Problem statement of Inter-domain Traffic Redirection Risks

An significant reason of inter-domain routing and forwarding inconsistency is inter-domain traffic redirection, where the AS redirects traffic to a new next-hop AS that is not in AS_PATH that source/upstream AS selects. Therefore, the draft uses inter-domain traffic redirection as an example to clarify the risk of this inconsistency. There are many network traffic redirection technologies, such as Policy Based Routing and BGP Flow Specification Rules etc. These technologies are widely used in carrier networks. For example, BGP Flow Specification provides various filtering conditions and processing actions to implement traffic control. This function is used not only to protect the device against denial-of-service (DoS) and distributed DoS (DDoS) attacks, but also used for network traffic optimization. As reliability and effectiveness of traffic redirection are artificially guaranteed, there are risks, such as traffic detour, traffic exposure, traffic black hole, traffic loop, as well as inconsistent traffic paths between the control plane and data plane. Currently, operators have applied the redirection technologies, such as BGP Flowspec, on a large scale. How to deal with these risks needs further discussion.

The operation and maintenance of redirection in the long run is a big challenge. Typo, limitation of out-of-band tool capabilities for configuration verification, network adjustment or failure may cause potential problems without system awareness. The following figure shows the risks of traffic redirection. Take the topology in Figure 1 as an example, In this application scenario, one campus network inter-connects to two providers Network. AS 65003 and AS 65500, AS 65105 and AS 65500 form a provider-customer adjacency relationship. Assume that the user needs to transmit data to the server. According to the routing information on the control plane, the traffic is transmitted through the path [User -- AS65001 -- AS65003 -- AS65104 -- AS65106 -- Server].

Risk 1: Violation of the valley-free principle leads to traffic detour and exposure. Assume ISP1 configures wrong traffic redirection rule, as a result, AS 65003 redirects traffic from AS 65104 to AS 65500. In this case, valley-free principle is violated as AS 65003 and AS 65500 form a provider-customer adjacency relationship. Traffic passes through the AS 65500 and exposes itself to the campus network. Risk 2: The traffic is redirected to a network node that has no route, which leads to a traffic black hole After traffic is redirected from AS 65003 to the AS 65500, the AS 65500 may not have a route to the destination server. In this case, the traffic is discarded, resulting in a traffic black hole. Assume traffic is redirected from AS65003 to AS65500, and AS65500 learns the route to server from AS65105. After the traffic is redirected to the AS65500, it is forwarded to the server according to the route. If the link between AS 65500 and AS 65105 fails, as a result, the route is lost. The traffic is redirected to AS 65500 anyway, which also causes a traffic black hole. Risk 3: One or more traffic redirections on the traffic transmission path may cause a traffic loop. Assume that traffic is redirected from AS 65104 to AS 65105. In AS 65105, the traffic is transmitted to AS 65500, or the traffic is redirected to AS 65500. In this case, if the traffic on AS 65500 is transmitted to AS 65003, then traffic loop occurs. Assume traffic is redirected from AS 65003 to AS 65500, and AS 65500 learns the route to server from AS 65105. After receiving the redirected traffic, the AS65500 forwards the traffic to the server according to the route. If the link between AS6500 and AS65105 fails, AS 65500 learns that the next hop of the route to the server is AS65003 through routing protocols. If the traffic is still redirected to AS 65500 at this time, AS 65500 will send packets back to AS65003, then traffic loop occurs. Risk 4: inconsistent paths on the control plane and data plane may cause O&M risks. The traffic owner expects traffic to be transmitted along the AS path carried in the route, but the actual transmission path is different from the AS path. If the network O&M control system does not obtain traffic redirection information on the network, unpredictable risks may occur during traffic optimization, for example, network congestion. For the risks mentioned above, it is not sufficient to rely on manual operation or automated management. Traffic redirection increases management difficulties and effectiveness requirements. It is necessary to explore technical solutions, such as redirection configuration verification, protocol extension, or path visualization, to reduce risks mentioned above.

Solution 1: AS path validation The root cause of the above risks is that BGP cannot perceive the AS_path generated by the inter-domain traffic redirection. The security of the AS_path only relies on human decision, which inevitably increases the risk of violating valley-free principle. There is the same problem in BGP flowspec, which only considers the verification of the source of the flow specification. The potential solution is to add a redirection path perception capability to BGP, and verify the validity of the redirection path before actually configuring the redirection. If the redirection could lead to the above risks, alert the configurator explicitly. Solution 2: Considerate routing Traffic redirection may generate unexpected inter-domain traffic models, causing some SLAs (service level agreement) to be unguaranteed. A potential solution is to extend BGP to allow redirection rules and the redirection AS_PATH to be advertised to other peers for routing decision. Solution 3: Network visualization Inter-domain redirection also hinders network visualization, making it impossible to determine the forwarding path of packets solely based on routing protocols. A potential solution is to extend the BMP protocol to allow redirection rules and redirected forwarding paths to be sent to the BMP server, improving administrators' ability to visualize and plan their own traffic.

This document makes no request of IANA.

TBD

The authors would like to thank Hao Li.