Abstract: This paper describes a method of designing and implementing a multiprocessor UNIX file system based on modified Dijkstra’s semaphore,which can provide two levels of concurrent and exclusive access to the flie:one based on the buffer level and the other based on the file body so as to achieve better parallel access to a multiprocessor file system.The model,structure and characters of the multiprocessor UNIX file system,are presented with emphasis on the principal data structure:low level buffer data structure and other internal data structures,and their management.Finally,data con- sistence in a multiprocessor environment is also discussed