This work addresses the difficult problem of fault tolerant control and closed-loop control allocation for spacecraft attitude control system with actuator failures, actuator saturation, and external disturbances. As a fundamental step, a modified fault diagnosis observer is proposed utilizing the iterative learning methodology to reconstruct the actuator failures in real-time. On the basis of the above reconstructed failure information, a fault tolerant control scheme incorporating with a parameter adjusting law is presented to enforce the spacecraft attitude control system to reach the real sliding mode surface in finite time. Meanwhile, the singularity of control command will be avoided using the saturation function, and the chattering problem restrained by the adjusting law. Furthermore, with the concept of control allocation, a novel on-line closed-loop constrained optimal fault tolerant control allocation scheme is employed to distribute the signals synthesized by the baseline controller over the redundant actuators with failures and constraints. The stability of the closed-loop fault tolerant control allocation process is guaranteed by the theory of the closedloop discrete-time feedback control system. The key achievement of the proposed systematic strategy is that the whole closedloop attitude control system can theoretically be guaranteed to be stable. Numerical simulations are carried out to verify the effectiveness and superiority of the developed approach.