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>
14 * this file is shared with liburing and that has to autodetect
15 * if linux/time_types.h is available or not, it can
16 * define UAPI_LINUX_IO_URING_H_SKIP_LINUX_TIME_TYPES_H
17 * if linux/time_types.h is not available
19 #ifndef UAPI_LINUX_IO_URING_H_SKIP_LINUX_TIME_TYPES_H
20 #include <linux/time_types.h>
28 * IO submission data structure (Submission Queue Entry)
31 __u8 opcode; /* type of operation for this sqe */
32 __u8 flags; /* IOSQE_ flags */
33 __u16 ioprio; /* ioprio for the request */
34 __s32 fd; /* file descriptor to do IO on */
36 __u64 off; /* offset into file */
44 __u64 addr; /* pointer to buffer or iovecs */
47 __u32 len; /* buffer size or number of iovecs */
49 __kernel_rwf_t rw_flags;
51 __u16 poll_events; /* compatibility */
52 __u32 poll32_events; /* word-reversed for BE */
53 __u32 sync_range_flags;
67 __u32 uring_cmd_flags;
70 __u64 user_data; /* data to be passed back at completion time */
71 /* pack this to avoid bogus arm OABI complaints */
73 /* index into fixed buffers, if used */
75 /* for grouped buffer selection */
77 } __attribute__((packed));
78 /* personality to use, if used */
94 * If the ring is initialized with IORING_SETUP_SQE128, then
95 * this field is used for 80 bytes of arbitrary command data
102 * If sqe->file_index is set to this for opcodes that instantiate a new
103 * direct descriptor (like openat/openat2/accept), then io_uring will allocate
104 * an available direct descriptor instead of having the application pass one
105 * in. The picked direct descriptor will be returned in cqe->res, or -ENFILE
106 * if the space is full.
108 #define IORING_FILE_INDEX_ALLOC (~0U)
111 IOSQE_FIXED_FILE_BIT,
114 IOSQE_IO_HARDLINK_BIT,
116 IOSQE_BUFFER_SELECT_BIT,
117 IOSQE_CQE_SKIP_SUCCESS_BIT,
123 /* use fixed fileset */
124 #define IOSQE_FIXED_FILE (1U << IOSQE_FIXED_FILE_BIT)
125 /* issue after inflight IO */
126 #define IOSQE_IO_DRAIN (1U << IOSQE_IO_DRAIN_BIT)
128 #define IOSQE_IO_LINK (1U << IOSQE_IO_LINK_BIT)
129 /* like LINK, but stronger */
130 #define IOSQE_IO_HARDLINK (1U << IOSQE_IO_HARDLINK_BIT)
131 /* always go async */
132 #define IOSQE_ASYNC (1U << IOSQE_ASYNC_BIT)
133 /* select buffer from sqe->buf_group */
134 #define IOSQE_BUFFER_SELECT (1U << IOSQE_BUFFER_SELECT_BIT)
135 /* don't post CQE if request succeeded */
136 #define IOSQE_CQE_SKIP_SUCCESS (1U << IOSQE_CQE_SKIP_SUCCESS_BIT)
139 * io_uring_setup() flags
141 #define IORING_SETUP_IOPOLL (1U << 0) /* io_context is polled */
142 #define IORING_SETUP_SQPOLL (1U << 1) /* SQ poll thread */
143 #define IORING_SETUP_SQ_AFF (1U << 2) /* sq_thread_cpu is valid */
144 #define IORING_SETUP_CQSIZE (1U << 3) /* app defines CQ size */
145 #define IORING_SETUP_CLAMP (1U << 4) /* clamp SQ/CQ ring sizes */
146 #define IORING_SETUP_ATTACH_WQ (1U << 5) /* attach to existing wq */
147 #define IORING_SETUP_R_DISABLED (1U << 6) /* start with ring disabled */
148 #define IORING_SETUP_SUBMIT_ALL (1U << 7) /* continue submit on error */
150 * Cooperative task running. When requests complete, they often require
151 * forcing the submitter to transition to the kernel to complete. If this
152 * flag is set, work will be done when the task transitions anyway, rather
153 * than force an inter-processor interrupt reschedule. This avoids interrupting
154 * a task running in userspace, and saves an IPI.
156 #define IORING_SETUP_COOP_TASKRUN (1U << 8)
158 * If COOP_TASKRUN is set, get notified if task work is available for
159 * running and a kernel transition would be needed to run it. This sets
160 * IORING_SQ_TASKRUN in the sq ring flags. Not valid with COOP_TASKRUN.
162 #define IORING_SETUP_TASKRUN_FLAG (1U << 9)
163 #define IORING_SETUP_SQE128 (1U << 10) /* SQEs are 128 byte */
164 #define IORING_SETUP_CQE32 (1U << 11) /* CQEs are 32 byte */
166 * Only one task is allowed to submit requests
168 #define IORING_SETUP_SINGLE_ISSUER (1U << 12)
171 * Defer running task work to get events.
172 * Rather than running bits of task work whenever the task transitions
173 * try to do it just before it is needed.
175 #define IORING_SETUP_DEFER_TASKRUN (1U << 13)
178 * Application provides the memory for the rings
180 #define IORING_SETUP_NO_MMAP (1U << 14)
183 * Register the ring fd in itself for use with
184 * IORING_REGISTER_USE_REGISTERED_RING; return a registered fd index rather
187 #define IORING_SETUP_REGISTERED_FD_ONLY (1U << 15)
190 * Removes indirection through the SQ index array.
192 #define IORING_SETUP_NO_SQARRAY (1U << 16)
199 IORING_OP_READ_FIXED,
200 IORING_OP_WRITE_FIXED,
202 IORING_OP_POLL_REMOVE,
203 IORING_OP_SYNC_FILE_RANGE,
207 IORING_OP_TIMEOUT_REMOVE,
209 IORING_OP_ASYNC_CANCEL,
210 IORING_OP_LINK_TIMEOUT,
215 IORING_OP_FILES_UPDATE,
226 IORING_OP_PROVIDE_BUFFERS,
227 IORING_OP_REMOVE_BUFFERS,
243 IORING_OP_SENDMSG_ZC,
244 IORING_OP_READ_MULTISHOT,
247 /* this goes last, obviously */
252 * sqe->uring_cmd_flags top 8bits aren't available for userspace
253 * IORING_URING_CMD_FIXED use registered buffer; pass this flag
254 * along with setting sqe->buf_index.
256 #define IORING_URING_CMD_FIXED (1U << 0)
257 #define IORING_URING_CMD_MASK IORING_URING_CMD_FIXED
263 #define IORING_FSYNC_DATASYNC (1U << 0)
268 #define IORING_TIMEOUT_ABS (1U << 0)
269 #define IORING_TIMEOUT_UPDATE (1U << 1)
270 #define IORING_TIMEOUT_BOOTTIME (1U << 2)
271 #define IORING_TIMEOUT_REALTIME (1U << 3)
272 #define IORING_LINK_TIMEOUT_UPDATE (1U << 4)
273 #define IORING_TIMEOUT_ETIME_SUCCESS (1U << 5)
274 #define IORING_TIMEOUT_MULTISHOT (1U << 6)
275 #define IORING_TIMEOUT_CLOCK_MASK (IORING_TIMEOUT_BOOTTIME | IORING_TIMEOUT_REALTIME)
276 #define IORING_TIMEOUT_UPDATE_MASK (IORING_TIMEOUT_UPDATE | IORING_LINK_TIMEOUT_UPDATE)
279 * extends splice(2) flags
281 #define SPLICE_F_FD_IN_FIXED (1U << 31) /* the last bit of __u32 */
284 * POLL_ADD flags. Note that since sqe->poll_events is the flag space, the
285 * command flags for POLL_ADD are stored in sqe->len.
287 * IORING_POLL_ADD_MULTI Multishot poll. Sets IORING_CQE_F_MORE if
288 * the poll handler will continue to report
289 * CQEs on behalf of the same SQE.
291 * IORING_POLL_UPDATE Update existing poll request, matching
292 * sqe->addr as the old user_data field.
294 * IORING_POLL_LEVEL Level triggered poll.
296 #define IORING_POLL_ADD_MULTI (1U << 0)
297 #define IORING_POLL_UPDATE_EVENTS (1U << 1)
298 #define IORING_POLL_UPDATE_USER_DATA (1U << 2)
299 #define IORING_POLL_ADD_LEVEL (1U << 3)
302 * ASYNC_CANCEL flags.
304 * IORING_ASYNC_CANCEL_ALL Cancel all requests that match the given key
305 * IORING_ASYNC_CANCEL_FD Key off 'fd' for cancelation rather than the
306 * request 'user_data'
307 * IORING_ASYNC_CANCEL_ANY Match any request
308 * IORING_ASYNC_CANCEL_FD_FIXED 'fd' passed in is a fixed descriptor
309 * IORING_ASYNC_CANCEL_USERDATA Match on user_data, default for no other key
310 * IORING_ASYNC_CANCEL_OP Match request based on opcode
312 #define IORING_ASYNC_CANCEL_ALL (1U << 0)
313 #define IORING_ASYNC_CANCEL_FD (1U << 1)
314 #define IORING_ASYNC_CANCEL_ANY (1U << 2)
315 #define IORING_ASYNC_CANCEL_FD_FIXED (1U << 3)
316 #define IORING_ASYNC_CANCEL_USERDATA (1U << 4)
317 #define IORING_ASYNC_CANCEL_OP (1U << 5)
320 * send/sendmsg and recv/recvmsg flags (sqe->ioprio)
322 * IORING_RECVSEND_POLL_FIRST If set, instead of first attempting to send
323 * or receive and arm poll if that yields an
324 * -EAGAIN result, arm poll upfront and skip
325 * the initial transfer attempt.
327 * IORING_RECV_MULTISHOT Multishot recv. Sets IORING_CQE_F_MORE if
328 * the handler will continue to report
329 * CQEs on behalf of the same SQE.
331 * IORING_RECVSEND_FIXED_BUF Use registered buffers, the index is stored in
332 * the buf_index field.
334 * IORING_SEND_ZC_REPORT_USAGE
335 * If set, SEND[MSG]_ZC should report
336 * the zerocopy usage in cqe.res
337 * for the IORING_CQE_F_NOTIF cqe.
338 * 0 is reported if zerocopy was actually possible.
339 * IORING_NOTIF_USAGE_ZC_COPIED if data was copied
340 * (at least partially).
342 #define IORING_RECVSEND_POLL_FIRST (1U << 0)
343 #define IORING_RECV_MULTISHOT (1U << 1)
344 #define IORING_RECVSEND_FIXED_BUF (1U << 2)
345 #define IORING_SEND_ZC_REPORT_USAGE (1U << 3)
348 * cqe.res for IORING_CQE_F_NOTIF if
349 * IORING_SEND_ZC_REPORT_USAGE was requested
351 * It should be treated as a flag, all other
352 * bits of cqe.res should be treated as reserved!
354 #define IORING_NOTIF_USAGE_ZC_COPIED (1U << 31)
357 * accept flags stored in sqe->ioprio
359 #define IORING_ACCEPT_MULTISHOT (1U << 0)
362 * IORING_OP_MSG_RING command types, stored in sqe->addr
365 IORING_MSG_DATA, /* pass sqe->len as 'res' and off as user_data */
366 IORING_MSG_SEND_FD, /* send a registered fd to another ring */
370 * IORING_OP_MSG_RING flags (sqe->msg_ring_flags)
372 * IORING_MSG_RING_CQE_SKIP Don't post a CQE to the target ring. Not
373 * applicable for IORING_MSG_DATA, obviously.
375 #define IORING_MSG_RING_CQE_SKIP (1U << 0)
376 /* Pass through the flags from sqe->file_index to cqe->flags */
377 #define IORING_MSG_RING_FLAGS_PASS (1U << 1)
380 * IO completion data structure (Completion Queue Entry)
382 struct io_uring_cqe {
383 __u64 user_data; /* sqe->data submission passed back */
384 __s32 res; /* result code for this event */
388 * If the ring is initialized with IORING_SETUP_CQE32, then this field
389 * contains 16-bytes of padding, doubling the size of the CQE.
397 * IORING_CQE_F_BUFFER If set, the upper 16 bits are the buffer ID
398 * IORING_CQE_F_MORE If set, parent SQE will generate more CQE entries
399 * IORING_CQE_F_SOCK_NONEMPTY If set, more data to read after socket recv
400 * IORING_CQE_F_NOTIF Set for notification CQEs. Can be used to distinct
403 #define IORING_CQE_F_BUFFER (1U << 0)
404 #define IORING_CQE_F_MORE (1U << 1)
405 #define IORING_CQE_F_SOCK_NONEMPTY (1U << 2)
406 #define IORING_CQE_F_NOTIF (1U << 3)
409 IORING_CQE_BUFFER_SHIFT = 16,
413 * Magic offsets for the application to mmap the data it needs
415 #define IORING_OFF_SQ_RING 0ULL
416 #define IORING_OFF_CQ_RING 0x8000000ULL
417 #define IORING_OFF_SQES 0x10000000ULL
418 #define IORING_OFF_PBUF_RING 0x80000000ULL
419 #define IORING_OFF_PBUF_SHIFT 16
420 #define IORING_OFF_MMAP_MASK 0xf8000000ULL
423 * Filled with the offset for mmap(2)
425 struct io_sqring_offsets {
440 #define IORING_SQ_NEED_WAKEUP (1U << 0) /* needs io_uring_enter wakeup */
441 #define IORING_SQ_CQ_OVERFLOW (1U << 1) /* CQ ring is overflown */
442 #define IORING_SQ_TASKRUN (1U << 2) /* task should enter the kernel */
444 struct io_cqring_offsets {
460 /* disable eventfd notifications */
461 #define IORING_CQ_EVENTFD_DISABLED (1U << 0)
464 * io_uring_enter(2) flags
466 #define IORING_ENTER_GETEVENTS (1U << 0)
467 #define IORING_ENTER_SQ_WAKEUP (1U << 1)
468 #define IORING_ENTER_SQ_WAIT (1U << 2)
469 #define IORING_ENTER_EXT_ARG (1U << 3)
470 #define IORING_ENTER_REGISTERED_RING (1U << 4)
473 * Passed in for io_uring_setup(2). Copied back with updated info on success
475 struct io_uring_params {
480 __u32 sq_thread_idle;
484 struct io_sqring_offsets sq_off;
485 struct io_cqring_offsets cq_off;
489 * io_uring_params->features flags
491 #define IORING_FEAT_SINGLE_MMAP (1U << 0)
492 #define IORING_FEAT_NODROP (1U << 1)
493 #define IORING_FEAT_SUBMIT_STABLE (1U << 2)
494 #define IORING_FEAT_RW_CUR_POS (1U << 3)
495 #define IORING_FEAT_CUR_PERSONALITY (1U << 4)
496 #define IORING_FEAT_FAST_POLL (1U << 5)
497 #define IORING_FEAT_POLL_32BITS (1U << 6)
498 #define IORING_FEAT_SQPOLL_NONFIXED (1U << 7)
499 #define IORING_FEAT_EXT_ARG (1U << 8)
500 #define IORING_FEAT_NATIVE_WORKERS (1U << 9)
501 #define IORING_FEAT_RSRC_TAGS (1U << 10)
502 #define IORING_FEAT_CQE_SKIP (1U << 11)
503 #define IORING_FEAT_LINKED_FILE (1U << 12)
504 #define IORING_FEAT_REG_REG_RING (1U << 13)
507 * io_uring_register(2) opcodes and arguments
510 IORING_REGISTER_BUFFERS = 0,
511 IORING_UNREGISTER_BUFFERS = 1,
512 IORING_REGISTER_FILES = 2,
513 IORING_UNREGISTER_FILES = 3,
514 IORING_REGISTER_EVENTFD = 4,
515 IORING_UNREGISTER_EVENTFD = 5,
516 IORING_REGISTER_FILES_UPDATE = 6,
517 IORING_REGISTER_EVENTFD_ASYNC = 7,
518 IORING_REGISTER_PROBE = 8,
519 IORING_REGISTER_PERSONALITY = 9,
520 IORING_UNREGISTER_PERSONALITY = 10,
521 IORING_REGISTER_RESTRICTIONS = 11,
522 IORING_REGISTER_ENABLE_RINGS = 12,
524 /* extended with tagging */
525 IORING_REGISTER_FILES2 = 13,
526 IORING_REGISTER_FILES_UPDATE2 = 14,
527 IORING_REGISTER_BUFFERS2 = 15,
528 IORING_REGISTER_BUFFERS_UPDATE = 16,
530 /* set/clear io-wq thread affinities */
531 IORING_REGISTER_IOWQ_AFF = 17,
532 IORING_UNREGISTER_IOWQ_AFF = 18,
534 /* set/get max number of io-wq workers */
535 IORING_REGISTER_IOWQ_MAX_WORKERS = 19,
537 /* register/unregister io_uring fd with the ring */
538 IORING_REGISTER_RING_FDS = 20,
539 IORING_UNREGISTER_RING_FDS = 21,
541 /* register ring based provide buffer group */
542 IORING_REGISTER_PBUF_RING = 22,
543 IORING_UNREGISTER_PBUF_RING = 23,
545 /* sync cancelation API */
546 IORING_REGISTER_SYNC_CANCEL = 24,
548 /* register a range of fixed file slots for automatic slot allocation */
549 IORING_REGISTER_FILE_ALLOC_RANGE = 25,
552 IORING_REGISTER_LAST,
554 /* flag added to the opcode to use a registered ring fd */
555 IORING_REGISTER_USE_REGISTERED_RING = 1U << 31
558 /* io-wq worker categories */
564 /* deprecated, see struct io_uring_rsrc_update */
565 struct io_uring_files_update {
568 __aligned_u64 /* __s32 * */ fds;
572 * Register a fully sparse file space, rather than pass in an array of all
573 * -1 file descriptors.
575 #define IORING_RSRC_REGISTER_SPARSE (1U << 0)
577 struct io_uring_rsrc_register {
585 struct io_uring_rsrc_update {
591 struct io_uring_rsrc_update2 {
600 /* Skip updating fd indexes set to this value in the fd table */
601 #define IORING_REGISTER_FILES_SKIP (-2)
603 #define IO_URING_OP_SUPPORTED (1U << 0)
605 struct io_uring_probe_op {
608 __u16 flags; /* IO_URING_OP_* flags */
612 struct io_uring_probe {
613 __u8 last_op; /* last opcode supported */
614 __u8 ops_len; /* length of ops[] array below */
617 struct io_uring_probe_op ops[];
620 struct io_uring_restriction {
623 __u8 register_op; /* IORING_RESTRICTION_REGISTER_OP */
624 __u8 sqe_op; /* IORING_RESTRICTION_SQE_OP */
625 __u8 sqe_flags; /* IORING_RESTRICTION_SQE_FLAGS_* */
631 struct io_uring_buf {
638 struct io_uring_buf_ring {
641 * To avoid spilling into more pages than we need to, the
642 * ring tail is overlaid with the io_uring_buf->resv field.
650 __DECLARE_FLEX_ARRAY(struct io_uring_buf, bufs);
655 * Flags for IORING_REGISTER_PBUF_RING.
657 * IOU_PBUF_RING_MMAP: If set, kernel will allocate the memory for the ring.
658 * The application must not set a ring_addr in struct
659 * io_uring_buf_reg, instead it must subsequently call
660 * mmap(2) with the offset set as:
661 * IORING_OFF_PBUF_RING | (bgid << IORING_OFF_PBUF_SHIFT)
662 * to get a virtual mapping for the ring.
665 IOU_PBUF_RING_MMAP = 1,
668 /* argument for IORING_(UN)REGISTER_PBUF_RING */
669 struct io_uring_buf_reg {
678 * io_uring_restriction->opcode values
681 /* Allow an io_uring_register(2) opcode */
682 IORING_RESTRICTION_REGISTER_OP = 0,
684 /* Allow an sqe opcode */
685 IORING_RESTRICTION_SQE_OP = 1,
687 /* Allow sqe flags */
688 IORING_RESTRICTION_SQE_FLAGS_ALLOWED = 2,
690 /* Require sqe flags (these flags must be set on each submission) */
691 IORING_RESTRICTION_SQE_FLAGS_REQUIRED = 3,
693 IORING_RESTRICTION_LAST
696 struct io_uring_getevents_arg {
704 * Argument for IORING_REGISTER_SYNC_CANCEL
706 struct io_uring_sync_cancel_reg {
710 struct __kernel_timespec timeout;
717 * Argument for IORING_REGISTER_FILE_ALLOC_RANGE
718 * The range is specified as [off, off + len)
720 struct io_uring_file_index_range {
726 struct io_uring_recvmsg_out {
734 * Argument for IORING_OP_URING_CMD when file is a socket
737 SOCKET_URING_OP_SIOCINQ = 0,
738 SOCKET_URING_OP_SIOCOUTQ,