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 __u32 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 __u32 pipe_flags; 77 }; 78 __u64 user_data; /* data to be passed back at completion time */ 79 /* pack this to avoid bogus arm OABI complaints */ 80 union { 81 /* index into fixed buffers, if used */ 82 __u16 buf_index; 83 /* for grouped buffer selection */ 84 __u16 buf_group; 85 } __attribute__((packed)); 86 /* personality to use, if used */ 87 __u16 personality; 88 union { 89 __s32 splice_fd_in; 90 __u32 file_index; 91 __u32 zcrx_ifq_idx; 92 __u32 optlen; 93 struct { 94 __u16 addr_len; 95 __u16 __pad3[1]; 96 }; 97 struct { 98 __u8 write_stream; 99 __u8 __pad4[3]; 100 }; 101 }; 102 union { 103 struct { 104 __u64 addr3; 105 __u64 __pad2[1]; 106 }; 107 struct { 108 __u64 attr_ptr; /* pointer to attribute information */ 109 __u64 attr_type_mask; /* bit mask of attributes */ 110 }; 111 __u64 optval; 112 /* 113 * If the ring is initialized with IORING_SETUP_SQE128, then 114 * this field is used for 80 bytes of arbitrary command data 115 */ 116 __u8 cmd[0]; 117 }; 118 }; 119 120 /* sqe->attr_type_mask flags */ 121 #define IORING_RW_ATTR_FLAG_PI (1U << 0) 122 /* PI attribute information */ 123 struct io_uring_attr_pi { 124 __u16 flags; 125 __u16 app_tag; 126 __u32 len; 127 __u64 addr; 128 __u64 seed; 129 __u64 rsvd; 130 }; 131 132 /* 133 * If sqe->file_index is set to this for opcodes that instantiate a new 134 * direct descriptor (like openat/openat2/accept), then io_uring will allocate 135 * an available direct descriptor instead of having the application pass one 136 * in. The picked direct descriptor will be returned in cqe->res, or -ENFILE 137 * if the space is full. 138 */ 139 #define IORING_FILE_INDEX_ALLOC (~0U) 140 141 enum io_uring_sqe_flags_bit { 142 IOSQE_FIXED_FILE_BIT, 143 IOSQE_IO_DRAIN_BIT, 144 IOSQE_IO_LINK_BIT, 145 IOSQE_IO_HARDLINK_BIT, 146 IOSQE_ASYNC_BIT, 147 IOSQE_BUFFER_SELECT_BIT, 148 IOSQE_CQE_SKIP_SUCCESS_BIT, 149 }; 150 151 /* 152 * sqe->flags 153 */ 154 /* use fixed fileset */ 155 #define IOSQE_FIXED_FILE (1U << IOSQE_FIXED_FILE_BIT) 156 /* issue after inflight IO */ 157 #define IOSQE_IO_DRAIN (1U << IOSQE_IO_DRAIN_BIT) 158 /* links next sqe */ 159 #define IOSQE_IO_LINK (1U << IOSQE_IO_LINK_BIT) 160 /* like LINK, but stronger */ 161 #define IOSQE_IO_HARDLINK (1U << IOSQE_IO_HARDLINK_BIT) 162 /* always go async */ 163 #define IOSQE_ASYNC (1U << IOSQE_ASYNC_BIT) 164 /* select buffer from sqe->buf_group */ 165 #define IOSQE_BUFFER_SELECT (1U << IOSQE_BUFFER_SELECT_BIT) 166 /* don't post CQE if request succeeded */ 167 #define IOSQE_CQE_SKIP_SUCCESS (1U << IOSQE_CQE_SKIP_SUCCESS_BIT) 168 169 /* 170 * io_uring_setup() flags 171 */ 172 #define IORING_SETUP_IOPOLL (1U << 0) /* io_context is polled */ 173 #define IORING_SETUP_SQPOLL (1U << 1) /* SQ poll thread */ 174 #define IORING_SETUP_SQ_AFF (1U << 2) /* sq_thread_cpu is valid */ 175 #define IORING_SETUP_CQSIZE (1U << 3) /* app defines CQ size */ 176 #define IORING_SETUP_CLAMP (1U << 4) /* clamp SQ/CQ ring sizes */ 177 #define IORING_SETUP_ATTACH_WQ (1U << 5) /* attach to existing wq */ 178 #define IORING_SETUP_R_DISABLED (1U << 6) /* start with ring disabled */ 179 #define IORING_SETUP_SUBMIT_ALL (1U << 7) /* continue submit on error */ 180 /* 181 * Cooperative task running. When requests complete, they often require 182 * forcing the submitter to transition to the kernel to complete. If this 183 * flag is set, work will be done when the task transitions anyway, rather 184 * than force an inter-processor interrupt reschedule. This avoids interrupting 185 * a task running in userspace, and saves an IPI. 186 */ 187 #define IORING_SETUP_COOP_TASKRUN (1U << 8) 188 /* 189 * If COOP_TASKRUN is set, get notified if task work is available for 190 * running and a kernel transition would be needed to run it. This sets 191 * IORING_SQ_TASKRUN in the sq ring flags. Not valid with COOP_TASKRUN. 192 */ 193 #define IORING_SETUP_TASKRUN_FLAG (1U << 9) 194 #define IORING_SETUP_SQE128 (1U << 10) /* SQEs are 128 byte */ 195 #define IORING_SETUP_CQE32 (1U << 11) /* CQEs are 32 byte */ 196 /* 197 * Only one task is allowed to submit requests 198 */ 199 #define IORING_SETUP_SINGLE_ISSUER (1U << 12) 200 201 /* 202 * Defer running task work to get events. 203 * Rather than running bits of task work whenever the task transitions 204 * try to do it just before it is needed. 205 */ 206 #define IORING_SETUP_DEFER_TASKRUN (1U << 13) 207 208 /* 209 * Application provides the memory for the rings 210 */ 211 #define IORING_SETUP_NO_MMAP (1U << 14) 212 213 /* 214 * Register the ring fd in itself for use with 215 * IORING_REGISTER_USE_REGISTERED_RING; return a registered fd index rather 216 * than an fd. 217 */ 218 #define IORING_SETUP_REGISTERED_FD_ONLY (1U << 15) 219 220 /* 221 * Removes indirection through the SQ index array. 222 */ 223 #define IORING_SETUP_NO_SQARRAY (1U << 16) 224 225 /* Use hybrid poll in iopoll process */ 226 #define IORING_SETUP_HYBRID_IOPOLL (1U << 17) 227 228 enum io_uring_op { 229 IORING_OP_NOP, 230 IORING_OP_READV, 231 IORING_OP_WRITEV, 232 IORING_OP_FSYNC, 233 IORING_OP_READ_FIXED, 234 IORING_OP_WRITE_FIXED, 235 IORING_OP_POLL_ADD, 236 IORING_OP_POLL_REMOVE, 237 IORING_OP_SYNC_FILE_RANGE, 238 IORING_OP_SENDMSG, 239 IORING_OP_RECVMSG, 240 IORING_OP_TIMEOUT, 241 IORING_OP_TIMEOUT_REMOVE, 242 IORING_OP_ACCEPT, 243 IORING_OP_ASYNC_CANCEL, 244 IORING_OP_LINK_TIMEOUT, 245 IORING_OP_CONNECT, 246 IORING_OP_FALLOCATE, 247 IORING_OP_OPENAT, 248 IORING_OP_CLOSE, 249 IORING_OP_FILES_UPDATE, 250 IORING_OP_STATX, 251 IORING_OP_READ, 252 IORING_OP_WRITE, 253 IORING_OP_FADVISE, 254 IORING_OP_MADVISE, 255 IORING_OP_SEND, 256 IORING_OP_RECV, 257 IORING_OP_OPENAT2, 258 IORING_OP_EPOLL_CTL, 259 IORING_OP_SPLICE, 260 IORING_OP_PROVIDE_BUFFERS, 261 IORING_OP_REMOVE_BUFFERS, 262 IORING_OP_TEE, 263 IORING_OP_SHUTDOWN, 264 IORING_OP_RENAMEAT, 265 IORING_OP_UNLINKAT, 266 IORING_OP_MKDIRAT, 267 IORING_OP_SYMLINKAT, 268 IORING_OP_LINKAT, 269 IORING_OP_MSG_RING, 270 IORING_OP_FSETXATTR, 271 IORING_OP_SETXATTR, 272 IORING_OP_FGETXATTR, 273 IORING_OP_GETXATTR, 274 IORING_OP_SOCKET, 275 IORING_OP_URING_CMD, 276 IORING_OP_SEND_ZC, 277 IORING_OP_SENDMSG_ZC, 278 IORING_OP_READ_MULTISHOT, 279 IORING_OP_WAITID, 280 IORING_OP_FUTEX_WAIT, 281 IORING_OP_FUTEX_WAKE, 282 IORING_OP_FUTEX_WAITV, 283 IORING_OP_FIXED_FD_INSTALL, 284 IORING_OP_FTRUNCATE, 285 IORING_OP_BIND, 286 IORING_OP_LISTEN, 287 IORING_OP_RECV_ZC, 288 IORING_OP_EPOLL_WAIT, 289 IORING_OP_READV_FIXED, 290 IORING_OP_WRITEV_FIXED, 291 IORING_OP_PIPE, 292 293 /* this goes last, obviously */ 294 IORING_OP_LAST, 295 }; 296 297 /* 298 * sqe->uring_cmd_flags top 8bits aren't available for userspace 299 * IORING_URING_CMD_FIXED use registered buffer; pass this flag 300 * along with setting sqe->buf_index. 301 */ 302 #define IORING_URING_CMD_FIXED (1U << 0) 303 #define IORING_URING_CMD_MASK IORING_URING_CMD_FIXED 304 305 306 /* 307 * sqe->fsync_flags 308 */ 309 #define IORING_FSYNC_DATASYNC (1U << 0) 310 311 /* 312 * sqe->timeout_flags 313 */ 314 #define IORING_TIMEOUT_ABS (1U << 0) 315 #define IORING_TIMEOUT_UPDATE (1U << 1) 316 #define IORING_TIMEOUT_BOOTTIME (1U << 2) 317 #define IORING_TIMEOUT_REALTIME (1U << 3) 318 #define IORING_LINK_TIMEOUT_UPDATE (1U << 4) 319 #define IORING_TIMEOUT_ETIME_SUCCESS (1U << 5) 320 #define IORING_TIMEOUT_MULTISHOT (1U << 6) 321 #define IORING_TIMEOUT_CLOCK_MASK (IORING_TIMEOUT_BOOTTIME | IORING_TIMEOUT_REALTIME) 322 #define IORING_TIMEOUT_UPDATE_MASK (IORING_TIMEOUT_UPDATE | IORING_LINK_TIMEOUT_UPDATE) 323 /* 324 * sqe->splice_flags 325 * extends splice(2) flags 326 */ 327 #define SPLICE_F_FD_IN_FIXED (1U << 31) /* the last bit of __u32 */ 328 329 /* 330 * POLL_ADD flags. Note that since sqe->poll_events is the flag space, the 331 * command flags for POLL_ADD are stored in sqe->len. 332 * 333 * IORING_POLL_ADD_MULTI Multishot poll. Sets IORING_CQE_F_MORE if 334 * the poll handler will continue to report 335 * CQEs on behalf of the same SQE. 336 * 337 * IORING_POLL_UPDATE Update existing poll request, matching 338 * sqe->addr as the old user_data field. 339 * 340 * IORING_POLL_LEVEL Level triggered poll. 341 */ 342 #define IORING_POLL_ADD_MULTI (1U << 0) 343 #define IORING_POLL_UPDATE_EVENTS (1U << 1) 344 #define IORING_POLL_UPDATE_USER_DATA (1U << 2) 345 #define IORING_POLL_ADD_LEVEL (1U << 3) 346 347 /* 348 * ASYNC_CANCEL flags. 349 * 350 * IORING_ASYNC_CANCEL_ALL Cancel all requests that match the given key 351 * IORING_ASYNC_CANCEL_FD Key off 'fd' for cancelation rather than the 352 * request 'user_data' 353 * IORING_ASYNC_CANCEL_ANY Match any request 354 * IORING_ASYNC_CANCEL_FD_FIXED 'fd' passed in is a fixed descriptor 355 * IORING_ASYNC_CANCEL_USERDATA Match on user_data, default for no other key 356 * IORING_ASYNC_CANCEL_OP Match request based on opcode 357 */ 358 #define IORING_ASYNC_CANCEL_ALL (1U << 0) 359 #define IORING_ASYNC_CANCEL_FD (1U << 1) 360 #define IORING_ASYNC_CANCEL_ANY (1U << 2) 361 #define IORING_ASYNC_CANCEL_FD_FIXED (1U << 3) 362 #define IORING_ASYNC_CANCEL_USERDATA (1U << 4) 363 #define IORING_ASYNC_CANCEL_OP (1U << 5) 364 365 /* 366 * send/sendmsg and recv/recvmsg flags (sqe->ioprio) 367 * 368 * IORING_RECVSEND_POLL_FIRST If set, instead of first attempting to send 369 * or receive and arm poll if that yields an 370 * -EAGAIN result, arm poll upfront and skip 371 * the initial transfer attempt. 372 * 373 * IORING_RECV_MULTISHOT Multishot recv. Sets IORING_CQE_F_MORE if 374 * the handler will continue to report 375 * CQEs on behalf of the same SQE. 376 * 377 * IORING_RECVSEND_FIXED_BUF Use registered buffers, the index is stored in 378 * the buf_index field. 379 * 380 * IORING_SEND_ZC_REPORT_USAGE 381 * If set, SEND[MSG]_ZC should report 382 * the zerocopy usage in cqe.res 383 * for the IORING_CQE_F_NOTIF cqe. 384 * 0 is reported if zerocopy was actually possible. 385 * IORING_NOTIF_USAGE_ZC_COPIED if data was copied 386 * (at least partially). 387 * 388 * IORING_RECVSEND_BUNDLE Used with IOSQE_BUFFER_SELECT. If set, send or 389 * recv will grab as many buffers from the buffer 390 * group ID given and send them all. The completion 391 * result will be the number of buffers send, with 392 * the starting buffer ID in cqe->flags as per 393 * usual for provided buffer usage. The buffers 394 * will be contiguous from the starting buffer ID. 395 * 396 * IORING_SEND_VECTORIZED If set, SEND[_ZC] will take a pointer to a io_vec 397 * to allow vectorized send operations. 398 */ 399 #define IORING_RECVSEND_POLL_FIRST (1U << 0) 400 #define IORING_RECV_MULTISHOT (1U << 1) 401 #define IORING_RECVSEND_FIXED_BUF (1U << 2) 402 #define IORING_SEND_ZC_REPORT_USAGE (1U << 3) 403 #define IORING_RECVSEND_BUNDLE (1U << 4) 404 #define IORING_SEND_VECTORIZED (1U << 5) 405 406 /* 407 * cqe.res for IORING_CQE_F_NOTIF if 408 * IORING_SEND_ZC_REPORT_USAGE was requested 409 * 410 * It should be treated as a flag, all other 411 * bits of cqe.res should be treated as reserved! 412 */ 413 #define IORING_NOTIF_USAGE_ZC_COPIED (1U << 31) 414 415 /* 416 * accept flags stored in sqe->ioprio 417 */ 418 #define IORING_ACCEPT_MULTISHOT (1U << 0) 419 #define IORING_ACCEPT_DONTWAIT (1U << 1) 420 #define IORING_ACCEPT_POLL_FIRST (1U << 2) 421 422 /* 423 * IORING_OP_MSG_RING command types, stored in sqe->addr 424 */ 425 enum io_uring_msg_ring_flags { 426 IORING_MSG_DATA, /* pass sqe->len as 'res' and off as user_data */ 427 IORING_MSG_SEND_FD, /* send a registered fd to another ring */ 428 }; 429 430 /* 431 * IORING_OP_MSG_RING flags (sqe->msg_ring_flags) 432 * 433 * IORING_MSG_RING_CQE_SKIP Don't post a CQE to the target ring. Not 434 * applicable for IORING_MSG_DATA, obviously. 435 */ 436 #define IORING_MSG_RING_CQE_SKIP (1U << 0) 437 /* Pass through the flags from sqe->file_index to cqe->flags */ 438 #define IORING_MSG_RING_FLAGS_PASS (1U << 1) 439 440 /* 441 * IORING_OP_FIXED_FD_INSTALL flags (sqe->install_fd_flags) 442 * 443 * IORING_FIXED_FD_NO_CLOEXEC Don't mark the fd as O_CLOEXEC 444 */ 445 #define IORING_FIXED_FD_NO_CLOEXEC (1U << 0) 446 447 /* 448 * IORING_OP_NOP flags (sqe->nop_flags) 449 * 450 * IORING_NOP_INJECT_RESULT Inject result from sqe->result 451 */ 452 #define IORING_NOP_INJECT_RESULT (1U << 0) 453 #define IORING_NOP_FILE (1U << 1) 454 #define IORING_NOP_FIXED_FILE (1U << 2) 455 #define IORING_NOP_FIXED_BUFFER (1U << 3) 456 #define IORING_NOP_TW (1U << 4) 457 458 /* 459 * IO completion data structure (Completion Queue Entry) 460 */ 461 struct io_uring_cqe { 462 __u64 user_data; /* sqe->user_data value passed back */ 463 __s32 res; /* result code for this event */ 464 __u32 flags; 465 466 /* 467 * If the ring is initialized with IORING_SETUP_CQE32, then this field 468 * contains 16-bytes of padding, doubling the size of the CQE. 469 */ 470 __u64 big_cqe[]; 471 }; 472 473 /* 474 * cqe->flags 475 * 476 * IORING_CQE_F_BUFFER If set, the upper 16 bits are the buffer ID 477 * IORING_CQE_F_MORE If set, parent SQE will generate more CQE entries 478 * IORING_CQE_F_SOCK_NONEMPTY If set, more data to read after socket recv 479 * IORING_CQE_F_NOTIF Set for notification CQEs. Can be used to distinct 480 * them from sends. 481 * IORING_CQE_F_BUF_MORE If set, the buffer ID set in the completion will get 482 * more completions. In other words, the buffer is being 483 * partially consumed, and will be used by the kernel for 484 * more completions. This is only set for buffers used via 485 * the incremental buffer consumption, as provided by 486 * a ring buffer setup with IOU_PBUF_RING_INC. For any 487 * other provided buffer type, all completions with a 488 * buffer passed back is automatically returned to the 489 * application. 490 */ 491 #define IORING_CQE_F_BUFFER (1U << 0) 492 #define IORING_CQE_F_MORE (1U << 1) 493 #define IORING_CQE_F_SOCK_NONEMPTY (1U << 2) 494 #define IORING_CQE_F_NOTIF (1U << 3) 495 #define IORING_CQE_F_BUF_MORE (1U << 4) 496 497 #define IORING_CQE_BUFFER_SHIFT 16 498 499 /* 500 * Magic offsets for the application to mmap the data it needs 501 */ 502 #define IORING_OFF_SQ_RING 0ULL 503 #define IORING_OFF_CQ_RING 0x8000000ULL 504 #define IORING_OFF_SQES 0x10000000ULL 505 #define IORING_OFF_PBUF_RING 0x80000000ULL 506 #define IORING_OFF_PBUF_SHIFT 16 507 #define IORING_OFF_MMAP_MASK 0xf8000000ULL 508 509 /* 510 * Filled with the offset for mmap(2) 511 */ 512 struct io_sqring_offsets { 513 __u32 head; 514 __u32 tail; 515 __u32 ring_mask; 516 __u32 ring_entries; 517 __u32 flags; 518 __u32 dropped; 519 __u32 array; 520 __u32 resv1; 521 __u64 user_addr; 522 }; 523 524 /* 525 * sq_ring->flags 526 */ 527 #define IORING_SQ_NEED_WAKEUP (1U << 0) /* needs io_uring_enter wakeup */ 528 #define IORING_SQ_CQ_OVERFLOW (1U << 1) /* CQ ring is overflown */ 529 #define IORING_SQ_TASKRUN (1U << 2) /* task should enter the kernel */ 530 531 struct io_cqring_offsets { 532 __u32 head; 533 __u32 tail; 534 __u32 ring_mask; 535 __u32 ring_entries; 536 __u32 overflow; 537 __u32 cqes; 538 __u32 flags; 539 __u32 resv1; 540 __u64 user_addr; 541 }; 542 543 /* 544 * cq_ring->flags 545 */ 546 547 /* disable eventfd notifications */ 548 #define IORING_CQ_EVENTFD_DISABLED (1U << 0) 549 550 /* 551 * io_uring_enter(2) flags 552 */ 553 #define IORING_ENTER_GETEVENTS (1U << 0) 554 #define IORING_ENTER_SQ_WAKEUP (1U << 1) 555 #define IORING_ENTER_SQ_WAIT (1U << 2) 556 #define IORING_ENTER_EXT_ARG (1U << 3) 557 #define IORING_ENTER_REGISTERED_RING (1U << 4) 558 #define IORING_ENTER_ABS_TIMER (1U << 5) 559 #define IORING_ENTER_EXT_ARG_REG (1U << 6) 560 #define IORING_ENTER_NO_IOWAIT (1U << 7) 561 562 /* 563 * Passed in for io_uring_setup(2). Copied back with updated info on success 564 */ 565 struct io_uring_params { 566 __u32 sq_entries; 567 __u32 cq_entries; 568 __u32 flags; 569 __u32 sq_thread_cpu; 570 __u32 sq_thread_idle; 571 __u32 features; 572 __u32 wq_fd; 573 __u32 resv[3]; 574 struct io_sqring_offsets sq_off; 575 struct io_cqring_offsets cq_off; 576 }; 577 578 /* 579 * io_uring_params->features flags 580 */ 581 #define IORING_FEAT_SINGLE_MMAP (1U << 0) 582 #define IORING_FEAT_NODROP (1U << 1) 583 #define IORING_FEAT_SUBMIT_STABLE (1U << 2) 584 #define IORING_FEAT_RW_CUR_POS (1U << 3) 585 #define IORING_FEAT_CUR_PERSONALITY (1U << 4) 586 #define IORING_FEAT_FAST_POLL (1U << 5) 587 #define IORING_FEAT_POLL_32BITS (1U << 6) 588 #define IORING_FEAT_SQPOLL_NONFIXED (1U << 7) 589 #define IORING_FEAT_EXT_ARG (1U << 8) 590 #define IORING_FEAT_NATIVE_WORKERS (1U << 9) 591 #define IORING_FEAT_RSRC_TAGS (1U << 10) 592 #define IORING_FEAT_CQE_SKIP (1U << 11) 593 #define IORING_FEAT_LINKED_FILE (1U << 12) 594 #define IORING_FEAT_REG_REG_RING (1U << 13) 595 #define IORING_FEAT_RECVSEND_BUNDLE (1U << 14) 596 #define IORING_FEAT_MIN_TIMEOUT (1U << 15) 597 #define IORING_FEAT_RW_ATTR (1U << 16) 598 #define IORING_FEAT_NO_IOWAIT (1U << 17) 599 600 /* 601 * io_uring_register(2) opcodes and arguments 602 */ 603 enum io_uring_register_op { 604 IORING_REGISTER_BUFFERS = 0, 605 IORING_UNREGISTER_BUFFERS = 1, 606 IORING_REGISTER_FILES = 2, 607 IORING_UNREGISTER_FILES = 3, 608 IORING_REGISTER_EVENTFD = 4, 609 IORING_UNREGISTER_EVENTFD = 5, 610 IORING_REGISTER_FILES_UPDATE = 6, 611 IORING_REGISTER_EVENTFD_ASYNC = 7, 612 IORING_REGISTER_PROBE = 8, 613 IORING_REGISTER_PERSONALITY = 9, 614 IORING_UNREGISTER_PERSONALITY = 10, 615 IORING_REGISTER_RESTRICTIONS = 11, 616 IORING_REGISTER_ENABLE_RINGS = 12, 617 618 /* extended with tagging */ 619 IORING_REGISTER_FILES2 = 13, 620 IORING_REGISTER_FILES_UPDATE2 = 14, 621 IORING_REGISTER_BUFFERS2 = 15, 622 IORING_REGISTER_BUFFERS_UPDATE = 16, 623 624 /* set/clear io-wq thread affinities */ 625 IORING_REGISTER_IOWQ_AFF = 17, 626 IORING_UNREGISTER_IOWQ_AFF = 18, 627 628 /* set/get max number of io-wq workers */ 629 IORING_REGISTER_IOWQ_MAX_WORKERS = 19, 630 631 /* register/unregister io_uring fd with the ring */ 632 IORING_REGISTER_RING_FDS = 20, 633 IORING_UNREGISTER_RING_FDS = 21, 634 635 /* register ring based provide buffer group */ 636 IORING_REGISTER_PBUF_RING = 22, 637 IORING_UNREGISTER_PBUF_RING = 23, 638 639 /* sync cancelation API */ 640 IORING_REGISTER_SYNC_CANCEL = 24, 641 642 /* register a range of fixed file slots for automatic slot allocation */ 643 IORING_REGISTER_FILE_ALLOC_RANGE = 25, 644 645 /* return status information for a buffer group */ 646 IORING_REGISTER_PBUF_STATUS = 26, 647 648 /* set/clear busy poll settings */ 649 IORING_REGISTER_NAPI = 27, 650 IORING_UNREGISTER_NAPI = 28, 651 652 IORING_REGISTER_CLOCK = 29, 653 654 /* clone registered buffers from source ring to current ring */ 655 IORING_REGISTER_CLONE_BUFFERS = 30, 656 657 /* send MSG_RING without having a ring */ 658 IORING_REGISTER_SEND_MSG_RING = 31, 659 660 /* register a netdev hw rx queue for zerocopy */ 661 IORING_REGISTER_ZCRX_IFQ = 32, 662 663 /* resize CQ ring */ 664 IORING_REGISTER_RESIZE_RINGS = 33, 665 666 IORING_REGISTER_MEM_REGION = 34, 667 668 /* this goes last */ 669 IORING_REGISTER_LAST, 670 671 /* flag added to the opcode to use a registered ring fd */ 672 IORING_REGISTER_USE_REGISTERED_RING = 1U << 31 673 }; 674 675 /* io-wq worker categories */ 676 enum io_wq_type { 677 IO_WQ_BOUND, 678 IO_WQ_UNBOUND, 679 }; 680 681 /* deprecated, see struct io_uring_rsrc_update */ 682 struct io_uring_files_update { 683 __u32 offset; 684 __u32 resv; 685 __aligned_u64 /* __s32 * */ fds; 686 }; 687 688 enum { 689 /* initialise with user provided memory pointed by user_addr */ 690 IORING_MEM_REGION_TYPE_USER = 1, 691 }; 692 693 struct io_uring_region_desc { 694 __u64 user_addr; 695 __u64 size; 696 __u32 flags; 697 __u32 id; 698 __u64 mmap_offset; 699 __u64 __resv[4]; 700 }; 701 702 enum { 703 /* expose the region as registered wait arguments */ 704 IORING_MEM_REGION_REG_WAIT_ARG = 1, 705 }; 706 707 struct io_uring_mem_region_reg { 708 __u64 region_uptr; /* struct io_uring_region_desc * */ 709 __u64 flags; 710 __u64 __resv[2]; 711 }; 712 713 /* 714 * Register a fully sparse file space, rather than pass in an array of all 715 * -1 file descriptors. 716 */ 717 #define IORING_RSRC_REGISTER_SPARSE (1U << 0) 718 719 struct io_uring_rsrc_register { 720 __u32 nr; 721 __u32 flags; 722 __u64 resv2; 723 __aligned_u64 data; 724 __aligned_u64 tags; 725 }; 726 727 struct io_uring_rsrc_update { 728 __u32 offset; 729 __u32 resv; 730 __aligned_u64 data; 731 }; 732 733 struct io_uring_rsrc_update2 { 734 __u32 offset; 735 __u32 resv; 736 __aligned_u64 data; 737 __aligned_u64 tags; 738 __u32 nr; 739 __u32 resv2; 740 }; 741 742 /* Skip updating fd indexes set to this value in the fd table */ 743 #define IORING_REGISTER_FILES_SKIP (-2) 744 745 #define IO_URING_OP_SUPPORTED (1U << 0) 746 747 struct io_uring_probe_op { 748 __u8 op; 749 __u8 resv; 750 __u16 flags; /* IO_URING_OP_* flags */ 751 __u32 resv2; 752 }; 753 754 struct io_uring_probe { 755 __u8 last_op; /* last opcode supported */ 756 __u8 ops_len; /* length of ops[] array below */ 757 __u16 resv; 758 __u32 resv2[3]; 759 struct io_uring_probe_op ops[]; 760 }; 761 762 struct io_uring_restriction { 763 __u16 opcode; 764 union { 765 __u8 register_op; /* IORING_RESTRICTION_REGISTER_OP */ 766 __u8 sqe_op; /* IORING_RESTRICTION_SQE_OP */ 767 __u8 sqe_flags; /* IORING_RESTRICTION_SQE_FLAGS_* */ 768 }; 769 __u8 resv; 770 __u32 resv2[3]; 771 }; 772 773 struct io_uring_clock_register { 774 __u32 clockid; 775 __u32 __resv[3]; 776 }; 777 778 enum { 779 IORING_REGISTER_SRC_REGISTERED = (1U << 0), 780 IORING_REGISTER_DST_REPLACE = (1U << 1), 781 }; 782 783 struct io_uring_clone_buffers { 784 __u32 src_fd; 785 __u32 flags; 786 __u32 src_off; 787 __u32 dst_off; 788 __u32 nr; 789 __u32 pad[3]; 790 }; 791 792 struct io_uring_buf { 793 __u64 addr; 794 __u32 len; 795 __u16 bid; 796 __u16 resv; 797 }; 798 799 struct io_uring_buf_ring { 800 union { 801 /* 802 * To avoid spilling into more pages than we need to, the 803 * ring tail is overlaid with the io_uring_buf->resv field. 804 */ 805 struct { 806 __u64 resv1; 807 __u32 resv2; 808 __u16 resv3; 809 __u16 tail; 810 }; 811 __DECLARE_FLEX_ARRAY(struct io_uring_buf, bufs); 812 }; 813 }; 814 815 /* 816 * Flags for IORING_REGISTER_PBUF_RING. 817 * 818 * IOU_PBUF_RING_MMAP: If set, kernel will allocate the memory for the ring. 819 * The application must not set a ring_addr in struct 820 * io_uring_buf_reg, instead it must subsequently call 821 * mmap(2) with the offset set as: 822 * IORING_OFF_PBUF_RING | (bgid << IORING_OFF_PBUF_SHIFT) 823 * to get a virtual mapping for the ring. 824 * IOU_PBUF_RING_INC: If set, buffers consumed from this buffer ring can be 825 * consumed incrementally. Normally one (or more) buffers 826 * are fully consumed. With incremental consumptions, it's 827 * feasible to register big ranges of buffers, and each 828 * use of it will consume only as much as it needs. This 829 * requires that both the kernel and application keep 830 * track of where the current read/recv index is at. 831 */ 832 enum io_uring_register_pbuf_ring_flags { 833 IOU_PBUF_RING_MMAP = 1, 834 IOU_PBUF_RING_INC = 2, 835 }; 836 837 /* argument for IORING_(UN)REGISTER_PBUF_RING */ 838 struct io_uring_buf_reg { 839 __u64 ring_addr; 840 __u32 ring_entries; 841 __u16 bgid; 842 __u16 flags; 843 __u64 resv[3]; 844 }; 845 846 /* argument for IORING_REGISTER_PBUF_STATUS */ 847 struct io_uring_buf_status { 848 __u32 buf_group; /* input */ 849 __u32 head; /* output */ 850 __u32 resv[8]; 851 }; 852 853 enum io_uring_napi_op { 854 /* register/ungister backward compatible opcode */ 855 IO_URING_NAPI_REGISTER_OP = 0, 856 857 /* opcodes to update napi_list when static tracking is used */ 858 IO_URING_NAPI_STATIC_ADD_ID = 1, 859 IO_URING_NAPI_STATIC_DEL_ID = 2 860 }; 861 862 enum io_uring_napi_tracking_strategy { 863 /* value must be 0 for backward compatibility */ 864 IO_URING_NAPI_TRACKING_DYNAMIC = 0, 865 IO_URING_NAPI_TRACKING_STATIC = 1, 866 IO_URING_NAPI_TRACKING_INACTIVE = 255 867 }; 868 869 /* argument for IORING_(UN)REGISTER_NAPI */ 870 struct io_uring_napi { 871 __u32 busy_poll_to; 872 __u8 prefer_busy_poll; 873 874 /* a io_uring_napi_op value */ 875 __u8 opcode; 876 __u8 pad[2]; 877 878 /* 879 * for IO_URING_NAPI_REGISTER_OP, it is a 880 * io_uring_napi_tracking_strategy value. 881 * 882 * for IO_URING_NAPI_STATIC_ADD_ID/IO_URING_NAPI_STATIC_DEL_ID 883 * it is the napi id to add/del from napi_list. 884 */ 885 __u32 op_param; 886 __u32 resv; 887 }; 888 889 /* 890 * io_uring_restriction->opcode values 891 */ 892 enum io_uring_register_restriction_op { 893 /* Allow an io_uring_register(2) opcode */ 894 IORING_RESTRICTION_REGISTER_OP = 0, 895 896 /* Allow an sqe opcode */ 897 IORING_RESTRICTION_SQE_OP = 1, 898 899 /* Allow sqe flags */ 900 IORING_RESTRICTION_SQE_FLAGS_ALLOWED = 2, 901 902 /* Require sqe flags (these flags must be set on each submission) */ 903 IORING_RESTRICTION_SQE_FLAGS_REQUIRED = 3, 904 905 IORING_RESTRICTION_LAST 906 }; 907 908 enum { 909 IORING_REG_WAIT_TS = (1U << 0), 910 }; 911 912 /* 913 * Argument for io_uring_enter(2) with 914 * IORING_GETEVENTS | IORING_ENTER_EXT_ARG_REG set, where the actual argument 915 * is an index into a previously registered fixed wait region described by 916 * the below structure. 917 */ 918 struct io_uring_reg_wait { 919 struct __kernel_timespec ts; 920 __u32 min_wait_usec; 921 __u32 flags; 922 __u64 sigmask; 923 __u32 sigmask_sz; 924 __u32 pad[3]; 925 __u64 pad2[2]; 926 }; 927 928 /* 929 * Argument for io_uring_enter(2) with IORING_GETEVENTS | IORING_ENTER_EXT_ARG 930 */ 931 struct io_uring_getevents_arg { 932 __u64 sigmask; 933 __u32 sigmask_sz; 934 __u32 min_wait_usec; 935 __u64 ts; 936 }; 937 938 /* 939 * Argument for IORING_REGISTER_SYNC_CANCEL 940 */ 941 struct io_uring_sync_cancel_reg { 942 __u64 addr; 943 __s32 fd; 944 __u32 flags; 945 struct __kernel_timespec timeout; 946 __u8 opcode; 947 __u8 pad[7]; 948 __u64 pad2[3]; 949 }; 950 951 /* 952 * Argument for IORING_REGISTER_FILE_ALLOC_RANGE 953 * The range is specified as [off, off + len) 954 */ 955 struct io_uring_file_index_range { 956 __u32 off; 957 __u32 len; 958 __u64 resv; 959 }; 960 961 struct io_uring_recvmsg_out { 962 __u32 namelen; 963 __u32 controllen; 964 __u32 payloadlen; 965 __u32 flags; 966 }; 967 968 /* 969 * Argument for IORING_OP_URING_CMD when file is a socket 970 */ 971 enum io_uring_socket_op { 972 SOCKET_URING_OP_SIOCINQ = 0, 973 SOCKET_URING_OP_SIOCOUTQ, 974 SOCKET_URING_OP_GETSOCKOPT, 975 SOCKET_URING_OP_SETSOCKOPT, 976 SOCKET_URING_OP_TX_TIMESTAMP, 977 }; 978 979 /* 980 * SOCKET_URING_OP_TX_TIMESTAMP definitions 981 */ 982 983 #define IORING_TIMESTAMP_HW_SHIFT 16 984 /* The cqe->flags bit from which the timestamp type is stored */ 985 #define IORING_TIMESTAMP_TYPE_SHIFT (IORING_TIMESTAMP_HW_SHIFT + 1) 986 /* The cqe->flags flag signifying whether it's a hardware timestamp */ 987 #define IORING_CQE_F_TSTAMP_HW ((__u32)1 << IORING_TIMESTAMP_HW_SHIFT) 988 989 struct io_timespec { 990 __u64 tv_sec; 991 __u64 tv_nsec; 992 }; 993 994 /* Zero copy receive refill queue entry */ 995 struct io_uring_zcrx_rqe { 996 __u64 off; 997 __u32 len; 998 __u32 __pad; 999 }; 1000 1001 struct io_uring_zcrx_cqe { 1002 __u64 off; 1003 __u64 __pad; 1004 }; 1005 1006 /* The bit from which area id is encoded into offsets */ 1007 #define IORING_ZCRX_AREA_SHIFT 48 1008 #define IORING_ZCRX_AREA_MASK (~(((__u64)1 << IORING_ZCRX_AREA_SHIFT) - 1)) 1009 1010 struct io_uring_zcrx_offsets { 1011 __u32 head; 1012 __u32 tail; 1013 __u32 rqes; 1014 __u32 __resv2; 1015 __u64 __resv[2]; 1016 }; 1017 1018 enum io_uring_zcrx_area_flags { 1019 IORING_ZCRX_AREA_DMABUF = 1, 1020 }; 1021 1022 struct io_uring_zcrx_area_reg { 1023 __u64 addr; 1024 __u64 len; 1025 __u64 rq_area_token; 1026 __u32 flags; 1027 __u32 dmabuf_fd; 1028 __u64 __resv2[2]; 1029 }; 1030 1031 /* 1032 * Argument for IORING_REGISTER_ZCRX_IFQ 1033 */ 1034 struct io_uring_zcrx_ifq_reg { 1035 __u32 if_idx; 1036 __u32 if_rxq; 1037 __u32 rq_entries; 1038 __u32 flags; 1039 1040 __u64 area_ptr; /* pointer to struct io_uring_zcrx_area_reg */ 1041 __u64 region_ptr; /* struct io_uring_region_desc * */ 1042 1043 struct io_uring_zcrx_offsets offsets; 1044 __u32 zcrx_id; 1045 __u32 __resv2; 1046 __u64 __resv[3]; 1047 }; 1048 1049 #ifdef __cplusplus 1050 } 1051 #endif 1052 1053 #endif 1054