xref: /linux-6.15/include/uapi/linux/io_uring.h (revision 19f7e942)
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 
285 	/* this goes last, obviously */
286 	IORING_OP_LAST,
287 };
288 
289 /*
290  * sqe->uring_cmd_flags		top 8bits aren't available for userspace
291  * IORING_URING_CMD_FIXED	use registered buffer; pass this flag
292  *				along with setting sqe->buf_index.
293  */
294 #define IORING_URING_CMD_FIXED	(1U << 0)
295 #define IORING_URING_CMD_MASK	IORING_URING_CMD_FIXED
296 
297 
298 /*
299  * sqe->fsync_flags
300  */
301 #define IORING_FSYNC_DATASYNC	(1U << 0)
302 
303 /*
304  * sqe->timeout_flags
305  */
306 #define IORING_TIMEOUT_ABS		(1U << 0)
307 #define IORING_TIMEOUT_UPDATE		(1U << 1)
308 #define IORING_TIMEOUT_BOOTTIME		(1U << 2)
309 #define IORING_TIMEOUT_REALTIME		(1U << 3)
310 #define IORING_LINK_TIMEOUT_UPDATE	(1U << 4)
311 #define IORING_TIMEOUT_ETIME_SUCCESS	(1U << 5)
312 #define IORING_TIMEOUT_MULTISHOT	(1U << 6)
313 #define IORING_TIMEOUT_CLOCK_MASK	(IORING_TIMEOUT_BOOTTIME | IORING_TIMEOUT_REALTIME)
314 #define IORING_TIMEOUT_UPDATE_MASK	(IORING_TIMEOUT_UPDATE | IORING_LINK_TIMEOUT_UPDATE)
315 /*
316  * sqe->splice_flags
317  * extends splice(2) flags
318  */
319 #define SPLICE_F_FD_IN_FIXED	(1U << 31) /* the last bit of __u32 */
320 
321 /*
322  * POLL_ADD flags. Note that since sqe->poll_events is the flag space, the
323  * command flags for POLL_ADD are stored in sqe->len.
324  *
325  * IORING_POLL_ADD_MULTI	Multishot poll. Sets IORING_CQE_F_MORE if
326  *				the poll handler will continue to report
327  *				CQEs on behalf of the same SQE.
328  *
329  * IORING_POLL_UPDATE		Update existing poll request, matching
330  *				sqe->addr as the old user_data field.
331  *
332  * IORING_POLL_LEVEL		Level triggered poll.
333  */
334 #define IORING_POLL_ADD_MULTI	(1U << 0)
335 #define IORING_POLL_UPDATE_EVENTS	(1U << 1)
336 #define IORING_POLL_UPDATE_USER_DATA	(1U << 2)
337 #define IORING_POLL_ADD_LEVEL		(1U << 3)
338 
339 /*
340  * ASYNC_CANCEL flags.
341  *
342  * IORING_ASYNC_CANCEL_ALL	Cancel all requests that match the given key
343  * IORING_ASYNC_CANCEL_FD	Key off 'fd' for cancelation rather than the
344  *				request 'user_data'
345  * IORING_ASYNC_CANCEL_ANY	Match any request
346  * IORING_ASYNC_CANCEL_FD_FIXED	'fd' passed in is a fixed descriptor
347  * IORING_ASYNC_CANCEL_USERDATA	Match on user_data, default for no other key
348  * IORING_ASYNC_CANCEL_OP	Match request based on opcode
349  */
350 #define IORING_ASYNC_CANCEL_ALL	(1U << 0)
351 #define IORING_ASYNC_CANCEL_FD	(1U << 1)
352 #define IORING_ASYNC_CANCEL_ANY	(1U << 2)
353 #define IORING_ASYNC_CANCEL_FD_FIXED	(1U << 3)
354 #define IORING_ASYNC_CANCEL_USERDATA	(1U << 4)
355 #define IORING_ASYNC_CANCEL_OP	(1U << 5)
356 
357 /*
358  * send/sendmsg and recv/recvmsg flags (sqe->ioprio)
359  *
360  * IORING_RECVSEND_POLL_FIRST	If set, instead of first attempting to send
361  *				or receive and arm poll if that yields an
362  *				-EAGAIN result, arm poll upfront and skip
363  *				the initial transfer attempt.
364  *
365  * IORING_RECV_MULTISHOT	Multishot recv. Sets IORING_CQE_F_MORE if
366  *				the handler will continue to report
367  *				CQEs on behalf of the same SQE.
368  *
369  * IORING_RECVSEND_FIXED_BUF	Use registered buffers, the index is stored in
370  *				the buf_index field.
371  *
372  * IORING_SEND_ZC_REPORT_USAGE
373  *				If set, SEND[MSG]_ZC should report
374  *				the zerocopy usage in cqe.res
375  *				for the IORING_CQE_F_NOTIF cqe.
376  *				0 is reported if zerocopy was actually possible.
377  *				IORING_NOTIF_USAGE_ZC_COPIED if data was copied
378  *				(at least partially).
379  *
380  * IORING_RECVSEND_BUNDLE	Used with IOSQE_BUFFER_SELECT. If set, send or
381  *				recv will grab as many buffers from the buffer
382  *				group ID given and send them all. The completion
383  *				result 	will be the number of buffers send, with
384  *				the starting buffer ID in cqe->flags as per
385  *				usual for provided buffer usage. The buffers
386  *				will be	contigious from the starting buffer ID.
387  */
388 #define IORING_RECVSEND_POLL_FIRST	(1U << 0)
389 #define IORING_RECV_MULTISHOT		(1U << 1)
390 #define IORING_RECVSEND_FIXED_BUF	(1U << 2)
391 #define IORING_SEND_ZC_REPORT_USAGE	(1U << 3)
392 #define IORING_RECVSEND_BUNDLE		(1U << 4)
393 
394 /*
395  * cqe.res for IORING_CQE_F_NOTIF if
396  * IORING_SEND_ZC_REPORT_USAGE was requested
397  *
398  * It should be treated as a flag, all other
399  * bits of cqe.res should be treated as reserved!
400  */
401 #define IORING_NOTIF_USAGE_ZC_COPIED    (1U << 31)
402 
403 /*
404  * accept flags stored in sqe->ioprio
405  */
406 #define IORING_ACCEPT_MULTISHOT	(1U << 0)
407 #define IORING_ACCEPT_DONTWAIT	(1U << 1)
408 #define IORING_ACCEPT_POLL_FIRST	(1U << 2)
409 
410 /*
411  * IORING_OP_MSG_RING command types, stored in sqe->addr
412  */
413 enum io_uring_msg_ring_flags {
414 	IORING_MSG_DATA,	/* pass sqe->len as 'res' and off as user_data */
415 	IORING_MSG_SEND_FD,	/* send a registered fd to another ring */
416 };
417 
418 /*
419  * IORING_OP_MSG_RING flags (sqe->msg_ring_flags)
420  *
421  * IORING_MSG_RING_CQE_SKIP	Don't post a CQE to the target ring. Not
422  *				applicable for IORING_MSG_DATA, obviously.
423  */
424 #define IORING_MSG_RING_CQE_SKIP	(1U << 0)
425 /* Pass through the flags from sqe->file_index to cqe->flags */
426 #define IORING_MSG_RING_FLAGS_PASS	(1U << 1)
427 
428 /*
429  * IORING_OP_FIXED_FD_INSTALL flags (sqe->install_fd_flags)
430  *
431  * IORING_FIXED_FD_NO_CLOEXEC	Don't mark the fd as O_CLOEXEC
432  */
433 #define IORING_FIXED_FD_NO_CLOEXEC	(1U << 0)
434 
435 /*
436  * IORING_OP_NOP flags (sqe->nop_flags)
437  *
438  * IORING_NOP_INJECT_RESULT	Inject result from sqe->result
439  */
440 #define IORING_NOP_INJECT_RESULT	(1U << 0)
441 #define IORING_NOP_FILE			(1U << 1)
442 #define IORING_NOP_FIXED_FILE		(1U << 2)
443 #define IORING_NOP_FIXED_BUFFER		(1U << 3)
444 
445 /*
446  * IO completion data structure (Completion Queue Entry)
447  */
448 struct io_uring_cqe {
449 	__u64	user_data;	/* sqe->user_data value passed back */
450 	__s32	res;		/* result code for this event */
451 	__u32	flags;
452 
453 	/*
454 	 * If the ring is initialized with IORING_SETUP_CQE32, then this field
455 	 * contains 16-bytes of padding, doubling the size of the CQE.
456 	 */
457 	__u64 big_cqe[];
458 };
459 
460 /*
461  * cqe->flags
462  *
463  * IORING_CQE_F_BUFFER	If set, the upper 16 bits are the buffer ID
464  * IORING_CQE_F_MORE	If set, parent SQE will generate more CQE entries
465  * IORING_CQE_F_SOCK_NONEMPTY	If set, more data to read after socket recv
466  * IORING_CQE_F_NOTIF	Set for notification CQEs. Can be used to distinct
467  * 			them from sends.
468  * IORING_CQE_F_BUF_MORE If set, the buffer ID set in the completion will get
469  *			more completions. In other words, the buffer is being
470  *			partially consumed, and will be used by the kernel for
471  *			more completions. This is only set for buffers used via
472  *			the incremental buffer consumption, as provided by
473  *			a ring buffer setup with IOU_PBUF_RING_INC. For any
474  *			other provided buffer type, all completions with a
475  *			buffer passed back is automatically returned to the
476  *			application.
477  */
478 #define IORING_CQE_F_BUFFER		(1U << 0)
479 #define IORING_CQE_F_MORE		(1U << 1)
480 #define IORING_CQE_F_SOCK_NONEMPTY	(1U << 2)
481 #define IORING_CQE_F_NOTIF		(1U << 3)
482 #define IORING_CQE_F_BUF_MORE		(1U << 4)
483 
484 #define IORING_CQE_BUFFER_SHIFT		16
485 
486 /*
487  * Magic offsets for the application to mmap the data it needs
488  */
489 #define IORING_OFF_SQ_RING		0ULL
490 #define IORING_OFF_CQ_RING		0x8000000ULL
491 #define IORING_OFF_SQES			0x10000000ULL
492 #define IORING_OFF_PBUF_RING		0x80000000ULL
493 #define IORING_OFF_PBUF_SHIFT		16
494 #define IORING_OFF_MMAP_MASK		0xf8000000ULL
495 
496 /*
497  * Filled with the offset for mmap(2)
498  */
499 struct io_sqring_offsets {
500 	__u32 head;
501 	__u32 tail;
502 	__u32 ring_mask;
503 	__u32 ring_entries;
504 	__u32 flags;
505 	__u32 dropped;
506 	__u32 array;
507 	__u32 resv1;
508 	__u64 user_addr;
509 };
510 
511 /*
512  * sq_ring->flags
513  */
514 #define IORING_SQ_NEED_WAKEUP	(1U << 0) /* needs io_uring_enter wakeup */
515 #define IORING_SQ_CQ_OVERFLOW	(1U << 1) /* CQ ring is overflown */
516 #define IORING_SQ_TASKRUN	(1U << 2) /* task should enter the kernel */
517 
518 struct io_cqring_offsets {
519 	__u32 head;
520 	__u32 tail;
521 	__u32 ring_mask;
522 	__u32 ring_entries;
523 	__u32 overflow;
524 	__u32 cqes;
525 	__u32 flags;
526 	__u32 resv1;
527 	__u64 user_addr;
528 };
529 
530 /*
531  * cq_ring->flags
532  */
533 
534 /* disable eventfd notifications */
535 #define IORING_CQ_EVENTFD_DISABLED	(1U << 0)
536 
537 /*
538  * io_uring_enter(2) flags
539  */
540 #define IORING_ENTER_GETEVENTS		(1U << 0)
541 #define IORING_ENTER_SQ_WAKEUP		(1U << 1)
542 #define IORING_ENTER_SQ_WAIT		(1U << 2)
543 #define IORING_ENTER_EXT_ARG		(1U << 3)
544 #define IORING_ENTER_REGISTERED_RING	(1U << 4)
545 #define IORING_ENTER_ABS_TIMER		(1U << 5)
546 #define IORING_ENTER_EXT_ARG_REG	(1U << 6)
547 
548 /*
549  * Passed in for io_uring_setup(2). Copied back with updated info on success
550  */
551 struct io_uring_params {
552 	__u32 sq_entries;
553 	__u32 cq_entries;
554 	__u32 flags;
555 	__u32 sq_thread_cpu;
556 	__u32 sq_thread_idle;
557 	__u32 features;
558 	__u32 wq_fd;
559 	__u32 resv[3];
560 	struct io_sqring_offsets sq_off;
561 	struct io_cqring_offsets cq_off;
562 };
563 
564 /*
565  * io_uring_params->features flags
566  */
567 #define IORING_FEAT_SINGLE_MMAP		(1U << 0)
568 #define IORING_FEAT_NODROP		(1U << 1)
569 #define IORING_FEAT_SUBMIT_STABLE	(1U << 2)
570 #define IORING_FEAT_RW_CUR_POS		(1U << 3)
571 #define IORING_FEAT_CUR_PERSONALITY	(1U << 4)
572 #define IORING_FEAT_FAST_POLL		(1U << 5)
573 #define IORING_FEAT_POLL_32BITS 	(1U << 6)
574 #define IORING_FEAT_SQPOLL_NONFIXED	(1U << 7)
575 #define IORING_FEAT_EXT_ARG		(1U << 8)
576 #define IORING_FEAT_NATIVE_WORKERS	(1U << 9)
577 #define IORING_FEAT_RSRC_TAGS		(1U << 10)
578 #define IORING_FEAT_CQE_SKIP		(1U << 11)
579 #define IORING_FEAT_LINKED_FILE		(1U << 12)
580 #define IORING_FEAT_REG_REG_RING	(1U << 13)
581 #define IORING_FEAT_RECVSEND_BUNDLE	(1U << 14)
582 #define IORING_FEAT_MIN_TIMEOUT		(1U << 15)
583 #define IORING_FEAT_RW_ATTR		(1U << 16)
584 
585 /*
586  * io_uring_register(2) opcodes and arguments
587  */
588 enum io_uring_register_op {
589 	IORING_REGISTER_BUFFERS			= 0,
590 	IORING_UNREGISTER_BUFFERS		= 1,
591 	IORING_REGISTER_FILES			= 2,
592 	IORING_UNREGISTER_FILES			= 3,
593 	IORING_REGISTER_EVENTFD			= 4,
594 	IORING_UNREGISTER_EVENTFD		= 5,
595 	IORING_REGISTER_FILES_UPDATE		= 6,
596 	IORING_REGISTER_EVENTFD_ASYNC		= 7,
597 	IORING_REGISTER_PROBE			= 8,
598 	IORING_REGISTER_PERSONALITY		= 9,
599 	IORING_UNREGISTER_PERSONALITY		= 10,
600 	IORING_REGISTER_RESTRICTIONS		= 11,
601 	IORING_REGISTER_ENABLE_RINGS		= 12,
602 
603 	/* extended with tagging */
604 	IORING_REGISTER_FILES2			= 13,
605 	IORING_REGISTER_FILES_UPDATE2		= 14,
606 	IORING_REGISTER_BUFFERS2		= 15,
607 	IORING_REGISTER_BUFFERS_UPDATE		= 16,
608 
609 	/* set/clear io-wq thread affinities */
610 	IORING_REGISTER_IOWQ_AFF		= 17,
611 	IORING_UNREGISTER_IOWQ_AFF		= 18,
612 
613 	/* set/get max number of io-wq workers */
614 	IORING_REGISTER_IOWQ_MAX_WORKERS	= 19,
615 
616 	/* register/unregister io_uring fd with the ring */
617 	IORING_REGISTER_RING_FDS		= 20,
618 	IORING_UNREGISTER_RING_FDS		= 21,
619 
620 	/* register ring based provide buffer group */
621 	IORING_REGISTER_PBUF_RING		= 22,
622 	IORING_UNREGISTER_PBUF_RING		= 23,
623 
624 	/* sync cancelation API */
625 	IORING_REGISTER_SYNC_CANCEL		= 24,
626 
627 	/* register a range of fixed file slots for automatic slot allocation */
628 	IORING_REGISTER_FILE_ALLOC_RANGE	= 25,
629 
630 	/* return status information for a buffer group */
631 	IORING_REGISTER_PBUF_STATUS		= 26,
632 
633 	/* set/clear busy poll settings */
634 	IORING_REGISTER_NAPI			= 27,
635 	IORING_UNREGISTER_NAPI			= 28,
636 
637 	IORING_REGISTER_CLOCK			= 29,
638 
639 	/* clone registered buffers from source ring to current ring */
640 	IORING_REGISTER_CLONE_BUFFERS		= 30,
641 
642 	/* send MSG_RING without having a ring */
643 	IORING_REGISTER_SEND_MSG_RING		= 31,
644 
645 	/* register a netdev hw rx queue for zerocopy */
646 	IORING_REGISTER_ZCRX_IFQ		= 32,
647 
648 	/* resize CQ ring */
649 	IORING_REGISTER_RESIZE_RINGS		= 33,
650 
651 	IORING_REGISTER_MEM_REGION		= 34,
652 
653 	/* this goes last */
654 	IORING_REGISTER_LAST,
655 
656 	/* flag added to the opcode to use a registered ring fd */
657 	IORING_REGISTER_USE_REGISTERED_RING	= 1U << 31
658 };
659 
660 /* io-wq worker categories */
661 enum io_wq_type {
662 	IO_WQ_BOUND,
663 	IO_WQ_UNBOUND,
664 };
665 
666 /* deprecated, see struct io_uring_rsrc_update */
667 struct io_uring_files_update {
668 	__u32 offset;
669 	__u32 resv;
670 	__aligned_u64 /* __s32 * */ fds;
671 };
672 
673 enum {
674 	/* initialise with user provided memory pointed by user_addr */
675 	IORING_MEM_REGION_TYPE_USER		= 1,
676 };
677 
678 struct io_uring_region_desc {
679 	__u64 user_addr;
680 	__u64 size;
681 	__u32 flags;
682 	__u32 id;
683 	__u64 mmap_offset;
684 	__u64 __resv[4];
685 };
686 
687 enum {
688 	/* expose the region as registered wait arguments */
689 	IORING_MEM_REGION_REG_WAIT_ARG		= 1,
690 };
691 
692 struct io_uring_mem_region_reg {
693 	__u64 region_uptr; /* struct io_uring_region_desc * */
694 	__u64 flags;
695 	__u64 __resv[2];
696 };
697 
698 /*
699  * Register a fully sparse file space, rather than pass in an array of all
700  * -1 file descriptors.
701  */
702 #define IORING_RSRC_REGISTER_SPARSE	(1U << 0)
703 
704 struct io_uring_rsrc_register {
705 	__u32 nr;
706 	__u32 flags;
707 	__u64 resv2;
708 	__aligned_u64 data;
709 	__aligned_u64 tags;
710 };
711 
712 struct io_uring_rsrc_update {
713 	__u32 offset;
714 	__u32 resv;
715 	__aligned_u64 data;
716 };
717 
718 struct io_uring_rsrc_update2 {
719 	__u32 offset;
720 	__u32 resv;
721 	__aligned_u64 data;
722 	__aligned_u64 tags;
723 	__u32 nr;
724 	__u32 resv2;
725 };
726 
727 /* Skip updating fd indexes set to this value in the fd table */
728 #define IORING_REGISTER_FILES_SKIP	(-2)
729 
730 #define IO_URING_OP_SUPPORTED	(1U << 0)
731 
732 struct io_uring_probe_op {
733 	__u8 op;
734 	__u8 resv;
735 	__u16 flags;	/* IO_URING_OP_* flags */
736 	__u32 resv2;
737 };
738 
739 struct io_uring_probe {
740 	__u8 last_op;	/* last opcode supported */
741 	__u8 ops_len;	/* length of ops[] array below */
742 	__u16 resv;
743 	__u32 resv2[3];
744 	struct io_uring_probe_op ops[];
745 };
746 
747 struct io_uring_restriction {
748 	__u16 opcode;
749 	union {
750 		__u8 register_op; /* IORING_RESTRICTION_REGISTER_OP */
751 		__u8 sqe_op;      /* IORING_RESTRICTION_SQE_OP */
752 		__u8 sqe_flags;   /* IORING_RESTRICTION_SQE_FLAGS_* */
753 	};
754 	__u8 resv;
755 	__u32 resv2[3];
756 };
757 
758 struct io_uring_clock_register {
759 	__u32	clockid;
760 	__u32	__resv[3];
761 };
762 
763 enum {
764 	IORING_REGISTER_SRC_REGISTERED	= (1U << 0),
765 	IORING_REGISTER_DST_REPLACE	= (1U << 1),
766 };
767 
768 struct io_uring_clone_buffers {
769 	__u32	src_fd;
770 	__u32	flags;
771 	__u32	src_off;
772 	__u32	dst_off;
773 	__u32	nr;
774 	__u32	pad[3];
775 };
776 
777 struct io_uring_buf {
778 	__u64	addr;
779 	__u32	len;
780 	__u16	bid;
781 	__u16	resv;
782 };
783 
784 struct io_uring_buf_ring {
785 	union {
786 		/*
787 		 * To avoid spilling into more pages than we need to, the
788 		 * ring tail is overlaid with the io_uring_buf->resv field.
789 		 */
790 		struct {
791 			__u64	resv1;
792 			__u32	resv2;
793 			__u16	resv3;
794 			__u16	tail;
795 		};
796 		__DECLARE_FLEX_ARRAY(struct io_uring_buf, bufs);
797 	};
798 };
799 
800 /*
801  * Flags for IORING_REGISTER_PBUF_RING.
802  *
803  * IOU_PBUF_RING_MMAP:	If set, kernel will allocate the memory for the ring.
804  *			The application must not set a ring_addr in struct
805  *			io_uring_buf_reg, instead it must subsequently call
806  *			mmap(2) with the offset set as:
807  *			IORING_OFF_PBUF_RING | (bgid << IORING_OFF_PBUF_SHIFT)
808  *			to get a virtual mapping for the ring.
809  * IOU_PBUF_RING_INC:	If set, buffers consumed from this buffer ring can be
810  *			consumed incrementally. Normally one (or more) buffers
811  *			are fully consumed. With incremental consumptions, it's
812  *			feasible to register big ranges of buffers, and each
813  *			use of it will consume only as much as it needs. This
814  *			requires that both the kernel and application keep
815  *			track of where the current read/recv index is at.
816  */
817 enum io_uring_register_pbuf_ring_flags {
818 	IOU_PBUF_RING_MMAP	= 1,
819 	IOU_PBUF_RING_INC	= 2,
820 };
821 
822 /* argument for IORING_(UN)REGISTER_PBUF_RING */
823 struct io_uring_buf_reg {
824 	__u64	ring_addr;
825 	__u32	ring_entries;
826 	__u16	bgid;
827 	__u16	flags;
828 	__u64	resv[3];
829 };
830 
831 /* argument for IORING_REGISTER_PBUF_STATUS */
832 struct io_uring_buf_status {
833 	__u32	buf_group;	/* input */
834 	__u32	head;		/* output */
835 	__u32	resv[8];
836 };
837 
838 enum io_uring_napi_op {
839 	/* register/ungister backward compatible opcode */
840 	IO_URING_NAPI_REGISTER_OP = 0,
841 
842 	/* opcodes to update napi_list when static tracking is used */
843 	IO_URING_NAPI_STATIC_ADD_ID = 1,
844 	IO_URING_NAPI_STATIC_DEL_ID = 2
845 };
846 
847 enum io_uring_napi_tracking_strategy {
848 	/* value must be 0 for backward compatibility */
849 	IO_URING_NAPI_TRACKING_DYNAMIC = 0,
850 	IO_URING_NAPI_TRACKING_STATIC = 1,
851 	IO_URING_NAPI_TRACKING_INACTIVE = 255
852 };
853 
854 /* argument for IORING_(UN)REGISTER_NAPI */
855 struct io_uring_napi {
856 	__u32	busy_poll_to;
857 	__u8	prefer_busy_poll;
858 
859 	/* a io_uring_napi_op value */
860 	__u8	opcode;
861 	__u8	pad[2];
862 
863 	/*
864 	 * for IO_URING_NAPI_REGISTER_OP, it is a
865 	 * io_uring_napi_tracking_strategy value.
866 	 *
867 	 * for IO_URING_NAPI_STATIC_ADD_ID/IO_URING_NAPI_STATIC_DEL_ID
868 	 * it is the napi id to add/del from napi_list.
869 	 */
870 	__u32	op_param;
871 	__u32	resv;
872 };
873 
874 /*
875  * io_uring_restriction->opcode values
876  */
877 enum io_uring_register_restriction_op {
878 	/* Allow an io_uring_register(2) opcode */
879 	IORING_RESTRICTION_REGISTER_OP		= 0,
880 
881 	/* Allow an sqe opcode */
882 	IORING_RESTRICTION_SQE_OP		= 1,
883 
884 	/* Allow sqe flags */
885 	IORING_RESTRICTION_SQE_FLAGS_ALLOWED	= 2,
886 
887 	/* Require sqe flags (these flags must be set on each submission) */
888 	IORING_RESTRICTION_SQE_FLAGS_REQUIRED	= 3,
889 
890 	IORING_RESTRICTION_LAST
891 };
892 
893 enum {
894 	IORING_REG_WAIT_TS		= (1U << 0),
895 };
896 
897 /*
898  * Argument for io_uring_enter(2) with
899  * IORING_GETEVENTS | IORING_ENTER_EXT_ARG_REG set, where the actual argument
900  * is an index into a previously registered fixed wait region described by
901  * the below structure.
902  */
903 struct io_uring_reg_wait {
904 	struct __kernel_timespec	ts;
905 	__u32				min_wait_usec;
906 	__u32				flags;
907 	__u64				sigmask;
908 	__u32				sigmask_sz;
909 	__u32				pad[3];
910 	__u64				pad2[2];
911 };
912 
913 /*
914  * Argument for io_uring_enter(2) with IORING_GETEVENTS | IORING_ENTER_EXT_ARG
915  */
916 struct io_uring_getevents_arg {
917 	__u64	sigmask;
918 	__u32	sigmask_sz;
919 	__u32	min_wait_usec;
920 	__u64	ts;
921 };
922 
923 /*
924  * Argument for IORING_REGISTER_SYNC_CANCEL
925  */
926 struct io_uring_sync_cancel_reg {
927 	__u64				addr;
928 	__s32				fd;
929 	__u32				flags;
930 	struct __kernel_timespec	timeout;
931 	__u8				opcode;
932 	__u8				pad[7];
933 	__u64				pad2[3];
934 };
935 
936 /*
937  * Argument for IORING_REGISTER_FILE_ALLOC_RANGE
938  * The range is specified as [off, off + len)
939  */
940 struct io_uring_file_index_range {
941 	__u32	off;
942 	__u32	len;
943 	__u64	resv;
944 };
945 
946 struct io_uring_recvmsg_out {
947 	__u32 namelen;
948 	__u32 controllen;
949 	__u32 payloadlen;
950 	__u32 flags;
951 };
952 
953 /*
954  * Argument for IORING_OP_URING_CMD when file is a socket
955  */
956 enum io_uring_socket_op {
957 	SOCKET_URING_OP_SIOCINQ		= 0,
958 	SOCKET_URING_OP_SIOCOUTQ,
959 	SOCKET_URING_OP_GETSOCKOPT,
960 	SOCKET_URING_OP_SETSOCKOPT,
961 };
962 
963 /* Zero copy receive refill queue entry */
964 struct io_uring_zcrx_rqe {
965 	__u64	off;
966 	__u32	len;
967 	__u32	__pad;
968 };
969 
970 struct io_uring_zcrx_cqe {
971 	__u64	off;
972 	__u64	__pad;
973 };
974 
975 /* The bit from which area id is encoded into offsets */
976 #define IORING_ZCRX_AREA_SHIFT	48
977 #define IORING_ZCRX_AREA_MASK	(~(((__u64)1 << IORING_ZCRX_AREA_SHIFT) - 1))
978 
979 struct io_uring_zcrx_offsets {
980 	__u32	head;
981 	__u32	tail;
982 	__u32	rqes;
983 	__u32	__resv2;
984 	__u64	__resv[2];
985 };
986 
987 struct io_uring_zcrx_area_reg {
988 	__u64	addr;
989 	__u64	len;
990 	__u64	rq_area_token;
991 	__u32	flags;
992 	__u32	__resv1;
993 	__u64	__resv2[2];
994 };
995 
996 /*
997  * Argument for IORING_REGISTER_ZCRX_IFQ
998  */
999 struct io_uring_zcrx_ifq_reg {
1000 	__u32	if_idx;
1001 	__u32	if_rxq;
1002 	__u32	rq_entries;
1003 	__u32	flags;
1004 
1005 	__u64	area_ptr; /* pointer to struct io_uring_zcrx_area_reg */
1006 	__u64	region_ptr; /* struct io_uring_region_desc * */
1007 
1008 	struct io_uring_zcrx_offsets offsets;
1009 	__u64	__resv[4];
1010 };
1011 
1012 #ifdef __cplusplus
1013 }
1014 #endif
1015 
1016 #endif
1017