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 }; 76 __u64 user_data; /* data to be passed back at completion time */ 77 /* pack this to avoid bogus arm OABI complaints */ 78 union { 79 /* index into fixed buffers, if used */ 80 __u16 buf_index; 81 /* for grouped buffer selection */ 82 __u16 buf_group; 83 } __attribute__((packed)); 84 /* personality to use, if used */ 85 __u16 personality; 86 union { 87 __s32 splice_fd_in; 88 __u32 file_index; 89 __u32 optlen; 90 struct { 91 __u16 addr_len; 92 __u16 __pad3[1]; 93 }; 94 }; 95 union { 96 struct { 97 __u64 addr3; 98 __u64 __pad2[1]; 99 }; 100 __u64 optval; 101 /* 102 * If the ring is initialized with IORING_SETUP_SQE128, then 103 * this field is used for 80 bytes of arbitrary command data 104 */ 105 __u8 cmd[0]; 106 }; 107 }; 108 109 /* 110 * If sqe->file_index is set to this for opcodes that instantiate a new 111 * direct descriptor (like openat/openat2/accept), then io_uring will allocate 112 * an available direct descriptor instead of having the application pass one 113 * in. The picked direct descriptor will be returned in cqe->res, or -ENFILE 114 * if the space is full. 115 */ 116 #define IORING_FILE_INDEX_ALLOC (~0U) 117 118 enum { 119 IOSQE_FIXED_FILE_BIT, 120 IOSQE_IO_DRAIN_BIT, 121 IOSQE_IO_LINK_BIT, 122 IOSQE_IO_HARDLINK_BIT, 123 IOSQE_ASYNC_BIT, 124 IOSQE_BUFFER_SELECT_BIT, 125 IOSQE_CQE_SKIP_SUCCESS_BIT, 126 }; 127 128 /* 129 * sqe->flags 130 */ 131 /* use fixed fileset */ 132 #define IOSQE_FIXED_FILE (1U << IOSQE_FIXED_FILE_BIT) 133 /* issue after inflight IO */ 134 #define IOSQE_IO_DRAIN (1U << IOSQE_IO_DRAIN_BIT) 135 /* links next sqe */ 136 #define IOSQE_IO_LINK (1U << IOSQE_IO_LINK_BIT) 137 /* like LINK, but stronger */ 138 #define IOSQE_IO_HARDLINK (1U << IOSQE_IO_HARDLINK_BIT) 139 /* always go async */ 140 #define IOSQE_ASYNC (1U << IOSQE_ASYNC_BIT) 141 /* select buffer from sqe->buf_group */ 142 #define IOSQE_BUFFER_SELECT (1U << IOSQE_BUFFER_SELECT_BIT) 143 /* don't post CQE if request succeeded */ 144 #define IOSQE_CQE_SKIP_SUCCESS (1U << IOSQE_CQE_SKIP_SUCCESS_BIT) 145 146 /* 147 * io_uring_setup() flags 148 */ 149 #define IORING_SETUP_IOPOLL (1U << 0) /* io_context is polled */ 150 #define IORING_SETUP_SQPOLL (1U << 1) /* SQ poll thread */ 151 #define IORING_SETUP_SQ_AFF (1U << 2) /* sq_thread_cpu is valid */ 152 #define IORING_SETUP_CQSIZE (1U << 3) /* app defines CQ size */ 153 #define IORING_SETUP_CLAMP (1U << 4) /* clamp SQ/CQ ring sizes */ 154 #define IORING_SETUP_ATTACH_WQ (1U << 5) /* attach to existing wq */ 155 #define IORING_SETUP_R_DISABLED (1U << 6) /* start with ring disabled */ 156 #define IORING_SETUP_SUBMIT_ALL (1U << 7) /* continue submit on error */ 157 /* 158 * Cooperative task running. When requests complete, they often require 159 * forcing the submitter to transition to the kernel to complete. If this 160 * flag is set, work will be done when the task transitions anyway, rather 161 * than force an inter-processor interrupt reschedule. This avoids interrupting 162 * a task running in userspace, and saves an IPI. 163 */ 164 #define IORING_SETUP_COOP_TASKRUN (1U << 8) 165 /* 166 * If COOP_TASKRUN is set, get notified if task work is available for 167 * running and a kernel transition would be needed to run it. This sets 168 * IORING_SQ_TASKRUN in the sq ring flags. Not valid with COOP_TASKRUN. 169 */ 170 #define IORING_SETUP_TASKRUN_FLAG (1U << 9) 171 #define IORING_SETUP_SQE128 (1U << 10) /* SQEs are 128 byte */ 172 #define IORING_SETUP_CQE32 (1U << 11) /* CQEs are 32 byte */ 173 /* 174 * Only one task is allowed to submit requests 175 */ 176 #define IORING_SETUP_SINGLE_ISSUER (1U << 12) 177 178 /* 179 * Defer running task work to get events. 180 * Rather than running bits of task work whenever the task transitions 181 * try to do it just before it is needed. 182 */ 183 #define IORING_SETUP_DEFER_TASKRUN (1U << 13) 184 185 /* 186 * Application provides the memory for the rings 187 */ 188 #define IORING_SETUP_NO_MMAP (1U << 14) 189 190 /* 191 * Register the ring fd in itself for use with 192 * IORING_REGISTER_USE_REGISTERED_RING; return a registered fd index rather 193 * than an fd. 194 */ 195 #define IORING_SETUP_REGISTERED_FD_ONLY (1U << 15) 196 197 /* 198 * Removes indirection through the SQ index array. 199 */ 200 #define IORING_SETUP_NO_SQARRAY (1U << 16) 201 202 enum io_uring_op { 203 IORING_OP_NOP, 204 IORING_OP_READV, 205 IORING_OP_WRITEV, 206 IORING_OP_FSYNC, 207 IORING_OP_READ_FIXED, 208 IORING_OP_WRITE_FIXED, 209 IORING_OP_POLL_ADD, 210 IORING_OP_POLL_REMOVE, 211 IORING_OP_SYNC_FILE_RANGE, 212 IORING_OP_SENDMSG, 213 IORING_OP_RECVMSG, 214 IORING_OP_TIMEOUT, 215 IORING_OP_TIMEOUT_REMOVE, 216 IORING_OP_ACCEPT, 217 IORING_OP_ASYNC_CANCEL, 218 IORING_OP_LINK_TIMEOUT, 219 IORING_OP_CONNECT, 220 IORING_OP_FALLOCATE, 221 IORING_OP_OPENAT, 222 IORING_OP_CLOSE, 223 IORING_OP_FILES_UPDATE, 224 IORING_OP_STATX, 225 IORING_OP_READ, 226 IORING_OP_WRITE, 227 IORING_OP_FADVISE, 228 IORING_OP_MADVISE, 229 IORING_OP_SEND, 230 IORING_OP_RECV, 231 IORING_OP_OPENAT2, 232 IORING_OP_EPOLL_CTL, 233 IORING_OP_SPLICE, 234 IORING_OP_PROVIDE_BUFFERS, 235 IORING_OP_REMOVE_BUFFERS, 236 IORING_OP_TEE, 237 IORING_OP_SHUTDOWN, 238 IORING_OP_RENAMEAT, 239 IORING_OP_UNLINKAT, 240 IORING_OP_MKDIRAT, 241 IORING_OP_SYMLINKAT, 242 IORING_OP_LINKAT, 243 IORING_OP_MSG_RING, 244 IORING_OP_FSETXATTR, 245 IORING_OP_SETXATTR, 246 IORING_OP_FGETXATTR, 247 IORING_OP_GETXATTR, 248 IORING_OP_SOCKET, 249 IORING_OP_URING_CMD, 250 IORING_OP_SEND_ZC, 251 IORING_OP_SENDMSG_ZC, 252 IORING_OP_READ_MULTISHOT, 253 IORING_OP_WAITID, 254 IORING_OP_FUTEX_WAIT, 255 IORING_OP_FUTEX_WAKE, 256 IORING_OP_FUTEX_WAITV, 257 IORING_OP_FIXED_FD_INSTALL, 258 IORING_OP_FTRUNCATE, 259 260 /* this goes last, obviously */ 261 IORING_OP_LAST, 262 }; 263 264 /* 265 * sqe->uring_cmd_flags top 8bits aren't available for userspace 266 * IORING_URING_CMD_FIXED use registered buffer; pass this flag 267 * along with setting sqe->buf_index. 268 */ 269 #define IORING_URING_CMD_FIXED (1U << 0) 270 #define IORING_URING_CMD_MASK IORING_URING_CMD_FIXED 271 272 273 /* 274 * sqe->fsync_flags 275 */ 276 #define IORING_FSYNC_DATASYNC (1U << 0) 277 278 /* 279 * sqe->timeout_flags 280 */ 281 #define IORING_TIMEOUT_ABS (1U << 0) 282 #define IORING_TIMEOUT_UPDATE (1U << 1) 283 #define IORING_TIMEOUT_BOOTTIME (1U << 2) 284 #define IORING_TIMEOUT_REALTIME (1U << 3) 285 #define IORING_LINK_TIMEOUT_UPDATE (1U << 4) 286 #define IORING_TIMEOUT_ETIME_SUCCESS (1U << 5) 287 #define IORING_TIMEOUT_MULTISHOT (1U << 6) 288 #define IORING_TIMEOUT_CLOCK_MASK (IORING_TIMEOUT_BOOTTIME | IORING_TIMEOUT_REALTIME) 289 #define IORING_TIMEOUT_UPDATE_MASK (IORING_TIMEOUT_UPDATE | IORING_LINK_TIMEOUT_UPDATE) 290 /* 291 * sqe->splice_flags 292 * extends splice(2) flags 293 */ 294 #define SPLICE_F_FD_IN_FIXED (1U << 31) /* the last bit of __u32 */ 295 296 /* 297 * POLL_ADD flags. Note that since sqe->poll_events is the flag space, the 298 * command flags for POLL_ADD are stored in sqe->len. 299 * 300 * IORING_POLL_ADD_MULTI Multishot poll. Sets IORING_CQE_F_MORE if 301 * the poll handler will continue to report 302 * CQEs on behalf of the same SQE. 303 * 304 * IORING_POLL_UPDATE Update existing poll request, matching 305 * sqe->addr as the old user_data field. 306 * 307 * IORING_POLL_LEVEL Level triggered poll. 308 */ 309 #define IORING_POLL_ADD_MULTI (1U << 0) 310 #define IORING_POLL_UPDATE_EVENTS (1U << 1) 311 #define IORING_POLL_UPDATE_USER_DATA (1U << 2) 312 #define IORING_POLL_ADD_LEVEL (1U << 3) 313 314 /* 315 * ASYNC_CANCEL flags. 316 * 317 * IORING_ASYNC_CANCEL_ALL Cancel all requests that match the given key 318 * IORING_ASYNC_CANCEL_FD Key off 'fd' for cancelation rather than the 319 * request 'user_data' 320 * IORING_ASYNC_CANCEL_ANY Match any request 321 * IORING_ASYNC_CANCEL_FD_FIXED 'fd' passed in is a fixed descriptor 322 * IORING_ASYNC_CANCEL_USERDATA Match on user_data, default for no other key 323 * IORING_ASYNC_CANCEL_OP Match request based on opcode 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 #define IORING_ASYNC_CANCEL_USERDATA (1U << 4) 330 #define IORING_ASYNC_CANCEL_OP (1U << 5) 331 332 /* 333 * send/sendmsg and recv/recvmsg flags (sqe->ioprio) 334 * 335 * IORING_RECVSEND_POLL_FIRST If set, instead of first attempting to send 336 * or receive and arm poll if that yields an 337 * -EAGAIN result, arm poll upfront and skip 338 * the initial transfer attempt. 339 * 340 * IORING_RECV_MULTISHOT Multishot recv. Sets IORING_CQE_F_MORE if 341 * the handler will continue to report 342 * CQEs on behalf of the same SQE. 343 * 344 * IORING_RECVSEND_FIXED_BUF Use registered buffers, the index is stored in 345 * the buf_index field. 346 * 347 * IORING_SEND_ZC_REPORT_USAGE 348 * If set, SEND[MSG]_ZC should report 349 * the zerocopy usage in cqe.res 350 * for the IORING_CQE_F_NOTIF cqe. 351 * 0 is reported if zerocopy was actually possible. 352 * IORING_NOTIF_USAGE_ZC_COPIED if data was copied 353 * (at least partially). 354 */ 355 #define IORING_RECVSEND_POLL_FIRST (1U << 0) 356 #define IORING_RECV_MULTISHOT (1U << 1) 357 #define IORING_RECVSEND_FIXED_BUF (1U << 2) 358 #define IORING_SEND_ZC_REPORT_USAGE (1U << 3) 359 360 /* 361 * cqe.res for IORING_CQE_F_NOTIF if 362 * IORING_SEND_ZC_REPORT_USAGE was requested 363 * 364 * It should be treated as a flag, all other 365 * bits of cqe.res should be treated as reserved! 366 */ 367 #define IORING_NOTIF_USAGE_ZC_COPIED (1U << 31) 368 369 /* 370 * accept flags stored in sqe->ioprio 371 */ 372 #define IORING_ACCEPT_MULTISHOT (1U << 0) 373 374 /* 375 * IORING_OP_MSG_RING command types, stored in sqe->addr 376 */ 377 enum { 378 IORING_MSG_DATA, /* pass sqe->len as 'res' and off as user_data */ 379 IORING_MSG_SEND_FD, /* send a registered fd to another ring */ 380 }; 381 382 /* 383 * IORING_OP_MSG_RING flags (sqe->msg_ring_flags) 384 * 385 * IORING_MSG_RING_CQE_SKIP Don't post a CQE to the target ring. Not 386 * applicable for IORING_MSG_DATA, obviously. 387 */ 388 #define IORING_MSG_RING_CQE_SKIP (1U << 0) 389 /* Pass through the flags from sqe->file_index to cqe->flags */ 390 #define IORING_MSG_RING_FLAGS_PASS (1U << 1) 391 392 /* 393 * IORING_OP_FIXED_FD_INSTALL flags (sqe->install_fd_flags) 394 * 395 * IORING_FIXED_FD_NO_CLOEXEC Don't mark the fd as O_CLOEXEC 396 */ 397 #define IORING_FIXED_FD_NO_CLOEXEC (1U << 0) 398 399 /* 400 * IO completion data structure (Completion Queue Entry) 401 */ 402 struct io_uring_cqe { 403 __u64 user_data; /* sqe->data submission passed back */ 404 __s32 res; /* result code for this event */ 405 __u32 flags; 406 407 /* 408 * If the ring is initialized with IORING_SETUP_CQE32, then this field 409 * contains 16-bytes of padding, doubling the size of the CQE. 410 */ 411 __u64 big_cqe[]; 412 }; 413 414 /* 415 * cqe->flags 416 * 417 * IORING_CQE_F_BUFFER If set, the upper 16 bits are the buffer ID 418 * IORING_CQE_F_MORE If set, parent SQE will generate more CQE entries 419 * IORING_CQE_F_SOCK_NONEMPTY If set, more data to read after socket recv 420 * IORING_CQE_F_NOTIF Set for notification CQEs. Can be used to distinct 421 * them from sends. 422 */ 423 #define IORING_CQE_F_BUFFER (1U << 0) 424 #define IORING_CQE_F_MORE (1U << 1) 425 #define IORING_CQE_F_SOCK_NONEMPTY (1U << 2) 426 #define IORING_CQE_F_NOTIF (1U << 3) 427 428 enum { 429 IORING_CQE_BUFFER_SHIFT = 16, 430 }; 431 432 /* 433 * Magic offsets for the application to mmap the data it needs 434 */ 435 #define IORING_OFF_SQ_RING 0ULL 436 #define IORING_OFF_CQ_RING 0x8000000ULL 437 #define IORING_OFF_SQES 0x10000000ULL 438 #define IORING_OFF_PBUF_RING 0x80000000ULL 439 #define IORING_OFF_PBUF_SHIFT 16 440 #define IORING_OFF_MMAP_MASK 0xf8000000ULL 441 442 /* 443 * Filled with the offset for mmap(2) 444 */ 445 struct io_sqring_offsets { 446 __u32 head; 447 __u32 tail; 448 __u32 ring_mask; 449 __u32 ring_entries; 450 __u32 flags; 451 __u32 dropped; 452 __u32 array; 453 __u32 resv1; 454 __u64 user_addr; 455 }; 456 457 /* 458 * sq_ring->flags 459 */ 460 #define IORING_SQ_NEED_WAKEUP (1U << 0) /* needs io_uring_enter wakeup */ 461 #define IORING_SQ_CQ_OVERFLOW (1U << 1) /* CQ ring is overflown */ 462 #define IORING_SQ_TASKRUN (1U << 2) /* task should enter the kernel */ 463 464 struct io_cqring_offsets { 465 __u32 head; 466 __u32 tail; 467 __u32 ring_mask; 468 __u32 ring_entries; 469 __u32 overflow; 470 __u32 cqes; 471 __u32 flags; 472 __u32 resv1; 473 __u64 user_addr; 474 }; 475 476 /* 477 * cq_ring->flags 478 */ 479 480 /* disable eventfd notifications */ 481 #define IORING_CQ_EVENTFD_DISABLED (1U << 0) 482 483 /* 484 * io_uring_enter(2) flags 485 */ 486 #define IORING_ENTER_GETEVENTS (1U << 0) 487 #define IORING_ENTER_SQ_WAKEUP (1U << 1) 488 #define IORING_ENTER_SQ_WAIT (1U << 2) 489 #define IORING_ENTER_EXT_ARG (1U << 3) 490 #define IORING_ENTER_REGISTERED_RING (1U << 4) 491 492 /* 493 * Passed in for io_uring_setup(2). Copied back with updated info on success 494 */ 495 struct io_uring_params { 496 __u32 sq_entries; 497 __u32 cq_entries; 498 __u32 flags; 499 __u32 sq_thread_cpu; 500 __u32 sq_thread_idle; 501 __u32 features; 502 __u32 wq_fd; 503 __u32 resv[3]; 504 struct io_sqring_offsets sq_off; 505 struct io_cqring_offsets cq_off; 506 }; 507 508 /* 509 * io_uring_params->features flags 510 */ 511 #define IORING_FEAT_SINGLE_MMAP (1U << 0) 512 #define IORING_FEAT_NODROP (1U << 1) 513 #define IORING_FEAT_SUBMIT_STABLE (1U << 2) 514 #define IORING_FEAT_RW_CUR_POS (1U << 3) 515 #define IORING_FEAT_CUR_PERSONALITY (1U << 4) 516 #define IORING_FEAT_FAST_POLL (1U << 5) 517 #define IORING_FEAT_POLL_32BITS (1U << 6) 518 #define IORING_FEAT_SQPOLL_NONFIXED (1U << 7) 519 #define IORING_FEAT_EXT_ARG (1U << 8) 520 #define IORING_FEAT_NATIVE_WORKERS (1U << 9) 521 #define IORING_FEAT_RSRC_TAGS (1U << 10) 522 #define IORING_FEAT_CQE_SKIP (1U << 11) 523 #define IORING_FEAT_LINKED_FILE (1U << 12) 524 #define IORING_FEAT_REG_REG_RING (1U << 13) 525 526 /* 527 * io_uring_register(2) opcodes and arguments 528 */ 529 enum { 530 IORING_REGISTER_BUFFERS = 0, 531 IORING_UNREGISTER_BUFFERS = 1, 532 IORING_REGISTER_FILES = 2, 533 IORING_UNREGISTER_FILES = 3, 534 IORING_REGISTER_EVENTFD = 4, 535 IORING_UNREGISTER_EVENTFD = 5, 536 IORING_REGISTER_FILES_UPDATE = 6, 537 IORING_REGISTER_EVENTFD_ASYNC = 7, 538 IORING_REGISTER_PROBE = 8, 539 IORING_REGISTER_PERSONALITY = 9, 540 IORING_UNREGISTER_PERSONALITY = 10, 541 IORING_REGISTER_RESTRICTIONS = 11, 542 IORING_REGISTER_ENABLE_RINGS = 12, 543 544 /* extended with tagging */ 545 IORING_REGISTER_FILES2 = 13, 546 IORING_REGISTER_FILES_UPDATE2 = 14, 547 IORING_REGISTER_BUFFERS2 = 15, 548 IORING_REGISTER_BUFFERS_UPDATE = 16, 549 550 /* set/clear io-wq thread affinities */ 551 IORING_REGISTER_IOWQ_AFF = 17, 552 IORING_UNREGISTER_IOWQ_AFF = 18, 553 554 /* set/get max number of io-wq workers */ 555 IORING_REGISTER_IOWQ_MAX_WORKERS = 19, 556 557 /* register/unregister io_uring fd with the ring */ 558 IORING_REGISTER_RING_FDS = 20, 559 IORING_UNREGISTER_RING_FDS = 21, 560 561 /* register ring based provide buffer group */ 562 IORING_REGISTER_PBUF_RING = 22, 563 IORING_UNREGISTER_PBUF_RING = 23, 564 565 /* sync cancelation API */ 566 IORING_REGISTER_SYNC_CANCEL = 24, 567 568 /* register a range of fixed file slots for automatic slot allocation */ 569 IORING_REGISTER_FILE_ALLOC_RANGE = 25, 570 571 /* return status information for a buffer group */ 572 IORING_REGISTER_PBUF_STATUS = 26, 573 574 /* set/clear busy poll settings */ 575 IORING_REGISTER_NAPI = 27, 576 IORING_UNREGISTER_NAPI = 28, 577 578 /* this goes last */ 579 IORING_REGISTER_LAST, 580 581 /* flag added to the opcode to use a registered ring fd */ 582 IORING_REGISTER_USE_REGISTERED_RING = 1U << 31 583 }; 584 585 /* io-wq worker categories */ 586 enum { 587 IO_WQ_BOUND, 588 IO_WQ_UNBOUND, 589 }; 590 591 /* deprecated, see struct io_uring_rsrc_update */ 592 struct io_uring_files_update { 593 __u32 offset; 594 __u32 resv; 595 __aligned_u64 /* __s32 * */ fds; 596 }; 597 598 /* 599 * Register a fully sparse file space, rather than pass in an array of all 600 * -1 file descriptors. 601 */ 602 #define IORING_RSRC_REGISTER_SPARSE (1U << 0) 603 604 struct io_uring_rsrc_register { 605 __u32 nr; 606 __u32 flags; 607 __u64 resv2; 608 __aligned_u64 data; 609 __aligned_u64 tags; 610 }; 611 612 struct io_uring_rsrc_update { 613 __u32 offset; 614 __u32 resv; 615 __aligned_u64 data; 616 }; 617 618 struct io_uring_rsrc_update2 { 619 __u32 offset; 620 __u32 resv; 621 __aligned_u64 data; 622 __aligned_u64 tags; 623 __u32 nr; 624 __u32 resv2; 625 }; 626 627 /* Skip updating fd indexes set to this value in the fd table */ 628 #define IORING_REGISTER_FILES_SKIP (-2) 629 630 #define IO_URING_OP_SUPPORTED (1U << 0) 631 632 struct io_uring_probe_op { 633 __u8 op; 634 __u8 resv; 635 __u16 flags; /* IO_URING_OP_* flags */ 636 __u32 resv2; 637 }; 638 639 struct io_uring_probe { 640 __u8 last_op; /* last opcode supported */ 641 __u8 ops_len; /* length of ops[] array below */ 642 __u16 resv; 643 __u32 resv2[3]; 644 struct io_uring_probe_op ops[]; 645 }; 646 647 struct io_uring_restriction { 648 __u16 opcode; 649 union { 650 __u8 register_op; /* IORING_RESTRICTION_REGISTER_OP */ 651 __u8 sqe_op; /* IORING_RESTRICTION_SQE_OP */ 652 __u8 sqe_flags; /* IORING_RESTRICTION_SQE_FLAGS_* */ 653 }; 654 __u8 resv; 655 __u32 resv2[3]; 656 }; 657 658 struct io_uring_buf { 659 __u64 addr; 660 __u32 len; 661 __u16 bid; 662 __u16 resv; 663 }; 664 665 struct io_uring_buf_ring { 666 union { 667 /* 668 * To avoid spilling into more pages than we need to, the 669 * ring tail is overlaid with the io_uring_buf->resv field. 670 */ 671 struct { 672 __u64 resv1; 673 __u32 resv2; 674 __u16 resv3; 675 __u16 tail; 676 }; 677 __DECLARE_FLEX_ARRAY(struct io_uring_buf, bufs); 678 }; 679 }; 680 681 /* 682 * Flags for IORING_REGISTER_PBUF_RING. 683 * 684 * IOU_PBUF_RING_MMAP: If set, kernel will allocate the memory for the ring. 685 * The application must not set a ring_addr in struct 686 * io_uring_buf_reg, instead it must subsequently call 687 * mmap(2) with the offset set as: 688 * IORING_OFF_PBUF_RING | (bgid << IORING_OFF_PBUF_SHIFT) 689 * to get a virtual mapping for the ring. 690 */ 691 enum { 692 IOU_PBUF_RING_MMAP = 1, 693 }; 694 695 /* argument for IORING_(UN)REGISTER_PBUF_RING */ 696 struct io_uring_buf_reg { 697 __u64 ring_addr; 698 __u32 ring_entries; 699 __u16 bgid; 700 __u16 flags; 701 __u64 resv[3]; 702 }; 703 704 /* argument for IORING_REGISTER_PBUF_STATUS */ 705 struct io_uring_buf_status { 706 __u32 buf_group; /* input */ 707 __u32 head; /* output */ 708 __u32 resv[8]; 709 }; 710 711 /* argument for IORING_(UN)REGISTER_NAPI */ 712 struct io_uring_napi { 713 __u32 busy_poll_to; 714 __u8 prefer_busy_poll; 715 __u8 pad[3]; 716 __u64 resv; 717 }; 718 719 /* 720 * io_uring_restriction->opcode values 721 */ 722 enum { 723 /* Allow an io_uring_register(2) opcode */ 724 IORING_RESTRICTION_REGISTER_OP = 0, 725 726 /* Allow an sqe opcode */ 727 IORING_RESTRICTION_SQE_OP = 1, 728 729 /* Allow sqe flags */ 730 IORING_RESTRICTION_SQE_FLAGS_ALLOWED = 2, 731 732 /* Require sqe flags (these flags must be set on each submission) */ 733 IORING_RESTRICTION_SQE_FLAGS_REQUIRED = 3, 734 735 IORING_RESTRICTION_LAST 736 }; 737 738 struct io_uring_getevents_arg { 739 __u64 sigmask; 740 __u32 sigmask_sz; 741 __u32 pad; 742 __u64 ts; 743 }; 744 745 /* 746 * Argument for IORING_REGISTER_SYNC_CANCEL 747 */ 748 struct io_uring_sync_cancel_reg { 749 __u64 addr; 750 __s32 fd; 751 __u32 flags; 752 struct __kernel_timespec timeout; 753 __u8 opcode; 754 __u8 pad[7]; 755 __u64 pad2[3]; 756 }; 757 758 /* 759 * Argument for IORING_REGISTER_FILE_ALLOC_RANGE 760 * The range is specified as [off, off + len) 761 */ 762 struct io_uring_file_index_range { 763 __u32 off; 764 __u32 len; 765 __u64 resv; 766 }; 767 768 struct io_uring_recvmsg_out { 769 __u32 namelen; 770 __u32 controllen; 771 __u32 payloadlen; 772 __u32 flags; 773 }; 774 775 /* 776 * Argument for IORING_OP_URING_CMD when file is a socket 777 */ 778 enum { 779 SOCKET_URING_OP_SIOCINQ = 0, 780 SOCKET_URING_OP_SIOCOUTQ, 781 SOCKET_URING_OP_GETSOCKOPT, 782 SOCKET_URING_OP_SETSOCKOPT, 783 }; 784 785 #ifdef __cplusplus 786 } 787 #endif 788 789 #endif 790