3 * Copyright (C) 2010 Google Inc.
5 * This file is part of Quagga
7 * Quagga is free software; you can redistribute it and/or modify it
8 * under the terms of the GNU General Public License as published by the
9 * Free Software Foundation; either version 2, or (at your option) any
12 * Quagga is distributed in the hope that it will be useful, but
13 * WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * General Public License for more details.
17 * You should have received a copy of the GNU General Public License
18 * along with Quagga; see the file COPYING. If not, write to the Free
19 * Software Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA
28 #include "sockunion.h"
33 #include "bgpd/bgpd.h"
34 #include "bgpd/bgp_table.h"
35 #include "bgpd/bgp_route.h"
36 #include "bgpd/bgp_attr.h"
37 #include "bgpd/bgp_debug.h"
38 #include "bgpd/bgp_aspath.h"
39 #include "bgpd/bgp_community.h"
40 #include "bgpd/bgp_ecommunity.h"
41 #include "bgpd/bgp_mpath.h"
44 * bgp_maximum_paths_set
46 * Record maximum-paths configuration for BGP instance
49 bgp_maximum_paths_set (struct bgp
*bgp
, afi_t afi
, safi_t safi
,
50 int peertype
, u_int16_t maxpaths
, u_int16_t options
)
52 if (!bgp
|| (afi
>= AFI_MAX
) || (safi
>= SAFI_MAX
))
58 bgp
->maxpaths
[afi
][safi
].maxpaths_ibgp
= maxpaths
;
59 bgp
->maxpaths
[afi
][safi
].ibgp_flags
|= options
;
62 bgp
->maxpaths
[afi
][safi
].maxpaths_ebgp
= maxpaths
;
72 * bgp_maximum_paths_unset
74 * Remove maximum-paths configuration from BGP instance
77 bgp_maximum_paths_unset (struct bgp
*bgp
, afi_t afi
, safi_t safi
,
80 if (!bgp
|| (afi
>= AFI_MAX
) || (safi
>= SAFI_MAX
))
86 bgp
->maxpaths
[afi
][safi
].maxpaths_ibgp
= MULTIPATH_NUM
;
87 bgp
->maxpaths
[afi
][safi
].ibgp_flags
= 0;
90 bgp
->maxpaths
[afi
][safi
].maxpaths_ebgp
= MULTIPATH_NUM
;
100 * bgp_info_nexthop_cmp
102 * Compare the nexthops of two paths. Return value is less than, equal to,
103 * or greater than zero if bi1 is respectively less than, equal to,
104 * or greater than bi2.
107 bgp_info_nexthop_cmp (struct bgp_info
*bi1
, struct bgp_info
*bi2
)
109 struct attr_extra
*ae1
, *ae2
;
112 ae1
= bi1
->attr
->extra
;
113 ae2
= bi2
->attr
->extra
;
115 compare
= IPV4_ADDR_CMP (&bi1
->attr
->nexthop
, &bi2
->attr
->nexthop
);
116 if (!compare
&& ae1
&& ae2
)
118 if (ae1
->mp_nexthop_len
== ae2
->mp_nexthop_len
)
120 switch (ae1
->mp_nexthop_len
)
122 case BGP_ATTR_NHLEN_IPV4
:
123 case BGP_ATTR_NHLEN_VPNV4
:
124 compare
= IPV4_ADDR_CMP (&ae1
->mp_nexthop_global_in
,
125 &ae2
->mp_nexthop_global_in
);
128 case BGP_ATTR_NHLEN_IPV6_GLOBAL
:
129 case BGP_ATTR_NHLEN_VPNV6_GLOBAL
:
130 compare
= IPV6_ADDR_CMP (&ae1
->mp_nexthop_global
,
131 &ae2
->mp_nexthop_global
);
133 case BGP_ATTR_NHLEN_IPV6_GLOBAL_AND_LL
:
134 compare
= IPV6_ADDR_CMP (&ae1
->mp_nexthop_global
,
135 &ae2
->mp_nexthop_global
);
137 compare
= IPV6_ADDR_CMP (&ae1
->mp_nexthop_local
,
138 &ae2
->mp_nexthop_local
);
140 #endif /* HAVE_IPV6 */
145 /* This can happen if one IPv6 peer sends you global and link-local
146 * nexthops but another IPv6 peer only sends you global
148 else if (ae1
->mp_nexthop_len
== BGP_ATTR_NHLEN_IPV6_GLOBAL
||
149 ae1
->mp_nexthop_len
== BGP_ATTR_NHLEN_IPV6_GLOBAL_AND_LL
)
151 compare
= IPV6_ADDR_CMP (&ae1
->mp_nexthop_global
,
152 &ae2
->mp_nexthop_global
);
155 if (ae1
->mp_nexthop_len
< ae2
->mp_nexthop_len
)
161 #endif /* HAVE_IPV6 */
170 * This function determines our multipath list ordering. By ordering
171 * the list we can deterministically select which paths are included
172 * in the multipath set. The ordering also helps in detecting changes
173 * in the multipath selection so we can detect whether to send an
176 * The order of paths is determined first by received nexthop, and then
177 * by peer address if the nexthops are the same.
180 bgp_info_mpath_cmp (void *val1
, void *val2
)
182 struct bgp_info
*bi1
, *bi2
;
188 compare
= bgp_info_nexthop_cmp (bi1
, bi2
);
192 if (!bi1
->peer
->su_remote
&& !bi2
->peer
->su_remote
)
194 else if (!bi1
->peer
->su_remote
)
196 else if (!bi2
->peer
->su_remote
)
199 compare
= sockunion_cmp (bi1
->peer
->su_remote
, bi2
->peer
->su_remote
);
208 * Initialize the mp_list, which holds the list of multipaths
209 * selected by bgp_best_selection
212 bgp_mp_list_init (struct list
*mp_list
)
215 memset (mp_list
, 0, sizeof (struct list
));
216 mp_list
->cmp
= bgp_info_mpath_cmp
;
222 * Clears all entries out of the mp_list
225 bgp_mp_list_clear (struct list
*mp_list
)
228 list_delete_all_node (mp_list
);
234 * Adds a multipath entry to the mp_list
237 bgp_mp_list_add (struct list
*mp_list
, struct bgp_info
*mpinfo
)
239 assert (mp_list
&& mpinfo
);
240 listnode_add_sort (mp_list
, mpinfo
);
246 * Allocate and zero memory for a new bgp_info_mpath element
248 static struct bgp_info_mpath
*
249 bgp_info_mpath_new (void)
251 struct bgp_info_mpath
*new_mpath
;
252 new_mpath
= XCALLOC (MTYPE_BGP_MPATH_INFO
, sizeof (struct bgp_info_mpath
));
257 * bgp_info_mpath_free
259 * Release resources for a bgp_info_mpath element and zero out pointer
262 bgp_info_mpath_free (struct bgp_info_mpath
**mpath
)
266 if ((*mpath
)->mp_attr
)
267 bgp_attr_unintern (&(*mpath
)->mp_attr
);
268 XFREE (MTYPE_BGP_MPATH_INFO
, *mpath
);
276 * Fetch the mpath element for the given bgp_info. Used for
277 * doing lazy allocation.
279 static struct bgp_info_mpath
*
280 bgp_info_mpath_get (struct bgp_info
*binfo
)
282 struct bgp_info_mpath
*mpath
;
285 mpath
= bgp_info_mpath_new();
288 binfo
->mpath
= mpath
;
289 mpath
->mp_info
= binfo
;
295 * bgp_info_mpath_enqueue
297 * Enqueue a path onto the multipath list given the previous multipath
301 bgp_info_mpath_enqueue (struct bgp_info
*prev_info
, struct bgp_info
*binfo
)
303 struct bgp_info_mpath
*prev
, *mpath
;
305 prev
= bgp_info_mpath_get (prev_info
);
306 mpath
= bgp_info_mpath_get (binfo
);
310 mpath
->mp_next
= prev
->mp_next
;
311 mpath
->mp_prev
= prev
;
313 prev
->mp_next
->mp_prev
= mpath
;
314 prev
->mp_next
= mpath
;
316 SET_FLAG (binfo
->flags
, BGP_INFO_MULTIPATH
);
320 * bgp_info_mpath_dequeue
322 * Remove a path from the multipath list
325 bgp_info_mpath_dequeue (struct bgp_info
*binfo
)
327 struct bgp_info_mpath
*mpath
= binfo
->mpath
;
331 mpath
->mp_prev
->mp_next
= mpath
->mp_next
;
333 mpath
->mp_next
->mp_prev
= mpath
->mp_prev
;
334 mpath
->mp_next
= mpath
->mp_prev
= NULL
;
335 UNSET_FLAG (binfo
->flags
, BGP_INFO_MULTIPATH
);
339 * bgp_info_mpath_next
341 * Given a bgp_info, return the next multipath entry
344 bgp_info_mpath_next (struct bgp_info
*binfo
)
346 if (!binfo
->mpath
|| !binfo
->mpath
->mp_next
)
348 return binfo
->mpath
->mp_next
->mp_info
;
352 * bgp_info_mpath_first
354 * Given bestpath bgp_info, return the first multipath entry.
357 bgp_info_mpath_first (struct bgp_info
*binfo
)
359 return bgp_info_mpath_next (binfo
);
363 * bgp_info_mpath_count
365 * Given the bestpath bgp_info, return the number of multipath entries
368 bgp_info_mpath_count (struct bgp_info
*binfo
)
372 return binfo
->mpath
->mp_count
;
376 * bgp_info_mpath_count_set
378 * Sets the count of multipaths into bestpath's mpath element
381 bgp_info_mpath_count_set (struct bgp_info
*binfo
, u_int32_t count
)
383 struct bgp_info_mpath
*mpath
;
384 if (!count
&& !binfo
->mpath
)
386 mpath
= bgp_info_mpath_get (binfo
);
389 mpath
->mp_count
= count
;
393 * bgp_info_mpath_attr
395 * Given bestpath bgp_info, return aggregated attribute set used
396 * for advertising the multipath route
399 bgp_info_mpath_attr (struct bgp_info
*binfo
)
403 return binfo
->mpath
->mp_attr
;
407 * bgp_info_mpath_attr_set
409 * Sets the aggregated attribute into bestpath's mpath element
412 bgp_info_mpath_attr_set (struct bgp_info
*binfo
, struct attr
*attr
)
414 struct bgp_info_mpath
*mpath
;
415 if (!attr
&& !binfo
->mpath
)
417 mpath
= bgp_info_mpath_get (binfo
);
420 mpath
->mp_attr
= attr
;
424 * bgp_info_mpath_update
426 * Compare and sync up the multipath list with the mp_list generated by
430 bgp_info_mpath_update (struct bgp_node
*rn
, struct bgp_info
*new_best
,
431 struct bgp_info
*old_best
, struct list
*mp_list
,
432 struct bgp_maxpaths_cfg
*mpath_cfg
)
434 u_int16_t maxpaths
, mpath_count
, old_mpath_count
;
435 struct listnode
*mp_node
, *mp_next_node
;
436 struct bgp_info
*cur_mpath
, *new_mpath
, *next_mpath
, *prev_mpath
;
437 int mpath_changed
, debug
;
438 char pfx_buf
[PREFIX2STR_BUFFER
], nh_buf
[2][INET6_ADDRSTRLEN
];
439 char path_buf
[PATH_ADDPATH_STR_BUFFER
];
442 maxpaths
= MULTIPATH_NUM
;
446 prev_mpath
= new_best
;
447 mp_node
= listhead (mp_list
);
448 debug
= bgp_debug_bestpath(&rn
->p
);
451 prefix2str (&rn
->p
, pfx_buf
, sizeof (pfx_buf
));
456 if (new_best
!= old_best
)
457 bgp_info_mpath_dequeue (new_best
);
458 maxpaths
= (new_best
->peer
->sort
== BGP_PEER_IBGP
) ?
459 mpath_cfg
->maxpaths_ibgp
: mpath_cfg
->maxpaths_ebgp
;
464 cur_mpath
= bgp_info_mpath_first (old_best
);
465 old_mpath_count
= bgp_info_mpath_count (old_best
);
466 bgp_info_mpath_count_set (old_best
, 0);
467 bgp_info_mpath_dequeue (old_best
);
471 zlog_debug("%s: starting mpath update, newbest %s num candidates %d old-mpath-count %d",
472 pfx_buf
, new_best
? new_best
->peer
->host
: "NONE",
473 listcount (mp_list
), old_mpath_count
);
476 * We perform an ordered walk through both lists in parallel.
477 * The reason for the ordered walk is that if there are paths
478 * that were previously multipaths and are still multipaths, the walk
479 * should encounter them in both lists at the same time. Otherwise
480 * there will be paths that are in one list or another, and we
481 * will deal with these separately.
483 * Note that new_best might be somewhere in the mp_list, so we need
486 while (mp_node
|| cur_mpath
)
488 struct bgp_info
*tmp_info
;
491 * We can bail out of this loop if all existing paths on the
492 * multipath list have been visited (for cleanup purposes) and
493 * the maxpath requirement is fulfulled
495 if (!cur_mpath
&& (mpath_count
>= maxpaths
))
498 mp_next_node
= mp_node
? listnextnode (mp_node
) : NULL
;
499 next_mpath
= cur_mpath
? bgp_info_mpath_next (cur_mpath
) : NULL
;
500 tmp_info
= mp_node
? listgetdata (mp_node
) : NULL
;
503 zlog_debug("%s: comparing candidate %s with existing mpath %s",
504 pfx_buf
, tmp_info
? tmp_info
->peer
->host
: "NONE",
505 cur_mpath
? cur_mpath
->peer
->host
: "NONE");
508 * If equal, the path was a multipath and is still a multipath.
509 * Insert onto new multipath list if maxpaths allows.
511 if (mp_node
&& (listgetdata (mp_node
) == cur_mpath
))
513 list_delete_node (mp_list
, mp_node
);
514 bgp_info_mpath_dequeue (cur_mpath
);
515 if ((mpath_count
< maxpaths
) &&
516 bgp_info_nexthop_cmp (prev_mpath
, cur_mpath
))
518 bgp_info_mpath_enqueue (prev_mpath
, cur_mpath
);
519 prev_mpath
= cur_mpath
;
523 bgp_info_path_with_addpath_rx_str(cur_mpath
, path_buf
);
524 zlog_debug("%s: %s is still multipath, cur count %d",
525 pfx_buf
, path_buf
, mpath_count
);
533 bgp_info_path_with_addpath_rx_str(cur_mpath
, path_buf
);
534 zlog_debug ("%s: remove mpath %s nexthop %s, cur count %d",
536 inet_ntop (AF_INET
, &cur_mpath
->attr
->nexthop
,
537 nh_buf
[0], sizeof (nh_buf
[0])),
541 mp_node
= mp_next_node
;
542 cur_mpath
= next_mpath
;
546 if (cur_mpath
&& (!mp_node
||
547 (bgp_info_mpath_cmp (cur_mpath
,
548 listgetdata (mp_node
)) < 0)))
551 * If here, we have an old multipath and either the mp_list
552 * is finished or the next mp_node points to a later
553 * multipath, so we need to purge this path from the
556 bgp_info_mpath_dequeue (cur_mpath
);
560 bgp_info_path_with_addpath_rx_str(cur_mpath
, path_buf
);
561 zlog_debug ("%s: remove mpath %s nexthop %s, cur count %d",
563 inet_ntop (AF_INET
, &cur_mpath
->attr
->nexthop
,
564 nh_buf
[0], sizeof (nh_buf
[0])),
567 cur_mpath
= next_mpath
;
572 * If here, we have a path on the mp_list that was not previously
573 * a multipath (due to non-equivalance or maxpaths exceeded),
574 * or the matching multipath is sorted later in the multipath
575 * list. Before we enqueue the path on the new multipath list,
576 * make sure its not on the old_best multipath list or referenced
578 * - If next_mpath points to this new path, update next_mpath to
579 * point to the multipath after this one
580 * - Dequeue the path from the multipath list just to make sure
582 new_mpath
= listgetdata (mp_node
);
583 list_delete_node (mp_list
, mp_node
);
584 if ((mpath_count
< maxpaths
) && (new_mpath
!= new_best
) &&
585 bgp_info_nexthop_cmp (prev_mpath
, new_mpath
))
587 if (new_mpath
== next_mpath
)
588 bgp_info_mpath_next (new_mpath
);
589 bgp_info_mpath_dequeue (new_mpath
);
591 bgp_info_mpath_enqueue (prev_mpath
, new_mpath
);
592 prev_mpath
= new_mpath
;
597 bgp_info_path_with_addpath_rx_str(new_mpath
, path_buf
);
598 zlog_debug ("%s: add mpath %s nexthop %s, cur count %d",
600 inet_ntop (AF_INET
, &new_mpath
->attr
->nexthop
,
601 nh_buf
[0], sizeof (nh_buf
[0])),
605 mp_node
= mp_next_node
;
612 zlog_debug("%s: New mpath count (incl newbest) %d mpath-change %s",
613 pfx_buf
, mpath_count
, mpath_changed
? "YES" : "NO");
615 bgp_info_mpath_count_set (new_best
, mpath_count
-1);
616 if (mpath_changed
|| (bgp_info_mpath_count (new_best
) != old_mpath_count
))
617 SET_FLAG (new_best
->flags
, BGP_INFO_MULTIPATH_CHG
);
622 * bgp_mp_dmed_deselect
624 * Clean up multipath information for BGP_INFO_DMED_SELECTED path that
625 * is not selected as best path
628 bgp_mp_dmed_deselect (struct bgp_info
*dmed_best
)
630 struct bgp_info
*mpinfo
, *mpnext
;
635 for (mpinfo
= bgp_info_mpath_first (dmed_best
); mpinfo
; mpinfo
= mpnext
)
637 mpnext
= bgp_info_mpath_next (mpinfo
);
638 bgp_info_mpath_dequeue (mpinfo
);
641 bgp_info_mpath_count_set (dmed_best
, 0);
642 UNSET_FLAG (dmed_best
->flags
, BGP_INFO_MULTIPATH_CHG
);
643 assert (bgp_info_mpath_first (dmed_best
) == 0);
647 * bgp_info_mpath_aggregate_update
649 * Set the multipath aggregate attribute. We need to see if the
650 * aggregate has changed and then set the ATTR_CHANGED flag on the
651 * bestpath info so that a peer update will be generated. The
652 * change is detected by generating the current attribute,
653 * interning it, and then comparing the interned pointer with the
654 * current value. We can skip this generate/compare step if there
655 * is no change in multipath selection and no attribute change in
659 bgp_info_mpath_aggregate_update (struct bgp_info
*new_best
,
660 struct bgp_info
*old_best
)
662 struct bgp_info
*mpinfo
;
663 struct aspath
*aspath
;
664 struct aspath
*asmerge
;
665 struct attr
*new_attr
, *old_attr
;
667 struct community
*community
, *commerge
;
668 struct ecommunity
*ecomm
, *ecommerge
;
669 struct attr_extra
*ae
;
670 struct attr attr
= { 0 };
672 if (old_best
&& (old_best
!= new_best
) &&
673 (old_attr
= bgp_info_mpath_attr (old_best
)))
675 bgp_attr_unintern (&old_attr
);
676 bgp_info_mpath_attr_set (old_best
, NULL
);
682 if (!bgp_info_mpath_count (new_best
))
684 if ((new_attr
= bgp_info_mpath_attr (new_best
)))
686 bgp_attr_unintern (&new_attr
);
687 bgp_info_mpath_attr_set (new_best
, NULL
);
688 SET_FLAG (new_best
->flags
, BGP_INFO_ATTR_CHANGED
);
693 bgp_attr_dup (&attr
, new_best
->attr
);
695 if (new_best
->peer
&&
696 bgp_flag_check (new_best
->peer
->bgp
, BGP_FLAG_MULTIPATH_RELAX_AS_SET
))
699 /* aggregate attribute from multipath constituents */
700 aspath
= aspath_dup (attr
.aspath
);
701 origin
= attr
.origin
;
702 community
= attr
.community
? community_dup (attr
.community
) : NULL
;
704 ecomm
= (ae
&& ae
->ecommunity
) ? ecommunity_dup (ae
->ecommunity
) : NULL
;
706 for (mpinfo
= bgp_info_mpath_first (new_best
); mpinfo
;
707 mpinfo
= bgp_info_mpath_next (mpinfo
))
709 asmerge
= aspath_aggregate (aspath
, mpinfo
->attr
->aspath
);
710 aspath_free (aspath
);
713 if (origin
< mpinfo
->attr
->origin
)
714 origin
= mpinfo
->attr
->origin
;
716 if (mpinfo
->attr
->community
)
720 commerge
= community_merge (community
, mpinfo
->attr
->community
);
721 community
= community_uniq_sort (commerge
);
722 community_free (commerge
);
725 community
= community_dup (mpinfo
->attr
->community
);
728 ae
= mpinfo
->attr
->extra
;
729 if (ae
&& ae
->ecommunity
)
733 ecommerge
= ecommunity_merge (ecomm
, ae
->ecommunity
);
734 ecomm
= ecommunity_uniq_sort (ecommerge
);
735 ecommunity_free (&ecommerge
);
738 ecomm
= ecommunity_dup (ae
->ecommunity
);
742 attr
.aspath
= aspath
;
743 attr
.origin
= origin
;
746 attr
.community
= community
;
747 attr
.flag
|= ATTR_FLAG_BIT (BGP_ATTR_COMMUNITIES
);
751 ae
= bgp_attr_extra_get (&attr
);
752 ae
->ecommunity
= ecomm
;
753 attr
.flag
|= ATTR_FLAG_BIT (BGP_ATTR_EXT_COMMUNITIES
);
756 /* Zap multipath attr nexthop so we set nexthop to self */
757 attr
.nexthop
.s_addr
= 0;
760 memset (&attr
.extra
->mp_nexthop_global
, 0, sizeof (struct in6_addr
));
761 #endif /* HAVE_IPV6 */
763 /* TODO: should we set ATOMIC_AGGREGATE and AGGREGATOR? */
766 new_attr
= bgp_attr_intern (&attr
);
767 bgp_attr_extra_free (&attr
);
769 if (new_attr
!= bgp_info_mpath_attr (new_best
))
771 if ((old_attr
= bgp_info_mpath_attr (new_best
)))
772 bgp_attr_unintern (&old_attr
);
773 bgp_info_mpath_attr_set (new_best
, new_attr
);
774 SET_FLAG (new_best
->flags
, BGP_INFO_ATTR_CHANGED
);
777 bgp_attr_unintern (&new_attr
);