Request for Comments: 6138
W. Lu, Ed.
LDP IGP Synchronization for Broadcast Networks
RFC 5443 describes a mechanism to achieve LDP IGP synchronization to prevent black-holing traffic (e.g., VPN) when an Interior Gateway Protocol (IGP) is operational on a link but Label Distribution Protocol (LDP) is not. If this mechanism is applied to broadcast links that have more than one LDP peer, the metric increase procedure can only be applied to the link as a whole but not to an individual peer. When a new LDP peer comes up on a broadcast network, this can result in loss of traffic through other established peers on that network. This document describes a mechanism to address that use- case without dropping traffic. The mechanism does not introduce any protocol message changes.
Status of This Memo
This document is not an Internet Standards Track specification; it is published for informational purposes.
This document is a product of the Internet Engineering Task Force (IETF). It represents the consensus of the IETF community. It has received public review and has been approved for publication by the Internet Engineering Steering Group (IESG). Not all documents approved by the IESG are a candidate for any level of Internet Standard; see Section 2 of RFC 5741.
Information about the current status of this document, any errata, and how to provide feedback on it may be obtained at http://www.rfc-editor.org/info/rfc6138.
Copyright © 2011 IETF Trust and the persons identified as the document authors. All rights reserved.
This document is subject to BCP 78 and the IETF Trust's Legal Provisions Relating to IETF Documents (http://trustee.ietf.org/license-info) in effect on the date of publication of this document. Please review these documents carefully, as they describe your rights and restrictions with respect to this document. Code Components extracted from this document must include Simplified BSD License text as described in Section 4.e of the Trust Legal Provisions and are provided without warranty as described in the Simplified BSD License.
Table of Contents
1. Introduction ....................................................2 2. Conventions Used in This Document ...............................2 3. Problem Statement ...............................................2 4. Solution ........................................................4 5. Scope ...........................................................5 6. Applicability ...................................................5 7. Security Considerations .........................................6 8. Conclusions .....................................................6 9. References ......................................................7 9.1. Normative References .......................................7 9.2. Informative References .....................................7 Acknowledgments ....................................................7 Appendix A. Computation of "Cut-Edge" ..............................8 Appendix B. Sync without Support at One End ........................8
In RFC 5443 [LDP-IGP-SYNC], when [LDP] is not fully operational on a link, the IGP advertises the link with maximum cost to avoid any transit traffic on the link if possible. When LDP becomes operational, i.e., all the label bindings have been exchanged, the link is advertised with its correct cost. This tries to ensure that the LDP Label Switch Path (LSP) is available all along the IGP shortest path. The mechanisms in [LDP-IGP-SYNC] have limitations when applied to a broadcast link. These are described in Section 3. A solution is defined in Section 4.
2. Conventions Used in This Document
The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT", "SHOULD", "SHOULD NOT", "RECOMMENDED", "MAY", and "OPTIONAL" in this document are to be interpreted as described in [RFC2119].
3. Problem Statement
On broadcast networks, a router's Link State Advertisement (LSA) contains a single cost to the broadcast network rather than a separate cost to each peer on the broadcast network. The operation of the mechanism in [LDP-IGP-SYNC] is analyzed using the sample topology in Figure 1, where routers A, B, C, and E are attached to a common broadcast network. Say all links in that topology have a cost of 1 except the link A-PE3, which has a cost of 10. The use-case when router B's link to the broadcast network comes up is analyzed. Before that link comes up, traffic between PE1 and PE2 flows along the bi-directional path PE1-A-C-D-PE2, and traffic between PE1 and PE3 flows along the bi-directional path PE1-A-E-PE3.
| +---+ +---+ |----| B |-----------|PE2| | +---+ +---+ +---+ +---+ | | |PE1|----| A |----| | +---+ +---+ | | | | +---+ +---+ | | |----| C |----| D |----+ | | +---+ +---+ | | | | | | | | +---+ | |----| E |-------------+ | | +---+ | | | | | | | +---+ +---------------------------|PE3| +---+
Figure 1: LDP IGP Sync on a Broadcast Network
In one interpretation of the applicability of [LDP-IGP-SYNC] to broadcast networks, when a new router is discovered on a broadcast network, that network should avoid transit traffic until LDP becomes operational between all routers on that network. This can be achieved by having all the attached routers advertise maximum cost to that network. This should result in traffic that is being sent via that broadcast network to be diverted. However, traffic might be inadvertently diverted to the link that just came up. Until LDP becomes operational, that traffic will be black-holed. An additional problem is route churn in the entire network that results in traffic that should be unaffected taking sub-optimal paths until the high- cost metric is reverted to the normal cost. In Figure 1, when B's link to the broadcast network comes up and it is discovered by routers A, C and E, then A, B, C, and E can all start advertising maximum cost to the broadcast network. A will have B as next-hop to PE2 and will not have a LDP LSP to PE2, resulting in VPN traffic from PE1 to PE2 to be black-holed at A. The route churn at A also results in traffic between PE1 and PE3 to be unnecessarily diverted to the sub-optimal path PE1-A-PE3 until the maximum-cost advertisement is reverted to the normal cost.
This interpretation has the additional complexity of requiring the maximum-cost advertisement to be reverted by all routers after LDP peering between all the routers on the broadcast network is operational. This is non-trivial and needs coordination between all the routers.
In another alternative interpretation of the applicability of [LDP-IGP-SYNC] to broadcast networks, only the router whose link to the broadcast network comes up advertises maximum cost for that link, but other routers continue to advertise the normal cost. In Figure 1, when B's link to the broadcast network comes up, it advertises a high cost to the broadcast network. After the IGP has converged but the LDP peering A-B is not yet operational, A will have B as the next-hop for PE2 and will not have a LDP LSP to PE2. Since A's cost to reach B is not high, A-B-PE2 becomes the shortest path. VPN traffic from PE1 to PE2 will be dropped at A.
The problem described above exists because the Link State Database (LSDB) of the IGP does not describe a link coming up on a broadcast network with a high bi-directional cost to all other routers on that broadcast network. A broadcast network is advertised as a pseudonode containing a list of routers to which the broadcast network is connected, and the cost of all these links from the pseudonode to each router is zero when computing SPF (Shortest Path First).
The solution proposed below removes the link that is coming up from the LSDB unless absolutely necessary. Only the router whose link is coming up plays a role in ensuring this. The other routers on the broadcast network are not involved. The following text describes this in more detail.
During the intra-area SPF algorithm execution, an additional computation is made to detect an alternate path to a directly connected network that does not have any IGP adjacencies.
If a router has a directly connected network that does not have an alternate path to reach it, then the interface to that network is a "cut-edge" in the topology for that router. When a "cut-edge" goes down, the network is partitioned into two disjoint sub-graphs. This property of whether or not an interface is a "cut-edge" is used when an IGP adjacency comes up on that interface. The method to determine whether an interface is a "cut-edge" is described in Appendix A.
During IGP procedures, when the router's first adjacency to the broadcast network is coming up and the LSA is about to be updated with a link to the pseudonode of the broadcast interface, a check is made whether that interface is a "cut-edge". If it is not a "cut-edge", then the updating of the LSA with that link to the pseudonode is postponed until LDP is operational with all the LDP peers on that broadcast interface. After LDP is operational, the LSA is updated with that link to the pseudonode (and the LSA is flooded). If the interface is a "cut-edge", then the updating of the LSA MUST NOT be delayed by LDP's operational state. Note that the IGP and LDP adjacency bring-up procedures are unchanged. The conditional check of whether the interface is a "cut-edge" must be done just before the adjacency is about to be reflected in the LSA.
If the IGP is [OSPF], the Router-LSA is not updated with a "Link Type 2" (link to transit network) for that subnet until LDP is operational with all neighboring routers on that subnet.
Similarly, if the IGP is [IS-IS], the "Link State PDU" is updated with an "IS Reachability TLV" (or an "Extended IS Reachability TLV") to the pseudonode after LDP is operational with all neighboring routers on that subnet.
Note that this solution can be introduced in a gradual manner in a network without any backward compatibility issues.
This document is agnostic to the method that detects LDP to be operational with a neighbor. It does not define any new method to detect that LDP is operational. At the time of publishing this document, LDP End-of-LIB [LDP-EOL] seems to be the preferred method.
Issues arising out of LDP not being configured on some routers or on some interfaces are not specific to the method described in this document and are considered outside the scope of this solution.
The method described in this document can be easily extended to point-to-point (P2P) links. However, an implementation may continue to apply the method described in [LDP-IGP-SYNC] to P2P links but apply the method described in this document to broadcast networks. Both methods can coexist in a network.
The techniques used in this document's solution enable LDP IGP synchronization in many scenarios where one end of the IGP adjacency does not support any LDP IGP sync method. This is an optional benefit and is for further study. Some ways to apply this technique to achieve that benefit are discussed in Appendix B.
7. Security Considerations
This document does not introduce any new security considerations beyond those already described in [LDP-IGP-SYNC].
Note that in [LDP-IGP-SYNC], when a link is advertised with a high metric, an alternate path with a large number of hops can result in the end-to-end path having more than 255 hops and thus result in unreachability. This fact could be exploited if control of metrics falls into the hands of an attacker.
This problem can even exist in a plain IP network with a link-state IGP. If the directly connected path has a higher metric than an alternate path with Time to Live (TTL) greater than 255 hops, then the standard SPF algorithm will conclude that the shortest path is the alternate path although the neighboring node is unreachable through this path. In this case, the link is advertised with its normal metric yet there is unreachability in the network. Thus, this document does not introduce any new issues beyond those in a standard IGP-based IP network, and operators need to apply policy and security to the techniques used to determine and distribute the metrics used on links in their networks.
This document complements [LDP-IGP-SYNC] by providing a solution to achieve LDP IGP synchronization for broadcast networks. It can also coexist with that solution in a network that has a combination of P2P links and broadcast networks. It can also be introduced into a network without backward compatibility issues. The solution in this document can also be used exclusively to achieve LDP IGP synchronization since this solution applies to both P2P links and broadcast networks.
This solution also has useful properties that can be optionally used to achieve LDP IGP synchronization when only one end of the IGP adjacency supports this solution but the other end supports neither this solution nor the one in [LDP-IGP-SYNC].
9.1. Normative References
[RFC2119] Bradner, S., "Key words for use in RFCs to Indicate Requirement Levels", BCP 14, RFC 2119, March 1997.
Jork, M., Atlas, A., and L. Fang, "LDP IGP
Synchronization", RFC 5443, March 2009.
[LDP] Andersson, L., Ed., Minei, I., Ed., and B. Thomas, Ed., "LDP Specification", RFC 5036, October 2007. [OSPF] Moy, J., "OSPF Version 2", STD 54, RFC 2328, April 1998. [IS-IS] International Organization for Standardization, "Intermediate System to Intermediate System intra-domain routeing information exchange protocol for use in conjunction with the protocol for providing the connectionless-mode network service (ISO 8473)", ISO Standard 10589, 2002.
9.2. Informative References
[LDP-EOL] Asati, R., Mohapatra, P., Chen, E., and B. Thomas, "Signaling LDP Label Advertisement Completion", RFC 5919, August 2010.
The authors would like to thank Luyuan Fang, Mikael Abrahamsson, Ben Niven-Jenkins, Bruno Decraene, Jeff Tantsura, and Acee Lindem for their review and useful comments.
Appendix A. Computation of "Cut-Edge"
A "cut-edge" can be computed during an intra-area SPF run or by using results of the previous SPF run. If an SPF run was scheduled but is pending execution, that SPF MUST be executed immediately before any procedure checks whether an interface is a "cut-edge".
An interface is considered a "cut-edge" if, during intra-area SPF (using Dijkstra's algorithm described in Section 16.1 of [OSPF]), there is no alternate path for the directly connected network. Alternately, a "cut-edge" can be detected by the last run of SPF if there is a lack of connectivity to the router-id of a directly connected peer via an alternate path. The router-id can be known during the adjacency bring-up process.
A "cut-edge" computation should not require any extra SPF runs. It should not increase the algorithmic complexity of SPF.
Appendix B. Sync without Support at One End
A useful property of the solution described in this document is that LDP IGP synchronization is achievable in many scenarios where one end of the IGP adjacency does not support any LDP IGP sync method.
For P2P links (or broadcast links on which the IGP operates in P2P mode) the applicability is straightforward. An IGP can establish a P2P adjacency on a P2P link or a broadcast link with the IGP in P2P mode. When a P2P adjacency comes up, the end of the adjacency that supports the solution in this document would not advertise the link to the other router in its LSA unless the edge is a "cut-edge" or until LDP becomes operational. Hence, neither of the two routers will have IGP next-hop as the other router unless the link is a "cut-edge". Consider Figure 1 modified such that the broadcast network is replaced by P2P links between each of A, B, C, and E. Say link A-B is coming up, but only A has implemented the solution in this document whereas B has implemented neither the solution in this document nor the solution in [LDP-IGP-SYNC]. Since A's LSA does not advertise a link to B until LDP is operational, B does not have A as next-hop. After LDP is operational, A advertises the link to B in its LSA. Hence, there is no traffic loss due to LDP LSP not being present.
For broadcast networks, the applicability is not straightforward and should be considered a topic for future study. One way is for the designated router (DR) to stop advertising the link in the pseudonode to the router whose link is coming up until LDP is operational.
Sriganesh Kini (editor) Ericsson 300 Holger Way San Jose, CA 95134 EMail: firstname.lastname@example.org Wenhu Lu (editor) Ericsson 300 Holger Way San Jose, CA 95134 EMail: email@example.com