Linux kernel mirror (for testing)
git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git
kernel
os
linux
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
234/*
235 * Allow both 64b and 128b SQEs. If a 128b SQE is posted, it will have
236 * a 128b opcode.
237 */
238#define IORING_SETUP_SQE_MIXED (1U << 19)
239
240/*
241 * When set, io_uring ignores SQ head and tail and fetches SQEs to submit
242 * starting from index 0 instead from the index stored in the head pointer.
243 * IOW, the user should place all SQE at the beginning of the SQ memory
244 * before issuing a submission syscall.
245 *
246 * It requires IORING_SETUP_NO_SQARRAY and is incompatible with
247 * IORING_SETUP_SQPOLL. The user must also never change the SQ head and tail
248 * values and keep it set to 0. Any other value is undefined behaviour.
249 */
250#define IORING_SETUP_SQ_REWIND (1U << 20)
251
252enum io_uring_op {
253 IORING_OP_NOP,
254 IORING_OP_READV,
255 IORING_OP_WRITEV,
256 IORING_OP_FSYNC,
257 IORING_OP_READ_FIXED,
258 IORING_OP_WRITE_FIXED,
259 IORING_OP_POLL_ADD,
260 IORING_OP_POLL_REMOVE,
261 IORING_OP_SYNC_FILE_RANGE,
262 IORING_OP_SENDMSG,
263 IORING_OP_RECVMSG,
264 IORING_OP_TIMEOUT,
265 IORING_OP_TIMEOUT_REMOVE,
266 IORING_OP_ACCEPT,
267 IORING_OP_ASYNC_CANCEL,
268 IORING_OP_LINK_TIMEOUT,
269 IORING_OP_CONNECT,
270 IORING_OP_FALLOCATE,
271 IORING_OP_OPENAT,
272 IORING_OP_CLOSE,
273 IORING_OP_FILES_UPDATE,
274 IORING_OP_STATX,
275 IORING_OP_READ,
276 IORING_OP_WRITE,
277 IORING_OP_FADVISE,
278 IORING_OP_MADVISE,
279 IORING_OP_SEND,
280 IORING_OP_RECV,
281 IORING_OP_OPENAT2,
282 IORING_OP_EPOLL_CTL,
283 IORING_OP_SPLICE,
284 IORING_OP_PROVIDE_BUFFERS,
285 IORING_OP_REMOVE_BUFFERS,
286 IORING_OP_TEE,
287 IORING_OP_SHUTDOWN,
288 IORING_OP_RENAMEAT,
289 IORING_OP_UNLINKAT,
290 IORING_OP_MKDIRAT,
291 IORING_OP_SYMLINKAT,
292 IORING_OP_LINKAT,
293 IORING_OP_MSG_RING,
294 IORING_OP_FSETXATTR,
295 IORING_OP_SETXATTR,
296 IORING_OP_FGETXATTR,
297 IORING_OP_GETXATTR,
298 IORING_OP_SOCKET,
299 IORING_OP_URING_CMD,
300 IORING_OP_SEND_ZC,
301 IORING_OP_SENDMSG_ZC,
302 IORING_OP_READ_MULTISHOT,
303 IORING_OP_WAITID,
304 IORING_OP_FUTEX_WAIT,
305 IORING_OP_FUTEX_WAKE,
306 IORING_OP_FUTEX_WAITV,
307 IORING_OP_FIXED_FD_INSTALL,
308 IORING_OP_FTRUNCATE,
309 IORING_OP_BIND,
310 IORING_OP_LISTEN,
311 IORING_OP_RECV_ZC,
312 IORING_OP_EPOLL_WAIT,
313 IORING_OP_READV_FIXED,
314 IORING_OP_WRITEV_FIXED,
315 IORING_OP_PIPE,
316 IORING_OP_NOP128,
317 IORING_OP_URING_CMD128,
318
319 /* this goes last, obviously */
320 IORING_OP_LAST,
321};
322
323/*
324 * sqe->uring_cmd_flags top 8bits aren't available for userspace
325 * IORING_URING_CMD_FIXED use registered buffer; pass this flag
326 * along with setting sqe->buf_index.
327 * IORING_URING_CMD_MULTISHOT must be used with buffer select, like other
328 * multishot commands. Not compatible with
329 * IORING_URING_CMD_FIXED, for now.
330 */
331#define IORING_URING_CMD_FIXED (1U << 0)
332#define IORING_URING_CMD_MULTISHOT (1U << 1)
333#define IORING_URING_CMD_MASK (IORING_URING_CMD_FIXED | IORING_URING_CMD_MULTISHOT)
334
335
336/*
337 * sqe->fsync_flags
338 */
339#define IORING_FSYNC_DATASYNC (1U << 0)
340
341/*
342 * sqe->timeout_flags
343 */
344#define IORING_TIMEOUT_ABS (1U << 0)
345#define IORING_TIMEOUT_UPDATE (1U << 1)
346#define IORING_TIMEOUT_BOOTTIME (1U << 2)
347#define IORING_TIMEOUT_REALTIME (1U << 3)
348#define IORING_LINK_TIMEOUT_UPDATE (1U << 4)
349#define IORING_TIMEOUT_ETIME_SUCCESS (1U << 5)
350#define IORING_TIMEOUT_MULTISHOT (1U << 6)
351#define IORING_TIMEOUT_CLOCK_MASK (IORING_TIMEOUT_BOOTTIME | IORING_TIMEOUT_REALTIME)
352#define IORING_TIMEOUT_UPDATE_MASK (IORING_TIMEOUT_UPDATE | IORING_LINK_TIMEOUT_UPDATE)
353/*
354 * sqe->splice_flags
355 * extends splice(2) flags
356 */
357#define SPLICE_F_FD_IN_FIXED (1U << 31) /* the last bit of __u32 */
358
359/*
360 * POLL_ADD flags. Note that since sqe->poll_events is the flag space, the
361 * command flags for POLL_ADD are stored in sqe->len.
362 *
363 * IORING_POLL_ADD_MULTI Multishot poll. Sets IORING_CQE_F_MORE if
364 * the poll handler will continue to report
365 * CQEs on behalf of the same SQE.
366 *
367 * IORING_POLL_UPDATE Update existing poll request, matching
368 * sqe->addr as the old user_data field.
369 *
370 * IORING_POLL_LEVEL Level triggered poll.
371 */
372#define IORING_POLL_ADD_MULTI (1U << 0)
373#define IORING_POLL_UPDATE_EVENTS (1U << 1)
374#define IORING_POLL_UPDATE_USER_DATA (1U << 2)
375#define IORING_POLL_ADD_LEVEL (1U << 3)
376
377/*
378 * ASYNC_CANCEL flags.
379 *
380 * IORING_ASYNC_CANCEL_ALL Cancel all requests that match the given key
381 * IORING_ASYNC_CANCEL_FD Key off 'fd' for cancelation rather than the
382 * request 'user_data'
383 * IORING_ASYNC_CANCEL_ANY Match any request
384 * IORING_ASYNC_CANCEL_FD_FIXED 'fd' passed in is a fixed descriptor
385 * IORING_ASYNC_CANCEL_USERDATA Match on user_data, default for no other key
386 * IORING_ASYNC_CANCEL_OP Match request based on opcode
387 */
388#define IORING_ASYNC_CANCEL_ALL (1U << 0)
389#define IORING_ASYNC_CANCEL_FD (1U << 1)
390#define IORING_ASYNC_CANCEL_ANY (1U << 2)
391#define IORING_ASYNC_CANCEL_FD_FIXED (1U << 3)
392#define IORING_ASYNC_CANCEL_USERDATA (1U << 4)
393#define IORING_ASYNC_CANCEL_OP (1U << 5)
394
395/*
396 * send/sendmsg and recv/recvmsg flags (sqe->ioprio)
397 *
398 * IORING_RECVSEND_POLL_FIRST If set, instead of first attempting to send
399 * or receive and arm poll if that yields an
400 * -EAGAIN result, arm poll upfront and skip
401 * the initial transfer attempt.
402 *
403 * IORING_RECV_MULTISHOT Multishot recv. Sets IORING_CQE_F_MORE if
404 * the handler will continue to report
405 * CQEs on behalf of the same SQE.
406 *
407 * IORING_RECVSEND_FIXED_BUF Use registered buffers, the index is stored in
408 * the buf_index field.
409 *
410 * IORING_SEND_ZC_REPORT_USAGE
411 * If set, SEND[MSG]_ZC should report
412 * the zerocopy usage in cqe.res
413 * for the IORING_CQE_F_NOTIF cqe.
414 * 0 is reported if zerocopy was actually possible.
415 * IORING_NOTIF_USAGE_ZC_COPIED if data was copied
416 * (at least partially).
417 *
418 * IORING_RECVSEND_BUNDLE Used with IOSQE_BUFFER_SELECT. If set, send or
419 * recv will grab as many buffers from the buffer
420 * group ID given and send them all. The completion
421 * result will be the number of buffers send, with
422 * the starting buffer ID in cqe->flags as per
423 * usual for provided buffer usage. The buffers
424 * will be contiguous from the starting buffer ID.
425 *
426 * IORING_SEND_VECTORIZED If set, SEND[_ZC] will take a pointer to a io_vec
427 * to allow vectorized send operations.
428 */
429#define IORING_RECVSEND_POLL_FIRST (1U << 0)
430#define IORING_RECV_MULTISHOT (1U << 1)
431#define IORING_RECVSEND_FIXED_BUF (1U << 2)
432#define IORING_SEND_ZC_REPORT_USAGE (1U << 3)
433#define IORING_RECVSEND_BUNDLE (1U << 4)
434#define IORING_SEND_VECTORIZED (1U << 5)
435
436/*
437 * cqe.res for IORING_CQE_F_NOTIF if
438 * IORING_SEND_ZC_REPORT_USAGE was requested
439 *
440 * It should be treated as a flag, all other
441 * bits of cqe.res should be treated as reserved!
442 */
443#define IORING_NOTIF_USAGE_ZC_COPIED (1U << 31)
444
445/*
446 * accept flags stored in sqe->ioprio
447 */
448#define IORING_ACCEPT_MULTISHOT (1U << 0)
449#define IORING_ACCEPT_DONTWAIT (1U << 1)
450#define IORING_ACCEPT_POLL_FIRST (1U << 2)
451
452/*
453 * IORING_OP_MSG_RING command types, stored in sqe->addr
454 */
455enum io_uring_msg_ring_flags {
456 IORING_MSG_DATA, /* pass sqe->len as 'res' and off as user_data */
457 IORING_MSG_SEND_FD, /* send a registered fd to another ring */
458};
459
460/*
461 * IORING_OP_MSG_RING flags (sqe->msg_ring_flags)
462 *
463 * IORING_MSG_RING_CQE_SKIP Don't post a CQE to the target ring. Not
464 * applicable for IORING_MSG_DATA, obviously.
465 */
466#define IORING_MSG_RING_CQE_SKIP (1U << 0)
467/* Pass through the flags from sqe->file_index to cqe->flags */
468#define IORING_MSG_RING_FLAGS_PASS (1U << 1)
469
470/*
471 * IORING_OP_FIXED_FD_INSTALL flags (sqe->install_fd_flags)
472 *
473 * IORING_FIXED_FD_NO_CLOEXEC Don't mark the fd as O_CLOEXEC
474 */
475#define IORING_FIXED_FD_NO_CLOEXEC (1U << 0)
476
477/*
478 * IORING_OP_NOP flags (sqe->nop_flags)
479 *
480 * IORING_NOP_INJECT_RESULT Inject result from sqe->result
481 */
482#define IORING_NOP_INJECT_RESULT (1U << 0)
483#define IORING_NOP_FILE (1U << 1)
484#define IORING_NOP_FIXED_FILE (1U << 2)
485#define IORING_NOP_FIXED_BUFFER (1U << 3)
486#define IORING_NOP_TW (1U << 4)
487#define IORING_NOP_CQE32 (1U << 5)
488
489/*
490 * IO completion data structure (Completion Queue Entry)
491 */
492struct io_uring_cqe {
493 __u64 user_data; /* sqe->user_data value passed back */
494 __s32 res; /* result code for this event */
495 __u32 flags;
496
497 /*
498 * If the ring is initialized with IORING_SETUP_CQE32, then this field
499 * contains 16-bytes of padding, doubling the size of the CQE.
500 */
501 __u64 big_cqe[];
502};
503
504/*
505 * cqe->flags
506 *
507 * IORING_CQE_F_BUFFER If set, the upper 16 bits are the buffer ID
508 * IORING_CQE_F_MORE If set, parent SQE will generate more CQE entries
509 * IORING_CQE_F_SOCK_NONEMPTY If set, more data to read after socket recv
510 * IORING_CQE_F_NOTIF Set for notification CQEs. Can be used to distinct
511 * them from sends.
512 * IORING_CQE_F_BUF_MORE If set, the buffer ID set in the completion will get
513 * more completions. In other words, the buffer is being
514 * partially consumed, and will be used by the kernel for
515 * more completions. This is only set for buffers used via
516 * the incremental buffer consumption, as provided by
517 * a ring buffer setup with IOU_PBUF_RING_INC. For any
518 * other provided buffer type, all completions with a
519 * buffer passed back is automatically returned to the
520 * application.
521 * IORING_CQE_F_SKIP If set, then the application/liburing must ignore this
522 * CQE. It's only purpose is to fill a gap in the ring,
523 * if a large CQE is attempted posted when the ring has
524 * just a single small CQE worth of space left before
525 * wrapping.
526 * IORING_CQE_F_32 If set, this is a 32b/big-cqe posting. Use with rings
527 * setup in a mixed CQE mode, where both 16b and 32b
528 * CQEs may be posted to the CQ ring.
529 */
530#define IORING_CQE_F_BUFFER (1U << 0)
531#define IORING_CQE_F_MORE (1U << 1)
532#define IORING_CQE_F_SOCK_NONEMPTY (1U << 2)
533#define IORING_CQE_F_NOTIF (1U << 3)
534#define IORING_CQE_F_BUF_MORE (1U << 4)
535#define IORING_CQE_F_SKIP (1U << 5)
536#define IORING_CQE_F_32 (1U << 15)
537
538#define IORING_CQE_BUFFER_SHIFT 16
539
540/*
541 * Magic offsets for the application to mmap the data it needs
542 */
543#define IORING_OFF_SQ_RING 0ULL
544#define IORING_OFF_CQ_RING 0x8000000ULL
545#define IORING_OFF_SQES 0x10000000ULL
546#define IORING_OFF_PBUF_RING 0x80000000ULL
547#define IORING_OFF_PBUF_SHIFT 16
548#define IORING_OFF_MMAP_MASK 0xf8000000ULL
549
550/*
551 * Filled with the offset for mmap(2)
552 */
553struct io_sqring_offsets {
554 __u32 head;
555 __u32 tail;
556 __u32 ring_mask;
557 __u32 ring_entries;
558 __u32 flags;
559 __u32 dropped;
560 __u32 array;
561 __u32 resv1;
562 __u64 user_addr;
563};
564
565/*
566 * sq_ring->flags
567 */
568#define IORING_SQ_NEED_WAKEUP (1U << 0) /* needs io_uring_enter wakeup */
569#define IORING_SQ_CQ_OVERFLOW (1U << 1) /* CQ ring is overflown */
570#define IORING_SQ_TASKRUN (1U << 2) /* task should enter the kernel */
571
572struct io_cqring_offsets {
573 __u32 head;
574 __u32 tail;
575 __u32 ring_mask;
576 __u32 ring_entries;
577 __u32 overflow;
578 __u32 cqes;
579 __u32 flags;
580 __u32 resv1;
581 __u64 user_addr;
582};
583
584/*
585 * cq_ring->flags
586 */
587
588/* disable eventfd notifications */
589#define IORING_CQ_EVENTFD_DISABLED (1U << 0)
590
591/*
592 * io_uring_enter(2) flags
593 */
594#define IORING_ENTER_GETEVENTS (1U << 0)
595#define IORING_ENTER_SQ_WAKEUP (1U << 1)
596#define IORING_ENTER_SQ_WAIT (1U << 2)
597#define IORING_ENTER_EXT_ARG (1U << 3)
598#define IORING_ENTER_REGISTERED_RING (1U << 4)
599#define IORING_ENTER_ABS_TIMER (1U << 5)
600#define IORING_ENTER_EXT_ARG_REG (1U << 6)
601#define IORING_ENTER_NO_IOWAIT (1U << 7)
602
603/*
604 * Passed in for io_uring_setup(2). Copied back with updated info on success
605 */
606struct io_uring_params {
607 __u32 sq_entries;
608 __u32 cq_entries;
609 __u32 flags;
610 __u32 sq_thread_cpu;
611 __u32 sq_thread_idle;
612 __u32 features;
613 __u32 wq_fd;
614 __u32 resv[3];
615 struct io_sqring_offsets sq_off;
616 struct io_cqring_offsets cq_off;
617};
618
619/*
620 * io_uring_params->features flags
621 */
622#define IORING_FEAT_SINGLE_MMAP (1U << 0)
623#define IORING_FEAT_NODROP (1U << 1)
624#define IORING_FEAT_SUBMIT_STABLE (1U << 2)
625#define IORING_FEAT_RW_CUR_POS (1U << 3)
626#define IORING_FEAT_CUR_PERSONALITY (1U << 4)
627#define IORING_FEAT_FAST_POLL (1U << 5)
628#define IORING_FEAT_POLL_32BITS (1U << 6)
629#define IORING_FEAT_SQPOLL_NONFIXED (1U << 7)
630#define IORING_FEAT_EXT_ARG (1U << 8)
631#define IORING_FEAT_NATIVE_WORKERS (1U << 9)
632#define IORING_FEAT_RSRC_TAGS (1U << 10)
633#define IORING_FEAT_CQE_SKIP (1U << 11)
634#define IORING_FEAT_LINKED_FILE (1U << 12)
635#define IORING_FEAT_REG_REG_RING (1U << 13)
636#define IORING_FEAT_RECVSEND_BUNDLE (1U << 14)
637#define IORING_FEAT_MIN_TIMEOUT (1U << 15)
638#define IORING_FEAT_RW_ATTR (1U << 16)
639#define IORING_FEAT_NO_IOWAIT (1U << 17)
640
641/*
642 * io_uring_register(2) opcodes and arguments
643 */
644enum io_uring_register_op {
645 IORING_REGISTER_BUFFERS = 0,
646 IORING_UNREGISTER_BUFFERS = 1,
647 IORING_REGISTER_FILES = 2,
648 IORING_UNREGISTER_FILES = 3,
649 IORING_REGISTER_EVENTFD = 4,
650 IORING_UNREGISTER_EVENTFD = 5,
651 IORING_REGISTER_FILES_UPDATE = 6,
652 IORING_REGISTER_EVENTFD_ASYNC = 7,
653 IORING_REGISTER_PROBE = 8,
654 IORING_REGISTER_PERSONALITY = 9,
655 IORING_UNREGISTER_PERSONALITY = 10,
656 IORING_REGISTER_RESTRICTIONS = 11,
657 IORING_REGISTER_ENABLE_RINGS = 12,
658
659 /* extended with tagging */
660 IORING_REGISTER_FILES2 = 13,
661 IORING_REGISTER_FILES_UPDATE2 = 14,
662 IORING_REGISTER_BUFFERS2 = 15,
663 IORING_REGISTER_BUFFERS_UPDATE = 16,
664
665 /* set/clear io-wq thread affinities */
666 IORING_REGISTER_IOWQ_AFF = 17,
667 IORING_UNREGISTER_IOWQ_AFF = 18,
668
669 /* set/get max number of io-wq workers */
670 IORING_REGISTER_IOWQ_MAX_WORKERS = 19,
671
672 /* register/unregister io_uring fd with the ring */
673 IORING_REGISTER_RING_FDS = 20,
674 IORING_UNREGISTER_RING_FDS = 21,
675
676 /* register ring based provide buffer group */
677 IORING_REGISTER_PBUF_RING = 22,
678 IORING_UNREGISTER_PBUF_RING = 23,
679
680 /* sync cancelation API */
681 IORING_REGISTER_SYNC_CANCEL = 24,
682
683 /* register a range of fixed file slots for automatic slot allocation */
684 IORING_REGISTER_FILE_ALLOC_RANGE = 25,
685
686 /* return status information for a buffer group */
687 IORING_REGISTER_PBUF_STATUS = 26,
688
689 /* set/clear busy poll settings */
690 IORING_REGISTER_NAPI = 27,
691 IORING_UNREGISTER_NAPI = 28,
692
693 IORING_REGISTER_CLOCK = 29,
694
695 /* clone registered buffers from source ring to current ring */
696 IORING_REGISTER_CLONE_BUFFERS = 30,
697
698 /* send MSG_RING without having a ring */
699 IORING_REGISTER_SEND_MSG_RING = 31,
700
701 /* register a netdev hw rx queue for zerocopy */
702 IORING_REGISTER_ZCRX_IFQ = 32,
703
704 /* resize CQ ring */
705 IORING_REGISTER_RESIZE_RINGS = 33,
706
707 IORING_REGISTER_MEM_REGION = 34,
708
709 /* query various aspects of io_uring, see linux/io_uring/query.h */
710 IORING_REGISTER_QUERY = 35,
711
712 /* auxiliary zcrx configuration, see enum zcrx_ctrl_op */
713 IORING_REGISTER_ZCRX_CTRL = 36,
714
715 /* register bpf filtering programs */
716 IORING_REGISTER_BPF_FILTER = 37,
717
718 /* this goes last */
719 IORING_REGISTER_LAST,
720
721 /* flag added to the opcode to use a registered ring fd */
722 IORING_REGISTER_USE_REGISTERED_RING = 1U << 31
723};
724
725/* io-wq worker categories */
726enum io_wq_type {
727 IO_WQ_BOUND,
728 IO_WQ_UNBOUND,
729};
730
731/* deprecated, see struct io_uring_rsrc_update */
732struct io_uring_files_update {
733 __u32 offset;
734 __u32 resv;
735 __aligned_u64 /* __s32 * */ fds;
736};
737
738enum {
739 /* initialise with user provided memory pointed by user_addr */
740 IORING_MEM_REGION_TYPE_USER = 1,
741};
742
743struct io_uring_region_desc {
744 __u64 user_addr;
745 __u64 size;
746 __u32 flags;
747 __u32 id;
748 __u64 mmap_offset;
749 __u64 __resv[4];
750};
751
752enum {
753 /* expose the region as registered wait arguments */
754 IORING_MEM_REGION_REG_WAIT_ARG = 1,
755};
756
757struct io_uring_mem_region_reg {
758 __u64 region_uptr; /* struct io_uring_region_desc * */
759 __u64 flags;
760 __u64 __resv[2];
761};
762
763/*
764 * Register a fully sparse file space, rather than pass in an array of all
765 * -1 file descriptors.
766 */
767#define IORING_RSRC_REGISTER_SPARSE (1U << 0)
768
769struct io_uring_rsrc_register {
770 __u32 nr;
771 __u32 flags;
772 __u64 resv2;
773 __aligned_u64 data;
774 __aligned_u64 tags;
775};
776
777struct io_uring_rsrc_update {
778 __u32 offset;
779 __u32 resv;
780 __aligned_u64 data;
781};
782
783struct io_uring_rsrc_update2 {
784 __u32 offset;
785 __u32 resv;
786 __aligned_u64 data;
787 __aligned_u64 tags;
788 __u32 nr;
789 __u32 resv2;
790};
791
792/* Skip updating fd indexes set to this value in the fd table */
793#define IORING_REGISTER_FILES_SKIP (-2)
794
795#define IO_URING_OP_SUPPORTED (1U << 0)
796
797struct io_uring_probe_op {
798 __u8 op;
799 __u8 resv;
800 __u16 flags; /* IO_URING_OP_* flags */
801 __u32 resv2;
802};
803
804struct io_uring_probe {
805 __u8 last_op; /* last opcode supported */
806 __u8 ops_len; /* length of ops[] array below */
807 __u16 resv;
808 __u32 resv2[3];
809 struct io_uring_probe_op ops[];
810};
811
812struct io_uring_restriction {
813 __u16 opcode;
814 union {
815 __u8 register_op; /* IORING_RESTRICTION_REGISTER_OP */
816 __u8 sqe_op; /* IORING_RESTRICTION_SQE_OP */
817 __u8 sqe_flags; /* IORING_RESTRICTION_SQE_FLAGS_* */
818 };
819 __u8 resv;
820 __u32 resv2[3];
821};
822
823struct io_uring_task_restriction {
824 __u16 flags;
825 __u16 nr_res;
826 __u32 resv[3];
827 __DECLARE_FLEX_ARRAY(struct io_uring_restriction, restrictions);
828};
829
830struct io_uring_clock_register {
831 __u32 clockid;
832 __u32 __resv[3];
833};
834
835enum {
836 IORING_REGISTER_SRC_REGISTERED = (1U << 0),
837 IORING_REGISTER_DST_REPLACE = (1U << 1),
838};
839
840struct io_uring_clone_buffers {
841 __u32 src_fd;
842 __u32 flags;
843 __u32 src_off;
844 __u32 dst_off;
845 __u32 nr;
846 __u32 pad[3];
847};
848
849struct io_uring_buf {
850 __u64 addr;
851 __u32 len;
852 __u16 bid;
853 __u16 resv;
854};
855
856struct io_uring_buf_ring {
857 union {
858 /*
859 * To avoid spilling into more pages than we need to, the
860 * ring tail is overlaid with the io_uring_buf->resv field.
861 */
862 struct {
863 __u64 resv1;
864 __u32 resv2;
865 __u16 resv3;
866 __u16 tail;
867 };
868 __DECLARE_FLEX_ARRAY(struct io_uring_buf, bufs);
869 };
870};
871
872/*
873 * Flags for IORING_REGISTER_PBUF_RING.
874 *
875 * IOU_PBUF_RING_MMAP: If set, kernel will allocate the memory for the ring.
876 * The application must not set a ring_addr in struct
877 * io_uring_buf_reg, instead it must subsequently call
878 * mmap(2) with the offset set as:
879 * IORING_OFF_PBUF_RING | (bgid << IORING_OFF_PBUF_SHIFT)
880 * to get a virtual mapping for the ring.
881 * IOU_PBUF_RING_INC: If set, buffers consumed from this buffer ring can be
882 * consumed incrementally. Normally one (or more) buffers
883 * are fully consumed. With incremental consumptions, it's
884 * feasible to register big ranges of buffers, and each
885 * use of it will consume only as much as it needs. This
886 * requires that both the kernel and application keep
887 * track of where the current read/recv index is at.
888 */
889enum io_uring_register_pbuf_ring_flags {
890 IOU_PBUF_RING_MMAP = 1,
891 IOU_PBUF_RING_INC = 2,
892};
893
894/* argument for IORING_(UN)REGISTER_PBUF_RING */
895struct io_uring_buf_reg {
896 __u64 ring_addr;
897 __u32 ring_entries;
898 __u16 bgid;
899 __u16 flags;
900 __u64 resv[3];
901};
902
903/* argument for IORING_REGISTER_PBUF_STATUS */
904struct io_uring_buf_status {
905 __u32 buf_group; /* input */
906 __u32 head; /* output */
907 __u32 resv[8];
908};
909
910enum io_uring_napi_op {
911 /* register/ungister backward compatible opcode */
912 IO_URING_NAPI_REGISTER_OP = 0,
913
914 /* opcodes to update napi_list when static tracking is used */
915 IO_URING_NAPI_STATIC_ADD_ID = 1,
916 IO_URING_NAPI_STATIC_DEL_ID = 2
917};
918
919enum io_uring_napi_tracking_strategy {
920 /* value must be 0 for backward compatibility */
921 IO_URING_NAPI_TRACKING_DYNAMIC = 0,
922 IO_URING_NAPI_TRACKING_STATIC = 1,
923 IO_URING_NAPI_TRACKING_INACTIVE = 255
924};
925
926/* argument for IORING_(UN)REGISTER_NAPI */
927struct io_uring_napi {
928 __u32 busy_poll_to;
929 __u8 prefer_busy_poll;
930
931 /* a io_uring_napi_op value */
932 __u8 opcode;
933 __u8 pad[2];
934
935 /*
936 * for IO_URING_NAPI_REGISTER_OP, it is a
937 * io_uring_napi_tracking_strategy value.
938 *
939 * for IO_URING_NAPI_STATIC_ADD_ID/IO_URING_NAPI_STATIC_DEL_ID
940 * it is the napi id to add/del from napi_list.
941 */
942 __u32 op_param;
943 __u32 resv;
944};
945
946/*
947 * io_uring_restriction->opcode values
948 */
949enum io_uring_register_restriction_op {
950 /* Allow an io_uring_register(2) opcode */
951 IORING_RESTRICTION_REGISTER_OP = 0,
952
953 /* Allow an sqe opcode */
954 IORING_RESTRICTION_SQE_OP = 1,
955
956 /* Allow sqe flags */
957 IORING_RESTRICTION_SQE_FLAGS_ALLOWED = 2,
958
959 /* Require sqe flags (these flags must be set on each submission) */
960 IORING_RESTRICTION_SQE_FLAGS_REQUIRED = 3,
961
962 IORING_RESTRICTION_LAST
963};
964
965enum {
966 IORING_REG_WAIT_TS = (1U << 0),
967};
968
969/*
970 * Argument for io_uring_enter(2) with
971 * IORING_GETEVENTS | IORING_ENTER_EXT_ARG_REG set, where the actual argument
972 * is an index into a previously registered fixed wait region described by
973 * the below structure.
974 */
975struct io_uring_reg_wait {
976 struct __kernel_timespec ts;
977 __u32 min_wait_usec;
978 __u32 flags;
979 __u64 sigmask;
980 __u32 sigmask_sz;
981 __u32 pad[3];
982 __u64 pad2[2];
983};
984
985/*
986 * Argument for io_uring_enter(2) with IORING_GETEVENTS | IORING_ENTER_EXT_ARG
987 */
988struct io_uring_getevents_arg {
989 __u64 sigmask;
990 __u32 sigmask_sz;
991 __u32 min_wait_usec;
992 __u64 ts;
993};
994
995/*
996 * Argument for IORING_REGISTER_SYNC_CANCEL
997 */
998struct io_uring_sync_cancel_reg {
999 __u64 addr;
1000 __s32 fd;
1001 __u32 flags;
1002 struct __kernel_timespec timeout;
1003 __u8 opcode;
1004 __u8 pad[7];
1005 __u64 pad2[3];
1006};
1007
1008/*
1009 * Argument for IORING_REGISTER_FILE_ALLOC_RANGE
1010 * The range is specified as [off, off + len)
1011 */
1012struct io_uring_file_index_range {
1013 __u32 off;
1014 __u32 len;
1015 __u64 resv;
1016};
1017
1018struct io_uring_recvmsg_out {
1019 __u32 namelen;
1020 __u32 controllen;
1021 __u32 payloadlen;
1022 __u32 flags;
1023};
1024
1025/*
1026 * Argument for IORING_OP_URING_CMD when file is a socket
1027 */
1028enum io_uring_socket_op {
1029 SOCKET_URING_OP_SIOCINQ = 0,
1030 SOCKET_URING_OP_SIOCOUTQ,
1031 SOCKET_URING_OP_GETSOCKOPT,
1032 SOCKET_URING_OP_SETSOCKOPT,
1033 SOCKET_URING_OP_TX_TIMESTAMP,
1034 SOCKET_URING_OP_GETSOCKNAME,
1035};
1036
1037/*
1038 * SOCKET_URING_OP_TX_TIMESTAMP definitions
1039 */
1040
1041#define IORING_TIMESTAMP_HW_SHIFT 16
1042/* The cqe->flags bit from which the timestamp type is stored */
1043#define IORING_TIMESTAMP_TYPE_SHIFT (IORING_TIMESTAMP_HW_SHIFT + 1)
1044/* The cqe->flags flag signifying whether it's a hardware timestamp */
1045#define IORING_CQE_F_TSTAMP_HW ((__u32)1 << IORING_TIMESTAMP_HW_SHIFT)
1046
1047struct io_timespec {
1048 __u64 tv_sec;
1049 __u64 tv_nsec;
1050};
1051
1052/* Zero copy receive refill queue entry */
1053struct io_uring_zcrx_rqe {
1054 __u64 off;
1055 __u32 len;
1056 __u32 __pad;
1057};
1058
1059struct io_uring_zcrx_cqe {
1060 __u64 off;
1061 __u64 __pad;
1062};
1063
1064/* The bit from which area id is encoded into offsets */
1065#define IORING_ZCRX_AREA_SHIFT 48
1066#define IORING_ZCRX_AREA_MASK (~(((__u64)1 << IORING_ZCRX_AREA_SHIFT) - 1))
1067
1068struct io_uring_zcrx_offsets {
1069 __u32 head;
1070 __u32 tail;
1071 __u32 rqes;
1072 __u32 __resv2;
1073 __u64 __resv[2];
1074};
1075
1076enum io_uring_zcrx_area_flags {
1077 IORING_ZCRX_AREA_DMABUF = 1,
1078};
1079
1080struct io_uring_zcrx_area_reg {
1081 __u64 addr;
1082 __u64 len;
1083 __u64 rq_area_token;
1084 __u32 flags;
1085 __u32 dmabuf_fd;
1086 __u64 __resv2[2];
1087};
1088
1089enum zcrx_reg_flags {
1090 ZCRX_REG_IMPORT = 1,
1091};
1092
1093enum zcrx_features {
1094 /*
1095 * The user can ask for the desired rx page size by passing the
1096 * value in struct io_uring_zcrx_ifq_reg::rx_buf_len.
1097 */
1098 ZCRX_FEATURE_RX_PAGE_SIZE = 1 << 0,
1099};
1100
1101/*
1102 * Argument for IORING_REGISTER_ZCRX_IFQ
1103 */
1104struct io_uring_zcrx_ifq_reg {
1105 __u32 if_idx;
1106 __u32 if_rxq;
1107 __u32 rq_entries;
1108 __u32 flags;
1109
1110 __u64 area_ptr; /* pointer to struct io_uring_zcrx_area_reg */
1111 __u64 region_ptr; /* struct io_uring_region_desc * */
1112
1113 struct io_uring_zcrx_offsets offsets;
1114 __u32 zcrx_id;
1115 __u32 rx_buf_len;
1116 __u64 __resv[3];
1117};
1118
1119enum zcrx_ctrl_op {
1120 ZCRX_CTRL_FLUSH_RQ,
1121 ZCRX_CTRL_EXPORT,
1122
1123 __ZCRX_CTRL_LAST,
1124};
1125
1126struct zcrx_ctrl_flush_rq {
1127 __u64 __resv[6];
1128};
1129
1130struct zcrx_ctrl_export {
1131 __u32 zcrx_fd;
1132 __u32 __resv1[11];
1133};
1134
1135struct zcrx_ctrl {
1136 __u32 zcrx_id;
1137 __u32 op; /* see enum zcrx_ctrl_op */
1138 __u64 __resv[2];
1139
1140 union {
1141 struct zcrx_ctrl_export zc_export;
1142 struct zcrx_ctrl_flush_rq zc_flush;
1143 };
1144};
1145
1146#ifdef __cplusplus
1147}
1148#endif
1149
1150#endif