The system security of linear cyber-physical systems (CPSs) with repetitive movement characteristics under injection attacks is investigated in this chapter. The optimal data-driven injection attack strategies through iterative learning are proposed without utilizing the knowledge of the system model. A switching attack strategy with high stealthiness, flexibility, and low energy consumption is established. Finding the optimal attack strategy that maximizes the quadratic cost function of the system is the main objective of this chapter. Different from the existing attack strategies, the proposed one applies the attack input data from the previous iteration at the same sampling time to update the current attack input, thereby improving the attack effect. Furthermore, in order to overcome the difficulty that the attacker is unaware of the attacked system model, a system identification method based on input and output (I/O) data is used to construct a virtual model. An improved projection estimation approach is developed to estimate the parameters of the virtual model with the assistance of the matrix inverse lemma. Finally, a networked direct current (DC) motor simulation is applied to verify the effectiveness of the presented schemes.