To Migrate or Not to Migrate: KOxI, an Evidence-Gated Pipeline for Rust-for-Linux Storage Drivers
TL;DR
KOxI, an evidence-gated pipeline that turns four lenses into three explicit per-driver gates (safety, fuzzing, performance) so an R4L migration decision can be grounded in numbers rather than intuition, is proposed.
Abstract
Rust-for-Linux (R4L) is landing storage code in mainline, but per-driver migration decisions are still largely a judgment call: maintainers must weigh historical bug exposure, residual unsafe surface, dynamic robustness, and execution cost without a shared evidence pipeline. We propose KOxI, an evidence-gated pipeline that turns those four lenses into three explicit per-driver gates (safety, fuzzing, performance) so an R4L migration decision can be grounded in numbers rather than intuition. We exercise KOxI on null_blk and Hindborg’s upstream rnull, the matched C/Rust pair giving the smallest faithful end-to-end test of the pipeline. Phase 1 screening flags null_blk as a strong Rust candidate, and the gates then produce a discriminating profile: Rust’s type system retires 2 of 11 CWE-classified historical fix commits (18.2%), fuzzing shows no observed large target-attributable regression (Mann–Whitney p = 0.37, A12 = 0.45), and matched fio (Flexible I/O Tester) workloads regress on all 18 cases (median IOPS \(-14.4\%\), worst \(-25.5\%\)). The contribution is the pipeline; the lesson for R4L is that even on a deliberately small, well-matched candidate, the three gates disagree in informative ways — which is exactly the signal upstreaming should want before it scales.