XFS Large Extent Counts Feature No Longer Considered Experimental

As another feature on the way for the upcoming Linux 6.5 kernel cycle is the large extent counts feature of the XFS file-system no longer being treated as "experimental" but is now considered safe to deploy.

Last year in Linux 5.19 there was lots of new XFS code merged, including this "Large Extent Count" feature. XFS' Large Extent Count raises the inode data limit from 2^31 - 1 to a much greater value of to 2^48 - 1. The attr fork extent counter limit also increases from 2^15 - 1 to 2^32 - 1.

This feature is implemented by adding a 64-bit inode field for the data fork extent counter and the existing 32-bit inode data fork extent counter is then used to hold the attribute fork extent counter. But this is a backwards incompatible change and thus making or repairing an XFS file-system with large extent counts does prevent being able to mount the XFS file-system on older kernels.

Enabling the XFS large extent counts requires using the "nrext64" option when running xfs_mkfs or xfs_repair also has support for upgrading an existing file-system to large extent counters with "nrext64" as well.

In any case, given the feature has been upstream in the Linux kernel for about one year and there has been no bug reports, the XFS developers are ready to declare it stable.

This patch in xfs-linux.git's for-next branch ahead of the Linux 6.5 cycle removes the experimental warning on large extent counts. Regardless though keep in mind though if using the "nrext64" option you lose the ability for mounting XFS file-systems on kernels prior to last year.