Commit | Line | Data |
---|---|---|
2b188cc1 JA |
1 | // SPDX-License-Identifier: GPL-2.0 |
2 | /* | |
3 | * Shared application/kernel submission and completion ring pairs, for | |
4 | * supporting fast/efficient IO. | |
5 | * | |
6 | * A note on the read/write ordering memory barriers that are matched between | |
1e84b97b SB |
7 | * the application and kernel side. |
8 | * | |
9 | * After the application reads the CQ ring tail, it must use an | |
10 | * appropriate smp_rmb() to pair with the smp_wmb() the kernel uses | |
11 | * before writing the tail (using smp_load_acquire to read the tail will | |
12 | * do). It also needs a smp_mb() before updating CQ head (ordering the | |
13 | * entry load(s) with the head store), pairing with an implicit barrier | |
14 | * through a control-dependency in io_get_cqring (smp_store_release to | |
15 | * store head will do). Failure to do so could lead to reading invalid | |
16 | * CQ entries. | |
17 | * | |
18 | * Likewise, the application must use an appropriate smp_wmb() before | |
19 | * writing the SQ tail (ordering SQ entry stores with the tail store), | |
20 | * which pairs with smp_load_acquire in io_get_sqring (smp_store_release | |
21 | * to store the tail will do). And it needs a barrier ordering the SQ | |
22 | * head load before writing new SQ entries (smp_load_acquire to read | |
23 | * head will do). | |
24 | * | |
25 | * When using the SQ poll thread (IORING_SETUP_SQPOLL), the application | |
26 | * needs to check the SQ flags for IORING_SQ_NEED_WAKEUP *after* | |
27 | * updating the SQ tail; a full memory barrier smp_mb() is needed | |
28 | * between. | |
2b188cc1 JA |
29 | * |
30 | * Also see the examples in the liburing library: | |
31 | * | |
32 | * git://git.kernel.dk/liburing | |
33 | * | |
34 | * io_uring also uses READ/WRITE_ONCE() for _any_ store or load that happens | |
35 | * from data shared between the kernel and application. This is done both | |
36 | * for ordering purposes, but also to ensure that once a value is loaded from | |
37 | * data that the application could potentially modify, it remains stable. | |
38 | * | |
39 | * Copyright (C) 2018-2019 Jens Axboe | |
c992fe29 | 40 | * Copyright (c) 2018-2019 Christoph Hellwig |
2b188cc1 JA |
41 | */ |
42 | #include <linux/kernel.h> | |
43 | #include <linux/init.h> | |
44 | #include <linux/errno.h> | |
45 | #include <linux/syscalls.h> | |
46 | #include <linux/compat.h> | |
47 | #include <linux/refcount.h> | |
48 | #include <linux/uio.h> | |
6b47ee6e | 49 | #include <linux/bits.h> |
2b188cc1 JA |
50 | |
51 | #include <linux/sched/signal.h> | |
52 | #include <linux/fs.h> | |
53 | #include <linux/file.h> | |
54 | #include <linux/fdtable.h> | |
55 | #include <linux/mm.h> | |
56 | #include <linux/mman.h> | |
57 | #include <linux/mmu_context.h> | |
58 | #include <linux/percpu.h> | |
59 | #include <linux/slab.h> | |
6c271ce2 | 60 | #include <linux/kthread.h> |
2b188cc1 | 61 | #include <linux/blkdev.h> |
edafccee | 62 | #include <linux/bvec.h> |
2b188cc1 JA |
63 | #include <linux/net.h> |
64 | #include <net/sock.h> | |
65 | #include <net/af_unix.h> | |
6b06314c | 66 | #include <net/scm.h> |
2b188cc1 JA |
67 | #include <linux/anon_inodes.h> |
68 | #include <linux/sched/mm.h> | |
69 | #include <linux/uaccess.h> | |
70 | #include <linux/nospec.h> | |
edafccee JA |
71 | #include <linux/sizes.h> |
72 | #include <linux/hugetlb.h> | |
aa4c3967 | 73 | #include <linux/highmem.h> |
15b71abe JA |
74 | #include <linux/namei.h> |
75 | #include <linux/fsnotify.h> | |
4840e418 | 76 | #include <linux/fadvise.h> |
3e4827b0 | 77 | #include <linux/eventpoll.h> |
ff002b30 | 78 | #include <linux/fs_struct.h> |
7d67af2c | 79 | #include <linux/splice.h> |
b41e9852 | 80 | #include <linux/task_work.h> |
2b188cc1 | 81 | |
c826bd7a DD |
82 | #define CREATE_TRACE_POINTS |
83 | #include <trace/events/io_uring.h> | |
84 | ||
2b188cc1 JA |
85 | #include <uapi/linux/io_uring.h> |
86 | ||
87 | #include "internal.h" | |
561fb04a | 88 | #include "io-wq.h" |
2b188cc1 | 89 | |
5277deaa | 90 | #define IORING_MAX_ENTRIES 32768 |
33a107f0 | 91 | #define IORING_MAX_CQ_ENTRIES (2 * IORING_MAX_ENTRIES) |
65e19f54 JA |
92 | |
93 | /* | |
94 | * Shift of 9 is 512 entries, or exactly one page on 64-bit archs | |
95 | */ | |
96 | #define IORING_FILE_TABLE_SHIFT 9 | |
97 | #define IORING_MAX_FILES_TABLE (1U << IORING_FILE_TABLE_SHIFT) | |
98 | #define IORING_FILE_TABLE_MASK (IORING_MAX_FILES_TABLE - 1) | |
99 | #define IORING_MAX_FIXED_FILES (64 * IORING_MAX_FILES_TABLE) | |
2b188cc1 JA |
100 | |
101 | struct io_uring { | |
102 | u32 head ____cacheline_aligned_in_smp; | |
103 | u32 tail ____cacheline_aligned_in_smp; | |
104 | }; | |
105 | ||
1e84b97b | 106 | /* |
75b28aff HV |
107 | * This data is shared with the application through the mmap at offsets |
108 | * IORING_OFF_SQ_RING and IORING_OFF_CQ_RING. | |
1e84b97b SB |
109 | * |
110 | * The offsets to the member fields are published through struct | |
111 | * io_sqring_offsets when calling io_uring_setup. | |
112 | */ | |
75b28aff | 113 | struct io_rings { |
1e84b97b SB |
114 | /* |
115 | * Head and tail offsets into the ring; the offsets need to be | |
116 | * masked to get valid indices. | |
117 | * | |
75b28aff HV |
118 | * The kernel controls head of the sq ring and the tail of the cq ring, |
119 | * and the application controls tail of the sq ring and the head of the | |
120 | * cq ring. | |
1e84b97b | 121 | */ |
75b28aff | 122 | struct io_uring sq, cq; |
1e84b97b | 123 | /* |
75b28aff | 124 | * Bitmasks to apply to head and tail offsets (constant, equals |
1e84b97b SB |
125 | * ring_entries - 1) |
126 | */ | |
75b28aff HV |
127 | u32 sq_ring_mask, cq_ring_mask; |
128 | /* Ring sizes (constant, power of 2) */ | |
129 | u32 sq_ring_entries, cq_ring_entries; | |
1e84b97b SB |
130 | /* |
131 | * Number of invalid entries dropped by the kernel due to | |
132 | * invalid index stored in array | |
133 | * | |
134 | * Written by the kernel, shouldn't be modified by the | |
135 | * application (i.e. get number of "new events" by comparing to | |
136 | * cached value). | |
137 | * | |
138 | * After a new SQ head value was read by the application this | |
139 | * counter includes all submissions that were dropped reaching | |
140 | * the new SQ head (and possibly more). | |
141 | */ | |
75b28aff | 142 | u32 sq_dropped; |
1e84b97b SB |
143 | /* |
144 | * Runtime flags | |
145 | * | |
146 | * Written by the kernel, shouldn't be modified by the | |
147 | * application. | |
148 | * | |
149 | * The application needs a full memory barrier before checking | |
150 | * for IORING_SQ_NEED_WAKEUP after updating the sq tail. | |
151 | */ | |
75b28aff | 152 | u32 sq_flags; |
1e84b97b SB |
153 | /* |
154 | * Number of completion events lost because the queue was full; | |
155 | * this should be avoided by the application by making sure | |
0b4295b5 | 156 | * there are not more requests pending than there is space in |
1e84b97b SB |
157 | * the completion queue. |
158 | * | |
159 | * Written by the kernel, shouldn't be modified by the | |
160 | * application (i.e. get number of "new events" by comparing to | |
161 | * cached value). | |
162 | * | |
163 | * As completion events come in out of order this counter is not | |
164 | * ordered with any other data. | |
165 | */ | |
75b28aff | 166 | u32 cq_overflow; |
1e84b97b SB |
167 | /* |
168 | * Ring buffer of completion events. | |
169 | * | |
170 | * The kernel writes completion events fresh every time they are | |
171 | * produced, so the application is allowed to modify pending | |
172 | * entries. | |
173 | */ | |
75b28aff | 174 | struct io_uring_cqe cqes[] ____cacheline_aligned_in_smp; |
2b188cc1 JA |
175 | }; |
176 | ||
edafccee JA |
177 | struct io_mapped_ubuf { |
178 | u64 ubuf; | |
179 | size_t len; | |
180 | struct bio_vec *bvec; | |
181 | unsigned int nr_bvecs; | |
182 | }; | |
183 | ||
65e19f54 JA |
184 | struct fixed_file_table { |
185 | struct file **files; | |
31b51510 JA |
186 | }; |
187 | ||
05f3fb3c JA |
188 | struct fixed_file_data { |
189 | struct fixed_file_table *table; | |
190 | struct io_ring_ctx *ctx; | |
191 | ||
192 | struct percpu_ref refs; | |
193 | struct llist_head put_llist; | |
05f3fb3c JA |
194 | struct work_struct ref_work; |
195 | struct completion done; | |
196 | }; | |
197 | ||
2b188cc1 JA |
198 | struct io_ring_ctx { |
199 | struct { | |
200 | struct percpu_ref refs; | |
201 | } ____cacheline_aligned_in_smp; | |
202 | ||
203 | struct { | |
204 | unsigned int flags; | |
e1d85334 RD |
205 | unsigned int compat: 1; |
206 | unsigned int account_mem: 1; | |
207 | unsigned int cq_overflow_flushed: 1; | |
208 | unsigned int drain_next: 1; | |
209 | unsigned int eventfd_async: 1; | |
2b188cc1 | 210 | |
75b28aff HV |
211 | /* |
212 | * Ring buffer of indices into array of io_uring_sqe, which is | |
213 | * mmapped by the application using the IORING_OFF_SQES offset. | |
214 | * | |
215 | * This indirection could e.g. be used to assign fixed | |
216 | * io_uring_sqe entries to operations and only submit them to | |
217 | * the queue when needed. | |
218 | * | |
219 | * The kernel modifies neither the indices array nor the entries | |
220 | * array. | |
221 | */ | |
222 | u32 *sq_array; | |
2b188cc1 JA |
223 | unsigned cached_sq_head; |
224 | unsigned sq_entries; | |
225 | unsigned sq_mask; | |
6c271ce2 | 226 | unsigned sq_thread_idle; |
498ccd9e | 227 | unsigned cached_sq_dropped; |
206aefde | 228 | atomic_t cached_cq_overflow; |
ad3eb2c8 | 229 | unsigned long sq_check_overflow; |
de0617e4 JA |
230 | |
231 | struct list_head defer_list; | |
5262f567 | 232 | struct list_head timeout_list; |
1d7bb1d5 | 233 | struct list_head cq_overflow_list; |
fcb323cc JA |
234 | |
235 | wait_queue_head_t inflight_wait; | |
ad3eb2c8 | 236 | struct io_uring_sqe *sq_sqes; |
2b188cc1 JA |
237 | } ____cacheline_aligned_in_smp; |
238 | ||
206aefde JA |
239 | struct io_rings *rings; |
240 | ||
2b188cc1 | 241 | /* IO offload */ |
561fb04a | 242 | struct io_wq *io_wq; |
6c271ce2 | 243 | struct task_struct *sqo_thread; /* if using sq thread polling */ |
2b188cc1 | 244 | struct mm_struct *sqo_mm; |
6c271ce2 | 245 | wait_queue_head_t sqo_wait; |
75b28aff | 246 | |
6b06314c JA |
247 | /* |
248 | * If used, fixed file set. Writers must ensure that ->refs is dead, | |
249 | * readers must ensure that ->refs is alive as long as the file* is | |
250 | * used. Only updated through io_uring_register(2). | |
251 | */ | |
05f3fb3c | 252 | struct fixed_file_data *file_data; |
6b06314c | 253 | unsigned nr_user_files; |
b14cca0c PB |
254 | int ring_fd; |
255 | struct file *ring_file; | |
6b06314c | 256 | |
edafccee JA |
257 | /* if used, fixed mapped user buffers */ |
258 | unsigned nr_user_bufs; | |
259 | struct io_mapped_ubuf *user_bufs; | |
260 | ||
2b188cc1 JA |
261 | struct user_struct *user; |
262 | ||
0b8c0ec7 | 263 | const struct cred *creds; |
181e448d | 264 | |
206aefde JA |
265 | /* 0 is for ctx quiesce/reinit/free, 1 is for sqo_thread started */ |
266 | struct completion *completions; | |
267 | ||
0ddf92e8 JA |
268 | /* if all else fails... */ |
269 | struct io_kiocb *fallback_req; | |
270 | ||
206aefde JA |
271 | #if defined(CONFIG_UNIX) |
272 | struct socket *ring_sock; | |
273 | #endif | |
274 | ||
071698e1 JA |
275 | struct idr personality_idr; |
276 | ||
206aefde JA |
277 | struct { |
278 | unsigned cached_cq_tail; | |
279 | unsigned cq_entries; | |
280 | unsigned cq_mask; | |
281 | atomic_t cq_timeouts; | |
ad3eb2c8 | 282 | unsigned long cq_check_overflow; |
206aefde JA |
283 | struct wait_queue_head cq_wait; |
284 | struct fasync_struct *cq_fasync; | |
285 | struct eventfd_ctx *cq_ev_fd; | |
286 | } ____cacheline_aligned_in_smp; | |
2b188cc1 JA |
287 | |
288 | struct { | |
289 | struct mutex uring_lock; | |
290 | wait_queue_head_t wait; | |
291 | } ____cacheline_aligned_in_smp; | |
292 | ||
293 | struct { | |
294 | spinlock_t completion_lock; | |
e94f141b | 295 | |
def596e9 JA |
296 | /* |
297 | * ->poll_list is protected by the ctx->uring_lock for | |
298 | * io_uring instances that don't use IORING_SETUP_SQPOLL. | |
299 | * For SQPOLL, only the single threaded io_sq_thread() will | |
300 | * manipulate the list, hence no extra locking is needed there. | |
301 | */ | |
302 | struct list_head poll_list; | |
78076bb6 JA |
303 | struct hlist_head *cancel_hash; |
304 | unsigned cancel_hash_bits; | |
e94f141b | 305 | bool poll_multi_file; |
31b51510 | 306 | |
fcb323cc JA |
307 | spinlock_t inflight_lock; |
308 | struct list_head inflight_list; | |
2b188cc1 | 309 | } ____cacheline_aligned_in_smp; |
2b188cc1 JA |
310 | }; |
311 | ||
09bb8394 JA |
312 | /* |
313 | * First field must be the file pointer in all the | |
314 | * iocb unions! See also 'struct kiocb' in <linux/fs.h> | |
315 | */ | |
221c5eb2 JA |
316 | struct io_poll_iocb { |
317 | struct file *file; | |
0969e783 JA |
318 | union { |
319 | struct wait_queue_head *head; | |
320 | u64 addr; | |
321 | }; | |
221c5eb2 | 322 | __poll_t events; |
8c838788 | 323 | bool done; |
221c5eb2 | 324 | bool canceled; |
392edb45 | 325 | struct wait_queue_entry wait; |
221c5eb2 JA |
326 | }; |
327 | ||
b5dba59e JA |
328 | struct io_close { |
329 | struct file *file; | |
330 | struct file *put_file; | |
331 | int fd; | |
332 | }; | |
333 | ||
ad8a48ac JA |
334 | struct io_timeout_data { |
335 | struct io_kiocb *req; | |
336 | struct hrtimer timer; | |
337 | struct timespec64 ts; | |
338 | enum hrtimer_mode mode; | |
cc42e0ac | 339 | u32 seq_offset; |
ad8a48ac JA |
340 | }; |
341 | ||
8ed8d3c3 JA |
342 | struct io_accept { |
343 | struct file *file; | |
344 | struct sockaddr __user *addr; | |
345 | int __user *addr_len; | |
346 | int flags; | |
347 | }; | |
348 | ||
349 | struct io_sync { | |
350 | struct file *file; | |
351 | loff_t len; | |
352 | loff_t off; | |
353 | int flags; | |
d63d1b5e | 354 | int mode; |
8ed8d3c3 JA |
355 | }; |
356 | ||
fbf23849 JA |
357 | struct io_cancel { |
358 | struct file *file; | |
359 | u64 addr; | |
360 | }; | |
361 | ||
b29472ee JA |
362 | struct io_timeout { |
363 | struct file *file; | |
364 | u64 addr; | |
365 | int flags; | |
26a61679 | 366 | unsigned count; |
b29472ee JA |
367 | }; |
368 | ||
9adbd45d JA |
369 | struct io_rw { |
370 | /* NOTE: kiocb has the file as the first member, so don't do it here */ | |
371 | struct kiocb kiocb; | |
372 | u64 addr; | |
373 | u64 len; | |
374 | }; | |
375 | ||
3fbb51c1 JA |
376 | struct io_connect { |
377 | struct file *file; | |
378 | struct sockaddr __user *addr; | |
379 | int addr_len; | |
380 | }; | |
381 | ||
e47293fd JA |
382 | struct io_sr_msg { |
383 | struct file *file; | |
fddaface JA |
384 | union { |
385 | struct user_msghdr __user *msg; | |
386 | void __user *buf; | |
387 | }; | |
e47293fd | 388 | int msg_flags; |
fddaface | 389 | size_t len; |
e47293fd JA |
390 | }; |
391 | ||
15b71abe JA |
392 | struct io_open { |
393 | struct file *file; | |
394 | int dfd; | |
eddc7ef5 | 395 | union { |
eddc7ef5 JA |
396 | unsigned mask; |
397 | }; | |
15b71abe | 398 | struct filename *filename; |
eddc7ef5 | 399 | struct statx __user *buffer; |
c12cedf2 | 400 | struct open_how how; |
15b71abe JA |
401 | }; |
402 | ||
05f3fb3c JA |
403 | struct io_files_update { |
404 | struct file *file; | |
405 | u64 arg; | |
406 | u32 nr_args; | |
407 | u32 offset; | |
408 | }; | |
409 | ||
4840e418 JA |
410 | struct io_fadvise { |
411 | struct file *file; | |
412 | u64 offset; | |
413 | u32 len; | |
414 | u32 advice; | |
415 | }; | |
416 | ||
c1ca757b JA |
417 | struct io_madvise { |
418 | struct file *file; | |
419 | u64 addr; | |
420 | u32 len; | |
421 | u32 advice; | |
422 | }; | |
423 | ||
3e4827b0 JA |
424 | struct io_epoll { |
425 | struct file *file; | |
426 | int epfd; | |
427 | int op; | |
428 | int fd; | |
429 | struct epoll_event event; | |
e47293fd JA |
430 | }; |
431 | ||
7d67af2c PB |
432 | struct io_splice { |
433 | struct file *file_out; | |
434 | struct file *file_in; | |
435 | loff_t off_out; | |
436 | loff_t off_in; | |
437 | u64 len; | |
438 | unsigned int flags; | |
439 | }; | |
440 | ||
f499a021 JA |
441 | struct io_async_connect { |
442 | struct sockaddr_storage address; | |
443 | }; | |
444 | ||
03b1230c JA |
445 | struct io_async_msghdr { |
446 | struct iovec fast_iov[UIO_FASTIOV]; | |
447 | struct iovec *iov; | |
448 | struct sockaddr __user *uaddr; | |
449 | struct msghdr msg; | |
b537916c | 450 | struct sockaddr_storage addr; |
03b1230c JA |
451 | }; |
452 | ||
f67676d1 JA |
453 | struct io_async_rw { |
454 | struct iovec fast_iov[UIO_FASTIOV]; | |
455 | struct iovec *iov; | |
456 | ssize_t nr_segs; | |
457 | ssize_t size; | |
458 | }; | |
459 | ||
1a6b74fc | 460 | struct io_async_ctx { |
f67676d1 JA |
461 | union { |
462 | struct io_async_rw rw; | |
03b1230c | 463 | struct io_async_msghdr msg; |
f499a021 | 464 | struct io_async_connect connect; |
2d28390a | 465 | struct io_timeout_data timeout; |
f67676d1 | 466 | }; |
1a6b74fc JA |
467 | }; |
468 | ||
6b47ee6e PB |
469 | enum { |
470 | REQ_F_FIXED_FILE_BIT = IOSQE_FIXED_FILE_BIT, | |
471 | REQ_F_IO_DRAIN_BIT = IOSQE_IO_DRAIN_BIT, | |
472 | REQ_F_LINK_BIT = IOSQE_IO_LINK_BIT, | |
473 | REQ_F_HARDLINK_BIT = IOSQE_IO_HARDLINK_BIT, | |
474 | REQ_F_FORCE_ASYNC_BIT = IOSQE_ASYNC_BIT, | |
475 | ||
476 | REQ_F_LINK_NEXT_BIT, | |
477 | REQ_F_FAIL_LINK_BIT, | |
478 | REQ_F_INFLIGHT_BIT, | |
479 | REQ_F_CUR_POS_BIT, | |
480 | REQ_F_NOWAIT_BIT, | |
481 | REQ_F_IOPOLL_COMPLETED_BIT, | |
482 | REQ_F_LINK_TIMEOUT_BIT, | |
483 | REQ_F_TIMEOUT_BIT, | |
484 | REQ_F_ISREG_BIT, | |
485 | REQ_F_MUST_PUNT_BIT, | |
486 | REQ_F_TIMEOUT_NOSEQ_BIT, | |
487 | REQ_F_COMP_LOCKED_BIT, | |
99bc4c38 | 488 | REQ_F_NEED_CLEANUP_BIT, |
2ca10259 | 489 | REQ_F_OVERFLOW_BIT, |
d7718a9d | 490 | REQ_F_POLLED_BIT, |
6b47ee6e PB |
491 | }; |
492 | ||
493 | enum { | |
494 | /* ctx owns file */ | |
495 | REQ_F_FIXED_FILE = BIT(REQ_F_FIXED_FILE_BIT), | |
496 | /* drain existing IO first */ | |
497 | REQ_F_IO_DRAIN = BIT(REQ_F_IO_DRAIN_BIT), | |
498 | /* linked sqes */ | |
499 | REQ_F_LINK = BIT(REQ_F_LINK_BIT), | |
500 | /* doesn't sever on completion < 0 */ | |
501 | REQ_F_HARDLINK = BIT(REQ_F_HARDLINK_BIT), | |
502 | /* IOSQE_ASYNC */ | |
503 | REQ_F_FORCE_ASYNC = BIT(REQ_F_FORCE_ASYNC_BIT), | |
504 | ||
505 | /* already grabbed next link */ | |
506 | REQ_F_LINK_NEXT = BIT(REQ_F_LINK_NEXT_BIT), | |
507 | /* fail rest of links */ | |
508 | REQ_F_FAIL_LINK = BIT(REQ_F_FAIL_LINK_BIT), | |
509 | /* on inflight list */ | |
510 | REQ_F_INFLIGHT = BIT(REQ_F_INFLIGHT_BIT), | |
511 | /* read/write uses file position */ | |
512 | REQ_F_CUR_POS = BIT(REQ_F_CUR_POS_BIT), | |
513 | /* must not punt to workers */ | |
514 | REQ_F_NOWAIT = BIT(REQ_F_NOWAIT_BIT), | |
515 | /* polled IO has completed */ | |
516 | REQ_F_IOPOLL_COMPLETED = BIT(REQ_F_IOPOLL_COMPLETED_BIT), | |
517 | /* has linked timeout */ | |
518 | REQ_F_LINK_TIMEOUT = BIT(REQ_F_LINK_TIMEOUT_BIT), | |
519 | /* timeout request */ | |
520 | REQ_F_TIMEOUT = BIT(REQ_F_TIMEOUT_BIT), | |
521 | /* regular file */ | |
522 | REQ_F_ISREG = BIT(REQ_F_ISREG_BIT), | |
523 | /* must be punted even for NONBLOCK */ | |
524 | REQ_F_MUST_PUNT = BIT(REQ_F_MUST_PUNT_BIT), | |
525 | /* no timeout sequence */ | |
526 | REQ_F_TIMEOUT_NOSEQ = BIT(REQ_F_TIMEOUT_NOSEQ_BIT), | |
527 | /* completion under lock */ | |
528 | REQ_F_COMP_LOCKED = BIT(REQ_F_COMP_LOCKED_BIT), | |
99bc4c38 PB |
529 | /* needs cleanup */ |
530 | REQ_F_NEED_CLEANUP = BIT(REQ_F_NEED_CLEANUP_BIT), | |
2ca10259 JA |
531 | /* in overflow list */ |
532 | REQ_F_OVERFLOW = BIT(REQ_F_OVERFLOW_BIT), | |
d7718a9d JA |
533 | /* already went through poll handler */ |
534 | REQ_F_POLLED = BIT(REQ_F_POLLED_BIT), | |
535 | }; | |
536 | ||
537 | struct async_poll { | |
538 | struct io_poll_iocb poll; | |
539 | struct io_wq_work work; | |
6b47ee6e PB |
540 | }; |
541 | ||
09bb8394 JA |
542 | /* |
543 | * NOTE! Each of the iocb union members has the file pointer | |
544 | * as the first entry in their struct definition. So you can | |
545 | * access the file pointer through any of the sub-structs, | |
546 | * or directly as just 'ki_filp' in this struct. | |
547 | */ | |
2b188cc1 | 548 | struct io_kiocb { |
221c5eb2 | 549 | union { |
09bb8394 | 550 | struct file *file; |
9adbd45d | 551 | struct io_rw rw; |
221c5eb2 | 552 | struct io_poll_iocb poll; |
8ed8d3c3 JA |
553 | struct io_accept accept; |
554 | struct io_sync sync; | |
fbf23849 | 555 | struct io_cancel cancel; |
b29472ee | 556 | struct io_timeout timeout; |
3fbb51c1 | 557 | struct io_connect connect; |
e47293fd | 558 | struct io_sr_msg sr_msg; |
15b71abe | 559 | struct io_open open; |
b5dba59e | 560 | struct io_close close; |
05f3fb3c | 561 | struct io_files_update files_update; |
4840e418 | 562 | struct io_fadvise fadvise; |
c1ca757b | 563 | struct io_madvise madvise; |
3e4827b0 | 564 | struct io_epoll epoll; |
7d67af2c | 565 | struct io_splice splice; |
221c5eb2 | 566 | }; |
2b188cc1 | 567 | |
1a6b74fc | 568 | struct io_async_ctx *io; |
cf6fd4bd | 569 | bool needs_fixed_file; |
d625c6ee | 570 | u8 opcode; |
2b188cc1 JA |
571 | |
572 | struct io_ring_ctx *ctx; | |
d7718a9d | 573 | struct list_head list; |
2b188cc1 | 574 | unsigned int flags; |
c16361c1 | 575 | refcount_t refs; |
d7718a9d | 576 | struct task_struct *task; |
2b188cc1 | 577 | u64 user_data; |
9e645e11 | 578 | u32 result; |
de0617e4 | 579 | u32 sequence; |
2b188cc1 | 580 | |
d7718a9d JA |
581 | struct list_head link_list; |
582 | ||
fcb323cc JA |
583 | struct list_head inflight_entry; |
584 | ||
b41e9852 JA |
585 | union { |
586 | /* | |
587 | * Only commands that never go async can use the below fields, | |
d7718a9d JA |
588 | * obviously. Right now only IORING_OP_POLL_ADD uses them, and |
589 | * async armed poll handlers for regular commands. The latter | |
590 | * restore the work, if needed. | |
b41e9852 JA |
591 | */ |
592 | struct { | |
b41e9852 | 593 | struct callback_head task_work; |
d7718a9d JA |
594 | struct hlist_node hash_node; |
595 | struct async_poll *apoll; | |
b41e9852 JA |
596 | }; |
597 | struct io_wq_work work; | |
598 | }; | |
2b188cc1 JA |
599 | }; |
600 | ||
601 | #define IO_PLUG_THRESHOLD 2 | |
def596e9 | 602 | #define IO_IOPOLL_BATCH 8 |
2b188cc1 | 603 | |
9a56a232 JA |
604 | struct io_submit_state { |
605 | struct blk_plug plug; | |
606 | ||
2579f913 JA |
607 | /* |
608 | * io_kiocb alloc cache | |
609 | */ | |
610 | void *reqs[IO_IOPOLL_BATCH]; | |
6c8a3134 | 611 | unsigned int free_reqs; |
2579f913 | 612 | |
9a56a232 JA |
613 | /* |
614 | * File reference cache | |
615 | */ | |
616 | struct file *file; | |
617 | unsigned int fd; | |
618 | unsigned int has_refs; | |
619 | unsigned int used_refs; | |
620 | unsigned int ios_left; | |
621 | }; | |
622 | ||
d3656344 JA |
623 | struct io_op_def { |
624 | /* needs req->io allocated for deferral/async */ | |
625 | unsigned async_ctx : 1; | |
626 | /* needs current->mm setup, does mm access */ | |
627 | unsigned needs_mm : 1; | |
628 | /* needs req->file assigned */ | |
629 | unsigned needs_file : 1; | |
630 | /* needs req->file assigned IFF fd is >= 0 */ | |
631 | unsigned fd_non_neg : 1; | |
632 | /* hash wq insertion if file is a regular file */ | |
633 | unsigned hash_reg_file : 1; | |
634 | /* unbound wq insertion if file is a non-regular file */ | |
635 | unsigned unbound_nonreg_file : 1; | |
66f4af93 JA |
636 | /* opcode is not supported by this kernel */ |
637 | unsigned not_supported : 1; | |
f86cd20c JA |
638 | /* needs file table */ |
639 | unsigned file_table : 1; | |
ff002b30 JA |
640 | /* needs ->fs */ |
641 | unsigned needs_fs : 1; | |
8a72758c JA |
642 | /* set if opcode supports polled "wait" */ |
643 | unsigned pollin : 1; | |
644 | unsigned pollout : 1; | |
d3656344 JA |
645 | }; |
646 | ||
647 | static const struct io_op_def io_op_defs[] = { | |
0463b6c5 PB |
648 | [IORING_OP_NOP] = {}, |
649 | [IORING_OP_READV] = { | |
d3656344 JA |
650 | .async_ctx = 1, |
651 | .needs_mm = 1, | |
652 | .needs_file = 1, | |
653 | .unbound_nonreg_file = 1, | |
8a72758c | 654 | .pollin = 1, |
d3656344 | 655 | }, |
0463b6c5 | 656 | [IORING_OP_WRITEV] = { |
d3656344 JA |
657 | .async_ctx = 1, |
658 | .needs_mm = 1, | |
659 | .needs_file = 1, | |
660 | .hash_reg_file = 1, | |
661 | .unbound_nonreg_file = 1, | |
8a72758c | 662 | .pollout = 1, |
d3656344 | 663 | }, |
0463b6c5 | 664 | [IORING_OP_FSYNC] = { |
d3656344 JA |
665 | .needs_file = 1, |
666 | }, | |
0463b6c5 | 667 | [IORING_OP_READ_FIXED] = { |
d3656344 JA |
668 | .needs_file = 1, |
669 | .unbound_nonreg_file = 1, | |
8a72758c | 670 | .pollin = 1, |
d3656344 | 671 | }, |
0463b6c5 | 672 | [IORING_OP_WRITE_FIXED] = { |
d3656344 JA |
673 | .needs_file = 1, |
674 | .hash_reg_file = 1, | |
675 | .unbound_nonreg_file = 1, | |
8a72758c | 676 | .pollout = 1, |
d3656344 | 677 | }, |
0463b6c5 | 678 | [IORING_OP_POLL_ADD] = { |
d3656344 JA |
679 | .needs_file = 1, |
680 | .unbound_nonreg_file = 1, | |
681 | }, | |
0463b6c5 PB |
682 | [IORING_OP_POLL_REMOVE] = {}, |
683 | [IORING_OP_SYNC_FILE_RANGE] = { | |
d3656344 JA |
684 | .needs_file = 1, |
685 | }, | |
0463b6c5 | 686 | [IORING_OP_SENDMSG] = { |
d3656344 JA |
687 | .async_ctx = 1, |
688 | .needs_mm = 1, | |
689 | .needs_file = 1, | |
690 | .unbound_nonreg_file = 1, | |
ff002b30 | 691 | .needs_fs = 1, |
8a72758c | 692 | .pollout = 1, |
d3656344 | 693 | }, |
0463b6c5 | 694 | [IORING_OP_RECVMSG] = { |
d3656344 JA |
695 | .async_ctx = 1, |
696 | .needs_mm = 1, | |
697 | .needs_file = 1, | |
698 | .unbound_nonreg_file = 1, | |
ff002b30 | 699 | .needs_fs = 1, |
8a72758c | 700 | .pollin = 1, |
d3656344 | 701 | }, |
0463b6c5 | 702 | [IORING_OP_TIMEOUT] = { |
d3656344 JA |
703 | .async_ctx = 1, |
704 | .needs_mm = 1, | |
705 | }, | |
0463b6c5 PB |
706 | [IORING_OP_TIMEOUT_REMOVE] = {}, |
707 | [IORING_OP_ACCEPT] = { | |
d3656344 JA |
708 | .needs_mm = 1, |
709 | .needs_file = 1, | |
710 | .unbound_nonreg_file = 1, | |
f86cd20c | 711 | .file_table = 1, |
8a72758c | 712 | .pollin = 1, |
d3656344 | 713 | }, |
0463b6c5 PB |
714 | [IORING_OP_ASYNC_CANCEL] = {}, |
715 | [IORING_OP_LINK_TIMEOUT] = { | |
d3656344 JA |
716 | .async_ctx = 1, |
717 | .needs_mm = 1, | |
718 | }, | |
0463b6c5 | 719 | [IORING_OP_CONNECT] = { |
d3656344 JA |
720 | .async_ctx = 1, |
721 | .needs_mm = 1, | |
722 | .needs_file = 1, | |
723 | .unbound_nonreg_file = 1, | |
8a72758c | 724 | .pollout = 1, |
d3656344 | 725 | }, |
0463b6c5 | 726 | [IORING_OP_FALLOCATE] = { |
d3656344 JA |
727 | .needs_file = 1, |
728 | }, | |
0463b6c5 | 729 | [IORING_OP_OPENAT] = { |
d3656344 JA |
730 | .needs_file = 1, |
731 | .fd_non_neg = 1, | |
f86cd20c | 732 | .file_table = 1, |
ff002b30 | 733 | .needs_fs = 1, |
d3656344 | 734 | }, |
0463b6c5 | 735 | [IORING_OP_CLOSE] = { |
d3656344 | 736 | .needs_file = 1, |
f86cd20c | 737 | .file_table = 1, |
d3656344 | 738 | }, |
0463b6c5 | 739 | [IORING_OP_FILES_UPDATE] = { |
d3656344 | 740 | .needs_mm = 1, |
f86cd20c | 741 | .file_table = 1, |
d3656344 | 742 | }, |
0463b6c5 | 743 | [IORING_OP_STATX] = { |
d3656344 JA |
744 | .needs_mm = 1, |
745 | .needs_file = 1, | |
746 | .fd_non_neg = 1, | |
ff002b30 | 747 | .needs_fs = 1, |
d3656344 | 748 | }, |
0463b6c5 | 749 | [IORING_OP_READ] = { |
3a6820f2 JA |
750 | .needs_mm = 1, |
751 | .needs_file = 1, | |
752 | .unbound_nonreg_file = 1, | |
8a72758c | 753 | .pollin = 1, |
3a6820f2 | 754 | }, |
0463b6c5 | 755 | [IORING_OP_WRITE] = { |
3a6820f2 JA |
756 | .needs_mm = 1, |
757 | .needs_file = 1, | |
758 | .unbound_nonreg_file = 1, | |
8a72758c | 759 | .pollout = 1, |
3a6820f2 | 760 | }, |
0463b6c5 | 761 | [IORING_OP_FADVISE] = { |
4840e418 JA |
762 | .needs_file = 1, |
763 | }, | |
0463b6c5 | 764 | [IORING_OP_MADVISE] = { |
c1ca757b JA |
765 | .needs_mm = 1, |
766 | }, | |
0463b6c5 | 767 | [IORING_OP_SEND] = { |
fddaface JA |
768 | .needs_mm = 1, |
769 | .needs_file = 1, | |
770 | .unbound_nonreg_file = 1, | |
8a72758c | 771 | .pollout = 1, |
fddaface | 772 | }, |
0463b6c5 | 773 | [IORING_OP_RECV] = { |
fddaface JA |
774 | .needs_mm = 1, |
775 | .needs_file = 1, | |
776 | .unbound_nonreg_file = 1, | |
8a72758c | 777 | .pollin = 1, |
fddaface | 778 | }, |
0463b6c5 | 779 | [IORING_OP_OPENAT2] = { |
cebdb986 JA |
780 | .needs_file = 1, |
781 | .fd_non_neg = 1, | |
f86cd20c | 782 | .file_table = 1, |
ff002b30 | 783 | .needs_fs = 1, |
cebdb986 | 784 | }, |
3e4827b0 JA |
785 | [IORING_OP_EPOLL_CTL] = { |
786 | .unbound_nonreg_file = 1, | |
787 | .file_table = 1, | |
788 | }, | |
7d67af2c PB |
789 | [IORING_OP_SPLICE] = { |
790 | .needs_file = 1, | |
791 | .hash_reg_file = 1, | |
792 | .unbound_nonreg_file = 1, | |
793 | } | |
d3656344 JA |
794 | }; |
795 | ||
561fb04a | 796 | static void io_wq_submit_work(struct io_wq_work **workptr); |
78e19bbe | 797 | static void io_cqring_fill_event(struct io_kiocb *req, long res); |
ec9c02ad | 798 | static void io_put_req(struct io_kiocb *req); |
978db57e | 799 | static void __io_double_put_req(struct io_kiocb *req); |
94ae5e77 JA |
800 | static struct io_kiocb *io_prep_linked_timeout(struct io_kiocb *req); |
801 | static void io_queue_linked_timeout(struct io_kiocb *req); | |
05f3fb3c JA |
802 | static int __io_sqe_files_update(struct io_ring_ctx *ctx, |
803 | struct io_uring_files_update *ip, | |
804 | unsigned nr_args); | |
f86cd20c | 805 | static int io_grab_files(struct io_kiocb *req); |
2faf852d | 806 | static void io_ring_file_ref_flush(struct fixed_file_data *data); |
99bc4c38 | 807 | static void io_cleanup_req(struct io_kiocb *req); |
b41e9852 JA |
808 | static int io_file_get(struct io_submit_state *state, struct io_kiocb *req, |
809 | int fd, struct file **out_file, bool fixed); | |
810 | static void __io_queue_sqe(struct io_kiocb *req, | |
811 | const struct io_uring_sqe *sqe); | |
de0617e4 | 812 | |
2b188cc1 JA |
813 | static struct kmem_cache *req_cachep; |
814 | ||
815 | static const struct file_operations io_uring_fops; | |
816 | ||
817 | struct sock *io_uring_get_socket(struct file *file) | |
818 | { | |
819 | #if defined(CONFIG_UNIX) | |
820 | if (file->f_op == &io_uring_fops) { | |
821 | struct io_ring_ctx *ctx = file->private_data; | |
822 | ||
823 | return ctx->ring_sock->sk; | |
824 | } | |
825 | #endif | |
826 | return NULL; | |
827 | } | |
828 | EXPORT_SYMBOL(io_uring_get_socket); | |
829 | ||
830 | static void io_ring_ctx_ref_free(struct percpu_ref *ref) | |
831 | { | |
832 | struct io_ring_ctx *ctx = container_of(ref, struct io_ring_ctx, refs); | |
833 | ||
206aefde | 834 | complete(&ctx->completions[0]); |
2b188cc1 JA |
835 | } |
836 | ||
837 | static struct io_ring_ctx *io_ring_ctx_alloc(struct io_uring_params *p) | |
838 | { | |
839 | struct io_ring_ctx *ctx; | |
78076bb6 | 840 | int hash_bits; |
2b188cc1 JA |
841 | |
842 | ctx = kzalloc(sizeof(*ctx), GFP_KERNEL); | |
843 | if (!ctx) | |
844 | return NULL; | |
845 | ||
0ddf92e8 JA |
846 | ctx->fallback_req = kmem_cache_alloc(req_cachep, GFP_KERNEL); |
847 | if (!ctx->fallback_req) | |
848 | goto err; | |
849 | ||
206aefde JA |
850 | ctx->completions = kmalloc(2 * sizeof(struct completion), GFP_KERNEL); |
851 | if (!ctx->completions) | |
852 | goto err; | |
853 | ||
78076bb6 JA |
854 | /* |
855 | * Use 5 bits less than the max cq entries, that should give us around | |
856 | * 32 entries per hash list if totally full and uniformly spread. | |
857 | */ | |
858 | hash_bits = ilog2(p->cq_entries); | |
859 | hash_bits -= 5; | |
860 | if (hash_bits <= 0) | |
861 | hash_bits = 1; | |
862 | ctx->cancel_hash_bits = hash_bits; | |
863 | ctx->cancel_hash = kmalloc((1U << hash_bits) * sizeof(struct hlist_head), | |
864 | GFP_KERNEL); | |
865 | if (!ctx->cancel_hash) | |
866 | goto err; | |
867 | __hash_init(ctx->cancel_hash, 1U << hash_bits); | |
868 | ||
21482896 | 869 | if (percpu_ref_init(&ctx->refs, io_ring_ctx_ref_free, |
206aefde JA |
870 | PERCPU_REF_ALLOW_REINIT, GFP_KERNEL)) |
871 | goto err; | |
2b188cc1 JA |
872 | |
873 | ctx->flags = p->flags; | |
874 | init_waitqueue_head(&ctx->cq_wait); | |
1d7bb1d5 | 875 | INIT_LIST_HEAD(&ctx->cq_overflow_list); |
206aefde JA |
876 | init_completion(&ctx->completions[0]); |
877 | init_completion(&ctx->completions[1]); | |
071698e1 | 878 | idr_init(&ctx->personality_idr); |
2b188cc1 JA |
879 | mutex_init(&ctx->uring_lock); |
880 | init_waitqueue_head(&ctx->wait); | |
881 | spin_lock_init(&ctx->completion_lock); | |
def596e9 | 882 | INIT_LIST_HEAD(&ctx->poll_list); |
de0617e4 | 883 | INIT_LIST_HEAD(&ctx->defer_list); |
5262f567 | 884 | INIT_LIST_HEAD(&ctx->timeout_list); |
fcb323cc JA |
885 | init_waitqueue_head(&ctx->inflight_wait); |
886 | spin_lock_init(&ctx->inflight_lock); | |
887 | INIT_LIST_HEAD(&ctx->inflight_list); | |
2b188cc1 | 888 | return ctx; |
206aefde | 889 | err: |
0ddf92e8 JA |
890 | if (ctx->fallback_req) |
891 | kmem_cache_free(req_cachep, ctx->fallback_req); | |
206aefde | 892 | kfree(ctx->completions); |
78076bb6 | 893 | kfree(ctx->cancel_hash); |
206aefde JA |
894 | kfree(ctx); |
895 | return NULL; | |
2b188cc1 JA |
896 | } |
897 | ||
9d858b21 | 898 | static inline bool __req_need_defer(struct io_kiocb *req) |
7adf4eaf | 899 | { |
a197f664 JL |
900 | struct io_ring_ctx *ctx = req->ctx; |
901 | ||
498ccd9e JA |
902 | return req->sequence != ctx->cached_cq_tail + ctx->cached_sq_dropped |
903 | + atomic_read(&ctx->cached_cq_overflow); | |
7adf4eaf JA |
904 | } |
905 | ||
9d858b21 | 906 | static inline bool req_need_defer(struct io_kiocb *req) |
de0617e4 | 907 | { |
87987898 | 908 | if (unlikely(req->flags & REQ_F_IO_DRAIN)) |
9d858b21 | 909 | return __req_need_defer(req); |
de0617e4 | 910 | |
9d858b21 | 911 | return false; |
de0617e4 JA |
912 | } |
913 | ||
7adf4eaf | 914 | static struct io_kiocb *io_get_deferred_req(struct io_ring_ctx *ctx) |
de0617e4 JA |
915 | { |
916 | struct io_kiocb *req; | |
917 | ||
7adf4eaf | 918 | req = list_first_entry_or_null(&ctx->defer_list, struct io_kiocb, list); |
9d858b21 | 919 | if (req && !req_need_defer(req)) { |
de0617e4 JA |
920 | list_del_init(&req->list); |
921 | return req; | |
922 | } | |
923 | ||
924 | return NULL; | |
925 | } | |
926 | ||
5262f567 JA |
927 | static struct io_kiocb *io_get_timeout_req(struct io_ring_ctx *ctx) |
928 | { | |
7adf4eaf JA |
929 | struct io_kiocb *req; |
930 | ||
931 | req = list_first_entry_or_null(&ctx->timeout_list, struct io_kiocb, list); | |
93bd25bb JA |
932 | if (req) { |
933 | if (req->flags & REQ_F_TIMEOUT_NOSEQ) | |
934 | return NULL; | |
fb4b3d3f | 935 | if (!__req_need_defer(req)) { |
93bd25bb JA |
936 | list_del_init(&req->list); |
937 | return req; | |
938 | } | |
7adf4eaf JA |
939 | } |
940 | ||
941 | return NULL; | |
5262f567 JA |
942 | } |
943 | ||
de0617e4 | 944 | static void __io_commit_cqring(struct io_ring_ctx *ctx) |
2b188cc1 | 945 | { |
75b28aff | 946 | struct io_rings *rings = ctx->rings; |
2b188cc1 | 947 | |
07910158 PB |
948 | /* order cqe stores with ring update */ |
949 | smp_store_release(&rings->cq.tail, ctx->cached_cq_tail); | |
2b188cc1 | 950 | |
07910158 PB |
951 | if (wq_has_sleeper(&ctx->cq_wait)) { |
952 | wake_up_interruptible(&ctx->cq_wait); | |
953 | kill_fasync(&ctx->cq_fasync, SIGIO, POLL_IN); | |
2b188cc1 JA |
954 | } |
955 | } | |
956 | ||
cccf0ee8 JA |
957 | static inline void io_req_work_grab_env(struct io_kiocb *req, |
958 | const struct io_op_def *def) | |
959 | { | |
960 | if (!req->work.mm && def->needs_mm) { | |
961 | mmgrab(current->mm); | |
962 | req->work.mm = current->mm; | |
2b188cc1 | 963 | } |
cccf0ee8 JA |
964 | if (!req->work.creds) |
965 | req->work.creds = get_current_cred(); | |
ff002b30 JA |
966 | if (!req->work.fs && def->needs_fs) { |
967 | spin_lock(¤t->fs->lock); | |
968 | if (!current->fs->in_exec) { | |
969 | req->work.fs = current->fs; | |
970 | req->work.fs->users++; | |
971 | } else { | |
972 | req->work.flags |= IO_WQ_WORK_CANCEL; | |
973 | } | |
974 | spin_unlock(¤t->fs->lock); | |
975 | } | |
6ab23144 JA |
976 | if (!req->work.task_pid) |
977 | req->work.task_pid = task_pid_vnr(current); | |
2b188cc1 JA |
978 | } |
979 | ||
cccf0ee8 | 980 | static inline void io_req_work_drop_env(struct io_kiocb *req) |
18d9be1a | 981 | { |
cccf0ee8 JA |
982 | if (req->work.mm) { |
983 | mmdrop(req->work.mm); | |
984 | req->work.mm = NULL; | |
985 | } | |
986 | if (req->work.creds) { | |
987 | put_cred(req->work.creds); | |
988 | req->work.creds = NULL; | |
989 | } | |
ff002b30 JA |
990 | if (req->work.fs) { |
991 | struct fs_struct *fs = req->work.fs; | |
992 | ||
993 | spin_lock(&req->work.fs->lock); | |
994 | if (--fs->users) | |
995 | fs = NULL; | |
996 | spin_unlock(&req->work.fs->lock); | |
997 | if (fs) | |
998 | free_fs_struct(fs); | |
999 | } | |
561fb04a JA |
1000 | } |
1001 | ||
94ae5e77 JA |
1002 | static inline bool io_prep_async_work(struct io_kiocb *req, |
1003 | struct io_kiocb **link) | |
18d9be1a | 1004 | { |
d3656344 | 1005 | const struct io_op_def *def = &io_op_defs[req->opcode]; |
561fb04a | 1006 | bool do_hashed = false; |
54a91f3b | 1007 | |
d3656344 JA |
1008 | if (req->flags & REQ_F_ISREG) { |
1009 | if (def->hash_reg_file) | |
3529d8c2 | 1010 | do_hashed = true; |
d3656344 JA |
1011 | } else { |
1012 | if (def->unbound_nonreg_file) | |
3529d8c2 | 1013 | req->work.flags |= IO_WQ_WORK_UNBOUND; |
54a91f3b | 1014 | } |
cccf0ee8 JA |
1015 | |
1016 | io_req_work_grab_env(req, def); | |
54a91f3b | 1017 | |
94ae5e77 | 1018 | *link = io_prep_linked_timeout(req); |
561fb04a JA |
1019 | return do_hashed; |
1020 | } | |
1021 | ||
a197f664 | 1022 | static inline void io_queue_async_work(struct io_kiocb *req) |
561fb04a | 1023 | { |
a197f664 | 1024 | struct io_ring_ctx *ctx = req->ctx; |
94ae5e77 JA |
1025 | struct io_kiocb *link; |
1026 | bool do_hashed; | |
1027 | ||
1028 | do_hashed = io_prep_async_work(req, &link); | |
561fb04a JA |
1029 | |
1030 | trace_io_uring_queue_async_work(ctx, do_hashed, req, &req->work, | |
1031 | req->flags); | |
1032 | if (!do_hashed) { | |
1033 | io_wq_enqueue(ctx->io_wq, &req->work); | |
1034 | } else { | |
1035 | io_wq_enqueue_hashed(ctx->io_wq, &req->work, | |
1036 | file_inode(req->file)); | |
1037 | } | |
94ae5e77 JA |
1038 | |
1039 | if (link) | |
1040 | io_queue_linked_timeout(link); | |
18d9be1a JA |
1041 | } |
1042 | ||
5262f567 JA |
1043 | static void io_kill_timeout(struct io_kiocb *req) |
1044 | { | |
1045 | int ret; | |
1046 | ||
2d28390a | 1047 | ret = hrtimer_try_to_cancel(&req->io->timeout.timer); |
5262f567 JA |
1048 | if (ret != -1) { |
1049 | atomic_inc(&req->ctx->cq_timeouts); | |
842f9612 | 1050 | list_del_init(&req->list); |
78e19bbe | 1051 | io_cqring_fill_event(req, 0); |
ec9c02ad | 1052 | io_put_req(req); |
5262f567 JA |
1053 | } |
1054 | } | |
1055 | ||
1056 | static void io_kill_timeouts(struct io_ring_ctx *ctx) | |
1057 | { | |
1058 | struct io_kiocb *req, *tmp; | |
1059 | ||
1060 | spin_lock_irq(&ctx->completion_lock); | |
1061 | list_for_each_entry_safe(req, tmp, &ctx->timeout_list, list) | |
1062 | io_kill_timeout(req); | |
1063 | spin_unlock_irq(&ctx->completion_lock); | |
1064 | } | |
1065 | ||
de0617e4 JA |
1066 | static void io_commit_cqring(struct io_ring_ctx *ctx) |
1067 | { | |
1068 | struct io_kiocb *req; | |
1069 | ||
5262f567 JA |
1070 | while ((req = io_get_timeout_req(ctx)) != NULL) |
1071 | io_kill_timeout(req); | |
1072 | ||
de0617e4 JA |
1073 | __io_commit_cqring(ctx); |
1074 | ||
87987898 | 1075 | while ((req = io_get_deferred_req(ctx)) != NULL) |
a197f664 | 1076 | io_queue_async_work(req); |
de0617e4 JA |
1077 | } |
1078 | ||
2b188cc1 JA |
1079 | static struct io_uring_cqe *io_get_cqring(struct io_ring_ctx *ctx) |
1080 | { | |
75b28aff | 1081 | struct io_rings *rings = ctx->rings; |
2b188cc1 JA |
1082 | unsigned tail; |
1083 | ||
1084 | tail = ctx->cached_cq_tail; | |
115e12e5 SB |
1085 | /* |
1086 | * writes to the cq entry need to come after reading head; the | |
1087 | * control dependency is enough as we're using WRITE_ONCE to | |
1088 | * fill the cq entry | |
1089 | */ | |
75b28aff | 1090 | if (tail - READ_ONCE(rings->cq.head) == rings->cq_ring_entries) |
2b188cc1 JA |
1091 | return NULL; |
1092 | ||
1093 | ctx->cached_cq_tail++; | |
75b28aff | 1094 | return &rings->cqes[tail & ctx->cq_mask]; |
2b188cc1 JA |
1095 | } |
1096 | ||
f2842ab5 JA |
1097 | static inline bool io_should_trigger_evfd(struct io_ring_ctx *ctx) |
1098 | { | |
f0b493e6 JA |
1099 | if (!ctx->cq_ev_fd) |
1100 | return false; | |
f2842ab5 JA |
1101 | if (!ctx->eventfd_async) |
1102 | return true; | |
b41e9852 | 1103 | return io_wq_current_is_worker(); |
f2842ab5 JA |
1104 | } |
1105 | ||
b41e9852 | 1106 | static void io_cqring_ev_posted(struct io_ring_ctx *ctx) |
1d7bb1d5 JA |
1107 | { |
1108 | if (waitqueue_active(&ctx->wait)) | |
1109 | wake_up(&ctx->wait); | |
1110 | if (waitqueue_active(&ctx->sqo_wait)) | |
1111 | wake_up(&ctx->sqo_wait); | |
b41e9852 | 1112 | if (io_should_trigger_evfd(ctx)) |
1d7bb1d5 JA |
1113 | eventfd_signal(ctx->cq_ev_fd, 1); |
1114 | } | |
1115 | ||
c4a2ed72 JA |
1116 | /* Returns true if there are no backlogged entries after the flush */ |
1117 | static bool io_cqring_overflow_flush(struct io_ring_ctx *ctx, bool force) | |
1d7bb1d5 JA |
1118 | { |
1119 | struct io_rings *rings = ctx->rings; | |
1120 | struct io_uring_cqe *cqe; | |
1121 | struct io_kiocb *req; | |
1122 | unsigned long flags; | |
1123 | LIST_HEAD(list); | |
1124 | ||
1125 | if (!force) { | |
1126 | if (list_empty_careful(&ctx->cq_overflow_list)) | |
c4a2ed72 | 1127 | return true; |
1d7bb1d5 JA |
1128 | if ((ctx->cached_cq_tail - READ_ONCE(rings->cq.head) == |
1129 | rings->cq_ring_entries)) | |
c4a2ed72 | 1130 | return false; |
1d7bb1d5 JA |
1131 | } |
1132 | ||
1133 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
1134 | ||
1135 | /* if force is set, the ring is going away. always drop after that */ | |
1136 | if (force) | |
69b3e546 | 1137 | ctx->cq_overflow_flushed = 1; |
1d7bb1d5 | 1138 | |
c4a2ed72 | 1139 | cqe = NULL; |
1d7bb1d5 JA |
1140 | while (!list_empty(&ctx->cq_overflow_list)) { |
1141 | cqe = io_get_cqring(ctx); | |
1142 | if (!cqe && !force) | |
1143 | break; | |
1144 | ||
1145 | req = list_first_entry(&ctx->cq_overflow_list, struct io_kiocb, | |
1146 | list); | |
1147 | list_move(&req->list, &list); | |
2ca10259 | 1148 | req->flags &= ~REQ_F_OVERFLOW; |
1d7bb1d5 JA |
1149 | if (cqe) { |
1150 | WRITE_ONCE(cqe->user_data, req->user_data); | |
1151 | WRITE_ONCE(cqe->res, req->result); | |
1152 | WRITE_ONCE(cqe->flags, 0); | |
1153 | } else { | |
1154 | WRITE_ONCE(ctx->rings->cq_overflow, | |
1155 | atomic_inc_return(&ctx->cached_cq_overflow)); | |
1156 | } | |
1157 | } | |
1158 | ||
1159 | io_commit_cqring(ctx); | |
ad3eb2c8 JA |
1160 | if (cqe) { |
1161 | clear_bit(0, &ctx->sq_check_overflow); | |
1162 | clear_bit(0, &ctx->cq_check_overflow); | |
1163 | } | |
1d7bb1d5 JA |
1164 | spin_unlock_irqrestore(&ctx->completion_lock, flags); |
1165 | io_cqring_ev_posted(ctx); | |
1166 | ||
1167 | while (!list_empty(&list)) { | |
1168 | req = list_first_entry(&list, struct io_kiocb, list); | |
1169 | list_del(&req->list); | |
ec9c02ad | 1170 | io_put_req(req); |
1d7bb1d5 | 1171 | } |
c4a2ed72 JA |
1172 | |
1173 | return cqe != NULL; | |
1d7bb1d5 JA |
1174 | } |
1175 | ||
78e19bbe | 1176 | static void io_cqring_fill_event(struct io_kiocb *req, long res) |
2b188cc1 | 1177 | { |
78e19bbe | 1178 | struct io_ring_ctx *ctx = req->ctx; |
2b188cc1 JA |
1179 | struct io_uring_cqe *cqe; |
1180 | ||
78e19bbe | 1181 | trace_io_uring_complete(ctx, req->user_data, res); |
51c3ff62 | 1182 | |
2b188cc1 JA |
1183 | /* |
1184 | * If we can't get a cq entry, userspace overflowed the | |
1185 | * submission (by quite a lot). Increment the overflow count in | |
1186 | * the ring. | |
1187 | */ | |
1188 | cqe = io_get_cqring(ctx); | |
1d7bb1d5 | 1189 | if (likely(cqe)) { |
78e19bbe | 1190 | WRITE_ONCE(cqe->user_data, req->user_data); |
2b188cc1 | 1191 | WRITE_ONCE(cqe->res, res); |
c71ffb67 | 1192 | WRITE_ONCE(cqe->flags, 0); |
1d7bb1d5 | 1193 | } else if (ctx->cq_overflow_flushed) { |
498ccd9e JA |
1194 | WRITE_ONCE(ctx->rings->cq_overflow, |
1195 | atomic_inc_return(&ctx->cached_cq_overflow)); | |
1d7bb1d5 | 1196 | } else { |
ad3eb2c8 JA |
1197 | if (list_empty(&ctx->cq_overflow_list)) { |
1198 | set_bit(0, &ctx->sq_check_overflow); | |
1199 | set_bit(0, &ctx->cq_check_overflow); | |
1200 | } | |
2ca10259 | 1201 | req->flags |= REQ_F_OVERFLOW; |
1d7bb1d5 JA |
1202 | refcount_inc(&req->refs); |
1203 | req->result = res; | |
1204 | list_add_tail(&req->list, &ctx->cq_overflow_list); | |
2b188cc1 JA |
1205 | } |
1206 | } | |
1207 | ||
78e19bbe | 1208 | static void io_cqring_add_event(struct io_kiocb *req, long res) |
2b188cc1 | 1209 | { |
78e19bbe | 1210 | struct io_ring_ctx *ctx = req->ctx; |
2b188cc1 JA |
1211 | unsigned long flags; |
1212 | ||
1213 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
78e19bbe | 1214 | io_cqring_fill_event(req, res); |
2b188cc1 JA |
1215 | io_commit_cqring(ctx); |
1216 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
1217 | ||
8c838788 | 1218 | io_cqring_ev_posted(ctx); |
2b188cc1 JA |
1219 | } |
1220 | ||
0ddf92e8 JA |
1221 | static inline bool io_is_fallback_req(struct io_kiocb *req) |
1222 | { | |
1223 | return req == (struct io_kiocb *) | |
1224 | ((unsigned long) req->ctx->fallback_req & ~1UL); | |
1225 | } | |
1226 | ||
1227 | static struct io_kiocb *io_get_fallback_req(struct io_ring_ctx *ctx) | |
1228 | { | |
1229 | struct io_kiocb *req; | |
1230 | ||
1231 | req = ctx->fallback_req; | |
1232 | if (!test_and_set_bit_lock(0, (unsigned long *) ctx->fallback_req)) | |
1233 | return req; | |
1234 | ||
1235 | return NULL; | |
1236 | } | |
1237 | ||
2579f913 JA |
1238 | static struct io_kiocb *io_get_req(struct io_ring_ctx *ctx, |
1239 | struct io_submit_state *state) | |
2b188cc1 | 1240 | { |
fd6fab2c | 1241 | gfp_t gfp = GFP_KERNEL | __GFP_NOWARN; |
2b188cc1 JA |
1242 | struct io_kiocb *req; |
1243 | ||
2579f913 | 1244 | if (!state) { |
fd6fab2c | 1245 | req = kmem_cache_alloc(req_cachep, gfp); |
2579f913 | 1246 | if (unlikely(!req)) |
0ddf92e8 | 1247 | goto fallback; |
2579f913 JA |
1248 | } else if (!state->free_reqs) { |
1249 | size_t sz; | |
1250 | int ret; | |
1251 | ||
1252 | sz = min_t(size_t, state->ios_left, ARRAY_SIZE(state->reqs)); | |
fd6fab2c JA |
1253 | ret = kmem_cache_alloc_bulk(req_cachep, gfp, sz, state->reqs); |
1254 | ||
1255 | /* | |
1256 | * Bulk alloc is all-or-nothing. If we fail to get a batch, | |
1257 | * retry single alloc to be on the safe side. | |
1258 | */ | |
1259 | if (unlikely(ret <= 0)) { | |
1260 | state->reqs[0] = kmem_cache_alloc(req_cachep, gfp); | |
1261 | if (!state->reqs[0]) | |
0ddf92e8 | 1262 | goto fallback; |
fd6fab2c JA |
1263 | ret = 1; |
1264 | } | |
2579f913 | 1265 | state->free_reqs = ret - 1; |
6c8a3134 | 1266 | req = state->reqs[ret - 1]; |
2579f913 | 1267 | } else { |
2579f913 | 1268 | state->free_reqs--; |
6c8a3134 | 1269 | req = state->reqs[state->free_reqs]; |
2b188cc1 JA |
1270 | } |
1271 | ||
0ddf92e8 | 1272 | got_it: |
1a6b74fc | 1273 | req->io = NULL; |
60c112b0 | 1274 | req->file = NULL; |
2579f913 JA |
1275 | req->ctx = ctx; |
1276 | req->flags = 0; | |
e65ef56d JA |
1277 | /* one is dropped after submission, the other at completion */ |
1278 | refcount_set(&req->refs, 2); | |
9e645e11 | 1279 | req->result = 0; |
561fb04a | 1280 | INIT_IO_WORK(&req->work, io_wq_submit_work); |
2579f913 | 1281 | return req; |
0ddf92e8 JA |
1282 | fallback: |
1283 | req = io_get_fallback_req(ctx); | |
1284 | if (req) | |
1285 | goto got_it; | |
6805b32e | 1286 | percpu_ref_put(&ctx->refs); |
2b188cc1 JA |
1287 | return NULL; |
1288 | } | |
1289 | ||
8da11c19 PB |
1290 | static inline void io_put_file(struct io_kiocb *req, struct file *file, |
1291 | bool fixed) | |
1292 | { | |
1293 | if (fixed) | |
1294 | percpu_ref_put(&req->ctx->file_data->refs); | |
1295 | else | |
1296 | fput(file); | |
1297 | } | |
1298 | ||
2b85edfc | 1299 | static void __io_req_do_free(struct io_kiocb *req) |
def596e9 | 1300 | { |
2b85edfc PB |
1301 | if (likely(!io_is_fallback_req(req))) |
1302 | kmem_cache_free(req_cachep, req); | |
1303 | else | |
1304 | clear_bit_unlock(0, (unsigned long *) req->ctx->fallback_req); | |
1305 | } | |
1306 | ||
c6ca97b3 | 1307 | static void __io_req_aux_free(struct io_kiocb *req) |
2b188cc1 | 1308 | { |
929a3af9 PB |
1309 | if (req->flags & REQ_F_NEED_CLEANUP) |
1310 | io_cleanup_req(req); | |
1311 | ||
96fd84d8 | 1312 | kfree(req->io); |
8da11c19 PB |
1313 | if (req->file) |
1314 | io_put_file(req, req->file, (req->flags & REQ_F_FIXED_FILE)); | |
cccf0ee8 JA |
1315 | |
1316 | io_req_work_drop_env(req); | |
def596e9 JA |
1317 | } |
1318 | ||
9e645e11 | 1319 | static void __io_free_req(struct io_kiocb *req) |
2b188cc1 | 1320 | { |
c6ca97b3 | 1321 | __io_req_aux_free(req); |
fcb323cc | 1322 | |
fcb323cc | 1323 | if (req->flags & REQ_F_INFLIGHT) { |
c6ca97b3 | 1324 | struct io_ring_ctx *ctx = req->ctx; |
fcb323cc JA |
1325 | unsigned long flags; |
1326 | ||
1327 | spin_lock_irqsave(&ctx->inflight_lock, flags); | |
1328 | list_del(&req->inflight_entry); | |
1329 | if (waitqueue_active(&ctx->inflight_wait)) | |
1330 | wake_up(&ctx->inflight_wait); | |
1331 | spin_unlock_irqrestore(&ctx->inflight_lock, flags); | |
1332 | } | |
2b85edfc PB |
1333 | |
1334 | percpu_ref_put(&req->ctx->refs); | |
1335 | __io_req_do_free(req); | |
e65ef56d JA |
1336 | } |
1337 | ||
c6ca97b3 JA |
1338 | struct req_batch { |
1339 | void *reqs[IO_IOPOLL_BATCH]; | |
1340 | int to_free; | |
1341 | int need_iter; | |
1342 | }; | |
1343 | ||
1344 | static void io_free_req_many(struct io_ring_ctx *ctx, struct req_batch *rb) | |
1345 | { | |
10fef4be JA |
1346 | int fixed_refs = rb->to_free; |
1347 | ||
c6ca97b3 JA |
1348 | if (!rb->to_free) |
1349 | return; | |
1350 | if (rb->need_iter) { | |
1351 | int i, inflight = 0; | |
1352 | unsigned long flags; | |
1353 | ||
10fef4be | 1354 | fixed_refs = 0; |
c6ca97b3 JA |
1355 | for (i = 0; i < rb->to_free; i++) { |
1356 | struct io_kiocb *req = rb->reqs[i]; | |
1357 | ||
10fef4be | 1358 | if (req->flags & REQ_F_FIXED_FILE) { |
c6ca97b3 | 1359 | req->file = NULL; |
10fef4be JA |
1360 | fixed_refs++; |
1361 | } | |
c6ca97b3 JA |
1362 | if (req->flags & REQ_F_INFLIGHT) |
1363 | inflight++; | |
c6ca97b3 JA |
1364 | __io_req_aux_free(req); |
1365 | } | |
1366 | if (!inflight) | |
1367 | goto do_free; | |
1368 | ||
1369 | spin_lock_irqsave(&ctx->inflight_lock, flags); | |
1370 | for (i = 0; i < rb->to_free; i++) { | |
1371 | struct io_kiocb *req = rb->reqs[i]; | |
1372 | ||
10fef4be | 1373 | if (req->flags & REQ_F_INFLIGHT) { |
c6ca97b3 JA |
1374 | list_del(&req->inflight_entry); |
1375 | if (!--inflight) | |
1376 | break; | |
1377 | } | |
1378 | } | |
1379 | spin_unlock_irqrestore(&ctx->inflight_lock, flags); | |
1380 | ||
1381 | if (waitqueue_active(&ctx->inflight_wait)) | |
1382 | wake_up(&ctx->inflight_wait); | |
1383 | } | |
1384 | do_free: | |
1385 | kmem_cache_free_bulk(req_cachep, rb->to_free, rb->reqs); | |
10fef4be JA |
1386 | if (fixed_refs) |
1387 | percpu_ref_put_many(&ctx->file_data->refs, fixed_refs); | |
c6ca97b3 | 1388 | percpu_ref_put_many(&ctx->refs, rb->to_free); |
c6ca97b3 | 1389 | rb->to_free = rb->need_iter = 0; |
e65ef56d JA |
1390 | } |
1391 | ||
a197f664 | 1392 | static bool io_link_cancel_timeout(struct io_kiocb *req) |
2665abfd | 1393 | { |
a197f664 | 1394 | struct io_ring_ctx *ctx = req->ctx; |
2665abfd JA |
1395 | int ret; |
1396 | ||
2d28390a | 1397 | ret = hrtimer_try_to_cancel(&req->io->timeout.timer); |
2665abfd | 1398 | if (ret != -1) { |
78e19bbe | 1399 | io_cqring_fill_event(req, -ECANCELED); |
2665abfd JA |
1400 | io_commit_cqring(ctx); |
1401 | req->flags &= ~REQ_F_LINK; | |
ec9c02ad | 1402 | io_put_req(req); |
2665abfd JA |
1403 | return true; |
1404 | } | |
1405 | ||
1406 | return false; | |
e65ef56d JA |
1407 | } |
1408 | ||
ba816ad6 | 1409 | static void io_req_link_next(struct io_kiocb *req, struct io_kiocb **nxtptr) |
9e645e11 | 1410 | { |
2665abfd | 1411 | struct io_ring_ctx *ctx = req->ctx; |
2665abfd | 1412 | bool wake_ev = false; |
9e645e11 | 1413 | |
4d7dd462 JA |
1414 | /* Already got next link */ |
1415 | if (req->flags & REQ_F_LINK_NEXT) | |
1416 | return; | |
1417 | ||
9e645e11 JA |
1418 | /* |
1419 | * The list should never be empty when we are called here. But could | |
1420 | * potentially happen if the chain is messed up, check to be on the | |
1421 | * safe side. | |
1422 | */ | |
4493233e PB |
1423 | while (!list_empty(&req->link_list)) { |
1424 | struct io_kiocb *nxt = list_first_entry(&req->link_list, | |
1425 | struct io_kiocb, link_list); | |
94ae5e77 | 1426 | |
4493233e PB |
1427 | if (unlikely((req->flags & REQ_F_LINK_TIMEOUT) && |
1428 | (nxt->flags & REQ_F_TIMEOUT))) { | |
1429 | list_del_init(&nxt->link_list); | |
94ae5e77 | 1430 | wake_ev |= io_link_cancel_timeout(nxt); |
94ae5e77 JA |
1431 | req->flags &= ~REQ_F_LINK_TIMEOUT; |
1432 | continue; | |
1433 | } | |
9e645e11 | 1434 | |
4493233e PB |
1435 | list_del_init(&req->link_list); |
1436 | if (!list_empty(&nxt->link_list)) | |
1437 | nxt->flags |= REQ_F_LINK; | |
b18fdf71 | 1438 | *nxtptr = nxt; |
94ae5e77 | 1439 | break; |
9e645e11 | 1440 | } |
2665abfd | 1441 | |
4d7dd462 | 1442 | req->flags |= REQ_F_LINK_NEXT; |
2665abfd JA |
1443 | if (wake_ev) |
1444 | io_cqring_ev_posted(ctx); | |
9e645e11 JA |
1445 | } |
1446 | ||
1447 | /* | |
1448 | * Called if REQ_F_LINK is set, and we fail the head request | |
1449 | */ | |
1450 | static void io_fail_links(struct io_kiocb *req) | |
1451 | { | |
2665abfd | 1452 | struct io_ring_ctx *ctx = req->ctx; |
2665abfd JA |
1453 | unsigned long flags; |
1454 | ||
1455 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
9e645e11 JA |
1456 | |
1457 | while (!list_empty(&req->link_list)) { | |
4493233e PB |
1458 | struct io_kiocb *link = list_first_entry(&req->link_list, |
1459 | struct io_kiocb, link_list); | |
9e645e11 | 1460 | |
4493233e | 1461 | list_del_init(&link->link_list); |
c826bd7a | 1462 | trace_io_uring_fail_link(req, link); |
2665abfd JA |
1463 | |
1464 | if ((req->flags & REQ_F_LINK_TIMEOUT) && | |
d625c6ee | 1465 | link->opcode == IORING_OP_LINK_TIMEOUT) { |
a197f664 | 1466 | io_link_cancel_timeout(link); |
2665abfd | 1467 | } else { |
78e19bbe | 1468 | io_cqring_fill_event(link, -ECANCELED); |
978db57e | 1469 | __io_double_put_req(link); |
2665abfd | 1470 | } |
5d960724 | 1471 | req->flags &= ~REQ_F_LINK_TIMEOUT; |
9e645e11 | 1472 | } |
2665abfd JA |
1473 | |
1474 | io_commit_cqring(ctx); | |
1475 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
1476 | io_cqring_ev_posted(ctx); | |
9e645e11 JA |
1477 | } |
1478 | ||
4d7dd462 | 1479 | static void io_req_find_next(struct io_kiocb *req, struct io_kiocb **nxt) |
9e645e11 | 1480 | { |
4d7dd462 | 1481 | if (likely(!(req->flags & REQ_F_LINK))) |
2665abfd | 1482 | return; |
2665abfd | 1483 | |
9e645e11 JA |
1484 | /* |
1485 | * If LINK is set, we have dependent requests in this chain. If we | |
1486 | * didn't fail this request, queue the first one up, moving any other | |
1487 | * dependencies to the next request. In case of failure, fail the rest | |
1488 | * of the chain. | |
1489 | */ | |
2665abfd JA |
1490 | if (req->flags & REQ_F_FAIL_LINK) { |
1491 | io_fail_links(req); | |
7c9e7f0f JA |
1492 | } else if ((req->flags & (REQ_F_LINK_TIMEOUT | REQ_F_COMP_LOCKED)) == |
1493 | REQ_F_LINK_TIMEOUT) { | |
2665abfd JA |
1494 | struct io_ring_ctx *ctx = req->ctx; |
1495 | unsigned long flags; | |
1496 | ||
1497 | /* | |
1498 | * If this is a timeout link, we could be racing with the | |
1499 | * timeout timer. Grab the completion lock for this case to | |
7c9e7f0f | 1500 | * protect against that. |
2665abfd JA |
1501 | */ |
1502 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
1503 | io_req_link_next(req, nxt); | |
1504 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
1505 | } else { | |
1506 | io_req_link_next(req, nxt); | |
9e645e11 | 1507 | } |
4d7dd462 | 1508 | } |
9e645e11 | 1509 | |
c69f8dbe JL |
1510 | static void io_free_req(struct io_kiocb *req) |
1511 | { | |
944e58bf PB |
1512 | struct io_kiocb *nxt = NULL; |
1513 | ||
1514 | io_req_find_next(req, &nxt); | |
70cf9f32 | 1515 | __io_free_req(req); |
944e58bf PB |
1516 | |
1517 | if (nxt) | |
1518 | io_queue_async_work(nxt); | |
c69f8dbe JL |
1519 | } |
1520 | ||
ba816ad6 JA |
1521 | /* |
1522 | * Drop reference to request, return next in chain (if there is one) if this | |
1523 | * was the last reference to this request. | |
1524 | */ | |
f9bd67f6 | 1525 | __attribute__((nonnull)) |
ec9c02ad | 1526 | static void io_put_req_find_next(struct io_kiocb *req, struct io_kiocb **nxtptr) |
e65ef56d | 1527 | { |
2a44f467 JA |
1528 | if (refcount_dec_and_test(&req->refs)) { |
1529 | io_req_find_next(req, nxtptr); | |
4d7dd462 | 1530 | __io_free_req(req); |
2a44f467 | 1531 | } |
2b188cc1 JA |
1532 | } |
1533 | ||
e65ef56d JA |
1534 | static void io_put_req(struct io_kiocb *req) |
1535 | { | |
1536 | if (refcount_dec_and_test(&req->refs)) | |
1537 | io_free_req(req); | |
2b188cc1 JA |
1538 | } |
1539 | ||
978db57e JA |
1540 | /* |
1541 | * Must only be used if we don't need to care about links, usually from | |
1542 | * within the completion handling itself. | |
1543 | */ | |
1544 | static void __io_double_put_req(struct io_kiocb *req) | |
78e19bbe JA |
1545 | { |
1546 | /* drop both submit and complete references */ | |
1547 | if (refcount_sub_and_test(2, &req->refs)) | |
1548 | __io_free_req(req); | |
1549 | } | |
1550 | ||
978db57e JA |
1551 | static void io_double_put_req(struct io_kiocb *req) |
1552 | { | |
1553 | /* drop both submit and complete references */ | |
1554 | if (refcount_sub_and_test(2, &req->refs)) | |
1555 | io_free_req(req); | |
1556 | } | |
1557 | ||
1d7bb1d5 | 1558 | static unsigned io_cqring_events(struct io_ring_ctx *ctx, bool noflush) |
a3a0e43f | 1559 | { |
84f97dc2 JA |
1560 | struct io_rings *rings = ctx->rings; |
1561 | ||
ad3eb2c8 JA |
1562 | if (test_bit(0, &ctx->cq_check_overflow)) { |
1563 | /* | |
1564 | * noflush == true is from the waitqueue handler, just ensure | |
1565 | * we wake up the task, and the next invocation will flush the | |
1566 | * entries. We cannot safely to it from here. | |
1567 | */ | |
1568 | if (noflush && !list_empty(&ctx->cq_overflow_list)) | |
1569 | return -1U; | |
1d7bb1d5 | 1570 | |
ad3eb2c8 JA |
1571 | io_cqring_overflow_flush(ctx, false); |
1572 | } | |
1d7bb1d5 | 1573 | |
a3a0e43f JA |
1574 | /* See comment at the top of this file */ |
1575 | smp_rmb(); | |
ad3eb2c8 | 1576 | return ctx->cached_cq_tail - READ_ONCE(rings->cq.head); |
a3a0e43f JA |
1577 | } |
1578 | ||
fb5ccc98 PB |
1579 | static inline unsigned int io_sqring_entries(struct io_ring_ctx *ctx) |
1580 | { | |
1581 | struct io_rings *rings = ctx->rings; | |
1582 | ||
1583 | /* make sure SQ entry isn't read before tail */ | |
1584 | return smp_load_acquire(&rings->sq.tail) - ctx->cached_sq_head; | |
1585 | } | |
1586 | ||
8237e045 | 1587 | static inline bool io_req_multi_free(struct req_batch *rb, struct io_kiocb *req) |
e94f141b | 1588 | { |
c6ca97b3 JA |
1589 | if ((req->flags & REQ_F_LINK) || io_is_fallback_req(req)) |
1590 | return false; | |
e94f141b | 1591 | |
c6ca97b3 JA |
1592 | if (!(req->flags & REQ_F_FIXED_FILE) || req->io) |
1593 | rb->need_iter++; | |
1594 | ||
1595 | rb->reqs[rb->to_free++] = req; | |
1596 | if (unlikely(rb->to_free == ARRAY_SIZE(rb->reqs))) | |
1597 | io_free_req_many(req->ctx, rb); | |
1598 | return true; | |
e94f141b JA |
1599 | } |
1600 | ||
def596e9 JA |
1601 | /* |
1602 | * Find and free completed poll iocbs | |
1603 | */ | |
1604 | static void io_iopoll_complete(struct io_ring_ctx *ctx, unsigned int *nr_events, | |
1605 | struct list_head *done) | |
1606 | { | |
8237e045 | 1607 | struct req_batch rb; |
def596e9 | 1608 | struct io_kiocb *req; |
def596e9 | 1609 | |
c6ca97b3 | 1610 | rb.to_free = rb.need_iter = 0; |
def596e9 JA |
1611 | while (!list_empty(done)) { |
1612 | req = list_first_entry(done, struct io_kiocb, list); | |
1613 | list_del(&req->list); | |
1614 | ||
78e19bbe | 1615 | io_cqring_fill_event(req, req->result); |
def596e9 JA |
1616 | (*nr_events)++; |
1617 | ||
8237e045 JA |
1618 | if (refcount_dec_and_test(&req->refs) && |
1619 | !io_req_multi_free(&rb, req)) | |
1620 | io_free_req(req); | |
def596e9 | 1621 | } |
def596e9 | 1622 | |
09bb8394 | 1623 | io_commit_cqring(ctx); |
8237e045 | 1624 | io_free_req_many(ctx, &rb); |
def596e9 JA |
1625 | } |
1626 | ||
1627 | static int io_do_iopoll(struct io_ring_ctx *ctx, unsigned int *nr_events, | |
1628 | long min) | |
1629 | { | |
1630 | struct io_kiocb *req, *tmp; | |
1631 | LIST_HEAD(done); | |
1632 | bool spin; | |
1633 | int ret; | |
1634 | ||
1635 | /* | |
1636 | * Only spin for completions if we don't have multiple devices hanging | |
1637 | * off our complete list, and we're under the requested amount. | |
1638 | */ | |
1639 | spin = !ctx->poll_multi_file && *nr_events < min; | |
1640 | ||
1641 | ret = 0; | |
1642 | list_for_each_entry_safe(req, tmp, &ctx->poll_list, list) { | |
9adbd45d | 1643 | struct kiocb *kiocb = &req->rw.kiocb; |
def596e9 JA |
1644 | |
1645 | /* | |
1646 | * Move completed entries to our local list. If we find a | |
1647 | * request that requires polling, break out and complete | |
1648 | * the done list first, if we have entries there. | |
1649 | */ | |
1650 | if (req->flags & REQ_F_IOPOLL_COMPLETED) { | |
1651 | list_move_tail(&req->list, &done); | |
1652 | continue; | |
1653 | } | |
1654 | if (!list_empty(&done)) | |
1655 | break; | |
1656 | ||
1657 | ret = kiocb->ki_filp->f_op->iopoll(kiocb, spin); | |
1658 | if (ret < 0) | |
1659 | break; | |
1660 | ||
1661 | if (ret && spin) | |
1662 | spin = false; | |
1663 | ret = 0; | |
1664 | } | |
1665 | ||
1666 | if (!list_empty(&done)) | |
1667 | io_iopoll_complete(ctx, nr_events, &done); | |
1668 | ||
1669 | return ret; | |
1670 | } | |
1671 | ||
1672 | /* | |
d195a66e | 1673 | * Poll for a minimum of 'min' events. Note that if min == 0 we consider that a |
def596e9 JA |
1674 | * non-spinning poll check - we'll still enter the driver poll loop, but only |
1675 | * as a non-spinning completion check. | |
1676 | */ | |
1677 | static int io_iopoll_getevents(struct io_ring_ctx *ctx, unsigned int *nr_events, | |
1678 | long min) | |
1679 | { | |
08f5439f | 1680 | while (!list_empty(&ctx->poll_list) && !need_resched()) { |
def596e9 JA |
1681 | int ret; |
1682 | ||
1683 | ret = io_do_iopoll(ctx, nr_events, min); | |
1684 | if (ret < 0) | |
1685 | return ret; | |
1686 | if (!min || *nr_events >= min) | |
1687 | return 0; | |
1688 | } | |
1689 | ||
1690 | return 1; | |
1691 | } | |
1692 | ||
1693 | /* | |
1694 | * We can't just wait for polled events to come to us, we have to actively | |
1695 | * find and complete them. | |
1696 | */ | |
1697 | static void io_iopoll_reap_events(struct io_ring_ctx *ctx) | |
1698 | { | |
1699 | if (!(ctx->flags & IORING_SETUP_IOPOLL)) | |
1700 | return; | |
1701 | ||
1702 | mutex_lock(&ctx->uring_lock); | |
1703 | while (!list_empty(&ctx->poll_list)) { | |
1704 | unsigned int nr_events = 0; | |
1705 | ||
1706 | io_iopoll_getevents(ctx, &nr_events, 1); | |
08f5439f JA |
1707 | |
1708 | /* | |
1709 | * Ensure we allow local-to-the-cpu processing to take place, | |
1710 | * in this case we need to ensure that we reap all events. | |
1711 | */ | |
1712 | cond_resched(); | |
def596e9 JA |
1713 | } |
1714 | mutex_unlock(&ctx->uring_lock); | |
1715 | } | |
1716 | ||
c7849be9 XW |
1717 | static int io_iopoll_check(struct io_ring_ctx *ctx, unsigned *nr_events, |
1718 | long min) | |
def596e9 | 1719 | { |
2b2ed975 | 1720 | int iters = 0, ret = 0; |
500f9fba | 1721 | |
c7849be9 XW |
1722 | /* |
1723 | * We disallow the app entering submit/complete with polling, but we | |
1724 | * still need to lock the ring to prevent racing with polled issue | |
1725 | * that got punted to a workqueue. | |
1726 | */ | |
1727 | mutex_lock(&ctx->uring_lock); | |
def596e9 JA |
1728 | do { |
1729 | int tmin = 0; | |
1730 | ||
a3a0e43f JA |
1731 | /* |
1732 | * Don't enter poll loop if we already have events pending. | |
1733 | * If we do, we can potentially be spinning for commands that | |
1734 | * already triggered a CQE (eg in error). | |
1735 | */ | |
1d7bb1d5 | 1736 | if (io_cqring_events(ctx, false)) |
a3a0e43f JA |
1737 | break; |
1738 | ||
500f9fba JA |
1739 | /* |
1740 | * If a submit got punted to a workqueue, we can have the | |
1741 | * application entering polling for a command before it gets | |
1742 | * issued. That app will hold the uring_lock for the duration | |
1743 | * of the poll right here, so we need to take a breather every | |
1744 | * now and then to ensure that the issue has a chance to add | |
1745 | * the poll to the issued list. Otherwise we can spin here | |
1746 | * forever, while the workqueue is stuck trying to acquire the | |
1747 | * very same mutex. | |
1748 | */ | |
1749 | if (!(++iters & 7)) { | |
1750 | mutex_unlock(&ctx->uring_lock); | |
1751 | mutex_lock(&ctx->uring_lock); | |
1752 | } | |
1753 | ||
def596e9 JA |
1754 | if (*nr_events < min) |
1755 | tmin = min - *nr_events; | |
1756 | ||
1757 | ret = io_iopoll_getevents(ctx, nr_events, tmin); | |
1758 | if (ret <= 0) | |
1759 | break; | |
1760 | ret = 0; | |
1761 | } while (min && !*nr_events && !need_resched()); | |
1762 | ||
500f9fba | 1763 | mutex_unlock(&ctx->uring_lock); |
def596e9 JA |
1764 | return ret; |
1765 | } | |
1766 | ||
491381ce | 1767 | static void kiocb_end_write(struct io_kiocb *req) |
2b188cc1 | 1768 | { |
491381ce JA |
1769 | /* |
1770 | * Tell lockdep we inherited freeze protection from submission | |
1771 | * thread. | |
1772 | */ | |
1773 | if (req->flags & REQ_F_ISREG) { | |
1774 | struct inode *inode = file_inode(req->file); | |
2b188cc1 | 1775 | |
491381ce | 1776 | __sb_writers_acquired(inode->i_sb, SB_FREEZE_WRITE); |
2b188cc1 | 1777 | } |
491381ce | 1778 | file_end_write(req->file); |
2b188cc1 JA |
1779 | } |
1780 | ||
4e88d6e7 JA |
1781 | static inline void req_set_fail_links(struct io_kiocb *req) |
1782 | { | |
1783 | if ((req->flags & (REQ_F_LINK | REQ_F_HARDLINK)) == REQ_F_LINK) | |
1784 | req->flags |= REQ_F_FAIL_LINK; | |
1785 | } | |
1786 | ||
ba816ad6 | 1787 | static void io_complete_rw_common(struct kiocb *kiocb, long res) |
2b188cc1 | 1788 | { |
9adbd45d | 1789 | struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb); |
2b188cc1 | 1790 | |
491381ce JA |
1791 | if (kiocb->ki_flags & IOCB_WRITE) |
1792 | kiocb_end_write(req); | |
2b188cc1 | 1793 | |
4e88d6e7 JA |
1794 | if (res != req->result) |
1795 | req_set_fail_links(req); | |
78e19bbe | 1796 | io_cqring_add_event(req, res); |
ba816ad6 JA |
1797 | } |
1798 | ||
1799 | static void io_complete_rw(struct kiocb *kiocb, long res, long res2) | |
1800 | { | |
9adbd45d | 1801 | struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb); |
ba816ad6 JA |
1802 | |
1803 | io_complete_rw_common(kiocb, res); | |
e65ef56d | 1804 | io_put_req(req); |
2b188cc1 JA |
1805 | } |
1806 | ||
def596e9 JA |
1807 | static void io_complete_rw_iopoll(struct kiocb *kiocb, long res, long res2) |
1808 | { | |
9adbd45d | 1809 | struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb); |
def596e9 | 1810 | |
491381ce JA |
1811 | if (kiocb->ki_flags & IOCB_WRITE) |
1812 | kiocb_end_write(req); | |
def596e9 | 1813 | |
4e88d6e7 JA |
1814 | if (res != req->result) |
1815 | req_set_fail_links(req); | |
9e645e11 | 1816 | req->result = res; |
def596e9 JA |
1817 | if (res != -EAGAIN) |
1818 | req->flags |= REQ_F_IOPOLL_COMPLETED; | |
1819 | } | |
1820 | ||
1821 | /* | |
1822 | * After the iocb has been issued, it's safe to be found on the poll list. | |
1823 | * Adding the kiocb to the list AFTER submission ensures that we don't | |
1824 | * find it from a io_iopoll_getevents() thread before the issuer is done | |
1825 | * accessing the kiocb cookie. | |
1826 | */ | |
1827 | static void io_iopoll_req_issued(struct io_kiocb *req) | |
1828 | { | |
1829 | struct io_ring_ctx *ctx = req->ctx; | |
1830 | ||
1831 | /* | |
1832 | * Track whether we have multiple files in our lists. This will impact | |
1833 | * how we do polling eventually, not spinning if we're on potentially | |
1834 | * different devices. | |
1835 | */ | |
1836 | if (list_empty(&ctx->poll_list)) { | |
1837 | ctx->poll_multi_file = false; | |
1838 | } else if (!ctx->poll_multi_file) { | |
1839 | struct io_kiocb *list_req; | |
1840 | ||
1841 | list_req = list_first_entry(&ctx->poll_list, struct io_kiocb, | |
1842 | list); | |
9adbd45d | 1843 | if (list_req->file != req->file) |
def596e9 JA |
1844 | ctx->poll_multi_file = true; |
1845 | } | |
1846 | ||
1847 | /* | |
1848 | * For fast devices, IO may have already completed. If it has, add | |
1849 | * it to the front so we find it first. | |
1850 | */ | |
1851 | if (req->flags & REQ_F_IOPOLL_COMPLETED) | |
1852 | list_add(&req->list, &ctx->poll_list); | |
1853 | else | |
1854 | list_add_tail(&req->list, &ctx->poll_list); | |
bdcd3eab XW |
1855 | |
1856 | if ((ctx->flags & IORING_SETUP_SQPOLL) && | |
1857 | wq_has_sleeper(&ctx->sqo_wait)) | |
1858 | wake_up(&ctx->sqo_wait); | |
def596e9 JA |
1859 | } |
1860 | ||
3d6770fb | 1861 | static void io_file_put(struct io_submit_state *state) |
9a56a232 | 1862 | { |
3d6770fb | 1863 | if (state->file) { |
9a56a232 JA |
1864 | int diff = state->has_refs - state->used_refs; |
1865 | ||
1866 | if (diff) | |
1867 | fput_many(state->file, diff); | |
1868 | state->file = NULL; | |
1869 | } | |
1870 | } | |
1871 | ||
1872 | /* | |
1873 | * Get as many references to a file as we have IOs left in this submission, | |
1874 | * assuming most submissions are for one file, or at least that each file | |
1875 | * has more than one submission. | |
1876 | */ | |
8da11c19 | 1877 | static struct file *__io_file_get(struct io_submit_state *state, int fd) |
9a56a232 JA |
1878 | { |
1879 | if (!state) | |
1880 | return fget(fd); | |
1881 | ||
1882 | if (state->file) { | |
1883 | if (state->fd == fd) { | |
1884 | state->used_refs++; | |
1885 | state->ios_left--; | |
1886 | return state->file; | |
1887 | } | |
3d6770fb | 1888 | io_file_put(state); |
9a56a232 JA |
1889 | } |
1890 | state->file = fget_many(fd, state->ios_left); | |
1891 | if (!state->file) | |
1892 | return NULL; | |
1893 | ||
1894 | state->fd = fd; | |
1895 | state->has_refs = state->ios_left; | |
1896 | state->used_refs = 1; | |
1897 | state->ios_left--; | |
1898 | return state->file; | |
1899 | } | |
1900 | ||
2b188cc1 JA |
1901 | /* |
1902 | * If we tracked the file through the SCM inflight mechanism, we could support | |
1903 | * any file. For now, just ensure that anything potentially problematic is done | |
1904 | * inline. | |
1905 | */ | |
1906 | static bool io_file_supports_async(struct file *file) | |
1907 | { | |
1908 | umode_t mode = file_inode(file)->i_mode; | |
1909 | ||
10d59345 | 1910 | if (S_ISBLK(mode) || S_ISCHR(mode) || S_ISSOCK(mode)) |
2b188cc1 JA |
1911 | return true; |
1912 | if (S_ISREG(mode) && file->f_op != &io_uring_fops) | |
1913 | return true; | |
1914 | ||
1915 | return false; | |
1916 | } | |
1917 | ||
3529d8c2 JA |
1918 | static int io_prep_rw(struct io_kiocb *req, const struct io_uring_sqe *sqe, |
1919 | bool force_nonblock) | |
2b188cc1 | 1920 | { |
def596e9 | 1921 | struct io_ring_ctx *ctx = req->ctx; |
9adbd45d | 1922 | struct kiocb *kiocb = &req->rw.kiocb; |
09bb8394 JA |
1923 | unsigned ioprio; |
1924 | int ret; | |
2b188cc1 | 1925 | |
491381ce JA |
1926 | if (S_ISREG(file_inode(req->file)->i_mode)) |
1927 | req->flags |= REQ_F_ISREG; | |
1928 | ||
2b188cc1 | 1929 | kiocb->ki_pos = READ_ONCE(sqe->off); |
ba04291e JA |
1930 | if (kiocb->ki_pos == -1 && !(req->file->f_mode & FMODE_STREAM)) { |
1931 | req->flags |= REQ_F_CUR_POS; | |
1932 | kiocb->ki_pos = req->file->f_pos; | |
1933 | } | |
2b188cc1 | 1934 | kiocb->ki_hint = ki_hint_validate(file_write_hint(kiocb->ki_filp)); |
3e577dcd PB |
1935 | kiocb->ki_flags = iocb_flags(kiocb->ki_filp); |
1936 | ret = kiocb_set_rw_flags(kiocb, READ_ONCE(sqe->rw_flags)); | |
1937 | if (unlikely(ret)) | |
1938 | return ret; | |
2b188cc1 JA |
1939 | |
1940 | ioprio = READ_ONCE(sqe->ioprio); | |
1941 | if (ioprio) { | |
1942 | ret = ioprio_check_cap(ioprio); | |
1943 | if (ret) | |
09bb8394 | 1944 | return ret; |
2b188cc1 JA |
1945 | |
1946 | kiocb->ki_ioprio = ioprio; | |
1947 | } else | |
1948 | kiocb->ki_ioprio = get_current_ioprio(); | |
1949 | ||
8449eeda | 1950 | /* don't allow async punt if RWF_NOWAIT was requested */ |
491381ce JA |
1951 | if ((kiocb->ki_flags & IOCB_NOWAIT) || |
1952 | (req->file->f_flags & O_NONBLOCK)) | |
8449eeda SB |
1953 | req->flags |= REQ_F_NOWAIT; |
1954 | ||
1955 | if (force_nonblock) | |
2b188cc1 | 1956 | kiocb->ki_flags |= IOCB_NOWAIT; |
8449eeda | 1957 | |
def596e9 | 1958 | if (ctx->flags & IORING_SETUP_IOPOLL) { |
def596e9 JA |
1959 | if (!(kiocb->ki_flags & IOCB_DIRECT) || |
1960 | !kiocb->ki_filp->f_op->iopoll) | |
09bb8394 | 1961 | return -EOPNOTSUPP; |
2b188cc1 | 1962 | |
def596e9 JA |
1963 | kiocb->ki_flags |= IOCB_HIPRI; |
1964 | kiocb->ki_complete = io_complete_rw_iopoll; | |
6873e0bd | 1965 | req->result = 0; |
def596e9 | 1966 | } else { |
09bb8394 JA |
1967 | if (kiocb->ki_flags & IOCB_HIPRI) |
1968 | return -EINVAL; | |
def596e9 JA |
1969 | kiocb->ki_complete = io_complete_rw; |
1970 | } | |
9adbd45d | 1971 | |
3529d8c2 JA |
1972 | req->rw.addr = READ_ONCE(sqe->addr); |
1973 | req->rw.len = READ_ONCE(sqe->len); | |
9adbd45d JA |
1974 | /* we own ->private, reuse it for the buffer index */ |
1975 | req->rw.kiocb.private = (void *) (unsigned long) | |
3529d8c2 | 1976 | READ_ONCE(sqe->buf_index); |
2b188cc1 | 1977 | return 0; |
2b188cc1 JA |
1978 | } |
1979 | ||
1980 | static inline void io_rw_done(struct kiocb *kiocb, ssize_t ret) | |
1981 | { | |
1982 | switch (ret) { | |
1983 | case -EIOCBQUEUED: | |
1984 | break; | |
1985 | case -ERESTARTSYS: | |
1986 | case -ERESTARTNOINTR: | |
1987 | case -ERESTARTNOHAND: | |
1988 | case -ERESTART_RESTARTBLOCK: | |
1989 | /* | |
1990 | * We can't just restart the syscall, since previously | |
1991 | * submitted sqes may already be in progress. Just fail this | |
1992 | * IO with EINTR. | |
1993 | */ | |
1994 | ret = -EINTR; | |
1995 | /* fall through */ | |
1996 | default: | |
1997 | kiocb->ki_complete(kiocb, ret, 0); | |
1998 | } | |
1999 | } | |
2000 | ||
014db007 | 2001 | static void kiocb_done(struct kiocb *kiocb, ssize_t ret) |
ba816ad6 | 2002 | { |
ba04291e JA |
2003 | struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb); |
2004 | ||
2005 | if (req->flags & REQ_F_CUR_POS) | |
2006 | req->file->f_pos = kiocb->ki_pos; | |
bcaec089 | 2007 | if (ret >= 0 && kiocb->ki_complete == io_complete_rw) |
014db007 | 2008 | io_complete_rw(kiocb, ret, 0); |
ba816ad6 JA |
2009 | else |
2010 | io_rw_done(kiocb, ret); | |
2011 | } | |
2012 | ||
9adbd45d | 2013 | static ssize_t io_import_fixed(struct io_kiocb *req, int rw, |
7d009165 | 2014 | struct iov_iter *iter) |
edafccee | 2015 | { |
9adbd45d JA |
2016 | struct io_ring_ctx *ctx = req->ctx; |
2017 | size_t len = req->rw.len; | |
edafccee JA |
2018 | struct io_mapped_ubuf *imu; |
2019 | unsigned index, buf_index; | |
2020 | size_t offset; | |
2021 | u64 buf_addr; | |
2022 | ||
2023 | /* attempt to use fixed buffers without having provided iovecs */ | |
2024 | if (unlikely(!ctx->user_bufs)) | |
2025 | return -EFAULT; | |
2026 | ||
9adbd45d | 2027 | buf_index = (unsigned long) req->rw.kiocb.private; |
edafccee JA |
2028 | if (unlikely(buf_index >= ctx->nr_user_bufs)) |
2029 | return -EFAULT; | |
2030 | ||
2031 | index = array_index_nospec(buf_index, ctx->nr_user_bufs); | |
2032 | imu = &ctx->user_bufs[index]; | |
9adbd45d | 2033 | buf_addr = req->rw.addr; |
edafccee JA |
2034 | |
2035 | /* overflow */ | |
2036 | if (buf_addr + len < buf_addr) | |
2037 | return -EFAULT; | |
2038 | /* not inside the mapped region */ | |
2039 | if (buf_addr < imu->ubuf || buf_addr + len > imu->ubuf + imu->len) | |
2040 | return -EFAULT; | |
2041 | ||
2042 | /* | |
2043 | * May not be a start of buffer, set size appropriately | |
2044 | * and advance us to the beginning. | |
2045 | */ | |
2046 | offset = buf_addr - imu->ubuf; | |
2047 | iov_iter_bvec(iter, rw, imu->bvec, imu->nr_bvecs, offset + len); | |
bd11b3a3 JA |
2048 | |
2049 | if (offset) { | |
2050 | /* | |
2051 | * Don't use iov_iter_advance() here, as it's really slow for | |
2052 | * using the latter parts of a big fixed buffer - it iterates | |
2053 | * over each segment manually. We can cheat a bit here, because | |
2054 | * we know that: | |
2055 | * | |
2056 | * 1) it's a BVEC iter, we set it up | |
2057 | * 2) all bvecs are PAGE_SIZE in size, except potentially the | |
2058 | * first and last bvec | |
2059 | * | |
2060 | * So just find our index, and adjust the iterator afterwards. | |
2061 | * If the offset is within the first bvec (or the whole first | |
2062 | * bvec, just use iov_iter_advance(). This makes it easier | |
2063 | * since we can just skip the first segment, which may not | |
2064 | * be PAGE_SIZE aligned. | |
2065 | */ | |
2066 | const struct bio_vec *bvec = imu->bvec; | |
2067 | ||
2068 | if (offset <= bvec->bv_len) { | |
2069 | iov_iter_advance(iter, offset); | |
2070 | } else { | |
2071 | unsigned long seg_skip; | |
2072 | ||
2073 | /* skip first vec */ | |
2074 | offset -= bvec->bv_len; | |
2075 | seg_skip = 1 + (offset >> PAGE_SHIFT); | |
2076 | ||
2077 | iter->bvec = bvec + seg_skip; | |
2078 | iter->nr_segs -= seg_skip; | |
99c79f66 | 2079 | iter->count -= bvec->bv_len + offset; |
bd11b3a3 | 2080 | iter->iov_offset = offset & ~PAGE_MASK; |
bd11b3a3 JA |
2081 | } |
2082 | } | |
2083 | ||
5e559561 | 2084 | return len; |
edafccee JA |
2085 | } |
2086 | ||
cf6fd4bd PB |
2087 | static ssize_t io_import_iovec(int rw, struct io_kiocb *req, |
2088 | struct iovec **iovec, struct iov_iter *iter) | |
2b188cc1 | 2089 | { |
9adbd45d JA |
2090 | void __user *buf = u64_to_user_ptr(req->rw.addr); |
2091 | size_t sqe_len = req->rw.len; | |
edafccee JA |
2092 | u8 opcode; |
2093 | ||
d625c6ee | 2094 | opcode = req->opcode; |
7d009165 | 2095 | if (opcode == IORING_OP_READ_FIXED || opcode == IORING_OP_WRITE_FIXED) { |
edafccee | 2096 | *iovec = NULL; |
9adbd45d | 2097 | return io_import_fixed(req, rw, iter); |
edafccee | 2098 | } |
2b188cc1 | 2099 | |
9adbd45d JA |
2100 | /* buffer index only valid with fixed read/write */ |
2101 | if (req->rw.kiocb.private) | |
2102 | return -EINVAL; | |
2103 | ||
3a6820f2 JA |
2104 | if (opcode == IORING_OP_READ || opcode == IORING_OP_WRITE) { |
2105 | ssize_t ret; | |
2106 | ret = import_single_range(rw, buf, sqe_len, *iovec, iter); | |
2107 | *iovec = NULL; | |
3a901598 | 2108 | return ret < 0 ? ret : sqe_len; |
3a6820f2 JA |
2109 | } |
2110 | ||
f67676d1 JA |
2111 | if (req->io) { |
2112 | struct io_async_rw *iorw = &req->io->rw; | |
2113 | ||
2114 | *iovec = iorw->iov; | |
2115 | iov_iter_init(iter, rw, *iovec, iorw->nr_segs, iorw->size); | |
2116 | if (iorw->iov == iorw->fast_iov) | |
2117 | *iovec = NULL; | |
2118 | return iorw->size; | |
2119 | } | |
2120 | ||
2b188cc1 | 2121 | #ifdef CONFIG_COMPAT |
cf6fd4bd | 2122 | if (req->ctx->compat) |
2b188cc1 JA |
2123 | return compat_import_iovec(rw, buf, sqe_len, UIO_FASTIOV, |
2124 | iovec, iter); | |
2125 | #endif | |
2126 | ||
2127 | return import_iovec(rw, buf, sqe_len, UIO_FASTIOV, iovec, iter); | |
2128 | } | |
2129 | ||
31b51510 | 2130 | /* |
32960613 JA |
2131 | * For files that don't have ->read_iter() and ->write_iter(), handle them |
2132 | * by looping over ->read() or ->write() manually. | |
31b51510 | 2133 | */ |
32960613 JA |
2134 | static ssize_t loop_rw_iter(int rw, struct file *file, struct kiocb *kiocb, |
2135 | struct iov_iter *iter) | |
2136 | { | |
2137 | ssize_t ret = 0; | |
2138 | ||
2139 | /* | |
2140 | * Don't support polled IO through this interface, and we can't | |
2141 | * support non-blocking either. For the latter, this just causes | |
2142 | * the kiocb to be handled from an async context. | |
2143 | */ | |
2144 | if (kiocb->ki_flags & IOCB_HIPRI) | |
2145 | return -EOPNOTSUPP; | |
2146 | if (kiocb->ki_flags & IOCB_NOWAIT) | |
2147 | return -EAGAIN; | |
2148 | ||
2149 | while (iov_iter_count(iter)) { | |
311ae9e1 | 2150 | struct iovec iovec; |
32960613 JA |
2151 | ssize_t nr; |
2152 | ||
311ae9e1 PB |
2153 | if (!iov_iter_is_bvec(iter)) { |
2154 | iovec = iov_iter_iovec(iter); | |
2155 | } else { | |
2156 | /* fixed buffers import bvec */ | |
2157 | iovec.iov_base = kmap(iter->bvec->bv_page) | |
2158 | + iter->iov_offset; | |
2159 | iovec.iov_len = min(iter->count, | |
2160 | iter->bvec->bv_len - iter->iov_offset); | |
2161 | } | |
2162 | ||
32960613 JA |
2163 | if (rw == READ) { |
2164 | nr = file->f_op->read(file, iovec.iov_base, | |
2165 | iovec.iov_len, &kiocb->ki_pos); | |
2166 | } else { | |
2167 | nr = file->f_op->write(file, iovec.iov_base, | |
2168 | iovec.iov_len, &kiocb->ki_pos); | |
2169 | } | |
2170 | ||
311ae9e1 PB |
2171 | if (iov_iter_is_bvec(iter)) |
2172 | kunmap(iter->bvec->bv_page); | |
2173 | ||
32960613 JA |
2174 | if (nr < 0) { |
2175 | if (!ret) | |
2176 | ret = nr; | |
2177 | break; | |
2178 | } | |
2179 | ret += nr; | |
2180 | if (nr != iovec.iov_len) | |
2181 | break; | |
2182 | iov_iter_advance(iter, nr); | |
2183 | } | |
2184 | ||
2185 | return ret; | |
2186 | } | |
2187 | ||
b7bb4f7d | 2188 | static void io_req_map_rw(struct io_kiocb *req, ssize_t io_size, |
f67676d1 JA |
2189 | struct iovec *iovec, struct iovec *fast_iov, |
2190 | struct iov_iter *iter) | |
2191 | { | |
2192 | req->io->rw.nr_segs = iter->nr_segs; | |
2193 | req->io->rw.size = io_size; | |
2194 | req->io->rw.iov = iovec; | |
2195 | if (!req->io->rw.iov) { | |
2196 | req->io->rw.iov = req->io->rw.fast_iov; | |
2197 | memcpy(req->io->rw.iov, fast_iov, | |
2198 | sizeof(struct iovec) * iter->nr_segs); | |
99bc4c38 PB |
2199 | } else { |
2200 | req->flags |= REQ_F_NEED_CLEANUP; | |
f67676d1 JA |
2201 | } |
2202 | } | |
2203 | ||
b7bb4f7d | 2204 | static int io_alloc_async_ctx(struct io_kiocb *req) |
f67676d1 | 2205 | { |
d3656344 JA |
2206 | if (!io_op_defs[req->opcode].async_ctx) |
2207 | return 0; | |
f67676d1 | 2208 | req->io = kmalloc(sizeof(*req->io), GFP_KERNEL); |
06b76d44 | 2209 | return req->io == NULL; |
b7bb4f7d JA |
2210 | } |
2211 | ||
b7bb4f7d JA |
2212 | static int io_setup_async_rw(struct io_kiocb *req, ssize_t io_size, |
2213 | struct iovec *iovec, struct iovec *fast_iov, | |
2214 | struct iov_iter *iter) | |
2215 | { | |
980ad263 | 2216 | if (!io_op_defs[req->opcode].async_ctx) |
74566df3 | 2217 | return 0; |
5d204bcf JA |
2218 | if (!req->io) { |
2219 | if (io_alloc_async_ctx(req)) | |
2220 | return -ENOMEM; | |
b7bb4f7d | 2221 | |
5d204bcf JA |
2222 | io_req_map_rw(req, io_size, iovec, fast_iov, iter); |
2223 | } | |
b7bb4f7d | 2224 | return 0; |
f67676d1 JA |
2225 | } |
2226 | ||
3529d8c2 JA |
2227 | static int io_read_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe, |
2228 | bool force_nonblock) | |
f67676d1 | 2229 | { |
3529d8c2 JA |
2230 | struct io_async_ctx *io; |
2231 | struct iov_iter iter; | |
f67676d1 JA |
2232 | ssize_t ret; |
2233 | ||
3529d8c2 JA |
2234 | ret = io_prep_rw(req, sqe, force_nonblock); |
2235 | if (ret) | |
2236 | return ret; | |
f67676d1 | 2237 | |
3529d8c2 JA |
2238 | if (unlikely(!(req->file->f_mode & FMODE_READ))) |
2239 | return -EBADF; | |
f67676d1 | 2240 | |
5f798bea PB |
2241 | /* either don't need iovec imported or already have it */ |
2242 | if (!req->io || req->flags & REQ_F_NEED_CLEANUP) | |
3529d8c2 JA |
2243 | return 0; |
2244 | ||
2245 | io = req->io; | |
2246 | io->rw.iov = io->rw.fast_iov; | |
2247 | req->io = NULL; | |
2248 | ret = io_import_iovec(READ, req, &io->rw.iov, &iter); | |
2249 | req->io = io; | |
2250 | if (ret < 0) | |
2251 | return ret; | |
2252 | ||
2253 | io_req_map_rw(req, ret, io->rw.iov, io->rw.fast_iov, &iter); | |
2254 | return 0; | |
f67676d1 JA |
2255 | } |
2256 | ||
014db007 | 2257 | static int io_read(struct io_kiocb *req, bool force_nonblock) |
2b188cc1 JA |
2258 | { |
2259 | struct iovec inline_vecs[UIO_FASTIOV], *iovec = inline_vecs; | |
9adbd45d | 2260 | struct kiocb *kiocb = &req->rw.kiocb; |
2b188cc1 | 2261 | struct iov_iter iter; |
31b51510 | 2262 | size_t iov_count; |
f67676d1 | 2263 | ssize_t io_size, ret; |
2b188cc1 | 2264 | |
3529d8c2 | 2265 | ret = io_import_iovec(READ, req, &iovec, &iter); |
06b76d44 JA |
2266 | if (ret < 0) |
2267 | return ret; | |
2b188cc1 | 2268 | |
fd6c2e4c JA |
2269 | /* Ensure we clear previously set non-block flag */ |
2270 | if (!force_nonblock) | |
29de5f6a | 2271 | kiocb->ki_flags &= ~IOCB_NOWAIT; |
fd6c2e4c | 2272 | |
797f3f53 | 2273 | req->result = 0; |
f67676d1 | 2274 | io_size = ret; |
9e645e11 | 2275 | if (req->flags & REQ_F_LINK) |
f67676d1 JA |
2276 | req->result = io_size; |
2277 | ||
2278 | /* | |
2279 | * If the file doesn't support async, mark it as REQ_F_MUST_PUNT so | |
2280 | * we know to async punt it even if it was opened O_NONBLOCK | |
2281 | */ | |
29de5f6a | 2282 | if (force_nonblock && !io_file_supports_async(req->file)) |
f67676d1 | 2283 | goto copy_iov; |
9e645e11 | 2284 | |
31b51510 | 2285 | iov_count = iov_iter_count(&iter); |
9adbd45d | 2286 | ret = rw_verify_area(READ, req->file, &kiocb->ki_pos, iov_count); |
2b188cc1 JA |
2287 | if (!ret) { |
2288 | ssize_t ret2; | |
2289 | ||
9adbd45d JA |
2290 | if (req->file->f_op->read_iter) |
2291 | ret2 = call_read_iter(req->file, kiocb, &iter); | |
32960613 | 2292 | else |
9adbd45d | 2293 | ret2 = loop_rw_iter(READ, req->file, kiocb, &iter); |
32960613 | 2294 | |
9d93a3f5 | 2295 | /* Catch -EAGAIN return for forced non-blocking submission */ |
f67676d1 | 2296 | if (!force_nonblock || ret2 != -EAGAIN) { |
014db007 | 2297 | kiocb_done(kiocb, ret2); |
f67676d1 JA |
2298 | } else { |
2299 | copy_iov: | |
b7bb4f7d | 2300 | ret = io_setup_async_rw(req, io_size, iovec, |
f67676d1 JA |
2301 | inline_vecs, &iter); |
2302 | if (ret) | |
2303 | goto out_free; | |
29de5f6a JA |
2304 | /* any defer here is final, must blocking retry */ |
2305 | if (!(req->flags & REQ_F_NOWAIT)) | |
2306 | req->flags |= REQ_F_MUST_PUNT; | |
f67676d1 JA |
2307 | return -EAGAIN; |
2308 | } | |
2b188cc1 | 2309 | } |
f67676d1 | 2310 | out_free: |
1e95081c | 2311 | kfree(iovec); |
99bc4c38 | 2312 | req->flags &= ~REQ_F_NEED_CLEANUP; |
2b188cc1 JA |
2313 | return ret; |
2314 | } | |
2315 | ||
3529d8c2 JA |
2316 | static int io_write_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe, |
2317 | bool force_nonblock) | |
f67676d1 | 2318 | { |
3529d8c2 JA |
2319 | struct io_async_ctx *io; |
2320 | struct iov_iter iter; | |
f67676d1 JA |
2321 | ssize_t ret; |
2322 | ||
3529d8c2 JA |
2323 | ret = io_prep_rw(req, sqe, force_nonblock); |
2324 | if (ret) | |
2325 | return ret; | |
f67676d1 | 2326 | |
3529d8c2 JA |
2327 | if (unlikely(!(req->file->f_mode & FMODE_WRITE))) |
2328 | return -EBADF; | |
f67676d1 | 2329 | |
5f798bea PB |
2330 | /* either don't need iovec imported or already have it */ |
2331 | if (!req->io || req->flags & REQ_F_NEED_CLEANUP) | |
3529d8c2 JA |
2332 | return 0; |
2333 | ||
2334 | io = req->io; | |
2335 | io->rw.iov = io->rw.fast_iov; | |
2336 | req->io = NULL; | |
2337 | ret = io_import_iovec(WRITE, req, &io->rw.iov, &iter); | |
2338 | req->io = io; | |
2339 | if (ret < 0) | |
2340 | return ret; | |
2341 | ||
2342 | io_req_map_rw(req, ret, io->rw.iov, io->rw.fast_iov, &iter); | |
2343 | return 0; | |
f67676d1 JA |
2344 | } |
2345 | ||
014db007 | 2346 | static int io_write(struct io_kiocb *req, bool force_nonblock) |
2b188cc1 JA |
2347 | { |
2348 | struct iovec inline_vecs[UIO_FASTIOV], *iovec = inline_vecs; | |
9adbd45d | 2349 | struct kiocb *kiocb = &req->rw.kiocb; |
2b188cc1 | 2350 | struct iov_iter iter; |
31b51510 | 2351 | size_t iov_count; |
f67676d1 | 2352 | ssize_t ret, io_size; |
2b188cc1 | 2353 | |
3529d8c2 | 2354 | ret = io_import_iovec(WRITE, req, &iovec, &iter); |
06b76d44 JA |
2355 | if (ret < 0) |
2356 | return ret; | |
2b188cc1 | 2357 | |
fd6c2e4c JA |
2358 | /* Ensure we clear previously set non-block flag */ |
2359 | if (!force_nonblock) | |
9adbd45d | 2360 | req->rw.kiocb.ki_flags &= ~IOCB_NOWAIT; |
fd6c2e4c | 2361 | |
797f3f53 | 2362 | req->result = 0; |
f67676d1 | 2363 | io_size = ret; |
9e645e11 | 2364 | if (req->flags & REQ_F_LINK) |
f67676d1 | 2365 | req->result = io_size; |
9e645e11 | 2366 | |
f67676d1 JA |
2367 | /* |
2368 | * If the file doesn't support async, mark it as REQ_F_MUST_PUNT so | |
2369 | * we know to async punt it even if it was opened O_NONBLOCK | |
2370 | */ | |
29de5f6a | 2371 | if (force_nonblock && !io_file_supports_async(req->file)) |
f67676d1 | 2372 | goto copy_iov; |
31b51510 | 2373 | |
10d59345 JA |
2374 | /* file path doesn't support NOWAIT for non-direct_IO */ |
2375 | if (force_nonblock && !(kiocb->ki_flags & IOCB_DIRECT) && | |
2376 | (req->flags & REQ_F_ISREG)) | |
f67676d1 | 2377 | goto copy_iov; |
31b51510 | 2378 | |
f67676d1 | 2379 | iov_count = iov_iter_count(&iter); |
9adbd45d | 2380 | ret = rw_verify_area(WRITE, req->file, &kiocb->ki_pos, iov_count); |
2b188cc1 | 2381 | if (!ret) { |
9bf7933f RP |
2382 | ssize_t ret2; |
2383 | ||
2b188cc1 JA |
2384 | /* |
2385 | * Open-code file_start_write here to grab freeze protection, | |
2386 | * which will be released by another thread in | |
2387 | * io_complete_rw(). Fool lockdep by telling it the lock got | |
2388 | * released so that it doesn't complain about the held lock when | |
2389 | * we return to userspace. | |
2390 | */ | |
491381ce | 2391 | if (req->flags & REQ_F_ISREG) { |
9adbd45d | 2392 | __sb_start_write(file_inode(req->file)->i_sb, |
2b188cc1 | 2393 | SB_FREEZE_WRITE, true); |
9adbd45d | 2394 | __sb_writers_release(file_inode(req->file)->i_sb, |
2b188cc1 JA |
2395 | SB_FREEZE_WRITE); |
2396 | } | |
2397 | kiocb->ki_flags |= IOCB_WRITE; | |
9bf7933f | 2398 | |
9adbd45d JA |
2399 | if (req->file->f_op->write_iter) |
2400 | ret2 = call_write_iter(req->file, kiocb, &iter); | |
32960613 | 2401 | else |
9adbd45d | 2402 | ret2 = loop_rw_iter(WRITE, req->file, kiocb, &iter); |
faac996c JA |
2403 | /* |
2404 | * Raw bdev writes will -EOPNOTSUPP for IOCB_NOWAIT. Just | |
2405 | * retry them without IOCB_NOWAIT. | |
2406 | */ | |
2407 | if (ret2 == -EOPNOTSUPP && (kiocb->ki_flags & IOCB_NOWAIT)) | |
2408 | ret2 = -EAGAIN; | |
f67676d1 | 2409 | if (!force_nonblock || ret2 != -EAGAIN) { |
014db007 | 2410 | kiocb_done(kiocb, ret2); |
f67676d1 JA |
2411 | } else { |
2412 | copy_iov: | |
b7bb4f7d | 2413 | ret = io_setup_async_rw(req, io_size, iovec, |
f67676d1 JA |
2414 | inline_vecs, &iter); |
2415 | if (ret) | |
2416 | goto out_free; | |
29de5f6a JA |
2417 | /* any defer here is final, must blocking retry */ |
2418 | req->flags |= REQ_F_MUST_PUNT; | |
f67676d1 JA |
2419 | return -EAGAIN; |
2420 | } | |
2b188cc1 | 2421 | } |
31b51510 | 2422 | out_free: |
99bc4c38 | 2423 | req->flags &= ~REQ_F_NEED_CLEANUP; |
1e95081c | 2424 | kfree(iovec); |
2b188cc1 JA |
2425 | return ret; |
2426 | } | |
2427 | ||
7d67af2c PB |
2428 | static int io_splice_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
2429 | { | |
2430 | struct io_splice* sp = &req->splice; | |
2431 | unsigned int valid_flags = SPLICE_F_FD_IN_FIXED | SPLICE_F_ALL; | |
2432 | int ret; | |
2433 | ||
2434 | if (req->flags & REQ_F_NEED_CLEANUP) | |
2435 | return 0; | |
2436 | ||
2437 | sp->file_in = NULL; | |
2438 | sp->off_in = READ_ONCE(sqe->splice_off_in); | |
2439 | sp->off_out = READ_ONCE(sqe->off); | |
2440 | sp->len = READ_ONCE(sqe->len); | |
2441 | sp->flags = READ_ONCE(sqe->splice_flags); | |
2442 | ||
2443 | if (unlikely(sp->flags & ~valid_flags)) | |
2444 | return -EINVAL; | |
2445 | ||
2446 | ret = io_file_get(NULL, req, READ_ONCE(sqe->splice_fd_in), &sp->file_in, | |
2447 | (sp->flags & SPLICE_F_FD_IN_FIXED)); | |
2448 | if (ret) | |
2449 | return ret; | |
2450 | req->flags |= REQ_F_NEED_CLEANUP; | |
2451 | ||
2452 | if (!S_ISREG(file_inode(sp->file_in)->i_mode)) | |
2453 | req->work.flags |= IO_WQ_WORK_UNBOUND; | |
2454 | ||
2455 | return 0; | |
2456 | } | |
2457 | ||
2458 | static bool io_splice_punt(struct file *file) | |
2459 | { | |
2460 | if (get_pipe_info(file)) | |
2461 | return false; | |
2462 | if (!io_file_supports_async(file)) | |
2463 | return true; | |
2464 | return !(file->f_mode & O_NONBLOCK); | |
2465 | } | |
2466 | ||
014db007 | 2467 | static int io_splice(struct io_kiocb *req, bool force_nonblock) |
7d67af2c PB |
2468 | { |
2469 | struct io_splice *sp = &req->splice; | |
2470 | struct file *in = sp->file_in; | |
2471 | struct file *out = sp->file_out; | |
2472 | unsigned int flags = sp->flags & ~SPLICE_F_FD_IN_FIXED; | |
2473 | loff_t *poff_in, *poff_out; | |
2474 | long ret; | |
2475 | ||
2476 | if (force_nonblock) { | |
2477 | if (io_splice_punt(in) || io_splice_punt(out)) | |
2478 | return -EAGAIN; | |
2479 | flags |= SPLICE_F_NONBLOCK; | |
2480 | } | |
2481 | ||
2482 | poff_in = (sp->off_in == -1) ? NULL : &sp->off_in; | |
2483 | poff_out = (sp->off_out == -1) ? NULL : &sp->off_out; | |
2484 | ret = do_splice(in, poff_in, out, poff_out, sp->len, flags); | |
2485 | if (force_nonblock && ret == -EAGAIN) | |
2486 | return -EAGAIN; | |
2487 | ||
2488 | io_put_file(req, in, (sp->flags & SPLICE_F_FD_IN_FIXED)); | |
2489 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
2490 | ||
2491 | io_cqring_add_event(req, ret); | |
2492 | if (ret != sp->len) | |
2493 | req_set_fail_links(req); | |
014db007 | 2494 | io_put_req(req); |
7d67af2c PB |
2495 | return 0; |
2496 | } | |
2497 | ||
2b188cc1 JA |
2498 | /* |
2499 | * IORING_OP_NOP just posts a completion event, nothing else. | |
2500 | */ | |
78e19bbe | 2501 | static int io_nop(struct io_kiocb *req) |
2b188cc1 JA |
2502 | { |
2503 | struct io_ring_ctx *ctx = req->ctx; | |
2b188cc1 | 2504 | |
def596e9 JA |
2505 | if (unlikely(ctx->flags & IORING_SETUP_IOPOLL)) |
2506 | return -EINVAL; | |
2507 | ||
78e19bbe | 2508 | io_cqring_add_event(req, 0); |
e65ef56d | 2509 | io_put_req(req); |
2b188cc1 JA |
2510 | return 0; |
2511 | } | |
2512 | ||
3529d8c2 | 2513 | static int io_prep_fsync(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
c992fe29 | 2514 | { |
6b06314c | 2515 | struct io_ring_ctx *ctx = req->ctx; |
c992fe29 | 2516 | |
09bb8394 JA |
2517 | if (!req->file) |
2518 | return -EBADF; | |
c992fe29 | 2519 | |
6b06314c | 2520 | if (unlikely(ctx->flags & IORING_SETUP_IOPOLL)) |
def596e9 | 2521 | return -EINVAL; |
edafccee | 2522 | if (unlikely(sqe->addr || sqe->ioprio || sqe->buf_index)) |
c992fe29 CH |
2523 | return -EINVAL; |
2524 | ||
8ed8d3c3 JA |
2525 | req->sync.flags = READ_ONCE(sqe->fsync_flags); |
2526 | if (unlikely(req->sync.flags & ~IORING_FSYNC_DATASYNC)) | |
2527 | return -EINVAL; | |
2528 | ||
2529 | req->sync.off = READ_ONCE(sqe->off); | |
2530 | req->sync.len = READ_ONCE(sqe->len); | |
c992fe29 CH |
2531 | return 0; |
2532 | } | |
2533 | ||
8ed8d3c3 JA |
2534 | static bool io_req_cancelled(struct io_kiocb *req) |
2535 | { | |
2536 | if (req->work.flags & IO_WQ_WORK_CANCEL) { | |
2537 | req_set_fail_links(req); | |
2538 | io_cqring_add_event(req, -ECANCELED); | |
594506fe | 2539 | io_double_put_req(req); |
8ed8d3c3 JA |
2540 | return true; |
2541 | } | |
2542 | ||
2543 | return false; | |
2544 | } | |
2545 | ||
78912934 JA |
2546 | static void io_link_work_cb(struct io_wq_work **workptr) |
2547 | { | |
2548 | struct io_wq_work *work = *workptr; | |
2549 | struct io_kiocb *link = work->data; | |
2550 | ||
2551 | io_queue_linked_timeout(link); | |
5eae8619 | 2552 | io_wq_submit_work(workptr); |
78912934 JA |
2553 | } |
2554 | ||
2555 | static void io_wq_assign_next(struct io_wq_work **workptr, struct io_kiocb *nxt) | |
2556 | { | |
2557 | struct io_kiocb *link; | |
2558 | ||
78912934 | 2559 | *workptr = &nxt->work; |
3b17cf5a | 2560 | link = io_prep_linked_timeout(nxt); |
78912934 | 2561 | if (link) { |
78912934 JA |
2562 | nxt->work.func = io_link_work_cb; |
2563 | nxt->work.data = link; | |
2564 | } | |
2565 | } | |
2566 | ||
014db007 | 2567 | static void __io_fsync(struct io_kiocb *req) |
8ed8d3c3 | 2568 | { |
8ed8d3c3 | 2569 | loff_t end = req->sync.off + req->sync.len; |
8ed8d3c3 JA |
2570 | int ret; |
2571 | ||
9adbd45d | 2572 | ret = vfs_fsync_range(req->file, req->sync.off, |
8ed8d3c3 JA |
2573 | end > 0 ? end : LLONG_MAX, |
2574 | req->sync.flags & IORING_FSYNC_DATASYNC); | |
2575 | if (ret < 0) | |
2576 | req_set_fail_links(req); | |
2577 | io_cqring_add_event(req, ret); | |
014db007 | 2578 | io_put_req(req); |
5ea62161 PB |
2579 | } |
2580 | ||
2581 | static void io_fsync_finish(struct io_wq_work **workptr) | |
2582 | { | |
2583 | struct io_kiocb *req = container_of(*workptr, struct io_kiocb, work); | |
2584 | struct io_kiocb *nxt = NULL; | |
2585 | ||
2586 | if (io_req_cancelled(req)) | |
2587 | return; | |
014db007 | 2588 | __io_fsync(req); |
594506fe | 2589 | io_put_req(req); /* drop submission reference */ |
8ed8d3c3 | 2590 | if (nxt) |
78912934 | 2591 | io_wq_assign_next(workptr, nxt); |
8ed8d3c3 JA |
2592 | } |
2593 | ||
014db007 | 2594 | static int io_fsync(struct io_kiocb *req, bool force_nonblock) |
c992fe29 | 2595 | { |
c992fe29 | 2596 | /* fsync always requires a blocking context */ |
8ed8d3c3 | 2597 | if (force_nonblock) { |
8ed8d3c3 | 2598 | req->work.func = io_fsync_finish; |
c992fe29 | 2599 | return -EAGAIN; |
8ed8d3c3 | 2600 | } |
014db007 | 2601 | __io_fsync(req); |
c992fe29 CH |
2602 | return 0; |
2603 | } | |
2604 | ||
014db007 | 2605 | static void __io_fallocate(struct io_kiocb *req) |
8ed8d3c3 | 2606 | { |
8ed8d3c3 JA |
2607 | int ret; |
2608 | ||
d63d1b5e JA |
2609 | ret = vfs_fallocate(req->file, req->sync.mode, req->sync.off, |
2610 | req->sync.len); | |
8ed8d3c3 JA |
2611 | if (ret < 0) |
2612 | req_set_fail_links(req); | |
2613 | io_cqring_add_event(req, ret); | |
014db007 | 2614 | io_put_req(req); |
5ea62161 PB |
2615 | } |
2616 | ||
2617 | static void io_fallocate_finish(struct io_wq_work **workptr) | |
2618 | { | |
2619 | struct io_kiocb *req = container_of(*workptr, struct io_kiocb, work); | |
2620 | struct io_kiocb *nxt = NULL; | |
2621 | ||
594506fe PB |
2622 | if (io_req_cancelled(req)) |
2623 | return; | |
014db007 | 2624 | __io_fallocate(req); |
594506fe | 2625 | io_put_req(req); /* drop submission reference */ |
8ed8d3c3 | 2626 | if (nxt) |
78912934 | 2627 | io_wq_assign_next(workptr, nxt); |
5d17b4a4 JA |
2628 | } |
2629 | ||
d63d1b5e JA |
2630 | static int io_fallocate_prep(struct io_kiocb *req, |
2631 | const struct io_uring_sqe *sqe) | |
2632 | { | |
2633 | if (sqe->ioprio || sqe->buf_index || sqe->rw_flags) | |
2634 | return -EINVAL; | |
2635 | ||
2636 | req->sync.off = READ_ONCE(sqe->off); | |
2637 | req->sync.len = READ_ONCE(sqe->addr); | |
2638 | req->sync.mode = READ_ONCE(sqe->len); | |
2639 | return 0; | |
2640 | } | |
2641 | ||
014db007 | 2642 | static int io_fallocate(struct io_kiocb *req, bool force_nonblock) |
5d17b4a4 | 2643 | { |
d63d1b5e | 2644 | /* fallocate always requiring blocking context */ |
8ed8d3c3 | 2645 | if (force_nonblock) { |
d63d1b5e | 2646 | req->work.func = io_fallocate_finish; |
5d17b4a4 | 2647 | return -EAGAIN; |
8ed8d3c3 | 2648 | } |
5d17b4a4 | 2649 | |
014db007 | 2650 | __io_fallocate(req); |
5d17b4a4 JA |
2651 | return 0; |
2652 | } | |
2653 | ||
15b71abe | 2654 | static int io_openat_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
b7bb4f7d | 2655 | { |
f8748881 | 2656 | const char __user *fname; |
15b71abe | 2657 | int ret; |
b7bb4f7d | 2658 | |
15b71abe JA |
2659 | if (sqe->ioprio || sqe->buf_index) |
2660 | return -EINVAL; | |
cf3040ca JA |
2661 | if (sqe->flags & IOSQE_FIXED_FILE) |
2662 | return -EBADF; | |
0bdbdd08 PB |
2663 | if (req->flags & REQ_F_NEED_CLEANUP) |
2664 | return 0; | |
03b1230c | 2665 | |
15b71abe | 2666 | req->open.dfd = READ_ONCE(sqe->fd); |
c12cedf2 | 2667 | req->open.how.mode = READ_ONCE(sqe->len); |
f8748881 | 2668 | fname = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
c12cedf2 | 2669 | req->open.how.flags = READ_ONCE(sqe->open_flags); |
3529d8c2 | 2670 | |
f8748881 | 2671 | req->open.filename = getname(fname); |
15b71abe JA |
2672 | if (IS_ERR(req->open.filename)) { |
2673 | ret = PTR_ERR(req->open.filename); | |
2674 | req->open.filename = NULL; | |
2675 | return ret; | |
2676 | } | |
3529d8c2 | 2677 | |
8fef80bf | 2678 | req->flags |= REQ_F_NEED_CLEANUP; |
15b71abe | 2679 | return 0; |
03b1230c JA |
2680 | } |
2681 | ||
cebdb986 | 2682 | static int io_openat2_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
aa1fa28f | 2683 | { |
cebdb986 JA |
2684 | struct open_how __user *how; |
2685 | const char __user *fname; | |
2686 | size_t len; | |
0fa03c62 JA |
2687 | int ret; |
2688 | ||
cebdb986 | 2689 | if (sqe->ioprio || sqe->buf_index) |
0fa03c62 | 2690 | return -EINVAL; |
cf3040ca JA |
2691 | if (sqe->flags & IOSQE_FIXED_FILE) |
2692 | return -EBADF; | |
0bdbdd08 PB |
2693 | if (req->flags & REQ_F_NEED_CLEANUP) |
2694 | return 0; | |
0fa03c62 | 2695 | |
cebdb986 JA |
2696 | req->open.dfd = READ_ONCE(sqe->fd); |
2697 | fname = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
2698 | how = u64_to_user_ptr(READ_ONCE(sqe->addr2)); | |
2699 | len = READ_ONCE(sqe->len); | |
0fa03c62 | 2700 | |
cebdb986 JA |
2701 | if (len < OPEN_HOW_SIZE_VER0) |
2702 | return -EINVAL; | |
3529d8c2 | 2703 | |
cebdb986 JA |
2704 | ret = copy_struct_from_user(&req->open.how, sizeof(req->open.how), how, |
2705 | len); | |
2706 | if (ret) | |
2707 | return ret; | |
3529d8c2 | 2708 | |
cebdb986 JA |
2709 | if (!(req->open.how.flags & O_PATH) && force_o_largefile()) |
2710 | req->open.how.flags |= O_LARGEFILE; | |
0fa03c62 | 2711 | |
cebdb986 JA |
2712 | req->open.filename = getname(fname); |
2713 | if (IS_ERR(req->open.filename)) { | |
2714 | ret = PTR_ERR(req->open.filename); | |
2715 | req->open.filename = NULL; | |
2716 | return ret; | |
2717 | } | |
2718 | ||
8fef80bf | 2719 | req->flags |= REQ_F_NEED_CLEANUP; |
cebdb986 JA |
2720 | return 0; |
2721 | } | |
2722 | ||
014db007 | 2723 | static int io_openat2(struct io_kiocb *req, bool force_nonblock) |
15b71abe JA |
2724 | { |
2725 | struct open_flags op; | |
15b71abe JA |
2726 | struct file *file; |
2727 | int ret; | |
2728 | ||
f86cd20c | 2729 | if (force_nonblock) |
15b71abe | 2730 | return -EAGAIN; |
15b71abe | 2731 | |
cebdb986 | 2732 | ret = build_open_flags(&req->open.how, &op); |
15b71abe JA |
2733 | if (ret) |
2734 | goto err; | |
2735 | ||
cebdb986 | 2736 | ret = get_unused_fd_flags(req->open.how.flags); |
15b71abe JA |
2737 | if (ret < 0) |
2738 | goto err; | |
2739 | ||
2740 | file = do_filp_open(req->open.dfd, req->open.filename, &op); | |
2741 | if (IS_ERR(file)) { | |
2742 | put_unused_fd(ret); | |
2743 | ret = PTR_ERR(file); | |
2744 | } else { | |
2745 | fsnotify_open(file); | |
2746 | fd_install(ret, file); | |
2747 | } | |
2748 | err: | |
2749 | putname(req->open.filename); | |
8fef80bf | 2750 | req->flags &= ~REQ_F_NEED_CLEANUP; |
15b71abe JA |
2751 | if (ret < 0) |
2752 | req_set_fail_links(req); | |
2753 | io_cqring_add_event(req, ret); | |
014db007 | 2754 | io_put_req(req); |
15b71abe JA |
2755 | return 0; |
2756 | } | |
2757 | ||
014db007 | 2758 | static int io_openat(struct io_kiocb *req, bool force_nonblock) |
cebdb986 JA |
2759 | { |
2760 | req->open.how = build_open_how(req->open.how.flags, req->open.how.mode); | |
014db007 | 2761 | return io_openat2(req, force_nonblock); |
cebdb986 JA |
2762 | } |
2763 | ||
3e4827b0 JA |
2764 | static int io_epoll_ctl_prep(struct io_kiocb *req, |
2765 | const struct io_uring_sqe *sqe) | |
2766 | { | |
2767 | #if defined(CONFIG_EPOLL) | |
2768 | if (sqe->ioprio || sqe->buf_index) | |
2769 | return -EINVAL; | |
2770 | ||
2771 | req->epoll.epfd = READ_ONCE(sqe->fd); | |
2772 | req->epoll.op = READ_ONCE(sqe->len); | |
2773 | req->epoll.fd = READ_ONCE(sqe->off); | |
2774 | ||
2775 | if (ep_op_has_event(req->epoll.op)) { | |
2776 | struct epoll_event __user *ev; | |
2777 | ||
2778 | ev = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
2779 | if (copy_from_user(&req->epoll.event, ev, sizeof(*ev))) | |
2780 | return -EFAULT; | |
2781 | } | |
2782 | ||
2783 | return 0; | |
2784 | #else | |
2785 | return -EOPNOTSUPP; | |
2786 | #endif | |
2787 | } | |
2788 | ||
014db007 | 2789 | static int io_epoll_ctl(struct io_kiocb *req, bool force_nonblock) |
3e4827b0 JA |
2790 | { |
2791 | #if defined(CONFIG_EPOLL) | |
2792 | struct io_epoll *ie = &req->epoll; | |
2793 | int ret; | |
2794 | ||
2795 | ret = do_epoll_ctl(ie->epfd, ie->op, ie->fd, &ie->event, force_nonblock); | |
2796 | if (force_nonblock && ret == -EAGAIN) | |
2797 | return -EAGAIN; | |
2798 | ||
2799 | if (ret < 0) | |
2800 | req_set_fail_links(req); | |
2801 | io_cqring_add_event(req, ret); | |
014db007 | 2802 | io_put_req(req); |
3e4827b0 JA |
2803 | return 0; |
2804 | #else | |
2805 | return -EOPNOTSUPP; | |
2806 | #endif | |
2807 | } | |
2808 | ||
c1ca757b JA |
2809 | static int io_madvise_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
2810 | { | |
2811 | #if defined(CONFIG_ADVISE_SYSCALLS) && defined(CONFIG_MMU) | |
2812 | if (sqe->ioprio || sqe->buf_index || sqe->off) | |
2813 | return -EINVAL; | |
2814 | ||
2815 | req->madvise.addr = READ_ONCE(sqe->addr); | |
2816 | req->madvise.len = READ_ONCE(sqe->len); | |
2817 | req->madvise.advice = READ_ONCE(sqe->fadvise_advice); | |
2818 | return 0; | |
2819 | #else | |
2820 | return -EOPNOTSUPP; | |
2821 | #endif | |
2822 | } | |
2823 | ||
014db007 | 2824 | static int io_madvise(struct io_kiocb *req, bool force_nonblock) |
c1ca757b JA |
2825 | { |
2826 | #if defined(CONFIG_ADVISE_SYSCALLS) && defined(CONFIG_MMU) | |
2827 | struct io_madvise *ma = &req->madvise; | |
2828 | int ret; | |
2829 | ||
2830 | if (force_nonblock) | |
2831 | return -EAGAIN; | |
2832 | ||
2833 | ret = do_madvise(ma->addr, ma->len, ma->advice); | |
2834 | if (ret < 0) | |
2835 | req_set_fail_links(req); | |
2836 | io_cqring_add_event(req, ret); | |
014db007 | 2837 | io_put_req(req); |
c1ca757b JA |
2838 | return 0; |
2839 | #else | |
2840 | return -EOPNOTSUPP; | |
2841 | #endif | |
2842 | } | |
2843 | ||
4840e418 JA |
2844 | static int io_fadvise_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
2845 | { | |
2846 | if (sqe->ioprio || sqe->buf_index || sqe->addr) | |
2847 | return -EINVAL; | |
2848 | ||
2849 | req->fadvise.offset = READ_ONCE(sqe->off); | |
2850 | req->fadvise.len = READ_ONCE(sqe->len); | |
2851 | req->fadvise.advice = READ_ONCE(sqe->fadvise_advice); | |
2852 | return 0; | |
2853 | } | |
2854 | ||
014db007 | 2855 | static int io_fadvise(struct io_kiocb *req, bool force_nonblock) |
4840e418 JA |
2856 | { |
2857 | struct io_fadvise *fa = &req->fadvise; | |
2858 | int ret; | |
2859 | ||
3e69426d JA |
2860 | if (force_nonblock) { |
2861 | switch (fa->advice) { | |
2862 | case POSIX_FADV_NORMAL: | |
2863 | case POSIX_FADV_RANDOM: | |
2864 | case POSIX_FADV_SEQUENTIAL: | |
2865 | break; | |
2866 | default: | |
2867 | return -EAGAIN; | |
2868 | } | |
2869 | } | |
4840e418 JA |
2870 | |
2871 | ret = vfs_fadvise(req->file, fa->offset, fa->len, fa->advice); | |
2872 | if (ret < 0) | |
2873 | req_set_fail_links(req); | |
2874 | io_cqring_add_event(req, ret); | |
014db007 | 2875 | io_put_req(req); |
4840e418 JA |
2876 | return 0; |
2877 | } | |
2878 | ||
eddc7ef5 JA |
2879 | static int io_statx_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
2880 | { | |
f8748881 | 2881 | const char __user *fname; |
eddc7ef5 JA |
2882 | unsigned lookup_flags; |
2883 | int ret; | |
2884 | ||
2885 | if (sqe->ioprio || sqe->buf_index) | |
2886 | return -EINVAL; | |
cf3040ca JA |
2887 | if (sqe->flags & IOSQE_FIXED_FILE) |
2888 | return -EBADF; | |
0bdbdd08 PB |
2889 | if (req->flags & REQ_F_NEED_CLEANUP) |
2890 | return 0; | |
eddc7ef5 JA |
2891 | |
2892 | req->open.dfd = READ_ONCE(sqe->fd); | |
2893 | req->open.mask = READ_ONCE(sqe->len); | |
f8748881 | 2894 | fname = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
eddc7ef5 | 2895 | req->open.buffer = u64_to_user_ptr(READ_ONCE(sqe->addr2)); |
c12cedf2 | 2896 | req->open.how.flags = READ_ONCE(sqe->statx_flags); |
eddc7ef5 | 2897 | |
c12cedf2 | 2898 | if (vfs_stat_set_lookup_flags(&lookup_flags, req->open.how.flags)) |
eddc7ef5 JA |
2899 | return -EINVAL; |
2900 | ||
f8748881 | 2901 | req->open.filename = getname_flags(fname, lookup_flags, NULL); |
eddc7ef5 JA |
2902 | if (IS_ERR(req->open.filename)) { |
2903 | ret = PTR_ERR(req->open.filename); | |
2904 | req->open.filename = NULL; | |
2905 | return ret; | |
2906 | } | |
2907 | ||
8fef80bf | 2908 | req->flags |= REQ_F_NEED_CLEANUP; |
eddc7ef5 JA |
2909 | return 0; |
2910 | } | |
2911 | ||
014db007 | 2912 | static int io_statx(struct io_kiocb *req, bool force_nonblock) |
eddc7ef5 JA |
2913 | { |
2914 | struct io_open *ctx = &req->open; | |
2915 | unsigned lookup_flags; | |
2916 | struct path path; | |
2917 | struct kstat stat; | |
2918 | int ret; | |
2919 | ||
2920 | if (force_nonblock) | |
2921 | return -EAGAIN; | |
2922 | ||
c12cedf2 | 2923 | if (vfs_stat_set_lookup_flags(&lookup_flags, ctx->how.flags)) |
eddc7ef5 JA |
2924 | return -EINVAL; |
2925 | ||
2926 | retry: | |
2927 | /* filename_lookup() drops it, keep a reference */ | |
2928 | ctx->filename->refcnt++; | |
2929 | ||
2930 | ret = filename_lookup(ctx->dfd, ctx->filename, lookup_flags, &path, | |
2931 | NULL); | |
2932 | if (ret) | |
2933 | goto err; | |
2934 | ||
c12cedf2 | 2935 | ret = vfs_getattr(&path, &stat, ctx->mask, ctx->how.flags); |
eddc7ef5 JA |
2936 | path_put(&path); |
2937 | if (retry_estale(ret, lookup_flags)) { | |
2938 | lookup_flags |= LOOKUP_REVAL; | |
2939 | goto retry; | |
2940 | } | |
2941 | if (!ret) | |
2942 | ret = cp_statx(&stat, ctx->buffer); | |
2943 | err: | |
2944 | putname(ctx->filename); | |
8fef80bf | 2945 | req->flags &= ~REQ_F_NEED_CLEANUP; |
eddc7ef5 JA |
2946 | if (ret < 0) |
2947 | req_set_fail_links(req); | |
2948 | io_cqring_add_event(req, ret); | |
014db007 | 2949 | io_put_req(req); |
eddc7ef5 JA |
2950 | return 0; |
2951 | } | |
2952 | ||
b5dba59e JA |
2953 | static int io_close_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
2954 | { | |
2955 | /* | |
2956 | * If we queue this for async, it must not be cancellable. That would | |
2957 | * leave the 'file' in an undeterminate state. | |
2958 | */ | |
2959 | req->work.flags |= IO_WQ_WORK_NO_CANCEL; | |
2960 | ||
2961 | if (sqe->ioprio || sqe->off || sqe->addr || sqe->len || | |
2962 | sqe->rw_flags || sqe->buf_index) | |
2963 | return -EINVAL; | |
2964 | if (sqe->flags & IOSQE_FIXED_FILE) | |
cf3040ca | 2965 | return -EBADF; |
b5dba59e JA |
2966 | |
2967 | req->close.fd = READ_ONCE(sqe->fd); | |
2968 | if (req->file->f_op == &io_uring_fops || | |
b14cca0c | 2969 | req->close.fd == req->ctx->ring_fd) |
b5dba59e JA |
2970 | return -EBADF; |
2971 | ||
2972 | return 0; | |
2973 | } | |
2974 | ||
a93b3331 | 2975 | /* only called when __close_fd_get_file() is done */ |
014db007 | 2976 | static void __io_close_finish(struct io_kiocb *req) |
a93b3331 PB |
2977 | { |
2978 | int ret; | |
2979 | ||
2980 | ret = filp_close(req->close.put_file, req->work.files); | |
2981 | if (ret < 0) | |
2982 | req_set_fail_links(req); | |
2983 | io_cqring_add_event(req, ret); | |
2984 | fput(req->close.put_file); | |
014db007 | 2985 | io_put_req(req); |
a93b3331 PB |
2986 | } |
2987 | ||
b5dba59e JA |
2988 | static void io_close_finish(struct io_wq_work **workptr) |
2989 | { | |
2990 | struct io_kiocb *req = container_of(*workptr, struct io_kiocb, work); | |
2991 | struct io_kiocb *nxt = NULL; | |
2992 | ||
7fbeb95d | 2993 | /* not cancellable, don't do io_req_cancelled() */ |
014db007 | 2994 | __io_close_finish(req); |
594506fe | 2995 | io_put_req(req); /* drop submission reference */ |
b5dba59e JA |
2996 | if (nxt) |
2997 | io_wq_assign_next(workptr, nxt); | |
2998 | } | |
2999 | ||
014db007 | 3000 | static int io_close(struct io_kiocb *req, bool force_nonblock) |
b5dba59e JA |
3001 | { |
3002 | int ret; | |
3003 | ||
3004 | req->close.put_file = NULL; | |
3005 | ret = __close_fd_get_file(req->close.fd, &req->close.put_file); | |
3006 | if (ret < 0) | |
3007 | return ret; | |
3008 | ||
3009 | /* if the file has a flush method, be safe and punt to async */ | |
a2100672 | 3010 | if (req->close.put_file->f_op->flush && force_nonblock) { |
594506fe PB |
3011 | /* submission ref will be dropped, take it for async */ |
3012 | refcount_inc(&req->refs); | |
3013 | ||
a2100672 PB |
3014 | req->work.func = io_close_finish; |
3015 | /* | |
3016 | * Do manual async queue here to avoid grabbing files - we don't | |
3017 | * need the files, and it'll cause io_close_finish() to close | |
3018 | * the file again and cause a double CQE entry for this request | |
3019 | */ | |
3020 | io_queue_async_work(req); | |
3021 | return 0; | |
3022 | } | |
b5dba59e JA |
3023 | |
3024 | /* | |
3025 | * No ->flush(), safely close from here and just punt the | |
3026 | * fput() to async context. | |
3027 | */ | |
014db007 | 3028 | __io_close_finish(req); |
a93b3331 | 3029 | return 0; |
b5dba59e JA |
3030 | } |
3031 | ||
3529d8c2 | 3032 | static int io_prep_sfr(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
5d17b4a4 JA |
3033 | { |
3034 | struct io_ring_ctx *ctx = req->ctx; | |
5d17b4a4 JA |
3035 | |
3036 | if (!req->file) | |
3037 | return -EBADF; | |
5d17b4a4 JA |
3038 | |
3039 | if (unlikely(ctx->flags & IORING_SETUP_IOPOLL)) | |
3040 | return -EINVAL; | |
3041 | if (unlikely(sqe->addr || sqe->ioprio || sqe->buf_index)) | |
3042 | return -EINVAL; | |
3043 | ||
8ed8d3c3 JA |
3044 | req->sync.off = READ_ONCE(sqe->off); |
3045 | req->sync.len = READ_ONCE(sqe->len); | |
3046 | req->sync.flags = READ_ONCE(sqe->sync_range_flags); | |
8ed8d3c3 JA |
3047 | return 0; |
3048 | } | |
3049 | ||
014db007 | 3050 | static void __io_sync_file_range(struct io_kiocb *req) |
8ed8d3c3 | 3051 | { |
8ed8d3c3 JA |
3052 | int ret; |
3053 | ||
9adbd45d | 3054 | ret = sync_file_range(req->file, req->sync.off, req->sync.len, |
8ed8d3c3 JA |
3055 | req->sync.flags); |
3056 | if (ret < 0) | |
3057 | req_set_fail_links(req); | |
3058 | io_cqring_add_event(req, ret); | |
014db007 | 3059 | io_put_req(req); |
5ea62161 PB |
3060 | } |
3061 | ||
3062 | ||
3063 | static void io_sync_file_range_finish(struct io_wq_work **workptr) | |
3064 | { | |
3065 | struct io_kiocb *req = container_of(*workptr, struct io_kiocb, work); | |
3066 | struct io_kiocb *nxt = NULL; | |
3067 | ||
3068 | if (io_req_cancelled(req)) | |
3069 | return; | |
014db007 | 3070 | __io_sync_file_range(req); |
594506fe | 3071 | io_put_req(req); /* put submission ref */ |
8ed8d3c3 | 3072 | if (nxt) |
78912934 | 3073 | io_wq_assign_next(workptr, nxt); |
5d17b4a4 JA |
3074 | } |
3075 | ||
014db007 | 3076 | static int io_sync_file_range(struct io_kiocb *req, bool force_nonblock) |
5d17b4a4 | 3077 | { |
5d17b4a4 | 3078 | /* sync_file_range always requires a blocking context */ |
8ed8d3c3 | 3079 | if (force_nonblock) { |
8ed8d3c3 | 3080 | req->work.func = io_sync_file_range_finish; |
5d17b4a4 | 3081 | return -EAGAIN; |
8ed8d3c3 | 3082 | } |
5d17b4a4 | 3083 | |
014db007 | 3084 | __io_sync_file_range(req); |
5d17b4a4 JA |
3085 | return 0; |
3086 | } | |
3087 | ||
02d27d89 PB |
3088 | static int io_setup_async_msg(struct io_kiocb *req, |
3089 | struct io_async_msghdr *kmsg) | |
3090 | { | |
3091 | if (req->io) | |
3092 | return -EAGAIN; | |
3093 | if (io_alloc_async_ctx(req)) { | |
3094 | if (kmsg->iov != kmsg->fast_iov) | |
3095 | kfree(kmsg->iov); | |
3096 | return -ENOMEM; | |
3097 | } | |
3098 | req->flags |= REQ_F_NEED_CLEANUP; | |
3099 | memcpy(&req->io->msg, kmsg, sizeof(*kmsg)); | |
3100 | return -EAGAIN; | |
3101 | } | |
3102 | ||
3529d8c2 | 3103 | static int io_sendmsg_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
03b1230c | 3104 | { |
0fa03c62 | 3105 | #if defined(CONFIG_NET) |
e47293fd | 3106 | struct io_sr_msg *sr = &req->sr_msg; |
3529d8c2 | 3107 | struct io_async_ctx *io = req->io; |
99bc4c38 | 3108 | int ret; |
03b1230c | 3109 | |
e47293fd JA |
3110 | sr->msg_flags = READ_ONCE(sqe->msg_flags); |
3111 | sr->msg = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
fddaface | 3112 | sr->len = READ_ONCE(sqe->len); |
3529d8c2 | 3113 | |
d8768362 JA |
3114 | #ifdef CONFIG_COMPAT |
3115 | if (req->ctx->compat) | |
3116 | sr->msg_flags |= MSG_CMSG_COMPAT; | |
3117 | #endif | |
3118 | ||
fddaface | 3119 | if (!io || req->opcode == IORING_OP_SEND) |
3529d8c2 | 3120 | return 0; |
5f798bea PB |
3121 | /* iovec is already imported */ |
3122 | if (req->flags & REQ_F_NEED_CLEANUP) | |
3123 | return 0; | |
3529d8c2 | 3124 | |
d9688565 | 3125 | io->msg.iov = io->msg.fast_iov; |
99bc4c38 | 3126 | ret = sendmsg_copy_msghdr(&io->msg.msg, sr->msg, sr->msg_flags, |
e47293fd | 3127 | &io->msg.iov); |
99bc4c38 PB |
3128 | if (!ret) |
3129 | req->flags |= REQ_F_NEED_CLEANUP; | |
3130 | return ret; | |
03b1230c | 3131 | #else |
e47293fd | 3132 | return -EOPNOTSUPP; |
03b1230c JA |
3133 | #endif |
3134 | } | |
3135 | ||
014db007 | 3136 | static int io_sendmsg(struct io_kiocb *req, bool force_nonblock) |
aa1fa28f | 3137 | { |
03b1230c | 3138 | #if defined(CONFIG_NET) |
0b416c3e | 3139 | struct io_async_msghdr *kmsg = NULL; |
0fa03c62 JA |
3140 | struct socket *sock; |
3141 | int ret; | |
3142 | ||
3143 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) | |
3144 | return -EINVAL; | |
3145 | ||
3146 | sock = sock_from_file(req->file, &ret); | |
3147 | if (sock) { | |
b7bb4f7d | 3148 | struct io_async_ctx io; |
0fa03c62 JA |
3149 | unsigned flags; |
3150 | ||
03b1230c | 3151 | if (req->io) { |
0b416c3e | 3152 | kmsg = &req->io->msg; |
b537916c | 3153 | kmsg->msg.msg_name = &req->io->msg.addr; |
0b416c3e JA |
3154 | /* if iov is set, it's allocated already */ |
3155 | if (!kmsg->iov) | |
3156 | kmsg->iov = kmsg->fast_iov; | |
3157 | kmsg->msg.msg_iter.iov = kmsg->iov; | |
03b1230c | 3158 | } else { |
3529d8c2 JA |
3159 | struct io_sr_msg *sr = &req->sr_msg; |
3160 | ||
0b416c3e | 3161 | kmsg = &io.msg; |
b537916c | 3162 | kmsg->msg.msg_name = &io.msg.addr; |
3529d8c2 JA |
3163 | |
3164 | io.msg.iov = io.msg.fast_iov; | |
3165 | ret = sendmsg_copy_msghdr(&io.msg.msg, sr->msg, | |
3166 | sr->msg_flags, &io.msg.iov); | |
03b1230c | 3167 | if (ret) |
3529d8c2 | 3168 | return ret; |
03b1230c | 3169 | } |
0fa03c62 | 3170 | |
e47293fd JA |
3171 | flags = req->sr_msg.msg_flags; |
3172 | if (flags & MSG_DONTWAIT) | |
3173 | req->flags |= REQ_F_NOWAIT; | |
3174 | else if (force_nonblock) | |
3175 | flags |= MSG_DONTWAIT; | |
3176 | ||
0b416c3e | 3177 | ret = __sys_sendmsg_sock(sock, &kmsg->msg, flags); |
02d27d89 PB |
3178 | if (force_nonblock && ret == -EAGAIN) |
3179 | return io_setup_async_msg(req, kmsg); | |
441cdbd5 JA |
3180 | if (ret == -ERESTARTSYS) |
3181 | ret = -EINTR; | |
0fa03c62 JA |
3182 | } |
3183 | ||
1e95081c | 3184 | if (kmsg && kmsg->iov != kmsg->fast_iov) |
0b416c3e | 3185 | kfree(kmsg->iov); |
99bc4c38 | 3186 | req->flags &= ~REQ_F_NEED_CLEANUP; |
78e19bbe | 3187 | io_cqring_add_event(req, ret); |
4e88d6e7 JA |
3188 | if (ret < 0) |
3189 | req_set_fail_links(req); | |
014db007 | 3190 | io_put_req(req); |
5d17b4a4 | 3191 | return 0; |
03b1230c JA |
3192 | #else |
3193 | return -EOPNOTSUPP; | |
aa1fa28f | 3194 | #endif |
03b1230c | 3195 | } |
aa1fa28f | 3196 | |
014db007 | 3197 | static int io_send(struct io_kiocb *req, bool force_nonblock) |
fddaface JA |
3198 | { |
3199 | #if defined(CONFIG_NET) | |
3200 | struct socket *sock; | |
3201 | int ret; | |
3202 | ||
3203 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) | |
3204 | return -EINVAL; | |
3205 | ||
3206 | sock = sock_from_file(req->file, &ret); | |
3207 | if (sock) { | |
3208 | struct io_sr_msg *sr = &req->sr_msg; | |
3209 | struct msghdr msg; | |
3210 | struct iovec iov; | |
3211 | unsigned flags; | |
3212 | ||
3213 | ret = import_single_range(WRITE, sr->buf, sr->len, &iov, | |
3214 | &msg.msg_iter); | |
3215 | if (ret) | |
3216 | return ret; | |
3217 | ||
3218 | msg.msg_name = NULL; | |
3219 | msg.msg_control = NULL; | |
3220 | msg.msg_controllen = 0; | |
3221 | msg.msg_namelen = 0; | |
3222 | ||
3223 | flags = req->sr_msg.msg_flags; | |
3224 | if (flags & MSG_DONTWAIT) | |
3225 | req->flags |= REQ_F_NOWAIT; | |
3226 | else if (force_nonblock) | |
3227 | flags |= MSG_DONTWAIT; | |
3228 | ||
0b7b21e4 JA |
3229 | msg.msg_flags = flags; |
3230 | ret = sock_sendmsg(sock, &msg); | |
fddaface JA |
3231 | if (force_nonblock && ret == -EAGAIN) |
3232 | return -EAGAIN; | |
3233 | if (ret == -ERESTARTSYS) | |
3234 | ret = -EINTR; | |
3235 | } | |
3236 | ||
3237 | io_cqring_add_event(req, ret); | |
3238 | if (ret < 0) | |
3239 | req_set_fail_links(req); | |
014db007 | 3240 | io_put_req(req); |
fddaface JA |
3241 | return 0; |
3242 | #else | |
3243 | return -EOPNOTSUPP; | |
3244 | #endif | |
3245 | } | |
3246 | ||
3529d8c2 JA |
3247 | static int io_recvmsg_prep(struct io_kiocb *req, |
3248 | const struct io_uring_sqe *sqe) | |
aa1fa28f JA |
3249 | { |
3250 | #if defined(CONFIG_NET) | |
e47293fd | 3251 | struct io_sr_msg *sr = &req->sr_msg; |
3529d8c2 | 3252 | struct io_async_ctx *io = req->io; |
99bc4c38 | 3253 | int ret; |
3529d8c2 JA |
3254 | |
3255 | sr->msg_flags = READ_ONCE(sqe->msg_flags); | |
3256 | sr->msg = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
0b7b21e4 | 3257 | sr->len = READ_ONCE(sqe->len); |
06b76d44 | 3258 | |
d8768362 JA |
3259 | #ifdef CONFIG_COMPAT |
3260 | if (req->ctx->compat) | |
3261 | sr->msg_flags |= MSG_CMSG_COMPAT; | |
3262 | #endif | |
3263 | ||
fddaface | 3264 | if (!io || req->opcode == IORING_OP_RECV) |
06b76d44 | 3265 | return 0; |
5f798bea PB |
3266 | /* iovec is already imported */ |
3267 | if (req->flags & REQ_F_NEED_CLEANUP) | |
3268 | return 0; | |
03b1230c | 3269 | |
d9688565 | 3270 | io->msg.iov = io->msg.fast_iov; |
99bc4c38 | 3271 | ret = recvmsg_copy_msghdr(&io->msg.msg, sr->msg, sr->msg_flags, |
e47293fd | 3272 | &io->msg.uaddr, &io->msg.iov); |
99bc4c38 PB |
3273 | if (!ret) |
3274 | req->flags |= REQ_F_NEED_CLEANUP; | |
3275 | return ret; | |
aa1fa28f | 3276 | #else |
e47293fd | 3277 | return -EOPNOTSUPP; |
aa1fa28f JA |
3278 | #endif |
3279 | } | |
3280 | ||
014db007 | 3281 | static int io_recvmsg(struct io_kiocb *req, bool force_nonblock) |
aa1fa28f JA |
3282 | { |
3283 | #if defined(CONFIG_NET) | |
0b416c3e | 3284 | struct io_async_msghdr *kmsg = NULL; |
03b1230c JA |
3285 | struct socket *sock; |
3286 | int ret; | |
3287 | ||
3288 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) | |
3289 | return -EINVAL; | |
3290 | ||
3291 | sock = sock_from_file(req->file, &ret); | |
3292 | if (sock) { | |
b7bb4f7d | 3293 | struct io_async_ctx io; |
03b1230c JA |
3294 | unsigned flags; |
3295 | ||
03b1230c | 3296 | if (req->io) { |
0b416c3e | 3297 | kmsg = &req->io->msg; |
b537916c | 3298 | kmsg->msg.msg_name = &req->io->msg.addr; |
0b416c3e JA |
3299 | /* if iov is set, it's allocated already */ |
3300 | if (!kmsg->iov) | |
3301 | kmsg->iov = kmsg->fast_iov; | |
3302 | kmsg->msg.msg_iter.iov = kmsg->iov; | |
03b1230c | 3303 | } else { |
3529d8c2 JA |
3304 | struct io_sr_msg *sr = &req->sr_msg; |
3305 | ||
0b416c3e | 3306 | kmsg = &io.msg; |
b537916c | 3307 | kmsg->msg.msg_name = &io.msg.addr; |
3529d8c2 JA |
3308 | |
3309 | io.msg.iov = io.msg.fast_iov; | |
3310 | ret = recvmsg_copy_msghdr(&io.msg.msg, sr->msg, | |
3311 | sr->msg_flags, &io.msg.uaddr, | |
3312 | &io.msg.iov); | |
03b1230c | 3313 | if (ret) |
3529d8c2 | 3314 | return ret; |
03b1230c JA |
3315 | } |
3316 | ||
e47293fd JA |
3317 | flags = req->sr_msg.msg_flags; |
3318 | if (flags & MSG_DONTWAIT) | |
3319 | req->flags |= REQ_F_NOWAIT; | |
3320 | else if (force_nonblock) | |
3321 | flags |= MSG_DONTWAIT; | |
3322 | ||
3323 | ret = __sys_recvmsg_sock(sock, &kmsg->msg, req->sr_msg.msg, | |
3324 | kmsg->uaddr, flags); | |
02d27d89 PB |
3325 | if (force_nonblock && ret == -EAGAIN) |
3326 | return io_setup_async_msg(req, kmsg); | |
03b1230c JA |
3327 | if (ret == -ERESTARTSYS) |
3328 | ret = -EINTR; | |
3329 | } | |
3330 | ||
1e95081c | 3331 | if (kmsg && kmsg->iov != kmsg->fast_iov) |
0b416c3e | 3332 | kfree(kmsg->iov); |
99bc4c38 | 3333 | req->flags &= ~REQ_F_NEED_CLEANUP; |
03b1230c | 3334 | io_cqring_add_event(req, ret); |
4e88d6e7 JA |
3335 | if (ret < 0) |
3336 | req_set_fail_links(req); | |
014db007 | 3337 | io_put_req(req); |
03b1230c | 3338 | return 0; |
0fa03c62 JA |
3339 | #else |
3340 | return -EOPNOTSUPP; | |
3341 | #endif | |
3342 | } | |
5d17b4a4 | 3343 | |
014db007 | 3344 | static int io_recv(struct io_kiocb *req, bool force_nonblock) |
fddaface JA |
3345 | { |
3346 | #if defined(CONFIG_NET) | |
3347 | struct socket *sock; | |
3348 | int ret; | |
3349 | ||
3350 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) | |
3351 | return -EINVAL; | |
3352 | ||
3353 | sock = sock_from_file(req->file, &ret); | |
3354 | if (sock) { | |
3355 | struct io_sr_msg *sr = &req->sr_msg; | |
3356 | struct msghdr msg; | |
3357 | struct iovec iov; | |
3358 | unsigned flags; | |
3359 | ||
3360 | ret = import_single_range(READ, sr->buf, sr->len, &iov, | |
3361 | &msg.msg_iter); | |
3362 | if (ret) | |
3363 | return ret; | |
3364 | ||
3365 | msg.msg_name = NULL; | |
3366 | msg.msg_control = NULL; | |
3367 | msg.msg_controllen = 0; | |
3368 | msg.msg_namelen = 0; | |
3369 | msg.msg_iocb = NULL; | |
3370 | msg.msg_flags = 0; | |
3371 | ||
3372 | flags = req->sr_msg.msg_flags; | |
3373 | if (flags & MSG_DONTWAIT) | |
3374 | req->flags |= REQ_F_NOWAIT; | |
3375 | else if (force_nonblock) | |
3376 | flags |= MSG_DONTWAIT; | |
3377 | ||
0b7b21e4 | 3378 | ret = sock_recvmsg(sock, &msg, flags); |
fddaface JA |
3379 | if (force_nonblock && ret == -EAGAIN) |
3380 | return -EAGAIN; | |
3381 | if (ret == -ERESTARTSYS) | |
3382 | ret = -EINTR; | |
3383 | } | |
3384 | ||
3385 | io_cqring_add_event(req, ret); | |
3386 | if (ret < 0) | |
3387 | req_set_fail_links(req); | |
014db007 | 3388 | io_put_req(req); |
fddaface JA |
3389 | return 0; |
3390 | #else | |
3391 | return -EOPNOTSUPP; | |
3392 | #endif | |
3393 | } | |
3394 | ||
3395 | ||
3529d8c2 | 3396 | static int io_accept_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
17f2fe35 JA |
3397 | { |
3398 | #if defined(CONFIG_NET) | |
8ed8d3c3 JA |
3399 | struct io_accept *accept = &req->accept; |
3400 | ||
17f2fe35 JA |
3401 | if (unlikely(req->ctx->flags & (IORING_SETUP_IOPOLL|IORING_SETUP_SQPOLL))) |
3402 | return -EINVAL; | |
8042d6ce | 3403 | if (sqe->ioprio || sqe->len || sqe->buf_index) |
17f2fe35 JA |
3404 | return -EINVAL; |
3405 | ||
d55e5f5b JA |
3406 | accept->addr = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
3407 | accept->addr_len = u64_to_user_ptr(READ_ONCE(sqe->addr2)); | |
8ed8d3c3 | 3408 | accept->flags = READ_ONCE(sqe->accept_flags); |
8ed8d3c3 JA |
3409 | return 0; |
3410 | #else | |
3411 | return -EOPNOTSUPP; | |
3412 | #endif | |
3413 | } | |
17f2fe35 | 3414 | |
8ed8d3c3 | 3415 | #if defined(CONFIG_NET) |
014db007 | 3416 | static int __io_accept(struct io_kiocb *req, bool force_nonblock) |
8ed8d3c3 JA |
3417 | { |
3418 | struct io_accept *accept = &req->accept; | |
3419 | unsigned file_flags; | |
3420 | int ret; | |
3421 | ||
3422 | file_flags = force_nonblock ? O_NONBLOCK : 0; | |
3423 | ret = __sys_accept4_file(req->file, file_flags, accept->addr, | |
3424 | accept->addr_len, accept->flags); | |
3425 | if (ret == -EAGAIN && force_nonblock) | |
17f2fe35 | 3426 | return -EAGAIN; |
8e3cca12 JA |
3427 | if (ret == -ERESTARTSYS) |
3428 | ret = -EINTR; | |
4e88d6e7 JA |
3429 | if (ret < 0) |
3430 | req_set_fail_links(req); | |
78e19bbe | 3431 | io_cqring_add_event(req, ret); |
014db007 | 3432 | io_put_req(req); |
17f2fe35 | 3433 | return 0; |
8ed8d3c3 JA |
3434 | } |
3435 | ||
3436 | static void io_accept_finish(struct io_wq_work **workptr) | |
3437 | { | |
3438 | struct io_kiocb *req = container_of(*workptr, struct io_kiocb, work); | |
3439 | struct io_kiocb *nxt = NULL; | |
3440 | ||
3441 | if (io_req_cancelled(req)) | |
3442 | return; | |
014db007 | 3443 | __io_accept(req, false); |
594506fe | 3444 | io_put_req(req); /* drop submission reference */ |
8ed8d3c3 | 3445 | if (nxt) |
78912934 | 3446 | io_wq_assign_next(workptr, nxt); |
8ed8d3c3 JA |
3447 | } |
3448 | #endif | |
3449 | ||
014db007 | 3450 | static int io_accept(struct io_kiocb *req, bool force_nonblock) |
8ed8d3c3 JA |
3451 | { |
3452 | #if defined(CONFIG_NET) | |
3453 | int ret; | |
3454 | ||
014db007 | 3455 | ret = __io_accept(req, force_nonblock); |
8ed8d3c3 JA |
3456 | if (ret == -EAGAIN && force_nonblock) { |
3457 | req->work.func = io_accept_finish; | |
8ed8d3c3 JA |
3458 | return -EAGAIN; |
3459 | } | |
3460 | return 0; | |
0fa03c62 JA |
3461 | #else |
3462 | return -EOPNOTSUPP; | |
3463 | #endif | |
3464 | } | |
5d17b4a4 | 3465 | |
3529d8c2 | 3466 | static int io_connect_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
f499a021 JA |
3467 | { |
3468 | #if defined(CONFIG_NET) | |
3529d8c2 JA |
3469 | struct io_connect *conn = &req->connect; |
3470 | struct io_async_ctx *io = req->io; | |
f499a021 | 3471 | |
3fbb51c1 JA |
3472 | if (unlikely(req->ctx->flags & (IORING_SETUP_IOPOLL|IORING_SETUP_SQPOLL))) |
3473 | return -EINVAL; | |
3474 | if (sqe->ioprio || sqe->len || sqe->buf_index || sqe->rw_flags) | |
3475 | return -EINVAL; | |
3476 | ||
3529d8c2 JA |
3477 | conn->addr = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
3478 | conn->addr_len = READ_ONCE(sqe->addr2); | |
3479 | ||
3480 | if (!io) | |
3481 | return 0; | |
3482 | ||
3483 | return move_addr_to_kernel(conn->addr, conn->addr_len, | |
3fbb51c1 | 3484 | &io->connect.address); |
f499a021 | 3485 | #else |
3fbb51c1 | 3486 | return -EOPNOTSUPP; |
f499a021 JA |
3487 | #endif |
3488 | } | |
3489 | ||
014db007 | 3490 | static int io_connect(struct io_kiocb *req, bool force_nonblock) |
f8e85cf2 JA |
3491 | { |
3492 | #if defined(CONFIG_NET) | |
f499a021 | 3493 | struct io_async_ctx __io, *io; |
f8e85cf2 | 3494 | unsigned file_flags; |
3fbb51c1 | 3495 | int ret; |
f8e85cf2 | 3496 | |
f499a021 JA |
3497 | if (req->io) { |
3498 | io = req->io; | |
3499 | } else { | |
3529d8c2 JA |
3500 | ret = move_addr_to_kernel(req->connect.addr, |
3501 | req->connect.addr_len, | |
3502 | &__io.connect.address); | |
f499a021 JA |
3503 | if (ret) |
3504 | goto out; | |
3505 | io = &__io; | |
3506 | } | |
3507 | ||
3fbb51c1 JA |
3508 | file_flags = force_nonblock ? O_NONBLOCK : 0; |
3509 | ||
3510 | ret = __sys_connect_file(req->file, &io->connect.address, | |
3511 | req->connect.addr_len, file_flags); | |
87f80d62 | 3512 | if ((ret == -EAGAIN || ret == -EINPROGRESS) && force_nonblock) { |
b7bb4f7d JA |
3513 | if (req->io) |
3514 | return -EAGAIN; | |
3515 | if (io_alloc_async_ctx(req)) { | |
f499a021 JA |
3516 | ret = -ENOMEM; |
3517 | goto out; | |
3518 | } | |
b7bb4f7d | 3519 | memcpy(&req->io->connect, &__io.connect, sizeof(__io.connect)); |
f8e85cf2 | 3520 | return -EAGAIN; |
f499a021 | 3521 | } |
f8e85cf2 JA |
3522 | if (ret == -ERESTARTSYS) |
3523 | ret = -EINTR; | |
f499a021 | 3524 | out: |
4e88d6e7 JA |
3525 | if (ret < 0) |
3526 | req_set_fail_links(req); | |
f8e85cf2 | 3527 | io_cqring_add_event(req, ret); |
014db007 | 3528 | io_put_req(req); |
f8e85cf2 JA |
3529 | return 0; |
3530 | #else | |
3531 | return -EOPNOTSUPP; | |
3532 | #endif | |
3533 | } | |
3534 | ||
d7718a9d JA |
3535 | struct io_poll_table { |
3536 | struct poll_table_struct pt; | |
3537 | struct io_kiocb *req; | |
3538 | int error; | |
3539 | }; | |
3540 | ||
3541 | static void __io_queue_proc(struct io_poll_iocb *poll, struct io_poll_table *pt, | |
3542 | struct wait_queue_head *head) | |
3543 | { | |
3544 | if (unlikely(poll->head)) { | |
3545 | pt->error = -EINVAL; | |
3546 | return; | |
3547 | } | |
3548 | ||
3549 | pt->error = 0; | |
3550 | poll->head = head; | |
3551 | add_wait_queue(head, &poll->wait); | |
3552 | } | |
3553 | ||
3554 | static void io_async_queue_proc(struct file *file, struct wait_queue_head *head, | |
3555 | struct poll_table_struct *p) | |
3556 | { | |
3557 | struct io_poll_table *pt = container_of(p, struct io_poll_table, pt); | |
3558 | ||
3559 | __io_queue_proc(&pt->req->apoll->poll, pt, head); | |
3560 | } | |
3561 | ||
3562 | static int __io_async_wake(struct io_kiocb *req, struct io_poll_iocb *poll, | |
3563 | __poll_t mask, task_work_func_t func) | |
3564 | { | |
3565 | struct task_struct *tsk; | |
3566 | ||
3567 | /* for instances that support it check for an event match first: */ | |
3568 | if (mask && !(mask & poll->events)) | |
3569 | return 0; | |
3570 | ||
3571 | trace_io_uring_task_add(req->ctx, req->opcode, req->user_data, mask); | |
3572 | ||
3573 | list_del_init(&poll->wait.entry); | |
3574 | ||
3575 | tsk = req->task; | |
3576 | req->result = mask; | |
3577 | init_task_work(&req->task_work, func); | |
3578 | /* | |
3579 | * If this fails, then the task is exiting. If that is the case, then | |
3580 | * the exit check will ultimately cancel these work items. Hence we | |
3581 | * don't need to check here and handle it specifically. | |
3582 | */ | |
3583 | task_work_add(tsk, &req->task_work, true); | |
3584 | wake_up_process(tsk); | |
3585 | return 1; | |
3586 | } | |
3587 | ||
3588 | static void io_async_task_func(struct callback_head *cb) | |
3589 | { | |
3590 | struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work); | |
3591 | struct async_poll *apoll = req->apoll; | |
3592 | struct io_ring_ctx *ctx = req->ctx; | |
3593 | ||
3594 | trace_io_uring_task_run(req->ctx, req->opcode, req->user_data); | |
3595 | ||
3596 | WARN_ON_ONCE(!list_empty(&req->apoll->poll.wait.entry)); | |
3597 | ||
3598 | if (hash_hashed(&req->hash_node)) { | |
3599 | spin_lock_irq(&ctx->completion_lock); | |
3600 | hash_del(&req->hash_node); | |
3601 | spin_unlock_irq(&ctx->completion_lock); | |
3602 | } | |
3603 | ||
3604 | /* restore ->work in case we need to retry again */ | |
3605 | memcpy(&req->work, &apoll->work, sizeof(req->work)); | |
3606 | ||
3607 | __set_current_state(TASK_RUNNING); | |
3608 | mutex_lock(&ctx->uring_lock); | |
3609 | __io_queue_sqe(req, NULL); | |
3610 | mutex_unlock(&ctx->uring_lock); | |
3611 | ||
3612 | kfree(apoll); | |
3613 | } | |
3614 | ||
3615 | static int io_async_wake(struct wait_queue_entry *wait, unsigned mode, int sync, | |
3616 | void *key) | |
3617 | { | |
3618 | struct io_kiocb *req = wait->private; | |
3619 | struct io_poll_iocb *poll = &req->apoll->poll; | |
3620 | ||
3621 | trace_io_uring_poll_wake(req->ctx, req->opcode, req->user_data, | |
3622 | key_to_poll(key)); | |
3623 | ||
3624 | return __io_async_wake(req, poll, key_to_poll(key), io_async_task_func); | |
3625 | } | |
3626 | ||
3627 | static void io_poll_req_insert(struct io_kiocb *req) | |
3628 | { | |
3629 | struct io_ring_ctx *ctx = req->ctx; | |
3630 | struct hlist_head *list; | |
3631 | ||
3632 | list = &ctx->cancel_hash[hash_long(req->user_data, ctx->cancel_hash_bits)]; | |
3633 | hlist_add_head(&req->hash_node, list); | |
3634 | } | |
3635 | ||
3636 | static __poll_t __io_arm_poll_handler(struct io_kiocb *req, | |
3637 | struct io_poll_iocb *poll, | |
3638 | struct io_poll_table *ipt, __poll_t mask, | |
3639 | wait_queue_func_t wake_func) | |
3640 | __acquires(&ctx->completion_lock) | |
3641 | { | |
3642 | struct io_ring_ctx *ctx = req->ctx; | |
3643 | bool cancel = false; | |
3644 | ||
3645 | poll->file = req->file; | |
3646 | poll->head = NULL; | |
3647 | poll->done = poll->canceled = false; | |
3648 | poll->events = mask; | |
3649 | ||
3650 | ipt->pt._key = mask; | |
3651 | ipt->req = req; | |
3652 | ipt->error = -EINVAL; | |
3653 | ||
3654 | INIT_LIST_HEAD(&poll->wait.entry); | |
3655 | init_waitqueue_func_entry(&poll->wait, wake_func); | |
3656 | poll->wait.private = req; | |
3657 | ||
3658 | mask = vfs_poll(req->file, &ipt->pt) & poll->events; | |
3659 | ||
3660 | spin_lock_irq(&ctx->completion_lock); | |
3661 | if (likely(poll->head)) { | |
3662 | spin_lock(&poll->head->lock); | |
3663 | if (unlikely(list_empty(&poll->wait.entry))) { | |
3664 | if (ipt->error) | |
3665 | cancel = true; | |
3666 | ipt->error = 0; | |
3667 | mask = 0; | |
3668 | } | |
3669 | if (mask || ipt->error) | |
3670 | list_del_init(&poll->wait.entry); | |
3671 | else if (cancel) | |
3672 | WRITE_ONCE(poll->canceled, true); | |
3673 | else if (!poll->done) /* actually waiting for an event */ | |
3674 | io_poll_req_insert(req); | |
3675 | spin_unlock(&poll->head->lock); | |
3676 | } | |
3677 | ||
3678 | return mask; | |
3679 | } | |
3680 | ||
3681 | static bool io_arm_poll_handler(struct io_kiocb *req) | |
3682 | { | |
3683 | const struct io_op_def *def = &io_op_defs[req->opcode]; | |
3684 | struct io_ring_ctx *ctx = req->ctx; | |
3685 | struct async_poll *apoll; | |
3686 | struct io_poll_table ipt; | |
3687 | __poll_t mask, ret; | |
3688 | ||
3689 | if (!req->file || !file_can_poll(req->file)) | |
3690 | return false; | |
3691 | if (req->flags & (REQ_F_MUST_PUNT | REQ_F_POLLED)) | |
3692 | return false; | |
3693 | if (!def->pollin && !def->pollout) | |
3694 | return false; | |
3695 | ||
3696 | apoll = kmalloc(sizeof(*apoll), GFP_ATOMIC); | |
3697 | if (unlikely(!apoll)) | |
3698 | return false; | |
3699 | ||
3700 | req->flags |= REQ_F_POLLED; | |
3701 | memcpy(&apoll->work, &req->work, sizeof(req->work)); | |
3702 | ||
3703 | /* | |
3704 | * Don't need a reference here, as we're adding it to the task | |
3705 | * task_works list. If the task exits, the list is pruned. | |
3706 | */ | |
3707 | req->task = current; | |
3708 | req->apoll = apoll; | |
3709 | INIT_HLIST_NODE(&req->hash_node); | |
3710 | ||
8755d97a | 3711 | mask = 0; |
d7718a9d | 3712 | if (def->pollin) |
8755d97a | 3713 | mask |= POLLIN | POLLRDNORM; |
d7718a9d JA |
3714 | if (def->pollout) |
3715 | mask |= POLLOUT | POLLWRNORM; | |
3716 | mask |= POLLERR | POLLPRI; | |
3717 | ||
3718 | ipt.pt._qproc = io_async_queue_proc; | |
3719 | ||
3720 | ret = __io_arm_poll_handler(req, &apoll->poll, &ipt, mask, | |
3721 | io_async_wake); | |
3722 | if (ret) { | |
3723 | ipt.error = 0; | |
3724 | apoll->poll.done = true; | |
3725 | spin_unlock_irq(&ctx->completion_lock); | |
3726 | memcpy(&req->work, &apoll->work, sizeof(req->work)); | |
3727 | kfree(apoll); | |
3728 | return false; | |
3729 | } | |
3730 | spin_unlock_irq(&ctx->completion_lock); | |
3731 | trace_io_uring_poll_arm(ctx, req->opcode, req->user_data, mask, | |
3732 | apoll->poll.events); | |
3733 | return true; | |
3734 | } | |
3735 | ||
3736 | static bool __io_poll_remove_one(struct io_kiocb *req, | |
3737 | struct io_poll_iocb *poll) | |
221c5eb2 | 3738 | { |
b41e9852 | 3739 | bool do_complete = false; |
221c5eb2 JA |
3740 | |
3741 | spin_lock(&poll->head->lock); | |
3742 | WRITE_ONCE(poll->canceled, true); | |
392edb45 JA |
3743 | if (!list_empty(&poll->wait.entry)) { |
3744 | list_del_init(&poll->wait.entry); | |
b41e9852 | 3745 | do_complete = true; |
221c5eb2 JA |
3746 | } |
3747 | spin_unlock(&poll->head->lock); | |
d7718a9d JA |
3748 | return do_complete; |
3749 | } | |
3750 | ||
3751 | static bool io_poll_remove_one(struct io_kiocb *req) | |
3752 | { | |
3753 | bool do_complete; | |
3754 | ||
3755 | if (req->opcode == IORING_OP_POLL_ADD) { | |
3756 | do_complete = __io_poll_remove_one(req, &req->poll); | |
3757 | } else { | |
3758 | /* non-poll requests have submit ref still */ | |
3759 | do_complete = __io_poll_remove_one(req, &req->apoll->poll); | |
3760 | if (do_complete) | |
3761 | io_put_req(req); | |
3762 | } | |
3763 | ||
78076bb6 | 3764 | hash_del(&req->hash_node); |
d7718a9d | 3765 | |
b41e9852 JA |
3766 | if (do_complete) { |
3767 | io_cqring_fill_event(req, -ECANCELED); | |
3768 | io_commit_cqring(req->ctx); | |
3769 | req->flags |= REQ_F_COMP_LOCKED; | |
3770 | io_put_req(req); | |
3771 | } | |
3772 | ||
3773 | return do_complete; | |
221c5eb2 JA |
3774 | } |
3775 | ||
3776 | static void io_poll_remove_all(struct io_ring_ctx *ctx) | |
3777 | { | |
78076bb6 | 3778 | struct hlist_node *tmp; |
221c5eb2 | 3779 | struct io_kiocb *req; |
78076bb6 | 3780 | int i; |
221c5eb2 JA |
3781 | |
3782 | spin_lock_irq(&ctx->completion_lock); | |
78076bb6 JA |
3783 | for (i = 0; i < (1U << ctx->cancel_hash_bits); i++) { |
3784 | struct hlist_head *list; | |
3785 | ||
3786 | list = &ctx->cancel_hash[i]; | |
3787 | hlist_for_each_entry_safe(req, tmp, list, hash_node) | |
3788 | io_poll_remove_one(req); | |
221c5eb2 JA |
3789 | } |
3790 | spin_unlock_irq(&ctx->completion_lock); | |
b41e9852 JA |
3791 | |
3792 | io_cqring_ev_posted(ctx); | |
221c5eb2 JA |
3793 | } |
3794 | ||
47f46768 JA |
3795 | static int io_poll_cancel(struct io_ring_ctx *ctx, __u64 sqe_addr) |
3796 | { | |
78076bb6 | 3797 | struct hlist_head *list; |
47f46768 JA |
3798 | struct io_kiocb *req; |
3799 | ||
78076bb6 JA |
3800 | list = &ctx->cancel_hash[hash_long(sqe_addr, ctx->cancel_hash_bits)]; |
3801 | hlist_for_each_entry(req, list, hash_node) { | |
b41e9852 JA |
3802 | if (sqe_addr != req->user_data) |
3803 | continue; | |
3804 | if (io_poll_remove_one(req)) | |
eac406c6 | 3805 | return 0; |
b41e9852 | 3806 | return -EALREADY; |
47f46768 JA |
3807 | } |
3808 | ||
3809 | return -ENOENT; | |
3810 | } | |
3811 | ||
3529d8c2 JA |
3812 | static int io_poll_remove_prep(struct io_kiocb *req, |
3813 | const struct io_uring_sqe *sqe) | |
0969e783 | 3814 | { |
0969e783 JA |
3815 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
3816 | return -EINVAL; | |
3817 | if (sqe->ioprio || sqe->off || sqe->len || sqe->buf_index || | |
3818 | sqe->poll_events) | |
3819 | return -EINVAL; | |
3820 | ||
3821 | req->poll.addr = READ_ONCE(sqe->addr); | |
0969e783 JA |
3822 | return 0; |
3823 | } | |
3824 | ||
221c5eb2 JA |
3825 | /* |
3826 | * Find a running poll command that matches one specified in sqe->addr, | |
3827 | * and remove it if found. | |
3828 | */ | |
fc4df999 | 3829 | static int io_poll_remove(struct io_kiocb *req) |
221c5eb2 JA |
3830 | { |
3831 | struct io_ring_ctx *ctx = req->ctx; | |
0969e783 | 3832 | u64 addr; |
47f46768 | 3833 | int ret; |
221c5eb2 | 3834 | |
0969e783 | 3835 | addr = req->poll.addr; |
221c5eb2 | 3836 | spin_lock_irq(&ctx->completion_lock); |
0969e783 | 3837 | ret = io_poll_cancel(ctx, addr); |
221c5eb2 JA |
3838 | spin_unlock_irq(&ctx->completion_lock); |
3839 | ||
78e19bbe | 3840 | io_cqring_add_event(req, ret); |
4e88d6e7 JA |
3841 | if (ret < 0) |
3842 | req_set_fail_links(req); | |
e65ef56d | 3843 | io_put_req(req); |
221c5eb2 JA |
3844 | return 0; |
3845 | } | |
3846 | ||
b0dd8a41 | 3847 | static void io_poll_complete(struct io_kiocb *req, __poll_t mask, int error) |
221c5eb2 | 3848 | { |
a197f664 JL |
3849 | struct io_ring_ctx *ctx = req->ctx; |
3850 | ||
8c838788 | 3851 | req->poll.done = true; |
b0a20349 | 3852 | io_cqring_fill_event(req, error ? error : mangle_poll(mask)); |
8c838788 | 3853 | io_commit_cqring(ctx); |
221c5eb2 JA |
3854 | } |
3855 | ||
b41e9852 | 3856 | static void io_poll_task_handler(struct io_kiocb *req, struct io_kiocb **nxt) |
221c5eb2 | 3857 | { |
221c5eb2 | 3858 | struct io_ring_ctx *ctx = req->ctx; |
221c5eb2 | 3859 | |
221c5eb2 | 3860 | spin_lock_irq(&ctx->completion_lock); |
78076bb6 | 3861 | hash_del(&req->hash_node); |
b41e9852 JA |
3862 | io_poll_complete(req, req->result, 0); |
3863 | req->flags |= REQ_F_COMP_LOCKED; | |
3864 | io_put_req_find_next(req, nxt); | |
e94f141b JA |
3865 | spin_unlock_irq(&ctx->completion_lock); |
3866 | ||
3867 | io_cqring_ev_posted(ctx); | |
e94f141b JA |
3868 | } |
3869 | ||
b41e9852 | 3870 | static void io_poll_task_func(struct callback_head *cb) |
f0b493e6 | 3871 | { |
b41e9852 JA |
3872 | struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work); |
3873 | struct io_kiocb *nxt = NULL; | |
f0b493e6 | 3874 | |
b41e9852 | 3875 | io_poll_task_handler(req, &nxt); |
d7718a9d JA |
3876 | if (nxt) { |
3877 | struct io_ring_ctx *ctx = nxt->ctx; | |
3878 | ||
3879 | mutex_lock(&ctx->uring_lock); | |
b41e9852 | 3880 | __io_queue_sqe(nxt, NULL); |
d7718a9d JA |
3881 | mutex_unlock(&ctx->uring_lock); |
3882 | } | |
f0b493e6 JA |
3883 | } |
3884 | ||
221c5eb2 JA |
3885 | static int io_poll_wake(struct wait_queue_entry *wait, unsigned mode, int sync, |
3886 | void *key) | |
3887 | { | |
c2f2eb7d JA |
3888 | struct io_kiocb *req = wait->private; |
3889 | struct io_poll_iocb *poll = &req->poll; | |
221c5eb2 | 3890 | |
d7718a9d | 3891 | return __io_async_wake(req, poll, key_to_poll(key), io_poll_task_func); |
221c5eb2 JA |
3892 | } |
3893 | ||
221c5eb2 JA |
3894 | static void io_poll_queue_proc(struct file *file, struct wait_queue_head *head, |
3895 | struct poll_table_struct *p) | |
3896 | { | |
3897 | struct io_poll_table *pt = container_of(p, struct io_poll_table, pt); | |
3898 | ||
d7718a9d | 3899 | __io_queue_proc(&pt->req->poll, pt, head); |
eac406c6 JA |
3900 | } |
3901 | ||
3529d8c2 | 3902 | static int io_poll_add_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
221c5eb2 JA |
3903 | { |
3904 | struct io_poll_iocb *poll = &req->poll; | |
221c5eb2 | 3905 | u16 events; |
221c5eb2 JA |
3906 | |
3907 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) | |
3908 | return -EINVAL; | |
3909 | if (sqe->addr || sqe->ioprio || sqe->off || sqe->len || sqe->buf_index) | |
3910 | return -EINVAL; | |
09bb8394 JA |
3911 | if (!poll->file) |
3912 | return -EBADF; | |
221c5eb2 | 3913 | |
221c5eb2 JA |
3914 | events = READ_ONCE(sqe->poll_events); |
3915 | poll->events = demangle_poll(events) | EPOLLERR | EPOLLHUP; | |
b41e9852 | 3916 | |
d7718a9d JA |
3917 | /* |
3918 | * Don't need a reference here, as we're adding it to the task | |
3919 | * task_works list. If the task exits, the list is pruned. | |
3920 | */ | |
b41e9852 | 3921 | req->task = current; |
0969e783 JA |
3922 | return 0; |
3923 | } | |
3924 | ||
014db007 | 3925 | static int io_poll_add(struct io_kiocb *req) |
0969e783 JA |
3926 | { |
3927 | struct io_poll_iocb *poll = &req->poll; | |
3928 | struct io_ring_ctx *ctx = req->ctx; | |
3929 | struct io_poll_table ipt; | |
0969e783 | 3930 | __poll_t mask; |
0969e783 | 3931 | |
78076bb6 | 3932 | INIT_HLIST_NODE(&req->hash_node); |
36703247 | 3933 | INIT_LIST_HEAD(&req->list); |
d7718a9d | 3934 | ipt.pt._qproc = io_poll_queue_proc; |
36703247 | 3935 | |
d7718a9d JA |
3936 | mask = __io_arm_poll_handler(req, &req->poll, &ipt, poll->events, |
3937 | io_poll_wake); | |
221c5eb2 | 3938 | |
8c838788 | 3939 | if (mask) { /* no async, we'd stolen it */ |
221c5eb2 | 3940 | ipt.error = 0; |
b0dd8a41 | 3941 | io_poll_complete(req, mask, 0); |
221c5eb2 | 3942 | } |
221c5eb2 JA |
3943 | spin_unlock_irq(&ctx->completion_lock); |
3944 | ||
8c838788 JA |
3945 | if (mask) { |
3946 | io_cqring_ev_posted(ctx); | |
014db007 | 3947 | io_put_req(req); |
221c5eb2 | 3948 | } |
8c838788 | 3949 | return ipt.error; |
221c5eb2 JA |
3950 | } |
3951 | ||
5262f567 JA |
3952 | static enum hrtimer_restart io_timeout_fn(struct hrtimer *timer) |
3953 | { | |
ad8a48ac JA |
3954 | struct io_timeout_data *data = container_of(timer, |
3955 | struct io_timeout_data, timer); | |
3956 | struct io_kiocb *req = data->req; | |
3957 | struct io_ring_ctx *ctx = req->ctx; | |
5262f567 JA |
3958 | unsigned long flags; |
3959 | ||
5262f567 JA |
3960 | atomic_inc(&ctx->cq_timeouts); |
3961 | ||
3962 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
ef03681a | 3963 | /* |
11365043 JA |
3964 | * We could be racing with timeout deletion. If the list is empty, |
3965 | * then timeout lookup already found it and will be handling it. | |
ef03681a | 3966 | */ |
842f9612 | 3967 | if (!list_empty(&req->list)) { |
11365043 | 3968 | struct io_kiocb *prev; |
5262f567 | 3969 | |
11365043 JA |
3970 | /* |
3971 | * Adjust the reqs sequence before the current one because it | |
d195a66e | 3972 | * will consume a slot in the cq_ring and the cq_tail |
11365043 JA |
3973 | * pointer will be increased, otherwise other timeout reqs may |
3974 | * return in advance without waiting for enough wait_nr. | |
3975 | */ | |
3976 | prev = req; | |
3977 | list_for_each_entry_continue_reverse(prev, &ctx->timeout_list, list) | |
3978 | prev->sequence++; | |
11365043 | 3979 | list_del_init(&req->list); |
11365043 | 3980 | } |
5262f567 | 3981 | |
78e19bbe | 3982 | io_cqring_fill_event(req, -ETIME); |
5262f567 JA |
3983 | io_commit_cqring(ctx); |
3984 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
3985 | ||
3986 | io_cqring_ev_posted(ctx); | |
4e88d6e7 | 3987 | req_set_fail_links(req); |
5262f567 JA |
3988 | io_put_req(req); |
3989 | return HRTIMER_NORESTART; | |
3990 | } | |
3991 | ||
47f46768 JA |
3992 | static int io_timeout_cancel(struct io_ring_ctx *ctx, __u64 user_data) |
3993 | { | |
3994 | struct io_kiocb *req; | |
3995 | int ret = -ENOENT; | |
3996 | ||
3997 | list_for_each_entry(req, &ctx->timeout_list, list) { | |
3998 | if (user_data == req->user_data) { | |
3999 | list_del_init(&req->list); | |
4000 | ret = 0; | |
4001 | break; | |
4002 | } | |
4003 | } | |
4004 | ||
4005 | if (ret == -ENOENT) | |
4006 | return ret; | |
4007 | ||
2d28390a | 4008 | ret = hrtimer_try_to_cancel(&req->io->timeout.timer); |
47f46768 JA |
4009 | if (ret == -1) |
4010 | return -EALREADY; | |
4011 | ||
4e88d6e7 | 4012 | req_set_fail_links(req); |
47f46768 JA |
4013 | io_cqring_fill_event(req, -ECANCELED); |
4014 | io_put_req(req); | |
4015 | return 0; | |
4016 | } | |
4017 | ||
3529d8c2 JA |
4018 | static int io_timeout_remove_prep(struct io_kiocb *req, |
4019 | const struct io_uring_sqe *sqe) | |
b29472ee | 4020 | { |
b29472ee JA |
4021 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4022 | return -EINVAL; | |
4023 | if (sqe->flags || sqe->ioprio || sqe->buf_index || sqe->len) | |
4024 | return -EINVAL; | |
4025 | ||
4026 | req->timeout.addr = READ_ONCE(sqe->addr); | |
4027 | req->timeout.flags = READ_ONCE(sqe->timeout_flags); | |
4028 | if (req->timeout.flags) | |
4029 | return -EINVAL; | |
4030 | ||
b29472ee JA |
4031 | return 0; |
4032 | } | |
4033 | ||
11365043 JA |
4034 | /* |
4035 | * Remove or update an existing timeout command | |
4036 | */ | |
fc4df999 | 4037 | static int io_timeout_remove(struct io_kiocb *req) |
11365043 JA |
4038 | { |
4039 | struct io_ring_ctx *ctx = req->ctx; | |
47f46768 | 4040 | int ret; |
11365043 | 4041 | |
11365043 | 4042 | spin_lock_irq(&ctx->completion_lock); |
b29472ee | 4043 | ret = io_timeout_cancel(ctx, req->timeout.addr); |
11365043 | 4044 | |
47f46768 | 4045 | io_cqring_fill_event(req, ret); |
11365043 JA |
4046 | io_commit_cqring(ctx); |
4047 | spin_unlock_irq(&ctx->completion_lock); | |
5262f567 | 4048 | io_cqring_ev_posted(ctx); |
4e88d6e7 JA |
4049 | if (ret < 0) |
4050 | req_set_fail_links(req); | |
ec9c02ad | 4051 | io_put_req(req); |
11365043 | 4052 | return 0; |
5262f567 JA |
4053 | } |
4054 | ||
3529d8c2 | 4055 | static int io_timeout_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe, |
2d28390a | 4056 | bool is_timeout_link) |
5262f567 | 4057 | { |
ad8a48ac | 4058 | struct io_timeout_data *data; |
a41525ab | 4059 | unsigned flags; |
5262f567 | 4060 | |
ad8a48ac | 4061 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
5262f567 | 4062 | return -EINVAL; |
ad8a48ac | 4063 | if (sqe->ioprio || sqe->buf_index || sqe->len != 1) |
a41525ab | 4064 | return -EINVAL; |
2d28390a JA |
4065 | if (sqe->off && is_timeout_link) |
4066 | return -EINVAL; | |
a41525ab JA |
4067 | flags = READ_ONCE(sqe->timeout_flags); |
4068 | if (flags & ~IORING_TIMEOUT_ABS) | |
5262f567 | 4069 | return -EINVAL; |
bdf20073 | 4070 | |
26a61679 JA |
4071 | req->timeout.count = READ_ONCE(sqe->off); |
4072 | ||
3529d8c2 | 4073 | if (!req->io && io_alloc_async_ctx(req)) |
26a61679 JA |
4074 | return -ENOMEM; |
4075 | ||
4076 | data = &req->io->timeout; | |
ad8a48ac | 4077 | data->req = req; |
ad8a48ac JA |
4078 | req->flags |= REQ_F_TIMEOUT; |
4079 | ||
4080 | if (get_timespec64(&data->ts, u64_to_user_ptr(sqe->addr))) | |
5262f567 JA |
4081 | return -EFAULT; |
4082 | ||
11365043 | 4083 | if (flags & IORING_TIMEOUT_ABS) |
ad8a48ac | 4084 | data->mode = HRTIMER_MODE_ABS; |
11365043 | 4085 | else |
ad8a48ac | 4086 | data->mode = HRTIMER_MODE_REL; |
11365043 | 4087 | |
ad8a48ac JA |
4088 | hrtimer_init(&data->timer, CLOCK_MONOTONIC, data->mode); |
4089 | return 0; | |
4090 | } | |
4091 | ||
fc4df999 | 4092 | static int io_timeout(struct io_kiocb *req) |
ad8a48ac JA |
4093 | { |
4094 | unsigned count; | |
4095 | struct io_ring_ctx *ctx = req->ctx; | |
4096 | struct io_timeout_data *data; | |
4097 | struct list_head *entry; | |
4098 | unsigned span = 0; | |
ad8a48ac | 4099 | |
2d28390a | 4100 | data = &req->io->timeout; |
93bd25bb | 4101 | |
5262f567 JA |
4102 | /* |
4103 | * sqe->off holds how many events that need to occur for this | |
93bd25bb JA |
4104 | * timeout event to be satisfied. If it isn't set, then this is |
4105 | * a pure timeout request, sequence isn't used. | |
5262f567 | 4106 | */ |
26a61679 | 4107 | count = req->timeout.count; |
93bd25bb JA |
4108 | if (!count) { |
4109 | req->flags |= REQ_F_TIMEOUT_NOSEQ; | |
4110 | spin_lock_irq(&ctx->completion_lock); | |
4111 | entry = ctx->timeout_list.prev; | |
4112 | goto add; | |
4113 | } | |
5262f567 JA |
4114 | |
4115 | req->sequence = ctx->cached_sq_head + count - 1; | |
2d28390a | 4116 | data->seq_offset = count; |
5262f567 JA |
4117 | |
4118 | /* | |
4119 | * Insertion sort, ensuring the first entry in the list is always | |
4120 | * the one we need first. | |
4121 | */ | |
5262f567 JA |
4122 | spin_lock_irq(&ctx->completion_lock); |
4123 | list_for_each_prev(entry, &ctx->timeout_list) { | |
4124 | struct io_kiocb *nxt = list_entry(entry, struct io_kiocb, list); | |
5da0fb1a | 4125 | unsigned nxt_sq_head; |
4126 | long long tmp, tmp_nxt; | |
2d28390a | 4127 | u32 nxt_offset = nxt->io->timeout.seq_offset; |
5262f567 | 4128 | |
93bd25bb JA |
4129 | if (nxt->flags & REQ_F_TIMEOUT_NOSEQ) |
4130 | continue; | |
4131 | ||
5da0fb1a | 4132 | /* |
4133 | * Since cached_sq_head + count - 1 can overflow, use type long | |
4134 | * long to store it. | |
4135 | */ | |
4136 | tmp = (long long)ctx->cached_sq_head + count - 1; | |
cc42e0ac PB |
4137 | nxt_sq_head = nxt->sequence - nxt_offset + 1; |
4138 | tmp_nxt = (long long)nxt_sq_head + nxt_offset - 1; | |
5da0fb1a | 4139 | |
4140 | /* | |
4141 | * cached_sq_head may overflow, and it will never overflow twice | |
4142 | * once there is some timeout req still be valid. | |
4143 | */ | |
4144 | if (ctx->cached_sq_head < nxt_sq_head) | |
8b07a65a | 4145 | tmp += UINT_MAX; |
5da0fb1a | 4146 | |
a1f58ba4 | 4147 | if (tmp > tmp_nxt) |
5262f567 | 4148 | break; |
a1f58ba4 | 4149 | |
4150 | /* | |
4151 | * Sequence of reqs after the insert one and itself should | |
4152 | * be adjusted because each timeout req consumes a slot. | |
4153 | */ | |
4154 | span++; | |
4155 | nxt->sequence++; | |
5262f567 | 4156 | } |
a1f58ba4 | 4157 | req->sequence -= span; |
93bd25bb | 4158 | add: |
5262f567 | 4159 | list_add(&req->list, entry); |
ad8a48ac JA |
4160 | data->timer.function = io_timeout_fn; |
4161 | hrtimer_start(&data->timer, timespec64_to_ktime(data->ts), data->mode); | |
5262f567 | 4162 | spin_unlock_irq(&ctx->completion_lock); |
5262f567 JA |
4163 | return 0; |
4164 | } | |
5262f567 | 4165 | |
62755e35 JA |
4166 | static bool io_cancel_cb(struct io_wq_work *work, void *data) |
4167 | { | |
4168 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); | |
4169 | ||
4170 | return req->user_data == (unsigned long) data; | |
4171 | } | |
4172 | ||
e977d6d3 | 4173 | static int io_async_cancel_one(struct io_ring_ctx *ctx, void *sqe_addr) |
62755e35 | 4174 | { |
62755e35 | 4175 | enum io_wq_cancel cancel_ret; |
62755e35 JA |
4176 | int ret = 0; |
4177 | ||
62755e35 JA |
4178 | cancel_ret = io_wq_cancel_cb(ctx->io_wq, io_cancel_cb, sqe_addr); |
4179 | switch (cancel_ret) { | |
4180 | case IO_WQ_CANCEL_OK: | |
4181 | ret = 0; | |
4182 | break; | |
4183 | case IO_WQ_CANCEL_RUNNING: | |
4184 | ret = -EALREADY; | |
4185 | break; | |
4186 | case IO_WQ_CANCEL_NOTFOUND: | |
4187 | ret = -ENOENT; | |
4188 | break; | |
4189 | } | |
4190 | ||
e977d6d3 JA |
4191 | return ret; |
4192 | } | |
4193 | ||
47f46768 JA |
4194 | static void io_async_find_and_cancel(struct io_ring_ctx *ctx, |
4195 | struct io_kiocb *req, __u64 sqe_addr, | |
014db007 | 4196 | int success_ret) |
47f46768 JA |
4197 | { |
4198 | unsigned long flags; | |
4199 | int ret; | |
4200 | ||
4201 | ret = io_async_cancel_one(ctx, (void *) (unsigned long) sqe_addr); | |
4202 | if (ret != -ENOENT) { | |
4203 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
4204 | goto done; | |
4205 | } | |
4206 | ||
4207 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
4208 | ret = io_timeout_cancel(ctx, sqe_addr); | |
4209 | if (ret != -ENOENT) | |
4210 | goto done; | |
4211 | ret = io_poll_cancel(ctx, sqe_addr); | |
4212 | done: | |
b0dd8a41 JA |
4213 | if (!ret) |
4214 | ret = success_ret; | |
47f46768 JA |
4215 | io_cqring_fill_event(req, ret); |
4216 | io_commit_cqring(ctx); | |
4217 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
4218 | io_cqring_ev_posted(ctx); | |
4219 | ||
4e88d6e7 JA |
4220 | if (ret < 0) |
4221 | req_set_fail_links(req); | |
014db007 | 4222 | io_put_req(req); |
47f46768 JA |
4223 | } |
4224 | ||
3529d8c2 JA |
4225 | static int io_async_cancel_prep(struct io_kiocb *req, |
4226 | const struct io_uring_sqe *sqe) | |
e977d6d3 | 4227 | { |
fbf23849 | 4228 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
e977d6d3 JA |
4229 | return -EINVAL; |
4230 | if (sqe->flags || sqe->ioprio || sqe->off || sqe->len || | |
4231 | sqe->cancel_flags) | |
4232 | return -EINVAL; | |
4233 | ||
fbf23849 JA |
4234 | req->cancel.addr = READ_ONCE(sqe->addr); |
4235 | return 0; | |
4236 | } | |
4237 | ||
014db007 | 4238 | static int io_async_cancel(struct io_kiocb *req) |
fbf23849 JA |
4239 | { |
4240 | struct io_ring_ctx *ctx = req->ctx; | |
fbf23849 | 4241 | |
014db007 | 4242 | io_async_find_and_cancel(ctx, req, req->cancel.addr, 0); |
5262f567 JA |
4243 | return 0; |
4244 | } | |
4245 | ||
05f3fb3c JA |
4246 | static int io_files_update_prep(struct io_kiocb *req, |
4247 | const struct io_uring_sqe *sqe) | |
4248 | { | |
4249 | if (sqe->flags || sqe->ioprio || sqe->rw_flags) | |
4250 | return -EINVAL; | |
4251 | ||
4252 | req->files_update.offset = READ_ONCE(sqe->off); | |
4253 | req->files_update.nr_args = READ_ONCE(sqe->len); | |
4254 | if (!req->files_update.nr_args) | |
4255 | return -EINVAL; | |
4256 | req->files_update.arg = READ_ONCE(sqe->addr); | |
4257 | return 0; | |
4258 | } | |
4259 | ||
4260 | static int io_files_update(struct io_kiocb *req, bool force_nonblock) | |
fbf23849 JA |
4261 | { |
4262 | struct io_ring_ctx *ctx = req->ctx; | |
05f3fb3c JA |
4263 | struct io_uring_files_update up; |
4264 | int ret; | |
fbf23849 | 4265 | |
f86cd20c | 4266 | if (force_nonblock) |
05f3fb3c | 4267 | return -EAGAIN; |
05f3fb3c JA |
4268 | |
4269 | up.offset = req->files_update.offset; | |
4270 | up.fds = req->files_update.arg; | |
4271 | ||
4272 | mutex_lock(&ctx->uring_lock); | |
4273 | ret = __io_sqe_files_update(ctx, &up, req->files_update.nr_args); | |
4274 | mutex_unlock(&ctx->uring_lock); | |
4275 | ||
4276 | if (ret < 0) | |
4277 | req_set_fail_links(req); | |
4278 | io_cqring_add_event(req, ret); | |
4279 | io_put_req(req); | |
5262f567 JA |
4280 | return 0; |
4281 | } | |
4282 | ||
3529d8c2 JA |
4283 | static int io_req_defer_prep(struct io_kiocb *req, |
4284 | const struct io_uring_sqe *sqe) | |
f67676d1 | 4285 | { |
e781573e | 4286 | ssize_t ret = 0; |
f67676d1 | 4287 | |
f86cd20c JA |
4288 | if (io_op_defs[req->opcode].file_table) { |
4289 | ret = io_grab_files(req); | |
4290 | if (unlikely(ret)) | |
4291 | return ret; | |
4292 | } | |
4293 | ||
cccf0ee8 JA |
4294 | io_req_work_grab_env(req, &io_op_defs[req->opcode]); |
4295 | ||
d625c6ee | 4296 | switch (req->opcode) { |
e781573e JA |
4297 | case IORING_OP_NOP: |
4298 | break; | |
f67676d1 JA |
4299 | case IORING_OP_READV: |
4300 | case IORING_OP_READ_FIXED: | |
3a6820f2 | 4301 | case IORING_OP_READ: |
3529d8c2 | 4302 | ret = io_read_prep(req, sqe, true); |
f67676d1 JA |
4303 | break; |
4304 | case IORING_OP_WRITEV: | |
4305 | case IORING_OP_WRITE_FIXED: | |
3a6820f2 | 4306 | case IORING_OP_WRITE: |
3529d8c2 | 4307 | ret = io_write_prep(req, sqe, true); |
f67676d1 | 4308 | break; |
0969e783 | 4309 | case IORING_OP_POLL_ADD: |
3529d8c2 | 4310 | ret = io_poll_add_prep(req, sqe); |
0969e783 JA |
4311 | break; |
4312 | case IORING_OP_POLL_REMOVE: | |
3529d8c2 | 4313 | ret = io_poll_remove_prep(req, sqe); |
0969e783 | 4314 | break; |
8ed8d3c3 | 4315 | case IORING_OP_FSYNC: |
3529d8c2 | 4316 | ret = io_prep_fsync(req, sqe); |
8ed8d3c3 JA |
4317 | break; |
4318 | case IORING_OP_SYNC_FILE_RANGE: | |
3529d8c2 | 4319 | ret = io_prep_sfr(req, sqe); |
8ed8d3c3 | 4320 | break; |
03b1230c | 4321 | case IORING_OP_SENDMSG: |
fddaface | 4322 | case IORING_OP_SEND: |
3529d8c2 | 4323 | ret = io_sendmsg_prep(req, sqe); |
03b1230c JA |
4324 | break; |
4325 | case IORING_OP_RECVMSG: | |
fddaface | 4326 | case IORING_OP_RECV: |
3529d8c2 | 4327 | ret = io_recvmsg_prep(req, sqe); |
03b1230c | 4328 | break; |
f499a021 | 4329 | case IORING_OP_CONNECT: |
3529d8c2 | 4330 | ret = io_connect_prep(req, sqe); |
f499a021 | 4331 | break; |
2d28390a | 4332 | case IORING_OP_TIMEOUT: |
3529d8c2 | 4333 | ret = io_timeout_prep(req, sqe, false); |
b7bb4f7d | 4334 | break; |
b29472ee | 4335 | case IORING_OP_TIMEOUT_REMOVE: |
3529d8c2 | 4336 | ret = io_timeout_remove_prep(req, sqe); |
b29472ee | 4337 | break; |
fbf23849 | 4338 | case IORING_OP_ASYNC_CANCEL: |
3529d8c2 | 4339 | ret = io_async_cancel_prep(req, sqe); |
fbf23849 | 4340 | break; |
2d28390a | 4341 | case IORING_OP_LINK_TIMEOUT: |
3529d8c2 | 4342 | ret = io_timeout_prep(req, sqe, true); |
b7bb4f7d | 4343 | break; |
8ed8d3c3 | 4344 | case IORING_OP_ACCEPT: |
3529d8c2 | 4345 | ret = io_accept_prep(req, sqe); |
8ed8d3c3 | 4346 | break; |
d63d1b5e JA |
4347 | case IORING_OP_FALLOCATE: |
4348 | ret = io_fallocate_prep(req, sqe); | |
4349 | break; | |
15b71abe JA |
4350 | case IORING_OP_OPENAT: |
4351 | ret = io_openat_prep(req, sqe); | |
4352 | break; | |
b5dba59e JA |
4353 | case IORING_OP_CLOSE: |
4354 | ret = io_close_prep(req, sqe); | |
4355 | break; | |
05f3fb3c JA |
4356 | case IORING_OP_FILES_UPDATE: |
4357 | ret = io_files_update_prep(req, sqe); | |
4358 | break; | |
eddc7ef5 JA |
4359 | case IORING_OP_STATX: |
4360 | ret = io_statx_prep(req, sqe); | |
4361 | break; | |
4840e418 JA |
4362 | case IORING_OP_FADVISE: |
4363 | ret = io_fadvise_prep(req, sqe); | |
4364 | break; | |
c1ca757b JA |
4365 | case IORING_OP_MADVISE: |
4366 | ret = io_madvise_prep(req, sqe); | |
4367 | break; | |
cebdb986 JA |
4368 | case IORING_OP_OPENAT2: |
4369 | ret = io_openat2_prep(req, sqe); | |
4370 | break; | |
3e4827b0 JA |
4371 | case IORING_OP_EPOLL_CTL: |
4372 | ret = io_epoll_ctl_prep(req, sqe); | |
4373 | break; | |
7d67af2c PB |
4374 | case IORING_OP_SPLICE: |
4375 | ret = io_splice_prep(req, sqe); | |
4376 | break; | |
f67676d1 | 4377 | default: |
e781573e JA |
4378 | printk_once(KERN_WARNING "io_uring: unhandled opcode %d\n", |
4379 | req->opcode); | |
4380 | ret = -EINVAL; | |
b7bb4f7d | 4381 | break; |
f67676d1 JA |
4382 | } |
4383 | ||
b7bb4f7d | 4384 | return ret; |
f67676d1 JA |
4385 | } |
4386 | ||
3529d8c2 | 4387 | static int io_req_defer(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
de0617e4 | 4388 | { |
a197f664 | 4389 | struct io_ring_ctx *ctx = req->ctx; |
f67676d1 | 4390 | int ret; |
de0617e4 | 4391 | |
9d858b21 BL |
4392 | /* Still need defer if there is pending req in defer list. */ |
4393 | if (!req_need_defer(req) && list_empty(&ctx->defer_list)) | |
de0617e4 JA |
4394 | return 0; |
4395 | ||
3529d8c2 | 4396 | if (!req->io && io_alloc_async_ctx(req)) |
de0617e4 JA |
4397 | return -EAGAIN; |
4398 | ||
3529d8c2 | 4399 | ret = io_req_defer_prep(req, sqe); |
b7bb4f7d | 4400 | if (ret < 0) |
2d28390a | 4401 | return ret; |
2d28390a | 4402 | |
de0617e4 | 4403 | spin_lock_irq(&ctx->completion_lock); |
9d858b21 | 4404 | if (!req_need_defer(req) && list_empty(&ctx->defer_list)) { |
de0617e4 | 4405 | spin_unlock_irq(&ctx->completion_lock); |
de0617e4 JA |
4406 | return 0; |
4407 | } | |
4408 | ||
915967f6 | 4409 | trace_io_uring_defer(ctx, req, req->user_data); |
de0617e4 JA |
4410 | list_add_tail(&req->list, &ctx->defer_list); |
4411 | spin_unlock_irq(&ctx->completion_lock); | |
4412 | return -EIOCBQUEUED; | |
4413 | } | |
4414 | ||
99bc4c38 PB |
4415 | static void io_cleanup_req(struct io_kiocb *req) |
4416 | { | |
4417 | struct io_async_ctx *io = req->io; | |
4418 | ||
4419 | switch (req->opcode) { | |
4420 | case IORING_OP_READV: | |
4421 | case IORING_OP_READ_FIXED: | |
4422 | case IORING_OP_READ: | |
4423 | case IORING_OP_WRITEV: | |
4424 | case IORING_OP_WRITE_FIXED: | |
4425 | case IORING_OP_WRITE: | |
4426 | if (io->rw.iov != io->rw.fast_iov) | |
4427 | kfree(io->rw.iov); | |
4428 | break; | |
4429 | case IORING_OP_SENDMSG: | |
4430 | case IORING_OP_RECVMSG: | |
4431 | if (io->msg.iov != io->msg.fast_iov) | |
4432 | kfree(io->msg.iov); | |
4433 | break; | |
8fef80bf PB |
4434 | case IORING_OP_OPENAT: |
4435 | case IORING_OP_OPENAT2: | |
4436 | case IORING_OP_STATX: | |
4437 | putname(req->open.filename); | |
4438 | break; | |
7d67af2c PB |
4439 | case IORING_OP_SPLICE: |
4440 | io_put_file(req, req->splice.file_in, | |
4441 | (req->splice.flags & SPLICE_F_FD_IN_FIXED)); | |
4442 | break; | |
99bc4c38 PB |
4443 | } |
4444 | ||
4445 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
4446 | } | |
4447 | ||
3529d8c2 | 4448 | static int io_issue_sqe(struct io_kiocb *req, const struct io_uring_sqe *sqe, |
014db007 | 4449 | bool force_nonblock) |
2b188cc1 | 4450 | { |
a197f664 | 4451 | struct io_ring_ctx *ctx = req->ctx; |
d625c6ee | 4452 | int ret; |
2b188cc1 | 4453 | |
d625c6ee | 4454 | switch (req->opcode) { |
2b188cc1 | 4455 | case IORING_OP_NOP: |
78e19bbe | 4456 | ret = io_nop(req); |
2b188cc1 JA |
4457 | break; |
4458 | case IORING_OP_READV: | |
edafccee | 4459 | case IORING_OP_READ_FIXED: |
3a6820f2 | 4460 | case IORING_OP_READ: |
3529d8c2 JA |
4461 | if (sqe) { |
4462 | ret = io_read_prep(req, sqe, force_nonblock); | |
4463 | if (ret < 0) | |
4464 | break; | |
4465 | } | |
014db007 | 4466 | ret = io_read(req, force_nonblock); |
edafccee | 4467 | break; |
3529d8c2 | 4468 | case IORING_OP_WRITEV: |
edafccee | 4469 | case IORING_OP_WRITE_FIXED: |
3a6820f2 | 4470 | case IORING_OP_WRITE: |
3529d8c2 JA |
4471 | if (sqe) { |
4472 | ret = io_write_prep(req, sqe, force_nonblock); | |
4473 | if (ret < 0) | |
4474 | break; | |
4475 | } | |
014db007 | 4476 | ret = io_write(req, force_nonblock); |
2b188cc1 | 4477 | break; |
c992fe29 | 4478 | case IORING_OP_FSYNC: |
3529d8c2 JA |
4479 | if (sqe) { |
4480 | ret = io_prep_fsync(req, sqe); | |
4481 | if (ret < 0) | |
4482 | break; | |
4483 | } | |
014db007 | 4484 | ret = io_fsync(req, force_nonblock); |
c992fe29 | 4485 | break; |
221c5eb2 | 4486 | case IORING_OP_POLL_ADD: |
3529d8c2 JA |
4487 | if (sqe) { |
4488 | ret = io_poll_add_prep(req, sqe); | |
4489 | if (ret) | |
4490 | break; | |
4491 | } | |
014db007 | 4492 | ret = io_poll_add(req); |
221c5eb2 JA |
4493 | break; |
4494 | case IORING_OP_POLL_REMOVE: | |
3529d8c2 JA |
4495 | if (sqe) { |
4496 | ret = io_poll_remove_prep(req, sqe); | |
4497 | if (ret < 0) | |
4498 | break; | |
4499 | } | |
fc4df999 | 4500 | ret = io_poll_remove(req); |
221c5eb2 | 4501 | break; |
5d17b4a4 | 4502 | case IORING_OP_SYNC_FILE_RANGE: |
3529d8c2 JA |
4503 | if (sqe) { |
4504 | ret = io_prep_sfr(req, sqe); | |
4505 | if (ret < 0) | |
4506 | break; | |
4507 | } | |
014db007 | 4508 | ret = io_sync_file_range(req, force_nonblock); |
5d17b4a4 | 4509 | break; |
0fa03c62 | 4510 | case IORING_OP_SENDMSG: |
fddaface | 4511 | case IORING_OP_SEND: |
3529d8c2 JA |
4512 | if (sqe) { |
4513 | ret = io_sendmsg_prep(req, sqe); | |
4514 | if (ret < 0) | |
4515 | break; | |
4516 | } | |
fddaface | 4517 | if (req->opcode == IORING_OP_SENDMSG) |
014db007 | 4518 | ret = io_sendmsg(req, force_nonblock); |
fddaface | 4519 | else |
014db007 | 4520 | ret = io_send(req, force_nonblock); |
0fa03c62 | 4521 | break; |
aa1fa28f | 4522 | case IORING_OP_RECVMSG: |
fddaface | 4523 | case IORING_OP_RECV: |
3529d8c2 JA |
4524 | if (sqe) { |
4525 | ret = io_recvmsg_prep(req, sqe); | |
4526 | if (ret) | |
4527 | break; | |
4528 | } | |
fddaface | 4529 | if (req->opcode == IORING_OP_RECVMSG) |
014db007 | 4530 | ret = io_recvmsg(req, force_nonblock); |
fddaface | 4531 | else |
014db007 | 4532 | ret = io_recv(req, force_nonblock); |
aa1fa28f | 4533 | break; |
5262f567 | 4534 | case IORING_OP_TIMEOUT: |
3529d8c2 JA |
4535 | if (sqe) { |
4536 | ret = io_timeout_prep(req, sqe, false); | |
4537 | if (ret) | |
4538 | break; | |
4539 | } | |
fc4df999 | 4540 | ret = io_timeout(req); |
5262f567 | 4541 | break; |
11365043 | 4542 | case IORING_OP_TIMEOUT_REMOVE: |
3529d8c2 JA |
4543 | if (sqe) { |
4544 | ret = io_timeout_remove_prep(req, sqe); | |
4545 | if (ret) | |
4546 | break; | |
4547 | } | |
fc4df999 | 4548 | ret = io_timeout_remove(req); |
11365043 | 4549 | break; |
17f2fe35 | 4550 | case IORING_OP_ACCEPT: |
3529d8c2 JA |
4551 | if (sqe) { |
4552 | ret = io_accept_prep(req, sqe); | |
4553 | if (ret) | |
4554 | break; | |
4555 | } | |
014db007 | 4556 | ret = io_accept(req, force_nonblock); |
17f2fe35 | 4557 | break; |
f8e85cf2 | 4558 | case IORING_OP_CONNECT: |
3529d8c2 JA |
4559 | if (sqe) { |
4560 | ret = io_connect_prep(req, sqe); | |
4561 | if (ret) | |
4562 | break; | |
4563 | } | |
014db007 | 4564 | ret = io_connect(req, force_nonblock); |
f8e85cf2 | 4565 | break; |
62755e35 | 4566 | case IORING_OP_ASYNC_CANCEL: |
3529d8c2 JA |
4567 | if (sqe) { |
4568 | ret = io_async_cancel_prep(req, sqe); | |
4569 | if (ret) | |
4570 | break; | |
4571 | } | |
014db007 | 4572 | ret = io_async_cancel(req); |
62755e35 | 4573 | break; |
d63d1b5e JA |
4574 | case IORING_OP_FALLOCATE: |
4575 | if (sqe) { | |
4576 | ret = io_fallocate_prep(req, sqe); | |
4577 | if (ret) | |
4578 | break; | |
4579 | } | |
014db007 | 4580 | ret = io_fallocate(req, force_nonblock); |
d63d1b5e | 4581 | break; |
15b71abe JA |
4582 | case IORING_OP_OPENAT: |
4583 | if (sqe) { | |
4584 | ret = io_openat_prep(req, sqe); | |
4585 | if (ret) | |
4586 | break; | |
4587 | } | |
014db007 | 4588 | ret = io_openat(req, force_nonblock); |
15b71abe | 4589 | break; |
b5dba59e JA |
4590 | case IORING_OP_CLOSE: |
4591 | if (sqe) { | |
4592 | ret = io_close_prep(req, sqe); | |
4593 | if (ret) | |
4594 | break; | |
4595 | } | |
014db007 | 4596 | ret = io_close(req, force_nonblock); |
b5dba59e | 4597 | break; |
05f3fb3c JA |
4598 | case IORING_OP_FILES_UPDATE: |
4599 | if (sqe) { | |
4600 | ret = io_files_update_prep(req, sqe); | |
4601 | if (ret) | |
4602 | break; | |
4603 | } | |
4604 | ret = io_files_update(req, force_nonblock); | |
4605 | break; | |
eddc7ef5 JA |
4606 | case IORING_OP_STATX: |
4607 | if (sqe) { | |
4608 | ret = io_statx_prep(req, sqe); | |
4609 | if (ret) | |
4610 | break; | |
4611 | } | |
014db007 | 4612 | ret = io_statx(req, force_nonblock); |
eddc7ef5 | 4613 | break; |
4840e418 JA |
4614 | case IORING_OP_FADVISE: |
4615 | if (sqe) { | |
4616 | ret = io_fadvise_prep(req, sqe); | |
4617 | if (ret) | |
4618 | break; | |
4619 | } | |
014db007 | 4620 | ret = io_fadvise(req, force_nonblock); |
4840e418 | 4621 | break; |
c1ca757b JA |
4622 | case IORING_OP_MADVISE: |
4623 | if (sqe) { | |
4624 | ret = io_madvise_prep(req, sqe); | |
4625 | if (ret) | |
4626 | break; | |
4627 | } | |
014db007 | 4628 | ret = io_madvise(req, force_nonblock); |
c1ca757b | 4629 | break; |
cebdb986 JA |
4630 | case IORING_OP_OPENAT2: |
4631 | if (sqe) { | |
4632 | ret = io_openat2_prep(req, sqe); | |
4633 | if (ret) | |
4634 | break; | |
4635 | } | |
014db007 | 4636 | ret = io_openat2(req, force_nonblock); |
cebdb986 | 4637 | break; |
3e4827b0 JA |
4638 | case IORING_OP_EPOLL_CTL: |
4639 | if (sqe) { | |
4640 | ret = io_epoll_ctl_prep(req, sqe); | |
4641 | if (ret) | |
4642 | break; | |
4643 | } | |
014db007 | 4644 | ret = io_epoll_ctl(req, force_nonblock); |
3e4827b0 | 4645 | break; |
7d67af2c PB |
4646 | case IORING_OP_SPLICE: |
4647 | if (sqe) { | |
4648 | ret = io_splice_prep(req, sqe); | |
4649 | if (ret < 0) | |
4650 | break; | |
4651 | } | |
014db007 | 4652 | ret = io_splice(req, force_nonblock); |
7d67af2c | 4653 | break; |
2b188cc1 JA |
4654 | default: |
4655 | ret = -EINVAL; | |
4656 | break; | |
4657 | } | |
4658 | ||
def596e9 JA |
4659 | if (ret) |
4660 | return ret; | |
4661 | ||
4662 | if (ctx->flags & IORING_SETUP_IOPOLL) { | |
11ba820b JA |
4663 | const bool in_async = io_wq_current_is_worker(); |
4664 | ||
9e645e11 | 4665 | if (req->result == -EAGAIN) |
def596e9 JA |
4666 | return -EAGAIN; |
4667 | ||
11ba820b JA |
4668 | /* workqueue context doesn't hold uring_lock, grab it now */ |
4669 | if (in_async) | |
4670 | mutex_lock(&ctx->uring_lock); | |
4671 | ||
def596e9 | 4672 | io_iopoll_req_issued(req); |
11ba820b JA |
4673 | |
4674 | if (in_async) | |
4675 | mutex_unlock(&ctx->uring_lock); | |
def596e9 JA |
4676 | } |
4677 | ||
4678 | return 0; | |
2b188cc1 JA |
4679 | } |
4680 | ||
561fb04a | 4681 | static void io_wq_submit_work(struct io_wq_work **workptr) |
2b188cc1 | 4682 | { |
561fb04a | 4683 | struct io_wq_work *work = *workptr; |
2b188cc1 | 4684 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); |
561fb04a JA |
4685 | struct io_kiocb *nxt = NULL; |
4686 | int ret = 0; | |
2b188cc1 | 4687 | |
0c9d5ccd JA |
4688 | /* if NO_CANCEL is set, we must still run the work */ |
4689 | if ((work->flags & (IO_WQ_WORK_CANCEL|IO_WQ_WORK_NO_CANCEL)) == | |
4690 | IO_WQ_WORK_CANCEL) { | |
561fb04a | 4691 | ret = -ECANCELED; |
0c9d5ccd | 4692 | } |
31b51510 | 4693 | |
561fb04a | 4694 | if (!ret) { |
561fb04a | 4695 | do { |
014db007 | 4696 | ret = io_issue_sqe(req, NULL, false); |
561fb04a JA |
4697 | /* |
4698 | * We can get EAGAIN for polled IO even though we're | |
4699 | * forcing a sync submission from here, since we can't | |
4700 | * wait for request slots on the block side. | |
4701 | */ | |
4702 | if (ret != -EAGAIN) | |
4703 | break; | |
4704 | cond_resched(); | |
4705 | } while (1); | |
4706 | } | |
31b51510 | 4707 | |
561fb04a | 4708 | if (ret) { |
4e88d6e7 | 4709 | req_set_fail_links(req); |
78e19bbe | 4710 | io_cqring_add_event(req, ret); |
817869d2 | 4711 | io_put_req(req); |
edafccee | 4712 | } |
2b188cc1 | 4713 | |
594506fe PB |
4714 | io_put_req(req); /* drop submission reference */ |
4715 | if (nxt) | |
78912934 | 4716 | io_wq_assign_next(workptr, nxt); |
2b188cc1 JA |
4717 | } |
4718 | ||
15b71abe | 4719 | static int io_req_needs_file(struct io_kiocb *req, int fd) |
9e3aa61a | 4720 | { |
d3656344 | 4721 | if (!io_op_defs[req->opcode].needs_file) |
9e3aa61a | 4722 | return 0; |
0b5faf6b | 4723 | if ((fd == -1 || fd == AT_FDCWD) && io_op_defs[req->opcode].fd_non_neg) |
d3656344 JA |
4724 | return 0; |
4725 | return 1; | |
09bb8394 JA |
4726 | } |
4727 | ||
65e19f54 JA |
4728 | static inline struct file *io_file_from_index(struct io_ring_ctx *ctx, |
4729 | int index) | |
4730 | { | |
4731 | struct fixed_file_table *table; | |
4732 | ||
05f3fb3c JA |
4733 | table = &ctx->file_data->table[index >> IORING_FILE_TABLE_SHIFT]; |
4734 | return table->files[index & IORING_FILE_TABLE_MASK];; | |
65e19f54 JA |
4735 | } |
4736 | ||
8da11c19 PB |
4737 | static int io_file_get(struct io_submit_state *state, struct io_kiocb *req, |
4738 | int fd, struct file **out_file, bool fixed) | |
09bb8394 | 4739 | { |
a197f664 | 4740 | struct io_ring_ctx *ctx = req->ctx; |
8da11c19 | 4741 | struct file *file; |
09bb8394 | 4742 | |
8da11c19 | 4743 | if (fixed) { |
05f3fb3c | 4744 | if (unlikely(!ctx->file_data || |
09bb8394 JA |
4745 | (unsigned) fd >= ctx->nr_user_files)) |
4746 | return -EBADF; | |
b7620121 | 4747 | fd = array_index_nospec(fd, ctx->nr_user_files); |
8da11c19 PB |
4748 | file = io_file_from_index(ctx, fd); |
4749 | if (!file) | |
08a45173 | 4750 | return -EBADF; |
05f3fb3c | 4751 | percpu_ref_get(&ctx->file_data->refs); |
09bb8394 | 4752 | } else { |
c826bd7a | 4753 | trace_io_uring_file_get(ctx, fd); |
8da11c19 PB |
4754 | file = __io_file_get(state, fd); |
4755 | if (unlikely(!file)) | |
09bb8394 JA |
4756 | return -EBADF; |
4757 | } | |
4758 | ||
8da11c19 | 4759 | *out_file = file; |
09bb8394 JA |
4760 | return 0; |
4761 | } | |
4762 | ||
8da11c19 PB |
4763 | static int io_req_set_file(struct io_submit_state *state, struct io_kiocb *req, |
4764 | const struct io_uring_sqe *sqe) | |
4765 | { | |
4766 | unsigned flags; | |
4767 | int fd; | |
4768 | bool fixed; | |
4769 | ||
4770 | flags = READ_ONCE(sqe->flags); | |
4771 | fd = READ_ONCE(sqe->fd); | |
4772 | ||
4773 | if (!io_req_needs_file(req, fd)) | |
4774 | return 0; | |
4775 | ||
4776 | fixed = (flags & IOSQE_FIXED_FILE); | |
4777 | if (unlikely(!fixed && req->needs_fixed_file)) | |
4778 | return -EBADF; | |
4779 | ||
4780 | return io_file_get(state, req, fd, &req->file, fixed); | |
4781 | } | |
4782 | ||
a197f664 | 4783 | static int io_grab_files(struct io_kiocb *req) |
fcb323cc JA |
4784 | { |
4785 | int ret = -EBADF; | |
a197f664 | 4786 | struct io_ring_ctx *ctx = req->ctx; |
fcb323cc | 4787 | |
f86cd20c JA |
4788 | if (req->work.files) |
4789 | return 0; | |
b14cca0c | 4790 | if (!ctx->ring_file) |
b5dba59e JA |
4791 | return -EBADF; |
4792 | ||
fcb323cc JA |
4793 | rcu_read_lock(); |
4794 | spin_lock_irq(&ctx->inflight_lock); | |
4795 | /* | |
4796 | * We use the f_ops->flush() handler to ensure that we can flush | |
4797 | * out work accessing these files if the fd is closed. Check if | |
4798 | * the fd has changed since we started down this path, and disallow | |
4799 | * this operation if it has. | |
4800 | */ | |
b14cca0c | 4801 | if (fcheck(ctx->ring_fd) == ctx->ring_file) { |
fcb323cc JA |
4802 | list_add(&req->inflight_entry, &ctx->inflight_list); |
4803 | req->flags |= REQ_F_INFLIGHT; | |
4804 | req->work.files = current->files; | |
4805 | ret = 0; | |
4806 | } | |
4807 | spin_unlock_irq(&ctx->inflight_lock); | |
4808 | rcu_read_unlock(); | |
4809 | ||
4810 | return ret; | |
4811 | } | |
4812 | ||
2665abfd | 4813 | static enum hrtimer_restart io_link_timeout_fn(struct hrtimer *timer) |
2b188cc1 | 4814 | { |
ad8a48ac JA |
4815 | struct io_timeout_data *data = container_of(timer, |
4816 | struct io_timeout_data, timer); | |
4817 | struct io_kiocb *req = data->req; | |
2665abfd JA |
4818 | struct io_ring_ctx *ctx = req->ctx; |
4819 | struct io_kiocb *prev = NULL; | |
4820 | unsigned long flags; | |
2665abfd JA |
4821 | |
4822 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
4823 | ||
4824 | /* | |
4825 | * We don't expect the list to be empty, that will only happen if we | |
4826 | * race with the completion of the linked work. | |
4827 | */ | |
4493233e PB |
4828 | if (!list_empty(&req->link_list)) { |
4829 | prev = list_entry(req->link_list.prev, struct io_kiocb, | |
4830 | link_list); | |
5d960724 | 4831 | if (refcount_inc_not_zero(&prev->refs)) { |
4493233e | 4832 | list_del_init(&req->link_list); |
5d960724 JA |
4833 | prev->flags &= ~REQ_F_LINK_TIMEOUT; |
4834 | } else | |
76a46e06 | 4835 | prev = NULL; |
2665abfd JA |
4836 | } |
4837 | ||
4838 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
4839 | ||
4840 | if (prev) { | |
4e88d6e7 | 4841 | req_set_fail_links(prev); |
014db007 | 4842 | io_async_find_and_cancel(ctx, req, prev->user_data, -ETIME); |
76a46e06 | 4843 | io_put_req(prev); |
47f46768 JA |
4844 | } else { |
4845 | io_cqring_add_event(req, -ETIME); | |
4846 | io_put_req(req); | |
2665abfd | 4847 | } |
2665abfd JA |
4848 | return HRTIMER_NORESTART; |
4849 | } | |
4850 | ||
ad8a48ac | 4851 | static void io_queue_linked_timeout(struct io_kiocb *req) |
2665abfd | 4852 | { |
76a46e06 | 4853 | struct io_ring_ctx *ctx = req->ctx; |
2665abfd | 4854 | |
76a46e06 JA |
4855 | /* |
4856 | * If the list is now empty, then our linked request finished before | |
4857 | * we got a chance to setup the timer | |
4858 | */ | |
4859 | spin_lock_irq(&ctx->completion_lock); | |
4493233e | 4860 | if (!list_empty(&req->link_list)) { |
2d28390a | 4861 | struct io_timeout_data *data = &req->io->timeout; |
94ae5e77 | 4862 | |
ad8a48ac JA |
4863 | data->timer.function = io_link_timeout_fn; |
4864 | hrtimer_start(&data->timer, timespec64_to_ktime(data->ts), | |
4865 | data->mode); | |
2665abfd | 4866 | } |
76a46e06 | 4867 | spin_unlock_irq(&ctx->completion_lock); |
2665abfd | 4868 | |
2665abfd | 4869 | /* drop submission reference */ |
76a46e06 JA |
4870 | io_put_req(req); |
4871 | } | |
2665abfd | 4872 | |
ad8a48ac | 4873 | static struct io_kiocb *io_prep_linked_timeout(struct io_kiocb *req) |
2665abfd JA |
4874 | { |
4875 | struct io_kiocb *nxt; | |
4876 | ||
4877 | if (!(req->flags & REQ_F_LINK)) | |
4878 | return NULL; | |
d7718a9d JA |
4879 | /* for polled retry, if flag is set, we already went through here */ |
4880 | if (req->flags & REQ_F_POLLED) | |
4881 | return NULL; | |
2665abfd | 4882 | |
4493233e PB |
4883 | nxt = list_first_entry_or_null(&req->link_list, struct io_kiocb, |
4884 | link_list); | |
d625c6ee | 4885 | if (!nxt || nxt->opcode != IORING_OP_LINK_TIMEOUT) |
76a46e06 | 4886 | return NULL; |
2665abfd | 4887 | |
76a46e06 | 4888 | req->flags |= REQ_F_LINK_TIMEOUT; |
76a46e06 | 4889 | return nxt; |
2665abfd JA |
4890 | } |
4891 | ||
3529d8c2 | 4892 | static void __io_queue_sqe(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
2b188cc1 | 4893 | { |
4a0a7a18 | 4894 | struct io_kiocb *linked_timeout; |
4bc4494e | 4895 | struct io_kiocb *nxt; |
193155c8 | 4896 | const struct cred *old_creds = NULL; |
e0c5c576 | 4897 | int ret; |
2b188cc1 | 4898 | |
4a0a7a18 JA |
4899 | again: |
4900 | linked_timeout = io_prep_linked_timeout(req); | |
4901 | ||
193155c8 JA |
4902 | if (req->work.creds && req->work.creds != current_cred()) { |
4903 | if (old_creds) | |
4904 | revert_creds(old_creds); | |
4905 | if (old_creds == req->work.creds) | |
4906 | old_creds = NULL; /* restored original creds */ | |
4907 | else | |
4908 | old_creds = override_creds(req->work.creds); | |
4909 | } | |
4910 | ||
014db007 | 4911 | ret = io_issue_sqe(req, sqe, true); |
491381ce JA |
4912 | |
4913 | /* | |
4914 | * We async punt it if the file wasn't marked NOWAIT, or if the file | |
4915 | * doesn't support non-blocking read/write attempts | |
4916 | */ | |
4917 | if (ret == -EAGAIN && (!(req->flags & REQ_F_NOWAIT) || | |
4918 | (req->flags & REQ_F_MUST_PUNT))) { | |
d7718a9d JA |
4919 | if (io_arm_poll_handler(req)) { |
4920 | if (linked_timeout) | |
4921 | io_queue_linked_timeout(linked_timeout); | |
4bc4494e | 4922 | goto exit; |
d7718a9d | 4923 | } |
86a761f8 | 4924 | punt: |
f86cd20c | 4925 | if (io_op_defs[req->opcode].file_table) { |
bbad27b2 PB |
4926 | ret = io_grab_files(req); |
4927 | if (ret) | |
4928 | goto err; | |
2b188cc1 | 4929 | } |
bbad27b2 PB |
4930 | |
4931 | /* | |
4932 | * Queued up for async execution, worker will release | |
4933 | * submit reference when the iocb is actually submitted. | |
4934 | */ | |
4935 | io_queue_async_work(req); | |
4bc4494e | 4936 | goto exit; |
2b188cc1 | 4937 | } |
e65ef56d | 4938 | |
fcb323cc | 4939 | err: |
4bc4494e | 4940 | nxt = NULL; |
76a46e06 | 4941 | /* drop submission reference */ |
2a44f467 | 4942 | io_put_req_find_next(req, &nxt); |
e65ef56d | 4943 | |
f9bd67f6 | 4944 | if (linked_timeout) { |
76a46e06 | 4945 | if (!ret) |
f9bd67f6 | 4946 | io_queue_linked_timeout(linked_timeout); |
76a46e06 | 4947 | else |
f9bd67f6 | 4948 | io_put_req(linked_timeout); |
76a46e06 JA |
4949 | } |
4950 | ||
e65ef56d | 4951 | /* and drop final reference, if we failed */ |
9e645e11 | 4952 | if (ret) { |
78e19bbe | 4953 | io_cqring_add_event(req, ret); |
4e88d6e7 | 4954 | req_set_fail_links(req); |
e65ef56d | 4955 | io_put_req(req); |
9e645e11 | 4956 | } |
4a0a7a18 JA |
4957 | if (nxt) { |
4958 | req = nxt; | |
86a761f8 PB |
4959 | |
4960 | if (req->flags & REQ_F_FORCE_ASYNC) | |
4961 | goto punt; | |
4a0a7a18 JA |
4962 | goto again; |
4963 | } | |
4bc4494e | 4964 | exit: |
193155c8 JA |
4965 | if (old_creds) |
4966 | revert_creds(old_creds); | |
2b188cc1 JA |
4967 | } |
4968 | ||
3529d8c2 | 4969 | static void io_queue_sqe(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
4fe2c963 JL |
4970 | { |
4971 | int ret; | |
4972 | ||
3529d8c2 | 4973 | ret = io_req_defer(req, sqe); |
4fe2c963 JL |
4974 | if (ret) { |
4975 | if (ret != -EIOCBQUEUED) { | |
1118591a | 4976 | fail_req: |
78e19bbe | 4977 | io_cqring_add_event(req, ret); |
4e88d6e7 | 4978 | req_set_fail_links(req); |
78e19bbe | 4979 | io_double_put_req(req); |
4fe2c963 | 4980 | } |
2550878f | 4981 | } else if (req->flags & REQ_F_FORCE_ASYNC) { |
1118591a PB |
4982 | ret = io_req_defer_prep(req, sqe); |
4983 | if (unlikely(ret < 0)) | |
4984 | goto fail_req; | |
ce35a47a JA |
4985 | /* |
4986 | * Never try inline submit of IOSQE_ASYNC is set, go straight | |
4987 | * to async execution. | |
4988 | */ | |
4989 | req->work.flags |= IO_WQ_WORK_CONCURRENT; | |
4990 | io_queue_async_work(req); | |
4991 | } else { | |
3529d8c2 | 4992 | __io_queue_sqe(req, sqe); |
ce35a47a | 4993 | } |
4fe2c963 JL |
4994 | } |
4995 | ||
1b4a51b6 | 4996 | static inline void io_queue_link_head(struct io_kiocb *req) |
4fe2c963 | 4997 | { |
94ae5e77 | 4998 | if (unlikely(req->flags & REQ_F_FAIL_LINK)) { |
1b4a51b6 PB |
4999 | io_cqring_add_event(req, -ECANCELED); |
5000 | io_double_put_req(req); | |
5001 | } else | |
3529d8c2 | 5002 | io_queue_sqe(req, NULL); |
4fe2c963 JL |
5003 | } |
5004 | ||
4e88d6e7 | 5005 | #define SQE_VALID_FLAGS (IOSQE_FIXED_FILE|IOSQE_IO_DRAIN|IOSQE_IO_LINK| \ |
ce35a47a | 5006 | IOSQE_IO_HARDLINK | IOSQE_ASYNC) |
9e645e11 | 5007 | |
3529d8c2 JA |
5008 | static bool io_submit_sqe(struct io_kiocb *req, const struct io_uring_sqe *sqe, |
5009 | struct io_submit_state *state, struct io_kiocb **link) | |
9e645e11 | 5010 | { |
a197f664 | 5011 | struct io_ring_ctx *ctx = req->ctx; |
32fe525b | 5012 | unsigned int sqe_flags; |
75c6a039 | 5013 | int ret, id; |
9e645e11 | 5014 | |
32fe525b | 5015 | sqe_flags = READ_ONCE(sqe->flags); |
9e645e11 JA |
5016 | |
5017 | /* enforce forwards compatibility on users */ | |
32fe525b | 5018 | if (unlikely(sqe_flags & ~SQE_VALID_FLAGS)) { |
9e645e11 | 5019 | ret = -EINVAL; |
196be95c | 5020 | goto err_req; |
9e645e11 JA |
5021 | } |
5022 | ||
75c6a039 JA |
5023 | id = READ_ONCE(sqe->personality); |
5024 | if (id) { | |
193155c8 JA |
5025 | req->work.creds = idr_find(&ctx->personality_idr, id); |
5026 | if (unlikely(!req->work.creds)) { | |
75c6a039 JA |
5027 | ret = -EINVAL; |
5028 | goto err_req; | |
5029 | } | |
193155c8 | 5030 | get_cred(req->work.creds); |
75c6a039 JA |
5031 | } |
5032 | ||
6b47ee6e | 5033 | /* same numerical values with corresponding REQ_F_*, safe to copy */ |
8da11c19 PB |
5034 | req->flags |= sqe_flags & (IOSQE_IO_DRAIN | IOSQE_IO_HARDLINK | |
5035 | IOSQE_ASYNC | IOSQE_FIXED_FILE); | |
9e645e11 | 5036 | |
3529d8c2 | 5037 | ret = io_req_set_file(state, req, sqe); |
9e645e11 JA |
5038 | if (unlikely(ret)) { |
5039 | err_req: | |
78e19bbe JA |
5040 | io_cqring_add_event(req, ret); |
5041 | io_double_put_req(req); | |
2e6e1fde | 5042 | return false; |
9e645e11 JA |
5043 | } |
5044 | ||
9e645e11 JA |
5045 | /* |
5046 | * If we already have a head request, queue this one for async | |
5047 | * submittal once the head completes. If we don't have a head but | |
5048 | * IOSQE_IO_LINK is set in the sqe, start a new head. This one will be | |
5049 | * submitted sync once the chain is complete. If none of those | |
5050 | * conditions are true (normal request), then just queue it. | |
5051 | */ | |
5052 | if (*link) { | |
9d76377f | 5053 | struct io_kiocb *head = *link; |
4e88d6e7 | 5054 | |
8cdf2193 PB |
5055 | /* |
5056 | * Taking sequential execution of a link, draining both sides | |
5057 | * of the link also fullfils IOSQE_IO_DRAIN semantics for all | |
5058 | * requests in the link. So, it drains the head and the | |
5059 | * next after the link request. The last one is done via | |
5060 | * drain_next flag to persist the effect across calls. | |
5061 | */ | |
711be031 PB |
5062 | if (sqe_flags & IOSQE_IO_DRAIN) { |
5063 | head->flags |= REQ_F_IO_DRAIN; | |
5064 | ctx->drain_next = 1; | |
5065 | } | |
b7bb4f7d | 5066 | if (io_alloc_async_ctx(req)) { |
9e645e11 JA |
5067 | ret = -EAGAIN; |
5068 | goto err_req; | |
5069 | } | |
5070 | ||
3529d8c2 | 5071 | ret = io_req_defer_prep(req, sqe); |
2d28390a | 5072 | if (ret) { |
4e88d6e7 | 5073 | /* fail even hard links since we don't submit */ |
9d76377f | 5074 | head->flags |= REQ_F_FAIL_LINK; |
f67676d1 | 5075 | goto err_req; |
2d28390a | 5076 | } |
9d76377f PB |
5077 | trace_io_uring_link(ctx, req, head); |
5078 | list_add_tail(&req->link_list, &head->link_list); | |
32fe525b PB |
5079 | |
5080 | /* last request of a link, enqueue the link */ | |
5081 | if (!(sqe_flags & (IOSQE_IO_LINK|IOSQE_IO_HARDLINK))) { | |
5082 | io_queue_link_head(head); | |
5083 | *link = NULL; | |
5084 | } | |
9e645e11 | 5085 | } else { |
711be031 PB |
5086 | if (unlikely(ctx->drain_next)) { |
5087 | req->flags |= REQ_F_IO_DRAIN; | |
5088 | req->ctx->drain_next = 0; | |
5089 | } | |
5090 | if (sqe_flags & (IOSQE_IO_LINK|IOSQE_IO_HARDLINK)) { | |
5091 | req->flags |= REQ_F_LINK; | |
711be031 PB |
5092 | INIT_LIST_HEAD(&req->link_list); |
5093 | ret = io_req_defer_prep(req, sqe); | |
5094 | if (ret) | |
5095 | req->flags |= REQ_F_FAIL_LINK; | |
5096 | *link = req; | |
5097 | } else { | |
5098 | io_queue_sqe(req, sqe); | |
5099 | } | |
9e645e11 | 5100 | } |
2e6e1fde PB |
5101 | |
5102 | return true; | |
9e645e11 JA |
5103 | } |
5104 | ||
9a56a232 JA |
5105 | /* |
5106 | * Batched submission is done, ensure local IO is flushed out. | |
5107 | */ | |
5108 | static void io_submit_state_end(struct io_submit_state *state) | |
5109 | { | |
5110 | blk_finish_plug(&state->plug); | |
3d6770fb | 5111 | io_file_put(state); |
2579f913 | 5112 | if (state->free_reqs) |
6c8a3134 | 5113 | kmem_cache_free_bulk(req_cachep, state->free_reqs, state->reqs); |
9a56a232 JA |
5114 | } |
5115 | ||
5116 | /* | |
5117 | * Start submission side cache. | |
5118 | */ | |
5119 | static void io_submit_state_start(struct io_submit_state *state, | |
22efde59 | 5120 | unsigned int max_ios) |
9a56a232 JA |
5121 | { |
5122 | blk_start_plug(&state->plug); | |
2579f913 | 5123 | state->free_reqs = 0; |
9a56a232 JA |
5124 | state->file = NULL; |
5125 | state->ios_left = max_ios; | |
5126 | } | |
5127 | ||
2b188cc1 JA |
5128 | static void io_commit_sqring(struct io_ring_ctx *ctx) |
5129 | { | |
75b28aff | 5130 | struct io_rings *rings = ctx->rings; |
2b188cc1 | 5131 | |
caf582c6 PB |
5132 | /* |
5133 | * Ensure any loads from the SQEs are done at this point, | |
5134 | * since once we write the new head, the application could | |
5135 | * write new data to them. | |
5136 | */ | |
5137 | smp_store_release(&rings->sq.head, ctx->cached_sq_head); | |
2b188cc1 JA |
5138 | } |
5139 | ||
2b188cc1 | 5140 | /* |
3529d8c2 | 5141 | * Fetch an sqe, if one is available. Note that sqe_ptr will point to memory |
2b188cc1 JA |
5142 | * that is mapped by userspace. This means that care needs to be taken to |
5143 | * ensure that reads are stable, as we cannot rely on userspace always | |
5144 | * being a good citizen. If members of the sqe are validated and then later | |
5145 | * used, it's important that those reads are done through READ_ONCE() to | |
5146 | * prevent a re-load down the line. | |
5147 | */ | |
3529d8c2 JA |
5148 | static bool io_get_sqring(struct io_ring_ctx *ctx, struct io_kiocb *req, |
5149 | const struct io_uring_sqe **sqe_ptr) | |
2b188cc1 | 5150 | { |
75b28aff | 5151 | u32 *sq_array = ctx->sq_array; |
2b188cc1 JA |
5152 | unsigned head; |
5153 | ||
5154 | /* | |
5155 | * The cached sq head (or cq tail) serves two purposes: | |
5156 | * | |
5157 | * 1) allows us to batch the cost of updating the user visible | |
5158 | * head updates. | |
5159 | * 2) allows the kernel side to track the head on its own, even | |
5160 | * though the application is the one updating it. | |
5161 | */ | |
ee7d46d9 | 5162 | head = READ_ONCE(sq_array[ctx->cached_sq_head & ctx->sq_mask]); |
9835d6fa | 5163 | if (likely(head < ctx->sq_entries)) { |
cf6fd4bd PB |
5164 | /* |
5165 | * All io need record the previous position, if LINK vs DARIN, | |
5166 | * it can be used to mark the position of the first IO in the | |
5167 | * link list. | |
5168 | */ | |
5169 | req->sequence = ctx->cached_sq_head; | |
3529d8c2 JA |
5170 | *sqe_ptr = &ctx->sq_sqes[head]; |
5171 | req->opcode = READ_ONCE((*sqe_ptr)->opcode); | |
5172 | req->user_data = READ_ONCE((*sqe_ptr)->user_data); | |
2b188cc1 JA |
5173 | ctx->cached_sq_head++; |
5174 | return true; | |
5175 | } | |
5176 | ||
5177 | /* drop invalid entries */ | |
5178 | ctx->cached_sq_head++; | |
498ccd9e | 5179 | ctx->cached_sq_dropped++; |
ee7d46d9 | 5180 | WRITE_ONCE(ctx->rings->sq_dropped, ctx->cached_sq_dropped); |
2b188cc1 JA |
5181 | return false; |
5182 | } | |
5183 | ||
fb5ccc98 | 5184 | static int io_submit_sqes(struct io_ring_ctx *ctx, unsigned int nr, |
ae9428ca PB |
5185 | struct file *ring_file, int ring_fd, |
5186 | struct mm_struct **mm, bool async) | |
6c271ce2 JA |
5187 | { |
5188 | struct io_submit_state state, *statep = NULL; | |
9e645e11 | 5189 | struct io_kiocb *link = NULL; |
9e645e11 | 5190 | int i, submitted = 0; |
95a1b3ff | 5191 | bool mm_fault = false; |
6c271ce2 | 5192 | |
c4a2ed72 | 5193 | /* if we have a backlog and couldn't flush it all, return BUSY */ |
ad3eb2c8 JA |
5194 | if (test_bit(0, &ctx->sq_check_overflow)) { |
5195 | if (!list_empty(&ctx->cq_overflow_list) && | |
5196 | !io_cqring_overflow_flush(ctx, false)) | |
5197 | return -EBUSY; | |
5198 | } | |
6c271ce2 | 5199 | |
ee7d46d9 PB |
5200 | /* make sure SQ entry isn't read before tail */ |
5201 | nr = min3(nr, ctx->sq_entries, io_sqring_entries(ctx)); | |
9ef4f124 | 5202 | |
2b85edfc PB |
5203 | if (!percpu_ref_tryget_many(&ctx->refs, nr)) |
5204 | return -EAGAIN; | |
6c271ce2 JA |
5205 | |
5206 | if (nr > IO_PLUG_THRESHOLD) { | |
22efde59 | 5207 | io_submit_state_start(&state, nr); |
6c271ce2 JA |
5208 | statep = &state; |
5209 | } | |
5210 | ||
b14cca0c PB |
5211 | ctx->ring_fd = ring_fd; |
5212 | ctx->ring_file = ring_file; | |
5213 | ||
6c271ce2 | 5214 | for (i = 0; i < nr; i++) { |
3529d8c2 | 5215 | const struct io_uring_sqe *sqe; |
196be95c | 5216 | struct io_kiocb *req; |
1cb1edb2 | 5217 | int err; |
fb5ccc98 | 5218 | |
196be95c PB |
5219 | req = io_get_req(ctx, statep); |
5220 | if (unlikely(!req)) { | |
5221 | if (!submitted) | |
5222 | submitted = -EAGAIN; | |
fb5ccc98 | 5223 | break; |
196be95c | 5224 | } |
3529d8c2 | 5225 | if (!io_get_sqring(ctx, req, &sqe)) { |
2b85edfc | 5226 | __io_req_do_free(req); |
196be95c PB |
5227 | break; |
5228 | } | |
fb5ccc98 | 5229 | |
d3656344 JA |
5230 | /* will complete beyond this point, count as submitted */ |
5231 | submitted++; | |
5232 | ||
5233 | if (unlikely(req->opcode >= IORING_OP_LAST)) { | |
1cb1edb2 PB |
5234 | err = -EINVAL; |
5235 | fail_req: | |
5236 | io_cqring_add_event(req, err); | |
d3656344 | 5237 | io_double_put_req(req); |
196be95c PB |
5238 | break; |
5239 | } | |
fb5ccc98 | 5240 | |
d3656344 | 5241 | if (io_op_defs[req->opcode].needs_mm && !*mm) { |
95a1b3ff | 5242 | mm_fault = mm_fault || !mmget_not_zero(ctx->sqo_mm); |
1cb1edb2 PB |
5243 | if (unlikely(mm_fault)) { |
5244 | err = -EFAULT; | |
5245 | goto fail_req; | |
95a1b3ff | 5246 | } |
1cb1edb2 PB |
5247 | use_mm(ctx->sqo_mm); |
5248 | *mm = ctx->sqo_mm; | |
9e645e11 | 5249 | } |
9e645e11 | 5250 | |
cf6fd4bd | 5251 | req->needs_fixed_file = async; |
354420f7 JA |
5252 | trace_io_uring_submit_sqe(ctx, req->opcode, req->user_data, |
5253 | true, async); | |
3529d8c2 | 5254 | if (!io_submit_sqe(req, sqe, statep, &link)) |
2e6e1fde | 5255 | break; |
6c271ce2 JA |
5256 | } |
5257 | ||
9466f437 PB |
5258 | if (unlikely(submitted != nr)) { |
5259 | int ref_used = (submitted == -EAGAIN) ? 0 : submitted; | |
5260 | ||
5261 | percpu_ref_put_many(&ctx->refs, nr - ref_used); | |
5262 | } | |
9e645e11 | 5263 | if (link) |
1b4a51b6 | 5264 | io_queue_link_head(link); |
6c271ce2 JA |
5265 | if (statep) |
5266 | io_submit_state_end(&state); | |
5267 | ||
ae9428ca PB |
5268 | /* Commit SQ ring head once we've consumed and submitted all SQEs */ |
5269 | io_commit_sqring(ctx); | |
5270 | ||
6c271ce2 JA |
5271 | return submitted; |
5272 | } | |
5273 | ||
5274 | static int io_sq_thread(void *data) | |
5275 | { | |
6c271ce2 JA |
5276 | struct io_ring_ctx *ctx = data; |
5277 | struct mm_struct *cur_mm = NULL; | |
181e448d | 5278 | const struct cred *old_cred; |
6c271ce2 JA |
5279 | mm_segment_t old_fs; |
5280 | DEFINE_WAIT(wait); | |
6c271ce2 | 5281 | unsigned long timeout; |
bdcd3eab | 5282 | int ret = 0; |
6c271ce2 | 5283 | |
206aefde | 5284 | complete(&ctx->completions[1]); |
a4c0b3de | 5285 | |
6c271ce2 JA |
5286 | old_fs = get_fs(); |
5287 | set_fs(USER_DS); | |
181e448d | 5288 | old_cred = override_creds(ctx->creds); |
6c271ce2 | 5289 | |
bdcd3eab | 5290 | timeout = jiffies + ctx->sq_thread_idle; |
2bbcd6d3 | 5291 | while (!kthread_should_park()) { |
fb5ccc98 | 5292 | unsigned int to_submit; |
6c271ce2 | 5293 | |
bdcd3eab | 5294 | if (!list_empty(&ctx->poll_list)) { |
6c271ce2 JA |
5295 | unsigned nr_events = 0; |
5296 | ||
bdcd3eab XW |
5297 | mutex_lock(&ctx->uring_lock); |
5298 | if (!list_empty(&ctx->poll_list)) | |
5299 | io_iopoll_getevents(ctx, &nr_events, 0); | |
5300 | else | |
6c271ce2 | 5301 | timeout = jiffies + ctx->sq_thread_idle; |
bdcd3eab | 5302 | mutex_unlock(&ctx->uring_lock); |
6c271ce2 JA |
5303 | } |
5304 | ||
fb5ccc98 | 5305 | to_submit = io_sqring_entries(ctx); |
c1edbf5f JA |
5306 | |
5307 | /* | |
5308 | * If submit got -EBUSY, flag us as needing the application | |
5309 | * to enter the kernel to reap and flush events. | |
5310 | */ | |
5311 | if (!to_submit || ret == -EBUSY) { | |
7143b5ac SG |
5312 | /* |
5313 | * Drop cur_mm before scheduling, we can't hold it for | |
5314 | * long periods (or over schedule()). Do this before | |
5315 | * adding ourselves to the waitqueue, as the unuse/drop | |
5316 | * may sleep. | |
5317 | */ | |
5318 | if (cur_mm) { | |
5319 | unuse_mm(cur_mm); | |
5320 | mmput(cur_mm); | |
5321 | cur_mm = NULL; | |
5322 | } | |
5323 | ||
6c271ce2 JA |
5324 | /* |
5325 | * We're polling. If we're within the defined idle | |
5326 | * period, then let us spin without work before going | |
c1edbf5f JA |
5327 | * to sleep. The exception is if we got EBUSY doing |
5328 | * more IO, we should wait for the application to | |
5329 | * reap events and wake us up. | |
6c271ce2 | 5330 | */ |
bdcd3eab | 5331 | if (!list_empty(&ctx->poll_list) || |
df069d80 JA |
5332 | (!time_after(jiffies, timeout) && ret != -EBUSY && |
5333 | !percpu_ref_is_dying(&ctx->refs))) { | |
b41e9852 JA |
5334 | if (current->task_works) |
5335 | task_work_run(); | |
9831a90c | 5336 | cond_resched(); |
6c271ce2 JA |
5337 | continue; |
5338 | } | |
5339 | ||
6c271ce2 JA |
5340 | prepare_to_wait(&ctx->sqo_wait, &wait, |
5341 | TASK_INTERRUPTIBLE); | |
5342 | ||
bdcd3eab XW |
5343 | /* |
5344 | * While doing polled IO, before going to sleep, we need | |
5345 | * to check if there are new reqs added to poll_list, it | |
5346 | * is because reqs may have been punted to io worker and | |
5347 | * will be added to poll_list later, hence check the | |
5348 | * poll_list again. | |
5349 | */ | |
5350 | if ((ctx->flags & IORING_SETUP_IOPOLL) && | |
5351 | !list_empty_careful(&ctx->poll_list)) { | |
5352 | finish_wait(&ctx->sqo_wait, &wait); | |
5353 | continue; | |
5354 | } | |
5355 | ||
6c271ce2 | 5356 | /* Tell userspace we may need a wakeup call */ |
75b28aff | 5357 | ctx->rings->sq_flags |= IORING_SQ_NEED_WAKEUP; |
0d7bae69 SB |
5358 | /* make sure to read SQ tail after writing flags */ |
5359 | smp_mb(); | |
6c271ce2 | 5360 | |
fb5ccc98 | 5361 | to_submit = io_sqring_entries(ctx); |
c1edbf5f | 5362 | if (!to_submit || ret == -EBUSY) { |
2bbcd6d3 | 5363 | if (kthread_should_park()) { |
6c271ce2 JA |
5364 | finish_wait(&ctx->sqo_wait, &wait); |
5365 | break; | |
5366 | } | |
b41e9852 JA |
5367 | if (current->task_works) { |
5368 | task_work_run(); | |
5369 | continue; | |
5370 | } | |
6c271ce2 JA |
5371 | if (signal_pending(current)) |
5372 | flush_signals(current); | |
5373 | schedule(); | |
5374 | finish_wait(&ctx->sqo_wait, &wait); | |
5375 | ||
75b28aff | 5376 | ctx->rings->sq_flags &= ~IORING_SQ_NEED_WAKEUP; |
6c271ce2 JA |
5377 | continue; |
5378 | } | |
5379 | finish_wait(&ctx->sqo_wait, &wait); | |
5380 | ||
75b28aff | 5381 | ctx->rings->sq_flags &= ~IORING_SQ_NEED_WAKEUP; |
6c271ce2 JA |
5382 | } |
5383 | ||
8a4955ff | 5384 | mutex_lock(&ctx->uring_lock); |
1d7bb1d5 | 5385 | ret = io_submit_sqes(ctx, to_submit, NULL, -1, &cur_mm, true); |
8a4955ff | 5386 | mutex_unlock(&ctx->uring_lock); |
bdcd3eab | 5387 | timeout = jiffies + ctx->sq_thread_idle; |
6c271ce2 JA |
5388 | } |
5389 | ||
b41e9852 JA |
5390 | if (current->task_works) |
5391 | task_work_run(); | |
5392 | ||
6c271ce2 JA |
5393 | set_fs(old_fs); |
5394 | if (cur_mm) { | |
5395 | unuse_mm(cur_mm); | |
5396 | mmput(cur_mm); | |
5397 | } | |
181e448d | 5398 | revert_creds(old_cred); |
06058632 | 5399 | |
2bbcd6d3 | 5400 | kthread_parkme(); |
06058632 | 5401 | |
6c271ce2 JA |
5402 | return 0; |
5403 | } | |
5404 | ||
bda52162 JA |
5405 | struct io_wait_queue { |
5406 | struct wait_queue_entry wq; | |
5407 | struct io_ring_ctx *ctx; | |
5408 | unsigned to_wait; | |
5409 | unsigned nr_timeouts; | |
5410 | }; | |
5411 | ||
1d7bb1d5 | 5412 | static inline bool io_should_wake(struct io_wait_queue *iowq, bool noflush) |
bda52162 JA |
5413 | { |
5414 | struct io_ring_ctx *ctx = iowq->ctx; | |
5415 | ||
5416 | /* | |
d195a66e | 5417 | * Wake up if we have enough events, or if a timeout occurred since we |
bda52162 JA |
5418 | * started waiting. For timeouts, we always want to return to userspace, |
5419 | * regardless of event count. | |
5420 | */ | |
1d7bb1d5 | 5421 | return io_cqring_events(ctx, noflush) >= iowq->to_wait || |
bda52162 JA |
5422 | atomic_read(&ctx->cq_timeouts) != iowq->nr_timeouts; |
5423 | } | |
5424 | ||
5425 | static int io_wake_function(struct wait_queue_entry *curr, unsigned int mode, | |
5426 | int wake_flags, void *key) | |
5427 | { | |
5428 | struct io_wait_queue *iowq = container_of(curr, struct io_wait_queue, | |
5429 | wq); | |
5430 | ||
1d7bb1d5 JA |
5431 | /* use noflush == true, as we can't safely rely on locking context */ |
5432 | if (!io_should_wake(iowq, true)) | |
bda52162 JA |
5433 | return -1; |
5434 | ||
5435 | return autoremove_wake_function(curr, mode, wake_flags, key); | |
5436 | } | |
5437 | ||
2b188cc1 JA |
5438 | /* |
5439 | * Wait until events become available, if we don't already have some. The | |
5440 | * application must reap them itself, as they reside on the shared cq ring. | |
5441 | */ | |
5442 | static int io_cqring_wait(struct io_ring_ctx *ctx, int min_events, | |
5443 | const sigset_t __user *sig, size_t sigsz) | |
5444 | { | |
bda52162 JA |
5445 | struct io_wait_queue iowq = { |
5446 | .wq = { | |
5447 | .private = current, | |
5448 | .func = io_wake_function, | |
5449 | .entry = LIST_HEAD_INIT(iowq.wq.entry), | |
5450 | }, | |
5451 | .ctx = ctx, | |
5452 | .to_wait = min_events, | |
5453 | }; | |
75b28aff | 5454 | struct io_rings *rings = ctx->rings; |
e9ffa5c2 | 5455 | int ret = 0; |
2b188cc1 | 5456 | |
b41e9852 JA |
5457 | do { |
5458 | if (io_cqring_events(ctx, false) >= min_events) | |
5459 | return 0; | |
5460 | if (!current->task_works) | |
5461 | break; | |
5462 | task_work_run(); | |
5463 | } while (1); | |
2b188cc1 JA |
5464 | |
5465 | if (sig) { | |
9e75ad5d AB |
5466 | #ifdef CONFIG_COMPAT |
5467 | if (in_compat_syscall()) | |
5468 | ret = set_compat_user_sigmask((const compat_sigset_t __user *)sig, | |
b772434b | 5469 | sigsz); |
9e75ad5d AB |
5470 | else |
5471 | #endif | |
b772434b | 5472 | ret = set_user_sigmask(sig, sigsz); |
9e75ad5d | 5473 | |
2b188cc1 JA |
5474 | if (ret) |
5475 | return ret; | |
5476 | } | |
5477 | ||
bda52162 | 5478 | iowq.nr_timeouts = atomic_read(&ctx->cq_timeouts); |
c826bd7a | 5479 | trace_io_uring_cqring_wait(ctx, min_events); |
bda52162 JA |
5480 | do { |
5481 | prepare_to_wait_exclusive(&ctx->wait, &iowq.wq, | |
5482 | TASK_INTERRUPTIBLE); | |
b41e9852 JA |
5483 | if (current->task_works) |
5484 | task_work_run(); | |
1d7bb1d5 | 5485 | if (io_should_wake(&iowq, false)) |
bda52162 JA |
5486 | break; |
5487 | schedule(); | |
5488 | if (signal_pending(current)) { | |
e9ffa5c2 | 5489 | ret = -EINTR; |
bda52162 JA |
5490 | break; |
5491 | } | |
5492 | } while (1); | |
5493 | finish_wait(&ctx->wait, &iowq.wq); | |
5494 | ||
e9ffa5c2 | 5495 | restore_saved_sigmask_unless(ret == -EINTR); |
2b188cc1 | 5496 | |
75b28aff | 5497 | return READ_ONCE(rings->cq.head) == READ_ONCE(rings->cq.tail) ? ret : 0; |
2b188cc1 JA |
5498 | } |
5499 | ||
6b06314c JA |
5500 | static void __io_sqe_files_unregister(struct io_ring_ctx *ctx) |
5501 | { | |
5502 | #if defined(CONFIG_UNIX) | |
5503 | if (ctx->ring_sock) { | |
5504 | struct sock *sock = ctx->ring_sock->sk; | |
5505 | struct sk_buff *skb; | |
5506 | ||
5507 | while ((skb = skb_dequeue(&sock->sk_receive_queue)) != NULL) | |
5508 | kfree_skb(skb); | |
5509 | } | |
5510 | #else | |
5511 | int i; | |
5512 | ||
65e19f54 JA |
5513 | for (i = 0; i < ctx->nr_user_files; i++) { |
5514 | struct file *file; | |
5515 | ||
5516 | file = io_file_from_index(ctx, i); | |
5517 | if (file) | |
5518 | fput(file); | |
5519 | } | |
6b06314c JA |
5520 | #endif |
5521 | } | |
5522 | ||
05f3fb3c JA |
5523 | static void io_file_ref_kill(struct percpu_ref *ref) |
5524 | { | |
5525 | struct fixed_file_data *data; | |
5526 | ||
5527 | data = container_of(ref, struct fixed_file_data, refs); | |
5528 | complete(&data->done); | |
5529 | } | |
5530 | ||
6b06314c JA |
5531 | static int io_sqe_files_unregister(struct io_ring_ctx *ctx) |
5532 | { | |
05f3fb3c | 5533 | struct fixed_file_data *data = ctx->file_data; |
65e19f54 JA |
5534 | unsigned nr_tables, i; |
5535 | ||
05f3fb3c | 5536 | if (!data) |
6b06314c JA |
5537 | return -ENXIO; |
5538 | ||
05f3fb3c | 5539 | percpu_ref_kill_and_confirm(&data->refs, io_file_ref_kill); |
e46a7950 | 5540 | flush_work(&data->ref_work); |
2faf852d JA |
5541 | wait_for_completion(&data->done); |
5542 | io_ring_file_ref_flush(data); | |
05f3fb3c JA |
5543 | percpu_ref_exit(&data->refs); |
5544 | ||
6b06314c | 5545 | __io_sqe_files_unregister(ctx); |
65e19f54 JA |
5546 | nr_tables = DIV_ROUND_UP(ctx->nr_user_files, IORING_MAX_FILES_TABLE); |
5547 | for (i = 0; i < nr_tables; i++) | |
05f3fb3c JA |
5548 | kfree(data->table[i].files); |
5549 | kfree(data->table); | |
5550 | kfree(data); | |
5551 | ctx->file_data = NULL; | |
6b06314c JA |
5552 | ctx->nr_user_files = 0; |
5553 | return 0; | |
5554 | } | |
5555 | ||
6c271ce2 JA |
5556 | static void io_sq_thread_stop(struct io_ring_ctx *ctx) |
5557 | { | |
5558 | if (ctx->sqo_thread) { | |
206aefde | 5559 | wait_for_completion(&ctx->completions[1]); |
2bbcd6d3 RP |
5560 | /* |
5561 | * The park is a bit of a work-around, without it we get | |
5562 | * warning spews on shutdown with SQPOLL set and affinity | |
5563 | * set to a single CPU. | |
5564 | */ | |
06058632 | 5565 | kthread_park(ctx->sqo_thread); |
6c271ce2 JA |
5566 | kthread_stop(ctx->sqo_thread); |
5567 | ctx->sqo_thread = NULL; | |
5568 | } | |
5569 | } | |
5570 | ||
6b06314c JA |
5571 | static void io_finish_async(struct io_ring_ctx *ctx) |
5572 | { | |
6c271ce2 JA |
5573 | io_sq_thread_stop(ctx); |
5574 | ||
561fb04a JA |
5575 | if (ctx->io_wq) { |
5576 | io_wq_destroy(ctx->io_wq); | |
5577 | ctx->io_wq = NULL; | |
6b06314c JA |
5578 | } |
5579 | } | |
5580 | ||
5581 | #if defined(CONFIG_UNIX) | |
6b06314c JA |
5582 | /* |
5583 | * Ensure the UNIX gc is aware of our file set, so we are certain that | |
5584 | * the io_uring can be safely unregistered on process exit, even if we have | |
5585 | * loops in the file referencing. | |
5586 | */ | |
5587 | static int __io_sqe_files_scm(struct io_ring_ctx *ctx, int nr, int offset) | |
5588 | { | |
5589 | struct sock *sk = ctx->ring_sock->sk; | |
5590 | struct scm_fp_list *fpl; | |
5591 | struct sk_buff *skb; | |
08a45173 | 5592 | int i, nr_files; |
6b06314c JA |
5593 | |
5594 | if (!capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN)) { | |
5595 | unsigned long inflight = ctx->user->unix_inflight + nr; | |
5596 | ||
5597 | if (inflight > task_rlimit(current, RLIMIT_NOFILE)) | |
5598 | return -EMFILE; | |
5599 | } | |
5600 | ||
5601 | fpl = kzalloc(sizeof(*fpl), GFP_KERNEL); | |
5602 | if (!fpl) | |
5603 | return -ENOMEM; | |
5604 | ||
5605 | skb = alloc_skb(0, GFP_KERNEL); | |
5606 | if (!skb) { | |
5607 | kfree(fpl); | |
5608 | return -ENOMEM; | |
5609 | } | |
5610 | ||
5611 | skb->sk = sk; | |
6b06314c | 5612 | |
08a45173 | 5613 | nr_files = 0; |
6b06314c JA |
5614 | fpl->user = get_uid(ctx->user); |
5615 | for (i = 0; i < nr; i++) { | |
65e19f54 JA |
5616 | struct file *file = io_file_from_index(ctx, i + offset); |
5617 | ||
5618 | if (!file) | |
08a45173 | 5619 | continue; |
65e19f54 | 5620 | fpl->fp[nr_files] = get_file(file); |
08a45173 JA |
5621 | unix_inflight(fpl->user, fpl->fp[nr_files]); |
5622 | nr_files++; | |
6b06314c JA |
5623 | } |
5624 | ||
08a45173 JA |
5625 | if (nr_files) { |
5626 | fpl->max = SCM_MAX_FD; | |
5627 | fpl->count = nr_files; | |
5628 | UNIXCB(skb).fp = fpl; | |
05f3fb3c | 5629 | skb->destructor = unix_destruct_scm; |
08a45173 JA |
5630 | refcount_add(skb->truesize, &sk->sk_wmem_alloc); |
5631 | skb_queue_head(&sk->sk_receive_queue, skb); | |
6b06314c | 5632 | |
08a45173 JA |
5633 | for (i = 0; i < nr_files; i++) |
5634 | fput(fpl->fp[i]); | |
5635 | } else { | |
5636 | kfree_skb(skb); | |
5637 | kfree(fpl); | |
5638 | } | |
6b06314c JA |
5639 | |
5640 | return 0; | |
5641 | } | |
5642 | ||
5643 | /* | |
5644 | * If UNIX sockets are enabled, fd passing can cause a reference cycle which | |
5645 | * causes regular reference counting to break down. We rely on the UNIX | |
5646 | * garbage collection to take care of this problem for us. | |
5647 | */ | |
5648 | static int io_sqe_files_scm(struct io_ring_ctx *ctx) | |
5649 | { | |
5650 | unsigned left, total; | |
5651 | int ret = 0; | |
5652 | ||
5653 | total = 0; | |
5654 | left = ctx->nr_user_files; | |
5655 | while (left) { | |
5656 | unsigned this_files = min_t(unsigned, left, SCM_MAX_FD); | |
6b06314c JA |
5657 | |
5658 | ret = __io_sqe_files_scm(ctx, this_files, total); | |
5659 | if (ret) | |
5660 | break; | |
5661 | left -= this_files; | |
5662 | total += this_files; | |
5663 | } | |
5664 | ||
5665 | if (!ret) | |
5666 | return 0; | |
5667 | ||
5668 | while (total < ctx->nr_user_files) { | |
65e19f54 JA |
5669 | struct file *file = io_file_from_index(ctx, total); |
5670 | ||
5671 | if (file) | |
5672 | fput(file); | |
6b06314c JA |
5673 | total++; |
5674 | } | |
5675 | ||
5676 | return ret; | |
5677 | } | |
5678 | #else | |
5679 | static int io_sqe_files_scm(struct io_ring_ctx *ctx) | |
5680 | { | |
5681 | return 0; | |
5682 | } | |
5683 | #endif | |
5684 | ||
65e19f54 JA |
5685 | static int io_sqe_alloc_file_tables(struct io_ring_ctx *ctx, unsigned nr_tables, |
5686 | unsigned nr_files) | |
5687 | { | |
5688 | int i; | |
5689 | ||
5690 | for (i = 0; i < nr_tables; i++) { | |
05f3fb3c | 5691 | struct fixed_file_table *table = &ctx->file_data->table[i]; |
65e19f54 JA |
5692 | unsigned this_files; |
5693 | ||
5694 | this_files = min(nr_files, IORING_MAX_FILES_TABLE); | |
5695 | table->files = kcalloc(this_files, sizeof(struct file *), | |
5696 | GFP_KERNEL); | |
5697 | if (!table->files) | |
5698 | break; | |
5699 | nr_files -= this_files; | |
5700 | } | |
5701 | ||
5702 | if (i == nr_tables) | |
5703 | return 0; | |
5704 | ||
5705 | for (i = 0; i < nr_tables; i++) { | |
05f3fb3c | 5706 | struct fixed_file_table *table = &ctx->file_data->table[i]; |
65e19f54 JA |
5707 | kfree(table->files); |
5708 | } | |
5709 | return 1; | |
5710 | } | |
5711 | ||
05f3fb3c JA |
5712 | static void io_ring_file_put(struct io_ring_ctx *ctx, struct file *file) |
5713 | { | |
5714 | #if defined(CONFIG_UNIX) | |
5715 | struct sock *sock = ctx->ring_sock->sk; | |
5716 | struct sk_buff_head list, *head = &sock->sk_receive_queue; | |
5717 | struct sk_buff *skb; | |
5718 | int i; | |
5719 | ||
5720 | __skb_queue_head_init(&list); | |
5721 | ||
5722 | /* | |
5723 | * Find the skb that holds this file in its SCM_RIGHTS. When found, | |
5724 | * remove this entry and rearrange the file array. | |
5725 | */ | |
5726 | skb = skb_dequeue(head); | |
5727 | while (skb) { | |
5728 | struct scm_fp_list *fp; | |
5729 | ||
5730 | fp = UNIXCB(skb).fp; | |
5731 | for (i = 0; i < fp->count; i++) { | |
5732 | int left; | |
5733 | ||
5734 | if (fp->fp[i] != file) | |
5735 | continue; | |
5736 | ||
5737 | unix_notinflight(fp->user, fp->fp[i]); | |
5738 | left = fp->count - 1 - i; | |
5739 | if (left) { | |
5740 | memmove(&fp->fp[i], &fp->fp[i + 1], | |
5741 | left * sizeof(struct file *)); | |
5742 | } | |
5743 | fp->count--; | |
5744 | if (!fp->count) { | |
5745 | kfree_skb(skb); | |
5746 | skb = NULL; | |
5747 | } else { | |
5748 | __skb_queue_tail(&list, skb); | |
5749 | } | |
5750 | fput(file); | |
5751 | file = NULL; | |
5752 | break; | |
5753 | } | |
5754 | ||
5755 | if (!file) | |
5756 | break; | |
5757 | ||
5758 | __skb_queue_tail(&list, skb); | |
5759 | ||
5760 | skb = skb_dequeue(head); | |
5761 | } | |
5762 | ||
5763 | if (skb_peek(&list)) { | |
5764 | spin_lock_irq(&head->lock); | |
5765 | while ((skb = __skb_dequeue(&list)) != NULL) | |
5766 | __skb_queue_tail(head, skb); | |
5767 | spin_unlock_irq(&head->lock); | |
5768 | } | |
5769 | #else | |
5770 | fput(file); | |
5771 | #endif | |
5772 | } | |
5773 | ||
5774 | struct io_file_put { | |
5775 | struct llist_node llist; | |
5776 | struct file *file; | |
5777 | struct completion *done; | |
5778 | }; | |
5779 | ||
2faf852d | 5780 | static void io_ring_file_ref_flush(struct fixed_file_data *data) |
65e19f54 | 5781 | { |
05f3fb3c | 5782 | struct io_file_put *pfile, *tmp; |
05f3fb3c | 5783 | struct llist_node *node; |
65e19f54 | 5784 | |
05f3fb3c JA |
5785 | while ((node = llist_del_all(&data->put_llist)) != NULL) { |
5786 | llist_for_each_entry_safe(pfile, tmp, node, llist) { | |
5787 | io_ring_file_put(data->ctx, pfile->file); | |
5788 | if (pfile->done) | |
5789 | complete(pfile->done); | |
5790 | else | |
5791 | kfree(pfile); | |
5792 | } | |
65e19f54 | 5793 | } |
2faf852d | 5794 | } |
65e19f54 | 5795 | |
2faf852d JA |
5796 | static void io_ring_file_ref_switch(struct work_struct *work) |
5797 | { | |
5798 | struct fixed_file_data *data; | |
65e19f54 | 5799 | |
2faf852d JA |
5800 | data = container_of(work, struct fixed_file_data, ref_work); |
5801 | io_ring_file_ref_flush(data); | |
05f3fb3c JA |
5802 | percpu_ref_switch_to_percpu(&data->refs); |
5803 | } | |
65e19f54 | 5804 | |
05f3fb3c JA |
5805 | static void io_file_data_ref_zero(struct percpu_ref *ref) |
5806 | { | |
5807 | struct fixed_file_data *data; | |
5808 | ||
5809 | data = container_of(ref, struct fixed_file_data, refs); | |
5810 | ||
2faf852d JA |
5811 | /* |
5812 | * We can't safely switch from inside this context, punt to wq. If | |
5813 | * the table ref is going away, the table is being unregistered. | |
5814 | * Don't queue up the async work for that case, the caller will | |
5815 | * handle it. | |
5816 | */ | |
5817 | if (!percpu_ref_is_dying(&data->refs)) | |
5818 | queue_work(system_wq, &data->ref_work); | |
65e19f54 JA |
5819 | } |
5820 | ||
6b06314c JA |
5821 | static int io_sqe_files_register(struct io_ring_ctx *ctx, void __user *arg, |
5822 | unsigned nr_args) | |
5823 | { | |
5824 | __s32 __user *fds = (__s32 __user *) arg; | |
65e19f54 | 5825 | unsigned nr_tables; |
05f3fb3c | 5826 | struct file *file; |
6b06314c JA |
5827 | int fd, ret = 0; |
5828 | unsigned i; | |
5829 | ||
05f3fb3c | 5830 | if (ctx->file_data) |
6b06314c JA |
5831 | return -EBUSY; |
5832 | if (!nr_args) | |
5833 | return -EINVAL; | |
5834 | if (nr_args > IORING_MAX_FIXED_FILES) | |
5835 | return -EMFILE; | |
5836 | ||
05f3fb3c JA |
5837 | ctx->file_data = kzalloc(sizeof(*ctx->file_data), GFP_KERNEL); |
5838 | if (!ctx->file_data) | |
5839 | return -ENOMEM; | |
5840 | ctx->file_data->ctx = ctx; | |
5841 | init_completion(&ctx->file_data->done); | |
5842 | ||
65e19f54 | 5843 | nr_tables = DIV_ROUND_UP(nr_args, IORING_MAX_FILES_TABLE); |
05f3fb3c JA |
5844 | ctx->file_data->table = kcalloc(nr_tables, |
5845 | sizeof(struct fixed_file_table), | |
65e19f54 | 5846 | GFP_KERNEL); |
05f3fb3c JA |
5847 | if (!ctx->file_data->table) { |
5848 | kfree(ctx->file_data); | |
5849 | ctx->file_data = NULL; | |
6b06314c | 5850 | return -ENOMEM; |
05f3fb3c JA |
5851 | } |
5852 | ||
5853 | if (percpu_ref_init(&ctx->file_data->refs, io_file_data_ref_zero, | |
5854 | PERCPU_REF_ALLOW_REINIT, GFP_KERNEL)) { | |
5855 | kfree(ctx->file_data->table); | |
5856 | kfree(ctx->file_data); | |
5857 | ctx->file_data = NULL; | |
6b06314c | 5858 | return -ENOMEM; |
05f3fb3c JA |
5859 | } |
5860 | ctx->file_data->put_llist.first = NULL; | |
5861 | INIT_WORK(&ctx->file_data->ref_work, io_ring_file_ref_switch); | |
6b06314c | 5862 | |
65e19f54 | 5863 | if (io_sqe_alloc_file_tables(ctx, nr_tables, nr_args)) { |
05f3fb3c JA |
5864 | percpu_ref_exit(&ctx->file_data->refs); |
5865 | kfree(ctx->file_data->table); | |
5866 | kfree(ctx->file_data); | |
5867 | ctx->file_data = NULL; | |
65e19f54 JA |
5868 | return -ENOMEM; |
5869 | } | |
5870 | ||
08a45173 | 5871 | for (i = 0; i < nr_args; i++, ctx->nr_user_files++) { |
65e19f54 JA |
5872 | struct fixed_file_table *table; |
5873 | unsigned index; | |
5874 | ||
6b06314c JA |
5875 | ret = -EFAULT; |
5876 | if (copy_from_user(&fd, &fds[i], sizeof(fd))) | |
5877 | break; | |
08a45173 JA |
5878 | /* allow sparse sets */ |
5879 | if (fd == -1) { | |
5880 | ret = 0; | |
5881 | continue; | |
5882 | } | |
6b06314c | 5883 | |
05f3fb3c | 5884 | table = &ctx->file_data->table[i >> IORING_FILE_TABLE_SHIFT]; |
65e19f54 | 5885 | index = i & IORING_FILE_TABLE_MASK; |
05f3fb3c | 5886 | file = fget(fd); |
6b06314c JA |
5887 | |
5888 | ret = -EBADF; | |
05f3fb3c | 5889 | if (!file) |
6b06314c | 5890 | break; |
05f3fb3c | 5891 | |
6b06314c JA |
5892 | /* |
5893 | * Don't allow io_uring instances to be registered. If UNIX | |
5894 | * isn't enabled, then this causes a reference cycle and this | |
5895 | * instance can never get freed. If UNIX is enabled we'll | |
5896 | * handle it just fine, but there's still no point in allowing | |
5897 | * a ring fd as it doesn't support regular read/write anyway. | |
5898 | */ | |
05f3fb3c JA |
5899 | if (file->f_op == &io_uring_fops) { |
5900 | fput(file); | |
6b06314c JA |
5901 | break; |
5902 | } | |
6b06314c | 5903 | ret = 0; |
05f3fb3c | 5904 | table->files[index] = file; |
6b06314c JA |
5905 | } |
5906 | ||
5907 | if (ret) { | |
65e19f54 | 5908 | for (i = 0; i < ctx->nr_user_files; i++) { |
65e19f54 JA |
5909 | file = io_file_from_index(ctx, i); |
5910 | if (file) | |
5911 | fput(file); | |
5912 | } | |
5913 | for (i = 0; i < nr_tables; i++) | |
05f3fb3c | 5914 | kfree(ctx->file_data->table[i].files); |
6b06314c | 5915 | |
05f3fb3c JA |
5916 | kfree(ctx->file_data->table); |
5917 | kfree(ctx->file_data); | |
5918 | ctx->file_data = NULL; | |
6b06314c JA |
5919 | ctx->nr_user_files = 0; |
5920 | return ret; | |
5921 | } | |
5922 | ||
5923 | ret = io_sqe_files_scm(ctx); | |
5924 | if (ret) | |
5925 | io_sqe_files_unregister(ctx); | |
5926 | ||
5927 | return ret; | |
5928 | } | |
5929 | ||
c3a31e60 JA |
5930 | static int io_sqe_file_register(struct io_ring_ctx *ctx, struct file *file, |
5931 | int index) | |
5932 | { | |
5933 | #if defined(CONFIG_UNIX) | |
5934 | struct sock *sock = ctx->ring_sock->sk; | |
5935 | struct sk_buff_head *head = &sock->sk_receive_queue; | |
5936 | struct sk_buff *skb; | |
5937 | ||
5938 | /* | |
5939 | * See if we can merge this file into an existing skb SCM_RIGHTS | |
5940 | * file set. If there's no room, fall back to allocating a new skb | |
5941 | * and filling it in. | |
5942 | */ | |
5943 | spin_lock_irq(&head->lock); | |
5944 | skb = skb_peek(head); | |
5945 | if (skb) { | |
5946 | struct scm_fp_list *fpl = UNIXCB(skb).fp; | |
5947 | ||
5948 | if (fpl->count < SCM_MAX_FD) { | |
5949 | __skb_unlink(skb, head); | |
5950 | spin_unlock_irq(&head->lock); | |
5951 | fpl->fp[fpl->count] = get_file(file); | |
5952 | unix_inflight(fpl->user, fpl->fp[fpl->count]); | |
5953 | fpl->count++; | |
5954 | spin_lock_irq(&head->lock); | |
5955 | __skb_queue_head(head, skb); | |
5956 | } else { | |
5957 | skb = NULL; | |
5958 | } | |
5959 | } | |
5960 | spin_unlock_irq(&head->lock); | |
5961 | ||
5962 | if (skb) { | |
5963 | fput(file); | |
5964 | return 0; | |
5965 | } | |
5966 | ||
5967 | return __io_sqe_files_scm(ctx, 1, index); | |
5968 | #else | |
5969 | return 0; | |
5970 | #endif | |
5971 | } | |
5972 | ||
05f3fb3c | 5973 | static void io_atomic_switch(struct percpu_ref *ref) |
c3a31e60 | 5974 | { |
05f3fb3c JA |
5975 | struct fixed_file_data *data; |
5976 | ||
dd3db2a3 JA |
5977 | /* |
5978 | * Juggle reference to ensure we hit zero, if needed, so we can | |
5979 | * switch back to percpu mode | |
5980 | */ | |
05f3fb3c | 5981 | data = container_of(ref, struct fixed_file_data, refs); |
dd3db2a3 JA |
5982 | percpu_ref_put(&data->refs); |
5983 | percpu_ref_get(&data->refs); | |
05f3fb3c JA |
5984 | } |
5985 | ||
5986 | static bool io_queue_file_removal(struct fixed_file_data *data, | |
5987 | struct file *file) | |
5988 | { | |
5989 | struct io_file_put *pfile, pfile_stack; | |
5990 | DECLARE_COMPLETION_ONSTACK(done); | |
5991 | ||
5992 | /* | |
5993 | * If we fail allocating the struct we need for doing async reomval | |
5994 | * of this file, just punt to sync and wait for it. | |
5995 | */ | |
5996 | pfile = kzalloc(sizeof(*pfile), GFP_KERNEL); | |
5997 | if (!pfile) { | |
5998 | pfile = &pfile_stack; | |
5999 | pfile->done = &done; | |
6000 | } | |
6001 | ||
6002 | pfile->file = file; | |
6003 | llist_add(&pfile->llist, &data->put_llist); | |
6004 | ||
6005 | if (pfile == &pfile_stack) { | |
dd3db2a3 | 6006 | percpu_ref_switch_to_atomic(&data->refs, io_atomic_switch); |
05f3fb3c JA |
6007 | wait_for_completion(&done); |
6008 | flush_work(&data->ref_work); | |
6009 | return false; | |
6010 | } | |
6011 | ||
6012 | return true; | |
6013 | } | |
6014 | ||
6015 | static int __io_sqe_files_update(struct io_ring_ctx *ctx, | |
6016 | struct io_uring_files_update *up, | |
6017 | unsigned nr_args) | |
6018 | { | |
6019 | struct fixed_file_data *data = ctx->file_data; | |
6020 | bool ref_switch = false; | |
6021 | struct file *file; | |
c3a31e60 JA |
6022 | __s32 __user *fds; |
6023 | int fd, i, err; | |
6024 | __u32 done; | |
6025 | ||
05f3fb3c | 6026 | if (check_add_overflow(up->offset, nr_args, &done)) |
c3a31e60 JA |
6027 | return -EOVERFLOW; |
6028 | if (done > ctx->nr_user_files) | |
6029 | return -EINVAL; | |
6030 | ||
6031 | done = 0; | |
05f3fb3c | 6032 | fds = u64_to_user_ptr(up->fds); |
c3a31e60 | 6033 | while (nr_args) { |
65e19f54 JA |
6034 | struct fixed_file_table *table; |
6035 | unsigned index; | |
6036 | ||
c3a31e60 JA |
6037 | err = 0; |
6038 | if (copy_from_user(&fd, &fds[done], sizeof(fd))) { | |
6039 | err = -EFAULT; | |
6040 | break; | |
6041 | } | |
05f3fb3c JA |
6042 | i = array_index_nospec(up->offset, ctx->nr_user_files); |
6043 | table = &ctx->file_data->table[i >> IORING_FILE_TABLE_SHIFT]; | |
65e19f54 JA |
6044 | index = i & IORING_FILE_TABLE_MASK; |
6045 | if (table->files[index]) { | |
05f3fb3c | 6046 | file = io_file_from_index(ctx, index); |
65e19f54 | 6047 | table->files[index] = NULL; |
05f3fb3c JA |
6048 | if (io_queue_file_removal(data, file)) |
6049 | ref_switch = true; | |
c3a31e60 JA |
6050 | } |
6051 | if (fd != -1) { | |
c3a31e60 JA |
6052 | file = fget(fd); |
6053 | if (!file) { | |
6054 | err = -EBADF; | |
6055 | break; | |
6056 | } | |
6057 | /* | |
6058 | * Don't allow io_uring instances to be registered. If | |
6059 | * UNIX isn't enabled, then this causes a reference | |
6060 | * cycle and this instance can never get freed. If UNIX | |
6061 | * is enabled we'll handle it just fine, but there's | |
6062 | * still no point in allowing a ring fd as it doesn't | |
6063 | * support regular read/write anyway. | |
6064 | */ | |
6065 | if (file->f_op == &io_uring_fops) { | |
6066 | fput(file); | |
6067 | err = -EBADF; | |
6068 | break; | |
6069 | } | |
65e19f54 | 6070 | table->files[index] = file; |
c3a31e60 JA |
6071 | err = io_sqe_file_register(ctx, file, i); |
6072 | if (err) | |
6073 | break; | |
6074 | } | |
6075 | nr_args--; | |
6076 | done++; | |
05f3fb3c JA |
6077 | up->offset++; |
6078 | } | |
6079 | ||
dd3db2a3 | 6080 | if (ref_switch) |
05f3fb3c | 6081 | percpu_ref_switch_to_atomic(&data->refs, io_atomic_switch); |
c3a31e60 JA |
6082 | |
6083 | return done ? done : err; | |
6084 | } | |
05f3fb3c JA |
6085 | static int io_sqe_files_update(struct io_ring_ctx *ctx, void __user *arg, |
6086 | unsigned nr_args) | |
6087 | { | |
6088 | struct io_uring_files_update up; | |
6089 | ||
6090 | if (!ctx->file_data) | |
6091 | return -ENXIO; | |
6092 | if (!nr_args) | |
6093 | return -EINVAL; | |
6094 | if (copy_from_user(&up, arg, sizeof(up))) | |
6095 | return -EFAULT; | |
6096 | if (up.resv) | |
6097 | return -EINVAL; | |
6098 | ||
6099 | return __io_sqe_files_update(ctx, &up, nr_args); | |
6100 | } | |
c3a31e60 | 6101 | |
7d723065 JA |
6102 | static void io_put_work(struct io_wq_work *work) |
6103 | { | |
6104 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); | |
6105 | ||
6106 | io_put_req(req); | |
6107 | } | |
6108 | ||
6109 | static void io_get_work(struct io_wq_work *work) | |
6110 | { | |
6111 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); | |
6112 | ||
6113 | refcount_inc(&req->refs); | |
6114 | } | |
6115 | ||
24369c2e PB |
6116 | static int io_init_wq_offload(struct io_ring_ctx *ctx, |
6117 | struct io_uring_params *p) | |
6118 | { | |
6119 | struct io_wq_data data; | |
6120 | struct fd f; | |
6121 | struct io_ring_ctx *ctx_attach; | |
6122 | unsigned int concurrency; | |
6123 | int ret = 0; | |
6124 | ||
6125 | data.user = ctx->user; | |
6126 | data.get_work = io_get_work; | |
6127 | data.put_work = io_put_work; | |
6128 | ||
6129 | if (!(p->flags & IORING_SETUP_ATTACH_WQ)) { | |
6130 | /* Do QD, or 4 * CPUS, whatever is smallest */ | |
6131 | concurrency = min(ctx->sq_entries, 4 * num_online_cpus()); | |
6132 | ||
6133 | ctx->io_wq = io_wq_create(concurrency, &data); | |
6134 | if (IS_ERR(ctx->io_wq)) { | |
6135 | ret = PTR_ERR(ctx->io_wq); | |
6136 | ctx->io_wq = NULL; | |
6137 | } | |
6138 | return ret; | |
6139 | } | |
6140 | ||
6141 | f = fdget(p->wq_fd); | |
6142 | if (!f.file) | |
6143 | return -EBADF; | |
6144 | ||
6145 | if (f.file->f_op != &io_uring_fops) { | |
6146 | ret = -EINVAL; | |
6147 | goto out_fput; | |
6148 | } | |
6149 | ||
6150 | ctx_attach = f.file->private_data; | |
6151 | /* @io_wq is protected by holding the fd */ | |
6152 | if (!io_wq_get(ctx_attach->io_wq, &data)) { | |
6153 | ret = -EINVAL; | |
6154 | goto out_fput; | |
6155 | } | |
6156 | ||
6157 | ctx->io_wq = ctx_attach->io_wq; | |
6158 | out_fput: | |
6159 | fdput(f); | |
6160 | return ret; | |
6161 | } | |
6162 | ||
6c271ce2 JA |
6163 | static int io_sq_offload_start(struct io_ring_ctx *ctx, |
6164 | struct io_uring_params *p) | |
2b188cc1 JA |
6165 | { |
6166 | int ret; | |
6167 | ||
6c271ce2 | 6168 | init_waitqueue_head(&ctx->sqo_wait); |
2b188cc1 JA |
6169 | mmgrab(current->mm); |
6170 | ctx->sqo_mm = current->mm; | |
6171 | ||
6c271ce2 | 6172 | if (ctx->flags & IORING_SETUP_SQPOLL) { |
3ec482d1 JA |
6173 | ret = -EPERM; |
6174 | if (!capable(CAP_SYS_ADMIN)) | |
6175 | goto err; | |
6176 | ||
917257da JA |
6177 | ctx->sq_thread_idle = msecs_to_jiffies(p->sq_thread_idle); |
6178 | if (!ctx->sq_thread_idle) | |
6179 | ctx->sq_thread_idle = HZ; | |
6180 | ||
6c271ce2 | 6181 | if (p->flags & IORING_SETUP_SQ_AFF) { |
44a9bd18 | 6182 | int cpu = p->sq_thread_cpu; |
6c271ce2 | 6183 | |
917257da | 6184 | ret = -EINVAL; |
44a9bd18 JA |
6185 | if (cpu >= nr_cpu_ids) |
6186 | goto err; | |
7889f44d | 6187 | if (!cpu_online(cpu)) |
917257da JA |
6188 | goto err; |
6189 | ||
6c271ce2 JA |
6190 | ctx->sqo_thread = kthread_create_on_cpu(io_sq_thread, |
6191 | ctx, cpu, | |
6192 | "io_uring-sq"); | |
6193 | } else { | |
6194 | ctx->sqo_thread = kthread_create(io_sq_thread, ctx, | |
6195 | "io_uring-sq"); | |
6196 | } | |
6197 | if (IS_ERR(ctx->sqo_thread)) { | |
6198 | ret = PTR_ERR(ctx->sqo_thread); | |
6199 | ctx->sqo_thread = NULL; | |
6200 | goto err; | |
6201 | } | |
6202 | wake_up_process(ctx->sqo_thread); | |
6203 | } else if (p->flags & IORING_SETUP_SQ_AFF) { | |
6204 | /* Can't have SQ_AFF without SQPOLL */ | |
6205 | ret = -EINVAL; | |
6206 | goto err; | |
6207 | } | |
6208 | ||
24369c2e PB |
6209 | ret = io_init_wq_offload(ctx, p); |
6210 | if (ret) | |
2b188cc1 | 6211 | goto err; |
2b188cc1 JA |
6212 | |
6213 | return 0; | |
6214 | err: | |
54a91f3b | 6215 | io_finish_async(ctx); |
2b188cc1 JA |
6216 | mmdrop(ctx->sqo_mm); |
6217 | ctx->sqo_mm = NULL; | |
6218 | return ret; | |
6219 | } | |
6220 | ||
6221 | static void io_unaccount_mem(struct user_struct *user, unsigned long nr_pages) | |
6222 | { | |
6223 | atomic_long_sub(nr_pages, &user->locked_vm); | |
6224 | } | |
6225 | ||
6226 | static int io_account_mem(struct user_struct *user, unsigned long nr_pages) | |
6227 | { | |
6228 | unsigned long page_limit, cur_pages, new_pages; | |
6229 | ||
6230 | /* Don't allow more pages than we can safely lock */ | |
6231 | page_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT; | |
6232 | ||
6233 | do { | |
6234 | cur_pages = atomic_long_read(&user->locked_vm); | |
6235 | new_pages = cur_pages + nr_pages; | |
6236 | if (new_pages > page_limit) | |
6237 | return -ENOMEM; | |
6238 | } while (atomic_long_cmpxchg(&user->locked_vm, cur_pages, | |
6239 | new_pages) != cur_pages); | |
6240 | ||
6241 | return 0; | |
6242 | } | |
6243 | ||
6244 | static void io_mem_free(void *ptr) | |
6245 | { | |
52e04ef4 MR |
6246 | struct page *page; |
6247 | ||
6248 | if (!ptr) | |
6249 | return; | |
2b188cc1 | 6250 | |
52e04ef4 | 6251 | page = virt_to_head_page(ptr); |
2b188cc1 JA |
6252 | if (put_page_testzero(page)) |
6253 | free_compound_page(page); | |
6254 | } | |
6255 | ||
6256 | static void *io_mem_alloc(size_t size) | |
6257 | { | |
6258 | gfp_t gfp_flags = GFP_KERNEL | __GFP_ZERO | __GFP_NOWARN | __GFP_COMP | | |
6259 | __GFP_NORETRY; | |
6260 | ||
6261 | return (void *) __get_free_pages(gfp_flags, get_order(size)); | |
6262 | } | |
6263 | ||
75b28aff HV |
6264 | static unsigned long rings_size(unsigned sq_entries, unsigned cq_entries, |
6265 | size_t *sq_offset) | |
6266 | { | |
6267 | struct io_rings *rings; | |
6268 | size_t off, sq_array_size; | |
6269 | ||
6270 | off = struct_size(rings, cqes, cq_entries); | |
6271 | if (off == SIZE_MAX) | |
6272 | return SIZE_MAX; | |
6273 | ||
6274 | #ifdef CONFIG_SMP | |
6275 | off = ALIGN(off, SMP_CACHE_BYTES); | |
6276 | if (off == 0) | |
6277 | return SIZE_MAX; | |
6278 | #endif | |
6279 | ||
6280 | sq_array_size = array_size(sizeof(u32), sq_entries); | |
6281 | if (sq_array_size == SIZE_MAX) | |
6282 | return SIZE_MAX; | |
6283 | ||
6284 | if (check_add_overflow(off, sq_array_size, &off)) | |
6285 | return SIZE_MAX; | |
6286 | ||
6287 | if (sq_offset) | |
6288 | *sq_offset = off; | |
6289 | ||
6290 | return off; | |
6291 | } | |
6292 | ||
2b188cc1 JA |
6293 | static unsigned long ring_pages(unsigned sq_entries, unsigned cq_entries) |
6294 | { | |
75b28aff | 6295 | size_t pages; |
2b188cc1 | 6296 | |
75b28aff HV |
6297 | pages = (size_t)1 << get_order( |
6298 | rings_size(sq_entries, cq_entries, NULL)); | |
6299 | pages += (size_t)1 << get_order( | |
6300 | array_size(sizeof(struct io_uring_sqe), sq_entries)); | |
2b188cc1 | 6301 | |
75b28aff | 6302 | return pages; |
2b188cc1 JA |
6303 | } |
6304 | ||
edafccee JA |
6305 | static int io_sqe_buffer_unregister(struct io_ring_ctx *ctx) |
6306 | { | |
6307 | int i, j; | |
6308 | ||
6309 | if (!ctx->user_bufs) | |
6310 | return -ENXIO; | |
6311 | ||
6312 | for (i = 0; i < ctx->nr_user_bufs; i++) { | |
6313 | struct io_mapped_ubuf *imu = &ctx->user_bufs[i]; | |
6314 | ||
6315 | for (j = 0; j < imu->nr_bvecs; j++) | |
f1f6a7dd | 6316 | unpin_user_page(imu->bvec[j].bv_page); |
edafccee JA |
6317 | |
6318 | if (ctx->account_mem) | |
6319 | io_unaccount_mem(ctx->user, imu->nr_bvecs); | |
d4ef6475 | 6320 | kvfree(imu->bvec); |
edafccee JA |
6321 | imu->nr_bvecs = 0; |
6322 | } | |
6323 | ||
6324 | kfree(ctx->user_bufs); | |
6325 | ctx->user_bufs = NULL; | |
6326 | ctx->nr_user_bufs = 0; | |
6327 | return 0; | |
6328 | } | |
6329 | ||
6330 | static int io_copy_iov(struct io_ring_ctx *ctx, struct iovec *dst, | |
6331 | void __user *arg, unsigned index) | |
6332 | { | |
6333 | struct iovec __user *src; | |
6334 | ||
6335 | #ifdef CONFIG_COMPAT | |
6336 | if (ctx->compat) { | |
6337 | struct compat_iovec __user *ciovs; | |
6338 | struct compat_iovec ciov; | |
6339 | ||
6340 | ciovs = (struct compat_iovec __user *) arg; | |
6341 | if (copy_from_user(&ciov, &ciovs[index], sizeof(ciov))) | |
6342 | return -EFAULT; | |
6343 | ||
d55e5f5b | 6344 | dst->iov_base = u64_to_user_ptr((u64)ciov.iov_base); |
edafccee JA |
6345 | dst->iov_len = ciov.iov_len; |
6346 | return 0; | |
6347 | } | |
6348 | #endif | |
6349 | src = (struct iovec __user *) arg; | |
6350 | if (copy_from_user(dst, &src[index], sizeof(*dst))) | |
6351 | return -EFAULT; | |
6352 | return 0; | |
6353 | } | |
6354 | ||
6355 | static int io_sqe_buffer_register(struct io_ring_ctx *ctx, void __user *arg, | |
6356 | unsigned nr_args) | |
6357 | { | |
6358 | struct vm_area_struct **vmas = NULL; | |
6359 | struct page **pages = NULL; | |
6360 | int i, j, got_pages = 0; | |
6361 | int ret = -EINVAL; | |
6362 | ||
6363 | if (ctx->user_bufs) | |
6364 | return -EBUSY; | |
6365 | if (!nr_args || nr_args > UIO_MAXIOV) | |
6366 | return -EINVAL; | |
6367 | ||
6368 | ctx->user_bufs = kcalloc(nr_args, sizeof(struct io_mapped_ubuf), | |
6369 | GFP_KERNEL); | |
6370 | if (!ctx->user_bufs) | |
6371 | return -ENOMEM; | |
6372 | ||
6373 | for (i = 0; i < nr_args; i++) { | |
6374 | struct io_mapped_ubuf *imu = &ctx->user_bufs[i]; | |
6375 | unsigned long off, start, end, ubuf; | |
6376 | int pret, nr_pages; | |
6377 | struct iovec iov; | |
6378 | size_t size; | |
6379 | ||
6380 | ret = io_copy_iov(ctx, &iov, arg, i); | |
6381 | if (ret) | |
a278682d | 6382 | goto err; |
edafccee JA |
6383 | |
6384 | /* | |
6385 | * Don't impose further limits on the size and buffer | |
6386 | * constraints here, we'll -EINVAL later when IO is | |
6387 | * submitted if they are wrong. | |
6388 | */ | |
6389 | ret = -EFAULT; | |
6390 | if (!iov.iov_base || !iov.iov_len) | |
6391 | goto err; | |
6392 | ||
6393 | /* arbitrary limit, but we need something */ | |
6394 | if (iov.iov_len > SZ_1G) | |
6395 | goto err; | |
6396 | ||
6397 | ubuf = (unsigned long) iov.iov_base; | |
6398 | end = (ubuf + iov.iov_len + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
6399 | start = ubuf >> PAGE_SHIFT; | |
6400 | nr_pages = end - start; | |
6401 | ||
6402 | if (ctx->account_mem) { | |
6403 | ret = io_account_mem(ctx->user, nr_pages); | |
6404 | if (ret) | |
6405 | goto err; | |
6406 | } | |
6407 | ||
6408 | ret = 0; | |
6409 | if (!pages || nr_pages > got_pages) { | |
6410 | kfree(vmas); | |
6411 | kfree(pages); | |
d4ef6475 | 6412 | pages = kvmalloc_array(nr_pages, sizeof(struct page *), |
edafccee | 6413 | GFP_KERNEL); |
d4ef6475 | 6414 | vmas = kvmalloc_array(nr_pages, |
edafccee JA |
6415 | sizeof(struct vm_area_struct *), |
6416 | GFP_KERNEL); | |
6417 | if (!pages || !vmas) { | |
6418 | ret = -ENOMEM; | |
6419 | if (ctx->account_mem) | |
6420 | io_unaccount_mem(ctx->user, nr_pages); | |
6421 | goto err; | |
6422 | } | |
6423 | got_pages = nr_pages; | |
6424 | } | |
6425 | ||
d4ef6475 | 6426 | imu->bvec = kvmalloc_array(nr_pages, sizeof(struct bio_vec), |
edafccee JA |
6427 | GFP_KERNEL); |
6428 | ret = -ENOMEM; | |
6429 | if (!imu->bvec) { | |
6430 | if (ctx->account_mem) | |
6431 | io_unaccount_mem(ctx->user, nr_pages); | |
6432 | goto err; | |
6433 | } | |
6434 | ||
6435 | ret = 0; | |
6436 | down_read(¤t->mm->mmap_sem); | |
2113b05d | 6437 | pret = pin_user_pages(ubuf, nr_pages, |
932f4a63 IW |
6438 | FOLL_WRITE | FOLL_LONGTERM, |
6439 | pages, vmas); | |
edafccee JA |
6440 | if (pret == nr_pages) { |
6441 | /* don't support file backed memory */ | |
6442 | for (j = 0; j < nr_pages; j++) { | |
6443 | struct vm_area_struct *vma = vmas[j]; | |
6444 | ||
6445 | if (vma->vm_file && | |
6446 | !is_file_hugepages(vma->vm_file)) { | |
6447 | ret = -EOPNOTSUPP; | |
6448 | break; | |
6449 | } | |
6450 | } | |
6451 | } else { | |
6452 | ret = pret < 0 ? pret : -EFAULT; | |
6453 | } | |
6454 | up_read(¤t->mm->mmap_sem); | |
6455 | if (ret) { | |
6456 | /* | |
6457 | * if we did partial map, or found file backed vmas, | |
6458 | * release any pages we did get | |
6459 | */ | |
27c4d3a3 | 6460 | if (pret > 0) |
f1f6a7dd | 6461 | unpin_user_pages(pages, pret); |
edafccee JA |
6462 | if (ctx->account_mem) |
6463 | io_unaccount_mem(ctx->user, nr_pages); | |
d4ef6475 | 6464 | kvfree(imu->bvec); |
edafccee JA |
6465 | goto err; |
6466 | } | |
6467 | ||
6468 | off = ubuf & ~PAGE_MASK; | |
6469 | size = iov.iov_len; | |
6470 | for (j = 0; j < nr_pages; j++) { | |
6471 | size_t vec_len; | |
6472 | ||
6473 | vec_len = min_t(size_t, size, PAGE_SIZE - off); | |
6474 | imu->bvec[j].bv_page = pages[j]; | |
6475 | imu->bvec[j].bv_len = vec_len; | |
6476 | imu->bvec[j].bv_offset = off; | |
6477 | off = 0; | |
6478 | size -= vec_len; | |
6479 | } | |
6480 | /* store original address for later verification */ | |
6481 | imu->ubuf = ubuf; | |
6482 | imu->len = iov.iov_len; | |
6483 | imu->nr_bvecs = nr_pages; | |
6484 | ||
6485 | ctx->nr_user_bufs++; | |
6486 | } | |
d4ef6475 MR |
6487 | kvfree(pages); |
6488 | kvfree(vmas); | |
edafccee JA |
6489 | return 0; |
6490 | err: | |
d4ef6475 MR |
6491 | kvfree(pages); |
6492 | kvfree(vmas); | |
edafccee JA |
6493 | io_sqe_buffer_unregister(ctx); |
6494 | return ret; | |
6495 | } | |
6496 | ||
9b402849 JA |
6497 | static int io_eventfd_register(struct io_ring_ctx *ctx, void __user *arg) |
6498 | { | |
6499 | __s32 __user *fds = arg; | |
6500 | int fd; | |
6501 | ||
6502 | if (ctx->cq_ev_fd) | |
6503 | return -EBUSY; | |
6504 | ||
6505 | if (copy_from_user(&fd, fds, sizeof(*fds))) | |
6506 | return -EFAULT; | |
6507 | ||
6508 | ctx->cq_ev_fd = eventfd_ctx_fdget(fd); | |
6509 | if (IS_ERR(ctx->cq_ev_fd)) { | |
6510 | int ret = PTR_ERR(ctx->cq_ev_fd); | |
6511 | ctx->cq_ev_fd = NULL; | |
6512 | return ret; | |
6513 | } | |
6514 | ||
6515 | return 0; | |
6516 | } | |
6517 | ||
6518 | static int io_eventfd_unregister(struct io_ring_ctx *ctx) | |
6519 | { | |
6520 | if (ctx->cq_ev_fd) { | |
6521 | eventfd_ctx_put(ctx->cq_ev_fd); | |
6522 | ctx->cq_ev_fd = NULL; | |
6523 | return 0; | |
6524 | } | |
6525 | ||
6526 | return -ENXIO; | |
6527 | } | |
6528 | ||
2b188cc1 JA |
6529 | static void io_ring_ctx_free(struct io_ring_ctx *ctx) |
6530 | { | |
6b06314c | 6531 | io_finish_async(ctx); |
2b188cc1 JA |
6532 | if (ctx->sqo_mm) |
6533 | mmdrop(ctx->sqo_mm); | |
def596e9 JA |
6534 | |
6535 | io_iopoll_reap_events(ctx); | |
edafccee | 6536 | io_sqe_buffer_unregister(ctx); |
6b06314c | 6537 | io_sqe_files_unregister(ctx); |
9b402849 | 6538 | io_eventfd_unregister(ctx); |
41726c9a | 6539 | idr_destroy(&ctx->personality_idr); |
def596e9 | 6540 | |
2b188cc1 | 6541 | #if defined(CONFIG_UNIX) |
355e8d26 EB |
6542 | if (ctx->ring_sock) { |
6543 | ctx->ring_sock->file = NULL; /* so that iput() is called */ | |
2b188cc1 | 6544 | sock_release(ctx->ring_sock); |
355e8d26 | 6545 | } |
2b188cc1 JA |
6546 | #endif |
6547 | ||
75b28aff | 6548 | io_mem_free(ctx->rings); |
2b188cc1 | 6549 | io_mem_free(ctx->sq_sqes); |
2b188cc1 JA |
6550 | |
6551 | percpu_ref_exit(&ctx->refs); | |
6552 | if (ctx->account_mem) | |
6553 | io_unaccount_mem(ctx->user, | |
6554 | ring_pages(ctx->sq_entries, ctx->cq_entries)); | |
6555 | free_uid(ctx->user); | |
181e448d | 6556 | put_cred(ctx->creds); |
206aefde | 6557 | kfree(ctx->completions); |
78076bb6 | 6558 | kfree(ctx->cancel_hash); |
0ddf92e8 | 6559 | kmem_cache_free(req_cachep, ctx->fallback_req); |
2b188cc1 JA |
6560 | kfree(ctx); |
6561 | } | |
6562 | ||
6563 | static __poll_t io_uring_poll(struct file *file, poll_table *wait) | |
6564 | { | |
6565 | struct io_ring_ctx *ctx = file->private_data; | |
6566 | __poll_t mask = 0; | |
6567 | ||
6568 | poll_wait(file, &ctx->cq_wait, wait); | |
4f7067c3 SB |
6569 | /* |
6570 | * synchronizes with barrier from wq_has_sleeper call in | |
6571 | * io_commit_cqring | |
6572 | */ | |
2b188cc1 | 6573 | smp_rmb(); |
75b28aff HV |
6574 | if (READ_ONCE(ctx->rings->sq.tail) - ctx->cached_sq_head != |
6575 | ctx->rings->sq_ring_entries) | |
2b188cc1 | 6576 | mask |= EPOLLOUT | EPOLLWRNORM; |
63e5d81f | 6577 | if (io_cqring_events(ctx, false)) |
2b188cc1 JA |
6578 | mask |= EPOLLIN | EPOLLRDNORM; |
6579 | ||
6580 | return mask; | |
6581 | } | |
6582 | ||
6583 | static int io_uring_fasync(int fd, struct file *file, int on) | |
6584 | { | |
6585 | struct io_ring_ctx *ctx = file->private_data; | |
6586 | ||
6587 | return fasync_helper(fd, file, on, &ctx->cq_fasync); | |
6588 | } | |
6589 | ||
071698e1 JA |
6590 | static int io_remove_personalities(int id, void *p, void *data) |
6591 | { | |
6592 | struct io_ring_ctx *ctx = data; | |
6593 | const struct cred *cred; | |
6594 | ||
6595 | cred = idr_remove(&ctx->personality_idr, id); | |
6596 | if (cred) | |
6597 | put_cred(cred); | |
6598 | return 0; | |
6599 | } | |
6600 | ||
2b188cc1 JA |
6601 | static void io_ring_ctx_wait_and_kill(struct io_ring_ctx *ctx) |
6602 | { | |
6603 | mutex_lock(&ctx->uring_lock); | |
6604 | percpu_ref_kill(&ctx->refs); | |
6605 | mutex_unlock(&ctx->uring_lock); | |
6606 | ||
df069d80 JA |
6607 | /* |
6608 | * Wait for sq thread to idle, if we have one. It won't spin on new | |
6609 | * work after we've killed the ctx ref above. This is important to do | |
6610 | * before we cancel existing commands, as the thread could otherwise | |
6611 | * be queueing new work post that. If that's work we need to cancel, | |
6612 | * it could cause shutdown to hang. | |
6613 | */ | |
6614 | while (ctx->sqo_thread && !wq_has_sleeper(&ctx->sqo_wait)) | |
6615 | cpu_relax(); | |
6616 | ||
5262f567 | 6617 | io_kill_timeouts(ctx); |
221c5eb2 | 6618 | io_poll_remove_all(ctx); |
561fb04a JA |
6619 | |
6620 | if (ctx->io_wq) | |
6621 | io_wq_cancel_all(ctx->io_wq); | |
6622 | ||
def596e9 | 6623 | io_iopoll_reap_events(ctx); |
15dff286 JA |
6624 | /* if we failed setting up the ctx, we might not have any rings */ |
6625 | if (ctx->rings) | |
6626 | io_cqring_overflow_flush(ctx, true); | |
071698e1 | 6627 | idr_for_each(&ctx->personality_idr, io_remove_personalities, ctx); |
206aefde | 6628 | wait_for_completion(&ctx->completions[0]); |
2b188cc1 JA |
6629 | io_ring_ctx_free(ctx); |
6630 | } | |
6631 | ||
6632 | static int io_uring_release(struct inode *inode, struct file *file) | |
6633 | { | |
6634 | struct io_ring_ctx *ctx = file->private_data; | |
6635 | ||
6636 | file->private_data = NULL; | |
6637 | io_ring_ctx_wait_and_kill(ctx); | |
6638 | return 0; | |
6639 | } | |
6640 | ||
fcb323cc JA |
6641 | static void io_uring_cancel_files(struct io_ring_ctx *ctx, |
6642 | struct files_struct *files) | |
6643 | { | |
6644 | struct io_kiocb *req; | |
6645 | DEFINE_WAIT(wait); | |
6646 | ||
6647 | while (!list_empty_careful(&ctx->inflight_list)) { | |
768134d4 | 6648 | struct io_kiocb *cancel_req = NULL; |
fcb323cc JA |
6649 | |
6650 | spin_lock_irq(&ctx->inflight_lock); | |
6651 | list_for_each_entry(req, &ctx->inflight_list, inflight_entry) { | |
768134d4 JA |
6652 | if (req->work.files != files) |
6653 | continue; | |
6654 | /* req is being completed, ignore */ | |
6655 | if (!refcount_inc_not_zero(&req->refs)) | |
6656 | continue; | |
6657 | cancel_req = req; | |
6658 | break; | |
fcb323cc | 6659 | } |
768134d4 | 6660 | if (cancel_req) |
fcb323cc | 6661 | prepare_to_wait(&ctx->inflight_wait, &wait, |
768134d4 | 6662 | TASK_UNINTERRUPTIBLE); |
fcb323cc JA |
6663 | spin_unlock_irq(&ctx->inflight_lock); |
6664 | ||
768134d4 JA |
6665 | /* We need to keep going until we don't find a matching req */ |
6666 | if (!cancel_req) | |
fcb323cc | 6667 | break; |
2f6d9b9d | 6668 | |
2ca10259 JA |
6669 | if (cancel_req->flags & REQ_F_OVERFLOW) { |
6670 | spin_lock_irq(&ctx->completion_lock); | |
6671 | list_del(&cancel_req->list); | |
6672 | cancel_req->flags &= ~REQ_F_OVERFLOW; | |
6673 | if (list_empty(&ctx->cq_overflow_list)) { | |
6674 | clear_bit(0, &ctx->sq_check_overflow); | |
6675 | clear_bit(0, &ctx->cq_check_overflow); | |
6676 | } | |
6677 | spin_unlock_irq(&ctx->completion_lock); | |
6678 | ||
6679 | WRITE_ONCE(ctx->rings->cq_overflow, | |
6680 | atomic_inc_return(&ctx->cached_cq_overflow)); | |
6681 | ||
6682 | /* | |
6683 | * Put inflight ref and overflow ref. If that's | |
6684 | * all we had, then we're done with this request. | |
6685 | */ | |
6686 | if (refcount_sub_and_test(2, &cancel_req->refs)) { | |
6687 | io_put_req(cancel_req); | |
6688 | continue; | |
6689 | } | |
6690 | } | |
6691 | ||
2f6d9b9d BL |
6692 | io_wq_cancel_work(ctx->io_wq, &cancel_req->work); |
6693 | io_put_req(cancel_req); | |
fcb323cc JA |
6694 | schedule(); |
6695 | } | |
768134d4 | 6696 | finish_wait(&ctx->inflight_wait, &wait); |
fcb323cc JA |
6697 | } |
6698 | ||
6699 | static int io_uring_flush(struct file *file, void *data) | |
6700 | { | |
6701 | struct io_ring_ctx *ctx = file->private_data; | |
6702 | ||
6703 | io_uring_cancel_files(ctx, data); | |
6ab23144 JA |
6704 | |
6705 | /* | |
6706 | * If the task is going away, cancel work it may have pending | |
6707 | */ | |
6708 | if (fatal_signal_pending(current) || (current->flags & PF_EXITING)) | |
6709 | io_wq_cancel_pid(ctx->io_wq, task_pid_vnr(current)); | |
6710 | ||
fcb323cc JA |
6711 | return 0; |
6712 | } | |
6713 | ||
6c5c240e RP |
6714 | static void *io_uring_validate_mmap_request(struct file *file, |
6715 | loff_t pgoff, size_t sz) | |
2b188cc1 | 6716 | { |
2b188cc1 | 6717 | struct io_ring_ctx *ctx = file->private_data; |
6c5c240e | 6718 | loff_t offset = pgoff << PAGE_SHIFT; |
2b188cc1 JA |
6719 | struct page *page; |
6720 | void *ptr; | |
6721 | ||
6722 | switch (offset) { | |
6723 | case IORING_OFF_SQ_RING: | |
75b28aff HV |
6724 | case IORING_OFF_CQ_RING: |
6725 | ptr = ctx->rings; | |
2b188cc1 JA |
6726 | break; |
6727 | case IORING_OFF_SQES: | |
6728 | ptr = ctx->sq_sqes; | |
6729 | break; | |
2b188cc1 | 6730 | default: |
6c5c240e | 6731 | return ERR_PTR(-EINVAL); |
2b188cc1 JA |
6732 | } |
6733 | ||
6734 | page = virt_to_head_page(ptr); | |
a50b854e | 6735 | if (sz > page_size(page)) |
6c5c240e RP |
6736 | return ERR_PTR(-EINVAL); |
6737 | ||
6738 | return ptr; | |
6739 | } | |
6740 | ||
6741 | #ifdef CONFIG_MMU | |
6742 | ||
6743 | static int io_uring_mmap(struct file *file, struct vm_area_struct *vma) | |
6744 | { | |
6745 | size_t sz = vma->vm_end - vma->vm_start; | |
6746 | unsigned long pfn; | |
6747 | void *ptr; | |
6748 | ||
6749 | ptr = io_uring_validate_mmap_request(file, vma->vm_pgoff, sz); | |
6750 | if (IS_ERR(ptr)) | |
6751 | return PTR_ERR(ptr); | |
2b188cc1 JA |
6752 | |
6753 | pfn = virt_to_phys(ptr) >> PAGE_SHIFT; | |
6754 | return remap_pfn_range(vma, vma->vm_start, pfn, sz, vma->vm_page_prot); | |
6755 | } | |
6756 | ||
6c5c240e RP |
6757 | #else /* !CONFIG_MMU */ |
6758 | ||
6759 | static int io_uring_mmap(struct file *file, struct vm_area_struct *vma) | |
6760 | { | |
6761 | return vma->vm_flags & (VM_SHARED | VM_MAYSHARE) ? 0 : -EINVAL; | |
6762 | } | |
6763 | ||
6764 | static unsigned int io_uring_nommu_mmap_capabilities(struct file *file) | |
6765 | { | |
6766 | return NOMMU_MAP_DIRECT | NOMMU_MAP_READ | NOMMU_MAP_WRITE; | |
6767 | } | |
6768 | ||
6769 | static unsigned long io_uring_nommu_get_unmapped_area(struct file *file, | |
6770 | unsigned long addr, unsigned long len, | |
6771 | unsigned long pgoff, unsigned long flags) | |
6772 | { | |
6773 | void *ptr; | |
6774 | ||
6775 | ptr = io_uring_validate_mmap_request(file, pgoff, len); | |
6776 | if (IS_ERR(ptr)) | |
6777 | return PTR_ERR(ptr); | |
6778 | ||
6779 | return (unsigned long) ptr; | |
6780 | } | |
6781 | ||
6782 | #endif /* !CONFIG_MMU */ | |
6783 | ||
2b188cc1 JA |
6784 | SYSCALL_DEFINE6(io_uring_enter, unsigned int, fd, u32, to_submit, |
6785 | u32, min_complete, u32, flags, const sigset_t __user *, sig, | |
6786 | size_t, sigsz) | |
6787 | { | |
6788 | struct io_ring_ctx *ctx; | |
6789 | long ret = -EBADF; | |
6790 | int submitted = 0; | |
6791 | struct fd f; | |
6792 | ||
b41e9852 JA |
6793 | if (current->task_works) |
6794 | task_work_run(); | |
6795 | ||
6c271ce2 | 6796 | if (flags & ~(IORING_ENTER_GETEVENTS | IORING_ENTER_SQ_WAKEUP)) |
2b188cc1 JA |
6797 | return -EINVAL; |
6798 | ||
6799 | f = fdget(fd); | |
6800 | if (!f.file) | |
6801 | return -EBADF; | |
6802 | ||
6803 | ret = -EOPNOTSUPP; | |
6804 | if (f.file->f_op != &io_uring_fops) | |
6805 | goto out_fput; | |
6806 | ||
6807 | ret = -ENXIO; | |
6808 | ctx = f.file->private_data; | |
6809 | if (!percpu_ref_tryget(&ctx->refs)) | |
6810 | goto out_fput; | |
6811 | ||
6c271ce2 JA |
6812 | /* |
6813 | * For SQ polling, the thread will do all submissions and completions. | |
6814 | * Just return the requested submit count, and wake the thread if | |
6815 | * we were asked to. | |
6816 | */ | |
b2a9eada | 6817 | ret = 0; |
6c271ce2 | 6818 | if (ctx->flags & IORING_SETUP_SQPOLL) { |
c1edbf5f JA |
6819 | if (!list_empty_careful(&ctx->cq_overflow_list)) |
6820 | io_cqring_overflow_flush(ctx, false); | |
6c271ce2 JA |
6821 | if (flags & IORING_ENTER_SQ_WAKEUP) |
6822 | wake_up(&ctx->sqo_wait); | |
6823 | submitted = to_submit; | |
b2a9eada | 6824 | } else if (to_submit) { |
ae9428ca | 6825 | struct mm_struct *cur_mm; |
2b188cc1 JA |
6826 | |
6827 | mutex_lock(&ctx->uring_lock); | |
ae9428ca PB |
6828 | /* already have mm, so io_submit_sqes() won't try to grab it */ |
6829 | cur_mm = ctx->sqo_mm; | |
6830 | submitted = io_submit_sqes(ctx, to_submit, f.file, fd, | |
6831 | &cur_mm, false); | |
2b188cc1 | 6832 | mutex_unlock(&ctx->uring_lock); |
7c504e65 PB |
6833 | |
6834 | if (submitted != to_submit) | |
6835 | goto out; | |
2b188cc1 JA |
6836 | } |
6837 | if (flags & IORING_ENTER_GETEVENTS) { | |
def596e9 JA |
6838 | unsigned nr_events = 0; |
6839 | ||
2b188cc1 JA |
6840 | min_complete = min(min_complete, ctx->cq_entries); |
6841 | ||
def596e9 | 6842 | if (ctx->flags & IORING_SETUP_IOPOLL) { |
def596e9 | 6843 | ret = io_iopoll_check(ctx, &nr_events, min_complete); |
def596e9 JA |
6844 | } else { |
6845 | ret = io_cqring_wait(ctx, min_complete, sig, sigsz); | |
6846 | } | |
2b188cc1 JA |
6847 | } |
6848 | ||
7c504e65 | 6849 | out: |
6805b32e | 6850 | percpu_ref_put(&ctx->refs); |
2b188cc1 JA |
6851 | out_fput: |
6852 | fdput(f); | |
6853 | return submitted ? submitted : ret; | |
6854 | } | |
6855 | ||
bebdb65e | 6856 | #ifdef CONFIG_PROC_FS |
87ce955b JA |
6857 | static int io_uring_show_cred(int id, void *p, void *data) |
6858 | { | |
6859 | const struct cred *cred = p; | |
6860 | struct seq_file *m = data; | |
6861 | struct user_namespace *uns = seq_user_ns(m); | |
6862 | struct group_info *gi; | |
6863 | kernel_cap_t cap; | |
6864 | unsigned __capi; | |
6865 | int g; | |
6866 | ||
6867 | seq_printf(m, "%5d\n", id); | |
6868 | seq_put_decimal_ull(m, "\tUid:\t", from_kuid_munged(uns, cred->uid)); | |
6869 | seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->euid)); | |
6870 | seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->suid)); | |
6871 | seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->fsuid)); | |
6872 | seq_put_decimal_ull(m, "\n\tGid:\t", from_kgid_munged(uns, cred->gid)); | |
6873 | seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->egid)); | |
6874 | seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->sgid)); | |
6875 | seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->fsgid)); | |
6876 | seq_puts(m, "\n\tGroups:\t"); | |
6877 | gi = cred->group_info; | |
6878 | for (g = 0; g < gi->ngroups; g++) { | |
6879 | seq_put_decimal_ull(m, g ? " " : "", | |
6880 | from_kgid_munged(uns, gi->gid[g])); | |
6881 | } | |
6882 | seq_puts(m, "\n\tCapEff:\t"); | |
6883 | cap = cred->cap_effective; | |
6884 | CAP_FOR_EACH_U32(__capi) | |
6885 | seq_put_hex_ll(m, NULL, cap.cap[CAP_LAST_U32 - __capi], 8); | |
6886 | seq_putc(m, '\n'); | |
6887 | return 0; | |
6888 | } | |
6889 | ||
6890 | static void __io_uring_show_fdinfo(struct io_ring_ctx *ctx, struct seq_file *m) | |
6891 | { | |
6892 | int i; | |
6893 | ||
6894 | mutex_lock(&ctx->uring_lock); | |
6895 | seq_printf(m, "UserFiles:\t%u\n", ctx->nr_user_files); | |
6896 | for (i = 0; i < ctx->nr_user_files; i++) { | |
6897 | struct fixed_file_table *table; | |
6898 | struct file *f; | |
6899 | ||
6900 | table = &ctx->file_data->table[i >> IORING_FILE_TABLE_SHIFT]; | |
6901 | f = table->files[i & IORING_FILE_TABLE_MASK]; | |
6902 | if (f) | |
6903 | seq_printf(m, "%5u: %s\n", i, file_dentry(f)->d_iname); | |
6904 | else | |
6905 | seq_printf(m, "%5u: <none>\n", i); | |
6906 | } | |
6907 | seq_printf(m, "UserBufs:\t%u\n", ctx->nr_user_bufs); | |
6908 | for (i = 0; i < ctx->nr_user_bufs; i++) { | |
6909 | struct io_mapped_ubuf *buf = &ctx->user_bufs[i]; | |
6910 | ||
6911 | seq_printf(m, "%5u: 0x%llx/%u\n", i, buf->ubuf, | |
6912 | (unsigned int) buf->len); | |
6913 | } | |
6914 | if (!idr_is_empty(&ctx->personality_idr)) { | |
6915 | seq_printf(m, "Personalities:\n"); | |
6916 | idr_for_each(&ctx->personality_idr, io_uring_show_cred, m); | |
6917 | } | |
d7718a9d JA |
6918 | seq_printf(m, "PollList:\n"); |
6919 | spin_lock_irq(&ctx->completion_lock); | |
6920 | for (i = 0; i < (1U << ctx->cancel_hash_bits); i++) { | |
6921 | struct hlist_head *list = &ctx->cancel_hash[i]; | |
6922 | struct io_kiocb *req; | |
6923 | ||
6924 | hlist_for_each_entry(req, list, hash_node) | |
6925 | seq_printf(m, " op=%d, task_works=%d\n", req->opcode, | |
6926 | req->task->task_works != NULL); | |
6927 | } | |
6928 | spin_unlock_irq(&ctx->completion_lock); | |
87ce955b JA |
6929 | mutex_unlock(&ctx->uring_lock); |
6930 | } | |
6931 | ||
6932 | static void io_uring_show_fdinfo(struct seq_file *m, struct file *f) | |
6933 | { | |
6934 | struct io_ring_ctx *ctx = f->private_data; | |
6935 | ||
6936 | if (percpu_ref_tryget(&ctx->refs)) { | |
6937 | __io_uring_show_fdinfo(ctx, m); | |
6938 | percpu_ref_put(&ctx->refs); | |
6939 | } | |
6940 | } | |
bebdb65e | 6941 | #endif |
87ce955b | 6942 | |
2b188cc1 JA |
6943 | static const struct file_operations io_uring_fops = { |
6944 | .release = io_uring_release, | |
fcb323cc | 6945 | .flush = io_uring_flush, |
2b188cc1 | 6946 | .mmap = io_uring_mmap, |
6c5c240e RP |
6947 | #ifndef CONFIG_MMU |
6948 | .get_unmapped_area = io_uring_nommu_get_unmapped_area, | |
6949 | .mmap_capabilities = io_uring_nommu_mmap_capabilities, | |
6950 | #endif | |
2b188cc1 JA |
6951 | .poll = io_uring_poll, |
6952 | .fasync = io_uring_fasync, | |
bebdb65e | 6953 | #ifdef CONFIG_PROC_FS |
87ce955b | 6954 | .show_fdinfo = io_uring_show_fdinfo, |
bebdb65e | 6955 | #endif |
2b188cc1 JA |
6956 | }; |
6957 | ||
6958 | static int io_allocate_scq_urings(struct io_ring_ctx *ctx, | |
6959 | struct io_uring_params *p) | |
6960 | { | |
75b28aff HV |
6961 | struct io_rings *rings; |
6962 | size_t size, sq_array_offset; | |
2b188cc1 | 6963 | |
75b28aff HV |
6964 | size = rings_size(p->sq_entries, p->cq_entries, &sq_array_offset); |
6965 | if (size == SIZE_MAX) | |
6966 | return -EOVERFLOW; | |
6967 | ||
6968 | rings = io_mem_alloc(size); | |
6969 | if (!rings) | |
2b188cc1 JA |
6970 | return -ENOMEM; |
6971 | ||
75b28aff HV |
6972 | ctx->rings = rings; |
6973 | ctx->sq_array = (u32 *)((char *)rings + sq_array_offset); | |
6974 | rings->sq_ring_mask = p->sq_entries - 1; | |
6975 | rings->cq_ring_mask = p->cq_entries - 1; | |
6976 | rings->sq_ring_entries = p->sq_entries; | |
6977 | rings->cq_ring_entries = p->cq_entries; | |
6978 | ctx->sq_mask = rings->sq_ring_mask; | |
6979 | ctx->cq_mask = rings->cq_ring_mask; | |
6980 | ctx->sq_entries = rings->sq_ring_entries; | |
6981 | ctx->cq_entries = rings->cq_ring_entries; | |
2b188cc1 JA |
6982 | |
6983 | size = array_size(sizeof(struct io_uring_sqe), p->sq_entries); | |
eb065d30 JA |
6984 | if (size == SIZE_MAX) { |
6985 | io_mem_free(ctx->rings); | |
6986 | ctx->rings = NULL; | |
2b188cc1 | 6987 | return -EOVERFLOW; |
eb065d30 | 6988 | } |
2b188cc1 JA |
6989 | |
6990 | ctx->sq_sqes = io_mem_alloc(size); | |
eb065d30 JA |
6991 | if (!ctx->sq_sqes) { |
6992 | io_mem_free(ctx->rings); | |
6993 | ctx->rings = NULL; | |
2b188cc1 | 6994 | return -ENOMEM; |
eb065d30 | 6995 | } |
2b188cc1 | 6996 | |
2b188cc1 JA |
6997 | return 0; |
6998 | } | |
6999 | ||
7000 | /* | |
7001 | * Allocate an anonymous fd, this is what constitutes the application | |
7002 | * visible backing of an io_uring instance. The application mmaps this | |
7003 | * fd to gain access to the SQ/CQ ring details. If UNIX sockets are enabled, | |
7004 | * we have to tie this fd to a socket for file garbage collection purposes. | |
7005 | */ | |
7006 | static int io_uring_get_fd(struct io_ring_ctx *ctx) | |
7007 | { | |
7008 | struct file *file; | |
7009 | int ret; | |
7010 | ||
7011 | #if defined(CONFIG_UNIX) | |
7012 | ret = sock_create_kern(&init_net, PF_UNIX, SOCK_RAW, IPPROTO_IP, | |
7013 | &ctx->ring_sock); | |
7014 | if (ret) | |
7015 | return ret; | |
7016 | #endif | |
7017 | ||
7018 | ret = get_unused_fd_flags(O_RDWR | O_CLOEXEC); | |
7019 | if (ret < 0) | |
7020 | goto err; | |
7021 | ||
7022 | file = anon_inode_getfile("[io_uring]", &io_uring_fops, ctx, | |
7023 | O_RDWR | O_CLOEXEC); | |
7024 | if (IS_ERR(file)) { | |
7025 | put_unused_fd(ret); | |
7026 | ret = PTR_ERR(file); | |
7027 | goto err; | |
7028 | } | |
7029 | ||
7030 | #if defined(CONFIG_UNIX) | |
7031 | ctx->ring_sock->file = file; | |
7032 | #endif | |
7033 | fd_install(ret, file); | |
7034 | return ret; | |
7035 | err: | |
7036 | #if defined(CONFIG_UNIX) | |
7037 | sock_release(ctx->ring_sock); | |
7038 | ctx->ring_sock = NULL; | |
7039 | #endif | |
7040 | return ret; | |
7041 | } | |
7042 | ||
7043 | static int io_uring_create(unsigned entries, struct io_uring_params *p) | |
7044 | { | |
7045 | struct user_struct *user = NULL; | |
7046 | struct io_ring_ctx *ctx; | |
7047 | bool account_mem; | |
7048 | int ret; | |
7049 | ||
8110c1a6 | 7050 | if (!entries) |
2b188cc1 | 7051 | return -EINVAL; |
8110c1a6 JA |
7052 | if (entries > IORING_MAX_ENTRIES) { |
7053 | if (!(p->flags & IORING_SETUP_CLAMP)) | |
7054 | return -EINVAL; | |
7055 | entries = IORING_MAX_ENTRIES; | |
7056 | } | |
2b188cc1 JA |
7057 | |
7058 | /* | |
7059 | * Use twice as many entries for the CQ ring. It's possible for the | |
7060 | * application to drive a higher depth than the size of the SQ ring, | |
7061 | * since the sqes are only used at submission time. This allows for | |
33a107f0 JA |
7062 | * some flexibility in overcommitting a bit. If the application has |
7063 | * set IORING_SETUP_CQSIZE, it will have passed in the desired number | |
7064 | * of CQ ring entries manually. | |
2b188cc1 JA |
7065 | */ |
7066 | p->sq_entries = roundup_pow_of_two(entries); | |
33a107f0 JA |
7067 | if (p->flags & IORING_SETUP_CQSIZE) { |
7068 | /* | |
7069 | * If IORING_SETUP_CQSIZE is set, we do the same roundup | |
7070 | * to a power-of-two, if it isn't already. We do NOT impose | |
7071 | * any cq vs sq ring sizing. | |
7072 | */ | |
8110c1a6 | 7073 | if (p->cq_entries < p->sq_entries) |
33a107f0 | 7074 | return -EINVAL; |
8110c1a6 JA |
7075 | if (p->cq_entries > IORING_MAX_CQ_ENTRIES) { |
7076 | if (!(p->flags & IORING_SETUP_CLAMP)) | |
7077 | return -EINVAL; | |
7078 | p->cq_entries = IORING_MAX_CQ_ENTRIES; | |
7079 | } | |
33a107f0 JA |
7080 | p->cq_entries = roundup_pow_of_two(p->cq_entries); |
7081 | } else { | |
7082 | p->cq_entries = 2 * p->sq_entries; | |
7083 | } | |
2b188cc1 JA |
7084 | |
7085 | user = get_uid(current_user()); | |
7086 | account_mem = !capable(CAP_IPC_LOCK); | |
7087 | ||
7088 | if (account_mem) { | |
7089 | ret = io_account_mem(user, | |
7090 | ring_pages(p->sq_entries, p->cq_entries)); | |
7091 | if (ret) { | |
7092 | free_uid(user); | |
7093 | return ret; | |
7094 | } | |
7095 | } | |
7096 | ||
7097 | ctx = io_ring_ctx_alloc(p); | |
7098 | if (!ctx) { | |
7099 | if (account_mem) | |
7100 | io_unaccount_mem(user, ring_pages(p->sq_entries, | |
7101 | p->cq_entries)); | |
7102 | free_uid(user); | |
7103 | return -ENOMEM; | |
7104 | } | |
7105 | ctx->compat = in_compat_syscall(); | |
7106 | ctx->account_mem = account_mem; | |
7107 | ctx->user = user; | |
0b8c0ec7 | 7108 | ctx->creds = get_current_cred(); |
2b188cc1 JA |
7109 | |
7110 | ret = io_allocate_scq_urings(ctx, p); | |
7111 | if (ret) | |
7112 | goto err; | |
7113 | ||
6c271ce2 | 7114 | ret = io_sq_offload_start(ctx, p); |
2b188cc1 JA |
7115 | if (ret) |
7116 | goto err; | |
7117 | ||
2b188cc1 | 7118 | memset(&p->sq_off, 0, sizeof(p->sq_off)); |
75b28aff HV |
7119 | p->sq_off.head = offsetof(struct io_rings, sq.head); |
7120 | p->sq_off.tail = offsetof(struct io_rings, sq.tail); | |
7121 | p->sq_off.ring_mask = offsetof(struct io_rings, sq_ring_mask); | |
7122 | p->sq_off.ring_entries = offsetof(struct io_rings, sq_ring_entries); | |
7123 | p->sq_off.flags = offsetof(struct io_rings, sq_flags); | |
7124 | p->sq_off.dropped = offsetof(struct io_rings, sq_dropped); | |
7125 | p->sq_off.array = (char *)ctx->sq_array - (char *)ctx->rings; | |
2b188cc1 JA |
7126 | |
7127 | memset(&p->cq_off, 0, sizeof(p->cq_off)); | |
75b28aff HV |
7128 | p->cq_off.head = offsetof(struct io_rings, cq.head); |
7129 | p->cq_off.tail = offsetof(struct io_rings, cq.tail); | |
7130 | p->cq_off.ring_mask = offsetof(struct io_rings, cq_ring_mask); | |
7131 | p->cq_off.ring_entries = offsetof(struct io_rings, cq_ring_entries); | |
7132 | p->cq_off.overflow = offsetof(struct io_rings, cq_overflow); | |
7133 | p->cq_off.cqes = offsetof(struct io_rings, cqes); | |
ac90f249 | 7134 | |
044c1ab3 JA |
7135 | /* |
7136 | * Install ring fd as the very last thing, so we don't risk someone | |
7137 | * having closed it before we finish setup | |
7138 | */ | |
7139 | ret = io_uring_get_fd(ctx); | |
7140 | if (ret < 0) | |
7141 | goto err; | |
7142 | ||
da8c9690 | 7143 | p->features = IORING_FEAT_SINGLE_MMAP | IORING_FEAT_NODROP | |
cccf0ee8 | 7144 | IORING_FEAT_SUBMIT_STABLE | IORING_FEAT_RW_CUR_POS | |
d7718a9d | 7145 | IORING_FEAT_CUR_PERSONALITY | IORING_FEAT_FAST_POLL; |
c826bd7a | 7146 | trace_io_uring_create(ret, ctx, p->sq_entries, p->cq_entries, p->flags); |
2b188cc1 JA |
7147 | return ret; |
7148 | err: | |
7149 | io_ring_ctx_wait_and_kill(ctx); | |
7150 | return ret; | |
7151 | } | |
7152 | ||
7153 | /* | |
7154 | * Sets up an aio uring context, and returns the fd. Applications asks for a | |
7155 | * ring size, we return the actual sq/cq ring sizes (among other things) in the | |
7156 | * params structure passed in. | |
7157 | */ | |
7158 | static long io_uring_setup(u32 entries, struct io_uring_params __user *params) | |
7159 | { | |
7160 | struct io_uring_params p; | |
7161 | long ret; | |
7162 | int i; | |
7163 | ||
7164 | if (copy_from_user(&p, params, sizeof(p))) | |
7165 | return -EFAULT; | |
7166 | for (i = 0; i < ARRAY_SIZE(p.resv); i++) { | |
7167 | if (p.resv[i]) | |
7168 | return -EINVAL; | |
7169 | } | |
7170 | ||
6c271ce2 | 7171 | if (p.flags & ~(IORING_SETUP_IOPOLL | IORING_SETUP_SQPOLL | |
8110c1a6 | 7172 | IORING_SETUP_SQ_AFF | IORING_SETUP_CQSIZE | |
24369c2e | 7173 | IORING_SETUP_CLAMP | IORING_SETUP_ATTACH_WQ)) |
2b188cc1 JA |
7174 | return -EINVAL; |
7175 | ||
7176 | ret = io_uring_create(entries, &p); | |
7177 | if (ret < 0) | |
7178 | return ret; | |
7179 | ||
7180 | if (copy_to_user(params, &p, sizeof(p))) | |
7181 | return -EFAULT; | |
7182 | ||
7183 | return ret; | |
7184 | } | |
7185 | ||
7186 | SYSCALL_DEFINE2(io_uring_setup, u32, entries, | |
7187 | struct io_uring_params __user *, params) | |
7188 | { | |
7189 | return io_uring_setup(entries, params); | |
7190 | } | |
7191 | ||
66f4af93 JA |
7192 | static int io_probe(struct io_ring_ctx *ctx, void __user *arg, unsigned nr_args) |
7193 | { | |
7194 | struct io_uring_probe *p; | |
7195 | size_t size; | |
7196 | int i, ret; | |
7197 | ||
7198 | size = struct_size(p, ops, nr_args); | |
7199 | if (size == SIZE_MAX) | |
7200 | return -EOVERFLOW; | |
7201 | p = kzalloc(size, GFP_KERNEL); | |
7202 | if (!p) | |
7203 | return -ENOMEM; | |
7204 | ||
7205 | ret = -EFAULT; | |
7206 | if (copy_from_user(p, arg, size)) | |
7207 | goto out; | |
7208 | ret = -EINVAL; | |
7209 | if (memchr_inv(p, 0, size)) | |
7210 | goto out; | |
7211 | ||
7212 | p->last_op = IORING_OP_LAST - 1; | |
7213 | if (nr_args > IORING_OP_LAST) | |
7214 | nr_args = IORING_OP_LAST; | |
7215 | ||
7216 | for (i = 0; i < nr_args; i++) { | |
7217 | p->ops[i].op = i; | |
7218 | if (!io_op_defs[i].not_supported) | |
7219 | p->ops[i].flags = IO_URING_OP_SUPPORTED; | |
7220 | } | |
7221 | p->ops_len = i; | |
7222 | ||
7223 | ret = 0; | |
7224 | if (copy_to_user(arg, p, size)) | |
7225 | ret = -EFAULT; | |
7226 | out: | |
7227 | kfree(p); | |
7228 | return ret; | |
7229 | } | |
7230 | ||
071698e1 JA |
7231 | static int io_register_personality(struct io_ring_ctx *ctx) |
7232 | { | |
7233 | const struct cred *creds = get_current_cred(); | |
7234 | int id; | |
7235 | ||
7236 | id = idr_alloc_cyclic(&ctx->personality_idr, (void *) creds, 1, | |
7237 | USHRT_MAX, GFP_KERNEL); | |
7238 | if (id < 0) | |
7239 | put_cred(creds); | |
7240 | return id; | |
7241 | } | |
7242 | ||
7243 | static int io_unregister_personality(struct io_ring_ctx *ctx, unsigned id) | |
7244 | { | |
7245 | const struct cred *old_creds; | |
7246 | ||
7247 | old_creds = idr_remove(&ctx->personality_idr, id); | |
7248 | if (old_creds) { | |
7249 | put_cred(old_creds); | |
7250 | return 0; | |
7251 | } | |
7252 | ||
7253 | return -EINVAL; | |
7254 | } | |
7255 | ||
7256 | static bool io_register_op_must_quiesce(int op) | |
7257 | { | |
7258 | switch (op) { | |
7259 | case IORING_UNREGISTER_FILES: | |
7260 | case IORING_REGISTER_FILES_UPDATE: | |
7261 | case IORING_REGISTER_PROBE: | |
7262 | case IORING_REGISTER_PERSONALITY: | |
7263 | case IORING_UNREGISTER_PERSONALITY: | |
7264 | return false; | |
7265 | default: | |
7266 | return true; | |
7267 | } | |
7268 | } | |
7269 | ||
edafccee JA |
7270 | static int __io_uring_register(struct io_ring_ctx *ctx, unsigned opcode, |
7271 | void __user *arg, unsigned nr_args) | |
b19062a5 JA |
7272 | __releases(ctx->uring_lock) |
7273 | __acquires(ctx->uring_lock) | |
edafccee JA |
7274 | { |
7275 | int ret; | |
7276 | ||
35fa71a0 JA |
7277 | /* |
7278 | * We're inside the ring mutex, if the ref is already dying, then | |
7279 | * someone else killed the ctx or is already going through | |
7280 | * io_uring_register(). | |
7281 | */ | |
7282 | if (percpu_ref_is_dying(&ctx->refs)) | |
7283 | return -ENXIO; | |
7284 | ||
071698e1 | 7285 | if (io_register_op_must_quiesce(opcode)) { |
05f3fb3c | 7286 | percpu_ref_kill(&ctx->refs); |
b19062a5 | 7287 | |
05f3fb3c JA |
7288 | /* |
7289 | * Drop uring mutex before waiting for references to exit. If | |
7290 | * another thread is currently inside io_uring_enter() it might | |
7291 | * need to grab the uring_lock to make progress. If we hold it | |
7292 | * here across the drain wait, then we can deadlock. It's safe | |
7293 | * to drop the mutex here, since no new references will come in | |
7294 | * after we've killed the percpu ref. | |
7295 | */ | |
7296 | mutex_unlock(&ctx->uring_lock); | |
c150368b | 7297 | ret = wait_for_completion_interruptible(&ctx->completions[0]); |
05f3fb3c | 7298 | mutex_lock(&ctx->uring_lock); |
c150368b JA |
7299 | if (ret) { |
7300 | percpu_ref_resurrect(&ctx->refs); | |
7301 | ret = -EINTR; | |
7302 | goto out; | |
7303 | } | |
05f3fb3c | 7304 | } |
edafccee JA |
7305 | |
7306 | switch (opcode) { | |
7307 | case IORING_REGISTER_BUFFERS: | |
7308 | ret = io_sqe_buffer_register(ctx, arg, nr_args); | |
7309 | break; | |
7310 | case IORING_UNREGISTER_BUFFERS: | |
7311 | ret = -EINVAL; | |
7312 | if (arg || nr_args) | |
7313 | break; | |
7314 | ret = io_sqe_buffer_unregister(ctx); | |
7315 | break; | |
6b06314c JA |
7316 | case IORING_REGISTER_FILES: |
7317 | ret = io_sqe_files_register(ctx, arg, nr_args); | |
7318 | break; | |
7319 | case IORING_UNREGISTER_FILES: | |
7320 | ret = -EINVAL; | |
7321 | if (arg || nr_args) | |
7322 | break; | |
7323 | ret = io_sqe_files_unregister(ctx); | |
7324 | break; | |
c3a31e60 JA |
7325 | case IORING_REGISTER_FILES_UPDATE: |
7326 | ret = io_sqe_files_update(ctx, arg, nr_args); | |
7327 | break; | |
9b402849 | 7328 | case IORING_REGISTER_EVENTFD: |
f2842ab5 | 7329 | case IORING_REGISTER_EVENTFD_ASYNC: |
9b402849 JA |
7330 | ret = -EINVAL; |
7331 | if (nr_args != 1) | |
7332 | break; | |
7333 | ret = io_eventfd_register(ctx, arg); | |
f2842ab5 JA |
7334 | if (ret) |
7335 | break; | |
7336 | if (opcode == IORING_REGISTER_EVENTFD_ASYNC) | |
7337 | ctx->eventfd_async = 1; | |
7338 | else | |
7339 | ctx->eventfd_async = 0; | |
9b402849 JA |
7340 | break; |
7341 | case IORING_UNREGISTER_EVENTFD: | |
7342 | ret = -EINVAL; | |
7343 | if (arg || nr_args) | |
7344 | break; | |
7345 | ret = io_eventfd_unregister(ctx); | |
7346 | break; | |
66f4af93 JA |
7347 | case IORING_REGISTER_PROBE: |
7348 | ret = -EINVAL; | |
7349 | if (!arg || nr_args > 256) | |
7350 | break; | |
7351 | ret = io_probe(ctx, arg, nr_args); | |
7352 | break; | |
071698e1 JA |
7353 | case IORING_REGISTER_PERSONALITY: |
7354 | ret = -EINVAL; | |
7355 | if (arg || nr_args) | |
7356 | break; | |
7357 | ret = io_register_personality(ctx); | |
7358 | break; | |
7359 | case IORING_UNREGISTER_PERSONALITY: | |
7360 | ret = -EINVAL; | |
7361 | if (arg) | |
7362 | break; | |
7363 | ret = io_unregister_personality(ctx, nr_args); | |
7364 | break; | |
edafccee JA |
7365 | default: |
7366 | ret = -EINVAL; | |
7367 | break; | |
7368 | } | |
7369 | ||
071698e1 | 7370 | if (io_register_op_must_quiesce(opcode)) { |
05f3fb3c | 7371 | /* bring the ctx back to life */ |
05f3fb3c | 7372 | percpu_ref_reinit(&ctx->refs); |
c150368b JA |
7373 | out: |
7374 | reinit_completion(&ctx->completions[0]); | |
05f3fb3c | 7375 | } |
edafccee JA |
7376 | return ret; |
7377 | } | |
7378 | ||
7379 | SYSCALL_DEFINE4(io_uring_register, unsigned int, fd, unsigned int, opcode, | |
7380 | void __user *, arg, unsigned int, nr_args) | |
7381 | { | |
7382 | struct io_ring_ctx *ctx; | |
7383 | long ret = -EBADF; | |
7384 | struct fd f; | |
7385 | ||
7386 | f = fdget(fd); | |
7387 | if (!f.file) | |
7388 | return -EBADF; | |
7389 | ||
7390 | ret = -EOPNOTSUPP; | |
7391 | if (f.file->f_op != &io_uring_fops) | |
7392 | goto out_fput; | |
7393 | ||
7394 | ctx = f.file->private_data; | |
7395 | ||
7396 | mutex_lock(&ctx->uring_lock); | |
7397 | ret = __io_uring_register(ctx, opcode, arg, nr_args); | |
7398 | mutex_unlock(&ctx->uring_lock); | |
c826bd7a DD |
7399 | trace_io_uring_register(ctx, opcode, ctx->nr_user_files, ctx->nr_user_bufs, |
7400 | ctx->cq_ev_fd != NULL, ret); | |
edafccee JA |
7401 | out_fput: |
7402 | fdput(f); | |
7403 | return ret; | |
7404 | } | |
7405 | ||
2b188cc1 JA |
7406 | static int __init io_uring_init(void) |
7407 | { | |
d7f62e82 SM |
7408 | #define __BUILD_BUG_VERIFY_ELEMENT(stype, eoffset, etype, ename) do { \ |
7409 | BUILD_BUG_ON(offsetof(stype, ename) != eoffset); \ | |
7410 | BUILD_BUG_ON(sizeof(etype) != sizeof_field(stype, ename)); \ | |
7411 | } while (0) | |
7412 | ||
7413 | #define BUILD_BUG_SQE_ELEM(eoffset, etype, ename) \ | |
7414 | __BUILD_BUG_VERIFY_ELEMENT(struct io_uring_sqe, eoffset, etype, ename) | |
7415 | BUILD_BUG_ON(sizeof(struct io_uring_sqe) != 64); | |
7416 | BUILD_BUG_SQE_ELEM(0, __u8, opcode); | |
7417 | BUILD_BUG_SQE_ELEM(1, __u8, flags); | |
7418 | BUILD_BUG_SQE_ELEM(2, __u16, ioprio); | |
7419 | BUILD_BUG_SQE_ELEM(4, __s32, fd); | |
7420 | BUILD_BUG_SQE_ELEM(8, __u64, off); | |
7421 | BUILD_BUG_SQE_ELEM(8, __u64, addr2); | |
7422 | BUILD_BUG_SQE_ELEM(16, __u64, addr); | |
7d67af2c | 7423 | BUILD_BUG_SQE_ELEM(16, __u64, splice_off_in); |
d7f62e82 SM |
7424 | BUILD_BUG_SQE_ELEM(24, __u32, len); |
7425 | BUILD_BUG_SQE_ELEM(28, __kernel_rwf_t, rw_flags); | |
7426 | BUILD_BUG_SQE_ELEM(28, /* compat */ int, rw_flags); | |
7427 | BUILD_BUG_SQE_ELEM(28, /* compat */ __u32, rw_flags); | |
7428 | BUILD_BUG_SQE_ELEM(28, __u32, fsync_flags); | |
7429 | BUILD_BUG_SQE_ELEM(28, __u16, poll_events); | |
7430 | BUILD_BUG_SQE_ELEM(28, __u32, sync_range_flags); | |
7431 | BUILD_BUG_SQE_ELEM(28, __u32, msg_flags); | |
7432 | BUILD_BUG_SQE_ELEM(28, __u32, timeout_flags); | |
7433 | BUILD_BUG_SQE_ELEM(28, __u32, accept_flags); | |
7434 | BUILD_BUG_SQE_ELEM(28, __u32, cancel_flags); | |
7435 | BUILD_BUG_SQE_ELEM(28, __u32, open_flags); | |
7436 | BUILD_BUG_SQE_ELEM(28, __u32, statx_flags); | |
7437 | BUILD_BUG_SQE_ELEM(28, __u32, fadvise_advice); | |
7d67af2c | 7438 | BUILD_BUG_SQE_ELEM(28, __u32, splice_flags); |
d7f62e82 SM |
7439 | BUILD_BUG_SQE_ELEM(32, __u64, user_data); |
7440 | BUILD_BUG_SQE_ELEM(40, __u16, buf_index); | |
7441 | BUILD_BUG_SQE_ELEM(42, __u16, personality); | |
7d67af2c | 7442 | BUILD_BUG_SQE_ELEM(44, __s32, splice_fd_in); |
d7f62e82 | 7443 | |
d3656344 | 7444 | BUILD_BUG_ON(ARRAY_SIZE(io_op_defs) != IORING_OP_LAST); |
2b188cc1 JA |
7445 | req_cachep = KMEM_CACHE(io_kiocb, SLAB_HWCACHE_ALIGN | SLAB_PANIC); |
7446 | return 0; | |
7447 | }; | |
7448 | __initcall(io_uring_init); |