Acceptor timeout config

[vasilakisfil]

Is there a way to limit how much time the acceptor is going to wait for the TLS handshake to take place from the client side? If a malicious client never sends its TLS handshake request part, will this block forever?

If yes then probably the example should be updated to use tokio's timeout.

[rfarr]

@vasilakisfil This is how I've done it. The key is to not await on the accept() call but instead wrap that future in the tokio timeout.

let tls_config = rustls::ServerConfig::builder()
    .with_safe_defaults()
    .with_no_client_auth()
    .with_single_cert(certs, private_key)
    .unwrap();

let tls_acceptor = tokio_rustls::TlsAcceptor::from(tls_config);

// ...
// Inside the acceptor loop after you have a tcp connection already accepted
// ...

let tls_stream = tls_acceptor.accept(tcp_stream).map_err(|e| {
    std::io::Error::new(std::io::ErrorKind::Other, format!("Client TLS connect fail: {:?}", e))
});

// Wrap the accept future with a timeout to put an upper limit on how long a client can take to complete
// the TLS handshake and get to a TLS "connected" state
let tls_handshake_timeout_seconds: u8 = 5;
let tls_stream = tokio::time::timeout(tokio::time::Duration::from_secs(tls_handshake_timeout_seconds.into()), tls_stream).map_err(move |_e| {
  std::io::Error::new(std::io::ErrorKind::TimedOut, format!("Client TLS handshake timeout after {} seconds", tls_handshake_timeout_seconds))
 });
 
let tls_stream = match tls_stream.await {
    // Happy path
    Ok(tls_stream) => tls_stream,
    // Some kind of error
    Err(e) => ...
    // Timeout
    Ok(Err(e)) => ...
};

[vasilakisfil]

Thanks, yes I have done the same thing. On event-driven architectures (aka channels), not having a timeout limit by default can lead to serious unexpected stalls in an application. Probably it's a good idea to add something on the documentation about it.

[nirbheek]

This seems like a serious issue to me, the lack of a TLS timeout means that applications built on tokio-rustls are vulnerable out of the box to a DoS attack to exhaust FDs, like a very basic Slowloris attack but with TLS.

I think by default there should be an aggressive configurable timeout for completion of the TLS handshake.

You can do a DoS on the default rustls server example by just spawning tons of nc <host> <port> instances that will never, ever timeout.

[Congyuwang]

I think this is indeed quite needed! I find my server with a lot of ESTABLISHED connection with 0/0 bytes communication that hangs forever, even though I have a timeout layer in my Axum service. It turns out that these connections await at tls_acceptor.accept(stream).await.

[djc]

(Make sure to also take care of the LazyConfigAcceptor when we tackle this.)

[stormshield-gt]

Just to mention it, another solution is to put the application behind a reverse proxy, as it was the general wisdom in the early days of Go.

[Congyuwang]

Just to mention it, another solution is to put the application behind a reverse proxy, as it was the general wisdom in the early days of Go.

That’s a good idea.