SUNRPC: Ensure we flush any closed sockets before xs_xprt_free()

We must ensure that all sockets are closed before we call xprt_free()
and release the reference to the net namespace. The problem is that
calling fput() will defer closing the socket until delayed_fput() gets
called.
Let's fix the situation by allowing rpciod and the transport teardown
code (which runs on the system wq) to call __fput_sync(), and directly
close the socket.

Reported-by: Felix Fu <[email protected]>
Acked-by: Al Viro <[email protected]>
Fixes: a73881c ("SUNRPC: Fix an Oops in udp_poll()")
Cc: [email protected] # 5.1.x: 3be232f: SUNRPC: Prevent immediate close+reconnect
Cc: [email protected] # 5.1.x: 89f4249: SUNRPC: Don't call connect() more than once on a TCP socket
Cc: [email protected] # 5.1.x
Signed-off-by: Trond Myklebust <[email protected]>

Author: Trond Myklebust

fs/file_table.c

@@ -412,6 +412,7 @@ void __fput_sync(struct file *file)
 }
 
 EXPORT_SYMBOL(fput);
+EXPORT_SYMBOL(__fput_sync);
 
 void __init files_init(void)
 {

include/trace/events/sunrpc.h

@@ -1004,7 +1004,6 @@ DEFINE_RPC_XPRT_LIFETIME_EVENT(connect);
 DEFINE_RPC_XPRT_LIFETIME_EVENT(disconnect_auto);
 DEFINE_RPC_XPRT_LIFETIME_EVENT(disconnect_done);
 DEFINE_RPC_XPRT_LIFETIME_EVENT(disconnect_force);
-DEFINE_RPC_XPRT_LIFETIME_EVENT(disconnect_cleanup);
 DEFINE_RPC_XPRT_LIFETIME_EVENT(destroy);
 
 DECLARE_EVENT_CLASS(rpc_xprt_event,

net/sunrpc/xprt.c

@@ -930,12 +930,7 @@ void xprt_connect(struct rpc_task *task)
 	if (!xprt_lock_write(xprt, task))
 		return;
 
-	if (test_and_clear_bit(XPRT_CLOSE_WAIT, &xprt->state)) {
-		trace_xprt_disconnect_cleanup(xprt);
-		xprt->ops->close(xprt);
-	}
-
-	if (!xprt_connected(xprt)) {
+	if (!xprt_connected(xprt) && !test_bit(XPRT_CLOSE_WAIT, &xprt->state)) {
 		task->tk_rqstp->rq_connect_cookie = xprt->connect_cookie;
 		rpc_sleep_on_timeout(&xprt->pending, task, NULL,
 				xprt_request_timeout(task->tk_rqstp));

net/sunrpc/xprtsock.c

@@ -879,7 +879,7 @@ static int xs_local_send_request(struct rpc_rqst *req)
 
 	/* Close the stream if the previous transmission was incomplete */
 	if (xs_send_request_was_aborted(transport, req)) {
-		xs_close(xprt);
+		xprt_force_disconnect(xprt);
 		return -ENOTCONN;
 	}
 
@@ -915,7 +915,7 @@ static int xs_local_send_request(struct rpc_rqst *req)
 			-status);
 		fallthrough;
 	case -EPIPE:
-		xs_close(xprt);
+		xprt_force_disconnect(xprt);
 		status = -ENOTCONN;
 	}
 
@@ -1185,6 +1185,16 @@ static void xs_reset_transport(struct sock_xprt *transport)
 
 	if (sk == NULL)
 		return;
+	/*
+	 * Make sure we're calling this in a context from which it is safe
+	 * to call __fput_sync(). In practice that means rpciod and the
+	 * system workqueue.
+	 */
+	if (!(current->flags & PF_WQ_WORKER)) {
+		WARN_ON_ONCE(1);
+		set_bit(XPRT_CLOSE_WAIT, &xprt->state);
+		return;
+	}
 
 	if (atomic_read(&transport->xprt.swapper))
 		sk_clear_memalloc(sk);
@@ -1208,7 +1218,7 @@ static void xs_reset_transport(struct sock_xprt *transport)
 	mutex_unlock(&transport->recv_mutex);
 
 	trace_rpc_socket_close(xprt, sock);
-	fput(filp);
+	__fput_sync(filp);
 
 	xprt_disconnect_done(xprt);
 }