- Support non-libc builds.
- Optimized syscall handling for x86-64/x86/aarch64.
- Enable non-lib function calls for fast path functions.
- Add support for multishot accept.
- io_uring_register_files() will set RLIMIT_NOFILE if necessary.
- Add support for registered ring fds, io_uring_register_ring_fd(),
reducing the overhead of an io_uring_enter() system call.
- Add support for the message ring opcode.
- Add support for newer request cancelation features.
- Add support for IORING_SETUP_COOP_TASKRUN, which can help reduce the
overhead of io_uring in general. Most applications should set this flag,
see the io_uring_setup.2 man page for details.
- Add support for registering a sparse buffer and file set.
- Add support for a new buffer provide scheme, see
io_uring_register_buf_ring.3 for details.
- Add io_uring_submit_and_wait_timeout() for submitting IO and waiting
for completions with a timeout.
- Add io_uring_prep_{read,write}v2 prep helpers.
- Add io_uring_prep_close_direct() helper.
- Add support for SQE128 and CQE32, which are doubly sized SQE and CQE
rings. This is needed for some cases of the new IORING_OP_URING_CMD,
notably for NVMe passthrough.
- ~5500 lines of man page additions, including adding ~90 new man pages.
- Synced with the 5.19 kernel release, supporting all the features of
5.19 and earlier.
- 24 new regression test cases, and ~7000 lines of new tests in general.
- General optimizations and fixes.