1 /* Routing Information Base.
2 * Copyright (C) 1997, 98, 99, 2001 Kunihiro Ishiguro
4 * This file is part of GNU Zebra.
6 * GNU Zebra is free software; you can redistribute it and/or modify it
7 * under the terms of the GNU General Public License as published by the
8 * Free Software Foundation; either version 2, or (at your option) any
11 * GNU Zebra is distributed in the hope that it will be useful, but
12 * WITHOUT ANY WARRANTY; without even the implied warranty of
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
14 * General Public License for more details.
16 * You should have received a copy of the GNU General Public License along
17 * with this program; see the file COPYING; if not, write to the Free Software
18 * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
33 #include "sockunion.h"
34 #include "srcdest_table.h"
38 #include "workqueue.h"
39 #include "nexthop_group_private.h"
40 #include "frr_pthread.h"
42 #include "zebra/zebra_router.h"
43 #include "zebra/connected.h"
44 #include "zebra/debug.h"
45 #include "zebra/interface.h"
46 #include "zebra/redistribute.h"
47 #include "zebra/rib.h"
49 #include "zebra/zapi_msg.h"
50 #include "zebra/zebra_errors.h"
51 #include "zebra/zebra_memory.h"
52 #include "zebra/zebra_ns.h"
53 #include "zebra/zebra_rnh.h"
54 #include "zebra/zebra_routemap.h"
55 #include "zebra/zebra_vrf.h"
56 #include "zebra/zebra_vxlan.h"
57 #include "zebra/zapi_msg.h"
58 #include "zebra/zebra_dplane.h"
60 DEFINE_MTYPE_STATIC(ZEBRA
, RIB_UPDATE_CTX
, "Rib update context object");
63 * Event, list, and mutex for delivery of dataplane results
65 static pthread_mutex_t dplane_mutex
;
66 static struct thread
*t_dplane
;
67 static struct dplane_ctx_q rib_dplane_q
;
69 DEFINE_HOOK(rib_update
, (struct route_node
* rn
, const char *reason
),
72 /* Should we allow non Quagga processes to delete our routes */
73 extern int allow_delete
;
75 /* Each route type's string and default distance value. */
80 } route_info
[ZEBRA_ROUTE_MAX
] = {
81 [ZEBRA_ROUTE_NHG
] = {ZEBRA_ROUTE_NHG
, 255 /* Uneeded for nhg's */, 0},
82 [ZEBRA_ROUTE_SYSTEM
] = {ZEBRA_ROUTE_SYSTEM
, 0, 5},
83 [ZEBRA_ROUTE_KERNEL
] = {ZEBRA_ROUTE_KERNEL
, 0, 1},
84 [ZEBRA_ROUTE_CONNECT
] = {ZEBRA_ROUTE_CONNECT
, 0, 1},
85 [ZEBRA_ROUTE_STATIC
] = {ZEBRA_ROUTE_STATIC
, 1, 2},
86 [ZEBRA_ROUTE_RIP
] = {ZEBRA_ROUTE_RIP
, 120, 3},
87 [ZEBRA_ROUTE_RIPNG
] = {ZEBRA_ROUTE_RIPNG
, 120, 3},
88 [ZEBRA_ROUTE_OSPF
] = {ZEBRA_ROUTE_OSPF
, 110, 3},
89 [ZEBRA_ROUTE_OSPF6
] = {ZEBRA_ROUTE_OSPF6
, 110, 3},
90 [ZEBRA_ROUTE_ISIS
] = {ZEBRA_ROUTE_ISIS
, 115, 3},
91 [ZEBRA_ROUTE_BGP
] = {ZEBRA_ROUTE_BGP
, 20 /* IBGP is 200. */, 4},
92 [ZEBRA_ROUTE_PIM
] = {ZEBRA_ROUTE_PIM
, 255, 5},
93 [ZEBRA_ROUTE_EIGRP
] = {ZEBRA_ROUTE_EIGRP
, 90, 3},
94 [ZEBRA_ROUTE_NHRP
] = {ZEBRA_ROUTE_NHRP
, 10, 3},
95 [ZEBRA_ROUTE_HSLS
] = {ZEBRA_ROUTE_HSLS
, 255, 5},
96 [ZEBRA_ROUTE_OLSR
] = {ZEBRA_ROUTE_OLSR
, 255, 5},
97 [ZEBRA_ROUTE_TABLE
] = {ZEBRA_ROUTE_TABLE
, 150, 2},
98 [ZEBRA_ROUTE_LDP
] = {ZEBRA_ROUTE_LDP
, 150, 5},
99 [ZEBRA_ROUTE_VNC
] = {ZEBRA_ROUTE_VNC
, 20, 4},
100 [ZEBRA_ROUTE_VNC_DIRECT
] = {ZEBRA_ROUTE_VNC_DIRECT
, 20, 4},
101 [ZEBRA_ROUTE_VNC_DIRECT_RH
] = {ZEBRA_ROUTE_VNC_DIRECT_RH
, 20, 4},
102 [ZEBRA_ROUTE_BGP_DIRECT
] = {ZEBRA_ROUTE_BGP_DIRECT
, 20, 4},
103 [ZEBRA_ROUTE_BGP_DIRECT_EXT
] = {ZEBRA_ROUTE_BGP_DIRECT_EXT
, 20, 4},
104 [ZEBRA_ROUTE_BABEL
] = {ZEBRA_ROUTE_BABEL
, 100, 3},
105 [ZEBRA_ROUTE_SHARP
] = {ZEBRA_ROUTE_SHARP
, 150, 5},
106 [ZEBRA_ROUTE_PBR
] = {ZEBRA_ROUTE_PBR
, 200, 5},
107 [ZEBRA_ROUTE_BFD
] = {ZEBRA_ROUTE_BFD
, 255, 5},
108 [ZEBRA_ROUTE_OPENFABRIC
] = {ZEBRA_ROUTE_OPENFABRIC
, 115, 3},
109 [ZEBRA_ROUTE_VRRP
] = {ZEBRA_ROUTE_VRRP
, 255, 5}
110 /* Any new route type added to zebra, should be mirrored here */
112 /* no entry/default: 150 */
115 static void __attribute__((format(printf
, 5, 6)))
116 _rnode_zlog(const char *_func
, vrf_id_t vrf_id
, struct route_node
*rn
,
117 int priority
, const char *msgfmt
, ...)
119 char buf
[SRCDEST2STR_BUFFER
+ sizeof(" (MRIB)")];
123 va_start(ap
, msgfmt
);
124 vsnprintf(msgbuf
, sizeof(msgbuf
), msgfmt
, ap
);
128 rib_table_info_t
*info
= srcdest_rnode_table_info(rn
);
129 srcdest_rnode2str(rn
, buf
, sizeof(buf
));
131 if (info
->safi
== SAFI_MULTICAST
)
132 strlcat(buf
, " (MRIB)", sizeof(buf
));
134 snprintf(buf
, sizeof(buf
), "{(route_node *) NULL}");
137 zlog(priority
, "%s: %d:%s: %s", _func
, vrf_id
, buf
, msgbuf
);
140 #define rnode_debug(node, vrf_id, ...) \
141 _rnode_zlog(__func__, vrf_id, node, LOG_DEBUG, __VA_ARGS__)
142 #define rnode_info(node, ...) \
143 _rnode_zlog(__func__, vrf_id, node, LOG_INFO, __VA_ARGS__)
145 uint8_t route_distance(int type
)
149 if ((unsigned)type
>= array_size(route_info
))
152 distance
= route_info
[type
].distance
;
157 int is_zebra_valid_kernel_table(uint32_t table_id
)
160 if ((table_id
== RT_TABLE_UNSPEC
) || (table_id
== RT_TABLE_LOCAL
)
161 || (table_id
== RT_TABLE_COMPAT
))
168 int is_zebra_main_routing_table(uint32_t table_id
)
170 if (table_id
== RT_TABLE_MAIN
)
175 int zebra_check_addr(const struct prefix
*p
)
177 if (p
->family
== AF_INET
) {
180 addr
= p
->u
.prefix4
.s_addr
;
183 if (IPV4_NET127(addr
) || IN_CLASSD(addr
)
184 || IPV4_LINKLOCAL(addr
))
187 if (p
->family
== AF_INET6
) {
188 if (IN6_IS_ADDR_LOOPBACK(&p
->u
.prefix6
))
190 if (IN6_IS_ADDR_LINKLOCAL(&p
->u
.prefix6
))
196 /* Add nexthop to the end of a rib node's nexthop list */
197 void route_entry_nexthop_add(struct route_entry
*re
, struct nexthop
*nexthop
)
199 _nexthop_group_add_sorted(re
->ng
, nexthop
);
204 * copy_nexthop - copy a nexthop to the rib structure.
206 void route_entry_copy_nexthops(struct route_entry
*re
, struct nexthop
*nh
)
208 assert(!re
->ng
->nexthop
);
209 copy_nexthops(&re
->ng
->nexthop
, nh
, NULL
);
212 /* Delete specified nexthop from the list. */
213 void route_entry_nexthop_delete(struct route_entry
*re
, struct nexthop
*nexthop
)
216 nexthop
->next
->prev
= nexthop
->prev
;
218 nexthop
->prev
->next
= nexthop
->next
;
220 re
->ng
->nexthop
= nexthop
->next
;
224 struct nexthop
*route_entry_nexthop_ifindex_add(struct route_entry
*re
,
228 struct nexthop
*nexthop
;
230 nexthop
= nexthop_new();
231 nexthop
->type
= NEXTHOP_TYPE_IFINDEX
;
232 nexthop
->ifindex
= ifindex
;
233 nexthop
->vrf_id
= nh_vrf_id
;
235 route_entry_nexthop_add(re
, nexthop
);
240 struct nexthop
*route_entry_nexthop_ipv4_add(struct route_entry
*re
,
241 struct in_addr
*ipv4
,
245 struct nexthop
*nexthop
;
247 nexthop
= nexthop_new();
248 nexthop
->type
= NEXTHOP_TYPE_IPV4
;
249 nexthop
->vrf_id
= nh_vrf_id
;
250 nexthop
->gate
.ipv4
= *ipv4
;
252 nexthop
->src
.ipv4
= *src
;
254 route_entry_nexthop_add(re
, nexthop
);
259 struct nexthop
*route_entry_nexthop_ipv4_ifindex_add(struct route_entry
*re
,
260 struct in_addr
*ipv4
,
265 struct nexthop
*nexthop
;
266 struct interface
*ifp
;
268 nexthop
= nexthop_new();
269 nexthop
->vrf_id
= nh_vrf_id
;
270 nexthop
->type
= NEXTHOP_TYPE_IPV4_IFINDEX
;
271 nexthop
->gate
.ipv4
= *ipv4
;
273 nexthop
->src
.ipv4
= *src
;
274 nexthop
->ifindex
= ifindex
;
275 ifp
= if_lookup_by_index(nexthop
->ifindex
, nh_vrf_id
);
276 /*Pending: need to think if null ifp here is ok during bootup?
277 There was a crash because ifp here was coming to be NULL */
279 if (connected_is_unnumbered(ifp
))
280 SET_FLAG(nexthop
->flags
, NEXTHOP_FLAG_ONLINK
);
282 route_entry_nexthop_add(re
, nexthop
);
287 struct nexthop
*route_entry_nexthop_ipv6_add(struct route_entry
*re
,
288 struct in6_addr
*ipv6
,
291 struct nexthop
*nexthop
;
293 nexthop
= nexthop_new();
294 nexthop
->vrf_id
= nh_vrf_id
;
295 nexthop
->type
= NEXTHOP_TYPE_IPV6
;
296 nexthop
->gate
.ipv6
= *ipv6
;
298 route_entry_nexthop_add(re
, nexthop
);
303 struct nexthop
*route_entry_nexthop_ipv6_ifindex_add(struct route_entry
*re
,
304 struct in6_addr
*ipv6
,
308 struct nexthop
*nexthop
;
310 nexthop
= nexthop_new();
311 nexthop
->vrf_id
= nh_vrf_id
;
312 nexthop
->type
= NEXTHOP_TYPE_IPV6_IFINDEX
;
313 nexthop
->gate
.ipv6
= *ipv6
;
314 nexthop
->ifindex
= ifindex
;
316 route_entry_nexthop_add(re
, nexthop
);
321 struct nexthop
*route_entry_nexthop_blackhole_add(struct route_entry
*re
,
322 enum blackhole_type bh_type
)
324 struct nexthop
*nexthop
;
326 nexthop
= nexthop_new();
327 nexthop
->vrf_id
= VRF_DEFAULT
;
328 nexthop
->type
= NEXTHOP_TYPE_BLACKHOLE
;
329 nexthop
->bh_type
= bh_type
;
331 route_entry_nexthop_add(re
, nexthop
);
336 struct route_entry
*rib_match(afi_t afi
, safi_t safi
, vrf_id_t vrf_id
,
337 union g_addr
*addr
, struct route_node
**rn_out
)
340 struct route_table
*table
;
341 struct route_node
*rn
;
342 struct route_entry
*match
= NULL
;
345 table
= zebra_vrf_table(afi
, safi
, vrf_id
);
349 memset(&p
, 0, sizeof(struct prefix
));
352 p
.u
.prefix4
= addr
->ipv4
;
353 p
.prefixlen
= IPV4_MAX_PREFIXLEN
;
355 p
.u
.prefix6
= addr
->ipv6
;
356 p
.prefixlen
= IPV6_MAX_PREFIXLEN
;
359 rn
= route_node_match(table
, (struct prefix
*)&p
);
364 route_unlock_node(rn
);
366 dest
= rib_dest_from_rnode(rn
);
367 if (dest
&& dest
->selected_fib
368 && !CHECK_FLAG(dest
->selected_fib
->status
,
369 ROUTE_ENTRY_REMOVED
))
370 match
= dest
->selected_fib
;
372 /* If there is no selected route or matched route is EGP, go up
377 } while (rn
&& rn
->info
== NULL
);
381 if (match
->type
!= ZEBRA_ROUTE_CONNECT
) {
382 if (!CHECK_FLAG(match
->status
,
383 ROUTE_ENTRY_INSTALLED
))
395 struct route_entry
*rib_match_ipv4_multicast(vrf_id_t vrf_id
,
397 struct route_node
**rn_out
)
399 struct route_entry
*re
= NULL
, *mre
= NULL
, *ure
= NULL
;
400 struct route_node
*m_rn
= NULL
, *u_rn
= NULL
;
401 union g_addr gaddr
= {.ipv4
= addr
};
403 switch (zrouter
.ipv4_multicast_mode
) {
404 case MCAST_MRIB_ONLY
:
405 return rib_match(AFI_IP
, SAFI_MULTICAST
, vrf_id
, &gaddr
,
407 case MCAST_URIB_ONLY
:
408 return rib_match(AFI_IP
, SAFI_UNICAST
, vrf_id
, &gaddr
, rn_out
);
409 case MCAST_NO_CONFIG
:
410 case MCAST_MIX_MRIB_FIRST
:
411 re
= mre
= rib_match(AFI_IP
, SAFI_MULTICAST
, vrf_id
, &gaddr
,
414 re
= ure
= rib_match(AFI_IP
, SAFI_UNICAST
, vrf_id
,
417 case MCAST_MIX_DISTANCE
:
418 mre
= rib_match(AFI_IP
, SAFI_MULTICAST
, vrf_id
, &gaddr
, &m_rn
);
419 ure
= rib_match(AFI_IP
, SAFI_UNICAST
, vrf_id
, &gaddr
, &u_rn
);
421 re
= ure
->distance
< mre
->distance
? ure
: mre
;
427 case MCAST_MIX_PFXLEN
:
428 mre
= rib_match(AFI_IP
, SAFI_MULTICAST
, vrf_id
, &gaddr
, &m_rn
);
429 ure
= rib_match(AFI_IP
, SAFI_UNICAST
, vrf_id
, &gaddr
, &u_rn
);
431 re
= u_rn
->p
.prefixlen
> m_rn
->p
.prefixlen
? ure
: mre
;
440 *rn_out
= (re
== mre
) ? m_rn
: u_rn
;
442 if (IS_ZEBRA_DEBUG_RIB
) {
444 inet_ntop(AF_INET
, &addr
, buf
, BUFSIZ
);
446 zlog_debug("%s: %s: vrf: %u found %s, using %s",
447 __func__
, buf
, vrf_id
,
448 mre
? (ure
? "MRIB+URIB" : "MRIB")
449 : ure
? "URIB" : "nothing",
450 re
== ure
? "URIB" : re
== mre
? "MRIB" : "none");
455 struct route_entry
*rib_lookup_ipv4(struct prefix_ipv4
*p
, vrf_id_t vrf_id
)
457 struct route_table
*table
;
458 struct route_node
*rn
;
459 struct route_entry
*match
= NULL
;
463 table
= zebra_vrf_table(AFI_IP
, SAFI_UNICAST
, vrf_id
);
467 rn
= route_node_lookup(table
, (struct prefix
*)p
);
469 /* No route for this prefix. */
474 route_unlock_node(rn
);
475 dest
= rib_dest_from_rnode(rn
);
477 if (dest
&& dest
->selected_fib
478 && !CHECK_FLAG(dest
->selected_fib
->status
, ROUTE_ENTRY_REMOVED
))
479 match
= dest
->selected_fib
;
484 if (match
->type
== ZEBRA_ROUTE_CONNECT
)
487 if (CHECK_FLAG(match
->status
, ROUTE_ENTRY_INSTALLED
))
494 * Is this RIB labeled-unicast? It must be of type BGP and all paths
495 * (nexthops) must have a label.
497 int zebra_rib_labeled_unicast(struct route_entry
*re
)
499 struct nexthop
*nexthop
= NULL
;
501 if (re
->type
!= ZEBRA_ROUTE_BGP
)
504 for (ALL_NEXTHOPS_PTR(re
->ng
, nexthop
))
505 if (!nexthop
->nh_label
|| !nexthop
->nh_label
->num_labels
)
511 /* Update flag indicates whether this is a "replace" or not. Currently, this
512 * is only used for IPv4.
514 void rib_install_kernel(struct route_node
*rn
, struct route_entry
*re
,
515 struct route_entry
*old
)
517 struct nexthop
*nexthop
;
518 rib_table_info_t
*info
= srcdest_rnode_table_info(rn
);
519 struct zebra_vrf
*zvrf
= vrf_info_lookup(re
->vrf_id
);
520 const struct prefix
*p
, *src_p
;
521 enum zebra_dplane_result ret
;
522 struct nhg_hash_entry
*nhe
;
524 rib_dest_t
*dest
= rib_dest_from_rnode(rn
);
526 srcdest_rnode_prefixes(rn
, &p
, &src_p
);
528 if (info
->safi
!= SAFI_UNICAST
) {
529 for (ALL_NEXTHOPS_PTR(re
->ng
, nexthop
))
530 SET_FLAG(nexthop
->flags
, NEXTHOP_FLAG_FIB
);
533 struct nexthop
*prev
;
535 for (ALL_NEXTHOPS_PTR(re
->ng
, nexthop
)) {
536 UNSET_FLAG (nexthop
->flags
, NEXTHOP_FLAG_DUPLICATE
);
537 for (ALL_NEXTHOPS_PTR(re
->ng
, prev
)) {
540 if (nexthop_same_firsthop(nexthop
, prev
)) {
541 SET_FLAG(nexthop
->flags
,
542 NEXTHOP_FLAG_DUPLICATE
);
550 * Install the resolved nexthop object first.
552 nhe
= zebra_nhg_resolve(zebra_nhg_lookup_id(re
->nhe_id
));
553 if (!nhe
->is_kernel_nh
)
554 zebra_nhg_install_kernel(nhe
);
557 * If this is a replace to a new RE let the originator of the RE
558 * know that they've lost
560 if (old
&& (old
!= re
) && (old
->type
!= re
->type
))
561 zsend_route_notify_owner(old
, p
, ZAPI_ROUTE_BETTER_ADMIN_WON
);
563 /* Update fib selection */
564 dest
->selected_fib
= re
;
567 * Make sure we update the FPM any time we send new information to
570 hook_call(rib_update
, rn
, "installing in kernel");
572 /* Send add or update */
574 ret
= dplane_route_update(rn
, re
, old
);
576 ret
= dplane_route_add(rn
, re
);
579 case ZEBRA_DPLANE_REQUEST_QUEUED
:
580 SET_FLAG(re
->status
, ROUTE_ENTRY_QUEUED
);
583 SET_FLAG(old
->status
, ROUTE_ENTRY_QUEUED
);
585 /* Free old FIB nexthop group */
586 if (old
->fib_ng
.nexthop
) {
587 nexthops_free(old
->fib_ng
.nexthop
);
588 old
->fib_ng
.nexthop
= NULL
;
591 if (!RIB_SYSTEM_ROUTE(old
)) {
592 /* Clear old route's FIB flags */
593 for (ALL_NEXTHOPS_PTR(old
->ng
, nexthop
)) {
594 UNSET_FLAG(nexthop
->flags
,
601 zvrf
->installs_queued
++;
603 case ZEBRA_DPLANE_REQUEST_FAILURE
:
605 char str
[SRCDEST2STR_BUFFER
];
607 srcdest_rnode2str(rn
, str
, sizeof(str
));
608 flog_err(EC_ZEBRA_DP_INSTALL_FAIL
,
609 "%u:%s: Failed to enqueue dataplane install",
613 case ZEBRA_DPLANE_REQUEST_SUCCESS
:
622 /* Uninstall the route from kernel. */
623 void rib_uninstall_kernel(struct route_node
*rn
, struct route_entry
*re
)
625 struct nexthop
*nexthop
;
626 rib_table_info_t
*info
= srcdest_rnode_table_info(rn
);
627 struct zebra_vrf
*zvrf
= vrf_info_lookup(re
->vrf_id
);
629 if (info
->safi
!= SAFI_UNICAST
) {
630 UNSET_FLAG(re
->status
, ROUTE_ENTRY_INSTALLED
);
631 for (ALL_NEXTHOPS_PTR(re
->ng
, nexthop
))
632 UNSET_FLAG(nexthop
->flags
, NEXTHOP_FLAG_FIB
);
637 * Make sure we update the FPM any time we send new information to
640 hook_call(rib_update
, rn
, "uninstalling from kernel");
642 switch (dplane_route_delete(rn
, re
)) {
643 case ZEBRA_DPLANE_REQUEST_QUEUED
:
645 zvrf
->removals_queued
++;
647 case ZEBRA_DPLANE_REQUEST_FAILURE
:
649 char str
[SRCDEST2STR_BUFFER
];
651 srcdest_rnode2str(rn
, str
, sizeof(str
));
652 flog_err(EC_ZEBRA_DP_INSTALL_FAIL
,
653 "%u:%s: Failed to enqueue dataplane uninstall",
657 case ZEBRA_DPLANE_REQUEST_SUCCESS
:
666 /* Uninstall the route from kernel. */
667 static void rib_uninstall(struct route_node
*rn
, struct route_entry
*re
)
669 rib_table_info_t
*info
= srcdest_rnode_table_info(rn
);
670 rib_dest_t
*dest
= rib_dest_from_rnode(rn
);
671 struct nexthop
*nexthop
;
673 if (dest
&& dest
->selected_fib
== re
) {
674 if (info
->safi
== SAFI_UNICAST
)
675 hook_call(rib_update
, rn
, "rib_uninstall");
677 /* If labeled-unicast route, uninstall transit LSP. */
678 if (zebra_rib_labeled_unicast(re
))
679 zebra_mpls_lsp_uninstall(info
->zvrf
, rn
, re
);
681 rib_uninstall_kernel(rn
, re
);
683 dest
->selected_fib
= NULL
;
685 /* Free FIB nexthop group, if present */
686 if (re
->fib_ng
.nexthop
) {
687 nexthops_free(re
->fib_ng
.nexthop
);
688 re
->fib_ng
.nexthop
= NULL
;
691 for (ALL_NEXTHOPS_PTR(re
->ng
, nexthop
))
692 UNSET_FLAG(nexthop
->flags
, NEXTHOP_FLAG_FIB
);
695 if (CHECK_FLAG(re
->flags
, ZEBRA_FLAG_SELECTED
)) {
696 const struct prefix
*p
, *src_p
;
698 srcdest_rnode_prefixes(rn
, &p
, &src_p
);
700 redistribute_delete(p
, src_p
, re
, NULL
);
701 UNSET_FLAG(re
->flags
, ZEBRA_FLAG_SELECTED
);
706 * rib_can_delete_dest
708 * Returns true if the given dest can be deleted from the table.
710 static int rib_can_delete_dest(rib_dest_t
*dest
)
712 if (re_list_first(&dest
->routes
)) {
717 * Unresolved rnh's are stored on the default route's list
719 * dest->rnode can also be the source prefix node in an
720 * ipv6 sourcedest table. Fortunately the prefix of a
721 * source prefix node can never be the default prefix.
723 if (is_default_prefix(&dest
->rnode
->p
))
727 * Don't delete the dest if we have to update the FPM about this
730 if (CHECK_FLAG(dest
->flags
, RIB_DEST_UPDATE_FPM
)
731 || CHECK_FLAG(dest
->flags
, RIB_DEST_SENT_TO_FPM
))
737 void zebra_rib_evaluate_rn_nexthops(struct route_node
*rn
, uint32_t seq
)
739 rib_dest_t
*dest
= rib_dest_from_rnode(rn
);
743 * We are storing the rnh's associated withb
744 * the tracked nexthop as a list of the rn's.
745 * Unresolved rnh's are placed at the top
746 * of the tree list.( 0.0.0.0/0 for v4 and 0::0/0 for v6 )
747 * As such for each rn we need to walk up the tree
748 * and see if any rnh's need to see if they
749 * would match a more specific route
752 if (IS_ZEBRA_DEBUG_NHT_DETAILED
) {
753 char buf
[PREFIX_STRLEN
];
755 zlog_debug("%s: %s Being examined for Nexthop Tracking Count: %zd",
757 srcdest_rnode2str(rn
, buf
, sizeof(buf
)),
758 dest
? rnh_list_count(&dest
->nht
) : 0);
763 dest
= rib_dest_from_rnode(rn
);
767 * If we have any rnh's stored in the nht list
768 * then we know that this route node was used for
769 * nht resolution and as such we need to call the
770 * nexthop tracking evaluation code
772 frr_each_safe(rnh_list
, &dest
->nht
, rnh
) {
773 struct zebra_vrf
*zvrf
=
774 zebra_vrf_lookup_by_id(rnh
->vrf_id
);
775 struct prefix
*p
= &rnh
->node
->p
;
777 if (IS_ZEBRA_DEBUG_NHT_DETAILED
) {
778 char buf1
[PREFIX_STRLEN
];
779 char buf2
[PREFIX_STRLEN
];
781 zlog_debug("%u:%s has Nexthop(%s) Type: %s depending on it, evaluating %u:%u",
783 srcdest_rnode2str(rn
, buf1
,
785 prefix2str(p
, buf2
, sizeof(buf2
)),
786 rnh_type2str(rnh
->type
),
791 * If we have evaluated this node on this pass
792 * already, due to following the tree up
793 * then we know that we can move onto the next
796 * Additionally we call zebra_evaluate_rnh
797 * when we gc the dest. In this case we know
798 * that there must be no other re's where
799 * we were originally as such we know that
800 * that sequence number is ok to respect.
802 if (rnh
->seqno
== seq
) {
803 if (IS_ZEBRA_DEBUG_NHT_DETAILED
)
805 "\tNode processed and moved already");
810 zebra_evaluate_rnh(zvrf
, family2afi(p
->family
), 0,
816 dest
= rib_dest_from_rnode(rn
);
823 * Garbage collect the rib dest corresponding to the given route node
826 * Returns true if the dest was deleted, false otherwise.
828 int rib_gc_dest(struct route_node
*rn
)
832 dest
= rib_dest_from_rnode(rn
);
836 if (!rib_can_delete_dest(dest
))
839 if (IS_ZEBRA_DEBUG_RIB
) {
840 struct zebra_vrf
*zvrf
;
842 zvrf
= rib_dest_vrf(dest
);
843 rnode_debug(rn
, zvrf_id(zvrf
), "removing dest from table");
846 zebra_rib_evaluate_rn_nexthops(rn
, zebra_router_get_next_sequence());
849 rnh_list_fini(&dest
->nht
);
850 XFREE(MTYPE_RIB_DEST
, dest
);
854 * Release the one reference that we keep on the route node.
856 route_unlock_node(rn
);
860 static void rib_process_add_fib(struct zebra_vrf
*zvrf
, struct route_node
*rn
,
861 struct route_entry
*new)
863 hook_call(rib_update
, rn
, "new route selected");
865 /* Update real nexthop. This may actually determine if nexthop is active
867 if (!nexthop_group_active_nexthop_num(new->ng
)) {
868 UNSET_FLAG(new->status
, ROUTE_ENTRY_CHANGED
);
872 if (IS_ZEBRA_DEBUG_RIB
) {
873 char buf
[SRCDEST2STR_BUFFER
];
874 srcdest_rnode2str(rn
, buf
, sizeof(buf
));
875 zlog_debug("%u:%s: Adding route rn %p, re %p (%s)",
876 zvrf_id(zvrf
), buf
, rn
, new,
877 zebra_route_string(new->type
));
880 /* If labeled-unicast route, install transit LSP. */
881 if (zebra_rib_labeled_unicast(new))
882 zebra_mpls_lsp_install(zvrf
, rn
, new);
884 rib_install_kernel(rn
, new, NULL
);
886 UNSET_FLAG(new->status
, ROUTE_ENTRY_CHANGED
);
889 static void rib_process_del_fib(struct zebra_vrf
*zvrf
, struct route_node
*rn
,
890 struct route_entry
*old
)
892 hook_call(rib_update
, rn
, "removing existing route");
894 /* Uninstall from kernel. */
895 if (IS_ZEBRA_DEBUG_RIB
) {
896 char buf
[SRCDEST2STR_BUFFER
];
897 srcdest_rnode2str(rn
, buf
, sizeof(buf
));
898 zlog_debug("%u:%s: Deleting route rn %p, re %p (%s)",
899 zvrf_id(zvrf
), buf
, rn
, old
,
900 zebra_route_string(old
->type
));
903 /* If labeled-unicast route, uninstall transit LSP. */
904 if (zebra_rib_labeled_unicast(old
))
905 zebra_mpls_lsp_uninstall(zvrf
, rn
, old
);
907 rib_uninstall_kernel(rn
, old
);
909 /* Update nexthop for route, reset changed flag. */
910 /* Note: this code also handles the Linux case when an interface goes
911 * down, causing the kernel to delete routes without sending DELROUTE
914 if (RIB_KERNEL_ROUTE(old
))
915 SET_FLAG(old
->status
, ROUTE_ENTRY_REMOVED
);
917 UNSET_FLAG(old
->status
, ROUTE_ENTRY_CHANGED
);
920 static void rib_process_update_fib(struct zebra_vrf
*zvrf
,
921 struct route_node
*rn
,
922 struct route_entry
*old
,
923 struct route_entry
*new)
928 * We have to install or update if a new route has been selected or
929 * something has changed.
931 if (new != old
|| CHECK_FLAG(new->status
, ROUTE_ENTRY_CHANGED
)) {
932 hook_call(rib_update
, rn
, "updating existing route");
934 /* Update the nexthop; we could determine here that nexthop is
936 if (nexthop_group_active_nexthop_num(new->ng
))
939 /* If nexthop is active, install the selected route, if
941 * the install succeeds, cleanup flags for prior route, if
946 if (IS_ZEBRA_DEBUG_RIB
) {
947 char buf
[SRCDEST2STR_BUFFER
];
948 srcdest_rnode2str(rn
, buf
, sizeof(buf
));
951 "%u:%s: Updating route rn %p, re %p (%s) old %p (%s)",
952 zvrf_id(zvrf
), buf
, rn
, new,
953 zebra_route_string(new->type
),
955 zebra_route_string(old
->type
));
958 "%u:%s: Updating route rn %p, re %p (%s)",
959 zvrf_id(zvrf
), buf
, rn
, new,
960 zebra_route_string(new->type
));
963 /* If labeled-unicast route, uninstall transit LSP. */
964 if (zebra_rib_labeled_unicast(old
))
965 zebra_mpls_lsp_uninstall(zvrf
, rn
, old
);
968 * Non-system route should be installed.
969 * If labeled-unicast route, install transit
972 if (zebra_rib_labeled_unicast(new))
973 zebra_mpls_lsp_install(zvrf
, rn
, new);
975 rib_install_kernel(rn
, new, old
);
979 * If nexthop for selected route is not active or install
981 * may need to uninstall and delete for redistribution.
984 if (IS_ZEBRA_DEBUG_RIB
) {
985 char buf
[SRCDEST2STR_BUFFER
];
986 srcdest_rnode2str(rn
, buf
, sizeof(buf
));
989 "%u:%s: Deleting route rn %p, re %p (%s) old %p (%s) - nexthop inactive",
990 zvrf_id(zvrf
), buf
, rn
, new,
991 zebra_route_string(new->type
),
993 zebra_route_string(old
->type
));
996 "%u:%s: Deleting route rn %p, re %p (%s) - nexthop inactive",
997 zvrf_id(zvrf
), buf
, rn
, new,
998 zebra_route_string(new->type
));
1001 /* If labeled-unicast route, uninstall transit LSP. */
1002 if (zebra_rib_labeled_unicast(old
))
1003 zebra_mpls_lsp_uninstall(zvrf
, rn
, old
);
1005 rib_uninstall_kernel(rn
, old
);
1009 * Same route selected; check if in the FIB and if not,
1010 * re-install. This is housekeeping code to deal with
1011 * race conditions in kernel with linux netlink reporting
1012 * interface up before IPv4 or IPv6 protocol is ready
1015 if (!CHECK_FLAG(new->status
, ROUTE_ENTRY_INSTALLED
) ||
1016 RIB_SYSTEM_ROUTE(new))
1017 rib_install_kernel(rn
, new, NULL
);
1020 /* Update prior route. */
1022 UNSET_FLAG(old
->status
, ROUTE_ENTRY_CHANGED
);
1024 /* Clear changed flag. */
1025 UNSET_FLAG(new->status
, ROUTE_ENTRY_CHANGED
);
1028 /* Check if 'alternate' RIB entry is better than 'current'. */
1029 static struct route_entry
*rib_choose_best(struct route_entry
*current
,
1030 struct route_entry
*alternate
)
1032 if (current
== NULL
)
1035 /* filter route selection in following order:
1036 * - connected beats other types
1037 * - if both connected, loopback or vrf wins
1038 * - lower distance beats higher
1039 * - lower metric beats higher for equal distance
1040 * - last, hence oldest, route wins tie break.
1043 /* Connected routes. Check to see if either are a vrf
1044 * or loopback interface. If not, pick the last connected
1045 * route of the set of lowest metric connected routes.
1047 if (alternate
->type
== ZEBRA_ROUTE_CONNECT
) {
1048 if (current
->type
!= ZEBRA_ROUTE_CONNECT
)
1051 /* both are connected. are either loop or vrf? */
1052 struct nexthop
*nexthop
= NULL
;
1054 for (ALL_NEXTHOPS_PTR(alternate
->ng
, nexthop
)) {
1055 struct interface
*ifp
= if_lookup_by_index(
1056 nexthop
->ifindex
, alternate
->vrf_id
);
1058 if (ifp
&& if_is_loopback_or_vrf(ifp
))
1062 for (ALL_NEXTHOPS_PTR(current
->ng
, nexthop
)) {
1063 struct interface
*ifp
= if_lookup_by_index(
1064 nexthop
->ifindex
, current
->vrf_id
);
1066 if (ifp
&& if_is_loopback_or_vrf(ifp
))
1070 /* Neither are loop or vrf so pick best metric */
1071 if (alternate
->metric
<= current
->metric
)
1077 if (current
->type
== ZEBRA_ROUTE_CONNECT
)
1080 /* higher distance loses */
1081 if (alternate
->distance
< current
->distance
)
1083 if (current
->distance
< alternate
->distance
)
1086 /* metric tie-breaks equal distance */
1087 if (alternate
->metric
<= current
->metric
)
1093 /* Core function for processing nexthop group contexts's off metaq */
1094 static void rib_nhg_process(struct nhg_ctx
*ctx
)
1096 nhg_ctx_process(ctx
);
1099 /* Core function for processing routing information base. */
1100 static void rib_process(struct route_node
*rn
)
1102 struct route_entry
*re
;
1103 struct route_entry
*next
;
1104 struct route_entry
*old_selected
= NULL
;
1105 struct route_entry
*new_selected
= NULL
;
1106 struct route_entry
*old_fib
= NULL
;
1107 struct route_entry
*new_fib
= NULL
;
1108 struct route_entry
*best
= NULL
;
1109 char buf
[SRCDEST2STR_BUFFER
];
1111 struct zebra_vrf
*zvrf
= NULL
;
1112 const struct prefix
*p
, *src_p
;
1114 srcdest_rnode_prefixes(rn
, &p
, &src_p
);
1115 vrf_id_t vrf_id
= VRF_UNKNOWN
;
1119 dest
= rib_dest_from_rnode(rn
);
1121 zvrf
= rib_dest_vrf(dest
);
1122 vrf_id
= zvrf_id(zvrf
);
1125 if (IS_ZEBRA_DEBUG_RIB
)
1126 srcdest_rnode2str(rn
, buf
, sizeof(buf
));
1128 if (IS_ZEBRA_DEBUG_RIB_DETAILED
)
1129 zlog_debug("%u:%s: Processing rn %p", vrf_id
, buf
, rn
);
1132 * we can have rn's that have a NULL info pointer
1133 * (dest). As such let's not let the deref happen
1134 * additionally we know RNODE_FOREACH_RE_SAFE
1135 * will not iterate so we are ok.
1138 old_fib
= dest
->selected_fib
;
1140 RNODE_FOREACH_RE_SAFE (rn
, re
, next
) {
1141 if (IS_ZEBRA_DEBUG_RIB_DETAILED
)
1143 "%u:%s: Examine re %p (%s) status %x flags %x dist %d metric %d",
1144 vrf_id
, buf
, re
, zebra_route_string(re
->type
),
1145 re
->status
, re
->flags
, re
->distance
,
1148 /* Currently selected re. */
1149 if (CHECK_FLAG(re
->flags
, ZEBRA_FLAG_SELECTED
)) {
1150 assert(old_selected
== NULL
);
1154 /* Skip deleted entries from selection */
1155 if (CHECK_FLAG(re
->status
, ROUTE_ENTRY_REMOVED
))
1158 /* Skip unreachable nexthop. */
1159 /* This first call to nexthop_active_update is merely to
1160 * determine if there's any change to nexthops associated
1161 * with this RIB entry. Now, rib_process() can be invoked due
1162 * to an external event such as link down or due to
1163 * next-hop-tracking evaluation. In the latter case,
1164 * a decision has already been made that the NHs have changed.
1165 * So, no need to invoke a potentially expensive call again.
1166 * Further, since the change might be in a recursive NH which
1167 * is not caught in the nexthop_active_update() code. Thus, we
1168 * might miss changes to recursive NHs.
1170 if (CHECK_FLAG(re
->status
, ROUTE_ENTRY_CHANGED
)
1171 && !nexthop_active_update(rn
, re
)) {
1172 if (re
->type
== ZEBRA_ROUTE_TABLE
) {
1173 /* XXX: HERE BE DRAGONS!!!!!
1174 * In all honesty, I have not yet figured out
1175 * what this part does or why the
1176 * ROUTE_ENTRY_CHANGED test above is correct
1177 * or why we need to delete a route here, and
1178 * also not whether this concerns both selected
1179 * and fib route, or only selected
1182 * This entry was denied by the 'ip protocol
1183 * table' route-map, we need to delete it */
1184 if (re
!= old_selected
) {
1185 if (IS_ZEBRA_DEBUG_RIB
)
1187 "%s: %u:%s: imported via import-table but denied "
1188 "by the ip protocol table route-map",
1189 __func__
, vrf_id
, buf
);
1192 SET_FLAG(re
->status
,
1193 ROUTE_ENTRY_REMOVED
);
1199 /* Infinite distance. */
1200 if (re
->distance
== DISTANCE_INFINITY
) {
1201 UNSET_FLAG(re
->status
, ROUTE_ENTRY_CHANGED
);
1205 if (CHECK_FLAG(re
->flags
, ZEBRA_FLAG_FIB_OVERRIDE
)) {
1206 best
= rib_choose_best(new_fib
, re
);
1207 if (new_fib
&& best
!= new_fib
)
1208 UNSET_FLAG(new_fib
->status
,
1209 ROUTE_ENTRY_CHANGED
);
1212 best
= rib_choose_best(new_selected
, re
);
1213 if (new_selected
&& best
!= new_selected
)
1214 UNSET_FLAG(new_selected
->status
,
1215 ROUTE_ENTRY_CHANGED
);
1216 new_selected
= best
;
1219 UNSET_FLAG(re
->status
, ROUTE_ENTRY_CHANGED
);
1220 } /* RNODE_FOREACH_RE */
1222 /* If no FIB override route, use the selected route also for FIB */
1223 if (new_fib
== NULL
)
1224 new_fib
= new_selected
;
1226 /* After the cycle is finished, the following pointers will be set:
1227 * old_selected --- RE entry currently having SELECTED
1228 * new_selected --- RE entry that is newly SELECTED
1229 * old_fib --- RE entry currently in kernel FIB
1230 * new_fib --- RE entry that is newly to be in kernel FIB
1232 * new_selected will get SELECTED flag, and is going to be redistributed
1233 * the zclients. new_fib (which can be new_selected) will be installed
1237 if (IS_ZEBRA_DEBUG_RIB_DETAILED
) {
1239 "%u:%s: After processing: old_selected %p new_selected %p old_fib %p new_fib %p",
1240 vrf_id
, buf
, (void *)old_selected
, (void *)new_selected
,
1241 (void *)old_fib
, (void *)new_fib
);
1244 /* Buffer ROUTE_ENTRY_CHANGED here, because it will get cleared if
1245 * fib == selected */
1246 bool selected_changed
= new_selected
&& CHECK_FLAG(new_selected
->status
,
1247 ROUTE_ENTRY_CHANGED
);
1249 /* Update fib according to selection results */
1250 if (new_fib
&& old_fib
)
1251 rib_process_update_fib(zvrf
, rn
, old_fib
, new_fib
);
1253 rib_process_add_fib(zvrf
, rn
, new_fib
);
1255 rib_process_del_fib(zvrf
, rn
, old_fib
);
1257 /* Update SELECTED entry */
1258 if (old_selected
!= new_selected
|| selected_changed
) {
1260 if (new_selected
&& new_selected
!= new_fib
)
1261 UNSET_FLAG(new_selected
->status
, ROUTE_ENTRY_CHANGED
);
1264 SET_FLAG(new_selected
->flags
, ZEBRA_FLAG_SELECTED
);
1268 * If we're removing the old entry, we should tell
1269 * redist subscribers about that *if* they aren't
1270 * going to see a redist for the new entry.
1272 if (!new_selected
|| CHECK_FLAG(old_selected
->status
,
1273 ROUTE_ENTRY_REMOVED
))
1274 redistribute_delete(p
, src_p
,
1278 if (old_selected
!= new_selected
)
1279 UNSET_FLAG(old_selected
->flags
,
1280 ZEBRA_FLAG_SELECTED
);
1284 /* Remove all RE entries queued for removal */
1285 RNODE_FOREACH_RE_SAFE (rn
, re
, next
) {
1286 if (CHECK_FLAG(re
->status
, ROUTE_ENTRY_REMOVED
)) {
1287 if (IS_ZEBRA_DEBUG_RIB
) {
1288 rnode_debug(rn
, vrf_id
, "rn %p, removing re %p",
1289 (void *)rn
, (void *)re
);
1296 * Check if the dest can be deleted now.
1301 static void zebra_rib_evaluate_mpls(struct route_node
*rn
)
1303 rib_dest_t
*dest
= rib_dest_from_rnode(rn
);
1304 struct zebra_vrf
*zvrf
= vrf_info_lookup(VRF_DEFAULT
);
1309 if (CHECK_FLAG(dest
->flags
, RIB_DEST_UPDATE_LSPS
)) {
1310 if (IS_ZEBRA_DEBUG_MPLS
)
1312 "%u: Scheduling all LSPs upon RIB completion",
1314 zebra_mpls_lsp_schedule(zvrf
);
1315 mpls_unmark_lsps_for_processing(rn
);
1320 * Utility to match route with dplane context data
1322 static bool rib_route_match_ctx(const struct route_entry
*re
,
1323 const struct zebra_dplane_ctx
*ctx
,
1326 bool result
= false;
1330 * In 'update' case, we test info about the 'previous' or
1333 if ((re
->type
== dplane_ctx_get_old_type(ctx
)) &&
1334 (re
->instance
== dplane_ctx_get_old_instance(ctx
))) {
1337 /* TODO -- we're using this extra test, but it's not
1338 * exactly clear why.
1340 if (re
->type
== ZEBRA_ROUTE_STATIC
&&
1341 (re
->distance
!= dplane_ctx_get_old_distance(ctx
) ||
1342 re
->tag
!= dplane_ctx_get_old_tag(ctx
))) {
1349 * Ordinary, single-route case using primary context info
1351 if ((dplane_ctx_get_op(ctx
) != DPLANE_OP_ROUTE_DELETE
) &&
1352 CHECK_FLAG(re
->status
, ROUTE_ENTRY_REMOVED
)) {
1353 /* Skip route that's been deleted */
1357 if ((re
->type
== dplane_ctx_get_type(ctx
)) &&
1358 (re
->instance
== dplane_ctx_get_instance(ctx
))) {
1361 /* TODO -- we're using this extra test, but it's not
1362 * exactly clear why.
1364 if (re
->type
== ZEBRA_ROUTE_STATIC
&&
1365 (re
->distance
!= dplane_ctx_get_distance(ctx
) ||
1366 re
->tag
!= dplane_ctx_get_tag(ctx
))) {
1377 static void zebra_rib_fixup_system(struct route_node
*rn
)
1379 struct route_entry
*re
;
1381 RNODE_FOREACH_RE(rn
, re
) {
1382 struct nexthop
*nhop
;
1384 if (!RIB_SYSTEM_ROUTE(re
))
1387 if (CHECK_FLAG(re
->status
, ROUTE_ENTRY_REMOVED
))
1390 SET_FLAG(re
->status
, ROUTE_ENTRY_INSTALLED
);
1391 UNSET_FLAG(re
->status
, ROUTE_ENTRY_QUEUED
);
1393 for (ALL_NEXTHOPS_PTR(re
->ng
, nhop
)) {
1394 if (CHECK_FLAG(nhop
->flags
, NEXTHOP_FLAG_RECURSIVE
))
1397 SET_FLAG(nhop
->flags
, NEXTHOP_FLAG_FIB
);
1403 * Update a route from a dplane context. This consolidates common code
1404 * that can be used in processing of results from FIB updates, and in
1405 * async notification processing.
1406 * The return is 'true' if the installed nexthops changed; 'false' otherwise.
1408 static bool rib_update_re_from_ctx(struct route_entry
*re
,
1409 struct route_node
*rn
,
1410 struct zebra_dplane_ctx
*ctx
)
1412 char dest_str
[PREFIX_STRLEN
] = "";
1413 char nh_str
[NEXTHOP_STRLEN
];
1414 struct nexthop
*nexthop
, *ctx_nexthop
;
1416 const struct nexthop_group
*ctxnhg
;
1417 bool is_selected
= false; /* Is 're' currently the selected re? */
1418 bool changed_p
= false; /* Change to nexthops? */
1421 /* Note well: only capturing the prefix string if debug is enabled here;
1422 * unconditional log messages will have to generate the string.
1424 if (IS_ZEBRA_DEBUG_RIB
)
1425 prefix2str(&(rn
->p
), dest_str
, sizeof(dest_str
));
1427 dest
= rib_dest_from_rnode(rn
);
1429 is_selected
= (re
== dest
->selected_fib
);
1431 if (IS_ZEBRA_DEBUG_RIB_DETAILED
)
1432 zlog_debug("update_from_ctx: %u:%s: %sSELECTED",
1433 re
->vrf_id
, dest_str
, (is_selected
? "" : "NOT "));
1435 /* Update zebra's nexthop FIB flag for each nexthop that was installed.
1436 * If the installed set differs from the set requested by the rib/owner,
1437 * we use the fib-specific nexthop-group to record the actual FIB
1442 * First check the fib nexthop-group, if it's present. The comparison
1443 * here is quite strict: we require that the fib sets match exactly.
1447 if (re
->fib_ng
.nexthop
== NULL
)
1452 /* First check the route's fib nexthops */
1453 for (ALL_NEXTHOPS(re
->fib_ng
, nexthop
)) {
1455 if (CHECK_FLAG(nexthop
->flags
, NEXTHOP_FLAG_RECURSIVE
))
1459 for (ALL_NEXTHOPS_PTR(dplane_ctx_get_ng(ctx
),
1461 if (nexthop_same(ctx_nexthop
, nexthop
))
1465 if (ctx_nexthop
== NULL
) {
1466 /* Nexthop not in the new installed set */
1467 if (IS_ZEBRA_DEBUG_RIB_DETAILED
) {
1468 nexthop2str(nexthop
, nh_str
,
1470 zlog_debug("update_from_ctx: no match for fib nh %s",
1482 /* Check the new installed set */
1484 for (ALL_NEXTHOPS_PTR(dplane_ctx_get_ng(ctx
), ctx_nexthop
)) {
1486 if (CHECK_FLAG(ctx_nexthop
->flags
,
1487 NEXTHOP_FLAG_RECURSIVE
))
1490 /* Compare with the current group's nexthops */
1492 for (ALL_NEXTHOPS(re
->fib_ng
, nexthop
)) {
1493 if (nexthop_same(nexthop
, ctx_nexthop
))
1497 if (nexthop
== NULL
) {
1498 /* Nexthop not in the old installed set */
1499 if (IS_ZEBRA_DEBUG_RIB_DETAILED
) {
1500 nexthop2str(ctx_nexthop
, nh_str
,
1502 zlog_debug("update_from_ctx: no fib match for notif nh %s",
1512 /* If the new FIB set matches the existing FIB set, we're done. */
1514 if (IS_ZEBRA_DEBUG_RIB
)
1515 zlog_debug("%u:%s update_from_ctx(): existing fib nhg, no change",
1516 re
->vrf_id
, dest_str
);
1519 } else if (re
->fib_ng
.nexthop
) {
1521 * Free stale fib list and move on to check the rib nhg.
1523 if (IS_ZEBRA_DEBUG_RIB
)
1524 zlog_debug("%u:%s update_from_ctx(): replacing fib nhg",
1525 re
->vrf_id
, dest_str
);
1526 nexthops_free(re
->fib_ng
.nexthop
);
1527 re
->fib_ng
.nexthop
= NULL
;
1529 /* Note that the installed nexthops have changed */
1532 if (IS_ZEBRA_DEBUG_RIB
)
1533 zlog_debug("%u:%s update_from_ctx(): no fib nhg",
1534 re
->vrf_id
, dest_str
);
1538 * Compare with the rib nexthop group. The comparison here is different:
1539 * the RIB group may be a superset of the list installed in the FIB. We
1540 * walk the RIB group, looking for the 'installable' candidate
1541 * nexthops, and then check those against the set
1542 * that is actually installed.
1545 for (ALL_NEXTHOPS_PTR(re
->ng
, nexthop
)) {
1547 if (CHECK_FLAG(nexthop
->flags
, NEXTHOP_FLAG_RECURSIVE
))
1550 if (!CHECK_FLAG(nexthop
->flags
, NEXTHOP_FLAG_ACTIVE
))
1553 /* Check for a FIB nexthop corresponding to the RIB nexthop */
1555 for (ALL_NEXTHOPS_PTR(dplane_ctx_get_ng(ctx
), ctx_nexthop
)) {
1556 if (nexthop_same(ctx_nexthop
, nexthop
))
1560 /* If the FIB doesn't know about the nexthop,
1561 * it's not installed
1563 if (ctx_nexthop
== NULL
) {
1564 if (IS_ZEBRA_DEBUG_RIB_DETAILED
) {
1565 nexthop2str(nexthop
, nh_str
, sizeof(nh_str
));
1566 zlog_debug("update_from_ctx: no notif match for rib nh %s",
1571 if (CHECK_FLAG(nexthop
->flags
, NEXTHOP_FLAG_FIB
))
1574 UNSET_FLAG(nexthop
->flags
, NEXTHOP_FLAG_FIB
);
1576 /* Keep checking nexthops */
1580 if (CHECK_FLAG(ctx_nexthop
->flags
, NEXTHOP_FLAG_FIB
)) {
1581 if (!CHECK_FLAG(nexthop
->flags
, NEXTHOP_FLAG_FIB
))
1584 SET_FLAG(nexthop
->flags
, NEXTHOP_FLAG_FIB
);
1586 if (CHECK_FLAG(nexthop
->flags
, NEXTHOP_FLAG_FIB
))
1589 UNSET_FLAG(nexthop
->flags
, NEXTHOP_FLAG_FIB
);
1593 /* If all nexthops were processed, we're done */
1595 if (IS_ZEBRA_DEBUG_RIB
)
1596 zlog_debug("%u:%s update_from_ctx(): rib nhg matched, changed '%s'",
1597 re
->vrf_id
, dest_str
,
1598 (changed_p
? "true" : "false"));
1602 /* FIB nexthop set differs from the RIB set:
1603 * create a fib-specific nexthop-group
1605 if (IS_ZEBRA_DEBUG_RIB
)
1606 zlog_debug("%u:%s update_from_ctx(): changed %s, adding new fib nhg",
1607 re
->vrf_id
, dest_str
,
1608 (changed_p
? "true" : "false"));
1610 ctxnhg
= dplane_ctx_get_ng(ctx
);
1612 if (ctxnhg
->nexthop
)
1613 copy_nexthops(&(re
->fib_ng
.nexthop
), ctxnhg
->nexthop
, NULL
);
1615 /* Bit of a special case when the fib has _no_ installed
1618 nexthop
= nexthop_new();
1619 nexthop
->type
= NEXTHOP_TYPE_IPV4
;
1620 _nexthop_add(&(re
->fib_ng
.nexthop
), nexthop
);
1628 * Helper to locate a zebra route-node from a dplane context. This is used
1629 * when processing dplane results, e.g. Note well: the route-node is returned
1630 * with a ref held - route_unlock_node() must be called eventually.
1632 static struct route_node
*
1633 rib_find_rn_from_ctx(const struct zebra_dplane_ctx
*ctx
)
1635 struct route_table
*table
= NULL
;
1636 struct route_node
*rn
= NULL
;
1637 const struct prefix
*dest_pfx
, *src_pfx
;
1639 /* Locate rn and re(s) from ctx */
1641 table
= zebra_vrf_table_with_table_id(dplane_ctx_get_afi(ctx
),
1642 dplane_ctx_get_safi(ctx
),
1643 dplane_ctx_get_vrf(ctx
),
1644 dplane_ctx_get_table(ctx
));
1645 if (table
== NULL
) {
1646 if (IS_ZEBRA_DEBUG_DPLANE
) {
1647 zlog_debug("Failed to find route for ctx: no table for afi %d, safi %d, vrf %u",
1648 dplane_ctx_get_afi(ctx
),
1649 dplane_ctx_get_safi(ctx
),
1650 dplane_ctx_get_vrf(ctx
));
1655 dest_pfx
= dplane_ctx_get_dest(ctx
);
1656 src_pfx
= dplane_ctx_get_src(ctx
);
1658 rn
= srcdest_rnode_get(table
, dest_pfx
,
1659 src_pfx
? (struct prefix_ipv6
*)src_pfx
: NULL
);
1668 * Route-update results processing after async dataplane update.
1670 static void rib_process_result(struct zebra_dplane_ctx
*ctx
)
1672 struct zebra_vrf
*zvrf
= NULL
;
1673 struct route_node
*rn
= NULL
;
1674 struct route_entry
*re
= NULL
, *old_re
= NULL
, *rib
;
1675 bool is_update
= false;
1676 char dest_str
[PREFIX_STRLEN
] = "";
1677 enum dplane_op_e op
;
1678 enum zebra_dplane_result status
;
1679 const struct prefix
*dest_pfx
, *src_pfx
;
1681 bool fib_changed
= false;
1683 zvrf
= vrf_info_lookup(dplane_ctx_get_vrf(ctx
));
1684 dest_pfx
= dplane_ctx_get_dest(ctx
);
1686 /* Note well: only capturing the prefix string if debug is enabled here;
1687 * unconditional log messages will have to generate the string.
1689 if (IS_ZEBRA_DEBUG_DPLANE
)
1690 prefix2str(dest_pfx
, dest_str
, sizeof(dest_str
));
1692 /* Locate rn and re(s) from ctx */
1693 rn
= rib_find_rn_from_ctx(ctx
);
1695 if (IS_ZEBRA_DEBUG_DPLANE
) {
1696 zlog_debug("Failed to process dplane results: no route for %u:%s",
1697 dplane_ctx_get_vrf(ctx
), dest_str
);
1702 srcdest_rnode_prefixes(rn
, &dest_pfx
, &src_pfx
);
1704 op
= dplane_ctx_get_op(ctx
);
1705 status
= dplane_ctx_get_status(ctx
);
1707 if (IS_ZEBRA_DEBUG_DPLANE_DETAIL
)
1708 zlog_debug("%u:%s Processing dplane ctx %p, op %s result %s",
1709 dplane_ctx_get_vrf(ctx
), dest_str
, ctx
,
1710 dplane_op2str(op
), dplane_res2str(status
));
1713 * Update is a bit of a special case, where we may have both old and new
1714 * routes to post-process.
1716 is_update
= dplane_ctx_is_update(ctx
);
1719 * Take a pass through the routes, look for matches with the context
1722 RNODE_FOREACH_RE(rn
, rib
) {
1725 if (rib_route_match_ctx(rib
, ctx
, false))
1729 /* Check for old route match */
1730 if (is_update
&& (old_re
== NULL
)) {
1731 if (rib_route_match_ctx(rib
, ctx
, true /*is_update*/))
1735 /* Have we found the routes we need to work on? */
1736 if (re
&& ((!is_update
|| old_re
)))
1740 seq
= dplane_ctx_get_seq(ctx
);
1743 * Check sequence number(s) to detect stale results before continuing
1746 if (re
->dplane_sequence
!= seq
) {
1747 if (IS_ZEBRA_DEBUG_DPLANE_DETAIL
)
1748 zlog_debug("%u:%s Stale dplane result for re %p",
1749 dplane_ctx_get_vrf(ctx
),
1752 UNSET_FLAG(re
->status
, ROUTE_ENTRY_QUEUED
);
1756 if (old_re
->dplane_sequence
!= dplane_ctx_get_old_seq(ctx
)) {
1757 if (IS_ZEBRA_DEBUG_DPLANE_DETAIL
)
1758 zlog_debug("%u:%s Stale dplane result for old_re %p",
1759 dplane_ctx_get_vrf(ctx
),
1762 UNSET_FLAG(old_re
->status
, ROUTE_ENTRY_QUEUED
);
1766 case DPLANE_OP_ROUTE_INSTALL
:
1767 case DPLANE_OP_ROUTE_UPDATE
:
1768 if (status
== ZEBRA_DPLANE_REQUEST_SUCCESS
) {
1770 UNSET_FLAG(re
->status
, ROUTE_ENTRY_FAILED
);
1771 SET_FLAG(re
->status
, ROUTE_ENTRY_INSTALLED
);
1774 * On an update operation from the same route type
1775 * context retrieval currently has no way to know
1776 * which was the old and which was the new.
1777 * So don't unset our flags that we just set.
1778 * We know redistribution is ok because the
1779 * old_re in this case is used for nothing
1780 * more than knowing whom to contact if necessary.
1782 if (old_re
&& old_re
!= re
) {
1783 UNSET_FLAG(old_re
->status
, ROUTE_ENTRY_FAILED
);
1784 UNSET_FLAG(old_re
->status
,
1785 ROUTE_ENTRY_INSTALLED
);
1788 /* Update zebra route based on the results in
1789 * the context struct.
1793 rib_update_re_from_ctx(re
, rn
, ctx
);
1796 if (IS_ZEBRA_DEBUG_DPLANE_DETAIL
)
1797 zlog_debug("%u:%s no fib change for re",
1804 redistribute_update(dest_pfx
, src_pfx
,
1809 * System routes are weird in that they
1810 * allow multiple to be installed that match
1811 * to the same prefix, so after we get the
1812 * result we need to clean them up so that
1813 * we can actually use them.
1815 if ((re
&& RIB_SYSTEM_ROUTE(re
)) ||
1816 (old_re
&& RIB_SYSTEM_ROUTE(old_re
)))
1817 zebra_rib_fixup_system(rn
);
1822 /* Notify route owner */
1823 zsend_route_notify_owner_ctx(ctx
, ZAPI_ROUTE_INSTALLED
);
1827 SET_FLAG(re
->status
, ROUTE_ENTRY_FAILED
);
1828 UNSET_FLAG(re
->status
, ROUTE_ENTRY_INSTALLED
);
1830 SET_FLAG(old_re
->status
, ROUTE_ENTRY_FAILED
);
1832 zsend_route_notify_owner(re
, dest_pfx
,
1833 ZAPI_ROUTE_FAIL_INSTALL
);
1835 zlog_warn("%u:%s: Route install failed",
1836 dplane_ctx_get_vrf(ctx
),
1837 prefix2str(dest_pfx
,
1838 dest_str
, sizeof(dest_str
)));
1841 case DPLANE_OP_ROUTE_DELETE
:
1843 SET_FLAG(re
->status
, ROUTE_ENTRY_FAILED
);
1845 * In the delete case, the zebra core datastructs were
1846 * updated (or removed) at the time the delete was issued,
1847 * so we're just notifying the route owner.
1849 if (status
== ZEBRA_DPLANE_REQUEST_SUCCESS
) {
1851 UNSET_FLAG(re
->status
, ROUTE_ENTRY_INSTALLED
);
1852 UNSET_FLAG(re
->status
, ROUTE_ENTRY_FAILED
);
1854 zsend_route_notify_owner_ctx(ctx
, ZAPI_ROUTE_REMOVED
);
1860 SET_FLAG(re
->status
, ROUTE_ENTRY_FAILED
);
1861 zsend_route_notify_owner_ctx(ctx
,
1862 ZAPI_ROUTE_REMOVE_FAIL
);
1864 zlog_warn("%u:%s: Route Deletion failure",
1865 dplane_ctx_get_vrf(ctx
),
1866 prefix2str(dest_pfx
,
1867 dest_str
, sizeof(dest_str
)));
1871 * System routes are weird in that they
1872 * allow multiple to be installed that match
1873 * to the same prefix, so after we get the
1874 * result we need to clean them up so that
1875 * we can actually use them.
1877 if ((re
&& RIB_SYSTEM_ROUTE(re
)) ||
1878 (old_re
&& RIB_SYSTEM_ROUTE(old_re
)))
1879 zebra_rib_fixup_system(rn
);
1885 zebra_rib_evaluate_rn_nexthops(rn
, seq
);
1886 zebra_rib_evaluate_mpls(rn
);
1890 route_unlock_node(rn
);
1892 /* Return context to dataplane module */
1893 dplane_ctx_fini(&ctx
);
1897 * Handle notification from async dataplane: the dataplane has detected
1898 * some change to a route, and notifies zebra so that the control plane
1899 * can reflect that change.
1901 static void rib_process_dplane_notify(struct zebra_dplane_ctx
*ctx
)
1903 struct route_node
*rn
= NULL
;
1904 struct route_entry
*re
= NULL
;
1905 struct nexthop
*nexthop
;
1906 char dest_str
[PREFIX_STRLEN
] = "";
1907 const struct prefix
*dest_pfx
, *src_pfx
;
1909 bool fib_changed
= false;
1910 bool debug_p
= IS_ZEBRA_DEBUG_DPLANE
| IS_ZEBRA_DEBUG_RIB
;
1911 int start_count
, end_count
;
1912 dest_pfx
= dplane_ctx_get_dest(ctx
);
1914 /* Note well: only capturing the prefix string if debug is enabled here;
1915 * unconditional log messages will have to generate the string.
1918 prefix2str(dest_pfx
, dest_str
, sizeof(dest_str
));
1920 /* Locate rn and re(s) from ctx */
1921 rn
= rib_find_rn_from_ctx(ctx
);
1924 zlog_debug("Failed to process dplane notification: no routes for %u:%s",
1925 dplane_ctx_get_vrf(ctx
), dest_str
);
1930 dest
= rib_dest_from_rnode(rn
);
1931 srcdest_rnode_prefixes(rn
, &dest_pfx
, &src_pfx
);
1934 zlog_debug("%u:%s Processing dplane notif ctx %p",
1935 dplane_ctx_get_vrf(ctx
), dest_str
, ctx
);
1938 * Take a pass through the routes, look for matches with the context
1941 RNODE_FOREACH_RE(rn
, re
) {
1942 if (rib_route_match_ctx(re
, ctx
, false /*!update*/))
1946 /* No match? Nothing we can do */
1949 zlog_debug("%u:%s Unable to process dplane notification: no entry for type %s",
1950 dplane_ctx_get_vrf(ctx
), dest_str
,
1952 dplane_ctx_get_type(ctx
)));
1957 /* Ensure we clear the QUEUED flag */
1958 UNSET_FLAG(re
->status
, ROUTE_ENTRY_QUEUED
);
1960 /* Is this a notification that ... matters? We only really care about
1961 * the route that is currently selected for installation.
1963 if (re
!= dest
->selected_fib
) {
1964 /* TODO -- don't skip processing entirely? We might like to
1965 * at least report on the event.
1968 zlog_debug("%u:%s dplane notif, but type %s not selected_fib",
1969 dplane_ctx_get_vrf(ctx
), dest_str
,
1971 dplane_ctx_get_type(ctx
)));
1975 /* We'll want to determine whether the installation status of the
1976 * route has changed: we'll check the status before processing,
1977 * and then again if there's been a change.
1980 for (ALL_NEXTHOPS_PTR(rib_active_nhg(re
), nexthop
)) {
1981 if (CHECK_FLAG(nexthop
->flags
, NEXTHOP_FLAG_FIB
))
1985 /* Update zebra's nexthop FIB flags based on the context struct's
1988 fib_changed
= rib_update_re_from_ctx(re
, rn
, ctx
);
1992 zlog_debug("%u:%s No change from dplane notification",
1993 dplane_ctx_get_vrf(ctx
), dest_str
);
1999 * Perform follow-up work if the actual status of the prefix
2004 for (ALL_NEXTHOPS_PTR(rib_active_nhg(re
), nexthop
)) {
2005 if (CHECK_FLAG(nexthop
->flags
, NEXTHOP_FLAG_FIB
))
2009 /* Various fib transitions: changed nexthops; from installed to
2010 * not-installed; or not-installed to installed.
2012 if (start_count
> 0 && end_count
> 0) {
2014 zlog_debug("%u:%s applied nexthop changes from dplane notification",
2015 dplane_ctx_get_vrf(ctx
), dest_str
);
2017 /* Changed nexthops - update kernel/others */
2018 dplane_route_notif_update(rn
, re
,
2019 DPLANE_OP_ROUTE_UPDATE
, ctx
);
2021 } else if (start_count
== 0 && end_count
> 0) {
2023 zlog_debug("%u:%s installed transition from dplane notification",
2024 dplane_ctx_get_vrf(ctx
), dest_str
);
2026 /* We expect this to be the selected route, so we want
2027 * to tell others about this transition.
2029 SET_FLAG(re
->status
, ROUTE_ENTRY_INSTALLED
);
2031 /* Changed nexthops - update kernel/others */
2032 dplane_route_notif_update(rn
, re
, DPLANE_OP_ROUTE_INSTALL
, ctx
);
2034 /* Redistribute, lsp, and nht update */
2035 redistribute_update(dest_pfx
, src_pfx
, re
, NULL
);
2037 zebra_rib_evaluate_rn_nexthops(
2038 rn
, zebra_router_get_next_sequence());
2040 zebra_rib_evaluate_mpls(rn
);
2042 } else if (start_count
> 0 && end_count
== 0) {
2044 zlog_debug("%u:%s un-installed transition from dplane notification",
2045 dplane_ctx_get_vrf(ctx
), dest_str
);
2047 /* Transition from _something_ installed to _nothing_
2050 /* We expect this to be the selected route, so we want
2051 * to tell others about this transistion.
2053 UNSET_FLAG(re
->status
, ROUTE_ENTRY_INSTALLED
);
2055 /* Changed nexthops - update kernel/others */
2056 dplane_route_notif_update(rn
, re
, DPLANE_OP_ROUTE_DELETE
, ctx
);
2058 /* Redistribute, lsp, and nht update */
2059 redistribute_delete(dest_pfx
, src_pfx
, re
, NULL
);
2061 zebra_rib_evaluate_rn_nexthops(
2062 rn
, zebra_router_get_next_sequence());
2064 zebra_rib_evaluate_mpls(rn
);
2069 route_unlock_node(rn
);
2071 /* Return context to dataplane module */
2072 dplane_ctx_fini(&ctx
);
2075 static void process_subq_nhg(struct listnode
*lnode
)
2077 struct nhg_ctx
*ctx
= NULL
;
2078 uint8_t qindex
= route_info
[ZEBRA_ROUTE_NHG
].meta_q_map
;
2080 ctx
= listgetdata(lnode
);
2085 if (IS_ZEBRA_DEBUG_RIB_DETAILED
)
2086 zlog_debug("NHG Context id=%u dequeued from sub-queue %u",
2089 rib_nhg_process(ctx
);
2092 static void process_subq_route(struct listnode
*lnode
, uint8_t qindex
)
2094 struct route_node
*rnode
= NULL
;
2095 rib_dest_t
*dest
= NULL
;
2096 struct zebra_vrf
*zvrf
= NULL
;
2098 rnode
= listgetdata(lnode
);
2099 dest
= rib_dest_from_rnode(rnode
);
2101 zvrf
= rib_dest_vrf(dest
);
2105 if (IS_ZEBRA_DEBUG_RIB_DETAILED
) {
2106 char buf
[SRCDEST2STR_BUFFER
];
2108 srcdest_rnode2str(rnode
, buf
, sizeof(buf
));
2109 zlog_debug("%u:%s: rn %p dequeued from sub-queue %u",
2110 zvrf
? zvrf_id(zvrf
) : 0, buf
, rnode
, qindex
);
2114 UNSET_FLAG(rib_dest_from_rnode(rnode
)->flags
,
2115 RIB_ROUTE_QUEUED(qindex
));
2120 zlog_debug ("%s: called for route_node (%p, %d) with no ribs",
2121 __func__
, rnode
, rnode
->lock
);
2122 zlog_backtrace(LOG_DEBUG
);
2125 route_unlock_node(rnode
);
2128 /* Take a list of route_node structs and return 1, if there was a record
2129 * picked from it and processed by rib_process(). Don't process more,
2130 * than one RN record; operate only in the specified sub-queue.
2132 static unsigned int process_subq(struct list
*subq
, uint8_t qindex
)
2134 struct listnode
*lnode
= listhead(subq
);
2139 if (qindex
== route_info
[ZEBRA_ROUTE_NHG
].meta_q_map
)
2140 process_subq_nhg(lnode
);
2142 process_subq_route(lnode
, qindex
);
2144 list_delete_node(subq
, lnode
);
2149 /* Dispatch the meta queue by picking, processing and unlocking the next RN from
2150 * a non-empty sub-queue with lowest priority. wq is equal to zebra->ribq and
2152 * is pointed to the meta queue structure.
2154 static wq_item_status
meta_queue_process(struct work_queue
*dummy
, void *data
)
2156 struct meta_queue
*mq
= data
;
2158 uint32_t queue_len
, queue_limit
;
2160 /* Ensure there's room for more dataplane updates */
2161 queue_limit
= dplane_get_in_queue_limit();
2162 queue_len
= dplane_get_in_queue_len();
2163 if (queue_len
> queue_limit
) {
2164 if (IS_ZEBRA_DEBUG_RIB_DETAILED
)
2165 zlog_debug("rib queue: dplane queue len %u, limit %u, retrying",
2166 queue_len
, queue_limit
);
2168 /* Ensure that the meta-queue is actually enqueued */
2169 if (work_queue_empty(zrouter
.ribq
))
2170 work_queue_add(zrouter
.ribq
, zrouter
.mq
);
2172 return WQ_QUEUE_BLOCKED
;
2175 for (i
= 0; i
< MQ_SIZE
; i
++)
2176 if (process_subq(mq
->subq
[i
], i
)) {
2180 return mq
->size
? WQ_REQUEUE
: WQ_SUCCESS
;
2185 * Look into the RN and queue it into the highest priority queue
2186 * at this point in time for processing.
2188 * We will enqueue a route node only once per invocation.
2190 * There are two possibilities here that should be kept in mind.
2191 * If the original invocation has not been pulled off for processing
2192 * yet, A subsuquent invocation can have a route entry with a better
2193 * meta queue index value and we can have a situation where
2194 * we might have the same node enqueued 2 times. Not necessarily
2195 * an optimal situation but it should be ok.
2197 * The other possibility is that the original invocation has not
2198 * been pulled off for processing yet, A subsusquent invocation
2199 * doesn't have a route_entry with a better meta-queue and the
2200 * original metaqueue index value will win and we'll end up with
2201 * the route node enqueued once.
2203 static int rib_meta_queue_add(struct meta_queue
*mq
, void *data
)
2205 struct route_node
*rn
= NULL
;
2206 struct route_entry
*re
= NULL
, *curr_re
= NULL
;
2207 uint8_t qindex
= MQ_SIZE
, curr_qindex
= MQ_SIZE
;
2209 rn
= (struct route_node
*)data
;
2211 RNODE_FOREACH_RE (rn
, curr_re
) {
2212 curr_qindex
= route_info
[curr_re
->type
].meta_q_map
;
2214 if (curr_qindex
<= qindex
) {
2216 qindex
= curr_qindex
;
2223 /* Invariant: at this point we always have rn->info set. */
2224 if (CHECK_FLAG(rib_dest_from_rnode(rn
)->flags
,
2225 RIB_ROUTE_QUEUED(qindex
))) {
2226 if (IS_ZEBRA_DEBUG_RIB_DETAILED
)
2227 rnode_debug(rn
, re
->vrf_id
,
2228 "rn %p is already queued in sub-queue %u",
2229 (void *)rn
, qindex
);
2233 SET_FLAG(rib_dest_from_rnode(rn
)->flags
, RIB_ROUTE_QUEUED(qindex
));
2234 listnode_add(mq
->subq
[qindex
], rn
);
2235 route_lock_node(rn
);
2238 if (IS_ZEBRA_DEBUG_RIB_DETAILED
)
2239 rnode_debug(rn
, re
->vrf_id
, "queued rn %p into sub-queue %u",
2240 (void *)rn
, qindex
);
2245 static int rib_meta_queue_nhg_add(struct meta_queue
*mq
, void *data
)
2247 struct nhg_ctx
*ctx
= NULL
;
2248 uint8_t qindex
= route_info
[ZEBRA_ROUTE_NHG
].meta_q_map
;
2250 ctx
= (struct nhg_ctx
*)data
;
2255 listnode_add(mq
->subq
[qindex
], ctx
);
2258 if (IS_ZEBRA_DEBUG_RIB_DETAILED
)
2259 zlog_debug("NHG Context id=%u queued into sub-queue %u",
2265 static int mq_add_handler(void *data
,
2266 int (*mq_add_func
)(struct meta_queue
*mq
, void *data
))
2268 if (zrouter
.ribq
== NULL
) {
2269 flog_err(EC_ZEBRA_WQ_NONEXISTENT
,
2270 "%s: work_queue does not exist!", __func__
);
2275 * The RIB queue should normally be either empty or holding the only
2276 * work_queue_item element. In the latter case this element would
2277 * hold a pointer to the meta queue structure, which must be used to
2278 * actually queue the route nodes to process. So create the MQ
2279 * holder, if necessary, then push the work into it in any case.
2280 * This semantics was introduced after 0.99.9 release.
2282 if (work_queue_empty(zrouter
.ribq
))
2283 work_queue_add(zrouter
.ribq
, zrouter
.mq
);
2285 return mq_add_func(zrouter
.mq
, data
);
2288 /* Add route_node to work queue and schedule processing */
2289 int rib_queue_add(struct route_node
*rn
)
2293 /* Pointless to queue a route_node with no RIB entries to add or remove
2295 if (!rnode_to_ribs(rn
)) {
2296 zlog_debug("%s: called for route_node (%p, %d) with no ribs",
2297 __func__
, (void *)rn
, rn
->lock
);
2298 zlog_backtrace(LOG_DEBUG
);
2302 return mq_add_handler(rn
, &rib_meta_queue_add
);
2305 int rib_queue_nhg_add(struct nhg_ctx
*ctx
)
2309 return mq_add_handler(ctx
, &rib_meta_queue_nhg_add
);
2312 /* Create new meta queue.
2313 A destructor function doesn't seem to be necessary here.
2315 static struct meta_queue
*meta_queue_new(void)
2317 struct meta_queue
*new;
2320 new = XCALLOC(MTYPE_WORK_QUEUE
, sizeof(struct meta_queue
));
2322 for (i
= 0; i
< MQ_SIZE
; i
++) {
2323 new->subq
[i
] = list_new();
2324 assert(new->subq
[i
]);
2330 void meta_queue_free(struct meta_queue
*mq
)
2334 for (i
= 0; i
< MQ_SIZE
; i
++)
2335 list_delete(&mq
->subq
[i
]);
2337 XFREE(MTYPE_WORK_QUEUE
, mq
);
2340 /* initialise zebra rib work queue */
2341 static void rib_queue_init(void)
2343 if (!(zrouter
.ribq
= work_queue_new(zrouter
.master
,
2344 "route_node processing"))) {
2345 flog_err(EC_ZEBRA_WQ_NONEXISTENT
,
2346 "%s: could not initialise work queue!", __func__
);
2350 /* fill in the work queue spec */
2351 zrouter
.ribq
->spec
.workfunc
= &meta_queue_process
;
2352 zrouter
.ribq
->spec
.errorfunc
= NULL
;
2353 zrouter
.ribq
->spec
.completion_func
= NULL
;
2354 /* XXX: TODO: These should be runtime configurable via vty */
2355 zrouter
.ribq
->spec
.max_retries
= 3;
2356 zrouter
.ribq
->spec
.hold
= ZEBRA_RIB_PROCESS_HOLD_TIME
;
2357 zrouter
.ribq
->spec
.retry
= ZEBRA_RIB_PROCESS_RETRY_TIME
;
2359 if (!(zrouter
.mq
= meta_queue_new())) {
2360 flog_err(EC_ZEBRA_WQ_NONEXISTENT
,
2361 "%s: could not initialise meta queue!", __func__
);
2367 rib_dest_t
*zebra_rib_create_dest(struct route_node
*rn
)
2371 dest
= XCALLOC(MTYPE_RIB_DEST
, sizeof(rib_dest_t
));
2372 rnh_list_init(&dest
->nht
);
2373 route_lock_node(rn
); /* rn route table reference */
2380 /* RIB updates are processed via a queue of pointers to route_nodes.
2382 * The queue length is bounded by the maximal size of the routing table,
2383 * as a route_node will not be requeued, if already queued.
2385 * REs are submitted via rib_addnode or rib_delnode which set minimal
2386 * state, or static_install_route (when an existing RE is updated)
2387 * and then submit route_node to queue for best-path selection later.
2388 * Order of add/delete state changes are preserved for any given RE.
2390 * Deleted REs are reaped during best-path selection.
2393 * |-> rib_link or unset ROUTE_ENTRY_REMOVE |->Update kernel with
2394 * |-------->| | best RE, if required
2396 * static_install->|->rib_addqueue...... -> rib_process
2398 * |-------->| |-> rib_unlink
2399 * |-> set ROUTE_ENTRY_REMOVE |
2400 * rib_delnode (RE freed)
2402 * The 'info' pointer of a route_node points to a rib_dest_t
2403 * ('dest'). Queueing state for a route_node is kept on the dest. The
2404 * dest is created on-demand by rib_link() and is kept around at least
2405 * as long as there are ribs hanging off it (@see rib_gc_dest()).
2407 * Refcounting (aka "locking" throughout the GNU Zebra and Quagga code):
2409 * - route_nodes: refcounted by:
2410 * - dest attached to route_node:
2411 * - managed by: rib_link/rib_gc_dest
2412 * - route_node processing queue
2413 * - managed by: rib_addqueue, rib_process.
2417 /* Add RE to head of the route node. */
2418 static void rib_link(struct route_node
*rn
, struct route_entry
*re
, int process
)
2422 const char *rmap_name
;
2426 dest
= rib_dest_from_rnode(rn
);
2428 if (IS_ZEBRA_DEBUG_RIB_DETAILED
)
2429 rnode_debug(rn
, re
->vrf_id
, "rn %p adding dest", rn
);
2431 dest
= zebra_rib_create_dest(rn
);
2434 re_list_add_head(&dest
->routes
, re
);
2436 afi
= (rn
->p
.family
== AF_INET
)
2438 : (rn
->p
.family
== AF_INET6
) ? AFI_IP6
: AFI_MAX
;
2439 if (is_zebra_import_table_enabled(afi
, re
->vrf_id
, re
->table
)) {
2440 struct zebra_vrf
*zvrf
= zebra_vrf_lookup_by_id(re
->vrf_id
);
2442 rmap_name
= zebra_get_import_table_route_map(afi
, re
->table
);
2443 zebra_add_import_table_entry(zvrf
, rn
, re
, rmap_name
);
2448 static void rib_addnode(struct route_node
*rn
,
2449 struct route_entry
*re
, int process
)
2451 /* RE node has been un-removed before route-node is processed.
2452 * route_node must hence already be on the queue for processing..
2454 if (CHECK_FLAG(re
->status
, ROUTE_ENTRY_REMOVED
)) {
2455 if (IS_ZEBRA_DEBUG_RIB
)
2456 rnode_debug(rn
, re
->vrf_id
, "rn %p, un-removed re %p",
2457 (void *)rn
, (void *)re
);
2459 UNSET_FLAG(re
->status
, ROUTE_ENTRY_REMOVED
);
2462 rib_link(rn
, re
, process
);
2468 * Detach a rib structure from a route_node.
2470 * Note that a call to rib_unlink() should be followed by a call to
2471 * rib_gc_dest() at some point. This allows a rib_dest_t that is no
2472 * longer required to be deleted.
2474 void rib_unlink(struct route_node
*rn
, struct route_entry
*re
)
2477 struct nhg_hash_entry
*nhe
= NULL
;
2481 if (IS_ZEBRA_DEBUG_RIB
)
2482 rnode_debug(rn
, re
->vrf_id
, "rn %p, re %p", (void *)rn
,
2485 dest
= rib_dest_from_rnode(rn
);
2487 re_list_del(&dest
->routes
, re
);
2489 if (dest
->selected_fib
== re
)
2490 dest
->selected_fib
= NULL
;
2493 nhe
= zebra_nhg_lookup_id(re
->nhe_id
);
2495 zebra_nhg_decrement_ref(nhe
);
2497 nexthop_group_free_delete(&re
->ng
);
2499 nexthops_free(re
->fib_ng
.nexthop
);
2501 XFREE(MTYPE_RE
, re
);
2504 void rib_delnode(struct route_node
*rn
, struct route_entry
*re
)
2508 if (IS_ZEBRA_DEBUG_RIB
)
2509 rnode_debug(rn
, re
->vrf_id
, "rn %p, re %p, removing",
2510 (void *)rn
, (void *)re
);
2511 SET_FLAG(re
->status
, ROUTE_ENTRY_REMOVED
);
2513 afi
= (rn
->p
.family
== AF_INET
)
2515 : (rn
->p
.family
== AF_INET6
) ? AFI_IP6
: AFI_MAX
;
2516 if (is_zebra_import_table_enabled(afi
, re
->vrf_id
, re
->table
)) {
2517 struct zebra_vrf
*zvrf
= zebra_vrf_lookup_by_id(re
->vrf_id
);
2519 zebra_del_import_table_entry(zvrf
, rn
, re
);
2520 /* Just clean up if non main table */
2521 if (IS_ZEBRA_DEBUG_RIB
) {
2522 char buf
[SRCDEST2STR_BUFFER
];
2523 srcdest_rnode2str(rn
, buf
, sizeof(buf
));
2524 zlog_debug("%u:%s: Freeing route rn %p, re %p (%s)",
2525 re
->vrf_id
, buf
, rn
, re
,
2526 zebra_route_string(re
->type
));
2535 /* This function dumps the contents of a given RE entry into
2536 * standard debug log. Calling function name and IP prefix in
2537 * question are passed as 1st and 2nd arguments.
2540 void _route_entry_dump(const char *func
, union prefixconstptr pp
,
2541 union prefixconstptr src_pp
,
2542 const struct route_entry
*re
)
2544 const struct prefix
*src_p
= src_pp
.p
;
2545 bool is_srcdst
= src_p
&& src_p
->prefixlen
;
2546 char straddr
[PREFIX_STRLEN
];
2547 char srcaddr
[PREFIX_STRLEN
];
2548 char nhname
[PREFIX_STRLEN
];
2549 struct nexthop
*nexthop
;
2551 zlog_debug("%s: dumping RE entry %p for %s%s%s vrf %u", func
,
2552 (const void *)re
, prefix2str(pp
, straddr
, sizeof(straddr
)),
2553 is_srcdst
? " from " : "",
2554 is_srcdst
? prefix2str(src_pp
, srcaddr
, sizeof(srcaddr
))
2557 zlog_debug("%s: uptime == %lu, type == %u, instance == %d, table == %d",
2558 straddr
, (unsigned long)re
->uptime
, re
->type
, re
->instance
,
2561 "%s: metric == %u, mtu == %u, distance == %u, flags == %u, status == %u",
2562 straddr
, re
->metric
, re
->mtu
, re
->distance
, re
->flags
, re
->status
);
2563 zlog_debug("%s: nexthop_num == %u, nexthop_active_num == %u", straddr
,
2564 nexthop_group_nexthop_num(re
->ng
),
2565 nexthop_group_active_nexthop_num(re
->ng
));
2567 for (ALL_NEXTHOPS_PTR(re
->ng
, nexthop
)) {
2568 struct interface
*ifp
;
2569 struct vrf
*vrf
= vrf_lookup_by_id(nexthop
->vrf_id
);
2571 switch (nexthop
->type
) {
2572 case NEXTHOP_TYPE_BLACKHOLE
:
2573 sprintf(nhname
, "Blackhole");
2575 case NEXTHOP_TYPE_IFINDEX
:
2576 ifp
= if_lookup_by_index(nexthop
->ifindex
,
2578 sprintf(nhname
, "%s", ifp
? ifp
->name
: "Unknown");
2580 case NEXTHOP_TYPE_IPV4
:
2582 case NEXTHOP_TYPE_IPV4_IFINDEX
:
2583 inet_ntop(AF_INET
, &nexthop
->gate
, nhname
,
2586 case NEXTHOP_TYPE_IPV6
:
2587 case NEXTHOP_TYPE_IPV6_IFINDEX
:
2588 inet_ntop(AF_INET6
, &nexthop
->gate
, nhname
,
2592 zlog_debug("%s: %s %s[%u] vrf %s(%u) with flags %s%s%s%s%s%s",
2593 straddr
, (nexthop
->rparent
? " NH" : "NH"), nhname
,
2594 nexthop
->ifindex
, vrf
? vrf
->name
: "Unknown",
2596 (CHECK_FLAG(nexthop
->flags
, NEXTHOP_FLAG_ACTIVE
)
2599 (CHECK_FLAG(re
->status
, ROUTE_ENTRY_INSTALLED
)
2602 (CHECK_FLAG(nexthop
->flags
, NEXTHOP_FLAG_RECURSIVE
)
2605 (CHECK_FLAG(nexthop
->flags
, NEXTHOP_FLAG_ONLINK
)
2608 (CHECK_FLAG(nexthop
->flags
, NEXTHOP_FLAG_MATCHED
)
2611 (CHECK_FLAG(nexthop
->flags
, NEXTHOP_FLAG_DUPLICATE
)
2615 zlog_debug("%s: dump complete", straddr
);
2618 /* This is an exported helper to rtm_read() to dump the strange
2619 * RE entry found by rib_lookup_ipv4_route()
2622 void rib_lookup_and_dump(struct prefix_ipv4
*p
, vrf_id_t vrf_id
)
2624 struct route_table
*table
;
2625 struct route_node
*rn
;
2626 struct route_entry
*re
;
2627 char prefix_buf
[INET_ADDRSTRLEN
];
2630 table
= zebra_vrf_table(AFI_IP
, SAFI_UNICAST
, vrf_id
);
2632 flog_err(EC_ZEBRA_TABLE_LOOKUP_FAILED
,
2633 "%s:%u zebra_vrf_table() returned NULL", __func__
,
2638 /* Scan the RIB table for exactly matching RE entry. */
2639 rn
= route_node_lookup(table
, (struct prefix
*)p
);
2641 /* No route for this prefix. */
2643 zlog_debug("%s:%u lookup failed for %s", __func__
, vrf_id
,
2644 prefix2str((struct prefix
*)p
, prefix_buf
,
2645 sizeof(prefix_buf
)));
2650 route_unlock_node(rn
);
2653 RNODE_FOREACH_RE (rn
, re
) {
2654 zlog_debug("%s:%u rn %p, re %p: %s, %s",
2656 (void *)rn
, (void *)re
,
2657 (CHECK_FLAG(re
->status
, ROUTE_ENTRY_REMOVED
)
2660 (CHECK_FLAG(re
->flags
, ZEBRA_FLAG_SELECTED
)
2663 route_entry_dump(p
, NULL
, re
);
2667 /* Check if requested address assignment will fail due to another
2668 * route being installed by zebra in FIB already. Take necessary
2669 * actions, if needed: remove such a route from FIB and deSELECT
2670 * corresponding RE entry. Then put affected RN into RIBQ head.
2672 void rib_lookup_and_pushup(struct prefix_ipv4
*p
, vrf_id_t vrf_id
)
2674 struct route_table
*table
;
2675 struct route_node
*rn
;
2678 if (NULL
== (table
= zebra_vrf_table(AFI_IP
, SAFI_UNICAST
, vrf_id
))) {
2679 flog_err(EC_ZEBRA_TABLE_LOOKUP_FAILED
,
2680 "%s:%u zebra_vrf_table() returned NULL", __func__
,
2685 /* No matches would be the simplest case. */
2686 if (NULL
== (rn
= route_node_lookup(table
, (struct prefix
*)p
)))
2690 route_unlock_node(rn
);
2692 dest
= rib_dest_from_rnode(rn
);
2693 /* Check all RE entries. In case any changes have to be done, requeue
2694 * the RN into RIBQ head. If the routing message about the new connected
2695 * route (generated by the IP address we are going to assign very soon)
2696 * comes before the RIBQ is processed, the new RE entry will join
2697 * RIBQ record already on head. This is necessary for proper
2699 * of the rest of the RE.
2701 if (dest
->selected_fib
) {
2702 if (IS_ZEBRA_DEBUG_RIB
) {
2703 char buf
[PREFIX_STRLEN
];
2705 zlog_debug("%u:%s: freeing way for connected prefix",
2706 dest
->selected_fib
->vrf_id
,
2707 prefix2str(&rn
->p
, buf
, sizeof(buf
)));
2708 route_entry_dump(&rn
->p
, NULL
, dest
->selected_fib
);
2710 rib_uninstall(rn
, dest
->selected_fib
);
2715 int rib_add_multipath(afi_t afi
, safi_t safi
, struct prefix
*p
,
2716 struct prefix_ipv6
*src_p
, struct route_entry
*re
)
2718 struct nhg_hash_entry
*nhe
= NULL
;
2719 struct route_table
*table
;
2720 struct route_node
*rn
;
2721 struct route_entry
*same
= NULL
;
2727 assert(!src_p
|| !src_p
->prefixlen
|| afi
== AFI_IP6
);
2730 table
= zebra_vrf_table_with_table_id(afi
, safi
, re
->vrf_id
, re
->table
);
2733 nexthop_group_free_delete(&re
->ng
);
2734 XFREE(MTYPE_RE
, re
);
2739 nhe
= zebra_nhg_lookup_id(re
->nhe_id
);
2743 EC_ZEBRA_TABLE_LOOKUP_FAILED
,
2744 "Zebra failed to find the nexthop hash entry for id=%u in a route entry",
2746 XFREE(MTYPE_RE
, re
);
2750 nhe
= zebra_nhg_rib_find(0, re
->ng
, afi
);
2753 * The nexthops got copied over into an nhe,
2756 nexthop_group_free_delete(&re
->ng
);
2759 char buf
[PREFIX_STRLEN
] = "";
2760 char buf2
[PREFIX_STRLEN
] = "";
2763 EC_ZEBRA_TABLE_LOOKUP_FAILED
,
2764 "Zebra failed to find or create a nexthop hash entry for %s%s%s",
2765 prefix2str(p
, buf
, sizeof(buf
)),
2766 src_p
? " from " : "",
2767 src_p
? prefix2str(src_p
, buf2
, sizeof(buf2
))
2770 XFREE(MTYPE_RE
, re
);
2775 /* Attach the re to the nhe's nexthop group */
2776 zebra_nhg_re_update_ref(re
, nhe
);
2778 /* Make it sure prefixlen is applied to the prefix. */
2781 apply_mask_ipv6(src_p
);
2783 /* Set default distance by route type. */
2784 if (re
->distance
== 0)
2785 re
->distance
= route_distance(re
->type
);
2787 /* Lookup route node.*/
2788 rn
= srcdest_rnode_get(table
, p
, src_p
);
2791 * If same type of route are installed, treat it as a implicit
2793 * If the user has specified the No route replace semantics
2794 * for the install don't do a route replace.
2796 RNODE_FOREACH_RE (rn
, same
) {
2797 if (CHECK_FLAG(same
->status
, ROUTE_ENTRY_REMOVED
))
2800 if (same
->type
!= re
->type
)
2802 if (same
->instance
!= re
->instance
)
2804 if (same
->type
== ZEBRA_ROUTE_KERNEL
2805 && same
->metric
!= re
->metric
)
2808 if (CHECK_FLAG(re
->flags
, ZEBRA_FLAG_RR_USE_DISTANCE
) &&
2809 same
->distance
!= re
->distance
)
2813 * We should allow duplicate connected routes
2814 * because of IPv6 link-local routes and unnumbered
2815 * interfaces on Linux.
2817 if (same
->type
!= ZEBRA_ROUTE_CONNECT
)
2821 /* If this route is kernel/connected route, notify the dataplane. */
2822 if (RIB_SYSTEM_ROUTE(re
)) {
2823 /* Notify dataplane */
2824 dplane_sys_route_add(rn
, re
);
2827 /* Link new re to node.*/
2828 if (IS_ZEBRA_DEBUG_RIB
) {
2829 rnode_debug(rn
, re
->vrf_id
,
2830 "Inserting route rn %p, re %p (%s) existing %p",
2831 rn
, re
, zebra_route_string(re
->type
), same
);
2833 if (IS_ZEBRA_DEBUG_RIB_DETAILED
)
2834 route_entry_dump(p
, src_p
, re
);
2837 SET_FLAG(re
->status
, ROUTE_ENTRY_CHANGED
);
2838 rib_addnode(rn
, re
, 1);
2841 /* Free implicit route.*/
2843 rib_delnode(rn
, same
);
2847 route_unlock_node(rn
);
2851 void rib_delete(afi_t afi
, safi_t safi
, vrf_id_t vrf_id
, int type
,
2852 unsigned short instance
, int flags
, struct prefix
*p
,
2853 struct prefix_ipv6
*src_p
, const struct nexthop
*nh
,
2854 uint32_t table_id
, uint32_t metric
, uint8_t distance
,
2857 struct route_table
*table
;
2858 struct route_node
*rn
;
2859 struct route_entry
*re
;
2860 struct route_entry
*fib
= NULL
;
2861 struct route_entry
*same
= NULL
;
2862 struct nexthop
*rtnh
;
2863 char buf2
[INET6_ADDRSTRLEN
];
2866 assert(!src_p
|| !src_p
->prefixlen
|| afi
== AFI_IP6
);
2869 table
= zebra_vrf_table_with_table_id(afi
, safi
, vrf_id
, table_id
);
2876 apply_mask_ipv6(src_p
);
2878 /* Lookup route node. */
2879 rn
= srcdest_rnode_lookup(table
, p
, src_p
);
2881 char dst_buf
[PREFIX_STRLEN
], src_buf
[PREFIX_STRLEN
];
2883 prefix2str(p
, dst_buf
, sizeof(dst_buf
));
2884 if (src_p
&& src_p
->prefixlen
)
2885 prefix2str(src_p
, src_buf
, sizeof(src_buf
));
2889 if (IS_ZEBRA_DEBUG_RIB
) {
2890 struct vrf
*vrf
= vrf_lookup_by_id(vrf_id
);
2892 zlog_debug("%s[%d]:%s%s%s doesn't exist in rib",
2893 vrf
->name
, table_id
, dst_buf
,
2894 (src_buf
[0] != '\0') ? " from " : "",
2900 dest
= rib_dest_from_rnode(rn
);
2901 fib
= dest
->selected_fib
;
2903 /* Lookup same type route. */
2904 RNODE_FOREACH_RE (rn
, re
) {
2905 if (CHECK_FLAG(re
->status
, ROUTE_ENTRY_REMOVED
))
2908 if (re
->type
!= type
)
2910 if (re
->instance
!= instance
)
2912 if (CHECK_FLAG(re
->flags
, ZEBRA_FLAG_RR_USE_DISTANCE
) &&
2913 distance
!= re
->distance
)
2916 if (re
->type
== ZEBRA_ROUTE_KERNEL
&& re
->metric
!= metric
)
2918 if (re
->type
== ZEBRA_ROUTE_CONNECT
&& (rtnh
= re
->ng
->nexthop
)
2919 && rtnh
->type
== NEXTHOP_TYPE_IFINDEX
&& nh
) {
2920 if (rtnh
->ifindex
!= nh
->ifindex
)
2925 /* Make sure that the route found has the same gateway. */
2931 for (ALL_NEXTHOPS_PTR(re
->ng
, rtnh
))
2933 * No guarantee all kernel send nh with labels
2936 if (nexthop_same_no_labels(rtnh
, nh
)) {
2944 /* If same type of route can't be found and this message is from
2948 * In the past(HA!) we could get here because
2949 * we were receiving a route delete from the
2950 * kernel and we're not marking the proto
2951 * as coming from it's appropriate originator.
2952 * Now that we are properly noticing the fact
2953 * that the kernel has deleted our route we
2954 * are not going to get called in this path
2955 * I am going to leave this here because
2956 * this might still work this way on non-linux
2957 * platforms as well as some weird state I have
2958 * not properly thought of yet.
2959 * If we can show that this code path is
2960 * dead then we can remove it.
2962 if (fib
&& CHECK_FLAG(flags
, ZEBRA_FLAG_SELFROUTE
)) {
2963 if (IS_ZEBRA_DEBUG_RIB
) {
2964 rnode_debug(rn
, vrf_id
,
2965 "rn %p, re %p (%s) was deleted from kernel, adding",
2967 zebra_route_string(fib
->type
));
2970 UNSET_FLAG(fib
->status
, ROUTE_ENTRY_INSTALLED
);
2972 for (rtnh
= fib
->ng
->nexthop
; rtnh
;
2974 UNSET_FLAG(rtnh
->flags
,
2978 * This is a non FRR route
2979 * as such we should mark
2982 dest
->selected_fib
= NULL
;
2984 /* This means someone else, other than Zebra,
2986 * a Zebra router from the kernel. We will add
2988 rib_install_kernel(rn
, fib
, NULL
);
2991 if (IS_ZEBRA_DEBUG_RIB
) {
2995 "via %s ifindex %d type %d "
2996 "doesn't exist in rib",
2997 inet_ntop(afi2family(afi
),
3004 "type %d doesn't exist in rib",
3007 route_unlock_node(rn
);
3013 if (fromkernel
&& CHECK_FLAG(flags
, ZEBRA_FLAG_SELFROUTE
)
3015 rib_install_kernel(rn
, same
, NULL
);
3016 route_unlock_node(rn
);
3021 /* Special handling for IPv4 or IPv6 routes sourced from
3022 * EVPN - the nexthop (and associated MAC) need to be
3023 * uninstalled if no more refs.
3025 if (CHECK_FLAG(flags
, ZEBRA_FLAG_EVPN_ROUTE
)) {
3026 struct nexthop
*tmp_nh
;
3028 for (ALL_NEXTHOPS_PTR(re
->ng
, tmp_nh
)) {
3029 struct ipaddr vtep_ip
;
3031 memset(&vtep_ip
, 0, sizeof(struct ipaddr
));
3032 if (afi
== AFI_IP
) {
3033 vtep_ip
.ipa_type
= IPADDR_V4
;
3034 memcpy(&(vtep_ip
.ipaddr_v4
),
3035 &(tmp_nh
->gate
.ipv4
),
3036 sizeof(struct in_addr
));
3038 vtep_ip
.ipa_type
= IPADDR_V6
;
3039 memcpy(&(vtep_ip
.ipaddr_v6
),
3040 &(tmp_nh
->gate
.ipv6
),
3041 sizeof(struct in6_addr
));
3043 zebra_vxlan_evpn_vrf_route_del(re
->vrf_id
,
3048 /* Notify dplane if system route changes */
3049 if (RIB_SYSTEM_ROUTE(re
))
3050 dplane_sys_route_del(rn
, same
);
3052 rib_delnode(rn
, same
);
3055 route_unlock_node(rn
);
3060 int rib_add(afi_t afi
, safi_t safi
, vrf_id_t vrf_id
, int type
,
3061 unsigned short instance
, int flags
, struct prefix
*p
,
3062 struct prefix_ipv6
*src_p
, const struct nexthop
*nh
,
3063 uint32_t nhe_id
, uint32_t table_id
, uint32_t metric
, uint32_t mtu
,
3064 uint8_t distance
, route_tag_t tag
)
3066 struct route_entry
*re
= NULL
;
3067 struct nexthop
*nexthop
= NULL
;
3069 /* Allocate new route_entry structure. */
3070 re
= XCALLOC(MTYPE_RE
, sizeof(struct route_entry
));
3072 re
->instance
= instance
;
3073 re
->distance
= distance
;
3075 re
->metric
= metric
;
3077 re
->table
= table_id
;
3078 re
->vrf_id
= vrf_id
;
3079 re
->uptime
= monotime(NULL
);
3081 re
->nhe_id
= nhe_id
;
3084 re
->ng
= nexthop_group_new();
3087 nexthop
= nexthop_new();
3089 route_entry_nexthop_add(re
, nexthop
);
3092 return rib_add_multipath(afi
, safi
, p
, src_p
, re
);
3095 static const char *rib_update_event2str(rib_update_event_t event
)
3097 const char *ret
= "UNKNOWN";
3100 case RIB_UPDATE_KERNEL
:
3101 ret
= "RIB_UPDATE_KERNEL";
3103 case RIB_UPDATE_RMAP_CHANGE
:
3104 ret
= "RIB_UPDATE_RMAP_CHANGE";
3106 case RIB_UPDATE_OTHER
:
3107 ret
= "RIB_UPDATE_OTHER";
3109 case RIB_UPDATE_MAX
:
3117 /* Schedule route nodes to be processed if they match the type */
3118 static void rib_update_route_node(struct route_node
*rn
, int type
)
3120 struct route_entry
*re
, *next
;
3121 bool re_changed
= false;
3123 RNODE_FOREACH_RE_SAFE (rn
, re
, next
) {
3124 if (type
== ZEBRA_ROUTE_ALL
|| type
== re
->type
) {
3125 SET_FLAG(re
->status
, ROUTE_ENTRY_CHANGED
);
3134 /* Schedule routes of a particular table (address-family) based on event. */
3135 void rib_update_table(struct route_table
*table
, rib_update_event_t event
)
3137 struct route_node
*rn
;
3139 if (IS_ZEBRA_DEBUG_EVENT
) {
3140 struct zebra_vrf
*zvrf
;
3143 zvrf
= table
->info
? ((rib_table_info_t
*)table
->info
)->zvrf
3145 vrf
= zvrf
? zvrf
->vrf
: NULL
;
3147 zlog_debug("%s: %s VRF %s Table %u event %s", __func__
,
3148 table
->info
? afi2str(
3149 ((rib_table_info_t
*)table
->info
)->afi
)
3151 vrf
? vrf
->name
: "Unknown",
3152 zvrf
? zvrf
->table_id
: 0,
3153 rib_update_event2str(event
));
3156 /* Walk all routes and queue for processing, if appropriate for
3157 * the trigger event.
3159 for (rn
= route_top(table
); rn
; rn
= srcdest_route_next(rn
)) {
3161 * If we are looking at a route node and the node
3162 * has already been queued we don't
3163 * need to queue it up again
3166 && CHECK_FLAG(rib_dest_from_rnode(rn
)->flags
,
3167 RIB_ROUTE_ANY_QUEUED
))
3171 case RIB_UPDATE_KERNEL
:
3172 rib_update_route_node(rn
, ZEBRA_ROUTE_KERNEL
);
3174 case RIB_UPDATE_RMAP_CHANGE
:
3175 case RIB_UPDATE_OTHER
:
3176 rib_update_route_node(rn
, ZEBRA_ROUTE_ALL
);
3184 static void rib_update_handle_vrf(vrf_id_t vrf_id
, rib_update_event_t event
)
3186 struct route_table
*table
;
3188 if (IS_ZEBRA_DEBUG_EVENT
)
3189 zlog_debug("%s: Handling VRF %s event %s", __func__
,
3190 vrf_id_to_name(vrf_id
), rib_update_event2str(event
));
3192 /* Process routes of interested address-families. */
3193 table
= zebra_vrf_table(AFI_IP
, SAFI_UNICAST
, vrf_id
);
3195 rib_update_table(table
, event
);
3197 table
= zebra_vrf_table(AFI_IP6
, SAFI_UNICAST
, vrf_id
);
3199 rib_update_table(table
, event
);
3202 static void rib_update_handle_vrf_all(rib_update_event_t event
)
3204 struct zebra_router_table
*zrt
;
3206 if (IS_ZEBRA_DEBUG_EVENT
)
3207 zlog_debug("%s: Handling VRF (ALL) event %s", __func__
,
3208 rib_update_event2str(event
));
3210 /* Just iterate over all the route tables, rather than vrf lookups */
3211 RB_FOREACH (zrt
, zebra_router_table_head
, &zrouter
.tables
)
3212 rib_update_table(zrt
->table
, event
);
3215 struct rib_update_ctx
{
3216 rib_update_event_t event
;
3221 static struct rib_update_ctx
*rib_update_ctx_init(vrf_id_t vrf_id
,
3222 rib_update_event_t event
)
3224 struct rib_update_ctx
*ctx
;
3226 ctx
= XCALLOC(MTYPE_RIB_UPDATE_CTX
, sizeof(struct rib_update_ctx
));
3229 ctx
->vrf_id
= vrf_id
;
3234 static void rib_update_ctx_fini(struct rib_update_ctx
**ctx
)
3236 XFREE(MTYPE_RIB_UPDATE_CTX
, *ctx
);
3241 static int rib_update_handler(struct thread
*thread
)
3243 struct rib_update_ctx
*ctx
;
3245 ctx
= THREAD_ARG(thread
);
3248 rib_update_handle_vrf_all(ctx
->event
);
3250 rib_update_handle_vrf(ctx
->vrf_id
, ctx
->event
);
3252 rib_update_ctx_fini(&ctx
);
3258 * Thread list to ensure we don't schedule a ton of events
3259 * if interfaces are flapping for instance.
3261 static struct thread
*t_rib_update_threads
[RIB_UPDATE_MAX
];
3263 /* Schedule a RIB update event for specific vrf */
3264 void rib_update_vrf(vrf_id_t vrf_id
, rib_update_event_t event
)
3266 struct rib_update_ctx
*ctx
;
3268 ctx
= rib_update_ctx_init(vrf_id
, event
);
3270 /* Don't worry about making sure multiple rib updates for specific vrf
3271 * are scheduled at once for now. If it becomes a problem, we can use a
3272 * lookup of some sort to keep track of running threads via t_vrf_id
3273 * like how we are doing it in t_rib_update_threads[].
3275 thread_add_event(zrouter
.master
, rib_update_handler
, ctx
, 0, NULL
);
3277 if (IS_ZEBRA_DEBUG_EVENT
)
3278 zlog_debug("%s: Scheduled VRF %s, event %s", __func__
,
3279 vrf_id_to_name(ctx
->vrf_id
),
3280 rib_update_event2str(event
));
3283 /* Schedule a RIB update event for all vrfs */
3284 void rib_update(rib_update_event_t event
)
3286 struct rib_update_ctx
*ctx
;
3288 ctx
= rib_update_ctx_init(0, event
);
3290 ctx
->vrf_all
= true;
3292 if (!thread_add_event(zrouter
.master
, rib_update_handler
, ctx
, 0,
3293 &t_rib_update_threads
[event
]))
3294 rib_update_ctx_fini(&ctx
); /* Already scheduled */
3295 else if (IS_ZEBRA_DEBUG_EVENT
)
3296 zlog_debug("%s: Schedued VRF (ALL), event %s", __func__
,
3297 rib_update_event2str(event
));
3300 /* Delete self installed routes after zebra is relaunched. */
3301 void rib_sweep_table(struct route_table
*table
)
3303 struct route_node
*rn
;
3304 struct route_entry
*re
;
3305 struct route_entry
*next
;
3306 struct nexthop
*nexthop
;
3311 for (rn
= route_top(table
); rn
; rn
= srcdest_route_next(rn
)) {
3312 RNODE_FOREACH_RE_SAFE (rn
, re
, next
) {
3314 if (IS_ZEBRA_DEBUG_RIB
)
3315 route_entry_dump(&rn
->p
, NULL
, re
);
3317 if (CHECK_FLAG(re
->status
, ROUTE_ENTRY_REMOVED
))
3320 if (!CHECK_FLAG(re
->flags
, ZEBRA_FLAG_SELFROUTE
))
3324 * If routes are older than startup_time then
3325 * we know we read them in from the kernel.
3326 * As such we can safely remove them.
3328 if (zrouter
.startup_time
< re
->uptime
)
3332 * So we are starting up and have received
3333 * routes from the kernel that we have installed
3334 * from a previous run of zebra but not cleaned
3335 * up ( say a kill -9 )
3336 * But since we haven't actually installed
3337 * them yet( we received them from the kernel )
3338 * we don't think they are active.
3339 * So let's pretend they are active to actually
3341 * In all honesty I'm not sure if we should
3342 * mark them as active when we receive them
3343 * This is startup only so probably ok.
3345 * If we ever decide to move rib_sweep_table
3346 * to a different spot (ie startup )
3347 * this decision needs to be revisited
3349 SET_FLAG(re
->status
, ROUTE_ENTRY_INSTALLED
);
3350 for (ALL_NEXTHOPS_PTR(re
->ng
, nexthop
))
3351 SET_FLAG(nexthop
->flags
, NEXTHOP_FLAG_FIB
);
3353 rib_uninstall_kernel(rn
, re
);
3354 rib_delnode(rn
, re
);
3359 /* Sweep all RIB tables. */
3360 int rib_sweep_route(struct thread
*t
)
3363 struct zebra_vrf
*zvrf
;
3365 RB_FOREACH (vrf
, vrf_id_head
, &vrfs_by_id
) {
3366 if ((zvrf
= vrf
->info
) == NULL
)
3369 rib_sweep_table(zvrf
->table
[AFI_IP
][SAFI_UNICAST
]);
3370 rib_sweep_table(zvrf
->table
[AFI_IP6
][SAFI_UNICAST
]);
3373 zebra_router_sweep_route();
3378 /* Remove specific by protocol routes from 'table'. */
3379 unsigned long rib_score_proto_table(uint8_t proto
, unsigned short instance
,
3380 struct route_table
*table
)
3382 struct route_node
*rn
;
3383 struct route_entry
*re
;
3384 struct route_entry
*next
;
3385 unsigned long n
= 0;
3388 for (rn
= route_top(table
); rn
; rn
= srcdest_route_next(rn
))
3389 RNODE_FOREACH_RE_SAFE (rn
, re
, next
) {
3390 if (CHECK_FLAG(re
->status
, ROUTE_ENTRY_REMOVED
))
3392 if (re
->type
== proto
3393 && re
->instance
== instance
) {
3394 rib_delnode(rn
, re
);
3401 /* Remove specific by protocol routes. */
3402 unsigned long rib_score_proto(uint8_t proto
, unsigned short instance
)
3405 struct zebra_vrf
*zvrf
;
3406 struct other_route_table
*ort
;
3407 unsigned long cnt
= 0;
3409 RB_FOREACH (vrf
, vrf_id_head
, &vrfs_by_id
) {
3414 cnt
+= rib_score_proto_table(proto
, instance
,
3415 zvrf
->table
[AFI_IP
][SAFI_UNICAST
])
3416 + rib_score_proto_table(
3418 zvrf
->table
[AFI_IP6
][SAFI_UNICAST
]);
3420 frr_each(otable
, &zvrf
->other_tables
, ort
) cnt
+=
3421 rib_score_proto_table(proto
, instance
, ort
->table
);
3427 /* Close RIB and clean up kernel routes. */
3428 void rib_close_table(struct route_table
*table
)
3430 struct route_node
*rn
;
3431 rib_table_info_t
*info
;
3437 info
= route_table_get_info(table
);
3439 for (rn
= route_top(table
); rn
; rn
= srcdest_route_next(rn
)) {
3440 dest
= rib_dest_from_rnode(rn
);
3442 if (dest
&& dest
->selected_fib
) {
3443 if (info
->safi
== SAFI_UNICAST
)
3444 hook_call(rib_update
, rn
, NULL
);
3446 rib_uninstall_kernel(rn
, dest
->selected_fib
);
3447 dest
->selected_fib
= NULL
;
3453 * Handler for async dataplane results after a pseudowire installation
3455 static int handle_pw_result(struct zebra_dplane_ctx
*ctx
)
3457 struct zebra_pw
*pw
;
3458 struct zebra_vrf
*vrf
;
3460 /* The pseudowire code assumes success - we act on an error
3461 * result for installation attempts here.
3463 if (dplane_ctx_get_op(ctx
) != DPLANE_OP_PW_INSTALL
)
3466 if (dplane_ctx_get_status(ctx
) != ZEBRA_DPLANE_REQUEST_SUCCESS
) {
3467 vrf
= zebra_vrf_lookup_by_id(dplane_ctx_get_vrf(ctx
));
3468 pw
= zebra_pw_find(vrf
, dplane_ctx_get_ifname(ctx
));
3470 zebra_pw_install_failure(pw
);
3480 * Handle results from the dataplane system. Dequeue update context
3481 * structs, dispatch to appropriate internal handlers.
3483 static int rib_process_dplane_results(struct thread
*thread
)
3485 struct zebra_dplane_ctx
*ctx
;
3486 struct dplane_ctx_q ctxlist
;
3487 bool shut_p
= false;
3489 /* Dequeue a list of completed updates with one lock/unlock cycle */
3492 TAILQ_INIT(&ctxlist
);
3494 /* Take lock controlling queue of results */
3495 frr_with_mutex(&dplane_mutex
) {
3496 /* Dequeue list of context structs */
3497 dplane_ctx_list_append(&ctxlist
, &rib_dplane_q
);
3500 /* Dequeue context block */
3501 ctx
= dplane_ctx_dequeue(&ctxlist
);
3503 /* If we've emptied the results queue, we're done */
3507 /* If zebra is shutting down, avoid processing results,
3508 * just drain the results queue.
3510 shut_p
= atomic_load_explicit(&zrouter
.in_shutdown
,
3511 memory_order_relaxed
);
3514 dplane_ctx_fini(&ctx
);
3516 ctx
= dplane_ctx_dequeue(&ctxlist
);
3523 switch (dplane_ctx_get_op(ctx
)) {
3524 case DPLANE_OP_ROUTE_INSTALL
:
3525 case DPLANE_OP_ROUTE_UPDATE
:
3526 case DPLANE_OP_ROUTE_DELETE
:
3528 /* Bit of special case for route updates
3529 * that were generated by async notifications:
3530 * we don't want to continue processing these
3533 if (dplane_ctx_get_notif_provider(ctx
) == 0)
3534 rib_process_result(ctx
);
3536 dplane_ctx_fini(&ctx
);
3540 case DPLANE_OP_ROUTE_NOTIFY
:
3541 rib_process_dplane_notify(ctx
);
3544 case DPLANE_OP_NH_INSTALL
:
3545 case DPLANE_OP_NH_UPDATE
:
3546 case DPLANE_OP_NH_DELETE
:
3547 zebra_nhg_dplane_result(ctx
);
3550 case DPLANE_OP_LSP_INSTALL
:
3551 case DPLANE_OP_LSP_UPDATE
:
3552 case DPLANE_OP_LSP_DELETE
:
3554 /* Bit of special case for LSP updates
3555 * that were generated by async notifications:
3556 * we don't want to continue processing these.
3558 if (dplane_ctx_get_notif_provider(ctx
) == 0)
3559 zebra_mpls_lsp_dplane_result(ctx
);
3561 dplane_ctx_fini(&ctx
);
3565 case DPLANE_OP_LSP_NOTIFY
:
3566 zebra_mpls_process_dplane_notify(ctx
);
3569 case DPLANE_OP_PW_INSTALL
:
3570 case DPLANE_OP_PW_UNINSTALL
:
3571 handle_pw_result(ctx
);
3574 case DPLANE_OP_SYS_ROUTE_ADD
:
3575 case DPLANE_OP_SYS_ROUTE_DELETE
:
3576 /* No further processing in zebra for these. */
3577 dplane_ctx_fini(&ctx
);
3580 case DPLANE_OP_MAC_INSTALL
:
3581 case DPLANE_OP_MAC_DELETE
:
3582 zebra_vxlan_handle_result(ctx
);
3585 /* Some op codes not handled here */
3586 case DPLANE_OP_ADDR_INSTALL
:
3587 case DPLANE_OP_ADDR_UNINSTALL
:
3588 case DPLANE_OP_NEIGH_INSTALL
:
3589 case DPLANE_OP_NEIGH_UPDATE
:
3590 case DPLANE_OP_NEIGH_DELETE
:
3591 case DPLANE_OP_VTEP_ADD
:
3592 case DPLANE_OP_VTEP_DELETE
:
3593 case DPLANE_OP_NONE
:
3594 /* Don't expect this: just return the struct? */
3595 dplane_ctx_fini(&ctx
);
3598 } /* Dispatch by op code */
3600 ctx
= dplane_ctx_dequeue(&ctxlist
);
3609 * Results are returned from the dataplane subsystem, in the context of
3610 * the dataplane pthread. We enqueue the results here for processing by
3611 * the main thread later.
3613 static int rib_dplane_results(struct dplane_ctx_q
*ctxlist
)
3615 /* Take lock controlling queue of results */
3616 frr_with_mutex(&dplane_mutex
) {
3617 /* Enqueue context blocks */
3618 dplane_ctx_list_append(&rib_dplane_q
, ctxlist
);
3621 /* Ensure event is signalled to zebra main pthread */
3622 thread_add_event(zrouter
.master
, rib_process_dplane_results
, NULL
, 0,
3629 * Ensure there are no empty slots in the route_info array.
3630 * Every route type in zebra should be present there.
3632 static void check_route_info(void)
3634 int len
= array_size(route_info
);
3637 * ZEBRA_ROUTE_SYSTEM is special cased since
3638 * its key is 0 anyway.
3640 * ZEBRA_ROUTE_ALL is also ignored.
3642 for (int i
= 0; i
< len
; i
++) {
3643 if (i
== ZEBRA_ROUTE_SYSTEM
|| i
== ZEBRA_ROUTE_ALL
)
3645 assert(route_info
[i
].key
);
3646 assert(route_info
[i
].meta_q_map
< MQ_SIZE
);
3650 /* Routing information base initialize. */
3657 /* Init dataplane, and register for results */
3658 pthread_mutex_init(&dplane_mutex
, NULL
);
3659 TAILQ_INIT(&rib_dplane_q
);
3660 zebra_dplane_init(rib_dplane_results
);
3666 * Get the first vrf id that is greater than the given vrf id if any.
3668 * Returns true if a vrf id was found, false otherwise.
3670 static inline int vrf_id_get_next(vrf_id_t vrf_id
, vrf_id_t
*next_id_p
)
3674 vrf
= vrf_lookup_by_id(vrf_id
);
3676 vrf
= RB_NEXT(vrf_id_head
, vrf
);
3678 *next_id_p
= vrf
->vrf_id
;
3687 * rib_tables_iter_next
3689 * Returns the next table in the iteration.
3691 struct route_table
*rib_tables_iter_next(rib_tables_iter_t
*iter
)
3693 struct route_table
*table
;
3696 * Array that helps us go over all AFI/SAFI combinations via one
3703 {AFI_IP
, SAFI_UNICAST
}, {AFI_IP
, SAFI_MULTICAST
},
3704 {AFI_IP
, SAFI_LABELED_UNICAST
}, {AFI_IP6
, SAFI_UNICAST
},
3705 {AFI_IP6
, SAFI_MULTICAST
}, {AFI_IP6
, SAFI_LABELED_UNICAST
},
3710 switch (iter
->state
) {
3712 case RIB_TABLES_ITER_S_INIT
:
3713 iter
->vrf_id
= VRF_DEFAULT
;
3714 iter
->afi_safi_ix
= -1;
3718 case RIB_TABLES_ITER_S_ITERATING
:
3719 iter
->afi_safi_ix
++;
3722 while (iter
->afi_safi_ix
3723 < (int)array_size(afi_safis
)) {
3724 table
= zebra_vrf_table(
3725 afi_safis
[iter
->afi_safi_ix
].afi
,
3726 afi_safis
[iter
->afi_safi_ix
].safi
,
3731 iter
->afi_safi_ix
++;
3735 * Found another table in this vrf.
3741 * Done with all tables in the current vrf, go to the
3745 if (!vrf_id_get_next(iter
->vrf_id
, &iter
->vrf_id
))
3748 iter
->afi_safi_ix
= 0;
3753 case RIB_TABLES_ITER_S_DONE
:
3758 iter
->state
= RIB_TABLES_ITER_S_ITERATING
;
3760 iter
->state
= RIB_TABLES_ITER_S_DONE
;