Abstract Specification of the UBIFS File System for Flash Memory

Abstract

Today we see an increasing demand for flash memory because it has certain advantages like resistance against kinetic shock. However, reliable data storage also requires a specialized file system knowing and handling the limitations of flash memory. This paper develops a formal, abstract model for the UBIFS flash file system, which has recently been included in the Linux kernel. We develop formal specifications for the core components of the file system: the inode-based file store, the flash index, its cached copy in the RAM and the journal to save the differences. Based on these data structures we give an abstract specification of the interface operations of UBIFS and prove some of the most important properties using the interactive verification system KIV.

Development of a Verified Flash File System

“…This layer was also the first we have looked at [39]. The idea of UBIFS is to store a collection of data items, called "nodes", that represent files, directories, directory entries, and pages that store the content of files.…”

Formalizing a hierarchical file system

“…In [KJ08], Kang and Jackson presented an elaborated work on flash memory formalism using Alloy. More recently in 2009, Schierl et al [SSHR09] contemplated the formalization of flash memory with reference to UBIFS implementation for Linux. They verify pre and postconditions, but do not consider refinements.…”

An Integrated Formal Methods Tool-Chain and Its Application to Verifying a File System Model

“…Reference [25] already contemplates NAND flash memory peculiarities, such as wear levelling, erase unit reclamation, and tolerance to power loss. More recently, new papers [33,18] on file system formalization have become available. Theorem proving is used in [18], which follows a top down approach in formalizing a hierarchical file system.…”

“…Theorem proving is used in [18], which follows a top down approach in formalizing a hierarchical file system. [33] reports on the bottom up verification of the UBIFS implementation for Linux.…”

“…As for current work on the VFS project itself, our implementation (refinement) strategy is based on the following design principle: whatever abstract model one writes for file systems, it can be refined into diagrams of "atomic" (1NF) simple relations using data transformation by calculation [29]. Inspired by [33], one just has to consider a further, generic refinement step in moving towards the flash level: that of implementing every simple relation by a 4-tuple made of the relation itself, the corresponding RAM and flash indices and the journal. We are currently busy in proving the correctness of this refinement strategy [8].…”