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
24extern "C" {
25#endif
26
27/*
28 * IO submission data structure (Submission Queue Entry)
29 */
30struct 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 */
123struct 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
141enum 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/*
229 * Allow both 16b and 32b CQEs. If a 32b CQE is posted, it will have
230 * IORING_CQE_F_32 set in cqe->flags.
231 */
232#define IORING_SETUP_CQE_MIXED (1U << 18)
233
234enum io_uring_op {
235 IORING_OP_NOP,
236 IORING_OP_READV,
237 IORING_OP_WRITEV,
238 IORING_OP_FSYNC,
239 IORING_OP_READ_FIXED,
240 IORING_OP_WRITE_FIXED,
241 IORING_OP_POLL_ADD,
242 IORING_OP_POLL_REMOVE,
243 IORING_OP_SYNC_FILE_RANGE,
244 IORING_OP_SENDMSG,
245 IORING_OP_RECVMSG,
246 IORING_OP_TIMEOUT,
247 IORING_OP_TIMEOUT_REMOVE,
248 IORING_OP_ACCEPT,
249 IORING_OP_ASYNC_CANCEL,
250 IORING_OP_LINK_TIMEOUT,
251 IORING_OP_CONNECT,
252 IORING_OP_FALLOCATE,
253 IORING_OP_OPENAT,
254 IORING_OP_CLOSE,
255 IORING_OP_FILES_UPDATE,
256 IORING_OP_STATX,
257 IORING_OP_READ,
258 IORING_OP_WRITE,
259 IORING_OP_FADVISE,
260 IORING_OP_MADVISE,
261 IORING_OP_SEND,
262 IORING_OP_RECV,
263 IORING_OP_OPENAT2,
264 IORING_OP_EPOLL_CTL,
265 IORING_OP_SPLICE,
266 IORING_OP_PROVIDE_BUFFERS,
267 IORING_OP_REMOVE_BUFFERS,
268 IORING_OP_TEE,
269 IORING_OP_SHUTDOWN,
270 IORING_OP_RENAMEAT,
271 IORING_OP_UNLINKAT,
272 IORING_OP_MKDIRAT,
273 IORING_OP_SYMLINKAT,
274 IORING_OP_LINKAT,
275 IORING_OP_MSG_RING,
276 IORING_OP_FSETXATTR,
277 IORING_OP_SETXATTR,
278 IORING_OP_FGETXATTR,
279 IORING_OP_GETXATTR,
280 IORING_OP_SOCKET,
281 IORING_OP_URING_CMD,
282 IORING_OP_SEND_ZC,
283 IORING_OP_SENDMSG_ZC,
284 IORING_OP_READ_MULTISHOT,
285 IORING_OP_WAITID,
286 IORING_OP_FUTEX_WAIT,
287 IORING_OP_FUTEX_WAKE,
288 IORING_OP_FUTEX_WAITV,
289 IORING_OP_FIXED_FD_INSTALL,
290 IORING_OP_FTRUNCATE,
291 IORING_OP_BIND,
292 IORING_OP_LISTEN,
293 IORING_OP_RECV_ZC,
294 IORING_OP_EPOLL_WAIT,
295 IORING_OP_READV_FIXED,
296 IORING_OP_WRITEV_FIXED,
297 IORING_OP_PIPE,
298
299 /* this goes last, obviously */
300 IORING_OP_LAST,
301};
302
303/*
304 * sqe->uring_cmd_flags top 8bits aren't available for userspace
305 * IORING_URING_CMD_FIXED use registered buffer; pass this flag
306 * along with setting sqe->buf_index.
307 * IORING_URING_CMD_MULTISHOT must be used with buffer select, like other
308 * multishot commands. Not compatible with
309 * IORING_URING_CMD_FIXED, for now.
310 */
311#define IORING_URING_CMD_FIXED (1U << 0)
312#define IORING_URING_CMD_MULTISHOT (1U << 1)
313#define IORING_URING_CMD_MASK (IORING_URING_CMD_FIXED | IORING_URING_CMD_MULTISHOT)
314
315
316/*
317 * sqe->fsync_flags
318 */
319#define IORING_FSYNC_DATASYNC (1U << 0)
320
321/*
322 * sqe->timeout_flags
323 */
324#define IORING_TIMEOUT_ABS (1U << 0)
325#define IORING_TIMEOUT_UPDATE (1U << 1)
326#define IORING_TIMEOUT_BOOTTIME (1U << 2)
327#define IORING_TIMEOUT_REALTIME (1U << 3)
328#define IORING_LINK_TIMEOUT_UPDATE (1U << 4)
329#define IORING_TIMEOUT_ETIME_SUCCESS (1U << 5)
330#define IORING_TIMEOUT_MULTISHOT (1U << 6)
331#define IORING_TIMEOUT_CLOCK_MASK (IORING_TIMEOUT_BOOTTIME | IORING_TIMEOUT_REALTIME)
332#define IORING_TIMEOUT_UPDATE_MASK (IORING_TIMEOUT_UPDATE | IORING_LINK_TIMEOUT_UPDATE)
333/*
334 * sqe->splice_flags
335 * extends splice(2) flags
336 */
337#define SPLICE_F_FD_IN_FIXED (1U << 31) /* the last bit of __u32 */
338
339/*
340 * POLL_ADD flags. Note that since sqe->poll_events is the flag space, the
341 * command flags for POLL_ADD are stored in sqe->len.
342 *
343 * IORING_POLL_ADD_MULTI Multishot poll. Sets IORING_CQE_F_MORE if
344 * the poll handler will continue to report
345 * CQEs on behalf of the same SQE.
346 *
347 * IORING_POLL_UPDATE Update existing poll request, matching
348 * sqe->addr as the old user_data field.
349 *
350 * IORING_POLL_LEVEL Level triggered poll.
351 */
352#define IORING_POLL_ADD_MULTI (1U << 0)
353#define IORING_POLL_UPDATE_EVENTS (1U << 1)
354#define IORING_POLL_UPDATE_USER_DATA (1U << 2)
355#define IORING_POLL_ADD_LEVEL (1U << 3)
356
357/*
358 * ASYNC_CANCEL flags.
359 *
360 * IORING_ASYNC_CANCEL_ALL Cancel all requests that match the given key
361 * IORING_ASYNC_CANCEL_FD Key off 'fd' for cancelation rather than the
362 * request 'user_data'
363 * IORING_ASYNC_CANCEL_ANY Match any request
364 * IORING_ASYNC_CANCEL_FD_FIXED 'fd' passed in is a fixed descriptor
365 * IORING_ASYNC_CANCEL_USERDATA Match on user_data, default for no other key
366 * IORING_ASYNC_CANCEL_OP Match request based on opcode
367 */
368#define IORING_ASYNC_CANCEL_ALL (1U << 0)
369#define IORING_ASYNC_CANCEL_FD (1U << 1)
370#define IORING_ASYNC_CANCEL_ANY (1U << 2)
371#define IORING_ASYNC_CANCEL_FD_FIXED (1U << 3)
372#define IORING_ASYNC_CANCEL_USERDATA (1U << 4)
373#define IORING_ASYNC_CANCEL_OP (1U << 5)
374
375/*
376 * send/sendmsg and recv/recvmsg flags (sqe->ioprio)
377 *
378 * IORING_RECVSEND_POLL_FIRST If set, instead of first attempting to send
379 * or receive and arm poll if that yields an
380 * -EAGAIN result, arm poll upfront and skip
381 * the initial transfer attempt.
382 *
383 * IORING_RECV_MULTISHOT Multishot recv. Sets IORING_CQE_F_MORE if
384 * the handler will continue to report
385 * CQEs on behalf of the same SQE.
386 *
387 * IORING_RECVSEND_FIXED_BUF Use registered buffers, the index is stored in
388 * the buf_index field.
389 *
390 * IORING_SEND_ZC_REPORT_USAGE
391 * If set, SEND[MSG]_ZC should report
392 * the zerocopy usage in cqe.res
393 * for the IORING_CQE_F_NOTIF cqe.
394 * 0 is reported if zerocopy was actually possible.
395 * IORING_NOTIF_USAGE_ZC_COPIED if data was copied
396 * (at least partially).
397 *
398 * IORING_RECVSEND_BUNDLE Used with IOSQE_BUFFER_SELECT. If set, send or
399 * recv will grab as many buffers from the buffer
400 * group ID given and send them all. The completion
401 * result will be the number of buffers send, with
402 * the starting buffer ID in cqe->flags as per
403 * usual for provided buffer usage. The buffers
404 * will be contiguous from the starting buffer ID.
405 *
406 * IORING_SEND_VECTORIZED If set, SEND[_ZC] will take a pointer to a io_vec
407 * to allow vectorized send operations.
408 */
409#define IORING_RECVSEND_POLL_FIRST (1U << 0)
410#define IORING_RECV_MULTISHOT (1U << 1)
411#define IORING_RECVSEND_FIXED_BUF (1U << 2)
412#define IORING_SEND_ZC_REPORT_USAGE (1U << 3)
413#define IORING_RECVSEND_BUNDLE (1U << 4)
414#define IORING_SEND_VECTORIZED (1U << 5)
415
416/*
417 * cqe.res for IORING_CQE_F_NOTIF if
418 * IORING_SEND_ZC_REPORT_USAGE was requested
419 *
420 * It should be treated as a flag, all other
421 * bits of cqe.res should be treated as reserved!
422 */
423#define IORING_NOTIF_USAGE_ZC_COPIED (1U << 31)
424
425/*
426 * accept flags stored in sqe->ioprio
427 */
428#define IORING_ACCEPT_MULTISHOT (1U << 0)
429#define IORING_ACCEPT_DONTWAIT (1U << 1)
430#define IORING_ACCEPT_POLL_FIRST (1U << 2)
431
432/*
433 * IORING_OP_MSG_RING command types, stored in sqe->addr
434 */
435enum io_uring_msg_ring_flags {
436 IORING_MSG_DATA, /* pass sqe->len as 'res' and off as user_data */
437 IORING_MSG_SEND_FD, /* send a registered fd to another ring */
438};
439
440/*
441 * IORING_OP_MSG_RING flags (sqe->msg_ring_flags)
442 *
443 * IORING_MSG_RING_CQE_SKIP Don't post a CQE to the target ring. Not
444 * applicable for IORING_MSG_DATA, obviously.
445 */
446#define IORING_MSG_RING_CQE_SKIP (1U << 0)
447/* Pass through the flags from sqe->file_index to cqe->flags */
448#define IORING_MSG_RING_FLAGS_PASS (1U << 1)
449
450/*
451 * IORING_OP_FIXED_FD_INSTALL flags (sqe->install_fd_flags)
452 *
453 * IORING_FIXED_FD_NO_CLOEXEC Don't mark the fd as O_CLOEXEC
454 */
455#define IORING_FIXED_FD_NO_CLOEXEC (1U << 0)
456
457/*
458 * IORING_OP_NOP flags (sqe->nop_flags)
459 *
460 * IORING_NOP_INJECT_RESULT Inject result from sqe->result
461 */
462#define IORING_NOP_INJECT_RESULT (1U << 0)
463#define IORING_NOP_FILE (1U << 1)
464#define IORING_NOP_FIXED_FILE (1U << 2)
465#define IORING_NOP_FIXED_BUFFER (1U << 3)
466#define IORING_NOP_TW (1U << 4)
467#define IORING_NOP_CQE32 (1U << 5)
468
469/*
470 * IO completion data structure (Completion Queue Entry)
471 */
472struct io_uring_cqe {
473 __u64 user_data; /* sqe->user_data value passed back */
474 __s32 res; /* result code for this event */
475 __u32 flags;
476
477 /*
478 * If the ring is initialized with IORING_SETUP_CQE32, then this field
479 * contains 16-bytes of padding, doubling the size of the CQE.
480 */
481 __u64 big_cqe[];
482};
483
484/*
485 * cqe->flags
486 *
487 * IORING_CQE_F_BUFFER If set, the upper 16 bits are the buffer ID
488 * IORING_CQE_F_MORE If set, parent SQE will generate more CQE entries
489 * IORING_CQE_F_SOCK_NONEMPTY If set, more data to read after socket recv
490 * IORING_CQE_F_NOTIF Set for notification CQEs. Can be used to distinct
491 * them from sends.
492 * IORING_CQE_F_BUF_MORE If set, the buffer ID set in the completion will get
493 * more completions. In other words, the buffer is being
494 * partially consumed, and will be used by the kernel for
495 * more completions. This is only set for buffers used via
496 * the incremental buffer consumption, as provided by
497 * a ring buffer setup with IOU_PBUF_RING_INC. For any
498 * other provided buffer type, all completions with a
499 * buffer passed back is automatically returned to the
500 * application.
501 * IORING_CQE_F_SKIP If set, then the application/liburing must ignore this
502 * CQE. It's only purpose is to fill a gap in the ring,
503 * if a large CQE is attempted posted when the ring has
504 * just a single small CQE worth of space left before
505 * wrapping.
506 * IORING_CQE_F_32 If set, this is a 32b/big-cqe posting. Use with rings
507 * setup in a mixed CQE mode, where both 16b and 32b
508 * CQEs may be posted to the CQ ring.
509 */
510#define IORING_CQE_F_BUFFER (1U << 0)
511#define IORING_CQE_F_MORE (1U << 1)
512#define IORING_CQE_F_SOCK_NONEMPTY (1U << 2)
513#define IORING_CQE_F_NOTIF (1U << 3)
514#define IORING_CQE_F_BUF_MORE (1U << 4)
515#define IORING_CQE_F_SKIP (1U << 5)
516#define IORING_CQE_F_32 (1U << 15)
517
518#define IORING_CQE_BUFFER_SHIFT 16
519
520/*
521 * Magic offsets for the application to mmap the data it needs
522 */
523#define IORING_OFF_SQ_RING 0ULL
524#define IORING_OFF_CQ_RING 0x8000000ULL
525#define IORING_OFF_SQES 0x10000000ULL
526#define IORING_OFF_PBUF_RING 0x80000000ULL
527#define IORING_OFF_PBUF_SHIFT 16
528#define IORING_OFF_MMAP_MASK 0xf8000000ULL
529
530/*
531 * Filled with the offset for mmap(2)
532 */
533struct io_sqring_offsets {
534 __u32 head;
535 __u32 tail;
536 __u32 ring_mask;
537 __u32 ring_entries;
538 __u32 flags;
539 __u32 dropped;
540 __u32 array;
541 __u32 resv1;
542 __u64 user_addr;
543};
544
545/*
546 * sq_ring->flags
547 */
548#define IORING_SQ_NEED_WAKEUP (1U << 0) /* needs io_uring_enter wakeup */
549#define IORING_SQ_CQ_OVERFLOW (1U << 1) /* CQ ring is overflown */
550#define IORING_SQ_TASKRUN (1U << 2) /* task should enter the kernel */
551
552struct io_cqring_offsets {
553 __u32 head;
554 __u32 tail;
555 __u32 ring_mask;
556 __u32 ring_entries;
557 __u32 overflow;
558 __u32 cqes;
559 __u32 flags;
560 __u32 resv1;
561 __u64 user_addr;
562};
563
564/*
565 * cq_ring->flags
566 */
567
568/* disable eventfd notifications */
569#define IORING_CQ_EVENTFD_DISABLED (1U << 0)
570
571/*
572 * io_uring_enter(2) flags
573 */
574#define IORING_ENTER_GETEVENTS (1U << 0)
575#define IORING_ENTER_SQ_WAKEUP (1U << 1)
576#define IORING_ENTER_SQ_WAIT (1U << 2)
577#define IORING_ENTER_EXT_ARG (1U << 3)
578#define IORING_ENTER_REGISTERED_RING (1U << 4)
579#define IORING_ENTER_ABS_TIMER (1U << 5)
580#define IORING_ENTER_EXT_ARG_REG (1U << 6)
581#define IORING_ENTER_NO_IOWAIT (1U << 7)
582
583/*
584 * Passed in for io_uring_setup(2). Copied back with updated info on success
585 */
586struct io_uring_params {
587 __u32 sq_entries;
588 __u32 cq_entries;
589 __u32 flags;
590 __u32 sq_thread_cpu;
591 __u32 sq_thread_idle;
592 __u32 features;
593 __u32 wq_fd;
594 __u32 resv[3];
595 struct io_sqring_offsets sq_off;
596 struct io_cqring_offsets cq_off;
597};
598
599/*
600 * io_uring_params->features flags
601 */
602#define IORING_FEAT_SINGLE_MMAP (1U << 0)
603#define IORING_FEAT_NODROP (1U << 1)
604#define IORING_FEAT_SUBMIT_STABLE (1U << 2)
605#define IORING_FEAT_RW_CUR_POS (1U << 3)
606#define IORING_FEAT_CUR_PERSONALITY (1U << 4)
607#define IORING_FEAT_FAST_POLL (1U << 5)
608#define IORING_FEAT_POLL_32BITS (1U << 6)
609#define IORING_FEAT_SQPOLL_NONFIXED (1U << 7)
610#define IORING_FEAT_EXT_ARG (1U << 8)
611#define IORING_FEAT_NATIVE_WORKERS (1U << 9)
612#define IORING_FEAT_RSRC_TAGS (1U << 10)
613#define IORING_FEAT_CQE_SKIP (1U << 11)
614#define IORING_FEAT_LINKED_FILE (1U << 12)
615#define IORING_FEAT_REG_REG_RING (1U << 13)
616#define IORING_FEAT_RECVSEND_BUNDLE (1U << 14)
617#define IORING_FEAT_MIN_TIMEOUT (1U << 15)
618#define IORING_FEAT_RW_ATTR (1U << 16)
619#define IORING_FEAT_NO_IOWAIT (1U << 17)
620
621/*
622 * io_uring_register(2) opcodes and arguments
623 */
624enum io_uring_register_op {
625 IORING_REGISTER_BUFFERS = 0,
626 IORING_UNREGISTER_BUFFERS = 1,
627 IORING_REGISTER_FILES = 2,
628 IORING_UNREGISTER_FILES = 3,
629 IORING_REGISTER_EVENTFD = 4,
630 IORING_UNREGISTER_EVENTFD = 5,
631 IORING_REGISTER_FILES_UPDATE = 6,
632 IORING_REGISTER_EVENTFD_ASYNC = 7,
633 IORING_REGISTER_PROBE = 8,
634 IORING_REGISTER_PERSONALITY = 9,
635 IORING_UNREGISTER_PERSONALITY = 10,
636 IORING_REGISTER_RESTRICTIONS = 11,
637 IORING_REGISTER_ENABLE_RINGS = 12,
638
639 /* extended with tagging */
640 IORING_REGISTER_FILES2 = 13,
641 IORING_REGISTER_FILES_UPDATE2 = 14,
642 IORING_REGISTER_BUFFERS2 = 15,
643 IORING_REGISTER_BUFFERS_UPDATE = 16,
644
645 /* set/clear io-wq thread affinities */
646 IORING_REGISTER_IOWQ_AFF = 17,
647 IORING_UNREGISTER_IOWQ_AFF = 18,
648
649 /* set/get max number of io-wq workers */
650 IORING_REGISTER_IOWQ_MAX_WORKERS = 19,
651
652 /* register/unregister io_uring fd with the ring */
653 IORING_REGISTER_RING_FDS = 20,
654 IORING_UNREGISTER_RING_FDS = 21,
655
656 /* register ring based provide buffer group */
657 IORING_REGISTER_PBUF_RING = 22,
658 IORING_UNREGISTER_PBUF_RING = 23,
659
660 /* sync cancelation API */
661 IORING_REGISTER_SYNC_CANCEL = 24,
662
663 /* register a range of fixed file slots for automatic slot allocation */
664 IORING_REGISTER_FILE_ALLOC_RANGE = 25,
665
666 /* return status information for a buffer group */
667 IORING_REGISTER_PBUF_STATUS = 26,
668
669 /* set/clear busy poll settings */
670 IORING_REGISTER_NAPI = 27,
671 IORING_UNREGISTER_NAPI = 28,
672
673 IORING_REGISTER_CLOCK = 29,
674
675 /* clone registered buffers from source ring to current ring */
676 IORING_REGISTER_CLONE_BUFFERS = 30,
677
678 /* send MSG_RING without having a ring */
679 IORING_REGISTER_SEND_MSG_RING = 31,
680
681 /* register a netdev hw rx queue for zerocopy */
682 IORING_REGISTER_ZCRX_IFQ = 32,
683
684 /* resize CQ ring */
685 IORING_REGISTER_RESIZE_RINGS = 33,
686
687 IORING_REGISTER_MEM_REGION = 34,
688
689 /* query various aspects of io_uring, see linux/io_uring/query.h */
690 IORING_REGISTER_QUERY = 35,
691
692 /* return zcrx buffers back into circulation */
693 IORING_REGISTER_ZCRX_REFILL = 36,
694
695 /* this goes last */
696 IORING_REGISTER_LAST,
697
698 /* flag added to the opcode to use a registered ring fd */
699 IORING_REGISTER_USE_REGISTERED_RING = 1U << 31
700};
701
702/* io-wq worker categories */
703enum io_wq_type {
704 IO_WQ_BOUND,
705 IO_WQ_UNBOUND,
706};
707
708/* deprecated, see struct io_uring_rsrc_update */
709struct io_uring_files_update {
710 __u32 offset;
711 __u32 resv;
712 __aligned_u64 /* __s32 * */ fds;
713};
714
715enum {
716 /* initialise with user provided memory pointed by user_addr */
717 IORING_MEM_REGION_TYPE_USER = 1,
718};
719
720struct io_uring_region_desc {
721 __u64 user_addr;
722 __u64 size;
723 __u32 flags;
724 __u32 id;
725 __u64 mmap_offset;
726 __u64 __resv[4];
727};
728
729enum {
730 /* expose the region as registered wait arguments */
731 IORING_MEM_REGION_REG_WAIT_ARG = 1,
732};
733
734struct io_uring_mem_region_reg {
735 __u64 region_uptr; /* struct io_uring_region_desc * */
736 __u64 flags;
737 __u64 __resv[2];
738};
739
740/*
741 * Register a fully sparse file space, rather than pass in an array of all
742 * -1 file descriptors.
743 */
744#define IORING_RSRC_REGISTER_SPARSE (1U << 0)
745
746struct io_uring_rsrc_register {
747 __u32 nr;
748 __u32 flags;
749 __u64 resv2;
750 __aligned_u64 data;
751 __aligned_u64 tags;
752};
753
754struct io_uring_rsrc_update {
755 __u32 offset;
756 __u32 resv;
757 __aligned_u64 data;
758};
759
760struct io_uring_rsrc_update2 {
761 __u32 offset;
762 __u32 resv;
763 __aligned_u64 data;
764 __aligned_u64 tags;
765 __u32 nr;
766 __u32 resv2;
767};
768
769/* Skip updating fd indexes set to this value in the fd table */
770#define IORING_REGISTER_FILES_SKIP (-2)
771
772#define IO_URING_OP_SUPPORTED (1U << 0)
773
774struct io_uring_probe_op {
775 __u8 op;
776 __u8 resv;
777 __u16 flags; /* IO_URING_OP_* flags */
778 __u32 resv2;
779};
780
781struct io_uring_probe {
782 __u8 last_op; /* last opcode supported */
783 __u8 ops_len; /* length of ops[] array below */
784 __u16 resv;
785 __u32 resv2[3];
786 struct io_uring_probe_op ops[];
787};
788
789struct io_uring_restriction {
790 __u16 opcode;
791 union {
792 __u8 register_op; /* IORING_RESTRICTION_REGISTER_OP */
793 __u8 sqe_op; /* IORING_RESTRICTION_SQE_OP */
794 __u8 sqe_flags; /* IORING_RESTRICTION_SQE_FLAGS_* */
795 };
796 __u8 resv;
797 __u32 resv2[3];
798};
799
800struct io_uring_clock_register {
801 __u32 clockid;
802 __u32 __resv[3];
803};
804
805enum {
806 IORING_REGISTER_SRC_REGISTERED = (1U << 0),
807 IORING_REGISTER_DST_REPLACE = (1U << 1),
808};
809
810struct io_uring_clone_buffers {
811 __u32 src_fd;
812 __u32 flags;
813 __u32 src_off;
814 __u32 dst_off;
815 __u32 nr;
816 __u32 pad[3];
817};
818
819struct io_uring_buf {
820 __u64 addr;
821 __u32 len;
822 __u16 bid;
823 __u16 resv;
824};
825
826struct io_uring_buf_ring {
827 union {
828 /*
829 * To avoid spilling into more pages than we need to, the
830 * ring tail is overlaid with the io_uring_buf->resv field.
831 */
832 struct {
833 __u64 resv1;
834 __u32 resv2;
835 __u16 resv3;
836 __u16 tail;
837 };
838 __DECLARE_FLEX_ARRAY(struct io_uring_buf, bufs);
839 };
840};
841
842/*
843 * Flags for IORING_REGISTER_PBUF_RING.
844 *
845 * IOU_PBUF_RING_MMAP: If set, kernel will allocate the memory for the ring.
846 * The application must not set a ring_addr in struct
847 * io_uring_buf_reg, instead it must subsequently call
848 * mmap(2) with the offset set as:
849 * IORING_OFF_PBUF_RING | (bgid << IORING_OFF_PBUF_SHIFT)
850 * to get a virtual mapping for the ring.
851 * IOU_PBUF_RING_INC: If set, buffers consumed from this buffer ring can be
852 * consumed incrementally. Normally one (or more) buffers
853 * are fully consumed. With incremental consumptions, it's
854 * feasible to register big ranges of buffers, and each
855 * use of it will consume only as much as it needs. This
856 * requires that both the kernel and application keep
857 * track of where the current read/recv index is at.
858 */
859enum io_uring_register_pbuf_ring_flags {
860 IOU_PBUF_RING_MMAP = 1,
861 IOU_PBUF_RING_INC = 2,
862};
863
864/* argument for IORING_(UN)REGISTER_PBUF_RING */
865struct io_uring_buf_reg {
866 __u64 ring_addr;
867 __u32 ring_entries;
868 __u16 bgid;
869 __u16 flags;
870 __u64 resv[3];
871};
872
873/* argument for IORING_REGISTER_PBUF_STATUS */
874struct io_uring_buf_status {
875 __u32 buf_group; /* input */
876 __u32 head; /* output */
877 __u32 resv[8];
878};
879
880enum io_uring_napi_op {
881 /* register/ungister backward compatible opcode */
882 IO_URING_NAPI_REGISTER_OP = 0,
883
884 /* opcodes to update napi_list when static tracking is used */
885 IO_URING_NAPI_STATIC_ADD_ID = 1,
886 IO_URING_NAPI_STATIC_DEL_ID = 2
887};
888
889enum io_uring_napi_tracking_strategy {
890 /* value must be 0 for backward compatibility */
891 IO_URING_NAPI_TRACKING_DYNAMIC = 0,
892 IO_URING_NAPI_TRACKING_STATIC = 1,
893 IO_URING_NAPI_TRACKING_INACTIVE = 255
894};
895
896/* argument for IORING_(UN)REGISTER_NAPI */
897struct io_uring_napi {
898 __u32 busy_poll_to;
899 __u8 prefer_busy_poll;
900
901 /* a io_uring_napi_op value */
902 __u8 opcode;
903 __u8 pad[2];
904
905 /*
906 * for IO_URING_NAPI_REGISTER_OP, it is a
907 * io_uring_napi_tracking_strategy value.
908 *
909 * for IO_URING_NAPI_STATIC_ADD_ID/IO_URING_NAPI_STATIC_DEL_ID
910 * it is the napi id to add/del from napi_list.
911 */
912 __u32 op_param;
913 __u32 resv;
914};
915
916/*
917 * io_uring_restriction->opcode values
918 */
919enum io_uring_register_restriction_op {
920 /* Allow an io_uring_register(2) opcode */
921 IORING_RESTRICTION_REGISTER_OP = 0,
922
923 /* Allow an sqe opcode */
924 IORING_RESTRICTION_SQE_OP = 1,
925
926 /* Allow sqe flags */
927 IORING_RESTRICTION_SQE_FLAGS_ALLOWED = 2,
928
929 /* Require sqe flags (these flags must be set on each submission) */
930 IORING_RESTRICTION_SQE_FLAGS_REQUIRED = 3,
931
932 IORING_RESTRICTION_LAST
933};
934
935enum {
936 IORING_REG_WAIT_TS = (1U << 0),
937};
938
939/*
940 * Argument for io_uring_enter(2) with
941 * IORING_GETEVENTS | IORING_ENTER_EXT_ARG_REG set, where the actual argument
942 * is an index into a previously registered fixed wait region described by
943 * the below structure.
944 */
945struct io_uring_reg_wait {
946 struct __kernel_timespec ts;
947 __u32 min_wait_usec;
948 __u32 flags;
949 __u64 sigmask;
950 __u32 sigmask_sz;
951 __u32 pad[3];
952 __u64 pad2[2];
953};
954
955/*
956 * Argument for io_uring_enter(2) with IORING_GETEVENTS | IORING_ENTER_EXT_ARG
957 */
958struct io_uring_getevents_arg {
959 __u64 sigmask;
960 __u32 sigmask_sz;
961 __u32 min_wait_usec;
962 __u64 ts;
963};
964
965/*
966 * Argument for IORING_REGISTER_SYNC_CANCEL
967 */
968struct io_uring_sync_cancel_reg {
969 __u64 addr;
970 __s32 fd;
971 __u32 flags;
972 struct __kernel_timespec timeout;
973 __u8 opcode;
974 __u8 pad[7];
975 __u64 pad2[3];
976};
977
978/*
979 * Argument for IORING_REGISTER_FILE_ALLOC_RANGE
980 * The range is specified as [off, off + len)
981 */
982struct io_uring_file_index_range {
983 __u32 off;
984 __u32 len;
985 __u64 resv;
986};
987
988struct io_uring_recvmsg_out {
989 __u32 namelen;
990 __u32 controllen;
991 __u32 payloadlen;
992 __u32 flags;
993};
994
995/*
996 * Argument for IORING_OP_URING_CMD when file is a socket
997 */
998enum io_uring_socket_op {
999 SOCKET_URING_OP_SIOCINQ = 0,
1000 SOCKET_URING_OP_SIOCOUTQ,
1001 SOCKET_URING_OP_GETSOCKOPT,
1002 SOCKET_URING_OP_SETSOCKOPT,
1003 SOCKET_URING_OP_TX_TIMESTAMP,
1004};
1005
1006/*
1007 * SOCKET_URING_OP_TX_TIMESTAMP definitions
1008 */
1009
1010#define IORING_TIMESTAMP_HW_SHIFT 16
1011/* The cqe->flags bit from which the timestamp type is stored */
1012#define IORING_TIMESTAMP_TYPE_SHIFT (IORING_TIMESTAMP_HW_SHIFT + 1)
1013/* The cqe->flags flag signifying whether it's a hardware timestamp */
1014#define IORING_CQE_F_TSTAMP_HW ((__u32)1 << IORING_TIMESTAMP_HW_SHIFT)
1015
1016struct io_timespec {
1017 __u64 tv_sec;
1018 __u64 tv_nsec;
1019};
1020
1021/* Zero copy receive refill queue entry */
1022struct io_uring_zcrx_rqe {
1023 __u64 off;
1024 __u32 len;
1025 __u32 __pad;
1026};
1027
1028struct io_uring_zcrx_cqe {
1029 __u64 off;
1030 __u64 __pad;
1031};
1032
1033/* The bit from which area id is encoded into offsets */
1034#define IORING_ZCRX_AREA_SHIFT 48
1035#define IORING_ZCRX_AREA_MASK (~(((__u64)1 << IORING_ZCRX_AREA_SHIFT) - 1))
1036
1037struct io_uring_zcrx_offsets {
1038 __u32 head;
1039 __u32 tail;
1040 __u32 rqes;
1041 __u32 __resv2;
1042 __u64 __resv[2];
1043};
1044
1045enum io_uring_zcrx_area_flags {
1046 IORING_ZCRX_AREA_DMABUF = 1,
1047};
1048
1049struct io_uring_zcrx_area_reg {
1050 __u64 addr;
1051 __u64 len;
1052 __u64 rq_area_token;
1053 __u32 flags;
1054 __u32 dmabuf_fd;
1055 __u64 __resv2[2];
1056};
1057
1058/*
1059 * Argument for IORING_REGISTER_ZCRX_IFQ
1060 */
1061struct io_uring_zcrx_ifq_reg {
1062 __u32 if_idx;
1063 __u32 if_rxq;
1064 __u32 rq_entries;
1065 __u32 flags;
1066
1067 __u64 area_ptr; /* pointer to struct io_uring_zcrx_area_reg */
1068 __u64 region_ptr; /* struct io_uring_region_desc * */
1069
1070 struct io_uring_zcrx_offsets offsets;
1071 __u32 zcrx_id;
1072 __u32 __resv2;
1073 __u64 __resv[3];
1074};
1075
1076struct io_uring_zcrx_sync_refill {
1077 __u32 zcrx_id;
1078 /* the number of entries to return */
1079 __u32 nr_entries;
1080 /* pointer to an array of struct io_uring_zcrx_rqe */
1081 __u64 rqes;
1082 __u64 __resv[2];
1083};
1084
1085#ifdef __cplusplus
1086}
1087#endif
1088
1089#endif
1090