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>
20 * IO submission data structure (Submission Queue Entry)
23 __u8 opcode; /* type of operation for this sqe */
24 __u8 flags; /* IOSQE_ flags */
25 __u16 ioprio; /* ioprio for the request */
26 __s32 fd; /* file descriptor to do IO on */
28 __u64 off; /* offset into file */
36 __u64 addr; /* pointer to buffer or iovecs */
39 __u32 len; /* buffer size or number of iovecs */
41 __kernel_rwf_t rw_flags;
43 __u16 poll_events; /* compatibility */
44 __u32 poll32_events; /* word-reversed for BE */
45 __u32 sync_range_flags;
60 __u64 user_data; /* data to be passed back at completion time */
61 /* pack this to avoid bogus arm OABI complaints */
63 /* index into fixed buffers, if used */
65 /* for grouped buffer selection */
67 } __attribute__((packed));
68 /* personality to use, if used */
84 * If the ring is initialized with IORING_SETUP_SQE128, then
85 * this field is used for 80 bytes of arbitrary command data
92 * If sqe->file_index is set to this for opcodes that instantiate a new
93 * direct descriptor (like openat/openat2/accept), then io_uring will allocate
94 * an available direct descriptor instead of having the application pass one
95 * in. The picked direct descriptor will be returned in cqe->res, or -ENFILE
96 * if the space is full.
98 #define IORING_FILE_INDEX_ALLOC (~0U)
101 IOSQE_FIXED_FILE_BIT,
104 IOSQE_IO_HARDLINK_BIT,
106 IOSQE_BUFFER_SELECT_BIT,
107 IOSQE_CQE_SKIP_SUCCESS_BIT,
113 /* use fixed fileset */
114 #define IOSQE_FIXED_FILE (1U << IOSQE_FIXED_FILE_BIT)
115 /* issue after inflight IO */
116 #define IOSQE_IO_DRAIN (1U << IOSQE_IO_DRAIN_BIT)
118 #define IOSQE_IO_LINK (1U << IOSQE_IO_LINK_BIT)
119 /* like LINK, but stronger */
120 #define IOSQE_IO_HARDLINK (1U << IOSQE_IO_HARDLINK_BIT)
121 /* always go async */
122 #define IOSQE_ASYNC (1U << IOSQE_ASYNC_BIT)
123 /* select buffer from sqe->buf_group */
124 #define IOSQE_BUFFER_SELECT (1U << IOSQE_BUFFER_SELECT_BIT)
125 /* don't post CQE if request succeeded */
126 #define IOSQE_CQE_SKIP_SUCCESS (1U << IOSQE_CQE_SKIP_SUCCESS_BIT)
129 * io_uring_setup() flags
131 #define IORING_SETUP_IOPOLL (1U << 0) /* io_context is polled */
132 #define IORING_SETUP_SQPOLL (1U << 1) /* SQ poll thread */
133 #define IORING_SETUP_SQ_AFF (1U << 2) /* sq_thread_cpu is valid */
134 #define IORING_SETUP_CQSIZE (1U << 3) /* app defines CQ size */
135 #define IORING_SETUP_CLAMP (1U << 4) /* clamp SQ/CQ ring sizes */
136 #define IORING_SETUP_ATTACH_WQ (1U << 5) /* attach to existing wq */
137 #define IORING_SETUP_R_DISABLED (1U << 6) /* start with ring disabled */
138 #define IORING_SETUP_SUBMIT_ALL (1U << 7) /* continue submit on error */
140 * Cooperative task running. When requests complete, they often require
141 * forcing the submitter to transition to the kernel to complete. If this
142 * flag is set, work will be done when the task transitions anyway, rather
143 * than force an inter-processor interrupt reschedule. This avoids interrupting
144 * a task running in userspace, and saves an IPI.
146 #define IORING_SETUP_COOP_TASKRUN (1U << 8)
148 * If COOP_TASKRUN is set, get notified if task work is available for
149 * running and a kernel transition would be needed to run it. This sets
150 * IORING_SQ_TASKRUN in the sq ring flags. Not valid with COOP_TASKRUN.
152 #define IORING_SETUP_TASKRUN_FLAG (1U << 9)
153 #define IORING_SETUP_SQE128 (1U << 10) /* SQEs are 128 byte */
154 #define IORING_SETUP_CQE32 (1U << 11) /* CQEs are 32 byte */
156 * Only one task is allowed to submit requests
158 #define IORING_SETUP_SINGLE_ISSUER (1U << 12)
161 * Defer running task work to get events.
162 * Rather than running bits of task work whenever the task transitions
163 * try to do it just before it is needed.
165 #define IORING_SETUP_DEFER_TASKRUN (1U << 13)
172 IORING_OP_READ_FIXED,
173 IORING_OP_WRITE_FIXED,
175 IORING_OP_POLL_REMOVE,
176 IORING_OP_SYNC_FILE_RANGE,
180 IORING_OP_TIMEOUT_REMOVE,
182 IORING_OP_ASYNC_CANCEL,
183 IORING_OP_LINK_TIMEOUT,
188 IORING_OP_FILES_UPDATE,
199 IORING_OP_PROVIDE_BUFFERS,
200 IORING_OP_REMOVE_BUFFERS,
217 /* this goes last, obviously */
224 #define IORING_FSYNC_DATASYNC (1U << 0)
229 #define IORING_TIMEOUT_ABS (1U << 0)
230 #define IORING_TIMEOUT_UPDATE (1U << 1)
231 #define IORING_TIMEOUT_BOOTTIME (1U << 2)
232 #define IORING_TIMEOUT_REALTIME (1U << 3)
233 #define IORING_LINK_TIMEOUT_UPDATE (1U << 4)
234 #define IORING_TIMEOUT_ETIME_SUCCESS (1U << 5)
235 #define IORING_TIMEOUT_CLOCK_MASK (IORING_TIMEOUT_BOOTTIME | IORING_TIMEOUT_REALTIME)
236 #define IORING_TIMEOUT_UPDATE_MASK (IORING_TIMEOUT_UPDATE | IORING_LINK_TIMEOUT_UPDATE)
239 * extends splice(2) flags
241 #define SPLICE_F_FD_IN_FIXED (1U << 31) /* the last bit of __u32 */
244 * POLL_ADD flags. Note that since sqe->poll_events is the flag space, the
245 * command flags for POLL_ADD are stored in sqe->len.
247 * IORING_POLL_ADD_MULTI Multishot poll. Sets IORING_CQE_F_MORE if
248 * the poll handler will continue to report
249 * CQEs on behalf of the same SQE.
251 * IORING_POLL_UPDATE Update existing poll request, matching
252 * sqe->addr as the old user_data field.
254 * IORING_POLL_LEVEL Level triggered poll.
256 #define IORING_POLL_ADD_MULTI (1U << 0)
257 #define IORING_POLL_UPDATE_EVENTS (1U << 1)
258 #define IORING_POLL_UPDATE_USER_DATA (1U << 2)
259 #define IORING_POLL_ADD_LEVEL (1U << 3)
262 * ASYNC_CANCEL flags.
264 * IORING_ASYNC_CANCEL_ALL Cancel all requests that match the given key
265 * IORING_ASYNC_CANCEL_FD Key off 'fd' for cancelation rather than the
266 * request 'user_data'
267 * IORING_ASYNC_CANCEL_ANY Match any request
268 * IORING_ASYNC_CANCEL_FD_FIXED 'fd' passed in is a fixed descriptor
270 #define IORING_ASYNC_CANCEL_ALL (1U << 0)
271 #define IORING_ASYNC_CANCEL_FD (1U << 1)
272 #define IORING_ASYNC_CANCEL_ANY (1U << 2)
273 #define IORING_ASYNC_CANCEL_FD_FIXED (1U << 3)
276 * send/sendmsg and recv/recvmsg flags (sqe->ioprio)
278 * IORING_RECVSEND_POLL_FIRST If set, instead of first attempting to send
279 * or receive and arm poll if that yields an
280 * -EAGAIN result, arm poll upfront and skip
281 * the initial transfer attempt.
283 * IORING_RECV_MULTISHOT Multishot recv. Sets IORING_CQE_F_MORE if
284 * the handler will continue to report
285 * CQEs on behalf of the same SQE.
287 * IORING_RECVSEND_FIXED_BUF Use registered buffers, the index is stored in
288 * the buf_index field.
290 #define IORING_RECVSEND_POLL_FIRST (1U << 0)
291 #define IORING_RECV_MULTISHOT (1U << 1)
292 #define IORING_RECVSEND_FIXED_BUF (1U << 2)
295 * accept flags stored in sqe->ioprio
297 #define IORING_ACCEPT_MULTISHOT (1U << 0)
300 * IORING_OP_MSG_RING command types, stored in sqe->addr
303 IORING_MSG_DATA, /* pass sqe->len as 'res' and off as user_data */
304 IORING_MSG_SEND_FD, /* send a registered fd to another ring */
308 * IORING_OP_MSG_RING flags (sqe->msg_ring_flags)
310 * IORING_MSG_RING_CQE_SKIP Don't post a CQE to the target ring. Not
311 * applicable for IORING_MSG_DATA, obviously.
313 #define IORING_MSG_RING_CQE_SKIP (1U << 0)
316 * IO completion data structure (Completion Queue Entry)
318 struct io_uring_cqe {
319 __u64 user_data; /* sqe->data submission passed back */
320 __s32 res; /* result code for this event */
324 * If the ring is initialized with IORING_SETUP_CQE32, then this field
325 * contains 16-bytes of padding, doubling the size of the CQE.
333 * IORING_CQE_F_BUFFER If set, the upper 16 bits are the buffer ID
334 * IORING_CQE_F_MORE If set, parent SQE will generate more CQE entries
335 * IORING_CQE_F_SOCK_NONEMPTY If set, more data to read after socket recv
336 * IORING_CQE_F_NOTIF Set for notification CQEs. Can be used to distinct
339 #define IORING_CQE_F_BUFFER (1U << 0)
340 #define IORING_CQE_F_MORE (1U << 1)
341 #define IORING_CQE_F_SOCK_NONEMPTY (1U << 2)
342 #define IORING_CQE_F_NOTIF (1U << 3)
345 IORING_CQE_BUFFER_SHIFT = 16,
349 * Magic offsets for the application to mmap the data it needs
351 #define IORING_OFF_SQ_RING 0ULL
352 #define IORING_OFF_CQ_RING 0x8000000ULL
353 #define IORING_OFF_SQES 0x10000000ULL
356 * Filled with the offset for mmap(2)
358 struct io_sqring_offsets {
373 #define IORING_SQ_NEED_WAKEUP (1U << 0) /* needs io_uring_enter wakeup */
374 #define IORING_SQ_CQ_OVERFLOW (1U << 1) /* CQ ring is overflown */
375 #define IORING_SQ_TASKRUN (1U << 2) /* task should enter the kernel */
377 struct io_cqring_offsets {
393 /* disable eventfd notifications */
394 #define IORING_CQ_EVENTFD_DISABLED (1U << 0)
397 * io_uring_enter(2) flags
399 #define IORING_ENTER_GETEVENTS (1U << 0)
400 #define IORING_ENTER_SQ_WAKEUP (1U << 1)
401 #define IORING_ENTER_SQ_WAIT (1U << 2)
402 #define IORING_ENTER_EXT_ARG (1U << 3)
403 #define IORING_ENTER_REGISTERED_RING (1U << 4)
406 * Passed in for io_uring_setup(2). Copied back with updated info on success
408 struct io_uring_params {
413 __u32 sq_thread_idle;
417 struct io_sqring_offsets sq_off;
418 struct io_cqring_offsets cq_off;
422 * io_uring_params->features flags
424 #define IORING_FEAT_SINGLE_MMAP (1U << 0)
425 #define IORING_FEAT_NODROP (1U << 1)
426 #define IORING_FEAT_SUBMIT_STABLE (1U << 2)
427 #define IORING_FEAT_RW_CUR_POS (1U << 3)
428 #define IORING_FEAT_CUR_PERSONALITY (1U << 4)
429 #define IORING_FEAT_FAST_POLL (1U << 5)
430 #define IORING_FEAT_POLL_32BITS (1U << 6)
431 #define IORING_FEAT_SQPOLL_NONFIXED (1U << 7)
432 #define IORING_FEAT_EXT_ARG (1U << 8)
433 #define IORING_FEAT_NATIVE_WORKERS (1U << 9)
434 #define IORING_FEAT_RSRC_TAGS (1U << 10)
435 #define IORING_FEAT_CQE_SKIP (1U << 11)
436 #define IORING_FEAT_LINKED_FILE (1U << 12)
439 * io_uring_register(2) opcodes and arguments
442 IORING_REGISTER_BUFFERS = 0,
443 IORING_UNREGISTER_BUFFERS = 1,
444 IORING_REGISTER_FILES = 2,
445 IORING_UNREGISTER_FILES = 3,
446 IORING_REGISTER_EVENTFD = 4,
447 IORING_UNREGISTER_EVENTFD = 5,
448 IORING_REGISTER_FILES_UPDATE = 6,
449 IORING_REGISTER_EVENTFD_ASYNC = 7,
450 IORING_REGISTER_PROBE = 8,
451 IORING_REGISTER_PERSONALITY = 9,
452 IORING_UNREGISTER_PERSONALITY = 10,
453 IORING_REGISTER_RESTRICTIONS = 11,
454 IORING_REGISTER_ENABLE_RINGS = 12,
456 /* extended with tagging */
457 IORING_REGISTER_FILES2 = 13,
458 IORING_REGISTER_FILES_UPDATE2 = 14,
459 IORING_REGISTER_BUFFERS2 = 15,
460 IORING_REGISTER_BUFFERS_UPDATE = 16,
462 /* set/clear io-wq thread affinities */
463 IORING_REGISTER_IOWQ_AFF = 17,
464 IORING_UNREGISTER_IOWQ_AFF = 18,
466 /* set/get max number of io-wq workers */
467 IORING_REGISTER_IOWQ_MAX_WORKERS = 19,
469 /* register/unregister io_uring fd with the ring */
470 IORING_REGISTER_RING_FDS = 20,
471 IORING_UNREGISTER_RING_FDS = 21,
473 /* register ring based provide buffer group */
474 IORING_REGISTER_PBUF_RING = 22,
475 IORING_UNREGISTER_PBUF_RING = 23,
477 /* sync cancelation API */
478 IORING_REGISTER_SYNC_CANCEL = 24,
480 /* register a range of fixed file slots for automatic slot allocation */
481 IORING_REGISTER_FILE_ALLOC_RANGE = 25,
487 /* io-wq worker categories */
493 /* deprecated, see struct io_uring_rsrc_update */
494 struct io_uring_files_update {
497 __aligned_u64 /* __s32 * */ fds;
501 * Register a fully sparse file space, rather than pass in an array of all
502 * -1 file descriptors.
504 #define IORING_RSRC_REGISTER_SPARSE (1U << 0)
506 struct io_uring_rsrc_register {
514 struct io_uring_rsrc_update {
520 struct io_uring_rsrc_update2 {
529 struct io_uring_notification_slot {
534 struct io_uring_notification_register {
542 /* Skip updating fd indexes set to this value in the fd table */
543 #define IORING_REGISTER_FILES_SKIP (-2)
545 #define IO_URING_OP_SUPPORTED (1U << 0)
547 struct io_uring_probe_op {
550 __u16 flags; /* IO_URING_OP_* flags */
554 struct io_uring_probe {
555 __u8 last_op; /* last opcode supported */
556 __u8 ops_len; /* length of ops[] array below */
559 struct io_uring_probe_op ops[];
562 struct io_uring_restriction {
565 __u8 register_op; /* IORING_RESTRICTION_REGISTER_OP */
566 __u8 sqe_op; /* IORING_RESTRICTION_SQE_OP */
567 __u8 sqe_flags; /* IORING_RESTRICTION_SQE_FLAGS_* */
573 struct io_uring_buf {
580 struct io_uring_buf_ring {
583 * To avoid spilling into more pages than we need to, the
584 * ring tail is overlaid with the io_uring_buf->resv field.
592 struct io_uring_buf bufs[0];
596 /* argument for IORING_(UN)REGISTER_PBUF_RING */
597 struct io_uring_buf_reg {
606 * io_uring_restriction->opcode values
609 /* Allow an io_uring_register(2) opcode */
610 IORING_RESTRICTION_REGISTER_OP = 0,
612 /* Allow an sqe opcode */
613 IORING_RESTRICTION_SQE_OP = 1,
615 /* Allow sqe flags */
616 IORING_RESTRICTION_SQE_FLAGS_ALLOWED = 2,
618 /* Require sqe flags (these flags must be set on each submission) */
619 IORING_RESTRICTION_SQE_FLAGS_REQUIRED = 3,
621 IORING_RESTRICTION_LAST
624 struct io_uring_getevents_arg {
632 * Argument for IORING_REGISTER_SYNC_CANCEL
634 struct io_uring_sync_cancel_reg {
638 struct __kernel_timespec timeout;
643 * Argument for IORING_REGISTER_FILE_ALLOC_RANGE
644 * The range is specified as [off, off + len)
646 struct io_uring_file_index_range {
652 struct io_uring_recvmsg_out {