The XFS filesystem could gain exciting new features, and more specifically, autonomous self-healing capabilities in the upcoming Linux kernel 7.0 cycle, following a pull request submitted by XFS maintainer Darrick J. Wong.
The proposal, called “xfs: autonomous self-healing of filesystems,” aims for the 7.0 merge window. It brings a new system to report filesystem problems in real time and lets userspace programs fix them automatically. This patchset uses new VFS error-reporting tools from Amutable’s CTO Christian Brauner, which are also planned for Linux kernel 7.0.
The main change adds a kernel feature that sends out health events when XFS finds problems like metadata corruption, file I/O errors, media check failures, or big changes such as shutdowns and unmounts.
Rather than just using kernel logs, these events go through a special anonymous file descriptor. Userspace programs with CAP_SYS_ADMIN rights can open this descriptor and read the event data as C structs.
Each event is queued internally, with limits in place to prevent resource exhaustion. The design allows multiple health conditions to be reported without blocking normal filesystem operation.
The patchset also introduces a new media verification ioctl. If media verification finds problems, those results go into the same health monitoring system, making sure integrity issues are reported in a consistent way.
On the userspace side, a new daemon called xfs_healer will handle these health events and start repairs automatically. The daemon is set up to be managed by systemd and can start on its own using fanotify. It will only block unmounts if a repair is happening.
If accepted, this change would update how XFS deals with errors while running. In the past, repairs were mostly reactive and needed manual work with tools like xfs_repair. The new system allows for real-time monitoring and could let some problems be fixed automatically.
As of now, these changes are proposed for Linux kernel 7.0 but have not been merged into the mainline yet.
