Large-scale parallel computing applications frequently encounter I/O performance bottlenecks during execution, which adversely impact overall computational efficiency. However, existing I/O tracing tools suffer from high overhead and strong intrusiveness when capturing fine-grained I/O behaviors and performing multi-level analyses. To address this challenge, this study proposes BpfioToolkit, a low-overhead, non-intrusive I/O tracing and analysis toolkit based on eBPF technology. BpfioToolkit aims to support precise analysis of complex parallel I/O patterns by tracing I/O requests issued by parallel applications and recording detailed I/O behavior logs. I/O operations at the MPI-IO layer, system call layer, and virtual file system layer within the I/O stack are efficiently traced, with key metrics such as I/O request frequency, read/write sizes, and file offsets accurately recorded. By correlating I/O behavior data across these layers, BpfioToolkit provides a precise and comprehensive view of I/O behaviors. Experimental evaluations on multiple typical parallel applications and benchmark programs demonstrate that BpfioToolkit maintains extremely low system overhead (only 0.54% to 1.68%) across different I/O intensity scenarios while generating rich I/O behavior data. These data facilitate the identification of I/O performance bottlenecks, such as inefficient I/O access patterns and I/O load imbalance, validating the practicality of BpfioToolkit. BpfioToolkit offers robust technical support for I/O performance analysis and optimization in large-scale parallel computing environments and exhibits broad application prospects.