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 IORING_OP_MKDIRAT, 141 IORING_OP_SYMLINKAT, 142 IORING_OP_LINKAT, 143 144 /* this goes last, obviously */ 145 IORING_OP_LAST, 146 }; 147 148 /* 149 * sqe->fsync_flags 150 */ 151 #define IORING_FSYNC_DATASYNC (1U << 0) 152 153 /* 154 * sqe->timeout_flags 155 */ 156 #define IORING_TIMEOUT_ABS (1U << 0) 157 #define IORING_TIMEOUT_UPDATE (1U << 1) 158 #define IORING_TIMEOUT_BOOTTIME (1U << 2) 159 #define IORING_TIMEOUT_REALTIME (1U << 3) 160 #define IORING_LINK_TIMEOUT_UPDATE (1U << 4) 161 #define IORING_TIMEOUT_CLOCK_MASK (IORING_TIMEOUT_BOOTTIME | IORING_TIMEOUT_REALTIME) 162 #define IORING_TIMEOUT_UPDATE_MASK (IORING_TIMEOUT_UPDATE | IORING_LINK_TIMEOUT_UPDATE) 163 /* 164 * sqe->splice_flags 165 * extends splice(2) flags 166 */ 167 #define SPLICE_F_FD_IN_FIXED (1U << 31) /* the last bit of __u32 */ 168 169 /* 170 * POLL_ADD flags. Note that since sqe->poll_events is the flag space, the 171 * command flags for POLL_ADD are stored in sqe->len. 172 * 173 * IORING_POLL_ADD_MULTI Multishot poll. Sets IORING_CQE_F_MORE if 174 * the poll handler will continue to report 175 * CQEs on behalf of the same SQE. 176 * 177 * IORING_POLL_UPDATE Update existing poll request, matching 178 * sqe->addr as the old user_data field. 179 */ 180 #define IORING_POLL_ADD_MULTI (1U << 0) 181 #define IORING_POLL_UPDATE_EVENTS (1U << 1) 182 #define IORING_POLL_UPDATE_USER_DATA (1U << 2) 183 184 /* 185 * IO completion data structure (Completion Queue Entry) 186 */ 187 struct io_uring_cqe { 188 __u64 user_data; /* sqe->data submission passed back */ 189 __s32 res; /* result code for this event */ 190 __u32 flags; 191 }; 192 193 /* 194 * cqe->flags 195 * 196 * IORING_CQE_F_BUFFER If set, the upper 16 bits are the buffer ID 197 * IORING_CQE_F_MORE If set, parent SQE will generate more CQE entries 198 */ 199 #define IORING_CQE_F_BUFFER (1U << 0) 200 #define IORING_CQE_F_MORE (1U << 1) 201 202 enum { 203 IORING_CQE_BUFFER_SHIFT = 16, 204 }; 205 206 /* 207 * Magic offsets for the application to mmap the data it needs 208 */ 209 #define IORING_OFF_SQ_RING 0ULL 210 #define IORING_OFF_CQ_RING 0x8000000ULL 211 #define IORING_OFF_SQES 0x10000000ULL 212 213 /* 214 * Filled with the offset for mmap(2) 215 */ 216 struct io_sqring_offsets { 217 __u32 head; 218 __u32 tail; 219 __u32 ring_mask; 220 __u32 ring_entries; 221 __u32 flags; 222 __u32 dropped; 223 __u32 array; 224 __u32 resv1; 225 __u64 resv2; 226 }; 227 228 /* 229 * sq_ring->flags 230 */ 231 #define IORING_SQ_NEED_WAKEUP (1U << 0) /* needs io_uring_enter wakeup */ 232 #define IORING_SQ_CQ_OVERFLOW (1U << 1) /* CQ ring is overflown */ 233 234 struct io_cqring_offsets { 235 __u32 head; 236 __u32 tail; 237 __u32 ring_mask; 238 __u32 ring_entries; 239 __u32 overflow; 240 __u32 cqes; 241 __u32 flags; 242 __u32 resv1; 243 __u64 resv2; 244 }; 245 246 /* 247 * cq_ring->flags 248 */ 249 250 /* disable eventfd notifications */ 251 #define IORING_CQ_EVENTFD_DISABLED (1U << 0) 252 253 /* 254 * io_uring_enter(2) flags 255 */ 256 #define IORING_ENTER_GETEVENTS (1U << 0) 257 #define IORING_ENTER_SQ_WAKEUP (1U << 1) 258 #define IORING_ENTER_SQ_WAIT (1U << 2) 259 #define IORING_ENTER_EXT_ARG (1U << 3) 260 261 /* 262 * Passed in for io_uring_setup(2). Copied back with updated info on success 263 */ 264 struct io_uring_params { 265 __u32 sq_entries; 266 __u32 cq_entries; 267 __u32 flags; 268 __u32 sq_thread_cpu; 269 __u32 sq_thread_idle; 270 __u32 features; 271 __u32 wq_fd; 272 __u32 resv[3]; 273 struct io_sqring_offsets sq_off; 274 struct io_cqring_offsets cq_off; 275 }; 276 277 /* 278 * io_uring_params->features flags 279 */ 280 #define IORING_FEAT_SINGLE_MMAP (1U << 0) 281 #define IORING_FEAT_NODROP (1U << 1) 282 #define IORING_FEAT_SUBMIT_STABLE (1U << 2) 283 #define IORING_FEAT_RW_CUR_POS (1U << 3) 284 #define IORING_FEAT_CUR_PERSONALITY (1U << 4) 285 #define IORING_FEAT_FAST_POLL (1U << 5) 286 #define IORING_FEAT_POLL_32BITS (1U << 6) 287 #define IORING_FEAT_SQPOLL_NONFIXED (1U << 7) 288 #define IORING_FEAT_EXT_ARG (1U << 8) 289 #define IORING_FEAT_NATIVE_WORKERS (1U << 9) 290 #define IORING_FEAT_RSRC_TAGS (1U << 10) 291 292 /* 293 * io_uring_register(2) opcodes and arguments 294 */ 295 enum { 296 IORING_REGISTER_BUFFERS = 0, 297 IORING_UNREGISTER_BUFFERS = 1, 298 IORING_REGISTER_FILES = 2, 299 IORING_UNREGISTER_FILES = 3, 300 IORING_REGISTER_EVENTFD = 4, 301 IORING_UNREGISTER_EVENTFD = 5, 302 IORING_REGISTER_FILES_UPDATE = 6, 303 IORING_REGISTER_EVENTFD_ASYNC = 7, 304 IORING_REGISTER_PROBE = 8, 305 IORING_REGISTER_PERSONALITY = 9, 306 IORING_UNREGISTER_PERSONALITY = 10, 307 IORING_REGISTER_RESTRICTIONS = 11, 308 IORING_REGISTER_ENABLE_RINGS = 12, 309 310 /* extended with tagging */ 311 IORING_REGISTER_FILES2 = 13, 312 IORING_REGISTER_FILES_UPDATE2 = 14, 313 IORING_REGISTER_BUFFERS2 = 15, 314 IORING_REGISTER_BUFFERS_UPDATE = 16, 315 316 /* set/clear io-wq thread affinities */ 317 IORING_REGISTER_IOWQ_AFF = 17, 318 IORING_UNREGISTER_IOWQ_AFF = 18, 319 320 /* set/get max number of io-wq workers */ 321 IORING_REGISTER_IOWQ_MAX_WORKERS = 19, 322 323 /* this goes last */ 324 IORING_REGISTER_LAST 325 }; 326 327 /* io-wq worker categories */ 328 enum { 329 IO_WQ_BOUND, 330 IO_WQ_UNBOUND, 331 }; 332 333 /* deprecated, see struct io_uring_rsrc_update */ 334 struct io_uring_files_update { 335 __u32 offset; 336 __u32 resv; 337 __aligned_u64 /* __s32 * */ fds; 338 }; 339 340 struct io_uring_rsrc_register { 341 __u32 nr; 342 __u32 resv; 343 __u64 resv2; 344 __aligned_u64 data; 345 __aligned_u64 tags; 346 }; 347 348 struct io_uring_rsrc_update { 349 __u32 offset; 350 __u32 resv; 351 __aligned_u64 data; 352 }; 353 354 struct io_uring_rsrc_update2 { 355 __u32 offset; 356 __u32 resv; 357 __aligned_u64 data; 358 __aligned_u64 tags; 359 __u32 nr; 360 __u32 resv2; 361 }; 362 363 /* Skip updating fd indexes set to this value in the fd table */ 364 #define IORING_REGISTER_FILES_SKIP (-2) 365 366 #define IO_URING_OP_SUPPORTED (1U << 0) 367 368 struct io_uring_probe_op { 369 __u8 op; 370 __u8 resv; 371 __u16 flags; /* IO_URING_OP_* flags */ 372 __u32 resv2; 373 }; 374 375 struct io_uring_probe { 376 __u8 last_op; /* last opcode supported */ 377 __u8 ops_len; /* length of ops[] array below */ 378 __u16 resv; 379 __u32 resv2[3]; 380 struct io_uring_probe_op ops[0]; 381 }; 382 383 struct io_uring_restriction { 384 __u16 opcode; 385 union { 386 __u8 register_op; /* IORING_RESTRICTION_REGISTER_OP */ 387 __u8 sqe_op; /* IORING_RESTRICTION_SQE_OP */ 388 __u8 sqe_flags; /* IORING_RESTRICTION_SQE_FLAGS_* */ 389 }; 390 __u8 resv; 391 __u32 resv2[3]; 392 }; 393 394 /* 395 * io_uring_restriction->opcode values 396 */ 397 enum { 398 /* Allow an io_uring_register(2) opcode */ 399 IORING_RESTRICTION_REGISTER_OP = 0, 400 401 /* Allow an sqe opcode */ 402 IORING_RESTRICTION_SQE_OP = 1, 403 404 /* Allow sqe flags */ 405 IORING_RESTRICTION_SQE_FLAGS_ALLOWED = 2, 406 407 /* Require sqe flags (these flags must be set on each submission) */ 408 IORING_RESTRICTION_SQE_FLAGS_REQUIRED = 3, 409 410 IORING_RESTRICTION_LAST 411 }; 412 413 struct io_uring_getevents_arg { 414 __u64 sigmask; 415 __u32 sigmask_sz; 416 __u32 pad; 417 __u64 ts; 418 }; 419 420 #endif 421
