Close the client side when the upstream closes, in inferproxy

Implemented-By: OpenCode session (model recorded in docs/implementer-log.md)

crates/inferproxy/src/lib.rs

@@ -108,28 +108,34 @@ pub fn serve(listener: UnixListener, upstream: String, limits: Limits) -> io::Re
 }
 
 fn handle(client: UnixStream, upstream: String, open: Arc<AtomicUsize>) {
-    let _guard = OpenGuard::new(open);
     let server = match TcpStream::connect(&upstream) {
         Ok(s) => s,
         Err(_) => return,
     };
-    let _ = forward(&client, &server);
-    drop(_guard);
+    let c2s = {
+        let _guard = OpenGuard::new(open);
+        match forward(&client, &server) {
+            Ok(c2s) => c2s,
+            Err(_) => return,
+        }
+    };
+    let _ = client.shutdown(Shutdown::Both);
+    let _ = server.shutdown(Shutdown::Both);
+    let _ = c2s.join();
 }
 
-fn forward(client: &UnixStream, server: &TcpStream) -> io::Result<()> {
+fn forward(client: &UnixStream, server: &TcpStream) -> io::Result<thread::JoinHandle<()>> {
     let mut c2s_in = client.try_clone()?;
     let mut s2c_out = client.try_clone()?;
     let mut s2c_in = server.try_clone()?;
     let mut c2s_out = server.try_clone()?;
-    thread::scope(|s| {
-        s.spawn(move || {
-            let _ = copy(&mut c2s_in, &mut c2s_out);
-            let _ = c2s_out.shutdown(Shutdown::Write);
-        });
-        s.spawn(move || {
-            let _ = copy(&mut s2c_in, &mut s2c_out);
-        });
+    let c2s = thread::spawn(move || {
+        let _ = copy(&mut c2s_in, &mut c2s_out);
+        let _ = c2s_out.shutdown(Shutdown::Write);
     });
-    Ok(())
+    let _ = thread::spawn(move || {
+        let _ = copy(&mut s2c_in, &mut s2c_out);
+    })
+    .join();
+    Ok(c2s)
 }