Confounding a $2^n$ design into $2^r$ blocks — how many effects are sacrificed

Suppose we wish to have a confounding arrangement in $2^r$ blocks of a $2^n$ factorial experiment. Then, how many interactions have to be confounded?