Harden supervisor shutdown: process group isolation, write-back drain

- Spawn all children (rclone, smbd, webdav) in isolated process groups
  so Ctrl+C doesn't reach them directly — supervisor controls shutdown order
- Wait for rclone VFS write-back queue to drain before unmounting (5min cap)
- Prefer fusermount3 over fusermount, skip if already unmounted

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

warpgate/src/supervisor.rs

@@ -4,6 +4,7 @@
 //! coordinated startup and shutdown. Designed to run as a systemd unit
 //! or standalone (Docker-friendly).
 
+use std::os::unix::process::CommandExt;
 use std::process::{Child, Command};
 use std::sync::atomic::{AtomicBool, Ordering};
 use std::sync::Arc;
@@ -26,6 +27,10 @@ const SIGTERM_GRACE: Duration = Duration::from_secs(3);
 const MAX_RESTARTS: u32 = 3;
 /// Reset restart counter after this period of stable running.
 const RESTART_STABLE_PERIOD: Duration = Duration::from_secs(300);
+/// Max time to wait for write-back queue to drain on shutdown.
+const WRITEBACK_DRAIN_TIMEOUT: Duration = Duration::from_secs(300);
+/// Poll interval when waiting for write-back drain.
+const WRITEBACK_POLL_INTERVAL: Duration = Duration::from_secs(2);
 
 /// Tracks restart attempts for a supervised child process.
 struct RestartTracker {
@@ -170,6 +175,7 @@ fn start_and_wait_mount(config: &Config, shutdown: &AtomicBool) -> Result<Child>
 
     let mut child = Command::new("rclone")
         .args(&args)
+        .process_group(0) // isolate from terminal SIGINT
         .spawn()
         .context("Failed to spawn rclone mount")?;
 
@@ -221,6 +227,7 @@ fn spawn_smbd() -> Result<Child> {
     Command::new("smbd")
         .args(["--foreground", "--debug-stdout", "--no-process-group",
                "--configfile", samba::SMB_CONF_PATH])
+        .process_group(0)
         .spawn()
         .context("Failed to spawn smbd")
 }
@@ -266,6 +273,7 @@ fn spawn_webdav(config: &Config) -> Result<Child> {
     let args = webdav::build_serve_args(config);
     Command::new("rclone")
         .args(&args)
+        .process_group(0)
         .spawn()
         .context("Failed to spawn rclone serve webdav")
 }
@@ -401,6 +409,55 @@ fn graceful_kill(child: &mut Child) {
     let _ = child.wait();
 }
 
+/// Wait for rclone VFS write-back queue to drain.
+///
+/// Polls `vfs/stats` every 2s. Exits when uploads_in_progress + uploads_queued
+/// reaches 0, or after 5 minutes (safety cap to avoid hanging forever).
+fn wait_writeback_drain() {
+    use crate::rclone::rc;
+
+    let deadline = Instant::now() + WRITEBACK_DRAIN_TIMEOUT;
+    let mut first = true;
+
+    loop {
+        match rc::vfs_stats() {
+            Ok(vfs) => {
+                if let Some(dc) = &vfs.disk_cache {
+                    let pending = dc.uploads_in_progress + dc.uploads_queued;
+                    if pending == 0 {
+                        if !first {
+                            println!("  Write-back queue drained.");
+                        }
+                        return;
+                    }
+                    if first {
+                        println!(
+                            "  Waiting for write-back queue ({pending} files pending)..."
+                        );
+                        first = false;
+                    } else {
+                        eprint!("\r  Write-back: {pending} files remaining...  ");
+                    }
+                } else {
+                    return; // no cache info → nothing to wait for
+                }
+            }
+            Err(_) => return, // RC API unavailable → rclone already gone
+        }
+
+        if Instant::now() > deadline {
+            eprintln!();
+            eprintln!(
+                "  Warning: write-back drain timed out after {}s, proceeding with shutdown.",
+                WRITEBACK_DRAIN_TIMEOUT.as_secs()
+            );
+            return;
+        }
+
+        thread::sleep(WRITEBACK_POLL_INTERVAL);
+    }
+}
+
 /// Reverse-order teardown of all services.
 ///
 /// Order: stop smbd → unexport NFS → kill WebDAV → unmount FUSE → kill rclone.
@@ -423,11 +480,23 @@ fn shutdown_services(config: &Config, mount: &mut Child, protocols: &mut Protoco
         println!("  WebDAV: stopped");
     }
 
-    // Lazy unmount FUSE
+    // Wait for write-back queue to drain before unmounting
-    let mount_point = config.mount.point.display().to_string();
+    wait_writeback_drain();
-    let _ = Command::new("fusermount")
+
-        .args(["-uz", &mount_point])
+    // Lazy unmount FUSE (skip if rclone already unmounted on signal)
-        .status();
+    if is_mounted(config).unwrap_or(false) {
+        let mount_point = config.mount.point.display().to_string();
+        let unmounted = Command::new("fusermount3")
+            .args(["-uz", &mount_point])
+            .status()
+            .map(|s| s.success())
+            .unwrap_or(false);
+        if !unmounted {
+            let _ = Command::new("fusermount")
+                .args(["-uz", &mount_point])
+                .status();
+        }
+    }
     println!("  FUSE: unmounted");
 
     // Gracefully stop rclone