The error amplifier is in a feedback loop. You might miss it if you are looking for local feedback directly between the input and output of the error amplifier, because the loop extends through the current amplifier, the pass device, and the resistive divider formed by RA and RB. As long as this larger loop is stable and has sufficient gain, it will correctly bias the error amplifier in it's linear mode.
Multi-stage feedback loops can employ a variety of combinations of local feedback (around a single stage) and global feedback (around the greater loop) to achieve their requirements for transfer function, stability, and gain. The analysis of some configurations can get very complicated indeed, but the same principles apply as with a simple feedback loop around a single op amp.
In this case, analysis is relatively simple. The current amplifier and pass device both operate (presumably) at approximately unity voltage gain. Essentially, they just buffer the voltage at the output of the error amplifier to VCC. So the first-order small signal analysis proceeds as if RA were simply connected from the output of the amp to it's inverting input.