1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3 * Copyright (c) 2020 ARM. All rights reserved.
4 * Copyright (c) 2020 Petr Vorel <[email protected]>
5 *
6 * Mostly copied/adapted from <linux/io_uring.h>
7 */
8
9 #ifndef LAPI_IO_URING_H__
10 #define LAPI_IO_URING_H__
11
12 #include <unistd.h>
13 #include <fcntl.h>
14 #include <sys/uio.h>
15 #include <stdlib.h>
16 #include <linux/types.h>
17
18 #include "lapi/syscalls.h"
19
20 #ifdef HAVE_LINUX_IO_URING_H
21 #include <linux/io_uring.h>
22 #endif
23
24 #ifndef IOSQE_FIXED_FILE
25
26 #ifndef __kernel_rwf_t
27 typedef int __kernel_rwf_t;
28 #endif
29
30 /*
31 * IO submission data structure (Submission Queue Entry)
32 */
33 struct io_uring_sqe {
34 uint8_t opcode; /* type of operation for this sqe */
35 uint8_t flags; /* IOSQE_ flags */
36 uint16_t ioprio; /* ioprio for the request */
37 int32_t fd; /* file descriptor to do IO on */
38 union {
39 uint64_t off; /* offset into file */
40 uint64_t addr2;
41 };
42 uint64_t addr; /* pointer to buffer or iovecs */
43 uint32_t len; /* buffer size or number of iovecs */
44 union {
45 __kernel_rwf_t rw_flags;
46 uint32_t fsync_flags;
47 uint16_t poll_events;
48 uint32_t sync_range_flags;
49 uint32_t msg_flags;
50 uint32_t timeout_flags;
51 uint32_t accept_flags;
52 uint32_t cancel_flags;
53 uint32_t open_flags;
54 uint32_t statx_flags;
55 uint32_t fadvise_advice;
56 };
57 uint64_t user_data; /* data to be passed back at completion time */
58 union {
59 struct {
60 /* index into fixed buffers, if used */
61 uint16_t buf_index;
62 /* personality to use, if used */
63 uint16_t personality;
64 };
65 uint64_t __pad2[3];
66 };
67 };
68
69 enum {
70 IOSQE_FIXED_FILE_BIT,
71 IOSQE_IO_DRAIN_BIT,
72 IOSQE_IO_LINK_BIT,
73 };
74
75 /*
76 * sqe->flags
77 */
78 /* use fixed fileset */
79 #define IOSQE_FIXED_FILE (1U << IOSQE_FIXED_FILE_BIT)
80 /* issue after inflight IO */
81 #define IOSQE_IO_DRAIN (1U << IOSQE_IO_DRAIN_BIT)
82 /* links next sqe */
83 #define IOSQE_IO_LINK (1U << IOSQE_IO_LINK_BIT)
84
85 /*
86 * io_uring_setup() flags
87 */
88 #define IORING_SETUP_IOPOLL (1U << 0) /* io_context is polled */
89 #define IORING_SETUP_SQPOLL (1U << 1) /* SQ poll thread */
90 #define IORING_SETUP_SQ_AFF (1U << 2) /* sq_thread_cpu is valid */
91 #define IORING_SETUP_CQSIZE (1U << 3) /* app defines CQ size */
92 #define IORING_SETUP_CLAMP (1U << 4) /* clamp SQ/CQ ring sizes */
93 #define IORING_SETUP_ATTACH_WQ (1U << 5) /* attach to existing wq */
94
95 enum {
96 IORING_OP_NOP,
97 IORING_OP_READV,
98 IORING_OP_WRITEV,
99 IORING_OP_FSYNC,
100 IORING_OP_READ_FIXED,
101 IORING_OP_WRITE_FIXED,
102 IORING_OP_POLL_ADD,
103 IORING_OP_POLL_REMOVE,
104 IORING_OP_SYNC_FILE_RANGE,
105 IORING_OP_SENDMSG,
106 IORING_OP_RECVMSG,
107 IORING_OP_TIMEOUT,
108 IORING_OP_TIMEOUT_REMOVE,
109 IORING_OP_ACCEPT,
110 IORING_OP_ASYNC_CANCEL,
111 IORING_OP_LINK_TIMEOUT,
112 IORING_OP_CONNECT,
113 IORING_OP_FALLOCATE,
114 IORING_OP_OPENAT,
115 IORING_OP_CLOSE,
116 IORING_OP_FILES_UPDATE,
117 IORING_OP_STATX,
118 IORING_OP_READ,
119 IORING_OP_WRITE,
120 IORING_OP_FADVISE,
121 IORING_OP_MADVISE,
122 IORING_OP_SEND,
123 IORING_OP_RECV,
124 IORING_OP_OPENAT2,
125 IORING_OP_EPOLL_CTL,
126
127 /* this goes last, obviously */
128 IORING_OP_LAST,
129 };
130
131 /*
132 * sqe->fsync_flags
133 */
134 #define IORING_FSYNC_DATASYNC (1U << 0)
135
136 /*
137 * sqe->timeout_flags
138 */
139 #define IORING_TIMEOUT_ABS (1U << 0)
140
141 /*
142 * IO completion data structure (Completion Queue Entry)
143 */
144 struct io_uring_cqe {
145 uint64_t user_data; /* sqe->data submission passed back */
146 int32_t res; /* result code for this event */
147 uint32_t flags;
148 };
149
150 /*
151 * Magic offsets for the application to mmap the data it needs
152 */
153 #define IORING_OFF_SQ_RING 0ULL
154 #define IORING_OFF_CQ_RING 0x8000000ULL
155 #define IORING_OFF_SQES 0x10000000ULL
156
157 /*
158 * Filled with the offset for mmap(2)
159 */
160 struct io_sqring_offsets {
161 uint32_t head;
162 uint32_t tail;
163 uint32_t ring_mask;
164 uint32_t ring_entries;
165 uint32_t flags;
166 uint32_t dropped;
167 uint32_t array;
168 uint32_t resv1;
169 uint64_t resv2;
170 };
171
172 /*
173 * sq_ring->flags
174 */
175 #define IORING_SQ_NEED_WAKEUP (1U << 0) /* needs io_uring_enter wakeup */
176
177 struct io_cqring_offsets {
178 uint32_t head;
179 uint32_t tail;
180 uint32_t ring_mask;
181 uint32_t ring_entries;
182 uint32_t overflow;
183 uint32_t cqes;
184 uint64_t resv[2];
185 };
186
187 /*
188 * io_uring_enter(2) flags
189 */
190 #define IORING_ENTER_GETEVENTS (1U << 0)
191 #define IORING_ENTER_SQ_WAKEUP (1U << 1)
192
193 /*
194 * Passed in for io_uring_setup(2). Copied back with updated info on success
195 */
196 struct io_uring_params {
197 uint32_t sq_entries;
198 uint32_t cq_entries;
199 uint32_t flags;
200 uint32_t sq_thread_cpu;
201 uint32_t sq_thread_idle;
202 uint32_t features;
203 uint32_t wq_fd;
204 uint32_t resv[3];
205 struct io_sqring_offsets sq_off;
206 struct io_cqring_offsets cq_off;
207 };
208
209 /*
210 * io_uring_params->features flags
211 */
212 #define IORING_FEAT_SINGLE_MMAP (1U << 0)
213 #define IORING_FEAT_NODROP (1U << 1)
214 #define IORING_FEAT_SUBMIT_STABLE (1U << 2)
215 #define IORING_FEAT_RW_CUR_POS (1U << 3)
216 #define IORING_FEAT_CUR_PERSONALITY (1U << 4)
217
218 /*
219 * io_uring_register(2) opcodes and arguments
220 */
221 #define IORING_REGISTER_BUFFERS 0
222 #define IORING_UNREGISTER_BUFFERS 1
223 #define IORING_REGISTER_FILES 2
224 #define IORING_UNREGISTER_FILES 3
225 #define IORING_REGISTER_EVENTFD 4
226 #define IORING_UNREGISTER_EVENTFD 5
227 #define IORING_REGISTER_FILES_UPDATE 6
228 #define IORING_REGISTER_EVENTFD_ASYNC 7
229 #define IORING_REGISTER_PROBE 8
230 #define IORING_REGISTER_PERSONALITY 9
231 #define IORING_UNREGISTER_PERSONALITY 10
232
233 struct io_uring_files_update {
234 uint32_t offset;
235 uint32_t resv;
236 uint64_t __attribute__((aligned(8))) fds;
237 };
238
239 #define IO_URING_OP_SUPPORTED (1U << 0)
240
241 struct io_uring_probe_op {
242 uint8_t op;
243 uint8_t resv;
244 uint16_t flags; /* IO_URING_OP_* flags */
245 uint32_t resv2;
246 };
247
248 struct io_uring_probe {
249 uint8_t last_op; /* last opcode supported */
250 uint8_t ops_len; /* length of ops[] array below */
251 uint16_t resv;
252 uint32_t resv2[3];
253 struct io_uring_probe_op ops[0];
254 };
255
256 #endif /* IOSQE_FIXED_FILE */
257
258 #ifndef IOSQE_IO_HADRLINK
259 /* like LINK, but stronger */
260 #define IOSQE_IO_HARDLINK_BIT 3
261 #define IOSQE_IO_HARDLINK (1U << IOSQE_IO_HARDLINK_BIT)
262 #endif /* IOSQE_IO_HADRLINK */
263
264 #ifndef IOSQE_ASYNC
265 /* always go async */
266 #define IOSQE_ASYNC_BIT 4
267 #define IOSQE_ASYNC (1U << IOSQE_ASYNC_BIT)
268 #endif /* IOSQE_ASYNC */
269
270 #ifndef HAVE_IO_URING_REGISTER
io_uring_register(int fd,unsigned int opcode,void * arg,unsigned int nr_args)271 static inline int io_uring_register(int fd, unsigned int opcode, void *arg,
272 unsigned int nr_args)
273 {
274 return tst_syscall(__NR_io_uring_register, fd, opcode, arg, nr_args);
275 }
276 #endif /* HAVE_IO_URING_REGISTER */
277
278
279 #ifndef HAVE_IO_URING_SETUP
io_uring_setup(unsigned int entries,struct io_uring_params * p)280 static inline int io_uring_setup(unsigned int entries,
281 struct io_uring_params *p)
282 {
283 return tst_syscall(__NR_io_uring_setup, entries, p);
284 }
285 #endif /* HAVE_IO_URING_SETUP */
286
287 #ifndef HAVE_IO_URING_ENTER
io_uring_enter(int fd,unsigned int to_submit,unsigned int min_complete,unsigned int flags,sigset_t * sig)288 static inline int io_uring_enter(int fd, unsigned int to_submit,
289 unsigned int min_complete, unsigned int flags, sigset_t *sig)
290 {
291 return tst_syscall(__NR_io_uring_enter, fd, to_submit, min_complete,
292 flags, sig, _NSIG / 8);
293 }
294 #endif /* HAVE_IO_URING_ENTER */
295
io_uring_setup_supported_by_kernel(void)296 static inline void io_uring_setup_supported_by_kernel(void)
297 {
298 long ret;
299 ret = syscall(__NR_io_uring_setup, NULL, 0);
300 if (ret != -1) {
301 SAFE_CLOSE(ret);
302 return;
303 }
304
305 if (errno == ENOSYS) {
306 if ((tst_kvercmp(5, 1, 0)) < 0) {
307 tst_brk(TCONF,
308 "Test not supported on kernel version < v5.1");
309 }
310 tst_brk(TCONF, "CONFIG_IO_URING not set?");
311 }
312 }
313
314 #endif /* LAPI_IO_URING_H__ */
315