1 /* SPDX-License-Identifier: (GPL-2.0 WITH Linux-syscall-note) OR MIT */
3 * Header file for the io_uring interface.
5 * Copyright (C) 2019 Jens Axboe
6 * Copyright (C) 2019 Christoph Hellwig
8 #ifndef LINUX_IO_URING_H
9 #define LINUX_IO_URING_H
12 #include <linux/types.h>
15 * IO submission data structure (Submission Queue Entry)
18 __u8 opcode; /* type of operation for this sqe */
19 __u8 flags; /* IOSQE_ flags */
20 __u16 ioprio; /* ioprio for the request */
21 __s32 fd; /* file descriptor to do IO on */
23 __u64 off; /* offset into file */
28 __u64 addr; /* pointer to buffer or iovecs */
31 __u32 len; /* buffer size or number of iovecs */
33 __kernel_rwf_t rw_flags;
35 __u16 poll_events; /* compatibility */
36 __u32 poll32_events; /* word-reversed for BE */
37 __u32 sync_range_flags;
50 __u64 user_data; /* data to be passed back at completion time */
51 /* pack this to avoid bogus arm OABI complaints */
53 /* index into fixed buffers, if used */
55 /* for grouped buffer selection */
57 } __attribute__((packed));
58 /* personality to use, if used */
70 * If the ring is initialized with IORING_SETUP_SQE128, then
71 * this field is used for 80 bytes of arbitrary command data
81 IOSQE_IO_HARDLINK_BIT,
83 IOSQE_BUFFER_SELECT_BIT,
84 IOSQE_CQE_SKIP_SUCCESS_BIT,
90 /* use fixed fileset */
91 #define IOSQE_FIXED_FILE (1U << IOSQE_FIXED_FILE_BIT)
92 /* issue after inflight IO */
93 #define IOSQE_IO_DRAIN (1U << IOSQE_IO_DRAIN_BIT)
95 #define IOSQE_IO_LINK (1U << IOSQE_IO_LINK_BIT)
96 /* like LINK, but stronger */
97 #define IOSQE_IO_HARDLINK (1U << IOSQE_IO_HARDLINK_BIT)
99 #define IOSQE_ASYNC (1U << IOSQE_ASYNC_BIT)
100 /* select buffer from sqe->buf_group */
101 #define IOSQE_BUFFER_SELECT (1U << IOSQE_BUFFER_SELECT_BIT)
102 /* don't post CQE if request succeeded */
103 #define IOSQE_CQE_SKIP_SUCCESS (1U << IOSQE_CQE_SKIP_SUCCESS_BIT)
106 * io_uring_setup() flags
108 #define IORING_SETUP_IOPOLL (1U << 0) /* io_context is polled */
109 #define IORING_SETUP_SQPOLL (1U << 1) /* SQ poll thread */
110 #define IORING_SETUP_SQ_AFF (1U << 2) /* sq_thread_cpu is valid */
111 #define IORING_SETUP_CQSIZE (1U << 3) /* app defines CQ size */
112 #define IORING_SETUP_CLAMP (1U << 4) /* clamp SQ/CQ ring sizes */
113 #define IORING_SETUP_ATTACH_WQ (1U << 5) /* attach to existing wq */
114 #define IORING_SETUP_R_DISABLED (1U << 6) /* start with ring disabled */
115 #define IORING_SETUP_SUBMIT_ALL (1U << 7) /* continue submit on error */
117 * Cooperative task running. When requests complete, they often require
118 * forcing the submitter to transition to the kernel to complete. If this
119 * flag is set, work will be done when the task transitions anyway, rather
120 * than force an inter-processor interrupt reschedule. This avoids interrupting
121 * a task running in userspace, and saves an IPI.
123 #define IORING_SETUP_COOP_TASKRUN (1U << 8)
125 * If COOP_TASKRUN is set, get notified if task work is available for
126 * running and a kernel transition would be needed to run it. This sets
127 * IORING_SQ_TASKRUN in the sq ring flags. Not valid with COOP_TASKRUN.
129 #define IORING_SETUP_TASKRUN_FLAG (1U << 9)
131 #define IORING_SETUP_SQE128 (1U << 10) /* SQEs are 128 byte */
132 #define IORING_SETUP_CQE32 (1U << 11) /* CQEs are 32 byte */
139 IORING_OP_READ_FIXED,
140 IORING_OP_WRITE_FIXED,
142 IORING_OP_POLL_REMOVE,
143 IORING_OP_SYNC_FILE_RANGE,
147 IORING_OP_TIMEOUT_REMOVE,
149 IORING_OP_ASYNC_CANCEL,
150 IORING_OP_LINK_TIMEOUT,
155 IORING_OP_FILES_UPDATE,
166 IORING_OP_PROVIDE_BUFFERS,
167 IORING_OP_REMOVE_BUFFERS,
184 /* this goes last, obviously */
191 #define IORING_FSYNC_DATASYNC (1U << 0)
196 #define IORING_TIMEOUT_ABS (1U << 0)
197 #define IORING_TIMEOUT_UPDATE (1U << 1)
198 #define IORING_TIMEOUT_BOOTTIME (1U << 2)
199 #define IORING_TIMEOUT_REALTIME (1U << 3)
200 #define IORING_LINK_TIMEOUT_UPDATE (1U << 4)
201 #define IORING_TIMEOUT_ETIME_SUCCESS (1U << 5)
202 #define IORING_TIMEOUT_CLOCK_MASK (IORING_TIMEOUT_BOOTTIME | IORING_TIMEOUT_REALTIME)
203 #define IORING_TIMEOUT_UPDATE_MASK (IORING_TIMEOUT_UPDATE | IORING_LINK_TIMEOUT_UPDATE)
206 * extends splice(2) flags
208 #define SPLICE_F_FD_IN_FIXED (1U << 31) /* the last bit of __u32 */
211 * POLL_ADD flags. Note that since sqe->poll_events is the flag space, the
212 * command flags for POLL_ADD are stored in sqe->len.
214 * IORING_POLL_ADD_MULTI Multishot poll. Sets IORING_CQE_F_MORE if
215 * the poll handler will continue to report
216 * CQEs on behalf of the same SQE.
218 * IORING_POLL_UPDATE Update existing poll request, matching
219 * sqe->addr as the old user_data field.
221 #define IORING_POLL_ADD_MULTI (1U << 0)
222 #define IORING_POLL_UPDATE_EVENTS (1U << 1)
223 #define IORING_POLL_UPDATE_USER_DATA (1U << 2)
226 * IO completion data structure (Completion Queue Entry)
228 struct io_uring_cqe {
229 __u64 user_data; /* sqe->data submission passed back */
230 __s32 res; /* result code for this event */
234 * If the ring is initialized with IORING_SETUP_CQE32, then this field
235 * contains 16-bytes of padding, doubling the size of the CQE.
243 * IORING_CQE_F_BUFFER If set, the upper 16 bits are the buffer ID
244 * IORING_CQE_F_MORE If set, parent SQE will generate more CQE entries
246 #define IORING_CQE_F_BUFFER (1U << 0)
247 #define IORING_CQE_F_MORE (1U << 1)
250 IORING_CQE_BUFFER_SHIFT = 16,
254 * Magic offsets for the application to mmap the data it needs
256 #define IORING_OFF_SQ_RING 0ULL
257 #define IORING_OFF_CQ_RING 0x8000000ULL
258 #define IORING_OFF_SQES 0x10000000ULL
261 * Filled with the offset for mmap(2)
263 struct io_sqring_offsets {
278 #define IORING_SQ_NEED_WAKEUP (1U << 0) /* needs io_uring_enter wakeup */
279 #define IORING_SQ_CQ_OVERFLOW (1U << 1) /* CQ ring is overflown */
281 struct io_cqring_offsets {
297 /* disable eventfd notifications */
298 #define IORING_CQ_EVENTFD_DISABLED (1U << 0)
301 * io_uring_enter(2) flags
303 #define IORING_ENTER_GETEVENTS (1U << 0)
304 #define IORING_ENTER_SQ_WAKEUP (1U << 1)
305 #define IORING_ENTER_SQ_WAIT (1U << 2)
306 #define IORING_ENTER_EXT_ARG (1U << 3)
307 #define IORING_ENTER_REGISTERED_RING (1U << 4)
310 * Passed in for io_uring_setup(2). Copied back with updated info on success
312 struct io_uring_params {
317 __u32 sq_thread_idle;
321 struct io_sqring_offsets sq_off;
322 struct io_cqring_offsets cq_off;
326 * io_uring_params->features flags
328 #define IORING_FEAT_SINGLE_MMAP (1U << 0)
329 #define IORING_FEAT_NODROP (1U << 1)
330 #define IORING_FEAT_SUBMIT_STABLE (1U << 2)
331 #define IORING_FEAT_RW_CUR_POS (1U << 3)
332 #define IORING_FEAT_CUR_PERSONALITY (1U << 4)
333 #define IORING_FEAT_FAST_POLL (1U << 5)
334 #define IORING_FEAT_POLL_32BITS (1U << 6)
335 #define IORING_FEAT_SQPOLL_NONFIXED (1U << 7)
336 #define IORING_FEAT_EXT_ARG (1U << 8)
337 #define IORING_FEAT_NATIVE_WORKERS (1U << 9)
338 #define IORING_FEAT_RSRC_TAGS (1U << 10)
339 #define IORING_FEAT_CQE_SKIP (1U << 11)
342 * io_uring_register(2) opcodes and arguments
345 IORING_REGISTER_BUFFERS = 0,
346 IORING_UNREGISTER_BUFFERS = 1,
347 IORING_REGISTER_FILES = 2,
348 IORING_UNREGISTER_FILES = 3,
349 IORING_REGISTER_EVENTFD = 4,
350 IORING_UNREGISTER_EVENTFD = 5,
351 IORING_REGISTER_FILES_UPDATE = 6,
352 IORING_REGISTER_EVENTFD_ASYNC = 7,
353 IORING_REGISTER_PROBE = 8,
354 IORING_REGISTER_PERSONALITY = 9,
355 IORING_UNREGISTER_PERSONALITY = 10,
356 IORING_REGISTER_RESTRICTIONS = 11,
357 IORING_REGISTER_ENABLE_RINGS = 12,
359 /* extended with tagging */
360 IORING_REGISTER_FILES2 = 13,
361 IORING_REGISTER_FILES_UPDATE2 = 14,
362 IORING_REGISTER_BUFFERS2 = 15,
363 IORING_REGISTER_BUFFERS_UPDATE = 16,
365 /* set/clear io-wq thread affinities */
366 IORING_REGISTER_IOWQ_AFF = 17,
367 IORING_UNREGISTER_IOWQ_AFF = 18,
369 /* set/get max number of io-wq workers */
370 IORING_REGISTER_IOWQ_MAX_WORKERS = 19,
372 /* register/unregister io_uring fd with the ring */
373 IORING_REGISTER_RING_FDS = 20,
374 IORING_UNREGISTER_RING_FDS = 21,
380 /* io-wq worker categories */
386 /* deprecated, see struct io_uring_rsrc_update */
387 struct io_uring_files_update {
390 __aligned_u64 /* __s32 * */ fds;
393 struct io_uring_rsrc_register {
401 struct io_uring_rsrc_update {
407 struct io_uring_rsrc_update2 {
416 /* Skip updating fd indexes set to this value in the fd table */
417 #define IORING_REGISTER_FILES_SKIP (-2)
419 #define IO_URING_OP_SUPPORTED (1U << 0)
421 struct io_uring_probe_op {
424 __u16 flags; /* IO_URING_OP_* flags */
428 struct io_uring_probe {
429 __u8 last_op; /* last opcode supported */
430 __u8 ops_len; /* length of ops[] array below */
433 struct io_uring_probe_op ops[0];
436 struct io_uring_restriction {
439 __u8 register_op; /* IORING_RESTRICTION_REGISTER_OP */
440 __u8 sqe_op; /* IORING_RESTRICTION_SQE_OP */
441 __u8 sqe_flags; /* IORING_RESTRICTION_SQE_FLAGS_* */
448 * io_uring_restriction->opcode values
451 /* Allow an io_uring_register(2) opcode */
452 IORING_RESTRICTION_REGISTER_OP = 0,
454 /* Allow an sqe opcode */
455 IORING_RESTRICTION_SQE_OP = 1,
457 /* Allow sqe flags */
458 IORING_RESTRICTION_SQE_FLAGS_ALLOWED = 2,
460 /* Require sqe flags (these flags must be set on each submission) */
461 IORING_RESTRICTION_SQE_FLAGS_REQUIRED = 3,
463 IORING_RESTRICTION_LAST
466 struct io_uring_getevents_arg {