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 __u32 hardlink_flags; 48 }; 49 __u64 user_data; /* data to be passed back at completion time */ 50 /* pack this to avoid bogus arm OABI complaints */ 51 union { 52 /* index into fixed buffers, if used */ 53 __u16 buf_index; 54 /* for grouped buffer selection */ 55 __u16 buf_group; 56 } __attribute__((packed)); 57 /* personality to use, if used */ 58 __u16 personality; 59 union { 60 __s32 splice_fd_in; 61 __u32 file_index; 62 }; 63 __u64 __pad2[2]; 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 IOSQE_CQE_SKIP_SUCCESS_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 /* like LINK, but stronger */ 86 #define IOSQE_IO_HARDLINK (1U << IOSQE_IO_HARDLINK_BIT) 87 /* always go async */ 88 #define IOSQE_ASYNC (1U << IOSQE_ASYNC_BIT) 89 /* select buffer from sqe->buf_group */ 90 #define IOSQE_BUFFER_SELECT (1U << IOSQE_BUFFER_SELECT_BIT) 91 /* don't post CQE if request succeeded */ 92 #define IOSQE_CQE_SKIP_SUCCESS (1U << IOSQE_CQE_SKIP_SUCCESS_BIT) 93 94 /* 95 * io_uring_setup() flags 96 */ 97 #define IORING_SETUP_IOPOLL (1U << 0) /* io_context is polled */ 98 #define IORING_SETUP_SQPOLL (1U << 1) /* SQ poll thread */ 99 #define IORING_SETUP_SQ_AFF (1U << 2) /* sq_thread_cpu is valid */ 100 #define IORING_SETUP_CQSIZE (1U << 3) /* app defines CQ size */ 101 #define IORING_SETUP_CLAMP (1U << 4) /* clamp SQ/CQ ring sizes */ 102 #define IORING_SETUP_ATTACH_WQ (1U << 5) /* attach to existing wq */ 103 #define IORING_SETUP_R_DISABLED (1U << 6) /* start with ring disabled */ 104 105 enum { 106 IORING_OP_NOP, 107 IORING_OP_READV, 108 IORING_OP_WRITEV, 109 IORING_OP_FSYNC, 110 IORING_OP_READ_FIXED, 111 IORING_OP_WRITE_FIXED, 112 IORING_OP_POLL_ADD, 113 IORING_OP_POLL_REMOVE, 114 IORING_OP_SYNC_FILE_RANGE, 115 IORING_OP_SENDMSG, 116 IORING_OP_RECVMSG, 117 IORING_OP_TIMEOUT, 118 IORING_OP_TIMEOUT_REMOVE, 119 IORING_OP_ACCEPT, 120 IORING_OP_ASYNC_CANCEL, 121 IORING_OP_LINK_TIMEOUT, 122 IORING_OP_CONNECT, 123 IORING_OP_FALLOCATE, 124 IORING_OP_OPENAT, 125 IORING_OP_CLOSE, 126 IORING_OP_FILES_UPDATE, 127 IORING_OP_STATX, 128 IORING_OP_READ, 129 IORING_OP_WRITE, 130 IORING_OP_FADVISE, 131 IORING_OP_MADVISE, 132 IORING_OP_SEND, 133 IORING_OP_RECV, 134 IORING_OP_OPENAT2, 135 IORING_OP_EPOLL_CTL, 136 IORING_OP_SPLICE, 137 IORING_OP_PROVIDE_BUFFERS, 138 IORING_OP_REMOVE_BUFFERS, 139 IORING_OP_TEE, 140 IORING_OP_SHUTDOWN, 141 IORING_OP_RENAMEAT, 142 IORING_OP_UNLINKAT, 143 IORING_OP_MKDIRAT, 144 IORING_OP_SYMLINKAT, 145 IORING_OP_LINKAT, 146 147 /* this goes last, obviously */ 148 IORING_OP_LAST, 149 }; 150 151 /* 152 * sqe->fsync_flags 153 */ 154 #define IORING_FSYNC_DATASYNC (1U << 0) 155 156 /* 157 * sqe->timeout_flags 158 */ 159 #define IORING_TIMEOUT_ABS (1U << 0) 160 #define IORING_TIMEOUT_UPDATE (1U << 1) 161 #define IORING_TIMEOUT_BOOTTIME (1U << 2) 162 #define IORING_TIMEOUT_REALTIME (1U << 3) 163 #define IORING_LINK_TIMEOUT_UPDATE (1U << 4) 164 #define IORING_TIMEOUT_ETIME_SUCCESS (1U << 5) 165 #define IORING_TIMEOUT_CLOCK_MASK (IORING_TIMEOUT_BOOTTIME | IORING_TIMEOUT_REALTIME) 166 #define IORING_TIMEOUT_UPDATE_MASK (IORING_TIMEOUT_UPDATE | IORING_LINK_TIMEOUT_UPDATE) 167 /* 168 * sqe->splice_flags 169 * extends splice(2) flags 170 */ 171 #define SPLICE_F_FD_IN_FIXED (1U << 31) /* the last bit of __u32 */ 172 173 /* 174 * POLL_ADD flags. Note that since sqe->poll_events is the flag space, the 175 * command flags for POLL_ADD are stored in sqe->len. 176 * 177 * IORING_POLL_ADD_MULTI Multishot poll. Sets IORING_CQE_F_MORE if 178 * the poll handler will continue to report 179 * CQEs on behalf of the same SQE. 180 * 181 * IORING_POLL_UPDATE Update existing poll request, matching 182 * sqe->addr as the old user_data field. 183 */ 184 #define IORING_POLL_ADD_MULTI (1U << 0) 185 #define IORING_POLL_UPDATE_EVENTS (1U << 1) 186 #define IORING_POLL_UPDATE_USER_DATA (1U << 2) 187 188 /* 189 * IO completion data structure (Completion Queue Entry) 190 */ 191 struct io_uring_cqe { 192 __u64 user_data; /* sqe->data submission passed back */ 193 __s32 res; /* result code for this event */ 194 __u32 flags; 195 }; 196 197 /* 198 * cqe->flags 199 * 200 * IORING_CQE_F_BUFFER If set, the upper 16 bits are the buffer ID 201 * IORING_CQE_F_MORE If set, parent SQE will generate more CQE entries 202 */ 203 #define IORING_CQE_F_BUFFER (1U << 0) 204 #define IORING_CQE_F_MORE (1U << 1) 205 206 enum { 207 IORING_CQE_BUFFER_SHIFT = 16, 208 }; 209 210 /* 211 * Magic offsets for the application to mmap the data it needs 212 */ 213 #define IORING_OFF_SQ_RING 0ULL 214 #define IORING_OFF_CQ_RING 0x8000000ULL 215 #define IORING_OFF_SQES 0x10000000ULL 216 217 /* 218 * Filled with the offset for mmap(2) 219 */ 220 struct io_sqring_offsets { 221 __u32 head; 222 __u32 tail; 223 __u32 ring_mask; 224 __u32 ring_entries; 225 __u32 flags; 226 __u32 dropped; 227 __u32 array; 228 __u32 resv1; 229 __u64 resv2; 230 }; 231 232 /* 233 * sq_ring->flags 234 */ 235 #define IORING_SQ_NEED_WAKEUP (1U << 0) /* needs io_uring_enter wakeup */ 236 #define IORING_SQ_CQ_OVERFLOW (1U << 1) /* CQ ring is overflown */ 237 238 struct io_cqring_offsets { 239 __u32 head; 240 __u32 tail; 241 __u32 ring_mask; 242 __u32 ring_entries; 243 __u32 overflow; 244 __u32 cqes; 245 __u32 flags; 246 __u32 resv1; 247 __u64 resv2; 248 }; 249 250 /* 251 * cq_ring->flags 252 */ 253 254 /* disable eventfd notifications */ 255 #define IORING_CQ_EVENTFD_DISABLED (1U << 0) 256 257 /* 258 * io_uring_enter(2) flags 259 */ 260 #define IORING_ENTER_GETEVENTS (1U << 0) 261 #define IORING_ENTER_SQ_WAKEUP (1U << 1) 262 #define IORING_ENTER_SQ_WAIT (1U << 2) 263 #define IORING_ENTER_EXT_ARG (1U << 3) 264 265 /* 266 * Passed in for io_uring_setup(2). Copied back with updated info on success 267 */ 268 struct io_uring_params { 269 __u32 sq_entries; 270 __u32 cq_entries; 271 __u32 flags; 272 __u32 sq_thread_cpu; 273 __u32 sq_thread_idle; 274 __u32 features; 275 __u32 wq_fd; 276 __u32 resv[3]; 277 struct io_sqring_offsets sq_off; 278 struct io_cqring_offsets cq_off; 279 }; 280 281 /* 282 * io_uring_params->features flags 283 */ 284 #define IORING_FEAT_SINGLE_MMAP (1U << 0) 285 #define IORING_FEAT_NODROP (1U << 1) 286 #define IORING_FEAT_SUBMIT_STABLE (1U << 2) 287 #define IORING_FEAT_RW_CUR_POS (1U << 3) 288 #define IORING_FEAT_CUR_PERSONALITY (1U << 4) 289 #define IORING_FEAT_FAST_POLL (1U << 5) 290 #define IORING_FEAT_POLL_32BITS (1U << 6) 291 #define IORING_FEAT_SQPOLL_NONFIXED (1U << 7) 292 #define IORING_FEAT_EXT_ARG (1U << 8) 293 #define IORING_FEAT_NATIVE_WORKERS (1U << 9) 294 #define IORING_FEAT_RSRC_TAGS (1U << 10) 295 #define IORING_FEAT_CQE_SKIP (1U << 11) 296 297 /* 298 * io_uring_register(2) opcodes and arguments 299 */ 300 enum { 301 IORING_REGISTER_BUFFERS = 0, 302 IORING_UNREGISTER_BUFFERS = 1, 303 IORING_REGISTER_FILES = 2, 304 IORING_UNREGISTER_FILES = 3, 305 IORING_REGISTER_EVENTFD = 4, 306 IORING_UNREGISTER_EVENTFD = 5, 307 IORING_REGISTER_FILES_UPDATE = 6, 308 IORING_REGISTER_EVENTFD_ASYNC = 7, 309 IORING_REGISTER_PROBE = 8, 310 IORING_REGISTER_PERSONALITY = 9, 311 IORING_UNREGISTER_PERSONALITY = 10, 312 IORING_REGISTER_RESTRICTIONS = 11, 313 IORING_REGISTER_ENABLE_RINGS = 12, 314 315 /* extended with tagging */ 316 IORING_REGISTER_FILES2 = 13, 317 IORING_REGISTER_FILES_UPDATE2 = 14, 318 IORING_REGISTER_BUFFERS2 = 15, 319 IORING_REGISTER_BUFFERS_UPDATE = 16, 320 321 /* set/clear io-wq thread affinities */ 322 IORING_REGISTER_IOWQ_AFF = 17, 323 IORING_UNREGISTER_IOWQ_AFF = 18, 324 325 /* set/get max number of io-wq workers */ 326 IORING_REGISTER_IOWQ_MAX_WORKERS = 19, 327 328 /* this goes last */ 329 IORING_REGISTER_LAST 330 }; 331 332 /* io-wq worker categories */ 333 enum { 334 IO_WQ_BOUND, 335 IO_WQ_UNBOUND, 336 }; 337 338 /* deprecated, see struct io_uring_rsrc_update */ 339 struct io_uring_files_update { 340 __u32 offset; 341 __u32 resv; 342 __aligned_u64 /* __s32 * */ fds; 343 }; 344 345 struct io_uring_rsrc_register { 346 __u32 nr; 347 __u32 resv; 348 __u64 resv2; 349 __aligned_u64 data; 350 __aligned_u64 tags; 351 }; 352 353 struct io_uring_rsrc_update { 354 __u32 offset; 355 __u32 resv; 356 __aligned_u64 data; 357 }; 358 359 struct io_uring_rsrc_update2 { 360 __u32 offset; 361 __u32 resv; 362 __aligned_u64 data; 363 __aligned_u64 tags; 364 __u32 nr; 365 __u32 resv2; 366 }; 367 368 /* Skip updating fd indexes set to this value in the fd table */ 369 #define IORING_REGISTER_FILES_SKIP (-2) 370 371 #define IO_URING_OP_SUPPORTED (1U << 0) 372 373 struct io_uring_probe_op { 374 __u8 op; 375 __u8 resv; 376 __u16 flags; /* IO_URING_OP_* flags */ 377 __u32 resv2; 378 }; 379 380 struct io_uring_probe { 381 __u8 last_op; /* last opcode supported */ 382 __u8 ops_len; /* length of ops[] array below */ 383 __u16 resv; 384 __u32 resv2[3]; 385 struct io_uring_probe_op ops[0]; 386 }; 387 388 struct io_uring_restriction { 389 __u16 opcode; 390 union { 391 __u8 register_op; /* IORING_RESTRICTION_REGISTER_OP */ 392 __u8 sqe_op; /* IORING_RESTRICTION_SQE_OP */ 393 __u8 sqe_flags; /* IORING_RESTRICTION_SQE_FLAGS_* */ 394 }; 395 __u8 resv; 396 __u32 resv2[3]; 397 }; 398 399 /* 400 * io_uring_restriction->opcode values 401 */ 402 enum { 403 /* Allow an io_uring_register(2) opcode */ 404 IORING_RESTRICTION_REGISTER_OP = 0, 405 406 /* Allow an sqe opcode */ 407 IORING_RESTRICTION_SQE_OP = 1, 408 409 /* Allow sqe flags */ 410 IORING_RESTRICTION_SQE_FLAGS_ALLOWED = 2, 411 412 /* Require sqe flags (these flags must be set on each submission) */ 413 IORING_RESTRICTION_SQE_FLAGS_REQUIRED = 3, 414 415 IORING_RESTRICTION_LAST 416 }; 417 418 struct io_uring_getevents_arg { 419 __u64 sigmask; 420 __u32 sigmask_sz; 421 __u32 pad; 422 __u64 ts; 423 }; 424 425 #endif 426