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;
49 __u32 uring_cmd_flags;
51 __u64 user_data; /* data to be passed back at completion time */
52 /* pack this to avoid bogus arm OABI complaints */
54 /* index into fixed buffers, if used */
56 /* for grouped buffer selection */
58 } __attribute__((packed));
59 /* personality to use, if used */
71 * If the ring is initialized with IORING_SETUP_SQE128, then
72 * this field is used for 80 bytes of arbitrary command data
82 IOSQE_IO_HARDLINK_BIT,
84 IOSQE_BUFFER_SELECT_BIT,
85 IOSQE_CQE_SKIP_SUCCESS_BIT,
91 /* use fixed fileset */
92 #define IOSQE_FIXED_FILE (1U << IOSQE_FIXED_FILE_BIT)
93 /* issue after inflight IO */
94 #define IOSQE_IO_DRAIN (1U << IOSQE_IO_DRAIN_BIT)
96 #define IOSQE_IO_LINK (1U << IOSQE_IO_LINK_BIT)
97 /* like LINK, but stronger */
98 #define IOSQE_IO_HARDLINK (1U << IOSQE_IO_HARDLINK_BIT)
100 #define IOSQE_ASYNC (1U << IOSQE_ASYNC_BIT)
101 /* select buffer from sqe->buf_group */
102 #define IOSQE_BUFFER_SELECT (1U << IOSQE_BUFFER_SELECT_BIT)
103 /* don't post CQE if request succeeded */
104 #define IOSQE_CQE_SKIP_SUCCESS (1U << IOSQE_CQE_SKIP_SUCCESS_BIT)
107 * io_uring_setup() flags
109 #define IORING_SETUP_IOPOLL (1U << 0) /* io_context is polled */
110 #define IORING_SETUP_SQPOLL (1U << 1) /* SQ poll thread */
111 #define IORING_SETUP_SQ_AFF (1U << 2) /* sq_thread_cpu is valid */
112 #define IORING_SETUP_CQSIZE (1U << 3) /* app defines CQ size */
113 #define IORING_SETUP_CLAMP (1U << 4) /* clamp SQ/CQ ring sizes */
114 #define IORING_SETUP_ATTACH_WQ (1U << 5) /* attach to existing wq */
115 #define IORING_SETUP_R_DISABLED (1U << 6) /* start with ring disabled */
116 #define IORING_SETUP_SUBMIT_ALL (1U << 7) /* continue submit on error */
118 * Cooperative task running. When requests complete, they often require
119 * forcing the submitter to transition to the kernel to complete. If this
120 * flag is set, work will be done when the task transitions anyway, rather
121 * than force an inter-processor interrupt reschedule. This avoids interrupting
122 * a task running in userspace, and saves an IPI.
124 #define IORING_SETUP_COOP_TASKRUN (1U << 8)
126 * If COOP_TASKRUN is set, get notified if task work is available for
127 * running and a kernel transition would be needed to run it. This sets
128 * IORING_SQ_TASKRUN in the sq ring flags. Not valid with COOP_TASKRUN.
130 #define IORING_SETUP_TASKRUN_FLAG (1U << 9)
132 #define IORING_SETUP_SQE128 (1U << 10) /* SQEs are 128 byte */
133 #define IORING_SETUP_CQE32 (1U << 11) /* CQEs are 32 byte */
136 * Only one task is allowed to submit requests
138 #define IORING_SETUP_SINGLE_ISSUER (1U << 12)
141 * Defer running task work to get events.
142 * Rather than running bits of task work whenever the task transitions
143 * try to do it just before it is needed.
145 #define IORING_SETUP_DEFER_TASKRUN (1U << 13)
152 IORING_OP_READ_FIXED,
153 IORING_OP_WRITE_FIXED,
155 IORING_OP_POLL_REMOVE,
156 IORING_OP_SYNC_FILE_RANGE,
160 IORING_OP_TIMEOUT_REMOVE,
162 IORING_OP_ASYNC_CANCEL,
163 IORING_OP_LINK_TIMEOUT,
168 IORING_OP_FILES_UPDATE,
179 IORING_OP_PROVIDE_BUFFERS,
180 IORING_OP_REMOVE_BUFFERS,
197 /* this goes last, obviously */
202 * sqe->uring_cmd_flags
203 * IORING_URING_CMD_FIXED use registered buffer; pass thig flag
204 * along with setting sqe->buf_index.
206 #define IORING_URING_CMD_FIXED (1U << 0)
211 #define IORING_FSYNC_DATASYNC (1U << 0)
216 #define IORING_TIMEOUT_ABS (1U << 0)
217 #define IORING_TIMEOUT_UPDATE (1U << 1)
218 #define IORING_TIMEOUT_BOOTTIME (1U << 2)
219 #define IORING_TIMEOUT_REALTIME (1U << 3)
220 #define IORING_LINK_TIMEOUT_UPDATE (1U << 4)
221 #define IORING_TIMEOUT_ETIME_SUCCESS (1U << 5)
222 #define IORING_TIMEOUT_CLOCK_MASK (IORING_TIMEOUT_BOOTTIME | IORING_TIMEOUT_REALTIME)
223 #define IORING_TIMEOUT_UPDATE_MASK (IORING_TIMEOUT_UPDATE | IORING_LINK_TIMEOUT_UPDATE)
226 * extends splice(2) flags
228 #define SPLICE_F_FD_IN_FIXED (1U << 31) /* the last bit of __u32 */
231 * POLL_ADD flags. Note that since sqe->poll_events is the flag space, the
232 * command flags for POLL_ADD are stored in sqe->len.
234 * IORING_POLL_ADD_MULTI Multishot poll. Sets IORING_CQE_F_MORE if
235 * the poll handler will continue to report
236 * CQEs on behalf of the same SQE.
238 * IORING_POLL_UPDATE Update existing poll request, matching
239 * sqe->addr as the old user_data field.
241 #define IORING_POLL_ADD_MULTI (1U << 0)
242 #define IORING_POLL_UPDATE_EVENTS (1U << 1)
243 #define IORING_POLL_UPDATE_USER_DATA (1U << 2)
246 * IO completion data structure (Completion Queue Entry)
248 struct io_uring_cqe {
249 __u64 user_data; /* sqe->data submission passed back */
250 __s32 res; /* result code for this event */
254 * If the ring is initialized with IORING_SETUP_CQE32, then this field
255 * contains 16-bytes of padding, doubling the size of the CQE.
263 * IORING_CQE_F_BUFFER If set, the upper 16 bits are the buffer ID
264 * IORING_CQE_F_MORE If set, parent SQE will generate more CQE entries
266 #define IORING_CQE_F_BUFFER (1U << 0)
267 #define IORING_CQE_F_MORE (1U << 1)
270 IORING_CQE_BUFFER_SHIFT = 16,
274 * Magic offsets for the application to mmap the data it needs
276 #define IORING_OFF_SQ_RING 0ULL
277 #define IORING_OFF_CQ_RING 0x8000000ULL
278 #define IORING_OFF_SQES 0x10000000ULL
281 * Filled with the offset for mmap(2)
283 struct io_sqring_offsets {
298 #define IORING_SQ_NEED_WAKEUP (1U << 0) /* needs io_uring_enter wakeup */
299 #define IORING_SQ_CQ_OVERFLOW (1U << 1) /* CQ ring is overflown */
301 struct io_cqring_offsets {
317 /* disable eventfd notifications */
318 #define IORING_CQ_EVENTFD_DISABLED (1U << 0)
321 * io_uring_enter(2) flags
323 #define IORING_ENTER_GETEVENTS (1U << 0)
324 #define IORING_ENTER_SQ_WAKEUP (1U << 1)
325 #define IORING_ENTER_SQ_WAIT (1U << 2)
326 #define IORING_ENTER_EXT_ARG (1U << 3)
327 #define IORING_ENTER_REGISTERED_RING (1U << 4)
330 * Passed in for io_uring_setup(2). Copied back with updated info on success
332 struct io_uring_params {
337 __u32 sq_thread_idle;
341 struct io_sqring_offsets sq_off;
342 struct io_cqring_offsets cq_off;
346 * io_uring_params->features flags
348 #define IORING_FEAT_SINGLE_MMAP (1U << 0)
349 #define IORING_FEAT_NODROP (1U << 1)
350 #define IORING_FEAT_SUBMIT_STABLE (1U << 2)
351 #define IORING_FEAT_RW_CUR_POS (1U << 3)
352 #define IORING_FEAT_CUR_PERSONALITY (1U << 4)
353 #define IORING_FEAT_FAST_POLL (1U << 5)
354 #define IORING_FEAT_POLL_32BITS (1U << 6)
355 #define IORING_FEAT_SQPOLL_NONFIXED (1U << 7)
356 #define IORING_FEAT_EXT_ARG (1U << 8)
357 #define IORING_FEAT_NATIVE_WORKERS (1U << 9)
358 #define IORING_FEAT_RSRC_TAGS (1U << 10)
359 #define IORING_FEAT_CQE_SKIP (1U << 11)
362 * io_uring_register(2) opcodes and arguments
365 IORING_REGISTER_BUFFERS = 0,
366 IORING_UNREGISTER_BUFFERS = 1,
367 IORING_REGISTER_FILES = 2,
368 IORING_UNREGISTER_FILES = 3,
369 IORING_REGISTER_EVENTFD = 4,
370 IORING_UNREGISTER_EVENTFD = 5,
371 IORING_REGISTER_FILES_UPDATE = 6,
372 IORING_REGISTER_EVENTFD_ASYNC = 7,
373 IORING_REGISTER_PROBE = 8,
374 IORING_REGISTER_PERSONALITY = 9,
375 IORING_UNREGISTER_PERSONALITY = 10,
376 IORING_REGISTER_RESTRICTIONS = 11,
377 IORING_REGISTER_ENABLE_RINGS = 12,
379 /* extended with tagging */
380 IORING_REGISTER_FILES2 = 13,
381 IORING_REGISTER_FILES_UPDATE2 = 14,
382 IORING_REGISTER_BUFFERS2 = 15,
383 IORING_REGISTER_BUFFERS_UPDATE = 16,
385 /* set/clear io-wq thread affinities */
386 IORING_REGISTER_IOWQ_AFF = 17,
387 IORING_UNREGISTER_IOWQ_AFF = 18,
389 /* set/get max number of io-wq workers */
390 IORING_REGISTER_IOWQ_MAX_WORKERS = 19,
392 /* register/unregister io_uring fd with the ring */
393 IORING_REGISTER_RING_FDS = 20,
394 IORING_UNREGISTER_RING_FDS = 21,
400 /* io-wq worker categories */
406 /* deprecated, see struct io_uring_rsrc_update */
407 struct io_uring_files_update {
410 __aligned_u64 /* __s32 * */ fds;
413 struct io_uring_rsrc_register {
421 struct io_uring_rsrc_update {
427 struct io_uring_rsrc_update2 {
436 /* Skip updating fd indexes set to this value in the fd table */
437 #define IORING_REGISTER_FILES_SKIP (-2)
439 #define IO_URING_OP_SUPPORTED (1U << 0)
441 struct io_uring_probe_op {
444 __u16 flags; /* IO_URING_OP_* flags */
448 struct io_uring_probe {
449 __u8 last_op; /* last opcode supported */
450 __u8 ops_len; /* length of ops[] array below */
453 struct io_uring_probe_op ops[0];
456 struct io_uring_restriction {
459 __u8 register_op; /* IORING_RESTRICTION_REGISTER_OP */
460 __u8 sqe_op; /* IORING_RESTRICTION_SQE_OP */
461 __u8 sqe_flags; /* IORING_RESTRICTION_SQE_FLAGS_* */
468 * io_uring_restriction->opcode values
471 /* Allow an io_uring_register(2) opcode */
472 IORING_RESTRICTION_REGISTER_OP = 0,
474 /* Allow an sqe opcode */
475 IORING_RESTRICTION_SQE_OP = 1,
477 /* Allow sqe flags */
478 IORING_RESTRICTION_SQE_FLAGS_ALLOWED = 2,
480 /* Require sqe flags (these flags must be set on each submission) */
481 IORING_RESTRICTION_SQE_FLAGS_REQUIRED = 3,
483 IORING_RESTRICTION_LAST
486 struct io_uring_getevents_arg {