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>
13 #include <linux/time_types.h>
16 * IO submission data structure (Submission Queue Entry)
19 __u8 opcode; /* type of operation for this sqe */
20 __u8 flags; /* IOSQE_ flags */
21 __u16 ioprio; /* ioprio for the request */
22 __s32 fd; /* file descriptor to do IO on */
24 __u64 off; /* offset into file */
32 __u64 addr; /* pointer to buffer or iovecs */
35 __u32 len; /* buffer size or number of iovecs */
37 __kernel_rwf_t rw_flags;
39 __u16 poll_events; /* compatibility */
40 __u32 poll32_events; /* word-reversed for BE */
41 __u32 sync_range_flags;
56 __u64 user_data; /* data to be passed back at completion time */
57 /* pack this to avoid bogus arm OABI complaints */
59 /* index into fixed buffers, if used */
61 /* for grouped buffer selection */
63 } __attribute__((packed));
64 /* personality to use, if used */
76 * If the ring is initialized with IORING_SETUP_SQE128, then
77 * this field is used for 80 bytes of arbitrary command data
84 * If sqe->file_index is set to this for opcodes that instantiate a new
85 * direct descriptor (like openat/openat2/accept), then io_uring will allocate
86 * an available direct descriptor instead of having the application pass one
87 * in. The picked direct descriptor will be returned in cqe->res, or -ENFILE
88 * if the space is full.
90 #define IORING_FILE_INDEX_ALLOC (~0U)
96 IOSQE_IO_HARDLINK_BIT,
98 IOSQE_BUFFER_SELECT_BIT,
99 IOSQE_CQE_SKIP_SUCCESS_BIT,
105 /* use fixed fileset */
106 #define IOSQE_FIXED_FILE (1U << IOSQE_FIXED_FILE_BIT)
107 /* issue after inflight IO */
108 #define IOSQE_IO_DRAIN (1U << IOSQE_IO_DRAIN_BIT)
110 #define IOSQE_IO_LINK (1U << IOSQE_IO_LINK_BIT)
111 /* like LINK, but stronger */
112 #define IOSQE_IO_HARDLINK (1U << IOSQE_IO_HARDLINK_BIT)
113 /* always go async */
114 #define IOSQE_ASYNC (1U << IOSQE_ASYNC_BIT)
115 /* select buffer from sqe->buf_group */
116 #define IOSQE_BUFFER_SELECT (1U << IOSQE_BUFFER_SELECT_BIT)
117 /* don't post CQE if request succeeded */
118 #define IOSQE_CQE_SKIP_SUCCESS (1U << IOSQE_CQE_SKIP_SUCCESS_BIT)
121 * io_uring_setup() flags
123 #define IORING_SETUP_IOPOLL (1U << 0) /* io_context is polled */
124 #define IORING_SETUP_SQPOLL (1U << 1) /* SQ poll thread */
125 #define IORING_SETUP_SQ_AFF (1U << 2) /* sq_thread_cpu is valid */
126 #define IORING_SETUP_CQSIZE (1U << 3) /* app defines CQ size */
127 #define IORING_SETUP_CLAMP (1U << 4) /* clamp SQ/CQ ring sizes */
128 #define IORING_SETUP_ATTACH_WQ (1U << 5) /* attach to existing wq */
129 #define IORING_SETUP_R_DISABLED (1U << 6) /* start with ring disabled */
130 #define IORING_SETUP_SUBMIT_ALL (1U << 7) /* continue submit on error */
132 * Cooperative task running. When requests complete, they often require
133 * forcing the submitter to transition to the kernel to complete. If this
134 * flag is set, work will be done when the task transitions anyway, rather
135 * than force an inter-processor interrupt reschedule. This avoids interrupting
136 * a task running in userspace, and saves an IPI.
138 #define IORING_SETUP_COOP_TASKRUN (1U << 8)
140 * If COOP_TASKRUN is set, get notified if task work is available for
141 * running and a kernel transition would be needed to run it. This sets
142 * IORING_SQ_TASKRUN in the sq ring flags. Not valid with COOP_TASKRUN.
144 #define IORING_SETUP_TASKRUN_FLAG (1U << 9)
145 #define IORING_SETUP_SQE128 (1U << 10) /* SQEs are 128 byte */
146 #define IORING_SETUP_CQE32 (1U << 11) /* CQEs are 32 byte */
148 * Only one task is allowed to submit requests
150 #define IORING_SETUP_SINGLE_ISSUER (1U << 12)
157 IORING_OP_READ_FIXED,
158 IORING_OP_WRITE_FIXED,
160 IORING_OP_POLL_REMOVE,
161 IORING_OP_SYNC_FILE_RANGE,
165 IORING_OP_TIMEOUT_REMOVE,
167 IORING_OP_ASYNC_CANCEL,
168 IORING_OP_LINK_TIMEOUT,
173 IORING_OP_FILES_UPDATE,
184 IORING_OP_PROVIDE_BUFFERS,
185 IORING_OP_REMOVE_BUFFERS,
201 /* this goes last, obviously */
208 #define IORING_FSYNC_DATASYNC (1U << 0)
213 #define IORING_TIMEOUT_ABS (1U << 0)
214 #define IORING_TIMEOUT_UPDATE (1U << 1)
215 #define IORING_TIMEOUT_BOOTTIME (1U << 2)
216 #define IORING_TIMEOUT_REALTIME (1U << 3)
217 #define IORING_LINK_TIMEOUT_UPDATE (1U << 4)
218 #define IORING_TIMEOUT_ETIME_SUCCESS (1U << 5)
219 #define IORING_TIMEOUT_CLOCK_MASK (IORING_TIMEOUT_BOOTTIME | IORING_TIMEOUT_REALTIME)
220 #define IORING_TIMEOUT_UPDATE_MASK (IORING_TIMEOUT_UPDATE | IORING_LINK_TIMEOUT_UPDATE)
223 * extends splice(2) flags
225 #define SPLICE_F_FD_IN_FIXED (1U << 31) /* the last bit of __u32 */
228 * POLL_ADD flags. Note that since sqe->poll_events is the flag space, the
229 * command flags for POLL_ADD are stored in sqe->len.
231 * IORING_POLL_ADD_MULTI Multishot poll. Sets IORING_CQE_F_MORE if
232 * the poll handler will continue to report
233 * CQEs on behalf of the same SQE.
235 * IORING_POLL_UPDATE Update existing poll request, matching
236 * sqe->addr as the old user_data field.
238 * IORING_POLL_LEVEL Level triggered poll.
240 #define IORING_POLL_ADD_MULTI (1U << 0)
241 #define IORING_POLL_UPDATE_EVENTS (1U << 1)
242 #define IORING_POLL_UPDATE_USER_DATA (1U << 2)
243 #define IORING_POLL_ADD_LEVEL (1U << 3)
246 * ASYNC_CANCEL flags.
248 * IORING_ASYNC_CANCEL_ALL Cancel all requests that match the given key
249 * IORING_ASYNC_CANCEL_FD Key off 'fd' for cancelation rather than the
250 * request 'user_data'
251 * IORING_ASYNC_CANCEL_ANY Match any request
252 * IORING_ASYNC_CANCEL_FD_FIXED 'fd' passed in is a fixed descriptor
254 #define IORING_ASYNC_CANCEL_ALL (1U << 0)
255 #define IORING_ASYNC_CANCEL_FD (1U << 1)
256 #define IORING_ASYNC_CANCEL_ANY (1U << 2)
257 #define IORING_ASYNC_CANCEL_FD_FIXED (1U << 3)
260 * send/sendmsg and recv/recvmsg flags (sqe->ioprio)
262 * IORING_RECVSEND_POLL_FIRST If set, instead of first attempting to send
263 * or receive and arm poll if that yields an
264 * -EAGAIN result, arm poll upfront and skip
265 * the initial transfer attempt.
267 #define IORING_RECVSEND_POLL_FIRST (1U << 0)
270 * accept flags stored in sqe->ioprio
272 #define IORING_ACCEPT_MULTISHOT (1U << 0)
275 * IORING_OP_MSG_RING command types, stored in sqe->addr
278 IORING_MSG_DATA, /* pass sqe->len as 'res' and off as user_data */
279 IORING_MSG_SEND_FD, /* send a registered fd to another ring */
283 * IORING_OP_MSG_RING flags (sqe->msg_ring_flags)
285 * IORING_MSG_RING_CQE_SKIP Don't post a CQE to the target ring. Not
286 * applicable for IORING_MSG_DATA, obviously.
288 #define IORING_MSG_RING_CQE_SKIP (1U << 0)
291 * IO completion data structure (Completion Queue Entry)
293 struct io_uring_cqe {
294 __u64 user_data; /* sqe->data submission passed back */
295 __s32 res; /* result code for this event */
299 * If the ring is initialized with IORING_SETUP_CQE32, then this field
300 * contains 16-bytes of padding, doubling the size of the CQE.
308 * IORING_CQE_F_BUFFER If set, the upper 16 bits are the buffer ID
309 * IORING_CQE_F_MORE If set, parent SQE will generate more CQE entries
310 * IORING_CQE_F_SOCK_NONEMPTY If set, more data to read after socket recv
312 #define IORING_CQE_F_BUFFER (1U << 0)
313 #define IORING_CQE_F_MORE (1U << 1)
314 #define IORING_CQE_F_SOCK_NONEMPTY (1U << 2)
317 IORING_CQE_BUFFER_SHIFT = 16,
321 * Magic offsets for the application to mmap the data it needs
323 #define IORING_OFF_SQ_RING 0ULL
324 #define IORING_OFF_CQ_RING 0x8000000ULL
325 #define IORING_OFF_SQES 0x10000000ULL
328 * Filled with the offset for mmap(2)
330 struct io_sqring_offsets {
345 #define IORING_SQ_NEED_WAKEUP (1U << 0) /* needs io_uring_enter wakeup */
346 #define IORING_SQ_CQ_OVERFLOW (1U << 1) /* CQ ring is overflown */
347 #define IORING_SQ_TASKRUN (1U << 2) /* task should enter the kernel */
349 struct io_cqring_offsets {
365 /* disable eventfd notifications */
366 #define IORING_CQ_EVENTFD_DISABLED (1U << 0)
369 * io_uring_enter(2) flags
371 #define IORING_ENTER_GETEVENTS (1U << 0)
372 #define IORING_ENTER_SQ_WAKEUP (1U << 1)
373 #define IORING_ENTER_SQ_WAIT (1U << 2)
374 #define IORING_ENTER_EXT_ARG (1U << 3)
375 #define IORING_ENTER_REGISTERED_RING (1U << 4)
378 * Passed in for io_uring_setup(2). Copied back with updated info on success
380 struct io_uring_params {
385 __u32 sq_thread_idle;
389 struct io_sqring_offsets sq_off;
390 struct io_cqring_offsets cq_off;
394 * io_uring_params->features flags
396 #define IORING_FEAT_SINGLE_MMAP (1U << 0)
397 #define IORING_FEAT_NODROP (1U << 1)
398 #define IORING_FEAT_SUBMIT_STABLE (1U << 2)
399 #define IORING_FEAT_RW_CUR_POS (1U << 3)
400 #define IORING_FEAT_CUR_PERSONALITY (1U << 4)
401 #define IORING_FEAT_FAST_POLL (1U << 5)
402 #define IORING_FEAT_POLL_32BITS (1U << 6)
403 #define IORING_FEAT_SQPOLL_NONFIXED (1U << 7)
404 #define IORING_FEAT_EXT_ARG (1U << 8)
405 #define IORING_FEAT_NATIVE_WORKERS (1U << 9)
406 #define IORING_FEAT_RSRC_TAGS (1U << 10)
407 #define IORING_FEAT_CQE_SKIP (1U << 11)
408 #define IORING_FEAT_LINKED_FILE (1U << 12)
411 * io_uring_register(2) opcodes and arguments
414 IORING_REGISTER_BUFFERS = 0,
415 IORING_UNREGISTER_BUFFERS = 1,
416 IORING_REGISTER_FILES = 2,
417 IORING_UNREGISTER_FILES = 3,
418 IORING_REGISTER_EVENTFD = 4,
419 IORING_UNREGISTER_EVENTFD = 5,
420 IORING_REGISTER_FILES_UPDATE = 6,
421 IORING_REGISTER_EVENTFD_ASYNC = 7,
422 IORING_REGISTER_PROBE = 8,
423 IORING_REGISTER_PERSONALITY = 9,
424 IORING_UNREGISTER_PERSONALITY = 10,
425 IORING_REGISTER_RESTRICTIONS = 11,
426 IORING_REGISTER_ENABLE_RINGS = 12,
428 /* extended with tagging */
429 IORING_REGISTER_FILES2 = 13,
430 IORING_REGISTER_FILES_UPDATE2 = 14,
431 IORING_REGISTER_BUFFERS2 = 15,
432 IORING_REGISTER_BUFFERS_UPDATE = 16,
434 /* set/clear io-wq thread affinities */
435 IORING_REGISTER_IOWQ_AFF = 17,
436 IORING_UNREGISTER_IOWQ_AFF = 18,
438 /* set/get max number of io-wq workers */
439 IORING_REGISTER_IOWQ_MAX_WORKERS = 19,
441 /* register/unregister io_uring fd with the ring */
442 IORING_REGISTER_RING_FDS = 20,
443 IORING_UNREGISTER_RING_FDS = 21,
445 /* register ring based provide buffer group */
446 IORING_REGISTER_PBUF_RING = 22,
447 IORING_UNREGISTER_PBUF_RING = 23,
449 /* sync cancelation API */
450 IORING_REGISTER_SYNC_CANCEL = 24,
452 /* register a range of fixed file slots for automatic slot allocation */
453 IORING_REGISTER_FILE_ALLOC_RANGE = 25,
459 /* io-wq worker categories */
465 /* deprecated, see struct io_uring_rsrc_update */
466 struct io_uring_files_update {
469 __aligned_u64 /* __s32 * */ fds;
473 * Register a fully sparse file space, rather than pass in an array of all
474 * -1 file descriptors.
476 #define IORING_RSRC_REGISTER_SPARSE (1U << 0)
478 struct io_uring_rsrc_register {
486 struct io_uring_rsrc_update {
492 struct io_uring_rsrc_update2 {
501 /* Skip updating fd indexes set to this value in the fd table */
502 #define IORING_REGISTER_FILES_SKIP (-2)
504 #define IO_URING_OP_SUPPORTED (1U << 0)
506 struct io_uring_probe_op {
509 __u16 flags; /* IO_URING_OP_* flags */
513 struct io_uring_probe {
514 __u8 last_op; /* last opcode supported */
515 __u8 ops_len; /* length of ops[] array below */
518 struct io_uring_probe_op ops[];
521 struct io_uring_restriction {
524 __u8 register_op; /* IORING_RESTRICTION_REGISTER_OP */
525 __u8 sqe_op; /* IORING_RESTRICTION_SQE_OP */
526 __u8 sqe_flags; /* IORING_RESTRICTION_SQE_FLAGS_* */
532 struct io_uring_buf {
539 struct io_uring_buf_ring {
542 * To avoid spilling into more pages than we need to, the
543 * ring tail is overlaid with the io_uring_buf->resv field.
551 struct io_uring_buf bufs[0];
555 /* argument for IORING_(UN)REGISTER_PBUF_RING */
556 struct io_uring_buf_reg {
565 * io_uring_restriction->opcode values
568 /* Allow an io_uring_register(2) opcode */
569 IORING_RESTRICTION_REGISTER_OP = 0,
571 /* Allow an sqe opcode */
572 IORING_RESTRICTION_SQE_OP = 1,
574 /* Allow sqe flags */
575 IORING_RESTRICTION_SQE_FLAGS_ALLOWED = 2,
577 /* Require sqe flags (these flags must be set on each submission) */
578 IORING_RESTRICTION_SQE_FLAGS_REQUIRED = 3,
580 IORING_RESTRICTION_LAST
583 struct io_uring_getevents_arg {
591 * Argument for IORING_REGISTER_SYNC_CANCEL
593 struct io_uring_sync_cancel_reg {
597 struct __kernel_timespec timeout;
602 * Argument for IORING_REGISTER_FILE_ALLOC_RANGE
603 * The range is specified as [off, off + len)
605 struct io_uring_file_index_range {