Linux Kernel Filesystem Feud: Torvalds Rejects Bcachefs Feature, Citing Merge Window Rules

A dispute has arisen within the Linux kernel development community over recent modifications proposed for the Bcachefs file system. These changes were submitted after the official feature merge window for the 6.16 release had already closed. According to established protocol, only bug fixes are permitted during this phase; however, Bcachefs developers continue to push for the inclusion of additional code, citing the need to improve the system’s reliability.

A new patch series for Bcachefs was recently submitted, encompassing both minor bug fixes and regression corrections, as well as a new feature dubbed journal_rewind. This capability enables the file system to revert to an earlier state, offering a mechanism for emergency recovery. Although the implementation remains incomplete and currently includes known limitations, its authors regard it as a significant advancement in enhancing the resilience of Bcachefs.

Linus Torvalds responded with sharp criticism. He reiterated that the merge window exists solely for the inclusion of new features and that it had long since closed. In his view, introducing functionality under the guise of bug fixes is unacceptable—especially for a system that is still classified as experimental. He emphasized that only genuine fixes should appear in release candidates.

Lead Bcachefs developer Kent Overstreet voiced his disagreement with this stance. He argued that most users simply want a stable and functional file system, particularly after negative experiences with alternatives like Btrfs and even XFS. In his opinion, user trust is paramount—and to earn that trust, he is willing to implement improvements swiftly, even if they are technically categorized as new features.

Kent further noted that file systems do not have the luxury of mitigating errors through simple reboots, as might be possible in other parts of the kernel. In some instances, he has had to introduce new data-handling mechanisms urgently to avert potential data loss. He underscored that the current patch in question spans merely 70 lines of code—code that, in his view, could safeguard user data—arguing that delaying its inclusion by three months would be unjustified.

He also highlighted his active engagement with the community, swift response to bug reports, and continuous enhancement of debugging tools. These efforts, he believes, are crucial for swiftly identifying and resolving issues, particularly when the code is being used in real-world, user-facing environments. Kent maintains that user support and feedback are every bit as vital as the code itself.

As of publication, Linus Torvalds has not yet decided whether to include the latest Bcachefs changes—journal_rewind among them—in the upcoming release candidate.