]>
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> | |
2b188cc1 JA |
58 | #include <linux/percpu.h> |
59 | #include <linux/slab.h> | |
6c271ce2 | 60 | #include <linux/kthread.h> |
2b188cc1 | 61 | #include <linux/blkdev.h> |
edafccee | 62 | #include <linux/bvec.h> |
2b188cc1 JA |
63 | #include <linux/net.h> |
64 | #include <net/sock.h> | |
65 | #include <net/af_unix.h> | |
6b06314c | 66 | #include <net/scm.h> |
2b188cc1 JA |
67 | #include <linux/anon_inodes.h> |
68 | #include <linux/sched/mm.h> | |
69 | #include <linux/uaccess.h> | |
70 | #include <linux/nospec.h> | |
edafccee JA |
71 | #include <linux/sizes.h> |
72 | #include <linux/hugetlb.h> | |
aa4c3967 | 73 | #include <linux/highmem.h> |
15b71abe JA |
74 | #include <linux/namei.h> |
75 | #include <linux/fsnotify.h> | |
4840e418 | 76 | #include <linux/fadvise.h> |
3e4827b0 | 77 | #include <linux/eventpoll.h> |
ff002b30 | 78 | #include <linux/fs_struct.h> |
7d67af2c | 79 | #include <linux/splice.h> |
b41e9852 | 80 | #include <linux/task_work.h> |
bcf5a063 | 81 | #include <linux/pagemap.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 | 144 | /* |
0d9b5b3a | 145 | * Runtime SQ flags |
1e84b97b SB |
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; |
0d9b5b3a SG |
154 | /* |
155 | * Runtime CQ flags | |
156 | * | |
157 | * Written by the application, shouldn't be modified by the | |
158 | * kernel. | |
159 | */ | |
160 | u32 cq_flags; | |
1e84b97b SB |
161 | /* |
162 | * Number of completion events lost because the queue was full; | |
163 | * this should be avoided by the application by making sure | |
0b4295b5 | 164 | * there are not more requests pending than there is space in |
1e84b97b SB |
165 | * the completion queue. |
166 | * | |
167 | * Written by the kernel, shouldn't be modified by the | |
168 | * application (i.e. get number of "new events" by comparing to | |
169 | * cached value). | |
170 | * | |
171 | * As completion events come in out of order this counter is not | |
172 | * ordered with any other data. | |
173 | */ | |
75b28aff | 174 | u32 cq_overflow; |
1e84b97b SB |
175 | /* |
176 | * Ring buffer of completion events. | |
177 | * | |
178 | * The kernel writes completion events fresh every time they are | |
179 | * produced, so the application is allowed to modify pending | |
180 | * entries. | |
181 | */ | |
75b28aff | 182 | struct io_uring_cqe cqes[] ____cacheline_aligned_in_smp; |
2b188cc1 JA |
183 | }; |
184 | ||
edafccee JA |
185 | struct io_mapped_ubuf { |
186 | u64 ubuf; | |
187 | size_t len; | |
188 | struct bio_vec *bvec; | |
189 | unsigned int nr_bvecs; | |
190 | }; | |
191 | ||
65e19f54 JA |
192 | struct fixed_file_table { |
193 | struct file **files; | |
31b51510 JA |
194 | }; |
195 | ||
05589553 XW |
196 | struct fixed_file_ref_node { |
197 | struct percpu_ref refs; | |
198 | struct list_head node; | |
199 | struct list_head file_list; | |
200 | struct fixed_file_data *file_data; | |
4a38aed2 | 201 | struct llist_node llist; |
05589553 XW |
202 | }; |
203 | ||
05f3fb3c JA |
204 | struct fixed_file_data { |
205 | struct fixed_file_table *table; | |
206 | struct io_ring_ctx *ctx; | |
207 | ||
05589553 | 208 | struct percpu_ref *cur_refs; |
05f3fb3c | 209 | struct percpu_ref refs; |
05f3fb3c | 210 | struct completion done; |
05589553 XW |
211 | struct list_head ref_list; |
212 | spinlock_t lock; | |
05f3fb3c JA |
213 | }; |
214 | ||
5a2e745d JA |
215 | struct io_buffer { |
216 | struct list_head list; | |
217 | __u64 addr; | |
218 | __s32 len; | |
219 | __u16 bid; | |
220 | }; | |
221 | ||
2b188cc1 JA |
222 | struct io_ring_ctx { |
223 | struct { | |
224 | struct percpu_ref refs; | |
225 | } ____cacheline_aligned_in_smp; | |
226 | ||
227 | struct { | |
228 | unsigned int flags; | |
e1d85334 | 229 | unsigned int compat: 1; |
aad5d8da | 230 | unsigned int limit_mem: 1; |
e1d85334 RD |
231 | unsigned int cq_overflow_flushed: 1; |
232 | unsigned int drain_next: 1; | |
233 | unsigned int eventfd_async: 1; | |
2b188cc1 | 234 | |
75b28aff HV |
235 | /* |
236 | * Ring buffer of indices into array of io_uring_sqe, which is | |
237 | * mmapped by the application using the IORING_OFF_SQES offset. | |
238 | * | |
239 | * This indirection could e.g. be used to assign fixed | |
240 | * io_uring_sqe entries to operations and only submit them to | |
241 | * the queue when needed. | |
242 | * | |
243 | * The kernel modifies neither the indices array nor the entries | |
244 | * array. | |
245 | */ | |
246 | u32 *sq_array; | |
2b188cc1 JA |
247 | unsigned cached_sq_head; |
248 | unsigned sq_entries; | |
249 | unsigned sq_mask; | |
6c271ce2 | 250 | unsigned sq_thread_idle; |
498ccd9e | 251 | unsigned cached_sq_dropped; |
206aefde | 252 | atomic_t cached_cq_overflow; |
ad3eb2c8 | 253 | unsigned long sq_check_overflow; |
de0617e4 JA |
254 | |
255 | struct list_head defer_list; | |
5262f567 | 256 | struct list_head timeout_list; |
1d7bb1d5 | 257 | struct list_head cq_overflow_list; |
fcb323cc JA |
258 | |
259 | wait_queue_head_t inflight_wait; | |
ad3eb2c8 | 260 | struct io_uring_sqe *sq_sqes; |
2b188cc1 JA |
261 | } ____cacheline_aligned_in_smp; |
262 | ||
206aefde JA |
263 | struct io_rings *rings; |
264 | ||
2b188cc1 | 265 | /* IO offload */ |
561fb04a | 266 | struct io_wq *io_wq; |
6c271ce2 | 267 | struct task_struct *sqo_thread; /* if using sq thread polling */ |
2aede0e4 JA |
268 | |
269 | /* | |
270 | * For SQPOLL usage - we hold a reference to the parent task, so we | |
271 | * have access to the ->files | |
272 | */ | |
273 | struct task_struct *sqo_task; | |
274 | ||
275 | /* Only used for accounting purposes */ | |
276 | struct mm_struct *mm_account; | |
277 | ||
6c271ce2 | 278 | wait_queue_head_t sqo_wait; |
75b28aff | 279 | |
6b06314c JA |
280 | /* |
281 | * If used, fixed file set. Writers must ensure that ->refs is dead, | |
282 | * readers must ensure that ->refs is alive as long as the file* is | |
283 | * used. Only updated through io_uring_register(2). | |
284 | */ | |
05f3fb3c | 285 | struct fixed_file_data *file_data; |
6b06314c | 286 | unsigned nr_user_files; |
b14cca0c PB |
287 | int ring_fd; |
288 | struct file *ring_file; | |
6b06314c | 289 | |
edafccee JA |
290 | /* if used, fixed mapped user buffers */ |
291 | unsigned nr_user_bufs; | |
292 | struct io_mapped_ubuf *user_bufs; | |
293 | ||
2b188cc1 JA |
294 | struct user_struct *user; |
295 | ||
0b8c0ec7 | 296 | const struct cred *creds; |
181e448d | 297 | |
0f158b4c JA |
298 | struct completion ref_comp; |
299 | struct completion sq_thread_comp; | |
206aefde | 300 | |
0ddf92e8 JA |
301 | /* if all else fails... */ |
302 | struct io_kiocb *fallback_req; | |
303 | ||
206aefde JA |
304 | #if defined(CONFIG_UNIX) |
305 | struct socket *ring_sock; | |
306 | #endif | |
307 | ||
5a2e745d JA |
308 | struct idr io_buffer_idr; |
309 | ||
071698e1 JA |
310 | struct idr personality_idr; |
311 | ||
206aefde JA |
312 | struct { |
313 | unsigned cached_cq_tail; | |
314 | unsigned cq_entries; | |
315 | unsigned cq_mask; | |
316 | atomic_t cq_timeouts; | |
ad3eb2c8 | 317 | unsigned long cq_check_overflow; |
206aefde JA |
318 | struct wait_queue_head cq_wait; |
319 | struct fasync_struct *cq_fasync; | |
320 | struct eventfd_ctx *cq_ev_fd; | |
321 | } ____cacheline_aligned_in_smp; | |
2b188cc1 JA |
322 | |
323 | struct { | |
324 | struct mutex uring_lock; | |
325 | wait_queue_head_t wait; | |
326 | } ____cacheline_aligned_in_smp; | |
327 | ||
328 | struct { | |
329 | spinlock_t completion_lock; | |
e94f141b | 330 | |
def596e9 | 331 | /* |
540e32a0 | 332 | * ->iopoll_list is protected by the ctx->uring_lock for |
def596e9 JA |
333 | * io_uring instances that don't use IORING_SETUP_SQPOLL. |
334 | * For SQPOLL, only the single threaded io_sq_thread() will | |
335 | * manipulate the list, hence no extra locking is needed there. | |
336 | */ | |
540e32a0 | 337 | struct list_head iopoll_list; |
78076bb6 JA |
338 | struct hlist_head *cancel_hash; |
339 | unsigned cancel_hash_bits; | |
e94f141b | 340 | bool poll_multi_file; |
31b51510 | 341 | |
fcb323cc JA |
342 | spinlock_t inflight_lock; |
343 | struct list_head inflight_list; | |
2b188cc1 | 344 | } ____cacheline_aligned_in_smp; |
85faa7b8 | 345 | |
4a38aed2 JA |
346 | struct delayed_work file_put_work; |
347 | struct llist_head file_put_llist; | |
348 | ||
85faa7b8 | 349 | struct work_struct exit_work; |
2b188cc1 JA |
350 | }; |
351 | ||
09bb8394 JA |
352 | /* |
353 | * First field must be the file pointer in all the | |
354 | * iocb unions! See also 'struct kiocb' in <linux/fs.h> | |
355 | */ | |
221c5eb2 JA |
356 | struct io_poll_iocb { |
357 | struct file *file; | |
0969e783 JA |
358 | union { |
359 | struct wait_queue_head *head; | |
360 | u64 addr; | |
361 | }; | |
221c5eb2 | 362 | __poll_t events; |
8c838788 | 363 | bool done; |
221c5eb2 | 364 | bool canceled; |
392edb45 | 365 | struct wait_queue_entry wait; |
221c5eb2 JA |
366 | }; |
367 | ||
b5dba59e JA |
368 | struct io_close { |
369 | struct file *file; | |
370 | struct file *put_file; | |
371 | int fd; | |
372 | }; | |
373 | ||
ad8a48ac JA |
374 | struct io_timeout_data { |
375 | struct io_kiocb *req; | |
376 | struct hrtimer timer; | |
377 | struct timespec64 ts; | |
378 | enum hrtimer_mode mode; | |
379 | }; | |
380 | ||
8ed8d3c3 JA |
381 | struct io_accept { |
382 | struct file *file; | |
383 | struct sockaddr __user *addr; | |
384 | int __user *addr_len; | |
385 | int flags; | |
09952e3e | 386 | unsigned long nofile; |
8ed8d3c3 JA |
387 | }; |
388 | ||
389 | struct io_sync { | |
390 | struct file *file; | |
391 | loff_t len; | |
392 | loff_t off; | |
393 | int flags; | |
d63d1b5e | 394 | int mode; |
8ed8d3c3 JA |
395 | }; |
396 | ||
fbf23849 JA |
397 | struct io_cancel { |
398 | struct file *file; | |
399 | u64 addr; | |
400 | }; | |
401 | ||
b29472ee JA |
402 | struct io_timeout { |
403 | struct file *file; | |
404 | u64 addr; | |
405 | int flags; | |
bfe68a22 PB |
406 | u32 off; |
407 | u32 target_seq; | |
135fcde8 | 408 | struct list_head list; |
b29472ee JA |
409 | }; |
410 | ||
9adbd45d JA |
411 | struct io_rw { |
412 | /* NOTE: kiocb has the file as the first member, so don't do it here */ | |
413 | struct kiocb kiocb; | |
414 | u64 addr; | |
415 | u64 len; | |
416 | }; | |
417 | ||
3fbb51c1 JA |
418 | struct io_connect { |
419 | struct file *file; | |
420 | struct sockaddr __user *addr; | |
421 | int addr_len; | |
422 | }; | |
423 | ||
e47293fd JA |
424 | struct io_sr_msg { |
425 | struct file *file; | |
fddaface | 426 | union { |
270a5940 | 427 | struct user_msghdr __user *umsg; |
fddaface JA |
428 | void __user *buf; |
429 | }; | |
e47293fd | 430 | int msg_flags; |
bcda7baa | 431 | int bgid; |
fddaface | 432 | size_t len; |
bcda7baa | 433 | struct io_buffer *kbuf; |
e47293fd JA |
434 | }; |
435 | ||
15b71abe JA |
436 | struct io_open { |
437 | struct file *file; | |
438 | int dfd; | |
15b71abe | 439 | struct filename *filename; |
c12cedf2 | 440 | struct open_how how; |
4022e7af | 441 | unsigned long nofile; |
15b71abe JA |
442 | }; |
443 | ||
05f3fb3c JA |
444 | struct io_files_update { |
445 | struct file *file; | |
446 | u64 arg; | |
447 | u32 nr_args; | |
448 | u32 offset; | |
449 | }; | |
450 | ||
4840e418 JA |
451 | struct io_fadvise { |
452 | struct file *file; | |
453 | u64 offset; | |
454 | u32 len; | |
455 | u32 advice; | |
456 | }; | |
457 | ||
c1ca757b JA |
458 | struct io_madvise { |
459 | struct file *file; | |
460 | u64 addr; | |
461 | u32 len; | |
462 | u32 advice; | |
463 | }; | |
464 | ||
3e4827b0 JA |
465 | struct io_epoll { |
466 | struct file *file; | |
467 | int epfd; | |
468 | int op; | |
469 | int fd; | |
470 | struct epoll_event event; | |
e47293fd JA |
471 | }; |
472 | ||
7d67af2c PB |
473 | struct io_splice { |
474 | struct file *file_out; | |
475 | struct file *file_in; | |
476 | loff_t off_out; | |
477 | loff_t off_in; | |
478 | u64 len; | |
479 | unsigned int flags; | |
480 | }; | |
481 | ||
ddf0322d JA |
482 | struct io_provide_buf { |
483 | struct file *file; | |
484 | __u64 addr; | |
485 | __s32 len; | |
486 | __u32 bgid; | |
487 | __u16 nbufs; | |
488 | __u16 bid; | |
489 | }; | |
490 | ||
1d9e1288 BM |
491 | struct io_statx { |
492 | struct file *file; | |
493 | int dfd; | |
494 | unsigned int mask; | |
495 | unsigned int flags; | |
e62753e4 | 496 | const char __user *filename; |
1d9e1288 BM |
497 | struct statx __user *buffer; |
498 | }; | |
499 | ||
3ca405eb PB |
500 | struct io_completion { |
501 | struct file *file; | |
502 | struct list_head list; | |
0f7e466b | 503 | int cflags; |
3ca405eb PB |
504 | }; |
505 | ||
f499a021 JA |
506 | struct io_async_connect { |
507 | struct sockaddr_storage address; | |
508 | }; | |
509 | ||
03b1230c JA |
510 | struct io_async_msghdr { |
511 | struct iovec fast_iov[UIO_FASTIOV]; | |
512 | struct iovec *iov; | |
513 | struct sockaddr __user *uaddr; | |
514 | struct msghdr msg; | |
b537916c | 515 | struct sockaddr_storage addr; |
03b1230c JA |
516 | }; |
517 | ||
f67676d1 JA |
518 | struct io_async_rw { |
519 | struct iovec fast_iov[UIO_FASTIOV]; | |
ff6165b2 JA |
520 | const struct iovec *free_iovec; |
521 | struct iov_iter iter; | |
227c0c96 | 522 | size_t bytes_done; |
bcf5a063 | 523 | struct wait_page_queue wpq; |
f67676d1 JA |
524 | }; |
525 | ||
1a6b74fc | 526 | struct io_async_ctx { |
f67676d1 JA |
527 | union { |
528 | struct io_async_rw rw; | |
03b1230c | 529 | struct io_async_msghdr msg; |
f499a021 | 530 | struct io_async_connect connect; |
2d28390a | 531 | struct io_timeout_data timeout; |
f67676d1 | 532 | }; |
1a6b74fc JA |
533 | }; |
534 | ||
6b47ee6e PB |
535 | enum { |
536 | REQ_F_FIXED_FILE_BIT = IOSQE_FIXED_FILE_BIT, | |
537 | REQ_F_IO_DRAIN_BIT = IOSQE_IO_DRAIN_BIT, | |
538 | REQ_F_LINK_BIT = IOSQE_IO_LINK_BIT, | |
539 | REQ_F_HARDLINK_BIT = IOSQE_IO_HARDLINK_BIT, | |
540 | REQ_F_FORCE_ASYNC_BIT = IOSQE_ASYNC_BIT, | |
bcda7baa | 541 | REQ_F_BUFFER_SELECT_BIT = IOSQE_BUFFER_SELECT_BIT, |
6b47ee6e | 542 | |
dea3b49c | 543 | REQ_F_LINK_HEAD_BIT, |
6b47ee6e PB |
544 | REQ_F_FAIL_LINK_BIT, |
545 | REQ_F_INFLIGHT_BIT, | |
546 | REQ_F_CUR_POS_BIT, | |
547 | REQ_F_NOWAIT_BIT, | |
6b47ee6e | 548 | REQ_F_LINK_TIMEOUT_BIT, |
6b47ee6e | 549 | REQ_F_ISREG_BIT, |
6b47ee6e | 550 | REQ_F_COMP_LOCKED_BIT, |
99bc4c38 | 551 | REQ_F_NEED_CLEANUP_BIT, |
d7718a9d | 552 | REQ_F_POLLED_BIT, |
bcda7baa | 553 | REQ_F_BUFFER_SELECTED_BIT, |
5b0bbee4 | 554 | REQ_F_NO_FILE_TABLE_BIT, |
7cdaf587 | 555 | REQ_F_WORK_INITIALIZED_BIT, |
84557871 JA |
556 | |
557 | /* not a real bit, just to check we're not overflowing the space */ | |
558 | __REQ_F_LAST_BIT, | |
6b47ee6e PB |
559 | }; |
560 | ||
561 | enum { | |
562 | /* ctx owns file */ | |
563 | REQ_F_FIXED_FILE = BIT(REQ_F_FIXED_FILE_BIT), | |
564 | /* drain existing IO first */ | |
565 | REQ_F_IO_DRAIN = BIT(REQ_F_IO_DRAIN_BIT), | |
566 | /* linked sqes */ | |
567 | REQ_F_LINK = BIT(REQ_F_LINK_BIT), | |
568 | /* doesn't sever on completion < 0 */ | |
569 | REQ_F_HARDLINK = BIT(REQ_F_HARDLINK_BIT), | |
570 | /* IOSQE_ASYNC */ | |
571 | REQ_F_FORCE_ASYNC = BIT(REQ_F_FORCE_ASYNC_BIT), | |
bcda7baa JA |
572 | /* IOSQE_BUFFER_SELECT */ |
573 | REQ_F_BUFFER_SELECT = BIT(REQ_F_BUFFER_SELECT_BIT), | |
6b47ee6e | 574 | |
dea3b49c PB |
575 | /* head of a link */ |
576 | REQ_F_LINK_HEAD = BIT(REQ_F_LINK_HEAD_BIT), | |
6b47ee6e PB |
577 | /* fail rest of links */ |
578 | REQ_F_FAIL_LINK = BIT(REQ_F_FAIL_LINK_BIT), | |
579 | /* on inflight list */ | |
580 | REQ_F_INFLIGHT = BIT(REQ_F_INFLIGHT_BIT), | |
581 | /* read/write uses file position */ | |
582 | REQ_F_CUR_POS = BIT(REQ_F_CUR_POS_BIT), | |
583 | /* must not punt to workers */ | |
584 | REQ_F_NOWAIT = BIT(REQ_F_NOWAIT_BIT), | |
6b47ee6e PB |
585 | /* has linked timeout */ |
586 | REQ_F_LINK_TIMEOUT = BIT(REQ_F_LINK_TIMEOUT_BIT), | |
6b47ee6e PB |
587 | /* regular file */ |
588 | REQ_F_ISREG = BIT(REQ_F_ISREG_BIT), | |
6b47ee6e PB |
589 | /* completion under lock */ |
590 | REQ_F_COMP_LOCKED = BIT(REQ_F_COMP_LOCKED_BIT), | |
99bc4c38 PB |
591 | /* needs cleanup */ |
592 | REQ_F_NEED_CLEANUP = BIT(REQ_F_NEED_CLEANUP_BIT), | |
d7718a9d JA |
593 | /* already went through poll handler */ |
594 | REQ_F_POLLED = BIT(REQ_F_POLLED_BIT), | |
bcda7baa JA |
595 | /* buffer already selected */ |
596 | REQ_F_BUFFER_SELECTED = BIT(REQ_F_BUFFER_SELECTED_BIT), | |
5b0bbee4 JA |
597 | /* doesn't need file table for this request */ |
598 | REQ_F_NO_FILE_TABLE = BIT(REQ_F_NO_FILE_TABLE_BIT), | |
7cdaf587 XW |
599 | /* io_wq_work is initialized */ |
600 | REQ_F_WORK_INITIALIZED = BIT(REQ_F_WORK_INITIALIZED_BIT), | |
d7718a9d JA |
601 | }; |
602 | ||
603 | struct async_poll { | |
604 | struct io_poll_iocb poll; | |
807abcb0 | 605 | struct io_poll_iocb *double_poll; |
6b47ee6e PB |
606 | }; |
607 | ||
09bb8394 JA |
608 | /* |
609 | * NOTE! Each of the iocb union members has the file pointer | |
610 | * as the first entry in their struct definition. So you can | |
611 | * access the file pointer through any of the sub-structs, | |
612 | * or directly as just 'ki_filp' in this struct. | |
613 | */ | |
2b188cc1 | 614 | struct io_kiocb { |
221c5eb2 | 615 | union { |
09bb8394 | 616 | struct file *file; |
9adbd45d | 617 | struct io_rw rw; |
221c5eb2 | 618 | struct io_poll_iocb poll; |
8ed8d3c3 JA |
619 | struct io_accept accept; |
620 | struct io_sync sync; | |
fbf23849 | 621 | struct io_cancel cancel; |
b29472ee | 622 | struct io_timeout timeout; |
3fbb51c1 | 623 | struct io_connect connect; |
e47293fd | 624 | struct io_sr_msg sr_msg; |
15b71abe | 625 | struct io_open open; |
b5dba59e | 626 | struct io_close close; |
05f3fb3c | 627 | struct io_files_update files_update; |
4840e418 | 628 | struct io_fadvise fadvise; |
c1ca757b | 629 | struct io_madvise madvise; |
3e4827b0 | 630 | struct io_epoll epoll; |
7d67af2c | 631 | struct io_splice splice; |
ddf0322d | 632 | struct io_provide_buf pbuf; |
1d9e1288 | 633 | struct io_statx statx; |
3ca405eb PB |
634 | /* use only after cleaning per-op data, see io_clean_op() */ |
635 | struct io_completion compl; | |
221c5eb2 | 636 | }; |
2b188cc1 | 637 | |
1a6b74fc | 638 | struct io_async_ctx *io; |
d625c6ee | 639 | u8 opcode; |
65a6543d XW |
640 | /* polled IO has completed */ |
641 | u8 iopoll_completed; | |
2b188cc1 | 642 | |
4f4eeba8 | 643 | u16 buf_index; |
9cf7c104 | 644 | u32 result; |
4f4eeba8 | 645 | |
010e8e6b PB |
646 | struct io_ring_ctx *ctx; |
647 | unsigned int flags; | |
648 | refcount_t refs; | |
649 | struct task_struct *task; | |
650 | u64 user_data; | |
d7718a9d | 651 | |
010e8e6b | 652 | struct list_head link_list; |
fcb323cc | 653 | |
d21ffe7e PB |
654 | /* |
655 | * 1. used with ctx->iopoll_list with reads/writes | |
656 | * 2. to track reqs with ->files (see io_op_def::file_table) | |
657 | */ | |
010e8e6b PB |
658 | struct list_head inflight_entry; |
659 | ||
660 | struct percpu_ref *fixed_file_refs; | |
661 | struct callback_head task_work; | |
662 | /* for polled requests, i.e. IORING_OP_POLL_ADD and async armed poll */ | |
663 | struct hlist_node hash_node; | |
664 | struct async_poll *apoll; | |
665 | struct io_wq_work work; | |
2b188cc1 | 666 | }; |
05589553 | 667 | |
27dc8338 PB |
668 | struct io_defer_entry { |
669 | struct list_head list; | |
670 | struct io_kiocb *req; | |
9cf7c104 | 671 | u32 seq; |
2b188cc1 JA |
672 | }; |
673 | ||
def596e9 | 674 | #define IO_IOPOLL_BATCH 8 |
2b188cc1 | 675 | |
013538bd JA |
676 | struct io_comp_state { |
677 | unsigned int nr; | |
678 | struct list_head list; | |
679 | struct io_ring_ctx *ctx; | |
680 | }; | |
681 | ||
9a56a232 JA |
682 | struct io_submit_state { |
683 | struct blk_plug plug; | |
684 | ||
2579f913 JA |
685 | /* |
686 | * io_kiocb alloc cache | |
687 | */ | |
688 | void *reqs[IO_IOPOLL_BATCH]; | |
6c8a3134 | 689 | unsigned int free_reqs; |
2579f913 | 690 | |
013538bd JA |
691 | /* |
692 | * Batch completion logic | |
693 | */ | |
694 | struct io_comp_state comp; | |
695 | ||
9a56a232 JA |
696 | /* |
697 | * File reference cache | |
698 | */ | |
699 | struct file *file; | |
700 | unsigned int fd; | |
701 | unsigned int has_refs; | |
9a56a232 JA |
702 | unsigned int ios_left; |
703 | }; | |
704 | ||
d3656344 JA |
705 | struct io_op_def { |
706 | /* needs req->io allocated for deferral/async */ | |
707 | unsigned async_ctx : 1; | |
708 | /* needs current->mm setup, does mm access */ | |
709 | unsigned needs_mm : 1; | |
710 | /* needs req->file assigned */ | |
711 | unsigned needs_file : 1; | |
fd2206e4 JA |
712 | /* don't fail if file grab fails */ |
713 | unsigned needs_file_no_error : 1; | |
d3656344 JA |
714 | /* hash wq insertion if file is a regular file */ |
715 | unsigned hash_reg_file : 1; | |
716 | /* unbound wq insertion if file is a non-regular file */ | |
717 | unsigned unbound_nonreg_file : 1; | |
66f4af93 JA |
718 | /* opcode is not supported by this kernel */ |
719 | unsigned not_supported : 1; | |
f86cd20c JA |
720 | /* needs file table */ |
721 | unsigned file_table : 1; | |
ff002b30 JA |
722 | /* needs ->fs */ |
723 | unsigned needs_fs : 1; | |
8a72758c JA |
724 | /* set if opcode supports polled "wait" */ |
725 | unsigned pollin : 1; | |
726 | unsigned pollout : 1; | |
bcda7baa JA |
727 | /* op supports buffer selection */ |
728 | unsigned buffer_select : 1; | |
57f1a649 | 729 | unsigned needs_fsize : 1; |
d3656344 JA |
730 | }; |
731 | ||
732 | static const struct io_op_def io_op_defs[] = { | |
0463b6c5 PB |
733 | [IORING_OP_NOP] = {}, |
734 | [IORING_OP_READV] = { | |
d3656344 JA |
735 | .async_ctx = 1, |
736 | .needs_mm = 1, | |
737 | .needs_file = 1, | |
738 | .unbound_nonreg_file = 1, | |
8a72758c | 739 | .pollin = 1, |
4d954c25 | 740 | .buffer_select = 1, |
d3656344 | 741 | }, |
0463b6c5 | 742 | [IORING_OP_WRITEV] = { |
d3656344 JA |
743 | .async_ctx = 1, |
744 | .needs_mm = 1, | |
745 | .needs_file = 1, | |
746 | .hash_reg_file = 1, | |
747 | .unbound_nonreg_file = 1, | |
8a72758c | 748 | .pollout = 1, |
57f1a649 | 749 | .needs_fsize = 1, |
d3656344 | 750 | }, |
0463b6c5 | 751 | [IORING_OP_FSYNC] = { |
d3656344 JA |
752 | .needs_file = 1, |
753 | }, | |
0463b6c5 | 754 | [IORING_OP_READ_FIXED] = { |
d3656344 JA |
755 | .needs_file = 1, |
756 | .unbound_nonreg_file = 1, | |
8a72758c | 757 | .pollin = 1, |
d3656344 | 758 | }, |
0463b6c5 | 759 | [IORING_OP_WRITE_FIXED] = { |
d3656344 JA |
760 | .needs_file = 1, |
761 | .hash_reg_file = 1, | |
762 | .unbound_nonreg_file = 1, | |
8a72758c | 763 | .pollout = 1, |
57f1a649 | 764 | .needs_fsize = 1, |
d3656344 | 765 | }, |
0463b6c5 | 766 | [IORING_OP_POLL_ADD] = { |
d3656344 JA |
767 | .needs_file = 1, |
768 | .unbound_nonreg_file = 1, | |
769 | }, | |
0463b6c5 PB |
770 | [IORING_OP_POLL_REMOVE] = {}, |
771 | [IORING_OP_SYNC_FILE_RANGE] = { | |
d3656344 JA |
772 | .needs_file = 1, |
773 | }, | |
0463b6c5 | 774 | [IORING_OP_SENDMSG] = { |
d3656344 JA |
775 | .async_ctx = 1, |
776 | .needs_mm = 1, | |
777 | .needs_file = 1, | |
778 | .unbound_nonreg_file = 1, | |
ff002b30 | 779 | .needs_fs = 1, |
8a72758c | 780 | .pollout = 1, |
d3656344 | 781 | }, |
0463b6c5 | 782 | [IORING_OP_RECVMSG] = { |
d3656344 JA |
783 | .async_ctx = 1, |
784 | .needs_mm = 1, | |
785 | .needs_file = 1, | |
786 | .unbound_nonreg_file = 1, | |
ff002b30 | 787 | .needs_fs = 1, |
8a72758c | 788 | .pollin = 1, |
52de1fe1 | 789 | .buffer_select = 1, |
d3656344 | 790 | }, |
0463b6c5 | 791 | [IORING_OP_TIMEOUT] = { |
d3656344 JA |
792 | .async_ctx = 1, |
793 | .needs_mm = 1, | |
794 | }, | |
0463b6c5 PB |
795 | [IORING_OP_TIMEOUT_REMOVE] = {}, |
796 | [IORING_OP_ACCEPT] = { | |
d3656344 JA |
797 | .needs_mm = 1, |
798 | .needs_file = 1, | |
799 | .unbound_nonreg_file = 1, | |
f86cd20c | 800 | .file_table = 1, |
8a72758c | 801 | .pollin = 1, |
d3656344 | 802 | }, |
0463b6c5 PB |
803 | [IORING_OP_ASYNC_CANCEL] = {}, |
804 | [IORING_OP_LINK_TIMEOUT] = { | |
d3656344 JA |
805 | .async_ctx = 1, |
806 | .needs_mm = 1, | |
807 | }, | |
0463b6c5 | 808 | [IORING_OP_CONNECT] = { |
d3656344 JA |
809 | .async_ctx = 1, |
810 | .needs_mm = 1, | |
811 | .needs_file = 1, | |
812 | .unbound_nonreg_file = 1, | |
8a72758c | 813 | .pollout = 1, |
d3656344 | 814 | }, |
0463b6c5 | 815 | [IORING_OP_FALLOCATE] = { |
d3656344 | 816 | .needs_file = 1, |
57f1a649 | 817 | .needs_fsize = 1, |
d3656344 | 818 | }, |
0463b6c5 | 819 | [IORING_OP_OPENAT] = { |
f86cd20c | 820 | .file_table = 1, |
ff002b30 | 821 | .needs_fs = 1, |
d3656344 | 822 | }, |
0463b6c5 | 823 | [IORING_OP_CLOSE] = { |
fd2206e4 JA |
824 | .needs_file = 1, |
825 | .needs_file_no_error = 1, | |
f86cd20c | 826 | .file_table = 1, |
d3656344 | 827 | }, |
0463b6c5 | 828 | [IORING_OP_FILES_UPDATE] = { |
d3656344 | 829 | .needs_mm = 1, |
f86cd20c | 830 | .file_table = 1, |
d3656344 | 831 | }, |
0463b6c5 | 832 | [IORING_OP_STATX] = { |
d3656344 | 833 | .needs_mm = 1, |
ff002b30 | 834 | .needs_fs = 1, |
5b0bbee4 | 835 | .file_table = 1, |
d3656344 | 836 | }, |
0463b6c5 | 837 | [IORING_OP_READ] = { |
3a6820f2 JA |
838 | .needs_mm = 1, |
839 | .needs_file = 1, | |
840 | .unbound_nonreg_file = 1, | |
8a72758c | 841 | .pollin = 1, |
bcda7baa | 842 | .buffer_select = 1, |
3a6820f2 | 843 | }, |
0463b6c5 | 844 | [IORING_OP_WRITE] = { |
3a6820f2 JA |
845 | .needs_mm = 1, |
846 | .needs_file = 1, | |
847 | .unbound_nonreg_file = 1, | |
8a72758c | 848 | .pollout = 1, |
57f1a649 | 849 | .needs_fsize = 1, |
3a6820f2 | 850 | }, |
0463b6c5 | 851 | [IORING_OP_FADVISE] = { |
4840e418 JA |
852 | .needs_file = 1, |
853 | }, | |
0463b6c5 | 854 | [IORING_OP_MADVISE] = { |
c1ca757b JA |
855 | .needs_mm = 1, |
856 | }, | |
0463b6c5 | 857 | [IORING_OP_SEND] = { |
fddaface JA |
858 | .needs_mm = 1, |
859 | .needs_file = 1, | |
860 | .unbound_nonreg_file = 1, | |
8a72758c | 861 | .pollout = 1, |
fddaface | 862 | }, |
0463b6c5 | 863 | [IORING_OP_RECV] = { |
fddaface JA |
864 | .needs_mm = 1, |
865 | .needs_file = 1, | |
866 | .unbound_nonreg_file = 1, | |
8a72758c | 867 | .pollin = 1, |
bcda7baa | 868 | .buffer_select = 1, |
fddaface | 869 | }, |
0463b6c5 | 870 | [IORING_OP_OPENAT2] = { |
f86cd20c | 871 | .file_table = 1, |
ff002b30 | 872 | .needs_fs = 1, |
cebdb986 | 873 | }, |
3e4827b0 JA |
874 | [IORING_OP_EPOLL_CTL] = { |
875 | .unbound_nonreg_file = 1, | |
876 | .file_table = 1, | |
877 | }, | |
7d67af2c PB |
878 | [IORING_OP_SPLICE] = { |
879 | .needs_file = 1, | |
880 | .hash_reg_file = 1, | |
881 | .unbound_nonreg_file = 1, | |
ddf0322d JA |
882 | }, |
883 | [IORING_OP_PROVIDE_BUFFERS] = {}, | |
067524e9 | 884 | [IORING_OP_REMOVE_BUFFERS] = {}, |
f2a8d5c7 PB |
885 | [IORING_OP_TEE] = { |
886 | .needs_file = 1, | |
887 | .hash_reg_file = 1, | |
888 | .unbound_nonreg_file = 1, | |
889 | }, | |
d3656344 JA |
890 | }; |
891 | ||
2e0464d4 BM |
892 | enum io_mem_account { |
893 | ACCT_LOCKED, | |
894 | ACCT_PINNED, | |
895 | }; | |
896 | ||
81b68a5c PB |
897 | static void __io_complete_rw(struct io_kiocb *req, long res, long res2, |
898 | struct io_comp_state *cs); | |
78e19bbe | 899 | static void io_cqring_fill_event(struct io_kiocb *req, long res); |
ec9c02ad | 900 | static void io_put_req(struct io_kiocb *req); |
c40f6379 | 901 | static void io_double_put_req(struct io_kiocb *req); |
978db57e | 902 | static void __io_double_put_req(struct io_kiocb *req); |
94ae5e77 | 903 | static struct io_kiocb *io_prep_linked_timeout(struct io_kiocb *req); |
7271ef3a | 904 | static void __io_queue_linked_timeout(struct io_kiocb *req); |
94ae5e77 | 905 | static void io_queue_linked_timeout(struct io_kiocb *req); |
05f3fb3c JA |
906 | static int __io_sqe_files_update(struct io_ring_ctx *ctx, |
907 | struct io_uring_files_update *ip, | |
908 | unsigned nr_args); | |
f56040b8 | 909 | static int io_prep_work_files(struct io_kiocb *req); |
3ca405eb | 910 | static void __io_clean_op(struct io_kiocb *req); |
b41e9852 JA |
911 | static int io_file_get(struct io_submit_state *state, struct io_kiocb *req, |
912 | int fd, struct file **out_file, bool fixed); | |
913 | static void __io_queue_sqe(struct io_kiocb *req, | |
f13fad7b JA |
914 | const struct io_uring_sqe *sqe, |
915 | struct io_comp_state *cs); | |
4349f30e | 916 | static void io_file_put_work(struct work_struct *work); |
de0617e4 | 917 | |
b63534c4 JA |
918 | static ssize_t io_import_iovec(int rw, struct io_kiocb *req, |
919 | struct iovec **iovec, struct iov_iter *iter, | |
920 | bool needs_lock); | |
ff6165b2 JA |
921 | static int io_setup_async_rw(struct io_kiocb *req, const struct iovec *iovec, |
922 | const struct iovec *fast_iov, | |
227c0c96 | 923 | struct iov_iter *iter, bool force); |
de0617e4 | 924 | |
2b188cc1 JA |
925 | static struct kmem_cache *req_cachep; |
926 | ||
927 | static const struct file_operations io_uring_fops; | |
928 | ||
929 | struct sock *io_uring_get_socket(struct file *file) | |
930 | { | |
931 | #if defined(CONFIG_UNIX) | |
932 | if (file->f_op == &io_uring_fops) { | |
933 | struct io_ring_ctx *ctx = file->private_data; | |
934 | ||
935 | return ctx->ring_sock->sk; | |
936 | } | |
937 | #endif | |
938 | return NULL; | |
939 | } | |
940 | EXPORT_SYMBOL(io_uring_get_socket); | |
941 | ||
3ca405eb PB |
942 | static inline void io_clean_op(struct io_kiocb *req) |
943 | { | |
bb175342 PB |
944 | if (req->flags & (REQ_F_NEED_CLEANUP | REQ_F_BUFFER_SELECTED | |
945 | REQ_F_INFLIGHT)) | |
3ca405eb PB |
946 | __io_clean_op(req); |
947 | } | |
948 | ||
4349f30e | 949 | static void io_sq_thread_drop_mm(void) |
c40f6379 JA |
950 | { |
951 | struct mm_struct *mm = current->mm; | |
952 | ||
953 | if (mm) { | |
954 | kthread_unuse_mm(mm); | |
955 | mmput(mm); | |
956 | } | |
957 | } | |
958 | ||
959 | static int __io_sq_thread_acquire_mm(struct io_ring_ctx *ctx) | |
960 | { | |
961 | if (!current->mm) { | |
cbcf7214 | 962 | if (unlikely(!(ctx->flags & IORING_SETUP_SQPOLL) || |
2aede0e4 JA |
963 | !ctx->sqo_task->mm || |
964 | !mmget_not_zero(ctx->sqo_task->mm))) | |
c40f6379 | 965 | return -EFAULT; |
2aede0e4 | 966 | kthread_use_mm(ctx->sqo_task->mm); |
c40f6379 JA |
967 | } |
968 | ||
969 | return 0; | |
970 | } | |
971 | ||
972 | static int io_sq_thread_acquire_mm(struct io_ring_ctx *ctx, | |
973 | struct io_kiocb *req) | |
974 | { | |
975 | if (!io_op_defs[req->opcode].needs_mm) | |
976 | return 0; | |
977 | return __io_sq_thread_acquire_mm(ctx); | |
978 | } | |
979 | ||
980 | static inline void req_set_fail_links(struct io_kiocb *req) | |
981 | { | |
982 | if ((req->flags & (REQ_F_LINK | REQ_F_HARDLINK)) == REQ_F_LINK) | |
983 | req->flags |= REQ_F_FAIL_LINK; | |
984 | } | |
4a38aed2 | 985 | |
7cdaf587 XW |
986 | /* |
987 | * Note: must call io_req_init_async() for the first time you | |
988 | * touch any members of io_wq_work. | |
989 | */ | |
990 | static inline void io_req_init_async(struct io_kiocb *req) | |
991 | { | |
992 | if (req->flags & REQ_F_WORK_INITIALIZED) | |
993 | return; | |
994 | ||
995 | memset(&req->work, 0, sizeof(req->work)); | |
996 | req->flags |= REQ_F_WORK_INITIALIZED; | |
997 | } | |
998 | ||
0cdaf760 PB |
999 | static inline bool io_async_submit(struct io_ring_ctx *ctx) |
1000 | { | |
1001 | return ctx->flags & IORING_SETUP_SQPOLL; | |
1002 | } | |
1003 | ||
2b188cc1 JA |
1004 | static void io_ring_ctx_ref_free(struct percpu_ref *ref) |
1005 | { | |
1006 | struct io_ring_ctx *ctx = container_of(ref, struct io_ring_ctx, refs); | |
1007 | ||
0f158b4c | 1008 | complete(&ctx->ref_comp); |
2b188cc1 JA |
1009 | } |
1010 | ||
8eb7e2d0 PB |
1011 | static inline bool io_is_timeout_noseq(struct io_kiocb *req) |
1012 | { | |
1013 | return !req->timeout.off; | |
1014 | } | |
1015 | ||
2b188cc1 JA |
1016 | static struct io_ring_ctx *io_ring_ctx_alloc(struct io_uring_params *p) |
1017 | { | |
1018 | struct io_ring_ctx *ctx; | |
78076bb6 | 1019 | int hash_bits; |
2b188cc1 JA |
1020 | |
1021 | ctx = kzalloc(sizeof(*ctx), GFP_KERNEL); | |
1022 | if (!ctx) | |
1023 | return NULL; | |
1024 | ||
0ddf92e8 JA |
1025 | ctx->fallback_req = kmem_cache_alloc(req_cachep, GFP_KERNEL); |
1026 | if (!ctx->fallback_req) | |
1027 | goto err; | |
1028 | ||
78076bb6 JA |
1029 | /* |
1030 | * Use 5 bits less than the max cq entries, that should give us around | |
1031 | * 32 entries per hash list if totally full and uniformly spread. | |
1032 | */ | |
1033 | hash_bits = ilog2(p->cq_entries); | |
1034 | hash_bits -= 5; | |
1035 | if (hash_bits <= 0) | |
1036 | hash_bits = 1; | |
1037 | ctx->cancel_hash_bits = hash_bits; | |
1038 | ctx->cancel_hash = kmalloc((1U << hash_bits) * sizeof(struct hlist_head), | |
1039 | GFP_KERNEL); | |
1040 | if (!ctx->cancel_hash) | |
1041 | goto err; | |
1042 | __hash_init(ctx->cancel_hash, 1U << hash_bits); | |
1043 | ||
21482896 | 1044 | if (percpu_ref_init(&ctx->refs, io_ring_ctx_ref_free, |
206aefde JA |
1045 | PERCPU_REF_ALLOW_REINIT, GFP_KERNEL)) |
1046 | goto err; | |
2b188cc1 JA |
1047 | |
1048 | ctx->flags = p->flags; | |
583863ed | 1049 | init_waitqueue_head(&ctx->sqo_wait); |
2b188cc1 | 1050 | init_waitqueue_head(&ctx->cq_wait); |
1d7bb1d5 | 1051 | INIT_LIST_HEAD(&ctx->cq_overflow_list); |
0f158b4c JA |
1052 | init_completion(&ctx->ref_comp); |
1053 | init_completion(&ctx->sq_thread_comp); | |
5a2e745d | 1054 | idr_init(&ctx->io_buffer_idr); |
071698e1 | 1055 | idr_init(&ctx->personality_idr); |
2b188cc1 JA |
1056 | mutex_init(&ctx->uring_lock); |
1057 | init_waitqueue_head(&ctx->wait); | |
1058 | spin_lock_init(&ctx->completion_lock); | |
540e32a0 | 1059 | INIT_LIST_HEAD(&ctx->iopoll_list); |
de0617e4 | 1060 | INIT_LIST_HEAD(&ctx->defer_list); |
5262f567 | 1061 | INIT_LIST_HEAD(&ctx->timeout_list); |
fcb323cc JA |
1062 | init_waitqueue_head(&ctx->inflight_wait); |
1063 | spin_lock_init(&ctx->inflight_lock); | |
1064 | INIT_LIST_HEAD(&ctx->inflight_list); | |
4a38aed2 JA |
1065 | INIT_DELAYED_WORK(&ctx->file_put_work, io_file_put_work); |
1066 | init_llist_head(&ctx->file_put_llist); | |
2b188cc1 | 1067 | return ctx; |
206aefde | 1068 | err: |
0ddf92e8 JA |
1069 | if (ctx->fallback_req) |
1070 | kmem_cache_free(req_cachep, ctx->fallback_req); | |
78076bb6 | 1071 | kfree(ctx->cancel_hash); |
206aefde JA |
1072 | kfree(ctx); |
1073 | return NULL; | |
2b188cc1 JA |
1074 | } |
1075 | ||
9cf7c104 | 1076 | static bool req_need_defer(struct io_kiocb *req, u32 seq) |
7adf4eaf | 1077 | { |
2bc9930e JA |
1078 | if (unlikely(req->flags & REQ_F_IO_DRAIN)) { |
1079 | struct io_ring_ctx *ctx = req->ctx; | |
a197f664 | 1080 | |
9cf7c104 | 1081 | return seq != ctx->cached_cq_tail |
31af27c7 | 1082 | + atomic_read(&ctx->cached_cq_overflow); |
2bc9930e | 1083 | } |
de0617e4 | 1084 | |
9d858b21 | 1085 | return false; |
de0617e4 JA |
1086 | } |
1087 | ||
de0617e4 | 1088 | static void __io_commit_cqring(struct io_ring_ctx *ctx) |
2b188cc1 | 1089 | { |
75b28aff | 1090 | struct io_rings *rings = ctx->rings; |
2b188cc1 | 1091 | |
07910158 PB |
1092 | /* order cqe stores with ring update */ |
1093 | smp_store_release(&rings->cq.tail, ctx->cached_cq_tail); | |
2b188cc1 | 1094 | |
07910158 PB |
1095 | if (wq_has_sleeper(&ctx->cq_wait)) { |
1096 | wake_up_interruptible(&ctx->cq_wait); | |
1097 | kill_fasync(&ctx->cq_fasync, SIGIO, POLL_IN); | |
2b188cc1 JA |
1098 | } |
1099 | } | |
1100 | ||
51a4cc11 JA |
1101 | /* |
1102 | * Returns true if we need to defer file table putting. This can only happen | |
1103 | * from the error path with REQ_F_COMP_LOCKED set. | |
1104 | */ | |
1105 | static bool io_req_clean_work(struct io_kiocb *req) | |
18d9be1a | 1106 | { |
7cdaf587 | 1107 | if (!(req->flags & REQ_F_WORK_INITIALIZED)) |
51a4cc11 JA |
1108 | return false; |
1109 | ||
1110 | req->flags &= ~REQ_F_WORK_INITIALIZED; | |
7cdaf587 | 1111 | |
cccf0ee8 JA |
1112 | if (req->work.mm) { |
1113 | mmdrop(req->work.mm); | |
1114 | req->work.mm = NULL; | |
1115 | } | |
1116 | if (req->work.creds) { | |
1117 | put_cred(req->work.creds); | |
1118 | req->work.creds = NULL; | |
1119 | } | |
ff002b30 JA |
1120 | if (req->work.fs) { |
1121 | struct fs_struct *fs = req->work.fs; | |
1122 | ||
51a4cc11 JA |
1123 | if (req->flags & REQ_F_COMP_LOCKED) |
1124 | return true; | |
1125 | ||
ff002b30 JA |
1126 | spin_lock(&req->work.fs->lock); |
1127 | if (--fs->users) | |
1128 | fs = NULL; | |
1129 | spin_unlock(&req->work.fs->lock); | |
1130 | if (fs) | |
1131 | free_fs_struct(fs); | |
b65e0dd6 | 1132 | req->work.fs = NULL; |
ff002b30 | 1133 | } |
51a4cc11 JA |
1134 | |
1135 | return false; | |
561fb04a JA |
1136 | } |
1137 | ||
cbdcb435 | 1138 | static void io_prep_async_work(struct io_kiocb *req) |
18d9be1a | 1139 | { |
d3656344 | 1140 | const struct io_op_def *def = &io_op_defs[req->opcode]; |
54a91f3b | 1141 | |
16d59803 PB |
1142 | io_req_init_async(req); |
1143 | ||
d3656344 | 1144 | if (req->flags & REQ_F_ISREG) { |
eefdf30f | 1145 | if (def->hash_reg_file || (req->ctx->flags & IORING_SETUP_IOPOLL)) |
8766dd51 | 1146 | io_wq_hash_work(&req->work, file_inode(req->file)); |
d3656344 JA |
1147 | } else { |
1148 | if (def->unbound_nonreg_file) | |
3529d8c2 | 1149 | req->work.flags |= IO_WQ_WORK_UNBOUND; |
54a91f3b | 1150 | } |
dca9cf8b PB |
1151 | if (!req->work.mm && def->needs_mm) { |
1152 | mmgrab(current->mm); | |
1153 | req->work.mm = current->mm; | |
1154 | } | |
1155 | if (!req->work.creds) | |
1156 | req->work.creds = get_current_cred(); | |
1157 | if (!req->work.fs && def->needs_fs) { | |
1158 | spin_lock(¤t->fs->lock); | |
1159 | if (!current->fs->in_exec) { | |
1160 | req->work.fs = current->fs; | |
1161 | req->work.fs->users++; | |
1162 | } else { | |
1163 | req->work.flags |= IO_WQ_WORK_CANCEL; | |
1164 | } | |
1165 | spin_unlock(¤t->fs->lock); | |
1166 | } | |
57f1a649 PB |
1167 | if (def->needs_fsize) |
1168 | req->work.fsize = rlimit(RLIMIT_FSIZE); | |
1169 | else | |
1170 | req->work.fsize = RLIM_INFINITY; | |
561fb04a | 1171 | } |
cccf0ee8 | 1172 | |
cbdcb435 | 1173 | static void io_prep_async_link(struct io_kiocb *req) |
561fb04a | 1174 | { |
cbdcb435 | 1175 | struct io_kiocb *cur; |
54a91f3b | 1176 | |
cbdcb435 PB |
1177 | io_prep_async_work(req); |
1178 | if (req->flags & REQ_F_LINK_HEAD) | |
1179 | list_for_each_entry(cur, &req->link_list, link_list) | |
1180 | io_prep_async_work(cur); | |
561fb04a JA |
1181 | } |
1182 | ||
7271ef3a | 1183 | static struct io_kiocb *__io_queue_async_work(struct io_kiocb *req) |
561fb04a | 1184 | { |
a197f664 | 1185 | struct io_ring_ctx *ctx = req->ctx; |
cbdcb435 | 1186 | struct io_kiocb *link = io_prep_linked_timeout(req); |
561fb04a | 1187 | |
8766dd51 PB |
1188 | trace_io_uring_queue_async_work(ctx, io_wq_is_hashed(&req->work), req, |
1189 | &req->work, req->flags); | |
1190 | io_wq_enqueue(ctx->io_wq, &req->work); | |
7271ef3a | 1191 | return link; |
18d9be1a JA |
1192 | } |
1193 | ||
cbdcb435 PB |
1194 | static void io_queue_async_work(struct io_kiocb *req) |
1195 | { | |
7271ef3a JA |
1196 | struct io_kiocb *link; |
1197 | ||
cbdcb435 PB |
1198 | /* init ->work of the whole link before punting */ |
1199 | io_prep_async_link(req); | |
7271ef3a JA |
1200 | link = __io_queue_async_work(req); |
1201 | ||
1202 | if (link) | |
1203 | io_queue_linked_timeout(link); | |
cbdcb435 PB |
1204 | } |
1205 | ||
5262f567 JA |
1206 | static void io_kill_timeout(struct io_kiocb *req) |
1207 | { | |
1208 | int ret; | |
1209 | ||
2d28390a | 1210 | ret = hrtimer_try_to_cancel(&req->io->timeout.timer); |
5262f567 | 1211 | if (ret != -1) { |
01cec8c1 PB |
1212 | atomic_set(&req->ctx->cq_timeouts, |
1213 | atomic_read(&req->ctx->cq_timeouts) + 1); | |
135fcde8 | 1214 | list_del_init(&req->timeout.list); |
f0e20b89 | 1215 | req->flags |= REQ_F_COMP_LOCKED; |
78e19bbe | 1216 | io_cqring_fill_event(req, 0); |
ec9c02ad | 1217 | io_put_req(req); |
5262f567 JA |
1218 | } |
1219 | } | |
1220 | ||
f3606e3a JA |
1221 | static bool io_task_match(struct io_kiocb *req, struct task_struct *tsk) |
1222 | { | |
1223 | struct io_ring_ctx *ctx = req->ctx; | |
1224 | ||
1225 | if (!tsk || req->task == tsk) | |
1226 | return true; | |
1227 | if ((ctx->flags & IORING_SETUP_SQPOLL) && req->task == ctx->sqo_thread) | |
1228 | return true; | |
1229 | return false; | |
1230 | } | |
1231 | ||
76e1b642 JA |
1232 | /* |
1233 | * Returns true if we found and killed one or more timeouts | |
1234 | */ | |
1235 | static bool io_kill_timeouts(struct io_ring_ctx *ctx, struct task_struct *tsk) | |
5262f567 JA |
1236 | { |
1237 | struct io_kiocb *req, *tmp; | |
76e1b642 | 1238 | int canceled = 0; |
5262f567 JA |
1239 | |
1240 | spin_lock_irq(&ctx->completion_lock); | |
f3606e3a | 1241 | list_for_each_entry_safe(req, tmp, &ctx->timeout_list, timeout.list) { |
76e1b642 | 1242 | if (io_task_match(req, tsk)) { |
f3606e3a | 1243 | io_kill_timeout(req); |
76e1b642 JA |
1244 | canceled++; |
1245 | } | |
f3606e3a | 1246 | } |
5262f567 | 1247 | spin_unlock_irq(&ctx->completion_lock); |
76e1b642 | 1248 | return canceled != 0; |
5262f567 JA |
1249 | } |
1250 | ||
04518945 | 1251 | static void __io_queue_deferred(struct io_ring_ctx *ctx) |
de0617e4 | 1252 | { |
04518945 | 1253 | do { |
27dc8338 PB |
1254 | struct io_defer_entry *de = list_first_entry(&ctx->defer_list, |
1255 | struct io_defer_entry, list); | |
7271ef3a | 1256 | struct io_kiocb *link; |
de0617e4 | 1257 | |
9cf7c104 | 1258 | if (req_need_defer(de->req, de->seq)) |
04518945 | 1259 | break; |
27dc8338 | 1260 | list_del_init(&de->list); |
cbdcb435 | 1261 | /* punt-init is done before queueing for defer */ |
7271ef3a JA |
1262 | link = __io_queue_async_work(de->req); |
1263 | if (link) { | |
1264 | __io_queue_linked_timeout(link); | |
1265 | /* drop submission reference */ | |
1266 | link->flags |= REQ_F_COMP_LOCKED; | |
1267 | io_put_req(link); | |
1268 | } | |
27dc8338 | 1269 | kfree(de); |
04518945 PB |
1270 | } while (!list_empty(&ctx->defer_list)); |
1271 | } | |
1272 | ||
360428f8 | 1273 | static void io_flush_timeouts(struct io_ring_ctx *ctx) |
de0617e4 | 1274 | { |
360428f8 PB |
1275 | while (!list_empty(&ctx->timeout_list)) { |
1276 | struct io_kiocb *req = list_first_entry(&ctx->timeout_list, | |
135fcde8 | 1277 | struct io_kiocb, timeout.list); |
de0617e4 | 1278 | |
8eb7e2d0 | 1279 | if (io_is_timeout_noseq(req)) |
360428f8 | 1280 | break; |
bfe68a22 PB |
1281 | if (req->timeout.target_seq != ctx->cached_cq_tail |
1282 | - atomic_read(&ctx->cq_timeouts)) | |
360428f8 | 1283 | break; |
bfe68a22 | 1284 | |
135fcde8 | 1285 | list_del_init(&req->timeout.list); |
5262f567 | 1286 | io_kill_timeout(req); |
360428f8 PB |
1287 | } |
1288 | } | |
5262f567 | 1289 | |
360428f8 PB |
1290 | static void io_commit_cqring(struct io_ring_ctx *ctx) |
1291 | { | |
1292 | io_flush_timeouts(ctx); | |
de0617e4 JA |
1293 | __io_commit_cqring(ctx); |
1294 | ||
04518945 PB |
1295 | if (unlikely(!list_empty(&ctx->defer_list))) |
1296 | __io_queue_deferred(ctx); | |
de0617e4 JA |
1297 | } |
1298 | ||
2b188cc1 JA |
1299 | static struct io_uring_cqe *io_get_cqring(struct io_ring_ctx *ctx) |
1300 | { | |
75b28aff | 1301 | struct io_rings *rings = ctx->rings; |
2b188cc1 JA |
1302 | unsigned tail; |
1303 | ||
1304 | tail = ctx->cached_cq_tail; | |
115e12e5 SB |
1305 | /* |
1306 | * writes to the cq entry need to come after reading head; the | |
1307 | * control dependency is enough as we're using WRITE_ONCE to | |
1308 | * fill the cq entry | |
1309 | */ | |
75b28aff | 1310 | if (tail - READ_ONCE(rings->cq.head) == rings->cq_ring_entries) |
2b188cc1 JA |
1311 | return NULL; |
1312 | ||
1313 | ctx->cached_cq_tail++; | |
75b28aff | 1314 | return &rings->cqes[tail & ctx->cq_mask]; |
2b188cc1 JA |
1315 | } |
1316 | ||
f2842ab5 JA |
1317 | static inline bool io_should_trigger_evfd(struct io_ring_ctx *ctx) |
1318 | { | |
f0b493e6 JA |
1319 | if (!ctx->cq_ev_fd) |
1320 | return false; | |
7e55a19c SG |
1321 | if (READ_ONCE(ctx->rings->cq_flags) & IORING_CQ_EVENTFD_DISABLED) |
1322 | return false; | |
f2842ab5 JA |
1323 | if (!ctx->eventfd_async) |
1324 | return true; | |
b41e9852 | 1325 | return io_wq_current_is_worker(); |
f2842ab5 JA |
1326 | } |
1327 | ||
b41e9852 | 1328 | static void io_cqring_ev_posted(struct io_ring_ctx *ctx) |
1d7bb1d5 JA |
1329 | { |
1330 | if (waitqueue_active(&ctx->wait)) | |
1331 | wake_up(&ctx->wait); | |
1332 | if (waitqueue_active(&ctx->sqo_wait)) | |
1333 | wake_up(&ctx->sqo_wait); | |
b41e9852 | 1334 | if (io_should_trigger_evfd(ctx)) |
1d7bb1d5 JA |
1335 | eventfd_signal(ctx->cq_ev_fd, 1); |
1336 | } | |
1337 | ||
46930143 PB |
1338 | static void io_cqring_mark_overflow(struct io_ring_ctx *ctx) |
1339 | { | |
1340 | if (list_empty(&ctx->cq_overflow_list)) { | |
1341 | clear_bit(0, &ctx->sq_check_overflow); | |
1342 | clear_bit(0, &ctx->cq_check_overflow); | |
1343 | ctx->rings->sq_flags &= ~IORING_SQ_CQ_OVERFLOW; | |
1344 | } | |
1345 | } | |
1346 | ||
c4a2ed72 JA |
1347 | /* Returns true if there are no backlogged entries after the flush */ |
1348 | static bool io_cqring_overflow_flush(struct io_ring_ctx *ctx, bool force) | |
1d7bb1d5 JA |
1349 | { |
1350 | struct io_rings *rings = ctx->rings; | |
1351 | struct io_uring_cqe *cqe; | |
1352 | struct io_kiocb *req; | |
1353 | unsigned long flags; | |
1354 | LIST_HEAD(list); | |
1355 | ||
1356 | if (!force) { | |
1357 | if (list_empty_careful(&ctx->cq_overflow_list)) | |
c4a2ed72 | 1358 | return true; |
1d7bb1d5 JA |
1359 | if ((ctx->cached_cq_tail - READ_ONCE(rings->cq.head) == |
1360 | rings->cq_ring_entries)) | |
c4a2ed72 | 1361 | return false; |
1d7bb1d5 JA |
1362 | } |
1363 | ||
1364 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
1365 | ||
1366 | /* if force is set, the ring is going away. always drop after that */ | |
1367 | if (force) | |
69b3e546 | 1368 | ctx->cq_overflow_flushed = 1; |
1d7bb1d5 | 1369 | |
c4a2ed72 | 1370 | cqe = NULL; |
1d7bb1d5 JA |
1371 | while (!list_empty(&ctx->cq_overflow_list)) { |
1372 | cqe = io_get_cqring(ctx); | |
1373 | if (!cqe && !force) | |
1374 | break; | |
1375 | ||
1376 | req = list_first_entry(&ctx->cq_overflow_list, struct io_kiocb, | |
40d8ddd4 PB |
1377 | compl.list); |
1378 | list_move(&req->compl.list, &list); | |
1d7bb1d5 JA |
1379 | if (cqe) { |
1380 | WRITE_ONCE(cqe->user_data, req->user_data); | |
1381 | WRITE_ONCE(cqe->res, req->result); | |
0f7e466b | 1382 | WRITE_ONCE(cqe->flags, req->compl.cflags); |
1d7bb1d5 JA |
1383 | } else { |
1384 | WRITE_ONCE(ctx->rings->cq_overflow, | |
1385 | atomic_inc_return(&ctx->cached_cq_overflow)); | |
1386 | } | |
1387 | } | |
1388 | ||
1389 | io_commit_cqring(ctx); | |
46930143 PB |
1390 | io_cqring_mark_overflow(ctx); |
1391 | ||
1d7bb1d5 JA |
1392 | spin_unlock_irqrestore(&ctx->completion_lock, flags); |
1393 | io_cqring_ev_posted(ctx); | |
1394 | ||
1395 | while (!list_empty(&list)) { | |
40d8ddd4 PB |
1396 | req = list_first_entry(&list, struct io_kiocb, compl.list); |
1397 | list_del(&req->compl.list); | |
ec9c02ad | 1398 | io_put_req(req); |
1d7bb1d5 | 1399 | } |
c4a2ed72 JA |
1400 | |
1401 | return cqe != NULL; | |
1d7bb1d5 JA |
1402 | } |
1403 | ||
bcda7baa | 1404 | static void __io_cqring_fill_event(struct io_kiocb *req, long res, long cflags) |
2b188cc1 | 1405 | { |
78e19bbe | 1406 | struct io_ring_ctx *ctx = req->ctx; |
2b188cc1 JA |
1407 | struct io_uring_cqe *cqe; |
1408 | ||
78e19bbe | 1409 | trace_io_uring_complete(ctx, req->user_data, res); |
51c3ff62 | 1410 | |
2b188cc1 JA |
1411 | /* |
1412 | * If we can't get a cq entry, userspace overflowed the | |
1413 | * submission (by quite a lot). Increment the overflow count in | |
1414 | * the ring. | |
1415 | */ | |
1416 | cqe = io_get_cqring(ctx); | |
1d7bb1d5 | 1417 | if (likely(cqe)) { |
78e19bbe | 1418 | WRITE_ONCE(cqe->user_data, req->user_data); |
2b188cc1 | 1419 | WRITE_ONCE(cqe->res, res); |
bcda7baa | 1420 | WRITE_ONCE(cqe->flags, cflags); |
1d7bb1d5 | 1421 | } else if (ctx->cq_overflow_flushed) { |
498ccd9e JA |
1422 | WRITE_ONCE(ctx->rings->cq_overflow, |
1423 | atomic_inc_return(&ctx->cached_cq_overflow)); | |
1d7bb1d5 | 1424 | } else { |
ad3eb2c8 JA |
1425 | if (list_empty(&ctx->cq_overflow_list)) { |
1426 | set_bit(0, &ctx->sq_check_overflow); | |
1427 | set_bit(0, &ctx->cq_check_overflow); | |
6d5f9049 | 1428 | ctx->rings->sq_flags |= IORING_SQ_CQ_OVERFLOW; |
ad3eb2c8 | 1429 | } |
40d8ddd4 | 1430 | io_clean_op(req); |
1d7bb1d5 | 1431 | req->result = res; |
0f7e466b | 1432 | req->compl.cflags = cflags; |
40d8ddd4 PB |
1433 | refcount_inc(&req->refs); |
1434 | list_add_tail(&req->compl.list, &ctx->cq_overflow_list); | |
2b188cc1 JA |
1435 | } |
1436 | } | |
1437 | ||
bcda7baa JA |
1438 | static void io_cqring_fill_event(struct io_kiocb *req, long res) |
1439 | { | |
1440 | __io_cqring_fill_event(req, res, 0); | |
1441 | } | |
1442 | ||
e1e16097 | 1443 | static void io_cqring_add_event(struct io_kiocb *req, long res, long cflags) |
2b188cc1 | 1444 | { |
78e19bbe | 1445 | struct io_ring_ctx *ctx = req->ctx; |
2b188cc1 JA |
1446 | unsigned long flags; |
1447 | ||
1448 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
bcda7baa | 1449 | __io_cqring_fill_event(req, res, cflags); |
2b188cc1 JA |
1450 | io_commit_cqring(ctx); |
1451 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
1452 | ||
8c838788 | 1453 | io_cqring_ev_posted(ctx); |
2b188cc1 JA |
1454 | } |
1455 | ||
229a7b63 | 1456 | static void io_submit_flush_completions(struct io_comp_state *cs) |
bcda7baa | 1457 | { |
229a7b63 JA |
1458 | struct io_ring_ctx *ctx = cs->ctx; |
1459 | ||
1460 | spin_lock_irq(&ctx->completion_lock); | |
1461 | while (!list_empty(&cs->list)) { | |
1462 | struct io_kiocb *req; | |
1463 | ||
3ca405eb PB |
1464 | req = list_first_entry(&cs->list, struct io_kiocb, compl.list); |
1465 | list_del(&req->compl.list); | |
0f7e466b | 1466 | __io_cqring_fill_event(req, req->result, req->compl.cflags); |
229a7b63 JA |
1467 | if (!(req->flags & REQ_F_LINK_HEAD)) { |
1468 | req->flags |= REQ_F_COMP_LOCKED; | |
1469 | io_put_req(req); | |
1470 | } else { | |
1471 | spin_unlock_irq(&ctx->completion_lock); | |
1472 | io_put_req(req); | |
1473 | spin_lock_irq(&ctx->completion_lock); | |
1474 | } | |
1475 | } | |
1476 | io_commit_cqring(ctx); | |
1477 | spin_unlock_irq(&ctx->completion_lock); | |
1478 | ||
1479 | io_cqring_ev_posted(ctx); | |
1480 | cs->nr = 0; | |
1481 | } | |
1482 | ||
1483 | static void __io_req_complete(struct io_kiocb *req, long res, unsigned cflags, | |
1484 | struct io_comp_state *cs) | |
1485 | { | |
1486 | if (!cs) { | |
1487 | io_cqring_add_event(req, res, cflags); | |
1488 | io_put_req(req); | |
1489 | } else { | |
3ca405eb | 1490 | io_clean_op(req); |
229a7b63 | 1491 | req->result = res; |
0f7e466b | 1492 | req->compl.cflags = cflags; |
3ca405eb | 1493 | list_add_tail(&req->compl.list, &cs->list); |
229a7b63 JA |
1494 | if (++cs->nr >= 32) |
1495 | io_submit_flush_completions(cs); | |
1496 | } | |
e1e16097 JA |
1497 | } |
1498 | ||
1499 | static void io_req_complete(struct io_kiocb *req, long res) | |
bcda7baa | 1500 | { |
229a7b63 | 1501 | __io_req_complete(req, res, 0, NULL); |
bcda7baa JA |
1502 | } |
1503 | ||
0ddf92e8 JA |
1504 | static inline bool io_is_fallback_req(struct io_kiocb *req) |
1505 | { | |
1506 | return req == (struct io_kiocb *) | |
1507 | ((unsigned long) req->ctx->fallback_req & ~1UL); | |
1508 | } | |
1509 | ||
1510 | static struct io_kiocb *io_get_fallback_req(struct io_ring_ctx *ctx) | |
1511 | { | |
1512 | struct io_kiocb *req; | |
1513 | ||
1514 | req = ctx->fallback_req; | |
dd461af6 | 1515 | if (!test_and_set_bit_lock(0, (unsigned long *) &ctx->fallback_req)) |
0ddf92e8 JA |
1516 | return req; |
1517 | ||
1518 | return NULL; | |
1519 | } | |
1520 | ||
0553b8bd PB |
1521 | static struct io_kiocb *io_alloc_req(struct io_ring_ctx *ctx, |
1522 | struct io_submit_state *state) | |
2b188cc1 | 1523 | { |
fd6fab2c | 1524 | gfp_t gfp = GFP_KERNEL | __GFP_NOWARN; |
2b188cc1 JA |
1525 | struct io_kiocb *req; |
1526 | ||
f6b6c7d6 | 1527 | if (!state->free_reqs) { |
2579f913 JA |
1528 | size_t sz; |
1529 | int ret; | |
1530 | ||
1531 | sz = min_t(size_t, state->ios_left, ARRAY_SIZE(state->reqs)); | |
fd6fab2c JA |
1532 | ret = kmem_cache_alloc_bulk(req_cachep, gfp, sz, state->reqs); |
1533 | ||
1534 | /* | |
1535 | * Bulk alloc is all-or-nothing. If we fail to get a batch, | |
1536 | * retry single alloc to be on the safe side. | |
1537 | */ | |
1538 | if (unlikely(ret <= 0)) { | |
1539 | state->reqs[0] = kmem_cache_alloc(req_cachep, gfp); | |
1540 | if (!state->reqs[0]) | |
0ddf92e8 | 1541 | goto fallback; |
fd6fab2c JA |
1542 | ret = 1; |
1543 | } | |
2579f913 | 1544 | state->free_reqs = ret - 1; |
6c8a3134 | 1545 | req = state->reqs[ret - 1]; |
2579f913 | 1546 | } else { |
2579f913 | 1547 | state->free_reqs--; |
6c8a3134 | 1548 | req = state->reqs[state->free_reqs]; |
2b188cc1 JA |
1549 | } |
1550 | ||
2579f913 | 1551 | return req; |
0ddf92e8 | 1552 | fallback: |
0553b8bd | 1553 | return io_get_fallback_req(ctx); |
2b188cc1 JA |
1554 | } |
1555 | ||
8da11c19 PB |
1556 | static inline void io_put_file(struct io_kiocb *req, struct file *file, |
1557 | bool fixed) | |
1558 | { | |
1559 | if (fixed) | |
05589553 | 1560 | percpu_ref_put(req->fixed_file_refs); |
8da11c19 PB |
1561 | else |
1562 | fput(file); | |
1563 | } | |
1564 | ||
51a4cc11 | 1565 | static bool io_dismantle_req(struct io_kiocb *req) |
2b188cc1 | 1566 | { |
3ca405eb | 1567 | io_clean_op(req); |
929a3af9 | 1568 | |
5acbbc8e JA |
1569 | if (req->io) |
1570 | kfree(req->io); | |
8da11c19 PB |
1571 | if (req->file) |
1572 | io_put_file(req, req->file, (req->flags & REQ_F_FIXED_FILE)); | |
fcb323cc | 1573 | |
51a4cc11 | 1574 | return io_req_clean_work(req); |
e65ef56d JA |
1575 | } |
1576 | ||
51a4cc11 | 1577 | static void __io_free_req_finish(struct io_kiocb *req) |
c6ca97b3 | 1578 | { |
51a4cc11 | 1579 | struct io_ring_ctx *ctx = req->ctx; |
c6ca97b3 | 1580 | |
e3bc8e9d JA |
1581 | put_task_struct(req->task); |
1582 | ||
b1e50e54 PB |
1583 | if (likely(!io_is_fallback_req(req))) |
1584 | kmem_cache_free(req_cachep, req); | |
1585 | else | |
ecfc5177 PB |
1586 | clear_bit_unlock(0, (unsigned long *) &ctx->fallback_req); |
1587 | percpu_ref_put(&ctx->refs); | |
e65ef56d JA |
1588 | } |
1589 | ||
51a4cc11 JA |
1590 | static void io_req_task_file_table_put(struct callback_head *cb) |
1591 | { | |
1592 | struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work); | |
1593 | struct fs_struct *fs = req->work.fs; | |
1594 | ||
1595 | spin_lock(&req->work.fs->lock); | |
1596 | if (--fs->users) | |
1597 | fs = NULL; | |
1598 | spin_unlock(&req->work.fs->lock); | |
1599 | if (fs) | |
1600 | free_fs_struct(fs); | |
1601 | req->work.fs = NULL; | |
1602 | __io_free_req_finish(req); | |
1603 | } | |
1604 | ||
1605 | static void __io_free_req(struct io_kiocb *req) | |
1606 | { | |
1607 | if (!io_dismantle_req(req)) { | |
1608 | __io_free_req_finish(req); | |
1609 | } else { | |
1610 | int ret; | |
1611 | ||
1612 | init_task_work(&req->task_work, io_req_task_file_table_put); | |
1613 | ret = task_work_add(req->task, &req->task_work, TWA_RESUME); | |
1614 | if (unlikely(ret)) { | |
1615 | struct task_struct *tsk; | |
1616 | ||
1617 | tsk = io_wq_get_task(req->ctx->io_wq); | |
1618 | task_work_add(tsk, &req->task_work, 0); | |
1619 | } | |
1620 | } | |
1621 | } | |
1622 | ||
a197f664 | 1623 | static bool io_link_cancel_timeout(struct io_kiocb *req) |
2665abfd | 1624 | { |
a197f664 | 1625 | struct io_ring_ctx *ctx = req->ctx; |
2665abfd JA |
1626 | int ret; |
1627 | ||
2d28390a | 1628 | ret = hrtimer_try_to_cancel(&req->io->timeout.timer); |
2665abfd | 1629 | if (ret != -1) { |
78e19bbe | 1630 | io_cqring_fill_event(req, -ECANCELED); |
2665abfd | 1631 | io_commit_cqring(ctx); |
dea3b49c | 1632 | req->flags &= ~REQ_F_LINK_HEAD; |
ec9c02ad | 1633 | io_put_req(req); |
2665abfd JA |
1634 | return true; |
1635 | } | |
1636 | ||
1637 | return false; | |
e65ef56d JA |
1638 | } |
1639 | ||
ab0b6451 | 1640 | static bool __io_kill_linked_timeout(struct io_kiocb *req) |
9e645e11 | 1641 | { |
7c86ffee | 1642 | struct io_kiocb *link; |
ab0b6451 | 1643 | bool wake_ev; |
7c86ffee PB |
1644 | |
1645 | if (list_empty(&req->link_list)) | |
ab0b6451 | 1646 | return false; |
7c86ffee PB |
1647 | link = list_first_entry(&req->link_list, struct io_kiocb, link_list); |
1648 | if (link->opcode != IORING_OP_LINK_TIMEOUT) | |
ab0b6451 | 1649 | return false; |
7c86ffee PB |
1650 | |
1651 | list_del_init(&link->link_list); | |
9b7adba9 | 1652 | link->flags |= REQ_F_COMP_LOCKED; |
7c86ffee PB |
1653 | wake_ev = io_link_cancel_timeout(link); |
1654 | req->flags &= ~REQ_F_LINK_TIMEOUT; | |
ab0b6451 JA |
1655 | return wake_ev; |
1656 | } | |
1657 | ||
1658 | static void io_kill_linked_timeout(struct io_kiocb *req) | |
9e645e11 | 1659 | { |
2665abfd | 1660 | struct io_ring_ctx *ctx = req->ctx; |
ab0b6451 | 1661 | bool wake_ev; |
9e645e11 | 1662 | |
ab0b6451 JA |
1663 | if (!(req->flags & REQ_F_COMP_LOCKED)) { |
1664 | unsigned long flags; | |
1665 | ||
1666 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
1667 | wake_ev = __io_kill_linked_timeout(req); | |
7c86ffee | 1668 | spin_unlock_irqrestore(&ctx->completion_lock, flags); |
ab0b6451 JA |
1669 | } else { |
1670 | wake_ev = __io_kill_linked_timeout(req); | |
1671 | } | |
1672 | ||
7c86ffee PB |
1673 | if (wake_ev) |
1674 | io_cqring_ev_posted(ctx); | |
1675 | } | |
1676 | ||
9b5f7bd9 | 1677 | static struct io_kiocb *io_req_link_next(struct io_kiocb *req) |
7c86ffee PB |
1678 | { |
1679 | struct io_kiocb *nxt; | |
4d7dd462 | 1680 | |
9e645e11 JA |
1681 | /* |
1682 | * The list should never be empty when we are called here. But could | |
1683 | * potentially happen if the chain is messed up, check to be on the | |
1684 | * safe side. | |
1685 | */ | |
7c86ffee | 1686 | if (unlikely(list_empty(&req->link_list))) |
9b5f7bd9 | 1687 | return NULL; |
2665abfd | 1688 | |
7c86ffee PB |
1689 | nxt = list_first_entry(&req->link_list, struct io_kiocb, link_list); |
1690 | list_del_init(&req->link_list); | |
1691 | if (!list_empty(&nxt->link_list)) | |
1692 | nxt->flags |= REQ_F_LINK_HEAD; | |
9b5f7bd9 | 1693 | return nxt; |
9e645e11 JA |
1694 | } |
1695 | ||
1696 | /* | |
dea3b49c | 1697 | * Called if REQ_F_LINK_HEAD is set, and we fail the head request |
9e645e11 | 1698 | */ |
7c86ffee | 1699 | static void __io_fail_links(struct io_kiocb *req) |
9e645e11 | 1700 | { |
2665abfd | 1701 | struct io_ring_ctx *ctx = req->ctx; |
9e645e11 JA |
1702 | |
1703 | while (!list_empty(&req->link_list)) { | |
4493233e PB |
1704 | struct io_kiocb *link = list_first_entry(&req->link_list, |
1705 | struct io_kiocb, link_list); | |
9e645e11 | 1706 | |
4493233e | 1707 | list_del_init(&link->link_list); |
c826bd7a | 1708 | trace_io_uring_fail_link(req, link); |
2665abfd | 1709 | |
7c86ffee | 1710 | io_cqring_fill_event(link, -ECANCELED); |
9b7adba9 | 1711 | link->flags |= REQ_F_COMP_LOCKED; |
7c86ffee | 1712 | __io_double_put_req(link); |
5d960724 | 1713 | req->flags &= ~REQ_F_LINK_TIMEOUT; |
9e645e11 | 1714 | } |
2665abfd JA |
1715 | |
1716 | io_commit_cqring(ctx); | |
2665abfd | 1717 | io_cqring_ev_posted(ctx); |
9e645e11 JA |
1718 | } |
1719 | ||
7c86ffee | 1720 | static void io_fail_links(struct io_kiocb *req) |
9e645e11 | 1721 | { |
7c86ffee | 1722 | struct io_ring_ctx *ctx = req->ctx; |
2665abfd | 1723 | |
7c86ffee | 1724 | if (!(req->flags & REQ_F_COMP_LOCKED)) { |
2665abfd JA |
1725 | unsigned long flags; |
1726 | ||
2665abfd | 1727 | spin_lock_irqsave(&ctx->completion_lock, flags); |
7c86ffee | 1728 | __io_fail_links(req); |
2665abfd JA |
1729 | spin_unlock_irqrestore(&ctx->completion_lock, flags); |
1730 | } else { | |
7c86ffee | 1731 | __io_fail_links(req); |
9e645e11 JA |
1732 | } |
1733 | ||
2665abfd | 1734 | io_cqring_ev_posted(ctx); |
9e645e11 JA |
1735 | } |
1736 | ||
3fa5e0f3 | 1737 | static struct io_kiocb *__io_req_find_next(struct io_kiocb *req) |
c69f8dbe | 1738 | { |
9b0d911a | 1739 | req->flags &= ~REQ_F_LINK_HEAD; |
7c86ffee PB |
1740 | if (req->flags & REQ_F_LINK_TIMEOUT) |
1741 | io_kill_linked_timeout(req); | |
944e58bf | 1742 | |
9e645e11 JA |
1743 | /* |
1744 | * If LINK is set, we have dependent requests in this chain. If we | |
1745 | * didn't fail this request, queue the first one up, moving any other | |
1746 | * dependencies to the next request. In case of failure, fail the rest | |
1747 | * of the chain. | |
1748 | */ | |
9b5f7bd9 PB |
1749 | if (likely(!(req->flags & REQ_F_FAIL_LINK))) |
1750 | return io_req_link_next(req); | |
1751 | io_fail_links(req); | |
1752 | return NULL; | |
4d7dd462 | 1753 | } |
9e645e11 | 1754 | |
3fa5e0f3 PB |
1755 | static struct io_kiocb *io_req_find_next(struct io_kiocb *req) |
1756 | { | |
1757 | if (likely(!(req->flags & REQ_F_LINK_HEAD))) | |
1758 | return NULL; | |
1759 | return __io_req_find_next(req); | |
1760 | } | |
1761 | ||
fd7d6de2 JA |
1762 | static int io_req_task_work_add(struct io_kiocb *req, struct callback_head *cb, |
1763 | bool twa_signal_ok) | |
c2c4c83c JA |
1764 | { |
1765 | struct task_struct *tsk = req->task; | |
1766 | struct io_ring_ctx *ctx = req->ctx; | |
0ba9c9ed | 1767 | int ret, notify; |
c2c4c83c | 1768 | |
6200b0ae JA |
1769 | if (tsk->flags & PF_EXITING) |
1770 | return -ESRCH; | |
1771 | ||
c2c4c83c | 1772 | /* |
0ba9c9ed JA |
1773 | * SQPOLL kernel thread doesn't need notification, just a wakeup. For |
1774 | * all other cases, use TWA_SIGNAL unconditionally to ensure we're | |
1775 | * processing task_work. There's no reliable way to tell if TWA_RESUME | |
1776 | * will do the job. | |
c2c4c83c | 1777 | */ |
0ba9c9ed | 1778 | notify = 0; |
fd7d6de2 | 1779 | if (!(ctx->flags & IORING_SETUP_SQPOLL) && twa_signal_ok) |
c2c4c83c JA |
1780 | notify = TWA_SIGNAL; |
1781 | ||
1782 | ret = task_work_add(tsk, cb, notify); | |
1783 | if (!ret) | |
1784 | wake_up_process(tsk); | |
0ba9c9ed | 1785 | |
c2c4c83c JA |
1786 | return ret; |
1787 | } | |
1788 | ||
c40f6379 JA |
1789 | static void __io_req_task_cancel(struct io_kiocb *req, int error) |
1790 | { | |
1791 | struct io_ring_ctx *ctx = req->ctx; | |
1792 | ||
1793 | spin_lock_irq(&ctx->completion_lock); | |
1794 | io_cqring_fill_event(req, error); | |
1795 | io_commit_cqring(ctx); | |
1796 | spin_unlock_irq(&ctx->completion_lock); | |
1797 | ||
1798 | io_cqring_ev_posted(ctx); | |
1799 | req_set_fail_links(req); | |
1800 | io_double_put_req(req); | |
1801 | } | |
1802 | ||
1803 | static void io_req_task_cancel(struct callback_head *cb) | |
1804 | { | |
1805 | struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work); | |
87ceb6a6 | 1806 | struct io_ring_ctx *ctx = req->ctx; |
c40f6379 JA |
1807 | |
1808 | __io_req_task_cancel(req, -ECANCELED); | |
87ceb6a6 | 1809 | percpu_ref_put(&ctx->refs); |
c40f6379 JA |
1810 | } |
1811 | ||
1812 | static void __io_req_task_submit(struct io_kiocb *req) | |
1813 | { | |
1814 | struct io_ring_ctx *ctx = req->ctx; | |
1815 | ||
c40f6379 JA |
1816 | if (!__io_sq_thread_acquire_mm(ctx)) { |
1817 | mutex_lock(&ctx->uring_lock); | |
1818 | __io_queue_sqe(req, NULL, NULL); | |
1819 | mutex_unlock(&ctx->uring_lock); | |
1820 | } else { | |
1821 | __io_req_task_cancel(req, -EFAULT); | |
1822 | } | |
1823 | } | |
1824 | ||
1825 | static void io_req_task_submit(struct callback_head *cb) | |
1826 | { | |
1827 | struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work); | |
6d816e08 | 1828 | struct io_ring_ctx *ctx = req->ctx; |
c40f6379 JA |
1829 | |
1830 | __io_req_task_submit(req); | |
6d816e08 | 1831 | percpu_ref_put(&ctx->refs); |
c40f6379 JA |
1832 | } |
1833 | ||
1834 | static void io_req_task_queue(struct io_kiocb *req) | |
1835 | { | |
c40f6379 JA |
1836 | int ret; |
1837 | ||
1838 | init_task_work(&req->task_work, io_req_task_submit); | |
6d816e08 | 1839 | percpu_ref_get(&req->ctx->refs); |
c40f6379 | 1840 | |
fd7d6de2 | 1841 | ret = io_req_task_work_add(req, &req->task_work, true); |
c40f6379 | 1842 | if (unlikely(ret)) { |
c2c4c83c JA |
1843 | struct task_struct *tsk; |
1844 | ||
c40f6379 JA |
1845 | init_task_work(&req->task_work, io_req_task_cancel); |
1846 | tsk = io_wq_get_task(req->ctx->io_wq); | |
c2c4c83c JA |
1847 | task_work_add(tsk, &req->task_work, 0); |
1848 | wake_up_process(tsk); | |
c40f6379 | 1849 | } |
c40f6379 JA |
1850 | } |
1851 | ||
c3524383 | 1852 | static void io_queue_next(struct io_kiocb *req) |
c69f8dbe | 1853 | { |
9b5f7bd9 | 1854 | struct io_kiocb *nxt = io_req_find_next(req); |
944e58bf PB |
1855 | |
1856 | if (nxt) | |
906a8c3f | 1857 | io_req_task_queue(nxt); |
c69f8dbe JL |
1858 | } |
1859 | ||
c3524383 | 1860 | static void io_free_req(struct io_kiocb *req) |
7a743e22 | 1861 | { |
c3524383 PB |
1862 | io_queue_next(req); |
1863 | __io_free_req(req); | |
1864 | } | |
8766dd51 | 1865 | |
2d6500d4 PB |
1866 | struct req_batch { |
1867 | void *reqs[IO_IOPOLL_BATCH]; | |
1868 | int to_free; | |
7a743e22 | 1869 | |
5af1d13e PB |
1870 | struct task_struct *task; |
1871 | int task_refs; | |
2d6500d4 PB |
1872 | }; |
1873 | ||
5af1d13e PB |
1874 | static inline void io_init_req_batch(struct req_batch *rb) |
1875 | { | |
1876 | rb->to_free = 0; | |
1877 | rb->task_refs = 0; | |
1878 | rb->task = NULL; | |
1879 | } | |
1880 | ||
2d6500d4 PB |
1881 | static void __io_req_free_batch_flush(struct io_ring_ctx *ctx, |
1882 | struct req_batch *rb) | |
1883 | { | |
1884 | kmem_cache_free_bulk(req_cachep, rb->to_free, rb->reqs); | |
1885 | percpu_ref_put_many(&ctx->refs, rb->to_free); | |
1886 | rb->to_free = 0; | |
1887 | } | |
1888 | ||
1889 | static void io_req_free_batch_finish(struct io_ring_ctx *ctx, | |
1890 | struct req_batch *rb) | |
1891 | { | |
1892 | if (rb->to_free) | |
1893 | __io_req_free_batch_flush(ctx, rb); | |
5af1d13e PB |
1894 | if (rb->task) { |
1895 | put_task_struct_many(rb->task, rb->task_refs); | |
1896 | rb->task = NULL; | |
1897 | } | |
2d6500d4 PB |
1898 | } |
1899 | ||
1900 | static void io_req_free_batch(struct req_batch *rb, struct io_kiocb *req) | |
1901 | { | |
1902 | if (unlikely(io_is_fallback_req(req))) { | |
1903 | io_free_req(req); | |
1904 | return; | |
1905 | } | |
1906 | if (req->flags & REQ_F_LINK_HEAD) | |
1907 | io_queue_next(req); | |
1908 | ||
e3bc8e9d JA |
1909 | if (req->task != rb->task) { |
1910 | if (rb->task) | |
1911 | put_task_struct_many(rb->task, rb->task_refs); | |
1912 | rb->task = req->task; | |
1913 | rb->task_refs = 0; | |
5af1d13e | 1914 | } |
e3bc8e9d | 1915 | rb->task_refs++; |
5af1d13e | 1916 | |
51a4cc11 | 1917 | WARN_ON_ONCE(io_dismantle_req(req)); |
2d6500d4 PB |
1918 | rb->reqs[rb->to_free++] = req; |
1919 | if (unlikely(rb->to_free == ARRAY_SIZE(rb->reqs))) | |
1920 | __io_req_free_batch_flush(req->ctx, rb); | |
7a743e22 PB |
1921 | } |
1922 | ||
ba816ad6 JA |
1923 | /* |
1924 | * Drop reference to request, return next in chain (if there is one) if this | |
1925 | * was the last reference to this request. | |
1926 | */ | |
9b5f7bd9 | 1927 | static struct io_kiocb *io_put_req_find_next(struct io_kiocb *req) |
e65ef56d | 1928 | { |
9b5f7bd9 PB |
1929 | struct io_kiocb *nxt = NULL; |
1930 | ||
2a44f467 | 1931 | if (refcount_dec_and_test(&req->refs)) { |
9b5f7bd9 | 1932 | nxt = io_req_find_next(req); |
4d7dd462 | 1933 | __io_free_req(req); |
2a44f467 | 1934 | } |
9b5f7bd9 | 1935 | return nxt; |
2b188cc1 JA |
1936 | } |
1937 | ||
e65ef56d JA |
1938 | static void io_put_req(struct io_kiocb *req) |
1939 | { | |
1940 | if (refcount_dec_and_test(&req->refs)) | |
1941 | io_free_req(req); | |
2b188cc1 JA |
1942 | } |
1943 | ||
f4db7182 | 1944 | static struct io_wq_work *io_steal_work(struct io_kiocb *req) |
7a743e22 | 1945 | { |
6df1db6b | 1946 | struct io_kiocb *nxt; |
f4db7182 | 1947 | |
7a743e22 | 1948 | /* |
f4db7182 PB |
1949 | * A ref is owned by io-wq in which context we're. So, if that's the |
1950 | * last one, it's safe to steal next work. False negatives are Ok, | |
1951 | * it just will be re-punted async in io_put_work() | |
7a743e22 | 1952 | */ |
f4db7182 PB |
1953 | if (refcount_read(&req->refs) != 1) |
1954 | return NULL; | |
7a743e22 | 1955 | |
9b5f7bd9 | 1956 | nxt = io_req_find_next(req); |
6df1db6b | 1957 | return nxt ? &nxt->work : NULL; |
7a743e22 PB |
1958 | } |
1959 | ||
978db57e JA |
1960 | /* |
1961 | * Must only be used if we don't need to care about links, usually from | |
1962 | * within the completion handling itself. | |
1963 | */ | |
1964 | static void __io_double_put_req(struct io_kiocb *req) | |
78e19bbe JA |
1965 | { |
1966 | /* drop both submit and complete references */ | |
1967 | if (refcount_sub_and_test(2, &req->refs)) | |
1968 | __io_free_req(req); | |
1969 | } | |
1970 | ||
978db57e JA |
1971 | static void io_double_put_req(struct io_kiocb *req) |
1972 | { | |
1973 | /* drop both submit and complete references */ | |
1974 | if (refcount_sub_and_test(2, &req->refs)) | |
1975 | io_free_req(req); | |
1976 | } | |
1977 | ||
1d7bb1d5 | 1978 | static unsigned io_cqring_events(struct io_ring_ctx *ctx, bool noflush) |
a3a0e43f | 1979 | { |
84f97dc2 JA |
1980 | struct io_rings *rings = ctx->rings; |
1981 | ||
ad3eb2c8 JA |
1982 | if (test_bit(0, &ctx->cq_check_overflow)) { |
1983 | /* | |
1984 | * noflush == true is from the waitqueue handler, just ensure | |
1985 | * we wake up the task, and the next invocation will flush the | |
1986 | * entries. We cannot safely to it from here. | |
1987 | */ | |
1988 | if (noflush && !list_empty(&ctx->cq_overflow_list)) | |
1989 | return -1U; | |
1d7bb1d5 | 1990 | |
ad3eb2c8 JA |
1991 | io_cqring_overflow_flush(ctx, false); |
1992 | } | |
1d7bb1d5 | 1993 | |
a3a0e43f JA |
1994 | /* See comment at the top of this file */ |
1995 | smp_rmb(); | |
ad3eb2c8 | 1996 | return ctx->cached_cq_tail - READ_ONCE(rings->cq.head); |
a3a0e43f JA |
1997 | } |
1998 | ||
fb5ccc98 PB |
1999 | static inline unsigned int io_sqring_entries(struct io_ring_ctx *ctx) |
2000 | { | |
2001 | struct io_rings *rings = ctx->rings; | |
2002 | ||
2003 | /* make sure SQ entry isn't read before tail */ | |
2004 | return smp_load_acquire(&rings->sq.tail) - ctx->cached_sq_head; | |
2005 | } | |
2006 | ||
8ff069bf | 2007 | static unsigned int io_put_kbuf(struct io_kiocb *req, struct io_buffer *kbuf) |
e94f141b | 2008 | { |
8ff069bf | 2009 | unsigned int cflags; |
e94f141b | 2010 | |
bcda7baa JA |
2011 | cflags = kbuf->bid << IORING_CQE_BUFFER_SHIFT; |
2012 | cflags |= IORING_CQE_F_BUFFER; | |
0e1b6fe3 | 2013 | req->flags &= ~REQ_F_BUFFER_SELECTED; |
bcda7baa JA |
2014 | kfree(kbuf); |
2015 | return cflags; | |
e94f141b JA |
2016 | } |
2017 | ||
8ff069bf | 2018 | static inline unsigned int io_put_rw_kbuf(struct io_kiocb *req) |
bcda7baa | 2019 | { |
4d954c25 | 2020 | struct io_buffer *kbuf; |
bcda7baa | 2021 | |
4d954c25 | 2022 | kbuf = (struct io_buffer *) (unsigned long) req->rw.addr; |
8ff069bf PB |
2023 | return io_put_kbuf(req, kbuf); |
2024 | } | |
2025 | ||
4c6e277c JA |
2026 | static inline bool io_run_task_work(void) |
2027 | { | |
6200b0ae JA |
2028 | /* |
2029 | * Not safe to run on exiting task, and the task_work handling will | |
2030 | * not add work to such a task. | |
2031 | */ | |
2032 | if (unlikely(current->flags & PF_EXITING)) | |
2033 | return false; | |
4c6e277c JA |
2034 | if (current->task_works) { |
2035 | __set_current_state(TASK_RUNNING); | |
2036 | task_work_run(); | |
2037 | return true; | |
2038 | } | |
2039 | ||
2040 | return false; | |
bcda7baa JA |
2041 | } |
2042 | ||
bbde017a XW |
2043 | static void io_iopoll_queue(struct list_head *again) |
2044 | { | |
2045 | struct io_kiocb *req; | |
2046 | ||
2047 | do { | |
d21ffe7e PB |
2048 | req = list_first_entry(again, struct io_kiocb, inflight_entry); |
2049 | list_del(&req->inflight_entry); | |
81b68a5c | 2050 | __io_complete_rw(req, -EAGAIN, 0, NULL); |
bbde017a XW |
2051 | } while (!list_empty(again)); |
2052 | } | |
2053 | ||
def596e9 JA |
2054 | /* |
2055 | * Find and free completed poll iocbs | |
2056 | */ | |
2057 | static void io_iopoll_complete(struct io_ring_ctx *ctx, unsigned int *nr_events, | |
2058 | struct list_head *done) | |
2059 | { | |
8237e045 | 2060 | struct req_batch rb; |
def596e9 | 2061 | struct io_kiocb *req; |
bbde017a XW |
2062 | LIST_HEAD(again); |
2063 | ||
2064 | /* order with ->result store in io_complete_rw_iopoll() */ | |
2065 | smp_rmb(); | |
def596e9 | 2066 | |
5af1d13e | 2067 | io_init_req_batch(&rb); |
def596e9 | 2068 | while (!list_empty(done)) { |
bcda7baa JA |
2069 | int cflags = 0; |
2070 | ||
d21ffe7e | 2071 | req = list_first_entry(done, struct io_kiocb, inflight_entry); |
bbde017a | 2072 | if (READ_ONCE(req->result) == -EAGAIN) { |
56450c20 | 2073 | req->result = 0; |
bbde017a | 2074 | req->iopoll_completed = 0; |
d21ffe7e | 2075 | list_move_tail(&req->inflight_entry, &again); |
bbde017a XW |
2076 | continue; |
2077 | } | |
d21ffe7e | 2078 | list_del(&req->inflight_entry); |
def596e9 | 2079 | |
bcda7baa | 2080 | if (req->flags & REQ_F_BUFFER_SELECTED) |
8ff069bf | 2081 | cflags = io_put_rw_kbuf(req); |
bcda7baa JA |
2082 | |
2083 | __io_cqring_fill_event(req, req->result, cflags); | |
def596e9 JA |
2084 | (*nr_events)++; |
2085 | ||
c3524383 | 2086 | if (refcount_dec_and_test(&req->refs)) |
2d6500d4 | 2087 | io_req_free_batch(&rb, req); |
def596e9 | 2088 | } |
def596e9 | 2089 | |
09bb8394 | 2090 | io_commit_cqring(ctx); |
32b2244a XW |
2091 | if (ctx->flags & IORING_SETUP_SQPOLL) |
2092 | io_cqring_ev_posted(ctx); | |
2d6500d4 | 2093 | io_req_free_batch_finish(ctx, &rb); |
581f9810 | 2094 | |
bbde017a XW |
2095 | if (!list_empty(&again)) |
2096 | io_iopoll_queue(&again); | |
581f9810 BM |
2097 | } |
2098 | ||
def596e9 JA |
2099 | static int io_do_iopoll(struct io_ring_ctx *ctx, unsigned int *nr_events, |
2100 | long min) | |
2101 | { | |
2102 | struct io_kiocb *req, *tmp; | |
2103 | LIST_HEAD(done); | |
2104 | bool spin; | |
2105 | int ret; | |
2106 | ||
2107 | /* | |
2108 | * Only spin for completions if we don't have multiple devices hanging | |
2109 | * off our complete list, and we're under the requested amount. | |
2110 | */ | |
2111 | spin = !ctx->poll_multi_file && *nr_events < min; | |
2112 | ||
2113 | ret = 0; | |
d21ffe7e | 2114 | list_for_each_entry_safe(req, tmp, &ctx->iopoll_list, inflight_entry) { |
9adbd45d | 2115 | struct kiocb *kiocb = &req->rw.kiocb; |
def596e9 JA |
2116 | |
2117 | /* | |
581f9810 BM |
2118 | * Move completed and retryable entries to our local lists. |
2119 | * If we find a request that requires polling, break out | |
2120 | * and complete those lists first, if we have entries there. | |
def596e9 | 2121 | */ |
65a6543d | 2122 | if (READ_ONCE(req->iopoll_completed)) { |
d21ffe7e | 2123 | list_move_tail(&req->inflight_entry, &done); |
def596e9 JA |
2124 | continue; |
2125 | } | |
2126 | if (!list_empty(&done)) | |
2127 | break; | |
2128 | ||
2129 | ret = kiocb->ki_filp->f_op->iopoll(kiocb, spin); | |
2130 | if (ret < 0) | |
2131 | break; | |
2132 | ||
3aadc23e PB |
2133 | /* iopoll may have completed current req */ |
2134 | if (READ_ONCE(req->iopoll_completed)) | |
d21ffe7e | 2135 | list_move_tail(&req->inflight_entry, &done); |
3aadc23e | 2136 | |
def596e9 JA |
2137 | if (ret && spin) |
2138 | spin = false; | |
2139 | ret = 0; | |
2140 | } | |
2141 | ||
2142 | if (!list_empty(&done)) | |
2143 | io_iopoll_complete(ctx, nr_events, &done); | |
2144 | ||
2145 | return ret; | |
2146 | } | |
2147 | ||
2148 | /* | |
d195a66e | 2149 | * Poll for a minimum of 'min' events. Note that if min == 0 we consider that a |
def596e9 JA |
2150 | * non-spinning poll check - we'll still enter the driver poll loop, but only |
2151 | * as a non-spinning completion check. | |
2152 | */ | |
2153 | static int io_iopoll_getevents(struct io_ring_ctx *ctx, unsigned int *nr_events, | |
2154 | long min) | |
2155 | { | |
540e32a0 | 2156 | while (!list_empty(&ctx->iopoll_list) && !need_resched()) { |
def596e9 JA |
2157 | int ret; |
2158 | ||
2159 | ret = io_do_iopoll(ctx, nr_events, min); | |
2160 | if (ret < 0) | |
2161 | return ret; | |
eba0a4dd | 2162 | if (*nr_events >= min) |
def596e9 JA |
2163 | return 0; |
2164 | } | |
2165 | ||
2166 | return 1; | |
2167 | } | |
2168 | ||
2169 | /* | |
2170 | * We can't just wait for polled events to come to us, we have to actively | |
2171 | * find and complete them. | |
2172 | */ | |
b2edc0a7 | 2173 | static void io_iopoll_try_reap_events(struct io_ring_ctx *ctx) |
def596e9 JA |
2174 | { |
2175 | if (!(ctx->flags & IORING_SETUP_IOPOLL)) | |
2176 | return; | |
2177 | ||
2178 | mutex_lock(&ctx->uring_lock); | |
540e32a0 | 2179 | while (!list_empty(&ctx->iopoll_list)) { |
def596e9 JA |
2180 | unsigned int nr_events = 0; |
2181 | ||
b2edc0a7 | 2182 | io_do_iopoll(ctx, &nr_events, 0); |
08f5439f | 2183 | |
b2edc0a7 PB |
2184 | /* let it sleep and repeat later if can't complete a request */ |
2185 | if (nr_events == 0) | |
2186 | break; | |
08f5439f JA |
2187 | /* |
2188 | * Ensure we allow local-to-the-cpu processing to take place, | |
2189 | * in this case we need to ensure that we reap all events. | |
3fcee5a6 | 2190 | * Also let task_work, etc. to progress by releasing the mutex |
08f5439f | 2191 | */ |
3fcee5a6 PB |
2192 | if (need_resched()) { |
2193 | mutex_unlock(&ctx->uring_lock); | |
2194 | cond_resched(); | |
2195 | mutex_lock(&ctx->uring_lock); | |
2196 | } | |
def596e9 JA |
2197 | } |
2198 | mutex_unlock(&ctx->uring_lock); | |
2199 | } | |
2200 | ||
7668b92a | 2201 | static int io_iopoll_check(struct io_ring_ctx *ctx, long min) |
def596e9 | 2202 | { |
7668b92a | 2203 | unsigned int nr_events = 0; |
2b2ed975 | 2204 | int iters = 0, ret = 0; |
500f9fba | 2205 | |
c7849be9 XW |
2206 | /* |
2207 | * We disallow the app entering submit/complete with polling, but we | |
2208 | * still need to lock the ring to prevent racing with polled issue | |
2209 | * that got punted to a workqueue. | |
2210 | */ | |
2211 | mutex_lock(&ctx->uring_lock); | |
def596e9 | 2212 | do { |
a3a0e43f JA |
2213 | /* |
2214 | * Don't enter poll loop if we already have events pending. | |
2215 | * If we do, we can potentially be spinning for commands that | |
2216 | * already triggered a CQE (eg in error). | |
2217 | */ | |
1d7bb1d5 | 2218 | if (io_cqring_events(ctx, false)) |
a3a0e43f JA |
2219 | break; |
2220 | ||
500f9fba JA |
2221 | /* |
2222 | * If a submit got punted to a workqueue, we can have the | |
2223 | * application entering polling for a command before it gets | |
2224 | * issued. That app will hold the uring_lock for the duration | |
2225 | * of the poll right here, so we need to take a breather every | |
2226 | * now and then to ensure that the issue has a chance to add | |
2227 | * the poll to the issued list. Otherwise we can spin here | |
2228 | * forever, while the workqueue is stuck trying to acquire the | |
2229 | * very same mutex. | |
2230 | */ | |
2231 | if (!(++iters & 7)) { | |
2232 | mutex_unlock(&ctx->uring_lock); | |
4c6e277c | 2233 | io_run_task_work(); |
500f9fba JA |
2234 | mutex_lock(&ctx->uring_lock); |
2235 | } | |
2236 | ||
7668b92a | 2237 | ret = io_iopoll_getevents(ctx, &nr_events, min); |
def596e9 JA |
2238 | if (ret <= 0) |
2239 | break; | |
2240 | ret = 0; | |
7668b92a | 2241 | } while (min && !nr_events && !need_resched()); |
def596e9 | 2242 | |
500f9fba | 2243 | mutex_unlock(&ctx->uring_lock); |
def596e9 JA |
2244 | return ret; |
2245 | } | |
2246 | ||
491381ce | 2247 | static void kiocb_end_write(struct io_kiocb *req) |
2b188cc1 | 2248 | { |
491381ce JA |
2249 | /* |
2250 | * Tell lockdep we inherited freeze protection from submission | |
2251 | * thread. | |
2252 | */ | |
2253 | if (req->flags & REQ_F_ISREG) { | |
2254 | struct inode *inode = file_inode(req->file); | |
2b188cc1 | 2255 | |
491381ce | 2256 | __sb_writers_acquired(inode->i_sb, SB_FREEZE_WRITE); |
2b188cc1 | 2257 | } |
491381ce | 2258 | file_end_write(req->file); |
2b188cc1 JA |
2259 | } |
2260 | ||
a1d7c393 JA |
2261 | static void io_complete_rw_common(struct kiocb *kiocb, long res, |
2262 | struct io_comp_state *cs) | |
2b188cc1 | 2263 | { |
9adbd45d | 2264 | struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb); |
bcda7baa | 2265 | int cflags = 0; |
2b188cc1 | 2266 | |
491381ce JA |
2267 | if (kiocb->ki_flags & IOCB_WRITE) |
2268 | kiocb_end_write(req); | |
2b188cc1 | 2269 | |
4e88d6e7 JA |
2270 | if (res != req->result) |
2271 | req_set_fail_links(req); | |
bcda7baa | 2272 | if (req->flags & REQ_F_BUFFER_SELECTED) |
8ff069bf | 2273 | cflags = io_put_rw_kbuf(req); |
a1d7c393 | 2274 | __io_req_complete(req, res, cflags, cs); |
ba816ad6 JA |
2275 | } |
2276 | ||
b63534c4 JA |
2277 | #ifdef CONFIG_BLOCK |
2278 | static bool io_resubmit_prep(struct io_kiocb *req, int error) | |
2279 | { | |
2280 | struct iovec inline_vecs[UIO_FASTIOV], *iovec = inline_vecs; | |
2281 | ssize_t ret = -ECANCELED; | |
2282 | struct iov_iter iter; | |
2283 | int rw; | |
2284 | ||
2285 | if (error) { | |
2286 | ret = error; | |
2287 | goto end_req; | |
2288 | } | |
2289 | ||
2290 | switch (req->opcode) { | |
2291 | case IORING_OP_READV: | |
2292 | case IORING_OP_READ_FIXED: | |
2293 | case IORING_OP_READ: | |
2294 | rw = READ; | |
2295 | break; | |
2296 | case IORING_OP_WRITEV: | |
2297 | case IORING_OP_WRITE_FIXED: | |
2298 | case IORING_OP_WRITE: | |
2299 | rw = WRITE; | |
2300 | break; | |
2301 | default: | |
2302 | printk_once(KERN_WARNING "io_uring: bad opcode in resubmit %d\n", | |
2303 | req->opcode); | |
2304 | goto end_req; | |
2305 | } | |
2306 | ||
8f3d7496 JA |
2307 | if (!req->io) { |
2308 | ret = io_import_iovec(rw, req, &iovec, &iter, false); | |
2309 | if (ret < 0) | |
2310 | goto end_req; | |
2311 | ret = io_setup_async_rw(req, iovec, inline_vecs, &iter, false); | |
2312 | if (!ret) | |
2313 | return true; | |
2314 | kfree(iovec); | |
2315 | } else { | |
b63534c4 | 2316 | return true; |
8f3d7496 | 2317 | } |
b63534c4 | 2318 | end_req: |
b63534c4 | 2319 | req_set_fail_links(req); |
e1e16097 | 2320 | io_req_complete(req, ret); |
b63534c4 JA |
2321 | return false; |
2322 | } | |
b63534c4 JA |
2323 | #endif |
2324 | ||
2325 | static bool io_rw_reissue(struct io_kiocb *req, long res) | |
2326 | { | |
2327 | #ifdef CONFIG_BLOCK | |
355afaeb | 2328 | umode_t mode = file_inode(req->file)->i_mode; |
b63534c4 JA |
2329 | int ret; |
2330 | ||
355afaeb JA |
2331 | if (!S_ISBLK(mode) && !S_ISREG(mode)) |
2332 | return false; | |
b63534c4 JA |
2333 | if ((res != -EAGAIN && res != -EOPNOTSUPP) || io_wq_current_is_worker()) |
2334 | return false; | |
2335 | ||
fdee946d | 2336 | ret = io_sq_thread_acquire_mm(req->ctx, req); |
6d816e08 | 2337 | |
fdee946d JA |
2338 | if (io_resubmit_prep(req, ret)) { |
2339 | refcount_inc(&req->refs); | |
2340 | io_queue_async_work(req); | |
b63534c4 | 2341 | return true; |
fdee946d JA |
2342 | } |
2343 | ||
b63534c4 JA |
2344 | #endif |
2345 | return false; | |
2346 | } | |
2347 | ||
a1d7c393 JA |
2348 | static void __io_complete_rw(struct io_kiocb *req, long res, long res2, |
2349 | struct io_comp_state *cs) | |
2350 | { | |
2351 | if (!io_rw_reissue(req, res)) | |
2352 | io_complete_rw_common(&req->rw.kiocb, res, cs); | |
ba816ad6 JA |
2353 | } |
2354 | ||
2355 | static void io_complete_rw(struct kiocb *kiocb, long res, long res2) | |
2356 | { | |
9adbd45d | 2357 | struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb); |
ba816ad6 | 2358 | |
a1d7c393 | 2359 | __io_complete_rw(req, res, res2, NULL); |
2b188cc1 JA |
2360 | } |
2361 | ||
def596e9 JA |
2362 | static void io_complete_rw_iopoll(struct kiocb *kiocb, long res, long res2) |
2363 | { | |
9adbd45d | 2364 | struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb); |
def596e9 | 2365 | |
491381ce JA |
2366 | if (kiocb->ki_flags & IOCB_WRITE) |
2367 | kiocb_end_write(req); | |
def596e9 | 2368 | |
2d7d6792 | 2369 | if (res != -EAGAIN && res != req->result) |
4e88d6e7 | 2370 | req_set_fail_links(req); |
bbde017a XW |
2371 | |
2372 | WRITE_ONCE(req->result, res); | |
2373 | /* order with io_poll_complete() checking ->result */ | |
cd664b0e PB |
2374 | smp_wmb(); |
2375 | WRITE_ONCE(req->iopoll_completed, 1); | |
def596e9 JA |
2376 | } |
2377 | ||
2378 | /* | |
2379 | * After the iocb has been issued, it's safe to be found on the poll list. | |
2380 | * Adding the kiocb to the list AFTER submission ensures that we don't | |
2381 | * find it from a io_iopoll_getevents() thread before the issuer is done | |
2382 | * accessing the kiocb cookie. | |
2383 | */ | |
2384 | static void io_iopoll_req_issued(struct io_kiocb *req) | |
2385 | { | |
2386 | struct io_ring_ctx *ctx = req->ctx; | |
2387 | ||
2388 | /* | |
2389 | * Track whether we have multiple files in our lists. This will impact | |
2390 | * how we do polling eventually, not spinning if we're on potentially | |
2391 | * different devices. | |
2392 | */ | |
540e32a0 | 2393 | if (list_empty(&ctx->iopoll_list)) { |
def596e9 JA |
2394 | ctx->poll_multi_file = false; |
2395 | } else if (!ctx->poll_multi_file) { | |
2396 | struct io_kiocb *list_req; | |
2397 | ||
540e32a0 | 2398 | list_req = list_first_entry(&ctx->iopoll_list, struct io_kiocb, |
d21ffe7e | 2399 | inflight_entry); |
9adbd45d | 2400 | if (list_req->file != req->file) |
def596e9 JA |
2401 | ctx->poll_multi_file = true; |
2402 | } | |
2403 | ||
2404 | /* | |
2405 | * For fast devices, IO may have already completed. If it has, add | |
2406 | * it to the front so we find it first. | |
2407 | */ | |
65a6543d | 2408 | if (READ_ONCE(req->iopoll_completed)) |
d21ffe7e | 2409 | list_add(&req->inflight_entry, &ctx->iopoll_list); |
def596e9 | 2410 | else |
d21ffe7e | 2411 | list_add_tail(&req->inflight_entry, &ctx->iopoll_list); |
bdcd3eab XW |
2412 | |
2413 | if ((ctx->flags & IORING_SETUP_SQPOLL) && | |
2414 | wq_has_sleeper(&ctx->sqo_wait)) | |
2415 | wake_up(&ctx->sqo_wait); | |
def596e9 JA |
2416 | } |
2417 | ||
9f13c35b | 2418 | static void __io_state_file_put(struct io_submit_state *state) |
9a56a232 | 2419 | { |
06ef3608 PB |
2420 | if (state->has_refs) |
2421 | fput_many(state->file, state->has_refs); | |
9f13c35b PB |
2422 | state->file = NULL; |
2423 | } | |
2424 | ||
2425 | static inline void io_state_file_put(struct io_submit_state *state) | |
2426 | { | |
2427 | if (state->file) | |
2428 | __io_state_file_put(state); | |
9a56a232 JA |
2429 | } |
2430 | ||
2431 | /* | |
2432 | * Get as many references to a file as we have IOs left in this submission, | |
2433 | * assuming most submissions are for one file, or at least that each file | |
2434 | * has more than one submission. | |
2435 | */ | |
8da11c19 | 2436 | static struct file *__io_file_get(struct io_submit_state *state, int fd) |
9a56a232 JA |
2437 | { |
2438 | if (!state) | |
2439 | return fget(fd); | |
2440 | ||
2441 | if (state->file) { | |
2442 | if (state->fd == fd) { | |
06ef3608 | 2443 | state->has_refs--; |
9a56a232 JA |
2444 | state->ios_left--; |
2445 | return state->file; | |
2446 | } | |
9f13c35b | 2447 | __io_state_file_put(state); |
9a56a232 JA |
2448 | } |
2449 | state->file = fget_many(fd, state->ios_left); | |
2450 | if (!state->file) | |
2451 | return NULL; | |
2452 | ||
2453 | state->fd = fd; | |
9a56a232 | 2454 | state->ios_left--; |
06ef3608 | 2455 | state->has_refs = state->ios_left; |
9a56a232 JA |
2456 | return state->file; |
2457 | } | |
2458 | ||
4503b767 JA |
2459 | static bool io_bdev_nowait(struct block_device *bdev) |
2460 | { | |
2461 | #ifdef CONFIG_BLOCK | |
2462 | return !bdev || queue_is_mq(bdev_get_queue(bdev)); | |
2463 | #else | |
2464 | return true; | |
2465 | #endif | |
2466 | } | |
2467 | ||
2b188cc1 JA |
2468 | /* |
2469 | * If we tracked the file through the SCM inflight mechanism, we could support | |
2470 | * any file. For now, just ensure that anything potentially problematic is done | |
2471 | * inline. | |
2472 | */ | |
af197f50 | 2473 | static bool io_file_supports_async(struct file *file, int rw) |
2b188cc1 JA |
2474 | { |
2475 | umode_t mode = file_inode(file)->i_mode; | |
2476 | ||
4503b767 JA |
2477 | if (S_ISBLK(mode)) { |
2478 | if (io_bdev_nowait(file->f_inode->i_bdev)) | |
2479 | return true; | |
2480 | return false; | |
2481 | } | |
2482 | if (S_ISCHR(mode) || S_ISSOCK(mode)) | |
2b188cc1 | 2483 | return true; |
4503b767 JA |
2484 | if (S_ISREG(mode)) { |
2485 | if (io_bdev_nowait(file->f_inode->i_sb->s_bdev) && | |
2486 | file->f_op != &io_uring_fops) | |
2487 | return true; | |
2488 | return false; | |
2489 | } | |
2b188cc1 | 2490 | |
c5b85625 JA |
2491 | /* any ->read/write should understand O_NONBLOCK */ |
2492 | if (file->f_flags & O_NONBLOCK) | |
2493 | return true; | |
2494 | ||
af197f50 JA |
2495 | if (!(file->f_mode & FMODE_NOWAIT)) |
2496 | return false; | |
2497 | ||
2498 | if (rw == READ) | |
2499 | return file->f_op->read_iter != NULL; | |
2500 | ||
2501 | return file->f_op->write_iter != NULL; | |
2b188cc1 JA |
2502 | } |
2503 | ||
3529d8c2 JA |
2504 | static int io_prep_rw(struct io_kiocb *req, const struct io_uring_sqe *sqe, |
2505 | bool force_nonblock) | |
2b188cc1 | 2506 | { |
def596e9 | 2507 | struct io_ring_ctx *ctx = req->ctx; |
9adbd45d | 2508 | struct kiocb *kiocb = &req->rw.kiocb; |
09bb8394 JA |
2509 | unsigned ioprio; |
2510 | int ret; | |
2b188cc1 | 2511 | |
491381ce JA |
2512 | if (S_ISREG(file_inode(req->file)->i_mode)) |
2513 | req->flags |= REQ_F_ISREG; | |
2514 | ||
2b188cc1 | 2515 | kiocb->ki_pos = READ_ONCE(sqe->off); |
ba04291e JA |
2516 | if (kiocb->ki_pos == -1 && !(req->file->f_mode & FMODE_STREAM)) { |
2517 | req->flags |= REQ_F_CUR_POS; | |
2518 | kiocb->ki_pos = req->file->f_pos; | |
2519 | } | |
2b188cc1 | 2520 | kiocb->ki_hint = ki_hint_validate(file_write_hint(kiocb->ki_filp)); |
3e577dcd PB |
2521 | kiocb->ki_flags = iocb_flags(kiocb->ki_filp); |
2522 | ret = kiocb_set_rw_flags(kiocb, READ_ONCE(sqe->rw_flags)); | |
2523 | if (unlikely(ret)) | |
2524 | return ret; | |
2b188cc1 JA |
2525 | |
2526 | ioprio = READ_ONCE(sqe->ioprio); | |
2527 | if (ioprio) { | |
2528 | ret = ioprio_check_cap(ioprio); | |
2529 | if (ret) | |
09bb8394 | 2530 | return ret; |
2b188cc1 JA |
2531 | |
2532 | kiocb->ki_ioprio = ioprio; | |
2533 | } else | |
2534 | kiocb->ki_ioprio = get_current_ioprio(); | |
2535 | ||
8449eeda | 2536 | /* don't allow async punt if RWF_NOWAIT was requested */ |
c5b85625 | 2537 | if (kiocb->ki_flags & IOCB_NOWAIT) |
8449eeda SB |
2538 | req->flags |= REQ_F_NOWAIT; |
2539 | ||
2540 | if (force_nonblock) | |
2b188cc1 | 2541 | kiocb->ki_flags |= IOCB_NOWAIT; |
8449eeda | 2542 | |
def596e9 | 2543 | if (ctx->flags & IORING_SETUP_IOPOLL) { |
def596e9 JA |
2544 | if (!(kiocb->ki_flags & IOCB_DIRECT) || |
2545 | !kiocb->ki_filp->f_op->iopoll) | |
09bb8394 | 2546 | return -EOPNOTSUPP; |
2b188cc1 | 2547 | |
def596e9 JA |
2548 | kiocb->ki_flags |= IOCB_HIPRI; |
2549 | kiocb->ki_complete = io_complete_rw_iopoll; | |
65a6543d | 2550 | req->iopoll_completed = 0; |
def596e9 | 2551 | } else { |
09bb8394 JA |
2552 | if (kiocb->ki_flags & IOCB_HIPRI) |
2553 | return -EINVAL; | |
def596e9 JA |
2554 | kiocb->ki_complete = io_complete_rw; |
2555 | } | |
9adbd45d | 2556 | |
3529d8c2 JA |
2557 | req->rw.addr = READ_ONCE(sqe->addr); |
2558 | req->rw.len = READ_ONCE(sqe->len); | |
4f4eeba8 | 2559 | req->buf_index = READ_ONCE(sqe->buf_index); |
2b188cc1 | 2560 | return 0; |
2b188cc1 JA |
2561 | } |
2562 | ||
2563 | static inline void io_rw_done(struct kiocb *kiocb, ssize_t ret) | |
2564 | { | |
2565 | switch (ret) { | |
2566 | case -EIOCBQUEUED: | |
2567 | break; | |
2568 | case -ERESTARTSYS: | |
2569 | case -ERESTARTNOINTR: | |
2570 | case -ERESTARTNOHAND: | |
2571 | case -ERESTART_RESTARTBLOCK: | |
2572 | /* | |
2573 | * We can't just restart the syscall, since previously | |
2574 | * submitted sqes may already be in progress. Just fail this | |
2575 | * IO with EINTR. | |
2576 | */ | |
2577 | ret = -EINTR; | |
df561f66 | 2578 | fallthrough; |
2b188cc1 JA |
2579 | default: |
2580 | kiocb->ki_complete(kiocb, ret, 0); | |
2581 | } | |
2582 | } | |
2583 | ||
a1d7c393 JA |
2584 | static void kiocb_done(struct kiocb *kiocb, ssize_t ret, |
2585 | struct io_comp_state *cs) | |
ba816ad6 | 2586 | { |
ba04291e JA |
2587 | struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb); |
2588 | ||
227c0c96 JA |
2589 | /* add previously done IO, if any */ |
2590 | if (req->io && req->io->rw.bytes_done > 0) { | |
2591 | if (ret < 0) | |
2592 | ret = req->io->rw.bytes_done; | |
2593 | else | |
2594 | ret += req->io->rw.bytes_done; | |
2595 | } | |
2596 | ||
ba04291e JA |
2597 | if (req->flags & REQ_F_CUR_POS) |
2598 | req->file->f_pos = kiocb->ki_pos; | |
bcaec089 | 2599 | if (ret >= 0 && kiocb->ki_complete == io_complete_rw) |
a1d7c393 | 2600 | __io_complete_rw(req, ret, 0, cs); |
ba816ad6 JA |
2601 | else |
2602 | io_rw_done(kiocb, ret); | |
2603 | } | |
2604 | ||
9adbd45d | 2605 | static ssize_t io_import_fixed(struct io_kiocb *req, int rw, |
7d009165 | 2606 | struct iov_iter *iter) |
edafccee | 2607 | { |
9adbd45d JA |
2608 | struct io_ring_ctx *ctx = req->ctx; |
2609 | size_t len = req->rw.len; | |
edafccee | 2610 | struct io_mapped_ubuf *imu; |
4f4eeba8 | 2611 | u16 index, buf_index; |
edafccee JA |
2612 | size_t offset; |
2613 | u64 buf_addr; | |
2614 | ||
2615 | /* attempt to use fixed buffers without having provided iovecs */ | |
2616 | if (unlikely(!ctx->user_bufs)) | |
2617 | return -EFAULT; | |
2618 | ||
4f4eeba8 | 2619 | buf_index = req->buf_index; |
edafccee JA |
2620 | if (unlikely(buf_index >= ctx->nr_user_bufs)) |
2621 | return -EFAULT; | |
2622 | ||
2623 | index = array_index_nospec(buf_index, ctx->nr_user_bufs); | |
2624 | imu = &ctx->user_bufs[index]; | |
9adbd45d | 2625 | buf_addr = req->rw.addr; |
edafccee JA |
2626 | |
2627 | /* overflow */ | |
2628 | if (buf_addr + len < buf_addr) | |
2629 | return -EFAULT; | |
2630 | /* not inside the mapped region */ | |
2631 | if (buf_addr < imu->ubuf || buf_addr + len > imu->ubuf + imu->len) | |
2632 | return -EFAULT; | |
2633 | ||
2634 | /* | |
2635 | * May not be a start of buffer, set size appropriately | |
2636 | * and advance us to the beginning. | |
2637 | */ | |
2638 | offset = buf_addr - imu->ubuf; | |
2639 | iov_iter_bvec(iter, rw, imu->bvec, imu->nr_bvecs, offset + len); | |
bd11b3a3 JA |
2640 | |
2641 | if (offset) { | |
2642 | /* | |
2643 | * Don't use iov_iter_advance() here, as it's really slow for | |
2644 | * using the latter parts of a big fixed buffer - it iterates | |
2645 | * over each segment manually. We can cheat a bit here, because | |
2646 | * we know that: | |
2647 | * | |
2648 | * 1) it's a BVEC iter, we set it up | |
2649 | * 2) all bvecs are PAGE_SIZE in size, except potentially the | |
2650 | * first and last bvec | |
2651 | * | |
2652 | * So just find our index, and adjust the iterator afterwards. | |
2653 | * If the offset is within the first bvec (or the whole first | |
2654 | * bvec, just use iov_iter_advance(). This makes it easier | |
2655 | * since we can just skip the first segment, which may not | |
2656 | * be PAGE_SIZE aligned. | |
2657 | */ | |
2658 | const struct bio_vec *bvec = imu->bvec; | |
2659 | ||
2660 | if (offset <= bvec->bv_len) { | |
2661 | iov_iter_advance(iter, offset); | |
2662 | } else { | |
2663 | unsigned long seg_skip; | |
2664 | ||
2665 | /* skip first vec */ | |
2666 | offset -= bvec->bv_len; | |
2667 | seg_skip = 1 + (offset >> PAGE_SHIFT); | |
2668 | ||
2669 | iter->bvec = bvec + seg_skip; | |
2670 | iter->nr_segs -= seg_skip; | |
99c79f66 | 2671 | iter->count -= bvec->bv_len + offset; |
bd11b3a3 | 2672 | iter->iov_offset = offset & ~PAGE_MASK; |
bd11b3a3 JA |
2673 | } |
2674 | } | |
2675 | ||
5e559561 | 2676 | return len; |
edafccee JA |
2677 | } |
2678 | ||
bcda7baa JA |
2679 | static void io_ring_submit_unlock(struct io_ring_ctx *ctx, bool needs_lock) |
2680 | { | |
2681 | if (needs_lock) | |
2682 | mutex_unlock(&ctx->uring_lock); | |
2683 | } | |
2684 | ||
2685 | static void io_ring_submit_lock(struct io_ring_ctx *ctx, bool needs_lock) | |
2686 | { | |
2687 | /* | |
2688 | * "Normal" inline submissions always hold the uring_lock, since we | |
2689 | * grab it from the system call. Same is true for the SQPOLL offload. | |
2690 | * The only exception is when we've detached the request and issue it | |
2691 | * from an async worker thread, grab the lock for that case. | |
2692 | */ | |
2693 | if (needs_lock) | |
2694 | mutex_lock(&ctx->uring_lock); | |
2695 | } | |
2696 | ||
2697 | static struct io_buffer *io_buffer_select(struct io_kiocb *req, size_t *len, | |
2698 | int bgid, struct io_buffer *kbuf, | |
2699 | bool needs_lock) | |
2700 | { | |
2701 | struct io_buffer *head; | |
2702 | ||
2703 | if (req->flags & REQ_F_BUFFER_SELECTED) | |
2704 | return kbuf; | |
2705 | ||
2706 | io_ring_submit_lock(req->ctx, needs_lock); | |
2707 | ||
2708 | lockdep_assert_held(&req->ctx->uring_lock); | |
2709 | ||
2710 | head = idr_find(&req->ctx->io_buffer_idr, bgid); | |
2711 | if (head) { | |
2712 | if (!list_empty(&head->list)) { | |
2713 | kbuf = list_last_entry(&head->list, struct io_buffer, | |
2714 | list); | |
2715 | list_del(&kbuf->list); | |
2716 | } else { | |
2717 | kbuf = head; | |
2718 | idr_remove(&req->ctx->io_buffer_idr, bgid); | |
2719 | } | |
2720 | if (*len > kbuf->len) | |
2721 | *len = kbuf->len; | |
2722 | } else { | |
2723 | kbuf = ERR_PTR(-ENOBUFS); | |
2724 | } | |
2725 | ||
2726 | io_ring_submit_unlock(req->ctx, needs_lock); | |
2727 | ||
2728 | return kbuf; | |
2729 | } | |
2730 | ||
4d954c25 JA |
2731 | static void __user *io_rw_buffer_select(struct io_kiocb *req, size_t *len, |
2732 | bool needs_lock) | |
2733 | { | |
2734 | struct io_buffer *kbuf; | |
4f4eeba8 | 2735 | u16 bgid; |
4d954c25 JA |
2736 | |
2737 | kbuf = (struct io_buffer *) (unsigned long) req->rw.addr; | |
4f4eeba8 | 2738 | bgid = req->buf_index; |
4d954c25 JA |
2739 | kbuf = io_buffer_select(req, len, bgid, kbuf, needs_lock); |
2740 | if (IS_ERR(kbuf)) | |
2741 | return kbuf; | |
2742 | req->rw.addr = (u64) (unsigned long) kbuf; | |
2743 | req->flags |= REQ_F_BUFFER_SELECTED; | |
2744 | return u64_to_user_ptr(kbuf->addr); | |
2745 | } | |
2746 | ||
2747 | #ifdef CONFIG_COMPAT | |
2748 | static ssize_t io_compat_import(struct io_kiocb *req, struct iovec *iov, | |
2749 | bool needs_lock) | |
2750 | { | |
2751 | struct compat_iovec __user *uiov; | |
2752 | compat_ssize_t clen; | |
2753 | void __user *buf; | |
2754 | ssize_t len; | |
2755 | ||
2756 | uiov = u64_to_user_ptr(req->rw.addr); | |
2757 | if (!access_ok(uiov, sizeof(*uiov))) | |
2758 | return -EFAULT; | |
2759 | if (__get_user(clen, &uiov->iov_len)) | |
2760 | return -EFAULT; | |
2761 | if (clen < 0) | |
2762 | return -EINVAL; | |
2763 | ||
2764 | len = clen; | |
2765 | buf = io_rw_buffer_select(req, &len, needs_lock); | |
2766 | if (IS_ERR(buf)) | |
2767 | return PTR_ERR(buf); | |
2768 | iov[0].iov_base = buf; | |
2769 | iov[0].iov_len = (compat_size_t) len; | |
2770 | return 0; | |
2771 | } | |
2772 | #endif | |
2773 | ||
2774 | static ssize_t __io_iov_buffer_select(struct io_kiocb *req, struct iovec *iov, | |
2775 | bool needs_lock) | |
2776 | { | |
2777 | struct iovec __user *uiov = u64_to_user_ptr(req->rw.addr); | |
2778 | void __user *buf; | |
2779 | ssize_t len; | |
2780 | ||
2781 | if (copy_from_user(iov, uiov, sizeof(*uiov))) | |
2782 | return -EFAULT; | |
2783 | ||
2784 | len = iov[0].iov_len; | |
2785 | if (len < 0) | |
2786 | return -EINVAL; | |
2787 | buf = io_rw_buffer_select(req, &len, needs_lock); | |
2788 | if (IS_ERR(buf)) | |
2789 | return PTR_ERR(buf); | |
2790 | iov[0].iov_base = buf; | |
2791 | iov[0].iov_len = len; | |
2792 | return 0; | |
2793 | } | |
2794 | ||
2795 | static ssize_t io_iov_buffer_select(struct io_kiocb *req, struct iovec *iov, | |
2796 | bool needs_lock) | |
2797 | { | |
dddb3e26 JA |
2798 | if (req->flags & REQ_F_BUFFER_SELECTED) { |
2799 | struct io_buffer *kbuf; | |
2800 | ||
2801 | kbuf = (struct io_buffer *) (unsigned long) req->rw.addr; | |
2802 | iov[0].iov_base = u64_to_user_ptr(kbuf->addr); | |
2803 | iov[0].iov_len = kbuf->len; | |
4d954c25 | 2804 | return 0; |
dddb3e26 | 2805 | } |
4d954c25 JA |
2806 | if (!req->rw.len) |
2807 | return 0; | |
2808 | else if (req->rw.len > 1) | |
2809 | return -EINVAL; | |
2810 | ||
2811 | #ifdef CONFIG_COMPAT | |
2812 | if (req->ctx->compat) | |
2813 | return io_compat_import(req, iov, needs_lock); | |
2814 | #endif | |
2815 | ||
2816 | return __io_iov_buffer_select(req, iov, needs_lock); | |
2817 | } | |
2818 | ||
8452fd0c JA |
2819 | static ssize_t __io_import_iovec(int rw, struct io_kiocb *req, |
2820 | struct iovec **iovec, struct iov_iter *iter, | |
2821 | bool needs_lock) | |
2b188cc1 | 2822 | { |
9adbd45d JA |
2823 | void __user *buf = u64_to_user_ptr(req->rw.addr); |
2824 | size_t sqe_len = req->rw.len; | |
4d954c25 | 2825 | ssize_t ret; |
edafccee JA |
2826 | u8 opcode; |
2827 | ||
d625c6ee | 2828 | opcode = req->opcode; |
7d009165 | 2829 | if (opcode == IORING_OP_READ_FIXED || opcode == IORING_OP_WRITE_FIXED) { |
edafccee | 2830 | *iovec = NULL; |
9adbd45d | 2831 | return io_import_fixed(req, rw, iter); |
edafccee | 2832 | } |
2b188cc1 | 2833 | |
bcda7baa | 2834 | /* buffer index only valid with fixed read/write, or buffer select */ |
4f4eeba8 | 2835 | if (req->buf_index && !(req->flags & REQ_F_BUFFER_SELECT)) |
9adbd45d JA |
2836 | return -EINVAL; |
2837 | ||
3a6820f2 | 2838 | if (opcode == IORING_OP_READ || opcode == IORING_OP_WRITE) { |
bcda7baa | 2839 | if (req->flags & REQ_F_BUFFER_SELECT) { |
4d954c25 | 2840 | buf = io_rw_buffer_select(req, &sqe_len, needs_lock); |
867a23ea | 2841 | if (IS_ERR(buf)) |
4d954c25 | 2842 | return PTR_ERR(buf); |
3f9d6441 | 2843 | req->rw.len = sqe_len; |
bcda7baa JA |
2844 | } |
2845 | ||
3a6820f2 JA |
2846 | ret = import_single_range(rw, buf, sqe_len, *iovec, iter); |
2847 | *iovec = NULL; | |
3a901598 | 2848 | return ret < 0 ? ret : sqe_len; |
3a6820f2 JA |
2849 | } |
2850 | ||
4d954c25 JA |
2851 | if (req->flags & REQ_F_BUFFER_SELECT) { |
2852 | ret = io_iov_buffer_select(req, *iovec, needs_lock); | |
3f9d6441 JA |
2853 | if (!ret) { |
2854 | ret = (*iovec)->iov_len; | |
2855 | iov_iter_init(iter, rw, *iovec, 1, ret); | |
2856 | } | |
4d954c25 JA |
2857 | *iovec = NULL; |
2858 | return ret; | |
2859 | } | |
2860 | ||
2b188cc1 | 2861 | #ifdef CONFIG_COMPAT |
cf6fd4bd | 2862 | if (req->ctx->compat) |
2b188cc1 JA |
2863 | return compat_import_iovec(rw, buf, sqe_len, UIO_FASTIOV, |
2864 | iovec, iter); | |
2865 | #endif | |
2866 | ||
2867 | return import_iovec(rw, buf, sqe_len, UIO_FASTIOV, iovec, iter); | |
2868 | } | |
2869 | ||
8452fd0c JA |
2870 | static ssize_t io_import_iovec(int rw, struct io_kiocb *req, |
2871 | struct iovec **iovec, struct iov_iter *iter, | |
2872 | bool needs_lock) | |
2873 | { | |
2874 | if (!req->io) | |
2875 | return __io_import_iovec(rw, req, iovec, iter, needs_lock); | |
2876 | *iovec = NULL; | |
2877 | return iov_iter_count(&req->io->rw.iter); | |
2878 | } | |
2879 | ||
0fef9483 JA |
2880 | static inline loff_t *io_kiocb_ppos(struct kiocb *kiocb) |
2881 | { | |
2882 | return kiocb->ki_filp->f_mode & FMODE_STREAM ? NULL : &kiocb->ki_pos; | |
2883 | } | |
2884 | ||
31b51510 | 2885 | /* |
32960613 JA |
2886 | * For files that don't have ->read_iter() and ->write_iter(), handle them |
2887 | * by looping over ->read() or ->write() manually. | |
31b51510 | 2888 | */ |
32960613 JA |
2889 | static ssize_t loop_rw_iter(int rw, struct file *file, struct kiocb *kiocb, |
2890 | struct iov_iter *iter) | |
2891 | { | |
2892 | ssize_t ret = 0; | |
2893 | ||
2894 | /* | |
2895 | * Don't support polled IO through this interface, and we can't | |
2896 | * support non-blocking either. For the latter, this just causes | |
2897 | * the kiocb to be handled from an async context. | |
2898 | */ | |
2899 | if (kiocb->ki_flags & IOCB_HIPRI) | |
2900 | return -EOPNOTSUPP; | |
2901 | if (kiocb->ki_flags & IOCB_NOWAIT) | |
2902 | return -EAGAIN; | |
2903 | ||
2904 | while (iov_iter_count(iter)) { | |
311ae9e1 | 2905 | struct iovec iovec; |
32960613 JA |
2906 | ssize_t nr; |
2907 | ||
311ae9e1 PB |
2908 | if (!iov_iter_is_bvec(iter)) { |
2909 | iovec = iov_iter_iovec(iter); | |
2910 | } else { | |
2911 | /* fixed buffers import bvec */ | |
2912 | iovec.iov_base = kmap(iter->bvec->bv_page) | |
2913 | + iter->iov_offset; | |
2914 | iovec.iov_len = min(iter->count, | |
2915 | iter->bvec->bv_len - iter->iov_offset); | |
2916 | } | |
2917 | ||
32960613 JA |
2918 | if (rw == READ) { |
2919 | nr = file->f_op->read(file, iovec.iov_base, | |
0fef9483 | 2920 | iovec.iov_len, io_kiocb_ppos(kiocb)); |
32960613 JA |
2921 | } else { |
2922 | nr = file->f_op->write(file, iovec.iov_base, | |
0fef9483 | 2923 | iovec.iov_len, io_kiocb_ppos(kiocb)); |
32960613 JA |
2924 | } |
2925 | ||
311ae9e1 PB |
2926 | if (iov_iter_is_bvec(iter)) |
2927 | kunmap(iter->bvec->bv_page); | |
2928 | ||
32960613 JA |
2929 | if (nr < 0) { |
2930 | if (!ret) | |
2931 | ret = nr; | |
2932 | break; | |
2933 | } | |
2934 | ret += nr; | |
2935 | if (nr != iovec.iov_len) | |
2936 | break; | |
2937 | iov_iter_advance(iter, nr); | |
2938 | } | |
2939 | ||
2940 | return ret; | |
2941 | } | |
2942 | ||
ff6165b2 JA |
2943 | static void io_req_map_rw(struct io_kiocb *req, const struct iovec *iovec, |
2944 | const struct iovec *fast_iov, struct iov_iter *iter) | |
f67676d1 | 2945 | { |
b64e3444 PB |
2946 | struct io_async_rw *rw = &req->io->rw; |
2947 | ||
ff6165b2 JA |
2948 | memcpy(&rw->iter, iter, sizeof(*iter)); |
2949 | rw->free_iovec = NULL; | |
227c0c96 | 2950 | rw->bytes_done = 0; |
ff6165b2 JA |
2951 | /* can only be fixed buffers, no need to do anything */ |
2952 | if (iter->type == ITER_BVEC) | |
2953 | return; | |
b64e3444 | 2954 | if (!iovec) { |
ff6165b2 JA |
2955 | unsigned iov_off = 0; |
2956 | ||
2957 | rw->iter.iov = rw->fast_iov; | |
2958 | if (iter->iov != fast_iov) { | |
2959 | iov_off = iter->iov - fast_iov; | |
2960 | rw->iter.iov += iov_off; | |
2961 | } | |
2962 | if (rw->fast_iov != fast_iov) | |
2963 | memcpy(rw->fast_iov + iov_off, fast_iov + iov_off, | |
45097dae | 2964 | sizeof(struct iovec) * iter->nr_segs); |
99bc4c38 | 2965 | } else { |
ff6165b2 | 2966 | rw->free_iovec = iovec; |
99bc4c38 | 2967 | req->flags |= REQ_F_NEED_CLEANUP; |
f67676d1 JA |
2968 | } |
2969 | } | |
2970 | ||
3d9932a8 XW |
2971 | static inline int __io_alloc_async_ctx(struct io_kiocb *req) |
2972 | { | |
2973 | req->io = kmalloc(sizeof(*req->io), GFP_KERNEL); | |
2974 | return req->io == NULL; | |
2975 | } | |
2976 | ||
b7bb4f7d | 2977 | static int io_alloc_async_ctx(struct io_kiocb *req) |
f67676d1 | 2978 | { |
d3656344 JA |
2979 | if (!io_op_defs[req->opcode].async_ctx) |
2980 | return 0; | |
3d9932a8 XW |
2981 | |
2982 | return __io_alloc_async_ctx(req); | |
b7bb4f7d JA |
2983 | } |
2984 | ||
ff6165b2 JA |
2985 | static int io_setup_async_rw(struct io_kiocb *req, const struct iovec *iovec, |
2986 | const struct iovec *fast_iov, | |
227c0c96 | 2987 | struct iov_iter *iter, bool force) |
b7bb4f7d | 2988 | { |
227c0c96 | 2989 | if (!force && !io_op_defs[req->opcode].async_ctx) |
74566df3 | 2990 | return 0; |
5d204bcf | 2991 | if (!req->io) { |
3d9932a8 | 2992 | if (__io_alloc_async_ctx(req)) |
5d204bcf | 2993 | return -ENOMEM; |
b7bb4f7d | 2994 | |
ff6165b2 | 2995 | io_req_map_rw(req, iovec, fast_iov, iter); |
5d204bcf | 2996 | } |
b7bb4f7d | 2997 | return 0; |
f67676d1 JA |
2998 | } |
2999 | ||
c3e330a4 PB |
3000 | static inline int io_rw_prep_async(struct io_kiocb *req, int rw, |
3001 | bool force_nonblock) | |
3002 | { | |
ff6165b2 | 3003 | struct io_async_rw *iorw = &req->io->rw; |
c183edff | 3004 | struct iovec *iov; |
c3e330a4 PB |
3005 | ssize_t ret; |
3006 | ||
c183edff JA |
3007 | iorw->iter.iov = iov = iorw->fast_iov; |
3008 | ret = __io_import_iovec(rw, req, &iov, &iorw->iter, !force_nonblock); | |
c3e330a4 PB |
3009 | if (unlikely(ret < 0)) |
3010 | return ret; | |
3011 | ||
c183edff | 3012 | iorw->iter.iov = iov; |
ff6165b2 | 3013 | io_req_map_rw(req, iorw->iter.iov, iorw->fast_iov, &iorw->iter); |
c3e330a4 PB |
3014 | return 0; |
3015 | } | |
3016 | ||
3529d8c2 JA |
3017 | static int io_read_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe, |
3018 | bool force_nonblock) | |
f67676d1 JA |
3019 | { |
3020 | ssize_t ret; | |
3021 | ||
3529d8c2 JA |
3022 | ret = io_prep_rw(req, sqe, force_nonblock); |
3023 | if (ret) | |
3024 | return ret; | |
f67676d1 | 3025 | |
3529d8c2 JA |
3026 | if (unlikely(!(req->file->f_mode & FMODE_READ))) |
3027 | return -EBADF; | |
f67676d1 | 3028 | |
5f798bea PB |
3029 | /* either don't need iovec imported or already have it */ |
3030 | if (!req->io || req->flags & REQ_F_NEED_CLEANUP) | |
3529d8c2 | 3031 | return 0; |
c3e330a4 | 3032 | return io_rw_prep_async(req, READ, force_nonblock); |
f67676d1 JA |
3033 | } |
3034 | ||
c1dd91d1 JA |
3035 | /* |
3036 | * This is our waitqueue callback handler, registered through lock_page_async() | |
3037 | * when we initially tried to do the IO with the iocb armed our waitqueue. | |
3038 | * This gets called when the page is unlocked, and we generally expect that to | |
3039 | * happen when the page IO is completed and the page is now uptodate. This will | |
3040 | * queue a task_work based retry of the operation, attempting to copy the data | |
3041 | * again. If the latter fails because the page was NOT uptodate, then we will | |
3042 | * do a thread based blocking retry of the operation. That's the unexpected | |
3043 | * slow path. | |
3044 | */ | |
bcf5a063 JA |
3045 | static int io_async_buf_func(struct wait_queue_entry *wait, unsigned mode, |
3046 | int sync, void *arg) | |
3047 | { | |
3048 | struct wait_page_queue *wpq; | |
3049 | struct io_kiocb *req = wait->private; | |
bcf5a063 | 3050 | struct wait_page_key *key = arg; |
bcf5a063 JA |
3051 | int ret; |
3052 | ||
3053 | wpq = container_of(wait, struct wait_page_queue, wait); | |
3054 | ||
cdc8fcb4 LT |
3055 | if (!wake_page_match(wpq, key)) |
3056 | return 0; | |
3057 | ||
c8d317aa | 3058 | req->rw.kiocb.ki_flags &= ~IOCB_WAITQ; |
bcf5a063 JA |
3059 | list_del_init(&wait->entry); |
3060 | ||
e7375122 | 3061 | init_task_work(&req->task_work, io_req_task_submit); |
6d816e08 JA |
3062 | percpu_ref_get(&req->ctx->refs); |
3063 | ||
bcf5a063 JA |
3064 | /* submit ref gets dropped, acquire a new one */ |
3065 | refcount_inc(&req->refs); | |
fd7d6de2 | 3066 | ret = io_req_task_work_add(req, &req->task_work, true); |
bcf5a063 | 3067 | if (unlikely(ret)) { |
c2c4c83c JA |
3068 | struct task_struct *tsk; |
3069 | ||
bcf5a063 | 3070 | /* queue just for cancelation */ |
e7375122 | 3071 | init_task_work(&req->task_work, io_req_task_cancel); |
bcf5a063 | 3072 | tsk = io_wq_get_task(req->ctx->io_wq); |
e7375122 | 3073 | task_work_add(tsk, &req->task_work, 0); |
c2c4c83c | 3074 | wake_up_process(tsk); |
bcf5a063 | 3075 | } |
bcf5a063 JA |
3076 | return 1; |
3077 | } | |
3078 | ||
c1dd91d1 JA |
3079 | /* |
3080 | * This controls whether a given IO request should be armed for async page | |
3081 | * based retry. If we return false here, the request is handed to the async | |
3082 | * worker threads for retry. If we're doing buffered reads on a regular file, | |
3083 | * we prepare a private wait_page_queue entry and retry the operation. This | |
3084 | * will either succeed because the page is now uptodate and unlocked, or it | |
3085 | * will register a callback when the page is unlocked at IO completion. Through | |
3086 | * that callback, io_uring uses task_work to setup a retry of the operation. | |
3087 | * That retry will attempt the buffered read again. The retry will generally | |
3088 | * succeed, or in rare cases where it fails, we then fall back to using the | |
3089 | * async worker threads for a blocking retry. | |
3090 | */ | |
227c0c96 | 3091 | static bool io_rw_should_retry(struct io_kiocb *req) |
f67676d1 | 3092 | { |
3b2a4439 | 3093 | struct wait_page_queue *wait = &req->io->rw.wpq; |
bcf5a063 | 3094 | struct kiocb *kiocb = &req->rw.kiocb; |
f67676d1 | 3095 | |
bcf5a063 JA |
3096 | /* never retry for NOWAIT, we just complete with -EAGAIN */ |
3097 | if (req->flags & REQ_F_NOWAIT) | |
3098 | return false; | |
f67676d1 | 3099 | |
227c0c96 | 3100 | /* Only for buffered IO */ |
3b2a4439 | 3101 | if (kiocb->ki_flags & (IOCB_DIRECT | IOCB_HIPRI)) |
bcf5a063 | 3102 | return false; |
3b2a4439 | 3103 | |
bcf5a063 JA |
3104 | /* |
3105 | * just use poll if we can, and don't attempt if the fs doesn't | |
3106 | * support callback based unlocks | |
3107 | */ | |
3108 | if (file_can_poll(req->file) || !(req->file->f_mode & FMODE_BUF_RASYNC)) | |
3109 | return false; | |
f67676d1 | 3110 | |
3b2a4439 JA |
3111 | wait->wait.func = io_async_buf_func; |
3112 | wait->wait.private = req; | |
3113 | wait->wait.flags = 0; | |
3114 | INIT_LIST_HEAD(&wait->wait.entry); | |
3115 | kiocb->ki_flags |= IOCB_WAITQ; | |
c8d317aa | 3116 | kiocb->ki_flags &= ~IOCB_NOWAIT; |
3b2a4439 | 3117 | kiocb->ki_waitq = wait; |
3b2a4439 | 3118 | return true; |
bcf5a063 JA |
3119 | } |
3120 | ||
3121 | static int io_iter_do_read(struct io_kiocb *req, struct iov_iter *iter) | |
3122 | { | |
3123 | if (req->file->f_op->read_iter) | |
3124 | return call_read_iter(req->file, &req->rw.kiocb, iter); | |
2dd2111d GH |
3125 | else if (req->file->f_op->read) |
3126 | return loop_rw_iter(READ, req->file, &req->rw.kiocb, iter); | |
3127 | else | |
3128 | return -EINVAL; | |
f67676d1 JA |
3129 | } |
3130 | ||
a1d7c393 JA |
3131 | static int io_read(struct io_kiocb *req, bool force_nonblock, |
3132 | struct io_comp_state *cs) | |
2b188cc1 JA |
3133 | { |
3134 | struct iovec inline_vecs[UIO_FASTIOV], *iovec = inline_vecs; | |
9adbd45d | 3135 | struct kiocb *kiocb = &req->rw.kiocb; |
ff6165b2 | 3136 | struct iov_iter __iter, *iter = &__iter; |
227c0c96 | 3137 | ssize_t io_size, ret, ret2; |
31b51510 | 3138 | size_t iov_count; |
f5cac8b1 | 3139 | bool no_async; |
ff6165b2 JA |
3140 | |
3141 | if (req->io) | |
3142 | iter = &req->io->rw.iter; | |
2b188cc1 | 3143 | |
ff6165b2 | 3144 | ret = io_import_iovec(READ, req, &iovec, iter, !force_nonblock); |
06b76d44 JA |
3145 | if (ret < 0) |
3146 | return ret; | |
eefdf30f | 3147 | iov_count = iov_iter_count(iter); |
fa15bafb PB |
3148 | io_size = ret; |
3149 | req->result = io_size; | |
227c0c96 | 3150 | ret = 0; |
2b188cc1 | 3151 | |
fd6c2e4c JA |
3152 | /* Ensure we clear previously set non-block flag */ |
3153 | if (!force_nonblock) | |
29de5f6a | 3154 | kiocb->ki_flags &= ~IOCB_NOWAIT; |
fd6c2e4c | 3155 | |
24c74678 | 3156 | /* If the file doesn't support async, just async punt */ |
f5cac8b1 JA |
3157 | no_async = force_nonblock && !io_file_supports_async(req->file, READ); |
3158 | if (no_async) | |
f67676d1 | 3159 | goto copy_iov; |
9e645e11 | 3160 | |
0fef9483 | 3161 | ret = rw_verify_area(READ, req->file, io_kiocb_ppos(kiocb), iov_count); |
fa15bafb PB |
3162 | if (unlikely(ret)) |
3163 | goto out_free; | |
2b188cc1 | 3164 | |
227c0c96 | 3165 | ret = io_iter_do_read(req, iter); |
32960613 | 3166 | |
227c0c96 JA |
3167 | if (!ret) { |
3168 | goto done; | |
3169 | } else if (ret == -EIOCBQUEUED) { | |
3170 | ret = 0; | |
3171 | goto out_free; | |
3172 | } else if (ret == -EAGAIN) { | |
eefdf30f JA |
3173 | /* IOPOLL retry should happen for io-wq threads */ |
3174 | if (!force_nonblock && !(req->ctx->flags & IORING_SETUP_IOPOLL)) | |
f91daf56 | 3175 | goto done; |
355afaeb JA |
3176 | /* no retry on NONBLOCK marked file */ |
3177 | if (req->file->f_flags & O_NONBLOCK) | |
3178 | goto done; | |
84216315 JA |
3179 | /* some cases will consume bytes even on error returns */ |
3180 | iov_iter_revert(iter, iov_count - iov_iter_count(iter)); | |
f38c7e3a JA |
3181 | ret = 0; |
3182 | goto copy_iov; | |
227c0c96 | 3183 | } else if (ret < 0) { |
00d23d51 JA |
3184 | /* make sure -ERESTARTSYS -> -EINTR is done */ |
3185 | goto done; | |
227c0c96 JA |
3186 | } |
3187 | ||
3188 | /* read it all, or we did blocking attempt. no retry. */ | |
f91daf56 JA |
3189 | if (!iov_iter_count(iter) || !force_nonblock || |
3190 | (req->file->f_flags & O_NONBLOCK)) | |
227c0c96 JA |
3191 | goto done; |
3192 | ||
3193 | io_size -= ret; | |
3194 | copy_iov: | |
3195 | ret2 = io_setup_async_rw(req, iovec, inline_vecs, iter, true); | |
3196 | if (ret2) { | |
3197 | ret = ret2; | |
3198 | goto out_free; | |
3199 | } | |
f5cac8b1 JA |
3200 | if (no_async) |
3201 | return -EAGAIN; | |
227c0c96 JA |
3202 | /* it's copied and will be cleaned with ->io */ |
3203 | iovec = NULL; | |
3204 | /* now use our persistent iterator, if we aren't already */ | |
3205 | iter = &req->io->rw.iter; | |
3206 | retry: | |
3207 | req->io->rw.bytes_done += ret; | |
3208 | /* if we can retry, do so with the callbacks armed */ | |
3209 | if (!io_rw_should_retry(req)) { | |
fa15bafb PB |
3210 | kiocb->ki_flags &= ~IOCB_WAITQ; |
3211 | return -EAGAIN; | |
2b188cc1 | 3212 | } |
227c0c96 JA |
3213 | |
3214 | /* | |
3215 | * Now retry read with the IOCB_WAITQ parts set in the iocb. If we | |
3216 | * get -EIOCBQUEUED, then we'll get a notification when the desired | |
3217 | * page gets unlocked. We can also get a partial read here, and if we | |
3218 | * do, then just retry at the new offset. | |
3219 | */ | |
3220 | ret = io_iter_do_read(req, iter); | |
3221 | if (ret == -EIOCBQUEUED) { | |
3222 | ret = 0; | |
3223 | goto out_free; | |
3224 | } else if (ret > 0 && ret < io_size) { | |
3225 | /* we got some bytes, but not all. retry. */ | |
3226 | goto retry; | |
3227 | } | |
3228 | done: | |
3229 | kiocb_done(kiocb, ret, cs); | |
3230 | ret = 0; | |
f67676d1 | 3231 | out_free: |
f261c168 | 3232 | /* it's reportedly faster than delegating the null check to kfree() */ |
252917c3 | 3233 | if (iovec) |
6f2cc166 | 3234 | kfree(iovec); |
2b188cc1 JA |
3235 | return ret; |
3236 | } | |
3237 | ||
3529d8c2 JA |
3238 | static int io_write_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe, |
3239 | bool force_nonblock) | |
f67676d1 JA |
3240 | { |
3241 | ssize_t ret; | |
3242 | ||
3529d8c2 JA |
3243 | ret = io_prep_rw(req, sqe, force_nonblock); |
3244 | if (ret) | |
3245 | return ret; | |
f67676d1 | 3246 | |
3529d8c2 JA |
3247 | if (unlikely(!(req->file->f_mode & FMODE_WRITE))) |
3248 | return -EBADF; | |
f67676d1 | 3249 | |
5f798bea PB |
3250 | /* either don't need iovec imported or already have it */ |
3251 | if (!req->io || req->flags & REQ_F_NEED_CLEANUP) | |
3529d8c2 | 3252 | return 0; |
c3e330a4 | 3253 | return io_rw_prep_async(req, WRITE, force_nonblock); |
f67676d1 JA |
3254 | } |
3255 | ||
a1d7c393 JA |
3256 | static int io_write(struct io_kiocb *req, bool force_nonblock, |
3257 | struct io_comp_state *cs) | |
2b188cc1 JA |
3258 | { |
3259 | struct iovec inline_vecs[UIO_FASTIOV], *iovec = inline_vecs; | |
9adbd45d | 3260 | struct kiocb *kiocb = &req->rw.kiocb; |
ff6165b2 | 3261 | struct iov_iter __iter, *iter = &__iter; |
31b51510 | 3262 | size_t iov_count; |
fa15bafb | 3263 | ssize_t ret, ret2, io_size; |
2b188cc1 | 3264 | |
ff6165b2 JA |
3265 | if (req->io) |
3266 | iter = &req->io->rw.iter; | |
3267 | ||
3268 | ret = io_import_iovec(WRITE, req, &iovec, iter, !force_nonblock); | |
06b76d44 JA |
3269 | if (ret < 0) |
3270 | return ret; | |
eefdf30f | 3271 | iov_count = iov_iter_count(iter); |
fa15bafb PB |
3272 | io_size = ret; |
3273 | req->result = io_size; | |
2b188cc1 | 3274 | |
fd6c2e4c JA |
3275 | /* Ensure we clear previously set non-block flag */ |
3276 | if (!force_nonblock) | |
9adbd45d | 3277 | req->rw.kiocb.ki_flags &= ~IOCB_NOWAIT; |
fd6c2e4c | 3278 | |
24c74678 | 3279 | /* If the file doesn't support async, just async punt */ |
af197f50 | 3280 | if (force_nonblock && !io_file_supports_async(req->file, WRITE)) |
f67676d1 | 3281 | goto copy_iov; |
31b51510 | 3282 | |
10d59345 JA |
3283 | /* file path doesn't support NOWAIT for non-direct_IO */ |
3284 | if (force_nonblock && !(kiocb->ki_flags & IOCB_DIRECT) && | |
3285 | (req->flags & REQ_F_ISREG)) | |
f67676d1 | 3286 | goto copy_iov; |
31b51510 | 3287 | |
0fef9483 | 3288 | ret = rw_verify_area(WRITE, req->file, io_kiocb_ppos(kiocb), iov_count); |
fa15bafb PB |
3289 | if (unlikely(ret)) |
3290 | goto out_free; | |
4ed734b0 | 3291 | |
fa15bafb PB |
3292 | /* |
3293 | * Open-code file_start_write here to grab freeze protection, | |
3294 | * which will be released by another thread in | |
3295 | * io_complete_rw(). Fool lockdep by telling it the lock got | |
3296 | * released so that it doesn't complain about the held lock when | |
3297 | * we return to userspace. | |
3298 | */ | |
3299 | if (req->flags & REQ_F_ISREG) { | |
3300 | __sb_start_write(file_inode(req->file)->i_sb, | |
3301 | SB_FREEZE_WRITE, true); | |
3302 | __sb_writers_release(file_inode(req->file)->i_sb, | |
3303 | SB_FREEZE_WRITE); | |
3304 | } | |
3305 | kiocb->ki_flags |= IOCB_WRITE; | |
4ed734b0 | 3306 | |
fa15bafb | 3307 | if (req->file->f_op->write_iter) |
ff6165b2 | 3308 | ret2 = call_write_iter(req->file, kiocb, iter); |
2dd2111d | 3309 | else if (req->file->f_op->write) |
ff6165b2 | 3310 | ret2 = loop_rw_iter(WRITE, req->file, kiocb, iter); |
2dd2111d GH |
3311 | else |
3312 | ret2 = -EINVAL; | |
4ed734b0 | 3313 | |
fa15bafb PB |
3314 | /* |
3315 | * Raw bdev writes will return -EOPNOTSUPP for IOCB_NOWAIT. Just | |
3316 | * retry them without IOCB_NOWAIT. | |
3317 | */ | |
3318 | if (ret2 == -EOPNOTSUPP && (kiocb->ki_flags & IOCB_NOWAIT)) | |
3319 | ret2 = -EAGAIN; | |
355afaeb JA |
3320 | /* no retry on NONBLOCK marked file */ |
3321 | if (ret2 == -EAGAIN && (req->file->f_flags & O_NONBLOCK)) | |
3322 | goto done; | |
fa15bafb | 3323 | if (!force_nonblock || ret2 != -EAGAIN) { |
eefdf30f JA |
3324 | /* IOPOLL retry should happen for io-wq threads */ |
3325 | if ((req->ctx->flags & IORING_SETUP_IOPOLL) && ret2 == -EAGAIN) | |
3326 | goto copy_iov; | |
355afaeb | 3327 | done: |
fa15bafb PB |
3328 | kiocb_done(kiocb, ret2, cs); |
3329 | } else { | |
f67676d1 | 3330 | copy_iov: |
84216315 JA |
3331 | /* some cases will consume bytes even on error returns */ |
3332 | iov_iter_revert(iter, iov_count - iov_iter_count(iter)); | |
227c0c96 | 3333 | ret = io_setup_async_rw(req, iovec, inline_vecs, iter, false); |
ff6165b2 JA |
3334 | if (!ret) |
3335 | return -EAGAIN; | |
2b188cc1 | 3336 | } |
31b51510 | 3337 | out_free: |
f261c168 | 3338 | /* it's reportedly faster than delegating the null check to kfree() */ |
252917c3 | 3339 | if (iovec) |
6f2cc166 | 3340 | kfree(iovec); |
2b188cc1 JA |
3341 | return ret; |
3342 | } | |
3343 | ||
f2a8d5c7 PB |
3344 | static int __io_splice_prep(struct io_kiocb *req, |
3345 | const struct io_uring_sqe *sqe) | |
7d67af2c PB |
3346 | { |
3347 | struct io_splice* sp = &req->splice; | |
3348 | unsigned int valid_flags = SPLICE_F_FD_IN_FIXED | SPLICE_F_ALL; | |
3349 | int ret; | |
3350 | ||
3351 | if (req->flags & REQ_F_NEED_CLEANUP) | |
3352 | return 0; | |
3232dd02 PB |
3353 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
3354 | return -EINVAL; | |
7d67af2c PB |
3355 | |
3356 | sp->file_in = NULL; | |
7d67af2c PB |
3357 | sp->len = READ_ONCE(sqe->len); |
3358 | sp->flags = READ_ONCE(sqe->splice_flags); | |
3359 | ||
3360 | if (unlikely(sp->flags & ~valid_flags)) | |
3361 | return -EINVAL; | |
3362 | ||
3363 | ret = io_file_get(NULL, req, READ_ONCE(sqe->splice_fd_in), &sp->file_in, | |
3364 | (sp->flags & SPLICE_F_FD_IN_FIXED)); | |
3365 | if (ret) | |
3366 | return ret; | |
3367 | req->flags |= REQ_F_NEED_CLEANUP; | |
3368 | ||
7cdaf587 XW |
3369 | if (!S_ISREG(file_inode(sp->file_in)->i_mode)) { |
3370 | /* | |
3371 | * Splice operation will be punted aync, and here need to | |
3372 | * modify io_wq_work.flags, so initialize io_wq_work firstly. | |
3373 | */ | |
3374 | io_req_init_async(req); | |
7d67af2c | 3375 | req->work.flags |= IO_WQ_WORK_UNBOUND; |
7cdaf587 | 3376 | } |
7d67af2c PB |
3377 | |
3378 | return 0; | |
3379 | } | |
3380 | ||
f2a8d5c7 PB |
3381 | static int io_tee_prep(struct io_kiocb *req, |
3382 | const struct io_uring_sqe *sqe) | |
3383 | { | |
3384 | if (READ_ONCE(sqe->splice_off_in) || READ_ONCE(sqe->off)) | |
3385 | return -EINVAL; | |
3386 | return __io_splice_prep(req, sqe); | |
3387 | } | |
3388 | ||
3389 | static int io_tee(struct io_kiocb *req, bool force_nonblock) | |
3390 | { | |
3391 | struct io_splice *sp = &req->splice; | |
3392 | struct file *in = sp->file_in; | |
3393 | struct file *out = sp->file_out; | |
3394 | unsigned int flags = sp->flags & ~SPLICE_F_FD_IN_FIXED; | |
3395 | long ret = 0; | |
3396 | ||
3397 | if (force_nonblock) | |
3398 | return -EAGAIN; | |
3399 | if (sp->len) | |
3400 | ret = do_tee(in, out, sp->len, flags); | |
3401 | ||
3402 | io_put_file(req, in, (sp->flags & SPLICE_F_FD_IN_FIXED)); | |
3403 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
3404 | ||
f2a8d5c7 PB |
3405 | if (ret != sp->len) |
3406 | req_set_fail_links(req); | |
e1e16097 | 3407 | io_req_complete(req, ret); |
f2a8d5c7 PB |
3408 | return 0; |
3409 | } | |
3410 | ||
3411 | static int io_splice_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) | |
3412 | { | |
3413 | struct io_splice* sp = &req->splice; | |
3414 | ||
3415 | sp->off_in = READ_ONCE(sqe->splice_off_in); | |
3416 | sp->off_out = READ_ONCE(sqe->off); | |
3417 | return __io_splice_prep(req, sqe); | |
3418 | } | |
3419 | ||
014db007 | 3420 | static int io_splice(struct io_kiocb *req, bool force_nonblock) |
7d67af2c PB |
3421 | { |
3422 | struct io_splice *sp = &req->splice; | |
3423 | struct file *in = sp->file_in; | |
3424 | struct file *out = sp->file_out; | |
3425 | unsigned int flags = sp->flags & ~SPLICE_F_FD_IN_FIXED; | |
3426 | loff_t *poff_in, *poff_out; | |
c9687426 | 3427 | long ret = 0; |
7d67af2c | 3428 | |
2fb3e822 PB |
3429 | if (force_nonblock) |
3430 | return -EAGAIN; | |
7d67af2c PB |
3431 | |
3432 | poff_in = (sp->off_in == -1) ? NULL : &sp->off_in; | |
3433 | poff_out = (sp->off_out == -1) ? NULL : &sp->off_out; | |
c9687426 | 3434 | |
948a7749 | 3435 | if (sp->len) |
c9687426 | 3436 | ret = do_splice(in, poff_in, out, poff_out, sp->len, flags); |
7d67af2c PB |
3437 | |
3438 | io_put_file(req, in, (sp->flags & SPLICE_F_FD_IN_FIXED)); | |
3439 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
3440 | ||
7d67af2c PB |
3441 | if (ret != sp->len) |
3442 | req_set_fail_links(req); | |
e1e16097 | 3443 | io_req_complete(req, ret); |
7d67af2c PB |
3444 | return 0; |
3445 | } | |
3446 | ||
2b188cc1 JA |
3447 | /* |
3448 | * IORING_OP_NOP just posts a completion event, nothing else. | |
3449 | */ | |
229a7b63 | 3450 | static int io_nop(struct io_kiocb *req, struct io_comp_state *cs) |
2b188cc1 JA |
3451 | { |
3452 | struct io_ring_ctx *ctx = req->ctx; | |
2b188cc1 | 3453 | |
def596e9 JA |
3454 | if (unlikely(ctx->flags & IORING_SETUP_IOPOLL)) |
3455 | return -EINVAL; | |
3456 | ||
229a7b63 | 3457 | __io_req_complete(req, 0, 0, cs); |
2b188cc1 JA |
3458 | return 0; |
3459 | } | |
3460 | ||
3529d8c2 | 3461 | static int io_prep_fsync(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
c992fe29 | 3462 | { |
6b06314c | 3463 | struct io_ring_ctx *ctx = req->ctx; |
c992fe29 | 3464 | |
09bb8394 JA |
3465 | if (!req->file) |
3466 | return -EBADF; | |
c992fe29 | 3467 | |
6b06314c | 3468 | if (unlikely(ctx->flags & IORING_SETUP_IOPOLL)) |
def596e9 | 3469 | return -EINVAL; |
edafccee | 3470 | if (unlikely(sqe->addr || sqe->ioprio || sqe->buf_index)) |
c992fe29 CH |
3471 | return -EINVAL; |
3472 | ||
8ed8d3c3 JA |
3473 | req->sync.flags = READ_ONCE(sqe->fsync_flags); |
3474 | if (unlikely(req->sync.flags & ~IORING_FSYNC_DATASYNC)) | |
3475 | return -EINVAL; | |
3476 | ||
3477 | req->sync.off = READ_ONCE(sqe->off); | |
3478 | req->sync.len = READ_ONCE(sqe->len); | |
c992fe29 CH |
3479 | return 0; |
3480 | } | |
3481 | ||
ac45abc0 | 3482 | static int io_fsync(struct io_kiocb *req, bool force_nonblock) |
8ed8d3c3 | 3483 | { |
8ed8d3c3 | 3484 | loff_t end = req->sync.off + req->sync.len; |
8ed8d3c3 JA |
3485 | int ret; |
3486 | ||
ac45abc0 PB |
3487 | /* fsync always requires a blocking context */ |
3488 | if (force_nonblock) | |
3489 | return -EAGAIN; | |
3490 | ||
9adbd45d | 3491 | ret = vfs_fsync_range(req->file, req->sync.off, |
8ed8d3c3 JA |
3492 | end > 0 ? end : LLONG_MAX, |
3493 | req->sync.flags & IORING_FSYNC_DATASYNC); | |
3494 | if (ret < 0) | |
3495 | req_set_fail_links(req); | |
e1e16097 | 3496 | io_req_complete(req, ret); |
c992fe29 CH |
3497 | return 0; |
3498 | } | |
3499 | ||
d63d1b5e JA |
3500 | static int io_fallocate_prep(struct io_kiocb *req, |
3501 | const struct io_uring_sqe *sqe) | |
3502 | { | |
3503 | if (sqe->ioprio || sqe->buf_index || sqe->rw_flags) | |
3504 | return -EINVAL; | |
3232dd02 PB |
3505 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
3506 | return -EINVAL; | |
d63d1b5e JA |
3507 | |
3508 | req->sync.off = READ_ONCE(sqe->off); | |
3509 | req->sync.len = READ_ONCE(sqe->addr); | |
3510 | req->sync.mode = READ_ONCE(sqe->len); | |
3511 | return 0; | |
3512 | } | |
3513 | ||
014db007 | 3514 | static int io_fallocate(struct io_kiocb *req, bool force_nonblock) |
5d17b4a4 | 3515 | { |
ac45abc0 PB |
3516 | int ret; |
3517 | ||
d63d1b5e | 3518 | /* fallocate always requiring blocking context */ |
ac45abc0 | 3519 | if (force_nonblock) |
5d17b4a4 | 3520 | return -EAGAIN; |
ac45abc0 PB |
3521 | ret = vfs_fallocate(req->file, req->sync.mode, req->sync.off, |
3522 | req->sync.len); | |
ac45abc0 PB |
3523 | if (ret < 0) |
3524 | req_set_fail_links(req); | |
e1e16097 | 3525 | io_req_complete(req, ret); |
5d17b4a4 JA |
3526 | return 0; |
3527 | } | |
3528 | ||
ec65fea5 | 3529 | static int __io_openat_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
b7bb4f7d | 3530 | { |
f8748881 | 3531 | const char __user *fname; |
15b71abe | 3532 | int ret; |
b7bb4f7d | 3533 | |
ec65fea5 | 3534 | if (unlikely(sqe->ioprio || sqe->buf_index)) |
15b71abe | 3535 | return -EINVAL; |
ec65fea5 | 3536 | if (unlikely(req->flags & REQ_F_FIXED_FILE)) |
cf3040ca | 3537 | return -EBADF; |
03b1230c | 3538 | |
ec65fea5 PB |
3539 | /* open.how should be already initialised */ |
3540 | if (!(req->open.how.flags & O_PATH) && force_o_largefile()) | |
08a1d26e | 3541 | req->open.how.flags |= O_LARGEFILE; |
3529d8c2 | 3542 | |
25e72d10 PB |
3543 | req->open.dfd = READ_ONCE(sqe->fd); |
3544 | fname = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
f8748881 | 3545 | req->open.filename = getname(fname); |
15b71abe JA |
3546 | if (IS_ERR(req->open.filename)) { |
3547 | ret = PTR_ERR(req->open.filename); | |
3548 | req->open.filename = NULL; | |
3549 | return ret; | |
3550 | } | |
4022e7af | 3551 | req->open.nofile = rlimit(RLIMIT_NOFILE); |
8fef80bf | 3552 | req->flags |= REQ_F_NEED_CLEANUP; |
15b71abe | 3553 | return 0; |
03b1230c JA |
3554 | } |
3555 | ||
ec65fea5 PB |
3556 | static int io_openat_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
3557 | { | |
3558 | u64 flags, mode; | |
3559 | ||
4eb8dded JA |
3560 | if (unlikely(req->ctx->flags & (IORING_SETUP_IOPOLL|IORING_SETUP_SQPOLL))) |
3561 | return -EINVAL; | |
ec65fea5 PB |
3562 | if (req->flags & REQ_F_NEED_CLEANUP) |
3563 | return 0; | |
3564 | mode = READ_ONCE(sqe->len); | |
3565 | flags = READ_ONCE(sqe->open_flags); | |
3566 | req->open.how = build_open_how(flags, mode); | |
3567 | return __io_openat_prep(req, sqe); | |
3568 | } | |
3569 | ||
cebdb986 | 3570 | static int io_openat2_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
aa1fa28f | 3571 | { |
cebdb986 | 3572 | struct open_how __user *how; |
cebdb986 | 3573 | size_t len; |
0fa03c62 JA |
3574 | int ret; |
3575 | ||
4eb8dded JA |
3576 | if (unlikely(req->ctx->flags & (IORING_SETUP_IOPOLL|IORING_SETUP_SQPOLL))) |
3577 | return -EINVAL; | |
0bdbdd08 PB |
3578 | if (req->flags & REQ_F_NEED_CLEANUP) |
3579 | return 0; | |
cebdb986 JA |
3580 | how = u64_to_user_ptr(READ_ONCE(sqe->addr2)); |
3581 | len = READ_ONCE(sqe->len); | |
cebdb986 JA |
3582 | if (len < OPEN_HOW_SIZE_VER0) |
3583 | return -EINVAL; | |
3529d8c2 | 3584 | |
cebdb986 JA |
3585 | ret = copy_struct_from_user(&req->open.how, sizeof(req->open.how), how, |
3586 | len); | |
3587 | if (ret) | |
3588 | return ret; | |
3529d8c2 | 3589 | |
ec65fea5 | 3590 | return __io_openat_prep(req, sqe); |
cebdb986 JA |
3591 | } |
3592 | ||
014db007 | 3593 | static int io_openat2(struct io_kiocb *req, bool force_nonblock) |
15b71abe JA |
3594 | { |
3595 | struct open_flags op; | |
15b71abe JA |
3596 | struct file *file; |
3597 | int ret; | |
3598 | ||
f86cd20c | 3599 | if (force_nonblock) |
15b71abe | 3600 | return -EAGAIN; |
15b71abe | 3601 | |
cebdb986 | 3602 | ret = build_open_flags(&req->open.how, &op); |
15b71abe JA |
3603 | if (ret) |
3604 | goto err; | |
3605 | ||
4022e7af | 3606 | ret = __get_unused_fd_flags(req->open.how.flags, req->open.nofile); |
15b71abe JA |
3607 | if (ret < 0) |
3608 | goto err; | |
3609 | ||
3610 | file = do_filp_open(req->open.dfd, req->open.filename, &op); | |
3611 | if (IS_ERR(file)) { | |
3612 | put_unused_fd(ret); | |
3613 | ret = PTR_ERR(file); | |
3614 | } else { | |
3615 | fsnotify_open(file); | |
3616 | fd_install(ret, file); | |
3617 | } | |
3618 | err: | |
3619 | putname(req->open.filename); | |
8fef80bf | 3620 | req->flags &= ~REQ_F_NEED_CLEANUP; |
15b71abe JA |
3621 | if (ret < 0) |
3622 | req_set_fail_links(req); | |
e1e16097 | 3623 | io_req_complete(req, ret); |
15b71abe JA |
3624 | return 0; |
3625 | } | |
3626 | ||
014db007 | 3627 | static int io_openat(struct io_kiocb *req, bool force_nonblock) |
cebdb986 | 3628 | { |
014db007 | 3629 | return io_openat2(req, force_nonblock); |
cebdb986 JA |
3630 | } |
3631 | ||
067524e9 JA |
3632 | static int io_remove_buffers_prep(struct io_kiocb *req, |
3633 | const struct io_uring_sqe *sqe) | |
3634 | { | |
3635 | struct io_provide_buf *p = &req->pbuf; | |
3636 | u64 tmp; | |
3637 | ||
3638 | if (sqe->ioprio || sqe->rw_flags || sqe->addr || sqe->len || sqe->off) | |
3639 | return -EINVAL; | |
3640 | ||
3641 | tmp = READ_ONCE(sqe->fd); | |
3642 | if (!tmp || tmp > USHRT_MAX) | |
3643 | return -EINVAL; | |
3644 | ||
3645 | memset(p, 0, sizeof(*p)); | |
3646 | p->nbufs = tmp; | |
3647 | p->bgid = READ_ONCE(sqe->buf_group); | |
3648 | return 0; | |
3649 | } | |
3650 | ||
3651 | static int __io_remove_buffers(struct io_ring_ctx *ctx, struct io_buffer *buf, | |
3652 | int bgid, unsigned nbufs) | |
3653 | { | |
3654 | unsigned i = 0; | |
3655 | ||
3656 | /* shouldn't happen */ | |
3657 | if (!nbufs) | |
3658 | return 0; | |
3659 | ||
3660 | /* the head kbuf is the list itself */ | |
3661 | while (!list_empty(&buf->list)) { | |
3662 | struct io_buffer *nxt; | |
3663 | ||
3664 | nxt = list_first_entry(&buf->list, struct io_buffer, list); | |
3665 | list_del(&nxt->list); | |
3666 | kfree(nxt); | |
3667 | if (++i == nbufs) | |
3668 | return i; | |
3669 | } | |
3670 | i++; | |
3671 | kfree(buf); | |
3672 | idr_remove(&ctx->io_buffer_idr, bgid); | |
3673 | ||
3674 | return i; | |
3675 | } | |
3676 | ||
229a7b63 JA |
3677 | static int io_remove_buffers(struct io_kiocb *req, bool force_nonblock, |
3678 | struct io_comp_state *cs) | |
067524e9 JA |
3679 | { |
3680 | struct io_provide_buf *p = &req->pbuf; | |
3681 | struct io_ring_ctx *ctx = req->ctx; | |
3682 | struct io_buffer *head; | |
3683 | int ret = 0; | |
3684 | ||
3685 | io_ring_submit_lock(ctx, !force_nonblock); | |
3686 | ||
3687 | lockdep_assert_held(&ctx->uring_lock); | |
3688 | ||
3689 | ret = -ENOENT; | |
3690 | head = idr_find(&ctx->io_buffer_idr, p->bgid); | |
3691 | if (head) | |
3692 | ret = __io_remove_buffers(ctx, head, p->bgid, p->nbufs); | |
3693 | ||
3694 | io_ring_submit_lock(ctx, !force_nonblock); | |
3695 | if (ret < 0) | |
3696 | req_set_fail_links(req); | |
229a7b63 | 3697 | __io_req_complete(req, ret, 0, cs); |
067524e9 JA |
3698 | return 0; |
3699 | } | |
3700 | ||
ddf0322d JA |
3701 | static int io_provide_buffers_prep(struct io_kiocb *req, |
3702 | const struct io_uring_sqe *sqe) | |
3703 | { | |
3704 | struct io_provide_buf *p = &req->pbuf; | |
3705 | u64 tmp; | |
3706 | ||
3707 | if (sqe->ioprio || sqe->rw_flags) | |
3708 | return -EINVAL; | |
3709 | ||
3710 | tmp = READ_ONCE(sqe->fd); | |
3711 | if (!tmp || tmp > USHRT_MAX) | |
3712 | return -E2BIG; | |
3713 | p->nbufs = tmp; | |
3714 | p->addr = READ_ONCE(sqe->addr); | |
3715 | p->len = READ_ONCE(sqe->len); | |
3716 | ||
efe68c1c | 3717 | if (!access_ok(u64_to_user_ptr(p->addr), (p->len * p->nbufs))) |
ddf0322d JA |
3718 | return -EFAULT; |
3719 | ||
3720 | p->bgid = READ_ONCE(sqe->buf_group); | |
3721 | tmp = READ_ONCE(sqe->off); | |
3722 | if (tmp > USHRT_MAX) | |
3723 | return -E2BIG; | |
3724 | p->bid = tmp; | |
3725 | return 0; | |
3726 | } | |
3727 | ||
3728 | static int io_add_buffers(struct io_provide_buf *pbuf, struct io_buffer **head) | |
3729 | { | |
3730 | struct io_buffer *buf; | |
3731 | u64 addr = pbuf->addr; | |
3732 | int i, bid = pbuf->bid; | |
3733 | ||
3734 | for (i = 0; i < pbuf->nbufs; i++) { | |
3735 | buf = kmalloc(sizeof(*buf), GFP_KERNEL); | |
3736 | if (!buf) | |
3737 | break; | |
3738 | ||
3739 | buf->addr = addr; | |
3740 | buf->len = pbuf->len; | |
3741 | buf->bid = bid; | |
3742 | addr += pbuf->len; | |
3743 | bid++; | |
3744 | if (!*head) { | |
3745 | INIT_LIST_HEAD(&buf->list); | |
3746 | *head = buf; | |
3747 | } else { | |
3748 | list_add_tail(&buf->list, &(*head)->list); | |
3749 | } | |
3750 | } | |
3751 | ||
3752 | return i ? i : -ENOMEM; | |
3753 | } | |
3754 | ||
229a7b63 JA |
3755 | static int io_provide_buffers(struct io_kiocb *req, bool force_nonblock, |
3756 | struct io_comp_state *cs) | |
ddf0322d JA |
3757 | { |
3758 | struct io_provide_buf *p = &req->pbuf; | |
3759 | struct io_ring_ctx *ctx = req->ctx; | |
3760 | struct io_buffer *head, *list; | |
3761 | int ret = 0; | |
3762 | ||
3763 | io_ring_submit_lock(ctx, !force_nonblock); | |
3764 | ||
3765 | lockdep_assert_held(&ctx->uring_lock); | |
3766 | ||
3767 | list = head = idr_find(&ctx->io_buffer_idr, p->bgid); | |
3768 | ||
3769 | ret = io_add_buffers(p, &head); | |
3770 | if (ret < 0) | |
3771 | goto out; | |
3772 | ||
3773 | if (!list) { | |
3774 | ret = idr_alloc(&ctx->io_buffer_idr, head, p->bgid, p->bgid + 1, | |
3775 | GFP_KERNEL); | |
3776 | if (ret < 0) { | |
067524e9 | 3777 | __io_remove_buffers(ctx, head, p->bgid, -1U); |
ddf0322d JA |
3778 | goto out; |
3779 | } | |
3780 | } | |
3781 | out: | |
3782 | io_ring_submit_unlock(ctx, !force_nonblock); | |
3783 | if (ret < 0) | |
3784 | req_set_fail_links(req); | |
229a7b63 | 3785 | __io_req_complete(req, ret, 0, cs); |
ddf0322d | 3786 | return 0; |
cebdb986 JA |
3787 | } |
3788 | ||
3e4827b0 JA |
3789 | static int io_epoll_ctl_prep(struct io_kiocb *req, |
3790 | const struct io_uring_sqe *sqe) | |
3791 | { | |
3792 | #if defined(CONFIG_EPOLL) | |
3793 | if (sqe->ioprio || sqe->buf_index) | |
3794 | return -EINVAL; | |
6ca56f84 | 3795 | if (unlikely(req->ctx->flags & (IORING_SETUP_IOPOLL | IORING_SETUP_SQPOLL))) |
3232dd02 | 3796 | return -EINVAL; |
3e4827b0 JA |
3797 | |
3798 | req->epoll.epfd = READ_ONCE(sqe->fd); | |
3799 | req->epoll.op = READ_ONCE(sqe->len); | |
3800 | req->epoll.fd = READ_ONCE(sqe->off); | |
3801 | ||
3802 | if (ep_op_has_event(req->epoll.op)) { | |
3803 | struct epoll_event __user *ev; | |
3804 | ||
3805 | ev = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
3806 | if (copy_from_user(&req->epoll.event, ev, sizeof(*ev))) | |
3807 | return -EFAULT; | |
3808 | } | |
3809 | ||
3810 | return 0; | |
3811 | #else | |
3812 | return -EOPNOTSUPP; | |
3813 | #endif | |
3814 | } | |
3815 | ||
229a7b63 JA |
3816 | static int io_epoll_ctl(struct io_kiocb *req, bool force_nonblock, |
3817 | struct io_comp_state *cs) | |
3e4827b0 JA |
3818 | { |
3819 | #if defined(CONFIG_EPOLL) | |
3820 | struct io_epoll *ie = &req->epoll; | |
3821 | int ret; | |
3822 | ||
3823 | ret = do_epoll_ctl(ie->epfd, ie->op, ie->fd, &ie->event, force_nonblock); | |
3824 | if (force_nonblock && ret == -EAGAIN) | |
3825 | return -EAGAIN; | |
3826 | ||
3827 | if (ret < 0) | |
3828 | req_set_fail_links(req); | |
229a7b63 | 3829 | __io_req_complete(req, ret, 0, cs); |
3e4827b0 JA |
3830 | return 0; |
3831 | #else | |
3832 | return -EOPNOTSUPP; | |
3833 | #endif | |
3834 | } | |
3835 | ||
c1ca757b JA |
3836 | static int io_madvise_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
3837 | { | |
3838 | #if defined(CONFIG_ADVISE_SYSCALLS) && defined(CONFIG_MMU) | |
3839 | if (sqe->ioprio || sqe->buf_index || sqe->off) | |
3840 | return -EINVAL; | |
3232dd02 PB |
3841 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
3842 | return -EINVAL; | |
c1ca757b JA |
3843 | |
3844 | req->madvise.addr = READ_ONCE(sqe->addr); | |
3845 | req->madvise.len = READ_ONCE(sqe->len); | |
3846 | req->madvise.advice = READ_ONCE(sqe->fadvise_advice); | |
3847 | return 0; | |
3848 | #else | |
3849 | return -EOPNOTSUPP; | |
3850 | #endif | |
3851 | } | |
3852 | ||
014db007 | 3853 | static int io_madvise(struct io_kiocb *req, bool force_nonblock) |
c1ca757b JA |
3854 | { |
3855 | #if defined(CONFIG_ADVISE_SYSCALLS) && defined(CONFIG_MMU) | |
3856 | struct io_madvise *ma = &req->madvise; | |
3857 | int ret; | |
3858 | ||
3859 | if (force_nonblock) | |
3860 | return -EAGAIN; | |
3861 | ||
3862 | ret = do_madvise(ma->addr, ma->len, ma->advice); | |
3863 | if (ret < 0) | |
3864 | req_set_fail_links(req); | |
e1e16097 | 3865 | io_req_complete(req, ret); |
c1ca757b JA |
3866 | return 0; |
3867 | #else | |
3868 | return -EOPNOTSUPP; | |
3869 | #endif | |
3870 | } | |
3871 | ||
4840e418 JA |
3872 | static int io_fadvise_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
3873 | { | |
3874 | if (sqe->ioprio || sqe->buf_index || sqe->addr) | |
3875 | return -EINVAL; | |
3232dd02 PB |
3876 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
3877 | return -EINVAL; | |
4840e418 JA |
3878 | |
3879 | req->fadvise.offset = READ_ONCE(sqe->off); | |
3880 | req->fadvise.len = READ_ONCE(sqe->len); | |
3881 | req->fadvise.advice = READ_ONCE(sqe->fadvise_advice); | |
3882 | return 0; | |
3883 | } | |
3884 | ||
014db007 | 3885 | static int io_fadvise(struct io_kiocb *req, bool force_nonblock) |
4840e418 JA |
3886 | { |
3887 | struct io_fadvise *fa = &req->fadvise; | |
3888 | int ret; | |
3889 | ||
3e69426d JA |
3890 | if (force_nonblock) { |
3891 | switch (fa->advice) { | |
3892 | case POSIX_FADV_NORMAL: | |
3893 | case POSIX_FADV_RANDOM: | |
3894 | case POSIX_FADV_SEQUENTIAL: | |
3895 | break; | |
3896 | default: | |
3897 | return -EAGAIN; | |
3898 | } | |
3899 | } | |
4840e418 JA |
3900 | |
3901 | ret = vfs_fadvise(req->file, fa->offset, fa->len, fa->advice); | |
3902 | if (ret < 0) | |
3903 | req_set_fail_links(req); | |
e1e16097 | 3904 | io_req_complete(req, ret); |
4840e418 JA |
3905 | return 0; |
3906 | } | |
3907 | ||
eddc7ef5 JA |
3908 | static int io_statx_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
3909 | { | |
6ca56f84 | 3910 | if (unlikely(req->ctx->flags & (IORING_SETUP_IOPOLL | IORING_SETUP_SQPOLL))) |
3232dd02 | 3911 | return -EINVAL; |
eddc7ef5 JA |
3912 | if (sqe->ioprio || sqe->buf_index) |
3913 | return -EINVAL; | |
9c280f90 | 3914 | if (req->flags & REQ_F_FIXED_FILE) |
cf3040ca | 3915 | return -EBADF; |
eddc7ef5 | 3916 | |
1d9e1288 BM |
3917 | req->statx.dfd = READ_ONCE(sqe->fd); |
3918 | req->statx.mask = READ_ONCE(sqe->len); | |
e62753e4 | 3919 | req->statx.filename = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
1d9e1288 BM |
3920 | req->statx.buffer = u64_to_user_ptr(READ_ONCE(sqe->addr2)); |
3921 | req->statx.flags = READ_ONCE(sqe->statx_flags); | |
eddc7ef5 JA |
3922 | |
3923 | return 0; | |
3924 | } | |
3925 | ||
014db007 | 3926 | static int io_statx(struct io_kiocb *req, bool force_nonblock) |
eddc7ef5 | 3927 | { |
1d9e1288 | 3928 | struct io_statx *ctx = &req->statx; |
eddc7ef5 JA |
3929 | int ret; |
3930 | ||
5b0bbee4 JA |
3931 | if (force_nonblock) { |
3932 | /* only need file table for an actual valid fd */ | |
3933 | if (ctx->dfd == -1 || ctx->dfd == AT_FDCWD) | |
3934 | req->flags |= REQ_F_NO_FILE_TABLE; | |
eddc7ef5 | 3935 | return -EAGAIN; |
5b0bbee4 | 3936 | } |
eddc7ef5 | 3937 | |
e62753e4 BM |
3938 | ret = do_statx(ctx->dfd, ctx->filename, ctx->flags, ctx->mask, |
3939 | ctx->buffer); | |
eddc7ef5 | 3940 | |
eddc7ef5 JA |
3941 | if (ret < 0) |
3942 | req_set_fail_links(req); | |
e1e16097 | 3943 | io_req_complete(req, ret); |
eddc7ef5 JA |
3944 | return 0; |
3945 | } | |
3946 | ||
b5dba59e JA |
3947 | static int io_close_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
3948 | { | |
3949 | /* | |
3950 | * If we queue this for async, it must not be cancellable. That would | |
7cdaf587 XW |
3951 | * leave the 'file' in an undeterminate state, and here need to modify |
3952 | * io_wq_work.flags, so initialize io_wq_work firstly. | |
b5dba59e | 3953 | */ |
7cdaf587 | 3954 | io_req_init_async(req); |
b5dba59e JA |
3955 | req->work.flags |= IO_WQ_WORK_NO_CANCEL; |
3956 | ||
3232dd02 PB |
3957 | if (unlikely(req->ctx->flags & (IORING_SETUP_IOPOLL|IORING_SETUP_SQPOLL))) |
3958 | return -EINVAL; | |
b5dba59e JA |
3959 | if (sqe->ioprio || sqe->off || sqe->addr || sqe->len || |
3960 | sqe->rw_flags || sqe->buf_index) | |
3961 | return -EINVAL; | |
9c280f90 | 3962 | if (req->flags & REQ_F_FIXED_FILE) |
cf3040ca | 3963 | return -EBADF; |
b5dba59e JA |
3964 | |
3965 | req->close.fd = READ_ONCE(sqe->fd); | |
fd2206e4 JA |
3966 | if ((req->file && req->file->f_op == &io_uring_fops) || |
3967 | req->close.fd == req->ctx->ring_fd) | |
3968 | return -EBADF; | |
b5dba59e | 3969 | |
3af73b28 | 3970 | req->close.put_file = NULL; |
b5dba59e | 3971 | return 0; |
b5dba59e JA |
3972 | } |
3973 | ||
229a7b63 JA |
3974 | static int io_close(struct io_kiocb *req, bool force_nonblock, |
3975 | struct io_comp_state *cs) | |
b5dba59e | 3976 | { |
3af73b28 | 3977 | struct io_close *close = &req->close; |
b5dba59e JA |
3978 | int ret; |
3979 | ||
3af73b28 PB |
3980 | /* might be already done during nonblock submission */ |
3981 | if (!close->put_file) { | |
3982 | ret = __close_fd_get_file(close->fd, &close->put_file); | |
3983 | if (ret < 0) | |
3984 | return (ret == -ENOENT) ? -EBADF : ret; | |
3985 | } | |
b5dba59e JA |
3986 | |
3987 | /* if the file has a flush method, be safe and punt to async */ | |
3af73b28 | 3988 | if (close->put_file->f_op->flush && force_nonblock) { |
24c74678 PB |
3989 | /* was never set, but play safe */ |
3990 | req->flags &= ~REQ_F_NOWAIT; | |
0bf0eefd | 3991 | /* avoid grabbing files - we don't need the files */ |
24c74678 | 3992 | req->flags |= REQ_F_NO_FILE_TABLE; |
0bf0eefd | 3993 | return -EAGAIN; |
a2100672 | 3994 | } |
b5dba59e | 3995 | |
3af73b28 PB |
3996 | /* No ->flush() or already async, safely close from here */ |
3997 | ret = filp_close(close->put_file, req->work.files); | |
3998 | if (ret < 0) | |
3999 | req_set_fail_links(req); | |
3af73b28 PB |
4000 | fput(close->put_file); |
4001 | close->put_file = NULL; | |
229a7b63 | 4002 | __io_req_complete(req, ret, 0, cs); |
1a417f4e | 4003 | return 0; |
b5dba59e JA |
4004 | } |
4005 | ||
3529d8c2 | 4006 | static int io_prep_sfr(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
5d17b4a4 JA |
4007 | { |
4008 | struct io_ring_ctx *ctx = req->ctx; | |
5d17b4a4 JA |
4009 | |
4010 | if (!req->file) | |
4011 | return -EBADF; | |
5d17b4a4 JA |
4012 | |
4013 | if (unlikely(ctx->flags & IORING_SETUP_IOPOLL)) | |
4014 | return -EINVAL; | |
4015 | if (unlikely(sqe->addr || sqe->ioprio || sqe->buf_index)) | |
4016 | return -EINVAL; | |
4017 | ||
8ed8d3c3 JA |
4018 | req->sync.off = READ_ONCE(sqe->off); |
4019 | req->sync.len = READ_ONCE(sqe->len); | |
4020 | req->sync.flags = READ_ONCE(sqe->sync_range_flags); | |
8ed8d3c3 JA |
4021 | return 0; |
4022 | } | |
4023 | ||
ac45abc0 | 4024 | static int io_sync_file_range(struct io_kiocb *req, bool force_nonblock) |
8ed8d3c3 | 4025 | { |
8ed8d3c3 JA |
4026 | int ret; |
4027 | ||
ac45abc0 PB |
4028 | /* sync_file_range always requires a blocking context */ |
4029 | if (force_nonblock) | |
4030 | return -EAGAIN; | |
4031 | ||
9adbd45d | 4032 | ret = sync_file_range(req->file, req->sync.off, req->sync.len, |
8ed8d3c3 JA |
4033 | req->sync.flags); |
4034 | if (ret < 0) | |
4035 | req_set_fail_links(req); | |
e1e16097 | 4036 | io_req_complete(req, ret); |
5d17b4a4 JA |
4037 | return 0; |
4038 | } | |
4039 | ||
469956e8 | 4040 | #if defined(CONFIG_NET) |
02d27d89 PB |
4041 | static int io_setup_async_msg(struct io_kiocb *req, |
4042 | struct io_async_msghdr *kmsg) | |
4043 | { | |
4044 | if (req->io) | |
4045 | return -EAGAIN; | |
4046 | if (io_alloc_async_ctx(req)) { | |
4047 | if (kmsg->iov != kmsg->fast_iov) | |
4048 | kfree(kmsg->iov); | |
4049 | return -ENOMEM; | |
4050 | } | |
4051 | req->flags |= REQ_F_NEED_CLEANUP; | |
4052 | memcpy(&req->io->msg, kmsg, sizeof(*kmsg)); | |
4053 | return -EAGAIN; | |
4054 | } | |
4055 | ||
2ae523ed PB |
4056 | static int io_sendmsg_copy_hdr(struct io_kiocb *req, |
4057 | struct io_async_msghdr *iomsg) | |
4058 | { | |
4059 | iomsg->iov = iomsg->fast_iov; | |
4060 | iomsg->msg.msg_name = &iomsg->addr; | |
4061 | return sendmsg_copy_msghdr(&iomsg->msg, req->sr_msg.umsg, | |
4062 | req->sr_msg.msg_flags, &iomsg->iov); | |
4063 | } | |
4064 | ||
3529d8c2 | 4065 | static int io_sendmsg_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
03b1230c | 4066 | { |
e47293fd | 4067 | struct io_sr_msg *sr = &req->sr_msg; |
3529d8c2 | 4068 | struct io_async_ctx *io = req->io; |
99bc4c38 | 4069 | int ret; |
03b1230c | 4070 | |
d2b6f48b PB |
4071 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4072 | return -EINVAL; | |
4073 | ||
e47293fd | 4074 | sr->msg_flags = READ_ONCE(sqe->msg_flags); |
270a5940 | 4075 | sr->umsg = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
fddaface | 4076 | sr->len = READ_ONCE(sqe->len); |
3529d8c2 | 4077 | |
d8768362 JA |
4078 | #ifdef CONFIG_COMPAT |
4079 | if (req->ctx->compat) | |
4080 | sr->msg_flags |= MSG_CMSG_COMPAT; | |
4081 | #endif | |
4082 | ||
fddaface | 4083 | if (!io || req->opcode == IORING_OP_SEND) |
3529d8c2 | 4084 | return 0; |
5f798bea PB |
4085 | /* iovec is already imported */ |
4086 | if (req->flags & REQ_F_NEED_CLEANUP) | |
4087 | return 0; | |
3529d8c2 | 4088 | |
2ae523ed | 4089 | ret = io_sendmsg_copy_hdr(req, &io->msg); |
99bc4c38 PB |
4090 | if (!ret) |
4091 | req->flags |= REQ_F_NEED_CLEANUP; | |
4092 | return ret; | |
03b1230c JA |
4093 | } |
4094 | ||
229a7b63 JA |
4095 | static int io_sendmsg(struct io_kiocb *req, bool force_nonblock, |
4096 | struct io_comp_state *cs) | |
aa1fa28f | 4097 | { |
6b754c8b | 4098 | struct io_async_msghdr iomsg, *kmsg; |
0fa03c62 | 4099 | struct socket *sock; |
7a7cacba | 4100 | unsigned flags; |
0fa03c62 JA |
4101 | int ret; |
4102 | ||
0fa03c62 | 4103 | sock = sock_from_file(req->file, &ret); |
7a7cacba PB |
4104 | if (unlikely(!sock)) |
4105 | return ret; | |
3529d8c2 | 4106 | |
7a7cacba PB |
4107 | if (req->io) { |
4108 | kmsg = &req->io->msg; | |
4109 | kmsg->msg.msg_name = &req->io->msg.addr; | |
4110 | /* if iov is set, it's allocated already */ | |
4111 | if (!kmsg->iov) | |
4112 | kmsg->iov = kmsg->fast_iov; | |
4113 | kmsg->msg.msg_iter.iov = kmsg->iov; | |
4114 | } else { | |
4115 | ret = io_sendmsg_copy_hdr(req, &iomsg); | |
4116 | if (ret) | |
4117 | return ret; | |
4118 | kmsg = &iomsg; | |
0fa03c62 | 4119 | } |
0fa03c62 | 4120 | |
7a7cacba PB |
4121 | flags = req->sr_msg.msg_flags; |
4122 | if (flags & MSG_DONTWAIT) | |
4123 | req->flags |= REQ_F_NOWAIT; | |
4124 | else if (force_nonblock) | |
4125 | flags |= MSG_DONTWAIT; | |
e47293fd | 4126 | |
7a7cacba PB |
4127 | ret = __sys_sendmsg_sock(sock, &kmsg->msg, flags); |
4128 | if (force_nonblock && ret == -EAGAIN) | |
4129 | return io_setup_async_msg(req, kmsg); | |
4130 | if (ret == -ERESTARTSYS) | |
4131 | ret = -EINTR; | |
0fa03c62 | 4132 | |
6b754c8b | 4133 | if (kmsg->iov != kmsg->fast_iov) |
0b416c3e | 4134 | kfree(kmsg->iov); |
99bc4c38 | 4135 | req->flags &= ~REQ_F_NEED_CLEANUP; |
4e88d6e7 JA |
4136 | if (ret < 0) |
4137 | req_set_fail_links(req); | |
229a7b63 | 4138 | __io_req_complete(req, ret, 0, cs); |
5d17b4a4 | 4139 | return 0; |
03b1230c | 4140 | } |
aa1fa28f | 4141 | |
229a7b63 JA |
4142 | static int io_send(struct io_kiocb *req, bool force_nonblock, |
4143 | struct io_comp_state *cs) | |
fddaface | 4144 | { |
7a7cacba PB |
4145 | struct io_sr_msg *sr = &req->sr_msg; |
4146 | struct msghdr msg; | |
4147 | struct iovec iov; | |
fddaface | 4148 | struct socket *sock; |
7a7cacba | 4149 | unsigned flags; |
fddaface JA |
4150 | int ret; |
4151 | ||
fddaface | 4152 | sock = sock_from_file(req->file, &ret); |
7a7cacba PB |
4153 | if (unlikely(!sock)) |
4154 | return ret; | |
fddaface | 4155 | |
7a7cacba PB |
4156 | ret = import_single_range(WRITE, sr->buf, sr->len, &iov, &msg.msg_iter); |
4157 | if (unlikely(ret)) | |
14c32eee | 4158 | return ret;; |
fddaface | 4159 | |
7a7cacba PB |
4160 | msg.msg_name = NULL; |
4161 | msg.msg_control = NULL; | |
4162 | msg.msg_controllen = 0; | |
4163 | msg.msg_namelen = 0; | |
fddaface | 4164 | |
7a7cacba PB |
4165 | flags = req->sr_msg.msg_flags; |
4166 | if (flags & MSG_DONTWAIT) | |
4167 | req->flags |= REQ_F_NOWAIT; | |
4168 | else if (force_nonblock) | |
4169 | flags |= MSG_DONTWAIT; | |
fddaface | 4170 | |
7a7cacba PB |
4171 | msg.msg_flags = flags; |
4172 | ret = sock_sendmsg(sock, &msg); | |
4173 | if (force_nonblock && ret == -EAGAIN) | |
4174 | return -EAGAIN; | |
4175 | if (ret == -ERESTARTSYS) | |
4176 | ret = -EINTR; | |
fddaface | 4177 | |
fddaface JA |
4178 | if (ret < 0) |
4179 | req_set_fail_links(req); | |
229a7b63 | 4180 | __io_req_complete(req, ret, 0, cs); |
fddaface | 4181 | return 0; |
fddaface JA |
4182 | } |
4183 | ||
1400e697 PB |
4184 | static int __io_recvmsg_copy_hdr(struct io_kiocb *req, |
4185 | struct io_async_msghdr *iomsg) | |
52de1fe1 JA |
4186 | { |
4187 | struct io_sr_msg *sr = &req->sr_msg; | |
4188 | struct iovec __user *uiov; | |
4189 | size_t iov_len; | |
4190 | int ret; | |
4191 | ||
1400e697 PB |
4192 | ret = __copy_msghdr_from_user(&iomsg->msg, sr->umsg, |
4193 | &iomsg->uaddr, &uiov, &iov_len); | |
52de1fe1 JA |
4194 | if (ret) |
4195 | return ret; | |
4196 | ||
4197 | if (req->flags & REQ_F_BUFFER_SELECT) { | |
4198 | if (iov_len > 1) | |
4199 | return -EINVAL; | |
1400e697 | 4200 | if (copy_from_user(iomsg->iov, uiov, sizeof(*uiov))) |
52de1fe1 | 4201 | return -EFAULT; |
1400e697 PB |
4202 | sr->len = iomsg->iov[0].iov_len; |
4203 | iov_iter_init(&iomsg->msg.msg_iter, READ, iomsg->iov, 1, | |
52de1fe1 | 4204 | sr->len); |
1400e697 | 4205 | iomsg->iov = NULL; |
52de1fe1 JA |
4206 | } else { |
4207 | ret = import_iovec(READ, uiov, iov_len, UIO_FASTIOV, | |
1400e697 | 4208 | &iomsg->iov, &iomsg->msg.msg_iter); |
52de1fe1 JA |
4209 | if (ret > 0) |
4210 | ret = 0; | |
4211 | } | |
4212 | ||
4213 | return ret; | |
4214 | } | |
4215 | ||
4216 | #ifdef CONFIG_COMPAT | |
4217 | static int __io_compat_recvmsg_copy_hdr(struct io_kiocb *req, | |
1400e697 | 4218 | struct io_async_msghdr *iomsg) |
52de1fe1 JA |
4219 | { |
4220 | struct compat_msghdr __user *msg_compat; | |
4221 | struct io_sr_msg *sr = &req->sr_msg; | |
4222 | struct compat_iovec __user *uiov; | |
4223 | compat_uptr_t ptr; | |
4224 | compat_size_t len; | |
4225 | int ret; | |
4226 | ||
270a5940 | 4227 | msg_compat = (struct compat_msghdr __user *) sr->umsg; |
1400e697 | 4228 | ret = __get_compat_msghdr(&iomsg->msg, msg_compat, &iomsg->uaddr, |
52de1fe1 JA |
4229 | &ptr, &len); |
4230 | if (ret) | |
4231 | return ret; | |
4232 | ||
4233 | uiov = compat_ptr(ptr); | |
4234 | if (req->flags & REQ_F_BUFFER_SELECT) { | |
4235 | compat_ssize_t clen; | |
4236 | ||
4237 | if (len > 1) | |
4238 | return -EINVAL; | |
4239 | if (!access_ok(uiov, sizeof(*uiov))) | |
4240 | return -EFAULT; | |
4241 | if (__get_user(clen, &uiov->iov_len)) | |
4242 | return -EFAULT; | |
4243 | if (clen < 0) | |
4244 | return -EINVAL; | |
1400e697 PB |
4245 | sr->len = iomsg->iov[0].iov_len; |
4246 | iomsg->iov = NULL; | |
52de1fe1 JA |
4247 | } else { |
4248 | ret = compat_import_iovec(READ, uiov, len, UIO_FASTIOV, | |
1400e697 PB |
4249 | &iomsg->iov, |
4250 | &iomsg->msg.msg_iter); | |
52de1fe1 JA |
4251 | if (ret < 0) |
4252 | return ret; | |
4253 | } | |
4254 | ||
4255 | return 0; | |
4256 | } | |
4257 | #endif | |
4258 | ||
1400e697 PB |
4259 | static int io_recvmsg_copy_hdr(struct io_kiocb *req, |
4260 | struct io_async_msghdr *iomsg) | |
52de1fe1 | 4261 | { |
1400e697 PB |
4262 | iomsg->msg.msg_name = &iomsg->addr; |
4263 | iomsg->iov = iomsg->fast_iov; | |
52de1fe1 JA |
4264 | |
4265 | #ifdef CONFIG_COMPAT | |
4266 | if (req->ctx->compat) | |
1400e697 | 4267 | return __io_compat_recvmsg_copy_hdr(req, iomsg); |
fddaface | 4268 | #endif |
52de1fe1 | 4269 | |
1400e697 | 4270 | return __io_recvmsg_copy_hdr(req, iomsg); |
52de1fe1 JA |
4271 | } |
4272 | ||
bcda7baa | 4273 | static struct io_buffer *io_recv_buffer_select(struct io_kiocb *req, |
7fbb1b54 | 4274 | bool needs_lock) |
bcda7baa JA |
4275 | { |
4276 | struct io_sr_msg *sr = &req->sr_msg; | |
4277 | struct io_buffer *kbuf; | |
4278 | ||
bcda7baa JA |
4279 | kbuf = io_buffer_select(req, &sr->len, sr->bgid, sr->kbuf, needs_lock); |
4280 | if (IS_ERR(kbuf)) | |
4281 | return kbuf; | |
4282 | ||
4283 | sr->kbuf = kbuf; | |
4284 | req->flags |= REQ_F_BUFFER_SELECTED; | |
bcda7baa | 4285 | return kbuf; |
fddaface JA |
4286 | } |
4287 | ||
7fbb1b54 PB |
4288 | static inline unsigned int io_put_recv_kbuf(struct io_kiocb *req) |
4289 | { | |
4290 | return io_put_kbuf(req, req->sr_msg.kbuf); | |
4291 | } | |
4292 | ||
3529d8c2 JA |
4293 | static int io_recvmsg_prep(struct io_kiocb *req, |
4294 | const struct io_uring_sqe *sqe) | |
aa1fa28f | 4295 | { |
e47293fd | 4296 | struct io_sr_msg *sr = &req->sr_msg; |
3529d8c2 | 4297 | struct io_async_ctx *io = req->io; |
99bc4c38 | 4298 | int ret; |
3529d8c2 | 4299 | |
d2b6f48b PB |
4300 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4301 | return -EINVAL; | |
4302 | ||
3529d8c2 | 4303 | sr->msg_flags = READ_ONCE(sqe->msg_flags); |
270a5940 | 4304 | sr->umsg = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
0b7b21e4 | 4305 | sr->len = READ_ONCE(sqe->len); |
bcda7baa | 4306 | sr->bgid = READ_ONCE(sqe->buf_group); |
06b76d44 | 4307 | |
d8768362 JA |
4308 | #ifdef CONFIG_COMPAT |
4309 | if (req->ctx->compat) | |
4310 | sr->msg_flags |= MSG_CMSG_COMPAT; | |
4311 | #endif | |
4312 | ||
fddaface | 4313 | if (!io || req->opcode == IORING_OP_RECV) |
06b76d44 | 4314 | return 0; |
5f798bea PB |
4315 | /* iovec is already imported */ |
4316 | if (req->flags & REQ_F_NEED_CLEANUP) | |
4317 | return 0; | |
03b1230c | 4318 | |
1400e697 | 4319 | ret = io_recvmsg_copy_hdr(req, &io->msg); |
99bc4c38 PB |
4320 | if (!ret) |
4321 | req->flags |= REQ_F_NEED_CLEANUP; | |
4322 | return ret; | |
aa1fa28f JA |
4323 | } |
4324 | ||
229a7b63 JA |
4325 | static int io_recvmsg(struct io_kiocb *req, bool force_nonblock, |
4326 | struct io_comp_state *cs) | |
aa1fa28f | 4327 | { |
6b754c8b | 4328 | struct io_async_msghdr iomsg, *kmsg; |
03b1230c | 4329 | struct socket *sock; |
7fbb1b54 | 4330 | struct io_buffer *kbuf; |
7a7cacba | 4331 | unsigned flags; |
52de1fe1 | 4332 | int ret, cflags = 0; |
03b1230c | 4333 | |
03b1230c | 4334 | sock = sock_from_file(req->file, &ret); |
7a7cacba PB |
4335 | if (unlikely(!sock)) |
4336 | return ret; | |
3529d8c2 | 4337 | |
7a7cacba PB |
4338 | if (req->io) { |
4339 | kmsg = &req->io->msg; | |
4340 | kmsg->msg.msg_name = &req->io->msg.addr; | |
4341 | /* if iov is set, it's allocated already */ | |
4342 | if (!kmsg->iov) | |
4343 | kmsg->iov = kmsg->fast_iov; | |
4344 | kmsg->msg.msg_iter.iov = kmsg->iov; | |
4345 | } else { | |
4346 | ret = io_recvmsg_copy_hdr(req, &iomsg); | |
4347 | if (ret) | |
681fda8d | 4348 | return ret; |
7a7cacba PB |
4349 | kmsg = &iomsg; |
4350 | } | |
03b1230c | 4351 | |
bc02ef33 | 4352 | if (req->flags & REQ_F_BUFFER_SELECT) { |
7fbb1b54 | 4353 | kbuf = io_recv_buffer_select(req, !force_nonblock); |
bc02ef33 | 4354 | if (IS_ERR(kbuf)) |
52de1fe1 | 4355 | return PTR_ERR(kbuf); |
7a7cacba PB |
4356 | kmsg->fast_iov[0].iov_base = u64_to_user_ptr(kbuf->addr); |
4357 | iov_iter_init(&kmsg->msg.msg_iter, READ, kmsg->iov, | |
4358 | 1, req->sr_msg.len); | |
4359 | } | |
52de1fe1 | 4360 | |
7a7cacba PB |
4361 | flags = req->sr_msg.msg_flags; |
4362 | if (flags & MSG_DONTWAIT) | |
4363 | req->flags |= REQ_F_NOWAIT; | |
4364 | else if (force_nonblock) | |
4365 | flags |= MSG_DONTWAIT; | |
e47293fd | 4366 | |
7a7cacba PB |
4367 | ret = __sys_recvmsg_sock(sock, &kmsg->msg, req->sr_msg.umsg, |
4368 | kmsg->uaddr, flags); | |
0e1b6fe3 PB |
4369 | if (force_nonblock && ret == -EAGAIN) |
4370 | return io_setup_async_msg(req, kmsg); | |
7a7cacba PB |
4371 | if (ret == -ERESTARTSYS) |
4372 | ret = -EINTR; | |
03b1230c | 4373 | |
7fbb1b54 PB |
4374 | if (req->flags & REQ_F_BUFFER_SELECTED) |
4375 | cflags = io_put_recv_kbuf(req); | |
6b754c8b | 4376 | if (kmsg->iov != kmsg->fast_iov) |
0b416c3e | 4377 | kfree(kmsg->iov); |
99bc4c38 | 4378 | req->flags &= ~REQ_F_NEED_CLEANUP; |
4e88d6e7 JA |
4379 | if (ret < 0) |
4380 | req_set_fail_links(req); | |
229a7b63 | 4381 | __io_req_complete(req, ret, cflags, cs); |
03b1230c | 4382 | return 0; |
0fa03c62 | 4383 | } |
5d17b4a4 | 4384 | |
229a7b63 JA |
4385 | static int io_recv(struct io_kiocb *req, bool force_nonblock, |
4386 | struct io_comp_state *cs) | |
fddaface | 4387 | { |
6b754c8b | 4388 | struct io_buffer *kbuf; |
7a7cacba PB |
4389 | struct io_sr_msg *sr = &req->sr_msg; |
4390 | struct msghdr msg; | |
4391 | void __user *buf = sr->buf; | |
fddaface | 4392 | struct socket *sock; |
7a7cacba PB |
4393 | struct iovec iov; |
4394 | unsigned flags; | |
bcda7baa | 4395 | int ret, cflags = 0; |
fddaface | 4396 | |
fddaface | 4397 | sock = sock_from_file(req->file, &ret); |
7a7cacba PB |
4398 | if (unlikely(!sock)) |
4399 | return ret; | |
fddaface | 4400 | |
bc02ef33 | 4401 | if (req->flags & REQ_F_BUFFER_SELECT) { |
7fbb1b54 | 4402 | kbuf = io_recv_buffer_select(req, !force_nonblock); |
bcda7baa JA |
4403 | if (IS_ERR(kbuf)) |
4404 | return PTR_ERR(kbuf); | |
7a7cacba | 4405 | buf = u64_to_user_ptr(kbuf->addr); |
bc02ef33 | 4406 | } |
bcda7baa | 4407 | |
7a7cacba | 4408 | ret = import_single_range(READ, buf, sr->len, &iov, &msg.msg_iter); |
14c32eee PB |
4409 | if (unlikely(ret)) |
4410 | goto out_free; | |
fddaface | 4411 | |
7a7cacba PB |
4412 | msg.msg_name = NULL; |
4413 | msg.msg_control = NULL; | |
4414 | msg.msg_controllen = 0; | |
4415 | msg.msg_namelen = 0; | |
4416 | msg.msg_iocb = NULL; | |
4417 | msg.msg_flags = 0; | |
fddaface | 4418 | |
7a7cacba PB |
4419 | flags = req->sr_msg.msg_flags; |
4420 | if (flags & MSG_DONTWAIT) | |
4421 | req->flags |= REQ_F_NOWAIT; | |
4422 | else if (force_nonblock) | |
4423 | flags |= MSG_DONTWAIT; | |
4424 | ||
4425 | ret = sock_recvmsg(sock, &msg, flags); | |
4426 | if (force_nonblock && ret == -EAGAIN) | |
4427 | return -EAGAIN; | |
4428 | if (ret == -ERESTARTSYS) | |
4429 | ret = -EINTR; | |
14c32eee | 4430 | out_free: |
7fbb1b54 PB |
4431 | if (req->flags & REQ_F_BUFFER_SELECTED) |
4432 | cflags = io_put_recv_kbuf(req); | |
fddaface JA |
4433 | if (ret < 0) |
4434 | req_set_fail_links(req); | |
229a7b63 | 4435 | __io_req_complete(req, ret, cflags, cs); |
fddaface | 4436 | return 0; |
fddaface JA |
4437 | } |
4438 | ||
3529d8c2 | 4439 | static int io_accept_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
17f2fe35 | 4440 | { |
8ed8d3c3 JA |
4441 | struct io_accept *accept = &req->accept; |
4442 | ||
17f2fe35 JA |
4443 | if (unlikely(req->ctx->flags & (IORING_SETUP_IOPOLL|IORING_SETUP_SQPOLL))) |
4444 | return -EINVAL; | |
8042d6ce | 4445 | if (sqe->ioprio || sqe->len || sqe->buf_index) |
17f2fe35 JA |
4446 | return -EINVAL; |
4447 | ||
d55e5f5b JA |
4448 | accept->addr = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
4449 | accept->addr_len = u64_to_user_ptr(READ_ONCE(sqe->addr2)); | |
8ed8d3c3 | 4450 | accept->flags = READ_ONCE(sqe->accept_flags); |
09952e3e | 4451 | accept->nofile = rlimit(RLIMIT_NOFILE); |
8ed8d3c3 | 4452 | return 0; |
8ed8d3c3 | 4453 | } |
17f2fe35 | 4454 | |
229a7b63 JA |
4455 | static int io_accept(struct io_kiocb *req, bool force_nonblock, |
4456 | struct io_comp_state *cs) | |
8ed8d3c3 JA |
4457 | { |
4458 | struct io_accept *accept = &req->accept; | |
ac45abc0 | 4459 | unsigned int file_flags = force_nonblock ? O_NONBLOCK : 0; |
8ed8d3c3 JA |
4460 | int ret; |
4461 | ||
e697deed JX |
4462 | if (req->file->f_flags & O_NONBLOCK) |
4463 | req->flags |= REQ_F_NOWAIT; | |
4464 | ||
8ed8d3c3 | 4465 | ret = __sys_accept4_file(req->file, file_flags, accept->addr, |
09952e3e JA |
4466 | accept->addr_len, accept->flags, |
4467 | accept->nofile); | |
8ed8d3c3 | 4468 | if (ret == -EAGAIN && force_nonblock) |
17f2fe35 | 4469 | return -EAGAIN; |
ac45abc0 PB |
4470 | if (ret < 0) { |
4471 | if (ret == -ERESTARTSYS) | |
4472 | ret = -EINTR; | |
4e88d6e7 | 4473 | req_set_fail_links(req); |
ac45abc0 | 4474 | } |
229a7b63 | 4475 | __io_req_complete(req, ret, 0, cs); |
17f2fe35 | 4476 | return 0; |
8ed8d3c3 JA |
4477 | } |
4478 | ||
3529d8c2 | 4479 | static int io_connect_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
f499a021 | 4480 | { |
3529d8c2 JA |
4481 | struct io_connect *conn = &req->connect; |
4482 | struct io_async_ctx *io = req->io; | |
f499a021 | 4483 | |
3fbb51c1 JA |
4484 | if (unlikely(req->ctx->flags & (IORING_SETUP_IOPOLL|IORING_SETUP_SQPOLL))) |
4485 | return -EINVAL; | |
4486 | if (sqe->ioprio || sqe->len || sqe->buf_index || sqe->rw_flags) | |
4487 | return -EINVAL; | |
4488 | ||
3529d8c2 JA |
4489 | conn->addr = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
4490 | conn->addr_len = READ_ONCE(sqe->addr2); | |
4491 | ||
4492 | if (!io) | |
4493 | return 0; | |
4494 | ||
4495 | return move_addr_to_kernel(conn->addr, conn->addr_len, | |
3fbb51c1 | 4496 | &io->connect.address); |
f499a021 JA |
4497 | } |
4498 | ||
229a7b63 JA |
4499 | static int io_connect(struct io_kiocb *req, bool force_nonblock, |
4500 | struct io_comp_state *cs) | |
f8e85cf2 | 4501 | { |
f499a021 | 4502 | struct io_async_ctx __io, *io; |
f8e85cf2 | 4503 | unsigned file_flags; |
3fbb51c1 | 4504 | int ret; |
f8e85cf2 | 4505 | |
f499a021 JA |
4506 | if (req->io) { |
4507 | io = req->io; | |
4508 | } else { | |
3529d8c2 JA |
4509 | ret = move_addr_to_kernel(req->connect.addr, |
4510 | req->connect.addr_len, | |
4511 | &__io.connect.address); | |
f499a021 JA |
4512 | if (ret) |
4513 | goto out; | |
4514 | io = &__io; | |
4515 | } | |
4516 | ||
3fbb51c1 JA |
4517 | file_flags = force_nonblock ? O_NONBLOCK : 0; |
4518 | ||
4519 | ret = __sys_connect_file(req->file, &io->connect.address, | |
4520 | req->connect.addr_len, file_flags); | |
87f80d62 | 4521 | if ((ret == -EAGAIN || ret == -EINPROGRESS) && force_nonblock) { |
b7bb4f7d JA |
4522 | if (req->io) |
4523 | return -EAGAIN; | |
4524 | if (io_alloc_async_ctx(req)) { | |
f499a021 JA |
4525 | ret = -ENOMEM; |
4526 | goto out; | |
4527 | } | |
b7bb4f7d | 4528 | memcpy(&req->io->connect, &__io.connect, sizeof(__io.connect)); |
f8e85cf2 | 4529 | return -EAGAIN; |
f499a021 | 4530 | } |
f8e85cf2 JA |
4531 | if (ret == -ERESTARTSYS) |
4532 | ret = -EINTR; | |
f499a021 | 4533 | out: |
4e88d6e7 JA |
4534 | if (ret < 0) |
4535 | req_set_fail_links(req); | |
229a7b63 | 4536 | __io_req_complete(req, ret, 0, cs); |
f8e85cf2 | 4537 | return 0; |
469956e8 Y |
4538 | } |
4539 | #else /* !CONFIG_NET */ | |
4540 | static int io_sendmsg_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) | |
4541 | { | |
f8e85cf2 | 4542 | return -EOPNOTSUPP; |
f8e85cf2 JA |
4543 | } |
4544 | ||
1e16c2f9 RD |
4545 | static int io_sendmsg(struct io_kiocb *req, bool force_nonblock, |
4546 | struct io_comp_state *cs) | |
469956e8 Y |
4547 | { |
4548 | return -EOPNOTSUPP; | |
4549 | } | |
4550 | ||
1e16c2f9 RD |
4551 | static int io_send(struct io_kiocb *req, bool force_nonblock, |
4552 | struct io_comp_state *cs) | |
469956e8 Y |
4553 | { |
4554 | return -EOPNOTSUPP; | |
4555 | } | |
4556 | ||
4557 | static int io_recvmsg_prep(struct io_kiocb *req, | |
4558 | const struct io_uring_sqe *sqe) | |
4559 | { | |
4560 | return -EOPNOTSUPP; | |
4561 | } | |
4562 | ||
1e16c2f9 RD |
4563 | static int io_recvmsg(struct io_kiocb *req, bool force_nonblock, |
4564 | struct io_comp_state *cs) | |
469956e8 Y |
4565 | { |
4566 | return -EOPNOTSUPP; | |
4567 | } | |
4568 | ||
1e16c2f9 RD |
4569 | static int io_recv(struct io_kiocb *req, bool force_nonblock, |
4570 | struct io_comp_state *cs) | |
469956e8 Y |
4571 | { |
4572 | return -EOPNOTSUPP; | |
4573 | } | |
4574 | ||
4575 | static int io_accept_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) | |
4576 | { | |
4577 | return -EOPNOTSUPP; | |
4578 | } | |
4579 | ||
1e16c2f9 RD |
4580 | static int io_accept(struct io_kiocb *req, bool force_nonblock, |
4581 | struct io_comp_state *cs) | |
469956e8 Y |
4582 | { |
4583 | return -EOPNOTSUPP; | |
4584 | } | |
ce593a6c | 4585 | |
469956e8 Y |
4586 | static int io_connect_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
4587 | { | |
4588 | return -EOPNOTSUPP; | |
4589 | } | |
4590 | ||
1e16c2f9 RD |
4591 | static int io_connect(struct io_kiocb *req, bool force_nonblock, |
4592 | struct io_comp_state *cs) | |
469956e8 | 4593 | { |
f8e85cf2 | 4594 | return -EOPNOTSUPP; |
ce593a6c | 4595 | } |
469956e8 | 4596 | #endif /* CONFIG_NET */ |
f8e85cf2 | 4597 | |
d7718a9d JA |
4598 | struct io_poll_table { |
4599 | struct poll_table_struct pt; | |
4600 | struct io_kiocb *req; | |
4601 | int error; | |
4602 | }; | |
ce593a6c | 4603 | |
d7718a9d JA |
4604 | static int __io_async_wake(struct io_kiocb *req, struct io_poll_iocb *poll, |
4605 | __poll_t mask, task_work_func_t func) | |
4606 | { | |
fd7d6de2 | 4607 | bool twa_signal_ok; |
aa96bf8a | 4608 | int ret; |
d7718a9d JA |
4609 | |
4610 | /* for instances that support it check for an event match first: */ | |
4611 | if (mask && !(mask & poll->events)) | |
4612 | return 0; | |
4613 | ||
4614 | trace_io_uring_task_add(req->ctx, req->opcode, req->user_data, mask); | |
4615 | ||
4616 | list_del_init(&poll->wait.entry); | |
4617 | ||
d7718a9d JA |
4618 | req->result = mask; |
4619 | init_task_work(&req->task_work, func); | |
6d816e08 JA |
4620 | percpu_ref_get(&req->ctx->refs); |
4621 | ||
fd7d6de2 JA |
4622 | /* |
4623 | * If we using the signalfd wait_queue_head for this wakeup, then | |
4624 | * it's not safe to use TWA_SIGNAL as we could be recursing on the | |
4625 | * tsk->sighand->siglock on doing the wakeup. Should not be needed | |
4626 | * either, as the normal wakeup will suffice. | |
4627 | */ | |
4628 | twa_signal_ok = (poll->head != &req->task->sighand->signalfd_wqh); | |
4629 | ||
d7718a9d | 4630 | /* |
e3aabf95 JA |
4631 | * If this fails, then the task is exiting. When a task exits, the |
4632 | * work gets canceled, so just cancel this request as well instead | |
4633 | * of executing it. We can't safely execute it anyway, as we may not | |
4634 | * have the needed state needed for it anyway. | |
d7718a9d | 4635 | */ |
fd7d6de2 | 4636 | ret = io_req_task_work_add(req, &req->task_work, twa_signal_ok); |
aa96bf8a | 4637 | if (unlikely(ret)) { |
c2c4c83c JA |
4638 | struct task_struct *tsk; |
4639 | ||
e3aabf95 | 4640 | WRITE_ONCE(poll->canceled, true); |
aa96bf8a | 4641 | tsk = io_wq_get_task(req->ctx->io_wq); |
ce593a6c JA |
4642 | task_work_add(tsk, &req->task_work, 0); |
4643 | wake_up_process(tsk); | |
aa96bf8a | 4644 | } |
d7718a9d JA |
4645 | return 1; |
4646 | } | |
4647 | ||
74ce6ce4 JA |
4648 | static bool io_poll_rewait(struct io_kiocb *req, struct io_poll_iocb *poll) |
4649 | __acquires(&req->ctx->completion_lock) | |
4650 | { | |
4651 | struct io_ring_ctx *ctx = req->ctx; | |
4652 | ||
4653 | if (!req->result && !READ_ONCE(poll->canceled)) { | |
4654 | struct poll_table_struct pt = { ._key = poll->events }; | |
4655 | ||
4656 | req->result = vfs_poll(req->file, &pt) & poll->events; | |
4657 | } | |
4658 | ||
4659 | spin_lock_irq(&ctx->completion_lock); | |
4660 | if (!req->result && !READ_ONCE(poll->canceled)) { | |
4661 | add_wait_queue(poll->head, &poll->wait); | |
4662 | return true; | |
4663 | } | |
4664 | ||
4665 | return false; | |
4666 | } | |
4667 | ||
d4e7cd36 | 4668 | static struct io_poll_iocb *io_poll_get_double(struct io_kiocb *req) |
18bceab1 | 4669 | { |
d4e7cd36 JA |
4670 | /* pure poll stashes this in ->io, poll driven retry elsewhere */ |
4671 | if (req->opcode == IORING_OP_POLL_ADD) | |
4672 | return (struct io_poll_iocb *) req->io; | |
4673 | return req->apoll->double_poll; | |
4674 | } | |
4675 | ||
4676 | static struct io_poll_iocb *io_poll_get_single(struct io_kiocb *req) | |
4677 | { | |
4678 | if (req->opcode == IORING_OP_POLL_ADD) | |
4679 | return &req->poll; | |
4680 | return &req->apoll->poll; | |
4681 | } | |
4682 | ||
4683 | static void io_poll_remove_double(struct io_kiocb *req) | |
4684 | { | |
4685 | struct io_poll_iocb *poll = io_poll_get_double(req); | |
18bceab1 JA |
4686 | |
4687 | lockdep_assert_held(&req->ctx->completion_lock); | |
4688 | ||
4689 | if (poll && poll->head) { | |
4690 | struct wait_queue_head *head = poll->head; | |
4691 | ||
4692 | spin_lock(&head->lock); | |
4693 | list_del_init(&poll->wait.entry); | |
4694 | if (poll->wait.private) | |
4695 | refcount_dec(&req->refs); | |
4696 | poll->head = NULL; | |
4697 | spin_unlock(&head->lock); | |
4698 | } | |
4699 | } | |
4700 | ||
4701 | static void io_poll_complete(struct io_kiocb *req, __poll_t mask, int error) | |
4702 | { | |
4703 | struct io_ring_ctx *ctx = req->ctx; | |
4704 | ||
d4e7cd36 | 4705 | io_poll_remove_double(req); |
18bceab1 JA |
4706 | req->poll.done = true; |
4707 | io_cqring_fill_event(req, error ? error : mangle_poll(mask)); | |
4708 | io_commit_cqring(ctx); | |
4709 | } | |
4710 | ||
4711 | static void io_poll_task_handler(struct io_kiocb *req, struct io_kiocb **nxt) | |
4712 | { | |
4713 | struct io_ring_ctx *ctx = req->ctx; | |
4714 | ||
4715 | if (io_poll_rewait(req, &req->poll)) { | |
4716 | spin_unlock_irq(&ctx->completion_lock); | |
4717 | return; | |
4718 | } | |
4719 | ||
4720 | hash_del(&req->hash_node); | |
4721 | io_poll_complete(req, req->result, 0); | |
4722 | req->flags |= REQ_F_COMP_LOCKED; | |
9b5f7bd9 | 4723 | *nxt = io_put_req_find_next(req); |
18bceab1 JA |
4724 | spin_unlock_irq(&ctx->completion_lock); |
4725 | ||
4726 | io_cqring_ev_posted(ctx); | |
4727 | } | |
4728 | ||
4729 | static void io_poll_task_func(struct callback_head *cb) | |
4730 | { | |
4731 | struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work); | |
6d816e08 | 4732 | struct io_ring_ctx *ctx = req->ctx; |
18bceab1 JA |
4733 | struct io_kiocb *nxt = NULL; |
4734 | ||
4735 | io_poll_task_handler(req, &nxt); | |
ea1164e5 PB |
4736 | if (nxt) |
4737 | __io_req_task_submit(nxt); | |
6d816e08 | 4738 | percpu_ref_put(&ctx->refs); |
18bceab1 JA |
4739 | } |
4740 | ||
4741 | static int io_poll_double_wake(struct wait_queue_entry *wait, unsigned mode, | |
4742 | int sync, void *key) | |
4743 | { | |
4744 | struct io_kiocb *req = wait->private; | |
d4e7cd36 | 4745 | struct io_poll_iocb *poll = io_poll_get_single(req); |
18bceab1 JA |
4746 | __poll_t mask = key_to_poll(key); |
4747 | ||
4748 | /* for instances that support it check for an event match first: */ | |
4749 | if (mask && !(mask & poll->events)) | |
4750 | return 0; | |
4751 | ||
8706e04e JA |
4752 | list_del_init(&wait->entry); |
4753 | ||
807abcb0 | 4754 | if (poll && poll->head) { |
18bceab1 JA |
4755 | bool done; |
4756 | ||
807abcb0 JA |
4757 | spin_lock(&poll->head->lock); |
4758 | done = list_empty(&poll->wait.entry); | |
18bceab1 | 4759 | if (!done) |
807abcb0 | 4760 | list_del_init(&poll->wait.entry); |
d4e7cd36 JA |
4761 | /* make sure double remove sees this as being gone */ |
4762 | wait->private = NULL; | |
807abcb0 | 4763 | spin_unlock(&poll->head->lock); |
18bceab1 JA |
4764 | if (!done) |
4765 | __io_async_wake(req, poll, mask, io_poll_task_func); | |
4766 | } | |
4767 | refcount_dec(&req->refs); | |
4768 | return 1; | |
4769 | } | |
4770 | ||
4771 | static void io_init_poll_iocb(struct io_poll_iocb *poll, __poll_t events, | |
4772 | wait_queue_func_t wake_func) | |
4773 | { | |
4774 | poll->head = NULL; | |
4775 | poll->done = false; | |
4776 | poll->canceled = false; | |
4777 | poll->events = events; | |
4778 | INIT_LIST_HEAD(&poll->wait.entry); | |
4779 | init_waitqueue_func_entry(&poll->wait, wake_func); | |
4780 | } | |
4781 | ||
4782 | static void __io_queue_proc(struct io_poll_iocb *poll, struct io_poll_table *pt, | |
807abcb0 JA |
4783 | struct wait_queue_head *head, |
4784 | struct io_poll_iocb **poll_ptr) | |
18bceab1 JA |
4785 | { |
4786 | struct io_kiocb *req = pt->req; | |
4787 | ||
4788 | /* | |
4789 | * If poll->head is already set, it's because the file being polled | |
4790 | * uses multiple waitqueues for poll handling (eg one for read, one | |
4791 | * for write). Setup a separate io_poll_iocb if this happens. | |
4792 | */ | |
4793 | if (unlikely(poll->head)) { | |
4794 | /* already have a 2nd entry, fail a third attempt */ | |
807abcb0 | 4795 | if (*poll_ptr) { |
18bceab1 JA |
4796 | pt->error = -EINVAL; |
4797 | return; | |
4798 | } | |
4799 | poll = kmalloc(sizeof(*poll), GFP_ATOMIC); | |
4800 | if (!poll) { | |
4801 | pt->error = -ENOMEM; | |
4802 | return; | |
4803 | } | |
4804 | io_init_poll_iocb(poll, req->poll.events, io_poll_double_wake); | |
4805 | refcount_inc(&req->refs); | |
4806 | poll->wait.private = req; | |
807abcb0 | 4807 | *poll_ptr = poll; |
18bceab1 JA |
4808 | } |
4809 | ||
4810 | pt->error = 0; | |
4811 | poll->head = head; | |
a31eb4a2 JX |
4812 | |
4813 | if (poll->events & EPOLLEXCLUSIVE) | |
4814 | add_wait_queue_exclusive(head, &poll->wait); | |
4815 | else | |
4816 | add_wait_queue(head, &poll->wait); | |
18bceab1 JA |
4817 | } |
4818 | ||
4819 | static void io_async_queue_proc(struct file *file, struct wait_queue_head *head, | |
4820 | struct poll_table_struct *p) | |
4821 | { | |
4822 | struct io_poll_table *pt = container_of(p, struct io_poll_table, pt); | |
807abcb0 | 4823 | struct async_poll *apoll = pt->req->apoll; |
18bceab1 | 4824 | |
807abcb0 | 4825 | __io_queue_proc(&apoll->poll, pt, head, &apoll->double_poll); |
18bceab1 JA |
4826 | } |
4827 | ||
d7718a9d JA |
4828 | static void io_async_task_func(struct callback_head *cb) |
4829 | { | |
4830 | struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work); | |
4831 | struct async_poll *apoll = req->apoll; | |
4832 | struct io_ring_ctx *ctx = req->ctx; | |
4833 | ||
4834 | trace_io_uring_task_run(req->ctx, req->opcode, req->user_data); | |
4835 | ||
74ce6ce4 | 4836 | if (io_poll_rewait(req, &apoll->poll)) { |
d7718a9d | 4837 | spin_unlock_irq(&ctx->completion_lock); |
6d816e08 | 4838 | percpu_ref_put(&ctx->refs); |
74ce6ce4 | 4839 | return; |
d7718a9d JA |
4840 | } |
4841 | ||
31067255 | 4842 | /* If req is still hashed, it cannot have been canceled. Don't check. */ |
0be0b0e3 | 4843 | if (hash_hashed(&req->hash_node)) |
74ce6ce4 | 4844 | hash_del(&req->hash_node); |
2bae047e | 4845 | |
d4e7cd36 | 4846 | io_poll_remove_double(req); |
74ce6ce4 JA |
4847 | spin_unlock_irq(&ctx->completion_lock); |
4848 | ||
0be0b0e3 PB |
4849 | if (!READ_ONCE(apoll->poll.canceled)) |
4850 | __io_req_task_submit(req); | |
4851 | else | |
4852 | __io_req_task_cancel(req, -ECANCELED); | |
aa340845 | 4853 | |
6d816e08 | 4854 | percpu_ref_put(&ctx->refs); |
807abcb0 | 4855 | kfree(apoll->double_poll); |
31067255 | 4856 | kfree(apoll); |
d7718a9d JA |
4857 | } |
4858 | ||
4859 | static int io_async_wake(struct wait_queue_entry *wait, unsigned mode, int sync, | |
4860 | void *key) | |
4861 | { | |
4862 | struct io_kiocb *req = wait->private; | |
4863 | struct io_poll_iocb *poll = &req->apoll->poll; | |
4864 | ||
4865 | trace_io_uring_poll_wake(req->ctx, req->opcode, req->user_data, | |
4866 | key_to_poll(key)); | |
4867 | ||
4868 | return __io_async_wake(req, poll, key_to_poll(key), io_async_task_func); | |
4869 | } | |
4870 | ||
4871 | static void io_poll_req_insert(struct io_kiocb *req) | |
4872 | { | |
4873 | struct io_ring_ctx *ctx = req->ctx; | |
4874 | struct hlist_head *list; | |
4875 | ||
4876 | list = &ctx->cancel_hash[hash_long(req->user_data, ctx->cancel_hash_bits)]; | |
4877 | hlist_add_head(&req->hash_node, list); | |
4878 | } | |
4879 | ||
4880 | static __poll_t __io_arm_poll_handler(struct io_kiocb *req, | |
4881 | struct io_poll_iocb *poll, | |
4882 | struct io_poll_table *ipt, __poll_t mask, | |
4883 | wait_queue_func_t wake_func) | |
4884 | __acquires(&ctx->completion_lock) | |
4885 | { | |
4886 | struct io_ring_ctx *ctx = req->ctx; | |
4887 | bool cancel = false; | |
4888 | ||
18bceab1 | 4889 | io_init_poll_iocb(poll, mask, wake_func); |
b90cd197 | 4890 | poll->file = req->file; |
18bceab1 | 4891 | poll->wait.private = req; |
d7718a9d JA |
4892 | |
4893 | ipt->pt._key = mask; | |
4894 | ipt->req = req; | |
4895 | ipt->error = -EINVAL; | |
4896 | ||
d7718a9d JA |
4897 | mask = vfs_poll(req->file, &ipt->pt) & poll->events; |
4898 | ||
4899 | spin_lock_irq(&ctx->completion_lock); | |
4900 | if (likely(poll->head)) { | |
4901 | spin_lock(&poll->head->lock); | |
4902 | if (unlikely(list_empty(&poll->wait.entry))) { | |
4903 | if (ipt->error) | |
4904 | cancel = true; | |
4905 | ipt->error = 0; | |
4906 | mask = 0; | |
4907 | } | |
4908 | if (mask || ipt->error) | |
4909 | list_del_init(&poll->wait.entry); | |
4910 | else if (cancel) | |
4911 | WRITE_ONCE(poll->canceled, true); | |
4912 | else if (!poll->done) /* actually waiting for an event */ | |
4913 | io_poll_req_insert(req); | |
4914 | spin_unlock(&poll->head->lock); | |
4915 | } | |
4916 | ||
4917 | return mask; | |
4918 | } | |
4919 | ||
4920 | static bool io_arm_poll_handler(struct io_kiocb *req) | |
4921 | { | |
4922 | const struct io_op_def *def = &io_op_defs[req->opcode]; | |
4923 | struct io_ring_ctx *ctx = req->ctx; | |
4924 | struct async_poll *apoll; | |
4925 | struct io_poll_table ipt; | |
4926 | __poll_t mask, ret; | |
9dab14b8 | 4927 | int rw; |
d7718a9d JA |
4928 | |
4929 | if (!req->file || !file_can_poll(req->file)) | |
4930 | return false; | |
24c74678 | 4931 | if (req->flags & REQ_F_POLLED) |
d7718a9d | 4932 | return false; |
9dab14b8 JA |
4933 | if (def->pollin) |
4934 | rw = READ; | |
4935 | else if (def->pollout) | |
4936 | rw = WRITE; | |
4937 | else | |
4938 | return false; | |
4939 | /* if we can't nonblock try, then no point in arming a poll handler */ | |
4940 | if (!io_file_supports_async(req->file, rw)) | |
d7718a9d JA |
4941 | return false; |
4942 | ||
4943 | apoll = kmalloc(sizeof(*apoll), GFP_ATOMIC); | |
4944 | if (unlikely(!apoll)) | |
4945 | return false; | |
807abcb0 | 4946 | apoll->double_poll = NULL; |
d7718a9d JA |
4947 | |
4948 | req->flags |= REQ_F_POLLED; | |
d7718a9d JA |
4949 | req->apoll = apoll; |
4950 | INIT_HLIST_NODE(&req->hash_node); | |
4951 | ||
8755d97a | 4952 | mask = 0; |
d7718a9d | 4953 | if (def->pollin) |
8755d97a | 4954 | mask |= POLLIN | POLLRDNORM; |
d7718a9d JA |
4955 | if (def->pollout) |
4956 | mask |= POLLOUT | POLLWRNORM; | |
4957 | mask |= POLLERR | POLLPRI; | |
4958 | ||
4959 | ipt.pt._qproc = io_async_queue_proc; | |
4960 | ||
4961 | ret = __io_arm_poll_handler(req, &apoll->poll, &ipt, mask, | |
4962 | io_async_wake); | |
a36da65c | 4963 | if (ret || ipt.error) { |
d4e7cd36 | 4964 | io_poll_remove_double(req); |
d7718a9d | 4965 | spin_unlock_irq(&ctx->completion_lock); |
807abcb0 | 4966 | kfree(apoll->double_poll); |
d7718a9d JA |
4967 | kfree(apoll); |
4968 | return false; | |
4969 | } | |
4970 | spin_unlock_irq(&ctx->completion_lock); | |
4971 | trace_io_uring_poll_arm(ctx, req->opcode, req->user_data, mask, | |
4972 | apoll->poll.events); | |
4973 | return true; | |
4974 | } | |
4975 | ||
4976 | static bool __io_poll_remove_one(struct io_kiocb *req, | |
4977 | struct io_poll_iocb *poll) | |
221c5eb2 | 4978 | { |
b41e9852 | 4979 | bool do_complete = false; |
221c5eb2 JA |
4980 | |
4981 | spin_lock(&poll->head->lock); | |
4982 | WRITE_ONCE(poll->canceled, true); | |
392edb45 JA |
4983 | if (!list_empty(&poll->wait.entry)) { |
4984 | list_del_init(&poll->wait.entry); | |
b41e9852 | 4985 | do_complete = true; |
221c5eb2 JA |
4986 | } |
4987 | spin_unlock(&poll->head->lock); | |
3bfa5bcb | 4988 | hash_del(&req->hash_node); |
d7718a9d JA |
4989 | return do_complete; |
4990 | } | |
4991 | ||
4992 | static bool io_poll_remove_one(struct io_kiocb *req) | |
4993 | { | |
4994 | bool do_complete; | |
4995 | ||
d4e7cd36 JA |
4996 | io_poll_remove_double(req); |
4997 | ||
d7718a9d JA |
4998 | if (req->opcode == IORING_OP_POLL_ADD) { |
4999 | do_complete = __io_poll_remove_one(req, &req->poll); | |
5000 | } else { | |
3bfa5bcb JA |
5001 | struct async_poll *apoll = req->apoll; |
5002 | ||
d7718a9d | 5003 | /* non-poll requests have submit ref still */ |
3bfa5bcb JA |
5004 | do_complete = __io_poll_remove_one(req, &apoll->poll); |
5005 | if (do_complete) { | |
d7718a9d | 5006 | io_put_req(req); |
807abcb0 | 5007 | kfree(apoll->double_poll); |
3bfa5bcb JA |
5008 | kfree(apoll); |
5009 | } | |
b1f573bd XW |
5010 | } |
5011 | ||
b41e9852 JA |
5012 | if (do_complete) { |
5013 | io_cqring_fill_event(req, -ECANCELED); | |
5014 | io_commit_cqring(req->ctx); | |
5015 | req->flags |= REQ_F_COMP_LOCKED; | |
f254ac04 | 5016 | req_set_fail_links(req); |
b41e9852 JA |
5017 | io_put_req(req); |
5018 | } | |
5019 | ||
5020 | return do_complete; | |
221c5eb2 JA |
5021 | } |
5022 | ||
76e1b642 JA |
5023 | /* |
5024 | * Returns true if we found and killed one or more poll requests | |
5025 | */ | |
5026 | static bool io_poll_remove_all(struct io_ring_ctx *ctx, struct task_struct *tsk) | |
221c5eb2 | 5027 | { |
78076bb6 | 5028 | struct hlist_node *tmp; |
221c5eb2 | 5029 | struct io_kiocb *req; |
8e2e1faf | 5030 | int posted = 0, i; |
221c5eb2 JA |
5031 | |
5032 | spin_lock_irq(&ctx->completion_lock); | |
78076bb6 JA |
5033 | for (i = 0; i < (1U << ctx->cancel_hash_bits); i++) { |
5034 | struct hlist_head *list; | |
5035 | ||
5036 | list = &ctx->cancel_hash[i]; | |
f3606e3a JA |
5037 | hlist_for_each_entry_safe(req, tmp, list, hash_node) { |
5038 | if (io_task_match(req, tsk)) | |
5039 | posted += io_poll_remove_one(req); | |
5040 | } | |
221c5eb2 JA |
5041 | } |
5042 | spin_unlock_irq(&ctx->completion_lock); | |
b41e9852 | 5043 | |
8e2e1faf JA |
5044 | if (posted) |
5045 | io_cqring_ev_posted(ctx); | |
76e1b642 JA |
5046 | |
5047 | return posted != 0; | |
221c5eb2 JA |
5048 | } |
5049 | ||
47f46768 JA |
5050 | static int io_poll_cancel(struct io_ring_ctx *ctx, __u64 sqe_addr) |
5051 | { | |
78076bb6 | 5052 | struct hlist_head *list; |
47f46768 JA |
5053 | struct io_kiocb *req; |
5054 | ||
78076bb6 JA |
5055 | list = &ctx->cancel_hash[hash_long(sqe_addr, ctx->cancel_hash_bits)]; |
5056 | hlist_for_each_entry(req, list, hash_node) { | |
b41e9852 JA |
5057 | if (sqe_addr != req->user_data) |
5058 | continue; | |
5059 | if (io_poll_remove_one(req)) | |
eac406c6 | 5060 | return 0; |
b41e9852 | 5061 | return -EALREADY; |
47f46768 JA |
5062 | } |
5063 | ||
5064 | return -ENOENT; | |
5065 | } | |
5066 | ||
3529d8c2 JA |
5067 | static int io_poll_remove_prep(struct io_kiocb *req, |
5068 | const struct io_uring_sqe *sqe) | |
0969e783 | 5069 | { |
0969e783 JA |
5070 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
5071 | return -EINVAL; | |
5072 | if (sqe->ioprio || sqe->off || sqe->len || sqe->buf_index || | |
5073 | sqe->poll_events) | |
5074 | return -EINVAL; | |
5075 | ||
5076 | req->poll.addr = READ_ONCE(sqe->addr); | |
0969e783 JA |
5077 | return 0; |
5078 | } | |
5079 | ||
221c5eb2 JA |
5080 | /* |
5081 | * Find a running poll command that matches one specified in sqe->addr, | |
5082 | * and remove it if found. | |
5083 | */ | |
fc4df999 | 5084 | static int io_poll_remove(struct io_kiocb *req) |
221c5eb2 JA |
5085 | { |
5086 | struct io_ring_ctx *ctx = req->ctx; | |
0969e783 | 5087 | u64 addr; |
47f46768 | 5088 | int ret; |
221c5eb2 | 5089 | |
0969e783 | 5090 | addr = req->poll.addr; |
221c5eb2 | 5091 | spin_lock_irq(&ctx->completion_lock); |
0969e783 | 5092 | ret = io_poll_cancel(ctx, addr); |
221c5eb2 JA |
5093 | spin_unlock_irq(&ctx->completion_lock); |
5094 | ||
4e88d6e7 JA |
5095 | if (ret < 0) |
5096 | req_set_fail_links(req); | |
e1e16097 | 5097 | io_req_complete(req, ret); |
221c5eb2 JA |
5098 | return 0; |
5099 | } | |
5100 | ||
221c5eb2 JA |
5101 | static int io_poll_wake(struct wait_queue_entry *wait, unsigned mode, int sync, |
5102 | void *key) | |
5103 | { | |
c2f2eb7d JA |
5104 | struct io_kiocb *req = wait->private; |
5105 | struct io_poll_iocb *poll = &req->poll; | |
221c5eb2 | 5106 | |
d7718a9d | 5107 | return __io_async_wake(req, poll, key_to_poll(key), io_poll_task_func); |
221c5eb2 JA |
5108 | } |
5109 | ||
221c5eb2 JA |
5110 | static void io_poll_queue_proc(struct file *file, struct wait_queue_head *head, |
5111 | struct poll_table_struct *p) | |
5112 | { | |
5113 | struct io_poll_table *pt = container_of(p, struct io_poll_table, pt); | |
5114 | ||
807abcb0 | 5115 | __io_queue_proc(&pt->req->poll, pt, head, (struct io_poll_iocb **) &pt->req->io); |
eac406c6 JA |
5116 | } |
5117 | ||
3529d8c2 | 5118 | static int io_poll_add_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
221c5eb2 JA |
5119 | { |
5120 | struct io_poll_iocb *poll = &req->poll; | |
5769a351 | 5121 | u32 events; |
221c5eb2 JA |
5122 | |
5123 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) | |
5124 | return -EINVAL; | |
5125 | if (sqe->addr || sqe->ioprio || sqe->off || sqe->len || sqe->buf_index) | |
5126 | return -EINVAL; | |
09bb8394 JA |
5127 | if (!poll->file) |
5128 | return -EBADF; | |
221c5eb2 | 5129 | |
5769a351 JX |
5130 | events = READ_ONCE(sqe->poll32_events); |
5131 | #ifdef __BIG_ENDIAN | |
5132 | events = swahw32(events); | |
5133 | #endif | |
a31eb4a2 JX |
5134 | poll->events = demangle_poll(events) | EPOLLERR | EPOLLHUP | |
5135 | (events & EPOLLEXCLUSIVE); | |
0969e783 JA |
5136 | return 0; |
5137 | } | |
5138 | ||
014db007 | 5139 | static int io_poll_add(struct io_kiocb *req) |
0969e783 JA |
5140 | { |
5141 | struct io_poll_iocb *poll = &req->poll; | |
5142 | struct io_ring_ctx *ctx = req->ctx; | |
5143 | struct io_poll_table ipt; | |
0969e783 | 5144 | __poll_t mask; |
0969e783 | 5145 | |
78076bb6 | 5146 | INIT_HLIST_NODE(&req->hash_node); |
d7718a9d | 5147 | ipt.pt._qproc = io_poll_queue_proc; |
36703247 | 5148 | |
d7718a9d JA |
5149 | mask = __io_arm_poll_handler(req, &req->poll, &ipt, poll->events, |
5150 | io_poll_wake); | |
221c5eb2 | 5151 | |
8c838788 | 5152 | if (mask) { /* no async, we'd stolen it */ |
221c5eb2 | 5153 | ipt.error = 0; |
b0dd8a41 | 5154 | io_poll_complete(req, mask, 0); |
221c5eb2 | 5155 | } |
221c5eb2 JA |
5156 | spin_unlock_irq(&ctx->completion_lock); |
5157 | ||
8c838788 JA |
5158 | if (mask) { |
5159 | io_cqring_ev_posted(ctx); | |
014db007 | 5160 | io_put_req(req); |
221c5eb2 | 5161 | } |
8c838788 | 5162 | return ipt.error; |
221c5eb2 JA |
5163 | } |
5164 | ||
5262f567 JA |
5165 | static enum hrtimer_restart io_timeout_fn(struct hrtimer *timer) |
5166 | { | |
ad8a48ac JA |
5167 | struct io_timeout_data *data = container_of(timer, |
5168 | struct io_timeout_data, timer); | |
5169 | struct io_kiocb *req = data->req; | |
5170 | struct io_ring_ctx *ctx = req->ctx; | |
5262f567 JA |
5171 | unsigned long flags; |
5172 | ||
5262f567 | 5173 | spin_lock_irqsave(&ctx->completion_lock, flags); |
01cec8c1 PB |
5174 | atomic_set(&req->ctx->cq_timeouts, |
5175 | atomic_read(&req->ctx->cq_timeouts) + 1); | |
5176 | ||
ef03681a | 5177 | /* |
11365043 JA |
5178 | * We could be racing with timeout deletion. If the list is empty, |
5179 | * then timeout lookup already found it and will be handling it. | |
ef03681a | 5180 | */ |
135fcde8 PB |
5181 | if (!list_empty(&req->timeout.list)) |
5182 | list_del_init(&req->timeout.list); | |
5262f567 | 5183 | |
78e19bbe | 5184 | io_cqring_fill_event(req, -ETIME); |
5262f567 JA |
5185 | io_commit_cqring(ctx); |
5186 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
5187 | ||
5188 | io_cqring_ev_posted(ctx); | |
4e88d6e7 | 5189 | req_set_fail_links(req); |
5262f567 JA |
5190 | io_put_req(req); |
5191 | return HRTIMER_NORESTART; | |
5192 | } | |
5193 | ||
f254ac04 JA |
5194 | static int __io_timeout_cancel(struct io_kiocb *req) |
5195 | { | |
5196 | int ret; | |
5197 | ||
5198 | list_del_init(&req->timeout.list); | |
5199 | ||
5200 | ret = hrtimer_try_to_cancel(&req->io->timeout.timer); | |
5201 | if (ret == -1) | |
5202 | return -EALREADY; | |
5203 | ||
5204 | req_set_fail_links(req); | |
5205 | req->flags |= REQ_F_COMP_LOCKED; | |
5206 | io_cqring_fill_event(req, -ECANCELED); | |
5207 | io_put_req(req); | |
5208 | return 0; | |
5209 | } | |
5210 | ||
47f46768 JA |
5211 | static int io_timeout_cancel(struct io_ring_ctx *ctx, __u64 user_data) |
5212 | { | |
5213 | struct io_kiocb *req; | |
5214 | int ret = -ENOENT; | |
5215 | ||
135fcde8 | 5216 | list_for_each_entry(req, &ctx->timeout_list, timeout.list) { |
47f46768 | 5217 | if (user_data == req->user_data) { |
47f46768 JA |
5218 | ret = 0; |
5219 | break; | |
5220 | } | |
5221 | } | |
5222 | ||
5223 | if (ret == -ENOENT) | |
5224 | return ret; | |
5225 | ||
f254ac04 | 5226 | return __io_timeout_cancel(req); |
47f46768 JA |
5227 | } |
5228 | ||
3529d8c2 JA |
5229 | static int io_timeout_remove_prep(struct io_kiocb *req, |
5230 | const struct io_uring_sqe *sqe) | |
b29472ee | 5231 | { |
b29472ee JA |
5232 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
5233 | return -EINVAL; | |
61710e43 DA |
5234 | if (unlikely(req->flags & (REQ_F_FIXED_FILE | REQ_F_BUFFER_SELECT))) |
5235 | return -EINVAL; | |
5236 | if (sqe->ioprio || sqe->buf_index || sqe->len) | |
b29472ee JA |
5237 | return -EINVAL; |
5238 | ||
5239 | req->timeout.addr = READ_ONCE(sqe->addr); | |
5240 | req->timeout.flags = READ_ONCE(sqe->timeout_flags); | |
5241 | if (req->timeout.flags) | |
5242 | return -EINVAL; | |
5243 | ||
b29472ee JA |
5244 | return 0; |
5245 | } | |
5246 | ||
11365043 JA |
5247 | /* |
5248 | * Remove or update an existing timeout command | |
5249 | */ | |
fc4df999 | 5250 | static int io_timeout_remove(struct io_kiocb *req) |
11365043 JA |
5251 | { |
5252 | struct io_ring_ctx *ctx = req->ctx; | |
47f46768 | 5253 | int ret; |
11365043 | 5254 | |
11365043 | 5255 | spin_lock_irq(&ctx->completion_lock); |
b29472ee | 5256 | ret = io_timeout_cancel(ctx, req->timeout.addr); |
11365043 | 5257 | |
47f46768 | 5258 | io_cqring_fill_event(req, ret); |
11365043 JA |
5259 | io_commit_cqring(ctx); |
5260 | spin_unlock_irq(&ctx->completion_lock); | |
5262f567 | 5261 | io_cqring_ev_posted(ctx); |
4e88d6e7 JA |
5262 | if (ret < 0) |
5263 | req_set_fail_links(req); | |
ec9c02ad | 5264 | io_put_req(req); |
11365043 | 5265 | return 0; |
5262f567 JA |
5266 | } |
5267 | ||
3529d8c2 | 5268 | static int io_timeout_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe, |
2d28390a | 5269 | bool is_timeout_link) |
5262f567 | 5270 | { |
ad8a48ac | 5271 | struct io_timeout_data *data; |
a41525ab | 5272 | unsigned flags; |
56080b02 | 5273 | u32 off = READ_ONCE(sqe->off); |
5262f567 | 5274 | |
ad8a48ac | 5275 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
5262f567 | 5276 | return -EINVAL; |
ad8a48ac | 5277 | if (sqe->ioprio || sqe->buf_index || sqe->len != 1) |
a41525ab | 5278 | return -EINVAL; |
56080b02 | 5279 | if (off && is_timeout_link) |
2d28390a | 5280 | return -EINVAL; |
a41525ab JA |
5281 | flags = READ_ONCE(sqe->timeout_flags); |
5282 | if (flags & ~IORING_TIMEOUT_ABS) | |
5262f567 | 5283 | return -EINVAL; |
bdf20073 | 5284 | |
bfe68a22 | 5285 | req->timeout.off = off; |
26a61679 | 5286 | |
3529d8c2 | 5287 | if (!req->io && io_alloc_async_ctx(req)) |
26a61679 JA |
5288 | return -ENOMEM; |
5289 | ||
5290 | data = &req->io->timeout; | |
ad8a48ac | 5291 | data->req = req; |
ad8a48ac JA |
5292 | |
5293 | if (get_timespec64(&data->ts, u64_to_user_ptr(sqe->addr))) | |
5262f567 JA |
5294 | return -EFAULT; |
5295 | ||
11365043 | 5296 | if (flags & IORING_TIMEOUT_ABS) |
ad8a48ac | 5297 | data->mode = HRTIMER_MODE_ABS; |
11365043 | 5298 | else |
ad8a48ac | 5299 | data->mode = HRTIMER_MODE_REL; |
11365043 | 5300 | |
ad8a48ac JA |
5301 | hrtimer_init(&data->timer, CLOCK_MONOTONIC, data->mode); |
5302 | return 0; | |
5303 | } | |
5304 | ||
fc4df999 | 5305 | static int io_timeout(struct io_kiocb *req) |
ad8a48ac | 5306 | { |
ad8a48ac | 5307 | struct io_ring_ctx *ctx = req->ctx; |
bfe68a22 | 5308 | struct io_timeout_data *data = &req->io->timeout; |
ad8a48ac | 5309 | struct list_head *entry; |
bfe68a22 | 5310 | u32 tail, off = req->timeout.off; |
ad8a48ac | 5311 | |
733f5c95 | 5312 | spin_lock_irq(&ctx->completion_lock); |
93bd25bb | 5313 | |
5262f567 JA |
5314 | /* |
5315 | * sqe->off holds how many events that need to occur for this | |
93bd25bb JA |
5316 | * timeout event to be satisfied. If it isn't set, then this is |
5317 | * a pure timeout request, sequence isn't used. | |
5262f567 | 5318 | */ |
8eb7e2d0 | 5319 | if (io_is_timeout_noseq(req)) { |
93bd25bb JA |
5320 | entry = ctx->timeout_list.prev; |
5321 | goto add; | |
5322 | } | |
5262f567 | 5323 | |
bfe68a22 PB |
5324 | tail = ctx->cached_cq_tail - atomic_read(&ctx->cq_timeouts); |
5325 | req->timeout.target_seq = tail + off; | |
5262f567 JA |
5326 | |
5327 | /* | |
5328 | * Insertion sort, ensuring the first entry in the list is always | |
5329 | * the one we need first. | |
5330 | */ | |
5262f567 | 5331 | list_for_each_prev(entry, &ctx->timeout_list) { |
135fcde8 PB |
5332 | struct io_kiocb *nxt = list_entry(entry, struct io_kiocb, |
5333 | timeout.list); | |
5262f567 | 5334 | |
8eb7e2d0 | 5335 | if (io_is_timeout_noseq(nxt)) |
93bd25bb | 5336 | continue; |
bfe68a22 PB |
5337 | /* nxt.seq is behind @tail, otherwise would've been completed */ |
5338 | if (off >= nxt->timeout.target_seq - tail) | |
5262f567 JA |
5339 | break; |
5340 | } | |
93bd25bb | 5341 | add: |
135fcde8 | 5342 | list_add(&req->timeout.list, entry); |
ad8a48ac JA |
5343 | data->timer.function = io_timeout_fn; |
5344 | hrtimer_start(&data->timer, timespec64_to_ktime(data->ts), data->mode); | |
5262f567 | 5345 | spin_unlock_irq(&ctx->completion_lock); |
5262f567 JA |
5346 | return 0; |
5347 | } | |
5262f567 | 5348 | |
62755e35 JA |
5349 | static bool io_cancel_cb(struct io_wq_work *work, void *data) |
5350 | { | |
5351 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); | |
5352 | ||
5353 | return req->user_data == (unsigned long) data; | |
5354 | } | |
5355 | ||
e977d6d3 | 5356 | static int io_async_cancel_one(struct io_ring_ctx *ctx, void *sqe_addr) |
62755e35 | 5357 | { |
62755e35 | 5358 | enum io_wq_cancel cancel_ret; |
62755e35 JA |
5359 | int ret = 0; |
5360 | ||
4f26bda1 | 5361 | cancel_ret = io_wq_cancel_cb(ctx->io_wq, io_cancel_cb, sqe_addr, false); |
62755e35 JA |
5362 | switch (cancel_ret) { |
5363 | case IO_WQ_CANCEL_OK: | |
5364 | ret = 0; | |
5365 | break; | |
5366 | case IO_WQ_CANCEL_RUNNING: | |
5367 | ret = -EALREADY; | |
5368 | break; | |
5369 | case IO_WQ_CANCEL_NOTFOUND: | |
5370 | ret = -ENOENT; | |
5371 | break; | |
5372 | } | |
5373 | ||
e977d6d3 JA |
5374 | return ret; |
5375 | } | |
5376 | ||
47f46768 JA |
5377 | static void io_async_find_and_cancel(struct io_ring_ctx *ctx, |
5378 | struct io_kiocb *req, __u64 sqe_addr, | |
014db007 | 5379 | int success_ret) |
47f46768 JA |
5380 | { |
5381 | unsigned long flags; | |
5382 | int ret; | |
5383 | ||
5384 | ret = io_async_cancel_one(ctx, (void *) (unsigned long) sqe_addr); | |
5385 | if (ret != -ENOENT) { | |
5386 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
5387 | goto done; | |
5388 | } | |
5389 | ||
5390 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
5391 | ret = io_timeout_cancel(ctx, sqe_addr); | |
5392 | if (ret != -ENOENT) | |
5393 | goto done; | |
5394 | ret = io_poll_cancel(ctx, sqe_addr); | |
5395 | done: | |
b0dd8a41 JA |
5396 | if (!ret) |
5397 | ret = success_ret; | |
47f46768 JA |
5398 | io_cqring_fill_event(req, ret); |
5399 | io_commit_cqring(ctx); | |
5400 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
5401 | io_cqring_ev_posted(ctx); | |
5402 | ||
4e88d6e7 JA |
5403 | if (ret < 0) |
5404 | req_set_fail_links(req); | |
014db007 | 5405 | io_put_req(req); |
47f46768 JA |
5406 | } |
5407 | ||
3529d8c2 JA |
5408 | static int io_async_cancel_prep(struct io_kiocb *req, |
5409 | const struct io_uring_sqe *sqe) | |
e977d6d3 | 5410 | { |
fbf23849 | 5411 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
e977d6d3 | 5412 | return -EINVAL; |
61710e43 DA |
5413 | if (unlikely(req->flags & (REQ_F_FIXED_FILE | REQ_F_BUFFER_SELECT))) |
5414 | return -EINVAL; | |
5415 | if (sqe->ioprio || sqe->off || sqe->len || sqe->cancel_flags) | |
e977d6d3 JA |
5416 | return -EINVAL; |
5417 | ||
fbf23849 JA |
5418 | req->cancel.addr = READ_ONCE(sqe->addr); |
5419 | return 0; | |
5420 | } | |
5421 | ||
014db007 | 5422 | static int io_async_cancel(struct io_kiocb *req) |
fbf23849 JA |
5423 | { |
5424 | struct io_ring_ctx *ctx = req->ctx; | |
fbf23849 | 5425 | |
014db007 | 5426 | io_async_find_and_cancel(ctx, req, req->cancel.addr, 0); |
5262f567 JA |
5427 | return 0; |
5428 | } | |
5429 | ||
05f3fb3c JA |
5430 | static int io_files_update_prep(struct io_kiocb *req, |
5431 | const struct io_uring_sqe *sqe) | |
5432 | { | |
6ca56f84 JA |
5433 | if (unlikely(req->ctx->flags & IORING_SETUP_SQPOLL)) |
5434 | return -EINVAL; | |
61710e43 DA |
5435 | if (unlikely(req->flags & (REQ_F_FIXED_FILE | REQ_F_BUFFER_SELECT))) |
5436 | return -EINVAL; | |
5437 | if (sqe->ioprio || sqe->rw_flags) | |
05f3fb3c JA |
5438 | return -EINVAL; |
5439 | ||
5440 | req->files_update.offset = READ_ONCE(sqe->off); | |
5441 | req->files_update.nr_args = READ_ONCE(sqe->len); | |
5442 | if (!req->files_update.nr_args) | |
5443 | return -EINVAL; | |
5444 | req->files_update.arg = READ_ONCE(sqe->addr); | |
5445 | return 0; | |
5446 | } | |
5447 | ||
229a7b63 JA |
5448 | static int io_files_update(struct io_kiocb *req, bool force_nonblock, |
5449 | struct io_comp_state *cs) | |
fbf23849 JA |
5450 | { |
5451 | struct io_ring_ctx *ctx = req->ctx; | |
05f3fb3c JA |
5452 | struct io_uring_files_update up; |
5453 | int ret; | |
fbf23849 | 5454 | |
f86cd20c | 5455 | if (force_nonblock) |
05f3fb3c | 5456 | return -EAGAIN; |
05f3fb3c JA |
5457 | |
5458 | up.offset = req->files_update.offset; | |
5459 | up.fds = req->files_update.arg; | |
5460 | ||
5461 | mutex_lock(&ctx->uring_lock); | |
5462 | ret = __io_sqe_files_update(ctx, &up, req->files_update.nr_args); | |
5463 | mutex_unlock(&ctx->uring_lock); | |
5464 | ||
5465 | if (ret < 0) | |
5466 | req_set_fail_links(req); | |
229a7b63 | 5467 | __io_req_complete(req, ret, 0, cs); |
5262f567 JA |
5468 | return 0; |
5469 | } | |
5470 | ||
3529d8c2 JA |
5471 | static int io_req_defer_prep(struct io_kiocb *req, |
5472 | const struct io_uring_sqe *sqe) | |
f67676d1 | 5473 | { |
e781573e | 5474 | ssize_t ret = 0; |
f67676d1 | 5475 | |
f1d96a8f PB |
5476 | if (!sqe) |
5477 | return 0; | |
5478 | ||
327d6d96 PB |
5479 | if (io_alloc_async_ctx(req)) |
5480 | return -EAGAIN; | |
f56040b8 PB |
5481 | ret = io_prep_work_files(req); |
5482 | if (unlikely(ret)) | |
5483 | return ret; | |
cccf0ee8 | 5484 | |
202700e1 JA |
5485 | io_prep_async_work(req); |
5486 | ||
d625c6ee | 5487 | switch (req->opcode) { |
e781573e JA |
5488 | case IORING_OP_NOP: |
5489 | break; | |
f67676d1 JA |
5490 | case IORING_OP_READV: |
5491 | case IORING_OP_READ_FIXED: | |
3a6820f2 | 5492 | case IORING_OP_READ: |
3529d8c2 | 5493 | ret = io_read_prep(req, sqe, true); |
f67676d1 JA |
5494 | break; |
5495 | case IORING_OP_WRITEV: | |
5496 | case IORING_OP_WRITE_FIXED: | |
3a6820f2 | 5497 | case IORING_OP_WRITE: |
3529d8c2 | 5498 | ret = io_write_prep(req, sqe, true); |
f67676d1 | 5499 | break; |
0969e783 | 5500 | case IORING_OP_POLL_ADD: |
3529d8c2 | 5501 | ret = io_poll_add_prep(req, sqe); |
0969e783 JA |
5502 | break; |
5503 | case IORING_OP_POLL_REMOVE: | |
3529d8c2 | 5504 | ret = io_poll_remove_prep(req, sqe); |
0969e783 | 5505 | break; |
8ed8d3c3 | 5506 | case IORING_OP_FSYNC: |
3529d8c2 | 5507 | ret = io_prep_fsync(req, sqe); |
8ed8d3c3 JA |
5508 | break; |
5509 | case IORING_OP_SYNC_FILE_RANGE: | |
3529d8c2 | 5510 | ret = io_prep_sfr(req, sqe); |
8ed8d3c3 | 5511 | break; |
03b1230c | 5512 | case IORING_OP_SENDMSG: |
fddaface | 5513 | case IORING_OP_SEND: |
3529d8c2 | 5514 | ret = io_sendmsg_prep(req, sqe); |
03b1230c JA |
5515 | break; |
5516 | case IORING_OP_RECVMSG: | |
fddaface | 5517 | case IORING_OP_RECV: |
3529d8c2 | 5518 | ret = io_recvmsg_prep(req, sqe); |
03b1230c | 5519 | break; |
f499a021 | 5520 | case IORING_OP_CONNECT: |
3529d8c2 | 5521 | ret = io_connect_prep(req, sqe); |
f499a021 | 5522 | break; |
2d28390a | 5523 | case IORING_OP_TIMEOUT: |
3529d8c2 | 5524 | ret = io_timeout_prep(req, sqe, false); |
b7bb4f7d | 5525 | break; |
b29472ee | 5526 | case IORING_OP_TIMEOUT_REMOVE: |
3529d8c2 | 5527 | ret = io_timeout_remove_prep(req, sqe); |
b29472ee | 5528 | break; |
fbf23849 | 5529 | case IORING_OP_ASYNC_CANCEL: |
3529d8c2 | 5530 | ret = io_async_cancel_prep(req, sqe); |
fbf23849 | 5531 | break; |
2d28390a | 5532 | case IORING_OP_LINK_TIMEOUT: |
3529d8c2 | 5533 | ret = io_timeout_prep(req, sqe, true); |
b7bb4f7d | 5534 | break; |
8ed8d3c3 | 5535 | case IORING_OP_ACCEPT: |
3529d8c2 | 5536 | ret = io_accept_prep(req, sqe); |
8ed8d3c3 | 5537 | break; |
d63d1b5e JA |
5538 | case IORING_OP_FALLOCATE: |
5539 | ret = io_fallocate_prep(req, sqe); | |
5540 | break; | |
15b71abe JA |
5541 | case IORING_OP_OPENAT: |
5542 | ret = io_openat_prep(req, sqe); | |
5543 | break; | |
b5dba59e JA |
5544 | case IORING_OP_CLOSE: |
5545 | ret = io_close_prep(req, sqe); | |
5546 | break; | |
05f3fb3c JA |
5547 | case IORING_OP_FILES_UPDATE: |
5548 | ret = io_files_update_prep(req, sqe); | |
5549 | break; | |
eddc7ef5 JA |
5550 | case IORING_OP_STATX: |
5551 | ret = io_statx_prep(req, sqe); | |
5552 | break; | |
4840e418 JA |
5553 | case IORING_OP_FADVISE: |
5554 | ret = io_fadvise_prep(req, sqe); | |
5555 | break; | |
c1ca757b JA |
5556 | case IORING_OP_MADVISE: |
5557 | ret = io_madvise_prep(req, sqe); | |
5558 | break; | |
cebdb986 JA |
5559 | case IORING_OP_OPENAT2: |
5560 | ret = io_openat2_prep(req, sqe); | |
5561 | break; | |
3e4827b0 JA |
5562 | case IORING_OP_EPOLL_CTL: |
5563 | ret = io_epoll_ctl_prep(req, sqe); | |
5564 | break; | |
7d67af2c PB |
5565 | case IORING_OP_SPLICE: |
5566 | ret = io_splice_prep(req, sqe); | |
5567 | break; | |
ddf0322d JA |
5568 | case IORING_OP_PROVIDE_BUFFERS: |
5569 | ret = io_provide_buffers_prep(req, sqe); | |
5570 | break; | |
067524e9 JA |
5571 | case IORING_OP_REMOVE_BUFFERS: |
5572 | ret = io_remove_buffers_prep(req, sqe); | |
5573 | break; | |
f2a8d5c7 PB |
5574 | case IORING_OP_TEE: |
5575 | ret = io_tee_prep(req, sqe); | |
5576 | break; | |
f67676d1 | 5577 | default: |
e781573e JA |
5578 | printk_once(KERN_WARNING "io_uring: unhandled opcode %d\n", |
5579 | req->opcode); | |
5580 | ret = -EINVAL; | |
b7bb4f7d | 5581 | break; |
f67676d1 JA |
5582 | } |
5583 | ||
b7bb4f7d | 5584 | return ret; |
f67676d1 JA |
5585 | } |
5586 | ||
9cf7c104 PB |
5587 | static u32 io_get_sequence(struct io_kiocb *req) |
5588 | { | |
5589 | struct io_kiocb *pos; | |
5590 | struct io_ring_ctx *ctx = req->ctx; | |
5591 | u32 total_submitted, nr_reqs = 1; | |
5592 | ||
5593 | if (req->flags & REQ_F_LINK_HEAD) | |
5594 | list_for_each_entry(pos, &req->link_list, link_list) | |
5595 | nr_reqs++; | |
5596 | ||
5597 | total_submitted = ctx->cached_sq_head - ctx->cached_sq_dropped; | |
5598 | return total_submitted - nr_reqs; | |
5599 | } | |
5600 | ||
3529d8c2 | 5601 | static int io_req_defer(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
de0617e4 | 5602 | { |
a197f664 | 5603 | struct io_ring_ctx *ctx = req->ctx; |
27dc8338 | 5604 | struct io_defer_entry *de; |
f67676d1 | 5605 | int ret; |
9cf7c104 | 5606 | u32 seq; |
de0617e4 | 5607 | |
9d858b21 | 5608 | /* Still need defer if there is pending req in defer list. */ |
9cf7c104 PB |
5609 | if (likely(list_empty_careful(&ctx->defer_list) && |
5610 | !(req->flags & REQ_F_IO_DRAIN))) | |
5611 | return 0; | |
5612 | ||
5613 | seq = io_get_sequence(req); | |
5614 | /* Still a chance to pass the sequence check */ | |
5615 | if (!req_need_defer(req, seq) && list_empty_careful(&ctx->defer_list)) | |
de0617e4 JA |
5616 | return 0; |
5617 | ||
650b5481 | 5618 | if (!req->io) { |
650b5481 | 5619 | ret = io_req_defer_prep(req, sqe); |
327d6d96 | 5620 | if (ret) |
650b5481 PB |
5621 | return ret; |
5622 | } | |
cbdcb435 | 5623 | io_prep_async_link(req); |
27dc8338 PB |
5624 | de = kmalloc(sizeof(*de), GFP_KERNEL); |
5625 | if (!de) | |
5626 | return -ENOMEM; | |
2d28390a | 5627 | |
de0617e4 | 5628 | spin_lock_irq(&ctx->completion_lock); |
9cf7c104 | 5629 | if (!req_need_defer(req, seq) && list_empty(&ctx->defer_list)) { |
de0617e4 | 5630 | spin_unlock_irq(&ctx->completion_lock); |
27dc8338 | 5631 | kfree(de); |
ae34817b PB |
5632 | io_queue_async_work(req); |
5633 | return -EIOCBQUEUED; | |
de0617e4 JA |
5634 | } |
5635 | ||
915967f6 | 5636 | trace_io_uring_defer(ctx, req, req->user_data); |
27dc8338 | 5637 | de->req = req; |
9cf7c104 | 5638 | de->seq = seq; |
27dc8338 | 5639 | list_add_tail(&de->list, &ctx->defer_list); |
de0617e4 JA |
5640 | spin_unlock_irq(&ctx->completion_lock); |
5641 | return -EIOCBQUEUED; | |
5642 | } | |
5643 | ||
f573d384 JA |
5644 | static void io_req_drop_files(struct io_kiocb *req) |
5645 | { | |
5646 | struct io_ring_ctx *ctx = req->ctx; | |
5647 | unsigned long flags; | |
5648 | ||
5649 | spin_lock_irqsave(&ctx->inflight_lock, flags); | |
5650 | list_del(&req->inflight_entry); | |
5651 | if (waitqueue_active(&ctx->inflight_wait)) | |
5652 | wake_up(&ctx->inflight_wait); | |
5653 | spin_unlock_irqrestore(&ctx->inflight_lock, flags); | |
5654 | req->flags &= ~REQ_F_INFLIGHT; | |
5655 | req->work.files = NULL; | |
5656 | } | |
5657 | ||
3ca405eb | 5658 | static void __io_clean_op(struct io_kiocb *req) |
99bc4c38 PB |
5659 | { |
5660 | struct io_async_ctx *io = req->io; | |
5661 | ||
0e1b6fe3 PB |
5662 | if (req->flags & REQ_F_BUFFER_SELECTED) { |
5663 | switch (req->opcode) { | |
5664 | case IORING_OP_READV: | |
5665 | case IORING_OP_READ_FIXED: | |
5666 | case IORING_OP_READ: | |
bcda7baa | 5667 | kfree((void *)(unsigned long)req->rw.addr); |
0e1b6fe3 PB |
5668 | break; |
5669 | case IORING_OP_RECVMSG: | |
5670 | case IORING_OP_RECV: | |
bcda7baa | 5671 | kfree(req->sr_msg.kbuf); |
0e1b6fe3 PB |
5672 | break; |
5673 | } | |
5674 | req->flags &= ~REQ_F_BUFFER_SELECTED; | |
99bc4c38 PB |
5675 | } |
5676 | ||
0e1b6fe3 PB |
5677 | if (req->flags & REQ_F_NEED_CLEANUP) { |
5678 | switch (req->opcode) { | |
5679 | case IORING_OP_READV: | |
5680 | case IORING_OP_READ_FIXED: | |
5681 | case IORING_OP_READ: | |
5682 | case IORING_OP_WRITEV: | |
5683 | case IORING_OP_WRITE_FIXED: | |
5684 | case IORING_OP_WRITE: | |
ff6165b2 JA |
5685 | if (io->rw.free_iovec) |
5686 | kfree(io->rw.free_iovec); | |
0e1b6fe3 PB |
5687 | break; |
5688 | case IORING_OP_RECVMSG: | |
5689 | case IORING_OP_SENDMSG: | |
5690 | if (io->msg.iov != io->msg.fast_iov) | |
5691 | kfree(io->msg.iov); | |
5692 | break; | |
5693 | case IORING_OP_SPLICE: | |
5694 | case IORING_OP_TEE: | |
5695 | io_put_file(req, req->splice.file_in, | |
5696 | (req->splice.flags & SPLICE_F_FD_IN_FIXED)); | |
5697 | break; | |
f3cd4850 JA |
5698 | case IORING_OP_OPENAT: |
5699 | case IORING_OP_OPENAT2: | |
5700 | if (req->open.filename) | |
5701 | putname(req->open.filename); | |
5702 | break; | |
0e1b6fe3 PB |
5703 | } |
5704 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
99bc4c38 | 5705 | } |
bb175342 | 5706 | |
f573d384 JA |
5707 | if (req->flags & REQ_F_INFLIGHT) |
5708 | io_req_drop_files(req); | |
99bc4c38 PB |
5709 | } |
5710 | ||
3529d8c2 | 5711 | static int io_issue_sqe(struct io_kiocb *req, const struct io_uring_sqe *sqe, |
f13fad7b | 5712 | bool force_nonblock, struct io_comp_state *cs) |
2b188cc1 | 5713 | { |
a197f664 | 5714 | struct io_ring_ctx *ctx = req->ctx; |
d625c6ee | 5715 | int ret; |
2b188cc1 | 5716 | |
d625c6ee | 5717 | switch (req->opcode) { |
2b188cc1 | 5718 | case IORING_OP_NOP: |
229a7b63 | 5719 | ret = io_nop(req, cs); |
2b188cc1 JA |
5720 | break; |
5721 | case IORING_OP_READV: | |
edafccee | 5722 | case IORING_OP_READ_FIXED: |
3a6820f2 | 5723 | case IORING_OP_READ: |
3529d8c2 JA |
5724 | if (sqe) { |
5725 | ret = io_read_prep(req, sqe, force_nonblock); | |
5726 | if (ret < 0) | |
5727 | break; | |
5728 | } | |
a1d7c393 | 5729 | ret = io_read(req, force_nonblock, cs); |
edafccee | 5730 | break; |
3529d8c2 | 5731 | case IORING_OP_WRITEV: |
edafccee | 5732 | case IORING_OP_WRITE_FIXED: |
3a6820f2 | 5733 | case IORING_OP_WRITE: |
3529d8c2 JA |
5734 | if (sqe) { |
5735 | ret = io_write_prep(req, sqe, force_nonblock); | |
5736 | if (ret < 0) | |
5737 | break; | |
5738 | } | |
a1d7c393 | 5739 | ret = io_write(req, force_nonblock, cs); |
2b188cc1 | 5740 | break; |
c992fe29 | 5741 | case IORING_OP_FSYNC: |
3529d8c2 JA |
5742 | if (sqe) { |
5743 | ret = io_prep_fsync(req, sqe); | |
5744 | if (ret < 0) | |
5745 | break; | |
5746 | } | |
014db007 | 5747 | ret = io_fsync(req, force_nonblock); |
c992fe29 | 5748 | break; |
221c5eb2 | 5749 | case IORING_OP_POLL_ADD: |
3529d8c2 JA |
5750 | if (sqe) { |
5751 | ret = io_poll_add_prep(req, sqe); | |
5752 | if (ret) | |
5753 | break; | |
5754 | } | |
014db007 | 5755 | ret = io_poll_add(req); |
221c5eb2 JA |
5756 | break; |
5757 | case IORING_OP_POLL_REMOVE: | |
3529d8c2 JA |
5758 | if (sqe) { |
5759 | ret = io_poll_remove_prep(req, sqe); | |
5760 | if (ret < 0) | |
5761 | break; | |
5762 | } | |
fc4df999 | 5763 | ret = io_poll_remove(req); |
221c5eb2 | 5764 | break; |
5d17b4a4 | 5765 | case IORING_OP_SYNC_FILE_RANGE: |
3529d8c2 JA |
5766 | if (sqe) { |
5767 | ret = io_prep_sfr(req, sqe); | |
5768 | if (ret < 0) | |
5769 | break; | |
5770 | } | |
014db007 | 5771 | ret = io_sync_file_range(req, force_nonblock); |
5d17b4a4 | 5772 | break; |
0fa03c62 | 5773 | case IORING_OP_SENDMSG: |
fddaface | 5774 | case IORING_OP_SEND: |
3529d8c2 JA |
5775 | if (sqe) { |
5776 | ret = io_sendmsg_prep(req, sqe); | |
5777 | if (ret < 0) | |
5778 | break; | |
5779 | } | |
fddaface | 5780 | if (req->opcode == IORING_OP_SENDMSG) |
229a7b63 | 5781 | ret = io_sendmsg(req, force_nonblock, cs); |
fddaface | 5782 | else |
229a7b63 | 5783 | ret = io_send(req, force_nonblock, cs); |
0fa03c62 | 5784 | break; |
aa1fa28f | 5785 | case IORING_OP_RECVMSG: |
fddaface | 5786 | case IORING_OP_RECV: |
3529d8c2 JA |
5787 | if (sqe) { |
5788 | ret = io_recvmsg_prep(req, sqe); | |
5789 | if (ret) | |
5790 | break; | |
5791 | } | |
fddaface | 5792 | if (req->opcode == IORING_OP_RECVMSG) |
229a7b63 | 5793 | ret = io_recvmsg(req, force_nonblock, cs); |
fddaface | 5794 | else |
229a7b63 | 5795 | ret = io_recv(req, force_nonblock, cs); |
aa1fa28f | 5796 | break; |
5262f567 | 5797 | case IORING_OP_TIMEOUT: |
3529d8c2 JA |
5798 | if (sqe) { |
5799 | ret = io_timeout_prep(req, sqe, false); | |
5800 | if (ret) | |
5801 | break; | |
5802 | } | |
fc4df999 | 5803 | ret = io_timeout(req); |
5262f567 | 5804 | break; |
11365043 | 5805 | case IORING_OP_TIMEOUT_REMOVE: |
3529d8c2 JA |
5806 | if (sqe) { |
5807 | ret = io_timeout_remove_prep(req, sqe); | |
5808 | if (ret) | |
5809 | break; | |
5810 | } | |
fc4df999 | 5811 | ret = io_timeout_remove(req); |
11365043 | 5812 | break; |
17f2fe35 | 5813 | case IORING_OP_ACCEPT: |
3529d8c2 JA |
5814 | if (sqe) { |
5815 | ret = io_accept_prep(req, sqe); | |
5816 | if (ret) | |
5817 | break; | |
5818 | } | |
229a7b63 | 5819 | ret = io_accept(req, force_nonblock, cs); |
17f2fe35 | 5820 | break; |
f8e85cf2 | 5821 | case IORING_OP_CONNECT: |
3529d8c2 JA |
5822 | if (sqe) { |
5823 | ret = io_connect_prep(req, sqe); | |
5824 | if (ret) | |
5825 | break; | |
5826 | } | |
229a7b63 | 5827 | ret = io_connect(req, force_nonblock, cs); |
f8e85cf2 | 5828 | break; |
62755e35 | 5829 | case IORING_OP_ASYNC_CANCEL: |
3529d8c2 JA |
5830 | if (sqe) { |
5831 | ret = io_async_cancel_prep(req, sqe); | |
5832 | if (ret) | |
5833 | break; | |
5834 | } | |
014db007 | 5835 | ret = io_async_cancel(req); |
62755e35 | 5836 | break; |
d63d1b5e JA |
5837 | case IORING_OP_FALLOCATE: |
5838 | if (sqe) { | |
5839 | ret = io_fallocate_prep(req, sqe); | |
5840 | if (ret) | |
5841 | break; | |
5842 | } | |
014db007 | 5843 | ret = io_fallocate(req, force_nonblock); |
d63d1b5e | 5844 | break; |
15b71abe JA |
5845 | case IORING_OP_OPENAT: |
5846 | if (sqe) { | |
5847 | ret = io_openat_prep(req, sqe); | |
5848 | if (ret) | |
5849 | break; | |
5850 | } | |
014db007 | 5851 | ret = io_openat(req, force_nonblock); |
15b71abe | 5852 | break; |
b5dba59e JA |
5853 | case IORING_OP_CLOSE: |
5854 | if (sqe) { | |
5855 | ret = io_close_prep(req, sqe); | |
5856 | if (ret) | |
5857 | break; | |
5858 | } | |
229a7b63 | 5859 | ret = io_close(req, force_nonblock, cs); |
b5dba59e | 5860 | break; |
05f3fb3c JA |
5861 | case IORING_OP_FILES_UPDATE: |
5862 | if (sqe) { | |
5863 | ret = io_files_update_prep(req, sqe); | |
5864 | if (ret) | |
5865 | break; | |
5866 | } | |
229a7b63 | 5867 | ret = io_files_update(req, force_nonblock, cs); |
05f3fb3c | 5868 | break; |
eddc7ef5 JA |
5869 | case IORING_OP_STATX: |
5870 | if (sqe) { | |
5871 | ret = io_statx_prep(req, sqe); | |
5872 | if (ret) | |
5873 | break; | |
5874 | } | |
014db007 | 5875 | ret = io_statx(req, force_nonblock); |
eddc7ef5 | 5876 | break; |
4840e418 JA |
5877 | case IORING_OP_FADVISE: |
5878 | if (sqe) { | |
5879 | ret = io_fadvise_prep(req, sqe); | |
5880 | if (ret) | |
5881 | break; | |
5882 | } | |
014db007 | 5883 | ret = io_fadvise(req, force_nonblock); |
4840e418 | 5884 | break; |
c1ca757b JA |
5885 | case IORING_OP_MADVISE: |
5886 | if (sqe) { | |
5887 | ret = io_madvise_prep(req, sqe); | |
5888 | if (ret) | |
5889 | break; | |
5890 | } | |
014db007 | 5891 | ret = io_madvise(req, force_nonblock); |
c1ca757b | 5892 | break; |
cebdb986 JA |
5893 | case IORING_OP_OPENAT2: |
5894 | if (sqe) { | |
5895 | ret = io_openat2_prep(req, sqe); | |
5896 | if (ret) | |
5897 | break; | |
5898 | } | |
014db007 | 5899 | ret = io_openat2(req, force_nonblock); |
cebdb986 | 5900 | break; |
3e4827b0 JA |
5901 | case IORING_OP_EPOLL_CTL: |
5902 | if (sqe) { | |
5903 | ret = io_epoll_ctl_prep(req, sqe); | |
5904 | if (ret) | |
5905 | break; | |
5906 | } | |
229a7b63 | 5907 | ret = io_epoll_ctl(req, force_nonblock, cs); |
3e4827b0 | 5908 | break; |
7d67af2c PB |
5909 | case IORING_OP_SPLICE: |
5910 | if (sqe) { | |
5911 | ret = io_splice_prep(req, sqe); | |
5912 | if (ret < 0) | |
5913 | break; | |
5914 | } | |
014db007 | 5915 | ret = io_splice(req, force_nonblock); |
7d67af2c | 5916 | break; |
ddf0322d JA |
5917 | case IORING_OP_PROVIDE_BUFFERS: |
5918 | if (sqe) { | |
5919 | ret = io_provide_buffers_prep(req, sqe); | |
5920 | if (ret) | |
5921 | break; | |
5922 | } | |
229a7b63 | 5923 | ret = io_provide_buffers(req, force_nonblock, cs); |
ddf0322d | 5924 | break; |
067524e9 JA |
5925 | case IORING_OP_REMOVE_BUFFERS: |
5926 | if (sqe) { | |
5927 | ret = io_remove_buffers_prep(req, sqe); | |
5928 | if (ret) | |
5929 | break; | |
5930 | } | |
229a7b63 | 5931 | ret = io_remove_buffers(req, force_nonblock, cs); |
3e4827b0 | 5932 | break; |
f2a8d5c7 PB |
5933 | case IORING_OP_TEE: |
5934 | if (sqe) { | |
5935 | ret = io_tee_prep(req, sqe); | |
5936 | if (ret < 0) | |
5937 | break; | |
5938 | } | |
5939 | ret = io_tee(req, force_nonblock); | |
5940 | break; | |
2b188cc1 JA |
5941 | default: |
5942 | ret = -EINVAL; | |
5943 | break; | |
5944 | } | |
5945 | ||
def596e9 JA |
5946 | if (ret) |
5947 | return ret; | |
5948 | ||
b532576e JA |
5949 | /* If the op doesn't have a file, we're not polling for it */ |
5950 | if ((ctx->flags & IORING_SETUP_IOPOLL) && req->file) { | |
11ba820b JA |
5951 | const bool in_async = io_wq_current_is_worker(); |
5952 | ||
11ba820b JA |
5953 | /* workqueue context doesn't hold uring_lock, grab it now */ |
5954 | if (in_async) | |
5955 | mutex_lock(&ctx->uring_lock); | |
5956 | ||
def596e9 | 5957 | io_iopoll_req_issued(req); |
11ba820b JA |
5958 | |
5959 | if (in_async) | |
5960 | mutex_unlock(&ctx->uring_lock); | |
def596e9 JA |
5961 | } |
5962 | ||
5963 | return 0; | |
2b188cc1 JA |
5964 | } |
5965 | ||
f4db7182 | 5966 | static struct io_wq_work *io_wq_submit_work(struct io_wq_work *work) |
2b188cc1 JA |
5967 | { |
5968 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); | |
6df1db6b | 5969 | struct io_kiocb *timeout; |
561fb04a | 5970 | int ret = 0; |
2b188cc1 | 5971 | |
6df1db6b PB |
5972 | timeout = io_prep_linked_timeout(req); |
5973 | if (timeout) | |
5974 | io_queue_linked_timeout(timeout); | |
d4c81f38 | 5975 | |
0c9d5ccd JA |
5976 | /* if NO_CANCEL is set, we must still run the work */ |
5977 | if ((work->flags & (IO_WQ_WORK_CANCEL|IO_WQ_WORK_NO_CANCEL)) == | |
5978 | IO_WQ_WORK_CANCEL) { | |
561fb04a | 5979 | ret = -ECANCELED; |
0c9d5ccd | 5980 | } |
31b51510 | 5981 | |
561fb04a | 5982 | if (!ret) { |
561fb04a | 5983 | do { |
f13fad7b | 5984 | ret = io_issue_sqe(req, NULL, false, NULL); |
561fb04a JA |
5985 | /* |
5986 | * We can get EAGAIN for polled IO even though we're | |
5987 | * forcing a sync submission from here, since we can't | |
5988 | * wait for request slots on the block side. | |
5989 | */ | |
5990 | if (ret != -EAGAIN) | |
5991 | break; | |
5992 | cond_resched(); | |
5993 | } while (1); | |
5994 | } | |
31b51510 | 5995 | |
561fb04a | 5996 | if (ret) { |
4e88d6e7 | 5997 | req_set_fail_links(req); |
e1e16097 | 5998 | io_req_complete(req, ret); |
edafccee | 5999 | } |
2b188cc1 | 6000 | |
f4db7182 | 6001 | return io_steal_work(req); |
2b188cc1 JA |
6002 | } |
6003 | ||
65e19f54 JA |
6004 | static inline struct file *io_file_from_index(struct io_ring_ctx *ctx, |
6005 | int index) | |
6006 | { | |
6007 | struct fixed_file_table *table; | |
6008 | ||
05f3fb3c | 6009 | table = &ctx->file_data->table[index >> IORING_FILE_TABLE_SHIFT]; |
84695089 | 6010 | return table->files[index & IORING_FILE_TABLE_MASK]; |
65e19f54 JA |
6011 | } |
6012 | ||
8da11c19 PB |
6013 | static int io_file_get(struct io_submit_state *state, struct io_kiocb *req, |
6014 | int fd, struct file **out_file, bool fixed) | |
09bb8394 | 6015 | { |
a197f664 | 6016 | struct io_ring_ctx *ctx = req->ctx; |
8da11c19 | 6017 | struct file *file; |
09bb8394 | 6018 | |
8da11c19 | 6019 | if (fixed) { |
05f3fb3c | 6020 | if (unlikely(!ctx->file_data || |
09bb8394 JA |
6021 | (unsigned) fd >= ctx->nr_user_files)) |
6022 | return -EBADF; | |
b7620121 | 6023 | fd = array_index_nospec(fd, ctx->nr_user_files); |
8da11c19 | 6024 | file = io_file_from_index(ctx, fd); |
fd2206e4 JA |
6025 | if (file) { |
6026 | req->fixed_file_refs = ctx->file_data->cur_refs; | |
6027 | percpu_ref_get(req->fixed_file_refs); | |
6028 | } | |
09bb8394 | 6029 | } else { |
c826bd7a | 6030 | trace_io_uring_file_get(ctx, fd); |
8da11c19 | 6031 | file = __io_file_get(state, fd); |
09bb8394 JA |
6032 | } |
6033 | ||
fd2206e4 JA |
6034 | if (file || io_op_defs[req->opcode].needs_file_no_error) { |
6035 | *out_file = file; | |
6036 | return 0; | |
6037 | } | |
6038 | return -EBADF; | |
09bb8394 JA |
6039 | } |
6040 | ||
8da11c19 | 6041 | static int io_req_set_file(struct io_submit_state *state, struct io_kiocb *req, |
63ff8223 | 6042 | int fd) |
8da11c19 | 6043 | { |
8da11c19 PB |
6044 | bool fixed; |
6045 | ||
63ff8223 | 6046 | fixed = (req->flags & REQ_F_FIXED_FILE) != 0; |
0cdaf760 | 6047 | if (unlikely(!fixed && io_async_submit(req->ctx))) |
8da11c19 PB |
6048 | return -EBADF; |
6049 | ||
6050 | return io_file_get(state, req, fd, &req->file, fixed); | |
6051 | } | |
6052 | ||
a197f664 | 6053 | static int io_grab_files(struct io_kiocb *req) |
fcb323cc JA |
6054 | { |
6055 | int ret = -EBADF; | |
a197f664 | 6056 | struct io_ring_ctx *ctx = req->ctx; |
fcb323cc | 6057 | |
f56040b8 PB |
6058 | io_req_init_async(req); |
6059 | ||
5b0bbee4 | 6060 | if (req->work.files || (req->flags & REQ_F_NO_FILE_TABLE)) |
f86cd20c | 6061 | return 0; |
b14cca0c | 6062 | if (!ctx->ring_file) |
b5dba59e JA |
6063 | return -EBADF; |
6064 | ||
fcb323cc JA |
6065 | rcu_read_lock(); |
6066 | spin_lock_irq(&ctx->inflight_lock); | |
6067 | /* | |
6068 | * We use the f_ops->flush() handler to ensure that we can flush | |
6069 | * out work accessing these files if the fd is closed. Check if | |
6070 | * the fd has changed since we started down this path, and disallow | |
6071 | * this operation if it has. | |
6072 | */ | |
b14cca0c | 6073 | if (fcheck(ctx->ring_fd) == ctx->ring_file) { |
fcb323cc JA |
6074 | list_add(&req->inflight_entry, &ctx->inflight_list); |
6075 | req->flags |= REQ_F_INFLIGHT; | |
6076 | req->work.files = current->files; | |
6077 | ret = 0; | |
6078 | } | |
6079 | spin_unlock_irq(&ctx->inflight_lock); | |
6080 | rcu_read_unlock(); | |
6081 | ||
6082 | return ret; | |
6083 | } | |
6084 | ||
f56040b8 PB |
6085 | static inline int io_prep_work_files(struct io_kiocb *req) |
6086 | { | |
6087 | if (!io_op_defs[req->opcode].file_table) | |
6088 | return 0; | |
6089 | return io_grab_files(req); | |
6090 | } | |
6091 | ||
2665abfd | 6092 | static enum hrtimer_restart io_link_timeout_fn(struct hrtimer *timer) |
2b188cc1 | 6093 | { |
ad8a48ac JA |
6094 | struct io_timeout_data *data = container_of(timer, |
6095 | struct io_timeout_data, timer); | |
6096 | struct io_kiocb *req = data->req; | |
2665abfd JA |
6097 | struct io_ring_ctx *ctx = req->ctx; |
6098 | struct io_kiocb *prev = NULL; | |
6099 | unsigned long flags; | |
2665abfd JA |
6100 | |
6101 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
6102 | ||
6103 | /* | |
6104 | * We don't expect the list to be empty, that will only happen if we | |
6105 | * race with the completion of the linked work. | |
6106 | */ | |
4493233e PB |
6107 | if (!list_empty(&req->link_list)) { |
6108 | prev = list_entry(req->link_list.prev, struct io_kiocb, | |
6109 | link_list); | |
5d960724 | 6110 | if (refcount_inc_not_zero(&prev->refs)) { |
4493233e | 6111 | list_del_init(&req->link_list); |
5d960724 JA |
6112 | prev->flags &= ~REQ_F_LINK_TIMEOUT; |
6113 | } else | |
76a46e06 | 6114 | prev = NULL; |
2665abfd JA |
6115 | } |
6116 | ||
6117 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
6118 | ||
6119 | if (prev) { | |
4e88d6e7 | 6120 | req_set_fail_links(prev); |
014db007 | 6121 | io_async_find_and_cancel(ctx, req, prev->user_data, -ETIME); |
76a46e06 | 6122 | io_put_req(prev); |
47f46768 | 6123 | } else { |
e1e16097 | 6124 | io_req_complete(req, -ETIME); |
2665abfd | 6125 | } |
2665abfd JA |
6126 | return HRTIMER_NORESTART; |
6127 | } | |
6128 | ||
7271ef3a | 6129 | static void __io_queue_linked_timeout(struct io_kiocb *req) |
2665abfd | 6130 | { |
76a46e06 JA |
6131 | /* |
6132 | * If the list is now empty, then our linked request finished before | |
6133 | * we got a chance to setup the timer | |
6134 | */ | |
4493233e | 6135 | if (!list_empty(&req->link_list)) { |
2d28390a | 6136 | struct io_timeout_data *data = &req->io->timeout; |
94ae5e77 | 6137 | |
ad8a48ac JA |
6138 | data->timer.function = io_link_timeout_fn; |
6139 | hrtimer_start(&data->timer, timespec64_to_ktime(data->ts), | |
6140 | data->mode); | |
2665abfd | 6141 | } |
7271ef3a JA |
6142 | } |
6143 | ||
6144 | static void io_queue_linked_timeout(struct io_kiocb *req) | |
6145 | { | |
6146 | struct io_ring_ctx *ctx = req->ctx; | |
6147 | ||
6148 | spin_lock_irq(&ctx->completion_lock); | |
6149 | __io_queue_linked_timeout(req); | |
76a46e06 | 6150 | spin_unlock_irq(&ctx->completion_lock); |
2665abfd | 6151 | |
2665abfd | 6152 | /* drop submission reference */ |
76a46e06 JA |
6153 | io_put_req(req); |
6154 | } | |
2665abfd | 6155 | |
ad8a48ac | 6156 | static struct io_kiocb *io_prep_linked_timeout(struct io_kiocb *req) |
2665abfd JA |
6157 | { |
6158 | struct io_kiocb *nxt; | |
6159 | ||
dea3b49c | 6160 | if (!(req->flags & REQ_F_LINK_HEAD)) |
2665abfd | 6161 | return NULL; |
6df1db6b | 6162 | if (req->flags & REQ_F_LINK_TIMEOUT) |
d7718a9d | 6163 | return NULL; |
2665abfd | 6164 | |
4493233e PB |
6165 | nxt = list_first_entry_or_null(&req->link_list, struct io_kiocb, |
6166 | link_list); | |
d625c6ee | 6167 | if (!nxt || nxt->opcode != IORING_OP_LINK_TIMEOUT) |
76a46e06 | 6168 | return NULL; |
2665abfd | 6169 | |
76a46e06 | 6170 | req->flags |= REQ_F_LINK_TIMEOUT; |
76a46e06 | 6171 | return nxt; |
2665abfd JA |
6172 | } |
6173 | ||
f13fad7b JA |
6174 | static void __io_queue_sqe(struct io_kiocb *req, const struct io_uring_sqe *sqe, |
6175 | struct io_comp_state *cs) | |
2b188cc1 | 6176 | { |
4a0a7a18 | 6177 | struct io_kiocb *linked_timeout; |
4bc4494e | 6178 | struct io_kiocb *nxt; |
193155c8 | 6179 | const struct cred *old_creds = NULL; |
e0c5c576 | 6180 | int ret; |
2b188cc1 | 6181 | |
4a0a7a18 JA |
6182 | again: |
6183 | linked_timeout = io_prep_linked_timeout(req); | |
6184 | ||
7cdaf587 XW |
6185 | if ((req->flags & REQ_F_WORK_INITIALIZED) && req->work.creds && |
6186 | req->work.creds != current_cred()) { | |
193155c8 JA |
6187 | if (old_creds) |
6188 | revert_creds(old_creds); | |
6189 | if (old_creds == req->work.creds) | |
6190 | old_creds = NULL; /* restored original creds */ | |
6191 | else | |
6192 | old_creds = override_creds(req->work.creds); | |
6193 | } | |
6194 | ||
f13fad7b | 6195 | ret = io_issue_sqe(req, sqe, true, cs); |
491381ce JA |
6196 | |
6197 | /* | |
6198 | * We async punt it if the file wasn't marked NOWAIT, or if the file | |
6199 | * doesn't support non-blocking read/write attempts | |
6200 | */ | |
24c74678 | 6201 | if (ret == -EAGAIN && !(req->flags & REQ_F_NOWAIT)) { |
f063c547 | 6202 | if (!io_arm_poll_handler(req)) { |
86a761f8 | 6203 | punt: |
f063c547 PB |
6204 | ret = io_prep_work_files(req); |
6205 | if (unlikely(ret)) | |
bbad27b2 | 6206 | goto err; |
f063c547 PB |
6207 | /* |
6208 | * Queued up for async execution, worker will release | |
6209 | * submit reference when the iocb is actually submitted. | |
6210 | */ | |
6211 | io_queue_async_work(req); | |
2b188cc1 | 6212 | } |
bbad27b2 | 6213 | |
f063c547 PB |
6214 | if (linked_timeout) |
6215 | io_queue_linked_timeout(linked_timeout); | |
4bc4494e | 6216 | goto exit; |
2b188cc1 | 6217 | } |
e65ef56d | 6218 | |
652532ad | 6219 | if (unlikely(ret)) { |
fcb323cc | 6220 | err: |
652532ad PB |
6221 | /* un-prep timeout, so it'll be killed as any other linked */ |
6222 | req->flags &= ~REQ_F_LINK_TIMEOUT; | |
4e88d6e7 | 6223 | req_set_fail_links(req); |
e65ef56d | 6224 | io_put_req(req); |
e1e16097 | 6225 | io_req_complete(req, ret); |
652532ad | 6226 | goto exit; |
9e645e11 | 6227 | } |
652532ad PB |
6228 | |
6229 | /* drop submission reference */ | |
6230 | nxt = io_put_req_find_next(req); | |
6231 | if (linked_timeout) | |
6232 | io_queue_linked_timeout(linked_timeout); | |
6233 | ||
4a0a7a18 JA |
6234 | if (nxt) { |
6235 | req = nxt; | |
86a761f8 PB |
6236 | |
6237 | if (req->flags & REQ_F_FORCE_ASYNC) | |
6238 | goto punt; | |
4a0a7a18 JA |
6239 | goto again; |
6240 | } | |
4bc4494e | 6241 | exit: |
193155c8 JA |
6242 | if (old_creds) |
6243 | revert_creds(old_creds); | |
2b188cc1 JA |
6244 | } |
6245 | ||
f13fad7b JA |
6246 | static void io_queue_sqe(struct io_kiocb *req, const struct io_uring_sqe *sqe, |
6247 | struct io_comp_state *cs) | |
4fe2c963 JL |
6248 | { |
6249 | int ret; | |
6250 | ||
3529d8c2 | 6251 | ret = io_req_defer(req, sqe); |
4fe2c963 JL |
6252 | if (ret) { |
6253 | if (ret != -EIOCBQUEUED) { | |
1118591a | 6254 | fail_req: |
4e88d6e7 | 6255 | req_set_fail_links(req); |
e1e16097 JA |
6256 | io_put_req(req); |
6257 | io_req_complete(req, ret); | |
4fe2c963 | 6258 | } |
2550878f | 6259 | } else if (req->flags & REQ_F_FORCE_ASYNC) { |
bd2ab18a | 6260 | if (!req->io) { |
bd2ab18a | 6261 | ret = io_req_defer_prep(req, sqe); |
327d6d96 | 6262 | if (unlikely(ret)) |
bd2ab18a PB |
6263 | goto fail_req; |
6264 | } | |
6265 | ||
ce35a47a JA |
6266 | /* |
6267 | * Never try inline submit of IOSQE_ASYNC is set, go straight | |
6268 | * to async execution. | |
6269 | */ | |
3e863ea3 | 6270 | io_req_init_async(req); |
ce35a47a JA |
6271 | req->work.flags |= IO_WQ_WORK_CONCURRENT; |
6272 | io_queue_async_work(req); | |
6273 | } else { | |
f13fad7b | 6274 | __io_queue_sqe(req, sqe, cs); |
ce35a47a | 6275 | } |
4fe2c963 JL |
6276 | } |
6277 | ||
f13fad7b JA |
6278 | static inline void io_queue_link_head(struct io_kiocb *req, |
6279 | struct io_comp_state *cs) | |
4fe2c963 | 6280 | { |
94ae5e77 | 6281 | if (unlikely(req->flags & REQ_F_FAIL_LINK)) { |
e1e16097 JA |
6282 | io_put_req(req); |
6283 | io_req_complete(req, -ECANCELED); | |
1b4a51b6 | 6284 | } else |
f13fad7b | 6285 | io_queue_sqe(req, NULL, cs); |
4fe2c963 JL |
6286 | } |
6287 | ||
1d4240cc | 6288 | static int io_submit_sqe(struct io_kiocb *req, const struct io_uring_sqe *sqe, |
f13fad7b | 6289 | struct io_kiocb **link, struct io_comp_state *cs) |
9e645e11 | 6290 | { |
a197f664 | 6291 | struct io_ring_ctx *ctx = req->ctx; |
ef4ff581 | 6292 | int ret; |
9e645e11 | 6293 | |
9e645e11 JA |
6294 | /* |
6295 | * If we already have a head request, queue this one for async | |
6296 | * submittal once the head completes. If we don't have a head but | |
6297 | * IOSQE_IO_LINK is set in the sqe, start a new head. This one will be | |
6298 | * submitted sync once the chain is complete. If none of those | |
6299 | * conditions are true (normal request), then just queue it. | |
6300 | */ | |
6301 | if (*link) { | |
9d76377f | 6302 | struct io_kiocb *head = *link; |
4e88d6e7 | 6303 | |
8cdf2193 PB |
6304 | /* |
6305 | * Taking sequential execution of a link, draining both sides | |
6306 | * of the link also fullfils IOSQE_IO_DRAIN semantics for all | |
6307 | * requests in the link. So, it drains the head and the | |
6308 | * next after the link request. The last one is done via | |
6309 | * drain_next flag to persist the effect across calls. | |
6310 | */ | |
ef4ff581 | 6311 | if (req->flags & REQ_F_IO_DRAIN) { |
711be031 PB |
6312 | head->flags |= REQ_F_IO_DRAIN; |
6313 | ctx->drain_next = 1; | |
6314 | } | |
3529d8c2 | 6315 | ret = io_req_defer_prep(req, sqe); |
327d6d96 | 6316 | if (unlikely(ret)) { |
4e88d6e7 | 6317 | /* fail even hard links since we don't submit */ |
9d76377f | 6318 | head->flags |= REQ_F_FAIL_LINK; |
1d4240cc | 6319 | return ret; |
2d28390a | 6320 | } |
9d76377f PB |
6321 | trace_io_uring_link(ctx, req, head); |
6322 | list_add_tail(&req->link_list, &head->link_list); | |
32fe525b PB |
6323 | |
6324 | /* last request of a link, enqueue the link */ | |
ef4ff581 | 6325 | if (!(req->flags & (REQ_F_LINK | REQ_F_HARDLINK))) { |
f13fad7b | 6326 | io_queue_link_head(head, cs); |
32fe525b PB |
6327 | *link = NULL; |
6328 | } | |
9e645e11 | 6329 | } else { |
711be031 PB |
6330 | if (unlikely(ctx->drain_next)) { |
6331 | req->flags |= REQ_F_IO_DRAIN; | |
ef4ff581 | 6332 | ctx->drain_next = 0; |
711be031 | 6333 | } |
ef4ff581 | 6334 | if (req->flags & (REQ_F_LINK | REQ_F_HARDLINK)) { |
dea3b49c | 6335 | req->flags |= REQ_F_LINK_HEAD; |
711be031 | 6336 | INIT_LIST_HEAD(&req->link_list); |
f1d96a8f | 6337 | |
711be031 | 6338 | ret = io_req_defer_prep(req, sqe); |
327d6d96 | 6339 | if (unlikely(ret)) |
711be031 PB |
6340 | req->flags |= REQ_F_FAIL_LINK; |
6341 | *link = req; | |
6342 | } else { | |
f13fad7b | 6343 | io_queue_sqe(req, sqe, cs); |
711be031 | 6344 | } |
9e645e11 | 6345 | } |
2e6e1fde | 6346 | |
1d4240cc | 6347 | return 0; |
9e645e11 JA |
6348 | } |
6349 | ||
9a56a232 JA |
6350 | /* |
6351 | * Batched submission is done, ensure local IO is flushed out. | |
6352 | */ | |
6353 | static void io_submit_state_end(struct io_submit_state *state) | |
6354 | { | |
f13fad7b JA |
6355 | if (!list_empty(&state->comp.list)) |
6356 | io_submit_flush_completions(&state->comp); | |
9a56a232 | 6357 | blk_finish_plug(&state->plug); |
9f13c35b | 6358 | io_state_file_put(state); |
2579f913 | 6359 | if (state->free_reqs) |
6c8a3134 | 6360 | kmem_cache_free_bulk(req_cachep, state->free_reqs, state->reqs); |
9a56a232 JA |
6361 | } |
6362 | ||
6363 | /* | |
6364 | * Start submission side cache. | |
6365 | */ | |
6366 | static void io_submit_state_start(struct io_submit_state *state, | |
013538bd | 6367 | struct io_ring_ctx *ctx, unsigned int max_ios) |
9a56a232 JA |
6368 | { |
6369 | blk_start_plug(&state->plug); | |
013538bd JA |
6370 | state->comp.nr = 0; |
6371 | INIT_LIST_HEAD(&state->comp.list); | |
6372 | state->comp.ctx = ctx; | |
2579f913 | 6373 | state->free_reqs = 0; |
9a56a232 JA |
6374 | state->file = NULL; |
6375 | state->ios_left = max_ios; | |
6376 | } | |
6377 | ||
2b188cc1 JA |
6378 | static void io_commit_sqring(struct io_ring_ctx *ctx) |
6379 | { | |
75b28aff | 6380 | struct io_rings *rings = ctx->rings; |
2b188cc1 | 6381 | |
caf582c6 PB |
6382 | /* |
6383 | * Ensure any loads from the SQEs are done at this point, | |
6384 | * since once we write the new head, the application could | |
6385 | * write new data to them. | |
6386 | */ | |
6387 | smp_store_release(&rings->sq.head, ctx->cached_sq_head); | |
2b188cc1 JA |
6388 | } |
6389 | ||
2b188cc1 | 6390 | /* |
3529d8c2 | 6391 | * Fetch an sqe, if one is available. Note that sqe_ptr will point to memory |
2b188cc1 JA |
6392 | * that is mapped by userspace. This means that care needs to be taken to |
6393 | * ensure that reads are stable, as we cannot rely on userspace always | |
6394 | * being a good citizen. If members of the sqe are validated and then later | |
6395 | * used, it's important that those reads are done through READ_ONCE() to | |
6396 | * prevent a re-load down the line. | |
6397 | */ | |
709b302f | 6398 | static const struct io_uring_sqe *io_get_sqe(struct io_ring_ctx *ctx) |
2b188cc1 | 6399 | { |
75b28aff | 6400 | u32 *sq_array = ctx->sq_array; |
2b188cc1 JA |
6401 | unsigned head; |
6402 | ||
6403 | /* | |
6404 | * The cached sq head (or cq tail) serves two purposes: | |
6405 | * | |
6406 | * 1) allows us to batch the cost of updating the user visible | |
6407 | * head updates. | |
6408 | * 2) allows the kernel side to track the head on its own, even | |
6409 | * though the application is the one updating it. | |
6410 | */ | |
ee7d46d9 | 6411 | head = READ_ONCE(sq_array[ctx->cached_sq_head & ctx->sq_mask]); |
709b302f PB |
6412 | if (likely(head < ctx->sq_entries)) |
6413 | return &ctx->sq_sqes[head]; | |
2b188cc1 JA |
6414 | |
6415 | /* drop invalid entries */ | |
498ccd9e | 6416 | ctx->cached_sq_dropped++; |
ee7d46d9 | 6417 | WRITE_ONCE(ctx->rings->sq_dropped, ctx->cached_sq_dropped); |
709b302f PB |
6418 | return NULL; |
6419 | } | |
6420 | ||
6421 | static inline void io_consume_sqe(struct io_ring_ctx *ctx) | |
6422 | { | |
6423 | ctx->cached_sq_head++; | |
2b188cc1 JA |
6424 | } |
6425 | ||
ef4ff581 PB |
6426 | #define SQE_VALID_FLAGS (IOSQE_FIXED_FILE|IOSQE_IO_DRAIN|IOSQE_IO_LINK| \ |
6427 | IOSQE_IO_HARDLINK | IOSQE_ASYNC | \ | |
6428 | IOSQE_BUFFER_SELECT) | |
6429 | ||
6430 | static int io_init_req(struct io_ring_ctx *ctx, struct io_kiocb *req, | |
6431 | const struct io_uring_sqe *sqe, | |
0cdaf760 | 6432 | struct io_submit_state *state) |
0553b8bd | 6433 | { |
ef4ff581 | 6434 | unsigned int sqe_flags; |
63ff8223 | 6435 | int id; |
ef4ff581 | 6436 | |
0553b8bd PB |
6437 | req->opcode = READ_ONCE(sqe->opcode); |
6438 | req->user_data = READ_ONCE(sqe->user_data); | |
6439 | req->io = NULL; | |
6440 | req->file = NULL; | |
6441 | req->ctx = ctx; | |
6442 | req->flags = 0; | |
6443 | /* one is dropped after submission, the other at completion */ | |
6444 | refcount_set(&req->refs, 2); | |
4dd2824d | 6445 | req->task = current; |
e3bc8e9d | 6446 | get_task_struct(req->task); |
0553b8bd | 6447 | req->result = 0; |
ef4ff581 PB |
6448 | |
6449 | if (unlikely(req->opcode >= IORING_OP_LAST)) | |
6450 | return -EINVAL; | |
6451 | ||
9d8426a0 JA |
6452 | if (unlikely(io_sq_thread_acquire_mm(ctx, req))) |
6453 | return -EFAULT; | |
ef4ff581 PB |
6454 | |
6455 | sqe_flags = READ_ONCE(sqe->flags); | |
6456 | /* enforce forwards compatibility on users */ | |
6457 | if (unlikely(sqe_flags & ~SQE_VALID_FLAGS)) | |
6458 | return -EINVAL; | |
6459 | ||
6460 | if ((sqe_flags & IOSQE_BUFFER_SELECT) && | |
6461 | !io_op_defs[req->opcode].buffer_select) | |
6462 | return -EOPNOTSUPP; | |
6463 | ||
6464 | id = READ_ONCE(sqe->personality); | |
6465 | if (id) { | |
7cdaf587 | 6466 | io_req_init_async(req); |
ef4ff581 PB |
6467 | req->work.creds = idr_find(&ctx->personality_idr, id); |
6468 | if (unlikely(!req->work.creds)) | |
6469 | return -EINVAL; | |
6470 | get_cred(req->work.creds); | |
6471 | } | |
6472 | ||
6473 | /* same numerical values with corresponding REQ_F_*, safe to copy */ | |
c11368a5 | 6474 | req->flags |= sqe_flags; |
ef4ff581 | 6475 | |
63ff8223 JA |
6476 | if (!io_op_defs[req->opcode].needs_file) |
6477 | return 0; | |
6478 | ||
6479 | return io_req_set_file(state, req, READ_ONCE(sqe->fd)); | |
0553b8bd PB |
6480 | } |
6481 | ||
fb5ccc98 | 6482 | static int io_submit_sqes(struct io_ring_ctx *ctx, unsigned int nr, |
0cdaf760 | 6483 | struct file *ring_file, int ring_fd) |
6c271ce2 | 6484 | { |
ac8691c4 | 6485 | struct io_submit_state state; |
9e645e11 | 6486 | struct io_kiocb *link = NULL; |
9e645e11 | 6487 | int i, submitted = 0; |
6c271ce2 | 6488 | |
c4a2ed72 | 6489 | /* if we have a backlog and couldn't flush it all, return BUSY */ |
ad3eb2c8 JA |
6490 | if (test_bit(0, &ctx->sq_check_overflow)) { |
6491 | if (!list_empty(&ctx->cq_overflow_list) && | |
6492 | !io_cqring_overflow_flush(ctx, false)) | |
6493 | return -EBUSY; | |
6494 | } | |
6c271ce2 | 6495 | |
ee7d46d9 PB |
6496 | /* make sure SQ entry isn't read before tail */ |
6497 | nr = min3(nr, ctx->sq_entries, io_sqring_entries(ctx)); | |
9ef4f124 | 6498 | |
2b85edfc PB |
6499 | if (!percpu_ref_tryget_many(&ctx->refs, nr)) |
6500 | return -EAGAIN; | |
6c271ce2 | 6501 | |
013538bd | 6502 | io_submit_state_start(&state, ctx, nr); |
6c271ce2 | 6503 | |
b14cca0c PB |
6504 | ctx->ring_fd = ring_fd; |
6505 | ctx->ring_file = ring_file; | |
6506 | ||
6c271ce2 | 6507 | for (i = 0; i < nr; i++) { |
3529d8c2 | 6508 | const struct io_uring_sqe *sqe; |
196be95c | 6509 | struct io_kiocb *req; |
1cb1edb2 | 6510 | int err; |
fb5ccc98 | 6511 | |
b1e50e54 PB |
6512 | sqe = io_get_sqe(ctx); |
6513 | if (unlikely(!sqe)) { | |
6514 | io_consume_sqe(ctx); | |
6515 | break; | |
6516 | } | |
ac8691c4 | 6517 | req = io_alloc_req(ctx, &state); |
196be95c PB |
6518 | if (unlikely(!req)) { |
6519 | if (!submitted) | |
6520 | submitted = -EAGAIN; | |
fb5ccc98 | 6521 | break; |
196be95c | 6522 | } |
fb5ccc98 | 6523 | |
ac8691c4 | 6524 | err = io_init_req(ctx, req, sqe, &state); |
709b302f | 6525 | io_consume_sqe(ctx); |
d3656344 JA |
6526 | /* will complete beyond this point, count as submitted */ |
6527 | submitted++; | |
6528 | ||
ef4ff581 | 6529 | if (unlikely(err)) { |
1cb1edb2 | 6530 | fail_req: |
e1e16097 JA |
6531 | io_put_req(req); |
6532 | io_req_complete(req, err); | |
196be95c PB |
6533 | break; |
6534 | } | |
fb5ccc98 | 6535 | |
354420f7 | 6536 | trace_io_uring_submit_sqe(ctx, req->opcode, req->user_data, |
0cdaf760 | 6537 | true, io_async_submit(ctx)); |
f13fad7b | 6538 | err = io_submit_sqe(req, sqe, &link, &state.comp); |
1d4240cc PB |
6539 | if (err) |
6540 | goto fail_req; | |
6c271ce2 JA |
6541 | } |
6542 | ||
9466f437 PB |
6543 | if (unlikely(submitted != nr)) { |
6544 | int ref_used = (submitted == -EAGAIN) ? 0 : submitted; | |
6545 | ||
6546 | percpu_ref_put_many(&ctx->refs, nr - ref_used); | |
6547 | } | |
9e645e11 | 6548 | if (link) |
f13fad7b | 6549 | io_queue_link_head(link, &state.comp); |
ac8691c4 | 6550 | io_submit_state_end(&state); |
6c271ce2 | 6551 | |
ae9428ca PB |
6552 | /* Commit SQ ring head once we've consumed and submitted all SQEs */ |
6553 | io_commit_sqring(ctx); | |
6554 | ||
6c271ce2 JA |
6555 | return submitted; |
6556 | } | |
6557 | ||
23b3628e XW |
6558 | static inline void io_ring_set_wakeup_flag(struct io_ring_ctx *ctx) |
6559 | { | |
6560 | /* Tell userspace we may need a wakeup call */ | |
6561 | spin_lock_irq(&ctx->completion_lock); | |
6562 | ctx->rings->sq_flags |= IORING_SQ_NEED_WAKEUP; | |
6563 | spin_unlock_irq(&ctx->completion_lock); | |
6564 | } | |
6565 | ||
6566 | static inline void io_ring_clear_wakeup_flag(struct io_ring_ctx *ctx) | |
6567 | { | |
6568 | spin_lock_irq(&ctx->completion_lock); | |
6569 | ctx->rings->sq_flags &= ~IORING_SQ_NEED_WAKEUP; | |
6570 | spin_unlock_irq(&ctx->completion_lock); | |
6571 | } | |
6572 | ||
6c271ce2 JA |
6573 | static int io_sq_thread(void *data) |
6574 | { | |
6c271ce2 | 6575 | struct io_ring_ctx *ctx = data; |
181e448d | 6576 | const struct cred *old_cred; |
6c271ce2 | 6577 | DEFINE_WAIT(wait); |
6c271ce2 | 6578 | unsigned long timeout; |
bdcd3eab | 6579 | int ret = 0; |
6c271ce2 | 6580 | |
0f158b4c | 6581 | complete(&ctx->sq_thread_comp); |
a4c0b3de | 6582 | |
181e448d | 6583 | old_cred = override_creds(ctx->creds); |
6c271ce2 | 6584 | |
bdcd3eab | 6585 | timeout = jiffies + ctx->sq_thread_idle; |
2bbcd6d3 | 6586 | while (!kthread_should_park()) { |
fb5ccc98 | 6587 | unsigned int to_submit; |
6c271ce2 | 6588 | |
540e32a0 | 6589 | if (!list_empty(&ctx->iopoll_list)) { |
6c271ce2 JA |
6590 | unsigned nr_events = 0; |
6591 | ||
bdcd3eab | 6592 | mutex_lock(&ctx->uring_lock); |
540e32a0 | 6593 | if (!list_empty(&ctx->iopoll_list) && !need_resched()) |
9dedd563 | 6594 | io_do_iopoll(ctx, &nr_events, 0); |
bdcd3eab | 6595 | else |
6c271ce2 | 6596 | timeout = jiffies + ctx->sq_thread_idle; |
bdcd3eab | 6597 | mutex_unlock(&ctx->uring_lock); |
6c271ce2 JA |
6598 | } |
6599 | ||
fb5ccc98 | 6600 | to_submit = io_sqring_entries(ctx); |
c1edbf5f JA |
6601 | |
6602 | /* | |
6603 | * If submit got -EBUSY, flag us as needing the application | |
6604 | * to enter the kernel to reap and flush events. | |
6605 | */ | |
b772f07a | 6606 | if (!to_submit || ret == -EBUSY || need_resched()) { |
7143b5ac SG |
6607 | /* |
6608 | * Drop cur_mm before scheduling, we can't hold it for | |
6609 | * long periods (or over schedule()). Do this before | |
6610 | * adding ourselves to the waitqueue, as the unuse/drop | |
6611 | * may sleep. | |
6612 | */ | |
4349f30e | 6613 | io_sq_thread_drop_mm(); |
7143b5ac | 6614 | |
6c271ce2 JA |
6615 | /* |
6616 | * We're polling. If we're within the defined idle | |
6617 | * period, then let us spin without work before going | |
c1edbf5f JA |
6618 | * to sleep. The exception is if we got EBUSY doing |
6619 | * more IO, we should wait for the application to | |
6620 | * reap events and wake us up. | |
6c271ce2 | 6621 | */ |
540e32a0 | 6622 | if (!list_empty(&ctx->iopoll_list) || need_resched() || |
df069d80 JA |
6623 | (!time_after(jiffies, timeout) && ret != -EBUSY && |
6624 | !percpu_ref_is_dying(&ctx->refs))) { | |
4c6e277c | 6625 | io_run_task_work(); |
9831a90c | 6626 | cond_resched(); |
6c271ce2 JA |
6627 | continue; |
6628 | } | |
6629 | ||
6c271ce2 JA |
6630 | prepare_to_wait(&ctx->sqo_wait, &wait, |
6631 | TASK_INTERRUPTIBLE); | |
6632 | ||
bdcd3eab XW |
6633 | /* |
6634 | * While doing polled IO, before going to sleep, we need | |
540e32a0 PB |
6635 | * to check if there are new reqs added to iopoll_list, |
6636 | * it is because reqs may have been punted to io worker | |
6637 | * and will be added to iopoll_list later, hence check | |
6638 | * the iopoll_list again. | |
bdcd3eab XW |
6639 | */ |
6640 | if ((ctx->flags & IORING_SETUP_IOPOLL) && | |
540e32a0 | 6641 | !list_empty_careful(&ctx->iopoll_list)) { |
bdcd3eab XW |
6642 | finish_wait(&ctx->sqo_wait, &wait); |
6643 | continue; | |
6644 | } | |
6645 | ||
23b3628e | 6646 | io_ring_set_wakeup_flag(ctx); |
6c271ce2 | 6647 | |
fb5ccc98 | 6648 | to_submit = io_sqring_entries(ctx); |
c1edbf5f | 6649 | if (!to_submit || ret == -EBUSY) { |
2bbcd6d3 | 6650 | if (kthread_should_park()) { |
6c271ce2 JA |
6651 | finish_wait(&ctx->sqo_wait, &wait); |
6652 | break; | |
6653 | } | |
4c6e277c | 6654 | if (io_run_task_work()) { |
10bea96d | 6655 | finish_wait(&ctx->sqo_wait, &wait); |
23b3628e | 6656 | io_ring_clear_wakeup_flag(ctx); |
b41e9852 JA |
6657 | continue; |
6658 | } | |
6c271ce2 JA |
6659 | if (signal_pending(current)) |
6660 | flush_signals(current); | |
6661 | schedule(); | |
6662 | finish_wait(&ctx->sqo_wait, &wait); | |
6663 | ||
23b3628e | 6664 | io_ring_clear_wakeup_flag(ctx); |
d4ae271d | 6665 | ret = 0; |
6c271ce2 JA |
6666 | continue; |
6667 | } | |
6668 | finish_wait(&ctx->sqo_wait, &wait); | |
6669 | ||
23b3628e | 6670 | io_ring_clear_wakeup_flag(ctx); |
6c271ce2 JA |
6671 | } |
6672 | ||
8a4955ff | 6673 | mutex_lock(&ctx->uring_lock); |
6b668c9b XW |
6674 | if (likely(!percpu_ref_is_dying(&ctx->refs))) |
6675 | ret = io_submit_sqes(ctx, to_submit, NULL, -1); | |
8a4955ff | 6676 | mutex_unlock(&ctx->uring_lock); |
bdcd3eab | 6677 | timeout = jiffies + ctx->sq_thread_idle; |
6c271ce2 JA |
6678 | } |
6679 | ||
4c6e277c | 6680 | io_run_task_work(); |
b41e9852 | 6681 | |
4349f30e | 6682 | io_sq_thread_drop_mm(); |
181e448d | 6683 | revert_creds(old_cred); |
06058632 | 6684 | |
2bbcd6d3 | 6685 | kthread_parkme(); |
06058632 | 6686 | |
6c271ce2 JA |
6687 | return 0; |
6688 | } | |
6689 | ||
bda52162 JA |
6690 | struct io_wait_queue { |
6691 | struct wait_queue_entry wq; | |
6692 | struct io_ring_ctx *ctx; | |
6693 | unsigned to_wait; | |
6694 | unsigned nr_timeouts; | |
6695 | }; | |
6696 | ||
1d7bb1d5 | 6697 | static inline bool io_should_wake(struct io_wait_queue *iowq, bool noflush) |
bda52162 JA |
6698 | { |
6699 | struct io_ring_ctx *ctx = iowq->ctx; | |
6700 | ||
6701 | /* | |
d195a66e | 6702 | * Wake up if we have enough events, or if a timeout occurred since we |
bda52162 JA |
6703 | * started waiting. For timeouts, we always want to return to userspace, |
6704 | * regardless of event count. | |
6705 | */ | |
1d7bb1d5 | 6706 | return io_cqring_events(ctx, noflush) >= iowq->to_wait || |
bda52162 JA |
6707 | atomic_read(&ctx->cq_timeouts) != iowq->nr_timeouts; |
6708 | } | |
6709 | ||
6710 | static int io_wake_function(struct wait_queue_entry *curr, unsigned int mode, | |
6711 | int wake_flags, void *key) | |
6712 | { | |
6713 | struct io_wait_queue *iowq = container_of(curr, struct io_wait_queue, | |
6714 | wq); | |
6715 | ||
1d7bb1d5 JA |
6716 | /* use noflush == true, as we can't safely rely on locking context */ |
6717 | if (!io_should_wake(iowq, true)) | |
bda52162 JA |
6718 | return -1; |
6719 | ||
6720 | return autoremove_wake_function(curr, mode, wake_flags, key); | |
6721 | } | |
6722 | ||
2b188cc1 JA |
6723 | /* |
6724 | * Wait until events become available, if we don't already have some. The | |
6725 | * application must reap them itself, as they reside on the shared cq ring. | |
6726 | */ | |
6727 | static int io_cqring_wait(struct io_ring_ctx *ctx, int min_events, | |
6728 | const sigset_t __user *sig, size_t sigsz) | |
6729 | { | |
bda52162 JA |
6730 | struct io_wait_queue iowq = { |
6731 | .wq = { | |
6732 | .private = current, | |
6733 | .func = io_wake_function, | |
6734 | .entry = LIST_HEAD_INIT(iowq.wq.entry), | |
6735 | }, | |
6736 | .ctx = ctx, | |
6737 | .to_wait = min_events, | |
6738 | }; | |
75b28aff | 6739 | struct io_rings *rings = ctx->rings; |
e9ffa5c2 | 6740 | int ret = 0; |
2b188cc1 | 6741 | |
b41e9852 JA |
6742 | do { |
6743 | if (io_cqring_events(ctx, false) >= min_events) | |
6744 | return 0; | |
4c6e277c | 6745 | if (!io_run_task_work()) |
b41e9852 | 6746 | break; |
b41e9852 | 6747 | } while (1); |
2b188cc1 JA |
6748 | |
6749 | if (sig) { | |
9e75ad5d AB |
6750 | #ifdef CONFIG_COMPAT |
6751 | if (in_compat_syscall()) | |
6752 | ret = set_compat_user_sigmask((const compat_sigset_t __user *)sig, | |
b772434b | 6753 | sigsz); |
9e75ad5d AB |
6754 | else |
6755 | #endif | |
b772434b | 6756 | ret = set_user_sigmask(sig, sigsz); |
9e75ad5d | 6757 | |
2b188cc1 JA |
6758 | if (ret) |
6759 | return ret; | |
6760 | } | |
6761 | ||
bda52162 | 6762 | iowq.nr_timeouts = atomic_read(&ctx->cq_timeouts); |
c826bd7a | 6763 | trace_io_uring_cqring_wait(ctx, min_events); |
bda52162 JA |
6764 | do { |
6765 | prepare_to_wait_exclusive(&ctx->wait, &iowq.wq, | |
6766 | TASK_INTERRUPTIBLE); | |
ce593a6c | 6767 | /* make sure we run task_work before checking for signals */ |
4c6e277c JA |
6768 | if (io_run_task_work()) |
6769 | continue; | |
bda52162 | 6770 | if (signal_pending(current)) { |
b7db41c9 JA |
6771 | if (current->jobctl & JOBCTL_TASK_WORK) { |
6772 | spin_lock_irq(¤t->sighand->siglock); | |
6773 | current->jobctl &= ~JOBCTL_TASK_WORK; | |
6774 | recalc_sigpending(); | |
6775 | spin_unlock_irq(¤t->sighand->siglock); | |
6776 | continue; | |
6777 | } | |
6778 | ret = -EINTR; | |
bda52162 JA |
6779 | break; |
6780 | } | |
ce593a6c JA |
6781 | if (io_should_wake(&iowq, false)) |
6782 | break; | |
6783 | schedule(); | |
bda52162 JA |
6784 | } while (1); |
6785 | finish_wait(&ctx->wait, &iowq.wq); | |
6786 | ||
b7db41c9 | 6787 | restore_saved_sigmask_unless(ret == -EINTR); |
2b188cc1 | 6788 | |
75b28aff | 6789 | return READ_ONCE(rings->cq.head) == READ_ONCE(rings->cq.tail) ? ret : 0; |
2b188cc1 JA |
6790 | } |
6791 | ||
6b06314c JA |
6792 | static void __io_sqe_files_unregister(struct io_ring_ctx *ctx) |
6793 | { | |
6794 | #if defined(CONFIG_UNIX) | |
6795 | if (ctx->ring_sock) { | |
6796 | struct sock *sock = ctx->ring_sock->sk; | |
6797 | struct sk_buff *skb; | |
6798 | ||
6799 | while ((skb = skb_dequeue(&sock->sk_receive_queue)) != NULL) | |
6800 | kfree_skb(skb); | |
6801 | } | |
6802 | #else | |
6803 | int i; | |
6804 | ||
65e19f54 JA |
6805 | for (i = 0; i < ctx->nr_user_files; i++) { |
6806 | struct file *file; | |
6807 | ||
6808 | file = io_file_from_index(ctx, i); | |
6809 | if (file) | |
6810 | fput(file); | |
6811 | } | |
6b06314c JA |
6812 | #endif |
6813 | } | |
6814 | ||
05f3fb3c JA |
6815 | static void io_file_ref_kill(struct percpu_ref *ref) |
6816 | { | |
6817 | struct fixed_file_data *data; | |
6818 | ||
6819 | data = container_of(ref, struct fixed_file_data, refs); | |
6820 | complete(&data->done); | |
6821 | } | |
6822 | ||
6b06314c JA |
6823 | static int io_sqe_files_unregister(struct io_ring_ctx *ctx) |
6824 | { | |
05f3fb3c | 6825 | struct fixed_file_data *data = ctx->file_data; |
05589553 | 6826 | struct fixed_file_ref_node *ref_node = NULL; |
65e19f54 JA |
6827 | unsigned nr_tables, i; |
6828 | ||
05f3fb3c | 6829 | if (!data) |
6b06314c JA |
6830 | return -ENXIO; |
6831 | ||
6a4d07cd | 6832 | spin_lock(&data->lock); |
05589553 XW |
6833 | if (!list_empty(&data->ref_list)) |
6834 | ref_node = list_first_entry(&data->ref_list, | |
6835 | struct fixed_file_ref_node, node); | |
6a4d07cd | 6836 | spin_unlock(&data->lock); |
05589553 XW |
6837 | if (ref_node) |
6838 | percpu_ref_kill(&ref_node->refs); | |
6839 | ||
6840 | percpu_ref_kill(&data->refs); | |
6841 | ||
6842 | /* wait for all refs nodes to complete */ | |
4a38aed2 | 6843 | flush_delayed_work(&ctx->file_put_work); |
2faf852d | 6844 | wait_for_completion(&data->done); |
05f3fb3c | 6845 | |
6b06314c | 6846 | __io_sqe_files_unregister(ctx); |
65e19f54 JA |
6847 | nr_tables = DIV_ROUND_UP(ctx->nr_user_files, IORING_MAX_FILES_TABLE); |
6848 | for (i = 0; i < nr_tables; i++) | |
05f3fb3c JA |
6849 | kfree(data->table[i].files); |
6850 | kfree(data->table); | |
05589553 XW |
6851 | percpu_ref_exit(&data->refs); |
6852 | kfree(data); | |
05f3fb3c | 6853 | ctx->file_data = NULL; |
6b06314c JA |
6854 | ctx->nr_user_files = 0; |
6855 | return 0; | |
6856 | } | |
6857 | ||
6c271ce2 JA |
6858 | static void io_sq_thread_stop(struct io_ring_ctx *ctx) |
6859 | { | |
6860 | if (ctx->sqo_thread) { | |
0f158b4c | 6861 | wait_for_completion(&ctx->sq_thread_comp); |
2bbcd6d3 RP |
6862 | /* |
6863 | * The park is a bit of a work-around, without it we get | |
6864 | * warning spews on shutdown with SQPOLL set and affinity | |
6865 | * set to a single CPU. | |
6866 | */ | |
06058632 | 6867 | kthread_park(ctx->sqo_thread); |
6c271ce2 JA |
6868 | kthread_stop(ctx->sqo_thread); |
6869 | ctx->sqo_thread = NULL; | |
6870 | } | |
6871 | } | |
6872 | ||
6b06314c JA |
6873 | static void io_finish_async(struct io_ring_ctx *ctx) |
6874 | { | |
6c271ce2 JA |
6875 | io_sq_thread_stop(ctx); |
6876 | ||
561fb04a JA |
6877 | if (ctx->io_wq) { |
6878 | io_wq_destroy(ctx->io_wq); | |
6879 | ctx->io_wq = NULL; | |
6b06314c JA |
6880 | } |
6881 | } | |
6882 | ||
6883 | #if defined(CONFIG_UNIX) | |
6b06314c JA |
6884 | /* |
6885 | * Ensure the UNIX gc is aware of our file set, so we are certain that | |
6886 | * the io_uring can be safely unregistered on process exit, even if we have | |
6887 | * loops in the file referencing. | |
6888 | */ | |
6889 | static int __io_sqe_files_scm(struct io_ring_ctx *ctx, int nr, int offset) | |
6890 | { | |
6891 | struct sock *sk = ctx->ring_sock->sk; | |
6892 | struct scm_fp_list *fpl; | |
6893 | struct sk_buff *skb; | |
08a45173 | 6894 | int i, nr_files; |
6b06314c | 6895 | |
6b06314c JA |
6896 | fpl = kzalloc(sizeof(*fpl), GFP_KERNEL); |
6897 | if (!fpl) | |
6898 | return -ENOMEM; | |
6899 | ||
6900 | skb = alloc_skb(0, GFP_KERNEL); | |
6901 | if (!skb) { | |
6902 | kfree(fpl); | |
6903 | return -ENOMEM; | |
6904 | } | |
6905 | ||
6906 | skb->sk = sk; | |
6b06314c | 6907 | |
08a45173 | 6908 | nr_files = 0; |
6b06314c JA |
6909 | fpl->user = get_uid(ctx->user); |
6910 | for (i = 0; i < nr; i++) { | |
65e19f54 JA |
6911 | struct file *file = io_file_from_index(ctx, i + offset); |
6912 | ||
6913 | if (!file) | |
08a45173 | 6914 | continue; |
65e19f54 | 6915 | fpl->fp[nr_files] = get_file(file); |
08a45173 JA |
6916 | unix_inflight(fpl->user, fpl->fp[nr_files]); |
6917 | nr_files++; | |
6b06314c JA |
6918 | } |
6919 | ||
08a45173 JA |
6920 | if (nr_files) { |
6921 | fpl->max = SCM_MAX_FD; | |
6922 | fpl->count = nr_files; | |
6923 | UNIXCB(skb).fp = fpl; | |
05f3fb3c | 6924 | skb->destructor = unix_destruct_scm; |
08a45173 JA |
6925 | refcount_add(skb->truesize, &sk->sk_wmem_alloc); |
6926 | skb_queue_head(&sk->sk_receive_queue, skb); | |
6b06314c | 6927 | |
08a45173 JA |
6928 | for (i = 0; i < nr_files; i++) |
6929 | fput(fpl->fp[i]); | |
6930 | } else { | |
6931 | kfree_skb(skb); | |
6932 | kfree(fpl); | |
6933 | } | |
6b06314c JA |
6934 | |
6935 | return 0; | |
6936 | } | |
6937 | ||
6938 | /* | |
6939 | * If UNIX sockets are enabled, fd passing can cause a reference cycle which | |
6940 | * causes regular reference counting to break down. We rely on the UNIX | |
6941 | * garbage collection to take care of this problem for us. | |
6942 | */ | |
6943 | static int io_sqe_files_scm(struct io_ring_ctx *ctx) | |
6944 | { | |
6945 | unsigned left, total; | |
6946 | int ret = 0; | |
6947 | ||
6948 | total = 0; | |
6949 | left = ctx->nr_user_files; | |
6950 | while (left) { | |
6951 | unsigned this_files = min_t(unsigned, left, SCM_MAX_FD); | |
6b06314c JA |
6952 | |
6953 | ret = __io_sqe_files_scm(ctx, this_files, total); | |
6954 | if (ret) | |
6955 | break; | |
6956 | left -= this_files; | |
6957 | total += this_files; | |
6958 | } | |
6959 | ||
6960 | if (!ret) | |
6961 | return 0; | |
6962 | ||
6963 | while (total < ctx->nr_user_files) { | |
65e19f54 JA |
6964 | struct file *file = io_file_from_index(ctx, total); |
6965 | ||
6966 | if (file) | |
6967 | fput(file); | |
6b06314c JA |
6968 | total++; |
6969 | } | |
6970 | ||
6971 | return ret; | |
6972 | } | |
6973 | #else | |
6974 | static int io_sqe_files_scm(struct io_ring_ctx *ctx) | |
6975 | { | |
6976 | return 0; | |
6977 | } | |
6978 | #endif | |
6979 | ||
65e19f54 JA |
6980 | static int io_sqe_alloc_file_tables(struct io_ring_ctx *ctx, unsigned nr_tables, |
6981 | unsigned nr_files) | |
6982 | { | |
6983 | int i; | |
6984 | ||
6985 | for (i = 0; i < nr_tables; i++) { | |
05f3fb3c | 6986 | struct fixed_file_table *table = &ctx->file_data->table[i]; |
65e19f54 JA |
6987 | unsigned this_files; |
6988 | ||
6989 | this_files = min(nr_files, IORING_MAX_FILES_TABLE); | |
6990 | table->files = kcalloc(this_files, sizeof(struct file *), | |
6991 | GFP_KERNEL); | |
6992 | if (!table->files) | |
6993 | break; | |
6994 | nr_files -= this_files; | |
6995 | } | |
6996 | ||
6997 | if (i == nr_tables) | |
6998 | return 0; | |
6999 | ||
7000 | for (i = 0; i < nr_tables; i++) { | |
05f3fb3c | 7001 | struct fixed_file_table *table = &ctx->file_data->table[i]; |
65e19f54 JA |
7002 | kfree(table->files); |
7003 | } | |
7004 | return 1; | |
7005 | } | |
7006 | ||
05f3fb3c JA |
7007 | static void io_ring_file_put(struct io_ring_ctx *ctx, struct file *file) |
7008 | { | |
7009 | #if defined(CONFIG_UNIX) | |
7010 | struct sock *sock = ctx->ring_sock->sk; | |
7011 | struct sk_buff_head list, *head = &sock->sk_receive_queue; | |
7012 | struct sk_buff *skb; | |
7013 | int i; | |
7014 | ||
7015 | __skb_queue_head_init(&list); | |
7016 | ||
7017 | /* | |
7018 | * Find the skb that holds this file in its SCM_RIGHTS. When found, | |
7019 | * remove this entry and rearrange the file array. | |
7020 | */ | |
7021 | skb = skb_dequeue(head); | |
7022 | while (skb) { | |
7023 | struct scm_fp_list *fp; | |
7024 | ||
7025 | fp = UNIXCB(skb).fp; | |
7026 | for (i = 0; i < fp->count; i++) { | |
7027 | int left; | |
7028 | ||
7029 | if (fp->fp[i] != file) | |
7030 | continue; | |
7031 | ||
7032 | unix_notinflight(fp->user, fp->fp[i]); | |
7033 | left = fp->count - 1 - i; | |
7034 | if (left) { | |
7035 | memmove(&fp->fp[i], &fp->fp[i + 1], | |
7036 | left * sizeof(struct file *)); | |
7037 | } | |
7038 | fp->count--; | |
7039 | if (!fp->count) { | |
7040 | kfree_skb(skb); | |
7041 | skb = NULL; | |
7042 | } else { | |
7043 | __skb_queue_tail(&list, skb); | |
7044 | } | |
7045 | fput(file); | |
7046 | file = NULL; | |
7047 | break; | |
7048 | } | |
7049 | ||
7050 | if (!file) | |
7051 | break; | |
7052 | ||
7053 | __skb_queue_tail(&list, skb); | |
7054 | ||
7055 | skb = skb_dequeue(head); | |
7056 | } | |
7057 | ||
7058 | if (skb_peek(&list)) { | |
7059 | spin_lock_irq(&head->lock); | |
7060 | while ((skb = __skb_dequeue(&list)) != NULL) | |
7061 | __skb_queue_tail(head, skb); | |
7062 | spin_unlock_irq(&head->lock); | |
7063 | } | |
7064 | #else | |
7065 | fput(file); | |
7066 | #endif | |
7067 | } | |
7068 | ||
7069 | struct io_file_put { | |
05589553 | 7070 | struct list_head list; |
05f3fb3c | 7071 | struct file *file; |
05f3fb3c JA |
7072 | }; |
7073 | ||
4a38aed2 | 7074 | static void __io_file_put_work(struct fixed_file_ref_node *ref_node) |
65e19f54 | 7075 | { |
4a38aed2 JA |
7076 | struct fixed_file_data *file_data = ref_node->file_data; |
7077 | struct io_ring_ctx *ctx = file_data->ctx; | |
05f3fb3c | 7078 | struct io_file_put *pfile, *tmp; |
05589553 XW |
7079 | |
7080 | list_for_each_entry_safe(pfile, tmp, &ref_node->file_list, list) { | |
6a4d07cd | 7081 | list_del(&pfile->list); |
05589553 XW |
7082 | io_ring_file_put(ctx, pfile->file); |
7083 | kfree(pfile); | |
65e19f54 | 7084 | } |
05589553 | 7085 | |
6a4d07cd JA |
7086 | spin_lock(&file_data->lock); |
7087 | list_del(&ref_node->node); | |
7088 | spin_unlock(&file_data->lock); | |
05589553 XW |
7089 | |
7090 | percpu_ref_exit(&ref_node->refs); | |
7091 | kfree(ref_node); | |
7092 | percpu_ref_put(&file_data->refs); | |
2faf852d | 7093 | } |
65e19f54 | 7094 | |
4a38aed2 JA |
7095 | static void io_file_put_work(struct work_struct *work) |
7096 | { | |
7097 | struct io_ring_ctx *ctx; | |
7098 | struct llist_node *node; | |
7099 | ||
7100 | ctx = container_of(work, struct io_ring_ctx, file_put_work.work); | |
7101 | node = llist_del_all(&ctx->file_put_llist); | |
7102 | ||
7103 | while (node) { | |
7104 | struct fixed_file_ref_node *ref_node; | |
7105 | struct llist_node *next = node->next; | |
7106 | ||
7107 | ref_node = llist_entry(node, struct fixed_file_ref_node, llist); | |
7108 | __io_file_put_work(ref_node); | |
7109 | node = next; | |
7110 | } | |
7111 | } | |
7112 | ||
05589553 | 7113 | static void io_file_data_ref_zero(struct percpu_ref *ref) |
2faf852d | 7114 | { |
05589553 | 7115 | struct fixed_file_ref_node *ref_node; |
4a38aed2 JA |
7116 | struct io_ring_ctx *ctx; |
7117 | bool first_add; | |
7118 | int delay = HZ; | |
65e19f54 | 7119 | |
05589553 | 7120 | ref_node = container_of(ref, struct fixed_file_ref_node, refs); |
4a38aed2 | 7121 | ctx = ref_node->file_data->ctx; |
05589553 | 7122 | |
4a38aed2 JA |
7123 | if (percpu_ref_is_dying(&ctx->file_data->refs)) |
7124 | delay = 0; | |
05589553 | 7125 | |
4a38aed2 JA |
7126 | first_add = llist_add(&ref_node->llist, &ctx->file_put_llist); |
7127 | if (!delay) | |
7128 | mod_delayed_work(system_wq, &ctx->file_put_work, 0); | |
7129 | else if (first_add) | |
7130 | queue_delayed_work(system_wq, &ctx->file_put_work, delay); | |
05f3fb3c | 7131 | } |
65e19f54 | 7132 | |
05589553 XW |
7133 | static struct fixed_file_ref_node *alloc_fixed_file_ref_node( |
7134 | struct io_ring_ctx *ctx) | |
05f3fb3c | 7135 | { |
05589553 | 7136 | struct fixed_file_ref_node *ref_node; |
05f3fb3c | 7137 | |
05589553 XW |
7138 | ref_node = kzalloc(sizeof(*ref_node), GFP_KERNEL); |
7139 | if (!ref_node) | |
7140 | return ERR_PTR(-ENOMEM); | |
05f3fb3c | 7141 | |
05589553 XW |
7142 | if (percpu_ref_init(&ref_node->refs, io_file_data_ref_zero, |
7143 | 0, GFP_KERNEL)) { | |
7144 | kfree(ref_node); | |
7145 | return ERR_PTR(-ENOMEM); | |
7146 | } | |
7147 | INIT_LIST_HEAD(&ref_node->node); | |
7148 | INIT_LIST_HEAD(&ref_node->file_list); | |
05589553 XW |
7149 | ref_node->file_data = ctx->file_data; |
7150 | return ref_node; | |
05589553 XW |
7151 | } |
7152 | ||
7153 | static void destroy_fixed_file_ref_node(struct fixed_file_ref_node *ref_node) | |
7154 | { | |
7155 | percpu_ref_exit(&ref_node->refs); | |
7156 | kfree(ref_node); | |
65e19f54 JA |
7157 | } |
7158 | ||
6b06314c JA |
7159 | static int io_sqe_files_register(struct io_ring_ctx *ctx, void __user *arg, |
7160 | unsigned nr_args) | |
7161 | { | |
7162 | __s32 __user *fds = (__s32 __user *) arg; | |
65e19f54 | 7163 | unsigned nr_tables; |
05f3fb3c | 7164 | struct file *file; |
6b06314c JA |
7165 | int fd, ret = 0; |
7166 | unsigned i; | |
05589553 | 7167 | struct fixed_file_ref_node *ref_node; |
6b06314c | 7168 | |
05f3fb3c | 7169 | if (ctx->file_data) |
6b06314c JA |
7170 | return -EBUSY; |
7171 | if (!nr_args) | |
7172 | return -EINVAL; | |
7173 | if (nr_args > IORING_MAX_FIXED_FILES) | |
7174 | return -EMFILE; | |
7175 | ||
05f3fb3c JA |
7176 | ctx->file_data = kzalloc(sizeof(*ctx->file_data), GFP_KERNEL); |
7177 | if (!ctx->file_data) | |
7178 | return -ENOMEM; | |
7179 | ctx->file_data->ctx = ctx; | |
7180 | init_completion(&ctx->file_data->done); | |
05589553 | 7181 | INIT_LIST_HEAD(&ctx->file_data->ref_list); |
f7fe9346 | 7182 | spin_lock_init(&ctx->file_data->lock); |
05f3fb3c | 7183 | |
65e19f54 | 7184 | nr_tables = DIV_ROUND_UP(nr_args, IORING_MAX_FILES_TABLE); |
05f3fb3c JA |
7185 | ctx->file_data->table = kcalloc(nr_tables, |
7186 | sizeof(struct fixed_file_table), | |
65e19f54 | 7187 | GFP_KERNEL); |
05f3fb3c JA |
7188 | if (!ctx->file_data->table) { |
7189 | kfree(ctx->file_data); | |
7190 | ctx->file_data = NULL; | |
6b06314c | 7191 | return -ENOMEM; |
05f3fb3c JA |
7192 | } |
7193 | ||
05589553 | 7194 | if (percpu_ref_init(&ctx->file_data->refs, io_file_ref_kill, |
05f3fb3c JA |
7195 | PERCPU_REF_ALLOW_REINIT, GFP_KERNEL)) { |
7196 | kfree(ctx->file_data->table); | |
7197 | kfree(ctx->file_data); | |
7198 | ctx->file_data = NULL; | |
6b06314c | 7199 | return -ENOMEM; |
05f3fb3c | 7200 | } |
6b06314c | 7201 | |
65e19f54 | 7202 | if (io_sqe_alloc_file_tables(ctx, nr_tables, nr_args)) { |
05f3fb3c JA |
7203 | percpu_ref_exit(&ctx->file_data->refs); |
7204 | kfree(ctx->file_data->table); | |
7205 | kfree(ctx->file_data); | |
7206 | ctx->file_data = NULL; | |
65e19f54 JA |
7207 | return -ENOMEM; |
7208 | } | |
7209 | ||
08a45173 | 7210 | for (i = 0; i < nr_args; i++, ctx->nr_user_files++) { |
65e19f54 JA |
7211 | struct fixed_file_table *table; |
7212 | unsigned index; | |
7213 | ||
6b06314c JA |
7214 | ret = -EFAULT; |
7215 | if (copy_from_user(&fd, &fds[i], sizeof(fd))) | |
7216 | break; | |
08a45173 JA |
7217 | /* allow sparse sets */ |
7218 | if (fd == -1) { | |
7219 | ret = 0; | |
7220 | continue; | |
7221 | } | |
6b06314c | 7222 | |
05f3fb3c | 7223 | table = &ctx->file_data->table[i >> IORING_FILE_TABLE_SHIFT]; |
65e19f54 | 7224 | index = i & IORING_FILE_TABLE_MASK; |
05f3fb3c | 7225 | file = fget(fd); |
6b06314c JA |
7226 | |
7227 | ret = -EBADF; | |
05f3fb3c | 7228 | if (!file) |
6b06314c | 7229 | break; |
05f3fb3c | 7230 | |
6b06314c JA |
7231 | /* |
7232 | * Don't allow io_uring instances to be registered. If UNIX | |
7233 | * isn't enabled, then this causes a reference cycle and this | |
7234 | * instance can never get freed. If UNIX is enabled we'll | |
7235 | * handle it just fine, but there's still no point in allowing | |
7236 | * a ring fd as it doesn't support regular read/write anyway. | |
7237 | */ | |
05f3fb3c JA |
7238 | if (file->f_op == &io_uring_fops) { |
7239 | fput(file); | |
6b06314c JA |
7240 | break; |
7241 | } | |
6b06314c | 7242 | ret = 0; |
05f3fb3c | 7243 | table->files[index] = file; |
6b06314c JA |
7244 | } |
7245 | ||
7246 | if (ret) { | |
65e19f54 | 7247 | for (i = 0; i < ctx->nr_user_files; i++) { |
65e19f54 JA |
7248 | file = io_file_from_index(ctx, i); |
7249 | if (file) | |
7250 | fput(file); | |
7251 | } | |
7252 | for (i = 0; i < nr_tables; i++) | |
05f3fb3c | 7253 | kfree(ctx->file_data->table[i].files); |
6b06314c | 7254 | |
667e57da | 7255 | percpu_ref_exit(&ctx->file_data->refs); |
05f3fb3c JA |
7256 | kfree(ctx->file_data->table); |
7257 | kfree(ctx->file_data); | |
7258 | ctx->file_data = NULL; | |
6b06314c JA |
7259 | ctx->nr_user_files = 0; |
7260 | return ret; | |
7261 | } | |
7262 | ||
7263 | ret = io_sqe_files_scm(ctx); | |
05589553 | 7264 | if (ret) { |
6b06314c | 7265 | io_sqe_files_unregister(ctx); |
05589553 XW |
7266 | return ret; |
7267 | } | |
6b06314c | 7268 | |
05589553 XW |
7269 | ref_node = alloc_fixed_file_ref_node(ctx); |
7270 | if (IS_ERR(ref_node)) { | |
7271 | io_sqe_files_unregister(ctx); | |
7272 | return PTR_ERR(ref_node); | |
7273 | } | |
7274 | ||
7275 | ctx->file_data->cur_refs = &ref_node->refs; | |
6a4d07cd | 7276 | spin_lock(&ctx->file_data->lock); |
05589553 | 7277 | list_add(&ref_node->node, &ctx->file_data->ref_list); |
6a4d07cd | 7278 | spin_unlock(&ctx->file_data->lock); |
05589553 | 7279 | percpu_ref_get(&ctx->file_data->refs); |
6b06314c JA |
7280 | return ret; |
7281 | } | |
7282 | ||
c3a31e60 JA |
7283 | static int io_sqe_file_register(struct io_ring_ctx *ctx, struct file *file, |
7284 | int index) | |
7285 | { | |
7286 | #if defined(CONFIG_UNIX) | |
7287 | struct sock *sock = ctx->ring_sock->sk; | |
7288 | struct sk_buff_head *head = &sock->sk_receive_queue; | |
7289 | struct sk_buff *skb; | |
7290 | ||
7291 | /* | |
7292 | * See if we can merge this file into an existing skb SCM_RIGHTS | |
7293 | * file set. If there's no room, fall back to allocating a new skb | |
7294 | * and filling it in. | |
7295 | */ | |
7296 | spin_lock_irq(&head->lock); | |
7297 | skb = skb_peek(head); | |
7298 | if (skb) { | |
7299 | struct scm_fp_list *fpl = UNIXCB(skb).fp; | |
7300 | ||
7301 | if (fpl->count < SCM_MAX_FD) { | |
7302 | __skb_unlink(skb, head); | |
7303 | spin_unlock_irq(&head->lock); | |
7304 | fpl->fp[fpl->count] = get_file(file); | |
7305 | unix_inflight(fpl->user, fpl->fp[fpl->count]); | |
7306 | fpl->count++; | |
7307 | spin_lock_irq(&head->lock); | |
7308 | __skb_queue_head(head, skb); | |
7309 | } else { | |
7310 | skb = NULL; | |
7311 | } | |
7312 | } | |
7313 | spin_unlock_irq(&head->lock); | |
7314 | ||
7315 | if (skb) { | |
7316 | fput(file); | |
7317 | return 0; | |
7318 | } | |
7319 | ||
7320 | return __io_sqe_files_scm(ctx, 1, index); | |
7321 | #else | |
7322 | return 0; | |
7323 | #endif | |
7324 | } | |
7325 | ||
a5318d3c | 7326 | static int io_queue_file_removal(struct fixed_file_data *data, |
05589553 | 7327 | struct file *file) |
05f3fb3c | 7328 | { |
a5318d3c | 7329 | struct io_file_put *pfile; |
05589553 XW |
7330 | struct percpu_ref *refs = data->cur_refs; |
7331 | struct fixed_file_ref_node *ref_node; | |
05f3fb3c | 7332 | |
05f3fb3c | 7333 | pfile = kzalloc(sizeof(*pfile), GFP_KERNEL); |
a5318d3c HD |
7334 | if (!pfile) |
7335 | return -ENOMEM; | |
05f3fb3c | 7336 | |
05589553 | 7337 | ref_node = container_of(refs, struct fixed_file_ref_node, refs); |
05f3fb3c | 7338 | pfile->file = file; |
05589553 XW |
7339 | list_add(&pfile->list, &ref_node->file_list); |
7340 | ||
a5318d3c | 7341 | return 0; |
05f3fb3c JA |
7342 | } |
7343 | ||
7344 | static int __io_sqe_files_update(struct io_ring_ctx *ctx, | |
7345 | struct io_uring_files_update *up, | |
7346 | unsigned nr_args) | |
7347 | { | |
7348 | struct fixed_file_data *data = ctx->file_data; | |
05589553 | 7349 | struct fixed_file_ref_node *ref_node; |
05f3fb3c | 7350 | struct file *file; |
c3a31e60 JA |
7351 | __s32 __user *fds; |
7352 | int fd, i, err; | |
7353 | __u32 done; | |
05589553 | 7354 | bool needs_switch = false; |
c3a31e60 | 7355 | |
05f3fb3c | 7356 | if (check_add_overflow(up->offset, nr_args, &done)) |
c3a31e60 JA |
7357 | return -EOVERFLOW; |
7358 | if (done > ctx->nr_user_files) | |
7359 | return -EINVAL; | |
7360 | ||
05589553 XW |
7361 | ref_node = alloc_fixed_file_ref_node(ctx); |
7362 | if (IS_ERR(ref_node)) | |
7363 | return PTR_ERR(ref_node); | |
7364 | ||
c3a31e60 | 7365 | done = 0; |
05f3fb3c | 7366 | fds = u64_to_user_ptr(up->fds); |
c3a31e60 | 7367 | while (nr_args) { |
65e19f54 JA |
7368 | struct fixed_file_table *table; |
7369 | unsigned index; | |
7370 | ||
c3a31e60 JA |
7371 | err = 0; |
7372 | if (copy_from_user(&fd, &fds[done], sizeof(fd))) { | |
7373 | err = -EFAULT; | |
7374 | break; | |
7375 | } | |
05f3fb3c JA |
7376 | i = array_index_nospec(up->offset, ctx->nr_user_files); |
7377 | table = &ctx->file_data->table[i >> IORING_FILE_TABLE_SHIFT]; | |
65e19f54 JA |
7378 | index = i & IORING_FILE_TABLE_MASK; |
7379 | if (table->files[index]) { | |
98dfd502 | 7380 | file = table->files[index]; |
a5318d3c HD |
7381 | err = io_queue_file_removal(data, file); |
7382 | if (err) | |
7383 | break; | |
65e19f54 | 7384 | table->files[index] = NULL; |
05589553 | 7385 | needs_switch = true; |
c3a31e60 JA |
7386 | } |
7387 | if (fd != -1) { | |
c3a31e60 JA |
7388 | file = fget(fd); |
7389 | if (!file) { | |
7390 | err = -EBADF; | |
7391 | break; | |
7392 | } | |
7393 | /* | |
7394 | * Don't allow io_uring instances to be registered. If | |
7395 | * UNIX isn't enabled, then this causes a reference | |
7396 | * cycle and this instance can never get freed. If UNIX | |
7397 | * is enabled we'll handle it just fine, but there's | |
7398 | * still no point in allowing a ring fd as it doesn't | |
7399 | * support regular read/write anyway. | |
7400 | */ | |
7401 | if (file->f_op == &io_uring_fops) { | |
7402 | fput(file); | |
7403 | err = -EBADF; | |
7404 | break; | |
7405 | } | |
65e19f54 | 7406 | table->files[index] = file; |
c3a31e60 | 7407 | err = io_sqe_file_register(ctx, file, i); |
f3bd9dae | 7408 | if (err) { |
95d1c8e5 | 7409 | table->files[index] = NULL; |
f3bd9dae | 7410 | fput(file); |
c3a31e60 | 7411 | break; |
f3bd9dae | 7412 | } |
c3a31e60 JA |
7413 | } |
7414 | nr_args--; | |
7415 | done++; | |
05f3fb3c JA |
7416 | up->offset++; |
7417 | } | |
7418 | ||
05589553 XW |
7419 | if (needs_switch) { |
7420 | percpu_ref_kill(data->cur_refs); | |
6a4d07cd | 7421 | spin_lock(&data->lock); |
05589553 XW |
7422 | list_add(&ref_node->node, &data->ref_list); |
7423 | data->cur_refs = &ref_node->refs; | |
6a4d07cd | 7424 | spin_unlock(&data->lock); |
05589553 XW |
7425 | percpu_ref_get(&ctx->file_data->refs); |
7426 | } else | |
7427 | destroy_fixed_file_ref_node(ref_node); | |
c3a31e60 JA |
7428 | |
7429 | return done ? done : err; | |
7430 | } | |
05589553 | 7431 | |
05f3fb3c JA |
7432 | static int io_sqe_files_update(struct io_ring_ctx *ctx, void __user *arg, |
7433 | unsigned nr_args) | |
7434 | { | |
7435 | struct io_uring_files_update up; | |
7436 | ||
7437 | if (!ctx->file_data) | |
7438 | return -ENXIO; | |
7439 | if (!nr_args) | |
7440 | return -EINVAL; | |
7441 | if (copy_from_user(&up, arg, sizeof(up))) | |
7442 | return -EFAULT; | |
7443 | if (up.resv) | |
7444 | return -EINVAL; | |
7445 | ||
7446 | return __io_sqe_files_update(ctx, &up, nr_args); | |
7447 | } | |
c3a31e60 | 7448 | |
e9fd9396 | 7449 | static void io_free_work(struct io_wq_work *work) |
7d723065 JA |
7450 | { |
7451 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); | |
7452 | ||
e9fd9396 | 7453 | /* Consider that io_steal_work() relies on this ref */ |
7d723065 JA |
7454 | io_put_req(req); |
7455 | } | |
7456 | ||
24369c2e PB |
7457 | static int io_init_wq_offload(struct io_ring_ctx *ctx, |
7458 | struct io_uring_params *p) | |
7459 | { | |
7460 | struct io_wq_data data; | |
7461 | struct fd f; | |
7462 | struct io_ring_ctx *ctx_attach; | |
7463 | unsigned int concurrency; | |
7464 | int ret = 0; | |
7465 | ||
7466 | data.user = ctx->user; | |
e9fd9396 | 7467 | data.free_work = io_free_work; |
f5fa38c5 | 7468 | data.do_work = io_wq_submit_work; |
24369c2e PB |
7469 | |
7470 | if (!(p->flags & IORING_SETUP_ATTACH_WQ)) { | |
7471 | /* Do QD, or 4 * CPUS, whatever is smallest */ | |
7472 | concurrency = min(ctx->sq_entries, 4 * num_online_cpus()); | |
7473 | ||
7474 | ctx->io_wq = io_wq_create(concurrency, &data); | |
7475 | if (IS_ERR(ctx->io_wq)) { | |
7476 | ret = PTR_ERR(ctx->io_wq); | |
7477 | ctx->io_wq = NULL; | |
7478 | } | |
7479 | return ret; | |
7480 | } | |
7481 | ||
7482 | f = fdget(p->wq_fd); | |
7483 | if (!f.file) | |
7484 | return -EBADF; | |
7485 | ||
7486 | if (f.file->f_op != &io_uring_fops) { | |
7487 | ret = -EINVAL; | |
7488 | goto out_fput; | |
7489 | } | |
7490 | ||
7491 | ctx_attach = f.file->private_data; | |
7492 | /* @io_wq is protected by holding the fd */ | |
7493 | if (!io_wq_get(ctx_attach->io_wq, &data)) { | |
7494 | ret = -EINVAL; | |
7495 | goto out_fput; | |
7496 | } | |
7497 | ||
7498 | ctx->io_wq = ctx_attach->io_wq; | |
7499 | out_fput: | |
7500 | fdput(f); | |
7501 | return ret; | |
7502 | } | |
7503 | ||
6c271ce2 JA |
7504 | static int io_sq_offload_start(struct io_ring_ctx *ctx, |
7505 | struct io_uring_params *p) | |
2b188cc1 JA |
7506 | { |
7507 | int ret; | |
7508 | ||
6c271ce2 | 7509 | if (ctx->flags & IORING_SETUP_SQPOLL) { |
3ec482d1 JA |
7510 | ret = -EPERM; |
7511 | if (!capable(CAP_SYS_ADMIN)) | |
7512 | goto err; | |
7513 | ||
917257da JA |
7514 | ctx->sq_thread_idle = msecs_to_jiffies(p->sq_thread_idle); |
7515 | if (!ctx->sq_thread_idle) | |
7516 | ctx->sq_thread_idle = HZ; | |
7517 | ||
6c271ce2 | 7518 | if (p->flags & IORING_SETUP_SQ_AFF) { |
44a9bd18 | 7519 | int cpu = p->sq_thread_cpu; |
6c271ce2 | 7520 | |
917257da | 7521 | ret = -EINVAL; |
44a9bd18 JA |
7522 | if (cpu >= nr_cpu_ids) |
7523 | goto err; | |
7889f44d | 7524 | if (!cpu_online(cpu)) |
917257da JA |
7525 | goto err; |
7526 | ||
6c271ce2 JA |
7527 | ctx->sqo_thread = kthread_create_on_cpu(io_sq_thread, |
7528 | ctx, cpu, | |
7529 | "io_uring-sq"); | |
7530 | } else { | |
7531 | ctx->sqo_thread = kthread_create(io_sq_thread, ctx, | |
7532 | "io_uring-sq"); | |
7533 | } | |
7534 | if (IS_ERR(ctx->sqo_thread)) { | |
7535 | ret = PTR_ERR(ctx->sqo_thread); | |
7536 | ctx->sqo_thread = NULL; | |
7537 | goto err; | |
7538 | } | |
7539 | wake_up_process(ctx->sqo_thread); | |
7540 | } else if (p->flags & IORING_SETUP_SQ_AFF) { | |
7541 | /* Can't have SQ_AFF without SQPOLL */ | |
7542 | ret = -EINVAL; | |
7543 | goto err; | |
7544 | } | |
7545 | ||
24369c2e PB |
7546 | ret = io_init_wq_offload(ctx, p); |
7547 | if (ret) | |
2b188cc1 | 7548 | goto err; |
2b188cc1 JA |
7549 | |
7550 | return 0; | |
7551 | err: | |
54a91f3b | 7552 | io_finish_async(ctx); |
2b188cc1 JA |
7553 | return ret; |
7554 | } | |
7555 | ||
a087e2b5 BM |
7556 | static inline void __io_unaccount_mem(struct user_struct *user, |
7557 | unsigned long nr_pages) | |
2b188cc1 JA |
7558 | { |
7559 | atomic_long_sub(nr_pages, &user->locked_vm); | |
7560 | } | |
7561 | ||
a087e2b5 BM |
7562 | static inline int __io_account_mem(struct user_struct *user, |
7563 | unsigned long nr_pages) | |
2b188cc1 JA |
7564 | { |
7565 | unsigned long page_limit, cur_pages, new_pages; | |
7566 | ||
7567 | /* Don't allow more pages than we can safely lock */ | |
7568 | page_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT; | |
7569 | ||
7570 | do { | |
7571 | cur_pages = atomic_long_read(&user->locked_vm); | |
7572 | new_pages = cur_pages + nr_pages; | |
7573 | if (new_pages > page_limit) | |
7574 | return -ENOMEM; | |
7575 | } while (atomic_long_cmpxchg(&user->locked_vm, cur_pages, | |
7576 | new_pages) != cur_pages); | |
7577 | ||
7578 | return 0; | |
7579 | } | |
7580 | ||
2e0464d4 BM |
7581 | static void io_unaccount_mem(struct io_ring_ctx *ctx, unsigned long nr_pages, |
7582 | enum io_mem_account acct) | |
a087e2b5 | 7583 | { |
aad5d8da | 7584 | if (ctx->limit_mem) |
a087e2b5 | 7585 | __io_unaccount_mem(ctx->user, nr_pages); |
30975825 | 7586 | |
2aede0e4 | 7587 | if (ctx->mm_account) { |
2e0464d4 | 7588 | if (acct == ACCT_LOCKED) |
2aede0e4 | 7589 | ctx->mm_account->locked_vm -= nr_pages; |
2e0464d4 | 7590 | else if (acct == ACCT_PINNED) |
2aede0e4 | 7591 | atomic64_sub(nr_pages, &ctx->mm_account->pinned_vm); |
2e0464d4 | 7592 | } |
a087e2b5 BM |
7593 | } |
7594 | ||
2e0464d4 BM |
7595 | static int io_account_mem(struct io_ring_ctx *ctx, unsigned long nr_pages, |
7596 | enum io_mem_account acct) | |
a087e2b5 | 7597 | { |
30975825 BM |
7598 | int ret; |
7599 | ||
7600 | if (ctx->limit_mem) { | |
7601 | ret = __io_account_mem(ctx->user, nr_pages); | |
7602 | if (ret) | |
7603 | return ret; | |
7604 | } | |
7605 | ||
2aede0e4 | 7606 | if (ctx->mm_account) { |
2e0464d4 | 7607 | if (acct == ACCT_LOCKED) |
2aede0e4 | 7608 | ctx->mm_account->locked_vm += nr_pages; |
2e0464d4 | 7609 | else if (acct == ACCT_PINNED) |
2aede0e4 | 7610 | atomic64_add(nr_pages, &ctx->mm_account->pinned_vm); |
2e0464d4 | 7611 | } |
a087e2b5 BM |
7612 | |
7613 | return 0; | |
7614 | } | |
7615 | ||
2b188cc1 JA |
7616 | static void io_mem_free(void *ptr) |
7617 | { | |
52e04ef4 MR |
7618 | struct page *page; |
7619 | ||
7620 | if (!ptr) | |
7621 | return; | |
2b188cc1 | 7622 | |
52e04ef4 | 7623 | page = virt_to_head_page(ptr); |
2b188cc1 JA |
7624 | if (put_page_testzero(page)) |
7625 | free_compound_page(page); | |
7626 | } | |
7627 | ||
7628 | static void *io_mem_alloc(size_t size) | |
7629 | { | |
7630 | gfp_t gfp_flags = GFP_KERNEL | __GFP_ZERO | __GFP_NOWARN | __GFP_COMP | | |
7631 | __GFP_NORETRY; | |
7632 | ||
7633 | return (void *) __get_free_pages(gfp_flags, get_order(size)); | |
7634 | } | |
7635 | ||
75b28aff HV |
7636 | static unsigned long rings_size(unsigned sq_entries, unsigned cq_entries, |
7637 | size_t *sq_offset) | |
7638 | { | |
7639 | struct io_rings *rings; | |
7640 | size_t off, sq_array_size; | |
7641 | ||
7642 | off = struct_size(rings, cqes, cq_entries); | |
7643 | if (off == SIZE_MAX) | |
7644 | return SIZE_MAX; | |
7645 | ||
7646 | #ifdef CONFIG_SMP | |
7647 | off = ALIGN(off, SMP_CACHE_BYTES); | |
7648 | if (off == 0) | |
7649 | return SIZE_MAX; | |
7650 | #endif | |
7651 | ||
b36200f5 DV |
7652 | if (sq_offset) |
7653 | *sq_offset = off; | |
7654 | ||
75b28aff HV |
7655 | sq_array_size = array_size(sizeof(u32), sq_entries); |
7656 | if (sq_array_size == SIZE_MAX) | |
7657 | return SIZE_MAX; | |
7658 | ||
7659 | if (check_add_overflow(off, sq_array_size, &off)) | |
7660 | return SIZE_MAX; | |
7661 | ||
75b28aff HV |
7662 | return off; |
7663 | } | |
7664 | ||
2b188cc1 JA |
7665 | static unsigned long ring_pages(unsigned sq_entries, unsigned cq_entries) |
7666 | { | |
75b28aff | 7667 | size_t pages; |
2b188cc1 | 7668 | |
75b28aff HV |
7669 | pages = (size_t)1 << get_order( |
7670 | rings_size(sq_entries, cq_entries, NULL)); | |
7671 | pages += (size_t)1 << get_order( | |
7672 | array_size(sizeof(struct io_uring_sqe), sq_entries)); | |
2b188cc1 | 7673 | |
75b28aff | 7674 | return pages; |
2b188cc1 JA |
7675 | } |
7676 | ||
edafccee JA |
7677 | static int io_sqe_buffer_unregister(struct io_ring_ctx *ctx) |
7678 | { | |
7679 | int i, j; | |
7680 | ||
7681 | if (!ctx->user_bufs) | |
7682 | return -ENXIO; | |
7683 | ||
7684 | for (i = 0; i < ctx->nr_user_bufs; i++) { | |
7685 | struct io_mapped_ubuf *imu = &ctx->user_bufs[i]; | |
7686 | ||
7687 | for (j = 0; j < imu->nr_bvecs; j++) | |
f1f6a7dd | 7688 | unpin_user_page(imu->bvec[j].bv_page); |
edafccee | 7689 | |
2e0464d4 | 7690 | io_unaccount_mem(ctx, imu->nr_bvecs, ACCT_PINNED); |
d4ef6475 | 7691 | kvfree(imu->bvec); |
edafccee JA |
7692 | imu->nr_bvecs = 0; |
7693 | } | |
7694 | ||
7695 | kfree(ctx->user_bufs); | |
7696 | ctx->user_bufs = NULL; | |
7697 | ctx->nr_user_bufs = 0; | |
7698 | return 0; | |
7699 | } | |
7700 | ||
7701 | static int io_copy_iov(struct io_ring_ctx *ctx, struct iovec *dst, | |
7702 | void __user *arg, unsigned index) | |
7703 | { | |
7704 | struct iovec __user *src; | |
7705 | ||
7706 | #ifdef CONFIG_COMPAT | |
7707 | if (ctx->compat) { | |
7708 | struct compat_iovec __user *ciovs; | |
7709 | struct compat_iovec ciov; | |
7710 | ||
7711 | ciovs = (struct compat_iovec __user *) arg; | |
7712 | if (copy_from_user(&ciov, &ciovs[index], sizeof(ciov))) | |
7713 | return -EFAULT; | |
7714 | ||
d55e5f5b | 7715 | dst->iov_base = u64_to_user_ptr((u64)ciov.iov_base); |
edafccee JA |
7716 | dst->iov_len = ciov.iov_len; |
7717 | return 0; | |
7718 | } | |
7719 | #endif | |
7720 | src = (struct iovec __user *) arg; | |
7721 | if (copy_from_user(dst, &src[index], sizeof(*dst))) | |
7722 | return -EFAULT; | |
7723 | return 0; | |
7724 | } | |
7725 | ||
7726 | static int io_sqe_buffer_register(struct io_ring_ctx *ctx, void __user *arg, | |
7727 | unsigned nr_args) | |
7728 | { | |
7729 | struct vm_area_struct **vmas = NULL; | |
7730 | struct page **pages = NULL; | |
7731 | int i, j, got_pages = 0; | |
7732 | int ret = -EINVAL; | |
7733 | ||
7734 | if (ctx->user_bufs) | |
7735 | return -EBUSY; | |
7736 | if (!nr_args || nr_args > UIO_MAXIOV) | |
7737 | return -EINVAL; | |
7738 | ||
7739 | ctx->user_bufs = kcalloc(nr_args, sizeof(struct io_mapped_ubuf), | |
7740 | GFP_KERNEL); | |
7741 | if (!ctx->user_bufs) | |
7742 | return -ENOMEM; | |
7743 | ||
7744 | for (i = 0; i < nr_args; i++) { | |
7745 | struct io_mapped_ubuf *imu = &ctx->user_bufs[i]; | |
7746 | unsigned long off, start, end, ubuf; | |
7747 | int pret, nr_pages; | |
7748 | struct iovec iov; | |
7749 | size_t size; | |
7750 | ||
7751 | ret = io_copy_iov(ctx, &iov, arg, i); | |
7752 | if (ret) | |
a278682d | 7753 | goto err; |
edafccee JA |
7754 | |
7755 | /* | |
7756 | * Don't impose further limits on the size and buffer | |
7757 | * constraints here, we'll -EINVAL later when IO is | |
7758 | * submitted if they are wrong. | |
7759 | */ | |
7760 | ret = -EFAULT; | |
7761 | if (!iov.iov_base || !iov.iov_len) | |
7762 | goto err; | |
7763 | ||
7764 | /* arbitrary limit, but we need something */ | |
7765 | if (iov.iov_len > SZ_1G) | |
7766 | goto err; | |
7767 | ||
7768 | ubuf = (unsigned long) iov.iov_base; | |
7769 | end = (ubuf + iov.iov_len + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
7770 | start = ubuf >> PAGE_SHIFT; | |
7771 | nr_pages = end - start; | |
7772 | ||
2e0464d4 | 7773 | ret = io_account_mem(ctx, nr_pages, ACCT_PINNED); |
a087e2b5 BM |
7774 | if (ret) |
7775 | goto err; | |
edafccee JA |
7776 | |
7777 | ret = 0; | |
7778 | if (!pages || nr_pages > got_pages) { | |
a8c73c1a DE |
7779 | kvfree(vmas); |
7780 | kvfree(pages); | |
d4ef6475 | 7781 | pages = kvmalloc_array(nr_pages, sizeof(struct page *), |
edafccee | 7782 | GFP_KERNEL); |
d4ef6475 | 7783 | vmas = kvmalloc_array(nr_pages, |
edafccee JA |
7784 | sizeof(struct vm_area_struct *), |
7785 | GFP_KERNEL); | |
7786 | if (!pages || !vmas) { | |
7787 | ret = -ENOMEM; | |
2e0464d4 | 7788 | io_unaccount_mem(ctx, nr_pages, ACCT_PINNED); |
edafccee JA |
7789 | goto err; |
7790 | } | |
7791 | got_pages = nr_pages; | |
7792 | } | |
7793 | ||
d4ef6475 | 7794 | imu->bvec = kvmalloc_array(nr_pages, sizeof(struct bio_vec), |
edafccee JA |
7795 | GFP_KERNEL); |
7796 | ret = -ENOMEM; | |
7797 | if (!imu->bvec) { | |
2e0464d4 | 7798 | io_unaccount_mem(ctx, nr_pages, ACCT_PINNED); |
edafccee JA |
7799 | goto err; |
7800 | } | |
7801 | ||
7802 | ret = 0; | |
d8ed45c5 | 7803 | mmap_read_lock(current->mm); |
2113b05d | 7804 | pret = pin_user_pages(ubuf, nr_pages, |
932f4a63 IW |
7805 | FOLL_WRITE | FOLL_LONGTERM, |
7806 | pages, vmas); | |
edafccee JA |
7807 | if (pret == nr_pages) { |
7808 | /* don't support file backed memory */ | |
7809 | for (j = 0; j < nr_pages; j++) { | |
7810 | struct vm_area_struct *vma = vmas[j]; | |
7811 | ||
7812 | if (vma->vm_file && | |
7813 | !is_file_hugepages(vma->vm_file)) { | |
7814 | ret = -EOPNOTSUPP; | |
7815 | break; | |
7816 | } | |
7817 | } | |
7818 | } else { | |
7819 | ret = pret < 0 ? pret : -EFAULT; | |
7820 | } | |
d8ed45c5 | 7821 | mmap_read_unlock(current->mm); |
edafccee JA |
7822 | if (ret) { |
7823 | /* | |
7824 | * if we did partial map, or found file backed vmas, | |
7825 | * release any pages we did get | |
7826 | */ | |
27c4d3a3 | 7827 | if (pret > 0) |
f1f6a7dd | 7828 | unpin_user_pages(pages, pret); |
2e0464d4 | 7829 | io_unaccount_mem(ctx, nr_pages, ACCT_PINNED); |
d4ef6475 | 7830 | kvfree(imu->bvec); |
edafccee JA |
7831 | goto err; |
7832 | } | |
7833 | ||
7834 | off = ubuf & ~PAGE_MASK; | |
7835 | size = iov.iov_len; | |
7836 | for (j = 0; j < nr_pages; j++) { | |
7837 | size_t vec_len; | |
7838 | ||
7839 | vec_len = min_t(size_t, size, PAGE_SIZE - off); | |
7840 | imu->bvec[j].bv_page = pages[j]; | |
7841 | imu->bvec[j].bv_len = vec_len; | |
7842 | imu->bvec[j].bv_offset = off; | |
7843 | off = 0; | |
7844 | size -= vec_len; | |
7845 | } | |
7846 | /* store original address for later verification */ | |
7847 | imu->ubuf = ubuf; | |
7848 | imu->len = iov.iov_len; | |
7849 | imu->nr_bvecs = nr_pages; | |
7850 | ||
7851 | ctx->nr_user_bufs++; | |
7852 | } | |
d4ef6475 MR |
7853 | kvfree(pages); |
7854 | kvfree(vmas); | |
edafccee JA |
7855 | return 0; |
7856 | err: | |
d4ef6475 MR |
7857 | kvfree(pages); |
7858 | kvfree(vmas); | |
edafccee JA |
7859 | io_sqe_buffer_unregister(ctx); |
7860 | return ret; | |
7861 | } | |
7862 | ||
9b402849 JA |
7863 | static int io_eventfd_register(struct io_ring_ctx *ctx, void __user *arg) |
7864 | { | |
7865 | __s32 __user *fds = arg; | |
7866 | int fd; | |
7867 | ||
7868 | if (ctx->cq_ev_fd) | |
7869 | return -EBUSY; | |
7870 | ||
7871 | if (copy_from_user(&fd, fds, sizeof(*fds))) | |
7872 | return -EFAULT; | |
7873 | ||
7874 | ctx->cq_ev_fd = eventfd_ctx_fdget(fd); | |
7875 | if (IS_ERR(ctx->cq_ev_fd)) { | |
7876 | int ret = PTR_ERR(ctx->cq_ev_fd); | |
7877 | ctx->cq_ev_fd = NULL; | |
7878 | return ret; | |
7879 | } | |
7880 | ||
7881 | return 0; | |
7882 | } | |
7883 | ||
7884 | static int io_eventfd_unregister(struct io_ring_ctx *ctx) | |
7885 | { | |
7886 | if (ctx->cq_ev_fd) { | |
7887 | eventfd_ctx_put(ctx->cq_ev_fd); | |
7888 | ctx->cq_ev_fd = NULL; | |
7889 | return 0; | |
7890 | } | |
7891 | ||
7892 | return -ENXIO; | |
7893 | } | |
7894 | ||
5a2e745d JA |
7895 | static int __io_destroy_buffers(int id, void *p, void *data) |
7896 | { | |
7897 | struct io_ring_ctx *ctx = data; | |
7898 | struct io_buffer *buf = p; | |
7899 | ||
067524e9 | 7900 | __io_remove_buffers(ctx, buf, id, -1U); |
5a2e745d JA |
7901 | return 0; |
7902 | } | |
7903 | ||
7904 | static void io_destroy_buffers(struct io_ring_ctx *ctx) | |
7905 | { | |
7906 | idr_for_each(&ctx->io_buffer_idr, __io_destroy_buffers, ctx); | |
7907 | idr_destroy(&ctx->io_buffer_idr); | |
7908 | } | |
7909 | ||
2b188cc1 JA |
7910 | static void io_ring_ctx_free(struct io_ring_ctx *ctx) |
7911 | { | |
6b06314c | 7912 | io_finish_async(ctx); |
5dbcad51 | 7913 | io_sqe_buffer_unregister(ctx); |
2aede0e4 JA |
7914 | |
7915 | if (ctx->sqo_task) { | |
7916 | put_task_struct(ctx->sqo_task); | |
7917 | ctx->sqo_task = NULL; | |
7918 | mmdrop(ctx->mm_account); | |
7919 | ctx->mm_account = NULL; | |
30975825 | 7920 | } |
def596e9 | 7921 | |
6b06314c | 7922 | io_sqe_files_unregister(ctx); |
9b402849 | 7923 | io_eventfd_unregister(ctx); |
5a2e745d | 7924 | io_destroy_buffers(ctx); |
41726c9a | 7925 | idr_destroy(&ctx->personality_idr); |
def596e9 | 7926 | |
2b188cc1 | 7927 | #if defined(CONFIG_UNIX) |
355e8d26 EB |
7928 | if (ctx->ring_sock) { |
7929 | ctx->ring_sock->file = NULL; /* so that iput() is called */ | |
2b188cc1 | 7930 | sock_release(ctx->ring_sock); |
355e8d26 | 7931 | } |
2b188cc1 JA |
7932 | #endif |
7933 | ||
75b28aff | 7934 | io_mem_free(ctx->rings); |
2b188cc1 | 7935 | io_mem_free(ctx->sq_sqes); |
2b188cc1 JA |
7936 | |
7937 | percpu_ref_exit(&ctx->refs); | |
2b188cc1 | 7938 | free_uid(ctx->user); |
181e448d | 7939 | put_cred(ctx->creds); |
78076bb6 | 7940 | kfree(ctx->cancel_hash); |
0ddf92e8 | 7941 | kmem_cache_free(req_cachep, ctx->fallback_req); |
2b188cc1 JA |
7942 | kfree(ctx); |
7943 | } | |
7944 | ||
7945 | static __poll_t io_uring_poll(struct file *file, poll_table *wait) | |
7946 | { | |
7947 | struct io_ring_ctx *ctx = file->private_data; | |
7948 | __poll_t mask = 0; | |
7949 | ||
7950 | poll_wait(file, &ctx->cq_wait, wait); | |
4f7067c3 SB |
7951 | /* |
7952 | * synchronizes with barrier from wq_has_sleeper call in | |
7953 | * io_commit_cqring | |
7954 | */ | |
2b188cc1 | 7955 | smp_rmb(); |
75b28aff HV |
7956 | if (READ_ONCE(ctx->rings->sq.tail) - ctx->cached_sq_head != |
7957 | ctx->rings->sq_ring_entries) | |
2b188cc1 | 7958 | mask |= EPOLLOUT | EPOLLWRNORM; |
63e5d81f | 7959 | if (io_cqring_events(ctx, false)) |
2b188cc1 JA |
7960 | mask |= EPOLLIN | EPOLLRDNORM; |
7961 | ||
7962 | return mask; | |
7963 | } | |
7964 | ||
7965 | static int io_uring_fasync(int fd, struct file *file, int on) | |
7966 | { | |
7967 | struct io_ring_ctx *ctx = file->private_data; | |
7968 | ||
7969 | return fasync_helper(fd, file, on, &ctx->cq_fasync); | |
7970 | } | |
7971 | ||
071698e1 JA |
7972 | static int io_remove_personalities(int id, void *p, void *data) |
7973 | { | |
7974 | struct io_ring_ctx *ctx = data; | |
7975 | const struct cred *cred; | |
7976 | ||
7977 | cred = idr_remove(&ctx->personality_idr, id); | |
7978 | if (cred) | |
7979 | put_cred(cred); | |
7980 | return 0; | |
7981 | } | |
7982 | ||
85faa7b8 JA |
7983 | static void io_ring_exit_work(struct work_struct *work) |
7984 | { | |
b2edc0a7 PB |
7985 | struct io_ring_ctx *ctx = container_of(work, struct io_ring_ctx, |
7986 | exit_work); | |
85faa7b8 | 7987 | |
56952e91 JA |
7988 | /* |
7989 | * If we're doing polled IO and end up having requests being | |
7990 | * submitted async (out-of-line), then completions can come in while | |
7991 | * we're waiting for refs to drop. We need to reap these manually, | |
7992 | * as nobody else will be looking for them. | |
7993 | */ | |
b2edc0a7 | 7994 | do { |
56952e91 JA |
7995 | if (ctx->rings) |
7996 | io_cqring_overflow_flush(ctx, true); | |
b2edc0a7 PB |
7997 | io_iopoll_try_reap_events(ctx); |
7998 | } while (!wait_for_completion_timeout(&ctx->ref_comp, HZ/20)); | |
85faa7b8 JA |
7999 | io_ring_ctx_free(ctx); |
8000 | } | |
8001 | ||
2b188cc1 JA |
8002 | static void io_ring_ctx_wait_and_kill(struct io_ring_ctx *ctx) |
8003 | { | |
8004 | mutex_lock(&ctx->uring_lock); | |
8005 | percpu_ref_kill(&ctx->refs); | |
8006 | mutex_unlock(&ctx->uring_lock); | |
8007 | ||
f3606e3a JA |
8008 | io_kill_timeouts(ctx, NULL); |
8009 | io_poll_remove_all(ctx, NULL); | |
561fb04a JA |
8010 | |
8011 | if (ctx->io_wq) | |
8012 | io_wq_cancel_all(ctx->io_wq); | |
8013 | ||
15dff286 JA |
8014 | /* if we failed setting up the ctx, we might not have any rings */ |
8015 | if (ctx->rings) | |
8016 | io_cqring_overflow_flush(ctx, true); | |
b2edc0a7 | 8017 | io_iopoll_try_reap_events(ctx); |
071698e1 | 8018 | idr_for_each(&ctx->personality_idr, io_remove_personalities, ctx); |
309fc03a JA |
8019 | |
8020 | /* | |
8021 | * Do this upfront, so we won't have a grace period where the ring | |
8022 | * is closed but resources aren't reaped yet. This can cause | |
8023 | * spurious failure in setting up a new ring. | |
8024 | */ | |
760618f7 JA |
8025 | io_unaccount_mem(ctx, ring_pages(ctx->sq_entries, ctx->cq_entries), |
8026 | ACCT_LOCKED); | |
309fc03a | 8027 | |
85faa7b8 | 8028 | INIT_WORK(&ctx->exit_work, io_ring_exit_work); |
fc666777 JA |
8029 | /* |
8030 | * Use system_unbound_wq to avoid spawning tons of event kworkers | |
8031 | * if we're exiting a ton of rings at the same time. It just adds | |
8032 | * noise and overhead, there's no discernable change in runtime | |
8033 | * over using system_wq. | |
8034 | */ | |
8035 | queue_work(system_unbound_wq, &ctx->exit_work); | |
2b188cc1 JA |
8036 | } |
8037 | ||
8038 | static int io_uring_release(struct inode *inode, struct file *file) | |
8039 | { | |
8040 | struct io_ring_ctx *ctx = file->private_data; | |
8041 | ||
8042 | file->private_data = NULL; | |
8043 | io_ring_ctx_wait_and_kill(ctx); | |
8044 | return 0; | |
8045 | } | |
8046 | ||
67c4d9e6 PB |
8047 | static bool io_wq_files_match(struct io_wq_work *work, void *data) |
8048 | { | |
8049 | struct files_struct *files = data; | |
8050 | ||
8051 | return work->files == files; | |
8052 | } | |
8053 | ||
f254ac04 JA |
8054 | /* |
8055 | * Returns true if 'preq' is the link parent of 'req' | |
8056 | */ | |
8057 | static bool io_match_link(struct io_kiocb *preq, struct io_kiocb *req) | |
8058 | { | |
8059 | struct io_kiocb *link; | |
8060 | ||
8061 | if (!(preq->flags & REQ_F_LINK_HEAD)) | |
8062 | return false; | |
8063 | ||
8064 | list_for_each_entry(link, &preq->link_list, link_list) { | |
8065 | if (link == req) | |
8066 | return true; | |
8067 | } | |
8068 | ||
8069 | return false; | |
8070 | } | |
8071 | ||
c127a2a1 PB |
8072 | static inline bool io_match_files(struct io_kiocb *req, |
8073 | struct files_struct *files) | |
8074 | { | |
8075 | return (req->flags & REQ_F_WORK_INITIALIZED) && req->work.files == files; | |
8076 | } | |
8077 | ||
8078 | static bool io_match_link_files(struct io_kiocb *req, | |
8079 | struct files_struct *files) | |
8080 | { | |
8081 | struct io_kiocb *link; | |
8082 | ||
8083 | if (io_match_files(req, files)) | |
8084 | return true; | |
8085 | if (req->flags & REQ_F_LINK_HEAD) { | |
8086 | list_for_each_entry(link, &req->link_list, link_list) { | |
8087 | if (io_match_files(link, files)) | |
8088 | return true; | |
8089 | } | |
8090 | } | |
8091 | return false; | |
8092 | } | |
8093 | ||
f254ac04 JA |
8094 | /* |
8095 | * We're looking to cancel 'req' because it's holding on to our files, but | |
8096 | * 'req' could be a link to another request. See if it is, and cancel that | |
8097 | * parent request if so. | |
8098 | */ | |
8099 | static bool io_poll_remove_link(struct io_ring_ctx *ctx, struct io_kiocb *req) | |
8100 | { | |
8101 | struct hlist_node *tmp; | |
8102 | struct io_kiocb *preq; | |
8103 | bool found = false; | |
8104 | int i; | |
8105 | ||
8106 | spin_lock_irq(&ctx->completion_lock); | |
8107 | for (i = 0; i < (1U << ctx->cancel_hash_bits); i++) { | |
8108 | struct hlist_head *list; | |
8109 | ||
8110 | list = &ctx->cancel_hash[i]; | |
8111 | hlist_for_each_entry_safe(preq, tmp, list, hash_node) { | |
8112 | found = io_match_link(preq, req); | |
8113 | if (found) { | |
8114 | io_poll_remove_one(preq); | |
8115 | break; | |
8116 | } | |
8117 | } | |
8118 | } | |
8119 | spin_unlock_irq(&ctx->completion_lock); | |
8120 | return found; | |
8121 | } | |
8122 | ||
8123 | static bool io_timeout_remove_link(struct io_ring_ctx *ctx, | |
8124 | struct io_kiocb *req) | |
8125 | { | |
8126 | struct io_kiocb *preq; | |
8127 | bool found = false; | |
8128 | ||
8129 | spin_lock_irq(&ctx->completion_lock); | |
8130 | list_for_each_entry(preq, &ctx->timeout_list, timeout.list) { | |
8131 | found = io_match_link(preq, req); | |
8132 | if (found) { | |
8133 | __io_timeout_cancel(preq); | |
8134 | break; | |
8135 | } | |
8136 | } | |
8137 | spin_unlock_irq(&ctx->completion_lock); | |
8138 | return found; | |
8139 | } | |
8140 | ||
b711d4ea JA |
8141 | static bool io_cancel_link_cb(struct io_wq_work *work, void *data) |
8142 | { | |
8143 | return io_match_link(container_of(work, struct io_kiocb, work), data); | |
8144 | } | |
8145 | ||
8146 | static void io_attempt_cancel(struct io_ring_ctx *ctx, struct io_kiocb *req) | |
8147 | { | |
8148 | enum io_wq_cancel cret; | |
8149 | ||
8150 | /* cancel this particular work, if it's running */ | |
8151 | cret = io_wq_cancel_work(ctx->io_wq, &req->work); | |
8152 | if (cret != IO_WQ_CANCEL_NOTFOUND) | |
8153 | return; | |
8154 | ||
8155 | /* find links that hold this pending, cancel those */ | |
8156 | cret = io_wq_cancel_cb(ctx->io_wq, io_cancel_link_cb, req, true); | |
8157 | if (cret != IO_WQ_CANCEL_NOTFOUND) | |
8158 | return; | |
8159 | ||
8160 | /* if we have a poll link holding this pending, cancel that */ | |
8161 | if (io_poll_remove_link(ctx, req)) | |
8162 | return; | |
8163 | ||
8164 | /* final option, timeout link is holding this req pending */ | |
8165 | io_timeout_remove_link(ctx, req); | |
8166 | } | |
8167 | ||
b7ddce3c PB |
8168 | static void io_cancel_defer_files(struct io_ring_ctx *ctx, |
8169 | struct files_struct *files) | |
8170 | { | |
8171 | struct io_defer_entry *de = NULL; | |
8172 | LIST_HEAD(list); | |
8173 | ||
8174 | spin_lock_irq(&ctx->completion_lock); | |
8175 | list_for_each_entry_reverse(de, &ctx->defer_list, list) { | |
c127a2a1 | 8176 | if (io_match_link_files(de->req, files)) { |
b7ddce3c PB |
8177 | list_cut_position(&list, &ctx->defer_list, &de->list); |
8178 | break; | |
8179 | } | |
8180 | } | |
8181 | spin_unlock_irq(&ctx->completion_lock); | |
8182 | ||
8183 | while (!list_empty(&list)) { | |
8184 | de = list_first_entry(&list, struct io_defer_entry, list); | |
8185 | list_del_init(&de->list); | |
8186 | req_set_fail_links(de->req); | |
8187 | io_put_req(de->req); | |
8188 | io_req_complete(de->req, -ECANCELED); | |
8189 | kfree(de); | |
8190 | } | |
8191 | } | |
8192 | ||
76e1b642 JA |
8193 | /* |
8194 | * Returns true if we found and killed one or more files pinning requests | |
8195 | */ | |
8196 | static bool io_uring_cancel_files(struct io_ring_ctx *ctx, | |
fcb323cc JA |
8197 | struct files_struct *files) |
8198 | { | |
67c4d9e6 | 8199 | if (list_empty_careful(&ctx->inflight_list)) |
76e1b642 | 8200 | return false; |
67c4d9e6 | 8201 | |
b7ddce3c | 8202 | io_cancel_defer_files(ctx, files); |
67c4d9e6 PB |
8203 | /* cancel all at once, should be faster than doing it one by one*/ |
8204 | io_wq_cancel_cb(ctx->io_wq, io_wq_files_match, files, true); | |
8205 | ||
fcb323cc | 8206 | while (!list_empty_careful(&ctx->inflight_list)) { |
d8f1b971 XW |
8207 | struct io_kiocb *cancel_req = NULL, *req; |
8208 | DEFINE_WAIT(wait); | |
fcb323cc JA |
8209 | |
8210 | spin_lock_irq(&ctx->inflight_lock); | |
8211 | list_for_each_entry(req, &ctx->inflight_list, inflight_entry) { | |
768134d4 JA |
8212 | if (req->work.files != files) |
8213 | continue; | |
8214 | /* req is being completed, ignore */ | |
8215 | if (!refcount_inc_not_zero(&req->refs)) | |
8216 | continue; | |
8217 | cancel_req = req; | |
8218 | break; | |
fcb323cc | 8219 | } |
768134d4 | 8220 | if (cancel_req) |
fcb323cc | 8221 | prepare_to_wait(&ctx->inflight_wait, &wait, |
768134d4 | 8222 | TASK_UNINTERRUPTIBLE); |
fcb323cc JA |
8223 | spin_unlock_irq(&ctx->inflight_lock); |
8224 | ||
768134d4 JA |
8225 | /* We need to keep going until we don't find a matching req */ |
8226 | if (!cancel_req) | |
fcb323cc | 8227 | break; |
bb175342 PB |
8228 | /* cancel this request, or head link requests */ |
8229 | io_attempt_cancel(ctx, cancel_req); | |
8230 | io_put_req(cancel_req); | |
6200b0ae JA |
8231 | /* cancellations _may_ trigger task work */ |
8232 | io_run_task_work(); | |
fcb323cc | 8233 | schedule(); |
d8f1b971 | 8234 | finish_wait(&ctx->inflight_wait, &wait); |
fcb323cc | 8235 | } |
76e1b642 JA |
8236 | |
8237 | return true; | |
fcb323cc JA |
8238 | } |
8239 | ||
801dd57b | 8240 | static bool io_cancel_task_cb(struct io_wq_work *work, void *data) |
44e728b8 | 8241 | { |
801dd57b PB |
8242 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); |
8243 | struct task_struct *task = data; | |
44e728b8 | 8244 | |
f3606e3a | 8245 | return io_task_match(req, task); |
44e728b8 PB |
8246 | } |
8247 | ||
fcb323cc JA |
8248 | static int io_uring_flush(struct file *file, void *data) |
8249 | { | |
8250 | struct io_ring_ctx *ctx = file->private_data; | |
8251 | ||
8252 | io_uring_cancel_files(ctx, data); | |
6ab23144 JA |
8253 | |
8254 | /* | |
8255 | * If the task is going away, cancel work it may have pending | |
8256 | */ | |
801dd57b PB |
8257 | if (fatal_signal_pending(current) || (current->flags & PF_EXITING)) |
8258 | io_wq_cancel_cb(ctx->io_wq, io_cancel_task_cb, current, true); | |
6ab23144 | 8259 | |
fcb323cc JA |
8260 | return 0; |
8261 | } | |
8262 | ||
6c5c240e RP |
8263 | static void *io_uring_validate_mmap_request(struct file *file, |
8264 | loff_t pgoff, size_t sz) | |
2b188cc1 | 8265 | { |
2b188cc1 | 8266 | struct io_ring_ctx *ctx = file->private_data; |
6c5c240e | 8267 | loff_t offset = pgoff << PAGE_SHIFT; |
2b188cc1 JA |
8268 | struct page *page; |
8269 | void *ptr; | |
8270 | ||
8271 | switch (offset) { | |
8272 | case IORING_OFF_SQ_RING: | |
75b28aff HV |
8273 | case IORING_OFF_CQ_RING: |
8274 | ptr = ctx->rings; | |
2b188cc1 JA |
8275 | break; |
8276 | case IORING_OFF_SQES: | |
8277 | ptr = ctx->sq_sqes; | |
8278 | break; | |
2b188cc1 | 8279 | default: |
6c5c240e | 8280 | return ERR_PTR(-EINVAL); |
2b188cc1 JA |
8281 | } |
8282 | ||
8283 | page = virt_to_head_page(ptr); | |
a50b854e | 8284 | if (sz > page_size(page)) |
6c5c240e RP |
8285 | return ERR_PTR(-EINVAL); |
8286 | ||
8287 | return ptr; | |
8288 | } | |
8289 | ||
8290 | #ifdef CONFIG_MMU | |
8291 | ||
8292 | static int io_uring_mmap(struct file *file, struct vm_area_struct *vma) | |
8293 | { | |
8294 | size_t sz = vma->vm_end - vma->vm_start; | |
8295 | unsigned long pfn; | |
8296 | void *ptr; | |
8297 | ||
8298 | ptr = io_uring_validate_mmap_request(file, vma->vm_pgoff, sz); | |
8299 | if (IS_ERR(ptr)) | |
8300 | return PTR_ERR(ptr); | |
2b188cc1 JA |
8301 | |
8302 | pfn = virt_to_phys(ptr) >> PAGE_SHIFT; | |
8303 | return remap_pfn_range(vma, vma->vm_start, pfn, sz, vma->vm_page_prot); | |
8304 | } | |
8305 | ||
6c5c240e RP |
8306 | #else /* !CONFIG_MMU */ |
8307 | ||
8308 | static int io_uring_mmap(struct file *file, struct vm_area_struct *vma) | |
8309 | { | |
8310 | return vma->vm_flags & (VM_SHARED | VM_MAYSHARE) ? 0 : -EINVAL; | |
8311 | } | |
8312 | ||
8313 | static unsigned int io_uring_nommu_mmap_capabilities(struct file *file) | |
8314 | { | |
8315 | return NOMMU_MAP_DIRECT | NOMMU_MAP_READ | NOMMU_MAP_WRITE; | |
8316 | } | |
8317 | ||
8318 | static unsigned long io_uring_nommu_get_unmapped_area(struct file *file, | |
8319 | unsigned long addr, unsigned long len, | |
8320 | unsigned long pgoff, unsigned long flags) | |
8321 | { | |
8322 | void *ptr; | |
8323 | ||
8324 | ptr = io_uring_validate_mmap_request(file, pgoff, len); | |
8325 | if (IS_ERR(ptr)) | |
8326 | return PTR_ERR(ptr); | |
8327 | ||
8328 | return (unsigned long) ptr; | |
8329 | } | |
8330 | ||
8331 | #endif /* !CONFIG_MMU */ | |
8332 | ||
2b188cc1 JA |
8333 | SYSCALL_DEFINE6(io_uring_enter, unsigned int, fd, u32, to_submit, |
8334 | u32, min_complete, u32, flags, const sigset_t __user *, sig, | |
8335 | size_t, sigsz) | |
8336 | { | |
8337 | struct io_ring_ctx *ctx; | |
8338 | long ret = -EBADF; | |
8339 | int submitted = 0; | |
8340 | struct fd f; | |
8341 | ||
4c6e277c | 8342 | io_run_task_work(); |
b41e9852 | 8343 | |
6c271ce2 | 8344 | if (flags & ~(IORING_ENTER_GETEVENTS | IORING_ENTER_SQ_WAKEUP)) |
2b188cc1 JA |
8345 | return -EINVAL; |
8346 | ||
8347 | f = fdget(fd); | |
8348 | if (!f.file) | |
8349 | return -EBADF; | |
8350 | ||
8351 | ret = -EOPNOTSUPP; | |
8352 | if (f.file->f_op != &io_uring_fops) | |
8353 | goto out_fput; | |
8354 | ||
8355 | ret = -ENXIO; | |
8356 | ctx = f.file->private_data; | |
8357 | if (!percpu_ref_tryget(&ctx->refs)) | |
8358 | goto out_fput; | |
8359 | ||
6c271ce2 JA |
8360 | /* |
8361 | * For SQ polling, the thread will do all submissions and completions. | |
8362 | * Just return the requested submit count, and wake the thread if | |
8363 | * we were asked to. | |
8364 | */ | |
b2a9eada | 8365 | ret = 0; |
6c271ce2 | 8366 | if (ctx->flags & IORING_SETUP_SQPOLL) { |
c1edbf5f JA |
8367 | if (!list_empty_careful(&ctx->cq_overflow_list)) |
8368 | io_cqring_overflow_flush(ctx, false); | |
6c271ce2 JA |
8369 | if (flags & IORING_ENTER_SQ_WAKEUP) |
8370 | wake_up(&ctx->sqo_wait); | |
8371 | submitted = to_submit; | |
b2a9eada | 8372 | } else if (to_submit) { |
2b188cc1 | 8373 | mutex_lock(&ctx->uring_lock); |
0cdaf760 | 8374 | submitted = io_submit_sqes(ctx, to_submit, f.file, fd); |
2b188cc1 | 8375 | mutex_unlock(&ctx->uring_lock); |
7c504e65 PB |
8376 | |
8377 | if (submitted != to_submit) | |
8378 | goto out; | |
2b188cc1 JA |
8379 | } |
8380 | if (flags & IORING_ENTER_GETEVENTS) { | |
8381 | min_complete = min(min_complete, ctx->cq_entries); | |
8382 | ||
32b2244a XW |
8383 | /* |
8384 | * When SETUP_IOPOLL and SETUP_SQPOLL are both enabled, user | |
8385 | * space applications don't need to do io completion events | |
8386 | * polling again, they can rely on io_sq_thread to do polling | |
8387 | * work, which can reduce cpu usage and uring_lock contention. | |
8388 | */ | |
8389 | if (ctx->flags & IORING_SETUP_IOPOLL && | |
8390 | !(ctx->flags & IORING_SETUP_SQPOLL)) { | |
7668b92a | 8391 | ret = io_iopoll_check(ctx, min_complete); |
def596e9 JA |
8392 | } else { |
8393 | ret = io_cqring_wait(ctx, min_complete, sig, sigsz); | |
8394 | } | |
2b188cc1 JA |
8395 | } |
8396 | ||
7c504e65 | 8397 | out: |
6805b32e | 8398 | percpu_ref_put(&ctx->refs); |
2b188cc1 JA |
8399 | out_fput: |
8400 | fdput(f); | |
8401 | return submitted ? submitted : ret; | |
8402 | } | |
8403 | ||
bebdb65e | 8404 | #ifdef CONFIG_PROC_FS |
87ce955b JA |
8405 | static int io_uring_show_cred(int id, void *p, void *data) |
8406 | { | |
8407 | const struct cred *cred = p; | |
8408 | struct seq_file *m = data; | |
8409 | struct user_namespace *uns = seq_user_ns(m); | |
8410 | struct group_info *gi; | |
8411 | kernel_cap_t cap; | |
8412 | unsigned __capi; | |
8413 | int g; | |
8414 | ||
8415 | seq_printf(m, "%5d\n", id); | |
8416 | seq_put_decimal_ull(m, "\tUid:\t", from_kuid_munged(uns, cred->uid)); | |
8417 | seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->euid)); | |
8418 | seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->suid)); | |
8419 | seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->fsuid)); | |
8420 | seq_put_decimal_ull(m, "\n\tGid:\t", from_kgid_munged(uns, cred->gid)); | |
8421 | seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->egid)); | |
8422 | seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->sgid)); | |
8423 | seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->fsgid)); | |
8424 | seq_puts(m, "\n\tGroups:\t"); | |
8425 | gi = cred->group_info; | |
8426 | for (g = 0; g < gi->ngroups; g++) { | |
8427 | seq_put_decimal_ull(m, g ? " " : "", | |
8428 | from_kgid_munged(uns, gi->gid[g])); | |
8429 | } | |
8430 | seq_puts(m, "\n\tCapEff:\t"); | |
8431 | cap = cred->cap_effective; | |
8432 | CAP_FOR_EACH_U32(__capi) | |
8433 | seq_put_hex_ll(m, NULL, cap.cap[CAP_LAST_U32 - __capi], 8); | |
8434 | seq_putc(m, '\n'); | |
8435 | return 0; | |
8436 | } | |
8437 | ||
8438 | static void __io_uring_show_fdinfo(struct io_ring_ctx *ctx, struct seq_file *m) | |
8439 | { | |
fad8e0de | 8440 | bool has_lock; |
87ce955b JA |
8441 | int i; |
8442 | ||
fad8e0de JA |
8443 | /* |
8444 | * Avoid ABBA deadlock between the seq lock and the io_uring mutex, | |
8445 | * since fdinfo case grabs it in the opposite direction of normal use | |
8446 | * cases. If we fail to get the lock, we just don't iterate any | |
8447 | * structures that could be going away outside the io_uring mutex. | |
8448 | */ | |
8449 | has_lock = mutex_trylock(&ctx->uring_lock); | |
8450 | ||
87ce955b | 8451 | seq_printf(m, "UserFiles:\t%u\n", ctx->nr_user_files); |
fad8e0de | 8452 | for (i = 0; has_lock && i < ctx->nr_user_files; i++) { |
87ce955b JA |
8453 | struct fixed_file_table *table; |
8454 | struct file *f; | |
8455 | ||
8456 | table = &ctx->file_data->table[i >> IORING_FILE_TABLE_SHIFT]; | |
8457 | f = table->files[i & IORING_FILE_TABLE_MASK]; | |
8458 | if (f) | |
8459 | seq_printf(m, "%5u: %s\n", i, file_dentry(f)->d_iname); | |
8460 | else | |
8461 | seq_printf(m, "%5u: <none>\n", i); | |
8462 | } | |
8463 | seq_printf(m, "UserBufs:\t%u\n", ctx->nr_user_bufs); | |
fad8e0de | 8464 | for (i = 0; has_lock && i < ctx->nr_user_bufs; i++) { |
87ce955b JA |
8465 | struct io_mapped_ubuf *buf = &ctx->user_bufs[i]; |
8466 | ||
8467 | seq_printf(m, "%5u: 0x%llx/%u\n", i, buf->ubuf, | |
8468 | (unsigned int) buf->len); | |
8469 | } | |
fad8e0de | 8470 | if (has_lock && !idr_is_empty(&ctx->personality_idr)) { |
87ce955b JA |
8471 | seq_printf(m, "Personalities:\n"); |
8472 | idr_for_each(&ctx->personality_idr, io_uring_show_cred, m); | |
8473 | } | |
d7718a9d JA |
8474 | seq_printf(m, "PollList:\n"); |
8475 | spin_lock_irq(&ctx->completion_lock); | |
8476 | for (i = 0; i < (1U << ctx->cancel_hash_bits); i++) { | |
8477 | struct hlist_head *list = &ctx->cancel_hash[i]; | |
8478 | struct io_kiocb *req; | |
8479 | ||
8480 | hlist_for_each_entry(req, list, hash_node) | |
8481 | seq_printf(m, " op=%d, task_works=%d\n", req->opcode, | |
8482 | req->task->task_works != NULL); | |
8483 | } | |
8484 | spin_unlock_irq(&ctx->completion_lock); | |
fad8e0de JA |
8485 | if (has_lock) |
8486 | mutex_unlock(&ctx->uring_lock); | |
87ce955b JA |
8487 | } |
8488 | ||
8489 | static void io_uring_show_fdinfo(struct seq_file *m, struct file *f) | |
8490 | { | |
8491 | struct io_ring_ctx *ctx = f->private_data; | |
8492 | ||
8493 | if (percpu_ref_tryget(&ctx->refs)) { | |
8494 | __io_uring_show_fdinfo(ctx, m); | |
8495 | percpu_ref_put(&ctx->refs); | |
8496 | } | |
8497 | } | |
bebdb65e | 8498 | #endif |
87ce955b | 8499 | |
2b188cc1 JA |
8500 | static const struct file_operations io_uring_fops = { |
8501 | .release = io_uring_release, | |
fcb323cc | 8502 | .flush = io_uring_flush, |
2b188cc1 | 8503 | .mmap = io_uring_mmap, |
6c5c240e RP |
8504 | #ifndef CONFIG_MMU |
8505 | .get_unmapped_area = io_uring_nommu_get_unmapped_area, | |
8506 | .mmap_capabilities = io_uring_nommu_mmap_capabilities, | |
8507 | #endif | |
2b188cc1 JA |
8508 | .poll = io_uring_poll, |
8509 | .fasync = io_uring_fasync, | |
bebdb65e | 8510 | #ifdef CONFIG_PROC_FS |
87ce955b | 8511 | .show_fdinfo = io_uring_show_fdinfo, |
bebdb65e | 8512 | #endif |
2b188cc1 JA |
8513 | }; |
8514 | ||
8515 | static int io_allocate_scq_urings(struct io_ring_ctx *ctx, | |
8516 | struct io_uring_params *p) | |
8517 | { | |
75b28aff HV |
8518 | struct io_rings *rings; |
8519 | size_t size, sq_array_offset; | |
2b188cc1 | 8520 | |
bd740481 JA |
8521 | /* make sure these are sane, as we already accounted them */ |
8522 | ctx->sq_entries = p->sq_entries; | |
8523 | ctx->cq_entries = p->cq_entries; | |
8524 | ||
75b28aff HV |
8525 | size = rings_size(p->sq_entries, p->cq_entries, &sq_array_offset); |
8526 | if (size == SIZE_MAX) | |
8527 | return -EOVERFLOW; | |
8528 | ||
8529 | rings = io_mem_alloc(size); | |
8530 | if (!rings) | |
2b188cc1 JA |
8531 | return -ENOMEM; |
8532 | ||
75b28aff HV |
8533 | ctx->rings = rings; |
8534 | ctx->sq_array = (u32 *)((char *)rings + sq_array_offset); | |
8535 | rings->sq_ring_mask = p->sq_entries - 1; | |
8536 | rings->cq_ring_mask = p->cq_entries - 1; | |
8537 | rings->sq_ring_entries = p->sq_entries; | |
8538 | rings->cq_ring_entries = p->cq_entries; | |
8539 | ctx->sq_mask = rings->sq_ring_mask; | |
8540 | ctx->cq_mask = rings->cq_ring_mask; | |
2b188cc1 JA |
8541 | |
8542 | size = array_size(sizeof(struct io_uring_sqe), p->sq_entries); | |
eb065d30 JA |
8543 | if (size == SIZE_MAX) { |
8544 | io_mem_free(ctx->rings); | |
8545 | ctx->rings = NULL; | |
2b188cc1 | 8546 | return -EOVERFLOW; |
eb065d30 | 8547 | } |
2b188cc1 JA |
8548 | |
8549 | ctx->sq_sqes = io_mem_alloc(size); | |
eb065d30 JA |
8550 | if (!ctx->sq_sqes) { |
8551 | io_mem_free(ctx->rings); | |
8552 | ctx->rings = NULL; | |
2b188cc1 | 8553 | return -ENOMEM; |
eb065d30 | 8554 | } |
2b188cc1 | 8555 | |
2b188cc1 JA |
8556 | return 0; |
8557 | } | |
8558 | ||
8559 | /* | |
8560 | * Allocate an anonymous fd, this is what constitutes the application | |
8561 | * visible backing of an io_uring instance. The application mmaps this | |
8562 | * fd to gain access to the SQ/CQ ring details. If UNIX sockets are enabled, | |
8563 | * we have to tie this fd to a socket for file garbage collection purposes. | |
8564 | */ | |
8565 | static int io_uring_get_fd(struct io_ring_ctx *ctx) | |
8566 | { | |
8567 | struct file *file; | |
8568 | int ret; | |
8569 | ||
8570 | #if defined(CONFIG_UNIX) | |
8571 | ret = sock_create_kern(&init_net, PF_UNIX, SOCK_RAW, IPPROTO_IP, | |
8572 | &ctx->ring_sock); | |
8573 | if (ret) | |
8574 | return ret; | |
8575 | #endif | |
8576 | ||
8577 | ret = get_unused_fd_flags(O_RDWR | O_CLOEXEC); | |
8578 | if (ret < 0) | |
8579 | goto err; | |
8580 | ||
8581 | file = anon_inode_getfile("[io_uring]", &io_uring_fops, ctx, | |
8582 | O_RDWR | O_CLOEXEC); | |
8583 | if (IS_ERR(file)) { | |
8584 | put_unused_fd(ret); | |
8585 | ret = PTR_ERR(file); | |
8586 | goto err; | |
8587 | } | |
8588 | ||
8589 | #if defined(CONFIG_UNIX) | |
8590 | ctx->ring_sock->file = file; | |
8591 | #endif | |
8592 | fd_install(ret, file); | |
8593 | return ret; | |
8594 | err: | |
8595 | #if defined(CONFIG_UNIX) | |
8596 | sock_release(ctx->ring_sock); | |
8597 | ctx->ring_sock = NULL; | |
8598 | #endif | |
8599 | return ret; | |
8600 | } | |
8601 | ||
7f13657d XW |
8602 | static int io_uring_create(unsigned entries, struct io_uring_params *p, |
8603 | struct io_uring_params __user *params) | |
2b188cc1 JA |
8604 | { |
8605 | struct user_struct *user = NULL; | |
8606 | struct io_ring_ctx *ctx; | |
aad5d8da | 8607 | bool limit_mem; |
2b188cc1 JA |
8608 | int ret; |
8609 | ||
8110c1a6 | 8610 | if (!entries) |
2b188cc1 | 8611 | return -EINVAL; |
8110c1a6 JA |
8612 | if (entries > IORING_MAX_ENTRIES) { |
8613 | if (!(p->flags & IORING_SETUP_CLAMP)) | |
8614 | return -EINVAL; | |
8615 | entries = IORING_MAX_ENTRIES; | |
8616 | } | |
2b188cc1 JA |
8617 | |
8618 | /* | |
8619 | * Use twice as many entries for the CQ ring. It's possible for the | |
8620 | * application to drive a higher depth than the size of the SQ ring, | |
8621 | * since the sqes are only used at submission time. This allows for | |
33a107f0 JA |
8622 | * some flexibility in overcommitting a bit. If the application has |
8623 | * set IORING_SETUP_CQSIZE, it will have passed in the desired number | |
8624 | * of CQ ring entries manually. | |
2b188cc1 JA |
8625 | */ |
8626 | p->sq_entries = roundup_pow_of_two(entries); | |
33a107f0 JA |
8627 | if (p->flags & IORING_SETUP_CQSIZE) { |
8628 | /* | |
8629 | * If IORING_SETUP_CQSIZE is set, we do the same roundup | |
8630 | * to a power-of-two, if it isn't already. We do NOT impose | |
8631 | * any cq vs sq ring sizing. | |
8632 | */ | |
8110c1a6 | 8633 | if (p->cq_entries < p->sq_entries) |
33a107f0 | 8634 | return -EINVAL; |
8110c1a6 JA |
8635 | if (p->cq_entries > IORING_MAX_CQ_ENTRIES) { |
8636 | if (!(p->flags & IORING_SETUP_CLAMP)) | |
8637 | return -EINVAL; | |
8638 | p->cq_entries = IORING_MAX_CQ_ENTRIES; | |
8639 | } | |
33a107f0 JA |
8640 | p->cq_entries = roundup_pow_of_two(p->cq_entries); |
8641 | } else { | |
8642 | p->cq_entries = 2 * p->sq_entries; | |
8643 | } | |
2b188cc1 JA |
8644 | |
8645 | user = get_uid(current_user()); | |
aad5d8da | 8646 | limit_mem = !capable(CAP_IPC_LOCK); |
2b188cc1 | 8647 | |
aad5d8da | 8648 | if (limit_mem) { |
a087e2b5 | 8649 | ret = __io_account_mem(user, |
2b188cc1 JA |
8650 | ring_pages(p->sq_entries, p->cq_entries)); |
8651 | if (ret) { | |
8652 | free_uid(user); | |
8653 | return ret; | |
8654 | } | |
8655 | } | |
8656 | ||
8657 | ctx = io_ring_ctx_alloc(p); | |
8658 | if (!ctx) { | |
aad5d8da | 8659 | if (limit_mem) |
a087e2b5 | 8660 | __io_unaccount_mem(user, ring_pages(p->sq_entries, |
2b188cc1 JA |
8661 | p->cq_entries)); |
8662 | free_uid(user); | |
8663 | return -ENOMEM; | |
8664 | } | |
8665 | ctx->compat = in_compat_syscall(); | |
2b188cc1 | 8666 | ctx->user = user; |
0b8c0ec7 | 8667 | ctx->creds = get_current_cred(); |
2b188cc1 | 8668 | |
2aede0e4 JA |
8669 | ctx->sqo_task = get_task_struct(current); |
8670 | ||
8671 | /* | |
8672 | * This is just grabbed for accounting purposes. When a process exits, | |
8673 | * the mm is exited and dropped before the files, hence we need to hang | |
8674 | * on to this mm purely for the purposes of being able to unaccount | |
8675 | * memory (locked/pinned vm). It's not used for anything else. | |
8676 | */ | |
6b7898eb | 8677 | mmgrab(current->mm); |
2aede0e4 | 8678 | ctx->mm_account = current->mm; |
6b7898eb | 8679 | |
f74441e6 JA |
8680 | /* |
8681 | * Account memory _before_ installing the file descriptor. Once | |
8682 | * the descriptor is installed, it can get closed at any time. Also | |
8683 | * do this before hitting the general error path, as ring freeing | |
8684 | * will un-account as well. | |
8685 | */ | |
8686 | io_account_mem(ctx, ring_pages(p->sq_entries, p->cq_entries), | |
8687 | ACCT_LOCKED); | |
8688 | ctx->limit_mem = limit_mem; | |
8689 | ||
2b188cc1 JA |
8690 | ret = io_allocate_scq_urings(ctx, p); |
8691 | if (ret) | |
8692 | goto err; | |
8693 | ||
6c271ce2 | 8694 | ret = io_sq_offload_start(ctx, p); |
2b188cc1 JA |
8695 | if (ret) |
8696 | goto err; | |
8697 | ||
2b188cc1 | 8698 | memset(&p->sq_off, 0, sizeof(p->sq_off)); |
75b28aff HV |
8699 | p->sq_off.head = offsetof(struct io_rings, sq.head); |
8700 | p->sq_off.tail = offsetof(struct io_rings, sq.tail); | |
8701 | p->sq_off.ring_mask = offsetof(struct io_rings, sq_ring_mask); | |
8702 | p->sq_off.ring_entries = offsetof(struct io_rings, sq_ring_entries); | |
8703 | p->sq_off.flags = offsetof(struct io_rings, sq_flags); | |
8704 | p->sq_off.dropped = offsetof(struct io_rings, sq_dropped); | |
8705 | p->sq_off.array = (char *)ctx->sq_array - (char *)ctx->rings; | |
2b188cc1 JA |
8706 | |
8707 | memset(&p->cq_off, 0, sizeof(p->cq_off)); | |
75b28aff HV |
8708 | p->cq_off.head = offsetof(struct io_rings, cq.head); |
8709 | p->cq_off.tail = offsetof(struct io_rings, cq.tail); | |
8710 | p->cq_off.ring_mask = offsetof(struct io_rings, cq_ring_mask); | |
8711 | p->cq_off.ring_entries = offsetof(struct io_rings, cq_ring_entries); | |
8712 | p->cq_off.overflow = offsetof(struct io_rings, cq_overflow); | |
8713 | p->cq_off.cqes = offsetof(struct io_rings, cqes); | |
0d9b5b3a | 8714 | p->cq_off.flags = offsetof(struct io_rings, cq_flags); |
ac90f249 | 8715 | |
7f13657d XW |
8716 | p->features = IORING_FEAT_SINGLE_MMAP | IORING_FEAT_NODROP | |
8717 | IORING_FEAT_SUBMIT_STABLE | IORING_FEAT_RW_CUR_POS | | |
5769a351 JX |
8718 | IORING_FEAT_CUR_PERSONALITY | IORING_FEAT_FAST_POLL | |
8719 | IORING_FEAT_POLL_32BITS; | |
7f13657d XW |
8720 | |
8721 | if (copy_to_user(params, p, sizeof(*p))) { | |
8722 | ret = -EFAULT; | |
8723 | goto err; | |
8724 | } | |
d1719f70 | 8725 | |
044c1ab3 JA |
8726 | /* |
8727 | * Install ring fd as the very last thing, so we don't risk someone | |
8728 | * having closed it before we finish setup | |
8729 | */ | |
8730 | ret = io_uring_get_fd(ctx); | |
8731 | if (ret < 0) | |
8732 | goto err; | |
8733 | ||
c826bd7a | 8734 | trace_io_uring_create(ret, ctx, p->sq_entries, p->cq_entries, p->flags); |
2b188cc1 JA |
8735 | return ret; |
8736 | err: | |
8737 | io_ring_ctx_wait_and_kill(ctx); | |
8738 | return ret; | |
8739 | } | |
8740 | ||
8741 | /* | |
8742 | * Sets up an aio uring context, and returns the fd. Applications asks for a | |
8743 | * ring size, we return the actual sq/cq ring sizes (among other things) in the | |
8744 | * params structure passed in. | |
8745 | */ | |
8746 | static long io_uring_setup(u32 entries, struct io_uring_params __user *params) | |
8747 | { | |
8748 | struct io_uring_params p; | |
2b188cc1 JA |
8749 | int i; |
8750 | ||
8751 | if (copy_from_user(&p, params, sizeof(p))) | |
8752 | return -EFAULT; | |
8753 | for (i = 0; i < ARRAY_SIZE(p.resv); i++) { | |
8754 | if (p.resv[i]) | |
8755 | return -EINVAL; | |
8756 | } | |
8757 | ||
6c271ce2 | 8758 | if (p.flags & ~(IORING_SETUP_IOPOLL | IORING_SETUP_SQPOLL | |
8110c1a6 | 8759 | IORING_SETUP_SQ_AFF | IORING_SETUP_CQSIZE | |
24369c2e | 8760 | IORING_SETUP_CLAMP | IORING_SETUP_ATTACH_WQ)) |
2b188cc1 JA |
8761 | return -EINVAL; |
8762 | ||
7f13657d | 8763 | return io_uring_create(entries, &p, params); |
2b188cc1 JA |
8764 | } |
8765 | ||
8766 | SYSCALL_DEFINE2(io_uring_setup, u32, entries, | |
8767 | struct io_uring_params __user *, params) | |
8768 | { | |
8769 | return io_uring_setup(entries, params); | |
8770 | } | |
8771 | ||
66f4af93 JA |
8772 | static int io_probe(struct io_ring_ctx *ctx, void __user *arg, unsigned nr_args) |
8773 | { | |
8774 | struct io_uring_probe *p; | |
8775 | size_t size; | |
8776 | int i, ret; | |
8777 | ||
8778 | size = struct_size(p, ops, nr_args); | |
8779 | if (size == SIZE_MAX) | |
8780 | return -EOVERFLOW; | |
8781 | p = kzalloc(size, GFP_KERNEL); | |
8782 | if (!p) | |
8783 | return -ENOMEM; | |
8784 | ||
8785 | ret = -EFAULT; | |
8786 | if (copy_from_user(p, arg, size)) | |
8787 | goto out; | |
8788 | ret = -EINVAL; | |
8789 | if (memchr_inv(p, 0, size)) | |
8790 | goto out; | |
8791 | ||
8792 | p->last_op = IORING_OP_LAST - 1; | |
8793 | if (nr_args > IORING_OP_LAST) | |
8794 | nr_args = IORING_OP_LAST; | |
8795 | ||
8796 | for (i = 0; i < nr_args; i++) { | |
8797 | p->ops[i].op = i; | |
8798 | if (!io_op_defs[i].not_supported) | |
8799 | p->ops[i].flags = IO_URING_OP_SUPPORTED; | |
8800 | } | |
8801 | p->ops_len = i; | |
8802 | ||
8803 | ret = 0; | |
8804 | if (copy_to_user(arg, p, size)) | |
8805 | ret = -EFAULT; | |
8806 | out: | |
8807 | kfree(p); | |
8808 | return ret; | |
8809 | } | |
8810 | ||
071698e1 JA |
8811 | static int io_register_personality(struct io_ring_ctx *ctx) |
8812 | { | |
8813 | const struct cred *creds = get_current_cred(); | |
8814 | int id; | |
8815 | ||
8816 | id = idr_alloc_cyclic(&ctx->personality_idr, (void *) creds, 1, | |
8817 | USHRT_MAX, GFP_KERNEL); | |
8818 | if (id < 0) | |
8819 | put_cred(creds); | |
8820 | return id; | |
8821 | } | |
8822 | ||
8823 | static int io_unregister_personality(struct io_ring_ctx *ctx, unsigned id) | |
8824 | { | |
8825 | const struct cred *old_creds; | |
8826 | ||
8827 | old_creds = idr_remove(&ctx->personality_idr, id); | |
8828 | if (old_creds) { | |
8829 | put_cred(old_creds); | |
8830 | return 0; | |
8831 | } | |
8832 | ||
8833 | return -EINVAL; | |
8834 | } | |
8835 | ||
8836 | static bool io_register_op_must_quiesce(int op) | |
8837 | { | |
8838 | switch (op) { | |
8839 | case IORING_UNREGISTER_FILES: | |
8840 | case IORING_REGISTER_FILES_UPDATE: | |
8841 | case IORING_REGISTER_PROBE: | |
8842 | case IORING_REGISTER_PERSONALITY: | |
8843 | case IORING_UNREGISTER_PERSONALITY: | |
8844 | return false; | |
8845 | default: | |
8846 | return true; | |
8847 | } | |
8848 | } | |
8849 | ||
edafccee JA |
8850 | static int __io_uring_register(struct io_ring_ctx *ctx, unsigned opcode, |
8851 | void __user *arg, unsigned nr_args) | |
b19062a5 JA |
8852 | __releases(ctx->uring_lock) |
8853 | __acquires(ctx->uring_lock) | |
edafccee JA |
8854 | { |
8855 | int ret; | |
8856 | ||
35fa71a0 JA |
8857 | /* |
8858 | * We're inside the ring mutex, if the ref is already dying, then | |
8859 | * someone else killed the ctx or is already going through | |
8860 | * io_uring_register(). | |
8861 | */ | |
8862 | if (percpu_ref_is_dying(&ctx->refs)) | |
8863 | return -ENXIO; | |
8864 | ||
071698e1 | 8865 | if (io_register_op_must_quiesce(opcode)) { |
05f3fb3c | 8866 | percpu_ref_kill(&ctx->refs); |
b19062a5 | 8867 | |
05f3fb3c JA |
8868 | /* |
8869 | * Drop uring mutex before waiting for references to exit. If | |
8870 | * another thread is currently inside io_uring_enter() it might | |
8871 | * need to grab the uring_lock to make progress. If we hold it | |
8872 | * here across the drain wait, then we can deadlock. It's safe | |
8873 | * to drop the mutex here, since no new references will come in | |
8874 | * after we've killed the percpu ref. | |
8875 | */ | |
8876 | mutex_unlock(&ctx->uring_lock); | |
0f158b4c | 8877 | ret = wait_for_completion_interruptible(&ctx->ref_comp); |
05f3fb3c | 8878 | mutex_lock(&ctx->uring_lock); |
c150368b JA |
8879 | if (ret) { |
8880 | percpu_ref_resurrect(&ctx->refs); | |
8881 | ret = -EINTR; | |
8882 | goto out; | |
8883 | } | |
05f3fb3c | 8884 | } |
edafccee JA |
8885 | |
8886 | switch (opcode) { | |
8887 | case IORING_REGISTER_BUFFERS: | |
8888 | ret = io_sqe_buffer_register(ctx, arg, nr_args); | |
8889 | break; | |
8890 | case IORING_UNREGISTER_BUFFERS: | |
8891 | ret = -EINVAL; | |
8892 | if (arg || nr_args) | |
8893 | break; | |
8894 | ret = io_sqe_buffer_unregister(ctx); | |
8895 | break; | |
6b06314c JA |
8896 | case IORING_REGISTER_FILES: |
8897 | ret = io_sqe_files_register(ctx, arg, nr_args); | |
8898 | break; | |
8899 | case IORING_UNREGISTER_FILES: | |
8900 | ret = -EINVAL; | |
8901 | if (arg || nr_args) | |
8902 | break; | |
8903 | ret = io_sqe_files_unregister(ctx); | |
8904 | break; | |
c3a31e60 JA |
8905 | case IORING_REGISTER_FILES_UPDATE: |
8906 | ret = io_sqe_files_update(ctx, arg, nr_args); | |
8907 | break; | |
9b402849 | 8908 | case IORING_REGISTER_EVENTFD: |
f2842ab5 | 8909 | case IORING_REGISTER_EVENTFD_ASYNC: |
9b402849 JA |
8910 | ret = -EINVAL; |
8911 | if (nr_args != 1) | |
8912 | break; | |
8913 | ret = io_eventfd_register(ctx, arg); | |
f2842ab5 JA |
8914 | if (ret) |
8915 | break; | |
8916 | if (opcode == IORING_REGISTER_EVENTFD_ASYNC) | |
8917 | ctx->eventfd_async = 1; | |
8918 | else | |
8919 | ctx->eventfd_async = 0; | |
9b402849 JA |
8920 | break; |
8921 | case IORING_UNREGISTER_EVENTFD: | |
8922 | ret = -EINVAL; | |
8923 | if (arg || nr_args) | |
8924 | break; | |
8925 | ret = io_eventfd_unregister(ctx); | |
8926 | break; | |
66f4af93 JA |
8927 | case IORING_REGISTER_PROBE: |
8928 | ret = -EINVAL; | |
8929 | if (!arg || nr_args > 256) | |
8930 | break; | |
8931 | ret = io_probe(ctx, arg, nr_args); | |
8932 | break; | |
071698e1 JA |
8933 | case IORING_REGISTER_PERSONALITY: |
8934 | ret = -EINVAL; | |
8935 | if (arg || nr_args) | |
8936 | break; | |
8937 | ret = io_register_personality(ctx); | |
8938 | break; | |
8939 | case IORING_UNREGISTER_PERSONALITY: | |
8940 | ret = -EINVAL; | |
8941 | if (arg) | |
8942 | break; | |
8943 | ret = io_unregister_personality(ctx, nr_args); | |
8944 | break; | |
edafccee JA |
8945 | default: |
8946 | ret = -EINVAL; | |
8947 | break; | |
8948 | } | |
8949 | ||
071698e1 | 8950 | if (io_register_op_must_quiesce(opcode)) { |
05f3fb3c | 8951 | /* bring the ctx back to life */ |
05f3fb3c | 8952 | percpu_ref_reinit(&ctx->refs); |
c150368b | 8953 | out: |
0f158b4c | 8954 | reinit_completion(&ctx->ref_comp); |
05f3fb3c | 8955 | } |
edafccee JA |
8956 | return ret; |
8957 | } | |
8958 | ||
8959 | SYSCALL_DEFINE4(io_uring_register, unsigned int, fd, unsigned int, opcode, | |
8960 | void __user *, arg, unsigned int, nr_args) | |
8961 | { | |
8962 | struct io_ring_ctx *ctx; | |
8963 | long ret = -EBADF; | |
8964 | struct fd f; | |
8965 | ||
8966 | f = fdget(fd); | |
8967 | if (!f.file) | |
8968 | return -EBADF; | |
8969 | ||
8970 | ret = -EOPNOTSUPP; | |
8971 | if (f.file->f_op != &io_uring_fops) | |
8972 | goto out_fput; | |
8973 | ||
8974 | ctx = f.file->private_data; | |
8975 | ||
8976 | mutex_lock(&ctx->uring_lock); | |
8977 | ret = __io_uring_register(ctx, opcode, arg, nr_args); | |
8978 | mutex_unlock(&ctx->uring_lock); | |
c826bd7a DD |
8979 | trace_io_uring_register(ctx, opcode, ctx->nr_user_files, ctx->nr_user_bufs, |
8980 | ctx->cq_ev_fd != NULL, ret); | |
edafccee JA |
8981 | out_fput: |
8982 | fdput(f); | |
8983 | return ret; | |
8984 | } | |
8985 | ||
2b188cc1 JA |
8986 | static int __init io_uring_init(void) |
8987 | { | |
d7f62e82 SM |
8988 | #define __BUILD_BUG_VERIFY_ELEMENT(stype, eoffset, etype, ename) do { \ |
8989 | BUILD_BUG_ON(offsetof(stype, ename) != eoffset); \ | |
8990 | BUILD_BUG_ON(sizeof(etype) != sizeof_field(stype, ename)); \ | |
8991 | } while (0) | |
8992 | ||
8993 | #define BUILD_BUG_SQE_ELEM(eoffset, etype, ename) \ | |
8994 | __BUILD_BUG_VERIFY_ELEMENT(struct io_uring_sqe, eoffset, etype, ename) | |
8995 | BUILD_BUG_ON(sizeof(struct io_uring_sqe) != 64); | |
8996 | BUILD_BUG_SQE_ELEM(0, __u8, opcode); | |
8997 | BUILD_BUG_SQE_ELEM(1, __u8, flags); | |
8998 | BUILD_BUG_SQE_ELEM(2, __u16, ioprio); | |
8999 | BUILD_BUG_SQE_ELEM(4, __s32, fd); | |
9000 | BUILD_BUG_SQE_ELEM(8, __u64, off); | |
9001 | BUILD_BUG_SQE_ELEM(8, __u64, addr2); | |
9002 | BUILD_BUG_SQE_ELEM(16, __u64, addr); | |
7d67af2c | 9003 | BUILD_BUG_SQE_ELEM(16, __u64, splice_off_in); |
d7f62e82 SM |
9004 | BUILD_BUG_SQE_ELEM(24, __u32, len); |
9005 | BUILD_BUG_SQE_ELEM(28, __kernel_rwf_t, rw_flags); | |
9006 | BUILD_BUG_SQE_ELEM(28, /* compat */ int, rw_flags); | |
9007 | BUILD_BUG_SQE_ELEM(28, /* compat */ __u32, rw_flags); | |
9008 | BUILD_BUG_SQE_ELEM(28, __u32, fsync_flags); | |
5769a351 JX |
9009 | BUILD_BUG_SQE_ELEM(28, /* compat */ __u16, poll_events); |
9010 | BUILD_BUG_SQE_ELEM(28, __u32, poll32_events); | |
d7f62e82 SM |
9011 | BUILD_BUG_SQE_ELEM(28, __u32, sync_range_flags); |
9012 | BUILD_BUG_SQE_ELEM(28, __u32, msg_flags); | |
9013 | BUILD_BUG_SQE_ELEM(28, __u32, timeout_flags); | |
9014 | BUILD_BUG_SQE_ELEM(28, __u32, accept_flags); | |
9015 | BUILD_BUG_SQE_ELEM(28, __u32, cancel_flags); | |
9016 | BUILD_BUG_SQE_ELEM(28, __u32, open_flags); | |
9017 | BUILD_BUG_SQE_ELEM(28, __u32, statx_flags); | |
9018 | BUILD_BUG_SQE_ELEM(28, __u32, fadvise_advice); | |
7d67af2c | 9019 | BUILD_BUG_SQE_ELEM(28, __u32, splice_flags); |
d7f62e82 SM |
9020 | BUILD_BUG_SQE_ELEM(32, __u64, user_data); |
9021 | BUILD_BUG_SQE_ELEM(40, __u16, buf_index); | |
9022 | BUILD_BUG_SQE_ELEM(42, __u16, personality); | |
7d67af2c | 9023 | BUILD_BUG_SQE_ELEM(44, __s32, splice_fd_in); |
d7f62e82 | 9024 | |
d3656344 | 9025 | BUILD_BUG_ON(ARRAY_SIZE(io_op_defs) != IORING_OP_LAST); |
84557871 | 9026 | BUILD_BUG_ON(__REQ_F_LAST_BIT >= 8 * sizeof(int)); |
2b188cc1 JA |
9027 | req_cachep = KMEM_CACHE(io_kiocb, SLAB_HWCACHE_ALIGN | SLAB_PANIC); |
9028 | return 0; | |
9029 | }; | |
9030 | __initcall(io_uring_init); |