xref: /linux/include/uapi/linux/io_uring.h (revision eff5f16bfd87ae48c56751741af41a825d5d4618)
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 	};
77 	__u64	user_data;	/* data to be passed back at completion time */
78 	/* pack this to avoid bogus arm OABI complaints */
79 	union {
80 		/* index into fixed buffers, if used */
81 		__u16	buf_index;
82 		/* for grouped buffer selection */
83 		__u16	buf_group;
84 	} __attribute__((packed));
85 	/* personality to use, if used */
86 	__u16	personality;
87 	union {
88 		__s32	splice_fd_in;
89 		__u32	file_index;
90 		__u32	zcrx_ifq_idx;
91 		__u32	optlen;
92 		struct {
93 			__u16	addr_len;
94 			__u16	__pad3[1];
95 		};
96 	};
97 	union {
98 		struct {
99 			__u64	addr3;
100 			__u64	__pad2[1];
101 		};
102 		struct {
103 			__u64	attr_ptr; /* pointer to attribute information */
104 			__u64	attr_type_mask; /* bit mask of attributes */
105 		};
106 		__u64	optval;
107 		/*
108 		 * If the ring is initialized with IORING_SETUP_SQE128, then
109 		 * this field is used for 80 bytes of arbitrary command data
110 		 */
111 		__u8	cmd[0];
112 	};
113 };
114 
115 /* sqe->attr_type_mask flags */
116 #define IORING_RW_ATTR_FLAG_PI	(1U << 0)
117 /* PI attribute information */
118 struct io_uring_attr_pi {
119 		__u16	flags;
120 		__u16	app_tag;
121 		__u32	len;
122 		__u64	addr;
123 		__u64	seed;
124 		__u64	rsvd;
125 };
126 
127 /*
128  * If sqe->file_index is set to this for opcodes that instantiate a new
129  * direct descriptor (like openat/openat2/accept), then io_uring will allocate
130  * an available direct descriptor instead of having the application pass one
131  * in. The picked direct descriptor will be returned in cqe->res, or -ENFILE
132  * if the space is full.
133  */
134 #define IORING_FILE_INDEX_ALLOC		(~0U)
135 
136 enum io_uring_sqe_flags_bit {
137 	IOSQE_FIXED_FILE_BIT,
138 	IOSQE_IO_DRAIN_BIT,
139 	IOSQE_IO_LINK_BIT,
140 	IOSQE_IO_HARDLINK_BIT,
141 	IOSQE_ASYNC_BIT,
142 	IOSQE_BUFFER_SELECT_BIT,
143 	IOSQE_CQE_SKIP_SUCCESS_BIT,
144 };
145 
146 /*
147  * sqe->flags
148  */
149 /* use fixed fileset */
150 #define IOSQE_FIXED_FILE	(1U << IOSQE_FIXED_FILE_BIT)
151 /* issue after inflight IO */
152 #define IOSQE_IO_DRAIN		(1U << IOSQE_IO_DRAIN_BIT)
153 /* links next sqe */
154 #define IOSQE_IO_LINK		(1U << IOSQE_IO_LINK_BIT)
155 /* like LINK, but stronger */
156 #define IOSQE_IO_HARDLINK	(1U << IOSQE_IO_HARDLINK_BIT)
157 /* always go async */
158 #define IOSQE_ASYNC		(1U << IOSQE_ASYNC_BIT)
159 /* select buffer from sqe->buf_group */
160 #define IOSQE_BUFFER_SELECT	(1U << IOSQE_BUFFER_SELECT_BIT)
161 /* don't post CQE if request succeeded */
162 #define IOSQE_CQE_SKIP_SUCCESS	(1U << IOSQE_CQE_SKIP_SUCCESS_BIT)
163 
164 /*
165  * io_uring_setup() flags
166  */
167 #define IORING_SETUP_IOPOLL	(1U << 0)	/* io_context is polled */
168 #define IORING_SETUP_SQPOLL	(1U << 1)	/* SQ poll thread */
169 #define IORING_SETUP_SQ_AFF	(1U << 2)	/* sq_thread_cpu is valid */
170 #define IORING_SETUP_CQSIZE	(1U << 3)	/* app defines CQ size */
171 #define IORING_SETUP_CLAMP	(1U << 4)	/* clamp SQ/CQ ring sizes */
172 #define IORING_SETUP_ATTACH_WQ	(1U << 5)	/* attach to existing wq */
173 #define IORING_SETUP_R_DISABLED	(1U << 6)	/* start with ring disabled */
174 #define IORING_SETUP_SUBMIT_ALL	(1U << 7)	/* continue submit on error */
175 /*
176  * Cooperative task running. When requests complete, they often require
177  * forcing the submitter to transition to the kernel to complete. If this
178  * flag is set, work will be done when the task transitions anyway, rather
179  * than force an inter-processor interrupt reschedule. This avoids interrupting
180  * a task running in userspace, and saves an IPI.
181  */
182 #define IORING_SETUP_COOP_TASKRUN	(1U << 8)
183 /*
184  * If COOP_TASKRUN is set, get notified if task work is available for
185  * running and a kernel transition would be needed to run it. This sets
186  * IORING_SQ_TASKRUN in the sq ring flags. Not valid with COOP_TASKRUN.
187  */
188 #define IORING_SETUP_TASKRUN_FLAG	(1U << 9)
189 #define IORING_SETUP_SQE128		(1U << 10) /* SQEs are 128 byte */
190 #define IORING_SETUP_CQE32		(1U << 11) /* CQEs are 32 byte */
191 /*
192  * Only one task is allowed to submit requests
193  */
194 #define IORING_SETUP_SINGLE_ISSUER	(1U << 12)
195 
196 /*
197  * Defer running task work to get events.
198  * Rather than running bits of task work whenever the task transitions
199  * try to do it just before it is needed.
200  */
201 #define IORING_SETUP_DEFER_TASKRUN	(1U << 13)
202 
203 /*
204  * Application provides the memory for the rings
205  */
206 #define IORING_SETUP_NO_MMAP		(1U << 14)
207 
208 /*
209  * Register the ring fd in itself for use with
210  * IORING_REGISTER_USE_REGISTERED_RING; return a registered fd index rather
211  * than an fd.
212  */
213 #define IORING_SETUP_REGISTERED_FD_ONLY	(1U << 15)
214 
215 /*
216  * Removes indirection through the SQ index array.
217  */
218 #define IORING_SETUP_NO_SQARRAY		(1U << 16)
219 
220 /* Use hybrid poll in iopoll process */
221 #define IORING_SETUP_HYBRID_IOPOLL	(1U << 17)
222 
223 enum io_uring_op {
224 	IORING_OP_NOP,
225 	IORING_OP_READV,
226 	IORING_OP_WRITEV,
227 	IORING_OP_FSYNC,
228 	IORING_OP_READ_FIXED,
229 	IORING_OP_WRITE_FIXED,
230 	IORING_OP_POLL_ADD,
231 	IORING_OP_POLL_REMOVE,
232 	IORING_OP_SYNC_FILE_RANGE,
233 	IORING_OP_SENDMSG,
234 	IORING_OP_RECVMSG,
235 	IORING_OP_TIMEOUT,
236 	IORING_OP_TIMEOUT_REMOVE,
237 	IORING_OP_ACCEPT,
238 	IORING_OP_ASYNC_CANCEL,
239 	IORING_OP_LINK_TIMEOUT,
240 	IORING_OP_CONNECT,
241 	IORING_OP_FALLOCATE,
242 	IORING_OP_OPENAT,
243 	IORING_OP_CLOSE,
244 	IORING_OP_FILES_UPDATE,
245 	IORING_OP_STATX,
246 	IORING_OP_READ,
247 	IORING_OP_WRITE,
248 	IORING_OP_FADVISE,
249 	IORING_OP_MADVISE,
250 	IORING_OP_SEND,
251 	IORING_OP_RECV,
252 	IORING_OP_OPENAT2,
253 	IORING_OP_EPOLL_CTL,
254 	IORING_OP_SPLICE,
255 	IORING_OP_PROVIDE_BUFFERS,
256 	IORING_OP_REMOVE_BUFFERS,
257 	IORING_OP_TEE,
258 	IORING_OP_SHUTDOWN,
259 	IORING_OP_RENAMEAT,
260 	IORING_OP_UNLINKAT,
261 	IORING_OP_MKDIRAT,
262 	IORING_OP_SYMLINKAT,
263 	IORING_OP_LINKAT,
264 	IORING_OP_MSG_RING,
265 	IORING_OP_FSETXATTR,
266 	IORING_OP_SETXATTR,
267 	IORING_OP_FGETXATTR,
268 	IORING_OP_GETXATTR,
269 	IORING_OP_SOCKET,
270 	IORING_OP_URING_CMD,
271 	IORING_OP_SEND_ZC,
272 	IORING_OP_SENDMSG_ZC,
273 	IORING_OP_READ_MULTISHOT,
274 	IORING_OP_WAITID,
275 	IORING_OP_FUTEX_WAIT,
276 	IORING_OP_FUTEX_WAKE,
277 	IORING_OP_FUTEX_WAITV,
278 	IORING_OP_FIXED_FD_INSTALL,
279 	IORING_OP_FTRUNCATE,
280 	IORING_OP_BIND,
281 	IORING_OP_LISTEN,
282 	IORING_OP_RECV_ZC,
283 	IORING_OP_EPOLL_WAIT,
284 	IORING_OP_READV_FIXED,
285 	IORING_OP_WRITEV_FIXED,
286 
287 	/* this goes last, obviously */
288 	IORING_OP_LAST,
289 };
290 
291 /*
292  * sqe->uring_cmd_flags		top 8bits aren't available for userspace
293  * IORING_URING_CMD_FIXED	use registered buffer; pass this flag
294  *				along with setting sqe->buf_index.
295  */
296 #define IORING_URING_CMD_FIXED	(1U << 0)
297 #define IORING_URING_CMD_MASK	IORING_URING_CMD_FIXED
298 
299 
300 /*
301  * sqe->fsync_flags
302  */
303 #define IORING_FSYNC_DATASYNC	(1U << 0)
304 
305 /*
306  * sqe->timeout_flags
307  */
308 #define IORING_TIMEOUT_ABS		(1U << 0)
309 #define IORING_TIMEOUT_UPDATE		(1U << 1)
310 #define IORING_TIMEOUT_BOOTTIME		(1U << 2)
311 #define IORING_TIMEOUT_REALTIME		(1U << 3)
312 #define IORING_LINK_TIMEOUT_UPDATE	(1U << 4)
313 #define IORING_TIMEOUT_ETIME_SUCCESS	(1U << 5)
314 #define IORING_TIMEOUT_MULTISHOT	(1U << 6)
315 #define IORING_TIMEOUT_CLOCK_MASK	(IORING_TIMEOUT_BOOTTIME | IORING_TIMEOUT_REALTIME)
316 #define IORING_TIMEOUT_UPDATE_MASK	(IORING_TIMEOUT_UPDATE | IORING_LINK_TIMEOUT_UPDATE)
317 /*
318  * sqe->splice_flags
319  * extends splice(2) flags
320  */
321 #define SPLICE_F_FD_IN_FIXED	(1U << 31) /* the last bit of __u32 */
322 
323 /*
324  * POLL_ADD flags. Note that since sqe->poll_events is the flag space, the
325  * command flags for POLL_ADD are stored in sqe->len.
326  *
327  * IORING_POLL_ADD_MULTI	Multishot poll. Sets IORING_CQE_F_MORE if
328  *				the poll handler will continue to report
329  *				CQEs on behalf of the same SQE.
330  *
331  * IORING_POLL_UPDATE		Update existing poll request, matching
332  *				sqe->addr as the old user_data field.
333  *
334  * IORING_POLL_LEVEL		Level triggered poll.
335  */
336 #define IORING_POLL_ADD_MULTI	(1U << 0)
337 #define IORING_POLL_UPDATE_EVENTS	(1U << 1)
338 #define IORING_POLL_UPDATE_USER_DATA	(1U << 2)
339 #define IORING_POLL_ADD_LEVEL		(1U << 3)
340 
341 /*
342  * ASYNC_CANCEL flags.
343  *
344  * IORING_ASYNC_CANCEL_ALL	Cancel all requests that match the given key
345  * IORING_ASYNC_CANCEL_FD	Key off 'fd' for cancelation rather than the
346  *				request 'user_data'
347  * IORING_ASYNC_CANCEL_ANY	Match any request
348  * IORING_ASYNC_CANCEL_FD_FIXED	'fd' passed in is a fixed descriptor
349  * IORING_ASYNC_CANCEL_USERDATA	Match on user_data, default for no other key
350  * IORING_ASYNC_CANCEL_OP	Match request based on opcode
351  */
352 #define IORING_ASYNC_CANCEL_ALL	(1U << 0)
353 #define IORING_ASYNC_CANCEL_FD	(1U << 1)
354 #define IORING_ASYNC_CANCEL_ANY	(1U << 2)
355 #define IORING_ASYNC_CANCEL_FD_FIXED	(1U << 3)
356 #define IORING_ASYNC_CANCEL_USERDATA	(1U << 4)
357 #define IORING_ASYNC_CANCEL_OP	(1U << 5)
358 
359 /*
360  * send/sendmsg and recv/recvmsg flags (sqe->ioprio)
361  *
362  * IORING_RECVSEND_POLL_FIRST	If set, instead of first attempting to send
363  *				or receive and arm poll if that yields an
364  *				-EAGAIN result, arm poll upfront and skip
365  *				the initial transfer attempt.
366  *
367  * IORING_RECV_MULTISHOT	Multishot recv. Sets IORING_CQE_F_MORE if
368  *				the handler will continue to report
369  *				CQEs on behalf of the same SQE.
370  *
371  * IORING_RECVSEND_FIXED_BUF	Use registered buffers, the index is stored in
372  *				the buf_index field.
373  *
374  * IORING_SEND_ZC_REPORT_USAGE
375  *				If set, SEND[MSG]_ZC should report
376  *				the zerocopy usage in cqe.res
377  *				for the IORING_CQE_F_NOTIF cqe.
378  *				0 is reported if zerocopy was actually possible.
379  *				IORING_NOTIF_USAGE_ZC_COPIED if data was copied
380  *				(at least partially).
381  *
382  * IORING_RECVSEND_BUNDLE	Used with IOSQE_BUFFER_SELECT. If set, send or
383  *				recv will grab as many buffers from the buffer
384  *				group ID given and send them all. The completion
385  *				result 	will be the number of buffers send, with
386  *				the starting buffer ID in cqe->flags as per
387  *				usual for provided buffer usage. The buffers
388  *				will be	contiguous from the starting buffer ID.
389  */
390 #define IORING_RECVSEND_POLL_FIRST	(1U << 0)
391 #define IORING_RECV_MULTISHOT		(1U << 1)
392 #define IORING_RECVSEND_FIXED_BUF	(1U << 2)
393 #define IORING_SEND_ZC_REPORT_USAGE	(1U << 3)
394 #define IORING_RECVSEND_BUNDLE		(1U << 4)
395 
396 /*
397  * cqe.res for IORING_CQE_F_NOTIF if
398  * IORING_SEND_ZC_REPORT_USAGE was requested
399  *
400  * It should be treated as a flag, all other
401  * bits of cqe.res should be treated as reserved!
402  */
403 #define IORING_NOTIF_USAGE_ZC_COPIED    (1U << 31)
404 
405 /*
406  * accept flags stored in sqe->ioprio
407  */
408 #define IORING_ACCEPT_MULTISHOT	(1U << 0)
409 #define IORING_ACCEPT_DONTWAIT	(1U << 1)
410 #define IORING_ACCEPT_POLL_FIRST	(1U << 2)
411 
412 /*
413  * IORING_OP_MSG_RING command types, stored in sqe->addr
414  */
415 enum io_uring_msg_ring_flags {
416 	IORING_MSG_DATA,	/* pass sqe->len as 'res' and off as user_data */
417 	IORING_MSG_SEND_FD,	/* send a registered fd to another ring */
418 };
419 
420 /*
421  * IORING_OP_MSG_RING flags (sqe->msg_ring_flags)
422  *
423  * IORING_MSG_RING_CQE_SKIP	Don't post a CQE to the target ring. Not
424  *				applicable for IORING_MSG_DATA, obviously.
425  */
426 #define IORING_MSG_RING_CQE_SKIP	(1U << 0)
427 /* Pass through the flags from sqe->file_index to cqe->flags */
428 #define IORING_MSG_RING_FLAGS_PASS	(1U << 1)
429 
430 /*
431  * IORING_OP_FIXED_FD_INSTALL flags (sqe->install_fd_flags)
432  *
433  * IORING_FIXED_FD_NO_CLOEXEC	Don't mark the fd as O_CLOEXEC
434  */
435 #define IORING_FIXED_FD_NO_CLOEXEC	(1U << 0)
436 
437 /*
438  * IORING_OP_NOP flags (sqe->nop_flags)
439  *
440  * IORING_NOP_INJECT_RESULT	Inject result from sqe->result
441  */
442 #define IORING_NOP_INJECT_RESULT	(1U << 0)
443 #define IORING_NOP_FILE			(1U << 1)
444 #define IORING_NOP_FIXED_FILE		(1U << 2)
445 #define IORING_NOP_FIXED_BUFFER		(1U << 3)
446 
447 /*
448  * IO completion data structure (Completion Queue Entry)
449  */
450 struct io_uring_cqe {
451 	__u64	user_data;	/* sqe->user_data value passed back */
452 	__s32	res;		/* result code for this event */
453 	__u32	flags;
454 
455 	/*
456 	 * If the ring is initialized with IORING_SETUP_CQE32, then this field
457 	 * contains 16-bytes of padding, doubling the size of the CQE.
458 	 */
459 	__u64 big_cqe[];
460 };
461 
462 /*
463  * cqe->flags
464  *
465  * IORING_CQE_F_BUFFER	If set, the upper 16 bits are the buffer ID
466  * IORING_CQE_F_MORE	If set, parent SQE will generate more CQE entries
467  * IORING_CQE_F_SOCK_NONEMPTY	If set, more data to read after socket recv
468  * IORING_CQE_F_NOTIF	Set for notification CQEs. Can be used to distinct
469  * 			them from sends.
470  * IORING_CQE_F_BUF_MORE If set, the buffer ID set in the completion will get
471  *			more completions. In other words, the buffer is being
472  *			partially consumed, and will be used by the kernel for
473  *			more completions. This is only set for buffers used via
474  *			the incremental buffer consumption, as provided by
475  *			a ring buffer setup with IOU_PBUF_RING_INC. For any
476  *			other provided buffer type, all completions with a
477  *			buffer passed back is automatically returned to the
478  *			application.
479  */
480 #define IORING_CQE_F_BUFFER		(1U << 0)
481 #define IORING_CQE_F_MORE		(1U << 1)
482 #define IORING_CQE_F_SOCK_NONEMPTY	(1U << 2)
483 #define IORING_CQE_F_NOTIF		(1U << 3)
484 #define IORING_CQE_F_BUF_MORE		(1U << 4)
485 
486 #define IORING_CQE_BUFFER_SHIFT		16
487 
488 /*
489  * Magic offsets for the application to mmap the data it needs
490  */
491 #define IORING_OFF_SQ_RING		0ULL
492 #define IORING_OFF_CQ_RING		0x8000000ULL
493 #define IORING_OFF_SQES			0x10000000ULL
494 #define IORING_OFF_PBUF_RING		0x80000000ULL
495 #define IORING_OFF_PBUF_SHIFT		16
496 #define IORING_OFF_MMAP_MASK		0xf8000000ULL
497 
498 /*
499  * Filled with the offset for mmap(2)
500  */
501 struct io_sqring_offsets {
502 	__u32 head;
503 	__u32 tail;
504 	__u32 ring_mask;
505 	__u32 ring_entries;
506 	__u32 flags;
507 	__u32 dropped;
508 	__u32 array;
509 	__u32 resv1;
510 	__u64 user_addr;
511 };
512 
513 /*
514  * sq_ring->flags
515  */
516 #define IORING_SQ_NEED_WAKEUP	(1U << 0) /* needs io_uring_enter wakeup */
517 #define IORING_SQ_CQ_OVERFLOW	(1U << 1) /* CQ ring is overflown */
518 #define IORING_SQ_TASKRUN	(1U << 2) /* task should enter the kernel */
519 
520 struct io_cqring_offsets {
521 	__u32 head;
522 	__u32 tail;
523 	__u32 ring_mask;
524 	__u32 ring_entries;
525 	__u32 overflow;
526 	__u32 cqes;
527 	__u32 flags;
528 	__u32 resv1;
529 	__u64 user_addr;
530 };
531 
532 /*
533  * cq_ring->flags
534  */
535 
536 /* disable eventfd notifications */
537 #define IORING_CQ_EVENTFD_DISABLED	(1U << 0)
538 
539 /*
540  * io_uring_enter(2) flags
541  */
542 #define IORING_ENTER_GETEVENTS		(1U << 0)
543 #define IORING_ENTER_SQ_WAKEUP		(1U << 1)
544 #define IORING_ENTER_SQ_WAIT		(1U << 2)
545 #define IORING_ENTER_EXT_ARG		(1U << 3)
546 #define IORING_ENTER_REGISTERED_RING	(1U << 4)
547 #define IORING_ENTER_ABS_TIMER		(1U << 5)
548 #define IORING_ENTER_EXT_ARG_REG	(1U << 6)
549 #define IORING_ENTER_NO_IOWAIT		(1U << 7)
550 
551 /*
552  * Passed in for io_uring_setup(2). Copied back with updated info on success
553  */
554 struct io_uring_params {
555 	__u32 sq_entries;
556 	__u32 cq_entries;
557 	__u32 flags;
558 	__u32 sq_thread_cpu;
559 	__u32 sq_thread_idle;
560 	__u32 features;
561 	__u32 wq_fd;
562 	__u32 resv[3];
563 	struct io_sqring_offsets sq_off;
564 	struct io_cqring_offsets cq_off;
565 };
566 
567 /*
568  * io_uring_params->features flags
569  */
570 #define IORING_FEAT_SINGLE_MMAP		(1U << 0)
571 #define IORING_FEAT_NODROP		(1U << 1)
572 #define IORING_FEAT_SUBMIT_STABLE	(1U << 2)
573 #define IORING_FEAT_RW_CUR_POS		(1U << 3)
574 #define IORING_FEAT_CUR_PERSONALITY	(1U << 4)
575 #define IORING_FEAT_FAST_POLL		(1U << 5)
576 #define IORING_FEAT_POLL_32BITS 	(1U << 6)
577 #define IORING_FEAT_SQPOLL_NONFIXED	(1U << 7)
578 #define IORING_FEAT_EXT_ARG		(1U << 8)
579 #define IORING_FEAT_NATIVE_WORKERS	(1U << 9)
580 #define IORING_FEAT_RSRC_TAGS		(1U << 10)
581 #define IORING_FEAT_CQE_SKIP		(1U << 11)
582 #define IORING_FEAT_LINKED_FILE		(1U << 12)
583 #define IORING_FEAT_REG_REG_RING	(1U << 13)
584 #define IORING_FEAT_RECVSEND_BUNDLE	(1U << 14)
585 #define IORING_FEAT_MIN_TIMEOUT		(1U << 15)
586 #define IORING_FEAT_RW_ATTR		(1U << 16)
587 #define IORING_FEAT_NO_IOWAIT		(1U << 17)
588 
589 /*
590  * io_uring_register(2) opcodes and arguments
591  */
592 enum io_uring_register_op {
593 	IORING_REGISTER_BUFFERS			= 0,
594 	IORING_UNREGISTER_BUFFERS		= 1,
595 	IORING_REGISTER_FILES			= 2,
596 	IORING_UNREGISTER_FILES			= 3,
597 	IORING_REGISTER_EVENTFD			= 4,
598 	IORING_UNREGISTER_EVENTFD		= 5,
599 	IORING_REGISTER_FILES_UPDATE		= 6,
600 	IORING_REGISTER_EVENTFD_ASYNC		= 7,
601 	IORING_REGISTER_PROBE			= 8,
602 	IORING_REGISTER_PERSONALITY		= 9,
603 	IORING_UNREGISTER_PERSONALITY		= 10,
604 	IORING_REGISTER_RESTRICTIONS		= 11,
605 	IORING_REGISTER_ENABLE_RINGS		= 12,
606 
607 	/* extended with tagging */
608 	IORING_REGISTER_FILES2			= 13,
609 	IORING_REGISTER_FILES_UPDATE2		= 14,
610 	IORING_REGISTER_BUFFERS2		= 15,
611 	IORING_REGISTER_BUFFERS_UPDATE		= 16,
612 
613 	/* set/clear io-wq thread affinities */
614 	IORING_REGISTER_IOWQ_AFF		= 17,
615 	IORING_UNREGISTER_IOWQ_AFF		= 18,
616 
617 	/* set/get max number of io-wq workers */
618 	IORING_REGISTER_IOWQ_MAX_WORKERS	= 19,
619 
620 	/* register/unregister io_uring fd with the ring */
621 	IORING_REGISTER_RING_FDS		= 20,
622 	IORING_UNREGISTER_RING_FDS		= 21,
623 
624 	/* register ring based provide buffer group */
625 	IORING_REGISTER_PBUF_RING		= 22,
626 	IORING_UNREGISTER_PBUF_RING		= 23,
627 
628 	/* sync cancelation API */
629 	IORING_REGISTER_SYNC_CANCEL		= 24,
630 
631 	/* register a range of fixed file slots for automatic slot allocation */
632 	IORING_REGISTER_FILE_ALLOC_RANGE	= 25,
633 
634 	/* return status information for a buffer group */
635 	IORING_REGISTER_PBUF_STATUS		= 26,
636 
637 	/* set/clear busy poll settings */
638 	IORING_REGISTER_NAPI			= 27,
639 	IORING_UNREGISTER_NAPI			= 28,
640 
641 	IORING_REGISTER_CLOCK			= 29,
642 
643 	/* clone registered buffers from source ring to current ring */
644 	IORING_REGISTER_CLONE_BUFFERS		= 30,
645 
646 	/* send MSG_RING without having a ring */
647 	IORING_REGISTER_SEND_MSG_RING		= 31,
648 
649 	/* register a netdev hw rx queue for zerocopy */
650 	IORING_REGISTER_ZCRX_IFQ		= 32,
651 
652 	/* resize CQ ring */
653 	IORING_REGISTER_RESIZE_RINGS		= 33,
654 
655 	IORING_REGISTER_MEM_REGION		= 34,
656 
657 	/* this goes last */
658 	IORING_REGISTER_LAST,
659 
660 	/* flag added to the opcode to use a registered ring fd */
661 	IORING_REGISTER_USE_REGISTERED_RING	= 1U << 31
662 };
663 
664 /* io-wq worker categories */
665 enum io_wq_type {
666 	IO_WQ_BOUND,
667 	IO_WQ_UNBOUND,
668 };
669 
670 /* deprecated, see struct io_uring_rsrc_update */
671 struct io_uring_files_update {
672 	__u32 offset;
673 	__u32 resv;
674 	__aligned_u64 /* __s32 * */ fds;
675 };
676 
677 enum {
678 	/* initialise with user provided memory pointed by user_addr */
679 	IORING_MEM_REGION_TYPE_USER		= 1,
680 };
681 
682 struct io_uring_region_desc {
683 	__u64 user_addr;
684 	__u64 size;
685 	__u32 flags;
686 	__u32 id;
687 	__u64 mmap_offset;
688 	__u64 __resv[4];
689 };
690 
691 enum {
692 	/* expose the region as registered wait arguments */
693 	IORING_MEM_REGION_REG_WAIT_ARG		= 1,
694 };
695 
696 struct io_uring_mem_region_reg {
697 	__u64 region_uptr; /* struct io_uring_region_desc * */
698 	__u64 flags;
699 	__u64 __resv[2];
700 };
701 
702 /*
703  * Register a fully sparse file space, rather than pass in an array of all
704  * -1 file descriptors.
705  */
706 #define IORING_RSRC_REGISTER_SPARSE	(1U << 0)
707 
708 struct io_uring_rsrc_register {
709 	__u32 nr;
710 	__u32 flags;
711 	__u64 resv2;
712 	__aligned_u64 data;
713 	__aligned_u64 tags;
714 };
715 
716 struct io_uring_rsrc_update {
717 	__u32 offset;
718 	__u32 resv;
719 	__aligned_u64 data;
720 };
721 
722 struct io_uring_rsrc_update2 {
723 	__u32 offset;
724 	__u32 resv;
725 	__aligned_u64 data;
726 	__aligned_u64 tags;
727 	__u32 nr;
728 	__u32 resv2;
729 };
730 
731 /* Skip updating fd indexes set to this value in the fd table */
732 #define IORING_REGISTER_FILES_SKIP	(-2)
733 
734 #define IO_URING_OP_SUPPORTED	(1U << 0)
735 
736 struct io_uring_probe_op {
737 	__u8 op;
738 	__u8 resv;
739 	__u16 flags;	/* IO_URING_OP_* flags */
740 	__u32 resv2;
741 };
742 
743 struct io_uring_probe {
744 	__u8 last_op;	/* last opcode supported */
745 	__u8 ops_len;	/* length of ops[] array below */
746 	__u16 resv;
747 	__u32 resv2[3];
748 	struct io_uring_probe_op ops[];
749 };
750 
751 struct io_uring_restriction {
752 	__u16 opcode;
753 	union {
754 		__u8 register_op; /* IORING_RESTRICTION_REGISTER_OP */
755 		__u8 sqe_op;      /* IORING_RESTRICTION_SQE_OP */
756 		__u8 sqe_flags;   /* IORING_RESTRICTION_SQE_FLAGS_* */
757 	};
758 	__u8 resv;
759 	__u32 resv2[3];
760 };
761 
762 struct io_uring_clock_register {
763 	__u32	clockid;
764 	__u32	__resv[3];
765 };
766 
767 enum {
768 	IORING_REGISTER_SRC_REGISTERED	= (1U << 0),
769 	IORING_REGISTER_DST_REPLACE	= (1U << 1),
770 };
771 
772 struct io_uring_clone_buffers {
773 	__u32	src_fd;
774 	__u32	flags;
775 	__u32	src_off;
776 	__u32	dst_off;
777 	__u32	nr;
778 	__u32	pad[3];
779 };
780 
781 struct io_uring_buf {
782 	__u64	addr;
783 	__u32	len;
784 	__u16	bid;
785 	__u16	resv;
786 };
787 
788 struct io_uring_buf_ring {
789 	union {
790 		/*
791 		 * To avoid spilling into more pages than we need to, the
792 		 * ring tail is overlaid with the io_uring_buf->resv field.
793 		 */
794 		struct {
795 			__u64	resv1;
796 			__u32	resv2;
797 			__u16	resv3;
798 			__u16	tail;
799 		};
800 		__DECLARE_FLEX_ARRAY(struct io_uring_buf, bufs);
801 	};
802 };
803 
804 /*
805  * Flags for IORING_REGISTER_PBUF_RING.
806  *
807  * IOU_PBUF_RING_MMAP:	If set, kernel will allocate the memory for the ring.
808  *			The application must not set a ring_addr in struct
809  *			io_uring_buf_reg, instead it must subsequently call
810  *			mmap(2) with the offset set as:
811  *			IORING_OFF_PBUF_RING | (bgid << IORING_OFF_PBUF_SHIFT)
812  *			to get a virtual mapping for the ring.
813  * IOU_PBUF_RING_INC:	If set, buffers consumed from this buffer ring can be
814  *			consumed incrementally. Normally one (or more) buffers
815  *			are fully consumed. With incremental consumptions, it's
816  *			feasible to register big ranges of buffers, and each
817  *			use of it will consume only as much as it needs. This
818  *			requires that both the kernel and application keep
819  *			track of where the current read/recv index is at.
820  */
821 enum io_uring_register_pbuf_ring_flags {
822 	IOU_PBUF_RING_MMAP	= 1,
823 	IOU_PBUF_RING_INC	= 2,
824 };
825 
826 /* argument for IORING_(UN)REGISTER_PBUF_RING */
827 struct io_uring_buf_reg {
828 	__u64	ring_addr;
829 	__u32	ring_entries;
830 	__u16	bgid;
831 	__u16	flags;
832 	__u64	resv[3];
833 };
834 
835 /* argument for IORING_REGISTER_PBUF_STATUS */
836 struct io_uring_buf_status {
837 	__u32	buf_group;	/* input */
838 	__u32	head;		/* output */
839 	__u32	resv[8];
840 };
841 
842 enum io_uring_napi_op {
843 	/* register/ungister backward compatible opcode */
844 	IO_URING_NAPI_REGISTER_OP = 0,
845 
846 	/* opcodes to update napi_list when static tracking is used */
847 	IO_URING_NAPI_STATIC_ADD_ID = 1,
848 	IO_URING_NAPI_STATIC_DEL_ID = 2
849 };
850 
851 enum io_uring_napi_tracking_strategy {
852 	/* value must be 0 for backward compatibility */
853 	IO_URING_NAPI_TRACKING_DYNAMIC = 0,
854 	IO_URING_NAPI_TRACKING_STATIC = 1,
855 	IO_URING_NAPI_TRACKING_INACTIVE = 255
856 };
857 
858 /* argument for IORING_(UN)REGISTER_NAPI */
859 struct io_uring_napi {
860 	__u32	busy_poll_to;
861 	__u8	prefer_busy_poll;
862 
863 	/* a io_uring_napi_op value */
864 	__u8	opcode;
865 	__u8	pad[2];
866 
867 	/*
868 	 * for IO_URING_NAPI_REGISTER_OP, it is a
869 	 * io_uring_napi_tracking_strategy value.
870 	 *
871 	 * for IO_URING_NAPI_STATIC_ADD_ID/IO_URING_NAPI_STATIC_DEL_ID
872 	 * it is the napi id to add/del from napi_list.
873 	 */
874 	__u32	op_param;
875 	__u32	resv;
876 };
877 
878 /*
879  * io_uring_restriction->opcode values
880  */
881 enum io_uring_register_restriction_op {
882 	/* Allow an io_uring_register(2) opcode */
883 	IORING_RESTRICTION_REGISTER_OP		= 0,
884 
885 	/* Allow an sqe opcode */
886 	IORING_RESTRICTION_SQE_OP		= 1,
887 
888 	/* Allow sqe flags */
889 	IORING_RESTRICTION_SQE_FLAGS_ALLOWED	= 2,
890 
891 	/* Require sqe flags (these flags must be set on each submission) */
892 	IORING_RESTRICTION_SQE_FLAGS_REQUIRED	= 3,
893 
894 	IORING_RESTRICTION_LAST
895 };
896 
897 enum {
898 	IORING_REG_WAIT_TS		= (1U << 0),
899 };
900 
901 /*
902  * Argument for io_uring_enter(2) with
903  * IORING_GETEVENTS | IORING_ENTER_EXT_ARG_REG set, where the actual argument
904  * is an index into a previously registered fixed wait region described by
905  * the below structure.
906  */
907 struct io_uring_reg_wait {
908 	struct __kernel_timespec	ts;
909 	__u32				min_wait_usec;
910 	__u32				flags;
911 	__u64				sigmask;
912 	__u32				sigmask_sz;
913 	__u32				pad[3];
914 	__u64				pad2[2];
915 };
916 
917 /*
918  * Argument for io_uring_enter(2) with IORING_GETEVENTS | IORING_ENTER_EXT_ARG
919  */
920 struct io_uring_getevents_arg {
921 	__u64	sigmask;
922 	__u32	sigmask_sz;
923 	__u32	min_wait_usec;
924 	__u64	ts;
925 };
926 
927 /*
928  * Argument for IORING_REGISTER_SYNC_CANCEL
929  */
930 struct io_uring_sync_cancel_reg {
931 	__u64				addr;
932 	__s32				fd;
933 	__u32				flags;
934 	struct __kernel_timespec	timeout;
935 	__u8				opcode;
936 	__u8				pad[7];
937 	__u64				pad2[3];
938 };
939 
940 /*
941  * Argument for IORING_REGISTER_FILE_ALLOC_RANGE
942  * The range is specified as [off, off + len)
943  */
944 struct io_uring_file_index_range {
945 	__u32	off;
946 	__u32	len;
947 	__u64	resv;
948 };
949 
950 struct io_uring_recvmsg_out {
951 	__u32 namelen;
952 	__u32 controllen;
953 	__u32 payloadlen;
954 	__u32 flags;
955 };
956 
957 /*
958  * Argument for IORING_OP_URING_CMD when file is a socket
959  */
960 enum io_uring_socket_op {
961 	SOCKET_URING_OP_SIOCINQ		= 0,
962 	SOCKET_URING_OP_SIOCOUTQ,
963 	SOCKET_URING_OP_GETSOCKOPT,
964 	SOCKET_URING_OP_SETSOCKOPT,
965 };
966 
967 /* Zero copy receive refill queue entry */
968 struct io_uring_zcrx_rqe {
969 	__u64	off;
970 	__u32	len;
971 	__u32	__pad;
972 };
973 
974 struct io_uring_zcrx_cqe {
975 	__u64	off;
976 	__u64	__pad;
977 };
978 
979 /* The bit from which area id is encoded into offsets */
980 #define IORING_ZCRX_AREA_SHIFT	48
981 #define IORING_ZCRX_AREA_MASK	(~(((__u64)1 << IORING_ZCRX_AREA_SHIFT) - 1))
982 
983 struct io_uring_zcrx_offsets {
984 	__u32	head;
985 	__u32	tail;
986 	__u32	rqes;
987 	__u32	__resv2;
988 	__u64	__resv[2];
989 };
990 
991 struct io_uring_zcrx_area_reg {
992 	__u64	addr;
993 	__u64	len;
994 	__u64	rq_area_token;
995 	__u32	flags;
996 	__u32	__resv1;
997 	__u64	__resv2[2];
998 };
999 
1000 /*
1001  * Argument for IORING_REGISTER_ZCRX_IFQ
1002  */
1003 struct io_uring_zcrx_ifq_reg {
1004 	__u32	if_idx;
1005 	__u32	if_rxq;
1006 	__u32	rq_entries;
1007 	__u32	flags;
1008 
1009 	__u64	area_ptr; /* pointer to struct io_uring_zcrx_area_reg */
1010 	__u64	region_ptr; /* struct io_uring_region_desc * */
1011 
1012 	struct io_uring_zcrx_offsets offsets;
1013 	__u64	__resv[4];
1014 };
1015 
1016 #ifdef __cplusplus
1017 }
1018 #endif
1019 
1020 #endif
1021