1 /* SPDX-License-Identifier: MIT */
2 #ifndef LIB_URING_H
3 #define LIB_URING_H
4
5 #ifndef _XOPEN_SOURCE
6 #define _XOPEN_SOURCE 500 /* Required for glibc to expose sigset_t */
7 #endif
8
9 #include <sys/socket.h>
10 #include <sys/uio.h>
11 #include <sys/stat.h>
12 #include <errno.h>
13 #include <signal.h>
14 #include <stdbool.h>
15 #include <inttypes.h>
16 #include <time.h>
17 #include <linux/swab.h>
18 #include "liburing/compat.h"
19 #include "liburing/io_uring.h"
20 #include "liburing/barrier.h"
21
22 #ifndef uring_unlikely
23 # define uring_unlikely(cond) __builtin_expect(!!(cond), 0)
24 #endif
25
26 #ifndef uring_likely
27 # define uring_likely(cond) __builtin_expect(!!(cond), 1)
28 #endif
29
30 #ifdef __cplusplus
31 extern "C" {
32 #endif
33
34 /*
35 * Library interface to io_uring
36 */
37 struct io_uring_sq {
38 unsigned *khead;
39 unsigned *ktail;
40 unsigned *kring_mask;
41 unsigned *kring_entries;
42 unsigned *kflags;
43 unsigned *kdropped;
44 unsigned *array;
45 struct io_uring_sqe *sqes;
46
47 unsigned sqe_head;
48 unsigned sqe_tail;
49
50 size_t ring_sz;
51 void *ring_ptr;
52
53 unsigned pad[4];
54 };
55
56 struct io_uring_cq {
57 unsigned *khead;
58 unsigned *ktail;
59 unsigned *kring_mask;
60 unsigned *kring_entries;
61 unsigned *kflags;
62 unsigned *koverflow;
63 struct io_uring_cqe *cqes;
64
65 size_t ring_sz;
66 void *ring_ptr;
67
68 unsigned pad[4];
69 };
70
71 struct io_uring {
72 struct io_uring_sq sq;
73 struct io_uring_cq cq;
74 unsigned flags;
75 int ring_fd;
76
77 unsigned features;
78 unsigned pad[3];
79 };
80
81 /*
82 * Library interface
83 */
84
85 /*
86 * return an allocated io_uring_probe structure, or NULL if probe fails (for
87 * example, if it is not available). The caller is responsible for freeing it
88 */
89 extern struct io_uring_probe *io_uring_get_probe_ring(struct io_uring *ring);
90 /* same as io_uring_get_probe_ring, but takes care of ring init and teardown */
91 extern struct io_uring_probe *io_uring_get_probe(void);
92
93 /*
94 * frees a probe allocated through io_uring_get_probe() or
95 * io_uring_get_probe_ring()
96 */
97 extern void io_uring_free_probe(struct io_uring_probe *probe);
98
io_uring_opcode_supported(const struct io_uring_probe * p,int op)99 static inline int io_uring_opcode_supported(const struct io_uring_probe *p, int op)
100 {
101 if (op > p->last_op)
102 return 0;
103 return (p->ops[op].flags & IO_URING_OP_SUPPORTED) != 0;
104 }
105
106 extern int io_uring_queue_init_params(unsigned entries, struct io_uring *ring,
107 struct io_uring_params *p);
108 extern int io_uring_queue_init(unsigned entries, struct io_uring *ring,
109 unsigned flags);
110 extern int io_uring_queue_mmap(int fd, struct io_uring_params *p,
111 struct io_uring *ring);
112 extern int io_uring_ring_dontfork(struct io_uring *ring);
113 extern void io_uring_queue_exit(struct io_uring *ring);
114 unsigned io_uring_peek_batch_cqe(struct io_uring *ring,
115 struct io_uring_cqe **cqes, unsigned count);
116 extern int io_uring_wait_cqes(struct io_uring *ring,
117 struct io_uring_cqe **cqe_ptr, unsigned wait_nr,
118 struct __kernel_timespec *ts, sigset_t *sigmask);
119 extern int io_uring_wait_cqe_timeout(struct io_uring *ring,
120 struct io_uring_cqe **cqe_ptr, struct __kernel_timespec *ts);
121 extern int io_uring_submit(struct io_uring *ring);
122 extern int io_uring_submit_and_wait(struct io_uring *ring, unsigned wait_nr);
123 extern struct io_uring_sqe *io_uring_get_sqe(struct io_uring *ring);
124
125 extern int io_uring_register_buffers(struct io_uring *ring,
126 const struct iovec *iovecs,
127 unsigned nr_iovecs);
128 extern int io_uring_unregister_buffers(struct io_uring *ring);
129 extern int io_uring_register_files(struct io_uring *ring, const int *files,
130 unsigned nr_files);
131 extern int io_uring_unregister_files(struct io_uring *ring);
132 extern int io_uring_register_files_update(struct io_uring *ring, unsigned off,
133 int *files, unsigned nr_files);
134 extern int io_uring_register_eventfd(struct io_uring *ring, int fd);
135 extern int io_uring_register_eventfd_async(struct io_uring *ring, int fd);
136 extern int io_uring_unregister_eventfd(struct io_uring *ring);
137 extern int io_uring_register_probe(struct io_uring *ring,
138 struct io_uring_probe *p, unsigned nr);
139 extern int io_uring_register_personality(struct io_uring *ring);
140 extern int io_uring_unregister_personality(struct io_uring *ring, int id);
141 extern int io_uring_register_restrictions(struct io_uring *ring,
142 struct io_uring_restriction *res,
143 unsigned int nr_res);
144 extern int io_uring_enable_rings(struct io_uring *ring);
145 extern int __io_uring_sqring_wait(struct io_uring *ring);
146
147 /*
148 * Helper for the peek/wait single cqe functions. Exported because of that,
149 * but probably shouldn't be used directly in an application.
150 */
151 extern int __io_uring_get_cqe(struct io_uring *ring,
152 struct io_uring_cqe **cqe_ptr, unsigned submit,
153 unsigned wait_nr, sigset_t *sigmask);
154
155 #define LIBURING_UDATA_TIMEOUT ((__u64) -1)
156
157 #define io_uring_for_each_cqe(ring, head, cqe) \
158 /* \
159 * io_uring_smp_load_acquire() enforces the order of tail \
160 * and CQE reads. \
161 */ \
162 for (head = *(ring)->cq.khead; \
163 (cqe = (head != io_uring_smp_load_acquire((ring)->cq.ktail) ? \
164 &(ring)->cq.cqes[head & (*(ring)->cq.kring_mask)] : NULL)); \
165 head++) \
166
167 /*
168 * Must be called after io_uring_for_each_cqe()
169 */
io_uring_cq_advance(struct io_uring * ring,unsigned nr)170 static inline void io_uring_cq_advance(struct io_uring *ring,
171 unsigned nr)
172 {
173 if (nr) {
174 struct io_uring_cq *cq = &ring->cq;
175
176 /*
177 * Ensure that the kernel only sees the new value of the head
178 * index after the CQEs have been read.
179 */
180 io_uring_smp_store_release(cq->khead, *cq->khead + nr);
181 }
182 }
183
184 /*
185 * Must be called after io_uring_{peek,wait}_cqe() after the cqe has
186 * been processed by the application.
187 */
io_uring_cqe_seen(struct io_uring * ring,struct io_uring_cqe * cqe)188 static inline void io_uring_cqe_seen(struct io_uring *ring,
189 struct io_uring_cqe *cqe)
190 {
191 if (cqe)
192 io_uring_cq_advance(ring, 1);
193 }
194
195 /*
196 * Command prep helpers
197 */
io_uring_sqe_set_data(struct io_uring_sqe * sqe,void * data)198 static inline void io_uring_sqe_set_data(struct io_uring_sqe *sqe, void *data)
199 {
200 sqe->user_data = (unsigned long) data;
201 }
202
io_uring_cqe_get_data(const struct io_uring_cqe * cqe)203 static inline void *io_uring_cqe_get_data(const struct io_uring_cqe *cqe)
204 {
205 return (void *) (uintptr_t) cqe->user_data;
206 }
207
io_uring_sqe_set_flags(struct io_uring_sqe * sqe,unsigned flags)208 static inline void io_uring_sqe_set_flags(struct io_uring_sqe *sqe,
209 unsigned flags)
210 {
211 sqe->flags = flags;
212 }
213
io_uring_prep_rw(int op,struct io_uring_sqe * sqe,int fd,const void * addr,unsigned len,__u64 offset)214 static inline void io_uring_prep_rw(int op, struct io_uring_sqe *sqe, int fd,
215 const void *addr, unsigned len,
216 __u64 offset)
217 {
218 sqe->opcode = op;
219 sqe->flags = 0;
220 sqe->ioprio = 0;
221 sqe->fd = fd;
222 sqe->off = offset;
223 sqe->addr = (unsigned long) addr;
224 sqe->len = len;
225 sqe->rw_flags = 0;
226 sqe->user_data = 0;
227 sqe->__pad2[0] = sqe->__pad2[1] = sqe->__pad2[2] = 0;
228 }
229
230 /**
231 * @pre Either fd_in or fd_out must be a pipe.
232 * @param off_in If fd_in refers to a pipe, off_in must be (int64_t) -1;
233 * If fd_in does not refer to a pipe and off_in is (int64_t) -1, then bytes are read
234 * from fd_in starting from the file offset and it is adjust appropriately;
235 * If fd_in does not refer to a pipe and off_in is not (int64_t) -1, then the
236 * starting offset of fd_in will be off_in.
237 * @param off_out The description of off_in also applied to off_out.
238 * @param splice_flags see man splice(2) for description of flags.
239 *
240 * This splice operation can be used to implement sendfile by splicing to an intermediate pipe
241 * first, then splice to the final destination.
242 * In fact, the implementation of sendfile in kernel uses splice internally.
243 *
244 * NOTE that even if fd_in or fd_out refers to a pipe, the splice operation can still failed with
245 * EINVAL if one of the fd doesn't explicitly support splice operation, e.g. reading from terminal
246 * is unsupported from kernel 5.7 to 5.11.
247 * Check issue #291 for more information.
248 */
io_uring_prep_splice(struct io_uring_sqe * sqe,int fd_in,int64_t off_in,int fd_out,int64_t off_out,unsigned int nbytes,unsigned int splice_flags)249 static inline void io_uring_prep_splice(struct io_uring_sqe *sqe,
250 int fd_in, int64_t off_in,
251 int fd_out, int64_t off_out,
252 unsigned int nbytes,
253 unsigned int splice_flags)
254 {
255 io_uring_prep_rw(IORING_OP_SPLICE, sqe, fd_out, NULL, nbytes, off_out);
256 sqe->splice_off_in = off_in;
257 sqe->splice_fd_in = fd_in;
258 sqe->splice_flags = splice_flags;
259 }
260
io_uring_prep_tee(struct io_uring_sqe * sqe,int fd_in,int fd_out,unsigned int nbytes,unsigned int splice_flags)261 static inline void io_uring_prep_tee(struct io_uring_sqe *sqe,
262 int fd_in, int fd_out,
263 unsigned int nbytes,
264 unsigned int splice_flags)
265 {
266 io_uring_prep_rw(IORING_OP_TEE, sqe, fd_out, NULL, nbytes, 0);
267 sqe->splice_off_in = 0;
268 sqe->splice_fd_in = fd_in;
269 sqe->splice_flags = splice_flags;
270 }
271
io_uring_prep_readv(struct io_uring_sqe * sqe,int fd,const struct iovec * iovecs,unsigned nr_vecs,off_t offset)272 static inline void io_uring_prep_readv(struct io_uring_sqe *sqe, int fd,
273 const struct iovec *iovecs,
274 unsigned nr_vecs, off_t offset)
275 {
276 io_uring_prep_rw(IORING_OP_READV, sqe, fd, iovecs, nr_vecs, offset);
277 }
278
io_uring_prep_read_fixed(struct io_uring_sqe * sqe,int fd,void * buf,unsigned nbytes,off_t offset,int buf_index)279 static inline void io_uring_prep_read_fixed(struct io_uring_sqe *sqe, int fd,
280 void *buf, unsigned nbytes,
281 off_t offset, int buf_index)
282 {
283 io_uring_prep_rw(IORING_OP_READ_FIXED, sqe, fd, buf, nbytes, offset);
284 sqe->buf_index = buf_index;
285 }
286
io_uring_prep_writev(struct io_uring_sqe * sqe,int fd,const struct iovec * iovecs,unsigned nr_vecs,off_t offset)287 static inline void io_uring_prep_writev(struct io_uring_sqe *sqe, int fd,
288 const struct iovec *iovecs,
289 unsigned nr_vecs, off_t offset)
290 {
291 io_uring_prep_rw(IORING_OP_WRITEV, sqe, fd, iovecs, nr_vecs, offset);
292 }
293
io_uring_prep_write_fixed(struct io_uring_sqe * sqe,int fd,const void * buf,unsigned nbytes,off_t offset,int buf_index)294 static inline void io_uring_prep_write_fixed(struct io_uring_sqe *sqe, int fd,
295 const void *buf, unsigned nbytes,
296 off_t offset, int buf_index)
297 {
298 io_uring_prep_rw(IORING_OP_WRITE_FIXED, sqe, fd, buf, nbytes, offset);
299 sqe->buf_index = buf_index;
300 }
301
io_uring_prep_recvmsg(struct io_uring_sqe * sqe,int fd,struct msghdr * msg,unsigned flags)302 static inline void io_uring_prep_recvmsg(struct io_uring_sqe *sqe, int fd,
303 struct msghdr *msg, unsigned flags)
304 {
305 io_uring_prep_rw(IORING_OP_RECVMSG, sqe, fd, msg, 1, 0);
306 sqe->msg_flags = flags;
307 }
308
io_uring_prep_sendmsg(struct io_uring_sqe * sqe,int fd,const struct msghdr * msg,unsigned flags)309 static inline void io_uring_prep_sendmsg(struct io_uring_sqe *sqe, int fd,
310 const struct msghdr *msg, unsigned flags)
311 {
312 io_uring_prep_rw(IORING_OP_SENDMSG, sqe, fd, msg, 1, 0);
313 sqe->msg_flags = flags;
314 }
315
io_uring_prep_poll_add(struct io_uring_sqe * sqe,int fd,unsigned poll_mask)316 static inline void io_uring_prep_poll_add(struct io_uring_sqe *sqe, int fd,
317 unsigned poll_mask)
318 {
319 io_uring_prep_rw(IORING_OP_POLL_ADD, sqe, fd, NULL, 0, 0);
320 #if __BYTE_ORDER == __BIG_ENDIAN
321 poll_mask = __swahw32(poll_mask);
322 #endif
323 sqe->poll32_events = poll_mask;
324 }
325
io_uring_prep_poll_remove(struct io_uring_sqe * sqe,void * user_data)326 static inline void io_uring_prep_poll_remove(struct io_uring_sqe *sqe,
327 void *user_data)
328 {
329 io_uring_prep_rw(IORING_OP_POLL_REMOVE, sqe, -1, user_data, 0, 0);
330 }
331
io_uring_prep_poll_update(struct io_uring_sqe * sqe,void * old_user_data,void * new_user_data,unsigned poll_mask,unsigned flags)332 static inline void io_uring_prep_poll_update(struct io_uring_sqe *sqe,
333 void *old_user_data,
334 void *new_user_data,
335 unsigned poll_mask, unsigned flags)
336 {
337 io_uring_prep_rw(IORING_OP_POLL_REMOVE, sqe, -1, old_user_data, flags,
338 (__u64)new_user_data);
339 #if __BYTE_ORDER == __BIG_ENDIAN
340 poll_mask = __swahw32(poll_mask);
341 #endif
342 sqe->poll32_events = poll_mask;
343 }
344
io_uring_prep_fsync(struct io_uring_sqe * sqe,int fd,unsigned fsync_flags)345 static inline void io_uring_prep_fsync(struct io_uring_sqe *sqe, int fd,
346 unsigned fsync_flags)
347 {
348 io_uring_prep_rw(IORING_OP_FSYNC, sqe, fd, NULL, 0, 0);
349 sqe->fsync_flags = fsync_flags;
350 }
351
io_uring_prep_nop(struct io_uring_sqe * sqe)352 static inline void io_uring_prep_nop(struct io_uring_sqe *sqe)
353 {
354 io_uring_prep_rw(IORING_OP_NOP, sqe, -1, NULL, 0, 0);
355 }
356
io_uring_prep_timeout(struct io_uring_sqe * sqe,struct __kernel_timespec * ts,unsigned count,unsigned flags)357 static inline void io_uring_prep_timeout(struct io_uring_sqe *sqe,
358 struct __kernel_timespec *ts,
359 unsigned count, unsigned flags)
360 {
361 io_uring_prep_rw(IORING_OP_TIMEOUT, sqe, -1, ts, 1, count);
362 sqe->timeout_flags = flags;
363 }
364
io_uring_prep_timeout_remove(struct io_uring_sqe * sqe,__u64 user_data,unsigned flags)365 static inline void io_uring_prep_timeout_remove(struct io_uring_sqe *sqe,
366 __u64 user_data, unsigned flags)
367 {
368 io_uring_prep_rw(IORING_OP_TIMEOUT_REMOVE, sqe, -1,
369 (void *)(unsigned long)user_data, 0, 0);
370 sqe->timeout_flags = flags;
371 }
372
io_uring_prep_timeout_update(struct io_uring_sqe * sqe,struct __kernel_timespec * ts,__u64 user_data,unsigned flags)373 static inline void io_uring_prep_timeout_update(struct io_uring_sqe *sqe,
374 struct __kernel_timespec *ts,
375 __u64 user_data, unsigned flags)
376 {
377 io_uring_prep_rw(IORING_OP_TIMEOUT_REMOVE, sqe, -1,
378 (void *)(unsigned long)user_data, 0,
379 (uintptr_t)ts);
380 sqe->timeout_flags = flags | IORING_TIMEOUT_UPDATE;
381 }
382
io_uring_prep_accept(struct io_uring_sqe * sqe,int fd,struct sockaddr * addr,socklen_t * addrlen,int flags)383 static inline void io_uring_prep_accept(struct io_uring_sqe *sqe, int fd,
384 struct sockaddr *addr,
385 socklen_t *addrlen, int flags)
386 {
387 io_uring_prep_rw(IORING_OP_ACCEPT, sqe, fd, addr, 0,
388 (__u64) (unsigned long) addrlen);
389 sqe->accept_flags = flags;
390 }
391
io_uring_prep_cancel(struct io_uring_sqe * sqe,void * user_data,int flags)392 static inline void io_uring_prep_cancel(struct io_uring_sqe *sqe, void *user_data,
393 int flags)
394 {
395 io_uring_prep_rw(IORING_OP_ASYNC_CANCEL, sqe, -1, user_data, 0, 0);
396 sqe->cancel_flags = flags;
397 }
398
io_uring_prep_link_timeout(struct io_uring_sqe * sqe,struct __kernel_timespec * ts,unsigned flags)399 static inline void io_uring_prep_link_timeout(struct io_uring_sqe *sqe,
400 struct __kernel_timespec *ts,
401 unsigned flags)
402 {
403 io_uring_prep_rw(IORING_OP_LINK_TIMEOUT, sqe, -1, ts, 1, 0);
404 sqe->timeout_flags = flags;
405 }
406
io_uring_prep_connect(struct io_uring_sqe * sqe,int fd,const struct sockaddr * addr,socklen_t addrlen)407 static inline void io_uring_prep_connect(struct io_uring_sqe *sqe, int fd,
408 const struct sockaddr *addr,
409 socklen_t addrlen)
410 {
411 io_uring_prep_rw(IORING_OP_CONNECT, sqe, fd, addr, 0, addrlen);
412 }
413
io_uring_prep_files_update(struct io_uring_sqe * sqe,int * fds,unsigned nr_fds,int offset)414 static inline void io_uring_prep_files_update(struct io_uring_sqe *sqe,
415 int *fds, unsigned nr_fds,
416 int offset)
417 {
418 io_uring_prep_rw(IORING_OP_FILES_UPDATE, sqe, -1, fds, nr_fds, offset);
419 }
420
io_uring_prep_fallocate(struct io_uring_sqe * sqe,int fd,int mode,off_t offset,off_t len)421 static inline void io_uring_prep_fallocate(struct io_uring_sqe *sqe, int fd,
422 int mode, off_t offset, off_t len)
423 {
424
425 io_uring_prep_rw(IORING_OP_FALLOCATE, sqe, fd,
426 (const uintptr_t *) (unsigned long) len, mode, offset);
427 }
428
io_uring_prep_openat(struct io_uring_sqe * sqe,int dfd,const char * path,int flags,mode_t mode)429 static inline void io_uring_prep_openat(struct io_uring_sqe *sqe, int dfd,
430 const char *path, int flags, mode_t mode)
431 {
432 io_uring_prep_rw(IORING_OP_OPENAT, sqe, dfd, path, mode, 0);
433 sqe->open_flags = flags;
434 }
435
io_uring_prep_close(struct io_uring_sqe * sqe,int fd)436 static inline void io_uring_prep_close(struct io_uring_sqe *sqe, int fd)
437 {
438 io_uring_prep_rw(IORING_OP_CLOSE, sqe, fd, NULL, 0, 0);
439 }
440
io_uring_prep_read(struct io_uring_sqe * sqe,int fd,void * buf,unsigned nbytes,off_t offset)441 static inline void io_uring_prep_read(struct io_uring_sqe *sqe, int fd,
442 void *buf, unsigned nbytes, off_t offset)
443 {
444 io_uring_prep_rw(IORING_OP_READ, sqe, fd, buf, nbytes, offset);
445 }
446
io_uring_prep_write(struct io_uring_sqe * sqe,int fd,const void * buf,unsigned nbytes,off_t offset)447 static inline void io_uring_prep_write(struct io_uring_sqe *sqe, int fd,
448 const void *buf, unsigned nbytes, off_t offset)
449 {
450 io_uring_prep_rw(IORING_OP_WRITE, sqe, fd, buf, nbytes, offset);
451 }
452
453 struct statx;
io_uring_prep_statx(struct io_uring_sqe * sqe,int dfd,const char * path,int flags,unsigned mask,struct statx * statxbuf)454 static inline void io_uring_prep_statx(struct io_uring_sqe *sqe, int dfd,
455 const char *path, int flags, unsigned mask,
456 struct statx *statxbuf)
457 {
458 io_uring_prep_rw(IORING_OP_STATX, sqe, dfd, path, mask,
459 (__u64) (unsigned long) statxbuf);
460 sqe->statx_flags = flags;
461 }
462
io_uring_prep_fadvise(struct io_uring_sqe * sqe,int fd,off_t offset,off_t len,int advice)463 static inline void io_uring_prep_fadvise(struct io_uring_sqe *sqe, int fd,
464 off_t offset, off_t len, int advice)
465 {
466 io_uring_prep_rw(IORING_OP_FADVISE, sqe, fd, NULL, len, offset);
467 sqe->fadvise_advice = advice;
468 }
469
io_uring_prep_madvise(struct io_uring_sqe * sqe,void * addr,off_t length,int advice)470 static inline void io_uring_prep_madvise(struct io_uring_sqe *sqe, void *addr,
471 off_t length, int advice)
472 {
473 io_uring_prep_rw(IORING_OP_MADVISE, sqe, -1, addr, length, 0);
474 sqe->fadvise_advice = advice;
475 }
476
io_uring_prep_send(struct io_uring_sqe * sqe,int sockfd,const void * buf,size_t len,int flags)477 static inline void io_uring_prep_send(struct io_uring_sqe *sqe, int sockfd,
478 const void *buf, size_t len, int flags)
479 {
480 io_uring_prep_rw(IORING_OP_SEND, sqe, sockfd, buf, len, 0);
481 sqe->msg_flags = flags;
482 }
483
io_uring_prep_recv(struct io_uring_sqe * sqe,int sockfd,void * buf,size_t len,int flags)484 static inline void io_uring_prep_recv(struct io_uring_sqe *sqe, int sockfd,
485 void *buf, size_t len, int flags)
486 {
487 io_uring_prep_rw(IORING_OP_RECV, sqe, sockfd, buf, len, 0);
488 sqe->msg_flags = flags;
489 }
490
io_uring_prep_openat2(struct io_uring_sqe * sqe,int dfd,const char * path,struct open_how * how)491 static inline void io_uring_prep_openat2(struct io_uring_sqe *sqe, int dfd,
492 const char *path, struct open_how *how)
493 {
494 io_uring_prep_rw(IORING_OP_OPENAT2, sqe, dfd, path, sizeof(*how),
495 (uint64_t) (uintptr_t) how);
496 }
497
498 struct epoll_event;
io_uring_prep_epoll_ctl(struct io_uring_sqe * sqe,int epfd,int fd,int op,struct epoll_event * ev)499 static inline void io_uring_prep_epoll_ctl(struct io_uring_sqe *sqe, int epfd,
500 int fd, int op,
501 struct epoll_event *ev)
502 {
503 io_uring_prep_rw(IORING_OP_EPOLL_CTL, sqe, epfd, ev, op, fd);
504 }
505
io_uring_prep_provide_buffers(struct io_uring_sqe * sqe,void * addr,int len,int nr,int bgid,int bid)506 static inline void io_uring_prep_provide_buffers(struct io_uring_sqe *sqe,
507 void *addr, int len, int nr,
508 int bgid, int bid)
509 {
510 io_uring_prep_rw(IORING_OP_PROVIDE_BUFFERS, sqe, nr, addr, len, bid);
511 sqe->buf_group = bgid;
512 }
513
io_uring_prep_remove_buffers(struct io_uring_sqe * sqe,int nr,int bgid)514 static inline void io_uring_prep_remove_buffers(struct io_uring_sqe *sqe,
515 int nr, int bgid)
516 {
517 io_uring_prep_rw(IORING_OP_REMOVE_BUFFERS, sqe, nr, NULL, 0, 0);
518 sqe->buf_group = bgid;
519 }
520
io_uring_prep_shutdown(struct io_uring_sqe * sqe,int fd,int how)521 static inline void io_uring_prep_shutdown(struct io_uring_sqe *sqe, int fd,
522 int how)
523 {
524 io_uring_prep_rw(IORING_OP_SHUTDOWN, sqe, fd, NULL, how, 0);
525 }
526
io_uring_prep_unlinkat(struct io_uring_sqe * sqe,int dfd,const char * path,int flags)527 static inline void io_uring_prep_unlinkat(struct io_uring_sqe *sqe, int dfd,
528 const char *path, int flags)
529 {
530 io_uring_prep_rw(IORING_OP_UNLINKAT, sqe, dfd, path, 0, 0);
531 sqe->unlink_flags = flags;
532 }
533
io_uring_prep_renameat(struct io_uring_sqe * sqe,int olddfd,const char * oldpath,int newdfd,const char * newpath,int flags)534 static inline void io_uring_prep_renameat(struct io_uring_sqe *sqe, int olddfd,
535 const char *oldpath, int newdfd,
536 const char *newpath, int flags)
537 {
538 io_uring_prep_rw(IORING_OP_RENAMEAT, sqe, olddfd, oldpath, newdfd,
539 (uint64_t) (uintptr_t) newpath);
540 sqe->rename_flags = flags;
541 }
542
io_uring_prep_sync_file_range(struct io_uring_sqe * sqe,int fd,unsigned len,off_t offset,int flags)543 static inline void io_uring_prep_sync_file_range(struct io_uring_sqe *sqe,
544 int fd, unsigned len,
545 off_t offset, int flags)
546 {
547 io_uring_prep_rw(IORING_OP_SYNC_FILE_RANGE, sqe, fd, NULL, len, offset);
548 sqe->sync_range_flags = flags;
549 }
550
io_uring_prep_mkdirat(struct io_uring_sqe * sqe,int dfd,const char * path,mode_t mode)551 static inline void io_uring_prep_mkdirat(struct io_uring_sqe *sqe, int dfd,
552 const char *path, mode_t mode)
553 {
554 io_uring_prep_rw(IORING_OP_MKDIRAT, sqe, dfd, path, mode, 0);
555 }
556
io_uring_prep_symlinkat(struct io_uring_sqe * sqe,const char * target,int newdirfd,const char * linkpath)557 static inline void io_uring_prep_symlinkat(struct io_uring_sqe *sqe,
558 const char *target, int newdirfd, const char *linkpath)
559 {
560 io_uring_prep_rw(IORING_OP_SYMLINKAT, sqe, newdirfd, target, 0,
561 (uint64_t) (uintptr_t) linkpath);
562 }
563
io_uring_prep_linkat(struct io_uring_sqe * sqe,int olddfd,const char * oldpath,int newdfd,const char * newpath,int flags)564 static inline void io_uring_prep_linkat(struct io_uring_sqe *sqe, int olddfd,
565 const char *oldpath, int newdfd,
566 const char *newpath, int flags)
567 {
568 io_uring_prep_rw(IORING_OP_LINKAT, sqe, olddfd, oldpath, newdfd,
569 (uint64_t) (uintptr_t) newpath);
570 sqe->hardlink_flags = flags;
571 }
572
573 /*
574 * Returns number of unconsumed (if SQPOLL) or unsubmitted entries exist in
575 * the SQ ring
576 */
io_uring_sq_ready(const struct io_uring * ring)577 static inline unsigned io_uring_sq_ready(const struct io_uring *ring)
578 {
579 /*
580 * Without a barrier, we could miss an update and think the SQ wasn't ready.
581 * We don't need the load acquire for non-SQPOLL since then we drive updates.
582 */
583 if (ring->flags & IORING_SETUP_SQPOLL)
584 return ring->sq.sqe_tail - io_uring_smp_load_acquire(ring->sq.khead);
585
586 /* always use real head, to avoid losing sync for short submit */
587 return ring->sq.sqe_tail - *ring->sq.khead;
588 }
589
590 /*
591 * Returns how much space is left in the SQ ring.
592 */
io_uring_sq_space_left(const struct io_uring * ring)593 static inline unsigned io_uring_sq_space_left(const struct io_uring *ring)
594 {
595 return *ring->sq.kring_entries - io_uring_sq_ready(ring);
596 }
597
598 /*
599 * Only applicable when using SQPOLL - allows the caller to wait for space
600 * to free up in the SQ ring, which happens when the kernel side thread has
601 * consumed one or more entries. If the SQ ring is currently non-full, no
602 * action is taken. Note: may return -EINVAL if the kernel doesn't support
603 * this feature.
604 */
io_uring_sqring_wait(struct io_uring * ring)605 static inline int io_uring_sqring_wait(struct io_uring *ring)
606 {
607 if (!(ring->flags & IORING_SETUP_SQPOLL))
608 return 0;
609 if (io_uring_sq_space_left(ring))
610 return 0;
611
612 return __io_uring_sqring_wait(ring);
613 }
614
615 /*
616 * Returns how many unconsumed entries are ready in the CQ ring
617 */
io_uring_cq_ready(const struct io_uring * ring)618 static inline unsigned io_uring_cq_ready(const struct io_uring *ring)
619 {
620 return io_uring_smp_load_acquire(ring->cq.ktail) - *ring->cq.khead;
621 }
622
623 /*
624 * Returns true if the eventfd notification is currently enabled
625 */
io_uring_cq_eventfd_enabled(const struct io_uring * ring)626 static inline bool io_uring_cq_eventfd_enabled(const struct io_uring *ring)
627 {
628 if (!ring->cq.kflags)
629 return true;
630
631 return !(*ring->cq.kflags & IORING_CQ_EVENTFD_DISABLED);
632 }
633
634 /*
635 * Toggle eventfd notification on or off, if an eventfd is registered with
636 * the ring.
637 */
io_uring_cq_eventfd_toggle(struct io_uring * ring,bool enabled)638 static inline int io_uring_cq_eventfd_toggle(struct io_uring *ring,
639 bool enabled)
640 {
641 uint32_t flags;
642
643 if (!!enabled == io_uring_cq_eventfd_enabled(ring))
644 return 0;
645
646 if (!ring->cq.kflags)
647 return -EOPNOTSUPP;
648
649 flags = *ring->cq.kflags;
650
651 if (enabled)
652 flags &= ~IORING_CQ_EVENTFD_DISABLED;
653 else
654 flags |= IORING_CQ_EVENTFD_DISABLED;
655
656 IO_URING_WRITE_ONCE(*ring->cq.kflags, flags);
657
658 return 0;
659 }
660
661 /*
662 * Return an IO completion, waiting for 'wait_nr' completions if one isn't
663 * readily available. Returns 0 with cqe_ptr filled in on success, -errno on
664 * failure.
665 */
io_uring_wait_cqe_nr(struct io_uring * ring,struct io_uring_cqe ** cqe_ptr,unsigned wait_nr)666 static inline int io_uring_wait_cqe_nr(struct io_uring *ring,
667 struct io_uring_cqe **cqe_ptr,
668 unsigned wait_nr)
669 {
670 return __io_uring_get_cqe(ring, cqe_ptr, 0, wait_nr, NULL);
671 }
672
673 /*
674 * Return an IO completion, if one is readily available. Returns 0 with
675 * cqe_ptr filled in on success, -errno on failure.
676 */
io_uring_peek_cqe(struct io_uring * ring,struct io_uring_cqe ** cqe_ptr)677 static inline int io_uring_peek_cqe(struct io_uring *ring,
678 struct io_uring_cqe **cqe_ptr)
679 {
680 return io_uring_wait_cqe_nr(ring, cqe_ptr, 0);
681 }
682
683 /*
684 * Return an IO completion, waiting for it if necessary. Returns 0 with
685 * cqe_ptr filled in on success, -errno on failure.
686 */
io_uring_wait_cqe(struct io_uring * ring,struct io_uring_cqe ** cqe_ptr)687 static inline int io_uring_wait_cqe(struct io_uring *ring,
688 struct io_uring_cqe **cqe_ptr)
689 {
690 return io_uring_wait_cqe_nr(ring, cqe_ptr, 1);
691 }
692
693 ssize_t io_uring_mlock_size(unsigned entries, unsigned flags);
694 ssize_t io_uring_mlock_size_params(unsigned entries, struct io_uring_params *p);
695
696 #ifdef __cplusplus
697 }
698 #endif
699
700 #endif
701