xref: /openbmc/linux/include/uapi/linux/io_uring.h (revision 9b4469410cf9a0fcbccc92c480fd42f7c815a745)
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
24 extern "C" {
25 #endif
26 
27 /*
28  * IO submission data structure (Submission Queue Entry)
29  */
30 struct 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 	};
47 	__u32	len;		/* buffer size or number of iovecs */
48 	union {
49 		__kernel_rwf_t	rw_flags;
50 		__u32		fsync_flags;
51 		__u16		poll_events;	/* compatibility */
52 		__u32		poll32_events;	/* word-reversed for BE */
53 		__u32		sync_range_flags;
54 		__u32		msg_flags;
55 		__u32		timeout_flags;
56 		__u32		accept_flags;
57 		__u32		cancel_flags;
58 		__u32		open_flags;
59 		__u32		statx_flags;
60 		__u32		fadvise_advice;
61 		__u32		splice_flags;
62 		__u32		rename_flags;
63 		__u32		unlink_flags;
64 		__u32		hardlink_flags;
65 		__u32		xattr_flags;
66 		__u32		msg_ring_flags;
67 		__u32		uring_cmd_flags;
68 	};
69 	__u64	user_data;	/* data to be passed back at completion time */
70 	/* pack this to avoid bogus arm OABI complaints */
71 	union {
72 		/* index into fixed buffers, if used */
73 		__u16	buf_index;
74 		/* for grouped buffer selection */
75 		__u16	buf_group;
76 	} __attribute__((packed));
77 	/* personality to use, if used */
78 	__u16	personality;
79 	union {
80 		__s32	splice_fd_in;
81 		__u32	file_index;
82 		struct {
83 			__u16	addr_len;
84 			__u16	__pad3[1];
85 		};
86 	};
87 	union {
88 		struct {
89 			__u64	addr3;
90 			__u64	__pad2[1];
91 		};
92 		/*
93 		 * If the ring is initialized with IORING_SETUP_SQE128, then
94 		 * this field is used for 80 bytes of arbitrary command data
95 		 */
96 		__u8	cmd[0];
97 	};
98 };
99 
100 /*
101  * If sqe->file_index is set to this for opcodes that instantiate a new
102  * direct descriptor (like openat/openat2/accept), then io_uring will allocate
103  * an available direct descriptor instead of having the application pass one
104  * in. The picked direct descriptor will be returned in cqe->res, or -ENFILE
105  * if the space is full.
106  */
107 #define IORING_FILE_INDEX_ALLOC		(~0U)
108 
109 enum {
110 	IOSQE_FIXED_FILE_BIT,
111 	IOSQE_IO_DRAIN_BIT,
112 	IOSQE_IO_LINK_BIT,
113 	IOSQE_IO_HARDLINK_BIT,
114 	IOSQE_ASYNC_BIT,
115 	IOSQE_BUFFER_SELECT_BIT,
116 	IOSQE_CQE_SKIP_SUCCESS_BIT,
117 };
118 
119 /*
120  * sqe->flags
121  */
122 /* use fixed fileset */
123 #define IOSQE_FIXED_FILE	(1U << IOSQE_FIXED_FILE_BIT)
124 /* issue after inflight IO */
125 #define IOSQE_IO_DRAIN		(1U << IOSQE_IO_DRAIN_BIT)
126 /* links next sqe */
127 #define IOSQE_IO_LINK		(1U << IOSQE_IO_LINK_BIT)
128 /* like LINK, but stronger */
129 #define IOSQE_IO_HARDLINK	(1U << IOSQE_IO_HARDLINK_BIT)
130 /* always go async */
131 #define IOSQE_ASYNC		(1U << IOSQE_ASYNC_BIT)
132 /* select buffer from sqe->buf_group */
133 #define IOSQE_BUFFER_SELECT	(1U << IOSQE_BUFFER_SELECT_BIT)
134 /* don't post CQE if request succeeded */
135 #define IOSQE_CQE_SKIP_SUCCESS	(1U << IOSQE_CQE_SKIP_SUCCESS_BIT)
136 
137 /*
138  * io_uring_setup() flags
139  */
140 #define IORING_SETUP_IOPOLL	(1U << 0)	/* io_context is polled */
141 #define IORING_SETUP_SQPOLL	(1U << 1)	/* SQ poll thread */
142 #define IORING_SETUP_SQ_AFF	(1U << 2)	/* sq_thread_cpu is valid */
143 #define IORING_SETUP_CQSIZE	(1U << 3)	/* app defines CQ size */
144 #define IORING_SETUP_CLAMP	(1U << 4)	/* clamp SQ/CQ ring sizes */
145 #define IORING_SETUP_ATTACH_WQ	(1U << 5)	/* attach to existing wq */
146 #define IORING_SETUP_R_DISABLED	(1U << 6)	/* start with ring disabled */
147 #define IORING_SETUP_SUBMIT_ALL	(1U << 7)	/* continue submit on error */
148 /*
149  * Cooperative task running. When requests complete, they often require
150  * forcing the submitter to transition to the kernel to complete. If this
151  * flag is set, work will be done when the task transitions anyway, rather
152  * than force an inter-processor interrupt reschedule. This avoids interrupting
153  * a task running in userspace, and saves an IPI.
154  */
155 #define IORING_SETUP_COOP_TASKRUN	(1U << 8)
156 /*
157  * If COOP_TASKRUN is set, get notified if task work is available for
158  * running and a kernel transition would be needed to run it. This sets
159  * IORING_SQ_TASKRUN in the sq ring flags. Not valid with COOP_TASKRUN.
160  */
161 #define IORING_SETUP_TASKRUN_FLAG	(1U << 9)
162 #define IORING_SETUP_SQE128		(1U << 10) /* SQEs are 128 byte */
163 #define IORING_SETUP_CQE32		(1U << 11) /* CQEs are 32 byte */
164 /*
165  * Only one task is allowed to submit requests
166  */
167 #define IORING_SETUP_SINGLE_ISSUER	(1U << 12)
168 
169 /*
170  * Defer running task work to get events.
171  * Rather than running bits of task work whenever the task transitions
172  * try to do it just before it is needed.
173  */
174 #define IORING_SETUP_DEFER_TASKRUN	(1U << 13)
175 
176 enum io_uring_op {
177 	IORING_OP_NOP,
178 	IORING_OP_READV,
179 	IORING_OP_WRITEV,
180 	IORING_OP_FSYNC,
181 	IORING_OP_READ_FIXED,
182 	IORING_OP_WRITE_FIXED,
183 	IORING_OP_POLL_ADD,
184 	IORING_OP_POLL_REMOVE,
185 	IORING_OP_SYNC_FILE_RANGE,
186 	IORING_OP_SENDMSG,
187 	IORING_OP_RECVMSG,
188 	IORING_OP_TIMEOUT,
189 	IORING_OP_TIMEOUT_REMOVE,
190 	IORING_OP_ACCEPT,
191 	IORING_OP_ASYNC_CANCEL,
192 	IORING_OP_LINK_TIMEOUT,
193 	IORING_OP_CONNECT,
194 	IORING_OP_FALLOCATE,
195 	IORING_OP_OPENAT,
196 	IORING_OP_CLOSE,
197 	IORING_OP_FILES_UPDATE,
198 	IORING_OP_STATX,
199 	IORING_OP_READ,
200 	IORING_OP_WRITE,
201 	IORING_OP_FADVISE,
202 	IORING_OP_MADVISE,
203 	IORING_OP_SEND,
204 	IORING_OP_RECV,
205 	IORING_OP_OPENAT2,
206 	IORING_OP_EPOLL_CTL,
207 	IORING_OP_SPLICE,
208 	IORING_OP_PROVIDE_BUFFERS,
209 	IORING_OP_REMOVE_BUFFERS,
210 	IORING_OP_TEE,
211 	IORING_OP_SHUTDOWN,
212 	IORING_OP_RENAMEAT,
213 	IORING_OP_UNLINKAT,
214 	IORING_OP_MKDIRAT,
215 	IORING_OP_SYMLINKAT,
216 	IORING_OP_LINKAT,
217 	IORING_OP_MSG_RING,
218 	IORING_OP_FSETXATTR,
219 	IORING_OP_SETXATTR,
220 	IORING_OP_FGETXATTR,
221 	IORING_OP_GETXATTR,
222 	IORING_OP_SOCKET,
223 	IORING_OP_URING_CMD,
224 	IORING_OP_SEND_ZC,
225 	IORING_OP_SENDMSG_ZC,
226 
227 	/* this goes last, obviously */
228 	IORING_OP_LAST,
229 };
230 
231 /*
232  * sqe->uring_cmd_flags
233  * IORING_URING_CMD_FIXED	use registered buffer; pass this flag
234  *				along with setting sqe->buf_index.
235  */
236 #define IORING_URING_CMD_FIXED	(1U << 0)
237 
238 
239 /*
240  * sqe->fsync_flags
241  */
242 #define IORING_FSYNC_DATASYNC	(1U << 0)
243 
244 /*
245  * sqe->timeout_flags
246  */
247 #define IORING_TIMEOUT_ABS		(1U << 0)
248 #define IORING_TIMEOUT_UPDATE		(1U << 1)
249 #define IORING_TIMEOUT_BOOTTIME		(1U << 2)
250 #define IORING_TIMEOUT_REALTIME		(1U << 3)
251 #define IORING_LINK_TIMEOUT_UPDATE	(1U << 4)
252 #define IORING_TIMEOUT_ETIME_SUCCESS	(1U << 5)
253 #define IORING_TIMEOUT_MULTISHOT	(1U << 6)
254 #define IORING_TIMEOUT_CLOCK_MASK	(IORING_TIMEOUT_BOOTTIME | IORING_TIMEOUT_REALTIME)
255 #define IORING_TIMEOUT_UPDATE_MASK	(IORING_TIMEOUT_UPDATE | IORING_LINK_TIMEOUT_UPDATE)
256 /*
257  * sqe->splice_flags
258  * extends splice(2) flags
259  */
260 #define SPLICE_F_FD_IN_FIXED	(1U << 31) /* the last bit of __u32 */
261 
262 /*
263  * POLL_ADD flags. Note that since sqe->poll_events is the flag space, the
264  * command flags for POLL_ADD are stored in sqe->len.
265  *
266  * IORING_POLL_ADD_MULTI	Multishot poll. Sets IORING_CQE_F_MORE if
267  *				the poll handler will continue to report
268  *				CQEs on behalf of the same SQE.
269  *
270  * IORING_POLL_UPDATE		Update existing poll request, matching
271  *				sqe->addr as the old user_data field.
272  *
273  * IORING_POLL_LEVEL		Level triggered poll.
274  */
275 #define IORING_POLL_ADD_MULTI	(1U << 0)
276 #define IORING_POLL_UPDATE_EVENTS	(1U << 1)
277 #define IORING_POLL_UPDATE_USER_DATA	(1U << 2)
278 #define IORING_POLL_ADD_LEVEL		(1U << 3)
279 
280 /*
281  * ASYNC_CANCEL flags.
282  *
283  * IORING_ASYNC_CANCEL_ALL	Cancel all requests that match the given key
284  * IORING_ASYNC_CANCEL_FD	Key off 'fd' for cancelation rather than the
285  *				request 'user_data'
286  * IORING_ASYNC_CANCEL_ANY	Match any request
287  * IORING_ASYNC_CANCEL_FD_FIXED	'fd' passed in is a fixed descriptor
288  */
289 #define IORING_ASYNC_CANCEL_ALL	(1U << 0)
290 #define IORING_ASYNC_CANCEL_FD	(1U << 1)
291 #define IORING_ASYNC_CANCEL_ANY	(1U << 2)
292 #define IORING_ASYNC_CANCEL_FD_FIXED	(1U << 3)
293 
294 /*
295  * send/sendmsg and recv/recvmsg flags (sqe->ioprio)
296  *
297  * IORING_RECVSEND_POLL_FIRST	If set, instead of first attempting to send
298  *				or receive and arm poll if that yields an
299  *				-EAGAIN result, arm poll upfront and skip
300  *				the initial transfer attempt.
301  *
302  * IORING_RECV_MULTISHOT	Multishot recv. Sets IORING_CQE_F_MORE if
303  *				the handler will continue to report
304  *				CQEs on behalf of the same SQE.
305  *
306  * IORING_RECVSEND_FIXED_BUF	Use registered buffers, the index is stored in
307  *				the buf_index field.
308  *
309  * IORING_SEND_ZC_REPORT_USAGE
310  *				If set, SEND[MSG]_ZC should report
311  *				the zerocopy usage in cqe.res
312  *				for the IORING_CQE_F_NOTIF cqe.
313  *				0 is reported if zerocopy was actually possible.
314  *				IORING_NOTIF_USAGE_ZC_COPIED if data was copied
315  *				(at least partially).
316  */
317 #define IORING_RECVSEND_POLL_FIRST	(1U << 0)
318 #define IORING_RECV_MULTISHOT		(1U << 1)
319 #define IORING_RECVSEND_FIXED_BUF	(1U << 2)
320 #define IORING_SEND_ZC_REPORT_USAGE	(1U << 3)
321 
322 /*
323  * cqe.res for IORING_CQE_F_NOTIF if
324  * IORING_SEND_ZC_REPORT_USAGE was requested
325  *
326  * It should be treated as a flag, all other
327  * bits of cqe.res should be treated as reserved!
328  */
329 #define IORING_NOTIF_USAGE_ZC_COPIED    (1U << 31)
330 
331 /*
332  * accept flags stored in sqe->ioprio
333  */
334 #define IORING_ACCEPT_MULTISHOT	(1U << 0)
335 
336 /*
337  * IORING_OP_MSG_RING command types, stored in sqe->addr
338  */
339 enum {
340 	IORING_MSG_DATA,	/* pass sqe->len as 'res' and off as user_data */
341 	IORING_MSG_SEND_FD,	/* send a registered fd to another ring */
342 };
343 
344 /*
345  * IORING_OP_MSG_RING flags (sqe->msg_ring_flags)
346  *
347  * IORING_MSG_RING_CQE_SKIP	Don't post a CQE to the target ring. Not
348  *				applicable for IORING_MSG_DATA, obviously.
349  */
350 #define IORING_MSG_RING_CQE_SKIP	(1U << 0)
351 /* Pass through the flags from sqe->file_index to cqe->flags */
352 #define IORING_MSG_RING_FLAGS_PASS	(1U << 1)
353 
354 /*
355  * IO completion data structure (Completion Queue Entry)
356  */
357 struct io_uring_cqe {
358 	__u64	user_data;	/* sqe->data submission passed back */
359 	__s32	res;		/* result code for this event */
360 	__u32	flags;
361 
362 	/*
363 	 * If the ring is initialized with IORING_SETUP_CQE32, then this field
364 	 * contains 16-bytes of padding, doubling the size of the CQE.
365 	 */
366 	__u64 big_cqe[];
367 };
368 
369 /*
370  * cqe->flags
371  *
372  * IORING_CQE_F_BUFFER	If set, the upper 16 bits are the buffer ID
373  * IORING_CQE_F_MORE	If set, parent SQE will generate more CQE entries
374  * IORING_CQE_F_SOCK_NONEMPTY	If set, more data to read after socket recv
375  * IORING_CQE_F_NOTIF	Set for notification CQEs. Can be used to distinct
376  * 			them from sends.
377  */
378 #define IORING_CQE_F_BUFFER		(1U << 0)
379 #define IORING_CQE_F_MORE		(1U << 1)
380 #define IORING_CQE_F_SOCK_NONEMPTY	(1U << 2)
381 #define IORING_CQE_F_NOTIF		(1U << 3)
382 
383 enum {
384 	IORING_CQE_BUFFER_SHIFT		= 16,
385 };
386 
387 /*
388  * Magic offsets for the application to mmap the data it needs
389  */
390 #define IORING_OFF_SQ_RING		0ULL
391 #define IORING_OFF_CQ_RING		0x8000000ULL
392 #define IORING_OFF_SQES			0x10000000ULL
393 #define IORING_OFF_PBUF_RING		0x80000000ULL
394 #define IORING_OFF_PBUF_SHIFT		16
395 #define IORING_OFF_MMAP_MASK		0xf8000000ULL
396 
397 /*
398  * Filled with the offset for mmap(2)
399  */
400 struct io_sqring_offsets {
401 	__u32 head;
402 	__u32 tail;
403 	__u32 ring_mask;
404 	__u32 ring_entries;
405 	__u32 flags;
406 	__u32 dropped;
407 	__u32 array;
408 	__u32 resv1;
409 	__u64 resv2;
410 };
411 
412 /*
413  * sq_ring->flags
414  */
415 #define IORING_SQ_NEED_WAKEUP	(1U << 0) /* needs io_uring_enter wakeup */
416 #define IORING_SQ_CQ_OVERFLOW	(1U << 1) /* CQ ring is overflown */
417 #define IORING_SQ_TASKRUN	(1U << 2) /* task should enter the kernel */
418 
419 struct io_cqring_offsets {
420 	__u32 head;
421 	__u32 tail;
422 	__u32 ring_mask;
423 	__u32 ring_entries;
424 	__u32 overflow;
425 	__u32 cqes;
426 	__u32 flags;
427 	__u32 resv1;
428 	__u64 resv2;
429 };
430 
431 /*
432  * cq_ring->flags
433  */
434 
435 /* disable eventfd notifications */
436 #define IORING_CQ_EVENTFD_DISABLED	(1U << 0)
437 
438 /*
439  * io_uring_enter(2) flags
440  */
441 #define IORING_ENTER_GETEVENTS		(1U << 0)
442 #define IORING_ENTER_SQ_WAKEUP		(1U << 1)
443 #define IORING_ENTER_SQ_WAIT		(1U << 2)
444 #define IORING_ENTER_EXT_ARG		(1U << 3)
445 #define IORING_ENTER_REGISTERED_RING	(1U << 4)
446 
447 /*
448  * Passed in for io_uring_setup(2). Copied back with updated info on success
449  */
450 struct io_uring_params {
451 	__u32 sq_entries;
452 	__u32 cq_entries;
453 	__u32 flags;
454 	__u32 sq_thread_cpu;
455 	__u32 sq_thread_idle;
456 	__u32 features;
457 	__u32 wq_fd;
458 	__u32 resv[3];
459 	struct io_sqring_offsets sq_off;
460 	struct io_cqring_offsets cq_off;
461 };
462 
463 /*
464  * io_uring_params->features flags
465  */
466 #define IORING_FEAT_SINGLE_MMAP		(1U << 0)
467 #define IORING_FEAT_NODROP		(1U << 1)
468 #define IORING_FEAT_SUBMIT_STABLE	(1U << 2)
469 #define IORING_FEAT_RW_CUR_POS		(1U << 3)
470 #define IORING_FEAT_CUR_PERSONALITY	(1U << 4)
471 #define IORING_FEAT_FAST_POLL		(1U << 5)
472 #define IORING_FEAT_POLL_32BITS 	(1U << 6)
473 #define IORING_FEAT_SQPOLL_NONFIXED	(1U << 7)
474 #define IORING_FEAT_EXT_ARG		(1U << 8)
475 #define IORING_FEAT_NATIVE_WORKERS	(1U << 9)
476 #define IORING_FEAT_RSRC_TAGS		(1U << 10)
477 #define IORING_FEAT_CQE_SKIP		(1U << 11)
478 #define IORING_FEAT_LINKED_FILE		(1U << 12)
479 #define IORING_FEAT_REG_REG_RING	(1U << 13)
480 
481 /*
482  * io_uring_register(2) opcodes and arguments
483  */
484 enum {
485 	IORING_REGISTER_BUFFERS			= 0,
486 	IORING_UNREGISTER_BUFFERS		= 1,
487 	IORING_REGISTER_FILES			= 2,
488 	IORING_UNREGISTER_FILES			= 3,
489 	IORING_REGISTER_EVENTFD			= 4,
490 	IORING_UNREGISTER_EVENTFD		= 5,
491 	IORING_REGISTER_FILES_UPDATE		= 6,
492 	IORING_REGISTER_EVENTFD_ASYNC		= 7,
493 	IORING_REGISTER_PROBE			= 8,
494 	IORING_REGISTER_PERSONALITY		= 9,
495 	IORING_UNREGISTER_PERSONALITY		= 10,
496 	IORING_REGISTER_RESTRICTIONS		= 11,
497 	IORING_REGISTER_ENABLE_RINGS		= 12,
498 
499 	/* extended with tagging */
500 	IORING_REGISTER_FILES2			= 13,
501 	IORING_REGISTER_FILES_UPDATE2		= 14,
502 	IORING_REGISTER_BUFFERS2		= 15,
503 	IORING_REGISTER_BUFFERS_UPDATE		= 16,
504 
505 	/* set/clear io-wq thread affinities */
506 	IORING_REGISTER_IOWQ_AFF		= 17,
507 	IORING_UNREGISTER_IOWQ_AFF		= 18,
508 
509 	/* set/get max number of io-wq workers */
510 	IORING_REGISTER_IOWQ_MAX_WORKERS	= 19,
511 
512 	/* register/unregister io_uring fd with the ring */
513 	IORING_REGISTER_RING_FDS		= 20,
514 	IORING_UNREGISTER_RING_FDS		= 21,
515 
516 	/* register ring based provide buffer group */
517 	IORING_REGISTER_PBUF_RING		= 22,
518 	IORING_UNREGISTER_PBUF_RING		= 23,
519 
520 	/* sync cancelation API */
521 	IORING_REGISTER_SYNC_CANCEL		= 24,
522 
523 	/* register a range of fixed file slots for automatic slot allocation */
524 	IORING_REGISTER_FILE_ALLOC_RANGE	= 25,
525 
526 	/* this goes last */
527 	IORING_REGISTER_LAST,
528 
529 	/* flag added to the opcode to use a registered ring fd */
530 	IORING_REGISTER_USE_REGISTERED_RING	= 1U << 31
531 };
532 
533 /* io-wq worker categories */
534 enum {
535 	IO_WQ_BOUND,
536 	IO_WQ_UNBOUND,
537 };
538 
539 /* deprecated, see struct io_uring_rsrc_update */
540 struct io_uring_files_update {
541 	__u32 offset;
542 	__u32 resv;
543 	__aligned_u64 /* __s32 * */ fds;
544 };
545 
546 /*
547  * Register a fully sparse file space, rather than pass in an array of all
548  * -1 file descriptors.
549  */
550 #define IORING_RSRC_REGISTER_SPARSE	(1U << 0)
551 
552 struct io_uring_rsrc_register {
553 	__u32 nr;
554 	__u32 flags;
555 	__u64 resv2;
556 	__aligned_u64 data;
557 	__aligned_u64 tags;
558 };
559 
560 struct io_uring_rsrc_update {
561 	__u32 offset;
562 	__u32 resv;
563 	__aligned_u64 data;
564 };
565 
566 struct io_uring_rsrc_update2 {
567 	__u32 offset;
568 	__u32 resv;
569 	__aligned_u64 data;
570 	__aligned_u64 tags;
571 	__u32 nr;
572 	__u32 resv2;
573 };
574 
575 /* Skip updating fd indexes set to this value in the fd table */
576 #define IORING_REGISTER_FILES_SKIP	(-2)
577 
578 #define IO_URING_OP_SUPPORTED	(1U << 0)
579 
580 struct io_uring_probe_op {
581 	__u8 op;
582 	__u8 resv;
583 	__u16 flags;	/* IO_URING_OP_* flags */
584 	__u32 resv2;
585 };
586 
587 struct io_uring_probe {
588 	__u8 last_op;	/* last opcode supported */
589 	__u8 ops_len;	/* length of ops[] array below */
590 	__u16 resv;
591 	__u32 resv2[3];
592 	struct io_uring_probe_op ops[];
593 };
594 
595 struct io_uring_restriction {
596 	__u16 opcode;
597 	union {
598 		__u8 register_op; /* IORING_RESTRICTION_REGISTER_OP */
599 		__u8 sqe_op;      /* IORING_RESTRICTION_SQE_OP */
600 		__u8 sqe_flags;   /* IORING_RESTRICTION_SQE_FLAGS_* */
601 	};
602 	__u8 resv;
603 	__u32 resv2[3];
604 };
605 
606 struct io_uring_buf {
607 	__u64	addr;
608 	__u32	len;
609 	__u16	bid;
610 	__u16	resv;
611 };
612 
613 struct io_uring_buf_ring {
614 	union {
615 		/*
616 		 * To avoid spilling into more pages than we need to, the
617 		 * ring tail is overlaid with the io_uring_buf->resv field.
618 		 */
619 		struct {
620 			__u64	resv1;
621 			__u32	resv2;
622 			__u16	resv3;
623 			__u16	tail;
624 		};
625 		__DECLARE_FLEX_ARRAY(struct io_uring_buf, bufs);
626 	};
627 };
628 
629 /*
630  * Flags for IORING_REGISTER_PBUF_RING.
631  *
632  * IOU_PBUF_RING_MMAP:	If set, kernel will allocate the memory for the ring.
633  *			The application must not set a ring_addr in struct
634  *			io_uring_buf_reg, instead it must subsequently call
635  *			mmap(2) with the offset set as:
636  *			IORING_OFF_PBUF_RING | (bgid << IORING_OFF_PBUF_SHIFT)
637  *			to get a virtual mapping for the ring.
638  */
639 enum {
640 	IOU_PBUF_RING_MMAP	= 1,
641 };
642 
643 /* argument for IORING_(UN)REGISTER_PBUF_RING */
644 struct io_uring_buf_reg {
645 	__u64	ring_addr;
646 	__u32	ring_entries;
647 	__u16	bgid;
648 	__u16	flags;
649 	__u64	resv[3];
650 };
651 
652 /*
653  * io_uring_restriction->opcode values
654  */
655 enum {
656 	/* Allow an io_uring_register(2) opcode */
657 	IORING_RESTRICTION_REGISTER_OP		= 0,
658 
659 	/* Allow an sqe opcode */
660 	IORING_RESTRICTION_SQE_OP		= 1,
661 
662 	/* Allow sqe flags */
663 	IORING_RESTRICTION_SQE_FLAGS_ALLOWED	= 2,
664 
665 	/* Require sqe flags (these flags must be set on each submission) */
666 	IORING_RESTRICTION_SQE_FLAGS_REQUIRED	= 3,
667 
668 	IORING_RESTRICTION_LAST
669 };
670 
671 struct io_uring_getevents_arg {
672 	__u64	sigmask;
673 	__u32	sigmask_sz;
674 	__u32	pad;
675 	__u64	ts;
676 };
677 
678 /*
679  * Argument for IORING_REGISTER_SYNC_CANCEL
680  */
681 struct io_uring_sync_cancel_reg {
682 	__u64				addr;
683 	__s32				fd;
684 	__u32				flags;
685 	struct __kernel_timespec	timeout;
686 	__u64				pad[4];
687 };
688 
689 /*
690  * Argument for IORING_REGISTER_FILE_ALLOC_RANGE
691  * The range is specified as [off, off + len)
692  */
693 struct io_uring_file_index_range {
694 	__u32	off;
695 	__u32	len;
696 	__u64	resv;
697 };
698 
699 struct io_uring_recvmsg_out {
700 	__u32 namelen;
701 	__u32 controllen;
702 	__u32 payloadlen;
703 	__u32 flags;
704 };
705 
706 #ifdef __cplusplus
707 }
708 #endif
709 
710 #endif
711