2 * BIRD -- BGP Attributes
4 * (c) 2000 Martin Mares <mj@ucw.cz>
6 * Can be freely distributed and used under the terms of the GNU GPL.
13 #include "nest/bird.h"
14 #include "nest/iface.h"
15 #include "nest/protocol.h"
16 #include "nest/route.h"
17 #include "nest/attrs.h"
18 #include "conf/conf.h"
19 #include "lib/resource.h"
20 #include "lib/string.h"
21 #include "lib/unaligned.h"
26 * UPDATE message error handling
28 * All checks from RFC 4271 6.3 are done as specified with these exceptions:
29 * - The semantic check of an IP address from NEXT_HOP attribute is missing.
30 * - Checks of some optional attribute values are missing.
31 * - Syntactic and semantic checks of NLRIs (done in DECODE_PREFIX())
32 * are probably inadequate.
34 * Loop detection based on AS_PATH causes updates to be withdrawn. RFC
35 * 4271 does not explicitly specifiy the behavior in that case.
37 * Loop detection related to route reflection (based on ORIGINATOR_ID
38 * and CLUSTER_LIST) causes updates to be withdrawn. RFC 4456 8
39 * specifies that such updates should be ignored, but that is generally
42 * Error checking of optional transitive attributes is done according to
43 * draft-ietf-idr-optional-transitive-03, but errors are handled always
46 * Unexpected AS_CONFED_* segments in AS_PATH are logged and removed,
47 * but unknown segments cause a session drop with Malformed AS_PATH
48 * error (see validate_path()). The behavior in such case is not
49 * explicitly specified by RFC 4271. RFC 5065 specifies that
50 * inconsistent AS_CONFED_* segments should cause a session drop, but
51 * implementations that pass invalid AS_CONFED_* segments are
54 * Error handling of AS4_* attributes is done as specified by
55 * draft-ietf-idr-rfc4893bis-03. There are several possible
56 * inconsistencies between AGGREGATOR and AS4_AGGREGATOR that are not
57 * handled by that draft, these are logged and ignored (see
58 * bgp_reconstruct_4b_attrs()).
62 static byte bgp_mandatory_attrs
[] = { BA_ORIGIN
, BA_AS_PATH
74 int (*validate
)(struct bgp_proto
*p
, byte
*attr
, int len
);
75 void (*format
)(eattr
*ea
, byte
*buf
, int buflen
);
82 bgp_check_origin(struct bgp_proto
*p UNUSED
, byte
*a
, int len UNUSED
)
90 bgp_format_origin(eattr
*a
, byte
*buf
, int buflen UNUSED
)
92 static char *bgp_origin_names
[] = { "IGP", "EGP", "Incomplete" };
94 bsprintf(buf
, bgp_origin_names
[a
->u
.data
]);
98 path_segment_contains(byte
*p
, int bs
, u32 asn
)
106 u32 asn2
= (bs
== 4) ? get_u32(p
) : get_u16(p
);
115 /* Validates path attribute, removes AS_CONFED_* segments, and also returns path length */
117 validate_path(struct bgp_proto
*p
, int as_path
, int bs
, byte
*idata
, uint
*ilength
)
131 plen
= 2 + bs
* a
[1];
137 log(L_WARN
"%s: %s_PATH attribute contains empty segment, skipping it",
138 p
->p
.name
, as_path
? "AS" : "AS4");
148 case AS_PATH_SEQUENCE
:
152 case AS_PATH_CONFED_SEQUENCE
:
153 case AS_PATH_CONFED_SET
:
154 if (as_path
&& path_segment_contains(a
, bs
, p
->remote_as
))
156 log(L_WARN
"%s: AS_CONFED_* segment with peer ASN found, misconfigured confederation?", p
->p
.name
);
160 log(L_WARN
"%s: %s_PATH attribute contains AS_CONFED_* segment, skipping segment",
161 p
->p
.name
, as_path
? "AS" : "AS4");
169 memmove(dst
, a
, plen
);
177 *ilength
= dst
- idata
;
182 validate_as_path(struct bgp_proto
*p
, byte
*a
, int *len
)
184 return validate_path(p
, 1, p
->as4_session
? 4 : 2, a
, len
);
188 validate_as4_path(struct bgp_proto
*p
, struct adata
*path
)
190 return validate_path(p
, 0, 4, path
->data
, &path
->length
);
194 bgp_check_next_hop(struct bgp_proto
*p UNUSED
, byte
*a
, int len
)
201 memcpy(&addr
, a
, len
);
203 if (ipa_classify(addr
) & IADDR_HOST
)
211 bgp_format_next_hop(eattr
*a
, byte
*buf
, int buflen UNUSED
)
213 ip_addr
*ipp
= (ip_addr
*) a
->u
.ptr
->data
;
215 /* in IPv6, we might have two addresses in NEXT HOP */
216 if ((a
->u
.ptr
->length
== NEXT_HOP_LENGTH
) && ipa_nonzero(ipp
[1]))
218 bsprintf(buf
, "%I %I", ipp
[0], ipp
[1]);
223 bsprintf(buf
, "%I", ipp
[0]);
227 bgp_check_aggregator(struct bgp_proto
*p
, byte
*a UNUSED
, int len
)
229 int exp_len
= p
->as4_session
? 8 : 6;
231 return (len
== exp_len
) ? 0 : WITHDRAW
;
235 bgp_format_aggregator(eattr
*a
, byte
*buf
, int buflen UNUSED
)
237 struct adata
*ad
= a
->u
.ptr
;
238 byte
*data
= ad
->data
;
244 bsprintf(buf
, "%d.%d.%d.%d AS%u", data
[0], data
[1], data
[2], data
[3], as
);
248 bgp_check_community(struct bgp_proto
*p UNUSED
, byte
*a UNUSED
, int len
)
250 return ((len
% 4) == 0) ? 0 : WITHDRAW
;
254 bgp_check_cluster_list(struct bgp_proto
*p UNUSED
, byte
*a UNUSED
, int len
)
256 return ((len
% 4) == 0) ? 0 : 5;
260 bgp_format_cluster_list(eattr
*a
, byte
*buf
, int buflen
)
262 /* Truncates cluster lists larger than buflen, probably not a problem */
263 int_set_format(a
->u
.ptr
, 0, -1, buf
, buflen
);
267 bgp_check_reach_nlri(struct bgp_proto
*p UNUSED
, byte
*a UNUSED
, int len UNUSED
)
270 p
->mp_reach_start
= a
;
271 p
->mp_reach_len
= len
;
277 bgp_check_unreach_nlri(struct bgp_proto
*p UNUSED
, byte
*a UNUSED
, int len UNUSED
)
280 p
->mp_unreach_start
= a
;
281 p
->mp_unreach_len
= len
;
287 bgp_check_ext_community(struct bgp_proto
*p UNUSED
, byte
*a UNUSED
, int len
)
289 return ((len
% 8) == 0) ? 0 : WITHDRAW
;
293 static struct attr_desc bgp_attr_table
[] = {
294 { NULL
, -1, 0, 0, 0, /* Undefined */
296 { "origin", 1, BAF_TRANSITIVE
, EAF_TYPE_INT
, 1, /* BA_ORIGIN */
297 bgp_check_origin
, bgp_format_origin
},
298 { "as_path", -1, BAF_TRANSITIVE
, EAF_TYPE_AS_PATH
, 1, /* BA_AS_PATH */
299 NULL
, NULL
}, /* is checked by validate_as_path() as a special case */
300 { "next_hop", 4, BAF_TRANSITIVE
, EAF_TYPE_IP_ADDRESS
, 1, /* BA_NEXT_HOP */
301 bgp_check_next_hop
, bgp_format_next_hop
},
302 { "med", 4, BAF_OPTIONAL
, EAF_TYPE_INT
, 1, /* BA_MULTI_EXIT_DISC */
304 { "local_pref", 4, BAF_TRANSITIVE
, EAF_TYPE_INT
, 0, /* BA_LOCAL_PREF */
306 { "atomic_aggr", 0, BAF_TRANSITIVE
, EAF_TYPE_OPAQUE
, 1, /* BA_ATOMIC_AGGR */
308 { "aggregator", -1, BAF_OPTIONAL
| BAF_TRANSITIVE
, EAF_TYPE_OPAQUE
, 1, /* BA_AGGREGATOR */
309 bgp_check_aggregator
, bgp_format_aggregator
},
310 { "community", -1, BAF_OPTIONAL
| BAF_TRANSITIVE
, EAF_TYPE_INT_SET
, 1, /* BA_COMMUNITY */
311 bgp_check_community
, NULL
},
312 { "originator_id", 4, BAF_OPTIONAL
, EAF_TYPE_ROUTER_ID
, 0, /* BA_ORIGINATOR_ID */
314 { "cluster_list", -1, BAF_OPTIONAL
, EAF_TYPE_INT_SET
, 0, /* BA_CLUSTER_LIST */
315 bgp_check_cluster_list
, bgp_format_cluster_list
},
316 { .name
= NULL
}, /* BA_DPA */
317 { .name
= NULL
}, /* BA_ADVERTISER */
318 { .name
= NULL
}, /* BA_RCID_PATH */
319 { "mp_reach_nlri", -1, BAF_OPTIONAL
, EAF_TYPE_OPAQUE
, 1, /* BA_MP_REACH_NLRI */
320 bgp_check_reach_nlri
, NULL
},
321 { "mp_unreach_nlri", -1, BAF_OPTIONAL
, EAF_TYPE_OPAQUE
, 1, /* BA_MP_UNREACH_NLRI */
322 bgp_check_unreach_nlri
, NULL
},
323 { "ext_community", -1, BAF_OPTIONAL
| BAF_TRANSITIVE
, EAF_TYPE_EC_SET
, 1, /* BA_EXT_COMMUNITY */
324 bgp_check_ext_community
, NULL
},
325 { "as4_path", -1, BAF_OPTIONAL
| BAF_TRANSITIVE
, EAF_TYPE_OPAQUE
, 1, /* BA_AS4_PATH */
327 { "as4_aggregator", -1, BAF_OPTIONAL
| BAF_TRANSITIVE
, EAF_TYPE_OPAQUE
, 1, /* BA_AS4_PATH */
331 /* BA_AS4_PATH is type EAF_TYPE_OPAQUE and not type EAF_TYPE_AS_PATH.
332 * It does not matter as this attribute does not appear on routes in the routing table.
335 #define ATTR_KNOWN(code) ((code) < ARRAY_SIZE(bgp_attr_table) && bgp_attr_table[code].name)
337 static inline struct adata
*
338 bgp_alloc_adata(struct linpool
*pool
, unsigned len
)
340 struct adata
*ad
= lp_alloc(pool
, sizeof(struct adata
) + len
);
346 bgp_set_attr(eattr
*e
, unsigned attr
, uintptr_t val
)
348 ASSERT(ATTR_KNOWN(attr
));
349 e
->id
= EA_CODE(EAP_BGP
, attr
);
350 e
->type
= bgp_attr_table
[attr
].type
;
351 e
->flags
= bgp_attr_table
[attr
].expected_flags
;
352 if (e
->type
& EAF_EMBEDDED
)
355 e
->u
.ptr
= (struct adata
*) val
;
359 bgp_set_attr_wa(eattr
*e
, struct linpool
*pool
, unsigned attr
, unsigned len
)
361 struct adata
*ad
= bgp_alloc_adata(pool
, len
);
362 bgp_set_attr(e
, attr
, (uintptr_t) ad
);
367 bgp_attach_attr(ea_list
**to
, struct linpool
*pool
, unsigned attr
, uintptr_t val
)
369 ea_list
*a
= lp_alloc(pool
, sizeof(ea_list
) + sizeof(eattr
));
372 a
->flags
= EALF_SORTED
;
374 bgp_set_attr(a
->attrs
, attr
, val
);
378 bgp_attach_attr_wa(ea_list
**to
, struct linpool
*pool
, unsigned attr
, unsigned len
)
380 struct adata
*ad
= bgp_alloc_adata(pool
, len
);
381 bgp_attach_attr(to
, pool
, attr
, (uintptr_t) ad
);
386 bgp_encode_attr_hdr(byte
*dst
, uint flags
, unsigned code
, int len
)
390 DBG("\tAttribute %02x (%d bytes, flags %02x)\n", code
, len
, flags
);
401 *dst
++ = flags
| BAF_EXT_LEN
;
411 aggregator_convert_to_old(struct adata
*aggr
, byte
*dst
, int *new_used
)
413 byte
*src
= aggr
->data
;
416 u32 as
= get_u32(src
);
424 /* Copy IPv4 address */
425 memcpy(dst
+ 2, src
+ 4, 4);
429 aggregator_convert_to_new(struct adata
*aggr
, byte
*dst
)
431 byte
*src
= aggr
->data
;
433 u32 as
= get_u16(src
);
436 /* Copy IPv4 address */
437 memcpy(dst
+ 4, src
+ 2, 4);
441 bgp_get_attr_len(eattr
*a
)
444 if (ATTR_KNOWN(EA_ID(a
->id
)))
446 int code
= EA_ID(a
->id
);
447 struct attr_desc
*desc
= &bgp_attr_table
[code
];
448 len
= desc
->expected_length
;
451 ASSERT(!(a
->type
& EAF_EMBEDDED
));
452 len
= a
->u
.ptr
->length
;
457 ASSERT((a
->type
& EAF_TYPE_MASK
) == EAF_TYPE_OPAQUE
);
458 len
= a
->u
.ptr
->length
;
464 #define ADVANCE(w, r, l) do { r -= l; w += l; } while (0)
467 * bgp_encode_attrs - encode BGP attributes
470 * @attrs: a list of extended attributes
471 * @remains: remaining space in the buffer
473 * The bgp_encode_attrs() function takes a list of extended attributes
474 * and converts it to its BGP representation (a part of an Update message).
476 * Result: Length of the attribute block generated or -1 if not enough space.
479 bgp_encode_attrs(struct bgp_proto
*p
, byte
*w
, ea_list
*attrs
, int remains
)
481 uint i
, code
, type
, flags
;
485 for(i
=0; i
<attrs
->count
; i
++)
487 eattr
*a
= &attrs
->attrs
[i
];
488 ASSERT(EA_PROTO(a
->id
) == EAP_BGP
);
492 /* When talking multiprotocol BGP, the NEXT_HOP attributes are used only temporarily. */
493 if (code
== BA_NEXT_HOP
)
497 /* When AS4-aware BGP speaker is talking to non-AS4-aware BGP speaker,
498 * we have to convert our 4B AS_PATH to 2B AS_PATH and send our AS_PATH
499 * as optional AS4_PATH attribute.
501 if ((code
== BA_AS_PATH
) && (! p
->as4_session
))
503 len
= a
->u
.ptr
->length
;
505 if (remains
< (len
+ 4))
508 /* Using temporary buffer because don't know a length of created attr
509 * and therefore a length of a header. Perhaps i should better always
510 * use BAF_EXT_LEN. */
514 int nl
= as_path_convert_to_old(a
->u
.ptr
, buf
, &new_used
);
516 DBG("BGP: Encoding old AS_PATH\n");
517 rv
= bgp_encode_attr_hdr(w
, BAF_TRANSITIVE
, BA_AS_PATH
, nl
);
518 ADVANCE(w
, remains
, rv
);
520 ADVANCE(w
, remains
, nl
);
525 if (remains
< (len
+ 4))
528 /* We should discard AS_CONFED_SEQUENCE or AS_CONFED_SET path segments
529 * here but we don't support confederations and such paths we already
530 * discarded in bgp_check_as_path().
533 DBG("BGP: Encoding AS4_PATH\n");
534 rv
= bgp_encode_attr_hdr(w
, BAF_OPTIONAL
| BAF_TRANSITIVE
, BA_AS4_PATH
, len
);
535 ADVANCE(w
, remains
, rv
);
536 memcpy(w
, a
->u
.ptr
->data
, len
);
537 ADVANCE(w
, remains
, len
);
542 /* The same issue with AGGREGATOR attribute */
543 if ((code
== BA_AGGREGATOR
) && (! p
->as4_session
))
548 if (remains
< (len
+ 3))
551 rv
= bgp_encode_attr_hdr(w
, BAF_OPTIONAL
| BAF_TRANSITIVE
, BA_AGGREGATOR
, len
);
552 ADVANCE(w
, remains
, rv
);
553 aggregator_convert_to_old(a
->u
.ptr
, w
, &new_used
);
554 ADVANCE(w
, remains
, len
);
560 if (remains
< (len
+ 3))
563 rv
= bgp_encode_attr_hdr(w
, BAF_OPTIONAL
| BAF_TRANSITIVE
, BA_AS4_AGGREGATOR
, len
);
564 ADVANCE(w
, remains
, rv
);
565 memcpy(w
, a
->u
.ptr
->data
, len
);
566 ADVANCE(w
, remains
, len
);
571 /* Standard path continues here ... */
573 type
= a
->type
& EAF_TYPE_MASK
;
574 flags
= a
->flags
& (BAF_OPTIONAL
| BAF_TRANSITIVE
| BAF_PARTIAL
);
575 len
= bgp_get_attr_len(a
);
577 /* Skip empty sets */
578 if (((type
== EAF_TYPE_INT_SET
) || (type
== EAF_TYPE_EC_SET
)) && (len
== 0))
581 if (remains
< len
+ 4)
584 rv
= bgp_encode_attr_hdr(w
, flags
, code
, len
);
585 ADVANCE(w
, remains
, rv
);
590 case EAF_TYPE_ROUTER_ID
:
592 put_u32(w
, a
->u
.data
);
596 case EAF_TYPE_IP_ADDRESS
:
598 ip_addr ip
= *(ip_addr
*)a
->u
.ptr
->data
;
603 case EAF_TYPE_INT_SET
:
604 case EAF_TYPE_EC_SET
:
606 u32
*z
= int_set_get_data(a
->u
.ptr
);
608 for(i
=0; i
<len
; i
+=4)
612 case EAF_TYPE_OPAQUE
:
613 case EAF_TYPE_AS_PATH
:
614 memcpy(w
, a
->u
.ptr
->data
, len
);
617 bug("bgp_encode_attrs: unknown attribute type %02x", a
->type
);
619 ADVANCE(w
, remains
, len
);
629 bgp_init_prefix(struct fib_node *N)
631 struct bgp_prefix *p = (struct bgp_prefix *) N;
632 p->bucket_node.next = NULL;
637 bgp_compare_u32(const u32
*x
, const u32
*y
)
639 return (*x
< *y
) ? -1 : (*x
> *y
) ? 1 : 0;
643 bgp_normalize_int_set(u32
*dest
, u32
*src
, unsigned cnt
)
645 memcpy(dest
, src
, sizeof(u32
) * cnt
);
646 qsort(dest
, cnt
, sizeof(u32
), (int(*)(const void *, const void *)) bgp_compare_u32
);
650 bgp_compare_ec(const u32
*xp
, const u32
*yp
)
652 u64 x
= ec_get(xp
, 0);
653 u64 y
= ec_get(yp
, 0);
654 return (x
< y
) ? -1 : (x
> y
) ? 1 : 0;
658 bgp_normalize_ec_set(struct adata
*ad
, u32
*src
, int internal
)
660 u32
*dst
= int_set_get_data(ad
);
662 /* Remove non-transitive communities (EC_TBIT active) on external sessions */
665 int len
= int_set_get_size(ad
);
669 for (i
=0; i
< len
; i
+= 2)
671 if (src
[i
] & EC_TBIT
)
678 ad
->length
= (t
- dst
) * 4;
681 memcpy(dst
, src
, ad
->length
);
683 qsort(dst
, ad
->length
/ 8, 8, (int(*)(const void *, const void *)) bgp_compare_ec
);
687 bgp_rehash_buckets(struct bgp_proto
*p
)
689 struct bgp_bucket
**old
= p
->bucket_hash
;
690 struct bgp_bucket
**new;
691 unsigned oldn
= p
->hash_size
;
693 struct bgp_bucket
*b
;
695 p
->hash_size
= p
->hash_limit
;
696 DBG("BGP: Rehashing bucket table from %d to %d\n", oldn
, p
->hash_size
);
698 if (p
->hash_limit
>= 65536)
700 new = p
->bucket_hash
= mb_allocz(p
->p
.pool
, p
->hash_size
* sizeof(struct bgp_bucket
*));
701 mask
= p
->hash_size
- 1;
702 for (i
=0; i
<oldn
; i
++)
705 old
[i
] = b
->hash_next
;
707 b
->hash_next
= new[e
];
709 b
->hash_next
->hash_prev
= b
;
716 static struct bgp_bucket
*
717 bgp_new_bucket(struct bgp_proto
*p
, ea_list
*new, unsigned hash
)
719 struct bgp_bucket
*b
;
720 unsigned ea_size
= sizeof(ea_list
) + new->count
* sizeof(eattr
);
721 unsigned ea_size_aligned
= BIRD_ALIGN(ea_size
, CPU_STRUCT_ALIGN
);
722 unsigned size
= sizeof(struct bgp_bucket
) + ea_size_aligned
;
725 unsigned index
= hash
& (p
->hash_size
- 1);
727 /* Gather total size of non-inline attributes */
728 for (i
=0; i
<new->count
; i
++)
730 eattr
*a
= &new->attrs
[i
];
731 if (!(a
->type
& EAF_EMBEDDED
))
732 size
+= BIRD_ALIGN(sizeof(struct adata
) + a
->u
.ptr
->length
, CPU_STRUCT_ALIGN
);
735 /* Create the bucket and hash it */
736 b
= mb_alloc(p
->p
.pool
, size
);
737 b
->hash_next
= p
->bucket_hash
[index
];
739 b
->hash_next
->hash_prev
= b
;
740 p
->bucket_hash
[index
] = b
;
743 add_tail(&p
->bucket_queue
, &b
->send_node
);
744 init_list(&b
->prefixes
);
745 memcpy(b
->eattrs
, new, ea_size
);
746 dest
= ((byte
*)b
->eattrs
) + ea_size_aligned
;
748 /* Copy values of non-inline attributes */
749 for (i
=0; i
<new->count
; i
++)
751 eattr
*a
= &b
->eattrs
->attrs
[i
];
752 if (!(a
->type
& EAF_EMBEDDED
))
754 struct adata
*oa
= a
->u
.ptr
;
755 struct adata
*na
= (struct adata
*) dest
;
756 memcpy(na
, oa
, sizeof(struct adata
) + oa
->length
);
758 dest
+= BIRD_ALIGN(sizeof(struct adata
) + na
->length
, CPU_STRUCT_ALIGN
);
762 /* If needed, rehash */
764 if (p
->hash_count
> p
->hash_limit
)
765 bgp_rehash_buckets(p
);
770 static struct bgp_bucket
*
771 bgp_get_bucket(struct bgp_proto
*p
, net
*n
, ea_list
*attrs
, int originate
)
774 unsigned i
, cnt
, hash
, code
;
777 struct bgp_bucket
*b
;
779 /* Merge the attribute list */
780 new = alloca(ea_scan(attrs
));
781 ea_merge(attrs
, new);
784 /* Normalize attributes */
791 if (EA_PROTO(a
->id
) != EAP_BGP
)
794 if (ATTR_KNOWN(code
))
796 if (!bgp_attr_table
[code
].allow_in_ebgp
&& !p
->is_internal
)
798 /* The flags might have been zero if the attr was added by filters */
799 a
->flags
= (a
->flags
& BAF_PARTIAL
) | bgp_attr_table
[code
].expected_flags
;
805 /* Don't re-export unknown non-transitive attributes */
806 if (!(a
->flags
& BAF_TRANSITIVE
))
810 if ((d
->type
& EAF_ORIGINATED
) && !originate
&& (d
->flags
& BAF_TRANSITIVE
) && (d
->flags
& BAF_OPTIONAL
))
811 d
->flags
|= BAF_PARTIAL
;
812 switch (d
->type
& EAF_TYPE_MASK
)
814 case EAF_TYPE_INT_SET
:
816 struct adata
*z
= alloca(sizeof(struct adata
) + d
->u
.ptr
->length
);
817 z
->length
= d
->u
.ptr
->length
;
818 bgp_normalize_int_set((u32
*) z
->data
, (u32
*) d
->u
.ptr
->data
, z
->length
/ 4);
822 case EAF_TYPE_EC_SET
:
824 struct adata
*z
= alloca(sizeof(struct adata
) + d
->u
.ptr
->length
);
825 z
->length
= d
->u
.ptr
->length
;
826 bgp_normalize_ec_set(z
, (u32
*) d
->u
.ptr
->data
, p
->is_internal
);
838 for(b
=p
->bucket_hash
[hash
& (p
->hash_size
- 1)]; b
; b
=b
->hash_next
)
839 if (b
->hash
== hash
&& ea_same(b
->eattrs
, new))
841 DBG("Found bucket.\n");
845 /* Ensure that there are all mandatory attributes */
846 for(i
=0; i
<ARRAY_SIZE(bgp_mandatory_attrs
); i
++)
847 if (!(seen
& (1 << bgp_mandatory_attrs
[i
])))
849 log(L_ERR
"%s: Mandatory attribute %s missing in route %I/%d", p
->p
.name
, bgp_attr_table
[bgp_mandatory_attrs
[i
]].name
, n
->n
.prefix
, n
->n
.pxlen
);
853 /* Check if next hop is valid */
854 a
= ea_find(new, EA_CODE(EAP_BGP
, BA_NEXT_HOP
));
855 if (!a
|| ipa_equal(p
->cf
->remote_ip
, *(ip_addr
*)a
->u
.ptr
->data
))
857 log(L_ERR
"%s: Invalid NEXT_HOP attribute in route %I/%d", p
->p
.name
, n
->n
.prefix
, n
->n
.pxlen
);
861 /* Create new bucket */
862 DBG("Creating bucket.\n");
863 return bgp_new_bucket(p
, new, hash
);
867 bgp_free_bucket(struct bgp_proto
*p
, struct bgp_bucket
*buck
)
870 buck
->hash_next
->hash_prev
= buck
->hash_prev
;
872 buck
->hash_prev
->hash_next
= buck
->hash_next
;
874 p
->bucket_hash
[buck
->hash
& (p
->hash_size
-1)] = buck
->hash_next
;
879 /* Prefix hash table */
881 #define PXH_KEY(n1) n1->n.prefix, n1->n.pxlen, n1->path_id
882 #define PXH_NEXT(n) n->next
883 #define PXH_EQ(p1,l1,i1,p2,l2,i2) ipa_equal(p1, p2) && l1 == l2 && i1 == i2
884 #define PXH_FN(p,l,i) ipa_hash32(p) ^ u32_hash((l << 16) ^ i)
886 #define PXH_REHASH bgp_pxh_rehash
887 #define PXH_PARAMS /8, *2, 2, 2, 8, 20
890 HASH_DEFINE_REHASH_FN(PXH
, struct bgp_prefix
)
893 bgp_init_prefix_table(struct bgp_proto
*p
, u32 order
)
895 HASH_INIT(p
->prefix_hash
, p
->p
.pool
, order
);
897 p
->prefix_slab
= sl_new(p
->p
.pool
, sizeof(struct bgp_prefix
));
900 static struct bgp_prefix
*
901 bgp_get_prefix(struct bgp_proto
*p
, ip_addr prefix
, int pxlen
, u32 path_id
)
903 struct bgp_prefix
*bp
= HASH_FIND(p
->prefix_hash
, PXH
, prefix
, pxlen
, path_id
);
908 bp
= sl_alloc(p
->prefix_slab
);
909 bp
->n
.prefix
= prefix
;
911 bp
->path_id
= path_id
;
912 bp
->bucket_node
.next
= NULL
;
914 HASH_INSERT2(p
->prefix_hash
, PXH
, p
->p
.pool
, bp
);
920 bgp_free_prefix(struct bgp_proto
*p
, struct bgp_prefix
*bp
)
922 HASH_REMOVE2(p
->prefix_hash
, PXH
, p
->p
.pool
, bp
);
923 sl_free(p
->prefix_slab
, bp
);
928 bgp_rt_notify(struct proto
*P
, rtable
*tbl UNUSED
, net
*n
, rte
*new, rte
*old UNUSED
, ea_list
*attrs
)
930 struct bgp_proto
*p
= (struct bgp_proto
*) P
;
931 struct bgp_bucket
*buck
;
932 struct bgp_prefix
*px
;
936 DBG("BGP: Got route %I/%d %s\n", n
->n
.prefix
, n
->n
.pxlen
, new ? "up" : "down");
941 buck
= bgp_get_bucket(p
, n
, attrs
, new->attrs
->source
!= RTS_BGP
);
942 if (!buck
) /* Inconsistent attribute list */
948 if (!(buck
= p
->withdraw_bucket
))
950 buck
= p
->withdraw_bucket
= mb_alloc(P
->pool
, sizeof(struct bgp_bucket
));
951 init_list(&buck
->prefixes
);
954 path_id
= p
->add_path_tx
? key
->attrs
->src
->global_id
: 0;
955 px
= bgp_get_prefix(p
, n
->n
.prefix
, n
->n
.pxlen
, path_id
);
956 if (px
->bucket_node
.next
)
958 DBG("\tRemoving old entry.\n");
959 rem_node(&px
->bucket_node
);
961 add_tail(&buck
->prefixes
, &px
->bucket_node
);
962 bgp_schedule_packet(p
->conn
, PKT_UPDATE
);
966 bgp_create_attrs(struct bgp_proto
*p
, rte
*e
, ea_list
**attrs
, struct linpool
*pool
)
968 ea_list
*ea
= lp_alloc(pool
, sizeof(ea_list
) + 4*sizeof(eattr
));
974 ea
->flags
= EALF_SORTED
;
977 bgp_set_attr(ea
->attrs
, BA_ORIGIN
,
978 ((rta
->source
== RTS_OSPF_EXT1
) || (rta
->source
== RTS_OSPF_EXT2
)) ? ORIGIN_INCOMPLETE
: ORIGIN_IGP
);
981 bgp_set_attr_wa(ea
->attrs
+1, pool
, BA_AS_PATH
, 0);
984 z
= bgp_set_attr_wa(ea
->attrs
+1, pool
, BA_AS_PATH
, 6);
985 z
[0] = AS_PATH_SEQUENCE
;
987 put_u32(z
+2, p
->local_as
);
990 /* iBGP -> use gw, eBGP multi-hop -> use source_addr,
991 eBGP single-hop -> use gw if on the same iface */
992 z
= bgp_set_attr_wa(ea
->attrs
+2, pool
, BA_NEXT_HOP
, NEXT_HOP_LENGTH
);
993 if (p
->cf
->next_hop_self
||
994 rta
->dest
!= RTD_ROUTER
||
995 ipa_equal(rta
->gw
, IPA_NONE
) ||
996 ipa_is_link_local(rta
->gw
) ||
997 (!p
->is_internal
&& !p
->cf
->next_hop_keep
&&
998 (!p
->neigh
|| (rta
->iface
!= p
->neigh
->iface
))))
999 set_next_hop(z
, p
->source_addr
);
1001 set_next_hop(z
, rta
->gw
);
1003 bgp_set_attr(ea
->attrs
+3, BA_LOCAL_PREF
, p
->cf
->default_local_pref
);
1005 return 0; /* Leave decision to the filters */
1010 bgp_as_path_loopy(struct bgp_proto
*p
, rta
*a
)
1012 int num
= p
->cf
->allow_local_as
+ 1;
1013 eattr
*e
= ea_find(a
->eattrs
, EA_CODE(EAP_BGP
, BA_AS_PATH
));
1014 return (e
&& (num
> 0) && as_path_contains(e
->u
.ptr
, p
->local_as
, num
));
1018 bgp_originator_id_loopy(struct bgp_proto
*p
, rta
*a
)
1020 eattr
*e
= ea_find(a
->eattrs
, EA_CODE(EAP_BGP
, BA_ORIGINATOR_ID
));
1021 return (e
&& (e
->u
.data
== p
->local_id
));
1025 bgp_cluster_list_loopy(struct bgp_proto
*p
, rta
*a
)
1027 eattr
*e
= ea_find(a
->eattrs
, EA_CODE(EAP_BGP
, BA_CLUSTER_LIST
));
1028 return (e
&& p
->rr_client
&& int_set_contains(e
->u
.ptr
, p
->rr_cluster_id
));
1033 bgp_path_prepend(rte
*e
, ea_list
**attrs
, struct linpool
*pool
, u32 as
)
1035 eattr
*a
= ea_find(e
->attrs
->eattrs
, EA_CODE(EAP_BGP
, BA_AS_PATH
));
1036 bgp_attach_attr(attrs
, pool
, BA_AS_PATH
, (uintptr_t) as_path_prepend(pool
, a
->u
.ptr
, as
));
1040 bgp_cluster_list_prepend(rte
*e
, ea_list
**attrs
, struct linpool
*pool
, u32 cid
)
1042 eattr
*a
= ea_find(e
->attrs
->eattrs
, EA_CODE(EAP_BGP
, BA_CLUSTER_LIST
));
1043 bgp_attach_attr(attrs
, pool
, BA_CLUSTER_LIST
, (uintptr_t) int_set_add(pool
, a
? a
->u
.ptr
: NULL
, cid
));
1047 bgp_update_attrs(struct bgp_proto
*p
, rte
*e
, ea_list
**attrs
, struct linpool
*pool
, int rr
)
1051 if (!p
->is_internal
&& !p
->rs_client
)
1053 bgp_path_prepend(e
, attrs
, pool
, p
->local_as
);
1055 /* The MULTI_EXIT_DISC attribute received from a neighboring AS MUST NOT be
1056 * propagated to other neighboring ASes.
1057 * Perhaps it would be better to undefine it.
1059 a
= ea_find(e
->attrs
->eattrs
, EA_CODE(EAP_BGP
, BA_MULTI_EXIT_DISC
));
1061 bgp_attach_attr(attrs
, pool
, BA_MULTI_EXIT_DISC
, 0);
1064 /* iBGP -> keep next_hop, eBGP multi-hop -> use source_addr,
1065 * eBGP single-hop -> keep next_hop if on the same iface.
1066 * If the next_hop is zero (i.e. link-local), keep only if on the same iface.
1068 * Note that same-iface-check uses iface from route, which is based on gw.
1070 a
= ea_find(e
->attrs
->eattrs
, EA_CODE(EAP_BGP
, BA_NEXT_HOP
));
1071 if (a
&& !p
->cf
->next_hop_self
&&
1072 (p
->cf
->next_hop_keep
||
1073 (p
->is_internal
&& ipa_nonzero(*((ip_addr
*) a
->u
.ptr
->data
))) ||
1074 (p
->neigh
&& (e
->attrs
->iface
== p
->neigh
->iface
))))
1076 /* Leave the original next hop attribute, will check later where does it point */
1080 /* Need to create new one */
1081 byte
*b
= bgp_attach_attr_wa(attrs
, pool
, BA_NEXT_HOP
, NEXT_HOP_LENGTH
);
1082 set_next_hop(b
, p
->source_addr
);
1087 /* Handling route reflection, RFC 4456 */
1088 struct bgp_proto
*src
= (struct bgp_proto
*) e
->attrs
->src
->proto
;
1090 a
= ea_find(e
->attrs
->eattrs
, EA_CODE(EAP_BGP
, BA_ORIGINATOR_ID
));
1092 bgp_attach_attr(attrs
, pool
, BA_ORIGINATOR_ID
, src
->remote_id
);
1094 /* We attach proper cluster ID according to whether the route is entering or leaving the cluster */
1095 bgp_cluster_list_prepend(e
, attrs
, pool
, src
->rr_client
? src
->rr_cluster_id
: p
->rr_cluster_id
);
1097 /* Two RR clients with different cluster ID, hmmm */
1098 if (src
->rr_client
&& p
->rr_client
&& (src
->rr_cluster_id
!= p
->rr_cluster_id
))
1099 bgp_cluster_list_prepend(e
, attrs
, pool
, p
->rr_cluster_id
);
1102 return 0; /* Leave decision to the filters */
1106 bgp_community_filter(struct bgp_proto
*p
, rte
*e
)
1111 /* Check if we aren't forbidden to export the route by communities */
1112 a
= ea_find(e
->attrs
->eattrs
, EA_CODE(EAP_BGP
, BA_COMMUNITY
));
1116 if (int_set_contains(d
, BGP_COMM_NO_ADVERTISE
))
1118 DBG("\tNO_ADVERTISE\n");
1121 if (!p
->is_internal
&&
1122 (int_set_contains(d
, BGP_COMM_NO_EXPORT
) ||
1123 int_set_contains(d
, BGP_COMM_NO_EXPORT_SUBCONFED
)))
1125 DBG("\tNO_EXPORT\n");
1134 bgp_import_control(struct proto
*P
, rte
**new, ea_list
**attrs
, struct linpool
*pool
)
1137 struct bgp_proto
*p
= (struct bgp_proto
*) P
;
1138 struct bgp_proto
*new_bgp
= (e
->attrs
->src
->proto
->proto
== &proto_bgp
) ?
1139 (struct bgp_proto
*) e
->attrs
->src
->proto
: NULL
;
1141 if (p
== new_bgp
) /* Poison reverse updates */
1145 /* We should check here for cluster list loop, because the receiving BGP instance
1146 might have different cluster ID */
1147 if (bgp_cluster_list_loopy(p
, e
->attrs
))
1150 if (p
->cf
->interpret_communities
&& bgp_community_filter(p
, e
))
1153 if (p
->local_as
== new_bgp
->local_as
&& p
->is_internal
&& new_bgp
->is_internal
)
1155 /* Redistribution of internal routes with IBGP */
1156 if (p
->rr_client
|| new_bgp
->rr_client
)
1157 /* Route reflection, RFC 4456 */
1158 return bgp_update_attrs(p
, e
, attrs
, pool
, 1);
1163 return bgp_update_attrs(p
, e
, attrs
, pool
, 0);
1166 return bgp_create_attrs(p
, e
, attrs
, pool
);
1170 bgp_get_neighbor(rte
*r
)
1172 eattr
*e
= ea_find(r
->attrs
->eattrs
, EA_CODE(EAP_BGP
, BA_AS_PATH
));
1175 if (e
&& as_path_get_first(e
->u
.ptr
, &as
))
1178 return ((struct bgp_proto
*) r
->attrs
->src
->proto
)->remote_as
;
1182 rte_resolvable(rte
*rt
)
1184 int rd
= rt
->attrs
->dest
;
1185 return (rd
== RTD_ROUTER
) || (rd
== RTD_DEVICE
) || (rd
== RTD_MULTIPATH
);
1189 bgp_rte_better(rte
*new, rte
*old
)
1191 struct bgp_proto
*new_bgp
= (struct bgp_proto
*) new->attrs
->src
->proto
;
1192 struct bgp_proto
*old_bgp
= (struct bgp_proto
*) old
->attrs
->src
->proto
;
1196 /* Skip suppressed routes (see bgp_rte_recalculate()) */
1197 n
= new->u
.bgp
.suppressed
;
1198 o
= old
->u
.bgp
.suppressed
;
1204 /* RFC 4271 9.1.2.1. Route resolvability test */
1205 n
= rte_resolvable(new);
1206 o
= rte_resolvable(old
);
1212 /* Start with local preferences */
1213 x
= ea_find(new->attrs
->eattrs
, EA_CODE(EAP_BGP
, BA_LOCAL_PREF
));
1214 y
= ea_find(old
->attrs
->eattrs
, EA_CODE(EAP_BGP
, BA_LOCAL_PREF
));
1215 n
= x
? x
->u
.data
: new_bgp
->cf
->default_local_pref
;
1216 o
= y
? y
->u
.data
: old_bgp
->cf
->default_local_pref
;
1222 /* RFC 4271 9.1.2.2. a) Use AS path lengths */
1223 if (new_bgp
->cf
->compare_path_lengths
|| old_bgp
->cf
->compare_path_lengths
)
1225 x
= ea_find(new->attrs
->eattrs
, EA_CODE(EAP_BGP
, BA_AS_PATH
));
1226 y
= ea_find(old
->attrs
->eattrs
, EA_CODE(EAP_BGP
, BA_AS_PATH
));
1227 n
= x
? as_path_getlen(x
->u
.ptr
) : AS_PATH_MAXLEN
;
1228 o
= y
? as_path_getlen(y
->u
.ptr
) : AS_PATH_MAXLEN
;
1235 /* RFC 4271 9.1.2.2. b) Use origins */
1236 x
= ea_find(new->attrs
->eattrs
, EA_CODE(EAP_BGP
, BA_ORIGIN
));
1237 y
= ea_find(old
->attrs
->eattrs
, EA_CODE(EAP_BGP
, BA_ORIGIN
));
1238 n
= x
? x
->u
.data
: ORIGIN_INCOMPLETE
;
1239 o
= y
? y
->u
.data
: ORIGIN_INCOMPLETE
;
1245 /* RFC 4271 9.1.2.2. c) Compare MED's */
1246 /* Proper RFC 4271 path selection cannot be interpreted as finding
1247 * the best path in some ordering. It is implemented partially in
1248 * bgp_rte_recalculate() when deterministic_med option is
1249 * active. Without that option, the behavior is just an
1250 * approximation, which in specific situations may lead to
1251 * persistent routing loops, because it is nondeterministic - it
1252 * depends on the order in which routes appeared. But it is also the
1253 * same behavior as used by default in Cisco routers, so it is
1254 * probably not a big issue.
1256 if (new_bgp
->cf
->med_metric
|| old_bgp
->cf
->med_metric
||
1257 (bgp_get_neighbor(new) == bgp_get_neighbor(old
)))
1259 x
= ea_find(new->attrs
->eattrs
, EA_CODE(EAP_BGP
, BA_MULTI_EXIT_DISC
));
1260 y
= ea_find(old
->attrs
->eattrs
, EA_CODE(EAP_BGP
, BA_MULTI_EXIT_DISC
));
1261 n
= x
? x
->u
.data
: new_bgp
->cf
->default_med
;
1262 o
= y
? y
->u
.data
: old_bgp
->cf
->default_med
;
1269 /* RFC 4271 9.1.2.2. d) Prefer external peers */
1270 if (new_bgp
->is_internal
> old_bgp
->is_internal
)
1272 if (new_bgp
->is_internal
< old_bgp
->is_internal
)
1275 /* RFC 4271 9.1.2.2. e) Compare IGP metrics */
1276 n
= new_bgp
->cf
->igp_metric
? new->attrs
->igp_metric
: 0;
1277 o
= old_bgp
->cf
->igp_metric
? old
->attrs
->igp_metric
: 0;
1283 /* RFC 4271 9.1.2.2. f) Compare BGP identifiers */
1284 /* RFC 4456 9. a) Use ORIGINATOR_ID instead of local neighor ID */
1285 x
= ea_find(new->attrs
->eattrs
, EA_CODE(EAP_BGP
, BA_ORIGINATOR_ID
));
1286 y
= ea_find(old
->attrs
->eattrs
, EA_CODE(EAP_BGP
, BA_ORIGINATOR_ID
));
1287 n
= x
? x
->u
.data
: new_bgp
->remote_id
;
1288 o
= y
? y
->u
.data
: old_bgp
->remote_id
;
1290 /* RFC 5004 - prefer older routes */
1291 /* (if both are external and from different peer) */
1292 if ((new_bgp
->cf
->prefer_older
|| old_bgp
->cf
->prefer_older
) &&
1293 !new_bgp
->is_internal
&& n
!= o
)
1296 /* rest of RFC 4271 9.1.2.2. f) */
1302 /* RFC 4456 9. b) Compare cluster list lengths */
1303 x
= ea_find(new->attrs
->eattrs
, EA_CODE(EAP_BGP
, BA_CLUSTER_LIST
));
1304 y
= ea_find(old
->attrs
->eattrs
, EA_CODE(EAP_BGP
, BA_CLUSTER_LIST
));
1305 n
= x
? int_set_get_size(x
->u
.ptr
) : 0;
1306 o
= y
? int_set_get_size(y
->u
.ptr
) : 0;
1312 /* RFC 4271 9.1.2.2. g) Compare peer IP adresses */
1313 return (ipa_compare(new_bgp
->cf
->remote_ip
, old_bgp
->cf
->remote_ip
) < 0);
1318 bgp_rte_mergable(rte
*pri
, rte
*sec
)
1320 struct bgp_proto
*pri_bgp
= (struct bgp_proto
*) pri
->attrs
->src
->proto
;
1321 struct bgp_proto
*sec_bgp
= (struct bgp_proto
*) sec
->attrs
->src
->proto
;
1325 /* Skip suppressed routes (see bgp_rte_recalculate()) */
1326 if (pri
->u
.bgp
.suppressed
!= sec
->u
.bgp
.suppressed
)
1329 /* RFC 4271 9.1.2.1. Route resolvability test */
1330 if (!rte_resolvable(sec
))
1333 /* Start with local preferences */
1334 x
= ea_find(pri
->attrs
->eattrs
, EA_CODE(EAP_BGP
, BA_LOCAL_PREF
));
1335 y
= ea_find(sec
->attrs
->eattrs
, EA_CODE(EAP_BGP
, BA_LOCAL_PREF
));
1336 p
= x
? x
->u
.data
: pri_bgp
->cf
->default_local_pref
;
1337 s
= y
? y
->u
.data
: sec_bgp
->cf
->default_local_pref
;
1341 /* RFC 4271 9.1.2.2. a) Use AS path lengths */
1342 if (pri_bgp
->cf
->compare_path_lengths
|| sec_bgp
->cf
->compare_path_lengths
)
1344 x
= ea_find(pri
->attrs
->eattrs
, EA_CODE(EAP_BGP
, BA_AS_PATH
));
1345 y
= ea_find(sec
->attrs
->eattrs
, EA_CODE(EAP_BGP
, BA_AS_PATH
));
1346 p
= x
? as_path_getlen(x
->u
.ptr
) : AS_PATH_MAXLEN
;
1347 s
= y
? as_path_getlen(y
->u
.ptr
) : AS_PATH_MAXLEN
;
1352 // if (DELTA(p, s) > pri_bgp->cf->relax_multipath)
1356 /* RFC 4271 9.1.2.2. b) Use origins */
1357 x
= ea_find(pri
->attrs
->eattrs
, EA_CODE(EAP_BGP
, BA_ORIGIN
));
1358 y
= ea_find(sec
->attrs
->eattrs
, EA_CODE(EAP_BGP
, BA_ORIGIN
));
1359 p
= x
? x
->u
.data
: ORIGIN_INCOMPLETE
;
1360 s
= y
? y
->u
.data
: ORIGIN_INCOMPLETE
;
1364 /* RFC 4271 9.1.2.2. c) Compare MED's */
1365 if (pri_bgp
->cf
->med_metric
|| sec_bgp
->cf
->med_metric
||
1366 (bgp_get_neighbor(pri
) == bgp_get_neighbor(sec
)))
1368 x
= ea_find(pri
->attrs
->eattrs
, EA_CODE(EAP_BGP
, BA_MULTI_EXIT_DISC
));
1369 y
= ea_find(sec
->attrs
->eattrs
, EA_CODE(EAP_BGP
, BA_MULTI_EXIT_DISC
));
1370 p
= x
? x
->u
.data
: pri_bgp
->cf
->default_med
;
1371 s
= y
? y
->u
.data
: sec_bgp
->cf
->default_med
;
1376 /* RFC 4271 9.1.2.2. d) Prefer external peers */
1377 if (pri_bgp
->is_internal
!= sec_bgp
->is_internal
)
1380 /* RFC 4271 9.1.2.2. e) Compare IGP metrics */
1381 p
= pri_bgp
->cf
->igp_metric
? pri
->attrs
->igp_metric
: 0;
1382 s
= sec_bgp
->cf
->igp_metric
? sec
->attrs
->igp_metric
: 0;
1386 /* Remaining criteria are ignored */
1394 same_group(rte
*r
, u32 lpref
, u32 lasn
)
1396 return (r
->pref
== lpref
) && (bgp_get_neighbor(r
) == lasn
);
1400 use_deterministic_med(rte
*r
)
1402 struct proto
*P
= r
->attrs
->src
->proto
;
1403 return (P
->proto
== &proto_bgp
) && ((struct bgp_proto
*) P
)->cf
->deterministic_med
;
1407 bgp_rte_recalculate(rtable
*table
, net
*net
, rte
*new, rte
*old
, rte
*old_best
)
1410 rte
*key
= new ? new : old
;
1411 u32 lpref
= key
->pref
;
1412 u32 lasn
= bgp_get_neighbor(key
);
1413 int old_is_group_best
= 0;
1416 * Proper RFC 4271 path selection is a bit complicated, it cannot be
1417 * implemented just by rte_better(), because it is not a linear
1418 * ordering. But it can be splitted to two levels, where the lower
1419 * level chooses the best routes in each group of routes from the
1420 * same neighboring AS and higher level chooses the best route (with
1421 * a slightly different ordering) between the best-in-group routes.
1423 * When deterministic_med is disabled, we just ignore this issue and
1424 * choose the best route by bgp_rte_better() alone. If enabled, the
1425 * lower level of the route selection is done here (for the group
1426 * to which the changed route belongs), all routes in group are
1427 * marked as suppressed, just chosen best-in-group is not.
1429 * Global best route selection then implements higher level by
1430 * choosing between non-suppressed routes (as they are always
1431 * preferred over suppressed routes). Routes from BGP protocols
1432 * that do not set deterministic_med are just never suppressed. As
1433 * they do not participate in the lower level selection, it is OK
1434 * that this fn is not called for them.
1436 * The idea is simple, the implementation is more problematic,
1437 * mostly because of optimizations in rte_recalculate() that
1438 * avoids full recalculation in most cases.
1440 * We can assume that at least one of new, old is non-NULL and both
1441 * are from the same protocol with enabled deterministic_med. We
1442 * group routes by both neighbor AS (lasn) and preference (lpref),
1443 * because bgp_rte_better() does not handle preference itself.
1446 /* If new and old are from different groups, we just process that
1447 as two independent events */
1448 if (new && old
&& !same_group(old
, lpref
, lasn
))
1451 i1
= bgp_rte_recalculate(table
, net
, NULL
, old
, old_best
);
1452 i2
= bgp_rte_recalculate(table
, net
, new, NULL
, old_best
);
1457 * We could find the best-in-group and then make some shortcuts like
1458 * in rte_recalculate, but as we would have to walk through all
1459 * net->routes just to find it, it is probably not worth. So we
1460 * just have two simpler fast cases that use just the old route.
1461 * We also set suppressed flag to avoid using it in bgp_rte_better().
1465 new->u
.bgp
.suppressed
= 1;
1469 old_is_group_best
= !old
->u
.bgp
.suppressed
;
1470 old
->u
.bgp
.suppressed
= 1;
1471 int new_is_better
= new && bgp_rte_better(new, old
);
1473 /* The first case - replace not best with worse (or remove not best) */
1474 if (!old_is_group_best
&& !new_is_better
)
1477 /* The second case - replace the best with better */
1478 if (old_is_group_best
&& new_is_better
)
1480 /* new is best-in-group, the see discussion below - this is
1481 a special variant of NBG && OBG. From OBG we can deduce
1482 that same_group(old_best) iff (old == old_best) */
1483 new->u
.bgp
.suppressed
= 0;
1484 return (old
== old_best
);
1488 /* The default case - find a new best-in-group route */
1489 r
= new; /* new may not be in the list */
1490 for (s
=net
->routes
; rte_is_valid(s
); s
=s
->next
)
1491 if (use_deterministic_med(s
) && same_group(s
, lpref
, lasn
))
1493 s
->u
.bgp
.suppressed
= 1;
1494 if (!r
|| bgp_rte_better(s
, r
))
1498 /* Simple case - the last route in group disappears */
1502 /* Found best-in-group */
1503 r
->u
.bgp
.suppressed
= 0;
1506 * There are generally two reasons why we have to force
1507 * recalculation (return 1): First, the new route may be wrongfully
1508 * chosen to be the best in the first case check in
1509 * rte_recalculate(), this may happen only if old_best is from the
1510 * same group. Second, another (different than new route)
1511 * best-in-group is chosen and that may be the proper best (although
1512 * rte_recalculate() without ignore that possibility).
1514 * There are three possible cases according to whether the old route
1515 * was the best in group (OBG, stored in old_is_group_best) and
1516 * whether the new route is the best in group (NBG, tested by r == new).
1517 * These cases work even if old or new is NULL.
1519 * NBG -> new is a possible candidate for the best route, so we just
1520 * check for the first reason using same_group().
1522 * !NBG && OBG -> Second reason applies, return 1
1524 * !NBG && !OBG -> Best in group does not change, old != old_best,
1525 * rte_better(new, old_best) is false and therefore
1526 * the first reason does not apply, return 0
1530 return old_best
&& same_group(old_best
, lpref
, lasn
);
1532 return old_is_group_best
;
1535 static struct adata
*
1536 bgp_aggregator_convert_to_new(struct adata
*old
, struct linpool
*pool
)
1538 struct adata
*newa
= lp_alloc(pool
, sizeof(struct adata
) + 8);
1540 aggregator_convert_to_new(old
, newa
->data
);
1545 /* Take last req_as ASNs from path old2 (in 2B format), convert to 4B format
1546 * and append path old4 (in 4B format).
1548 static struct adata
*
1549 bgp_merge_as_paths(struct adata
*old2
, struct adata
*old4
, int req_as
, struct linpool
*pool
)
1551 byte buf
[old2
->length
* 2];
1553 int ol
= as_path_convert_to_new(old2
, buf
, req_as
);
1554 int nl
= ol
+ (old4
? old4
->length
: 0);
1556 struct adata
*newa
= lp_alloc(pool
, sizeof(struct adata
) + nl
);
1558 memcpy(newa
->data
, buf
, ol
);
1559 if (old4
) memcpy(newa
->data
+ ol
, old4
->data
, old4
->length
);
1565 as4_aggregator_valid(struct adata
*aggr
)
1567 return aggr
->length
== 8;
1571 /* Reconstruct 4B AS_PATH and AGGREGATOR according to RFC 4893 4.2.3 */
1573 bgp_reconstruct_4b_atts(struct bgp_proto
*p
, rta
*a
, struct linpool
*pool
)
1575 eattr
*p2
=ea_find(a
->eattrs
, EA_CODE(EAP_BGP
, BA_AS_PATH
));
1576 eattr
*p4
=ea_find(a
->eattrs
, EA_CODE(EAP_BGP
, BA_AS4_PATH
));
1577 eattr
*a2
=ea_find(a
->eattrs
, EA_CODE(EAP_BGP
, BA_AGGREGATOR
));
1578 eattr
*a4
=ea_find(a
->eattrs
, EA_CODE(EAP_BGP
, BA_AS4_AGGREGATOR
));
1581 if (a4
&& !as4_aggregator_valid(a4
->u
.ptr
))
1583 log(L_WARN
"%s: AS4_AGGREGATOR attribute is invalid, skipping attribute", p
->p
.name
);
1590 u32 a2_as
= get_u16(a2
->u
.ptr
->data
);
1594 if (a2_as
!= AS_TRANS
)
1596 /* Routes were aggregated by old router and therefore AS4_PATH
1597 * and AS4_AGGREGATOR is invalid
1599 * Convert AS_PATH and AGGREGATOR to 4B format and finish.
1602 a2
->u
.ptr
= bgp_aggregator_convert_to_new(a2
->u
.ptr
, pool
);
1603 p2
->u
.ptr
= bgp_merge_as_paths(p2
->u
.ptr
, NULL
, AS_PATH_MAXLEN
, pool
);
1609 /* Common case, use AS4_AGGREGATOR attribute */
1610 a2
->u
.ptr
= a4
->u
.ptr
;
1615 /* Common case, use old AGGREGATOR attribute */
1616 a2
->u
.ptr
= bgp_aggregator_convert_to_new(a2
->u
.ptr
, pool
);
1618 if ((a2_as
== AS_TRANS
) && !a4_removed
)
1619 log(L_WARN
"%s: AGGREGATOR attribute contain AS_TRANS, but AS4_AGGREGATOR is missing", p
->p
.name
);
1624 log(L_WARN
"%s: AS4_AGGREGATOR attribute received, but AGGREGATOR attribute is missing", p
->p
.name
);
1626 int p2_len
= as_path_getlen_int(p2
->u
.ptr
, 2);
1627 int p4_len
= p4
? validate_as4_path(p
, p4
->u
.ptr
) : -1;
1629 if (p4
&& (p4_len
< 0))
1630 log(L_WARN
"%s: AS4_PATH attribute is malformed, skipping attribute", p
->p
.name
);
1632 if ((p4_len
<= 0) || (p2_len
< p4_len
))
1633 p2
->u
.ptr
= bgp_merge_as_paths(p2
->u
.ptr
, NULL
, AS_PATH_MAXLEN
, pool
);
1635 p2
->u
.ptr
= bgp_merge_as_paths(p2
->u
.ptr
, p4
->u
.ptr
, p2_len
- p4_len
, pool
);
1639 bgp_remove_as4_attrs(struct bgp_proto
*p
, rta
*a
)
1641 unsigned id1
= EA_CODE(EAP_BGP
, BA_AS4_PATH
);
1642 unsigned id2
= EA_CODE(EAP_BGP
, BA_AS4_AGGREGATOR
);
1643 ea_list
**el
= &(a
->eattrs
);
1645 /* We know that ea_lists constructed in bgp_decode attrs have one attribute per ea_list struct */
1648 unsigned fid
= (*el
)->attrs
[0].id
;
1650 if ((fid
== id1
) || (fid
== id2
))
1654 log(L_WARN
"%s: Unexpected AS4_* attributes received", p
->p
.name
);
1657 el
= &((*el
)->next
);
1662 * bgp_decode_attrs - check and decode BGP attributes
1664 * @attr: start of attribute block
1665 * @len: length of attribute block
1666 * @pool: linear pool to make all the allocations in
1667 * @mandatory: 1 iff presence of mandatory attributes has to be checked
1669 * This function takes a BGP attribute block (a part of an Update message), checks
1670 * its consistency and converts it to a list of BIRD route attributes represented
1674 bgp_decode_attrs(struct bgp_conn
*conn
, byte
*attr
, uint len
, struct linpool
*pool
, int mandatory
)
1676 struct bgp_proto
*bgp
= conn
->bgp
;
1677 rta
*a
= lp_alloc(pool
, sizeof(struct rta
));
1678 uint flags
, code
, l
, i
, type
;
1680 byte
*z
, *attr_start
;
1686 bzero(a
, sizeof(rta
));
1687 a
->source
= RTS_BGP
;
1688 a
->scope
= SCOPE_UNIVERSE
;
1689 a
->cast
= RTC_UNICAST
;
1690 /* a->dest = RTD_ROUTER; -- set in bgp_set_next_hop() */
1691 a
->from
= bgp
->cf
->remote_ip
;
1693 /* Parse the attributes */
1694 bzero(seen
, sizeof(seen
));
1695 DBG("BGP: Parsing attributes\n");
1704 if (flags
& BAF_EXT_LEN
)
1724 DBG("Attr %02x %02x %d\n", code
, flags
, l
);
1725 if (seen
[code
/8] & (1 << (code
%8)))
1727 if (ATTR_KNOWN(code
))
1729 struct attr_desc
*desc
= &bgp_attr_table
[code
];
1730 if (desc
->expected_length
>= 0 && desc
->expected_length
!= (int) l
)
1731 { errcode
= 5; goto err
; }
1732 if ((desc
->expected_flags
^ flags
) & (BAF_OPTIONAL
| BAF_TRANSITIVE
))
1733 { errcode
= 4; goto err
; }
1734 if (!desc
->allow_in_ebgp
&& !bgp
->is_internal
)
1738 errcode
= desc
->validate(bgp
, z
, l
);
1741 if (errcode
== IGNORE
)
1743 if (errcode
<= WITHDRAW
)
1745 log(L_WARN
"%s: Attribute %s is malformed, withdrawing update",
1746 bgp
->p
.name
, desc
->name
);
1750 else if (code
== BA_AS_PATH
)
1752 /* Special case as it might also trim the attribute */
1753 if (validate_as_path(bgp
, z
, &l
) < 0)
1754 { errcode
= 11; goto err
; }
1758 else /* Unknown attribute */
1760 if (!(flags
& BAF_OPTIONAL
))
1761 { errcode
= 2; goto err
; }
1762 type
= EAF_TYPE_OPAQUE
;
1765 // Only OPTIONAL and TRANSITIVE attributes may have non-zero PARTIAL flag
1766 // if (!((flags & BAF_OPTIONAL) && (flags & BAF_TRANSITIVE)) && (flags & BAF_PARTIAL))
1767 // { errcode = 4; goto err; }
1769 seen
[code
/8] |= (1 << (code
%8));
1770 ea
= lp_alloc(pool
, sizeof(ea_list
) + sizeof(eattr
));
1771 ea
->next
= a
->eattrs
;
1775 ea
->attrs
[0].id
= EA_CODE(EAP_BGP
, code
);
1776 ea
->attrs
[0].flags
= flags
;
1777 ea
->attrs
[0].type
= type
;
1778 if (type
& EAF_EMBEDDED
)
1782 ad
= lp_alloc(pool
, sizeof(struct adata
) + l
);
1783 ea
->attrs
[0].u
.ptr
= ad
;
1785 memcpy(ad
->data
, z
, l
);
1789 case EAF_TYPE_ROUTER_ID
:
1792 ea
->attrs
[0].u
.data
= *z
;
1794 ea
->attrs
[0].u
.data
= get_u32(z
);
1796 case EAF_TYPE_IP_ADDRESS
:
1797 ipa_ntoh(*(ip_addr
*)ad
->data
);
1799 case EAF_TYPE_INT_SET
:
1800 case EAF_TYPE_EC_SET
:
1802 u32
*z
= (u32
*) ad
->data
;
1803 for(i
=0; i
<ad
->length
/4; i
++)
1814 /* If we received MP_REACH_NLRI we should check mandatory attributes */
1815 if (bgp
->mp_reach_len
!= 0)
1819 /* If there is no (reachability) NLRI, we should exit now */
1823 /* Check if all mandatory attributes are present */
1824 for(i
=0; i
< ARRAY_SIZE(bgp_mandatory_attrs
); i
++)
1826 code
= bgp_mandatory_attrs
[i
];
1827 if (!(seen
[code
/8] & (1 << (code
%8))))
1829 bgp_error(conn
, 3, 3, &bgp_mandatory_attrs
[i
], 1);
1834 /* When receiving attributes from non-AS4-aware BGP speaker,
1835 * we have to reconstruct 4B AS_PATH and AGGREGATOR attributes
1837 if (! bgp
->as4_session
)
1838 bgp_reconstruct_4b_atts(bgp
, a
, pool
);
1840 bgp_remove_as4_attrs(bgp
, a
);
1842 /* If the AS path attribute contains our AS, reject the routes */
1843 if (bgp_as_path_loopy(bgp
, a
))
1846 /* Two checks for IBGP loops caused by route reflection, RFC 4456 */
1847 if (bgp_originator_id_loopy(bgp
, a
) ||
1848 bgp_cluster_list_loopy(bgp
, a
))
1851 /* If there's no local preference, define one */
1852 if (!(seen
[0] & (1 << BA_LOCAL_PREF
)))
1853 bgp_attach_attr(&a
->eattrs
, pool
, BA_LOCAL_PREF
, bgp
->cf
->default_local_pref
);
1861 bgp_error(conn
, 3, 1, NULL
, 0);
1865 bgp_error(conn
, 3, errcode
, attr_start
, z
+l
-attr_start
);
1870 bgp_get_attr(eattr
*a
, byte
*buf
, int buflen
)
1872 uint i
= EA_ID(a
->id
);
1873 struct attr_desc
*d
;
1878 d
= &bgp_attr_table
[i
];
1879 len
= bsprintf(buf
, "%s", d
->name
);
1885 d
->format(a
, buf
, buflen
- len
- 2);
1890 bsprintf(buf
, "%02x%s", i
, (a
->flags
& BAF_TRANSITIVE
) ? " [t]" : "");
1895 bgp_init_bucket_table(struct bgp_proto
*p
)
1898 p
->hash_limit
= p
->hash_size
* 4;
1899 p
->bucket_hash
= mb_allocz(p
->p
.pool
, p
->hash_size
* sizeof(struct bgp_bucket
*));
1900 init_list(&p
->bucket_queue
);
1901 p
->withdraw_bucket
= NULL
;
1902 // fib_init(&p->prefix_fib, p->p.pool, sizeof(struct bgp_prefix), 0, bgp_init_prefix);
1906 bgp_get_route_info(rte
*e
, byte
*buf
, ea_list
*attrs
)
1908 eattr
*p
= ea_find(attrs
, EA_CODE(EAP_BGP
, BA_AS_PATH
));
1909 eattr
*o
= ea_find(attrs
, EA_CODE(EAP_BGP
, BA_ORIGIN
));
1912 buf
+= bsprintf(buf
, " (%d", e
->pref
);
1914 if (e
->u
.bgp
.suppressed
)
1915 buf
+= bsprintf(buf
, "-");
1917 if (e
->attrs
->hostentry
)
1919 if (!rte_resolvable(e
))
1920 buf
+= bsprintf(buf
, "/-");
1921 else if (e
->attrs
->igp_metric
>= IGP_METRIC_UNKNOWN
)
1922 buf
+= bsprintf(buf
, "/?");
1924 buf
+= bsprintf(buf
, "/%d", e
->attrs
->igp_metric
);
1926 buf
+= bsprintf(buf
, ") [");
1928 if (p
&& as_path_get_last(p
->u
.ptr
, &origas
))
1929 buf
+= bsprintf(buf
, "AS%u", origas
);
1931 buf
+= bsprintf(buf
, "%c", "ie?"[o
->u
.data
]);