From 1b63f015e6a1c9c28b64f4468ec855a951f343cb Mon Sep 17 00:00:00 2001
From: Ryan Patterson <cgamesplay@cgamesplay.com>
Date: Sat, 28 Feb 2026 19:28:43 +0800
Subject: [PATCH] fix stale daemon cleanup

---
 libshpool/src/daemon/mod.rs |  8 +++++++
 libshpool/src/daemonize.rs  | 23 ++++++++++++++----
 shpool/tests/daemon.rs      | 47 ++++++++++++++++++++++++++++++++++++-
 3 files changed, 72 insertions(+), 6 deletions(-)

diff --git a/libshpool/src/daemon/mod.rs b/libshpool/src/daemon/mod.rs
index 5f18e473..1f014db6 100644
--- a/libshpool/src/daemon/mod.rs
+++ b/libshpool/src/daemon/mod.rs
@@ -65,6 +65,14 @@ pub fn run(
         }
         Err(e) => {
             info!("no systemd activation socket: {:?}", e);
+            // If a stale socket file exists (file on disk, nothing listening),
+            // remove it before binding so we don't get EADDRINUSE.
+            if let Err(connect_err) = std::os::unix::net::UnixStream::connect(&socket) {
+                if connect_err.kind() == std::io::ErrorKind::ConnectionRefused {
+                    info!("removing stale socket file at {:?}", socket);
+                    std::fs::remove_file(&socket).context("removing stale socket before bind")?;
+                }
+            }
             (Some(socket.clone()), UnixListener::bind(&socket).context("binding to socket")?)
         }
     };
diff --git a/libshpool/src/daemonize.rs b/libshpool/src/daemonize.rs
index 0fde9416..867aa5a6 100644
--- a/libshpool/src/daemonize.rs
+++ b/libshpool/src/daemonize.rs
@@ -33,11 +33,24 @@ where
 {
     let control_sock = control_sock.as_ref();
 
-    if UnixStream::connect(control_sock).is_ok() {
-        info!("daemon already running on {:?}, no need to autodaemonize", control_sock);
-        // There is already a daemon listening on the control socket, we
-        // don't need to do anything.
-        return Ok(());
+    match UnixStream::connect(control_sock) {
+        Ok(_) => {
+            info!("daemon already running on {:?}, no need to autodaemonize", control_sock);
+            // There is already a daemon listening on the control socket, we
+            // don't need to do anything.
+            return Ok(());
+        }
+        Err(e) if e.kind() == std::io::ErrorKind::ConnectionRefused => {
+            // The socket file exists but nothing is listening (stale socket).
+            // Remove it so the new daemon can bind successfully.
+            info!("stale socket at {:?}, removing before autodaemonizing", control_sock);
+            std::fs::remove_file(control_sock)
+                .with_context(|| format!("removing stale socket at {:?}", control_sock))?;
+        }
+        Err(_) => {
+            // Socket file does not exist or other error; fall through to spawn
+            // daemon.
+        }
     }
     info!("no daemon running on {:?}, autodaemonizing", control_sock);
 
diff --git a/shpool/tests/daemon.rs b/shpool/tests/daemon.rs
index 0957ba48..5a041baf 100644
--- a/shpool/tests/daemon.rs
+++ b/shpool/tests/daemon.rs
@@ -1,7 +1,10 @@
 use std::{
     fmt::Write,
     io::Read,
-    os::unix::{net::UnixListener, process::CommandExt as _},
+    os::unix::{
+        net::{UnixListener, UnixStream},
+        process::CommandExt as _,
+    },
     path,
     process::{Command, Stdio},
     time,
@@ -254,6 +257,48 @@ fn cleanup_socket() -> anyhow::Result<()> {
     Ok(())
 }
 
+#[test]
+#[timeout(30000)]
+fn stale_socket_autodaemonize() -> anyhow::Result<()> {
+    let tmp_dir = tmpdir::Dir::new("/tmp/shpool-test")?;
+    let socket_path = tmp_dir.path().join("shpool.socket");
+
+    // Create a stale socket: UnixListener::drop() closes the fd but does NOT
+    // remove the socket file on Linux/macOS, leaving a dead file on disk.
+    {
+        let _listener = UnixListener::bind(&socket_path).context("binding stale socket")?;
+    }
+    assert!(socket_path.exists(), "stale socket file should persist after listener drop");
+    assert!(
+        matches!(
+            UnixStream::connect(&socket_path).map_err(|e| e.kind()),
+            Err(std::io::ErrorKind::ConnectionRefused)
+        ),
+        "expected ConnectionRefused on stale socket"
+    );
+
+    // shpool list --daemonize should succeed despite the stale socket file.
+    // Without the fix this times out with "control socket never came up".
+    let log_file = tmp_dir.path().join("shpool.log");
+    let out = Command::new(support::shpool_bin()?)
+        .arg("--daemonize")
+        .arg("--socket")
+        .arg(&socket_path)
+        .arg("--log-file")
+        .arg(&log_file)
+        .arg("--config-file")
+        .arg(support::testdata_file("norc.toml"))
+        .arg("list")
+        .output()
+        .context("running shpool list --daemonize")?;
+
+    // Best-effort cleanup of the background daemon spawned by --daemonize.
+    Command::new("pkill").arg("-f").arg(socket_path.to_string_lossy().as_ref()).output().ok();
+
+    assert!(out.status.success(), "shpool list should succeed despite stale socket");
+    Ok(())
+}
+
 #[test]
 #[timeout(30000)]
 fn echo_sentinel() -> anyhow::Result<()> {