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;
52 __u64 user_data; /* data to be passed back at completion time */
53 /* pack this to avoid bogus arm OABI complaints */
55 /* index into fixed buffers, if used */
57 /* for grouped buffer selection */
59 } __attribute__((packed));
60 /* personality to use, if used */
72 * If the ring is initialized with IORING_SETUP_SQE128, then
73 * this field is used for 80 bytes of arbitrary command data
80 * If sqe->file_index is set to this for opcodes that instantiate a new
81 * direct descriptor (like openat/openat2/accept), then io_uring will allocate
82 * an available direct descriptor instead of having the application pass one
83 * in. The picked direct descriptor will be returned in cqe->res, or -ENFILE
84 * if the space is full.
86 #define IORING_FILE_INDEX_ALLOC (~0U)
92 IOSQE_IO_HARDLINK_BIT,
94 IOSQE_BUFFER_SELECT_BIT,
95 IOSQE_CQE_SKIP_SUCCESS_BIT,
101 /* use fixed fileset */
102 #define IOSQE_FIXED_FILE (1U << IOSQE_FIXED_FILE_BIT)
103 /* issue after inflight IO */
104 #define IOSQE_IO_DRAIN (1U << IOSQE_IO_DRAIN_BIT)
106 #define IOSQE_IO_LINK (1U << IOSQE_IO_LINK_BIT)
107 /* like LINK, but stronger */
108 #define IOSQE_IO_HARDLINK (1U << IOSQE_IO_HARDLINK_BIT)
109 /* always go async */
110 #define IOSQE_ASYNC (1U << IOSQE_ASYNC_BIT)
111 /* select buffer from sqe->buf_group */
112 #define IOSQE_BUFFER_SELECT (1U << IOSQE_BUFFER_SELECT_BIT)
113 /* don't post CQE if request succeeded */
114 #define IOSQE_CQE_SKIP_SUCCESS (1U << IOSQE_CQE_SKIP_SUCCESS_BIT)
117 * io_uring_setup() flags
119 #define IORING_SETUP_IOPOLL (1U << 0) /* io_context is polled */
120 #define IORING_SETUP_SQPOLL (1U << 1) /* SQ poll thread */
121 #define IORING_SETUP_SQ_AFF (1U << 2) /* sq_thread_cpu is valid */
122 #define IORING_SETUP_CQSIZE (1U << 3) /* app defines CQ size */
123 #define IORING_SETUP_CLAMP (1U << 4) /* clamp SQ/CQ ring sizes */
124 #define IORING_SETUP_ATTACH_WQ (1U << 5) /* attach to existing wq */
125 #define IORING_SETUP_R_DISABLED (1U << 6) /* start with ring disabled */
126 #define IORING_SETUP_SUBMIT_ALL (1U << 7) /* continue submit on error */
128 * Cooperative task running. When requests complete, they often require
129 * forcing the submitter to transition to the kernel to complete. If this
130 * flag is set, work will be done when the task transitions anyway, rather
131 * than force an inter-processor interrupt reschedule. This avoids interrupting
132 * a task running in userspace, and saves an IPI.
134 #define IORING_SETUP_COOP_TASKRUN (1U << 8)
136 * If COOP_TASKRUN is set, get notified if task work is available for
137 * running and a kernel transition would be needed to run it. This sets
138 * IORING_SQ_TASKRUN in the sq ring flags. Not valid with COOP_TASKRUN.
140 #define IORING_SETUP_TASKRUN_FLAG (1U << 9)
142 #define IORING_SETUP_SQE128 (1U << 10) /* SQEs are 128 byte */
143 #define IORING_SETUP_CQE32 (1U << 11) /* CQEs are 32 byte */
150 IORING_OP_READ_FIXED,
151 IORING_OP_WRITE_FIXED,
153 IORING_OP_POLL_REMOVE,
154 IORING_OP_SYNC_FILE_RANGE,
158 IORING_OP_TIMEOUT_REMOVE,
160 IORING_OP_ASYNC_CANCEL,
161 IORING_OP_LINK_TIMEOUT,
166 IORING_OP_FILES_UPDATE,
177 IORING_OP_PROVIDE_BUFFERS,
178 IORING_OP_REMOVE_BUFFERS,
194 /* this goes last, obviously */
201 #define IORING_FSYNC_DATASYNC (1U << 0)
206 #define IORING_TIMEOUT_ABS (1U << 0)
207 #define IORING_TIMEOUT_UPDATE (1U << 1)
208 #define IORING_TIMEOUT_BOOTTIME (1U << 2)
209 #define IORING_TIMEOUT_REALTIME (1U << 3)
210 #define IORING_LINK_TIMEOUT_UPDATE (1U << 4)
211 #define IORING_TIMEOUT_ETIME_SUCCESS (1U << 5)
212 #define IORING_TIMEOUT_CLOCK_MASK (IORING_TIMEOUT_BOOTTIME | IORING_TIMEOUT_REALTIME)
213 #define IORING_TIMEOUT_UPDATE_MASK (IORING_TIMEOUT_UPDATE | IORING_LINK_TIMEOUT_UPDATE)
216 * extends splice(2) flags
218 #define SPLICE_F_FD_IN_FIXED (1U << 31) /* the last bit of __u32 */
221 * POLL_ADD flags. Note that since sqe->poll_events is the flag space, the
222 * command flags for POLL_ADD are stored in sqe->len.
224 * IORING_POLL_ADD_MULTI Multishot poll. Sets IORING_CQE_F_MORE if
225 * the poll handler will continue to report
226 * CQEs on behalf of the same SQE.
228 * IORING_POLL_UPDATE Update existing poll request, matching
229 * sqe->addr as the old user_data field.
231 #define IORING_POLL_ADD_MULTI (1U << 0)
232 #define IORING_POLL_UPDATE_EVENTS (1U << 1)
233 #define IORING_POLL_UPDATE_USER_DATA (1U << 2)
236 * ASYNC_CANCEL flags.
238 * IORING_ASYNC_CANCEL_ALL Cancel all requests that match the given key
239 * IORING_ASYNC_CANCEL_FD Key off 'fd' for cancelation rather than the
240 * request 'user_data'
241 * IORING_ASYNC_CANCEL_ANY Match any request
243 #define IORING_ASYNC_CANCEL_ALL (1U << 0)
244 #define IORING_ASYNC_CANCEL_FD (1U << 1)
245 #define IORING_ASYNC_CANCEL_ANY (1U << 2)
248 * send/sendmsg and recv/recvmsg flags (sqe->addr2)
250 * IORING_RECVSEND_POLL_FIRST If set, instead of first attempting to send
251 * or receive and arm poll if that yields an
252 * -EAGAIN result, arm poll upfront and skip
253 * the initial transfer attempt.
255 #define IORING_RECVSEND_POLL_FIRST (1U << 0)
258 * accept flags stored in sqe->ioprio
260 #define IORING_ACCEPT_MULTISHOT (1U << 0)
263 * close flags, store in sqe->close_flags
265 #define IORING_CLOSE_FD_AND_FILE_SLOT (1U << 0)
268 * IO completion data structure (Completion Queue Entry)
270 struct io_uring_cqe {
271 __u64 user_data; /* sqe->data submission passed back */
272 __s32 res; /* result code for this event */
276 * If the ring is initialized with IORING_SETUP_CQE32, then this field
277 * contains 16-bytes of padding, doubling the size of the CQE.
285 * IORING_CQE_F_BUFFER If set, the upper 16 bits are the buffer ID
286 * IORING_CQE_F_MORE If set, parent SQE will generate more CQE entries
287 * IORING_CQE_F_SOCK_NONEMPTY If set, more data to read after socket recv
289 #define IORING_CQE_F_BUFFER (1U << 0)
290 #define IORING_CQE_F_MORE (1U << 1)
291 #define IORING_CQE_F_SOCK_NONEMPTY (1U << 2)
294 IORING_CQE_BUFFER_SHIFT = 16,
298 * Magic offsets for the application to mmap the data it needs
300 #define IORING_OFF_SQ_RING 0ULL
301 #define IORING_OFF_CQ_RING 0x8000000ULL
302 #define IORING_OFF_SQES 0x10000000ULL
305 * Filled with the offset for mmap(2)
307 struct io_sqring_offsets {
322 #define IORING_SQ_NEED_WAKEUP (1U << 0) /* needs io_uring_enter wakeup */
323 #define IORING_SQ_CQ_OVERFLOW (1U << 1) /* CQ ring is overflown */
324 #define IORING_SQ_TASKRUN (1U << 2) /* task should enter the kernel */
326 struct io_cqring_offsets {
342 /* disable eventfd notifications */
343 #define IORING_CQ_EVENTFD_DISABLED (1U << 0)
346 * io_uring_enter(2) flags
348 #define IORING_ENTER_GETEVENTS (1U << 0)
349 #define IORING_ENTER_SQ_WAKEUP (1U << 1)
350 #define IORING_ENTER_SQ_WAIT (1U << 2)
351 #define IORING_ENTER_EXT_ARG (1U << 3)
352 #define IORING_ENTER_REGISTERED_RING (1U << 4)
355 * Passed in for io_uring_setup(2). Copied back with updated info on success
357 struct io_uring_params {
362 __u32 sq_thread_idle;
366 struct io_sqring_offsets sq_off;
367 struct io_cqring_offsets cq_off;
371 * io_uring_params->features flags
373 #define IORING_FEAT_SINGLE_MMAP (1U << 0)
374 #define IORING_FEAT_NODROP (1U << 1)
375 #define IORING_FEAT_SUBMIT_STABLE (1U << 2)
376 #define IORING_FEAT_RW_CUR_POS (1U << 3)
377 #define IORING_FEAT_CUR_PERSONALITY (1U << 4)
378 #define IORING_FEAT_FAST_POLL (1U << 5)
379 #define IORING_FEAT_POLL_32BITS (1U << 6)
380 #define IORING_FEAT_SQPOLL_NONFIXED (1U << 7)
381 #define IORING_FEAT_EXT_ARG (1U << 8)
382 #define IORING_FEAT_NATIVE_WORKERS (1U << 9)
383 #define IORING_FEAT_RSRC_TAGS (1U << 10)
384 #define IORING_FEAT_CQE_SKIP (1U << 11)
385 #define IORING_FEAT_LINKED_FILE (1U << 12)
388 * io_uring_register(2) opcodes and arguments
391 IORING_REGISTER_BUFFERS = 0,
392 IORING_UNREGISTER_BUFFERS = 1,
393 IORING_REGISTER_FILES = 2,
394 IORING_UNREGISTER_FILES = 3,
395 IORING_REGISTER_EVENTFD = 4,
396 IORING_UNREGISTER_EVENTFD = 5,
397 IORING_REGISTER_FILES_UPDATE = 6,
398 IORING_REGISTER_EVENTFD_ASYNC = 7,
399 IORING_REGISTER_PROBE = 8,
400 IORING_REGISTER_PERSONALITY = 9,
401 IORING_UNREGISTER_PERSONALITY = 10,
402 IORING_REGISTER_RESTRICTIONS = 11,
403 IORING_REGISTER_ENABLE_RINGS = 12,
405 /* extended with tagging */
406 IORING_REGISTER_FILES2 = 13,
407 IORING_REGISTER_FILES_UPDATE2 = 14,
408 IORING_REGISTER_BUFFERS2 = 15,
409 IORING_REGISTER_BUFFERS_UPDATE = 16,
411 /* set/clear io-wq thread affinities */
412 IORING_REGISTER_IOWQ_AFF = 17,
413 IORING_UNREGISTER_IOWQ_AFF = 18,
415 /* set/get max number of io-wq workers */
416 IORING_REGISTER_IOWQ_MAX_WORKERS = 19,
418 /* register/unregister io_uring fd with the ring */
419 IORING_REGISTER_RING_FDS = 20,
420 IORING_UNREGISTER_RING_FDS = 21,
422 /* register ring based provide buffer group */
423 IORING_REGISTER_PBUF_RING = 22,
424 IORING_UNREGISTER_PBUF_RING = 23,
430 /* io-wq worker categories */
436 /* deprecated, see struct io_uring_rsrc_update */
437 struct io_uring_files_update {
440 __aligned_u64 /* __s32 * */ fds;
444 * Register a fully sparse file space, rather than pass in an array of all
445 * -1 file descriptors.
447 #define IORING_RSRC_REGISTER_SPARSE (1U << 0)
449 struct io_uring_rsrc_register {
457 struct io_uring_rsrc_update {
463 struct io_uring_rsrc_update2 {
472 /* Skip updating fd indexes set to this value in the fd table */
473 #define IORING_REGISTER_FILES_SKIP (-2)
475 #define IO_URING_OP_SUPPORTED (1U << 0)
477 struct io_uring_probe_op {
480 __u16 flags; /* IO_URING_OP_* flags */
484 struct io_uring_probe {
485 __u8 last_op; /* last opcode supported */
486 __u8 ops_len; /* length of ops[] array below */
489 struct io_uring_probe_op ops[0];
492 struct io_uring_restriction {
495 __u8 register_op; /* IORING_RESTRICTION_REGISTER_OP */
496 __u8 sqe_op; /* IORING_RESTRICTION_SQE_OP */
497 __u8 sqe_flags; /* IORING_RESTRICTION_SQE_FLAGS_* */
503 struct io_uring_buf {
510 struct io_uring_buf_ring {
513 * To avoid spilling into more pages than we need to, the
514 * ring tail is overlaid with the io_uring_buf->resv field.
522 struct io_uring_buf bufs[0];
526 /* argument for IORING_(UN)REGISTER_PBUF_RING */
527 struct io_uring_buf_reg {
536 * io_uring_restriction->opcode values
539 /* Allow an io_uring_register(2) opcode */
540 IORING_RESTRICTION_REGISTER_OP = 0,
542 /* Allow an sqe opcode */
543 IORING_RESTRICTION_SQE_OP = 1,
545 /* Allow sqe flags */
546 IORING_RESTRICTION_SQE_FLAGS_ALLOWED = 2,
548 /* Require sqe flags (these flags must be set on each submission) */
549 IORING_RESTRICTION_SQE_FLAGS_REQUIRED = 3,
551 IORING_RESTRICTION_LAST
554 struct io_uring_getevents_arg {