This paper considers cell-free massive MIMO (cfm-MIMO) for the downlink ultra reliable and low-latency communication (URLLC). At the time of writing, cfm-MIMO has only been considered for communication in the long block regime (LBR), whose throughput is determined by the Shannon function rate with the interference treated as Gaussian noise. Conjugate beamforming (CB) is often used as it requires only local channel state information (CSI) for implementation but its design is based on a large-scale nonconvex problem, which is computationally intractable. The rate function in URLLC is much more complex than the Shannon function rate. The paper proposes a special class of CB, which admits a low-scale optimization formulation for computational tractability. Accordingly, a new path-following algorithm, which generates a sequence of better feasible points and converges at least to a locally optimal solution, is developed for optimizing URLLC rates and cfm-MIMO energy efficiency. Furthermore, the paper also develops improper Gaussian signaling to improve both Shannon function rate and URLLC rate.