remove locking around tcpSesAllocCount atomic variable

The global tcpSesAllocCount variable is an atomic already and doesn't
really need the extra locking around it. Remove the locking and just use
the atomic_inc_return and atomic_dec_return functions to make sure we
access it correctly.

Signed-off-by: Jeff Layton <jlayton@redhat.com>
Signed-off-by: Steve French <sfrench@us.ibm.com>

fs/cifs/connect.c

@@ -351,11 +351,9 @@ cifs_demultiplex_thread(struct TCP_Server_Info *server)
 
 	current->flags |= PF_MEMALLOC;
 	cFYI(1, ("Demultiplex PID: %d", task_pid_nr(current)));
-	write_lock(&GlobalSMBSeslock);
-	atomic_inc(&tcpSesAllocCount);
-	length = tcpSesAllocCount.counter;
-	write_unlock(&GlobalSMBSeslock);
-	if (length  > 1)
+
+	length = atomic_inc_return(&tcpSesAllocCount);
+	if (length > 1)
 		mempool_resize(cifs_req_poolp, length + cifs_min_rcv,
 				GFP_KERNEL);
 
@@ -745,14 +743,11 @@ multi_t2_fnd:
 		coming home not much else we can do but free the memory */
 	}
 
-	write_lock(&GlobalSMBSeslock);
-	atomic_dec(&tcpSesAllocCount);
-	length = tcpSesAllocCount.counter;
-
 	/* last chance to mark ses pointers invalid
 	if there are any pointing to this (e.g
 	if a crazy root user tried to kill cifsd
 	kernel thread explicitly this might happen) */
+	write_lock(&GlobalSMBSeslock);
 	list_for_each(tmp, &GlobalSMBSessionList) {
 		ses = list_entry(tmp, struct cifsSesInfo,
 				cifsSessionList);
@@ -763,6 +758,8 @@ multi_t2_fnd:
 
 	kfree(server->hostname);
 	kfree(server);
+
+	length = atomic_dec_return(&tcpSesAllocCount);
 	if (length  > 0)
 		mempool_resize(cifs_req_poolp, length + cifs_min_rcv,
 				GFP_KERNEL);