Hi Peng,
Oh good, just checking. I was feeling a bit decaffeinated this morning and didn't dig into it originally. You lose one degree of freedom for the intercept and one for your covariate. Yes, the DF is consistent across voxels. You could allow DFs to vary across voxels with options in 3dttest++, but then my ability to guide you in the process of interpreting the results requires incantations and a deck of tarot cards. Best to stick to tried and true.
Worth noting that you can also find this correlation using other methods (though I prefer 3dttest++), outlined
here.
-Peter