MASQUE moves proxy authorization to the destination decision
MASQUE carries proxied transport through HTTP. RFC 9298 defines UDP proxying with the connect-udp protocol, using an HTTP request to name the destination host and port and then carrying UDP payloads in HTTP Datagrams. The tunnel is convenient because it can reuse an HTTP authentication and policy layer. The difficult control is deciding which destinations that layer should permit.
An authenticated client is not automatically an authorized client for every target. A proxy that accepts any host and port becomes an open relay with a TLS or HTTP front door. An attacker can use it to reach internal services, obscure source addresses, or turn the proxy's network position into an SSRF primitive. The proxy needs a destination policy evaluated after DNS resolution and again when the connection is established.
Separate client identity from destination authority
The client certificate, token, or account identifies who is asking. The target URI identifies what the request wants to reach. A policy should combine those inputs with a tenant or device context and an explicit purpose. A remote-access product may allow private application ranges for one group while a general browsing service allows only public destinations for another.
Host allowlists built only on a pre-resolution hostname can be bypassed by a changing DNS answer or a redirect at the destination. The proxy should resolve under its own policy, reject address families that the service does not permit, and apply consistent rules to UDP targets and any related TCP or IP proxying. A public proxy should normally deny private, loopback, link-local, and metadata addresses unless a separate controlled service explicitly needs them.
Datagrams still need accounting
HTTP Datagrams can carry application-controlled payloads after the request succeeds. The proxy should set per-client and per-target rate limits, maximum concurrent flows, idle timeouts, and a clear response when policy changes. It should not log full payloads as a default debugging measure because the traffic may contain credentials, DNS questions, or application data.
The authentication layer should not trust a destination encoded by an untrusted browser without checking the final authority. Reverse proxies and gateways need to pass the intended target through a single canonical representation and reject conflicting host headers or URI components. A request that changes meaning between the front end and the MASQUE service has lost the policy boundary.
Review the deployment
Start with a list of networks the proxy can reach, not a list of ports that happen to be open. Test direct IP requests, hostnames, IPv6, private ranges, metadata services, and a target that changes DNS between validation and connection. Verify what the proxy records for client, target, policy decision, and duration without recording secrets.
MASQUE can reduce application-specific tunnels and give one place to apply egress policy. That value depends on the proxy treating destination authorization as a first-class decision. The tunnel protects the path. It does not decide whether the path should exist.
References
[1] RFC 9298, Proxying UDP in HTTP. https://www.rfc-editor.org/rfc/rfc9298.html
[2] RFC 9484, Proxying IP in HTTP. https://www.rfc-editor.org/rfc/rfc9484.html
[3] RFC 9931, Security Considerations for Optimistic Protocol Transitions in HTTP/1.1. https://www.rfc-editor.org/rfc/rfc9931.html
Top comments (0)