The DNG spec covers both mosaic (color filter array, or CFA) and non-mosaic (e.g., RGB) formats. As noted above, for mosaic (CFA) images the PhotometricInterpretation tag needs to be CFA and the # of samples per pixel is usually 1, whereas in the latter case it would be LinearRaw and the # of samples per pixel could be variable (1 for a monochrome image like the Leica M MONOCHROM, 3 for a RGB image, 4 for a GMCY image from a Sony F828, etc.).
In the case of a CFA image, there are additional tags to specify the mosaic pattern and plane layout (CFARepeatPatternDim, CFAPattern, CFAPlaneColor, and CFALayout).
You mentioned the difference between a raw RGB image versus, say, a TIFF (which is typically already rendered). This is specified via the ColorimetricReference tag. For raw files (typically scene referred) this value should be 0. For rendered files like most JPEGs or TIFFs this value should be 1 (output referred, using the ICC profile perceptual dynamic range).