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 /* pack this to avoid bogus arm OABI complaints */ 50 union { 51 /* index into fixed buffers, if used */ 52 __u16 buf_index; 53 /* for grouped buffer selection */ 54 __u16 buf_group; 55 } __attribute__((packed)); 56 /* personality to use, if used */ 57 __u16 personality; 58 __s32 splice_fd_in; 59 __u64 __pad2[2]; 60 }; 61 62 enum { 63 IOSQE_FIXED_FILE_BIT, 64 IOSQE_IO_DRAIN_BIT, 65 IOSQE_IO_LINK_BIT, 66 IOSQE_IO_HARDLINK_BIT, 67 IOSQE_ASYNC_BIT, 68 IOSQE_BUFFER_SELECT_BIT, 69 }; 70 71 /* 72 * sqe->flags 73 */ 74 /* use fixed fileset */ 75 #define IOSQE_FIXED_FILE (1U << IOSQE_FIXED_FILE_BIT) 76 /* issue after inflight IO */ 77 #define IOSQE_IO_DRAIN (1U << IOSQE_IO_DRAIN_BIT) 78 /* links next sqe */ 79 #define IOSQE_IO_LINK (1U << IOSQE_IO_LINK_BIT) 80 /* like LINK, but stronger */ 81 #define IOSQE_IO_HARDLINK (1U << IOSQE_IO_HARDLINK_BIT) 82 /* always go async */ 83 #define IOSQE_ASYNC (1U << IOSQE_ASYNC_BIT) 84 /* select buffer from sqe->buf_group */ 85 #define IOSQE_BUFFER_SELECT (1U << IOSQE_BUFFER_SELECT_BIT) 86 87 /* 88 * io_uring_setup() flags 89 */ 90 #define IORING_SETUP_IOPOLL (1U << 0) /* io_context is polled */ 91 #define IORING_SETUP_SQPOLL (1U << 1) /* SQ poll thread */ 92 #define IORING_SETUP_SQ_AFF (1U << 2) /* sq_thread_cpu is valid */ 93 #define IORING_SETUP_CQSIZE (1U << 3) /* app defines CQ size */ 94 #define IORING_SETUP_CLAMP (1U << 4) /* clamp SQ/CQ ring sizes */ 95 #define IORING_SETUP_ATTACH_WQ (1U << 5) /* attach to existing wq */ 96 #define IORING_SETUP_R_DISABLED (1U << 6) /* start with ring disabled */ 97 98 enum { 99 IORING_OP_NOP, 100 IORING_OP_READV, 101 IORING_OP_WRITEV, 102 IORING_OP_FSYNC, 103 IORING_OP_READ_FIXED, 104 IORING_OP_WRITE_FIXED, 105 IORING_OP_POLL_ADD, 106 IORING_OP_POLL_REMOVE, 107 IORING_OP_SYNC_FILE_RANGE, 108 IORING_OP_SENDMSG, 109 IORING_OP_RECVMSG, 110 IORING_OP_TIMEOUT, 111 IORING_OP_TIMEOUT_REMOVE, 112 IORING_OP_ACCEPT, 113 IORING_OP_ASYNC_CANCEL, 114 IORING_OP_LINK_TIMEOUT, 115 IORING_OP_CONNECT, 116 IORING_OP_FALLOCATE, 117 IORING_OP_OPENAT, 118 IORING_OP_CLOSE, 119 IORING_OP_FILES_UPDATE, 120 IORING_OP_STATX, 121 IORING_OP_READ, 122 IORING_OP_WRITE, 123 IORING_OP_FADVISE, 124 IORING_OP_MADVISE, 125 IORING_OP_SEND, 126 IORING_OP_RECV, 127 IORING_OP_OPENAT2, 128 IORING_OP_EPOLL_CTL, 129 IORING_OP_SPLICE, 130 IORING_OP_PROVIDE_BUFFERS, 131 IORING_OP_REMOVE_BUFFERS, 132 IORING_OP_TEE, 133 IORING_OP_SHUTDOWN, 134 IORING_OP_RENAMEAT, 135 IORING_OP_UNLINKAT, 136 137 /* this goes last, obviously */ 138 IORING_OP_LAST, 139 }; 140 141 /* 142 * sqe->fsync_flags 143 */ 144 #define IORING_FSYNC_DATASYNC (1U << 0) 145 146 /* 147 * sqe->timeout_flags 148 */ 149 #define IORING_TIMEOUT_ABS (1U << 0) 150 #define IORING_TIMEOUT_UPDATE (1U << 1) 151 152 /* 153 * sqe->splice_flags 154 * extends splice(2) flags 155 */ 156 #define SPLICE_F_FD_IN_FIXED (1U << 31) /* the last bit of __u32 */ 157 158 /* 159 * POLL_ADD flags. Note that since sqe->poll_events is the flag space, the 160 * command flags for POLL_ADD are stored in sqe->len. 161 * 162 * IORING_POLL_ADD_MULTI Multishot poll. Sets IORING_CQE_F_MORE if 163 * the poll handler will continue to report 164 * CQEs on behalf of the same SQE. 165 * 166 * IORING_POLL_UPDATE Update existing poll request, matching 167 * sqe->addr as the old user_data field. 168 */ 169 #define IORING_POLL_ADD_MULTI (1U << 0) 170 #define IORING_POLL_UPDATE_EVENTS (1U << 1) 171 #define IORING_POLL_UPDATE_USER_DATA (1U << 2) 172 173 /* 174 * IO completion data structure (Completion Queue Entry) 175 */ 176 struct io_uring_cqe { 177 __u64 user_data; /* sqe->data submission passed back */ 178 __s32 res; /* result code for this event */ 179 __u32 flags; 180 }; 181 182 /* 183 * cqe->flags 184 * 185 * IORING_CQE_F_BUFFER If set, the upper 16 bits are the buffer ID 186 * IORING_CQE_F_MORE If set, parent SQE will generate more CQE entries 187 */ 188 #define IORING_CQE_F_BUFFER (1U << 0) 189 #define IORING_CQE_F_MORE (1U << 1) 190 191 enum { 192 IORING_CQE_BUFFER_SHIFT = 16, 193 }; 194 195 /* 196 * Magic offsets for the application to mmap the data it needs 197 */ 198 #define IORING_OFF_SQ_RING 0ULL 199 #define IORING_OFF_CQ_RING 0x8000000ULL 200 #define IORING_OFF_SQES 0x10000000ULL 201 202 /* 203 * Filled with the offset for mmap(2) 204 */ 205 struct io_sqring_offsets { 206 __u32 head; 207 __u32 tail; 208 __u32 ring_mask; 209 __u32 ring_entries; 210 __u32 flags; 211 __u32 dropped; 212 __u32 array; 213 __u32 resv1; 214 __u64 resv2; 215 }; 216 217 /* 218 * sq_ring->flags 219 */ 220 #define IORING_SQ_NEED_WAKEUP (1U << 0) /* needs io_uring_enter wakeup */ 221 #define IORING_SQ_CQ_OVERFLOW (1U << 1) /* CQ ring is overflown */ 222 223 struct io_cqring_offsets { 224 __u32 head; 225 __u32 tail; 226 __u32 ring_mask; 227 __u32 ring_entries; 228 __u32 overflow; 229 __u32 cqes; 230 __u32 flags; 231 __u32 resv1; 232 __u64 resv2; 233 }; 234 235 /* 236 * cq_ring->flags 237 */ 238 239 /* disable eventfd notifications */ 240 #define IORING_CQ_EVENTFD_DISABLED (1U << 0) 241 242 /* 243 * io_uring_enter(2) flags 244 */ 245 #define IORING_ENTER_GETEVENTS (1U << 0) 246 #define IORING_ENTER_SQ_WAKEUP (1U << 1) 247 #define IORING_ENTER_SQ_WAIT (1U << 2) 248 #define IORING_ENTER_EXT_ARG (1U << 3) 249 250 /* 251 * Passed in for io_uring_setup(2). Copied back with updated info on success 252 */ 253 struct io_uring_params { 254 __u32 sq_entries; 255 __u32 cq_entries; 256 __u32 flags; 257 __u32 sq_thread_cpu; 258 __u32 sq_thread_idle; 259 __u32 features; 260 __u32 wq_fd; 261 __u32 resv[3]; 262 struct io_sqring_offsets sq_off; 263 struct io_cqring_offsets cq_off; 264 }; 265 266 /* 267 * io_uring_params->features flags 268 */ 269 #define IORING_FEAT_SINGLE_MMAP (1U << 0) 270 #define IORING_FEAT_NODROP (1U << 1) 271 #define IORING_FEAT_SUBMIT_STABLE (1U << 2) 272 #define IORING_FEAT_RW_CUR_POS (1U << 3) 273 #define IORING_FEAT_CUR_PERSONALITY (1U << 4) 274 #define IORING_FEAT_FAST_POLL (1U << 5) 275 #define IORING_FEAT_POLL_32BITS (1U << 6) 276 #define IORING_FEAT_SQPOLL_NONFIXED (1U << 7) 277 #define IORING_FEAT_EXT_ARG (1U << 8) 278 #define IORING_FEAT_NATIVE_WORKERS (1U << 9) 279 #define IORING_FEAT_RSRC_TAGS (1U << 10) 280 281 /* 282 * io_uring_register(2) opcodes and arguments 283 */ 284 enum { 285 IORING_REGISTER_BUFFERS = 0, 286 IORING_UNREGISTER_BUFFERS = 1, 287 IORING_REGISTER_FILES = 2, 288 IORING_UNREGISTER_FILES = 3, 289 IORING_REGISTER_EVENTFD = 4, 290 IORING_UNREGISTER_EVENTFD = 5, 291 IORING_REGISTER_FILES_UPDATE = 6, 292 IORING_REGISTER_EVENTFD_ASYNC = 7, 293 IORING_REGISTER_PROBE = 8, 294 IORING_REGISTER_PERSONALITY = 9, 295 IORING_UNREGISTER_PERSONALITY = 10, 296 IORING_REGISTER_RESTRICTIONS = 11, 297 IORING_REGISTER_ENABLE_RINGS = 12, 298 299 /* extended with tagging */ 300 IORING_REGISTER_FILES2 = 13, 301 IORING_REGISTER_FILES_UPDATE2 = 14, 302 IORING_REGISTER_BUFFERS2 = 15, 303 IORING_REGISTER_BUFFERS_UPDATE = 16, 304 305 /* set/clear io-wq thread affinities */ 306 IORING_REGISTER_IOWQ_AFF = 17, 307 IORING_UNREGISTER_IOWQ_AFF = 18, 308 309 /* this goes last */ 310 IORING_REGISTER_LAST 311 }; 312 313 /* deprecated, see struct io_uring_rsrc_update */ 314 struct io_uring_files_update { 315 __u32 offset; 316 __u32 resv; 317 __aligned_u64 /* __s32 * */ fds; 318 }; 319 320 struct io_uring_rsrc_register { 321 __u32 nr; 322 __u32 resv; 323 __u64 resv2; 324 __aligned_u64 data; 325 __aligned_u64 tags; 326 }; 327 328 struct io_uring_rsrc_update { 329 __u32 offset; 330 __u32 resv; 331 __aligned_u64 data; 332 }; 333 334 struct io_uring_rsrc_update2 { 335 __u32 offset; 336 __u32 resv; 337 __aligned_u64 data; 338 __aligned_u64 tags; 339 __u32 nr; 340 __u32 resv2; 341 }; 342 343 /* Skip updating fd indexes set to this value in the fd table */ 344 #define IORING_REGISTER_FILES_SKIP (-2) 345 346 #define IO_URING_OP_SUPPORTED (1U << 0) 347 348 struct io_uring_probe_op { 349 __u8 op; 350 __u8 resv; 351 __u16 flags; /* IO_URING_OP_* flags */ 352 __u32 resv2; 353 }; 354 355 struct io_uring_probe { 356 __u8 last_op; /* last opcode supported */ 357 __u8 ops_len; /* length of ops[] array below */ 358 __u16 resv; 359 __u32 resv2[3]; 360 struct io_uring_probe_op ops[0]; 361 }; 362 363 struct io_uring_restriction { 364 __u16 opcode; 365 union { 366 __u8 register_op; /* IORING_RESTRICTION_REGISTER_OP */ 367 __u8 sqe_op; /* IORING_RESTRICTION_SQE_OP */ 368 __u8 sqe_flags; /* IORING_RESTRICTION_SQE_FLAGS_* */ 369 }; 370 __u8 resv; 371 __u32 resv2[3]; 372 }; 373 374 /* 375 * io_uring_restriction->opcode values 376 */ 377 enum { 378 /* Allow an io_uring_register(2) opcode */ 379 IORING_RESTRICTION_REGISTER_OP = 0, 380 381 /* Allow an sqe opcode */ 382 IORING_RESTRICTION_SQE_OP = 1, 383 384 /* Allow sqe flags */ 385 IORING_RESTRICTION_SQE_FLAGS_ALLOWED = 2, 386 387 /* Require sqe flags (these flags must be set on each submission) */ 388 IORING_RESTRICTION_SQE_FLAGS_REQUIRED = 3, 389 390 IORING_RESTRICTION_LAST 391 }; 392 393 struct io_uring_getevents_arg { 394 __u64 sigmask; 395 __u32 sigmask_sz; 396 __u32 pad; 397 __u64 ts; 398 }; 399 400 #endif 401