xref: /linux/include/uapi/linux/io_uring.h (revision 3f1c07fc21c68bd3bd2df9d2c9441f6485e934d9)
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 		__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 */
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 /*
229  * Allow both 16b and 32b CQEs. If a 32b CQE is posted, it will have
230  * IORING_CQE_F_32 set in cqe->flags.
231  */
232 #define IORING_SETUP_CQE_MIXED		(1U << 18)
233 
234 /*
235  * Allow both 64b and 128b SQEs. If a 128b SQE is posted, it will have
236  * a 128b opcode.
237  */
238 #define IORING_SETUP_SQE_MIXED		(1U << 19)
239 
240 enum io_uring_op {
241 	IORING_OP_NOP,
242 	IORING_OP_READV,
243 	IORING_OP_WRITEV,
244 	IORING_OP_FSYNC,
245 	IORING_OP_READ_FIXED,
246 	IORING_OP_WRITE_FIXED,
247 	IORING_OP_POLL_ADD,
248 	IORING_OP_POLL_REMOVE,
249 	IORING_OP_SYNC_FILE_RANGE,
250 	IORING_OP_SENDMSG,
251 	IORING_OP_RECVMSG,
252 	IORING_OP_TIMEOUT,
253 	IORING_OP_TIMEOUT_REMOVE,
254 	IORING_OP_ACCEPT,
255 	IORING_OP_ASYNC_CANCEL,
256 	IORING_OP_LINK_TIMEOUT,
257 	IORING_OP_CONNECT,
258 	IORING_OP_FALLOCATE,
259 	IORING_OP_OPENAT,
260 	IORING_OP_CLOSE,
261 	IORING_OP_FILES_UPDATE,
262 	IORING_OP_STATX,
263 	IORING_OP_READ,
264 	IORING_OP_WRITE,
265 	IORING_OP_FADVISE,
266 	IORING_OP_MADVISE,
267 	IORING_OP_SEND,
268 	IORING_OP_RECV,
269 	IORING_OP_OPENAT2,
270 	IORING_OP_EPOLL_CTL,
271 	IORING_OP_SPLICE,
272 	IORING_OP_PROVIDE_BUFFERS,
273 	IORING_OP_REMOVE_BUFFERS,
274 	IORING_OP_TEE,
275 	IORING_OP_SHUTDOWN,
276 	IORING_OP_RENAMEAT,
277 	IORING_OP_UNLINKAT,
278 	IORING_OP_MKDIRAT,
279 	IORING_OP_SYMLINKAT,
280 	IORING_OP_LINKAT,
281 	IORING_OP_MSG_RING,
282 	IORING_OP_FSETXATTR,
283 	IORING_OP_SETXATTR,
284 	IORING_OP_FGETXATTR,
285 	IORING_OP_GETXATTR,
286 	IORING_OP_SOCKET,
287 	IORING_OP_URING_CMD,
288 	IORING_OP_SEND_ZC,
289 	IORING_OP_SENDMSG_ZC,
290 	IORING_OP_READ_MULTISHOT,
291 	IORING_OP_WAITID,
292 	IORING_OP_FUTEX_WAIT,
293 	IORING_OP_FUTEX_WAKE,
294 	IORING_OP_FUTEX_WAITV,
295 	IORING_OP_FIXED_FD_INSTALL,
296 	IORING_OP_FTRUNCATE,
297 	IORING_OP_BIND,
298 	IORING_OP_LISTEN,
299 	IORING_OP_RECV_ZC,
300 	IORING_OP_EPOLL_WAIT,
301 	IORING_OP_READV_FIXED,
302 	IORING_OP_WRITEV_FIXED,
303 	IORING_OP_PIPE,
304 	IORING_OP_NOP128,
305 	IORING_OP_URING_CMD128,
306 
307 	/* this goes last, obviously */
308 	IORING_OP_LAST,
309 };
310 
311 /*
312  * sqe->uring_cmd_flags		top 8bits aren't available for userspace
313  * IORING_URING_CMD_FIXED	use registered buffer; pass this flag
314  *				along with setting sqe->buf_index.
315  * IORING_URING_CMD_MULTISHOT	must be used with buffer select, like other
316  *				multishot commands. Not compatible with
317  *				IORING_URING_CMD_FIXED, for now.
318  */
319 #define IORING_URING_CMD_FIXED	(1U << 0)
320 #define IORING_URING_CMD_MULTISHOT	(1U << 1)
321 #define IORING_URING_CMD_MASK	(IORING_URING_CMD_FIXED | IORING_URING_CMD_MULTISHOT)
322 
323 
324 /*
325  * sqe->fsync_flags
326  */
327 #define IORING_FSYNC_DATASYNC	(1U << 0)
328 
329 /*
330  * sqe->timeout_flags
331  */
332 #define IORING_TIMEOUT_ABS		(1U << 0)
333 #define IORING_TIMEOUT_UPDATE		(1U << 1)
334 #define IORING_TIMEOUT_BOOTTIME		(1U << 2)
335 #define IORING_TIMEOUT_REALTIME		(1U << 3)
336 #define IORING_LINK_TIMEOUT_UPDATE	(1U << 4)
337 #define IORING_TIMEOUT_ETIME_SUCCESS	(1U << 5)
338 #define IORING_TIMEOUT_MULTISHOT	(1U << 6)
339 #define IORING_TIMEOUT_CLOCK_MASK	(IORING_TIMEOUT_BOOTTIME | IORING_TIMEOUT_REALTIME)
340 #define IORING_TIMEOUT_UPDATE_MASK	(IORING_TIMEOUT_UPDATE | IORING_LINK_TIMEOUT_UPDATE)
341 /*
342  * sqe->splice_flags
343  * extends splice(2) flags
344  */
345 #define SPLICE_F_FD_IN_FIXED	(1U << 31) /* the last bit of __u32 */
346 
347 /*
348  * POLL_ADD flags. Note that since sqe->poll_events is the flag space, the
349  * command flags for POLL_ADD are stored in sqe->len.
350  *
351  * IORING_POLL_ADD_MULTI	Multishot poll. Sets IORING_CQE_F_MORE if
352  *				the poll handler will continue to report
353  *				CQEs on behalf of the same SQE.
354  *
355  * IORING_POLL_UPDATE		Update existing poll request, matching
356  *				sqe->addr as the old user_data field.
357  *
358  * IORING_POLL_LEVEL		Level triggered poll.
359  */
360 #define IORING_POLL_ADD_MULTI	(1U << 0)
361 #define IORING_POLL_UPDATE_EVENTS	(1U << 1)
362 #define IORING_POLL_UPDATE_USER_DATA	(1U << 2)
363 #define IORING_POLL_ADD_LEVEL		(1U << 3)
364 
365 /*
366  * ASYNC_CANCEL flags.
367  *
368  * IORING_ASYNC_CANCEL_ALL	Cancel all requests that match the given key
369  * IORING_ASYNC_CANCEL_FD	Key off 'fd' for cancelation rather than the
370  *				request 'user_data'
371  * IORING_ASYNC_CANCEL_ANY	Match any request
372  * IORING_ASYNC_CANCEL_FD_FIXED	'fd' passed in is a fixed descriptor
373  * IORING_ASYNC_CANCEL_USERDATA	Match on user_data, default for no other key
374  * IORING_ASYNC_CANCEL_OP	Match request based on opcode
375  */
376 #define IORING_ASYNC_CANCEL_ALL	(1U << 0)
377 #define IORING_ASYNC_CANCEL_FD	(1U << 1)
378 #define IORING_ASYNC_CANCEL_ANY	(1U << 2)
379 #define IORING_ASYNC_CANCEL_FD_FIXED	(1U << 3)
380 #define IORING_ASYNC_CANCEL_USERDATA	(1U << 4)
381 #define IORING_ASYNC_CANCEL_OP	(1U << 5)
382 
383 /*
384  * send/sendmsg and recv/recvmsg flags (sqe->ioprio)
385  *
386  * IORING_RECVSEND_POLL_FIRST	If set, instead of first attempting to send
387  *				or receive and arm poll if that yields an
388  *				-EAGAIN result, arm poll upfront and skip
389  *				the initial transfer attempt.
390  *
391  * IORING_RECV_MULTISHOT	Multishot recv. Sets IORING_CQE_F_MORE if
392  *				the handler will continue to report
393  *				CQEs on behalf of the same SQE.
394  *
395  * IORING_RECVSEND_FIXED_BUF	Use registered buffers, the index is stored in
396  *				the buf_index field.
397  *
398  * IORING_SEND_ZC_REPORT_USAGE
399  *				If set, SEND[MSG]_ZC should report
400  *				the zerocopy usage in cqe.res
401  *				for the IORING_CQE_F_NOTIF cqe.
402  *				0 is reported if zerocopy was actually possible.
403  *				IORING_NOTIF_USAGE_ZC_COPIED if data was copied
404  *				(at least partially).
405  *
406  * IORING_RECVSEND_BUNDLE	Used with IOSQE_BUFFER_SELECT. If set, send or
407  *				recv will grab as many buffers from the buffer
408  *				group ID given and send them all. The completion
409  *				result 	will be the number of buffers send, with
410  *				the starting buffer ID in cqe->flags as per
411  *				usual for provided buffer usage. The buffers
412  *				will be	contiguous from the starting buffer ID.
413  *
414  * IORING_SEND_VECTORIZED	If set, SEND[_ZC] will take a pointer to a io_vec
415  *				to allow vectorized send operations.
416  */
417 #define IORING_RECVSEND_POLL_FIRST	(1U << 0)
418 #define IORING_RECV_MULTISHOT		(1U << 1)
419 #define IORING_RECVSEND_FIXED_BUF	(1U << 2)
420 #define IORING_SEND_ZC_REPORT_USAGE	(1U << 3)
421 #define IORING_RECVSEND_BUNDLE		(1U << 4)
422 #define IORING_SEND_VECTORIZED		(1U << 5)
423 
424 /*
425  * cqe.res for IORING_CQE_F_NOTIF if
426  * IORING_SEND_ZC_REPORT_USAGE was requested
427  *
428  * It should be treated as a flag, all other
429  * bits of cqe.res should be treated as reserved!
430  */
431 #define IORING_NOTIF_USAGE_ZC_COPIED    (1U << 31)
432 
433 /*
434  * accept flags stored in sqe->ioprio
435  */
436 #define IORING_ACCEPT_MULTISHOT	(1U << 0)
437 #define IORING_ACCEPT_DONTWAIT	(1U << 1)
438 #define IORING_ACCEPT_POLL_FIRST	(1U << 2)
439 
440 /*
441  * IORING_OP_MSG_RING command types, stored in sqe->addr
442  */
443 enum io_uring_msg_ring_flags {
444 	IORING_MSG_DATA,	/* pass sqe->len as 'res' and off as user_data */
445 	IORING_MSG_SEND_FD,	/* send a registered fd to another ring */
446 };
447 
448 /*
449  * IORING_OP_MSG_RING flags (sqe->msg_ring_flags)
450  *
451  * IORING_MSG_RING_CQE_SKIP	Don't post a CQE to the target ring. Not
452  *				applicable for IORING_MSG_DATA, obviously.
453  */
454 #define IORING_MSG_RING_CQE_SKIP	(1U << 0)
455 /* Pass through the flags from sqe->file_index to cqe->flags */
456 #define IORING_MSG_RING_FLAGS_PASS	(1U << 1)
457 
458 /*
459  * IORING_OP_FIXED_FD_INSTALL flags (sqe->install_fd_flags)
460  *
461  * IORING_FIXED_FD_NO_CLOEXEC	Don't mark the fd as O_CLOEXEC
462  */
463 #define IORING_FIXED_FD_NO_CLOEXEC	(1U << 0)
464 
465 /*
466  * IORING_OP_NOP flags (sqe->nop_flags)
467  *
468  * IORING_NOP_INJECT_RESULT	Inject result from sqe->result
469  */
470 #define IORING_NOP_INJECT_RESULT	(1U << 0)
471 #define IORING_NOP_FILE			(1U << 1)
472 #define IORING_NOP_FIXED_FILE		(1U << 2)
473 #define IORING_NOP_FIXED_BUFFER		(1U << 3)
474 #define IORING_NOP_TW			(1U << 4)
475 #define IORING_NOP_CQE32		(1U << 5)
476 
477 /*
478  * IO completion data structure (Completion Queue Entry)
479  */
480 struct io_uring_cqe {
481 	__u64	user_data;	/* sqe->user_data value passed back */
482 	__s32	res;		/* result code for this event */
483 	__u32	flags;
484 
485 	/*
486 	 * If the ring is initialized with IORING_SETUP_CQE32, then this field
487 	 * contains 16-bytes of padding, doubling the size of the CQE.
488 	 */
489 	__u64 big_cqe[];
490 };
491 
492 /*
493  * cqe->flags
494  *
495  * IORING_CQE_F_BUFFER	If set, the upper 16 bits are the buffer ID
496  * IORING_CQE_F_MORE	If set, parent SQE will generate more CQE entries
497  * IORING_CQE_F_SOCK_NONEMPTY	If set, more data to read after socket recv
498  * IORING_CQE_F_NOTIF	Set for notification CQEs. Can be used to distinct
499  * 			them from sends.
500  * IORING_CQE_F_BUF_MORE If set, the buffer ID set in the completion will get
501  *			more completions. In other words, the buffer is being
502  *			partially consumed, and will be used by the kernel for
503  *			more completions. This is only set for buffers used via
504  *			the incremental buffer consumption, as provided by
505  *			a ring buffer setup with IOU_PBUF_RING_INC. For any
506  *			other provided buffer type, all completions with a
507  *			buffer passed back is automatically returned to the
508  *			application.
509  * IORING_CQE_F_SKIP	If set, then the application/liburing must ignore this
510  *			CQE. It's only purpose is to fill a gap in the ring,
511  *			if a large CQE is attempted posted when the ring has
512  *			just a single small CQE worth of space left before
513  *			wrapping.
514  * IORING_CQE_F_32	If set, this is a 32b/big-cqe posting. Use with rings
515  *			setup in a mixed CQE mode, where both 16b and 32b
516  *			CQEs may be posted to the CQ ring.
517  */
518 #define IORING_CQE_F_BUFFER		(1U << 0)
519 #define IORING_CQE_F_MORE		(1U << 1)
520 #define IORING_CQE_F_SOCK_NONEMPTY	(1U << 2)
521 #define IORING_CQE_F_NOTIF		(1U << 3)
522 #define IORING_CQE_F_BUF_MORE		(1U << 4)
523 #define IORING_CQE_F_SKIP		(1U << 5)
524 #define IORING_CQE_F_32			(1U << 15)
525 
526 #define IORING_CQE_BUFFER_SHIFT		16
527 
528 /*
529  * Magic offsets for the application to mmap the data it needs
530  */
531 #define IORING_OFF_SQ_RING		0ULL
532 #define IORING_OFF_CQ_RING		0x8000000ULL
533 #define IORING_OFF_SQES			0x10000000ULL
534 #define IORING_OFF_PBUF_RING		0x80000000ULL
535 #define IORING_OFF_PBUF_SHIFT		16
536 #define IORING_OFF_MMAP_MASK		0xf8000000ULL
537 
538 /*
539  * Filled with the offset for mmap(2)
540  */
541 struct io_sqring_offsets {
542 	__u32 head;
543 	__u32 tail;
544 	__u32 ring_mask;
545 	__u32 ring_entries;
546 	__u32 flags;
547 	__u32 dropped;
548 	__u32 array;
549 	__u32 resv1;
550 	__u64 user_addr;
551 };
552 
553 /*
554  * sq_ring->flags
555  */
556 #define IORING_SQ_NEED_WAKEUP	(1U << 0) /* needs io_uring_enter wakeup */
557 #define IORING_SQ_CQ_OVERFLOW	(1U << 1) /* CQ ring is overflown */
558 #define IORING_SQ_TASKRUN	(1U << 2) /* task should enter the kernel */
559 
560 struct io_cqring_offsets {
561 	__u32 head;
562 	__u32 tail;
563 	__u32 ring_mask;
564 	__u32 ring_entries;
565 	__u32 overflow;
566 	__u32 cqes;
567 	__u32 flags;
568 	__u32 resv1;
569 	__u64 user_addr;
570 };
571 
572 /*
573  * cq_ring->flags
574  */
575 
576 /* disable eventfd notifications */
577 #define IORING_CQ_EVENTFD_DISABLED	(1U << 0)
578 
579 /*
580  * io_uring_enter(2) flags
581  */
582 #define IORING_ENTER_GETEVENTS		(1U << 0)
583 #define IORING_ENTER_SQ_WAKEUP		(1U << 1)
584 #define IORING_ENTER_SQ_WAIT		(1U << 2)
585 #define IORING_ENTER_EXT_ARG		(1U << 3)
586 #define IORING_ENTER_REGISTERED_RING	(1U << 4)
587 #define IORING_ENTER_ABS_TIMER		(1U << 5)
588 #define IORING_ENTER_EXT_ARG_REG	(1U << 6)
589 #define IORING_ENTER_NO_IOWAIT		(1U << 7)
590 
591 /*
592  * Passed in for io_uring_setup(2). Copied back with updated info on success
593  */
594 struct io_uring_params {
595 	__u32 sq_entries;
596 	__u32 cq_entries;
597 	__u32 flags;
598 	__u32 sq_thread_cpu;
599 	__u32 sq_thread_idle;
600 	__u32 features;
601 	__u32 wq_fd;
602 	__u32 resv[3];
603 	struct io_sqring_offsets sq_off;
604 	struct io_cqring_offsets cq_off;
605 };
606 
607 /*
608  * io_uring_params->features flags
609  */
610 #define IORING_FEAT_SINGLE_MMAP		(1U << 0)
611 #define IORING_FEAT_NODROP		(1U << 1)
612 #define IORING_FEAT_SUBMIT_STABLE	(1U << 2)
613 #define IORING_FEAT_RW_CUR_POS		(1U << 3)
614 #define IORING_FEAT_CUR_PERSONALITY	(1U << 4)
615 #define IORING_FEAT_FAST_POLL		(1U << 5)
616 #define IORING_FEAT_POLL_32BITS 	(1U << 6)
617 #define IORING_FEAT_SQPOLL_NONFIXED	(1U << 7)
618 #define IORING_FEAT_EXT_ARG		(1U << 8)
619 #define IORING_FEAT_NATIVE_WORKERS	(1U << 9)
620 #define IORING_FEAT_RSRC_TAGS		(1U << 10)
621 #define IORING_FEAT_CQE_SKIP		(1U << 11)
622 #define IORING_FEAT_LINKED_FILE		(1U << 12)
623 #define IORING_FEAT_REG_REG_RING	(1U << 13)
624 #define IORING_FEAT_RECVSEND_BUNDLE	(1U << 14)
625 #define IORING_FEAT_MIN_TIMEOUT		(1U << 15)
626 #define IORING_FEAT_RW_ATTR		(1U << 16)
627 #define IORING_FEAT_NO_IOWAIT		(1U << 17)
628 
629 /*
630  * io_uring_register(2) opcodes and arguments
631  */
632 enum io_uring_register_op {
633 	IORING_REGISTER_BUFFERS			= 0,
634 	IORING_UNREGISTER_BUFFERS		= 1,
635 	IORING_REGISTER_FILES			= 2,
636 	IORING_UNREGISTER_FILES			= 3,
637 	IORING_REGISTER_EVENTFD			= 4,
638 	IORING_UNREGISTER_EVENTFD		= 5,
639 	IORING_REGISTER_FILES_UPDATE		= 6,
640 	IORING_REGISTER_EVENTFD_ASYNC		= 7,
641 	IORING_REGISTER_PROBE			= 8,
642 	IORING_REGISTER_PERSONALITY		= 9,
643 	IORING_UNREGISTER_PERSONALITY		= 10,
644 	IORING_REGISTER_RESTRICTIONS		= 11,
645 	IORING_REGISTER_ENABLE_RINGS		= 12,
646 
647 	/* extended with tagging */
648 	IORING_REGISTER_FILES2			= 13,
649 	IORING_REGISTER_FILES_UPDATE2		= 14,
650 	IORING_REGISTER_BUFFERS2		= 15,
651 	IORING_REGISTER_BUFFERS_UPDATE		= 16,
652 
653 	/* set/clear io-wq thread affinities */
654 	IORING_REGISTER_IOWQ_AFF		= 17,
655 	IORING_UNREGISTER_IOWQ_AFF		= 18,
656 
657 	/* set/get max number of io-wq workers */
658 	IORING_REGISTER_IOWQ_MAX_WORKERS	= 19,
659 
660 	/* register/unregister io_uring fd with the ring */
661 	IORING_REGISTER_RING_FDS		= 20,
662 	IORING_UNREGISTER_RING_FDS		= 21,
663 
664 	/* register ring based provide buffer group */
665 	IORING_REGISTER_PBUF_RING		= 22,
666 	IORING_UNREGISTER_PBUF_RING		= 23,
667 
668 	/* sync cancelation API */
669 	IORING_REGISTER_SYNC_CANCEL		= 24,
670 
671 	/* register a range of fixed file slots for automatic slot allocation */
672 	IORING_REGISTER_FILE_ALLOC_RANGE	= 25,
673 
674 	/* return status information for a buffer group */
675 	IORING_REGISTER_PBUF_STATUS		= 26,
676 
677 	/* set/clear busy poll settings */
678 	IORING_REGISTER_NAPI			= 27,
679 	IORING_UNREGISTER_NAPI			= 28,
680 
681 	IORING_REGISTER_CLOCK			= 29,
682 
683 	/* clone registered buffers from source ring to current ring */
684 	IORING_REGISTER_CLONE_BUFFERS		= 30,
685 
686 	/* send MSG_RING without having a ring */
687 	IORING_REGISTER_SEND_MSG_RING		= 31,
688 
689 	/* register a netdev hw rx queue for zerocopy */
690 	IORING_REGISTER_ZCRX_IFQ		= 32,
691 
692 	/* resize CQ ring */
693 	IORING_REGISTER_RESIZE_RINGS		= 33,
694 
695 	IORING_REGISTER_MEM_REGION		= 34,
696 
697 	/* query various aspects of io_uring, see linux/io_uring/query.h */
698 	IORING_REGISTER_QUERY			= 35,
699 
700 	/* auxiliary zcrx configuration, see enum zcrx_ctrl_op */
701 	IORING_REGISTER_ZCRX_CTRL		= 36,
702 
703 	/* this goes last */
704 	IORING_REGISTER_LAST,
705 
706 	/* flag added to the opcode to use a registered ring fd */
707 	IORING_REGISTER_USE_REGISTERED_RING	= 1U << 31
708 };
709 
710 /* io-wq worker categories */
711 enum io_wq_type {
712 	IO_WQ_BOUND,
713 	IO_WQ_UNBOUND,
714 };
715 
716 /* deprecated, see struct io_uring_rsrc_update */
717 struct io_uring_files_update {
718 	__u32 offset;
719 	__u32 resv;
720 	__aligned_u64 /* __s32 * */ fds;
721 };
722 
723 enum {
724 	/* initialise with user provided memory pointed by user_addr */
725 	IORING_MEM_REGION_TYPE_USER		= 1,
726 };
727 
728 struct io_uring_region_desc {
729 	__u64 user_addr;
730 	__u64 size;
731 	__u32 flags;
732 	__u32 id;
733 	__u64 mmap_offset;
734 	__u64 __resv[4];
735 };
736 
737 enum {
738 	/* expose the region as registered wait arguments */
739 	IORING_MEM_REGION_REG_WAIT_ARG		= 1,
740 };
741 
742 struct io_uring_mem_region_reg {
743 	__u64 region_uptr; /* struct io_uring_region_desc * */
744 	__u64 flags;
745 	__u64 __resv[2];
746 };
747 
748 /*
749  * Register a fully sparse file space, rather than pass in an array of all
750  * -1 file descriptors.
751  */
752 #define IORING_RSRC_REGISTER_SPARSE	(1U << 0)
753 
754 struct io_uring_rsrc_register {
755 	__u32 nr;
756 	__u32 flags;
757 	__u64 resv2;
758 	__aligned_u64 data;
759 	__aligned_u64 tags;
760 };
761 
762 struct io_uring_rsrc_update {
763 	__u32 offset;
764 	__u32 resv;
765 	__aligned_u64 data;
766 };
767 
768 struct io_uring_rsrc_update2 {
769 	__u32 offset;
770 	__u32 resv;
771 	__aligned_u64 data;
772 	__aligned_u64 tags;
773 	__u32 nr;
774 	__u32 resv2;
775 };
776 
777 /* Skip updating fd indexes set to this value in the fd table */
778 #define IORING_REGISTER_FILES_SKIP	(-2)
779 
780 #define IO_URING_OP_SUPPORTED	(1U << 0)
781 
782 struct io_uring_probe_op {
783 	__u8 op;
784 	__u8 resv;
785 	__u16 flags;	/* IO_URING_OP_* flags */
786 	__u32 resv2;
787 };
788 
789 struct io_uring_probe {
790 	__u8 last_op;	/* last opcode supported */
791 	__u8 ops_len;	/* length of ops[] array below */
792 	__u16 resv;
793 	__u32 resv2[3];
794 	struct io_uring_probe_op ops[];
795 };
796 
797 struct io_uring_restriction {
798 	__u16 opcode;
799 	union {
800 		__u8 register_op; /* IORING_RESTRICTION_REGISTER_OP */
801 		__u8 sqe_op;      /* IORING_RESTRICTION_SQE_OP */
802 		__u8 sqe_flags;   /* IORING_RESTRICTION_SQE_FLAGS_* */
803 	};
804 	__u8 resv;
805 	__u32 resv2[3];
806 };
807 
808 struct io_uring_clock_register {
809 	__u32	clockid;
810 	__u32	__resv[3];
811 };
812 
813 enum {
814 	IORING_REGISTER_SRC_REGISTERED	= (1U << 0),
815 	IORING_REGISTER_DST_REPLACE	= (1U << 1),
816 };
817 
818 struct io_uring_clone_buffers {
819 	__u32	src_fd;
820 	__u32	flags;
821 	__u32	src_off;
822 	__u32	dst_off;
823 	__u32	nr;
824 	__u32	pad[3];
825 };
826 
827 struct io_uring_buf {
828 	__u64	addr;
829 	__u32	len;
830 	__u16	bid;
831 	__u16	resv;
832 };
833 
834 struct io_uring_buf_ring {
835 	union {
836 		/*
837 		 * To avoid spilling into more pages than we need to, the
838 		 * ring tail is overlaid with the io_uring_buf->resv field.
839 		 */
840 		struct {
841 			__u64	resv1;
842 			__u32	resv2;
843 			__u16	resv3;
844 			__u16	tail;
845 		};
846 		__DECLARE_FLEX_ARRAY(struct io_uring_buf, bufs);
847 	};
848 };
849 
850 /*
851  * Flags for IORING_REGISTER_PBUF_RING.
852  *
853  * IOU_PBUF_RING_MMAP:	If set, kernel will allocate the memory for the ring.
854  *			The application must not set a ring_addr in struct
855  *			io_uring_buf_reg, instead it must subsequently call
856  *			mmap(2) with the offset set as:
857  *			IORING_OFF_PBUF_RING | (bgid << IORING_OFF_PBUF_SHIFT)
858  *			to get a virtual mapping for the ring.
859  * IOU_PBUF_RING_INC:	If set, buffers consumed from this buffer ring can be
860  *			consumed incrementally. Normally one (or more) buffers
861  *			are fully consumed. With incremental consumptions, it's
862  *			feasible to register big ranges of buffers, and each
863  *			use of it will consume only as much as it needs. This
864  *			requires that both the kernel and application keep
865  *			track of where the current read/recv index is at.
866  */
867 enum io_uring_register_pbuf_ring_flags {
868 	IOU_PBUF_RING_MMAP	= 1,
869 	IOU_PBUF_RING_INC	= 2,
870 };
871 
872 /* argument for IORING_(UN)REGISTER_PBUF_RING */
873 struct io_uring_buf_reg {
874 	__u64	ring_addr;
875 	__u32	ring_entries;
876 	__u16	bgid;
877 	__u16	flags;
878 	__u64	resv[3];
879 };
880 
881 /* argument for IORING_REGISTER_PBUF_STATUS */
882 struct io_uring_buf_status {
883 	__u32	buf_group;	/* input */
884 	__u32	head;		/* output */
885 	__u32	resv[8];
886 };
887 
888 enum io_uring_napi_op {
889 	/* register/ungister backward compatible opcode */
890 	IO_URING_NAPI_REGISTER_OP = 0,
891 
892 	/* opcodes to update napi_list when static tracking is used */
893 	IO_URING_NAPI_STATIC_ADD_ID = 1,
894 	IO_URING_NAPI_STATIC_DEL_ID = 2
895 };
896 
897 enum io_uring_napi_tracking_strategy {
898 	/* value must be 0 for backward compatibility */
899 	IO_URING_NAPI_TRACKING_DYNAMIC = 0,
900 	IO_URING_NAPI_TRACKING_STATIC = 1,
901 	IO_URING_NAPI_TRACKING_INACTIVE = 255
902 };
903 
904 /* argument for IORING_(UN)REGISTER_NAPI */
905 struct io_uring_napi {
906 	__u32	busy_poll_to;
907 	__u8	prefer_busy_poll;
908 
909 	/* a io_uring_napi_op value */
910 	__u8	opcode;
911 	__u8	pad[2];
912 
913 	/*
914 	 * for IO_URING_NAPI_REGISTER_OP, it is a
915 	 * io_uring_napi_tracking_strategy value.
916 	 *
917 	 * for IO_URING_NAPI_STATIC_ADD_ID/IO_URING_NAPI_STATIC_DEL_ID
918 	 * it is the napi id to add/del from napi_list.
919 	 */
920 	__u32	op_param;
921 	__u32	resv;
922 };
923 
924 /*
925  * io_uring_restriction->opcode values
926  */
927 enum io_uring_register_restriction_op {
928 	/* Allow an io_uring_register(2) opcode */
929 	IORING_RESTRICTION_REGISTER_OP		= 0,
930 
931 	/* Allow an sqe opcode */
932 	IORING_RESTRICTION_SQE_OP		= 1,
933 
934 	/* Allow sqe flags */
935 	IORING_RESTRICTION_SQE_FLAGS_ALLOWED	= 2,
936 
937 	/* Require sqe flags (these flags must be set on each submission) */
938 	IORING_RESTRICTION_SQE_FLAGS_REQUIRED	= 3,
939 
940 	IORING_RESTRICTION_LAST
941 };
942 
943 enum {
944 	IORING_REG_WAIT_TS		= (1U << 0),
945 };
946 
947 /*
948  * Argument for io_uring_enter(2) with
949  * IORING_GETEVENTS | IORING_ENTER_EXT_ARG_REG set, where the actual argument
950  * is an index into a previously registered fixed wait region described by
951  * the below structure.
952  */
953 struct io_uring_reg_wait {
954 	struct __kernel_timespec	ts;
955 	__u32				min_wait_usec;
956 	__u32				flags;
957 	__u64				sigmask;
958 	__u32				sigmask_sz;
959 	__u32				pad[3];
960 	__u64				pad2[2];
961 };
962 
963 /*
964  * Argument for io_uring_enter(2) with IORING_GETEVENTS | IORING_ENTER_EXT_ARG
965  */
966 struct io_uring_getevents_arg {
967 	__u64	sigmask;
968 	__u32	sigmask_sz;
969 	__u32	min_wait_usec;
970 	__u64	ts;
971 };
972 
973 /*
974  * Argument for IORING_REGISTER_SYNC_CANCEL
975  */
976 struct io_uring_sync_cancel_reg {
977 	__u64				addr;
978 	__s32				fd;
979 	__u32				flags;
980 	struct __kernel_timespec	timeout;
981 	__u8				opcode;
982 	__u8				pad[7];
983 	__u64				pad2[3];
984 };
985 
986 /*
987  * Argument for IORING_REGISTER_FILE_ALLOC_RANGE
988  * The range is specified as [off, off + len)
989  */
990 struct io_uring_file_index_range {
991 	__u32	off;
992 	__u32	len;
993 	__u64	resv;
994 };
995 
996 struct io_uring_recvmsg_out {
997 	__u32 namelen;
998 	__u32 controllen;
999 	__u32 payloadlen;
1000 	__u32 flags;
1001 };
1002 
1003 /*
1004  * Argument for IORING_OP_URING_CMD when file is a socket
1005  */
1006 enum io_uring_socket_op {
1007 	SOCKET_URING_OP_SIOCINQ		= 0,
1008 	SOCKET_URING_OP_SIOCOUTQ,
1009 	SOCKET_URING_OP_GETSOCKOPT,
1010 	SOCKET_URING_OP_SETSOCKOPT,
1011 	SOCKET_URING_OP_TX_TIMESTAMP,
1012 	SOCKET_URING_OP_GETSOCKNAME,
1013 };
1014 
1015 /*
1016  * SOCKET_URING_OP_TX_TIMESTAMP definitions
1017  */
1018 
1019 #define IORING_TIMESTAMP_HW_SHIFT	16
1020 /* The cqe->flags bit from which the timestamp type is stored */
1021 #define IORING_TIMESTAMP_TYPE_SHIFT	(IORING_TIMESTAMP_HW_SHIFT + 1)
1022 /* The cqe->flags flag signifying whether it's a hardware timestamp */
1023 #define IORING_CQE_F_TSTAMP_HW		((__u32)1 << IORING_TIMESTAMP_HW_SHIFT)
1024 
1025 struct io_timespec {
1026 	__u64		tv_sec;
1027 	__u64		tv_nsec;
1028 };
1029 
1030 /* Zero copy receive refill queue entry */
1031 struct io_uring_zcrx_rqe {
1032 	__u64	off;
1033 	__u32	len;
1034 	__u32	__pad;
1035 };
1036 
1037 struct io_uring_zcrx_cqe {
1038 	__u64	off;
1039 	__u64	__pad;
1040 };
1041 
1042 /* The bit from which area id is encoded into offsets */
1043 #define IORING_ZCRX_AREA_SHIFT	48
1044 #define IORING_ZCRX_AREA_MASK	(~(((__u64)1 << IORING_ZCRX_AREA_SHIFT) - 1))
1045 
1046 struct io_uring_zcrx_offsets {
1047 	__u32	head;
1048 	__u32	tail;
1049 	__u32	rqes;
1050 	__u32	__resv2;
1051 	__u64	__resv[2];
1052 };
1053 
1054 enum io_uring_zcrx_area_flags {
1055 	IORING_ZCRX_AREA_DMABUF		= 1,
1056 };
1057 
1058 struct io_uring_zcrx_area_reg {
1059 	__u64	addr;
1060 	__u64	len;
1061 	__u64	rq_area_token;
1062 	__u32	flags;
1063 	__u32	dmabuf_fd;
1064 	__u64	__resv2[2];
1065 };
1066 
1067 enum zcrx_reg_flags {
1068 	ZCRX_REG_IMPORT	= 1,
1069 };
1070 
1071 /*
1072  * Argument for IORING_REGISTER_ZCRX_IFQ
1073  */
1074 struct io_uring_zcrx_ifq_reg {
1075 	__u32	if_idx;
1076 	__u32	if_rxq;
1077 	__u32	rq_entries;
1078 	__u32	flags;
1079 
1080 	__u64	area_ptr; /* pointer to struct io_uring_zcrx_area_reg */
1081 	__u64	region_ptr; /* struct io_uring_region_desc * */
1082 
1083 	struct io_uring_zcrx_offsets offsets;
1084 	__u32	zcrx_id;
1085 	__u32	__resv2;
1086 	__u64	__resv[3];
1087 };
1088 
1089 enum zcrx_ctrl_op {
1090 	ZCRX_CTRL_FLUSH_RQ,
1091 	ZCRX_CTRL_EXPORT,
1092 
1093 	__ZCRX_CTRL_LAST,
1094 };
1095 
1096 struct zcrx_ctrl_flush_rq {
1097 	__u64		__resv[6];
1098 };
1099 
1100 struct zcrx_ctrl_export {
1101 	__u32		zcrx_fd;
1102 	__u32 		__resv1[11];
1103 };
1104 
1105 struct zcrx_ctrl {
1106 	__u32	zcrx_id;
1107 	__u32	op; /* see enum zcrx_ctrl_op */
1108 	__u64	__resv[2];
1109 
1110 	union {
1111 		struct zcrx_ctrl_export		zc_export;
1112 		struct zcrx_ctrl_flush_rq	zc_flush;
1113 	};
1114 };
1115 
1116 #ifdef __cplusplus
1117 }
1118 #endif
1119 
1120 #endif
1121