2 * BIRD -- BGP Attributes
4 * (c) 2000 Martin Mares <mj@ucw.cz>
5 * (c) 2008--2016 Ondrej Zajicek <santiago@crfreenet.org>
6 * (c) 2008--2016 CZ.NIC z.s.p.o.
8 * Can be freely distributed and used under the terms of the GNU GPL.
15 #include "nest/bird.h"
16 #include "nest/iface.h"
17 #include "nest/protocol.h"
18 #include "nest/route.h"
19 #include "nest/attrs.h"
20 #include "conf/conf.h"
21 #include "lib/resource.h"
22 #include "lib/string.h"
23 #include "lib/unaligned.h"
28 * UPDATE message error handling
30 * All checks from RFC 4271 6.3 are done as specified with these exceptions:
31 * - The semantic check of an IP address from NEXT_HOP attribute is missing.
32 * - Checks of some optional attribute values are missing.
33 * - Syntactic and semantic checks of NLRIs (done in DECODE_PREFIX())
34 * are probably inadequate.
36 * Loop detection based on AS_PATH causes updates to be withdrawn. RFC
37 * 4271 does not explicitly specifiy the behavior in that case.
39 * Loop detection related to route reflection (based on ORIGINATOR_ID
40 * and CLUSTER_LIST) causes updates to be withdrawn. RFC 4456 8
41 * specifies that such updates should be ignored, but that is generally
44 * BGP attribute table has several hooks:
46 * export - Hook that validates and normalizes attribute during export phase.
47 * Receives eattr, may modify it (e.g., sort community lists for canonical
48 * representation), UNSET() it (e.g., skip empty lists), or WITHDRAW() it if
49 * necessary. May assume that eattr has value valid w.r.t. its type, but may be
50 * invalid w.r.t. BGP constraints. Optional.
52 * encode - Hook that converts internal representation to external one during
53 * packet writing. Receives eattr and puts it in the buffer (including attribute
54 * header). Returns number of bytes, or -1 if not enough space. May assume that
55 * eattr has value valid w.r.t. its type and validated by export hook. Mandatory
56 * for all known attributes that exist internally after export phase (i.e., all
57 * except pseudoattributes MP_(UN)REACH_NLRI).
59 * decode - Hook that converts external representation to internal one during
60 * packet parsing. Receives attribute data in buffer, validates it and adds
61 * attribute to ea_list. If data are invalid, steps DISCARD(), WITHDRAW() or
62 * bgp_parse_error() may be used to escape. Mandatory for all known attributes.
64 * format - Optional hook that converts eattr to textual representation.
67 // XXXX review pool usage : c->c.proto->pool
70 struct bgp_attr_desc
{
74 void (*export
)(struct bgp_export_state
*s
, eattr
*a
);
75 int (*encode
)(struct bgp_write_state
*s
, eattr
*a
, byte
*buf
, uint size
);
76 void (*decode
)(struct bgp_parse_state
*s
, uint code
, uint flags
, byte
*data
, uint len
, ea_list
**to
);
77 void (*format
)(eattr
*ea
, byte
*buf
, uint size
);
80 static const struct bgp_attr_desc bgp_attr_table
[];
82 static inline int bgp_attr_known(uint code
);
85 bgp_set_attr(ea_list
**attrs
, struct linpool
*pool
, uint code
, uint flags
, uintptr_t val
)
87 ASSERT(bgp_attr_known(code
));
89 ea_list
*a
= lp_alloc(pool
, sizeof(ea_list
) + sizeof(eattr
));
90 eattr
*e
= &a
->attrs
[0];
92 a
->flags
= EALF_SORTED
;
97 e
->id
= EA_CODE(EAP_BGP
, code
);
98 e
->type
= bgp_attr_table
[code
].type
;
101 if (e
->type
& EAF_EMBEDDED
)
102 e
->u
.data
= (u32
) val
;
104 e
->u
.ptr
= (struct adata
*) val
;
111 #define REPORT(msg, args...) \
112 ({ log(L_REMOTE "%s: " msg, s->proto->p.name, ## args); })
114 #define DISCARD(msg, args...) \
115 ({ REPORT(msg, ## args); return; })
117 #define WITHDRAW(msg, args...) \
118 ({ REPORT(msg, ## args); s->err_withdraw = 1; return; })
121 ({ a->type = EAF_TYPE_UNDEF; return; })
123 #define NEW_BGP "Discarding %s attribute received from AS4-aware neighbor"
124 #define BAD_EBGP "Discarding %s attribute received from EBGP neighbor"
125 #define BAD_LENGTH "Malformed %s attribute - invalid length (%u)"
126 #define BAD_VALUE "Malformed %s attribute - invalid value (%u)"
127 #define NO_MANDATORY "Missing mandatory %s attribute"
131 bgp_put_attr_hdr3(byte
*buf
, uint code
, uint flags
, uint len
)
140 bgp_put_attr_hdr4(byte
*buf
, uint code
, uint flags
, uint len
)
142 *buf
++ = flags
| BAF_EXT_LEN
;
149 bgp_put_attr_hdr(byte
*buf
, uint code
, uint flags
, uint len
)
152 return bgp_put_attr_hdr3(buf
, code
, flags
, len
);
154 return bgp_put_attr_hdr4(buf
, code
, flags
, len
);
158 bgp_encode_u8(struct bgp_write_state
*s UNUSED
, eattr
*a
, byte
*buf
, uint size
)
163 bgp_put_attr_hdr3(buf
, EA_ID(a
->id
), a
->flags
, 1);
170 bgp_encode_u32(struct bgp_write_state
*s UNUSED
, eattr
*a
, byte
*buf
, uint size
)
175 bgp_put_attr_hdr3(buf
, EA_ID(a
->id
), a
->flags
, 4);
176 put_u32(buf
+3, a
->u
.data
);
182 bgp_encode_u32s(struct bgp_write_state
*s UNUSED
, eattr
*a
, byte
*buf
, uint size
)
184 uint len
= a
->u
.ptr
->length
;
189 uint hdr
= bgp_put_attr_hdr(buf
, EA_ID(a
->id
), a
->flags
, len
);
190 put_u32s(buf
+ hdr
, (u32
*) a
->u
.ptr
->data
, len
/ 4);
196 bgp_put_attr(byte
*buf
, uint size
, uint code
, uint flags
, byte
*data
, uint len
)
201 uint hdr
= bgp_put_attr_hdr(buf
, code
, flags
, len
);
202 memcpy(buf
+ hdr
, data
, len
);
208 bgp_encode_raw(struct bgp_write_state
*s UNUSED
, eattr
*a
, byte
*buf
, uint size
)
210 return bgp_put_attr(buf
, size
, EA_ID(a
->id
), a
->flags
, a
->u
.ptr
->data
, a
->u
.ptr
->length
);
219 bgp_export_origin(struct bgp_export_state
*s
, eattr
*a
)
222 WITHDRAW(BAD_VALUE
, "ORIGIN", a
->u
.data
);
226 bgp_decode_origin(struct bgp_parse_state
*s
, uint code UNUSED
, uint flags
, byte
*data
, uint len
, ea_list
**to
)
229 WITHDRAW(BAD_LENGTH
, "ORIGIN", len
);
232 WITHDRAW(BAD_VALUE
, "ORIGIN", data
[0]);
234 bgp_set_attr_u32(to
, s
->pool
, BA_ORIGIN
, flags
, data
[0]);
238 bgp_format_origin(eattr
*a
, byte
*buf
, uint size UNUSED
)
240 static const char *bgp_origin_names
[] = { "IGP", "EGP", "Incomplete" };
242 bsprintf(buf
, (a
->u
.data
<= 2) ? bgp_origin_names
[a
->u
.data
] : "?");
247 bgp_encode_as_path(struct bgp_write_state
*s
, eattr
*a
, byte
*buf
, uint size
)
249 byte
*data
= a
->u
.ptr
->data
;
250 uint len
= a
->u
.ptr
->length
;
254 /* Prepare 16-bit AS_PATH (from 32-bit one) in a temporary buffer */
257 len
= as_path_32to16(data
, src
, len
);
260 return bgp_put_attr(buf
, size
, BA_AS_PATH
, a
->flags
, data
, len
);
264 bgp_decode_as_path(struct bgp_parse_state
*s
, uint code UNUSED
, uint flags
, byte
*data
, uint len
, ea_list
**to
)
266 struct bgp_proto
*p
= s
->proto
;
267 int as_length
= s
->as4_session
? 4 : 2;
268 int as_confed
= p
->cf
->confederation
&& p
->is_interior
;
271 if (!as_path_valid(data
, len
, as_length
, as_confed
, err
, sizeof(err
)))
272 WITHDRAW("Malformed AS_PATH attribute - %s", err
);
274 /* In some circumstances check for initial AS_CONFED_SEQUENCE; RFC 5065 5.0 */
275 if (p
->is_interior
&& !p
->is_internal
&&
276 ((len
< 2) || (data
[0] != AS_PATH_CONFED_SEQUENCE
)))
277 WITHDRAW("Malformed AS_PATH attribute - %s", "missing initial AS_CONFED_SEQUENCE");
281 /* Prepare 32-bit AS_PATH (from 16-bit one) in a temporary buffer */
283 data
= alloca(2*len
);
284 len
= as_path_16to32(data
, src
, len
);
287 bgp_set_attr_data(to
, s
->pool
, BA_AS_PATH
, flags
, data
, len
);
292 bgp_encode_next_hop(struct bgp_write_state
*s
, eattr
*a
, byte
*buf
, uint size
)
295 * The NEXT_HOP attribute is used only in traditional (IPv4) BGP. In MP-BGP,
296 * the next hop is encoded as a part of the MP_REACH_NLRI attribute, so we
297 * store it and encode it later by AFI-specific hooks.
300 if (s
->channel
->afi
== BGP_AF_IPV4
)
302 ASSERT(a
->u
.ptr
->length
== sizeof(ip_addr
));
307 bgp_put_attr_hdr3(buf
, BA_NEXT_HOP
, a
->flags
, 4);
308 put_ip4(buf
+3, ipa_to_ip4( *(ip_addr
*) a
->u
.ptr
->data
));
320 bgp_decode_next_hop(struct bgp_parse_state
*s
, uint code UNUSED
, uint flags UNUSED
, byte
*data
, uint len
, ea_list
**to UNUSED
)
323 WITHDRAW(BAD_LENGTH
, "NEXT_HOP", len
);
325 /* Semantic checks are done later */
326 s
->ip_next_hop_len
= len
;
327 s
->ip_next_hop_data
= data
;
330 /* TODO: This function should use AF-specific hook */
332 bgp_format_next_hop(eattr
*a
, byte
*buf
, uint size UNUSED
)
334 ip_addr
*nh
= (void *) a
->u
.ptr
->data
;
335 uint len
= a
->u
.ptr
->length
;
337 ASSERT((len
== 16) || (len
== 32));
339 /* in IPv6, we may have two addresses in NEXT HOP */
340 if ((len
== 16) || ipa_zero(nh
[1]))
341 bsprintf(buf
, "%I", nh
[0]);
343 bsprintf(buf
, "%I %I", nh
[0], nh
[1]);
348 bgp_decode_med(struct bgp_parse_state
*s
, uint code UNUSED
, uint flags
, byte
*data
, uint len
, ea_list
**to
)
351 WITHDRAW(BAD_LENGTH
, "MULTI_EXIT_DISC", len
);
353 u32 val
= get_u32(data
);
354 bgp_set_attr_u32(to
, s
->pool
, BA_MULTI_EXIT_DISC
, flags
, val
);
359 bgp_export_local_pref(struct bgp_export_state
*s
, eattr
*a
)
361 if (!s
->proto
->is_interior
)
366 bgp_decode_local_pref(struct bgp_parse_state
*s
, uint code UNUSED
, uint flags
, byte
*data
, uint len
, ea_list
**to
)
368 if (!s
->proto
->is_interior
)
369 DISCARD(BAD_EBGP
, "LOCAL_PREF");
372 WITHDRAW(BAD_LENGTH
, "LOCAL_PREF", len
);
374 u32 val
= get_u32(data
);
375 bgp_set_attr_u32(to
, s
->pool
, BA_LOCAL_PREF
, flags
, val
);
380 bgp_decode_atomic_aggr(struct bgp_parse_state
*s
, uint code UNUSED
, uint flags
, byte
*data UNUSED
, uint len
, ea_list
**to
)
383 DISCARD(BAD_LENGTH
, "ATOMIC_AGGR", len
);
385 bgp_set_attr_data(to
, s
->pool
, BA_ATOMIC_AGGR
, flags
, NULL
, 0);
389 bgp_encode_aggregator(struct bgp_write_state
*s
, eattr
*a
, byte
*buf
, uint size
)
391 byte
*data
= a
->u
.ptr
->data
;
392 uint len
= a
->u
.ptr
->length
;
396 /* Prepare 16-bit AGGREGATOR (from 32-bit one) in a temporary buffer */
399 len
= aggregator_32to16(data
, src
);
402 return bgp_put_attr(buf
, size
, BA_AGGREGATOR
, a
->flags
, data
, len
);
406 bgp_decode_aggregator(struct bgp_parse_state
*s
, uint code UNUSED
, uint flags
, byte
*data
, uint len
, ea_list
**to
)
408 if (len
!= (s
->as4_session
? 8 : 6))
409 DISCARD(BAD_LENGTH
, "AGGREGATOR", len
);
413 /* Prepare 32-bit AGGREGATOR (from 16-bit one) in a temporary buffer */
416 len
= aggregator_16to32(data
, src
);
419 bgp_set_attr_data(to
, s
->pool
, BA_AGGREGATOR
, flags
, data
, len
);
423 bgp_format_aggregator(eattr
*a
, byte
*buf
, uint size UNUSED
)
425 byte
*data
= a
->u
.ptr
->data
;
427 bsprintf(buf
, "%I4 AS%u", get_ip4(data
+4), get_u32(data
+0));
432 bgp_export_community(struct bgp_export_state
*s
, eattr
*a
)
434 if (a
->u
.ptr
->length
== 0)
437 a
->u
.ptr
= int_set_sort(s
->pool
, a
->u
.ptr
);
441 bgp_decode_community(struct bgp_parse_state
*s
, uint code UNUSED
, uint flags
, byte
*data
, uint len
, ea_list
**to
)
443 if (!len
|| (len
% 4))
444 WITHDRAW(BAD_LENGTH
, "COMMUNITY", len
);
446 struct adata
*ad
= lp_alloc_adata(s
->pool
, len
);
447 get_u32s(data
, (u32
*) ad
->data
, len
/ 4);
448 bgp_set_attr_ptr(to
, s
->pool
, BA_COMMUNITY
, flags
, ad
);
453 bgp_export_originator_id(struct bgp_export_state
*s
, eattr
*a
)
455 if (!s
->proto
->is_internal
)
460 bgp_decode_originator_id(struct bgp_parse_state
*s
, uint code UNUSED
, uint flags
, byte
*data
, uint len
, ea_list
**to
)
462 if (!s
->proto
->is_internal
)
463 DISCARD(BAD_EBGP
, "ORIGINATOR_ID");
466 WITHDRAW(BAD_LENGTH
, "ORIGINATOR_ID", len
);
468 u32 val
= get_u32(data
);
469 bgp_set_attr_u32(to
, s
->pool
, BA_ORIGINATOR_ID
, flags
, val
);
474 bgp_export_cluster_list(struct bgp_export_state
*s UNUSED
, eattr
*a
)
476 if (!s
->proto
->is_internal
)
479 if (a
->u
.ptr
->length
== 0)
484 bgp_decode_cluster_list(struct bgp_parse_state
*s
, uint code UNUSED
, uint flags
, byte
*data
, uint len
, ea_list
**to
)
486 if (!s
->proto
->is_internal
)
487 DISCARD(BAD_EBGP
, "CLUSTER_LIST");
489 if (!len
|| (len
% 4))
490 WITHDRAW(BAD_LENGTH
, "CLUSTER_LIST", len
);
492 struct adata
*ad
= lp_alloc_adata(s
->pool
, len
);
493 get_u32s(data
, (u32
*) ad
->data
, len
/ 4);
494 bgp_set_attr_ptr(to
, s
->pool
, BA_CLUSTER_LIST
, flags
, ad
);
498 bgp_format_cluster_list(eattr
*a
, byte
*buf
, uint size
)
500 /* Truncates cluster lists larger than buflen, probably not a problem */
501 int_set_format(a
->u
.ptr
, 0, -1, buf
, size
);
508 return (get_u16(buf
) << 16) | buf
[2];
512 bgp_decode_mp_reach_nlri(struct bgp_parse_state
*s
, uint code UNUSED
, uint flags UNUSED
, byte
*data
, uint len
, ea_list
**to UNUSED
)
515 * 2 B MP_REACH_NLRI data - Address Family Identifier
516 * 1 B MP_REACH_NLRI data - Subsequent Address Family Identifier
517 * 1 B MP_REACH_NLRI data - Length of Next Hop Network Address
518 * var MP_REACH_NLRI data - Network Address of Next Hop
519 * 1 B MP_REACH_NLRI data - Reserved (zero)
520 * var MP_REACH_NLRI data - Network Layer Reachability Information
523 if ((len
< 5) || (len
< (5 + (uint
) data
[3])))
524 bgp_parse_error(s
, 9);
526 s
->mp_reach_af
= get_af3(data
);
527 s
->mp_next_hop_len
= data
[3];
528 s
->mp_next_hop_data
= data
+ 4;
529 s
->mp_reach_len
= len
- 5 - s
->mp_next_hop_len
;
530 s
->mp_reach_nlri
= data
+ 5 + s
->mp_next_hop_len
;
535 bgp_decode_mp_unreach_nlri(struct bgp_parse_state
*s
, uint code UNUSED
, uint flags UNUSED
, byte
*data
, uint len
, ea_list
**to UNUSED
)
538 * 2 B MP_UNREACH_NLRI data - Address Family Identifier
539 * 1 B MP_UNREACH_NLRI data - Subsequent Address Family Identifier
540 * var MP_UNREACH_NLRI data - Network Layer Reachability Information
544 bgp_parse_error(s
, 9);
546 s
->mp_unreach_af
= get_af3(data
);
547 s
->mp_unreach_len
= len
- 3;
548 s
->mp_unreach_nlri
= data
+ 3;
553 bgp_export_ext_community(struct bgp_export_state
*s
, eattr
*a
)
555 if (a
->u
.ptr
->length
== 0)
558 a
->u
.ptr
= ec_set_sort(s
->pool
, a
->u
.ptr
);
562 bgp_decode_ext_community(struct bgp_parse_state
*s
, uint code UNUSED
, uint flags
, byte
*data
, uint len
, ea_list
**to
)
564 if (!len
|| (len
% 8))
565 WITHDRAW(BAD_LENGTH
, "EXT_COMMUNITY", len
);
567 struct adata
*ad
= lp_alloc_adata(s
->pool
, len
);
568 get_u32s(data
, (u32
*) ad
->data
, len
/ 4);
569 bgp_set_attr_ptr(to
, s
->pool
, BA_EXT_COMMUNITY
, flags
, ad
);
574 bgp_decode_as4_aggregator(struct bgp_parse_state
*s
, uint code UNUSED
, uint flags
, byte
*data
, uint len
, ea_list
**to
)
577 DISCARD(NEW_BGP
, "AS4_AGGREGATOR");
580 DISCARD(BAD_LENGTH
, "AS4_AGGREGATOR", len
);
582 bgp_set_attr_data(to
, s
->pool
, BA_AS4_AGGREGATOR
, flags
, data
, len
);
586 bgp_decode_as4_path(struct bgp_parse_state
*s
, uint code UNUSED
, uint flags
, byte
*data
, uint len
, ea_list
**to
)
591 DISCARD(NEW_BGP
, "AS4_PATH");
594 DISCARD(BAD_LENGTH
, "AS4_PATH", len
);
596 if (!as_path_valid(data
, len
, 4, 1, err
, sizeof(err
)))
597 DISCARD("Malformed AS4_PATH attribute - %s", err
);
599 struct adata
*a
= lp_alloc_adata(s
->pool
, len
);
600 memcpy(a
->data
, data
, len
);
602 /* AS_CONFED* segments are invalid in AS4_PATH; RFC 6793 6 */
603 if (as_path_contains_confed(a
))
605 REPORT("Discarding AS_CONFED* segment from AS4_PATH attribute");
606 a
= as_path_strip_confed(s
->pool
, a
);
609 bgp_set_attr_ptr(to
, s
->pool
, BA_AS4_PATH
, flags
, a
);
613 bgp_export_large_community(struct bgp_export_state
*s
, eattr
*a
)
615 if (a
->u
.ptr
->length
== 0)
618 a
->u
.ptr
= lc_set_sort(s
->pool
, a
->u
.ptr
);
622 bgp_decode_large_community(struct bgp_parse_state
*s
, uint code UNUSED
, uint flags
, byte
*data
, uint len
, ea_list
**to
)
624 if (!len
|| (len
% 12))
625 WITHDRAW(BAD_LENGTH
, "LARGE_COMMUNITY", len
);
627 struct adata
*ad
= lp_alloc_adata(s
->pool
, len
);
628 get_u32s(data
, (u32
*) ad
->data
, len
/ 4);
629 bgp_set_attr_ptr(to
, s
->pool
, BA_LARGE_COMMUNITY
, flags
, ad
);
633 bgp_decode_unknown(struct bgp_parse_state
*s
, uint code
, uint flags
, byte
*data
, uint len
, ea_list
**to
)
635 bgp_set_attr_data(to
, s
->pool
, code
, flags
, data
, len
);
643 static const struct bgp_attr_desc bgp_attr_table
[] = {
646 .type
= EAF_TYPE_INT
,
647 .flags
= BAF_TRANSITIVE
,
648 .export
= bgp_export_origin
,
649 .encode
= bgp_encode_u8
,
650 .decode
= bgp_decode_origin
,
651 .format
= bgp_format_origin
,
655 .type
= EAF_TYPE_AS_PATH
,
656 .flags
= BAF_TRANSITIVE
,
657 .encode
= bgp_encode_as_path
,
658 .decode
= bgp_decode_as_path
,
662 .type
= EAF_TYPE_IP_ADDRESS
,
663 .flags
= BAF_TRANSITIVE
,
664 .encode
= bgp_encode_next_hop
,
665 .decode
= bgp_decode_next_hop
,
666 .format
= bgp_format_next_hop
,
668 [BA_MULTI_EXIT_DISC
] = {
670 .type
= EAF_TYPE_INT
,
671 .flags
= BAF_OPTIONAL
,
672 .encode
= bgp_encode_u32
,
673 .decode
= bgp_decode_med
,
676 .name
= "local_pref",
677 .type
= EAF_TYPE_INT
,
678 .flags
= BAF_TRANSITIVE
,
679 .export
= bgp_export_local_pref
,
680 .encode
= bgp_encode_u32
,
681 .decode
= bgp_decode_local_pref
,
684 .name
= "atomic_aggr",
685 .type
= EAF_TYPE_OPAQUE
,
686 .flags
= BAF_TRANSITIVE
,
687 .encode
= bgp_encode_raw
,
688 .decode
= bgp_decode_atomic_aggr
,
691 .name
= "aggregator",
692 .type
= EAF_TYPE_OPAQUE
,
693 .flags
= BAF_OPTIONAL
| BAF_TRANSITIVE
,
694 .encode
= bgp_encode_aggregator
,
695 .decode
= bgp_decode_aggregator
,
696 .format
= bgp_format_aggregator
,
700 .type
= EAF_TYPE_INT_SET
,
701 .flags
= BAF_OPTIONAL
| BAF_TRANSITIVE
,
702 .export
= bgp_export_community
,
703 .encode
= bgp_encode_u32s
,
704 .decode
= bgp_decode_community
,
706 [BA_ORIGINATOR_ID
] = {
707 .name
= "originator_id",
708 .type
= EAF_TYPE_ROUTER_ID
,
709 .flags
= BAF_OPTIONAL
,
710 .export
= bgp_export_originator_id
,
711 .encode
= bgp_encode_u32
,
712 .decode
= bgp_decode_originator_id
,
714 [BA_CLUSTER_LIST
] = {
715 .name
= "cluster_list",
716 .type
= EAF_TYPE_INT_SET
,
717 .flags
= BAF_OPTIONAL
,
718 .export
= bgp_export_cluster_list
,
719 .encode
= bgp_encode_u32s
,
720 .decode
= bgp_decode_cluster_list
,
721 .format
= bgp_format_cluster_list
,
723 [BA_MP_REACH_NLRI
] = {
724 .name
= "mp_reach_nlri",
725 .type
= EAF_TYPE_OPAQUE
,
726 .flags
= BAF_OPTIONAL
,
727 .decode
= bgp_decode_mp_reach_nlri
,
729 [BA_MP_UNREACH_NLRI
] = {
730 .name
= "mp_unreach_nlri",
731 .type
= EAF_TYPE_OPAQUE
,
732 .flags
= BAF_OPTIONAL
,
733 .decode
= bgp_decode_mp_unreach_nlri
,
735 [BA_EXT_COMMUNITY
] = {
736 .name
= "ext_community",
737 .type
= EAF_TYPE_EC_SET
,
738 .flags
= BAF_OPTIONAL
| BAF_TRANSITIVE
,
739 .export
= bgp_export_ext_community
,
740 .encode
= bgp_encode_u32s
,
741 .decode
= bgp_decode_ext_community
,
745 .type
= EAF_TYPE_AS_PATH
,
746 .flags
= BAF_OPTIONAL
| BAF_TRANSITIVE
,
747 .encode
= bgp_encode_raw
,
748 .decode
= bgp_decode_as4_path
,
750 [BA_AS4_AGGREGATOR
] = {
751 .name
= "as4_aggregator",
752 .type
= EAF_TYPE_OPAQUE
,
753 .flags
= BAF_OPTIONAL
| BAF_TRANSITIVE
,
754 .encode
= bgp_encode_raw
,
755 .decode
= bgp_decode_as4_aggregator
,
756 .format
= bgp_format_aggregator
,
758 [BA_LARGE_COMMUNITY
] = {
759 .name
= "large_community",
760 .type
= EAF_TYPE_LC_SET
,
761 .flags
= BAF_OPTIONAL
| BAF_TRANSITIVE
,
762 .export
= bgp_export_large_community
,
763 .encode
= bgp_encode_u32s
,
764 .decode
= bgp_decode_large_community
,
769 bgp_attr_known(uint code
)
771 return (code
< ARRAY_SIZE(bgp_attr_table
)) && bgp_attr_table
[code
].name
;
780 bgp_export_attr(struct bgp_export_state
*s
, eattr
*a
, ea_list
*to
)
782 if (EA_PROTO(a
->id
) != EAP_BGP
)
785 uint code
= EA_ID(a
->id
);
787 if (bgp_attr_known(code
))
789 const struct bgp_attr_desc
*desc
= &bgp_attr_table
[code
];
791 /* The flags might have been zero if the attr was added by filters */
792 a
->flags
= (a
->flags
& BAF_PARTIAL
) | desc
->flags
;
794 /* Set partial bit if new opt-trans attribute is attached to non-local route */
795 if ((s
->src
!= NULL
) && (a
->type
& EAF_ORIGINATED
) &&
796 (a
->flags
& BAF_OPTIONAL
) && (a
->flags
& BAF_TRANSITIVE
))
797 a
->flags
|= BAF_PARTIAL
;
799 /* Call specific hook */
800 CALL(desc
->export
, s
, a
);
802 /* Attribute might become undefined in hook */
803 if ((a
->type
& EAF_TYPE_MASK
) == EAF_TYPE_UNDEF
)
808 /* Don't re-export unknown non-transitive attributes */
809 if (!(a
->flags
& BAF_TRANSITIVE
))
812 a
->flags
|= BAF_PARTIAL
;
815 /* Append updated attribute */
816 to
->attrs
[to
->count
++] = *a
;
820 * bgp_export_attrs - export BGP attributes
821 * @s: BGP export state
822 * @attrs: a list of extended attributes
824 * The bgp_export_attrs() function takes a list of attributes and merges it to
825 * one newly allocated and sorted segment. Attributes are validated and
826 * normalized by type-specific export hooks and attribute flags are updated.
827 * Some attributes may be eliminated (e.g. unknown non-tranitive attributes, or
828 * empty community sets).
830 * Result: one sorted attribute list segment, or NULL if attributes are unsuitable.
832 static inline ea_list
*
833 bgp_export_attrs(struct bgp_export_state
*s
, ea_list
*attrs
)
835 /* Merge the attribute list */
836 ea_list
*new = lp_alloc(s
->pool
, ea_scan(attrs
));
837 ea_merge(attrs
, new);
844 /* Export each attribute */
845 for (i
= 0; i
< count
; i
++)
846 bgp_export_attr(s
, &new->attrs
[i
], new);
861 bgp_encode_attr(struct bgp_write_state
*s
, eattr
*a
, byte
*buf
, uint size
)
863 ASSERT(EA_PROTO(a
->id
) == EAP_BGP
);
865 uint code
= EA_ID(a
->id
);
867 if (bgp_attr_known(code
))
868 return bgp_attr_table
[code
].encode(s
, a
, buf
, size
);
870 return bgp_encode_raw(s
, a
, buf
, size
);
874 * bgp_encode_attrs - encode BGP attributes
875 * @s: BGP write state
876 * @attrs: a list of extended attributes
880 * The bgp_encode_attrs() function takes a list of extended attributes
881 * and converts it to its BGP representation (a part of an Update message).
883 * Result: Length of the attribute block generated or -1 if not enough space.
886 bgp_encode_attrs(struct bgp_write_state
*s
, ea_list
*attrs
, byte
*buf
, byte
*end
)
891 for (i
= 0; i
< attrs
->count
; i
++)
893 len
= bgp_encode_attr(s
, &attrs
->attrs
[i
], pos
, end
- pos
);
909 static void bgp_process_as4_attrs(ea_list
**attrs
, struct linpool
*pool
);
912 bgp_as_path_loopy(struct bgp_proto
*p
, ea_list
*attrs
, u32 asn
)
914 eattr
*e
= bgp_find_attr(attrs
, BA_AS_PATH
);
915 int num
= p
->cf
->allow_local_as
+ 1;
916 return (e
&& (num
> 0) && as_path_contains(e
->u
.ptr
, asn
, num
));
920 bgp_originator_id_loopy(struct bgp_proto
*p
, ea_list
*attrs
)
922 eattr
*e
= bgp_find_attr(attrs
, BA_ORIGINATOR_ID
);
923 return (e
&& (e
->u
.data
== p
->local_id
));
927 bgp_cluster_list_loopy(struct bgp_proto
*p
, ea_list
*attrs
)
929 eattr
*e
= bgp_find_attr(attrs
, BA_CLUSTER_LIST
);
930 return (e
&& int_set_contains(e
->u
.ptr
, p
->rr_cluster_id
));
934 bgp_decode_attr(struct bgp_parse_state
*s
, uint code
, uint flags
, byte
*data
, uint len
, ea_list
**to
)
936 /* Handle duplicate attributes; RFC 7606 3 (g) */
937 if (BIT32_TEST(s
->attrs_seen
, code
))
939 if ((code
== BA_MP_REACH_NLRI
) || (code
== BA_MP_UNREACH_NLRI
))
940 bgp_parse_error(s
, 1);
942 DISCARD("Discarding duplicate attribute (code %u)", code
);
944 BIT32_SET(s
->attrs_seen
, code
);
946 if (bgp_attr_known(code
))
948 const struct bgp_attr_desc
*desc
= &bgp_attr_table
[code
];
950 /* Handle conflicting flags; RFC 7606 3 (c) */
951 if ((flags
^ desc
->flags
) & (BAF_OPTIONAL
| BAF_TRANSITIVE
))
952 WITHDRAW("Malformed %s attribute - conflicting flags (%02x)", desc
->name
, flags
);
954 desc
->decode(s
, code
, flags
, data
, len
, to
);
956 else /* Unknown attribute */
958 if (!(flags
& BAF_OPTIONAL
))
959 WITHDRAW("Unknown attribute (code %u) - conflicting flags (%02x)", code
, flags
);
961 bgp_decode_unknown(s
, code
, flags
, data
, len
, to
);
966 * bgp_decode_attrs - check and decode BGP attributes
967 * @s: BGP parse state
968 * @data: start of attribute block
969 * @len: length of attribute block
971 * This function takes a BGP attribute block (a part of an Update message), checks
972 * its consistency and converts it to a list of BIRD route attributes represented
973 * by an (uncached) &rta.
976 bgp_decode_attrs(struct bgp_parse_state
*s
, byte
*data
, uint len
)
978 struct bgp_proto
*p
= s
->proto
;
979 ea_list
*attrs
= NULL
;
980 uint code
, flags
, alen
;
983 /* Parse the attributes */
988 /* Read attribute type */
993 ADVANCE(pos
, len
, 2);
995 /* Read attribute length */
996 if (flags
& BAF_EXT_LEN
)
1000 alen
= get_u16(pos
);
1001 ADVANCE(pos
, len
, 2);
1008 ADVANCE(pos
, len
, 1);
1014 DBG("Attr %02x %02x %u\n", code
, flags
, alen
);
1016 bgp_decode_attr(s
, code
, flags
, pos
, alen
, &attrs
);
1017 ADVANCE(pos
, len
, alen
);
1020 if (s
->err_withdraw
)
1023 /* If there is no reachability NLRI, we are finished */
1024 if (!s
->ip_reach_len
&& !s
->mp_reach_len
)
1028 /* Handle missing mandatory attributes; RFC 7606 3 (d) */
1029 if (!BIT32_TEST(s
->attrs_seen
, BA_ORIGIN
))
1030 { REPORT(NO_MANDATORY
, "ORIGIN"); goto withdraw
; }
1032 if (!BIT32_TEST(s
->attrs_seen
, BA_AS_PATH
))
1033 { REPORT(NO_MANDATORY
, "AS_PATH"); goto withdraw
; }
1035 /* When receiving attributes from non-AS4-aware BGP speaker, we have to
1036 reconstruct AS_PATH and AGGREGATOR attributes; RFC 6793 4.2.3 */
1037 if (!p
->as4_session
)
1038 bgp_process_as4_attrs(&attrs
, s
->pool
);
1040 /* Reject routes with our ASN in AS_PATH attribute */
1041 if (bgp_as_path_loopy(p
, attrs
, p
->local_as
))
1044 /* Reject routes with our Confederation ID in AS_PATH attribute; RFC 5065 4.0 */
1045 if ((p
->public_as
!= p
->local_as
) && bgp_as_path_loopy(p
, attrs
, p
->public_as
))
1048 /* Reject routes with our Router ID in ORIGINATOR_ID attribute; RFC 4456 8 */
1049 if (p
->is_internal
&& bgp_originator_id_loopy(p
, attrs
))
1052 /* Reject routes with our Cluster ID in CLUSTER_LIST attribute; RFC 4456 8 */
1053 if (p
->rr_client
&& bgp_cluster_list_loopy(p
, attrs
))
1056 /* If there is no local preference, define one */
1057 if (!BIT32_TEST(s
->attrs_seen
, BA_LOCAL_PREF
))
1058 bgp_set_attr_u32(&attrs
, s
->pool
, BA_LOCAL_PREF
, 0, p
->cf
->default_local_pref
);
1064 /* RFC 7606 4 - handle attribute framing errors */
1065 REPORT("Malformed attribute list - framing error (%u/%u) at %d",
1066 alen
, len
, (int) (pos
- s
->attrs
));
1069 /* RFC 7606 5.2 - handle missing NLRI during errors */
1070 if (!s
->ip_reach_len
&& !s
->mp_reach_len
)
1071 bgp_parse_error(s
, 1);
1073 s
->err_withdraw
= 1;
1079 * Route bucket hash table
1082 #define RBH_KEY(b) b->eattrs, b->hash
1083 #define RBH_NEXT(b) b->next
1084 #define RBH_EQ(a1,h1,a2,h2) h1 == h2 && ea_same(a1, a2)
1085 #define RBH_FN(a,h) h
1087 #define RBH_REHASH bgp_rbh_rehash
1088 #define RBH_PARAMS /8, *2, 2, 2, 8, 20
1091 HASH_DEFINE_REHASH_FN(RBH
, struct bgp_bucket
)
1094 bgp_init_bucket_table(struct bgp_channel
*c
)
1096 HASH_INIT(c
->bucket_hash
, c
->pool
, 8);
1098 init_list(&c
->bucket_queue
);
1099 c
->withdraw_bucket
= NULL
;
1102 static struct bgp_bucket
*
1103 bgp_get_bucket(struct bgp_channel
*c
, ea_list
*new)
1105 /* Hash and lookup */
1106 u32 hash
= ea_hash(new);
1107 struct bgp_bucket
*b
= HASH_FIND(c
->bucket_hash
, RBH
, new, hash
);
1112 uint ea_size
= sizeof(ea_list
) + new->count
* sizeof(eattr
);
1113 uint ea_size_aligned
= BIRD_ALIGN(ea_size
, CPU_STRUCT_ALIGN
);
1114 uint size
= sizeof(struct bgp_bucket
) + ea_size_aligned
;
1118 /* Gather total size of non-inline attributes */
1119 for (i
= 0; i
< new->count
; i
++)
1121 eattr
*a
= &new->attrs
[i
];
1123 if (!(a
->type
& EAF_EMBEDDED
))
1124 size
+= BIRD_ALIGN(sizeof(struct adata
) + a
->u
.ptr
->length
, CPU_STRUCT_ALIGN
);
1127 /* Create the bucket */
1128 b
= mb_alloc(c
->pool
, size
);
1129 init_list(&b
->prefixes
);
1132 /* Copy list of extended attributes */
1133 memcpy(b
->eattrs
, new, ea_size
);
1134 dest
= ((byte
*) b
->eattrs
) + ea_size_aligned
;
1136 /* Copy values of non-inline attributes */
1137 for (i
= 0; i
< new->count
; i
++)
1139 eattr
*a
= &b
->eattrs
->attrs
[i
];
1141 if (!(a
->type
& EAF_EMBEDDED
))
1143 struct adata
*oa
= a
->u
.ptr
;
1144 struct adata
*na
= (struct adata
*) dest
;
1145 memcpy(na
, oa
, sizeof(struct adata
) + oa
->length
);
1147 dest
+= BIRD_ALIGN(sizeof(struct adata
) + na
->length
, CPU_STRUCT_ALIGN
);
1151 /* Insert the bucket to send queue and bucket hash */
1152 add_tail(&c
->bucket_queue
, &b
->send_node
);
1153 HASH_INSERT2(c
->bucket_hash
, RBH
, c
->pool
, b
);
1158 static struct bgp_bucket
*
1159 bgp_get_withdraw_bucket(struct bgp_channel
*c
)
1161 if (!c
->withdraw_bucket
)
1163 c
->withdraw_bucket
= mb_allocz(c
->pool
, sizeof(struct bgp_bucket
));
1164 init_list(&c
->withdraw_bucket
->prefixes
);
1167 return c
->withdraw_bucket
;
1171 bgp_free_bucket(struct bgp_channel
*c
, struct bgp_bucket
*b
)
1173 rem_node(&b
->send_node
);
1174 HASH_REMOVE2(c
->bucket_hash
, RBH
, c
->pool
, b
);
1179 bgp_defer_bucket(struct bgp_channel
*c
, struct bgp_bucket
*b
)
1181 rem_node(&b
->send_node
);
1182 add_tail(&c
->bucket_queue
, &b
->send_node
);
1186 bgp_withdraw_bucket(struct bgp_channel
*c
, struct bgp_bucket
*b
)
1188 struct bgp_proto
*p
= (void *) c
->c
.proto
;
1189 struct bgp_bucket
*wb
= bgp_get_withdraw_bucket(c
);
1191 log(L_ERR
"%s: Attribute list too long", p
->p
.name
);
1192 while (!EMPTY_LIST(b
->prefixes
))
1194 struct bgp_prefix
*px
= HEAD(b
->prefixes
);
1196 log(L_ERR
"%s: - withdrawing %N", p
->p
.name
, &px
->net
);
1197 rem_node(&px
->buck_node
);
1198 add_tail(&wb
->prefixes
, &px
->buck_node
);
1207 #define PXH_KEY(px) px->net, px->path_id, px->hash
1208 #define PXH_NEXT(px) px->next
1209 #define PXH_EQ(n1,i1,h1,n2,i2,h2) h1 == h2 && i1 == i2 && net_equal(n1, n2)
1210 #define PXH_FN(n,i,h) h
1212 #define PXH_REHASH bgp_pxh_rehash
1213 #define PXH_PARAMS /8, *2, 2, 2, 8, 20
1216 HASH_DEFINE_REHASH_FN(PXH
, struct bgp_prefix
)
1219 bgp_init_prefix_table(struct bgp_channel
*c
)
1221 HASH_INIT(c
->prefix_hash
, c
->pool
, 8);
1223 uint alen
= net_addr_length
[c
->c
.net_type
];
1224 c
->prefix_slab
= alen
? sl_new(c
->pool
, sizeof(struct bgp_prefix
) + alen
) : NULL
;
1227 static struct bgp_prefix
*
1228 bgp_get_prefix(struct bgp_channel
*c
, net_addr
*net
, u32 path_id
)
1230 u32 hash
= net_hash(net
) ^ u32_hash(path_id
);
1231 struct bgp_prefix
*px
= HASH_FIND(c
->prefix_hash
, PXH
, net
, path_id
, hash
);
1235 rem_node(&px
->buck_node
);
1240 px
= sl_alloc(c
->prefix_slab
);
1242 px
= mb_alloc(c
->pool
, sizeof(struct bgp_prefix
) + net
->length
);
1244 px
->buck_node
.next
= NULL
;
1245 px
->buck_node
.prev
= NULL
;
1247 px
->path_id
= path_id
;
1248 net_copy(px
->net
, net
);
1250 HASH_INSERT2(c
->prefix_hash
, PXH
, c
->pool
, px
);
1256 bgp_free_prefix(struct bgp_channel
*c
, struct bgp_prefix
*px
)
1258 rem_node(&px
->buck_node
);
1259 HASH_REMOVE2(c
->prefix_hash
, PXH
, c
->pool
, px
);
1262 sl_free(c
->prefix_slab
, px
);
1273 bgp_import_control(struct proto
*P
, rte
**new, ea_list
**attrs UNUSED
, struct linpool
*pool UNUSED
)
1276 struct proto
*SRC
= e
->attrs
->src
->proto
;
1277 struct bgp_proto
*p
= (struct bgp_proto
*) P
;
1278 struct bgp_proto
*src
= (SRC
->proto
== &proto_bgp
) ? (struct bgp_proto
*) SRC
: NULL
;
1280 /* Reject our routes */
1284 /* Accept non-BGP routes */
1288 // XXXX: Check next hop AF
1290 /* IBGP route reflection, RFC 4456 */
1291 if (p
->is_internal
&& src
->is_internal
&& (p
->local_as
== src
->local_as
))
1293 /* Rejected unless configured as route reflector */
1294 if (!p
->rr_client
&& !src
->rr_client
)
1297 /* Generally, this should be handled when path is received, but we check it
1298 also here as rr_cluster_id may be undefined or different in src. */
1299 if (p
->rr_cluster_id
&& bgp_cluster_list_loopy(p
, e
->attrs
->eattrs
))
1303 /* Handle well-known communities, RFC 1997 */
1305 if (p
->cf
->interpret_communities
&&
1306 (c
= ea_find(e
->attrs
->eattrs
, EA_CODE(EAP_BGP
, BA_COMMUNITY
))))
1308 struct adata
*d
= c
->u
.ptr
;
1310 /* Do not export anywhere */
1311 if (int_set_contains(d
, BGP_COMM_NO_ADVERTISE
))
1314 /* Do not export outside of AS (or member-AS) */
1315 if (!p
->is_internal
&& int_set_contains(d
, BGP_COMM_NO_EXPORT_SUBCONFED
))
1318 /* Do not export outside of AS (or confederation) */
1319 if (!p
->is_interior
&& int_set_contains(d
, BGP_COMM_NO_EXPORT
))
1326 static adata null_adata
; /* adata of length 0 */
1329 bgp_update_attrs(struct bgp_proto
*p
, struct bgp_channel
*c
, rte
*e
, ea_list
*attrs0
, struct linpool
*pool
)
1331 struct proto
*SRC
= e
->attrs
->src
->proto
;
1332 struct bgp_proto
*src
= (SRC
->proto
== &proto_bgp
) ? (void *) SRC
: NULL
;
1333 struct bgp_export_state s
= { .proto
= p
, .channel
=c
, .pool
= pool
, .src
= src
, .route
= e
};
1334 ea_list
*attrs
= attrs0
;
1338 /* ORIGIN attribute - mandatory, attach if missing */
1339 if (! bgp_find_attr(attrs0
, BA_ORIGIN
))
1340 bgp_set_attr_u32(&attrs
, pool
, BA_ORIGIN
, 0, src
? ORIGIN_INCOMPLETE
: ORIGIN_IGP
);
1342 /* AS_PATH attribute - mandatory */
1343 a
= bgp_find_attr(attrs0
, BA_AS_PATH
);
1344 ad
= a
? a
->u
.ptr
: &null_adata
;
1346 /* AS_PATH attribute - strip AS_CONFED* segments outside confederation */
1347 if ((!p
->cf
->confederation
|| !p
->is_interior
) && as_path_contains_confed(ad
))
1348 ad
= as_path_strip_confed(pool
, ad
);
1350 /* AS_PATH attribute - keep or prepend ASN */
1351 if (p
->is_internal
||
1352 (p
->rs_client
&& src
&& src
->rs_client
))
1354 /* IBGP or route server -> just ensure there is one */
1356 bgp_set_attr_ptr(&attrs
, pool
, BA_AS_PATH
, 0, &null_adata
);
1358 else if (p
->is_interior
)
1360 /* Confederation -> prepend ASN as AS_CONFED_SEQUENCE */
1361 ad
= as_path_prepend2(pool
, ad
, AS_PATH_CONFED_SEQUENCE
, p
->public_as
);
1362 bgp_set_attr_ptr(&attrs
, pool
, BA_AS_PATH
, 0, ad
);
1364 else /* Regular EBGP (no RS, no confederation) */
1366 /* Regular EBGP -> prepend ASN as regular sequence */
1367 ad
= as_path_prepend2(pool
, ad
, AS_PATH_SEQUENCE
, p
->public_as
);
1368 bgp_set_attr_ptr(&attrs
, pool
, BA_AS_PATH
, 0, ad
);
1370 /* MULTI_EXIT_DESC attribute - accept only if set in export filter */
1371 a
= bgp_find_attr(attrs0
, BA_MULTI_EXIT_DISC
);
1372 if (a
&& !(a
->type
& EAF_FRESH
))
1373 bgp_unset_attr(&attrs
, pool
, BA_MULTI_EXIT_DISC
);
1376 /* NEXT_HOP attribute - delegated to AF-specific hook */
1377 a
= bgp_find_attr(attrs0
, BA_NEXT_HOP
);
1378 bgp_update_next_hop(&s
, a
, &attrs
);
1380 /* LOCAL_PREF attribute - required for IBGP, attach if missing */
1381 if (p
->is_interior
&& ! bgp_find_attr(attrs0
, BA_LOCAL_PREF
))
1382 bgp_set_attr_u32(&attrs
, pool
, BA_LOCAL_PREF
, 0, p
->cf
->default_local_pref
);
1384 /* IBGP route reflection, RFC 4456 */
1385 if (src
&& src
->is_internal
&& p
->is_internal
&& (src
->local_as
== p
->local_as
))
1387 /* ORIGINATOR_ID attribute - attach if not already set */
1388 if (! bgp_find_attr(attrs0
, BA_ORIGINATOR_ID
))
1389 bgp_set_attr_u32(&attrs
, pool
, BA_ORIGINATOR_ID
, 0, src
->remote_id
);
1391 /* CLUSTER_LIST attribute - prepend cluster ID */
1392 a
= bgp_find_attr(attrs0
, BA_CLUSTER_LIST
);
1393 ad
= a
? a
->u
.ptr
: NULL
;
1395 /* Prepend src cluster ID */
1396 if (src
->rr_cluster_id
)
1397 ad
= int_set_add(pool
, ad
, src
->rr_cluster_id
);
1399 /* Prepend dst cluster ID if src and dst clusters are different */
1400 if (p
->rr_cluster_id
&& (src
->rr_cluster_id
!= p
->rr_cluster_id
))
1401 ad
= int_set_add(pool
, ad
, p
->rr_cluster_id
);
1403 /* Should be at least one prepended cluster ID */
1404 bgp_set_attr_ptr(&attrs
, pool
, BA_CLUSTER_LIST
, 0, ad
);
1407 /* AS4_* transition attributes, RFC 6793 4.2.2 */
1408 if (! p
->as4_session
)
1410 a
= bgp_find_attr(attrs
, BA_AS_PATH
);
1411 if (a
&& as_path_contains_as4(a
->u
.ptr
))
1413 bgp_set_attr_ptr(&attrs
, pool
, BA_AS_PATH
, 0, as_path_to_old(pool
, a
->u
.ptr
));
1414 bgp_set_attr_ptr(&attrs
, pool
, BA_AS4_PATH
, 0, as_path_strip_confed(pool
, a
->u
.ptr
));
1417 a
= bgp_find_attr(attrs
, BA_AGGREGATOR
);
1418 if (a
&& aggregator_contains_as4(a
->u
.ptr
))
1420 bgp_set_attr_ptr(&attrs
, pool
, BA_AGGREGATOR
, 0, aggregator_to_old(pool
, a
->u
.ptr
));
1421 bgp_set_attr_ptr(&attrs
, pool
, BA_AS4_AGGREGATOR
, 0, a
->u
.ptr
);
1426 * Presence of mandatory attributes ORIGIN and AS_PATH is ensured by above
1427 * conditions. Presence and validity of quasi-mandatory NEXT_HOP attribute
1428 * should be checked in AF-specific hooks.
1431 /* Apply per-attribute export hooks for validatation and normalization */
1432 return bgp_export_attrs(&s
, attrs
);
1436 bgp_rt_notify(struct proto
*P
, struct channel
*C
, net
*n
, rte
*new, rte
*old
, ea_list
*attrs
)
1438 struct bgp_proto
*p
= (void *) P
;
1439 struct bgp_channel
*c
= (void *) C
;
1440 struct bgp_bucket
*buck
;
1441 struct bgp_prefix
*px
;
1446 attrs
= bgp_update_attrs(p
, c
, new, attrs
, bgp_linpool
);
1448 /* If attributes are invalid, we fail back to withdraw */
1449 buck
= attrs
? bgp_get_bucket(c
, attrs
) : bgp_get_withdraw_bucket(c
);
1450 path
= new->attrs
->src
->global_id
;
1452 lp_flush(bgp_linpool
);
1456 buck
= bgp_get_withdraw_bucket(c
);
1457 path
= old
->attrs
->src
->global_id
;
1460 px
= bgp_get_prefix(c
, n
->n
.addr
, c
->add_path_tx
? path
: 0);
1461 add_tail(&buck
->prefixes
, &px
->buck_node
);
1463 bgp_schedule_packet(p
->conn
, c
, PKT_UPDATE
);
1468 bgp_get_neighbor(rte
*r
)
1470 eattr
*e
= ea_find(r
->attrs
->eattrs
, EA_CODE(EAP_BGP
, BA_AS_PATH
));
1473 if (e
&& as_path_get_first_regular(e
->u
.ptr
, &as
))
1476 /* If AS_PATH is not defined, we treat rte as locally originated */
1477 struct bgp_proto
*p
= (void *) r
->attrs
->src
->proto
;
1478 return p
->cf
->confederation
?: p
->local_as
;
1482 rte_resolvable(rte
*rt
)
1484 int rd
= rt
->attrs
->dest
;
1485 return (rd
== RTD_ROUTER
) || (rd
== RTD_DEVICE
) || (rd
== RTD_MULTIPATH
);
1489 bgp_rte_better(rte
*new, rte
*old
)
1491 struct bgp_proto
*new_bgp
= (struct bgp_proto
*) new->attrs
->src
->proto
;
1492 struct bgp_proto
*old_bgp
= (struct bgp_proto
*) old
->attrs
->src
->proto
;
1496 /* Skip suppressed routes (see bgp_rte_recalculate()) */
1497 n
= new->u
.bgp
.suppressed
;
1498 o
= old
->u
.bgp
.suppressed
;
1504 /* RFC 4271 9.1.2.1. Route resolvability test */
1505 n
= rte_resolvable(new);
1506 o
= rte_resolvable(old
);
1512 /* Start with local preferences */
1513 x
= ea_find(new->attrs
->eattrs
, EA_CODE(EAP_BGP
, BA_LOCAL_PREF
));
1514 y
= ea_find(old
->attrs
->eattrs
, EA_CODE(EAP_BGP
, BA_LOCAL_PREF
));
1515 n
= x
? x
->u
.data
: new_bgp
->cf
->default_local_pref
;
1516 o
= y
? y
->u
.data
: old_bgp
->cf
->default_local_pref
;
1522 /* RFC 4271 9.1.2.2. a) Use AS path lengths */
1523 if (new_bgp
->cf
->compare_path_lengths
|| old_bgp
->cf
->compare_path_lengths
)
1525 x
= ea_find(new->attrs
->eattrs
, EA_CODE(EAP_BGP
, BA_AS_PATH
));
1526 y
= ea_find(old
->attrs
->eattrs
, EA_CODE(EAP_BGP
, BA_AS_PATH
));
1527 n
= x
? as_path_getlen(x
->u
.ptr
) : AS_PATH_MAXLEN
;
1528 o
= y
? as_path_getlen(y
->u
.ptr
) : AS_PATH_MAXLEN
;
1535 /* RFC 4271 9.1.2.2. b) Use origins */
1536 x
= ea_find(new->attrs
->eattrs
, EA_CODE(EAP_BGP
, BA_ORIGIN
));
1537 y
= ea_find(old
->attrs
->eattrs
, EA_CODE(EAP_BGP
, BA_ORIGIN
));
1538 n
= x
? x
->u
.data
: ORIGIN_INCOMPLETE
;
1539 o
= y
? y
->u
.data
: ORIGIN_INCOMPLETE
;
1545 /* RFC 4271 9.1.2.2. c) Compare MED's */
1546 /* Proper RFC 4271 path selection cannot be interpreted as finding
1547 * the best path in some ordering. It is implemented partially in
1548 * bgp_rte_recalculate() when deterministic_med option is
1549 * active. Without that option, the behavior is just an
1550 * approximation, which in specific situations may lead to
1551 * persistent routing loops, because it is nondeterministic - it
1552 * depends on the order in which routes appeared. But it is also the
1553 * same behavior as used by default in Cisco routers, so it is
1554 * probably not a big issue.
1556 if (new_bgp
->cf
->med_metric
|| old_bgp
->cf
->med_metric
||
1557 (bgp_get_neighbor(new) == bgp_get_neighbor(old
)))
1559 x
= ea_find(new->attrs
->eattrs
, EA_CODE(EAP_BGP
, BA_MULTI_EXIT_DISC
));
1560 y
= ea_find(old
->attrs
->eattrs
, EA_CODE(EAP_BGP
, BA_MULTI_EXIT_DISC
));
1561 n
= x
? x
->u
.data
: new_bgp
->cf
->default_med
;
1562 o
= y
? y
->u
.data
: old_bgp
->cf
->default_med
;
1569 /* RFC 4271 9.1.2.2. d) Prefer external peers */
1570 if (new_bgp
->is_interior
> old_bgp
->is_interior
)
1572 if (new_bgp
->is_interior
< old_bgp
->is_interior
)
1575 /* RFC 4271 9.1.2.2. e) Compare IGP metrics */
1576 n
= new_bgp
->cf
->igp_metric
? new->attrs
->igp_metric
: 0;
1577 o
= old_bgp
->cf
->igp_metric
? old
->attrs
->igp_metric
: 0;
1583 /* RFC 4271 9.1.2.2. f) Compare BGP identifiers */
1584 /* RFC 4456 9. a) Use ORIGINATOR_ID instead of local neighbor ID */
1585 x
= ea_find(new->attrs
->eattrs
, EA_CODE(EAP_BGP
, BA_ORIGINATOR_ID
));
1586 y
= ea_find(old
->attrs
->eattrs
, EA_CODE(EAP_BGP
, BA_ORIGINATOR_ID
));
1587 n
= x
? x
->u
.data
: new_bgp
->remote_id
;
1588 o
= y
? y
->u
.data
: old_bgp
->remote_id
;
1590 /* RFC 5004 - prefer older routes */
1591 /* (if both are external and from different peer) */
1592 if ((new_bgp
->cf
->prefer_older
|| old_bgp
->cf
->prefer_older
) &&
1593 !new_bgp
->is_internal
&& n
!= o
)
1596 /* rest of RFC 4271 9.1.2.2. f) */
1602 /* RFC 4456 9. b) Compare cluster list lengths */
1603 x
= ea_find(new->attrs
->eattrs
, EA_CODE(EAP_BGP
, BA_CLUSTER_LIST
));
1604 y
= ea_find(old
->attrs
->eattrs
, EA_CODE(EAP_BGP
, BA_CLUSTER_LIST
));
1605 n
= x
? int_set_get_size(x
->u
.ptr
) : 0;
1606 o
= y
? int_set_get_size(y
->u
.ptr
) : 0;
1612 /* RFC 4271 9.1.2.2. g) Compare peer IP adresses */
1613 return (ipa_compare(new_bgp
->cf
->remote_ip
, old_bgp
->cf
->remote_ip
) < 0);
1618 bgp_rte_mergable(rte
*pri
, rte
*sec
)
1620 struct bgp_proto
*pri_bgp
= (struct bgp_proto
*) pri
->attrs
->src
->proto
;
1621 struct bgp_proto
*sec_bgp
= (struct bgp_proto
*) sec
->attrs
->src
->proto
;
1625 /* Skip suppressed routes (see bgp_rte_recalculate()) */
1626 if (pri
->u
.bgp
.suppressed
!= sec
->u
.bgp
.suppressed
)
1629 /* RFC 4271 9.1.2.1. Route resolvability test */
1630 if (!rte_resolvable(sec
))
1633 /* Start with local preferences */
1634 x
= ea_find(pri
->attrs
->eattrs
, EA_CODE(EAP_BGP
, BA_LOCAL_PREF
));
1635 y
= ea_find(sec
->attrs
->eattrs
, EA_CODE(EAP_BGP
, BA_LOCAL_PREF
));
1636 p
= x
? x
->u
.data
: pri_bgp
->cf
->default_local_pref
;
1637 s
= y
? y
->u
.data
: sec_bgp
->cf
->default_local_pref
;
1641 /* RFC 4271 9.1.2.2. a) Use AS path lengths */
1642 if (pri_bgp
->cf
->compare_path_lengths
|| sec_bgp
->cf
->compare_path_lengths
)
1644 x
= ea_find(pri
->attrs
->eattrs
, EA_CODE(EAP_BGP
, BA_AS_PATH
));
1645 y
= ea_find(sec
->attrs
->eattrs
, EA_CODE(EAP_BGP
, BA_AS_PATH
));
1646 p
= x
? as_path_getlen(x
->u
.ptr
) : AS_PATH_MAXLEN
;
1647 s
= y
? as_path_getlen(y
->u
.ptr
) : AS_PATH_MAXLEN
;
1652 // if (DELTA(p, s) > pri_bgp->cf->relax_multipath)
1656 /* RFC 4271 9.1.2.2. b) Use origins */
1657 x
= ea_find(pri
->attrs
->eattrs
, EA_CODE(EAP_BGP
, BA_ORIGIN
));
1658 y
= ea_find(sec
->attrs
->eattrs
, EA_CODE(EAP_BGP
, BA_ORIGIN
));
1659 p
= x
? x
->u
.data
: ORIGIN_INCOMPLETE
;
1660 s
= y
? y
->u
.data
: ORIGIN_INCOMPLETE
;
1664 /* RFC 4271 9.1.2.2. c) Compare MED's */
1665 if (pri_bgp
->cf
->med_metric
|| sec_bgp
->cf
->med_metric
||
1666 (bgp_get_neighbor(pri
) == bgp_get_neighbor(sec
)))
1668 x
= ea_find(pri
->attrs
->eattrs
, EA_CODE(EAP_BGP
, BA_MULTI_EXIT_DISC
));
1669 y
= ea_find(sec
->attrs
->eattrs
, EA_CODE(EAP_BGP
, BA_MULTI_EXIT_DISC
));
1670 p
= x
? x
->u
.data
: pri_bgp
->cf
->default_med
;
1671 s
= y
? y
->u
.data
: sec_bgp
->cf
->default_med
;
1676 /* RFC 4271 9.1.2.2. d) Prefer external peers */
1677 if (pri_bgp
->is_interior
!= sec_bgp
->is_interior
)
1680 /* RFC 4271 9.1.2.2. e) Compare IGP metrics */
1681 p
= pri_bgp
->cf
->igp_metric
? pri
->attrs
->igp_metric
: 0;
1682 s
= sec_bgp
->cf
->igp_metric
? sec
->attrs
->igp_metric
: 0;
1686 /* Remaining criteria are ignored */
1693 same_group(rte
*r
, u32 lpref
, u32 lasn
)
1695 return (r
->pref
== lpref
) && (bgp_get_neighbor(r
) == lasn
);
1699 use_deterministic_med(rte
*r
)
1701 struct proto
*P
= r
->attrs
->src
->proto
;
1702 return (P
->proto
== &proto_bgp
) && ((struct bgp_proto
*) P
)->cf
->deterministic_med
;
1706 bgp_rte_recalculate(rtable
*table
, net
*net
, rte
*new, rte
*old
, rte
*old_best
)
1709 rte
*key
= new ? new : old
;
1710 u32 lpref
= key
->pref
;
1711 u32 lasn
= bgp_get_neighbor(key
);
1712 int old_is_group_best
= 0;
1715 * Proper RFC 4271 path selection is a bit complicated, it cannot be
1716 * implemented just by rte_better(), because it is not a linear
1717 * ordering. But it can be splitted to two levels, where the lower
1718 * level chooses the best routes in each group of routes from the
1719 * same neighboring AS and higher level chooses the best route (with
1720 * a slightly different ordering) between the best-in-group routes.
1722 * When deterministic_med is disabled, we just ignore this issue and
1723 * choose the best route by bgp_rte_better() alone. If enabled, the
1724 * lower level of the route selection is done here (for the group
1725 * to which the changed route belongs), all routes in group are
1726 * marked as suppressed, just chosen best-in-group is not.
1728 * Global best route selection then implements higher level by
1729 * choosing between non-suppressed routes (as they are always
1730 * preferred over suppressed routes). Routes from BGP protocols
1731 * that do not set deterministic_med are just never suppressed. As
1732 * they do not participate in the lower level selection, it is OK
1733 * that this fn is not called for them.
1735 * The idea is simple, the implementation is more problematic,
1736 * mostly because of optimizations in rte_recalculate() that
1737 * avoids full recalculation in most cases.
1739 * We can assume that at least one of new, old is non-NULL and both
1740 * are from the same protocol with enabled deterministic_med. We
1741 * group routes by both neighbor AS (lasn) and preference (lpref),
1742 * because bgp_rte_better() does not handle preference itself.
1745 /* If new and old are from different groups, we just process that
1746 as two independent events */
1747 if (new && old
&& !same_group(old
, lpref
, lasn
))
1750 i1
= bgp_rte_recalculate(table
, net
, NULL
, old
, old_best
);
1751 i2
= bgp_rte_recalculate(table
, net
, new, NULL
, old_best
);
1756 * We could find the best-in-group and then make some shortcuts like
1757 * in rte_recalculate, but as we would have to walk through all
1758 * net->routes just to find it, it is probably not worth. So we
1759 * just have two simpler fast cases that use just the old route.
1760 * We also set suppressed flag to avoid using it in bgp_rte_better().
1764 new->u
.bgp
.suppressed
= 1;
1768 old_is_group_best
= !old
->u
.bgp
.suppressed
;
1769 old
->u
.bgp
.suppressed
= 1;
1770 int new_is_better
= new && bgp_rte_better(new, old
);
1772 /* The first case - replace not best with worse (or remove not best) */
1773 if (!old_is_group_best
&& !new_is_better
)
1776 /* The second case - replace the best with better */
1777 if (old_is_group_best
&& new_is_better
)
1779 /* new is best-in-group, the see discussion below - this is
1780 a special variant of NBG && OBG. From OBG we can deduce
1781 that same_group(old_best) iff (old == old_best) */
1782 new->u
.bgp
.suppressed
= 0;
1783 return (old
== old_best
);
1787 /* The default case - find a new best-in-group route */
1788 r
= new; /* new may not be in the list */
1789 for (s
=net
->routes
; rte_is_valid(s
); s
=s
->next
)
1790 if (use_deterministic_med(s
) && same_group(s
, lpref
, lasn
))
1792 s
->u
.bgp
.suppressed
= 1;
1793 if (!r
|| bgp_rte_better(s
, r
))
1797 /* Simple case - the last route in group disappears */
1801 /* Found best-in-group */
1802 r
->u
.bgp
.suppressed
= 0;
1805 * There are generally two reasons why we have to force
1806 * recalculation (return 1): First, the new route may be wrongfully
1807 * chosen to be the best in the first case check in
1808 * rte_recalculate(), this may happen only if old_best is from the
1809 * same group. Second, another (different than new route)
1810 * best-in-group is chosen and that may be the proper best (although
1811 * rte_recalculate() without ignore that possibility).
1813 * There are three possible cases according to whether the old route
1814 * was the best in group (OBG, stored in old_is_group_best) and
1815 * whether the new route is the best in group (NBG, tested by r == new).
1816 * These cases work even if old or new is NULL.
1818 * NBG -> new is a possible candidate for the best route, so we just
1819 * check for the first reason using same_group().
1821 * !NBG && OBG -> Second reason applies, return 1
1823 * !NBG && !OBG -> Best in group does not change, old != old_best,
1824 * rte_better(new, old_best) is false and therefore
1825 * the first reason does not apply, return 0
1829 return old_best
&& same_group(old_best
, lpref
, lasn
);
1831 return old_is_group_best
;
1836 * Reconstruct AS_PATH and AGGREGATOR according to RFC 6793 4.2.3
1839 bgp_process_as4_attrs(ea_list
**attrs
, struct linpool
*pool
)
1841 eattr
*p2
= bgp_find_attr(*attrs
, BA_AS_PATH
);
1842 eattr
*p4
= bgp_find_attr(*attrs
, BA_AS4_PATH
);
1843 eattr
*a2
= bgp_find_attr(*attrs
, BA_AGGREGATOR
);
1844 eattr
*a4
= bgp_find_attr(*attrs
, BA_AS4_AGGREGATOR
);
1846 /* First, unset AS4_* attributes */
1847 if (p4
) bgp_unset_attr(attrs
, pool
, BA_AS4_PATH
);
1848 if (a4
) bgp_unset_attr(attrs
, pool
, BA_AS4_AGGREGATOR
);
1850 /* Handle AGGREGATOR attribute */
1853 u32 a2_asn
= get_u32(a2
->u
.ptr
->data
);
1855 /* If routes were aggregated by an old router, then AS4_PATH and
1856 AS4_AGGREGATOR are invalid. In that case we give up. */
1857 if (a2_asn
!= AS_TRANS
)
1860 /* Use AS4_AGGREGATOR instead of AGGREGATOR */
1861 a2
->u
.ptr
= a4
->u
.ptr
;
1864 /* Handle AS_PATH attribute */
1867 /* Both as_path_getlen() and as_path_cut() take AS_CONFED* as zero length */
1868 int p2_len
= as_path_getlen(p2
->u
.ptr
);
1869 int p4_len
= as_path_getlen(p4
->u
.ptr
);
1871 /* AS_PATH is too short, give up */
1872 if (p2_len
< p4_len
)
1875 /* Merge AS_PATH and AS4_PATH */
1876 as_path_cut(p2
->u
.ptr
, p2_len
- p4_len
);
1877 p2
->u
.ptr
= as_path_merge(pool
, p2
->u
.ptr
, p4
->u
.ptr
);
1882 bgp_get_attr(eattr
*a
, byte
*buf
, int buflen
)
1884 uint i
= EA_ID(a
->id
);
1885 const struct bgp_attr_desc
*d
;
1888 if (bgp_attr_known(i
))
1890 d
= &bgp_attr_table
[i
];
1891 len
= bsprintf(buf
, "%s", d
->name
);
1897 d
->format(a
, buf
, buflen
- len
- 2);
1903 bsprintf(buf
, "%02x%s", i
, (a
->flags
& BAF_TRANSITIVE
) ? " [t]" : "");
1908 bgp_get_route_info(rte
*e
, byte
*buf
, ea_list
*attrs
)
1910 eattr
*p
= ea_find(attrs
, EA_CODE(EAP_BGP
, BA_AS_PATH
));
1911 eattr
*o
= ea_find(attrs
, EA_CODE(EAP_BGP
, BA_ORIGIN
));
1914 buf
+= bsprintf(buf
, " (%d", e
->pref
);
1916 if (e
->u
.bgp
.suppressed
)
1917 buf
+= bsprintf(buf
, "-");
1919 if (e
->attrs
->hostentry
)
1921 if (!rte_resolvable(e
))
1922 buf
+= bsprintf(buf
, "/-");
1923 else if (e
->attrs
->igp_metric
>= IGP_METRIC_UNKNOWN
)
1924 buf
+= bsprintf(buf
, "/?");
1926 buf
+= bsprintf(buf
, "/%d", e
->attrs
->igp_metric
);
1928 buf
+= bsprintf(buf
, ") [");
1930 if (p
&& as_path_get_last(p
->u
.ptr
, &origas
))
1931 buf
+= bsprintf(buf
, "AS%u", origas
);
1933 buf
+= bsprintf(buf
, "%c", "ie?"[o
->u
.data
]);