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"
59 #include "zebra/zebra_nhg.h"
62 * Event, list, and mutex for delivery of dataplane results
64 static pthread_mutex_t dplane_mutex
;
65 static struct thread
*t_dplane
;
66 static struct dplane_ctx_q rib_dplane_q
;
68 DEFINE_HOOK(rib_update
, (struct route_node
* rn
, const char *reason
),
71 /* Should we allow non Quagga processes to delete our routes */
72 extern int allow_delete
;
74 /* Each route type's string and default distance value. */
79 } route_info
[ZEBRA_ROUTE_MAX
] = {
80 [ZEBRA_ROUTE_SYSTEM
] = {ZEBRA_ROUTE_SYSTEM
, 0, 4},
81 [ZEBRA_ROUTE_KERNEL
] = {ZEBRA_ROUTE_KERNEL
, 0, 0},
82 [ZEBRA_ROUTE_CONNECT
] = {ZEBRA_ROUTE_CONNECT
, 0, 0},
83 [ZEBRA_ROUTE_STATIC
] = {ZEBRA_ROUTE_STATIC
, 1, 1},
84 [ZEBRA_ROUTE_RIP
] = {ZEBRA_ROUTE_RIP
, 120, 2},
85 [ZEBRA_ROUTE_RIPNG
] = {ZEBRA_ROUTE_RIPNG
, 120, 2},
86 [ZEBRA_ROUTE_OSPF
] = {ZEBRA_ROUTE_OSPF
, 110, 2},
87 [ZEBRA_ROUTE_OSPF6
] = {ZEBRA_ROUTE_OSPF6
, 110, 2},
88 [ZEBRA_ROUTE_ISIS
] = {ZEBRA_ROUTE_ISIS
, 115, 2},
89 [ZEBRA_ROUTE_BGP
] = {ZEBRA_ROUTE_BGP
, 20 /* IBGP is 200. */, 3},
90 [ZEBRA_ROUTE_PIM
] = {ZEBRA_ROUTE_PIM
, 255, 4},
91 [ZEBRA_ROUTE_EIGRP
] = {ZEBRA_ROUTE_EIGRP
, 90, 2},
92 [ZEBRA_ROUTE_NHRP
] = {ZEBRA_ROUTE_NHRP
, 10, 2},
93 [ZEBRA_ROUTE_HSLS
] = {ZEBRA_ROUTE_HSLS
, 255, 4},
94 [ZEBRA_ROUTE_OLSR
] = {ZEBRA_ROUTE_OLSR
, 255, 4},
95 [ZEBRA_ROUTE_TABLE
] = {ZEBRA_ROUTE_TABLE
, 150, 1},
96 [ZEBRA_ROUTE_LDP
] = {ZEBRA_ROUTE_LDP
, 150, 4},
97 [ZEBRA_ROUTE_VNC
] = {ZEBRA_ROUTE_VNC
, 20, 3},
98 [ZEBRA_ROUTE_VNC_DIRECT
] = {ZEBRA_ROUTE_VNC_DIRECT
, 20, 3},
99 [ZEBRA_ROUTE_VNC_DIRECT_RH
] = {ZEBRA_ROUTE_VNC_DIRECT_RH
, 20, 3},
100 [ZEBRA_ROUTE_BGP_DIRECT
] = {ZEBRA_ROUTE_BGP_DIRECT
, 20, 3},
101 [ZEBRA_ROUTE_BGP_DIRECT_EXT
] = {ZEBRA_ROUTE_BGP_DIRECT_EXT
, 20, 3},
102 [ZEBRA_ROUTE_BABEL
] = {ZEBRA_ROUTE_BABEL
, 100, 2},
103 [ZEBRA_ROUTE_SHARP
] = {ZEBRA_ROUTE_SHARP
, 150, 4},
104 [ZEBRA_ROUTE_PBR
] = {ZEBRA_ROUTE_PBR
, 200, 4},
105 [ZEBRA_ROUTE_BFD
] = {ZEBRA_ROUTE_BFD
, 255, 4},
106 [ZEBRA_ROUTE_OPENFABRIC
] = {ZEBRA_ROUTE_OPENFABRIC
, 115, 2},
107 [ZEBRA_ROUTE_VRRP
] = {ZEBRA_ROUTE_VRRP
, 255, 4}
108 /* Any new route type added to zebra, should be mirrored here */
110 /* no entry/default: 150 */
113 static void __attribute__((format(printf
, 5, 6)))
114 _rnode_zlog(const char *_func
, vrf_id_t vrf_id
, struct route_node
*rn
,
115 int priority
, const char *msgfmt
, ...)
117 char buf
[SRCDEST2STR_BUFFER
+ sizeof(" (MRIB)")];
121 va_start(ap
, msgfmt
);
122 vsnprintf(msgbuf
, sizeof(msgbuf
), msgfmt
, ap
);
126 rib_table_info_t
*info
= srcdest_rnode_table_info(rn
);
127 srcdest_rnode2str(rn
, buf
, sizeof(buf
));
129 if (info
->safi
== SAFI_MULTICAST
)
130 strlcat(buf
, " (MRIB)", sizeof(buf
));
132 snprintf(buf
, sizeof(buf
), "{(route_node *) NULL}");
135 zlog(priority
, "%s: %d:%s: %s", _func
, vrf_id
, buf
, msgbuf
);
138 #define rnode_debug(node, vrf_id, ...) \
139 _rnode_zlog(__func__, vrf_id, node, LOG_DEBUG, __VA_ARGS__)
140 #define rnode_info(node, ...) \
141 _rnode_zlog(__func__, vrf_id, node, LOG_INFO, __VA_ARGS__)
143 uint8_t route_distance(int type
)
147 if ((unsigned)type
>= array_size(route_info
))
150 distance
= route_info
[type
].distance
;
155 int is_zebra_valid_kernel_table(uint32_t table_id
)
158 if ((table_id
== RT_TABLE_UNSPEC
) || (table_id
== RT_TABLE_LOCAL
)
159 || (table_id
== RT_TABLE_COMPAT
))
166 int is_zebra_main_routing_table(uint32_t table_id
)
168 if (table_id
== RT_TABLE_MAIN
)
173 int zebra_check_addr(const struct prefix
*p
)
175 if (p
->family
== AF_INET
) {
178 addr
= p
->u
.prefix4
.s_addr
;
181 if (IPV4_NET127(addr
) || IN_CLASSD(addr
)
182 || IPV4_LINKLOCAL(addr
))
185 if (p
->family
== AF_INET6
) {
186 if (IN6_IS_ADDR_LOOPBACK(&p
->u
.prefix6
))
188 if (IN6_IS_ADDR_LINKLOCAL(&p
->u
.prefix6
))
194 /* Add nexthop to the end of a rib node's nexthop list */
195 void route_entry_nexthop_add(struct route_entry
*re
, struct nexthop
*nexthop
)
197 _nexthop_group_add_sorted(&re
->ng
, nexthop
);
203 * copy_nexthop - copy a nexthop to the rib structure.
205 void route_entry_copy_nexthops(struct route_entry
*re
, struct nexthop
*nh
)
207 assert(!re
->ng
.nexthop
);
208 copy_nexthops(&re
->ng
.nexthop
, nh
, NULL
);
209 for (struct nexthop
*nexthop
= nh
; nexthop
; nexthop
= nexthop
->next
)
213 /* Delete specified nexthop from the list. */
214 void route_entry_nexthop_delete(struct route_entry
*re
, struct nexthop
*nexthop
)
217 nexthop
->next
->prev
= nexthop
->prev
;
219 nexthop
->prev
->next
= nexthop
->next
;
221 re
->ng
.nexthop
= nexthop
->next
;
226 struct nexthop
*route_entry_nexthop_ifindex_add(struct route_entry
*re
,
230 struct nexthop
*nexthop
;
232 nexthop
= nexthop_new();
233 nexthop
->type
= NEXTHOP_TYPE_IFINDEX
;
234 nexthop
->ifindex
= ifindex
;
235 nexthop
->vrf_id
= nh_vrf_id
;
237 route_entry_nexthop_add(re
, nexthop
);
242 struct nexthop
*route_entry_nexthop_ipv4_add(struct route_entry
*re
,
243 struct in_addr
*ipv4
,
247 struct nexthop
*nexthop
;
249 nexthop
= nexthop_new();
250 nexthop
->type
= NEXTHOP_TYPE_IPV4
;
251 nexthop
->vrf_id
= nh_vrf_id
;
252 nexthop
->gate
.ipv4
= *ipv4
;
254 nexthop
->src
.ipv4
= *src
;
256 route_entry_nexthop_add(re
, nexthop
);
261 struct nexthop
*route_entry_nexthop_ipv4_ifindex_add(struct route_entry
*re
,
262 struct in_addr
*ipv4
,
267 struct nexthop
*nexthop
;
268 struct interface
*ifp
;
270 nexthop
= nexthop_new();
271 nexthop
->vrf_id
= nh_vrf_id
;
272 nexthop
->type
= NEXTHOP_TYPE_IPV4_IFINDEX
;
273 nexthop
->gate
.ipv4
= *ipv4
;
275 nexthop
->src
.ipv4
= *src
;
276 nexthop
->ifindex
= ifindex
;
277 ifp
= if_lookup_by_index(nexthop
->ifindex
, nh_vrf_id
);
278 /*Pending: need to think if null ifp here is ok during bootup?
279 There was a crash because ifp here was coming to be NULL */
281 if (connected_is_unnumbered(ifp
))
282 SET_FLAG(nexthop
->flags
, NEXTHOP_FLAG_ONLINK
);
284 route_entry_nexthop_add(re
, nexthop
);
289 struct nexthop
*route_entry_nexthop_ipv6_add(struct route_entry
*re
,
290 struct in6_addr
*ipv6
,
293 struct nexthop
*nexthop
;
295 nexthop
= nexthop_new();
296 nexthop
->vrf_id
= nh_vrf_id
;
297 nexthop
->type
= NEXTHOP_TYPE_IPV6
;
298 nexthop
->gate
.ipv6
= *ipv6
;
300 route_entry_nexthop_add(re
, nexthop
);
305 struct nexthop
*route_entry_nexthop_ipv6_ifindex_add(struct route_entry
*re
,
306 struct in6_addr
*ipv6
,
310 struct nexthop
*nexthop
;
312 nexthop
= nexthop_new();
313 nexthop
->vrf_id
= nh_vrf_id
;
314 nexthop
->type
= NEXTHOP_TYPE_IPV6_IFINDEX
;
315 nexthop
->gate
.ipv6
= *ipv6
;
316 nexthop
->ifindex
= ifindex
;
318 route_entry_nexthop_add(re
, nexthop
);
323 struct nexthop
*route_entry_nexthop_blackhole_add(struct route_entry
*re
,
324 enum blackhole_type bh_type
)
326 struct nexthop
*nexthop
;
328 nexthop
= nexthop_new();
329 nexthop
->vrf_id
= VRF_DEFAULT
;
330 nexthop
->type
= NEXTHOP_TYPE_BLACKHOLE
;
331 nexthop
->bh_type
= bh_type
;
333 route_entry_nexthop_add(re
, nexthop
);
338 struct route_entry
*rib_match(afi_t afi
, safi_t safi
, vrf_id_t vrf_id
,
339 union g_addr
*addr
, struct route_node
**rn_out
)
342 struct route_table
*table
;
343 struct route_node
*rn
;
344 struct route_entry
*match
= NULL
;
347 table
= zebra_vrf_table(afi
, safi
, vrf_id
);
351 memset(&p
, 0, sizeof(struct prefix
));
354 p
.u
.prefix4
= addr
->ipv4
;
355 p
.prefixlen
= IPV4_MAX_PREFIXLEN
;
357 p
.u
.prefix6
= addr
->ipv6
;
358 p
.prefixlen
= IPV6_MAX_PREFIXLEN
;
361 rn
= route_node_match(table
, (struct prefix
*)&p
);
366 route_unlock_node(rn
);
368 dest
= rib_dest_from_rnode(rn
);
369 if (dest
&& dest
->selected_fib
370 && !CHECK_FLAG(dest
->selected_fib
->status
,
371 ROUTE_ENTRY_REMOVED
))
372 match
= dest
->selected_fib
;
374 /* If there is no selected route or matched route is EGP, go up
379 } while (rn
&& rn
->info
== NULL
);
383 if (match
->type
!= ZEBRA_ROUTE_CONNECT
) {
384 if (!CHECK_FLAG(match
->status
,
385 ROUTE_ENTRY_INSTALLED
))
397 struct route_entry
*rib_match_ipv4_multicast(vrf_id_t vrf_id
,
399 struct route_node
**rn_out
)
401 struct route_entry
*re
= NULL
, *mre
= NULL
, *ure
= NULL
;
402 struct route_node
*m_rn
= NULL
, *u_rn
= NULL
;
403 union g_addr gaddr
= {.ipv4
= addr
};
405 switch (zrouter
.ipv4_multicast_mode
) {
406 case MCAST_MRIB_ONLY
:
407 return rib_match(AFI_IP
, SAFI_MULTICAST
, vrf_id
, &gaddr
,
409 case MCAST_URIB_ONLY
:
410 return rib_match(AFI_IP
, SAFI_UNICAST
, vrf_id
, &gaddr
, rn_out
);
411 case MCAST_NO_CONFIG
:
412 case MCAST_MIX_MRIB_FIRST
:
413 re
= mre
= rib_match(AFI_IP
, SAFI_MULTICAST
, vrf_id
, &gaddr
,
416 re
= ure
= rib_match(AFI_IP
, SAFI_UNICAST
, vrf_id
,
419 case MCAST_MIX_DISTANCE
:
420 mre
= rib_match(AFI_IP
, SAFI_MULTICAST
, vrf_id
, &gaddr
, &m_rn
);
421 ure
= rib_match(AFI_IP
, SAFI_UNICAST
, vrf_id
, &gaddr
, &u_rn
);
423 re
= ure
->distance
< mre
->distance
? ure
: mre
;
429 case MCAST_MIX_PFXLEN
:
430 mre
= rib_match(AFI_IP
, SAFI_MULTICAST
, vrf_id
, &gaddr
, &m_rn
);
431 ure
= rib_match(AFI_IP
, SAFI_UNICAST
, vrf_id
, &gaddr
, &u_rn
);
433 re
= u_rn
->p
.prefixlen
> m_rn
->p
.prefixlen
? ure
: mre
;
442 *rn_out
= (re
== mre
) ? m_rn
: u_rn
;
444 if (IS_ZEBRA_DEBUG_RIB
) {
446 inet_ntop(AF_INET
, &addr
, buf
, BUFSIZ
);
448 zlog_debug("%s: %s: vrf: %u found %s, using %s",
449 __func__
, buf
, vrf_id
,
450 mre
? (ure
? "MRIB+URIB" : "MRIB")
451 : ure
? "URIB" : "nothing",
452 re
== ure
? "URIB" : re
== mre
? "MRIB" : "none");
457 struct route_entry
*rib_lookup_ipv4(struct prefix_ipv4
*p
, vrf_id_t vrf_id
)
459 struct route_table
*table
;
460 struct route_node
*rn
;
461 struct route_entry
*match
= NULL
;
465 table
= zebra_vrf_table(AFI_IP
, SAFI_UNICAST
, vrf_id
);
469 rn
= route_node_lookup(table
, (struct prefix
*)p
);
471 /* No route for this prefix. */
476 route_unlock_node(rn
);
477 dest
= rib_dest_from_rnode(rn
);
479 if (dest
&& dest
->selected_fib
480 && !CHECK_FLAG(dest
->selected_fib
->status
, ROUTE_ENTRY_REMOVED
))
481 match
= dest
->selected_fib
;
486 if (match
->type
== ZEBRA_ROUTE_CONNECT
)
489 if (CHECK_FLAG(match
->status
, ROUTE_ENTRY_INSTALLED
))
496 * Is this RIB labeled-unicast? It must be of type BGP and all paths
497 * (nexthops) must have a label.
499 int zebra_rib_labeled_unicast(struct route_entry
*re
)
501 struct nexthop
*nexthop
= NULL
;
503 if (re
->type
!= ZEBRA_ROUTE_BGP
)
506 for (ALL_NEXTHOPS(re
->ng
, nexthop
))
507 if (!nexthop
->nh_label
|| !nexthop
->nh_label
->num_labels
)
513 /* Update flag indicates whether this is a "replace" or not. Currently, this
514 * is only used for IPv4.
516 void rib_install_kernel(struct route_node
*rn
, struct route_entry
*re
,
517 struct route_entry
*old
)
519 struct nexthop
*nexthop
;
520 rib_table_info_t
*info
= srcdest_rnode_table_info(rn
);
521 struct zebra_vrf
*zvrf
= vrf_info_lookup(re
->vrf_id
);
522 const struct prefix
*p
, *src_p
;
523 enum zebra_dplane_result ret
;
525 rib_dest_t
*dest
= rib_dest_from_rnode(rn
);
527 srcdest_rnode_prefixes(rn
, &p
, &src_p
);
529 if (info
->safi
!= SAFI_UNICAST
) {
530 for (ALL_NEXTHOPS(re
->ng
, nexthop
))
531 SET_FLAG(nexthop
->flags
, NEXTHOP_FLAG_FIB
);
534 struct nexthop
*prev
;
536 for (ALL_NEXTHOPS(re
->ng
, nexthop
)) {
537 UNSET_FLAG (nexthop
->flags
, NEXTHOP_FLAG_DUPLICATE
);
538 for (ALL_NEXTHOPS(re
->ng
, prev
)) {
541 if (nexthop_same_firsthop(nexthop
, prev
)) {
542 SET_FLAG(nexthop
->flags
,
543 NEXTHOP_FLAG_DUPLICATE
);
551 * If this is a replace to a new RE let the originator of the RE
552 * know that they've lost
554 if (old
&& (old
!= re
) && (old
->type
!= re
->type
))
555 zsend_route_notify_owner(old
, p
, ZAPI_ROUTE_BETTER_ADMIN_WON
);
557 /* Update fib selection */
558 dest
->selected_fib
= re
;
561 * Make sure we update the FPM any time we send new information to
564 hook_call(rib_update
, rn
, "installing in kernel");
566 /* Send add or update */
568 ret
= dplane_route_update(rn
, re
, old
);
570 ret
= dplane_route_add(rn
, re
);
573 case ZEBRA_DPLANE_REQUEST_QUEUED
:
574 SET_FLAG(re
->status
, ROUTE_ENTRY_QUEUED
);
577 SET_FLAG(old
->status
, ROUTE_ENTRY_QUEUED
);
579 /* Free old FIB nexthop group */
580 if (old
->fib_ng
.nexthop
) {
581 nexthops_free(old
->fib_ng
.nexthop
);
582 old
->fib_ng
.nexthop
= NULL
;
585 if (!RIB_SYSTEM_ROUTE(old
)) {
586 /* Clear old route's FIB flags */
587 for (ALL_NEXTHOPS(old
->ng
, nexthop
)) {
588 UNSET_FLAG(nexthop
->flags
,
595 zvrf
->installs_queued
++;
597 case ZEBRA_DPLANE_REQUEST_FAILURE
:
599 char str
[SRCDEST2STR_BUFFER
];
601 srcdest_rnode2str(rn
, str
, sizeof(str
));
602 flog_err(EC_ZEBRA_DP_INSTALL_FAIL
,
603 "%u:%s: Failed to enqueue dataplane install",
607 case ZEBRA_DPLANE_REQUEST_SUCCESS
:
616 /* Uninstall the route from kernel. */
617 void rib_uninstall_kernel(struct route_node
*rn
, struct route_entry
*re
)
619 struct nexthop
*nexthop
;
620 rib_table_info_t
*info
= srcdest_rnode_table_info(rn
);
621 struct zebra_vrf
*zvrf
= vrf_info_lookup(re
->vrf_id
);
623 if (info
->safi
!= SAFI_UNICAST
) {
624 UNSET_FLAG(re
->status
, ROUTE_ENTRY_INSTALLED
);
625 for (ALL_NEXTHOPS(re
->ng
, nexthop
))
626 UNSET_FLAG(nexthop
->flags
, NEXTHOP_FLAG_FIB
);
631 * Make sure we update the FPM any time we send new information to
634 hook_call(rib_update
, rn
, "uninstalling from kernel");
636 switch (dplane_route_delete(rn
, re
)) {
637 case ZEBRA_DPLANE_REQUEST_QUEUED
:
639 zvrf
->removals_queued
++;
641 case ZEBRA_DPLANE_REQUEST_FAILURE
:
643 char str
[SRCDEST2STR_BUFFER
];
645 srcdest_rnode2str(rn
, str
, sizeof(str
));
646 flog_err(EC_ZEBRA_DP_INSTALL_FAIL
,
647 "%u:%s: Failed to enqueue dataplane uninstall",
651 case ZEBRA_DPLANE_REQUEST_SUCCESS
:
660 /* Uninstall the route from kernel. */
661 static void rib_uninstall(struct route_node
*rn
, struct route_entry
*re
)
663 rib_table_info_t
*info
= srcdest_rnode_table_info(rn
);
664 rib_dest_t
*dest
= rib_dest_from_rnode(rn
);
665 struct nexthop
*nexthop
;
667 if (dest
&& dest
->selected_fib
== re
) {
668 if (info
->safi
== SAFI_UNICAST
)
669 hook_call(rib_update
, rn
, "rib_uninstall");
671 /* If labeled-unicast route, uninstall transit LSP. */
672 if (zebra_rib_labeled_unicast(re
))
673 zebra_mpls_lsp_uninstall(info
->zvrf
, rn
, re
);
675 rib_uninstall_kernel(rn
, re
);
677 dest
->selected_fib
= NULL
;
679 /* Free FIB nexthop group, if present */
680 if (re
->fib_ng
.nexthop
) {
681 nexthops_free(re
->fib_ng
.nexthop
);
682 re
->fib_ng
.nexthop
= NULL
;
685 for (ALL_NEXTHOPS(re
->ng
, nexthop
))
686 UNSET_FLAG(nexthop
->flags
, NEXTHOP_FLAG_FIB
);
689 if (CHECK_FLAG(re
->flags
, ZEBRA_FLAG_SELECTED
)) {
690 const struct prefix
*p
, *src_p
;
692 srcdest_rnode_prefixes(rn
, &p
, &src_p
);
694 redistribute_delete(p
, src_p
, re
);
695 UNSET_FLAG(re
->flags
, ZEBRA_FLAG_SELECTED
);
700 * rib_can_delete_dest
702 * Returns true if the given dest can be deleted from the table.
704 static int rib_can_delete_dest(rib_dest_t
*dest
)
706 if (re_list_first(&dest
->routes
)) {
711 * Unresolved rnh's are stored on the default route's list
713 * dest->rnode can also be the source prefix node in an
714 * ipv6 sourcedest table. Fortunately the prefix of a
715 * source prefix node can never be the default prefix.
717 if (is_default_prefix(&dest
->rnode
->p
))
721 * Don't delete the dest if we have to update the FPM about this
724 if (CHECK_FLAG(dest
->flags
, RIB_DEST_UPDATE_FPM
)
725 || CHECK_FLAG(dest
->flags
, RIB_DEST_SENT_TO_FPM
))
731 void zebra_rib_evaluate_rn_nexthops(struct route_node
*rn
, uint32_t seq
)
733 rib_dest_t
*dest
= rib_dest_from_rnode(rn
);
737 * We are storing the rnh's associated withb
738 * the tracked nexthop as a list of the rn's.
739 * Unresolved rnh's are placed at the top
740 * of the tree list.( 0.0.0.0/0 for v4 and 0::0/0 for v6 )
741 * As such for each rn we need to walk up the tree
742 * and see if any rnh's need to see if they
743 * would match a more specific route
746 if (IS_ZEBRA_DEBUG_NHT_DETAILED
) {
747 char buf
[PREFIX_STRLEN
];
749 zlog_debug("%s: %s Being examined for Nexthop Tracking Count: %zd",
751 srcdest_rnode2str(rn
, buf
, sizeof(buf
)),
752 dest
? rnh_list_count(&dest
->nht
) : 0);
757 dest
= rib_dest_from_rnode(rn
);
761 * If we have any rnh's stored in the nht list
762 * then we know that this route node was used for
763 * nht resolution and as such we need to call the
764 * nexthop tracking evaluation code
766 frr_each_safe(rnh_list
, &dest
->nht
, rnh
) {
767 struct zebra_vrf
*zvrf
=
768 zebra_vrf_lookup_by_id(rnh
->vrf_id
);
769 struct prefix
*p
= &rnh
->node
->p
;
771 if (IS_ZEBRA_DEBUG_NHT_DETAILED
) {
772 char buf1
[PREFIX_STRLEN
];
773 char buf2
[PREFIX_STRLEN
];
775 zlog_debug("%u:%s has Nexthop(%s) Type: %s depending on it, evaluating %u:%u",
777 srcdest_rnode2str(rn
, buf1
,
779 prefix2str(p
, buf2
, sizeof(buf2
)),
780 rnh_type2str(rnh
->type
),
785 * If we have evaluated this node on this pass
786 * already, due to following the tree up
787 * then we know that we can move onto the next
790 * Additionally we call zebra_evaluate_rnh
791 * when we gc the dest. In this case we know
792 * that there must be no other re's where
793 * we were originally as such we know that
794 * that sequence number is ok to respect.
796 if (rnh
->seqno
== seq
) {
797 if (IS_ZEBRA_DEBUG_NHT_DETAILED
)
799 "\tNode processed and moved already");
804 zebra_evaluate_rnh(zvrf
, family2afi(p
->family
), 0,
810 dest
= rib_dest_from_rnode(rn
);
817 * Garbage collect the rib dest corresponding to the given route node
820 * Returns true if the dest was deleted, false otherwise.
822 int rib_gc_dest(struct route_node
*rn
)
826 dest
= rib_dest_from_rnode(rn
);
830 if (!rib_can_delete_dest(dest
))
833 if (IS_ZEBRA_DEBUG_RIB
) {
834 struct zebra_vrf
*zvrf
;
836 zvrf
= rib_dest_vrf(dest
);
837 rnode_debug(rn
, zvrf_id(zvrf
), "removing dest from table");
840 zebra_rib_evaluate_rn_nexthops(rn
, zebra_router_get_next_sequence());
843 rnh_list_fini(&dest
->nht
);
844 XFREE(MTYPE_RIB_DEST
, dest
);
848 * Release the one reference that we keep on the route node.
850 route_unlock_node(rn
);
854 static void rib_process_add_fib(struct zebra_vrf
*zvrf
, struct route_node
*rn
,
855 struct route_entry
*new)
857 hook_call(rib_update
, rn
, "new route selected");
859 /* Update real nexthop. This may actually determine if nexthop is active
861 if (!nexthop_group_active_nexthop_num(&new->ng
)) {
862 UNSET_FLAG(new->status
, ROUTE_ENTRY_CHANGED
);
866 if (IS_ZEBRA_DEBUG_RIB
) {
867 char buf
[SRCDEST2STR_BUFFER
];
868 srcdest_rnode2str(rn
, buf
, sizeof(buf
));
869 zlog_debug("%u:%s: Adding route rn %p, re %p (%s)",
870 zvrf_id(zvrf
), buf
, rn
, new,
871 zebra_route_string(new->type
));
874 /* If labeled-unicast route, install transit LSP. */
875 if (zebra_rib_labeled_unicast(new))
876 zebra_mpls_lsp_install(zvrf
, rn
, new);
878 rib_install_kernel(rn
, new, NULL
);
880 UNSET_FLAG(new->status
, ROUTE_ENTRY_CHANGED
);
883 static void rib_process_del_fib(struct zebra_vrf
*zvrf
, struct route_node
*rn
,
884 struct route_entry
*old
)
886 hook_call(rib_update
, rn
, "removing existing route");
888 /* Uninstall from kernel. */
889 if (IS_ZEBRA_DEBUG_RIB
) {
890 char buf
[SRCDEST2STR_BUFFER
];
891 srcdest_rnode2str(rn
, buf
, sizeof(buf
));
892 zlog_debug("%u:%s: Deleting route rn %p, re %p (%s)",
893 zvrf_id(zvrf
), buf
, rn
, old
,
894 zebra_route_string(old
->type
));
897 /* If labeled-unicast route, uninstall transit LSP. */
898 if (zebra_rib_labeled_unicast(old
))
899 zebra_mpls_lsp_uninstall(zvrf
, rn
, old
);
901 rib_uninstall_kernel(rn
, old
);
903 /* Update nexthop for route, reset changed flag. */
904 /* Note: this code also handles the Linux case when an interface goes
905 * down, causing the kernel to delete routes without sending DELROUTE
908 if (RIB_KERNEL_ROUTE(old
))
909 SET_FLAG(old
->status
, ROUTE_ENTRY_REMOVED
);
911 UNSET_FLAG(old
->status
, ROUTE_ENTRY_CHANGED
);
914 static void rib_process_update_fib(struct zebra_vrf
*zvrf
,
915 struct route_node
*rn
,
916 struct route_entry
*old
,
917 struct route_entry
*new)
922 * We have to install or update if a new route has been selected or
923 * something has changed.
925 if (new != old
|| CHECK_FLAG(new->status
, ROUTE_ENTRY_CHANGED
)) {
926 hook_call(rib_update
, rn
, "updating existing route");
928 /* Update the nexthop; we could determine here that nexthop is
930 if (nexthop_group_active_nexthop_num(&new->ng
))
933 /* If nexthop is active, install the selected route, if
935 * the install succeeds, cleanup flags for prior route, if
940 if (IS_ZEBRA_DEBUG_RIB
) {
941 char buf
[SRCDEST2STR_BUFFER
];
942 srcdest_rnode2str(rn
, buf
, sizeof(buf
));
945 "%u:%s: Updating route rn %p, re %p (%s) old %p (%s)",
946 zvrf_id(zvrf
), buf
, rn
, new,
947 zebra_route_string(new->type
),
949 zebra_route_string(old
->type
));
952 "%u:%s: Updating route rn %p, re %p (%s)",
953 zvrf_id(zvrf
), buf
, rn
, new,
954 zebra_route_string(new->type
));
957 /* If labeled-unicast route, uninstall transit LSP. */
958 if (zebra_rib_labeled_unicast(old
))
959 zebra_mpls_lsp_uninstall(zvrf
, rn
, old
);
962 * Non-system route should be installed.
963 * If labeled-unicast route, install transit
966 if (zebra_rib_labeled_unicast(new))
967 zebra_mpls_lsp_install(zvrf
, rn
, new);
969 rib_install_kernel(rn
, new, old
);
973 * If nexthop for selected route is not active or install
975 * may need to uninstall and delete for redistribution.
978 if (IS_ZEBRA_DEBUG_RIB
) {
979 char buf
[SRCDEST2STR_BUFFER
];
980 srcdest_rnode2str(rn
, buf
, sizeof(buf
));
983 "%u:%s: Deleting route rn %p, re %p (%s) old %p (%s) - nexthop inactive",
984 zvrf_id(zvrf
), buf
, rn
, new,
985 zebra_route_string(new->type
),
987 zebra_route_string(old
->type
));
990 "%u:%s: Deleting route rn %p, re %p (%s) - nexthop inactive",
991 zvrf_id(zvrf
), buf
, rn
, new,
992 zebra_route_string(new->type
));
995 /* If labeled-unicast route, uninstall transit LSP. */
996 if (zebra_rib_labeled_unicast(old
))
997 zebra_mpls_lsp_uninstall(zvrf
, rn
, old
);
999 rib_uninstall_kernel(rn
, old
);
1003 * Same route selected; check if in the FIB and if not,
1004 * re-install. This is housekeeping code to deal with
1005 * race conditions in kernel with linux netlink reporting
1006 * interface up before IPv4 or IPv6 protocol is ready
1009 if (!CHECK_FLAG(new->status
, ROUTE_ENTRY_INSTALLED
) ||
1010 RIB_SYSTEM_ROUTE(new))
1011 rib_install_kernel(rn
, new, NULL
);
1014 /* Update prior route. */
1016 UNSET_FLAG(old
->status
, ROUTE_ENTRY_CHANGED
);
1018 /* Clear changed flag. */
1019 UNSET_FLAG(new->status
, ROUTE_ENTRY_CHANGED
);
1022 /* Check if 'alternate' RIB entry is better than 'current'. */
1023 static struct route_entry
*rib_choose_best(struct route_entry
*current
,
1024 struct route_entry
*alternate
)
1026 if (current
== NULL
)
1029 /* filter route selection in following order:
1030 * - connected beats other types
1031 * - if both connected, loopback or vrf wins
1032 * - lower distance beats higher
1033 * - lower metric beats higher for equal distance
1034 * - last, hence oldest, route wins tie break.
1037 /* Connected routes. Check to see if either are a vrf
1038 * or loopback interface. If not, pick the last connected
1039 * route of the set of lowest metric connected routes.
1041 if (alternate
->type
== ZEBRA_ROUTE_CONNECT
) {
1042 if (current
->type
!= ZEBRA_ROUTE_CONNECT
)
1045 /* both are connected. are either loop or vrf? */
1046 struct nexthop
*nexthop
= NULL
;
1048 for (ALL_NEXTHOPS(alternate
->ng
, nexthop
)) {
1049 if (if_is_loopback_or_vrf(if_lookup_by_index(
1050 nexthop
->ifindex
, alternate
->vrf_id
)))
1054 for (ALL_NEXTHOPS(current
->ng
, nexthop
)) {
1055 if (if_is_loopback_or_vrf(if_lookup_by_index(
1056 nexthop
->ifindex
, current
->vrf_id
)))
1060 /* Neither are loop or vrf so pick best metric */
1061 if (alternate
->metric
<= current
->metric
)
1067 if (current
->type
== ZEBRA_ROUTE_CONNECT
)
1070 /* higher distance loses */
1071 if (alternate
->distance
< current
->distance
)
1073 if (current
->distance
< alternate
->distance
)
1076 /* metric tie-breaks equal distance */
1077 if (alternate
->metric
<= current
->metric
)
1083 /* Core function for processing routing information base. */
1084 static void rib_process(struct route_node
*rn
)
1086 struct route_entry
*re
;
1087 struct route_entry
*next
;
1088 struct route_entry
*old_selected
= NULL
;
1089 struct route_entry
*new_selected
= NULL
;
1090 struct route_entry
*old_fib
= NULL
;
1091 struct route_entry
*new_fib
= NULL
;
1092 struct route_entry
*best
= NULL
;
1093 char buf
[SRCDEST2STR_BUFFER
];
1095 struct zebra_vrf
*zvrf
= NULL
;
1096 const struct prefix
*p
, *src_p
;
1098 srcdest_rnode_prefixes(rn
, &p
, &src_p
);
1099 vrf_id_t vrf_id
= VRF_UNKNOWN
;
1103 dest
= rib_dest_from_rnode(rn
);
1105 zvrf
= rib_dest_vrf(dest
);
1106 vrf_id
= zvrf_id(zvrf
);
1109 if (IS_ZEBRA_DEBUG_RIB
)
1110 srcdest_rnode2str(rn
, buf
, sizeof(buf
));
1112 if (IS_ZEBRA_DEBUG_RIB_DETAILED
)
1113 zlog_debug("%u:%s: Processing rn %p", vrf_id
, buf
, rn
);
1116 * we can have rn's that have a NULL info pointer
1117 * (dest). As such let's not let the deref happen
1118 * additionally we know RNODE_FOREACH_RE_SAFE
1119 * will not iterate so we are ok.
1122 old_fib
= dest
->selected_fib
;
1124 RNODE_FOREACH_RE_SAFE (rn
, re
, next
) {
1125 if (IS_ZEBRA_DEBUG_RIB_DETAILED
)
1127 "%u:%s: Examine re %p (%s) status %x flags %x dist %d metric %d",
1128 vrf_id
, buf
, re
, zebra_route_string(re
->type
),
1129 re
->status
, re
->flags
, re
->distance
,
1132 /* Currently selected re. */
1133 if (CHECK_FLAG(re
->flags
, ZEBRA_FLAG_SELECTED
)) {
1134 assert(old_selected
== NULL
);
1138 /* Skip deleted entries from selection */
1139 if (CHECK_FLAG(re
->status
, ROUTE_ENTRY_REMOVED
))
1142 /* Skip unreachable nexthop. */
1143 /* This first call to nexthop_active_update is merely to
1144 * determine if there's any change to nexthops associated
1145 * with this RIB entry. Now, rib_process() can be invoked due
1146 * to an external event such as link down or due to
1147 * next-hop-tracking evaluation. In the latter case,
1148 * a decision has already been made that the NHs have changed.
1149 * So, no need to invoke a potentially expensive call again.
1150 * Further, since the change might be in a recursive NH which
1151 * is not caught in the nexthop_active_update() code. Thus, we
1152 * might miss changes to recursive NHs.
1154 if (CHECK_FLAG(re
->status
, ROUTE_ENTRY_CHANGED
)
1155 && !nexthop_active_update(rn
, re
)) {
1156 if (re
->type
== ZEBRA_ROUTE_TABLE
) {
1157 /* XXX: HERE BE DRAGONS!!!!!
1158 * In all honesty, I have not yet figured out
1159 * what this part does or why the
1160 * ROUTE_ENTRY_CHANGED test above is correct
1161 * or why we need to delete a route here, and
1162 * also not whether this concerns both selected
1163 * and fib route, or only selected
1166 * This entry was denied by the 'ip protocol
1167 * table' route-map, we need to delete it */
1168 if (re
!= old_selected
) {
1169 if (IS_ZEBRA_DEBUG_RIB
)
1171 "%s: %u:%s: imported via import-table but denied "
1172 "by the ip protocol table route-map",
1173 __func__
, vrf_id
, buf
);
1176 SET_FLAG(re
->status
,
1177 ROUTE_ENTRY_REMOVED
);
1183 /* Infinite distance. */
1184 if (re
->distance
== DISTANCE_INFINITY
) {
1185 UNSET_FLAG(re
->status
, ROUTE_ENTRY_CHANGED
);
1189 if (CHECK_FLAG(re
->flags
, ZEBRA_FLAG_FIB_OVERRIDE
)) {
1190 best
= rib_choose_best(new_fib
, re
);
1191 if (new_fib
&& best
!= new_fib
)
1192 UNSET_FLAG(new_fib
->status
,
1193 ROUTE_ENTRY_CHANGED
);
1196 best
= rib_choose_best(new_selected
, re
);
1197 if (new_selected
&& best
!= new_selected
)
1198 UNSET_FLAG(new_selected
->status
,
1199 ROUTE_ENTRY_CHANGED
);
1200 new_selected
= best
;
1203 UNSET_FLAG(re
->status
, ROUTE_ENTRY_CHANGED
);
1204 } /* RNODE_FOREACH_RE */
1206 /* If no FIB override route, use the selected route also for FIB */
1207 if (new_fib
== NULL
)
1208 new_fib
= new_selected
;
1210 /* After the cycle is finished, the following pointers will be set:
1211 * old_selected --- RE entry currently having SELECTED
1212 * new_selected --- RE entry that is newly SELECTED
1213 * old_fib --- RE entry currently in kernel FIB
1214 * new_fib --- RE entry that is newly to be in kernel FIB
1216 * new_selected will get SELECTED flag, and is going to be redistributed
1217 * the zclients. new_fib (which can be new_selected) will be installed
1221 if (IS_ZEBRA_DEBUG_RIB_DETAILED
) {
1223 "%u:%s: After processing: old_selected %p new_selected %p old_fib %p new_fib %p",
1224 vrf_id
, buf
, (void *)old_selected
, (void *)new_selected
,
1225 (void *)old_fib
, (void *)new_fib
);
1228 /* Buffer ROUTE_ENTRY_CHANGED here, because it will get cleared if
1229 * fib == selected */
1230 bool selected_changed
= new_selected
&& CHECK_FLAG(new_selected
->status
,
1231 ROUTE_ENTRY_CHANGED
);
1233 /* Update fib according to selection results */
1234 if (new_fib
&& old_fib
)
1235 rib_process_update_fib(zvrf
, rn
, old_fib
, new_fib
);
1237 rib_process_add_fib(zvrf
, rn
, new_fib
);
1239 rib_process_del_fib(zvrf
, rn
, old_fib
);
1241 /* Update SELECTED entry */
1242 if (old_selected
!= new_selected
|| selected_changed
) {
1244 if (new_selected
&& new_selected
!= new_fib
)
1245 UNSET_FLAG(new_selected
->status
, ROUTE_ENTRY_CHANGED
);
1248 SET_FLAG(new_selected
->flags
, ZEBRA_FLAG_SELECTED
);
1252 redistribute_delete(p
, src_p
, old_selected
);
1253 if (old_selected
!= new_selected
)
1254 UNSET_FLAG(old_selected
->flags
,
1255 ZEBRA_FLAG_SELECTED
);
1259 /* Remove all RE entries queued for removal */
1260 RNODE_FOREACH_RE_SAFE (rn
, re
, next
) {
1261 if (CHECK_FLAG(re
->status
, ROUTE_ENTRY_REMOVED
)) {
1262 if (IS_ZEBRA_DEBUG_RIB
) {
1263 rnode_debug(rn
, vrf_id
, "rn %p, removing re %p",
1264 (void *)rn
, (void *)re
);
1271 * Check if the dest can be deleted now.
1276 static void zebra_rib_evaluate_mpls(struct route_node
*rn
)
1278 rib_dest_t
*dest
= rib_dest_from_rnode(rn
);
1279 struct zebra_vrf
*zvrf
= vrf_info_lookup(VRF_DEFAULT
);
1284 if (CHECK_FLAG(dest
->flags
, RIB_DEST_UPDATE_LSPS
)) {
1285 if (IS_ZEBRA_DEBUG_MPLS
)
1287 "%u: Scheduling all LSPs upon RIB completion",
1289 zebra_mpls_lsp_schedule(zvrf
);
1290 mpls_unmark_lsps_for_processing(rn
);
1295 * Utility to match route with dplane context data
1297 static bool rib_route_match_ctx(const struct route_entry
*re
,
1298 const struct zebra_dplane_ctx
*ctx
,
1301 bool result
= false;
1305 * In 'update' case, we test info about the 'previous' or
1308 if ((re
->type
== dplane_ctx_get_old_type(ctx
)) &&
1309 (re
->instance
== dplane_ctx_get_old_instance(ctx
))) {
1312 /* TODO -- we're using this extra test, but it's not
1313 * exactly clear why.
1315 if (re
->type
== ZEBRA_ROUTE_STATIC
&&
1316 (re
->distance
!= dplane_ctx_get_old_distance(ctx
) ||
1317 re
->tag
!= dplane_ctx_get_old_tag(ctx
))) {
1324 * Ordinary, single-route case using primary context info
1326 if ((dplane_ctx_get_op(ctx
) != DPLANE_OP_ROUTE_DELETE
) &&
1327 CHECK_FLAG(re
->status
, ROUTE_ENTRY_REMOVED
)) {
1328 /* Skip route that's been deleted */
1332 if ((re
->type
== dplane_ctx_get_type(ctx
)) &&
1333 (re
->instance
== dplane_ctx_get_instance(ctx
))) {
1336 /* TODO -- we're using this extra test, but it's not
1337 * exactly clear why.
1339 if (re
->type
== ZEBRA_ROUTE_STATIC
&&
1340 (re
->distance
!= dplane_ctx_get_distance(ctx
) ||
1341 re
->tag
!= dplane_ctx_get_tag(ctx
))) {
1352 static void zebra_rib_fixup_system(struct route_node
*rn
)
1354 struct route_entry
*re
;
1356 RNODE_FOREACH_RE(rn
, re
) {
1357 struct nexthop
*nhop
;
1359 if (!RIB_SYSTEM_ROUTE(re
))
1362 if (CHECK_FLAG(re
->status
, ROUTE_ENTRY_REMOVED
))
1365 SET_FLAG(re
->status
, ROUTE_ENTRY_INSTALLED
);
1366 UNSET_FLAG(re
->status
, ROUTE_ENTRY_QUEUED
);
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 /* Ensure we clear the QUEUED flag */
1931 UNSET_FLAG(re
->status
, ROUTE_ENTRY_QUEUED
);
1933 /* Is this a notification that ... matters? We only really care about
1934 * the route that is currently selected for installation.
1936 if (re
!= dest
->selected_fib
) {
1937 /* TODO -- don't skip processing entirely? We might like to
1938 * at least report on the event.
1941 zlog_debug("%u:%s dplane notif, but type %s not selected_fib",
1942 dplane_ctx_get_vrf(ctx
), dest_str
,
1944 dplane_ctx_get_type(ctx
)));
1948 /* We'll want to determine whether the installation status of the
1949 * route has changed: we'll check the status before processing,
1950 * and then again if there's been a change.
1953 for (ALL_NEXTHOPS_PTR(rib_active_nhg(re
), nexthop
)) {
1954 if (CHECK_FLAG(nexthop
->flags
, NEXTHOP_FLAG_FIB
))
1958 /* Update zebra's nexthop FIB flags based on the context struct's
1961 fib_changed
= rib_update_re_from_ctx(re
, rn
, ctx
);
1965 zlog_debug("%u:%s No change from dplane notification",
1966 dplane_ctx_get_vrf(ctx
), dest_str
);
1972 * Perform follow-up work if the actual status of the prefix
1977 for (ALL_NEXTHOPS_PTR(rib_active_nhg(re
), nexthop
)) {
1978 if (CHECK_FLAG(nexthop
->flags
, NEXTHOP_FLAG_FIB
))
1982 /* Various fib transitions: changed nexthops; from installed to
1983 * not-installed; or not-installed to installed.
1985 if (start_count
> 0 && end_count
> 0) {
1987 /* Changed nexthops - update kernel/others */
1988 dplane_route_notif_update(rn
, re
,
1989 DPLANE_OP_ROUTE_UPDATE
, ctx
);
1991 } else if (start_count
== 0 && end_count
> 0) {
1993 zlog_debug("%u:%s installed transition from dplane notification",
1994 dplane_ctx_get_vrf(ctx
), dest_str
);
1996 /* We expect this to be the selected route, so we want
1997 * to tell others about this transition.
1999 SET_FLAG(re
->status
, ROUTE_ENTRY_INSTALLED
);
2001 /* Changed nexthops - update kernel/others */
2002 dplane_route_notif_update(rn
, re
, DPLANE_OP_ROUTE_INSTALL
, ctx
);
2004 /* Redistribute, lsp, and nht update */
2005 redistribute_update(dest_pfx
, src_pfx
, re
, NULL
);
2007 zebra_rib_evaluate_rn_nexthops(
2008 rn
, zebra_router_get_next_sequence());
2010 zebra_rib_evaluate_mpls(rn
);
2012 } else if (start_count
> 0 && end_count
== 0) {
2014 zlog_debug("%u:%s un-installed transition from dplane notification",
2015 dplane_ctx_get_vrf(ctx
), dest_str
);
2017 /* Transition from _something_ installed to _nothing_
2020 /* We expect this to be the selected route, so we want
2021 * to tell others about this transistion.
2023 UNSET_FLAG(re
->status
, ROUTE_ENTRY_INSTALLED
);
2025 /* Changed nexthops - update kernel/others */
2026 dplane_route_notif_update(rn
, re
, DPLANE_OP_ROUTE_DELETE
, ctx
);
2028 /* Redistribute, lsp, and nht update */
2029 redistribute_delete(dest_pfx
, src_pfx
, re
);
2031 zebra_rib_evaluate_rn_nexthops(
2032 rn
, zebra_router_get_next_sequence());
2034 zebra_rib_evaluate_mpls(rn
);
2039 route_unlock_node(rn
);
2041 /* Return context to dataplane module */
2042 dplane_ctx_fini(&ctx
);
2045 /* Take a list of route_node structs and return 1, if there was a record
2046 * picked from it and processed by rib_process(). Don't process more,
2047 * than one RN record; operate only in the specified sub-queue.
2049 static unsigned int process_subq(struct list
*subq
, uint8_t qindex
)
2051 struct listnode
*lnode
= listhead(subq
);
2052 struct route_node
*rnode
;
2054 struct zebra_vrf
*zvrf
= NULL
;
2059 rnode
= listgetdata(lnode
);
2060 dest
= rib_dest_from_rnode(rnode
);
2062 zvrf
= rib_dest_vrf(dest
);
2066 if (IS_ZEBRA_DEBUG_RIB_DETAILED
) {
2067 char buf
[SRCDEST2STR_BUFFER
];
2069 srcdest_rnode2str(rnode
, buf
, sizeof(buf
));
2070 zlog_debug("%u:%s: rn %p dequeued from sub-queue %u",
2071 zvrf
? zvrf_id(zvrf
) : 0, buf
, rnode
, qindex
);
2075 UNSET_FLAG(rib_dest_from_rnode(rnode
)->flags
,
2076 RIB_ROUTE_QUEUED(qindex
));
2081 zlog_debug ("%s: called for route_node (%p, %d) with no ribs",
2082 __func__
, rnode
, rnode
->lock
);
2083 zlog_backtrace(LOG_DEBUG
);
2086 route_unlock_node(rnode
);
2087 list_delete_node(subq
, lnode
);
2093 * Perform next-hop tracking processing after RIB updates.
2095 static void do_nht_processing(void)
2099 /* Dispatch the meta queue by picking, processing and unlocking the next RN from
2100 * a non-empty sub-queue with lowest priority. wq is equal to zebra->ribq and
2102 * is pointed to the meta queue structure.
2104 static wq_item_status
meta_queue_process(struct work_queue
*dummy
, void *data
)
2106 struct meta_queue
*mq
= data
;
2108 uint32_t queue_len
, queue_limit
;
2110 /* Ensure there's room for more dataplane updates */
2111 queue_limit
= dplane_get_in_queue_limit();
2112 queue_len
= dplane_get_in_queue_len();
2113 if (queue_len
> queue_limit
) {
2114 if (IS_ZEBRA_DEBUG_RIB_DETAILED
)
2115 zlog_debug("rib queue: dplane queue len %u, limit %u, retrying",
2116 queue_len
, queue_limit
);
2118 /* Ensure that the meta-queue is actually enqueued */
2119 if (work_queue_empty(zrouter
.ribq
))
2120 work_queue_add(zrouter
.ribq
, zrouter
.mq
);
2122 return WQ_QUEUE_BLOCKED
;
2125 for (i
= 0; i
< MQ_SIZE
; i
++)
2126 if (process_subq(mq
->subq
[i
], i
)) {
2130 return mq
->size
? WQ_REQUEUE
: WQ_SUCCESS
;
2135 * Look into the RN and queue it into the highest priority queue
2136 * at this point in time for processing.
2138 * We will enqueue a route node only once per invocation.
2140 * There are two possibilities here that should be kept in mind.
2141 * If the original invocation has not been pulled off for processing
2142 * yet, A subsuquent invocation can have a route entry with a better
2143 * meta queue index value and we can have a situation where
2144 * we might have the same node enqueued 2 times. Not necessarily
2145 * an optimal situation but it should be ok.
2147 * The other possibility is that the original invocation has not
2148 * been pulled off for processing yet, A subsusquent invocation
2149 * doesn't have a route_entry with a better meta-queue and the
2150 * original metaqueue index value will win and we'll end up with
2151 * the route node enqueued once.
2153 static void rib_meta_queue_add(struct meta_queue
*mq
, struct route_node
*rn
)
2155 struct route_entry
*re
= NULL
, *curr_re
= NULL
;
2156 uint8_t qindex
= MQ_SIZE
, curr_qindex
= MQ_SIZE
;
2158 RNODE_FOREACH_RE (rn
, curr_re
) {
2159 curr_qindex
= route_info
[curr_re
->type
].meta_q_map
;
2161 if (curr_qindex
<= qindex
) {
2163 qindex
= curr_qindex
;
2170 /* Invariant: at this point we always have rn->info set. */
2171 if (CHECK_FLAG(rib_dest_from_rnode(rn
)->flags
,
2172 RIB_ROUTE_QUEUED(qindex
))) {
2173 if (IS_ZEBRA_DEBUG_RIB_DETAILED
)
2174 rnode_debug(rn
, re
->vrf_id
,
2175 "rn %p is already queued in sub-queue %u",
2176 (void *)rn
, qindex
);
2180 SET_FLAG(rib_dest_from_rnode(rn
)->flags
, RIB_ROUTE_QUEUED(qindex
));
2181 listnode_add(mq
->subq
[qindex
], rn
);
2182 route_lock_node(rn
);
2185 if (IS_ZEBRA_DEBUG_RIB_DETAILED
)
2186 rnode_debug(rn
, re
->vrf_id
, "queued rn %p into sub-queue %u",
2187 (void *)rn
, qindex
);
2190 /* Add route_node to work queue and schedule processing */
2191 void rib_queue_add(struct route_node
*rn
)
2195 /* Pointless to queue a route_node with no RIB entries to add or remove
2197 if (!rnode_to_ribs(rn
)) {
2198 zlog_debug("%s: called for route_node (%p, %d) with no ribs",
2199 __func__
, (void *)rn
, rn
->lock
);
2200 zlog_backtrace(LOG_DEBUG
);
2204 if (zrouter
.ribq
== NULL
) {
2205 flog_err(EC_ZEBRA_WQ_NONEXISTENT
,
2206 "%s: work_queue does not exist!", __func__
);
2211 * The RIB queue should normally be either empty or holding the only
2212 * work_queue_item element. In the latter case this element would
2213 * hold a pointer to the meta queue structure, which must be used to
2214 * actually queue the route nodes to process. So create the MQ
2215 * holder, if necessary, then push the work into it in any case.
2216 * This semantics was introduced after 0.99.9 release.
2218 if (work_queue_empty(zrouter
.ribq
))
2219 work_queue_add(zrouter
.ribq
, zrouter
.mq
);
2221 rib_meta_queue_add(zrouter
.mq
, rn
);
2226 /* Create new meta queue.
2227 A destructor function doesn't seem to be necessary here.
2229 static struct meta_queue
*meta_queue_new(void)
2231 struct meta_queue
*new;
2234 new = XCALLOC(MTYPE_WORK_QUEUE
, sizeof(struct meta_queue
));
2236 for (i
= 0; i
< MQ_SIZE
; i
++) {
2237 new->subq
[i
] = list_new();
2238 assert(new->subq
[i
]);
2244 void meta_queue_free(struct meta_queue
*mq
)
2248 for (i
= 0; i
< MQ_SIZE
; i
++)
2249 list_delete(&mq
->subq
[i
]);
2251 XFREE(MTYPE_WORK_QUEUE
, mq
);
2254 /* initialise zebra rib work queue */
2255 static void rib_queue_init(void)
2257 if (!(zrouter
.ribq
= work_queue_new(zrouter
.master
,
2258 "route_node processing"))) {
2259 flog_err(EC_ZEBRA_WQ_NONEXISTENT
,
2260 "%s: could not initialise work queue!", __func__
);
2264 /* fill in the work queue spec */
2265 zrouter
.ribq
->spec
.workfunc
= &meta_queue_process
;
2266 zrouter
.ribq
->spec
.errorfunc
= NULL
;
2267 zrouter
.ribq
->spec
.completion_func
= NULL
;
2268 /* XXX: TODO: These should be runtime configurable via vty */
2269 zrouter
.ribq
->spec
.max_retries
= 3;
2270 zrouter
.ribq
->spec
.hold
= ZEBRA_RIB_PROCESS_HOLD_TIME
;
2271 zrouter
.ribq
->spec
.retry
= ZEBRA_RIB_PROCESS_RETRY_TIME
;
2273 if (!(zrouter
.mq
= meta_queue_new())) {
2274 flog_err(EC_ZEBRA_WQ_NONEXISTENT
,
2275 "%s: could not initialise meta queue!", __func__
);
2281 rib_dest_t
*zebra_rib_create_dest(struct route_node
*rn
)
2285 dest
= XCALLOC(MTYPE_RIB_DEST
, sizeof(rib_dest_t
));
2286 rnh_list_init(&dest
->nht
);
2287 route_lock_node(rn
); /* rn route table reference */
2294 /* RIB updates are processed via a queue of pointers to route_nodes.
2296 * The queue length is bounded by the maximal size of the routing table,
2297 * as a route_node will not be requeued, if already queued.
2299 * REs are submitted via rib_addnode or rib_delnode which set minimal
2300 * state, or static_install_route (when an existing RE is updated)
2301 * and then submit route_node to queue for best-path selection later.
2302 * Order of add/delete state changes are preserved for any given RE.
2304 * Deleted REs are reaped during best-path selection.
2307 * |-> rib_link or unset ROUTE_ENTRY_REMOVE |->Update kernel with
2308 * |-------->| | best RE, if required
2310 * static_install->|->rib_addqueue...... -> rib_process
2312 * |-------->| |-> rib_unlink
2313 * |-> set ROUTE_ENTRY_REMOVE |
2314 * rib_delnode (RE freed)
2316 * The 'info' pointer of a route_node points to a rib_dest_t
2317 * ('dest'). Queueing state for a route_node is kept on the dest. The
2318 * dest is created on-demand by rib_link() and is kept around at least
2319 * as long as there are ribs hanging off it (@see rib_gc_dest()).
2321 * Refcounting (aka "locking" throughout the GNU Zebra and Quagga code):
2323 * - route_nodes: refcounted by:
2324 * - dest attached to route_node:
2325 * - managed by: rib_link/rib_gc_dest
2326 * - route_node processing queue
2327 * - managed by: rib_addqueue, rib_process.
2331 /* Add RE to head of the route node. */
2332 static void rib_link(struct route_node
*rn
, struct route_entry
*re
, int process
)
2336 const char *rmap_name
;
2340 dest
= rib_dest_from_rnode(rn
);
2342 if (IS_ZEBRA_DEBUG_RIB_DETAILED
)
2343 rnode_debug(rn
, re
->vrf_id
, "rn %p adding dest", rn
);
2345 dest
= zebra_rib_create_dest(rn
);
2348 re_list_add_head(&dest
->routes
, re
);
2350 afi
= (rn
->p
.family
== AF_INET
)
2352 : (rn
->p
.family
== AF_INET6
) ? AFI_IP6
: AFI_MAX
;
2353 if (is_zebra_import_table_enabled(afi
, re
->vrf_id
, re
->table
)) {
2354 struct zebra_vrf
*zvrf
= zebra_vrf_lookup_by_id(re
->vrf_id
);
2356 rmap_name
= zebra_get_import_table_route_map(afi
, re
->table
);
2357 zebra_add_import_table_entry(zvrf
, rn
, re
, rmap_name
);
2362 static void rib_addnode(struct route_node
*rn
,
2363 struct route_entry
*re
, int process
)
2365 /* RE node has been un-removed before route-node is processed.
2366 * route_node must hence already be on the queue for processing..
2368 if (CHECK_FLAG(re
->status
, ROUTE_ENTRY_REMOVED
)) {
2369 if (IS_ZEBRA_DEBUG_RIB
)
2370 rnode_debug(rn
, re
->vrf_id
, "rn %p, un-removed re %p",
2371 (void *)rn
, (void *)re
);
2373 UNSET_FLAG(re
->status
, ROUTE_ENTRY_REMOVED
);
2376 rib_link(rn
, re
, process
);
2382 * Detach a rib structure from a route_node.
2384 * Note that a call to rib_unlink() should be followed by a call to
2385 * rib_gc_dest() at some point. This allows a rib_dest_t that is no
2386 * longer required to be deleted.
2388 void rib_unlink(struct route_node
*rn
, struct route_entry
*re
)
2394 if (IS_ZEBRA_DEBUG_RIB
)
2395 rnode_debug(rn
, re
->vrf_id
, "rn %p, re %p", (void *)rn
,
2398 dest
= rib_dest_from_rnode(rn
);
2400 re_list_del(&dest
->routes
, re
);
2402 if (dest
->selected_fib
== re
)
2403 dest
->selected_fib
= NULL
;
2405 nexthops_free(re
->ng
.nexthop
);
2406 nexthops_free(re
->fib_ng
.nexthop
);
2408 XFREE(MTYPE_RE
, re
);
2411 void rib_delnode(struct route_node
*rn
, struct route_entry
*re
)
2415 if (IS_ZEBRA_DEBUG_RIB
)
2416 rnode_debug(rn
, re
->vrf_id
, "rn %p, re %p, removing",
2417 (void *)rn
, (void *)re
);
2418 SET_FLAG(re
->status
, ROUTE_ENTRY_REMOVED
);
2420 afi
= (rn
->p
.family
== AF_INET
)
2422 : (rn
->p
.family
== AF_INET6
) ? AFI_IP6
: AFI_MAX
;
2423 if (is_zebra_import_table_enabled(afi
, re
->vrf_id
, re
->table
)) {
2424 struct zebra_vrf
*zvrf
= zebra_vrf_lookup_by_id(re
->vrf_id
);
2426 zebra_del_import_table_entry(zvrf
, rn
, re
);
2427 /* Just clean up if non main table */
2428 if (IS_ZEBRA_DEBUG_RIB
) {
2429 char buf
[SRCDEST2STR_BUFFER
];
2430 srcdest_rnode2str(rn
, buf
, sizeof(buf
));
2431 zlog_debug("%u:%s: Freeing route rn %p, re %p (%s)",
2432 re
->vrf_id
, buf
, rn
, re
,
2433 zebra_route_string(re
->type
));
2442 /* This function dumps the contents of a given RE entry into
2443 * standard debug log. Calling function name and IP prefix in
2444 * question are passed as 1st and 2nd arguments.
2447 void _route_entry_dump(const char *func
, union prefixconstptr pp
,
2448 union prefixconstptr src_pp
,
2449 const struct route_entry
*re
)
2451 const struct prefix
*src_p
= src_pp
.p
;
2452 bool is_srcdst
= src_p
&& src_p
->prefixlen
;
2453 char straddr
[PREFIX_STRLEN
];
2454 char srcaddr
[PREFIX_STRLEN
];
2455 char nhname
[PREFIX_STRLEN
];
2456 struct nexthop
*nexthop
;
2458 zlog_debug("%s: dumping RE entry %p for %s%s%s vrf %u", func
,
2459 (const void *)re
, prefix2str(pp
, straddr
, sizeof(straddr
)),
2460 is_srcdst
? " from " : "",
2461 is_srcdst
? prefix2str(src_pp
, srcaddr
, sizeof(srcaddr
))
2464 zlog_debug("%s: uptime == %lu, type == %u, instance == %d, table == %d",
2465 straddr
, (unsigned long)re
->uptime
, re
->type
, re
->instance
,
2468 "%s: metric == %u, mtu == %u, distance == %u, flags == %u, status == %u",
2469 straddr
, re
->metric
, re
->mtu
, re
->distance
, re
->flags
, re
->status
);
2470 zlog_debug("%s: nexthop_num == %u, nexthop_active_num == %u", straddr
,
2471 re
->nexthop_num
, re
->nexthop_active_num
);
2473 for (ALL_NEXTHOPS(re
->ng
, nexthop
)) {
2474 struct interface
*ifp
;
2475 struct vrf
*vrf
= vrf_lookup_by_id(nexthop
->vrf_id
);
2477 switch (nexthop
->type
) {
2478 case NEXTHOP_TYPE_BLACKHOLE
:
2479 sprintf(nhname
, "Blackhole");
2481 case NEXTHOP_TYPE_IFINDEX
:
2482 ifp
= if_lookup_by_index(nexthop
->ifindex
,
2484 sprintf(nhname
, "%s", ifp
? ifp
->name
: "Unknown");
2486 case NEXTHOP_TYPE_IPV4
:
2488 case NEXTHOP_TYPE_IPV4_IFINDEX
:
2489 inet_ntop(AF_INET
, &nexthop
->gate
, nhname
,
2492 case NEXTHOP_TYPE_IPV6
:
2493 case NEXTHOP_TYPE_IPV6_IFINDEX
:
2494 inet_ntop(AF_INET6
, &nexthop
->gate
, nhname
,
2498 zlog_debug("%s: %s %s[%u] vrf %s(%u) with flags %s%s%s%s%s%s",
2499 straddr
, (nexthop
->rparent
? " NH" : "NH"), nhname
,
2500 nexthop
->ifindex
, vrf
? vrf
->name
: "Unknown",
2502 (CHECK_FLAG(nexthop
->flags
, NEXTHOP_FLAG_ACTIVE
)
2505 (CHECK_FLAG(re
->status
, ROUTE_ENTRY_INSTALLED
)
2508 (CHECK_FLAG(nexthop
->flags
, NEXTHOP_FLAG_RECURSIVE
)
2511 (CHECK_FLAG(nexthop
->flags
, NEXTHOP_FLAG_ONLINK
)
2514 (CHECK_FLAG(nexthop
->flags
, NEXTHOP_FLAG_MATCHED
)
2517 (CHECK_FLAG(nexthop
->flags
, NEXTHOP_FLAG_DUPLICATE
)
2521 zlog_debug("%s: dump complete", straddr
);
2524 /* This is an exported helper to rtm_read() to dump the strange
2525 * RE entry found by rib_lookup_ipv4_route()
2528 void rib_lookup_and_dump(struct prefix_ipv4
*p
, vrf_id_t vrf_id
)
2530 struct route_table
*table
;
2531 struct route_node
*rn
;
2532 struct route_entry
*re
;
2533 char prefix_buf
[INET_ADDRSTRLEN
];
2536 table
= zebra_vrf_table(AFI_IP
, SAFI_UNICAST
, vrf_id
);
2538 flog_err(EC_ZEBRA_TABLE_LOOKUP_FAILED
,
2539 "%s:%u zebra_vrf_table() returned NULL", __func__
,
2544 /* Scan the RIB table for exactly matching RE entry. */
2545 rn
= route_node_lookup(table
, (struct prefix
*)p
);
2547 /* No route for this prefix. */
2549 zlog_debug("%s:%u lookup failed for %s", __func__
, vrf_id
,
2550 prefix2str((struct prefix
*)p
, prefix_buf
,
2551 sizeof(prefix_buf
)));
2556 route_unlock_node(rn
);
2559 RNODE_FOREACH_RE (rn
, re
) {
2560 zlog_debug("%s:%u rn %p, re %p: %s, %s",
2562 (void *)rn
, (void *)re
,
2563 (CHECK_FLAG(re
->status
, ROUTE_ENTRY_REMOVED
)
2566 (CHECK_FLAG(re
->flags
, ZEBRA_FLAG_SELECTED
)
2569 route_entry_dump(p
, NULL
, re
);
2573 /* Check if requested address assignment will fail due to another
2574 * route being installed by zebra in FIB already. Take necessary
2575 * actions, if needed: remove such a route from FIB and deSELECT
2576 * corresponding RE entry. Then put affected RN into RIBQ head.
2578 void rib_lookup_and_pushup(struct prefix_ipv4
*p
, vrf_id_t vrf_id
)
2580 struct route_table
*table
;
2581 struct route_node
*rn
;
2584 if (NULL
== (table
= zebra_vrf_table(AFI_IP
, SAFI_UNICAST
, vrf_id
))) {
2585 flog_err(EC_ZEBRA_TABLE_LOOKUP_FAILED
,
2586 "%s:%u zebra_vrf_table() returned NULL", __func__
,
2591 /* No matches would be the simplest case. */
2592 if (NULL
== (rn
= route_node_lookup(table
, (struct prefix
*)p
)))
2596 route_unlock_node(rn
);
2598 dest
= rib_dest_from_rnode(rn
);
2599 /* Check all RE entries. In case any changes have to be done, requeue
2600 * the RN into RIBQ head. If the routing message about the new connected
2601 * route (generated by the IP address we are going to assign very soon)
2602 * comes before the RIBQ is processed, the new RE entry will join
2603 * RIBQ record already on head. This is necessary for proper
2605 * of the rest of the RE.
2607 if (dest
->selected_fib
) {
2608 if (IS_ZEBRA_DEBUG_RIB
) {
2609 char buf
[PREFIX_STRLEN
];
2611 zlog_debug("%u:%s: freeing way for connected prefix",
2612 dest
->selected_fib
->vrf_id
,
2613 prefix2str(&rn
->p
, buf
, sizeof(buf
)));
2614 route_entry_dump(&rn
->p
, NULL
, dest
->selected_fib
);
2616 rib_uninstall(rn
, dest
->selected_fib
);
2621 int rib_add_multipath(afi_t afi
, safi_t safi
, struct prefix
*p
,
2622 struct prefix_ipv6
*src_p
, struct route_entry
*re
)
2624 struct route_table
*table
;
2625 struct route_node
*rn
;
2626 struct route_entry
*same
= NULL
;
2632 assert(!src_p
|| !src_p
->prefixlen
|| afi
== AFI_IP6
);
2635 table
= zebra_vrf_table_with_table_id(afi
, safi
, re
->vrf_id
, re
->table
);
2637 XFREE(MTYPE_RE
, re
);
2641 /* Make it sure prefixlen is applied to the prefix. */
2644 apply_mask_ipv6(src_p
);
2646 /* Set default distance by route type. */
2647 if (re
->distance
== 0)
2648 re
->distance
= route_distance(re
->type
);
2650 /* Lookup route node.*/
2651 rn
= srcdest_rnode_get(table
, p
, src_p
);
2654 * If same type of route are installed, treat it as a implicit
2656 * If the user has specified the No route replace semantics
2657 * for the install don't do a route replace.
2659 RNODE_FOREACH_RE (rn
, same
) {
2660 if (CHECK_FLAG(same
->status
, ROUTE_ENTRY_REMOVED
))
2663 if (same
->type
!= re
->type
)
2665 if (same
->instance
!= re
->instance
)
2667 if (same
->type
== ZEBRA_ROUTE_KERNEL
2668 && same
->metric
!= re
->metric
)
2671 if (CHECK_FLAG(re
->flags
, ZEBRA_FLAG_RR_USE_DISTANCE
) &&
2672 same
->distance
!= re
->distance
)
2676 * We should allow duplicate connected routes
2677 * because of IPv6 link-local routes and unnumbered
2678 * interfaces on Linux.
2680 if (same
->type
!= ZEBRA_ROUTE_CONNECT
)
2684 /* If this route is kernel/connected route, notify the dataplane. */
2685 if (RIB_SYSTEM_ROUTE(re
)) {
2686 /* Notify dataplane */
2687 dplane_sys_route_add(rn
, re
);
2690 /* Link new re to node.*/
2691 if (IS_ZEBRA_DEBUG_RIB
) {
2692 rnode_debug(rn
, re
->vrf_id
,
2693 "Inserting route rn %p, re %p (%s) existing %p",
2694 rn
, re
, zebra_route_string(re
->type
), same
);
2696 if (IS_ZEBRA_DEBUG_RIB_DETAILED
)
2697 route_entry_dump(p
, src_p
, re
);
2700 SET_FLAG(re
->status
, ROUTE_ENTRY_CHANGED
);
2701 rib_addnode(rn
, re
, 1);
2704 /* Free implicit route.*/
2706 rib_delnode(rn
, same
);
2710 route_unlock_node(rn
);
2714 void rib_delete(afi_t afi
, safi_t safi
, vrf_id_t vrf_id
, int type
,
2715 unsigned short instance
, int flags
, struct prefix
*p
,
2716 struct prefix_ipv6
*src_p
, const struct nexthop
*nh
,
2717 uint32_t table_id
, uint32_t metric
, uint8_t distance
,
2720 struct route_table
*table
;
2721 struct route_node
*rn
;
2722 struct route_entry
*re
;
2723 struct route_entry
*fib
= NULL
;
2724 struct route_entry
*same
= NULL
;
2725 struct nexthop
*rtnh
;
2726 char buf2
[INET6_ADDRSTRLEN
];
2729 assert(!src_p
|| !src_p
->prefixlen
|| afi
== AFI_IP6
);
2732 table
= zebra_vrf_table_with_table_id(afi
, safi
, vrf_id
, table_id
);
2739 apply_mask_ipv6(src_p
);
2741 /* Lookup route node. */
2742 rn
= srcdest_rnode_lookup(table
, p
, src_p
);
2744 char dst_buf
[PREFIX_STRLEN
], src_buf
[PREFIX_STRLEN
];
2746 prefix2str(p
, dst_buf
, sizeof(dst_buf
));
2747 if (src_p
&& src_p
->prefixlen
)
2748 prefix2str(src_p
, src_buf
, sizeof(src_buf
));
2752 if (IS_ZEBRA_DEBUG_RIB
) {
2753 struct vrf
*vrf
= vrf_lookup_by_id(vrf_id
);
2755 zlog_debug("%s[%d]:%s%s%s doesn't exist in rib",
2756 vrf
->name
, table_id
, dst_buf
,
2757 (src_buf
[0] != '\0') ? " from " : "",
2763 dest
= rib_dest_from_rnode(rn
);
2764 fib
= dest
->selected_fib
;
2766 /* Lookup same type route. */
2767 RNODE_FOREACH_RE (rn
, re
) {
2768 if (CHECK_FLAG(re
->status
, ROUTE_ENTRY_REMOVED
))
2771 if (re
->type
!= type
)
2773 if (re
->instance
!= instance
)
2775 if (CHECK_FLAG(re
->flags
, ZEBRA_FLAG_RR_USE_DISTANCE
) &&
2776 distance
!= re
->distance
)
2779 if (re
->type
== ZEBRA_ROUTE_KERNEL
&& re
->metric
!= metric
)
2781 if (re
->type
== ZEBRA_ROUTE_CONNECT
&& (rtnh
= re
->ng
.nexthop
)
2782 && rtnh
->type
== NEXTHOP_TYPE_IFINDEX
&& nh
) {
2783 if (rtnh
->ifindex
!= nh
->ifindex
)
2788 /* Make sure that the route found has the same gateway. */
2794 for (ALL_NEXTHOPS(re
->ng
, rtnh
))
2796 * No guarantee all kernel send nh with labels
2799 if (nexthop_same_no_labels(rtnh
, nh
)) {
2807 /* If same type of route can't be found and this message is from
2811 * In the past(HA!) we could get here because
2812 * we were receiving a route delete from the
2813 * kernel and we're not marking the proto
2814 * as coming from it's appropriate originator.
2815 * Now that we are properly noticing the fact
2816 * that the kernel has deleted our route we
2817 * are not going to get called in this path
2818 * I am going to leave this here because
2819 * this might still work this way on non-linux
2820 * platforms as well as some weird state I have
2821 * not properly thought of yet.
2822 * If we can show that this code path is
2823 * dead then we can remove it.
2825 if (fib
&& CHECK_FLAG(flags
, ZEBRA_FLAG_SELFROUTE
)) {
2826 if (IS_ZEBRA_DEBUG_RIB
) {
2827 rnode_debug(rn
, vrf_id
,
2828 "rn %p, re %p (%s) was deleted from kernel, adding",
2830 zebra_route_string(fib
->type
));
2833 UNSET_FLAG(fib
->status
, ROUTE_ENTRY_INSTALLED
);
2835 for (rtnh
= fib
->ng
.nexthop
; rtnh
;
2837 UNSET_FLAG(rtnh
->flags
,
2841 * This is a non FRR route
2842 * as such we should mark
2845 dest
->selected_fib
= NULL
;
2847 /* This means someone else, other than Zebra,
2849 * a Zebra router from the kernel. We will add
2851 rib_install_kernel(rn
, fib
, NULL
);
2854 if (IS_ZEBRA_DEBUG_RIB
) {
2858 "via %s ifindex %d type %d "
2859 "doesn't exist in rib",
2860 inet_ntop(afi2family(afi
),
2867 "type %d doesn't exist in rib",
2870 route_unlock_node(rn
);
2876 if (fromkernel
&& CHECK_FLAG(flags
, ZEBRA_FLAG_SELFROUTE
)
2878 rib_install_kernel(rn
, same
, NULL
);
2879 route_unlock_node(rn
);
2884 /* Special handling for IPv4 or IPv6 routes sourced from
2885 * EVPN - the nexthop (and associated MAC) need to be
2886 * uninstalled if no more refs.
2888 if (CHECK_FLAG(flags
, ZEBRA_FLAG_EVPN_ROUTE
)) {
2889 struct nexthop
*tmp_nh
;
2891 for (ALL_NEXTHOPS(re
->ng
, tmp_nh
)) {
2892 struct ipaddr vtep_ip
;
2894 memset(&vtep_ip
, 0, sizeof(struct ipaddr
));
2895 if (afi
== AFI_IP
) {
2896 vtep_ip
.ipa_type
= IPADDR_V4
;
2897 memcpy(&(vtep_ip
.ipaddr_v4
),
2898 &(tmp_nh
->gate
.ipv4
),
2899 sizeof(struct in_addr
));
2901 vtep_ip
.ipa_type
= IPADDR_V6
;
2902 memcpy(&(vtep_ip
.ipaddr_v6
),
2903 &(tmp_nh
->gate
.ipv6
),
2904 sizeof(struct in6_addr
));
2906 zebra_vxlan_evpn_vrf_route_del(re
->vrf_id
,
2911 /* Notify dplane if system route changes */
2912 if (RIB_SYSTEM_ROUTE(re
))
2913 dplane_sys_route_del(rn
, same
);
2915 rib_delnode(rn
, same
);
2918 route_unlock_node(rn
);
2923 int rib_add(afi_t afi
, safi_t safi
, vrf_id_t vrf_id
, int type
,
2924 unsigned short instance
, int flags
, struct prefix
*p
,
2925 struct prefix_ipv6
*src_p
, const struct nexthop
*nh
,
2926 uint32_t table_id
, uint32_t metric
, uint32_t mtu
, uint8_t distance
,
2929 struct route_entry
*re
;
2930 struct nexthop
*nexthop
;
2932 /* Allocate new route_entry structure. */
2933 re
= XCALLOC(MTYPE_RE
, sizeof(struct route_entry
));
2935 re
->instance
= instance
;
2936 re
->distance
= distance
;
2938 re
->metric
= metric
;
2940 re
->table
= table_id
;
2941 re
->vrf_id
= vrf_id
;
2942 re
->nexthop_num
= 0;
2943 re
->uptime
= monotime(NULL
);
2947 nexthop
= nexthop_new();
2949 route_entry_nexthop_add(re
, nexthop
);
2951 return rib_add_multipath(afi
, safi
, p
, src_p
, re
);
2954 /* Schedule routes of a particular table (address-family) based on event. */
2955 void rib_update_table(struct route_table
*table
, rib_update_event_t event
)
2957 struct route_node
*rn
;
2958 struct route_entry
*re
, *next
;
2960 /* Walk all routes and queue for processing, if appropriate for
2961 * the trigger event.
2963 for (rn
= route_top(table
); rn
; rn
= srcdest_route_next(rn
)) {
2965 * If we are looking at a route node and the node
2966 * has already been queued we don't
2967 * need to queue it up again
2969 if (rn
->info
&& CHECK_FLAG(rib_dest_from_rnode(rn
)->flags
,
2970 RIB_ROUTE_ANY_QUEUED
))
2973 case RIB_UPDATE_RMAP_CHANGE
:
2974 case RIB_UPDATE_OTHER
:
2975 /* Right now, examine all routes. Can restrict to a
2977 * some cases (TODO).
2979 if (rnode_to_ribs(rn
)) {
2980 RNODE_FOREACH_RE_SAFE (rn
, re
, next
)
2981 SET_FLAG(re
->status
,
2982 ROUTE_ENTRY_CHANGED
);
2993 /* RIB update function. */
2994 void rib_update(vrf_id_t vrf_id
, rib_update_event_t event
)
2996 struct route_table
*table
;
2998 /* Process routes of interested address-families. */
2999 table
= zebra_vrf_table(AFI_IP
, SAFI_UNICAST
, vrf_id
);
3001 if (IS_ZEBRA_DEBUG_EVENT
)
3002 zlog_debug("%s : AFI_IP event %d", __func__
, event
);
3003 rib_update_table(table
, event
);
3006 table
= zebra_vrf_table(AFI_IP6
, SAFI_UNICAST
, vrf_id
);
3008 if (IS_ZEBRA_DEBUG_EVENT
)
3009 zlog_debug("%s : AFI_IP6 event %d", __func__
, event
);
3010 rib_update_table(table
, event
);
3014 /* Delete self installed routes after zebra is relaunched. */
3015 void rib_sweep_table(struct route_table
*table
)
3017 struct route_node
*rn
;
3018 struct route_entry
*re
;
3019 struct route_entry
*next
;
3020 struct nexthop
*nexthop
;
3025 for (rn
= route_top(table
); rn
; rn
= srcdest_route_next(rn
)) {
3026 RNODE_FOREACH_RE_SAFE (rn
, re
, next
) {
3028 if (IS_ZEBRA_DEBUG_RIB
)
3029 route_entry_dump(&rn
->p
, NULL
, re
);
3031 if (CHECK_FLAG(re
->status
, ROUTE_ENTRY_REMOVED
))
3034 if (!CHECK_FLAG(re
->flags
, ZEBRA_FLAG_SELFROUTE
))
3038 * If routes are older than startup_time then
3039 * we know we read them in from the kernel.
3040 * As such we can safely remove them.
3042 if (zrouter
.startup_time
< re
->uptime
)
3046 * So we are starting up and have received
3047 * routes from the kernel that we have installed
3048 * from a previous run of zebra but not cleaned
3049 * up ( say a kill -9 )
3050 * But since we haven't actually installed
3051 * them yet( we received them from the kernel )
3052 * we don't think they are active.
3053 * So let's pretend they are active to actually
3055 * In all honesty I'm not sure if we should
3056 * mark them as active when we receive them
3057 * This is startup only so probably ok.
3059 * If we ever decide to move rib_sweep_table
3060 * to a different spot (ie startup )
3061 * this decision needs to be revisited
3063 SET_FLAG(re
->status
, ROUTE_ENTRY_INSTALLED
);
3064 for (ALL_NEXTHOPS(re
->ng
, nexthop
))
3065 SET_FLAG(nexthop
->flags
, NEXTHOP_FLAG_FIB
);
3067 rib_uninstall_kernel(rn
, re
);
3068 rib_delnode(rn
, re
);
3073 /* Sweep all RIB tables. */
3074 int rib_sweep_route(struct thread
*t
)
3077 struct zebra_vrf
*zvrf
;
3079 RB_FOREACH (vrf
, vrf_id_head
, &vrfs_by_id
) {
3080 if ((zvrf
= vrf
->info
) == NULL
)
3083 rib_sweep_table(zvrf
->table
[AFI_IP
][SAFI_UNICAST
]);
3084 rib_sweep_table(zvrf
->table
[AFI_IP6
][SAFI_UNICAST
]);
3087 zebra_router_sweep_route();
3092 /* Remove specific by protocol routes from 'table'. */
3093 unsigned long rib_score_proto_table(uint8_t proto
, unsigned short instance
,
3094 struct route_table
*table
)
3096 struct route_node
*rn
;
3097 struct route_entry
*re
;
3098 struct route_entry
*next
;
3099 unsigned long n
= 0;
3102 for (rn
= route_top(table
); rn
; rn
= srcdest_route_next(rn
))
3103 RNODE_FOREACH_RE_SAFE (rn
, re
, next
) {
3104 if (CHECK_FLAG(re
->status
, ROUTE_ENTRY_REMOVED
))
3106 if (re
->type
== proto
3107 && re
->instance
== instance
) {
3108 rib_delnode(rn
, re
);
3115 /* Remove specific by protocol routes. */
3116 unsigned long rib_score_proto(uint8_t proto
, unsigned short instance
)
3119 struct zebra_vrf
*zvrf
;
3120 struct other_route_table
*ort
;
3121 unsigned long cnt
= 0;
3123 RB_FOREACH (vrf
, vrf_id_head
, &vrfs_by_id
) {
3128 cnt
+= rib_score_proto_table(proto
, instance
,
3129 zvrf
->table
[AFI_IP
][SAFI_UNICAST
])
3130 + rib_score_proto_table(
3132 zvrf
->table
[AFI_IP6
][SAFI_UNICAST
]);
3134 frr_each(otable
, &zvrf
->other_tables
, ort
) cnt
+=
3135 rib_score_proto_table(proto
, instance
, ort
->table
);
3141 /* Close RIB and clean up kernel routes. */
3142 void rib_close_table(struct route_table
*table
)
3144 struct route_node
*rn
;
3145 rib_table_info_t
*info
;
3151 info
= route_table_get_info(table
);
3153 for (rn
= route_top(table
); rn
; rn
= srcdest_route_next(rn
)) {
3154 dest
= rib_dest_from_rnode(rn
);
3156 if (dest
&& dest
->selected_fib
) {
3157 if (info
->safi
== SAFI_UNICAST
)
3158 hook_call(rib_update
, rn
, NULL
);
3160 rib_uninstall_kernel(rn
, dest
->selected_fib
);
3161 dest
->selected_fib
= NULL
;
3167 * Handler for async dataplane results after a pseudowire installation
3169 static int handle_pw_result(struct zebra_dplane_ctx
*ctx
)
3171 struct zebra_pw
*pw
;
3172 struct zebra_vrf
*vrf
;
3174 /* The pseudowire code assumes success - we act on an error
3175 * result for installation attempts here.
3177 if (dplane_ctx_get_op(ctx
) != DPLANE_OP_PW_INSTALL
)
3180 if (dplane_ctx_get_status(ctx
) != ZEBRA_DPLANE_REQUEST_SUCCESS
) {
3181 vrf
= zebra_vrf_lookup_by_id(dplane_ctx_get_vrf(ctx
));
3182 pw
= zebra_pw_find(vrf
, dplane_ctx_get_ifname(ctx
));
3184 zebra_pw_install_failure(pw
);
3194 * Handle results from the dataplane system. Dequeue update context
3195 * structs, dispatch to appropriate internal handlers.
3197 static int rib_process_dplane_results(struct thread
*thread
)
3199 struct zebra_dplane_ctx
*ctx
;
3200 struct dplane_ctx_q ctxlist
;
3202 /* Dequeue a list of completed updates with one lock/unlock cycle */
3205 TAILQ_INIT(&ctxlist
);
3207 /* Take lock controlling queue of results */
3208 frr_with_mutex(&dplane_mutex
) {
3209 /* Dequeue list of context structs */
3210 dplane_ctx_list_append(&ctxlist
, &rib_dplane_q
);
3213 /* Dequeue context block */
3214 ctx
= dplane_ctx_dequeue(&ctxlist
);
3216 /* If we've emptied the results queue, we're done */
3221 switch (dplane_ctx_get_op(ctx
)) {
3222 case DPLANE_OP_ROUTE_INSTALL
:
3223 case DPLANE_OP_ROUTE_UPDATE
:
3224 case DPLANE_OP_ROUTE_DELETE
:
3226 /* Bit of special case for route updates
3227 * that were generated by async notifications:
3228 * we don't want to continue processing these
3231 if (dplane_ctx_get_notif_provider(ctx
) == 0)
3232 rib_process_result(ctx
);
3234 dplane_ctx_fini(&ctx
);
3238 case DPLANE_OP_ROUTE_NOTIFY
:
3239 rib_process_dplane_notify(ctx
);
3242 case DPLANE_OP_LSP_INSTALL
:
3243 case DPLANE_OP_LSP_UPDATE
:
3244 case DPLANE_OP_LSP_DELETE
:
3246 /* Bit of special case for LSP updates
3247 * that were generated by async notifications:
3248 * we don't want to continue processing these.
3250 if (dplane_ctx_get_notif_provider(ctx
) == 0)
3251 zebra_mpls_lsp_dplane_result(ctx
);
3253 dplane_ctx_fini(&ctx
);
3257 case DPLANE_OP_LSP_NOTIFY
:
3258 zebra_mpls_process_dplane_notify(ctx
);
3261 case DPLANE_OP_PW_INSTALL
:
3262 case DPLANE_OP_PW_UNINSTALL
:
3263 handle_pw_result(ctx
);
3266 case DPLANE_OP_SYS_ROUTE_ADD
:
3267 case DPLANE_OP_SYS_ROUTE_DELETE
:
3268 /* No further processing in zebra for these. */
3269 dplane_ctx_fini(&ctx
);
3272 case DPLANE_OP_MAC_INSTALL
:
3273 case DPLANE_OP_MAC_DELETE
:
3274 zebra_vxlan_handle_result(ctx
);
3277 /* Some op codes not handled here */
3278 case DPLANE_OP_ADDR_INSTALL
:
3279 case DPLANE_OP_ADDR_UNINSTALL
:
3280 case DPLANE_OP_NEIGH_INSTALL
:
3281 case DPLANE_OP_NEIGH_UPDATE
:
3282 case DPLANE_OP_NEIGH_DELETE
:
3283 case DPLANE_OP_VTEP_ADD
:
3284 case DPLANE_OP_VTEP_DELETE
:
3285 case DPLANE_OP_NONE
:
3286 /* Don't expect this: just return the struct? */
3287 dplane_ctx_fini(&ctx
);
3290 } /* Dispatch by op code */
3292 ctx
= dplane_ctx_dequeue(&ctxlist
);
3297 /* Check for nexthop tracking processing after finishing with results */
3298 do_nht_processing();
3304 * Results are returned from the dataplane subsystem, in the context of
3305 * the dataplane pthread. We enqueue the results here for processing by
3306 * the main thread later.
3308 static int rib_dplane_results(struct dplane_ctx_q
*ctxlist
)
3310 /* Take lock controlling queue of results */
3311 frr_with_mutex(&dplane_mutex
) {
3312 /* Enqueue context blocks */
3313 dplane_ctx_list_append(&rib_dplane_q
, ctxlist
);
3316 /* Ensure event is signalled to zebra main pthread */
3317 thread_add_event(zrouter
.master
, rib_process_dplane_results
, NULL
, 0,
3324 * Ensure there are no empty slots in the route_info array.
3325 * Every route type in zebra should be present there.
3327 static void check_route_info(void)
3329 int len
= array_size(route_info
);
3332 * ZEBRA_ROUTE_SYSTEM is special cased since
3333 * its key is 0 anyway.
3335 * ZEBRA_ROUTE_ALL is also ignored.
3337 for (int i
= 0; i
< len
; i
++) {
3338 if (i
== ZEBRA_ROUTE_SYSTEM
|| i
== ZEBRA_ROUTE_ALL
)
3340 assert(route_info
[i
].key
);
3341 assert(route_info
[i
].meta_q_map
< MQ_SIZE
);
3345 /* Routing information base initialize. */
3352 /* Init dataplane, and register for results */
3353 pthread_mutex_init(&dplane_mutex
, NULL
);
3354 TAILQ_INIT(&rib_dplane_q
);
3355 zebra_dplane_init(rib_dplane_results
);
3361 * Get the first vrf id that is greater than the given vrf id if any.
3363 * Returns true if a vrf id was found, false otherwise.
3365 static inline int vrf_id_get_next(vrf_id_t vrf_id
, vrf_id_t
*next_id_p
)
3369 vrf
= vrf_lookup_by_id(vrf_id
);
3371 vrf
= RB_NEXT(vrf_id_head
, vrf
);
3373 *next_id_p
= vrf
->vrf_id
;
3382 * rib_tables_iter_next
3384 * Returns the next table in the iteration.
3386 struct route_table
*rib_tables_iter_next(rib_tables_iter_t
*iter
)
3388 struct route_table
*table
;
3391 * Array that helps us go over all AFI/SAFI combinations via one
3398 {AFI_IP
, SAFI_UNICAST
}, {AFI_IP
, SAFI_MULTICAST
},
3399 {AFI_IP
, SAFI_LABELED_UNICAST
}, {AFI_IP6
, SAFI_UNICAST
},
3400 {AFI_IP6
, SAFI_MULTICAST
}, {AFI_IP6
, SAFI_LABELED_UNICAST
},
3405 switch (iter
->state
) {
3407 case RIB_TABLES_ITER_S_INIT
:
3408 iter
->vrf_id
= VRF_DEFAULT
;
3409 iter
->afi_safi_ix
= -1;
3413 case RIB_TABLES_ITER_S_ITERATING
:
3414 iter
->afi_safi_ix
++;
3417 while (iter
->afi_safi_ix
3418 < (int)array_size(afi_safis
)) {
3419 table
= zebra_vrf_table(
3420 afi_safis
[iter
->afi_safi_ix
].afi
,
3421 afi_safis
[iter
->afi_safi_ix
].safi
,
3426 iter
->afi_safi_ix
++;
3430 * Found another table in this vrf.
3436 * Done with all tables in the current vrf, go to the
3440 if (!vrf_id_get_next(iter
->vrf_id
, &iter
->vrf_id
))
3443 iter
->afi_safi_ix
= 0;
3448 case RIB_TABLES_ITER_S_DONE
:
3453 iter
->state
= RIB_TABLES_ITER_S_ITERATING
;
3455 iter
->state
= RIB_TABLES_ITER_S_DONE
;