]>
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> | |
52de1fe1 | 47 | #include <net/compat.h> |
2b188cc1 JA |
48 | #include <linux/refcount.h> |
49 | #include <linux/uio.h> | |
6b47ee6e | 50 | #include <linux/bits.h> |
2b188cc1 JA |
51 | |
52 | #include <linux/sched/signal.h> | |
53 | #include <linux/fs.h> | |
54 | #include <linux/file.h> | |
55 | #include <linux/fdtable.h> | |
56 | #include <linux/mm.h> | |
57 | #include <linux/mman.h> | |
58 | #include <linux/mmu_context.h> | |
59 | #include <linux/percpu.h> | |
60 | #include <linux/slab.h> | |
6c271ce2 | 61 | #include <linux/kthread.h> |
2b188cc1 | 62 | #include <linux/blkdev.h> |
edafccee | 63 | #include <linux/bvec.h> |
2b188cc1 JA |
64 | #include <linux/net.h> |
65 | #include <net/sock.h> | |
66 | #include <net/af_unix.h> | |
6b06314c | 67 | #include <net/scm.h> |
2b188cc1 JA |
68 | #include <linux/anon_inodes.h> |
69 | #include <linux/sched/mm.h> | |
70 | #include <linux/uaccess.h> | |
71 | #include <linux/nospec.h> | |
edafccee JA |
72 | #include <linux/sizes.h> |
73 | #include <linux/hugetlb.h> | |
aa4c3967 | 74 | #include <linux/highmem.h> |
15b71abe JA |
75 | #include <linux/namei.h> |
76 | #include <linux/fsnotify.h> | |
4840e418 | 77 | #include <linux/fadvise.h> |
3e4827b0 | 78 | #include <linux/eventpoll.h> |
ff002b30 | 79 | #include <linux/fs_struct.h> |
7d67af2c | 80 | #include <linux/splice.h> |
b41e9852 | 81 | #include <linux/task_work.h> |
2b188cc1 | 82 | |
c826bd7a DD |
83 | #define CREATE_TRACE_POINTS |
84 | #include <trace/events/io_uring.h> | |
85 | ||
2b188cc1 JA |
86 | #include <uapi/linux/io_uring.h> |
87 | ||
88 | #include "internal.h" | |
561fb04a | 89 | #include "io-wq.h" |
2b188cc1 | 90 | |
5277deaa | 91 | #define IORING_MAX_ENTRIES 32768 |
33a107f0 | 92 | #define IORING_MAX_CQ_ENTRIES (2 * IORING_MAX_ENTRIES) |
65e19f54 JA |
93 | |
94 | /* | |
95 | * Shift of 9 is 512 entries, or exactly one page on 64-bit archs | |
96 | */ | |
97 | #define IORING_FILE_TABLE_SHIFT 9 | |
98 | #define IORING_MAX_FILES_TABLE (1U << IORING_FILE_TABLE_SHIFT) | |
99 | #define IORING_FILE_TABLE_MASK (IORING_MAX_FILES_TABLE - 1) | |
100 | #define IORING_MAX_FIXED_FILES (64 * IORING_MAX_FILES_TABLE) | |
2b188cc1 JA |
101 | |
102 | struct io_uring { | |
103 | u32 head ____cacheline_aligned_in_smp; | |
104 | u32 tail ____cacheline_aligned_in_smp; | |
105 | }; | |
106 | ||
1e84b97b | 107 | /* |
75b28aff HV |
108 | * This data is shared with the application through the mmap at offsets |
109 | * IORING_OFF_SQ_RING and IORING_OFF_CQ_RING. | |
1e84b97b SB |
110 | * |
111 | * The offsets to the member fields are published through struct | |
112 | * io_sqring_offsets when calling io_uring_setup. | |
113 | */ | |
75b28aff | 114 | struct io_rings { |
1e84b97b SB |
115 | /* |
116 | * Head and tail offsets into the ring; the offsets need to be | |
117 | * masked to get valid indices. | |
118 | * | |
75b28aff HV |
119 | * The kernel controls head of the sq ring and the tail of the cq ring, |
120 | * and the application controls tail of the sq ring and the head of the | |
121 | * cq ring. | |
1e84b97b | 122 | */ |
75b28aff | 123 | struct io_uring sq, cq; |
1e84b97b | 124 | /* |
75b28aff | 125 | * Bitmasks to apply to head and tail offsets (constant, equals |
1e84b97b SB |
126 | * ring_entries - 1) |
127 | */ | |
75b28aff HV |
128 | u32 sq_ring_mask, cq_ring_mask; |
129 | /* Ring sizes (constant, power of 2) */ | |
130 | u32 sq_ring_entries, cq_ring_entries; | |
1e84b97b SB |
131 | /* |
132 | * Number of invalid entries dropped by the kernel due to | |
133 | * invalid index stored in array | |
134 | * | |
135 | * Written by the kernel, shouldn't be modified by the | |
136 | * application (i.e. get number of "new events" by comparing to | |
137 | * cached value). | |
138 | * | |
139 | * After a new SQ head value was read by the application this | |
140 | * counter includes all submissions that were dropped reaching | |
141 | * the new SQ head (and possibly more). | |
142 | */ | |
75b28aff | 143 | u32 sq_dropped; |
1e84b97b SB |
144 | /* |
145 | * Runtime flags | |
146 | * | |
147 | * Written by the kernel, shouldn't be modified by the | |
148 | * application. | |
149 | * | |
150 | * The application needs a full memory barrier before checking | |
151 | * for IORING_SQ_NEED_WAKEUP after updating the sq tail. | |
152 | */ | |
75b28aff | 153 | u32 sq_flags; |
1e84b97b SB |
154 | /* |
155 | * Number of completion events lost because the queue was full; | |
156 | * this should be avoided by the application by making sure | |
0b4295b5 | 157 | * there are not more requests pending than there is space in |
1e84b97b SB |
158 | * the completion queue. |
159 | * | |
160 | * Written by the kernel, shouldn't be modified by the | |
161 | * application (i.e. get number of "new events" by comparing to | |
162 | * cached value). | |
163 | * | |
164 | * As completion events come in out of order this counter is not | |
165 | * ordered with any other data. | |
166 | */ | |
75b28aff | 167 | u32 cq_overflow; |
1e84b97b SB |
168 | /* |
169 | * Ring buffer of completion events. | |
170 | * | |
171 | * The kernel writes completion events fresh every time they are | |
172 | * produced, so the application is allowed to modify pending | |
173 | * entries. | |
174 | */ | |
75b28aff | 175 | struct io_uring_cqe cqes[] ____cacheline_aligned_in_smp; |
2b188cc1 JA |
176 | }; |
177 | ||
edafccee JA |
178 | struct io_mapped_ubuf { |
179 | u64 ubuf; | |
180 | size_t len; | |
181 | struct bio_vec *bvec; | |
182 | unsigned int nr_bvecs; | |
183 | }; | |
184 | ||
65e19f54 JA |
185 | struct fixed_file_table { |
186 | struct file **files; | |
31b51510 JA |
187 | }; |
188 | ||
05589553 XW |
189 | struct fixed_file_ref_node { |
190 | struct percpu_ref refs; | |
191 | struct list_head node; | |
192 | struct list_head file_list; | |
193 | struct fixed_file_data *file_data; | |
194 | struct work_struct work; | |
195 | }; | |
196 | ||
05f3fb3c JA |
197 | struct fixed_file_data { |
198 | struct fixed_file_table *table; | |
199 | struct io_ring_ctx *ctx; | |
200 | ||
05589553 | 201 | struct percpu_ref *cur_refs; |
05f3fb3c | 202 | struct percpu_ref refs; |
05f3fb3c | 203 | struct completion done; |
05589553 XW |
204 | struct list_head ref_list; |
205 | spinlock_t lock; | |
05f3fb3c JA |
206 | }; |
207 | ||
5a2e745d JA |
208 | struct io_buffer { |
209 | struct list_head list; | |
210 | __u64 addr; | |
211 | __s32 len; | |
212 | __u16 bid; | |
213 | }; | |
214 | ||
2b188cc1 JA |
215 | struct io_ring_ctx { |
216 | struct { | |
217 | struct percpu_ref refs; | |
218 | } ____cacheline_aligned_in_smp; | |
219 | ||
220 | struct { | |
221 | unsigned int flags; | |
e1d85334 RD |
222 | unsigned int compat: 1; |
223 | unsigned int account_mem: 1; | |
224 | unsigned int cq_overflow_flushed: 1; | |
225 | unsigned int drain_next: 1; | |
226 | unsigned int eventfd_async: 1; | |
2b188cc1 | 227 | |
75b28aff HV |
228 | /* |
229 | * Ring buffer of indices into array of io_uring_sqe, which is | |
230 | * mmapped by the application using the IORING_OFF_SQES offset. | |
231 | * | |
232 | * This indirection could e.g. be used to assign fixed | |
233 | * io_uring_sqe entries to operations and only submit them to | |
234 | * the queue when needed. | |
235 | * | |
236 | * The kernel modifies neither the indices array nor the entries | |
237 | * array. | |
238 | */ | |
239 | u32 *sq_array; | |
2b188cc1 JA |
240 | unsigned cached_sq_head; |
241 | unsigned sq_entries; | |
242 | unsigned sq_mask; | |
6c271ce2 | 243 | unsigned sq_thread_idle; |
498ccd9e | 244 | unsigned cached_sq_dropped; |
206aefde | 245 | atomic_t cached_cq_overflow; |
ad3eb2c8 | 246 | unsigned long sq_check_overflow; |
de0617e4 JA |
247 | |
248 | struct list_head defer_list; | |
5262f567 | 249 | struct list_head timeout_list; |
1d7bb1d5 | 250 | struct list_head cq_overflow_list; |
fcb323cc JA |
251 | |
252 | wait_queue_head_t inflight_wait; | |
ad3eb2c8 | 253 | struct io_uring_sqe *sq_sqes; |
2b188cc1 JA |
254 | } ____cacheline_aligned_in_smp; |
255 | ||
206aefde JA |
256 | struct io_rings *rings; |
257 | ||
2b188cc1 | 258 | /* IO offload */ |
561fb04a | 259 | struct io_wq *io_wq; |
6c271ce2 | 260 | struct task_struct *sqo_thread; /* if using sq thread polling */ |
2b188cc1 | 261 | struct mm_struct *sqo_mm; |
6c271ce2 | 262 | wait_queue_head_t sqo_wait; |
75b28aff | 263 | |
6b06314c JA |
264 | /* |
265 | * If used, fixed file set. Writers must ensure that ->refs is dead, | |
266 | * readers must ensure that ->refs is alive as long as the file* is | |
267 | * used. Only updated through io_uring_register(2). | |
268 | */ | |
05f3fb3c | 269 | struct fixed_file_data *file_data; |
6b06314c | 270 | unsigned nr_user_files; |
b14cca0c PB |
271 | int ring_fd; |
272 | struct file *ring_file; | |
6b06314c | 273 | |
edafccee JA |
274 | /* if used, fixed mapped user buffers */ |
275 | unsigned nr_user_bufs; | |
276 | struct io_mapped_ubuf *user_bufs; | |
277 | ||
2b188cc1 JA |
278 | struct user_struct *user; |
279 | ||
0b8c0ec7 | 280 | const struct cred *creds; |
181e448d | 281 | |
0f158b4c JA |
282 | struct completion ref_comp; |
283 | struct completion sq_thread_comp; | |
206aefde | 284 | |
0ddf92e8 JA |
285 | /* if all else fails... */ |
286 | struct io_kiocb *fallback_req; | |
287 | ||
206aefde JA |
288 | #if defined(CONFIG_UNIX) |
289 | struct socket *ring_sock; | |
290 | #endif | |
291 | ||
5a2e745d JA |
292 | struct idr io_buffer_idr; |
293 | ||
071698e1 JA |
294 | struct idr personality_idr; |
295 | ||
206aefde JA |
296 | struct { |
297 | unsigned cached_cq_tail; | |
298 | unsigned cq_entries; | |
299 | unsigned cq_mask; | |
300 | atomic_t cq_timeouts; | |
ad3eb2c8 | 301 | unsigned long cq_check_overflow; |
206aefde JA |
302 | struct wait_queue_head cq_wait; |
303 | struct fasync_struct *cq_fasync; | |
304 | struct eventfd_ctx *cq_ev_fd; | |
305 | } ____cacheline_aligned_in_smp; | |
2b188cc1 JA |
306 | |
307 | struct { | |
308 | struct mutex uring_lock; | |
309 | wait_queue_head_t wait; | |
310 | } ____cacheline_aligned_in_smp; | |
311 | ||
312 | struct { | |
313 | spinlock_t completion_lock; | |
e94f141b | 314 | |
def596e9 JA |
315 | /* |
316 | * ->poll_list is protected by the ctx->uring_lock for | |
317 | * io_uring instances that don't use IORING_SETUP_SQPOLL. | |
318 | * For SQPOLL, only the single threaded io_sq_thread() will | |
319 | * manipulate the list, hence no extra locking is needed there. | |
320 | */ | |
321 | struct list_head poll_list; | |
78076bb6 JA |
322 | struct hlist_head *cancel_hash; |
323 | unsigned cancel_hash_bits; | |
e94f141b | 324 | bool poll_multi_file; |
31b51510 | 325 | |
fcb323cc JA |
326 | spinlock_t inflight_lock; |
327 | struct list_head inflight_list; | |
2b188cc1 | 328 | } ____cacheline_aligned_in_smp; |
85faa7b8 JA |
329 | |
330 | struct work_struct exit_work; | |
2b188cc1 JA |
331 | }; |
332 | ||
09bb8394 JA |
333 | /* |
334 | * First field must be the file pointer in all the | |
335 | * iocb unions! See also 'struct kiocb' in <linux/fs.h> | |
336 | */ | |
221c5eb2 JA |
337 | struct io_poll_iocb { |
338 | struct file *file; | |
0969e783 JA |
339 | union { |
340 | struct wait_queue_head *head; | |
341 | u64 addr; | |
342 | }; | |
221c5eb2 | 343 | __poll_t events; |
8c838788 | 344 | bool done; |
221c5eb2 | 345 | bool canceled; |
392edb45 | 346 | struct wait_queue_entry wait; |
221c5eb2 JA |
347 | }; |
348 | ||
b5dba59e JA |
349 | struct io_close { |
350 | struct file *file; | |
351 | struct file *put_file; | |
352 | int fd; | |
353 | }; | |
354 | ||
ad8a48ac JA |
355 | struct io_timeout_data { |
356 | struct io_kiocb *req; | |
357 | struct hrtimer timer; | |
358 | struct timespec64 ts; | |
359 | enum hrtimer_mode mode; | |
360 | }; | |
361 | ||
8ed8d3c3 JA |
362 | struct io_accept { |
363 | struct file *file; | |
364 | struct sockaddr __user *addr; | |
365 | int __user *addr_len; | |
366 | int flags; | |
09952e3e | 367 | unsigned long nofile; |
8ed8d3c3 JA |
368 | }; |
369 | ||
370 | struct io_sync { | |
371 | struct file *file; | |
372 | loff_t len; | |
373 | loff_t off; | |
374 | int flags; | |
d63d1b5e | 375 | int mode; |
8ed8d3c3 JA |
376 | }; |
377 | ||
fbf23849 JA |
378 | struct io_cancel { |
379 | struct file *file; | |
380 | u64 addr; | |
381 | }; | |
382 | ||
b29472ee JA |
383 | struct io_timeout { |
384 | struct file *file; | |
385 | u64 addr; | |
386 | int flags; | |
b55ce732 | 387 | u32 count; |
b29472ee JA |
388 | }; |
389 | ||
9adbd45d JA |
390 | struct io_rw { |
391 | /* NOTE: kiocb has the file as the first member, so don't do it here */ | |
392 | struct kiocb kiocb; | |
393 | u64 addr; | |
394 | u64 len; | |
395 | }; | |
396 | ||
3fbb51c1 JA |
397 | struct io_connect { |
398 | struct file *file; | |
399 | struct sockaddr __user *addr; | |
400 | int addr_len; | |
401 | }; | |
402 | ||
e47293fd JA |
403 | struct io_sr_msg { |
404 | struct file *file; | |
fddaface JA |
405 | union { |
406 | struct user_msghdr __user *msg; | |
407 | void __user *buf; | |
408 | }; | |
e47293fd | 409 | int msg_flags; |
bcda7baa | 410 | int bgid; |
fddaface | 411 | size_t len; |
bcda7baa | 412 | struct io_buffer *kbuf; |
e47293fd JA |
413 | }; |
414 | ||
15b71abe JA |
415 | struct io_open { |
416 | struct file *file; | |
417 | int dfd; | |
eddc7ef5 | 418 | union { |
eddc7ef5 JA |
419 | unsigned mask; |
420 | }; | |
15b71abe | 421 | struct filename *filename; |
eddc7ef5 | 422 | struct statx __user *buffer; |
c12cedf2 | 423 | struct open_how how; |
4022e7af | 424 | unsigned long nofile; |
15b71abe JA |
425 | }; |
426 | ||
05f3fb3c JA |
427 | struct io_files_update { |
428 | struct file *file; | |
429 | u64 arg; | |
430 | u32 nr_args; | |
431 | u32 offset; | |
432 | }; | |
433 | ||
4840e418 JA |
434 | struct io_fadvise { |
435 | struct file *file; | |
436 | u64 offset; | |
437 | u32 len; | |
438 | u32 advice; | |
439 | }; | |
440 | ||
c1ca757b JA |
441 | struct io_madvise { |
442 | struct file *file; | |
443 | u64 addr; | |
444 | u32 len; | |
445 | u32 advice; | |
446 | }; | |
447 | ||
3e4827b0 JA |
448 | struct io_epoll { |
449 | struct file *file; | |
450 | int epfd; | |
451 | int op; | |
452 | int fd; | |
453 | struct epoll_event event; | |
e47293fd JA |
454 | }; |
455 | ||
7d67af2c PB |
456 | struct io_splice { |
457 | struct file *file_out; | |
458 | struct file *file_in; | |
459 | loff_t off_out; | |
460 | loff_t off_in; | |
461 | u64 len; | |
462 | unsigned int flags; | |
463 | }; | |
464 | ||
ddf0322d JA |
465 | struct io_provide_buf { |
466 | struct file *file; | |
467 | __u64 addr; | |
468 | __s32 len; | |
469 | __u32 bgid; | |
470 | __u16 nbufs; | |
471 | __u16 bid; | |
472 | }; | |
473 | ||
f499a021 JA |
474 | struct io_async_connect { |
475 | struct sockaddr_storage address; | |
476 | }; | |
477 | ||
03b1230c JA |
478 | struct io_async_msghdr { |
479 | struct iovec fast_iov[UIO_FASTIOV]; | |
480 | struct iovec *iov; | |
481 | struct sockaddr __user *uaddr; | |
482 | struct msghdr msg; | |
b537916c | 483 | struct sockaddr_storage addr; |
03b1230c JA |
484 | }; |
485 | ||
f67676d1 JA |
486 | struct io_async_rw { |
487 | struct iovec fast_iov[UIO_FASTIOV]; | |
488 | struct iovec *iov; | |
489 | ssize_t nr_segs; | |
490 | ssize_t size; | |
491 | }; | |
492 | ||
1a6b74fc | 493 | struct io_async_ctx { |
f67676d1 JA |
494 | union { |
495 | struct io_async_rw rw; | |
03b1230c | 496 | struct io_async_msghdr msg; |
f499a021 | 497 | struct io_async_connect connect; |
2d28390a | 498 | struct io_timeout_data timeout; |
f67676d1 | 499 | }; |
1a6b74fc JA |
500 | }; |
501 | ||
6b47ee6e PB |
502 | enum { |
503 | REQ_F_FIXED_FILE_BIT = IOSQE_FIXED_FILE_BIT, | |
504 | REQ_F_IO_DRAIN_BIT = IOSQE_IO_DRAIN_BIT, | |
505 | REQ_F_LINK_BIT = IOSQE_IO_LINK_BIT, | |
506 | REQ_F_HARDLINK_BIT = IOSQE_IO_HARDLINK_BIT, | |
507 | REQ_F_FORCE_ASYNC_BIT = IOSQE_ASYNC_BIT, | |
bcda7baa | 508 | REQ_F_BUFFER_SELECT_BIT = IOSQE_BUFFER_SELECT_BIT, |
6b47ee6e | 509 | |
dea3b49c | 510 | REQ_F_LINK_HEAD_BIT, |
6b47ee6e PB |
511 | REQ_F_LINK_NEXT_BIT, |
512 | REQ_F_FAIL_LINK_BIT, | |
513 | REQ_F_INFLIGHT_BIT, | |
514 | REQ_F_CUR_POS_BIT, | |
515 | REQ_F_NOWAIT_BIT, | |
516 | REQ_F_IOPOLL_COMPLETED_BIT, | |
517 | REQ_F_LINK_TIMEOUT_BIT, | |
518 | REQ_F_TIMEOUT_BIT, | |
519 | REQ_F_ISREG_BIT, | |
520 | REQ_F_MUST_PUNT_BIT, | |
521 | REQ_F_TIMEOUT_NOSEQ_BIT, | |
522 | REQ_F_COMP_LOCKED_BIT, | |
99bc4c38 | 523 | REQ_F_NEED_CLEANUP_BIT, |
2ca10259 | 524 | REQ_F_OVERFLOW_BIT, |
d7718a9d | 525 | REQ_F_POLLED_BIT, |
bcda7baa | 526 | REQ_F_BUFFER_SELECTED_BIT, |
5b0bbee4 | 527 | REQ_F_NO_FILE_TABLE_BIT, |
84557871 JA |
528 | |
529 | /* not a real bit, just to check we're not overflowing the space */ | |
530 | __REQ_F_LAST_BIT, | |
6b47ee6e PB |
531 | }; |
532 | ||
533 | enum { | |
534 | /* ctx owns file */ | |
535 | REQ_F_FIXED_FILE = BIT(REQ_F_FIXED_FILE_BIT), | |
536 | /* drain existing IO first */ | |
537 | REQ_F_IO_DRAIN = BIT(REQ_F_IO_DRAIN_BIT), | |
538 | /* linked sqes */ | |
539 | REQ_F_LINK = BIT(REQ_F_LINK_BIT), | |
540 | /* doesn't sever on completion < 0 */ | |
541 | REQ_F_HARDLINK = BIT(REQ_F_HARDLINK_BIT), | |
542 | /* IOSQE_ASYNC */ | |
543 | REQ_F_FORCE_ASYNC = BIT(REQ_F_FORCE_ASYNC_BIT), | |
bcda7baa JA |
544 | /* IOSQE_BUFFER_SELECT */ |
545 | REQ_F_BUFFER_SELECT = BIT(REQ_F_BUFFER_SELECT_BIT), | |
6b47ee6e | 546 | |
dea3b49c PB |
547 | /* head of a link */ |
548 | REQ_F_LINK_HEAD = BIT(REQ_F_LINK_HEAD_BIT), | |
6b47ee6e PB |
549 | /* already grabbed next link */ |
550 | REQ_F_LINK_NEXT = BIT(REQ_F_LINK_NEXT_BIT), | |
551 | /* fail rest of links */ | |
552 | REQ_F_FAIL_LINK = BIT(REQ_F_FAIL_LINK_BIT), | |
553 | /* on inflight list */ | |
554 | REQ_F_INFLIGHT = BIT(REQ_F_INFLIGHT_BIT), | |
555 | /* read/write uses file position */ | |
556 | REQ_F_CUR_POS = BIT(REQ_F_CUR_POS_BIT), | |
557 | /* must not punt to workers */ | |
558 | REQ_F_NOWAIT = BIT(REQ_F_NOWAIT_BIT), | |
559 | /* polled IO has completed */ | |
560 | REQ_F_IOPOLL_COMPLETED = BIT(REQ_F_IOPOLL_COMPLETED_BIT), | |
561 | /* has linked timeout */ | |
562 | REQ_F_LINK_TIMEOUT = BIT(REQ_F_LINK_TIMEOUT_BIT), | |
563 | /* timeout request */ | |
564 | REQ_F_TIMEOUT = BIT(REQ_F_TIMEOUT_BIT), | |
565 | /* regular file */ | |
566 | REQ_F_ISREG = BIT(REQ_F_ISREG_BIT), | |
567 | /* must be punted even for NONBLOCK */ | |
568 | REQ_F_MUST_PUNT = BIT(REQ_F_MUST_PUNT_BIT), | |
569 | /* no timeout sequence */ | |
570 | REQ_F_TIMEOUT_NOSEQ = BIT(REQ_F_TIMEOUT_NOSEQ_BIT), | |
571 | /* completion under lock */ | |
572 | REQ_F_COMP_LOCKED = BIT(REQ_F_COMP_LOCKED_BIT), | |
99bc4c38 PB |
573 | /* needs cleanup */ |
574 | REQ_F_NEED_CLEANUP = BIT(REQ_F_NEED_CLEANUP_BIT), | |
2ca10259 JA |
575 | /* in overflow list */ |
576 | REQ_F_OVERFLOW = BIT(REQ_F_OVERFLOW_BIT), | |
d7718a9d JA |
577 | /* already went through poll handler */ |
578 | REQ_F_POLLED = BIT(REQ_F_POLLED_BIT), | |
bcda7baa JA |
579 | /* buffer already selected */ |
580 | REQ_F_BUFFER_SELECTED = BIT(REQ_F_BUFFER_SELECTED_BIT), | |
5b0bbee4 JA |
581 | /* doesn't need file table for this request */ |
582 | REQ_F_NO_FILE_TABLE = BIT(REQ_F_NO_FILE_TABLE_BIT), | |
d7718a9d JA |
583 | }; |
584 | ||
585 | struct async_poll { | |
586 | struct io_poll_iocb poll; | |
587 | struct io_wq_work work; | |
6b47ee6e PB |
588 | }; |
589 | ||
09bb8394 JA |
590 | /* |
591 | * NOTE! Each of the iocb union members has the file pointer | |
592 | * as the first entry in their struct definition. So you can | |
593 | * access the file pointer through any of the sub-structs, | |
594 | * or directly as just 'ki_filp' in this struct. | |
595 | */ | |
2b188cc1 | 596 | struct io_kiocb { |
221c5eb2 | 597 | union { |
09bb8394 | 598 | struct file *file; |
9adbd45d | 599 | struct io_rw rw; |
221c5eb2 | 600 | struct io_poll_iocb poll; |
8ed8d3c3 JA |
601 | struct io_accept accept; |
602 | struct io_sync sync; | |
fbf23849 | 603 | struct io_cancel cancel; |
b29472ee | 604 | struct io_timeout timeout; |
3fbb51c1 | 605 | struct io_connect connect; |
e47293fd | 606 | struct io_sr_msg sr_msg; |
15b71abe | 607 | struct io_open open; |
b5dba59e | 608 | struct io_close close; |
05f3fb3c | 609 | struct io_files_update files_update; |
4840e418 | 610 | struct io_fadvise fadvise; |
c1ca757b | 611 | struct io_madvise madvise; |
3e4827b0 | 612 | struct io_epoll epoll; |
7d67af2c | 613 | struct io_splice splice; |
ddf0322d | 614 | struct io_provide_buf pbuf; |
221c5eb2 | 615 | }; |
2b188cc1 | 616 | |
1a6b74fc | 617 | struct io_async_ctx *io; |
c398ecb3 | 618 | int cflags; |
cf6fd4bd | 619 | bool needs_fixed_file; |
d625c6ee | 620 | u8 opcode; |
2b188cc1 JA |
621 | |
622 | struct io_ring_ctx *ctx; | |
d7718a9d | 623 | struct list_head list; |
2b188cc1 | 624 | unsigned int flags; |
c16361c1 | 625 | refcount_t refs; |
3537b6a7 JA |
626 | struct task_struct *task; |
627 | unsigned long fsize; | |
2b188cc1 | 628 | u64 user_data; |
9e645e11 | 629 | u32 result; |
de0617e4 | 630 | u32 sequence; |
2b188cc1 | 631 | |
d7718a9d JA |
632 | struct list_head link_list; |
633 | ||
fcb323cc JA |
634 | struct list_head inflight_entry; |
635 | ||
05589553 XW |
636 | struct percpu_ref *fixed_file_refs; |
637 | ||
b41e9852 JA |
638 | union { |
639 | /* | |
640 | * Only commands that never go async can use the below fields, | |
d7718a9d JA |
641 | * obviously. Right now only IORING_OP_POLL_ADD uses them, and |
642 | * async armed poll handlers for regular commands. The latter | |
643 | * restore the work, if needed. | |
b41e9852 JA |
644 | */ |
645 | struct { | |
b41e9852 | 646 | struct callback_head task_work; |
d7718a9d JA |
647 | struct hlist_node hash_node; |
648 | struct async_poll *apoll; | |
b41e9852 JA |
649 | }; |
650 | struct io_wq_work work; | |
651 | }; | |
2b188cc1 JA |
652 | }; |
653 | ||
654 | #define IO_PLUG_THRESHOLD 2 | |
def596e9 | 655 | #define IO_IOPOLL_BATCH 8 |
2b188cc1 | 656 | |
9a56a232 JA |
657 | struct io_submit_state { |
658 | struct blk_plug plug; | |
659 | ||
2579f913 JA |
660 | /* |
661 | * io_kiocb alloc cache | |
662 | */ | |
663 | void *reqs[IO_IOPOLL_BATCH]; | |
6c8a3134 | 664 | unsigned int free_reqs; |
2579f913 | 665 | |
9a56a232 JA |
666 | /* |
667 | * File reference cache | |
668 | */ | |
669 | struct file *file; | |
670 | unsigned int fd; | |
671 | unsigned int has_refs; | |
672 | unsigned int used_refs; | |
673 | unsigned int ios_left; | |
674 | }; | |
675 | ||
d3656344 JA |
676 | struct io_op_def { |
677 | /* needs req->io allocated for deferral/async */ | |
678 | unsigned async_ctx : 1; | |
679 | /* needs current->mm setup, does mm access */ | |
680 | unsigned needs_mm : 1; | |
681 | /* needs req->file assigned */ | |
682 | unsigned needs_file : 1; | |
d3656344 JA |
683 | /* hash wq insertion if file is a regular file */ |
684 | unsigned hash_reg_file : 1; | |
685 | /* unbound wq insertion if file is a non-regular file */ | |
686 | unsigned unbound_nonreg_file : 1; | |
66f4af93 JA |
687 | /* opcode is not supported by this kernel */ |
688 | unsigned not_supported : 1; | |
f86cd20c JA |
689 | /* needs file table */ |
690 | unsigned file_table : 1; | |
ff002b30 JA |
691 | /* needs ->fs */ |
692 | unsigned needs_fs : 1; | |
8a72758c JA |
693 | /* set if opcode supports polled "wait" */ |
694 | unsigned pollin : 1; | |
695 | unsigned pollout : 1; | |
bcda7baa JA |
696 | /* op supports buffer selection */ |
697 | unsigned buffer_select : 1; | |
d3656344 JA |
698 | }; |
699 | ||
700 | static const struct io_op_def io_op_defs[] = { | |
0463b6c5 PB |
701 | [IORING_OP_NOP] = {}, |
702 | [IORING_OP_READV] = { | |
d3656344 JA |
703 | .async_ctx = 1, |
704 | .needs_mm = 1, | |
705 | .needs_file = 1, | |
706 | .unbound_nonreg_file = 1, | |
8a72758c | 707 | .pollin = 1, |
4d954c25 | 708 | .buffer_select = 1, |
d3656344 | 709 | }, |
0463b6c5 | 710 | [IORING_OP_WRITEV] = { |
d3656344 JA |
711 | .async_ctx = 1, |
712 | .needs_mm = 1, | |
713 | .needs_file = 1, | |
714 | .hash_reg_file = 1, | |
715 | .unbound_nonreg_file = 1, | |
8a72758c | 716 | .pollout = 1, |
d3656344 | 717 | }, |
0463b6c5 | 718 | [IORING_OP_FSYNC] = { |
d3656344 JA |
719 | .needs_file = 1, |
720 | }, | |
0463b6c5 | 721 | [IORING_OP_READ_FIXED] = { |
d3656344 JA |
722 | .needs_file = 1, |
723 | .unbound_nonreg_file = 1, | |
8a72758c | 724 | .pollin = 1, |
d3656344 | 725 | }, |
0463b6c5 | 726 | [IORING_OP_WRITE_FIXED] = { |
d3656344 JA |
727 | .needs_file = 1, |
728 | .hash_reg_file = 1, | |
729 | .unbound_nonreg_file = 1, | |
8a72758c | 730 | .pollout = 1, |
d3656344 | 731 | }, |
0463b6c5 | 732 | [IORING_OP_POLL_ADD] = { |
d3656344 JA |
733 | .needs_file = 1, |
734 | .unbound_nonreg_file = 1, | |
735 | }, | |
0463b6c5 PB |
736 | [IORING_OP_POLL_REMOVE] = {}, |
737 | [IORING_OP_SYNC_FILE_RANGE] = { | |
d3656344 JA |
738 | .needs_file = 1, |
739 | }, | |
0463b6c5 | 740 | [IORING_OP_SENDMSG] = { |
d3656344 JA |
741 | .async_ctx = 1, |
742 | .needs_mm = 1, | |
743 | .needs_file = 1, | |
744 | .unbound_nonreg_file = 1, | |
ff002b30 | 745 | .needs_fs = 1, |
8a72758c | 746 | .pollout = 1, |
d3656344 | 747 | }, |
0463b6c5 | 748 | [IORING_OP_RECVMSG] = { |
d3656344 JA |
749 | .async_ctx = 1, |
750 | .needs_mm = 1, | |
751 | .needs_file = 1, | |
752 | .unbound_nonreg_file = 1, | |
ff002b30 | 753 | .needs_fs = 1, |
8a72758c | 754 | .pollin = 1, |
52de1fe1 | 755 | .buffer_select = 1, |
d3656344 | 756 | }, |
0463b6c5 | 757 | [IORING_OP_TIMEOUT] = { |
d3656344 JA |
758 | .async_ctx = 1, |
759 | .needs_mm = 1, | |
760 | }, | |
0463b6c5 PB |
761 | [IORING_OP_TIMEOUT_REMOVE] = {}, |
762 | [IORING_OP_ACCEPT] = { | |
d3656344 JA |
763 | .needs_mm = 1, |
764 | .needs_file = 1, | |
765 | .unbound_nonreg_file = 1, | |
f86cd20c | 766 | .file_table = 1, |
8a72758c | 767 | .pollin = 1, |
d3656344 | 768 | }, |
0463b6c5 PB |
769 | [IORING_OP_ASYNC_CANCEL] = {}, |
770 | [IORING_OP_LINK_TIMEOUT] = { | |
d3656344 JA |
771 | .async_ctx = 1, |
772 | .needs_mm = 1, | |
773 | }, | |
0463b6c5 | 774 | [IORING_OP_CONNECT] = { |
d3656344 JA |
775 | .async_ctx = 1, |
776 | .needs_mm = 1, | |
777 | .needs_file = 1, | |
778 | .unbound_nonreg_file = 1, | |
8a72758c | 779 | .pollout = 1, |
d3656344 | 780 | }, |
0463b6c5 | 781 | [IORING_OP_FALLOCATE] = { |
d3656344 JA |
782 | .needs_file = 1, |
783 | }, | |
0463b6c5 | 784 | [IORING_OP_OPENAT] = { |
f86cd20c | 785 | .file_table = 1, |
ff002b30 | 786 | .needs_fs = 1, |
d3656344 | 787 | }, |
0463b6c5 | 788 | [IORING_OP_CLOSE] = { |
f86cd20c | 789 | .file_table = 1, |
d3656344 | 790 | }, |
0463b6c5 | 791 | [IORING_OP_FILES_UPDATE] = { |
d3656344 | 792 | .needs_mm = 1, |
f86cd20c | 793 | .file_table = 1, |
d3656344 | 794 | }, |
0463b6c5 | 795 | [IORING_OP_STATX] = { |
d3656344 | 796 | .needs_mm = 1, |
ff002b30 | 797 | .needs_fs = 1, |
5b0bbee4 | 798 | .file_table = 1, |
d3656344 | 799 | }, |
0463b6c5 | 800 | [IORING_OP_READ] = { |
3a6820f2 JA |
801 | .needs_mm = 1, |
802 | .needs_file = 1, | |
803 | .unbound_nonreg_file = 1, | |
8a72758c | 804 | .pollin = 1, |
bcda7baa | 805 | .buffer_select = 1, |
3a6820f2 | 806 | }, |
0463b6c5 | 807 | [IORING_OP_WRITE] = { |
3a6820f2 JA |
808 | .needs_mm = 1, |
809 | .needs_file = 1, | |
810 | .unbound_nonreg_file = 1, | |
8a72758c | 811 | .pollout = 1, |
3a6820f2 | 812 | }, |
0463b6c5 | 813 | [IORING_OP_FADVISE] = { |
4840e418 JA |
814 | .needs_file = 1, |
815 | }, | |
0463b6c5 | 816 | [IORING_OP_MADVISE] = { |
c1ca757b JA |
817 | .needs_mm = 1, |
818 | }, | |
0463b6c5 | 819 | [IORING_OP_SEND] = { |
fddaface JA |
820 | .needs_mm = 1, |
821 | .needs_file = 1, | |
822 | .unbound_nonreg_file = 1, | |
8a72758c | 823 | .pollout = 1, |
fddaface | 824 | }, |
0463b6c5 | 825 | [IORING_OP_RECV] = { |
fddaface JA |
826 | .needs_mm = 1, |
827 | .needs_file = 1, | |
828 | .unbound_nonreg_file = 1, | |
8a72758c | 829 | .pollin = 1, |
bcda7baa | 830 | .buffer_select = 1, |
fddaface | 831 | }, |
0463b6c5 | 832 | [IORING_OP_OPENAT2] = { |
f86cd20c | 833 | .file_table = 1, |
ff002b30 | 834 | .needs_fs = 1, |
cebdb986 | 835 | }, |
3e4827b0 JA |
836 | [IORING_OP_EPOLL_CTL] = { |
837 | .unbound_nonreg_file = 1, | |
838 | .file_table = 1, | |
839 | }, | |
7d67af2c PB |
840 | [IORING_OP_SPLICE] = { |
841 | .needs_file = 1, | |
842 | .hash_reg_file = 1, | |
843 | .unbound_nonreg_file = 1, | |
ddf0322d JA |
844 | }, |
845 | [IORING_OP_PROVIDE_BUFFERS] = {}, | |
067524e9 | 846 | [IORING_OP_REMOVE_BUFFERS] = {}, |
d3656344 JA |
847 | }; |
848 | ||
561fb04a | 849 | static void io_wq_submit_work(struct io_wq_work **workptr); |
78e19bbe | 850 | static void io_cqring_fill_event(struct io_kiocb *req, long res); |
ec9c02ad | 851 | static void io_put_req(struct io_kiocb *req); |
978db57e | 852 | static void __io_double_put_req(struct io_kiocb *req); |
94ae5e77 JA |
853 | static struct io_kiocb *io_prep_linked_timeout(struct io_kiocb *req); |
854 | static void io_queue_linked_timeout(struct io_kiocb *req); | |
05f3fb3c JA |
855 | static int __io_sqe_files_update(struct io_ring_ctx *ctx, |
856 | struct io_uring_files_update *ip, | |
857 | unsigned nr_args); | |
f86cd20c | 858 | static int io_grab_files(struct io_kiocb *req); |
99bc4c38 | 859 | static void io_cleanup_req(struct io_kiocb *req); |
b41e9852 JA |
860 | static int io_file_get(struct io_submit_state *state, struct io_kiocb *req, |
861 | int fd, struct file **out_file, bool fixed); | |
862 | static void __io_queue_sqe(struct io_kiocb *req, | |
863 | const struct io_uring_sqe *sqe); | |
de0617e4 | 864 | |
2b188cc1 JA |
865 | static struct kmem_cache *req_cachep; |
866 | ||
867 | static const struct file_operations io_uring_fops; | |
868 | ||
869 | struct sock *io_uring_get_socket(struct file *file) | |
870 | { | |
871 | #if defined(CONFIG_UNIX) | |
872 | if (file->f_op == &io_uring_fops) { | |
873 | struct io_ring_ctx *ctx = file->private_data; | |
874 | ||
875 | return ctx->ring_sock->sk; | |
876 | } | |
877 | #endif | |
878 | return NULL; | |
879 | } | |
880 | EXPORT_SYMBOL(io_uring_get_socket); | |
881 | ||
882 | static void io_ring_ctx_ref_free(struct percpu_ref *ref) | |
883 | { | |
884 | struct io_ring_ctx *ctx = container_of(ref, struct io_ring_ctx, refs); | |
885 | ||
0f158b4c | 886 | complete(&ctx->ref_comp); |
2b188cc1 JA |
887 | } |
888 | ||
889 | static struct io_ring_ctx *io_ring_ctx_alloc(struct io_uring_params *p) | |
890 | { | |
891 | struct io_ring_ctx *ctx; | |
78076bb6 | 892 | int hash_bits; |
2b188cc1 JA |
893 | |
894 | ctx = kzalloc(sizeof(*ctx), GFP_KERNEL); | |
895 | if (!ctx) | |
896 | return NULL; | |
897 | ||
0ddf92e8 JA |
898 | ctx->fallback_req = kmem_cache_alloc(req_cachep, GFP_KERNEL); |
899 | if (!ctx->fallback_req) | |
900 | goto err; | |
901 | ||
78076bb6 JA |
902 | /* |
903 | * Use 5 bits less than the max cq entries, that should give us around | |
904 | * 32 entries per hash list if totally full and uniformly spread. | |
905 | */ | |
906 | hash_bits = ilog2(p->cq_entries); | |
907 | hash_bits -= 5; | |
908 | if (hash_bits <= 0) | |
909 | hash_bits = 1; | |
910 | ctx->cancel_hash_bits = hash_bits; | |
911 | ctx->cancel_hash = kmalloc((1U << hash_bits) * sizeof(struct hlist_head), | |
912 | GFP_KERNEL); | |
913 | if (!ctx->cancel_hash) | |
914 | goto err; | |
915 | __hash_init(ctx->cancel_hash, 1U << hash_bits); | |
916 | ||
21482896 | 917 | if (percpu_ref_init(&ctx->refs, io_ring_ctx_ref_free, |
206aefde JA |
918 | PERCPU_REF_ALLOW_REINIT, GFP_KERNEL)) |
919 | goto err; | |
2b188cc1 JA |
920 | |
921 | ctx->flags = p->flags; | |
922 | init_waitqueue_head(&ctx->cq_wait); | |
1d7bb1d5 | 923 | INIT_LIST_HEAD(&ctx->cq_overflow_list); |
0f158b4c JA |
924 | init_completion(&ctx->ref_comp); |
925 | init_completion(&ctx->sq_thread_comp); | |
5a2e745d | 926 | idr_init(&ctx->io_buffer_idr); |
071698e1 | 927 | idr_init(&ctx->personality_idr); |
2b188cc1 JA |
928 | mutex_init(&ctx->uring_lock); |
929 | init_waitqueue_head(&ctx->wait); | |
930 | spin_lock_init(&ctx->completion_lock); | |
def596e9 | 931 | INIT_LIST_HEAD(&ctx->poll_list); |
de0617e4 | 932 | INIT_LIST_HEAD(&ctx->defer_list); |
5262f567 | 933 | INIT_LIST_HEAD(&ctx->timeout_list); |
fcb323cc JA |
934 | init_waitqueue_head(&ctx->inflight_wait); |
935 | spin_lock_init(&ctx->inflight_lock); | |
936 | INIT_LIST_HEAD(&ctx->inflight_list); | |
2b188cc1 | 937 | return ctx; |
206aefde | 938 | err: |
0ddf92e8 JA |
939 | if (ctx->fallback_req) |
940 | kmem_cache_free(req_cachep, ctx->fallback_req); | |
78076bb6 | 941 | kfree(ctx->cancel_hash); |
206aefde JA |
942 | kfree(ctx); |
943 | return NULL; | |
2b188cc1 JA |
944 | } |
945 | ||
9d858b21 | 946 | static inline bool __req_need_defer(struct io_kiocb *req) |
7adf4eaf | 947 | { |
a197f664 JL |
948 | struct io_ring_ctx *ctx = req->ctx; |
949 | ||
31af27c7 PB |
950 | return req->sequence != ctx->cached_cq_tail |
951 | + atomic_read(&ctx->cached_cq_overflow); | |
7adf4eaf JA |
952 | } |
953 | ||
9d858b21 | 954 | static inline bool req_need_defer(struct io_kiocb *req) |
de0617e4 | 955 | { |
87987898 | 956 | if (unlikely(req->flags & REQ_F_IO_DRAIN)) |
9d858b21 | 957 | return __req_need_defer(req); |
de0617e4 | 958 | |
9d858b21 | 959 | return false; |
de0617e4 JA |
960 | } |
961 | ||
7adf4eaf | 962 | static struct io_kiocb *io_get_deferred_req(struct io_ring_ctx *ctx) |
de0617e4 JA |
963 | { |
964 | struct io_kiocb *req; | |
965 | ||
7adf4eaf | 966 | req = list_first_entry_or_null(&ctx->defer_list, struct io_kiocb, list); |
9d858b21 | 967 | if (req && !req_need_defer(req)) { |
de0617e4 JA |
968 | list_del_init(&req->list); |
969 | return req; | |
970 | } | |
971 | ||
972 | return NULL; | |
973 | } | |
974 | ||
5262f567 JA |
975 | static struct io_kiocb *io_get_timeout_req(struct io_ring_ctx *ctx) |
976 | { | |
7adf4eaf JA |
977 | struct io_kiocb *req; |
978 | ||
979 | req = list_first_entry_or_null(&ctx->timeout_list, struct io_kiocb, list); | |
93bd25bb JA |
980 | if (req) { |
981 | if (req->flags & REQ_F_TIMEOUT_NOSEQ) | |
982 | return NULL; | |
fb4b3d3f | 983 | if (!__req_need_defer(req)) { |
93bd25bb JA |
984 | list_del_init(&req->list); |
985 | return req; | |
986 | } | |
7adf4eaf JA |
987 | } |
988 | ||
989 | return NULL; | |
5262f567 JA |
990 | } |
991 | ||
de0617e4 | 992 | static void __io_commit_cqring(struct io_ring_ctx *ctx) |
2b188cc1 | 993 | { |
75b28aff | 994 | struct io_rings *rings = ctx->rings; |
2b188cc1 | 995 | |
07910158 PB |
996 | /* order cqe stores with ring update */ |
997 | smp_store_release(&rings->cq.tail, ctx->cached_cq_tail); | |
2b188cc1 | 998 | |
07910158 PB |
999 | if (wq_has_sleeper(&ctx->cq_wait)) { |
1000 | wake_up_interruptible(&ctx->cq_wait); | |
1001 | kill_fasync(&ctx->cq_fasync, SIGIO, POLL_IN); | |
2b188cc1 JA |
1002 | } |
1003 | } | |
1004 | ||
cccf0ee8 JA |
1005 | static inline void io_req_work_grab_env(struct io_kiocb *req, |
1006 | const struct io_op_def *def) | |
1007 | { | |
1008 | if (!req->work.mm && def->needs_mm) { | |
1009 | mmgrab(current->mm); | |
1010 | req->work.mm = current->mm; | |
2b188cc1 | 1011 | } |
cccf0ee8 JA |
1012 | if (!req->work.creds) |
1013 | req->work.creds = get_current_cred(); | |
ff002b30 JA |
1014 | if (!req->work.fs && def->needs_fs) { |
1015 | spin_lock(¤t->fs->lock); | |
1016 | if (!current->fs->in_exec) { | |
1017 | req->work.fs = current->fs; | |
1018 | req->work.fs->users++; | |
1019 | } else { | |
1020 | req->work.flags |= IO_WQ_WORK_CANCEL; | |
1021 | } | |
1022 | spin_unlock(¤t->fs->lock); | |
1023 | } | |
6ab23144 JA |
1024 | if (!req->work.task_pid) |
1025 | req->work.task_pid = task_pid_vnr(current); | |
2b188cc1 JA |
1026 | } |
1027 | ||
cccf0ee8 | 1028 | static inline void io_req_work_drop_env(struct io_kiocb *req) |
18d9be1a | 1029 | { |
cccf0ee8 JA |
1030 | if (req->work.mm) { |
1031 | mmdrop(req->work.mm); | |
1032 | req->work.mm = NULL; | |
1033 | } | |
1034 | if (req->work.creds) { | |
1035 | put_cred(req->work.creds); | |
1036 | req->work.creds = NULL; | |
1037 | } | |
ff002b30 JA |
1038 | if (req->work.fs) { |
1039 | struct fs_struct *fs = req->work.fs; | |
1040 | ||
1041 | spin_lock(&req->work.fs->lock); | |
1042 | if (--fs->users) | |
1043 | fs = NULL; | |
1044 | spin_unlock(&req->work.fs->lock); | |
1045 | if (fs) | |
1046 | free_fs_struct(fs); | |
1047 | } | |
561fb04a JA |
1048 | } |
1049 | ||
8766dd51 | 1050 | static inline void io_prep_async_work(struct io_kiocb *req, |
94ae5e77 | 1051 | struct io_kiocb **link) |
18d9be1a | 1052 | { |
d3656344 | 1053 | const struct io_op_def *def = &io_op_defs[req->opcode]; |
54a91f3b | 1054 | |
d3656344 JA |
1055 | if (req->flags & REQ_F_ISREG) { |
1056 | if (def->hash_reg_file) | |
8766dd51 | 1057 | io_wq_hash_work(&req->work, file_inode(req->file)); |
d3656344 JA |
1058 | } else { |
1059 | if (def->unbound_nonreg_file) | |
3529d8c2 | 1060 | req->work.flags |= IO_WQ_WORK_UNBOUND; |
54a91f3b | 1061 | } |
cccf0ee8 JA |
1062 | |
1063 | io_req_work_grab_env(req, def); | |
54a91f3b | 1064 | |
94ae5e77 | 1065 | *link = io_prep_linked_timeout(req); |
561fb04a JA |
1066 | } |
1067 | ||
a197f664 | 1068 | static inline void io_queue_async_work(struct io_kiocb *req) |
561fb04a | 1069 | { |
a197f664 | 1070 | struct io_ring_ctx *ctx = req->ctx; |
94ae5e77 | 1071 | struct io_kiocb *link; |
94ae5e77 | 1072 | |
8766dd51 | 1073 | io_prep_async_work(req, &link); |
561fb04a | 1074 | |
8766dd51 PB |
1075 | trace_io_uring_queue_async_work(ctx, io_wq_is_hashed(&req->work), req, |
1076 | &req->work, req->flags); | |
1077 | io_wq_enqueue(ctx->io_wq, &req->work); | |
94ae5e77 JA |
1078 | |
1079 | if (link) | |
1080 | io_queue_linked_timeout(link); | |
18d9be1a JA |
1081 | } |
1082 | ||
5262f567 JA |
1083 | static void io_kill_timeout(struct io_kiocb *req) |
1084 | { | |
1085 | int ret; | |
1086 | ||
2d28390a | 1087 | ret = hrtimer_try_to_cancel(&req->io->timeout.timer); |
5262f567 JA |
1088 | if (ret != -1) { |
1089 | atomic_inc(&req->ctx->cq_timeouts); | |
842f9612 | 1090 | list_del_init(&req->list); |
f0e20b89 | 1091 | req->flags |= REQ_F_COMP_LOCKED; |
78e19bbe | 1092 | io_cqring_fill_event(req, 0); |
ec9c02ad | 1093 | io_put_req(req); |
5262f567 JA |
1094 | } |
1095 | } | |
1096 | ||
1097 | static void io_kill_timeouts(struct io_ring_ctx *ctx) | |
1098 | { | |
1099 | struct io_kiocb *req, *tmp; | |
1100 | ||
1101 | spin_lock_irq(&ctx->completion_lock); | |
1102 | list_for_each_entry_safe(req, tmp, &ctx->timeout_list, list) | |
1103 | io_kill_timeout(req); | |
1104 | spin_unlock_irq(&ctx->completion_lock); | |
1105 | } | |
1106 | ||
de0617e4 JA |
1107 | static void io_commit_cqring(struct io_ring_ctx *ctx) |
1108 | { | |
1109 | struct io_kiocb *req; | |
1110 | ||
5262f567 JA |
1111 | while ((req = io_get_timeout_req(ctx)) != NULL) |
1112 | io_kill_timeout(req); | |
1113 | ||
de0617e4 JA |
1114 | __io_commit_cqring(ctx); |
1115 | ||
87987898 | 1116 | while ((req = io_get_deferred_req(ctx)) != NULL) |
a197f664 | 1117 | io_queue_async_work(req); |
de0617e4 JA |
1118 | } |
1119 | ||
2b188cc1 JA |
1120 | static struct io_uring_cqe *io_get_cqring(struct io_ring_ctx *ctx) |
1121 | { | |
75b28aff | 1122 | struct io_rings *rings = ctx->rings; |
2b188cc1 JA |
1123 | unsigned tail; |
1124 | ||
1125 | tail = ctx->cached_cq_tail; | |
115e12e5 SB |
1126 | /* |
1127 | * writes to the cq entry need to come after reading head; the | |
1128 | * control dependency is enough as we're using WRITE_ONCE to | |
1129 | * fill the cq entry | |
1130 | */ | |
75b28aff | 1131 | if (tail - READ_ONCE(rings->cq.head) == rings->cq_ring_entries) |
2b188cc1 JA |
1132 | return NULL; |
1133 | ||
1134 | ctx->cached_cq_tail++; | |
75b28aff | 1135 | return &rings->cqes[tail & ctx->cq_mask]; |
2b188cc1 JA |
1136 | } |
1137 | ||
f2842ab5 JA |
1138 | static inline bool io_should_trigger_evfd(struct io_ring_ctx *ctx) |
1139 | { | |
f0b493e6 JA |
1140 | if (!ctx->cq_ev_fd) |
1141 | return false; | |
f2842ab5 JA |
1142 | if (!ctx->eventfd_async) |
1143 | return true; | |
b41e9852 | 1144 | return io_wq_current_is_worker(); |
f2842ab5 JA |
1145 | } |
1146 | ||
b41e9852 | 1147 | static void io_cqring_ev_posted(struct io_ring_ctx *ctx) |
1d7bb1d5 JA |
1148 | { |
1149 | if (waitqueue_active(&ctx->wait)) | |
1150 | wake_up(&ctx->wait); | |
1151 | if (waitqueue_active(&ctx->sqo_wait)) | |
1152 | wake_up(&ctx->sqo_wait); | |
b41e9852 | 1153 | if (io_should_trigger_evfd(ctx)) |
1d7bb1d5 JA |
1154 | eventfd_signal(ctx->cq_ev_fd, 1); |
1155 | } | |
1156 | ||
c4a2ed72 JA |
1157 | /* Returns true if there are no backlogged entries after the flush */ |
1158 | static bool io_cqring_overflow_flush(struct io_ring_ctx *ctx, bool force) | |
1d7bb1d5 JA |
1159 | { |
1160 | struct io_rings *rings = ctx->rings; | |
1161 | struct io_uring_cqe *cqe; | |
1162 | struct io_kiocb *req; | |
1163 | unsigned long flags; | |
1164 | LIST_HEAD(list); | |
1165 | ||
1166 | if (!force) { | |
1167 | if (list_empty_careful(&ctx->cq_overflow_list)) | |
c4a2ed72 | 1168 | return true; |
1d7bb1d5 JA |
1169 | if ((ctx->cached_cq_tail - READ_ONCE(rings->cq.head) == |
1170 | rings->cq_ring_entries)) | |
c4a2ed72 | 1171 | return false; |
1d7bb1d5 JA |
1172 | } |
1173 | ||
1174 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
1175 | ||
1176 | /* if force is set, the ring is going away. always drop after that */ | |
1177 | if (force) | |
69b3e546 | 1178 | ctx->cq_overflow_flushed = 1; |
1d7bb1d5 | 1179 | |
c4a2ed72 | 1180 | cqe = NULL; |
1d7bb1d5 JA |
1181 | while (!list_empty(&ctx->cq_overflow_list)) { |
1182 | cqe = io_get_cqring(ctx); | |
1183 | if (!cqe && !force) | |
1184 | break; | |
1185 | ||
1186 | req = list_first_entry(&ctx->cq_overflow_list, struct io_kiocb, | |
1187 | list); | |
1188 | list_move(&req->list, &list); | |
2ca10259 | 1189 | req->flags &= ~REQ_F_OVERFLOW; |
1d7bb1d5 JA |
1190 | if (cqe) { |
1191 | WRITE_ONCE(cqe->user_data, req->user_data); | |
1192 | WRITE_ONCE(cqe->res, req->result); | |
bcda7baa | 1193 | WRITE_ONCE(cqe->flags, req->cflags); |
1d7bb1d5 JA |
1194 | } else { |
1195 | WRITE_ONCE(ctx->rings->cq_overflow, | |
1196 | atomic_inc_return(&ctx->cached_cq_overflow)); | |
1197 | } | |
1198 | } | |
1199 | ||
1200 | io_commit_cqring(ctx); | |
ad3eb2c8 JA |
1201 | if (cqe) { |
1202 | clear_bit(0, &ctx->sq_check_overflow); | |
1203 | clear_bit(0, &ctx->cq_check_overflow); | |
1204 | } | |
1d7bb1d5 JA |
1205 | spin_unlock_irqrestore(&ctx->completion_lock, flags); |
1206 | io_cqring_ev_posted(ctx); | |
1207 | ||
1208 | while (!list_empty(&list)) { | |
1209 | req = list_first_entry(&list, struct io_kiocb, list); | |
1210 | list_del(&req->list); | |
ec9c02ad | 1211 | io_put_req(req); |
1d7bb1d5 | 1212 | } |
c4a2ed72 JA |
1213 | |
1214 | return cqe != NULL; | |
1d7bb1d5 JA |
1215 | } |
1216 | ||
bcda7baa | 1217 | static void __io_cqring_fill_event(struct io_kiocb *req, long res, long cflags) |
2b188cc1 | 1218 | { |
78e19bbe | 1219 | struct io_ring_ctx *ctx = req->ctx; |
2b188cc1 JA |
1220 | struct io_uring_cqe *cqe; |
1221 | ||
78e19bbe | 1222 | trace_io_uring_complete(ctx, req->user_data, res); |
51c3ff62 | 1223 | |
2b188cc1 JA |
1224 | /* |
1225 | * If we can't get a cq entry, userspace overflowed the | |
1226 | * submission (by quite a lot). Increment the overflow count in | |
1227 | * the ring. | |
1228 | */ | |
1229 | cqe = io_get_cqring(ctx); | |
1d7bb1d5 | 1230 | if (likely(cqe)) { |
78e19bbe | 1231 | WRITE_ONCE(cqe->user_data, req->user_data); |
2b188cc1 | 1232 | WRITE_ONCE(cqe->res, res); |
bcda7baa | 1233 | WRITE_ONCE(cqe->flags, cflags); |
1d7bb1d5 | 1234 | } else if (ctx->cq_overflow_flushed) { |
498ccd9e JA |
1235 | WRITE_ONCE(ctx->rings->cq_overflow, |
1236 | atomic_inc_return(&ctx->cached_cq_overflow)); | |
1d7bb1d5 | 1237 | } else { |
ad3eb2c8 JA |
1238 | if (list_empty(&ctx->cq_overflow_list)) { |
1239 | set_bit(0, &ctx->sq_check_overflow); | |
1240 | set_bit(0, &ctx->cq_check_overflow); | |
1241 | } | |
2ca10259 | 1242 | req->flags |= REQ_F_OVERFLOW; |
1d7bb1d5 JA |
1243 | refcount_inc(&req->refs); |
1244 | req->result = res; | |
bcda7baa | 1245 | req->cflags = cflags; |
1d7bb1d5 | 1246 | list_add_tail(&req->list, &ctx->cq_overflow_list); |
2b188cc1 JA |
1247 | } |
1248 | } | |
1249 | ||
bcda7baa JA |
1250 | static void io_cqring_fill_event(struct io_kiocb *req, long res) |
1251 | { | |
1252 | __io_cqring_fill_event(req, res, 0); | |
1253 | } | |
1254 | ||
1255 | static void __io_cqring_add_event(struct io_kiocb *req, long res, long cflags) | |
2b188cc1 | 1256 | { |
78e19bbe | 1257 | struct io_ring_ctx *ctx = req->ctx; |
2b188cc1 JA |
1258 | unsigned long flags; |
1259 | ||
1260 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
bcda7baa | 1261 | __io_cqring_fill_event(req, res, cflags); |
2b188cc1 JA |
1262 | io_commit_cqring(ctx); |
1263 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
1264 | ||
8c838788 | 1265 | io_cqring_ev_posted(ctx); |
2b188cc1 JA |
1266 | } |
1267 | ||
bcda7baa JA |
1268 | static void io_cqring_add_event(struct io_kiocb *req, long res) |
1269 | { | |
1270 | __io_cqring_add_event(req, res, 0); | |
1271 | } | |
1272 | ||
0ddf92e8 JA |
1273 | static inline bool io_is_fallback_req(struct io_kiocb *req) |
1274 | { | |
1275 | return req == (struct io_kiocb *) | |
1276 | ((unsigned long) req->ctx->fallback_req & ~1UL); | |
1277 | } | |
1278 | ||
1279 | static struct io_kiocb *io_get_fallback_req(struct io_ring_ctx *ctx) | |
1280 | { | |
1281 | struct io_kiocb *req; | |
1282 | ||
1283 | req = ctx->fallback_req; | |
dd461af6 | 1284 | if (!test_and_set_bit_lock(0, (unsigned long *) &ctx->fallback_req)) |
0ddf92e8 JA |
1285 | return req; |
1286 | ||
1287 | return NULL; | |
1288 | } | |
1289 | ||
0553b8bd PB |
1290 | static struct io_kiocb *io_alloc_req(struct io_ring_ctx *ctx, |
1291 | struct io_submit_state *state) | |
2b188cc1 | 1292 | { |
fd6fab2c | 1293 | gfp_t gfp = GFP_KERNEL | __GFP_NOWARN; |
2b188cc1 JA |
1294 | struct io_kiocb *req; |
1295 | ||
2579f913 | 1296 | if (!state) { |
fd6fab2c | 1297 | req = kmem_cache_alloc(req_cachep, gfp); |
2579f913 | 1298 | if (unlikely(!req)) |
0ddf92e8 | 1299 | goto fallback; |
2579f913 JA |
1300 | } else if (!state->free_reqs) { |
1301 | size_t sz; | |
1302 | int ret; | |
1303 | ||
1304 | sz = min_t(size_t, state->ios_left, ARRAY_SIZE(state->reqs)); | |
fd6fab2c JA |
1305 | ret = kmem_cache_alloc_bulk(req_cachep, gfp, sz, state->reqs); |
1306 | ||
1307 | /* | |
1308 | * Bulk alloc is all-or-nothing. If we fail to get a batch, | |
1309 | * retry single alloc to be on the safe side. | |
1310 | */ | |
1311 | if (unlikely(ret <= 0)) { | |
1312 | state->reqs[0] = kmem_cache_alloc(req_cachep, gfp); | |
1313 | if (!state->reqs[0]) | |
0ddf92e8 | 1314 | goto fallback; |
fd6fab2c JA |
1315 | ret = 1; |
1316 | } | |
2579f913 | 1317 | state->free_reqs = ret - 1; |
6c8a3134 | 1318 | req = state->reqs[ret - 1]; |
2579f913 | 1319 | } else { |
2579f913 | 1320 | state->free_reqs--; |
6c8a3134 | 1321 | req = state->reqs[state->free_reqs]; |
2b188cc1 JA |
1322 | } |
1323 | ||
2579f913 | 1324 | return req; |
0ddf92e8 | 1325 | fallback: |
0553b8bd | 1326 | return io_get_fallback_req(ctx); |
2b188cc1 JA |
1327 | } |
1328 | ||
8da11c19 PB |
1329 | static inline void io_put_file(struct io_kiocb *req, struct file *file, |
1330 | bool fixed) | |
1331 | { | |
1332 | if (fixed) | |
05589553 | 1333 | percpu_ref_put(req->fixed_file_refs); |
8da11c19 PB |
1334 | else |
1335 | fput(file); | |
1336 | } | |
1337 | ||
c6ca97b3 | 1338 | static void __io_req_aux_free(struct io_kiocb *req) |
2b188cc1 | 1339 | { |
929a3af9 PB |
1340 | if (req->flags & REQ_F_NEED_CLEANUP) |
1341 | io_cleanup_req(req); | |
1342 | ||
96fd84d8 | 1343 | kfree(req->io); |
8da11c19 PB |
1344 | if (req->file) |
1345 | io_put_file(req, req->file, (req->flags & REQ_F_FIXED_FILE)); | |
3537b6a7 JA |
1346 | if (req->task) |
1347 | put_task_struct(req->task); | |
cccf0ee8 JA |
1348 | |
1349 | io_req_work_drop_env(req); | |
def596e9 JA |
1350 | } |
1351 | ||
9e645e11 | 1352 | static void __io_free_req(struct io_kiocb *req) |
2b188cc1 | 1353 | { |
c6ca97b3 | 1354 | __io_req_aux_free(req); |
fcb323cc | 1355 | |
fcb323cc | 1356 | if (req->flags & REQ_F_INFLIGHT) { |
c6ca97b3 | 1357 | struct io_ring_ctx *ctx = req->ctx; |
fcb323cc JA |
1358 | unsigned long flags; |
1359 | ||
1360 | spin_lock_irqsave(&ctx->inflight_lock, flags); | |
1361 | list_del(&req->inflight_entry); | |
1362 | if (waitqueue_active(&ctx->inflight_wait)) | |
1363 | wake_up(&ctx->inflight_wait); | |
1364 | spin_unlock_irqrestore(&ctx->inflight_lock, flags); | |
1365 | } | |
2b85edfc PB |
1366 | |
1367 | percpu_ref_put(&req->ctx->refs); | |
b1e50e54 PB |
1368 | if (likely(!io_is_fallback_req(req))) |
1369 | kmem_cache_free(req_cachep, req); | |
1370 | else | |
dd461af6 | 1371 | clear_bit_unlock(0, (unsigned long *) &req->ctx->fallback_req); |
e65ef56d JA |
1372 | } |
1373 | ||
c6ca97b3 JA |
1374 | struct req_batch { |
1375 | void *reqs[IO_IOPOLL_BATCH]; | |
1376 | int to_free; | |
1377 | int need_iter; | |
1378 | }; | |
1379 | ||
1380 | static void io_free_req_many(struct io_ring_ctx *ctx, struct req_batch *rb) | |
1381 | { | |
1382 | if (!rb->to_free) | |
1383 | return; | |
1384 | if (rb->need_iter) { | |
1385 | int i, inflight = 0; | |
1386 | unsigned long flags; | |
1387 | ||
1388 | for (i = 0; i < rb->to_free; i++) { | |
1389 | struct io_kiocb *req = rb->reqs[i]; | |
1390 | ||
10fef4be | 1391 | if (req->flags & REQ_F_FIXED_FILE) { |
c6ca97b3 | 1392 | req->file = NULL; |
05589553 | 1393 | percpu_ref_put(req->fixed_file_refs); |
10fef4be | 1394 | } |
c6ca97b3 JA |
1395 | if (req->flags & REQ_F_INFLIGHT) |
1396 | inflight++; | |
c6ca97b3 JA |
1397 | __io_req_aux_free(req); |
1398 | } | |
1399 | if (!inflight) | |
1400 | goto do_free; | |
1401 | ||
1402 | spin_lock_irqsave(&ctx->inflight_lock, flags); | |
1403 | for (i = 0; i < rb->to_free; i++) { | |
1404 | struct io_kiocb *req = rb->reqs[i]; | |
1405 | ||
10fef4be | 1406 | if (req->flags & REQ_F_INFLIGHT) { |
c6ca97b3 JA |
1407 | list_del(&req->inflight_entry); |
1408 | if (!--inflight) | |
1409 | break; | |
1410 | } | |
1411 | } | |
1412 | spin_unlock_irqrestore(&ctx->inflight_lock, flags); | |
1413 | ||
1414 | if (waitqueue_active(&ctx->inflight_wait)) | |
1415 | wake_up(&ctx->inflight_wait); | |
1416 | } | |
1417 | do_free: | |
1418 | kmem_cache_free_bulk(req_cachep, rb->to_free, rb->reqs); | |
1419 | percpu_ref_put_many(&ctx->refs, rb->to_free); | |
c6ca97b3 | 1420 | rb->to_free = rb->need_iter = 0; |
e65ef56d JA |
1421 | } |
1422 | ||
a197f664 | 1423 | static bool io_link_cancel_timeout(struct io_kiocb *req) |
2665abfd | 1424 | { |
a197f664 | 1425 | struct io_ring_ctx *ctx = req->ctx; |
2665abfd JA |
1426 | int ret; |
1427 | ||
2d28390a | 1428 | ret = hrtimer_try_to_cancel(&req->io->timeout.timer); |
2665abfd | 1429 | if (ret != -1) { |
78e19bbe | 1430 | io_cqring_fill_event(req, -ECANCELED); |
2665abfd | 1431 | io_commit_cqring(ctx); |
dea3b49c | 1432 | req->flags &= ~REQ_F_LINK_HEAD; |
ec9c02ad | 1433 | io_put_req(req); |
2665abfd JA |
1434 | return true; |
1435 | } | |
1436 | ||
1437 | return false; | |
e65ef56d JA |
1438 | } |
1439 | ||
ba816ad6 | 1440 | static void io_req_link_next(struct io_kiocb *req, struct io_kiocb **nxtptr) |
9e645e11 | 1441 | { |
2665abfd | 1442 | struct io_ring_ctx *ctx = req->ctx; |
2665abfd | 1443 | bool wake_ev = false; |
9e645e11 | 1444 | |
4d7dd462 JA |
1445 | /* Already got next link */ |
1446 | if (req->flags & REQ_F_LINK_NEXT) | |
1447 | return; | |
1448 | ||
9e645e11 JA |
1449 | /* |
1450 | * The list should never be empty when we are called here. But could | |
1451 | * potentially happen if the chain is messed up, check to be on the | |
1452 | * safe side. | |
1453 | */ | |
4493233e PB |
1454 | while (!list_empty(&req->link_list)) { |
1455 | struct io_kiocb *nxt = list_first_entry(&req->link_list, | |
1456 | struct io_kiocb, link_list); | |
94ae5e77 | 1457 | |
4493233e PB |
1458 | if (unlikely((req->flags & REQ_F_LINK_TIMEOUT) && |
1459 | (nxt->flags & REQ_F_TIMEOUT))) { | |
1460 | list_del_init(&nxt->link_list); | |
94ae5e77 | 1461 | wake_ev |= io_link_cancel_timeout(nxt); |
94ae5e77 JA |
1462 | req->flags &= ~REQ_F_LINK_TIMEOUT; |
1463 | continue; | |
1464 | } | |
9e645e11 | 1465 | |
4493233e PB |
1466 | list_del_init(&req->link_list); |
1467 | if (!list_empty(&nxt->link_list)) | |
dea3b49c | 1468 | nxt->flags |= REQ_F_LINK_HEAD; |
b18fdf71 | 1469 | *nxtptr = nxt; |
94ae5e77 | 1470 | break; |
9e645e11 | 1471 | } |
2665abfd | 1472 | |
4d7dd462 | 1473 | req->flags |= REQ_F_LINK_NEXT; |
2665abfd JA |
1474 | if (wake_ev) |
1475 | io_cqring_ev_posted(ctx); | |
9e645e11 JA |
1476 | } |
1477 | ||
1478 | /* | |
dea3b49c | 1479 | * Called if REQ_F_LINK_HEAD is set, and we fail the head request |
9e645e11 JA |
1480 | */ |
1481 | static void io_fail_links(struct io_kiocb *req) | |
1482 | { | |
2665abfd | 1483 | struct io_ring_ctx *ctx = req->ctx; |
2665abfd JA |
1484 | unsigned long flags; |
1485 | ||
1486 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
9e645e11 JA |
1487 | |
1488 | while (!list_empty(&req->link_list)) { | |
4493233e PB |
1489 | struct io_kiocb *link = list_first_entry(&req->link_list, |
1490 | struct io_kiocb, link_list); | |
9e645e11 | 1491 | |
4493233e | 1492 | list_del_init(&link->link_list); |
c826bd7a | 1493 | trace_io_uring_fail_link(req, link); |
2665abfd JA |
1494 | |
1495 | if ((req->flags & REQ_F_LINK_TIMEOUT) && | |
d625c6ee | 1496 | link->opcode == IORING_OP_LINK_TIMEOUT) { |
a197f664 | 1497 | io_link_cancel_timeout(link); |
2665abfd | 1498 | } else { |
78e19bbe | 1499 | io_cqring_fill_event(link, -ECANCELED); |
978db57e | 1500 | __io_double_put_req(link); |
2665abfd | 1501 | } |
5d960724 | 1502 | req->flags &= ~REQ_F_LINK_TIMEOUT; |
9e645e11 | 1503 | } |
2665abfd JA |
1504 | |
1505 | io_commit_cqring(ctx); | |
1506 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
1507 | io_cqring_ev_posted(ctx); | |
9e645e11 JA |
1508 | } |
1509 | ||
4d7dd462 | 1510 | static void io_req_find_next(struct io_kiocb *req, struct io_kiocb **nxt) |
9e645e11 | 1511 | { |
dea3b49c | 1512 | if (likely(!(req->flags & REQ_F_LINK_HEAD))) |
2665abfd | 1513 | return; |
2665abfd | 1514 | |
9e645e11 JA |
1515 | /* |
1516 | * If LINK is set, we have dependent requests in this chain. If we | |
1517 | * didn't fail this request, queue the first one up, moving any other | |
1518 | * dependencies to the next request. In case of failure, fail the rest | |
1519 | * of the chain. | |
1520 | */ | |
2665abfd JA |
1521 | if (req->flags & REQ_F_FAIL_LINK) { |
1522 | io_fail_links(req); | |
7c9e7f0f JA |
1523 | } else if ((req->flags & (REQ_F_LINK_TIMEOUT | REQ_F_COMP_LOCKED)) == |
1524 | REQ_F_LINK_TIMEOUT) { | |
2665abfd JA |
1525 | struct io_ring_ctx *ctx = req->ctx; |
1526 | unsigned long flags; | |
1527 | ||
1528 | /* | |
1529 | * If this is a timeout link, we could be racing with the | |
1530 | * timeout timer. Grab the completion lock for this case to | |
7c9e7f0f | 1531 | * protect against that. |
2665abfd JA |
1532 | */ |
1533 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
1534 | io_req_link_next(req, nxt); | |
1535 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
1536 | } else { | |
1537 | io_req_link_next(req, nxt); | |
9e645e11 | 1538 | } |
4d7dd462 | 1539 | } |
9e645e11 | 1540 | |
c69f8dbe JL |
1541 | static void io_free_req(struct io_kiocb *req) |
1542 | { | |
944e58bf PB |
1543 | struct io_kiocb *nxt = NULL; |
1544 | ||
1545 | io_req_find_next(req, &nxt); | |
70cf9f32 | 1546 | __io_free_req(req); |
944e58bf PB |
1547 | |
1548 | if (nxt) | |
1549 | io_queue_async_work(nxt); | |
c69f8dbe JL |
1550 | } |
1551 | ||
7a743e22 PB |
1552 | static void io_link_work_cb(struct io_wq_work **workptr) |
1553 | { | |
18a542ff PB |
1554 | struct io_kiocb *req = container_of(*workptr, struct io_kiocb, work); |
1555 | struct io_kiocb *link; | |
7a743e22 | 1556 | |
18a542ff | 1557 | link = list_first_entry(&req->link_list, struct io_kiocb, link_list); |
7a743e22 PB |
1558 | io_queue_linked_timeout(link); |
1559 | io_wq_submit_work(workptr); | |
1560 | } | |
1561 | ||
1562 | static void io_wq_assign_next(struct io_wq_work **workptr, struct io_kiocb *nxt) | |
1563 | { | |
1564 | struct io_kiocb *link; | |
8766dd51 PB |
1565 | const struct io_op_def *def = &io_op_defs[nxt->opcode]; |
1566 | ||
1567 | if ((nxt->flags & REQ_F_ISREG) && def->hash_reg_file) | |
1568 | io_wq_hash_work(&nxt->work, file_inode(nxt->file)); | |
7a743e22 PB |
1569 | |
1570 | *workptr = &nxt->work; | |
1571 | link = io_prep_linked_timeout(nxt); | |
18a542ff | 1572 | if (link) |
7a743e22 | 1573 | nxt->work.func = io_link_work_cb; |
7a743e22 PB |
1574 | } |
1575 | ||
ba816ad6 JA |
1576 | /* |
1577 | * Drop reference to request, return next in chain (if there is one) if this | |
1578 | * was the last reference to this request. | |
1579 | */ | |
f9bd67f6 | 1580 | __attribute__((nonnull)) |
ec9c02ad | 1581 | static void io_put_req_find_next(struct io_kiocb *req, struct io_kiocb **nxtptr) |
e65ef56d | 1582 | { |
2a44f467 JA |
1583 | if (refcount_dec_and_test(&req->refs)) { |
1584 | io_req_find_next(req, nxtptr); | |
4d7dd462 | 1585 | __io_free_req(req); |
2a44f467 | 1586 | } |
2b188cc1 JA |
1587 | } |
1588 | ||
e65ef56d JA |
1589 | static void io_put_req(struct io_kiocb *req) |
1590 | { | |
1591 | if (refcount_dec_and_test(&req->refs)) | |
1592 | io_free_req(req); | |
2b188cc1 JA |
1593 | } |
1594 | ||
e9fd9396 PB |
1595 | static void io_steal_work(struct io_kiocb *req, |
1596 | struct io_wq_work **workptr) | |
7a743e22 PB |
1597 | { |
1598 | /* | |
1599 | * It's in an io-wq worker, so there always should be at least | |
1600 | * one reference, which will be dropped in io_put_work() just | |
1601 | * after the current handler returns. | |
1602 | * | |
1603 | * It also means, that if the counter dropped to 1, then there is | |
1604 | * no asynchronous users left, so it's safe to steal the next work. | |
1605 | */ | |
7a743e22 PB |
1606 | if (refcount_read(&req->refs) == 1) { |
1607 | struct io_kiocb *nxt = NULL; | |
1608 | ||
1609 | io_req_find_next(req, &nxt); | |
1610 | if (nxt) | |
1611 | io_wq_assign_next(workptr, nxt); | |
1612 | } | |
1613 | } | |
1614 | ||
978db57e JA |
1615 | /* |
1616 | * Must only be used if we don't need to care about links, usually from | |
1617 | * within the completion handling itself. | |
1618 | */ | |
1619 | static void __io_double_put_req(struct io_kiocb *req) | |
78e19bbe JA |
1620 | { |
1621 | /* drop both submit and complete references */ | |
1622 | if (refcount_sub_and_test(2, &req->refs)) | |
1623 | __io_free_req(req); | |
1624 | } | |
1625 | ||
978db57e JA |
1626 | static void io_double_put_req(struct io_kiocb *req) |
1627 | { | |
1628 | /* drop both submit and complete references */ | |
1629 | if (refcount_sub_and_test(2, &req->refs)) | |
1630 | io_free_req(req); | |
1631 | } | |
1632 | ||
1d7bb1d5 | 1633 | static unsigned io_cqring_events(struct io_ring_ctx *ctx, bool noflush) |
a3a0e43f | 1634 | { |
84f97dc2 JA |
1635 | struct io_rings *rings = ctx->rings; |
1636 | ||
ad3eb2c8 JA |
1637 | if (test_bit(0, &ctx->cq_check_overflow)) { |
1638 | /* | |
1639 | * noflush == true is from the waitqueue handler, just ensure | |
1640 | * we wake up the task, and the next invocation will flush the | |
1641 | * entries. We cannot safely to it from here. | |
1642 | */ | |
1643 | if (noflush && !list_empty(&ctx->cq_overflow_list)) | |
1644 | return -1U; | |
1d7bb1d5 | 1645 | |
ad3eb2c8 JA |
1646 | io_cqring_overflow_flush(ctx, false); |
1647 | } | |
1d7bb1d5 | 1648 | |
a3a0e43f JA |
1649 | /* See comment at the top of this file */ |
1650 | smp_rmb(); | |
ad3eb2c8 | 1651 | return ctx->cached_cq_tail - READ_ONCE(rings->cq.head); |
a3a0e43f JA |
1652 | } |
1653 | ||
fb5ccc98 PB |
1654 | static inline unsigned int io_sqring_entries(struct io_ring_ctx *ctx) |
1655 | { | |
1656 | struct io_rings *rings = ctx->rings; | |
1657 | ||
1658 | /* make sure SQ entry isn't read before tail */ | |
1659 | return smp_load_acquire(&rings->sq.tail) - ctx->cached_sq_head; | |
1660 | } | |
1661 | ||
8237e045 | 1662 | static inline bool io_req_multi_free(struct req_batch *rb, struct io_kiocb *req) |
e94f141b | 1663 | { |
dea3b49c | 1664 | if ((req->flags & REQ_F_LINK_HEAD) || io_is_fallback_req(req)) |
c6ca97b3 | 1665 | return false; |
e94f141b | 1666 | |
c6ca97b3 JA |
1667 | if (!(req->flags & REQ_F_FIXED_FILE) || req->io) |
1668 | rb->need_iter++; | |
1669 | ||
1670 | rb->reqs[rb->to_free++] = req; | |
1671 | if (unlikely(rb->to_free == ARRAY_SIZE(rb->reqs))) | |
1672 | io_free_req_many(req->ctx, rb); | |
1673 | return true; | |
e94f141b JA |
1674 | } |
1675 | ||
bcda7baa JA |
1676 | static int io_put_kbuf(struct io_kiocb *req) |
1677 | { | |
4d954c25 | 1678 | struct io_buffer *kbuf; |
bcda7baa JA |
1679 | int cflags; |
1680 | ||
4d954c25 | 1681 | kbuf = (struct io_buffer *) (unsigned long) req->rw.addr; |
bcda7baa JA |
1682 | cflags = kbuf->bid << IORING_CQE_BUFFER_SHIFT; |
1683 | cflags |= IORING_CQE_F_BUFFER; | |
1684 | req->rw.addr = 0; | |
1685 | kfree(kbuf); | |
1686 | return cflags; | |
1687 | } | |
1688 | ||
def596e9 JA |
1689 | /* |
1690 | * Find and free completed poll iocbs | |
1691 | */ | |
1692 | static void io_iopoll_complete(struct io_ring_ctx *ctx, unsigned int *nr_events, | |
1693 | struct list_head *done) | |
1694 | { | |
8237e045 | 1695 | struct req_batch rb; |
def596e9 | 1696 | struct io_kiocb *req; |
def596e9 | 1697 | |
c6ca97b3 | 1698 | rb.to_free = rb.need_iter = 0; |
def596e9 | 1699 | while (!list_empty(done)) { |
bcda7baa JA |
1700 | int cflags = 0; |
1701 | ||
def596e9 JA |
1702 | req = list_first_entry(done, struct io_kiocb, list); |
1703 | list_del(&req->list); | |
1704 | ||
bcda7baa JA |
1705 | if (req->flags & REQ_F_BUFFER_SELECTED) |
1706 | cflags = io_put_kbuf(req); | |
1707 | ||
1708 | __io_cqring_fill_event(req, req->result, cflags); | |
def596e9 JA |
1709 | (*nr_events)++; |
1710 | ||
8237e045 JA |
1711 | if (refcount_dec_and_test(&req->refs) && |
1712 | !io_req_multi_free(&rb, req)) | |
1713 | io_free_req(req); | |
def596e9 | 1714 | } |
def596e9 | 1715 | |
09bb8394 | 1716 | io_commit_cqring(ctx); |
32b2244a XW |
1717 | if (ctx->flags & IORING_SETUP_SQPOLL) |
1718 | io_cqring_ev_posted(ctx); | |
8237e045 | 1719 | io_free_req_many(ctx, &rb); |
def596e9 JA |
1720 | } |
1721 | ||
581f9810 BM |
1722 | static void io_iopoll_queue(struct list_head *again) |
1723 | { | |
1724 | struct io_kiocb *req; | |
1725 | ||
1726 | do { | |
1727 | req = list_first_entry(again, struct io_kiocb, list); | |
1728 | list_del(&req->list); | |
1729 | refcount_inc(&req->refs); | |
1730 | io_queue_async_work(req); | |
1731 | } while (!list_empty(again)); | |
1732 | } | |
1733 | ||
def596e9 JA |
1734 | static int io_do_iopoll(struct io_ring_ctx *ctx, unsigned int *nr_events, |
1735 | long min) | |
1736 | { | |
1737 | struct io_kiocb *req, *tmp; | |
1738 | LIST_HEAD(done); | |
581f9810 | 1739 | LIST_HEAD(again); |
def596e9 JA |
1740 | bool spin; |
1741 | int ret; | |
1742 | ||
1743 | /* | |
1744 | * Only spin for completions if we don't have multiple devices hanging | |
1745 | * off our complete list, and we're under the requested amount. | |
1746 | */ | |
1747 | spin = !ctx->poll_multi_file && *nr_events < min; | |
1748 | ||
1749 | ret = 0; | |
1750 | list_for_each_entry_safe(req, tmp, &ctx->poll_list, list) { | |
9adbd45d | 1751 | struct kiocb *kiocb = &req->rw.kiocb; |
def596e9 JA |
1752 | |
1753 | /* | |
581f9810 BM |
1754 | * Move completed and retryable entries to our local lists. |
1755 | * If we find a request that requires polling, break out | |
1756 | * and complete those lists first, if we have entries there. | |
def596e9 JA |
1757 | */ |
1758 | if (req->flags & REQ_F_IOPOLL_COMPLETED) { | |
1759 | list_move_tail(&req->list, &done); | |
1760 | continue; | |
1761 | } | |
1762 | if (!list_empty(&done)) | |
1763 | break; | |
1764 | ||
581f9810 BM |
1765 | if (req->result == -EAGAIN) { |
1766 | list_move_tail(&req->list, &again); | |
1767 | continue; | |
1768 | } | |
1769 | if (!list_empty(&again)) | |
1770 | break; | |
1771 | ||
def596e9 JA |
1772 | ret = kiocb->ki_filp->f_op->iopoll(kiocb, spin); |
1773 | if (ret < 0) | |
1774 | break; | |
1775 | ||
1776 | if (ret && spin) | |
1777 | spin = false; | |
1778 | ret = 0; | |
1779 | } | |
1780 | ||
1781 | if (!list_empty(&done)) | |
1782 | io_iopoll_complete(ctx, nr_events, &done); | |
1783 | ||
581f9810 BM |
1784 | if (!list_empty(&again)) |
1785 | io_iopoll_queue(&again); | |
1786 | ||
def596e9 JA |
1787 | return ret; |
1788 | } | |
1789 | ||
1790 | /* | |
d195a66e | 1791 | * Poll for a minimum of 'min' events. Note that if min == 0 we consider that a |
def596e9 JA |
1792 | * non-spinning poll check - we'll still enter the driver poll loop, but only |
1793 | * as a non-spinning completion check. | |
1794 | */ | |
1795 | static int io_iopoll_getevents(struct io_ring_ctx *ctx, unsigned int *nr_events, | |
1796 | long min) | |
1797 | { | |
08f5439f | 1798 | while (!list_empty(&ctx->poll_list) && !need_resched()) { |
def596e9 JA |
1799 | int ret; |
1800 | ||
1801 | ret = io_do_iopoll(ctx, nr_events, min); | |
1802 | if (ret < 0) | |
1803 | return ret; | |
1804 | if (!min || *nr_events >= min) | |
1805 | return 0; | |
1806 | } | |
1807 | ||
1808 | return 1; | |
1809 | } | |
1810 | ||
1811 | /* | |
1812 | * We can't just wait for polled events to come to us, we have to actively | |
1813 | * find and complete them. | |
1814 | */ | |
1815 | static void io_iopoll_reap_events(struct io_ring_ctx *ctx) | |
1816 | { | |
1817 | if (!(ctx->flags & IORING_SETUP_IOPOLL)) | |
1818 | return; | |
1819 | ||
1820 | mutex_lock(&ctx->uring_lock); | |
1821 | while (!list_empty(&ctx->poll_list)) { | |
1822 | unsigned int nr_events = 0; | |
1823 | ||
1824 | io_iopoll_getevents(ctx, &nr_events, 1); | |
08f5439f JA |
1825 | |
1826 | /* | |
1827 | * Ensure we allow local-to-the-cpu processing to take place, | |
1828 | * in this case we need to ensure that we reap all events. | |
1829 | */ | |
1830 | cond_resched(); | |
def596e9 JA |
1831 | } |
1832 | mutex_unlock(&ctx->uring_lock); | |
1833 | } | |
1834 | ||
c7849be9 XW |
1835 | static int io_iopoll_check(struct io_ring_ctx *ctx, unsigned *nr_events, |
1836 | long min) | |
def596e9 | 1837 | { |
2b2ed975 | 1838 | int iters = 0, ret = 0; |
500f9fba | 1839 | |
c7849be9 XW |
1840 | /* |
1841 | * We disallow the app entering submit/complete with polling, but we | |
1842 | * still need to lock the ring to prevent racing with polled issue | |
1843 | * that got punted to a workqueue. | |
1844 | */ | |
1845 | mutex_lock(&ctx->uring_lock); | |
def596e9 JA |
1846 | do { |
1847 | int tmin = 0; | |
1848 | ||
a3a0e43f JA |
1849 | /* |
1850 | * Don't enter poll loop if we already have events pending. | |
1851 | * If we do, we can potentially be spinning for commands that | |
1852 | * already triggered a CQE (eg in error). | |
1853 | */ | |
1d7bb1d5 | 1854 | if (io_cqring_events(ctx, false)) |
a3a0e43f JA |
1855 | break; |
1856 | ||
500f9fba JA |
1857 | /* |
1858 | * If a submit got punted to a workqueue, we can have the | |
1859 | * application entering polling for a command before it gets | |
1860 | * issued. That app will hold the uring_lock for the duration | |
1861 | * of the poll right here, so we need to take a breather every | |
1862 | * now and then to ensure that the issue has a chance to add | |
1863 | * the poll to the issued list. Otherwise we can spin here | |
1864 | * forever, while the workqueue is stuck trying to acquire the | |
1865 | * very same mutex. | |
1866 | */ | |
1867 | if (!(++iters & 7)) { | |
1868 | mutex_unlock(&ctx->uring_lock); | |
1869 | mutex_lock(&ctx->uring_lock); | |
1870 | } | |
1871 | ||
def596e9 JA |
1872 | if (*nr_events < min) |
1873 | tmin = min - *nr_events; | |
1874 | ||
1875 | ret = io_iopoll_getevents(ctx, nr_events, tmin); | |
1876 | if (ret <= 0) | |
1877 | break; | |
1878 | ret = 0; | |
1879 | } while (min && !*nr_events && !need_resched()); | |
1880 | ||
500f9fba | 1881 | mutex_unlock(&ctx->uring_lock); |
def596e9 JA |
1882 | return ret; |
1883 | } | |
1884 | ||
491381ce | 1885 | static void kiocb_end_write(struct io_kiocb *req) |
2b188cc1 | 1886 | { |
491381ce JA |
1887 | /* |
1888 | * Tell lockdep we inherited freeze protection from submission | |
1889 | * thread. | |
1890 | */ | |
1891 | if (req->flags & REQ_F_ISREG) { | |
1892 | struct inode *inode = file_inode(req->file); | |
2b188cc1 | 1893 | |
491381ce | 1894 | __sb_writers_acquired(inode->i_sb, SB_FREEZE_WRITE); |
2b188cc1 | 1895 | } |
491381ce | 1896 | file_end_write(req->file); |
2b188cc1 JA |
1897 | } |
1898 | ||
4e88d6e7 JA |
1899 | static inline void req_set_fail_links(struct io_kiocb *req) |
1900 | { | |
1901 | if ((req->flags & (REQ_F_LINK | REQ_F_HARDLINK)) == REQ_F_LINK) | |
1902 | req->flags |= REQ_F_FAIL_LINK; | |
1903 | } | |
1904 | ||
ba816ad6 | 1905 | static void io_complete_rw_common(struct kiocb *kiocb, long res) |
2b188cc1 | 1906 | { |
9adbd45d | 1907 | struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb); |
bcda7baa | 1908 | int cflags = 0; |
2b188cc1 | 1909 | |
491381ce JA |
1910 | if (kiocb->ki_flags & IOCB_WRITE) |
1911 | kiocb_end_write(req); | |
2b188cc1 | 1912 | |
4e88d6e7 JA |
1913 | if (res != req->result) |
1914 | req_set_fail_links(req); | |
bcda7baa JA |
1915 | if (req->flags & REQ_F_BUFFER_SELECTED) |
1916 | cflags = io_put_kbuf(req); | |
1917 | __io_cqring_add_event(req, res, cflags); | |
ba816ad6 JA |
1918 | } |
1919 | ||
1920 | static void io_complete_rw(struct kiocb *kiocb, long res, long res2) | |
1921 | { | |
9adbd45d | 1922 | struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb); |
ba816ad6 JA |
1923 | |
1924 | io_complete_rw_common(kiocb, res); | |
e65ef56d | 1925 | io_put_req(req); |
2b188cc1 JA |
1926 | } |
1927 | ||
def596e9 JA |
1928 | static void io_complete_rw_iopoll(struct kiocb *kiocb, long res, long res2) |
1929 | { | |
9adbd45d | 1930 | struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb); |
def596e9 | 1931 | |
491381ce JA |
1932 | if (kiocb->ki_flags & IOCB_WRITE) |
1933 | kiocb_end_write(req); | |
def596e9 | 1934 | |
4e88d6e7 JA |
1935 | if (res != req->result) |
1936 | req_set_fail_links(req); | |
9e645e11 | 1937 | req->result = res; |
def596e9 JA |
1938 | if (res != -EAGAIN) |
1939 | req->flags |= REQ_F_IOPOLL_COMPLETED; | |
1940 | } | |
1941 | ||
1942 | /* | |
1943 | * After the iocb has been issued, it's safe to be found on the poll list. | |
1944 | * Adding the kiocb to the list AFTER submission ensures that we don't | |
1945 | * find it from a io_iopoll_getevents() thread before the issuer is done | |
1946 | * accessing the kiocb cookie. | |
1947 | */ | |
1948 | static void io_iopoll_req_issued(struct io_kiocb *req) | |
1949 | { | |
1950 | struct io_ring_ctx *ctx = req->ctx; | |
1951 | ||
1952 | /* | |
1953 | * Track whether we have multiple files in our lists. This will impact | |
1954 | * how we do polling eventually, not spinning if we're on potentially | |
1955 | * different devices. | |
1956 | */ | |
1957 | if (list_empty(&ctx->poll_list)) { | |
1958 | ctx->poll_multi_file = false; | |
1959 | } else if (!ctx->poll_multi_file) { | |
1960 | struct io_kiocb *list_req; | |
1961 | ||
1962 | list_req = list_first_entry(&ctx->poll_list, struct io_kiocb, | |
1963 | list); | |
9adbd45d | 1964 | if (list_req->file != req->file) |
def596e9 JA |
1965 | ctx->poll_multi_file = true; |
1966 | } | |
1967 | ||
1968 | /* | |
1969 | * For fast devices, IO may have already completed. If it has, add | |
1970 | * it to the front so we find it first. | |
1971 | */ | |
1972 | if (req->flags & REQ_F_IOPOLL_COMPLETED) | |
1973 | list_add(&req->list, &ctx->poll_list); | |
1974 | else | |
1975 | list_add_tail(&req->list, &ctx->poll_list); | |
bdcd3eab XW |
1976 | |
1977 | if ((ctx->flags & IORING_SETUP_SQPOLL) && | |
1978 | wq_has_sleeper(&ctx->sqo_wait)) | |
1979 | wake_up(&ctx->sqo_wait); | |
def596e9 JA |
1980 | } |
1981 | ||
3d6770fb | 1982 | static void io_file_put(struct io_submit_state *state) |
9a56a232 | 1983 | { |
3d6770fb | 1984 | if (state->file) { |
9a56a232 JA |
1985 | int diff = state->has_refs - state->used_refs; |
1986 | ||
1987 | if (diff) | |
1988 | fput_many(state->file, diff); | |
1989 | state->file = NULL; | |
1990 | } | |
1991 | } | |
1992 | ||
1993 | /* | |
1994 | * Get as many references to a file as we have IOs left in this submission, | |
1995 | * assuming most submissions are for one file, or at least that each file | |
1996 | * has more than one submission. | |
1997 | */ | |
8da11c19 | 1998 | static struct file *__io_file_get(struct io_submit_state *state, int fd) |
9a56a232 JA |
1999 | { |
2000 | if (!state) | |
2001 | return fget(fd); | |
2002 | ||
2003 | if (state->file) { | |
2004 | if (state->fd == fd) { | |
2005 | state->used_refs++; | |
2006 | state->ios_left--; | |
2007 | return state->file; | |
2008 | } | |
3d6770fb | 2009 | io_file_put(state); |
9a56a232 JA |
2010 | } |
2011 | state->file = fget_many(fd, state->ios_left); | |
2012 | if (!state->file) | |
2013 | return NULL; | |
2014 | ||
2015 | state->fd = fd; | |
2016 | state->has_refs = state->ios_left; | |
2017 | state->used_refs = 1; | |
2018 | state->ios_left--; | |
2019 | return state->file; | |
2020 | } | |
2021 | ||
2b188cc1 JA |
2022 | /* |
2023 | * If we tracked the file through the SCM inflight mechanism, we could support | |
2024 | * any file. For now, just ensure that anything potentially problematic is done | |
2025 | * inline. | |
2026 | */ | |
af197f50 | 2027 | static bool io_file_supports_async(struct file *file, int rw) |
2b188cc1 JA |
2028 | { |
2029 | umode_t mode = file_inode(file)->i_mode; | |
2030 | ||
10d59345 | 2031 | if (S_ISBLK(mode) || S_ISCHR(mode) || S_ISSOCK(mode)) |
2b188cc1 JA |
2032 | return true; |
2033 | if (S_ISREG(mode) && file->f_op != &io_uring_fops) | |
2034 | return true; | |
2035 | ||
af197f50 JA |
2036 | if (!(file->f_mode & FMODE_NOWAIT)) |
2037 | return false; | |
2038 | ||
2039 | if (rw == READ) | |
2040 | return file->f_op->read_iter != NULL; | |
2041 | ||
2042 | return file->f_op->write_iter != NULL; | |
2b188cc1 JA |
2043 | } |
2044 | ||
3529d8c2 JA |
2045 | static int io_prep_rw(struct io_kiocb *req, const struct io_uring_sqe *sqe, |
2046 | bool force_nonblock) | |
2b188cc1 | 2047 | { |
def596e9 | 2048 | struct io_ring_ctx *ctx = req->ctx; |
9adbd45d | 2049 | struct kiocb *kiocb = &req->rw.kiocb; |
09bb8394 JA |
2050 | unsigned ioprio; |
2051 | int ret; | |
2b188cc1 | 2052 | |
491381ce JA |
2053 | if (S_ISREG(file_inode(req->file)->i_mode)) |
2054 | req->flags |= REQ_F_ISREG; | |
2055 | ||
2b188cc1 | 2056 | kiocb->ki_pos = READ_ONCE(sqe->off); |
ba04291e JA |
2057 | if (kiocb->ki_pos == -1 && !(req->file->f_mode & FMODE_STREAM)) { |
2058 | req->flags |= REQ_F_CUR_POS; | |
2059 | kiocb->ki_pos = req->file->f_pos; | |
2060 | } | |
2b188cc1 | 2061 | kiocb->ki_hint = ki_hint_validate(file_write_hint(kiocb->ki_filp)); |
3e577dcd PB |
2062 | kiocb->ki_flags = iocb_flags(kiocb->ki_filp); |
2063 | ret = kiocb_set_rw_flags(kiocb, READ_ONCE(sqe->rw_flags)); | |
2064 | if (unlikely(ret)) | |
2065 | return ret; | |
2b188cc1 JA |
2066 | |
2067 | ioprio = READ_ONCE(sqe->ioprio); | |
2068 | if (ioprio) { | |
2069 | ret = ioprio_check_cap(ioprio); | |
2070 | if (ret) | |
09bb8394 | 2071 | return ret; |
2b188cc1 JA |
2072 | |
2073 | kiocb->ki_ioprio = ioprio; | |
2074 | } else | |
2075 | kiocb->ki_ioprio = get_current_ioprio(); | |
2076 | ||
8449eeda | 2077 | /* don't allow async punt if RWF_NOWAIT was requested */ |
491381ce JA |
2078 | if ((kiocb->ki_flags & IOCB_NOWAIT) || |
2079 | (req->file->f_flags & O_NONBLOCK)) | |
8449eeda SB |
2080 | req->flags |= REQ_F_NOWAIT; |
2081 | ||
2082 | if (force_nonblock) | |
2b188cc1 | 2083 | kiocb->ki_flags |= IOCB_NOWAIT; |
8449eeda | 2084 | |
def596e9 | 2085 | if (ctx->flags & IORING_SETUP_IOPOLL) { |
def596e9 JA |
2086 | if (!(kiocb->ki_flags & IOCB_DIRECT) || |
2087 | !kiocb->ki_filp->f_op->iopoll) | |
09bb8394 | 2088 | return -EOPNOTSUPP; |
2b188cc1 | 2089 | |
def596e9 JA |
2090 | kiocb->ki_flags |= IOCB_HIPRI; |
2091 | kiocb->ki_complete = io_complete_rw_iopoll; | |
6873e0bd | 2092 | req->result = 0; |
def596e9 | 2093 | } else { |
09bb8394 JA |
2094 | if (kiocb->ki_flags & IOCB_HIPRI) |
2095 | return -EINVAL; | |
def596e9 JA |
2096 | kiocb->ki_complete = io_complete_rw; |
2097 | } | |
9adbd45d | 2098 | |
3529d8c2 JA |
2099 | req->rw.addr = READ_ONCE(sqe->addr); |
2100 | req->rw.len = READ_ONCE(sqe->len); | |
bcda7baa | 2101 | /* we own ->private, reuse it for the buffer index / buffer ID */ |
9adbd45d | 2102 | req->rw.kiocb.private = (void *) (unsigned long) |
3529d8c2 | 2103 | READ_ONCE(sqe->buf_index); |
2b188cc1 | 2104 | return 0; |
2b188cc1 JA |
2105 | } |
2106 | ||
2107 | static inline void io_rw_done(struct kiocb *kiocb, ssize_t ret) | |
2108 | { | |
2109 | switch (ret) { | |
2110 | case -EIOCBQUEUED: | |
2111 | break; | |
2112 | case -ERESTARTSYS: | |
2113 | case -ERESTARTNOINTR: | |
2114 | case -ERESTARTNOHAND: | |
2115 | case -ERESTART_RESTARTBLOCK: | |
2116 | /* | |
2117 | * We can't just restart the syscall, since previously | |
2118 | * submitted sqes may already be in progress. Just fail this | |
2119 | * IO with EINTR. | |
2120 | */ | |
2121 | ret = -EINTR; | |
2122 | /* fall through */ | |
2123 | default: | |
2124 | kiocb->ki_complete(kiocb, ret, 0); | |
2125 | } | |
2126 | } | |
2127 | ||
014db007 | 2128 | static void kiocb_done(struct kiocb *kiocb, ssize_t ret) |
ba816ad6 | 2129 | { |
ba04291e JA |
2130 | struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb); |
2131 | ||
2132 | if (req->flags & REQ_F_CUR_POS) | |
2133 | req->file->f_pos = kiocb->ki_pos; | |
bcaec089 | 2134 | if (ret >= 0 && kiocb->ki_complete == io_complete_rw) |
014db007 | 2135 | io_complete_rw(kiocb, ret, 0); |
ba816ad6 JA |
2136 | else |
2137 | io_rw_done(kiocb, ret); | |
2138 | } | |
2139 | ||
9adbd45d | 2140 | static ssize_t io_import_fixed(struct io_kiocb *req, int rw, |
7d009165 | 2141 | struct iov_iter *iter) |
edafccee | 2142 | { |
9adbd45d JA |
2143 | struct io_ring_ctx *ctx = req->ctx; |
2144 | size_t len = req->rw.len; | |
edafccee JA |
2145 | struct io_mapped_ubuf *imu; |
2146 | unsigned index, buf_index; | |
2147 | size_t offset; | |
2148 | u64 buf_addr; | |
2149 | ||
2150 | /* attempt to use fixed buffers without having provided iovecs */ | |
2151 | if (unlikely(!ctx->user_bufs)) | |
2152 | return -EFAULT; | |
2153 | ||
9adbd45d | 2154 | buf_index = (unsigned long) req->rw.kiocb.private; |
edafccee JA |
2155 | if (unlikely(buf_index >= ctx->nr_user_bufs)) |
2156 | return -EFAULT; | |
2157 | ||
2158 | index = array_index_nospec(buf_index, ctx->nr_user_bufs); | |
2159 | imu = &ctx->user_bufs[index]; | |
9adbd45d | 2160 | buf_addr = req->rw.addr; |
edafccee JA |
2161 | |
2162 | /* overflow */ | |
2163 | if (buf_addr + len < buf_addr) | |
2164 | return -EFAULT; | |
2165 | /* not inside the mapped region */ | |
2166 | if (buf_addr < imu->ubuf || buf_addr + len > imu->ubuf + imu->len) | |
2167 | return -EFAULT; | |
2168 | ||
2169 | /* | |
2170 | * May not be a start of buffer, set size appropriately | |
2171 | * and advance us to the beginning. | |
2172 | */ | |
2173 | offset = buf_addr - imu->ubuf; | |
2174 | iov_iter_bvec(iter, rw, imu->bvec, imu->nr_bvecs, offset + len); | |
bd11b3a3 JA |
2175 | |
2176 | if (offset) { | |
2177 | /* | |
2178 | * Don't use iov_iter_advance() here, as it's really slow for | |
2179 | * using the latter parts of a big fixed buffer - it iterates | |
2180 | * over each segment manually. We can cheat a bit here, because | |
2181 | * we know that: | |
2182 | * | |
2183 | * 1) it's a BVEC iter, we set it up | |
2184 | * 2) all bvecs are PAGE_SIZE in size, except potentially the | |
2185 | * first and last bvec | |
2186 | * | |
2187 | * So just find our index, and adjust the iterator afterwards. | |
2188 | * If the offset is within the first bvec (or the whole first | |
2189 | * bvec, just use iov_iter_advance(). This makes it easier | |
2190 | * since we can just skip the first segment, which may not | |
2191 | * be PAGE_SIZE aligned. | |
2192 | */ | |
2193 | const struct bio_vec *bvec = imu->bvec; | |
2194 | ||
2195 | if (offset <= bvec->bv_len) { | |
2196 | iov_iter_advance(iter, offset); | |
2197 | } else { | |
2198 | unsigned long seg_skip; | |
2199 | ||
2200 | /* skip first vec */ | |
2201 | offset -= bvec->bv_len; | |
2202 | seg_skip = 1 + (offset >> PAGE_SHIFT); | |
2203 | ||
2204 | iter->bvec = bvec + seg_skip; | |
2205 | iter->nr_segs -= seg_skip; | |
99c79f66 | 2206 | iter->count -= bvec->bv_len + offset; |
bd11b3a3 | 2207 | iter->iov_offset = offset & ~PAGE_MASK; |
bd11b3a3 JA |
2208 | } |
2209 | } | |
2210 | ||
5e559561 | 2211 | return len; |
edafccee JA |
2212 | } |
2213 | ||
bcda7baa JA |
2214 | static void io_ring_submit_unlock(struct io_ring_ctx *ctx, bool needs_lock) |
2215 | { | |
2216 | if (needs_lock) | |
2217 | mutex_unlock(&ctx->uring_lock); | |
2218 | } | |
2219 | ||
2220 | static void io_ring_submit_lock(struct io_ring_ctx *ctx, bool needs_lock) | |
2221 | { | |
2222 | /* | |
2223 | * "Normal" inline submissions always hold the uring_lock, since we | |
2224 | * grab it from the system call. Same is true for the SQPOLL offload. | |
2225 | * The only exception is when we've detached the request and issue it | |
2226 | * from an async worker thread, grab the lock for that case. | |
2227 | */ | |
2228 | if (needs_lock) | |
2229 | mutex_lock(&ctx->uring_lock); | |
2230 | } | |
2231 | ||
2232 | static struct io_buffer *io_buffer_select(struct io_kiocb *req, size_t *len, | |
2233 | int bgid, struct io_buffer *kbuf, | |
2234 | bool needs_lock) | |
2235 | { | |
2236 | struct io_buffer *head; | |
2237 | ||
2238 | if (req->flags & REQ_F_BUFFER_SELECTED) | |
2239 | return kbuf; | |
2240 | ||
2241 | io_ring_submit_lock(req->ctx, needs_lock); | |
2242 | ||
2243 | lockdep_assert_held(&req->ctx->uring_lock); | |
2244 | ||
2245 | head = idr_find(&req->ctx->io_buffer_idr, bgid); | |
2246 | if (head) { | |
2247 | if (!list_empty(&head->list)) { | |
2248 | kbuf = list_last_entry(&head->list, struct io_buffer, | |
2249 | list); | |
2250 | list_del(&kbuf->list); | |
2251 | } else { | |
2252 | kbuf = head; | |
2253 | idr_remove(&req->ctx->io_buffer_idr, bgid); | |
2254 | } | |
2255 | if (*len > kbuf->len) | |
2256 | *len = kbuf->len; | |
2257 | } else { | |
2258 | kbuf = ERR_PTR(-ENOBUFS); | |
2259 | } | |
2260 | ||
2261 | io_ring_submit_unlock(req->ctx, needs_lock); | |
2262 | ||
2263 | return kbuf; | |
2264 | } | |
2265 | ||
4d954c25 JA |
2266 | static void __user *io_rw_buffer_select(struct io_kiocb *req, size_t *len, |
2267 | bool needs_lock) | |
2268 | { | |
2269 | struct io_buffer *kbuf; | |
2270 | int bgid; | |
2271 | ||
2272 | kbuf = (struct io_buffer *) (unsigned long) req->rw.addr; | |
2273 | bgid = (int) (unsigned long) req->rw.kiocb.private; | |
2274 | kbuf = io_buffer_select(req, len, bgid, kbuf, needs_lock); | |
2275 | if (IS_ERR(kbuf)) | |
2276 | return kbuf; | |
2277 | req->rw.addr = (u64) (unsigned long) kbuf; | |
2278 | req->flags |= REQ_F_BUFFER_SELECTED; | |
2279 | return u64_to_user_ptr(kbuf->addr); | |
2280 | } | |
2281 | ||
2282 | #ifdef CONFIG_COMPAT | |
2283 | static ssize_t io_compat_import(struct io_kiocb *req, struct iovec *iov, | |
2284 | bool needs_lock) | |
2285 | { | |
2286 | struct compat_iovec __user *uiov; | |
2287 | compat_ssize_t clen; | |
2288 | void __user *buf; | |
2289 | ssize_t len; | |
2290 | ||
2291 | uiov = u64_to_user_ptr(req->rw.addr); | |
2292 | if (!access_ok(uiov, sizeof(*uiov))) | |
2293 | return -EFAULT; | |
2294 | if (__get_user(clen, &uiov->iov_len)) | |
2295 | return -EFAULT; | |
2296 | if (clen < 0) | |
2297 | return -EINVAL; | |
2298 | ||
2299 | len = clen; | |
2300 | buf = io_rw_buffer_select(req, &len, needs_lock); | |
2301 | if (IS_ERR(buf)) | |
2302 | return PTR_ERR(buf); | |
2303 | iov[0].iov_base = buf; | |
2304 | iov[0].iov_len = (compat_size_t) len; | |
2305 | return 0; | |
2306 | } | |
2307 | #endif | |
2308 | ||
2309 | static ssize_t __io_iov_buffer_select(struct io_kiocb *req, struct iovec *iov, | |
2310 | bool needs_lock) | |
2311 | { | |
2312 | struct iovec __user *uiov = u64_to_user_ptr(req->rw.addr); | |
2313 | void __user *buf; | |
2314 | ssize_t len; | |
2315 | ||
2316 | if (copy_from_user(iov, uiov, sizeof(*uiov))) | |
2317 | return -EFAULT; | |
2318 | ||
2319 | len = iov[0].iov_len; | |
2320 | if (len < 0) | |
2321 | return -EINVAL; | |
2322 | buf = io_rw_buffer_select(req, &len, needs_lock); | |
2323 | if (IS_ERR(buf)) | |
2324 | return PTR_ERR(buf); | |
2325 | iov[0].iov_base = buf; | |
2326 | iov[0].iov_len = len; | |
2327 | return 0; | |
2328 | } | |
2329 | ||
2330 | static ssize_t io_iov_buffer_select(struct io_kiocb *req, struct iovec *iov, | |
2331 | bool needs_lock) | |
2332 | { | |
2333 | if (req->flags & REQ_F_BUFFER_SELECTED) | |
2334 | return 0; | |
2335 | if (!req->rw.len) | |
2336 | return 0; | |
2337 | else if (req->rw.len > 1) | |
2338 | return -EINVAL; | |
2339 | ||
2340 | #ifdef CONFIG_COMPAT | |
2341 | if (req->ctx->compat) | |
2342 | return io_compat_import(req, iov, needs_lock); | |
2343 | #endif | |
2344 | ||
2345 | return __io_iov_buffer_select(req, iov, needs_lock); | |
2346 | } | |
2347 | ||
cf6fd4bd | 2348 | static ssize_t io_import_iovec(int rw, struct io_kiocb *req, |
bcda7baa JA |
2349 | struct iovec **iovec, struct iov_iter *iter, |
2350 | bool needs_lock) | |
2b188cc1 | 2351 | { |
9adbd45d JA |
2352 | void __user *buf = u64_to_user_ptr(req->rw.addr); |
2353 | size_t sqe_len = req->rw.len; | |
4d954c25 | 2354 | ssize_t ret; |
edafccee JA |
2355 | u8 opcode; |
2356 | ||
d625c6ee | 2357 | opcode = req->opcode; |
7d009165 | 2358 | if (opcode == IORING_OP_READ_FIXED || opcode == IORING_OP_WRITE_FIXED) { |
edafccee | 2359 | *iovec = NULL; |
9adbd45d | 2360 | return io_import_fixed(req, rw, iter); |
edafccee | 2361 | } |
2b188cc1 | 2362 | |
bcda7baa JA |
2363 | /* buffer index only valid with fixed read/write, or buffer select */ |
2364 | if (req->rw.kiocb.private && !(req->flags & REQ_F_BUFFER_SELECT)) | |
9adbd45d JA |
2365 | return -EINVAL; |
2366 | ||
3a6820f2 | 2367 | if (opcode == IORING_OP_READ || opcode == IORING_OP_WRITE) { |
bcda7baa | 2368 | if (req->flags & REQ_F_BUFFER_SELECT) { |
4d954c25 JA |
2369 | buf = io_rw_buffer_select(req, &sqe_len, needs_lock); |
2370 | if (IS_ERR(buf)) { | |
bcda7baa | 2371 | *iovec = NULL; |
4d954c25 | 2372 | return PTR_ERR(buf); |
bcda7baa | 2373 | } |
3f9d6441 | 2374 | req->rw.len = sqe_len; |
bcda7baa JA |
2375 | } |
2376 | ||
3a6820f2 JA |
2377 | ret = import_single_range(rw, buf, sqe_len, *iovec, iter); |
2378 | *iovec = NULL; | |
3a901598 | 2379 | return ret < 0 ? ret : sqe_len; |
3a6820f2 JA |
2380 | } |
2381 | ||
f67676d1 JA |
2382 | if (req->io) { |
2383 | struct io_async_rw *iorw = &req->io->rw; | |
2384 | ||
2385 | *iovec = iorw->iov; | |
2386 | iov_iter_init(iter, rw, *iovec, iorw->nr_segs, iorw->size); | |
2387 | if (iorw->iov == iorw->fast_iov) | |
2388 | *iovec = NULL; | |
2389 | return iorw->size; | |
2390 | } | |
2391 | ||
4d954c25 JA |
2392 | if (req->flags & REQ_F_BUFFER_SELECT) { |
2393 | ret = io_iov_buffer_select(req, *iovec, needs_lock); | |
3f9d6441 JA |
2394 | if (!ret) { |
2395 | ret = (*iovec)->iov_len; | |
2396 | iov_iter_init(iter, rw, *iovec, 1, ret); | |
2397 | } | |
4d954c25 JA |
2398 | *iovec = NULL; |
2399 | return ret; | |
2400 | } | |
2401 | ||
2b188cc1 | 2402 | #ifdef CONFIG_COMPAT |
cf6fd4bd | 2403 | if (req->ctx->compat) |
2b188cc1 JA |
2404 | return compat_import_iovec(rw, buf, sqe_len, UIO_FASTIOV, |
2405 | iovec, iter); | |
2406 | #endif | |
2407 | ||
2408 | return import_iovec(rw, buf, sqe_len, UIO_FASTIOV, iovec, iter); | |
2409 | } | |
2410 | ||
31b51510 | 2411 | /* |
32960613 JA |
2412 | * For files that don't have ->read_iter() and ->write_iter(), handle them |
2413 | * by looping over ->read() or ->write() manually. | |
31b51510 | 2414 | */ |
32960613 JA |
2415 | static ssize_t loop_rw_iter(int rw, struct file *file, struct kiocb *kiocb, |
2416 | struct iov_iter *iter) | |
2417 | { | |
2418 | ssize_t ret = 0; | |
2419 | ||
2420 | /* | |
2421 | * Don't support polled IO through this interface, and we can't | |
2422 | * support non-blocking either. For the latter, this just causes | |
2423 | * the kiocb to be handled from an async context. | |
2424 | */ | |
2425 | if (kiocb->ki_flags & IOCB_HIPRI) | |
2426 | return -EOPNOTSUPP; | |
2427 | if (kiocb->ki_flags & IOCB_NOWAIT) | |
2428 | return -EAGAIN; | |
2429 | ||
2430 | while (iov_iter_count(iter)) { | |
311ae9e1 | 2431 | struct iovec iovec; |
32960613 JA |
2432 | ssize_t nr; |
2433 | ||
311ae9e1 PB |
2434 | if (!iov_iter_is_bvec(iter)) { |
2435 | iovec = iov_iter_iovec(iter); | |
2436 | } else { | |
2437 | /* fixed buffers import bvec */ | |
2438 | iovec.iov_base = kmap(iter->bvec->bv_page) | |
2439 | + iter->iov_offset; | |
2440 | iovec.iov_len = min(iter->count, | |
2441 | iter->bvec->bv_len - iter->iov_offset); | |
2442 | } | |
2443 | ||
32960613 JA |
2444 | if (rw == READ) { |
2445 | nr = file->f_op->read(file, iovec.iov_base, | |
2446 | iovec.iov_len, &kiocb->ki_pos); | |
2447 | } else { | |
2448 | nr = file->f_op->write(file, iovec.iov_base, | |
2449 | iovec.iov_len, &kiocb->ki_pos); | |
2450 | } | |
2451 | ||
311ae9e1 PB |
2452 | if (iov_iter_is_bvec(iter)) |
2453 | kunmap(iter->bvec->bv_page); | |
2454 | ||
32960613 JA |
2455 | if (nr < 0) { |
2456 | if (!ret) | |
2457 | ret = nr; | |
2458 | break; | |
2459 | } | |
2460 | ret += nr; | |
2461 | if (nr != iovec.iov_len) | |
2462 | break; | |
2463 | iov_iter_advance(iter, nr); | |
2464 | } | |
2465 | ||
2466 | return ret; | |
2467 | } | |
2468 | ||
b7bb4f7d | 2469 | static void io_req_map_rw(struct io_kiocb *req, ssize_t io_size, |
f67676d1 JA |
2470 | struct iovec *iovec, struct iovec *fast_iov, |
2471 | struct iov_iter *iter) | |
2472 | { | |
2473 | req->io->rw.nr_segs = iter->nr_segs; | |
2474 | req->io->rw.size = io_size; | |
2475 | req->io->rw.iov = iovec; | |
2476 | if (!req->io->rw.iov) { | |
2477 | req->io->rw.iov = req->io->rw.fast_iov; | |
45097dae XW |
2478 | if (req->io->rw.iov != fast_iov) |
2479 | memcpy(req->io->rw.iov, fast_iov, | |
2480 | sizeof(struct iovec) * iter->nr_segs); | |
99bc4c38 PB |
2481 | } else { |
2482 | req->flags |= REQ_F_NEED_CLEANUP; | |
f67676d1 JA |
2483 | } |
2484 | } | |
2485 | ||
3d9932a8 XW |
2486 | static inline int __io_alloc_async_ctx(struct io_kiocb *req) |
2487 | { | |
2488 | req->io = kmalloc(sizeof(*req->io), GFP_KERNEL); | |
2489 | return req->io == NULL; | |
2490 | } | |
2491 | ||
b7bb4f7d | 2492 | static int io_alloc_async_ctx(struct io_kiocb *req) |
f67676d1 | 2493 | { |
d3656344 JA |
2494 | if (!io_op_defs[req->opcode].async_ctx) |
2495 | return 0; | |
3d9932a8 XW |
2496 | |
2497 | return __io_alloc_async_ctx(req); | |
b7bb4f7d JA |
2498 | } |
2499 | ||
b7bb4f7d JA |
2500 | static int io_setup_async_rw(struct io_kiocb *req, ssize_t io_size, |
2501 | struct iovec *iovec, struct iovec *fast_iov, | |
2502 | struct iov_iter *iter) | |
2503 | { | |
980ad263 | 2504 | if (!io_op_defs[req->opcode].async_ctx) |
74566df3 | 2505 | return 0; |
5d204bcf | 2506 | if (!req->io) { |
3d9932a8 | 2507 | if (__io_alloc_async_ctx(req)) |
5d204bcf | 2508 | return -ENOMEM; |
b7bb4f7d | 2509 | |
5d204bcf JA |
2510 | io_req_map_rw(req, io_size, iovec, fast_iov, iter); |
2511 | } | |
b7bb4f7d | 2512 | return 0; |
f67676d1 JA |
2513 | } |
2514 | ||
3529d8c2 JA |
2515 | static int io_read_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe, |
2516 | bool force_nonblock) | |
f67676d1 | 2517 | { |
3529d8c2 JA |
2518 | struct io_async_ctx *io; |
2519 | struct iov_iter iter; | |
f67676d1 JA |
2520 | ssize_t ret; |
2521 | ||
3529d8c2 JA |
2522 | ret = io_prep_rw(req, sqe, force_nonblock); |
2523 | if (ret) | |
2524 | return ret; | |
f67676d1 | 2525 | |
3529d8c2 JA |
2526 | if (unlikely(!(req->file->f_mode & FMODE_READ))) |
2527 | return -EBADF; | |
f67676d1 | 2528 | |
5f798bea PB |
2529 | /* either don't need iovec imported or already have it */ |
2530 | if (!req->io || req->flags & REQ_F_NEED_CLEANUP) | |
3529d8c2 JA |
2531 | return 0; |
2532 | ||
2533 | io = req->io; | |
2534 | io->rw.iov = io->rw.fast_iov; | |
2535 | req->io = NULL; | |
bcda7baa | 2536 | ret = io_import_iovec(READ, req, &io->rw.iov, &iter, !force_nonblock); |
3529d8c2 JA |
2537 | req->io = io; |
2538 | if (ret < 0) | |
2539 | return ret; | |
2540 | ||
2541 | io_req_map_rw(req, ret, io->rw.iov, io->rw.fast_iov, &iter); | |
2542 | return 0; | |
f67676d1 JA |
2543 | } |
2544 | ||
014db007 | 2545 | static int io_read(struct io_kiocb *req, bool force_nonblock) |
2b188cc1 JA |
2546 | { |
2547 | struct iovec inline_vecs[UIO_FASTIOV], *iovec = inline_vecs; | |
9adbd45d | 2548 | struct kiocb *kiocb = &req->rw.kiocb; |
2b188cc1 | 2549 | struct iov_iter iter; |
31b51510 | 2550 | size_t iov_count; |
f67676d1 | 2551 | ssize_t io_size, ret; |
2b188cc1 | 2552 | |
bcda7baa | 2553 | ret = io_import_iovec(READ, req, &iovec, &iter, !force_nonblock); |
06b76d44 JA |
2554 | if (ret < 0) |
2555 | return ret; | |
2b188cc1 | 2556 | |
fd6c2e4c JA |
2557 | /* Ensure we clear previously set non-block flag */ |
2558 | if (!force_nonblock) | |
29de5f6a | 2559 | kiocb->ki_flags &= ~IOCB_NOWAIT; |
fd6c2e4c | 2560 | |
797f3f53 | 2561 | req->result = 0; |
f67676d1 | 2562 | io_size = ret; |
dea3b49c | 2563 | if (req->flags & REQ_F_LINK_HEAD) |
f67676d1 JA |
2564 | req->result = io_size; |
2565 | ||
2566 | /* | |
2567 | * If the file doesn't support async, mark it as REQ_F_MUST_PUNT so | |
2568 | * we know to async punt it even if it was opened O_NONBLOCK | |
2569 | */ | |
af197f50 | 2570 | if (force_nonblock && !io_file_supports_async(req->file, READ)) |
f67676d1 | 2571 | goto copy_iov; |
9e645e11 | 2572 | |
31b51510 | 2573 | iov_count = iov_iter_count(&iter); |
9adbd45d | 2574 | ret = rw_verify_area(READ, req->file, &kiocb->ki_pos, iov_count); |
2b188cc1 JA |
2575 | if (!ret) { |
2576 | ssize_t ret2; | |
2577 | ||
9adbd45d JA |
2578 | if (req->file->f_op->read_iter) |
2579 | ret2 = call_read_iter(req->file, kiocb, &iter); | |
32960613 | 2580 | else |
9adbd45d | 2581 | ret2 = loop_rw_iter(READ, req->file, kiocb, &iter); |
32960613 | 2582 | |
9d93a3f5 | 2583 | /* Catch -EAGAIN return for forced non-blocking submission */ |
f67676d1 | 2584 | if (!force_nonblock || ret2 != -EAGAIN) { |
014db007 | 2585 | kiocb_done(kiocb, ret2); |
f67676d1 JA |
2586 | } else { |
2587 | copy_iov: | |
b7bb4f7d | 2588 | ret = io_setup_async_rw(req, io_size, iovec, |
f67676d1 JA |
2589 | inline_vecs, &iter); |
2590 | if (ret) | |
2591 | goto out_free; | |
29de5f6a | 2592 | /* any defer here is final, must blocking retry */ |
490e8967 JA |
2593 | if (!(req->flags & REQ_F_NOWAIT) && |
2594 | !file_can_poll(req->file)) | |
29de5f6a | 2595 | req->flags |= REQ_F_MUST_PUNT; |
f67676d1 JA |
2596 | return -EAGAIN; |
2597 | } | |
2b188cc1 | 2598 | } |
f67676d1 | 2599 | out_free: |
1e95081c | 2600 | kfree(iovec); |
99bc4c38 | 2601 | req->flags &= ~REQ_F_NEED_CLEANUP; |
2b188cc1 JA |
2602 | return ret; |
2603 | } | |
2604 | ||
3529d8c2 JA |
2605 | static int io_write_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe, |
2606 | bool force_nonblock) | |
f67676d1 | 2607 | { |
3529d8c2 JA |
2608 | struct io_async_ctx *io; |
2609 | struct iov_iter iter; | |
f67676d1 JA |
2610 | ssize_t ret; |
2611 | ||
3529d8c2 JA |
2612 | ret = io_prep_rw(req, sqe, force_nonblock); |
2613 | if (ret) | |
2614 | return ret; | |
f67676d1 | 2615 | |
3529d8c2 JA |
2616 | if (unlikely(!(req->file->f_mode & FMODE_WRITE))) |
2617 | return -EBADF; | |
f67676d1 | 2618 | |
4ed734b0 JA |
2619 | req->fsize = rlimit(RLIMIT_FSIZE); |
2620 | ||
5f798bea PB |
2621 | /* either don't need iovec imported or already have it */ |
2622 | if (!req->io || req->flags & REQ_F_NEED_CLEANUP) | |
3529d8c2 JA |
2623 | return 0; |
2624 | ||
2625 | io = req->io; | |
2626 | io->rw.iov = io->rw.fast_iov; | |
2627 | req->io = NULL; | |
bcda7baa | 2628 | ret = io_import_iovec(WRITE, req, &io->rw.iov, &iter, !force_nonblock); |
3529d8c2 JA |
2629 | req->io = io; |
2630 | if (ret < 0) | |
2631 | return ret; | |
2632 | ||
2633 | io_req_map_rw(req, ret, io->rw.iov, io->rw.fast_iov, &iter); | |
2634 | return 0; | |
f67676d1 JA |
2635 | } |
2636 | ||
014db007 | 2637 | static int io_write(struct io_kiocb *req, bool force_nonblock) |
2b188cc1 JA |
2638 | { |
2639 | struct iovec inline_vecs[UIO_FASTIOV], *iovec = inline_vecs; | |
9adbd45d | 2640 | struct kiocb *kiocb = &req->rw.kiocb; |
2b188cc1 | 2641 | struct iov_iter iter; |
31b51510 | 2642 | size_t iov_count; |
f67676d1 | 2643 | ssize_t ret, io_size; |
2b188cc1 | 2644 | |
bcda7baa | 2645 | ret = io_import_iovec(WRITE, req, &iovec, &iter, !force_nonblock); |
06b76d44 JA |
2646 | if (ret < 0) |
2647 | return ret; | |
2b188cc1 | 2648 | |
fd6c2e4c JA |
2649 | /* Ensure we clear previously set non-block flag */ |
2650 | if (!force_nonblock) | |
9adbd45d | 2651 | req->rw.kiocb.ki_flags &= ~IOCB_NOWAIT; |
fd6c2e4c | 2652 | |
797f3f53 | 2653 | req->result = 0; |
f67676d1 | 2654 | io_size = ret; |
dea3b49c | 2655 | if (req->flags & REQ_F_LINK_HEAD) |
f67676d1 | 2656 | req->result = io_size; |
9e645e11 | 2657 | |
f67676d1 JA |
2658 | /* |
2659 | * If the file doesn't support async, mark it as REQ_F_MUST_PUNT so | |
2660 | * we know to async punt it even if it was opened O_NONBLOCK | |
2661 | */ | |
af197f50 | 2662 | if (force_nonblock && !io_file_supports_async(req->file, WRITE)) |
f67676d1 | 2663 | goto copy_iov; |
31b51510 | 2664 | |
10d59345 JA |
2665 | /* file path doesn't support NOWAIT for non-direct_IO */ |
2666 | if (force_nonblock && !(kiocb->ki_flags & IOCB_DIRECT) && | |
2667 | (req->flags & REQ_F_ISREG)) | |
f67676d1 | 2668 | goto copy_iov; |
31b51510 | 2669 | |
f67676d1 | 2670 | iov_count = iov_iter_count(&iter); |
9adbd45d | 2671 | ret = rw_verify_area(WRITE, req->file, &kiocb->ki_pos, iov_count); |
2b188cc1 | 2672 | if (!ret) { |
9bf7933f RP |
2673 | ssize_t ret2; |
2674 | ||
2b188cc1 JA |
2675 | /* |
2676 | * Open-code file_start_write here to grab freeze protection, | |
2677 | * which will be released by another thread in | |
2678 | * io_complete_rw(). Fool lockdep by telling it the lock got | |
2679 | * released so that it doesn't complain about the held lock when | |
2680 | * we return to userspace. | |
2681 | */ | |
491381ce | 2682 | if (req->flags & REQ_F_ISREG) { |
9adbd45d | 2683 | __sb_start_write(file_inode(req->file)->i_sb, |
2b188cc1 | 2684 | SB_FREEZE_WRITE, true); |
9adbd45d | 2685 | __sb_writers_release(file_inode(req->file)->i_sb, |
2b188cc1 JA |
2686 | SB_FREEZE_WRITE); |
2687 | } | |
2688 | kiocb->ki_flags |= IOCB_WRITE; | |
9bf7933f | 2689 | |
4ed734b0 JA |
2690 | if (!force_nonblock) |
2691 | current->signal->rlim[RLIMIT_FSIZE].rlim_cur = req->fsize; | |
2692 | ||
9adbd45d JA |
2693 | if (req->file->f_op->write_iter) |
2694 | ret2 = call_write_iter(req->file, kiocb, &iter); | |
32960613 | 2695 | else |
9adbd45d | 2696 | ret2 = loop_rw_iter(WRITE, req->file, kiocb, &iter); |
4ed734b0 JA |
2697 | |
2698 | if (!force_nonblock) | |
2699 | current->signal->rlim[RLIMIT_FSIZE].rlim_cur = RLIM_INFINITY; | |
2700 | ||
faac996c | 2701 | /* |
bff6035d | 2702 | * Raw bdev writes will return -EOPNOTSUPP for IOCB_NOWAIT. Just |
faac996c JA |
2703 | * retry them without IOCB_NOWAIT. |
2704 | */ | |
2705 | if (ret2 == -EOPNOTSUPP && (kiocb->ki_flags & IOCB_NOWAIT)) | |
2706 | ret2 = -EAGAIN; | |
f67676d1 | 2707 | if (!force_nonblock || ret2 != -EAGAIN) { |
014db007 | 2708 | kiocb_done(kiocb, ret2); |
f67676d1 JA |
2709 | } else { |
2710 | copy_iov: | |
b7bb4f7d | 2711 | ret = io_setup_async_rw(req, io_size, iovec, |
f67676d1 JA |
2712 | inline_vecs, &iter); |
2713 | if (ret) | |
2714 | goto out_free; | |
29de5f6a | 2715 | /* any defer here is final, must blocking retry */ |
490e8967 JA |
2716 | if (!file_can_poll(req->file)) |
2717 | req->flags |= REQ_F_MUST_PUNT; | |
f67676d1 JA |
2718 | return -EAGAIN; |
2719 | } | |
2b188cc1 | 2720 | } |
31b51510 | 2721 | out_free: |
99bc4c38 | 2722 | req->flags &= ~REQ_F_NEED_CLEANUP; |
1e95081c | 2723 | kfree(iovec); |
2b188cc1 JA |
2724 | return ret; |
2725 | } | |
2726 | ||
7d67af2c PB |
2727 | static int io_splice_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
2728 | { | |
2729 | struct io_splice* sp = &req->splice; | |
2730 | unsigned int valid_flags = SPLICE_F_FD_IN_FIXED | SPLICE_F_ALL; | |
2731 | int ret; | |
2732 | ||
2733 | if (req->flags & REQ_F_NEED_CLEANUP) | |
2734 | return 0; | |
2735 | ||
2736 | sp->file_in = NULL; | |
2737 | sp->off_in = READ_ONCE(sqe->splice_off_in); | |
2738 | sp->off_out = READ_ONCE(sqe->off); | |
2739 | sp->len = READ_ONCE(sqe->len); | |
2740 | sp->flags = READ_ONCE(sqe->splice_flags); | |
2741 | ||
2742 | if (unlikely(sp->flags & ~valid_flags)) | |
2743 | return -EINVAL; | |
2744 | ||
2745 | ret = io_file_get(NULL, req, READ_ONCE(sqe->splice_fd_in), &sp->file_in, | |
2746 | (sp->flags & SPLICE_F_FD_IN_FIXED)); | |
2747 | if (ret) | |
2748 | return ret; | |
2749 | req->flags |= REQ_F_NEED_CLEANUP; | |
2750 | ||
2751 | if (!S_ISREG(file_inode(sp->file_in)->i_mode)) | |
2752 | req->work.flags |= IO_WQ_WORK_UNBOUND; | |
2753 | ||
2754 | return 0; | |
2755 | } | |
2756 | ||
014db007 | 2757 | static int io_splice(struct io_kiocb *req, bool force_nonblock) |
7d67af2c PB |
2758 | { |
2759 | struct io_splice *sp = &req->splice; | |
2760 | struct file *in = sp->file_in; | |
2761 | struct file *out = sp->file_out; | |
2762 | unsigned int flags = sp->flags & ~SPLICE_F_FD_IN_FIXED; | |
2763 | loff_t *poff_in, *poff_out; | |
2764 | long ret; | |
2765 | ||
2fb3e822 PB |
2766 | if (force_nonblock) |
2767 | return -EAGAIN; | |
7d67af2c PB |
2768 | |
2769 | poff_in = (sp->off_in == -1) ? NULL : &sp->off_in; | |
2770 | poff_out = (sp->off_out == -1) ? NULL : &sp->off_out; | |
2771 | ret = do_splice(in, poff_in, out, poff_out, sp->len, flags); | |
2772 | if (force_nonblock && ret == -EAGAIN) | |
2773 | return -EAGAIN; | |
2774 | ||
2775 | io_put_file(req, in, (sp->flags & SPLICE_F_FD_IN_FIXED)); | |
2776 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
2777 | ||
2778 | io_cqring_add_event(req, ret); | |
2779 | if (ret != sp->len) | |
2780 | req_set_fail_links(req); | |
014db007 | 2781 | io_put_req(req); |
7d67af2c PB |
2782 | return 0; |
2783 | } | |
2784 | ||
2b188cc1 JA |
2785 | /* |
2786 | * IORING_OP_NOP just posts a completion event, nothing else. | |
2787 | */ | |
78e19bbe | 2788 | static int io_nop(struct io_kiocb *req) |
2b188cc1 JA |
2789 | { |
2790 | struct io_ring_ctx *ctx = req->ctx; | |
2b188cc1 | 2791 | |
def596e9 JA |
2792 | if (unlikely(ctx->flags & IORING_SETUP_IOPOLL)) |
2793 | return -EINVAL; | |
2794 | ||
78e19bbe | 2795 | io_cqring_add_event(req, 0); |
e65ef56d | 2796 | io_put_req(req); |
2b188cc1 JA |
2797 | return 0; |
2798 | } | |
2799 | ||
3529d8c2 | 2800 | static int io_prep_fsync(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
c992fe29 | 2801 | { |
6b06314c | 2802 | struct io_ring_ctx *ctx = req->ctx; |
c992fe29 | 2803 | |
09bb8394 JA |
2804 | if (!req->file) |
2805 | return -EBADF; | |
c992fe29 | 2806 | |
6b06314c | 2807 | if (unlikely(ctx->flags & IORING_SETUP_IOPOLL)) |
def596e9 | 2808 | return -EINVAL; |
edafccee | 2809 | if (unlikely(sqe->addr || sqe->ioprio || sqe->buf_index)) |
c992fe29 CH |
2810 | return -EINVAL; |
2811 | ||
8ed8d3c3 JA |
2812 | req->sync.flags = READ_ONCE(sqe->fsync_flags); |
2813 | if (unlikely(req->sync.flags & ~IORING_FSYNC_DATASYNC)) | |
2814 | return -EINVAL; | |
2815 | ||
2816 | req->sync.off = READ_ONCE(sqe->off); | |
2817 | req->sync.len = READ_ONCE(sqe->len); | |
c992fe29 CH |
2818 | return 0; |
2819 | } | |
2820 | ||
8ed8d3c3 JA |
2821 | static bool io_req_cancelled(struct io_kiocb *req) |
2822 | { | |
2823 | if (req->work.flags & IO_WQ_WORK_CANCEL) { | |
2824 | req_set_fail_links(req); | |
2825 | io_cqring_add_event(req, -ECANCELED); | |
2826 | io_put_req(req); | |
2827 | return true; | |
2828 | } | |
2829 | ||
2830 | return false; | |
2831 | } | |
2832 | ||
014db007 | 2833 | static void __io_fsync(struct io_kiocb *req) |
8ed8d3c3 | 2834 | { |
8ed8d3c3 | 2835 | loff_t end = req->sync.off + req->sync.len; |
8ed8d3c3 JA |
2836 | int ret; |
2837 | ||
9adbd45d | 2838 | ret = vfs_fsync_range(req->file, req->sync.off, |
8ed8d3c3 JA |
2839 | end > 0 ? end : LLONG_MAX, |
2840 | req->sync.flags & IORING_FSYNC_DATASYNC); | |
2841 | if (ret < 0) | |
2842 | req_set_fail_links(req); | |
2843 | io_cqring_add_event(req, ret); | |
014db007 | 2844 | io_put_req(req); |
8ed8d3c3 JA |
2845 | } |
2846 | ||
5ea62161 | 2847 | static void io_fsync_finish(struct io_wq_work **workptr) |
c992fe29 | 2848 | { |
5ea62161 | 2849 | struct io_kiocb *req = container_of(*workptr, struct io_kiocb, work); |
5ea62161 PB |
2850 | |
2851 | if (io_req_cancelled(req)) | |
2852 | return; | |
014db007 | 2853 | __io_fsync(req); |
e9fd9396 | 2854 | io_steal_work(req, workptr); |
8ed8d3c3 | 2855 | } |
c992fe29 | 2856 | |
014db007 | 2857 | static int io_fsync(struct io_kiocb *req, bool force_nonblock) |
c992fe29 | 2858 | { |
c992fe29 | 2859 | /* fsync always requires a blocking context */ |
8ed8d3c3 | 2860 | if (force_nonblock) { |
8ed8d3c3 | 2861 | req->work.func = io_fsync_finish; |
c992fe29 | 2862 | return -EAGAIN; |
8ed8d3c3 | 2863 | } |
014db007 | 2864 | __io_fsync(req); |
c992fe29 CH |
2865 | return 0; |
2866 | } | |
2867 | ||
014db007 | 2868 | static void __io_fallocate(struct io_kiocb *req) |
8ed8d3c3 | 2869 | { |
8ed8d3c3 JA |
2870 | int ret; |
2871 | ||
4ed734b0 | 2872 | current->signal->rlim[RLIMIT_FSIZE].rlim_cur = req->fsize; |
d63d1b5e JA |
2873 | ret = vfs_fallocate(req->file, req->sync.mode, req->sync.off, |
2874 | req->sync.len); | |
4ed734b0 | 2875 | current->signal->rlim[RLIMIT_FSIZE].rlim_cur = RLIM_INFINITY; |
8ed8d3c3 JA |
2876 | if (ret < 0) |
2877 | req_set_fail_links(req); | |
2878 | io_cqring_add_event(req, ret); | |
014db007 | 2879 | io_put_req(req); |
5ea62161 PB |
2880 | } |
2881 | ||
2882 | static void io_fallocate_finish(struct io_wq_work **workptr) | |
2883 | { | |
2884 | struct io_kiocb *req = container_of(*workptr, struct io_kiocb, work); | |
5ea62161 | 2885 | |
594506fe PB |
2886 | if (io_req_cancelled(req)) |
2887 | return; | |
014db007 | 2888 | __io_fallocate(req); |
e9fd9396 | 2889 | io_steal_work(req, workptr); |
5d17b4a4 JA |
2890 | } |
2891 | ||
d63d1b5e JA |
2892 | static int io_fallocate_prep(struct io_kiocb *req, |
2893 | const struct io_uring_sqe *sqe) | |
2894 | { | |
2895 | if (sqe->ioprio || sqe->buf_index || sqe->rw_flags) | |
2896 | return -EINVAL; | |
2897 | ||
2898 | req->sync.off = READ_ONCE(sqe->off); | |
2899 | req->sync.len = READ_ONCE(sqe->addr); | |
2900 | req->sync.mode = READ_ONCE(sqe->len); | |
4ed734b0 | 2901 | req->fsize = rlimit(RLIMIT_FSIZE); |
d63d1b5e JA |
2902 | return 0; |
2903 | } | |
2904 | ||
014db007 | 2905 | static int io_fallocate(struct io_kiocb *req, bool force_nonblock) |
5d17b4a4 | 2906 | { |
d63d1b5e | 2907 | /* fallocate always requiring blocking context */ |
8ed8d3c3 | 2908 | if (force_nonblock) { |
d63d1b5e | 2909 | req->work.func = io_fallocate_finish; |
5d17b4a4 | 2910 | return -EAGAIN; |
8ed8d3c3 | 2911 | } |
5d17b4a4 | 2912 | |
014db007 | 2913 | __io_fallocate(req); |
5d17b4a4 JA |
2914 | return 0; |
2915 | } | |
2916 | ||
15b71abe | 2917 | static int io_openat_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
b7bb4f7d | 2918 | { |
f8748881 | 2919 | const char __user *fname; |
15b71abe | 2920 | int ret; |
b7bb4f7d | 2921 | |
15b71abe JA |
2922 | if (sqe->ioprio || sqe->buf_index) |
2923 | return -EINVAL; | |
9c280f90 | 2924 | if (req->flags & REQ_F_FIXED_FILE) |
cf3040ca | 2925 | return -EBADF; |
0bdbdd08 PB |
2926 | if (req->flags & REQ_F_NEED_CLEANUP) |
2927 | return 0; | |
03b1230c | 2928 | |
15b71abe | 2929 | req->open.dfd = READ_ONCE(sqe->fd); |
c12cedf2 | 2930 | req->open.how.mode = READ_ONCE(sqe->len); |
f8748881 | 2931 | fname = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
c12cedf2 | 2932 | req->open.how.flags = READ_ONCE(sqe->open_flags); |
08a1d26e JA |
2933 | if (force_o_largefile()) |
2934 | req->open.how.flags |= O_LARGEFILE; | |
3529d8c2 | 2935 | |
f8748881 | 2936 | req->open.filename = getname(fname); |
15b71abe JA |
2937 | if (IS_ERR(req->open.filename)) { |
2938 | ret = PTR_ERR(req->open.filename); | |
2939 | req->open.filename = NULL; | |
2940 | return ret; | |
2941 | } | |
3529d8c2 | 2942 | |
4022e7af | 2943 | req->open.nofile = rlimit(RLIMIT_NOFILE); |
8fef80bf | 2944 | req->flags |= REQ_F_NEED_CLEANUP; |
15b71abe | 2945 | return 0; |
03b1230c JA |
2946 | } |
2947 | ||
cebdb986 | 2948 | static int io_openat2_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
aa1fa28f | 2949 | { |
cebdb986 JA |
2950 | struct open_how __user *how; |
2951 | const char __user *fname; | |
2952 | size_t len; | |
0fa03c62 JA |
2953 | int ret; |
2954 | ||
cebdb986 | 2955 | if (sqe->ioprio || sqe->buf_index) |
0fa03c62 | 2956 | return -EINVAL; |
9c280f90 | 2957 | if (req->flags & REQ_F_FIXED_FILE) |
cf3040ca | 2958 | return -EBADF; |
0bdbdd08 PB |
2959 | if (req->flags & REQ_F_NEED_CLEANUP) |
2960 | return 0; | |
0fa03c62 | 2961 | |
cebdb986 JA |
2962 | req->open.dfd = READ_ONCE(sqe->fd); |
2963 | fname = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
2964 | how = u64_to_user_ptr(READ_ONCE(sqe->addr2)); | |
2965 | len = READ_ONCE(sqe->len); | |
0fa03c62 | 2966 | |
cebdb986 JA |
2967 | if (len < OPEN_HOW_SIZE_VER0) |
2968 | return -EINVAL; | |
3529d8c2 | 2969 | |
cebdb986 JA |
2970 | ret = copy_struct_from_user(&req->open.how, sizeof(req->open.how), how, |
2971 | len); | |
2972 | if (ret) | |
2973 | return ret; | |
3529d8c2 | 2974 | |
cebdb986 JA |
2975 | if (!(req->open.how.flags & O_PATH) && force_o_largefile()) |
2976 | req->open.how.flags |= O_LARGEFILE; | |
0fa03c62 | 2977 | |
cebdb986 JA |
2978 | req->open.filename = getname(fname); |
2979 | if (IS_ERR(req->open.filename)) { | |
2980 | ret = PTR_ERR(req->open.filename); | |
2981 | req->open.filename = NULL; | |
2982 | return ret; | |
2983 | } | |
2984 | ||
4022e7af | 2985 | req->open.nofile = rlimit(RLIMIT_NOFILE); |
8fef80bf | 2986 | req->flags |= REQ_F_NEED_CLEANUP; |
cebdb986 JA |
2987 | return 0; |
2988 | } | |
2989 | ||
014db007 | 2990 | static int io_openat2(struct io_kiocb *req, bool force_nonblock) |
15b71abe JA |
2991 | { |
2992 | struct open_flags op; | |
15b71abe JA |
2993 | struct file *file; |
2994 | int ret; | |
2995 | ||
f86cd20c | 2996 | if (force_nonblock) |
15b71abe | 2997 | return -EAGAIN; |
15b71abe | 2998 | |
cebdb986 | 2999 | ret = build_open_flags(&req->open.how, &op); |
15b71abe JA |
3000 | if (ret) |
3001 | goto err; | |
3002 | ||
4022e7af | 3003 | ret = __get_unused_fd_flags(req->open.how.flags, req->open.nofile); |
15b71abe JA |
3004 | if (ret < 0) |
3005 | goto err; | |
3006 | ||
3007 | file = do_filp_open(req->open.dfd, req->open.filename, &op); | |
3008 | if (IS_ERR(file)) { | |
3009 | put_unused_fd(ret); | |
3010 | ret = PTR_ERR(file); | |
3011 | } else { | |
3012 | fsnotify_open(file); | |
3013 | fd_install(ret, file); | |
3014 | } | |
3015 | err: | |
3016 | putname(req->open.filename); | |
8fef80bf | 3017 | req->flags &= ~REQ_F_NEED_CLEANUP; |
15b71abe JA |
3018 | if (ret < 0) |
3019 | req_set_fail_links(req); | |
3020 | io_cqring_add_event(req, ret); | |
014db007 | 3021 | io_put_req(req); |
15b71abe JA |
3022 | return 0; |
3023 | } | |
3024 | ||
014db007 | 3025 | static int io_openat(struct io_kiocb *req, bool force_nonblock) |
cebdb986 JA |
3026 | { |
3027 | req->open.how = build_open_how(req->open.how.flags, req->open.how.mode); | |
014db007 | 3028 | return io_openat2(req, force_nonblock); |
cebdb986 JA |
3029 | } |
3030 | ||
067524e9 JA |
3031 | static int io_remove_buffers_prep(struct io_kiocb *req, |
3032 | const struct io_uring_sqe *sqe) | |
3033 | { | |
3034 | struct io_provide_buf *p = &req->pbuf; | |
3035 | u64 tmp; | |
3036 | ||
3037 | if (sqe->ioprio || sqe->rw_flags || sqe->addr || sqe->len || sqe->off) | |
3038 | return -EINVAL; | |
3039 | ||
3040 | tmp = READ_ONCE(sqe->fd); | |
3041 | if (!tmp || tmp > USHRT_MAX) | |
3042 | return -EINVAL; | |
3043 | ||
3044 | memset(p, 0, sizeof(*p)); | |
3045 | p->nbufs = tmp; | |
3046 | p->bgid = READ_ONCE(sqe->buf_group); | |
3047 | return 0; | |
3048 | } | |
3049 | ||
3050 | static int __io_remove_buffers(struct io_ring_ctx *ctx, struct io_buffer *buf, | |
3051 | int bgid, unsigned nbufs) | |
3052 | { | |
3053 | unsigned i = 0; | |
3054 | ||
3055 | /* shouldn't happen */ | |
3056 | if (!nbufs) | |
3057 | return 0; | |
3058 | ||
3059 | /* the head kbuf is the list itself */ | |
3060 | while (!list_empty(&buf->list)) { | |
3061 | struct io_buffer *nxt; | |
3062 | ||
3063 | nxt = list_first_entry(&buf->list, struct io_buffer, list); | |
3064 | list_del(&nxt->list); | |
3065 | kfree(nxt); | |
3066 | if (++i == nbufs) | |
3067 | return i; | |
3068 | } | |
3069 | i++; | |
3070 | kfree(buf); | |
3071 | idr_remove(&ctx->io_buffer_idr, bgid); | |
3072 | ||
3073 | return i; | |
3074 | } | |
3075 | ||
3076 | static int io_remove_buffers(struct io_kiocb *req, bool force_nonblock) | |
3077 | { | |
3078 | struct io_provide_buf *p = &req->pbuf; | |
3079 | struct io_ring_ctx *ctx = req->ctx; | |
3080 | struct io_buffer *head; | |
3081 | int ret = 0; | |
3082 | ||
3083 | io_ring_submit_lock(ctx, !force_nonblock); | |
3084 | ||
3085 | lockdep_assert_held(&ctx->uring_lock); | |
3086 | ||
3087 | ret = -ENOENT; | |
3088 | head = idr_find(&ctx->io_buffer_idr, p->bgid); | |
3089 | if (head) | |
3090 | ret = __io_remove_buffers(ctx, head, p->bgid, p->nbufs); | |
3091 | ||
3092 | io_ring_submit_lock(ctx, !force_nonblock); | |
3093 | if (ret < 0) | |
3094 | req_set_fail_links(req); | |
3095 | io_cqring_add_event(req, ret); | |
3096 | io_put_req(req); | |
3097 | return 0; | |
3098 | } | |
3099 | ||
ddf0322d JA |
3100 | static int io_provide_buffers_prep(struct io_kiocb *req, |
3101 | const struct io_uring_sqe *sqe) | |
3102 | { | |
3103 | struct io_provide_buf *p = &req->pbuf; | |
3104 | u64 tmp; | |
3105 | ||
3106 | if (sqe->ioprio || sqe->rw_flags) | |
3107 | return -EINVAL; | |
3108 | ||
3109 | tmp = READ_ONCE(sqe->fd); | |
3110 | if (!tmp || tmp > USHRT_MAX) | |
3111 | return -E2BIG; | |
3112 | p->nbufs = tmp; | |
3113 | p->addr = READ_ONCE(sqe->addr); | |
3114 | p->len = READ_ONCE(sqe->len); | |
3115 | ||
3116 | if (!access_ok(u64_to_user_ptr(p->addr), p->len)) | |
3117 | return -EFAULT; | |
3118 | ||
3119 | p->bgid = READ_ONCE(sqe->buf_group); | |
3120 | tmp = READ_ONCE(sqe->off); | |
3121 | if (tmp > USHRT_MAX) | |
3122 | return -E2BIG; | |
3123 | p->bid = tmp; | |
3124 | return 0; | |
3125 | } | |
3126 | ||
3127 | static int io_add_buffers(struct io_provide_buf *pbuf, struct io_buffer **head) | |
3128 | { | |
3129 | struct io_buffer *buf; | |
3130 | u64 addr = pbuf->addr; | |
3131 | int i, bid = pbuf->bid; | |
3132 | ||
3133 | for (i = 0; i < pbuf->nbufs; i++) { | |
3134 | buf = kmalloc(sizeof(*buf), GFP_KERNEL); | |
3135 | if (!buf) | |
3136 | break; | |
3137 | ||
3138 | buf->addr = addr; | |
3139 | buf->len = pbuf->len; | |
3140 | buf->bid = bid; | |
3141 | addr += pbuf->len; | |
3142 | bid++; | |
3143 | if (!*head) { | |
3144 | INIT_LIST_HEAD(&buf->list); | |
3145 | *head = buf; | |
3146 | } else { | |
3147 | list_add_tail(&buf->list, &(*head)->list); | |
3148 | } | |
3149 | } | |
3150 | ||
3151 | return i ? i : -ENOMEM; | |
3152 | } | |
3153 | ||
ddf0322d JA |
3154 | static int io_provide_buffers(struct io_kiocb *req, bool force_nonblock) |
3155 | { | |
3156 | struct io_provide_buf *p = &req->pbuf; | |
3157 | struct io_ring_ctx *ctx = req->ctx; | |
3158 | struct io_buffer *head, *list; | |
3159 | int ret = 0; | |
3160 | ||
3161 | io_ring_submit_lock(ctx, !force_nonblock); | |
3162 | ||
3163 | lockdep_assert_held(&ctx->uring_lock); | |
3164 | ||
3165 | list = head = idr_find(&ctx->io_buffer_idr, p->bgid); | |
3166 | ||
3167 | ret = io_add_buffers(p, &head); | |
3168 | if (ret < 0) | |
3169 | goto out; | |
3170 | ||
3171 | if (!list) { | |
3172 | ret = idr_alloc(&ctx->io_buffer_idr, head, p->bgid, p->bgid + 1, | |
3173 | GFP_KERNEL); | |
3174 | if (ret < 0) { | |
067524e9 | 3175 | __io_remove_buffers(ctx, head, p->bgid, -1U); |
ddf0322d JA |
3176 | goto out; |
3177 | } | |
3178 | } | |
3179 | out: | |
3180 | io_ring_submit_unlock(ctx, !force_nonblock); | |
3181 | if (ret < 0) | |
3182 | req_set_fail_links(req); | |
3183 | io_cqring_add_event(req, ret); | |
3184 | io_put_req(req); | |
3185 | return 0; | |
cebdb986 JA |
3186 | } |
3187 | ||
3e4827b0 JA |
3188 | static int io_epoll_ctl_prep(struct io_kiocb *req, |
3189 | const struct io_uring_sqe *sqe) | |
3190 | { | |
3191 | #if defined(CONFIG_EPOLL) | |
3192 | if (sqe->ioprio || sqe->buf_index) | |
3193 | return -EINVAL; | |
3194 | ||
3195 | req->epoll.epfd = READ_ONCE(sqe->fd); | |
3196 | req->epoll.op = READ_ONCE(sqe->len); | |
3197 | req->epoll.fd = READ_ONCE(sqe->off); | |
3198 | ||
3199 | if (ep_op_has_event(req->epoll.op)) { | |
3200 | struct epoll_event __user *ev; | |
3201 | ||
3202 | ev = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
3203 | if (copy_from_user(&req->epoll.event, ev, sizeof(*ev))) | |
3204 | return -EFAULT; | |
3205 | } | |
3206 | ||
3207 | return 0; | |
3208 | #else | |
3209 | return -EOPNOTSUPP; | |
3210 | #endif | |
3211 | } | |
3212 | ||
014db007 | 3213 | static int io_epoll_ctl(struct io_kiocb *req, bool force_nonblock) |
3e4827b0 JA |
3214 | { |
3215 | #if defined(CONFIG_EPOLL) | |
3216 | struct io_epoll *ie = &req->epoll; | |
3217 | int ret; | |
3218 | ||
3219 | ret = do_epoll_ctl(ie->epfd, ie->op, ie->fd, &ie->event, force_nonblock); | |
3220 | if (force_nonblock && ret == -EAGAIN) | |
3221 | return -EAGAIN; | |
3222 | ||
3223 | if (ret < 0) | |
3224 | req_set_fail_links(req); | |
3225 | io_cqring_add_event(req, ret); | |
014db007 | 3226 | io_put_req(req); |
3e4827b0 JA |
3227 | return 0; |
3228 | #else | |
3229 | return -EOPNOTSUPP; | |
3230 | #endif | |
3231 | } | |
3232 | ||
c1ca757b JA |
3233 | static int io_madvise_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
3234 | { | |
3235 | #if defined(CONFIG_ADVISE_SYSCALLS) && defined(CONFIG_MMU) | |
3236 | if (sqe->ioprio || sqe->buf_index || sqe->off) | |
3237 | return -EINVAL; | |
3238 | ||
3239 | req->madvise.addr = READ_ONCE(sqe->addr); | |
3240 | req->madvise.len = READ_ONCE(sqe->len); | |
3241 | req->madvise.advice = READ_ONCE(sqe->fadvise_advice); | |
3242 | return 0; | |
3243 | #else | |
3244 | return -EOPNOTSUPP; | |
3245 | #endif | |
3246 | } | |
3247 | ||
014db007 | 3248 | static int io_madvise(struct io_kiocb *req, bool force_nonblock) |
c1ca757b JA |
3249 | { |
3250 | #if defined(CONFIG_ADVISE_SYSCALLS) && defined(CONFIG_MMU) | |
3251 | struct io_madvise *ma = &req->madvise; | |
3252 | int ret; | |
3253 | ||
3254 | if (force_nonblock) | |
3255 | return -EAGAIN; | |
3256 | ||
3257 | ret = do_madvise(ma->addr, ma->len, ma->advice); | |
3258 | if (ret < 0) | |
3259 | req_set_fail_links(req); | |
3260 | io_cqring_add_event(req, ret); | |
014db007 | 3261 | io_put_req(req); |
c1ca757b JA |
3262 | return 0; |
3263 | #else | |
3264 | return -EOPNOTSUPP; | |
3265 | #endif | |
3266 | } | |
3267 | ||
4840e418 JA |
3268 | static int io_fadvise_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
3269 | { | |
3270 | if (sqe->ioprio || sqe->buf_index || sqe->addr) | |
3271 | return -EINVAL; | |
3272 | ||
3273 | req->fadvise.offset = READ_ONCE(sqe->off); | |
3274 | req->fadvise.len = READ_ONCE(sqe->len); | |
3275 | req->fadvise.advice = READ_ONCE(sqe->fadvise_advice); | |
3276 | return 0; | |
3277 | } | |
3278 | ||
014db007 | 3279 | static int io_fadvise(struct io_kiocb *req, bool force_nonblock) |
4840e418 JA |
3280 | { |
3281 | struct io_fadvise *fa = &req->fadvise; | |
3282 | int ret; | |
3283 | ||
3e69426d JA |
3284 | if (force_nonblock) { |
3285 | switch (fa->advice) { | |
3286 | case POSIX_FADV_NORMAL: | |
3287 | case POSIX_FADV_RANDOM: | |
3288 | case POSIX_FADV_SEQUENTIAL: | |
3289 | break; | |
3290 | default: | |
3291 | return -EAGAIN; | |
3292 | } | |
3293 | } | |
4840e418 JA |
3294 | |
3295 | ret = vfs_fadvise(req->file, fa->offset, fa->len, fa->advice); | |
3296 | if (ret < 0) | |
3297 | req_set_fail_links(req); | |
3298 | io_cqring_add_event(req, ret); | |
014db007 | 3299 | io_put_req(req); |
4840e418 JA |
3300 | return 0; |
3301 | } | |
3302 | ||
eddc7ef5 JA |
3303 | static int io_statx_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
3304 | { | |
f8748881 | 3305 | const char __user *fname; |
eddc7ef5 JA |
3306 | unsigned lookup_flags; |
3307 | int ret; | |
3308 | ||
3309 | if (sqe->ioprio || sqe->buf_index) | |
3310 | return -EINVAL; | |
9c280f90 | 3311 | if (req->flags & REQ_F_FIXED_FILE) |
cf3040ca | 3312 | return -EBADF; |
0bdbdd08 PB |
3313 | if (req->flags & REQ_F_NEED_CLEANUP) |
3314 | return 0; | |
eddc7ef5 JA |
3315 | |
3316 | req->open.dfd = READ_ONCE(sqe->fd); | |
3317 | req->open.mask = READ_ONCE(sqe->len); | |
f8748881 | 3318 | fname = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
eddc7ef5 | 3319 | req->open.buffer = u64_to_user_ptr(READ_ONCE(sqe->addr2)); |
c12cedf2 | 3320 | req->open.how.flags = READ_ONCE(sqe->statx_flags); |
eddc7ef5 | 3321 | |
c12cedf2 | 3322 | if (vfs_stat_set_lookup_flags(&lookup_flags, req->open.how.flags)) |
eddc7ef5 JA |
3323 | return -EINVAL; |
3324 | ||
f8748881 | 3325 | req->open.filename = getname_flags(fname, lookup_flags, NULL); |
eddc7ef5 JA |
3326 | if (IS_ERR(req->open.filename)) { |
3327 | ret = PTR_ERR(req->open.filename); | |
3328 | req->open.filename = NULL; | |
3329 | return ret; | |
3330 | } | |
3331 | ||
8fef80bf | 3332 | req->flags |= REQ_F_NEED_CLEANUP; |
eddc7ef5 JA |
3333 | return 0; |
3334 | } | |
3335 | ||
014db007 | 3336 | static int io_statx(struct io_kiocb *req, bool force_nonblock) |
eddc7ef5 JA |
3337 | { |
3338 | struct io_open *ctx = &req->open; | |
3339 | unsigned lookup_flags; | |
3340 | struct path path; | |
3341 | struct kstat stat; | |
3342 | int ret; | |
3343 | ||
5b0bbee4 JA |
3344 | if (force_nonblock) { |
3345 | /* only need file table for an actual valid fd */ | |
3346 | if (ctx->dfd == -1 || ctx->dfd == AT_FDCWD) | |
3347 | req->flags |= REQ_F_NO_FILE_TABLE; | |
eddc7ef5 | 3348 | return -EAGAIN; |
5b0bbee4 | 3349 | } |
eddc7ef5 | 3350 | |
c12cedf2 | 3351 | if (vfs_stat_set_lookup_flags(&lookup_flags, ctx->how.flags)) |
eddc7ef5 JA |
3352 | return -EINVAL; |
3353 | ||
3354 | retry: | |
3355 | /* filename_lookup() drops it, keep a reference */ | |
3356 | ctx->filename->refcnt++; | |
3357 | ||
3358 | ret = filename_lookup(ctx->dfd, ctx->filename, lookup_flags, &path, | |
3359 | NULL); | |
3360 | if (ret) | |
3361 | goto err; | |
3362 | ||
c12cedf2 | 3363 | ret = vfs_getattr(&path, &stat, ctx->mask, ctx->how.flags); |
eddc7ef5 JA |
3364 | path_put(&path); |
3365 | if (retry_estale(ret, lookup_flags)) { | |
3366 | lookup_flags |= LOOKUP_REVAL; | |
3367 | goto retry; | |
3368 | } | |
3369 | if (!ret) | |
3370 | ret = cp_statx(&stat, ctx->buffer); | |
3371 | err: | |
3372 | putname(ctx->filename); | |
8fef80bf | 3373 | req->flags &= ~REQ_F_NEED_CLEANUP; |
eddc7ef5 JA |
3374 | if (ret < 0) |
3375 | req_set_fail_links(req); | |
3376 | io_cqring_add_event(req, ret); | |
014db007 | 3377 | io_put_req(req); |
eddc7ef5 JA |
3378 | return 0; |
3379 | } | |
3380 | ||
b5dba59e JA |
3381 | static int io_close_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
3382 | { | |
3383 | /* | |
3384 | * If we queue this for async, it must not be cancellable. That would | |
3385 | * leave the 'file' in an undeterminate state. | |
3386 | */ | |
3387 | req->work.flags |= IO_WQ_WORK_NO_CANCEL; | |
3388 | ||
3389 | if (sqe->ioprio || sqe->off || sqe->addr || sqe->len || | |
3390 | sqe->rw_flags || sqe->buf_index) | |
3391 | return -EINVAL; | |
9c280f90 | 3392 | if (req->flags & REQ_F_FIXED_FILE) |
cf3040ca | 3393 | return -EBADF; |
b5dba59e JA |
3394 | |
3395 | req->close.fd = READ_ONCE(sqe->fd); | |
b5dba59e JA |
3396 | return 0; |
3397 | } | |
3398 | ||
a93b3331 | 3399 | /* only called when __close_fd_get_file() is done */ |
014db007 | 3400 | static void __io_close_finish(struct io_kiocb *req) |
a93b3331 PB |
3401 | { |
3402 | int ret; | |
3403 | ||
3404 | ret = filp_close(req->close.put_file, req->work.files); | |
3405 | if (ret < 0) | |
3406 | req_set_fail_links(req); | |
3407 | io_cqring_add_event(req, ret); | |
3408 | fput(req->close.put_file); | |
014db007 | 3409 | io_put_req(req); |
a93b3331 PB |
3410 | } |
3411 | ||
b5dba59e JA |
3412 | static void io_close_finish(struct io_wq_work **workptr) |
3413 | { | |
3414 | struct io_kiocb *req = container_of(*workptr, struct io_kiocb, work); | |
b5dba59e | 3415 | |
7fbeb95d | 3416 | /* not cancellable, don't do io_req_cancelled() */ |
014db007 | 3417 | __io_close_finish(req); |
e9fd9396 | 3418 | io_steal_work(req, workptr); |
b5dba59e JA |
3419 | } |
3420 | ||
014db007 | 3421 | static int io_close(struct io_kiocb *req, bool force_nonblock) |
b5dba59e JA |
3422 | { |
3423 | int ret; | |
3424 | ||
3425 | req->close.put_file = NULL; | |
3426 | ret = __close_fd_get_file(req->close.fd, &req->close.put_file); | |
904fbcb1 JA |
3427 | if (ret < 0) { |
3428 | if (ret == -ENOENT) | |
3429 | ret = -EBADF; | |
b5dba59e | 3430 | return ret; |
904fbcb1 | 3431 | } |
b5dba59e JA |
3432 | |
3433 | /* if the file has a flush method, be safe and punt to async */ | |
a2100672 | 3434 | if (req->close.put_file->f_op->flush && force_nonblock) { |
594506fe PB |
3435 | /* submission ref will be dropped, take it for async */ |
3436 | refcount_inc(&req->refs); | |
3437 | ||
a2100672 PB |
3438 | req->work.func = io_close_finish; |
3439 | /* | |
3440 | * Do manual async queue here to avoid grabbing files - we don't | |
3441 | * need the files, and it'll cause io_close_finish() to close | |
3442 | * the file again and cause a double CQE entry for this request | |
3443 | */ | |
3444 | io_queue_async_work(req); | |
3445 | return 0; | |
3446 | } | |
b5dba59e JA |
3447 | |
3448 | /* | |
3449 | * No ->flush(), safely close from here and just punt the | |
3450 | * fput() to async context. | |
3451 | */ | |
014db007 | 3452 | __io_close_finish(req); |
1a417f4e | 3453 | return 0; |
b5dba59e JA |
3454 | } |
3455 | ||
3529d8c2 | 3456 | static int io_prep_sfr(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
5d17b4a4 JA |
3457 | { |
3458 | struct io_ring_ctx *ctx = req->ctx; | |
5d17b4a4 JA |
3459 | |
3460 | if (!req->file) | |
3461 | return -EBADF; | |
5d17b4a4 JA |
3462 | |
3463 | if (unlikely(ctx->flags & IORING_SETUP_IOPOLL)) | |
3464 | return -EINVAL; | |
3465 | if (unlikely(sqe->addr || sqe->ioprio || sqe->buf_index)) | |
3466 | return -EINVAL; | |
3467 | ||
8ed8d3c3 JA |
3468 | req->sync.off = READ_ONCE(sqe->off); |
3469 | req->sync.len = READ_ONCE(sqe->len); | |
3470 | req->sync.flags = READ_ONCE(sqe->sync_range_flags); | |
8ed8d3c3 JA |
3471 | return 0; |
3472 | } | |
3473 | ||
014db007 | 3474 | static void __io_sync_file_range(struct io_kiocb *req) |
8ed8d3c3 | 3475 | { |
8ed8d3c3 JA |
3476 | int ret; |
3477 | ||
9adbd45d | 3478 | ret = sync_file_range(req->file, req->sync.off, req->sync.len, |
8ed8d3c3 JA |
3479 | req->sync.flags); |
3480 | if (ret < 0) | |
3481 | req_set_fail_links(req); | |
3482 | io_cqring_add_event(req, ret); | |
014db007 | 3483 | io_put_req(req); |
5ea62161 PB |
3484 | } |
3485 | ||
3486 | ||
3487 | static void io_sync_file_range_finish(struct io_wq_work **workptr) | |
3488 | { | |
3489 | struct io_kiocb *req = container_of(*workptr, struct io_kiocb, work); | |
5ea62161 PB |
3490 | |
3491 | if (io_req_cancelled(req)) | |
3492 | return; | |
014db007 | 3493 | __io_sync_file_range(req); |
7759a0bf | 3494 | io_steal_work(req, workptr); |
5d17b4a4 JA |
3495 | } |
3496 | ||
014db007 | 3497 | static int io_sync_file_range(struct io_kiocb *req, bool force_nonblock) |
5d17b4a4 | 3498 | { |
5d17b4a4 | 3499 | /* sync_file_range always requires a blocking context */ |
8ed8d3c3 | 3500 | if (force_nonblock) { |
8ed8d3c3 | 3501 | req->work.func = io_sync_file_range_finish; |
5d17b4a4 | 3502 | return -EAGAIN; |
8ed8d3c3 | 3503 | } |
5d17b4a4 | 3504 | |
014db007 | 3505 | __io_sync_file_range(req); |
5d17b4a4 JA |
3506 | return 0; |
3507 | } | |
3508 | ||
469956e8 | 3509 | #if defined(CONFIG_NET) |
02d27d89 PB |
3510 | static int io_setup_async_msg(struct io_kiocb *req, |
3511 | struct io_async_msghdr *kmsg) | |
3512 | { | |
3513 | if (req->io) | |
3514 | return -EAGAIN; | |
3515 | if (io_alloc_async_ctx(req)) { | |
3516 | if (kmsg->iov != kmsg->fast_iov) | |
3517 | kfree(kmsg->iov); | |
3518 | return -ENOMEM; | |
3519 | } | |
3520 | req->flags |= REQ_F_NEED_CLEANUP; | |
3521 | memcpy(&req->io->msg, kmsg, sizeof(*kmsg)); | |
3522 | return -EAGAIN; | |
3523 | } | |
3524 | ||
3529d8c2 | 3525 | static int io_sendmsg_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
03b1230c | 3526 | { |
e47293fd | 3527 | struct io_sr_msg *sr = &req->sr_msg; |
3529d8c2 | 3528 | struct io_async_ctx *io = req->io; |
99bc4c38 | 3529 | int ret; |
03b1230c | 3530 | |
e47293fd JA |
3531 | sr->msg_flags = READ_ONCE(sqe->msg_flags); |
3532 | sr->msg = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
fddaface | 3533 | sr->len = READ_ONCE(sqe->len); |
3529d8c2 | 3534 | |
d8768362 JA |
3535 | #ifdef CONFIG_COMPAT |
3536 | if (req->ctx->compat) | |
3537 | sr->msg_flags |= MSG_CMSG_COMPAT; | |
3538 | #endif | |
3539 | ||
fddaface | 3540 | if (!io || req->opcode == IORING_OP_SEND) |
3529d8c2 | 3541 | return 0; |
5f798bea PB |
3542 | /* iovec is already imported */ |
3543 | if (req->flags & REQ_F_NEED_CLEANUP) | |
3544 | return 0; | |
3529d8c2 | 3545 | |
d9688565 | 3546 | io->msg.iov = io->msg.fast_iov; |
99bc4c38 | 3547 | ret = sendmsg_copy_msghdr(&io->msg.msg, sr->msg, sr->msg_flags, |
e47293fd | 3548 | &io->msg.iov); |
99bc4c38 PB |
3549 | if (!ret) |
3550 | req->flags |= REQ_F_NEED_CLEANUP; | |
3551 | return ret; | |
03b1230c JA |
3552 | } |
3553 | ||
014db007 | 3554 | static int io_sendmsg(struct io_kiocb *req, bool force_nonblock) |
aa1fa28f | 3555 | { |
0b416c3e | 3556 | struct io_async_msghdr *kmsg = NULL; |
0fa03c62 JA |
3557 | struct socket *sock; |
3558 | int ret; | |
3559 | ||
3560 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) | |
3561 | return -EINVAL; | |
3562 | ||
3563 | sock = sock_from_file(req->file, &ret); | |
3564 | if (sock) { | |
b7bb4f7d | 3565 | struct io_async_ctx io; |
0fa03c62 JA |
3566 | unsigned flags; |
3567 | ||
03b1230c | 3568 | if (req->io) { |
0b416c3e | 3569 | kmsg = &req->io->msg; |
b537916c | 3570 | kmsg->msg.msg_name = &req->io->msg.addr; |
0b416c3e JA |
3571 | /* if iov is set, it's allocated already */ |
3572 | if (!kmsg->iov) | |
3573 | kmsg->iov = kmsg->fast_iov; | |
3574 | kmsg->msg.msg_iter.iov = kmsg->iov; | |
03b1230c | 3575 | } else { |
3529d8c2 JA |
3576 | struct io_sr_msg *sr = &req->sr_msg; |
3577 | ||
0b416c3e | 3578 | kmsg = &io.msg; |
b537916c | 3579 | kmsg->msg.msg_name = &io.msg.addr; |
3529d8c2 JA |
3580 | |
3581 | io.msg.iov = io.msg.fast_iov; | |
3582 | ret = sendmsg_copy_msghdr(&io.msg.msg, sr->msg, | |
3583 | sr->msg_flags, &io.msg.iov); | |
03b1230c | 3584 | if (ret) |
3529d8c2 | 3585 | return ret; |
03b1230c | 3586 | } |
0fa03c62 | 3587 | |
e47293fd JA |
3588 | flags = req->sr_msg.msg_flags; |
3589 | if (flags & MSG_DONTWAIT) | |
3590 | req->flags |= REQ_F_NOWAIT; | |
3591 | else if (force_nonblock) | |
3592 | flags |= MSG_DONTWAIT; | |
3593 | ||
0b416c3e | 3594 | ret = __sys_sendmsg_sock(sock, &kmsg->msg, flags); |
02d27d89 PB |
3595 | if (force_nonblock && ret == -EAGAIN) |
3596 | return io_setup_async_msg(req, kmsg); | |
441cdbd5 JA |
3597 | if (ret == -ERESTARTSYS) |
3598 | ret = -EINTR; | |
0fa03c62 JA |
3599 | } |
3600 | ||
1e95081c | 3601 | if (kmsg && kmsg->iov != kmsg->fast_iov) |
0b416c3e | 3602 | kfree(kmsg->iov); |
99bc4c38 | 3603 | req->flags &= ~REQ_F_NEED_CLEANUP; |
78e19bbe | 3604 | io_cqring_add_event(req, ret); |
4e88d6e7 JA |
3605 | if (ret < 0) |
3606 | req_set_fail_links(req); | |
014db007 | 3607 | io_put_req(req); |
5d17b4a4 | 3608 | return 0; |
03b1230c | 3609 | } |
aa1fa28f | 3610 | |
014db007 | 3611 | static int io_send(struct io_kiocb *req, bool force_nonblock) |
fddaface | 3612 | { |
fddaface JA |
3613 | struct socket *sock; |
3614 | int ret; | |
3615 | ||
3616 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) | |
3617 | return -EINVAL; | |
3618 | ||
3619 | sock = sock_from_file(req->file, &ret); | |
3620 | if (sock) { | |
3621 | struct io_sr_msg *sr = &req->sr_msg; | |
3622 | struct msghdr msg; | |
3623 | struct iovec iov; | |
3624 | unsigned flags; | |
3625 | ||
3626 | ret = import_single_range(WRITE, sr->buf, sr->len, &iov, | |
3627 | &msg.msg_iter); | |
3628 | if (ret) | |
3629 | return ret; | |
3630 | ||
3631 | msg.msg_name = NULL; | |
3632 | msg.msg_control = NULL; | |
3633 | msg.msg_controllen = 0; | |
3634 | msg.msg_namelen = 0; | |
3635 | ||
3636 | flags = req->sr_msg.msg_flags; | |
3637 | if (flags & MSG_DONTWAIT) | |
3638 | req->flags |= REQ_F_NOWAIT; | |
3639 | else if (force_nonblock) | |
3640 | flags |= MSG_DONTWAIT; | |
3641 | ||
0b7b21e4 JA |
3642 | msg.msg_flags = flags; |
3643 | ret = sock_sendmsg(sock, &msg); | |
fddaface JA |
3644 | if (force_nonblock && ret == -EAGAIN) |
3645 | return -EAGAIN; | |
3646 | if (ret == -ERESTARTSYS) | |
3647 | ret = -EINTR; | |
3648 | } | |
3649 | ||
3650 | io_cqring_add_event(req, ret); | |
3651 | if (ret < 0) | |
3652 | req_set_fail_links(req); | |
014db007 | 3653 | io_put_req(req); |
fddaface | 3654 | return 0; |
fddaface JA |
3655 | } |
3656 | ||
52de1fe1 JA |
3657 | static int __io_recvmsg_copy_hdr(struct io_kiocb *req, struct io_async_ctx *io) |
3658 | { | |
3659 | struct io_sr_msg *sr = &req->sr_msg; | |
3660 | struct iovec __user *uiov; | |
3661 | size_t iov_len; | |
3662 | int ret; | |
3663 | ||
3664 | ret = __copy_msghdr_from_user(&io->msg.msg, sr->msg, &io->msg.uaddr, | |
3665 | &uiov, &iov_len); | |
3666 | if (ret) | |
3667 | return ret; | |
3668 | ||
3669 | if (req->flags & REQ_F_BUFFER_SELECT) { | |
3670 | if (iov_len > 1) | |
3671 | return -EINVAL; | |
3672 | if (copy_from_user(io->msg.iov, uiov, sizeof(*uiov))) | |
3673 | return -EFAULT; | |
3674 | sr->len = io->msg.iov[0].iov_len; | |
3675 | iov_iter_init(&io->msg.msg.msg_iter, READ, io->msg.iov, 1, | |
3676 | sr->len); | |
3677 | io->msg.iov = NULL; | |
3678 | } else { | |
3679 | ret = import_iovec(READ, uiov, iov_len, UIO_FASTIOV, | |
3680 | &io->msg.iov, &io->msg.msg.msg_iter); | |
3681 | if (ret > 0) | |
3682 | ret = 0; | |
3683 | } | |
3684 | ||
3685 | return ret; | |
3686 | } | |
3687 | ||
3688 | #ifdef CONFIG_COMPAT | |
3689 | static int __io_compat_recvmsg_copy_hdr(struct io_kiocb *req, | |
3690 | struct io_async_ctx *io) | |
3691 | { | |
3692 | struct compat_msghdr __user *msg_compat; | |
3693 | struct io_sr_msg *sr = &req->sr_msg; | |
3694 | struct compat_iovec __user *uiov; | |
3695 | compat_uptr_t ptr; | |
3696 | compat_size_t len; | |
3697 | int ret; | |
3698 | ||
3699 | msg_compat = (struct compat_msghdr __user *) sr->msg; | |
3700 | ret = __get_compat_msghdr(&io->msg.msg, msg_compat, &io->msg.uaddr, | |
3701 | &ptr, &len); | |
3702 | if (ret) | |
3703 | return ret; | |
3704 | ||
3705 | uiov = compat_ptr(ptr); | |
3706 | if (req->flags & REQ_F_BUFFER_SELECT) { | |
3707 | compat_ssize_t clen; | |
3708 | ||
3709 | if (len > 1) | |
3710 | return -EINVAL; | |
3711 | if (!access_ok(uiov, sizeof(*uiov))) | |
3712 | return -EFAULT; | |
3713 | if (__get_user(clen, &uiov->iov_len)) | |
3714 | return -EFAULT; | |
3715 | if (clen < 0) | |
3716 | return -EINVAL; | |
3717 | sr->len = io->msg.iov[0].iov_len; | |
3718 | io->msg.iov = NULL; | |
3719 | } else { | |
3720 | ret = compat_import_iovec(READ, uiov, len, UIO_FASTIOV, | |
3721 | &io->msg.iov, | |
3722 | &io->msg.msg.msg_iter); | |
3723 | if (ret < 0) | |
3724 | return ret; | |
3725 | } | |
3726 | ||
3727 | return 0; | |
3728 | } | |
3729 | #endif | |
3730 | ||
3731 | static int io_recvmsg_copy_hdr(struct io_kiocb *req, struct io_async_ctx *io) | |
3732 | { | |
3733 | io->msg.iov = io->msg.fast_iov; | |
3734 | ||
3735 | #ifdef CONFIG_COMPAT | |
3736 | if (req->ctx->compat) | |
3737 | return __io_compat_recvmsg_copy_hdr(req, io); | |
fddaface | 3738 | #endif |
52de1fe1 JA |
3739 | |
3740 | return __io_recvmsg_copy_hdr(req, io); | |
3741 | } | |
3742 | ||
bcda7baa JA |
3743 | static struct io_buffer *io_recv_buffer_select(struct io_kiocb *req, |
3744 | int *cflags, bool needs_lock) | |
3745 | { | |
3746 | struct io_sr_msg *sr = &req->sr_msg; | |
3747 | struct io_buffer *kbuf; | |
3748 | ||
3749 | if (!(req->flags & REQ_F_BUFFER_SELECT)) | |
3750 | return NULL; | |
3751 | ||
3752 | kbuf = io_buffer_select(req, &sr->len, sr->bgid, sr->kbuf, needs_lock); | |
3753 | if (IS_ERR(kbuf)) | |
3754 | return kbuf; | |
3755 | ||
3756 | sr->kbuf = kbuf; | |
3757 | req->flags |= REQ_F_BUFFER_SELECTED; | |
3758 | ||
3759 | *cflags = kbuf->bid << IORING_CQE_BUFFER_SHIFT; | |
3760 | *cflags |= IORING_CQE_F_BUFFER; | |
3761 | return kbuf; | |
fddaface JA |
3762 | } |
3763 | ||
3529d8c2 JA |
3764 | static int io_recvmsg_prep(struct io_kiocb *req, |
3765 | const struct io_uring_sqe *sqe) | |
aa1fa28f | 3766 | { |
e47293fd | 3767 | struct io_sr_msg *sr = &req->sr_msg; |
3529d8c2 | 3768 | struct io_async_ctx *io = req->io; |
99bc4c38 | 3769 | int ret; |
3529d8c2 JA |
3770 | |
3771 | sr->msg_flags = READ_ONCE(sqe->msg_flags); | |
3772 | sr->msg = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
0b7b21e4 | 3773 | sr->len = READ_ONCE(sqe->len); |
bcda7baa | 3774 | sr->bgid = READ_ONCE(sqe->buf_group); |
06b76d44 | 3775 | |
d8768362 JA |
3776 | #ifdef CONFIG_COMPAT |
3777 | if (req->ctx->compat) | |
3778 | sr->msg_flags |= MSG_CMSG_COMPAT; | |
3779 | #endif | |
3780 | ||
fddaface | 3781 | if (!io || req->opcode == IORING_OP_RECV) |
06b76d44 | 3782 | return 0; |
5f798bea PB |
3783 | /* iovec is already imported */ |
3784 | if (req->flags & REQ_F_NEED_CLEANUP) | |
3785 | return 0; | |
03b1230c | 3786 | |
52de1fe1 | 3787 | ret = io_recvmsg_copy_hdr(req, io); |
99bc4c38 PB |
3788 | if (!ret) |
3789 | req->flags |= REQ_F_NEED_CLEANUP; | |
3790 | return ret; | |
aa1fa28f JA |
3791 | } |
3792 | ||
014db007 | 3793 | static int io_recvmsg(struct io_kiocb *req, bool force_nonblock) |
aa1fa28f | 3794 | { |
0b416c3e | 3795 | struct io_async_msghdr *kmsg = NULL; |
03b1230c | 3796 | struct socket *sock; |
52de1fe1 | 3797 | int ret, cflags = 0; |
03b1230c JA |
3798 | |
3799 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) | |
3800 | return -EINVAL; | |
3801 | ||
3802 | sock = sock_from_file(req->file, &ret); | |
3803 | if (sock) { | |
52de1fe1 | 3804 | struct io_buffer *kbuf; |
b7bb4f7d | 3805 | struct io_async_ctx io; |
03b1230c JA |
3806 | unsigned flags; |
3807 | ||
03b1230c | 3808 | if (req->io) { |
0b416c3e | 3809 | kmsg = &req->io->msg; |
b537916c | 3810 | kmsg->msg.msg_name = &req->io->msg.addr; |
0b416c3e JA |
3811 | /* if iov is set, it's allocated already */ |
3812 | if (!kmsg->iov) | |
3813 | kmsg->iov = kmsg->fast_iov; | |
3814 | kmsg->msg.msg_iter.iov = kmsg->iov; | |
03b1230c | 3815 | } else { |
0b416c3e | 3816 | kmsg = &io.msg; |
b537916c | 3817 | kmsg->msg.msg_name = &io.msg.addr; |
3529d8c2 | 3818 | |
52de1fe1 | 3819 | ret = io_recvmsg_copy_hdr(req, &io); |
03b1230c | 3820 | if (ret) |
3529d8c2 | 3821 | return ret; |
03b1230c JA |
3822 | } |
3823 | ||
52de1fe1 JA |
3824 | kbuf = io_recv_buffer_select(req, &cflags, !force_nonblock); |
3825 | if (IS_ERR(kbuf)) { | |
3826 | return PTR_ERR(kbuf); | |
3827 | } else if (kbuf) { | |
3828 | kmsg->fast_iov[0].iov_base = u64_to_user_ptr(kbuf->addr); | |
3829 | iov_iter_init(&kmsg->msg.msg_iter, READ, kmsg->iov, | |
3830 | 1, req->sr_msg.len); | |
3831 | } | |
3832 | ||
e47293fd JA |
3833 | flags = req->sr_msg.msg_flags; |
3834 | if (flags & MSG_DONTWAIT) | |
3835 | req->flags |= REQ_F_NOWAIT; | |
3836 | else if (force_nonblock) | |
3837 | flags |= MSG_DONTWAIT; | |
3838 | ||
3839 | ret = __sys_recvmsg_sock(sock, &kmsg->msg, req->sr_msg.msg, | |
3840 | kmsg->uaddr, flags); | |
02d27d89 PB |
3841 | if (force_nonblock && ret == -EAGAIN) |
3842 | return io_setup_async_msg(req, kmsg); | |
03b1230c JA |
3843 | if (ret == -ERESTARTSYS) |
3844 | ret = -EINTR; | |
3845 | } | |
3846 | ||
1e95081c | 3847 | if (kmsg && kmsg->iov != kmsg->fast_iov) |
0b416c3e | 3848 | kfree(kmsg->iov); |
99bc4c38 | 3849 | req->flags &= ~REQ_F_NEED_CLEANUP; |
52de1fe1 | 3850 | __io_cqring_add_event(req, ret, cflags); |
4e88d6e7 JA |
3851 | if (ret < 0) |
3852 | req_set_fail_links(req); | |
014db007 | 3853 | io_put_req(req); |
03b1230c | 3854 | return 0; |
0fa03c62 | 3855 | } |
5d17b4a4 | 3856 | |
014db007 | 3857 | static int io_recv(struct io_kiocb *req, bool force_nonblock) |
fddaface | 3858 | { |
bcda7baa | 3859 | struct io_buffer *kbuf = NULL; |
fddaface | 3860 | struct socket *sock; |
bcda7baa | 3861 | int ret, cflags = 0; |
fddaface JA |
3862 | |
3863 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) | |
3864 | return -EINVAL; | |
3865 | ||
3866 | sock = sock_from_file(req->file, &ret); | |
3867 | if (sock) { | |
3868 | struct io_sr_msg *sr = &req->sr_msg; | |
bcda7baa | 3869 | void __user *buf = sr->buf; |
fddaface JA |
3870 | struct msghdr msg; |
3871 | struct iovec iov; | |
3872 | unsigned flags; | |
3873 | ||
bcda7baa JA |
3874 | kbuf = io_recv_buffer_select(req, &cflags, !force_nonblock); |
3875 | if (IS_ERR(kbuf)) | |
3876 | return PTR_ERR(kbuf); | |
3877 | else if (kbuf) | |
3878 | buf = u64_to_user_ptr(kbuf->addr); | |
3879 | ||
3880 | ret = import_single_range(READ, buf, sr->len, &iov, | |
fddaface | 3881 | &msg.msg_iter); |
bcda7baa JA |
3882 | if (ret) { |
3883 | kfree(kbuf); | |
fddaface | 3884 | return ret; |
bcda7baa | 3885 | } |
fddaface | 3886 | |
bcda7baa | 3887 | req->flags |= REQ_F_NEED_CLEANUP; |
fddaface JA |
3888 | msg.msg_name = NULL; |
3889 | msg.msg_control = NULL; | |
3890 | msg.msg_controllen = 0; | |
3891 | msg.msg_namelen = 0; | |
3892 | msg.msg_iocb = NULL; | |
3893 | msg.msg_flags = 0; | |
3894 | ||
3895 | flags = req->sr_msg.msg_flags; | |
3896 | if (flags & MSG_DONTWAIT) | |
3897 | req->flags |= REQ_F_NOWAIT; | |
3898 | else if (force_nonblock) | |
3899 | flags |= MSG_DONTWAIT; | |
3900 | ||
0b7b21e4 | 3901 | ret = sock_recvmsg(sock, &msg, flags); |
fddaface JA |
3902 | if (force_nonblock && ret == -EAGAIN) |
3903 | return -EAGAIN; | |
3904 | if (ret == -ERESTARTSYS) | |
3905 | ret = -EINTR; | |
3906 | } | |
3907 | ||
bcda7baa JA |
3908 | kfree(kbuf); |
3909 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
3910 | __io_cqring_add_event(req, ret, cflags); | |
fddaface JA |
3911 | if (ret < 0) |
3912 | req_set_fail_links(req); | |
014db007 | 3913 | io_put_req(req); |
fddaface | 3914 | return 0; |
fddaface JA |
3915 | } |
3916 | ||
3529d8c2 | 3917 | static int io_accept_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
17f2fe35 | 3918 | { |
8ed8d3c3 JA |
3919 | struct io_accept *accept = &req->accept; |
3920 | ||
17f2fe35 JA |
3921 | if (unlikely(req->ctx->flags & (IORING_SETUP_IOPOLL|IORING_SETUP_SQPOLL))) |
3922 | return -EINVAL; | |
8042d6ce | 3923 | if (sqe->ioprio || sqe->len || sqe->buf_index) |
17f2fe35 JA |
3924 | return -EINVAL; |
3925 | ||
d55e5f5b JA |
3926 | accept->addr = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
3927 | accept->addr_len = u64_to_user_ptr(READ_ONCE(sqe->addr2)); | |
8ed8d3c3 | 3928 | accept->flags = READ_ONCE(sqe->accept_flags); |
09952e3e | 3929 | accept->nofile = rlimit(RLIMIT_NOFILE); |
8ed8d3c3 | 3930 | return 0; |
8ed8d3c3 | 3931 | } |
17f2fe35 | 3932 | |
014db007 | 3933 | static int __io_accept(struct io_kiocb *req, bool force_nonblock) |
8ed8d3c3 JA |
3934 | { |
3935 | struct io_accept *accept = &req->accept; | |
3936 | unsigned file_flags; | |
3937 | int ret; | |
3938 | ||
3939 | file_flags = force_nonblock ? O_NONBLOCK : 0; | |
3940 | ret = __sys_accept4_file(req->file, file_flags, accept->addr, | |
09952e3e JA |
3941 | accept->addr_len, accept->flags, |
3942 | accept->nofile); | |
8ed8d3c3 | 3943 | if (ret == -EAGAIN && force_nonblock) |
17f2fe35 | 3944 | return -EAGAIN; |
8e3cca12 JA |
3945 | if (ret == -ERESTARTSYS) |
3946 | ret = -EINTR; | |
4e88d6e7 JA |
3947 | if (ret < 0) |
3948 | req_set_fail_links(req); | |
78e19bbe | 3949 | io_cqring_add_event(req, ret); |
014db007 | 3950 | io_put_req(req); |
17f2fe35 | 3951 | return 0; |
8ed8d3c3 JA |
3952 | } |
3953 | ||
3954 | static void io_accept_finish(struct io_wq_work **workptr) | |
3955 | { | |
3956 | struct io_kiocb *req = container_of(*workptr, struct io_kiocb, work); | |
8ed8d3c3 JA |
3957 | |
3958 | if (io_req_cancelled(req)) | |
3959 | return; | |
014db007 | 3960 | __io_accept(req, false); |
e9fd9396 | 3961 | io_steal_work(req, workptr); |
8ed8d3c3 | 3962 | } |
8ed8d3c3 | 3963 | |
014db007 | 3964 | static int io_accept(struct io_kiocb *req, bool force_nonblock) |
8ed8d3c3 | 3965 | { |
8ed8d3c3 JA |
3966 | int ret; |
3967 | ||
014db007 | 3968 | ret = __io_accept(req, force_nonblock); |
8ed8d3c3 JA |
3969 | if (ret == -EAGAIN && force_nonblock) { |
3970 | req->work.func = io_accept_finish; | |
8ed8d3c3 JA |
3971 | return -EAGAIN; |
3972 | } | |
3973 | return 0; | |
0fa03c62 | 3974 | } |
5d17b4a4 | 3975 | |
3529d8c2 | 3976 | static int io_connect_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
f499a021 | 3977 | { |
3529d8c2 JA |
3978 | struct io_connect *conn = &req->connect; |
3979 | struct io_async_ctx *io = req->io; | |
f499a021 | 3980 | |
3fbb51c1 JA |
3981 | if (unlikely(req->ctx->flags & (IORING_SETUP_IOPOLL|IORING_SETUP_SQPOLL))) |
3982 | return -EINVAL; | |
3983 | if (sqe->ioprio || sqe->len || sqe->buf_index || sqe->rw_flags) | |
3984 | return -EINVAL; | |
3985 | ||
3529d8c2 JA |
3986 | conn->addr = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
3987 | conn->addr_len = READ_ONCE(sqe->addr2); | |
3988 | ||
3989 | if (!io) | |
3990 | return 0; | |
3991 | ||
3992 | return move_addr_to_kernel(conn->addr, conn->addr_len, | |
3fbb51c1 | 3993 | &io->connect.address); |
f499a021 JA |
3994 | } |
3995 | ||
014db007 | 3996 | static int io_connect(struct io_kiocb *req, bool force_nonblock) |
f8e85cf2 | 3997 | { |
f499a021 | 3998 | struct io_async_ctx __io, *io; |
f8e85cf2 | 3999 | unsigned file_flags; |
3fbb51c1 | 4000 | int ret; |
f8e85cf2 | 4001 | |
f499a021 JA |
4002 | if (req->io) { |
4003 | io = req->io; | |
4004 | } else { | |
3529d8c2 JA |
4005 | ret = move_addr_to_kernel(req->connect.addr, |
4006 | req->connect.addr_len, | |
4007 | &__io.connect.address); | |
f499a021 JA |
4008 | if (ret) |
4009 | goto out; | |
4010 | io = &__io; | |
4011 | } | |
4012 | ||
3fbb51c1 JA |
4013 | file_flags = force_nonblock ? O_NONBLOCK : 0; |
4014 | ||
4015 | ret = __sys_connect_file(req->file, &io->connect.address, | |
4016 | req->connect.addr_len, file_flags); | |
87f80d62 | 4017 | if ((ret == -EAGAIN || ret == -EINPROGRESS) && force_nonblock) { |
b7bb4f7d JA |
4018 | if (req->io) |
4019 | return -EAGAIN; | |
4020 | if (io_alloc_async_ctx(req)) { | |
f499a021 JA |
4021 | ret = -ENOMEM; |
4022 | goto out; | |
4023 | } | |
b7bb4f7d | 4024 | memcpy(&req->io->connect, &__io.connect, sizeof(__io.connect)); |
f8e85cf2 | 4025 | return -EAGAIN; |
f499a021 | 4026 | } |
f8e85cf2 JA |
4027 | if (ret == -ERESTARTSYS) |
4028 | ret = -EINTR; | |
f499a021 | 4029 | out: |
4e88d6e7 JA |
4030 | if (ret < 0) |
4031 | req_set_fail_links(req); | |
f8e85cf2 | 4032 | io_cqring_add_event(req, ret); |
014db007 | 4033 | io_put_req(req); |
f8e85cf2 | 4034 | return 0; |
469956e8 Y |
4035 | } |
4036 | #else /* !CONFIG_NET */ | |
4037 | static int io_sendmsg_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) | |
4038 | { | |
f8e85cf2 | 4039 | return -EOPNOTSUPP; |
f8e85cf2 JA |
4040 | } |
4041 | ||
469956e8 Y |
4042 | static int io_sendmsg(struct io_kiocb *req, bool force_nonblock) |
4043 | { | |
4044 | return -EOPNOTSUPP; | |
4045 | } | |
4046 | ||
4047 | static int io_send(struct io_kiocb *req, bool force_nonblock) | |
4048 | { | |
4049 | return -EOPNOTSUPP; | |
4050 | } | |
4051 | ||
4052 | static int io_recvmsg_prep(struct io_kiocb *req, | |
4053 | const struct io_uring_sqe *sqe) | |
4054 | { | |
4055 | return -EOPNOTSUPP; | |
4056 | } | |
4057 | ||
4058 | static int io_recvmsg(struct io_kiocb *req, bool force_nonblock) | |
4059 | { | |
4060 | return -EOPNOTSUPP; | |
4061 | } | |
4062 | ||
4063 | static int io_recv(struct io_kiocb *req, bool force_nonblock) | |
4064 | { | |
4065 | return -EOPNOTSUPP; | |
4066 | } | |
4067 | ||
4068 | static int io_accept_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) | |
4069 | { | |
4070 | return -EOPNOTSUPP; | |
4071 | } | |
4072 | ||
4073 | static int io_accept(struct io_kiocb *req, bool force_nonblock) | |
4074 | { | |
4075 | return -EOPNOTSUPP; | |
4076 | } | |
4077 | ||
4078 | static int io_connect_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) | |
4079 | { | |
4080 | return -EOPNOTSUPP; | |
4081 | } | |
4082 | ||
4083 | static int io_connect(struct io_kiocb *req, bool force_nonblock) | |
4084 | { | |
f8e85cf2 | 4085 | return -EOPNOTSUPP; |
f8e85cf2 | 4086 | } |
469956e8 | 4087 | #endif /* CONFIG_NET */ |
f8e85cf2 | 4088 | |
d7718a9d JA |
4089 | struct io_poll_table { |
4090 | struct poll_table_struct pt; | |
4091 | struct io_kiocb *req; | |
4092 | int error; | |
4093 | }; | |
4094 | ||
4095 | static void __io_queue_proc(struct io_poll_iocb *poll, struct io_poll_table *pt, | |
4096 | struct wait_queue_head *head) | |
4097 | { | |
4098 | if (unlikely(poll->head)) { | |
4099 | pt->error = -EINVAL; | |
4100 | return; | |
4101 | } | |
4102 | ||
4103 | pt->error = 0; | |
4104 | poll->head = head; | |
4105 | add_wait_queue(head, &poll->wait); | |
4106 | } | |
4107 | ||
4108 | static void io_async_queue_proc(struct file *file, struct wait_queue_head *head, | |
4109 | struct poll_table_struct *p) | |
4110 | { | |
4111 | struct io_poll_table *pt = container_of(p, struct io_poll_table, pt); | |
4112 | ||
4113 | __io_queue_proc(&pt->req->apoll->poll, pt, head); | |
4114 | } | |
4115 | ||
4116 | static int __io_async_wake(struct io_kiocb *req, struct io_poll_iocb *poll, | |
4117 | __poll_t mask, task_work_func_t func) | |
4118 | { | |
4119 | struct task_struct *tsk; | |
aa96bf8a | 4120 | int ret; |
d7718a9d JA |
4121 | |
4122 | /* for instances that support it check for an event match first: */ | |
4123 | if (mask && !(mask & poll->events)) | |
4124 | return 0; | |
4125 | ||
4126 | trace_io_uring_task_add(req->ctx, req->opcode, req->user_data, mask); | |
4127 | ||
4128 | list_del_init(&poll->wait.entry); | |
4129 | ||
4130 | tsk = req->task; | |
4131 | req->result = mask; | |
4132 | init_task_work(&req->task_work, func); | |
4133 | /* | |
aa96bf8a JA |
4134 | * If this fails, then the task is exiting. Punt to one of the io-wq |
4135 | * threads to ensure the work gets run, we can't always rely on exit | |
4136 | * cancelation taking care of this. | |
d7718a9d | 4137 | */ |
aa96bf8a JA |
4138 | ret = task_work_add(tsk, &req->task_work, true); |
4139 | if (unlikely(ret)) { | |
4140 | tsk = io_wq_get_task(req->ctx->io_wq); | |
4141 | task_work_add(tsk, &req->task_work, true); | |
4142 | } | |
d7718a9d JA |
4143 | wake_up_process(tsk); |
4144 | return 1; | |
4145 | } | |
4146 | ||
74ce6ce4 JA |
4147 | static bool io_poll_rewait(struct io_kiocb *req, struct io_poll_iocb *poll) |
4148 | __acquires(&req->ctx->completion_lock) | |
4149 | { | |
4150 | struct io_ring_ctx *ctx = req->ctx; | |
4151 | ||
4152 | if (!req->result && !READ_ONCE(poll->canceled)) { | |
4153 | struct poll_table_struct pt = { ._key = poll->events }; | |
4154 | ||
4155 | req->result = vfs_poll(req->file, &pt) & poll->events; | |
4156 | } | |
4157 | ||
4158 | spin_lock_irq(&ctx->completion_lock); | |
4159 | if (!req->result && !READ_ONCE(poll->canceled)) { | |
4160 | add_wait_queue(poll->head, &poll->wait); | |
4161 | return true; | |
4162 | } | |
4163 | ||
4164 | return false; | |
4165 | } | |
4166 | ||
d7718a9d JA |
4167 | static void io_async_task_func(struct callback_head *cb) |
4168 | { | |
4169 | struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work); | |
4170 | struct async_poll *apoll = req->apoll; | |
4171 | struct io_ring_ctx *ctx = req->ctx; | |
2bae047e | 4172 | bool canceled; |
d7718a9d JA |
4173 | |
4174 | trace_io_uring_task_run(req->ctx, req->opcode, req->user_data); | |
4175 | ||
74ce6ce4 | 4176 | if (io_poll_rewait(req, &apoll->poll)) { |
d7718a9d | 4177 | spin_unlock_irq(&ctx->completion_lock); |
74ce6ce4 | 4178 | return; |
d7718a9d JA |
4179 | } |
4180 | ||
74ce6ce4 JA |
4181 | if (hash_hashed(&req->hash_node)) |
4182 | hash_del(&req->hash_node); | |
4183 | ||
2bae047e JA |
4184 | canceled = READ_ONCE(apoll->poll.canceled); |
4185 | if (canceled) { | |
4186 | io_cqring_fill_event(req, -ECANCELED); | |
4187 | io_commit_cqring(ctx); | |
4188 | } | |
4189 | ||
74ce6ce4 JA |
4190 | spin_unlock_irq(&ctx->completion_lock); |
4191 | ||
44575a67 XW |
4192 | /* restore ->work in case we need to retry again */ |
4193 | memcpy(&req->work, &apoll->work, sizeof(req->work)); | |
4194 | ||
2bae047e JA |
4195 | if (canceled) { |
4196 | kfree(apoll); | |
4197 | io_cqring_ev_posted(ctx); | |
4198 | req_set_fail_links(req); | |
44575a67 | 4199 | io_double_put_req(req); |
2bae047e JA |
4200 | return; |
4201 | } | |
4202 | ||
d7718a9d JA |
4203 | __set_current_state(TASK_RUNNING); |
4204 | mutex_lock(&ctx->uring_lock); | |
4205 | __io_queue_sqe(req, NULL); | |
4206 | mutex_unlock(&ctx->uring_lock); | |
4207 | ||
4208 | kfree(apoll); | |
4209 | } | |
4210 | ||
4211 | static int io_async_wake(struct wait_queue_entry *wait, unsigned mode, int sync, | |
4212 | void *key) | |
4213 | { | |
4214 | struct io_kiocb *req = wait->private; | |
4215 | struct io_poll_iocb *poll = &req->apoll->poll; | |
4216 | ||
4217 | trace_io_uring_poll_wake(req->ctx, req->opcode, req->user_data, | |
4218 | key_to_poll(key)); | |
4219 | ||
4220 | return __io_async_wake(req, poll, key_to_poll(key), io_async_task_func); | |
4221 | } | |
4222 | ||
4223 | static void io_poll_req_insert(struct io_kiocb *req) | |
4224 | { | |
4225 | struct io_ring_ctx *ctx = req->ctx; | |
4226 | struct hlist_head *list; | |
4227 | ||
4228 | list = &ctx->cancel_hash[hash_long(req->user_data, ctx->cancel_hash_bits)]; | |
4229 | hlist_add_head(&req->hash_node, list); | |
4230 | } | |
4231 | ||
4232 | static __poll_t __io_arm_poll_handler(struct io_kiocb *req, | |
4233 | struct io_poll_iocb *poll, | |
4234 | struct io_poll_table *ipt, __poll_t mask, | |
4235 | wait_queue_func_t wake_func) | |
4236 | __acquires(&ctx->completion_lock) | |
4237 | { | |
4238 | struct io_ring_ctx *ctx = req->ctx; | |
4239 | bool cancel = false; | |
4240 | ||
4241 | poll->file = req->file; | |
4242 | poll->head = NULL; | |
4243 | poll->done = poll->canceled = false; | |
4244 | poll->events = mask; | |
4245 | ||
4246 | ipt->pt._key = mask; | |
4247 | ipt->req = req; | |
4248 | ipt->error = -EINVAL; | |
4249 | ||
4250 | INIT_LIST_HEAD(&poll->wait.entry); | |
4251 | init_waitqueue_func_entry(&poll->wait, wake_func); | |
4252 | poll->wait.private = req; | |
4253 | ||
4254 | mask = vfs_poll(req->file, &ipt->pt) & poll->events; | |
4255 | ||
4256 | spin_lock_irq(&ctx->completion_lock); | |
4257 | if (likely(poll->head)) { | |
4258 | spin_lock(&poll->head->lock); | |
4259 | if (unlikely(list_empty(&poll->wait.entry))) { | |
4260 | if (ipt->error) | |
4261 | cancel = true; | |
4262 | ipt->error = 0; | |
4263 | mask = 0; | |
4264 | } | |
4265 | if (mask || ipt->error) | |
4266 | list_del_init(&poll->wait.entry); | |
4267 | else if (cancel) | |
4268 | WRITE_ONCE(poll->canceled, true); | |
4269 | else if (!poll->done) /* actually waiting for an event */ | |
4270 | io_poll_req_insert(req); | |
4271 | spin_unlock(&poll->head->lock); | |
4272 | } | |
4273 | ||
4274 | return mask; | |
4275 | } | |
4276 | ||
4277 | static bool io_arm_poll_handler(struct io_kiocb *req) | |
4278 | { | |
4279 | const struct io_op_def *def = &io_op_defs[req->opcode]; | |
4280 | struct io_ring_ctx *ctx = req->ctx; | |
4281 | struct async_poll *apoll; | |
4282 | struct io_poll_table ipt; | |
4283 | __poll_t mask, ret; | |
4284 | ||
4285 | if (!req->file || !file_can_poll(req->file)) | |
4286 | return false; | |
4287 | if (req->flags & (REQ_F_MUST_PUNT | REQ_F_POLLED)) | |
4288 | return false; | |
4289 | if (!def->pollin && !def->pollout) | |
4290 | return false; | |
4291 | ||
4292 | apoll = kmalloc(sizeof(*apoll), GFP_ATOMIC); | |
4293 | if (unlikely(!apoll)) | |
4294 | return false; | |
4295 | ||
4296 | req->flags |= REQ_F_POLLED; | |
4297 | memcpy(&apoll->work, &req->work, sizeof(req->work)); | |
4298 | ||
3537b6a7 | 4299 | get_task_struct(current); |
d7718a9d JA |
4300 | req->task = current; |
4301 | req->apoll = apoll; | |
4302 | INIT_HLIST_NODE(&req->hash_node); | |
4303 | ||
8755d97a | 4304 | mask = 0; |
d7718a9d | 4305 | if (def->pollin) |
8755d97a | 4306 | mask |= POLLIN | POLLRDNORM; |
d7718a9d JA |
4307 | if (def->pollout) |
4308 | mask |= POLLOUT | POLLWRNORM; | |
4309 | mask |= POLLERR | POLLPRI; | |
4310 | ||
4311 | ipt.pt._qproc = io_async_queue_proc; | |
4312 | ||
4313 | ret = __io_arm_poll_handler(req, &apoll->poll, &ipt, mask, | |
4314 | io_async_wake); | |
4315 | if (ret) { | |
4316 | ipt.error = 0; | |
4317 | apoll->poll.done = true; | |
4318 | spin_unlock_irq(&ctx->completion_lock); | |
4319 | memcpy(&req->work, &apoll->work, sizeof(req->work)); | |
4320 | kfree(apoll); | |
4321 | return false; | |
4322 | } | |
4323 | spin_unlock_irq(&ctx->completion_lock); | |
4324 | trace_io_uring_poll_arm(ctx, req->opcode, req->user_data, mask, | |
4325 | apoll->poll.events); | |
4326 | return true; | |
4327 | } | |
4328 | ||
4329 | static bool __io_poll_remove_one(struct io_kiocb *req, | |
4330 | struct io_poll_iocb *poll) | |
221c5eb2 | 4331 | { |
b41e9852 | 4332 | bool do_complete = false; |
221c5eb2 JA |
4333 | |
4334 | spin_lock(&poll->head->lock); | |
4335 | WRITE_ONCE(poll->canceled, true); | |
392edb45 JA |
4336 | if (!list_empty(&poll->wait.entry)) { |
4337 | list_del_init(&poll->wait.entry); | |
b41e9852 | 4338 | do_complete = true; |
221c5eb2 JA |
4339 | } |
4340 | spin_unlock(&poll->head->lock); | |
d7718a9d JA |
4341 | return do_complete; |
4342 | } | |
4343 | ||
4344 | static bool io_poll_remove_one(struct io_kiocb *req) | |
4345 | { | |
b1f573bd | 4346 | struct async_poll *apoll = NULL; |
d7718a9d JA |
4347 | bool do_complete; |
4348 | ||
4349 | if (req->opcode == IORING_OP_POLL_ADD) { | |
4350 | do_complete = __io_poll_remove_one(req, &req->poll); | |
4351 | } else { | |
b1f573bd | 4352 | apoll = req->apoll; |
d7718a9d JA |
4353 | /* non-poll requests have submit ref still */ |
4354 | do_complete = __io_poll_remove_one(req, &req->apoll->poll); | |
4355 | if (do_complete) | |
4356 | io_put_req(req); | |
4357 | } | |
4358 | ||
78076bb6 | 4359 | hash_del(&req->hash_node); |
d7718a9d | 4360 | |
44575a67 | 4361 | if (do_complete && apoll) { |
b1f573bd XW |
4362 | /* |
4363 | * restore ->work because we need to call io_req_work_drop_env. | |
4364 | */ | |
4365 | memcpy(&req->work, &apoll->work, sizeof(req->work)); | |
4366 | kfree(apoll); | |
4367 | } | |
4368 | ||
b41e9852 JA |
4369 | if (do_complete) { |
4370 | io_cqring_fill_event(req, -ECANCELED); | |
4371 | io_commit_cqring(req->ctx); | |
4372 | req->flags |= REQ_F_COMP_LOCKED; | |
4373 | io_put_req(req); | |
4374 | } | |
4375 | ||
4376 | return do_complete; | |
221c5eb2 JA |
4377 | } |
4378 | ||
4379 | static void io_poll_remove_all(struct io_ring_ctx *ctx) | |
4380 | { | |
78076bb6 | 4381 | struct hlist_node *tmp; |
221c5eb2 | 4382 | struct io_kiocb *req; |
8e2e1faf | 4383 | int posted = 0, i; |
221c5eb2 JA |
4384 | |
4385 | spin_lock_irq(&ctx->completion_lock); | |
78076bb6 JA |
4386 | for (i = 0; i < (1U << ctx->cancel_hash_bits); i++) { |
4387 | struct hlist_head *list; | |
4388 | ||
4389 | list = &ctx->cancel_hash[i]; | |
4390 | hlist_for_each_entry_safe(req, tmp, list, hash_node) | |
8e2e1faf | 4391 | posted += io_poll_remove_one(req); |
221c5eb2 JA |
4392 | } |
4393 | spin_unlock_irq(&ctx->completion_lock); | |
b41e9852 | 4394 | |
8e2e1faf JA |
4395 | if (posted) |
4396 | io_cqring_ev_posted(ctx); | |
221c5eb2 JA |
4397 | } |
4398 | ||
47f46768 JA |
4399 | static int io_poll_cancel(struct io_ring_ctx *ctx, __u64 sqe_addr) |
4400 | { | |
78076bb6 | 4401 | struct hlist_head *list; |
47f46768 JA |
4402 | struct io_kiocb *req; |
4403 | ||
78076bb6 JA |
4404 | list = &ctx->cancel_hash[hash_long(sqe_addr, ctx->cancel_hash_bits)]; |
4405 | hlist_for_each_entry(req, list, hash_node) { | |
b41e9852 JA |
4406 | if (sqe_addr != req->user_data) |
4407 | continue; | |
4408 | if (io_poll_remove_one(req)) | |
eac406c6 | 4409 | return 0; |
b41e9852 | 4410 | return -EALREADY; |
47f46768 JA |
4411 | } |
4412 | ||
4413 | return -ENOENT; | |
4414 | } | |
4415 | ||
3529d8c2 JA |
4416 | static int io_poll_remove_prep(struct io_kiocb *req, |
4417 | const struct io_uring_sqe *sqe) | |
0969e783 | 4418 | { |
0969e783 JA |
4419 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4420 | return -EINVAL; | |
4421 | if (sqe->ioprio || sqe->off || sqe->len || sqe->buf_index || | |
4422 | sqe->poll_events) | |
4423 | return -EINVAL; | |
4424 | ||
4425 | req->poll.addr = READ_ONCE(sqe->addr); | |
0969e783 JA |
4426 | return 0; |
4427 | } | |
4428 | ||
221c5eb2 JA |
4429 | /* |
4430 | * Find a running poll command that matches one specified in sqe->addr, | |
4431 | * and remove it if found. | |
4432 | */ | |
fc4df999 | 4433 | static int io_poll_remove(struct io_kiocb *req) |
221c5eb2 JA |
4434 | { |
4435 | struct io_ring_ctx *ctx = req->ctx; | |
0969e783 | 4436 | u64 addr; |
47f46768 | 4437 | int ret; |
221c5eb2 | 4438 | |
0969e783 | 4439 | addr = req->poll.addr; |
221c5eb2 | 4440 | spin_lock_irq(&ctx->completion_lock); |
0969e783 | 4441 | ret = io_poll_cancel(ctx, addr); |
221c5eb2 JA |
4442 | spin_unlock_irq(&ctx->completion_lock); |
4443 | ||
78e19bbe | 4444 | io_cqring_add_event(req, ret); |
4e88d6e7 JA |
4445 | if (ret < 0) |
4446 | req_set_fail_links(req); | |
e65ef56d | 4447 | io_put_req(req); |
221c5eb2 JA |
4448 | return 0; |
4449 | } | |
4450 | ||
b0dd8a41 | 4451 | static void io_poll_complete(struct io_kiocb *req, __poll_t mask, int error) |
221c5eb2 | 4452 | { |
a197f664 JL |
4453 | struct io_ring_ctx *ctx = req->ctx; |
4454 | ||
8c838788 | 4455 | req->poll.done = true; |
b0a20349 | 4456 | io_cqring_fill_event(req, error ? error : mangle_poll(mask)); |
8c838788 | 4457 | io_commit_cqring(ctx); |
221c5eb2 JA |
4458 | } |
4459 | ||
b41e9852 | 4460 | static void io_poll_task_handler(struct io_kiocb *req, struct io_kiocb **nxt) |
221c5eb2 | 4461 | { |
221c5eb2 | 4462 | struct io_ring_ctx *ctx = req->ctx; |
a6ba632d JA |
4463 | struct io_poll_iocb *poll = &req->poll; |
4464 | ||
74ce6ce4 | 4465 | if (io_poll_rewait(req, poll)) { |
a6ba632d JA |
4466 | spin_unlock_irq(&ctx->completion_lock); |
4467 | return; | |
4468 | } | |
74ce6ce4 | 4469 | |
78076bb6 | 4470 | hash_del(&req->hash_node); |
b41e9852 JA |
4471 | io_poll_complete(req, req->result, 0); |
4472 | req->flags |= REQ_F_COMP_LOCKED; | |
4473 | io_put_req_find_next(req, nxt); | |
221c5eb2 JA |
4474 | spin_unlock_irq(&ctx->completion_lock); |
4475 | ||
8c838788 | 4476 | io_cqring_ev_posted(ctx); |
221c5eb2 JA |
4477 | } |
4478 | ||
b41e9852 | 4479 | static void io_poll_task_func(struct callback_head *cb) |
e94f141b | 4480 | { |
b41e9852 JA |
4481 | struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work); |
4482 | struct io_kiocb *nxt = NULL; | |
e94f141b | 4483 | |
b41e9852 | 4484 | io_poll_task_handler(req, &nxt); |
d7718a9d JA |
4485 | if (nxt) { |
4486 | struct io_ring_ctx *ctx = nxt->ctx; | |
e94f141b | 4487 | |
d7718a9d | 4488 | mutex_lock(&ctx->uring_lock); |
b41e9852 | 4489 | __io_queue_sqe(nxt, NULL); |
d7718a9d | 4490 | mutex_unlock(&ctx->uring_lock); |
e94f141b | 4491 | } |
f0b493e6 JA |
4492 | } |
4493 | ||
221c5eb2 JA |
4494 | static int io_poll_wake(struct wait_queue_entry *wait, unsigned mode, int sync, |
4495 | void *key) | |
4496 | { | |
c2f2eb7d JA |
4497 | struct io_kiocb *req = wait->private; |
4498 | struct io_poll_iocb *poll = &req->poll; | |
221c5eb2 | 4499 | |
d7718a9d | 4500 | return __io_async_wake(req, poll, key_to_poll(key), io_poll_task_func); |
221c5eb2 JA |
4501 | } |
4502 | ||
221c5eb2 JA |
4503 | static void io_poll_queue_proc(struct file *file, struct wait_queue_head *head, |
4504 | struct poll_table_struct *p) | |
4505 | { | |
4506 | struct io_poll_table *pt = container_of(p, struct io_poll_table, pt); | |
4507 | ||
d7718a9d | 4508 | __io_queue_proc(&pt->req->poll, pt, head); |
eac406c6 JA |
4509 | } |
4510 | ||
3529d8c2 | 4511 | static int io_poll_add_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
221c5eb2 JA |
4512 | { |
4513 | struct io_poll_iocb *poll = &req->poll; | |
221c5eb2 | 4514 | u16 events; |
221c5eb2 JA |
4515 | |
4516 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) | |
4517 | return -EINVAL; | |
4518 | if (sqe->addr || sqe->ioprio || sqe->off || sqe->len || sqe->buf_index) | |
4519 | return -EINVAL; | |
09bb8394 JA |
4520 | if (!poll->file) |
4521 | return -EBADF; | |
221c5eb2 | 4522 | |
221c5eb2 JA |
4523 | events = READ_ONCE(sqe->poll_events); |
4524 | poll->events = demangle_poll(events) | EPOLLERR | EPOLLHUP; | |
b41e9852 | 4525 | |
3537b6a7 | 4526 | get_task_struct(current); |
b41e9852 | 4527 | req->task = current; |
0969e783 JA |
4528 | return 0; |
4529 | } | |
4530 | ||
014db007 | 4531 | static int io_poll_add(struct io_kiocb *req) |
0969e783 JA |
4532 | { |
4533 | struct io_poll_iocb *poll = &req->poll; | |
4534 | struct io_ring_ctx *ctx = req->ctx; | |
4535 | struct io_poll_table ipt; | |
0969e783 | 4536 | __poll_t mask; |
0969e783 | 4537 | |
78076bb6 | 4538 | INIT_HLIST_NODE(&req->hash_node); |
36703247 | 4539 | INIT_LIST_HEAD(&req->list); |
d7718a9d | 4540 | ipt.pt._qproc = io_poll_queue_proc; |
36703247 | 4541 | |
d7718a9d JA |
4542 | mask = __io_arm_poll_handler(req, &req->poll, &ipt, poll->events, |
4543 | io_poll_wake); | |
221c5eb2 | 4544 | |
8c838788 | 4545 | if (mask) { /* no async, we'd stolen it */ |
221c5eb2 | 4546 | ipt.error = 0; |
b0dd8a41 | 4547 | io_poll_complete(req, mask, 0); |
221c5eb2 | 4548 | } |
221c5eb2 JA |
4549 | spin_unlock_irq(&ctx->completion_lock); |
4550 | ||
8c838788 JA |
4551 | if (mask) { |
4552 | io_cqring_ev_posted(ctx); | |
014db007 | 4553 | io_put_req(req); |
221c5eb2 | 4554 | } |
8c838788 | 4555 | return ipt.error; |
221c5eb2 JA |
4556 | } |
4557 | ||
5262f567 JA |
4558 | static enum hrtimer_restart io_timeout_fn(struct hrtimer *timer) |
4559 | { | |
ad8a48ac JA |
4560 | struct io_timeout_data *data = container_of(timer, |
4561 | struct io_timeout_data, timer); | |
4562 | struct io_kiocb *req = data->req; | |
4563 | struct io_ring_ctx *ctx = req->ctx; | |
5262f567 JA |
4564 | unsigned long flags; |
4565 | ||
5262f567 JA |
4566 | atomic_inc(&ctx->cq_timeouts); |
4567 | ||
4568 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
ef03681a | 4569 | /* |
11365043 JA |
4570 | * We could be racing with timeout deletion. If the list is empty, |
4571 | * then timeout lookup already found it and will be handling it. | |
ef03681a | 4572 | */ |
842f9612 | 4573 | if (!list_empty(&req->list)) { |
11365043 | 4574 | struct io_kiocb *prev; |
5262f567 | 4575 | |
11365043 JA |
4576 | /* |
4577 | * Adjust the reqs sequence before the current one because it | |
d195a66e | 4578 | * will consume a slot in the cq_ring and the cq_tail |
11365043 JA |
4579 | * pointer will be increased, otherwise other timeout reqs may |
4580 | * return in advance without waiting for enough wait_nr. | |
4581 | */ | |
4582 | prev = req; | |
4583 | list_for_each_entry_continue_reverse(prev, &ctx->timeout_list, list) | |
4584 | prev->sequence++; | |
11365043 | 4585 | list_del_init(&req->list); |
11365043 | 4586 | } |
5262f567 | 4587 | |
78e19bbe | 4588 | io_cqring_fill_event(req, -ETIME); |
5262f567 JA |
4589 | io_commit_cqring(ctx); |
4590 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
4591 | ||
4592 | io_cqring_ev_posted(ctx); | |
4e88d6e7 | 4593 | req_set_fail_links(req); |
5262f567 JA |
4594 | io_put_req(req); |
4595 | return HRTIMER_NORESTART; | |
4596 | } | |
4597 | ||
47f46768 JA |
4598 | static int io_timeout_cancel(struct io_ring_ctx *ctx, __u64 user_data) |
4599 | { | |
4600 | struct io_kiocb *req; | |
4601 | int ret = -ENOENT; | |
4602 | ||
4603 | list_for_each_entry(req, &ctx->timeout_list, list) { | |
4604 | if (user_data == req->user_data) { | |
4605 | list_del_init(&req->list); | |
4606 | ret = 0; | |
4607 | break; | |
4608 | } | |
4609 | } | |
4610 | ||
4611 | if (ret == -ENOENT) | |
4612 | return ret; | |
4613 | ||
2d28390a | 4614 | ret = hrtimer_try_to_cancel(&req->io->timeout.timer); |
47f46768 JA |
4615 | if (ret == -1) |
4616 | return -EALREADY; | |
4617 | ||
4e88d6e7 | 4618 | req_set_fail_links(req); |
47f46768 JA |
4619 | io_cqring_fill_event(req, -ECANCELED); |
4620 | io_put_req(req); | |
4621 | return 0; | |
4622 | } | |
4623 | ||
3529d8c2 JA |
4624 | static int io_timeout_remove_prep(struct io_kiocb *req, |
4625 | const struct io_uring_sqe *sqe) | |
b29472ee | 4626 | { |
b29472ee JA |
4627 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4628 | return -EINVAL; | |
4629 | if (sqe->flags || sqe->ioprio || sqe->buf_index || sqe->len) | |
4630 | return -EINVAL; | |
4631 | ||
4632 | req->timeout.addr = READ_ONCE(sqe->addr); | |
4633 | req->timeout.flags = READ_ONCE(sqe->timeout_flags); | |
4634 | if (req->timeout.flags) | |
4635 | return -EINVAL; | |
4636 | ||
b29472ee JA |
4637 | return 0; |
4638 | } | |
4639 | ||
11365043 JA |
4640 | /* |
4641 | * Remove or update an existing timeout command | |
4642 | */ | |
fc4df999 | 4643 | static int io_timeout_remove(struct io_kiocb *req) |
11365043 JA |
4644 | { |
4645 | struct io_ring_ctx *ctx = req->ctx; | |
47f46768 | 4646 | int ret; |
11365043 | 4647 | |
11365043 | 4648 | spin_lock_irq(&ctx->completion_lock); |
b29472ee | 4649 | ret = io_timeout_cancel(ctx, req->timeout.addr); |
11365043 | 4650 | |
47f46768 | 4651 | io_cqring_fill_event(req, ret); |
11365043 JA |
4652 | io_commit_cqring(ctx); |
4653 | spin_unlock_irq(&ctx->completion_lock); | |
5262f567 | 4654 | io_cqring_ev_posted(ctx); |
4e88d6e7 JA |
4655 | if (ret < 0) |
4656 | req_set_fail_links(req); | |
ec9c02ad | 4657 | io_put_req(req); |
11365043 | 4658 | return 0; |
5262f567 JA |
4659 | } |
4660 | ||
3529d8c2 | 4661 | static int io_timeout_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe, |
2d28390a | 4662 | bool is_timeout_link) |
5262f567 | 4663 | { |
ad8a48ac | 4664 | struct io_timeout_data *data; |
a41525ab | 4665 | unsigned flags; |
5262f567 | 4666 | |
ad8a48ac | 4667 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
5262f567 | 4668 | return -EINVAL; |
ad8a48ac | 4669 | if (sqe->ioprio || sqe->buf_index || sqe->len != 1) |
a41525ab | 4670 | return -EINVAL; |
2d28390a JA |
4671 | if (sqe->off && is_timeout_link) |
4672 | return -EINVAL; | |
a41525ab JA |
4673 | flags = READ_ONCE(sqe->timeout_flags); |
4674 | if (flags & ~IORING_TIMEOUT_ABS) | |
5262f567 | 4675 | return -EINVAL; |
bdf20073 | 4676 | |
26a61679 JA |
4677 | req->timeout.count = READ_ONCE(sqe->off); |
4678 | ||
3529d8c2 | 4679 | if (!req->io && io_alloc_async_ctx(req)) |
26a61679 JA |
4680 | return -ENOMEM; |
4681 | ||
4682 | data = &req->io->timeout; | |
ad8a48ac | 4683 | data->req = req; |
ad8a48ac JA |
4684 | req->flags |= REQ_F_TIMEOUT; |
4685 | ||
4686 | if (get_timespec64(&data->ts, u64_to_user_ptr(sqe->addr))) | |
5262f567 JA |
4687 | return -EFAULT; |
4688 | ||
11365043 | 4689 | if (flags & IORING_TIMEOUT_ABS) |
ad8a48ac | 4690 | data->mode = HRTIMER_MODE_ABS; |
11365043 | 4691 | else |
ad8a48ac | 4692 | data->mode = HRTIMER_MODE_REL; |
11365043 | 4693 | |
ad8a48ac JA |
4694 | hrtimer_init(&data->timer, CLOCK_MONOTONIC, data->mode); |
4695 | return 0; | |
4696 | } | |
4697 | ||
fc4df999 | 4698 | static int io_timeout(struct io_kiocb *req) |
ad8a48ac | 4699 | { |
ad8a48ac JA |
4700 | struct io_ring_ctx *ctx = req->ctx; |
4701 | struct io_timeout_data *data; | |
4702 | struct list_head *entry; | |
4703 | unsigned span = 0; | |
b55ce732 | 4704 | u32 count = req->timeout.count; |
22cad158 | 4705 | u32 seq = req->sequence; |
ad8a48ac | 4706 | |
2d28390a | 4707 | data = &req->io->timeout; |
93bd25bb | 4708 | |
5262f567 JA |
4709 | /* |
4710 | * sqe->off holds how many events that need to occur for this | |
93bd25bb JA |
4711 | * timeout event to be satisfied. If it isn't set, then this is |
4712 | * a pure timeout request, sequence isn't used. | |
5262f567 | 4713 | */ |
93bd25bb JA |
4714 | if (!count) { |
4715 | req->flags |= REQ_F_TIMEOUT_NOSEQ; | |
4716 | spin_lock_irq(&ctx->completion_lock); | |
4717 | entry = ctx->timeout_list.prev; | |
4718 | goto add; | |
4719 | } | |
5262f567 | 4720 | |
22cad158 | 4721 | req->sequence = seq + count; |
5262f567 JA |
4722 | |
4723 | /* | |
4724 | * Insertion sort, ensuring the first entry in the list is always | |
4725 | * the one we need first. | |
4726 | */ | |
5262f567 JA |
4727 | spin_lock_irq(&ctx->completion_lock); |
4728 | list_for_each_prev(entry, &ctx->timeout_list) { | |
4729 | struct io_kiocb *nxt = list_entry(entry, struct io_kiocb, list); | |
22cad158 | 4730 | unsigned nxt_seq; |
5da0fb1a | 4731 | long long tmp, tmp_nxt; |
b55ce732 | 4732 | u32 nxt_offset = nxt->timeout.count; |
5262f567 | 4733 | |
93bd25bb JA |
4734 | if (nxt->flags & REQ_F_TIMEOUT_NOSEQ) |
4735 | continue; | |
4736 | ||
5da0fb1a | 4737 | /* |
22cad158 | 4738 | * Since seq + count can overflow, use type long |
5da0fb1a | 4739 | * long to store it. |
4740 | */ | |
22cad158 PB |
4741 | tmp = (long long)seq + count; |
4742 | nxt_seq = nxt->sequence - nxt_offset; | |
4743 | tmp_nxt = (long long)nxt_seq + nxt_offset; | |
5da0fb1a | 4744 | |
4745 | /* | |
4746 | * cached_sq_head may overflow, and it will never overflow twice | |
4747 | * once there is some timeout req still be valid. | |
4748 | */ | |
22cad158 | 4749 | if (seq < nxt_seq) |
8b07a65a | 4750 | tmp += UINT_MAX; |
5da0fb1a | 4751 | |
a1f58ba4 | 4752 | if (tmp > tmp_nxt) |
5262f567 | 4753 | break; |
a1f58ba4 | 4754 | |
4755 | /* | |
4756 | * Sequence of reqs after the insert one and itself should | |
4757 | * be adjusted because each timeout req consumes a slot. | |
4758 | */ | |
4759 | span++; | |
4760 | nxt->sequence++; | |
5262f567 | 4761 | } |
a1f58ba4 | 4762 | req->sequence -= span; |
93bd25bb | 4763 | add: |
5262f567 | 4764 | list_add(&req->list, entry); |
ad8a48ac JA |
4765 | data->timer.function = io_timeout_fn; |
4766 | hrtimer_start(&data->timer, timespec64_to_ktime(data->ts), data->mode); | |
5262f567 | 4767 | spin_unlock_irq(&ctx->completion_lock); |
5262f567 JA |
4768 | return 0; |
4769 | } | |
5262f567 | 4770 | |
62755e35 JA |
4771 | static bool io_cancel_cb(struct io_wq_work *work, void *data) |
4772 | { | |
4773 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); | |
4774 | ||
4775 | return req->user_data == (unsigned long) data; | |
4776 | } | |
4777 | ||
e977d6d3 | 4778 | static int io_async_cancel_one(struct io_ring_ctx *ctx, void *sqe_addr) |
62755e35 | 4779 | { |
62755e35 | 4780 | enum io_wq_cancel cancel_ret; |
62755e35 JA |
4781 | int ret = 0; |
4782 | ||
62755e35 JA |
4783 | cancel_ret = io_wq_cancel_cb(ctx->io_wq, io_cancel_cb, sqe_addr); |
4784 | switch (cancel_ret) { | |
4785 | case IO_WQ_CANCEL_OK: | |
4786 | ret = 0; | |
4787 | break; | |
4788 | case IO_WQ_CANCEL_RUNNING: | |
4789 | ret = -EALREADY; | |
4790 | break; | |
4791 | case IO_WQ_CANCEL_NOTFOUND: | |
4792 | ret = -ENOENT; | |
4793 | break; | |
4794 | } | |
4795 | ||
e977d6d3 JA |
4796 | return ret; |
4797 | } | |
4798 | ||
47f46768 JA |
4799 | static void io_async_find_and_cancel(struct io_ring_ctx *ctx, |
4800 | struct io_kiocb *req, __u64 sqe_addr, | |
014db007 | 4801 | int success_ret) |
47f46768 JA |
4802 | { |
4803 | unsigned long flags; | |
4804 | int ret; | |
4805 | ||
4806 | ret = io_async_cancel_one(ctx, (void *) (unsigned long) sqe_addr); | |
4807 | if (ret != -ENOENT) { | |
4808 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
4809 | goto done; | |
4810 | } | |
4811 | ||
4812 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
4813 | ret = io_timeout_cancel(ctx, sqe_addr); | |
4814 | if (ret != -ENOENT) | |
4815 | goto done; | |
4816 | ret = io_poll_cancel(ctx, sqe_addr); | |
4817 | done: | |
b0dd8a41 JA |
4818 | if (!ret) |
4819 | ret = success_ret; | |
47f46768 JA |
4820 | io_cqring_fill_event(req, ret); |
4821 | io_commit_cqring(ctx); | |
4822 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
4823 | io_cqring_ev_posted(ctx); | |
4824 | ||
4e88d6e7 JA |
4825 | if (ret < 0) |
4826 | req_set_fail_links(req); | |
014db007 | 4827 | io_put_req(req); |
47f46768 JA |
4828 | } |
4829 | ||
3529d8c2 JA |
4830 | static int io_async_cancel_prep(struct io_kiocb *req, |
4831 | const struct io_uring_sqe *sqe) | |
e977d6d3 | 4832 | { |
fbf23849 | 4833 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
e977d6d3 JA |
4834 | return -EINVAL; |
4835 | if (sqe->flags || sqe->ioprio || sqe->off || sqe->len || | |
4836 | sqe->cancel_flags) | |
4837 | return -EINVAL; | |
4838 | ||
fbf23849 JA |
4839 | req->cancel.addr = READ_ONCE(sqe->addr); |
4840 | return 0; | |
4841 | } | |
4842 | ||
014db007 | 4843 | static int io_async_cancel(struct io_kiocb *req) |
fbf23849 JA |
4844 | { |
4845 | struct io_ring_ctx *ctx = req->ctx; | |
fbf23849 | 4846 | |
014db007 | 4847 | io_async_find_and_cancel(ctx, req, req->cancel.addr, 0); |
5262f567 JA |
4848 | return 0; |
4849 | } | |
4850 | ||
05f3fb3c JA |
4851 | static int io_files_update_prep(struct io_kiocb *req, |
4852 | const struct io_uring_sqe *sqe) | |
4853 | { | |
4854 | if (sqe->flags || sqe->ioprio || sqe->rw_flags) | |
4855 | return -EINVAL; | |
4856 | ||
4857 | req->files_update.offset = READ_ONCE(sqe->off); | |
4858 | req->files_update.nr_args = READ_ONCE(sqe->len); | |
4859 | if (!req->files_update.nr_args) | |
4860 | return -EINVAL; | |
4861 | req->files_update.arg = READ_ONCE(sqe->addr); | |
4862 | return 0; | |
4863 | } | |
4864 | ||
4865 | static int io_files_update(struct io_kiocb *req, bool force_nonblock) | |
fbf23849 JA |
4866 | { |
4867 | struct io_ring_ctx *ctx = req->ctx; | |
05f3fb3c JA |
4868 | struct io_uring_files_update up; |
4869 | int ret; | |
fbf23849 | 4870 | |
f86cd20c | 4871 | if (force_nonblock) |
05f3fb3c | 4872 | return -EAGAIN; |
05f3fb3c JA |
4873 | |
4874 | up.offset = req->files_update.offset; | |
4875 | up.fds = req->files_update.arg; | |
4876 | ||
4877 | mutex_lock(&ctx->uring_lock); | |
4878 | ret = __io_sqe_files_update(ctx, &up, req->files_update.nr_args); | |
4879 | mutex_unlock(&ctx->uring_lock); | |
4880 | ||
4881 | if (ret < 0) | |
4882 | req_set_fail_links(req); | |
4883 | io_cqring_add_event(req, ret); | |
4884 | io_put_req(req); | |
5262f567 JA |
4885 | return 0; |
4886 | } | |
4887 | ||
3529d8c2 JA |
4888 | static int io_req_defer_prep(struct io_kiocb *req, |
4889 | const struct io_uring_sqe *sqe) | |
f67676d1 | 4890 | { |
e781573e | 4891 | ssize_t ret = 0; |
f67676d1 | 4892 | |
f1d96a8f PB |
4893 | if (!sqe) |
4894 | return 0; | |
4895 | ||
f86cd20c JA |
4896 | if (io_op_defs[req->opcode].file_table) { |
4897 | ret = io_grab_files(req); | |
4898 | if (unlikely(ret)) | |
4899 | return ret; | |
4900 | } | |
4901 | ||
cccf0ee8 JA |
4902 | io_req_work_grab_env(req, &io_op_defs[req->opcode]); |
4903 | ||
d625c6ee | 4904 | switch (req->opcode) { |
e781573e JA |
4905 | case IORING_OP_NOP: |
4906 | break; | |
f67676d1 JA |
4907 | case IORING_OP_READV: |
4908 | case IORING_OP_READ_FIXED: | |
3a6820f2 | 4909 | case IORING_OP_READ: |
3529d8c2 | 4910 | ret = io_read_prep(req, sqe, true); |
f67676d1 JA |
4911 | break; |
4912 | case IORING_OP_WRITEV: | |
4913 | case IORING_OP_WRITE_FIXED: | |
3a6820f2 | 4914 | case IORING_OP_WRITE: |
3529d8c2 | 4915 | ret = io_write_prep(req, sqe, true); |
f67676d1 | 4916 | break; |
0969e783 | 4917 | case IORING_OP_POLL_ADD: |
3529d8c2 | 4918 | ret = io_poll_add_prep(req, sqe); |
0969e783 JA |
4919 | break; |
4920 | case IORING_OP_POLL_REMOVE: | |
3529d8c2 | 4921 | ret = io_poll_remove_prep(req, sqe); |
0969e783 | 4922 | break; |
8ed8d3c3 | 4923 | case IORING_OP_FSYNC: |
3529d8c2 | 4924 | ret = io_prep_fsync(req, sqe); |
8ed8d3c3 JA |
4925 | break; |
4926 | case IORING_OP_SYNC_FILE_RANGE: | |
3529d8c2 | 4927 | ret = io_prep_sfr(req, sqe); |
8ed8d3c3 | 4928 | break; |
03b1230c | 4929 | case IORING_OP_SENDMSG: |
fddaface | 4930 | case IORING_OP_SEND: |
3529d8c2 | 4931 | ret = io_sendmsg_prep(req, sqe); |
03b1230c JA |
4932 | break; |
4933 | case IORING_OP_RECVMSG: | |
fddaface | 4934 | case IORING_OP_RECV: |
3529d8c2 | 4935 | ret = io_recvmsg_prep(req, sqe); |
03b1230c | 4936 | break; |
f499a021 | 4937 | case IORING_OP_CONNECT: |
3529d8c2 | 4938 | ret = io_connect_prep(req, sqe); |
f499a021 | 4939 | break; |
2d28390a | 4940 | case IORING_OP_TIMEOUT: |
3529d8c2 | 4941 | ret = io_timeout_prep(req, sqe, false); |
b7bb4f7d | 4942 | break; |
b29472ee | 4943 | case IORING_OP_TIMEOUT_REMOVE: |
3529d8c2 | 4944 | ret = io_timeout_remove_prep(req, sqe); |
b29472ee | 4945 | break; |
fbf23849 | 4946 | case IORING_OP_ASYNC_CANCEL: |
3529d8c2 | 4947 | ret = io_async_cancel_prep(req, sqe); |
fbf23849 | 4948 | break; |
2d28390a | 4949 | case IORING_OP_LINK_TIMEOUT: |
3529d8c2 | 4950 | ret = io_timeout_prep(req, sqe, true); |
b7bb4f7d | 4951 | break; |
8ed8d3c3 | 4952 | case IORING_OP_ACCEPT: |
3529d8c2 | 4953 | ret = io_accept_prep(req, sqe); |
8ed8d3c3 | 4954 | break; |
d63d1b5e JA |
4955 | case IORING_OP_FALLOCATE: |
4956 | ret = io_fallocate_prep(req, sqe); | |
4957 | break; | |
15b71abe JA |
4958 | case IORING_OP_OPENAT: |
4959 | ret = io_openat_prep(req, sqe); | |
4960 | break; | |
b5dba59e JA |
4961 | case IORING_OP_CLOSE: |
4962 | ret = io_close_prep(req, sqe); | |
4963 | break; | |
05f3fb3c JA |
4964 | case IORING_OP_FILES_UPDATE: |
4965 | ret = io_files_update_prep(req, sqe); | |
4966 | break; | |
eddc7ef5 JA |
4967 | case IORING_OP_STATX: |
4968 | ret = io_statx_prep(req, sqe); | |
4969 | break; | |
4840e418 JA |
4970 | case IORING_OP_FADVISE: |
4971 | ret = io_fadvise_prep(req, sqe); | |
4972 | break; | |
c1ca757b JA |
4973 | case IORING_OP_MADVISE: |
4974 | ret = io_madvise_prep(req, sqe); | |
4975 | break; | |
cebdb986 JA |
4976 | case IORING_OP_OPENAT2: |
4977 | ret = io_openat2_prep(req, sqe); | |
4978 | break; | |
3e4827b0 JA |
4979 | case IORING_OP_EPOLL_CTL: |
4980 | ret = io_epoll_ctl_prep(req, sqe); | |
4981 | break; | |
7d67af2c PB |
4982 | case IORING_OP_SPLICE: |
4983 | ret = io_splice_prep(req, sqe); | |
4984 | break; | |
ddf0322d JA |
4985 | case IORING_OP_PROVIDE_BUFFERS: |
4986 | ret = io_provide_buffers_prep(req, sqe); | |
4987 | break; | |
067524e9 JA |
4988 | case IORING_OP_REMOVE_BUFFERS: |
4989 | ret = io_remove_buffers_prep(req, sqe); | |
4990 | break; | |
f67676d1 | 4991 | default: |
e781573e JA |
4992 | printk_once(KERN_WARNING "io_uring: unhandled opcode %d\n", |
4993 | req->opcode); | |
4994 | ret = -EINVAL; | |
b7bb4f7d | 4995 | break; |
f67676d1 JA |
4996 | } |
4997 | ||
b7bb4f7d | 4998 | return ret; |
f67676d1 JA |
4999 | } |
5000 | ||
3529d8c2 | 5001 | static int io_req_defer(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
de0617e4 | 5002 | { |
a197f664 | 5003 | struct io_ring_ctx *ctx = req->ctx; |
f67676d1 | 5004 | int ret; |
de0617e4 | 5005 | |
9d858b21 | 5006 | /* Still need defer if there is pending req in defer list. */ |
4ee36314 | 5007 | if (!req_need_defer(req) && list_empty_careful(&ctx->defer_list)) |
de0617e4 JA |
5008 | return 0; |
5009 | ||
3529d8c2 | 5010 | if (!req->io && io_alloc_async_ctx(req)) |
de0617e4 JA |
5011 | return -EAGAIN; |
5012 | ||
3529d8c2 | 5013 | ret = io_req_defer_prep(req, sqe); |
b7bb4f7d | 5014 | if (ret < 0) |
2d28390a | 5015 | return ret; |
2d28390a | 5016 | |
de0617e4 | 5017 | spin_lock_irq(&ctx->completion_lock); |
9d858b21 | 5018 | if (!req_need_defer(req) && list_empty(&ctx->defer_list)) { |
de0617e4 | 5019 | spin_unlock_irq(&ctx->completion_lock); |
de0617e4 JA |
5020 | return 0; |
5021 | } | |
5022 | ||
915967f6 | 5023 | trace_io_uring_defer(ctx, req, req->user_data); |
de0617e4 JA |
5024 | list_add_tail(&req->list, &ctx->defer_list); |
5025 | spin_unlock_irq(&ctx->completion_lock); | |
5026 | return -EIOCBQUEUED; | |
5027 | } | |
5028 | ||
99bc4c38 PB |
5029 | static void io_cleanup_req(struct io_kiocb *req) |
5030 | { | |
5031 | struct io_async_ctx *io = req->io; | |
5032 | ||
5033 | switch (req->opcode) { | |
5034 | case IORING_OP_READV: | |
5035 | case IORING_OP_READ_FIXED: | |
5036 | case IORING_OP_READ: | |
bcda7baa JA |
5037 | if (req->flags & REQ_F_BUFFER_SELECTED) |
5038 | kfree((void *)(unsigned long)req->rw.addr); | |
5039 | /* fallthrough */ | |
99bc4c38 PB |
5040 | case IORING_OP_WRITEV: |
5041 | case IORING_OP_WRITE_FIXED: | |
5042 | case IORING_OP_WRITE: | |
5043 | if (io->rw.iov != io->rw.fast_iov) | |
5044 | kfree(io->rw.iov); | |
5045 | break; | |
99bc4c38 | 5046 | case IORING_OP_RECVMSG: |
52de1fe1 JA |
5047 | if (req->flags & REQ_F_BUFFER_SELECTED) |
5048 | kfree(req->sr_msg.kbuf); | |
5049 | /* fallthrough */ | |
5050 | case IORING_OP_SENDMSG: | |
99bc4c38 PB |
5051 | if (io->msg.iov != io->msg.fast_iov) |
5052 | kfree(io->msg.iov); | |
5053 | break; | |
bcda7baa JA |
5054 | case IORING_OP_RECV: |
5055 | if (req->flags & REQ_F_BUFFER_SELECTED) | |
5056 | kfree(req->sr_msg.kbuf); | |
5057 | break; | |
8fef80bf PB |
5058 | case IORING_OP_OPENAT: |
5059 | case IORING_OP_OPENAT2: | |
5060 | case IORING_OP_STATX: | |
5061 | putname(req->open.filename); | |
5062 | break; | |
7d67af2c PB |
5063 | case IORING_OP_SPLICE: |
5064 | io_put_file(req, req->splice.file_in, | |
5065 | (req->splice.flags & SPLICE_F_FD_IN_FIXED)); | |
5066 | break; | |
99bc4c38 PB |
5067 | } |
5068 | ||
5069 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
5070 | } | |
5071 | ||
3529d8c2 | 5072 | static int io_issue_sqe(struct io_kiocb *req, const struct io_uring_sqe *sqe, |
014db007 | 5073 | bool force_nonblock) |
2b188cc1 | 5074 | { |
a197f664 | 5075 | struct io_ring_ctx *ctx = req->ctx; |
d625c6ee | 5076 | int ret; |
2b188cc1 | 5077 | |
d625c6ee | 5078 | switch (req->opcode) { |
2b188cc1 | 5079 | case IORING_OP_NOP: |
78e19bbe | 5080 | ret = io_nop(req); |
2b188cc1 JA |
5081 | break; |
5082 | case IORING_OP_READV: | |
edafccee | 5083 | case IORING_OP_READ_FIXED: |
3a6820f2 | 5084 | case IORING_OP_READ: |
3529d8c2 JA |
5085 | if (sqe) { |
5086 | ret = io_read_prep(req, sqe, force_nonblock); | |
5087 | if (ret < 0) | |
5088 | break; | |
5089 | } | |
014db007 | 5090 | ret = io_read(req, force_nonblock); |
edafccee | 5091 | break; |
3529d8c2 | 5092 | case IORING_OP_WRITEV: |
edafccee | 5093 | case IORING_OP_WRITE_FIXED: |
3a6820f2 | 5094 | case IORING_OP_WRITE: |
3529d8c2 JA |
5095 | if (sqe) { |
5096 | ret = io_write_prep(req, sqe, force_nonblock); | |
5097 | if (ret < 0) | |
5098 | break; | |
5099 | } | |
014db007 | 5100 | ret = io_write(req, force_nonblock); |
2b188cc1 | 5101 | break; |
c992fe29 | 5102 | case IORING_OP_FSYNC: |
3529d8c2 JA |
5103 | if (sqe) { |
5104 | ret = io_prep_fsync(req, sqe); | |
5105 | if (ret < 0) | |
5106 | break; | |
5107 | } | |
014db007 | 5108 | ret = io_fsync(req, force_nonblock); |
c992fe29 | 5109 | break; |
221c5eb2 | 5110 | case IORING_OP_POLL_ADD: |
3529d8c2 JA |
5111 | if (sqe) { |
5112 | ret = io_poll_add_prep(req, sqe); | |
5113 | if (ret) | |
5114 | break; | |
5115 | } | |
014db007 | 5116 | ret = io_poll_add(req); |
221c5eb2 JA |
5117 | break; |
5118 | case IORING_OP_POLL_REMOVE: | |
3529d8c2 JA |
5119 | if (sqe) { |
5120 | ret = io_poll_remove_prep(req, sqe); | |
5121 | if (ret < 0) | |
5122 | break; | |
5123 | } | |
fc4df999 | 5124 | ret = io_poll_remove(req); |
221c5eb2 | 5125 | break; |
5d17b4a4 | 5126 | case IORING_OP_SYNC_FILE_RANGE: |
3529d8c2 JA |
5127 | if (sqe) { |
5128 | ret = io_prep_sfr(req, sqe); | |
5129 | if (ret < 0) | |
5130 | break; | |
5131 | } | |
014db007 | 5132 | ret = io_sync_file_range(req, force_nonblock); |
5d17b4a4 | 5133 | break; |
0fa03c62 | 5134 | case IORING_OP_SENDMSG: |
fddaface | 5135 | case IORING_OP_SEND: |
3529d8c2 JA |
5136 | if (sqe) { |
5137 | ret = io_sendmsg_prep(req, sqe); | |
5138 | if (ret < 0) | |
5139 | break; | |
5140 | } | |
fddaface | 5141 | if (req->opcode == IORING_OP_SENDMSG) |
014db007 | 5142 | ret = io_sendmsg(req, force_nonblock); |
fddaface | 5143 | else |
014db007 | 5144 | ret = io_send(req, force_nonblock); |
0fa03c62 | 5145 | break; |
aa1fa28f | 5146 | case IORING_OP_RECVMSG: |
fddaface | 5147 | case IORING_OP_RECV: |
3529d8c2 JA |
5148 | if (sqe) { |
5149 | ret = io_recvmsg_prep(req, sqe); | |
5150 | if (ret) | |
5151 | break; | |
5152 | } | |
fddaface | 5153 | if (req->opcode == IORING_OP_RECVMSG) |
014db007 | 5154 | ret = io_recvmsg(req, force_nonblock); |
fddaface | 5155 | else |
014db007 | 5156 | ret = io_recv(req, force_nonblock); |
aa1fa28f | 5157 | break; |
5262f567 | 5158 | case IORING_OP_TIMEOUT: |
3529d8c2 JA |
5159 | if (sqe) { |
5160 | ret = io_timeout_prep(req, sqe, false); | |
5161 | if (ret) | |
5162 | break; | |
5163 | } | |
fc4df999 | 5164 | ret = io_timeout(req); |
5262f567 | 5165 | break; |
11365043 | 5166 | case IORING_OP_TIMEOUT_REMOVE: |
3529d8c2 JA |
5167 | if (sqe) { |
5168 | ret = io_timeout_remove_prep(req, sqe); | |
5169 | if (ret) | |
5170 | break; | |
5171 | } | |
fc4df999 | 5172 | ret = io_timeout_remove(req); |
11365043 | 5173 | break; |
17f2fe35 | 5174 | case IORING_OP_ACCEPT: |
3529d8c2 JA |
5175 | if (sqe) { |
5176 | ret = io_accept_prep(req, sqe); | |
5177 | if (ret) | |
5178 | break; | |
5179 | } | |
014db007 | 5180 | ret = io_accept(req, force_nonblock); |
17f2fe35 | 5181 | break; |
f8e85cf2 | 5182 | case IORING_OP_CONNECT: |
3529d8c2 JA |
5183 | if (sqe) { |
5184 | ret = io_connect_prep(req, sqe); | |
5185 | if (ret) | |
5186 | break; | |
5187 | } | |
014db007 | 5188 | ret = io_connect(req, force_nonblock); |
f8e85cf2 | 5189 | break; |
62755e35 | 5190 | case IORING_OP_ASYNC_CANCEL: |
3529d8c2 JA |
5191 | if (sqe) { |
5192 | ret = io_async_cancel_prep(req, sqe); | |
5193 | if (ret) | |
5194 | break; | |
5195 | } | |
014db007 | 5196 | ret = io_async_cancel(req); |
62755e35 | 5197 | break; |
d63d1b5e JA |
5198 | case IORING_OP_FALLOCATE: |
5199 | if (sqe) { | |
5200 | ret = io_fallocate_prep(req, sqe); | |
5201 | if (ret) | |
5202 | break; | |
5203 | } | |
014db007 | 5204 | ret = io_fallocate(req, force_nonblock); |
d63d1b5e | 5205 | break; |
15b71abe JA |
5206 | case IORING_OP_OPENAT: |
5207 | if (sqe) { | |
5208 | ret = io_openat_prep(req, sqe); | |
5209 | if (ret) | |
5210 | break; | |
5211 | } | |
014db007 | 5212 | ret = io_openat(req, force_nonblock); |
15b71abe | 5213 | break; |
b5dba59e JA |
5214 | case IORING_OP_CLOSE: |
5215 | if (sqe) { | |
5216 | ret = io_close_prep(req, sqe); | |
5217 | if (ret) | |
5218 | break; | |
5219 | } | |
014db007 | 5220 | ret = io_close(req, force_nonblock); |
b5dba59e | 5221 | break; |
05f3fb3c JA |
5222 | case IORING_OP_FILES_UPDATE: |
5223 | if (sqe) { | |
5224 | ret = io_files_update_prep(req, sqe); | |
5225 | if (ret) | |
5226 | break; | |
5227 | } | |
5228 | ret = io_files_update(req, force_nonblock); | |
5229 | break; | |
eddc7ef5 JA |
5230 | case IORING_OP_STATX: |
5231 | if (sqe) { | |
5232 | ret = io_statx_prep(req, sqe); | |
5233 | if (ret) | |
5234 | break; | |
5235 | } | |
014db007 | 5236 | ret = io_statx(req, force_nonblock); |
eddc7ef5 | 5237 | break; |
4840e418 JA |
5238 | case IORING_OP_FADVISE: |
5239 | if (sqe) { | |
5240 | ret = io_fadvise_prep(req, sqe); | |
5241 | if (ret) | |
5242 | break; | |
5243 | } | |
014db007 | 5244 | ret = io_fadvise(req, force_nonblock); |
4840e418 | 5245 | break; |
c1ca757b JA |
5246 | case IORING_OP_MADVISE: |
5247 | if (sqe) { | |
5248 | ret = io_madvise_prep(req, sqe); | |
5249 | if (ret) | |
5250 | break; | |
5251 | } | |
014db007 | 5252 | ret = io_madvise(req, force_nonblock); |
c1ca757b | 5253 | break; |
cebdb986 JA |
5254 | case IORING_OP_OPENAT2: |
5255 | if (sqe) { | |
5256 | ret = io_openat2_prep(req, sqe); | |
5257 | if (ret) | |
5258 | break; | |
5259 | } | |
014db007 | 5260 | ret = io_openat2(req, force_nonblock); |
cebdb986 | 5261 | break; |
3e4827b0 JA |
5262 | case IORING_OP_EPOLL_CTL: |
5263 | if (sqe) { | |
5264 | ret = io_epoll_ctl_prep(req, sqe); | |
5265 | if (ret) | |
5266 | break; | |
5267 | } | |
014db007 | 5268 | ret = io_epoll_ctl(req, force_nonblock); |
3e4827b0 | 5269 | break; |
7d67af2c PB |
5270 | case IORING_OP_SPLICE: |
5271 | if (sqe) { | |
5272 | ret = io_splice_prep(req, sqe); | |
5273 | if (ret < 0) | |
5274 | break; | |
5275 | } | |
014db007 | 5276 | ret = io_splice(req, force_nonblock); |
7d67af2c | 5277 | break; |
ddf0322d JA |
5278 | case IORING_OP_PROVIDE_BUFFERS: |
5279 | if (sqe) { | |
5280 | ret = io_provide_buffers_prep(req, sqe); | |
5281 | if (ret) | |
5282 | break; | |
5283 | } | |
5284 | ret = io_provide_buffers(req, force_nonblock); | |
5285 | break; | |
067524e9 JA |
5286 | case IORING_OP_REMOVE_BUFFERS: |
5287 | if (sqe) { | |
5288 | ret = io_remove_buffers_prep(req, sqe); | |
5289 | if (ret) | |
5290 | break; | |
5291 | } | |
5292 | ret = io_remove_buffers(req, force_nonblock); | |
3e4827b0 | 5293 | break; |
2b188cc1 JA |
5294 | default: |
5295 | ret = -EINVAL; | |
5296 | break; | |
5297 | } | |
5298 | ||
def596e9 JA |
5299 | if (ret) |
5300 | return ret; | |
5301 | ||
5302 | if (ctx->flags & IORING_SETUP_IOPOLL) { | |
11ba820b JA |
5303 | const bool in_async = io_wq_current_is_worker(); |
5304 | ||
9e645e11 | 5305 | if (req->result == -EAGAIN) |
def596e9 JA |
5306 | return -EAGAIN; |
5307 | ||
11ba820b JA |
5308 | /* workqueue context doesn't hold uring_lock, grab it now */ |
5309 | if (in_async) | |
5310 | mutex_lock(&ctx->uring_lock); | |
5311 | ||
def596e9 | 5312 | io_iopoll_req_issued(req); |
11ba820b JA |
5313 | |
5314 | if (in_async) | |
5315 | mutex_unlock(&ctx->uring_lock); | |
def596e9 JA |
5316 | } |
5317 | ||
5318 | return 0; | |
2b188cc1 JA |
5319 | } |
5320 | ||
561fb04a | 5321 | static void io_wq_submit_work(struct io_wq_work **workptr) |
2b188cc1 | 5322 | { |
561fb04a | 5323 | struct io_wq_work *work = *workptr; |
2b188cc1 | 5324 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); |
561fb04a | 5325 | int ret = 0; |
2b188cc1 | 5326 | |
0c9d5ccd JA |
5327 | /* if NO_CANCEL is set, we must still run the work */ |
5328 | if ((work->flags & (IO_WQ_WORK_CANCEL|IO_WQ_WORK_NO_CANCEL)) == | |
5329 | IO_WQ_WORK_CANCEL) { | |
561fb04a | 5330 | ret = -ECANCELED; |
0c9d5ccd | 5331 | } |
31b51510 | 5332 | |
561fb04a | 5333 | if (!ret) { |
561fb04a | 5334 | do { |
014db007 | 5335 | ret = io_issue_sqe(req, NULL, false); |
561fb04a JA |
5336 | /* |
5337 | * We can get EAGAIN for polled IO even though we're | |
5338 | * forcing a sync submission from here, since we can't | |
5339 | * wait for request slots on the block side. | |
5340 | */ | |
5341 | if (ret != -EAGAIN) | |
5342 | break; | |
5343 | cond_resched(); | |
5344 | } while (1); | |
5345 | } | |
31b51510 | 5346 | |
561fb04a | 5347 | if (ret) { |
4e88d6e7 | 5348 | req_set_fail_links(req); |
78e19bbe | 5349 | io_cqring_add_event(req, ret); |
817869d2 | 5350 | io_put_req(req); |
edafccee | 5351 | } |
2b188cc1 | 5352 | |
e9fd9396 | 5353 | io_steal_work(req, workptr); |
2b188cc1 JA |
5354 | } |
5355 | ||
65e19f54 JA |
5356 | static inline struct file *io_file_from_index(struct io_ring_ctx *ctx, |
5357 | int index) | |
5358 | { | |
5359 | struct fixed_file_table *table; | |
5360 | ||
05f3fb3c | 5361 | table = &ctx->file_data->table[index >> IORING_FILE_TABLE_SHIFT]; |
84695089 | 5362 | return table->files[index & IORING_FILE_TABLE_MASK]; |
65e19f54 JA |
5363 | } |
5364 | ||
8da11c19 PB |
5365 | static int io_file_get(struct io_submit_state *state, struct io_kiocb *req, |
5366 | int fd, struct file **out_file, bool fixed) | |
09bb8394 | 5367 | { |
a197f664 | 5368 | struct io_ring_ctx *ctx = req->ctx; |
8da11c19 | 5369 | struct file *file; |
09bb8394 | 5370 | |
8da11c19 | 5371 | if (fixed) { |
05f3fb3c | 5372 | if (unlikely(!ctx->file_data || |
09bb8394 JA |
5373 | (unsigned) fd >= ctx->nr_user_files)) |
5374 | return -EBADF; | |
b7620121 | 5375 | fd = array_index_nospec(fd, ctx->nr_user_files); |
8da11c19 PB |
5376 | file = io_file_from_index(ctx, fd); |
5377 | if (!file) | |
08a45173 | 5378 | return -EBADF; |
05589553 XW |
5379 | req->fixed_file_refs = ctx->file_data->cur_refs; |
5380 | percpu_ref_get(req->fixed_file_refs); | |
09bb8394 | 5381 | } else { |
c826bd7a | 5382 | trace_io_uring_file_get(ctx, fd); |
8da11c19 PB |
5383 | file = __io_file_get(state, fd); |
5384 | if (unlikely(!file)) | |
09bb8394 JA |
5385 | return -EBADF; |
5386 | } | |
5387 | ||
8da11c19 | 5388 | *out_file = file; |
09bb8394 JA |
5389 | return 0; |
5390 | } | |
5391 | ||
8da11c19 | 5392 | static int io_req_set_file(struct io_submit_state *state, struct io_kiocb *req, |
63ff8223 | 5393 | int fd) |
8da11c19 | 5394 | { |
8da11c19 PB |
5395 | bool fixed; |
5396 | ||
63ff8223 | 5397 | fixed = (req->flags & REQ_F_FIXED_FILE) != 0; |
8da11c19 PB |
5398 | if (unlikely(!fixed && req->needs_fixed_file)) |
5399 | return -EBADF; | |
5400 | ||
5401 | return io_file_get(state, req, fd, &req->file, fixed); | |
5402 | } | |
5403 | ||
a197f664 | 5404 | static int io_grab_files(struct io_kiocb *req) |
fcb323cc JA |
5405 | { |
5406 | int ret = -EBADF; | |
a197f664 | 5407 | struct io_ring_ctx *ctx = req->ctx; |
fcb323cc | 5408 | |
5b0bbee4 | 5409 | if (req->work.files || (req->flags & REQ_F_NO_FILE_TABLE)) |
f86cd20c | 5410 | return 0; |
b14cca0c | 5411 | if (!ctx->ring_file) |
b5dba59e JA |
5412 | return -EBADF; |
5413 | ||
fcb323cc JA |
5414 | rcu_read_lock(); |
5415 | spin_lock_irq(&ctx->inflight_lock); | |
5416 | /* | |
5417 | * We use the f_ops->flush() handler to ensure that we can flush | |
5418 | * out work accessing these files if the fd is closed. Check if | |
5419 | * the fd has changed since we started down this path, and disallow | |
5420 | * this operation if it has. | |
5421 | */ | |
b14cca0c | 5422 | if (fcheck(ctx->ring_fd) == ctx->ring_file) { |
fcb323cc JA |
5423 | list_add(&req->inflight_entry, &ctx->inflight_list); |
5424 | req->flags |= REQ_F_INFLIGHT; | |
5425 | req->work.files = current->files; | |
5426 | ret = 0; | |
5427 | } | |
5428 | spin_unlock_irq(&ctx->inflight_lock); | |
5429 | rcu_read_unlock(); | |
5430 | ||
5431 | return ret; | |
5432 | } | |
5433 | ||
2665abfd | 5434 | static enum hrtimer_restart io_link_timeout_fn(struct hrtimer *timer) |
2b188cc1 | 5435 | { |
ad8a48ac JA |
5436 | struct io_timeout_data *data = container_of(timer, |
5437 | struct io_timeout_data, timer); | |
5438 | struct io_kiocb *req = data->req; | |
2665abfd JA |
5439 | struct io_ring_ctx *ctx = req->ctx; |
5440 | struct io_kiocb *prev = NULL; | |
5441 | unsigned long flags; | |
2665abfd JA |
5442 | |
5443 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
5444 | ||
5445 | /* | |
5446 | * We don't expect the list to be empty, that will only happen if we | |
5447 | * race with the completion of the linked work. | |
5448 | */ | |
4493233e PB |
5449 | if (!list_empty(&req->link_list)) { |
5450 | prev = list_entry(req->link_list.prev, struct io_kiocb, | |
5451 | link_list); | |
5d960724 | 5452 | if (refcount_inc_not_zero(&prev->refs)) { |
4493233e | 5453 | list_del_init(&req->link_list); |
5d960724 JA |
5454 | prev->flags &= ~REQ_F_LINK_TIMEOUT; |
5455 | } else | |
76a46e06 | 5456 | prev = NULL; |
2665abfd JA |
5457 | } |
5458 | ||
5459 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
5460 | ||
5461 | if (prev) { | |
4e88d6e7 | 5462 | req_set_fail_links(prev); |
014db007 | 5463 | io_async_find_and_cancel(ctx, req, prev->user_data, -ETIME); |
76a46e06 | 5464 | io_put_req(prev); |
47f46768 JA |
5465 | } else { |
5466 | io_cqring_add_event(req, -ETIME); | |
5467 | io_put_req(req); | |
2665abfd | 5468 | } |
2665abfd JA |
5469 | return HRTIMER_NORESTART; |
5470 | } | |
5471 | ||
ad8a48ac | 5472 | static void io_queue_linked_timeout(struct io_kiocb *req) |
2665abfd | 5473 | { |
76a46e06 | 5474 | struct io_ring_ctx *ctx = req->ctx; |
2665abfd | 5475 | |
76a46e06 JA |
5476 | /* |
5477 | * If the list is now empty, then our linked request finished before | |
5478 | * we got a chance to setup the timer | |
5479 | */ | |
5480 | spin_lock_irq(&ctx->completion_lock); | |
4493233e | 5481 | if (!list_empty(&req->link_list)) { |
2d28390a | 5482 | struct io_timeout_data *data = &req->io->timeout; |
94ae5e77 | 5483 | |
ad8a48ac JA |
5484 | data->timer.function = io_link_timeout_fn; |
5485 | hrtimer_start(&data->timer, timespec64_to_ktime(data->ts), | |
5486 | data->mode); | |
2665abfd | 5487 | } |
76a46e06 | 5488 | spin_unlock_irq(&ctx->completion_lock); |
2665abfd | 5489 | |
2665abfd | 5490 | /* drop submission reference */ |
76a46e06 JA |
5491 | io_put_req(req); |
5492 | } | |
2665abfd | 5493 | |
ad8a48ac | 5494 | static struct io_kiocb *io_prep_linked_timeout(struct io_kiocb *req) |
2665abfd JA |
5495 | { |
5496 | struct io_kiocb *nxt; | |
5497 | ||
dea3b49c | 5498 | if (!(req->flags & REQ_F_LINK_HEAD)) |
2665abfd | 5499 | return NULL; |
d7718a9d JA |
5500 | /* for polled retry, if flag is set, we already went through here */ |
5501 | if (req->flags & REQ_F_POLLED) | |
5502 | return NULL; | |
2665abfd | 5503 | |
4493233e PB |
5504 | nxt = list_first_entry_or_null(&req->link_list, struct io_kiocb, |
5505 | link_list); | |
d625c6ee | 5506 | if (!nxt || nxt->opcode != IORING_OP_LINK_TIMEOUT) |
76a46e06 | 5507 | return NULL; |
2665abfd | 5508 | |
76a46e06 | 5509 | req->flags |= REQ_F_LINK_TIMEOUT; |
76a46e06 | 5510 | return nxt; |
2665abfd JA |
5511 | } |
5512 | ||
3529d8c2 | 5513 | static void __io_queue_sqe(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
2b188cc1 | 5514 | { |
4a0a7a18 | 5515 | struct io_kiocb *linked_timeout; |
4bc4494e | 5516 | struct io_kiocb *nxt; |
193155c8 | 5517 | const struct cred *old_creds = NULL; |
e0c5c576 | 5518 | int ret; |
2b188cc1 | 5519 | |
4a0a7a18 JA |
5520 | again: |
5521 | linked_timeout = io_prep_linked_timeout(req); | |
5522 | ||
193155c8 JA |
5523 | if (req->work.creds && req->work.creds != current_cred()) { |
5524 | if (old_creds) | |
5525 | revert_creds(old_creds); | |
5526 | if (old_creds == req->work.creds) | |
5527 | old_creds = NULL; /* restored original creds */ | |
5528 | else | |
5529 | old_creds = override_creds(req->work.creds); | |
5530 | } | |
5531 | ||
014db007 | 5532 | ret = io_issue_sqe(req, sqe, true); |
491381ce JA |
5533 | |
5534 | /* | |
5535 | * We async punt it if the file wasn't marked NOWAIT, or if the file | |
5536 | * doesn't support non-blocking read/write attempts | |
5537 | */ | |
5538 | if (ret == -EAGAIN && (!(req->flags & REQ_F_NOWAIT) || | |
5539 | (req->flags & REQ_F_MUST_PUNT))) { | |
d7718a9d JA |
5540 | if (io_arm_poll_handler(req)) { |
5541 | if (linked_timeout) | |
5542 | io_queue_linked_timeout(linked_timeout); | |
4bc4494e | 5543 | goto exit; |
d7718a9d | 5544 | } |
86a761f8 | 5545 | punt: |
f86cd20c | 5546 | if (io_op_defs[req->opcode].file_table) { |
bbad27b2 PB |
5547 | ret = io_grab_files(req); |
5548 | if (ret) | |
5549 | goto err; | |
2b188cc1 | 5550 | } |
bbad27b2 PB |
5551 | |
5552 | /* | |
5553 | * Queued up for async execution, worker will release | |
5554 | * submit reference when the iocb is actually submitted. | |
5555 | */ | |
5556 | io_queue_async_work(req); | |
4bc4494e | 5557 | goto exit; |
2b188cc1 | 5558 | } |
e65ef56d | 5559 | |
fcb323cc | 5560 | err: |
4bc4494e | 5561 | nxt = NULL; |
76a46e06 | 5562 | /* drop submission reference */ |
2a44f467 | 5563 | io_put_req_find_next(req, &nxt); |
e65ef56d | 5564 | |
f9bd67f6 | 5565 | if (linked_timeout) { |
76a46e06 | 5566 | if (!ret) |
f9bd67f6 | 5567 | io_queue_linked_timeout(linked_timeout); |
76a46e06 | 5568 | else |
f9bd67f6 | 5569 | io_put_req(linked_timeout); |
76a46e06 JA |
5570 | } |
5571 | ||
e65ef56d | 5572 | /* and drop final reference, if we failed */ |
9e645e11 | 5573 | if (ret) { |
78e19bbe | 5574 | io_cqring_add_event(req, ret); |
4e88d6e7 | 5575 | req_set_fail_links(req); |
e65ef56d | 5576 | io_put_req(req); |
9e645e11 | 5577 | } |
4a0a7a18 JA |
5578 | if (nxt) { |
5579 | req = nxt; | |
86a761f8 PB |
5580 | |
5581 | if (req->flags & REQ_F_FORCE_ASYNC) | |
5582 | goto punt; | |
4a0a7a18 JA |
5583 | goto again; |
5584 | } | |
4bc4494e | 5585 | exit: |
193155c8 JA |
5586 | if (old_creds) |
5587 | revert_creds(old_creds); | |
2b188cc1 JA |
5588 | } |
5589 | ||
3529d8c2 | 5590 | static void io_queue_sqe(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
4fe2c963 JL |
5591 | { |
5592 | int ret; | |
5593 | ||
3529d8c2 | 5594 | ret = io_req_defer(req, sqe); |
4fe2c963 JL |
5595 | if (ret) { |
5596 | if (ret != -EIOCBQUEUED) { | |
1118591a | 5597 | fail_req: |
78e19bbe | 5598 | io_cqring_add_event(req, ret); |
4e88d6e7 | 5599 | req_set_fail_links(req); |
78e19bbe | 5600 | io_double_put_req(req); |
4fe2c963 | 5601 | } |
2550878f | 5602 | } else if (req->flags & REQ_F_FORCE_ASYNC) { |
1118591a PB |
5603 | ret = io_req_defer_prep(req, sqe); |
5604 | if (unlikely(ret < 0)) | |
5605 | goto fail_req; | |
ce35a47a JA |
5606 | /* |
5607 | * Never try inline submit of IOSQE_ASYNC is set, go straight | |
5608 | * to async execution. | |
5609 | */ | |
5610 | req->work.flags |= IO_WQ_WORK_CONCURRENT; | |
5611 | io_queue_async_work(req); | |
5612 | } else { | |
3529d8c2 | 5613 | __io_queue_sqe(req, sqe); |
ce35a47a | 5614 | } |
4fe2c963 JL |
5615 | } |
5616 | ||
1b4a51b6 | 5617 | static inline void io_queue_link_head(struct io_kiocb *req) |
4fe2c963 | 5618 | { |
94ae5e77 | 5619 | if (unlikely(req->flags & REQ_F_FAIL_LINK)) { |
1b4a51b6 PB |
5620 | io_cqring_add_event(req, -ECANCELED); |
5621 | io_double_put_req(req); | |
5622 | } else | |
3529d8c2 | 5623 | io_queue_sqe(req, NULL); |
4fe2c963 JL |
5624 | } |
5625 | ||
1d4240cc | 5626 | static int io_submit_sqe(struct io_kiocb *req, const struct io_uring_sqe *sqe, |
7d01bd74 | 5627 | struct io_kiocb **link) |
9e645e11 | 5628 | { |
a197f664 | 5629 | struct io_ring_ctx *ctx = req->ctx; |
ef4ff581 | 5630 | int ret; |
9e645e11 | 5631 | |
9e645e11 JA |
5632 | /* |
5633 | * If we already have a head request, queue this one for async | |
5634 | * submittal once the head completes. If we don't have a head but | |
5635 | * IOSQE_IO_LINK is set in the sqe, start a new head. This one will be | |
5636 | * submitted sync once the chain is complete. If none of those | |
5637 | * conditions are true (normal request), then just queue it. | |
5638 | */ | |
5639 | if (*link) { | |
9d76377f | 5640 | struct io_kiocb *head = *link; |
4e88d6e7 | 5641 | |
8cdf2193 PB |
5642 | /* |
5643 | * Taking sequential execution of a link, draining both sides | |
5644 | * of the link also fullfils IOSQE_IO_DRAIN semantics for all | |
5645 | * requests in the link. So, it drains the head and the | |
5646 | * next after the link request. The last one is done via | |
5647 | * drain_next flag to persist the effect across calls. | |
5648 | */ | |
ef4ff581 | 5649 | if (req->flags & REQ_F_IO_DRAIN) { |
711be031 PB |
5650 | head->flags |= REQ_F_IO_DRAIN; |
5651 | ctx->drain_next = 1; | |
5652 | } | |
1d4240cc PB |
5653 | if (io_alloc_async_ctx(req)) |
5654 | return -EAGAIN; | |
9e645e11 | 5655 | |
3529d8c2 | 5656 | ret = io_req_defer_prep(req, sqe); |
2d28390a | 5657 | if (ret) { |
4e88d6e7 | 5658 | /* fail even hard links since we don't submit */ |
9d76377f | 5659 | head->flags |= REQ_F_FAIL_LINK; |
1d4240cc | 5660 | return ret; |
2d28390a | 5661 | } |
9d76377f PB |
5662 | trace_io_uring_link(ctx, req, head); |
5663 | list_add_tail(&req->link_list, &head->link_list); | |
32fe525b PB |
5664 | |
5665 | /* last request of a link, enqueue the link */ | |
ef4ff581 | 5666 | if (!(req->flags & (REQ_F_LINK | REQ_F_HARDLINK))) { |
32fe525b PB |
5667 | io_queue_link_head(head); |
5668 | *link = NULL; | |
5669 | } | |
9e645e11 | 5670 | } else { |
711be031 PB |
5671 | if (unlikely(ctx->drain_next)) { |
5672 | req->flags |= REQ_F_IO_DRAIN; | |
ef4ff581 | 5673 | ctx->drain_next = 0; |
711be031 | 5674 | } |
ef4ff581 | 5675 | if (req->flags & (REQ_F_LINK | REQ_F_HARDLINK)) { |
dea3b49c | 5676 | req->flags |= REQ_F_LINK_HEAD; |
711be031 | 5677 | INIT_LIST_HEAD(&req->link_list); |
f1d96a8f | 5678 | |
1d4240cc PB |
5679 | if (io_alloc_async_ctx(req)) |
5680 | return -EAGAIN; | |
5681 | ||
711be031 PB |
5682 | ret = io_req_defer_prep(req, sqe); |
5683 | if (ret) | |
5684 | req->flags |= REQ_F_FAIL_LINK; | |
5685 | *link = req; | |
5686 | } else { | |
5687 | io_queue_sqe(req, sqe); | |
5688 | } | |
9e645e11 | 5689 | } |
2e6e1fde | 5690 | |
1d4240cc | 5691 | return 0; |
9e645e11 JA |
5692 | } |
5693 | ||
9a56a232 JA |
5694 | /* |
5695 | * Batched submission is done, ensure local IO is flushed out. | |
5696 | */ | |
5697 | static void io_submit_state_end(struct io_submit_state *state) | |
5698 | { | |
5699 | blk_finish_plug(&state->plug); | |
3d6770fb | 5700 | io_file_put(state); |
2579f913 | 5701 | if (state->free_reqs) |
6c8a3134 | 5702 | kmem_cache_free_bulk(req_cachep, state->free_reqs, state->reqs); |
9a56a232 JA |
5703 | } |
5704 | ||
5705 | /* | |
5706 | * Start submission side cache. | |
5707 | */ | |
5708 | static void io_submit_state_start(struct io_submit_state *state, | |
22efde59 | 5709 | unsigned int max_ios) |
9a56a232 JA |
5710 | { |
5711 | blk_start_plug(&state->plug); | |
2579f913 | 5712 | state->free_reqs = 0; |
9a56a232 JA |
5713 | state->file = NULL; |
5714 | state->ios_left = max_ios; | |
5715 | } | |
5716 | ||
2b188cc1 JA |
5717 | static void io_commit_sqring(struct io_ring_ctx *ctx) |
5718 | { | |
75b28aff | 5719 | struct io_rings *rings = ctx->rings; |
2b188cc1 | 5720 | |
caf582c6 PB |
5721 | /* |
5722 | * Ensure any loads from the SQEs are done at this point, | |
5723 | * since once we write the new head, the application could | |
5724 | * write new data to them. | |
5725 | */ | |
5726 | smp_store_release(&rings->sq.head, ctx->cached_sq_head); | |
2b188cc1 JA |
5727 | } |
5728 | ||
2b188cc1 | 5729 | /* |
3529d8c2 | 5730 | * Fetch an sqe, if one is available. Note that sqe_ptr will point to memory |
2b188cc1 JA |
5731 | * that is mapped by userspace. This means that care needs to be taken to |
5732 | * ensure that reads are stable, as we cannot rely on userspace always | |
5733 | * being a good citizen. If members of the sqe are validated and then later | |
5734 | * used, it's important that those reads are done through READ_ONCE() to | |
5735 | * prevent a re-load down the line. | |
5736 | */ | |
709b302f | 5737 | static const struct io_uring_sqe *io_get_sqe(struct io_ring_ctx *ctx) |
2b188cc1 | 5738 | { |
75b28aff | 5739 | u32 *sq_array = ctx->sq_array; |
2b188cc1 JA |
5740 | unsigned head; |
5741 | ||
5742 | /* | |
5743 | * The cached sq head (or cq tail) serves two purposes: | |
5744 | * | |
5745 | * 1) allows us to batch the cost of updating the user visible | |
5746 | * head updates. | |
5747 | * 2) allows the kernel side to track the head on its own, even | |
5748 | * though the application is the one updating it. | |
5749 | */ | |
ee7d46d9 | 5750 | head = READ_ONCE(sq_array[ctx->cached_sq_head & ctx->sq_mask]); |
709b302f PB |
5751 | if (likely(head < ctx->sq_entries)) |
5752 | return &ctx->sq_sqes[head]; | |
2b188cc1 JA |
5753 | |
5754 | /* drop invalid entries */ | |
498ccd9e | 5755 | ctx->cached_sq_dropped++; |
ee7d46d9 | 5756 | WRITE_ONCE(ctx->rings->sq_dropped, ctx->cached_sq_dropped); |
709b302f PB |
5757 | return NULL; |
5758 | } | |
5759 | ||
5760 | static inline void io_consume_sqe(struct io_ring_ctx *ctx) | |
5761 | { | |
5762 | ctx->cached_sq_head++; | |
2b188cc1 JA |
5763 | } |
5764 | ||
ef4ff581 PB |
5765 | #define SQE_VALID_FLAGS (IOSQE_FIXED_FILE|IOSQE_IO_DRAIN|IOSQE_IO_LINK| \ |
5766 | IOSQE_IO_HARDLINK | IOSQE_ASYNC | \ | |
5767 | IOSQE_BUFFER_SELECT) | |
5768 | ||
5769 | static int io_init_req(struct io_ring_ctx *ctx, struct io_kiocb *req, | |
5770 | const struct io_uring_sqe *sqe, | |
5771 | struct io_submit_state *state, bool async) | |
0553b8bd | 5772 | { |
ef4ff581 | 5773 | unsigned int sqe_flags; |
63ff8223 | 5774 | int id; |
ef4ff581 | 5775 | |
0553b8bd PB |
5776 | /* |
5777 | * All io need record the previous position, if LINK vs DARIN, | |
5778 | * it can be used to mark the position of the first IO in the | |
5779 | * link list. | |
5780 | */ | |
31af27c7 | 5781 | req->sequence = ctx->cached_sq_head - ctx->cached_sq_dropped; |
0553b8bd PB |
5782 | req->opcode = READ_ONCE(sqe->opcode); |
5783 | req->user_data = READ_ONCE(sqe->user_data); | |
5784 | req->io = NULL; | |
5785 | req->file = NULL; | |
5786 | req->ctx = ctx; | |
5787 | req->flags = 0; | |
5788 | /* one is dropped after submission, the other at completion */ | |
5789 | refcount_set(&req->refs, 2); | |
5790 | req->task = NULL; | |
5791 | req->result = 0; | |
ef4ff581 | 5792 | req->needs_fixed_file = async; |
0553b8bd | 5793 | INIT_IO_WORK(&req->work, io_wq_submit_work); |
ef4ff581 PB |
5794 | |
5795 | if (unlikely(req->opcode >= IORING_OP_LAST)) | |
5796 | return -EINVAL; | |
5797 | ||
5798 | if (io_op_defs[req->opcode].needs_mm && !current->mm) { | |
5799 | if (unlikely(!mmget_not_zero(ctx->sqo_mm))) | |
5800 | return -EFAULT; | |
5801 | use_mm(ctx->sqo_mm); | |
5802 | } | |
5803 | ||
5804 | sqe_flags = READ_ONCE(sqe->flags); | |
5805 | /* enforce forwards compatibility on users */ | |
5806 | if (unlikely(sqe_flags & ~SQE_VALID_FLAGS)) | |
5807 | return -EINVAL; | |
5808 | ||
5809 | if ((sqe_flags & IOSQE_BUFFER_SELECT) && | |
5810 | !io_op_defs[req->opcode].buffer_select) | |
5811 | return -EOPNOTSUPP; | |
5812 | ||
5813 | id = READ_ONCE(sqe->personality); | |
5814 | if (id) { | |
5815 | req->work.creds = idr_find(&ctx->personality_idr, id); | |
5816 | if (unlikely(!req->work.creds)) | |
5817 | return -EINVAL; | |
5818 | get_cred(req->work.creds); | |
5819 | } | |
5820 | ||
5821 | /* same numerical values with corresponding REQ_F_*, safe to copy */ | |
5822 | req->flags |= sqe_flags & (IOSQE_IO_DRAIN | IOSQE_IO_HARDLINK | | |
5823 | IOSQE_ASYNC | IOSQE_FIXED_FILE | | |
5824 | IOSQE_BUFFER_SELECT | IOSQE_IO_LINK); | |
5825 | ||
63ff8223 JA |
5826 | if (!io_op_defs[req->opcode].needs_file) |
5827 | return 0; | |
5828 | ||
5829 | return io_req_set_file(state, req, READ_ONCE(sqe->fd)); | |
0553b8bd PB |
5830 | } |
5831 | ||
fb5ccc98 | 5832 | static int io_submit_sqes(struct io_ring_ctx *ctx, unsigned int nr, |
bf9c2f1c | 5833 | struct file *ring_file, int ring_fd, bool async) |
6c271ce2 JA |
5834 | { |
5835 | struct io_submit_state state, *statep = NULL; | |
9e645e11 | 5836 | struct io_kiocb *link = NULL; |
9e645e11 | 5837 | int i, submitted = 0; |
6c271ce2 | 5838 | |
c4a2ed72 | 5839 | /* if we have a backlog and couldn't flush it all, return BUSY */ |
ad3eb2c8 JA |
5840 | if (test_bit(0, &ctx->sq_check_overflow)) { |
5841 | if (!list_empty(&ctx->cq_overflow_list) && | |
5842 | !io_cqring_overflow_flush(ctx, false)) | |
5843 | return -EBUSY; | |
5844 | } | |
6c271ce2 | 5845 | |
ee7d46d9 PB |
5846 | /* make sure SQ entry isn't read before tail */ |
5847 | nr = min3(nr, ctx->sq_entries, io_sqring_entries(ctx)); | |
9ef4f124 | 5848 | |
2b85edfc PB |
5849 | if (!percpu_ref_tryget_many(&ctx->refs, nr)) |
5850 | return -EAGAIN; | |
6c271ce2 JA |
5851 | |
5852 | if (nr > IO_PLUG_THRESHOLD) { | |
22efde59 | 5853 | io_submit_state_start(&state, nr); |
6c271ce2 JA |
5854 | statep = &state; |
5855 | } | |
5856 | ||
b14cca0c PB |
5857 | ctx->ring_fd = ring_fd; |
5858 | ctx->ring_file = ring_file; | |
5859 | ||
6c271ce2 | 5860 | for (i = 0; i < nr; i++) { |
3529d8c2 | 5861 | const struct io_uring_sqe *sqe; |
196be95c | 5862 | struct io_kiocb *req; |
1cb1edb2 | 5863 | int err; |
fb5ccc98 | 5864 | |
b1e50e54 PB |
5865 | sqe = io_get_sqe(ctx); |
5866 | if (unlikely(!sqe)) { | |
5867 | io_consume_sqe(ctx); | |
5868 | break; | |
5869 | } | |
0553b8bd | 5870 | req = io_alloc_req(ctx, statep); |
196be95c PB |
5871 | if (unlikely(!req)) { |
5872 | if (!submitted) | |
5873 | submitted = -EAGAIN; | |
fb5ccc98 | 5874 | break; |
196be95c | 5875 | } |
fb5ccc98 | 5876 | |
ef4ff581 | 5877 | err = io_init_req(ctx, req, sqe, statep, async); |
709b302f | 5878 | io_consume_sqe(ctx); |
d3656344 JA |
5879 | /* will complete beyond this point, count as submitted */ |
5880 | submitted++; | |
5881 | ||
ef4ff581 | 5882 | if (unlikely(err)) { |
1cb1edb2 PB |
5883 | fail_req: |
5884 | io_cqring_add_event(req, err); | |
d3656344 | 5885 | io_double_put_req(req); |
196be95c PB |
5886 | break; |
5887 | } | |
fb5ccc98 | 5888 | |
354420f7 JA |
5889 | trace_io_uring_submit_sqe(ctx, req->opcode, req->user_data, |
5890 | true, async); | |
7d01bd74 | 5891 | err = io_submit_sqe(req, sqe, &link); |
1d4240cc PB |
5892 | if (err) |
5893 | goto fail_req; | |
6c271ce2 JA |
5894 | } |
5895 | ||
9466f437 PB |
5896 | if (unlikely(submitted != nr)) { |
5897 | int ref_used = (submitted == -EAGAIN) ? 0 : submitted; | |
5898 | ||
5899 | percpu_ref_put_many(&ctx->refs, nr - ref_used); | |
5900 | } | |
9e645e11 | 5901 | if (link) |
1b4a51b6 | 5902 | io_queue_link_head(link); |
6c271ce2 JA |
5903 | if (statep) |
5904 | io_submit_state_end(&state); | |
5905 | ||
ae9428ca PB |
5906 | /* Commit SQ ring head once we've consumed and submitted all SQEs */ |
5907 | io_commit_sqring(ctx); | |
5908 | ||
6c271ce2 JA |
5909 | return submitted; |
5910 | } | |
5911 | ||
bf9c2f1c PB |
5912 | static inline void io_sq_thread_drop_mm(struct io_ring_ctx *ctx) |
5913 | { | |
5914 | struct mm_struct *mm = current->mm; | |
5915 | ||
5916 | if (mm) { | |
5917 | unuse_mm(mm); | |
5918 | mmput(mm); | |
5919 | } | |
5920 | } | |
5921 | ||
6c271ce2 JA |
5922 | static int io_sq_thread(void *data) |
5923 | { | |
6c271ce2 | 5924 | struct io_ring_ctx *ctx = data; |
181e448d | 5925 | const struct cred *old_cred; |
6c271ce2 JA |
5926 | mm_segment_t old_fs; |
5927 | DEFINE_WAIT(wait); | |
6c271ce2 | 5928 | unsigned long timeout; |
bdcd3eab | 5929 | int ret = 0; |
6c271ce2 | 5930 | |
0f158b4c | 5931 | complete(&ctx->sq_thread_comp); |
a4c0b3de | 5932 | |
6c271ce2 JA |
5933 | old_fs = get_fs(); |
5934 | set_fs(USER_DS); | |
181e448d | 5935 | old_cred = override_creds(ctx->creds); |
6c271ce2 | 5936 | |
bdcd3eab | 5937 | timeout = jiffies + ctx->sq_thread_idle; |
2bbcd6d3 | 5938 | while (!kthread_should_park()) { |
fb5ccc98 | 5939 | unsigned int to_submit; |
6c271ce2 | 5940 | |
bdcd3eab | 5941 | if (!list_empty(&ctx->poll_list)) { |
6c271ce2 JA |
5942 | unsigned nr_events = 0; |
5943 | ||
bdcd3eab XW |
5944 | mutex_lock(&ctx->uring_lock); |
5945 | if (!list_empty(&ctx->poll_list)) | |
5946 | io_iopoll_getevents(ctx, &nr_events, 0); | |
5947 | else | |
6c271ce2 | 5948 | timeout = jiffies + ctx->sq_thread_idle; |
bdcd3eab | 5949 | mutex_unlock(&ctx->uring_lock); |
6c271ce2 JA |
5950 | } |
5951 | ||
fb5ccc98 | 5952 | to_submit = io_sqring_entries(ctx); |
c1edbf5f JA |
5953 | |
5954 | /* | |
5955 | * If submit got -EBUSY, flag us as needing the application | |
5956 | * to enter the kernel to reap and flush events. | |
5957 | */ | |
5958 | if (!to_submit || ret == -EBUSY) { | |
7143b5ac SG |
5959 | /* |
5960 | * Drop cur_mm before scheduling, we can't hold it for | |
5961 | * long periods (or over schedule()). Do this before | |
5962 | * adding ourselves to the waitqueue, as the unuse/drop | |
5963 | * may sleep. | |
5964 | */ | |
bf9c2f1c | 5965 | io_sq_thread_drop_mm(ctx); |
7143b5ac | 5966 | |
6c271ce2 JA |
5967 | /* |
5968 | * We're polling. If we're within the defined idle | |
5969 | * period, then let us spin without work before going | |
c1edbf5f JA |
5970 | * to sleep. The exception is if we got EBUSY doing |
5971 | * more IO, we should wait for the application to | |
5972 | * reap events and wake us up. | |
6c271ce2 | 5973 | */ |
bdcd3eab | 5974 | if (!list_empty(&ctx->poll_list) || |
df069d80 JA |
5975 | (!time_after(jiffies, timeout) && ret != -EBUSY && |
5976 | !percpu_ref_is_dying(&ctx->refs))) { | |
b41e9852 JA |
5977 | if (current->task_works) |
5978 | task_work_run(); | |
9831a90c | 5979 | cond_resched(); |
6c271ce2 JA |
5980 | continue; |
5981 | } | |
5982 | ||
6c271ce2 JA |
5983 | prepare_to_wait(&ctx->sqo_wait, &wait, |
5984 | TASK_INTERRUPTIBLE); | |
5985 | ||
bdcd3eab XW |
5986 | /* |
5987 | * While doing polled IO, before going to sleep, we need | |
5988 | * to check if there are new reqs added to poll_list, it | |
5989 | * is because reqs may have been punted to io worker and | |
5990 | * will be added to poll_list later, hence check the | |
5991 | * poll_list again. | |
5992 | */ | |
5993 | if ((ctx->flags & IORING_SETUP_IOPOLL) && | |
5994 | !list_empty_careful(&ctx->poll_list)) { | |
5995 | finish_wait(&ctx->sqo_wait, &wait); | |
5996 | continue; | |
5997 | } | |
5998 | ||
6c271ce2 | 5999 | /* Tell userspace we may need a wakeup call */ |
75b28aff | 6000 | ctx->rings->sq_flags |= IORING_SQ_NEED_WAKEUP; |
0d7bae69 SB |
6001 | /* make sure to read SQ tail after writing flags */ |
6002 | smp_mb(); | |
6c271ce2 | 6003 | |
fb5ccc98 | 6004 | to_submit = io_sqring_entries(ctx); |
c1edbf5f | 6005 | if (!to_submit || ret == -EBUSY) { |
2bbcd6d3 | 6006 | if (kthread_should_park()) { |
6c271ce2 JA |
6007 | finish_wait(&ctx->sqo_wait, &wait); |
6008 | break; | |
6009 | } | |
b41e9852 JA |
6010 | if (current->task_works) { |
6011 | task_work_run(); | |
10bea96d | 6012 | finish_wait(&ctx->sqo_wait, &wait); |
b41e9852 JA |
6013 | continue; |
6014 | } | |
6c271ce2 JA |
6015 | if (signal_pending(current)) |
6016 | flush_signals(current); | |
6017 | schedule(); | |
6018 | finish_wait(&ctx->sqo_wait, &wait); | |
6019 | ||
75b28aff | 6020 | ctx->rings->sq_flags &= ~IORING_SQ_NEED_WAKEUP; |
6c271ce2 JA |
6021 | continue; |
6022 | } | |
6023 | finish_wait(&ctx->sqo_wait, &wait); | |
6024 | ||
75b28aff | 6025 | ctx->rings->sq_flags &= ~IORING_SQ_NEED_WAKEUP; |
6c271ce2 JA |
6026 | } |
6027 | ||
8a4955ff | 6028 | mutex_lock(&ctx->uring_lock); |
bf9c2f1c | 6029 | ret = io_submit_sqes(ctx, to_submit, NULL, -1, true); |
8a4955ff | 6030 | mutex_unlock(&ctx->uring_lock); |
bdcd3eab | 6031 | timeout = jiffies + ctx->sq_thread_idle; |
6c271ce2 JA |
6032 | } |
6033 | ||
b41e9852 JA |
6034 | if (current->task_works) |
6035 | task_work_run(); | |
6036 | ||
6c271ce2 | 6037 | set_fs(old_fs); |
bf9c2f1c | 6038 | io_sq_thread_drop_mm(ctx); |
181e448d | 6039 | revert_creds(old_cred); |
06058632 | 6040 | |
2bbcd6d3 | 6041 | kthread_parkme(); |
06058632 | 6042 | |
6c271ce2 JA |
6043 | return 0; |
6044 | } | |
6045 | ||
bda52162 JA |
6046 | struct io_wait_queue { |
6047 | struct wait_queue_entry wq; | |
6048 | struct io_ring_ctx *ctx; | |
6049 | unsigned to_wait; | |
6050 | unsigned nr_timeouts; | |
6051 | }; | |
6052 | ||
1d7bb1d5 | 6053 | static inline bool io_should_wake(struct io_wait_queue *iowq, bool noflush) |
bda52162 JA |
6054 | { |
6055 | struct io_ring_ctx *ctx = iowq->ctx; | |
6056 | ||
6057 | /* | |
d195a66e | 6058 | * Wake up if we have enough events, or if a timeout occurred since we |
bda52162 JA |
6059 | * started waiting. For timeouts, we always want to return to userspace, |
6060 | * regardless of event count. | |
6061 | */ | |
1d7bb1d5 | 6062 | return io_cqring_events(ctx, noflush) >= iowq->to_wait || |
bda52162 JA |
6063 | atomic_read(&ctx->cq_timeouts) != iowq->nr_timeouts; |
6064 | } | |
6065 | ||
6066 | static int io_wake_function(struct wait_queue_entry *curr, unsigned int mode, | |
6067 | int wake_flags, void *key) | |
6068 | { | |
6069 | struct io_wait_queue *iowq = container_of(curr, struct io_wait_queue, | |
6070 | wq); | |
6071 | ||
1d7bb1d5 JA |
6072 | /* use noflush == true, as we can't safely rely on locking context */ |
6073 | if (!io_should_wake(iowq, true)) | |
bda52162 JA |
6074 | return -1; |
6075 | ||
6076 | return autoremove_wake_function(curr, mode, wake_flags, key); | |
6077 | } | |
6078 | ||
2b188cc1 JA |
6079 | /* |
6080 | * Wait until events become available, if we don't already have some. The | |
6081 | * application must reap them itself, as they reside on the shared cq ring. | |
6082 | */ | |
6083 | static int io_cqring_wait(struct io_ring_ctx *ctx, int min_events, | |
6084 | const sigset_t __user *sig, size_t sigsz) | |
6085 | { | |
bda52162 JA |
6086 | struct io_wait_queue iowq = { |
6087 | .wq = { | |
6088 | .private = current, | |
6089 | .func = io_wake_function, | |
6090 | .entry = LIST_HEAD_INIT(iowq.wq.entry), | |
6091 | }, | |
6092 | .ctx = ctx, | |
6093 | .to_wait = min_events, | |
6094 | }; | |
75b28aff | 6095 | struct io_rings *rings = ctx->rings; |
e9ffa5c2 | 6096 | int ret = 0; |
2b188cc1 | 6097 | |
b41e9852 JA |
6098 | do { |
6099 | if (io_cqring_events(ctx, false) >= min_events) | |
6100 | return 0; | |
6101 | if (!current->task_works) | |
6102 | break; | |
6103 | task_work_run(); | |
6104 | } while (1); | |
2b188cc1 JA |
6105 | |
6106 | if (sig) { | |
9e75ad5d AB |
6107 | #ifdef CONFIG_COMPAT |
6108 | if (in_compat_syscall()) | |
6109 | ret = set_compat_user_sigmask((const compat_sigset_t __user *)sig, | |
b772434b | 6110 | sigsz); |
9e75ad5d AB |
6111 | else |
6112 | #endif | |
b772434b | 6113 | ret = set_user_sigmask(sig, sigsz); |
9e75ad5d | 6114 | |
2b188cc1 JA |
6115 | if (ret) |
6116 | return ret; | |
6117 | } | |
6118 | ||
bda52162 | 6119 | iowq.nr_timeouts = atomic_read(&ctx->cq_timeouts); |
c826bd7a | 6120 | trace_io_uring_cqring_wait(ctx, min_events); |
bda52162 JA |
6121 | do { |
6122 | prepare_to_wait_exclusive(&ctx->wait, &iowq.wq, | |
6123 | TASK_INTERRUPTIBLE); | |
b41e9852 JA |
6124 | if (current->task_works) |
6125 | task_work_run(); | |
1d7bb1d5 | 6126 | if (io_should_wake(&iowq, false)) |
bda52162 JA |
6127 | break; |
6128 | schedule(); | |
6129 | if (signal_pending(current)) { | |
e9ffa5c2 | 6130 | ret = -EINTR; |
bda52162 JA |
6131 | break; |
6132 | } | |
6133 | } while (1); | |
6134 | finish_wait(&ctx->wait, &iowq.wq); | |
6135 | ||
e9ffa5c2 | 6136 | restore_saved_sigmask_unless(ret == -EINTR); |
2b188cc1 | 6137 | |
75b28aff | 6138 | return READ_ONCE(rings->cq.head) == READ_ONCE(rings->cq.tail) ? ret : 0; |
2b188cc1 JA |
6139 | } |
6140 | ||
6b06314c JA |
6141 | static void __io_sqe_files_unregister(struct io_ring_ctx *ctx) |
6142 | { | |
6143 | #if defined(CONFIG_UNIX) | |
6144 | if (ctx->ring_sock) { | |
6145 | struct sock *sock = ctx->ring_sock->sk; | |
6146 | struct sk_buff *skb; | |
6147 | ||
6148 | while ((skb = skb_dequeue(&sock->sk_receive_queue)) != NULL) | |
6149 | kfree_skb(skb); | |
6150 | } | |
6151 | #else | |
6152 | int i; | |
6153 | ||
65e19f54 JA |
6154 | for (i = 0; i < ctx->nr_user_files; i++) { |
6155 | struct file *file; | |
6156 | ||
6157 | file = io_file_from_index(ctx, i); | |
6158 | if (file) | |
6159 | fput(file); | |
6160 | } | |
6b06314c JA |
6161 | #endif |
6162 | } | |
6163 | ||
05f3fb3c JA |
6164 | static void io_file_ref_kill(struct percpu_ref *ref) |
6165 | { | |
6166 | struct fixed_file_data *data; | |
6167 | ||
6168 | data = container_of(ref, struct fixed_file_data, refs); | |
6169 | complete(&data->done); | |
6170 | } | |
6171 | ||
6b06314c JA |
6172 | static int io_sqe_files_unregister(struct io_ring_ctx *ctx) |
6173 | { | |
05f3fb3c | 6174 | struct fixed_file_data *data = ctx->file_data; |
05589553 | 6175 | struct fixed_file_ref_node *ref_node = NULL; |
65e19f54 | 6176 | unsigned nr_tables, i; |
05589553 | 6177 | unsigned long flags; |
65e19f54 | 6178 | |
05f3fb3c | 6179 | if (!data) |
6b06314c JA |
6180 | return -ENXIO; |
6181 | ||
05589553 XW |
6182 | spin_lock_irqsave(&data->lock, flags); |
6183 | if (!list_empty(&data->ref_list)) | |
6184 | ref_node = list_first_entry(&data->ref_list, | |
6185 | struct fixed_file_ref_node, node); | |
6186 | spin_unlock_irqrestore(&data->lock, flags); | |
6187 | if (ref_node) | |
6188 | percpu_ref_kill(&ref_node->refs); | |
6189 | ||
6190 | percpu_ref_kill(&data->refs); | |
6191 | ||
6192 | /* wait for all refs nodes to complete */ | |
2faf852d | 6193 | wait_for_completion(&data->done); |
05f3fb3c | 6194 | |
6b06314c | 6195 | __io_sqe_files_unregister(ctx); |
65e19f54 JA |
6196 | nr_tables = DIV_ROUND_UP(ctx->nr_user_files, IORING_MAX_FILES_TABLE); |
6197 | for (i = 0; i < nr_tables; i++) | |
05f3fb3c JA |
6198 | kfree(data->table[i].files); |
6199 | kfree(data->table); | |
05589553 XW |
6200 | percpu_ref_exit(&data->refs); |
6201 | kfree(data); | |
05f3fb3c | 6202 | ctx->file_data = NULL; |
6b06314c JA |
6203 | ctx->nr_user_files = 0; |
6204 | return 0; | |
6205 | } | |
6206 | ||
6c271ce2 JA |
6207 | static void io_sq_thread_stop(struct io_ring_ctx *ctx) |
6208 | { | |
6209 | if (ctx->sqo_thread) { | |
0f158b4c | 6210 | wait_for_completion(&ctx->sq_thread_comp); |
2bbcd6d3 RP |
6211 | /* |
6212 | * The park is a bit of a work-around, without it we get | |
6213 | * warning spews on shutdown with SQPOLL set and affinity | |
6214 | * set to a single CPU. | |
6215 | */ | |
06058632 | 6216 | kthread_park(ctx->sqo_thread); |
6c271ce2 JA |
6217 | kthread_stop(ctx->sqo_thread); |
6218 | ctx->sqo_thread = NULL; | |
6219 | } | |
6220 | } | |
6221 | ||
6b06314c JA |
6222 | static void io_finish_async(struct io_ring_ctx *ctx) |
6223 | { | |
6c271ce2 JA |
6224 | io_sq_thread_stop(ctx); |
6225 | ||
561fb04a JA |
6226 | if (ctx->io_wq) { |
6227 | io_wq_destroy(ctx->io_wq); | |
6228 | ctx->io_wq = NULL; | |
6b06314c JA |
6229 | } |
6230 | } | |
6231 | ||
6232 | #if defined(CONFIG_UNIX) | |
6b06314c JA |
6233 | /* |
6234 | * Ensure the UNIX gc is aware of our file set, so we are certain that | |
6235 | * the io_uring can be safely unregistered on process exit, even if we have | |
6236 | * loops in the file referencing. | |
6237 | */ | |
6238 | static int __io_sqe_files_scm(struct io_ring_ctx *ctx, int nr, int offset) | |
6239 | { | |
6240 | struct sock *sk = ctx->ring_sock->sk; | |
6241 | struct scm_fp_list *fpl; | |
6242 | struct sk_buff *skb; | |
08a45173 | 6243 | int i, nr_files; |
6b06314c | 6244 | |
6b06314c JA |
6245 | fpl = kzalloc(sizeof(*fpl), GFP_KERNEL); |
6246 | if (!fpl) | |
6247 | return -ENOMEM; | |
6248 | ||
6249 | skb = alloc_skb(0, GFP_KERNEL); | |
6250 | if (!skb) { | |
6251 | kfree(fpl); | |
6252 | return -ENOMEM; | |
6253 | } | |
6254 | ||
6255 | skb->sk = sk; | |
6b06314c | 6256 | |
08a45173 | 6257 | nr_files = 0; |
6b06314c JA |
6258 | fpl->user = get_uid(ctx->user); |
6259 | for (i = 0; i < nr; i++) { | |
65e19f54 JA |
6260 | struct file *file = io_file_from_index(ctx, i + offset); |
6261 | ||
6262 | if (!file) | |
08a45173 | 6263 | continue; |
65e19f54 | 6264 | fpl->fp[nr_files] = get_file(file); |
08a45173 JA |
6265 | unix_inflight(fpl->user, fpl->fp[nr_files]); |
6266 | nr_files++; | |
6b06314c JA |
6267 | } |
6268 | ||
08a45173 JA |
6269 | if (nr_files) { |
6270 | fpl->max = SCM_MAX_FD; | |
6271 | fpl->count = nr_files; | |
6272 | UNIXCB(skb).fp = fpl; | |
05f3fb3c | 6273 | skb->destructor = unix_destruct_scm; |
08a45173 JA |
6274 | refcount_add(skb->truesize, &sk->sk_wmem_alloc); |
6275 | skb_queue_head(&sk->sk_receive_queue, skb); | |
6b06314c | 6276 | |
08a45173 JA |
6277 | for (i = 0; i < nr_files; i++) |
6278 | fput(fpl->fp[i]); | |
6279 | } else { | |
6280 | kfree_skb(skb); | |
6281 | kfree(fpl); | |
6282 | } | |
6b06314c JA |
6283 | |
6284 | return 0; | |
6285 | } | |
6286 | ||
6287 | /* | |
6288 | * If UNIX sockets are enabled, fd passing can cause a reference cycle which | |
6289 | * causes regular reference counting to break down. We rely on the UNIX | |
6290 | * garbage collection to take care of this problem for us. | |
6291 | */ | |
6292 | static int io_sqe_files_scm(struct io_ring_ctx *ctx) | |
6293 | { | |
6294 | unsigned left, total; | |
6295 | int ret = 0; | |
6296 | ||
6297 | total = 0; | |
6298 | left = ctx->nr_user_files; | |
6299 | while (left) { | |
6300 | unsigned this_files = min_t(unsigned, left, SCM_MAX_FD); | |
6b06314c JA |
6301 | |
6302 | ret = __io_sqe_files_scm(ctx, this_files, total); | |
6303 | if (ret) | |
6304 | break; | |
6305 | left -= this_files; | |
6306 | total += this_files; | |
6307 | } | |
6308 | ||
6309 | if (!ret) | |
6310 | return 0; | |
6311 | ||
6312 | while (total < ctx->nr_user_files) { | |
65e19f54 JA |
6313 | struct file *file = io_file_from_index(ctx, total); |
6314 | ||
6315 | if (file) | |
6316 | fput(file); | |
6b06314c JA |
6317 | total++; |
6318 | } | |
6319 | ||
6320 | return ret; | |
6321 | } | |
6322 | #else | |
6323 | static int io_sqe_files_scm(struct io_ring_ctx *ctx) | |
6324 | { | |
6325 | return 0; | |
6326 | } | |
6327 | #endif | |
6328 | ||
65e19f54 JA |
6329 | static int io_sqe_alloc_file_tables(struct io_ring_ctx *ctx, unsigned nr_tables, |
6330 | unsigned nr_files) | |
6331 | { | |
6332 | int i; | |
6333 | ||
6334 | for (i = 0; i < nr_tables; i++) { | |
05f3fb3c | 6335 | struct fixed_file_table *table = &ctx->file_data->table[i]; |
65e19f54 JA |
6336 | unsigned this_files; |
6337 | ||
6338 | this_files = min(nr_files, IORING_MAX_FILES_TABLE); | |
6339 | table->files = kcalloc(this_files, sizeof(struct file *), | |
6340 | GFP_KERNEL); | |
6341 | if (!table->files) | |
6342 | break; | |
6343 | nr_files -= this_files; | |
6344 | } | |
6345 | ||
6346 | if (i == nr_tables) | |
6347 | return 0; | |
6348 | ||
6349 | for (i = 0; i < nr_tables; i++) { | |
05f3fb3c | 6350 | struct fixed_file_table *table = &ctx->file_data->table[i]; |
65e19f54 JA |
6351 | kfree(table->files); |
6352 | } | |
6353 | return 1; | |
6354 | } | |
6355 | ||
05f3fb3c JA |
6356 | static void io_ring_file_put(struct io_ring_ctx *ctx, struct file *file) |
6357 | { | |
6358 | #if defined(CONFIG_UNIX) | |
6359 | struct sock *sock = ctx->ring_sock->sk; | |
6360 | struct sk_buff_head list, *head = &sock->sk_receive_queue; | |
6361 | struct sk_buff *skb; | |
6362 | int i; | |
6363 | ||
6364 | __skb_queue_head_init(&list); | |
6365 | ||
6366 | /* | |
6367 | * Find the skb that holds this file in its SCM_RIGHTS. When found, | |
6368 | * remove this entry and rearrange the file array. | |
6369 | */ | |
6370 | skb = skb_dequeue(head); | |
6371 | while (skb) { | |
6372 | struct scm_fp_list *fp; | |
6373 | ||
6374 | fp = UNIXCB(skb).fp; | |
6375 | for (i = 0; i < fp->count; i++) { | |
6376 | int left; | |
6377 | ||
6378 | if (fp->fp[i] != file) | |
6379 | continue; | |
6380 | ||
6381 | unix_notinflight(fp->user, fp->fp[i]); | |
6382 | left = fp->count - 1 - i; | |
6383 | if (left) { | |
6384 | memmove(&fp->fp[i], &fp->fp[i + 1], | |
6385 | left * sizeof(struct file *)); | |
6386 | } | |
6387 | fp->count--; | |
6388 | if (!fp->count) { | |
6389 | kfree_skb(skb); | |
6390 | skb = NULL; | |
6391 | } else { | |
6392 | __skb_queue_tail(&list, skb); | |
6393 | } | |
6394 | fput(file); | |
6395 | file = NULL; | |
6396 | break; | |
6397 | } | |
6398 | ||
6399 | if (!file) | |
6400 | break; | |
6401 | ||
6402 | __skb_queue_tail(&list, skb); | |
6403 | ||
6404 | skb = skb_dequeue(head); | |
6405 | } | |
6406 | ||
6407 | if (skb_peek(&list)) { | |
6408 | spin_lock_irq(&head->lock); | |
6409 | while ((skb = __skb_dequeue(&list)) != NULL) | |
6410 | __skb_queue_tail(head, skb); | |
6411 | spin_unlock_irq(&head->lock); | |
6412 | } | |
6413 | #else | |
6414 | fput(file); | |
6415 | #endif | |
6416 | } | |
6417 | ||
6418 | struct io_file_put { | |
05589553 | 6419 | struct list_head list; |
05f3fb3c | 6420 | struct file *file; |
05f3fb3c JA |
6421 | }; |
6422 | ||
05589553 | 6423 | static void io_file_put_work(struct work_struct *work) |
65e19f54 | 6424 | { |
05589553 XW |
6425 | struct fixed_file_ref_node *ref_node; |
6426 | struct fixed_file_data *file_data; | |
6427 | struct io_ring_ctx *ctx; | |
05f3fb3c | 6428 | struct io_file_put *pfile, *tmp; |
05589553 | 6429 | unsigned long flags; |
65e19f54 | 6430 | |
05589553 XW |
6431 | ref_node = container_of(work, struct fixed_file_ref_node, work); |
6432 | file_data = ref_node->file_data; | |
6433 | ctx = file_data->ctx; | |
6434 | ||
6435 | list_for_each_entry_safe(pfile, tmp, &ref_node->file_list, list) { | |
6436 | list_del_init(&pfile->list); | |
6437 | io_ring_file_put(ctx, pfile->file); | |
6438 | kfree(pfile); | |
65e19f54 | 6439 | } |
05589553 XW |
6440 | |
6441 | spin_lock_irqsave(&file_data->lock, flags); | |
6442 | list_del_init(&ref_node->node); | |
6443 | spin_unlock_irqrestore(&file_data->lock, flags); | |
6444 | ||
6445 | percpu_ref_exit(&ref_node->refs); | |
6446 | kfree(ref_node); | |
6447 | percpu_ref_put(&file_data->refs); | |
2faf852d | 6448 | } |
65e19f54 | 6449 | |
05589553 | 6450 | static void io_file_data_ref_zero(struct percpu_ref *ref) |
2faf852d | 6451 | { |
05589553 | 6452 | struct fixed_file_ref_node *ref_node; |
65e19f54 | 6453 | |
05589553 XW |
6454 | ref_node = container_of(ref, struct fixed_file_ref_node, refs); |
6455 | ||
6456 | queue_work(system_wq, &ref_node->work); | |
05f3fb3c | 6457 | } |
65e19f54 | 6458 | |
05589553 XW |
6459 | static struct fixed_file_ref_node *alloc_fixed_file_ref_node( |
6460 | struct io_ring_ctx *ctx) | |
05f3fb3c | 6461 | { |
05589553 | 6462 | struct fixed_file_ref_node *ref_node; |
05f3fb3c | 6463 | |
05589553 XW |
6464 | ref_node = kzalloc(sizeof(*ref_node), GFP_KERNEL); |
6465 | if (!ref_node) | |
6466 | return ERR_PTR(-ENOMEM); | |
05f3fb3c | 6467 | |
05589553 XW |
6468 | if (percpu_ref_init(&ref_node->refs, io_file_data_ref_zero, |
6469 | 0, GFP_KERNEL)) { | |
6470 | kfree(ref_node); | |
6471 | return ERR_PTR(-ENOMEM); | |
6472 | } | |
6473 | INIT_LIST_HEAD(&ref_node->node); | |
6474 | INIT_LIST_HEAD(&ref_node->file_list); | |
6475 | INIT_WORK(&ref_node->work, io_file_put_work); | |
6476 | ref_node->file_data = ctx->file_data; | |
6477 | return ref_node; | |
6478 | ||
6479 | } | |
6480 | ||
6481 | static void destroy_fixed_file_ref_node(struct fixed_file_ref_node *ref_node) | |
6482 | { | |
6483 | percpu_ref_exit(&ref_node->refs); | |
6484 | kfree(ref_node); | |
65e19f54 JA |
6485 | } |
6486 | ||
6b06314c JA |
6487 | static int io_sqe_files_register(struct io_ring_ctx *ctx, void __user *arg, |
6488 | unsigned nr_args) | |
6489 | { | |
6490 | __s32 __user *fds = (__s32 __user *) arg; | |
65e19f54 | 6491 | unsigned nr_tables; |
05f3fb3c | 6492 | struct file *file; |
6b06314c JA |
6493 | int fd, ret = 0; |
6494 | unsigned i; | |
05589553 XW |
6495 | struct fixed_file_ref_node *ref_node; |
6496 | unsigned long flags; | |
6b06314c | 6497 | |
05f3fb3c | 6498 | if (ctx->file_data) |
6b06314c JA |
6499 | return -EBUSY; |
6500 | if (!nr_args) | |
6501 | return -EINVAL; | |
6502 | if (nr_args > IORING_MAX_FIXED_FILES) | |
6503 | return -EMFILE; | |
6504 | ||
05f3fb3c JA |
6505 | ctx->file_data = kzalloc(sizeof(*ctx->file_data), GFP_KERNEL); |
6506 | if (!ctx->file_data) | |
6507 | return -ENOMEM; | |
6508 | ctx->file_data->ctx = ctx; | |
6509 | init_completion(&ctx->file_data->done); | |
05589553 | 6510 | INIT_LIST_HEAD(&ctx->file_data->ref_list); |
f7fe9346 | 6511 | spin_lock_init(&ctx->file_data->lock); |
05f3fb3c | 6512 | |
65e19f54 | 6513 | nr_tables = DIV_ROUND_UP(nr_args, IORING_MAX_FILES_TABLE); |
05f3fb3c JA |
6514 | ctx->file_data->table = kcalloc(nr_tables, |
6515 | sizeof(struct fixed_file_table), | |
65e19f54 | 6516 | GFP_KERNEL); |
05f3fb3c JA |
6517 | if (!ctx->file_data->table) { |
6518 | kfree(ctx->file_data); | |
6519 | ctx->file_data = NULL; | |
6b06314c | 6520 | return -ENOMEM; |
05f3fb3c JA |
6521 | } |
6522 | ||
05589553 | 6523 | if (percpu_ref_init(&ctx->file_data->refs, io_file_ref_kill, |
05f3fb3c JA |
6524 | PERCPU_REF_ALLOW_REINIT, GFP_KERNEL)) { |
6525 | kfree(ctx->file_data->table); | |
6526 | kfree(ctx->file_data); | |
6527 | ctx->file_data = NULL; | |
6b06314c | 6528 | return -ENOMEM; |
05f3fb3c | 6529 | } |
6b06314c | 6530 | |
65e19f54 | 6531 | if (io_sqe_alloc_file_tables(ctx, nr_tables, nr_args)) { |
05f3fb3c JA |
6532 | percpu_ref_exit(&ctx->file_data->refs); |
6533 | kfree(ctx->file_data->table); | |
6534 | kfree(ctx->file_data); | |
6535 | ctx->file_data = NULL; | |
65e19f54 JA |
6536 | return -ENOMEM; |
6537 | } | |
6538 | ||
08a45173 | 6539 | for (i = 0; i < nr_args; i++, ctx->nr_user_files++) { |
65e19f54 JA |
6540 | struct fixed_file_table *table; |
6541 | unsigned index; | |
6542 | ||
6b06314c JA |
6543 | ret = -EFAULT; |
6544 | if (copy_from_user(&fd, &fds[i], sizeof(fd))) | |
6545 | break; | |
08a45173 JA |
6546 | /* allow sparse sets */ |
6547 | if (fd == -1) { | |
6548 | ret = 0; | |
6549 | continue; | |
6550 | } | |
6b06314c | 6551 | |
05f3fb3c | 6552 | table = &ctx->file_data->table[i >> IORING_FILE_TABLE_SHIFT]; |
65e19f54 | 6553 | index = i & IORING_FILE_TABLE_MASK; |
05f3fb3c | 6554 | file = fget(fd); |
6b06314c JA |
6555 | |
6556 | ret = -EBADF; | |
05f3fb3c | 6557 | if (!file) |
6b06314c | 6558 | break; |
05f3fb3c | 6559 | |
6b06314c JA |
6560 | /* |
6561 | * Don't allow io_uring instances to be registered. If UNIX | |
6562 | * isn't enabled, then this causes a reference cycle and this | |
6563 | * instance can never get freed. If UNIX is enabled we'll | |
6564 | * handle it just fine, but there's still no point in allowing | |
6565 | * a ring fd as it doesn't support regular read/write anyway. | |
6566 | */ | |
05f3fb3c JA |
6567 | if (file->f_op == &io_uring_fops) { |
6568 | fput(file); | |
6b06314c JA |
6569 | break; |
6570 | } | |
6b06314c | 6571 | ret = 0; |
05f3fb3c | 6572 | table->files[index] = file; |
6b06314c JA |
6573 | } |
6574 | ||
6575 | if (ret) { | |
65e19f54 | 6576 | for (i = 0; i < ctx->nr_user_files; i++) { |
65e19f54 JA |
6577 | file = io_file_from_index(ctx, i); |
6578 | if (file) | |
6579 | fput(file); | |
6580 | } | |
6581 | for (i = 0; i < nr_tables; i++) | |
05f3fb3c | 6582 | kfree(ctx->file_data->table[i].files); |
6b06314c | 6583 | |
05f3fb3c JA |
6584 | kfree(ctx->file_data->table); |
6585 | kfree(ctx->file_data); | |
6586 | ctx->file_data = NULL; | |
6b06314c JA |
6587 | ctx->nr_user_files = 0; |
6588 | return ret; | |
6589 | } | |
6590 | ||
6591 | ret = io_sqe_files_scm(ctx); | |
05589553 | 6592 | if (ret) { |
6b06314c | 6593 | io_sqe_files_unregister(ctx); |
05589553 XW |
6594 | return ret; |
6595 | } | |
6b06314c | 6596 | |
05589553 XW |
6597 | ref_node = alloc_fixed_file_ref_node(ctx); |
6598 | if (IS_ERR(ref_node)) { | |
6599 | io_sqe_files_unregister(ctx); | |
6600 | return PTR_ERR(ref_node); | |
6601 | } | |
6602 | ||
6603 | ctx->file_data->cur_refs = &ref_node->refs; | |
6604 | spin_lock_irqsave(&ctx->file_data->lock, flags); | |
6605 | list_add(&ref_node->node, &ctx->file_data->ref_list); | |
6606 | spin_unlock_irqrestore(&ctx->file_data->lock, flags); | |
6607 | percpu_ref_get(&ctx->file_data->refs); | |
6b06314c JA |
6608 | return ret; |
6609 | } | |
6610 | ||
c3a31e60 JA |
6611 | static int io_sqe_file_register(struct io_ring_ctx *ctx, struct file *file, |
6612 | int index) | |
6613 | { | |
6614 | #if defined(CONFIG_UNIX) | |
6615 | struct sock *sock = ctx->ring_sock->sk; | |
6616 | struct sk_buff_head *head = &sock->sk_receive_queue; | |
6617 | struct sk_buff *skb; | |
6618 | ||
6619 | /* | |
6620 | * See if we can merge this file into an existing skb SCM_RIGHTS | |
6621 | * file set. If there's no room, fall back to allocating a new skb | |
6622 | * and filling it in. | |
6623 | */ | |
6624 | spin_lock_irq(&head->lock); | |
6625 | skb = skb_peek(head); | |
6626 | if (skb) { | |
6627 | struct scm_fp_list *fpl = UNIXCB(skb).fp; | |
6628 | ||
6629 | if (fpl->count < SCM_MAX_FD) { | |
6630 | __skb_unlink(skb, head); | |
6631 | spin_unlock_irq(&head->lock); | |
6632 | fpl->fp[fpl->count] = get_file(file); | |
6633 | unix_inflight(fpl->user, fpl->fp[fpl->count]); | |
6634 | fpl->count++; | |
6635 | spin_lock_irq(&head->lock); | |
6636 | __skb_queue_head(head, skb); | |
6637 | } else { | |
6638 | skb = NULL; | |
6639 | } | |
6640 | } | |
6641 | spin_unlock_irq(&head->lock); | |
6642 | ||
6643 | if (skb) { | |
6644 | fput(file); | |
6645 | return 0; | |
6646 | } | |
6647 | ||
6648 | return __io_sqe_files_scm(ctx, 1, index); | |
6649 | #else | |
6650 | return 0; | |
6651 | #endif | |
6652 | } | |
6653 | ||
a5318d3c | 6654 | static int io_queue_file_removal(struct fixed_file_data *data, |
05589553 | 6655 | struct file *file) |
05f3fb3c | 6656 | { |
a5318d3c | 6657 | struct io_file_put *pfile; |
05589553 XW |
6658 | struct percpu_ref *refs = data->cur_refs; |
6659 | struct fixed_file_ref_node *ref_node; | |
05f3fb3c | 6660 | |
05f3fb3c | 6661 | pfile = kzalloc(sizeof(*pfile), GFP_KERNEL); |
a5318d3c HD |
6662 | if (!pfile) |
6663 | return -ENOMEM; | |
05f3fb3c | 6664 | |
05589553 | 6665 | ref_node = container_of(refs, struct fixed_file_ref_node, refs); |
05f3fb3c | 6666 | pfile->file = file; |
05589553 XW |
6667 | list_add(&pfile->list, &ref_node->file_list); |
6668 | ||
a5318d3c | 6669 | return 0; |
05f3fb3c JA |
6670 | } |
6671 | ||
6672 | static int __io_sqe_files_update(struct io_ring_ctx *ctx, | |
6673 | struct io_uring_files_update *up, | |
6674 | unsigned nr_args) | |
6675 | { | |
6676 | struct fixed_file_data *data = ctx->file_data; | |
05589553 | 6677 | struct fixed_file_ref_node *ref_node; |
05f3fb3c | 6678 | struct file *file; |
c3a31e60 JA |
6679 | __s32 __user *fds; |
6680 | int fd, i, err; | |
6681 | __u32 done; | |
05589553 XW |
6682 | unsigned long flags; |
6683 | bool needs_switch = false; | |
c3a31e60 | 6684 | |
05f3fb3c | 6685 | if (check_add_overflow(up->offset, nr_args, &done)) |
c3a31e60 JA |
6686 | return -EOVERFLOW; |
6687 | if (done > ctx->nr_user_files) | |
6688 | return -EINVAL; | |
6689 | ||
05589553 XW |
6690 | ref_node = alloc_fixed_file_ref_node(ctx); |
6691 | if (IS_ERR(ref_node)) | |
6692 | return PTR_ERR(ref_node); | |
6693 | ||
c3a31e60 | 6694 | done = 0; |
05f3fb3c | 6695 | fds = u64_to_user_ptr(up->fds); |
c3a31e60 | 6696 | while (nr_args) { |
65e19f54 JA |
6697 | struct fixed_file_table *table; |
6698 | unsigned index; | |
6699 | ||
c3a31e60 JA |
6700 | err = 0; |
6701 | if (copy_from_user(&fd, &fds[done], sizeof(fd))) { | |
6702 | err = -EFAULT; | |
6703 | break; | |
6704 | } | |
05f3fb3c JA |
6705 | i = array_index_nospec(up->offset, ctx->nr_user_files); |
6706 | table = &ctx->file_data->table[i >> IORING_FILE_TABLE_SHIFT]; | |
65e19f54 JA |
6707 | index = i & IORING_FILE_TABLE_MASK; |
6708 | if (table->files[index]) { | |
05f3fb3c | 6709 | file = io_file_from_index(ctx, index); |
a5318d3c HD |
6710 | err = io_queue_file_removal(data, file); |
6711 | if (err) | |
6712 | break; | |
65e19f54 | 6713 | table->files[index] = NULL; |
05589553 | 6714 | needs_switch = true; |
c3a31e60 JA |
6715 | } |
6716 | if (fd != -1) { | |
c3a31e60 JA |
6717 | file = fget(fd); |
6718 | if (!file) { | |
6719 | err = -EBADF; | |
6720 | break; | |
6721 | } | |
6722 | /* | |
6723 | * Don't allow io_uring instances to be registered. If | |
6724 | * UNIX isn't enabled, then this causes a reference | |
6725 | * cycle and this instance can never get freed. If UNIX | |
6726 | * is enabled we'll handle it just fine, but there's | |
6727 | * still no point in allowing a ring fd as it doesn't | |
6728 | * support regular read/write anyway. | |
6729 | */ | |
6730 | if (file->f_op == &io_uring_fops) { | |
6731 | fput(file); | |
6732 | err = -EBADF; | |
6733 | break; | |
6734 | } | |
65e19f54 | 6735 | table->files[index] = file; |
c3a31e60 JA |
6736 | err = io_sqe_file_register(ctx, file, i); |
6737 | if (err) | |
6738 | break; | |
6739 | } | |
6740 | nr_args--; | |
6741 | done++; | |
05f3fb3c JA |
6742 | up->offset++; |
6743 | } | |
6744 | ||
05589553 XW |
6745 | if (needs_switch) { |
6746 | percpu_ref_kill(data->cur_refs); | |
6747 | spin_lock_irqsave(&data->lock, flags); | |
6748 | list_add(&ref_node->node, &data->ref_list); | |
6749 | data->cur_refs = &ref_node->refs; | |
6750 | spin_unlock_irqrestore(&data->lock, flags); | |
6751 | percpu_ref_get(&ctx->file_data->refs); | |
6752 | } else | |
6753 | destroy_fixed_file_ref_node(ref_node); | |
c3a31e60 JA |
6754 | |
6755 | return done ? done : err; | |
6756 | } | |
05589553 | 6757 | |
05f3fb3c JA |
6758 | static int io_sqe_files_update(struct io_ring_ctx *ctx, void __user *arg, |
6759 | unsigned nr_args) | |
6760 | { | |
6761 | struct io_uring_files_update up; | |
6762 | ||
6763 | if (!ctx->file_data) | |
6764 | return -ENXIO; | |
6765 | if (!nr_args) | |
6766 | return -EINVAL; | |
6767 | if (copy_from_user(&up, arg, sizeof(up))) | |
6768 | return -EFAULT; | |
6769 | if (up.resv) | |
6770 | return -EINVAL; | |
6771 | ||
6772 | return __io_sqe_files_update(ctx, &up, nr_args); | |
6773 | } | |
c3a31e60 | 6774 | |
e9fd9396 | 6775 | static void io_free_work(struct io_wq_work *work) |
7d723065 JA |
6776 | { |
6777 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); | |
6778 | ||
e9fd9396 | 6779 | /* Consider that io_steal_work() relies on this ref */ |
7d723065 JA |
6780 | io_put_req(req); |
6781 | } | |
6782 | ||
24369c2e PB |
6783 | static int io_init_wq_offload(struct io_ring_ctx *ctx, |
6784 | struct io_uring_params *p) | |
6785 | { | |
6786 | struct io_wq_data data; | |
6787 | struct fd f; | |
6788 | struct io_ring_ctx *ctx_attach; | |
6789 | unsigned int concurrency; | |
6790 | int ret = 0; | |
6791 | ||
6792 | data.user = ctx->user; | |
e9fd9396 | 6793 | data.free_work = io_free_work; |
24369c2e PB |
6794 | |
6795 | if (!(p->flags & IORING_SETUP_ATTACH_WQ)) { | |
6796 | /* Do QD, or 4 * CPUS, whatever is smallest */ | |
6797 | concurrency = min(ctx->sq_entries, 4 * num_online_cpus()); | |
6798 | ||
6799 | ctx->io_wq = io_wq_create(concurrency, &data); | |
6800 | if (IS_ERR(ctx->io_wq)) { | |
6801 | ret = PTR_ERR(ctx->io_wq); | |
6802 | ctx->io_wq = NULL; | |
6803 | } | |
6804 | return ret; | |
6805 | } | |
6806 | ||
6807 | f = fdget(p->wq_fd); | |
6808 | if (!f.file) | |
6809 | return -EBADF; | |
6810 | ||
6811 | if (f.file->f_op != &io_uring_fops) { | |
6812 | ret = -EINVAL; | |
6813 | goto out_fput; | |
6814 | } | |
6815 | ||
6816 | ctx_attach = f.file->private_data; | |
6817 | /* @io_wq is protected by holding the fd */ | |
6818 | if (!io_wq_get(ctx_attach->io_wq, &data)) { | |
6819 | ret = -EINVAL; | |
6820 | goto out_fput; | |
6821 | } | |
6822 | ||
6823 | ctx->io_wq = ctx_attach->io_wq; | |
6824 | out_fput: | |
6825 | fdput(f); | |
6826 | return ret; | |
6827 | } | |
6828 | ||
6c271ce2 JA |
6829 | static int io_sq_offload_start(struct io_ring_ctx *ctx, |
6830 | struct io_uring_params *p) | |
2b188cc1 JA |
6831 | { |
6832 | int ret; | |
6833 | ||
6c271ce2 | 6834 | init_waitqueue_head(&ctx->sqo_wait); |
2b188cc1 JA |
6835 | mmgrab(current->mm); |
6836 | ctx->sqo_mm = current->mm; | |
6837 | ||
6c271ce2 | 6838 | if (ctx->flags & IORING_SETUP_SQPOLL) { |
3ec482d1 JA |
6839 | ret = -EPERM; |
6840 | if (!capable(CAP_SYS_ADMIN)) | |
6841 | goto err; | |
6842 | ||
917257da JA |
6843 | ctx->sq_thread_idle = msecs_to_jiffies(p->sq_thread_idle); |
6844 | if (!ctx->sq_thread_idle) | |
6845 | ctx->sq_thread_idle = HZ; | |
6846 | ||
6c271ce2 | 6847 | if (p->flags & IORING_SETUP_SQ_AFF) { |
44a9bd18 | 6848 | int cpu = p->sq_thread_cpu; |
6c271ce2 | 6849 | |
917257da | 6850 | ret = -EINVAL; |
44a9bd18 JA |
6851 | if (cpu >= nr_cpu_ids) |
6852 | goto err; | |
7889f44d | 6853 | if (!cpu_online(cpu)) |
917257da JA |
6854 | goto err; |
6855 | ||
6c271ce2 JA |
6856 | ctx->sqo_thread = kthread_create_on_cpu(io_sq_thread, |
6857 | ctx, cpu, | |
6858 | "io_uring-sq"); | |
6859 | } else { | |
6860 | ctx->sqo_thread = kthread_create(io_sq_thread, ctx, | |
6861 | "io_uring-sq"); | |
6862 | } | |
6863 | if (IS_ERR(ctx->sqo_thread)) { | |
6864 | ret = PTR_ERR(ctx->sqo_thread); | |
6865 | ctx->sqo_thread = NULL; | |
6866 | goto err; | |
6867 | } | |
6868 | wake_up_process(ctx->sqo_thread); | |
6869 | } else if (p->flags & IORING_SETUP_SQ_AFF) { | |
6870 | /* Can't have SQ_AFF without SQPOLL */ | |
6871 | ret = -EINVAL; | |
6872 | goto err; | |
6873 | } | |
6874 | ||
24369c2e PB |
6875 | ret = io_init_wq_offload(ctx, p); |
6876 | if (ret) | |
2b188cc1 | 6877 | goto err; |
2b188cc1 JA |
6878 | |
6879 | return 0; | |
6880 | err: | |
54a91f3b | 6881 | io_finish_async(ctx); |
2b188cc1 JA |
6882 | mmdrop(ctx->sqo_mm); |
6883 | ctx->sqo_mm = NULL; | |
6884 | return ret; | |
6885 | } | |
6886 | ||
6887 | static void io_unaccount_mem(struct user_struct *user, unsigned long nr_pages) | |
6888 | { | |
6889 | atomic_long_sub(nr_pages, &user->locked_vm); | |
6890 | } | |
6891 | ||
6892 | static int io_account_mem(struct user_struct *user, unsigned long nr_pages) | |
6893 | { | |
6894 | unsigned long page_limit, cur_pages, new_pages; | |
6895 | ||
6896 | /* Don't allow more pages than we can safely lock */ | |
6897 | page_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT; | |
6898 | ||
6899 | do { | |
6900 | cur_pages = atomic_long_read(&user->locked_vm); | |
6901 | new_pages = cur_pages + nr_pages; | |
6902 | if (new_pages > page_limit) | |
6903 | return -ENOMEM; | |
6904 | } while (atomic_long_cmpxchg(&user->locked_vm, cur_pages, | |
6905 | new_pages) != cur_pages); | |
6906 | ||
6907 | return 0; | |
6908 | } | |
6909 | ||
6910 | static void io_mem_free(void *ptr) | |
6911 | { | |
52e04ef4 MR |
6912 | struct page *page; |
6913 | ||
6914 | if (!ptr) | |
6915 | return; | |
2b188cc1 | 6916 | |
52e04ef4 | 6917 | page = virt_to_head_page(ptr); |
2b188cc1 JA |
6918 | if (put_page_testzero(page)) |
6919 | free_compound_page(page); | |
6920 | } | |
6921 | ||
6922 | static void *io_mem_alloc(size_t size) | |
6923 | { | |
6924 | gfp_t gfp_flags = GFP_KERNEL | __GFP_ZERO | __GFP_NOWARN | __GFP_COMP | | |
6925 | __GFP_NORETRY; | |
6926 | ||
6927 | return (void *) __get_free_pages(gfp_flags, get_order(size)); | |
6928 | } | |
6929 | ||
75b28aff HV |
6930 | static unsigned long rings_size(unsigned sq_entries, unsigned cq_entries, |
6931 | size_t *sq_offset) | |
6932 | { | |
6933 | struct io_rings *rings; | |
6934 | size_t off, sq_array_size; | |
6935 | ||
6936 | off = struct_size(rings, cqes, cq_entries); | |
6937 | if (off == SIZE_MAX) | |
6938 | return SIZE_MAX; | |
6939 | ||
6940 | #ifdef CONFIG_SMP | |
6941 | off = ALIGN(off, SMP_CACHE_BYTES); | |
6942 | if (off == 0) | |
6943 | return SIZE_MAX; | |
6944 | #endif | |
6945 | ||
6946 | sq_array_size = array_size(sizeof(u32), sq_entries); | |
6947 | if (sq_array_size == SIZE_MAX) | |
6948 | return SIZE_MAX; | |
6949 | ||
6950 | if (check_add_overflow(off, sq_array_size, &off)) | |
6951 | return SIZE_MAX; | |
6952 | ||
6953 | if (sq_offset) | |
6954 | *sq_offset = off; | |
6955 | ||
6956 | return off; | |
6957 | } | |
6958 | ||
2b188cc1 JA |
6959 | static unsigned long ring_pages(unsigned sq_entries, unsigned cq_entries) |
6960 | { | |
75b28aff | 6961 | size_t pages; |
2b188cc1 | 6962 | |
75b28aff HV |
6963 | pages = (size_t)1 << get_order( |
6964 | rings_size(sq_entries, cq_entries, NULL)); | |
6965 | pages += (size_t)1 << get_order( | |
6966 | array_size(sizeof(struct io_uring_sqe), sq_entries)); | |
2b188cc1 | 6967 | |
75b28aff | 6968 | return pages; |
2b188cc1 JA |
6969 | } |
6970 | ||
edafccee JA |
6971 | static int io_sqe_buffer_unregister(struct io_ring_ctx *ctx) |
6972 | { | |
6973 | int i, j; | |
6974 | ||
6975 | if (!ctx->user_bufs) | |
6976 | return -ENXIO; | |
6977 | ||
6978 | for (i = 0; i < ctx->nr_user_bufs; i++) { | |
6979 | struct io_mapped_ubuf *imu = &ctx->user_bufs[i]; | |
6980 | ||
6981 | for (j = 0; j < imu->nr_bvecs; j++) | |
f1f6a7dd | 6982 | unpin_user_page(imu->bvec[j].bv_page); |
edafccee JA |
6983 | |
6984 | if (ctx->account_mem) | |
6985 | io_unaccount_mem(ctx->user, imu->nr_bvecs); | |
d4ef6475 | 6986 | kvfree(imu->bvec); |
edafccee JA |
6987 | imu->nr_bvecs = 0; |
6988 | } | |
6989 | ||
6990 | kfree(ctx->user_bufs); | |
6991 | ctx->user_bufs = NULL; | |
6992 | ctx->nr_user_bufs = 0; | |
6993 | return 0; | |
6994 | } | |
6995 | ||
6996 | static int io_copy_iov(struct io_ring_ctx *ctx, struct iovec *dst, | |
6997 | void __user *arg, unsigned index) | |
6998 | { | |
6999 | struct iovec __user *src; | |
7000 | ||
7001 | #ifdef CONFIG_COMPAT | |
7002 | if (ctx->compat) { | |
7003 | struct compat_iovec __user *ciovs; | |
7004 | struct compat_iovec ciov; | |
7005 | ||
7006 | ciovs = (struct compat_iovec __user *) arg; | |
7007 | if (copy_from_user(&ciov, &ciovs[index], sizeof(ciov))) | |
7008 | return -EFAULT; | |
7009 | ||
d55e5f5b | 7010 | dst->iov_base = u64_to_user_ptr((u64)ciov.iov_base); |
edafccee JA |
7011 | dst->iov_len = ciov.iov_len; |
7012 | return 0; | |
7013 | } | |
7014 | #endif | |
7015 | src = (struct iovec __user *) arg; | |
7016 | if (copy_from_user(dst, &src[index], sizeof(*dst))) | |
7017 | return -EFAULT; | |
7018 | return 0; | |
7019 | } | |
7020 | ||
7021 | static int io_sqe_buffer_register(struct io_ring_ctx *ctx, void __user *arg, | |
7022 | unsigned nr_args) | |
7023 | { | |
7024 | struct vm_area_struct **vmas = NULL; | |
7025 | struct page **pages = NULL; | |
7026 | int i, j, got_pages = 0; | |
7027 | int ret = -EINVAL; | |
7028 | ||
7029 | if (ctx->user_bufs) | |
7030 | return -EBUSY; | |
7031 | if (!nr_args || nr_args > UIO_MAXIOV) | |
7032 | return -EINVAL; | |
7033 | ||
7034 | ctx->user_bufs = kcalloc(nr_args, sizeof(struct io_mapped_ubuf), | |
7035 | GFP_KERNEL); | |
7036 | if (!ctx->user_bufs) | |
7037 | return -ENOMEM; | |
7038 | ||
7039 | for (i = 0; i < nr_args; i++) { | |
7040 | struct io_mapped_ubuf *imu = &ctx->user_bufs[i]; | |
7041 | unsigned long off, start, end, ubuf; | |
7042 | int pret, nr_pages; | |
7043 | struct iovec iov; | |
7044 | size_t size; | |
7045 | ||
7046 | ret = io_copy_iov(ctx, &iov, arg, i); | |
7047 | if (ret) | |
a278682d | 7048 | goto err; |
edafccee JA |
7049 | |
7050 | /* | |
7051 | * Don't impose further limits on the size and buffer | |
7052 | * constraints here, we'll -EINVAL later when IO is | |
7053 | * submitted if they are wrong. | |
7054 | */ | |
7055 | ret = -EFAULT; | |
7056 | if (!iov.iov_base || !iov.iov_len) | |
7057 | goto err; | |
7058 | ||
7059 | /* arbitrary limit, but we need something */ | |
7060 | if (iov.iov_len > SZ_1G) | |
7061 | goto err; | |
7062 | ||
7063 | ubuf = (unsigned long) iov.iov_base; | |
7064 | end = (ubuf + iov.iov_len + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
7065 | start = ubuf >> PAGE_SHIFT; | |
7066 | nr_pages = end - start; | |
7067 | ||
7068 | if (ctx->account_mem) { | |
7069 | ret = io_account_mem(ctx->user, nr_pages); | |
7070 | if (ret) | |
7071 | goto err; | |
7072 | } | |
7073 | ||
7074 | ret = 0; | |
7075 | if (!pages || nr_pages > got_pages) { | |
7076 | kfree(vmas); | |
7077 | kfree(pages); | |
d4ef6475 | 7078 | pages = kvmalloc_array(nr_pages, sizeof(struct page *), |
edafccee | 7079 | GFP_KERNEL); |
d4ef6475 | 7080 | vmas = kvmalloc_array(nr_pages, |
edafccee JA |
7081 | sizeof(struct vm_area_struct *), |
7082 | GFP_KERNEL); | |
7083 | if (!pages || !vmas) { | |
7084 | ret = -ENOMEM; | |
7085 | if (ctx->account_mem) | |
7086 | io_unaccount_mem(ctx->user, nr_pages); | |
7087 | goto err; | |
7088 | } | |
7089 | got_pages = nr_pages; | |
7090 | } | |
7091 | ||
d4ef6475 | 7092 | imu->bvec = kvmalloc_array(nr_pages, sizeof(struct bio_vec), |
edafccee JA |
7093 | GFP_KERNEL); |
7094 | ret = -ENOMEM; | |
7095 | if (!imu->bvec) { | |
7096 | if (ctx->account_mem) | |
7097 | io_unaccount_mem(ctx->user, nr_pages); | |
7098 | goto err; | |
7099 | } | |
7100 | ||
7101 | ret = 0; | |
7102 | down_read(¤t->mm->mmap_sem); | |
2113b05d | 7103 | pret = pin_user_pages(ubuf, nr_pages, |
932f4a63 IW |
7104 | FOLL_WRITE | FOLL_LONGTERM, |
7105 | pages, vmas); | |
edafccee JA |
7106 | if (pret == nr_pages) { |
7107 | /* don't support file backed memory */ | |
7108 | for (j = 0; j < nr_pages; j++) { | |
7109 | struct vm_area_struct *vma = vmas[j]; | |
7110 | ||
7111 | if (vma->vm_file && | |
7112 | !is_file_hugepages(vma->vm_file)) { | |
7113 | ret = -EOPNOTSUPP; | |
7114 | break; | |
7115 | } | |
7116 | } | |
7117 | } else { | |
7118 | ret = pret < 0 ? pret : -EFAULT; | |
7119 | } | |
7120 | up_read(¤t->mm->mmap_sem); | |
7121 | if (ret) { | |
7122 | /* | |
7123 | * if we did partial map, or found file backed vmas, | |
7124 | * release any pages we did get | |
7125 | */ | |
27c4d3a3 | 7126 | if (pret > 0) |
f1f6a7dd | 7127 | unpin_user_pages(pages, pret); |
edafccee JA |
7128 | if (ctx->account_mem) |
7129 | io_unaccount_mem(ctx->user, nr_pages); | |
d4ef6475 | 7130 | kvfree(imu->bvec); |
edafccee JA |
7131 | goto err; |
7132 | } | |
7133 | ||
7134 | off = ubuf & ~PAGE_MASK; | |
7135 | size = iov.iov_len; | |
7136 | for (j = 0; j < nr_pages; j++) { | |
7137 | size_t vec_len; | |
7138 | ||
7139 | vec_len = min_t(size_t, size, PAGE_SIZE - off); | |
7140 | imu->bvec[j].bv_page = pages[j]; | |
7141 | imu->bvec[j].bv_len = vec_len; | |
7142 | imu->bvec[j].bv_offset = off; | |
7143 | off = 0; | |
7144 | size -= vec_len; | |
7145 | } | |
7146 | /* store original address for later verification */ | |
7147 | imu->ubuf = ubuf; | |
7148 | imu->len = iov.iov_len; | |
7149 | imu->nr_bvecs = nr_pages; | |
7150 | ||
7151 | ctx->nr_user_bufs++; | |
7152 | } | |
d4ef6475 MR |
7153 | kvfree(pages); |
7154 | kvfree(vmas); | |
edafccee JA |
7155 | return 0; |
7156 | err: | |
d4ef6475 MR |
7157 | kvfree(pages); |
7158 | kvfree(vmas); | |
edafccee JA |
7159 | io_sqe_buffer_unregister(ctx); |
7160 | return ret; | |
7161 | } | |
7162 | ||
9b402849 JA |
7163 | static int io_eventfd_register(struct io_ring_ctx *ctx, void __user *arg) |
7164 | { | |
7165 | __s32 __user *fds = arg; | |
7166 | int fd; | |
7167 | ||
7168 | if (ctx->cq_ev_fd) | |
7169 | return -EBUSY; | |
7170 | ||
7171 | if (copy_from_user(&fd, fds, sizeof(*fds))) | |
7172 | return -EFAULT; | |
7173 | ||
7174 | ctx->cq_ev_fd = eventfd_ctx_fdget(fd); | |
7175 | if (IS_ERR(ctx->cq_ev_fd)) { | |
7176 | int ret = PTR_ERR(ctx->cq_ev_fd); | |
7177 | ctx->cq_ev_fd = NULL; | |
7178 | return ret; | |
7179 | } | |
7180 | ||
7181 | return 0; | |
7182 | } | |
7183 | ||
7184 | static int io_eventfd_unregister(struct io_ring_ctx *ctx) | |
7185 | { | |
7186 | if (ctx->cq_ev_fd) { | |
7187 | eventfd_ctx_put(ctx->cq_ev_fd); | |
7188 | ctx->cq_ev_fd = NULL; | |
7189 | return 0; | |
7190 | } | |
7191 | ||
7192 | return -ENXIO; | |
7193 | } | |
7194 | ||
5a2e745d JA |
7195 | static int __io_destroy_buffers(int id, void *p, void *data) |
7196 | { | |
7197 | struct io_ring_ctx *ctx = data; | |
7198 | struct io_buffer *buf = p; | |
7199 | ||
067524e9 | 7200 | __io_remove_buffers(ctx, buf, id, -1U); |
5a2e745d JA |
7201 | return 0; |
7202 | } | |
7203 | ||
7204 | static void io_destroy_buffers(struct io_ring_ctx *ctx) | |
7205 | { | |
7206 | idr_for_each(&ctx->io_buffer_idr, __io_destroy_buffers, ctx); | |
7207 | idr_destroy(&ctx->io_buffer_idr); | |
7208 | } | |
7209 | ||
2b188cc1 JA |
7210 | static void io_ring_ctx_free(struct io_ring_ctx *ctx) |
7211 | { | |
6b06314c | 7212 | io_finish_async(ctx); |
2b188cc1 JA |
7213 | if (ctx->sqo_mm) |
7214 | mmdrop(ctx->sqo_mm); | |
def596e9 JA |
7215 | |
7216 | io_iopoll_reap_events(ctx); | |
edafccee | 7217 | io_sqe_buffer_unregister(ctx); |
6b06314c | 7218 | io_sqe_files_unregister(ctx); |
9b402849 | 7219 | io_eventfd_unregister(ctx); |
5a2e745d | 7220 | io_destroy_buffers(ctx); |
41726c9a | 7221 | idr_destroy(&ctx->personality_idr); |
def596e9 | 7222 | |
2b188cc1 | 7223 | #if defined(CONFIG_UNIX) |
355e8d26 EB |
7224 | if (ctx->ring_sock) { |
7225 | ctx->ring_sock->file = NULL; /* so that iput() is called */ | |
2b188cc1 | 7226 | sock_release(ctx->ring_sock); |
355e8d26 | 7227 | } |
2b188cc1 JA |
7228 | #endif |
7229 | ||
75b28aff | 7230 | io_mem_free(ctx->rings); |
2b188cc1 | 7231 | io_mem_free(ctx->sq_sqes); |
2b188cc1 JA |
7232 | |
7233 | percpu_ref_exit(&ctx->refs); | |
7234 | if (ctx->account_mem) | |
7235 | io_unaccount_mem(ctx->user, | |
7236 | ring_pages(ctx->sq_entries, ctx->cq_entries)); | |
7237 | free_uid(ctx->user); | |
181e448d | 7238 | put_cred(ctx->creds); |
78076bb6 | 7239 | kfree(ctx->cancel_hash); |
0ddf92e8 | 7240 | kmem_cache_free(req_cachep, ctx->fallback_req); |
2b188cc1 JA |
7241 | kfree(ctx); |
7242 | } | |
7243 | ||
7244 | static __poll_t io_uring_poll(struct file *file, poll_table *wait) | |
7245 | { | |
7246 | struct io_ring_ctx *ctx = file->private_data; | |
7247 | __poll_t mask = 0; | |
7248 | ||
7249 | poll_wait(file, &ctx->cq_wait, wait); | |
4f7067c3 SB |
7250 | /* |
7251 | * synchronizes with barrier from wq_has_sleeper call in | |
7252 | * io_commit_cqring | |
7253 | */ | |
2b188cc1 | 7254 | smp_rmb(); |
75b28aff HV |
7255 | if (READ_ONCE(ctx->rings->sq.tail) - ctx->cached_sq_head != |
7256 | ctx->rings->sq_ring_entries) | |
2b188cc1 | 7257 | mask |= EPOLLOUT | EPOLLWRNORM; |
63e5d81f | 7258 | if (io_cqring_events(ctx, false)) |
2b188cc1 JA |
7259 | mask |= EPOLLIN | EPOLLRDNORM; |
7260 | ||
7261 | return mask; | |
7262 | } | |
7263 | ||
7264 | static int io_uring_fasync(int fd, struct file *file, int on) | |
7265 | { | |
7266 | struct io_ring_ctx *ctx = file->private_data; | |
7267 | ||
7268 | return fasync_helper(fd, file, on, &ctx->cq_fasync); | |
7269 | } | |
7270 | ||
071698e1 JA |
7271 | static int io_remove_personalities(int id, void *p, void *data) |
7272 | { | |
7273 | struct io_ring_ctx *ctx = data; | |
7274 | const struct cred *cred; | |
7275 | ||
7276 | cred = idr_remove(&ctx->personality_idr, id); | |
7277 | if (cred) | |
7278 | put_cred(cred); | |
7279 | return 0; | |
7280 | } | |
7281 | ||
85faa7b8 JA |
7282 | static void io_ring_exit_work(struct work_struct *work) |
7283 | { | |
7284 | struct io_ring_ctx *ctx; | |
7285 | ||
7286 | ctx = container_of(work, struct io_ring_ctx, exit_work); | |
7287 | if (ctx->rings) | |
7288 | io_cqring_overflow_flush(ctx, true); | |
7289 | ||
0f158b4c | 7290 | wait_for_completion(&ctx->ref_comp); |
85faa7b8 JA |
7291 | io_ring_ctx_free(ctx); |
7292 | } | |
7293 | ||
2b188cc1 JA |
7294 | static void io_ring_ctx_wait_and_kill(struct io_ring_ctx *ctx) |
7295 | { | |
7296 | mutex_lock(&ctx->uring_lock); | |
7297 | percpu_ref_kill(&ctx->refs); | |
7298 | mutex_unlock(&ctx->uring_lock); | |
7299 | ||
df069d80 JA |
7300 | /* |
7301 | * Wait for sq thread to idle, if we have one. It won't spin on new | |
7302 | * work after we've killed the ctx ref above. This is important to do | |
7303 | * before we cancel existing commands, as the thread could otherwise | |
7304 | * be queueing new work post that. If that's work we need to cancel, | |
7305 | * it could cause shutdown to hang. | |
7306 | */ | |
7307 | while (ctx->sqo_thread && !wq_has_sleeper(&ctx->sqo_wait)) | |
3fd44c86 | 7308 | cond_resched(); |
df069d80 | 7309 | |
5262f567 | 7310 | io_kill_timeouts(ctx); |
221c5eb2 | 7311 | io_poll_remove_all(ctx); |
561fb04a JA |
7312 | |
7313 | if (ctx->io_wq) | |
7314 | io_wq_cancel_all(ctx->io_wq); | |
7315 | ||
def596e9 | 7316 | io_iopoll_reap_events(ctx); |
15dff286 JA |
7317 | /* if we failed setting up the ctx, we might not have any rings */ |
7318 | if (ctx->rings) | |
7319 | io_cqring_overflow_flush(ctx, true); | |
071698e1 | 7320 | idr_for_each(&ctx->personality_idr, io_remove_personalities, ctx); |
85faa7b8 JA |
7321 | INIT_WORK(&ctx->exit_work, io_ring_exit_work); |
7322 | queue_work(system_wq, &ctx->exit_work); | |
2b188cc1 JA |
7323 | } |
7324 | ||
7325 | static int io_uring_release(struct inode *inode, struct file *file) | |
7326 | { | |
7327 | struct io_ring_ctx *ctx = file->private_data; | |
7328 | ||
7329 | file->private_data = NULL; | |
7330 | io_ring_ctx_wait_and_kill(ctx); | |
7331 | return 0; | |
7332 | } | |
7333 | ||
fcb323cc JA |
7334 | static void io_uring_cancel_files(struct io_ring_ctx *ctx, |
7335 | struct files_struct *files) | |
7336 | { | |
fcb323cc | 7337 | while (!list_empty_careful(&ctx->inflight_list)) { |
d8f1b971 XW |
7338 | struct io_kiocb *cancel_req = NULL, *req; |
7339 | DEFINE_WAIT(wait); | |
fcb323cc JA |
7340 | |
7341 | spin_lock_irq(&ctx->inflight_lock); | |
7342 | list_for_each_entry(req, &ctx->inflight_list, inflight_entry) { | |
768134d4 JA |
7343 | if (req->work.files != files) |
7344 | continue; | |
7345 | /* req is being completed, ignore */ | |
7346 | if (!refcount_inc_not_zero(&req->refs)) | |
7347 | continue; | |
7348 | cancel_req = req; | |
7349 | break; | |
fcb323cc | 7350 | } |
768134d4 | 7351 | if (cancel_req) |
fcb323cc | 7352 | prepare_to_wait(&ctx->inflight_wait, &wait, |
768134d4 | 7353 | TASK_UNINTERRUPTIBLE); |
fcb323cc JA |
7354 | spin_unlock_irq(&ctx->inflight_lock); |
7355 | ||
768134d4 JA |
7356 | /* We need to keep going until we don't find a matching req */ |
7357 | if (!cancel_req) | |
fcb323cc | 7358 | break; |
2f6d9b9d | 7359 | |
2ca10259 JA |
7360 | if (cancel_req->flags & REQ_F_OVERFLOW) { |
7361 | spin_lock_irq(&ctx->completion_lock); | |
7362 | list_del(&cancel_req->list); | |
7363 | cancel_req->flags &= ~REQ_F_OVERFLOW; | |
7364 | if (list_empty(&ctx->cq_overflow_list)) { | |
7365 | clear_bit(0, &ctx->sq_check_overflow); | |
7366 | clear_bit(0, &ctx->cq_check_overflow); | |
7367 | } | |
7368 | spin_unlock_irq(&ctx->completion_lock); | |
7369 | ||
7370 | WRITE_ONCE(ctx->rings->cq_overflow, | |
7371 | atomic_inc_return(&ctx->cached_cq_overflow)); | |
7372 | ||
7373 | /* | |
7374 | * Put inflight ref and overflow ref. If that's | |
7375 | * all we had, then we're done with this request. | |
7376 | */ | |
7377 | if (refcount_sub_and_test(2, &cancel_req->refs)) { | |
7378 | io_put_req(cancel_req); | |
d8f1b971 | 7379 | finish_wait(&ctx->inflight_wait, &wait); |
2ca10259 JA |
7380 | continue; |
7381 | } | |
7382 | } | |
7383 | ||
2f6d9b9d BL |
7384 | io_wq_cancel_work(ctx->io_wq, &cancel_req->work); |
7385 | io_put_req(cancel_req); | |
fcb323cc | 7386 | schedule(); |
d8f1b971 | 7387 | finish_wait(&ctx->inflight_wait, &wait); |
fcb323cc JA |
7388 | } |
7389 | } | |
7390 | ||
7391 | static int io_uring_flush(struct file *file, void *data) | |
7392 | { | |
7393 | struct io_ring_ctx *ctx = file->private_data; | |
7394 | ||
7395 | io_uring_cancel_files(ctx, data); | |
6ab23144 JA |
7396 | |
7397 | /* | |
7398 | * If the task is going away, cancel work it may have pending | |
7399 | */ | |
7400 | if (fatal_signal_pending(current) || (current->flags & PF_EXITING)) | |
7401 | io_wq_cancel_pid(ctx->io_wq, task_pid_vnr(current)); | |
7402 | ||
fcb323cc JA |
7403 | return 0; |
7404 | } | |
7405 | ||
6c5c240e RP |
7406 | static void *io_uring_validate_mmap_request(struct file *file, |
7407 | loff_t pgoff, size_t sz) | |
2b188cc1 | 7408 | { |
2b188cc1 | 7409 | struct io_ring_ctx *ctx = file->private_data; |
6c5c240e | 7410 | loff_t offset = pgoff << PAGE_SHIFT; |
2b188cc1 JA |
7411 | struct page *page; |
7412 | void *ptr; | |
7413 | ||
7414 | switch (offset) { | |
7415 | case IORING_OFF_SQ_RING: | |
75b28aff HV |
7416 | case IORING_OFF_CQ_RING: |
7417 | ptr = ctx->rings; | |
2b188cc1 JA |
7418 | break; |
7419 | case IORING_OFF_SQES: | |
7420 | ptr = ctx->sq_sqes; | |
7421 | break; | |
2b188cc1 | 7422 | default: |
6c5c240e | 7423 | return ERR_PTR(-EINVAL); |
2b188cc1 JA |
7424 | } |
7425 | ||
7426 | page = virt_to_head_page(ptr); | |
a50b854e | 7427 | if (sz > page_size(page)) |
6c5c240e RP |
7428 | return ERR_PTR(-EINVAL); |
7429 | ||
7430 | return ptr; | |
7431 | } | |
7432 | ||
7433 | #ifdef CONFIG_MMU | |
7434 | ||
7435 | static int io_uring_mmap(struct file *file, struct vm_area_struct *vma) | |
7436 | { | |
7437 | size_t sz = vma->vm_end - vma->vm_start; | |
7438 | unsigned long pfn; | |
7439 | void *ptr; | |
7440 | ||
7441 | ptr = io_uring_validate_mmap_request(file, vma->vm_pgoff, sz); | |
7442 | if (IS_ERR(ptr)) | |
7443 | return PTR_ERR(ptr); | |
2b188cc1 JA |
7444 | |
7445 | pfn = virt_to_phys(ptr) >> PAGE_SHIFT; | |
7446 | return remap_pfn_range(vma, vma->vm_start, pfn, sz, vma->vm_page_prot); | |
7447 | } | |
7448 | ||
6c5c240e RP |
7449 | #else /* !CONFIG_MMU */ |
7450 | ||
7451 | static int io_uring_mmap(struct file *file, struct vm_area_struct *vma) | |
7452 | { | |
7453 | return vma->vm_flags & (VM_SHARED | VM_MAYSHARE) ? 0 : -EINVAL; | |
7454 | } | |
7455 | ||
7456 | static unsigned int io_uring_nommu_mmap_capabilities(struct file *file) | |
7457 | { | |
7458 | return NOMMU_MAP_DIRECT | NOMMU_MAP_READ | NOMMU_MAP_WRITE; | |
7459 | } | |
7460 | ||
7461 | static unsigned long io_uring_nommu_get_unmapped_area(struct file *file, | |
7462 | unsigned long addr, unsigned long len, | |
7463 | unsigned long pgoff, unsigned long flags) | |
7464 | { | |
7465 | void *ptr; | |
7466 | ||
7467 | ptr = io_uring_validate_mmap_request(file, pgoff, len); | |
7468 | if (IS_ERR(ptr)) | |
7469 | return PTR_ERR(ptr); | |
7470 | ||
7471 | return (unsigned long) ptr; | |
7472 | } | |
7473 | ||
7474 | #endif /* !CONFIG_MMU */ | |
7475 | ||
2b188cc1 JA |
7476 | SYSCALL_DEFINE6(io_uring_enter, unsigned int, fd, u32, to_submit, |
7477 | u32, min_complete, u32, flags, const sigset_t __user *, sig, | |
7478 | size_t, sigsz) | |
7479 | { | |
7480 | struct io_ring_ctx *ctx; | |
7481 | long ret = -EBADF; | |
7482 | int submitted = 0; | |
7483 | struct fd f; | |
7484 | ||
b41e9852 JA |
7485 | if (current->task_works) |
7486 | task_work_run(); | |
7487 | ||
6c271ce2 | 7488 | if (flags & ~(IORING_ENTER_GETEVENTS | IORING_ENTER_SQ_WAKEUP)) |
2b188cc1 JA |
7489 | return -EINVAL; |
7490 | ||
7491 | f = fdget(fd); | |
7492 | if (!f.file) | |
7493 | return -EBADF; | |
7494 | ||
7495 | ret = -EOPNOTSUPP; | |
7496 | if (f.file->f_op != &io_uring_fops) | |
7497 | goto out_fput; | |
7498 | ||
7499 | ret = -ENXIO; | |
7500 | ctx = f.file->private_data; | |
7501 | if (!percpu_ref_tryget(&ctx->refs)) | |
7502 | goto out_fput; | |
7503 | ||
6c271ce2 JA |
7504 | /* |
7505 | * For SQ polling, the thread will do all submissions and completions. | |
7506 | * Just return the requested submit count, and wake the thread if | |
7507 | * we were asked to. | |
7508 | */ | |
b2a9eada | 7509 | ret = 0; |
6c271ce2 | 7510 | if (ctx->flags & IORING_SETUP_SQPOLL) { |
c1edbf5f JA |
7511 | if (!list_empty_careful(&ctx->cq_overflow_list)) |
7512 | io_cqring_overflow_flush(ctx, false); | |
6c271ce2 JA |
7513 | if (flags & IORING_ENTER_SQ_WAKEUP) |
7514 | wake_up(&ctx->sqo_wait); | |
7515 | submitted = to_submit; | |
b2a9eada | 7516 | } else if (to_submit) { |
2b188cc1 | 7517 | mutex_lock(&ctx->uring_lock); |
bf9c2f1c | 7518 | submitted = io_submit_sqes(ctx, to_submit, f.file, fd, false); |
2b188cc1 | 7519 | mutex_unlock(&ctx->uring_lock); |
7c504e65 PB |
7520 | |
7521 | if (submitted != to_submit) | |
7522 | goto out; | |
2b188cc1 JA |
7523 | } |
7524 | if (flags & IORING_ENTER_GETEVENTS) { | |
def596e9 JA |
7525 | unsigned nr_events = 0; |
7526 | ||
2b188cc1 JA |
7527 | min_complete = min(min_complete, ctx->cq_entries); |
7528 | ||
32b2244a XW |
7529 | /* |
7530 | * When SETUP_IOPOLL and SETUP_SQPOLL are both enabled, user | |
7531 | * space applications don't need to do io completion events | |
7532 | * polling again, they can rely on io_sq_thread to do polling | |
7533 | * work, which can reduce cpu usage and uring_lock contention. | |
7534 | */ | |
7535 | if (ctx->flags & IORING_SETUP_IOPOLL && | |
7536 | !(ctx->flags & IORING_SETUP_SQPOLL)) { | |
def596e9 | 7537 | ret = io_iopoll_check(ctx, &nr_events, min_complete); |
def596e9 JA |
7538 | } else { |
7539 | ret = io_cqring_wait(ctx, min_complete, sig, sigsz); | |
7540 | } | |
2b188cc1 JA |
7541 | } |
7542 | ||
7c504e65 | 7543 | out: |
6805b32e | 7544 | percpu_ref_put(&ctx->refs); |
2b188cc1 JA |
7545 | out_fput: |
7546 | fdput(f); | |
7547 | return submitted ? submitted : ret; | |
7548 | } | |
7549 | ||
bebdb65e | 7550 | #ifdef CONFIG_PROC_FS |
87ce955b JA |
7551 | static int io_uring_show_cred(int id, void *p, void *data) |
7552 | { | |
7553 | const struct cred *cred = p; | |
7554 | struct seq_file *m = data; | |
7555 | struct user_namespace *uns = seq_user_ns(m); | |
7556 | struct group_info *gi; | |
7557 | kernel_cap_t cap; | |
7558 | unsigned __capi; | |
7559 | int g; | |
7560 | ||
7561 | seq_printf(m, "%5d\n", id); | |
7562 | seq_put_decimal_ull(m, "\tUid:\t", from_kuid_munged(uns, cred->uid)); | |
7563 | seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->euid)); | |
7564 | seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->suid)); | |
7565 | seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->fsuid)); | |
7566 | seq_put_decimal_ull(m, "\n\tGid:\t", from_kgid_munged(uns, cred->gid)); | |
7567 | seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->egid)); | |
7568 | seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->sgid)); | |
7569 | seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->fsgid)); | |
7570 | seq_puts(m, "\n\tGroups:\t"); | |
7571 | gi = cred->group_info; | |
7572 | for (g = 0; g < gi->ngroups; g++) { | |
7573 | seq_put_decimal_ull(m, g ? " " : "", | |
7574 | from_kgid_munged(uns, gi->gid[g])); | |
7575 | } | |
7576 | seq_puts(m, "\n\tCapEff:\t"); | |
7577 | cap = cred->cap_effective; | |
7578 | CAP_FOR_EACH_U32(__capi) | |
7579 | seq_put_hex_ll(m, NULL, cap.cap[CAP_LAST_U32 - __capi], 8); | |
7580 | seq_putc(m, '\n'); | |
7581 | return 0; | |
7582 | } | |
7583 | ||
7584 | static void __io_uring_show_fdinfo(struct io_ring_ctx *ctx, struct seq_file *m) | |
7585 | { | |
7586 | int i; | |
7587 | ||
7588 | mutex_lock(&ctx->uring_lock); | |
7589 | seq_printf(m, "UserFiles:\t%u\n", ctx->nr_user_files); | |
7590 | for (i = 0; i < ctx->nr_user_files; i++) { | |
7591 | struct fixed_file_table *table; | |
7592 | struct file *f; | |
7593 | ||
7594 | table = &ctx->file_data->table[i >> IORING_FILE_TABLE_SHIFT]; | |
7595 | f = table->files[i & IORING_FILE_TABLE_MASK]; | |
7596 | if (f) | |
7597 | seq_printf(m, "%5u: %s\n", i, file_dentry(f)->d_iname); | |
7598 | else | |
7599 | seq_printf(m, "%5u: <none>\n", i); | |
7600 | } | |
7601 | seq_printf(m, "UserBufs:\t%u\n", ctx->nr_user_bufs); | |
7602 | for (i = 0; i < ctx->nr_user_bufs; i++) { | |
7603 | struct io_mapped_ubuf *buf = &ctx->user_bufs[i]; | |
7604 | ||
7605 | seq_printf(m, "%5u: 0x%llx/%u\n", i, buf->ubuf, | |
7606 | (unsigned int) buf->len); | |
7607 | } | |
7608 | if (!idr_is_empty(&ctx->personality_idr)) { | |
7609 | seq_printf(m, "Personalities:\n"); | |
7610 | idr_for_each(&ctx->personality_idr, io_uring_show_cred, m); | |
7611 | } | |
d7718a9d JA |
7612 | seq_printf(m, "PollList:\n"); |
7613 | spin_lock_irq(&ctx->completion_lock); | |
7614 | for (i = 0; i < (1U << ctx->cancel_hash_bits); i++) { | |
7615 | struct hlist_head *list = &ctx->cancel_hash[i]; | |
7616 | struct io_kiocb *req; | |
7617 | ||
7618 | hlist_for_each_entry(req, list, hash_node) | |
7619 | seq_printf(m, " op=%d, task_works=%d\n", req->opcode, | |
7620 | req->task->task_works != NULL); | |
7621 | } | |
7622 | spin_unlock_irq(&ctx->completion_lock); | |
87ce955b JA |
7623 | mutex_unlock(&ctx->uring_lock); |
7624 | } | |
7625 | ||
7626 | static void io_uring_show_fdinfo(struct seq_file *m, struct file *f) | |
7627 | { | |
7628 | struct io_ring_ctx *ctx = f->private_data; | |
7629 | ||
7630 | if (percpu_ref_tryget(&ctx->refs)) { | |
7631 | __io_uring_show_fdinfo(ctx, m); | |
7632 | percpu_ref_put(&ctx->refs); | |
7633 | } | |
7634 | } | |
bebdb65e | 7635 | #endif |
87ce955b | 7636 | |
2b188cc1 JA |
7637 | static const struct file_operations io_uring_fops = { |
7638 | .release = io_uring_release, | |
fcb323cc | 7639 | .flush = io_uring_flush, |
2b188cc1 | 7640 | .mmap = io_uring_mmap, |
6c5c240e RP |
7641 | #ifndef CONFIG_MMU |
7642 | .get_unmapped_area = io_uring_nommu_get_unmapped_area, | |
7643 | .mmap_capabilities = io_uring_nommu_mmap_capabilities, | |
7644 | #endif | |
2b188cc1 JA |
7645 | .poll = io_uring_poll, |
7646 | .fasync = io_uring_fasync, | |
bebdb65e | 7647 | #ifdef CONFIG_PROC_FS |
87ce955b | 7648 | .show_fdinfo = io_uring_show_fdinfo, |
bebdb65e | 7649 | #endif |
2b188cc1 JA |
7650 | }; |
7651 | ||
7652 | static int io_allocate_scq_urings(struct io_ring_ctx *ctx, | |
7653 | struct io_uring_params *p) | |
7654 | { | |
75b28aff HV |
7655 | struct io_rings *rings; |
7656 | size_t size, sq_array_offset; | |
2b188cc1 | 7657 | |
75b28aff HV |
7658 | size = rings_size(p->sq_entries, p->cq_entries, &sq_array_offset); |
7659 | if (size == SIZE_MAX) | |
7660 | return -EOVERFLOW; | |
7661 | ||
7662 | rings = io_mem_alloc(size); | |
7663 | if (!rings) | |
2b188cc1 JA |
7664 | return -ENOMEM; |
7665 | ||
75b28aff HV |
7666 | ctx->rings = rings; |
7667 | ctx->sq_array = (u32 *)((char *)rings + sq_array_offset); | |
7668 | rings->sq_ring_mask = p->sq_entries - 1; | |
7669 | rings->cq_ring_mask = p->cq_entries - 1; | |
7670 | rings->sq_ring_entries = p->sq_entries; | |
7671 | rings->cq_ring_entries = p->cq_entries; | |
7672 | ctx->sq_mask = rings->sq_ring_mask; | |
7673 | ctx->cq_mask = rings->cq_ring_mask; | |
7674 | ctx->sq_entries = rings->sq_ring_entries; | |
7675 | ctx->cq_entries = rings->cq_ring_entries; | |
2b188cc1 JA |
7676 | |
7677 | size = array_size(sizeof(struct io_uring_sqe), p->sq_entries); | |
eb065d30 JA |
7678 | if (size == SIZE_MAX) { |
7679 | io_mem_free(ctx->rings); | |
7680 | ctx->rings = NULL; | |
2b188cc1 | 7681 | return -EOVERFLOW; |
eb065d30 | 7682 | } |
2b188cc1 JA |
7683 | |
7684 | ctx->sq_sqes = io_mem_alloc(size); | |
eb065d30 JA |
7685 | if (!ctx->sq_sqes) { |
7686 | io_mem_free(ctx->rings); | |
7687 | ctx->rings = NULL; | |
2b188cc1 | 7688 | return -ENOMEM; |
eb065d30 | 7689 | } |
2b188cc1 | 7690 | |
2b188cc1 JA |
7691 | return 0; |
7692 | } | |
7693 | ||
7694 | /* | |
7695 | * Allocate an anonymous fd, this is what constitutes the application | |
7696 | * visible backing of an io_uring instance. The application mmaps this | |
7697 | * fd to gain access to the SQ/CQ ring details. If UNIX sockets are enabled, | |
7698 | * we have to tie this fd to a socket for file garbage collection purposes. | |
7699 | */ | |
7700 | static int io_uring_get_fd(struct io_ring_ctx *ctx) | |
7701 | { | |
7702 | struct file *file; | |
7703 | int ret; | |
7704 | ||
7705 | #if defined(CONFIG_UNIX) | |
7706 | ret = sock_create_kern(&init_net, PF_UNIX, SOCK_RAW, IPPROTO_IP, | |
7707 | &ctx->ring_sock); | |
7708 | if (ret) | |
7709 | return ret; | |
7710 | #endif | |
7711 | ||
7712 | ret = get_unused_fd_flags(O_RDWR | O_CLOEXEC); | |
7713 | if (ret < 0) | |
7714 | goto err; | |
7715 | ||
7716 | file = anon_inode_getfile("[io_uring]", &io_uring_fops, ctx, | |
7717 | O_RDWR | O_CLOEXEC); | |
7718 | if (IS_ERR(file)) { | |
7719 | put_unused_fd(ret); | |
7720 | ret = PTR_ERR(file); | |
7721 | goto err; | |
7722 | } | |
7723 | ||
7724 | #if defined(CONFIG_UNIX) | |
7725 | ctx->ring_sock->file = file; | |
7726 | #endif | |
7727 | fd_install(ret, file); | |
7728 | return ret; | |
7729 | err: | |
7730 | #if defined(CONFIG_UNIX) | |
7731 | sock_release(ctx->ring_sock); | |
7732 | ctx->ring_sock = NULL; | |
7733 | #endif | |
7734 | return ret; | |
7735 | } | |
7736 | ||
7f13657d XW |
7737 | static int io_uring_create(unsigned entries, struct io_uring_params *p, |
7738 | struct io_uring_params __user *params) | |
2b188cc1 JA |
7739 | { |
7740 | struct user_struct *user = NULL; | |
7741 | struct io_ring_ctx *ctx; | |
7742 | bool account_mem; | |
7743 | int ret; | |
7744 | ||
8110c1a6 | 7745 | if (!entries) |
2b188cc1 | 7746 | return -EINVAL; |
8110c1a6 JA |
7747 | if (entries > IORING_MAX_ENTRIES) { |
7748 | if (!(p->flags & IORING_SETUP_CLAMP)) | |
7749 | return -EINVAL; | |
7750 | entries = IORING_MAX_ENTRIES; | |
7751 | } | |
2b188cc1 JA |
7752 | |
7753 | /* | |
7754 | * Use twice as many entries for the CQ ring. It's possible for the | |
7755 | * application to drive a higher depth than the size of the SQ ring, | |
7756 | * since the sqes are only used at submission time. This allows for | |
33a107f0 JA |
7757 | * some flexibility in overcommitting a bit. If the application has |
7758 | * set IORING_SETUP_CQSIZE, it will have passed in the desired number | |
7759 | * of CQ ring entries manually. | |
2b188cc1 JA |
7760 | */ |
7761 | p->sq_entries = roundup_pow_of_two(entries); | |
33a107f0 JA |
7762 | if (p->flags & IORING_SETUP_CQSIZE) { |
7763 | /* | |
7764 | * If IORING_SETUP_CQSIZE is set, we do the same roundup | |
7765 | * to a power-of-two, if it isn't already. We do NOT impose | |
7766 | * any cq vs sq ring sizing. | |
7767 | */ | |
8110c1a6 | 7768 | if (p->cq_entries < p->sq_entries) |
33a107f0 | 7769 | return -EINVAL; |
8110c1a6 JA |
7770 | if (p->cq_entries > IORING_MAX_CQ_ENTRIES) { |
7771 | if (!(p->flags & IORING_SETUP_CLAMP)) | |
7772 | return -EINVAL; | |
7773 | p->cq_entries = IORING_MAX_CQ_ENTRIES; | |
7774 | } | |
33a107f0 JA |
7775 | p->cq_entries = roundup_pow_of_two(p->cq_entries); |
7776 | } else { | |
7777 | p->cq_entries = 2 * p->sq_entries; | |
7778 | } | |
2b188cc1 JA |
7779 | |
7780 | user = get_uid(current_user()); | |
7781 | account_mem = !capable(CAP_IPC_LOCK); | |
7782 | ||
7783 | if (account_mem) { | |
7784 | ret = io_account_mem(user, | |
7785 | ring_pages(p->sq_entries, p->cq_entries)); | |
7786 | if (ret) { | |
7787 | free_uid(user); | |
7788 | return ret; | |
7789 | } | |
7790 | } | |
7791 | ||
7792 | ctx = io_ring_ctx_alloc(p); | |
7793 | if (!ctx) { | |
7794 | if (account_mem) | |
7795 | io_unaccount_mem(user, ring_pages(p->sq_entries, | |
7796 | p->cq_entries)); | |
7797 | free_uid(user); | |
7798 | return -ENOMEM; | |
7799 | } | |
7800 | ctx->compat = in_compat_syscall(); | |
7801 | ctx->account_mem = account_mem; | |
7802 | ctx->user = user; | |
0b8c0ec7 | 7803 | ctx->creds = get_current_cred(); |
2b188cc1 JA |
7804 | |
7805 | ret = io_allocate_scq_urings(ctx, p); | |
7806 | if (ret) | |
7807 | goto err; | |
7808 | ||
6c271ce2 | 7809 | ret = io_sq_offload_start(ctx, p); |
2b188cc1 JA |
7810 | if (ret) |
7811 | goto err; | |
7812 | ||
2b188cc1 | 7813 | memset(&p->sq_off, 0, sizeof(p->sq_off)); |
75b28aff HV |
7814 | p->sq_off.head = offsetof(struct io_rings, sq.head); |
7815 | p->sq_off.tail = offsetof(struct io_rings, sq.tail); | |
7816 | p->sq_off.ring_mask = offsetof(struct io_rings, sq_ring_mask); | |
7817 | p->sq_off.ring_entries = offsetof(struct io_rings, sq_ring_entries); | |
7818 | p->sq_off.flags = offsetof(struct io_rings, sq_flags); | |
7819 | p->sq_off.dropped = offsetof(struct io_rings, sq_dropped); | |
7820 | p->sq_off.array = (char *)ctx->sq_array - (char *)ctx->rings; | |
2b188cc1 JA |
7821 | |
7822 | memset(&p->cq_off, 0, sizeof(p->cq_off)); | |
75b28aff HV |
7823 | p->cq_off.head = offsetof(struct io_rings, cq.head); |
7824 | p->cq_off.tail = offsetof(struct io_rings, cq.tail); | |
7825 | p->cq_off.ring_mask = offsetof(struct io_rings, cq_ring_mask); | |
7826 | p->cq_off.ring_entries = offsetof(struct io_rings, cq_ring_entries); | |
7827 | p->cq_off.overflow = offsetof(struct io_rings, cq_overflow); | |
7828 | p->cq_off.cqes = offsetof(struct io_rings, cqes); | |
ac90f249 | 7829 | |
7f13657d XW |
7830 | p->features = IORING_FEAT_SINGLE_MMAP | IORING_FEAT_NODROP | |
7831 | IORING_FEAT_SUBMIT_STABLE | IORING_FEAT_RW_CUR_POS | | |
7832 | IORING_FEAT_CUR_PERSONALITY | IORING_FEAT_FAST_POLL; | |
7833 | ||
7834 | if (copy_to_user(params, p, sizeof(*p))) { | |
7835 | ret = -EFAULT; | |
7836 | goto err; | |
7837 | } | |
044c1ab3 JA |
7838 | /* |
7839 | * Install ring fd as the very last thing, so we don't risk someone | |
7840 | * having closed it before we finish setup | |
7841 | */ | |
7842 | ret = io_uring_get_fd(ctx); | |
7843 | if (ret < 0) | |
7844 | goto err; | |
7845 | ||
c826bd7a | 7846 | trace_io_uring_create(ret, ctx, p->sq_entries, p->cq_entries, p->flags); |
2b188cc1 JA |
7847 | return ret; |
7848 | err: | |
7849 | io_ring_ctx_wait_and_kill(ctx); | |
7850 | return ret; | |
7851 | } | |
7852 | ||
7853 | /* | |
7854 | * Sets up an aio uring context, and returns the fd. Applications asks for a | |
7855 | * ring size, we return the actual sq/cq ring sizes (among other things) in the | |
7856 | * params structure passed in. | |
7857 | */ | |
7858 | static long io_uring_setup(u32 entries, struct io_uring_params __user *params) | |
7859 | { | |
7860 | struct io_uring_params p; | |
2b188cc1 JA |
7861 | int i; |
7862 | ||
7863 | if (copy_from_user(&p, params, sizeof(p))) | |
7864 | return -EFAULT; | |
7865 | for (i = 0; i < ARRAY_SIZE(p.resv); i++) { | |
7866 | if (p.resv[i]) | |
7867 | return -EINVAL; | |
7868 | } | |
7869 | ||
6c271ce2 | 7870 | if (p.flags & ~(IORING_SETUP_IOPOLL | IORING_SETUP_SQPOLL | |
8110c1a6 | 7871 | IORING_SETUP_SQ_AFF | IORING_SETUP_CQSIZE | |
24369c2e | 7872 | IORING_SETUP_CLAMP | IORING_SETUP_ATTACH_WQ)) |
2b188cc1 JA |
7873 | return -EINVAL; |
7874 | ||
7f13657d | 7875 | return io_uring_create(entries, &p, params); |
2b188cc1 JA |
7876 | } |
7877 | ||
7878 | SYSCALL_DEFINE2(io_uring_setup, u32, entries, | |
7879 | struct io_uring_params __user *, params) | |
7880 | { | |
7881 | return io_uring_setup(entries, params); | |
7882 | } | |
7883 | ||
66f4af93 JA |
7884 | static int io_probe(struct io_ring_ctx *ctx, void __user *arg, unsigned nr_args) |
7885 | { | |
7886 | struct io_uring_probe *p; | |
7887 | size_t size; | |
7888 | int i, ret; | |
7889 | ||
7890 | size = struct_size(p, ops, nr_args); | |
7891 | if (size == SIZE_MAX) | |
7892 | return -EOVERFLOW; | |
7893 | p = kzalloc(size, GFP_KERNEL); | |
7894 | if (!p) | |
7895 | return -ENOMEM; | |
7896 | ||
7897 | ret = -EFAULT; | |
7898 | if (copy_from_user(p, arg, size)) | |
7899 | goto out; | |
7900 | ret = -EINVAL; | |
7901 | if (memchr_inv(p, 0, size)) | |
7902 | goto out; | |
7903 | ||
7904 | p->last_op = IORING_OP_LAST - 1; | |
7905 | if (nr_args > IORING_OP_LAST) | |
7906 | nr_args = IORING_OP_LAST; | |
7907 | ||
7908 | for (i = 0; i < nr_args; i++) { | |
7909 | p->ops[i].op = i; | |
7910 | if (!io_op_defs[i].not_supported) | |
7911 | p->ops[i].flags = IO_URING_OP_SUPPORTED; | |
7912 | } | |
7913 | p->ops_len = i; | |
7914 | ||
7915 | ret = 0; | |
7916 | if (copy_to_user(arg, p, size)) | |
7917 | ret = -EFAULT; | |
7918 | out: | |
7919 | kfree(p); | |
7920 | return ret; | |
7921 | } | |
7922 | ||
071698e1 JA |
7923 | static int io_register_personality(struct io_ring_ctx *ctx) |
7924 | { | |
7925 | const struct cred *creds = get_current_cred(); | |
7926 | int id; | |
7927 | ||
7928 | id = idr_alloc_cyclic(&ctx->personality_idr, (void *) creds, 1, | |
7929 | USHRT_MAX, GFP_KERNEL); | |
7930 | if (id < 0) | |
7931 | put_cred(creds); | |
7932 | return id; | |
7933 | } | |
7934 | ||
7935 | static int io_unregister_personality(struct io_ring_ctx *ctx, unsigned id) | |
7936 | { | |
7937 | const struct cred *old_creds; | |
7938 | ||
7939 | old_creds = idr_remove(&ctx->personality_idr, id); | |
7940 | if (old_creds) { | |
7941 | put_cred(old_creds); | |
7942 | return 0; | |
7943 | } | |
7944 | ||
7945 | return -EINVAL; | |
7946 | } | |
7947 | ||
7948 | static bool io_register_op_must_quiesce(int op) | |
7949 | { | |
7950 | switch (op) { | |
7951 | case IORING_UNREGISTER_FILES: | |
7952 | case IORING_REGISTER_FILES_UPDATE: | |
7953 | case IORING_REGISTER_PROBE: | |
7954 | case IORING_REGISTER_PERSONALITY: | |
7955 | case IORING_UNREGISTER_PERSONALITY: | |
7956 | return false; | |
7957 | default: | |
7958 | return true; | |
7959 | } | |
7960 | } | |
7961 | ||
edafccee JA |
7962 | static int __io_uring_register(struct io_ring_ctx *ctx, unsigned opcode, |
7963 | void __user *arg, unsigned nr_args) | |
b19062a5 JA |
7964 | __releases(ctx->uring_lock) |
7965 | __acquires(ctx->uring_lock) | |
edafccee JA |
7966 | { |
7967 | int ret; | |
7968 | ||
35fa71a0 JA |
7969 | /* |
7970 | * We're inside the ring mutex, if the ref is already dying, then | |
7971 | * someone else killed the ctx or is already going through | |
7972 | * io_uring_register(). | |
7973 | */ | |
7974 | if (percpu_ref_is_dying(&ctx->refs)) | |
7975 | return -ENXIO; | |
7976 | ||
071698e1 | 7977 | if (io_register_op_must_quiesce(opcode)) { |
05f3fb3c | 7978 | percpu_ref_kill(&ctx->refs); |
b19062a5 | 7979 | |
05f3fb3c JA |
7980 | /* |
7981 | * Drop uring mutex before waiting for references to exit. If | |
7982 | * another thread is currently inside io_uring_enter() it might | |
7983 | * need to grab the uring_lock to make progress. If we hold it | |
7984 | * here across the drain wait, then we can deadlock. It's safe | |
7985 | * to drop the mutex here, since no new references will come in | |
7986 | * after we've killed the percpu ref. | |
7987 | */ | |
7988 | mutex_unlock(&ctx->uring_lock); | |
0f158b4c | 7989 | ret = wait_for_completion_interruptible(&ctx->ref_comp); |
05f3fb3c | 7990 | mutex_lock(&ctx->uring_lock); |
c150368b JA |
7991 | if (ret) { |
7992 | percpu_ref_resurrect(&ctx->refs); | |
7993 | ret = -EINTR; | |
7994 | goto out; | |
7995 | } | |
05f3fb3c | 7996 | } |
edafccee JA |
7997 | |
7998 | switch (opcode) { | |
7999 | case IORING_REGISTER_BUFFERS: | |
8000 | ret = io_sqe_buffer_register(ctx, arg, nr_args); | |
8001 | break; | |
8002 | case IORING_UNREGISTER_BUFFERS: | |
8003 | ret = -EINVAL; | |
8004 | if (arg || nr_args) | |
8005 | break; | |
8006 | ret = io_sqe_buffer_unregister(ctx); | |
8007 | break; | |
6b06314c JA |
8008 | case IORING_REGISTER_FILES: |
8009 | ret = io_sqe_files_register(ctx, arg, nr_args); | |
8010 | break; | |
8011 | case IORING_UNREGISTER_FILES: | |
8012 | ret = -EINVAL; | |
8013 | if (arg || nr_args) | |
8014 | break; | |
8015 | ret = io_sqe_files_unregister(ctx); | |
8016 | break; | |
c3a31e60 JA |
8017 | case IORING_REGISTER_FILES_UPDATE: |
8018 | ret = io_sqe_files_update(ctx, arg, nr_args); | |
8019 | break; | |
9b402849 | 8020 | case IORING_REGISTER_EVENTFD: |
f2842ab5 | 8021 | case IORING_REGISTER_EVENTFD_ASYNC: |
9b402849 JA |
8022 | ret = -EINVAL; |
8023 | if (nr_args != 1) | |
8024 | break; | |
8025 | ret = io_eventfd_register(ctx, arg); | |
f2842ab5 JA |
8026 | if (ret) |
8027 | break; | |
8028 | if (opcode == IORING_REGISTER_EVENTFD_ASYNC) | |
8029 | ctx->eventfd_async = 1; | |
8030 | else | |
8031 | ctx->eventfd_async = 0; | |
9b402849 JA |
8032 | break; |
8033 | case IORING_UNREGISTER_EVENTFD: | |
8034 | ret = -EINVAL; | |
8035 | if (arg || nr_args) | |
8036 | break; | |
8037 | ret = io_eventfd_unregister(ctx); | |
8038 | break; | |
66f4af93 JA |
8039 | case IORING_REGISTER_PROBE: |
8040 | ret = -EINVAL; | |
8041 | if (!arg || nr_args > 256) | |
8042 | break; | |
8043 | ret = io_probe(ctx, arg, nr_args); | |
8044 | break; | |
071698e1 JA |
8045 | case IORING_REGISTER_PERSONALITY: |
8046 | ret = -EINVAL; | |
8047 | if (arg || nr_args) | |
8048 | break; | |
8049 | ret = io_register_personality(ctx); | |
8050 | break; | |
8051 | case IORING_UNREGISTER_PERSONALITY: | |
8052 | ret = -EINVAL; | |
8053 | if (arg) | |
8054 | break; | |
8055 | ret = io_unregister_personality(ctx, nr_args); | |
8056 | break; | |
edafccee JA |
8057 | default: |
8058 | ret = -EINVAL; | |
8059 | break; | |
8060 | } | |
8061 | ||
071698e1 | 8062 | if (io_register_op_must_quiesce(opcode)) { |
05f3fb3c | 8063 | /* bring the ctx back to life */ |
05f3fb3c | 8064 | percpu_ref_reinit(&ctx->refs); |
c150368b | 8065 | out: |
0f158b4c | 8066 | reinit_completion(&ctx->ref_comp); |
05f3fb3c | 8067 | } |
edafccee JA |
8068 | return ret; |
8069 | } | |
8070 | ||
8071 | SYSCALL_DEFINE4(io_uring_register, unsigned int, fd, unsigned int, opcode, | |
8072 | void __user *, arg, unsigned int, nr_args) | |
8073 | { | |
8074 | struct io_ring_ctx *ctx; | |
8075 | long ret = -EBADF; | |
8076 | struct fd f; | |
8077 | ||
8078 | f = fdget(fd); | |
8079 | if (!f.file) | |
8080 | return -EBADF; | |
8081 | ||
8082 | ret = -EOPNOTSUPP; | |
8083 | if (f.file->f_op != &io_uring_fops) | |
8084 | goto out_fput; | |
8085 | ||
8086 | ctx = f.file->private_data; | |
8087 | ||
8088 | mutex_lock(&ctx->uring_lock); | |
8089 | ret = __io_uring_register(ctx, opcode, arg, nr_args); | |
8090 | mutex_unlock(&ctx->uring_lock); | |
c826bd7a DD |
8091 | trace_io_uring_register(ctx, opcode, ctx->nr_user_files, ctx->nr_user_bufs, |
8092 | ctx->cq_ev_fd != NULL, ret); | |
edafccee JA |
8093 | out_fput: |
8094 | fdput(f); | |
8095 | return ret; | |
8096 | } | |
8097 | ||
2b188cc1 JA |
8098 | static int __init io_uring_init(void) |
8099 | { | |
d7f62e82 SM |
8100 | #define __BUILD_BUG_VERIFY_ELEMENT(stype, eoffset, etype, ename) do { \ |
8101 | BUILD_BUG_ON(offsetof(stype, ename) != eoffset); \ | |
8102 | BUILD_BUG_ON(sizeof(etype) != sizeof_field(stype, ename)); \ | |
8103 | } while (0) | |
8104 | ||
8105 | #define BUILD_BUG_SQE_ELEM(eoffset, etype, ename) \ | |
8106 | __BUILD_BUG_VERIFY_ELEMENT(struct io_uring_sqe, eoffset, etype, ename) | |
8107 | BUILD_BUG_ON(sizeof(struct io_uring_sqe) != 64); | |
8108 | BUILD_BUG_SQE_ELEM(0, __u8, opcode); | |
8109 | BUILD_BUG_SQE_ELEM(1, __u8, flags); | |
8110 | BUILD_BUG_SQE_ELEM(2, __u16, ioprio); | |
8111 | BUILD_BUG_SQE_ELEM(4, __s32, fd); | |
8112 | BUILD_BUG_SQE_ELEM(8, __u64, off); | |
8113 | BUILD_BUG_SQE_ELEM(8, __u64, addr2); | |
8114 | BUILD_BUG_SQE_ELEM(16, __u64, addr); | |
7d67af2c | 8115 | BUILD_BUG_SQE_ELEM(16, __u64, splice_off_in); |
d7f62e82 SM |
8116 | BUILD_BUG_SQE_ELEM(24, __u32, len); |
8117 | BUILD_BUG_SQE_ELEM(28, __kernel_rwf_t, rw_flags); | |
8118 | BUILD_BUG_SQE_ELEM(28, /* compat */ int, rw_flags); | |
8119 | BUILD_BUG_SQE_ELEM(28, /* compat */ __u32, rw_flags); | |
8120 | BUILD_BUG_SQE_ELEM(28, __u32, fsync_flags); | |
8121 | BUILD_BUG_SQE_ELEM(28, __u16, poll_events); | |
8122 | BUILD_BUG_SQE_ELEM(28, __u32, sync_range_flags); | |
8123 | BUILD_BUG_SQE_ELEM(28, __u32, msg_flags); | |
8124 | BUILD_BUG_SQE_ELEM(28, __u32, timeout_flags); | |
8125 | BUILD_BUG_SQE_ELEM(28, __u32, accept_flags); | |
8126 | BUILD_BUG_SQE_ELEM(28, __u32, cancel_flags); | |
8127 | BUILD_BUG_SQE_ELEM(28, __u32, open_flags); | |
8128 | BUILD_BUG_SQE_ELEM(28, __u32, statx_flags); | |
8129 | BUILD_BUG_SQE_ELEM(28, __u32, fadvise_advice); | |
7d67af2c | 8130 | BUILD_BUG_SQE_ELEM(28, __u32, splice_flags); |
d7f62e82 SM |
8131 | BUILD_BUG_SQE_ELEM(32, __u64, user_data); |
8132 | BUILD_BUG_SQE_ELEM(40, __u16, buf_index); | |
8133 | BUILD_BUG_SQE_ELEM(42, __u16, personality); | |
7d67af2c | 8134 | BUILD_BUG_SQE_ELEM(44, __s32, splice_fd_in); |
d7f62e82 | 8135 | |
d3656344 | 8136 | BUILD_BUG_ON(ARRAY_SIZE(io_op_defs) != IORING_OP_LAST); |
84557871 | 8137 | BUILD_BUG_ON(__REQ_F_LAST_BIT >= 8 * sizeof(int)); |
2b188cc1 JA |
8138 | req_cachep = KMEM_CACHE(io_kiocb, SLAB_HWCACHE_ALIGN | SLAB_PANIC); |
8139 | return 0; | |
8140 | }; | |
8141 | __initcall(io_uring_init); |