Btrfs Async Buffered Writes Double Throughput in Linux 6.1

The latest Btrfs async buffered‑write patch can deliver over twice the throughput of the previous implementation and is currently queued in the Btrfs for‑next Git repository, expected to be merged in Linux 6.1.

This patch series adds support for asynchronous buffered writes when using Btrfs together with io_uring. Previously, io_uring only supported buffered writes on the slow path for Btrfs; the new patches enable buffered writes on the fast path as well.

Patch series adds support for async buffered writes when using Btrfs and io_uring.

Meta (Facebook) engineer Stefan Roesch has been developing these patches, and benchmark results reported by Phoronix are impressive.

Performance testing shows that at an I/O depth of 1, the new patches more than double throughput compared to the existing behavior (where buffered writes are handled by an io‑worker process), while also significantly reducing latency. Achieving comparable performance with the old code would require an I/O depth of 4, and increasing depth further yields no additional gains.

Before the Linux 6.1 merge window opened in early October, a large collection of optimization patches already existed in Kdave’s for‑next Btrfs branch, with the async buffered‑write series being the primary feature.

For more information, see the kernel mailing list thread https://lore.kernel.org/linux-btrfs/[email protected]/ , the Phoronix article https://www.phoronix.com/news/Btrfs-Async-Buffered-Writes-6.1 , and the related git commit https://git.kernel.org/pub/scm/linux/kernel/git/kdave/linux.git/commit/?h=for-next&id=55121cf062ef609ce3ccc0f7b9f72e9df2f4a23c .