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 struct { 102 __u64 attr_ptr; /* pointer to attribute information */ 103 __u64 attr_type_mask; /* bit mask of attributes */ 104 }; 105 __u64 optval; 106 /* 107 * If the ring is initialized with IORING_SETUP_SQE128, then 108 * this field is used for 80 bytes of arbitrary command data 109 */ 110 __u8 cmd[0]; 111 }; 112 }; 113 114 /* sqe->attr_type_mask flags */ 115 #define IORING_RW_ATTR_FLAG_PI (1U << 0) 116 /* PI attribute information */ 117 struct io_uring_attr_pi { 118 __u16 flags; 119 __u16 app_tag; 120 __u32 len; 121 __u64 addr; 122 __u64 seed; 123 __u64 rsvd; 124 }; 125 126 /* 127 * If sqe->file_index is set to this for opcodes that instantiate a new 128 * direct descriptor (like openat/openat2/accept), then io_uring will allocate 129 * an available direct descriptor instead of having the application pass one 130 * in. The picked direct descriptor will be returned in cqe->res, or -ENFILE 131 * if the space is full. 132 */ 133 #define IORING_FILE_INDEX_ALLOC (~0U) 134 135 enum io_uring_sqe_flags_bit { 136 IOSQE_FIXED_FILE_BIT, 137 IOSQE_IO_DRAIN_BIT, 138 IOSQE_IO_LINK_BIT, 139 IOSQE_IO_HARDLINK_BIT, 140 IOSQE_ASYNC_BIT, 141 IOSQE_BUFFER_SELECT_BIT, 142 IOSQE_CQE_SKIP_SUCCESS_BIT, 143 }; 144 145 /* 146 * sqe->flags 147 */ 148 /* use fixed fileset */ 149 #define IOSQE_FIXED_FILE (1U << IOSQE_FIXED_FILE_BIT) 150 /* issue after inflight IO */ 151 #define IOSQE_IO_DRAIN (1U << IOSQE_IO_DRAIN_BIT) 152 /* links next sqe */ 153 #define IOSQE_IO_LINK (1U << IOSQE_IO_LINK_BIT) 154 /* like LINK, but stronger */ 155 #define IOSQE_IO_HARDLINK (1U << IOSQE_IO_HARDLINK_BIT) 156 /* always go async */ 157 #define IOSQE_ASYNC (1U << IOSQE_ASYNC_BIT) 158 /* select buffer from sqe->buf_group */ 159 #define IOSQE_BUFFER_SELECT (1U << IOSQE_BUFFER_SELECT_BIT) 160 /* don't post CQE if request succeeded */ 161 #define IOSQE_CQE_SKIP_SUCCESS (1U << IOSQE_CQE_SKIP_SUCCESS_BIT) 162 163 /* 164 * io_uring_setup() flags 165 */ 166 #define IORING_SETUP_IOPOLL (1U << 0) /* io_context is polled */ 167 #define IORING_SETUP_SQPOLL (1U << 1) /* SQ poll thread */ 168 #define IORING_SETUP_SQ_AFF (1U << 2) /* sq_thread_cpu is valid */ 169 #define IORING_SETUP_CQSIZE (1U << 3) /* app defines CQ size */ 170 #define IORING_SETUP_CLAMP (1U << 4) /* clamp SQ/CQ ring sizes */ 171 #define IORING_SETUP_ATTACH_WQ (1U << 5) /* attach to existing wq */ 172 #define IORING_SETUP_R_DISABLED (1U << 6) /* start with ring disabled */ 173 #define IORING_SETUP_SUBMIT_ALL (1U << 7) /* continue submit on error */ 174 /* 175 * Cooperative task running. When requests complete, they often require 176 * forcing the submitter to transition to the kernel to complete. If this 177 * flag is set, work will be done when the task transitions anyway, rather 178 * than force an inter-processor interrupt reschedule. This avoids interrupting 179 * a task running in userspace, and saves an IPI. 180 */ 181 #define IORING_SETUP_COOP_TASKRUN (1U << 8) 182 /* 183 * If COOP_TASKRUN is set, get notified if task work is available for 184 * running and a kernel transition would be needed to run it. This sets 185 * IORING_SQ_TASKRUN in the sq ring flags. Not valid with COOP_TASKRUN. 186 */ 187 #define IORING_SETUP_TASKRUN_FLAG (1U << 9) 188 #define IORING_SETUP_SQE128 (1U << 10) /* SQEs are 128 byte */ 189 #define IORING_SETUP_CQE32 (1U << 11) /* CQEs are 32 byte */ 190 /* 191 * Only one task is allowed to submit requests 192 */ 193 #define IORING_SETUP_SINGLE_ISSUER (1U << 12) 194 195 /* 196 * Defer running task work to get events. 197 * Rather than running bits of task work whenever the task transitions 198 * try to do it just before it is needed. 199 */ 200 #define IORING_SETUP_DEFER_TASKRUN (1U << 13) 201 202 /* 203 * Application provides the memory for the rings 204 */ 205 #define IORING_SETUP_NO_MMAP (1U << 14) 206 207 /* 208 * Register the ring fd in itself for use with 209 * IORING_REGISTER_USE_REGISTERED_RING; return a registered fd index rather 210 * than an fd. 211 */ 212 #define IORING_SETUP_REGISTERED_FD_ONLY (1U << 15) 213 214 /* 215 * Removes indirection through the SQ index array. 216 */ 217 #define IORING_SETUP_NO_SQARRAY (1U << 16) 218 219 /* Use hybrid poll in iopoll process */ 220 #define IORING_SETUP_HYBRID_IOPOLL (1U << 17) 221 222 enum io_uring_op { 223 IORING_OP_NOP, 224 IORING_OP_READV, 225 IORING_OP_WRITEV, 226 IORING_OP_FSYNC, 227 IORING_OP_READ_FIXED, 228 IORING_OP_WRITE_FIXED, 229 IORING_OP_POLL_ADD, 230 IORING_OP_POLL_REMOVE, 231 IORING_OP_SYNC_FILE_RANGE, 232 IORING_OP_SENDMSG, 233 IORING_OP_RECVMSG, 234 IORING_OP_TIMEOUT, 235 IORING_OP_TIMEOUT_REMOVE, 236 IORING_OP_ACCEPT, 237 IORING_OP_ASYNC_CANCEL, 238 IORING_OP_LINK_TIMEOUT, 239 IORING_OP_CONNECT, 240 IORING_OP_FALLOCATE, 241 IORING_OP_OPENAT, 242 IORING_OP_CLOSE, 243 IORING_OP_FILES_UPDATE, 244 IORING_OP_STATX, 245 IORING_OP_READ, 246 IORING_OP_WRITE, 247 IORING_OP_FADVISE, 248 IORING_OP_MADVISE, 249 IORING_OP_SEND, 250 IORING_OP_RECV, 251 IORING_OP_OPENAT2, 252 IORING_OP_EPOLL_CTL, 253 IORING_OP_SPLICE, 254 IORING_OP_PROVIDE_BUFFERS, 255 IORING_OP_REMOVE_BUFFERS, 256 IORING_OP_TEE, 257 IORING_OP_SHUTDOWN, 258 IORING_OP_RENAMEAT, 259 IORING_OP_UNLINKAT, 260 IORING_OP_MKDIRAT, 261 IORING_OP_SYMLINKAT, 262 IORING_OP_LINKAT, 263 IORING_OP_MSG_RING, 264 IORING_OP_FSETXATTR, 265 IORING_OP_SETXATTR, 266 IORING_OP_FGETXATTR, 267 IORING_OP_GETXATTR, 268 IORING_OP_SOCKET, 269 IORING_OP_URING_CMD, 270 IORING_OP_SEND_ZC, 271 IORING_OP_SENDMSG_ZC, 272 IORING_OP_READ_MULTISHOT, 273 IORING_OP_WAITID, 274 IORING_OP_FUTEX_WAIT, 275 IORING_OP_FUTEX_WAKE, 276 IORING_OP_FUTEX_WAITV, 277 IORING_OP_FIXED_FD_INSTALL, 278 IORING_OP_FTRUNCATE, 279 IORING_OP_BIND, 280 IORING_OP_LISTEN, 281 282 /* this goes last, obviously */ 283 IORING_OP_LAST, 284 }; 285 286 /* 287 * sqe->uring_cmd_flags top 8bits aren't available for userspace 288 * IORING_URING_CMD_FIXED use registered buffer; pass this flag 289 * along with setting sqe->buf_index. 290 */ 291 #define IORING_URING_CMD_FIXED (1U << 0) 292 #define IORING_URING_CMD_MASK IORING_URING_CMD_FIXED 293 294 295 /* 296 * sqe->fsync_flags 297 */ 298 #define IORING_FSYNC_DATASYNC (1U << 0) 299 300 /* 301 * sqe->timeout_flags 302 */ 303 #define IORING_TIMEOUT_ABS (1U << 0) 304 #define IORING_TIMEOUT_UPDATE (1U << 1) 305 #define IORING_TIMEOUT_BOOTTIME (1U << 2) 306 #define IORING_TIMEOUT_REALTIME (1U << 3) 307 #define IORING_LINK_TIMEOUT_UPDATE (1U << 4) 308 #define IORING_TIMEOUT_ETIME_SUCCESS (1U << 5) 309 #define IORING_TIMEOUT_MULTISHOT (1U << 6) 310 #define IORING_TIMEOUT_CLOCK_MASK (IORING_TIMEOUT_BOOTTIME | IORING_TIMEOUT_REALTIME) 311 #define IORING_TIMEOUT_UPDATE_MASK (IORING_TIMEOUT_UPDATE | IORING_LINK_TIMEOUT_UPDATE) 312 /* 313 * sqe->splice_flags 314 * extends splice(2) flags 315 */ 316 #define SPLICE_F_FD_IN_FIXED (1U << 31) /* the last bit of __u32 */ 317 318 /* 319 * POLL_ADD flags. Note that since sqe->poll_events is the flag space, the 320 * command flags for POLL_ADD are stored in sqe->len. 321 * 322 * IORING_POLL_ADD_MULTI Multishot poll. Sets IORING_CQE_F_MORE if 323 * the poll handler will continue to report 324 * CQEs on behalf of the same SQE. 325 * 326 * IORING_POLL_UPDATE Update existing poll request, matching 327 * sqe->addr as the old user_data field. 328 * 329 * IORING_POLL_LEVEL Level triggered poll. 330 */ 331 #define IORING_POLL_ADD_MULTI (1U << 0) 332 #define IORING_POLL_UPDATE_EVENTS (1U << 1) 333 #define IORING_POLL_UPDATE_USER_DATA (1U << 2) 334 #define IORING_POLL_ADD_LEVEL (1U << 3) 335 336 /* 337 * ASYNC_CANCEL flags. 338 * 339 * IORING_ASYNC_CANCEL_ALL Cancel all requests that match the given key 340 * IORING_ASYNC_CANCEL_FD Key off 'fd' for cancelation rather than the 341 * request 'user_data' 342 * IORING_ASYNC_CANCEL_ANY Match any request 343 * IORING_ASYNC_CANCEL_FD_FIXED 'fd' passed in is a fixed descriptor 344 * IORING_ASYNC_CANCEL_USERDATA Match on user_data, default for no other key 345 * IORING_ASYNC_CANCEL_OP Match request based on opcode 346 */ 347 #define IORING_ASYNC_CANCEL_ALL (1U << 0) 348 #define IORING_ASYNC_CANCEL_FD (1U << 1) 349 #define IORING_ASYNC_CANCEL_ANY (1U << 2) 350 #define IORING_ASYNC_CANCEL_FD_FIXED (1U << 3) 351 #define IORING_ASYNC_CANCEL_USERDATA (1U << 4) 352 #define IORING_ASYNC_CANCEL_OP (1U << 5) 353 354 /* 355 * send/sendmsg and recv/recvmsg flags (sqe->ioprio) 356 * 357 * IORING_RECVSEND_POLL_FIRST If set, instead of first attempting to send 358 * or receive and arm poll if that yields an 359 * -EAGAIN result, arm poll upfront and skip 360 * the initial transfer attempt. 361 * 362 * IORING_RECV_MULTISHOT Multishot recv. Sets IORING_CQE_F_MORE if 363 * the handler will continue to report 364 * CQEs on behalf of the same SQE. 365 * 366 * IORING_RECVSEND_FIXED_BUF Use registered buffers, the index is stored in 367 * the buf_index field. 368 * 369 * IORING_SEND_ZC_REPORT_USAGE 370 * If set, SEND[MSG]_ZC should report 371 * the zerocopy usage in cqe.res 372 * for the IORING_CQE_F_NOTIF cqe. 373 * 0 is reported if zerocopy was actually possible. 374 * IORING_NOTIF_USAGE_ZC_COPIED if data was copied 375 * (at least partially). 376 * 377 * IORING_RECVSEND_BUNDLE Used with IOSQE_BUFFER_SELECT. If set, send or 378 * recv will grab as many buffers from the buffer 379 * group ID given and send them all. The completion 380 * result will be the number of buffers send, with 381 * the starting buffer ID in cqe->flags as per 382 * usual for provided buffer usage. The buffers 383 * will be contiguous from the starting buffer ID. 384 */ 385 #define IORING_RECVSEND_POLL_FIRST (1U << 0) 386 #define IORING_RECV_MULTISHOT (1U << 1) 387 #define IORING_RECVSEND_FIXED_BUF (1U << 2) 388 #define IORING_SEND_ZC_REPORT_USAGE (1U << 3) 389 #define IORING_RECVSEND_BUNDLE (1U << 4) 390 391 /* 392 * cqe.res for IORING_CQE_F_NOTIF if 393 * IORING_SEND_ZC_REPORT_USAGE was requested 394 * 395 * It should be treated as a flag, all other 396 * bits of cqe.res should be treated as reserved! 397 */ 398 #define IORING_NOTIF_USAGE_ZC_COPIED (1U << 31) 399 400 /* 401 * accept flags stored in sqe->ioprio 402 */ 403 #define IORING_ACCEPT_MULTISHOT (1U << 0) 404 #define IORING_ACCEPT_DONTWAIT (1U << 1) 405 #define IORING_ACCEPT_POLL_FIRST (1U << 2) 406 407 /* 408 * IORING_OP_MSG_RING command types, stored in sqe->addr 409 */ 410 enum io_uring_msg_ring_flags { 411 IORING_MSG_DATA, /* pass sqe->len as 'res' and off as user_data */ 412 IORING_MSG_SEND_FD, /* send a registered fd to another ring */ 413 }; 414 415 /* 416 * IORING_OP_MSG_RING flags (sqe->msg_ring_flags) 417 * 418 * IORING_MSG_RING_CQE_SKIP Don't post a CQE to the target ring. Not 419 * applicable for IORING_MSG_DATA, obviously. 420 */ 421 #define IORING_MSG_RING_CQE_SKIP (1U << 0) 422 /* Pass through the flags from sqe->file_index to cqe->flags */ 423 #define IORING_MSG_RING_FLAGS_PASS (1U << 1) 424 425 /* 426 * IORING_OP_FIXED_FD_INSTALL flags (sqe->install_fd_flags) 427 * 428 * IORING_FIXED_FD_NO_CLOEXEC Don't mark the fd as O_CLOEXEC 429 */ 430 #define IORING_FIXED_FD_NO_CLOEXEC (1U << 0) 431 432 /* 433 * IORING_OP_NOP flags (sqe->nop_flags) 434 * 435 * IORING_NOP_INJECT_RESULT Inject result from sqe->result 436 */ 437 #define IORING_NOP_INJECT_RESULT (1U << 0) 438 #define IORING_NOP_FILE (1U << 1) 439 #define IORING_NOP_FIXED_FILE (1U << 2) 440 #define IORING_NOP_FIXED_BUFFER (1U << 3) 441 442 /* 443 * IO completion data structure (Completion Queue Entry) 444 */ 445 struct io_uring_cqe { 446 __u64 user_data; /* sqe->user_data value passed back */ 447 __s32 res; /* result code for this event */ 448 __u32 flags; 449 450 /* 451 * If the ring is initialized with IORING_SETUP_CQE32, then this field 452 * contains 16-bytes of padding, doubling the size of the CQE. 453 */ 454 __u64 big_cqe[]; 455 }; 456 457 /* 458 * cqe->flags 459 * 460 * IORING_CQE_F_BUFFER If set, the upper 16 bits are the buffer ID 461 * IORING_CQE_F_MORE If set, parent SQE will generate more CQE entries 462 * IORING_CQE_F_SOCK_NONEMPTY If set, more data to read after socket recv 463 * IORING_CQE_F_NOTIF Set for notification CQEs. Can be used to distinct 464 * them from sends. 465 * IORING_CQE_F_BUF_MORE If set, the buffer ID set in the completion will get 466 * more completions. In other words, the buffer is being 467 * partially consumed, and will be used by the kernel for 468 * more completions. This is only set for buffers used via 469 * the incremental buffer consumption, as provided by 470 * a ring buffer setup with IOU_PBUF_RING_INC. For any 471 * other provided buffer type, all completions with a 472 * buffer passed back is automatically returned to the 473 * application. 474 */ 475 #define IORING_CQE_F_BUFFER (1U << 0) 476 #define IORING_CQE_F_MORE (1U << 1) 477 #define IORING_CQE_F_SOCK_NONEMPTY (1U << 2) 478 #define IORING_CQE_F_NOTIF (1U << 3) 479 #define IORING_CQE_F_BUF_MORE (1U << 4) 480 481 #define IORING_CQE_BUFFER_SHIFT 16 482 483 /* 484 * Magic offsets for the application to mmap the data it needs 485 */ 486 #define IORING_OFF_SQ_RING 0ULL 487 #define IORING_OFF_CQ_RING 0x8000000ULL 488 #define IORING_OFF_SQES 0x10000000ULL 489 #define IORING_OFF_PBUF_RING 0x80000000ULL 490 #define IORING_OFF_PBUF_SHIFT 16 491 #define IORING_OFF_MMAP_MASK 0xf8000000ULL 492 493 /* 494 * Filled with the offset for mmap(2) 495 */ 496 struct io_sqring_offsets { 497 __u32 head; 498 __u32 tail; 499 __u32 ring_mask; 500 __u32 ring_entries; 501 __u32 flags; 502 __u32 dropped; 503 __u32 array; 504 __u32 resv1; 505 __u64 user_addr; 506 }; 507 508 /* 509 * sq_ring->flags 510 */ 511 #define IORING_SQ_NEED_WAKEUP (1U << 0) /* needs io_uring_enter wakeup */ 512 #define IORING_SQ_CQ_OVERFLOW (1U << 1) /* CQ ring is overflown */ 513 #define IORING_SQ_TASKRUN (1U << 2) /* task should enter the kernel */ 514 515 struct io_cqring_offsets { 516 __u32 head; 517 __u32 tail; 518 __u32 ring_mask; 519 __u32 ring_entries; 520 __u32 overflow; 521 __u32 cqes; 522 __u32 flags; 523 __u32 resv1; 524 __u64 user_addr; 525 }; 526 527 /* 528 * cq_ring->flags 529 */ 530 531 /* disable eventfd notifications */ 532 #define IORING_CQ_EVENTFD_DISABLED (1U << 0) 533 534 /* 535 * io_uring_enter(2) flags 536 */ 537 #define IORING_ENTER_GETEVENTS (1U << 0) 538 #define IORING_ENTER_SQ_WAKEUP (1U << 1) 539 #define IORING_ENTER_SQ_WAIT (1U << 2) 540 #define IORING_ENTER_EXT_ARG (1U << 3) 541 #define IORING_ENTER_REGISTERED_RING (1U << 4) 542 #define IORING_ENTER_ABS_TIMER (1U << 5) 543 #define IORING_ENTER_EXT_ARG_REG (1U << 6) 544 #define IORING_ENTER_NO_IOWAIT (1U << 7) 545 546 /* 547 * Passed in for io_uring_setup(2). Copied back with updated info on success 548 */ 549 struct io_uring_params { 550 __u32 sq_entries; 551 __u32 cq_entries; 552 __u32 flags; 553 __u32 sq_thread_cpu; 554 __u32 sq_thread_idle; 555 __u32 features; 556 __u32 wq_fd; 557 __u32 resv[3]; 558 struct io_sqring_offsets sq_off; 559 struct io_cqring_offsets cq_off; 560 }; 561 562 /* 563 * io_uring_params->features flags 564 */ 565 #define IORING_FEAT_SINGLE_MMAP (1U << 0) 566 #define IORING_FEAT_NODROP (1U << 1) 567 #define IORING_FEAT_SUBMIT_STABLE (1U << 2) 568 #define IORING_FEAT_RW_CUR_POS (1U << 3) 569 #define IORING_FEAT_CUR_PERSONALITY (1U << 4) 570 #define IORING_FEAT_FAST_POLL (1U << 5) 571 #define IORING_FEAT_POLL_32BITS (1U << 6) 572 #define IORING_FEAT_SQPOLL_NONFIXED (1U << 7) 573 #define IORING_FEAT_EXT_ARG (1U << 8) 574 #define IORING_FEAT_NATIVE_WORKERS (1U << 9) 575 #define IORING_FEAT_RSRC_TAGS (1U << 10) 576 #define IORING_FEAT_CQE_SKIP (1U << 11) 577 #define IORING_FEAT_LINKED_FILE (1U << 12) 578 #define IORING_FEAT_REG_REG_RING (1U << 13) 579 #define IORING_FEAT_RECVSEND_BUNDLE (1U << 14) 580 #define IORING_FEAT_MIN_TIMEOUT (1U << 15) 581 #define IORING_FEAT_RW_ATTR (1U << 16) 582 #define IORING_FEAT_NO_IOWAIT (1U << 17) 583 584 /* 585 * io_uring_register(2) opcodes and arguments 586 */ 587 enum io_uring_register_op { 588 IORING_REGISTER_BUFFERS = 0, 589 IORING_UNREGISTER_BUFFERS = 1, 590 IORING_REGISTER_FILES = 2, 591 IORING_UNREGISTER_FILES = 3, 592 IORING_REGISTER_EVENTFD = 4, 593 IORING_UNREGISTER_EVENTFD = 5, 594 IORING_REGISTER_FILES_UPDATE = 6, 595 IORING_REGISTER_EVENTFD_ASYNC = 7, 596 IORING_REGISTER_PROBE = 8, 597 IORING_REGISTER_PERSONALITY = 9, 598 IORING_UNREGISTER_PERSONALITY = 10, 599 IORING_REGISTER_RESTRICTIONS = 11, 600 IORING_REGISTER_ENABLE_RINGS = 12, 601 602 /* extended with tagging */ 603 IORING_REGISTER_FILES2 = 13, 604 IORING_REGISTER_FILES_UPDATE2 = 14, 605 IORING_REGISTER_BUFFERS2 = 15, 606 IORING_REGISTER_BUFFERS_UPDATE = 16, 607 608 /* set/clear io-wq thread affinities */ 609 IORING_REGISTER_IOWQ_AFF = 17, 610 IORING_UNREGISTER_IOWQ_AFF = 18, 611 612 /* set/get max number of io-wq workers */ 613 IORING_REGISTER_IOWQ_MAX_WORKERS = 19, 614 615 /* register/unregister io_uring fd with the ring */ 616 IORING_REGISTER_RING_FDS = 20, 617 IORING_UNREGISTER_RING_FDS = 21, 618 619 /* register ring based provide buffer group */ 620 IORING_REGISTER_PBUF_RING = 22, 621 IORING_UNREGISTER_PBUF_RING = 23, 622 623 /* sync cancelation API */ 624 IORING_REGISTER_SYNC_CANCEL = 24, 625 626 /* register a range of fixed file slots for automatic slot allocation */ 627 IORING_REGISTER_FILE_ALLOC_RANGE = 25, 628 629 /* return status information for a buffer group */ 630 IORING_REGISTER_PBUF_STATUS = 26, 631 632 /* set/clear busy poll settings */ 633 IORING_REGISTER_NAPI = 27, 634 IORING_UNREGISTER_NAPI = 28, 635 636 IORING_REGISTER_CLOCK = 29, 637 638 /* clone registered buffers from source ring to current ring */ 639 IORING_REGISTER_CLONE_BUFFERS = 30, 640 641 /* send MSG_RING without having a ring */ 642 IORING_REGISTER_SEND_MSG_RING = 31, 643 644 /* 32 reserved for zc rx */ 645 646 /* resize CQ ring */ 647 IORING_REGISTER_RESIZE_RINGS = 33, 648 649 IORING_REGISTER_MEM_REGION = 34, 650 651 /* this goes last */ 652 IORING_REGISTER_LAST, 653 654 /* flag added to the opcode to use a registered ring fd */ 655 IORING_REGISTER_USE_REGISTERED_RING = 1U << 31 656 }; 657 658 /* io-wq worker categories */ 659 enum io_wq_type { 660 IO_WQ_BOUND, 661 IO_WQ_UNBOUND, 662 }; 663 664 /* deprecated, see struct io_uring_rsrc_update */ 665 struct io_uring_files_update { 666 __u32 offset; 667 __u32 resv; 668 __aligned_u64 /* __s32 * */ fds; 669 }; 670 671 enum { 672 /* initialise with user provided memory pointed by user_addr */ 673 IORING_MEM_REGION_TYPE_USER = 1, 674 }; 675 676 struct io_uring_region_desc { 677 __u64 user_addr; 678 __u64 size; 679 __u32 flags; 680 __u32 id; 681 __u64 mmap_offset; 682 __u64 __resv[4]; 683 }; 684 685 enum { 686 /* expose the region as registered wait arguments */ 687 IORING_MEM_REGION_REG_WAIT_ARG = 1, 688 }; 689 690 struct io_uring_mem_region_reg { 691 __u64 region_uptr; /* struct io_uring_region_desc * */ 692 __u64 flags; 693 __u64 __resv[2]; 694 }; 695 696 /* 697 * Register a fully sparse file space, rather than pass in an array of all 698 * -1 file descriptors. 699 */ 700 #define IORING_RSRC_REGISTER_SPARSE (1U << 0) 701 702 struct io_uring_rsrc_register { 703 __u32 nr; 704 __u32 flags; 705 __u64 resv2; 706 __aligned_u64 data; 707 __aligned_u64 tags; 708 }; 709 710 struct io_uring_rsrc_update { 711 __u32 offset; 712 __u32 resv; 713 __aligned_u64 data; 714 }; 715 716 struct io_uring_rsrc_update2 { 717 __u32 offset; 718 __u32 resv; 719 __aligned_u64 data; 720 __aligned_u64 tags; 721 __u32 nr; 722 __u32 resv2; 723 }; 724 725 /* Skip updating fd indexes set to this value in the fd table */ 726 #define IORING_REGISTER_FILES_SKIP (-2) 727 728 #define IO_URING_OP_SUPPORTED (1U << 0) 729 730 struct io_uring_probe_op { 731 __u8 op; 732 __u8 resv; 733 __u16 flags; /* IO_URING_OP_* flags */ 734 __u32 resv2; 735 }; 736 737 struct io_uring_probe { 738 __u8 last_op; /* last opcode supported */ 739 __u8 ops_len; /* length of ops[] array below */ 740 __u16 resv; 741 __u32 resv2[3]; 742 struct io_uring_probe_op ops[]; 743 }; 744 745 struct io_uring_restriction { 746 __u16 opcode; 747 union { 748 __u8 register_op; /* IORING_RESTRICTION_REGISTER_OP */ 749 __u8 sqe_op; /* IORING_RESTRICTION_SQE_OP */ 750 __u8 sqe_flags; /* IORING_RESTRICTION_SQE_FLAGS_* */ 751 }; 752 __u8 resv; 753 __u32 resv2[3]; 754 }; 755 756 struct io_uring_clock_register { 757 __u32 clockid; 758 __u32 __resv[3]; 759 }; 760 761 enum { 762 IORING_REGISTER_SRC_REGISTERED = (1U << 0), 763 IORING_REGISTER_DST_REPLACE = (1U << 1), 764 }; 765 766 struct io_uring_clone_buffers { 767 __u32 src_fd; 768 __u32 flags; 769 __u32 src_off; 770 __u32 dst_off; 771 __u32 nr; 772 __u32 pad[3]; 773 }; 774 775 struct io_uring_buf { 776 __u64 addr; 777 __u32 len; 778 __u16 bid; 779 __u16 resv; 780 }; 781 782 struct io_uring_buf_ring { 783 union { 784 /* 785 * To avoid spilling into more pages than we need to, the 786 * ring tail is overlaid with the io_uring_buf->resv field. 787 */ 788 struct { 789 __u64 resv1; 790 __u32 resv2; 791 __u16 resv3; 792 __u16 tail; 793 }; 794 __DECLARE_FLEX_ARRAY(struct io_uring_buf, bufs); 795 }; 796 }; 797 798 /* 799 * Flags for IORING_REGISTER_PBUF_RING. 800 * 801 * IOU_PBUF_RING_MMAP: If set, kernel will allocate the memory for the ring. 802 * The application must not set a ring_addr in struct 803 * io_uring_buf_reg, instead it must subsequently call 804 * mmap(2) with the offset set as: 805 * IORING_OFF_PBUF_RING | (bgid << IORING_OFF_PBUF_SHIFT) 806 * to get a virtual mapping for the ring. 807 * IOU_PBUF_RING_INC: If set, buffers consumed from this buffer ring can be 808 * consumed incrementally. Normally one (or more) buffers 809 * are fully consumed. With incremental consumptions, it's 810 * feasible to register big ranges of buffers, and each 811 * use of it will consume only as much as it needs. This 812 * requires that both the kernel and application keep 813 * track of where the current read/recv index is at. 814 */ 815 enum io_uring_register_pbuf_ring_flags { 816 IOU_PBUF_RING_MMAP = 1, 817 IOU_PBUF_RING_INC = 2, 818 }; 819 820 /* argument for IORING_(UN)REGISTER_PBUF_RING */ 821 struct io_uring_buf_reg { 822 __u64 ring_addr; 823 __u32 ring_entries; 824 __u16 bgid; 825 __u16 flags; 826 __u64 resv[3]; 827 }; 828 829 /* argument for IORING_REGISTER_PBUF_STATUS */ 830 struct io_uring_buf_status { 831 __u32 buf_group; /* input */ 832 __u32 head; /* output */ 833 __u32 resv[8]; 834 }; 835 836 enum io_uring_napi_op { 837 /* register/ungister backward compatible opcode */ 838 IO_URING_NAPI_REGISTER_OP = 0, 839 840 /* opcodes to update napi_list when static tracking is used */ 841 IO_URING_NAPI_STATIC_ADD_ID = 1, 842 IO_URING_NAPI_STATIC_DEL_ID = 2 843 }; 844 845 enum io_uring_napi_tracking_strategy { 846 /* value must be 0 for backward compatibility */ 847 IO_URING_NAPI_TRACKING_DYNAMIC = 0, 848 IO_URING_NAPI_TRACKING_STATIC = 1, 849 IO_URING_NAPI_TRACKING_INACTIVE = 255 850 }; 851 852 /* argument for IORING_(UN)REGISTER_NAPI */ 853 struct io_uring_napi { 854 __u32 busy_poll_to; 855 __u8 prefer_busy_poll; 856 857 /* a io_uring_napi_op value */ 858 __u8 opcode; 859 __u8 pad[2]; 860 861 /* 862 * for IO_URING_NAPI_REGISTER_OP, it is a 863 * io_uring_napi_tracking_strategy value. 864 * 865 * for IO_URING_NAPI_STATIC_ADD_ID/IO_URING_NAPI_STATIC_DEL_ID 866 * it is the napi id to add/del from napi_list. 867 */ 868 __u32 op_param; 869 __u32 resv; 870 }; 871 872 /* 873 * io_uring_restriction->opcode values 874 */ 875 enum io_uring_register_restriction_op { 876 /* Allow an io_uring_register(2) opcode */ 877 IORING_RESTRICTION_REGISTER_OP = 0, 878 879 /* Allow an sqe opcode */ 880 IORING_RESTRICTION_SQE_OP = 1, 881 882 /* Allow sqe flags */ 883 IORING_RESTRICTION_SQE_FLAGS_ALLOWED = 2, 884 885 /* Require sqe flags (these flags must be set on each submission) */ 886 IORING_RESTRICTION_SQE_FLAGS_REQUIRED = 3, 887 888 IORING_RESTRICTION_LAST 889 }; 890 891 enum { 892 IORING_REG_WAIT_TS = (1U << 0), 893 }; 894 895 /* 896 * Argument for io_uring_enter(2) with 897 * IORING_GETEVENTS | IORING_ENTER_EXT_ARG_REG set, where the actual argument 898 * is an index into a previously registered fixed wait region described by 899 * the below structure. 900 */ 901 struct io_uring_reg_wait { 902 struct __kernel_timespec ts; 903 __u32 min_wait_usec; 904 __u32 flags; 905 __u64 sigmask; 906 __u32 sigmask_sz; 907 __u32 pad[3]; 908 __u64 pad2[2]; 909 }; 910 911 /* 912 * Argument for io_uring_enter(2) with IORING_GETEVENTS | IORING_ENTER_EXT_ARG 913 */ 914 struct io_uring_getevents_arg { 915 __u64 sigmask; 916 __u32 sigmask_sz; 917 __u32 min_wait_usec; 918 __u64 ts; 919 }; 920 921 /* 922 * Argument for IORING_REGISTER_SYNC_CANCEL 923 */ 924 struct io_uring_sync_cancel_reg { 925 __u64 addr; 926 __s32 fd; 927 __u32 flags; 928 struct __kernel_timespec timeout; 929 __u8 opcode; 930 __u8 pad[7]; 931 __u64 pad2[3]; 932 }; 933 934 /* 935 * Argument for IORING_REGISTER_FILE_ALLOC_RANGE 936 * The range is specified as [off, off + len) 937 */ 938 struct io_uring_file_index_range { 939 __u32 off; 940 __u32 len; 941 __u64 resv; 942 }; 943 944 struct io_uring_recvmsg_out { 945 __u32 namelen; 946 __u32 controllen; 947 __u32 payloadlen; 948 __u32 flags; 949 }; 950 951 /* 952 * Argument for IORING_OP_URING_CMD when file is a socket 953 */ 954 enum io_uring_socket_op { 955 SOCKET_URING_OP_SIOCINQ = 0, 956 SOCKET_URING_OP_SIOCOUTQ, 957 SOCKET_URING_OP_GETSOCKOPT, 958 SOCKET_URING_OP_SETSOCKOPT, 959 }; 960 961 #ifdef __cplusplus 962 } 963 #endif 964 965 #endif 966