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 io_uring_sqe_flags_bit { 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 the memory for the rings 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 top 8bits aren't available for userspace 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 #define IORING_URING_CMD_MASK IORING_URING_CMD_FIXED 274 275 276 /* 277 * sqe->fsync_flags 278 */ 279 #define IORING_FSYNC_DATASYNC (1U << 0) 280 281 /* 282 * sqe->timeout_flags 283 */ 284 #define IORING_TIMEOUT_ABS (1U << 0) 285 #define IORING_TIMEOUT_UPDATE (1U << 1) 286 #define IORING_TIMEOUT_BOOTTIME (1U << 2) 287 #define IORING_TIMEOUT_REALTIME (1U << 3) 288 #define IORING_LINK_TIMEOUT_UPDATE (1U << 4) 289 #define IORING_TIMEOUT_ETIME_SUCCESS (1U << 5) 290 #define IORING_TIMEOUT_MULTISHOT (1U << 6) 291 #define IORING_TIMEOUT_CLOCK_MASK (IORING_TIMEOUT_BOOTTIME | IORING_TIMEOUT_REALTIME) 292 #define IORING_TIMEOUT_UPDATE_MASK (IORING_TIMEOUT_UPDATE | IORING_LINK_TIMEOUT_UPDATE) 293 /* 294 * sqe->splice_flags 295 * extends splice(2) flags 296 */ 297 #define SPLICE_F_FD_IN_FIXED (1U << 31) /* the last bit of __u32 */ 298 299 /* 300 * POLL_ADD flags. Note that since sqe->poll_events is the flag space, the 301 * command flags for POLL_ADD are stored in sqe->len. 302 * 303 * IORING_POLL_ADD_MULTI Multishot poll. Sets IORING_CQE_F_MORE if 304 * the poll handler will continue to report 305 * CQEs on behalf of the same SQE. 306 * 307 * IORING_POLL_UPDATE Update existing poll request, matching 308 * sqe->addr as the old user_data field. 309 * 310 * IORING_POLL_LEVEL Level triggered poll. 311 */ 312 #define IORING_POLL_ADD_MULTI (1U << 0) 313 #define IORING_POLL_UPDATE_EVENTS (1U << 1) 314 #define IORING_POLL_UPDATE_USER_DATA (1U << 2) 315 #define IORING_POLL_ADD_LEVEL (1U << 3) 316 317 /* 318 * ASYNC_CANCEL flags. 319 * 320 * IORING_ASYNC_CANCEL_ALL Cancel all requests that match the given key 321 * IORING_ASYNC_CANCEL_FD Key off 'fd' for cancelation rather than the 322 * request 'user_data' 323 * IORING_ASYNC_CANCEL_ANY Match any request 324 * IORING_ASYNC_CANCEL_FD_FIXED 'fd' passed in is a fixed descriptor 325 * IORING_ASYNC_CANCEL_USERDATA Match on user_data, default for no other key 326 * IORING_ASYNC_CANCEL_OP Match request based on opcode 327 */ 328 #define IORING_ASYNC_CANCEL_ALL (1U << 0) 329 #define IORING_ASYNC_CANCEL_FD (1U << 1) 330 #define IORING_ASYNC_CANCEL_ANY (1U << 2) 331 #define IORING_ASYNC_CANCEL_FD_FIXED (1U << 3) 332 #define IORING_ASYNC_CANCEL_USERDATA (1U << 4) 333 #define IORING_ASYNC_CANCEL_OP (1U << 5) 334 335 /* 336 * send/sendmsg and recv/recvmsg flags (sqe->ioprio) 337 * 338 * IORING_RECVSEND_POLL_FIRST If set, instead of first attempting to send 339 * or receive and arm poll if that yields an 340 * -EAGAIN result, arm poll upfront and skip 341 * the initial transfer attempt. 342 * 343 * IORING_RECV_MULTISHOT Multishot recv. Sets IORING_CQE_F_MORE if 344 * the handler will continue to report 345 * CQEs on behalf of the same SQE. 346 * 347 * IORING_RECVSEND_FIXED_BUF Use registered buffers, the index is stored in 348 * the buf_index field. 349 * 350 * IORING_SEND_ZC_REPORT_USAGE 351 * If set, SEND[MSG]_ZC should report 352 * the zerocopy usage in cqe.res 353 * for the IORING_CQE_F_NOTIF cqe. 354 * 0 is reported if zerocopy was actually possible. 355 * IORING_NOTIF_USAGE_ZC_COPIED if data was copied 356 * (at least partially). 357 * 358 * IORING_RECVSEND_BUNDLE Used with IOSQE_BUFFER_SELECT. If set, send or 359 * recv will grab as many buffers from the buffer 360 * group ID given and send them all. The completion 361 * result will be the number of buffers send, with 362 * the starting buffer ID in cqe->flags as per 363 * usual for provided buffer usage. The buffers 364 * will be contigious from the starting buffer ID. 365 */ 366 #define IORING_RECVSEND_POLL_FIRST (1U << 0) 367 #define IORING_RECV_MULTISHOT (1U << 1) 368 #define IORING_RECVSEND_FIXED_BUF (1U << 2) 369 #define IORING_SEND_ZC_REPORT_USAGE (1U << 3) 370 #define IORING_RECVSEND_BUNDLE (1U << 4) 371 372 /* 373 * cqe.res for IORING_CQE_F_NOTIF if 374 * IORING_SEND_ZC_REPORT_USAGE was requested 375 * 376 * It should be treated as a flag, all other 377 * bits of cqe.res should be treated as reserved! 378 */ 379 #define IORING_NOTIF_USAGE_ZC_COPIED (1U << 31) 380 381 /* 382 * accept flags stored in sqe->ioprio 383 */ 384 #define IORING_ACCEPT_MULTISHOT (1U << 0) 385 #define IORING_ACCEPT_DONTWAIT (1U << 1) 386 #define IORING_ACCEPT_POLL_FIRST (1U << 2) 387 388 /* 389 * IORING_OP_MSG_RING command types, stored in sqe->addr 390 */ 391 enum io_uring_msg_ring_flags { 392 IORING_MSG_DATA, /* pass sqe->len as 'res' and off as user_data */ 393 IORING_MSG_SEND_FD, /* send a registered fd to another ring */ 394 }; 395 396 /* 397 * IORING_OP_MSG_RING flags (sqe->msg_ring_flags) 398 * 399 * IORING_MSG_RING_CQE_SKIP Don't post a CQE to the target ring. Not 400 * applicable for IORING_MSG_DATA, obviously. 401 */ 402 #define IORING_MSG_RING_CQE_SKIP (1U << 0) 403 /* Pass through the flags from sqe->file_index to cqe->flags */ 404 #define IORING_MSG_RING_FLAGS_PASS (1U << 1) 405 406 /* 407 * IORING_OP_FIXED_FD_INSTALL flags (sqe->install_fd_flags) 408 * 409 * IORING_FIXED_FD_NO_CLOEXEC Don't mark the fd as O_CLOEXEC 410 */ 411 #define IORING_FIXED_FD_NO_CLOEXEC (1U << 0) 412 413 /* 414 * IORING_OP_NOP flags (sqe->nop_flags) 415 * 416 * IORING_NOP_INJECT_RESULT Inject result from sqe->result 417 */ 418 #define IORING_NOP_INJECT_RESULT (1U << 0) 419 #define IORING_NOP_FILE (1U << 1) 420 #define IORING_NOP_FIXED_FILE (1U << 2) 421 #define IORING_NOP_FIXED_BUFFER (1U << 3) 422 423 /* 424 * IO completion data structure (Completion Queue Entry) 425 */ 426 struct io_uring_cqe { 427 __u64 user_data; /* sqe->user_data value passed back */ 428 __s32 res; /* result code for this event */ 429 __u32 flags; 430 431 /* 432 * If the ring is initialized with IORING_SETUP_CQE32, then this field 433 * contains 16-bytes of padding, doubling the size of the CQE. 434 */ 435 __u64 big_cqe[]; 436 }; 437 438 /* 439 * cqe->flags 440 * 441 * IORING_CQE_F_BUFFER If set, the upper 16 bits are the buffer ID 442 * IORING_CQE_F_MORE If set, parent SQE will generate more CQE entries 443 * IORING_CQE_F_SOCK_NONEMPTY If set, more data to read after socket recv 444 * IORING_CQE_F_NOTIF Set for notification CQEs. Can be used to distinct 445 * them from sends. 446 * IORING_CQE_F_BUF_MORE If set, the buffer ID set in the completion will get 447 * more completions. In other words, the buffer is being 448 * partially consumed, and will be used by the kernel for 449 * more completions. This is only set for buffers used via 450 * the incremental buffer consumption, as provided by 451 * a ring buffer setup with IOU_PBUF_RING_INC. For any 452 * other provided buffer type, all completions with a 453 * buffer passed back is automatically returned to the 454 * application. 455 */ 456 #define IORING_CQE_F_BUFFER (1U << 0) 457 #define IORING_CQE_F_MORE (1U << 1) 458 #define IORING_CQE_F_SOCK_NONEMPTY (1U << 2) 459 #define IORING_CQE_F_NOTIF (1U << 3) 460 #define IORING_CQE_F_BUF_MORE (1U << 4) 461 462 #define IORING_CQE_BUFFER_SHIFT 16 463 464 /* 465 * Magic offsets for the application to mmap the data it needs 466 */ 467 #define IORING_OFF_SQ_RING 0ULL 468 #define IORING_OFF_CQ_RING 0x8000000ULL 469 #define IORING_OFF_SQES 0x10000000ULL 470 #define IORING_OFF_PBUF_RING 0x80000000ULL 471 #define IORING_OFF_PBUF_SHIFT 16 472 #define IORING_OFF_MMAP_MASK 0xf8000000ULL 473 474 /* 475 * Filled with the offset for mmap(2) 476 */ 477 struct io_sqring_offsets { 478 __u32 head; 479 __u32 tail; 480 __u32 ring_mask; 481 __u32 ring_entries; 482 __u32 flags; 483 __u32 dropped; 484 __u32 array; 485 __u32 resv1; 486 __u64 user_addr; 487 }; 488 489 /* 490 * sq_ring->flags 491 */ 492 #define IORING_SQ_NEED_WAKEUP (1U << 0) /* needs io_uring_enter wakeup */ 493 #define IORING_SQ_CQ_OVERFLOW (1U << 1) /* CQ ring is overflown */ 494 #define IORING_SQ_TASKRUN (1U << 2) /* task should enter the kernel */ 495 496 struct io_cqring_offsets { 497 __u32 head; 498 __u32 tail; 499 __u32 ring_mask; 500 __u32 ring_entries; 501 __u32 overflow; 502 __u32 cqes; 503 __u32 flags; 504 __u32 resv1; 505 __u64 user_addr; 506 }; 507 508 /* 509 * cq_ring->flags 510 */ 511 512 /* disable eventfd notifications */ 513 #define IORING_CQ_EVENTFD_DISABLED (1U << 0) 514 515 /* 516 * io_uring_enter(2) flags 517 */ 518 #define IORING_ENTER_GETEVENTS (1U << 0) 519 #define IORING_ENTER_SQ_WAKEUP (1U << 1) 520 #define IORING_ENTER_SQ_WAIT (1U << 2) 521 #define IORING_ENTER_EXT_ARG (1U << 3) 522 #define IORING_ENTER_REGISTERED_RING (1U << 4) 523 #define IORING_ENTER_ABS_TIMER (1U << 5) 524 #define IORING_ENTER_EXT_ARG_REG (1U << 6) 525 526 /* 527 * Passed in for io_uring_setup(2). Copied back with updated info on success 528 */ 529 struct io_uring_params { 530 __u32 sq_entries; 531 __u32 cq_entries; 532 __u32 flags; 533 __u32 sq_thread_cpu; 534 __u32 sq_thread_idle; 535 __u32 features; 536 __u32 wq_fd; 537 __u32 resv[3]; 538 struct io_sqring_offsets sq_off; 539 struct io_cqring_offsets cq_off; 540 }; 541 542 /* 543 * io_uring_params->features flags 544 */ 545 #define IORING_FEAT_SINGLE_MMAP (1U << 0) 546 #define IORING_FEAT_NODROP (1U << 1) 547 #define IORING_FEAT_SUBMIT_STABLE (1U << 2) 548 #define IORING_FEAT_RW_CUR_POS (1U << 3) 549 #define IORING_FEAT_CUR_PERSONALITY (1U << 4) 550 #define IORING_FEAT_FAST_POLL (1U << 5) 551 #define IORING_FEAT_POLL_32BITS (1U << 6) 552 #define IORING_FEAT_SQPOLL_NONFIXED (1U << 7) 553 #define IORING_FEAT_EXT_ARG (1U << 8) 554 #define IORING_FEAT_NATIVE_WORKERS (1U << 9) 555 #define IORING_FEAT_RSRC_TAGS (1U << 10) 556 #define IORING_FEAT_CQE_SKIP (1U << 11) 557 #define IORING_FEAT_LINKED_FILE (1U << 12) 558 #define IORING_FEAT_REG_REG_RING (1U << 13) 559 #define IORING_FEAT_RECVSEND_BUNDLE (1U << 14) 560 #define IORING_FEAT_MIN_TIMEOUT (1U << 15) 561 562 /* 563 * io_uring_register(2) opcodes and arguments 564 */ 565 enum io_uring_register_op { 566 IORING_REGISTER_BUFFERS = 0, 567 IORING_UNREGISTER_BUFFERS = 1, 568 IORING_REGISTER_FILES = 2, 569 IORING_UNREGISTER_FILES = 3, 570 IORING_REGISTER_EVENTFD = 4, 571 IORING_UNREGISTER_EVENTFD = 5, 572 IORING_REGISTER_FILES_UPDATE = 6, 573 IORING_REGISTER_EVENTFD_ASYNC = 7, 574 IORING_REGISTER_PROBE = 8, 575 IORING_REGISTER_PERSONALITY = 9, 576 IORING_UNREGISTER_PERSONALITY = 10, 577 IORING_REGISTER_RESTRICTIONS = 11, 578 IORING_REGISTER_ENABLE_RINGS = 12, 579 580 /* extended with tagging */ 581 IORING_REGISTER_FILES2 = 13, 582 IORING_REGISTER_FILES_UPDATE2 = 14, 583 IORING_REGISTER_BUFFERS2 = 15, 584 IORING_REGISTER_BUFFERS_UPDATE = 16, 585 586 /* set/clear io-wq thread affinities */ 587 IORING_REGISTER_IOWQ_AFF = 17, 588 IORING_UNREGISTER_IOWQ_AFF = 18, 589 590 /* set/get max number of io-wq workers */ 591 IORING_REGISTER_IOWQ_MAX_WORKERS = 19, 592 593 /* register/unregister io_uring fd with the ring */ 594 IORING_REGISTER_RING_FDS = 20, 595 IORING_UNREGISTER_RING_FDS = 21, 596 597 /* register ring based provide buffer group */ 598 IORING_REGISTER_PBUF_RING = 22, 599 IORING_UNREGISTER_PBUF_RING = 23, 600 601 /* sync cancelation API */ 602 IORING_REGISTER_SYNC_CANCEL = 24, 603 604 /* register a range of fixed file slots for automatic slot allocation */ 605 IORING_REGISTER_FILE_ALLOC_RANGE = 25, 606 607 /* return status information for a buffer group */ 608 IORING_REGISTER_PBUF_STATUS = 26, 609 610 /* set/clear busy poll settings */ 611 IORING_REGISTER_NAPI = 27, 612 IORING_UNREGISTER_NAPI = 28, 613 614 IORING_REGISTER_CLOCK = 29, 615 616 /* clone registered buffers from source ring to current ring */ 617 IORING_REGISTER_CLONE_BUFFERS = 30, 618 619 /* send MSG_RING without having a ring */ 620 IORING_REGISTER_SEND_MSG_RING = 31, 621 622 /* 32 reserved for zc rx */ 623 624 /* resize CQ ring */ 625 IORING_REGISTER_RESIZE_RINGS = 33, 626 627 /* register fixed io_uring_reg_wait arguments */ 628 IORING_REGISTER_CQWAIT_REG = 34, 629 630 /* this goes last */ 631 IORING_REGISTER_LAST, 632 633 /* flag added to the opcode to use a registered ring fd */ 634 IORING_REGISTER_USE_REGISTERED_RING = 1U << 31 635 }; 636 637 /* io-wq worker categories */ 638 enum io_wq_type { 639 IO_WQ_BOUND, 640 IO_WQ_UNBOUND, 641 }; 642 643 /* deprecated, see struct io_uring_rsrc_update */ 644 struct io_uring_files_update { 645 __u32 offset; 646 __u32 resv; 647 __aligned_u64 /* __s32 * */ fds; 648 }; 649 650 /* 651 * Register a fully sparse file space, rather than pass in an array of all 652 * -1 file descriptors. 653 */ 654 #define IORING_RSRC_REGISTER_SPARSE (1U << 0) 655 656 struct io_uring_rsrc_register { 657 __u32 nr; 658 __u32 flags; 659 __u64 resv2; 660 __aligned_u64 data; 661 __aligned_u64 tags; 662 }; 663 664 struct io_uring_rsrc_update { 665 __u32 offset; 666 __u32 resv; 667 __aligned_u64 data; 668 }; 669 670 struct io_uring_rsrc_update2 { 671 __u32 offset; 672 __u32 resv; 673 __aligned_u64 data; 674 __aligned_u64 tags; 675 __u32 nr; 676 __u32 resv2; 677 }; 678 679 /* Skip updating fd indexes set to this value in the fd table */ 680 #define IORING_REGISTER_FILES_SKIP (-2) 681 682 #define IO_URING_OP_SUPPORTED (1U << 0) 683 684 struct io_uring_probe_op { 685 __u8 op; 686 __u8 resv; 687 __u16 flags; /* IO_URING_OP_* flags */ 688 __u32 resv2; 689 }; 690 691 struct io_uring_probe { 692 __u8 last_op; /* last opcode supported */ 693 __u8 ops_len; /* length of ops[] array below */ 694 __u16 resv; 695 __u32 resv2[3]; 696 struct io_uring_probe_op ops[]; 697 }; 698 699 struct io_uring_restriction { 700 __u16 opcode; 701 union { 702 __u8 register_op; /* IORING_RESTRICTION_REGISTER_OP */ 703 __u8 sqe_op; /* IORING_RESTRICTION_SQE_OP */ 704 __u8 sqe_flags; /* IORING_RESTRICTION_SQE_FLAGS_* */ 705 }; 706 __u8 resv; 707 __u32 resv2[3]; 708 }; 709 710 struct io_uring_clock_register { 711 __u32 clockid; 712 __u32 __resv[3]; 713 }; 714 715 enum { 716 IORING_REGISTER_SRC_REGISTERED = 1, 717 }; 718 719 struct io_uring_clone_buffers { 720 __u32 src_fd; 721 __u32 flags; 722 __u32 pad[6]; 723 }; 724 725 struct io_uring_buf { 726 __u64 addr; 727 __u32 len; 728 __u16 bid; 729 __u16 resv; 730 }; 731 732 struct io_uring_buf_ring { 733 union { 734 /* 735 * To avoid spilling into more pages than we need to, the 736 * ring tail is overlaid with the io_uring_buf->resv field. 737 */ 738 struct { 739 __u64 resv1; 740 __u32 resv2; 741 __u16 resv3; 742 __u16 tail; 743 }; 744 __DECLARE_FLEX_ARRAY(struct io_uring_buf, bufs); 745 }; 746 }; 747 748 /* 749 * Flags for IORING_REGISTER_PBUF_RING. 750 * 751 * IOU_PBUF_RING_MMAP: If set, kernel will allocate the memory for the ring. 752 * The application must not set a ring_addr in struct 753 * io_uring_buf_reg, instead it must subsequently call 754 * mmap(2) with the offset set as: 755 * IORING_OFF_PBUF_RING | (bgid << IORING_OFF_PBUF_SHIFT) 756 * to get a virtual mapping for the ring. 757 * IOU_PBUF_RING_INC: If set, buffers consumed from this buffer ring can be 758 * consumed incrementally. Normally one (or more) buffers 759 * are fully consumed. With incremental consumptions, it's 760 * feasible to register big ranges of buffers, and each 761 * use of it will consume only as much as it needs. This 762 * requires that both the kernel and application keep 763 * track of where the current read/recv index is at. 764 */ 765 enum io_uring_register_pbuf_ring_flags { 766 IOU_PBUF_RING_MMAP = 1, 767 IOU_PBUF_RING_INC = 2, 768 }; 769 770 /* argument for IORING_(UN)REGISTER_PBUF_RING */ 771 struct io_uring_buf_reg { 772 __u64 ring_addr; 773 __u32 ring_entries; 774 __u16 bgid; 775 __u16 flags; 776 __u64 resv[3]; 777 }; 778 779 /* argument for IORING_REGISTER_PBUF_STATUS */ 780 struct io_uring_buf_status { 781 __u32 buf_group; /* input */ 782 __u32 head; /* output */ 783 __u32 resv[8]; 784 }; 785 786 /* argument for IORING_(UN)REGISTER_NAPI */ 787 struct io_uring_napi { 788 __u32 busy_poll_to; 789 __u8 prefer_busy_poll; 790 __u8 pad[3]; 791 __u64 resv; 792 }; 793 794 /* 795 * io_uring_restriction->opcode values 796 */ 797 enum io_uring_register_restriction_op { 798 /* Allow an io_uring_register(2) opcode */ 799 IORING_RESTRICTION_REGISTER_OP = 0, 800 801 /* Allow an sqe opcode */ 802 IORING_RESTRICTION_SQE_OP = 1, 803 804 /* Allow sqe flags */ 805 IORING_RESTRICTION_SQE_FLAGS_ALLOWED = 2, 806 807 /* Require sqe flags (these flags must be set on each submission) */ 808 IORING_RESTRICTION_SQE_FLAGS_REQUIRED = 3, 809 810 IORING_RESTRICTION_LAST 811 }; 812 813 enum { 814 IORING_REG_WAIT_TS = (1U << 0), 815 }; 816 817 /* 818 * Argument for IORING_REGISTER_CQWAIT_REG, registering a region of 819 * struct io_uring_reg_wait that can be indexed when io_uring_enter(2) is 820 * called rather than pass in a wait argument structure separately. 821 */ 822 struct io_uring_cqwait_reg_arg { 823 __u32 flags; 824 __u32 struct_size; 825 __u32 nr_entries; 826 __u32 pad; 827 __u64 user_addr; 828 __u64 pad2[3]; 829 }; 830 831 /* 832 * Argument for io_uring_enter(2) with 833 * IORING_GETEVENTS | IORING_ENTER_EXT_ARG_REG set, where the actual argument 834 * is an index into a previously registered fixed wait region described by 835 * the below structure. 836 */ 837 struct io_uring_reg_wait { 838 struct __kernel_timespec ts; 839 __u32 min_wait_usec; 840 __u32 flags; 841 __u64 sigmask; 842 __u32 sigmask_sz; 843 __u32 pad[3]; 844 __u64 pad2[2]; 845 }; 846 847 /* 848 * Argument for io_uring_enter(2) with IORING_GETEVENTS | IORING_ENTER_EXT_ARG 849 */ 850 struct io_uring_getevents_arg { 851 __u64 sigmask; 852 __u32 sigmask_sz; 853 __u32 min_wait_usec; 854 __u64 ts; 855 }; 856 857 /* 858 * Argument for IORING_REGISTER_SYNC_CANCEL 859 */ 860 struct io_uring_sync_cancel_reg { 861 __u64 addr; 862 __s32 fd; 863 __u32 flags; 864 struct __kernel_timespec timeout; 865 __u8 opcode; 866 __u8 pad[7]; 867 __u64 pad2[3]; 868 }; 869 870 /* 871 * Argument for IORING_REGISTER_FILE_ALLOC_RANGE 872 * The range is specified as [off, off + len) 873 */ 874 struct io_uring_file_index_range { 875 __u32 off; 876 __u32 len; 877 __u64 resv; 878 }; 879 880 struct io_uring_recvmsg_out { 881 __u32 namelen; 882 __u32 controllen; 883 __u32 payloadlen; 884 __u32 flags; 885 }; 886 887 /* 888 * Argument for IORING_OP_URING_CMD when file is a socket 889 */ 890 enum io_uring_socket_op { 891 SOCKET_URING_OP_SIOCINQ = 0, 892 SOCKET_URING_OP_SIOCOUTQ, 893 SOCKET_URING_OP_GETSOCKOPT, 894 SOCKET_URING_OP_SETSOCKOPT, 895 }; 896 897 #ifdef __cplusplus 898 } 899 #endif 900 901 #endif 902