Let $L$ be a subspace of $C^n$ and $P_L$ be the orthogonal projector of $C^n$ onto $L$. For $A\in C^{n\times m}$, the generalized Bott-Duffin (B-D) inverse $A^{(+)}_{(L)}$ is given by $A^{(+)}_{(L)}=P_L(AP_L +P_{L\bot})^{+}$. By defined a nonstandard inner product, a finite formulae is presented to compute Bott-Duffin inverse $A^{(-1)}_{(L)}=P_L(AP_L+P_{L\bot})^{-1}$ and generalized Bott-Duffin inverse $A^{(+)}_{(L)}=P_L(AP_L +P_{L\bot})^{+}$ under the condition $A$ is $L-zero$ (i.e., $AL\cap L^{\bot}=\{0\}$). By this iterative method, when taken the initial matrix $X_0=P_LA^*P_L$, the Bott-duffin inverse $A^{(-1)}_{(L)}$ and generalized Bott-duffin inverse $A^{(+)}_{(L)}$ can be obtained within a finite number of iterations in absence of roundoff errors. Finally a given numerical example illustrates that the iterative algorithm dose converge.