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36 * Loop detection based on AS_PATH causes updates to be withdrawn.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40 * and CLUSTER_LIST) causes updates to be withdrawn.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.
68 struct bgp_attr_desc
{
72 void (*export
)(struct bgp_export_state
*s
, eattr
*a
);
73 int (*encode
)(struct bgp_write_state
*s
, eattr
*a
, byte
*buf
, uint size
);
74 void (*decode
)(struct bgp_parse_state
*s
, uint code
, uint flags
, byte
*data
, uint len
, ea_list
**to
);
75 void (*format
)(eattr
*ea
, byte
*buf
, uint size
);
78 static const struct bgp_attr_desc bgp_attr_table
[];
80 static inline int bgp_attr_known(uint code
);
83 bgp_set_attr(ea_list
**attrs
, struct linpool
*pool
, uint code
, uint flags
, uintptr_t val
)
85 ASSERT(bgp_attr_known(code
));
87 ea_list
*a
= lp_alloc(pool
, sizeof(ea_list
) + sizeof(eattr
));
88 eattr
*e
= &a
->attrs
[0];
90 a
->flags
= EALF_SORTED
;
95 e
->id
= EA_CODE(EAP_BGP
, code
);
96 e
->type
= bgp_attr_table
[code
].type
;
99 if (e
->type
& EAF_EMBEDDED
)
100 e
->u
.data
= (u32
) val
;
102 e
->u
.ptr
= (struct adata
*) val
;
109 #define REPORT(msg, args...) \
110 ({ log(L_REMOTE "%s: " msg, s->proto->p.name, ## args); })
112 #define DISCARD(msg, args...) \
113 ({ REPORT(msg, ## args); return; })
115 #define WITHDRAW(msg, args...) \
116 ({ REPORT(msg, ## args); s->err_withdraw = 1; return; })
119 ({ a->type = EAF_TYPE_UNDEF; return; })
121 #define NEW_BGP "Discarding %s attribute received from AS4-aware neighbor"
122 #define BAD_EBGP "Discarding %s attribute received from EBGP neighbor"
123 #define BAD_LENGTH "Malformed %s attribute - invalid length (%u)"
124 #define BAD_VALUE "Malformed %s attribute - invalid value (%u)"
125 #define NO_MANDATORY "Missing mandatory %s attribute"
129 bgp_put_attr_hdr3(byte
*buf
, uint code
, uint flags
, uint len
)
138 bgp_put_attr_hdr4(byte
*buf
, uint code
, uint flags
, uint len
)
140 *buf
++ = flags
| BAF_EXT_LEN
;
147 bgp_put_attr_hdr(byte
*buf
, uint code
, uint flags
, uint len
)
150 return bgp_put_attr_hdr3(buf
, code
, flags
, len
);
152 return bgp_put_attr_hdr4(buf
, code
, flags
, len
);
156 bgp_encode_u8(struct bgp_write_state
*s UNUSED
, eattr
*a
, byte
*buf
, uint size
)
161 bgp_put_attr_hdr3(buf
, EA_ID(a
->id
), a
->flags
, 1);
168 bgp_encode_u32(struct bgp_write_state
*s UNUSED
, eattr
*a
, byte
*buf
, uint size
)
173 bgp_put_attr_hdr3(buf
, EA_ID(a
->id
), a
->flags
, 4);
174 put_u32(buf
+3, a
->u
.data
);
180 bgp_encode_u32s(struct bgp_write_state
*s UNUSED
, eattr
*a
, byte
*buf
, uint size
)
182 uint len
= a
->u
.ptr
->length
;
187 uint hdr
= bgp_put_attr_hdr(buf
, EA_ID(a
->id
), a
->flags
, len
);
188 put_u32s(buf
+ hdr
, (u32
*) a
->u
.ptr
->data
, len
/ 4);
194 bgp_put_attr(byte
*buf
, uint size
, uint code
, uint flags
, byte
*data
, uint len
)
199 uint hdr
= bgp_put_attr_hdr(buf
, code
, flags
, len
);
200 memcpy(buf
+ hdr
, data
, len
);
206 bgp_encode_raw(struct bgp_write_state
*s UNUSED
, eattr
*a
, byte
*buf
, uint size
)
208 return bgp_put_attr(buf
, size
, EA_ID(a
->id
), a
->flags
, a
->u
.ptr
->data
, a
->u
.ptr
->length
);
217 bgp_export_origin(struct bgp_export_state
*s
, eattr
*a
)
220 WITHDRAW(BAD_VALUE
, "ORIGIN", a
->u
.data
);
224 bgp_decode_origin(struct bgp_parse_state
*s
, uint code UNUSED
, uint flags
, byte
*data
, uint len
, ea_list
**to
)
227 WITHDRAW(BAD_LENGTH
, "ORIGIN", len
);
230 WITHDRAW(BAD_VALUE
, "ORIGIN", data
[0]);
232 bgp_set_attr_u32(to
, s
->pool
, BA_ORIGIN
, flags
, data
[0]);
236 bgp_format_origin(eattr
*a
, byte
*buf
, uint size UNUSED
)
238 static const char *bgp_origin_names
[] = { "IGP", "EGP", "Incomplete" };
240 bsprintf(buf
, (a
->u
.data
<= 2) ? bgp_origin_names
[a
->u
.data
] : "?");
245 bgp_encode_as_path(struct bgp_write_state
*s
, eattr
*a
, byte
*buf
, uint size
)
247 byte
*data
= a
->u
.ptr
->data
;
248 uint len
= a
->u
.ptr
->length
;
252 /* Prepare 16-bit AS_PATH (from 32-bit one) in a temporary buffer */
255 len
= as_path_32to16(data
, src
, len
);
258 return bgp_put_attr(buf
, size
, BA_AS_PATH
, a
->flags
, data
, len
);
262 bgp_decode_as_path(struct bgp_parse_state
*s
, uint code UNUSED
, uint flags
, byte
*data
, uint len
, ea_list
**to
)
264 struct bgp_proto
*p
= s
->proto
;
265 int as_length
= s
->as4_session
? 4 : 2;
266 int as_confed
= p
->cf
->confederation
&& p
->is_interior
;
269 if (!as_path_valid(data
, len
, as_length
, as_confed
, err
, sizeof(err
)))
270 WITHDRAW("Malformed AS_PATH attribute - %s", err
);
272 /* In some circumstances check for initial AS_CONFED_SEQUENCE; RFC 5065 5.0 */
273 if (p
->is_interior
&& !p
->is_internal
&&
274 ((len
< 2) || (data
[0] != AS_PATH_CONFED_SEQUENCE
)))
275 WITHDRAW("Malformed AS_PATH attribute - %s", "missing initial AS_CONFED_SEQUENCE");
279 /* Prepare 32-bit AS_PATH (from 16-bit one) in a temporary buffer */
281 data
= alloca(2*len
);
282 len
= as_path_16to32(data
, src
, len
);
285 bgp_set_attr_data(to
, s
->pool
, BA_AS_PATH
, flags
, data
, len
);
290 bgp_encode_next_hop(struct bgp_write_state
*s
, eattr
*a
, byte
*buf
, uint size
)
293 * The NEXT_HOP attribute is used only in traditional (IPv4) BGP. In MP-BGP,
294 * the next hop is encoded as a part of the MP_REACH_NLRI attribute, so we
295 * store it and encode it later by AFI-specific hooks.
298 if ((s
->channel
->afi
== BGP_AF_IPV4
) && !s
->channel
->ext_next_hop
)
300 ASSERT(a
->u
.ptr
->length
== sizeof(ip_addr
));
305 bgp_put_attr_hdr3(buf
, BA_NEXT_HOP
, a
->flags
, 4);
306 put_ip4(buf
+3, ipa_to_ip4( *(ip_addr
*) a
->u
.ptr
->data
));
318 bgp_decode_next_hop(struct bgp_parse_state
*s
, uint code UNUSED
, uint flags UNUSED
, byte
*data
, uint len
, ea_list
**to UNUSED
)
321 WITHDRAW(BAD_LENGTH
, "NEXT_HOP", len
);
323 /* Semantic checks are done later */
324 s
->ip_next_hop_len
= len
;
325 s
->ip_next_hop_data
= data
;
328 /* TODO: This function should use AF-specific hook */
330 bgp_format_next_hop(eattr
*a
, byte
*buf
, uint size UNUSED
)
332 ip_addr
*nh
= (void *) a
->u
.ptr
->data
;
333 uint len
= a
->u
.ptr
->length
;
335 ASSERT((len
== 16) || (len
== 32));
337 /* in IPv6, we may have two addresses in NEXT HOP */
338 if ((len
== 16) || ipa_zero(nh
[1]))
339 bsprintf(buf
, "%I", nh
[0]);
341 bsprintf(buf
, "%I %I", nh
[0], nh
[1]);
346 bgp_decode_med(struct bgp_parse_state
*s
, uint code UNUSED
, uint flags
, byte
*data
, uint len
, ea_list
**to
)
349 WITHDRAW(BAD_LENGTH
, "MULTI_EXIT_DISC", len
);
351 u32 val
= get_u32(data
);
352 bgp_set_attr_u32(to
, s
->pool
, BA_MULTI_EXIT_DISC
, flags
, val
);
357 bgp_export_local_pref(struct bgp_export_state
*s
, eattr
*a
)
359 if (!s
->proto
->is_interior
&& !s
->proto
->cf
->allow_local_pref
)
364 bgp_decode_local_pref(struct bgp_parse_state
*s
, uint code UNUSED
, uint flags
, byte
*data
, uint len
, ea_list
**to
)
366 if (!s
->proto
->is_interior
&& !s
->proto
->cf
->allow_local_pref
)
367 DISCARD(BAD_EBGP
, "LOCAL_PREF");
370 WITHDRAW(BAD_LENGTH
, "LOCAL_PREF", len
);
372 u32 val
= get_u32(data
);
373 bgp_set_attr_u32(to
, s
->pool
, BA_LOCAL_PREF
, flags
, val
);
378 bgp_decode_atomic_aggr(struct bgp_parse_state
*s
, uint code UNUSED
, uint flags
, byte
*data UNUSED
, uint len
, ea_list
**to
)
381 DISCARD(BAD_LENGTH
, "ATOMIC_AGGR", len
);
383 bgp_set_attr_data(to
, s
->pool
, BA_ATOMIC_AGGR
, flags
, NULL
, 0);
387 bgp_encode_aggregator(struct bgp_write_state
*s
, eattr
*a
, byte
*buf
, uint size
)
389 byte
*data
= a
->u
.ptr
->data
;
390 uint len
= a
->u
.ptr
->length
;
394 /* Prepare 16-bit AGGREGATOR (from 32-bit one) in a temporary buffer */
397 len
= aggregator_32to16(data
, src
);
400 return bgp_put_attr(buf
, size
, BA_AGGREGATOR
, a
->flags
, data
, len
);
404 bgp_decode_aggregator(struct bgp_parse_state
*s
, uint code UNUSED
, uint flags
, byte
*data
, uint len
, ea_list
**to
)
406 if (len
!= (s
->as4_session
? 8 : 6))
407 DISCARD(BAD_LENGTH
, "AGGREGATOR", len
);
411 /* Prepare 32-bit AGGREGATOR (from 16-bit one) in a temporary buffer */
414 len
= aggregator_16to32(data
, src
);
417 bgp_set_attr_data(to
, s
->pool
, BA_AGGREGATOR
, flags
, data
, len
);
421 bgp_format_aggregator(eattr
*a
, byte
*buf
, uint size UNUSED
)
423 byte
*data
= a
->u
.ptr
->data
;
425 bsprintf(buf
, "%I4 AS%u", get_ip4(data
+4), get_u32(data
+0));
430 bgp_export_community(struct bgp_export_state
*s
, eattr
*a
)
432 if (a
->u
.ptr
->length
== 0)
435 a
->u
.ptr
= int_set_sort(s
->pool
, a
->u
.ptr
);
439 bgp_decode_community(struct bgp_parse_state
*s
, uint code UNUSED
, uint flags
, byte
*data
, uint len
, ea_list
**to
)
441 if (!len
|| (len
% 4))
442 WITHDRAW(BAD_LENGTH
, "COMMUNITY", len
);
444 struct adata
*ad
= lp_alloc_adata(s
->pool
, len
);
445 get_u32s(data
, (u32
*) ad
->data
, len
/ 4);
446 bgp_set_attr_ptr(to
, s
->pool
, BA_COMMUNITY
, flags
, ad
);
451 bgp_export_originator_id(struct bgp_export_state
*s
, eattr
*a
)
453 if (!s
->proto
->is_internal
)
458 bgp_decode_originator_id(struct bgp_parse_state
*s
, uint code UNUSED
, uint flags
, byte
*data
, uint len
, ea_list
**to
)
460 if (!s
->proto
->is_internal
)
461 DISCARD(BAD_EBGP
, "ORIGINATOR_ID");
464 WITHDRAW(BAD_LENGTH
, "ORIGINATOR_ID", len
);
466 u32 val
= get_u32(data
);
467 bgp_set_attr_u32(to
, s
->pool
, BA_ORIGINATOR_ID
, flags
, val
);
472 bgp_export_cluster_list(struct bgp_export_state
*s UNUSED
, eattr
*a
)
474 if (!s
->proto
->is_internal
)
477 if (a
->u
.ptr
->length
== 0)
482 bgp_decode_cluster_list(struct bgp_parse_state
*s
, uint code UNUSED
, uint flags
, byte
*data
, uint len
, ea_list
**to
)
484 if (!s
->proto
->is_internal
)
485 DISCARD(BAD_EBGP
, "CLUSTER_LIST");
487 if (!len
|| (len
% 4))
488 WITHDRAW(BAD_LENGTH
, "CLUSTER_LIST", len
);
490 struct adata
*ad
= lp_alloc_adata(s
->pool
, len
);
491 get_u32s(data
, (u32
*) ad
->data
, len
/ 4);
492 bgp_set_attr_ptr(to
, s
->pool
, BA_CLUSTER_LIST
, flags
, ad
);
496 bgp_format_cluster_list(eattr
*a
, byte
*buf
, uint size
)
498 /* Truncates cluster lists larger than buflen, probably not a problem */
499 int_set_format(a
->u
.ptr
, 0, -1, buf
, size
);
506 return (get_u16(buf
) << 16) | buf
[2];
510 bgp_decode_mp_reach_nlri(struct bgp_parse_state
*s
, uint code UNUSED
, uint flags UNUSED
, byte
*data
, uint len
, ea_list
**to UNUSED
)
513 * 2 B MP_REACH_NLRI data - Address Family Identifier
514 * 1 B MP_REACH_NLRI data - Subsequent Address Family Identifier
515 * 1 B MP_REACH_NLRI data - Length of Next Hop Network Address
516 * var MP_REACH_NLRI data - Network Address of Next Hop
517 * 1 B MP_REACH_NLRI data - Reserved (zero)
518 * var MP_REACH_NLRI data - Network Layer Reachability Information
521 if ((len
< 5) || (len
< (5 + (uint
) data
[3])))
522 bgp_parse_error(s
, 9);
524 s
->mp_reach_af
= get_af3(data
);
525 s
->mp_next_hop_len
= data
[3];
526 s
->mp_next_hop_data
= data
+ 4;
527 s
->mp_reach_len
= len
- 5 - s
->mp_next_hop_len
;
528 s
->mp_reach_nlri
= data
+ 5 + s
->mp_next_hop_len
;
533 bgp_decode_mp_unreach_nlri(struct bgp_parse_state
*s
, uint code UNUSED
, uint flags UNUSED
, byte
*data
, uint len
, ea_list
**to UNUSED
)
536 * 2 B MP_UNREACH_NLRI data - Address Family Identifier
537 * 1 B MP_UNREACH_NLRI data - Subsequent Address Family Identifier
538 * var MP_UNREACH_NLRI data - Network Layer Reachability Information
542 bgp_parse_error(s
, 9);
544 s
->mp_unreach_af
= get_af3(data
);
545 s
->mp_unreach_len
= len
- 3;
546 s
->mp_unreach_nlri
= data
+ 3;
551 bgp_export_ext_community(struct bgp_export_state
*s
, eattr
*a
)
553 a
->u
.ptr
= ec_set_del_nontrans(s
->pool
, a
->u
.ptr
);
555 if (a
->u
.ptr
->length
== 0)
558 ec_set_sort_x(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_export_mpls_label_stack(struct bgp_export_state
*s
, eattr
*a
)
635 net_addr
*n
= s
->route
->net
->n
.addr
;
636 u32
*labels
= (u32
*) a
->u
.ptr
->data
;
637 uint lnum
= a
->u
.ptr
->length
/ 4;
639 /* Perhaps we should just ignore it? */
641 WITHDRAW("Unexpected MPLS stack");
643 /* Empty MPLS stack is not allowed */
645 WITHDRAW("Malformed MPLS stack - empty");
647 /* This is ugly, but we must ensure that labels fit into NLRI field */
648 if ((24*lnum
+ (net_is_vpn(n
) ? 64 : 0) + net_pxlen(n
)) > 255)
649 WITHDRAW("Malformed MPLS stack - too many labels (%u)", lnum
);
651 for (uint i
= 0; i
< lnum
; i
++)
653 if (labels
[i
] > 0xfffff)
654 WITHDRAW("Malformed MPLS stack - invalid label (%u)", labels
[i
]);
656 /* TODO: Check for special-purpose label values? */
661 bgp_encode_mpls_label_stack(struct bgp_write_state
*s
, eattr
*a
, byte
*buf UNUSED
, uint size UNUSED
)
664 * MPLS labels are encoded as a part of the NLRI in MP_REACH_NLRI attribute,
665 * so we store MPLS_LABEL_STACK and encode it later by AFI-specific hooks.
668 s
->mpls_labels
= a
->u
.ptr
;
673 bgp_decode_mpls_label_stack(struct bgp_parse_state
*s
, uint code UNUSED
, uint flags UNUSED
, byte
*data UNUSED
, uint len UNUSED
, ea_list
**to UNUSED
)
675 DISCARD("Discarding received attribute #0");
679 bgp_format_mpls_label_stack(eattr
*a
, byte
*buf
, uint size
)
681 u32
*labels
= (u32
*) a
->u
.ptr
->data
;
682 uint lnum
= a
->u
.ptr
->length
/ 4;
685 for (uint i
= 0; i
< lnum
; i
++)
689 bsprintf(pos
, "...");
693 uint l
= bsprintf(pos
, "%d/", labels
[i
]);
694 ADVANCE(pos
, size
, l
);
697 /* Clear last slash or terminate empty string */
698 pos
[lnum
? -1 : 0] = 0;
702 bgp_decode_unknown(struct bgp_parse_state
*s
, uint code
, uint flags
, byte
*data
, uint len
, ea_list
**to
)
704 /* Cannot use bgp_set_attr_data() as it works on known attributes only */
705 ea_set_attr_data(to
, s
->pool
, EA_CODE(EAP_BGP
, code
), flags
, EAF_TYPE_OPAQUE
, data
, len
);
713 static const struct bgp_attr_desc bgp_attr_table
[] = {
716 .type
= EAF_TYPE_INT
,
717 .flags
= BAF_TRANSITIVE
,
718 .export
= bgp_export_origin
,
719 .encode
= bgp_encode_u8
,
720 .decode
= bgp_decode_origin
,
721 .format
= bgp_format_origin
,
725 .type
= EAF_TYPE_AS_PATH
,
726 .flags
= BAF_TRANSITIVE
,
727 .encode
= bgp_encode_as_path
,
728 .decode
= bgp_decode_as_path
,
732 .type
= EAF_TYPE_IP_ADDRESS
,
733 .flags
= BAF_TRANSITIVE
,
734 .encode
= bgp_encode_next_hop
,
735 .decode
= bgp_decode_next_hop
,
736 .format
= bgp_format_next_hop
,
738 [BA_MULTI_EXIT_DISC
] = {
740 .type
= EAF_TYPE_INT
,
741 .flags
= BAF_OPTIONAL
,
742 .encode
= bgp_encode_u32
,
743 .decode
= bgp_decode_med
,
746 .name
= "local_pref",
747 .type
= EAF_TYPE_INT
,
748 .flags
= BAF_TRANSITIVE
,
749 .export
= bgp_export_local_pref
,
750 .encode
= bgp_encode_u32
,
751 .decode
= bgp_decode_local_pref
,
754 .name
= "atomic_aggr",
755 .type
= EAF_TYPE_OPAQUE
,
756 .flags
= BAF_TRANSITIVE
,
757 .encode
= bgp_encode_raw
,
758 .decode
= bgp_decode_atomic_aggr
,
761 .name
= "aggregator",
762 .type
= EAF_TYPE_OPAQUE
,
763 .flags
= BAF_OPTIONAL
| BAF_TRANSITIVE
,
764 .encode
= bgp_encode_aggregator
,
765 .decode
= bgp_decode_aggregator
,
766 .format
= bgp_format_aggregator
,
770 .type
= EAF_TYPE_INT_SET
,
771 .flags
= BAF_OPTIONAL
| BAF_TRANSITIVE
,
772 .export
= bgp_export_community
,
773 .encode
= bgp_encode_u32s
,
774 .decode
= bgp_decode_community
,
776 [BA_ORIGINATOR_ID
] = {
777 .name
= "originator_id",
778 .type
= EAF_TYPE_ROUTER_ID
,
779 .flags
= BAF_OPTIONAL
,
780 .export
= bgp_export_originator_id
,
781 .encode
= bgp_encode_u32
,
782 .decode
= bgp_decode_originator_id
,
784 [BA_CLUSTER_LIST
] = {
785 .name
= "cluster_list",
786 .type
= EAF_TYPE_INT_SET
,
787 .flags
= BAF_OPTIONAL
,
788 .export
= bgp_export_cluster_list
,
789 .encode
= bgp_encode_u32s
,
790 .decode
= bgp_decode_cluster_list
,
791 .format
= bgp_format_cluster_list
,
793 [BA_MP_REACH_NLRI
] = {
794 .name
= "mp_reach_nlri",
795 .type
= EAF_TYPE_OPAQUE
,
796 .flags
= BAF_OPTIONAL
,
797 .decode
= bgp_decode_mp_reach_nlri
,
799 [BA_MP_UNREACH_NLRI
] = {
800 .name
= "mp_unreach_nlri",
801 .type
= EAF_TYPE_OPAQUE
,
802 .flags
= BAF_OPTIONAL
,
803 .decode
= bgp_decode_mp_unreach_nlri
,
805 [BA_EXT_COMMUNITY
] = {
806 .name
= "ext_community",
807 .type
= EAF_TYPE_EC_SET
,
808 .flags
= BAF_OPTIONAL
| BAF_TRANSITIVE
,
809 .export
= bgp_export_ext_community
,
810 .encode
= bgp_encode_u32s
,
811 .decode
= bgp_decode_ext_community
,
815 .type
= EAF_TYPE_AS_PATH
,
816 .flags
= BAF_OPTIONAL
| BAF_TRANSITIVE
,
817 .encode
= bgp_encode_raw
,
818 .decode
= bgp_decode_as4_path
,
820 [BA_AS4_AGGREGATOR
] = {
821 .name
= "as4_aggregator",
822 .type
= EAF_TYPE_OPAQUE
,
823 .flags
= BAF_OPTIONAL
| BAF_TRANSITIVE
,
824 .encode
= bgp_encode_raw
,
825 .decode
= bgp_decode_as4_aggregator
,
826 .format
= bgp_format_aggregator
,
828 [BA_LARGE_COMMUNITY
] = {
829 .name
= "large_community",
830 .type
= EAF_TYPE_LC_SET
,
831 .flags
= BAF_OPTIONAL
| BAF_TRANSITIVE
,
832 .export
= bgp_export_large_community
,
833 .encode
= bgp_encode_u32s
,
834 .decode
= bgp_decode_large_community
,
836 [BA_MPLS_LABEL_STACK
] = {
837 .name
= "mpls_label_stack",
838 .type
= EAF_TYPE_INT_SET
,
839 .export
= bgp_export_mpls_label_stack
,
840 .encode
= bgp_encode_mpls_label_stack
,
841 .decode
= bgp_decode_mpls_label_stack
,
842 .format
= bgp_format_mpls_label_stack
,
847 bgp_attr_known(uint code
)
849 return (code
< ARRAY_SIZE(bgp_attr_table
)) && bgp_attr_table
[code
].name
;
858 bgp_export_attr(struct bgp_export_state
*s
, eattr
*a
, ea_list
*to
)
860 if (EA_PROTO(a
->id
) != EAP_BGP
)
863 uint code
= EA_ID(a
->id
);
865 if (bgp_attr_known(code
))
867 const struct bgp_attr_desc
*desc
= &bgp_attr_table
[code
];
869 /* The flags might have been zero if the attr was added by filters */
870 a
->flags
= (a
->flags
& BAF_PARTIAL
) | desc
->flags
;
872 /* Set partial bit if new opt-trans attribute is attached to non-local route */
873 if ((s
->src
!= NULL
) && (a
->type
& EAF_ORIGINATED
) &&
874 (a
->flags
& BAF_OPTIONAL
) && (a
->flags
& BAF_TRANSITIVE
))
875 a
->flags
|= BAF_PARTIAL
;
877 /* Call specific hook */
878 CALL(desc
->export
, s
, a
);
880 /* Attribute might become undefined in hook */
881 if ((a
->type
& EAF_TYPE_MASK
) == EAF_TYPE_UNDEF
)
886 /* Don't re-export unknown non-transitive attributes */
887 if (!(a
->flags
& BAF_TRANSITIVE
))
890 a
->flags
|= BAF_PARTIAL
;
893 /* Append updated attribute */
894 to
->attrs
[to
->count
++] = *a
;
898 * bgp_export_attrs - export BGP attributes
899 * @s: BGP export state
900 * @attrs: a list of extended attributes
902 * The bgp_export_attrs() function takes a list of attributes and merges it to
903 * one newly allocated and sorted segment. Attributes are validated and
904 * normalized by type-specific export hooks and attribute flags are updated.
905 * Some attributes may be eliminated (e.g. unknown non-tranitive attributes, or
906 * empty community sets).
908 * Result: one sorted attribute list segment, or NULL if attributes are unsuitable.
910 static inline ea_list
*
911 bgp_export_attrs(struct bgp_export_state
*s
, ea_list
*attrs
)
913 /* Merge the attribute list */
914 ea_list
*new = lp_alloc(s
->pool
, ea_scan(attrs
));
915 ea_merge(attrs
, new);
922 /* Export each attribute */
923 for (i
= 0; i
< count
; i
++)
924 bgp_export_attr(s
, &new->attrs
[i
], new);
938 bgp_encode_attr(struct bgp_write_state
*s
, eattr
*a
, byte
*buf
, uint size
)
940 ASSERT(EA_PROTO(a
->id
) == EAP_BGP
);
942 uint code
= EA_ID(a
->id
);
944 if (bgp_attr_known(code
))
945 return bgp_attr_table
[code
].encode(s
, a
, buf
, size
);
947 return bgp_encode_raw(s
, a
, buf
, size
);
951 * bgp_encode_attrs - encode BGP attributes
952 * @s: BGP write state
953 * @attrs: a list of extended attributes
957 * The bgp_encode_attrs() function takes a list of extended attributes
958 * and converts it to its BGP representation (a part of an Update message).
960 * Result: Length of the attribute block generated or -1 if not enough space.
963 bgp_encode_attrs(struct bgp_write_state
*s
, ea_list
*attrs
, byte
*buf
, byte
*end
)
968 for (i
= 0; i
< attrs
->count
; i
++)
970 len
= bgp_encode_attr(s
, &attrs
->attrs
[i
], pos
, end
- pos
);
986 static void bgp_process_as4_attrs(ea_list
**attrs
, struct linpool
*pool
);
989 bgp_as_path_loopy(struct bgp_proto
*p
, ea_list
*attrs
, u32 asn
)
991 eattr
*e
= bgp_find_attr(attrs
, BA_AS_PATH
);
992 int num
= p
->cf
->allow_local_as
+ 1;
993 return (e
&& (num
> 0) && as_path_contains(e
->u
.ptr
, asn
, num
));
997 bgp_originator_id_loopy(struct bgp_proto
*p
, ea_list
*attrs
)
999 eattr
*e
= bgp_find_attr(attrs
, BA_ORIGINATOR_ID
);
1000 return (e
&& (e
->u
.data
== p
->local_id
));
1004 bgp_cluster_list_loopy(struct bgp_proto
*p
, ea_list
*attrs
)
1006 eattr
*e
= bgp_find_attr(attrs
, BA_CLUSTER_LIST
);
1007 return (e
&& int_set_contains(e
->u
.ptr
, p
->rr_cluster_id
));
1011 bgp_decode_attr(struct bgp_parse_state
*s
, uint code
, uint flags
, byte
*data
, uint len
, ea_list
**to
)
1013 /* Handle duplicate attributes; RFC 7606 3 (g) */
1014 if (BIT32_TEST(s
->attrs_seen
, code
))
1016 if ((code
== BA_MP_REACH_NLRI
) || (code
== BA_MP_UNREACH_NLRI
))
1017 bgp_parse_error(s
, 1);
1019 DISCARD("Discarding duplicate attribute (code %u)", code
);
1021 BIT32_SET(s
->attrs_seen
, code
);
1023 if (bgp_attr_known(code
))
1025 const struct bgp_attr_desc
*desc
= &bgp_attr_table
[code
];
1027 /* Handle conflicting flags; RFC 7606 3 (c) */
1028 if ((flags
^ desc
->flags
) & (BAF_OPTIONAL
| BAF_TRANSITIVE
))
1029 WITHDRAW("Malformed %s attribute - conflicting flags (%02x)", desc
->name
, flags
);
1031 desc
->decode(s
, code
, flags
, data
, len
, to
);
1033 else /* Unknown attribute */
1035 if (!(flags
& BAF_OPTIONAL
))
1036 WITHDRAW("Unknown attribute (code %u) - conflicting flags (%02x)", code
, flags
);
1038 bgp_decode_unknown(s
, code
, flags
, data
, len
, to
);
1043 * bgp_decode_attrs - check and decode BGP attributes
1044 * @s: BGP parse state
1045 * @data: start of attribute block
1046 * @len: length of attribute block
1048 * This function takes a BGP attribute block (a part of an Update message), checks
1049 * its consistency and converts it to a list of BIRD route attributes represented
1050 * by an (uncached) &rta.
1053 bgp_decode_attrs(struct bgp_parse_state
*s
, byte
*data
, uint len
)
1055 struct bgp_proto
*p
= s
->proto
;
1056 ea_list
*attrs
= NULL
;
1057 uint code
, flags
, alen
;
1060 /* Parse the attributes */
1065 /* Read attribute type */
1070 ADVANCE(pos
, len
, 2);
1072 /* Read attribute length */
1073 if (flags
& BAF_EXT_LEN
)
1077 alen
= get_u16(pos
);
1078 ADVANCE(pos
, len
, 2);
1085 ADVANCE(pos
, len
, 1);
1091 DBG("Attr %02x %02x %u\n", code
, flags
, alen
);
1093 bgp_decode_attr(s
, code
, flags
, pos
, alen
, &attrs
);
1094 ADVANCE(pos
, len
, alen
);
1097 if (s
->err_withdraw
)
1100 /* If there is no reachability NLRI, we are finished */
1101 if (!s
->ip_reach_len
&& !s
->mp_reach_len
)
1105 /* Handle missing mandatory attributes; RFC 7606 3 (d) */
1106 if (!BIT32_TEST(s
->attrs_seen
, BA_ORIGIN
))
1107 { REPORT(NO_MANDATORY
, "ORIGIN"); goto withdraw
; }
1109 if (!BIT32_TEST(s
->attrs_seen
, BA_AS_PATH
))
1110 { REPORT(NO_MANDATORY
, "AS_PATH"); goto withdraw
; }
1112 /* When receiving attributes from non-AS4-aware BGP speaker, we have to
1113 reconstruct AS_PATH and AGGREGATOR attributes; RFC 6793 4.2.3 */
1114 if (!p
->as4_session
)
1115 bgp_process_as4_attrs(&attrs
, s
->pool
);
1117 /* Reject routes with our ASN in AS_PATH attribute */
1118 if (bgp_as_path_loopy(p
, attrs
, p
->local_as
))
1121 /* Reject routes with our Confederation ID in AS_PATH attribute; RFC 5065 4.0 */
1122 if ((p
->public_as
!= p
->local_as
) && bgp_as_path_loopy(p
, attrs
, p
->public_as
))
1125 /* Reject routes with our Router ID in ORIGINATOR_ID attribute; RFC 4456 8 */
1126 if (p
->is_internal
&& bgp_originator_id_loopy(p
, attrs
))
1129 /* Reject routes with our Cluster ID in CLUSTER_LIST attribute; RFC 4456 8 */
1130 if (p
->rr_client
&& bgp_cluster_list_loopy(p
, attrs
))
1133 /* If there is no local preference, define one */
1134 if (!BIT32_TEST(s
->attrs_seen
, BA_LOCAL_PREF
))
1135 bgp_set_attr_u32(&attrs
, s
->pool
, BA_LOCAL_PREF
, 0, p
->cf
->default_local_pref
);
1141 /* RFC 7606 4 - handle attribute framing errors */
1142 REPORT("Malformed attribute list - framing error (%u/%u) at %d",
1143 alen
, len
, (int) (pos
- s
->attrs
));
1146 /* RFC 7606 5.2 - handle missing NLRI during errors */
1147 if (!s
->ip_reach_len
&& !s
->mp_reach_len
)
1148 bgp_parse_error(s
, 1);
1150 s
->err_withdraw
= 1;
1156 * Route bucket hash table
1159 #define RBH_KEY(b) b->eattrs, b->hash
1160 #define RBH_NEXT(b) b->next
1161 #define RBH_EQ(a1,h1,a2,h2) h1 == h2 && ea_same(a1, a2)
1162 #define RBH_FN(a,h) h
1164 #define RBH_REHASH bgp_rbh_rehash
1165 #define RBH_PARAMS /8, *2, 2, 2, 8, 20
1168 HASH_DEFINE_REHASH_FN(RBH
, struct bgp_bucket
)
1171 bgp_init_bucket_table(struct bgp_channel
*c
)
1173 HASH_INIT(c
->bucket_hash
, c
->pool
, 8);
1175 init_list(&c
->bucket_queue
);
1176 c
->withdraw_bucket
= NULL
;
1180 bgp_free_bucket_table(struct bgp_channel
*c
)
1182 HASH_FREE(c
->bucket_hash
);
1184 struct bgp_bucket
*b
;
1185 WALK_LIST_FIRST(b
, c
->bucket_queue
)
1187 rem_node(&b
->send_node
);
1191 mb_free(c
->withdraw_bucket
);
1192 c
->withdraw_bucket
= NULL
;
1195 static struct bgp_bucket
*
1196 bgp_get_bucket(struct bgp_channel
*c
, ea_list
*new)
1198 /* Hash and lookup */
1199 u32 hash
= ea_hash(new);
1200 struct bgp_bucket
*b
= HASH_FIND(c
->bucket_hash
, RBH
, new, hash
);
1205 uint ea_size
= sizeof(ea_list
) + new->count
* sizeof(eattr
);
1206 uint ea_size_aligned
= BIRD_ALIGN(ea_size
, CPU_STRUCT_ALIGN
);
1207 uint size
= sizeof(struct bgp_bucket
) + ea_size_aligned
;
1211 /* Gather total size of non-inline attributes */
1212 for (i
= 0; i
< new->count
; i
++)
1214 eattr
*a
= &new->attrs
[i
];
1216 if (!(a
->type
& EAF_EMBEDDED
))
1217 size
+= BIRD_ALIGN(sizeof(struct adata
) + a
->u
.ptr
->length
, CPU_STRUCT_ALIGN
);
1220 /* Create the bucket */
1221 b
= mb_alloc(c
->pool
, size
);
1222 init_list(&b
->prefixes
);
1225 /* Copy list of extended attributes */
1226 memcpy(b
->eattrs
, new, ea_size
);
1227 dest
= ((byte
*) b
->eattrs
) + ea_size_aligned
;
1229 /* Copy values of non-inline attributes */
1230 for (i
= 0; i
< new->count
; i
++)
1232 eattr
*a
= &b
->eattrs
->attrs
[i
];
1234 if (!(a
->type
& EAF_EMBEDDED
))
1236 struct adata
*oa
= a
->u
.ptr
;
1237 struct adata
*na
= (struct adata
*) dest
;
1238 memcpy(na
, oa
, sizeof(struct adata
) + oa
->length
);
1240 dest
+= BIRD_ALIGN(sizeof(struct adata
) + na
->length
, CPU_STRUCT_ALIGN
);
1244 /* Insert the bucket to send queue and bucket hash */
1245 add_tail(&c
->bucket_queue
, &b
->send_node
);
1246 HASH_INSERT2(c
->bucket_hash
, RBH
, c
->pool
, b
);
1251 static struct bgp_bucket
*
1252 bgp_get_withdraw_bucket(struct bgp_channel
*c
)
1254 if (!c
->withdraw_bucket
)
1256 c
->withdraw_bucket
= mb_allocz(c
->pool
, sizeof(struct bgp_bucket
));
1257 init_list(&c
->withdraw_bucket
->prefixes
);
1260 return c
->withdraw_bucket
;
1264 bgp_free_bucket(struct bgp_channel
*c
, struct bgp_bucket
*b
)
1266 rem_node(&b
->send_node
);
1267 HASH_REMOVE2(c
->bucket_hash
, RBH
, c
->pool
, b
);
1272 bgp_defer_bucket(struct bgp_channel
*c
, struct bgp_bucket
*b
)
1274 rem_node(&b
->send_node
);
1275 add_tail(&c
->bucket_queue
, &b
->send_node
);
1279 bgp_withdraw_bucket(struct bgp_channel
*c
, struct bgp_bucket
*b
)
1281 struct bgp_proto
*p
= (void *) c
->c
.proto
;
1282 struct bgp_bucket
*wb
= bgp_get_withdraw_bucket(c
);
1284 log(L_ERR
"%s: Attribute list too long", p
->p
.name
);
1285 while (!EMPTY_LIST(b
->prefixes
))
1287 struct bgp_prefix
*px
= HEAD(b
->prefixes
);
1289 log(L_ERR
"%s: - withdrawing %N", p
->p
.name
, &px
->net
);
1290 rem_node(&px
->buck_node
);
1291 add_tail(&wb
->prefixes
, &px
->buck_node
);
1300 #define PXH_KEY(px) px->net, px->path_id, px->hash
1301 #define PXH_NEXT(px) px->next
1302 #define PXH_EQ(n1,i1,h1,n2,i2,h2) h1 == h2 && i1 == i2 && net_equal(n1, n2)
1303 #define PXH_FN(n,i,h) h
1305 #define PXH_REHASH bgp_pxh_rehash
1306 #define PXH_PARAMS /8, *2, 2, 2, 8, 20
1309 HASH_DEFINE_REHASH_FN(PXH
, struct bgp_prefix
)
1312 bgp_init_prefix_table(struct bgp_channel
*c
)
1314 HASH_INIT(c
->prefix_hash
, c
->pool
, 8);
1316 uint alen
= net_addr_length
[c
->c
.net_type
];
1317 c
->prefix_slab
= alen
? sl_new(c
->pool
, sizeof(struct bgp_prefix
) + alen
) : NULL
;
1321 bgp_free_prefix_table(struct bgp_channel
*c
)
1323 HASH_FREE(c
->prefix_hash
);
1325 rfree(c
->prefix_slab
);
1326 c
->prefix_slab
= NULL
;
1329 static struct bgp_prefix
*
1330 bgp_get_prefix(struct bgp_channel
*c
, net_addr
*net
, u32 path_id
)
1332 u32 hash
= net_hash(net
) ^ u32_hash(path_id
);
1333 struct bgp_prefix
*px
= HASH_FIND(c
->prefix_hash
, PXH
, net
, path_id
, hash
);
1337 rem_node(&px
->buck_node
);
1342 px
= sl_alloc(c
->prefix_slab
);
1344 px
= mb_alloc(c
->pool
, sizeof(struct bgp_prefix
) + net
->length
);
1346 px
->buck_node
.next
= NULL
;
1347 px
->buck_node
.prev
= NULL
;
1349 px
->path_id
= path_id
;
1350 net_copy(px
->net
, net
);
1352 HASH_INSERT2(c
->prefix_hash
, PXH
, c
->pool
, px
);
1358 bgp_free_prefix(struct bgp_channel
*c
, struct bgp_prefix
*px
)
1360 rem_node(&px
->buck_node
);
1361 HASH_REMOVE2(c
->prefix_hash
, PXH
, c
->pool
, px
);
1364 sl_free(c
->prefix_slab
, px
);
1375 bgp_import_control(struct proto
*P
, rte
**new, ea_list
**attrs UNUSED
, struct linpool
*pool UNUSED
)
1378 struct proto
*SRC
= e
->attrs
->src
->proto
;
1379 struct bgp_proto
*p
= (struct bgp_proto
*) P
;
1380 struct bgp_proto
*src
= (SRC
->proto
== &proto_bgp
) ? (struct bgp_proto
*) SRC
: NULL
;
1382 /* Reject our routes */
1386 /* Accept non-BGP routes */
1390 // XXXX: Check next hop AF
1392 /* IBGP route reflection, RFC 4456 */
1393 if (p
->is_internal
&& src
->is_internal
&& (p
->local_as
== src
->local_as
))
1395 /* Rejected unless configured as route reflector */
1396 if (!p
->rr_client
&& !src
->rr_client
)
1399 /* Generally, this should be handled when path is received, but we check it
1400 also here as rr_cluster_id may be undefined or different in src. */
1401 if (p
->rr_cluster_id
&& bgp_cluster_list_loopy(p
, e
->attrs
->eattrs
))
1405 /* Handle well-known communities, RFC 1997 */
1407 if (p
->cf
->interpret_communities
&&
1408 (c
= ea_find(e
->attrs
->eattrs
, EA_CODE(EAP_BGP
, BA_COMMUNITY
))))
1410 struct adata
*d
= c
->u
.ptr
;
1412 /* Do not export anywhere */
1413 if (int_set_contains(d
, BGP_COMM_NO_ADVERTISE
))
1416 /* Do not export outside of AS (or member-AS) */
1417 if (!p
->is_internal
&& int_set_contains(d
, BGP_COMM_NO_EXPORT_SUBCONFED
))
1420 /* Do not export outside of AS (or confederation) */
1421 if (!p
->is_interior
&& int_set_contains(d
, BGP_COMM_NO_EXPORT
))
1429 static adata null_adata
; /* adata of length 0 */
1432 bgp_update_attrs(struct bgp_proto
*p
, struct bgp_channel
*c
, rte
*e
, ea_list
*attrs0
, struct linpool
*pool
)
1434 struct proto
*SRC
= e
->attrs
->src
->proto
;
1435 struct bgp_proto
*src
= (SRC
->proto
== &proto_bgp
) ? (void *) SRC
: NULL
;
1436 struct bgp_export_state s
= { .proto
= p
, .channel
= c
, .pool
= pool
, .src
= src
, .route
= e
, .mpls
= c
->desc
->mpls
};
1437 ea_list
*attrs
= attrs0
;
1441 /* ORIGIN attribute - mandatory, attach if missing */
1442 if (! bgp_find_attr(attrs0
, BA_ORIGIN
))
1443 bgp_set_attr_u32(&attrs
, pool
, BA_ORIGIN
, 0, src
? ORIGIN_INCOMPLETE
: ORIGIN_IGP
);
1445 /* AS_PATH attribute - mandatory */
1446 a
= bgp_find_attr(attrs0
, BA_AS_PATH
);
1447 ad
= a
? a
->u
.ptr
: &null_adata
;
1449 /* AS_PATH attribute - strip AS_CONFED* segments outside confederation */
1450 if ((!p
->cf
->confederation
|| !p
->is_interior
) && as_path_contains_confed(ad
))
1451 ad
= as_path_strip_confed(pool
, ad
);
1453 /* AS_PATH attribute - keep or prepend ASN */
1454 if (p
->is_internal
||
1455 (p
->rs_client
&& src
&& src
->rs_client
))
1457 /* IBGP or route server -> just ensure there is one */
1459 bgp_set_attr_ptr(&attrs
, pool
, BA_AS_PATH
, 0, &null_adata
);
1461 else if (p
->is_interior
)
1463 /* Confederation -> prepend ASN as AS_CONFED_SEQUENCE */
1464 ad
= as_path_prepend2(pool
, ad
, AS_PATH_CONFED_SEQUENCE
, p
->public_as
);
1465 bgp_set_attr_ptr(&attrs
, pool
, BA_AS_PATH
, 0, ad
);
1467 else /* Regular EBGP (no RS, no confederation) */
1469 /* Regular EBGP -> prepend ASN as regular sequence */
1470 ad
= as_path_prepend2(pool
, ad
, AS_PATH_SEQUENCE
, p
->public_as
);
1471 bgp_set_attr_ptr(&attrs
, pool
, BA_AS_PATH
, 0, ad
);
1473 /* MULTI_EXIT_DESC attribute - accept only if set in export filter */
1474 a
= bgp_find_attr(attrs0
, BA_MULTI_EXIT_DISC
);
1475 if (a
&& !(a
->type
& EAF_FRESH
))
1476 bgp_unset_attr(&attrs
, pool
, BA_MULTI_EXIT_DISC
);
1479 /* NEXT_HOP attribute - delegated to AF-specific hook */
1480 a
= bgp_find_attr(attrs0
, BA_NEXT_HOP
);
1481 bgp_update_next_hop(&s
, a
, &attrs
);
1483 /* LOCAL_PREF attribute - required for IBGP, attach if missing */
1484 if (p
->is_interior
&& ! bgp_find_attr(attrs0
, BA_LOCAL_PREF
))
1485 bgp_set_attr_u32(&attrs
, pool
, BA_LOCAL_PREF
, 0, p
->cf
->default_local_pref
);
1487 /* IBGP route reflection, RFC 4456 */
1488 if (src
&& src
->is_internal
&& p
->is_internal
&& (src
->local_as
== p
->local_as
))
1490 /* ORIGINATOR_ID attribute - attach if not already set */
1491 if (! bgp_find_attr(attrs0
, BA_ORIGINATOR_ID
))
1492 bgp_set_attr_u32(&attrs
, pool
, BA_ORIGINATOR_ID
, 0, src
->remote_id
);
1494 /* CLUSTER_LIST attribute - prepend cluster ID */
1495 a
= bgp_find_attr(attrs0
, BA_CLUSTER_LIST
);
1496 ad
= a
? a
->u
.ptr
: NULL
;
1498 /* Prepend src cluster ID */
1499 if (src
->rr_cluster_id
)
1500 ad
= int_set_prepend(pool
, ad
, src
->rr_cluster_id
);
1502 /* Prepend dst cluster ID if src and dst clusters are different */
1503 if (p
->rr_cluster_id
&& (src
->rr_cluster_id
!= p
->rr_cluster_id
))
1504 ad
= int_set_prepend(pool
, ad
, p
->rr_cluster_id
);
1506 /* Should be at least one prepended cluster ID */
1507 bgp_set_attr_ptr(&attrs
, pool
, BA_CLUSTER_LIST
, 0, ad
);
1510 /* AS4_* transition attributes, RFC 6793 4.2.2 */
1511 if (! p
->as4_session
)
1513 a
= bgp_find_attr(attrs
, BA_AS_PATH
);
1514 if (a
&& as_path_contains_as4(a
->u
.ptr
))
1516 bgp_set_attr_ptr(&attrs
, pool
, BA_AS_PATH
, 0, as_path_to_old(pool
, a
->u
.ptr
));
1517 bgp_set_attr_ptr(&attrs
, pool
, BA_AS4_PATH
, 0, as_path_strip_confed(pool
, a
->u
.ptr
));
1520 a
= bgp_find_attr(attrs
, BA_AGGREGATOR
);
1521 if (a
&& aggregator_contains_as4(a
->u
.ptr
))
1523 bgp_set_attr_ptr(&attrs
, pool
, BA_AGGREGATOR
, 0, aggregator_to_old(pool
, a
->u
.ptr
));
1524 bgp_set_attr_ptr(&attrs
, pool
, BA_AS4_AGGREGATOR
, 0, a
->u
.ptr
);
1529 * Presence of mandatory attributes ORIGIN and AS_PATH is ensured by above
1530 * conditions. Presence and validity of quasi-mandatory NEXT_HOP attribute
1531 * should be checked in AF-specific hooks.
1534 /* Apply per-attribute export hooks for validatation and normalization */
1535 return bgp_export_attrs(&s
, attrs
);
1539 bgp_rt_notify(struct proto
*P
, struct channel
*C
, net
*n
, rte
*new, rte
*old
, ea_list
*attrs
)
1541 struct bgp_proto
*p
= (void *) P
;
1542 struct bgp_channel
*c
= (void *) C
;
1543 struct bgp_bucket
*buck
;
1544 struct bgp_prefix
*px
;
1549 attrs
= bgp_update_attrs(p
, c
, new, attrs
, bgp_linpool2
);
1551 /* If attributes are invalid, we fail back to withdraw */
1552 buck
= attrs
? bgp_get_bucket(c
, attrs
) : bgp_get_withdraw_bucket(c
);
1553 path
= new->attrs
->src
->global_id
;
1555 lp_flush(bgp_linpool2
);
1559 buck
= bgp_get_withdraw_bucket(c
);
1560 path
= old
->attrs
->src
->global_id
;
1563 px
= bgp_get_prefix(c
, n
->n
.addr
, c
->add_path_tx
? path
: 0);
1564 add_tail(&buck
->prefixes
, &px
->buck_node
);
1566 bgp_schedule_packet(p
->conn
, c
, PKT_UPDATE
);
1571 bgp_get_neighbor(rte
*r
)
1573 eattr
*e
= ea_find(r
->attrs
->eattrs
, EA_CODE(EAP_BGP
, BA_AS_PATH
));
1576 if (e
&& as_path_get_first_regular(e
->u
.ptr
, &as
))
1579 /* If AS_PATH is not defined, we treat rte as locally originated */
1580 struct bgp_proto
*p
= (void *) r
->attrs
->src
->proto
;
1581 return p
->cf
->confederation
?: p
->local_as
;
1585 rte_resolvable(rte
*rt
)
1587 return rt
->attrs
->dest
== RTD_UNICAST
;
1591 bgp_rte_better(rte
*new, rte
*old
)
1593 struct bgp_proto
*new_bgp
= (struct bgp_proto
*) new->attrs
->src
->proto
;
1594 struct bgp_proto
*old_bgp
= (struct bgp_proto
*) old
->attrs
->src
->proto
;
1598 /* Skip suppressed routes (see bgp_rte_recalculate()) */
1599 n
= new->u
.bgp
.suppressed
;
1600 o
= old
->u
.bgp
.suppressed
;
1606 /* RFC 4271 9.1.2.1. Route resolvability test */
1607 n
= rte_resolvable(new);
1608 o
= rte_resolvable(old
);
1614 /* Start with local preferences */
1615 x
= ea_find(new->attrs
->eattrs
, EA_CODE(EAP_BGP
, BA_LOCAL_PREF
));
1616 y
= ea_find(old
->attrs
->eattrs
, EA_CODE(EAP_BGP
, BA_LOCAL_PREF
));
1617 n
= x
? x
->u
.data
: new_bgp
->cf
->default_local_pref
;
1618 o
= y
? y
->u
.data
: old_bgp
->cf
->default_local_pref
;
1624 /* RFC 4271 9.1.2.2. a) Use AS path lengths */
1625 if (new_bgp
->cf
->compare_path_lengths
|| old_bgp
->cf
->compare_path_lengths
)
1627 x
= ea_find(new->attrs
->eattrs
, EA_CODE(EAP_BGP
, BA_AS_PATH
));
1628 y
= ea_find(old
->attrs
->eattrs
, EA_CODE(EAP_BGP
, BA_AS_PATH
));
1629 n
= x
? as_path_getlen(x
->u
.ptr
) : AS_PATH_MAXLEN
;
1630 o
= y
? as_path_getlen(y
->u
.ptr
) : AS_PATH_MAXLEN
;
1637 /* RFC 4271 9.1.2.2. b) Use origins */
1638 x
= ea_find(new->attrs
->eattrs
, EA_CODE(EAP_BGP
, BA_ORIGIN
));
1639 y
= ea_find(old
->attrs
->eattrs
, EA_CODE(EAP_BGP
, BA_ORIGIN
));
1640 n
= x
? x
->u
.data
: ORIGIN_INCOMPLETE
;
1641 o
= y
? y
->u
.data
: ORIGIN_INCOMPLETE
;
1647 /* RFC 4271 9.1.2.2. c) Compare MED's */
1648 /* Proper RFC 4271 path selection cannot be interpreted as finding
1649 * the best path in some ordering. It is implemented partially in
1650 * bgp_rte_recalculate() when deterministic_med option is
1651 * active. Without that option, the behavior is just an
1652 * approximation, which in specific situations may lead to
1653 * persistent routing loops, because it is nondeterministic - it
1654 * depends on the order in which routes appeared. But it is also the
1655 * same behavior as used by default in Cisco routers, so it is
1656 * probably not a big issue.
1658 if (new_bgp
->cf
->med_metric
|| old_bgp
->cf
->med_metric
||
1659 (bgp_get_neighbor(new) == bgp_get_neighbor(old
)))
1661 x
= ea_find(new->attrs
->eattrs
, EA_CODE(EAP_BGP
, BA_MULTI_EXIT_DISC
));
1662 y
= ea_find(old
->attrs
->eattrs
, EA_CODE(EAP_BGP
, BA_MULTI_EXIT_DISC
));
1663 n
= x
? x
->u
.data
: new_bgp
->cf
->default_med
;
1664 o
= y
? y
->u
.data
: old_bgp
->cf
->default_med
;
1671 /* RFC 4271 9.1.2.2. d) Prefer external peers */
1672 if (new_bgp
->is_interior
> old_bgp
->is_interior
)
1674 if (new_bgp
->is_interior
< old_bgp
->is_interior
)
1677 /* RFC 4271 9.1.2.2. e) Compare IGP metrics */
1678 n
= new_bgp
->cf
->igp_metric
? new->attrs
->igp_metric
: 0;
1679 o
= old_bgp
->cf
->igp_metric
? old
->attrs
->igp_metric
: 0;
1685 /* RFC 4271 9.1.2.2. f) Compare BGP identifiers */
1686 /* RFC 4456 9. a) Use ORIGINATOR_ID instead of local neighbor ID */
1687 x
= ea_find(new->attrs
->eattrs
, EA_CODE(EAP_BGP
, BA_ORIGINATOR_ID
));
1688 y
= ea_find(old
->attrs
->eattrs
, EA_CODE(EAP_BGP
, BA_ORIGINATOR_ID
));
1689 n
= x
? x
->u
.data
: new_bgp
->remote_id
;
1690 o
= y
? y
->u
.data
: old_bgp
->remote_id
;
1692 /* RFC 5004 - prefer older routes */
1693 /* (if both are external and from different peer) */
1694 if ((new_bgp
->cf
->prefer_older
|| old_bgp
->cf
->prefer_older
) &&
1695 !new_bgp
->is_internal
&& n
!= o
)
1698 /* rest of RFC 4271 9.1.2.2. f) */
1704 /* RFC 4456 9. b) Compare cluster list lengths */
1705 x
= ea_find(new->attrs
->eattrs
, EA_CODE(EAP_BGP
, BA_CLUSTER_LIST
));
1706 y
= ea_find(old
->attrs
->eattrs
, EA_CODE(EAP_BGP
, BA_CLUSTER_LIST
));
1707 n
= x
? int_set_get_size(x
->u
.ptr
) : 0;
1708 o
= y
? int_set_get_size(y
->u
.ptr
) : 0;
1714 /* RFC 4271 9.1.2.2. g) Compare peer IP adresses */
1715 return (ipa_compare(new_bgp
->cf
->remote_ip
, old_bgp
->cf
->remote_ip
) < 0);
1720 bgp_rte_mergable(rte
*pri
, rte
*sec
)
1722 struct bgp_proto
*pri_bgp
= (struct bgp_proto
*) pri
->attrs
->src
->proto
;
1723 struct bgp_proto
*sec_bgp
= (struct bgp_proto
*) sec
->attrs
->src
->proto
;
1727 /* Skip suppressed routes (see bgp_rte_recalculate()) */
1728 if (pri
->u
.bgp
.suppressed
!= sec
->u
.bgp
.suppressed
)
1731 /* RFC 4271 9.1.2.1. Route resolvability test */
1732 if (!rte_resolvable(sec
))
1735 /* Start with local preferences */
1736 x
= ea_find(pri
->attrs
->eattrs
, EA_CODE(EAP_BGP
, BA_LOCAL_PREF
));
1737 y
= ea_find(sec
->attrs
->eattrs
, EA_CODE(EAP_BGP
, BA_LOCAL_PREF
));
1738 p
= x
? x
->u
.data
: pri_bgp
->cf
->default_local_pref
;
1739 s
= y
? y
->u
.data
: sec_bgp
->cf
->default_local_pref
;
1743 /* RFC 4271 9.1.2.2. a) Use AS path lengths */
1744 if (pri_bgp
->cf
->compare_path_lengths
|| sec_bgp
->cf
->compare_path_lengths
)
1746 x
= ea_find(pri
->attrs
->eattrs
, EA_CODE(EAP_BGP
, BA_AS_PATH
));
1747 y
= ea_find(sec
->attrs
->eattrs
, EA_CODE(EAP_BGP
, BA_AS_PATH
));
1748 p
= x
? as_path_getlen(x
->u
.ptr
) : AS_PATH_MAXLEN
;
1749 s
= y
? as_path_getlen(y
->u
.ptr
) : AS_PATH_MAXLEN
;
1754 // if (DELTA(p, s) > pri_bgp->cf->relax_multipath)
1758 /* RFC 4271 9.1.2.2. b) Use origins */
1759 x
= ea_find(pri
->attrs
->eattrs
, EA_CODE(EAP_BGP
, BA_ORIGIN
));
1760 y
= ea_find(sec
->attrs
->eattrs
, EA_CODE(EAP_BGP
, BA_ORIGIN
));
1761 p
= x
? x
->u
.data
: ORIGIN_INCOMPLETE
;
1762 s
= y
? y
->u
.data
: ORIGIN_INCOMPLETE
;
1766 /* RFC 4271 9.1.2.2. c) Compare MED's */
1767 if (pri_bgp
->cf
->med_metric
|| sec_bgp
->cf
->med_metric
||
1768 (bgp_get_neighbor(pri
) == bgp_get_neighbor(sec
)))
1770 x
= ea_find(pri
->attrs
->eattrs
, EA_CODE(EAP_BGP
, BA_MULTI_EXIT_DISC
));
1771 y
= ea_find(sec
->attrs
->eattrs
, EA_CODE(EAP_BGP
, BA_MULTI_EXIT_DISC
));
1772 p
= x
? x
->u
.data
: pri_bgp
->cf
->default_med
;
1773 s
= y
? y
->u
.data
: sec_bgp
->cf
->default_med
;
1778 /* RFC 4271 9.1.2.2. d) Prefer external peers */
1779 if (pri_bgp
->is_interior
!= sec_bgp
->is_interior
)
1782 /* RFC 4271 9.1.2.2. e) Compare IGP metrics */
1783 p
= pri_bgp
->cf
->igp_metric
? pri
->attrs
->igp_metric
: 0;
1784 s
= sec_bgp
->cf
->igp_metric
? sec
->attrs
->igp_metric
: 0;
1788 /* Remaining criteria are ignored */
1795 same_group(rte
*r
, u32 lpref
, u32 lasn
)
1797 return (r
->pref
== lpref
) && (bgp_get_neighbor(r
) == lasn
);
1801 use_deterministic_med(rte
*r
)
1803 struct proto
*P
= r
->attrs
->src
->proto
;
1804 return (P
->proto
== &proto_bgp
) && ((struct bgp_proto
*) P
)->cf
->deterministic_med
;
1808 bgp_rte_recalculate(rtable
*table
, net
*net
, rte
*new, rte
*old
, rte
*old_best
)
1811 rte
*key
= new ? new : old
;
1812 u32 lpref
= key
->pref
;
1813 u32 lasn
= bgp_get_neighbor(key
);
1814 int old_is_group_best
= 0;
1817 * Proper RFC 4271 path selection is a bit complicated, it cannot be
1818 * implemented just by rte_better(), because it is not a linear
1819 * ordering. But it can be splitted to two levels, where the lower
1820 * level chooses the best routes in each group of routes from the
1821 * same neighboring AS and higher level chooses the best route (with
1822 * a slightly different ordering) between the best-in-group routes.
1824 * When deterministic_med is disabled, we just ignore this issue and
1825 * choose the best route by bgp_rte_better() alone. If enabled, the
1826 * lower level of the route selection is done here (for the group
1827 * to which the changed route belongs), all routes in group are
1828 * marked as suppressed, just chosen best-in-group is not.
1830 * Global best route selection then implements higher level by
1831 * choosing between non-suppressed routes (as they are always
1832 * preferred over suppressed routes). Routes from BGP protocols
1833 * that do not set deterministic_med are just never suppressed. As
1834 * they do not participate in the lower level selection, it is OK
1835 * that this fn is not called for them.
1837 * The idea is simple, the implementation is more problematic,
1838 * mostly because of optimizations in rte_recalculate() that
1839 * avoids full recalculation in most cases.
1841 * We can assume that at least one of new, old is non-NULL and both
1842 * are from the same protocol with enabled deterministic_med. We
1843 * group routes by both neighbor AS (lasn) and preference (lpref),
1844 * because bgp_rte_better() does not handle preference itself.
1847 /* If new and old are from different groups, we just process that
1848 as two independent events */
1849 if (new && old
&& !same_group(old
, lpref
, lasn
))
1852 i1
= bgp_rte_recalculate(table
, net
, NULL
, old
, old_best
);
1853 i2
= bgp_rte_recalculate(table
, net
, new, NULL
, old_best
);
1858 * We could find the best-in-group and then make some shortcuts like
1859 * in rte_recalculate, but as we would have to walk through all
1860 * net->routes just to find it, it is probably not worth. So we
1861 * just have two simpler fast cases that use just the old route.
1862 * We also set suppressed flag to avoid using it in bgp_rte_better().
1866 new->u
.bgp
.suppressed
= 1;
1870 old_is_group_best
= !old
->u
.bgp
.suppressed
;
1871 old
->u
.bgp
.suppressed
= 1;
1872 int new_is_better
= new && bgp_rte_better(new, old
);
1874 /* The first case - replace not best with worse (or remove not best) */
1875 if (!old_is_group_best
&& !new_is_better
)
1878 /* The second case - replace the best with better */
1879 if (old_is_group_best
&& new_is_better
)
1881 /* new is best-in-group, the see discussion below - this is
1882 a special variant of NBG && OBG. From OBG we can deduce
1883 that same_group(old_best) iff (old == old_best) */
1884 new->u
.bgp
.suppressed
= 0;
1885 return (old
== old_best
);
1889 /* The default case - find a new best-in-group route */
1890 r
= new; /* new may not be in the list */
1891 for (s
=net
->routes
; rte_is_valid(s
); s
=s
->next
)
1892 if (use_deterministic_med(s
) && same_group(s
, lpref
, lasn
))
1894 s
->u
.bgp
.suppressed
= 1;
1895 if (!r
|| bgp_rte_better(s
, r
))
1899 /* Simple case - the last route in group disappears */
1903 /* Found best-in-group */
1904 r
->u
.bgp
.suppressed
= 0;
1907 * There are generally two reasons why we have to force
1908 * recalculation (return 1): First, the new route may be wrongfully
1909 * chosen to be the best in the first case check in
1910 * rte_recalculate(), this may happen only if old_best is from the
1911 * same group. Second, another (different than new route)
1912 * best-in-group is chosen and that may be the proper best (although
1913 * rte_recalculate() without ignore that possibility).
1915 * There are three possible cases according to whether the old route
1916 * was the best in group (OBG, stored in old_is_group_best) and
1917 * whether the new route is the best in group (NBG, tested by r == new).
1918 * These cases work even if old or new is NULL.
1920 * NBG -> new is a possible candidate for the best route, so we just
1921 * check for the first reason using same_group().
1923 * !NBG && OBG -> Second reason applies, return 1
1925 * !NBG && !OBG -> Best in group does not change, old != old_best,
1926 * rte_better(new, old_best) is false and therefore
1927 * the first reason does not apply, return 0
1931 return old_best
&& same_group(old_best
, lpref
, lasn
);
1933 return old_is_group_best
;
1938 * Reconstruct AS_PATH and AGGREGATOR according to RFC 6793 4.2.3
1941 bgp_process_as4_attrs(ea_list
**attrs
, struct linpool
*pool
)
1943 eattr
*p2
= bgp_find_attr(*attrs
, BA_AS_PATH
);
1944 eattr
*p4
= bgp_find_attr(*attrs
, BA_AS4_PATH
);
1945 eattr
*a2
= bgp_find_attr(*attrs
, BA_AGGREGATOR
);
1946 eattr
*a4
= bgp_find_attr(*attrs
, BA_AS4_AGGREGATOR
);
1948 /* First, unset AS4_* attributes */
1949 if (p4
) bgp_unset_attr(attrs
, pool
, BA_AS4_PATH
);
1950 if (a4
) bgp_unset_attr(attrs
, pool
, BA_AS4_AGGREGATOR
);
1952 /* Handle AGGREGATOR attribute */
1955 u32 a2_asn
= get_u32(a2
->u
.ptr
->data
);
1957 /* If routes were aggregated by an old router, then AS4_PATH and
1958 AS4_AGGREGATOR are invalid. In that case we give up. */
1959 if (a2_asn
!= AS_TRANS
)
1962 /* Use AS4_AGGREGATOR instead of AGGREGATOR */
1963 a2
->u
.ptr
= a4
->u
.ptr
;
1966 /* Handle AS_PATH attribute */
1969 /* Both as_path_getlen() and as_path_cut() take AS_CONFED* as zero length */
1970 int p2_len
= as_path_getlen(p2
->u
.ptr
);
1971 int p4_len
= as_path_getlen(p4
->u
.ptr
);
1973 /* AS_PATH is too short, give up */
1974 if (p2_len
< p4_len
)
1977 /* Merge AS_PATH and AS4_PATH */
1978 as_path_cut(p2
->u
.ptr
, p2_len
- p4_len
);
1979 p2
->u
.ptr
= as_path_merge(pool
, p2
->u
.ptr
, p4
->u
.ptr
);
1984 bgp_get_attr(eattr
*a
, byte
*buf
, int buflen
)
1986 uint i
= EA_ID(a
->id
);
1987 const struct bgp_attr_desc
*d
;
1990 if (bgp_attr_known(i
))
1992 d
= &bgp_attr_table
[i
];
1993 len
= bsprintf(buf
, "%s", d
->name
);
1999 d
->format(a
, buf
, buflen
- len
- 2);
2005 bsprintf(buf
, "%02x%s", i
, (a
->flags
& BAF_TRANSITIVE
) ? " [t]" : "");
2010 bgp_get_route_info(rte
*e
, byte
*buf
, ea_list
*attrs
)
2012 eattr
*p
= ea_find(attrs
, EA_CODE(EAP_BGP
, BA_AS_PATH
));
2013 eattr
*o
= ea_find(attrs
, EA_CODE(EAP_BGP
, BA_ORIGIN
));
2016 buf
+= bsprintf(buf
, " (%d", e
->pref
);
2018 if (e
->u
.bgp
.suppressed
)
2019 buf
+= bsprintf(buf
, "-");
2021 if (e
->attrs
->hostentry
)
2023 if (!rte_resolvable(e
))
2024 buf
+= bsprintf(buf
, "/-");
2025 else if (e
->attrs
->igp_metric
>= IGP_METRIC_UNKNOWN
)
2026 buf
+= bsprintf(buf
, "/?");
2028 buf
+= bsprintf(buf
, "/%d", e
->attrs
->igp_metric
);
2030 buf
+= bsprintf(buf
, ") [");
2032 if (p
&& as_path_get_last(p
->u
.ptr
, &origas
))
2033 buf
+= bsprintf(buf
, "AS%u", origas
);
2035 buf
+= bsprintf(buf
, "%c", "ie?"[o
->u
.data
]);