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 }; 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 #define IORING_TIMEOUT_BOOTTIME (1U << 2) 156 #define IORING_TIMEOUT_REALTIME (1U << 3) 157 #define IORING_LINK_TIMEOUT_UPDATE (1U << 4) 158 #define IORING_TIMEOUT_CLOCK_MASK (IORING_TIMEOUT_BOOTTIME | IORING_TIMEOUT_REALTIME) 159 #define IORING_TIMEOUT_UPDATE_MASK (IORING_TIMEOUT_UPDATE | IORING_LINK_TIMEOUT_UPDATE) 160 /* 161 * sqe->splice_flags 162 * extends splice(2) flags 163 */ 164 #define SPLICE_F_FD_IN_FIXED (1U << 31) /* the last bit of __u32 */ 165 166 /* 167 * POLL_ADD flags. Note that since sqe->poll_events is the flag space, the 168 * command flags for POLL_ADD are stored in sqe->len. 169 * 170 * IORING_POLL_ADD_MULTI Multishot poll. Sets IORING_CQE_F_MORE if 171 * the poll handler will continue to report 172 * CQEs on behalf of the same SQE. 173 * 174 * IORING_POLL_UPDATE Update existing poll request, matching 175 * sqe->addr as the old user_data field. 176 */ 177 #define IORING_POLL_ADD_MULTI (1U << 0) 178 #define IORING_POLL_UPDATE_EVENTS (1U << 1) 179 #define IORING_POLL_UPDATE_USER_DATA (1U << 2) 180 181 /* 182 * IO completion data structure (Completion Queue Entry) 183 */ 184 struct io_uring_cqe { 185 __u64 user_data; /* sqe->data submission passed back */ 186 __s32 res; /* result code for this event */ 187 __u32 flags; 188 }; 189 190 /* 191 * cqe->flags 192 * 193 * IORING_CQE_F_BUFFER If set, the upper 16 bits are the buffer ID 194 * IORING_CQE_F_MORE If set, parent SQE will generate more CQE entries 195 */ 196 #define IORING_CQE_F_BUFFER (1U << 0) 197 #define IORING_CQE_F_MORE (1U << 1) 198 199 enum { 200 IORING_CQE_BUFFER_SHIFT = 16, 201 }; 202 203 /* 204 * Magic offsets for the application to mmap the data it needs 205 */ 206 #define IORING_OFF_SQ_RING 0ULL 207 #define IORING_OFF_CQ_RING 0x8000000ULL 208 #define IORING_OFF_SQES 0x10000000ULL 209 210 /* 211 * Filled with the offset for mmap(2) 212 */ 213 struct io_sqring_offsets { 214 __u32 head; 215 __u32 tail; 216 __u32 ring_mask; 217 __u32 ring_entries; 218 __u32 flags; 219 __u32 dropped; 220 __u32 array; 221 __u32 resv1; 222 __u64 resv2; 223 }; 224 225 /* 226 * sq_ring->flags 227 */ 228 #define IORING_SQ_NEED_WAKEUP (1U << 0) /* needs io_uring_enter wakeup */ 229 #define IORING_SQ_CQ_OVERFLOW (1U << 1) /* CQ ring is overflown */ 230 231 struct io_cqring_offsets { 232 __u32 head; 233 __u32 tail; 234 __u32 ring_mask; 235 __u32 ring_entries; 236 __u32 overflow; 237 __u32 cqes; 238 __u32 flags; 239 __u32 resv1; 240 __u64 resv2; 241 }; 242 243 /* 244 * cq_ring->flags 245 */ 246 247 /* disable eventfd notifications */ 248 #define IORING_CQ_EVENTFD_DISABLED (1U << 0) 249 250 /* 251 * io_uring_enter(2) flags 252 */ 253 #define IORING_ENTER_GETEVENTS (1U << 0) 254 #define IORING_ENTER_SQ_WAKEUP (1U << 1) 255 #define IORING_ENTER_SQ_WAIT (1U << 2) 256 #define IORING_ENTER_EXT_ARG (1U << 3) 257 258 /* 259 * Passed in for io_uring_setup(2). Copied back with updated info on success 260 */ 261 struct io_uring_params { 262 __u32 sq_entries; 263 __u32 cq_entries; 264 __u32 flags; 265 __u32 sq_thread_cpu; 266 __u32 sq_thread_idle; 267 __u32 features; 268 __u32 wq_fd; 269 __u32 resv[3]; 270 struct io_sqring_offsets sq_off; 271 struct io_cqring_offsets cq_off; 272 }; 273 274 /* 275 * io_uring_params->features flags 276 */ 277 #define IORING_FEAT_SINGLE_MMAP (1U << 0) 278 #define IORING_FEAT_NODROP (1U << 1) 279 #define IORING_FEAT_SUBMIT_STABLE (1U << 2) 280 #define IORING_FEAT_RW_CUR_POS (1U << 3) 281 #define IORING_FEAT_CUR_PERSONALITY (1U << 4) 282 #define IORING_FEAT_FAST_POLL (1U << 5) 283 #define IORING_FEAT_POLL_32BITS (1U << 6) 284 #define IORING_FEAT_SQPOLL_NONFIXED (1U << 7) 285 #define IORING_FEAT_EXT_ARG (1U << 8) 286 #define IORING_FEAT_NATIVE_WORKERS (1U << 9) 287 #define IORING_FEAT_RSRC_TAGS (1U << 10) 288 289 /* 290 * io_uring_register(2) opcodes and arguments 291 */ 292 enum { 293 IORING_REGISTER_BUFFERS = 0, 294 IORING_UNREGISTER_BUFFERS = 1, 295 IORING_REGISTER_FILES = 2, 296 IORING_UNREGISTER_FILES = 3, 297 IORING_REGISTER_EVENTFD = 4, 298 IORING_UNREGISTER_EVENTFD = 5, 299 IORING_REGISTER_FILES_UPDATE = 6, 300 IORING_REGISTER_EVENTFD_ASYNC = 7, 301 IORING_REGISTER_PROBE = 8, 302 IORING_REGISTER_PERSONALITY = 9, 303 IORING_UNREGISTER_PERSONALITY = 10, 304 IORING_REGISTER_RESTRICTIONS = 11, 305 IORING_REGISTER_ENABLE_RINGS = 12, 306 307 /* extended with tagging */ 308 IORING_REGISTER_FILES2 = 13, 309 IORING_REGISTER_FILES_UPDATE2 = 14, 310 IORING_REGISTER_BUFFERS2 = 15, 311 IORING_REGISTER_BUFFERS_UPDATE = 16, 312 313 /* set/clear io-wq thread affinities */ 314 IORING_REGISTER_IOWQ_AFF = 17, 315 IORING_UNREGISTER_IOWQ_AFF = 18, 316 317 /* set/get max number of io-wq workers */ 318 IORING_REGISTER_IOWQ_MAX_WORKERS = 19, 319 320 /* this goes last */ 321 IORING_REGISTER_LAST 322 }; 323 324 /* io-wq worker categories */ 325 enum { 326 IO_WQ_BOUND, 327 IO_WQ_UNBOUND, 328 }; 329 330 /* deprecated, see struct io_uring_rsrc_update */ 331 struct io_uring_files_update { 332 __u32 offset; 333 __u32 resv; 334 __aligned_u64 /* __s32 * */ fds; 335 }; 336 337 struct io_uring_rsrc_register { 338 __u32 nr; 339 __u32 resv; 340 __u64 resv2; 341 __aligned_u64 data; 342 __aligned_u64 tags; 343 }; 344 345 struct io_uring_rsrc_update { 346 __u32 offset; 347 __u32 resv; 348 __aligned_u64 data; 349 }; 350 351 struct io_uring_rsrc_update2 { 352 __u32 offset; 353 __u32 resv; 354 __aligned_u64 data; 355 __aligned_u64 tags; 356 __u32 nr; 357 __u32 resv2; 358 }; 359 360 /* Skip updating fd indexes set to this value in the fd table */ 361 #define IORING_REGISTER_FILES_SKIP (-2) 362 363 #define IO_URING_OP_SUPPORTED (1U << 0) 364 365 struct io_uring_probe_op { 366 __u8 op; 367 __u8 resv; 368 __u16 flags; /* IO_URING_OP_* flags */ 369 __u32 resv2; 370 }; 371 372 struct io_uring_probe { 373 __u8 last_op; /* last opcode supported */ 374 __u8 ops_len; /* length of ops[] array below */ 375 __u16 resv; 376 __u32 resv2[3]; 377 struct io_uring_probe_op ops[0]; 378 }; 379 380 struct io_uring_restriction { 381 __u16 opcode; 382 union { 383 __u8 register_op; /* IORING_RESTRICTION_REGISTER_OP */ 384 __u8 sqe_op; /* IORING_RESTRICTION_SQE_OP */ 385 __u8 sqe_flags; /* IORING_RESTRICTION_SQE_FLAGS_* */ 386 }; 387 __u8 resv; 388 __u32 resv2[3]; 389 }; 390 391 /* 392 * io_uring_restriction->opcode values 393 */ 394 enum { 395 /* Allow an io_uring_register(2) opcode */ 396 IORING_RESTRICTION_REGISTER_OP = 0, 397 398 /* Allow an sqe opcode */ 399 IORING_RESTRICTION_SQE_OP = 1, 400 401 /* Allow sqe flags */ 402 IORING_RESTRICTION_SQE_FLAGS_ALLOWED = 2, 403 404 /* Require sqe flags (these flags must be set on each submission) */ 405 IORING_RESTRICTION_SQE_FLAGS_REQUIRED = 3, 406 407 IORING_RESTRICTION_LAST 408 }; 409 410 struct io_uring_getevents_arg { 411 __u64 sigmask; 412 __u32 sigmask_sz; 413 __u32 pad; 414 __u64 ts; 415 }; 416 417 #endif 418