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"
41 #include "zebra/zebra_router.h"
42 #include "zebra/connected.h"
43 #include "zebra/debug.h"
44 #include "zebra/interface.h"
45 #include "zebra/redistribute.h"
46 #include "zebra/rib.h"
48 #include "zebra/zapi_msg.h"
49 #include "zebra/zebra_errors.h"
50 #include "zebra/zebra_memory.h"
51 #include "zebra/zebra_ns.h"
52 #include "zebra/zebra_rnh.h"
53 #include "zebra/zebra_routemap.h"
54 #include "zebra/zebra_vrf.h"
55 #include "zebra/zebra_vxlan.h"
56 #include "zebra/zapi_msg.h"
57 #include "zebra/zebra_dplane.h"
58 #include "zebra/zebra_nhg.h"
61 * Event, list, and mutex for delivery of dataplane results
63 static pthread_mutex_t dplane_mutex
;
64 static struct thread
*t_dplane
;
65 static struct dplane_ctx_q rib_dplane_q
;
67 DEFINE_HOOK(rib_update
, (struct route_node
* rn
, const char *reason
),
70 /* Should we allow non Quagga processes to delete our routes */
71 extern int allow_delete
;
73 /* Each route type's string and default distance value. */
78 } route_info
[ZEBRA_ROUTE_MAX
] = {
79 [ZEBRA_ROUTE_SYSTEM
] = {ZEBRA_ROUTE_SYSTEM
, 0, 4},
80 [ZEBRA_ROUTE_KERNEL
] = {ZEBRA_ROUTE_KERNEL
, 0, 0},
81 [ZEBRA_ROUTE_CONNECT
] = {ZEBRA_ROUTE_CONNECT
, 0, 0},
82 [ZEBRA_ROUTE_STATIC
] = {ZEBRA_ROUTE_STATIC
, 1, 1},
83 [ZEBRA_ROUTE_RIP
] = {ZEBRA_ROUTE_RIP
, 120, 2},
84 [ZEBRA_ROUTE_RIPNG
] = {ZEBRA_ROUTE_RIPNG
, 120, 2},
85 [ZEBRA_ROUTE_OSPF
] = {ZEBRA_ROUTE_OSPF
, 110, 2},
86 [ZEBRA_ROUTE_OSPF6
] = {ZEBRA_ROUTE_OSPF6
, 110, 2},
87 [ZEBRA_ROUTE_ISIS
] = {ZEBRA_ROUTE_ISIS
, 115, 2},
88 [ZEBRA_ROUTE_BGP
] = {ZEBRA_ROUTE_BGP
, 20 /* IBGP is 200. */, 3},
89 [ZEBRA_ROUTE_PIM
] = {ZEBRA_ROUTE_PIM
, 255, 4},
90 [ZEBRA_ROUTE_EIGRP
] = {ZEBRA_ROUTE_EIGRP
, 90, 2},
91 [ZEBRA_ROUTE_NHRP
] = {ZEBRA_ROUTE_NHRP
, 10, 2},
92 [ZEBRA_ROUTE_HSLS
] = {ZEBRA_ROUTE_HSLS
, 255, 4},
93 [ZEBRA_ROUTE_OLSR
] = {ZEBRA_ROUTE_OLSR
, 255, 4},
94 [ZEBRA_ROUTE_TABLE
] = {ZEBRA_ROUTE_TABLE
, 150, 1},
95 [ZEBRA_ROUTE_LDP
] = {ZEBRA_ROUTE_LDP
, 150, 4},
96 [ZEBRA_ROUTE_VNC
] = {ZEBRA_ROUTE_VNC
, 20, 3},
97 [ZEBRA_ROUTE_VNC_DIRECT
] = {ZEBRA_ROUTE_VNC_DIRECT
, 20, 3},
98 [ZEBRA_ROUTE_VNC_DIRECT_RH
] = {ZEBRA_ROUTE_VNC_DIRECT_RH
, 20, 3},
99 [ZEBRA_ROUTE_BGP_DIRECT
] = {ZEBRA_ROUTE_BGP_DIRECT
, 20, 3},
100 [ZEBRA_ROUTE_BGP_DIRECT_EXT
] = {ZEBRA_ROUTE_BGP_DIRECT_EXT
, 20, 3},
101 [ZEBRA_ROUTE_BABEL
] = {ZEBRA_ROUTE_BABEL
, 100, 2},
102 [ZEBRA_ROUTE_SHARP
] = {ZEBRA_ROUTE_SHARP
, 150, 4},
103 [ZEBRA_ROUTE_PBR
] = {ZEBRA_ROUTE_PBR
, 200, 4},
104 [ZEBRA_ROUTE_BFD
] = {ZEBRA_ROUTE_BFD
, 255, 4},
105 [ZEBRA_ROUTE_OPENFABRIC
] = {ZEBRA_ROUTE_OPENFABRIC
, 115, 2},
106 [ZEBRA_ROUTE_VRRP
] = {ZEBRA_ROUTE_VRRP
, 255, 4}
107 /* Any new route type added to zebra, should be mirrored here */
109 /* no entry/default: 150 */
112 static void __attribute__((format(printf
, 5, 6)))
113 _rnode_zlog(const char *_func
, vrf_id_t vrf_id
, struct route_node
*rn
,
114 int priority
, const char *msgfmt
, ...)
116 char buf
[SRCDEST2STR_BUFFER
+ sizeof(" (MRIB)")];
120 va_start(ap
, msgfmt
);
121 vsnprintf(msgbuf
, sizeof(msgbuf
), msgfmt
, ap
);
125 rib_table_info_t
*info
= srcdest_rnode_table_info(rn
);
126 srcdest_rnode2str(rn
, buf
, sizeof(buf
));
128 if (info
->safi
== SAFI_MULTICAST
)
129 strlcat(buf
, " (MRIB)", sizeof(buf
));
131 snprintf(buf
, sizeof(buf
), "{(route_node *) NULL}");
134 zlog(priority
, "%s: %d:%s: %s", _func
, vrf_id
, buf
, msgbuf
);
137 #define rnode_debug(node, vrf_id, ...) \
138 _rnode_zlog(__func__, vrf_id, node, LOG_DEBUG, __VA_ARGS__)
139 #define rnode_info(node, ...) \
140 _rnode_zlog(__func__, vrf_id, node, LOG_INFO, __VA_ARGS__)
142 uint8_t route_distance(int type
)
146 if ((unsigned)type
>= array_size(route_info
))
149 distance
= route_info
[type
].distance
;
154 int is_zebra_valid_kernel_table(uint32_t table_id
)
157 if ((table_id
== RT_TABLE_UNSPEC
) || (table_id
== RT_TABLE_LOCAL
)
158 || (table_id
== RT_TABLE_COMPAT
))
165 int is_zebra_main_routing_table(uint32_t table_id
)
167 if (table_id
== RT_TABLE_MAIN
)
172 int zebra_check_addr(const struct prefix
*p
)
174 if (p
->family
== AF_INET
) {
177 addr
= p
->u
.prefix4
.s_addr
;
180 if (IPV4_NET127(addr
) || IN_CLASSD(addr
)
181 || IPV4_LINKLOCAL(addr
))
184 if (p
->family
== AF_INET6
) {
185 if (IN6_IS_ADDR_LOOPBACK(&p
->u
.prefix6
))
187 if (IN6_IS_ADDR_LINKLOCAL(&p
->u
.prefix6
))
193 /* Add nexthop to the end of a rib node's nexthop list */
194 void route_entry_nexthop_add(struct route_entry
*re
, struct nexthop
*nexthop
)
196 _nexthop_group_add_sorted(&re
->ng
, nexthop
);
202 * copy_nexthop - copy a nexthop to the rib structure.
204 void route_entry_copy_nexthops(struct route_entry
*re
, struct nexthop
*nh
)
206 assert(!re
->ng
.nexthop
);
207 copy_nexthops(&re
->ng
.nexthop
, nh
, NULL
);
208 for (struct nexthop
*nexthop
= nh
; nexthop
; nexthop
= nexthop
->next
)
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
;
225 struct nexthop
*route_entry_nexthop_ifindex_add(struct route_entry
*re
,
229 struct nexthop
*nexthop
;
231 nexthop
= nexthop_new();
232 nexthop
->type
= NEXTHOP_TYPE_IFINDEX
;
233 nexthop
->ifindex
= ifindex
;
234 nexthop
->vrf_id
= nh_vrf_id
;
236 route_entry_nexthop_add(re
, nexthop
);
241 struct nexthop
*route_entry_nexthop_ipv4_add(struct route_entry
*re
,
242 struct in_addr
*ipv4
,
246 struct nexthop
*nexthop
;
248 nexthop
= nexthop_new();
249 nexthop
->type
= NEXTHOP_TYPE_IPV4
;
250 nexthop
->vrf_id
= nh_vrf_id
;
251 nexthop
->gate
.ipv4
= *ipv4
;
253 nexthop
->src
.ipv4
= *src
;
255 route_entry_nexthop_add(re
, nexthop
);
260 struct nexthop
*route_entry_nexthop_ipv4_ifindex_add(struct route_entry
*re
,
261 struct in_addr
*ipv4
,
266 struct nexthop
*nexthop
;
267 struct interface
*ifp
;
269 nexthop
= nexthop_new();
270 nexthop
->vrf_id
= nh_vrf_id
;
271 nexthop
->type
= NEXTHOP_TYPE_IPV4_IFINDEX
;
272 nexthop
->gate
.ipv4
= *ipv4
;
274 nexthop
->src
.ipv4
= *src
;
275 nexthop
->ifindex
= ifindex
;
276 ifp
= if_lookup_by_index(nexthop
->ifindex
, nh_vrf_id
);
277 /*Pending: need to think if null ifp here is ok during bootup?
278 There was a crash because ifp here was coming to be NULL */
280 if (connected_is_unnumbered(ifp
))
281 SET_FLAG(nexthop
->flags
, NEXTHOP_FLAG_ONLINK
);
283 route_entry_nexthop_add(re
, nexthop
);
288 struct nexthop
*route_entry_nexthop_ipv6_add(struct route_entry
*re
,
289 struct in6_addr
*ipv6
,
292 struct nexthop
*nexthop
;
294 nexthop
= nexthop_new();
295 nexthop
->vrf_id
= nh_vrf_id
;
296 nexthop
->type
= NEXTHOP_TYPE_IPV6
;
297 nexthop
->gate
.ipv6
= *ipv6
;
299 route_entry_nexthop_add(re
, nexthop
);
304 struct nexthop
*route_entry_nexthop_ipv6_ifindex_add(struct route_entry
*re
,
305 struct in6_addr
*ipv6
,
309 struct nexthop
*nexthop
;
311 nexthop
= nexthop_new();
312 nexthop
->vrf_id
= nh_vrf_id
;
313 nexthop
->type
= NEXTHOP_TYPE_IPV6_IFINDEX
;
314 nexthop
->gate
.ipv6
= *ipv6
;
315 nexthop
->ifindex
= ifindex
;
317 route_entry_nexthop_add(re
, nexthop
);
322 struct nexthop
*route_entry_nexthop_blackhole_add(struct route_entry
*re
,
323 enum blackhole_type bh_type
)
325 struct nexthop
*nexthop
;
327 nexthop
= nexthop_new();
328 nexthop
->vrf_id
= VRF_DEFAULT
;
329 nexthop
->type
= NEXTHOP_TYPE_BLACKHOLE
;
330 nexthop
->bh_type
= bh_type
;
332 route_entry_nexthop_add(re
, nexthop
);
337 struct route_entry
*rib_match(afi_t afi
, safi_t safi
, vrf_id_t vrf_id
,
338 union g_addr
*addr
, struct route_node
**rn_out
)
341 struct route_table
*table
;
342 struct route_node
*rn
;
343 struct route_entry
*match
= NULL
;
346 table
= zebra_vrf_table(afi
, safi
, vrf_id
);
350 memset(&p
, 0, sizeof(struct prefix
));
353 p
.u
.prefix4
= addr
->ipv4
;
354 p
.prefixlen
= IPV4_MAX_PREFIXLEN
;
356 p
.u
.prefix6
= addr
->ipv6
;
357 p
.prefixlen
= IPV6_MAX_PREFIXLEN
;
360 rn
= route_node_match(table
, (struct prefix
*)&p
);
365 route_unlock_node(rn
);
367 dest
= rib_dest_from_rnode(rn
);
368 if (dest
&& dest
->selected_fib
369 && !CHECK_FLAG(dest
->selected_fib
->status
,
370 ROUTE_ENTRY_REMOVED
))
371 match
= dest
->selected_fib
;
373 /* If there is no selected route or matched route is EGP, go up
378 } while (rn
&& rn
->info
== NULL
);
382 if (match
->type
!= ZEBRA_ROUTE_CONNECT
) {
383 if (!CHECK_FLAG(match
->status
,
384 ROUTE_ENTRY_INSTALLED
))
396 struct route_entry
*rib_match_ipv4_multicast(vrf_id_t vrf_id
,
398 struct route_node
**rn_out
)
400 struct route_entry
*re
= NULL
, *mre
= NULL
, *ure
= NULL
;
401 struct route_node
*m_rn
= NULL
, *u_rn
= NULL
;
402 union g_addr gaddr
= {.ipv4
= addr
};
404 switch (zrouter
.ipv4_multicast_mode
) {
405 case MCAST_MRIB_ONLY
:
406 return rib_match(AFI_IP
, SAFI_MULTICAST
, vrf_id
, &gaddr
,
408 case MCAST_URIB_ONLY
:
409 return rib_match(AFI_IP
, SAFI_UNICAST
, vrf_id
, &gaddr
, rn_out
);
410 case MCAST_NO_CONFIG
:
411 case MCAST_MIX_MRIB_FIRST
:
412 re
= mre
= rib_match(AFI_IP
, SAFI_MULTICAST
, vrf_id
, &gaddr
,
415 re
= ure
= rib_match(AFI_IP
, SAFI_UNICAST
, vrf_id
,
418 case MCAST_MIX_DISTANCE
:
419 mre
= rib_match(AFI_IP
, SAFI_MULTICAST
, vrf_id
, &gaddr
, &m_rn
);
420 ure
= rib_match(AFI_IP
, SAFI_UNICAST
, vrf_id
, &gaddr
, &u_rn
);
422 re
= ure
->distance
< mre
->distance
? ure
: mre
;
428 case MCAST_MIX_PFXLEN
:
429 mre
= rib_match(AFI_IP
, SAFI_MULTICAST
, vrf_id
, &gaddr
, &m_rn
);
430 ure
= rib_match(AFI_IP
, SAFI_UNICAST
, vrf_id
, &gaddr
, &u_rn
);
432 re
= u_rn
->p
.prefixlen
> m_rn
->p
.prefixlen
? ure
: mre
;
441 *rn_out
= (re
== mre
) ? m_rn
: u_rn
;
443 if (IS_ZEBRA_DEBUG_RIB
) {
445 inet_ntop(AF_INET
, &addr
, buf
, BUFSIZ
);
447 zlog_debug("%s: %s: vrf: %u found %s, using %s",
448 __func__
, buf
, vrf_id
,
449 mre
? (ure
? "MRIB+URIB" : "MRIB")
450 : ure
? "URIB" : "nothing",
451 re
== ure
? "URIB" : re
== mre
? "MRIB" : "none");
456 struct route_entry
*rib_lookup_ipv4(struct prefix_ipv4
*p
, vrf_id_t vrf_id
)
458 struct route_table
*table
;
459 struct route_node
*rn
;
460 struct route_entry
*match
= NULL
;
464 table
= zebra_vrf_table(AFI_IP
, SAFI_UNICAST
, vrf_id
);
468 rn
= route_node_lookup(table
, (struct prefix
*)p
);
470 /* No route for this prefix. */
475 route_unlock_node(rn
);
476 dest
= rib_dest_from_rnode(rn
);
478 if (dest
&& dest
->selected_fib
479 && !CHECK_FLAG(dest
->selected_fib
->status
, ROUTE_ENTRY_REMOVED
))
480 match
= dest
->selected_fib
;
485 if (match
->type
== ZEBRA_ROUTE_CONNECT
)
488 if (CHECK_FLAG(match
->status
, ROUTE_ENTRY_INSTALLED
))
495 * Is this RIB labeled-unicast? It must be of type BGP and all paths
496 * (nexthops) must have a label.
498 int zebra_rib_labeled_unicast(struct route_entry
*re
)
500 struct nexthop
*nexthop
= NULL
;
502 if (re
->type
!= ZEBRA_ROUTE_BGP
)
505 for (ALL_NEXTHOPS(re
->ng
, nexthop
))
506 if (!nexthop
->nh_label
|| !nexthop
->nh_label
->num_labels
)
512 /* Update flag indicates whether this is a "replace" or not. Currently, this
513 * is only used for IPv4.
515 void rib_install_kernel(struct route_node
*rn
, struct route_entry
*re
,
516 struct route_entry
*old
)
518 struct nexthop
*nexthop
;
519 rib_table_info_t
*info
= srcdest_rnode_table_info(rn
);
520 struct zebra_vrf
*zvrf
= vrf_info_lookup(re
->vrf_id
);
521 const struct prefix
*p
, *src_p
;
522 enum zebra_dplane_result ret
;
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(re
->ng
, nexthop
))
530 SET_FLAG(nexthop
->flags
, NEXTHOP_FLAG_FIB
);
533 struct nexthop
*prev
;
535 for (ALL_NEXTHOPS(re
->ng
, nexthop
)) {
536 UNSET_FLAG (nexthop
->flags
, NEXTHOP_FLAG_DUPLICATE
);
537 for (ALL_NEXTHOPS(re
->ng
, prev
)) {
540 if (nexthop_same_firsthop(nexthop
, prev
)) {
541 SET_FLAG(nexthop
->flags
,
542 NEXTHOP_FLAG_DUPLICATE
);
550 * If this is a replace to a new RE let the originator of the RE
551 * know that they've lost
553 if (old
&& (old
!= re
) && (old
->type
!= re
->type
))
554 zsend_route_notify_owner(old
, p
, ZAPI_ROUTE_BETTER_ADMIN_WON
);
556 /* Update fib selection */
557 dest
->selected_fib
= re
;
560 * Make sure we update the FPM any time we send new information to
563 hook_call(rib_update
, rn
, "installing in kernel");
565 /* Send add or update */
567 ret
= dplane_route_update(rn
, re
, old
);
569 ret
= dplane_route_add(rn
, re
);
572 case ZEBRA_DPLANE_REQUEST_QUEUED
:
573 SET_FLAG(re
->status
, ROUTE_ENTRY_QUEUED
);
576 SET_FLAG(old
->status
, ROUTE_ENTRY_QUEUED
);
578 /* Free old FIB nexthop group */
579 if (old
->fib_ng
.nexthop
) {
580 nexthops_free(old
->fib_ng
.nexthop
);
581 old
->fib_ng
.nexthop
= NULL
;
584 if (!RIB_SYSTEM_ROUTE(old
)) {
585 /* Clear old route's FIB flags */
586 for (ALL_NEXTHOPS(old
->ng
, nexthop
)) {
587 UNSET_FLAG(nexthop
->flags
,
594 zvrf
->installs_queued
++;
596 case ZEBRA_DPLANE_REQUEST_FAILURE
:
598 char str
[SRCDEST2STR_BUFFER
];
600 srcdest_rnode2str(rn
, str
, sizeof(str
));
601 flog_err(EC_ZEBRA_DP_INSTALL_FAIL
,
602 "%u:%s: Failed to enqueue dataplane install",
606 case ZEBRA_DPLANE_REQUEST_SUCCESS
:
615 /* Uninstall the route from kernel. */
616 void rib_uninstall_kernel(struct route_node
*rn
, struct route_entry
*re
)
618 struct nexthop
*nexthop
;
619 rib_table_info_t
*info
= srcdest_rnode_table_info(rn
);
620 struct zebra_vrf
*zvrf
= vrf_info_lookup(re
->vrf_id
);
622 if (info
->safi
!= SAFI_UNICAST
) {
623 UNSET_FLAG(re
->status
, ROUTE_ENTRY_INSTALLED
);
624 for (ALL_NEXTHOPS(re
->ng
, nexthop
))
625 UNSET_FLAG(nexthop
->flags
, NEXTHOP_FLAG_FIB
);
630 * Make sure we update the FPM any time we send new information to
633 hook_call(rib_update
, rn
, "uninstalling from kernel");
635 switch (dplane_route_delete(rn
, re
)) {
636 case ZEBRA_DPLANE_REQUEST_QUEUED
:
638 zvrf
->removals_queued
++;
640 case ZEBRA_DPLANE_REQUEST_FAILURE
:
642 char str
[SRCDEST2STR_BUFFER
];
644 srcdest_rnode2str(rn
, str
, sizeof(str
));
645 flog_err(EC_ZEBRA_DP_INSTALL_FAIL
,
646 "%u:%s: Failed to enqueue dataplane uninstall",
650 case ZEBRA_DPLANE_REQUEST_SUCCESS
:
659 /* Uninstall the route from kernel. */
660 static void rib_uninstall(struct route_node
*rn
, struct route_entry
*re
)
662 rib_table_info_t
*info
= srcdest_rnode_table_info(rn
);
663 rib_dest_t
*dest
= rib_dest_from_rnode(rn
);
664 struct nexthop
*nexthop
;
666 if (dest
&& dest
->selected_fib
== re
) {
667 if (info
->safi
== SAFI_UNICAST
)
668 hook_call(rib_update
, rn
, "rib_uninstall");
670 /* If labeled-unicast route, uninstall transit LSP. */
671 if (zebra_rib_labeled_unicast(re
))
672 zebra_mpls_lsp_uninstall(info
->zvrf
, rn
, re
);
674 rib_uninstall_kernel(rn
, re
);
676 dest
->selected_fib
= NULL
;
678 /* Free FIB nexthop group, if present */
679 if (re
->fib_ng
.nexthop
) {
680 nexthops_free(re
->fib_ng
.nexthop
);
681 re
->fib_ng
.nexthop
= NULL
;
684 for (ALL_NEXTHOPS(re
->ng
, nexthop
))
685 UNSET_FLAG(nexthop
->flags
, NEXTHOP_FLAG_FIB
);
688 if (CHECK_FLAG(re
->flags
, ZEBRA_FLAG_SELECTED
)) {
689 const struct prefix
*p
, *src_p
;
691 srcdest_rnode_prefixes(rn
, &p
, &src_p
);
693 redistribute_delete(p
, src_p
, re
);
694 UNSET_FLAG(re
->flags
, ZEBRA_FLAG_SELECTED
);
699 * rib_can_delete_dest
701 * Returns true if the given dest can be deleted from the table.
703 static int rib_can_delete_dest(rib_dest_t
*dest
)
705 if (re_list_first(&dest
->routes
)) {
710 * Unresolved rnh's are stored on the default route's list
712 * dest->rnode can also be the source prefix node in an
713 * ipv6 sourcedest table. Fortunately the prefix of a
714 * source prefix node can never be the default prefix.
716 if (is_default_prefix(&dest
->rnode
->p
))
720 * Don't delete the dest if we have to update the FPM about this
723 if (CHECK_FLAG(dest
->flags
, RIB_DEST_UPDATE_FPM
)
724 || CHECK_FLAG(dest
->flags
, RIB_DEST_SENT_TO_FPM
))
730 void zebra_rib_evaluate_rn_nexthops(struct route_node
*rn
, uint32_t seq
)
732 rib_dest_t
*dest
= rib_dest_from_rnode(rn
);
736 * We are storing the rnh's associated withb
737 * the tracked nexthop as a list of the rn's.
738 * Unresolved rnh's are placed at the top
739 * of the tree list.( 0.0.0.0/0 for v4 and 0::0/0 for v6 )
740 * As such for each rn we need to walk up the tree
741 * and see if any rnh's need to see if they
742 * would match a more specific route
745 if (IS_ZEBRA_DEBUG_NHT_DETAILED
) {
746 char buf
[PREFIX_STRLEN
];
748 zlog_debug("%s: %s Being examined for Nexthop Tracking Count: %zd",
750 srcdest_rnode2str(rn
, buf
, sizeof(buf
)),
751 dest
? rnh_list_count(&dest
->nht
) : 0);
756 dest
= rib_dest_from_rnode(rn
);
760 * If we have any rnh's stored in the nht list
761 * then we know that this route node was used for
762 * nht resolution and as such we need to call the
763 * nexthop tracking evaluation code
765 frr_each_safe(rnh_list
, &dest
->nht
, rnh
) {
766 struct zebra_vrf
*zvrf
=
767 zebra_vrf_lookup_by_id(rnh
->vrf_id
);
768 struct prefix
*p
= &rnh
->node
->p
;
770 if (IS_ZEBRA_DEBUG_NHT_DETAILED
) {
771 char buf1
[PREFIX_STRLEN
];
772 char buf2
[PREFIX_STRLEN
];
774 zlog_debug("%u:%s has Nexthop(%s) Type: %s depending on it, evaluating %u:%u",
776 srcdest_rnode2str(rn
, buf1
,
778 prefix2str(p
, buf2
, sizeof(buf2
)),
779 rnh_type2str(rnh
->type
),
784 * If we have evaluated this node on this pass
785 * already, due to following the tree up
786 * then we know that we can move onto the next
789 * Additionally we call zebra_evaluate_rnh
790 * when we gc the dest. In this case we know
791 * that there must be no other re's where
792 * we were originally as such we know that
793 * that sequence number is ok to respect.
795 if (rnh
->seqno
== seq
) {
796 if (IS_ZEBRA_DEBUG_NHT_DETAILED
)
798 "\tNode processed and moved already");
803 zebra_evaluate_rnh(zvrf
, family2afi(p
->family
), 0,
809 dest
= rib_dest_from_rnode(rn
);
816 * Garbage collect the rib dest corresponding to the given route node
819 * Returns true if the dest was deleted, false otherwise.
821 int rib_gc_dest(struct route_node
*rn
)
825 dest
= rib_dest_from_rnode(rn
);
829 if (!rib_can_delete_dest(dest
))
832 if (IS_ZEBRA_DEBUG_RIB
) {
833 struct zebra_vrf
*zvrf
;
835 zvrf
= rib_dest_vrf(dest
);
836 rnode_debug(rn
, zvrf_id(zvrf
), "removing dest from table");
839 zebra_rib_evaluate_rn_nexthops(rn
, zebra_router_get_next_sequence());
842 rnh_list_fini(&dest
->nht
);
843 XFREE(MTYPE_RIB_DEST
, dest
);
847 * Release the one reference that we keep on the route node.
849 route_unlock_node(rn
);
853 static void rib_process_add_fib(struct zebra_vrf
*zvrf
, struct route_node
*rn
,
854 struct route_entry
*new)
856 hook_call(rib_update
, rn
, "new route selected");
858 /* Update real nexthop. This may actually determine if nexthop is active
860 if (!nexthop_group_active_nexthop_num(&new->ng
)) {
861 UNSET_FLAG(new->status
, ROUTE_ENTRY_CHANGED
);
865 if (IS_ZEBRA_DEBUG_RIB
) {
866 char buf
[SRCDEST2STR_BUFFER
];
867 srcdest_rnode2str(rn
, buf
, sizeof(buf
));
868 zlog_debug("%u:%s: Adding route rn %p, re %p (%s)",
869 zvrf_id(zvrf
), buf
, rn
, new,
870 zebra_route_string(new->type
));
873 /* If labeled-unicast route, install transit LSP. */
874 if (zebra_rib_labeled_unicast(new))
875 zebra_mpls_lsp_install(zvrf
, rn
, new);
877 rib_install_kernel(rn
, new, NULL
);
879 UNSET_FLAG(new->status
, ROUTE_ENTRY_CHANGED
);
882 static void rib_process_del_fib(struct zebra_vrf
*zvrf
, struct route_node
*rn
,
883 struct route_entry
*old
)
885 hook_call(rib_update
, rn
, "removing existing route");
887 /* Uninstall from kernel. */
888 if (IS_ZEBRA_DEBUG_RIB
) {
889 char buf
[SRCDEST2STR_BUFFER
];
890 srcdest_rnode2str(rn
, buf
, sizeof(buf
));
891 zlog_debug("%u:%s: Deleting route rn %p, re %p (%s)",
892 zvrf_id(zvrf
), buf
, rn
, old
,
893 zebra_route_string(old
->type
));
896 /* If labeled-unicast route, uninstall transit LSP. */
897 if (zebra_rib_labeled_unicast(old
))
898 zebra_mpls_lsp_uninstall(zvrf
, rn
, old
);
900 rib_uninstall_kernel(rn
, old
);
902 /* Update nexthop for route, reset changed flag. */
903 /* Note: this code also handles the Linux case when an interface goes
904 * down, causing the kernel to delete routes without sending DELROUTE
907 if (RIB_KERNEL_ROUTE(old
))
908 SET_FLAG(old
->status
, ROUTE_ENTRY_REMOVED
);
910 UNSET_FLAG(old
->status
, ROUTE_ENTRY_CHANGED
);
913 static void rib_process_update_fib(struct zebra_vrf
*zvrf
,
914 struct route_node
*rn
,
915 struct route_entry
*old
,
916 struct route_entry
*new)
921 * We have to install or update if a new route has been selected or
922 * something has changed.
924 if (new != old
|| CHECK_FLAG(new->status
, ROUTE_ENTRY_CHANGED
)) {
925 hook_call(rib_update
, rn
, "updating existing route");
927 /* Update the nexthop; we could determine here that nexthop is
929 if (nexthop_group_active_nexthop_num(&new->ng
))
932 /* If nexthop is active, install the selected route, if
934 * the install succeeds, cleanup flags for prior route, if
939 if (IS_ZEBRA_DEBUG_RIB
) {
940 char buf
[SRCDEST2STR_BUFFER
];
941 srcdest_rnode2str(rn
, buf
, sizeof(buf
));
944 "%u:%s: Updating route rn %p, re %p (%s) old %p (%s)",
945 zvrf_id(zvrf
), buf
, rn
, new,
946 zebra_route_string(new->type
),
948 zebra_route_string(old
->type
));
951 "%u:%s: Updating route rn %p, re %p (%s)",
952 zvrf_id(zvrf
), buf
, rn
, new,
953 zebra_route_string(new->type
));
956 /* If labeled-unicast route, uninstall transit LSP. */
957 if (zebra_rib_labeled_unicast(old
))
958 zebra_mpls_lsp_uninstall(zvrf
, rn
, old
);
961 * Non-system route should be installed.
962 * If labeled-unicast route, install transit
965 if (zebra_rib_labeled_unicast(new))
966 zebra_mpls_lsp_install(zvrf
, rn
, new);
968 rib_install_kernel(rn
, new, old
);
972 * If nexthop for selected route is not active or install
974 * may need to uninstall and delete for redistribution.
977 if (IS_ZEBRA_DEBUG_RIB
) {
978 char buf
[SRCDEST2STR_BUFFER
];
979 srcdest_rnode2str(rn
, buf
, sizeof(buf
));
982 "%u:%s: Deleting route rn %p, re %p (%s) old %p (%s) - nexthop inactive",
983 zvrf_id(zvrf
), buf
, rn
, new,
984 zebra_route_string(new->type
),
986 zebra_route_string(old
->type
));
989 "%u:%s: Deleting route rn %p, re %p (%s) - nexthop inactive",
990 zvrf_id(zvrf
), buf
, rn
, new,
991 zebra_route_string(new->type
));
994 /* If labeled-unicast route, uninstall transit LSP. */
995 if (zebra_rib_labeled_unicast(old
))
996 zebra_mpls_lsp_uninstall(zvrf
, rn
, old
);
998 rib_uninstall_kernel(rn
, old
);
1002 * Same route selected; check if in the FIB and if not,
1003 * re-install. This is housekeeping code to deal with
1004 * race conditions in kernel with linux netlink reporting
1005 * interface up before IPv4 or IPv6 protocol is ready
1008 if (!CHECK_FLAG(new->status
, ROUTE_ENTRY_INSTALLED
) ||
1009 RIB_SYSTEM_ROUTE(new))
1010 rib_install_kernel(rn
, new, NULL
);
1013 /* Update prior route. */
1015 UNSET_FLAG(old
->status
, ROUTE_ENTRY_CHANGED
);
1017 /* Clear changed flag. */
1018 UNSET_FLAG(new->status
, ROUTE_ENTRY_CHANGED
);
1021 /* Check if 'alternate' RIB entry is better than 'current'. */
1022 static struct route_entry
*rib_choose_best(struct route_entry
*current
,
1023 struct route_entry
*alternate
)
1025 if (current
== NULL
)
1028 /* filter route selection in following order:
1029 * - connected beats other types
1030 * - if both connected, loopback or vrf wins
1031 * - lower distance beats higher
1032 * - lower metric beats higher for equal distance
1033 * - last, hence oldest, route wins tie break.
1036 /* Connected routes. Check to see if either are a vrf
1037 * or loopback interface. If not, pick the last connected
1038 * route of the set of lowest metric connected routes.
1040 if (alternate
->type
== ZEBRA_ROUTE_CONNECT
) {
1041 if (current
->type
!= ZEBRA_ROUTE_CONNECT
)
1044 /* both are connected. are either loop or vrf? */
1045 struct nexthop
*nexthop
= NULL
;
1047 for (ALL_NEXTHOPS(alternate
->ng
, nexthop
)) {
1048 if (if_is_loopback_or_vrf(if_lookup_by_index(
1049 nexthop
->ifindex
, alternate
->vrf_id
)))
1053 for (ALL_NEXTHOPS(current
->ng
, nexthop
)) {
1054 if (if_is_loopback_or_vrf(if_lookup_by_index(
1055 nexthop
->ifindex
, current
->vrf_id
)))
1059 /* Neither are loop or vrf so pick best metric */
1060 if (alternate
->metric
<= current
->metric
)
1066 if (current
->type
== ZEBRA_ROUTE_CONNECT
)
1069 /* higher distance loses */
1070 if (alternate
->distance
< current
->distance
)
1072 if (current
->distance
< alternate
->distance
)
1075 /* metric tie-breaks equal distance */
1076 if (alternate
->metric
<= current
->metric
)
1082 /* Core function for processing routing information base. */
1083 static void rib_process(struct route_node
*rn
)
1085 struct route_entry
*re
;
1086 struct route_entry
*next
;
1087 struct route_entry
*old_selected
= NULL
;
1088 struct route_entry
*new_selected
= NULL
;
1089 struct route_entry
*old_fib
= NULL
;
1090 struct route_entry
*new_fib
= NULL
;
1091 struct route_entry
*best
= NULL
;
1092 char buf
[SRCDEST2STR_BUFFER
];
1094 struct zebra_vrf
*zvrf
= NULL
;
1095 const struct prefix
*p
, *src_p
;
1097 srcdest_rnode_prefixes(rn
, &p
, &src_p
);
1098 vrf_id_t vrf_id
= VRF_UNKNOWN
;
1102 dest
= rib_dest_from_rnode(rn
);
1104 zvrf
= rib_dest_vrf(dest
);
1105 vrf_id
= zvrf_id(zvrf
);
1108 if (IS_ZEBRA_DEBUG_RIB
)
1109 srcdest_rnode2str(rn
, buf
, sizeof(buf
));
1111 if (IS_ZEBRA_DEBUG_RIB_DETAILED
)
1112 zlog_debug("%u:%s: Processing rn %p", vrf_id
, buf
, rn
);
1115 * we can have rn's that have a NULL info pointer
1116 * (dest). As such let's not let the deref happen
1117 * additionally we know RNODE_FOREACH_RE_SAFE
1118 * will not iterate so we are ok.
1121 old_fib
= dest
->selected_fib
;
1123 RNODE_FOREACH_RE_SAFE (rn
, re
, next
) {
1124 if (IS_ZEBRA_DEBUG_RIB_DETAILED
)
1126 "%u:%s: Examine re %p (%s) status %x flags %x dist %d metric %d",
1127 vrf_id
, buf
, re
, zebra_route_string(re
->type
),
1128 re
->status
, re
->flags
, re
->distance
,
1131 UNSET_FLAG(re
->status
, ROUTE_ENTRY_NEXTHOPS_CHANGED
);
1133 /* Currently selected re. */
1134 if (CHECK_FLAG(re
->flags
, ZEBRA_FLAG_SELECTED
)) {
1135 assert(old_selected
== NULL
);
1139 /* Skip deleted entries from selection */
1140 if (CHECK_FLAG(re
->status
, ROUTE_ENTRY_REMOVED
))
1143 /* Skip unreachable nexthop. */
1144 /* This first call to nexthop_active_update is merely to
1145 * determine if there's any change to nexthops associated
1146 * with this RIB entry. Now, rib_process() can be invoked due
1147 * to an external event such as link down or due to
1148 * next-hop-tracking evaluation. In the latter case,
1149 * a decision has already been made that the NHs have changed.
1150 * So, no need to invoke a potentially expensive call again.
1151 * Further, since the change might be in a recursive NH which
1152 * is not caught in the nexthop_active_update() code. Thus, we
1153 * might miss changes to recursive NHs.
1155 if (CHECK_FLAG(re
->status
, ROUTE_ENTRY_CHANGED
)
1156 && !nexthop_active_update(rn
, re
)) {
1157 if (re
->type
== ZEBRA_ROUTE_TABLE
) {
1158 /* XXX: HERE BE DRAGONS!!!!!
1159 * In all honesty, I have not yet figured out
1160 * what this part does or why the
1161 * ROUTE_ENTRY_CHANGED test above is correct
1162 * or why we need to delete a route here, and
1163 * also not whether this concerns both selected
1164 * and fib route, or only selected
1167 * This entry was denied by the 'ip protocol
1168 * table' route-map, we need to delete it */
1169 if (re
!= old_selected
) {
1170 if (IS_ZEBRA_DEBUG_RIB
)
1172 "%s: %u:%s: imported via import-table but denied "
1173 "by the ip protocol table route-map",
1174 __func__
, vrf_id
, buf
);
1177 SET_FLAG(re
->status
,
1178 ROUTE_ENTRY_REMOVED
);
1184 /* Infinite distance. */
1185 if (re
->distance
== DISTANCE_INFINITY
) {
1186 UNSET_FLAG(re
->status
, ROUTE_ENTRY_CHANGED
);
1190 if (CHECK_FLAG(re
->flags
, ZEBRA_FLAG_FIB_OVERRIDE
)) {
1191 best
= rib_choose_best(new_fib
, re
);
1192 if (new_fib
&& best
!= new_fib
)
1193 UNSET_FLAG(new_fib
->status
,
1194 ROUTE_ENTRY_CHANGED
);
1197 best
= rib_choose_best(new_selected
, re
);
1198 if (new_selected
&& best
!= new_selected
)
1199 UNSET_FLAG(new_selected
->status
,
1200 ROUTE_ENTRY_CHANGED
);
1201 new_selected
= best
;
1204 UNSET_FLAG(re
->status
, ROUTE_ENTRY_CHANGED
);
1205 } /* RNODE_FOREACH_RE */
1207 /* If no FIB override route, use the selected route also for FIB */
1208 if (new_fib
== NULL
)
1209 new_fib
= new_selected
;
1211 /* After the cycle is finished, the following pointers will be set:
1212 * old_selected --- RE entry currently having SELECTED
1213 * new_selected --- RE entry that is newly SELECTED
1214 * old_fib --- RE entry currently in kernel FIB
1215 * new_fib --- RE entry that is newly to be in kernel FIB
1217 * new_selected will get SELECTED flag, and is going to be redistributed
1218 * the zclients. new_fib (which can be new_selected) will be installed
1222 if (IS_ZEBRA_DEBUG_RIB_DETAILED
) {
1224 "%u:%s: After processing: old_selected %p new_selected %p old_fib %p new_fib %p",
1225 vrf_id
, buf
, (void *)old_selected
, (void *)new_selected
,
1226 (void *)old_fib
, (void *)new_fib
);
1229 /* Buffer ROUTE_ENTRY_CHANGED here, because it will get cleared if
1230 * fib == selected */
1231 bool selected_changed
= new_selected
&& CHECK_FLAG(new_selected
->status
,
1232 ROUTE_ENTRY_CHANGED
);
1234 /* Update fib according to selection results */
1235 if (new_fib
&& old_fib
)
1236 rib_process_update_fib(zvrf
, rn
, old_fib
, new_fib
);
1238 rib_process_add_fib(zvrf
, rn
, new_fib
);
1240 rib_process_del_fib(zvrf
, rn
, old_fib
);
1242 /* Update SELECTED entry */
1243 if (old_selected
!= new_selected
|| selected_changed
) {
1245 if (new_selected
&& new_selected
!= new_fib
)
1246 UNSET_FLAG(new_selected
->status
, ROUTE_ENTRY_CHANGED
);
1249 SET_FLAG(new_selected
->flags
, ZEBRA_FLAG_SELECTED
);
1253 redistribute_delete(p
, src_p
, old_selected
);
1254 if (old_selected
!= new_selected
)
1255 UNSET_FLAG(old_selected
->flags
,
1256 ZEBRA_FLAG_SELECTED
);
1260 /* Remove all RE entries queued for removal */
1261 RNODE_FOREACH_RE_SAFE (rn
, re
, next
) {
1262 if (CHECK_FLAG(re
->status
, ROUTE_ENTRY_REMOVED
)) {
1263 if (IS_ZEBRA_DEBUG_RIB
) {
1264 rnode_debug(rn
, vrf_id
, "rn %p, removing re %p",
1265 (void *)rn
, (void *)re
);
1272 * Check if the dest can be deleted now.
1277 static void zebra_rib_evaluate_mpls(struct route_node
*rn
)
1279 rib_dest_t
*dest
= rib_dest_from_rnode(rn
);
1280 struct zebra_vrf
*zvrf
= vrf_info_lookup(VRF_DEFAULT
);
1285 if (CHECK_FLAG(dest
->flags
, RIB_DEST_UPDATE_LSPS
)) {
1286 if (IS_ZEBRA_DEBUG_MPLS
)
1288 "%u: Scheduling all LSPs upon RIB completion",
1290 zebra_mpls_lsp_schedule(zvrf
);
1291 mpls_unmark_lsps_for_processing(rn
);
1296 * Utility to match route with dplane context data
1298 static bool rib_route_match_ctx(const struct route_entry
*re
,
1299 const struct zebra_dplane_ctx
*ctx
,
1302 bool result
= false;
1306 * In 'update' case, we test info about the 'previous' or
1309 if ((re
->type
== dplane_ctx_get_old_type(ctx
)) &&
1310 (re
->instance
== dplane_ctx_get_old_instance(ctx
))) {
1313 /* TODO -- we're using this extra test, but it's not
1314 * exactly clear why.
1316 if (re
->type
== ZEBRA_ROUTE_STATIC
&&
1317 (re
->distance
!= dplane_ctx_get_old_distance(ctx
) ||
1318 re
->tag
!= dplane_ctx_get_old_tag(ctx
))) {
1325 * Ordinary, single-route case using primary context info
1327 if ((dplane_ctx_get_op(ctx
) != DPLANE_OP_ROUTE_DELETE
) &&
1328 CHECK_FLAG(re
->status
, ROUTE_ENTRY_REMOVED
)) {
1329 /* Skip route that's been deleted */
1333 if ((re
->type
== dplane_ctx_get_type(ctx
)) &&
1334 (re
->instance
== dplane_ctx_get_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_distance(ctx
) ||
1342 re
->tag
!= dplane_ctx_get_tag(ctx
))) {
1353 static void zebra_rib_fixup_system(struct route_node
*rn
)
1355 struct route_entry
*re
;
1357 RNODE_FOREACH_RE(rn
, re
) {
1358 struct nexthop
*nhop
;
1360 if (!RIB_SYSTEM_ROUTE(re
))
1363 if (CHECK_FLAG(re
->status
, ROUTE_ENTRY_REMOVED
))
1366 SET_FLAG(re
->status
, ROUTE_ENTRY_INSTALLED
);
1368 for (ALL_NEXTHOPS(re
->ng
, nhop
)) {
1369 if (CHECK_FLAG(nhop
->flags
, NEXTHOP_FLAG_RECURSIVE
))
1372 SET_FLAG(nhop
->flags
, NEXTHOP_FLAG_FIB
);
1378 * Update a route from a dplane context. This consolidates common code
1379 * that can be used in processing of results from FIB updates, and in
1380 * async notification processing.
1381 * The return is 'true' if the installed nexthops changed; 'false' otherwise.
1383 static bool rib_update_re_from_ctx(struct route_entry
*re
,
1384 struct route_node
*rn
,
1385 struct zebra_dplane_ctx
*ctx
)
1387 char dest_str
[PREFIX_STRLEN
] = "";
1388 char nh_str
[NEXTHOP_STRLEN
];
1389 struct nexthop
*nexthop
, *ctx_nexthop
;
1391 const struct nexthop_group
*ctxnhg
;
1392 bool is_selected
= false; /* Is 're' currently the selected re? */
1393 bool changed_p
= false; /* Change to nexthops? */
1396 /* Note well: only capturing the prefix string if debug is enabled here;
1397 * unconditional log messages will have to generate the string.
1399 if (IS_ZEBRA_DEBUG_RIB
)
1400 prefix2str(&(rn
->p
), dest_str
, sizeof(dest_str
));
1402 dest
= rib_dest_from_rnode(rn
);
1404 is_selected
= (re
== dest
->selected_fib
);
1406 if (IS_ZEBRA_DEBUG_RIB_DETAILED
)
1407 zlog_debug("update_from_ctx: %u:%s: %sSELECTED",
1408 re
->vrf_id
, dest_str
, (is_selected
? "" : "NOT "));
1410 /* Update zebra's nexthop FIB flag for each nexthop that was installed.
1411 * If the installed set differs from the set requested by the rib/owner,
1412 * we use the fib-specific nexthop-group to record the actual FIB
1417 * First check the fib nexthop-group, if it's present. The comparison
1418 * here is quite strict: we require that the fib sets match exactly.
1422 if (re
->fib_ng
.nexthop
== NULL
)
1427 /* First check the route's fib nexthops */
1428 for (ALL_NEXTHOPS(re
->fib_ng
, nexthop
)) {
1430 if (CHECK_FLAG(nexthop
->flags
, NEXTHOP_FLAG_RECURSIVE
))
1434 for (ALL_NEXTHOPS_PTR(dplane_ctx_get_ng(ctx
),
1436 if (nexthop_same(ctx_nexthop
, nexthop
))
1440 if (ctx_nexthop
== NULL
) {
1441 /* Nexthop not in the new installed set */
1442 if (IS_ZEBRA_DEBUG_RIB_DETAILED
) {
1443 nexthop2str(nexthop
, nh_str
,
1445 zlog_debug("update_from_ctx: no match for fib nh %s",
1457 /* Check the new installed set */
1459 for (ALL_NEXTHOPS_PTR(dplane_ctx_get_ng(ctx
), ctx_nexthop
)) {
1461 if (CHECK_FLAG(ctx_nexthop
->flags
,
1462 NEXTHOP_FLAG_RECURSIVE
))
1465 /* Compare with the current group's nexthops */
1467 for (ALL_NEXTHOPS(re
->fib_ng
, nexthop
)) {
1468 if (nexthop_same(nexthop
, ctx_nexthop
))
1472 if (nexthop
== NULL
) {
1473 /* Nexthop not in the old installed set */
1474 if (IS_ZEBRA_DEBUG_RIB_DETAILED
) {
1475 nexthop2str(ctx_nexthop
, nh_str
,
1477 zlog_debug("update_from_ctx: no fib match for notif nh %s",
1487 /* If the new FIB set matches the existing FIB set, we're done. */
1489 if (IS_ZEBRA_DEBUG_RIB
)
1490 zlog_debug("%u:%s update_from_ctx(): existing fib nhg, no change",
1491 re
->vrf_id
, dest_str
);
1494 } else if (re
->fib_ng
.nexthop
) {
1496 * Free stale fib list and move on to check the rib nhg.
1498 if (IS_ZEBRA_DEBUG_RIB
)
1499 zlog_debug("%u:%s update_from_ctx(): replacing fib nhg",
1500 re
->vrf_id
, dest_str
);
1501 nexthops_free(re
->fib_ng
.nexthop
);
1502 re
->fib_ng
.nexthop
= NULL
;
1504 /* Note that the installed nexthops have changed */
1507 if (IS_ZEBRA_DEBUG_RIB
)
1508 zlog_debug("%u:%s update_from_ctx(): no fib nhg",
1509 re
->vrf_id
, dest_str
);
1513 * Compare with the rib nexthop group. The comparison here is different:
1514 * the RIB group may be a superset of the list installed in the FIB. We
1515 * walk the RIB group, looking for the 'installable' candidate
1516 * nexthops, and then check those against the set
1517 * that is actually installed.
1520 for (ALL_NEXTHOPS(re
->ng
, nexthop
)) {
1522 if (CHECK_FLAG(nexthop
->flags
, NEXTHOP_FLAG_RECURSIVE
))
1525 if (!CHECK_FLAG(nexthop
->flags
, NEXTHOP_FLAG_ACTIVE
))
1528 /* Check for a FIB nexthop corresponding to the RIB nexthop */
1530 for (ALL_NEXTHOPS_PTR(dplane_ctx_get_ng(ctx
), ctx_nexthop
)) {
1531 if (nexthop_same(ctx_nexthop
, nexthop
))
1535 /* If the FIB doesn't know about the nexthop,
1536 * it's not installed
1538 if (ctx_nexthop
== NULL
) {
1539 if (IS_ZEBRA_DEBUG_RIB_DETAILED
) {
1540 nexthop2str(nexthop
, nh_str
, sizeof(nh_str
));
1541 zlog_debug("update_from_ctx: no notif match for rib nh %s",
1546 if (CHECK_FLAG(nexthop
->flags
, NEXTHOP_FLAG_FIB
))
1549 UNSET_FLAG(nexthop
->flags
, NEXTHOP_FLAG_FIB
);
1553 if (CHECK_FLAG(ctx_nexthop
->flags
, NEXTHOP_FLAG_FIB
)) {
1554 if (!CHECK_FLAG(nexthop
->flags
, NEXTHOP_FLAG_FIB
))
1557 SET_FLAG(nexthop
->flags
, NEXTHOP_FLAG_FIB
);
1559 if (CHECK_FLAG(nexthop
->flags
, NEXTHOP_FLAG_FIB
))
1562 UNSET_FLAG(nexthop
->flags
, NEXTHOP_FLAG_FIB
);
1566 /* If all nexthops were processed, we're done */
1568 if (IS_ZEBRA_DEBUG_RIB
)
1569 zlog_debug("%u:%s update_from_ctx(): rib nhg matched, changed '%s'",
1570 re
->vrf_id
, dest_str
,
1571 (changed_p
? "true" : "false"));
1575 /* FIB nexthop set differs from the RIB set:
1576 * create a fib-specific nexthop-group
1578 if (IS_ZEBRA_DEBUG_RIB
)
1579 zlog_debug("%u:%s update_from_ctx(): changed %s, adding new fib nhg",
1580 re
->vrf_id
, dest_str
,
1581 (changed_p
? "true" : "false"));
1583 ctxnhg
= dplane_ctx_get_ng(ctx
);
1585 if (ctxnhg
->nexthop
)
1586 copy_nexthops(&(re
->fib_ng
.nexthop
), ctxnhg
->nexthop
, NULL
);
1588 /* Bit of a special case when the fib has _no_ installed
1591 nexthop
= nexthop_new();
1592 nexthop
->type
= NEXTHOP_TYPE_IPV4
;
1593 _nexthop_add(&(re
->fib_ng
.nexthop
), nexthop
);
1601 * Helper to locate a zebra route-node from a dplane context. This is used
1602 * when processing dplane results, e.g. Note well: the route-node is returned
1603 * with a ref held - route_unlock_node() must be called eventually.
1605 static struct route_node
*
1606 rib_find_rn_from_ctx(const struct zebra_dplane_ctx
*ctx
)
1608 struct route_table
*table
= NULL
;
1609 struct route_node
*rn
= NULL
;
1610 const struct prefix
*dest_pfx
, *src_pfx
;
1612 /* Locate rn and re(s) from ctx */
1614 table
= zebra_vrf_table_with_table_id(dplane_ctx_get_afi(ctx
),
1615 dplane_ctx_get_safi(ctx
),
1616 dplane_ctx_get_vrf(ctx
),
1617 dplane_ctx_get_table(ctx
));
1618 if (table
== NULL
) {
1619 if (IS_ZEBRA_DEBUG_DPLANE
) {
1620 zlog_debug("Failed to find route for ctx: no table for afi %d, safi %d, vrf %u",
1621 dplane_ctx_get_afi(ctx
),
1622 dplane_ctx_get_safi(ctx
),
1623 dplane_ctx_get_vrf(ctx
));
1628 dest_pfx
= dplane_ctx_get_dest(ctx
);
1629 src_pfx
= dplane_ctx_get_src(ctx
);
1631 rn
= srcdest_rnode_get(table
, dest_pfx
,
1632 src_pfx
? (struct prefix_ipv6
*)src_pfx
: NULL
);
1641 * Route-update results processing after async dataplane update.
1643 static void rib_process_result(struct zebra_dplane_ctx
*ctx
)
1645 struct zebra_vrf
*zvrf
= NULL
;
1646 struct route_node
*rn
= NULL
;
1647 struct route_entry
*re
= NULL
, *old_re
= NULL
, *rib
;
1648 bool is_update
= false;
1649 char dest_str
[PREFIX_STRLEN
] = "";
1650 enum dplane_op_e op
;
1651 enum zebra_dplane_result status
;
1652 const struct prefix
*dest_pfx
, *src_pfx
;
1654 bool fib_changed
= false;
1656 zvrf
= vrf_info_lookup(dplane_ctx_get_vrf(ctx
));
1657 dest_pfx
= dplane_ctx_get_dest(ctx
);
1659 /* Note well: only capturing the prefix string if debug is enabled here;
1660 * unconditional log messages will have to generate the string.
1662 if (IS_ZEBRA_DEBUG_DPLANE
)
1663 prefix2str(dest_pfx
, dest_str
, sizeof(dest_str
));
1665 /* Locate rn and re(s) from ctx */
1666 rn
= rib_find_rn_from_ctx(ctx
);
1668 if (IS_ZEBRA_DEBUG_DPLANE
) {
1669 zlog_debug("Failed to process dplane results: no route for %u:%s",
1670 dplane_ctx_get_vrf(ctx
), dest_str
);
1675 srcdest_rnode_prefixes(rn
, &dest_pfx
, &src_pfx
);
1677 op
= dplane_ctx_get_op(ctx
);
1678 status
= dplane_ctx_get_status(ctx
);
1680 if (IS_ZEBRA_DEBUG_DPLANE_DETAIL
)
1681 zlog_debug("%u:%s Processing dplane ctx %p, op %s result %s",
1682 dplane_ctx_get_vrf(ctx
), dest_str
, ctx
,
1683 dplane_op2str(op
), dplane_res2str(status
));
1686 * Update is a bit of a special case, where we may have both old and new
1687 * routes to post-process.
1689 is_update
= dplane_ctx_is_update(ctx
);
1692 * Take a pass through the routes, look for matches with the context
1695 RNODE_FOREACH_RE(rn
, rib
) {
1698 if (rib_route_match_ctx(rib
, ctx
, false))
1702 /* Check for old route match */
1703 if (is_update
&& (old_re
== NULL
)) {
1704 if (rib_route_match_ctx(rib
, ctx
, true /*is_update*/))
1708 /* Have we found the routes we need to work on? */
1709 if (re
&& ((!is_update
|| old_re
)))
1713 seq
= dplane_ctx_get_seq(ctx
);
1716 * Check sequence number(s) to detect stale results before continuing
1719 if (re
->dplane_sequence
!= seq
) {
1720 if (IS_ZEBRA_DEBUG_DPLANE_DETAIL
)
1721 zlog_debug("%u:%s Stale dplane result for re %p",
1722 dplane_ctx_get_vrf(ctx
),
1725 UNSET_FLAG(re
->status
, ROUTE_ENTRY_QUEUED
);
1729 if (old_re
->dplane_sequence
!= dplane_ctx_get_old_seq(ctx
)) {
1730 if (IS_ZEBRA_DEBUG_DPLANE_DETAIL
)
1731 zlog_debug("%u:%s Stale dplane result for old_re %p",
1732 dplane_ctx_get_vrf(ctx
),
1735 UNSET_FLAG(old_re
->status
, ROUTE_ENTRY_QUEUED
);
1739 case DPLANE_OP_ROUTE_INSTALL
:
1740 case DPLANE_OP_ROUTE_UPDATE
:
1741 if (status
== ZEBRA_DPLANE_REQUEST_SUCCESS
) {
1743 UNSET_FLAG(re
->status
, ROUTE_ENTRY_FAILED
);
1744 SET_FLAG(re
->status
, ROUTE_ENTRY_INSTALLED
);
1747 * On an update operation from the same route type
1748 * context retrieval currently has no way to know
1749 * which was the old and which was the new.
1750 * So don't unset our flags that we just set.
1751 * We know redistribution is ok because the
1752 * old_re in this case is used for nothing
1753 * more than knowing whom to contact if necessary.
1755 if (old_re
&& old_re
!= re
) {
1756 UNSET_FLAG(old_re
->status
, ROUTE_ENTRY_FAILED
);
1757 UNSET_FLAG(old_re
->status
,
1758 ROUTE_ENTRY_INSTALLED
);
1761 /* Update zebra route based on the results in
1762 * the context struct.
1766 rib_update_re_from_ctx(re
, rn
, ctx
);
1769 if (IS_ZEBRA_DEBUG_DPLANE_DETAIL
)
1770 zlog_debug("%u:%s no fib change for re",
1777 redistribute_update(dest_pfx
, src_pfx
,
1782 * System routes are weird in that they
1783 * allow multiple to be installed that match
1784 * to the same prefix, so after we get the
1785 * result we need to clean them up so that
1786 * we can actually use them.
1788 if ((re
&& RIB_SYSTEM_ROUTE(re
)) ||
1789 (old_re
&& RIB_SYSTEM_ROUTE(old_re
)))
1790 zebra_rib_fixup_system(rn
);
1795 /* Notify route owner */
1796 zsend_route_notify_owner_ctx(ctx
, ZAPI_ROUTE_INSTALLED
);
1800 SET_FLAG(re
->status
, ROUTE_ENTRY_FAILED
);
1801 UNSET_FLAG(re
->status
, ROUTE_ENTRY_INSTALLED
);
1803 SET_FLAG(old_re
->status
, ROUTE_ENTRY_FAILED
);
1805 zsend_route_notify_owner(re
, dest_pfx
,
1806 ZAPI_ROUTE_FAIL_INSTALL
);
1808 zlog_warn("%u:%s: Route install failed",
1809 dplane_ctx_get_vrf(ctx
),
1810 prefix2str(dest_pfx
,
1811 dest_str
, sizeof(dest_str
)));
1814 case DPLANE_OP_ROUTE_DELETE
:
1816 SET_FLAG(re
->status
, ROUTE_ENTRY_FAILED
);
1818 * In the delete case, the zebra core datastructs were
1819 * updated (or removed) at the time the delete was issued,
1820 * so we're just notifying the route owner.
1822 if (status
== ZEBRA_DPLANE_REQUEST_SUCCESS
) {
1824 UNSET_FLAG(re
->status
, ROUTE_ENTRY_INSTALLED
);
1825 UNSET_FLAG(re
->status
, ROUTE_ENTRY_FAILED
);
1827 zsend_route_notify_owner_ctx(ctx
, ZAPI_ROUTE_REMOVED
);
1833 SET_FLAG(re
->status
, ROUTE_ENTRY_FAILED
);
1834 zsend_route_notify_owner_ctx(ctx
,
1835 ZAPI_ROUTE_REMOVE_FAIL
);
1837 zlog_warn("%u:%s: Route Deletion failure",
1838 dplane_ctx_get_vrf(ctx
),
1839 prefix2str(dest_pfx
,
1840 dest_str
, sizeof(dest_str
)));
1844 * System routes are weird in that they
1845 * allow multiple to be installed that match
1846 * to the same prefix, so after we get the
1847 * result we need to clean them up so that
1848 * we can actually use them.
1850 if ((re
&& RIB_SYSTEM_ROUTE(re
)) ||
1851 (old_re
&& RIB_SYSTEM_ROUTE(old_re
)))
1852 zebra_rib_fixup_system(rn
);
1858 zebra_rib_evaluate_rn_nexthops(rn
, seq
);
1859 zebra_rib_evaluate_mpls(rn
);
1863 route_unlock_node(rn
);
1865 /* Return context to dataplane module */
1866 dplane_ctx_fini(&ctx
);
1870 * Handle notification from async dataplane: the dataplane has detected
1871 * some change to a route, and notifies zebra so that the control plane
1872 * can reflect that change.
1874 static void rib_process_dplane_notify(struct zebra_dplane_ctx
*ctx
)
1876 struct route_node
*rn
= NULL
;
1877 struct route_entry
*re
= NULL
;
1878 struct nexthop
*nexthop
;
1879 char dest_str
[PREFIX_STRLEN
] = "";
1880 const struct prefix
*dest_pfx
, *src_pfx
;
1882 bool fib_changed
= false;
1883 bool debug_p
= IS_ZEBRA_DEBUG_DPLANE
| IS_ZEBRA_DEBUG_RIB
;
1884 int start_count
, end_count
;
1885 dest_pfx
= dplane_ctx_get_dest(ctx
);
1887 /* Note well: only capturing the prefix string if debug is enabled here;
1888 * unconditional log messages will have to generate the string.
1891 prefix2str(dest_pfx
, dest_str
, sizeof(dest_str
));
1893 /* Locate rn and re(s) from ctx */
1894 rn
= rib_find_rn_from_ctx(ctx
);
1897 zlog_debug("Failed to process dplane notification: no routes for %u:%s",
1898 dplane_ctx_get_vrf(ctx
), dest_str
);
1903 dest
= rib_dest_from_rnode(rn
);
1904 srcdest_rnode_prefixes(rn
, &dest_pfx
, &src_pfx
);
1907 zlog_debug("%u:%s Processing dplane notif ctx %p",
1908 dplane_ctx_get_vrf(ctx
), dest_str
, ctx
);
1911 * Take a pass through the routes, look for matches with the context
1914 RNODE_FOREACH_RE(rn
, re
) {
1915 if (rib_route_match_ctx(re
, ctx
, false /*!update*/))
1919 /* No match? Nothing we can do */
1922 zlog_debug("%u:%s Unable to process dplane notification: no entry for type %s",
1923 dplane_ctx_get_vrf(ctx
), dest_str
,
1925 dplane_ctx_get_type(ctx
)));
1930 /* Is this a notification that ... matters? We only really care about
1931 * the route that is currently selected for installation.
1933 if (re
!= dest
->selected_fib
) {
1934 /* TODO -- don't skip processing entirely? We might like to
1935 * at least report on the event.
1938 zlog_debug("%u:%s dplane notif, but type %s not selected_fib",
1939 dplane_ctx_get_vrf(ctx
), dest_str
,
1941 dplane_ctx_get_type(ctx
)));
1945 /* We'll want to determine whether the installation status of the
1946 * route has changed: we'll check the status before processing,
1947 * and then again if there's been a change.
1950 for (ALL_NEXTHOPS_PTR(rib_active_nhg(re
), nexthop
)) {
1951 if (CHECK_FLAG(nexthop
->flags
, NEXTHOP_FLAG_FIB
))
1955 /* Update zebra's nexthop FIB flags based on the context struct's
1958 fib_changed
= rib_update_re_from_ctx(re
, rn
, ctx
);
1962 zlog_debug("%u:%s No change from dplane notification",
1963 dplane_ctx_get_vrf(ctx
), dest_str
);
1969 * Perform follow-up work if the actual status of the prefix
1974 for (ALL_NEXTHOPS_PTR(rib_active_nhg(re
), nexthop
)) {
1975 if (CHECK_FLAG(nexthop
->flags
, NEXTHOP_FLAG_FIB
))
1979 /* Various fib transitions: changed nexthops; from installed to
1980 * not-installed; or not-installed to installed.
1982 if (start_count
> 0 && end_count
> 0) {
1984 /* Changed nexthops - update kernel/others */
1985 dplane_route_notif_update(rn
, re
,
1986 DPLANE_OP_ROUTE_UPDATE
, ctx
);
1988 } else if (start_count
== 0 && end_count
> 0) {
1990 zlog_debug("%u:%s installed transition from dplane notification",
1991 dplane_ctx_get_vrf(ctx
), dest_str
);
1993 /* We expect this to be the selected route, so we want
1994 * to tell others about this transistion.
1996 SET_FLAG(re
->status
, ROUTE_ENTRY_INSTALLED
);
1998 /* Changed nexthops - update kernel/others */
1999 dplane_route_notif_update(rn
, re
, DPLANE_OP_ROUTE_INSTALL
, ctx
);
2001 /* Redistribute, lsp, and nht update */
2002 redistribute_update(dest_pfx
, src_pfx
, re
, NULL
);
2004 zebra_rib_evaluate_rn_nexthops(
2005 rn
, zebra_router_get_next_sequence());
2007 zebra_rib_evaluate_mpls(rn
);
2009 } else if (start_count
> 0 && end_count
== 0) {
2011 zlog_debug("%u:%s un-installed transition from dplane notification",
2012 dplane_ctx_get_vrf(ctx
), dest_str
);
2014 /* Transition from _something_ installed to _nothing_
2017 /* We expect this to be the selected route, so we want
2018 * to tell others about this transistion.
2020 UNSET_FLAG(re
->status
, ROUTE_ENTRY_INSTALLED
);
2022 /* Changed nexthops - update kernel/others */
2023 dplane_route_notif_update(rn
, re
, DPLANE_OP_ROUTE_DELETE
, ctx
);
2025 /* Redistribute, lsp, and nht update */
2026 redistribute_delete(dest_pfx
, src_pfx
, re
);
2028 zebra_rib_evaluate_rn_nexthops(
2029 rn
, zebra_router_get_next_sequence());
2031 zebra_rib_evaluate_mpls(rn
);
2036 route_unlock_node(rn
);
2038 /* Return context to dataplane module */
2039 dplane_ctx_fini(&ctx
);
2042 /* Take a list of route_node structs and return 1, if there was a record
2043 * picked from it and processed by rib_process(). Don't process more,
2044 * than one RN record; operate only in the specified sub-queue.
2046 static unsigned int process_subq(struct list
*subq
, uint8_t qindex
)
2048 struct listnode
*lnode
= listhead(subq
);
2049 struct route_node
*rnode
;
2051 struct zebra_vrf
*zvrf
= NULL
;
2056 rnode
= listgetdata(lnode
);
2057 dest
= rib_dest_from_rnode(rnode
);
2059 zvrf
= rib_dest_vrf(dest
);
2063 if (IS_ZEBRA_DEBUG_RIB_DETAILED
) {
2064 char buf
[SRCDEST2STR_BUFFER
];
2066 srcdest_rnode2str(rnode
, buf
, sizeof(buf
));
2067 zlog_debug("%u:%s: rn %p dequeued from sub-queue %u",
2068 zvrf
? zvrf_id(zvrf
) : 0, buf
, rnode
, qindex
);
2072 UNSET_FLAG(rib_dest_from_rnode(rnode
)->flags
,
2073 RIB_ROUTE_QUEUED(qindex
));
2078 zlog_debug ("%s: called for route_node (%p, %d) with no ribs",
2079 __func__
, rnode
, rnode
->lock
);
2080 zlog_backtrace(LOG_DEBUG
);
2083 route_unlock_node(rnode
);
2084 list_delete_node(subq
, lnode
);
2090 * Perform next-hop tracking processing after RIB updates.
2092 static void do_nht_processing(void)
2096 /* Dispatch the meta queue by picking, processing and unlocking the next RN from
2097 * a non-empty sub-queue with lowest priority. wq is equal to zebra->ribq and
2099 * is pointed to the meta queue structure.
2101 static wq_item_status
meta_queue_process(struct work_queue
*dummy
, void *data
)
2103 struct meta_queue
*mq
= data
;
2105 uint32_t queue_len
, queue_limit
;
2107 /* Ensure there's room for more dataplane updates */
2108 queue_limit
= dplane_get_in_queue_limit();
2109 queue_len
= dplane_get_in_queue_len();
2110 if (queue_len
> queue_limit
) {
2111 if (IS_ZEBRA_DEBUG_RIB_DETAILED
)
2112 zlog_debug("rib queue: dplane queue len %u, limit %u, retrying",
2113 queue_len
, queue_limit
);
2115 /* Ensure that the meta-queue is actually enqueued */
2116 if (work_queue_empty(zrouter
.ribq
))
2117 work_queue_add(zrouter
.ribq
, zrouter
.mq
);
2119 return WQ_QUEUE_BLOCKED
;
2122 for (i
= 0; i
< MQ_SIZE
; i
++)
2123 if (process_subq(mq
->subq
[i
], i
)) {
2127 return mq
->size
? WQ_REQUEUE
: WQ_SUCCESS
;
2132 * Look into the RN and queue it into the highest priority queue
2133 * at this point in time for processing.
2135 * We will enqueue a route node only once per invocation.
2137 * There are two possibilities here that should be kept in mind.
2138 * If the original invocation has not been pulled off for processing
2139 * yet, A subsuquent invocation can have a route entry with a better
2140 * meta queue index value and we can have a situation where
2141 * we might have the same node enqueued 2 times. Not necessarily
2142 * an optimal situation but it should be ok.
2144 * The other possibility is that the original invocation has not
2145 * been pulled off for processing yet, A subsusquent invocation
2146 * doesn't have a route_entry with a better meta-queue and the
2147 * original metaqueue index value will win and we'll end up with
2148 * the route node enqueued once.
2150 static void rib_meta_queue_add(struct meta_queue
*mq
, struct route_node
*rn
)
2152 struct route_entry
*re
= NULL
, *curr_re
= NULL
;
2153 uint8_t qindex
= MQ_SIZE
, curr_qindex
= MQ_SIZE
;
2155 RNODE_FOREACH_RE (rn
, curr_re
) {
2156 curr_qindex
= route_info
[curr_re
->type
].meta_q_map
;
2158 if (curr_qindex
<= qindex
) {
2160 qindex
= curr_qindex
;
2167 /* Invariant: at this point we always have rn->info set. */
2168 if (CHECK_FLAG(rib_dest_from_rnode(rn
)->flags
,
2169 RIB_ROUTE_QUEUED(qindex
))) {
2170 if (IS_ZEBRA_DEBUG_RIB_DETAILED
)
2171 rnode_debug(rn
, re
->vrf_id
,
2172 "rn %p is already queued in sub-queue %u",
2173 (void *)rn
, qindex
);
2177 SET_FLAG(rib_dest_from_rnode(rn
)->flags
, RIB_ROUTE_QUEUED(qindex
));
2178 listnode_add(mq
->subq
[qindex
], rn
);
2179 route_lock_node(rn
);
2182 if (IS_ZEBRA_DEBUG_RIB_DETAILED
)
2183 rnode_debug(rn
, re
->vrf_id
, "queued rn %p into sub-queue %u",
2184 (void *)rn
, qindex
);
2187 /* Add route_node to work queue and schedule processing */
2188 void rib_queue_add(struct route_node
*rn
)
2192 /* Pointless to queue a route_node with no RIB entries to add or remove
2194 if (!rnode_to_ribs(rn
)) {
2195 zlog_debug("%s: called for route_node (%p, %d) with no ribs",
2196 __func__
, (void *)rn
, rn
->lock
);
2197 zlog_backtrace(LOG_DEBUG
);
2201 if (zrouter
.ribq
== NULL
) {
2202 flog_err(EC_ZEBRA_WQ_NONEXISTENT
,
2203 "%s: work_queue does not exist!", __func__
);
2208 * The RIB queue should normally be either empty or holding the only
2209 * work_queue_item element. In the latter case this element would
2210 * hold a pointer to the meta queue structure, which must be used to
2211 * actually queue the route nodes to process. So create the MQ
2212 * holder, if necessary, then push the work into it in any case.
2213 * This semantics was introduced after 0.99.9 release.
2215 if (work_queue_empty(zrouter
.ribq
))
2216 work_queue_add(zrouter
.ribq
, zrouter
.mq
);
2218 rib_meta_queue_add(zrouter
.mq
, rn
);
2223 /* Create new meta queue.
2224 A destructor function doesn't seem to be necessary here.
2226 static struct meta_queue
*meta_queue_new(void)
2228 struct meta_queue
*new;
2231 new = XCALLOC(MTYPE_WORK_QUEUE
, sizeof(struct meta_queue
));
2233 for (i
= 0; i
< MQ_SIZE
; i
++) {
2234 new->subq
[i
] = list_new();
2235 assert(new->subq
[i
]);
2241 void meta_queue_free(struct meta_queue
*mq
)
2245 for (i
= 0; i
< MQ_SIZE
; i
++)
2246 list_delete(&mq
->subq
[i
]);
2248 XFREE(MTYPE_WORK_QUEUE
, mq
);
2251 /* initialise zebra rib work queue */
2252 static void rib_queue_init(void)
2254 if (!(zrouter
.ribq
= work_queue_new(zrouter
.master
,
2255 "route_node processing"))) {
2256 flog_err(EC_ZEBRA_WQ_NONEXISTENT
,
2257 "%s: could not initialise work queue!", __func__
);
2261 /* fill in the work queue spec */
2262 zrouter
.ribq
->spec
.workfunc
= &meta_queue_process
;
2263 zrouter
.ribq
->spec
.errorfunc
= NULL
;
2264 zrouter
.ribq
->spec
.completion_func
= NULL
;
2265 /* XXX: TODO: These should be runtime configurable via vty */
2266 zrouter
.ribq
->spec
.max_retries
= 3;
2267 zrouter
.ribq
->spec
.hold
= ZEBRA_RIB_PROCESS_HOLD_TIME
;
2268 zrouter
.ribq
->spec
.retry
= ZEBRA_RIB_PROCESS_RETRY_TIME
;
2270 if (!(zrouter
.mq
= meta_queue_new())) {
2271 flog_err(EC_ZEBRA_WQ_NONEXISTENT
,
2272 "%s: could not initialise meta queue!", __func__
);
2278 rib_dest_t
*zebra_rib_create_dest(struct route_node
*rn
)
2282 dest
= XCALLOC(MTYPE_RIB_DEST
, sizeof(rib_dest_t
));
2283 rnh_list_init(&dest
->nht
);
2284 route_lock_node(rn
); /* rn route table reference */
2291 /* RIB updates are processed via a queue of pointers to route_nodes.
2293 * The queue length is bounded by the maximal size of the routing table,
2294 * as a route_node will not be requeued, if already queued.
2296 * REs are submitted via rib_addnode or rib_delnode which set minimal
2297 * state, or static_install_route (when an existing RE is updated)
2298 * and then submit route_node to queue for best-path selection later.
2299 * Order of add/delete state changes are preserved for any given RE.
2301 * Deleted REs are reaped during best-path selection.
2304 * |-> rib_link or unset ROUTE_ENTRY_REMOVE |->Update kernel with
2305 * |-------->| | best RE, if required
2307 * static_install->|->rib_addqueue...... -> rib_process
2309 * |-------->| |-> rib_unlink
2310 * |-> set ROUTE_ENTRY_REMOVE |
2311 * rib_delnode (RE freed)
2313 * The 'info' pointer of a route_node points to a rib_dest_t
2314 * ('dest'). Queueing state for a route_node is kept on the dest. The
2315 * dest is created on-demand by rib_link() and is kept around at least
2316 * as long as there are ribs hanging off it (@see rib_gc_dest()).
2318 * Refcounting (aka "locking" throughout the GNU Zebra and Quagga code):
2320 * - route_nodes: refcounted by:
2321 * - dest attached to route_node:
2322 * - managed by: rib_link/rib_gc_dest
2323 * - route_node processing queue
2324 * - managed by: rib_addqueue, rib_process.
2328 /* Add RE to head of the route node. */
2329 static void rib_link(struct route_node
*rn
, struct route_entry
*re
, int process
)
2333 const char *rmap_name
;
2337 dest
= rib_dest_from_rnode(rn
);
2339 if (IS_ZEBRA_DEBUG_RIB_DETAILED
)
2340 rnode_debug(rn
, re
->vrf_id
, "rn %p adding dest", rn
);
2342 dest
= zebra_rib_create_dest(rn
);
2345 re_list_add_head(&dest
->routes
, re
);
2347 afi
= (rn
->p
.family
== AF_INET
)
2349 : (rn
->p
.family
== AF_INET6
) ? AFI_IP6
: AFI_MAX
;
2350 if (is_zebra_import_table_enabled(afi
, re
->vrf_id
, re
->table
)) {
2351 struct zebra_vrf
*zvrf
= zebra_vrf_lookup_by_id(re
->vrf_id
);
2353 rmap_name
= zebra_get_import_table_route_map(afi
, re
->table
);
2354 zebra_add_import_table_entry(zvrf
, rn
, re
, rmap_name
);
2359 static void rib_addnode(struct route_node
*rn
,
2360 struct route_entry
*re
, int process
)
2362 /* RE node has been un-removed before route-node is processed.
2363 * route_node must hence already be on the queue for processing..
2365 if (CHECK_FLAG(re
->status
, ROUTE_ENTRY_REMOVED
)) {
2366 if (IS_ZEBRA_DEBUG_RIB
)
2367 rnode_debug(rn
, re
->vrf_id
, "rn %p, un-removed re %p",
2368 (void *)rn
, (void *)re
);
2370 UNSET_FLAG(re
->status
, ROUTE_ENTRY_REMOVED
);
2373 rib_link(rn
, re
, process
);
2379 * Detach a rib structure from a route_node.
2381 * Note that a call to rib_unlink() should be followed by a call to
2382 * rib_gc_dest() at some point. This allows a rib_dest_t that is no
2383 * longer required to be deleted.
2385 void rib_unlink(struct route_node
*rn
, struct route_entry
*re
)
2391 if (IS_ZEBRA_DEBUG_RIB
)
2392 rnode_debug(rn
, re
->vrf_id
, "rn %p, re %p", (void *)rn
,
2395 dest
= rib_dest_from_rnode(rn
);
2397 re_list_del(&dest
->routes
, re
);
2399 if (dest
->selected_fib
== re
)
2400 dest
->selected_fib
= NULL
;
2402 nexthops_free(re
->ng
.nexthop
);
2403 nexthops_free(re
->fib_ng
.nexthop
);
2405 XFREE(MTYPE_RE
, re
);
2408 void rib_delnode(struct route_node
*rn
, struct route_entry
*re
)
2412 if (IS_ZEBRA_DEBUG_RIB
)
2413 rnode_debug(rn
, re
->vrf_id
, "rn %p, re %p, removing",
2414 (void *)rn
, (void *)re
);
2415 SET_FLAG(re
->status
, ROUTE_ENTRY_REMOVED
);
2417 afi
= (rn
->p
.family
== AF_INET
)
2419 : (rn
->p
.family
== AF_INET6
) ? AFI_IP6
: AFI_MAX
;
2420 if (is_zebra_import_table_enabled(afi
, re
->vrf_id
, re
->table
)) {
2421 struct zebra_vrf
*zvrf
= zebra_vrf_lookup_by_id(re
->vrf_id
);
2423 zebra_del_import_table_entry(zvrf
, rn
, re
);
2424 /* Just clean up if non main table */
2425 if (IS_ZEBRA_DEBUG_RIB
) {
2426 char buf
[SRCDEST2STR_BUFFER
];
2427 srcdest_rnode2str(rn
, buf
, sizeof(buf
));
2428 zlog_debug("%u:%s: Freeing route rn %p, re %p (%s)",
2429 re
->vrf_id
, buf
, rn
, re
,
2430 zebra_route_string(re
->type
));
2439 /* This function dumps the contents of a given RE entry into
2440 * standard debug log. Calling function name and IP prefix in
2441 * question are passed as 1st and 2nd arguments.
2444 void _route_entry_dump(const char *func
, union prefixconstptr pp
,
2445 union prefixconstptr src_pp
,
2446 const struct route_entry
*re
)
2448 const struct prefix
*src_p
= src_pp
.p
;
2449 bool is_srcdst
= src_p
&& src_p
->prefixlen
;
2450 char straddr
[PREFIX_STRLEN
];
2451 char srcaddr
[PREFIX_STRLEN
];
2452 char nhname
[PREFIX_STRLEN
];
2453 struct nexthop
*nexthop
;
2455 zlog_debug("%s: dumping RE entry %p for %s%s%s vrf %u", func
,
2456 (const void *)re
, prefix2str(pp
, straddr
, sizeof(straddr
)),
2457 is_srcdst
? " from " : "",
2458 is_srcdst
? prefix2str(src_pp
, srcaddr
, sizeof(srcaddr
))
2461 zlog_debug("%s: uptime == %lu, type == %u, instance == %d, table == %d",
2462 straddr
, (unsigned long)re
->uptime
, re
->type
, re
->instance
,
2465 "%s: metric == %u, mtu == %u, distance == %u, flags == %u, status == %u",
2466 straddr
, re
->metric
, re
->mtu
, re
->distance
, re
->flags
, re
->status
);
2467 zlog_debug("%s: nexthop_num == %u, nexthop_active_num == %u", straddr
,
2468 re
->nexthop_num
, re
->nexthop_active_num
);
2470 for (ALL_NEXTHOPS(re
->ng
, nexthop
)) {
2471 struct interface
*ifp
;
2472 struct vrf
*vrf
= vrf_lookup_by_id(nexthop
->vrf_id
);
2474 switch (nexthop
->type
) {
2475 case NEXTHOP_TYPE_BLACKHOLE
:
2476 sprintf(nhname
, "Blackhole");
2478 case NEXTHOP_TYPE_IFINDEX
:
2479 ifp
= if_lookup_by_index(nexthop
->ifindex
,
2481 sprintf(nhname
, "%s", ifp
? ifp
->name
: "Unknown");
2483 case NEXTHOP_TYPE_IPV4
:
2485 case NEXTHOP_TYPE_IPV4_IFINDEX
:
2486 inet_ntop(AF_INET
, &nexthop
->gate
, nhname
,
2489 case NEXTHOP_TYPE_IPV6
:
2490 case NEXTHOP_TYPE_IPV6_IFINDEX
:
2491 inet_ntop(AF_INET6
, &nexthop
->gate
, nhname
,
2495 zlog_debug("%s: %s %s[%u] vrf %s(%u) with flags %s%s%s%s%s%s",
2496 straddr
, (nexthop
->rparent
? " NH" : "NH"), nhname
,
2497 nexthop
->ifindex
, vrf
? vrf
->name
: "Unknown",
2499 (CHECK_FLAG(nexthop
->flags
, NEXTHOP_FLAG_ACTIVE
)
2502 (CHECK_FLAG(re
->status
, ROUTE_ENTRY_INSTALLED
)
2505 (CHECK_FLAG(nexthop
->flags
, NEXTHOP_FLAG_RECURSIVE
)
2508 (CHECK_FLAG(nexthop
->flags
, NEXTHOP_FLAG_ONLINK
)
2511 (CHECK_FLAG(nexthop
->flags
, NEXTHOP_FLAG_MATCHED
)
2514 (CHECK_FLAG(nexthop
->flags
, NEXTHOP_FLAG_DUPLICATE
)
2518 zlog_debug("%s: dump complete", straddr
);
2521 /* This is an exported helper to rtm_read() to dump the strange
2522 * RE entry found by rib_lookup_ipv4_route()
2525 void rib_lookup_and_dump(struct prefix_ipv4
*p
, vrf_id_t vrf_id
)
2527 struct route_table
*table
;
2528 struct route_node
*rn
;
2529 struct route_entry
*re
;
2530 char prefix_buf
[INET_ADDRSTRLEN
];
2533 table
= zebra_vrf_table(AFI_IP
, SAFI_UNICAST
, vrf_id
);
2535 flog_err(EC_ZEBRA_TABLE_LOOKUP_FAILED
,
2536 "%s:%u zebra_vrf_table() returned NULL", __func__
,
2541 /* Scan the RIB table for exactly matching RE entry. */
2542 rn
= route_node_lookup(table
, (struct prefix
*)p
);
2544 /* No route for this prefix. */
2546 zlog_debug("%s:%u lookup failed for %s", __func__
, vrf_id
,
2547 prefix2str((struct prefix
*)p
, prefix_buf
,
2548 sizeof(prefix_buf
)));
2553 route_unlock_node(rn
);
2556 RNODE_FOREACH_RE (rn
, re
) {
2557 zlog_debug("%s:%u rn %p, re %p: %s, %s",
2559 (void *)rn
, (void *)re
,
2560 (CHECK_FLAG(re
->status
, ROUTE_ENTRY_REMOVED
)
2563 (CHECK_FLAG(re
->flags
, ZEBRA_FLAG_SELECTED
)
2566 route_entry_dump(p
, NULL
, re
);
2570 /* Check if requested address assignment will fail due to another
2571 * route being installed by zebra in FIB already. Take necessary
2572 * actions, if needed: remove such a route from FIB and deSELECT
2573 * corresponding RE entry. Then put affected RN into RIBQ head.
2575 void rib_lookup_and_pushup(struct prefix_ipv4
*p
, vrf_id_t vrf_id
)
2577 struct route_table
*table
;
2578 struct route_node
*rn
;
2581 if (NULL
== (table
= zebra_vrf_table(AFI_IP
, SAFI_UNICAST
, vrf_id
))) {
2582 flog_err(EC_ZEBRA_TABLE_LOOKUP_FAILED
,
2583 "%s:%u zebra_vrf_table() returned NULL", __func__
,
2588 /* No matches would be the simplest case. */
2589 if (NULL
== (rn
= route_node_lookup(table
, (struct prefix
*)p
)))
2593 route_unlock_node(rn
);
2595 dest
= rib_dest_from_rnode(rn
);
2596 /* Check all RE entries. In case any changes have to be done, requeue
2597 * the RN into RIBQ head. If the routing message about the new connected
2598 * route (generated by the IP address we are going to assign very soon)
2599 * comes before the RIBQ is processed, the new RE entry will join
2600 * RIBQ record already on head. This is necessary for proper
2602 * of the rest of the RE.
2604 if (dest
->selected_fib
) {
2605 if (IS_ZEBRA_DEBUG_RIB
) {
2606 char buf
[PREFIX_STRLEN
];
2608 zlog_debug("%u:%s: freeing way for connected prefix",
2609 dest
->selected_fib
->vrf_id
,
2610 prefix2str(&rn
->p
, buf
, sizeof(buf
)));
2611 route_entry_dump(&rn
->p
, NULL
, dest
->selected_fib
);
2613 rib_uninstall(rn
, dest
->selected_fib
);
2618 int rib_add_multipath(afi_t afi
, safi_t safi
, struct prefix
*p
,
2619 struct prefix_ipv6
*src_p
, struct route_entry
*re
)
2621 struct route_table
*table
;
2622 struct route_node
*rn
;
2623 struct route_entry
*same
= NULL
;
2629 assert(!src_p
|| !src_p
->prefixlen
|| afi
== AFI_IP6
);
2632 table
= zebra_vrf_table_with_table_id(afi
, safi
, re
->vrf_id
, re
->table
);
2634 XFREE(MTYPE_RE
, re
);
2638 /* Make it sure prefixlen is applied to the prefix. */
2641 apply_mask_ipv6(src_p
);
2643 /* Set default distance by route type. */
2644 if (re
->distance
== 0)
2645 re
->distance
= route_distance(re
->type
);
2647 /* Lookup route node.*/
2648 rn
= srcdest_rnode_get(table
, p
, src_p
);
2651 * If same type of route are installed, treat it as a implicit
2653 * If the user has specified the No route replace semantics
2654 * for the install don't do a route replace.
2656 RNODE_FOREACH_RE (rn
, same
) {
2657 if (CHECK_FLAG(same
->status
, ROUTE_ENTRY_REMOVED
))
2660 if (same
->type
!= re
->type
)
2662 if (same
->instance
!= re
->instance
)
2664 if (same
->type
== ZEBRA_ROUTE_KERNEL
2665 && same
->metric
!= re
->metric
)
2668 if (CHECK_FLAG(re
->flags
, ZEBRA_FLAG_RR_USE_DISTANCE
) &&
2669 same
->distance
!= re
->distance
)
2673 * We should allow duplicate connected routes
2674 * because of IPv6 link-local routes and unnumbered
2675 * interfaces on Linux.
2677 if (same
->type
!= ZEBRA_ROUTE_CONNECT
)
2681 /* If this route is kernel/connected route, notify the dataplane. */
2682 if (RIB_SYSTEM_ROUTE(re
)) {
2683 /* Notify dataplane */
2684 dplane_sys_route_add(rn
, re
);
2687 /* Link new re to node.*/
2688 if (IS_ZEBRA_DEBUG_RIB
) {
2689 rnode_debug(rn
, re
->vrf_id
,
2690 "Inserting route rn %p, re %p (%s) existing %p",
2691 rn
, re
, zebra_route_string(re
->type
), same
);
2693 if (IS_ZEBRA_DEBUG_RIB_DETAILED
)
2694 route_entry_dump(p
, src_p
, re
);
2697 SET_FLAG(re
->status
, ROUTE_ENTRY_CHANGED
);
2698 rib_addnode(rn
, re
, 1);
2701 /* Free implicit route.*/
2703 rib_delnode(rn
, same
);
2707 route_unlock_node(rn
);
2711 void rib_delete(afi_t afi
, safi_t safi
, vrf_id_t vrf_id
, int type
,
2712 unsigned short instance
, int flags
, struct prefix
*p
,
2713 struct prefix_ipv6
*src_p
, const struct nexthop
*nh
,
2714 uint32_t table_id
, uint32_t metric
, uint8_t distance
,
2717 struct route_table
*table
;
2718 struct route_node
*rn
;
2719 struct route_entry
*re
;
2720 struct route_entry
*fib
= NULL
;
2721 struct route_entry
*same
= NULL
;
2722 struct nexthop
*rtnh
;
2723 char buf2
[INET6_ADDRSTRLEN
];
2726 assert(!src_p
|| !src_p
->prefixlen
|| afi
== AFI_IP6
);
2729 table
= zebra_vrf_table_with_table_id(afi
, safi
, vrf_id
, table_id
);
2736 apply_mask_ipv6(src_p
);
2738 /* Lookup route node. */
2739 rn
= srcdest_rnode_lookup(table
, p
, src_p
);
2741 char dst_buf
[PREFIX_STRLEN
], src_buf
[PREFIX_STRLEN
];
2743 prefix2str(p
, dst_buf
, sizeof(dst_buf
));
2744 if (src_p
&& src_p
->prefixlen
)
2745 prefix2str(src_p
, src_buf
, sizeof(src_buf
));
2749 if (IS_ZEBRA_DEBUG_RIB
) {
2750 struct vrf
*vrf
= vrf_lookup_by_id(vrf_id
);
2752 zlog_debug("%s[%d]:%s%s%s doesn't exist in rib",
2753 vrf
->name
, table_id
, dst_buf
,
2754 (src_buf
[0] != '\0') ? " from " : "",
2760 dest
= rib_dest_from_rnode(rn
);
2761 fib
= dest
->selected_fib
;
2763 /* Lookup same type route. */
2764 RNODE_FOREACH_RE (rn
, re
) {
2765 if (CHECK_FLAG(re
->status
, ROUTE_ENTRY_REMOVED
))
2768 if (re
->type
!= type
)
2770 if (re
->instance
!= instance
)
2772 if (CHECK_FLAG(re
->flags
, ZEBRA_FLAG_RR_USE_DISTANCE
) &&
2773 distance
!= re
->distance
)
2776 if (re
->type
== ZEBRA_ROUTE_KERNEL
&& re
->metric
!= metric
)
2778 if (re
->type
== ZEBRA_ROUTE_CONNECT
&& (rtnh
= re
->ng
.nexthop
)
2779 && rtnh
->type
== NEXTHOP_TYPE_IFINDEX
&& nh
) {
2780 if (rtnh
->ifindex
!= nh
->ifindex
)
2785 /* Make sure that the route found has the same gateway. */
2791 for (ALL_NEXTHOPS(re
->ng
, rtnh
))
2793 * No guarantee all kernel send nh with labels
2796 if (nexthop_same_no_labels(rtnh
, nh
)) {
2804 /* If same type of route can't be found and this message is from
2808 * In the past(HA!) we could get here because
2809 * we were receiving a route delete from the
2810 * kernel and we're not marking the proto
2811 * as coming from it's appropriate originator.
2812 * Now that we are properly noticing the fact
2813 * that the kernel has deleted our route we
2814 * are not going to get called in this path
2815 * I am going to leave this here because
2816 * this might still work this way on non-linux
2817 * platforms as well as some weird state I have
2818 * not properly thought of yet.
2819 * If we can show that this code path is
2820 * dead then we can remove it.
2822 if (fib
&& CHECK_FLAG(flags
, ZEBRA_FLAG_SELFROUTE
)) {
2823 if (IS_ZEBRA_DEBUG_RIB
) {
2824 rnode_debug(rn
, vrf_id
,
2825 "rn %p, re %p (%s) was deleted from kernel, adding",
2827 zebra_route_string(fib
->type
));
2830 UNSET_FLAG(fib
->status
, ROUTE_ENTRY_INSTALLED
);
2832 for (rtnh
= fib
->ng
.nexthop
; rtnh
;
2834 UNSET_FLAG(rtnh
->flags
,
2838 * This is a non FRR route
2839 * as such we should mark
2842 dest
->selected_fib
= NULL
;
2844 /* This means someone else, other than Zebra,
2846 * a Zebra router from the kernel. We will add
2848 rib_install_kernel(rn
, fib
, NULL
);
2851 if (IS_ZEBRA_DEBUG_RIB
) {
2855 "via %s ifindex %d type %d "
2856 "doesn't exist in rib",
2857 inet_ntop(afi2family(afi
),
2864 "type %d doesn't exist in rib",
2867 route_unlock_node(rn
);
2873 if (fromkernel
&& CHECK_FLAG(flags
, ZEBRA_FLAG_SELFROUTE
)
2875 rib_install_kernel(rn
, same
, NULL
);
2876 route_unlock_node(rn
);
2881 /* Special handling for IPv4 or IPv6 routes sourced from
2882 * EVPN - the nexthop (and associated MAC) need to be
2883 * uninstalled if no more refs.
2885 if (CHECK_FLAG(flags
, ZEBRA_FLAG_EVPN_ROUTE
)) {
2886 struct nexthop
*tmp_nh
;
2888 for (ALL_NEXTHOPS(re
->ng
, tmp_nh
)) {
2889 struct ipaddr vtep_ip
;
2891 memset(&vtep_ip
, 0, sizeof(struct ipaddr
));
2892 if (afi
== AFI_IP
) {
2893 vtep_ip
.ipa_type
= IPADDR_V4
;
2894 memcpy(&(vtep_ip
.ipaddr_v4
),
2895 &(tmp_nh
->gate
.ipv4
),
2896 sizeof(struct in_addr
));
2898 vtep_ip
.ipa_type
= IPADDR_V6
;
2899 memcpy(&(vtep_ip
.ipaddr_v6
),
2900 &(tmp_nh
->gate
.ipv6
),
2901 sizeof(struct in6_addr
));
2903 zebra_vxlan_evpn_vrf_route_del(re
->vrf_id
,
2908 /* Notify dplane if system route changes */
2909 if (RIB_SYSTEM_ROUTE(re
))
2910 dplane_sys_route_del(rn
, same
);
2912 rib_delnode(rn
, same
);
2915 route_unlock_node(rn
);
2920 int rib_add(afi_t afi
, safi_t safi
, vrf_id_t vrf_id
, int type
,
2921 unsigned short instance
, int flags
, struct prefix
*p
,
2922 struct prefix_ipv6
*src_p
, const struct nexthop
*nh
,
2923 uint32_t table_id
, uint32_t metric
, uint32_t mtu
, uint8_t distance
,
2926 struct route_entry
*re
;
2927 struct nexthop
*nexthop
;
2929 /* Allocate new route_entry structure. */
2930 re
= XCALLOC(MTYPE_RE
, sizeof(struct route_entry
));
2932 re
->instance
= instance
;
2933 re
->distance
= distance
;
2935 re
->metric
= metric
;
2937 re
->table
= table_id
;
2938 re
->vrf_id
= vrf_id
;
2939 re
->nexthop_num
= 0;
2940 re
->uptime
= monotime(NULL
);
2944 nexthop
= nexthop_new();
2946 route_entry_nexthop_add(re
, nexthop
);
2948 return rib_add_multipath(afi
, safi
, p
, src_p
, re
);
2951 /* Schedule routes of a particular table (address-family) based on event. */
2952 void rib_update_table(struct route_table
*table
, rib_update_event_t event
)
2954 struct route_node
*rn
;
2955 struct route_entry
*re
, *next
;
2957 /* Walk all routes and queue for processing, if appropriate for
2958 * the trigger event.
2960 for (rn
= route_top(table
); rn
; rn
= srcdest_route_next(rn
)) {
2962 * If we are looking at a route node and the node
2963 * has already been queued we don't
2964 * need to queue it up again
2966 if (rn
->info
&& CHECK_FLAG(rib_dest_from_rnode(rn
)->flags
,
2967 RIB_ROUTE_ANY_QUEUED
))
2970 case RIB_UPDATE_IF_CHANGE
:
2971 /* Examine all routes that won't get processed by the
2973 * triggered by nexthop evaluation (NHT). This would be
2975 * kernel and certain static routes. Note that NHT will
2977 * triggered upon an interface event as connected routes
2979 * get queued for processing.
2981 RNODE_FOREACH_RE_SAFE (rn
, re
, next
) {
2984 if (re
->type
!= ZEBRA_ROUTE_SYSTEM
2985 && re
->type
!= ZEBRA_ROUTE_KERNEL
2986 && re
->type
!= ZEBRA_ROUTE_CONNECT
2987 && re
->type
!= ZEBRA_ROUTE_STATIC
)
2990 if (re
->type
!= ZEBRA_ROUTE_STATIC
) {
2991 SET_FLAG(re
->status
,
2992 ROUTE_ENTRY_CHANGED
);
2997 for (nh
= re
->ng
.nexthop
; nh
; nh
= nh
->next
)
2998 if (!(nh
->type
== NEXTHOP_TYPE_IPV4
2999 || nh
->type
== NEXTHOP_TYPE_IPV6
))
3002 /* If we only have nexthops to a
3007 SET_FLAG(re
->status
,
3008 ROUTE_ENTRY_CHANGED
);
3014 case RIB_UPDATE_RMAP_CHANGE
:
3015 case RIB_UPDATE_OTHER
:
3016 /* Right now, examine all routes. Can restrict to a
3018 * some cases (TODO).
3020 if (rnode_to_ribs(rn
)) {
3021 RNODE_FOREACH_RE_SAFE (rn
, re
, next
)
3022 SET_FLAG(re
->status
,
3023 ROUTE_ENTRY_CHANGED
);
3034 /* RIB update function. */
3035 void rib_update(vrf_id_t vrf_id
, rib_update_event_t event
)
3037 struct route_table
*table
;
3039 /* Process routes of interested address-families. */
3040 table
= zebra_vrf_table(AFI_IP
, SAFI_UNICAST
, vrf_id
);
3042 if (IS_ZEBRA_DEBUG_EVENT
)
3043 zlog_debug("%s : AFI_IP event %d", __func__
, event
);
3044 rib_update_table(table
, event
);
3047 table
= zebra_vrf_table(AFI_IP6
, SAFI_UNICAST
, vrf_id
);
3049 if (IS_ZEBRA_DEBUG_EVENT
)
3050 zlog_debug("%s : AFI_IP6 event %d", __func__
, event
);
3051 rib_update_table(table
, event
);
3055 /* Delete self installed routes after zebra is relaunched. */
3056 void rib_sweep_table(struct route_table
*table
)
3058 struct route_node
*rn
;
3059 struct route_entry
*re
;
3060 struct route_entry
*next
;
3061 struct nexthop
*nexthop
;
3066 for (rn
= route_top(table
); rn
; rn
= srcdest_route_next(rn
)) {
3067 RNODE_FOREACH_RE_SAFE (rn
, re
, next
) {
3069 if (IS_ZEBRA_DEBUG_RIB
)
3070 route_entry_dump(&rn
->p
, NULL
, re
);
3072 if (CHECK_FLAG(re
->status
, ROUTE_ENTRY_REMOVED
))
3075 if (!CHECK_FLAG(re
->flags
, ZEBRA_FLAG_SELFROUTE
))
3079 * If routes are older than startup_time then
3080 * we know we read them in from the kernel.
3081 * As such we can safely remove them.
3083 if (zrouter
.startup_time
< re
->uptime
)
3087 * So we are starting up and have received
3088 * routes from the kernel that we have installed
3089 * from a previous run of zebra but not cleaned
3090 * up ( say a kill -9 )
3091 * But since we haven't actually installed
3092 * them yet( we received them from the kernel )
3093 * we don't think they are active.
3094 * So let's pretend they are active to actually
3096 * In all honesty I'm not sure if we should
3097 * mark them as active when we receive them
3098 * This is startup only so probably ok.
3100 * If we ever decide to move rib_sweep_table
3101 * to a different spot (ie startup )
3102 * this decision needs to be revisited
3104 SET_FLAG(re
->status
, ROUTE_ENTRY_INSTALLED
);
3105 for (ALL_NEXTHOPS(re
->ng
, nexthop
))
3106 SET_FLAG(nexthop
->flags
, NEXTHOP_FLAG_FIB
);
3108 rib_uninstall_kernel(rn
, re
);
3109 rib_delnode(rn
, re
);
3114 /* Sweep all RIB tables. */
3115 int rib_sweep_route(struct thread
*t
)
3118 struct zebra_vrf
*zvrf
;
3120 RB_FOREACH (vrf
, vrf_id_head
, &vrfs_by_id
) {
3121 if ((zvrf
= vrf
->info
) == NULL
)
3124 rib_sweep_table(zvrf
->table
[AFI_IP
][SAFI_UNICAST
]);
3125 rib_sweep_table(zvrf
->table
[AFI_IP6
][SAFI_UNICAST
]);
3128 zebra_router_sweep_route();
3133 /* Remove specific by protocol routes from 'table'. */
3134 unsigned long rib_score_proto_table(uint8_t proto
, unsigned short instance
,
3135 struct route_table
*table
)
3137 struct route_node
*rn
;
3138 struct route_entry
*re
;
3139 struct route_entry
*next
;
3140 unsigned long n
= 0;
3143 for (rn
= route_top(table
); rn
; rn
= srcdest_route_next(rn
))
3144 RNODE_FOREACH_RE_SAFE (rn
, re
, next
) {
3145 if (CHECK_FLAG(re
->status
, ROUTE_ENTRY_REMOVED
))
3147 if (re
->type
== proto
3148 && re
->instance
== instance
) {
3149 rib_delnode(rn
, re
);
3156 /* Remove specific by protocol routes. */
3157 unsigned long rib_score_proto(uint8_t proto
, unsigned short instance
)
3160 struct zebra_vrf
*zvrf
;
3161 struct other_route_table
*ort
;
3162 unsigned long cnt
= 0;
3164 RB_FOREACH (vrf
, vrf_id_head
, &vrfs_by_id
) {
3169 cnt
+= rib_score_proto_table(proto
, instance
,
3170 zvrf
->table
[AFI_IP
][SAFI_UNICAST
])
3171 + rib_score_proto_table(
3173 zvrf
->table
[AFI_IP6
][SAFI_UNICAST
]);
3175 frr_each(otable
, &zvrf
->other_tables
, ort
) cnt
+=
3176 rib_score_proto_table(proto
, instance
, ort
->table
);
3182 /* Close RIB and clean up kernel routes. */
3183 void rib_close_table(struct route_table
*table
)
3185 struct route_node
*rn
;
3186 rib_table_info_t
*info
;
3192 info
= route_table_get_info(table
);
3194 for (rn
= route_top(table
); rn
; rn
= srcdest_route_next(rn
)) {
3195 dest
= rib_dest_from_rnode(rn
);
3197 if (dest
&& dest
->selected_fib
) {
3198 if (info
->safi
== SAFI_UNICAST
)
3199 hook_call(rib_update
, rn
, NULL
);
3201 rib_uninstall_kernel(rn
, dest
->selected_fib
);
3202 dest
->selected_fib
= NULL
;
3208 * Handler for async dataplane results after a pseudowire installation
3210 static int handle_pw_result(struct zebra_dplane_ctx
*ctx
)
3212 struct zebra_pw
*pw
;
3213 struct zebra_vrf
*vrf
;
3215 /* The pseudowire code assumes success - we act on an error
3216 * result for installation attempts here.
3218 if (dplane_ctx_get_op(ctx
) != DPLANE_OP_PW_INSTALL
)
3221 if (dplane_ctx_get_status(ctx
) != ZEBRA_DPLANE_REQUEST_SUCCESS
) {
3222 vrf
= zebra_vrf_lookup_by_id(dplane_ctx_get_vrf(ctx
));
3223 pw
= zebra_pw_find(vrf
, dplane_ctx_get_pw_ifname(ctx
));
3225 zebra_pw_install_failure(pw
);
3235 * Handle results from the dataplane system. Dequeue update context
3236 * structs, dispatch to appropriate internal handlers.
3238 static int rib_process_dplane_results(struct thread
*thread
)
3240 struct zebra_dplane_ctx
*ctx
;
3241 struct dplane_ctx_q ctxlist
;
3243 /* Dequeue a list of completed updates with one lock/unlock cycle */
3246 TAILQ_INIT(&ctxlist
);
3248 /* Take lock controlling queue of results */
3249 pthread_mutex_lock(&dplane_mutex
);
3251 /* Dequeue list of context structs */
3252 dplane_ctx_list_append(&ctxlist
, &rib_dplane_q
);
3254 pthread_mutex_unlock(&dplane_mutex
);
3256 /* Dequeue context block */
3257 ctx
= dplane_ctx_dequeue(&ctxlist
);
3259 /* If we've emptied the results queue, we're done */
3264 switch (dplane_ctx_get_op(ctx
)) {
3265 case DPLANE_OP_ROUTE_INSTALL
:
3266 case DPLANE_OP_ROUTE_UPDATE
:
3267 case DPLANE_OP_ROUTE_DELETE
:
3269 /* Bit of special case for route updates
3270 * that were generated by async notifications:
3271 * we don't want to continue processing these
3274 if (dplane_ctx_get_notif_provider(ctx
) == 0)
3275 rib_process_result(ctx
);
3277 dplane_ctx_fini(&ctx
);
3281 case DPLANE_OP_ROUTE_NOTIFY
:
3282 rib_process_dplane_notify(ctx
);
3285 case DPLANE_OP_LSP_INSTALL
:
3286 case DPLANE_OP_LSP_UPDATE
:
3287 case DPLANE_OP_LSP_DELETE
:
3289 /* Bit of special case for LSP updates
3290 * that were generated by async notifications:
3291 * we don't want to continue processing these.
3293 if (dplane_ctx_get_notif_provider(ctx
) == 0)
3294 zebra_mpls_lsp_dplane_result(ctx
);
3296 dplane_ctx_fini(&ctx
);
3300 case DPLANE_OP_LSP_NOTIFY
:
3301 zebra_mpls_process_dplane_notify(ctx
);
3304 case DPLANE_OP_PW_INSTALL
:
3305 case DPLANE_OP_PW_UNINSTALL
:
3306 handle_pw_result(ctx
);
3309 case DPLANE_OP_SYS_ROUTE_ADD
:
3310 case DPLANE_OP_SYS_ROUTE_DELETE
:
3311 /* No further processing in zebra for these. */
3312 dplane_ctx_fini(&ctx
);
3316 /* Don't expect this: just return the struct? */
3317 dplane_ctx_fini(&ctx
);
3319 } /* Dispatch by op code */
3321 ctx
= dplane_ctx_dequeue(&ctxlist
);
3326 /* Check for nexthop tracking processing after finishing with results */
3327 do_nht_processing();
3333 * Results are returned from the dataplane subsystem, in the context of
3334 * the dataplane pthread. We enqueue the results here for processing by
3335 * the main thread later.
3337 static int rib_dplane_results(struct dplane_ctx_q
*ctxlist
)
3339 /* Take lock controlling queue of results */
3340 pthread_mutex_lock(&dplane_mutex
);
3342 /* Enqueue context blocks */
3343 dplane_ctx_list_append(&rib_dplane_q
, ctxlist
);
3345 pthread_mutex_unlock(&dplane_mutex
);
3347 /* Ensure event is signalled to zebra main pthread */
3348 thread_add_event(zrouter
.master
, rib_process_dplane_results
, NULL
, 0,
3355 * Ensure there are no empty slots in the route_info array.
3356 * Every route type in zebra should be present there.
3358 static void check_route_info(void)
3360 int len
= array_size(route_info
);
3363 * ZEBRA_ROUTE_SYSTEM is special cased since
3364 * its key is 0 anyway.
3366 * ZEBRA_ROUTE_ALL is also ignored.
3368 for (int i
= 0; i
< len
; i
++) {
3369 if (i
== ZEBRA_ROUTE_SYSTEM
|| i
== ZEBRA_ROUTE_ALL
)
3371 assert(route_info
[i
].key
);
3372 assert(route_info
[i
].meta_q_map
< MQ_SIZE
);
3376 /* Routing information base initialize. */
3383 /* Init dataplane, and register for results */
3384 pthread_mutex_init(&dplane_mutex
, NULL
);
3385 TAILQ_INIT(&rib_dplane_q
);
3386 zebra_dplane_init(rib_dplane_results
);
3392 * Get the first vrf id that is greater than the given vrf id if any.
3394 * Returns true if a vrf id was found, false otherwise.
3396 static inline int vrf_id_get_next(vrf_id_t vrf_id
, vrf_id_t
*next_id_p
)
3400 vrf
= vrf_lookup_by_id(vrf_id
);
3402 vrf
= RB_NEXT(vrf_id_head
, vrf
);
3404 *next_id_p
= vrf
->vrf_id
;
3413 * rib_tables_iter_next
3415 * Returns the next table in the iteration.
3417 struct route_table
*rib_tables_iter_next(rib_tables_iter_t
*iter
)
3419 struct route_table
*table
;
3422 * Array that helps us go over all AFI/SAFI combinations via one
3429 {AFI_IP
, SAFI_UNICAST
}, {AFI_IP
, SAFI_MULTICAST
},
3430 {AFI_IP
, SAFI_LABELED_UNICAST
}, {AFI_IP6
, SAFI_UNICAST
},
3431 {AFI_IP6
, SAFI_MULTICAST
}, {AFI_IP6
, SAFI_LABELED_UNICAST
},
3436 switch (iter
->state
) {
3438 case RIB_TABLES_ITER_S_INIT
:
3439 iter
->vrf_id
= VRF_DEFAULT
;
3440 iter
->afi_safi_ix
= -1;
3444 case RIB_TABLES_ITER_S_ITERATING
:
3445 iter
->afi_safi_ix
++;
3448 while (iter
->afi_safi_ix
3449 < (int)array_size(afi_safis
)) {
3450 table
= zebra_vrf_table(
3451 afi_safis
[iter
->afi_safi_ix
].afi
,
3452 afi_safis
[iter
->afi_safi_ix
].safi
,
3457 iter
->afi_safi_ix
++;
3461 * Found another table in this vrf.
3467 * Done with all tables in the current vrf, go to the
3471 if (!vrf_id_get_next(iter
->vrf_id
, &iter
->vrf_id
))
3474 iter
->afi_safi_ix
= 0;
3479 case RIB_TABLES_ITER_S_DONE
:
3484 iter
->state
= RIB_TABLES_ITER_S_ITERATING
;
3486 iter
->state
= RIB_TABLES_ITER_S_DONE
;