diff --git a/net/mptcp/protocol.h b/net/mptcp/protocol.h
index fc00dd587a297..d77b25636125b 100644
--- a/net/mptcp/protocol.h
+++ b/net/mptcp/protocol.h
@@ -312,7 +312,6 @@ struct mptcp_sock {
 
 	u32 setsockopt_seq;
 	char		ca_name[TCP_CA_NAME_MAX];
-	struct mptcp_sock	*dl_next;
 };
 
 #define mptcp_data_lock(sk) spin_lock_bh(&(sk)->sk_lock.slock)
diff --git a/net/mptcp/subflow.c b/net/mptcp/subflow.c
index 047e46dd028dd..52a747a80e88e 100644
--- a/net/mptcp/subflow.c
+++ b/net/mptcp/subflow.c
@@ -1724,16 +1724,31 @@ static void subflow_state_change(struct sock *sk)
 void mptcp_subflow_queue_clean(struct sock *listener_sk, struct sock *listener_ssk)
 {
 	struct request_sock_queue *queue = &inet_csk(listener_ssk)->icsk_accept_queue;
-	struct mptcp_sock *msk, *next, *head = NULL;
-	struct request_sock *req;
-	struct sock *sk;
+	struct request_sock *req, *head, *tail;
+	struct mptcp_subflow_context *subflow;
+	struct sock *sk, *ssk;
 
-	/* build a list of all unaccepted mptcp sockets */
+	/* Due to lock dependencies no relevant lock can be acquired under rskq_lock.
+	 * Splice the req list, so that accept() can not reach the pending ssk after
+	 * the listener socket is released below.
+	 */
 	spin_lock_bh(&queue->rskq_lock);
-	for (req = queue->rskq_accept_head; req; req = req->dl_next) {
-		struct mptcp_subflow_context *subflow;
-		struct sock *ssk = req->sk;
+	head = queue->rskq_accept_head;
+	tail = queue->rskq_accept_tail;
+	queue->rskq_accept_head = NULL;
+	queue->rskq_accept_tail = NULL;
+	spin_unlock_bh(&queue->rskq_lock);
+
+	if (!head)
+		return;
 
+	/* can't acquire the msk socket lock under the subflow one,
+	 * or will cause ABBA deadlock
+	 */
+	release_sock(listener_ssk);
+
+	for (req = head; req; req = req->dl_next) {
+		ssk = req->sk;
 		if (!sk_is_mptcp(ssk))
 			continue;
 
@@ -1741,32 +1756,10 @@ void mptcp_subflow_queue_clean(struct sock *listener_sk, struct sock *listener_s
 		if (!subflow || !subflow->conn)
 			continue;
 
-		/* skip if already in list */
 		sk = subflow->conn;
-		msk = mptcp_sk(sk);
-		if (msk->dl_next || msk == head)
-			continue;
-
 		sock_hold(sk);
-		msk->dl_next = head;
-		head = msk;
-	}
-	spin_unlock_bh(&queue->rskq_lock);
-	if (!head)
-		return;
-
-	/* can't acquire the msk socket lock under the subflow one,
-	 * or will cause ABBA deadlock
-	 */
-	release_sock(listener_ssk);
-
-	for (msk = head; msk; msk = next) {
-		sk = (struct sock *)msk;
 
 		lock_sock_nested(sk, SINGLE_DEPTH_NESTING);
-		next = msk->dl_next;
-		msk->dl_next = NULL;
-
 		__mptcp_unaccepted_force_close(sk);
 		release_sock(sk);
 
@@ -1790,6 +1783,13 @@ void mptcp_subflow_queue_clean(struct sock *listener_sk, struct sock *listener_s
 
 	/* we are still under the listener msk socket lock */
 	lock_sock_nested(listener_ssk, SINGLE_DEPTH_NESTING);
+
+	/* restore the listener queue, to let the TCP code clean it up */
+	spin_lock_bh(&queue->rskq_lock);
+	WARN_ON_ONCE(queue->rskq_accept_head);
+	queue->rskq_accept_head = head;
+	queue->rskq_accept_tail = tail;
+	spin_unlock_bh(&queue->rskq_lock);
 }
 
 static int subflow_ulp_init(struct sock *sk)