Reconfigurable intelligent surface (RIS)-aided near-field communications is investigated. First, the necessity of investigating RIS-aided near-field communications and the advantages brought about by the unique spherical-wave-based near-field propagation are discussed. Then, the family of patch-array-based RISs and metasurface-based RISs are introduced along with their respective near-field channel models. A pair of fundamental performance limits of RIS-aided near-field communications, namely their power scaling law and effective degrees-of-freedom, are analyzed for both patch-array-based and metasurface-based RISs, which reveals the potential performance gains that can be achieved. Furthermore, the associated near-field beam training and beamforming design issues are studied, where a two-stage hierarchical beam training approach and a low-complexity element-wise beamforming design are proposed for RIS-aided near-field communications. Finally, a suite of open research problems is highlighted for motivating future research.