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 /*
15  * IO submission data structure (Submission Queue Entry)
16  */
17 struct io_uring_sqe {
18 	__u8	opcode;		/* type of operation for this sqe */
19 	__u8	flags;		/* IOSQE_ flags */
20 	__u16	ioprio;		/* ioprio for the request */
21 	__s32	fd;		/* file descriptor to do IO on */
22 	union {
23 		__u64	off;	/* offset into file */
24 		__u64	addr2;
25 	};
26 	union {
27 		__u64	addr;	/* pointer to buffer or iovecs */
28 		__u64	splice_off_in;
29 	};
30 	__u32	len;		/* buffer size or number of iovecs */
31 	union {
32 		__kernel_rwf_t	rw_flags;
33 		__u32		fsync_flags;
34 		__u16		poll_events;	/* compatibility */
35 		__u32		poll32_events;	/* word-reversed for BE */
36 		__u32		sync_range_flags;
37 		__u32		msg_flags;
38 		__u32		timeout_flags;
39 		__u32		accept_flags;
40 		__u32		cancel_flags;
41 		__u32		open_flags;
42 		__u32		statx_flags;
43 		__u32		fadvise_advice;
44 		__u32		splice_flags;
45 	};
46 	__u64	user_data;	/* data to be passed back at completion time */
47 	union {
48 		struct {
49 			/* pack this to avoid bogus arm OABI complaints */
50 			union {
51 				/* index into fixed buffers, if used */
52 				__u16	buf_index;
53 				/* for grouped buffer selection */
54 				__u16	buf_group;
55 			} __attribute__((packed));
56 			/* personality to use, if used */
57 			__u16	personality;
58 			__s32	splice_fd_in;
59 		};
60 		__u64	__pad2[3];
61 	};
62 };
63 
64 enum {
65 	IOSQE_FIXED_FILE_BIT,
66 	IOSQE_IO_DRAIN_BIT,
67 	IOSQE_IO_LINK_BIT,
68 	IOSQE_IO_HARDLINK_BIT,
69 	IOSQE_ASYNC_BIT,
70 	IOSQE_BUFFER_SELECT_BIT,
71 };
72 
73 /*
74  * sqe->flags
75  */
76 /* use fixed fileset */
77 #define IOSQE_FIXED_FILE	(1U << IOSQE_FIXED_FILE_BIT)
78 /* issue after inflight IO */
79 #define IOSQE_IO_DRAIN		(1U << IOSQE_IO_DRAIN_BIT)
80 /* links next sqe */
81 #define IOSQE_IO_LINK		(1U << IOSQE_IO_LINK_BIT)
82 /* like LINK, but stronger */
83 #define IOSQE_IO_HARDLINK	(1U << IOSQE_IO_HARDLINK_BIT)
84 /* always go async */
85 #define IOSQE_ASYNC		(1U << IOSQE_ASYNC_BIT)
86 /* select buffer from sqe->buf_group */
87 #define IOSQE_BUFFER_SELECT	(1U << IOSQE_BUFFER_SELECT_BIT)
88 
89 /*
90  * io_uring_setup() flags
91  */
92 #define IORING_SETUP_IOPOLL	(1U << 0)	/* io_context is polled */
93 #define IORING_SETUP_SQPOLL	(1U << 1)	/* SQ poll thread */
94 #define IORING_SETUP_SQ_AFF	(1U << 2)	/* sq_thread_cpu is valid */
95 #define IORING_SETUP_CQSIZE	(1U << 3)	/* app defines CQ size */
96 #define IORING_SETUP_CLAMP	(1U << 4)	/* clamp SQ/CQ ring sizes */
97 #define IORING_SETUP_ATTACH_WQ	(1U << 5)	/* attach to existing wq */
98 #define IORING_SETUP_R_DISABLED	(1U << 6)	/* start with ring disabled */
99 
100 enum {
101 	IORING_OP_NOP,
102 	IORING_OP_READV,
103 	IORING_OP_WRITEV,
104 	IORING_OP_FSYNC,
105 	IORING_OP_READ_FIXED,
106 	IORING_OP_WRITE_FIXED,
107 	IORING_OP_POLL_ADD,
108 	IORING_OP_POLL_REMOVE,
109 	IORING_OP_SYNC_FILE_RANGE,
110 	IORING_OP_SENDMSG,
111 	IORING_OP_RECVMSG,
112 	IORING_OP_TIMEOUT,
113 	IORING_OP_TIMEOUT_REMOVE,
114 	IORING_OP_ACCEPT,
115 	IORING_OP_ASYNC_CANCEL,
116 	IORING_OP_LINK_TIMEOUT,
117 	IORING_OP_CONNECT,
118 	IORING_OP_FALLOCATE,
119 	IORING_OP_OPENAT,
120 	IORING_OP_CLOSE,
121 	IORING_OP_FILES_UPDATE,
122 	IORING_OP_STATX,
123 	IORING_OP_READ,
124 	IORING_OP_WRITE,
125 	IORING_OP_FADVISE,
126 	IORING_OP_MADVISE,
127 	IORING_OP_SEND,
128 	IORING_OP_RECV,
129 	IORING_OP_OPENAT2,
130 	IORING_OP_EPOLL_CTL,
131 	IORING_OP_SPLICE,
132 	IORING_OP_PROVIDE_BUFFERS,
133 	IORING_OP_REMOVE_BUFFERS,
134 	IORING_OP_TEE,
135 
136 	/* this goes last, obviously */
137 	IORING_OP_LAST,
138 };
139 
140 /*
141  * sqe->fsync_flags
142  */
143 #define IORING_FSYNC_DATASYNC	(1U << 0)
144 
145 /*
146  * sqe->timeout_flags
147  */
148 #define IORING_TIMEOUT_ABS	(1U << 0)
149 
150 /*
151  * sqe->splice_flags
152  * extends splice(2) flags
153  */
154 #define SPLICE_F_FD_IN_FIXED	(1U << 31) /* the last bit of __u32 */
155 
156 /*
157  * IO completion data structure (Completion Queue Entry)
158  */
159 struct io_uring_cqe {
160 	__u64	user_data;	/* sqe->data submission passed back */
161 	__s32	res;		/* result code for this event */
162 	__u32	flags;
163 };
164 
165 /*
166  * cqe->flags
167  *
168  * IORING_CQE_F_BUFFER	If set, the upper 16 bits are the buffer ID
169  */
170 #define IORING_CQE_F_BUFFER		(1U << 0)
171 
172 enum {
173 	IORING_CQE_BUFFER_SHIFT		= 16,
174 };
175 
176 /*
177  * Magic offsets for the application to mmap the data it needs
178  */
179 #define IORING_OFF_SQ_RING		0ULL
180 #define IORING_OFF_CQ_RING		0x8000000ULL
181 #define IORING_OFF_SQES			0x10000000ULL
182 
183 /*
184  * Filled with the offset for mmap(2)
185  */
186 struct io_sqring_offsets {
187 	__u32 head;
188 	__u32 tail;
189 	__u32 ring_mask;
190 	__u32 ring_entries;
191 	__u32 flags;
192 	__u32 dropped;
193 	__u32 array;
194 	__u32 resv1;
195 	__u64 resv2;
196 };
197 
198 /*
199  * sq_ring->flags
200  */
201 #define IORING_SQ_NEED_WAKEUP	(1U << 0) /* needs io_uring_enter wakeup */
202 #define IORING_SQ_CQ_OVERFLOW	(1U << 1) /* CQ ring is overflown */
203 
204 struct io_cqring_offsets {
205 	__u32 head;
206 	__u32 tail;
207 	__u32 ring_mask;
208 	__u32 ring_entries;
209 	__u32 overflow;
210 	__u32 cqes;
211 	__u32 flags;
212 	__u32 resv1;
213 	__u64 resv2;
214 };
215 
216 /*
217  * cq_ring->flags
218  */
219 
220 /* disable eventfd notifications */
221 #define IORING_CQ_EVENTFD_DISABLED	(1U << 0)
222 
223 /*
224  * io_uring_enter(2) flags
225  */
226 #define IORING_ENTER_GETEVENTS	(1U << 0)
227 #define IORING_ENTER_SQ_WAKEUP	(1U << 1)
228 #define IORING_ENTER_SQ_WAIT	(1U << 2)
229 
230 /*
231  * Passed in for io_uring_setup(2). Copied back with updated info on success
232  */
233 struct io_uring_params {
234 	__u32 sq_entries;
235 	__u32 cq_entries;
236 	__u32 flags;
237 	__u32 sq_thread_cpu;
238 	__u32 sq_thread_idle;
239 	__u32 features;
240 	__u32 wq_fd;
241 	__u32 resv[3];
242 	struct io_sqring_offsets sq_off;
243 	struct io_cqring_offsets cq_off;
244 };
245 
246 /*
247  * io_uring_params->features flags
248  */
249 #define IORING_FEAT_SINGLE_MMAP		(1U << 0)
250 #define IORING_FEAT_NODROP		(1U << 1)
251 #define IORING_FEAT_SUBMIT_STABLE	(1U << 2)
252 #define IORING_FEAT_RW_CUR_POS		(1U << 3)
253 #define IORING_FEAT_CUR_PERSONALITY	(1U << 4)
254 #define IORING_FEAT_FAST_POLL		(1U << 5)
255 #define IORING_FEAT_POLL_32BITS 	(1U << 6)
256 
257 /*
258  * io_uring_register(2) opcodes and arguments
259  */
260 enum {
261 	IORING_REGISTER_BUFFERS			= 0,
262 	IORING_UNREGISTER_BUFFERS		= 1,
263 	IORING_REGISTER_FILES			= 2,
264 	IORING_UNREGISTER_FILES			= 3,
265 	IORING_REGISTER_EVENTFD			= 4,
266 	IORING_UNREGISTER_EVENTFD		= 5,
267 	IORING_REGISTER_FILES_UPDATE		= 6,
268 	IORING_REGISTER_EVENTFD_ASYNC		= 7,
269 	IORING_REGISTER_PROBE			= 8,
270 	IORING_REGISTER_PERSONALITY		= 9,
271 	IORING_UNREGISTER_PERSONALITY		= 10,
272 	IORING_REGISTER_RESTRICTIONS		= 11,
273 	IORING_REGISTER_ENABLE_RINGS		= 12,
274 
275 	/* this goes last */
276 	IORING_REGISTER_LAST
277 };
278 
279 struct io_uring_files_update {
280 	__u32 offset;
281 	__u32 resv;
282 	__aligned_u64 /* __s32 * */ fds;
283 };
284 
285 #define IO_URING_OP_SUPPORTED	(1U << 0)
286 
287 struct io_uring_probe_op {
288 	__u8 op;
289 	__u8 resv;
290 	__u16 flags;	/* IO_URING_OP_* flags */
291 	__u32 resv2;
292 };
293 
294 struct io_uring_probe {
295 	__u8 last_op;	/* last opcode supported */
296 	__u8 ops_len;	/* length of ops[] array below */
297 	__u16 resv;
298 	__u32 resv2[3];
299 	struct io_uring_probe_op ops[0];
300 };
301 
302 struct io_uring_restriction {
303 	__u16 opcode;
304 	union {
305 		__u8 register_op; /* IORING_RESTRICTION_REGISTER_OP */
306 		__u8 sqe_op;      /* IORING_RESTRICTION_SQE_OP */
307 		__u8 sqe_flags;   /* IORING_RESTRICTION_SQE_FLAGS_* */
308 	};
309 	__u8 resv;
310 	__u32 resv2[3];
311 };
312 
313 /*
314  * io_uring_restriction->opcode values
315  */
316 enum {
317 	/* Allow an io_uring_register(2) opcode */
318 	IORING_RESTRICTION_REGISTER_OP		= 0,
319 
320 	/* Allow an sqe opcode */
321 	IORING_RESTRICTION_SQE_OP		= 1,
322 
323 	/* Allow sqe flags */
324 	IORING_RESTRICTION_SQE_FLAGS_ALLOWED	= 2,
325 
326 	/* Require sqe flags (these flags must be set on each submission) */
327 	IORING_RESTRICTION_SQE_FLAGS_REQUIRED	= 3,
328 
329 	IORING_RESTRICTION_LAST
330 };
331 
332 #endif
333