Bcachefs 1.38 Released With Faster Mounts and Discard Fixes

Bcachefs 1.38 has been released by Kent Overstreet, introducing a range of performance, scalability, and stability improvements to this modern copy-on-write filesystem designed to compete with Btrfs and ZFS. A key highlight is the redesign of the need_discard btree, which now indexes pending discard operations by journal sequence number instead of device and bucket, resolving deadlocks and mount-time allocator issues while improving discard efficiency under heavy workloads.

The release also brings major scalability gains through journal pipelining, increasing the number of in-flight journal writes from 16 to 256 via a dedicated FIFO queue—removing a critical bottleneck for large storage arrays. Systems with many snapshots will benefit from faster mount times thanks to a fix for an O(n²) memory issue in the snapshot table. Additional fixes address SSD misdetection, encryption and nocow incompatibility (now handled automatically), migration alignment problems, and edge cases in device shrink and repair operations.

Tooling improvements include more accurate output from dump sanitize and better handling of journal ranges, while packaging remains limited across distributions. Although support exists in systems like Arch Linux and Fedora, others rely on external repositories and DKMS modules, as Bcachefs is currently maintained outside the mainline Linux kernel.

For more details, see the changelog.