1 /* Routing Information Base.
2 * Copyright (C) 1997, 98, 99, 2001 Kunihiro Ishiguro
4 * This file is part of GNU Zebra.
6 * GNU Zebra is free software; you can redistribute it and/or modify it
7 * under the terms of the GNU General Public License as published by the
8 * Free Software Foundation; either version 2, or (at your option) any
11 * GNU Zebra is distributed in the hope that it will be useful, but
12 * WITHOUT ANY WARRANTY; without even the implied warranty of
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
14 * General Public License for more details.
16 * You should have received a copy of the GNU General Public License along
17 * with this program; see the file COPYING; if not, write to the Free Software
18 * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
33 #include "sockunion.h"
34 #include "srcdest_table.h"
38 #include "workqueue.h"
39 #include "nexthop_group_private.h"
40 #include "frr_pthread.h"
42 #include "zebra/zebra_router.h"
43 #include "zebra/connected.h"
44 #include "zebra/debug.h"
45 #include "zebra/interface.h"
46 #include "zebra/redistribute.h"
47 #include "zebra/rib.h"
49 #include "zebra/zapi_msg.h"
50 #include "zebra/zebra_errors.h"
51 #include "zebra/zebra_memory.h"
52 #include "zebra/zebra_ns.h"
53 #include "zebra/zebra_rnh.h"
54 #include "zebra/zebra_routemap.h"
55 #include "zebra/zebra_vrf.h"
56 #include "zebra/zebra_vxlan.h"
57 #include "zebra/zapi_msg.h"
58 #include "zebra/zebra_dplane.h"
60 DEFINE_MTYPE_STATIC(ZEBRA
, RIB_UPDATE_CTX
, "Rib update context object");
63 * Event, list, and mutex for delivery of dataplane results
65 static pthread_mutex_t dplane_mutex
;
66 static struct thread
*t_dplane
;
67 static struct dplane_ctx_q rib_dplane_q
;
69 DEFINE_HOOK(rib_update
, (struct route_node
* rn
, const char *reason
),
72 /* Should we allow non Quagga processes to delete our routes */
73 extern int allow_delete
;
75 /* Each route type's string and default distance value. */
80 } route_info
[ZEBRA_ROUTE_MAX
] = {
81 [ZEBRA_ROUTE_NHG
] = {ZEBRA_ROUTE_NHG
, 255 /* Uneeded for nhg's */, 0},
82 [ZEBRA_ROUTE_SYSTEM
] = {ZEBRA_ROUTE_SYSTEM
, 0, 5},
83 [ZEBRA_ROUTE_KERNEL
] = {ZEBRA_ROUTE_KERNEL
, 0, 1},
84 [ZEBRA_ROUTE_CONNECT
] = {ZEBRA_ROUTE_CONNECT
, 0, 1},
85 [ZEBRA_ROUTE_STATIC
] = {ZEBRA_ROUTE_STATIC
, 1, 2},
86 [ZEBRA_ROUTE_RIP
] = {ZEBRA_ROUTE_RIP
, 120, 3},
87 [ZEBRA_ROUTE_RIPNG
] = {ZEBRA_ROUTE_RIPNG
, 120, 3},
88 [ZEBRA_ROUTE_OSPF
] = {ZEBRA_ROUTE_OSPF
, 110, 3},
89 [ZEBRA_ROUTE_OSPF6
] = {ZEBRA_ROUTE_OSPF6
, 110, 3},
90 [ZEBRA_ROUTE_ISIS
] = {ZEBRA_ROUTE_ISIS
, 115, 3},
91 [ZEBRA_ROUTE_BGP
] = {ZEBRA_ROUTE_BGP
, 20 /* IBGP is 200. */, 4},
92 [ZEBRA_ROUTE_PIM
] = {ZEBRA_ROUTE_PIM
, 255, 5},
93 [ZEBRA_ROUTE_EIGRP
] = {ZEBRA_ROUTE_EIGRP
, 90, 3},
94 [ZEBRA_ROUTE_NHRP
] = {ZEBRA_ROUTE_NHRP
, 10, 3},
95 [ZEBRA_ROUTE_HSLS
] = {ZEBRA_ROUTE_HSLS
, 255, 5},
96 [ZEBRA_ROUTE_OLSR
] = {ZEBRA_ROUTE_OLSR
, 255, 5},
97 [ZEBRA_ROUTE_TABLE
] = {ZEBRA_ROUTE_TABLE
, 150, 2},
98 [ZEBRA_ROUTE_LDP
] = {ZEBRA_ROUTE_LDP
, 150, 5},
99 [ZEBRA_ROUTE_VNC
] = {ZEBRA_ROUTE_VNC
, 20, 4},
100 [ZEBRA_ROUTE_VNC_DIRECT
] = {ZEBRA_ROUTE_VNC_DIRECT
, 20, 4},
101 [ZEBRA_ROUTE_VNC_DIRECT_RH
] = {ZEBRA_ROUTE_VNC_DIRECT_RH
, 20, 4},
102 [ZEBRA_ROUTE_BGP_DIRECT
] = {ZEBRA_ROUTE_BGP_DIRECT
, 20, 4},
103 [ZEBRA_ROUTE_BGP_DIRECT_EXT
] = {ZEBRA_ROUTE_BGP_DIRECT_EXT
, 20, 4},
104 [ZEBRA_ROUTE_BABEL
] = {ZEBRA_ROUTE_BABEL
, 100, 3},
105 [ZEBRA_ROUTE_SHARP
] = {ZEBRA_ROUTE_SHARP
, 150, 5},
106 [ZEBRA_ROUTE_PBR
] = {ZEBRA_ROUTE_PBR
, 200, 5},
107 [ZEBRA_ROUTE_BFD
] = {ZEBRA_ROUTE_BFD
, 255, 5},
108 [ZEBRA_ROUTE_OPENFABRIC
] = {ZEBRA_ROUTE_OPENFABRIC
, 115, 3},
109 [ZEBRA_ROUTE_VRRP
] = {ZEBRA_ROUTE_VRRP
, 255, 5}
110 /* Any new route type added to zebra, should be mirrored here */
112 /* no entry/default: 150 */
115 static void __attribute__((format(printf
, 5, 6)))
116 _rnode_zlog(const char *_func
, vrf_id_t vrf_id
, struct route_node
*rn
,
117 int priority
, const char *msgfmt
, ...)
119 char buf
[SRCDEST2STR_BUFFER
+ sizeof(" (MRIB)")];
123 va_start(ap
, msgfmt
);
124 vsnprintf(msgbuf
, sizeof(msgbuf
), msgfmt
, ap
);
128 rib_table_info_t
*info
= srcdest_rnode_table_info(rn
);
129 srcdest_rnode2str(rn
, buf
, sizeof(buf
));
131 if (info
->safi
== SAFI_MULTICAST
)
132 strlcat(buf
, " (MRIB)", sizeof(buf
));
134 snprintf(buf
, sizeof(buf
), "{(route_node *) NULL}");
137 zlog(priority
, "%s: %d:%s: %s", _func
, vrf_id
, buf
, msgbuf
);
140 #define rnode_debug(node, vrf_id, ...) \
141 _rnode_zlog(__func__, vrf_id, node, LOG_DEBUG, __VA_ARGS__)
142 #define rnode_info(node, ...) \
143 _rnode_zlog(__func__, vrf_id, node, LOG_INFO, __VA_ARGS__)
145 uint8_t route_distance(int type
)
149 if ((unsigned)type
>= array_size(route_info
))
152 distance
= route_info
[type
].distance
;
157 int is_zebra_valid_kernel_table(uint32_t table_id
)
160 if ((table_id
== RT_TABLE_UNSPEC
) || (table_id
== RT_TABLE_LOCAL
)
161 || (table_id
== RT_TABLE_COMPAT
))
168 int is_zebra_main_routing_table(uint32_t table_id
)
170 if (table_id
== RT_TABLE_MAIN
)
175 int zebra_check_addr(const struct prefix
*p
)
177 if (p
->family
== AF_INET
) {
180 addr
= p
->u
.prefix4
.s_addr
;
183 if (IPV4_NET127(addr
) || IN_CLASSD(addr
)
184 || IPV4_LINKLOCAL(addr
))
187 if (p
->family
== AF_INET6
) {
188 if (IN6_IS_ADDR_LOOPBACK(&p
->u
.prefix6
))
190 if (IN6_IS_ADDR_LINKLOCAL(&p
->u
.prefix6
))
196 /* Add nexthop to the end of a rib node's nexthop list */
197 void route_entry_nexthop_add(struct route_entry
*re
, struct nexthop
*nexthop
)
199 _nexthop_group_add_sorted(re
->ng
, nexthop
);
204 * copy_nexthop - copy a nexthop to the rib structure.
206 void route_entry_copy_nexthops(struct route_entry
*re
, struct nexthop
*nh
)
208 assert(!re
->ng
->nexthop
);
209 copy_nexthops(&re
->ng
->nexthop
, nh
, NULL
);
212 /* Delete specified nexthop from the list. */
213 void route_entry_nexthop_delete(struct route_entry
*re
, struct nexthop
*nexthop
)
216 nexthop
->next
->prev
= nexthop
->prev
;
218 nexthop
->prev
->next
= nexthop
->next
;
220 re
->ng
->nexthop
= nexthop
->next
;
224 struct nexthop
*route_entry_nexthop_ifindex_add(struct route_entry
*re
,
228 struct nexthop
*nexthop
;
230 nexthop
= nexthop_new();
231 nexthop
->type
= NEXTHOP_TYPE_IFINDEX
;
232 nexthop
->ifindex
= ifindex
;
233 nexthop
->vrf_id
= nh_vrf_id
;
235 route_entry_nexthop_add(re
, nexthop
);
240 struct nexthop
*route_entry_nexthop_ipv4_add(struct route_entry
*re
,
241 struct in_addr
*ipv4
,
245 struct nexthop
*nexthop
;
247 nexthop
= nexthop_new();
248 nexthop
->type
= NEXTHOP_TYPE_IPV4
;
249 nexthop
->vrf_id
= nh_vrf_id
;
250 nexthop
->gate
.ipv4
= *ipv4
;
252 nexthop
->src
.ipv4
= *src
;
254 route_entry_nexthop_add(re
, nexthop
);
259 struct nexthop
*route_entry_nexthop_ipv4_ifindex_add(struct route_entry
*re
,
260 struct in_addr
*ipv4
,
265 struct nexthop
*nexthop
;
266 struct interface
*ifp
;
268 nexthop
= nexthop_new();
269 nexthop
->vrf_id
= nh_vrf_id
;
270 nexthop
->type
= NEXTHOP_TYPE_IPV4_IFINDEX
;
271 nexthop
->gate
.ipv4
= *ipv4
;
273 nexthop
->src
.ipv4
= *src
;
274 nexthop
->ifindex
= ifindex
;
275 ifp
= if_lookup_by_index(nexthop
->ifindex
, nh_vrf_id
);
276 /*Pending: need to think if null ifp here is ok during bootup?
277 There was a crash because ifp here was coming to be NULL */
279 if (connected_is_unnumbered(ifp
))
280 SET_FLAG(nexthop
->flags
, NEXTHOP_FLAG_ONLINK
);
282 route_entry_nexthop_add(re
, nexthop
);
287 struct nexthop
*route_entry_nexthop_ipv6_add(struct route_entry
*re
,
288 struct in6_addr
*ipv6
,
291 struct nexthop
*nexthop
;
293 nexthop
= nexthop_new();
294 nexthop
->vrf_id
= nh_vrf_id
;
295 nexthop
->type
= NEXTHOP_TYPE_IPV6
;
296 nexthop
->gate
.ipv6
= *ipv6
;
298 route_entry_nexthop_add(re
, nexthop
);
303 struct nexthop
*route_entry_nexthop_ipv6_ifindex_add(struct route_entry
*re
,
304 struct in6_addr
*ipv6
,
308 struct nexthop
*nexthop
;
310 nexthop
= nexthop_new();
311 nexthop
->vrf_id
= nh_vrf_id
;
312 nexthop
->type
= NEXTHOP_TYPE_IPV6_IFINDEX
;
313 nexthop
->gate
.ipv6
= *ipv6
;
314 nexthop
->ifindex
= ifindex
;
316 route_entry_nexthop_add(re
, nexthop
);
321 struct nexthop
*route_entry_nexthop_blackhole_add(struct route_entry
*re
,
322 enum blackhole_type bh_type
)
324 struct nexthop
*nexthop
;
326 nexthop
= nexthop_new();
327 nexthop
->vrf_id
= VRF_DEFAULT
;
328 nexthop
->type
= NEXTHOP_TYPE_BLACKHOLE
;
329 nexthop
->bh_type
= bh_type
;
331 route_entry_nexthop_add(re
, nexthop
);
336 struct route_entry
*rib_match(afi_t afi
, safi_t safi
, vrf_id_t vrf_id
,
337 union g_addr
*addr
, struct route_node
**rn_out
)
340 struct route_table
*table
;
341 struct route_node
*rn
;
342 struct route_entry
*match
= NULL
;
345 table
= zebra_vrf_table(afi
, safi
, vrf_id
);
349 memset(&p
, 0, sizeof(struct prefix
));
352 p
.u
.prefix4
= addr
->ipv4
;
353 p
.prefixlen
= IPV4_MAX_PREFIXLEN
;
355 p
.u
.prefix6
= addr
->ipv6
;
356 p
.prefixlen
= IPV6_MAX_PREFIXLEN
;
359 rn
= route_node_match(table
, (struct prefix
*)&p
);
364 route_unlock_node(rn
);
366 dest
= rib_dest_from_rnode(rn
);
367 if (dest
&& dest
->selected_fib
368 && !CHECK_FLAG(dest
->selected_fib
->status
,
369 ROUTE_ENTRY_REMOVED
))
370 match
= dest
->selected_fib
;
372 /* If there is no selected route or matched route is EGP, go up
377 } while (rn
&& rn
->info
== NULL
);
381 if (match
->type
!= ZEBRA_ROUTE_CONNECT
) {
382 if (!CHECK_FLAG(match
->status
,
383 ROUTE_ENTRY_INSTALLED
))
395 struct route_entry
*rib_match_ipv4_multicast(vrf_id_t vrf_id
,
397 struct route_node
**rn_out
)
399 struct route_entry
*re
= NULL
, *mre
= NULL
, *ure
= NULL
;
400 struct route_node
*m_rn
= NULL
, *u_rn
= NULL
;
401 union g_addr gaddr
= {.ipv4
= addr
};
403 switch (zrouter
.ipv4_multicast_mode
) {
404 case MCAST_MRIB_ONLY
:
405 return rib_match(AFI_IP
, SAFI_MULTICAST
, vrf_id
, &gaddr
,
407 case MCAST_URIB_ONLY
:
408 return rib_match(AFI_IP
, SAFI_UNICAST
, vrf_id
, &gaddr
, rn_out
);
409 case MCAST_NO_CONFIG
:
410 case MCAST_MIX_MRIB_FIRST
:
411 re
= mre
= rib_match(AFI_IP
, SAFI_MULTICAST
, vrf_id
, &gaddr
,
414 re
= ure
= rib_match(AFI_IP
, SAFI_UNICAST
, vrf_id
,
417 case MCAST_MIX_DISTANCE
:
418 mre
= rib_match(AFI_IP
, SAFI_MULTICAST
, vrf_id
, &gaddr
, &m_rn
);
419 ure
= rib_match(AFI_IP
, SAFI_UNICAST
, vrf_id
, &gaddr
, &u_rn
);
421 re
= ure
->distance
< mre
->distance
? ure
: mre
;
427 case MCAST_MIX_PFXLEN
:
428 mre
= rib_match(AFI_IP
, SAFI_MULTICAST
, vrf_id
, &gaddr
, &m_rn
);
429 ure
= rib_match(AFI_IP
, SAFI_UNICAST
, vrf_id
, &gaddr
, &u_rn
);
431 re
= u_rn
->p
.prefixlen
> m_rn
->p
.prefixlen
? ure
: mre
;
440 *rn_out
= (re
== mre
) ? m_rn
: u_rn
;
442 if (IS_ZEBRA_DEBUG_RIB
) {
444 inet_ntop(AF_INET
, &addr
, buf
, BUFSIZ
);
446 zlog_debug("%s: %s: vrf: %u found %s, using %s",
447 __func__
, buf
, vrf_id
,
448 mre
? (ure
? "MRIB+URIB" : "MRIB")
449 : ure
? "URIB" : "nothing",
450 re
== ure
? "URIB" : re
== mre
? "MRIB" : "none");
455 struct route_entry
*rib_lookup_ipv4(struct prefix_ipv4
*p
, vrf_id_t vrf_id
)
457 struct route_table
*table
;
458 struct route_node
*rn
;
459 struct route_entry
*match
= NULL
;
463 table
= zebra_vrf_table(AFI_IP
, SAFI_UNICAST
, vrf_id
);
467 rn
= route_node_lookup(table
, (struct prefix
*)p
);
469 /* No route for this prefix. */
474 route_unlock_node(rn
);
475 dest
= rib_dest_from_rnode(rn
);
477 if (dest
&& dest
->selected_fib
478 && !CHECK_FLAG(dest
->selected_fib
->status
, ROUTE_ENTRY_REMOVED
))
479 match
= dest
->selected_fib
;
484 if (match
->type
== ZEBRA_ROUTE_CONNECT
)
487 if (CHECK_FLAG(match
->status
, ROUTE_ENTRY_INSTALLED
))
494 * Is this RIB labeled-unicast? It must be of type BGP and all paths
495 * (nexthops) must have a label.
497 int zebra_rib_labeled_unicast(struct route_entry
*re
)
499 struct nexthop
*nexthop
= NULL
;
501 if (re
->type
!= ZEBRA_ROUTE_BGP
)
504 for (ALL_NEXTHOPS_PTR(re
->ng
, nexthop
))
505 if (!nexthop
->nh_label
|| !nexthop
->nh_label
->num_labels
)
511 /* Update flag indicates whether this is a "replace" or not. Currently, this
512 * is only used for IPv4.
514 void rib_install_kernel(struct route_node
*rn
, struct route_entry
*re
,
515 struct route_entry
*old
)
517 struct nexthop
*nexthop
;
518 rib_table_info_t
*info
= srcdest_rnode_table_info(rn
);
519 struct zebra_vrf
*zvrf
= vrf_info_lookup(re
->vrf_id
);
520 const struct prefix
*p
, *src_p
;
521 enum zebra_dplane_result ret
;
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 * Install the resolved nexthop object first.
551 zebra_nhg_install_kernel(zebra_nhg_lookup_id(re
->nhe_id
));
554 * If this is a replace to a new RE let the originator of the RE
555 * know that they've lost
557 if (old
&& (old
!= re
) && (old
->type
!= re
->type
))
558 zsend_route_notify_owner(old
, p
, ZAPI_ROUTE_BETTER_ADMIN_WON
);
560 /* Update fib selection */
561 dest
->selected_fib
= re
;
564 * Make sure we update the FPM any time we send new information to
567 hook_call(rib_update
, rn
, "installing in kernel");
569 /* Send add or update */
571 ret
= dplane_route_update(rn
, re
, old
);
573 ret
= dplane_route_add(rn
, re
);
576 case ZEBRA_DPLANE_REQUEST_QUEUED
:
577 SET_FLAG(re
->status
, ROUTE_ENTRY_QUEUED
);
580 SET_FLAG(old
->status
, ROUTE_ENTRY_QUEUED
);
582 /* Free old FIB nexthop group */
583 if (old
->fib_ng
.nexthop
) {
584 nexthops_free(old
->fib_ng
.nexthop
);
585 old
->fib_ng
.nexthop
= NULL
;
588 if (!RIB_SYSTEM_ROUTE(old
)) {
589 /* Clear old route's FIB flags */
590 for (ALL_NEXTHOPS_PTR(old
->ng
, nexthop
)) {
591 UNSET_FLAG(nexthop
->flags
,
598 zvrf
->installs_queued
++;
600 case ZEBRA_DPLANE_REQUEST_FAILURE
:
602 char str
[SRCDEST2STR_BUFFER
];
604 srcdest_rnode2str(rn
, str
, sizeof(str
));
605 flog_err(EC_ZEBRA_DP_INSTALL_FAIL
,
606 "%u:%s: Failed to enqueue dataplane install",
610 case ZEBRA_DPLANE_REQUEST_SUCCESS
:
619 /* Uninstall the route from kernel. */
620 void rib_uninstall_kernel(struct route_node
*rn
, struct route_entry
*re
)
622 struct nexthop
*nexthop
;
623 rib_table_info_t
*info
= srcdest_rnode_table_info(rn
);
624 struct zebra_vrf
*zvrf
= vrf_info_lookup(re
->vrf_id
);
626 if (info
->safi
!= SAFI_UNICAST
) {
627 UNSET_FLAG(re
->status
, ROUTE_ENTRY_INSTALLED
);
628 for (ALL_NEXTHOPS_PTR(re
->ng
, nexthop
))
629 UNSET_FLAG(nexthop
->flags
, NEXTHOP_FLAG_FIB
);
634 * Make sure we update the FPM any time we send new information to
637 hook_call(rib_update
, rn
, "uninstalling from kernel");
639 switch (dplane_route_delete(rn
, re
)) {
640 case ZEBRA_DPLANE_REQUEST_QUEUED
:
642 zvrf
->removals_queued
++;
644 case ZEBRA_DPLANE_REQUEST_FAILURE
:
646 char str
[SRCDEST2STR_BUFFER
];
648 srcdest_rnode2str(rn
, str
, sizeof(str
));
649 flog_err(EC_ZEBRA_DP_INSTALL_FAIL
,
650 "%u:%s: Failed to enqueue dataplane uninstall",
654 case ZEBRA_DPLANE_REQUEST_SUCCESS
:
663 /* Uninstall the route from kernel. */
664 static void rib_uninstall(struct route_node
*rn
, struct route_entry
*re
)
666 rib_table_info_t
*info
= srcdest_rnode_table_info(rn
);
667 rib_dest_t
*dest
= rib_dest_from_rnode(rn
);
668 struct nexthop
*nexthop
;
670 if (dest
&& dest
->selected_fib
== re
) {
671 if (info
->safi
== SAFI_UNICAST
)
672 hook_call(rib_update
, rn
, "rib_uninstall");
674 /* If labeled-unicast route, uninstall transit LSP. */
675 if (zebra_rib_labeled_unicast(re
))
676 zebra_mpls_lsp_uninstall(info
->zvrf
, rn
, re
);
678 rib_uninstall_kernel(rn
, re
);
680 dest
->selected_fib
= NULL
;
682 /* Free FIB nexthop group, if present */
683 if (re
->fib_ng
.nexthop
) {
684 nexthops_free(re
->fib_ng
.nexthop
);
685 re
->fib_ng
.nexthop
= NULL
;
688 for (ALL_NEXTHOPS_PTR(re
->ng
, nexthop
))
689 UNSET_FLAG(nexthop
->flags
, NEXTHOP_FLAG_FIB
);
692 if (CHECK_FLAG(re
->flags
, ZEBRA_FLAG_SELECTED
)) {
693 const struct prefix
*p
, *src_p
;
695 srcdest_rnode_prefixes(rn
, &p
, &src_p
);
697 redistribute_delete(p
, src_p
, re
, NULL
);
698 UNSET_FLAG(re
->flags
, ZEBRA_FLAG_SELECTED
);
703 * rib_can_delete_dest
705 * Returns true if the given dest can be deleted from the table.
707 static int rib_can_delete_dest(rib_dest_t
*dest
)
709 if (re_list_first(&dest
->routes
)) {
714 * Unresolved rnh's are stored on the default route's list
716 * dest->rnode can also be the source prefix node in an
717 * ipv6 sourcedest table. Fortunately the prefix of a
718 * source prefix node can never be the default prefix.
720 if (is_default_prefix(&dest
->rnode
->p
))
724 * Don't delete the dest if we have to update the FPM about this
727 if (CHECK_FLAG(dest
->flags
, RIB_DEST_UPDATE_FPM
)
728 || CHECK_FLAG(dest
->flags
, RIB_DEST_SENT_TO_FPM
))
734 void zebra_rib_evaluate_rn_nexthops(struct route_node
*rn
, uint32_t seq
)
736 rib_dest_t
*dest
= rib_dest_from_rnode(rn
);
740 * We are storing the rnh's associated withb
741 * the tracked nexthop as a list of the rn's.
742 * Unresolved rnh's are placed at the top
743 * of the tree list.( 0.0.0.0/0 for v4 and 0::0/0 for v6 )
744 * As such for each rn we need to walk up the tree
745 * and see if any rnh's need to see if they
746 * would match a more specific route
749 if (IS_ZEBRA_DEBUG_NHT_DETAILED
) {
750 char buf
[PREFIX_STRLEN
];
752 zlog_debug("%s: %s Being examined for Nexthop Tracking Count: %zd",
754 srcdest_rnode2str(rn
, buf
, sizeof(buf
)),
755 dest
? rnh_list_count(&dest
->nht
) : 0);
760 dest
= rib_dest_from_rnode(rn
);
764 * If we have any rnh's stored in the nht list
765 * then we know that this route node was used for
766 * nht resolution and as such we need to call the
767 * nexthop tracking evaluation code
769 frr_each_safe(rnh_list
, &dest
->nht
, rnh
) {
770 struct zebra_vrf
*zvrf
=
771 zebra_vrf_lookup_by_id(rnh
->vrf_id
);
772 struct prefix
*p
= &rnh
->node
->p
;
774 if (IS_ZEBRA_DEBUG_NHT_DETAILED
) {
775 char buf1
[PREFIX_STRLEN
];
776 char buf2
[PREFIX_STRLEN
];
778 zlog_debug("%u:%s has Nexthop(%s) Type: %s depending on it, evaluating %u:%u",
780 srcdest_rnode2str(rn
, buf1
,
782 prefix2str(p
, buf2
, sizeof(buf2
)),
783 rnh_type2str(rnh
->type
),
788 * If we have evaluated this node on this pass
789 * already, due to following the tree up
790 * then we know that we can move onto the next
793 * Additionally we call zebra_evaluate_rnh
794 * when we gc the dest. In this case we know
795 * that there must be no other re's where
796 * we were originally as such we know that
797 * that sequence number is ok to respect.
799 if (rnh
->seqno
== seq
) {
800 if (IS_ZEBRA_DEBUG_NHT_DETAILED
)
802 "\tNode processed and moved already");
807 zebra_evaluate_rnh(zvrf
, family2afi(p
->family
), 0,
813 dest
= rib_dest_from_rnode(rn
);
820 * Garbage collect the rib dest corresponding to the given route node
823 * Returns true if the dest was deleted, false otherwise.
825 int rib_gc_dest(struct route_node
*rn
)
829 dest
= rib_dest_from_rnode(rn
);
833 if (!rib_can_delete_dest(dest
))
836 if (IS_ZEBRA_DEBUG_RIB
) {
837 struct zebra_vrf
*zvrf
;
839 zvrf
= rib_dest_vrf(dest
);
840 rnode_debug(rn
, zvrf_id(zvrf
), "removing dest from table");
843 zebra_rib_evaluate_rn_nexthops(rn
, zebra_router_get_next_sequence());
846 rnh_list_fini(&dest
->nht
);
847 XFREE(MTYPE_RIB_DEST
, dest
);
851 * Release the one reference that we keep on the route node.
853 route_unlock_node(rn
);
857 static void rib_process_add_fib(struct zebra_vrf
*zvrf
, struct route_node
*rn
,
858 struct route_entry
*new)
860 hook_call(rib_update
, rn
, "new route selected");
862 /* Update real nexthop. This may actually determine if nexthop is active
864 if (!nexthop_group_active_nexthop_num(new->ng
)) {
865 UNSET_FLAG(new->status
, ROUTE_ENTRY_CHANGED
);
869 if (IS_ZEBRA_DEBUG_RIB
) {
870 char buf
[SRCDEST2STR_BUFFER
];
871 srcdest_rnode2str(rn
, buf
, sizeof(buf
));
872 zlog_debug("%u:%s: Adding route rn %p, re %p (%s)",
873 zvrf_id(zvrf
), buf
, rn
, new,
874 zebra_route_string(new->type
));
877 /* If labeled-unicast route, install transit LSP. */
878 if (zebra_rib_labeled_unicast(new))
879 zebra_mpls_lsp_install(zvrf
, rn
, new);
881 rib_install_kernel(rn
, new, NULL
);
883 UNSET_FLAG(new->status
, ROUTE_ENTRY_CHANGED
);
886 static void rib_process_del_fib(struct zebra_vrf
*zvrf
, struct route_node
*rn
,
887 struct route_entry
*old
)
889 hook_call(rib_update
, rn
, "removing existing route");
891 /* Uninstall from kernel. */
892 if (IS_ZEBRA_DEBUG_RIB
) {
893 char buf
[SRCDEST2STR_BUFFER
];
894 srcdest_rnode2str(rn
, buf
, sizeof(buf
));
895 zlog_debug("%u:%s: Deleting route rn %p, re %p (%s)",
896 zvrf_id(zvrf
), buf
, rn
, old
,
897 zebra_route_string(old
->type
));
900 /* If labeled-unicast route, uninstall transit LSP. */
901 if (zebra_rib_labeled_unicast(old
))
902 zebra_mpls_lsp_uninstall(zvrf
, rn
, old
);
904 rib_uninstall_kernel(rn
, old
);
906 /* Update nexthop for route, reset changed flag. */
907 /* Note: this code also handles the Linux case when an interface goes
908 * down, causing the kernel to delete routes without sending DELROUTE
911 if (RIB_KERNEL_ROUTE(old
))
912 SET_FLAG(old
->status
, ROUTE_ENTRY_REMOVED
);
914 UNSET_FLAG(old
->status
, ROUTE_ENTRY_CHANGED
);
917 static void rib_process_update_fib(struct zebra_vrf
*zvrf
,
918 struct route_node
*rn
,
919 struct route_entry
*old
,
920 struct route_entry
*new)
925 * We have to install or update if a new route has been selected or
926 * something has changed.
928 if (new != old
|| CHECK_FLAG(new->status
, ROUTE_ENTRY_CHANGED
)) {
929 hook_call(rib_update
, rn
, "updating existing route");
931 /* Update the nexthop; we could determine here that nexthop is
933 if (nexthop_group_active_nexthop_num(new->ng
))
936 /* If nexthop is active, install the selected route, if
938 * the install succeeds, cleanup flags for prior route, if
943 if (IS_ZEBRA_DEBUG_RIB
) {
944 char buf
[SRCDEST2STR_BUFFER
];
945 srcdest_rnode2str(rn
, buf
, sizeof(buf
));
948 "%u:%s: Updating route rn %p, re %p (%s) old %p (%s)",
949 zvrf_id(zvrf
), buf
, rn
, new,
950 zebra_route_string(new->type
),
952 zebra_route_string(old
->type
));
955 "%u:%s: Updating route rn %p, re %p (%s)",
956 zvrf_id(zvrf
), buf
, rn
, new,
957 zebra_route_string(new->type
));
960 /* If labeled-unicast route, uninstall transit LSP. */
961 if (zebra_rib_labeled_unicast(old
))
962 zebra_mpls_lsp_uninstall(zvrf
, rn
, old
);
965 * Non-system route should be installed.
966 * If labeled-unicast route, install transit
969 if (zebra_rib_labeled_unicast(new))
970 zebra_mpls_lsp_install(zvrf
, rn
, new);
972 rib_install_kernel(rn
, new, old
);
976 * If nexthop for selected route is not active or install
978 * may need to uninstall and delete for redistribution.
981 if (IS_ZEBRA_DEBUG_RIB
) {
982 char buf
[SRCDEST2STR_BUFFER
];
983 srcdest_rnode2str(rn
, buf
, sizeof(buf
));
986 "%u:%s: Deleting route rn %p, re %p (%s) old %p (%s) - nexthop inactive",
987 zvrf_id(zvrf
), buf
, rn
, new,
988 zebra_route_string(new->type
),
990 zebra_route_string(old
->type
));
993 "%u:%s: Deleting route rn %p, re %p (%s) - nexthop inactive",
994 zvrf_id(zvrf
), buf
, rn
, new,
995 zebra_route_string(new->type
));
998 /* If labeled-unicast route, uninstall transit LSP. */
999 if (zebra_rib_labeled_unicast(old
))
1000 zebra_mpls_lsp_uninstall(zvrf
, rn
, old
);
1002 rib_uninstall_kernel(rn
, old
);
1006 * Same route selected; check if in the FIB and if not,
1007 * re-install. This is housekeeping code to deal with
1008 * race conditions in kernel with linux netlink reporting
1009 * interface up before IPv4 or IPv6 protocol is ready
1012 if (!CHECK_FLAG(new->status
, ROUTE_ENTRY_INSTALLED
) ||
1013 RIB_SYSTEM_ROUTE(new))
1014 rib_install_kernel(rn
, new, NULL
);
1017 /* Update prior route. */
1019 UNSET_FLAG(old
->status
, ROUTE_ENTRY_CHANGED
);
1021 /* Clear changed flag. */
1022 UNSET_FLAG(new->status
, ROUTE_ENTRY_CHANGED
);
1025 /* Check if 'alternate' RIB entry is better than 'current'. */
1026 static struct route_entry
*rib_choose_best(struct route_entry
*current
,
1027 struct route_entry
*alternate
)
1029 if (current
== NULL
)
1032 /* filter route selection in following order:
1033 * - connected beats other types
1034 * - if both connected, loopback or vrf wins
1035 * - lower distance beats higher
1036 * - lower metric beats higher for equal distance
1037 * - last, hence oldest, route wins tie break.
1040 /* Connected routes. Check to see if either are a vrf
1041 * or loopback interface. If not, pick the last connected
1042 * route of the set of lowest metric connected routes.
1044 if (alternate
->type
== ZEBRA_ROUTE_CONNECT
) {
1045 if (current
->type
!= ZEBRA_ROUTE_CONNECT
)
1048 /* both are connected. are either loop or vrf? */
1049 struct nexthop
*nexthop
= NULL
;
1051 for (ALL_NEXTHOPS_PTR(alternate
->ng
, nexthop
)) {
1052 struct interface
*ifp
= if_lookup_by_index(
1053 nexthop
->ifindex
, alternate
->vrf_id
);
1055 if (ifp
&& if_is_loopback_or_vrf(ifp
))
1059 for (ALL_NEXTHOPS_PTR(current
->ng
, nexthop
)) {
1060 struct interface
*ifp
= if_lookup_by_index(
1061 nexthop
->ifindex
, current
->vrf_id
);
1063 if (ifp
&& if_is_loopback_or_vrf(ifp
))
1067 /* Neither are loop or vrf so pick best metric */
1068 if (alternate
->metric
<= current
->metric
)
1074 if (current
->type
== ZEBRA_ROUTE_CONNECT
)
1077 /* higher distance loses */
1078 if (alternate
->distance
< current
->distance
)
1080 if (current
->distance
< alternate
->distance
)
1083 /* metric tie-breaks equal distance */
1084 if (alternate
->metric
<= current
->metric
)
1090 /* Core function for processing nexthop group contexts's off metaq */
1091 static void rib_nhg_process(struct nhg_ctx
*ctx
)
1093 nhg_ctx_process(ctx
);
1096 /* Core function for processing routing information base. */
1097 static void rib_process(struct route_node
*rn
)
1099 struct route_entry
*re
;
1100 struct route_entry
*next
;
1101 struct route_entry
*old_selected
= NULL
;
1102 struct route_entry
*new_selected
= NULL
;
1103 struct route_entry
*old_fib
= NULL
;
1104 struct route_entry
*new_fib
= NULL
;
1105 struct route_entry
*best
= NULL
;
1106 char buf
[SRCDEST2STR_BUFFER
];
1108 struct zebra_vrf
*zvrf
= NULL
;
1109 const struct prefix
*p
, *src_p
;
1111 srcdest_rnode_prefixes(rn
, &p
, &src_p
);
1112 vrf_id_t vrf_id
= VRF_UNKNOWN
;
1116 dest
= rib_dest_from_rnode(rn
);
1118 zvrf
= rib_dest_vrf(dest
);
1119 vrf_id
= zvrf_id(zvrf
);
1122 if (IS_ZEBRA_DEBUG_RIB
)
1123 srcdest_rnode2str(rn
, buf
, sizeof(buf
));
1125 if (IS_ZEBRA_DEBUG_RIB_DETAILED
)
1126 zlog_debug("%u:%s: Processing rn %p", vrf_id
, buf
, rn
);
1129 * we can have rn's that have a NULL info pointer
1130 * (dest). As such let's not let the deref happen
1131 * additionally we know RNODE_FOREACH_RE_SAFE
1132 * will not iterate so we are ok.
1135 old_fib
= dest
->selected_fib
;
1137 RNODE_FOREACH_RE_SAFE (rn
, re
, next
) {
1138 if (IS_ZEBRA_DEBUG_RIB_DETAILED
)
1140 "%u:%s: Examine re %p (%s) status %x flags %x dist %d metric %d",
1141 vrf_id
, buf
, re
, zebra_route_string(re
->type
),
1142 re
->status
, re
->flags
, re
->distance
,
1145 /* Currently selected re. */
1146 if (CHECK_FLAG(re
->flags
, ZEBRA_FLAG_SELECTED
)) {
1147 assert(old_selected
== NULL
);
1151 /* Skip deleted entries from selection */
1152 if (CHECK_FLAG(re
->status
, ROUTE_ENTRY_REMOVED
))
1155 /* Skip unreachable nexthop. */
1156 /* This first call to nexthop_active_update is merely to
1157 * determine if there's any change to nexthops associated
1158 * with this RIB entry. Now, rib_process() can be invoked due
1159 * to an external event such as link down or due to
1160 * next-hop-tracking evaluation. In the latter case,
1161 * a decision has already been made that the NHs have changed.
1162 * So, no need to invoke a potentially expensive call again.
1163 * Further, since the change might be in a recursive NH which
1164 * is not caught in the nexthop_active_update() code. Thus, we
1165 * might miss changes to recursive NHs.
1167 if (CHECK_FLAG(re
->status
, ROUTE_ENTRY_CHANGED
)
1168 && !nexthop_active_update(rn
, re
)) {
1169 if (re
->type
== ZEBRA_ROUTE_TABLE
) {
1170 /* XXX: HERE BE DRAGONS!!!!!
1171 * In all honesty, I have not yet figured out
1172 * what this part does or why the
1173 * ROUTE_ENTRY_CHANGED test above is correct
1174 * or why we need to delete a route here, and
1175 * also not whether this concerns both selected
1176 * and fib route, or only selected
1179 * This entry was denied by the 'ip protocol
1180 * table' route-map, we need to delete it */
1181 if (re
!= old_selected
) {
1182 if (IS_ZEBRA_DEBUG_RIB
)
1184 "%s: %u:%s: imported via import-table but denied "
1185 "by the ip protocol table route-map",
1186 __func__
, vrf_id
, buf
);
1189 SET_FLAG(re
->status
,
1190 ROUTE_ENTRY_REMOVED
);
1196 /* Infinite distance. */
1197 if (re
->distance
== DISTANCE_INFINITY
) {
1198 UNSET_FLAG(re
->status
, ROUTE_ENTRY_CHANGED
);
1202 if (CHECK_FLAG(re
->flags
, ZEBRA_FLAG_FIB_OVERRIDE
)) {
1203 best
= rib_choose_best(new_fib
, re
);
1204 if (new_fib
&& best
!= new_fib
)
1205 UNSET_FLAG(new_fib
->status
,
1206 ROUTE_ENTRY_CHANGED
);
1209 best
= rib_choose_best(new_selected
, re
);
1210 if (new_selected
&& best
!= new_selected
)
1211 UNSET_FLAG(new_selected
->status
,
1212 ROUTE_ENTRY_CHANGED
);
1213 new_selected
= best
;
1216 UNSET_FLAG(re
->status
, ROUTE_ENTRY_CHANGED
);
1217 } /* RNODE_FOREACH_RE */
1219 /* If no FIB override route, use the selected route also for FIB */
1220 if (new_fib
== NULL
)
1221 new_fib
= new_selected
;
1223 /* After the cycle is finished, the following pointers will be set:
1224 * old_selected --- RE entry currently having SELECTED
1225 * new_selected --- RE entry that is newly SELECTED
1226 * old_fib --- RE entry currently in kernel FIB
1227 * new_fib --- RE entry that is newly to be in kernel FIB
1229 * new_selected will get SELECTED flag, and is going to be redistributed
1230 * the zclients. new_fib (which can be new_selected) will be installed
1234 if (IS_ZEBRA_DEBUG_RIB_DETAILED
) {
1236 "%u:%s: After processing: old_selected %p new_selected %p old_fib %p new_fib %p",
1237 vrf_id
, buf
, (void *)old_selected
, (void *)new_selected
,
1238 (void *)old_fib
, (void *)new_fib
);
1241 /* Buffer ROUTE_ENTRY_CHANGED here, because it will get cleared if
1242 * fib == selected */
1243 bool selected_changed
= new_selected
&& CHECK_FLAG(new_selected
->status
,
1244 ROUTE_ENTRY_CHANGED
);
1246 /* Update fib according to selection results */
1247 if (new_fib
&& old_fib
)
1248 rib_process_update_fib(zvrf
, rn
, old_fib
, new_fib
);
1250 rib_process_add_fib(zvrf
, rn
, new_fib
);
1252 rib_process_del_fib(zvrf
, rn
, old_fib
);
1254 /* Update SELECTED entry */
1255 if (old_selected
!= new_selected
|| selected_changed
) {
1257 if (new_selected
&& new_selected
!= new_fib
)
1258 UNSET_FLAG(new_selected
->status
, ROUTE_ENTRY_CHANGED
);
1261 SET_FLAG(new_selected
->flags
, ZEBRA_FLAG_SELECTED
);
1265 * If we're removing the old entry, we should tell
1266 * redist subscribers about that *if* they aren't
1267 * going to see a redist for the new entry.
1269 if (!new_selected
|| CHECK_FLAG(old_selected
->status
,
1270 ROUTE_ENTRY_REMOVED
))
1271 redistribute_delete(p
, src_p
,
1275 if (old_selected
!= new_selected
)
1276 UNSET_FLAG(old_selected
->flags
,
1277 ZEBRA_FLAG_SELECTED
);
1281 /* Remove all RE entries queued for removal */
1282 RNODE_FOREACH_RE_SAFE (rn
, re
, next
) {
1283 if (CHECK_FLAG(re
->status
, ROUTE_ENTRY_REMOVED
)) {
1284 if (IS_ZEBRA_DEBUG_RIB
) {
1285 rnode_debug(rn
, vrf_id
, "rn %p, removing re %p",
1286 (void *)rn
, (void *)re
);
1293 * Check if the dest can be deleted now.
1298 static void zebra_rib_evaluate_mpls(struct route_node
*rn
)
1300 rib_dest_t
*dest
= rib_dest_from_rnode(rn
);
1301 struct zebra_vrf
*zvrf
= vrf_info_lookup(VRF_DEFAULT
);
1306 if (CHECK_FLAG(dest
->flags
, RIB_DEST_UPDATE_LSPS
)) {
1307 if (IS_ZEBRA_DEBUG_MPLS
)
1309 "%u: Scheduling all LSPs upon RIB completion",
1311 zebra_mpls_lsp_schedule(zvrf
);
1312 mpls_unmark_lsps_for_processing(rn
);
1317 * Utility to match route with dplane context data
1319 static bool rib_route_match_ctx(const struct route_entry
*re
,
1320 const struct zebra_dplane_ctx
*ctx
,
1323 bool result
= false;
1327 * In 'update' case, we test info about the 'previous' or
1330 if ((re
->type
== dplane_ctx_get_old_type(ctx
)) &&
1331 (re
->instance
== dplane_ctx_get_old_instance(ctx
))) {
1334 /* TODO -- we're using this extra test, but it's not
1335 * exactly clear why.
1337 if (re
->type
== ZEBRA_ROUTE_STATIC
&&
1338 (re
->distance
!= dplane_ctx_get_old_distance(ctx
) ||
1339 re
->tag
!= dplane_ctx_get_old_tag(ctx
))) {
1346 * Ordinary, single-route case using primary context info
1348 if ((dplane_ctx_get_op(ctx
) != DPLANE_OP_ROUTE_DELETE
) &&
1349 CHECK_FLAG(re
->status
, ROUTE_ENTRY_REMOVED
)) {
1350 /* Skip route that's been deleted */
1354 if ((re
->type
== dplane_ctx_get_type(ctx
)) &&
1355 (re
->instance
== dplane_ctx_get_instance(ctx
))) {
1358 /* TODO -- we're using this extra test, but it's not
1359 * exactly clear why.
1361 if (re
->type
== ZEBRA_ROUTE_STATIC
&&
1362 (re
->distance
!= dplane_ctx_get_distance(ctx
) ||
1363 re
->tag
!= dplane_ctx_get_tag(ctx
))) {
1374 static void zebra_rib_fixup_system(struct route_node
*rn
)
1376 struct route_entry
*re
;
1378 RNODE_FOREACH_RE(rn
, re
) {
1379 struct nexthop
*nhop
;
1381 if (!RIB_SYSTEM_ROUTE(re
))
1384 if (CHECK_FLAG(re
->status
, ROUTE_ENTRY_REMOVED
))
1387 SET_FLAG(re
->status
, ROUTE_ENTRY_INSTALLED
);
1388 UNSET_FLAG(re
->status
, ROUTE_ENTRY_QUEUED
);
1390 for (ALL_NEXTHOPS_PTR(re
->ng
, nhop
)) {
1391 if (CHECK_FLAG(nhop
->flags
, NEXTHOP_FLAG_RECURSIVE
))
1394 SET_FLAG(nhop
->flags
, NEXTHOP_FLAG_FIB
);
1400 * Update a route from a dplane context. This consolidates common code
1401 * that can be used in processing of results from FIB updates, and in
1402 * async notification processing.
1403 * The return is 'true' if the installed nexthops changed; 'false' otherwise.
1405 static bool rib_update_re_from_ctx(struct route_entry
*re
,
1406 struct route_node
*rn
,
1407 struct zebra_dplane_ctx
*ctx
)
1409 char dest_str
[PREFIX_STRLEN
] = "";
1410 char nh_str
[NEXTHOP_STRLEN
];
1411 struct nexthop
*nexthop
, *ctx_nexthop
;
1413 const struct nexthop_group
*ctxnhg
;
1414 bool is_selected
= false; /* Is 're' currently the selected re? */
1415 bool changed_p
= false; /* Change to nexthops? */
1418 /* Note well: only capturing the prefix string if debug is enabled here;
1419 * unconditional log messages will have to generate the string.
1421 if (IS_ZEBRA_DEBUG_RIB
)
1422 prefix2str(&(rn
->p
), dest_str
, sizeof(dest_str
));
1424 dest
= rib_dest_from_rnode(rn
);
1426 is_selected
= (re
== dest
->selected_fib
);
1428 if (IS_ZEBRA_DEBUG_RIB_DETAILED
)
1429 zlog_debug("update_from_ctx: %u:%s: %sSELECTED",
1430 re
->vrf_id
, dest_str
, (is_selected
? "" : "NOT "));
1432 /* Update zebra's nexthop FIB flag for each nexthop that was installed.
1433 * If the installed set differs from the set requested by the rib/owner,
1434 * we use the fib-specific nexthop-group to record the actual FIB
1439 * First check the fib nexthop-group, if it's present. The comparison
1440 * here is quite strict: we require that the fib sets match exactly.
1444 if (re
->fib_ng
.nexthop
== NULL
)
1449 /* First check the route's fib nexthops */
1450 for (ALL_NEXTHOPS(re
->fib_ng
, nexthop
)) {
1452 if (CHECK_FLAG(nexthop
->flags
, NEXTHOP_FLAG_RECURSIVE
))
1456 for (ALL_NEXTHOPS_PTR(dplane_ctx_get_ng(ctx
),
1458 if (nexthop_same(ctx_nexthop
, nexthop
))
1462 if (ctx_nexthop
== NULL
) {
1463 /* Nexthop not in the new installed set */
1464 if (IS_ZEBRA_DEBUG_RIB_DETAILED
) {
1465 nexthop2str(nexthop
, nh_str
,
1467 zlog_debug("update_from_ctx: no match for fib nh %s",
1479 /* Check the new installed set */
1481 for (ALL_NEXTHOPS_PTR(dplane_ctx_get_ng(ctx
), ctx_nexthop
)) {
1483 if (CHECK_FLAG(ctx_nexthop
->flags
,
1484 NEXTHOP_FLAG_RECURSIVE
))
1487 /* Compare with the current group's nexthops */
1489 for (ALL_NEXTHOPS(re
->fib_ng
, nexthop
)) {
1490 if (nexthop_same(nexthop
, ctx_nexthop
))
1494 if (nexthop
== NULL
) {
1495 /* Nexthop not in the old installed set */
1496 if (IS_ZEBRA_DEBUG_RIB_DETAILED
) {
1497 nexthop2str(ctx_nexthop
, nh_str
,
1499 zlog_debug("update_from_ctx: no fib match for notif nh %s",
1509 /* If the new FIB set matches the existing FIB set, we're done. */
1511 if (IS_ZEBRA_DEBUG_RIB
)
1512 zlog_debug("%u:%s update_from_ctx(): existing fib nhg, no change",
1513 re
->vrf_id
, dest_str
);
1516 } else if (re
->fib_ng
.nexthop
) {
1518 * Free stale fib list and move on to check the rib nhg.
1520 if (IS_ZEBRA_DEBUG_RIB
)
1521 zlog_debug("%u:%s update_from_ctx(): replacing fib nhg",
1522 re
->vrf_id
, dest_str
);
1523 nexthops_free(re
->fib_ng
.nexthop
);
1524 re
->fib_ng
.nexthop
= NULL
;
1526 /* Note that the installed nexthops have changed */
1529 if (IS_ZEBRA_DEBUG_RIB
)
1530 zlog_debug("%u:%s update_from_ctx(): no fib nhg",
1531 re
->vrf_id
, dest_str
);
1535 * Compare with the rib nexthop group. The comparison here is different:
1536 * the RIB group may be a superset of the list installed in the FIB. We
1537 * walk the RIB group, looking for the 'installable' candidate
1538 * nexthops, and then check those against the set
1539 * that is actually installed.
1542 for (ALL_NEXTHOPS_PTR(re
->ng
, nexthop
)) {
1544 if (CHECK_FLAG(nexthop
->flags
, NEXTHOP_FLAG_RECURSIVE
))
1547 if (!CHECK_FLAG(nexthop
->flags
, NEXTHOP_FLAG_ACTIVE
))
1550 /* Check for a FIB nexthop corresponding to the RIB nexthop */
1552 for (ALL_NEXTHOPS_PTR(dplane_ctx_get_ng(ctx
), ctx_nexthop
)) {
1553 if (nexthop_same(ctx_nexthop
, nexthop
))
1557 /* If the FIB doesn't know about the nexthop,
1558 * it's not installed
1560 if (ctx_nexthop
== NULL
) {
1561 if (IS_ZEBRA_DEBUG_RIB_DETAILED
) {
1562 nexthop2str(nexthop
, nh_str
, sizeof(nh_str
));
1563 zlog_debug("update_from_ctx: no notif match for rib nh %s",
1568 if (CHECK_FLAG(nexthop
->flags
, NEXTHOP_FLAG_FIB
))
1571 UNSET_FLAG(nexthop
->flags
, NEXTHOP_FLAG_FIB
);
1573 /* Keep checking nexthops */
1577 if (CHECK_FLAG(ctx_nexthop
->flags
, NEXTHOP_FLAG_FIB
)) {
1578 if (!CHECK_FLAG(nexthop
->flags
, NEXTHOP_FLAG_FIB
))
1581 SET_FLAG(nexthop
->flags
, NEXTHOP_FLAG_FIB
);
1583 if (CHECK_FLAG(nexthop
->flags
, NEXTHOP_FLAG_FIB
))
1586 UNSET_FLAG(nexthop
->flags
, NEXTHOP_FLAG_FIB
);
1590 /* If all nexthops were processed, we're done */
1592 if (IS_ZEBRA_DEBUG_RIB
)
1593 zlog_debug("%u:%s update_from_ctx(): rib nhg matched, changed '%s'",
1594 re
->vrf_id
, dest_str
,
1595 (changed_p
? "true" : "false"));
1599 /* FIB nexthop set differs from the RIB set:
1600 * create a fib-specific nexthop-group
1602 if (IS_ZEBRA_DEBUG_RIB
)
1603 zlog_debug("%u:%s update_from_ctx(): changed %s, adding new fib nhg",
1604 re
->vrf_id
, dest_str
,
1605 (changed_p
? "true" : "false"));
1607 ctxnhg
= dplane_ctx_get_ng(ctx
);
1609 if (ctxnhg
->nexthop
)
1610 copy_nexthops(&(re
->fib_ng
.nexthop
), ctxnhg
->nexthop
, NULL
);
1612 /* Bit of a special case when the fib has _no_ installed
1615 nexthop
= nexthop_new();
1616 nexthop
->type
= NEXTHOP_TYPE_IPV4
;
1617 _nexthop_add(&(re
->fib_ng
.nexthop
), nexthop
);
1625 * Helper to locate a zebra route-node from a dplane context. This is used
1626 * when processing dplane results, e.g. Note well: the route-node is returned
1627 * with a ref held - route_unlock_node() must be called eventually.
1629 static struct route_node
*
1630 rib_find_rn_from_ctx(const struct zebra_dplane_ctx
*ctx
)
1632 struct route_table
*table
= NULL
;
1633 struct route_node
*rn
= NULL
;
1634 const struct prefix
*dest_pfx
, *src_pfx
;
1636 /* Locate rn and re(s) from ctx */
1638 table
= zebra_vrf_table_with_table_id(dplane_ctx_get_afi(ctx
),
1639 dplane_ctx_get_safi(ctx
),
1640 dplane_ctx_get_vrf(ctx
),
1641 dplane_ctx_get_table(ctx
));
1642 if (table
== NULL
) {
1643 if (IS_ZEBRA_DEBUG_DPLANE
) {
1644 zlog_debug("Failed to find route for ctx: no table for afi %d, safi %d, vrf %u",
1645 dplane_ctx_get_afi(ctx
),
1646 dplane_ctx_get_safi(ctx
),
1647 dplane_ctx_get_vrf(ctx
));
1652 dest_pfx
= dplane_ctx_get_dest(ctx
);
1653 src_pfx
= dplane_ctx_get_src(ctx
);
1655 rn
= srcdest_rnode_get(table
, dest_pfx
,
1656 src_pfx
? (struct prefix_ipv6
*)src_pfx
: NULL
);
1665 * Route-update results processing after async dataplane update.
1667 static void rib_process_result(struct zebra_dplane_ctx
*ctx
)
1669 struct zebra_vrf
*zvrf
= NULL
;
1670 struct route_node
*rn
= NULL
;
1671 struct route_entry
*re
= NULL
, *old_re
= NULL
, *rib
;
1672 bool is_update
= false;
1673 char dest_str
[PREFIX_STRLEN
] = "";
1674 enum dplane_op_e op
;
1675 enum zebra_dplane_result status
;
1676 const struct prefix
*dest_pfx
, *src_pfx
;
1678 bool fib_changed
= false;
1680 zvrf
= vrf_info_lookup(dplane_ctx_get_vrf(ctx
));
1681 dest_pfx
= dplane_ctx_get_dest(ctx
);
1683 /* Note well: only capturing the prefix string if debug is enabled here;
1684 * unconditional log messages will have to generate the string.
1686 if (IS_ZEBRA_DEBUG_DPLANE
)
1687 prefix2str(dest_pfx
, dest_str
, sizeof(dest_str
));
1689 /* Locate rn and re(s) from ctx */
1690 rn
= rib_find_rn_from_ctx(ctx
);
1692 if (IS_ZEBRA_DEBUG_DPLANE
) {
1693 zlog_debug("Failed to process dplane results: no route for %u:%s",
1694 dplane_ctx_get_vrf(ctx
), dest_str
);
1699 srcdest_rnode_prefixes(rn
, &dest_pfx
, &src_pfx
);
1701 op
= dplane_ctx_get_op(ctx
);
1702 status
= dplane_ctx_get_status(ctx
);
1704 if (IS_ZEBRA_DEBUG_DPLANE_DETAIL
)
1705 zlog_debug("%u:%s Processing dplane ctx %p, op %s result %s",
1706 dplane_ctx_get_vrf(ctx
), dest_str
, ctx
,
1707 dplane_op2str(op
), dplane_res2str(status
));
1710 * Update is a bit of a special case, where we may have both old and new
1711 * routes to post-process.
1713 is_update
= dplane_ctx_is_update(ctx
);
1716 * Take a pass through the routes, look for matches with the context
1719 RNODE_FOREACH_RE(rn
, rib
) {
1722 if (rib_route_match_ctx(rib
, ctx
, false))
1726 /* Check for old route match */
1727 if (is_update
&& (old_re
== NULL
)) {
1728 if (rib_route_match_ctx(rib
, ctx
, true /*is_update*/))
1732 /* Have we found the routes we need to work on? */
1733 if (re
&& ((!is_update
|| old_re
)))
1737 seq
= dplane_ctx_get_seq(ctx
);
1740 * Check sequence number(s) to detect stale results before continuing
1743 if (re
->dplane_sequence
!= seq
) {
1744 if (IS_ZEBRA_DEBUG_DPLANE_DETAIL
)
1745 zlog_debug("%u:%s Stale dplane result for re %p",
1746 dplane_ctx_get_vrf(ctx
),
1749 UNSET_FLAG(re
->status
, ROUTE_ENTRY_QUEUED
);
1753 if (old_re
->dplane_sequence
!= dplane_ctx_get_old_seq(ctx
)) {
1754 if (IS_ZEBRA_DEBUG_DPLANE_DETAIL
)
1755 zlog_debug("%u:%s Stale dplane result for old_re %p",
1756 dplane_ctx_get_vrf(ctx
),
1759 UNSET_FLAG(old_re
->status
, ROUTE_ENTRY_QUEUED
);
1763 case DPLANE_OP_ROUTE_INSTALL
:
1764 case DPLANE_OP_ROUTE_UPDATE
:
1765 if (status
== ZEBRA_DPLANE_REQUEST_SUCCESS
) {
1767 UNSET_FLAG(re
->status
, ROUTE_ENTRY_FAILED
);
1768 SET_FLAG(re
->status
, ROUTE_ENTRY_INSTALLED
);
1771 * On an update operation from the same route type
1772 * context retrieval currently has no way to know
1773 * which was the old and which was the new.
1774 * So don't unset our flags that we just set.
1775 * We know redistribution is ok because the
1776 * old_re in this case is used for nothing
1777 * more than knowing whom to contact if necessary.
1779 if (old_re
&& old_re
!= re
) {
1780 UNSET_FLAG(old_re
->status
, ROUTE_ENTRY_FAILED
);
1781 UNSET_FLAG(old_re
->status
,
1782 ROUTE_ENTRY_INSTALLED
);
1785 /* Update zebra route based on the results in
1786 * the context struct.
1790 rib_update_re_from_ctx(re
, rn
, ctx
);
1793 if (IS_ZEBRA_DEBUG_DPLANE_DETAIL
)
1794 zlog_debug("%u:%s no fib change for re",
1801 redistribute_update(dest_pfx
, src_pfx
,
1806 * System routes are weird in that they
1807 * allow multiple to be installed that match
1808 * to the same prefix, so after we get the
1809 * result we need to clean them up so that
1810 * we can actually use them.
1812 if ((re
&& RIB_SYSTEM_ROUTE(re
)) ||
1813 (old_re
&& RIB_SYSTEM_ROUTE(old_re
)))
1814 zebra_rib_fixup_system(rn
);
1819 /* Notify route owner */
1820 zsend_route_notify_owner_ctx(ctx
, ZAPI_ROUTE_INSTALLED
);
1824 SET_FLAG(re
->status
, ROUTE_ENTRY_FAILED
);
1825 UNSET_FLAG(re
->status
, ROUTE_ENTRY_INSTALLED
);
1827 SET_FLAG(old_re
->status
, ROUTE_ENTRY_FAILED
);
1829 zsend_route_notify_owner(re
, dest_pfx
,
1830 ZAPI_ROUTE_FAIL_INSTALL
);
1832 zlog_warn("%u:%s: Route install failed",
1833 dplane_ctx_get_vrf(ctx
),
1834 prefix2str(dest_pfx
,
1835 dest_str
, sizeof(dest_str
)));
1838 case DPLANE_OP_ROUTE_DELETE
:
1840 SET_FLAG(re
->status
, ROUTE_ENTRY_FAILED
);
1842 * In the delete case, the zebra core datastructs were
1843 * updated (or removed) at the time the delete was issued,
1844 * so we're just notifying the route owner.
1846 if (status
== ZEBRA_DPLANE_REQUEST_SUCCESS
) {
1848 UNSET_FLAG(re
->status
, ROUTE_ENTRY_INSTALLED
);
1849 UNSET_FLAG(re
->status
, ROUTE_ENTRY_FAILED
);
1851 zsend_route_notify_owner_ctx(ctx
, ZAPI_ROUTE_REMOVED
);
1857 SET_FLAG(re
->status
, ROUTE_ENTRY_FAILED
);
1858 zsend_route_notify_owner_ctx(ctx
,
1859 ZAPI_ROUTE_REMOVE_FAIL
);
1861 zlog_warn("%u:%s: Route Deletion failure",
1862 dplane_ctx_get_vrf(ctx
),
1863 prefix2str(dest_pfx
,
1864 dest_str
, sizeof(dest_str
)));
1868 * System routes are weird in that they
1869 * allow multiple to be installed that match
1870 * to the same prefix, so after we get the
1871 * result we need to clean them up so that
1872 * we can actually use them.
1874 if ((re
&& RIB_SYSTEM_ROUTE(re
)) ||
1875 (old_re
&& RIB_SYSTEM_ROUTE(old_re
)))
1876 zebra_rib_fixup_system(rn
);
1882 zebra_rib_evaluate_rn_nexthops(rn
, seq
);
1883 zebra_rib_evaluate_mpls(rn
);
1887 route_unlock_node(rn
);
1889 /* Return context to dataplane module */
1890 dplane_ctx_fini(&ctx
);
1894 * Handle notification from async dataplane: the dataplane has detected
1895 * some change to a route, and notifies zebra so that the control plane
1896 * can reflect that change.
1898 static void rib_process_dplane_notify(struct zebra_dplane_ctx
*ctx
)
1900 struct route_node
*rn
= NULL
;
1901 struct route_entry
*re
= NULL
;
1902 struct nexthop
*nexthop
;
1903 char dest_str
[PREFIX_STRLEN
] = "";
1904 const struct prefix
*dest_pfx
, *src_pfx
;
1906 bool fib_changed
= false;
1907 bool debug_p
= IS_ZEBRA_DEBUG_DPLANE
| IS_ZEBRA_DEBUG_RIB
;
1908 int start_count
, end_count
;
1909 dest_pfx
= dplane_ctx_get_dest(ctx
);
1911 /* Note well: only capturing the prefix string if debug is enabled here;
1912 * unconditional log messages will have to generate the string.
1915 prefix2str(dest_pfx
, dest_str
, sizeof(dest_str
));
1917 /* Locate rn and re(s) from ctx */
1918 rn
= rib_find_rn_from_ctx(ctx
);
1921 zlog_debug("Failed to process dplane notification: no routes for %u:%s",
1922 dplane_ctx_get_vrf(ctx
), dest_str
);
1927 dest
= rib_dest_from_rnode(rn
);
1928 srcdest_rnode_prefixes(rn
, &dest_pfx
, &src_pfx
);
1931 zlog_debug("%u:%s Processing dplane notif ctx %p",
1932 dplane_ctx_get_vrf(ctx
), dest_str
, ctx
);
1935 * Take a pass through the routes, look for matches with the context
1938 RNODE_FOREACH_RE(rn
, re
) {
1939 if (rib_route_match_ctx(re
, ctx
, false /*!update*/))
1943 /* No match? Nothing we can do */
1946 zlog_debug("%u:%s Unable to process dplane notification: no entry for type %s",
1947 dplane_ctx_get_vrf(ctx
), dest_str
,
1949 dplane_ctx_get_type(ctx
)));
1954 /* Ensure we clear the QUEUED flag */
1955 UNSET_FLAG(re
->status
, ROUTE_ENTRY_QUEUED
);
1957 /* Is this a notification that ... matters? We only really care about
1958 * the route that is currently selected for installation.
1960 if (re
!= dest
->selected_fib
) {
1961 /* TODO -- don't skip processing entirely? We might like to
1962 * at least report on the event.
1965 zlog_debug("%u:%s dplane notif, but type %s not selected_fib",
1966 dplane_ctx_get_vrf(ctx
), dest_str
,
1968 dplane_ctx_get_type(ctx
)));
1972 /* We'll want to determine whether the installation status of the
1973 * route has changed: we'll check the status before processing,
1974 * and then again if there's been a change.
1977 for (ALL_NEXTHOPS_PTR(rib_active_nhg(re
), nexthop
)) {
1978 if (CHECK_FLAG(nexthop
->flags
, NEXTHOP_FLAG_FIB
))
1982 /* Update zebra's nexthop FIB flags based on the context struct's
1985 fib_changed
= rib_update_re_from_ctx(re
, rn
, ctx
);
1989 zlog_debug("%u:%s No change from dplane notification",
1990 dplane_ctx_get_vrf(ctx
), dest_str
);
1996 * Perform follow-up work if the actual status of the prefix
2001 for (ALL_NEXTHOPS_PTR(rib_active_nhg(re
), nexthop
)) {
2002 if (CHECK_FLAG(nexthop
->flags
, NEXTHOP_FLAG_FIB
))
2006 /* Various fib transitions: changed nexthops; from installed to
2007 * not-installed; or not-installed to installed.
2009 if (start_count
> 0 && end_count
> 0) {
2011 zlog_debug("%u:%s applied nexthop changes from dplane notification",
2012 dplane_ctx_get_vrf(ctx
), dest_str
);
2014 /* Changed nexthops - update kernel/others */
2015 dplane_route_notif_update(rn
, re
,
2016 DPLANE_OP_ROUTE_UPDATE
, ctx
);
2018 } else if (start_count
== 0 && end_count
> 0) {
2020 zlog_debug("%u:%s installed transition from dplane notification",
2021 dplane_ctx_get_vrf(ctx
), dest_str
);
2023 /* We expect this to be the selected route, so we want
2024 * to tell others about this transition.
2026 SET_FLAG(re
->status
, ROUTE_ENTRY_INSTALLED
);
2028 /* Changed nexthops - update kernel/others */
2029 dplane_route_notif_update(rn
, re
, DPLANE_OP_ROUTE_INSTALL
, ctx
);
2031 /* Redistribute, lsp, and nht update */
2032 redistribute_update(dest_pfx
, src_pfx
, re
, NULL
);
2034 zebra_rib_evaluate_rn_nexthops(
2035 rn
, zebra_router_get_next_sequence());
2037 zebra_rib_evaluate_mpls(rn
);
2039 } else if (start_count
> 0 && end_count
== 0) {
2041 zlog_debug("%u:%s un-installed transition from dplane notification",
2042 dplane_ctx_get_vrf(ctx
), dest_str
);
2044 /* Transition from _something_ installed to _nothing_
2047 /* We expect this to be the selected route, so we want
2048 * to tell others about this transistion.
2050 UNSET_FLAG(re
->status
, ROUTE_ENTRY_INSTALLED
);
2052 /* Changed nexthops - update kernel/others */
2053 dplane_route_notif_update(rn
, re
, DPLANE_OP_ROUTE_DELETE
, ctx
);
2055 /* Redistribute, lsp, and nht update */
2056 redistribute_delete(dest_pfx
, src_pfx
, re
, NULL
);
2058 zebra_rib_evaluate_rn_nexthops(
2059 rn
, zebra_router_get_next_sequence());
2061 zebra_rib_evaluate_mpls(rn
);
2066 route_unlock_node(rn
);
2068 /* Return context to dataplane module */
2069 dplane_ctx_fini(&ctx
);
2072 static void process_subq_nhg(struct listnode
*lnode
)
2074 struct nhg_ctx
*ctx
= NULL
;
2075 uint8_t qindex
= route_info
[ZEBRA_ROUTE_NHG
].meta_q_map
;
2077 ctx
= listgetdata(lnode
);
2082 if (IS_ZEBRA_DEBUG_RIB_DETAILED
)
2083 zlog_debug("NHG Context id=%u dequeued from sub-queue %u",
2086 rib_nhg_process(ctx
);
2089 static void process_subq_route(struct listnode
*lnode
, uint8_t qindex
)
2091 struct route_node
*rnode
= NULL
;
2092 rib_dest_t
*dest
= NULL
;
2093 struct zebra_vrf
*zvrf
= NULL
;
2095 rnode
= listgetdata(lnode
);
2096 dest
= rib_dest_from_rnode(rnode
);
2098 zvrf
= rib_dest_vrf(dest
);
2102 if (IS_ZEBRA_DEBUG_RIB_DETAILED
) {
2103 char buf
[SRCDEST2STR_BUFFER
];
2105 srcdest_rnode2str(rnode
, buf
, sizeof(buf
));
2106 zlog_debug("%u:%s: rn %p dequeued from sub-queue %u",
2107 zvrf
? zvrf_id(zvrf
) : 0, buf
, rnode
, qindex
);
2111 UNSET_FLAG(rib_dest_from_rnode(rnode
)->flags
,
2112 RIB_ROUTE_QUEUED(qindex
));
2117 zlog_debug ("%s: called for route_node (%p, %d) with no ribs",
2118 __func__
, rnode
, rnode
->lock
);
2119 zlog_backtrace(LOG_DEBUG
);
2122 route_unlock_node(rnode
);
2125 /* Take a list of route_node structs and return 1, if there was a record
2126 * picked from it and processed by rib_process(). Don't process more,
2127 * than one RN record; operate only in the specified sub-queue.
2129 static unsigned int process_subq(struct list
*subq
, uint8_t qindex
)
2131 struct listnode
*lnode
= listhead(subq
);
2136 if (qindex
== route_info
[ZEBRA_ROUTE_NHG
].meta_q_map
)
2137 process_subq_nhg(lnode
);
2139 process_subq_route(lnode
, qindex
);
2141 list_delete_node(subq
, lnode
);
2146 /* Dispatch the meta queue by picking, processing and unlocking the next RN from
2147 * a non-empty sub-queue with lowest priority. wq is equal to zebra->ribq and
2149 * is pointed to the meta queue structure.
2151 static wq_item_status
meta_queue_process(struct work_queue
*dummy
, void *data
)
2153 struct meta_queue
*mq
= data
;
2155 uint32_t queue_len
, queue_limit
;
2157 /* Ensure there's room for more dataplane updates */
2158 queue_limit
= dplane_get_in_queue_limit();
2159 queue_len
= dplane_get_in_queue_len();
2160 if (queue_len
> queue_limit
) {
2161 if (IS_ZEBRA_DEBUG_RIB_DETAILED
)
2162 zlog_debug("rib queue: dplane queue len %u, limit %u, retrying",
2163 queue_len
, queue_limit
);
2165 /* Ensure that the meta-queue is actually enqueued */
2166 if (work_queue_empty(zrouter
.ribq
))
2167 work_queue_add(zrouter
.ribq
, zrouter
.mq
);
2169 return WQ_QUEUE_BLOCKED
;
2172 for (i
= 0; i
< MQ_SIZE
; i
++)
2173 if (process_subq(mq
->subq
[i
], i
)) {
2177 return mq
->size
? WQ_REQUEUE
: WQ_SUCCESS
;
2182 * Look into the RN and queue it into the highest priority queue
2183 * at this point in time for processing.
2185 * We will enqueue a route node only once per invocation.
2187 * There are two possibilities here that should be kept in mind.
2188 * If the original invocation has not been pulled off for processing
2189 * yet, A subsuquent invocation can have a route entry with a better
2190 * meta queue index value and we can have a situation where
2191 * we might have the same node enqueued 2 times. Not necessarily
2192 * an optimal situation but it should be ok.
2194 * The other possibility is that the original invocation has not
2195 * been pulled off for processing yet, A subsusquent invocation
2196 * doesn't have a route_entry with a better meta-queue and the
2197 * original metaqueue index value will win and we'll end up with
2198 * the route node enqueued once.
2200 static int rib_meta_queue_add(struct meta_queue
*mq
, void *data
)
2202 struct route_node
*rn
= NULL
;
2203 struct route_entry
*re
= NULL
, *curr_re
= NULL
;
2204 uint8_t qindex
= MQ_SIZE
, curr_qindex
= MQ_SIZE
;
2206 rn
= (struct route_node
*)data
;
2208 RNODE_FOREACH_RE (rn
, curr_re
) {
2209 curr_qindex
= route_info
[curr_re
->type
].meta_q_map
;
2211 if (curr_qindex
<= qindex
) {
2213 qindex
= curr_qindex
;
2220 /* Invariant: at this point we always have rn->info set. */
2221 if (CHECK_FLAG(rib_dest_from_rnode(rn
)->flags
,
2222 RIB_ROUTE_QUEUED(qindex
))) {
2223 if (IS_ZEBRA_DEBUG_RIB_DETAILED
)
2224 rnode_debug(rn
, re
->vrf_id
,
2225 "rn %p is already queued in sub-queue %u",
2226 (void *)rn
, qindex
);
2230 SET_FLAG(rib_dest_from_rnode(rn
)->flags
, RIB_ROUTE_QUEUED(qindex
));
2231 listnode_add(mq
->subq
[qindex
], rn
);
2232 route_lock_node(rn
);
2235 if (IS_ZEBRA_DEBUG_RIB_DETAILED
)
2236 rnode_debug(rn
, re
->vrf_id
, "queued rn %p into sub-queue %u",
2237 (void *)rn
, qindex
);
2242 static int rib_meta_queue_nhg_add(struct meta_queue
*mq
, void *data
)
2244 struct nhg_ctx
*ctx
= NULL
;
2245 uint8_t qindex
= route_info
[ZEBRA_ROUTE_NHG
].meta_q_map
;
2247 ctx
= (struct nhg_ctx
*)data
;
2252 listnode_add(mq
->subq
[qindex
], ctx
);
2255 if (IS_ZEBRA_DEBUG_RIB_DETAILED
)
2256 zlog_debug("NHG Context id=%u queued into sub-queue %u",
2262 static int mq_add_handler(void *data
,
2263 int (*mq_add_func
)(struct meta_queue
*mq
, void *data
))
2265 if (zrouter
.ribq
== NULL
) {
2266 flog_err(EC_ZEBRA_WQ_NONEXISTENT
,
2267 "%s: work_queue does not exist!", __func__
);
2272 * The RIB queue should normally be either empty or holding the only
2273 * work_queue_item element. In the latter case this element would
2274 * hold a pointer to the meta queue structure, which must be used to
2275 * actually queue the route nodes to process. So create the MQ
2276 * holder, if necessary, then push the work into it in any case.
2277 * This semantics was introduced after 0.99.9 release.
2279 if (work_queue_empty(zrouter
.ribq
))
2280 work_queue_add(zrouter
.ribq
, zrouter
.mq
);
2282 return mq_add_func(zrouter
.mq
, data
);
2285 /* Add route_node to work queue and schedule processing */
2286 int rib_queue_add(struct route_node
*rn
)
2290 /* Pointless to queue a route_node with no RIB entries to add or remove
2292 if (!rnode_to_ribs(rn
)) {
2293 zlog_debug("%s: called for route_node (%p, %d) with no ribs",
2294 __func__
, (void *)rn
, rn
->lock
);
2295 zlog_backtrace(LOG_DEBUG
);
2299 return mq_add_handler(rn
, &rib_meta_queue_add
);
2302 int rib_queue_nhg_add(struct nhg_ctx
*ctx
)
2306 return mq_add_handler(ctx
, &rib_meta_queue_nhg_add
);
2309 /* Create new meta queue.
2310 A destructor function doesn't seem to be necessary here.
2312 static struct meta_queue
*meta_queue_new(void)
2314 struct meta_queue
*new;
2317 new = XCALLOC(MTYPE_WORK_QUEUE
, sizeof(struct meta_queue
));
2319 for (i
= 0; i
< MQ_SIZE
; i
++) {
2320 new->subq
[i
] = list_new();
2321 assert(new->subq
[i
]);
2327 void meta_queue_free(struct meta_queue
*mq
)
2331 for (i
= 0; i
< MQ_SIZE
; i
++)
2332 list_delete(&mq
->subq
[i
]);
2334 XFREE(MTYPE_WORK_QUEUE
, mq
);
2337 /* initialise zebra rib work queue */
2338 static void rib_queue_init(void)
2340 if (!(zrouter
.ribq
= work_queue_new(zrouter
.master
,
2341 "route_node processing"))) {
2342 flog_err(EC_ZEBRA_WQ_NONEXISTENT
,
2343 "%s: could not initialise work queue!", __func__
);
2347 /* fill in the work queue spec */
2348 zrouter
.ribq
->spec
.workfunc
= &meta_queue_process
;
2349 zrouter
.ribq
->spec
.errorfunc
= NULL
;
2350 zrouter
.ribq
->spec
.completion_func
= NULL
;
2351 /* XXX: TODO: These should be runtime configurable via vty */
2352 zrouter
.ribq
->spec
.max_retries
= 3;
2353 zrouter
.ribq
->spec
.hold
= ZEBRA_RIB_PROCESS_HOLD_TIME
;
2354 zrouter
.ribq
->spec
.retry
= ZEBRA_RIB_PROCESS_RETRY_TIME
;
2356 if (!(zrouter
.mq
= meta_queue_new())) {
2357 flog_err(EC_ZEBRA_WQ_NONEXISTENT
,
2358 "%s: could not initialise meta queue!", __func__
);
2364 rib_dest_t
*zebra_rib_create_dest(struct route_node
*rn
)
2368 dest
= XCALLOC(MTYPE_RIB_DEST
, sizeof(rib_dest_t
));
2369 rnh_list_init(&dest
->nht
);
2370 route_lock_node(rn
); /* rn route table reference */
2377 /* RIB updates are processed via a queue of pointers to route_nodes.
2379 * The queue length is bounded by the maximal size of the routing table,
2380 * as a route_node will not be requeued, if already queued.
2382 * REs are submitted via rib_addnode or rib_delnode which set minimal
2383 * state, or static_install_route (when an existing RE is updated)
2384 * and then submit route_node to queue for best-path selection later.
2385 * Order of add/delete state changes are preserved for any given RE.
2387 * Deleted REs are reaped during best-path selection.
2390 * |-> rib_link or unset ROUTE_ENTRY_REMOVE |->Update kernel with
2391 * |-------->| | best RE, if required
2393 * static_install->|->rib_addqueue...... -> rib_process
2395 * |-------->| |-> rib_unlink
2396 * |-> set ROUTE_ENTRY_REMOVE |
2397 * rib_delnode (RE freed)
2399 * The 'info' pointer of a route_node points to a rib_dest_t
2400 * ('dest'). Queueing state for a route_node is kept on the dest. The
2401 * dest is created on-demand by rib_link() and is kept around at least
2402 * as long as there are ribs hanging off it (@see rib_gc_dest()).
2404 * Refcounting (aka "locking" throughout the GNU Zebra and Quagga code):
2406 * - route_nodes: refcounted by:
2407 * - dest attached to route_node:
2408 * - managed by: rib_link/rib_gc_dest
2409 * - route_node processing queue
2410 * - managed by: rib_addqueue, rib_process.
2414 /* Add RE to head of the route node. */
2415 static void rib_link(struct route_node
*rn
, struct route_entry
*re
, int process
)
2419 const char *rmap_name
;
2423 dest
= rib_dest_from_rnode(rn
);
2425 if (IS_ZEBRA_DEBUG_RIB_DETAILED
)
2426 rnode_debug(rn
, re
->vrf_id
, "rn %p adding dest", rn
);
2428 dest
= zebra_rib_create_dest(rn
);
2431 re_list_add_head(&dest
->routes
, re
);
2433 afi
= (rn
->p
.family
== AF_INET
)
2435 : (rn
->p
.family
== AF_INET6
) ? AFI_IP6
: AFI_MAX
;
2436 if (is_zebra_import_table_enabled(afi
, re
->vrf_id
, re
->table
)) {
2437 struct zebra_vrf
*zvrf
= zebra_vrf_lookup_by_id(re
->vrf_id
);
2439 rmap_name
= zebra_get_import_table_route_map(afi
, re
->table
);
2440 zebra_add_import_table_entry(zvrf
, rn
, re
, rmap_name
);
2445 static void rib_addnode(struct route_node
*rn
,
2446 struct route_entry
*re
, int process
)
2448 /* RE node has been un-removed before route-node is processed.
2449 * route_node must hence already be on the queue for processing..
2451 if (CHECK_FLAG(re
->status
, ROUTE_ENTRY_REMOVED
)) {
2452 if (IS_ZEBRA_DEBUG_RIB
)
2453 rnode_debug(rn
, re
->vrf_id
, "rn %p, un-removed re %p",
2454 (void *)rn
, (void *)re
);
2456 UNSET_FLAG(re
->status
, ROUTE_ENTRY_REMOVED
);
2459 rib_link(rn
, re
, process
);
2465 * Detach a rib structure from a route_node.
2467 * Note that a call to rib_unlink() should be followed by a call to
2468 * rib_gc_dest() at some point. This allows a rib_dest_t that is no
2469 * longer required to be deleted.
2471 void rib_unlink(struct route_node
*rn
, struct route_entry
*re
)
2474 struct nhg_hash_entry
*nhe
= NULL
;
2478 if (IS_ZEBRA_DEBUG_RIB
)
2479 rnode_debug(rn
, re
->vrf_id
, "rn %p, re %p", (void *)rn
,
2482 dest
= rib_dest_from_rnode(rn
);
2484 re_list_del(&dest
->routes
, re
);
2486 if (dest
->selected_fib
== re
)
2487 dest
->selected_fib
= NULL
;
2490 nhe
= zebra_nhg_lookup_id(re
->nhe_id
);
2492 zebra_nhg_decrement_ref(nhe
);
2494 nexthop_group_free_delete(&re
->ng
);
2496 nexthops_free(re
->fib_ng
.nexthop
);
2498 XFREE(MTYPE_RE
, re
);
2501 void rib_delnode(struct route_node
*rn
, struct route_entry
*re
)
2505 if (IS_ZEBRA_DEBUG_RIB
)
2506 rnode_debug(rn
, re
->vrf_id
, "rn %p, re %p, removing",
2507 (void *)rn
, (void *)re
);
2508 SET_FLAG(re
->status
, ROUTE_ENTRY_REMOVED
);
2510 afi
= (rn
->p
.family
== AF_INET
)
2512 : (rn
->p
.family
== AF_INET6
) ? AFI_IP6
: AFI_MAX
;
2513 if (is_zebra_import_table_enabled(afi
, re
->vrf_id
, re
->table
)) {
2514 struct zebra_vrf
*zvrf
= zebra_vrf_lookup_by_id(re
->vrf_id
);
2516 zebra_del_import_table_entry(zvrf
, rn
, re
);
2517 /* Just clean up if non main table */
2518 if (IS_ZEBRA_DEBUG_RIB
) {
2519 char buf
[SRCDEST2STR_BUFFER
];
2520 srcdest_rnode2str(rn
, buf
, sizeof(buf
));
2521 zlog_debug("%u:%s: Freeing route rn %p, re %p (%s)",
2522 re
->vrf_id
, buf
, rn
, re
,
2523 zebra_route_string(re
->type
));
2532 /* This function dumps the contents of a given RE entry into
2533 * standard debug log. Calling function name and IP prefix in
2534 * question are passed as 1st and 2nd arguments.
2537 void _route_entry_dump(const char *func
, union prefixconstptr pp
,
2538 union prefixconstptr src_pp
,
2539 const struct route_entry
*re
)
2541 const struct prefix
*src_p
= src_pp
.p
;
2542 bool is_srcdst
= src_p
&& src_p
->prefixlen
;
2543 char straddr
[PREFIX_STRLEN
];
2544 char srcaddr
[PREFIX_STRLEN
];
2545 char nhname
[PREFIX_STRLEN
];
2546 struct nexthop
*nexthop
;
2548 zlog_debug("%s: dumping RE entry %p for %s%s%s vrf %u", func
,
2549 (const void *)re
, prefix2str(pp
, straddr
, sizeof(straddr
)),
2550 is_srcdst
? " from " : "",
2551 is_srcdst
? prefix2str(src_pp
, srcaddr
, sizeof(srcaddr
))
2554 zlog_debug("%s: uptime == %lu, type == %u, instance == %d, table == %d",
2555 straddr
, (unsigned long)re
->uptime
, re
->type
, re
->instance
,
2558 "%s: metric == %u, mtu == %u, distance == %u, flags == %u, status == %u",
2559 straddr
, re
->metric
, re
->mtu
, re
->distance
, re
->flags
, re
->status
);
2560 zlog_debug("%s: nexthop_num == %u, nexthop_active_num == %u", straddr
,
2561 nexthop_group_nexthop_num(re
->ng
),
2562 nexthop_group_active_nexthop_num(re
->ng
));
2564 for (ALL_NEXTHOPS_PTR(re
->ng
, nexthop
)) {
2565 struct interface
*ifp
;
2566 struct vrf
*vrf
= vrf_lookup_by_id(nexthop
->vrf_id
);
2568 switch (nexthop
->type
) {
2569 case NEXTHOP_TYPE_BLACKHOLE
:
2570 sprintf(nhname
, "Blackhole");
2572 case NEXTHOP_TYPE_IFINDEX
:
2573 ifp
= if_lookup_by_index(nexthop
->ifindex
,
2575 sprintf(nhname
, "%s", ifp
? ifp
->name
: "Unknown");
2577 case NEXTHOP_TYPE_IPV4
:
2579 case NEXTHOP_TYPE_IPV4_IFINDEX
:
2580 inet_ntop(AF_INET
, &nexthop
->gate
, nhname
,
2583 case NEXTHOP_TYPE_IPV6
:
2584 case NEXTHOP_TYPE_IPV6_IFINDEX
:
2585 inet_ntop(AF_INET6
, &nexthop
->gate
, nhname
,
2589 zlog_debug("%s: %s %s[%u] vrf %s(%u) with flags %s%s%s%s%s%s",
2590 straddr
, (nexthop
->rparent
? " NH" : "NH"), nhname
,
2591 nexthop
->ifindex
, vrf
? vrf
->name
: "Unknown",
2593 (CHECK_FLAG(nexthop
->flags
, NEXTHOP_FLAG_ACTIVE
)
2596 (CHECK_FLAG(re
->status
, ROUTE_ENTRY_INSTALLED
)
2599 (CHECK_FLAG(nexthop
->flags
, NEXTHOP_FLAG_RECURSIVE
)
2602 (CHECK_FLAG(nexthop
->flags
, NEXTHOP_FLAG_ONLINK
)
2605 (CHECK_FLAG(nexthop
->flags
, NEXTHOP_FLAG_MATCHED
)
2608 (CHECK_FLAG(nexthop
->flags
, NEXTHOP_FLAG_DUPLICATE
)
2612 zlog_debug("%s: dump complete", straddr
);
2615 /* This is an exported helper to rtm_read() to dump the strange
2616 * RE entry found by rib_lookup_ipv4_route()
2619 void rib_lookup_and_dump(struct prefix_ipv4
*p
, vrf_id_t vrf_id
)
2621 struct route_table
*table
;
2622 struct route_node
*rn
;
2623 struct route_entry
*re
;
2624 char prefix_buf
[INET_ADDRSTRLEN
];
2627 table
= zebra_vrf_table(AFI_IP
, SAFI_UNICAST
, vrf_id
);
2629 flog_err(EC_ZEBRA_TABLE_LOOKUP_FAILED
,
2630 "%s:%u zebra_vrf_table() returned NULL", __func__
,
2635 /* Scan the RIB table for exactly matching RE entry. */
2636 rn
= route_node_lookup(table
, (struct prefix
*)p
);
2638 /* No route for this prefix. */
2640 zlog_debug("%s:%u lookup failed for %s", __func__
, vrf_id
,
2641 prefix2str((struct prefix
*)p
, prefix_buf
,
2642 sizeof(prefix_buf
)));
2647 route_unlock_node(rn
);
2650 RNODE_FOREACH_RE (rn
, re
) {
2651 zlog_debug("%s:%u rn %p, re %p: %s, %s",
2653 (void *)rn
, (void *)re
,
2654 (CHECK_FLAG(re
->status
, ROUTE_ENTRY_REMOVED
)
2657 (CHECK_FLAG(re
->flags
, ZEBRA_FLAG_SELECTED
)
2660 route_entry_dump(p
, NULL
, re
);
2664 /* Check if requested address assignment will fail due to another
2665 * route being installed by zebra in FIB already. Take necessary
2666 * actions, if needed: remove such a route from FIB and deSELECT
2667 * corresponding RE entry. Then put affected RN into RIBQ head.
2669 void rib_lookup_and_pushup(struct prefix_ipv4
*p
, vrf_id_t vrf_id
)
2671 struct route_table
*table
;
2672 struct route_node
*rn
;
2675 if (NULL
== (table
= zebra_vrf_table(AFI_IP
, SAFI_UNICAST
, vrf_id
))) {
2676 flog_err(EC_ZEBRA_TABLE_LOOKUP_FAILED
,
2677 "%s:%u zebra_vrf_table() returned NULL", __func__
,
2682 /* No matches would be the simplest case. */
2683 if (NULL
== (rn
= route_node_lookup(table
, (struct prefix
*)p
)))
2687 route_unlock_node(rn
);
2689 dest
= rib_dest_from_rnode(rn
);
2690 /* Check all RE entries. In case any changes have to be done, requeue
2691 * the RN into RIBQ head. If the routing message about the new connected
2692 * route (generated by the IP address we are going to assign very soon)
2693 * comes before the RIBQ is processed, the new RE entry will join
2694 * RIBQ record already on head. This is necessary for proper
2696 * of the rest of the RE.
2698 if (dest
->selected_fib
) {
2699 if (IS_ZEBRA_DEBUG_RIB
) {
2700 char buf
[PREFIX_STRLEN
];
2702 zlog_debug("%u:%s: freeing way for connected prefix",
2703 dest
->selected_fib
->vrf_id
,
2704 prefix2str(&rn
->p
, buf
, sizeof(buf
)));
2705 route_entry_dump(&rn
->p
, NULL
, dest
->selected_fib
);
2707 rib_uninstall(rn
, dest
->selected_fib
);
2712 int rib_add_multipath(afi_t afi
, safi_t safi
, struct prefix
*p
,
2713 struct prefix_ipv6
*src_p
, struct route_entry
*re
)
2715 struct nhg_hash_entry
*nhe
= NULL
;
2716 struct route_table
*table
;
2717 struct route_node
*rn
;
2718 struct route_entry
*same
= NULL
;
2724 assert(!src_p
|| !src_p
->prefixlen
|| afi
== AFI_IP6
);
2727 table
= zebra_vrf_table_with_table_id(afi
, safi
, re
->vrf_id
, re
->table
);
2730 nexthop_group_free_delete(&re
->ng
);
2731 XFREE(MTYPE_RE
, re
);
2736 nhe
= zebra_nhg_lookup_id(re
->nhe_id
);
2740 EC_ZEBRA_TABLE_LOOKUP_FAILED
,
2741 "Zebra failed to find the nexthop hash entry for id=%u in a route entry",
2743 XFREE(MTYPE_RE
, re
);
2747 nhe
= zebra_nhg_rib_find(0, re
->ng
, afi
);
2750 * The nexthops got copied over into an nhe,
2753 nexthop_group_free_delete(&re
->ng
);
2756 char buf
[PREFIX_STRLEN
] = "";
2757 char buf2
[PREFIX_STRLEN
] = "";
2760 EC_ZEBRA_TABLE_LOOKUP_FAILED
,
2761 "Zebra failed to find or create a nexthop hash entry for %s%s%s",
2762 prefix2str(p
, buf
, sizeof(buf
)),
2763 src_p
? " from " : "",
2764 src_p
? prefix2str(src_p
, buf2
, sizeof(buf2
))
2767 XFREE(MTYPE_RE
, re
);
2773 * Attach the re to the nhe's nexthop group.
2775 * TODO: This will need to change when we start getting IDs from upper
2776 * level protocols, as the refcnt might be wrong, since it checks
2777 * if old_id != new_id.
2779 zebra_nhg_re_update_ref(re
, nhe
);
2781 /* Make it sure prefixlen is applied to the prefix. */
2784 apply_mask_ipv6(src_p
);
2786 /* Set default distance by route type. */
2787 if (re
->distance
== 0)
2788 re
->distance
= route_distance(re
->type
);
2790 /* Lookup route node.*/
2791 rn
= srcdest_rnode_get(table
, p
, src_p
);
2794 * If same type of route are installed, treat it as a implicit
2796 * If the user has specified the No route replace semantics
2797 * for the install don't do a route replace.
2799 RNODE_FOREACH_RE (rn
, same
) {
2800 if (CHECK_FLAG(same
->status
, ROUTE_ENTRY_REMOVED
))
2803 if (same
->type
!= re
->type
)
2805 if (same
->instance
!= re
->instance
)
2807 if (same
->type
== ZEBRA_ROUTE_KERNEL
2808 && same
->metric
!= re
->metric
)
2811 if (CHECK_FLAG(re
->flags
, ZEBRA_FLAG_RR_USE_DISTANCE
) &&
2812 same
->distance
!= re
->distance
)
2816 * We should allow duplicate connected routes
2817 * because of IPv6 link-local routes and unnumbered
2818 * interfaces on Linux.
2820 if (same
->type
!= ZEBRA_ROUTE_CONNECT
)
2824 /* If this route is kernel/connected route, notify the dataplane. */
2825 if (RIB_SYSTEM_ROUTE(re
)) {
2826 /* Notify dataplane */
2827 dplane_sys_route_add(rn
, re
);
2830 /* Link new re to node.*/
2831 if (IS_ZEBRA_DEBUG_RIB
) {
2832 rnode_debug(rn
, re
->vrf_id
,
2833 "Inserting route rn %p, re %p (%s) existing %p",
2834 rn
, re
, zebra_route_string(re
->type
), same
);
2836 if (IS_ZEBRA_DEBUG_RIB_DETAILED
)
2837 route_entry_dump(p
, src_p
, re
);
2840 SET_FLAG(re
->status
, ROUTE_ENTRY_CHANGED
);
2841 rib_addnode(rn
, re
, 1);
2844 /* Free implicit route.*/
2846 rib_delnode(rn
, same
);
2850 route_unlock_node(rn
);
2854 void rib_delete(afi_t afi
, safi_t safi
, vrf_id_t vrf_id
, int type
,
2855 unsigned short instance
, int flags
, struct prefix
*p
,
2856 struct prefix_ipv6
*src_p
, const struct nexthop
*nh
,
2857 uint32_t nhe_id
, uint32_t table_id
, uint32_t metric
,
2858 uint8_t distance
, bool fromkernel
)
2860 struct route_table
*table
;
2861 struct route_node
*rn
;
2862 struct route_entry
*re
;
2863 struct route_entry
*fib
= NULL
;
2864 struct route_entry
*same
= NULL
;
2865 struct nexthop
*rtnh
;
2866 char buf2
[INET6_ADDRSTRLEN
];
2869 assert(!src_p
|| !src_p
->prefixlen
|| afi
== AFI_IP6
);
2872 table
= zebra_vrf_table_with_table_id(afi
, safi
, vrf_id
, table_id
);
2879 apply_mask_ipv6(src_p
);
2881 /* Lookup route node. */
2882 rn
= srcdest_rnode_lookup(table
, p
, src_p
);
2884 char dst_buf
[PREFIX_STRLEN
], src_buf
[PREFIX_STRLEN
];
2886 prefix2str(p
, dst_buf
, sizeof(dst_buf
));
2887 if (src_p
&& src_p
->prefixlen
)
2888 prefix2str(src_p
, src_buf
, sizeof(src_buf
));
2892 if (IS_ZEBRA_DEBUG_RIB
) {
2893 struct vrf
*vrf
= vrf_lookup_by_id(vrf_id
);
2895 zlog_debug("%s[%d]:%s%s%s doesn't exist in rib",
2896 vrf
->name
, table_id
, dst_buf
,
2897 (src_buf
[0] != '\0') ? " from " : "",
2903 dest
= rib_dest_from_rnode(rn
);
2904 fib
= dest
->selected_fib
;
2906 /* Lookup same type route. */
2907 RNODE_FOREACH_RE (rn
, re
) {
2908 if (CHECK_FLAG(re
->status
, ROUTE_ENTRY_REMOVED
))
2911 if (re
->type
!= type
)
2913 if (re
->instance
!= instance
)
2915 if (CHECK_FLAG(re
->flags
, ZEBRA_FLAG_RR_USE_DISTANCE
) &&
2916 distance
!= re
->distance
)
2919 if (re
->type
== ZEBRA_ROUTE_KERNEL
&& re
->metric
!= metric
)
2921 if (re
->type
== ZEBRA_ROUTE_CONNECT
&& (rtnh
= re
->ng
->nexthop
)
2922 && rtnh
->type
== NEXTHOP_TYPE_IFINDEX
&& nh
) {
2923 if (rtnh
->ifindex
!= nh
->ifindex
)
2928 /* Make sure that the route found has the same gateway. */
2929 else if (nhe_id
&& re
->nhe_id
== nhe_id
) {
2937 for (ALL_NEXTHOPS_PTR(re
->ng
, rtnh
))
2939 * No guarantee all kernel send nh with labels
2942 if (nexthop_same_no_labels(rtnh
, nh
)) {
2950 /* If same type of route can't be found and this message is from
2954 * In the past(HA!) we could get here because
2955 * we were receiving a route delete from the
2956 * kernel and we're not marking the proto
2957 * as coming from it's appropriate originator.
2958 * Now that we are properly noticing the fact
2959 * that the kernel has deleted our route we
2960 * are not going to get called in this path
2961 * I am going to leave this here because
2962 * this might still work this way on non-linux
2963 * platforms as well as some weird state I have
2964 * not properly thought of yet.
2965 * If we can show that this code path is
2966 * dead then we can remove it.
2968 if (fib
&& CHECK_FLAG(flags
, ZEBRA_FLAG_SELFROUTE
)) {
2969 if (IS_ZEBRA_DEBUG_RIB
) {
2970 rnode_debug(rn
, vrf_id
,
2971 "rn %p, re %p (%s) was deleted from kernel, adding",
2973 zebra_route_string(fib
->type
));
2976 UNSET_FLAG(fib
->status
, ROUTE_ENTRY_INSTALLED
);
2978 for (rtnh
= fib
->ng
->nexthop
; rtnh
;
2980 UNSET_FLAG(rtnh
->flags
,
2984 * This is a non FRR route
2985 * as such we should mark
2988 dest
->selected_fib
= NULL
;
2990 /* This means someone else, other than Zebra,
2992 * a Zebra router from the kernel. We will add
2994 rib_install_kernel(rn
, fib
, NULL
);
2997 if (IS_ZEBRA_DEBUG_RIB
) {
3001 "via %s ifindex %d type %d "
3002 "doesn't exist in rib",
3003 inet_ntop(afi2family(afi
),
3010 "type %d doesn't exist in rib",
3013 route_unlock_node(rn
);
3019 if (fromkernel
&& CHECK_FLAG(flags
, ZEBRA_FLAG_SELFROUTE
)
3021 rib_install_kernel(rn
, same
, NULL
);
3022 route_unlock_node(rn
);
3027 /* Special handling for IPv4 or IPv6 routes sourced from
3028 * EVPN - the nexthop (and associated MAC) need to be
3029 * uninstalled if no more refs.
3031 if (CHECK_FLAG(flags
, ZEBRA_FLAG_EVPN_ROUTE
)) {
3032 struct nexthop
*tmp_nh
;
3034 for (ALL_NEXTHOPS_PTR(re
->ng
, tmp_nh
)) {
3035 struct ipaddr vtep_ip
;
3037 memset(&vtep_ip
, 0, sizeof(struct ipaddr
));
3038 if (afi
== AFI_IP
) {
3039 vtep_ip
.ipa_type
= IPADDR_V4
;
3040 memcpy(&(vtep_ip
.ipaddr_v4
),
3041 &(tmp_nh
->gate
.ipv4
),
3042 sizeof(struct in_addr
));
3044 vtep_ip
.ipa_type
= IPADDR_V6
;
3045 memcpy(&(vtep_ip
.ipaddr_v6
),
3046 &(tmp_nh
->gate
.ipv6
),
3047 sizeof(struct in6_addr
));
3049 zebra_vxlan_evpn_vrf_route_del(re
->vrf_id
,
3054 /* Notify dplane if system route changes */
3055 if (RIB_SYSTEM_ROUTE(re
))
3056 dplane_sys_route_del(rn
, same
);
3058 rib_delnode(rn
, same
);
3061 route_unlock_node(rn
);
3066 int rib_add(afi_t afi
, safi_t safi
, vrf_id_t vrf_id
, int type
,
3067 unsigned short instance
, int flags
, struct prefix
*p
,
3068 struct prefix_ipv6
*src_p
, const struct nexthop
*nh
,
3069 uint32_t nhe_id
, uint32_t table_id
, uint32_t metric
, uint32_t mtu
,
3070 uint8_t distance
, route_tag_t tag
)
3072 struct route_entry
*re
= NULL
;
3073 struct nexthop
*nexthop
= NULL
;
3075 /* Allocate new route_entry structure. */
3076 re
= XCALLOC(MTYPE_RE
, sizeof(struct route_entry
));
3078 re
->instance
= instance
;
3079 re
->distance
= distance
;
3081 re
->metric
= metric
;
3083 re
->table
= table_id
;
3084 re
->vrf_id
= vrf_id
;
3085 re
->uptime
= monotime(NULL
);
3087 re
->nhe_id
= nhe_id
;
3090 re
->ng
= nexthop_group_new();
3093 nexthop
= nexthop_new();
3095 route_entry_nexthop_add(re
, nexthop
);
3098 return rib_add_multipath(afi
, safi
, p
, src_p
, re
);
3101 static const char *rib_update_event2str(rib_update_event_t event
)
3103 const char *ret
= "UNKNOWN";
3106 case RIB_UPDATE_KERNEL
:
3107 ret
= "RIB_UPDATE_KERNEL";
3109 case RIB_UPDATE_RMAP_CHANGE
:
3110 ret
= "RIB_UPDATE_RMAP_CHANGE";
3112 case RIB_UPDATE_OTHER
:
3113 ret
= "RIB_UPDATE_OTHER";
3115 case RIB_UPDATE_MAX
:
3123 /* Schedule route nodes to be processed if they match the type */
3124 static void rib_update_route_node(struct route_node
*rn
, int type
)
3126 struct route_entry
*re
, *next
;
3127 bool re_changed
= false;
3129 RNODE_FOREACH_RE_SAFE (rn
, re
, next
) {
3130 if (type
== ZEBRA_ROUTE_ALL
|| type
== re
->type
) {
3131 SET_FLAG(re
->status
, ROUTE_ENTRY_CHANGED
);
3140 /* Schedule routes of a particular table (address-family) based on event. */
3141 void rib_update_table(struct route_table
*table
, rib_update_event_t event
)
3143 struct route_node
*rn
;
3145 if (IS_ZEBRA_DEBUG_EVENT
) {
3146 struct zebra_vrf
*zvrf
;
3149 zvrf
= table
->info
? ((rib_table_info_t
*)table
->info
)->zvrf
3151 vrf
= zvrf
? zvrf
->vrf
: NULL
;
3153 zlog_debug("%s: %s VRF %s Table %u event %s", __func__
,
3154 table
->info
? afi2str(
3155 ((rib_table_info_t
*)table
->info
)->afi
)
3157 vrf
? vrf
->name
: "Unknown",
3158 zvrf
? zvrf
->table_id
: 0,
3159 rib_update_event2str(event
));
3162 /* Walk all routes and queue for processing, if appropriate for
3163 * the trigger event.
3165 for (rn
= route_top(table
); rn
; rn
= srcdest_route_next(rn
)) {
3167 * If we are looking at a route node and the node
3168 * has already been queued we don't
3169 * need to queue it up again
3172 && CHECK_FLAG(rib_dest_from_rnode(rn
)->flags
,
3173 RIB_ROUTE_ANY_QUEUED
))
3177 case RIB_UPDATE_KERNEL
:
3178 rib_update_route_node(rn
, ZEBRA_ROUTE_KERNEL
);
3180 case RIB_UPDATE_RMAP_CHANGE
:
3181 case RIB_UPDATE_OTHER
:
3182 rib_update_route_node(rn
, ZEBRA_ROUTE_ALL
);
3190 static void rib_update_handle_vrf(vrf_id_t vrf_id
, rib_update_event_t event
)
3192 struct route_table
*table
;
3194 if (IS_ZEBRA_DEBUG_EVENT
)
3195 zlog_debug("%s: Handling VRF %s event %s", __func__
,
3196 vrf_id_to_name(vrf_id
), rib_update_event2str(event
));
3198 /* Process routes of interested address-families. */
3199 table
= zebra_vrf_table(AFI_IP
, SAFI_UNICAST
, vrf_id
);
3201 rib_update_table(table
, event
);
3203 table
= zebra_vrf_table(AFI_IP6
, SAFI_UNICAST
, vrf_id
);
3205 rib_update_table(table
, event
);
3208 static void rib_update_handle_vrf_all(rib_update_event_t event
)
3210 struct zebra_router_table
*zrt
;
3212 if (IS_ZEBRA_DEBUG_EVENT
)
3213 zlog_debug("%s: Handling VRF (ALL) event %s", __func__
,
3214 rib_update_event2str(event
));
3216 /* Just iterate over all the route tables, rather than vrf lookups */
3217 RB_FOREACH (zrt
, zebra_router_table_head
, &zrouter
.tables
)
3218 rib_update_table(zrt
->table
, event
);
3221 struct rib_update_ctx
{
3222 rib_update_event_t event
;
3227 static struct rib_update_ctx
*rib_update_ctx_init(vrf_id_t vrf_id
,
3228 rib_update_event_t event
)
3230 struct rib_update_ctx
*ctx
;
3232 ctx
= XCALLOC(MTYPE_RIB_UPDATE_CTX
, sizeof(struct rib_update_ctx
));
3235 ctx
->vrf_id
= vrf_id
;
3240 static void rib_update_ctx_fini(struct rib_update_ctx
**ctx
)
3242 XFREE(MTYPE_RIB_UPDATE_CTX
, *ctx
);
3247 static int rib_update_handler(struct thread
*thread
)
3249 struct rib_update_ctx
*ctx
;
3251 ctx
= THREAD_ARG(thread
);
3254 rib_update_handle_vrf_all(ctx
->event
);
3256 rib_update_handle_vrf(ctx
->vrf_id
, ctx
->event
);
3258 rib_update_ctx_fini(&ctx
);
3264 * Thread list to ensure we don't schedule a ton of events
3265 * if interfaces are flapping for instance.
3267 static struct thread
*t_rib_update_threads
[RIB_UPDATE_MAX
];
3269 /* Schedule a RIB update event for specific vrf */
3270 void rib_update_vrf(vrf_id_t vrf_id
, rib_update_event_t event
)
3272 struct rib_update_ctx
*ctx
;
3274 ctx
= rib_update_ctx_init(vrf_id
, event
);
3276 /* Don't worry about making sure multiple rib updates for specific vrf
3277 * are scheduled at once for now. If it becomes a problem, we can use a
3278 * lookup of some sort to keep track of running threads via t_vrf_id
3279 * like how we are doing it in t_rib_update_threads[].
3281 thread_add_event(zrouter
.master
, rib_update_handler
, ctx
, 0, NULL
);
3283 if (IS_ZEBRA_DEBUG_EVENT
)
3284 zlog_debug("%s: Scheduled VRF %s, event %s", __func__
,
3285 vrf_id_to_name(ctx
->vrf_id
),
3286 rib_update_event2str(event
));
3289 /* Schedule a RIB update event for all vrfs */
3290 void rib_update(rib_update_event_t event
)
3292 struct rib_update_ctx
*ctx
;
3294 ctx
= rib_update_ctx_init(0, event
);
3296 ctx
->vrf_all
= true;
3298 if (!thread_add_event(zrouter
.master
, rib_update_handler
, ctx
, 0,
3299 &t_rib_update_threads
[event
]))
3300 rib_update_ctx_fini(&ctx
); /* Already scheduled */
3301 else if (IS_ZEBRA_DEBUG_EVENT
)
3302 zlog_debug("%s: Schedued VRF (ALL), event %s", __func__
,
3303 rib_update_event2str(event
));
3306 /* Delete self installed routes after zebra is relaunched. */
3307 void rib_sweep_table(struct route_table
*table
)
3309 struct route_node
*rn
;
3310 struct route_entry
*re
;
3311 struct route_entry
*next
;
3312 struct nexthop
*nexthop
;
3317 for (rn
= route_top(table
); rn
; rn
= srcdest_route_next(rn
)) {
3318 RNODE_FOREACH_RE_SAFE (rn
, re
, next
) {
3320 if (IS_ZEBRA_DEBUG_RIB
)
3321 route_entry_dump(&rn
->p
, NULL
, re
);
3323 if (CHECK_FLAG(re
->status
, ROUTE_ENTRY_REMOVED
))
3326 if (!CHECK_FLAG(re
->flags
, ZEBRA_FLAG_SELFROUTE
))
3330 * If routes are older than startup_time then
3331 * we know we read them in from the kernel.
3332 * As such we can safely remove them.
3334 if (zrouter
.startup_time
< re
->uptime
)
3338 * So we are starting up and have received
3339 * routes from the kernel that we have installed
3340 * from a previous run of zebra but not cleaned
3341 * up ( say a kill -9 )
3342 * But since we haven't actually installed
3343 * them yet( we received them from the kernel )
3344 * we don't think they are active.
3345 * So let's pretend they are active to actually
3347 * In all honesty I'm not sure if we should
3348 * mark them as active when we receive them
3349 * This is startup only so probably ok.
3351 * If we ever decide to move rib_sweep_table
3352 * to a different spot (ie startup )
3353 * this decision needs to be revisited
3355 SET_FLAG(re
->status
, ROUTE_ENTRY_INSTALLED
);
3356 for (ALL_NEXTHOPS_PTR(re
->ng
, nexthop
))
3357 SET_FLAG(nexthop
->flags
, NEXTHOP_FLAG_FIB
);
3359 rib_uninstall_kernel(rn
, re
);
3360 rib_delnode(rn
, re
);
3365 /* Sweep all RIB tables. */
3366 int rib_sweep_route(struct thread
*t
)
3369 struct zebra_vrf
*zvrf
;
3371 RB_FOREACH (vrf
, vrf_id_head
, &vrfs_by_id
) {
3372 if ((zvrf
= vrf
->info
) == NULL
)
3375 rib_sweep_table(zvrf
->table
[AFI_IP
][SAFI_UNICAST
]);
3376 rib_sweep_table(zvrf
->table
[AFI_IP6
][SAFI_UNICAST
]);
3379 zebra_router_sweep_route();
3380 zebra_router_sweep_nhgs();
3385 /* Remove specific by protocol routes from 'table'. */
3386 unsigned long rib_score_proto_table(uint8_t proto
, unsigned short instance
,
3387 struct route_table
*table
)
3389 struct route_node
*rn
;
3390 struct route_entry
*re
;
3391 struct route_entry
*next
;
3392 unsigned long n
= 0;
3395 for (rn
= route_top(table
); rn
; rn
= srcdest_route_next(rn
))
3396 RNODE_FOREACH_RE_SAFE (rn
, re
, next
) {
3397 if (CHECK_FLAG(re
->status
, ROUTE_ENTRY_REMOVED
))
3399 if (re
->type
== proto
3400 && re
->instance
== instance
) {
3401 rib_delnode(rn
, re
);
3408 /* Remove specific by protocol routes. */
3409 unsigned long rib_score_proto(uint8_t proto
, unsigned short instance
)
3412 struct zebra_vrf
*zvrf
;
3413 struct other_route_table
*ort
;
3414 unsigned long cnt
= 0;
3416 RB_FOREACH (vrf
, vrf_id_head
, &vrfs_by_id
) {
3421 cnt
+= rib_score_proto_table(proto
, instance
,
3422 zvrf
->table
[AFI_IP
][SAFI_UNICAST
])
3423 + rib_score_proto_table(
3425 zvrf
->table
[AFI_IP6
][SAFI_UNICAST
]);
3427 frr_each(otable
, &zvrf
->other_tables
, ort
) cnt
+=
3428 rib_score_proto_table(proto
, instance
, ort
->table
);
3434 /* Close RIB and clean up kernel routes. */
3435 void rib_close_table(struct route_table
*table
)
3437 struct route_node
*rn
;
3438 rib_table_info_t
*info
;
3444 info
= route_table_get_info(table
);
3446 for (rn
= route_top(table
); rn
; rn
= srcdest_route_next(rn
)) {
3447 dest
= rib_dest_from_rnode(rn
);
3449 if (dest
&& dest
->selected_fib
) {
3450 if (info
->safi
== SAFI_UNICAST
)
3451 hook_call(rib_update
, rn
, NULL
);
3453 rib_uninstall_kernel(rn
, dest
->selected_fib
);
3454 dest
->selected_fib
= NULL
;
3460 * Handler for async dataplane results after a pseudowire installation
3462 static int handle_pw_result(struct zebra_dplane_ctx
*ctx
)
3464 struct zebra_pw
*pw
;
3465 struct zebra_vrf
*vrf
;
3467 /* The pseudowire code assumes success - we act on an error
3468 * result for installation attempts here.
3470 if (dplane_ctx_get_op(ctx
) != DPLANE_OP_PW_INSTALL
)
3473 if (dplane_ctx_get_status(ctx
) != ZEBRA_DPLANE_REQUEST_SUCCESS
) {
3474 vrf
= zebra_vrf_lookup_by_id(dplane_ctx_get_vrf(ctx
));
3475 pw
= zebra_pw_find(vrf
, dplane_ctx_get_ifname(ctx
));
3477 zebra_pw_install_failure(pw
);
3487 * Handle results from the dataplane system. Dequeue update context
3488 * structs, dispatch to appropriate internal handlers.
3490 static int rib_process_dplane_results(struct thread
*thread
)
3492 struct zebra_dplane_ctx
*ctx
;
3493 struct dplane_ctx_q ctxlist
;
3494 bool shut_p
= false;
3496 /* Dequeue a list of completed updates with one lock/unlock cycle */
3499 TAILQ_INIT(&ctxlist
);
3501 /* Take lock controlling queue of results */
3502 frr_with_mutex(&dplane_mutex
) {
3503 /* Dequeue list of context structs */
3504 dplane_ctx_list_append(&ctxlist
, &rib_dplane_q
);
3507 /* Dequeue context block */
3508 ctx
= dplane_ctx_dequeue(&ctxlist
);
3510 /* If we've emptied the results queue, we're done */
3514 /* If zebra is shutting down, avoid processing results,
3515 * just drain the results queue.
3517 shut_p
= atomic_load_explicit(&zrouter
.in_shutdown
,
3518 memory_order_relaxed
);
3521 dplane_ctx_fini(&ctx
);
3523 ctx
= dplane_ctx_dequeue(&ctxlist
);
3530 switch (dplane_ctx_get_op(ctx
)) {
3531 case DPLANE_OP_ROUTE_INSTALL
:
3532 case DPLANE_OP_ROUTE_UPDATE
:
3533 case DPLANE_OP_ROUTE_DELETE
:
3535 /* Bit of special case for route updates
3536 * that were generated by async notifications:
3537 * we don't want to continue processing these
3540 if (dplane_ctx_get_notif_provider(ctx
) == 0)
3541 rib_process_result(ctx
);
3543 dplane_ctx_fini(&ctx
);
3547 case DPLANE_OP_ROUTE_NOTIFY
:
3548 rib_process_dplane_notify(ctx
);
3551 case DPLANE_OP_NH_INSTALL
:
3552 case DPLANE_OP_NH_UPDATE
:
3553 case DPLANE_OP_NH_DELETE
:
3554 zebra_nhg_dplane_result(ctx
);
3557 case DPLANE_OP_LSP_INSTALL
:
3558 case DPLANE_OP_LSP_UPDATE
:
3559 case DPLANE_OP_LSP_DELETE
:
3561 /* Bit of special case for LSP updates
3562 * that were generated by async notifications:
3563 * we don't want to continue processing these.
3565 if (dplane_ctx_get_notif_provider(ctx
) == 0)
3566 zebra_mpls_lsp_dplane_result(ctx
);
3568 dplane_ctx_fini(&ctx
);
3572 case DPLANE_OP_LSP_NOTIFY
:
3573 zebra_mpls_process_dplane_notify(ctx
);
3576 case DPLANE_OP_PW_INSTALL
:
3577 case DPLANE_OP_PW_UNINSTALL
:
3578 handle_pw_result(ctx
);
3581 case DPLANE_OP_SYS_ROUTE_ADD
:
3582 case DPLANE_OP_SYS_ROUTE_DELETE
:
3583 /* No further processing in zebra for these. */
3584 dplane_ctx_fini(&ctx
);
3587 case DPLANE_OP_MAC_INSTALL
:
3588 case DPLANE_OP_MAC_DELETE
:
3589 zebra_vxlan_handle_result(ctx
);
3592 /* Some op codes not handled here */
3593 case DPLANE_OP_ADDR_INSTALL
:
3594 case DPLANE_OP_ADDR_UNINSTALL
:
3595 case DPLANE_OP_NEIGH_INSTALL
:
3596 case DPLANE_OP_NEIGH_UPDATE
:
3597 case DPLANE_OP_NEIGH_DELETE
:
3598 case DPLANE_OP_VTEP_ADD
:
3599 case DPLANE_OP_VTEP_DELETE
:
3600 case DPLANE_OP_NONE
:
3601 /* Don't expect this: just return the struct? */
3602 dplane_ctx_fini(&ctx
);
3605 } /* Dispatch by op code */
3607 ctx
= dplane_ctx_dequeue(&ctxlist
);
3616 * Results are returned from the dataplane subsystem, in the context of
3617 * the dataplane pthread. We enqueue the results here for processing by
3618 * the main thread later.
3620 static int rib_dplane_results(struct dplane_ctx_q
*ctxlist
)
3622 /* Take lock controlling queue of results */
3623 frr_with_mutex(&dplane_mutex
) {
3624 /* Enqueue context blocks */
3625 dplane_ctx_list_append(&rib_dplane_q
, ctxlist
);
3628 /* Ensure event is signalled to zebra main pthread */
3629 thread_add_event(zrouter
.master
, rib_process_dplane_results
, NULL
, 0,
3636 * Ensure there are no empty slots in the route_info array.
3637 * Every route type in zebra should be present there.
3639 static void check_route_info(void)
3641 int len
= array_size(route_info
);
3644 * ZEBRA_ROUTE_SYSTEM is special cased since
3645 * its key is 0 anyway.
3647 * ZEBRA_ROUTE_ALL is also ignored.
3649 for (int i
= 0; i
< len
; i
++) {
3650 if (i
== ZEBRA_ROUTE_SYSTEM
|| i
== ZEBRA_ROUTE_ALL
)
3652 assert(route_info
[i
].key
);
3653 assert(route_info
[i
].meta_q_map
< MQ_SIZE
);
3657 /* Routing information base initialize. */
3664 /* Init dataplane, and register for results */
3665 pthread_mutex_init(&dplane_mutex
, NULL
);
3666 TAILQ_INIT(&rib_dplane_q
);
3667 zebra_dplane_init(rib_dplane_results
);
3673 * Get the first vrf id that is greater than the given vrf id if any.
3675 * Returns true if a vrf id was found, false otherwise.
3677 static inline int vrf_id_get_next(vrf_id_t vrf_id
, vrf_id_t
*next_id_p
)
3681 vrf
= vrf_lookup_by_id(vrf_id
);
3683 vrf
= RB_NEXT(vrf_id_head
, vrf
);
3685 *next_id_p
= vrf
->vrf_id
;
3694 * rib_tables_iter_next
3696 * Returns the next table in the iteration.
3698 struct route_table
*rib_tables_iter_next(rib_tables_iter_t
*iter
)
3700 struct route_table
*table
;
3703 * Array that helps us go over all AFI/SAFI combinations via one
3710 {AFI_IP
, SAFI_UNICAST
}, {AFI_IP
, SAFI_MULTICAST
},
3711 {AFI_IP
, SAFI_LABELED_UNICAST
}, {AFI_IP6
, SAFI_UNICAST
},
3712 {AFI_IP6
, SAFI_MULTICAST
}, {AFI_IP6
, SAFI_LABELED_UNICAST
},
3717 switch (iter
->state
) {
3719 case RIB_TABLES_ITER_S_INIT
:
3720 iter
->vrf_id
= VRF_DEFAULT
;
3721 iter
->afi_safi_ix
= -1;
3725 case RIB_TABLES_ITER_S_ITERATING
:
3726 iter
->afi_safi_ix
++;
3729 while (iter
->afi_safi_ix
3730 < (int)array_size(afi_safis
)) {
3731 table
= zebra_vrf_table(
3732 afi_safis
[iter
->afi_safi_ix
].afi
,
3733 afi_safis
[iter
->afi_safi_ix
].safi
,
3738 iter
->afi_safi_ix
++;
3742 * Found another table in this vrf.
3748 * Done with all tables in the current vrf, go to the
3752 if (!vrf_id_get_next(iter
->vrf_id
, &iter
->vrf_id
))
3755 iter
->afi_safi_ix
= 0;
3760 case RIB_TABLES_ITER_S_DONE
:
3765 iter
->state
= RIB_TABLES_ITER_S_ITERATING
;
3767 iter
->state
= RIB_TABLES_ITER_S_DONE
;