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 };
47 __u32 len; /* buffer size or number of iovecs */
48 union {
49 __kernel_rwf_t rw_flags;
50 __u32 fsync_flags;
51 __u16 poll_events; /* compatibility */
52 __u32 poll32_events; /* word-reversed for BE */
53 __u32 sync_range_flags;
54 __u32 msg_flags;
55 __u32 timeout_flags;
56 __u32 accept_flags;
57 __u32 cancel_flags;
58 __u32 open_flags;
59 __u32 statx_flags;
60 __u32 fadvise_advice;
61 __u32 splice_flags;
62 __u32 rename_flags;
63 __u32 unlink_flags;
64 __u32 hardlink_flags;
65 __u32 xattr_flags;
66 __u32 msg_ring_flags;
67 __u32 uring_cmd_flags;
68 };
69 __u64 user_data; /* data to be passed back at completion time */
70 /* pack this to avoid bogus arm OABI complaints */
71 union {
72 /* index into fixed buffers, if used */
73 __u16 buf_index;
74 /* for grouped buffer selection */
75 __u16 buf_group;
76 } __attribute__((packed));
77 /* personality to use, if used */
78 __u16 personality;
79 union {
80 __s32 splice_fd_in;
81 __u32 file_index;
82 struct {
83 __u16 addr_len;
84 __u16 __pad3[1];
85 };
86 };
87 union {
88 struct {
89 __u64 addr3;
90 __u64 __pad2[1];
91 };
92 /*
93 * If the ring is initialized with IORING_SETUP_SQE128, then
94 * this field is used for 80 bytes of arbitrary command data
95 */
96 __u8 cmd[0];
97 };
98};
99
100/*
101 * If sqe->file_index is set to this for opcodes that instantiate a new
102 * direct descriptor (like openat/openat2/accept), then io_uring will allocate
103 * an available direct descriptor instead of having the application pass one
104 * in. The picked direct descriptor will be returned in cqe->res, or -ENFILE
105 * if the space is full.
106 */
107#define IORING_FILE_INDEX_ALLOC (~0U)
108
109enum {
110 IOSQE_FIXED_FILE_BIT,
111 IOSQE_IO_DRAIN_BIT,
112 IOSQE_IO_LINK_BIT,
113 IOSQE_IO_HARDLINK_BIT,
114 IOSQE_ASYNC_BIT,
115 IOSQE_BUFFER_SELECT_BIT,
116 IOSQE_CQE_SKIP_SUCCESS_BIT,
117};
118
119/*
120 * sqe->flags
121 */
122/* use fixed fileset */
123#define IOSQE_FIXED_FILE (1U << IOSQE_FIXED_FILE_BIT)
124/* issue after inflight IO */
125#define IOSQE_IO_DRAIN (1U << IOSQE_IO_DRAIN_BIT)
126/* links next sqe */
127#define IOSQE_IO_LINK (1U << IOSQE_IO_LINK_BIT)
128/* like LINK, but stronger */
129#define IOSQE_IO_HARDLINK (1U << IOSQE_IO_HARDLINK_BIT)
130/* always go async */
131#define IOSQE_ASYNC (1U << IOSQE_ASYNC_BIT)
132/* select buffer from sqe->buf_group */
133#define IOSQE_BUFFER_SELECT (1U << IOSQE_BUFFER_SELECT_BIT)
134/* don't post CQE if request succeeded */
135#define IOSQE_CQE_SKIP_SUCCESS (1U << IOSQE_CQE_SKIP_SUCCESS_BIT)
136
137/*
138 * io_uring_setup() flags
139 */
140#define IORING_SETUP_IOPOLL (1U << 0) /* io_context is polled */
141#define IORING_SETUP_SQPOLL (1U << 1) /* SQ poll thread */
142#define IORING_SETUP_SQ_AFF (1U << 2) /* sq_thread_cpu is valid */
143#define IORING_SETUP_CQSIZE (1U << 3) /* app defines CQ size */
144#define IORING_SETUP_CLAMP (1U << 4) /* clamp SQ/CQ ring sizes */
145#define IORING_SETUP_ATTACH_WQ (1U << 5) /* attach to existing wq */
146#define IORING_SETUP_R_DISABLED (1U << 6) /* start with ring disabled */
147#define IORING_SETUP_SUBMIT_ALL (1U << 7) /* continue submit on error */
148/*
149 * Cooperative task running. When requests complete, they often require
150 * forcing the submitter to transition to the kernel to complete. If this
151 * flag is set, work will be done when the task transitions anyway, rather
152 * than force an inter-processor interrupt reschedule. This avoids interrupting
153 * a task running in userspace, and saves an IPI.
154 */
155#define IORING_SETUP_COOP_TASKRUN (1U << 8)
156/*
157 * If COOP_TASKRUN is set, get notified if task work is available for
158 * running and a kernel transition would be needed to run it. This sets
159 * IORING_SQ_TASKRUN in the sq ring flags. Not valid with COOP_TASKRUN.
160 */
161#define IORING_SETUP_TASKRUN_FLAG (1U << 9)
162#define IORING_SETUP_SQE128 (1U << 10) /* SQEs are 128 byte */
163#define IORING_SETUP_CQE32 (1U << 11) /* CQEs are 32 byte */
164/*
165 * Only one task is allowed to submit requests
166 */
167#define IORING_SETUP_SINGLE_ISSUER (1U << 12)
168
169/*
170 * Defer running task work to get events.
171 * Rather than running bits of task work whenever the task transitions
172 * try to do it just before it is needed.
173 */
174#define IORING_SETUP_DEFER_TASKRUN (1U << 13)
175
176/*
177 * Application provides the memory for the rings
178 */
179#define IORING_SETUP_NO_MMAP (1U << 14)
180
181/*
182 * Register the ring fd in itself for use with
183 * IORING_REGISTER_USE_REGISTERED_RING; return a registered fd index rather
184 * than an fd.
185 */
186#define IORING_SETUP_REGISTERED_FD_ONLY (1U << 15)
187
188/*
189 * Removes indirection through the SQ index array.
190 */
191#define IORING_SETUP_NO_SQARRAY (1U << 16)
192
193enum io_uring_op {
194 IORING_OP_NOP,
195 IORING_OP_READV,
196 IORING_OP_WRITEV,
197 IORING_OP_FSYNC,
198 IORING_OP_READ_FIXED,
199 IORING_OP_WRITE_FIXED,
200 IORING_OP_POLL_ADD,
201 IORING_OP_POLL_REMOVE,
202 IORING_OP_SYNC_FILE_RANGE,
203 IORING_OP_SENDMSG,
204 IORING_OP_RECVMSG,
205 IORING_OP_TIMEOUT,
206 IORING_OP_TIMEOUT_REMOVE,
207 IORING_OP_ACCEPT,
208 IORING_OP_ASYNC_CANCEL,
209 IORING_OP_LINK_TIMEOUT,
210 IORING_OP_CONNECT,
211 IORING_OP_FALLOCATE,
212 IORING_OP_OPENAT,
213 IORING_OP_CLOSE,
214 IORING_OP_FILES_UPDATE,
215 IORING_OP_STATX,
216 IORING_OP_READ,
217 IORING_OP_WRITE,
218 IORING_OP_FADVISE,
219 IORING_OP_MADVISE,
220 IORING_OP_SEND,
221 IORING_OP_RECV,
222 IORING_OP_OPENAT2,
223 IORING_OP_EPOLL_CTL,
224 IORING_OP_SPLICE,
225 IORING_OP_PROVIDE_BUFFERS,
226 IORING_OP_REMOVE_BUFFERS,
227 IORING_OP_TEE,
228 IORING_OP_SHUTDOWN,
229 IORING_OP_RENAMEAT,
230 IORING_OP_UNLINKAT,
231 IORING_OP_MKDIRAT,
232 IORING_OP_SYMLINKAT,
233 IORING_OP_LINKAT,
234 IORING_OP_MSG_RING,
235 IORING_OP_FSETXATTR,
236 IORING_OP_SETXATTR,
237 IORING_OP_FGETXATTR,
238 IORING_OP_GETXATTR,
239 IORING_OP_SOCKET,
240 IORING_OP_URING_CMD,
241 IORING_OP_SEND_ZC,
242 IORING_OP_SENDMSG_ZC,
243
244 /* this goes last, obviously */
245 IORING_OP_LAST,
246};
247
248/*
249 * sqe->uring_cmd_flags
250 * IORING_URING_CMD_FIXED use registered buffer; pass this flag
251 * along with setting sqe->buf_index.
252 * IORING_URING_CMD_POLLED driver use only
253 */
254#define IORING_URING_CMD_FIXED (1U << 0)
255#define IORING_URING_CMD_POLLED (1U << 31)
256
257
258/*
259 * sqe->fsync_flags
260 */
261#define IORING_FSYNC_DATASYNC (1U << 0)
262
263/*
264 * sqe->timeout_flags
265 */
266#define IORING_TIMEOUT_ABS (1U << 0)
267#define IORING_TIMEOUT_UPDATE (1U << 1)
268#define IORING_TIMEOUT_BOOTTIME (1U << 2)
269#define IORING_TIMEOUT_REALTIME (1U << 3)
270#define IORING_LINK_TIMEOUT_UPDATE (1U << 4)
271#define IORING_TIMEOUT_ETIME_SUCCESS (1U << 5)
272#define IORING_TIMEOUT_MULTISHOT (1U << 6)
273#define IORING_TIMEOUT_CLOCK_MASK (IORING_TIMEOUT_BOOTTIME | IORING_TIMEOUT_REALTIME)
274#define IORING_TIMEOUT_UPDATE_MASK (IORING_TIMEOUT_UPDATE | IORING_LINK_TIMEOUT_UPDATE)
275/*
276 * sqe->splice_flags
277 * extends splice(2) flags
278 */
279#define SPLICE_F_FD_IN_FIXED (1U << 31) /* the last bit of __u32 */
280
281/*
282 * POLL_ADD flags. Note that since sqe->poll_events is the flag space, the
283 * command flags for POLL_ADD are stored in sqe->len.
284 *
285 * IORING_POLL_ADD_MULTI Multishot poll. Sets IORING_CQE_F_MORE if
286 * the poll handler will continue to report
287 * CQEs on behalf of the same SQE.
288 *
289 * IORING_POLL_UPDATE Update existing poll request, matching
290 * sqe->addr as the old user_data field.
291 *
292 * IORING_POLL_LEVEL Level triggered poll.
293 */
294#define IORING_POLL_ADD_MULTI (1U << 0)
295#define IORING_POLL_UPDATE_EVENTS (1U << 1)
296#define IORING_POLL_UPDATE_USER_DATA (1U << 2)
297#define IORING_POLL_ADD_LEVEL (1U << 3)
298
299/*
300 * ASYNC_CANCEL flags.
301 *
302 * IORING_ASYNC_CANCEL_ALL Cancel all requests that match the given key
303 * IORING_ASYNC_CANCEL_FD Key off 'fd' for cancelation rather than the
304 * request 'user_data'
305 * IORING_ASYNC_CANCEL_ANY Match any request
306 * IORING_ASYNC_CANCEL_FD_FIXED 'fd' passed in is a fixed descriptor
307 * IORING_ASYNC_CANCEL_USERDATA Match on user_data, default for no other key
308 * IORING_ASYNC_CANCEL_OP Match request based on opcode
309 */
310#define IORING_ASYNC_CANCEL_ALL (1U << 0)
311#define IORING_ASYNC_CANCEL_FD (1U << 1)
312#define IORING_ASYNC_CANCEL_ANY (1U << 2)
313#define IORING_ASYNC_CANCEL_FD_FIXED (1U << 3)
314#define IORING_ASYNC_CANCEL_USERDATA (1U << 4)
315#define IORING_ASYNC_CANCEL_OP (1U << 5)
316
317/*
318 * send/sendmsg and recv/recvmsg flags (sqe->ioprio)
319 *
320 * IORING_RECVSEND_POLL_FIRST If set, instead of first attempting to send
321 * or receive and arm poll if that yields an
322 * -EAGAIN result, arm poll upfront and skip
323 * the initial transfer attempt.
324 *
325 * IORING_RECV_MULTISHOT Multishot recv. Sets IORING_CQE_F_MORE if
326 * the handler will continue to report
327 * CQEs on behalf of the same SQE.
328 *
329 * IORING_RECVSEND_FIXED_BUF Use registered buffers, the index is stored in
330 * the buf_index field.
331 *
332 * IORING_SEND_ZC_REPORT_USAGE
333 * If set, SEND[MSG]_ZC should report
334 * the zerocopy usage in cqe.res
335 * for the IORING_CQE_F_NOTIF cqe.
336 * 0 is reported if zerocopy was actually possible.
337 * IORING_NOTIF_USAGE_ZC_COPIED if data was copied
338 * (at least partially).
339 */
340#define IORING_RECVSEND_POLL_FIRST (1U << 0)
341#define IORING_RECV_MULTISHOT (1U << 1)
342#define IORING_RECVSEND_FIXED_BUF (1U << 2)
343#define IORING_SEND_ZC_REPORT_USAGE (1U << 3)
344
345/*
346 * cqe.res for IORING_CQE_F_NOTIF if
347 * IORING_SEND_ZC_REPORT_USAGE was requested
348 *
349 * It should be treated as a flag, all other
350 * bits of cqe.res should be treated as reserved!
351 */
352#define IORING_NOTIF_USAGE_ZC_COPIED (1U << 31)
353
354/*
355 * accept flags stored in sqe->ioprio
356 */
357#define IORING_ACCEPT_MULTISHOT (1U << 0)
358
359/*
360 * IORING_OP_MSG_RING command types, stored in sqe->addr
361 */
362enum {
363 IORING_MSG_DATA, /* pass sqe->len as 'res' and off as user_data */
364 IORING_MSG_SEND_FD, /* send a registered fd to another ring */
365};
366
367/*
368 * IORING_OP_MSG_RING flags (sqe->msg_ring_flags)
369 *
370 * IORING_MSG_RING_CQE_SKIP Don't post a CQE to the target ring. Not
371 * applicable for IORING_MSG_DATA, obviously.
372 */
373#define IORING_MSG_RING_CQE_SKIP (1U << 0)
374/* Pass through the flags from sqe->file_index to cqe->flags */
375#define IORING_MSG_RING_FLAGS_PASS (1U << 1)
376
377/*
378 * IO completion data structure (Completion Queue Entry)
379 */
380struct io_uring_cqe {
381 __u64 user_data; /* sqe->data submission passed back */
382 __s32 res; /* result code for this event */
383 __u32 flags;
384
385 /*
386 * If the ring is initialized with IORING_SETUP_CQE32, then this field
387 * contains 16-bytes of padding, doubling the size of the CQE.
388 */
389 __u64 big_cqe[];
390};
391
392/*
393 * cqe->flags
394 *
395 * IORING_CQE_F_BUFFER If set, the upper 16 bits are the buffer ID
396 * IORING_CQE_F_MORE If set, parent SQE will generate more CQE entries
397 * IORING_CQE_F_SOCK_NONEMPTY If set, more data to read after socket recv
398 * IORING_CQE_F_NOTIF Set for notification CQEs. Can be used to distinct
399 * them from sends.
400 */
401#define IORING_CQE_F_BUFFER (1U << 0)
402#define IORING_CQE_F_MORE (1U << 1)
403#define IORING_CQE_F_SOCK_NONEMPTY (1U << 2)
404#define IORING_CQE_F_NOTIF (1U << 3)
405
406enum {
407 IORING_CQE_BUFFER_SHIFT = 16,
408};
409
410/*
411 * Magic offsets for the application to mmap the data it needs
412 */
413#define IORING_OFF_SQ_RING 0ULL
414#define IORING_OFF_CQ_RING 0x8000000ULL
415#define IORING_OFF_SQES 0x10000000ULL
416#define IORING_OFF_PBUF_RING 0x80000000ULL
417#define IORING_OFF_PBUF_SHIFT 16
418#define IORING_OFF_MMAP_MASK 0xf8000000ULL
419
420/*
421 * Filled with the offset for mmap(2)
422 */
423struct io_sqring_offsets {
424 __u32 head;
425 __u32 tail;
426 __u32 ring_mask;
427 __u32 ring_entries;
428 __u32 flags;
429 __u32 dropped;
430 __u32 array;
431 __u32 resv1;
432 __u64 user_addr;
433};
434
435/*
436 * sq_ring->flags
437 */
438#define IORING_SQ_NEED_WAKEUP (1U << 0) /* needs io_uring_enter wakeup */
439#define IORING_SQ_CQ_OVERFLOW (1U << 1) /* CQ ring is overflown */
440#define IORING_SQ_TASKRUN (1U << 2) /* task should enter the kernel */
441
442struct io_cqring_offsets {
443 __u32 head;
444 __u32 tail;
445 __u32 ring_mask;
446 __u32 ring_entries;
447 __u32 overflow;
448 __u32 cqes;
449 __u32 flags;
450 __u32 resv1;
451 __u64 user_addr;
452};
453
454/*
455 * cq_ring->flags
456 */
457
458/* disable eventfd notifications */
459#define IORING_CQ_EVENTFD_DISABLED (1U << 0)
460
461/*
462 * io_uring_enter(2) flags
463 */
464#define IORING_ENTER_GETEVENTS (1U << 0)
465#define IORING_ENTER_SQ_WAKEUP (1U << 1)
466#define IORING_ENTER_SQ_WAIT (1U << 2)
467#define IORING_ENTER_EXT_ARG (1U << 3)
468#define IORING_ENTER_REGISTERED_RING (1U << 4)
469
470/*
471 * Passed in for io_uring_setup(2). Copied back with updated info on success
472 */
473struct io_uring_params {
474 __u32 sq_entries;
475 __u32 cq_entries;
476 __u32 flags;
477 __u32 sq_thread_cpu;
478 __u32 sq_thread_idle;
479 __u32 features;
480 __u32 wq_fd;
481 __u32 resv[3];
482 struct io_sqring_offsets sq_off;
483 struct io_cqring_offsets cq_off;
484};
485
486/*
487 * io_uring_params->features flags
488 */
489#define IORING_FEAT_SINGLE_MMAP (1U << 0)
490#define IORING_FEAT_NODROP (1U << 1)
491#define IORING_FEAT_SUBMIT_STABLE (1U << 2)
492#define IORING_FEAT_RW_CUR_POS (1U << 3)
493#define IORING_FEAT_CUR_PERSONALITY (1U << 4)
494#define IORING_FEAT_FAST_POLL (1U << 5)
495#define IORING_FEAT_POLL_32BITS (1U << 6)
496#define IORING_FEAT_SQPOLL_NONFIXED (1U << 7)
497#define IORING_FEAT_EXT_ARG (1U << 8)
498#define IORING_FEAT_NATIVE_WORKERS (1U << 9)
499#define IORING_FEAT_RSRC_TAGS (1U << 10)
500#define IORING_FEAT_CQE_SKIP (1U << 11)
501#define IORING_FEAT_LINKED_FILE (1U << 12)
502#define IORING_FEAT_REG_REG_RING (1U << 13)
503
504/*
505 * io_uring_register(2) opcodes and arguments
506 */
507enum {
508 IORING_REGISTER_BUFFERS = 0,
509 IORING_UNREGISTER_BUFFERS = 1,
510 IORING_REGISTER_FILES = 2,
511 IORING_UNREGISTER_FILES = 3,
512 IORING_REGISTER_EVENTFD = 4,
513 IORING_UNREGISTER_EVENTFD = 5,
514 IORING_REGISTER_FILES_UPDATE = 6,
515 IORING_REGISTER_EVENTFD_ASYNC = 7,
516 IORING_REGISTER_PROBE = 8,
517 IORING_REGISTER_PERSONALITY = 9,
518 IORING_UNREGISTER_PERSONALITY = 10,
519 IORING_REGISTER_RESTRICTIONS = 11,
520 IORING_REGISTER_ENABLE_RINGS = 12,
521
522 /* extended with tagging */
523 IORING_REGISTER_FILES2 = 13,
524 IORING_REGISTER_FILES_UPDATE2 = 14,
525 IORING_REGISTER_BUFFERS2 = 15,
526 IORING_REGISTER_BUFFERS_UPDATE = 16,
527
528 /* set/clear io-wq thread affinities */
529 IORING_REGISTER_IOWQ_AFF = 17,
530 IORING_UNREGISTER_IOWQ_AFF = 18,
531
532 /* set/get max number of io-wq workers */
533 IORING_REGISTER_IOWQ_MAX_WORKERS = 19,
534
535 /* register/unregister io_uring fd with the ring */
536 IORING_REGISTER_RING_FDS = 20,
537 IORING_UNREGISTER_RING_FDS = 21,
538
539 /* register ring based provide buffer group */
540 IORING_REGISTER_PBUF_RING = 22,
541 IORING_UNREGISTER_PBUF_RING = 23,
542
543 /* sync cancelation API */
544 IORING_REGISTER_SYNC_CANCEL = 24,
545
546 /* register a range of fixed file slots for automatic slot allocation */
547 IORING_REGISTER_FILE_ALLOC_RANGE = 25,
548
549 /* this goes last */
550 IORING_REGISTER_LAST,
551
552 /* flag added to the opcode to use a registered ring fd */
553 IORING_REGISTER_USE_REGISTERED_RING = 1U << 31
554};
555
556/* io-wq worker categories */
557enum {
558 IO_WQ_BOUND,
559 IO_WQ_UNBOUND,
560};
561
562/* deprecated, see struct io_uring_rsrc_update */
563struct io_uring_files_update {
564 __u32 offset;
565 __u32 resv;
566 __aligned_u64 /* __s32 * */ fds;
567};
568
569/*
570 * Register a fully sparse file space, rather than pass in an array of all
571 * -1 file descriptors.
572 */
573#define IORING_RSRC_REGISTER_SPARSE (1U << 0)
574
575struct io_uring_rsrc_register {
576 __u32 nr;
577 __u32 flags;
578 __u64 resv2;
579 __aligned_u64 data;
580 __aligned_u64 tags;
581};
582
583struct io_uring_rsrc_update {
584 __u32 offset;
585 __u32 resv;
586 __aligned_u64 data;
587};
588
589struct io_uring_rsrc_update2 {
590 __u32 offset;
591 __u32 resv;
592 __aligned_u64 data;
593 __aligned_u64 tags;
594 __u32 nr;
595 __u32 resv2;
596};
597
598/* Skip updating fd indexes set to this value in the fd table */
599#define IORING_REGISTER_FILES_SKIP (-2)
600
601#define IO_URING_OP_SUPPORTED (1U << 0)
602
603struct io_uring_probe_op {
604 __u8 op;
605 __u8 resv;
606 __u16 flags; /* IO_URING_OP_* flags */
607 __u32 resv2;
608};
609
610struct io_uring_probe {
611 __u8 last_op; /* last opcode supported */
612 __u8 ops_len; /* length of ops[] array below */
613 __u16 resv;
614 __u32 resv2[3];
615 struct io_uring_probe_op ops[];
616};
617
618struct io_uring_restriction {
619 __u16 opcode;
620 union {
621 __u8 register_op; /* IORING_RESTRICTION_REGISTER_OP */
622 __u8 sqe_op; /* IORING_RESTRICTION_SQE_OP */
623 __u8 sqe_flags; /* IORING_RESTRICTION_SQE_FLAGS_* */
624 };
625 __u8 resv;
626 __u32 resv2[3];
627};
628
629struct io_uring_buf {
630 __u64 addr;
631 __u32 len;
632 __u16 bid;
633 __u16 resv;
634};
635
636struct io_uring_buf_ring {
637 union {
638 /*
639 * To avoid spilling into more pages than we need to, the
640 * ring tail is overlaid with the io_uring_buf->resv field.
641 */
642 struct {
643 __u64 resv1;
644 __u32 resv2;
645 __u16 resv3;
646 __u16 tail;
647 };
648 __DECLARE_FLEX_ARRAY(struct io_uring_buf, bufs);
649 };
650};
651
652/*
653 * Flags for IORING_REGISTER_PBUF_RING.
654 *
655 * IOU_PBUF_RING_MMAP: If set, kernel will allocate the memory for the ring.
656 * The application must not set a ring_addr in struct
657 * io_uring_buf_reg, instead it must subsequently call
658 * mmap(2) with the offset set as:
659 * IORING_OFF_PBUF_RING | (bgid << IORING_OFF_PBUF_SHIFT)
660 * to get a virtual mapping for the ring.
661 */
662enum {
663 IOU_PBUF_RING_MMAP = 1,
664};
665
666/* argument for IORING_(UN)REGISTER_PBUF_RING */
667struct io_uring_buf_reg {
668 __u64 ring_addr;
669 __u32 ring_entries;
670 __u16 bgid;
671 __u16 flags;
672 __u64 resv[3];
673};
674
675/*
676 * io_uring_restriction->opcode values
677 */
678enum {
679 /* Allow an io_uring_register(2) opcode */
680 IORING_RESTRICTION_REGISTER_OP = 0,
681
682 /* Allow an sqe opcode */
683 IORING_RESTRICTION_SQE_OP = 1,
684
685 /* Allow sqe flags */
686 IORING_RESTRICTION_SQE_FLAGS_ALLOWED = 2,
687
688 /* Require sqe flags (these flags must be set on each submission) */
689 IORING_RESTRICTION_SQE_FLAGS_REQUIRED = 3,
690
691 IORING_RESTRICTION_LAST
692};
693
694struct io_uring_getevents_arg {
695 __u64 sigmask;
696 __u32 sigmask_sz;
697 __u32 pad;
698 __u64 ts;
699};
700
701/*
702 * Argument for IORING_REGISTER_SYNC_CANCEL
703 */
704struct io_uring_sync_cancel_reg {
705 __u64 addr;
706 __s32 fd;
707 __u32 flags;
708 struct __kernel_timespec timeout;
709 __u8 opcode;
710 __u8 pad[7];
711 __u64 pad2[3];
712};
713
714/*
715 * Argument for IORING_REGISTER_FILE_ALLOC_RANGE
716 * The range is specified as [off, off + len)
717 */
718struct io_uring_file_index_range {
719 __u32 off;
720 __u32 len;
721 __u64 resv;
722};
723
724struct io_uring_recvmsg_out {
725 __u32 namelen;
726 __u32 controllen;
727 __u32 payloadlen;
728 __u32 flags;
729};
730
731/*
732 * Argument for IORING_OP_URING_CMD when file is a socket
733 */
734enum {
735 SOCKET_URING_OP_SIOCINQ = 0,
736 SOCKET_URING_OP_SIOCOUTQ,
737};
738
739#ifdef __cplusplus
740}
741#endif
742
743#endif