as full-differential n-input fold-cascade amplifier in the picture, we can add the cmfb in the a, b or c point, what is the difference of the three method? which one is better

Hm, as with a lot of things in this world there are multiple ways to do the same thing.

That said, getting everything at proper common mode operating point early in a multiple stage system is generally considered good practice.

That said, location C, is somewhat downstream.

The most common place to add this would be location B. This does not disturb your common mode current below the ddifferential pair and will change the gain of the stage slightly (you are changing the impedance of the diff pair load a bit with control added at node B, but that should be slight) and location A tends to be controlled by a common current bias system.

In analysis all would work, in practice I would tend to use B, develop a multiple use location bias system to control A (all the differential pairs in a system) and would avoid C just due to the fact that you want to introduce CMFB early in the system.