Version 4.0 Beta 4 of The Global File System
Release Notes: Kernel patches have been updated and GFS now runs on 2.2.18. Initial work has begun on porting to 2.4. It is not ready yet, but it should be shortly after 2.4 is officially released. Initial support has also been added for exporting GFS filesystems to non-GFS nodes via kernel-space NFS. Support for banning a GNBD client from a GNBD server has been implemented, as well as a new STOMITH method that utilizes this functionality. Further enhanments have been made to the `memexpd' high availability failover setup and a new tool, `initds', has been introduced for initializing data segments that maintain lock state info. New tools for DMEP operations have also been introduced.
Other releases
Release Notes: Numerous minor bugfixes and code cleanups.
Release Notes: This version has been ported to recent Linux kernels. There are many speedups, enhancements, and bugfixes.
Release Notes: New kernel patches for Linux 2.4.5, a new hard_panic patch, a new STOMITH method (apc_ms), initialization script updates, and a slew of updates to the man pages. The following bugs have also been addressed: undefined rscsi_disks symbol, memexp kernel thread issue, memexpd performance degradation over time, problems with HIMEM systems, pool and passamble fixes, GNBD fixes, user-space LFS enhancements, and gfs_jadd improvements.
Release Notes: Support for Linux 2.4.4, addition of Lock Value Blocks (LVBs) for performance enhancement, updated flock and fcntl support, a rewrite of the Pool tools for enhanced functionality, performance improvements when GFS is used as a local vs. cluster FS, a fix for an atime bug, improved df performance, and new STOMITH methods. There are incompatibilities in the GFS modules between 4.0.1 and 4.1; please read the Release Notes on how to upgrade.
Release Notes: The following bugs have been fixed: change to the sorting method for directory code that caused stack problems on some architectures, the ability to mount mulitple GFS MemExp file systems, lockups on umount in rare situations have been eliminated, flock now performs correctly when under major contention loads, time stamp modifications on truncate() have been fixed, obscure bug that caused nodes to try and STOMITH themselves has been eliminated, and a reservation bug occasionally encountered when expanding a file system has been fixed.
