CVE-2026-109823

An attacker can read the name of filesystem paths of your application present via Tidewave configuration via /tidewave/connect or via the rendered toolbar.

Neither response checks the Host header. A web page that rebinds its own domain to the loopback address can therefore read the tag with a same-origin request, which carries no Origin header. The paths usually reveal the local username and the project layout.

This issue affects tidewave: from 0.8.0 before 0.9.2.

An attacker can read the name of filesystem paths of your application present via Tidewave configuration via /tidewave/connect or via the rendered toolbar.

Neither response checks the Host header. A web page that rebinds its own domain to the loopback address can therefore read the tag with a same-origin request, which carries no Origin header. The paths usually reveal the local username and the project layout.

This issue affects tidewave: from 0.8.0 before 0.9.2.

An attacker can read the name of filesystem paths of your application present via Tidewave configuration via /tidewave/connect or via the rendered toolbar.

Neither response checks the Host header. A web page that rebinds its own domain to the loopback address can therefore read the tag with a same-origin request, which carries no Origin header. The paths usually reveal the local username and the project layout.

This issue affects tidewave: from 0.8.0 before 0.9.2.

An attacker can read the name of filesystem paths of your application present via Tidewave configuration via /tidewave/connect or via the rendered toolbar.

Neither response checks the Host header. A web page that rebinds its own domain to the loopback address can therefore read the tag with a same-origin request, which carries no Origin header. The paths usually reveal the local username and the project layout.

This issue affects tidewave: from 0.8.0 before 0.9.2.

An attacker can read the name of filesystem paths of your application present via Tidewave configuration via /tidewave/connect or via the rendered toolbar.

Neither response checks the Host header. A web page that rebinds its own domain to the loopback address can therefore read the tag with a same-origin request, which carries no Origin header. The paths usually reveal the local username and the project layout.

This issue affects tidewave: from 0.8.0 before 0.9.2.

Summary

An attacker can read the name of filesystem paths of your application present via Tidewave configuration via /tidewave/connect or via the rendered toolbar.

Neither response checks the Host header. A web page that rebinds its own domain to the loopback address can therefore read the tag with a same-origin request, which carries no Origin header. The paths usually reveal the local username and the project layout.

This issue affects tidewave: from 0.8.0 before 0.9.2.

Details

1. Data collection. Tidewave.tidewave_config_meta/1 builds a map with root (Tidewave.MCP.root()), framework.app_paths (Mix.Project.deps_paths()), wsl_distro, and host_path. It HTML-escapes the JSON encoding of the map into <meta name="tidewave:config">.

2. Exposure. Tidewave.Router.control_html/1 serves the tag on GET /tidewave/connect. For all other routes, Tidewave.call/2 registers a before_send callback that injects the tag into every text/html response when toolbar: true.

3. Read access. check_origin/2 accepts any request without an Origin header, and no code validates Host. After DNS rebinding, the browser sends a same-origin GET from a loopback peer.

Proof of concept

1. Run a Phoenix or Plug application with plug Tidewave on 127.0.0.1:4001.
2. From a loopback peer, send GET /tidewave/connect with Host: attacker.example:4001 and Sec-Fetch-Site: same-origin, and no Origin header. This is the request a DNS-rebound page makes.
3. The server answers 200 OK. The tidewave:config meta tag contains the absolute project root and the dependency paths.
4. Send the same request to any HTML route of the application. The injected toolbar tag contains the same data.

Impact

A malicious web page that a developer visits while the Phoenix development server runs can learn the developer's username, project location, and dependency layout on disk. This information helps an attacker prepare further attacks against the machine.