Automatic generation of commercial floorplans can significantly improve spatial design efficiency and customer experiences. Traditional manual methods are labor-intensive, costly, and prone to inconsistency. While deep learning methods have shown promise, their application to commercial scenarios faces challenges, including limited datasets and insufficient controllability. To address these issues, we introduce a diffusion-based generative framework tailored specifically for commercial floorplan synthesis. We first create a synthetic dataset of diverse commercial floorplans using traditional geometric approaches. Then, leveraging a diffusion-based architecture, our method generates high-quality, realistic commercial floorplans. A subsequent vectorization step converts generated images into practical vector formats. Additionally, our approach supports practical constraints such as boundary masks, path constraints, and bubble diagrams, enabling precise and flexible control. Experiments demonstrate that our method outperforms existing generative models quantitatively and qualitatively. Our work represents a pioneering effort in applying diffusion-based generative methods to commercial floorplan design, providing a flexible and efficient solution for spatial designers.