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 __u32 rename_flags; 46 __u32 unlink_flags; 47 }; 48 __u64 user_data; /* data to be passed back at completion time */ 49 union { 50 struct { 51 /* pack this to avoid bogus arm OABI complaints */ 52 union { 53 /* index into fixed buffers, if used */ 54 __u16 buf_index; 55 /* for grouped buffer selection */ 56 __u16 buf_group; 57 } __attribute__((packed)); 58 /* personality to use, if used */ 59 __u16 personality; 60 __s32 splice_fd_in; 61 }; 62 __u64 __pad2[3]; 63 }; 64 }; 65 66 enum { 67 IOSQE_FIXED_FILE_BIT, 68 IOSQE_IO_DRAIN_BIT, 69 IOSQE_IO_LINK_BIT, 70 IOSQE_IO_HARDLINK_BIT, 71 IOSQE_ASYNC_BIT, 72 IOSQE_BUFFER_SELECT_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 /* like LINK, but stronger */ 85 #define IOSQE_IO_HARDLINK (1U << IOSQE_IO_HARDLINK_BIT) 86 /* always go async */ 87 #define IOSQE_ASYNC (1U << IOSQE_ASYNC_BIT) 88 /* select buffer from sqe->buf_group */ 89 #define IOSQE_BUFFER_SELECT (1U << IOSQE_BUFFER_SELECT_BIT) 90 91 /* 92 * io_uring_setup() flags 93 */ 94 #define IORING_SETUP_IOPOLL (1U << 0) /* io_context is polled */ 95 #define IORING_SETUP_SQPOLL (1U << 1) /* SQ poll thread */ 96 #define IORING_SETUP_SQ_AFF (1U << 2) /* sq_thread_cpu is valid */ 97 #define IORING_SETUP_CQSIZE (1U << 3) /* app defines CQ size */ 98 #define IORING_SETUP_CLAMP (1U << 4) /* clamp SQ/CQ ring sizes */ 99 #define IORING_SETUP_ATTACH_WQ (1U << 5) /* attach to existing wq */ 100 #define IORING_SETUP_R_DISABLED (1U << 6) /* start with ring disabled */ 101 102 enum { 103 IORING_OP_NOP, 104 IORING_OP_READV, 105 IORING_OP_WRITEV, 106 IORING_OP_FSYNC, 107 IORING_OP_READ_FIXED, 108 IORING_OP_WRITE_FIXED, 109 IORING_OP_POLL_ADD, 110 IORING_OP_POLL_REMOVE, 111 IORING_OP_SYNC_FILE_RANGE, 112 IORING_OP_SENDMSG, 113 IORING_OP_RECVMSG, 114 IORING_OP_TIMEOUT, 115 IORING_OP_TIMEOUT_REMOVE, 116 IORING_OP_ACCEPT, 117 IORING_OP_ASYNC_CANCEL, 118 IORING_OP_LINK_TIMEOUT, 119 IORING_OP_CONNECT, 120 IORING_OP_FALLOCATE, 121 IORING_OP_OPENAT, 122 IORING_OP_CLOSE, 123 IORING_OP_FILES_UPDATE, 124 IORING_OP_STATX, 125 IORING_OP_READ, 126 IORING_OP_WRITE, 127 IORING_OP_FADVISE, 128 IORING_OP_MADVISE, 129 IORING_OP_SEND, 130 IORING_OP_RECV, 131 IORING_OP_OPENAT2, 132 IORING_OP_EPOLL_CTL, 133 IORING_OP_SPLICE, 134 IORING_OP_PROVIDE_BUFFERS, 135 IORING_OP_REMOVE_BUFFERS, 136 IORING_OP_TEE, 137 IORING_OP_SHUTDOWN, 138 IORING_OP_RENAMEAT, 139 IORING_OP_UNLINKAT, 140 141 /* this goes last, obviously */ 142 IORING_OP_LAST, 143 }; 144 145 /* 146 * sqe->fsync_flags 147 */ 148 #define IORING_FSYNC_DATASYNC (1U << 0) 149 150 /* 151 * sqe->timeout_flags 152 */ 153 #define IORING_TIMEOUT_ABS (1U << 0) 154 #define IORING_TIMEOUT_UPDATE (1U << 1) 155 156 /* 157 * sqe->splice_flags 158 * extends splice(2) flags 159 */ 160 #define SPLICE_F_FD_IN_FIXED (1U << 31) /* the last bit of __u32 */ 161 162 /* 163 * POLL_ADD flags. Note that since sqe->poll_events is the flag space, the 164 * command flags for POLL_ADD are stored in sqe->len. 165 * 166 * IORING_POLL_ADD_MULTI Multishot poll. Sets IORING_CQE_F_MORE if 167 * the poll handler will continue to report 168 * CQEs on behalf of the same SQE. 169 * 170 * IORING_POLL_UPDATE Update existing poll request, matching 171 * sqe->addr as the old user_data field. 172 */ 173 #define IORING_POLL_ADD_MULTI (1U << 0) 174 #define IORING_POLL_UPDATE_EVENTS (1U << 1) 175 #define IORING_POLL_UPDATE_USER_DATA (1U << 2) 176 177 /* 178 * IO completion data structure (Completion Queue Entry) 179 */ 180 struct io_uring_cqe { 181 __u64 user_data; /* sqe->data submission passed back */ 182 __s32 res; /* result code for this event */ 183 __u32 flags; 184 }; 185 186 /* 187 * cqe->flags 188 * 189 * IORING_CQE_F_BUFFER If set, the upper 16 bits are the buffer ID 190 * IORING_CQE_F_MORE If set, parent SQE will generate more CQE entries 191 */ 192 #define IORING_CQE_F_BUFFER (1U << 0) 193 #define IORING_CQE_F_MORE (1U << 1) 194 195 enum { 196 IORING_CQE_BUFFER_SHIFT = 16, 197 }; 198 199 /* 200 * Magic offsets for the application to mmap the data it needs 201 */ 202 #define IORING_OFF_SQ_RING 0ULL 203 #define IORING_OFF_CQ_RING 0x8000000ULL 204 #define IORING_OFF_SQES 0x10000000ULL 205 206 /* 207 * Filled with the offset for mmap(2) 208 */ 209 struct io_sqring_offsets { 210 __u32 head; 211 __u32 tail; 212 __u32 ring_mask; 213 __u32 ring_entries; 214 __u32 flags; 215 __u32 dropped; 216 __u32 array; 217 __u32 resv1; 218 __u64 resv2; 219 }; 220 221 /* 222 * sq_ring->flags 223 */ 224 #define IORING_SQ_NEED_WAKEUP (1U << 0) /* needs io_uring_enter wakeup */ 225 #define IORING_SQ_CQ_OVERFLOW (1U << 1) /* CQ ring is overflown */ 226 227 struct io_cqring_offsets { 228 __u32 head; 229 __u32 tail; 230 __u32 ring_mask; 231 __u32 ring_entries; 232 __u32 overflow; 233 __u32 cqes; 234 __u32 flags; 235 __u32 resv1; 236 __u64 resv2; 237 }; 238 239 /* 240 * cq_ring->flags 241 */ 242 243 /* disable eventfd notifications */ 244 #define IORING_CQ_EVENTFD_DISABLED (1U << 0) 245 246 /* 247 * io_uring_enter(2) flags 248 */ 249 #define IORING_ENTER_GETEVENTS (1U << 0) 250 #define IORING_ENTER_SQ_WAKEUP (1U << 1) 251 #define IORING_ENTER_SQ_WAIT (1U << 2) 252 #define IORING_ENTER_EXT_ARG (1U << 3) 253 254 /* 255 * Passed in for io_uring_setup(2). Copied back with updated info on success 256 */ 257 struct io_uring_params { 258 __u32 sq_entries; 259 __u32 cq_entries; 260 __u32 flags; 261 __u32 sq_thread_cpu; 262 __u32 sq_thread_idle; 263 __u32 features; 264 __u32 wq_fd; 265 __u32 resv[3]; 266 struct io_sqring_offsets sq_off; 267 struct io_cqring_offsets cq_off; 268 }; 269 270 /* 271 * io_uring_params->features flags 272 */ 273 #define IORING_FEAT_SINGLE_MMAP (1U << 0) 274 #define IORING_FEAT_NODROP (1U << 1) 275 #define IORING_FEAT_SUBMIT_STABLE (1U << 2) 276 #define IORING_FEAT_RW_CUR_POS (1U << 3) 277 #define IORING_FEAT_CUR_PERSONALITY (1U << 4) 278 #define IORING_FEAT_FAST_POLL (1U << 5) 279 #define IORING_FEAT_POLL_32BITS (1U << 6) 280 #define IORING_FEAT_SQPOLL_NONFIXED (1U << 7) 281 #define IORING_FEAT_EXT_ARG (1U << 8) 282 #define IORING_FEAT_NATIVE_WORKERS (1U << 9) 283 284 /* 285 * io_uring_register(2) opcodes and arguments 286 */ 287 enum { 288 IORING_REGISTER_BUFFERS = 0, 289 IORING_UNREGISTER_BUFFERS = 1, 290 IORING_REGISTER_FILES = 2, 291 IORING_UNREGISTER_FILES = 3, 292 IORING_REGISTER_EVENTFD = 4, 293 IORING_UNREGISTER_EVENTFD = 5, 294 IORING_REGISTER_FILES_UPDATE = 6, 295 IORING_REGISTER_EVENTFD_ASYNC = 7, 296 IORING_REGISTER_PROBE = 8, 297 IORING_REGISTER_PERSONALITY = 9, 298 IORING_UNREGISTER_PERSONALITY = 10, 299 IORING_REGISTER_RESTRICTIONS = 11, 300 IORING_REGISTER_ENABLE_RINGS = 12, 301 IORING_REGISTER_RSRC = 13, 302 IORING_REGISTER_RSRC_UPDATE = 14, 303 304 /* this goes last */ 305 IORING_REGISTER_LAST 306 }; 307 308 /* deprecated, see struct io_uring_rsrc_update */ 309 struct io_uring_files_update { 310 __u32 offset; 311 __u32 resv; 312 __aligned_u64 /* __s32 * */ fds; 313 }; 314 315 enum { 316 IORING_RSRC_FILE = 0, 317 IORING_RSRC_BUFFER = 1, 318 }; 319 320 struct io_uring_rsrc_register { 321 __u32 type; 322 __u32 nr; 323 __aligned_u64 data; 324 __aligned_u64 tags; 325 }; 326 327 struct io_uring_rsrc_update { 328 __u32 offset; 329 __u32 resv; 330 __aligned_u64 data; 331 }; 332 333 struct io_uring_rsrc_update2 { 334 __u32 offset; 335 __u32 resv; 336 __aligned_u64 data; 337 __aligned_u64 tags; 338 __u32 type; 339 __u32 nr; 340 }; 341 342 /* Skip updating fd indexes set to this value in the fd table */ 343 #define IORING_REGISTER_FILES_SKIP (-2) 344 345 #define IO_URING_OP_SUPPORTED (1U << 0) 346 347 struct io_uring_probe_op { 348 __u8 op; 349 __u8 resv; 350 __u16 flags; /* IO_URING_OP_* flags */ 351 __u32 resv2; 352 }; 353 354 struct io_uring_probe { 355 __u8 last_op; /* last opcode supported */ 356 __u8 ops_len; /* length of ops[] array below */ 357 __u16 resv; 358 __u32 resv2[3]; 359 struct io_uring_probe_op ops[0]; 360 }; 361 362 struct io_uring_restriction { 363 __u16 opcode; 364 union { 365 __u8 register_op; /* IORING_RESTRICTION_REGISTER_OP */ 366 __u8 sqe_op; /* IORING_RESTRICTION_SQE_OP */ 367 __u8 sqe_flags; /* IORING_RESTRICTION_SQE_FLAGS_* */ 368 }; 369 __u8 resv; 370 __u32 resv2[3]; 371 }; 372 373 /* 374 * io_uring_restriction->opcode values 375 */ 376 enum { 377 /* Allow an io_uring_register(2) opcode */ 378 IORING_RESTRICTION_REGISTER_OP = 0, 379 380 /* Allow an sqe opcode */ 381 IORING_RESTRICTION_SQE_OP = 1, 382 383 /* Allow sqe flags */ 384 IORING_RESTRICTION_SQE_FLAGS_ALLOWED = 2, 385 386 /* Require sqe flags (these flags must be set on each submission) */ 387 IORING_RESTRICTION_SQE_FLAGS_REQUIRED = 3, 388 389 IORING_RESTRICTION_LAST 390 }; 391 392 struct io_uring_getevents_arg { 393 __u64 sigmask; 394 __u32 sigmask_sz; 395 __u32 pad; 396 __u64 ts; 397 }; 398 399 #endif 400