xref: /linux/include/uapi/linux/io_uring.h (revision a3d14d1602ca11429d242d230c31af8f822f614f)
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 		struct {
47 			__u32	level;
48 			__u32	optname;
49 		};
50 	};
51 	__u32	len;		/* buffer size or number of iovecs */
52 	union {
53 		__kernel_rwf_t	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 */
123 struct 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 
141 enum 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 with COOP_TASKRUN.
192  */
193 #define IORING_SETUP_TASKRUN_FLAG	(1U << 9)
194 #define IORING_SETUP_SQE128		(1U << 10) /* SQEs are 128 byte */
195 #define IORING_SETUP_CQE32		(1U << 11) /* CQEs are 32 byte */
196 /*
197  * Only one task is allowed to submit requests
198  */
199 #define IORING_SETUP_SINGLE_ISSUER	(1U << 12)
200 
201 /*
202  * Defer running task work to get events.
203  * Rather than running bits of task work whenever the task transitions
204  * try to do it just before it is needed.
205  */
206 #define IORING_SETUP_DEFER_TASKRUN	(1U << 13)
207 
208 /*
209  * Application provides the memory for the rings
210  */
211 #define IORING_SETUP_NO_MMAP		(1U << 14)
212 
213 /*
214  * Register the ring fd in itself for use with
215  * IORING_REGISTER_USE_REGISTERED_RING; return a registered fd index rather
216  * than an fd.
217  */
218 #define IORING_SETUP_REGISTERED_FD_ONLY	(1U << 15)
219 
220 /*
221  * Removes indirection through the SQ index array.
222  */
223 #define IORING_SETUP_NO_SQARRAY		(1U << 16)
224 
225 /* Use hybrid poll in iopoll process */
226 #define IORING_SETUP_HYBRID_IOPOLL	(1U << 17)
227 
228 enum io_uring_op {
229 	IORING_OP_NOP,
230 	IORING_OP_READV,
231 	IORING_OP_WRITEV,
232 	IORING_OP_FSYNC,
233 	IORING_OP_READ_FIXED,
234 	IORING_OP_WRITE_FIXED,
235 	IORING_OP_POLL_ADD,
236 	IORING_OP_POLL_REMOVE,
237 	IORING_OP_SYNC_FILE_RANGE,
238 	IORING_OP_SENDMSG,
239 	IORING_OP_RECVMSG,
240 	IORING_OP_TIMEOUT,
241 	IORING_OP_TIMEOUT_REMOVE,
242 	IORING_OP_ACCEPT,
243 	IORING_OP_ASYNC_CANCEL,
244 	IORING_OP_LINK_TIMEOUT,
245 	IORING_OP_CONNECT,
246 	IORING_OP_FALLOCATE,
247 	IORING_OP_OPENAT,
248 	IORING_OP_CLOSE,
249 	IORING_OP_FILES_UPDATE,
250 	IORING_OP_STATX,
251 	IORING_OP_READ,
252 	IORING_OP_WRITE,
253 	IORING_OP_FADVISE,
254 	IORING_OP_MADVISE,
255 	IORING_OP_SEND,
256 	IORING_OP_RECV,
257 	IORING_OP_OPENAT2,
258 	IORING_OP_EPOLL_CTL,
259 	IORING_OP_SPLICE,
260 	IORING_OP_PROVIDE_BUFFERS,
261 	IORING_OP_REMOVE_BUFFERS,
262 	IORING_OP_TEE,
263 	IORING_OP_SHUTDOWN,
264 	IORING_OP_RENAMEAT,
265 	IORING_OP_UNLINKAT,
266 	IORING_OP_MKDIRAT,
267 	IORING_OP_SYMLINKAT,
268 	IORING_OP_LINKAT,
269 	IORING_OP_MSG_RING,
270 	IORING_OP_FSETXATTR,
271 	IORING_OP_SETXATTR,
272 	IORING_OP_FGETXATTR,
273 	IORING_OP_GETXATTR,
274 	IORING_OP_SOCKET,
275 	IORING_OP_URING_CMD,
276 	IORING_OP_SEND_ZC,
277 	IORING_OP_SENDMSG_ZC,
278 	IORING_OP_READ_MULTISHOT,
279 	IORING_OP_WAITID,
280 	IORING_OP_FUTEX_WAIT,
281 	IORING_OP_FUTEX_WAKE,
282 	IORING_OP_FUTEX_WAITV,
283 	IORING_OP_FIXED_FD_INSTALL,
284 	IORING_OP_FTRUNCATE,
285 	IORING_OP_BIND,
286 	IORING_OP_LISTEN,
287 	IORING_OP_RECV_ZC,
288 	IORING_OP_EPOLL_WAIT,
289 	IORING_OP_READV_FIXED,
290 	IORING_OP_WRITEV_FIXED,
291 	IORING_OP_PIPE,
292 
293 	/* this goes last, obviously */
294 	IORING_OP_LAST,
295 };
296 
297 /*
298  * sqe->uring_cmd_flags		top 8bits aren't available for userspace
299  * IORING_URING_CMD_FIXED	use registered buffer; pass this flag
300  *				along with setting sqe->buf_index.
301  */
302 #define IORING_URING_CMD_FIXED	(1U << 0)
303 #define IORING_URING_CMD_MASK	IORING_URING_CMD_FIXED
304 
305 
306 /*
307  * sqe->fsync_flags
308  */
309 #define IORING_FSYNC_DATASYNC	(1U << 0)
310 
311 /*
312  * sqe->timeout_flags
313  */
314 #define IORING_TIMEOUT_ABS		(1U << 0)
315 #define IORING_TIMEOUT_UPDATE		(1U << 1)
316 #define IORING_TIMEOUT_BOOTTIME		(1U << 2)
317 #define IORING_TIMEOUT_REALTIME		(1U << 3)
318 #define IORING_LINK_TIMEOUT_UPDATE	(1U << 4)
319 #define IORING_TIMEOUT_ETIME_SUCCESS	(1U << 5)
320 #define IORING_TIMEOUT_MULTISHOT	(1U << 6)
321 #define IORING_TIMEOUT_CLOCK_MASK	(IORING_TIMEOUT_BOOTTIME | IORING_TIMEOUT_REALTIME)
322 #define IORING_TIMEOUT_UPDATE_MASK	(IORING_TIMEOUT_UPDATE | IORING_LINK_TIMEOUT_UPDATE)
323 /*
324  * sqe->splice_flags
325  * extends splice(2) flags
326  */
327 #define SPLICE_F_FD_IN_FIXED	(1U << 31) /* the last bit of __u32 */
328 
329 /*
330  * POLL_ADD flags. Note that since sqe->poll_events is the flag space, the
331  * command flags for POLL_ADD are stored in sqe->len.
332  *
333  * IORING_POLL_ADD_MULTI	Multishot poll. Sets IORING_CQE_F_MORE if
334  *				the poll handler will continue to report
335  *				CQEs on behalf of the same SQE.
336  *
337  * IORING_POLL_UPDATE		Update existing poll request, matching
338  *				sqe->addr as the old user_data field.
339  *
340  * IORING_POLL_LEVEL		Level triggered poll.
341  */
342 #define IORING_POLL_ADD_MULTI	(1U << 0)
343 #define IORING_POLL_UPDATE_EVENTS	(1U << 1)
344 #define IORING_POLL_UPDATE_USER_DATA	(1U << 2)
345 #define IORING_POLL_ADD_LEVEL		(1U << 3)
346 
347 /*
348  * ASYNC_CANCEL flags.
349  *
350  * IORING_ASYNC_CANCEL_ALL	Cancel all requests that match the given key
351  * IORING_ASYNC_CANCEL_FD	Key off 'fd' for cancelation rather than the
352  *				request 'user_data'
353  * IORING_ASYNC_CANCEL_ANY	Match any request
354  * IORING_ASYNC_CANCEL_FD_FIXED	'fd' passed in is a fixed descriptor
355  * IORING_ASYNC_CANCEL_USERDATA	Match on user_data, default for no other key
356  * IORING_ASYNC_CANCEL_OP	Match request based on opcode
357  */
358 #define IORING_ASYNC_CANCEL_ALL	(1U << 0)
359 #define IORING_ASYNC_CANCEL_FD	(1U << 1)
360 #define IORING_ASYNC_CANCEL_ANY	(1U << 2)
361 #define IORING_ASYNC_CANCEL_FD_FIXED	(1U << 3)
362 #define IORING_ASYNC_CANCEL_USERDATA	(1U << 4)
363 #define IORING_ASYNC_CANCEL_OP	(1U << 5)
364 
365 /*
366  * send/sendmsg and recv/recvmsg flags (sqe->ioprio)
367  *
368  * IORING_RECVSEND_POLL_FIRST	If set, instead of first attempting to send
369  *				or receive and arm poll if that yields an
370  *				-EAGAIN result, arm poll upfront and skip
371  *				the initial transfer attempt.
372  *
373  * IORING_RECV_MULTISHOT	Multishot recv. Sets IORING_CQE_F_MORE if
374  *				the handler will continue to report
375  *				CQEs on behalf of the same SQE.
376  *
377  * IORING_RECVSEND_FIXED_BUF	Use registered buffers, the index is stored in
378  *				the buf_index field.
379  *
380  * IORING_SEND_ZC_REPORT_USAGE
381  *				If set, SEND[MSG]_ZC should report
382  *				the zerocopy usage in cqe.res
383  *				for the IORING_CQE_F_NOTIF cqe.
384  *				0 is reported if zerocopy was actually possible.
385  *				IORING_NOTIF_USAGE_ZC_COPIED if data was copied
386  *				(at least partially).
387  *
388  * IORING_RECVSEND_BUNDLE	Used with IOSQE_BUFFER_SELECT. If set, send or
389  *				recv will grab as many buffers from the buffer
390  *				group ID given and send them all. The completion
391  *				result 	will be the number of buffers send, with
392  *				the starting buffer ID in cqe->flags as per
393  *				usual for provided buffer usage. The buffers
394  *				will be	contiguous from the starting buffer ID.
395  */
396 #define IORING_RECVSEND_POLL_FIRST	(1U << 0)
397 #define IORING_RECV_MULTISHOT		(1U << 1)
398 #define IORING_RECVSEND_FIXED_BUF	(1U << 2)
399 #define IORING_SEND_ZC_REPORT_USAGE	(1U << 3)
400 #define IORING_RECVSEND_BUNDLE		(1U << 4)
401 
402 /*
403  * cqe.res for IORING_CQE_F_NOTIF if
404  * IORING_SEND_ZC_REPORT_USAGE was requested
405  *
406  * It should be treated as a flag, all other
407  * bits of cqe.res should be treated as reserved!
408  */
409 #define IORING_NOTIF_USAGE_ZC_COPIED    (1U << 31)
410 
411 /*
412  * accept flags stored in sqe->ioprio
413  */
414 #define IORING_ACCEPT_MULTISHOT	(1U << 0)
415 #define IORING_ACCEPT_DONTWAIT	(1U << 1)
416 #define IORING_ACCEPT_POLL_FIRST	(1U << 2)
417 
418 /*
419  * IORING_OP_MSG_RING command types, stored in sqe->addr
420  */
421 enum io_uring_msg_ring_flags {
422 	IORING_MSG_DATA,	/* pass sqe->len as 'res' and off as user_data */
423 	IORING_MSG_SEND_FD,	/* send a registered fd to another ring */
424 };
425 
426 /*
427  * IORING_OP_MSG_RING flags (sqe->msg_ring_flags)
428  *
429  * IORING_MSG_RING_CQE_SKIP	Don't post a CQE to the target ring. Not
430  *				applicable for IORING_MSG_DATA, obviously.
431  */
432 #define IORING_MSG_RING_CQE_SKIP	(1U << 0)
433 /* Pass through the flags from sqe->file_index to cqe->flags */
434 #define IORING_MSG_RING_FLAGS_PASS	(1U << 1)
435 
436 /*
437  * IORING_OP_FIXED_FD_INSTALL flags (sqe->install_fd_flags)
438  *
439  * IORING_FIXED_FD_NO_CLOEXEC	Don't mark the fd as O_CLOEXEC
440  */
441 #define IORING_FIXED_FD_NO_CLOEXEC	(1U << 0)
442 
443 /*
444  * IORING_OP_NOP flags (sqe->nop_flags)
445  *
446  * IORING_NOP_INJECT_RESULT	Inject result from sqe->result
447  */
448 #define IORING_NOP_INJECT_RESULT	(1U << 0)
449 #define IORING_NOP_FILE			(1U << 1)
450 #define IORING_NOP_FIXED_FILE		(1U << 2)
451 #define IORING_NOP_FIXED_BUFFER		(1U << 3)
452 
453 /*
454  * IO completion data structure (Completion Queue Entry)
455  */
456 struct io_uring_cqe {
457 	__u64	user_data;	/* sqe->user_data value passed back */
458 	__s32	res;		/* result code for this event */
459 	__u32	flags;
460 
461 	/*
462 	 * If the ring is initialized with IORING_SETUP_CQE32, then this field
463 	 * contains 16-bytes of padding, doubling the size of the CQE.
464 	 */
465 	__u64 big_cqe[];
466 };
467 
468 /*
469  * cqe->flags
470  *
471  * IORING_CQE_F_BUFFER	If set, the upper 16 bits are the buffer ID
472  * IORING_CQE_F_MORE	If set, parent SQE will generate more CQE entries
473  * IORING_CQE_F_SOCK_NONEMPTY	If set, more data to read after socket recv
474  * IORING_CQE_F_NOTIF	Set for notification CQEs. Can be used to distinct
475  * 			them from sends.
476  * IORING_CQE_F_BUF_MORE If set, the buffer ID set in the completion will get
477  *			more completions. In other words, the buffer is being
478  *			partially consumed, and will be used by the kernel for
479  *			more completions. This is only set for buffers used via
480  *			the incremental buffer consumption, as provided by
481  *			a ring buffer setup with IOU_PBUF_RING_INC. For any
482  *			other provided buffer type, all completions with a
483  *			buffer passed back is automatically returned to the
484  *			application.
485  */
486 #define IORING_CQE_F_BUFFER		(1U << 0)
487 #define IORING_CQE_F_MORE		(1U << 1)
488 #define IORING_CQE_F_SOCK_NONEMPTY	(1U << 2)
489 #define IORING_CQE_F_NOTIF		(1U << 3)
490 #define IORING_CQE_F_BUF_MORE		(1U << 4)
491 
492 #define IORING_CQE_BUFFER_SHIFT		16
493 
494 /*
495  * Magic offsets for the application to mmap the data it needs
496  */
497 #define IORING_OFF_SQ_RING		0ULL
498 #define IORING_OFF_CQ_RING		0x8000000ULL
499 #define IORING_OFF_SQES			0x10000000ULL
500 #define IORING_OFF_PBUF_RING		0x80000000ULL
501 #define IORING_OFF_PBUF_SHIFT		16
502 #define IORING_OFF_MMAP_MASK		0xf8000000ULL
503 
504 /*
505  * Filled with the offset for mmap(2)
506  */
507 struct io_sqring_offsets {
508 	__u32 head;
509 	__u32 tail;
510 	__u32 ring_mask;
511 	__u32 ring_entries;
512 	__u32 flags;
513 	__u32 dropped;
514 	__u32 array;
515 	__u32 resv1;
516 	__u64 user_addr;
517 };
518 
519 /*
520  * sq_ring->flags
521  */
522 #define IORING_SQ_NEED_WAKEUP	(1U << 0) /* needs io_uring_enter wakeup */
523 #define IORING_SQ_CQ_OVERFLOW	(1U << 1) /* CQ ring is overflown */
524 #define IORING_SQ_TASKRUN	(1U << 2) /* task should enter the kernel */
525 
526 struct io_cqring_offsets {
527 	__u32 head;
528 	__u32 tail;
529 	__u32 ring_mask;
530 	__u32 ring_entries;
531 	__u32 overflow;
532 	__u32 cqes;
533 	__u32 flags;
534 	__u32 resv1;
535 	__u64 user_addr;
536 };
537 
538 /*
539  * cq_ring->flags
540  */
541 
542 /* disable eventfd notifications */
543 #define IORING_CQ_EVENTFD_DISABLED	(1U << 0)
544 
545 /*
546  * io_uring_enter(2) flags
547  */
548 #define IORING_ENTER_GETEVENTS		(1U << 0)
549 #define IORING_ENTER_SQ_WAKEUP		(1U << 1)
550 #define IORING_ENTER_SQ_WAIT		(1U << 2)
551 #define IORING_ENTER_EXT_ARG		(1U << 3)
552 #define IORING_ENTER_REGISTERED_RING	(1U << 4)
553 #define IORING_ENTER_ABS_TIMER		(1U << 5)
554 #define IORING_ENTER_EXT_ARG_REG	(1U << 6)
555 #define IORING_ENTER_NO_IOWAIT		(1U << 7)
556 
557 /*
558  * Passed in for io_uring_setup(2). Copied back with updated info on success
559  */
560 struct io_uring_params {
561 	__u32 sq_entries;
562 	__u32 cq_entries;
563 	__u32 flags;
564 	__u32 sq_thread_cpu;
565 	__u32 sq_thread_idle;
566 	__u32 features;
567 	__u32 wq_fd;
568 	__u32 resv[3];
569 	struct io_sqring_offsets sq_off;
570 	struct io_cqring_offsets cq_off;
571 };
572 
573 /*
574  * io_uring_params->features flags
575  */
576 #define IORING_FEAT_SINGLE_MMAP		(1U << 0)
577 #define IORING_FEAT_NODROP		(1U << 1)
578 #define IORING_FEAT_SUBMIT_STABLE	(1U << 2)
579 #define IORING_FEAT_RW_CUR_POS		(1U << 3)
580 #define IORING_FEAT_CUR_PERSONALITY	(1U << 4)
581 #define IORING_FEAT_FAST_POLL		(1U << 5)
582 #define IORING_FEAT_POLL_32BITS 	(1U << 6)
583 #define IORING_FEAT_SQPOLL_NONFIXED	(1U << 7)
584 #define IORING_FEAT_EXT_ARG		(1U << 8)
585 #define IORING_FEAT_NATIVE_WORKERS	(1U << 9)
586 #define IORING_FEAT_RSRC_TAGS		(1U << 10)
587 #define IORING_FEAT_CQE_SKIP		(1U << 11)
588 #define IORING_FEAT_LINKED_FILE		(1U << 12)
589 #define IORING_FEAT_REG_REG_RING	(1U << 13)
590 #define IORING_FEAT_RECVSEND_BUNDLE	(1U << 14)
591 #define IORING_FEAT_MIN_TIMEOUT		(1U << 15)
592 #define IORING_FEAT_RW_ATTR		(1U << 16)
593 #define IORING_FEAT_NO_IOWAIT		(1U << 17)
594 
595 /*
596  * io_uring_register(2) opcodes and arguments
597  */
598 enum io_uring_register_op {
599 	IORING_REGISTER_BUFFERS			= 0,
600 	IORING_UNREGISTER_BUFFERS		= 1,
601 	IORING_REGISTER_FILES			= 2,
602 	IORING_UNREGISTER_FILES			= 3,
603 	IORING_REGISTER_EVENTFD			= 4,
604 	IORING_UNREGISTER_EVENTFD		= 5,
605 	IORING_REGISTER_FILES_UPDATE		= 6,
606 	IORING_REGISTER_EVENTFD_ASYNC		= 7,
607 	IORING_REGISTER_PROBE			= 8,
608 	IORING_REGISTER_PERSONALITY		= 9,
609 	IORING_UNREGISTER_PERSONALITY		= 10,
610 	IORING_REGISTER_RESTRICTIONS		= 11,
611 	IORING_REGISTER_ENABLE_RINGS		= 12,
612 
613 	/* extended with tagging */
614 	IORING_REGISTER_FILES2			= 13,
615 	IORING_REGISTER_FILES_UPDATE2		= 14,
616 	IORING_REGISTER_BUFFERS2		= 15,
617 	IORING_REGISTER_BUFFERS_UPDATE		= 16,
618 
619 	/* set/clear io-wq thread affinities */
620 	IORING_REGISTER_IOWQ_AFF		= 17,
621 	IORING_UNREGISTER_IOWQ_AFF		= 18,
622 
623 	/* set/get max number of io-wq workers */
624 	IORING_REGISTER_IOWQ_MAX_WORKERS	= 19,
625 
626 	/* register/unregister io_uring fd with the ring */
627 	IORING_REGISTER_RING_FDS		= 20,
628 	IORING_UNREGISTER_RING_FDS		= 21,
629 
630 	/* register ring based provide buffer group */
631 	IORING_REGISTER_PBUF_RING		= 22,
632 	IORING_UNREGISTER_PBUF_RING		= 23,
633 
634 	/* sync cancelation API */
635 	IORING_REGISTER_SYNC_CANCEL		= 24,
636 
637 	/* register a range of fixed file slots for automatic slot allocation */
638 	IORING_REGISTER_FILE_ALLOC_RANGE	= 25,
639 
640 	/* return status information for a buffer group */
641 	IORING_REGISTER_PBUF_STATUS		= 26,
642 
643 	/* set/clear busy poll settings */
644 	IORING_REGISTER_NAPI			= 27,
645 	IORING_UNREGISTER_NAPI			= 28,
646 
647 	IORING_REGISTER_CLOCK			= 29,
648 
649 	/* clone registered buffers from source ring to current ring */
650 	IORING_REGISTER_CLONE_BUFFERS		= 30,
651 
652 	/* send MSG_RING without having a ring */
653 	IORING_REGISTER_SEND_MSG_RING		= 31,
654 
655 	/* register a netdev hw rx queue for zerocopy */
656 	IORING_REGISTER_ZCRX_IFQ		= 32,
657 
658 	/* resize CQ ring */
659 	IORING_REGISTER_RESIZE_RINGS		= 33,
660 
661 	IORING_REGISTER_MEM_REGION		= 34,
662 
663 	/* this goes last */
664 	IORING_REGISTER_LAST,
665 
666 	/* flag added to the opcode to use a registered ring fd */
667 	IORING_REGISTER_USE_REGISTERED_RING	= 1U << 31
668 };
669 
670 /* io-wq worker categories */
671 enum io_wq_type {
672 	IO_WQ_BOUND,
673 	IO_WQ_UNBOUND,
674 };
675 
676 /* deprecated, see struct io_uring_rsrc_update */
677 struct io_uring_files_update {
678 	__u32 offset;
679 	__u32 resv;
680 	__aligned_u64 /* __s32 * */ fds;
681 };
682 
683 enum {
684 	/* initialise with user provided memory pointed by user_addr */
685 	IORING_MEM_REGION_TYPE_USER		= 1,
686 };
687 
688 struct io_uring_region_desc {
689 	__u64 user_addr;
690 	__u64 size;
691 	__u32 flags;
692 	__u32 id;
693 	__u64 mmap_offset;
694 	__u64 __resv[4];
695 };
696 
697 enum {
698 	/* expose the region as registered wait arguments */
699 	IORING_MEM_REGION_REG_WAIT_ARG		= 1,
700 };
701 
702 struct io_uring_mem_region_reg {
703 	__u64 region_uptr; /* struct io_uring_region_desc * */
704 	__u64 flags;
705 	__u64 __resv[2];
706 };
707 
708 /*
709  * Register a fully sparse file space, rather than pass in an array of all
710  * -1 file descriptors.
711  */
712 #define IORING_RSRC_REGISTER_SPARSE	(1U << 0)
713 
714 struct io_uring_rsrc_register {
715 	__u32 nr;
716 	__u32 flags;
717 	__u64 resv2;
718 	__aligned_u64 data;
719 	__aligned_u64 tags;
720 };
721 
722 struct io_uring_rsrc_update {
723 	__u32 offset;
724 	__u32 resv;
725 	__aligned_u64 data;
726 };
727 
728 struct io_uring_rsrc_update2 {
729 	__u32 offset;
730 	__u32 resv;
731 	__aligned_u64 data;
732 	__aligned_u64 tags;
733 	__u32 nr;
734 	__u32 resv2;
735 };
736 
737 /* Skip updating fd indexes set to this value in the fd table */
738 #define IORING_REGISTER_FILES_SKIP	(-2)
739 
740 #define IO_URING_OP_SUPPORTED	(1U << 0)
741 
742 struct io_uring_probe_op {
743 	__u8 op;
744 	__u8 resv;
745 	__u16 flags;	/* IO_URING_OP_* flags */
746 	__u32 resv2;
747 };
748 
749 struct io_uring_probe {
750 	__u8 last_op;	/* last opcode supported */
751 	__u8 ops_len;	/* length of ops[] array below */
752 	__u16 resv;
753 	__u32 resv2[3];
754 	struct io_uring_probe_op ops[];
755 };
756 
757 struct io_uring_restriction {
758 	__u16 opcode;
759 	union {
760 		__u8 register_op; /* IORING_RESTRICTION_REGISTER_OP */
761 		__u8 sqe_op;      /* IORING_RESTRICTION_SQE_OP */
762 		__u8 sqe_flags;   /* IORING_RESTRICTION_SQE_FLAGS_* */
763 	};
764 	__u8 resv;
765 	__u32 resv2[3];
766 };
767 
768 struct io_uring_clock_register {
769 	__u32	clockid;
770 	__u32	__resv[3];
771 };
772 
773 enum {
774 	IORING_REGISTER_SRC_REGISTERED	= (1U << 0),
775 	IORING_REGISTER_DST_REPLACE	= (1U << 1),
776 };
777 
778 struct io_uring_clone_buffers {
779 	__u32	src_fd;
780 	__u32	flags;
781 	__u32	src_off;
782 	__u32	dst_off;
783 	__u32	nr;
784 	__u32	pad[3];
785 };
786 
787 struct io_uring_buf {
788 	__u64	addr;
789 	__u32	len;
790 	__u16	bid;
791 	__u16	resv;
792 };
793 
794 struct io_uring_buf_ring {
795 	union {
796 		/*
797 		 * To avoid spilling into more pages than we need to, the
798 		 * ring tail is overlaid with the io_uring_buf->resv field.
799 		 */
800 		struct {
801 			__u64	resv1;
802 			__u32	resv2;
803 			__u16	resv3;
804 			__u16	tail;
805 		};
806 		__DECLARE_FLEX_ARRAY(struct io_uring_buf, bufs);
807 	};
808 };
809 
810 /*
811  * Flags for IORING_REGISTER_PBUF_RING.
812  *
813  * IOU_PBUF_RING_MMAP:	If set, kernel will allocate the memory for the ring.
814  *			The application must not set a ring_addr in struct
815  *			io_uring_buf_reg, instead it must subsequently call
816  *			mmap(2) with the offset set as:
817  *			IORING_OFF_PBUF_RING | (bgid << IORING_OFF_PBUF_SHIFT)
818  *			to get a virtual mapping for the ring.
819  * IOU_PBUF_RING_INC:	If set, buffers consumed from this buffer ring can be
820  *			consumed incrementally. Normally one (or more) buffers
821  *			are fully consumed. With incremental consumptions, it's
822  *			feasible to register big ranges of buffers, and each
823  *			use of it will consume only as much as it needs. This
824  *			requires that both the kernel and application keep
825  *			track of where the current read/recv index is at.
826  */
827 enum io_uring_register_pbuf_ring_flags {
828 	IOU_PBUF_RING_MMAP	= 1,
829 	IOU_PBUF_RING_INC	= 2,
830 };
831 
832 /* argument for IORING_(UN)REGISTER_PBUF_RING */
833 struct io_uring_buf_reg {
834 	__u64	ring_addr;
835 	__u32	ring_entries;
836 	__u16	bgid;
837 	__u16	flags;
838 	__u64	resv[3];
839 };
840 
841 /* argument for IORING_REGISTER_PBUF_STATUS */
842 struct io_uring_buf_status {
843 	__u32	buf_group;	/* input */
844 	__u32	head;		/* output */
845 	__u32	resv[8];
846 };
847 
848 enum io_uring_napi_op {
849 	/* register/ungister backward compatible opcode */
850 	IO_URING_NAPI_REGISTER_OP = 0,
851 
852 	/* opcodes to update napi_list when static tracking is used */
853 	IO_URING_NAPI_STATIC_ADD_ID = 1,
854 	IO_URING_NAPI_STATIC_DEL_ID = 2
855 };
856 
857 enum io_uring_napi_tracking_strategy {
858 	/* value must be 0 for backward compatibility */
859 	IO_URING_NAPI_TRACKING_DYNAMIC = 0,
860 	IO_URING_NAPI_TRACKING_STATIC = 1,
861 	IO_URING_NAPI_TRACKING_INACTIVE = 255
862 };
863 
864 /* argument for IORING_(UN)REGISTER_NAPI */
865 struct io_uring_napi {
866 	__u32	busy_poll_to;
867 	__u8	prefer_busy_poll;
868 
869 	/* a io_uring_napi_op value */
870 	__u8	opcode;
871 	__u8	pad[2];
872 
873 	/*
874 	 * for IO_URING_NAPI_REGISTER_OP, it is a
875 	 * io_uring_napi_tracking_strategy value.
876 	 *
877 	 * for IO_URING_NAPI_STATIC_ADD_ID/IO_URING_NAPI_STATIC_DEL_ID
878 	 * it is the napi id to add/del from napi_list.
879 	 */
880 	__u32	op_param;
881 	__u32	resv;
882 };
883 
884 /*
885  * io_uring_restriction->opcode values
886  */
887 enum io_uring_register_restriction_op {
888 	/* Allow an io_uring_register(2) opcode */
889 	IORING_RESTRICTION_REGISTER_OP		= 0,
890 
891 	/* Allow an sqe opcode */
892 	IORING_RESTRICTION_SQE_OP		= 1,
893 
894 	/* Allow sqe flags */
895 	IORING_RESTRICTION_SQE_FLAGS_ALLOWED	= 2,
896 
897 	/* Require sqe flags (these flags must be set on each submission) */
898 	IORING_RESTRICTION_SQE_FLAGS_REQUIRED	= 3,
899 
900 	IORING_RESTRICTION_LAST
901 };
902 
903 enum {
904 	IORING_REG_WAIT_TS		= (1U << 0),
905 };
906 
907 /*
908  * Argument for io_uring_enter(2) with
909  * IORING_GETEVENTS | IORING_ENTER_EXT_ARG_REG set, where the actual argument
910  * is an index into a previously registered fixed wait region described by
911  * the below structure.
912  */
913 struct io_uring_reg_wait {
914 	struct __kernel_timespec	ts;
915 	__u32				min_wait_usec;
916 	__u32				flags;
917 	__u64				sigmask;
918 	__u32				sigmask_sz;
919 	__u32				pad[3];
920 	__u64				pad2[2];
921 };
922 
923 /*
924  * Argument for io_uring_enter(2) with IORING_GETEVENTS | IORING_ENTER_EXT_ARG
925  */
926 struct io_uring_getevents_arg {
927 	__u64	sigmask;
928 	__u32	sigmask_sz;
929 	__u32	min_wait_usec;
930 	__u64	ts;
931 };
932 
933 /*
934  * Argument for IORING_REGISTER_SYNC_CANCEL
935  */
936 struct io_uring_sync_cancel_reg {
937 	__u64				addr;
938 	__s32				fd;
939 	__u32				flags;
940 	struct __kernel_timespec	timeout;
941 	__u8				opcode;
942 	__u8				pad[7];
943 	__u64				pad2[3];
944 };
945 
946 /*
947  * Argument for IORING_REGISTER_FILE_ALLOC_RANGE
948  * The range is specified as [off, off + len)
949  */
950 struct io_uring_file_index_range {
951 	__u32	off;
952 	__u32	len;
953 	__u64	resv;
954 };
955 
956 struct io_uring_recvmsg_out {
957 	__u32 namelen;
958 	__u32 controllen;
959 	__u32 payloadlen;
960 	__u32 flags;
961 };
962 
963 /*
964  * Argument for IORING_OP_URING_CMD when file is a socket
965  */
966 enum io_uring_socket_op {
967 	SOCKET_URING_OP_SIOCINQ		= 0,
968 	SOCKET_URING_OP_SIOCOUTQ,
969 	SOCKET_URING_OP_GETSOCKOPT,
970 	SOCKET_URING_OP_SETSOCKOPT,
971 };
972 
973 /* Zero copy receive refill queue entry */
974 struct io_uring_zcrx_rqe {
975 	__u64	off;
976 	__u32	len;
977 	__u32	__pad;
978 };
979 
980 struct io_uring_zcrx_cqe {
981 	__u64	off;
982 	__u64	__pad;
983 };
984 
985 /* The bit from which area id is encoded into offsets */
986 #define IORING_ZCRX_AREA_SHIFT	48
987 #define IORING_ZCRX_AREA_MASK	(~(((__u64)1 << IORING_ZCRX_AREA_SHIFT) - 1))
988 
989 struct io_uring_zcrx_offsets {
990 	__u32	head;
991 	__u32	tail;
992 	__u32	rqes;
993 	__u32	__resv2;
994 	__u64	__resv[2];
995 };
996 
997 enum io_uring_zcrx_area_flags {
998 	IORING_ZCRX_AREA_DMABUF		= 1,
999 };
1000 
1001 struct io_uring_zcrx_area_reg {
1002 	__u64	addr;
1003 	__u64	len;
1004 	__u64	rq_area_token;
1005 	__u32	flags;
1006 	__u32	dmabuf_fd;
1007 	__u64	__resv2[2];
1008 };
1009 
1010 /*
1011  * Argument for IORING_REGISTER_ZCRX_IFQ
1012  */
1013 struct io_uring_zcrx_ifq_reg {
1014 	__u32	if_idx;
1015 	__u32	if_rxq;
1016 	__u32	rq_entries;
1017 	__u32	flags;
1018 
1019 	__u64	area_ptr; /* pointer to struct io_uring_zcrx_area_reg */
1020 	__u64	region_ptr; /* struct io_uring_region_desc * */
1021 
1022 	struct io_uring_zcrx_offsets offsets;
1023 	__u32	zcrx_id;
1024 	__u32	__resv2;
1025 	__u64	__resv[3];
1026 };
1027 
1028 #ifdef __cplusplus
1029 }
1030 #endif
1031 
1032 #endif
1033