opa-envoy-plugin has an Authorization Bypass via Double-Slash Path Misinterpretation in input.parsed_pathGHSA-9f29-v6mm-pw6w

A security vulnerability has been discovered in how the input.parsed_path field is constructed. HTTP request paths are treated as full URIs when parsed; interpreting leading path segments prefixed with double slashes (//) as authority components, and therefore dropping them from the parsed path. This creates a path interpretation mismatch between authorization policies and backend servers, enabling attackers to bypass access controls by crafting requests where the authorization filter evaluates a different path than the one ultimately served.

Attack example

HTTP request:

GET //admin/users HTTP/1.1
Host: example.com

Policy sees:

The leading //admin path segment is interpreted as an authority component, and dropped from input.parsed_path field:

{
  "parsed_path": ["users"]
}

Backend receives:

//admin/users path, normalized to /admin/users.

Affected Request Pattern Examples

Impact

Users are impacted if all the following conditions apply:

1. Protected resources are path-hierarchical (e.g., /admin/users vs /users)
2. Authorization policies use input.parsed_path for path-based decisions
3. Backend servers apply lenient path normalization

Patches

Go: v1.13.2-envoy-2 Docker: 1.13.2-envoy-2, 1.13.2-envoy-2-static

Workarounds

Users who cannot immediately upgrade opa-envoy-plugin are recommended to apply one, or more, of the workarrounds described below.

1. Enable the merge_slashes Envoy configuration option

As per Envoy best practices, enabling the merge_slashes configuration option in Envoy will remove redundant slashes from the request path before filtering is applied, effectively mitigating the input.parsed_path issue described in this advisory.

2. Use input.attributes.request.http.path instead of input.parsed_path in policies

The input.attributes.request.http.path field contains the unprocessed, raw request path. Users are recommended to update any policy using input.parsed_path to instead use the input.attributes.request.http.path field.

Example

package example

# Use instead of input.parsed_path
parsed_path := split(                                        # tokenize into array
	trim_left(                                               # drop leading slashes
		urlquery.decode(input.attributes.request.http.path), # url-decode the path
		"/",
	),
	"/",
)