xref: /linux-6.15/include/uapi/linux/io_uring.h (revision bbaf1ff0)
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 /*
177  * Application provides the memory for the rings
178  */
179 #define IORING_SETUP_NO_MMAP		(1U << 14)
180 
181 /*
182  * Register the ring fd in itself for use with
183  * IORING_REGISTER_USE_REGISTERED_RING; return a registered fd index rather
184  * than an fd.
185  */
186 #define IORING_SETUP_REGISTERED_FD_ONLY	(1U << 15)
187 
188 enum io_uring_op {
189 	IORING_OP_NOP,
190 	IORING_OP_READV,
191 	IORING_OP_WRITEV,
192 	IORING_OP_FSYNC,
193 	IORING_OP_READ_FIXED,
194 	IORING_OP_WRITE_FIXED,
195 	IORING_OP_POLL_ADD,
196 	IORING_OP_POLL_REMOVE,
197 	IORING_OP_SYNC_FILE_RANGE,
198 	IORING_OP_SENDMSG,
199 	IORING_OP_RECVMSG,
200 	IORING_OP_TIMEOUT,
201 	IORING_OP_TIMEOUT_REMOVE,
202 	IORING_OP_ACCEPT,
203 	IORING_OP_ASYNC_CANCEL,
204 	IORING_OP_LINK_TIMEOUT,
205 	IORING_OP_CONNECT,
206 	IORING_OP_FALLOCATE,
207 	IORING_OP_OPENAT,
208 	IORING_OP_CLOSE,
209 	IORING_OP_FILES_UPDATE,
210 	IORING_OP_STATX,
211 	IORING_OP_READ,
212 	IORING_OP_WRITE,
213 	IORING_OP_FADVISE,
214 	IORING_OP_MADVISE,
215 	IORING_OP_SEND,
216 	IORING_OP_RECV,
217 	IORING_OP_OPENAT2,
218 	IORING_OP_EPOLL_CTL,
219 	IORING_OP_SPLICE,
220 	IORING_OP_PROVIDE_BUFFERS,
221 	IORING_OP_REMOVE_BUFFERS,
222 	IORING_OP_TEE,
223 	IORING_OP_SHUTDOWN,
224 	IORING_OP_RENAMEAT,
225 	IORING_OP_UNLINKAT,
226 	IORING_OP_MKDIRAT,
227 	IORING_OP_SYMLINKAT,
228 	IORING_OP_LINKAT,
229 	IORING_OP_MSG_RING,
230 	IORING_OP_FSETXATTR,
231 	IORING_OP_SETXATTR,
232 	IORING_OP_FGETXATTR,
233 	IORING_OP_GETXATTR,
234 	IORING_OP_SOCKET,
235 	IORING_OP_URING_CMD,
236 	IORING_OP_SEND_ZC,
237 	IORING_OP_SENDMSG_ZC,
238 
239 	/* this goes last, obviously */
240 	IORING_OP_LAST,
241 };
242 
243 /*
244  * sqe->uring_cmd_flags
245  * IORING_URING_CMD_FIXED	use registered buffer; pass this flag
246  *				along with setting sqe->buf_index.
247  */
248 #define IORING_URING_CMD_FIXED	(1U << 0)
249 
250 
251 /*
252  * sqe->fsync_flags
253  */
254 #define IORING_FSYNC_DATASYNC	(1U << 0)
255 
256 /*
257  * sqe->timeout_flags
258  */
259 #define IORING_TIMEOUT_ABS		(1U << 0)
260 #define IORING_TIMEOUT_UPDATE		(1U << 1)
261 #define IORING_TIMEOUT_BOOTTIME		(1U << 2)
262 #define IORING_TIMEOUT_REALTIME		(1U << 3)
263 #define IORING_LINK_TIMEOUT_UPDATE	(1U << 4)
264 #define IORING_TIMEOUT_ETIME_SUCCESS	(1U << 5)
265 #define IORING_TIMEOUT_MULTISHOT	(1U << 6)
266 #define IORING_TIMEOUT_CLOCK_MASK	(IORING_TIMEOUT_BOOTTIME | IORING_TIMEOUT_REALTIME)
267 #define IORING_TIMEOUT_UPDATE_MASK	(IORING_TIMEOUT_UPDATE | IORING_LINK_TIMEOUT_UPDATE)
268 /*
269  * sqe->splice_flags
270  * extends splice(2) flags
271  */
272 #define SPLICE_F_FD_IN_FIXED	(1U << 31) /* the last bit of __u32 */
273 
274 /*
275  * POLL_ADD flags. Note that since sqe->poll_events is the flag space, the
276  * command flags for POLL_ADD are stored in sqe->len.
277  *
278  * IORING_POLL_ADD_MULTI	Multishot poll. Sets IORING_CQE_F_MORE if
279  *				the poll handler will continue to report
280  *				CQEs on behalf of the same SQE.
281  *
282  * IORING_POLL_UPDATE		Update existing poll request, matching
283  *				sqe->addr as the old user_data field.
284  *
285  * IORING_POLL_LEVEL		Level triggered poll.
286  */
287 #define IORING_POLL_ADD_MULTI	(1U << 0)
288 #define IORING_POLL_UPDATE_EVENTS	(1U << 1)
289 #define IORING_POLL_UPDATE_USER_DATA	(1U << 2)
290 #define IORING_POLL_ADD_LEVEL		(1U << 3)
291 
292 /*
293  * ASYNC_CANCEL flags.
294  *
295  * IORING_ASYNC_CANCEL_ALL	Cancel all requests that match the given key
296  * IORING_ASYNC_CANCEL_FD	Key off 'fd' for cancelation rather than the
297  *				request 'user_data'
298  * IORING_ASYNC_CANCEL_ANY	Match any request
299  * IORING_ASYNC_CANCEL_FD_FIXED	'fd' passed in is a fixed descriptor
300  */
301 #define IORING_ASYNC_CANCEL_ALL	(1U << 0)
302 #define IORING_ASYNC_CANCEL_FD	(1U << 1)
303 #define IORING_ASYNC_CANCEL_ANY	(1U << 2)
304 #define IORING_ASYNC_CANCEL_FD_FIXED	(1U << 3)
305 
306 /*
307  * send/sendmsg and recv/recvmsg flags (sqe->ioprio)
308  *
309  * IORING_RECVSEND_POLL_FIRST	If set, instead of first attempting to send
310  *				or receive and arm poll if that yields an
311  *				-EAGAIN result, arm poll upfront and skip
312  *				the initial transfer attempt.
313  *
314  * IORING_RECV_MULTISHOT	Multishot recv. Sets IORING_CQE_F_MORE if
315  *				the handler will continue to report
316  *				CQEs on behalf of the same SQE.
317  *
318  * IORING_RECVSEND_FIXED_BUF	Use registered buffers, the index is stored in
319  *				the buf_index field.
320  *
321  * IORING_SEND_ZC_REPORT_USAGE
322  *				If set, SEND[MSG]_ZC should report
323  *				the zerocopy usage in cqe.res
324  *				for the IORING_CQE_F_NOTIF cqe.
325  *				0 is reported if zerocopy was actually possible.
326  *				IORING_NOTIF_USAGE_ZC_COPIED if data was copied
327  *				(at least partially).
328  */
329 #define IORING_RECVSEND_POLL_FIRST	(1U << 0)
330 #define IORING_RECV_MULTISHOT		(1U << 1)
331 #define IORING_RECVSEND_FIXED_BUF	(1U << 2)
332 #define IORING_SEND_ZC_REPORT_USAGE	(1U << 3)
333 
334 /*
335  * cqe.res for IORING_CQE_F_NOTIF if
336  * IORING_SEND_ZC_REPORT_USAGE was requested
337  *
338  * It should be treated as a flag, all other
339  * bits of cqe.res should be treated as reserved!
340  */
341 #define IORING_NOTIF_USAGE_ZC_COPIED    (1U << 31)
342 
343 /*
344  * accept flags stored in sqe->ioprio
345  */
346 #define IORING_ACCEPT_MULTISHOT	(1U << 0)
347 
348 /*
349  * IORING_OP_MSG_RING command types, stored in sqe->addr
350  */
351 enum {
352 	IORING_MSG_DATA,	/* pass sqe->len as 'res' and off as user_data */
353 	IORING_MSG_SEND_FD,	/* send a registered fd to another ring */
354 };
355 
356 /*
357  * IORING_OP_MSG_RING flags (sqe->msg_ring_flags)
358  *
359  * IORING_MSG_RING_CQE_SKIP	Don't post a CQE to the target ring. Not
360  *				applicable for IORING_MSG_DATA, obviously.
361  */
362 #define IORING_MSG_RING_CQE_SKIP	(1U << 0)
363 /* Pass through the flags from sqe->file_index to cqe->flags */
364 #define IORING_MSG_RING_FLAGS_PASS	(1U << 1)
365 
366 /*
367  * IO completion data structure (Completion Queue Entry)
368  */
369 struct io_uring_cqe {
370 	__u64	user_data;	/* sqe->data submission passed back */
371 	__s32	res;		/* result code for this event */
372 	__u32	flags;
373 
374 	/*
375 	 * If the ring is initialized with IORING_SETUP_CQE32, then this field
376 	 * contains 16-bytes of padding, doubling the size of the CQE.
377 	 */
378 	__u64 big_cqe[];
379 };
380 
381 /*
382  * cqe->flags
383  *
384  * IORING_CQE_F_BUFFER	If set, the upper 16 bits are the buffer ID
385  * IORING_CQE_F_MORE	If set, parent SQE will generate more CQE entries
386  * IORING_CQE_F_SOCK_NONEMPTY	If set, more data to read after socket recv
387  * IORING_CQE_F_NOTIF	Set for notification CQEs. Can be used to distinct
388  * 			them from sends.
389  */
390 #define IORING_CQE_F_BUFFER		(1U << 0)
391 #define IORING_CQE_F_MORE		(1U << 1)
392 #define IORING_CQE_F_SOCK_NONEMPTY	(1U << 2)
393 #define IORING_CQE_F_NOTIF		(1U << 3)
394 
395 enum {
396 	IORING_CQE_BUFFER_SHIFT		= 16,
397 };
398 
399 /*
400  * Magic offsets for the application to mmap the data it needs
401  */
402 #define IORING_OFF_SQ_RING		0ULL
403 #define IORING_OFF_CQ_RING		0x8000000ULL
404 #define IORING_OFF_SQES			0x10000000ULL
405 #define IORING_OFF_PBUF_RING		0x80000000ULL
406 #define IORING_OFF_PBUF_SHIFT		16
407 #define IORING_OFF_MMAP_MASK		0xf8000000ULL
408 
409 /*
410  * Filled with the offset for mmap(2)
411  */
412 struct io_sqring_offsets {
413 	__u32 head;
414 	__u32 tail;
415 	__u32 ring_mask;
416 	__u32 ring_entries;
417 	__u32 flags;
418 	__u32 dropped;
419 	__u32 array;
420 	__u32 resv1;
421 	__u64 user_addr;
422 };
423 
424 /*
425  * sq_ring->flags
426  */
427 #define IORING_SQ_NEED_WAKEUP	(1U << 0) /* needs io_uring_enter wakeup */
428 #define IORING_SQ_CQ_OVERFLOW	(1U << 1) /* CQ ring is overflown */
429 #define IORING_SQ_TASKRUN	(1U << 2) /* task should enter the kernel */
430 
431 struct io_cqring_offsets {
432 	__u32 head;
433 	__u32 tail;
434 	__u32 ring_mask;
435 	__u32 ring_entries;
436 	__u32 overflow;
437 	__u32 cqes;
438 	__u32 flags;
439 	__u32 resv1;
440 	__u64 user_addr;
441 };
442 
443 /*
444  * cq_ring->flags
445  */
446 
447 /* disable eventfd notifications */
448 #define IORING_CQ_EVENTFD_DISABLED	(1U << 0)
449 
450 /*
451  * io_uring_enter(2) flags
452  */
453 #define IORING_ENTER_GETEVENTS		(1U << 0)
454 #define IORING_ENTER_SQ_WAKEUP		(1U << 1)
455 #define IORING_ENTER_SQ_WAIT		(1U << 2)
456 #define IORING_ENTER_EXT_ARG		(1U << 3)
457 #define IORING_ENTER_REGISTERED_RING	(1U << 4)
458 
459 /*
460  * Passed in for io_uring_setup(2). Copied back with updated info on success
461  */
462 struct io_uring_params {
463 	__u32 sq_entries;
464 	__u32 cq_entries;
465 	__u32 flags;
466 	__u32 sq_thread_cpu;
467 	__u32 sq_thread_idle;
468 	__u32 features;
469 	__u32 wq_fd;
470 	__u32 resv[3];
471 	struct io_sqring_offsets sq_off;
472 	struct io_cqring_offsets cq_off;
473 };
474 
475 /*
476  * io_uring_params->features flags
477  */
478 #define IORING_FEAT_SINGLE_MMAP		(1U << 0)
479 #define IORING_FEAT_NODROP		(1U << 1)
480 #define IORING_FEAT_SUBMIT_STABLE	(1U << 2)
481 #define IORING_FEAT_RW_CUR_POS		(1U << 3)
482 #define IORING_FEAT_CUR_PERSONALITY	(1U << 4)
483 #define IORING_FEAT_FAST_POLL		(1U << 5)
484 #define IORING_FEAT_POLL_32BITS 	(1U << 6)
485 #define IORING_FEAT_SQPOLL_NONFIXED	(1U << 7)
486 #define IORING_FEAT_EXT_ARG		(1U << 8)
487 #define IORING_FEAT_NATIVE_WORKERS	(1U << 9)
488 #define IORING_FEAT_RSRC_TAGS		(1U << 10)
489 #define IORING_FEAT_CQE_SKIP		(1U << 11)
490 #define IORING_FEAT_LINKED_FILE		(1U << 12)
491 #define IORING_FEAT_REG_REG_RING	(1U << 13)
492 
493 /*
494  * io_uring_register(2) opcodes and arguments
495  */
496 enum {
497 	IORING_REGISTER_BUFFERS			= 0,
498 	IORING_UNREGISTER_BUFFERS		= 1,
499 	IORING_REGISTER_FILES			= 2,
500 	IORING_UNREGISTER_FILES			= 3,
501 	IORING_REGISTER_EVENTFD			= 4,
502 	IORING_UNREGISTER_EVENTFD		= 5,
503 	IORING_REGISTER_FILES_UPDATE		= 6,
504 	IORING_REGISTER_EVENTFD_ASYNC		= 7,
505 	IORING_REGISTER_PROBE			= 8,
506 	IORING_REGISTER_PERSONALITY		= 9,
507 	IORING_UNREGISTER_PERSONALITY		= 10,
508 	IORING_REGISTER_RESTRICTIONS		= 11,
509 	IORING_REGISTER_ENABLE_RINGS		= 12,
510 
511 	/* extended with tagging */
512 	IORING_REGISTER_FILES2			= 13,
513 	IORING_REGISTER_FILES_UPDATE2		= 14,
514 	IORING_REGISTER_BUFFERS2		= 15,
515 	IORING_REGISTER_BUFFERS_UPDATE		= 16,
516 
517 	/* set/clear io-wq thread affinities */
518 	IORING_REGISTER_IOWQ_AFF		= 17,
519 	IORING_UNREGISTER_IOWQ_AFF		= 18,
520 
521 	/* set/get max number of io-wq workers */
522 	IORING_REGISTER_IOWQ_MAX_WORKERS	= 19,
523 
524 	/* register/unregister io_uring fd with the ring */
525 	IORING_REGISTER_RING_FDS		= 20,
526 	IORING_UNREGISTER_RING_FDS		= 21,
527 
528 	/* register ring based provide buffer group */
529 	IORING_REGISTER_PBUF_RING		= 22,
530 	IORING_UNREGISTER_PBUF_RING		= 23,
531 
532 	/* sync cancelation API */
533 	IORING_REGISTER_SYNC_CANCEL		= 24,
534 
535 	/* register a range of fixed file slots for automatic slot allocation */
536 	IORING_REGISTER_FILE_ALLOC_RANGE	= 25,
537 
538 	/* this goes last */
539 	IORING_REGISTER_LAST,
540 
541 	/* flag added to the opcode to use a registered ring fd */
542 	IORING_REGISTER_USE_REGISTERED_RING	= 1U << 31
543 };
544 
545 /* io-wq worker categories */
546 enum {
547 	IO_WQ_BOUND,
548 	IO_WQ_UNBOUND,
549 };
550 
551 /* deprecated, see struct io_uring_rsrc_update */
552 struct io_uring_files_update {
553 	__u32 offset;
554 	__u32 resv;
555 	__aligned_u64 /* __s32 * */ fds;
556 };
557 
558 /*
559  * Register a fully sparse file space, rather than pass in an array of all
560  * -1 file descriptors.
561  */
562 #define IORING_RSRC_REGISTER_SPARSE	(1U << 0)
563 
564 struct io_uring_rsrc_register {
565 	__u32 nr;
566 	__u32 flags;
567 	__u64 resv2;
568 	__aligned_u64 data;
569 	__aligned_u64 tags;
570 };
571 
572 struct io_uring_rsrc_update {
573 	__u32 offset;
574 	__u32 resv;
575 	__aligned_u64 data;
576 };
577 
578 struct io_uring_rsrc_update2 {
579 	__u32 offset;
580 	__u32 resv;
581 	__aligned_u64 data;
582 	__aligned_u64 tags;
583 	__u32 nr;
584 	__u32 resv2;
585 };
586 
587 /* Skip updating fd indexes set to this value in the fd table */
588 #define IORING_REGISTER_FILES_SKIP	(-2)
589 
590 #define IO_URING_OP_SUPPORTED	(1U << 0)
591 
592 struct io_uring_probe_op {
593 	__u8 op;
594 	__u8 resv;
595 	__u16 flags;	/* IO_URING_OP_* flags */
596 	__u32 resv2;
597 };
598 
599 struct io_uring_probe {
600 	__u8 last_op;	/* last opcode supported */
601 	__u8 ops_len;	/* length of ops[] array below */
602 	__u16 resv;
603 	__u32 resv2[3];
604 	struct io_uring_probe_op ops[];
605 };
606 
607 struct io_uring_restriction {
608 	__u16 opcode;
609 	union {
610 		__u8 register_op; /* IORING_RESTRICTION_REGISTER_OP */
611 		__u8 sqe_op;      /* IORING_RESTRICTION_SQE_OP */
612 		__u8 sqe_flags;   /* IORING_RESTRICTION_SQE_FLAGS_* */
613 	};
614 	__u8 resv;
615 	__u32 resv2[3];
616 };
617 
618 struct io_uring_buf {
619 	__u64	addr;
620 	__u32	len;
621 	__u16	bid;
622 	__u16	resv;
623 };
624 
625 struct io_uring_buf_ring {
626 	union {
627 		/*
628 		 * To avoid spilling into more pages than we need to, the
629 		 * ring tail is overlaid with the io_uring_buf->resv field.
630 		 */
631 		struct {
632 			__u64	resv1;
633 			__u32	resv2;
634 			__u16	resv3;
635 			__u16	tail;
636 		};
637 		__DECLARE_FLEX_ARRAY(struct io_uring_buf, bufs);
638 	};
639 };
640 
641 /*
642  * Flags for IORING_REGISTER_PBUF_RING.
643  *
644  * IOU_PBUF_RING_MMAP:	If set, kernel will allocate the memory for the ring.
645  *			The application must not set a ring_addr in struct
646  *			io_uring_buf_reg, instead it must subsequently call
647  *			mmap(2) with the offset set as:
648  *			IORING_OFF_PBUF_RING | (bgid << IORING_OFF_PBUF_SHIFT)
649  *			to get a virtual mapping for the ring.
650  */
651 enum {
652 	IOU_PBUF_RING_MMAP	= 1,
653 };
654 
655 /* argument for IORING_(UN)REGISTER_PBUF_RING */
656 struct io_uring_buf_reg {
657 	__u64	ring_addr;
658 	__u32	ring_entries;
659 	__u16	bgid;
660 	__u16	flags;
661 	__u64	resv[3];
662 };
663 
664 /*
665  * io_uring_restriction->opcode values
666  */
667 enum {
668 	/* Allow an io_uring_register(2) opcode */
669 	IORING_RESTRICTION_REGISTER_OP		= 0,
670 
671 	/* Allow an sqe opcode */
672 	IORING_RESTRICTION_SQE_OP		= 1,
673 
674 	/* Allow sqe flags */
675 	IORING_RESTRICTION_SQE_FLAGS_ALLOWED	= 2,
676 
677 	/* Require sqe flags (these flags must be set on each submission) */
678 	IORING_RESTRICTION_SQE_FLAGS_REQUIRED	= 3,
679 
680 	IORING_RESTRICTION_LAST
681 };
682 
683 struct io_uring_getevents_arg {
684 	__u64	sigmask;
685 	__u32	sigmask_sz;
686 	__u32	pad;
687 	__u64	ts;
688 };
689 
690 /*
691  * Argument for IORING_REGISTER_SYNC_CANCEL
692  */
693 struct io_uring_sync_cancel_reg {
694 	__u64				addr;
695 	__s32				fd;
696 	__u32				flags;
697 	struct __kernel_timespec	timeout;
698 	__u64				pad[4];
699 };
700 
701 /*
702  * Argument for IORING_REGISTER_FILE_ALLOC_RANGE
703  * The range is specified as [off, off + len)
704  */
705 struct io_uring_file_index_range {
706 	__u32	off;
707 	__u32	len;
708 	__u64	resv;
709 };
710 
711 struct io_uring_recvmsg_out {
712 	__u32 namelen;
713 	__u32 controllen;
714 	__u32 payloadlen;
715 	__u32 flags;
716 };
717 
718 #ifdef __cplusplus
719 }
720 #endif
721 
722 #endif
723