tools/include/uapi/linux/io_uring.h

   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         };
  75         __u64   user_data;      /* data to be passed back at completion time */
  76         /* pack this to avoid bogus arm OABI complaints */
  77         union {
  78                 /* index into fixed buffers, if used */
  79                 __u16   buf_index;
  80                 /* for grouped buffer selection */
  81                 __u16   buf_group;
  82         } __attribute__((packed));
  83         /* personality to use, if used */
  84         __u16   personality;
  85         union {
  86                 __s32   splice_fd_in;
  87                 __u32   file_index;
  88                 __u32   optlen;
  89                 struct {
  90                         __u16   addr_len;
  91                         __u16   __pad3[1];
  92                 };
  93         };
  94         union {
  95                 struct {
  96                         __u64   addr3;
  97                         __u64   __pad2[1];
  98                 };
  99                 __u64   optval;
 100                 /*
 101                  * If the ring is initialized with IORING_SETUP_SQE128, then
 102                  * this field is used for 80 bytes of arbitrary command data
 103                  */
 104                 __u8    cmd[0];
 105         };
 106 };
 107
 108 /*
 109  * If sqe->file_index is set to this for opcodes that instantiate a new
 110  * direct descriptor (like openat/openat2/accept), then io_uring will allocate
 111  * an available direct descriptor instead of having the application pass one
 112  * in. The picked direct descriptor will be returned in cqe->res, or -ENFILE
 113  * if the space is full.
 114  */
 115 #define IORING_FILE_INDEX_ALLOC         (~0U)
 116
 117 enum {
 118         IOSQE_FIXED_FILE_BIT,
 119         IOSQE_IO_DRAIN_BIT,
 120         IOSQE_IO_LINK_BIT,
 121         IOSQE_IO_HARDLINK_BIT,
 122         IOSQE_ASYNC_BIT,
 123         IOSQE_BUFFER_SELECT_BIT,
 124         IOSQE_CQE_SKIP_SUCCESS_BIT,
 125 };
 126
 127 /*
 128  * sqe->flags
 129  */
 130 /* use fixed fileset */
 131 #define IOSQE_FIXED_FILE        (1U << IOSQE_FIXED_FILE_BIT)
 132 /* issue after inflight IO */
 133 #define IOSQE_IO_DRAIN          (1U << IOSQE_IO_DRAIN_BIT)
 134 /* links next sqe */
 135 #define IOSQE_IO_LINK           (1U << IOSQE_IO_LINK_BIT)
 136 /* like LINK, but stronger */
 137 #define IOSQE_IO_HARDLINK       (1U << IOSQE_IO_HARDLINK_BIT)
 138 /* always go async */
 139 #define IOSQE_ASYNC             (1U << IOSQE_ASYNC_BIT)
 140 /* select buffer from sqe->buf_group */
 141 #define IOSQE_BUFFER_SELECT     (1U << IOSQE_BUFFER_SELECT_BIT)
 142 /* don't post CQE if request succeeded */
 143 #define IOSQE_CQE_SKIP_SUCCESS  (1U << IOSQE_CQE_SKIP_SUCCESS_BIT)
 144
 145 /*
 146  * io_uring_setup() flags
 147  */
 148 #define IORING_SETUP_IOPOLL     (1U << 0)       /* io_context is polled */
 149 #define IORING_SETUP_SQPOLL     (1U << 1)       /* SQ poll thread */
 150 #define IORING_SETUP_SQ_AFF     (1U << 2)       /* sq_thread_cpu is valid */
 151 #define IORING_SETUP_CQSIZE     (1U << 3)       /* app defines CQ size */
 152 #define IORING_SETUP_CLAMP      (1U << 4)       /* clamp SQ/CQ ring sizes */
 153 #define IORING_SETUP_ATTACH_WQ  (1U << 5)       /* attach to existing wq */
 154 #define IORING_SETUP_R_DISABLED (1U << 6)       /* start with ring disabled */
 155 #define IORING_SETUP_SUBMIT_ALL (1U << 7)       /* continue submit on error */
 156 /*
 157  * Cooperative task running. When requests complete, they often require
 158  * forcing the submitter to transition to the kernel to complete. If this
 159  * flag is set, work will be done when the task transitions anyway, rather
 160  * than force an inter-processor interrupt reschedule. This avoids interrupting
 161  * a task running in userspace, and saves an IPI.
 162  */
 163 #define IORING_SETUP_COOP_TASKRUN       (1U << 8)
 164 /*
 165  * If COOP_TASKRUN is set, get notified if task work is available for
 166  * running and a kernel transition would be needed to run it. This sets
 167  * IORING_SQ_TASKRUN in the sq ring flags. Not valid with COOP_TASKRUN.
 168  */
 169 #define IORING_SETUP_TASKRUN_FLAG       (1U << 9)
 170 #define IORING_SETUP_SQE128             (1U << 10) /* SQEs are 128 byte */
 171 #define IORING_SETUP_CQE32              (1U << 11) /* CQEs are 32 byte */
 172 /*
 173  * Only one task is allowed to submit requests
 174  */
 175 #define IORING_SETUP_SINGLE_ISSUER      (1U << 12)
 176
 177 /*
 178  * Defer running task work to get events.
 179  * Rather than running bits of task work whenever the task transitions
 180  * try to do it just before it is needed.
 181  */
 182 #define IORING_SETUP_DEFER_TASKRUN      (1U << 13)
 183
 184 /*
 185  * Application provides the memory for the rings
 186  */
 187 #define IORING_SETUP_NO_MMAP            (1U << 14)
 188
 189 /*
 190  * Register the ring fd in itself for use with
 191  * IORING_REGISTER_USE_REGISTERED_RING; return a registered fd index rather
 192  * than an fd.
 193  */
 194 #define IORING_SETUP_REGISTERED_FD_ONLY (1U << 15)
 195
 196 /*
 197  * Removes indirection through the SQ index array.
 198  */
 199 #define IORING_SETUP_NO_SQARRAY         (1U << 16)
 200
 201 enum io_uring_op {
 202         IORING_OP_NOP,
 203         IORING_OP_READV,
 204         IORING_OP_WRITEV,
 205         IORING_OP_FSYNC,
 206         IORING_OP_READ_FIXED,
 207         IORING_OP_WRITE_FIXED,
 208         IORING_OP_POLL_ADD,
 209         IORING_OP_POLL_REMOVE,
 210         IORING_OP_SYNC_FILE_RANGE,
 211         IORING_OP_SENDMSG,
 212         IORING_OP_RECVMSG,
 213         IORING_OP_TIMEOUT,
 214         IORING_OP_TIMEOUT_REMOVE,
 215         IORING_OP_ACCEPT,
 216         IORING_OP_ASYNC_CANCEL,
 217         IORING_OP_LINK_TIMEOUT,
 218         IORING_OP_CONNECT,
 219         IORING_OP_FALLOCATE,
 220         IORING_OP_OPENAT,
 221         IORING_OP_CLOSE,
 222         IORING_OP_FILES_UPDATE,
 223         IORING_OP_STATX,
 224         IORING_OP_READ,
 225         IORING_OP_WRITE,
 226         IORING_OP_FADVISE,
 227         IORING_OP_MADVISE,
 228         IORING_OP_SEND,
 229         IORING_OP_RECV,
 230         IORING_OP_OPENAT2,
 231         IORING_OP_EPOLL_CTL,
 232         IORING_OP_SPLICE,
 233         IORING_OP_PROVIDE_BUFFERS,
 234         IORING_OP_REMOVE_BUFFERS,
 235         IORING_OP_TEE,
 236         IORING_OP_SHUTDOWN,
 237         IORING_OP_RENAMEAT,
 238         IORING_OP_UNLINKAT,
 239         IORING_OP_MKDIRAT,
 240         IORING_OP_SYMLINKAT,
 241         IORING_OP_LINKAT,
 242         IORING_OP_MSG_RING,
 243         IORING_OP_FSETXATTR,
 244         IORING_OP_SETXATTR,
 245         IORING_OP_FGETXATTR,
 246         IORING_OP_GETXATTR,
 247         IORING_OP_SOCKET,
 248         IORING_OP_URING_CMD,
 249         IORING_OP_SEND_ZC,
 250         IORING_OP_SENDMSG_ZC,
 251         IORING_OP_READ_MULTISHOT,
 252         IORING_OP_WAITID,
 253         IORING_OP_FUTEX_WAIT,
 254         IORING_OP_FUTEX_WAKE,
 255         IORING_OP_FUTEX_WAITV,
 256
 257         /* this goes last, obviously */
 258         IORING_OP_LAST,
 259 };
 260
 261 /*
 262  * sqe->uring_cmd_flags         top 8bits aren't available for userspace
 263  * IORING_URING_CMD_FIXED       use registered buffer; pass this flag
 264  *                              along with setting sqe->buf_index.
 265  */
 266 #define IORING_URING_CMD_FIXED  (1U << 0)
 267 #define IORING_URING_CMD_MASK   IORING_URING_CMD_FIXED
 268
 269
 270 /*
 271  * sqe->fsync_flags
 272  */
 273 #define IORING_FSYNC_DATASYNC   (1U << 0)
 274
 275 /*
 276  * sqe->timeout_flags
 277  */
 278 #define IORING_TIMEOUT_ABS              (1U << 0)
 279 #define IORING_TIMEOUT_UPDATE           (1U << 1)
 280 #define IORING_TIMEOUT_BOOTTIME         (1U << 2)
 281 #define IORING_TIMEOUT_REALTIME         (1U << 3)
 282 #define IORING_LINK_TIMEOUT_UPDATE      (1U << 4)
 283 #define IORING_TIMEOUT_ETIME_SUCCESS    (1U << 5)
 284 #define IORING_TIMEOUT_MULTISHOT        (1U << 6)
 285 #define IORING_TIMEOUT_CLOCK_MASK       (IORING_TIMEOUT_BOOTTIME | IORING_TIMEOUT_REALTIME)
 286 #define IORING_TIMEOUT_UPDATE_MASK      (IORING_TIMEOUT_UPDATE | IORING_LINK_TIMEOUT_UPDATE)
 287 /*
 288  * sqe->splice_flags
 289  * extends splice(2) flags
 290  */
 291 #define SPLICE_F_FD_IN_FIXED    (1U << 31) /* the last bit of __u32 */
 292
 293 /*
 294  * POLL_ADD flags. Note that since sqe->poll_events is the flag space, the
 295  * command flags for POLL_ADD are stored in sqe->len.
 296  *
 297  * IORING_POLL_ADD_MULTI        Multishot poll. Sets IORING_CQE_F_MORE if
 298  *                              the poll handler will continue to report
 299  *                              CQEs on behalf of the same SQE.
 300  *
 301  * IORING_POLL_UPDATE           Update existing poll request, matching
 302  *                              sqe->addr as the old user_data field.
 303  *
 304  * IORING_POLL_LEVEL            Level triggered poll.
 305  */
 306 #define IORING_POLL_ADD_MULTI   (1U << 0)
 307 #define IORING_POLL_UPDATE_EVENTS       (1U << 1)
 308 #define IORING_POLL_UPDATE_USER_DATA    (1U << 2)
 309 #define IORING_POLL_ADD_LEVEL           (1U << 3)
 310
 311 /*
 312  * ASYNC_CANCEL flags.
 313  *
 314  * IORING_ASYNC_CANCEL_ALL      Cancel all requests that match the given key
 315  * IORING_ASYNC_CANCEL_FD       Key off 'fd' for cancelation rather than the
 316  *                              request 'user_data'
 317  * IORING_ASYNC_CANCEL_ANY      Match any request
 318  * IORING_ASYNC_CANCEL_FD_FIXED 'fd' passed in is a fixed descriptor
 319  * IORING_ASYNC_CANCEL_USERDATA Match on user_data, default for no other key
 320  * IORING_ASYNC_CANCEL_OP       Match request based on opcode
 321  */
 322 #define IORING_ASYNC_CANCEL_ALL (1U << 0)
 323 #define IORING_ASYNC_CANCEL_FD  (1U << 1)
 324 #define IORING_ASYNC_CANCEL_ANY (1U << 2)
 325 #define IORING_ASYNC_CANCEL_FD_FIXED    (1U << 3)
 326 #define IORING_ASYNC_CANCEL_USERDATA    (1U << 4)
 327 #define IORING_ASYNC_CANCEL_OP  (1U << 5)
 328
 329 /*
 330  * send/sendmsg and recv/recvmsg flags (sqe->ioprio)
 331  *
 332  * IORING_RECVSEND_POLL_FIRST   If set, instead of first attempting to send
 333  *                              or receive and arm poll if that yields an
 334  *                              -EAGAIN result, arm poll upfront and skip
 335  *                              the initial transfer attempt.
 336  *
 337  * IORING_RECV_MULTISHOT        Multishot recv. Sets IORING_CQE_F_MORE if
 338  *                              the handler will continue to report
 339  *                              CQEs on behalf of the same SQE.
 340  *
 341  * IORING_RECVSEND_FIXED_BUF    Use registered buffers, the index is stored in
 342  *                              the buf_index field.
 343  *
 344  * IORING_SEND_ZC_REPORT_USAGE
 345  *                              If set, SEND[MSG]_ZC should report
 346  *                              the zerocopy usage in cqe.res
 347  *                              for the IORING_CQE_F_NOTIF cqe.
 348  *                              0 is reported if zerocopy was actually possible.
 349  *                              IORING_NOTIF_USAGE_ZC_COPIED if data was copied
 350  *                              (at least partially).
 351  */
 352 #define IORING_RECVSEND_POLL_FIRST      (1U << 0)
 353 #define IORING_RECV_MULTISHOT           (1U << 1)
 354 #define IORING_RECVSEND_FIXED_BUF       (1U << 2)
 355 #define IORING_SEND_ZC_REPORT_USAGE     (1U << 3)
 356
 357 /*
 358  * cqe.res for IORING_CQE_F_NOTIF if
 359  * IORING_SEND_ZC_REPORT_USAGE was requested
 360  *
 361  * It should be treated as a flag, all other
 362  * bits of cqe.res should be treated as reserved!
 363  */
 364 #define IORING_NOTIF_USAGE_ZC_COPIED    (1U << 31)
 365
 366 /*
 367  * accept flags stored in sqe->ioprio
 368  */
 369 #define IORING_ACCEPT_MULTISHOT (1U << 0)
 370
 371 /*
 372  * IORING_OP_MSG_RING command types, stored in sqe->addr
 373  */
 374 enum {
 375         IORING_MSG_DATA,        /* pass sqe->len as 'res' and off as user_data */
 376         IORING_MSG_SEND_FD,     /* send a registered fd to another ring */
 377 };
 378
 379 /*
 380  * IORING_OP_MSG_RING flags (sqe->msg_ring_flags)
 381  *
 382  * IORING_MSG_RING_CQE_SKIP     Don't post a CQE to the target ring. Not
 383  *                              applicable for IORING_MSG_DATA, obviously.
 384  */
 385 #define IORING_MSG_RING_CQE_SKIP        (1U << 0)
 386 /* Pass through the flags from sqe->file_index to cqe->flags */
 387 #define IORING_MSG_RING_FLAGS_PASS      (1U << 1)
 388
 389 /*
 390  * IO completion data structure (Completion Queue Entry)
 391  */
 392 struct io_uring_cqe {
 393         __u64   user_data;      /* sqe->data submission passed back */
 394         __s32   res;            /* result code for this event */
 395         __u32   flags;
 396
 397         /*
 398          * If the ring is initialized with IORING_SETUP_CQE32, then this field
 399          * contains 16-bytes of padding, doubling the size of the CQE.
 400          */
 401         __u64 big_cqe[];
 402 };
 403
 404 /*
 405  * cqe->flags
 406  *
 407  * IORING_CQE_F_BUFFER  If set, the upper 16 bits are the buffer ID
 408  * IORING_CQE_F_MORE    If set, parent SQE will generate more CQE entries
 409  * IORING_CQE_F_SOCK_NONEMPTY   If set, more data to read after socket recv
 410  * IORING_CQE_F_NOTIF   Set for notification CQEs. Can be used to distinct
 411  *                      them from sends.
 412  */
 413 #define IORING_CQE_F_BUFFER             (1U << 0)
 414 #define IORING_CQE_F_MORE               (1U << 1)
 415 #define IORING_CQE_F_SOCK_NONEMPTY      (1U << 2)
 416 #define IORING_CQE_F_NOTIF              (1U << 3)
 417
 418 enum {
 419         IORING_CQE_BUFFER_SHIFT         = 16,
 420 };
 421
 422 /*
 423  * Magic offsets for the application to mmap the data it needs
 424  */
 425 #define IORING_OFF_SQ_RING              0ULL
 426 #define IORING_OFF_CQ_RING              0x8000000ULL
 427 #define IORING_OFF_SQES                 0x10000000ULL
 428 #define IORING_OFF_PBUF_RING            0x80000000ULL
 429 #define IORING_OFF_PBUF_SHIFT           16
 430 #define IORING_OFF_MMAP_MASK            0xf8000000ULL
 431
 432 /*
 433  * Filled with the offset for mmap(2)
 434  */
 435 struct io_sqring_offsets {
 436         __u32 head;
 437         __u32 tail;
 438         __u32 ring_mask;
 439         __u32 ring_entries;
 440         __u32 flags;
 441         __u32 dropped;
 442         __u32 array;
 443         __u32 resv1;
 444         __u64 user_addr;
 445 };
 446
 447 /*
 448  * sq_ring->flags
 449  */
 450 #define IORING_SQ_NEED_WAKEUP   (1U << 0) /* needs io_uring_enter wakeup */
 451 #define IORING_SQ_CQ_OVERFLOW   (1U << 1) /* CQ ring is overflown */
 452 #define IORING_SQ_TASKRUN       (1U << 2) /* task should enter the kernel */
 453
 454 struct io_cqring_offsets {
 455         __u32 head;
 456         __u32 tail;
 457         __u32 ring_mask;
 458         __u32 ring_entries;
 459         __u32 overflow;
 460         __u32 cqes;
 461         __u32 flags;
 462         __u32 resv1;
 463         __u64 user_addr;
 464 };
 465
 466 /*
 467  * cq_ring->flags
 468  */
 469
 470 /* disable eventfd notifications */
 471 #define IORING_CQ_EVENTFD_DISABLED      (1U << 0)
 472
 473 /*
 474  * io_uring_enter(2) flags
 475  */
 476 #define IORING_ENTER_GETEVENTS          (1U << 0)
 477 #define IORING_ENTER_SQ_WAKEUP          (1U << 1)
 478 #define IORING_ENTER_SQ_WAIT            (1U << 2)
 479 #define IORING_ENTER_EXT_ARG            (1U << 3)
 480 #define IORING_ENTER_REGISTERED_RING    (1U << 4)
 481
 482 /*
 483  * Passed in for io_uring_setup(2). Copied back with updated info on success
 484  */
 485 struct io_uring_params {
 486         __u32 sq_entries;
 487         __u32 cq_entries;
 488         __u32 flags;
 489         __u32 sq_thread_cpu;
 490         __u32 sq_thread_idle;
 491         __u32 features;
 492         __u32 wq_fd;
 493         __u32 resv[3];
 494         struct io_sqring_offsets sq_off;
 495         struct io_cqring_offsets cq_off;
 496 };
 497
 498 /*
 499  * io_uring_params->features flags
 500  */
 501 #define IORING_FEAT_SINGLE_MMAP         (1U << 0)
 502 #define IORING_FEAT_NODROP              (1U << 1)
 503 #define IORING_FEAT_SUBMIT_STABLE       (1U << 2)
 504 #define IORING_FEAT_RW_CUR_POS          (1U << 3)
 505 #define IORING_FEAT_CUR_PERSONALITY     (1U << 4)
 506 #define IORING_FEAT_FAST_POLL           (1U << 5)
 507 #define IORING_FEAT_POLL_32BITS         (1U << 6)
 508 #define IORING_FEAT_SQPOLL_NONFIXED     (1U << 7)
 509 #define IORING_FEAT_EXT_ARG             (1U << 8)
 510 #define IORING_FEAT_NATIVE_WORKERS      (1U << 9)
 511 #define IORING_FEAT_RSRC_TAGS           (1U << 10)
 512 #define IORING_FEAT_CQE_SKIP            (1U << 11)
 513 #define IORING_FEAT_LINKED_FILE         (1U << 12)
 514 #define IORING_FEAT_REG_REG_RING        (1U << 13)
 515
 516 /*
 517  * io_uring_register(2) opcodes and arguments
 518  */
 519 enum {
 520         IORING_REGISTER_BUFFERS                 = 0,
 521         IORING_UNREGISTER_BUFFERS               = 1,
 522         IORING_REGISTER_FILES                   = 2,
 523         IORING_UNREGISTER_FILES                 = 3,
 524         IORING_REGISTER_EVENTFD                 = 4,
 525         IORING_UNREGISTER_EVENTFD               = 5,
 526         IORING_REGISTER_FILES_UPDATE            = 6,
 527         IORING_REGISTER_EVENTFD_ASYNC           = 7,
 528         IORING_REGISTER_PROBE                   = 8,
 529         IORING_REGISTER_PERSONALITY             = 9,
 530         IORING_UNREGISTER_PERSONALITY           = 10,
 531         IORING_REGISTER_RESTRICTIONS            = 11,
 532         IORING_REGISTER_ENABLE_RINGS            = 12,
 533
 534         /* extended with tagging */
 535         IORING_REGISTER_FILES2                  = 13,
 536         IORING_REGISTER_FILES_UPDATE2           = 14,
 537         IORING_REGISTER_BUFFERS2                = 15,
 538         IORING_REGISTER_BUFFERS_UPDATE          = 16,
 539
 540         /* set/clear io-wq thread affinities */
 541         IORING_REGISTER_IOWQ_AFF                = 17,
 542         IORING_UNREGISTER_IOWQ_AFF              = 18,
 543
 544         /* set/get max number of io-wq workers */
 545         IORING_REGISTER_IOWQ_MAX_WORKERS        = 19,
 546
 547         /* register/unregister io_uring fd with the ring */
 548         IORING_REGISTER_RING_FDS                = 20,
 549         IORING_UNREGISTER_RING_FDS              = 21,
 550
 551         /* register ring based provide buffer group */
 552         IORING_REGISTER_PBUF_RING               = 22,
 553         IORING_UNREGISTER_PBUF_RING             = 23,
 554
 555         /* sync cancelation API */
 556         IORING_REGISTER_SYNC_CANCEL             = 24,
 557
 558         /* register a range of fixed file slots for automatic slot allocation */
 559         IORING_REGISTER_FILE_ALLOC_RANGE        = 25,
 560
 561         /* this goes last */
 562         IORING_REGISTER_LAST,
 563
 564         /* flag added to the opcode to use a registered ring fd */
 565         IORING_REGISTER_USE_REGISTERED_RING     = 1U << 31
 566 };
 567
 568 /* io-wq worker categories */
 569 enum {
 570         IO_WQ_BOUND,
 571         IO_WQ_UNBOUND,
 572 };
 573
 574 /* deprecated, see struct io_uring_rsrc_update */
 575 struct io_uring_files_update {
 576         __u32 offset;
 577         __u32 resv;
 578         __aligned_u64 /* __s32 * */ fds;
 579 };
 580
 581 /*
 582  * Register a fully sparse file space, rather than pass in an array of all
 583  * -1 file descriptors.
 584  */
 585 #define IORING_RSRC_REGISTER_SPARSE     (1U << 0)
 586
 587 struct io_uring_rsrc_register {
 588         __u32 nr;
 589         __u32 flags;
 590         __u64 resv2;
 591         __aligned_u64 data;
 592         __aligned_u64 tags;
 593 };
 594
 595 struct io_uring_rsrc_update {
 596         __u32 offset;
 597         __u32 resv;
 598         __aligned_u64 data;
 599 };
 600
 601 struct io_uring_rsrc_update2 {
 602         __u32 offset;
 603         __u32 resv;
 604         __aligned_u64 data;
 605         __aligned_u64 tags;
 606         __u32 nr;
 607         __u32 resv2;
 608 };
 609
 610 /* Skip updating fd indexes set to this value in the fd table */
 611 #define IORING_REGISTER_FILES_SKIP      (-2)
 612
 613 #define IO_URING_OP_SUPPORTED   (1U << 0)
 614
 615 struct io_uring_probe_op {
 616         __u8 op;
 617         __u8 resv;
 618         __u16 flags;    /* IO_URING_OP_* flags */
 619         __u32 resv2;
 620 };
 621
 622 struct io_uring_probe {
 623         __u8 last_op;   /* last opcode supported */
 624         __u8 ops_len;   /* length of ops[] array below */
 625         __u16 resv;
 626         __u32 resv2[3];
 627         struct io_uring_probe_op ops[];
 628 };
 629
 630 struct io_uring_restriction {
 631         __u16 opcode;
 632         union {
 633                 __u8 register_op; /* IORING_RESTRICTION_REGISTER_OP */
 634                 __u8 sqe_op;      /* IORING_RESTRICTION_SQE_OP */
 635                 __u8 sqe_flags;   /* IORING_RESTRICTION_SQE_FLAGS_* */
 636         };
 637         __u8 resv;
 638         __u32 resv2[3];
 639 };
 640
 641 struct io_uring_buf {
 642         __u64   addr;
 643         __u32   len;
 644         __u16   bid;
 645         __u16   resv;
 646 };
 647
 648 struct io_uring_buf_ring {
 649         union {
 650                 /*
 651                  * To avoid spilling into more pages than we need to, the
 652                  * ring tail is overlaid with the io_uring_buf->resv field.
 653                  */
 654                 struct {
 655                         __u64   resv1;
 656                         __u32   resv2;
 657                         __u16   resv3;
 658                         __u16   tail;
 659                 };
 660                 __DECLARE_FLEX_ARRAY(struct io_uring_buf, bufs);
 661         };
 662 };
 663
 664 /*
 665  * Flags for IORING_REGISTER_PBUF_RING.
 666  *
 667  * IOU_PBUF_RING_MMAP:  If set, kernel will allocate the memory for the ring.
 668  *                      The application must not set a ring_addr in struct
 669  *                      io_uring_buf_reg, instead it must subsequently call
 670  *                      mmap(2) with the offset set as:
 671  *                      IORING_OFF_PBUF_RING | (bgid << IORING_OFF_PBUF_SHIFT)
 672  *                      to get a virtual mapping for the ring.
 673  */
 674 enum {
 675         IOU_PBUF_RING_MMAP      = 1,
 676 };
 677
 678 /* argument for IORING_(UN)REGISTER_PBUF_RING */
 679 struct io_uring_buf_reg {
 680         __u64   ring_addr;
 681         __u32   ring_entries;
 682         __u16   bgid;
 683         __u16   flags;
 684         __u64   resv[3];
 685 };
 686
 687 /*
 688  * io_uring_restriction->opcode values
 689  */
 690 enum {
 691         /* Allow an io_uring_register(2) opcode */
 692         IORING_RESTRICTION_REGISTER_OP          = 0,
 693
 694         /* Allow an sqe opcode */
 695         IORING_RESTRICTION_SQE_OP               = 1,
 696
 697         /* Allow sqe flags */
 698         IORING_RESTRICTION_SQE_FLAGS_ALLOWED    = 2,
 699
 700         /* Require sqe flags (these flags must be set on each submission) */
 701         IORING_RESTRICTION_SQE_FLAGS_REQUIRED   = 3,
 702
 703         IORING_RESTRICTION_LAST
 704 };
 705
 706 struct io_uring_getevents_arg {
 707         __u64   sigmask;
 708         __u32   sigmask_sz;
 709         __u32   pad;
 710         __u64   ts;
 711 };
 712
 713 /*
 714  * Argument for IORING_REGISTER_SYNC_CANCEL
 715  */
 716 struct io_uring_sync_cancel_reg {
 717         __u64                           addr;
 718         __s32                           fd;
 719         __u32                           flags;
 720         struct __kernel_timespec        timeout;
 721         __u8                            opcode;
 722         __u8                            pad[7];
 723         __u64                           pad2[3];
 724 };
 725
 726 /*
 727  * Argument for IORING_REGISTER_FILE_ALLOC_RANGE
 728  * The range is specified as [off, off + len)
 729  */
 730 struct io_uring_file_index_range {
 731         __u32   off;
 732         __u32   len;
 733         __u64   resv;
 734 };
 735
 736 struct io_uring_recvmsg_out {
 737         __u32 namelen;
 738         __u32 controllen;
 739         __u32 payloadlen;
 740         __u32 flags;
 741 };
 742
 743 /*
 744  * Argument for IORING_OP_URING_CMD when file is a socket
 745  */
 746 enum {
 747         SOCKET_URING_OP_SIOCINQ         = 0,
 748         SOCKET_URING_OP_SIOCOUTQ,
 749         SOCKET_URING_OP_GETSOCKOPT,
 750         SOCKET_URING_OP_SETSOCKOPT,
 751 };
 752
 753 #ifdef __cplusplus
 754 }
 755 #endif
 756
 757 #endif