mptcp: add annotations around msk->subflow accesses

The MPTCP can access the first subflow socket in a few spots
outside the socket lock scope. That is actually safe, as MPTCP
will delete the socket itself only after the msk sock close().

Still the such accesses causes a few KCSAN splats, as reported
by Christoph. Silence the harmless warning adding a few annotation
around the relevant accesses.

Fixes: 71ba088c ("mptcp: cleanup accept and poll")
Reported-by: Christoph Paasch <cpaasch@apple.com>
Closes: https://github.com/multipath-tcp/mptcp_net-next/issues/402Reviewed-by: Mat Martineau <martineau@kernel.org>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Signed-off-by: Mat Martineau <martineau@kernel.org>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

net/mptcp/protocol.c

@@ -91,7 +91,7 @@ static int __mptcp_socket_create(struct mptcp_sock *msk)
 		return err;
 
 	msk->first = ssock->sk;
-	msk->subflow = ssock;
+	WRITE_ONCE(msk->subflow, ssock);
 	subflow = mptcp_subflow_ctx(ssock->sk);
 	list_add(&subflow->node, &msk->conn_list);
 	sock_hold(ssock->sk);
@@ -2282,7 +2282,7 @@ static void mptcp_dispose_initial_subflow(struct mptcp_sock *msk)
 {
 	if (msk->subflow) {
 		iput(SOCK_INODE(msk->subflow));
-		msk->subflow = NULL;
+		WRITE_ONCE(msk->subflow, NULL);
 	}
 }
 
@@ -3136,7 +3136,7 @@ struct sock *mptcp_sk_clone(const struct sock *sk,
 	msk = mptcp_sk(nsk);
 	msk->local_key = subflow_req->local_key;
 	msk->token = subflow_req->token;
-	msk->subflow = NULL;
+	WRITE_ONCE(msk->subflow, NULL);
 	msk->in_accept_queue = 1;
 	WRITE_ONCE(msk->fully_established, false);
 	if (mp_opt->suboptions & OPTION_MPTCP_CSUMREQD)
@@ -3184,7 +3184,7 @@ static struct sock *mptcp_accept(struct sock *sk, int flags, int *err,
 	struct socket *listener;
 	struct sock *newsk;
 
-	listener = msk->subflow;
+	listener = READ_ONCE(msk->subflow);
 	if (WARN_ON_ONCE(!listener)) {
 		*err = -EINVAL;
 		return NULL;
@@ -3736,10 +3736,10 @@ static int mptcp_stream_accept(struct socket *sock, struct socket *newsock,
 
 	pr_debug("msk=%p", msk);
 
-	/* buggy applications can call accept on socket states other then LISTEN
+	/* Buggy applications can call accept on socket states other then LISTEN
 	 * but no need to allocate the first subflow just to error out.
 	 */
-	ssock = msk->subflow;
+	ssock = READ_ONCE(msk->subflow);
 	if (!ssock)
 		return -EINVAL;
 
@@ -3813,10 +3813,12 @@ static __poll_t mptcp_poll(struct file *file, struct socket *sock,
 	state = inet_sk_state_load(sk);
 	pr_debug("msk=%p state=%d flags=%lx", msk, state, msk->flags);
 	if (state == TCP_LISTEN) {
-		if (WARN_ON_ONCE(!msk->subflow || !msk->subflow->sk))
+		struct socket *ssock = READ_ONCE(msk->subflow);
+
+		if (WARN_ON_ONCE(!ssock || !ssock->sk))
 			return 0;
 
-		return inet_csk_listen_poll(msk->subflow->sk);
+		return inet_csk_listen_poll(ssock->sk);
 	}
 
 	if (state != TCP_SYN_SENT && state != TCP_SYN_RECV) {

net/mptcp/protocol.h

@@ -305,7 +305,11 @@ struct mptcp_sock {
 	struct list_head rtx_queue;
 	struct mptcp_data_frag *first_pending;
 	struct list_head join_list;
-	struct socket	*subflow; /* outgoing connect/listener/!mp_capable */
+	struct socket	*subflow; /* outgoing connect/listener/!mp_capable
+				   * The mptcp ops can safely dereference, using suitable
+				   * ONCE annotation, the subflow outside the socket
+				   * lock as such sock is freed after close().
+				   */
 	struct sock	*first;
 	struct mptcp_pm_data	pm;
 	struct {