agent UDS: chmod per-agent socket dir 0777 + log web_ui::serve errors

The harness runs as the non-root agent user; the per-agent
/run/hive-agent/<name>/ dir lands at 0755 root:root after
create_dir_all, so bind(2) of web.sock failed with EACCES. The error
was invisible because the web_ui::serve future was tokio::spawn'd with
its JoinHandle dropped — no log, no socket, agent looks unreachable
through the gateway.
This commit is contained in:
müde 2026-05-31 21:38:28 +02:00
commit 8b238bbfaf
2 changed files with 29 additions and 3 deletions

View file

@ -564,15 +564,29 @@ async fn serve_main<S: Surface>(socket: &Path, poll_ms: u64) -> Result<()> {
socket.to_path_buf(),
S::FORGE_IS_MANAGER,
));
tokio::spawn(web_ui::serve(
label,
// Log web_ui::serve's error instead of dropping it. A bare
// `tokio::spawn(web_ui::serve(...))` discards the JoinHandle, so
// any Err (e.g. EACCES from `bind_unix` when HIVE_WEB_SOCKET points
// at a dir the agent user can't write) vanishes — leaving an
// operator with no log line and no socket, debuggable only by
// staring at lifecycle.rs.
let web_ui_args = (
label.clone(),
port,
login_state.clone(),
bus.clone(),
socket.to_path_buf(),
files.clone(),
turn_lock.clone(),
));
);
tokio::spawn(async move {
let (label, port, login_state, bus, socket, files, turn_lock) = web_ui_args;
if let Err(e) =
web_ui::serve(label, port, login_state, bus, socket, files, turn_lock).await
{
tracing::error!(error = %e, "web_ui::serve exited with error");
}
});
if matches!(initial, LoginState::NeedsLogin) {
turn::wait_for_login(&claude_dir, login_state.clone(), &bus, poll_ms).await;
} else {

View file

@ -1219,6 +1219,18 @@ fn set_nspawn_flags(
let socket_dir = crate::agent_sockets::agent_dir_for(agent_name);
std::fs::create_dir_all(&socket_dir)
.with_context(|| format!("create {}", socket_dir.display()))?;
// 0777 so the in-container agent user (`sock`, `iris`, …) can
// `bind(2)` the `web.sock` file here. `create_dir_all` lands
// the dir at 0755 root:root; the harness runs as the non-root
// agent user inside the container and otherwise fails the
// bind with EACCES — silently, since web_ui::serve's spawn
// drops its JoinHandle (the gateway then sees an empty
// agent-sockets.json and the agent appears unreachable).
// Per-agent dir + per-agent bind-mount means no other
// container ever sees this path, so the wide mode is contained.
use std::os::unix::fs::PermissionsExt;
std::fs::set_permissions(&socket_dir, std::fs::Permissions::from_mode(0o777))
.with_context(|| format!("chmod 0777 {}", socket_dir.display()))?;
let _ = write!(
binds,
" --bind={socket_dir}:{socket_dir}",