Module uring 

io_uring batched small-file copy backend (Linux only).

This is the one place in ripsync-core where raw unsafe is justified: the io_uring submission interface requires pointing the kernel at user buffers. The rest of the crate keeps #![forbid(unsafe_code)]; here we opt in locally and annotate every unsafe block with a // SAFETY: rationale.

Strategy: copy a chunk of files by (1) opening each source and temp file, (2) submitting one read per file and reaping all completions, then (3) submitting one write per file and reaping. Two ring round-trips replace 2·N read/write syscalls. Files larger than MAX_FILE but at most MAX_LARGE_FILE are handled via linked [Splice] SQEs when the kernel supports it (≥5.19). Anything larger or unsupported is reported back so the caller can fall back to the portable path.

Structs

Job

One copy request: read all of src, write it to the fresh temp file tmp.

Constants

MAX_FILE

Largest file handled by the single-shot uring path; bigger files fall back or use the splice path.

MAX_LARGE_FILE

Largest file handled by the linked-splice uring path (64 MiB).

Functions

copy_batch

Copy every job, returning a per-job result (bytes written, or the error that should trigger a portable-path fallback for that single file).

copy_single_large

Copy a single large file via linked splice SQEs.

kernel_supports_splice

Check whether the running kernel supports IORING_OP_SPLICE (≥5.19).

large_file_stats

Return (attempts, successes) counters for large-file splice copies.