# Ceph kernel client adds global lazy I/O mount option for Linux 7.4

> **Key takeaway:** A new 'lazyio' mount option allows Ceph users to enable buffered I/O via the page cache globally for all regular files.

**Published:** 2026-10-31T10:30:00+00:00 
**Source:** Phoronix Benchmarks & Drivers 
**Category:** kernel-os 
**Canonical URL:** https://fosswire.org/news/ceph-kernel-client-adds-global-lazy-io-mount-option-for-linu.html 

## Summary
The Ceph kernel client is receiving significant updates in the upcoming Linux 7.4 release. The primary addition is a new 'lazyio' mount option, which enables lazy I/O mode for all regular files at the time of opening. This feature allows for buffered I/O through the page cache even when multiple clients are writing to the same file.

Alongside this performance feature, the kernel is moving toward the removal of the long-deprecated inline data feature. To facilitate this transition, a new Kconfig option has been introduced to control inline data support. This change affects users of high-performance computing (HPC) workloads and those managing large-scale distributed storage via Ceph.

## Technical changes in the Ceph client
- **Lazy I/O Implementation:** The new 'lazyio' mount option provides a global alternative to the existing per-file ioctl, enabling buffered I/O via the page cache for all regular files.
- **Cache Coherency Trade-off:** The lazy I/O mode is specifically designed for workloads, such as HPC, that can tolerate relaxed cache coherency in exchange for increased throughput.
- **Inline Data Deprecation:** The experimental inline data feature, which stores small files directly within inode metadata, is being phased out. It is now controlled by the CEPH_FS_INLINE_DATA Kconfig option, which defaults to off.
- **Code Maintenance:** The Linux 7.4 pull request includes various bug fixes and general code clean-ups for the Ceph subsystem.

## Why it matters
This update provides HPC administrators with a more convenient way to optimize Ceph performance for specific distributed workloads. However, system engineers must now explicitly enable CEPH_FS_INLINE_DATA if they still rely on storing small files within metadata structures, as the feature will eventually be removed entirely.