MeshXFS: An AI-Written POSIX Clustered Filesystem Built on XFS

Summary

MeshXFS is an open-source GPL-2.0 clustered filesystem that allows multiple nodes to mount the same block device read-write while preserving POSIX semantics, coherent caching, and crash consistency. It is built as a heavily modified fork of Linux XFS and adds in-kernel coordination, peer discovery, distributed locking, heartbeat tracking, fencing, journal slicing, and survivor-led recovery. The project says all code that turns XFS into a multinode filesystem was written by AI: Claude Opus running in Claude Code produced a large share, while Claude Fable, GPT, Gemini, and Astra were consulted on lock protocols, recovery races, and fencing. Human involvement is described as setting requirements and running tests rather than writing or line-reviewing the clustering implementation. Version 0.90.130 adds or restores validation for network locking over single and multipath storage at 2, 4, 8, and 16 nodes, supports Ubuntu 26.04 and Debian 12, and fixes several recovery, buffered-write, and freeze/thaw issues. The project also supports a two-node DRBD dual-primary configuration with local disks, but this requires external fencing and has lower write performance than native XFS on the same DRBD device. Validation covers iSCSI direct and multipath configurations, selected Proxmox VE, RHEL 9.8, Ubuntu, and Debian kernels, and extensive coherency, crash, path-fault, and performance tests. The claims remain bounded: other kernels and storage technologies are untested, storage write caches must be power-loss protected, multi-hour endurance is not covered, and many larger or alternative configurations remain development-only. The repository therefore presents MeshXFS both as a usable, narrowly validated clustered filesystem and as an experiment in whether AI can produce complex kernel and distributed-systems code.