News › performance
By The Linux Camp News, October 26, 2026
The new buffered_iomap mount option enables an alternative EXT4 I/O path. Its maintainer reports gains for large writes, but feature exclusions and automatic fallback make the benefit workload-dependent.
EXT4's optional buffered iomap support has entered the Linux 7.4 development tree. A merge-tracking confirmation dated October 26, 2026 UTC records acceptance of the October 22 pull request targeting 7.4-rc1. This establishes development-tree inclusion, not availability in a stable Linux 7.4 release. 3
The alternative path replaces EXT4's buffer-head usage with iomap, an I/O framework, rather than replacing EXT4 itself. The patch list includes iomap implementations for buffered reads, buffered writes, writeback and memory-mapped I/O. The distinction matters because this is a change in how EXT4 handles those operations internally; the supplied descriptions do not establish precisely which reduced overheads account for the reported speedup. 1 2
Maintainer Theodore Ts'o reports significant performance improvements for large writes with buffered iomap enabled. That makes large-write workloads candidates for evaluation, not guaranteed beneficiaries. The supplied maintainer report gives no numerical results, hardware configuration or workload sizes, so it does not support a throughput estimate or an independently reproduced performance claim. 2
Compatibility is the main qualification. Ts'o lists online defragmentation, inline data, fscrypt, non-extent inodes and data=journal as unsupported by the new mode. His summary also names “fsverify”; the supplied material does not resolve that spelling. Operators evaluating the feature should check these dependencies first, because a faster path is not a substitute for required filesystem functionality. 2
The first configuration check is whether the buffered_iomap mount option is selected: merely using a kernel containing the change does not enable it. Even after opting in, EXT4 automatically falls back to its original I/O path when necessary. The operational inference is that enabling the option does not prove every operation uses iomap or gains performance. The summary does not enumerate fallback conditions, so evaluation must distinguish option selection from actual workload benefit. 2