Failure Analysis of SGI XFS File system
Krishna Pradeep Tamma, Shreepadma Venugopalan
Abstract: Commodity file systems expect a fail stop disk. But
todays disks fail in unexpected ways. Disks exhibit latent sector
errors, silent corruption and transient failures to name a few. In this
paper we study the behavior of SGI XFS to such errors. File systems
play a key role in handling most data and hence the failure
handling policy of the file system plays a major role in ensuring the
integrity of the data. XFS is a highly scalable journaling file system
developed by SGI. We analyze the failure handling policy of XFS file
system. We fail reads and writes of various blocks that originate from
the file system and analyze the file system behavior. Some such blocks
include super block, journal header block, commit block, data block,
index blocks. We classify our errors according to the extended IRON
Taxonomy. We see that XFS is vulnerable to these failures and does not
possess a uniform failure handling policy. Further the file system at
times silently corrupts data. We also see that XFS has very little
internal redundancy even for critical structures such as the super
block and B+Tree root. Finally we see that XFS, like most commodity
file systems, fails to recover from these failures putting the data at
risk.
Available as:
Postscript
or PDF
Click
here to download our software.