BGP Route Policy Pre-view and Intent Verification Using BGP Monitoring Protocol

BGP route policies (e.g., prefix-lists, AS-path filters, community manipulation) are the primary tool for inter-domain traffic engineering. However, modern policy structures are complex and highly error-prone. When a network operator or an automated orchestrator deploys a modified policy to a router, the resulting BGP updates instantly propagate throughout the network. If the policy contains a logical flaw, widespread traffic disruption occurs before monitoring systems can react. Existing network validation models rely on offline simulation tools. While useful, these simulations cannot perfectly replicate the internal, real-time state machine and multi-vendor nuances of a live router's RIB processing engine. This document defines a protocol-driven mechanism for "on-box" policy pre-viewing. By leveraging a control-plane sandbox, a router executes candidate policies against its live Adj-RIB-In without modifying its active Local-RIB or Forwarding Information Base (FIB). The simulated differences (Deltas) are asynchronously reported via an extended BGP Monitoring Protocol (BMP) session, providing a reliable, zero-risk framework for closed-loop intent verification.

This document introduces a new BMP message type, termed the Route Pre-view Monitoring (RPM) Message (Suggested Type Value: TBD1). The RPM message extends the standard BMP Route Monitoring (RM) structure by appending a mandatory Policy Transaction Header to encapsulate pre-view context.

Policy Transaction ID: A 32-bit unsigned integer linking the pre-viewed routes back to the configuration transaction initialized by the controller. Pre-view Flag: A 1-octet bitmask defined as follows: 0x01 (Add): The route inside the BGP UPDATE would be newly selected and installed due to the candidate policy. 0x02 (Withdraw/Deny): The route would be denied or filtered out by the candidate policy, causing a withdrawal from the active RIB. 0x03 (Modify): The route is still accepted, but its path attributes (e.g., Local-Pref, Communities) are mutated.

When a candidate policy configuration block is staged on the router with a designated Transaction ID, the router MUST NOT evaluate this policy against the active operational Local-RIB. Instead, the router spawns a temporary Sandbox context: The router feeds the existing Adj-RIB-In routes of the targeted peer(s) through the candidate policy engine. The resulting route outputs are gathered into a temporary Shadow Local-RIB. The router performs a mathematical matrix comparison: `Delta = Shadow Local-RIB - Operational Local-RIB`. For each non-zero Delta entry, the router generates an RPM message containing the simulated route state and streams it to the BMP receiver.

The centralized SDN controller listens to the RPM stream. By aggregating these messages across multiple monitored nodes, the controller can verify high-level intents (e.g., ensuring traffic does not transit a restricted Autonomous System).

Once verification concludes, the controller completes the transaction using the device's configuration channel (e.g., NETCONF/YANG): Commit Execution: If the Delta aligns with operator intent, the controller executes a configuration `<commit>`. The router merges the Shadow Local-RIB entries directly into the operational Local-RIB and updates the hardware FIB linecards. Rollback/Discard Execution: If the Delta violates intent, the controller issues a `<discard-changes>` or abort command. The router purges the Sandbox instance and drops the Shadow Local-RIB from memory, leaving active data-plane traffic entirely undisturbed.

Executing comprehensive shadow calculations across multi-million route tables can stress device CPU and memory. Implementations MUST follow these scoping rules: Incremental Evaluation: The router SHOULD limit policy execution only to routes belonging to neighbors directly referenced by the policy modification. Suppression of Unchanged Flows: If a route passes through the candidate policy with its primary path selection status and all attributes completely unchanged, the router MUST NOT generate an RPM message for that route.

This document requests IANA to allocate a new message type value from the "BGP Monitoring Protocol (BMP) Message Types" sub-registry:

RPM messages expose the hypothetical routing shifts of an enterprise fabric prior to deployment. If intercepted, an attacker could deduce intended traffic engineering patterns. The BMP sessions carrying RPM messages MUST be encrypted using TCP-AO or TLS profiles customized for BMP.

The following people made significant contributions to this document:

The authors would like to acknowledge the review and inputs from xxx.