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;
71 __u64 user_data; /* data to be passed back at completion time */
72 /* pack this to avoid bogus arm OABI complaints */
74 /* index into fixed buffers, if used */
76 /* for grouped buffer selection */
78 } __attribute__((packed));
79 /* personality to use, if used */
95 * If the ring is initialized with IORING_SETUP_SQE128, then
96 * this field is used for 80 bytes of arbitrary command data
103 * If sqe->file_index is set to this for opcodes that instantiate a new
104 * direct descriptor (like openat/openat2/accept), then io_uring will allocate
105 * an available direct descriptor instead of having the application pass one
106 * in. The picked direct descriptor will be returned in cqe->res, or -ENFILE
107 * if the space is full.
109 #define IORING_FILE_INDEX_ALLOC (~0U)
112 IOSQE_FIXED_FILE_BIT,
115 IOSQE_IO_HARDLINK_BIT,
117 IOSQE_BUFFER_SELECT_BIT,
118 IOSQE_CQE_SKIP_SUCCESS_BIT,
124 /* use fixed fileset */
125 #define IOSQE_FIXED_FILE (1U << IOSQE_FIXED_FILE_BIT)
126 /* issue after inflight IO */
127 #define IOSQE_IO_DRAIN (1U << IOSQE_IO_DRAIN_BIT)
129 #define IOSQE_IO_LINK (1U << IOSQE_IO_LINK_BIT)
130 /* like LINK, but stronger */
131 #define IOSQE_IO_HARDLINK (1U << IOSQE_IO_HARDLINK_BIT)
132 /* always go async */
133 #define IOSQE_ASYNC (1U << IOSQE_ASYNC_BIT)
134 /* select buffer from sqe->buf_group */
135 #define IOSQE_BUFFER_SELECT (1U << IOSQE_BUFFER_SELECT_BIT)
136 /* don't post CQE if request succeeded */
137 #define IOSQE_CQE_SKIP_SUCCESS (1U << IOSQE_CQE_SKIP_SUCCESS_BIT)
140 * io_uring_setup() flags
142 #define IORING_SETUP_IOPOLL (1U << 0) /* io_context is polled */
143 #define IORING_SETUP_SQPOLL (1U << 1) /* SQ poll thread */
144 #define IORING_SETUP_SQ_AFF (1U << 2) /* sq_thread_cpu is valid */
145 #define IORING_SETUP_CQSIZE (1U << 3) /* app defines CQ size */
146 #define IORING_SETUP_CLAMP (1U << 4) /* clamp SQ/CQ ring sizes */
147 #define IORING_SETUP_ATTACH_WQ (1U << 5) /* attach to existing wq */
148 #define IORING_SETUP_R_DISABLED (1U << 6) /* start with ring disabled */
149 #define IORING_SETUP_SUBMIT_ALL (1U << 7) /* continue submit on error */
151 * Cooperative task running. When requests complete, they often require
152 * forcing the submitter to transition to the kernel to complete. If this
153 * flag is set, work will be done when the task transitions anyway, rather
154 * than force an inter-processor interrupt reschedule. This avoids interrupting
155 * a task running in userspace, and saves an IPI.
157 #define IORING_SETUP_COOP_TASKRUN (1U << 8)
159 * If COOP_TASKRUN is set, get notified if task work is available for
160 * running and a kernel transition would be needed to run it. This sets
161 * IORING_SQ_TASKRUN in the sq ring flags. Not valid with COOP_TASKRUN.
163 #define IORING_SETUP_TASKRUN_FLAG (1U << 9)
164 #define IORING_SETUP_SQE128 (1U << 10) /* SQEs are 128 byte */
165 #define IORING_SETUP_CQE32 (1U << 11) /* CQEs are 32 byte */
167 * Only one task is allowed to submit requests
169 #define IORING_SETUP_SINGLE_ISSUER (1U << 12)
172 * Defer running task work to get events.
173 * Rather than running bits of task work whenever the task transitions
174 * try to do it just before it is needed.
176 #define IORING_SETUP_DEFER_TASKRUN (1U << 13)
179 * Application provides the memory for the rings
181 #define IORING_SETUP_NO_MMAP (1U << 14)
184 * Register the ring fd in itself for use with
185 * IORING_REGISTER_USE_REGISTERED_RING; return a registered fd index rather
188 #define IORING_SETUP_REGISTERED_FD_ONLY (1U << 15)
191 * Removes indirection through the SQ index array.
193 #define IORING_SETUP_NO_SQARRAY (1U << 16)
200 IORING_OP_READ_FIXED,
201 IORING_OP_WRITE_FIXED,
203 IORING_OP_POLL_REMOVE,
204 IORING_OP_SYNC_FILE_RANGE,
208 IORING_OP_TIMEOUT_REMOVE,
210 IORING_OP_ASYNC_CANCEL,
211 IORING_OP_LINK_TIMEOUT,
216 IORING_OP_FILES_UPDATE,
227 IORING_OP_PROVIDE_BUFFERS,
228 IORING_OP_REMOVE_BUFFERS,
244 IORING_OP_SENDMSG_ZC,
245 IORING_OP_READ_MULTISHOT,
247 IORING_OP_FUTEX_WAIT,
248 IORING_OP_FUTEX_WAKE,
250 /* this goes last, obviously */
255 * sqe->uring_cmd_flags top 8bits aren't available for userspace
256 * IORING_URING_CMD_FIXED use registered buffer; pass this flag
257 * along with setting sqe->buf_index.
259 #define IORING_URING_CMD_FIXED (1U << 0)
260 #define IORING_URING_CMD_MASK IORING_URING_CMD_FIXED
266 #define IORING_FSYNC_DATASYNC (1U << 0)
271 #define IORING_TIMEOUT_ABS (1U << 0)
272 #define IORING_TIMEOUT_UPDATE (1U << 1)
273 #define IORING_TIMEOUT_BOOTTIME (1U << 2)
274 #define IORING_TIMEOUT_REALTIME (1U << 3)
275 #define IORING_LINK_TIMEOUT_UPDATE (1U << 4)
276 #define IORING_TIMEOUT_ETIME_SUCCESS (1U << 5)
277 #define IORING_TIMEOUT_MULTISHOT (1U << 6)
278 #define IORING_TIMEOUT_CLOCK_MASK (IORING_TIMEOUT_BOOTTIME | IORING_TIMEOUT_REALTIME)
279 #define IORING_TIMEOUT_UPDATE_MASK (IORING_TIMEOUT_UPDATE | IORING_LINK_TIMEOUT_UPDATE)
282 * extends splice(2) flags
284 #define SPLICE_F_FD_IN_FIXED (1U << 31) /* the last bit of __u32 */
287 * POLL_ADD flags. Note that since sqe->poll_events is the flag space, the
288 * command flags for POLL_ADD are stored in sqe->len.
290 * IORING_POLL_ADD_MULTI Multishot poll. Sets IORING_CQE_F_MORE if
291 * the poll handler will continue to report
292 * CQEs on behalf of the same SQE.
294 * IORING_POLL_UPDATE Update existing poll request, matching
295 * sqe->addr as the old user_data field.
297 * IORING_POLL_LEVEL Level triggered poll.
299 #define IORING_POLL_ADD_MULTI (1U << 0)
300 #define IORING_POLL_UPDATE_EVENTS (1U << 1)
301 #define IORING_POLL_UPDATE_USER_DATA (1U << 2)
302 #define IORING_POLL_ADD_LEVEL (1U << 3)
305 * ASYNC_CANCEL flags.
307 * IORING_ASYNC_CANCEL_ALL Cancel all requests that match the given key
308 * IORING_ASYNC_CANCEL_FD Key off 'fd' for cancelation rather than the
309 * request 'user_data'
310 * IORING_ASYNC_CANCEL_ANY Match any request
311 * IORING_ASYNC_CANCEL_FD_FIXED 'fd' passed in is a fixed descriptor
312 * IORING_ASYNC_CANCEL_USERDATA Match on user_data, default for no other key
313 * IORING_ASYNC_CANCEL_OP Match request based on opcode
315 #define IORING_ASYNC_CANCEL_ALL (1U << 0)
316 #define IORING_ASYNC_CANCEL_FD (1U << 1)
317 #define IORING_ASYNC_CANCEL_ANY (1U << 2)
318 #define IORING_ASYNC_CANCEL_FD_FIXED (1U << 3)
319 #define IORING_ASYNC_CANCEL_USERDATA (1U << 4)
320 #define IORING_ASYNC_CANCEL_OP (1U << 5)
323 * send/sendmsg and recv/recvmsg flags (sqe->ioprio)
325 * IORING_RECVSEND_POLL_FIRST If set, instead of first attempting to send
326 * or receive and arm poll if that yields an
327 * -EAGAIN result, arm poll upfront and skip
328 * the initial transfer attempt.
330 * IORING_RECV_MULTISHOT Multishot recv. Sets IORING_CQE_F_MORE if
331 * the handler will continue to report
332 * CQEs on behalf of the same SQE.
334 * IORING_RECVSEND_FIXED_BUF Use registered buffers, the index is stored in
335 * the buf_index field.
337 * IORING_SEND_ZC_REPORT_USAGE
338 * If set, SEND[MSG]_ZC should report
339 * the zerocopy usage in cqe.res
340 * for the IORING_CQE_F_NOTIF cqe.
341 * 0 is reported if zerocopy was actually possible.
342 * IORING_NOTIF_USAGE_ZC_COPIED if data was copied
343 * (at least partially).
345 #define IORING_RECVSEND_POLL_FIRST (1U << 0)
346 #define IORING_RECV_MULTISHOT (1U << 1)
347 #define IORING_RECVSEND_FIXED_BUF (1U << 2)
348 #define IORING_SEND_ZC_REPORT_USAGE (1U << 3)
351 * cqe.res for IORING_CQE_F_NOTIF if
352 * IORING_SEND_ZC_REPORT_USAGE was requested
354 * It should be treated as a flag, all other
355 * bits of cqe.res should be treated as reserved!
357 #define IORING_NOTIF_USAGE_ZC_COPIED (1U << 31)
360 * accept flags stored in sqe->ioprio
362 #define IORING_ACCEPT_MULTISHOT (1U << 0)
365 * IORING_OP_MSG_RING command types, stored in sqe->addr
368 IORING_MSG_DATA, /* pass sqe->len as 'res' and off as user_data */
369 IORING_MSG_SEND_FD, /* send a registered fd to another ring */
373 * IORING_OP_MSG_RING flags (sqe->msg_ring_flags)
375 * IORING_MSG_RING_CQE_SKIP Don't post a CQE to the target ring. Not
376 * applicable for IORING_MSG_DATA, obviously.
378 #define IORING_MSG_RING_CQE_SKIP (1U << 0)
379 /* Pass through the flags from sqe->file_index to cqe->flags */
380 #define IORING_MSG_RING_FLAGS_PASS (1U << 1)
383 * IO completion data structure (Completion Queue Entry)
385 struct io_uring_cqe {
386 __u64 user_data; /* sqe->data submission passed back */
387 __s32 res; /* result code for this event */
391 * If the ring is initialized with IORING_SETUP_CQE32, then this field
392 * contains 16-bytes of padding, doubling the size of the CQE.
400 * IORING_CQE_F_BUFFER If set, the upper 16 bits are the buffer ID
401 * IORING_CQE_F_MORE If set, parent SQE will generate more CQE entries
402 * IORING_CQE_F_SOCK_NONEMPTY If set, more data to read after socket recv
403 * IORING_CQE_F_NOTIF Set for notification CQEs. Can be used to distinct
406 #define IORING_CQE_F_BUFFER (1U << 0)
407 #define IORING_CQE_F_MORE (1U << 1)
408 #define IORING_CQE_F_SOCK_NONEMPTY (1U << 2)
409 #define IORING_CQE_F_NOTIF (1U << 3)
412 IORING_CQE_BUFFER_SHIFT = 16,
416 * Magic offsets for the application to mmap the data it needs
418 #define IORING_OFF_SQ_RING 0ULL
419 #define IORING_OFF_CQ_RING 0x8000000ULL
420 #define IORING_OFF_SQES 0x10000000ULL
421 #define IORING_OFF_PBUF_RING 0x80000000ULL
422 #define IORING_OFF_PBUF_SHIFT 16
423 #define IORING_OFF_MMAP_MASK 0xf8000000ULL
426 * Filled with the offset for mmap(2)
428 struct io_sqring_offsets {
443 #define IORING_SQ_NEED_WAKEUP (1U << 0) /* needs io_uring_enter wakeup */
444 #define IORING_SQ_CQ_OVERFLOW (1U << 1) /* CQ ring is overflown */
445 #define IORING_SQ_TASKRUN (1U << 2) /* task should enter the kernel */
447 struct io_cqring_offsets {
463 /* disable eventfd notifications */
464 #define IORING_CQ_EVENTFD_DISABLED (1U << 0)
467 * io_uring_enter(2) flags
469 #define IORING_ENTER_GETEVENTS (1U << 0)
470 #define IORING_ENTER_SQ_WAKEUP (1U << 1)
471 #define IORING_ENTER_SQ_WAIT (1U << 2)
472 #define IORING_ENTER_EXT_ARG (1U << 3)
473 #define IORING_ENTER_REGISTERED_RING (1U << 4)
476 * Passed in for io_uring_setup(2). Copied back with updated info on success
478 struct io_uring_params {
483 __u32 sq_thread_idle;
487 struct io_sqring_offsets sq_off;
488 struct io_cqring_offsets cq_off;
492 * io_uring_params->features flags
494 #define IORING_FEAT_SINGLE_MMAP (1U << 0)
495 #define IORING_FEAT_NODROP (1U << 1)
496 #define IORING_FEAT_SUBMIT_STABLE (1U << 2)
497 #define IORING_FEAT_RW_CUR_POS (1U << 3)
498 #define IORING_FEAT_CUR_PERSONALITY (1U << 4)
499 #define IORING_FEAT_FAST_POLL (1U << 5)
500 #define IORING_FEAT_POLL_32BITS (1U << 6)
501 #define IORING_FEAT_SQPOLL_NONFIXED (1U << 7)
502 #define IORING_FEAT_EXT_ARG (1U << 8)
503 #define IORING_FEAT_NATIVE_WORKERS (1U << 9)
504 #define IORING_FEAT_RSRC_TAGS (1U << 10)
505 #define IORING_FEAT_CQE_SKIP (1U << 11)
506 #define IORING_FEAT_LINKED_FILE (1U << 12)
507 #define IORING_FEAT_REG_REG_RING (1U << 13)
510 * io_uring_register(2) opcodes and arguments
513 IORING_REGISTER_BUFFERS = 0,
514 IORING_UNREGISTER_BUFFERS = 1,
515 IORING_REGISTER_FILES = 2,
516 IORING_UNREGISTER_FILES = 3,
517 IORING_REGISTER_EVENTFD = 4,
518 IORING_UNREGISTER_EVENTFD = 5,
519 IORING_REGISTER_FILES_UPDATE = 6,
520 IORING_REGISTER_EVENTFD_ASYNC = 7,
521 IORING_REGISTER_PROBE = 8,
522 IORING_REGISTER_PERSONALITY = 9,
523 IORING_UNREGISTER_PERSONALITY = 10,
524 IORING_REGISTER_RESTRICTIONS = 11,
525 IORING_REGISTER_ENABLE_RINGS = 12,
527 /* extended with tagging */
528 IORING_REGISTER_FILES2 = 13,
529 IORING_REGISTER_FILES_UPDATE2 = 14,
530 IORING_REGISTER_BUFFERS2 = 15,
531 IORING_REGISTER_BUFFERS_UPDATE = 16,
533 /* set/clear io-wq thread affinities */
534 IORING_REGISTER_IOWQ_AFF = 17,
535 IORING_UNREGISTER_IOWQ_AFF = 18,
537 /* set/get max number of io-wq workers */
538 IORING_REGISTER_IOWQ_MAX_WORKERS = 19,
540 /* register/unregister io_uring fd with the ring */
541 IORING_REGISTER_RING_FDS = 20,
542 IORING_UNREGISTER_RING_FDS = 21,
544 /* register ring based provide buffer group */
545 IORING_REGISTER_PBUF_RING = 22,
546 IORING_UNREGISTER_PBUF_RING = 23,
548 /* sync cancelation API */
549 IORING_REGISTER_SYNC_CANCEL = 24,
551 /* register a range of fixed file slots for automatic slot allocation */
552 IORING_REGISTER_FILE_ALLOC_RANGE = 25,
555 IORING_REGISTER_LAST,
557 /* flag added to the opcode to use a registered ring fd */
558 IORING_REGISTER_USE_REGISTERED_RING = 1U << 31
561 /* io-wq worker categories */
567 /* deprecated, see struct io_uring_rsrc_update */
568 struct io_uring_files_update {
571 __aligned_u64 /* __s32 * */ fds;
575 * Register a fully sparse file space, rather than pass in an array of all
576 * -1 file descriptors.
578 #define IORING_RSRC_REGISTER_SPARSE (1U << 0)
580 struct io_uring_rsrc_register {
588 struct io_uring_rsrc_update {
594 struct io_uring_rsrc_update2 {
603 /* Skip updating fd indexes set to this value in the fd table */
604 #define IORING_REGISTER_FILES_SKIP (-2)
606 #define IO_URING_OP_SUPPORTED (1U << 0)
608 struct io_uring_probe_op {
611 __u16 flags; /* IO_URING_OP_* flags */
615 struct io_uring_probe {
616 __u8 last_op; /* last opcode supported */
617 __u8 ops_len; /* length of ops[] array below */
620 struct io_uring_probe_op ops[];
623 struct io_uring_restriction {
626 __u8 register_op; /* IORING_RESTRICTION_REGISTER_OP */
627 __u8 sqe_op; /* IORING_RESTRICTION_SQE_OP */
628 __u8 sqe_flags; /* IORING_RESTRICTION_SQE_FLAGS_* */
634 struct io_uring_buf {
641 struct io_uring_buf_ring {
644 * To avoid spilling into more pages than we need to, the
645 * ring tail is overlaid with the io_uring_buf->resv field.
653 __DECLARE_FLEX_ARRAY(struct io_uring_buf, bufs);
658 * Flags for IORING_REGISTER_PBUF_RING.
660 * IOU_PBUF_RING_MMAP: If set, kernel will allocate the memory for the ring.
661 * The application must not set a ring_addr in struct
662 * io_uring_buf_reg, instead it must subsequently call
663 * mmap(2) with the offset set as:
664 * IORING_OFF_PBUF_RING | (bgid << IORING_OFF_PBUF_SHIFT)
665 * to get a virtual mapping for the ring.
668 IOU_PBUF_RING_MMAP = 1,
671 /* argument for IORING_(UN)REGISTER_PBUF_RING */
672 struct io_uring_buf_reg {
681 * io_uring_restriction->opcode values
684 /* Allow an io_uring_register(2) opcode */
685 IORING_RESTRICTION_REGISTER_OP = 0,
687 /* Allow an sqe opcode */
688 IORING_RESTRICTION_SQE_OP = 1,
690 /* Allow sqe flags */
691 IORING_RESTRICTION_SQE_FLAGS_ALLOWED = 2,
693 /* Require sqe flags (these flags must be set on each submission) */
694 IORING_RESTRICTION_SQE_FLAGS_REQUIRED = 3,
696 IORING_RESTRICTION_LAST
699 struct io_uring_getevents_arg {
707 * Argument for IORING_REGISTER_SYNC_CANCEL
709 struct io_uring_sync_cancel_reg {
713 struct __kernel_timespec timeout;
720 * Argument for IORING_REGISTER_FILE_ALLOC_RANGE
721 * The range is specified as [off, off + len)
723 struct io_uring_file_index_range {
729 struct io_uring_recvmsg_out {
737 * Argument for IORING_OP_URING_CMD when file is a socket
740 SOCKET_URING_OP_SIOCINQ = 0,
741 SOCKET_URING_OP_SIOCOUTQ,