In practice, visible light signals undergo non-line-of-sight (NLOS) propagation, and in visible light-based positioning (VLP) methods, the NLOS links are usually treated as disturbance sources to simplify the associated signal processing. However, the impact of NLOS propagation on VLP performance is not fully understood. In this paper, we aim to reveal the performance limits of VLP systems in an NLOS propagation environment via Fisher information analysis. Firstly, the closed-form Cramer-Rao lower bound (CRLB) on the estimation error of user detector (UD) location and orientation is established to shed light on the NLOS-based VLP performance limits. Secondly, the information contribution from the NLOS channel is quantified to gain insights into the effect of the NLOS propagation on the VLP performance. It is shown that VLP can gain additional UD location information from the NLOS channel via leveraging the NLOS propagation knowledge. In other words, the NLOS channel can be exploited to improve VLP performance in addition to the line-of-sight (LOS) channel. The obtained closed-form VLP performance limits can not only provide theoretical foundations for the VLP algorithm design under NLOS propagation, but also provide a performance benchmark for various VLP algorithms.