100 likes | 327 Views
IPv6 Next Hop for IPv4 Prefix. In BGP Updates, NH not necessarily of same address family as NLRI Currently deployed examples: VPN-IPv4 family has IPv4 NH though with bizarre encoding to make it syntactically “look like” a VPN-IPv4 address IPv6 address family can have IPv4 next hop
E N D
IPv6 Next Hop for IPv4 Prefix • In BGP Updates, NH not necessarily of same address family as NLRI • Currently deployed examples: • VPN-IPv4 family has IPv4 NH • though with bizarre encoding to make it syntactically “look like” a VPN-IPv4 address • IPv6 address family can have IPv4 next hop • using "IPv4-mapped IPv6 address", defined in RFC 4291 as an address of an IPv4 node • Might become typical deployment scenario in dual stack core, so that single BGP session running on IPv4 carries v4 and v6 prefixes. • Other special purpose address families (l2vpn, rt-constraint) have been defined with IP NHs, usually allowing the NH to be either v4 or v6, depending on its length. IDR WG Meeting
Problem • IPv4 AFI/SAFI: • unique, no obvious encoding trick allowing one to use NH of a different AF • Softwire WG requirement: • carry IPv4 routing over IPv6-only single stack core • v4 routing at border routers, not in core • Softwire solution: • IBGP sessions running over v6, carrying v4 NLRI • v4 NLRI need v6 NHs • (data plane uses IPv6 or MPLS encaps) IDR WG Meeting
Solution • Modify IPv4 address family definition: • to allow that NH can be either IPv4 or IPv6, • use length field to distinguish • use BGP capability to determine whether a peer can support this IDR WG Meeting
Rejected Alternatives • Use TLV Encoding instead of length field • In general, a better solution, but • Simpler not to introduce a new type • Given existing deployment of strange encoding solutions for all cases but 4-over-6, hardly seems worthwhile to have a “purer” solution for this one case • New AFI/SAFI for 4/6 case • More complicated • Operationally • Update for given NLRI would change AFI/SAFI as it travels, introducing all kinds of corner cases to track down. IDR WG Meeting
Is It Needed? • Why not just force NH and NLRI to be same? • Support “v4-free core” as a overlay topology of v6 tunnels. • Run additional BGP sessions, on v4, over the overlay • Softwire rather than IDR problem, but even so: • additional BGP sessions a complication • overlay topology a complication • not much to be said for this alternative IDR WG Meeting
Not Without Cost • Simple solution, but not without cost: • when installing v4 route, need to go to v6 routing table to validate next hop • an implementation complexity • But: • already done in VPN support • operational simplicity trumps implementation simplicity IDR WG Meeting
Info-SAFI and Encapsulations • Softwire WG concerned with situations in which e.g., v4 packets need to be tunneled through a v4-free core. • Easily done with MPLS, but a more complicated solution is desired, in which: • IP encapsulations are used to tunnel packets through core • Choice of encaps technology is a matter of policy • Policies may be dependent on characteristics of tunnel endpoints • e.g., if both endpoints belong to a particular class of routers, use encaps X, else use encaps Y • Need way to advertise membership in a class. • Some IP encapsulations (GRE, L2TPv4) lack appropriate native signaling protocols: • Need way to pass info needed to create the encaps header. IDR WG Meeting
Proposal • Create new SAFI, info-SAFI, to carry info about a particular BGP speaker. • Address of BGP speaker embedded in NLRI. • Characteristics of BGP speaker advertised as (extended) communities attached to info-SAFI. • For particular encaps techniques, e.g., L2TPv3, define new attribute to carry signaling info. • Allows multiple info-SAFIs per speaker: • some can carry slowly changing attributes and others more rapidly changing attributes • but no semantics attached to how the attributes are distributed over the info-SAFIs. IDR WG Meeting
Obvious Alternative • Just use /32 prefix identifying BGP speaker, and attach attributes to that • Why not? • Route distribution constraints might be differrent • Don’t want to run into trouble with summarization policies IDR WG Meeting
What This Isn’t • Not Tunnel-SAFI • not “tunnel infrastructure”, not trying to replace NH with tunnel • Not “all purpose use BGP to send any random information about anything” scheme IDR WG Meeting