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