The effects of metadata corruption on nfs

Abstract

Distributed file systems need to be robust in the face of failures. In this work, we study the failure handling and recovery mechanisms of a widely used distributed file system, Linux NFS. We study the behavior of NFS under corruption of important metadata through fault injection. We find that the NFS protocol behaves in unexpected ways in the presence of these corruptions. On some occasions, incorrect errors are communicated to the client application; inothers, the system hangs applications or crashes outright; in a few cases, success is falsely reported when an operation has failed. We use the results of our study to draw lessons for future designs and implementations of the NFS protocol.