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
34 #include "sockunion.h"
35 #include "srcdest_table.h"
39 #include "workqueue.h"
41 #include "zebra/zebra_router.h"
42 #include "zebra/connected.h"
43 #include "zebra/debug.h"
44 #include "zebra/interface.h"
45 #include "zebra/redistribute.h"
46 #include "zebra/rib.h"
48 #include "zebra/zapi_msg.h"
49 #include "zebra/zebra_errors.h"
50 #include "zebra/zebra_memory.h"
51 #include "zebra/zebra_ns.h"
52 #include "zebra/zebra_rnh.h"
53 #include "zebra/zebra_routemap.h"
54 #include "zebra/zebra_vrf.h"
55 #include "zebra/zebra_vxlan.h"
56 #include "zebra/zapi_msg.h"
57 #include "zebra/zebra_dplane.h"
60 * Event, list, and mutex for delivery of dataplane results
62 static pthread_mutex_t dplane_mutex
;
63 static struct thread
*t_dplane
;
64 static struct dplane_ctx_q rib_dplane_q
;
66 DEFINE_HOOK(rib_update
, (struct route_node
* rn
, const char *reason
),
69 /* Should we allow non Quagga processes to delete our routes */
70 extern int allow_delete
;
72 /* Each route type's string and default distance value. */
77 } route_info
[ZEBRA_ROUTE_MAX
] = {
78 [ZEBRA_ROUTE_SYSTEM
] = {ZEBRA_ROUTE_SYSTEM
, 0, 4},
79 [ZEBRA_ROUTE_KERNEL
] = {ZEBRA_ROUTE_KERNEL
, 0, 0},
80 [ZEBRA_ROUTE_CONNECT
] = {ZEBRA_ROUTE_CONNECT
, 0, 0},
81 [ZEBRA_ROUTE_STATIC
] = {ZEBRA_ROUTE_STATIC
, 1, 1},
82 [ZEBRA_ROUTE_RIP
] = {ZEBRA_ROUTE_RIP
, 120, 2},
83 [ZEBRA_ROUTE_RIPNG
] = {ZEBRA_ROUTE_RIPNG
, 120, 2},
84 [ZEBRA_ROUTE_OSPF
] = {ZEBRA_ROUTE_OSPF
, 110, 2},
85 [ZEBRA_ROUTE_OSPF6
] = {ZEBRA_ROUTE_OSPF6
, 110, 2},
86 [ZEBRA_ROUTE_ISIS
] = {ZEBRA_ROUTE_ISIS
, 115, 2},
87 [ZEBRA_ROUTE_BGP
] = {ZEBRA_ROUTE_BGP
, 20 /* IBGP is 200. */, 3},
88 [ZEBRA_ROUTE_PIM
] = {ZEBRA_ROUTE_PIM
, 255, 4},
89 [ZEBRA_ROUTE_EIGRP
] = {ZEBRA_ROUTE_EIGRP
, 90, 2},
90 [ZEBRA_ROUTE_NHRP
] = {ZEBRA_ROUTE_NHRP
, 10, 2},
91 [ZEBRA_ROUTE_HSLS
] = {ZEBRA_ROUTE_HSLS
, 255, 4},
92 [ZEBRA_ROUTE_OLSR
] = {ZEBRA_ROUTE_OLSR
, 255, 4},
93 [ZEBRA_ROUTE_TABLE
] = {ZEBRA_ROUTE_TABLE
, 150, 1},
94 [ZEBRA_ROUTE_LDP
] = {ZEBRA_ROUTE_LDP
, 150, 4},
95 [ZEBRA_ROUTE_VNC
] = {ZEBRA_ROUTE_VNC
, 20, 3},
96 [ZEBRA_ROUTE_VNC_DIRECT
] = {ZEBRA_ROUTE_VNC_DIRECT
, 20, 3},
97 [ZEBRA_ROUTE_VNC_DIRECT_RH
] = {ZEBRA_ROUTE_VNC_DIRECT_RH
, 20, 3},
98 [ZEBRA_ROUTE_BGP_DIRECT
] = {ZEBRA_ROUTE_BGP_DIRECT
, 20, 3},
99 [ZEBRA_ROUTE_BGP_DIRECT_EXT
] = {ZEBRA_ROUTE_BGP_DIRECT_EXT
, 20, 3},
100 [ZEBRA_ROUTE_BABEL
] = {ZEBRA_ROUTE_BABEL
, 100, 2},
101 [ZEBRA_ROUTE_SHARP
] = {ZEBRA_ROUTE_SHARP
, 150, 4},
103 /* no entry/default: 150 */
106 /* RPF lookup behaviour */
107 static enum multicast_mode ipv4_multicast_mode
= MCAST_NO_CONFIG
;
110 static void __attribute__((format(printf
, 5, 6)))
111 _rnode_zlog(const char *_func
, vrf_id_t vrf_id
, struct route_node
*rn
,
112 int priority
, const char *msgfmt
, ...)
114 char buf
[SRCDEST2STR_BUFFER
+ sizeof(" (MRIB)")];
118 va_start(ap
, msgfmt
);
119 vsnprintf(msgbuf
, sizeof(msgbuf
), msgfmt
, ap
);
123 rib_table_info_t
*info
= srcdest_rnode_table_info(rn
);
124 srcdest_rnode2str(rn
, buf
, sizeof(buf
));
126 if (info
->safi
== SAFI_MULTICAST
)
127 strcat(buf
, " (MRIB)");
129 snprintf(buf
, sizeof(buf
), "{(route_node *) NULL}");
132 zlog(priority
, "%s: %d:%s: %s", _func
, vrf_id
, buf
, msgbuf
);
135 #define rnode_debug(node, vrf_id, ...) \
136 _rnode_zlog(__func__, vrf_id, node, LOG_DEBUG, __VA_ARGS__)
137 #define rnode_info(node, ...) \
138 _rnode_zlog(__func__, vrf_id, node, LOG_INFO, __VA_ARGS__)
140 uint8_t route_distance(int type
)
144 if ((unsigned)type
>= array_size(route_info
))
147 distance
= route_info
[type
].distance
;
152 int is_zebra_valid_kernel_table(uint32_t table_id
)
155 if ((table_id
== RT_TABLE_UNSPEC
) || (table_id
== RT_TABLE_LOCAL
)
156 || (table_id
== RT_TABLE_COMPAT
))
163 int is_zebra_main_routing_table(uint32_t table_id
)
165 if ((table_id
== RT_TABLE_MAIN
)
166 || (table_id
== zebrad
.rtm_table_default
))
171 int zebra_check_addr(const struct prefix
*p
)
173 if (p
->family
== AF_INET
) {
176 addr
= p
->u
.prefix4
.s_addr
;
179 if (IPV4_NET127(addr
) || IN_CLASSD(addr
)
180 || IPV4_LINKLOCAL(addr
))
183 if (p
->family
== AF_INET6
) {
184 if (IN6_IS_ADDR_LOOPBACK(&p
->u
.prefix6
))
186 if (IN6_IS_ADDR_LINKLOCAL(&p
->u
.prefix6
))
192 /* Add nexthop to the end of a rib node's nexthop list */
193 void route_entry_nexthop_add(struct route_entry
*re
, struct nexthop
*nexthop
)
195 nexthop_add(&re
->ng
.nexthop
, nexthop
);
201 * copy_nexthop - copy a nexthop to the rib structure.
203 void route_entry_copy_nexthops(struct route_entry
*re
, struct nexthop
*nh
)
205 assert(!re
->ng
.nexthop
);
206 copy_nexthops(&re
->ng
.nexthop
, nh
, NULL
);
207 for (struct nexthop
*nexthop
= nh
; nexthop
; nexthop
= nexthop
->next
)
211 /* Delete specified nexthop from the list. */
212 void route_entry_nexthop_delete(struct route_entry
*re
, struct nexthop
*nexthop
)
215 nexthop
->next
->prev
= nexthop
->prev
;
217 nexthop
->prev
->next
= nexthop
->next
;
219 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 || CHECK_FLAG(re
->flags
, ZEBRA_FLAG_EVPN_ROUTE
)) {
281 SET_FLAG(nexthop
->flags
, NEXTHOP_FLAG_ONLINK
);
284 route_entry_nexthop_add(re
, nexthop
);
289 struct nexthop
*route_entry_nexthop_ipv6_add(struct route_entry
*re
,
290 struct in6_addr
*ipv6
,
293 struct nexthop
*nexthop
;
295 nexthop
= nexthop_new();
296 nexthop
->vrf_id
= nh_vrf_id
;
297 nexthop
->type
= NEXTHOP_TYPE_IPV6
;
298 nexthop
->gate
.ipv6
= *ipv6
;
300 route_entry_nexthop_add(re
, nexthop
);
305 struct nexthop
*route_entry_nexthop_ipv6_ifindex_add(struct route_entry
*re
,
306 struct in6_addr
*ipv6
,
310 struct nexthop
*nexthop
;
312 nexthop
= nexthop_new();
313 nexthop
->vrf_id
= nh_vrf_id
;
314 nexthop
->type
= NEXTHOP_TYPE_IPV6_IFINDEX
;
315 nexthop
->gate
.ipv6
= *ipv6
;
316 nexthop
->ifindex
= ifindex
;
317 if (CHECK_FLAG(re
->flags
, ZEBRA_FLAG_EVPN_ROUTE
))
318 SET_FLAG(nexthop
->flags
, NEXTHOP_FLAG_ONLINK
);
320 route_entry_nexthop_add(re
, nexthop
);
325 struct nexthop
*route_entry_nexthop_blackhole_add(struct route_entry
*re
,
326 enum blackhole_type bh_type
)
328 struct nexthop
*nexthop
;
330 nexthop
= nexthop_new();
331 nexthop
->vrf_id
= VRF_DEFAULT
;
332 nexthop
->type
= NEXTHOP_TYPE_BLACKHOLE
;
333 nexthop
->bh_type
= bh_type
;
335 route_entry_nexthop_add(re
, nexthop
);
340 static void nexthop_set_resolved(afi_t afi
, const struct nexthop
*newhop
,
341 struct nexthop
*nexthop
)
343 struct nexthop
*resolved_hop
;
345 resolved_hop
= nexthop_new();
346 SET_FLAG(resolved_hop
->flags
, NEXTHOP_FLAG_ACTIVE
);
348 resolved_hop
->vrf_id
= nexthop
->vrf_id
;
349 switch (newhop
->type
) {
350 case NEXTHOP_TYPE_IPV4
:
351 case NEXTHOP_TYPE_IPV4_IFINDEX
:
352 /* If the resolving route specifies a gateway, use it */
353 resolved_hop
->type
= newhop
->type
;
354 resolved_hop
->gate
.ipv4
= newhop
->gate
.ipv4
;
356 if (newhop
->ifindex
) {
357 resolved_hop
->type
= NEXTHOP_TYPE_IPV4_IFINDEX
;
358 resolved_hop
->ifindex
= newhop
->ifindex
;
361 case NEXTHOP_TYPE_IPV6
:
362 case NEXTHOP_TYPE_IPV6_IFINDEX
:
363 resolved_hop
->type
= newhop
->type
;
364 resolved_hop
->gate
.ipv6
= newhop
->gate
.ipv6
;
366 if (newhop
->ifindex
) {
367 resolved_hop
->type
= NEXTHOP_TYPE_IPV6_IFINDEX
;
368 resolved_hop
->ifindex
= newhop
->ifindex
;
371 case NEXTHOP_TYPE_IFINDEX
:
372 /* If the resolving route is an interface route,
373 * it means the gateway we are looking up is connected
374 * to that interface. (The actual network is _not_ onlink).
375 * Therefore, the resolved route should have the original
376 * gateway as nexthop as it is directly connected.
378 * On Linux, we have to set the onlink netlink flag because
379 * otherwise, the kernel won't accept the route.
381 resolved_hop
->flags
|= NEXTHOP_FLAG_ONLINK
;
383 resolved_hop
->type
= NEXTHOP_TYPE_IPV4_IFINDEX
;
384 resolved_hop
->gate
.ipv4
= nexthop
->gate
.ipv4
;
385 } else if (afi
== AFI_IP6
) {
386 resolved_hop
->type
= NEXTHOP_TYPE_IPV6_IFINDEX
;
387 resolved_hop
->gate
.ipv6
= nexthop
->gate
.ipv6
;
389 resolved_hop
->ifindex
= newhop
->ifindex
;
391 case NEXTHOP_TYPE_BLACKHOLE
:
392 resolved_hop
->type
= NEXTHOP_TYPE_BLACKHOLE
;
393 resolved_hop
->bh_type
= nexthop
->bh_type
;
397 if (newhop
->flags
& NEXTHOP_FLAG_ONLINK
)
398 resolved_hop
->flags
|= NEXTHOP_FLAG_ONLINK
;
400 /* Copy labels of the resolved route */
401 if (newhop
->nh_label
)
402 nexthop_add_labels(resolved_hop
, newhop
->nh_label_type
,
403 newhop
->nh_label
->num_labels
,
404 &newhop
->nh_label
->label
[0]);
406 resolved_hop
->rparent
= nexthop
;
407 nexthop_add(&nexthop
->resolved
, resolved_hop
);
410 /* If force flag is not set, do not modify falgs at all for uninstall
411 the route from FIB. */
412 static int nexthop_active(afi_t afi
, struct route_entry
*re
,
413 struct nexthop
*nexthop
, bool set
,
414 struct route_node
*top
)
417 struct route_table
*table
;
418 struct route_node
*rn
;
419 struct route_entry
*match
= NULL
;
421 struct nexthop
*newhop
;
422 struct interface
*ifp
;
425 if ((nexthop
->type
== NEXTHOP_TYPE_IPV4
)
426 || nexthop
->type
== NEXTHOP_TYPE_IPV6
)
427 nexthop
->ifindex
= 0;
430 UNSET_FLAG(nexthop
->flags
, NEXTHOP_FLAG_RECURSIVE
);
431 nexthops_free(nexthop
->resolved
);
432 nexthop
->resolved
= NULL
;
436 /* Next hops (remote VTEPs) for EVPN routes are fully resolved. */
437 if (CHECK_FLAG(nexthop
->flags
, NEXTHOP_FLAG_EVPN_RVTEP
))
441 * If the kernel has sent us a route, then
442 * by golly gee whiz it's a good route.
444 if (re
->type
== ZEBRA_ROUTE_KERNEL
||
445 re
->type
== ZEBRA_ROUTE_SYSTEM
)
448 /* Skip nexthops that have been filtered out due to route-map */
449 /* The nexthops are specific to this route and so the same */
450 /* nexthop for a different route may not have this flag set */
451 if (CHECK_FLAG(nexthop
->flags
, NEXTHOP_FLAG_FILTERED
)) {
452 if (IS_ZEBRA_DEBUG_RIB_DETAILED
)
453 zlog_debug("\t%s: Nexthop Filtered",
454 __PRETTY_FUNCTION__
);
459 * Check to see if we should trust the passed in information
460 * for UNNUMBERED interfaces as that we won't find the GW
461 * address in the routing table.
463 if (CHECK_FLAG(nexthop
->flags
, NEXTHOP_FLAG_ONLINK
)) {
464 ifp
= if_lookup_by_index(nexthop
->ifindex
, nexthop
->vrf_id
);
466 if (IS_ZEBRA_DEBUG_RIB_DETAILED
)
468 "\t%s: Onlink and interface: %u[%u] does not exist",
469 __PRETTY_FUNCTION__
, nexthop
->ifindex
,
473 if (connected_is_unnumbered(ifp
)) {
474 if (if_is_operative(ifp
))
477 if (IS_ZEBRA_DEBUG_RIB_DETAILED
)
479 "\t%s: Onlink and interface %s is not operative",
480 __PRETTY_FUNCTION__
, ifp
->name
);
484 if (!if_is_operative(ifp
)) {
485 if (IS_ZEBRA_DEBUG_RIB_DETAILED
)
487 "\t%s: Interface %s is not unnumbered",
489 ifp
? ifp
->name
: "Unknown");
494 /* Make lookup prefix. */
495 memset(&p
, 0, sizeof(struct prefix
));
499 p
.prefixlen
= IPV4_MAX_PREFIXLEN
;
500 p
.u
.prefix4
= nexthop
->gate
.ipv4
;
504 p
.prefixlen
= IPV6_MAX_PREFIXLEN
;
505 p
.u
.prefix6
= nexthop
->gate
.ipv6
;
508 assert(afi
!= AFI_IP
&& afi
!= AFI_IP6
);
512 table
= zebra_vrf_table(afi
, SAFI_UNICAST
, nexthop
->vrf_id
);
514 if (IS_ZEBRA_DEBUG_RIB_DETAILED
)
515 zlog_debug("\t%s: Table not found",
516 __PRETTY_FUNCTION__
);
520 rn
= route_node_match(table
, (struct prefix
*)&p
);
522 route_unlock_node(rn
);
524 /* Lookup should halt if we've matched against ourselves ('top',
525 * if specified) - i.e., we cannot have a nexthop NH1 is
526 * resolved by a route NH1. The exception is if the route is a
529 if (top
&& rn
== top
)
530 if (((afi
== AFI_IP
) && (rn
->p
.prefixlen
!= 32))
531 || ((afi
== AFI_IP6
) && (rn
->p
.prefixlen
!= 128))) {
532 if (IS_ZEBRA_DEBUG_RIB_DETAILED
)
534 "\t%s: Matched against ourself and prefix length is not max bit length",
535 __PRETTY_FUNCTION__
);
539 /* Pick up selected route. */
540 /* However, do not resolve over default route unless explicitly
542 if (is_default_prefix(&rn
->p
)
543 && !rnh_resolve_via_default(p
.family
)) {
544 if (IS_ZEBRA_DEBUG_RIB_DETAILED
)
546 "\t:%s: Resolved against default route",
547 __PRETTY_FUNCTION__
);
551 dest
= rib_dest_from_rnode(rn
);
552 if (dest
&& dest
->selected_fib
553 && !CHECK_FLAG(dest
->selected_fib
->status
,
555 && dest
->selected_fib
->type
!= ZEBRA_ROUTE_TABLE
)
556 match
= dest
->selected_fib
;
558 /* If there is no selected route or matched route is EGP, go up
563 } while (rn
&& rn
->info
== NULL
);
570 if (match
->type
== ZEBRA_ROUTE_CONNECT
) {
571 /* Directly point connected route. */
572 newhop
= match
->ng
.nexthop
;
574 if (nexthop
->type
== NEXTHOP_TYPE_IPV4
575 || nexthop
->type
== NEXTHOP_TYPE_IPV6
)
576 nexthop
->ifindex
= newhop
->ifindex
;
579 } else if (CHECK_FLAG(re
->flags
, ZEBRA_FLAG_ALLOW_RECURSION
)) {
581 for (ALL_NEXTHOPS(match
->ng
, newhop
)) {
582 if (!CHECK_FLAG(match
->status
,
583 ROUTE_ENTRY_INSTALLED
))
585 if (CHECK_FLAG(newhop
->flags
,
586 NEXTHOP_FLAG_RECURSIVE
))
590 SET_FLAG(nexthop
->flags
,
591 NEXTHOP_FLAG_RECURSIVE
);
593 ROUTE_ENTRY_NEXTHOPS_CHANGED
);
594 nexthop_set_resolved(afi
, newhop
,
600 re
->nexthop_mtu
= match
->mtu
;
601 if (!resolved
&& IS_ZEBRA_DEBUG_RIB_DETAILED
)
602 zlog_debug("\t%s: Recursion failed to find",
603 __PRETTY_FUNCTION__
);
605 } else if (re
->type
== ZEBRA_ROUTE_STATIC
) {
607 for (ALL_NEXTHOPS(match
->ng
, newhop
)) {
608 if (!CHECK_FLAG(match
->status
,
609 ROUTE_ENTRY_INSTALLED
))
613 SET_FLAG(nexthop
->flags
,
614 NEXTHOP_FLAG_RECURSIVE
);
615 nexthop_set_resolved(afi
, newhop
,
621 re
->nexthop_mtu
= match
->mtu
;
623 if (!resolved
&& IS_ZEBRA_DEBUG_RIB_DETAILED
)
625 "\t%s: Static route unable to resolve",
626 __PRETTY_FUNCTION__
);
629 if (IS_ZEBRA_DEBUG_RIB_DETAILED
) {
630 zlog_debug("\t%s: Route Type %s has not turned on recursion",
632 zebra_route_string(re
->type
));
633 if (re
->type
== ZEBRA_ROUTE_BGP
&&
634 !CHECK_FLAG(re
->flags
, ZEBRA_FLAG_IBGP
))
635 zlog_debug("\tEBGP: see \"disable-ebgp-connected-route-check\" or \"disable-connected-check\"");
640 if (IS_ZEBRA_DEBUG_RIB_DETAILED
)
641 zlog_debug("\t%s: Nexthop did not lookup in table",
642 __PRETTY_FUNCTION__
);
646 struct route_entry
*rib_match(afi_t afi
, safi_t safi
, vrf_id_t vrf_id
,
647 union g_addr
*addr
, struct route_node
**rn_out
)
650 struct route_table
*table
;
651 struct route_node
*rn
;
652 struct route_entry
*match
= NULL
;
655 table
= zebra_vrf_table(afi
, safi
, vrf_id
);
659 memset(&p
, 0, sizeof(struct prefix
));
662 p
.u
.prefix4
= addr
->ipv4
;
663 p
.prefixlen
= IPV4_MAX_PREFIXLEN
;
665 p
.u
.prefix6
= addr
->ipv6
;
666 p
.prefixlen
= IPV6_MAX_PREFIXLEN
;
669 rn
= route_node_match(table
, (struct prefix
*)&p
);
674 route_unlock_node(rn
);
676 dest
= rib_dest_from_rnode(rn
);
677 if (dest
&& dest
->selected_fib
678 && !CHECK_FLAG(dest
->selected_fib
->status
,
679 ROUTE_ENTRY_REMOVED
))
680 match
= dest
->selected_fib
;
682 /* If there is no selected route or matched route is EGP, go up
687 } while (rn
&& rn
->info
== NULL
);
691 if (match
->type
!= ZEBRA_ROUTE_CONNECT
) {
692 if (!CHECK_FLAG(match
->status
,
693 ROUTE_ENTRY_INSTALLED
))
705 struct route_entry
*rib_match_ipv4_multicast(vrf_id_t vrf_id
,
707 struct route_node
**rn_out
)
709 struct route_entry
*re
= NULL
, *mre
= NULL
, *ure
= NULL
;
710 struct route_node
*m_rn
= NULL
, *u_rn
= NULL
;
711 union g_addr gaddr
= {.ipv4
= addr
};
713 switch (ipv4_multicast_mode
) {
714 case MCAST_MRIB_ONLY
:
715 return rib_match(AFI_IP
, SAFI_MULTICAST
, vrf_id
, &gaddr
,
717 case MCAST_URIB_ONLY
:
718 return rib_match(AFI_IP
, SAFI_UNICAST
, vrf_id
, &gaddr
, rn_out
);
719 case MCAST_NO_CONFIG
:
720 case MCAST_MIX_MRIB_FIRST
:
721 re
= mre
= rib_match(AFI_IP
, SAFI_MULTICAST
, vrf_id
, &gaddr
,
724 re
= ure
= rib_match(AFI_IP
, SAFI_UNICAST
, vrf_id
,
727 case MCAST_MIX_DISTANCE
:
728 mre
= rib_match(AFI_IP
, SAFI_MULTICAST
, vrf_id
, &gaddr
, &m_rn
);
729 ure
= rib_match(AFI_IP
, SAFI_UNICAST
, vrf_id
, &gaddr
, &u_rn
);
731 re
= ure
->distance
< mre
->distance
? ure
: mre
;
737 case MCAST_MIX_PFXLEN
:
738 mre
= rib_match(AFI_IP
, SAFI_MULTICAST
, vrf_id
, &gaddr
, &m_rn
);
739 ure
= rib_match(AFI_IP
, SAFI_UNICAST
, vrf_id
, &gaddr
, &u_rn
);
741 re
= u_rn
->p
.prefixlen
> m_rn
->p
.prefixlen
? ure
: mre
;
750 *rn_out
= (re
== mre
) ? m_rn
: u_rn
;
752 if (IS_ZEBRA_DEBUG_RIB
) {
754 inet_ntop(AF_INET
, &addr
, buf
, BUFSIZ
);
756 zlog_debug("%s: %s: vrf: %u found %s, using %s",
757 __func__
, buf
, vrf_id
,
758 mre
? (ure
? "MRIB+URIB" : "MRIB")
759 : ure
? "URIB" : "nothing",
760 re
== ure
? "URIB" : re
== mre
? "MRIB" : "none");
765 void multicast_mode_ipv4_set(enum multicast_mode mode
)
767 if (IS_ZEBRA_DEBUG_RIB
)
768 zlog_debug("%s: multicast lookup mode set (%d)", __func__
,
770 ipv4_multicast_mode
= mode
;
773 enum multicast_mode
multicast_mode_ipv4_get(void)
775 return ipv4_multicast_mode
;
778 struct route_entry
*rib_lookup_ipv4(struct prefix_ipv4
*p
, vrf_id_t vrf_id
)
780 struct route_table
*table
;
781 struct route_node
*rn
;
782 struct route_entry
*match
= NULL
;
786 table
= zebra_vrf_table(AFI_IP
, SAFI_UNICAST
, vrf_id
);
790 rn
= route_node_lookup(table
, (struct prefix
*)p
);
792 /* No route for this prefix. */
797 route_unlock_node(rn
);
798 dest
= rib_dest_from_rnode(rn
);
800 if (dest
&& dest
->selected_fib
801 && !CHECK_FLAG(dest
->selected_fib
->status
, ROUTE_ENTRY_REMOVED
))
802 match
= dest
->selected_fib
;
807 if (match
->type
== ZEBRA_ROUTE_CONNECT
)
810 if (CHECK_FLAG(match
->status
, ROUTE_ENTRY_INSTALLED
))
816 #define RIB_SYSTEM_ROUTE(R) \
817 ((R)->type == ZEBRA_ROUTE_KERNEL || (R)->type == ZEBRA_ROUTE_CONNECT)
819 #define RIB_KERNEL_ROUTE(R) \
820 ((R)->type == ZEBRA_ROUTE_KERNEL)
822 /* This function verifies reachability of one given nexthop, which can be
823 * numbered or unnumbered, IPv4 or IPv6. The result is unconditionally stored
824 * in nexthop->flags field. If the 4th parameter, 'set', is non-zero,
825 * nexthop->ifindex will be updated appropriately as well.
826 * An existing route map can turn (otherwise active) nexthop into inactive, but
829 * The return value is the final value of 'ACTIVE' flag.
832 static unsigned nexthop_active_check(struct route_node
*rn
,
833 struct route_entry
*re
,
834 struct nexthop
*nexthop
, bool set
)
836 struct interface
*ifp
;
837 route_map_result_t ret
= RMAP_MATCH
;
839 char buf
[SRCDEST2STR_BUFFER
];
840 const struct prefix
*p
, *src_p
;
841 struct zebra_vrf
*zvrf
;
843 srcdest_rnode_prefixes(rn
, &p
, &src_p
);
845 if (rn
->p
.family
== AF_INET
)
847 else if (rn
->p
.family
== AF_INET6
)
851 switch (nexthop
->type
) {
852 case NEXTHOP_TYPE_IFINDEX
:
853 ifp
= if_lookup_by_index(nexthop
->ifindex
, nexthop
->vrf_id
);
854 if (ifp
&& if_is_operative(ifp
))
855 SET_FLAG(nexthop
->flags
, NEXTHOP_FLAG_ACTIVE
);
857 UNSET_FLAG(nexthop
->flags
, NEXTHOP_FLAG_ACTIVE
);
859 case NEXTHOP_TYPE_IPV4
:
860 case NEXTHOP_TYPE_IPV4_IFINDEX
:
862 if (nexthop_active(AFI_IP
, re
, nexthop
, set
, rn
))
863 SET_FLAG(nexthop
->flags
, NEXTHOP_FLAG_ACTIVE
);
865 UNSET_FLAG(nexthop
->flags
, NEXTHOP_FLAG_ACTIVE
);
867 case NEXTHOP_TYPE_IPV6
:
869 if (nexthop_active(AFI_IP6
, re
, nexthop
, set
, rn
))
870 SET_FLAG(nexthop
->flags
, NEXTHOP_FLAG_ACTIVE
);
872 UNSET_FLAG(nexthop
->flags
, NEXTHOP_FLAG_ACTIVE
);
874 case NEXTHOP_TYPE_IPV6_IFINDEX
:
875 /* RFC 5549, v4 prefix with v6 NH */
876 if (rn
->p
.family
!= AF_INET
)
878 if (IN6_IS_ADDR_LINKLOCAL(&nexthop
->gate
.ipv6
)) {
879 ifp
= if_lookup_by_index(nexthop
->ifindex
,
881 if (ifp
&& if_is_operative(ifp
))
882 SET_FLAG(nexthop
->flags
, NEXTHOP_FLAG_ACTIVE
);
884 UNSET_FLAG(nexthop
->flags
, NEXTHOP_FLAG_ACTIVE
);
886 if (nexthop_active(AFI_IP6
, re
, nexthop
, set
, rn
))
887 SET_FLAG(nexthop
->flags
, NEXTHOP_FLAG_ACTIVE
);
889 UNSET_FLAG(nexthop
->flags
, NEXTHOP_FLAG_ACTIVE
);
892 case NEXTHOP_TYPE_BLACKHOLE
:
893 SET_FLAG(nexthop
->flags
, NEXTHOP_FLAG_ACTIVE
);
898 if (!CHECK_FLAG(nexthop
->flags
, NEXTHOP_FLAG_ACTIVE
)) {
899 if (IS_ZEBRA_DEBUG_RIB_DETAILED
)
900 zlog_debug("\t%s: Unable to find a active nexthop",
901 __PRETTY_FUNCTION__
);
905 /* XXX: What exactly do those checks do? Do we support
906 * e.g. IPv4 routes with IPv6 nexthops or vice versa?
908 if (RIB_SYSTEM_ROUTE(re
) || (family
== AFI_IP
&& p
->family
!= AF_INET
)
909 || (family
== AFI_IP6
&& p
->family
!= AF_INET6
))
910 return CHECK_FLAG(nexthop
->flags
, NEXTHOP_FLAG_ACTIVE
);
912 /* The original code didn't determine the family correctly
913 * e.g. for NEXTHOP_TYPE_IFINDEX. Retrieve the correct afi
914 * from the rib_table_info in those cases.
915 * Possibly it may be better to use only the rib_table_info
919 rib_table_info_t
*info
;
921 info
= srcdest_rnode_table_info(rn
);
925 memset(&nexthop
->rmap_src
.ipv6
, 0, sizeof(union g_addr
));
927 zvrf
= zebra_vrf_lookup_by_id(nexthop
->vrf_id
);
929 if (IS_ZEBRA_DEBUG_RIB_DETAILED
)
930 zlog_debug("\t%s: zvrf is NULL", __PRETTY_FUNCTION__
);
931 return CHECK_FLAG(nexthop
->flags
, NEXTHOP_FLAG_ACTIVE
);
934 /* It'll get set if required inside */
935 ret
= zebra_route_map_check(family
, re
->type
, re
->instance
, p
,
936 nexthop
, zvrf
, re
->tag
);
937 if (ret
== RMAP_DENYMATCH
) {
938 if (IS_ZEBRA_DEBUG_RIB
) {
939 srcdest_rnode2str(rn
, buf
, sizeof(buf
));
941 "%u:%s: Filtering out with NH out %s due to route map",
943 ifindex2ifname(nexthop
->ifindex
,
946 UNSET_FLAG(nexthop
->flags
, NEXTHOP_FLAG_ACTIVE
);
948 return CHECK_FLAG(nexthop
->flags
, NEXTHOP_FLAG_ACTIVE
);
951 /* Iterate over all nexthops of the given RIB entry and refresh their
952 * ACTIVE flag. re->nexthop_active_num is updated accordingly. If any
953 * nexthop is found to toggle the ACTIVE flag, the whole re structure
954 * is flagged with ROUTE_ENTRY_CHANGED. The 4th 'set' argument is
955 * transparently passed to nexthop_active_check().
957 * Return value is the new number of active nexthops.
960 static int nexthop_active_update(struct route_node
*rn
, struct route_entry
*re
,
963 struct nexthop
*nexthop
;
964 union g_addr prev_src
;
965 unsigned int prev_active
, new_active
, old_num_nh
;
966 ifindex_t prev_index
;
968 old_num_nh
= re
->nexthop_active_num
;
970 re
->nexthop_active_num
= 0;
971 UNSET_FLAG(re
->status
, ROUTE_ENTRY_CHANGED
);
973 for (nexthop
= re
->ng
.nexthop
; nexthop
; nexthop
= nexthop
->next
) {
974 /* No protocol daemon provides src and so we're skipping
976 prev_src
= nexthop
->rmap_src
;
977 prev_active
= CHECK_FLAG(nexthop
->flags
, NEXTHOP_FLAG_ACTIVE
);
978 prev_index
= nexthop
->ifindex
;
980 * We need to respect the multipath_num here
981 * as that what we should be able to install from
982 * a multipath perpsective should not be a data plane
985 new_active
= nexthop_active_check(rn
, re
, nexthop
, set
);
986 if (new_active
&& re
->nexthop_active_num
>= multipath_num
) {
987 UNSET_FLAG(nexthop
->flags
, NEXTHOP_FLAG_ACTIVE
);
991 re
->nexthop_active_num
++;
992 /* Don't allow src setting on IPv6 addr for now */
993 if (prev_active
!= new_active
|| prev_index
!= nexthop
->ifindex
994 || ((nexthop
->type
>= NEXTHOP_TYPE_IFINDEX
995 && nexthop
->type
< NEXTHOP_TYPE_IPV6
)
996 && prev_src
.ipv4
.s_addr
997 != nexthop
->rmap_src
.ipv4
.s_addr
)
998 || ((nexthop
->type
>= NEXTHOP_TYPE_IPV6
999 && nexthop
->type
< NEXTHOP_TYPE_BLACKHOLE
)
1000 && !(IPV6_ADDR_SAME(&prev_src
.ipv6
,
1001 &nexthop
->rmap_src
.ipv6
)))) {
1002 SET_FLAG(re
->status
, ROUTE_ENTRY_CHANGED
);
1003 SET_FLAG(re
->status
, ROUTE_ENTRY_NEXTHOPS_CHANGED
);
1007 if (old_num_nh
!= re
->nexthop_active_num
)
1008 SET_FLAG(re
->status
, ROUTE_ENTRY_CHANGED
);
1010 if (CHECK_FLAG(re
->status
, ROUTE_ENTRY_CHANGED
)) {
1011 SET_FLAG(re
->status
, ROUTE_ENTRY_NEXTHOPS_CHANGED
);
1014 return re
->nexthop_active_num
;
1018 * Is this RIB labeled-unicast? It must be of type BGP and all paths
1019 * (nexthops) must have a label.
1021 int zebra_rib_labeled_unicast(struct route_entry
*re
)
1023 struct nexthop
*nexthop
= NULL
;
1025 if (re
->type
!= ZEBRA_ROUTE_BGP
)
1028 for (ALL_NEXTHOPS(re
->ng
, nexthop
))
1029 if (!nexthop
->nh_label
|| !nexthop
->nh_label
->num_labels
)
1035 /* Update flag indicates whether this is a "replace" or not. Currently, this
1036 * is only used for IPv4.
1038 void rib_install_kernel(struct route_node
*rn
, struct route_entry
*re
,
1039 struct route_entry
*old
)
1041 struct nexthop
*nexthop
;
1042 rib_table_info_t
*info
= srcdest_rnode_table_info(rn
);
1043 struct zebra_vrf
*zvrf
= vrf_info_lookup(re
->vrf_id
);
1044 const struct prefix
*p
, *src_p
;
1045 enum zebra_dplane_result ret
;
1047 rib_dest_t
*dest
= rib_dest_from_rnode(rn
);
1049 srcdest_rnode_prefixes(rn
, &p
, &src_p
);
1051 if (info
->safi
!= SAFI_UNICAST
) {
1052 for (ALL_NEXTHOPS(re
->ng
, nexthop
))
1053 SET_FLAG(nexthop
->flags
, NEXTHOP_FLAG_FIB
);
1056 struct nexthop
*prev
;
1058 for (ALL_NEXTHOPS(re
->ng
, nexthop
)) {
1059 UNSET_FLAG (nexthop
->flags
, NEXTHOP_FLAG_DUPLICATE
);
1060 for (ALL_NEXTHOPS(re
->ng
, prev
)) {
1061 if (prev
== nexthop
)
1063 if (nexthop_same_firsthop(nexthop
, prev
)) {
1064 SET_FLAG(nexthop
->flags
,
1065 NEXTHOP_FLAG_DUPLICATE
);
1073 * If this is a replace to a new RE let the originator of the RE
1074 * know that they've lost
1076 if (old
&& (old
!= re
) && (old
->type
!= re
->type
))
1077 zsend_route_notify_owner(old
, p
, ZAPI_ROUTE_BETTER_ADMIN_WON
);
1079 /* Update fib selection */
1080 dest
->selected_fib
= re
;
1083 * Make sure we update the FPM any time we send new information to
1086 hook_call(rib_update
, rn
, "installing in kernel");
1088 /* Send add or update */
1089 if (old
&& (old
!= re
))
1090 ret
= dplane_route_update(rn
, re
, old
);
1092 ret
= dplane_route_add(rn
, re
);
1095 case ZEBRA_DPLANE_REQUEST_QUEUED
:
1096 SET_FLAG(re
->status
, ROUTE_ENTRY_QUEUED
);
1098 SET_FLAG(old
->status
, ROUTE_ENTRY_QUEUED
);
1100 zvrf
->installs_queued
++;
1102 case ZEBRA_DPLANE_REQUEST_FAILURE
:
1104 char str
[SRCDEST2STR_BUFFER
];
1106 srcdest_rnode2str(rn
, str
, sizeof(str
));
1107 flog_err(EC_ZEBRA_DP_INSTALL_FAIL
,
1108 "%u:%s: Failed to enqueue dataplane install",
1112 case ZEBRA_DPLANE_REQUEST_SUCCESS
:
1121 /* Uninstall the route from kernel. */
1122 void rib_uninstall_kernel(struct route_node
*rn
, struct route_entry
*re
)
1124 struct nexthop
*nexthop
;
1125 rib_table_info_t
*info
= srcdest_rnode_table_info(rn
);
1126 struct zebra_vrf
*zvrf
= vrf_info_lookup(re
->vrf_id
);
1128 if (info
->safi
!= SAFI_UNICAST
) {
1129 UNSET_FLAG(re
->status
, ROUTE_ENTRY_INSTALLED
);
1130 for (ALL_NEXTHOPS(re
->ng
, nexthop
))
1131 UNSET_FLAG(nexthop
->flags
, NEXTHOP_FLAG_FIB
);
1136 * Make sure we update the FPM any time we send new information to
1139 hook_call(rib_update
, rn
, "uninstalling from kernel");
1141 switch (dplane_route_delete(rn
, re
)) {
1142 case ZEBRA_DPLANE_REQUEST_QUEUED
:
1144 zvrf
->removals_queued
++;
1146 case ZEBRA_DPLANE_REQUEST_FAILURE
:
1148 char str
[SRCDEST2STR_BUFFER
];
1150 srcdest_rnode2str(rn
, str
, sizeof(str
));
1151 flog_err(EC_ZEBRA_DP_INSTALL_FAIL
,
1152 "%u:%s: Failed to enqueue dataplane uninstall",
1156 case ZEBRA_DPLANE_REQUEST_SUCCESS
:
1165 /* Uninstall the route from kernel. */
1166 static void rib_uninstall(struct route_node
*rn
, struct route_entry
*re
)
1168 rib_table_info_t
*info
= srcdest_rnode_table_info(rn
);
1169 rib_dest_t
*dest
= rib_dest_from_rnode(rn
);
1170 struct nexthop
*nexthop
;
1172 if (dest
&& dest
->selected_fib
== re
) {
1173 if (info
->safi
== SAFI_UNICAST
)
1174 hook_call(rib_update
, rn
, "rib_uninstall");
1176 /* If labeled-unicast route, uninstall transit LSP. */
1177 if (zebra_rib_labeled_unicast(re
))
1178 zebra_mpls_lsp_uninstall(info
->zvrf
, rn
, re
);
1180 if (!RIB_SYSTEM_ROUTE(re
))
1181 rib_uninstall_kernel(rn
, re
);
1183 UNSET_FLAG(re
->status
, ROUTE_ENTRY_INSTALLED
);
1185 dest
->selected_fib
= NULL
;
1187 for (ALL_NEXTHOPS(re
->ng
, nexthop
))
1188 UNSET_FLAG(nexthop
->flags
, NEXTHOP_FLAG_FIB
);
1191 if (CHECK_FLAG(re
->flags
, ZEBRA_FLAG_SELECTED
)) {
1192 const struct prefix
*p
, *src_p
;
1194 srcdest_rnode_prefixes(rn
, &p
, &src_p
);
1196 redistribute_delete(p
, src_p
, re
);
1197 UNSET_FLAG(re
->flags
, ZEBRA_FLAG_SELECTED
);
1202 * rib_can_delete_dest
1204 * Returns TRUE if the given dest can be deleted from the table.
1206 static int rib_can_delete_dest(rib_dest_t
*dest
)
1213 * Don't delete the dest if we have to update the FPM about this
1216 if (CHECK_FLAG(dest
->flags
, RIB_DEST_UPDATE_FPM
)
1217 || CHECK_FLAG(dest
->flags
, RIB_DEST_SENT_TO_FPM
))
1226 * Garbage collect the rib dest corresponding to the given route node
1229 * Returns TRUE if the dest was deleted, FALSE otherwise.
1231 int rib_gc_dest(struct route_node
*rn
)
1235 dest
= rib_dest_from_rnode(rn
);
1239 if (!rib_can_delete_dest(dest
))
1242 if (IS_ZEBRA_DEBUG_RIB
) {
1243 struct zebra_vrf
*zvrf
;
1245 zvrf
= rib_dest_vrf(dest
);
1246 rnode_debug(rn
, zvrf_id(zvrf
), "removing dest from table");
1250 XFREE(MTYPE_RIB_DEST
, dest
);
1254 * Release the one reference that we keep on the route node.
1256 route_unlock_node(rn
);
1260 static void rib_process_add_fib(struct zebra_vrf
*zvrf
, struct route_node
*rn
,
1261 struct route_entry
*new)
1263 rib_dest_t
*dest
= rib_dest_from_rnode(rn
);
1265 hook_call(rib_update
, rn
, "new route selected");
1267 /* Update real nexthop. This may actually determine if nexthop is active
1269 if (!nexthop_active_update(rn
, new, true)) {
1270 UNSET_FLAG(new->status
, ROUTE_ENTRY_CHANGED
);
1274 if (IS_ZEBRA_DEBUG_RIB
) {
1275 char buf
[SRCDEST2STR_BUFFER
];
1276 srcdest_rnode2str(rn
, buf
, sizeof(buf
));
1277 zlog_debug("%u:%s: Adding route rn %p, re %p (%s)",
1278 zvrf_id(zvrf
), buf
, rn
, new,
1279 zebra_route_string(new->type
));
1282 /* If labeled-unicast route, install transit LSP. */
1283 if (zebra_rib_labeled_unicast(new))
1284 zebra_mpls_lsp_install(zvrf
, rn
, new);
1286 if (!RIB_SYSTEM_ROUTE(new))
1287 rib_install_kernel(rn
, new, NULL
);
1289 dest
->selected_fib
= new;
1291 UNSET_FLAG(new->status
, ROUTE_ENTRY_CHANGED
);
1294 static void rib_process_del_fib(struct zebra_vrf
*zvrf
, struct route_node
*rn
,
1295 struct route_entry
*old
)
1297 rib_dest_t
*dest
= rib_dest_from_rnode(rn
);
1298 hook_call(rib_update
, rn
, "removing existing route");
1300 /* Uninstall from kernel. */
1301 if (IS_ZEBRA_DEBUG_RIB
) {
1302 char buf
[SRCDEST2STR_BUFFER
];
1303 srcdest_rnode2str(rn
, buf
, sizeof(buf
));
1304 zlog_debug("%u:%s: Deleting route rn %p, re %p (%s)",
1305 zvrf_id(zvrf
), buf
, rn
, old
,
1306 zebra_route_string(old
->type
));
1309 /* If labeled-unicast route, uninstall transit LSP. */
1310 if (zebra_rib_labeled_unicast(old
))
1311 zebra_mpls_lsp_uninstall(zvrf
, rn
, old
);
1313 if (!RIB_SYSTEM_ROUTE(old
))
1314 rib_uninstall_kernel(rn
, old
);
1316 UNSET_FLAG(old
->status
, ROUTE_ENTRY_INSTALLED
);
1318 * We are setting this to NULL here
1319 * because that is what we traditionally
1320 * have been doing. I am not positive
1321 * that this is the right thing to do
1322 * but let's leave the code alone
1323 * for the RIB_SYSTEM_ROUTE case
1325 dest
->selected_fib
= NULL
;
1328 /* Update nexthop for route, reset changed flag. */
1329 /* Note: this code also handles the Linux case when an interface goes
1330 * down, causing the kernel to delete routes without sending DELROUTE
1333 if (!nexthop_active_update(rn
, old
, true) &&
1334 (RIB_KERNEL_ROUTE(old
)))
1335 SET_FLAG(old
->status
, ROUTE_ENTRY_REMOVED
);
1337 UNSET_FLAG(old
->status
, ROUTE_ENTRY_CHANGED
);
1340 static void rib_process_update_fib(struct zebra_vrf
*zvrf
,
1341 struct route_node
*rn
,
1342 struct route_entry
*old
,
1343 struct route_entry
*new)
1345 struct nexthop
*nexthop
= NULL
;
1347 rib_dest_t
*dest
= rib_dest_from_rnode(rn
);
1350 * We have to install or update if a new route has been selected or
1351 * something has changed.
1353 if (new != old
|| CHECK_FLAG(new->status
, ROUTE_ENTRY_CHANGED
)) {
1354 hook_call(rib_update
, rn
, "updating existing route");
1356 /* Update the nexthop; we could determine here that nexthop is
1358 if (nexthop_active_update(rn
, new, true))
1361 /* If nexthop is active, install the selected route, if
1363 * the install succeeds, cleanup flags for prior route, if
1368 if (IS_ZEBRA_DEBUG_RIB
) {
1369 char buf
[SRCDEST2STR_BUFFER
];
1370 srcdest_rnode2str(rn
, buf
, sizeof(buf
));
1373 "%u:%s: Updating route rn %p, re %p (%s) old %p (%s)",
1374 zvrf_id(zvrf
), buf
, rn
, new,
1375 zebra_route_string(new->type
),
1377 zebra_route_string(old
->type
));
1380 "%u:%s: Updating route rn %p, re %p (%s)",
1381 zvrf_id(zvrf
), buf
, rn
, new,
1382 zebra_route_string(new->type
));
1385 /* If labeled-unicast route, uninstall transit LSP. */
1386 if (zebra_rib_labeled_unicast(old
))
1387 zebra_mpls_lsp_uninstall(zvrf
, rn
, old
);
1389 /* Non-system route should be installed. */
1390 if (!RIB_SYSTEM_ROUTE(new)) {
1391 /* If labeled-unicast route, install transit
1393 if (zebra_rib_labeled_unicast(new))
1394 zebra_mpls_lsp_install(zvrf
, rn
, new);
1396 rib_install_kernel(rn
, new, old
);
1398 UNSET_FLAG(new->status
, ROUTE_ENTRY_INSTALLED
);
1400 * We do not need to install the
1401 * selected route because it
1402 * is already isntalled by
1403 * the system( ie not us )
1404 * so just mark it as winning
1405 * we do need to ensure that
1406 * if we uninstall a route
1407 * from ourselves we don't
1408 * over write this pointer
1410 dest
->selected_fib
= NULL
;
1412 /* If install succeeded or system route, cleanup flags
1413 * for prior route. */
1415 if (RIB_SYSTEM_ROUTE(new)) {
1416 if (!RIB_SYSTEM_ROUTE(old
))
1417 rib_uninstall_kernel(rn
, old
);
1421 ROUTE_ENTRY_INSTALLED
);
1423 UNSET_FLAG(old
->status
,
1424 ROUTE_ENTRY_INSTALLED
);
1425 for (nexthop
= old
->ng
.nexthop
; nexthop
;
1426 nexthop
= nexthop
->next
)
1427 UNSET_FLAG(nexthop
->flags
,
1434 * If nexthop for selected route is not active or install
1436 * may need to uninstall and delete for redistribution.
1439 if (IS_ZEBRA_DEBUG_RIB
) {
1440 char buf
[SRCDEST2STR_BUFFER
];
1441 srcdest_rnode2str(rn
, buf
, sizeof(buf
));
1444 "%u:%s: Deleting route rn %p, re %p (%s) old %p (%s) - nexthop inactive",
1445 zvrf_id(zvrf
), buf
, rn
, new,
1446 zebra_route_string(new->type
),
1448 zebra_route_string(old
->type
));
1451 "%u:%s: Deleting route rn %p, re %p (%s) - nexthop inactive",
1452 zvrf_id(zvrf
), buf
, rn
, new,
1453 zebra_route_string(new->type
));
1456 /* If labeled-unicast route, uninstall transit LSP. */
1457 if (zebra_rib_labeled_unicast(old
))
1458 zebra_mpls_lsp_uninstall(zvrf
, rn
, old
);
1460 if (!RIB_SYSTEM_ROUTE(old
))
1461 rib_uninstall_kernel(rn
, old
);
1463 UNSET_FLAG(old
->status
, ROUTE_ENTRY_INSTALLED
);
1464 dest
->selected_fib
= NULL
;
1469 * Same route selected; check if in the FIB and if not,
1471 * is housekeeping code to deal with race conditions in kernel
1473 * netlink reporting interface up before IPv4 or IPv6 protocol
1477 if (!RIB_SYSTEM_ROUTE(new)
1478 && !CHECK_FLAG(new->status
, ROUTE_ENTRY_INSTALLED
))
1479 rib_install_kernel(rn
, new, NULL
);
1482 /* Update prior route. */
1484 /* Set real nexthop. */
1485 nexthop_active_update(rn
, old
, true);
1486 UNSET_FLAG(old
->status
, ROUTE_ENTRY_CHANGED
);
1489 /* Clear changed flag. */
1490 UNSET_FLAG(new->status
, ROUTE_ENTRY_CHANGED
);
1493 /* Check if 'alternate' RIB entry is better than 'current'. */
1494 static struct route_entry
*rib_choose_best(struct route_entry
*current
,
1495 struct route_entry
*alternate
)
1497 if (current
== NULL
)
1500 /* filter route selection in following order:
1501 * - connected beats other types
1502 * - if both connected, loopback or vrf wins
1503 * - lower distance beats higher
1504 * - lower metric beats higher for equal distance
1505 * - last, hence oldest, route wins tie break.
1508 /* Connected routes. Check to see if either are a vrf
1509 * or loopback interface. If not, pick the last connected
1510 * route of the set of lowest metric connected routes.
1512 if (alternate
->type
== ZEBRA_ROUTE_CONNECT
) {
1513 if (current
->type
!= ZEBRA_ROUTE_CONNECT
)
1516 /* both are connected. are either loop or vrf? */
1517 struct nexthop
*nexthop
= NULL
;
1519 for (ALL_NEXTHOPS(alternate
->ng
, nexthop
)) {
1520 if (if_is_loopback_or_vrf(if_lookup_by_index(
1521 nexthop
->ifindex
, alternate
->vrf_id
)))
1525 for (ALL_NEXTHOPS(current
->ng
, nexthop
)) {
1526 if (if_is_loopback_or_vrf(if_lookup_by_index(
1527 nexthop
->ifindex
, current
->vrf_id
)))
1531 /* Neither are loop or vrf so pick best metric */
1532 if (alternate
->metric
<= current
->metric
)
1538 if (current
->type
== ZEBRA_ROUTE_CONNECT
)
1541 /* higher distance loses */
1542 if (alternate
->distance
< current
->distance
)
1544 if (current
->distance
< alternate
->distance
)
1547 /* metric tie-breaks equal distance */
1548 if (alternate
->metric
<= current
->metric
)
1554 /* Core function for processing routing information base. */
1555 static void rib_process(struct route_node
*rn
)
1557 struct route_entry
*re
;
1558 struct route_entry
*next
;
1559 struct route_entry
*old_selected
= NULL
;
1560 struct route_entry
*new_selected
= NULL
;
1561 struct route_entry
*old_fib
= NULL
;
1562 struct route_entry
*new_fib
= NULL
;
1563 struct route_entry
*best
= NULL
;
1564 char buf
[SRCDEST2STR_BUFFER
];
1566 struct zebra_vrf
*zvrf
= NULL
;
1567 const struct prefix
*p
, *src_p
;
1569 srcdest_rnode_prefixes(rn
, &p
, &src_p
);
1570 vrf_id_t vrf_id
= VRF_UNKNOWN
;
1574 dest
= rib_dest_from_rnode(rn
);
1576 zvrf
= rib_dest_vrf(dest
);
1577 vrf_id
= zvrf_id(zvrf
);
1580 if (IS_ZEBRA_DEBUG_RIB
)
1581 srcdest_rnode2str(rn
, buf
, sizeof(buf
));
1583 if (IS_ZEBRA_DEBUG_RIB_DETAILED
)
1584 zlog_debug("%u:%s: Processing rn %p", vrf_id
, buf
, rn
);
1587 * we can have rn's that have a NULL info pointer
1588 * (dest). As such let's not let the deref happen
1589 * additionally we know RNODE_FOREACH_RE_SAFE
1590 * will not iterate so we are ok.
1593 old_fib
= dest
->selected_fib
;
1595 RNODE_FOREACH_RE_SAFE (rn
, re
, next
) {
1596 if (IS_ZEBRA_DEBUG_RIB_DETAILED
)
1598 "%u:%s: Examine re %p (%s) status %x flags %x dist %d metric %d",
1599 vrf_id
, buf
, re
, zebra_route_string(re
->type
),
1600 re
->status
, re
->flags
, re
->distance
,
1603 UNSET_FLAG(re
->status
, ROUTE_ENTRY_NEXTHOPS_CHANGED
);
1605 /* Currently selected re. */
1606 if (CHECK_FLAG(re
->flags
, ZEBRA_FLAG_SELECTED
)) {
1607 assert(old_selected
== NULL
);
1611 /* Skip deleted entries from selection */
1612 if (CHECK_FLAG(re
->status
, ROUTE_ENTRY_REMOVED
))
1615 /* Skip unreachable nexthop. */
1616 /* This first call to nexthop_active_update is merely to
1618 * there's any change to nexthops associated with this RIB
1620 * rib_process() can be invoked due to an external event such as
1622 * down or due to next-hop-tracking evaluation. In the latter
1624 * a decision has already been made that the NHs have changed.
1626 * need to invoke a potentially expensive call again. Further,
1628 * the change might be in a recursive NH which is not caught in
1629 * the nexthop_active_update() code. Thus, we might miss changes
1633 if (!CHECK_FLAG(re
->status
, ROUTE_ENTRY_CHANGED
)
1634 && !nexthop_active_update(rn
, re
, false)) {
1635 if (re
->type
== ZEBRA_ROUTE_TABLE
) {
1636 /* XXX: HERE BE DRAGONS!!!!!
1637 * In all honesty, I have not yet figured out
1639 * does or why the ROUTE_ENTRY_CHANGED test
1641 * or why we need to delete a route here, and
1643 * this concerns both selected and fib route, or
1646 /* This entry was denied by the 'ip protocol
1647 * table' route-map, we
1648 * need to delete it */
1649 if (re
!= old_selected
) {
1650 if (IS_ZEBRA_DEBUG_RIB
)
1652 "%s: %u:%s: imported via import-table but denied "
1653 "by the ip protocol table route-map",
1654 __func__
, vrf_id
, buf
);
1657 SET_FLAG(re
->status
,
1658 ROUTE_ENTRY_REMOVED
);
1664 /* Infinite distance. */
1665 if (re
->distance
== DISTANCE_INFINITY
) {
1666 UNSET_FLAG(re
->status
, ROUTE_ENTRY_CHANGED
);
1670 if (CHECK_FLAG(re
->flags
, ZEBRA_FLAG_FIB_OVERRIDE
)) {
1671 best
= rib_choose_best(new_fib
, re
);
1672 if (new_fib
&& best
!= new_fib
)
1673 UNSET_FLAG(new_fib
->status
,
1674 ROUTE_ENTRY_CHANGED
);
1677 best
= rib_choose_best(new_selected
, re
);
1678 if (new_selected
&& best
!= new_selected
)
1679 UNSET_FLAG(new_selected
->status
,
1680 ROUTE_ENTRY_CHANGED
);
1681 new_selected
= best
;
1684 UNSET_FLAG(re
->status
, ROUTE_ENTRY_CHANGED
);
1685 } /* RNODE_FOREACH_RE */
1687 /* If no FIB override route, use the selected route also for FIB */
1688 if (new_fib
== NULL
)
1689 new_fib
= new_selected
;
1691 /* After the cycle is finished, the following pointers will be set:
1692 * old_selected --- RE entry currently having SELECTED
1693 * new_selected --- RE entry that is newly SELECTED
1694 * old_fib --- RE entry currently in kernel FIB
1695 * new_fib --- RE entry that is newly to be in kernel FIB
1697 * new_selected will get SELECTED flag, and is going to be redistributed
1698 * the zclients. new_fib (which can be new_selected) will be installed
1702 if (IS_ZEBRA_DEBUG_RIB_DETAILED
) {
1704 "%u:%s: After processing: old_selected %p new_selected %p old_fib %p new_fib %p",
1705 vrf_id
, buf
, (void *)old_selected
, (void *)new_selected
,
1706 (void *)old_fib
, (void *)new_fib
);
1709 /* Buffer ROUTE_ENTRY_CHANGED here, because it will get cleared if
1710 * fib == selected */
1711 bool selected_changed
= new_selected
&& CHECK_FLAG(new_selected
->status
,
1712 ROUTE_ENTRY_CHANGED
);
1714 /* Update fib according to selection results */
1715 if (new_fib
&& old_fib
)
1716 rib_process_update_fib(zvrf
, rn
, old_fib
, new_fib
);
1718 rib_process_add_fib(zvrf
, rn
, new_fib
);
1720 rib_process_del_fib(zvrf
, rn
, old_fib
);
1722 /* Update SELECTED entry */
1723 if (old_selected
!= new_selected
|| selected_changed
) {
1725 if (new_selected
&& new_selected
!= new_fib
) {
1726 nexthop_active_update(rn
, new_selected
, true);
1727 UNSET_FLAG(new_selected
->status
, ROUTE_ENTRY_CHANGED
);
1731 SET_FLAG(new_selected
->flags
, ZEBRA_FLAG_SELECTED
);
1733 /* Special case: new route is system route, so
1734 * dataplane update will not be done - ensure we
1735 * redistribute the route.
1737 if (RIB_SYSTEM_ROUTE(new_selected
))
1738 redistribute_update(p
, src_p
, new_selected
,
1744 redistribute_delete(p
, src_p
, old_selected
);
1745 if (old_selected
!= new_selected
)
1746 UNSET_FLAG(old_selected
->flags
,
1747 ZEBRA_FLAG_SELECTED
);
1751 /* Remove all RE entries queued for removal */
1752 RNODE_FOREACH_RE_SAFE (rn
, re
, next
) {
1753 if (CHECK_FLAG(re
->status
, ROUTE_ENTRY_REMOVED
)) {
1754 if (IS_ZEBRA_DEBUG_RIB
) {
1755 rnode_debug(rn
, vrf_id
, "rn %p, removing re %p",
1756 (void *)rn
, (void *)re
);
1763 * Check if the dest can be deleted now.
1769 * Utility to match route with dplane context data
1771 static bool rib_route_match_ctx(const struct route_entry
*re
,
1772 const struct zebra_dplane_ctx
*ctx
,
1775 bool result
= false;
1779 * In 'update' case, we test info about the 'previous' or
1782 if ((re
->type
== dplane_ctx_get_old_type(ctx
)) &&
1783 (re
->instance
== dplane_ctx_get_old_instance(ctx
))) {
1786 /* TODO -- we're using this extra test, but it's not
1787 * exactly clear why.
1789 if (re
->type
== ZEBRA_ROUTE_STATIC
&&
1790 (re
->distance
!= dplane_ctx_get_old_distance(ctx
) ||
1791 re
->tag
!= dplane_ctx_get_old_tag(ctx
))) {
1798 * Ordinary, single-route case using primary context info
1800 if ((dplane_ctx_get_op(ctx
) != DPLANE_OP_ROUTE_DELETE
) &&
1801 CHECK_FLAG(re
->status
, ROUTE_ENTRY_REMOVED
)) {
1802 /* Skip route that's been deleted */
1806 if ((re
->type
== dplane_ctx_get_type(ctx
)) &&
1807 (re
->instance
== dplane_ctx_get_instance(ctx
))) {
1810 /* TODO -- we're using this extra test, but it's not
1811 * exactly clear why.
1813 if (re
->type
== ZEBRA_ROUTE_STATIC
&&
1814 (re
->distance
!= dplane_ctx_get_distance(ctx
) ||
1815 re
->tag
!= dplane_ctx_get_tag(ctx
))) {
1827 * Route-update results processing after async dataplane update.
1829 static void rib_process_result(struct zebra_dplane_ctx
*ctx
)
1831 struct route_table
*table
= NULL
;
1832 struct zebra_vrf
*zvrf
= NULL
;
1833 struct route_node
*rn
= NULL
;
1834 struct route_entry
*re
= NULL
, *old_re
= NULL
, *rib
;
1835 bool is_update
= false;
1836 struct nexthop
*nexthop
, *ctx_nexthop
;
1837 char dest_str
[PREFIX_STRLEN
] = "";
1838 enum dplane_op_e op
;
1839 enum zebra_dplane_result status
;
1840 const struct prefix
*dest_pfx
, *src_pfx
;
1842 /* Locate rn and re(s) from ctx */
1844 table
= zebra_vrf_table_with_table_id(dplane_ctx_get_afi(ctx
),
1845 dplane_ctx_get_safi(ctx
),
1846 dplane_ctx_get_vrf(ctx
),
1847 dplane_ctx_get_table(ctx
));
1848 if (table
== NULL
) {
1849 if (IS_ZEBRA_DEBUG_DPLANE
) {
1850 zlog_debug("Failed to process dplane results: no table for afi %d, safi %d, vrf %u",
1851 dplane_ctx_get_afi(ctx
),
1852 dplane_ctx_get_safi(ctx
),
1853 dplane_ctx_get_vrf(ctx
));
1858 zvrf
= vrf_info_lookup(dplane_ctx_get_vrf(ctx
));
1860 dest_pfx
= dplane_ctx_get_dest(ctx
);
1862 /* Note well: only capturing the prefix string if debug is enabled here;
1863 * unconditional log messages will have to generate the string.
1865 if (IS_ZEBRA_DEBUG_DPLANE
)
1866 prefix2str(dest_pfx
, dest_str
, sizeof(dest_str
));
1868 src_pfx
= dplane_ctx_get_src(ctx
);
1869 rn
= srcdest_rnode_get(table
, dplane_ctx_get_dest(ctx
),
1870 src_pfx
? (struct prefix_ipv6
*)src_pfx
: NULL
);
1872 if (IS_ZEBRA_DEBUG_DPLANE
) {
1873 zlog_debug("Failed to process dplane results: no route for %u:%s",
1874 dplane_ctx_get_vrf(ctx
), dest_str
);
1879 srcdest_rnode_prefixes(rn
, &dest_pfx
, &src_pfx
);
1881 op
= dplane_ctx_get_op(ctx
);
1882 status
= dplane_ctx_get_status(ctx
);
1884 if (IS_ZEBRA_DEBUG_DPLANE_DETAIL
)
1885 zlog_debug("%u:%s Processing dplane ctx %p, op %s result %s",
1886 dplane_ctx_get_vrf(ctx
), dest_str
, ctx
,
1887 dplane_op2str(op
), dplane_res2str(status
));
1890 * Update is a bit of a special case, where we may have both old and new
1891 * routes to post-process.
1893 is_update
= dplane_ctx_is_update(ctx
);
1896 * Take a pass through the routes, look for matches with the context
1899 RNODE_FOREACH_RE(rn
, rib
) {
1902 if (rib_route_match_ctx(rib
, ctx
, false))
1906 /* Check for old route match */
1907 if (is_update
&& (old_re
== NULL
)) {
1908 if (rib_route_match_ctx(rib
, ctx
, true /*is_update*/))
1912 /* Have we found the routes we need to work on? */
1913 if (re
&& ((!is_update
|| old_re
)))
1918 * Check sequence number(s) to detect stale results before continuing
1921 if (re
->dplane_sequence
!= dplane_ctx_get_seq(ctx
)) {
1922 if (IS_ZEBRA_DEBUG_DPLANE_DETAIL
)
1923 zlog_debug("%u:%s Stale dplane result for re %p",
1924 dplane_ctx_get_vrf(ctx
),
1927 UNSET_FLAG(re
->status
, ROUTE_ENTRY_QUEUED
);
1931 if (old_re
->dplane_sequence
!= dplane_ctx_get_old_seq(ctx
)) {
1932 if (IS_ZEBRA_DEBUG_DPLANE_DETAIL
)
1933 zlog_debug("%u:%s Stale dplane result for old_re %p",
1934 dplane_ctx_get_vrf(ctx
),
1937 UNSET_FLAG(re
->status
, ROUTE_ENTRY_QUEUED
);
1941 case DPLANE_OP_ROUTE_INSTALL
:
1942 case DPLANE_OP_ROUTE_UPDATE
:
1943 if (status
== ZEBRA_DPLANE_REQUEST_SUCCESS
) {
1945 UNSET_FLAG(re
->status
, ROUTE_ENTRY_FAILED
);
1946 SET_FLAG(re
->status
, ROUTE_ENTRY_INSTALLED
);
1949 UNSET_FLAG(old_re
->status
, ROUTE_ENTRY_FAILED
);
1950 UNSET_FLAG(old_re
->status
,
1951 ROUTE_ENTRY_INSTALLED
);
1953 /* Update zebra nexthop FIB flag for each
1954 * nexthop that was installed.
1956 for (ALL_NEXTHOPS_PTR(dplane_ctx_get_ng(ctx
),
1962 for (ALL_NEXTHOPS(re
->ng
, nexthop
)) {
1963 if (nexthop_same(ctx_nexthop
, nexthop
))
1967 if (nexthop
== NULL
)
1970 if (CHECK_FLAG(nexthop
->flags
,
1971 NEXTHOP_FLAG_RECURSIVE
))
1974 if (CHECK_FLAG(ctx_nexthop
->flags
,
1976 SET_FLAG(nexthop
->flags
,
1979 UNSET_FLAG(nexthop
->flags
,
1985 /* Set flag for nexthop tracking processing */
1986 zvrf
->flags
|= ZEBRA_VRF_RIB_SCHEDULED
;
1991 * TODO -- still calling the redist api using the
1992 * route_entries, and there's a corner-case here:
1993 * if there's no client for the 'new' route, a redist
1994 * deleting the 'old' route will be sent. But if the
1995 * 'old' context info was stale, 'old_re' will be
1996 * NULL here and that delete will not be sent.
1999 redistribute_update(dest_pfx
, src_pfx
,
2002 /* Notify route owner */
2003 zsend_route_notify_owner_ctx(ctx
, ZAPI_ROUTE_INSTALLED
);
2007 SET_FLAG(re
->status
, ROUTE_ENTRY_FAILED
);
2009 SET_FLAG(old_re
->status
, ROUTE_ENTRY_FAILED
);
2011 zsend_route_notify_owner(re
, dest_pfx
,
2012 ZAPI_ROUTE_FAIL_INSTALL
);
2014 zlog_warn("%u:%s: Route install failed",
2015 dplane_ctx_get_vrf(ctx
),
2016 prefix2str(dest_pfx
,
2017 dest_str
, sizeof(dest_str
)));
2020 case DPLANE_OP_ROUTE_DELETE
:
2022 SET_FLAG(re
->status
, ROUTE_ENTRY_FAILED
);
2024 * In the delete case, the zebra core datastructs were
2025 * updated (or removed) at the time the delete was issued,
2026 * so we're just notifying the route owner.
2028 if (status
== ZEBRA_DPLANE_REQUEST_SUCCESS
) {
2030 UNSET_FLAG(re
->status
, ROUTE_ENTRY_INSTALLED
);
2031 UNSET_FLAG(re
->status
, ROUTE_ENTRY_FAILED
);
2033 zsend_route_notify_owner_ctx(ctx
, ZAPI_ROUTE_REMOVED
);
2039 SET_FLAG(re
->status
, ROUTE_ENTRY_FAILED
);
2040 zsend_route_notify_owner_ctx(ctx
,
2041 ZAPI_ROUTE_REMOVE_FAIL
);
2043 zlog_warn("%u:%s: Route Deletion failure",
2044 dplane_ctx_get_vrf(ctx
),
2045 prefix2str(dest_pfx
,
2046 dest_str
, sizeof(dest_str
)));
2054 /* Return context to dataplane module */
2055 dplane_ctx_fini(&ctx
);
2058 /* Take a list of route_node structs and return 1, if there was a record
2059 * picked from it and processed by rib_process(). Don't process more,
2060 * than one RN record; operate only in the specified sub-queue.
2062 static unsigned int process_subq(struct list
*subq
, uint8_t qindex
)
2064 struct listnode
*lnode
= listhead(subq
);
2065 struct route_node
*rnode
;
2067 struct zebra_vrf
*zvrf
= NULL
;
2072 rnode
= listgetdata(lnode
);
2073 dest
= rib_dest_from_rnode(rnode
);
2075 zvrf
= rib_dest_vrf(dest
);
2079 if (IS_ZEBRA_DEBUG_RIB_DETAILED
) {
2080 char buf
[SRCDEST2STR_BUFFER
];
2081 srcdest_rnode2str(rnode
, buf
, sizeof(buf
));
2082 zlog_debug("%u:%s: rn %p dequeued from sub-queue %u",
2083 zvrf
? zvrf_id(zvrf
) : 0, buf
, rnode
, qindex
);
2087 UNSET_FLAG(rib_dest_from_rnode(rnode
)->flags
,
2088 RIB_ROUTE_QUEUED(qindex
));
2093 zlog_debug ("%s: called for route_node (%p, %d) with no ribs",
2094 __func__
, rnode
, rnode
->lock
);
2095 zlog_backtrace(LOG_DEBUG
);
2098 route_unlock_node(rnode
);
2099 list_delete_node(subq
, lnode
);
2104 * Perform next-hop tracking processing after RIB updates.
2106 static void do_nht_processing(void)
2109 struct zebra_vrf
*zvrf
;
2111 /* Evaluate nexthops for those VRFs which underwent route processing.
2113 * should limit the evaluation to the necessary VRFs in most common
2116 RB_FOREACH (vrf
, vrf_id_head
, &vrfs_by_id
) {
2118 if (zvrf
== NULL
|| !(zvrf
->flags
& ZEBRA_VRF_RIB_SCHEDULED
))
2121 if (IS_ZEBRA_DEBUG_RIB_DETAILED
|| IS_ZEBRA_DEBUG_NHT
)
2122 zlog_debug("NHT processing check for zvrf %s",
2125 zvrf
->flags
&= ~ZEBRA_VRF_RIB_SCHEDULED
;
2126 zebra_evaluate_rnh(zvrf
, AFI_IP
, 0, RNH_NEXTHOP_TYPE
, NULL
);
2127 zebra_evaluate_rnh(zvrf
, AFI_IP
, 0, RNH_IMPORT_CHECK_TYPE
,
2129 zebra_evaluate_rnh(zvrf
, AFI_IP6
, 0, RNH_NEXTHOP_TYPE
, NULL
);
2130 zebra_evaluate_rnh(zvrf
, AFI_IP6
, 0, RNH_IMPORT_CHECK_TYPE
,
2134 /* Schedule LSPs for processing, if needed. */
2135 zvrf
= vrf_info_lookup(VRF_DEFAULT
);
2136 if (mpls_should_lsps_be_processed(zvrf
)) {
2137 if (IS_ZEBRA_DEBUG_MPLS
)
2139 "%u: Scheduling all LSPs upon RIB completion",
2141 zebra_mpls_lsp_schedule(zvrf
);
2142 mpls_unmark_lsps_for_processing(zvrf
);
2147 * All meta queues have been processed. Trigger next-hop evaluation.
2149 static void meta_queue_process_complete(struct work_queue
*dummy
)
2151 do_nht_processing();
2154 /* Dispatch the meta queue by picking, processing and unlocking the next RN from
2155 * a non-empty sub-queue with lowest priority. wq is equal to zebra->ribq and
2157 * is pointed to the meta queue structure.
2159 static wq_item_status
meta_queue_process(struct work_queue
*dummy
, void *data
)
2161 struct meta_queue
*mq
= data
;
2163 uint32_t queue_len
, queue_limit
;
2165 /* Ensure there's room for more dataplane updates */
2166 queue_limit
= dplane_get_in_queue_limit();
2167 queue_len
= dplane_get_in_queue_len();
2168 if (queue_len
> queue_limit
) {
2169 if (IS_ZEBRA_DEBUG_RIB_DETAILED
)
2170 zlog_debug("rib queue: dplane queue len %u, limit %u, retrying",
2171 queue_len
, queue_limit
);
2173 /* Ensure that the meta-queue is actually enqueued */
2174 if (work_queue_empty(zebrad
.ribq
))
2175 work_queue_add(zebrad
.ribq
, zebrad
.mq
);
2177 return WQ_QUEUE_BLOCKED
;
2180 for (i
= 0; i
< MQ_SIZE
; i
++)
2181 if (process_subq(mq
->subq
[i
], i
)) {
2185 return mq
->size
? WQ_REQUEUE
: WQ_SUCCESS
;
2190 * Look into the RN and queue it into the highest priority queue
2191 * at this point in time for processing.
2193 * We will enqueue a route node only once per invocation.
2195 * There are two possibilities here that should be kept in mind.
2196 * If the original invocation has not been pulled off for processing
2197 * yet, A subsuquent invocation can have a route entry with a better
2198 * meta queue index value and we can have a situation where
2199 * we might have the same node enqueued 2 times. Not necessarily
2200 * an optimal situation but it should be ok.
2202 * The other possibility is that the original invocation has not
2203 * been pulled off for processing yet, A subsusquent invocation
2204 * doesn't have a route_entry with a better meta-queue and the
2205 * original metaqueue index value will win and we'll end up with
2206 * the route node enqueued once.
2208 static void rib_meta_queue_add(struct meta_queue
*mq
, struct route_node
*rn
)
2210 struct route_entry
*re
= NULL
, *curr_re
= NULL
;
2211 uint8_t qindex
= MQ_SIZE
, curr_qindex
= MQ_SIZE
;
2212 struct zebra_vrf
*zvrf
;
2214 RNODE_FOREACH_RE (rn
, curr_re
) {
2215 curr_qindex
= route_info
[curr_re
->type
].meta_q_map
;
2217 if (curr_qindex
<= qindex
) {
2219 qindex
= curr_qindex
;
2226 /* Invariant: at this point we always have rn->info set. */
2227 if (CHECK_FLAG(rib_dest_from_rnode(rn
)->flags
,
2228 RIB_ROUTE_QUEUED(qindex
))) {
2229 if (IS_ZEBRA_DEBUG_RIB_DETAILED
)
2230 rnode_debug(rn
, re
->vrf_id
,
2231 "rn %p is already queued in sub-queue %u",
2232 (void *)rn
, qindex
);
2236 SET_FLAG(rib_dest_from_rnode(rn
)->flags
, RIB_ROUTE_QUEUED(qindex
));
2237 listnode_add(mq
->subq
[qindex
], rn
);
2238 route_lock_node(rn
);
2241 if (IS_ZEBRA_DEBUG_RIB_DETAILED
)
2242 rnode_debug(rn
, re
->vrf_id
, "queued rn %p into sub-queue %u",
2243 (void *)rn
, qindex
);
2245 zvrf
= zebra_vrf_lookup_by_id(re
->vrf_id
);
2247 zvrf
->flags
|= ZEBRA_VRF_RIB_SCHEDULED
;
2250 /* Add route_node to work queue and schedule processing */
2251 void rib_queue_add(struct route_node
*rn
)
2255 /* Pointless to queue a route_node with no RIB entries to add or remove
2257 if (!rnode_to_ribs(rn
)) {
2258 zlog_debug("%s: called for route_node (%p, %d) with no ribs",
2259 __func__
, (void *)rn
, rn
->lock
);
2260 zlog_backtrace(LOG_DEBUG
);
2264 if (zebrad
.ribq
== NULL
) {
2265 flog_err(EC_ZEBRA_WQ_NONEXISTENT
,
2266 "%s: work_queue does not exist!", __func__
);
2271 * The RIB queue should normally be either empty or holding the only
2272 * work_queue_item element. In the latter case this element would
2273 * hold a pointer to the meta queue structure, which must be used to
2274 * actually queue the route nodes to process. So create the MQ
2275 * holder, if necessary, then push the work into it in any case.
2276 * This semantics was introduced after 0.99.9 release.
2278 if (work_queue_empty(zebrad
.ribq
))
2279 work_queue_add(zebrad
.ribq
, zebrad
.mq
);
2281 rib_meta_queue_add(zebrad
.mq
, rn
);
2286 /* Create new meta queue.
2287 A destructor function doesn't seem to be necessary here.
2289 static struct meta_queue
*meta_queue_new(void)
2291 struct meta_queue
*new;
2294 new = XCALLOC(MTYPE_WORK_QUEUE
, sizeof(struct meta_queue
));
2296 for (i
= 0; i
< MQ_SIZE
; i
++) {
2297 new->subq
[i
] = list_new();
2298 assert(new->subq
[i
]);
2304 void meta_queue_free(struct meta_queue
*mq
)
2308 for (i
= 0; i
< MQ_SIZE
; i
++)
2309 list_delete(&mq
->subq
[i
]);
2311 XFREE(MTYPE_WORK_QUEUE
, mq
);
2314 /* initialise zebra rib work queue */
2315 static void rib_queue_init(struct zebra_t
*zebra
)
2320 work_queue_new(zebra
->master
, "route_node processing"))) {
2321 flog_err(EC_ZEBRA_WQ_NONEXISTENT
,
2322 "%s: could not initialise work queue!", __func__
);
2326 /* fill in the work queue spec */
2327 zebra
->ribq
->spec
.workfunc
= &meta_queue_process
;
2328 zebra
->ribq
->spec
.errorfunc
= NULL
;
2329 zebra
->ribq
->spec
.completion_func
= &meta_queue_process_complete
;
2330 /* XXX: TODO: These should be runtime configurable via vty */
2331 zebra
->ribq
->spec
.max_retries
= 3;
2332 zebra
->ribq
->spec
.hold
= ZEBRA_RIB_PROCESS_HOLD_TIME
;
2333 zebra
->ribq
->spec
.retry
= ZEBRA_RIB_PROCESS_RETRY_TIME
;
2335 if (!(zebra
->mq
= meta_queue_new())) {
2336 flog_err(EC_ZEBRA_WQ_NONEXISTENT
,
2337 "%s: could not initialise meta queue!", __func__
);
2343 /* RIB updates are processed via a queue of pointers to route_nodes.
2345 * The queue length is bounded by the maximal size of the routing table,
2346 * as a route_node will not be requeued, if already queued.
2348 * REs are submitted via rib_addnode or rib_delnode which set minimal
2349 * state, or static_install_route (when an existing RE is updated)
2350 * and then submit route_node to queue for best-path selection later.
2351 * Order of add/delete state changes are preserved for any given RE.
2353 * Deleted REs are reaped during best-path selection.
2356 * |-> rib_link or unset ROUTE_ENTRY_REMOVE |->Update kernel with
2357 * |-------->| | best RE, if required
2359 * static_install->|->rib_addqueue...... -> rib_process
2361 * |-------->| |-> rib_unlink
2362 * |-> set ROUTE_ENTRY_REMOVE |
2363 * rib_delnode (RE freed)
2365 * The 'info' pointer of a route_node points to a rib_dest_t
2366 * ('dest'). Queueing state for a route_node is kept on the dest. The
2367 * dest is created on-demand by rib_link() and is kept around at least
2368 * as long as there are ribs hanging off it (@see rib_gc_dest()).
2370 * Refcounting (aka "locking" throughout the GNU Zebra and Quagga code):
2372 * - route_nodes: refcounted by:
2373 * - dest attached to route_node:
2374 * - managed by: rib_link/rib_gc_dest
2375 * - route_node processing queue
2376 * - managed by: rib_addqueue, rib_process.
2380 /* Add RE to head of the route node. */
2381 static void rib_link(struct route_node
*rn
, struct route_entry
*re
, int process
)
2383 struct route_entry
*head
;
2386 const char *rmap_name
;
2390 dest
= rib_dest_from_rnode(rn
);
2392 if (IS_ZEBRA_DEBUG_RIB_DETAILED
)
2393 rnode_debug(rn
, re
->vrf_id
, "rn %p adding dest", rn
);
2395 dest
= XCALLOC(MTYPE_RIB_DEST
, sizeof(rib_dest_t
));
2396 route_lock_node(rn
); /* rn route table reference */
2401 head
= dest
->routes
;
2408 afi
= (rn
->p
.family
== AF_INET
)
2410 : (rn
->p
.family
== AF_INET6
) ? AFI_IP6
: AFI_MAX
;
2411 if (is_zebra_import_table_enabled(afi
, re
->table
)) {
2412 rmap_name
= zebra_get_import_table_route_map(afi
, re
->table
);
2413 zebra_add_import_table_entry(rn
, re
, rmap_name
);
2418 static void rib_addnode(struct route_node
*rn
,
2419 struct route_entry
*re
, int process
)
2421 /* RE node has been un-removed before route-node is processed.
2422 * route_node must hence already be on the queue for processing..
2424 if (CHECK_FLAG(re
->status
, ROUTE_ENTRY_REMOVED
)) {
2425 if (IS_ZEBRA_DEBUG_RIB
)
2426 rnode_debug(rn
, re
->vrf_id
, "rn %p, un-removed re %p",
2427 (void *)rn
, (void *)re
);
2429 UNSET_FLAG(re
->status
, ROUTE_ENTRY_REMOVED
);
2432 rib_link(rn
, re
, process
);
2438 * Detach a rib structure from a route_node.
2440 * Note that a call to rib_unlink() should be followed by a call to
2441 * rib_gc_dest() at some point. This allows a rib_dest_t that is no
2442 * longer required to be deleted.
2444 void rib_unlink(struct route_node
*rn
, struct route_entry
*re
)
2450 if (IS_ZEBRA_DEBUG_RIB
)
2451 rnode_debug(rn
, re
->vrf_id
, "rn %p, re %p", (void *)rn
,
2454 dest
= rib_dest_from_rnode(rn
);
2457 re
->next
->prev
= re
->prev
;
2460 re
->prev
->next
= re
->next
;
2462 dest
->routes
= re
->next
;
2465 if (dest
->selected_fib
== re
)
2466 dest
->selected_fib
= NULL
;
2468 nexthops_free(re
->ng
.nexthop
);
2469 XFREE(MTYPE_RE
, re
);
2472 void rib_delnode(struct route_node
*rn
, struct route_entry
*re
)
2476 if (IS_ZEBRA_DEBUG_RIB
)
2477 rnode_debug(rn
, re
->vrf_id
, "rn %p, re %p, removing",
2478 (void *)rn
, (void *)re
);
2479 SET_FLAG(re
->status
, ROUTE_ENTRY_REMOVED
);
2481 afi
= (rn
->p
.family
== AF_INET
)
2483 : (rn
->p
.family
== AF_INET6
) ? AFI_IP6
: AFI_MAX
;
2484 if (is_zebra_import_table_enabled(afi
, re
->table
)) {
2485 zebra_del_import_table_entry(rn
, re
);
2486 /* Just clean up if non main table */
2487 if (IS_ZEBRA_DEBUG_RIB
) {
2488 char buf
[SRCDEST2STR_BUFFER
];
2489 srcdest_rnode2str(rn
, buf
, sizeof(buf
));
2490 zlog_debug("%u:%s: Freeing route rn %p, re %p (%s)",
2491 re
->vrf_id
, buf
, rn
, re
,
2492 zebra_route_string(re
->type
));
2501 /* This function dumps the contents of a given RE entry into
2502 * standard debug log. Calling function name and IP prefix in
2503 * question are passed as 1st and 2nd arguments.
2506 void _route_entry_dump(const char *func
, union prefixconstptr pp
,
2507 union prefixconstptr src_pp
,
2508 const struct route_entry
*re
)
2510 const struct prefix
*src_p
= src_pp
.p
;
2511 bool is_srcdst
= src_p
&& src_p
->prefixlen
;
2512 char straddr
[PREFIX_STRLEN
];
2513 char srcaddr
[PREFIX_STRLEN
];
2514 struct nexthop
*nexthop
;
2516 zlog_debug("%s: dumping RE entry %p for %s%s%s vrf %u", func
,
2517 (const void *)re
, prefix2str(pp
, straddr
, sizeof(straddr
)),
2518 is_srcdst
? " from " : "",
2519 is_srcdst
? prefix2str(src_pp
, srcaddr
, sizeof(srcaddr
))
2522 zlog_debug("%s: uptime == %lu, type == %u, instance == %d, table == %d",
2523 func
, (unsigned long)re
->uptime
, re
->type
, re
->instance
,
2526 "%s: metric == %u, mtu == %u, distance == %u, flags == %u, status == %u",
2527 func
, re
->metric
, re
->mtu
, re
->distance
, re
->flags
, re
->status
);
2528 zlog_debug("%s: nexthop_num == %u, nexthop_active_num == %u", func
,
2529 re
->nexthop_num
, re
->nexthop_active_num
);
2531 for (ALL_NEXTHOPS(re
->ng
, nexthop
)) {
2532 struct interface
*ifp
;
2533 struct vrf
*vrf
= vrf_lookup_by_id(nexthop
->vrf_id
);
2535 switch (nexthop
->type
) {
2536 case NEXTHOP_TYPE_BLACKHOLE
:
2537 sprintf(straddr
, "Blackhole");
2539 case NEXTHOP_TYPE_IFINDEX
:
2540 ifp
= if_lookup_by_index(nexthop
->ifindex
,
2542 sprintf(straddr
, "%s", ifp
? ifp
->name
: "Unknown");
2544 case NEXTHOP_TYPE_IPV4
:
2546 case NEXTHOP_TYPE_IPV4_IFINDEX
:
2547 inet_ntop(AF_INET
, &nexthop
->gate
, straddr
,
2550 case NEXTHOP_TYPE_IPV6
:
2551 case NEXTHOP_TYPE_IPV6_IFINDEX
:
2552 inet_ntop(AF_INET6
, &nexthop
->gate
, straddr
,
2556 zlog_debug("%s: %s %s[%u] vrf %s(%u) with flags %s%s%s", func
,
2557 (nexthop
->rparent
? " NH" : "NH"), straddr
,
2558 nexthop
->ifindex
, vrf
? vrf
->name
: "Unknown",
2560 (CHECK_FLAG(nexthop
->flags
, NEXTHOP_FLAG_ACTIVE
)
2563 (CHECK_FLAG(re
->status
, ROUTE_ENTRY_INSTALLED
)
2566 (CHECK_FLAG(nexthop
->flags
, NEXTHOP_FLAG_RECURSIVE
)
2570 zlog_debug("%s: dump complete", func
);
2573 /* This is an exported helper to rtm_read() to dump the strange
2574 * RE entry found by rib_lookup_ipv4_route()
2577 void rib_lookup_and_dump(struct prefix_ipv4
*p
, vrf_id_t vrf_id
)
2579 struct route_table
*table
;
2580 struct route_node
*rn
;
2581 struct route_entry
*re
;
2582 char prefix_buf
[INET_ADDRSTRLEN
];
2585 table
= zebra_vrf_table(AFI_IP
, SAFI_UNICAST
, vrf_id
);
2587 flog_err(EC_ZEBRA_TABLE_LOOKUP_FAILED
,
2588 "%s:%u zebra_vrf_table() returned NULL", __func__
,
2593 /* Scan the RIB table for exactly matching RE entry. */
2594 rn
= route_node_lookup(table
, (struct prefix
*)p
);
2596 /* No route for this prefix. */
2598 zlog_debug("%s:%u lookup failed for %s", __func__
, vrf_id
,
2599 prefix2str((struct prefix
*)p
, prefix_buf
,
2600 sizeof(prefix_buf
)));
2605 route_unlock_node(rn
);
2608 RNODE_FOREACH_RE (rn
, re
) {
2609 zlog_debug("%s:%u rn %p, re %p: %s, %s",
2611 (void *)rn
, (void *)re
,
2612 (CHECK_FLAG(re
->status
, ROUTE_ENTRY_REMOVED
)
2615 (CHECK_FLAG(re
->flags
, ZEBRA_FLAG_SELECTED
)
2618 route_entry_dump(p
, NULL
, re
);
2622 /* Check if requested address assignment will fail due to another
2623 * route being installed by zebra in FIB already. Take necessary
2624 * actions, if needed: remove such a route from FIB and deSELECT
2625 * corresponding RE entry. Then put affected RN into RIBQ head.
2627 void rib_lookup_and_pushup(struct prefix_ipv4
*p
, vrf_id_t vrf_id
)
2629 struct route_table
*table
;
2630 struct route_node
*rn
;
2631 unsigned changed
= 0;
2634 if (NULL
== (table
= zebra_vrf_table(AFI_IP
, SAFI_UNICAST
, vrf_id
))) {
2635 flog_err(EC_ZEBRA_TABLE_LOOKUP_FAILED
,
2636 "%s:%u zebra_vrf_table() returned NULL", __func__
,
2641 /* No matches would be the simplest case. */
2642 if (NULL
== (rn
= route_node_lookup(table
, (struct prefix
*)p
)))
2646 route_unlock_node(rn
);
2648 dest
= rib_dest_from_rnode(rn
);
2649 /* Check all RE entries. In case any changes have to be done, requeue
2650 * the RN into RIBQ head. If the routing message about the new connected
2651 * route (generated by the IP address we are going to assign very soon)
2652 * comes before the RIBQ is processed, the new RE entry will join
2653 * RIBQ record already on head. This is necessary for proper
2655 * of the rest of the RE.
2657 if (dest
->selected_fib
&& !RIB_SYSTEM_ROUTE(dest
->selected_fib
)) {
2659 if (IS_ZEBRA_DEBUG_RIB
) {
2660 char buf
[PREFIX_STRLEN
];
2662 zlog_debug("%u:%s: freeing way for connected prefix",
2663 dest
->selected_fib
->vrf_id
,
2664 prefix2str(&rn
->p
, buf
, sizeof(buf
)));
2665 route_entry_dump(&rn
->p
, NULL
, dest
->selected_fib
);
2667 rib_uninstall(rn
, dest
->selected_fib
);
2673 int rib_add_multipath(afi_t afi
, safi_t safi
, struct prefix
*p
,
2674 struct prefix_ipv6
*src_p
, struct route_entry
*re
)
2676 struct route_table
*table
;
2677 struct route_node
*rn
;
2678 struct route_entry
*same
= NULL
;
2679 struct nexthop
*nexthop
;
2685 assert(!src_p
|| !src_p
->prefixlen
|| afi
== AFI_IP6
);
2688 table
= zebra_vrf_table_with_table_id(afi
, safi
, re
->vrf_id
, re
->table
);
2690 XFREE(MTYPE_RE
, re
);
2694 /* Make it sure prefixlen is applied to the prefix. */
2697 apply_mask_ipv6(src_p
);
2699 /* Set default distance by route type. */
2700 if (re
->distance
== 0) {
2701 re
->distance
= route_distance(re
->type
);
2703 /* iBGP distance is 200. */
2704 if (re
->type
== ZEBRA_ROUTE_BGP
2705 && CHECK_FLAG(re
->flags
, ZEBRA_FLAG_IBGP
))
2709 /* Lookup route node.*/
2710 rn
= srcdest_rnode_get(table
, p
, src_p
);
2713 * If same type of route are installed, treat it as a implicit
2715 * If the user has specified the No route replace semantics
2716 * for the install don't do a route replace.
2718 RNODE_FOREACH_RE (rn
, same
) {
2719 if (CHECK_FLAG(same
->status
, ROUTE_ENTRY_REMOVED
))
2722 if (same
->type
!= re
->type
)
2724 if (same
->instance
!= re
->instance
)
2726 if (same
->type
== ZEBRA_ROUTE_KERNEL
2727 && same
->metric
!= re
->metric
)
2730 if (CHECK_FLAG(re
->flags
, ZEBRA_FLAG_RR_USE_DISTANCE
) &&
2731 same
->distance
!= re
->distance
)
2735 * We should allow duplicate connected routes
2736 * because of IPv6 link-local routes and unnumbered
2737 * interfaces on Linux.
2739 if (same
->type
!= ZEBRA_ROUTE_CONNECT
)
2743 /* If this route is kernel route, set FIB flag to the route. */
2744 if (RIB_SYSTEM_ROUTE(re
)) {
2745 SET_FLAG(re
->status
, ROUTE_ENTRY_INSTALLED
);
2746 for (nexthop
= re
->ng
.nexthop
; nexthop
; nexthop
= nexthop
->next
)
2747 SET_FLAG(nexthop
->flags
, NEXTHOP_FLAG_FIB
);
2750 /* Link new re to node.*/
2751 if (IS_ZEBRA_DEBUG_RIB
) {
2752 rnode_debug(rn
, re
->vrf_id
,
2753 "Inserting route rn %p, re %p (%s) existing %p",
2754 rn
, re
, zebra_route_string(re
->type
), same
);
2756 if (IS_ZEBRA_DEBUG_RIB_DETAILED
)
2757 route_entry_dump(p
, src_p
, re
);
2759 rib_addnode(rn
, re
, 1);
2762 /* Free implicit route.*/
2764 rib_delnode(rn
, same
);
2768 route_unlock_node(rn
);
2772 void rib_delete(afi_t afi
, safi_t safi
, vrf_id_t vrf_id
, int type
,
2773 unsigned short instance
, int flags
, struct prefix
*p
,
2774 struct prefix_ipv6
*src_p
, const struct nexthop
*nh
,
2775 uint32_t table_id
, uint32_t metric
, uint8_t distance
,
2778 struct route_table
*table
;
2779 struct route_node
*rn
;
2780 struct route_entry
*re
;
2781 struct route_entry
*fib
= NULL
;
2782 struct route_entry
*same
= NULL
;
2783 struct nexthop
*rtnh
;
2784 char buf2
[INET6_ADDRSTRLEN
];
2787 assert(!src_p
|| !src_p
->prefixlen
|| afi
== AFI_IP6
);
2790 table
= zebra_vrf_table_with_table_id(afi
, safi
, vrf_id
, table_id
);
2797 apply_mask_ipv6(src_p
);
2799 /* Lookup route node. */
2800 rn
= srcdest_rnode_lookup(table
, p
, src_p
);
2802 char dst_buf
[PREFIX_STRLEN
], src_buf
[PREFIX_STRLEN
];
2804 prefix2str(p
, dst_buf
, sizeof(dst_buf
));
2805 if (src_p
&& src_p
->prefixlen
)
2806 prefix2str(src_p
, src_buf
, sizeof(src_buf
));
2810 if (IS_ZEBRA_DEBUG_RIB
)
2811 zlog_debug("%u:%s%s%s doesn't exist in rib", vrf_id
,
2813 (src_buf
[0] != '\0') ? " from " : "",
2818 dest
= rib_dest_from_rnode(rn
);
2819 fib
= dest
->selected_fib
;
2821 /* Lookup same type route. */
2822 RNODE_FOREACH_RE (rn
, re
) {
2823 if (CHECK_FLAG(re
->status
, ROUTE_ENTRY_REMOVED
))
2826 if (re
->type
!= type
)
2828 if (re
->instance
!= instance
)
2830 if (CHECK_FLAG(re
->flags
, ZEBRA_FLAG_RR_USE_DISTANCE
) &&
2831 distance
!= re
->distance
)
2834 if (re
->type
== ZEBRA_ROUTE_KERNEL
&& re
->metric
!= metric
)
2836 if (re
->type
== ZEBRA_ROUTE_CONNECT
&& (rtnh
= re
->ng
.nexthop
)
2837 && rtnh
->type
== NEXTHOP_TYPE_IFINDEX
&& nh
) {
2838 if (rtnh
->ifindex
!= nh
->ifindex
)
2843 /* Make sure that the route found has the same gateway. */
2849 for (ALL_NEXTHOPS(re
->ng
, rtnh
))
2850 if (nexthop_same_no_recurse(rtnh
, nh
)) {
2858 /* If same type of route can't be found and this message is from
2862 * In the past(HA!) we could get here because
2863 * we were receiving a route delete from the
2864 * kernel and we're not marking the proto
2865 * as coming from it's appropriate originator.
2866 * Now that we are properly noticing the fact
2867 * that the kernel has deleted our route we
2868 * are not going to get called in this path
2869 * I am going to leave this here because
2870 * this might still work this way on non-linux
2871 * platforms as well as some weird state I have
2872 * not properly thought of yet.
2873 * If we can show that this code path is
2874 * dead then we can remove it.
2876 if (fib
&& CHECK_FLAG(flags
, ZEBRA_FLAG_SELFROUTE
)) {
2877 if (IS_ZEBRA_DEBUG_RIB
) {
2878 rnode_debug(rn
, vrf_id
,
2879 "rn %p, re %p (%s) was deleted from kernel, adding",
2881 zebra_route_string(fib
->type
));
2884 UNSET_FLAG(fib
->status
, ROUTE_ENTRY_INSTALLED
);
2886 for (rtnh
= fib
->ng
.nexthop
; rtnh
;
2888 UNSET_FLAG(rtnh
->flags
,
2892 * This is a non FRR route
2893 * as such we should mark
2896 dest
->selected_fib
= NULL
;
2898 /* This means someone else, other than Zebra,
2900 * a Zebra router from the kernel. We will add
2902 rib_install_kernel(rn
, fib
, NULL
);
2905 if (IS_ZEBRA_DEBUG_RIB
) {
2909 "via %s ifindex %d type %d "
2910 "doesn't exist in rib",
2911 inet_ntop(afi2family(afi
),
2918 "type %d doesn't exist in rib",
2921 route_unlock_node(rn
);
2927 if (fromkernel
&& CHECK_FLAG(flags
, ZEBRA_FLAG_SELFROUTE
)
2929 rib_install_kernel(rn
, same
, NULL
);
2930 route_unlock_node(rn
);
2935 if (CHECK_FLAG(flags
, ZEBRA_FLAG_EVPN_ROUTE
)) {
2936 struct nexthop
*tmp_nh
;
2938 for (ALL_NEXTHOPS(re
->ng
, tmp_nh
)) {
2939 struct ipaddr vtep_ip
;
2941 memset(&vtep_ip
, 0, sizeof(struct ipaddr
));
2942 if (afi
== AFI_IP
) {
2943 vtep_ip
.ipa_type
= IPADDR_V4
;
2944 memcpy(&(vtep_ip
.ipaddr_v4
),
2945 &(tmp_nh
->gate
.ipv4
),
2946 sizeof(struct in_addr
));
2948 vtep_ip
.ipa_type
= IPADDR_V6
;
2949 memcpy(&(vtep_ip
.ipaddr_v6
),
2950 &(tmp_nh
->gate
.ipv6
),
2951 sizeof(struct in6_addr
));
2953 zebra_vxlan_evpn_vrf_route_del(re
->vrf_id
,
2957 rib_delnode(rn
, same
);
2960 route_unlock_node(rn
);
2965 int rib_add(afi_t afi
, safi_t safi
, vrf_id_t vrf_id
, int type
,
2966 unsigned short instance
, int flags
, struct prefix
*p
,
2967 struct prefix_ipv6
*src_p
, const struct nexthop
*nh
,
2968 uint32_t table_id
, uint32_t metric
, uint32_t mtu
, uint8_t distance
,
2971 struct route_entry
*re
;
2972 struct nexthop
*nexthop
;
2974 /* Allocate new route_entry structure. */
2975 re
= XCALLOC(MTYPE_RE
, sizeof(struct route_entry
));
2977 re
->instance
= instance
;
2978 re
->distance
= distance
;
2980 re
->metric
= metric
;
2982 re
->table
= table_id
;
2983 re
->vrf_id
= vrf_id
;
2984 re
->nexthop_num
= 0;
2985 re
->uptime
= time(NULL
);
2989 nexthop
= nexthop_new();
2991 route_entry_nexthop_add(re
, nexthop
);
2993 return rib_add_multipath(afi
, safi
, p
, src_p
, re
);
2996 /* Schedule routes of a particular table (address-family) based on event. */
2997 void rib_update_table(struct route_table
*table
, rib_update_event_t event
)
2999 struct route_node
*rn
;
3000 struct route_entry
*re
, *next
;
3002 /* Walk all routes and queue for processing, if appropriate for
3003 * the trigger event.
3005 for (rn
= route_top(table
); rn
; rn
= srcdest_route_next(rn
)) {
3007 * If we are looking at a route node and the node
3008 * has already been queued we don't
3009 * need to queue it up again
3011 if (rn
->info
&& CHECK_FLAG(rib_dest_from_rnode(rn
)->flags
,
3012 RIB_ROUTE_ANY_QUEUED
))
3015 case RIB_UPDATE_IF_CHANGE
:
3016 /* Examine all routes that won't get processed by the
3018 * triggered by nexthop evaluation (NHT). This would be
3020 * kernel and certain static routes. Note that NHT will
3022 * triggered upon an interface event as connected routes
3024 * get queued for processing.
3026 RNODE_FOREACH_RE_SAFE (rn
, re
, next
) {
3029 if (re
->type
!= ZEBRA_ROUTE_SYSTEM
3030 && re
->type
!= ZEBRA_ROUTE_KERNEL
3031 && re
->type
!= ZEBRA_ROUTE_CONNECT
3032 && re
->type
!= ZEBRA_ROUTE_STATIC
)
3035 if (re
->type
!= ZEBRA_ROUTE_STATIC
) {
3040 for (nh
= re
->ng
.nexthop
; nh
; nh
= nh
->next
)
3041 if (!(nh
->type
== NEXTHOP_TYPE_IPV4
3042 || nh
->type
== NEXTHOP_TYPE_IPV6
))
3045 /* If we only have nexthops to a
3054 case RIB_UPDATE_RMAP_CHANGE
:
3055 case RIB_UPDATE_OTHER
:
3056 /* Right now, examine all routes. Can restrict to a
3058 * some cases (TODO).
3060 if (rnode_to_ribs(rn
))
3070 /* RIB update function. */
3071 void rib_update(vrf_id_t vrf_id
, rib_update_event_t event
)
3073 struct route_table
*table
;
3075 /* Process routes of interested address-families. */
3076 table
= zebra_vrf_table(AFI_IP
, SAFI_UNICAST
, vrf_id
);
3078 if (IS_ZEBRA_DEBUG_EVENT
)
3079 zlog_debug("%s : AFI_IP event %d", __func__
, event
);
3080 rib_update_table(table
, event
);
3083 table
= zebra_vrf_table(AFI_IP6
, SAFI_UNICAST
, vrf_id
);
3085 if (IS_ZEBRA_DEBUG_EVENT
)
3086 zlog_debug("%s : AFI_IP6 event %d", __func__
, event
);
3087 rib_update_table(table
, event
);
3091 /* Delete self installed routes after zebra is relaunched. */
3092 void rib_sweep_table(struct route_table
*table
)
3094 struct route_node
*rn
;
3095 struct route_entry
*re
;
3096 struct route_entry
*next
;
3097 struct nexthop
*nexthop
;
3102 for (rn
= route_top(table
); rn
; rn
= srcdest_route_next(rn
)) {
3103 RNODE_FOREACH_RE_SAFE (rn
, re
, next
) {
3104 if (IS_ZEBRA_DEBUG_RIB
)
3105 route_entry_dump(&rn
->p
, NULL
, re
);
3107 if (CHECK_FLAG(re
->status
, ROUTE_ENTRY_REMOVED
))
3110 if (!CHECK_FLAG(re
->flags
, ZEBRA_FLAG_SELFROUTE
))
3114 * So we are starting up and have received
3115 * routes from the kernel that we have installed
3116 * from a previous run of zebra but not cleaned
3117 * up ( say a kill -9 )
3118 * But since we haven't actually installed
3119 * them yet( we received them from the kernel )
3120 * we don't think they are active.
3121 * So let's pretend they are active to actually
3123 * In all honesty I'm not sure if we should
3124 * mark them as active when we receive them
3125 * This is startup only so probably ok.
3127 * If we ever decide to move rib_sweep_table
3128 * to a different spot (ie startup )
3129 * this decision needs to be revisited
3131 SET_FLAG(re
->status
, ROUTE_ENTRY_INSTALLED
);
3132 for (ALL_NEXTHOPS(re
->ng
, nexthop
))
3133 SET_FLAG(nexthop
->flags
, NEXTHOP_FLAG_FIB
);
3135 rib_uninstall_kernel(rn
, re
);
3136 rib_delnode(rn
, re
);
3141 /* Sweep all RIB tables. */
3142 void rib_sweep_route(void)
3145 struct zebra_vrf
*zvrf
;
3147 RB_FOREACH (vrf
, vrf_id_head
, &vrfs_by_id
) {
3148 if ((zvrf
= vrf
->info
) == NULL
)
3151 rib_sweep_table(zvrf
->table
[AFI_IP
][SAFI_UNICAST
]);
3152 rib_sweep_table(zvrf
->table
[AFI_IP6
][SAFI_UNICAST
]);
3155 zebra_router_sweep_route();
3158 /* Remove specific by protocol routes from 'table'. */
3159 unsigned long rib_score_proto_table(uint8_t proto
, unsigned short instance
,
3160 struct route_table
*table
)
3162 struct route_node
*rn
;
3163 struct route_entry
*re
;
3164 struct route_entry
*next
;
3165 unsigned long n
= 0;
3168 for (rn
= route_top(table
); rn
; rn
= srcdest_route_next(rn
))
3169 RNODE_FOREACH_RE_SAFE (rn
, re
, next
) {
3170 if (CHECK_FLAG(re
->status
, ROUTE_ENTRY_REMOVED
))
3172 if (re
->type
== proto
3173 && re
->instance
== instance
) {
3174 rib_delnode(rn
, re
);
3181 /* Remove specific by protocol routes. */
3182 unsigned long rib_score_proto(uint8_t proto
, unsigned short instance
)
3185 struct zebra_vrf
*zvrf
;
3186 unsigned long cnt
= 0;
3188 RB_FOREACH (vrf
, vrf_id_head
, &vrfs_by_id
)
3189 if ((zvrf
= vrf
->info
) != NULL
)
3190 cnt
+= rib_score_proto_table(
3192 zvrf
->table
[AFI_IP
][SAFI_UNICAST
])
3193 + rib_score_proto_table(
3195 zvrf
->table
[AFI_IP6
][SAFI_UNICAST
]);
3197 cnt
+= zebra_router_score_proto(proto
, instance
);
3202 /* Close RIB and clean up kernel routes. */
3203 void rib_close_table(struct route_table
*table
)
3205 struct route_node
*rn
;
3206 rib_table_info_t
*info
;
3212 info
= route_table_get_info(table
);
3214 for (rn
= route_top(table
); rn
; rn
= srcdest_route_next(rn
)) {
3215 dest
= rib_dest_from_rnode(rn
);
3217 if (dest
&& dest
->selected_fib
) {
3218 if (info
->safi
== SAFI_UNICAST
)
3219 hook_call(rib_update
, rn
, NULL
);
3221 if (!RIB_SYSTEM_ROUTE(dest
->selected_fib
)) {
3222 rib_uninstall_kernel(rn
, dest
->selected_fib
);
3223 dest
->selected_fib
= NULL
;
3230 * Handle results from the dataplane system. Dequeue update context
3231 * structs, dispatch to appropriate internal handlers.
3233 static int rib_process_dplane_results(struct thread
*thread
)
3235 struct zebra_dplane_ctx
*ctx
;
3236 struct dplane_ctx_q ctxlist
;
3238 /* Dequeue a list of completed updates with one lock/unlock cycle */
3240 /* TODO -- dequeue a list with one lock/unlock cycle? */
3243 TAILQ_INIT(&ctxlist
);
3245 /* Take lock controlling queue of results */
3246 pthread_mutex_lock(&dplane_mutex
);
3248 /* Dequeue list of context structs */
3249 dplane_ctx_list_append(&ctxlist
, &rib_dplane_q
);
3251 pthread_mutex_unlock(&dplane_mutex
);
3253 /* Dequeue context block */
3254 ctx
= dplane_ctx_dequeue(&ctxlist
);
3256 /* If we've emptied the results queue, we're done */
3261 switch (dplane_ctx_get_op(ctx
)) {
3262 case DPLANE_OP_ROUTE_INSTALL
:
3263 case DPLANE_OP_ROUTE_UPDATE
:
3264 case DPLANE_OP_ROUTE_DELETE
:
3265 rib_process_result(ctx
);
3268 case DPLANE_OP_LSP_INSTALL
:
3269 case DPLANE_OP_LSP_UPDATE
:
3270 case DPLANE_OP_LSP_DELETE
:
3271 zebra_mpls_lsp_dplane_result(ctx
);
3275 /* Don't expect this: just return the struct? */
3276 dplane_ctx_fini(&ctx
);
3278 } /* Dispatch by op code */
3280 ctx
= dplane_ctx_dequeue(&ctxlist
);
3285 /* Check for nexthop tracking processing after finishing with results */
3286 do_nht_processing();
3292 * Results are returned from the dataplane subsystem, in the context of
3293 * the dataplane pthread. We enqueue the results here for processing by
3294 * the main thread later.
3296 static int rib_dplane_results(struct dplane_ctx_q
*ctxlist
)
3298 /* Take lock controlling queue of results */
3299 pthread_mutex_lock(&dplane_mutex
);
3301 /* Enqueue context blocks */
3302 dplane_ctx_list_append(&rib_dplane_q
, ctxlist
);
3304 pthread_mutex_unlock(&dplane_mutex
);
3306 /* Ensure event is signalled to zebra main pthread */
3307 thread_add_event(zebrad
.master
, rib_process_dplane_results
, NULL
, 0,
3313 /* Routing information base initialize. */
3316 rib_queue_init(&zebrad
);
3318 /* Init dataplane, and register for results */
3319 pthread_mutex_init(&dplane_mutex
, NULL
);
3320 TAILQ_INIT(&rib_dplane_q
);
3321 zebra_dplane_init(rib_dplane_results
);
3327 * Get the first vrf id that is greater than the given vrf id if any.
3329 * Returns TRUE if a vrf id was found, FALSE otherwise.
3331 static inline int vrf_id_get_next(vrf_id_t vrf_id
, vrf_id_t
*next_id_p
)
3335 vrf
= vrf_lookup_by_id(vrf_id
);
3337 vrf
= RB_NEXT(vrf_id_head
, vrf
);
3339 *next_id_p
= vrf
->vrf_id
;
3348 * rib_tables_iter_next
3350 * Returns the next table in the iteration.
3352 struct route_table
*rib_tables_iter_next(rib_tables_iter_t
*iter
)
3354 struct route_table
*table
;
3357 * Array that helps us go over all AFI/SAFI combinations via one
3364 {AFI_IP
, SAFI_UNICAST
}, {AFI_IP
, SAFI_MULTICAST
},
3365 {AFI_IP
, SAFI_LABELED_UNICAST
}, {AFI_IP6
, SAFI_UNICAST
},
3366 {AFI_IP6
, SAFI_MULTICAST
}, {AFI_IP6
, SAFI_LABELED_UNICAST
},
3371 switch (iter
->state
) {
3373 case RIB_TABLES_ITER_S_INIT
:
3374 iter
->vrf_id
= VRF_DEFAULT
;
3375 iter
->afi_safi_ix
= -1;
3379 case RIB_TABLES_ITER_S_ITERATING
:
3380 iter
->afi_safi_ix
++;
3383 while (iter
->afi_safi_ix
3384 < (int)ZEBRA_NUM_OF(afi_safis
)) {
3385 table
= zebra_vrf_table(
3386 afi_safis
[iter
->afi_safi_ix
].afi
,
3387 afi_safis
[iter
->afi_safi_ix
].safi
,
3392 iter
->afi_safi_ix
++;
3396 * Found another table in this vrf.
3402 * Done with all tables in the current vrf, go to the
3406 if (!vrf_id_get_next(iter
->vrf_id
, &iter
->vrf_id
))
3409 iter
->afi_safi_ix
= 0;
3414 case RIB_TABLES_ITER_S_DONE
:
3419 iter
->state
= RIB_TABLES_ITER_S_ITERATING
;
3421 iter
->state
= RIB_TABLES_ITER_S_DONE
;