nexthops */
788 for (nh
= newnhe
->nhg
.nexthop
; nh
; nh
= nh
->next
) {
789 if (IS_ZEBRA_DEBUG_NHG_DETAIL
)
790 zlog_debug("%s: depends NH %pNHv %s",
792 CHECK_FLAG(nh
->flags
,
793 NEXTHOP_FLAG_RECURSIVE
) ?
796 depends_find_add(&newnhe
->nhg_depends
, nh
, afi
,
797 newnhe
->type
, from_dplane
);
802 SET_FLAG(newnhe
->flags
, NEXTHOP_GROUP_RECURSIVE
);
804 /* Attach dependent backpointers to singletons */
805 zebra_nhg_connect_depends(newnhe
, &newnhe
->nhg_depends
);
811 if (zebra_nhg_get_backup_nhg(newnhe
) == NULL
||
812 zebra_nhg_get_backup_nhg(newnhe
)->nexthop
== NULL
)
815 /* If there are backup nexthops, add them to the backup
816 * depends tree. The rules here are a little different.
819 backup_nhe
= newnhe
->backup_info
->nhe
;
821 nh
= backup_nhe
->nhg
.nexthop
;
823 /* Singleton recursive NH */
824 if (nh
->next
== NULL
&&
825 CHECK_FLAG(nh
->flags
, NEXTHOP_FLAG_RECURSIVE
)) {
826 if (IS_ZEBRA_DEBUG_NHG_DETAIL
)
827 zlog_debug("%s: backup depend NH %pNHv (R)",
830 /* Single recursive nexthop */
831 handle_recursive_depend(&backup_nhe
->nhg_depends
, nh
->resolved
,
832 afi
, backup_nhe
->type
);
835 /* One or more backup NHs */
836 for (; nh
; nh
= nh
->next
) {
837 if (IS_ZEBRA_DEBUG_NHG_DETAIL
)
838 zlog_debug("%s: backup depend NH %pNHv %s",
840 CHECK_FLAG(nh
->flags
,
841 NEXTHOP_FLAG_RECURSIVE
) ?
844 depends_find_add(&backup_nhe
->nhg_depends
, nh
, afi
,
845 backup_nhe
->type
, from_dplane
);
850 SET_FLAG(backup_nhe
->flags
, NEXTHOP_GROUP_RECURSIVE
);
853 /* Reset time since last update */
854 (*nhe
)->uptime
= monotime(NULL
);
860 * Lookup or create an nhe, based on an nhg or an nhe id.
862 static bool zebra_nhg_find(struct nhg_hash_entry
**nhe
, uint32_t id
,
863 struct nexthop_group
*nhg
,
864 struct nhg_connected_tree_head
*nhg_depends
,
865 vrf_id_t vrf_id
, afi_t afi
, int type
,
868 struct nhg_hash_entry lookup
= {};
869 bool created
= false;
871 if (IS_ZEBRA_DEBUG_NHG_DETAIL
)
872 zlog_debug("%s: id %u, nhg %p, vrf %d, type %d, depends %p",
873 __func__
, id
, nhg
, vrf_id
, type
,
876 /* Use a temporary nhe and call into the superset/common code */
878 lookup
.type
= type
? type
: ZEBRA_ROUTE_NHG
;
881 lookup
.vrf_id
= vrf_id
;
882 if (nhg_depends
|| lookup
.nhg
.nexthop
->next
) {
883 /* Groups can have all vrfs and AF's in them */
884 lookup
.afi
= AFI_UNSPEC
;
886 switch (lookup
.nhg
.nexthop
->type
) {
887 case (NEXTHOP_TYPE_IFINDEX
):
888 case (NEXTHOP_TYPE_BLACKHOLE
):
890 * This switch case handles setting the afi different
891 * for ipv4/v6 routes. Ifindex/blackhole nexthop
892 * objects cannot be ambiguous, they must be Address
893 * Family specific. If we get here, we will either use
894 * the AF of the route, or the one we got passed from
895 * here from the kernel.
899 case (NEXTHOP_TYPE_IPV4_IFINDEX
):
900 case (NEXTHOP_TYPE_IPV4
):
903 case (NEXTHOP_TYPE_IPV6_IFINDEX
):
904 case (NEXTHOP_TYPE_IPV6
):
905 lookup
.afi
= AFI_IP6
;
910 created
= zebra_nhe_find(nhe
, &lookup
, nhg_depends
, afi
, from_dplane
);
915 /* Find/create a single nexthop */
916 static struct nhg_hash_entry
*zebra_nhg_find_nexthop(uint32_t id
,
921 struct nhg_hash_entry
*nhe
= NULL
;
922 struct nexthop_group nhg
= {};
923 vrf_id_t vrf_id
= !vrf_is_backend_netns() ? VRF_DEFAULT
: nh
->vrf_id
;
925 nexthop_group_add_sorted(&nhg
, nh
);
927 zebra_nhg_find(&nhe
, id
, &nhg
, NULL
, vrf_id
, afi
, type
, from_dplane
);
929 if (IS_ZEBRA_DEBUG_NHG_DETAIL
)
930 zlog_debug("%s: nh %pNHv => %p (%u)",
931 __func__
, nh
, nhe
, nhe
? nhe
->id
: 0);
936 static uint32_t nhg_ctx_get_id(const struct nhg_ctx
*ctx
)
941 static void nhg_ctx_set_status(struct nhg_ctx
*ctx
, enum nhg_ctx_status status
)
943 ctx
->status
= status
;
946 static enum nhg_ctx_status
nhg_ctx_get_status(const struct nhg_ctx
*ctx
)
951 static void nhg_ctx_set_op(struct nhg_ctx
*ctx
, enum nhg_ctx_op_e op
)
956 static enum nhg_ctx_op_e
nhg_ctx_get_op(const struct nhg_ctx
*ctx
)
961 static vrf_id_t
nhg_ctx_get_vrf_id(const struct nhg_ctx
*ctx
)
966 static int nhg_ctx_get_type(const struct nhg_ctx
*ctx
)
971 static int nhg_ctx_get_afi(const struct nhg_ctx
*ctx
)
976 static struct nexthop
*nhg_ctx_get_nh(struct nhg_ctx
*ctx
)
981 static uint8_t nhg_ctx_get_count(const struct nhg_ctx
*ctx
)
986 static struct nh_grp
*nhg_ctx_get_grp(struct nhg_ctx
*ctx
)
991 static struct nhg_ctx
*nhg_ctx_new(void)
995 new = XCALLOC(MTYPE_NHG_CTX
, sizeof(struct nhg_ctx
));
1000 void nhg_ctx_free(struct nhg_ctx
**ctx
)
1007 assert((*ctx
) != NULL
);
1009 if (nhg_ctx_get_count(*ctx
))
1012 nh
= nhg_ctx_get_nh(*ctx
);
1014 nexthop_del_labels(nh
);
1015 nexthop_del_srv6_seg6local(nh
);
1016 nexthop_del_srv6_seg6(nh
);
1019 XFREE(MTYPE_NHG_CTX
, *ctx
);
1022 static struct nhg_ctx
*nhg_ctx_init(uint32_t id
, struct nexthop
*nh
,
1023 struct nh_grp
*grp
, vrf_id_t vrf_id
,
1024 afi_t afi
, int type
, uint8_t count
)
1026 struct nhg_ctx
*ctx
= NULL
;
1028 ctx
= nhg_ctx_new();
1031 ctx
->vrf_id
= vrf_id
;
1037 /* Copy over the array */
1038 memcpy(&ctx
->u
.grp
, grp
, count
* sizeof(struct nh_grp
));
1045 static void zebra_nhg_set_valid(struct nhg_hash_entry
*nhe
)
1047 struct nhg_connected
*rb_node_dep
;
1049 SET_FLAG(nhe
->flags
, NEXTHOP_GROUP_VALID
);
1051 frr_each(nhg_connected_tree
, &nhe
->nhg_dependents
, rb_node_dep
)
1052 zebra_nhg_set_valid(rb_node_dep
->nhe
);
1055 static void zebra_nhg_set_invalid(struct nhg_hash_entry
*nhe
)
1057 struct nhg_connected
*rb_node_dep
;
1059 UNSET_FLAG(nhe
->flags
, NEXTHOP_GROUP_VALID
);
1061 /* Update validity of nexthops depending on it */
1062 frr_each(nhg_connected_tree
, &nhe
->nhg_dependents
, rb_node_dep
)
1063 zebra_nhg_check_valid(rb_node_dep
->nhe
);
1066 void zebra_nhg_check_valid(struct nhg_hash_entry
*nhe
)
1068 struct nhg_connected
*rb_node_dep
= NULL
;
1071 /* If anthing else in the group is valid, the group is valid */
1072 frr_each(nhg_connected_tree
, &nhe
->nhg_depends
, rb_node_dep
) {
1073 if (CHECK_FLAG(rb_node_dep
->nhe
->flags
, NEXTHOP_GROUP_VALID
)) {
1081 zebra_nhg_set_valid(nhe
);
1083 zebra_nhg_set_invalid(nhe
);
1086 static void zebra_nhg_release_all_deps(struct nhg_hash_entry
*nhe
)
1088 /* Remove it from any lists it may be on */
1089 zebra_nhg_depends_release(nhe
);
1090 zebra_nhg_dependents_release(nhe
);
1092 if_nhg_dependents_del(nhe
->ifp
, nhe
);
1095 static void zebra_nhg_release(struct nhg_hash_entry
*nhe
)
1097 if (IS_ZEBRA_DEBUG_NHG_DETAIL
)
1098 zlog_debug("%s: nhe %p (%u)", __func__
, nhe
, nhe
->id
);
1100 zebra_nhg_release_all_deps(nhe
);
1103 * If its not zebra owned, we didn't store it here and have to be
1104 * sure we don't clear one thats actually being used.
1106 if (nhe
->id
< ZEBRA_NHG_PROTO_LOWER
)
1107 hash_release(zrouter
.nhgs
, nhe
);
1109 hash_release(zrouter
.nhgs_id
, nhe
);
1112 static void zebra_nhg_handle_uninstall(struct nhg_hash_entry
*nhe
)
1114 zebra_nhg_release(nhe
);
1115 zebra_nhg_free(nhe
);
1118 static void zebra_nhg_handle_install(struct nhg_hash_entry
*nhe
)
1120 /* Update validity of groups depending on it */
1121 struct nhg_connected
*rb_node_dep
;
1123 frr_each_safe(nhg_connected_tree
, &nhe
->nhg_dependents
, rb_node_dep
)
1124 zebra_nhg_set_valid(rb_node_dep
->nhe
);
1128 * The kernel/other program has changed the state of a nexthop object we are
1131 static void zebra_nhg_handle_kernel_state_change(struct nhg_hash_entry
*nhe
,
1137 "Kernel %s a nexthop group with ID (%u) that we are still using for a route, sending it back down",
1138 (is_delete
? "deleted" : "updated"), nhe
->id
);
1140 UNSET_FLAG(nhe
->flags
, NEXTHOP_GROUP_INSTALLED
);
1141 zebra_nhg_install_kernel(nhe
);
1143 zebra_nhg_handle_uninstall(nhe
);
1146 static int nhg_ctx_process_new(struct nhg_ctx
*ctx
)
1148 struct nexthop_group
*nhg
= NULL
;
1149 struct nhg_connected_tree_head nhg_depends
= {};
1150 struct nhg_hash_entry
*lookup
= NULL
;
1151 struct nhg_hash_entry
*nhe
= NULL
;
1153 uint32_t id
= nhg_ctx_get_id(ctx
);
1154 uint8_t count
= nhg_ctx_get_count(ctx
);
1155 vrf_id_t vrf_id
= nhg_ctx_get_vrf_id(ctx
);
1156 int type
= nhg_ctx_get_type(ctx
);
1157 afi_t afi
= nhg_ctx_get_afi(ctx
);
1159 lookup
= zebra_nhg_lookup_id(id
);
1161 if (IS_ZEBRA_DEBUG_NHG_DETAIL
)
1162 zlog_debug("%s: id %u, count %d, lookup => %p",
1163 __func__
, id
, count
, lookup
);
1166 /* This is already present in our table, hence an update
1167 * that we did not initate.
1169 zebra_nhg_handle_kernel_state_change(lookup
, false);
1173 if (nhg_ctx_get_count(ctx
)) {
1174 nhg
= nexthop_group_new();
1175 if (zebra_nhg_process_grp(nhg
, &nhg_depends
,
1176 nhg_ctx_get_grp(ctx
), count
)) {
1177 depends_decrement_free(&nhg_depends
);
1178 nexthop_group_delete(&nhg
);
1182 if (!zebra_nhg_find(&nhe
, id
, nhg
, &nhg_depends
, vrf_id
, afi
,
1184 depends_decrement_free(&nhg_depends
);
1186 /* These got copied over in zebra_nhg_alloc() */
1187 nexthop_group_delete(&nhg
);
1189 nhe
= zebra_nhg_find_nexthop(id
, nhg_ctx_get_nh(ctx
), afi
, type
,
1194 EC_ZEBRA_TABLE_LOOKUP_FAILED
,
1195 "Zebra failed to find or create a nexthop hash entry for ID (%u)",
1200 if (IS_ZEBRA_DEBUG_NHG_DETAIL
)
1201 zlog_debug("%s: nhe %p (%u) is new", __func__
, nhe
, nhe
->id
);
1204 * If daemon nhg from the kernel, add a refcnt here to indicate the
1207 if (PROTO_OWNED(nhe
))
1208 zebra_nhg_increment_ref(nhe
);
1210 SET_FLAG(nhe
->flags
, NEXTHOP_GROUP_VALID
);
1211 SET_FLAG(nhe
->flags
, NEXTHOP_GROUP_INSTALLED
);
1216 static int nhg_ctx_process_del(struct nhg_ctx
*ctx
)
1218 struct nhg_hash_entry
*nhe
= NULL
;
1219 uint32_t id
= nhg_ctx_get_id(ctx
);
1221 nhe
= zebra_nhg_lookup_id(id
);
1225 EC_ZEBRA_BAD_NHG_MESSAGE
,
1226 "Kernel delete message received for nexthop group ID (%u) that we do not have in our ID table",
1231 zebra_nhg_handle_kernel_state_change(nhe
, true);
1236 static void nhg_ctx_fini(struct nhg_ctx
**ctx
)
1239 * Just freeing for now, maybe do something more in the future
1246 static int queue_add(struct nhg_ctx
*ctx
)
1248 /* If its queued or already processed do nothing */
1249 if (nhg_ctx_get_status(ctx
) == NHG_CTX_QUEUED
)
1252 if (rib_queue_nhg_ctx_add(ctx
)) {
1253 nhg_ctx_set_status(ctx
, NHG_CTX_FAILURE
);
1257 nhg_ctx_set_status(ctx
, NHG_CTX_QUEUED
);
1262 int nhg_ctx_process(struct nhg_ctx
*ctx
)
1266 switch (nhg_ctx_get_op(ctx
)) {
1267 case NHG_CTX_OP_NEW
:
1268 ret
= nhg_ctx_process_new(ctx
);
1269 if (nhg_ctx_get_count(ctx
) && ret
== -ENOENT
1270 && nhg_ctx_get_status(ctx
) != NHG_CTX_REQUEUED
) {
1272 * We have entered a situation where we are
1273 * processing a group from the kernel
1274 * that has a contained nexthop which
1275 * we have not yet processed.
1277 * Re-enqueue this ctx to be handled exactly one
1278 * more time (indicated by the flag).
1280 * By the time we get back to it, we
1281 * should have processed its depends.
1283 nhg_ctx_set_status(ctx
, NHG_CTX_NONE
);
1284 if (queue_add(ctx
) == 0) {
1285 nhg_ctx_set_status(ctx
, NHG_CTX_REQUEUED
);
1290 case NHG_CTX_OP_DEL
:
1291 ret
= nhg_ctx_process_del(ctx
);
1292 case NHG_CTX_OP_NONE
:
1296 nhg_ctx_set_status(ctx
, (ret
? NHG_CTX_FAILURE
: NHG_CTX_SUCCESS
));
1303 /* Kernel-side, you either get a single new nexthop or a array of ID's */
1304 int zebra_nhg_kernel_find(uint32_t id
, struct nexthop
*nh
, struct nh_grp
*grp
,
1305 uint8_t count
, vrf_id_t vrf_id
, afi_t afi
, int type
,
1308 struct nhg_ctx
*ctx
= NULL
;
1310 if (IS_ZEBRA_DEBUG_NHG_DETAIL
)
1311 zlog_debug("%s: nh %pNHv, id %u, count %d",
1312 __func__
, nh
, id
, (int)count
);
1314 if (id
> id_counter
&& id
< ZEBRA_NHG_PROTO_LOWER
)
1315 /* Increase our counter so we don't try to create
1316 * an ID that already exists
1320 ctx
= nhg_ctx_init(id
, nh
, grp
, vrf_id
, afi
, type
, count
);
1321 nhg_ctx_set_op(ctx
, NHG_CTX_OP_NEW
);
1323 /* Under statup conditions, we need to handle them immediately
1324 * like we do for routes. Otherwise, we are going to get a route
1325 * with a nhe_id that we have not handled.
1328 return nhg_ctx_process(ctx
);
1330 if (queue_add(ctx
)) {
1338 /* Kernel-side, received delete message */
1339 int zebra_nhg_kernel_del(uint32_t id
, vrf_id_t vrf_id
)
1341 struct nhg_ctx
*ctx
= NULL
;
1343 ctx
= nhg_ctx_init(id
, NULL
, NULL
, vrf_id
, 0, 0, 0);
1345 nhg_ctx_set_op(ctx
, NHG_CTX_OP_DEL
);
1347 if (queue_add(ctx
)) {
1355 /* Some dependency helper functions */
1356 static struct nhg_hash_entry
*depends_find_recursive(const struct nexthop
*nh
,
1357 afi_t afi
, int type
)
1359 struct nhg_hash_entry
*nhe
;
1360 struct nexthop
*lookup
= NULL
;
1362 lookup
= nexthop_dup(nh
, NULL
);
1364 nhe
= zebra_nhg_find_nexthop(0, lookup
, afi
, type
, false);
1366 nexthops_free(lookup
);
1371 static struct nhg_hash_entry
*depends_find_singleton(const struct nexthop
*nh
,
1372 afi_t afi
, int type
,
1375 struct nhg_hash_entry
*nhe
;
1376 struct nexthop lookup
= {};
1378 /* Capture a snapshot of this single nh; it might be part of a list,
1379 * so we need to make a standalone copy.
1381 nexthop_copy_no_recurse(&lookup
, nh
, NULL
);
1383 nhe
= zebra_nhg_find_nexthop(0, &lookup
, afi
, type
, from_dplane
);
1385 /* The copy may have allocated labels; free them if necessary. */
1386 nexthop_del_labels(&lookup
);
1387 nexthop_del_srv6_seg6local(&lookup
);
1388 nexthop_del_srv6_seg6(&lookup
);
1390 if (IS_ZEBRA_DEBUG_NHG_DETAIL
)
1391 zlog_debug("%s: nh %pNHv => %p (%u)",
1392 __func__
, nh
, nhe
, nhe
? nhe
->id
: 0);
1397 static struct nhg_hash_entry
*depends_find(const struct nexthop
*nh
, afi_t afi
,
1398 int type
, bool from_dplane
)
1400 struct nhg_hash_entry
*nhe
= NULL
;
1405 /* We are separating these functions out to increase handling speed
1406 * in the non-recursive case (by not alloc/freeing)
1408 if (CHECK_FLAG(nh
->flags
, NEXTHOP_FLAG_RECURSIVE
))
1409 nhe
= depends_find_recursive(nh
, afi
, type
);
1411 nhe
= depends_find_singleton(nh
, afi
, type
, from_dplane
);
1414 if (IS_ZEBRA_DEBUG_NHG_DETAIL
) {
1415 zlog_debug("%s: nh %pNHv %s => %p (%u)", __func__
, nh
,
1416 CHECK_FLAG(nh
->flags
, NEXTHOP_FLAG_RECURSIVE
) ? "(R)"
1418 nhe
, nhe
? nhe
->id
: 0);
1425 static void depends_add(struct nhg_connected_tree_head
*head
,
1426 struct nhg_hash_entry
*depend
)
1428 if (IS_ZEBRA_DEBUG_NHG_DETAIL
)
1429 zlog_debug("%s: head %p nh %pNHv",
1430 __func__
, head
, depend
->nhg
.nexthop
);
1432 /* If NULL is returned, it was successfully added and
1433 * needs to have its refcnt incremented.
1435 * Else the NHE is already present in the tree and doesn't
1436 * need to increment the refcnt.
1438 if (nhg_connected_tree_add_nhe(head
, depend
) == NULL
)
1439 zebra_nhg_increment_ref(depend
);
1442 static struct nhg_hash_entry
*
1443 depends_find_add(struct nhg_connected_tree_head
*head
, struct nexthop
*nh
,
1444 afi_t afi
, int type
, bool from_dplane
)
1446 struct nhg_hash_entry
*depend
= NULL
;
1448 depend
= depends_find(nh
, afi
, type
, from_dplane
);
1450 if (IS_ZEBRA_DEBUG_NHG_DETAIL
)
1451 zlog_debug("%s: nh %pNHv => %p",
1452 __func__
, nh
, depend
);
1455 depends_add(head
, depend
);
1460 static struct nhg_hash_entry
*
1461 depends_find_id_add(struct nhg_connected_tree_head
*head
, uint32_t id
)
1463 struct nhg_hash_entry
*depend
= NULL
;
1465 depend
= zebra_nhg_lookup_id(id
);
1468 depends_add(head
, depend
);
1473 static void depends_decrement_free(struct nhg_connected_tree_head
*head
)
1475 nhg_connected_tree_decrement_ref(head
);
1476 nhg_connected_tree_free(head
);
1479 /* Find an nhe based on a list of nexthops */
1480 struct nhg_hash_entry
*zebra_nhg_rib_find(uint32_t id
,
1481 struct nexthop_group
*nhg
,
1482 afi_t rt_afi
, int type
)
1484 struct nhg_hash_entry
*nhe
= NULL
;
1488 * CLANG SA is complaining that nexthop may be NULL
1489 * Make it happy but this is ridonc
1491 assert(nhg
->nexthop
);
1492 vrf_id
= !vrf_is_backend_netns() ? VRF_DEFAULT
: nhg
->nexthop
->vrf_id
;
1494 zebra_nhg_find(&nhe
, id
, nhg
, NULL
, vrf_id
, rt_afi
, type
, false);
1496 if (IS_ZEBRA_DEBUG_NHG_DETAIL
)
1497 zlog_debug("%s: => nhe %p (%u)",
1498 __func__
, nhe
, nhe
? nhe
->id
: 0);
1503 /* Find an nhe based on a route's nhe */
1504 struct nhg_hash_entry
*
1505 zebra_nhg_rib_find_nhe(struct nhg_hash_entry
*rt_nhe
, afi_t rt_afi
)
1507 struct nhg_hash_entry
*nhe
= NULL
;
1509 if (!(rt_nhe
&& rt_nhe
->nhg
.nexthop
)) {
1510 flog_err(EC_ZEBRA_TABLE_LOOKUP_FAILED
,
1511 "No nexthop passed to %s", __func__
);
1515 if (IS_ZEBRA_DEBUG_NHG_DETAIL
)
1516 zlog_debug("%s: rt_nhe %p (%u)", __func__
, rt_nhe
, rt_nhe
->id
);
1518 zebra_nhe_find(&nhe
, rt_nhe
, NULL
, rt_afi
, false);
1520 if (IS_ZEBRA_DEBUG_NHG_DETAIL
)
1521 zlog_debug("%s: => nhe %p (%u)",
1522 __func__
, nhe
, nhe
? nhe
->id
: 0);
1528 * Allocate backup nexthop info object. Typically these are embedded in
1529 * nhg_hash_entry objects.
1531 struct nhg_backup_info
*zebra_nhg_backup_alloc(void)
1533 struct nhg_backup_info
*p
;
1535 p
= XCALLOC(MTYPE_NHG
, sizeof(struct nhg_backup_info
));
1537 p
->nhe
= zebra_nhg_alloc();
1539 /* Identify the embedded group used to hold the list of backups */
1540 SET_FLAG(p
->nhe
->flags
, NEXTHOP_GROUP_BACKUP
);
1546 * Free backup nexthop info object, deal with any embedded allocations
1548 void zebra_nhg_backup_free(struct nhg_backup_info
**p
)
1552 zebra_nhg_free((*p
)->nhe
);
1554 XFREE(MTYPE_NHG
, (*p
));
1558 /* Accessor for backup nexthop group */
1559 struct nexthop_group
*zebra_nhg_get_backup_nhg(struct nhg_hash_entry
*nhe
)
1561 struct nexthop_group
*p
= NULL
;
1564 if (nhe
->backup_info
&& nhe
->backup_info
->nhe
)
1565 p
= &(nhe
->backup_info
->nhe
->nhg
);
1572 * Helper to return a copy of a backup_info - note that this is a shallow
1573 * copy, meant to be used when creating a new nhe from info passed in with
1576 static struct nhg_backup_info
*
1577 nhg_backup_copy(const struct nhg_backup_info
*orig
)
1579 struct nhg_backup_info
*b
;
1581 b
= zebra_nhg_backup_alloc();
1583 /* Copy list of nexthops */
1584 nexthop_group_copy(&(b
->nhe
->nhg
), &(orig
->nhe
->nhg
));
1589 static void zebra_nhg_free_members(struct nhg_hash_entry
*nhe
)
1591 nexthops_free(nhe
->nhg
.nexthop
);
1593 zebra_nhg_backup_free(&nhe
->backup_info
);
1595 /* Decrement to remove connection ref */
1596 nhg_connected_tree_decrement_ref(&nhe
->nhg_depends
);
1597 nhg_connected_tree_free(&nhe
->nhg_depends
);
1598 nhg_connected_tree_free(&nhe
->nhg_dependents
);
1601 void zebra_nhg_free(struct nhg_hash_entry
*nhe
)
1603 if (IS_ZEBRA_DEBUG_NHG_DETAIL
) {
1604 /* Group or singleton? */
1605 if (nhe
->nhg
.nexthop
&& nhe
->nhg
.nexthop
->next
)
1606 zlog_debug("%s: nhe %p (%u), refcnt %d",
1607 __func__
, nhe
, nhe
->id
, nhe
->refcnt
);
1609 zlog_debug("%s: nhe %p (%u), refcnt %d, NH %pNHv",
1610 __func__
, nhe
, nhe
->id
, nhe
->refcnt
,
1615 zlog_debug("nhe_id=%u hash refcnt=%d", nhe
->id
, nhe
->refcnt
);
1617 zebra_nhg_free_members(nhe
);
1619 XFREE(MTYPE_NHG
, nhe
);
1622 void zebra_nhg_hash_free(void *p
)
1624 zebra_nhg_release_all_deps((struct nhg_hash_entry
*)p
);
1625 zebra_nhg_free((struct nhg_hash_entry
*)p
);
1628 void zebra_nhg_decrement_ref(struct nhg_hash_entry
*nhe
)
1630 if (IS_ZEBRA_DEBUG_NHG_DETAIL
)
1631 zlog_debug("%s: nhe %p (%u) %d => %d",
1632 __func__
, nhe
, nhe
->id
, nhe
->refcnt
,
1637 if (!zebra_nhg_depends_is_empty(nhe
))
1638 nhg_connected_tree_decrement_ref(&nhe
->nhg_depends
);
1640 if (ZEBRA_NHG_CREATED(nhe
) && nhe
->refcnt
<= 0)
1641 zebra_nhg_uninstall_kernel(nhe
);
1644 void zebra_nhg_increment_ref(struct nhg_hash_entry
*nhe
)
1646 if (IS_ZEBRA_DEBUG_NHG_DETAIL
)
1647 zlog_debug("%s: nhe %p (%u) %d => %d",
1648 __func__
, nhe
, nhe
->id
, nhe
->refcnt
,
1653 if (!zebra_nhg_depends_is_empty(nhe
))
1654 nhg_connected_tree_increment_ref(&nhe
->nhg_depends
);
1657 static struct nexthop
*nexthop_set_resolved(afi_t afi
,
1658 const struct nexthop
*newhop
,
1659 struct nexthop
*nexthop
,
1660 struct zebra_sr_policy
*policy
)
1662 struct nexthop
*resolved_hop
;
1663 uint8_t num_labels
= 0;
1664 mpls_label_t labels
[MPLS_MAX_LABELS
];
1665 enum lsp_types_t label_type
= ZEBRA_LSP_NONE
;
1668 resolved_hop
= nexthop_new();
1669 SET_FLAG(resolved_hop
->flags
, NEXTHOP_FLAG_ACTIVE
);
1671 resolved_hop
->vrf_id
= nexthop
->vrf_id
;
1672 switch (newhop
->type
) {
1673 case NEXTHOP_TYPE_IPV4
:
1674 case NEXTHOP_TYPE_IPV4_IFINDEX
:
1675 /* If the resolving route specifies a gateway, use it */
1676 resolved_hop
->type
= newhop
->type
;
1677 resolved_hop
->gate
.ipv4
= newhop
->gate
.ipv4
;
1679 if (newhop
->ifindex
) {
1680 resolved_hop
->type
= NEXTHOP_TYPE_IPV4_IFINDEX
;
1681 resolved_hop
->ifindex
= newhop
->ifindex
;
1684 case NEXTHOP_TYPE_IPV6
:
1685 case NEXTHOP_TYPE_IPV6_IFINDEX
:
1686 resolved_hop
->type
= newhop
->type
;
1687 resolved_hop
->gate
.ipv6
= newhop
->gate
.ipv6
;
1689 if (newhop
->ifindex
) {
1690 resolved_hop
->type
= NEXTHOP_TYPE_IPV6_IFINDEX
;
1691 resolved_hop
->ifindex
= newhop
->ifindex
;
1694 case NEXTHOP_TYPE_IFINDEX
:
1695 /* If the resolving route is an interface route,
1696 * it means the gateway we are looking up is connected
1697 * to that interface. (The actual network is _not_ onlink).
1698 * Therefore, the resolved route should have the original
1699 * gateway as nexthop as it is directly connected.
1701 * On Linux, we have to set the onlink netlink flag because
1702 * otherwise, the kernel won't accept the route.
1704 resolved_hop
->flags
|= NEXTHOP_FLAG_ONLINK
;
1705 if (afi
== AFI_IP
) {
1706 resolved_hop
->type
= NEXTHOP_TYPE_IPV4_IFINDEX
;
1707 resolved_hop
->gate
.ipv4
= nexthop
->gate
.ipv4
;
1708 } else if (afi
== AFI_IP6
) {
1709 resolved_hop
->type
= NEXTHOP_TYPE_IPV6_IFINDEX
;
1710 resolved_hop
->gate
.ipv6
= nexthop
->gate
.ipv6
;
1712 resolved_hop
->ifindex
= newhop
->ifindex
;
1714 case NEXTHOP_TYPE_BLACKHOLE
:
1715 resolved_hop
->type
= NEXTHOP_TYPE_BLACKHOLE
;
1716 resolved_hop
->bh_type
= newhop
->bh_type
;
1720 if (newhop
->flags
& NEXTHOP_FLAG_ONLINK
)
1721 resolved_hop
->flags
|= NEXTHOP_FLAG_ONLINK
;
1723 /* Copy labels of the resolved route and the parent resolving to it */
1728 * Don't push the first SID if the corresponding action in the
1731 if (!newhop
->nh_label
|| !newhop
->nh_label
->num_labels
1732 || newhop
->nh_label
->label
[0] == MPLS_LABEL_IMPLICIT_NULL
)
1735 for (; i
< policy
->segment_list
.label_num
; i
++)
1736 labels
[num_labels
++] = policy
->segment_list
.labels
[i
];
1737 label_type
= policy
->segment_list
.type
;
1738 } else if (newhop
->nh_label
) {
1739 for (i
= 0; i
< newhop
->nh_label
->num_labels
; i
++) {
1740 /* Be a bit picky about overrunning the local array */
1741 if (num_labels
>= MPLS_MAX_LABELS
) {
1742 if (IS_ZEBRA_DEBUG_NHG
|| IS_ZEBRA_DEBUG_RIB
)
1743 zlog_debug("%s: too many labels in newhop %pNHv",
1747 labels
[num_labels
++] = newhop
->nh_label
->label
[i
];
1749 /* Use the "outer" type */
1750 label_type
= newhop
->nh_label_type
;
1753 if (nexthop
->nh_label
) {
1754 for (i
= 0; i
< nexthop
->nh_label
->num_labels
; i
++) {
1755 /* Be a bit picky about overrunning the local array */
1756 if (num_labels
>= MPLS_MAX_LABELS
) {
1757 if (IS_ZEBRA_DEBUG_NHG
|| IS_ZEBRA_DEBUG_RIB
)
1758 zlog_debug("%s: too many labels in nexthop %pNHv",
1762 labels
[num_labels
++] = nexthop
->nh_label
->label
[i
];
1765 /* If the parent has labels, use its type if
1766 * we don't already have one.
1768 if (label_type
== ZEBRA_LSP_NONE
)
1769 label_type
= nexthop
->nh_label_type
;
1773 nexthop_add_labels(resolved_hop
, label_type
, num_labels
,
1776 if (nexthop
->nh_srv6
) {
1777 nexthop_add_srv6_seg6local(resolved_hop
,
1778 nexthop
->nh_srv6
->seg6local_action
,
1779 &nexthop
->nh_srv6
->seg6local_ctx
);
1780 nexthop_add_srv6_seg6(resolved_hop
,
1781 &nexthop
->nh_srv6
->seg6_segs
);
1784 resolved_hop
->rparent
= nexthop
;
1785 _nexthop_add(&nexthop
->resolved
, resolved_hop
);
1787 return resolved_hop
;
1790 /* Checks if nexthop we are trying to resolve to is valid */
1791 static bool nexthop_valid_resolve(const struct nexthop
*nexthop
,
1792 const struct nexthop
*resolved
)
1794 /* Can't resolve to a recursive nexthop */
1795 if (CHECK_FLAG(resolved
->flags
, NEXTHOP_FLAG_RECURSIVE
))
1798 /* Must be ACTIVE */
1799 if (!CHECK_FLAG(resolved
->flags
, NEXTHOP_FLAG_ACTIVE
))
1802 /* Must not be duplicate */
1803 if (CHECK_FLAG(resolved
->flags
, NEXTHOP_FLAG_DUPLICATE
))
1806 switch (nexthop
->type
) {
1807 case NEXTHOP_TYPE_IPV4_IFINDEX
:
1808 case NEXTHOP_TYPE_IPV6_IFINDEX
:
1809 /* If the nexthop we are resolving to does not match the
1810 * ifindex for the nexthop the route wanted, its not valid.
1812 if (nexthop
->ifindex
!= resolved
->ifindex
)
1815 case NEXTHOP_TYPE_IPV4
:
1816 case NEXTHOP_TYPE_IPV6
:
1817 case NEXTHOP_TYPE_IFINDEX
:
1818 case NEXTHOP_TYPE_BLACKHOLE
:
1826 * When resolving a recursive nexthop, capture backup nexthop(s) also
1827 * so they can be conveyed through the dataplane to the FIB. We'll look
1828 * at the backups in the resolving nh 'nexthop' and its nhe, and copy them
1829 * into the route's resolved nh 'resolved' and its nhe 'nhe'.
1831 static int resolve_backup_nexthops(const struct nexthop
*nexthop
,
1832 const struct nhg_hash_entry
*nhe
,
1833 struct nexthop
*resolved
,
1834 struct nhg_hash_entry
*resolve_nhe
,
1835 struct backup_nh_map_s
*map
)
1838 const struct nexthop
*bnh
;
1839 struct nexthop
*nh
, *newnh
;
1840 mpls_label_t labels
[MPLS_MAX_LABELS
];
1843 assert(nexthop
->backup_num
<= NEXTHOP_MAX_BACKUPS
);
1845 /* Locate backups from the original nexthop's backup index and nhe */
1846 for (i
= 0; i
< nexthop
->backup_num
; i
++) {
1847 idx
= nexthop
->backup_idx
[i
];
1849 /* Do we already know about this particular backup? */
1850 for (j
= 0; j
< map
->map_count
; j
++) {
1851 if (map
->map
[j
].orig_idx
== idx
)
1855 if (j
< map
->map_count
) {
1856 resolved
->backup_idx
[resolved
->backup_num
] =
1857 map
->map
[j
].new_idx
;
1858 resolved
->backup_num
++;
1860 SET_FLAG(resolved
->flags
, NEXTHOP_FLAG_HAS_BACKUP
);
1862 if (IS_ZEBRA_DEBUG_RIB_DETAILED
)
1863 zlog_debug("%s: found map idx orig %d, new %d",
1864 __func__
, map
->map
[j
].orig_idx
,
1865 map
->map
[j
].new_idx
);
1870 /* We can't handle any new map entries at this point. */
1871 if (map
->map_count
== MULTIPATH_NUM
)
1874 /* Need to create/copy a new backup */
1875 bnh
= nhe
->backup_info
->nhe
->nhg
.nexthop
;
1876 for (j
= 0; j
< idx
; j
++) {
1882 /* Whoops - bad index in the nexthop? */
1886 if (resolve_nhe
->backup_info
== NULL
)
1887 resolve_nhe
->backup_info
= zebra_nhg_backup_alloc();
1889 /* Update backup info in the resolving nexthop and its nhe */
1890 newnh
= nexthop_dup_no_recurse(bnh
, NULL
);
1892 /* We may need some special handling for mpls labels: the new
1893 * backup needs to carry the recursive nexthop's labels,
1894 * if any: they may be vrf labels e.g.
1895 * The original/inner labels are in the stack of 'resolve_nhe',
1896 * if that is longer than the stack in 'nexthop'.
1898 if (newnh
->nh_label
&& resolved
->nh_label
&&
1899 nexthop
->nh_label
) {
1900 if (resolved
->nh_label
->num_labels
>
1901 nexthop
->nh_label
->num_labels
) {
1902 /* Prepare new label stack */
1904 for (j
= 0; j
< newnh
->nh_label
->num_labels
;
1906 labels
[j
] = newnh
->nh_label
->label
[j
];
1910 /* Include inner labels */
1911 for (j
= nexthop
->nh_label
->num_labels
;
1912 j
< resolved
->nh_label
->num_labels
;
1914 labels
[num_labels
] =
1915 resolved
->nh_label
->label
[j
];
1919 /* Replace existing label stack in the backup */
1920 nexthop_del_labels(newnh
);
1921 nexthop_add_labels(newnh
, bnh
->nh_label_type
,
1922 num_labels
, labels
);
1926 /* Need to compute the new backup index in the new
1927 * backup list, and add to map struct.
1930 nh
= resolve_nhe
->backup_info
->nhe
->nhg
.nexthop
;
1940 } else /* First one */
1941 resolve_nhe
->backup_info
->nhe
->nhg
.nexthop
= newnh
;
1944 resolved
->backup_idx
[resolved
->backup_num
] = j
;
1945 resolved
->backup_num
++;
1947 SET_FLAG(resolved
->flags
, NEXTHOP_FLAG_HAS_BACKUP
);
1949 if (IS_ZEBRA_DEBUG_RIB_DETAILED
)
1950 zlog_debug("%s: added idx orig %d, new %d",
1953 /* Update map/cache */
1954 map
->map
[map
->map_count
].orig_idx
= idx
;
1955 map
->map
[map
->map_count
].new_idx
= j
;
1963 * Given a nexthop we need to properly recursively resolve,
1964 * do a table lookup to find and match if at all possible.
1965 * Set the nexthop->ifindex and resolution info as appropriate.
1967 static int nexthop_active(struct nexthop
*nexthop
, struct nhg_hash_entry
*nhe
,
1968 const struct prefix
*top
, int type
, uint32_t flags
,
1969 uint32_t *pmtu
, vrf_id_t vrf_id
)
1972 struct route_table
*table
;
1973 struct route_node
*rn
;
1974 struct route_entry
*match
= NULL
;
1976 struct zebra_nhlfe
*nhlfe
;
1977 struct nexthop
*newhop
;
1978 struct interface
*ifp
;
1980 struct zebra_vrf
*zvrf
;
1981 struct in_addr local_ipv4
;
1982 struct in_addr
*ipv4
;
1985 /* Reset some nexthop attributes that we'll recompute if necessary */
1986 if ((nexthop
->type
== NEXTHOP_TYPE_IPV4
)
1987 || (nexthop
->type
== NEXTHOP_TYPE_IPV6
))
1988 nexthop
->ifindex
= 0;
1990 UNSET_FLAG(nexthop
->flags
, NEXTHOP_FLAG_RECURSIVE
);
1991 nexthops_free(nexthop
->resolved
);
1992 nexthop
->resolved
= NULL
;
1995 * Set afi based on nexthop type.
1996 * Some nexthop types get special handling, possibly skipping
1997 * the normal processing.
1999 switch (nexthop
->type
) {
2000 case NEXTHOP_TYPE_IFINDEX
:
2002 ifp
= if_lookup_by_index(nexthop
->ifindex
, nexthop
->vrf_id
);
2004 * If the interface exists and its operative or its a kernel
2005 * route and interface is up, its active. We trust kernel routes
2009 && (if_is_operative(ifp
)
2011 && (type
== ZEBRA_ROUTE_KERNEL
2012 || type
== ZEBRA_ROUTE_SYSTEM
))))
2018 case NEXTHOP_TYPE_IPV6_IFINDEX
:
2021 if (IN6_IS_ADDR_LINKLOCAL(&nexthop
->gate
.ipv6
)) {
2022 ifp
= if_lookup_by_index(nexthop
->ifindex
,
2024 if (ifp
&& if_is_operative(ifp
))
2031 case NEXTHOP_TYPE_IPV4
:
2032 case NEXTHOP_TYPE_IPV4_IFINDEX
:
2035 case NEXTHOP_TYPE_IPV6
:
2039 case NEXTHOP_TYPE_BLACKHOLE
:
2044 * If the nexthop has been marked as 'onlink' we just need to make
2045 * sure the nexthop's interface is known and is operational.
2047 if (CHECK_FLAG(nexthop
->flags
, NEXTHOP_FLAG_ONLINK
)) {
2048 ifp
= if_lookup_by_index(nexthop
->ifindex
, nexthop
->vrf_id
);
2050 if (IS_ZEBRA_DEBUG_RIB_DETAILED
)
2051 zlog_debug("nexthop %pNHv marked onlink but nhif %u doesn't exist",
2052 nexthop
, nexthop
->ifindex
);
2055 if (!if_is_operative(ifp
)) {
2056 if (IS_ZEBRA_DEBUG_RIB_DETAILED
)
2057 zlog_debug("nexthop %pNHv marked onlink but nhif %s is not operational",
2058 nexthop
, ifp
->name
);
2065 ((top
->family
== AF_INET
&& top
->prefixlen
== IPV4_MAX_BITLEN
&&
2066 nexthop
->gate
.ipv4
.s_addr
== top
->u
.prefix4
.s_addr
) ||
2067 (top
->family
== AF_INET6
&& top
->prefixlen
== IPV6_MAX_BITLEN
&&
2068 memcmp(&nexthop
->gate
.ipv6
, &top
->u
.prefix6
, IPV6_MAX_BYTELEN
) ==
2070 nexthop
->vrf_id
== vrf_id
) {
2071 if (IS_ZEBRA_DEBUG_RIB_DETAILED
)
2073 " :%s: Attempting to install a max prefixlength route through itself",
2078 /* Validation for ipv4 mapped ipv6 nexthop. */
2079 if (IS_MAPPED_IPV6(&nexthop
->gate
.ipv6
)) {
2082 ipv4_mapped_ipv6_to_ipv4(&nexthop
->gate
.ipv6
, ipv4
);
2084 ipv4
= &nexthop
->gate
.ipv4
;
2087 /* Processing for nexthops with SR 'color' attribute, using
2088 * the corresponding SR policy object.
2090 if (nexthop
->srte_color
) {
2091 struct ipaddr endpoint
= {0};
2092 struct zebra_sr_policy
*policy
;
2096 endpoint
.ipa_type
= IPADDR_V4
;
2097 endpoint
.ipaddr_v4
= *ipv4
;
2100 endpoint
.ipa_type
= IPADDR_V6
;
2101 endpoint
.ipaddr_v6
= nexthop
->gate
.ipv6
;
2104 flog_err(EC_LIB_DEVELOPMENT
,
2105 "%s: unknown address-family: %u", __func__
,
2110 policy
= zebra_sr_policy_find(nexthop
->srte_color
, &endpoint
);
2111 if (policy
&& policy
->status
== ZEBRA_SR_POLICY_UP
) {
2113 frr_each_safe (nhlfe_list
, &policy
->lsp
->nhlfe_list
,
2115 if (!CHECK_FLAG(nhlfe
->flags
,
2116 NHLFE_FLAG_SELECTED
)
2117 || CHECK_FLAG(nhlfe
->flags
,
2118 NHLFE_FLAG_DELETED
))
2120 SET_FLAG(nexthop
->flags
,
2121 NEXTHOP_FLAG_RECURSIVE
);
2122 nexthop_set_resolved(afi
, nhlfe
->nexthop
,
2131 /* Make lookup prefix. */
2132 memset(&p
, 0, sizeof(struct prefix
));
2136 p
.prefixlen
= IPV4_MAX_BITLEN
;
2137 p
.u
.prefix4
= *ipv4
;
2140 p
.family
= AF_INET6
;
2141 p
.prefixlen
= IPV6_MAX_BITLEN
;
2142 p
.u
.prefix6
= nexthop
->gate
.ipv6
;
2145 assert(afi
!= AFI_IP
&& afi
!= AFI_IP6
);
2149 table
= zebra_vrf_table(afi
, SAFI_UNICAST
, nexthop
->vrf_id
);
2151 zvrf
= zebra_vrf_lookup_by_id(nexthop
->vrf_id
);
2152 if (!table
|| !zvrf
) {
2153 if (IS_ZEBRA_DEBUG_RIB_DETAILED
)
2154 zlog_debug(" %s: Table not found", __func__
);
2158 rn
= route_node_match(table
, (struct prefix
*)&p
);
2160 route_unlock_node(rn
);
2162 /* Lookup should halt if we've matched against ourselves ('top',
2163 * if specified) - i.e., we cannot have a nexthop NH1 is
2164 * resolved by a route NH1. The exception is if the route is a
2167 if (prefix_same(&rn
->p
, top
))
2168 if (((afi
== AFI_IP
)
2169 && (rn
->p
.prefixlen
!= IPV4_MAX_BITLEN
))
2170 || ((afi
== AFI_IP6
)
2171 && (rn
->p
.prefixlen
!= IPV6_MAX_BITLEN
))) {
2172 if (IS_ZEBRA_DEBUG_RIB_DETAILED
)
2174 " %s: Matched against ourself and prefix length is not max bit length",
2179 /* Pick up selected route. */
2180 /* However, do not resolve over default route unless explicitly
2183 if (is_default_prefix(&rn
->p
)
2184 && !rnh_resolve_via_default(zvrf
, p
.family
)) {
2185 if (IS_ZEBRA_DEBUG_RIB_DETAILED
)
2187 " :%s: Resolved against default route",
2192 dest
= rib_dest_from_rnode(rn
);
2193 if (dest
&& dest
->selected_fib
2194 && !CHECK_FLAG(dest
->selected_fib
->status
,
2195 ROUTE_ENTRY_REMOVED
)
2196 && dest
->selected_fib
->type
!= ZEBRA_ROUTE_TABLE
)
2197 match
= dest
->selected_fib
;
2199 /* If there is no selected route or matched route is EGP, go up
2205 } while (rn
&& rn
->info
== NULL
);
2207 route_lock_node(rn
);
2212 if (match
->type
== ZEBRA_ROUTE_CONNECT
) {
2213 /* Directly point connected route. */
2214 newhop
= match
->nhe
->nhg
.nexthop
;
2216 if (nexthop
->type
== NEXTHOP_TYPE_IPV4
2217 || nexthop
->type
== NEXTHOP_TYPE_IPV6
)
2218 nexthop
->ifindex
= newhop
->ifindex
;
2219 else if (nexthop
->ifindex
!= newhop
->ifindex
) {
2220 if (IS_ZEBRA_DEBUG_RIB_DETAILED
)
2222 "%s: %pNHv given ifindex does not match nexthops ifindex found found: %pNHv",
2226 * NEXTHOP_TYPE_*_IFINDEX but ifindex
2227 * doesn't match what we found.
2233 if (IS_ZEBRA_DEBUG_NHG_DETAIL
)
2234 zlog_debug("%s: CONNECT match %p (%u), newhop %pNHv",
2236 match
->nhe
->id
, newhop
);
2239 } else if (CHECK_FLAG(flags
, ZEBRA_FLAG_ALLOW_RECURSION
)) {
2240 struct nexthop_group
*nhg
;
2241 struct nexthop
*resolver
;
2242 struct backup_nh_map_s map
= {};
2246 /* Only useful if installed */
2247 if (!CHECK_FLAG(match
->status
, ROUTE_ENTRY_INSTALLED
)) {
2248 if (IS_ZEBRA_DEBUG_RIB_DETAILED
)
2249 zlog_debug("%s: match %p (%u) not installed",
2253 goto done_with_match
;
2256 /* Examine installed nexthops; note that there
2257 * may not be any installed primary nexthops if
2258 * only backups are installed.
2260 nhg
= rib_get_fib_nhg(match
);
2261 for (ALL_NEXTHOPS_PTR(nhg
, newhop
)) {
2262 if (!nexthop_valid_resolve(nexthop
, newhop
))
2265 if (IS_ZEBRA_DEBUG_NHG_DETAIL
)
2266 zlog_debug("%s: RECURSIVE match %p (%u), newhop %pNHv",
2268 match
->nhe
->id
, newhop
);
2270 SET_FLAG(nexthop
->flags
,
2271 NEXTHOP_FLAG_RECURSIVE
);
2272 resolver
= nexthop_set_resolved(afi
, newhop
,
2276 /* If there are backup nexthops, capture
2277 * that info with the resolving nexthop.
2279 if (resolver
&& newhop
->backup_num
> 0) {
2280 resolve_backup_nexthops(newhop
,
2287 /* Examine installed backup nexthops, if any. There
2288 * are only installed backups *if* there is a
2289 * dedicated fib list. The UI can also control use
2290 * of backups for resolution.
2292 nhg
= rib_get_fib_backup_nhg(match
);
2293 if (!use_recursive_backups
||
2294 nhg
== NULL
|| nhg
->nexthop
== NULL
)
2295 goto done_with_match
;
2297 for (ALL_NEXTHOPS_PTR(nhg
, newhop
)) {
2298 if (!nexthop_valid_resolve(nexthop
, newhop
))
2301 if (IS_ZEBRA_DEBUG_NHG_DETAIL
)
2302 zlog_debug("%s: RECURSIVE match backup %p (%u), newhop %pNHv",
2304 match
->nhe
->id
, newhop
);
2306 SET_FLAG(nexthop
->flags
,
2307 NEXTHOP_FLAG_RECURSIVE
);
2308 nexthop_set_resolved(afi
, newhop
, nexthop
,
2314 /* Capture resolving mtu */
2319 } else if (IS_ZEBRA_DEBUG_RIB_DETAILED
)
2321 " %s: Recursion failed to find",
2326 if (IS_ZEBRA_DEBUG_RIB_DETAILED
) {
2328 " %s: Route Type %s has not turned on recursion",
2329 __func__
, zebra_route_string(type
));
2330 if (type
== ZEBRA_ROUTE_BGP
2331 && !CHECK_FLAG(flags
, ZEBRA_FLAG_IBGP
))
2333 " EBGP: see \"disable-ebgp-connected-route-check\" or \"disable-connected-check\"");
2338 if (IS_ZEBRA_DEBUG_RIB_DETAILED
)
2339 zlog_debug(" %s: Nexthop did not lookup in table",
2344 /* This function verifies reachability of one given nexthop, which can be
2345 * numbered or unnumbered, IPv4 or IPv6. The result is unconditionally stored
2346 * in nexthop->flags field. The nexthop->ifindex will be updated
2347 * appropriately as well.
2349 * An existing route map can turn an otherwise active nexthop into inactive,
2350 * but not vice versa.
2352 * The return value is the final value of 'ACTIVE' flag.
2354 static unsigned nexthop_active_check(struct route_node
*rn
,
2355 struct route_entry
*re
,
2356 struct nexthop
*nexthop
,
2357 struct nhg_hash_entry
*nhe
)
2359 route_map_result_t ret
= RMAP_PERMITMATCH
;
2361 const struct prefix
*p
, *src_p
;
2362 struct zebra_vrf
*zvrf
;
2366 srcdest_rnode_prefixes(rn
, &p
, &src_p
);
2368 if (rn
->p
.family
== AF_INET
)
2370 else if (rn
->p
.family
== AF_INET6
)
2375 if (IS_ZEBRA_DEBUG_NHG_DETAIL
)
2376 zlog_debug("%s: re %p, nexthop %pNHv", __func__
, re
, nexthop
);
2379 * If the kernel has sent us a NEW route, then
2380 * by golly gee whiz it's a good route.
2382 * If its an already INSTALLED route we have already handled, then the
2383 * kernel route's nexthop might have became unreachable
2384 * and we have to handle that.
2386 if (!CHECK_FLAG(re
->status
, ROUTE_ENTRY_INSTALLED
) &&
2387 (re
->type
== ZEBRA_ROUTE_KERNEL
||
2388 re
->type
== ZEBRA_ROUTE_SYSTEM
)) {
2389 SET_FLAG(nexthop
->flags
, NEXTHOP_FLAG_ACTIVE
);
2394 vrf_id
= zvrf_id(rib_dest_vrf(rib_dest_from_rnode(rn
)));
2395 switch (nexthop
->type
) {
2396 case NEXTHOP_TYPE_IFINDEX
:
2397 if (nexthop_active(nexthop
, nhe
, &rn
->p
, re
->type
, re
->flags
,
2399 SET_FLAG(nexthop
->flags
, NEXTHOP_FLAG_ACTIVE
);
2401 UNSET_FLAG(nexthop
->flags
, NEXTHOP_FLAG_ACTIVE
);
2403 case NEXTHOP_TYPE_IPV4
:
2404 case NEXTHOP_TYPE_IPV4_IFINDEX
:
2406 if (nexthop_active(nexthop
, nhe
, &rn
->p
, re
->type
, re
->flags
,
2408 SET_FLAG(nexthop
->flags
, NEXTHOP_FLAG_ACTIVE
);
2410 UNSET_FLAG(nexthop
->flags
, NEXTHOP_FLAG_ACTIVE
);
2412 case NEXTHOP_TYPE_IPV6
:
2414 if (nexthop_active(nexthop
, nhe
, &rn
->p
, re
->type
, re
->flags
,
2416 SET_FLAG(nexthop
->flags
, NEXTHOP_FLAG_ACTIVE
);
2418 UNSET_FLAG(nexthop
->flags
, NEXTHOP_FLAG_ACTIVE
);
2420 case NEXTHOP_TYPE_IPV6_IFINDEX
:
2421 /* RFC 5549, v4 prefix with v6 NH */
2422 if (rn
->p
.family
!= AF_INET
)
2425 if (nexthop_active(nexthop
, nhe
, &rn
->p
, re
->type
, re
->flags
,
2427 SET_FLAG(nexthop
->flags
, NEXTHOP_FLAG_ACTIVE
);
2429 UNSET_FLAG(nexthop
->flags
, NEXTHOP_FLAG_ACTIVE
);
2431 case NEXTHOP_TYPE_BLACKHOLE
:
2432 SET_FLAG(nexthop
->flags
, NEXTHOP_FLAG_ACTIVE
);
2440 if (!CHECK_FLAG(nexthop
->flags
, NEXTHOP_FLAG_ACTIVE
)) {
2441 if (IS_ZEBRA_DEBUG_RIB_DETAILED
)
2442 zlog_debug(" %s: Unable to find active nexthop",
2447 /* Capture recursive nexthop mtu.
2448 * TODO -- the code used to just reset the re's value to zero
2449 * for each nexthop, and then jam any resolving route's mtu value in,
2450 * whether or not that was zero, or lt/gt any existing value? The
2451 * way this is used appears to be as a floor value, so let's try
2452 * using it that way here.
2455 if (re
->nexthop_mtu
== 0 || re
->nexthop_mtu
> mtu
)
2456 re
->nexthop_mtu
= mtu
;
2459 /* XXX: What exactly do those checks do? Do we support
2460 * e.g. IPv4 routes with IPv6 nexthops or vice versa?
2462 if (RIB_SYSTEM_ROUTE(re
) || (family
== AFI_IP
&& p
->family
!= AF_INET
)
2463 || (family
== AFI_IP6
&& p
->family
!= AF_INET6
))
2464 return CHECK_FLAG(nexthop
->flags
, NEXTHOP_FLAG_ACTIVE
);
2466 /* The original code didn't determine the family correctly
2467 * e.g. for NEXTHOP_TYPE_IFINDEX. Retrieve the correct afi
2468 * from the rib_table_info in those cases.
2469 * Possibly it may be better to use only the rib_table_info
2473 struct rib_table_info
*info
;
2475 info
= srcdest_rnode_table_info(rn
);
2479 memset(&nexthop
->rmap_src
.ipv6
, 0, sizeof(union g_addr
));
2481 zvrf
= zebra_vrf_lookup_by_id(re
->vrf_id
);
2483 if (IS_ZEBRA_DEBUG_RIB_DETAILED
)
2484 zlog_debug(" %s: zvrf is NULL", __func__
);
2485 return CHECK_FLAG(nexthop
->flags
, NEXTHOP_FLAG_ACTIVE
);
2488 /* It'll get set if required inside */
2489 ret
= zebra_route_map_check(family
, re
->type
, re
->instance
, p
, nexthop
,
2491 if (ret
== RMAP_DENYMATCH
) {
2492 if (IS_ZEBRA_DEBUG_RIB
) {
2494 "%u:%pRN: Filtering out with NH out %s due to route map",
2496 ifindex2ifname(nexthop
->ifindex
,
2499 UNSET_FLAG(nexthop
->flags
, NEXTHOP_FLAG_ACTIVE
);
2501 return CHECK_FLAG(nexthop
->flags
, NEXTHOP_FLAG_ACTIVE
);
2504 /* Helper function called after resolution to walk nhg rb trees
2505 * and toggle the NEXTHOP_GROUP_VALID flag if the nexthop
2506 * is active on singleton NHEs.
2508 static bool zebra_nhg_set_valid_if_active(struct nhg_hash_entry
*nhe
)
2510 struct nhg_connected
*rb_node_dep
= NULL
;
2513 if (!zebra_nhg_depends_is_empty(nhe
)) {
2514 /* Is at least one depend valid? */
2515 frr_each(nhg_connected_tree
, &nhe
->nhg_depends
, rb_node_dep
) {
2516 if (zebra_nhg_set_valid_if_active(rb_node_dep
->nhe
))
2523 /* should be fully resolved singleton at this point */
2524 if (CHECK_FLAG(nhe
->nhg
.nexthop
->flags
, NEXTHOP_FLAG_ACTIVE
))
2529 SET_FLAG(nhe
->flags
, NEXTHOP_GROUP_VALID
);
2535 * Process a list of nexthops, given an nhe, determining
2536 * whether each one is ACTIVE/installable at this time.
2538 static uint32_t nexthop_list_active_update(struct route_node
*rn
,
2539 struct route_entry
*re
,
2540 struct nhg_hash_entry
*nhe
,
2543 union g_addr prev_src
;
2544 unsigned int prev_active
, new_active
;
2545 ifindex_t prev_index
;
2546 uint32_t counter
= 0;
2547 struct nexthop
*nexthop
;
2548 struct nexthop_group
*nhg
= &nhe
->nhg
;
2550 nexthop
= nhg
->nexthop
;
2552 /* Init recursive nh mtu */
2553 re
->nexthop_mtu
= 0;
2555 /* Process nexthops one-by-one */
2556 for ( ; nexthop
; nexthop
= nexthop
->next
) {
2558 /* No protocol daemon provides src and so we're skipping
2561 prev_src
= nexthop
->rmap_src
;
2562 prev_active
= CHECK_FLAG(nexthop
->flags
, NEXTHOP_FLAG_ACTIVE
);
2563 prev_index
= nexthop
->ifindex
;
2565 /* Include the containing nhe for primary nexthops: if there's
2566 * recursive resolution, we capture the backup info also.
2569 nexthop_active_check(rn
, re
, nexthop
,
2570 (is_backup
? NULL
: nhe
));
2573 * We need to respect the multipath_num here
2574 * as that what we should be able to install from
2575 * a multipath perspective should not be a data plane
2578 if (new_active
&& counter
>= zrouter
.multipath_num
) {
2581 /* Set it and its resolved nexthop as inactive. */
2582 for (nh
= nexthop
; nh
; nh
= nh
->resolved
)
2583 UNSET_FLAG(nh
->flags
, NEXTHOP_FLAG_ACTIVE
);
2591 /* Check for changes to the nexthop - set ROUTE_ENTRY_CHANGED */
2592 if (prev_active
!= new_active
|| prev_index
!= nexthop
->ifindex
2593 || ((nexthop
->type
>= NEXTHOP_TYPE_IFINDEX
2594 && nexthop
->type
< NEXTHOP_TYPE_IPV6
)
2595 && prev_src
.ipv4
.s_addr
2596 != nexthop
->rmap_src
.ipv4
.s_addr
)
2597 || ((nexthop
->type
>= NEXTHOP_TYPE_IPV6
2598 && nexthop
->type
< NEXTHOP_TYPE_BLACKHOLE
)
2599 && !(IPV6_ADDR_SAME(&prev_src
.ipv6
,
2600 &nexthop
->rmap_src
.ipv6
)))
2601 || CHECK_FLAG(re
->status
, ROUTE_ENTRY_LABELS_CHANGED
))
2602 SET_FLAG(re
->status
, ROUTE_ENTRY_CHANGED
);
2609 static uint32_t proto_nhg_nexthop_active_update(struct nexthop_group
*nhg
)
2612 uint32_t curr_active
= 0;
2614 /* Assume all active for now */
2616 for (nh
= nhg
->nexthop
; nh
; nh
= nh
->next
) {
2617 SET_FLAG(nh
->flags
, NEXTHOP_FLAG_ACTIVE
);
2625 * Iterate over all nexthops of the given RIB entry and refresh their
2626 * ACTIVE flag. If any nexthop is found to toggle the ACTIVE flag,
2627 * the whole re structure is flagged with ROUTE_ENTRY_CHANGED.
2629 * Return value is the new number of active nexthops.
2631 int nexthop_active_update(struct route_node
*rn
, struct route_entry
*re
)
2633 struct nhg_hash_entry
*curr_nhe
;
2634 uint32_t curr_active
= 0, backup_active
= 0;
2636 if (PROTO_OWNED(re
->nhe
))
2637 return proto_nhg_nexthop_active_update(&re
->nhe
->nhg
);
2639 afi_t rt_afi
= family2afi(rn
->p
.family
);
2641 UNSET_FLAG(re
->status
, ROUTE_ENTRY_CHANGED
);
2643 /* Make a local copy of the existing nhe, so we don't work on/modify
2646 curr_nhe
= zebra_nhe_copy(re
->nhe
, re
->nhe
->id
);
2648 if (IS_ZEBRA_DEBUG_NHG_DETAIL
)
2649 zlog_debug("%s: re %p nhe %p (%u), curr_nhe %p",
2650 __func__
, re
, re
->nhe
, re
->nhe
->id
,
2653 /* Clear the existing id, if any: this will avoid any confusion
2654 * if the id exists, and will also force the creation
2655 * of a new nhe reflecting the changes we may make in this local copy.
2659 /* Process nexthops */
2660 curr_active
= nexthop_list_active_update(rn
, re
, curr_nhe
, false);
2662 if (IS_ZEBRA_DEBUG_NHG_DETAIL
)
2663 zlog_debug("%s: re %p curr_active %u", __func__
, re
,
2666 /* If there are no backup nexthops, we are done */
2667 if (zebra_nhg_get_backup_nhg(curr_nhe
) == NULL
)
2670 backup_active
= nexthop_list_active_update(
2671 rn
, re
, curr_nhe
->backup_info
->nhe
, true /*is_backup*/);
2673 if (IS_ZEBRA_DEBUG_NHG_DETAIL
)
2674 zlog_debug("%s: re %p backup_active %u", __func__
, re
,
2680 * Ref or create an nhe that matches the current state of the
2683 if (CHECK_FLAG(re
->status
, ROUTE_ENTRY_CHANGED
)) {
2684 struct nhg_hash_entry
*new_nhe
= NULL
;
2686 new_nhe
= zebra_nhg_rib_find_nhe(curr_nhe
, rt_afi
);
2688 if (IS_ZEBRA_DEBUG_NHG_DETAIL
)
2689 zlog_debug("%s: re %p CHANGED: nhe %p (%u) => new_nhe %p (%u)",
2690 __func__
, re
, re
->nhe
,
2691 re
->nhe
->id
, new_nhe
, new_nhe
->id
);
2693 route_entry_update_nhe(re
, new_nhe
);
2697 /* Walk the NHE depends tree and toggle NEXTHOP_GROUP_VALID
2698 * flag where appropriate.
2701 zebra_nhg_set_valid_if_active(re
->nhe
);
2704 * Do not need the old / copied nhe anymore since it
2705 * was either copied over into a new nhe or not
2708 zebra_nhg_free(curr_nhe
);
2712 /* Recursively construct a grp array of fully resolved IDs.
2714 * This function allows us to account for groups within groups,
2715 * by converting them into a flat array of IDs.
2717 * nh_grp is modified at every level of recursion to append
2718 * to it the next unique, fully resolved ID from the entire tree.
2722 * I'm pretty sure we only allow ONE level of group within group currently.
2723 * But making this recursive just in case that ever changes.
2725 static uint8_t zebra_nhg_nhe2grp_internal(struct nh_grp
*grp
,
2727 struct nhg_hash_entry
*nhe
,
2730 struct nhg_connected
*rb_node_dep
= NULL
;
2731 struct nhg_hash_entry
*depend
= NULL
;
2732 uint8_t i
= curr_index
;
2734 frr_each(nhg_connected_tree
, &nhe
->nhg_depends
, rb_node_dep
) {
2735 bool duplicate
= false;
2740 depend
= rb_node_dep
->nhe
;
2743 * If its recursive, use its resolved nhe in the group
2745 if (CHECK_FLAG(depend
->flags
, NEXTHOP_GROUP_RECURSIVE
)) {
2746 depend
= zebra_nhg_resolve(depend
);
2749 EC_ZEBRA_NHG_FIB_UPDATE
,
2750 "Failed to recursively resolve Nexthop Hash Entry in the group id=%u",
2756 if (!zebra_nhg_depends_is_empty(depend
)) {
2757 /* This is a group within a group */
2758 i
= zebra_nhg_nhe2grp_internal(grp
, i
, depend
, max_num
);
2760 if (!CHECK_FLAG(depend
->flags
, NEXTHOP_GROUP_VALID
)) {
2761 if (IS_ZEBRA_DEBUG_RIB_DETAILED
2762 || IS_ZEBRA_DEBUG_NHG
)
2764 "%s: Nexthop ID (%u) not valid, not appending to dataplane install group",
2765 __func__
, depend
->id
);
2769 /* If the nexthop not installed/queued for install don't
2770 * put in the ID array.
2772 if (!(CHECK_FLAG(depend
->flags
, NEXTHOP_GROUP_INSTALLED
)
2773 || CHECK_FLAG(depend
->flags
,
2774 NEXTHOP_GROUP_QUEUED
))) {
2775 if (IS_ZEBRA_DEBUG_RIB_DETAILED
2776 || IS_ZEBRA_DEBUG_NHG
)
2778 "%s: Nexthop ID (%u) not installed or queued for install, not appending to dataplane install group",
2779 __func__
, depend
->id
);
2783 /* Check for duplicate IDs, ignore if found. */
2784 for (int j
= 0; j
< i
; j
++) {
2785 if (depend
->id
== grp
[j
].id
) {
2792 if (IS_ZEBRA_DEBUG_RIB_DETAILED
2793 || IS_ZEBRA_DEBUG_NHG
)
2795 "%s: Nexthop ID (%u) is duplicate, not appending to dataplane install group",
2796 __func__
, depend
->id
);
2800 grp
[i
].id
= depend
->id
;
2801 grp
[i
].weight
= depend
->nhg
.nexthop
->weight
;
2806 if (nhe
->backup_info
== NULL
|| nhe
->backup_info
->nhe
== NULL
)
2809 /* TODO -- For now, we are not trying to use or install any
2810 * backup info in this nexthop-id path: we aren't prepared
2811 * to use the backups here yet. We're just debugging what we find.
2813 if (IS_ZEBRA_DEBUG_NHG_DETAIL
)
2814 zlog_debug("%s: skipping backup nhe", __func__
);
2820 /* Convert a nhe into a group array */
2821 uint8_t zebra_nhg_nhe2grp(struct nh_grp
*grp
, struct nhg_hash_entry
*nhe
,
2824 /* Call into the recursive function */
2825 return zebra_nhg_nhe2grp_internal(grp
, 0, nhe
, max_num
);
2828 void zebra_nhg_install_kernel(struct nhg_hash_entry
*nhe
)
2830 struct nhg_connected
*rb_node_dep
= NULL
;
2832 /* Resolve it first */
2833 nhe
= zebra_nhg_resolve(nhe
);
2835 /* Make sure all depends are installed/queued */
2836 frr_each(nhg_connected_tree
, &nhe
->nhg_depends
, rb_node_dep
) {
2837 zebra_nhg_install_kernel(rb_node_dep
->nhe
);
2840 if (CHECK_FLAG(nhe
->flags
, NEXTHOP_GROUP_VALID
)
2841 && !CHECK_FLAG(nhe
->flags
, NEXTHOP_GROUP_INSTALLED
)
2842 && !CHECK_FLAG(nhe
->flags
, NEXTHOP_GROUP_QUEUED
)) {
2843 /* Change its type to us since we are installing it */
2844 if (!ZEBRA_NHG_CREATED(nhe
))
2845 nhe
->type
= ZEBRA_ROUTE_NHG
;
2847 int ret
= dplane_nexthop_add(nhe
);
2850 case ZEBRA_DPLANE_REQUEST_QUEUED
:
2851 SET_FLAG(nhe
->flags
, NEXTHOP_GROUP_QUEUED
);
2853 case ZEBRA_DPLANE_REQUEST_FAILURE
:
2855 EC_ZEBRA_DP_INSTALL_FAIL
,
2856 "Failed to install Nexthop ID (%u) into the kernel",
2859 case ZEBRA_DPLANE_REQUEST_SUCCESS
:
2860 SET_FLAG(nhe
->flags
, NEXTHOP_GROUP_INSTALLED
);
2861 zebra_nhg_handle_install(nhe
);
2867 void zebra_nhg_uninstall_kernel(struct nhg_hash_entry
*nhe
)
2869 if (CHECK_FLAG(nhe
->flags
, NEXTHOP_GROUP_INSTALLED
)) {
2870 int ret
= dplane_nexthop_delete(nhe
);
2873 case ZEBRA_DPLANE_REQUEST_QUEUED
:
2874 SET_FLAG(nhe
->flags
, NEXTHOP_GROUP_QUEUED
);
2876 case ZEBRA_DPLANE_REQUEST_FAILURE
:
2878 EC_ZEBRA_DP_DELETE_FAIL
,
2879 "Failed to uninstall Nexthop ID (%u) from the kernel",
2882 case ZEBRA_DPLANE_REQUEST_SUCCESS
:
2883 UNSET_FLAG(nhe
->flags
, NEXTHOP_GROUP_INSTALLED
);
2888 zebra_nhg_handle_uninstall(nhe
);
2891 void zebra_nhg_dplane_result(struct zebra_dplane_ctx
*ctx
)
2893 enum dplane_op_e op
;
2894 enum zebra_dplane_result status
;
2896 struct nhg_hash_entry
*nhe
= NULL
;
2898 op
= dplane_ctx_get_op(ctx
);
2899 status
= dplane_ctx_get_status(ctx
);
2901 id
= dplane_ctx_get_nhe_id(ctx
);
2903 if (IS_ZEBRA_DEBUG_DPLANE_DETAIL
|| IS_ZEBRA_DEBUG_NHG_DETAIL
)
2905 "Nexthop dplane ctx %p, op %s, nexthop ID (%u), result %s",
2906 ctx
, dplane_op2str(op
), id
, dplane_res2str(status
));
2909 case DPLANE_OP_NH_DELETE
:
2910 if (status
!= ZEBRA_DPLANE_REQUEST_SUCCESS
)
2912 EC_ZEBRA_DP_DELETE_FAIL
,
2913 "Failed to uninstall Nexthop ID (%u) from the kernel",
2916 /* We already free'd the data, nothing to do */
2918 case DPLANE_OP_NH_INSTALL
:
2919 case DPLANE_OP_NH_UPDATE
:
2920 nhe
= zebra_nhg_lookup_id(id
);
2923 if (IS_ZEBRA_DEBUG_NHG
)
2925 "%s operation preformed on Nexthop ID (%u) in the kernel, that we no longer have in our table",
2926 dplane_op2str(op
), id
);
2931 UNSET_FLAG(nhe
->flags
, NEXTHOP_GROUP_QUEUED
);
2932 if (status
== ZEBRA_DPLANE_REQUEST_SUCCESS
) {
2933 SET_FLAG(nhe
->flags
, NEXTHOP_GROUP_VALID
);
2934 SET_FLAG(nhe
->flags
, NEXTHOP_GROUP_INSTALLED
);
2935 zebra_nhg_handle_install(nhe
);
2937 /* If daemon nhg, send it an update */
2938 if (PROTO_OWNED(nhe
))
2939 zsend_nhg_notify(nhe
->type
, nhe
->zapi_instance
,
2940 nhe
->zapi_session
, nhe
->id
,
2941 ZAPI_NHG_INSTALLED
);
2943 /* If daemon nhg, send it an update */
2944 if (PROTO_OWNED(nhe
))
2945 zsend_nhg_notify(nhe
->type
, nhe
->zapi_instance
,
2946 nhe
->zapi_session
, nhe
->id
,
2947 ZAPI_NHG_FAIL_INSTALL
);
2950 EC_ZEBRA_DP_INSTALL_FAIL
,
2951 "Failed to install Nexthop ID (%u) into the kernel",
2956 case DPLANE_OP_ROUTE_INSTALL
:
2957 case DPLANE_OP_ROUTE_UPDATE
:
2958 case DPLANE_OP_ROUTE_DELETE
:
2959 case DPLANE_OP_ROUTE_NOTIFY
:
2960 case DPLANE_OP_LSP_INSTALL
:
2961 case DPLANE_OP_LSP_UPDATE
:
2962 case DPLANE_OP_LSP_DELETE
:
2963 case DPLANE_OP_LSP_NOTIFY
:
2964 case DPLANE_OP_PW_INSTALL
:
2965 case DPLANE_OP_PW_UNINSTALL
:
2966 case DPLANE_OP_SYS_ROUTE_ADD
:
2967 case DPLANE_OP_SYS_ROUTE_DELETE
:
2968 case DPLANE_OP_ADDR_INSTALL
:
2969 case DPLANE_OP_ADDR_UNINSTALL
:
2970 case DPLANE_OP_MAC_INSTALL
:
2971 case DPLANE_OP_MAC_DELETE
:
2972 case DPLANE_OP_NEIGH_INSTALL
:
2973 case DPLANE_OP_NEIGH_UPDATE
:
2974 case DPLANE_OP_NEIGH_DELETE
:
2975 case DPLANE_OP_NEIGH_IP_INSTALL
:
2976 case DPLANE_OP_NEIGH_IP_DELETE
:
2977 case DPLANE_OP_VTEP_ADD
:
2978 case DPLANE_OP_VTEP_DELETE
:
2979 case DPLANE_OP_RULE_ADD
:
2980 case DPLANE_OP_RULE_DELETE
:
2981 case DPLANE_OP_RULE_UPDATE
:
2982 case DPLANE_OP_NEIGH_DISCOVER
:
2983 case DPLANE_OP_BR_PORT_UPDATE
:
2984 case DPLANE_OP_NONE
:
2985 case DPLANE_OP_IPTABLE_ADD
:
2986 case DPLANE_OP_IPTABLE_DELETE
:
2987 case DPLANE_OP_IPSET_ADD
:
2988 case DPLANE_OP_IPSET_DELETE
:
2989 case DPLANE_OP_IPSET_ENTRY_ADD
:
2990 case DPLANE_OP_IPSET_ENTRY_DELETE
:
2991 case DPLANE_OP_NEIGH_TABLE_UPDATE
:
2992 case DPLANE_OP_GRE_SET
:
2993 case DPLANE_OP_INTF_ADDR_ADD
:
2994 case DPLANE_OP_INTF_ADDR_DEL
:
2995 case DPLANE_OP_INTF_NETCONFIG
:
2999 dplane_ctx_fini(&ctx
);
3002 static int zebra_nhg_sweep_entry(struct hash_bucket
*bucket
, void *arg
)
3004 struct nhg_hash_entry
*nhe
= NULL
;
3006 nhe
= (struct nhg_hash_entry
*)bucket
->data
;
3009 * same logic as with routes.
3011 * If older than startup time, we know we read them in from the
3012 * kernel and have not gotten and update for them since startup
3013 * from an upper level proto.
3015 if (zrouter
.startup_time
< nhe
->uptime
)
3016 return HASHWALK_CONTINUE
;
3019 * If it's proto-owned and not being used by a route, remove it since
3020 * we haven't gotten an update about it from the proto since startup.
3021 * This means that either the config for it was removed or the daemon
3022 * didn't get started. This handles graceful restart & retain scenario.
3024 if (PROTO_OWNED(nhe
) && nhe
->refcnt
== 1) {
3025 zebra_nhg_decrement_ref(nhe
);
3026 return HASHWALK_ABORT
;
3030 * If its being ref'd by routes, just let it be uninstalled via a route
3033 if (ZEBRA_NHG_CREATED(nhe
) && nhe
->refcnt
<= 0) {
3034 zebra_nhg_uninstall_kernel(nhe
);
3035 return HASHWALK_ABORT
;
3038 return HASHWALK_CONTINUE
;
3041 void zebra_nhg_sweep_table(struct hash
*hash
)
3046 * Yes this is extremely odd. Effectively nhg's have
3047 * other nexthop groups that depend on them and when you
3048 * remove them, you can have other entries blown up.
3049 * our hash code does not work with deleting multiple
3050 * entries at a time and will possibly cause crashes
3051 * So what to do? Whenever zebra_nhg_sweep_entry
3052 * deletes an entry it will return HASHWALK_ABORT,
3053 * cause that deletion might have triggered more.
3054 * then we can just keep sweeping this table
3055 * until nothing more is found to do.
3058 count
= hashcount(hash
);
3059 hash_walk(hash
, zebra_nhg_sweep_entry
, NULL
);
3060 } while (count
!= hashcount(hash
));
3063 static void zebra_nhg_mark_keep_entry(struct hash_bucket
*bucket
, void *arg
)
3065 struct nhg_hash_entry
*nhe
= bucket
->data
;
3067 UNSET_FLAG(nhe
->flags
, NEXTHOP_GROUP_INSTALLED
);
3071 * When we are shutting down and we have retain mode enabled
3072 * in zebra the process is to mark each vrf that it's
3073 * routes should not be deleted. The problem with that
3074 * is that shutdown actually free's up memory which
3075 * causes the nexthop group's ref counts to go to zero
3076 * we need a way to subtly tell the system to not remove
3077 * the nexthop groups from the kernel at the same time.
3078 * The easiest just looks like that we should not mark
3079 * the nhg's as installed any more and when the ref count
3080 * goes to zero we'll attempt to delete and do nothing
3082 void zebra_nhg_mark_keep(void)
3084 hash_iterate(zrouter
.nhgs_id
, zebra_nhg_mark_keep_entry
, NULL
);
3087 /* Global control to disable use of kernel nexthops, if available. We can't
3088 * force the kernel to support nexthop ids, of course, but we can disable
3089 * zebra's use of them, for testing e.g. By default, if the kernel supports
3090 * nexthop ids, zebra uses them.
3092 void zebra_nhg_enable_kernel_nexthops(bool set
)
3094 g_nexthops_enabled
= set
;
3097 bool zebra_nhg_kernel_nexthops_enabled(void)
3099 return g_nexthops_enabled
;
3102 /* Global control for use of activated backups for recursive resolution. */
3103 void zebra_nhg_set_recursive_use_backups(bool set
)
3105 use_recursive_backups
= set
;
3108 bool zebra_nhg_recursive_use_backups(void)
3110 return use_recursive_backups
;
3114 * Global control to only use kernel nexthops for protocol created NHGs.
3115 * There are some use cases where you may not want zebra to implicitly
3116 * create kernel nexthops for all routes and only create them for NHGs
3117 * passed down by upper level protos.
3121 void zebra_nhg_set_proto_nexthops_only(bool set
)
3123 proto_nexthops_only
= set
;
3126 bool zebra_nhg_proto_nexthops_only(void)
3128 return proto_nexthops_only
;
3131 /* Add NHE from upper level proto */
3132 struct nhg_hash_entry
*zebra_nhg_proto_add(uint32_t id
, int type
,
3133 uint16_t instance
, uint32_t session
,
3134 struct nexthop_group
*nhg
, afi_t afi
)
3136 struct nhg_hash_entry lookup
;
3137 struct nhg_hash_entry
*new, *old
;
3138 struct nhg_connected
*rb_node_dep
= NULL
;
3139 struct nexthop
*newhop
;
3140 bool replace
= false;
3142 if (!nhg
->nexthop
) {
3143 if (IS_ZEBRA_DEBUG_NHG
)
3144 zlog_debug("%s: id %u, no nexthops passed to add",
3150 /* Set nexthop list as active, since they wont go through rib
3153 * Assuming valid/onlink for now.
3155 * Once resolution is figured out, we won't need this!
3157 for (ALL_NEXTHOPS_PTR(nhg
, newhop
)) {
3158 if (CHECK_FLAG(newhop
->flags
, NEXTHOP_FLAG_HAS_BACKUP
)) {
3159 if (IS_ZEBRA_DEBUG_NHG
)
3161 "%s: id %u, backup nexthops not supported",
3166 if (newhop
->type
== NEXTHOP_TYPE_BLACKHOLE
) {
3167 if (IS_ZEBRA_DEBUG_NHG
)
3169 "%s: id %u, blackhole nexthop not supported",
3174 if (newhop
->type
== NEXTHOP_TYPE_IFINDEX
) {
3175 if (IS_ZEBRA_DEBUG_NHG
)
3177 "%s: id %u, nexthop without gateway not supported",
3182 if (!newhop
->ifindex
) {
3183 if (IS_ZEBRA_DEBUG_NHG
)
3185 "%s: id %u, nexthop without ifindex is not supported",
3189 SET_FLAG(newhop
->flags
, NEXTHOP_FLAG_ACTIVE
);
3192 zebra_nhe_init(&lookup
, afi
, nhg
->nexthop
);
3193 lookup
.nhg
.nexthop
= nhg
->nexthop
;
3197 old
= zebra_nhg_lookup_id(id
);
3201 * This is a replace, just release NHE from ID for now, The
3202 * depends/dependents may still be used in the replacement so
3203 * we don't touch them other than to remove their refs to their
3207 hash_release(zrouter
.nhgs_id
, old
);
3209 /* Free all the things */
3210 zebra_nhg_release_all_deps(old
);
3213 new = zebra_nhg_rib_find_nhe(&lookup
, afi
);
3215 zebra_nhg_increment_ref(new);
3217 /* Capture zapi client info */
3218 new->zapi_instance
= instance
;
3219 new->zapi_session
= session
;
3221 zebra_nhg_set_valid_if_active(new);
3223 zebra_nhg_install_kernel(new);
3227 * Check to handle recving DEL while routes still in use then
3230 * In this case we would have decremented the refcnt already
3231 * but set the FLAG here. Go ahead and increment once to fix
3232 * the misordering we have been sent.
3234 if (CHECK_FLAG(old
->flags
, NEXTHOP_GROUP_PROTO_RELEASED
))
3235 zebra_nhg_increment_ref(old
);
3237 rib_handle_nhg_replace(old
, new);
3239 /* if this != 1 at this point, we have a bug */
3240 assert(old
->refcnt
== 1);
3242 /* We have to decrement its singletons
3243 * because some might not exist in NEW.
3245 if (!zebra_nhg_depends_is_empty(old
)) {
3246 frr_each (nhg_connected_tree
, &old
->nhg_depends
,
3248 zebra_nhg_decrement_ref(rb_node_dep
->nhe
);
3251 /* Dont call the dec API, we dont want to uninstall the ID */
3253 zebra_nhg_free(old
);
3257 if (IS_ZEBRA_DEBUG_NHG_DETAIL
)
3258 zlog_debug("%s: %s nhe %p (%u), vrf %d, type %s", __func__
,
3259 (replace
? "replaced" : "added"), new, new->id
,
3260 new->vrf_id
, zebra_route_string(new->type
));
3265 /* Delete NHE from upper level proto, caller must decrement ref */
3266 struct nhg_hash_entry
*zebra_nhg_proto_del(uint32_t id
, int type
)
3268 struct nhg_hash_entry
*nhe
;
3270 nhe
= zebra_nhg_lookup_id(id
);
3273 if (IS_ZEBRA_DEBUG_NHG
)
3274 zlog_debug("%s: id %u, lookup failed", __func__
, id
);
3279 if (type
!= nhe
->type
) {
3280 if (IS_ZEBRA_DEBUG_NHG
)
3282 "%s: id %u, type %s mismatch, sent by %s, ignoring",
3283 __func__
, id
, zebra_route_string(nhe
->type
),
3284 zebra_route_string(type
));
3288 if (CHECK_FLAG(nhe
->flags
, NEXTHOP_GROUP_PROTO_RELEASED
)) {
3289 if (IS_ZEBRA_DEBUG_NHG
)
3290 zlog_debug("%s: id %u, already released", __func__
, id
);
3295 SET_FLAG(nhe
->flags
, NEXTHOP_GROUP_PROTO_RELEASED
);
3297 if (nhe
->refcnt
> 1) {
3298 if (IS_ZEBRA_DEBUG_NHG
)
3300 "%s: id %u, still being used by routes refcnt %u",
3301 __func__
, nhe
->id
, nhe
->refcnt
);
3305 if (IS_ZEBRA_DEBUG_NHG_DETAIL
)
3306 zlog_debug("%s: deleted nhe %p (%u), vrf %d, type %s", __func__
,
3307 nhe
, nhe
->id
, nhe
->vrf_id
,
3308 zebra_route_string(nhe
->type
));
3313 struct nhg_score_proto_iter
{
3318 static void zebra_nhg_score_proto_entry(struct hash_bucket
*bucket
, void *arg
)
3320 struct nhg_hash_entry
*nhe
;
3321 struct nhg_score_proto_iter
*iter
;
3323 nhe
= (struct nhg_hash_entry
*)bucket
->data
;
3326 /* Needs to match type and outside zebra ID space */
3327 if (nhe
->type
== iter
->type
&& PROTO_OWNED(nhe
)) {
3328 if (IS_ZEBRA_DEBUG_NHG_DETAIL
)
3330 "%s: found nhe %p (%u), vrf %d, type %s after client disconnect",
3331 __func__
, nhe
, nhe
->id
, nhe
->vrf_id
,
3332 zebra_route_string(nhe
->type
));
3334 /* Add to removal list */
3335 listnode_add(iter
->found
, nhe
);
3339 /* Remove specific by proto NHGs */
3340 unsigned long zebra_nhg_score_proto(int type
)
3342 struct nhg_hash_entry
*nhe
;
3343 struct nhg_score_proto_iter iter
= {};
3344 struct listnode
*ln
;
3345 unsigned long count
;
3348 iter
.found
= list_new();
3350 /* Find matching entries to remove */
3351 hash_iterate(zrouter
.nhgs_id
, zebra_nhg_score_proto_entry
, &iter
);
3353 /* Now remove them */
3354 for (ALL_LIST_ELEMENTS_RO(iter
.found
, ln
, nhe
)) {
3356 * This should be the last ref if we remove client routes too,
3357 * and thus should remove and free them.
3359 zebra_nhg_decrement_ref(nhe
);
3362 count
= iter
.found
->count
;
3363 list_delete(&iter
.found
);