OSPF Opaque / SR, BGP MP / VPN / EVPN (P2)

Status: draft

Notes on the routing protocol extensions of M7: OSPFv2 Opaque LSAs (RFC 5250) with Traffic Engineering (RFC 3630), Router Information (RFC 7770), Extended Prefix / Extended Link (RFC 7684) and Segment Routing (RFC 8665), and the BGP multiprotocol NLRI (RFC 4760) for IPv4 / IPv6 unicast, VPN-IPv4 / VPN-IPv6 (RFC 4364 / RFC 4659) and EVPN (RFC 7432 / RFC 9136). The definitions are in crates/wpg-protocols/src/ospf_opaque.rs and crates/wpg-protocols/src/bgp_mp.rs; the basics of OSPF and BGP are on the routing protocols page.

OSPFv2 Opaque LSAs

ipv4 / ospfv2 / ospfv2_lsu / ospfv2_opaque_lsa / ospfv2_te_lsa
ipv4 / ospfv2 / ospfv2_lsu / ospfv2_opaque_lsa / ospfv2_ri_lsa
ipv4 / ospfv2 / ospfv2_lsu / ospfv2_opaque_lsa / ospfv2_ext_prefix_lsa
ipv4 / ospfv2 / ospfv2_lsu / ospfv2_opaque_lsa / ospfv2_ext_link_lsa

ospfv2_opaque_lsa is the LSA header with the Link State ID split into opaque_type (8 bits, Auto from the body layer) and opaque_id (24 bits). ls_type gives the flooding scope: 9 link-local, 10 area (default), 11 AS. The LSA length and the Fletcher ls_checksum are Auto as for the other LSAs. When parsing, an LSA of type 9 / 10 / 11 is first read as ospfv2_lsa and then re-decoded as ospfv2_opaque_lsa; an unknown opaque type is followed by a raw layer.

TLVs

Each body has one field, tlvs. TLVs and sub-TLVs have a 16-bit type and a 16-bit length counting the value only; the value is padded with zeros to a multiple of 4 octets (not counted in the length). Lengths and padding are Auto.

A SID is 3 octets (a label in the low 20 bits, V and L flags set) or 4 octets (an index, V and L clear). Every sub-TLV that carries one exists in both forms; pick the item by name (label or index field) and the parser tells them apart by the value length:

- protocol: ospfv2_ext_prefix_lsa
  fields:
    tlvs:
      - extended_prefix:
          route_type: intra_area
          prefix_length: 32
          flags: n
          prefix: 192.0.2.1
          sub_tlvs:
            - prefix_sid: { algorithm: 0, index: 1 }          # 4 octets
            # - prefix_sid_label: { flags: v|l, label: 16001 } # 3 octets

TE bandwidths are the raw 32-bit IEEE floating point values in octets per second (0x4b3ebc20 = 12.5e6 = 100 Mbit/s).

BGP multiprotocol NLRI

MP_REACH_NLRI (code 14) and MP_UNREACH_NLRI (code 15) have a layout that depends on the AFI / SAFI after them. Each modelled family is its own path attribute item; the parser picks it from the first three value octets (AFI, SAFI):

Other families use the generic mp_reach_nlri / mp_unreach_nlri with the next hop and NLRI as octets. afi / safi default to the family; next_hop_length is Auto.

• Prefixes: length is in bits and prefix has ceil(length / 8) octets. The UPDATE withdrawn_routes and nlri are prefix lists too, and both UPDATE length fields are Auto.
• VPN prefixes: length counts the label (24 bits), the route distinguisher (64 bits) and the prefix bits, so a /24 is 112. One label is modelled (bos: 1); a withdrawal usually carries label: 0x80000, bos: 0 (the 0x800000 compatibility value).
• Route distinguishers are 8 octets: 0000 fde9 00000064 is 65001:100 (type 0), 0001 c0000201 0001 is 192.0.2.1:1 (type 1).
• EVPN routes (8-bit type, 8-bit length): ethernet_ad (1), mac_ip (2), inclusive_multicast (3), ethernet_segment (4), ip_prefix (5, IPv4, 34 octets) and ipv6_prefix (5, IPv6, 58 octets). ESIs are 10 octets. Label fields are the 3-octet wire value: label << 4 | bos for MPLS, the VNI for VXLAN (RFC 8365). IP addresses with a length field (ip_length in bits) are octets.
• The PMSI Tunnel attribute (pmsi_tunnel, code 22) is typed; route targets are extended_communities values (0x0002fde900000064 = RT 65001:100).

- protocol: bgp_update
  fields:
    path_attributes:
      - { origin: 0 }
      - { as_path: { segments: [] } }
      - mp_reach_vpnv4:
          next_hop: [{ rd: "0000000000000000", address: 192.0.2.1 }]
          nlri:
            - { length: 112, label: 16, bos: 1, rd: "0000 fde9 00000064", prefix: "c6 33 64" }
    nlri: []

Extended Length

A path attribute whose value exceeds 255 octets gets the Extended Length flag (0x10) and a 16-bit length automatically. To send the long form for a short attribute (as FRR and GoBGP do for MP_REACH_NLRI), give the item an explicit type:

- { name: mp_reach_ipv4_unicast, type: 0x900e, fields: { next_hop: [192.0.2.1], nlri: [{ length: 24, prefix: "c63364" }] } }

Received attributes keep their form: a parsed Extended Length attribute is exported with its type.

Limits

• OSPFv3 Extended LSAs (RFC 8362) and OSPFv3 Segment Routing (RFC 8666) are not modelled.
• The OSPF TE GMPLS sub-TLVs (RFC 4203) and the RI TLVs other than those listed stay raw items.
• BGP ADD-PATH path identifiers (RFC 7911), label stacks of more than one label (RFC 8277) and the BGP Prefix-SID attribute (RFC 8669) are not modelled. Write such NLRI as octets with the generic mp_reach_nlri / mp_unreach_nlri items; when parsing such NLRI, the octets still rebuild identically but the decoded fields do not line up with them.
• Several BGP messages in one segment, and several typed LSAs in one LSU, are not supported (see the routing protocols page).

Generated reference

Field reference of the protocols on this page (generated from the definitions):

ospfv2_lsa · ospfv2_opaque_lsa · ospfv2_te_lsa · ospfv2_ri_lsa · ospfv2_ext_prefix_lsa · ospfv2_ext_link_lsa · raw