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 * this file is shared with liburing and that has to autodetect 15 * if linux/time_types.h is available or not, it can 16 * define UAPI_LINUX_IO_URING_H_SKIP_LINUX_TIME_TYPES_H 17 * if linux/time_types.h is not available 18 */ 19 #ifndef UAPI_LINUX_IO_URING_H_SKIP_LINUX_TIME_TYPES_H 20 #include <linux/time_types.h> 21 #endif 22 23 #ifdef __cplusplus 24 extern "C" { 25 #endif 26 27 /* 28 * IO submission data structure (Submission Queue Entry) 29 */ 30 struct io_uring_sqe { 31 __u8 opcode; /* type of operation for this sqe */ 32 __u8 flags; /* IOSQE_ flags */ 33 __u16 ioprio; /* ioprio for the request */ 34 __s32 fd; /* file descriptor to do IO on */ 35 union { 36 __u64 off; /* offset into file */ 37 __u64 addr2; 38 struct { 39 __u32 cmd_op; 40 __u32 __pad1; 41 }; 42 }; 43 union { 44 __u64 addr; /* pointer to buffer or iovecs */ 45 __u64 splice_off_in; 46 struct { 47 __u32 level; 48 __u32 optname; 49 }; 50 }; 51 __u32 len; /* buffer size or number of iovecs */ 52 union { 53 __kernel_rwf_t rw_flags; 54 __u32 fsync_flags; 55 __u16 poll_events; /* compatibility */ 56 __u32 poll32_events; /* word-reversed for BE */ 57 __u32 sync_range_flags; 58 __u32 msg_flags; 59 __u32 timeout_flags; 60 __u32 accept_flags; 61 __u32 cancel_flags; 62 __u32 open_flags; 63 __u32 statx_flags; 64 __u32 fadvise_advice; 65 __u32 splice_flags; 66 __u32 rename_flags; 67 __u32 unlink_flags; 68 __u32 hardlink_flags; 69 __u32 xattr_flags; 70 __u32 msg_ring_flags; 71 __u32 uring_cmd_flags; 72 __u32 waitid_flags; 73 __u32 futex_flags; 74 __u32 install_fd_flags; 75 __u32 nop_flags; 76 }; 77 __u64 user_data; /* data to be passed back at completion time */ 78 /* pack this to avoid bogus arm OABI complaints */ 79 union { 80 /* index into fixed buffers, if used */ 81 __u16 buf_index; 82 /* for grouped buffer selection */ 83 __u16 buf_group; 84 } __attribute__((packed)); 85 /* personality to use, if used */ 86 __u16 personality; 87 union { 88 __s32 splice_fd_in; 89 __u32 file_index; 90 __u32 optlen; 91 struct { 92 __u16 addr_len; 93 __u16 __pad3[1]; 94 }; 95 }; 96 union { 97 struct { 98 __u64 addr3; 99 __u64 __pad2[1]; 100 }; 101 __u64 optval; 102 /* 103 * If the ring is initialized with IORING_SETUP_SQE128, then 104 * this field is used for 80 bytes of arbitrary command data 105 */ 106 __u8 cmd[0]; 107 }; 108 }; 109 110 /* 111 * If sqe->file_index is set to this for opcodes that instantiate a new 112 * direct descriptor (like openat/openat2/accept), then io_uring will allocate 113 * an available direct descriptor instead of having the application pass one 114 * in. The picked direct descriptor will be returned in cqe->res, or -ENFILE 115 * if the space is full. 116 */ 117 #define IORING_FILE_INDEX_ALLOC (~0U) 118 119 enum { 120 IOSQE_FIXED_FILE_BIT, 121 IOSQE_IO_DRAIN_BIT, 122 IOSQE_IO_LINK_BIT, 123 IOSQE_IO_HARDLINK_BIT, 124 IOSQE_ASYNC_BIT, 125 IOSQE_BUFFER_SELECT_BIT, 126 IOSQE_CQE_SKIP_SUCCESS_BIT, 127 }; 128 129 /* 130 * sqe->flags 131 */ 132 /* use fixed fileset */ 133 #define IOSQE_FIXED_FILE (1U << IOSQE_FIXED_FILE_BIT) 134 /* issue after inflight IO */ 135 #define IOSQE_IO_DRAIN (1U << IOSQE_IO_DRAIN_BIT) 136 /* links next sqe */ 137 #define IOSQE_IO_LINK (1U << IOSQE_IO_LINK_BIT) 138 /* like LINK, but stronger */ 139 #define IOSQE_IO_HARDLINK (1U << IOSQE_IO_HARDLINK_BIT) 140 /* always go async */ 141 #define IOSQE_ASYNC (1U << IOSQE_ASYNC_BIT) 142 /* select buffer from sqe->buf_group */ 143 #define IOSQE_BUFFER_SELECT (1U << IOSQE_BUFFER_SELECT_BIT) 144 /* don't post CQE if request succeeded */ 145 #define IOSQE_CQE_SKIP_SUCCESS (1U << IOSQE_CQE_SKIP_SUCCESS_BIT) 146 147 /* 148 * io_uring_setup() flags 149 */ 150 #define IORING_SETUP_IOPOLL (1U << 0) /* io_context is polled */ 151 #define IORING_SETUP_SQPOLL (1U << 1) /* SQ poll thread */ 152 #define IORING_SETUP_SQ_AFF (1U << 2) /* sq_thread_cpu is valid */ 153 #define IORING_SETUP_CQSIZE (1U << 3) /* app defines CQ size */ 154 #define IORING_SETUP_CLAMP (1U << 4) /* clamp SQ/CQ ring sizes */ 155 #define IORING_SETUP_ATTACH_WQ (1U << 5) /* attach to existing wq */ 156 #define IORING_SETUP_R_DISABLED (1U << 6) /* start with ring disabled */ 157 #define IORING_SETUP_SUBMIT_ALL (1U << 7) /* continue submit on error */ 158 /* 159 * Cooperative task running. When requests complete, they often require 160 * forcing the submitter to transition to the kernel to complete. If this 161 * flag is set, work will be done when the task transitions anyway, rather 162 * than force an inter-processor interrupt reschedule. This avoids interrupting 163 * a task running in userspace, and saves an IPI. 164 */ 165 #define IORING_SETUP_COOP_TASKRUN (1U << 8) 166 /* 167 * If COOP_TASKRUN is set, get notified if task work is available for 168 * running and a kernel transition would be needed to run it. This sets 169 * IORING_SQ_TASKRUN in the sq ring flags. Not valid with COOP_TASKRUN. 170 */ 171 #define IORING_SETUP_TASKRUN_FLAG (1U << 9) 172 #define IORING_SETUP_SQE128 (1U << 10) /* SQEs are 128 byte */ 173 #define IORING_SETUP_CQE32 (1U << 11) /* CQEs are 32 byte */ 174 /* 175 * Only one task is allowed to submit requests 176 */ 177 #define IORING_SETUP_SINGLE_ISSUER (1U << 12) 178 179 /* 180 * Defer running task work to get events. 181 * Rather than running bits of task work whenever the task transitions 182 * try to do it just before it is needed. 183 */ 184 #define IORING_SETUP_DEFER_TASKRUN (1U << 13) 185 186 /* 187 * Application provides ring memory 188 */ 189 #define IORING_SETUP_NO_MMAP (1U << 14) 190 191 /* 192 * Register the ring fd in itself for use with 193 * IORING_REGISTER_USE_REGISTERED_RING; return a registered fd index rather 194 * than an fd. 195 */ 196 #define IORING_SETUP_REGISTERED_FD_ONLY (1U << 15) 197 198 /* 199 * Removes indirection through the SQ index array. 200 */ 201 #define IORING_SETUP_NO_SQARRAY (1U << 16) 202 203 enum io_uring_op { 204 IORING_OP_NOP, 205 IORING_OP_READV, 206 IORING_OP_WRITEV, 207 IORING_OP_FSYNC, 208 IORING_OP_READ_FIXED, 209 IORING_OP_WRITE_FIXED, 210 IORING_OP_POLL_ADD, 211 IORING_OP_POLL_REMOVE, 212 IORING_OP_SYNC_FILE_RANGE, 213 IORING_OP_SENDMSG, 214 IORING_OP_RECVMSG, 215 IORING_OP_TIMEOUT, 216 IORING_OP_TIMEOUT_REMOVE, 217 IORING_OP_ACCEPT, 218 IORING_OP_ASYNC_CANCEL, 219 IORING_OP_LINK_TIMEOUT, 220 IORING_OP_CONNECT, 221 IORING_OP_FALLOCATE, 222 IORING_OP_OPENAT, 223 IORING_OP_CLOSE, 224 IORING_OP_FILES_UPDATE, 225 IORING_OP_STATX, 226 IORING_OP_READ, 227 IORING_OP_WRITE, 228 IORING_OP_FADVISE, 229 IORING_OP_MADVISE, 230 IORING_OP_SEND, 231 IORING_OP_RECV, 232 IORING_OP_OPENAT2, 233 IORING_OP_EPOLL_CTL, 234 IORING_OP_SPLICE, 235 IORING_OP_PROVIDE_BUFFERS, 236 IORING_OP_REMOVE_BUFFERS, 237 IORING_OP_TEE, 238 IORING_OP_SHUTDOWN, 239 IORING_OP_RENAMEAT, 240 IORING_OP_UNLINKAT, 241 IORING_OP_MKDIRAT, 242 IORING_OP_SYMLINKAT, 243 IORING_OP_LINKAT, 244 IORING_OP_MSG_RING, 245 IORING_OP_FSETXATTR, 246 IORING_OP_SETXATTR, 247 IORING_OP_FGETXATTR, 248 IORING_OP_GETXATTR, 249 IORING_OP_SOCKET, 250 IORING_OP_URING_CMD, 251 IORING_OP_SEND_ZC, 252 IORING_OP_SENDMSG_ZC, 253 IORING_OP_READ_MULTISHOT, 254 IORING_OP_WAITID, 255 IORING_OP_FUTEX_WAIT, 256 IORING_OP_FUTEX_WAKE, 257 IORING_OP_FUTEX_WAITV, 258 IORING_OP_FIXED_FD_INSTALL, 259 IORING_OP_FTRUNCATE, 260 IORING_OP_BIND, 261 IORING_OP_LISTEN, 262 263 /* this goes last, obviously */ 264 IORING_OP_LAST, 265 }; 266 267 /* 268 * sqe->uring_cmd_flags 269 * IORING_URING_CMD_FIXED use registered buffer; pass this flag 270 * along with setting sqe->buf_index. 271 */ 272 #define IORING_URING_CMD_FIXED (1U << 0) 273 274 275 /* 276 * sqe->fsync_flags 277 */ 278 #define IORING_FSYNC_DATASYNC (1U << 0) 279 280 /* 281 * sqe->timeout_flags 282 */ 283 #define IORING_TIMEOUT_ABS (1U << 0) 284 #define IORING_TIMEOUT_UPDATE (1U << 1) 285 #define IORING_TIMEOUT_BOOTTIME (1U << 2) 286 #define IORING_TIMEOUT_REALTIME (1U << 3) 287 #define IORING_LINK_TIMEOUT_UPDATE (1U << 4) 288 #define IORING_TIMEOUT_ETIME_SUCCESS (1U << 5) 289 #define IORING_TIMEOUT_MULTISHOT (1U << 6) 290 #define IORING_TIMEOUT_CLOCK_MASK (IORING_TIMEOUT_BOOTTIME | IORING_TIMEOUT_REALTIME) 291 #define IORING_TIMEOUT_UPDATE_MASK (IORING_TIMEOUT_UPDATE | IORING_LINK_TIMEOUT_UPDATE) 292 /* 293 * sqe->splice_flags 294 * extends splice(2) flags 295 */ 296 #define SPLICE_F_FD_IN_FIXED (1U << 31) /* the last bit of __u32 */ 297 298 /* 299 * POLL_ADD flags. Note that since sqe->poll_events is the flag space, the 300 * command flags for POLL_ADD are stored in sqe->len. 301 * 302 * IORING_POLL_ADD_MULTI Multishot poll. Sets IORING_CQE_F_MORE if 303 * the poll handler will continue to report 304 * CQEs on behalf of the same SQE. 305 * 306 * IORING_POLL_UPDATE Update existing poll request, matching 307 * sqe->addr as the old user_data field. 308 * 309 * IORING_POLL_LEVEL Level triggered poll. 310 */ 311 #define IORING_POLL_ADD_MULTI (1U << 0) 312 #define IORING_POLL_UPDATE_EVENTS (1U << 1) 313 #define IORING_POLL_UPDATE_USER_DATA (1U << 2) 314 #define IORING_POLL_ADD_LEVEL (1U << 3) 315 316 /* 317 * ASYNC_CANCEL flags. 318 * 319 * IORING_ASYNC_CANCEL_ALL Cancel all requests that match the given key 320 * IORING_ASYNC_CANCEL_FD Key off 'fd' for cancelation rather than the 321 * request 'user_data' 322 * IORING_ASYNC_CANCEL_ANY Match any request 323 * IORING_ASYNC_CANCEL_FD_FIXED 'fd' passed in is a fixed descriptor 324 */ 325 #define IORING_ASYNC_CANCEL_ALL (1U << 0) 326 #define IORING_ASYNC_CANCEL_FD (1U << 1) 327 #define IORING_ASYNC_CANCEL_ANY (1U << 2) 328 #define IORING_ASYNC_CANCEL_FD_FIXED (1U << 3) 329 330 /* 331 * send/sendmsg and recv/recvmsg flags (sqe->ioprio) 332 * 333 * IORING_RECVSEND_POLL_FIRST If set, instead of first attempting to send 334 * or receive and arm poll if that yields an 335 * -EAGAIN result, arm poll upfront and skip 336 * the initial transfer attempt. 337 * 338 * IORING_RECV_MULTISHOT Multishot recv. Sets IORING_CQE_F_MORE if 339 * the handler will continue to report 340 * CQEs on behalf of the same SQE. 341 * 342 * IORING_RECVSEND_FIXED_BUF Use registered buffers, the index is stored in 343 * the buf_index field. 344 * 345 * IORING_SEND_ZC_REPORT_USAGE 346 * If set, SEND[MSG]_ZC should report 347 * the zerocopy usage in cqe.res 348 * for the IORING_CQE_F_NOTIF cqe. 349 * 0 is reported if zerocopy was actually possible. 350 * IORING_NOTIF_USAGE_ZC_COPIED if data was copied 351 * (at least partially). 352 * 353 * IORING_RECVSEND_BUNDLE Used with IOSQE_BUFFER_SELECT. If set, send wil 354 * grab as many buffers from the buffer group ID 355 * given and send them all. The completion result 356 * will be the number of buffers send, with the 357 * starting buffer ID in cqe->flags as per usual 358 * for provided buffer usage. The buffers will be 359 * contigious from the starting buffer ID. 360 */ 361 #define IORING_RECVSEND_POLL_FIRST (1U << 0) 362 #define IORING_RECV_MULTISHOT (1U << 1) 363 #define IORING_RECVSEND_FIXED_BUF (1U << 2) 364 #define IORING_SEND_ZC_REPORT_USAGE (1U << 3) 365 #define IORING_RECVSEND_BUNDLE (1U << 4) 366 367 /* 368 * cqe.res for IORING_CQE_F_NOTIF if 369 * IORING_SEND_ZC_REPORT_USAGE was requested 370 * 371 * It should be treated as a flag, all other 372 * bits of cqe.res should be treated as reserved! 373 */ 374 #define IORING_NOTIF_USAGE_ZC_COPIED (1U << 31) 375 376 /* 377 * accept flags stored in sqe->ioprio 378 */ 379 #define IORING_ACCEPT_MULTISHOT (1U << 0) 380 #define IORING_ACCEPT_DONTWAIT (1U << 1) 381 #define IORING_ACCEPT_POLL_FIRST (1U << 2) 382 383 /* 384 * IORING_OP_MSG_RING command types, stored in sqe->addr 385 */ 386 enum { 387 IORING_MSG_DATA, /* pass sqe->len as 'res' and off as user_data */ 388 IORING_MSG_SEND_FD, /* send a registered fd to another ring */ 389 }; 390 391 /* 392 * IORING_OP_MSG_RING flags (sqe->msg_ring_flags) 393 * 394 * IORING_MSG_RING_CQE_SKIP Don't post a CQE to the target ring. Not 395 * applicable for IORING_MSG_DATA, obviously. 396 */ 397 #define IORING_MSG_RING_CQE_SKIP (1U << 0) 398 /* Pass through the flags from sqe->file_index to cqe->flags */ 399 #define IORING_MSG_RING_FLAGS_PASS (1U << 1) 400 401 /* 402 * IORING_OP_FIXED_FD_INSTALL flags (sqe->install_fd_flags) 403 * 404 * IORING_FIXED_FD_NO_CLOEXEC Don't mark the fd as O_CLOEXEC 405 */ 406 #define IORING_FIXED_FD_NO_CLOEXEC (1U << 0) 407 408 /* 409 * IORING_OP_NOP flags (sqe->nop_flags) 410 * 411 * IORING_NOP_INJECT_RESULT Inject result from sqe->result 412 */ 413 #define IORING_NOP_INJECT_RESULT (1U << 0) 414 415 /* 416 * IO completion data structure (Completion Queue Entry) 417 */ 418 struct io_uring_cqe { 419 __u64 user_data; /* sqe->user_data value passed back */ 420 __s32 res; /* result code for this event */ 421 __u32 flags; 422 423 /* 424 * If the ring is initialized with IORING_SETUP_CQE32, then this field 425 * contains 16-bytes of padding, doubling the size of the CQE. 426 */ 427 __u64 big_cqe[]; 428 }; 429 430 /* 431 * cqe->flags 432 * 433 * IORING_CQE_F_BUFFER If set, the upper 16 bits are the buffer ID 434 * IORING_CQE_F_MORE If set, parent SQE will generate more CQE entries 435 * IORING_CQE_F_SOCK_NONEMPTY If set, more data to read after socket recv 436 * IORING_CQE_F_NOTIF Set for notification CQEs. Can be used to distinct 437 * them from sends. 438 */ 439 #define IORING_CQE_F_BUFFER (1U << 0) 440 #define IORING_CQE_F_MORE (1U << 1) 441 #define IORING_CQE_F_SOCK_NONEMPTY (1U << 2) 442 #define IORING_CQE_F_NOTIF (1U << 3) 443 444 enum { 445 IORING_CQE_BUFFER_SHIFT = 16, 446 }; 447 448 /* 449 * Magic offsets for the application to mmap the data it needs 450 */ 451 #define IORING_OFF_SQ_RING 0ULL 452 #define IORING_OFF_CQ_RING 0x8000000ULL 453 #define IORING_OFF_SQES 0x10000000ULL 454 #define IORING_OFF_PBUF_RING 0x80000000ULL 455 #define IORING_OFF_PBUF_SHIFT 16 456 #define IORING_OFF_MMAP_MASK 0xf8000000ULL 457 458 /* 459 * Filled with the offset for mmap(2) 460 */ 461 struct io_sqring_offsets { 462 __u32 head; 463 __u32 tail; 464 __u32 ring_mask; 465 __u32 ring_entries; 466 __u32 flags; 467 __u32 dropped; 468 __u32 array; 469 __u32 resv1; 470 __u64 user_addr; 471 }; 472 473 /* 474 * sq_ring->flags 475 */ 476 #define IORING_SQ_NEED_WAKEUP (1U << 0) /* needs io_uring_enter wakeup */ 477 #define IORING_SQ_CQ_OVERFLOW (1U << 1) /* CQ ring is overflown */ 478 #define IORING_SQ_TASKRUN (1U << 2) /* task should enter the kernel */ 479 480 struct io_cqring_offsets { 481 __u32 head; 482 __u32 tail; 483 __u32 ring_mask; 484 __u32 ring_entries; 485 __u32 overflow; 486 __u32 cqes; 487 __u32 flags; 488 __u32 resv1; 489 __u64 user_addr; 490 }; 491 492 /* 493 * cq_ring->flags 494 */ 495 496 /* disable eventfd notifications */ 497 #define IORING_CQ_EVENTFD_DISABLED (1U << 0) 498 499 /* 500 * io_uring_enter(2) flags 501 */ 502 #define IORING_ENTER_GETEVENTS (1U << 0) 503 #define IORING_ENTER_SQ_WAKEUP (1U << 1) 504 #define IORING_ENTER_SQ_WAIT (1U << 2) 505 #define IORING_ENTER_EXT_ARG (1U << 3) 506 #define IORING_ENTER_REGISTERED_RING (1U << 4) 507 508 /* 509 * Passed in for io_uring_setup(2). Copied back with updated info on success 510 */ 511 struct io_uring_params { 512 __u32 sq_entries; 513 __u32 cq_entries; 514 __u32 flags; 515 __u32 sq_thread_cpu; 516 __u32 sq_thread_idle; 517 __u32 features; 518 __u32 wq_fd; 519 __u32 resv[3]; 520 struct io_sqring_offsets sq_off; 521 struct io_cqring_offsets cq_off; 522 }; 523 524 /* 525 * io_uring_params->features flags 526 */ 527 #define IORING_FEAT_SINGLE_MMAP (1U << 0) 528 #define IORING_FEAT_NODROP (1U << 1) 529 #define IORING_FEAT_SUBMIT_STABLE (1U << 2) 530 #define IORING_FEAT_RW_CUR_POS (1U << 3) 531 #define IORING_FEAT_CUR_PERSONALITY (1U << 4) 532 #define IORING_FEAT_FAST_POLL (1U << 5) 533 #define IORING_FEAT_POLL_32BITS (1U << 6) 534 #define IORING_FEAT_SQPOLL_NONFIXED (1U << 7) 535 #define IORING_FEAT_EXT_ARG (1U << 8) 536 #define IORING_FEAT_NATIVE_WORKERS (1U << 9) 537 #define IORING_FEAT_RSRC_TAGS (1U << 10) 538 #define IORING_FEAT_CQE_SKIP (1U << 11) 539 #define IORING_FEAT_LINKED_FILE (1U << 12) 540 #define IORING_FEAT_REG_REG_RING (1U << 13) 541 #define IORING_FEAT_RECVSEND_BUNDLE (1U << 14) 542 543 /* 544 * io_uring_register(2) opcodes and arguments 545 */ 546 enum { 547 IORING_REGISTER_BUFFERS = 0, 548 IORING_UNREGISTER_BUFFERS = 1, 549 IORING_REGISTER_FILES = 2, 550 IORING_UNREGISTER_FILES = 3, 551 IORING_REGISTER_EVENTFD = 4, 552 IORING_UNREGISTER_EVENTFD = 5, 553 IORING_REGISTER_FILES_UPDATE = 6, 554 IORING_REGISTER_EVENTFD_ASYNC = 7, 555 IORING_REGISTER_PROBE = 8, 556 IORING_REGISTER_PERSONALITY = 9, 557 IORING_UNREGISTER_PERSONALITY = 10, 558 IORING_REGISTER_RESTRICTIONS = 11, 559 IORING_REGISTER_ENABLE_RINGS = 12, 560 561 /* extended with tagging */ 562 IORING_REGISTER_FILES2 = 13, 563 IORING_REGISTER_FILES_UPDATE2 = 14, 564 IORING_REGISTER_BUFFERS2 = 15, 565 IORING_REGISTER_BUFFERS_UPDATE = 16, 566 567 /* set/clear io-wq thread affinities */ 568 IORING_REGISTER_IOWQ_AFF = 17, 569 IORING_UNREGISTER_IOWQ_AFF = 18, 570 571 /* set/get max number of io-wq workers */ 572 IORING_REGISTER_IOWQ_MAX_WORKERS = 19, 573 574 /* register/unregister io_uring fd with the ring */ 575 IORING_REGISTER_RING_FDS = 20, 576 IORING_UNREGISTER_RING_FDS = 21, 577 578 /* register ring based provide buffer group */ 579 IORING_REGISTER_PBUF_RING = 22, 580 IORING_UNREGISTER_PBUF_RING = 23, 581 582 /* sync cancelation API */ 583 IORING_REGISTER_SYNC_CANCEL = 24, 584 585 /* register a range of fixed file slots for automatic slot allocation */ 586 IORING_REGISTER_FILE_ALLOC_RANGE = 25, 587 588 /* return status information for a buffer group */ 589 IORING_REGISTER_PBUF_STATUS = 26, 590 591 /* set/clear busy poll settings */ 592 IORING_REGISTER_NAPI = 27, 593 IORING_UNREGISTER_NAPI = 28, 594 595 /* this goes last */ 596 IORING_REGISTER_LAST, 597 598 /* flag added to the opcode to use a registered ring fd */ 599 IORING_REGISTER_USE_REGISTERED_RING = 1U << 31 600 }; 601 602 /* io-wq worker categories */ 603 enum { 604 IO_WQ_BOUND, 605 IO_WQ_UNBOUND, 606 }; 607 608 /* deprecated, see struct io_uring_rsrc_update */ 609 struct io_uring_files_update { 610 __u32 offset; 611 __u32 resv; 612 __aligned_u64 /* __s32 * */ fds; 613 }; 614 615 /* 616 * Register a fully sparse file space, rather than pass in an array of all 617 * -1 file descriptors. 618 */ 619 #define IORING_RSRC_REGISTER_SPARSE (1U << 0) 620 621 struct io_uring_rsrc_register { 622 __u32 nr; 623 __u32 flags; 624 __u64 resv2; 625 __aligned_u64 data; 626 __aligned_u64 tags; 627 }; 628 629 struct io_uring_rsrc_update { 630 __u32 offset; 631 __u32 resv; 632 __aligned_u64 data; 633 }; 634 635 struct io_uring_rsrc_update2 { 636 __u32 offset; 637 __u32 resv; 638 __aligned_u64 data; 639 __aligned_u64 tags; 640 __u32 nr; 641 __u32 resv2; 642 }; 643 644 /* Skip updating fd indexes set to this value in the fd table */ 645 #define IORING_REGISTER_FILES_SKIP (-2) 646 647 #define IO_URING_OP_SUPPORTED (1U << 0) 648 649 struct io_uring_probe_op { 650 __u8 op; 651 __u8 resv; 652 __u16 flags; /* IO_URING_OP_* flags */ 653 __u32 resv2; 654 }; 655 656 struct io_uring_probe { 657 __u8 last_op; /* last opcode supported */ 658 __u8 ops_len; /* length of ops[] array below */ 659 __u16 resv; 660 __u32 resv2[3]; 661 struct io_uring_probe_op ops[]; 662 }; 663 664 struct io_uring_restriction { 665 __u16 opcode; 666 union { 667 __u8 register_op; /* IORING_RESTRICTION_REGISTER_OP */ 668 __u8 sqe_op; /* IORING_RESTRICTION_SQE_OP */ 669 __u8 sqe_flags; /* IORING_RESTRICTION_SQE_FLAGS_* */ 670 }; 671 __u8 resv; 672 __u32 resv2[3]; 673 }; 674 675 struct io_uring_buf { 676 __u64 addr; 677 __u32 len; 678 __u16 bid; 679 __u16 resv; 680 }; 681 682 struct io_uring_buf_ring { 683 union { 684 /* 685 * To avoid spilling into more pages than we need to, the 686 * ring tail is overlaid with the io_uring_buf->resv field. 687 */ 688 struct { 689 __u64 resv1; 690 __u32 resv2; 691 __u16 resv3; 692 __u16 tail; 693 }; 694 struct io_uring_buf bufs[0]; 695 }; 696 }; 697 698 /* 699 * Flags for IORING_REGISTER_PBUF_RING. 700 * 701 * IOU_PBUF_RING_MMAP: If set, kernel will allocate the memory for the ring. 702 * The application must not set a ring_addr in struct 703 * io_uring_buf_reg, instead it must subsequently call 704 * mmap(2) with the offset set as: 705 * IORING_OFF_PBUF_RING | (bgid << IORING_OFF_PBUF_SHIFT) 706 * to get a virtual mapping for the ring. 707 */ 708 enum { 709 IOU_PBUF_RING_MMAP = 1, 710 }; 711 712 /* argument for IORING_(UN)REGISTER_PBUF_RING */ 713 struct io_uring_buf_reg { 714 __u64 ring_addr; 715 __u32 ring_entries; 716 __u16 bgid; 717 __u16 flags; 718 __u64 resv[3]; 719 }; 720 721 /* argument for IORING_REGISTER_PBUF_STATUS */ 722 struct io_uring_buf_status { 723 __u32 buf_group; /* input */ 724 __u32 head; /* output */ 725 __u32 resv[8]; 726 }; 727 728 /* argument for IORING_(UN)REGISTER_NAPI */ 729 struct io_uring_napi { 730 __u32 busy_poll_to; 731 __u8 prefer_busy_poll; 732 __u8 pad[3]; 733 __u64 resv; 734 }; 735 736 /* 737 * io_uring_restriction->opcode values 738 */ 739 enum { 740 /* Allow an io_uring_register(2) opcode */ 741 IORING_RESTRICTION_REGISTER_OP = 0, 742 743 /* Allow an sqe opcode */ 744 IORING_RESTRICTION_SQE_OP = 1, 745 746 /* Allow sqe flags */ 747 IORING_RESTRICTION_SQE_FLAGS_ALLOWED = 2, 748 749 /* Require sqe flags (these flags must be set on each submission) */ 750 IORING_RESTRICTION_SQE_FLAGS_REQUIRED = 3, 751 752 IORING_RESTRICTION_LAST 753 }; 754 755 struct io_uring_getevents_arg { 756 __u64 sigmask; 757 __u32 sigmask_sz; 758 __u32 pad; 759 __u64 ts; 760 }; 761 762 /* 763 * Argument for IORING_REGISTER_SYNC_CANCEL 764 */ 765 struct io_uring_sync_cancel_reg { 766 __u64 addr; 767 __s32 fd; 768 __u32 flags; 769 struct __kernel_timespec timeout; 770 __u64 pad[4]; 771 }; 772 773 /* 774 * Argument for IORING_REGISTER_FILE_ALLOC_RANGE 775 * The range is specified as [off, off + len) 776 */ 777 struct io_uring_file_index_range { 778 __u32 off; 779 __u32 len; 780 __u64 resv; 781 }; 782 783 struct io_uring_recvmsg_out { 784 __u32 namelen; 785 __u32 controllen; 786 __u32 payloadlen; 787 __u32 flags; 788 }; 789 790 /* 791 * Argument for IORING_OP_URING_CMD when file is a socket 792 */ 793 enum { 794 SOCKET_URING_OP_SIOCINQ = 0, 795 SOCKET_URING_OP_SIOCOUTQ, 796 SOCKET_URING_OP_GETSOCKOPT, 797 SOCKET_URING_OP_SETSOCKOPT, 798 }; 799 800 #ifdef __cplusplus 801 } 802 #endif 803 804 #endif 805