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 */
27 __u64 addr; /* pointer to buffer or iovecs */
30 __u32 len; /* buffer size or number of iovecs */
32 __kernel_rwf_t rw_flags;
34 __u16 poll_events; /* compatibility */
35 __u32 poll32_events; /* word-reversed for BE */
36 __u32 sync_range_flags;
49 __u64 user_data; /* data to be passed back at completion time */
50 /* pack this to avoid bogus arm OABI complaints */
52 /* index into fixed buffers, if used */
54 /* for grouped buffer selection */
56 } __attribute__((packed));
57 /* personality to use, if used */
70 IOSQE_IO_HARDLINK_BIT,
72 IOSQE_BUFFER_SELECT_BIT,
73 IOSQE_CQE_SKIP_SUCCESS_BIT,
79 /* use fixed fileset */
80 #define IOSQE_FIXED_FILE (1U << IOSQE_FIXED_FILE_BIT)
81 /* issue after inflight IO */
82 #define IOSQE_IO_DRAIN (1U << IOSQE_IO_DRAIN_BIT)
84 #define IOSQE_IO_LINK (1U << IOSQE_IO_LINK_BIT)
85 /* like LINK, but stronger */
86 #define IOSQE_IO_HARDLINK (1U << IOSQE_IO_HARDLINK_BIT)
88 #define IOSQE_ASYNC (1U << IOSQE_ASYNC_BIT)
89 /* select buffer from sqe->buf_group */
90 #define IOSQE_BUFFER_SELECT (1U << IOSQE_BUFFER_SELECT_BIT)
91 /* don't post CQE if request succeeded */
92 #define IOSQE_CQE_SKIP_SUCCESS (1U << IOSQE_CQE_SKIP_SUCCESS_BIT)
95 * io_uring_setup() flags
97 #define IORING_SETUP_IOPOLL (1U << 0) /* io_context is polled */
98 #define IORING_SETUP_SQPOLL (1U << 1) /* SQ poll thread */
99 #define IORING_SETUP_SQ_AFF (1U << 2) /* sq_thread_cpu is valid */
100 #define IORING_SETUP_CQSIZE (1U << 3) /* app defines CQ size */
101 #define IORING_SETUP_CLAMP (1U << 4) /* clamp SQ/CQ ring sizes */
102 #define IORING_SETUP_ATTACH_WQ (1U << 5) /* attach to existing wq */
103 #define IORING_SETUP_R_DISABLED (1U << 6) /* start with ring disabled */
110 IORING_OP_READ_FIXED,
111 IORING_OP_WRITE_FIXED,
113 IORING_OP_POLL_REMOVE,
114 IORING_OP_SYNC_FILE_RANGE,
118 IORING_OP_TIMEOUT_REMOVE,
120 IORING_OP_ASYNC_CANCEL,
121 IORING_OP_LINK_TIMEOUT,
126 IORING_OP_FILES_UPDATE,
137 IORING_OP_PROVIDE_BUFFERS,
138 IORING_OP_REMOVE_BUFFERS,
147 /* this goes last, obviously */
154 #define IORING_FSYNC_DATASYNC (1U << 0)
159 #define IORING_TIMEOUT_ABS (1U << 0)
160 #define IORING_TIMEOUT_UPDATE (1U << 1)
161 #define IORING_TIMEOUT_BOOTTIME (1U << 2)
162 #define IORING_TIMEOUT_REALTIME (1U << 3)
163 #define IORING_LINK_TIMEOUT_UPDATE (1U << 4)
164 #define IORING_TIMEOUT_ETIME_SUCCESS (1U << 5)
165 #define IORING_TIMEOUT_CLOCK_MASK (IORING_TIMEOUT_BOOTTIME | IORING_TIMEOUT_REALTIME)
166 #define IORING_TIMEOUT_UPDATE_MASK (IORING_TIMEOUT_UPDATE | IORING_LINK_TIMEOUT_UPDATE)
169 * extends splice(2) flags
171 #define SPLICE_F_FD_IN_FIXED (1U << 31) /* the last bit of __u32 */
174 * POLL_ADD flags. Note that since sqe->poll_events is the flag space, the
175 * command flags for POLL_ADD are stored in sqe->len.
177 * IORING_POLL_ADD_MULTI Multishot poll. Sets IORING_CQE_F_MORE if
178 * the poll handler will continue to report
179 * CQEs on behalf of the same SQE.
181 * IORING_POLL_UPDATE Update existing poll request, matching
182 * sqe->addr as the old user_data field.
184 #define IORING_POLL_ADD_MULTI (1U << 0)
185 #define IORING_POLL_UPDATE_EVENTS (1U << 1)
186 #define IORING_POLL_UPDATE_USER_DATA (1U << 2)
189 * IO completion data structure (Completion Queue Entry)
191 struct io_uring_cqe {
192 __u64 user_data; /* sqe->data submission passed back */
193 __s32 res; /* result code for this event */
200 * IORING_CQE_F_BUFFER If set, the upper 16 bits are the buffer ID
201 * IORING_CQE_F_MORE If set, parent SQE will generate more CQE entries
203 #define IORING_CQE_F_BUFFER (1U << 0)
204 #define IORING_CQE_F_MORE (1U << 1)
207 IORING_CQE_BUFFER_SHIFT = 16,
211 * Magic offsets for the application to mmap the data it needs
213 #define IORING_OFF_SQ_RING 0ULL
214 #define IORING_OFF_CQ_RING 0x8000000ULL
215 #define IORING_OFF_SQES 0x10000000ULL
218 * Filled with the offset for mmap(2)
220 struct io_sqring_offsets {
235 #define IORING_SQ_NEED_WAKEUP (1U << 0) /* needs io_uring_enter wakeup */
236 #define IORING_SQ_CQ_OVERFLOW (1U << 1) /* CQ ring is overflown */
238 struct io_cqring_offsets {
254 /* disable eventfd notifications */
255 #define IORING_CQ_EVENTFD_DISABLED (1U << 0)
258 * io_uring_enter(2) flags
260 #define IORING_ENTER_GETEVENTS (1U << 0)
261 #define IORING_ENTER_SQ_WAKEUP (1U << 1)
262 #define IORING_ENTER_SQ_WAIT (1U << 2)
263 #define IORING_ENTER_EXT_ARG (1U << 3)
264 #define IORING_ENTER_REGISTERED_RING (1U << 4)
267 * Passed in for io_uring_setup(2). Copied back with updated info on success
269 struct io_uring_params {
274 __u32 sq_thread_idle;
278 struct io_sqring_offsets sq_off;
279 struct io_cqring_offsets cq_off;
283 * io_uring_params->features flags
285 #define IORING_FEAT_SINGLE_MMAP (1U << 0)
286 #define IORING_FEAT_NODROP (1U << 1)
287 #define IORING_FEAT_SUBMIT_STABLE (1U << 2)
288 #define IORING_FEAT_RW_CUR_POS (1U << 3)
289 #define IORING_FEAT_CUR_PERSONALITY (1U << 4)
290 #define IORING_FEAT_FAST_POLL (1U << 5)
291 #define IORING_FEAT_POLL_32BITS (1U << 6)
292 #define IORING_FEAT_SQPOLL_NONFIXED (1U << 7)
293 #define IORING_FEAT_EXT_ARG (1U << 8)
294 #define IORING_FEAT_NATIVE_WORKERS (1U << 9)
295 #define IORING_FEAT_RSRC_TAGS (1U << 10)
296 #define IORING_FEAT_CQE_SKIP (1U << 11)
299 * io_uring_register(2) opcodes and arguments
302 IORING_REGISTER_BUFFERS = 0,
303 IORING_UNREGISTER_BUFFERS = 1,
304 IORING_REGISTER_FILES = 2,
305 IORING_UNREGISTER_FILES = 3,
306 IORING_REGISTER_EVENTFD = 4,
307 IORING_UNREGISTER_EVENTFD = 5,
308 IORING_REGISTER_FILES_UPDATE = 6,
309 IORING_REGISTER_EVENTFD_ASYNC = 7,
310 IORING_REGISTER_PROBE = 8,
311 IORING_REGISTER_PERSONALITY = 9,
312 IORING_UNREGISTER_PERSONALITY = 10,
313 IORING_REGISTER_RESTRICTIONS = 11,
314 IORING_REGISTER_ENABLE_RINGS = 12,
316 /* extended with tagging */
317 IORING_REGISTER_FILES2 = 13,
318 IORING_REGISTER_FILES_UPDATE2 = 14,
319 IORING_REGISTER_BUFFERS2 = 15,
320 IORING_REGISTER_BUFFERS_UPDATE = 16,
322 /* set/clear io-wq thread affinities */
323 IORING_REGISTER_IOWQ_AFF = 17,
324 IORING_UNREGISTER_IOWQ_AFF = 18,
326 /* set/get max number of io-wq workers */
327 IORING_REGISTER_IOWQ_MAX_WORKERS = 19,
329 /* register/unregister io_uring fd with the ring */
330 IORING_REGISTER_RING_FDS = 20,
331 IORING_UNREGISTER_RING_FDS = 21,
337 /* io-wq worker categories */
343 /* deprecated, see struct io_uring_rsrc_update */
344 struct io_uring_files_update {
347 __aligned_u64 /* __s32 * */ fds;
350 struct io_uring_rsrc_register {
358 struct io_uring_rsrc_update {
364 struct io_uring_rsrc_update2 {
373 /* Skip updating fd indexes set to this value in the fd table */
374 #define IORING_REGISTER_FILES_SKIP (-2)
376 #define IO_URING_OP_SUPPORTED (1U << 0)
378 struct io_uring_probe_op {
381 __u16 flags; /* IO_URING_OP_* flags */
385 struct io_uring_probe {
386 __u8 last_op; /* last opcode supported */
387 __u8 ops_len; /* length of ops[] array below */
390 struct io_uring_probe_op ops[0];
393 struct io_uring_restriction {
396 __u8 register_op; /* IORING_RESTRICTION_REGISTER_OP */
397 __u8 sqe_op; /* IORING_RESTRICTION_SQE_OP */
398 __u8 sqe_flags; /* IORING_RESTRICTION_SQE_FLAGS_* */
405 * io_uring_restriction->opcode values
408 /* Allow an io_uring_register(2) opcode */
409 IORING_RESTRICTION_REGISTER_OP = 0,
411 /* Allow an sqe opcode */
412 IORING_RESTRICTION_SQE_OP = 1,
414 /* Allow sqe flags */
415 IORING_RESTRICTION_SQE_FLAGS_ALLOWED = 2,
417 /* Require sqe flags (these flags must be set on each submission) */
418 IORING_RESTRICTION_SQE_FLAGS_REQUIRED = 3,
420 IORING_RESTRICTION_LAST
423 struct io_uring_getevents_arg {