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"
61 DEFINE_MTYPE_STATIC(ZEBRA
, RIB_UPDATE_CTX
, "Rib update context object");
64 * Event, list, and mutex for delivery of dataplane results
66 static pthread_mutex_t dplane_mutex
;
67 static struct thread
*t_dplane
;
68 static struct dplane_ctx_q rib_dplane_q
;
70 DEFINE_HOOK(rib_update
, (struct route_node
* rn
, const char *reason
),
73 /* Should we allow non Quagga processes to delete our routes */
74 extern int allow_delete
;
76 /* Each route type's string and default distance value. */
81 } route_info
[ZEBRA_ROUTE_MAX
] = {
82 [ZEBRA_ROUTE_SYSTEM
] = {ZEBRA_ROUTE_SYSTEM
, 0, 4},
83 [ZEBRA_ROUTE_KERNEL
] = {ZEBRA_ROUTE_KERNEL
, 0, 0},
84 [ZEBRA_ROUTE_CONNECT
] = {ZEBRA_ROUTE_CONNECT
, 0, 0},
85 [ZEBRA_ROUTE_STATIC
] = {ZEBRA_ROUTE_STATIC
, 1, 1},
86 [ZEBRA_ROUTE_RIP
] = {ZEBRA_ROUTE_RIP
, 120, 2},
87 [ZEBRA_ROUTE_RIPNG
] = {ZEBRA_ROUTE_RIPNG
, 120, 2},
88 [ZEBRA_ROUTE_OSPF
] = {ZEBRA_ROUTE_OSPF
, 110, 2},
89 [ZEBRA_ROUTE_OSPF6
] = {ZEBRA_ROUTE_OSPF6
, 110, 2},
90 [ZEBRA_ROUTE_ISIS
] = {ZEBRA_ROUTE_ISIS
, 115, 2},
91 [ZEBRA_ROUTE_BGP
] = {ZEBRA_ROUTE_BGP
, 20 /* IBGP is 200. */, 3},
92 [ZEBRA_ROUTE_PIM
] = {ZEBRA_ROUTE_PIM
, 255, 4},
93 [ZEBRA_ROUTE_EIGRP
] = {ZEBRA_ROUTE_EIGRP
, 90, 2},
94 [ZEBRA_ROUTE_NHRP
] = {ZEBRA_ROUTE_NHRP
, 10, 2},
95 [ZEBRA_ROUTE_HSLS
] = {ZEBRA_ROUTE_HSLS
, 255, 4},
96 [ZEBRA_ROUTE_OLSR
] = {ZEBRA_ROUTE_OLSR
, 255, 4},
97 [ZEBRA_ROUTE_TABLE
] = {ZEBRA_ROUTE_TABLE
, 150, 1},
98 [ZEBRA_ROUTE_LDP
] = {ZEBRA_ROUTE_LDP
, 150, 4},
99 [ZEBRA_ROUTE_VNC
] = {ZEBRA_ROUTE_VNC
, 20, 3},
100 [ZEBRA_ROUTE_VNC_DIRECT
] = {ZEBRA_ROUTE_VNC_DIRECT
, 20, 3},
101 [ZEBRA_ROUTE_VNC_DIRECT_RH
] = {ZEBRA_ROUTE_VNC_DIRECT_RH
, 20, 3},
102 [ZEBRA_ROUTE_BGP_DIRECT
] = {ZEBRA_ROUTE_BGP_DIRECT
, 20, 3},
103 [ZEBRA_ROUTE_BGP_DIRECT_EXT
] = {ZEBRA_ROUTE_BGP_DIRECT_EXT
, 20, 3},
104 [ZEBRA_ROUTE_BABEL
] = {ZEBRA_ROUTE_BABEL
, 100, 2},
105 [ZEBRA_ROUTE_SHARP
] = {ZEBRA_ROUTE_SHARP
, 150, 4},
106 [ZEBRA_ROUTE_PBR
] = {ZEBRA_ROUTE_PBR
, 200, 4},
107 [ZEBRA_ROUTE_BFD
] = {ZEBRA_ROUTE_BFD
, 255, 4},
108 [ZEBRA_ROUTE_OPENFABRIC
] = {ZEBRA_ROUTE_OPENFABRIC
, 115, 2},
109 [ZEBRA_ROUTE_VRRP
] = {ZEBRA_ROUTE_VRRP
, 255, 4}
110 /* Any new route type added to zebra, should be mirrored here */
112 /* no entry/default: 150 */
115 static void __attribute__((format(printf
, 5, 6)))
116 _rnode_zlog(const char *_func
, vrf_id_t vrf_id
, struct route_node
*rn
,
117 int priority
, const char *msgfmt
, ...)
119 char buf
[SRCDEST2STR_BUFFER
+ sizeof(" (MRIB)")];
123 va_start(ap
, msgfmt
);
124 vsnprintf(msgbuf
, sizeof(msgbuf
), msgfmt
, ap
);
128 rib_table_info_t
*info
= srcdest_rnode_table_info(rn
);
129 srcdest_rnode2str(rn
, buf
, sizeof(buf
));
131 if (info
->safi
== SAFI_MULTICAST
)
132 strlcat(buf
, " (MRIB)", sizeof(buf
));
134 snprintf(buf
, sizeof(buf
), "{(route_node *) NULL}");
137 zlog(priority
, "%s: %d:%s: %s", _func
, vrf_id
, buf
, msgbuf
);
140 #define rnode_debug(node, vrf_id, ...) \
141 _rnode_zlog(__func__, vrf_id, node, LOG_DEBUG, __VA_ARGS__)
142 #define rnode_info(node, ...) \
143 _rnode_zlog(__func__, vrf_id, node, LOG_INFO, __VA_ARGS__)
145 uint8_t route_distance(int type
)
149 if ((unsigned)type
>= array_size(route_info
))
152 distance
= route_info
[type
].distance
;
157 int is_zebra_valid_kernel_table(uint32_t table_id
)
160 if ((table_id
== RT_TABLE_UNSPEC
) || (table_id
== RT_TABLE_LOCAL
)
161 || (table_id
== RT_TABLE_COMPAT
))
168 int is_zebra_main_routing_table(uint32_t table_id
)
170 if (table_id
== RT_TABLE_MAIN
)
175 int zebra_check_addr(const struct prefix
*p
)
177 if (p
->family
== AF_INET
) {
180 addr
= p
->u
.prefix4
.s_addr
;
183 if (IPV4_NET127(addr
) || IN_CLASSD(addr
)
184 || IPV4_LINKLOCAL(addr
))
187 if (p
->family
== AF_INET6
) {
188 if (IN6_IS_ADDR_LOOPBACK(&p
->u
.prefix6
))
190 if (IN6_IS_ADDR_LINKLOCAL(&p
->u
.prefix6
))
196 /* Add nexthop to the end of a rib node's nexthop list */
197 void route_entry_nexthop_add(struct route_entry
*re
, struct nexthop
*nexthop
)
199 _nexthop_group_add_sorted(re
->ng
, nexthop
);
204 * copy_nexthop - copy a nexthop to the rib structure.
206 void route_entry_copy_nexthops(struct route_entry
*re
, struct nexthop
*nh
)
208 assert(!re
->ng
->nexthop
);
209 copy_nexthops(&re
->ng
->nexthop
, nh
, NULL
);
212 /* Delete specified nexthop from the list. */
213 void route_entry_nexthop_delete(struct route_entry
*re
, struct nexthop
*nexthop
)
216 nexthop
->next
->prev
= nexthop
->prev
;
218 nexthop
->prev
->next
= nexthop
->next
;
220 re
->ng
->nexthop
= nexthop
->next
;
224 struct nexthop
*route_entry_nexthop_ifindex_add(struct route_entry
*re
,
228 struct nexthop
*nexthop
;
230 nexthop
= nexthop_new();
231 nexthop
->type
= NEXTHOP_TYPE_IFINDEX
;
232 nexthop
->ifindex
= ifindex
;
233 nexthop
->vrf_id
= nh_vrf_id
;
235 route_entry_nexthop_add(re
, nexthop
);
240 struct nexthop
*route_entry_nexthop_ipv4_add(struct route_entry
*re
,
241 struct in_addr
*ipv4
,
245 struct nexthop
*nexthop
;
247 nexthop
= nexthop_new();
248 nexthop
->type
= NEXTHOP_TYPE_IPV4
;
249 nexthop
->vrf_id
= nh_vrf_id
;
250 nexthop
->gate
.ipv4
= *ipv4
;
252 nexthop
->src
.ipv4
= *src
;
254 route_entry_nexthop_add(re
, nexthop
);
259 struct nexthop
*route_entry_nexthop_ipv4_ifindex_add(struct route_entry
*re
,
260 struct in_addr
*ipv4
,
265 struct nexthop
*nexthop
;
266 struct interface
*ifp
;
268 nexthop
= nexthop_new();
269 nexthop
->vrf_id
= nh_vrf_id
;
270 nexthop
->type
= NEXTHOP_TYPE_IPV4_IFINDEX
;
271 nexthop
->gate
.ipv4
= *ipv4
;
273 nexthop
->src
.ipv4
= *src
;
274 nexthop
->ifindex
= ifindex
;
275 ifp
= if_lookup_by_index(nexthop
->ifindex
, nh_vrf_id
);
276 /*Pending: need to think if null ifp here is ok during bootup?
277 There was a crash because ifp here was coming to be NULL */
279 if (connected_is_unnumbered(ifp
))
280 SET_FLAG(nexthop
->flags
, NEXTHOP_FLAG_ONLINK
);
282 route_entry_nexthop_add(re
, nexthop
);
287 struct nexthop
*route_entry_nexthop_ipv6_add(struct route_entry
*re
,
288 struct in6_addr
*ipv6
,
291 struct nexthop
*nexthop
;
293 nexthop
= nexthop_new();
294 nexthop
->vrf_id
= nh_vrf_id
;
295 nexthop
->type
= NEXTHOP_TYPE_IPV6
;
296 nexthop
->gate
.ipv6
= *ipv6
;
298 route_entry_nexthop_add(re
, nexthop
);
303 struct nexthop
*route_entry_nexthop_ipv6_ifindex_add(struct route_entry
*re
,
304 struct in6_addr
*ipv6
,
308 struct nexthop
*nexthop
;
310 nexthop
= nexthop_new();
311 nexthop
->vrf_id
= nh_vrf_id
;
312 nexthop
->type
= NEXTHOP_TYPE_IPV6_IFINDEX
;
313 nexthop
->gate
.ipv6
= *ipv6
;
314 nexthop
->ifindex
= ifindex
;
316 route_entry_nexthop_add(re
, nexthop
);
321 struct nexthop
*route_entry_nexthop_blackhole_add(struct route_entry
*re
,
322 enum blackhole_type bh_type
)
324 struct nexthop
*nexthop
;
326 nexthop
= nexthop_new();
327 nexthop
->vrf_id
= VRF_DEFAULT
;
328 nexthop
->type
= NEXTHOP_TYPE_BLACKHOLE
;
329 nexthop
->bh_type
= bh_type
;
331 route_entry_nexthop_add(re
, nexthop
);
336 struct route_entry
*rib_match(afi_t afi
, safi_t safi
, vrf_id_t vrf_id
,
337 union g_addr
*addr
, struct route_node
**rn_out
)
340 struct route_table
*table
;
341 struct route_node
*rn
;
342 struct route_entry
*match
= NULL
;
345 table
= zebra_vrf_table(afi
, safi
, vrf_id
);
349 memset(&p
, 0, sizeof(struct prefix
));
352 p
.u
.prefix4
= addr
->ipv4
;
353 p
.prefixlen
= IPV4_MAX_PREFIXLEN
;
355 p
.u
.prefix6
= addr
->ipv6
;
356 p
.prefixlen
= IPV6_MAX_PREFIXLEN
;
359 rn
= route_node_match(table
, (struct prefix
*)&p
);
364 route_unlock_node(rn
);
366 dest
= rib_dest_from_rnode(rn
);
367 if (dest
&& dest
->selected_fib
368 && !CHECK_FLAG(dest
->selected_fib
->status
,
369 ROUTE_ENTRY_REMOVED
))
370 match
= dest
->selected_fib
;
372 /* If there is no selected route or matched route is EGP, go up
377 } while (rn
&& rn
->info
== NULL
);
381 if (match
->type
!= ZEBRA_ROUTE_CONNECT
) {
382 if (!CHECK_FLAG(match
->status
,
383 ROUTE_ENTRY_INSTALLED
))
395 struct route_entry
*rib_match_ipv4_multicast(vrf_id_t vrf_id
,
397 struct route_node
**rn_out
)
399 struct route_entry
*re
= NULL
, *mre
= NULL
, *ure
= NULL
;
400 struct route_node
*m_rn
= NULL
, *u_rn
= NULL
;
401 union g_addr gaddr
= {.ipv4
= addr
};
403 switch (zrouter
.ipv4_multicast_mode
) {
404 case MCAST_MRIB_ONLY
:
405 return rib_match(AFI_IP
, SAFI_MULTICAST
, vrf_id
, &gaddr
,
407 case MCAST_URIB_ONLY
:
408 return rib_match(AFI_IP
, SAFI_UNICAST
, vrf_id
, &gaddr
, rn_out
);
409 case MCAST_NO_CONFIG
:
410 case MCAST_MIX_MRIB_FIRST
:
411 re
= mre
= rib_match(AFI_IP
, SAFI_MULTICAST
, vrf_id
, &gaddr
,
414 re
= ure
= rib_match(AFI_IP
, SAFI_UNICAST
, vrf_id
,
417 case MCAST_MIX_DISTANCE
:
418 mre
= rib_match(AFI_IP
, SAFI_MULTICAST
, vrf_id
, &gaddr
, &m_rn
);
419 ure
= rib_match(AFI_IP
, SAFI_UNICAST
, vrf_id
, &gaddr
, &u_rn
);
421 re
= ure
->distance
< mre
->distance
? ure
: mre
;
427 case MCAST_MIX_PFXLEN
:
428 mre
= rib_match(AFI_IP
, SAFI_MULTICAST
, vrf_id
, &gaddr
, &m_rn
);
429 ure
= rib_match(AFI_IP
, SAFI_UNICAST
, vrf_id
, &gaddr
, &u_rn
);
431 re
= u_rn
->p
.prefixlen
> m_rn
->p
.prefixlen
? ure
: mre
;
440 *rn_out
= (re
== mre
) ? m_rn
: u_rn
;
442 if (IS_ZEBRA_DEBUG_RIB
) {
444 inet_ntop(AF_INET
, &addr
, buf
, BUFSIZ
);
446 zlog_debug("%s: %s: vrf: %u found %s, using %s",
447 __func__
, buf
, vrf_id
,
448 mre
? (ure
? "MRIB+URIB" : "MRIB")
449 : ure
? "URIB" : "nothing",
450 re
== ure
? "URIB" : re
== mre
? "MRIB" : "none");
455 struct route_entry
*rib_lookup_ipv4(struct prefix_ipv4
*p
, vrf_id_t vrf_id
)
457 struct route_table
*table
;
458 struct route_node
*rn
;
459 struct route_entry
*match
= NULL
;
463 table
= zebra_vrf_table(AFI_IP
, SAFI_UNICAST
, vrf_id
);
467 rn
= route_node_lookup(table
, (struct prefix
*)p
);
469 /* No route for this prefix. */
474 route_unlock_node(rn
);
475 dest
= rib_dest_from_rnode(rn
);
477 if (dest
&& dest
->selected_fib
478 && !CHECK_FLAG(dest
->selected_fib
->status
, ROUTE_ENTRY_REMOVED
))
479 match
= dest
->selected_fib
;
484 if (match
->type
== ZEBRA_ROUTE_CONNECT
)
487 if (CHECK_FLAG(match
->status
, ROUTE_ENTRY_INSTALLED
))
494 * Is this RIB labeled-unicast? It must be of type BGP and all paths
495 * (nexthops) must have a label.
497 int zebra_rib_labeled_unicast(struct route_entry
*re
)
499 struct nexthop
*nexthop
= NULL
;
501 if (re
->type
!= ZEBRA_ROUTE_BGP
)
504 for (ALL_NEXTHOPS_PTR(re
->ng
, nexthop
))
505 if (!nexthop
->nh_label
|| !nexthop
->nh_label
->num_labels
)
511 /* Update flag indicates whether this is a "replace" or not. Currently, this
512 * is only used for IPv4.
514 void rib_install_kernel(struct route_node
*rn
, struct route_entry
*re
,
515 struct route_entry
*old
)
517 struct nexthop
*nexthop
;
518 rib_table_info_t
*info
= srcdest_rnode_table_info(rn
);
519 struct zebra_vrf
*zvrf
= vrf_info_lookup(re
->vrf_id
);
520 const struct prefix
*p
, *src_p
;
521 enum zebra_dplane_result ret
;
523 rib_dest_t
*dest
= rib_dest_from_rnode(rn
);
525 srcdest_rnode_prefixes(rn
, &p
, &src_p
);
527 if (info
->safi
!= SAFI_UNICAST
) {
528 for (ALL_NEXTHOPS_PTR(re
->ng
, nexthop
))
529 SET_FLAG(nexthop
->flags
, NEXTHOP_FLAG_FIB
);
532 struct nexthop
*prev
;
534 for (ALL_NEXTHOPS_PTR(re
->ng
, nexthop
)) {
535 UNSET_FLAG (nexthop
->flags
, NEXTHOP_FLAG_DUPLICATE
);
536 for (ALL_NEXTHOPS_PTR(re
->ng
, prev
)) {
539 if (nexthop_same_firsthop(nexthop
, prev
)) {
540 SET_FLAG(nexthop
->flags
,
541 NEXTHOP_FLAG_DUPLICATE
);
549 * If this is a replace to a new RE let the originator of the RE
550 * know that they've lost
552 if (old
&& (old
!= re
) && (old
->type
!= re
->type
))
553 zsend_route_notify_owner(old
, p
, ZAPI_ROUTE_BETTER_ADMIN_WON
);
555 /* Update fib selection */
556 dest
->selected_fib
= re
;
559 * Make sure we update the FPM any time we send new information to
562 hook_call(rib_update
, rn
, "installing in kernel");
564 /* Send add or update */
566 ret
= dplane_route_update(rn
, re
, old
);
568 ret
= dplane_route_add(rn
, re
);
571 case ZEBRA_DPLANE_REQUEST_QUEUED
:
572 SET_FLAG(re
->status
, ROUTE_ENTRY_QUEUED
);
575 SET_FLAG(old
->status
, ROUTE_ENTRY_QUEUED
);
577 /* Free old FIB nexthop group */
578 if (old
->fib_ng
.nexthop
) {
579 nexthops_free(old
->fib_ng
.nexthop
);
580 old
->fib_ng
.nexthop
= NULL
;
583 if (!RIB_SYSTEM_ROUTE(old
)) {
584 /* Clear old route's FIB flags */
585 for (ALL_NEXTHOPS_PTR(old
->ng
, nexthop
)) {
586 UNSET_FLAG(nexthop
->flags
,
593 zvrf
->installs_queued
++;
595 case ZEBRA_DPLANE_REQUEST_FAILURE
:
597 char str
[SRCDEST2STR_BUFFER
];
599 srcdest_rnode2str(rn
, str
, sizeof(str
));
600 flog_err(EC_ZEBRA_DP_INSTALL_FAIL
,
601 "%u:%s: Failed to enqueue dataplane install",
605 case ZEBRA_DPLANE_REQUEST_SUCCESS
:
614 /* Uninstall the route from kernel. */
615 void rib_uninstall_kernel(struct route_node
*rn
, struct route_entry
*re
)
617 struct nexthop
*nexthop
;
618 rib_table_info_t
*info
= srcdest_rnode_table_info(rn
);
619 struct zebra_vrf
*zvrf
= vrf_info_lookup(re
->vrf_id
);
621 if (info
->safi
!= SAFI_UNICAST
) {
622 UNSET_FLAG(re
->status
, ROUTE_ENTRY_INSTALLED
);
623 for (ALL_NEXTHOPS_PTR(re
->ng
, nexthop
))
624 UNSET_FLAG(nexthop
->flags
, NEXTHOP_FLAG_FIB
);
629 * Make sure we update the FPM any time we send new information to
632 hook_call(rib_update
, rn
, "uninstalling from kernel");
634 switch (dplane_route_delete(rn
, re
)) {
635 case ZEBRA_DPLANE_REQUEST_QUEUED
:
637 zvrf
->removals_queued
++;
639 case ZEBRA_DPLANE_REQUEST_FAILURE
:
641 char str
[SRCDEST2STR_BUFFER
];
643 srcdest_rnode2str(rn
, str
, sizeof(str
));
644 flog_err(EC_ZEBRA_DP_INSTALL_FAIL
,
645 "%u:%s: Failed to enqueue dataplane uninstall",
649 case ZEBRA_DPLANE_REQUEST_SUCCESS
:
658 /* Uninstall the route from kernel. */
659 static void rib_uninstall(struct route_node
*rn
, struct route_entry
*re
)
661 rib_table_info_t
*info
= srcdest_rnode_table_info(rn
);
662 rib_dest_t
*dest
= rib_dest_from_rnode(rn
);
663 struct nexthop
*nexthop
;
665 if (dest
&& dest
->selected_fib
== re
) {
666 if (info
->safi
== SAFI_UNICAST
)
667 hook_call(rib_update
, rn
, "rib_uninstall");
669 /* If labeled-unicast route, uninstall transit LSP. */
670 if (zebra_rib_labeled_unicast(re
))
671 zebra_mpls_lsp_uninstall(info
->zvrf
, rn
, re
);
673 rib_uninstall_kernel(rn
, re
);
675 dest
->selected_fib
= NULL
;
677 /* Free FIB nexthop group, if present */
678 if (re
->fib_ng
.nexthop
) {
679 nexthops_free(re
->fib_ng
.nexthop
);
680 re
->fib_ng
.nexthop
= NULL
;
683 for (ALL_NEXTHOPS_PTR(re
->ng
, nexthop
))
684 UNSET_FLAG(nexthop
->flags
, NEXTHOP_FLAG_FIB
);
687 if (CHECK_FLAG(re
->flags
, ZEBRA_FLAG_SELECTED
)) {
688 const struct prefix
*p
, *src_p
;
690 srcdest_rnode_prefixes(rn
, &p
, &src_p
);
692 redistribute_delete(p
, src_p
, re
, NULL
);
693 UNSET_FLAG(re
->flags
, ZEBRA_FLAG_SELECTED
);
698 * rib_can_delete_dest
700 * Returns true if the given dest can be deleted from the table.
702 static int rib_can_delete_dest(rib_dest_t
*dest
)
704 if (re_list_first(&dest
->routes
)) {
709 * Unresolved rnh's are stored on the default route's list
711 * dest->rnode can also be the source prefix node in an
712 * ipv6 sourcedest table. Fortunately the prefix of a
713 * source prefix node can never be the default prefix.
715 if (is_default_prefix(&dest
->rnode
->p
))
719 * Don't delete the dest if we have to update the FPM about this
722 if (CHECK_FLAG(dest
->flags
, RIB_DEST_UPDATE_FPM
)
723 || CHECK_FLAG(dest
->flags
, RIB_DEST_SENT_TO_FPM
))
729 void zebra_rib_evaluate_rn_nexthops(struct route_node
*rn
, uint32_t seq
)
731 rib_dest_t
*dest
= rib_dest_from_rnode(rn
);
735 * We are storing the rnh's associated withb
736 * the tracked nexthop as a list of the rn's.
737 * Unresolved rnh's are placed at the top
738 * of the tree list.( 0.0.0.0/0 for v4 and 0::0/0 for v6 )
739 * As such for each rn we need to walk up the tree
740 * and see if any rnh's need to see if they
741 * would match a more specific route
744 if (IS_ZEBRA_DEBUG_NHT_DETAILED
) {
745 char buf
[PREFIX_STRLEN
];
747 zlog_debug("%s: %s Being examined for Nexthop Tracking Count: %zd",
749 srcdest_rnode2str(rn
, buf
, sizeof(buf
)),
750 dest
? rnh_list_count(&dest
->nht
) : 0);
755 dest
= rib_dest_from_rnode(rn
);
759 * If we have any rnh's stored in the nht list
760 * then we know that this route node was used for
761 * nht resolution and as such we need to call the
762 * nexthop tracking evaluation code
764 frr_each_safe(rnh_list
, &dest
->nht
, rnh
) {
765 struct zebra_vrf
*zvrf
=
766 zebra_vrf_lookup_by_id(rnh
->vrf_id
);
767 struct prefix
*p
= &rnh
->node
->p
;
769 if (IS_ZEBRA_DEBUG_NHT_DETAILED
) {
770 char buf1
[PREFIX_STRLEN
];
771 char buf2
[PREFIX_STRLEN
];
773 zlog_debug("%u:%s has Nexthop(%s) Type: %s depending on it, evaluating %u:%u",
775 srcdest_rnode2str(rn
, buf1
,
777 prefix2str(p
, buf2
, sizeof(buf2
)),
778 rnh_type2str(rnh
->type
),
783 * If we have evaluated this node on this pass
784 * already, due to following the tree up
785 * then we know that we can move onto the next
788 * Additionally we call zebra_evaluate_rnh
789 * when we gc the dest. In this case we know
790 * that there must be no other re's where
791 * we were originally as such we know that
792 * that sequence number is ok to respect.
794 if (rnh
->seqno
== seq
) {
795 if (IS_ZEBRA_DEBUG_NHT_DETAILED
)
797 "\tNode processed and moved already");
802 zebra_evaluate_rnh(zvrf
, family2afi(p
->family
), 0,
808 dest
= rib_dest_from_rnode(rn
);
815 * Garbage collect the rib dest corresponding to the given route node
818 * Returns true if the dest was deleted, false otherwise.
820 int rib_gc_dest(struct route_node
*rn
)
824 dest
= rib_dest_from_rnode(rn
);
828 if (!rib_can_delete_dest(dest
))
831 if (IS_ZEBRA_DEBUG_RIB
) {
832 struct zebra_vrf
*zvrf
;
834 zvrf
= rib_dest_vrf(dest
);
835 rnode_debug(rn
, zvrf_id(zvrf
), "removing dest from table");
838 zebra_rib_evaluate_rn_nexthops(rn
, zebra_router_get_next_sequence());
841 rnh_list_fini(&dest
->nht
);
842 XFREE(MTYPE_RIB_DEST
, dest
);
846 * Release the one reference that we keep on the route node.
848 route_unlock_node(rn
);
852 static void rib_process_add_fib(struct zebra_vrf
*zvrf
, struct route_node
*rn
,
853 struct route_entry
*new)
855 hook_call(rib_update
, rn
, "new route selected");
857 /* Update real nexthop. This may actually determine if nexthop is active
859 if (!nexthop_group_active_nexthop_num(new->ng
)) {
860 UNSET_FLAG(new->status
, ROUTE_ENTRY_CHANGED
);
864 if (IS_ZEBRA_DEBUG_RIB
) {
865 char buf
[SRCDEST2STR_BUFFER
];
866 srcdest_rnode2str(rn
, buf
, sizeof(buf
));
867 zlog_debug("%u:%s: Adding route rn %p, re %p (%s)",
868 zvrf_id(zvrf
), buf
, rn
, new,
869 zebra_route_string(new->type
));
872 /* If labeled-unicast route, install transit LSP. */
873 if (zebra_rib_labeled_unicast(new))
874 zebra_mpls_lsp_install(zvrf
, rn
, new);
876 rib_install_kernel(rn
, new, NULL
);
878 UNSET_FLAG(new->status
, ROUTE_ENTRY_CHANGED
);
881 static void rib_process_del_fib(struct zebra_vrf
*zvrf
, struct route_node
*rn
,
882 struct route_entry
*old
)
884 hook_call(rib_update
, rn
, "removing existing route");
886 /* Uninstall from kernel. */
887 if (IS_ZEBRA_DEBUG_RIB
) {
888 char buf
[SRCDEST2STR_BUFFER
];
889 srcdest_rnode2str(rn
, buf
, sizeof(buf
));
890 zlog_debug("%u:%s: Deleting route rn %p, re %p (%s)",
891 zvrf_id(zvrf
), buf
, rn
, old
,
892 zebra_route_string(old
->type
));
895 /* If labeled-unicast route, uninstall transit LSP. */
896 if (zebra_rib_labeled_unicast(old
))
897 zebra_mpls_lsp_uninstall(zvrf
, rn
, old
);
899 rib_uninstall_kernel(rn
, old
);
901 /* Update nexthop for route, reset changed flag. */
902 /* Note: this code also handles the Linux case when an interface goes
903 * down, causing the kernel to delete routes without sending DELROUTE
906 if (RIB_KERNEL_ROUTE(old
))
907 SET_FLAG(old
->status
, ROUTE_ENTRY_REMOVED
);
909 UNSET_FLAG(old
->status
, ROUTE_ENTRY_CHANGED
);
912 static void rib_process_update_fib(struct zebra_vrf
*zvrf
,
913 struct route_node
*rn
,
914 struct route_entry
*old
,
915 struct route_entry
*new)
920 * We have to install or update if a new route has been selected or
921 * something has changed.
923 if (new != old
|| CHECK_FLAG(new->status
, ROUTE_ENTRY_CHANGED
)) {
924 hook_call(rib_update
, rn
, "updating existing route");
926 /* Update the nexthop; we could determine here that nexthop is
928 if (nexthop_group_active_nexthop_num(new->ng
))
931 /* If nexthop is active, install the selected route, if
933 * the install succeeds, cleanup flags for prior route, if
938 if (IS_ZEBRA_DEBUG_RIB
) {
939 char buf
[SRCDEST2STR_BUFFER
];
940 srcdest_rnode2str(rn
, buf
, sizeof(buf
));
943 "%u:%s: Updating route rn %p, re %p (%s) old %p (%s)",
944 zvrf_id(zvrf
), buf
, rn
, new,
945 zebra_route_string(new->type
),
947 zebra_route_string(old
->type
));
950 "%u:%s: Updating route rn %p, re %p (%s)",
951 zvrf_id(zvrf
), buf
, rn
, new,
952 zebra_route_string(new->type
));
955 /* If labeled-unicast route, uninstall transit LSP. */
956 if (zebra_rib_labeled_unicast(old
))
957 zebra_mpls_lsp_uninstall(zvrf
, rn
, old
);
960 * Non-system route should be installed.
961 * If labeled-unicast route, install transit
964 if (zebra_rib_labeled_unicast(new))
965 zebra_mpls_lsp_install(zvrf
, rn
, new);
967 rib_install_kernel(rn
, new, old
);
971 * If nexthop for selected route is not active or install
973 * may need to uninstall and delete for redistribution.
976 if (IS_ZEBRA_DEBUG_RIB
) {
977 char buf
[SRCDEST2STR_BUFFER
];
978 srcdest_rnode2str(rn
, buf
, sizeof(buf
));
981 "%u:%s: Deleting route rn %p, re %p (%s) old %p (%s) - nexthop inactive",
982 zvrf_id(zvrf
), buf
, rn
, new,
983 zebra_route_string(new->type
),
985 zebra_route_string(old
->type
));
988 "%u:%s: Deleting route rn %p, re %p (%s) - nexthop inactive",
989 zvrf_id(zvrf
), buf
, rn
, new,
990 zebra_route_string(new->type
));
993 /* If labeled-unicast route, uninstall transit LSP. */
994 if (zebra_rib_labeled_unicast(old
))
995 zebra_mpls_lsp_uninstall(zvrf
, rn
, old
);
997 rib_uninstall_kernel(rn
, old
);
1001 * Same route selected; check if in the FIB and if not,
1002 * re-install. This is housekeeping code to deal with
1003 * race conditions in kernel with linux netlink reporting
1004 * interface up before IPv4 or IPv6 protocol is ready
1007 if (!CHECK_FLAG(new->status
, ROUTE_ENTRY_INSTALLED
) ||
1008 RIB_SYSTEM_ROUTE(new))
1009 rib_install_kernel(rn
, new, NULL
);
1012 /* Update prior route. */
1014 UNSET_FLAG(old
->status
, ROUTE_ENTRY_CHANGED
);
1016 /* Clear changed flag. */
1017 UNSET_FLAG(new->status
, ROUTE_ENTRY_CHANGED
);
1020 /* Check if 'alternate' RIB entry is better than 'current'. */
1021 static struct route_entry
*rib_choose_best(struct route_entry
*current
,
1022 struct route_entry
*alternate
)
1024 if (current
== NULL
)
1027 /* filter route selection in following order:
1028 * - connected beats other types
1029 * - if both connected, loopback or vrf wins
1030 * - lower distance beats higher
1031 * - lower metric beats higher for equal distance
1032 * - last, hence oldest, route wins tie break.
1035 /* Connected routes. Check to see if either are a vrf
1036 * or loopback interface. If not, pick the last connected
1037 * route of the set of lowest metric connected routes.
1039 if (alternate
->type
== ZEBRA_ROUTE_CONNECT
) {
1040 if (current
->type
!= ZEBRA_ROUTE_CONNECT
)
1043 /* both are connected. are either loop or vrf? */
1044 struct nexthop
*nexthop
= NULL
;
1046 for (ALL_NEXTHOPS_PTR(alternate
->ng
, nexthop
)) {
1047 struct interface
*ifp
= if_lookup_by_index(
1048 nexthop
->ifindex
, alternate
->vrf_id
);
1050 if (ifp
&& if_is_loopback_or_vrf(ifp
))
1054 for (ALL_NEXTHOPS_PTR(current
->ng
, nexthop
)) {
1055 struct interface
*ifp
= if_lookup_by_index(
1056 nexthop
->ifindex
, current
->vrf_id
);
1058 if (ifp
&& if_is_loopback_or_vrf(ifp
))
1062 /* Neither are loop or vrf so pick best metric */
1063 if (alternate
->metric
<= current
->metric
)
1069 if (current
->type
== ZEBRA_ROUTE_CONNECT
)
1072 /* higher distance loses */
1073 if (alternate
->distance
< current
->distance
)
1075 if (current
->distance
< alternate
->distance
)
1078 /* metric tie-breaks equal distance */
1079 if (alternate
->metric
<= current
->metric
)
1085 /* Core function for processing routing information base. */
1086 static void rib_process(struct route_node
*rn
)
1088 struct route_entry
*re
;
1089 struct route_entry
*next
;
1090 struct route_entry
*old_selected
= NULL
;
1091 struct route_entry
*new_selected
= NULL
;
1092 struct route_entry
*old_fib
= NULL
;
1093 struct route_entry
*new_fib
= NULL
;
1094 struct route_entry
*best
= NULL
;
1095 char buf
[SRCDEST2STR_BUFFER
];
1097 struct zebra_vrf
*zvrf
= NULL
;
1098 const struct prefix
*p
, *src_p
;
1100 srcdest_rnode_prefixes(rn
, &p
, &src_p
);
1101 vrf_id_t vrf_id
= VRF_UNKNOWN
;
1105 dest
= rib_dest_from_rnode(rn
);
1107 zvrf
= rib_dest_vrf(dest
);
1108 vrf_id
= zvrf_id(zvrf
);
1111 if (IS_ZEBRA_DEBUG_RIB
)
1112 srcdest_rnode2str(rn
, buf
, sizeof(buf
));
1114 if (IS_ZEBRA_DEBUG_RIB_DETAILED
)
1115 zlog_debug("%u:%s: Processing rn %p", vrf_id
, buf
, rn
);
1118 * we can have rn's that have a NULL info pointer
1119 * (dest). As such let's not let the deref happen
1120 * additionally we know RNODE_FOREACH_RE_SAFE
1121 * will not iterate so we are ok.
1124 old_fib
= dest
->selected_fib
;
1126 RNODE_FOREACH_RE_SAFE (rn
, re
, next
) {
1127 if (IS_ZEBRA_DEBUG_RIB_DETAILED
)
1129 "%u:%s: Examine re %p (%s) status %x flags %x dist %d metric %d",
1130 vrf_id
, buf
, re
, zebra_route_string(re
->type
),
1131 re
->status
, re
->flags
, re
->distance
,
1134 /* Currently selected re. */
1135 if (CHECK_FLAG(re
->flags
, ZEBRA_FLAG_SELECTED
)) {
1136 assert(old_selected
== NULL
);
1140 /* Skip deleted entries from selection */
1141 if (CHECK_FLAG(re
->status
, ROUTE_ENTRY_REMOVED
))
1144 /* Skip unreachable nexthop. */
1145 /* This first call to nexthop_active_update is merely to
1146 * determine if there's any change to nexthops associated
1147 * with this RIB entry. Now, rib_process() can be invoked due
1148 * to an external event such as link down or due to
1149 * next-hop-tracking evaluation. In the latter case,
1150 * a decision has already been made that the NHs have changed.
1151 * So, no need to invoke a potentially expensive call again.
1152 * Further, since the change might be in a recursive NH which
1153 * is not caught in the nexthop_active_update() code. Thus, we
1154 * might miss changes to recursive NHs.
1156 if (CHECK_FLAG(re
->status
, ROUTE_ENTRY_CHANGED
)
1157 && !nexthop_active_update(rn
, re
)) {
1158 if (re
->type
== ZEBRA_ROUTE_TABLE
) {
1159 /* XXX: HERE BE DRAGONS!!!!!
1160 * In all honesty, I have not yet figured out
1161 * what this part does or why the
1162 * ROUTE_ENTRY_CHANGED test above is correct
1163 * or why we need to delete a route here, and
1164 * also not whether this concerns both selected
1165 * and fib route, or only selected
1168 * This entry was denied by the 'ip protocol
1169 * table' route-map, we need to delete it */
1170 if (re
!= old_selected
) {
1171 if (IS_ZEBRA_DEBUG_RIB
)
1173 "%s: %u:%s: imported via import-table but denied "
1174 "by the ip protocol table route-map",
1175 __func__
, vrf_id
, buf
);
1178 SET_FLAG(re
->status
,
1179 ROUTE_ENTRY_REMOVED
);
1185 /* Infinite distance. */
1186 if (re
->distance
== DISTANCE_INFINITY
) {
1187 UNSET_FLAG(re
->status
, ROUTE_ENTRY_CHANGED
);
1191 if (CHECK_FLAG(re
->flags
, ZEBRA_FLAG_FIB_OVERRIDE
)) {
1192 best
= rib_choose_best(new_fib
, re
);
1193 if (new_fib
&& best
!= new_fib
)
1194 UNSET_FLAG(new_fib
->status
,
1195 ROUTE_ENTRY_CHANGED
);
1198 best
= rib_choose_best(new_selected
, re
);
1199 if (new_selected
&& best
!= new_selected
)
1200 UNSET_FLAG(new_selected
->status
,
1201 ROUTE_ENTRY_CHANGED
);
1202 new_selected
= best
;
1205 UNSET_FLAG(re
->status
, ROUTE_ENTRY_CHANGED
);
1206 } /* RNODE_FOREACH_RE */
1208 /* If no FIB override route, use the selected route also for FIB */
1209 if (new_fib
== NULL
)
1210 new_fib
= new_selected
;
1212 /* After the cycle is finished, the following pointers will be set:
1213 * old_selected --- RE entry currently having SELECTED
1214 * new_selected --- RE entry that is newly SELECTED
1215 * old_fib --- RE entry currently in kernel FIB
1216 * new_fib --- RE entry that is newly to be in kernel FIB
1218 * new_selected will get SELECTED flag, and is going to be redistributed
1219 * the zclients. new_fib (which can be new_selected) will be installed
1223 if (IS_ZEBRA_DEBUG_RIB_DETAILED
) {
1225 "%u:%s: After processing: old_selected %p new_selected %p old_fib %p new_fib %p",
1226 vrf_id
, buf
, (void *)old_selected
, (void *)new_selected
,
1227 (void *)old_fib
, (void *)new_fib
);
1230 /* Buffer ROUTE_ENTRY_CHANGED here, because it will get cleared if
1231 * fib == selected */
1232 bool selected_changed
= new_selected
&& CHECK_FLAG(new_selected
->status
,
1233 ROUTE_ENTRY_CHANGED
);
1235 /* Update fib according to selection results */
1236 if (new_fib
&& old_fib
)
1237 rib_process_update_fib(zvrf
, rn
, old_fib
, new_fib
);
1239 rib_process_add_fib(zvrf
, rn
, new_fib
);
1241 rib_process_del_fib(zvrf
, rn
, old_fib
);
1243 /* Update SELECTED entry */
1244 if (old_selected
!= new_selected
|| selected_changed
) {
1246 if (new_selected
&& new_selected
!= new_fib
)
1247 UNSET_FLAG(new_selected
->status
, ROUTE_ENTRY_CHANGED
);
1250 SET_FLAG(new_selected
->flags
, ZEBRA_FLAG_SELECTED
);
1254 * If we're removing the old entry, we should tell
1255 * redist subscribers about that *if* they aren't
1256 * going to see a redist for the new entry.
1258 if (!new_selected
|| CHECK_FLAG(old_selected
->status
,
1259 ROUTE_ENTRY_REMOVED
))
1260 redistribute_delete(p
, src_p
,
1264 if (old_selected
!= new_selected
)
1265 UNSET_FLAG(old_selected
->flags
,
1266 ZEBRA_FLAG_SELECTED
);
1270 /* Remove all RE entries queued for removal */
1271 RNODE_FOREACH_RE_SAFE (rn
, re
, next
) {
1272 if (CHECK_FLAG(re
->status
, ROUTE_ENTRY_REMOVED
)) {
1273 if (IS_ZEBRA_DEBUG_RIB
) {
1274 rnode_debug(rn
, vrf_id
, "rn %p, removing re %p",
1275 (void *)rn
, (void *)re
);
1282 * Check if the dest can be deleted now.
1287 static void zebra_rib_evaluate_mpls(struct route_node
*rn
)
1289 rib_dest_t
*dest
= rib_dest_from_rnode(rn
);
1290 struct zebra_vrf
*zvrf
= vrf_info_lookup(VRF_DEFAULT
);
1295 if (CHECK_FLAG(dest
->flags
, RIB_DEST_UPDATE_LSPS
)) {
1296 if (IS_ZEBRA_DEBUG_MPLS
)
1298 "%u: Scheduling all LSPs upon RIB completion",
1300 zebra_mpls_lsp_schedule(zvrf
);
1301 mpls_unmark_lsps_for_processing(rn
);
1306 * Utility to match route with dplane context data
1308 static bool rib_route_match_ctx(const struct route_entry
*re
,
1309 const struct zebra_dplane_ctx
*ctx
,
1312 bool result
= false;
1316 * In 'update' case, we test info about the 'previous' or
1319 if ((re
->type
== dplane_ctx_get_old_type(ctx
)) &&
1320 (re
->instance
== dplane_ctx_get_old_instance(ctx
))) {
1323 /* TODO -- we're using this extra test, but it's not
1324 * exactly clear why.
1326 if (re
->type
== ZEBRA_ROUTE_STATIC
&&
1327 (re
->distance
!= dplane_ctx_get_old_distance(ctx
) ||
1328 re
->tag
!= dplane_ctx_get_old_tag(ctx
))) {
1335 * Ordinary, single-route case using primary context info
1337 if ((dplane_ctx_get_op(ctx
) != DPLANE_OP_ROUTE_DELETE
) &&
1338 CHECK_FLAG(re
->status
, ROUTE_ENTRY_REMOVED
)) {
1339 /* Skip route that's been deleted */
1343 if ((re
->type
== dplane_ctx_get_type(ctx
)) &&
1344 (re
->instance
== dplane_ctx_get_instance(ctx
))) {
1347 /* TODO -- we're using this extra test, but it's not
1348 * exactly clear why.
1350 if (re
->type
== ZEBRA_ROUTE_STATIC
&&
1351 (re
->distance
!= dplane_ctx_get_distance(ctx
) ||
1352 re
->tag
!= dplane_ctx_get_tag(ctx
))) {
1363 static void zebra_rib_fixup_system(struct route_node
*rn
)
1365 struct route_entry
*re
;
1367 RNODE_FOREACH_RE(rn
, re
) {
1368 struct nexthop
*nhop
;
1370 if (!RIB_SYSTEM_ROUTE(re
))
1373 if (CHECK_FLAG(re
->status
, ROUTE_ENTRY_REMOVED
))
1376 SET_FLAG(re
->status
, ROUTE_ENTRY_INSTALLED
);
1377 UNSET_FLAG(re
->status
, ROUTE_ENTRY_QUEUED
);
1379 for (ALL_NEXTHOPS_PTR(re
->ng
, nhop
)) {
1380 if (CHECK_FLAG(nhop
->flags
, NEXTHOP_FLAG_RECURSIVE
))
1383 SET_FLAG(nhop
->flags
, NEXTHOP_FLAG_FIB
);
1389 * Update a route from a dplane context. This consolidates common code
1390 * that can be used in processing of results from FIB updates, and in
1391 * async notification processing.
1392 * The return is 'true' if the installed nexthops changed; 'false' otherwise.
1394 static bool rib_update_re_from_ctx(struct route_entry
*re
,
1395 struct route_node
*rn
,
1396 struct zebra_dplane_ctx
*ctx
)
1398 char dest_str
[PREFIX_STRLEN
] = "";
1399 char nh_str
[NEXTHOP_STRLEN
];
1400 struct nexthop
*nexthop
, *ctx_nexthop
;
1402 const struct nexthop_group
*ctxnhg
;
1403 bool is_selected
= false; /* Is 're' currently the selected re? */
1404 bool changed_p
= false; /* Change to nexthops? */
1407 /* Note well: only capturing the prefix string if debug is enabled here;
1408 * unconditional log messages will have to generate the string.
1410 if (IS_ZEBRA_DEBUG_RIB
)
1411 prefix2str(&(rn
->p
), dest_str
, sizeof(dest_str
));
1413 dest
= rib_dest_from_rnode(rn
);
1415 is_selected
= (re
== dest
->selected_fib
);
1417 if (IS_ZEBRA_DEBUG_RIB_DETAILED
)
1418 zlog_debug("update_from_ctx: %u:%s: %sSELECTED",
1419 re
->vrf_id
, dest_str
, (is_selected
? "" : "NOT "));
1421 /* Update zebra's nexthop FIB flag for each nexthop that was installed.
1422 * If the installed set differs from the set requested by the rib/owner,
1423 * we use the fib-specific nexthop-group to record the actual FIB
1428 * First check the fib nexthop-group, if it's present. The comparison
1429 * here is quite strict: we require that the fib sets match exactly.
1433 if (re
->fib_ng
.nexthop
== NULL
)
1438 /* First check the route's fib nexthops */
1439 for (ALL_NEXTHOPS(re
->fib_ng
, nexthop
)) {
1441 if (CHECK_FLAG(nexthop
->flags
, NEXTHOP_FLAG_RECURSIVE
))
1445 for (ALL_NEXTHOPS_PTR(dplane_ctx_get_ng(ctx
),
1447 if (nexthop_same(ctx_nexthop
, nexthop
))
1451 if (ctx_nexthop
== NULL
) {
1452 /* Nexthop not in the new installed set */
1453 if (IS_ZEBRA_DEBUG_RIB_DETAILED
) {
1454 nexthop2str(nexthop
, nh_str
,
1456 zlog_debug("update_from_ctx: no match for fib nh %s",
1468 /* Check the new installed set */
1470 for (ALL_NEXTHOPS_PTR(dplane_ctx_get_ng(ctx
), ctx_nexthop
)) {
1472 if (CHECK_FLAG(ctx_nexthop
->flags
,
1473 NEXTHOP_FLAG_RECURSIVE
))
1476 /* Compare with the current group's nexthops */
1478 for (ALL_NEXTHOPS(re
->fib_ng
, nexthop
)) {
1479 if (nexthop_same(nexthop
, ctx_nexthop
))
1483 if (nexthop
== NULL
) {
1484 /* Nexthop not in the old installed set */
1485 if (IS_ZEBRA_DEBUG_RIB_DETAILED
) {
1486 nexthop2str(ctx_nexthop
, nh_str
,
1488 zlog_debug("update_from_ctx: no fib match for notif nh %s",
1498 /* If the new FIB set matches the existing FIB set, we're done. */
1500 if (IS_ZEBRA_DEBUG_RIB
)
1501 zlog_debug("%u:%s update_from_ctx(): existing fib nhg, no change",
1502 re
->vrf_id
, dest_str
);
1505 } else if (re
->fib_ng
.nexthop
) {
1507 * Free stale fib list and move on to check the rib nhg.
1509 if (IS_ZEBRA_DEBUG_RIB
)
1510 zlog_debug("%u:%s update_from_ctx(): replacing fib nhg",
1511 re
->vrf_id
, dest_str
);
1512 nexthops_free(re
->fib_ng
.nexthop
);
1513 re
->fib_ng
.nexthop
= NULL
;
1515 /* Note that the installed nexthops have changed */
1518 if (IS_ZEBRA_DEBUG_RIB
)
1519 zlog_debug("%u:%s update_from_ctx(): no fib nhg",
1520 re
->vrf_id
, dest_str
);
1524 * Compare with the rib nexthop group. The comparison here is different:
1525 * the RIB group may be a superset of the list installed in the FIB. We
1526 * walk the RIB group, looking for the 'installable' candidate
1527 * nexthops, and then check those against the set
1528 * that is actually installed.
1531 for (ALL_NEXTHOPS_PTR(re
->ng
, nexthop
)) {
1533 if (CHECK_FLAG(nexthop
->flags
, NEXTHOP_FLAG_RECURSIVE
))
1536 if (!CHECK_FLAG(nexthop
->flags
, NEXTHOP_FLAG_ACTIVE
))
1539 /* Check for a FIB nexthop corresponding to the RIB nexthop */
1541 for (ALL_NEXTHOPS_PTR(dplane_ctx_get_ng(ctx
), ctx_nexthop
)) {
1542 if (nexthop_same(ctx_nexthop
, nexthop
))
1546 /* If the FIB doesn't know about the nexthop,
1547 * it's not installed
1549 if (ctx_nexthop
== NULL
) {
1550 if (IS_ZEBRA_DEBUG_RIB_DETAILED
) {
1551 nexthop2str(nexthop
, nh_str
, sizeof(nh_str
));
1552 zlog_debug("update_from_ctx: no notif match for rib nh %s",
1557 if (CHECK_FLAG(nexthop
->flags
, NEXTHOP_FLAG_FIB
))
1560 UNSET_FLAG(nexthop
->flags
, NEXTHOP_FLAG_FIB
);
1562 /* Keep checking nexthops */
1566 if (CHECK_FLAG(ctx_nexthop
->flags
, NEXTHOP_FLAG_FIB
)) {
1567 if (!CHECK_FLAG(nexthop
->flags
, NEXTHOP_FLAG_FIB
))
1570 SET_FLAG(nexthop
->flags
, NEXTHOP_FLAG_FIB
);
1572 if (CHECK_FLAG(nexthop
->flags
, NEXTHOP_FLAG_FIB
))
1575 UNSET_FLAG(nexthop
->flags
, NEXTHOP_FLAG_FIB
);
1579 /* If all nexthops were processed, we're done */
1581 if (IS_ZEBRA_DEBUG_RIB
)
1582 zlog_debug("%u:%s update_from_ctx(): rib nhg matched, changed '%s'",
1583 re
->vrf_id
, dest_str
,
1584 (changed_p
? "true" : "false"));
1588 /* FIB nexthop set differs from the RIB set:
1589 * create a fib-specific nexthop-group
1591 if (IS_ZEBRA_DEBUG_RIB
)
1592 zlog_debug("%u:%s update_from_ctx(): changed %s, adding new fib nhg",
1593 re
->vrf_id
, dest_str
,
1594 (changed_p
? "true" : "false"));
1596 ctxnhg
= dplane_ctx_get_ng(ctx
);
1598 if (ctxnhg
->nexthop
)
1599 copy_nexthops(&(re
->fib_ng
.nexthop
), ctxnhg
->nexthop
, NULL
);
1601 /* Bit of a special case when the fib has _no_ installed
1604 nexthop
= nexthop_new();
1605 nexthop
->type
= NEXTHOP_TYPE_IPV4
;
1606 _nexthop_add(&(re
->fib_ng
.nexthop
), nexthop
);
1614 * Helper to locate a zebra route-node from a dplane context. This is used
1615 * when processing dplane results, e.g. Note well: the route-node is returned
1616 * with a ref held - route_unlock_node() must be called eventually.
1618 static struct route_node
*
1619 rib_find_rn_from_ctx(const struct zebra_dplane_ctx
*ctx
)
1621 struct route_table
*table
= NULL
;
1622 struct route_node
*rn
= NULL
;
1623 const struct prefix
*dest_pfx
, *src_pfx
;
1625 /* Locate rn and re(s) from ctx */
1627 table
= zebra_vrf_table_with_table_id(dplane_ctx_get_afi(ctx
),
1628 dplane_ctx_get_safi(ctx
),
1629 dplane_ctx_get_vrf(ctx
),
1630 dplane_ctx_get_table(ctx
));
1631 if (table
== NULL
) {
1632 if (IS_ZEBRA_DEBUG_DPLANE
) {
1633 zlog_debug("Failed to find route for ctx: no table for afi %d, safi %d, vrf %u",
1634 dplane_ctx_get_afi(ctx
),
1635 dplane_ctx_get_safi(ctx
),
1636 dplane_ctx_get_vrf(ctx
));
1641 dest_pfx
= dplane_ctx_get_dest(ctx
);
1642 src_pfx
= dplane_ctx_get_src(ctx
);
1644 rn
= srcdest_rnode_get(table
, dest_pfx
,
1645 src_pfx
? (struct prefix_ipv6
*)src_pfx
: NULL
);
1654 * Route-update results processing after async dataplane update.
1656 static void rib_process_result(struct zebra_dplane_ctx
*ctx
)
1658 struct zebra_vrf
*zvrf
= NULL
;
1659 struct route_node
*rn
= NULL
;
1660 struct route_entry
*re
= NULL
, *old_re
= NULL
, *rib
;
1661 bool is_update
= false;
1662 char dest_str
[PREFIX_STRLEN
] = "";
1663 enum dplane_op_e op
;
1664 enum zebra_dplane_result status
;
1665 const struct prefix
*dest_pfx
, *src_pfx
;
1667 bool fib_changed
= false;
1669 zvrf
= vrf_info_lookup(dplane_ctx_get_vrf(ctx
));
1670 dest_pfx
= dplane_ctx_get_dest(ctx
);
1672 /* Note well: only capturing the prefix string if debug is enabled here;
1673 * unconditional log messages will have to generate the string.
1675 if (IS_ZEBRA_DEBUG_DPLANE
)
1676 prefix2str(dest_pfx
, dest_str
, sizeof(dest_str
));
1678 /* Locate rn and re(s) from ctx */
1679 rn
= rib_find_rn_from_ctx(ctx
);
1681 if (IS_ZEBRA_DEBUG_DPLANE
) {
1682 zlog_debug("Failed to process dplane results: no route for %u:%s",
1683 dplane_ctx_get_vrf(ctx
), dest_str
);
1688 srcdest_rnode_prefixes(rn
, &dest_pfx
, &src_pfx
);
1690 op
= dplane_ctx_get_op(ctx
);
1691 status
= dplane_ctx_get_status(ctx
);
1693 if (IS_ZEBRA_DEBUG_DPLANE_DETAIL
)
1694 zlog_debug("%u:%s Processing dplane ctx %p, op %s result %s",
1695 dplane_ctx_get_vrf(ctx
), dest_str
, ctx
,
1696 dplane_op2str(op
), dplane_res2str(status
));
1699 * Update is a bit of a special case, where we may have both old and new
1700 * routes to post-process.
1702 is_update
= dplane_ctx_is_update(ctx
);
1705 * Take a pass through the routes, look for matches with the context
1708 RNODE_FOREACH_RE(rn
, rib
) {
1711 if (rib_route_match_ctx(rib
, ctx
, false))
1715 /* Check for old route match */
1716 if (is_update
&& (old_re
== NULL
)) {
1717 if (rib_route_match_ctx(rib
, ctx
, true /*is_update*/))
1721 /* Have we found the routes we need to work on? */
1722 if (re
&& ((!is_update
|| old_re
)))
1726 seq
= dplane_ctx_get_seq(ctx
);
1729 * Check sequence number(s) to detect stale results before continuing
1732 if (re
->dplane_sequence
!= seq
) {
1733 if (IS_ZEBRA_DEBUG_DPLANE_DETAIL
)
1734 zlog_debug("%u:%s Stale dplane result for re %p",
1735 dplane_ctx_get_vrf(ctx
),
1738 UNSET_FLAG(re
->status
, ROUTE_ENTRY_QUEUED
);
1742 if (old_re
->dplane_sequence
!= dplane_ctx_get_old_seq(ctx
)) {
1743 if (IS_ZEBRA_DEBUG_DPLANE_DETAIL
)
1744 zlog_debug("%u:%s Stale dplane result for old_re %p",
1745 dplane_ctx_get_vrf(ctx
),
1748 UNSET_FLAG(old_re
->status
, ROUTE_ENTRY_QUEUED
);
1752 case DPLANE_OP_ROUTE_INSTALL
:
1753 case DPLANE_OP_ROUTE_UPDATE
:
1754 if (status
== ZEBRA_DPLANE_REQUEST_SUCCESS
) {
1756 UNSET_FLAG(re
->status
, ROUTE_ENTRY_FAILED
);
1757 SET_FLAG(re
->status
, ROUTE_ENTRY_INSTALLED
);
1760 * On an update operation from the same route type
1761 * context retrieval currently has no way to know
1762 * which was the old and which was the new.
1763 * So don't unset our flags that we just set.
1764 * We know redistribution is ok because the
1765 * old_re in this case is used for nothing
1766 * more than knowing whom to contact if necessary.
1768 if (old_re
&& old_re
!= re
) {
1769 UNSET_FLAG(old_re
->status
, ROUTE_ENTRY_FAILED
);
1770 UNSET_FLAG(old_re
->status
,
1771 ROUTE_ENTRY_INSTALLED
);
1774 /* Update zebra route based on the results in
1775 * the context struct.
1779 rib_update_re_from_ctx(re
, rn
, ctx
);
1782 if (IS_ZEBRA_DEBUG_DPLANE_DETAIL
)
1783 zlog_debug("%u:%s no fib change for re",
1790 redistribute_update(dest_pfx
, src_pfx
,
1795 * System routes are weird in that they
1796 * allow multiple to be installed that match
1797 * to the same prefix, so after we get the
1798 * result we need to clean them up so that
1799 * we can actually use them.
1801 if ((re
&& RIB_SYSTEM_ROUTE(re
)) ||
1802 (old_re
&& RIB_SYSTEM_ROUTE(old_re
)))
1803 zebra_rib_fixup_system(rn
);
1808 /* Notify route owner */
1809 zsend_route_notify_owner_ctx(ctx
, ZAPI_ROUTE_INSTALLED
);
1813 SET_FLAG(re
->status
, ROUTE_ENTRY_FAILED
);
1814 UNSET_FLAG(re
->status
, ROUTE_ENTRY_INSTALLED
);
1816 SET_FLAG(old_re
->status
, ROUTE_ENTRY_FAILED
);
1818 zsend_route_notify_owner(re
, dest_pfx
,
1819 ZAPI_ROUTE_FAIL_INSTALL
);
1821 zlog_warn("%u:%s: Route install failed",
1822 dplane_ctx_get_vrf(ctx
),
1823 prefix2str(dest_pfx
,
1824 dest_str
, sizeof(dest_str
)));
1827 case DPLANE_OP_ROUTE_DELETE
:
1829 SET_FLAG(re
->status
, ROUTE_ENTRY_FAILED
);
1831 * In the delete case, the zebra core datastructs were
1832 * updated (or removed) at the time the delete was issued,
1833 * so we're just notifying the route owner.
1835 if (status
== ZEBRA_DPLANE_REQUEST_SUCCESS
) {
1837 UNSET_FLAG(re
->status
, ROUTE_ENTRY_INSTALLED
);
1838 UNSET_FLAG(re
->status
, ROUTE_ENTRY_FAILED
);
1840 zsend_route_notify_owner_ctx(ctx
, ZAPI_ROUTE_REMOVED
);
1846 SET_FLAG(re
->status
, ROUTE_ENTRY_FAILED
);
1847 zsend_route_notify_owner_ctx(ctx
,
1848 ZAPI_ROUTE_REMOVE_FAIL
);
1850 zlog_warn("%u:%s: Route Deletion failure",
1851 dplane_ctx_get_vrf(ctx
),
1852 prefix2str(dest_pfx
,
1853 dest_str
, sizeof(dest_str
)));
1857 * System routes are weird in that they
1858 * allow multiple to be installed that match
1859 * to the same prefix, so after we get the
1860 * result we need to clean them up so that
1861 * we can actually use them.
1863 if ((re
&& RIB_SYSTEM_ROUTE(re
)) ||
1864 (old_re
&& RIB_SYSTEM_ROUTE(old_re
)))
1865 zebra_rib_fixup_system(rn
);
1871 zebra_rib_evaluate_rn_nexthops(rn
, seq
);
1872 zebra_rib_evaluate_mpls(rn
);
1876 route_unlock_node(rn
);
1878 /* Return context to dataplane module */
1879 dplane_ctx_fini(&ctx
);
1883 * Handle notification from async dataplane: the dataplane has detected
1884 * some change to a route, and notifies zebra so that the control plane
1885 * can reflect that change.
1887 static void rib_process_dplane_notify(struct zebra_dplane_ctx
*ctx
)
1889 struct route_node
*rn
= NULL
;
1890 struct route_entry
*re
= NULL
;
1891 struct nexthop
*nexthop
;
1892 char dest_str
[PREFIX_STRLEN
] = "";
1893 const struct prefix
*dest_pfx
, *src_pfx
;
1895 bool fib_changed
= false;
1896 bool debug_p
= IS_ZEBRA_DEBUG_DPLANE
| IS_ZEBRA_DEBUG_RIB
;
1897 int start_count
, end_count
;
1898 dest_pfx
= dplane_ctx_get_dest(ctx
);
1900 /* Note well: only capturing the prefix string if debug is enabled here;
1901 * unconditional log messages will have to generate the string.
1904 prefix2str(dest_pfx
, dest_str
, sizeof(dest_str
));
1906 /* Locate rn and re(s) from ctx */
1907 rn
= rib_find_rn_from_ctx(ctx
);
1910 zlog_debug("Failed to process dplane notification: no routes for %u:%s",
1911 dplane_ctx_get_vrf(ctx
), dest_str
);
1916 dest
= rib_dest_from_rnode(rn
);
1917 srcdest_rnode_prefixes(rn
, &dest_pfx
, &src_pfx
);
1920 zlog_debug("%u:%s Processing dplane notif ctx %p",
1921 dplane_ctx_get_vrf(ctx
), dest_str
, ctx
);
1924 * Take a pass through the routes, look for matches with the context
1927 RNODE_FOREACH_RE(rn
, re
) {
1928 if (rib_route_match_ctx(re
, ctx
, false /*!update*/))
1932 /* No match? Nothing we can do */
1935 zlog_debug("%u:%s Unable to process dplane notification: no entry for type %s",
1936 dplane_ctx_get_vrf(ctx
), dest_str
,
1938 dplane_ctx_get_type(ctx
)));
1943 /* Ensure we clear the QUEUED flag */
1944 UNSET_FLAG(re
->status
, ROUTE_ENTRY_QUEUED
);
1946 /* Is this a notification that ... matters? We only really care about
1947 * the route that is currently selected for installation.
1949 if (re
!= dest
->selected_fib
) {
1950 /* TODO -- don't skip processing entirely? We might like to
1951 * at least report on the event.
1954 zlog_debug("%u:%s dplane notif, but type %s not selected_fib",
1955 dplane_ctx_get_vrf(ctx
), dest_str
,
1957 dplane_ctx_get_type(ctx
)));
1961 /* We'll want to determine whether the installation status of the
1962 * route has changed: we'll check the status before processing,
1963 * and then again if there's been a change.
1966 for (ALL_NEXTHOPS_PTR(rib_active_nhg(re
), nexthop
)) {
1967 if (CHECK_FLAG(nexthop
->flags
, NEXTHOP_FLAG_FIB
))
1971 /* Update zebra's nexthop FIB flags based on the context struct's
1974 fib_changed
= rib_update_re_from_ctx(re
, rn
, ctx
);
1978 zlog_debug("%u:%s No change from dplane notification",
1979 dplane_ctx_get_vrf(ctx
), dest_str
);
1985 * Perform follow-up work if the actual status of the prefix
1990 for (ALL_NEXTHOPS_PTR(rib_active_nhg(re
), nexthop
)) {
1991 if (CHECK_FLAG(nexthop
->flags
, NEXTHOP_FLAG_FIB
))
1995 /* Various fib transitions: changed nexthops; from installed to
1996 * not-installed; or not-installed to installed.
1998 if (start_count
> 0 && end_count
> 0) {
2000 zlog_debug("%u:%s applied nexthop changes from dplane notification",
2001 dplane_ctx_get_vrf(ctx
), dest_str
);
2003 /* Changed nexthops - update kernel/others */
2004 dplane_route_notif_update(rn
, re
,
2005 DPLANE_OP_ROUTE_UPDATE
, ctx
);
2007 } else if (start_count
== 0 && end_count
> 0) {
2009 zlog_debug("%u:%s installed transition from dplane notification",
2010 dplane_ctx_get_vrf(ctx
), dest_str
);
2012 /* We expect this to be the selected route, so we want
2013 * to tell others about this transition.
2015 SET_FLAG(re
->status
, ROUTE_ENTRY_INSTALLED
);
2017 /* Changed nexthops - update kernel/others */
2018 dplane_route_notif_update(rn
, re
, DPLANE_OP_ROUTE_INSTALL
, ctx
);
2020 /* Redistribute, lsp, and nht update */
2021 redistribute_update(dest_pfx
, src_pfx
, re
, NULL
);
2023 zebra_rib_evaluate_rn_nexthops(
2024 rn
, zebra_router_get_next_sequence());
2026 zebra_rib_evaluate_mpls(rn
);
2028 } else if (start_count
> 0 && end_count
== 0) {
2030 zlog_debug("%u:%s un-installed transition from dplane notification",
2031 dplane_ctx_get_vrf(ctx
), dest_str
);
2033 /* Transition from _something_ installed to _nothing_
2036 /* We expect this to be the selected route, so we want
2037 * to tell others about this transistion.
2039 UNSET_FLAG(re
->status
, ROUTE_ENTRY_INSTALLED
);
2041 /* Changed nexthops - update kernel/others */
2042 dplane_route_notif_update(rn
, re
, DPLANE_OP_ROUTE_DELETE
, ctx
);
2044 /* Redistribute, lsp, and nht update */
2045 redistribute_delete(dest_pfx
, src_pfx
, re
, NULL
);
2047 zebra_rib_evaluate_rn_nexthops(
2048 rn
, zebra_router_get_next_sequence());
2050 zebra_rib_evaluate_mpls(rn
);
2055 route_unlock_node(rn
);
2057 /* Return context to dataplane module */
2058 dplane_ctx_fini(&ctx
);
2061 /* Take a list of route_node structs and return 1, if there was a record
2062 * picked from it and processed by rib_process(). Don't process more,
2063 * than one RN record; operate only in the specified sub-queue.
2065 static unsigned int process_subq(struct list
*subq
, uint8_t qindex
)
2067 struct listnode
*lnode
= listhead(subq
);
2068 struct route_node
*rnode
;
2070 struct zebra_vrf
*zvrf
= NULL
;
2075 rnode
= listgetdata(lnode
);
2076 dest
= rib_dest_from_rnode(rnode
);
2078 zvrf
= rib_dest_vrf(dest
);
2082 if (IS_ZEBRA_DEBUG_RIB_DETAILED
) {
2083 char buf
[SRCDEST2STR_BUFFER
];
2085 srcdest_rnode2str(rnode
, buf
, sizeof(buf
));
2086 zlog_debug("%u:%s: rn %p dequeued from sub-queue %u",
2087 zvrf
? zvrf_id(zvrf
) : 0, buf
, rnode
, qindex
);
2091 UNSET_FLAG(rib_dest_from_rnode(rnode
)->flags
,
2092 RIB_ROUTE_QUEUED(qindex
));
2097 zlog_debug ("%s: called for route_node (%p, %d) with no ribs",
2098 __func__
, rnode
, rnode
->lock
);
2099 zlog_backtrace(LOG_DEBUG
);
2102 route_unlock_node(rnode
);
2103 list_delete_node(subq
, lnode
);
2107 /* Dispatch the meta queue by picking, processing and unlocking the next RN from
2108 * a non-empty sub-queue with lowest priority. wq is equal to zebra->ribq and
2110 * is pointed to the meta queue structure.
2112 static wq_item_status
meta_queue_process(struct work_queue
*dummy
, void *data
)
2114 struct meta_queue
*mq
= data
;
2116 uint32_t queue_len
, queue_limit
;
2118 /* Ensure there's room for more dataplane updates */
2119 queue_limit
= dplane_get_in_queue_limit();
2120 queue_len
= dplane_get_in_queue_len();
2121 if (queue_len
> queue_limit
) {
2122 if (IS_ZEBRA_DEBUG_RIB_DETAILED
)
2123 zlog_debug("rib queue: dplane queue len %u, limit %u, retrying",
2124 queue_len
, queue_limit
);
2126 /* Ensure that the meta-queue is actually enqueued */
2127 if (work_queue_empty(zrouter
.ribq
))
2128 work_queue_add(zrouter
.ribq
, zrouter
.mq
);
2130 return WQ_QUEUE_BLOCKED
;
2133 for (i
= 0; i
< MQ_SIZE
; i
++)
2134 if (process_subq(mq
->subq
[i
], i
)) {
2138 return mq
->size
? WQ_REQUEUE
: WQ_SUCCESS
;
2143 * Look into the RN and queue it into the highest priority queue
2144 * at this point in time for processing.
2146 * We will enqueue a route node only once per invocation.
2148 * There are two possibilities here that should be kept in mind.
2149 * If the original invocation has not been pulled off for processing
2150 * yet, A subsuquent invocation can have a route entry with a better
2151 * meta queue index value and we can have a situation where
2152 * we might have the same node enqueued 2 times. Not necessarily
2153 * an optimal situation but it should be ok.
2155 * The other possibility is that the original invocation has not
2156 * been pulled off for processing yet, A subsusquent invocation
2157 * doesn't have a route_entry with a better meta-queue and the
2158 * original metaqueue index value will win and we'll end up with
2159 * the route node enqueued once.
2161 static void rib_meta_queue_add(struct meta_queue
*mq
, struct route_node
*rn
)
2163 struct route_entry
*re
= NULL
, *curr_re
= NULL
;
2164 uint8_t qindex
= MQ_SIZE
, curr_qindex
= MQ_SIZE
;
2166 RNODE_FOREACH_RE (rn
, curr_re
) {
2167 curr_qindex
= route_info
[curr_re
->type
].meta_q_map
;
2169 if (curr_qindex
<= qindex
) {
2171 qindex
= curr_qindex
;
2178 /* Invariant: at this point we always have rn->info set. */
2179 if (CHECK_FLAG(rib_dest_from_rnode(rn
)->flags
,
2180 RIB_ROUTE_QUEUED(qindex
))) {
2181 if (IS_ZEBRA_DEBUG_RIB_DETAILED
)
2182 rnode_debug(rn
, re
->vrf_id
,
2183 "rn %p is already queued in sub-queue %u",
2184 (void *)rn
, qindex
);
2188 SET_FLAG(rib_dest_from_rnode(rn
)->flags
, RIB_ROUTE_QUEUED(qindex
));
2189 listnode_add(mq
->subq
[qindex
], rn
);
2190 route_lock_node(rn
);
2193 if (IS_ZEBRA_DEBUG_RIB_DETAILED
)
2194 rnode_debug(rn
, re
->vrf_id
, "queued rn %p into sub-queue %u",
2195 (void *)rn
, qindex
);
2198 /* Add route_node to work queue and schedule processing */
2199 void rib_queue_add(struct route_node
*rn
)
2203 /* Pointless to queue a route_node with no RIB entries to add or remove
2205 if (!rnode_to_ribs(rn
)) {
2206 zlog_debug("%s: called for route_node (%p, %d) with no ribs",
2207 __func__
, (void *)rn
, rn
->lock
);
2208 zlog_backtrace(LOG_DEBUG
);
2212 if (zrouter
.ribq
== NULL
) {
2213 flog_err(EC_ZEBRA_WQ_NONEXISTENT
,
2214 "%s: work_queue does not exist!", __func__
);
2219 * The RIB queue should normally be either empty or holding the only
2220 * work_queue_item element. In the latter case this element would
2221 * hold a pointer to the meta queue structure, which must be used to
2222 * actually queue the route nodes to process. So create the MQ
2223 * holder, if necessary, then push the work into it in any case.
2224 * This semantics was introduced after 0.99.9 release.
2226 if (work_queue_empty(zrouter
.ribq
))
2227 work_queue_add(zrouter
.ribq
, zrouter
.mq
);
2229 rib_meta_queue_add(zrouter
.mq
, rn
);
2234 /* Create new meta queue.
2235 A destructor function doesn't seem to be necessary here.
2237 static struct meta_queue
*meta_queue_new(void)
2239 struct meta_queue
*new;
2242 new = XCALLOC(MTYPE_WORK_QUEUE
, sizeof(struct meta_queue
));
2244 for (i
= 0; i
< MQ_SIZE
; i
++) {
2245 new->subq
[i
] = list_new();
2246 assert(new->subq
[i
]);
2252 void meta_queue_free(struct meta_queue
*mq
)
2256 for (i
= 0; i
< MQ_SIZE
; i
++)
2257 list_delete(&mq
->subq
[i
]);
2259 XFREE(MTYPE_WORK_QUEUE
, mq
);
2262 /* initialise zebra rib work queue */
2263 static void rib_queue_init(void)
2265 if (!(zrouter
.ribq
= work_queue_new(zrouter
.master
,
2266 "route_node processing"))) {
2267 flog_err(EC_ZEBRA_WQ_NONEXISTENT
,
2268 "%s: could not initialise work queue!", __func__
);
2272 /* fill in the work queue spec */
2273 zrouter
.ribq
->spec
.workfunc
= &meta_queue_process
;
2274 zrouter
.ribq
->spec
.errorfunc
= NULL
;
2275 zrouter
.ribq
->spec
.completion_func
= NULL
;
2276 /* XXX: TODO: These should be runtime configurable via vty */
2277 zrouter
.ribq
->spec
.max_retries
= 3;
2278 zrouter
.ribq
->spec
.hold
= ZEBRA_RIB_PROCESS_HOLD_TIME
;
2279 zrouter
.ribq
->spec
.retry
= ZEBRA_RIB_PROCESS_RETRY_TIME
;
2281 if (!(zrouter
.mq
= meta_queue_new())) {
2282 flog_err(EC_ZEBRA_WQ_NONEXISTENT
,
2283 "%s: could not initialise meta queue!", __func__
);
2289 rib_dest_t
*zebra_rib_create_dest(struct route_node
*rn
)
2293 dest
= XCALLOC(MTYPE_RIB_DEST
, sizeof(rib_dest_t
));
2294 rnh_list_init(&dest
->nht
);
2295 route_lock_node(rn
); /* rn route table reference */
2302 /* RIB updates are processed via a queue of pointers to route_nodes.
2304 * The queue length is bounded by the maximal size of the routing table,
2305 * as a route_node will not be requeued, if already queued.
2307 * REs are submitted via rib_addnode or rib_delnode which set minimal
2308 * state, or static_install_route (when an existing RE is updated)
2309 * and then submit route_node to queue for best-path selection later.
2310 * Order of add/delete state changes are preserved for any given RE.
2312 * Deleted REs are reaped during best-path selection.
2315 * |-> rib_link or unset ROUTE_ENTRY_REMOVE |->Update kernel with
2316 * |-------->| | best RE, if required
2318 * static_install->|->rib_addqueue...... -> rib_process
2320 * |-------->| |-> rib_unlink
2321 * |-> set ROUTE_ENTRY_REMOVE |
2322 * rib_delnode (RE freed)
2324 * The 'info' pointer of a route_node points to a rib_dest_t
2325 * ('dest'). Queueing state for a route_node is kept on the dest. The
2326 * dest is created on-demand by rib_link() and is kept around at least
2327 * as long as there are ribs hanging off it (@see rib_gc_dest()).
2329 * Refcounting (aka "locking" throughout the GNU Zebra and Quagga code):
2331 * - route_nodes: refcounted by:
2332 * - dest attached to route_node:
2333 * - managed by: rib_link/rib_gc_dest
2334 * - route_node processing queue
2335 * - managed by: rib_addqueue, rib_process.
2339 /* Add RE to head of the route node. */
2340 static void rib_link(struct route_node
*rn
, struct route_entry
*re
, int process
)
2344 const char *rmap_name
;
2348 dest
= rib_dest_from_rnode(rn
);
2350 if (IS_ZEBRA_DEBUG_RIB_DETAILED
)
2351 rnode_debug(rn
, re
->vrf_id
, "rn %p adding dest", rn
);
2353 dest
= zebra_rib_create_dest(rn
);
2356 re_list_add_head(&dest
->routes
, re
);
2358 afi
= (rn
->p
.family
== AF_INET
)
2360 : (rn
->p
.family
== AF_INET6
) ? AFI_IP6
: AFI_MAX
;
2361 if (is_zebra_import_table_enabled(afi
, re
->vrf_id
, re
->table
)) {
2362 struct zebra_vrf
*zvrf
= zebra_vrf_lookup_by_id(re
->vrf_id
);
2364 rmap_name
= zebra_get_import_table_route_map(afi
, re
->table
);
2365 zebra_add_import_table_entry(zvrf
, rn
, re
, rmap_name
);
2370 static void rib_addnode(struct route_node
*rn
,
2371 struct route_entry
*re
, int process
)
2373 /* RE node has been un-removed before route-node is processed.
2374 * route_node must hence already be on the queue for processing..
2376 if (CHECK_FLAG(re
->status
, ROUTE_ENTRY_REMOVED
)) {
2377 if (IS_ZEBRA_DEBUG_RIB
)
2378 rnode_debug(rn
, re
->vrf_id
, "rn %p, un-removed re %p",
2379 (void *)rn
, (void *)re
);
2381 UNSET_FLAG(re
->status
, ROUTE_ENTRY_REMOVED
);
2384 rib_link(rn
, re
, process
);
2390 * Detach a rib structure from a route_node.
2392 * Note that a call to rib_unlink() should be followed by a call to
2393 * rib_gc_dest() at some point. This allows a rib_dest_t that is no
2394 * longer required to be deleted.
2396 void rib_unlink(struct route_node
*rn
, struct route_entry
*re
)
2399 struct nhg_hash_entry
*nhe
= NULL
;
2403 if (IS_ZEBRA_DEBUG_RIB
)
2404 rnode_debug(rn
, re
->vrf_id
, "rn %p, re %p", (void *)rn
,
2407 dest
= rib_dest_from_rnode(rn
);
2409 re_list_del(&dest
->routes
, re
);
2411 if (dest
->selected_fib
== re
)
2412 dest
->selected_fib
= NULL
;
2414 nhe
= zebra_nhg_lookup_id(re
->nhe_id
);
2416 zebra_nhg_decrement_ref(nhe
);
2418 nexthops_free(re
->fib_ng
.nexthop
);
2420 XFREE(MTYPE_RE
, re
);
2423 void rib_delnode(struct route_node
*rn
, struct route_entry
*re
)
2427 if (IS_ZEBRA_DEBUG_RIB
)
2428 rnode_debug(rn
, re
->vrf_id
, "rn %p, re %p, removing",
2429 (void *)rn
, (void *)re
);
2430 SET_FLAG(re
->status
, ROUTE_ENTRY_REMOVED
);
2432 afi
= (rn
->p
.family
== AF_INET
)
2434 : (rn
->p
.family
== AF_INET6
) ? AFI_IP6
: AFI_MAX
;
2435 if (is_zebra_import_table_enabled(afi
, re
->vrf_id
, re
->table
)) {
2436 struct zebra_vrf
*zvrf
= zebra_vrf_lookup_by_id(re
->vrf_id
);
2438 zebra_del_import_table_entry(zvrf
, rn
, re
);
2439 /* Just clean up if non main table */
2440 if (IS_ZEBRA_DEBUG_RIB
) {
2441 char buf
[SRCDEST2STR_BUFFER
];
2442 srcdest_rnode2str(rn
, buf
, sizeof(buf
));
2443 zlog_debug("%u:%s: Freeing route rn %p, re %p (%s)",
2444 re
->vrf_id
, buf
, rn
, re
,
2445 zebra_route_string(re
->type
));
2454 /* This function dumps the contents of a given RE entry into
2455 * standard debug log. Calling function name and IP prefix in
2456 * question are passed as 1st and 2nd arguments.
2459 void _route_entry_dump(const char *func
, union prefixconstptr pp
,
2460 union prefixconstptr src_pp
,
2461 const struct route_entry
*re
)
2463 const struct prefix
*src_p
= src_pp
.p
;
2464 bool is_srcdst
= src_p
&& src_p
->prefixlen
;
2465 char straddr
[PREFIX_STRLEN
];
2466 char srcaddr
[PREFIX_STRLEN
];
2467 char nhname
[PREFIX_STRLEN
];
2468 struct nexthop
*nexthop
;
2470 zlog_debug("%s: dumping RE entry %p for %s%s%s vrf %u", func
,
2471 (const void *)re
, prefix2str(pp
, straddr
, sizeof(straddr
)),
2472 is_srcdst
? " from " : "",
2473 is_srcdst
? prefix2str(src_pp
, srcaddr
, sizeof(srcaddr
))
2476 zlog_debug("%s: uptime == %lu, type == %u, instance == %d, table == %d",
2477 straddr
, (unsigned long)re
->uptime
, re
->type
, re
->instance
,
2480 "%s: metric == %u, mtu == %u, distance == %u, flags == %u, status == %u",
2481 straddr
, re
->metric
, re
->mtu
, re
->distance
, re
->flags
, re
->status
);
2482 zlog_debug("%s: nexthop_num == %u, nexthop_active_num == %u", straddr
,
2483 nexthop_group_nexthop_num(re
->ng
),
2484 nexthop_group_active_nexthop_num(re
->ng
));
2486 for (ALL_NEXTHOPS_PTR(re
->ng
, nexthop
)) {
2487 struct interface
*ifp
;
2488 struct vrf
*vrf
= vrf_lookup_by_id(nexthop
->vrf_id
);
2490 switch (nexthop
->type
) {
2491 case NEXTHOP_TYPE_BLACKHOLE
:
2492 sprintf(nhname
, "Blackhole");
2494 case NEXTHOP_TYPE_IFINDEX
:
2495 ifp
= if_lookup_by_index(nexthop
->ifindex
,
2497 sprintf(nhname
, "%s", ifp
? ifp
->name
: "Unknown");
2499 case NEXTHOP_TYPE_IPV4
:
2501 case NEXTHOP_TYPE_IPV4_IFINDEX
:
2502 inet_ntop(AF_INET
, &nexthop
->gate
, nhname
,
2505 case NEXTHOP_TYPE_IPV6
:
2506 case NEXTHOP_TYPE_IPV6_IFINDEX
:
2507 inet_ntop(AF_INET6
, &nexthop
->gate
, nhname
,
2511 zlog_debug("%s: %s %s[%u] vrf %s(%u) with flags %s%s%s%s%s%s",
2512 straddr
, (nexthop
->rparent
? " NH" : "NH"), nhname
,
2513 nexthop
->ifindex
, vrf
? vrf
->name
: "Unknown",
2515 (CHECK_FLAG(nexthop
->flags
, NEXTHOP_FLAG_ACTIVE
)
2518 (CHECK_FLAG(re
->status
, ROUTE_ENTRY_INSTALLED
)
2521 (CHECK_FLAG(nexthop
->flags
, NEXTHOP_FLAG_RECURSIVE
)
2524 (CHECK_FLAG(nexthop
->flags
, NEXTHOP_FLAG_ONLINK
)
2527 (CHECK_FLAG(nexthop
->flags
, NEXTHOP_FLAG_MATCHED
)
2530 (CHECK_FLAG(nexthop
->flags
, NEXTHOP_FLAG_DUPLICATE
)
2534 zlog_debug("%s: dump complete", straddr
);
2537 /* This is an exported helper to rtm_read() to dump the strange
2538 * RE entry found by rib_lookup_ipv4_route()
2541 void rib_lookup_and_dump(struct prefix_ipv4
*p
, vrf_id_t vrf_id
)
2543 struct route_table
*table
;
2544 struct route_node
*rn
;
2545 struct route_entry
*re
;
2546 char prefix_buf
[INET_ADDRSTRLEN
];
2549 table
= zebra_vrf_table(AFI_IP
, SAFI_UNICAST
, vrf_id
);
2551 flog_err(EC_ZEBRA_TABLE_LOOKUP_FAILED
,
2552 "%s:%u zebra_vrf_table() returned NULL", __func__
,
2557 /* Scan the RIB table for exactly matching RE entry. */
2558 rn
= route_node_lookup(table
, (struct prefix
*)p
);
2560 /* No route for this prefix. */
2562 zlog_debug("%s:%u lookup failed for %s", __func__
, vrf_id
,
2563 prefix2str((struct prefix
*)p
, prefix_buf
,
2564 sizeof(prefix_buf
)));
2569 route_unlock_node(rn
);
2572 RNODE_FOREACH_RE (rn
, re
) {
2573 zlog_debug("%s:%u rn %p, re %p: %s, %s",
2575 (void *)rn
, (void *)re
,
2576 (CHECK_FLAG(re
->status
, ROUTE_ENTRY_REMOVED
)
2579 (CHECK_FLAG(re
->flags
, ZEBRA_FLAG_SELECTED
)
2582 route_entry_dump(p
, NULL
, re
);
2586 /* Check if requested address assignment will fail due to another
2587 * route being installed by zebra in FIB already. Take necessary
2588 * actions, if needed: remove such a route from FIB and deSELECT
2589 * corresponding RE entry. Then put affected RN into RIBQ head.
2591 void rib_lookup_and_pushup(struct prefix_ipv4
*p
, vrf_id_t vrf_id
)
2593 struct route_table
*table
;
2594 struct route_node
*rn
;
2597 if (NULL
== (table
= zebra_vrf_table(AFI_IP
, SAFI_UNICAST
, vrf_id
))) {
2598 flog_err(EC_ZEBRA_TABLE_LOOKUP_FAILED
,
2599 "%s:%u zebra_vrf_table() returned NULL", __func__
,
2604 /* No matches would be the simplest case. */
2605 if (NULL
== (rn
= route_node_lookup(table
, (struct prefix
*)p
)))
2609 route_unlock_node(rn
);
2611 dest
= rib_dest_from_rnode(rn
);
2612 /* Check all RE entries. In case any changes have to be done, requeue
2613 * the RN into RIBQ head. If the routing message about the new connected
2614 * route (generated by the IP address we are going to assign very soon)
2615 * comes before the RIBQ is processed, the new RE entry will join
2616 * RIBQ record already on head. This is necessary for proper
2618 * of the rest of the RE.
2620 if (dest
->selected_fib
) {
2621 if (IS_ZEBRA_DEBUG_RIB
) {
2622 char buf
[PREFIX_STRLEN
];
2624 zlog_debug("%u:%s: freeing way for connected prefix",
2625 dest
->selected_fib
->vrf_id
,
2626 prefix2str(&rn
->p
, buf
, sizeof(buf
)));
2627 route_entry_dump(&rn
->p
, NULL
, dest
->selected_fib
);
2629 rib_uninstall(rn
, dest
->selected_fib
);
2634 int rib_add_multipath(afi_t afi
, safi_t safi
, struct prefix
*p
,
2635 struct prefix_ipv6
*src_p
, struct route_entry
*re
)
2637 struct route_table
*table
;
2638 struct route_node
*rn
;
2639 struct route_entry
*same
= NULL
;
2640 struct nhg_hash_entry
*nhe
= NULL
;
2641 struct list
*nhg_depends
= NULL
;
2642 /* Default to route afi */
2643 afi_t nhg_afi
= afi
;
2649 assert(!src_p
|| !src_p
->prefixlen
|| afi
== AFI_IP6
);
2652 table
= zebra_vrf_table_with_table_id(afi
, safi
, re
->vrf_id
, re
->table
);
2654 zebra_nhg_free_group_depends(re
->ng
, nhg_depends
);
2655 XFREE(MTYPE_RE
, re
);
2659 /* Make it sure prefixlen is applied to the prefix. */
2662 apply_mask_ipv6(src_p
);
2664 /* If its a group, create a dependency list */
2665 if (re
->ng
&& re
->ng
->nexthop
->next
) {
2666 struct nexthop
*nh
= NULL
;
2667 struct nexthop lookup
= {0};
2668 struct nhg_hash_entry
*depend
= NULL
;
2670 nhg_depends
= nhg_depend_new_list();
2672 for (ALL_NEXTHOPS_PTR(re
->ng
, nh
)) {
2674 /* Clear it, since its a group */
2676 /* Use the route afi here, since a single nh */
2677 depend
= zebra_nhg_find_nexthop(&lookup
, afi
);
2678 nhg_depend_add(nhg_depends
, depend
);
2681 /* change the afi for group */
2682 if (listcount(nhg_depends
))
2683 nhg_afi
= AFI_UNSPEC
;
2686 nhe
= zebra_nhg_find(re
->ng
, re
->vrf_id
, nhg_afi
, re
->nhe_id
,
2687 nhg_depends
, false);
2690 // TODO: Add interface pointer
2691 zebra_nhg_free_group_depends(re
->ng
, nhg_depends
);
2693 re
->nhe_id
= nhe
->id
;
2694 zebra_nhg_increment_ref(nhe
);
2697 EC_ZEBRA_TABLE_LOOKUP_FAILED
,
2698 "Zebra failed to find or create a nexthop hash entry for id=%u in a route entry",
2703 /* Set default distance by route type. */
2704 if (re
->distance
== 0)
2705 re
->distance
= route_distance(re
->type
);
2707 /* Lookup route node.*/
2708 rn
= srcdest_rnode_get(table
, p
, src_p
);
2711 * If same type of route are installed, treat it as a implicit
2713 * If the user has specified the No route replace semantics
2714 * for the install don't do a route replace.
2716 RNODE_FOREACH_RE (rn
, same
) {
2717 if (CHECK_FLAG(same
->status
, ROUTE_ENTRY_REMOVED
))
2720 if (same
->type
!= re
->type
)
2722 if (same
->instance
!= re
->instance
)
2724 if (same
->type
== ZEBRA_ROUTE_KERNEL
2725 && same
->metric
!= re
->metric
)
2728 if (CHECK_FLAG(re
->flags
, ZEBRA_FLAG_RR_USE_DISTANCE
) &&
2729 same
->distance
!= re
->distance
)
2733 * We should allow duplicate connected routes
2734 * because of IPv6 link-local routes and unnumbered
2735 * interfaces on Linux.
2737 if (same
->type
!= ZEBRA_ROUTE_CONNECT
)
2741 /* If this route is kernel/connected route, notify the dataplane. */
2742 if (RIB_SYSTEM_ROUTE(re
)) {
2743 /* Notify dataplane */
2744 dplane_sys_route_add(rn
, re
);
2747 /* Link new re to node.*/
2748 if (IS_ZEBRA_DEBUG_RIB
) {
2749 rnode_debug(rn
, re
->vrf_id
,
2750 "Inserting route rn %p, re %p (%s) existing %p",
2751 rn
, re
, zebra_route_string(re
->type
), same
);
2753 if (IS_ZEBRA_DEBUG_RIB_DETAILED
)
2754 route_entry_dump(p
, src_p
, re
);
2757 SET_FLAG(re
->status
, ROUTE_ENTRY_CHANGED
);
2758 rib_addnode(rn
, re
, 1);
2761 /* Free implicit route.*/
2763 rib_delnode(rn
, same
);
2767 route_unlock_node(rn
);
2771 void rib_delete(afi_t afi
, safi_t safi
, vrf_id_t vrf_id
, int type
,
2772 unsigned short instance
, int flags
, struct prefix
*p
,
2773 struct prefix_ipv6
*src_p
, const struct nexthop
*nh
,
2774 uint32_t table_id
, uint32_t metric
, uint8_t distance
,
2777 struct route_table
*table
;
2778 struct route_node
*rn
;
2779 struct route_entry
*re
;
2780 struct route_entry
*fib
= NULL
;
2781 struct route_entry
*same
= NULL
;
2782 struct nexthop
*rtnh
;
2783 char buf2
[INET6_ADDRSTRLEN
];
2786 assert(!src_p
|| !src_p
->prefixlen
|| afi
== AFI_IP6
);
2789 table
= zebra_vrf_table_with_table_id(afi
, safi
, vrf_id
, table_id
);
2796 apply_mask_ipv6(src_p
);
2798 /* Lookup route node. */
2799 rn
= srcdest_rnode_lookup(table
, p
, src_p
);
2801 char dst_buf
[PREFIX_STRLEN
], src_buf
[PREFIX_STRLEN
];
2803 prefix2str(p
, dst_buf
, sizeof(dst_buf
));
2804 if (src_p
&& src_p
->prefixlen
)
2805 prefix2str(src_p
, src_buf
, sizeof(src_buf
));
2809 if (IS_ZEBRA_DEBUG_RIB
) {
2810 struct vrf
*vrf
= vrf_lookup_by_id(vrf_id
);
2812 zlog_debug("%s[%d]:%s%s%s doesn't exist in rib",
2813 vrf
->name
, table_id
, dst_buf
,
2814 (src_buf
[0] != '\0') ? " from " : "",
2820 dest
= rib_dest_from_rnode(rn
);
2821 fib
= dest
->selected_fib
;
2823 /* Lookup same type route. */
2824 RNODE_FOREACH_RE (rn
, re
) {
2825 if (CHECK_FLAG(re
->status
, ROUTE_ENTRY_REMOVED
))
2828 if (re
->type
!= type
)
2830 if (re
->instance
!= instance
)
2832 if (CHECK_FLAG(re
->flags
, ZEBRA_FLAG_RR_USE_DISTANCE
) &&
2833 distance
!= re
->distance
)
2836 if (re
->type
== ZEBRA_ROUTE_KERNEL
&& re
->metric
!= metric
)
2838 if (re
->type
== ZEBRA_ROUTE_CONNECT
&& (rtnh
= re
->ng
->nexthop
)
2839 && rtnh
->type
== NEXTHOP_TYPE_IFINDEX
&& nh
) {
2840 if (rtnh
->ifindex
!= nh
->ifindex
)
2845 /* Make sure that the route found has the same gateway. */
2851 for (ALL_NEXTHOPS_PTR(re
->ng
, rtnh
))
2853 * No guarantee all kernel send nh with labels
2856 if (nexthop_same_no_labels(rtnh
, nh
)) {
2864 /* If same type of route can't be found and this message is from
2868 * In the past(HA!) we could get here because
2869 * we were receiving a route delete from the
2870 * kernel and we're not marking the proto
2871 * as coming from it's appropriate originator.
2872 * Now that we are properly noticing the fact
2873 * that the kernel has deleted our route we
2874 * are not going to get called in this path
2875 * I am going to leave this here because
2876 * this might still work this way on non-linux
2877 * platforms as well as some weird state I have
2878 * not properly thought of yet.
2879 * If we can show that this code path is
2880 * dead then we can remove it.
2882 if (fib
&& CHECK_FLAG(flags
, ZEBRA_FLAG_SELFROUTE
)) {
2883 if (IS_ZEBRA_DEBUG_RIB
) {
2884 rnode_debug(rn
, vrf_id
,
2885 "rn %p, re %p (%s) was deleted from kernel, adding",
2887 zebra_route_string(fib
->type
));
2890 UNSET_FLAG(fib
->status
, ROUTE_ENTRY_INSTALLED
);
2892 for (rtnh
= fib
->ng
->nexthop
; rtnh
;
2894 UNSET_FLAG(rtnh
->flags
,
2898 * This is a non FRR route
2899 * as such we should mark
2902 dest
->selected_fib
= NULL
;
2904 /* This means someone else, other than Zebra,
2906 * a Zebra router from the kernel. We will add
2908 rib_install_kernel(rn
, fib
, NULL
);
2911 if (IS_ZEBRA_DEBUG_RIB
) {
2915 "via %s ifindex %d type %d "
2916 "doesn't exist in rib",
2917 inet_ntop(afi2family(afi
),
2924 "type %d doesn't exist in rib",
2927 route_unlock_node(rn
);
2933 if (fromkernel
&& CHECK_FLAG(flags
, ZEBRA_FLAG_SELFROUTE
)
2935 rib_install_kernel(rn
, same
, NULL
);
2936 route_unlock_node(rn
);
2941 /* Special handling for IPv4 or IPv6 routes sourced from
2942 * EVPN - the nexthop (and associated MAC) need to be
2943 * uninstalled if no more refs.
2945 if (CHECK_FLAG(flags
, ZEBRA_FLAG_EVPN_ROUTE
)) {
2946 struct nexthop
*tmp_nh
;
2948 for (ALL_NEXTHOPS_PTR(re
->ng
, tmp_nh
)) {
2949 struct ipaddr vtep_ip
;
2951 memset(&vtep_ip
, 0, sizeof(struct ipaddr
));
2952 if (afi
== AFI_IP
) {
2953 vtep_ip
.ipa_type
= IPADDR_V4
;
2954 memcpy(&(vtep_ip
.ipaddr_v4
),
2955 &(tmp_nh
->gate
.ipv4
),
2956 sizeof(struct in_addr
));
2958 vtep_ip
.ipa_type
= IPADDR_V6
;
2959 memcpy(&(vtep_ip
.ipaddr_v6
),
2960 &(tmp_nh
->gate
.ipv6
),
2961 sizeof(struct in6_addr
));
2963 zebra_vxlan_evpn_vrf_route_del(re
->vrf_id
,
2968 /* Notify dplane if system route changes */
2969 if (RIB_SYSTEM_ROUTE(re
))
2970 dplane_sys_route_del(rn
, same
);
2972 rib_delnode(rn
, same
);
2975 route_unlock_node(rn
);
2980 int rib_add(afi_t afi
, safi_t safi
, vrf_id_t vrf_id
, int type
,
2981 unsigned short instance
, int flags
, struct prefix
*p
,
2982 struct prefix_ipv6
*src_p
, const struct nexthop
*nh
,
2983 uint32_t nhe_id
, uint32_t table_id
, uint32_t metric
, uint32_t mtu
,
2984 uint8_t distance
, route_tag_t tag
)
2986 struct route_entry
*re
= NULL
;
2987 struct nexthop
*nexthop
= NULL
;
2989 /* Allocate new route_entry structure. */
2990 re
= XCALLOC(MTYPE_RE
, sizeof(struct route_entry
));
2992 re
->instance
= instance
;
2993 re
->distance
= distance
;
2995 re
->metric
= metric
;
2997 re
->table
= table_id
;
2998 re
->vrf_id
= vrf_id
;
2999 re
->uptime
= monotime(NULL
);
3001 re
->nhe_id
= nhe_id
;
3003 re
->ng
= nexthop_group_new();
3006 nexthop
= nexthop_new();
3008 route_entry_nexthop_add(re
, nexthop
);
3010 return rib_add_multipath(afi
, safi
, p
, src_p
, re
);
3013 static const char *rib_update_event2str(rib_update_event_t event
)
3015 const char *ret
= "UNKNOWN";
3018 case RIB_UPDATE_KERNEL
:
3019 ret
= "RIB_UPDATE_KERNEL";
3021 case RIB_UPDATE_RMAP_CHANGE
:
3022 ret
= "RIB_UPDATE_RMAP_CHANGE";
3024 case RIB_UPDATE_OTHER
:
3025 ret
= "RIB_UPDATE_OTHER";
3027 case RIB_UPDATE_MAX
:
3035 /* Schedule route nodes to be processed if they match the type */
3036 static void rib_update_route_node(struct route_node
*rn
, int type
)
3038 struct route_entry
*re
, *next
;
3039 bool re_changed
= false;
3041 RNODE_FOREACH_RE_SAFE (rn
, re
, next
) {
3042 if (type
== ZEBRA_ROUTE_ALL
|| type
== re
->type
) {
3043 SET_FLAG(re
->status
, ROUTE_ENTRY_CHANGED
);
3052 /* Schedule routes of a particular table (address-family) based on event. */
3053 void rib_update_table(struct route_table
*table
, rib_update_event_t event
)
3055 struct route_node
*rn
;
3057 if (IS_ZEBRA_DEBUG_EVENT
) {
3058 struct zebra_vrf
*zvrf
;
3061 zvrf
= table
->info
? ((rib_table_info_t
*)table
->info
)->zvrf
3063 vrf
= zvrf
? zvrf
->vrf
: NULL
;
3065 zlog_debug("%s: %s VRF %s Table %u event %s", __func__
,
3066 table
->info
? afi2str(
3067 ((rib_table_info_t
*)table
->info
)->afi
)
3069 vrf
? vrf
->name
: "Unknown",
3070 zvrf
? zvrf
->table_id
: 0,
3071 rib_update_event2str(event
));
3074 /* Walk all routes and queue for processing, if appropriate for
3075 * the trigger event.
3077 for (rn
= route_top(table
); rn
; rn
= srcdest_route_next(rn
)) {
3079 * If we are looking at a route node and the node
3080 * has already been queued we don't
3081 * need to queue it up again
3084 && CHECK_FLAG(rib_dest_from_rnode(rn
)->flags
,
3085 RIB_ROUTE_ANY_QUEUED
))
3089 case RIB_UPDATE_KERNEL
:
3090 rib_update_route_node(rn
, ZEBRA_ROUTE_KERNEL
);
3092 case RIB_UPDATE_RMAP_CHANGE
:
3093 case RIB_UPDATE_OTHER
:
3094 rib_update_route_node(rn
, ZEBRA_ROUTE_ALL
);
3102 static void rib_update_handle_vrf(vrf_id_t vrf_id
, rib_update_event_t event
)
3104 struct route_table
*table
;
3106 if (IS_ZEBRA_DEBUG_EVENT
)
3107 zlog_debug("%s: Handling VRF %s event %s", __func__
,
3108 vrf_id_to_name(vrf_id
), rib_update_event2str(event
));
3110 /* Process routes of interested address-families. */
3111 table
= zebra_vrf_table(AFI_IP
, SAFI_UNICAST
, vrf_id
);
3113 rib_update_table(table
, event
);
3115 table
= zebra_vrf_table(AFI_IP6
, SAFI_UNICAST
, vrf_id
);
3117 rib_update_table(table
, event
);
3120 static void rib_update_handle_vrf_all(rib_update_event_t event
)
3122 struct zebra_router_table
*zrt
;
3124 if (IS_ZEBRA_DEBUG_EVENT
)
3125 zlog_debug("%s: Handling VRF (ALL) event %s", __func__
,
3126 rib_update_event2str(event
));
3128 /* Just iterate over all the route tables, rather than vrf lookups */
3129 RB_FOREACH (zrt
, zebra_router_table_head
, &zrouter
.tables
)
3130 rib_update_table(zrt
->table
, event
);
3133 struct rib_update_ctx
{
3134 rib_update_event_t event
;
3139 static struct rib_update_ctx
*rib_update_ctx_init(vrf_id_t vrf_id
,
3140 rib_update_event_t event
)
3142 struct rib_update_ctx
*ctx
;
3144 ctx
= XCALLOC(MTYPE_RIB_UPDATE_CTX
, sizeof(struct rib_update_ctx
));
3147 ctx
->vrf_id
= vrf_id
;
3152 static void rib_update_ctx_fini(struct rib_update_ctx
**ctx
)
3154 XFREE(MTYPE_RIB_UPDATE_CTX
, *ctx
);
3159 static int rib_update_handler(struct thread
*thread
)
3161 struct rib_update_ctx
*ctx
;
3163 ctx
= THREAD_ARG(thread
);
3166 rib_update_handle_vrf_all(ctx
->event
);
3168 rib_update_handle_vrf(ctx
->vrf_id
, ctx
->event
);
3170 rib_update_ctx_fini(&ctx
);
3176 * Thread list to ensure we don't schedule a ton of events
3177 * if interfaces are flapping for instance.
3179 static struct thread
*t_rib_update_threads
[RIB_UPDATE_MAX
];
3181 /* Schedule a RIB update event for specific vrf */
3182 void rib_update_vrf(vrf_id_t vrf_id
, rib_update_event_t event
)
3184 struct rib_update_ctx
*ctx
;
3186 ctx
= rib_update_ctx_init(vrf_id
, event
);
3188 /* Don't worry about making sure multiple rib updates for specific vrf
3189 * are scheduled at once for now. If it becomes a problem, we can use a
3190 * lookup of some sort to keep track of running threads via t_vrf_id
3191 * like how we are doing it in t_rib_update_threads[].
3193 thread_add_event(zrouter
.master
, rib_update_handler
, ctx
, 0, NULL
);
3195 if (IS_ZEBRA_DEBUG_EVENT
)
3196 zlog_debug("%s: Scheduled VRF %s, event %s", __func__
,
3197 vrf_id_to_name(ctx
->vrf_id
),
3198 rib_update_event2str(event
));
3201 /* Schedule a RIB update event for all vrfs */
3202 void rib_update(rib_update_event_t event
)
3204 struct rib_update_ctx
*ctx
;
3206 ctx
= rib_update_ctx_init(0, event
);
3208 ctx
->vrf_all
= true;
3210 if (!thread_add_event(zrouter
.master
, rib_update_handler
, ctx
, 0,
3211 &t_rib_update_threads
[event
]))
3212 rib_update_ctx_fini(&ctx
); /* Already scheduled */
3213 else if (IS_ZEBRA_DEBUG_EVENT
)
3214 zlog_debug("%s: Schedued VRF (ALL), event %s", __func__
,
3215 rib_update_event2str(event
));
3218 /* Delete self installed routes after zebra is relaunched. */
3219 void rib_sweep_table(struct route_table
*table
)
3221 struct route_node
*rn
;
3222 struct route_entry
*re
;
3223 struct route_entry
*next
;
3224 struct nexthop
*nexthop
;
3229 for (rn
= route_top(table
); rn
; rn
= srcdest_route_next(rn
)) {
3230 RNODE_FOREACH_RE_SAFE (rn
, re
, next
) {
3232 if (IS_ZEBRA_DEBUG_RIB
)
3233 route_entry_dump(&rn
->p
, NULL
, re
);
3235 if (CHECK_FLAG(re
->status
, ROUTE_ENTRY_REMOVED
))
3238 if (!CHECK_FLAG(re
->flags
, ZEBRA_FLAG_SELFROUTE
))
3242 * If routes are older than startup_time then
3243 * we know we read them in from the kernel.
3244 * As such we can safely remove them.
3246 if (zrouter
.startup_time
< re
->uptime
)
3250 * So we are starting up and have received
3251 * routes from the kernel that we have installed
3252 * from a previous run of zebra but not cleaned
3253 * up ( say a kill -9 )
3254 * But since we haven't actually installed
3255 * them yet( we received them from the kernel )
3256 * we don't think they are active.
3257 * So let's pretend they are active to actually
3259 * In all honesty I'm not sure if we should
3260 * mark them as active when we receive them
3261 * This is startup only so probably ok.
3263 * If we ever decide to move rib_sweep_table
3264 * to a different spot (ie startup )
3265 * this decision needs to be revisited
3267 SET_FLAG(re
->status
, ROUTE_ENTRY_INSTALLED
);
3268 for (ALL_NEXTHOPS_PTR(re
->ng
, nexthop
))
3269 SET_FLAG(nexthop
->flags
, NEXTHOP_FLAG_FIB
);
3271 rib_uninstall_kernel(rn
, re
);
3272 rib_delnode(rn
, re
);
3277 /* Sweep all RIB tables. */
3278 int rib_sweep_route(struct thread
*t
)
3281 struct zebra_vrf
*zvrf
;
3283 RB_FOREACH (vrf
, vrf_id_head
, &vrfs_by_id
) {
3284 if ((zvrf
= vrf
->info
) == NULL
)
3287 rib_sweep_table(zvrf
->table
[AFI_IP
][SAFI_UNICAST
]);
3288 rib_sweep_table(zvrf
->table
[AFI_IP6
][SAFI_UNICAST
]);
3291 zebra_router_sweep_route();
3296 /* Remove specific by protocol routes from 'table'. */
3297 unsigned long rib_score_proto_table(uint8_t proto
, unsigned short instance
,
3298 struct route_table
*table
)
3300 struct route_node
*rn
;
3301 struct route_entry
*re
;
3302 struct route_entry
*next
;
3303 unsigned long n
= 0;
3306 for (rn
= route_top(table
); rn
; rn
= srcdest_route_next(rn
))
3307 RNODE_FOREACH_RE_SAFE (rn
, re
, next
) {
3308 if (CHECK_FLAG(re
->status
, ROUTE_ENTRY_REMOVED
))
3310 if (re
->type
== proto
3311 && re
->instance
== instance
) {
3312 rib_delnode(rn
, re
);
3319 /* Remove specific by protocol routes. */
3320 unsigned long rib_score_proto(uint8_t proto
, unsigned short instance
)
3323 struct zebra_vrf
*zvrf
;
3324 struct other_route_table
*ort
;
3325 unsigned long cnt
= 0;
3327 RB_FOREACH (vrf
, vrf_id_head
, &vrfs_by_id
) {
3332 cnt
+= rib_score_proto_table(proto
, instance
,
3333 zvrf
->table
[AFI_IP
][SAFI_UNICAST
])
3334 + rib_score_proto_table(
3336 zvrf
->table
[AFI_IP6
][SAFI_UNICAST
]);
3338 frr_each(otable
, &zvrf
->other_tables
, ort
) cnt
+=
3339 rib_score_proto_table(proto
, instance
, ort
->table
);
3345 /* Close RIB and clean up kernel routes. */
3346 void rib_close_table(struct route_table
*table
)
3348 struct route_node
*rn
;
3349 rib_table_info_t
*info
;
3355 info
= route_table_get_info(table
);
3357 for (rn
= route_top(table
); rn
; rn
= srcdest_route_next(rn
)) {
3358 dest
= rib_dest_from_rnode(rn
);
3360 if (dest
&& dest
->selected_fib
) {
3361 if (info
->safi
== SAFI_UNICAST
)
3362 hook_call(rib_update
, rn
, NULL
);
3364 rib_uninstall_kernel(rn
, dest
->selected_fib
);
3365 dest
->selected_fib
= NULL
;
3371 * Handler for async dataplane results after a pseudowire installation
3373 static int handle_pw_result(struct zebra_dplane_ctx
*ctx
)
3375 struct zebra_pw
*pw
;
3376 struct zebra_vrf
*vrf
;
3378 /* The pseudowire code assumes success - we act on an error
3379 * result for installation attempts here.
3381 if (dplane_ctx_get_op(ctx
) != DPLANE_OP_PW_INSTALL
)
3384 if (dplane_ctx_get_status(ctx
) != ZEBRA_DPLANE_REQUEST_SUCCESS
) {
3385 vrf
= zebra_vrf_lookup_by_id(dplane_ctx_get_vrf(ctx
));
3386 pw
= zebra_pw_find(vrf
, dplane_ctx_get_ifname(ctx
));
3388 zebra_pw_install_failure(pw
);
3398 * Handle results from the dataplane system. Dequeue update context
3399 * structs, dispatch to appropriate internal handlers.
3401 static int rib_process_dplane_results(struct thread
*thread
)
3403 struct zebra_dplane_ctx
*ctx
;
3404 struct dplane_ctx_q ctxlist
;
3405 bool shut_p
= false;
3407 /* Dequeue a list of completed updates with one lock/unlock cycle */
3410 TAILQ_INIT(&ctxlist
);
3412 /* Take lock controlling queue of results */
3413 frr_with_mutex(&dplane_mutex
) {
3414 /* Dequeue list of context structs */
3415 dplane_ctx_list_append(&ctxlist
, &rib_dplane_q
);
3418 /* Dequeue context block */
3419 ctx
= dplane_ctx_dequeue(&ctxlist
);
3421 /* If we've emptied the results queue, we're done */
3425 /* If zebra is shutting down, avoid processing results,
3426 * just drain the results queue.
3428 shut_p
= atomic_load_explicit(&zrouter
.in_shutdown
,
3429 memory_order_relaxed
);
3432 dplane_ctx_fini(&ctx
);
3434 ctx
= dplane_ctx_dequeue(&ctxlist
);
3441 switch (dplane_ctx_get_op(ctx
)) {
3442 case DPLANE_OP_ROUTE_INSTALL
:
3443 case DPLANE_OP_ROUTE_UPDATE
:
3444 case DPLANE_OP_ROUTE_DELETE
:
3446 /* Bit of special case for route updates
3447 * that were generated by async notifications:
3448 * we don't want to continue processing these
3451 if (dplane_ctx_get_notif_provider(ctx
) == 0)
3452 rib_process_result(ctx
);
3454 dplane_ctx_fini(&ctx
);
3458 case DPLANE_OP_ROUTE_NOTIFY
:
3459 rib_process_dplane_notify(ctx
);
3462 case DPLANE_OP_NH_INSTALL
:
3463 case DPLANE_OP_NH_UPDATE
:
3464 case DPLANE_OP_NH_DELETE
:
3465 zebra_nhg_dplane_result(ctx
);
3468 case DPLANE_OP_LSP_INSTALL
:
3469 case DPLANE_OP_LSP_UPDATE
:
3470 case DPLANE_OP_LSP_DELETE
:
3472 /* Bit of special case for LSP updates
3473 * that were generated by async notifications:
3474 * we don't want to continue processing these.
3476 if (dplane_ctx_get_notif_provider(ctx
) == 0)
3477 zebra_mpls_lsp_dplane_result(ctx
);
3479 dplane_ctx_fini(&ctx
);
3483 case DPLANE_OP_LSP_NOTIFY
:
3484 zebra_mpls_process_dplane_notify(ctx
);
3487 case DPLANE_OP_PW_INSTALL
:
3488 case DPLANE_OP_PW_UNINSTALL
:
3489 handle_pw_result(ctx
);
3492 case DPLANE_OP_SYS_ROUTE_ADD
:
3493 case DPLANE_OP_SYS_ROUTE_DELETE
:
3494 /* No further processing in zebra for these. */
3495 dplane_ctx_fini(&ctx
);
3498 case DPLANE_OP_MAC_INSTALL
:
3499 case DPLANE_OP_MAC_DELETE
:
3500 zebra_vxlan_handle_result(ctx
);
3503 /* Some op codes not handled here */
3504 case DPLANE_OP_ADDR_INSTALL
:
3505 case DPLANE_OP_ADDR_UNINSTALL
:
3506 case DPLANE_OP_NEIGH_INSTALL
:
3507 case DPLANE_OP_NEIGH_UPDATE
:
3508 case DPLANE_OP_NEIGH_DELETE
:
3509 case DPLANE_OP_VTEP_ADD
:
3510 case DPLANE_OP_VTEP_DELETE
:
3511 case DPLANE_OP_NONE
:
3512 /* Don't expect this: just return the struct? */
3513 dplane_ctx_fini(&ctx
);
3516 } /* Dispatch by op code */
3518 ctx
= dplane_ctx_dequeue(&ctxlist
);
3527 * Results are returned from the dataplane subsystem, in the context of
3528 * the dataplane pthread. We enqueue the results here for processing by
3529 * the main thread later.
3531 static int rib_dplane_results(struct dplane_ctx_q
*ctxlist
)
3533 /* Take lock controlling queue of results */
3534 frr_with_mutex(&dplane_mutex
) {
3535 /* Enqueue context blocks */
3536 dplane_ctx_list_append(&rib_dplane_q
, ctxlist
);
3539 /* Ensure event is signalled to zebra main pthread */
3540 thread_add_event(zrouter
.master
, rib_process_dplane_results
, NULL
, 0,
3547 * Ensure there are no empty slots in the route_info array.
3548 * Every route type in zebra should be present there.
3550 static void check_route_info(void)
3552 int len
= array_size(route_info
);
3555 * ZEBRA_ROUTE_SYSTEM is special cased since
3556 * its key is 0 anyway.
3558 * ZEBRA_ROUTE_ALL is also ignored.
3560 for (int i
= 0; i
< len
; i
++) {
3561 if (i
== ZEBRA_ROUTE_SYSTEM
|| i
== ZEBRA_ROUTE_ALL
)
3563 assert(route_info
[i
].key
);
3564 assert(route_info
[i
].meta_q_map
< MQ_SIZE
);
3568 /* Routing information base initialize. */
3575 /* Init dataplane, and register for results */
3576 pthread_mutex_init(&dplane_mutex
, NULL
);
3577 TAILQ_INIT(&rib_dplane_q
);
3578 zebra_dplane_init(rib_dplane_results
);
3584 * Get the first vrf id that is greater than the given vrf id if any.
3586 * Returns true if a vrf id was found, false otherwise.
3588 static inline int vrf_id_get_next(vrf_id_t vrf_id
, vrf_id_t
*next_id_p
)
3592 vrf
= vrf_lookup_by_id(vrf_id
);
3594 vrf
= RB_NEXT(vrf_id_head
, vrf
);
3596 *next_id_p
= vrf
->vrf_id
;
3605 * rib_tables_iter_next
3607 * Returns the next table in the iteration.
3609 struct route_table
*rib_tables_iter_next(rib_tables_iter_t
*iter
)
3611 struct route_table
*table
;
3614 * Array that helps us go over all AFI/SAFI combinations via one
3621 {AFI_IP
, SAFI_UNICAST
}, {AFI_IP
, SAFI_MULTICAST
},
3622 {AFI_IP
, SAFI_LABELED_UNICAST
}, {AFI_IP6
, SAFI_UNICAST
},
3623 {AFI_IP6
, SAFI_MULTICAST
}, {AFI_IP6
, SAFI_LABELED_UNICAST
},
3628 switch (iter
->state
) {
3630 case RIB_TABLES_ITER_S_INIT
:
3631 iter
->vrf_id
= VRF_DEFAULT
;
3632 iter
->afi_safi_ix
= -1;
3636 case RIB_TABLES_ITER_S_ITERATING
:
3637 iter
->afi_safi_ix
++;
3640 while (iter
->afi_safi_ix
3641 < (int)array_size(afi_safis
)) {
3642 table
= zebra_vrf_table(
3643 afi_safis
[iter
->afi_safi_ix
].afi
,
3644 afi_safis
[iter
->afi_safi_ix
].safi
,
3649 iter
->afi_safi_ix
++;
3653 * Found another table in this vrf.
3659 * Done with all tables in the current vrf, go to the
3663 if (!vrf_id_get_next(iter
->vrf_id
, &iter
->vrf_id
))
3666 iter
->afi_safi_ix
= 0;
3671 case RIB_TABLES_ITER_S_DONE
:
3676 iter
->state
= RIB_TABLES_ITER_S_ITERATING
;
3678 iter
->state
= RIB_TABLES_ITER_S_DONE
;