.vscode | ||
contrib | ||
proto | ||
protogen | ||
src | ||
.gitignore | ||
Cargo.lock | ||
Cargo.toml | ||
config.toml | ||
generate.sh | ||
LICENSE | ||
README.md | ||
update-proto.sh |
wl-mitm
wl-mitm
is a filtering man-in-the-middle proxy for Wayland compositors.
Through a toml config file, it allows you to selectively enable only a necessary subset of Wayland protocols
for apps to function.
Wayland's core protocols are generally safe to expose to any program, but there are extensions that could potentially be abused. For example, wlr-screencopy-unstable-v1. A sandboxed program may be able to make use of these extensions to "escape", at least in the sense of accessing screen content, clipboard, etc.
Moreover, access control to sensitive protocols implemented in different compositors may be wildly different. Although something like security-context-v1 exists, what ends up being exposed is still determined by the compositor and this behavior is usually not programmable by the end user.
wl-mitm
solves this by proxying the Wayland socket and exposing a very flexible configuration format to
you, the user. You can elect to allow any subset of Wayland protocols you would like an application to have access
to, or even filter specific requests provided that clients can handle a missed message well. ask_cmd
and
notify_cmd
combines this filtering capability with arbitrary extensibility through external programs. For example,
an ask_cmd
may show a prompt to the user whether to allow a certain Wayland request to proceed. A notify_cmd
may send a desktop notification when an application performs a senstive action through a Wayland request.
This also makes it potentially a very useful debugging tool.
wl-mitm
is intended to be used on top of some other sandboxing system that, at the very least, limits
a program's access to the rest of the filesystem. Otherwise, a malicious application can simply access the original,
unadulterated Wayland socket directly under $XDG_RUNTIME_DIR
.
Building
wl-mitm
requires a Rust compiler supporting Rust 2024.
wl-mitm
relies on generated code and does not use build.rs
or proc_macro
s due to performance issues with
rust-analyzer
. Instead, run ./generate.sh
to generate Rust parser code based on Wayland protocol XMLs located
under proto/
.
After running ./generate.sh
, simply run cargo build --release
to produce a release binary.
Usage
Run wl-mitm
with
wl-mitm <path/to/configuration/file>
Path to the configuration file defaults to ./config.toml
.
This repo contains an example configuration at config.toml
that allows a few base Wayland protocols for standard
desktop apps to function. It also demonstrates the use of ask_cmd
and notify_cmd
by defining filters on clipboard-related
requests. Detailed explanation of the configuration format is also contained in the example.
To launch a program under wl-mitm
, set its WAYLAND_DISPLAY
env variable to whatever listen
is under [socket]
in config.toml
.
Note that you may want to use another container and pass only the wl-mitm
'd socket through for proper isolation.
A Word on Filtering
wl-mitm
gives you a ton of flexibility on filtering. In particular, you are allowed to filter any request on any object
in config.toml
.
However, please keep in mind that while most requests can be filtered, Wayland (and most of its protocols) is not designed to handle
this type of filtering. While a lot of requests can be filtered without consequences, there are a significant number of them that
will result in irrecoverable de-sync between the client and the server. For example, any message that creates a new object ID
(of the type new_id
) will result in undefined behavior if filtered.
wl-mitm
does provide the ability to reject a request with an explicit error. Unfortunately, errors in Wayland are usually
fatal, and clients are not designed to recover from them.
The most reliable (and the only protocol-compliant) way to prevent a client from using certain features is to block the
entire global object where that feature resides. These are usually "manager" objects that live in the global registry,
and can be blocked in allowed_globals
inside config.toml
. Each extension protocol has to expose some manager object
in order for clients to access their functionalities, and thus blocking them there will prevent clients from even knowing
their existence. However, globals can't be selectively blocked using ask_cmd
, because clients usually bind to them as soon
as they start.
Since notify_cmd
never blocks a request, it is safe to use on any request filter.
XWayland
wl-mitm
is explicitly not a sub-compositor. This also means that it can't support XWayland natively like some
subcompositors do, such as sommelier from
ChromeOS. The decision to not implement it as a sub-compositor is made to ensure maximum compatibility with Wayland
protocols and compositors, and to introduce as little additional quirks as possible over the host compositor.
If you would like to use XWayland over a socket filtered by wl-mitm
, here are a few options:
- Use Sommelier's X mode only;
- xwayland-satellite is a compositor-agnostic XWayland implementation in Rust;
- gamescope is an XWayland-only subcompositor specifically intended for games.
Supported Protocols
wl-mitm
aims to include support for all known, non-deprecated Wayland protocols. This is currently achieved by
pulling in all XMLs from the wayland-explorer project.
The update-proto.sh
script is responsible for updating the list of XMLs used to generate Wayland parsers.