Crate kevy_uring 

kevy-uring — pure-Rust io_uring bindings against the Linux kernel ABI.

A completion-based I/O engine. Where epoll/kqueue tell you when an fd is ready (then you do a read/write syscall each), io_uring lets you submit the reads/writes/accepts themselves into a shared submission queue (SQ) and later reap their results from a completion queue (CQ) — batching many operations into one io_uring_enter syscall, the lever toward the disk-I/O ceiling. Linux-only: on every other target this crate is an empty module that any caller can cfg-gate.

Hand-written against the kernel ABI — io_uring_setup/io_uring_enter/ io_uring_register are raw syscalls (no glibc wrappers, no liburing C dependency); the SQ/CQ/SQE regions are mmap’d and driven through the documented head/tail cursors. No libc crate, no third-party dependency.

Carved out of kevy-sys so the engine can be reused independently of kevy’s network internals. Part of the kevy key–value server.

Safety

The shared ring cursors are accessed as AtomicU32 over the mmap’d memory (the kernel is the other party): the producer publishes the SQ tail with Release and reads the SQ head with Acquire; the consumer reads the CQ tail with Acquire and publishes the CQ head with Release. IoUring owns its ring fd and three mappings, freed on drop.

Structs

Completion

One reaped completion (struct io_uring_cqe): the user_data you tagged the submission with, and res (bytes transferred / accepted fd when ≥ 0, else -errno).

IoUring

A Linux io_uring instance: one submission ring + one completion ring.

KernelTimespec

struct __kernel_timespec — the timeout payload an IORING_OP_TIMEOUT SQE points at (always 64-bit fields, independent of the C time_t width).

ProvidedBufRing

A registered provided-buffer ring (the destination pool for multishot recv). Owns the buf-ring mapping and the backing slab; the kernel fills a buffer per arrival, the app recycles it.