exec-server: default remote transport to Noise (#26245)

## Why

The transport in
[openai/codex#26242](https://github.com/openai/codex/pull/26242) needs
to be used by every remote orchestrator-to-executor connection before
JSON-RPC traffic starts.

## Changes

- Generates one executor Noise identity when remote exec-server starts
and registers its public key.
- Creates a harness identity for each physical remote environment
connection.
- Fetches a fresh registry bundle before connecting and validates the
authenticated harness key before completing the executor handshake.
- Multiplexes encrypted logical streams over the existing executor
WebSocket.
- Adds bounded stream, handshake-failure, and reassembly state.
- Adds safe lifecycle diagnostics without logging keys, authorizations,
plaintext, or ciphertext.
- Covers reconnects, replay rejection, validation failure, framing
limits, and encrypted JSON-RPC tool traffic.

## Stack

1. [openai/codex#26242](https://github.com/openai/codex/pull/26242):
Noise channel and relay transport
2. **[openai/codex#26245](https://github.com/openai/codex/pull/26245)**:
remote registration and runtime activation

## Verification

- `just test -p codex-exec-server`
- `just fix -p codex-exec-server`
- `just bazel-lock-check`
- `cargo shear`

---------

Co-authored-by: Codex <noreply@openai.com>
This commit is contained in:
viyatb-oai
2026-06-15 17:39:00 -07:00
committed by GitHub
co-authored by Codex
parent 1fe89de576
commit 6e50b22e55
14 changed files with 1902 additions and 491 deletions
+236 -274
View File
@@ -4,40 +4,49 @@ mod common;
mod relay_proto;
use std::collections::HashMap;
use std::sync::Arc;
use std::sync::Mutex;
use std::sync::atomic::AtomicUsize;
use std::sync::atomic::Ordering;
use std::time::Duration;
use anyhow::Context;
use anyhow::Result;
use anyhow::anyhow;
use anyhow::bail;
use base64::Engine as _;
use base64::engine::general_purpose::STANDARD;
use codex_api::AuthProvider;
use codex_app_server_protocol::JSONRPCError;
use codex_app_server_protocol::JSONRPCMessage;
use codex_app_server_protocol::JSONRPCNotification;
use codex_app_server_protocol::JSONRPCRequest;
use codex_app_server_protocol::JSONRPCResponse;
use codex_app_server_protocol::RequestId;
use codex_exec_server::EnvironmentManager;
use codex_exec_server::ExecParams;
use codex_exec_server::ExecResponse;
use codex_exec_server::ExecServerClient;
use codex_exec_server::ExecServerError;
use codex_exec_server::ExecServerRuntimePaths;
use codex_exec_server::InitializeParams;
use codex_exec_server::InitializeResponse;
use codex_exec_server::FsReadFileParams;
use codex_exec_server::NoiseChannelIdentity;
use codex_exec_server::NoiseChannelPublicKey;
use codex_exec_server::NoiseRendezvousConnectArgs;
use codex_exec_server::NoiseRendezvousConnectBundle;
use codex_exec_server::NoiseRendezvousConnectProvider;
use codex_exec_server::ProcessId;
use codex_exec_server::RemoteEnvironmentConfig;
use codex_utils_path_uri::PathUri;
use futures::FutureExt;
use futures::SinkExt;
use futures::StreamExt;
use futures::future::BoxFuture;
use http::HeaderMap;
use http::HeaderValue;
use pretty_assertions::assert_eq;
use prost::Message as ProstMessage;
use relay_proto::RelayData;
use relay_proto::RelayMessageFrame;
use relay_proto::RelayReset;
use relay_proto::relay_message_frame;
use std::sync::Arc;
use tempfile::TempDir;
use tokio::net::TcpListener;
use tokio::net::TcpStream;
use tokio::time::timeout;
use tokio_tungstenite::WebSocketStream;
use tokio_tungstenite::accept_async;
use tokio_tungstenite::tungstenite::Message;
use uuid::Uuid;
use wiremock::Mock;
use wiremock::MockServer;
use wiremock::ResponseTemplate;
@@ -45,10 +54,11 @@ use wiremock::matchers::header;
use wiremock::matchers::method;
use wiremock::matchers::path;
const ENVIRONMENT_ID: &str = "env-mux-test";
const ENVIRONMENT_ID: &str = "env-noise-relay-test";
const EXECUTOR_REGISTRATION_ID: &str = "registration-1";
const HARNESS_KEY_AUTHORIZATION: &str = "harness-key-authorization";
const REGISTRY_TOKEN: &str = "registry-token";
const RELAY_MESSAGE_FRAME_VERSION: u32 = 1;
const TEST_TIMEOUT: Duration = Duration::from_secs(5);
const TEST_TIMEOUT: Duration = Duration::from_secs(10);
#[derive(Debug)]
struct StaticRegistryAuthProvider;
@@ -62,12 +72,70 @@ impl AuthProvider for StaticRegistryAuthProvider {
}
}
struct FailingNoiseConnectProvider {
attempts: Arc<AtomicUsize>,
}
impl NoiseRendezvousConnectProvider for FailingNoiseConnectProvider {
fn connect_bundle(
&self,
_: NoiseChannelPublicKey,
) -> BoxFuture<'_, Result<NoiseRendezvousConnectBundle, ExecServerError>> {
self.attempts.fetch_add(1, Ordering::SeqCst);
async {
Err(ExecServerError::Protocol(
"test registry connect failure".to_string(),
))
}
.boxed()
}
}
fn static_registry_auth_provider() -> codex_api::SharedAuthProvider {
Arc::new(StaticRegistryAuthProvider)
}
#[tokio::test(flavor = "multi_thread", worker_threads = 2)]
async fn multiplexed_remote_environment_routes_independent_virtual_streams() -> Result<()> {
#[tokio::test]
async fn noise_environment_refreshes_bundle_for_each_connection_attempt() -> Result<()> {
let attempts = Arc::new(AtomicUsize::new(0));
let manager = EnvironmentManager::without_environments();
manager.upsert_noise_environment(
ENVIRONMENT_ID.to_string(),
Arc::new(FailingNoiseConnectProvider {
attempts: Arc::clone(&attempts),
}),
)?;
let backend = manager
.get_environment(ENVIRONMENT_ID)
.context("Noise environment should be materialized")?
.get_exec_backend();
for attempt in 1..=2 {
let result = backend
.start(ExecParams {
process_id: ProcessId::new(format!("proc-{attempt}")),
argv: vec!["true".to_string()],
cwd: PathUri::from_path(std::env::current_dir()?)?,
env_policy: None,
env: HashMap::new(),
tty: false,
pipe_stdin: false,
arg0: None,
})
.await;
assert!(matches!(
result,
Err(ExecServerError::Protocol(ref message))
if message == "test registry connect failure"
));
}
assert_eq!(attempts.load(Ordering::SeqCst), 2);
Ok(())
}
#[tokio::test(flavor = "multi_thread", worker_threads = 4)]
async fn remote_environment_routes_encrypted_exec_server_rpc() -> Result<()> {
let listener = TcpListener::bind("127.0.0.1:0").await?;
let rendezvous_url = format!("ws://{}", listener.local_addr()?);
let registry = MockServer::start().await;
@@ -78,7 +146,19 @@ async fn multiplexed_remote_environment_routes_independent_virtual_streams() ->
.and(header("authorization", format!("Bearer {REGISTRY_TOKEN}")))
.respond_with(ResponseTemplate::new(200).set_body_json(serde_json::json!({
"environment_id": ENVIRONMENT_ID,
"url": rendezvous_url,
"url": format!("{rendezvous_url}/relay?role=environment"),
"security_profile": "noise_hybrid_ik_v1",
"executor_registration_id": EXECUTOR_REGISTRATION_ID,
})))
.mount(&registry)
.await;
Mock::given(method("POST"))
.and(path(format!(
"/cloud/environment/{ENVIRONMENT_ID}/validate"
)))
.and(header("authorization", format!("Bearer {REGISTRY_TOKEN}")))
.respond_with(ResponseTemplate::new(200).set_body_json(serde_json::json!({
"valid": true,
})))
.mount(&registry)
.await;
@@ -95,281 +175,163 @@ async fn multiplexed_remote_environment_routes_independent_virtual_streams() ->
runtime_paths,
));
let (socket, _peer_addr) = timeout(TEST_TIMEOUT, listener.accept())
let environment_websocket = accept_websocket(&listener, "environment").await?;
let executor_public_key = registered_executor_public_key(&registry).await?;
let harness_identity = NoiseChannelIdentity::generate()?;
let client_args = NoiseRendezvousConnectArgs {
bundle: NoiseRendezvousConnectBundle {
websocket_url: format!("{rendezvous_url}/relay?role=harness"),
environment_id: ENVIRONMENT_ID.to_string(),
executor_registration_id: EXECUTOR_REGISTRATION_ID.to_string(),
executor_public_key,
harness_key_authorization: HARNESS_KEY_AUTHORIZATION.to_string(),
},
harness_identity,
client_name: "noise-relay-test".to_string(),
connect_timeout: TEST_TIMEOUT,
initialize_timeout: TEST_TIMEOUT,
resume_session_id: None,
};
let client_task =
tokio::spawn(async move { ExecServerClient::connect_noise_rendezvous(client_args).await });
let harness_websocket = accept_websocket(&listener, "harness").await?;
let captured_frames = Arc::new(Mutex::new(Vec::new()));
let relay_task = tokio::spawn(proxy_relay_frames(
environment_websocket,
harness_websocket,
Arc::clone(&captured_frames),
));
let client = timeout(TEST_TIMEOUT, client_task)
.await
.context("remote environment should connect to fake rendezvous")??;
let mut websocket = timeout(TEST_TIMEOUT, accept_async(socket))
.await
.context("fake rendezvous should accept environment websocket")??;
.context("Noise harness client should connect")???;
let stream_a = "stream-a";
let stream_b = "stream-b";
send_relay_message(
&mut websocket,
stream_a,
/*seq*/ 0,
initialize_request(/*id*/ 1, "relay-test-a")?,
)
.await?;
send_relay_message(
&mut websocket,
stream_b,
/*seq*/ 0,
initialize_request(/*id*/ 1, "relay-test-b")?,
)
.await?;
let response = client
.exec(ExecParams {
process_id: ProcessId::from("proc-1"),
argv: vec!["true".to_string()],
cwd: PathUri::from_path(std::env::current_dir()?)?,
env_policy: None,
env: HashMap::new(),
tty: false,
pipe_stdin: false,
arg0: None,
})
.await?;
assert_eq!(
response,
ExecResponse {
process_id: ProcessId::from("proc-1")
}
);
let initialize_responses = read_relay_messages_by_stream(&mut websocket, /*count*/ 2).await?;
let session_a =
assert_initialize_response(initialize_responses.get(stream_a), stream_a, /*id*/ 1)?;
let session_b =
assert_initialize_response(initialize_responses.get(stream_b), stream_b, /*id*/ 1)?;
assert_ne!(session_a, session_b);
let temp_dir = TempDir::new()?;
let large_file_path = temp_dir.path().join("large-response.bin");
let large_file_contents = vec![0x5a; 128 * 1024];
std::fs::write(&large_file_path, &large_file_contents)?;
let read_response = client
.fs_read_file(FsReadFileParams {
path: PathUri::from_path(large_file_path)?,
sandbox: None,
})
.await?;
assert_eq!(
STANDARD.decode(read_response.data_base64)?,
large_file_contents
);
send_relay_message(
&mut websocket,
stream_a,
/*seq*/ 1,
notification("initialized", serde_json::json!({})),
)
.await?;
send_relay_message(
&mut websocket,
stream_b,
/*seq*/ 1,
notification("initialized", serde_json::json!({})),
)
.await?;
assert_relay_data_is_encrypted(&captured_frames)?;
send_relay_message(
&mut websocket,
stream_a,
/*seq*/ 2,
request(/*id*/ 2, "test/unknown-a", serde_json::json!({})),
)
.await?;
send_relay_message(
&mut websocket,
stream_b,
/*seq*/ 2,
request(/*id*/ 2, "test/unknown-b", serde_json::json!({})),
)
.await?;
let unknown_method_responses =
read_relay_messages_by_stream(&mut websocket, /*count*/ 2).await?;
assert_error_response(
unknown_method_responses.get(stream_a),
stream_a,
/*id*/ 2,
"test/unknown-a",
)?;
assert_error_response(
unknown_method_responses.get(stream_b),
stream_b,
/*id*/ 2,
"test/unknown-b",
)?;
send_relay_reset(&mut websocket, stream_a, "test_reset").await?;
send_relay_message(
&mut websocket,
stream_b,
/*seq*/ 3,
request(
/*id*/ 3,
"test/unknown-b-after-reset",
serde_json::json!({}),
),
)
.await?;
let (stream_id, message) = read_relay_message(&mut websocket).await?;
assert_eq!(stream_id, stream_b);
assert_error_response(
Some(&message),
stream_b,
/*id*/ 3,
"test/unknown-b-after-reset",
)?;
websocket.close(None).await?;
drop(client);
relay_task.abort();
remote_environment.abort();
let _ = relay_task.await;
let _ = remote_environment.await;
Ok(())
}
async fn send_relay_message<S>(
websocket: &mut WebSocketStream<S>,
stream_id: &str,
seq: u32,
message: JSONRPCMessage,
) -> Result<()>
where
S: tokio::io::AsyncRead + tokio::io::AsyncWrite + Unpin,
{
let payload = serde_json::to_vec(&message)?;
let frame = RelayMessageFrame {
version: RELAY_MESSAGE_FRAME_VERSION,
stream_id: stream_id.to_string(),
ack: 0,
ack_bits: 0,
body: Some(relay_message_frame::Body::Data(RelayData {
seq,
segment_index: 0,
segment_count: 1,
payload,
})),
};
send_relay_frame(websocket, frame).await
async fn accept_websocket(
listener: &TcpListener,
role: &str,
) -> Result<WebSocketStream<TcpStream>> {
let (socket, _peer_addr) = timeout(TEST_TIMEOUT, listener.accept())
.await
.with_context(|| format!("remote {role} should connect to fake rendezvous"))??;
timeout(TEST_TIMEOUT, accept_async(socket))
.await
.with_context(|| format!("fake rendezvous should accept {role} websocket"))?
.map_err(Into::into)
}
async fn send_relay_reset<S>(
websocket: &mut WebSocketStream<S>,
stream_id: &str,
reason: &str,
) -> Result<()>
where
S: tokio::io::AsyncRead + tokio::io::AsyncWrite + Unpin,
{
send_relay_frame(
websocket,
RelayMessageFrame {
version: RELAY_MESSAGE_FRAME_VERSION,
stream_id: stream_id.to_string(),
ack: 0,
ack_bits: 0,
body: Some(relay_message_frame::Body::Reset(RelayReset {
reason: reason.to_string(),
})),
},
)
.await
async fn registered_executor_public_key(registry: &MockServer) -> Result<NoiseChannelPublicKey> {
let requests = registry
.received_requests()
.await
.context("wiremock should retain requests")?;
let request = requests
.iter()
.find(|request| request.url.path().ends_with("/register"))
.context("exec-server should register before connecting")?;
let body: serde_json::Value = serde_json::from_slice(&request.body)?;
let key = serde_json::from_value(body["executor_public_key"].clone())?;
Ok(key)
}
async fn send_relay_frame<S>(
websocket: &mut WebSocketStream<S>,
frame: RelayMessageFrame,
) -> Result<()>
where
S: tokio::io::AsyncRead + tokio::io::AsyncWrite + Unpin,
{
websocket
.send(Message::Binary(frame.encode_to_vec().into()))
.await?;
async fn proxy_relay_frames(
mut environment: WebSocketStream<TcpStream>,
mut harness: WebSocketStream<TcpStream>,
captured_frames: Arc<Mutex<Vec<Vec<u8>>>>,
) -> Result<()> {
loop {
tokio::select! {
message = environment.next() => {
let Some(message) = message else {
break;
};
let message = message?;
capture_binary_frame(&captured_frames, &message);
harness.send(message).await?;
}
message = harness.next() => {
let Some(message) = message else {
break;
};
let message = message?;
capture_binary_frame(&captured_frames, &message);
environment.send(message).await?;
}
}
}
Ok(())
}
async fn read_relay_messages_by_stream<S>(
websocket: &mut WebSocketStream<S>,
count: usize,
) -> Result<HashMap<String, JSONRPCMessage>>
where
S: tokio::io::AsyncRead + tokio::io::AsyncWrite + Unpin,
{
let mut messages = HashMap::new();
for _ in 0..count {
let (stream_id, message) = read_relay_message(websocket).await?;
if messages.insert(stream_id.clone(), message).is_some() {
bail!("received duplicate response for stream {stream_id}");
}
}
Ok(messages)
}
async fn read_relay_message<S>(
websocket: &mut WebSocketStream<S>,
) -> Result<(String, JSONRPCMessage)>
where
S: tokio::io::AsyncRead + tokio::io::AsyncWrite + Unpin,
{
loop {
let frame = timeout(TEST_TIMEOUT, websocket.next())
.await
.context("timed out waiting for relay frame")?
.ok_or_else(|| anyhow!("environment websocket closed"))??;
match frame {
Message::Binary(bytes) => {
let frame = RelayMessageFrame::decode(bytes.as_ref())?;
let stream_id = frame.stream_id;
let Some(relay_message_frame::Body::Data(data)) = frame.body else {
continue;
};
let message = serde_json::from_slice(&data.payload)?;
return Ok((stream_id, message));
}
Message::Ping(_) | Message::Pong(_) => {}
Message::Close(_) => bail!("environment websocket closed"),
Message::Text(_) => bail!("environment sent text frame on relay websocket"),
Message::Frame(_) => {}
}
fn capture_binary_frame(captured_frames: &Mutex<Vec<Vec<u8>>>, message: &Message) {
if let Message::Binary(bytes) = message {
captured_frames
.lock()
.unwrap_or_else(std::sync::PoisonError::into_inner)
.push(bytes.to_vec());
}
}
fn initialize_request(id: i64, client_name: &str) -> Result<JSONRPCMessage> {
Ok(request(
id,
"initialize",
serde_json::to_value(InitializeParams {
client_name: client_name.to_string(),
resume_session_id: None,
})?,
))
}
fn request(id: i64, method: &str, params: serde_json::Value) -> JSONRPCMessage {
JSONRPCMessage::Request(JSONRPCRequest {
id: RequestId::Integer(id),
method: method.to_string(),
params: Some(params),
trace: None,
})
}
fn notification(method: &str, params: serde_json::Value) -> JSONRPCMessage {
JSONRPCMessage::Notification(JSONRPCNotification {
method: method.to_string(),
params: Some(params),
})
}
fn assert_initialize_response(
message: Option<&JSONRPCMessage>,
stream_id: &str,
id: i64,
) -> Result<Uuid> {
let message = message.ok_or_else(|| anyhow!("missing initialize response for {stream_id}"))?;
let JSONRPCMessage::Response(JSONRPCResponse {
id: response_id,
result,
}) = message
else {
bail!("expected initialize response for {stream_id}, got {message:?}");
};
assert_eq!(response_id, &RequestId::Integer(id));
let response: InitializeResponse = serde_json::from_value(result.clone())?;
Ok(Uuid::parse_str(&response.session_id)?)
}
fn assert_error_response(
message: Option<&JSONRPCMessage>,
stream_id: &str,
id: i64,
expected_method: &str,
) -> Result<()> {
let message = message.ok_or_else(|| anyhow!("missing error response for {stream_id}"))?;
let JSONRPCMessage::Error(JSONRPCError {
id: response_id,
error,
}) = message
else {
bail!("expected error response for {stream_id}, got {message:?}");
};
assert_eq!(response_id, &RequestId::Integer(id));
fn assert_relay_data_is_encrypted(captured_frames: &Mutex<Vec<Vec<u8>>>) -> Result<()> {
let captured_frames = captured_frames
.lock()
.unwrap_or_else(std::sync::PoisonError::into_inner);
let mut data_frames = 0;
for encoded in captured_frames.iter() {
let frame = RelayMessageFrame::decode(encoded.as_slice())?;
let Some(relay_message_frame::Body::Data(data)) = frame.body else {
continue;
};
data_frames += 1;
let payload = String::from_utf8_lossy(&data.payload);
assert!(!payload.contains("initialize"));
assert!(!payload.contains("process/start"));
assert!(!payload.contains("noise-relay-test"));
}
assert!(
error.message.contains(expected_method),
"expected error for {stream_id} to mention {expected_method}, got {}",
error.message
data_frames >= 4,
"expected encrypted request and response frames"
);
Ok(())
}