io-uring

Use the tokio-rs/io-uring crate: create an IoUring instance with a queue size, build operations with opcode constructors like opcode::Read, push them onto the submission queue, then call submit_and_wait and read results from the completion queue.

Call io_uring_queue_init with a queue depth and a pointer to struct io_uring. Get an SQE with io_uring_get_sqe, prepare it with functions like io_uring_prep_read, submit with io_uring_submit, and wait for results with io_uring_wait_cqe.

Basic read, write, and fsync operations require kernel 5.1. Accept, connect, send, and recv need 5.5, statx needs 5.7, and zero-copy send requires 5.19. Check your target kernel before using newer operations.

liburing is the reference C library maintained by io_uring's author and exposes the full API directly. The tokio-rs/io-uring crate provides the same low-level ring access for Rust with an entry builder pattern and raw fd types.

EAGAIN occurs when the submission queue is full or the completion queue overflows. Check available SQ slots before pushing entries, consume CQEs promptly with io_uring_cqe_seen, and size the ring appropriately for your workload.

Register buffers and files with io_uring_register_buffers and io_uring_register_files on hot paths to avoid per-operation mapping overhead. Use IOSQE_FIXED_FILE and fixed buffer operations like io_uring_prep_read_fixed to reference them.