Linux 7.4 FUSE Large Folios Target Faster Buffered I/O

FUSE is preparing to enable large folios in the Linux 7.4 development cycle. The enabling change has reached the FUSE for-next branch, which makes it a candidate for the next kernel merge window rather than a feature already available in today's stable kernel.

A folio can represent multiple contiguous memory pages as one kernel memory-management unit. Letting FUSE use larger folios can reduce per-page overhead in buffered I/O paths, but the current enablement remains conditional rather than an unconditional switch for every FUSE workload.

The original enabling patch's fio testing reported roughly 15-20% higher write throughput across tested 256K, 1M and 5M block sizes. It also reported a much larger roughly 800% improvement for sequential reads after the page cache had already been populated.

That large cached-read number needs a strict boundary. The same patch reported no performance difference for the first sequential read that populated the cache, no difference for random reads and no difference for direct I/O. It therefore does not mean an NVMe drive, a FUSE mount or Linux storage in general becomes eight times faster.

Large-folio support in FUSE has been developed over multiple kernel cycles. Earlier work added the ability to operate on folios larger than one base page, while later iomap work addressed buffered reads, readahead, writes and writeback so partial-folio state can be tracked safely and efficiently.

That distinction matters because supporting a larger memory object and deciding when to enable it are separate steps. The current for-next change is the enablement milestone after that groundwork, not the beginning of FUSE large-folio development.

The change being present in FUSE for-next is strong upstream-development evidence, but it is not the same as a stable Linux 7.4 release. Code can still change before or during the merge window, and distribution availability will follow each distribution's own kernel cadence.

For users, the practical takeaway is narrower: FUSE is positioned to make better use of larger folios for buffered workloads in Linux 7.4, with promising workload-specific fio results and equally important cases where the submitted tests showed no gain.