Linux kernel mirror (for testing)
git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git
kernel
os
linux
1/* SPDX-License-Identifier: (GPL-2.0 WITH Linux-syscall-note) OR MIT */
2/*
3 * Header file for the io_uring interface.
4 *
5 * Copyright (C) 2019 Jens Axboe
6 * Copyright (C) 2019 Christoph Hellwig
7 */
8#ifndef LINUX_IO_URING_H
9#define LINUX_IO_URING_H
10
11#include <linux/fs.h>
12#include <linux/types.h>
13/*
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
18 */
19#ifndef UAPI_LINUX_IO_URING_H_SKIP_LINUX_TIME_TYPES_H
20#include <linux/time_types.h>
21#endif
22
23#ifdef __cplusplus
24extern "C" {
25#endif
26
27/*
28 * IO submission data structure (Submission Queue Entry)
29 */
30struct io_uring_sqe {
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 */
35 union {
36 __u64 off; /* offset into file */
37 __u64 addr2;
38 struct {
39 __u32 cmd_op;
40 __u32 __pad1;
41 };
42 };
43 union {
44 __u64 addr; /* pointer to buffer or iovecs */
45 __u64 splice_off_in;
46 struct {
47 __u32 level;
48 __u32 optname;
49 };
50 };
51 __u32 len; /* buffer size or number of iovecs */
52 union {
53 __u32 rw_flags;
54 __u32 fsync_flags;
55 __u16 poll_events; /* compatibility */
56 __u32 poll32_events; /* word-reversed for BE */
57 __u32 sync_range_flags;
58 __u32 msg_flags;
59 __u32 timeout_flags;
60 __u32 accept_flags;
61 __u32 cancel_flags;
62 __u32 open_flags;
63 __u32 statx_flags;
64 __u32 fadvise_advice;
65 __u32 splice_flags;
66 __u32 rename_flags;
67 __u32 unlink_flags;
68 __u32 hardlink_flags;
69 __u32 xattr_flags;
70 __u32 msg_ring_flags;
71 __u32 uring_cmd_flags;
72 __u32 waitid_flags;
73 __u32 futex_flags;
74 __u32 install_fd_flags;
75 __u32 nop_flags;
76 __u32 pipe_flags;
77 };
78 __u64 user_data; /* data to be passed back at completion time */
79 /* pack this to avoid bogus arm OABI complaints */
80 union {
81 /* index into fixed buffers, if used */
82 __u16 buf_index;
83 /* for grouped buffer selection */
84 __u16 buf_group;
85 } __attribute__((packed));
86 /* personality to use, if used */
87 __u16 personality;
88 union {
89 __s32 splice_fd_in;
90 __u32 file_index;
91 __u32 zcrx_ifq_idx;
92 __u32 optlen;
93 struct {
94 __u16 addr_len;
95 __u16 __pad3[1];
96 };
97 struct {
98 __u8 write_stream;
99 __u8 __pad4[3];
100 };
101 };
102 union {
103 struct {
104 __u64 addr3;
105 __u64 __pad2[1];
106 };
107 struct {
108 __u64 attr_ptr; /* pointer to attribute information */
109 __u64 attr_type_mask; /* bit mask of attributes */
110 };
111 __u64 optval;
112 /*
113 * If the ring is initialized with IORING_SETUP_SQE128, then
114 * this field is used for 80 bytes of arbitrary command data
115 */
116 __u8 cmd[0];
117 };
118};
119
120/* sqe->attr_type_mask flags */
121#define IORING_RW_ATTR_FLAG_PI (1U << 0)
122/* PI attribute information */
123struct io_uring_attr_pi {
124 __u16 flags;
125 __u16 app_tag;
126 __u32 len;
127 __u64 addr;
128 __u64 seed;
129 __u64 rsvd;
130};
131
132/*
133 * If sqe->file_index is set to this for opcodes that instantiate a new
134 * direct descriptor (like openat/openat2/accept), then io_uring will allocate
135 * an available direct descriptor instead of having the application pass one
136 * in. The picked direct descriptor will be returned in cqe->res, or -ENFILE
137 * if the space is full.
138 */
139#define IORING_FILE_INDEX_ALLOC (~0U)
140
141enum io_uring_sqe_flags_bit {
142 IOSQE_FIXED_FILE_BIT,
143 IOSQE_IO_DRAIN_BIT,
144 IOSQE_IO_LINK_BIT,
145 IOSQE_IO_HARDLINK_BIT,
146 IOSQE_ASYNC_BIT,
147 IOSQE_BUFFER_SELECT_BIT,
148 IOSQE_CQE_SKIP_SUCCESS_BIT,
149};
150
151/*
152 * sqe->flags
153 */
154/* use fixed fileset */
155#define IOSQE_FIXED_FILE (1U << IOSQE_FIXED_FILE_BIT)
156/* issue after inflight IO */
157#define IOSQE_IO_DRAIN (1U << IOSQE_IO_DRAIN_BIT)
158/* links next sqe */
159#define IOSQE_IO_LINK (1U << IOSQE_IO_LINK_BIT)
160/* like LINK, but stronger */
161#define IOSQE_IO_HARDLINK (1U << IOSQE_IO_HARDLINK_BIT)
162/* always go async */
163#define IOSQE_ASYNC (1U << IOSQE_ASYNC_BIT)
164/* select buffer from sqe->buf_group */
165#define IOSQE_BUFFER_SELECT (1U << IOSQE_BUFFER_SELECT_BIT)
166/* don't post CQE if request succeeded */
167#define IOSQE_CQE_SKIP_SUCCESS (1U << IOSQE_CQE_SKIP_SUCCESS_BIT)
168
169/*
170 * io_uring_setup() flags
171 */
172#define IORING_SETUP_IOPOLL (1U << 0) /* io_context is polled */
173#define IORING_SETUP_SQPOLL (1U << 1) /* SQ poll thread */
174#define IORING_SETUP_SQ_AFF (1U << 2) /* sq_thread_cpu is valid */
175#define IORING_SETUP_CQSIZE (1U << 3) /* app defines CQ size */
176#define IORING_SETUP_CLAMP (1U << 4) /* clamp SQ/CQ ring sizes */
177#define IORING_SETUP_ATTACH_WQ (1U << 5) /* attach to existing wq */
178#define IORING_SETUP_R_DISABLED (1U << 6) /* start with ring disabled */
179#define IORING_SETUP_SUBMIT_ALL (1U << 7) /* continue submit on error */
180/*
181 * Cooperative task running. When requests complete, they often require
182 * forcing the submitter to transition to the kernel to complete. If this
183 * flag is set, work will be done when the task transitions anyway, rather
184 * than force an inter-processor interrupt reschedule. This avoids interrupting
185 * a task running in userspace, and saves an IPI.
186 */
187#define IORING_SETUP_COOP_TASKRUN (1U << 8)
188/*
189 * If COOP_TASKRUN is set, get notified if task work is available for
190 * running and a kernel transition would be needed to run it. This sets
191 * IORING_SQ_TASKRUN in the sq ring flags. Not valid without COOP_TASKRUN
192 * or DEFER_TASKRUN.
193 */
194#define IORING_SETUP_TASKRUN_FLAG (1U << 9)
195#define IORING_SETUP_SQE128 (1U << 10) /* SQEs are 128 byte */
196#define IORING_SETUP_CQE32 (1U << 11) /* CQEs are 32 byte */
197/*
198 * Only one task is allowed to submit requests
199 */
200#define IORING_SETUP_SINGLE_ISSUER (1U << 12)
201
202/*
203 * Defer running task work to get events.
204 * Rather than running bits of task work whenever the task transitions
205 * try to do it just before it is needed.
206 */
207#define IORING_SETUP_DEFER_TASKRUN (1U << 13)
208
209/*
210 * Application provides the memory for the rings
211 */
212#define IORING_SETUP_NO_MMAP (1U << 14)
213
214/*
215 * Register the ring fd in itself for use with
216 * IORING_REGISTER_USE_REGISTERED_RING; return a registered fd index rather
217 * than an fd.
218 */
219#define IORING_SETUP_REGISTERED_FD_ONLY (1U << 15)
220
221/*
222 * Removes indirection through the SQ index array.
223 */
224#define IORING_SETUP_NO_SQARRAY (1U << 16)
225
226/* Use hybrid poll in iopoll process */
227#define IORING_SETUP_HYBRID_IOPOLL (1U << 17)
228
229/*
230 * Allow both 16b and 32b CQEs. If a 32b CQE is posted, it will have
231 * IORING_CQE_F_32 set in cqe->flags.
232 */
233#define IORING_SETUP_CQE_MIXED (1U << 18)
234
235/*
236 * Allow both 64b and 128b SQEs. If a 128b SQE is posted, it will have
237 * a 128b opcode.
238 */
239#define IORING_SETUP_SQE_MIXED (1U << 19)
240
241/*
242 * When set, io_uring ignores SQ head and tail and fetches SQEs to submit
243 * starting from index 0 instead from the index stored in the head pointer.
244 * IOW, the user should place all SQE at the beginning of the SQ memory
245 * before issuing a submission syscall.
246 *
247 * It requires IORING_SETUP_NO_SQARRAY and is incompatible with
248 * IORING_SETUP_SQPOLL. The user must also never change the SQ head and tail
249 * values and keep it set to 0. Any other value is undefined behaviour.
250 */
251#define IORING_SETUP_SQ_REWIND (1U << 20)
252
253enum io_uring_op {
254 IORING_OP_NOP,
255 IORING_OP_READV,
256 IORING_OP_WRITEV,
257 IORING_OP_FSYNC,
258 IORING_OP_READ_FIXED,
259 IORING_OP_WRITE_FIXED,
260 IORING_OP_POLL_ADD,
261 IORING_OP_POLL_REMOVE,
262 IORING_OP_SYNC_FILE_RANGE,
263 IORING_OP_SENDMSG,
264 IORING_OP_RECVMSG,
265 IORING_OP_TIMEOUT,
266 IORING_OP_TIMEOUT_REMOVE,
267 IORING_OP_ACCEPT,
268 IORING_OP_ASYNC_CANCEL,
269 IORING_OP_LINK_TIMEOUT,
270 IORING_OP_CONNECT,
271 IORING_OP_FALLOCATE,
272 IORING_OP_OPENAT,
273 IORING_OP_CLOSE,
274 IORING_OP_FILES_UPDATE,
275 IORING_OP_STATX,
276 IORING_OP_READ,
277 IORING_OP_WRITE,
278 IORING_OP_FADVISE,
279 IORING_OP_MADVISE,
280 IORING_OP_SEND,
281 IORING_OP_RECV,
282 IORING_OP_OPENAT2,
283 IORING_OP_EPOLL_CTL,
284 IORING_OP_SPLICE,
285 IORING_OP_PROVIDE_BUFFERS,
286 IORING_OP_REMOVE_BUFFERS,
287 IORING_OP_TEE,
288 IORING_OP_SHUTDOWN,
289 IORING_OP_RENAMEAT,
290 IORING_OP_UNLINKAT,
291 IORING_OP_MKDIRAT,
292 IORING_OP_SYMLINKAT,
293 IORING_OP_LINKAT,
294 IORING_OP_MSG_RING,
295 IORING_OP_FSETXATTR,
296 IORING_OP_SETXATTR,
297 IORING_OP_FGETXATTR,
298 IORING_OP_GETXATTR,
299 IORING_OP_SOCKET,
300 IORING_OP_URING_CMD,
301 IORING_OP_SEND_ZC,
302 IORING_OP_SENDMSG_ZC,
303 IORING_OP_READ_MULTISHOT,
304 IORING_OP_WAITID,
305 IORING_OP_FUTEX_WAIT,
306 IORING_OP_FUTEX_WAKE,
307 IORING_OP_FUTEX_WAITV,
308 IORING_OP_FIXED_FD_INSTALL,
309 IORING_OP_FTRUNCATE,
310 IORING_OP_BIND,
311 IORING_OP_LISTEN,
312 IORING_OP_RECV_ZC,
313 IORING_OP_EPOLL_WAIT,
314 IORING_OP_READV_FIXED,
315 IORING_OP_WRITEV_FIXED,
316 IORING_OP_PIPE,
317 IORING_OP_NOP128,
318 IORING_OP_URING_CMD128,
319
320 /* this goes last, obviously */
321 IORING_OP_LAST,
322};
323
324/*
325 * sqe->uring_cmd_flags top 8bits aren't available for userspace
326 * IORING_URING_CMD_FIXED use registered buffer; pass this flag
327 * along with setting sqe->buf_index.
328 * IORING_URING_CMD_MULTISHOT must be used with buffer select, like other
329 * multishot commands. Not compatible with
330 * IORING_URING_CMD_FIXED, for now.
331 */
332#define IORING_URING_CMD_FIXED (1U << 0)
333#define IORING_URING_CMD_MULTISHOT (1U << 1)
334#define IORING_URING_CMD_MASK (IORING_URING_CMD_FIXED | IORING_URING_CMD_MULTISHOT)
335
336
337/*
338 * sqe->fsync_flags
339 */
340#define IORING_FSYNC_DATASYNC (1U << 0)
341
342/*
343 * sqe->timeout_flags
344 */
345#define IORING_TIMEOUT_ABS (1U << 0)
346#define IORING_TIMEOUT_UPDATE (1U << 1)
347#define IORING_TIMEOUT_BOOTTIME (1U << 2)
348#define IORING_TIMEOUT_REALTIME (1U << 3)
349#define IORING_LINK_TIMEOUT_UPDATE (1U << 4)
350#define IORING_TIMEOUT_ETIME_SUCCESS (1U << 5)
351#define IORING_TIMEOUT_MULTISHOT (1U << 6)
352#define IORING_TIMEOUT_CLOCK_MASK (IORING_TIMEOUT_BOOTTIME | IORING_TIMEOUT_REALTIME)
353#define IORING_TIMEOUT_UPDATE_MASK (IORING_TIMEOUT_UPDATE | IORING_LINK_TIMEOUT_UPDATE)
354/*
355 * sqe->splice_flags
356 * extends splice(2) flags
357 */
358#define SPLICE_F_FD_IN_FIXED (1U << 31) /* the last bit of __u32 */
359
360/*
361 * POLL_ADD flags. Note that since sqe->poll_events is the flag space, the
362 * command flags for POLL_ADD are stored in sqe->len.
363 *
364 * IORING_POLL_ADD_MULTI Multishot poll. Sets IORING_CQE_F_MORE if
365 * the poll handler will continue to report
366 * CQEs on behalf of the same SQE.
367 *
368 * IORING_POLL_UPDATE Update existing poll request, matching
369 * sqe->addr as the old user_data field.
370 *
371 * IORING_POLL_LEVEL Level triggered poll.
372 */
373#define IORING_POLL_ADD_MULTI (1U << 0)
374#define IORING_POLL_UPDATE_EVENTS (1U << 1)
375#define IORING_POLL_UPDATE_USER_DATA (1U << 2)
376#define IORING_POLL_ADD_LEVEL (1U << 3)
377
378/*
379 * ASYNC_CANCEL flags.
380 *
381 * IORING_ASYNC_CANCEL_ALL Cancel all requests that match the given key
382 * IORING_ASYNC_CANCEL_FD Key off 'fd' for cancelation rather than the
383 * request 'user_data'
384 * IORING_ASYNC_CANCEL_ANY Match any request
385 * IORING_ASYNC_CANCEL_FD_FIXED 'fd' passed in is a fixed descriptor
386 * IORING_ASYNC_CANCEL_USERDATA Match on user_data, default for no other key
387 * IORING_ASYNC_CANCEL_OP Match request based on opcode
388 */
389#define IORING_ASYNC_CANCEL_ALL (1U << 0)
390#define IORING_ASYNC_CANCEL_FD (1U << 1)
391#define IORING_ASYNC_CANCEL_ANY (1U << 2)
392#define IORING_ASYNC_CANCEL_FD_FIXED (1U << 3)
393#define IORING_ASYNC_CANCEL_USERDATA (1U << 4)
394#define IORING_ASYNC_CANCEL_OP (1U << 5)
395
396/*
397 * send/sendmsg and recv/recvmsg flags (sqe->ioprio)
398 *
399 * IORING_RECVSEND_POLL_FIRST If set, instead of first attempting to send
400 * or receive and arm poll if that yields an
401 * -EAGAIN result, arm poll upfront and skip
402 * the initial transfer attempt.
403 *
404 * IORING_RECV_MULTISHOT Multishot recv. Sets IORING_CQE_F_MORE if
405 * the handler will continue to report
406 * CQEs on behalf of the same SQE.
407 *
408 * IORING_RECVSEND_FIXED_BUF Use registered buffers, the index is stored in
409 * the buf_index field.
410 *
411 * IORING_SEND_ZC_REPORT_USAGE
412 * If set, SEND[MSG]_ZC should report
413 * the zerocopy usage in cqe.res
414 * for the IORING_CQE_F_NOTIF cqe.
415 * 0 is reported if zerocopy was actually possible.
416 * IORING_NOTIF_USAGE_ZC_COPIED if data was copied
417 * (at least partially).
418 *
419 * IORING_RECVSEND_BUNDLE Used with IOSQE_BUFFER_SELECT. If set, send or
420 * recv will grab as many buffers from the buffer
421 * group ID given and send them all. The completion
422 * result will be the number of buffers send, with
423 * the starting buffer ID in cqe->flags as per
424 * usual for provided buffer usage. The buffers
425 * will be contiguous from the starting buffer ID.
426 *
427 * IORING_SEND_VECTORIZED If set, SEND[_ZC] will take a pointer to a io_vec
428 * to allow vectorized send operations.
429 */
430#define IORING_RECVSEND_POLL_FIRST (1U << 0)
431#define IORING_RECV_MULTISHOT (1U << 1)
432#define IORING_RECVSEND_FIXED_BUF (1U << 2)
433#define IORING_SEND_ZC_REPORT_USAGE (1U << 3)
434#define IORING_RECVSEND_BUNDLE (1U << 4)
435#define IORING_SEND_VECTORIZED (1U << 5)
436
437/*
438 * cqe.res for IORING_CQE_F_NOTIF if
439 * IORING_SEND_ZC_REPORT_USAGE was requested
440 *
441 * It should be treated as a flag, all other
442 * bits of cqe.res should be treated as reserved!
443 */
444#define IORING_NOTIF_USAGE_ZC_COPIED (1U << 31)
445
446/*
447 * accept flags stored in sqe->ioprio
448 */
449#define IORING_ACCEPT_MULTISHOT (1U << 0)
450#define IORING_ACCEPT_DONTWAIT (1U << 1)
451#define IORING_ACCEPT_POLL_FIRST (1U << 2)
452
453/*
454 * IORING_OP_MSG_RING command types, stored in sqe->addr
455 */
456enum io_uring_msg_ring_flags {
457 IORING_MSG_DATA, /* pass sqe->len as 'res' and off as user_data */
458 IORING_MSG_SEND_FD, /* send a registered fd to another ring */
459};
460
461/*
462 * IORING_OP_MSG_RING flags (sqe->msg_ring_flags)
463 *
464 * IORING_MSG_RING_CQE_SKIP Don't post a CQE to the target ring. Not
465 * applicable for IORING_MSG_DATA, obviously.
466 */
467#define IORING_MSG_RING_CQE_SKIP (1U << 0)
468/* Pass through the flags from sqe->file_index to cqe->flags */
469#define IORING_MSG_RING_FLAGS_PASS (1U << 1)
470
471/*
472 * IORING_OP_FIXED_FD_INSTALL flags (sqe->install_fd_flags)
473 *
474 * IORING_FIXED_FD_NO_CLOEXEC Don't mark the fd as O_CLOEXEC
475 */
476#define IORING_FIXED_FD_NO_CLOEXEC (1U << 0)
477
478/*
479 * IORING_OP_NOP flags (sqe->nop_flags)
480 *
481 * IORING_NOP_INJECT_RESULT Inject result from sqe->result
482 */
483#define IORING_NOP_INJECT_RESULT (1U << 0)
484#define IORING_NOP_FILE (1U << 1)
485#define IORING_NOP_FIXED_FILE (1U << 2)
486#define IORING_NOP_FIXED_BUFFER (1U << 3)
487#define IORING_NOP_TW (1U << 4)
488#define IORING_NOP_CQE32 (1U << 5)
489
490/*
491 * IO completion data structure (Completion Queue Entry)
492 */
493struct io_uring_cqe {
494 __u64 user_data; /* sqe->user_data value passed back */
495 __s32 res; /* result code for this event */
496 __u32 flags;
497
498 /*
499 * If the ring is initialized with IORING_SETUP_CQE32, then this field
500 * contains 16-bytes of padding, doubling the size of the CQE.
501 */
502 __u64 big_cqe[];
503};
504
505/*
506 * cqe->flags
507 *
508 * IORING_CQE_F_BUFFER If set, the upper 16 bits are the buffer ID
509 * IORING_CQE_F_MORE If set, parent SQE will generate more CQE entries
510 * IORING_CQE_F_SOCK_NONEMPTY If set, more data to read after socket recv
511 * IORING_CQE_F_NOTIF Set for notification CQEs. Can be used to distinct
512 * them from sends.
513 * IORING_CQE_F_BUF_MORE If set, the buffer ID set in the completion will get
514 * more completions. In other words, the buffer is being
515 * partially consumed, and will be used by the kernel for
516 * more completions. This is only set for buffers used via
517 * the incremental buffer consumption, as provided by
518 * a ring buffer setup with IOU_PBUF_RING_INC. For any
519 * other provided buffer type, all completions with a
520 * buffer passed back is automatically returned to the
521 * application.
522 * IORING_CQE_F_SKIP If set, then the application/liburing must ignore this
523 * CQE. It's only purpose is to fill a gap in the ring,
524 * if a large CQE is attempted posted when the ring has
525 * just a single small CQE worth of space left before
526 * wrapping.
527 * IORING_CQE_F_32 If set, this is a 32b/big-cqe posting. Use with rings
528 * setup in a mixed CQE mode, where both 16b and 32b
529 * CQEs may be posted to the CQ ring.
530 */
531#define IORING_CQE_F_BUFFER (1U << 0)
532#define IORING_CQE_F_MORE (1U << 1)
533#define IORING_CQE_F_SOCK_NONEMPTY (1U << 2)
534#define IORING_CQE_F_NOTIF (1U << 3)
535#define IORING_CQE_F_BUF_MORE (1U << 4)
536#define IORING_CQE_F_SKIP (1U << 5)
537#define IORING_CQE_F_32 (1U << 15)
538
539#define IORING_CQE_BUFFER_SHIFT 16
540
541/*
542 * Magic offsets for the application to mmap the data it needs
543 */
544#define IORING_OFF_SQ_RING 0ULL
545#define IORING_OFF_CQ_RING 0x8000000ULL
546#define IORING_OFF_SQES 0x10000000ULL
547#define IORING_OFF_PBUF_RING 0x80000000ULL
548#define IORING_OFF_PBUF_SHIFT 16
549#define IORING_OFF_MMAP_MASK 0xf8000000ULL
550
551/*
552 * Filled with the offset for mmap(2)
553 */
554struct io_sqring_offsets {
555 __u32 head;
556 __u32 tail;
557 __u32 ring_mask;
558 __u32 ring_entries;
559 __u32 flags;
560 __u32 dropped;
561 __u32 array;
562 __u32 resv1;
563 __u64 user_addr;
564};
565
566/*
567 * sq_ring->flags
568 */
569#define IORING_SQ_NEED_WAKEUP (1U << 0) /* needs io_uring_enter wakeup */
570#define IORING_SQ_CQ_OVERFLOW (1U << 1) /* CQ ring is overflown */
571#define IORING_SQ_TASKRUN (1U << 2) /* task should enter the kernel */
572
573struct io_cqring_offsets {
574 __u32 head;
575 __u32 tail;
576 __u32 ring_mask;
577 __u32 ring_entries;
578 __u32 overflow;
579 __u32 cqes;
580 __u32 flags;
581 __u32 resv1;
582 __u64 user_addr;
583};
584
585/*
586 * cq_ring->flags
587 */
588
589/* disable eventfd notifications */
590#define IORING_CQ_EVENTFD_DISABLED (1U << 0)
591
592/*
593 * io_uring_enter(2) flags
594 */
595#define IORING_ENTER_GETEVENTS (1U << 0)
596#define IORING_ENTER_SQ_WAKEUP (1U << 1)
597#define IORING_ENTER_SQ_WAIT (1U << 2)
598#define IORING_ENTER_EXT_ARG (1U << 3)
599#define IORING_ENTER_REGISTERED_RING (1U << 4)
600#define IORING_ENTER_ABS_TIMER (1U << 5)
601#define IORING_ENTER_EXT_ARG_REG (1U << 6)
602#define IORING_ENTER_NO_IOWAIT (1U << 7)
603
604/*
605 * Passed in for io_uring_setup(2). Copied back with updated info on success
606 */
607struct io_uring_params {
608 __u32 sq_entries;
609 __u32 cq_entries;
610 __u32 flags;
611 __u32 sq_thread_cpu;
612 __u32 sq_thread_idle;
613 __u32 features;
614 __u32 wq_fd;
615 __u32 resv[3];
616 struct io_sqring_offsets sq_off;
617 struct io_cqring_offsets cq_off;
618};
619
620/*
621 * io_uring_params->features flags
622 */
623#define IORING_FEAT_SINGLE_MMAP (1U << 0)
624#define IORING_FEAT_NODROP (1U << 1)
625#define IORING_FEAT_SUBMIT_STABLE (1U << 2)
626#define IORING_FEAT_RW_CUR_POS (1U << 3)
627#define IORING_FEAT_CUR_PERSONALITY (1U << 4)
628#define IORING_FEAT_FAST_POLL (1U << 5)
629#define IORING_FEAT_POLL_32BITS (1U << 6)
630#define IORING_FEAT_SQPOLL_NONFIXED (1U << 7)
631#define IORING_FEAT_EXT_ARG (1U << 8)
632#define IORING_FEAT_NATIVE_WORKERS (1U << 9)
633#define IORING_FEAT_RSRC_TAGS (1U << 10)
634#define IORING_FEAT_CQE_SKIP (1U << 11)
635#define IORING_FEAT_LINKED_FILE (1U << 12)
636#define IORING_FEAT_REG_REG_RING (1U << 13)
637#define IORING_FEAT_RECVSEND_BUNDLE (1U << 14)
638#define IORING_FEAT_MIN_TIMEOUT (1U << 15)
639#define IORING_FEAT_RW_ATTR (1U << 16)
640#define IORING_FEAT_NO_IOWAIT (1U << 17)
641
642/*
643 * io_uring_register(2) opcodes and arguments
644 */
645enum io_uring_register_op {
646 IORING_REGISTER_BUFFERS = 0,
647 IORING_UNREGISTER_BUFFERS = 1,
648 IORING_REGISTER_FILES = 2,
649 IORING_UNREGISTER_FILES = 3,
650 IORING_REGISTER_EVENTFD = 4,
651 IORING_UNREGISTER_EVENTFD = 5,
652 IORING_REGISTER_FILES_UPDATE = 6,
653 IORING_REGISTER_EVENTFD_ASYNC = 7,
654 IORING_REGISTER_PROBE = 8,
655 IORING_REGISTER_PERSONALITY = 9,
656 IORING_UNREGISTER_PERSONALITY = 10,
657 IORING_REGISTER_RESTRICTIONS = 11,
658 IORING_REGISTER_ENABLE_RINGS = 12,
659
660 /* extended with tagging */
661 IORING_REGISTER_FILES2 = 13,
662 IORING_REGISTER_FILES_UPDATE2 = 14,
663 IORING_REGISTER_BUFFERS2 = 15,
664 IORING_REGISTER_BUFFERS_UPDATE = 16,
665
666 /* set/clear io-wq thread affinities */
667 IORING_REGISTER_IOWQ_AFF = 17,
668 IORING_UNREGISTER_IOWQ_AFF = 18,
669
670 /* set/get max number of io-wq workers */
671 IORING_REGISTER_IOWQ_MAX_WORKERS = 19,
672
673 /* register/unregister io_uring fd with the ring */
674 IORING_REGISTER_RING_FDS = 20,
675 IORING_UNREGISTER_RING_FDS = 21,
676
677 /* register ring based provide buffer group */
678 IORING_REGISTER_PBUF_RING = 22,
679 IORING_UNREGISTER_PBUF_RING = 23,
680
681 /* sync cancelation API */
682 IORING_REGISTER_SYNC_CANCEL = 24,
683
684 /* register a range of fixed file slots for automatic slot allocation */
685 IORING_REGISTER_FILE_ALLOC_RANGE = 25,
686
687 /* return status information for a buffer group */
688 IORING_REGISTER_PBUF_STATUS = 26,
689
690 /* set/clear busy poll settings */
691 IORING_REGISTER_NAPI = 27,
692 IORING_UNREGISTER_NAPI = 28,
693
694 IORING_REGISTER_CLOCK = 29,
695
696 /* clone registered buffers from source ring to current ring */
697 IORING_REGISTER_CLONE_BUFFERS = 30,
698
699 /* send MSG_RING without having a ring */
700 IORING_REGISTER_SEND_MSG_RING = 31,
701
702 /* register a netdev hw rx queue for zerocopy */
703 IORING_REGISTER_ZCRX_IFQ = 32,
704
705 /* resize CQ ring */
706 IORING_REGISTER_RESIZE_RINGS = 33,
707
708 IORING_REGISTER_MEM_REGION = 34,
709
710 /* query various aspects of io_uring, see linux/io_uring/query.h */
711 IORING_REGISTER_QUERY = 35,
712
713 /* auxiliary zcrx configuration, see enum zcrx_ctrl_op */
714 IORING_REGISTER_ZCRX_CTRL = 36,
715
716 /* register bpf filtering programs */
717 IORING_REGISTER_BPF_FILTER = 37,
718
719 /* this goes last */
720 IORING_REGISTER_LAST,
721
722 /* flag added to the opcode to use a registered ring fd */
723 IORING_REGISTER_USE_REGISTERED_RING = 1U << 31
724};
725
726/* io-wq worker categories */
727enum io_wq_type {
728 IO_WQ_BOUND,
729 IO_WQ_UNBOUND,
730};
731
732/* deprecated, see struct io_uring_rsrc_update */
733struct io_uring_files_update {
734 __u32 offset;
735 __u32 resv;
736 __aligned_u64 /* __s32 * */ fds;
737};
738
739enum {
740 /* initialise with user provided memory pointed by user_addr */
741 IORING_MEM_REGION_TYPE_USER = 1,
742};
743
744struct io_uring_region_desc {
745 __u64 user_addr;
746 __u64 size;
747 __u32 flags;
748 __u32 id;
749 __u64 mmap_offset;
750 __u64 __resv[4];
751};
752
753enum {
754 /* expose the region as registered wait arguments */
755 IORING_MEM_REGION_REG_WAIT_ARG = 1,
756};
757
758struct io_uring_mem_region_reg {
759 __u64 region_uptr; /* struct io_uring_region_desc * */
760 __u64 flags;
761 __u64 __resv[2];
762};
763
764/*
765 * Register a fully sparse file space, rather than pass in an array of all
766 * -1 file descriptors.
767 */
768#define IORING_RSRC_REGISTER_SPARSE (1U << 0)
769
770struct io_uring_rsrc_register {
771 __u32 nr;
772 __u32 flags;
773 __u64 resv2;
774 __aligned_u64 data;
775 __aligned_u64 tags;
776};
777
778struct io_uring_rsrc_update {
779 __u32 offset;
780 __u32 resv;
781 __aligned_u64 data;
782};
783
784struct io_uring_rsrc_update2 {
785 __u32 offset;
786 __u32 resv;
787 __aligned_u64 data;
788 __aligned_u64 tags;
789 __u32 nr;
790 __u32 resv2;
791};
792
793/* Skip updating fd indexes set to this value in the fd table */
794#define IORING_REGISTER_FILES_SKIP (-2)
795
796#define IO_URING_OP_SUPPORTED (1U << 0)
797
798struct io_uring_probe_op {
799 __u8 op;
800 __u8 resv;
801 __u16 flags; /* IO_URING_OP_* flags */
802 __u32 resv2;
803};
804
805struct io_uring_probe {
806 __u8 last_op; /* last opcode supported */
807 __u8 ops_len; /* length of ops[] array below */
808 __u16 resv;
809 __u32 resv2[3];
810 struct io_uring_probe_op ops[];
811};
812
813struct io_uring_restriction {
814 __u16 opcode;
815 union {
816 __u8 register_op; /* IORING_RESTRICTION_REGISTER_OP */
817 __u8 sqe_op; /* IORING_RESTRICTION_SQE_OP */
818 __u8 sqe_flags; /* IORING_RESTRICTION_SQE_FLAGS_* */
819 };
820 __u8 resv;
821 __u32 resv2[3];
822};
823
824struct io_uring_task_restriction {
825 __u16 flags;
826 __u16 nr_res;
827 __u32 resv[3];
828 __DECLARE_FLEX_ARRAY(struct io_uring_restriction, restrictions);
829};
830
831struct io_uring_clock_register {
832 __u32 clockid;
833 __u32 __resv[3];
834};
835
836enum {
837 IORING_REGISTER_SRC_REGISTERED = (1U << 0),
838 IORING_REGISTER_DST_REPLACE = (1U << 1),
839};
840
841struct io_uring_clone_buffers {
842 __u32 src_fd;
843 __u32 flags;
844 __u32 src_off;
845 __u32 dst_off;
846 __u32 nr;
847 __u32 pad[3];
848};
849
850struct io_uring_buf {
851 __u64 addr;
852 __u32 len;
853 __u16 bid;
854 __u16 resv;
855};
856
857struct io_uring_buf_ring {
858 union {
859 /*
860 * To avoid spilling into more pages than we need to, the
861 * ring tail is overlaid with the io_uring_buf->resv field.
862 */
863 struct {
864 __u64 resv1;
865 __u32 resv2;
866 __u16 resv3;
867 __u16 tail;
868 };
869 __DECLARE_FLEX_ARRAY(struct io_uring_buf, bufs);
870 };
871};
872
873/*
874 * Flags for IORING_REGISTER_PBUF_RING.
875 *
876 * IOU_PBUF_RING_MMAP: If set, kernel will allocate the memory for the ring.
877 * The application must not set a ring_addr in struct
878 * io_uring_buf_reg, instead it must subsequently call
879 * mmap(2) with the offset set as:
880 * IORING_OFF_PBUF_RING | (bgid << IORING_OFF_PBUF_SHIFT)
881 * to get a virtual mapping for the ring.
882 * IOU_PBUF_RING_INC: If set, buffers consumed from this buffer ring can be
883 * consumed incrementally. Normally one (or more) buffers
884 * are fully consumed. With incremental consumptions, it's
885 * feasible to register big ranges of buffers, and each
886 * use of it will consume only as much as it needs. This
887 * requires that both the kernel and application keep
888 * track of where the current read/recv index is at.
889 */
890enum io_uring_register_pbuf_ring_flags {
891 IOU_PBUF_RING_MMAP = 1,
892 IOU_PBUF_RING_INC = 2,
893};
894
895/* argument for IORING_(UN)REGISTER_PBUF_RING */
896struct io_uring_buf_reg {
897 __u64 ring_addr;
898 __u32 ring_entries;
899 __u16 bgid;
900 __u16 flags;
901 __u64 resv[3];
902};
903
904/* argument for IORING_REGISTER_PBUF_STATUS */
905struct io_uring_buf_status {
906 __u32 buf_group; /* input */
907 __u32 head; /* output */
908 __u32 resv[8];
909};
910
911enum io_uring_napi_op {
912 /* register/ungister backward compatible opcode */
913 IO_URING_NAPI_REGISTER_OP = 0,
914
915 /* opcodes to update napi_list when static tracking is used */
916 IO_URING_NAPI_STATIC_ADD_ID = 1,
917 IO_URING_NAPI_STATIC_DEL_ID = 2
918};
919
920enum io_uring_napi_tracking_strategy {
921 /* value must be 0 for backward compatibility */
922 IO_URING_NAPI_TRACKING_DYNAMIC = 0,
923 IO_URING_NAPI_TRACKING_STATIC = 1,
924 IO_URING_NAPI_TRACKING_INACTIVE = 255
925};
926
927/* argument for IORING_(UN)REGISTER_NAPI */
928struct io_uring_napi {
929 __u32 busy_poll_to;
930 __u8 prefer_busy_poll;
931
932 /* a io_uring_napi_op value */
933 __u8 opcode;
934 __u8 pad[2];
935
936 /*
937 * for IO_URING_NAPI_REGISTER_OP, it is a
938 * io_uring_napi_tracking_strategy value.
939 *
940 * for IO_URING_NAPI_STATIC_ADD_ID/IO_URING_NAPI_STATIC_DEL_ID
941 * it is the napi id to add/del from napi_list.
942 */
943 __u32 op_param;
944 __u32 resv;
945};
946
947/*
948 * io_uring_restriction->opcode values
949 */
950enum io_uring_register_restriction_op {
951 /* Allow an io_uring_register(2) opcode */
952 IORING_RESTRICTION_REGISTER_OP = 0,
953
954 /* Allow an sqe opcode */
955 IORING_RESTRICTION_SQE_OP = 1,
956
957 /* Allow sqe flags */
958 IORING_RESTRICTION_SQE_FLAGS_ALLOWED = 2,
959
960 /* Require sqe flags (these flags must be set on each submission) */
961 IORING_RESTRICTION_SQE_FLAGS_REQUIRED = 3,
962
963 IORING_RESTRICTION_LAST
964};
965
966enum {
967 IORING_REG_WAIT_TS = (1U << 0),
968};
969
970/*
971 * Argument for io_uring_enter(2) with
972 * IORING_GETEVENTS | IORING_ENTER_EXT_ARG_REG set, where the actual argument
973 * is an index into a previously registered fixed wait region described by
974 * the below structure.
975 */
976struct io_uring_reg_wait {
977 struct __kernel_timespec ts;
978 __u32 min_wait_usec;
979 __u32 flags;
980 __u64 sigmask;
981 __u32 sigmask_sz;
982 __u32 pad[3];
983 __u64 pad2[2];
984};
985
986/*
987 * Argument for io_uring_enter(2) with IORING_GETEVENTS | IORING_ENTER_EXT_ARG
988 */
989struct io_uring_getevents_arg {
990 __u64 sigmask;
991 __u32 sigmask_sz;
992 __u32 min_wait_usec;
993 __u64 ts;
994};
995
996/*
997 * Argument for IORING_REGISTER_SYNC_CANCEL
998 */
999struct io_uring_sync_cancel_reg {
1000 __u64 addr;
1001 __s32 fd;
1002 __u32 flags;
1003 struct __kernel_timespec timeout;
1004 __u8 opcode;
1005 __u8 pad[7];
1006 __u64 pad2[3];
1007};
1008
1009/*
1010 * Argument for IORING_REGISTER_FILE_ALLOC_RANGE
1011 * The range is specified as [off, off + len)
1012 */
1013struct io_uring_file_index_range {
1014 __u32 off;
1015 __u32 len;
1016 __u64 resv;
1017};
1018
1019struct io_uring_recvmsg_out {
1020 __u32 namelen;
1021 __u32 controllen;
1022 __u32 payloadlen;
1023 __u32 flags;
1024};
1025
1026/*
1027 * Argument for IORING_OP_URING_CMD when file is a socket
1028 */
1029enum io_uring_socket_op {
1030 SOCKET_URING_OP_SIOCINQ = 0,
1031 SOCKET_URING_OP_SIOCOUTQ,
1032 SOCKET_URING_OP_GETSOCKOPT,
1033 SOCKET_URING_OP_SETSOCKOPT,
1034 SOCKET_URING_OP_TX_TIMESTAMP,
1035 SOCKET_URING_OP_GETSOCKNAME,
1036};
1037
1038/*
1039 * SOCKET_URING_OP_TX_TIMESTAMP definitions
1040 */
1041
1042#define IORING_TIMESTAMP_HW_SHIFT 16
1043/* The cqe->flags bit from which the timestamp type is stored */
1044#define IORING_TIMESTAMP_TYPE_SHIFT (IORING_TIMESTAMP_HW_SHIFT + 1)
1045/* The cqe->flags flag signifying whether it's a hardware timestamp */
1046#define IORING_CQE_F_TSTAMP_HW ((__u32)1 << IORING_TIMESTAMP_HW_SHIFT)
1047
1048struct io_timespec {
1049 __u64 tv_sec;
1050 __u64 tv_nsec;
1051};
1052
1053/* Zero copy receive refill queue entry */
1054struct io_uring_zcrx_rqe {
1055 __u64 off;
1056 __u32 len;
1057 __u32 __pad;
1058};
1059
1060struct io_uring_zcrx_cqe {
1061 __u64 off;
1062 __u64 __pad;
1063};
1064
1065/* The bit from which area id is encoded into offsets */
1066#define IORING_ZCRX_AREA_SHIFT 48
1067#define IORING_ZCRX_AREA_MASK (~(((__u64)1 << IORING_ZCRX_AREA_SHIFT) - 1))
1068
1069struct io_uring_zcrx_offsets {
1070 __u32 head;
1071 __u32 tail;
1072 __u32 rqes;
1073 __u32 __resv2;
1074 __u64 __resv[2];
1075};
1076
1077enum io_uring_zcrx_area_flags {
1078 IORING_ZCRX_AREA_DMABUF = 1,
1079};
1080
1081struct io_uring_zcrx_area_reg {
1082 __u64 addr;
1083 __u64 len;
1084 __u64 rq_area_token;
1085 __u32 flags;
1086 __u32 dmabuf_fd;
1087 __u64 __resv2[2];
1088};
1089
1090enum zcrx_reg_flags {
1091 ZCRX_REG_IMPORT = 1,
1092};
1093
1094enum zcrx_features {
1095 /*
1096 * The user can ask for the desired rx page size by passing the
1097 * value in struct io_uring_zcrx_ifq_reg::rx_buf_len.
1098 */
1099 ZCRX_FEATURE_RX_PAGE_SIZE = 1 << 0,
1100};
1101
1102/*
1103 * Argument for IORING_REGISTER_ZCRX_IFQ
1104 */
1105struct io_uring_zcrx_ifq_reg {
1106 __u32 if_idx;
1107 __u32 if_rxq;
1108 __u32 rq_entries;
1109 __u32 flags;
1110
1111 __u64 area_ptr; /* pointer to struct io_uring_zcrx_area_reg */
1112 __u64 region_ptr; /* struct io_uring_region_desc * */
1113
1114 struct io_uring_zcrx_offsets offsets;
1115 __u32 zcrx_id;
1116 __u32 rx_buf_len;
1117 __u64 __resv[3];
1118};
1119
1120enum zcrx_ctrl_op {
1121 ZCRX_CTRL_FLUSH_RQ,
1122 ZCRX_CTRL_EXPORT,
1123
1124 __ZCRX_CTRL_LAST,
1125};
1126
1127struct zcrx_ctrl_flush_rq {
1128 __u64 __resv[6];
1129};
1130
1131struct zcrx_ctrl_export {
1132 __u32 zcrx_fd;
1133 __u32 __resv1[11];
1134};
1135
1136struct zcrx_ctrl {
1137 __u32 zcrx_id;
1138 __u32 op; /* see enum zcrx_ctrl_op */
1139 __u64 __resv[2];
1140
1141 union {
1142 struct zcrx_ctrl_export zc_export;
1143 struct zcrx_ctrl_flush_rq zc_flush;
1144 };
1145};
1146
1147#ifdef __cplusplus
1148}
1149#endif
1150
1151#endif