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