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 */ |
2b188cc1 | 268 | struct mm_struct *sqo_mm; |
6c271ce2 | 269 | wait_queue_head_t sqo_wait; |
75b28aff | 270 | |
6b06314c JA |
271 | /* |
272 | * If used, fixed file set. Writers must ensure that ->refs is dead, | |
273 | * readers must ensure that ->refs is alive as long as the file* is | |
274 | * used. Only updated through io_uring_register(2). | |
275 | */ | |
05f3fb3c | 276 | struct fixed_file_data *file_data; |
6b06314c | 277 | unsigned nr_user_files; |
b14cca0c PB |
278 | int ring_fd; |
279 | struct file *ring_file; | |
6b06314c | 280 | |
edafccee JA |
281 | /* if used, fixed mapped user buffers */ |
282 | unsigned nr_user_bufs; | |
283 | struct io_mapped_ubuf *user_bufs; | |
284 | ||
2b188cc1 JA |
285 | struct user_struct *user; |
286 | ||
0b8c0ec7 | 287 | const struct cred *creds; |
181e448d | 288 | |
0f158b4c JA |
289 | struct completion ref_comp; |
290 | struct completion sq_thread_comp; | |
206aefde | 291 | |
0ddf92e8 JA |
292 | /* if all else fails... */ |
293 | struct io_kiocb *fallback_req; | |
294 | ||
206aefde JA |
295 | #if defined(CONFIG_UNIX) |
296 | struct socket *ring_sock; | |
297 | #endif | |
298 | ||
5a2e745d JA |
299 | struct idr io_buffer_idr; |
300 | ||
071698e1 JA |
301 | struct idr personality_idr; |
302 | ||
206aefde JA |
303 | struct { |
304 | unsigned cached_cq_tail; | |
305 | unsigned cq_entries; | |
306 | unsigned cq_mask; | |
307 | atomic_t cq_timeouts; | |
ad3eb2c8 | 308 | unsigned long cq_check_overflow; |
206aefde JA |
309 | struct wait_queue_head cq_wait; |
310 | struct fasync_struct *cq_fasync; | |
311 | struct eventfd_ctx *cq_ev_fd; | |
312 | } ____cacheline_aligned_in_smp; | |
2b188cc1 JA |
313 | |
314 | struct { | |
315 | struct mutex uring_lock; | |
316 | wait_queue_head_t wait; | |
317 | } ____cacheline_aligned_in_smp; | |
318 | ||
319 | struct { | |
320 | spinlock_t completion_lock; | |
e94f141b | 321 | |
def596e9 | 322 | /* |
540e32a0 | 323 | * ->iopoll_list is protected by the ctx->uring_lock for |
def596e9 JA |
324 | * io_uring instances that don't use IORING_SETUP_SQPOLL. |
325 | * For SQPOLL, only the single threaded io_sq_thread() will | |
326 | * manipulate the list, hence no extra locking is needed there. | |
327 | */ | |
540e32a0 | 328 | struct list_head iopoll_list; |
78076bb6 JA |
329 | struct hlist_head *cancel_hash; |
330 | unsigned cancel_hash_bits; | |
e94f141b | 331 | bool poll_multi_file; |
31b51510 | 332 | |
fcb323cc JA |
333 | spinlock_t inflight_lock; |
334 | struct list_head inflight_list; | |
2b188cc1 | 335 | } ____cacheline_aligned_in_smp; |
85faa7b8 | 336 | |
4a38aed2 JA |
337 | struct delayed_work file_put_work; |
338 | struct llist_head file_put_llist; | |
339 | ||
85faa7b8 | 340 | struct work_struct exit_work; |
2b188cc1 JA |
341 | }; |
342 | ||
09bb8394 JA |
343 | /* |
344 | * First field must be the file pointer in all the | |
345 | * iocb unions! See also 'struct kiocb' in <linux/fs.h> | |
346 | */ | |
221c5eb2 JA |
347 | struct io_poll_iocb { |
348 | struct file *file; | |
0969e783 JA |
349 | union { |
350 | struct wait_queue_head *head; | |
351 | u64 addr; | |
352 | }; | |
221c5eb2 | 353 | __poll_t events; |
8c838788 | 354 | bool done; |
221c5eb2 | 355 | bool canceled; |
392edb45 | 356 | struct wait_queue_entry wait; |
221c5eb2 JA |
357 | }; |
358 | ||
b5dba59e JA |
359 | struct io_close { |
360 | struct file *file; | |
361 | struct file *put_file; | |
362 | int fd; | |
363 | }; | |
364 | ||
ad8a48ac JA |
365 | struct io_timeout_data { |
366 | struct io_kiocb *req; | |
367 | struct hrtimer timer; | |
368 | struct timespec64 ts; | |
369 | enum hrtimer_mode mode; | |
370 | }; | |
371 | ||
8ed8d3c3 JA |
372 | struct io_accept { |
373 | struct file *file; | |
374 | struct sockaddr __user *addr; | |
375 | int __user *addr_len; | |
376 | int flags; | |
09952e3e | 377 | unsigned long nofile; |
8ed8d3c3 JA |
378 | }; |
379 | ||
380 | struct io_sync { | |
381 | struct file *file; | |
382 | loff_t len; | |
383 | loff_t off; | |
384 | int flags; | |
d63d1b5e | 385 | int mode; |
8ed8d3c3 JA |
386 | }; |
387 | ||
fbf23849 JA |
388 | struct io_cancel { |
389 | struct file *file; | |
390 | u64 addr; | |
391 | }; | |
392 | ||
b29472ee JA |
393 | struct io_timeout { |
394 | struct file *file; | |
395 | u64 addr; | |
396 | int flags; | |
bfe68a22 PB |
397 | u32 off; |
398 | u32 target_seq; | |
135fcde8 | 399 | struct list_head list; |
b29472ee JA |
400 | }; |
401 | ||
9adbd45d JA |
402 | struct io_rw { |
403 | /* NOTE: kiocb has the file as the first member, so don't do it here */ | |
404 | struct kiocb kiocb; | |
405 | u64 addr; | |
406 | u64 len; | |
407 | }; | |
408 | ||
3fbb51c1 JA |
409 | struct io_connect { |
410 | struct file *file; | |
411 | struct sockaddr __user *addr; | |
412 | int addr_len; | |
413 | }; | |
414 | ||
e47293fd JA |
415 | struct io_sr_msg { |
416 | struct file *file; | |
fddaface | 417 | union { |
270a5940 | 418 | struct user_msghdr __user *umsg; |
fddaface JA |
419 | void __user *buf; |
420 | }; | |
e47293fd | 421 | int msg_flags; |
bcda7baa | 422 | int bgid; |
fddaface | 423 | size_t len; |
bcda7baa | 424 | struct io_buffer *kbuf; |
e47293fd JA |
425 | }; |
426 | ||
15b71abe JA |
427 | struct io_open { |
428 | struct file *file; | |
429 | int dfd; | |
15b71abe | 430 | struct filename *filename; |
c12cedf2 | 431 | struct open_how how; |
4022e7af | 432 | unsigned long nofile; |
15b71abe JA |
433 | }; |
434 | ||
05f3fb3c JA |
435 | struct io_files_update { |
436 | struct file *file; | |
437 | u64 arg; | |
438 | u32 nr_args; | |
439 | u32 offset; | |
440 | }; | |
441 | ||
4840e418 JA |
442 | struct io_fadvise { |
443 | struct file *file; | |
444 | u64 offset; | |
445 | u32 len; | |
446 | u32 advice; | |
447 | }; | |
448 | ||
c1ca757b JA |
449 | struct io_madvise { |
450 | struct file *file; | |
451 | u64 addr; | |
452 | u32 len; | |
453 | u32 advice; | |
454 | }; | |
455 | ||
3e4827b0 JA |
456 | struct io_epoll { |
457 | struct file *file; | |
458 | int epfd; | |
459 | int op; | |
460 | int fd; | |
461 | struct epoll_event event; | |
e47293fd JA |
462 | }; |
463 | ||
7d67af2c PB |
464 | struct io_splice { |
465 | struct file *file_out; | |
466 | struct file *file_in; | |
467 | loff_t off_out; | |
468 | loff_t off_in; | |
469 | u64 len; | |
470 | unsigned int flags; | |
471 | }; | |
472 | ||
ddf0322d JA |
473 | struct io_provide_buf { |
474 | struct file *file; | |
475 | __u64 addr; | |
476 | __s32 len; | |
477 | __u32 bgid; | |
478 | __u16 nbufs; | |
479 | __u16 bid; | |
480 | }; | |
481 | ||
1d9e1288 BM |
482 | struct io_statx { |
483 | struct file *file; | |
484 | int dfd; | |
485 | unsigned int mask; | |
486 | unsigned int flags; | |
e62753e4 | 487 | const char __user *filename; |
1d9e1288 BM |
488 | struct statx __user *buffer; |
489 | }; | |
490 | ||
3ca405eb PB |
491 | struct io_completion { |
492 | struct file *file; | |
493 | struct list_head list; | |
0f7e466b | 494 | int cflags; |
3ca405eb PB |
495 | }; |
496 | ||
f499a021 JA |
497 | struct io_async_connect { |
498 | struct sockaddr_storage address; | |
499 | }; | |
500 | ||
03b1230c JA |
501 | struct io_async_msghdr { |
502 | struct iovec fast_iov[UIO_FASTIOV]; | |
503 | struct iovec *iov; | |
504 | struct sockaddr __user *uaddr; | |
505 | struct msghdr msg; | |
b537916c | 506 | struct sockaddr_storage addr; |
03b1230c JA |
507 | }; |
508 | ||
f67676d1 JA |
509 | struct io_async_rw { |
510 | struct iovec fast_iov[UIO_FASTIOV]; | |
ff6165b2 JA |
511 | const struct iovec *free_iovec; |
512 | struct iov_iter iter; | |
227c0c96 | 513 | size_t bytes_done; |
bcf5a063 | 514 | struct wait_page_queue wpq; |
f67676d1 JA |
515 | }; |
516 | ||
1a6b74fc | 517 | struct io_async_ctx { |
f67676d1 JA |
518 | union { |
519 | struct io_async_rw rw; | |
03b1230c | 520 | struct io_async_msghdr msg; |
f499a021 | 521 | struct io_async_connect connect; |
2d28390a | 522 | struct io_timeout_data timeout; |
f67676d1 | 523 | }; |
1a6b74fc JA |
524 | }; |
525 | ||
6b47ee6e PB |
526 | enum { |
527 | REQ_F_FIXED_FILE_BIT = IOSQE_FIXED_FILE_BIT, | |
528 | REQ_F_IO_DRAIN_BIT = IOSQE_IO_DRAIN_BIT, | |
529 | REQ_F_LINK_BIT = IOSQE_IO_LINK_BIT, | |
530 | REQ_F_HARDLINK_BIT = IOSQE_IO_HARDLINK_BIT, | |
531 | REQ_F_FORCE_ASYNC_BIT = IOSQE_ASYNC_BIT, | |
bcda7baa | 532 | REQ_F_BUFFER_SELECT_BIT = IOSQE_BUFFER_SELECT_BIT, |
6b47ee6e | 533 | |
dea3b49c | 534 | REQ_F_LINK_HEAD_BIT, |
6b47ee6e PB |
535 | REQ_F_FAIL_LINK_BIT, |
536 | REQ_F_INFLIGHT_BIT, | |
537 | REQ_F_CUR_POS_BIT, | |
538 | REQ_F_NOWAIT_BIT, | |
6b47ee6e | 539 | REQ_F_LINK_TIMEOUT_BIT, |
6b47ee6e | 540 | REQ_F_ISREG_BIT, |
6b47ee6e | 541 | REQ_F_COMP_LOCKED_BIT, |
99bc4c38 | 542 | REQ_F_NEED_CLEANUP_BIT, |
d7718a9d | 543 | REQ_F_POLLED_BIT, |
bcda7baa | 544 | REQ_F_BUFFER_SELECTED_BIT, |
5b0bbee4 | 545 | REQ_F_NO_FILE_TABLE_BIT, |
7cdaf587 | 546 | REQ_F_WORK_INITIALIZED_BIT, |
4dd2824d | 547 | REQ_F_TASK_PINNED_BIT, |
84557871 JA |
548 | |
549 | /* not a real bit, just to check we're not overflowing the space */ | |
550 | __REQ_F_LAST_BIT, | |
6b47ee6e PB |
551 | }; |
552 | ||
553 | enum { | |
554 | /* ctx owns file */ | |
555 | REQ_F_FIXED_FILE = BIT(REQ_F_FIXED_FILE_BIT), | |
556 | /* drain existing IO first */ | |
557 | REQ_F_IO_DRAIN = BIT(REQ_F_IO_DRAIN_BIT), | |
558 | /* linked sqes */ | |
559 | REQ_F_LINK = BIT(REQ_F_LINK_BIT), | |
560 | /* doesn't sever on completion < 0 */ | |
561 | REQ_F_HARDLINK = BIT(REQ_F_HARDLINK_BIT), | |
562 | /* IOSQE_ASYNC */ | |
563 | REQ_F_FORCE_ASYNC = BIT(REQ_F_FORCE_ASYNC_BIT), | |
bcda7baa JA |
564 | /* IOSQE_BUFFER_SELECT */ |
565 | REQ_F_BUFFER_SELECT = BIT(REQ_F_BUFFER_SELECT_BIT), | |
6b47ee6e | 566 | |
dea3b49c PB |
567 | /* head of a link */ |
568 | REQ_F_LINK_HEAD = BIT(REQ_F_LINK_HEAD_BIT), | |
6b47ee6e PB |
569 | /* fail rest of links */ |
570 | REQ_F_FAIL_LINK = BIT(REQ_F_FAIL_LINK_BIT), | |
571 | /* on inflight list */ | |
572 | REQ_F_INFLIGHT = BIT(REQ_F_INFLIGHT_BIT), | |
573 | /* read/write uses file position */ | |
574 | REQ_F_CUR_POS = BIT(REQ_F_CUR_POS_BIT), | |
575 | /* must not punt to workers */ | |
576 | REQ_F_NOWAIT = BIT(REQ_F_NOWAIT_BIT), | |
6b47ee6e PB |
577 | /* has linked timeout */ |
578 | REQ_F_LINK_TIMEOUT = BIT(REQ_F_LINK_TIMEOUT_BIT), | |
6b47ee6e PB |
579 | /* regular file */ |
580 | REQ_F_ISREG = BIT(REQ_F_ISREG_BIT), | |
6b47ee6e PB |
581 | /* completion under lock */ |
582 | REQ_F_COMP_LOCKED = BIT(REQ_F_COMP_LOCKED_BIT), | |
99bc4c38 PB |
583 | /* needs cleanup */ |
584 | REQ_F_NEED_CLEANUP = BIT(REQ_F_NEED_CLEANUP_BIT), | |
d7718a9d JA |
585 | /* already went through poll handler */ |
586 | REQ_F_POLLED = BIT(REQ_F_POLLED_BIT), | |
bcda7baa JA |
587 | /* buffer already selected */ |
588 | REQ_F_BUFFER_SELECTED = BIT(REQ_F_BUFFER_SELECTED_BIT), | |
5b0bbee4 JA |
589 | /* doesn't need file table for this request */ |
590 | REQ_F_NO_FILE_TABLE = BIT(REQ_F_NO_FILE_TABLE_BIT), | |
7cdaf587 XW |
591 | /* io_wq_work is initialized */ |
592 | REQ_F_WORK_INITIALIZED = BIT(REQ_F_WORK_INITIALIZED_BIT), | |
4dd2824d PB |
593 | /* req->task is refcounted */ |
594 | REQ_F_TASK_PINNED = BIT(REQ_F_TASK_PINNED_BIT), | |
d7718a9d JA |
595 | }; |
596 | ||
597 | struct async_poll { | |
598 | struct io_poll_iocb poll; | |
807abcb0 | 599 | struct io_poll_iocb *double_poll; |
6b47ee6e PB |
600 | }; |
601 | ||
09bb8394 JA |
602 | /* |
603 | * NOTE! Each of the iocb union members has the file pointer | |
604 | * as the first entry in their struct definition. So you can | |
605 | * access the file pointer through any of the sub-structs, | |
606 | * or directly as just 'ki_filp' in this struct. | |
607 | */ | |
2b188cc1 | 608 | struct io_kiocb { |
221c5eb2 | 609 | union { |
09bb8394 | 610 | struct file *file; |
9adbd45d | 611 | struct io_rw rw; |
221c5eb2 | 612 | struct io_poll_iocb poll; |
8ed8d3c3 JA |
613 | struct io_accept accept; |
614 | struct io_sync sync; | |
fbf23849 | 615 | struct io_cancel cancel; |
b29472ee | 616 | struct io_timeout timeout; |
3fbb51c1 | 617 | struct io_connect connect; |
e47293fd | 618 | struct io_sr_msg sr_msg; |
15b71abe | 619 | struct io_open open; |
b5dba59e | 620 | struct io_close close; |
05f3fb3c | 621 | struct io_files_update files_update; |
4840e418 | 622 | struct io_fadvise fadvise; |
c1ca757b | 623 | struct io_madvise madvise; |
3e4827b0 | 624 | struct io_epoll epoll; |
7d67af2c | 625 | struct io_splice splice; |
ddf0322d | 626 | struct io_provide_buf pbuf; |
1d9e1288 | 627 | struct io_statx statx; |
3ca405eb PB |
628 | /* use only after cleaning per-op data, see io_clean_op() */ |
629 | struct io_completion compl; | |
221c5eb2 | 630 | }; |
2b188cc1 | 631 | |
1a6b74fc | 632 | struct io_async_ctx *io; |
d625c6ee | 633 | u8 opcode; |
65a6543d XW |
634 | /* polled IO has completed */ |
635 | u8 iopoll_completed; | |
2b188cc1 | 636 | |
4f4eeba8 | 637 | u16 buf_index; |
9cf7c104 | 638 | u32 result; |
4f4eeba8 | 639 | |
010e8e6b PB |
640 | struct io_ring_ctx *ctx; |
641 | unsigned int flags; | |
642 | refcount_t refs; | |
643 | struct task_struct *task; | |
644 | u64 user_data; | |
d7718a9d | 645 | |
010e8e6b | 646 | struct list_head link_list; |
fcb323cc | 647 | |
d21ffe7e PB |
648 | /* |
649 | * 1. used with ctx->iopoll_list with reads/writes | |
650 | * 2. to track reqs with ->files (see io_op_def::file_table) | |
651 | */ | |
010e8e6b PB |
652 | struct list_head inflight_entry; |
653 | ||
654 | struct percpu_ref *fixed_file_refs; | |
655 | struct callback_head task_work; | |
656 | /* for polled requests, i.e. IORING_OP_POLL_ADD and async armed poll */ | |
657 | struct hlist_node hash_node; | |
658 | struct async_poll *apoll; | |
659 | struct io_wq_work work; | |
2b188cc1 | 660 | }; |
05589553 | 661 | |
27dc8338 PB |
662 | struct io_defer_entry { |
663 | struct list_head list; | |
664 | struct io_kiocb *req; | |
9cf7c104 | 665 | u32 seq; |
2b188cc1 JA |
666 | }; |
667 | ||
def596e9 | 668 | #define IO_IOPOLL_BATCH 8 |
2b188cc1 | 669 | |
013538bd JA |
670 | struct io_comp_state { |
671 | unsigned int nr; | |
672 | struct list_head list; | |
673 | struct io_ring_ctx *ctx; | |
674 | }; | |
675 | ||
9a56a232 JA |
676 | struct io_submit_state { |
677 | struct blk_plug plug; | |
678 | ||
2579f913 JA |
679 | /* |
680 | * io_kiocb alloc cache | |
681 | */ | |
682 | void *reqs[IO_IOPOLL_BATCH]; | |
6c8a3134 | 683 | unsigned int free_reqs; |
2579f913 | 684 | |
013538bd JA |
685 | /* |
686 | * Batch completion logic | |
687 | */ | |
688 | struct io_comp_state comp; | |
689 | ||
9a56a232 JA |
690 | /* |
691 | * File reference cache | |
692 | */ | |
693 | struct file *file; | |
694 | unsigned int fd; | |
695 | unsigned int has_refs; | |
9a56a232 JA |
696 | unsigned int ios_left; |
697 | }; | |
698 | ||
d3656344 JA |
699 | struct io_op_def { |
700 | /* needs req->io allocated for deferral/async */ | |
701 | unsigned async_ctx : 1; | |
702 | /* needs current->mm setup, does mm access */ | |
703 | unsigned needs_mm : 1; | |
704 | /* needs req->file assigned */ | |
705 | unsigned needs_file : 1; | |
fd2206e4 JA |
706 | /* don't fail if file grab fails */ |
707 | unsigned needs_file_no_error : 1; | |
d3656344 JA |
708 | /* hash wq insertion if file is a regular file */ |
709 | unsigned hash_reg_file : 1; | |
710 | /* unbound wq insertion if file is a non-regular file */ | |
711 | unsigned unbound_nonreg_file : 1; | |
66f4af93 JA |
712 | /* opcode is not supported by this kernel */ |
713 | unsigned not_supported : 1; | |
f86cd20c JA |
714 | /* needs file table */ |
715 | unsigned file_table : 1; | |
ff002b30 JA |
716 | /* needs ->fs */ |
717 | unsigned needs_fs : 1; | |
8a72758c JA |
718 | /* set if opcode supports polled "wait" */ |
719 | unsigned pollin : 1; | |
720 | unsigned pollout : 1; | |
bcda7baa JA |
721 | /* op supports buffer selection */ |
722 | unsigned buffer_select : 1; | |
57f1a649 | 723 | unsigned needs_fsize : 1; |
d3656344 JA |
724 | }; |
725 | ||
726 | static const struct io_op_def io_op_defs[] = { | |
0463b6c5 PB |
727 | [IORING_OP_NOP] = {}, |
728 | [IORING_OP_READV] = { | |
d3656344 JA |
729 | .async_ctx = 1, |
730 | .needs_mm = 1, | |
731 | .needs_file = 1, | |
732 | .unbound_nonreg_file = 1, | |
8a72758c | 733 | .pollin = 1, |
4d954c25 | 734 | .buffer_select = 1, |
d3656344 | 735 | }, |
0463b6c5 | 736 | [IORING_OP_WRITEV] = { |
d3656344 JA |
737 | .async_ctx = 1, |
738 | .needs_mm = 1, | |
739 | .needs_file = 1, | |
740 | .hash_reg_file = 1, | |
741 | .unbound_nonreg_file = 1, | |
8a72758c | 742 | .pollout = 1, |
57f1a649 | 743 | .needs_fsize = 1, |
d3656344 | 744 | }, |
0463b6c5 | 745 | [IORING_OP_FSYNC] = { |
d3656344 JA |
746 | .needs_file = 1, |
747 | }, | |
0463b6c5 | 748 | [IORING_OP_READ_FIXED] = { |
d3656344 JA |
749 | .needs_file = 1, |
750 | .unbound_nonreg_file = 1, | |
8a72758c | 751 | .pollin = 1, |
d3656344 | 752 | }, |
0463b6c5 | 753 | [IORING_OP_WRITE_FIXED] = { |
d3656344 JA |
754 | .needs_file = 1, |
755 | .hash_reg_file = 1, | |
756 | .unbound_nonreg_file = 1, | |
8a72758c | 757 | .pollout = 1, |
57f1a649 | 758 | .needs_fsize = 1, |
d3656344 | 759 | }, |
0463b6c5 | 760 | [IORING_OP_POLL_ADD] = { |
d3656344 JA |
761 | .needs_file = 1, |
762 | .unbound_nonreg_file = 1, | |
763 | }, | |
0463b6c5 PB |
764 | [IORING_OP_POLL_REMOVE] = {}, |
765 | [IORING_OP_SYNC_FILE_RANGE] = { | |
d3656344 JA |
766 | .needs_file = 1, |
767 | }, | |
0463b6c5 | 768 | [IORING_OP_SENDMSG] = { |
d3656344 JA |
769 | .async_ctx = 1, |
770 | .needs_mm = 1, | |
771 | .needs_file = 1, | |
772 | .unbound_nonreg_file = 1, | |
ff002b30 | 773 | .needs_fs = 1, |
8a72758c | 774 | .pollout = 1, |
d3656344 | 775 | }, |
0463b6c5 | 776 | [IORING_OP_RECVMSG] = { |
d3656344 JA |
777 | .async_ctx = 1, |
778 | .needs_mm = 1, | |
779 | .needs_file = 1, | |
780 | .unbound_nonreg_file = 1, | |
ff002b30 | 781 | .needs_fs = 1, |
8a72758c | 782 | .pollin = 1, |
52de1fe1 | 783 | .buffer_select = 1, |
d3656344 | 784 | }, |
0463b6c5 | 785 | [IORING_OP_TIMEOUT] = { |
d3656344 JA |
786 | .async_ctx = 1, |
787 | .needs_mm = 1, | |
788 | }, | |
0463b6c5 PB |
789 | [IORING_OP_TIMEOUT_REMOVE] = {}, |
790 | [IORING_OP_ACCEPT] = { | |
d3656344 JA |
791 | .needs_mm = 1, |
792 | .needs_file = 1, | |
793 | .unbound_nonreg_file = 1, | |
f86cd20c | 794 | .file_table = 1, |
8a72758c | 795 | .pollin = 1, |
d3656344 | 796 | }, |
0463b6c5 PB |
797 | [IORING_OP_ASYNC_CANCEL] = {}, |
798 | [IORING_OP_LINK_TIMEOUT] = { | |
d3656344 JA |
799 | .async_ctx = 1, |
800 | .needs_mm = 1, | |
801 | }, | |
0463b6c5 | 802 | [IORING_OP_CONNECT] = { |
d3656344 JA |
803 | .async_ctx = 1, |
804 | .needs_mm = 1, | |
805 | .needs_file = 1, | |
806 | .unbound_nonreg_file = 1, | |
8a72758c | 807 | .pollout = 1, |
d3656344 | 808 | }, |
0463b6c5 | 809 | [IORING_OP_FALLOCATE] = { |
d3656344 | 810 | .needs_file = 1, |
57f1a649 | 811 | .needs_fsize = 1, |
d3656344 | 812 | }, |
0463b6c5 | 813 | [IORING_OP_OPENAT] = { |
f86cd20c | 814 | .file_table = 1, |
ff002b30 | 815 | .needs_fs = 1, |
d3656344 | 816 | }, |
0463b6c5 | 817 | [IORING_OP_CLOSE] = { |
fd2206e4 JA |
818 | .needs_file = 1, |
819 | .needs_file_no_error = 1, | |
f86cd20c | 820 | .file_table = 1, |
d3656344 | 821 | }, |
0463b6c5 | 822 | [IORING_OP_FILES_UPDATE] = { |
d3656344 | 823 | .needs_mm = 1, |
f86cd20c | 824 | .file_table = 1, |
d3656344 | 825 | }, |
0463b6c5 | 826 | [IORING_OP_STATX] = { |
d3656344 | 827 | .needs_mm = 1, |
ff002b30 | 828 | .needs_fs = 1, |
5b0bbee4 | 829 | .file_table = 1, |
d3656344 | 830 | }, |
0463b6c5 | 831 | [IORING_OP_READ] = { |
3a6820f2 JA |
832 | .needs_mm = 1, |
833 | .needs_file = 1, | |
834 | .unbound_nonreg_file = 1, | |
8a72758c | 835 | .pollin = 1, |
bcda7baa | 836 | .buffer_select = 1, |
3a6820f2 | 837 | }, |
0463b6c5 | 838 | [IORING_OP_WRITE] = { |
3a6820f2 JA |
839 | .needs_mm = 1, |
840 | .needs_file = 1, | |
841 | .unbound_nonreg_file = 1, | |
8a72758c | 842 | .pollout = 1, |
57f1a649 | 843 | .needs_fsize = 1, |
3a6820f2 | 844 | }, |
0463b6c5 | 845 | [IORING_OP_FADVISE] = { |
4840e418 JA |
846 | .needs_file = 1, |
847 | }, | |
0463b6c5 | 848 | [IORING_OP_MADVISE] = { |
c1ca757b JA |
849 | .needs_mm = 1, |
850 | }, | |
0463b6c5 | 851 | [IORING_OP_SEND] = { |
fddaface JA |
852 | .needs_mm = 1, |
853 | .needs_file = 1, | |
854 | .unbound_nonreg_file = 1, | |
8a72758c | 855 | .pollout = 1, |
fddaface | 856 | }, |
0463b6c5 | 857 | [IORING_OP_RECV] = { |
fddaface JA |
858 | .needs_mm = 1, |
859 | .needs_file = 1, | |
860 | .unbound_nonreg_file = 1, | |
8a72758c | 861 | .pollin = 1, |
bcda7baa | 862 | .buffer_select = 1, |
fddaface | 863 | }, |
0463b6c5 | 864 | [IORING_OP_OPENAT2] = { |
f86cd20c | 865 | .file_table = 1, |
ff002b30 | 866 | .needs_fs = 1, |
cebdb986 | 867 | }, |
3e4827b0 JA |
868 | [IORING_OP_EPOLL_CTL] = { |
869 | .unbound_nonreg_file = 1, | |
870 | .file_table = 1, | |
871 | }, | |
7d67af2c PB |
872 | [IORING_OP_SPLICE] = { |
873 | .needs_file = 1, | |
874 | .hash_reg_file = 1, | |
875 | .unbound_nonreg_file = 1, | |
ddf0322d JA |
876 | }, |
877 | [IORING_OP_PROVIDE_BUFFERS] = {}, | |
067524e9 | 878 | [IORING_OP_REMOVE_BUFFERS] = {}, |
f2a8d5c7 PB |
879 | [IORING_OP_TEE] = { |
880 | .needs_file = 1, | |
881 | .hash_reg_file = 1, | |
882 | .unbound_nonreg_file = 1, | |
883 | }, | |
d3656344 JA |
884 | }; |
885 | ||
2e0464d4 BM |
886 | enum io_mem_account { |
887 | ACCT_LOCKED, | |
888 | ACCT_PINNED, | |
889 | }; | |
890 | ||
81b68a5c PB |
891 | static void __io_complete_rw(struct io_kiocb *req, long res, long res2, |
892 | struct io_comp_state *cs); | |
78e19bbe | 893 | static void io_cqring_fill_event(struct io_kiocb *req, long res); |
ec9c02ad | 894 | static void io_put_req(struct io_kiocb *req); |
c40f6379 | 895 | static void io_double_put_req(struct io_kiocb *req); |
978db57e | 896 | static void __io_double_put_req(struct io_kiocb *req); |
94ae5e77 | 897 | static struct io_kiocb *io_prep_linked_timeout(struct io_kiocb *req); |
7271ef3a | 898 | static void __io_queue_linked_timeout(struct io_kiocb *req); |
94ae5e77 | 899 | static void io_queue_linked_timeout(struct io_kiocb *req); |
05f3fb3c JA |
900 | static int __io_sqe_files_update(struct io_ring_ctx *ctx, |
901 | struct io_uring_files_update *ip, | |
902 | unsigned nr_args); | |
f56040b8 | 903 | static int io_prep_work_files(struct io_kiocb *req); |
3ca405eb | 904 | static void __io_clean_op(struct io_kiocb *req); |
b41e9852 JA |
905 | static int io_file_get(struct io_submit_state *state, struct io_kiocb *req, |
906 | int fd, struct file **out_file, bool fixed); | |
907 | static void __io_queue_sqe(struct io_kiocb *req, | |
f13fad7b JA |
908 | const struct io_uring_sqe *sqe, |
909 | struct io_comp_state *cs); | |
4349f30e | 910 | static void io_file_put_work(struct work_struct *work); |
de0617e4 | 911 | |
b63534c4 JA |
912 | static ssize_t io_import_iovec(int rw, struct io_kiocb *req, |
913 | struct iovec **iovec, struct iov_iter *iter, | |
914 | bool needs_lock); | |
ff6165b2 JA |
915 | static int io_setup_async_rw(struct io_kiocb *req, const struct iovec *iovec, |
916 | const struct iovec *fast_iov, | |
227c0c96 | 917 | struct iov_iter *iter, bool force); |
de0617e4 | 918 | |
2b188cc1 JA |
919 | static struct kmem_cache *req_cachep; |
920 | ||
921 | static const struct file_operations io_uring_fops; | |
922 | ||
923 | struct sock *io_uring_get_socket(struct file *file) | |
924 | { | |
925 | #if defined(CONFIG_UNIX) | |
926 | if (file->f_op == &io_uring_fops) { | |
927 | struct io_ring_ctx *ctx = file->private_data; | |
928 | ||
929 | return ctx->ring_sock->sk; | |
930 | } | |
931 | #endif | |
932 | return NULL; | |
933 | } | |
934 | EXPORT_SYMBOL(io_uring_get_socket); | |
935 | ||
4dd2824d PB |
936 | static void io_get_req_task(struct io_kiocb *req) |
937 | { | |
938 | if (req->flags & REQ_F_TASK_PINNED) | |
939 | return; | |
940 | get_task_struct(req->task); | |
941 | req->flags |= REQ_F_TASK_PINNED; | |
942 | } | |
943 | ||
3ca405eb PB |
944 | static inline void io_clean_op(struct io_kiocb *req) |
945 | { | |
bb175342 PB |
946 | if (req->flags & (REQ_F_NEED_CLEANUP | REQ_F_BUFFER_SELECTED | |
947 | REQ_F_INFLIGHT)) | |
3ca405eb PB |
948 | __io_clean_op(req); |
949 | } | |
950 | ||
4dd2824d PB |
951 | /* not idempotent -- it doesn't clear REQ_F_TASK_PINNED */ |
952 | static void __io_put_req_task(struct io_kiocb *req) | |
953 | { | |
954 | if (req->flags & REQ_F_TASK_PINNED) | |
955 | put_task_struct(req->task); | |
956 | } | |
957 | ||
4349f30e | 958 | static void io_sq_thread_drop_mm(void) |
c40f6379 JA |
959 | { |
960 | struct mm_struct *mm = current->mm; | |
961 | ||
962 | if (mm) { | |
963 | kthread_unuse_mm(mm); | |
964 | mmput(mm); | |
965 | } | |
966 | } | |
967 | ||
968 | static int __io_sq_thread_acquire_mm(struct io_ring_ctx *ctx) | |
969 | { | |
970 | if (!current->mm) { | |
cbcf7214 PB |
971 | if (unlikely(!(ctx->flags & IORING_SETUP_SQPOLL) || |
972 | !mmget_not_zero(ctx->sqo_mm))) | |
c40f6379 JA |
973 | return -EFAULT; |
974 | kthread_use_mm(ctx->sqo_mm); | |
975 | } | |
976 | ||
977 | return 0; | |
978 | } | |
979 | ||
980 | static int io_sq_thread_acquire_mm(struct io_ring_ctx *ctx, | |
981 | struct io_kiocb *req) | |
982 | { | |
983 | if (!io_op_defs[req->opcode].needs_mm) | |
984 | return 0; | |
985 | return __io_sq_thread_acquire_mm(ctx); | |
986 | } | |
987 | ||
988 | static inline void req_set_fail_links(struct io_kiocb *req) | |
989 | { | |
990 | if ((req->flags & (REQ_F_LINK | REQ_F_HARDLINK)) == REQ_F_LINK) | |
991 | req->flags |= REQ_F_FAIL_LINK; | |
992 | } | |
4a38aed2 | 993 | |
7cdaf587 XW |
994 | /* |
995 | * Note: must call io_req_init_async() for the first time you | |
996 | * touch any members of io_wq_work. | |
997 | */ | |
998 | static inline void io_req_init_async(struct io_kiocb *req) | |
999 | { | |
1000 | if (req->flags & REQ_F_WORK_INITIALIZED) | |
1001 | return; | |
1002 | ||
1003 | memset(&req->work, 0, sizeof(req->work)); | |
1004 | req->flags |= REQ_F_WORK_INITIALIZED; | |
1005 | } | |
1006 | ||
0cdaf760 PB |
1007 | static inline bool io_async_submit(struct io_ring_ctx *ctx) |
1008 | { | |
1009 | return ctx->flags & IORING_SETUP_SQPOLL; | |
1010 | } | |
1011 | ||
2b188cc1 JA |
1012 | static void io_ring_ctx_ref_free(struct percpu_ref *ref) |
1013 | { | |
1014 | struct io_ring_ctx *ctx = container_of(ref, struct io_ring_ctx, refs); | |
1015 | ||
0f158b4c | 1016 | complete(&ctx->ref_comp); |
2b188cc1 JA |
1017 | } |
1018 | ||
8eb7e2d0 PB |
1019 | static inline bool io_is_timeout_noseq(struct io_kiocb *req) |
1020 | { | |
1021 | return !req->timeout.off; | |
1022 | } | |
1023 | ||
2b188cc1 JA |
1024 | static struct io_ring_ctx *io_ring_ctx_alloc(struct io_uring_params *p) |
1025 | { | |
1026 | struct io_ring_ctx *ctx; | |
78076bb6 | 1027 | int hash_bits; |
2b188cc1 JA |
1028 | |
1029 | ctx = kzalloc(sizeof(*ctx), GFP_KERNEL); | |
1030 | if (!ctx) | |
1031 | return NULL; | |
1032 | ||
0ddf92e8 JA |
1033 | ctx->fallback_req = kmem_cache_alloc(req_cachep, GFP_KERNEL); |
1034 | if (!ctx->fallback_req) | |
1035 | goto err; | |
1036 | ||
78076bb6 JA |
1037 | /* |
1038 | * Use 5 bits less than the max cq entries, that should give us around | |
1039 | * 32 entries per hash list if totally full and uniformly spread. | |
1040 | */ | |
1041 | hash_bits = ilog2(p->cq_entries); | |
1042 | hash_bits -= 5; | |
1043 | if (hash_bits <= 0) | |
1044 | hash_bits = 1; | |
1045 | ctx->cancel_hash_bits = hash_bits; | |
1046 | ctx->cancel_hash = kmalloc((1U << hash_bits) * sizeof(struct hlist_head), | |
1047 | GFP_KERNEL); | |
1048 | if (!ctx->cancel_hash) | |
1049 | goto err; | |
1050 | __hash_init(ctx->cancel_hash, 1U << hash_bits); | |
1051 | ||
21482896 | 1052 | if (percpu_ref_init(&ctx->refs, io_ring_ctx_ref_free, |
206aefde JA |
1053 | PERCPU_REF_ALLOW_REINIT, GFP_KERNEL)) |
1054 | goto err; | |
2b188cc1 JA |
1055 | |
1056 | ctx->flags = p->flags; | |
583863ed | 1057 | init_waitqueue_head(&ctx->sqo_wait); |
2b188cc1 | 1058 | init_waitqueue_head(&ctx->cq_wait); |
1d7bb1d5 | 1059 | INIT_LIST_HEAD(&ctx->cq_overflow_list); |
0f158b4c JA |
1060 | init_completion(&ctx->ref_comp); |
1061 | init_completion(&ctx->sq_thread_comp); | |
5a2e745d | 1062 | idr_init(&ctx->io_buffer_idr); |
071698e1 | 1063 | idr_init(&ctx->personality_idr); |
2b188cc1 JA |
1064 | mutex_init(&ctx->uring_lock); |
1065 | init_waitqueue_head(&ctx->wait); | |
1066 | spin_lock_init(&ctx->completion_lock); | |
540e32a0 | 1067 | INIT_LIST_HEAD(&ctx->iopoll_list); |
de0617e4 | 1068 | INIT_LIST_HEAD(&ctx->defer_list); |
5262f567 | 1069 | INIT_LIST_HEAD(&ctx->timeout_list); |
fcb323cc JA |
1070 | init_waitqueue_head(&ctx->inflight_wait); |
1071 | spin_lock_init(&ctx->inflight_lock); | |
1072 | INIT_LIST_HEAD(&ctx->inflight_list); | |
4a38aed2 JA |
1073 | INIT_DELAYED_WORK(&ctx->file_put_work, io_file_put_work); |
1074 | init_llist_head(&ctx->file_put_llist); | |
2b188cc1 | 1075 | return ctx; |
206aefde | 1076 | err: |
0ddf92e8 JA |
1077 | if (ctx->fallback_req) |
1078 | kmem_cache_free(req_cachep, ctx->fallback_req); | |
78076bb6 | 1079 | kfree(ctx->cancel_hash); |
206aefde JA |
1080 | kfree(ctx); |
1081 | return NULL; | |
2b188cc1 JA |
1082 | } |
1083 | ||
9cf7c104 | 1084 | static bool req_need_defer(struct io_kiocb *req, u32 seq) |
7adf4eaf | 1085 | { |
2bc9930e JA |
1086 | if (unlikely(req->flags & REQ_F_IO_DRAIN)) { |
1087 | struct io_ring_ctx *ctx = req->ctx; | |
a197f664 | 1088 | |
9cf7c104 | 1089 | return seq != ctx->cached_cq_tail |
31af27c7 | 1090 | + atomic_read(&ctx->cached_cq_overflow); |
2bc9930e | 1091 | } |
de0617e4 | 1092 | |
9d858b21 | 1093 | return false; |
de0617e4 JA |
1094 | } |
1095 | ||
de0617e4 | 1096 | static void __io_commit_cqring(struct io_ring_ctx *ctx) |
2b188cc1 | 1097 | { |
75b28aff | 1098 | struct io_rings *rings = ctx->rings; |
2b188cc1 | 1099 | |
07910158 PB |
1100 | /* order cqe stores with ring update */ |
1101 | smp_store_release(&rings->cq.tail, ctx->cached_cq_tail); | |
2b188cc1 | 1102 | |
07910158 PB |
1103 | if (wq_has_sleeper(&ctx->cq_wait)) { |
1104 | wake_up_interruptible(&ctx->cq_wait); | |
1105 | kill_fasync(&ctx->cq_fasync, SIGIO, POLL_IN); | |
2b188cc1 JA |
1106 | } |
1107 | } | |
1108 | ||
51a4cc11 JA |
1109 | /* |
1110 | * Returns true if we need to defer file table putting. This can only happen | |
1111 | * from the error path with REQ_F_COMP_LOCKED set. | |
1112 | */ | |
1113 | static bool io_req_clean_work(struct io_kiocb *req) | |
18d9be1a | 1114 | { |
7cdaf587 | 1115 | if (!(req->flags & REQ_F_WORK_INITIALIZED)) |
51a4cc11 JA |
1116 | return false; |
1117 | ||
1118 | req->flags &= ~REQ_F_WORK_INITIALIZED; | |
7cdaf587 | 1119 | |
cccf0ee8 JA |
1120 | if (req->work.mm) { |
1121 | mmdrop(req->work.mm); | |
1122 | req->work.mm = NULL; | |
1123 | } | |
1124 | if (req->work.creds) { | |
1125 | put_cred(req->work.creds); | |
1126 | req->work.creds = NULL; | |
1127 | } | |
ff002b30 JA |
1128 | if (req->work.fs) { |
1129 | struct fs_struct *fs = req->work.fs; | |
1130 | ||
51a4cc11 JA |
1131 | if (req->flags & REQ_F_COMP_LOCKED) |
1132 | return true; | |
1133 | ||
ff002b30 JA |
1134 | spin_lock(&req->work.fs->lock); |
1135 | if (--fs->users) | |
1136 | fs = NULL; | |
1137 | spin_unlock(&req->work.fs->lock); | |
1138 | if (fs) | |
1139 | free_fs_struct(fs); | |
b65e0dd6 | 1140 | req->work.fs = NULL; |
ff002b30 | 1141 | } |
51a4cc11 JA |
1142 | |
1143 | return false; | |
561fb04a JA |
1144 | } |
1145 | ||
cbdcb435 | 1146 | static void io_prep_async_work(struct io_kiocb *req) |
18d9be1a | 1147 | { |
d3656344 | 1148 | const struct io_op_def *def = &io_op_defs[req->opcode]; |
54a91f3b | 1149 | |
16d59803 PB |
1150 | io_req_init_async(req); |
1151 | ||
d3656344 | 1152 | if (req->flags & REQ_F_ISREG) { |
eefdf30f | 1153 | if (def->hash_reg_file || (req->ctx->flags & IORING_SETUP_IOPOLL)) |
8766dd51 | 1154 | io_wq_hash_work(&req->work, file_inode(req->file)); |
d3656344 JA |
1155 | } else { |
1156 | if (def->unbound_nonreg_file) | |
3529d8c2 | 1157 | req->work.flags |= IO_WQ_WORK_UNBOUND; |
54a91f3b | 1158 | } |
dca9cf8b PB |
1159 | if (!req->work.mm && def->needs_mm) { |
1160 | mmgrab(current->mm); | |
1161 | req->work.mm = current->mm; | |
1162 | } | |
1163 | if (!req->work.creds) | |
1164 | req->work.creds = get_current_cred(); | |
1165 | if (!req->work.fs && def->needs_fs) { | |
1166 | spin_lock(¤t->fs->lock); | |
1167 | if (!current->fs->in_exec) { | |
1168 | req->work.fs = current->fs; | |
1169 | req->work.fs->users++; | |
1170 | } else { | |
1171 | req->work.flags |= IO_WQ_WORK_CANCEL; | |
1172 | } | |
1173 | spin_unlock(¤t->fs->lock); | |
1174 | } | |
57f1a649 PB |
1175 | if (def->needs_fsize) |
1176 | req->work.fsize = rlimit(RLIMIT_FSIZE); | |
1177 | else | |
1178 | req->work.fsize = RLIM_INFINITY; | |
561fb04a | 1179 | } |
cccf0ee8 | 1180 | |
cbdcb435 | 1181 | static void io_prep_async_link(struct io_kiocb *req) |
561fb04a | 1182 | { |
cbdcb435 | 1183 | struct io_kiocb *cur; |
54a91f3b | 1184 | |
cbdcb435 PB |
1185 | io_prep_async_work(req); |
1186 | if (req->flags & REQ_F_LINK_HEAD) | |
1187 | list_for_each_entry(cur, &req->link_list, link_list) | |
1188 | io_prep_async_work(cur); | |
561fb04a JA |
1189 | } |
1190 | ||
7271ef3a | 1191 | static struct io_kiocb *__io_queue_async_work(struct io_kiocb *req) |
561fb04a | 1192 | { |
a197f664 | 1193 | struct io_ring_ctx *ctx = req->ctx; |
cbdcb435 | 1194 | struct io_kiocb *link = io_prep_linked_timeout(req); |
561fb04a | 1195 | |
8766dd51 PB |
1196 | trace_io_uring_queue_async_work(ctx, io_wq_is_hashed(&req->work), req, |
1197 | &req->work, req->flags); | |
1198 | io_wq_enqueue(ctx->io_wq, &req->work); | |
7271ef3a | 1199 | return link; |
18d9be1a JA |
1200 | } |
1201 | ||
cbdcb435 PB |
1202 | static void io_queue_async_work(struct io_kiocb *req) |
1203 | { | |
7271ef3a JA |
1204 | struct io_kiocb *link; |
1205 | ||
cbdcb435 PB |
1206 | /* init ->work of the whole link before punting */ |
1207 | io_prep_async_link(req); | |
7271ef3a JA |
1208 | link = __io_queue_async_work(req); |
1209 | ||
1210 | if (link) | |
1211 | io_queue_linked_timeout(link); | |
cbdcb435 PB |
1212 | } |
1213 | ||
5262f567 JA |
1214 | static void io_kill_timeout(struct io_kiocb *req) |
1215 | { | |
1216 | int ret; | |
1217 | ||
2d28390a | 1218 | ret = hrtimer_try_to_cancel(&req->io->timeout.timer); |
5262f567 | 1219 | if (ret != -1) { |
01cec8c1 PB |
1220 | atomic_set(&req->ctx->cq_timeouts, |
1221 | atomic_read(&req->ctx->cq_timeouts) + 1); | |
135fcde8 | 1222 | list_del_init(&req->timeout.list); |
f0e20b89 | 1223 | req->flags |= REQ_F_COMP_LOCKED; |
78e19bbe | 1224 | io_cqring_fill_event(req, 0); |
ec9c02ad | 1225 | io_put_req(req); |
5262f567 JA |
1226 | } |
1227 | } | |
1228 | ||
1229 | static void io_kill_timeouts(struct io_ring_ctx *ctx) | |
1230 | { | |
1231 | struct io_kiocb *req, *tmp; | |
1232 | ||
1233 | spin_lock_irq(&ctx->completion_lock); | |
135fcde8 | 1234 | list_for_each_entry_safe(req, tmp, &ctx->timeout_list, timeout.list) |
5262f567 JA |
1235 | io_kill_timeout(req); |
1236 | spin_unlock_irq(&ctx->completion_lock); | |
1237 | } | |
1238 | ||
04518945 | 1239 | static void __io_queue_deferred(struct io_ring_ctx *ctx) |
de0617e4 | 1240 | { |
04518945 | 1241 | do { |
27dc8338 PB |
1242 | struct io_defer_entry *de = list_first_entry(&ctx->defer_list, |
1243 | struct io_defer_entry, list); | |
7271ef3a | 1244 | struct io_kiocb *link; |
de0617e4 | 1245 | |
9cf7c104 | 1246 | if (req_need_defer(de->req, de->seq)) |
04518945 | 1247 | break; |
27dc8338 | 1248 | list_del_init(&de->list); |
cbdcb435 | 1249 | /* punt-init is done before queueing for defer */ |
7271ef3a JA |
1250 | link = __io_queue_async_work(de->req); |
1251 | if (link) { | |
1252 | __io_queue_linked_timeout(link); | |
1253 | /* drop submission reference */ | |
1254 | link->flags |= REQ_F_COMP_LOCKED; | |
1255 | io_put_req(link); | |
1256 | } | |
27dc8338 | 1257 | kfree(de); |
04518945 PB |
1258 | } while (!list_empty(&ctx->defer_list)); |
1259 | } | |
1260 | ||
360428f8 | 1261 | static void io_flush_timeouts(struct io_ring_ctx *ctx) |
de0617e4 | 1262 | { |
360428f8 PB |
1263 | while (!list_empty(&ctx->timeout_list)) { |
1264 | struct io_kiocb *req = list_first_entry(&ctx->timeout_list, | |
135fcde8 | 1265 | struct io_kiocb, timeout.list); |
de0617e4 | 1266 | |
8eb7e2d0 | 1267 | if (io_is_timeout_noseq(req)) |
360428f8 | 1268 | break; |
bfe68a22 PB |
1269 | if (req->timeout.target_seq != ctx->cached_cq_tail |
1270 | - atomic_read(&ctx->cq_timeouts)) | |
360428f8 | 1271 | break; |
bfe68a22 | 1272 | |
135fcde8 | 1273 | list_del_init(&req->timeout.list); |
5262f567 | 1274 | io_kill_timeout(req); |
360428f8 PB |
1275 | } |
1276 | } | |
5262f567 | 1277 | |
360428f8 PB |
1278 | static void io_commit_cqring(struct io_ring_ctx *ctx) |
1279 | { | |
1280 | io_flush_timeouts(ctx); | |
de0617e4 JA |
1281 | __io_commit_cqring(ctx); |
1282 | ||
04518945 PB |
1283 | if (unlikely(!list_empty(&ctx->defer_list))) |
1284 | __io_queue_deferred(ctx); | |
de0617e4 JA |
1285 | } |
1286 | ||
2b188cc1 JA |
1287 | static struct io_uring_cqe *io_get_cqring(struct io_ring_ctx *ctx) |
1288 | { | |
75b28aff | 1289 | struct io_rings *rings = ctx->rings; |
2b188cc1 JA |
1290 | unsigned tail; |
1291 | ||
1292 | tail = ctx->cached_cq_tail; | |
115e12e5 SB |
1293 | /* |
1294 | * writes to the cq entry need to come after reading head; the | |
1295 | * control dependency is enough as we're using WRITE_ONCE to | |
1296 | * fill the cq entry | |
1297 | */ | |
75b28aff | 1298 | if (tail - READ_ONCE(rings->cq.head) == rings->cq_ring_entries) |
2b188cc1 JA |
1299 | return NULL; |
1300 | ||
1301 | ctx->cached_cq_tail++; | |
75b28aff | 1302 | return &rings->cqes[tail & ctx->cq_mask]; |
2b188cc1 JA |
1303 | } |
1304 | ||
f2842ab5 JA |
1305 | static inline bool io_should_trigger_evfd(struct io_ring_ctx *ctx) |
1306 | { | |
f0b493e6 JA |
1307 | if (!ctx->cq_ev_fd) |
1308 | return false; | |
7e55a19c SG |
1309 | if (READ_ONCE(ctx->rings->cq_flags) & IORING_CQ_EVENTFD_DISABLED) |
1310 | return false; | |
f2842ab5 JA |
1311 | if (!ctx->eventfd_async) |
1312 | return true; | |
b41e9852 | 1313 | return io_wq_current_is_worker(); |
f2842ab5 JA |
1314 | } |
1315 | ||
b41e9852 | 1316 | static void io_cqring_ev_posted(struct io_ring_ctx *ctx) |
1d7bb1d5 JA |
1317 | { |
1318 | if (waitqueue_active(&ctx->wait)) | |
1319 | wake_up(&ctx->wait); | |
1320 | if (waitqueue_active(&ctx->sqo_wait)) | |
1321 | wake_up(&ctx->sqo_wait); | |
b41e9852 | 1322 | if (io_should_trigger_evfd(ctx)) |
1d7bb1d5 JA |
1323 | eventfd_signal(ctx->cq_ev_fd, 1); |
1324 | } | |
1325 | ||
46930143 PB |
1326 | static void io_cqring_mark_overflow(struct io_ring_ctx *ctx) |
1327 | { | |
1328 | if (list_empty(&ctx->cq_overflow_list)) { | |
1329 | clear_bit(0, &ctx->sq_check_overflow); | |
1330 | clear_bit(0, &ctx->cq_check_overflow); | |
1331 | ctx->rings->sq_flags &= ~IORING_SQ_CQ_OVERFLOW; | |
1332 | } | |
1333 | } | |
1334 | ||
c4a2ed72 JA |
1335 | /* Returns true if there are no backlogged entries after the flush */ |
1336 | static bool io_cqring_overflow_flush(struct io_ring_ctx *ctx, bool force) | |
1d7bb1d5 JA |
1337 | { |
1338 | struct io_rings *rings = ctx->rings; | |
1339 | struct io_uring_cqe *cqe; | |
1340 | struct io_kiocb *req; | |
1341 | unsigned long flags; | |
1342 | LIST_HEAD(list); | |
1343 | ||
1344 | if (!force) { | |
1345 | if (list_empty_careful(&ctx->cq_overflow_list)) | |
c4a2ed72 | 1346 | return true; |
1d7bb1d5 JA |
1347 | if ((ctx->cached_cq_tail - READ_ONCE(rings->cq.head) == |
1348 | rings->cq_ring_entries)) | |
c4a2ed72 | 1349 | return false; |
1d7bb1d5 JA |
1350 | } |
1351 | ||
1352 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
1353 | ||
1354 | /* if force is set, the ring is going away. always drop after that */ | |
1355 | if (force) | |
69b3e546 | 1356 | ctx->cq_overflow_flushed = 1; |
1d7bb1d5 | 1357 | |
c4a2ed72 | 1358 | cqe = NULL; |
1d7bb1d5 JA |
1359 | while (!list_empty(&ctx->cq_overflow_list)) { |
1360 | cqe = io_get_cqring(ctx); | |
1361 | if (!cqe && !force) | |
1362 | break; | |
1363 | ||
1364 | req = list_first_entry(&ctx->cq_overflow_list, struct io_kiocb, | |
40d8ddd4 PB |
1365 | compl.list); |
1366 | list_move(&req->compl.list, &list); | |
1d7bb1d5 JA |
1367 | if (cqe) { |
1368 | WRITE_ONCE(cqe->user_data, req->user_data); | |
1369 | WRITE_ONCE(cqe->res, req->result); | |
0f7e466b | 1370 | WRITE_ONCE(cqe->flags, req->compl.cflags); |
1d7bb1d5 JA |
1371 | } else { |
1372 | WRITE_ONCE(ctx->rings->cq_overflow, | |
1373 | atomic_inc_return(&ctx->cached_cq_overflow)); | |
1374 | } | |
1375 | } | |
1376 | ||
1377 | io_commit_cqring(ctx); | |
46930143 PB |
1378 | io_cqring_mark_overflow(ctx); |
1379 | ||
1d7bb1d5 JA |
1380 | spin_unlock_irqrestore(&ctx->completion_lock, flags); |
1381 | io_cqring_ev_posted(ctx); | |
1382 | ||
1383 | while (!list_empty(&list)) { | |
40d8ddd4 PB |
1384 | req = list_first_entry(&list, struct io_kiocb, compl.list); |
1385 | list_del(&req->compl.list); | |
ec9c02ad | 1386 | io_put_req(req); |
1d7bb1d5 | 1387 | } |
c4a2ed72 JA |
1388 | |
1389 | return cqe != NULL; | |
1d7bb1d5 JA |
1390 | } |
1391 | ||
bcda7baa | 1392 | static void __io_cqring_fill_event(struct io_kiocb *req, long res, long cflags) |
2b188cc1 | 1393 | { |
78e19bbe | 1394 | struct io_ring_ctx *ctx = req->ctx; |
2b188cc1 JA |
1395 | struct io_uring_cqe *cqe; |
1396 | ||
78e19bbe | 1397 | trace_io_uring_complete(ctx, req->user_data, res); |
51c3ff62 | 1398 | |
2b188cc1 JA |
1399 | /* |
1400 | * If we can't get a cq entry, userspace overflowed the | |
1401 | * submission (by quite a lot). Increment the overflow count in | |
1402 | * the ring. | |
1403 | */ | |
1404 | cqe = io_get_cqring(ctx); | |
1d7bb1d5 | 1405 | if (likely(cqe)) { |
78e19bbe | 1406 | WRITE_ONCE(cqe->user_data, req->user_data); |
2b188cc1 | 1407 | WRITE_ONCE(cqe->res, res); |
bcda7baa | 1408 | WRITE_ONCE(cqe->flags, cflags); |
1d7bb1d5 | 1409 | } else if (ctx->cq_overflow_flushed) { |
498ccd9e JA |
1410 | WRITE_ONCE(ctx->rings->cq_overflow, |
1411 | atomic_inc_return(&ctx->cached_cq_overflow)); | |
1d7bb1d5 | 1412 | } else { |
ad3eb2c8 JA |
1413 | if (list_empty(&ctx->cq_overflow_list)) { |
1414 | set_bit(0, &ctx->sq_check_overflow); | |
1415 | set_bit(0, &ctx->cq_check_overflow); | |
6d5f9049 | 1416 | ctx->rings->sq_flags |= IORING_SQ_CQ_OVERFLOW; |
ad3eb2c8 | 1417 | } |
40d8ddd4 | 1418 | io_clean_op(req); |
1d7bb1d5 | 1419 | req->result = res; |
0f7e466b | 1420 | req->compl.cflags = cflags; |
40d8ddd4 PB |
1421 | refcount_inc(&req->refs); |
1422 | list_add_tail(&req->compl.list, &ctx->cq_overflow_list); | |
2b188cc1 JA |
1423 | } |
1424 | } | |
1425 | ||
bcda7baa JA |
1426 | static void io_cqring_fill_event(struct io_kiocb *req, long res) |
1427 | { | |
1428 | __io_cqring_fill_event(req, res, 0); | |
1429 | } | |
1430 | ||
e1e16097 | 1431 | static void io_cqring_add_event(struct io_kiocb *req, long res, long cflags) |
2b188cc1 | 1432 | { |
78e19bbe | 1433 | struct io_ring_ctx *ctx = req->ctx; |
2b188cc1 JA |
1434 | unsigned long flags; |
1435 | ||
1436 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
bcda7baa | 1437 | __io_cqring_fill_event(req, res, cflags); |
2b188cc1 JA |
1438 | io_commit_cqring(ctx); |
1439 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
1440 | ||
8c838788 | 1441 | io_cqring_ev_posted(ctx); |
2b188cc1 JA |
1442 | } |
1443 | ||
229a7b63 | 1444 | static void io_submit_flush_completions(struct io_comp_state *cs) |
bcda7baa | 1445 | { |
229a7b63 JA |
1446 | struct io_ring_ctx *ctx = cs->ctx; |
1447 | ||
1448 | spin_lock_irq(&ctx->completion_lock); | |
1449 | while (!list_empty(&cs->list)) { | |
1450 | struct io_kiocb *req; | |
1451 | ||
3ca405eb PB |
1452 | req = list_first_entry(&cs->list, struct io_kiocb, compl.list); |
1453 | list_del(&req->compl.list); | |
0f7e466b | 1454 | __io_cqring_fill_event(req, req->result, req->compl.cflags); |
229a7b63 JA |
1455 | if (!(req->flags & REQ_F_LINK_HEAD)) { |
1456 | req->flags |= REQ_F_COMP_LOCKED; | |
1457 | io_put_req(req); | |
1458 | } else { | |
1459 | spin_unlock_irq(&ctx->completion_lock); | |
1460 | io_put_req(req); | |
1461 | spin_lock_irq(&ctx->completion_lock); | |
1462 | } | |
1463 | } | |
1464 | io_commit_cqring(ctx); | |
1465 | spin_unlock_irq(&ctx->completion_lock); | |
1466 | ||
1467 | io_cqring_ev_posted(ctx); | |
1468 | cs->nr = 0; | |
1469 | } | |
1470 | ||
1471 | static void __io_req_complete(struct io_kiocb *req, long res, unsigned cflags, | |
1472 | struct io_comp_state *cs) | |
1473 | { | |
1474 | if (!cs) { | |
1475 | io_cqring_add_event(req, res, cflags); | |
1476 | io_put_req(req); | |
1477 | } else { | |
3ca405eb | 1478 | io_clean_op(req); |
229a7b63 | 1479 | req->result = res; |
0f7e466b | 1480 | req->compl.cflags = cflags; |
3ca405eb | 1481 | list_add_tail(&req->compl.list, &cs->list); |
229a7b63 JA |
1482 | if (++cs->nr >= 32) |
1483 | io_submit_flush_completions(cs); | |
1484 | } | |
e1e16097 JA |
1485 | } |
1486 | ||
1487 | static void io_req_complete(struct io_kiocb *req, long res) | |
bcda7baa | 1488 | { |
229a7b63 | 1489 | __io_req_complete(req, res, 0, NULL); |
bcda7baa JA |
1490 | } |
1491 | ||
0ddf92e8 JA |
1492 | static inline bool io_is_fallback_req(struct io_kiocb *req) |
1493 | { | |
1494 | return req == (struct io_kiocb *) | |
1495 | ((unsigned long) req->ctx->fallback_req & ~1UL); | |
1496 | } | |
1497 | ||
1498 | static struct io_kiocb *io_get_fallback_req(struct io_ring_ctx *ctx) | |
1499 | { | |
1500 | struct io_kiocb *req; | |
1501 | ||
1502 | req = ctx->fallback_req; | |
dd461af6 | 1503 | if (!test_and_set_bit_lock(0, (unsigned long *) &ctx->fallback_req)) |
0ddf92e8 JA |
1504 | return req; |
1505 | ||
1506 | return NULL; | |
1507 | } | |
1508 | ||
0553b8bd PB |
1509 | static struct io_kiocb *io_alloc_req(struct io_ring_ctx *ctx, |
1510 | struct io_submit_state *state) | |
2b188cc1 | 1511 | { |
fd6fab2c | 1512 | gfp_t gfp = GFP_KERNEL | __GFP_NOWARN; |
2b188cc1 JA |
1513 | struct io_kiocb *req; |
1514 | ||
f6b6c7d6 | 1515 | if (!state->free_reqs) { |
2579f913 JA |
1516 | size_t sz; |
1517 | int ret; | |
1518 | ||
1519 | sz = min_t(size_t, state->ios_left, ARRAY_SIZE(state->reqs)); | |
fd6fab2c JA |
1520 | ret = kmem_cache_alloc_bulk(req_cachep, gfp, sz, state->reqs); |
1521 | ||
1522 | /* | |
1523 | * Bulk alloc is all-or-nothing. If we fail to get a batch, | |
1524 | * retry single alloc to be on the safe side. | |
1525 | */ | |
1526 | if (unlikely(ret <= 0)) { | |
1527 | state->reqs[0] = kmem_cache_alloc(req_cachep, gfp); | |
1528 | if (!state->reqs[0]) | |
0ddf92e8 | 1529 | goto fallback; |
fd6fab2c JA |
1530 | ret = 1; |
1531 | } | |
2579f913 | 1532 | state->free_reqs = ret - 1; |
6c8a3134 | 1533 | req = state->reqs[ret - 1]; |
2579f913 | 1534 | } else { |
2579f913 | 1535 | state->free_reqs--; |
6c8a3134 | 1536 | req = state->reqs[state->free_reqs]; |
2b188cc1 JA |
1537 | } |
1538 | ||
2579f913 | 1539 | return req; |
0ddf92e8 | 1540 | fallback: |
0553b8bd | 1541 | return io_get_fallback_req(ctx); |
2b188cc1 JA |
1542 | } |
1543 | ||
8da11c19 PB |
1544 | static inline void io_put_file(struct io_kiocb *req, struct file *file, |
1545 | bool fixed) | |
1546 | { | |
1547 | if (fixed) | |
05589553 | 1548 | percpu_ref_put(req->fixed_file_refs); |
8da11c19 PB |
1549 | else |
1550 | fput(file); | |
1551 | } | |
1552 | ||
51a4cc11 | 1553 | static bool io_dismantle_req(struct io_kiocb *req) |
2b188cc1 | 1554 | { |
3ca405eb | 1555 | io_clean_op(req); |
929a3af9 | 1556 | |
5acbbc8e JA |
1557 | if (req->io) |
1558 | kfree(req->io); | |
8da11c19 PB |
1559 | if (req->file) |
1560 | io_put_file(req, req->file, (req->flags & REQ_F_FIXED_FILE)); | |
fcb323cc | 1561 | |
51a4cc11 | 1562 | return io_req_clean_work(req); |
e65ef56d JA |
1563 | } |
1564 | ||
51a4cc11 | 1565 | static void __io_free_req_finish(struct io_kiocb *req) |
c6ca97b3 | 1566 | { |
51a4cc11 | 1567 | struct io_ring_ctx *ctx = req->ctx; |
c6ca97b3 | 1568 | |
5af1d13e | 1569 | __io_put_req_task(req); |
b1e50e54 PB |
1570 | if (likely(!io_is_fallback_req(req))) |
1571 | kmem_cache_free(req_cachep, req); | |
1572 | else | |
ecfc5177 PB |
1573 | clear_bit_unlock(0, (unsigned long *) &ctx->fallback_req); |
1574 | percpu_ref_put(&ctx->refs); | |
e65ef56d JA |
1575 | } |
1576 | ||
51a4cc11 JA |
1577 | static void io_req_task_file_table_put(struct callback_head *cb) |
1578 | { | |
1579 | struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work); | |
1580 | struct fs_struct *fs = req->work.fs; | |
1581 | ||
1582 | spin_lock(&req->work.fs->lock); | |
1583 | if (--fs->users) | |
1584 | fs = NULL; | |
1585 | spin_unlock(&req->work.fs->lock); | |
1586 | if (fs) | |
1587 | free_fs_struct(fs); | |
1588 | req->work.fs = NULL; | |
1589 | __io_free_req_finish(req); | |
1590 | } | |
1591 | ||
1592 | static void __io_free_req(struct io_kiocb *req) | |
1593 | { | |
1594 | if (!io_dismantle_req(req)) { | |
1595 | __io_free_req_finish(req); | |
1596 | } else { | |
1597 | int ret; | |
1598 | ||
1599 | init_task_work(&req->task_work, io_req_task_file_table_put); | |
1600 | ret = task_work_add(req->task, &req->task_work, TWA_RESUME); | |
1601 | if (unlikely(ret)) { | |
1602 | struct task_struct *tsk; | |
1603 | ||
1604 | tsk = io_wq_get_task(req->ctx->io_wq); | |
1605 | task_work_add(tsk, &req->task_work, 0); | |
1606 | } | |
1607 | } | |
1608 | } | |
1609 | ||
a197f664 | 1610 | static bool io_link_cancel_timeout(struct io_kiocb *req) |
2665abfd | 1611 | { |
a197f664 | 1612 | struct io_ring_ctx *ctx = req->ctx; |
2665abfd JA |
1613 | int ret; |
1614 | ||
2d28390a | 1615 | ret = hrtimer_try_to_cancel(&req->io->timeout.timer); |
2665abfd | 1616 | if (ret != -1) { |
78e19bbe | 1617 | io_cqring_fill_event(req, -ECANCELED); |
2665abfd | 1618 | io_commit_cqring(ctx); |
dea3b49c | 1619 | req->flags &= ~REQ_F_LINK_HEAD; |
ec9c02ad | 1620 | io_put_req(req); |
2665abfd JA |
1621 | return true; |
1622 | } | |
1623 | ||
1624 | return false; | |
e65ef56d JA |
1625 | } |
1626 | ||
ab0b6451 | 1627 | static bool __io_kill_linked_timeout(struct io_kiocb *req) |
9e645e11 | 1628 | { |
7c86ffee | 1629 | struct io_kiocb *link; |
ab0b6451 | 1630 | bool wake_ev; |
7c86ffee PB |
1631 | |
1632 | if (list_empty(&req->link_list)) | |
ab0b6451 | 1633 | return false; |
7c86ffee PB |
1634 | link = list_first_entry(&req->link_list, struct io_kiocb, link_list); |
1635 | if (link->opcode != IORING_OP_LINK_TIMEOUT) | |
ab0b6451 | 1636 | return false; |
7c86ffee PB |
1637 | |
1638 | list_del_init(&link->link_list); | |
9b7adba9 | 1639 | link->flags |= REQ_F_COMP_LOCKED; |
7c86ffee PB |
1640 | wake_ev = io_link_cancel_timeout(link); |
1641 | req->flags &= ~REQ_F_LINK_TIMEOUT; | |
ab0b6451 JA |
1642 | return wake_ev; |
1643 | } | |
1644 | ||
1645 | static void io_kill_linked_timeout(struct io_kiocb *req) | |
9e645e11 | 1646 | { |
2665abfd | 1647 | struct io_ring_ctx *ctx = req->ctx; |
ab0b6451 | 1648 | bool wake_ev; |
9e645e11 | 1649 | |
ab0b6451 JA |
1650 | if (!(req->flags & REQ_F_COMP_LOCKED)) { |
1651 | unsigned long flags; | |
1652 | ||
1653 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
1654 | wake_ev = __io_kill_linked_timeout(req); | |
7c86ffee | 1655 | spin_unlock_irqrestore(&ctx->completion_lock, flags); |
ab0b6451 JA |
1656 | } else { |
1657 | wake_ev = __io_kill_linked_timeout(req); | |
1658 | } | |
1659 | ||
7c86ffee PB |
1660 | if (wake_ev) |
1661 | io_cqring_ev_posted(ctx); | |
1662 | } | |
1663 | ||
9b5f7bd9 | 1664 | static struct io_kiocb *io_req_link_next(struct io_kiocb *req) |
7c86ffee PB |
1665 | { |
1666 | struct io_kiocb *nxt; | |
4d7dd462 | 1667 | |
9e645e11 JA |
1668 | /* |
1669 | * The list should never be empty when we are called here. But could | |
1670 | * potentially happen if the chain is messed up, check to be on the | |
1671 | * safe side. | |
1672 | */ | |
7c86ffee | 1673 | if (unlikely(list_empty(&req->link_list))) |
9b5f7bd9 | 1674 | return NULL; |
2665abfd | 1675 | |
7c86ffee PB |
1676 | nxt = list_first_entry(&req->link_list, struct io_kiocb, link_list); |
1677 | list_del_init(&req->link_list); | |
1678 | if (!list_empty(&nxt->link_list)) | |
1679 | nxt->flags |= REQ_F_LINK_HEAD; | |
9b5f7bd9 | 1680 | return nxt; |
9e645e11 JA |
1681 | } |
1682 | ||
1683 | /* | |
dea3b49c | 1684 | * Called if REQ_F_LINK_HEAD is set, and we fail the head request |
9e645e11 | 1685 | */ |
7c86ffee | 1686 | static void __io_fail_links(struct io_kiocb *req) |
9e645e11 | 1687 | { |
2665abfd | 1688 | struct io_ring_ctx *ctx = req->ctx; |
9e645e11 JA |
1689 | |
1690 | while (!list_empty(&req->link_list)) { | |
4493233e PB |
1691 | struct io_kiocb *link = list_first_entry(&req->link_list, |
1692 | struct io_kiocb, link_list); | |
9e645e11 | 1693 | |
4493233e | 1694 | list_del_init(&link->link_list); |
c826bd7a | 1695 | trace_io_uring_fail_link(req, link); |
2665abfd | 1696 | |
7c86ffee | 1697 | io_cqring_fill_event(link, -ECANCELED); |
9b7adba9 | 1698 | link->flags |= REQ_F_COMP_LOCKED; |
7c86ffee | 1699 | __io_double_put_req(link); |
5d960724 | 1700 | req->flags &= ~REQ_F_LINK_TIMEOUT; |
9e645e11 | 1701 | } |
2665abfd JA |
1702 | |
1703 | io_commit_cqring(ctx); | |
2665abfd | 1704 | io_cqring_ev_posted(ctx); |
9e645e11 JA |
1705 | } |
1706 | ||
7c86ffee | 1707 | static void io_fail_links(struct io_kiocb *req) |
9e645e11 | 1708 | { |
7c86ffee | 1709 | struct io_ring_ctx *ctx = req->ctx; |
2665abfd | 1710 | |
7c86ffee | 1711 | if (!(req->flags & REQ_F_COMP_LOCKED)) { |
2665abfd JA |
1712 | unsigned long flags; |
1713 | ||
2665abfd | 1714 | spin_lock_irqsave(&ctx->completion_lock, flags); |
7c86ffee | 1715 | __io_fail_links(req); |
2665abfd JA |
1716 | spin_unlock_irqrestore(&ctx->completion_lock, flags); |
1717 | } else { | |
7c86ffee | 1718 | __io_fail_links(req); |
9e645e11 JA |
1719 | } |
1720 | ||
2665abfd | 1721 | io_cqring_ev_posted(ctx); |
9e645e11 JA |
1722 | } |
1723 | ||
3fa5e0f3 | 1724 | static struct io_kiocb *__io_req_find_next(struct io_kiocb *req) |
c69f8dbe | 1725 | { |
9b0d911a | 1726 | req->flags &= ~REQ_F_LINK_HEAD; |
7c86ffee PB |
1727 | if (req->flags & REQ_F_LINK_TIMEOUT) |
1728 | io_kill_linked_timeout(req); | |
944e58bf | 1729 | |
9e645e11 JA |
1730 | /* |
1731 | * If LINK is set, we have dependent requests in this chain. If we | |
1732 | * didn't fail this request, queue the first one up, moving any other | |
1733 | * dependencies to the next request. In case of failure, fail the rest | |
1734 | * of the chain. | |
1735 | */ | |
9b5f7bd9 PB |
1736 | if (likely(!(req->flags & REQ_F_FAIL_LINK))) |
1737 | return io_req_link_next(req); | |
1738 | io_fail_links(req); | |
1739 | return NULL; | |
4d7dd462 | 1740 | } |
9e645e11 | 1741 | |
3fa5e0f3 PB |
1742 | static struct io_kiocb *io_req_find_next(struct io_kiocb *req) |
1743 | { | |
1744 | if (likely(!(req->flags & REQ_F_LINK_HEAD))) | |
1745 | return NULL; | |
1746 | return __io_req_find_next(req); | |
1747 | } | |
1748 | ||
fd7d6de2 JA |
1749 | static int io_req_task_work_add(struct io_kiocb *req, struct callback_head *cb, |
1750 | bool twa_signal_ok) | |
c2c4c83c JA |
1751 | { |
1752 | struct task_struct *tsk = req->task; | |
1753 | struct io_ring_ctx *ctx = req->ctx; | |
0ba9c9ed | 1754 | int ret, notify; |
c2c4c83c JA |
1755 | |
1756 | /* | |
0ba9c9ed JA |
1757 | * SQPOLL kernel thread doesn't need notification, just a wakeup. For |
1758 | * all other cases, use TWA_SIGNAL unconditionally to ensure we're | |
1759 | * processing task_work. There's no reliable way to tell if TWA_RESUME | |
1760 | * will do the job. | |
c2c4c83c | 1761 | */ |
0ba9c9ed | 1762 | notify = 0; |
fd7d6de2 | 1763 | if (!(ctx->flags & IORING_SETUP_SQPOLL) && twa_signal_ok) |
c2c4c83c JA |
1764 | notify = TWA_SIGNAL; |
1765 | ||
1766 | ret = task_work_add(tsk, cb, notify); | |
1767 | if (!ret) | |
1768 | wake_up_process(tsk); | |
0ba9c9ed | 1769 | |
c2c4c83c JA |
1770 | return ret; |
1771 | } | |
1772 | ||
c40f6379 JA |
1773 | static void __io_req_task_cancel(struct io_kiocb *req, int error) |
1774 | { | |
1775 | struct io_ring_ctx *ctx = req->ctx; | |
1776 | ||
1777 | spin_lock_irq(&ctx->completion_lock); | |
1778 | io_cqring_fill_event(req, error); | |
1779 | io_commit_cqring(ctx); | |
1780 | spin_unlock_irq(&ctx->completion_lock); | |
1781 | ||
1782 | io_cqring_ev_posted(ctx); | |
1783 | req_set_fail_links(req); | |
1784 | io_double_put_req(req); | |
1785 | } | |
1786 | ||
1787 | static void io_req_task_cancel(struct callback_head *cb) | |
1788 | { | |
1789 | struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work); | |
1790 | ||
1791 | __io_req_task_cancel(req, -ECANCELED); | |
1792 | } | |
1793 | ||
1794 | static void __io_req_task_submit(struct io_kiocb *req) | |
1795 | { | |
1796 | struct io_ring_ctx *ctx = req->ctx; | |
1797 | ||
c40f6379 JA |
1798 | if (!__io_sq_thread_acquire_mm(ctx)) { |
1799 | mutex_lock(&ctx->uring_lock); | |
1800 | __io_queue_sqe(req, NULL, NULL); | |
1801 | mutex_unlock(&ctx->uring_lock); | |
1802 | } else { | |
1803 | __io_req_task_cancel(req, -EFAULT); | |
1804 | } | |
1805 | } | |
1806 | ||
1807 | static void io_req_task_submit(struct callback_head *cb) | |
1808 | { | |
1809 | struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work); | |
6d816e08 | 1810 | struct io_ring_ctx *ctx = req->ctx; |
c40f6379 JA |
1811 | |
1812 | __io_req_task_submit(req); | |
6d816e08 | 1813 | percpu_ref_put(&ctx->refs); |
c40f6379 JA |
1814 | } |
1815 | ||
1816 | static void io_req_task_queue(struct io_kiocb *req) | |
1817 | { | |
c40f6379 JA |
1818 | int ret; |
1819 | ||
1820 | init_task_work(&req->task_work, io_req_task_submit); | |
6d816e08 | 1821 | percpu_ref_get(&req->ctx->refs); |
c40f6379 | 1822 | |
fd7d6de2 | 1823 | ret = io_req_task_work_add(req, &req->task_work, true); |
c40f6379 | 1824 | if (unlikely(ret)) { |
c2c4c83c JA |
1825 | struct task_struct *tsk; |
1826 | ||
c40f6379 JA |
1827 | init_task_work(&req->task_work, io_req_task_cancel); |
1828 | tsk = io_wq_get_task(req->ctx->io_wq); | |
c2c4c83c JA |
1829 | task_work_add(tsk, &req->task_work, 0); |
1830 | wake_up_process(tsk); | |
c40f6379 | 1831 | } |
c40f6379 JA |
1832 | } |
1833 | ||
c3524383 | 1834 | static void io_queue_next(struct io_kiocb *req) |
c69f8dbe | 1835 | { |
9b5f7bd9 | 1836 | struct io_kiocb *nxt = io_req_find_next(req); |
944e58bf PB |
1837 | |
1838 | if (nxt) | |
906a8c3f | 1839 | io_req_task_queue(nxt); |
c69f8dbe JL |
1840 | } |
1841 | ||
c3524383 | 1842 | static void io_free_req(struct io_kiocb *req) |
7a743e22 | 1843 | { |
c3524383 PB |
1844 | io_queue_next(req); |
1845 | __io_free_req(req); | |
1846 | } | |
8766dd51 | 1847 | |
2d6500d4 PB |
1848 | struct req_batch { |
1849 | void *reqs[IO_IOPOLL_BATCH]; | |
1850 | int to_free; | |
7a743e22 | 1851 | |
5af1d13e PB |
1852 | struct task_struct *task; |
1853 | int task_refs; | |
2d6500d4 PB |
1854 | }; |
1855 | ||
5af1d13e PB |
1856 | static inline void io_init_req_batch(struct req_batch *rb) |
1857 | { | |
1858 | rb->to_free = 0; | |
1859 | rb->task_refs = 0; | |
1860 | rb->task = NULL; | |
1861 | } | |
1862 | ||
2d6500d4 PB |
1863 | static void __io_req_free_batch_flush(struct io_ring_ctx *ctx, |
1864 | struct req_batch *rb) | |
1865 | { | |
1866 | kmem_cache_free_bulk(req_cachep, rb->to_free, rb->reqs); | |
1867 | percpu_ref_put_many(&ctx->refs, rb->to_free); | |
1868 | rb->to_free = 0; | |
1869 | } | |
1870 | ||
1871 | static void io_req_free_batch_finish(struct io_ring_ctx *ctx, | |
1872 | struct req_batch *rb) | |
1873 | { | |
1874 | if (rb->to_free) | |
1875 | __io_req_free_batch_flush(ctx, rb); | |
5af1d13e PB |
1876 | if (rb->task) { |
1877 | put_task_struct_many(rb->task, rb->task_refs); | |
1878 | rb->task = NULL; | |
1879 | } | |
2d6500d4 PB |
1880 | } |
1881 | ||
1882 | static void io_req_free_batch(struct req_batch *rb, struct io_kiocb *req) | |
1883 | { | |
1884 | if (unlikely(io_is_fallback_req(req))) { | |
1885 | io_free_req(req); | |
1886 | return; | |
1887 | } | |
1888 | if (req->flags & REQ_F_LINK_HEAD) | |
1889 | io_queue_next(req); | |
1890 | ||
5af1d13e PB |
1891 | if (req->flags & REQ_F_TASK_PINNED) { |
1892 | if (req->task != rb->task) { | |
1893 | if (rb->task) | |
1894 | put_task_struct_many(rb->task, rb->task_refs); | |
1895 | rb->task = req->task; | |
1896 | rb->task_refs = 0; | |
1897 | } | |
1898 | rb->task_refs++; | |
1899 | req->flags &= ~REQ_F_TASK_PINNED; | |
1900 | } | |
1901 | ||
51a4cc11 | 1902 | WARN_ON_ONCE(io_dismantle_req(req)); |
2d6500d4 PB |
1903 | rb->reqs[rb->to_free++] = req; |
1904 | if (unlikely(rb->to_free == ARRAY_SIZE(rb->reqs))) | |
1905 | __io_req_free_batch_flush(req->ctx, rb); | |
7a743e22 PB |
1906 | } |
1907 | ||
ba816ad6 JA |
1908 | /* |
1909 | * Drop reference to request, return next in chain (if there is one) if this | |
1910 | * was the last reference to this request. | |
1911 | */ | |
9b5f7bd9 | 1912 | static struct io_kiocb *io_put_req_find_next(struct io_kiocb *req) |
e65ef56d | 1913 | { |
9b5f7bd9 PB |
1914 | struct io_kiocb *nxt = NULL; |
1915 | ||
2a44f467 | 1916 | if (refcount_dec_and_test(&req->refs)) { |
9b5f7bd9 | 1917 | nxt = io_req_find_next(req); |
4d7dd462 | 1918 | __io_free_req(req); |
2a44f467 | 1919 | } |
9b5f7bd9 | 1920 | return nxt; |
2b188cc1 JA |
1921 | } |
1922 | ||
e65ef56d JA |
1923 | static void io_put_req(struct io_kiocb *req) |
1924 | { | |
1925 | if (refcount_dec_and_test(&req->refs)) | |
1926 | io_free_req(req); | |
2b188cc1 JA |
1927 | } |
1928 | ||
f4db7182 | 1929 | static struct io_wq_work *io_steal_work(struct io_kiocb *req) |
7a743e22 | 1930 | { |
6df1db6b | 1931 | struct io_kiocb *nxt; |
f4db7182 | 1932 | |
7a743e22 | 1933 | /* |
f4db7182 PB |
1934 | * A ref is owned by io-wq in which context we're. So, if that's the |
1935 | * last one, it's safe to steal next work. False negatives are Ok, | |
1936 | * it just will be re-punted async in io_put_work() | |
7a743e22 | 1937 | */ |
f4db7182 PB |
1938 | if (refcount_read(&req->refs) != 1) |
1939 | return NULL; | |
7a743e22 | 1940 | |
9b5f7bd9 | 1941 | nxt = io_req_find_next(req); |
6df1db6b | 1942 | return nxt ? &nxt->work : NULL; |
7a743e22 PB |
1943 | } |
1944 | ||
978db57e JA |
1945 | /* |
1946 | * Must only be used if we don't need to care about links, usually from | |
1947 | * within the completion handling itself. | |
1948 | */ | |
1949 | static void __io_double_put_req(struct io_kiocb *req) | |
78e19bbe JA |
1950 | { |
1951 | /* drop both submit and complete references */ | |
1952 | if (refcount_sub_and_test(2, &req->refs)) | |
1953 | __io_free_req(req); | |
1954 | } | |
1955 | ||
978db57e JA |
1956 | static void io_double_put_req(struct io_kiocb *req) |
1957 | { | |
1958 | /* drop both submit and complete references */ | |
1959 | if (refcount_sub_and_test(2, &req->refs)) | |
1960 | io_free_req(req); | |
1961 | } | |
1962 | ||
1d7bb1d5 | 1963 | static unsigned io_cqring_events(struct io_ring_ctx *ctx, bool noflush) |
a3a0e43f | 1964 | { |
84f97dc2 JA |
1965 | struct io_rings *rings = ctx->rings; |
1966 | ||
ad3eb2c8 JA |
1967 | if (test_bit(0, &ctx->cq_check_overflow)) { |
1968 | /* | |
1969 | * noflush == true is from the waitqueue handler, just ensure | |
1970 | * we wake up the task, and the next invocation will flush the | |
1971 | * entries. We cannot safely to it from here. | |
1972 | */ | |
1973 | if (noflush && !list_empty(&ctx->cq_overflow_list)) | |
1974 | return -1U; | |
1d7bb1d5 | 1975 | |
ad3eb2c8 JA |
1976 | io_cqring_overflow_flush(ctx, false); |
1977 | } | |
1d7bb1d5 | 1978 | |
a3a0e43f JA |
1979 | /* See comment at the top of this file */ |
1980 | smp_rmb(); | |
ad3eb2c8 | 1981 | return ctx->cached_cq_tail - READ_ONCE(rings->cq.head); |
a3a0e43f JA |
1982 | } |
1983 | ||
fb5ccc98 PB |
1984 | static inline unsigned int io_sqring_entries(struct io_ring_ctx *ctx) |
1985 | { | |
1986 | struct io_rings *rings = ctx->rings; | |
1987 | ||
1988 | /* make sure SQ entry isn't read before tail */ | |
1989 | return smp_load_acquire(&rings->sq.tail) - ctx->cached_sq_head; | |
1990 | } | |
1991 | ||
8ff069bf | 1992 | static unsigned int io_put_kbuf(struct io_kiocb *req, struct io_buffer *kbuf) |
e94f141b | 1993 | { |
8ff069bf | 1994 | unsigned int cflags; |
e94f141b | 1995 | |
bcda7baa JA |
1996 | cflags = kbuf->bid << IORING_CQE_BUFFER_SHIFT; |
1997 | cflags |= IORING_CQE_F_BUFFER; | |
0e1b6fe3 | 1998 | req->flags &= ~REQ_F_BUFFER_SELECTED; |
bcda7baa JA |
1999 | kfree(kbuf); |
2000 | return cflags; | |
e94f141b JA |
2001 | } |
2002 | ||
8ff069bf | 2003 | static inline unsigned int io_put_rw_kbuf(struct io_kiocb *req) |
bcda7baa | 2004 | { |
4d954c25 | 2005 | struct io_buffer *kbuf; |
bcda7baa | 2006 | |
4d954c25 | 2007 | kbuf = (struct io_buffer *) (unsigned long) req->rw.addr; |
8ff069bf PB |
2008 | return io_put_kbuf(req, kbuf); |
2009 | } | |
2010 | ||
4c6e277c JA |
2011 | static inline bool io_run_task_work(void) |
2012 | { | |
2013 | if (current->task_works) { | |
2014 | __set_current_state(TASK_RUNNING); | |
2015 | task_work_run(); | |
2016 | return true; | |
2017 | } | |
2018 | ||
2019 | return false; | |
bcda7baa JA |
2020 | } |
2021 | ||
bbde017a XW |
2022 | static void io_iopoll_queue(struct list_head *again) |
2023 | { | |
2024 | struct io_kiocb *req; | |
2025 | ||
2026 | do { | |
d21ffe7e PB |
2027 | req = list_first_entry(again, struct io_kiocb, inflight_entry); |
2028 | list_del(&req->inflight_entry); | |
81b68a5c | 2029 | __io_complete_rw(req, -EAGAIN, 0, NULL); |
bbde017a XW |
2030 | } while (!list_empty(again)); |
2031 | } | |
2032 | ||
def596e9 JA |
2033 | /* |
2034 | * Find and free completed poll iocbs | |
2035 | */ | |
2036 | static void io_iopoll_complete(struct io_ring_ctx *ctx, unsigned int *nr_events, | |
2037 | struct list_head *done) | |
2038 | { | |
8237e045 | 2039 | struct req_batch rb; |
def596e9 | 2040 | struct io_kiocb *req; |
bbde017a XW |
2041 | LIST_HEAD(again); |
2042 | ||
2043 | /* order with ->result store in io_complete_rw_iopoll() */ | |
2044 | smp_rmb(); | |
def596e9 | 2045 | |
5af1d13e | 2046 | io_init_req_batch(&rb); |
def596e9 | 2047 | while (!list_empty(done)) { |
bcda7baa JA |
2048 | int cflags = 0; |
2049 | ||
d21ffe7e | 2050 | req = list_first_entry(done, struct io_kiocb, inflight_entry); |
bbde017a | 2051 | if (READ_ONCE(req->result) == -EAGAIN) { |
56450c20 | 2052 | req->result = 0; |
bbde017a | 2053 | req->iopoll_completed = 0; |
d21ffe7e | 2054 | list_move_tail(&req->inflight_entry, &again); |
bbde017a XW |
2055 | continue; |
2056 | } | |
d21ffe7e | 2057 | list_del(&req->inflight_entry); |
def596e9 | 2058 | |
bcda7baa | 2059 | if (req->flags & REQ_F_BUFFER_SELECTED) |
8ff069bf | 2060 | cflags = io_put_rw_kbuf(req); |
bcda7baa JA |
2061 | |
2062 | __io_cqring_fill_event(req, req->result, cflags); | |
def596e9 JA |
2063 | (*nr_events)++; |
2064 | ||
c3524383 | 2065 | if (refcount_dec_and_test(&req->refs)) |
2d6500d4 | 2066 | io_req_free_batch(&rb, req); |
def596e9 | 2067 | } |
def596e9 | 2068 | |
09bb8394 | 2069 | io_commit_cqring(ctx); |
32b2244a XW |
2070 | if (ctx->flags & IORING_SETUP_SQPOLL) |
2071 | io_cqring_ev_posted(ctx); | |
2d6500d4 | 2072 | io_req_free_batch_finish(ctx, &rb); |
581f9810 | 2073 | |
bbde017a XW |
2074 | if (!list_empty(&again)) |
2075 | io_iopoll_queue(&again); | |
581f9810 BM |
2076 | } |
2077 | ||
def596e9 JA |
2078 | static int io_do_iopoll(struct io_ring_ctx *ctx, unsigned int *nr_events, |
2079 | long min) | |
2080 | { | |
2081 | struct io_kiocb *req, *tmp; | |
2082 | LIST_HEAD(done); | |
2083 | bool spin; | |
2084 | int ret; | |
2085 | ||
2086 | /* | |
2087 | * Only spin for completions if we don't have multiple devices hanging | |
2088 | * off our complete list, and we're under the requested amount. | |
2089 | */ | |
2090 | spin = !ctx->poll_multi_file && *nr_events < min; | |
2091 | ||
2092 | ret = 0; | |
d21ffe7e | 2093 | list_for_each_entry_safe(req, tmp, &ctx->iopoll_list, inflight_entry) { |
9adbd45d | 2094 | struct kiocb *kiocb = &req->rw.kiocb; |
def596e9 JA |
2095 | |
2096 | /* | |
581f9810 BM |
2097 | * Move completed and retryable entries to our local lists. |
2098 | * If we find a request that requires polling, break out | |
2099 | * and complete those lists first, if we have entries there. | |
def596e9 | 2100 | */ |
65a6543d | 2101 | if (READ_ONCE(req->iopoll_completed)) { |
d21ffe7e | 2102 | list_move_tail(&req->inflight_entry, &done); |
def596e9 JA |
2103 | continue; |
2104 | } | |
2105 | if (!list_empty(&done)) | |
2106 | break; | |
2107 | ||
2108 | ret = kiocb->ki_filp->f_op->iopoll(kiocb, spin); | |
2109 | if (ret < 0) | |
2110 | break; | |
2111 | ||
3aadc23e PB |
2112 | /* iopoll may have completed current req */ |
2113 | if (READ_ONCE(req->iopoll_completed)) | |
d21ffe7e | 2114 | list_move_tail(&req->inflight_entry, &done); |
3aadc23e | 2115 | |
def596e9 JA |
2116 | if (ret && spin) |
2117 | spin = false; | |
2118 | ret = 0; | |
2119 | } | |
2120 | ||
2121 | if (!list_empty(&done)) | |
2122 | io_iopoll_complete(ctx, nr_events, &done); | |
2123 | ||
2124 | return ret; | |
2125 | } | |
2126 | ||
2127 | /* | |
d195a66e | 2128 | * Poll for a minimum of 'min' events. Note that if min == 0 we consider that a |
def596e9 JA |
2129 | * non-spinning poll check - we'll still enter the driver poll loop, but only |
2130 | * as a non-spinning completion check. | |
2131 | */ | |
2132 | static int io_iopoll_getevents(struct io_ring_ctx *ctx, unsigned int *nr_events, | |
2133 | long min) | |
2134 | { | |
540e32a0 | 2135 | while (!list_empty(&ctx->iopoll_list) && !need_resched()) { |
def596e9 JA |
2136 | int ret; |
2137 | ||
2138 | ret = io_do_iopoll(ctx, nr_events, min); | |
2139 | if (ret < 0) | |
2140 | return ret; | |
eba0a4dd | 2141 | if (*nr_events >= min) |
def596e9 JA |
2142 | return 0; |
2143 | } | |
2144 | ||
2145 | return 1; | |
2146 | } | |
2147 | ||
2148 | /* | |
2149 | * We can't just wait for polled events to come to us, we have to actively | |
2150 | * find and complete them. | |
2151 | */ | |
b2edc0a7 | 2152 | static void io_iopoll_try_reap_events(struct io_ring_ctx *ctx) |
def596e9 JA |
2153 | { |
2154 | if (!(ctx->flags & IORING_SETUP_IOPOLL)) | |
2155 | return; | |
2156 | ||
2157 | mutex_lock(&ctx->uring_lock); | |
540e32a0 | 2158 | while (!list_empty(&ctx->iopoll_list)) { |
def596e9 JA |
2159 | unsigned int nr_events = 0; |
2160 | ||
b2edc0a7 | 2161 | io_do_iopoll(ctx, &nr_events, 0); |
08f5439f | 2162 | |
b2edc0a7 PB |
2163 | /* let it sleep and repeat later if can't complete a request */ |
2164 | if (nr_events == 0) | |
2165 | break; | |
08f5439f JA |
2166 | /* |
2167 | * Ensure we allow local-to-the-cpu processing to take place, | |
2168 | * in this case we need to ensure that we reap all events. | |
3fcee5a6 | 2169 | * Also let task_work, etc. to progress by releasing the mutex |
08f5439f | 2170 | */ |
3fcee5a6 PB |
2171 | if (need_resched()) { |
2172 | mutex_unlock(&ctx->uring_lock); | |
2173 | cond_resched(); | |
2174 | mutex_lock(&ctx->uring_lock); | |
2175 | } | |
def596e9 JA |
2176 | } |
2177 | mutex_unlock(&ctx->uring_lock); | |
2178 | } | |
2179 | ||
7668b92a | 2180 | static int io_iopoll_check(struct io_ring_ctx *ctx, long min) |
def596e9 | 2181 | { |
7668b92a | 2182 | unsigned int nr_events = 0; |
2b2ed975 | 2183 | int iters = 0, ret = 0; |
500f9fba | 2184 | |
c7849be9 XW |
2185 | /* |
2186 | * We disallow the app entering submit/complete with polling, but we | |
2187 | * still need to lock the ring to prevent racing with polled issue | |
2188 | * that got punted to a workqueue. | |
2189 | */ | |
2190 | mutex_lock(&ctx->uring_lock); | |
def596e9 | 2191 | do { |
a3a0e43f JA |
2192 | /* |
2193 | * Don't enter poll loop if we already have events pending. | |
2194 | * If we do, we can potentially be spinning for commands that | |
2195 | * already triggered a CQE (eg in error). | |
2196 | */ | |
1d7bb1d5 | 2197 | if (io_cqring_events(ctx, false)) |
a3a0e43f JA |
2198 | break; |
2199 | ||
500f9fba JA |
2200 | /* |
2201 | * If a submit got punted to a workqueue, we can have the | |
2202 | * application entering polling for a command before it gets | |
2203 | * issued. That app will hold the uring_lock for the duration | |
2204 | * of the poll right here, so we need to take a breather every | |
2205 | * now and then to ensure that the issue has a chance to add | |
2206 | * the poll to the issued list. Otherwise we can spin here | |
2207 | * forever, while the workqueue is stuck trying to acquire the | |
2208 | * very same mutex. | |
2209 | */ | |
2210 | if (!(++iters & 7)) { | |
2211 | mutex_unlock(&ctx->uring_lock); | |
4c6e277c | 2212 | io_run_task_work(); |
500f9fba JA |
2213 | mutex_lock(&ctx->uring_lock); |
2214 | } | |
2215 | ||
7668b92a | 2216 | ret = io_iopoll_getevents(ctx, &nr_events, min); |
def596e9 JA |
2217 | if (ret <= 0) |
2218 | break; | |
2219 | ret = 0; | |
7668b92a | 2220 | } while (min && !nr_events && !need_resched()); |
def596e9 | 2221 | |
500f9fba | 2222 | mutex_unlock(&ctx->uring_lock); |
def596e9 JA |
2223 | return ret; |
2224 | } | |
2225 | ||
491381ce | 2226 | static void kiocb_end_write(struct io_kiocb *req) |
2b188cc1 | 2227 | { |
491381ce JA |
2228 | /* |
2229 | * Tell lockdep we inherited freeze protection from submission | |
2230 | * thread. | |
2231 | */ | |
2232 | if (req->flags & REQ_F_ISREG) { | |
2233 | struct inode *inode = file_inode(req->file); | |
2b188cc1 | 2234 | |
491381ce | 2235 | __sb_writers_acquired(inode->i_sb, SB_FREEZE_WRITE); |
2b188cc1 | 2236 | } |
491381ce | 2237 | file_end_write(req->file); |
2b188cc1 JA |
2238 | } |
2239 | ||
a1d7c393 JA |
2240 | static void io_complete_rw_common(struct kiocb *kiocb, long res, |
2241 | struct io_comp_state *cs) | |
2b188cc1 | 2242 | { |
9adbd45d | 2243 | struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb); |
bcda7baa | 2244 | int cflags = 0; |
2b188cc1 | 2245 | |
491381ce JA |
2246 | if (kiocb->ki_flags & IOCB_WRITE) |
2247 | kiocb_end_write(req); | |
2b188cc1 | 2248 | |
4e88d6e7 JA |
2249 | if (res != req->result) |
2250 | req_set_fail_links(req); | |
bcda7baa | 2251 | if (req->flags & REQ_F_BUFFER_SELECTED) |
8ff069bf | 2252 | cflags = io_put_rw_kbuf(req); |
a1d7c393 | 2253 | __io_req_complete(req, res, cflags, cs); |
ba816ad6 JA |
2254 | } |
2255 | ||
b63534c4 JA |
2256 | #ifdef CONFIG_BLOCK |
2257 | static bool io_resubmit_prep(struct io_kiocb *req, int error) | |
2258 | { | |
2259 | struct iovec inline_vecs[UIO_FASTIOV], *iovec = inline_vecs; | |
2260 | ssize_t ret = -ECANCELED; | |
2261 | struct iov_iter iter; | |
2262 | int rw; | |
2263 | ||
2264 | if (error) { | |
2265 | ret = error; | |
2266 | goto end_req; | |
2267 | } | |
2268 | ||
2269 | switch (req->opcode) { | |
2270 | case IORING_OP_READV: | |
2271 | case IORING_OP_READ_FIXED: | |
2272 | case IORING_OP_READ: | |
2273 | rw = READ; | |
2274 | break; | |
2275 | case IORING_OP_WRITEV: | |
2276 | case IORING_OP_WRITE_FIXED: | |
2277 | case IORING_OP_WRITE: | |
2278 | rw = WRITE; | |
2279 | break; | |
2280 | default: | |
2281 | printk_once(KERN_WARNING "io_uring: bad opcode in resubmit %d\n", | |
2282 | req->opcode); | |
2283 | goto end_req; | |
2284 | } | |
2285 | ||
2286 | ret = io_import_iovec(rw, req, &iovec, &iter, false); | |
2287 | if (ret < 0) | |
2288 | goto end_req; | |
227c0c96 | 2289 | ret = io_setup_async_rw(req, iovec, inline_vecs, &iter, false); |
b63534c4 JA |
2290 | if (!ret) |
2291 | return true; | |
2292 | kfree(iovec); | |
2293 | end_req: | |
b63534c4 | 2294 | req_set_fail_links(req); |
e1e16097 | 2295 | io_req_complete(req, ret); |
b63534c4 JA |
2296 | return false; |
2297 | } | |
b63534c4 JA |
2298 | #endif |
2299 | ||
2300 | static bool io_rw_reissue(struct io_kiocb *req, long res) | |
2301 | { | |
2302 | #ifdef CONFIG_BLOCK | |
355afaeb | 2303 | umode_t mode = file_inode(req->file)->i_mode; |
b63534c4 JA |
2304 | int ret; |
2305 | ||
355afaeb JA |
2306 | if (!S_ISBLK(mode) && !S_ISREG(mode)) |
2307 | return false; | |
b63534c4 JA |
2308 | if ((res != -EAGAIN && res != -EOPNOTSUPP) || io_wq_current_is_worker()) |
2309 | return false; | |
2310 | ||
fdee946d | 2311 | ret = io_sq_thread_acquire_mm(req->ctx, req); |
6d816e08 | 2312 | |
fdee946d JA |
2313 | if (io_resubmit_prep(req, ret)) { |
2314 | refcount_inc(&req->refs); | |
2315 | io_queue_async_work(req); | |
b63534c4 | 2316 | return true; |
fdee946d JA |
2317 | } |
2318 | ||
b63534c4 JA |
2319 | #endif |
2320 | return false; | |
2321 | } | |
2322 | ||
a1d7c393 JA |
2323 | static void __io_complete_rw(struct io_kiocb *req, long res, long res2, |
2324 | struct io_comp_state *cs) | |
2325 | { | |
2326 | if (!io_rw_reissue(req, res)) | |
2327 | io_complete_rw_common(&req->rw.kiocb, res, cs); | |
ba816ad6 JA |
2328 | } |
2329 | ||
2330 | static void io_complete_rw(struct kiocb *kiocb, long res, long res2) | |
2331 | { | |
9adbd45d | 2332 | struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb); |
ba816ad6 | 2333 | |
a1d7c393 | 2334 | __io_complete_rw(req, res, res2, NULL); |
2b188cc1 JA |
2335 | } |
2336 | ||
def596e9 JA |
2337 | static void io_complete_rw_iopoll(struct kiocb *kiocb, long res, long res2) |
2338 | { | |
9adbd45d | 2339 | struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb); |
def596e9 | 2340 | |
491381ce JA |
2341 | if (kiocb->ki_flags & IOCB_WRITE) |
2342 | kiocb_end_write(req); | |
def596e9 | 2343 | |
2d7d6792 | 2344 | if (res != -EAGAIN && res != req->result) |
4e88d6e7 | 2345 | req_set_fail_links(req); |
bbde017a XW |
2346 | |
2347 | WRITE_ONCE(req->result, res); | |
2348 | /* order with io_poll_complete() checking ->result */ | |
cd664b0e PB |
2349 | smp_wmb(); |
2350 | WRITE_ONCE(req->iopoll_completed, 1); | |
def596e9 JA |
2351 | } |
2352 | ||
2353 | /* | |
2354 | * After the iocb has been issued, it's safe to be found on the poll list. | |
2355 | * Adding the kiocb to the list AFTER submission ensures that we don't | |
2356 | * find it from a io_iopoll_getevents() thread before the issuer is done | |
2357 | * accessing the kiocb cookie. | |
2358 | */ | |
2359 | static void io_iopoll_req_issued(struct io_kiocb *req) | |
2360 | { | |
2361 | struct io_ring_ctx *ctx = req->ctx; | |
2362 | ||
2363 | /* | |
2364 | * Track whether we have multiple files in our lists. This will impact | |
2365 | * how we do polling eventually, not spinning if we're on potentially | |
2366 | * different devices. | |
2367 | */ | |
540e32a0 | 2368 | if (list_empty(&ctx->iopoll_list)) { |
def596e9 JA |
2369 | ctx->poll_multi_file = false; |
2370 | } else if (!ctx->poll_multi_file) { | |
2371 | struct io_kiocb *list_req; | |
2372 | ||
540e32a0 | 2373 | list_req = list_first_entry(&ctx->iopoll_list, struct io_kiocb, |
d21ffe7e | 2374 | inflight_entry); |
9adbd45d | 2375 | if (list_req->file != req->file) |
def596e9 JA |
2376 | ctx->poll_multi_file = true; |
2377 | } | |
2378 | ||
2379 | /* | |
2380 | * For fast devices, IO may have already completed. If it has, add | |
2381 | * it to the front so we find it first. | |
2382 | */ | |
65a6543d | 2383 | if (READ_ONCE(req->iopoll_completed)) |
d21ffe7e | 2384 | list_add(&req->inflight_entry, &ctx->iopoll_list); |
def596e9 | 2385 | else |
d21ffe7e | 2386 | list_add_tail(&req->inflight_entry, &ctx->iopoll_list); |
bdcd3eab XW |
2387 | |
2388 | if ((ctx->flags & IORING_SETUP_SQPOLL) && | |
2389 | wq_has_sleeper(&ctx->sqo_wait)) | |
2390 | wake_up(&ctx->sqo_wait); | |
def596e9 JA |
2391 | } |
2392 | ||
9f13c35b | 2393 | static void __io_state_file_put(struct io_submit_state *state) |
9a56a232 | 2394 | { |
06ef3608 PB |
2395 | if (state->has_refs) |
2396 | fput_many(state->file, state->has_refs); | |
9f13c35b PB |
2397 | state->file = NULL; |
2398 | } | |
2399 | ||
2400 | static inline void io_state_file_put(struct io_submit_state *state) | |
2401 | { | |
2402 | if (state->file) | |
2403 | __io_state_file_put(state); | |
9a56a232 JA |
2404 | } |
2405 | ||
2406 | /* | |
2407 | * Get as many references to a file as we have IOs left in this submission, | |
2408 | * assuming most submissions are for one file, or at least that each file | |
2409 | * has more than one submission. | |
2410 | */ | |
8da11c19 | 2411 | static struct file *__io_file_get(struct io_submit_state *state, int fd) |
9a56a232 JA |
2412 | { |
2413 | if (!state) | |
2414 | return fget(fd); | |
2415 | ||
2416 | if (state->file) { | |
2417 | if (state->fd == fd) { | |
06ef3608 | 2418 | state->has_refs--; |
9a56a232 JA |
2419 | state->ios_left--; |
2420 | return state->file; | |
2421 | } | |
9f13c35b | 2422 | __io_state_file_put(state); |
9a56a232 JA |
2423 | } |
2424 | state->file = fget_many(fd, state->ios_left); | |
2425 | if (!state->file) | |
2426 | return NULL; | |
2427 | ||
2428 | state->fd = fd; | |
9a56a232 | 2429 | state->ios_left--; |
06ef3608 | 2430 | state->has_refs = state->ios_left; |
9a56a232 JA |
2431 | return state->file; |
2432 | } | |
2433 | ||
4503b767 JA |
2434 | static bool io_bdev_nowait(struct block_device *bdev) |
2435 | { | |
2436 | #ifdef CONFIG_BLOCK | |
2437 | return !bdev || queue_is_mq(bdev_get_queue(bdev)); | |
2438 | #else | |
2439 | return true; | |
2440 | #endif | |
2441 | } | |
2442 | ||
2b188cc1 JA |
2443 | /* |
2444 | * If we tracked the file through the SCM inflight mechanism, we could support | |
2445 | * any file. For now, just ensure that anything potentially problematic is done | |
2446 | * inline. | |
2447 | */ | |
af197f50 | 2448 | static bool io_file_supports_async(struct file *file, int rw) |
2b188cc1 JA |
2449 | { |
2450 | umode_t mode = file_inode(file)->i_mode; | |
2451 | ||
4503b767 JA |
2452 | if (S_ISBLK(mode)) { |
2453 | if (io_bdev_nowait(file->f_inode->i_bdev)) | |
2454 | return true; | |
2455 | return false; | |
2456 | } | |
2457 | if (S_ISCHR(mode) || S_ISSOCK(mode)) | |
2b188cc1 | 2458 | return true; |
4503b767 JA |
2459 | if (S_ISREG(mode)) { |
2460 | if (io_bdev_nowait(file->f_inode->i_sb->s_bdev) && | |
2461 | file->f_op != &io_uring_fops) | |
2462 | return true; | |
2463 | return false; | |
2464 | } | |
2b188cc1 | 2465 | |
c5b85625 JA |
2466 | /* any ->read/write should understand O_NONBLOCK */ |
2467 | if (file->f_flags & O_NONBLOCK) | |
2468 | return true; | |
2469 | ||
af197f50 JA |
2470 | if (!(file->f_mode & FMODE_NOWAIT)) |
2471 | return false; | |
2472 | ||
2473 | if (rw == READ) | |
2474 | return file->f_op->read_iter != NULL; | |
2475 | ||
2476 | return file->f_op->write_iter != NULL; | |
2b188cc1 JA |
2477 | } |
2478 | ||
3529d8c2 JA |
2479 | static int io_prep_rw(struct io_kiocb *req, const struct io_uring_sqe *sqe, |
2480 | bool force_nonblock) | |
2b188cc1 | 2481 | { |
def596e9 | 2482 | struct io_ring_ctx *ctx = req->ctx; |
9adbd45d | 2483 | struct kiocb *kiocb = &req->rw.kiocb; |
09bb8394 JA |
2484 | unsigned ioprio; |
2485 | int ret; | |
2b188cc1 | 2486 | |
491381ce JA |
2487 | if (S_ISREG(file_inode(req->file)->i_mode)) |
2488 | req->flags |= REQ_F_ISREG; | |
2489 | ||
2b188cc1 | 2490 | kiocb->ki_pos = READ_ONCE(sqe->off); |
ba04291e JA |
2491 | if (kiocb->ki_pos == -1 && !(req->file->f_mode & FMODE_STREAM)) { |
2492 | req->flags |= REQ_F_CUR_POS; | |
2493 | kiocb->ki_pos = req->file->f_pos; | |
2494 | } | |
2b188cc1 | 2495 | kiocb->ki_hint = ki_hint_validate(file_write_hint(kiocb->ki_filp)); |
3e577dcd PB |
2496 | kiocb->ki_flags = iocb_flags(kiocb->ki_filp); |
2497 | ret = kiocb_set_rw_flags(kiocb, READ_ONCE(sqe->rw_flags)); | |
2498 | if (unlikely(ret)) | |
2499 | return ret; | |
2b188cc1 JA |
2500 | |
2501 | ioprio = READ_ONCE(sqe->ioprio); | |
2502 | if (ioprio) { | |
2503 | ret = ioprio_check_cap(ioprio); | |
2504 | if (ret) | |
09bb8394 | 2505 | return ret; |
2b188cc1 JA |
2506 | |
2507 | kiocb->ki_ioprio = ioprio; | |
2508 | } else | |
2509 | kiocb->ki_ioprio = get_current_ioprio(); | |
2510 | ||
8449eeda | 2511 | /* don't allow async punt if RWF_NOWAIT was requested */ |
c5b85625 | 2512 | if (kiocb->ki_flags & IOCB_NOWAIT) |
8449eeda SB |
2513 | req->flags |= REQ_F_NOWAIT; |
2514 | ||
b63534c4 JA |
2515 | if (kiocb->ki_flags & IOCB_DIRECT) |
2516 | io_get_req_task(req); | |
2517 | ||
8449eeda | 2518 | if (force_nonblock) |
2b188cc1 | 2519 | kiocb->ki_flags |= IOCB_NOWAIT; |
8449eeda | 2520 | |
def596e9 | 2521 | if (ctx->flags & IORING_SETUP_IOPOLL) { |
def596e9 JA |
2522 | if (!(kiocb->ki_flags & IOCB_DIRECT) || |
2523 | !kiocb->ki_filp->f_op->iopoll) | |
09bb8394 | 2524 | return -EOPNOTSUPP; |
2b188cc1 | 2525 | |
def596e9 JA |
2526 | kiocb->ki_flags |= IOCB_HIPRI; |
2527 | kiocb->ki_complete = io_complete_rw_iopoll; | |
65a6543d | 2528 | req->iopoll_completed = 0; |
f3a6fa22 | 2529 | io_get_req_task(req); |
def596e9 | 2530 | } else { |
09bb8394 JA |
2531 | if (kiocb->ki_flags & IOCB_HIPRI) |
2532 | return -EINVAL; | |
def596e9 JA |
2533 | kiocb->ki_complete = io_complete_rw; |
2534 | } | |
9adbd45d | 2535 | |
3529d8c2 JA |
2536 | req->rw.addr = READ_ONCE(sqe->addr); |
2537 | req->rw.len = READ_ONCE(sqe->len); | |
4f4eeba8 | 2538 | req->buf_index = READ_ONCE(sqe->buf_index); |
2b188cc1 | 2539 | return 0; |
2b188cc1 JA |
2540 | } |
2541 | ||
2542 | static inline void io_rw_done(struct kiocb *kiocb, ssize_t ret) | |
2543 | { | |
2544 | switch (ret) { | |
2545 | case -EIOCBQUEUED: | |
2546 | break; | |
2547 | case -ERESTARTSYS: | |
2548 | case -ERESTARTNOINTR: | |
2549 | case -ERESTARTNOHAND: | |
2550 | case -ERESTART_RESTARTBLOCK: | |
2551 | /* | |
2552 | * We can't just restart the syscall, since previously | |
2553 | * submitted sqes may already be in progress. Just fail this | |
2554 | * IO with EINTR. | |
2555 | */ | |
2556 | ret = -EINTR; | |
2557 | /* fall through */ | |
2558 | default: | |
2559 | kiocb->ki_complete(kiocb, ret, 0); | |
2560 | } | |
2561 | } | |
2562 | ||
a1d7c393 JA |
2563 | static void kiocb_done(struct kiocb *kiocb, ssize_t ret, |
2564 | struct io_comp_state *cs) | |
ba816ad6 | 2565 | { |
ba04291e JA |
2566 | struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb); |
2567 | ||
227c0c96 JA |
2568 | /* add previously done IO, if any */ |
2569 | if (req->io && req->io->rw.bytes_done > 0) { | |
2570 | if (ret < 0) | |
2571 | ret = req->io->rw.bytes_done; | |
2572 | else | |
2573 | ret += req->io->rw.bytes_done; | |
2574 | } | |
2575 | ||
ba04291e JA |
2576 | if (req->flags & REQ_F_CUR_POS) |
2577 | req->file->f_pos = kiocb->ki_pos; | |
bcaec089 | 2578 | if (ret >= 0 && kiocb->ki_complete == io_complete_rw) |
a1d7c393 | 2579 | __io_complete_rw(req, ret, 0, cs); |
ba816ad6 JA |
2580 | else |
2581 | io_rw_done(kiocb, ret); | |
2582 | } | |
2583 | ||
9adbd45d | 2584 | static ssize_t io_import_fixed(struct io_kiocb *req, int rw, |
7d009165 | 2585 | struct iov_iter *iter) |
edafccee | 2586 | { |
9adbd45d JA |
2587 | struct io_ring_ctx *ctx = req->ctx; |
2588 | size_t len = req->rw.len; | |
edafccee | 2589 | struct io_mapped_ubuf *imu; |
4f4eeba8 | 2590 | u16 index, buf_index; |
edafccee JA |
2591 | size_t offset; |
2592 | u64 buf_addr; | |
2593 | ||
2594 | /* attempt to use fixed buffers without having provided iovecs */ | |
2595 | if (unlikely(!ctx->user_bufs)) | |
2596 | return -EFAULT; | |
2597 | ||
4f4eeba8 | 2598 | buf_index = req->buf_index; |
edafccee JA |
2599 | if (unlikely(buf_index >= ctx->nr_user_bufs)) |
2600 | return -EFAULT; | |
2601 | ||
2602 | index = array_index_nospec(buf_index, ctx->nr_user_bufs); | |
2603 | imu = &ctx->user_bufs[index]; | |
9adbd45d | 2604 | buf_addr = req->rw.addr; |
edafccee JA |
2605 | |
2606 | /* overflow */ | |
2607 | if (buf_addr + len < buf_addr) | |
2608 | return -EFAULT; | |
2609 | /* not inside the mapped region */ | |
2610 | if (buf_addr < imu->ubuf || buf_addr + len > imu->ubuf + imu->len) | |
2611 | return -EFAULT; | |
2612 | ||
2613 | /* | |
2614 | * May not be a start of buffer, set size appropriately | |
2615 | * and advance us to the beginning. | |
2616 | */ | |
2617 | offset = buf_addr - imu->ubuf; | |
2618 | iov_iter_bvec(iter, rw, imu->bvec, imu->nr_bvecs, offset + len); | |
bd11b3a3 JA |
2619 | |
2620 | if (offset) { | |
2621 | /* | |
2622 | * Don't use iov_iter_advance() here, as it's really slow for | |
2623 | * using the latter parts of a big fixed buffer - it iterates | |
2624 | * over each segment manually. We can cheat a bit here, because | |
2625 | * we know that: | |
2626 | * | |
2627 | * 1) it's a BVEC iter, we set it up | |
2628 | * 2) all bvecs are PAGE_SIZE in size, except potentially the | |
2629 | * first and last bvec | |
2630 | * | |
2631 | * So just find our index, and adjust the iterator afterwards. | |
2632 | * If the offset is within the first bvec (or the whole first | |
2633 | * bvec, just use iov_iter_advance(). This makes it easier | |
2634 | * since we can just skip the first segment, which may not | |
2635 | * be PAGE_SIZE aligned. | |
2636 | */ | |
2637 | const struct bio_vec *bvec = imu->bvec; | |
2638 | ||
2639 | if (offset <= bvec->bv_len) { | |
2640 | iov_iter_advance(iter, offset); | |
2641 | } else { | |
2642 | unsigned long seg_skip; | |
2643 | ||
2644 | /* skip first vec */ | |
2645 | offset -= bvec->bv_len; | |
2646 | seg_skip = 1 + (offset >> PAGE_SHIFT); | |
2647 | ||
2648 | iter->bvec = bvec + seg_skip; | |
2649 | iter->nr_segs -= seg_skip; | |
99c79f66 | 2650 | iter->count -= bvec->bv_len + offset; |
bd11b3a3 | 2651 | iter->iov_offset = offset & ~PAGE_MASK; |
bd11b3a3 JA |
2652 | } |
2653 | } | |
2654 | ||
5e559561 | 2655 | return len; |
edafccee JA |
2656 | } |
2657 | ||
bcda7baa JA |
2658 | static void io_ring_submit_unlock(struct io_ring_ctx *ctx, bool needs_lock) |
2659 | { | |
2660 | if (needs_lock) | |
2661 | mutex_unlock(&ctx->uring_lock); | |
2662 | } | |
2663 | ||
2664 | static void io_ring_submit_lock(struct io_ring_ctx *ctx, bool needs_lock) | |
2665 | { | |
2666 | /* | |
2667 | * "Normal" inline submissions always hold the uring_lock, since we | |
2668 | * grab it from the system call. Same is true for the SQPOLL offload. | |
2669 | * The only exception is when we've detached the request and issue it | |
2670 | * from an async worker thread, grab the lock for that case. | |
2671 | */ | |
2672 | if (needs_lock) | |
2673 | mutex_lock(&ctx->uring_lock); | |
2674 | } | |
2675 | ||
2676 | static struct io_buffer *io_buffer_select(struct io_kiocb *req, size_t *len, | |
2677 | int bgid, struct io_buffer *kbuf, | |
2678 | bool needs_lock) | |
2679 | { | |
2680 | struct io_buffer *head; | |
2681 | ||
2682 | if (req->flags & REQ_F_BUFFER_SELECTED) | |
2683 | return kbuf; | |
2684 | ||
2685 | io_ring_submit_lock(req->ctx, needs_lock); | |
2686 | ||
2687 | lockdep_assert_held(&req->ctx->uring_lock); | |
2688 | ||
2689 | head = idr_find(&req->ctx->io_buffer_idr, bgid); | |
2690 | if (head) { | |
2691 | if (!list_empty(&head->list)) { | |
2692 | kbuf = list_last_entry(&head->list, struct io_buffer, | |
2693 | list); | |
2694 | list_del(&kbuf->list); | |
2695 | } else { | |
2696 | kbuf = head; | |
2697 | idr_remove(&req->ctx->io_buffer_idr, bgid); | |
2698 | } | |
2699 | if (*len > kbuf->len) | |
2700 | *len = kbuf->len; | |
2701 | } else { | |
2702 | kbuf = ERR_PTR(-ENOBUFS); | |
2703 | } | |
2704 | ||
2705 | io_ring_submit_unlock(req->ctx, needs_lock); | |
2706 | ||
2707 | return kbuf; | |
2708 | } | |
2709 | ||
4d954c25 JA |
2710 | static void __user *io_rw_buffer_select(struct io_kiocb *req, size_t *len, |
2711 | bool needs_lock) | |
2712 | { | |
2713 | struct io_buffer *kbuf; | |
4f4eeba8 | 2714 | u16 bgid; |
4d954c25 JA |
2715 | |
2716 | kbuf = (struct io_buffer *) (unsigned long) req->rw.addr; | |
4f4eeba8 | 2717 | bgid = req->buf_index; |
4d954c25 JA |
2718 | kbuf = io_buffer_select(req, len, bgid, kbuf, needs_lock); |
2719 | if (IS_ERR(kbuf)) | |
2720 | return kbuf; | |
2721 | req->rw.addr = (u64) (unsigned long) kbuf; | |
2722 | req->flags |= REQ_F_BUFFER_SELECTED; | |
2723 | return u64_to_user_ptr(kbuf->addr); | |
2724 | } | |
2725 | ||
2726 | #ifdef CONFIG_COMPAT | |
2727 | static ssize_t io_compat_import(struct io_kiocb *req, struct iovec *iov, | |
2728 | bool needs_lock) | |
2729 | { | |
2730 | struct compat_iovec __user *uiov; | |
2731 | compat_ssize_t clen; | |
2732 | void __user *buf; | |
2733 | ssize_t len; | |
2734 | ||
2735 | uiov = u64_to_user_ptr(req->rw.addr); | |
2736 | if (!access_ok(uiov, sizeof(*uiov))) | |
2737 | return -EFAULT; | |
2738 | if (__get_user(clen, &uiov->iov_len)) | |
2739 | return -EFAULT; | |
2740 | if (clen < 0) | |
2741 | return -EINVAL; | |
2742 | ||
2743 | len = clen; | |
2744 | buf = io_rw_buffer_select(req, &len, needs_lock); | |
2745 | if (IS_ERR(buf)) | |
2746 | return PTR_ERR(buf); | |
2747 | iov[0].iov_base = buf; | |
2748 | iov[0].iov_len = (compat_size_t) len; | |
2749 | return 0; | |
2750 | } | |
2751 | #endif | |
2752 | ||
2753 | static ssize_t __io_iov_buffer_select(struct io_kiocb *req, struct iovec *iov, | |
2754 | bool needs_lock) | |
2755 | { | |
2756 | struct iovec __user *uiov = u64_to_user_ptr(req->rw.addr); | |
2757 | void __user *buf; | |
2758 | ssize_t len; | |
2759 | ||
2760 | if (copy_from_user(iov, uiov, sizeof(*uiov))) | |
2761 | return -EFAULT; | |
2762 | ||
2763 | len = iov[0].iov_len; | |
2764 | if (len < 0) | |
2765 | return -EINVAL; | |
2766 | buf = io_rw_buffer_select(req, &len, needs_lock); | |
2767 | if (IS_ERR(buf)) | |
2768 | return PTR_ERR(buf); | |
2769 | iov[0].iov_base = buf; | |
2770 | iov[0].iov_len = len; | |
2771 | return 0; | |
2772 | } | |
2773 | ||
2774 | static ssize_t io_iov_buffer_select(struct io_kiocb *req, struct iovec *iov, | |
2775 | bool needs_lock) | |
2776 | { | |
dddb3e26 JA |
2777 | if (req->flags & REQ_F_BUFFER_SELECTED) { |
2778 | struct io_buffer *kbuf; | |
2779 | ||
2780 | kbuf = (struct io_buffer *) (unsigned long) req->rw.addr; | |
2781 | iov[0].iov_base = u64_to_user_ptr(kbuf->addr); | |
2782 | iov[0].iov_len = kbuf->len; | |
4d954c25 | 2783 | return 0; |
dddb3e26 | 2784 | } |
4d954c25 JA |
2785 | if (!req->rw.len) |
2786 | return 0; | |
2787 | else if (req->rw.len > 1) | |
2788 | return -EINVAL; | |
2789 | ||
2790 | #ifdef CONFIG_COMPAT | |
2791 | if (req->ctx->compat) | |
2792 | return io_compat_import(req, iov, needs_lock); | |
2793 | #endif | |
2794 | ||
2795 | return __io_iov_buffer_select(req, iov, needs_lock); | |
2796 | } | |
2797 | ||
8452fd0c JA |
2798 | static ssize_t __io_import_iovec(int rw, struct io_kiocb *req, |
2799 | struct iovec **iovec, struct iov_iter *iter, | |
2800 | bool needs_lock) | |
2b188cc1 | 2801 | { |
9adbd45d JA |
2802 | void __user *buf = u64_to_user_ptr(req->rw.addr); |
2803 | size_t sqe_len = req->rw.len; | |
4d954c25 | 2804 | ssize_t ret; |
edafccee JA |
2805 | u8 opcode; |
2806 | ||
d625c6ee | 2807 | opcode = req->opcode; |
7d009165 | 2808 | if (opcode == IORING_OP_READ_FIXED || opcode == IORING_OP_WRITE_FIXED) { |
edafccee | 2809 | *iovec = NULL; |
9adbd45d | 2810 | return io_import_fixed(req, rw, iter); |
edafccee | 2811 | } |
2b188cc1 | 2812 | |
bcda7baa | 2813 | /* buffer index only valid with fixed read/write, or buffer select */ |
4f4eeba8 | 2814 | if (req->buf_index && !(req->flags & REQ_F_BUFFER_SELECT)) |
9adbd45d JA |
2815 | return -EINVAL; |
2816 | ||
3a6820f2 | 2817 | if (opcode == IORING_OP_READ || opcode == IORING_OP_WRITE) { |
bcda7baa | 2818 | if (req->flags & REQ_F_BUFFER_SELECT) { |
4d954c25 | 2819 | buf = io_rw_buffer_select(req, &sqe_len, needs_lock); |
867a23ea | 2820 | if (IS_ERR(buf)) |
4d954c25 | 2821 | return PTR_ERR(buf); |
3f9d6441 | 2822 | req->rw.len = sqe_len; |
bcda7baa JA |
2823 | } |
2824 | ||
3a6820f2 JA |
2825 | ret = import_single_range(rw, buf, sqe_len, *iovec, iter); |
2826 | *iovec = NULL; | |
3a901598 | 2827 | return ret < 0 ? ret : sqe_len; |
3a6820f2 JA |
2828 | } |
2829 | ||
4d954c25 JA |
2830 | if (req->flags & REQ_F_BUFFER_SELECT) { |
2831 | ret = io_iov_buffer_select(req, *iovec, needs_lock); | |
3f9d6441 JA |
2832 | if (!ret) { |
2833 | ret = (*iovec)->iov_len; | |
2834 | iov_iter_init(iter, rw, *iovec, 1, ret); | |
2835 | } | |
4d954c25 JA |
2836 | *iovec = NULL; |
2837 | return ret; | |
2838 | } | |
2839 | ||
2b188cc1 | 2840 | #ifdef CONFIG_COMPAT |
cf6fd4bd | 2841 | if (req->ctx->compat) |
2b188cc1 JA |
2842 | return compat_import_iovec(rw, buf, sqe_len, UIO_FASTIOV, |
2843 | iovec, iter); | |
2844 | #endif | |
2845 | ||
2846 | return import_iovec(rw, buf, sqe_len, UIO_FASTIOV, iovec, iter); | |
2847 | } | |
2848 | ||
8452fd0c JA |
2849 | static ssize_t io_import_iovec(int rw, struct io_kiocb *req, |
2850 | struct iovec **iovec, struct iov_iter *iter, | |
2851 | bool needs_lock) | |
2852 | { | |
2853 | if (!req->io) | |
2854 | return __io_import_iovec(rw, req, iovec, iter, needs_lock); | |
2855 | *iovec = NULL; | |
2856 | return iov_iter_count(&req->io->rw.iter); | |
2857 | } | |
2858 | ||
0fef9483 JA |
2859 | static inline loff_t *io_kiocb_ppos(struct kiocb *kiocb) |
2860 | { | |
2861 | return kiocb->ki_filp->f_mode & FMODE_STREAM ? NULL : &kiocb->ki_pos; | |
2862 | } | |
2863 | ||
31b51510 | 2864 | /* |
32960613 JA |
2865 | * For files that don't have ->read_iter() and ->write_iter(), handle them |
2866 | * by looping over ->read() or ->write() manually. | |
31b51510 | 2867 | */ |
32960613 JA |
2868 | static ssize_t loop_rw_iter(int rw, struct file *file, struct kiocb *kiocb, |
2869 | struct iov_iter *iter) | |
2870 | { | |
2871 | ssize_t ret = 0; | |
2872 | ||
2873 | /* | |
2874 | * Don't support polled IO through this interface, and we can't | |
2875 | * support non-blocking either. For the latter, this just causes | |
2876 | * the kiocb to be handled from an async context. | |
2877 | */ | |
2878 | if (kiocb->ki_flags & IOCB_HIPRI) | |
2879 | return -EOPNOTSUPP; | |
2880 | if (kiocb->ki_flags & IOCB_NOWAIT) | |
2881 | return -EAGAIN; | |
2882 | ||
2883 | while (iov_iter_count(iter)) { | |
311ae9e1 | 2884 | struct iovec iovec; |
32960613 JA |
2885 | ssize_t nr; |
2886 | ||
311ae9e1 PB |
2887 | if (!iov_iter_is_bvec(iter)) { |
2888 | iovec = iov_iter_iovec(iter); | |
2889 | } else { | |
2890 | /* fixed buffers import bvec */ | |
2891 | iovec.iov_base = kmap(iter->bvec->bv_page) | |
2892 | + iter->iov_offset; | |
2893 | iovec.iov_len = min(iter->count, | |
2894 | iter->bvec->bv_len - iter->iov_offset); | |
2895 | } | |
2896 | ||
32960613 JA |
2897 | if (rw == READ) { |
2898 | nr = file->f_op->read(file, iovec.iov_base, | |
0fef9483 | 2899 | iovec.iov_len, io_kiocb_ppos(kiocb)); |
32960613 JA |
2900 | } else { |
2901 | nr = file->f_op->write(file, iovec.iov_base, | |
0fef9483 | 2902 | iovec.iov_len, io_kiocb_ppos(kiocb)); |
32960613 JA |
2903 | } |
2904 | ||
311ae9e1 PB |
2905 | if (iov_iter_is_bvec(iter)) |
2906 | kunmap(iter->bvec->bv_page); | |
2907 | ||
32960613 JA |
2908 | if (nr < 0) { |
2909 | if (!ret) | |
2910 | ret = nr; | |
2911 | break; | |
2912 | } | |
2913 | ret += nr; | |
2914 | if (nr != iovec.iov_len) | |
2915 | break; | |
2916 | iov_iter_advance(iter, nr); | |
2917 | } | |
2918 | ||
2919 | return ret; | |
2920 | } | |
2921 | ||
ff6165b2 JA |
2922 | static void io_req_map_rw(struct io_kiocb *req, const struct iovec *iovec, |
2923 | const struct iovec *fast_iov, struct iov_iter *iter) | |
f67676d1 | 2924 | { |
b64e3444 PB |
2925 | struct io_async_rw *rw = &req->io->rw; |
2926 | ||
ff6165b2 JA |
2927 | memcpy(&rw->iter, iter, sizeof(*iter)); |
2928 | rw->free_iovec = NULL; | |
227c0c96 | 2929 | rw->bytes_done = 0; |
ff6165b2 JA |
2930 | /* can only be fixed buffers, no need to do anything */ |
2931 | if (iter->type == ITER_BVEC) | |
2932 | return; | |
b64e3444 | 2933 | if (!iovec) { |
ff6165b2 JA |
2934 | unsigned iov_off = 0; |
2935 | ||
2936 | rw->iter.iov = rw->fast_iov; | |
2937 | if (iter->iov != fast_iov) { | |
2938 | iov_off = iter->iov - fast_iov; | |
2939 | rw->iter.iov += iov_off; | |
2940 | } | |
2941 | if (rw->fast_iov != fast_iov) | |
2942 | memcpy(rw->fast_iov + iov_off, fast_iov + iov_off, | |
45097dae | 2943 | sizeof(struct iovec) * iter->nr_segs); |
99bc4c38 | 2944 | } else { |
ff6165b2 | 2945 | rw->free_iovec = iovec; |
99bc4c38 | 2946 | req->flags |= REQ_F_NEED_CLEANUP; |
f67676d1 JA |
2947 | } |
2948 | } | |
2949 | ||
3d9932a8 XW |
2950 | static inline int __io_alloc_async_ctx(struct io_kiocb *req) |
2951 | { | |
2952 | req->io = kmalloc(sizeof(*req->io), GFP_KERNEL); | |
2953 | return req->io == NULL; | |
2954 | } | |
2955 | ||
b7bb4f7d | 2956 | static int io_alloc_async_ctx(struct io_kiocb *req) |
f67676d1 | 2957 | { |
d3656344 JA |
2958 | if (!io_op_defs[req->opcode].async_ctx) |
2959 | return 0; | |
3d9932a8 XW |
2960 | |
2961 | return __io_alloc_async_ctx(req); | |
b7bb4f7d JA |
2962 | } |
2963 | ||
ff6165b2 JA |
2964 | static int io_setup_async_rw(struct io_kiocb *req, const struct iovec *iovec, |
2965 | const struct iovec *fast_iov, | |
227c0c96 | 2966 | struct iov_iter *iter, bool force) |
b7bb4f7d | 2967 | { |
227c0c96 | 2968 | if (!force && !io_op_defs[req->opcode].async_ctx) |
74566df3 | 2969 | return 0; |
5d204bcf | 2970 | if (!req->io) { |
3d9932a8 | 2971 | if (__io_alloc_async_ctx(req)) |
5d204bcf | 2972 | return -ENOMEM; |
b7bb4f7d | 2973 | |
ff6165b2 | 2974 | io_req_map_rw(req, iovec, fast_iov, iter); |
5d204bcf | 2975 | } |
b7bb4f7d | 2976 | return 0; |
f67676d1 JA |
2977 | } |
2978 | ||
c3e330a4 PB |
2979 | static inline int io_rw_prep_async(struct io_kiocb *req, int rw, |
2980 | bool force_nonblock) | |
2981 | { | |
ff6165b2 | 2982 | struct io_async_rw *iorw = &req->io->rw; |
c183edff | 2983 | struct iovec *iov; |
c3e330a4 PB |
2984 | ssize_t ret; |
2985 | ||
c183edff JA |
2986 | iorw->iter.iov = iov = iorw->fast_iov; |
2987 | ret = __io_import_iovec(rw, req, &iov, &iorw->iter, !force_nonblock); | |
c3e330a4 PB |
2988 | if (unlikely(ret < 0)) |
2989 | return ret; | |
2990 | ||
c183edff | 2991 | iorw->iter.iov = iov; |
ff6165b2 | 2992 | io_req_map_rw(req, iorw->iter.iov, iorw->fast_iov, &iorw->iter); |
c3e330a4 PB |
2993 | return 0; |
2994 | } | |
2995 | ||
3529d8c2 JA |
2996 | static int io_read_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe, |
2997 | bool force_nonblock) | |
f67676d1 JA |
2998 | { |
2999 | ssize_t ret; | |
3000 | ||
3529d8c2 JA |
3001 | ret = io_prep_rw(req, sqe, force_nonblock); |
3002 | if (ret) | |
3003 | return ret; | |
f67676d1 | 3004 | |
3529d8c2 JA |
3005 | if (unlikely(!(req->file->f_mode & FMODE_READ))) |
3006 | return -EBADF; | |
f67676d1 | 3007 | |
5f798bea PB |
3008 | /* either don't need iovec imported or already have it */ |
3009 | if (!req->io || req->flags & REQ_F_NEED_CLEANUP) | |
3529d8c2 | 3010 | return 0; |
c3e330a4 | 3011 | return io_rw_prep_async(req, READ, force_nonblock); |
f67676d1 JA |
3012 | } |
3013 | ||
c1dd91d1 JA |
3014 | /* |
3015 | * This is our waitqueue callback handler, registered through lock_page_async() | |
3016 | * when we initially tried to do the IO with the iocb armed our waitqueue. | |
3017 | * This gets called when the page is unlocked, and we generally expect that to | |
3018 | * happen when the page IO is completed and the page is now uptodate. This will | |
3019 | * queue a task_work based retry of the operation, attempting to copy the data | |
3020 | * again. If the latter fails because the page was NOT uptodate, then we will | |
3021 | * do a thread based blocking retry of the operation. That's the unexpected | |
3022 | * slow path. | |
3023 | */ | |
bcf5a063 JA |
3024 | static int io_async_buf_func(struct wait_queue_entry *wait, unsigned mode, |
3025 | int sync, void *arg) | |
3026 | { | |
3027 | struct wait_page_queue *wpq; | |
3028 | struct io_kiocb *req = wait->private; | |
bcf5a063 | 3029 | struct wait_page_key *key = arg; |
bcf5a063 JA |
3030 | int ret; |
3031 | ||
3032 | wpq = container_of(wait, struct wait_page_queue, wait); | |
3033 | ||
cdc8fcb4 LT |
3034 | if (!wake_page_match(wpq, key)) |
3035 | return 0; | |
3036 | ||
bcf5a063 JA |
3037 | list_del_init(&wait->entry); |
3038 | ||
e7375122 | 3039 | init_task_work(&req->task_work, io_req_task_submit); |
6d816e08 JA |
3040 | percpu_ref_get(&req->ctx->refs); |
3041 | ||
bcf5a063 JA |
3042 | /* submit ref gets dropped, acquire a new one */ |
3043 | refcount_inc(&req->refs); | |
fd7d6de2 | 3044 | ret = io_req_task_work_add(req, &req->task_work, true); |
bcf5a063 | 3045 | if (unlikely(ret)) { |
c2c4c83c JA |
3046 | struct task_struct *tsk; |
3047 | ||
bcf5a063 | 3048 | /* queue just for cancelation */ |
e7375122 | 3049 | init_task_work(&req->task_work, io_req_task_cancel); |
bcf5a063 | 3050 | tsk = io_wq_get_task(req->ctx->io_wq); |
e7375122 | 3051 | task_work_add(tsk, &req->task_work, 0); |
c2c4c83c | 3052 | wake_up_process(tsk); |
bcf5a063 | 3053 | } |
bcf5a063 JA |
3054 | return 1; |
3055 | } | |
3056 | ||
c1dd91d1 JA |
3057 | /* |
3058 | * This controls whether a given IO request should be armed for async page | |
3059 | * based retry. If we return false here, the request is handed to the async | |
3060 | * worker threads for retry. If we're doing buffered reads on a regular file, | |
3061 | * we prepare a private wait_page_queue entry and retry the operation. This | |
3062 | * will either succeed because the page is now uptodate and unlocked, or it | |
3063 | * will register a callback when the page is unlocked at IO completion. Through | |
3064 | * that callback, io_uring uses task_work to setup a retry of the operation. | |
3065 | * That retry will attempt the buffered read again. The retry will generally | |
3066 | * succeed, or in rare cases where it fails, we then fall back to using the | |
3067 | * async worker threads for a blocking retry. | |
3068 | */ | |
227c0c96 | 3069 | static bool io_rw_should_retry(struct io_kiocb *req) |
f67676d1 | 3070 | { |
3b2a4439 | 3071 | struct wait_page_queue *wait = &req->io->rw.wpq; |
bcf5a063 | 3072 | struct kiocb *kiocb = &req->rw.kiocb; |
f67676d1 | 3073 | |
bcf5a063 JA |
3074 | /* never retry for NOWAIT, we just complete with -EAGAIN */ |
3075 | if (req->flags & REQ_F_NOWAIT) | |
3076 | return false; | |
f67676d1 | 3077 | |
227c0c96 | 3078 | /* Only for buffered IO */ |
3b2a4439 | 3079 | if (kiocb->ki_flags & (IOCB_DIRECT | IOCB_HIPRI)) |
bcf5a063 | 3080 | return false; |
3b2a4439 | 3081 | |
bcf5a063 JA |
3082 | /* |
3083 | * just use poll if we can, and don't attempt if the fs doesn't | |
3084 | * support callback based unlocks | |
3085 | */ | |
3086 | if (file_can_poll(req->file) || !(req->file->f_mode & FMODE_BUF_RASYNC)) | |
3087 | return false; | |
f67676d1 | 3088 | |
3b2a4439 JA |
3089 | wait->wait.func = io_async_buf_func; |
3090 | wait->wait.private = req; | |
3091 | wait->wait.flags = 0; | |
3092 | INIT_LIST_HEAD(&wait->wait.entry); | |
3093 | kiocb->ki_flags |= IOCB_WAITQ; | |
3094 | kiocb->ki_waitq = wait; | |
3529d8c2 | 3095 | |
3b2a4439 JA |
3096 | io_get_req_task(req); |
3097 | return true; | |
bcf5a063 JA |
3098 | } |
3099 | ||
3100 | static int io_iter_do_read(struct io_kiocb *req, struct iov_iter *iter) | |
3101 | { | |
3102 | if (req->file->f_op->read_iter) | |
3103 | return call_read_iter(req->file, &req->rw.kiocb, iter); | |
2dd2111d GH |
3104 | else if (req->file->f_op->read) |
3105 | return loop_rw_iter(READ, req->file, &req->rw.kiocb, iter); | |
3106 | else | |
3107 | return -EINVAL; | |
f67676d1 JA |
3108 | } |
3109 | ||
a1d7c393 JA |
3110 | static int io_read(struct io_kiocb *req, bool force_nonblock, |
3111 | struct io_comp_state *cs) | |
2b188cc1 JA |
3112 | { |
3113 | struct iovec inline_vecs[UIO_FASTIOV], *iovec = inline_vecs; | |
9adbd45d | 3114 | struct kiocb *kiocb = &req->rw.kiocb; |
ff6165b2 | 3115 | struct iov_iter __iter, *iter = &__iter; |
227c0c96 | 3116 | ssize_t io_size, ret, ret2; |
31b51510 | 3117 | size_t iov_count; |
ff6165b2 JA |
3118 | |
3119 | if (req->io) | |
3120 | iter = &req->io->rw.iter; | |
2b188cc1 | 3121 | |
ff6165b2 | 3122 | ret = io_import_iovec(READ, req, &iovec, iter, !force_nonblock); |
06b76d44 JA |
3123 | if (ret < 0) |
3124 | return ret; | |
eefdf30f | 3125 | iov_count = iov_iter_count(iter); |
fa15bafb PB |
3126 | io_size = ret; |
3127 | req->result = io_size; | |
227c0c96 | 3128 | ret = 0; |
2b188cc1 | 3129 | |
fd6c2e4c JA |
3130 | /* Ensure we clear previously set non-block flag */ |
3131 | if (!force_nonblock) | |
29de5f6a | 3132 | kiocb->ki_flags &= ~IOCB_NOWAIT; |
fd6c2e4c | 3133 | |
24c74678 | 3134 | /* If the file doesn't support async, just async punt */ |
af197f50 | 3135 | if (force_nonblock && !io_file_supports_async(req->file, READ)) |
f67676d1 | 3136 | goto copy_iov; |
9e645e11 | 3137 | |
0fef9483 | 3138 | ret = rw_verify_area(READ, req->file, io_kiocb_ppos(kiocb), iov_count); |
fa15bafb PB |
3139 | if (unlikely(ret)) |
3140 | goto out_free; | |
2b188cc1 | 3141 | |
227c0c96 | 3142 | ret = io_iter_do_read(req, iter); |
32960613 | 3143 | |
227c0c96 JA |
3144 | if (!ret) { |
3145 | goto done; | |
3146 | } else if (ret == -EIOCBQUEUED) { | |
3147 | ret = 0; | |
3148 | goto out_free; | |
3149 | } else if (ret == -EAGAIN) { | |
eefdf30f JA |
3150 | /* IOPOLL retry should happen for io-wq threads */ |
3151 | if (!force_nonblock && !(req->ctx->flags & IORING_SETUP_IOPOLL)) | |
f91daf56 | 3152 | goto done; |
355afaeb JA |
3153 | /* no retry on NONBLOCK marked file */ |
3154 | if (req->file->f_flags & O_NONBLOCK) | |
3155 | goto done; | |
84216315 JA |
3156 | /* some cases will consume bytes even on error returns */ |
3157 | iov_iter_revert(iter, iov_count - iov_iter_count(iter)); | |
227c0c96 | 3158 | ret = io_setup_async_rw(req, iovec, inline_vecs, iter, false); |
fa15bafb PB |
3159 | if (ret) |
3160 | goto out_free; | |
227c0c96 JA |
3161 | return -EAGAIN; |
3162 | } else if (ret < 0) { | |
00d23d51 JA |
3163 | /* make sure -ERESTARTSYS -> -EINTR is done */ |
3164 | goto done; | |
227c0c96 JA |
3165 | } |
3166 | ||
3167 | /* read it all, or we did blocking attempt. no retry. */ | |
f91daf56 JA |
3168 | if (!iov_iter_count(iter) || !force_nonblock || |
3169 | (req->file->f_flags & O_NONBLOCK)) | |
227c0c96 JA |
3170 | goto done; |
3171 | ||
3172 | io_size -= ret; | |
3173 | copy_iov: | |
3174 | ret2 = io_setup_async_rw(req, iovec, inline_vecs, iter, true); | |
3175 | if (ret2) { | |
3176 | ret = ret2; | |
3177 | goto out_free; | |
3178 | } | |
3179 | /* it's copied and will be cleaned with ->io */ | |
3180 | iovec = NULL; | |
3181 | /* now use our persistent iterator, if we aren't already */ | |
3182 | iter = &req->io->rw.iter; | |
3183 | retry: | |
3184 | req->io->rw.bytes_done += ret; | |
3185 | /* if we can retry, do so with the callbacks armed */ | |
3186 | if (!io_rw_should_retry(req)) { | |
fa15bafb PB |
3187 | kiocb->ki_flags &= ~IOCB_WAITQ; |
3188 | return -EAGAIN; | |
2b188cc1 | 3189 | } |
227c0c96 JA |
3190 | |
3191 | /* | |
3192 | * Now retry read with the IOCB_WAITQ parts set in the iocb. If we | |
3193 | * get -EIOCBQUEUED, then we'll get a notification when the desired | |
3194 | * page gets unlocked. We can also get a partial read here, and if we | |
3195 | * do, then just retry at the new offset. | |
3196 | */ | |
3197 | ret = io_iter_do_read(req, iter); | |
3198 | if (ret == -EIOCBQUEUED) { | |
3199 | ret = 0; | |
3200 | goto out_free; | |
3201 | } else if (ret > 0 && ret < io_size) { | |
3202 | /* we got some bytes, but not all. retry. */ | |
3203 | goto retry; | |
3204 | } | |
3205 | done: | |
3206 | kiocb_done(kiocb, ret, cs); | |
3207 | ret = 0; | |
f67676d1 | 3208 | out_free: |
f261c168 | 3209 | /* it's reportedly faster than delegating the null check to kfree() */ |
252917c3 | 3210 | if (iovec) |
6f2cc166 | 3211 | kfree(iovec); |
2b188cc1 JA |
3212 | return ret; |
3213 | } | |
3214 | ||
3529d8c2 JA |
3215 | static int io_write_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe, |
3216 | bool force_nonblock) | |
f67676d1 JA |
3217 | { |
3218 | ssize_t ret; | |
3219 | ||
3529d8c2 JA |
3220 | ret = io_prep_rw(req, sqe, force_nonblock); |
3221 | if (ret) | |
3222 | return ret; | |
f67676d1 | 3223 | |
3529d8c2 JA |
3224 | if (unlikely(!(req->file->f_mode & FMODE_WRITE))) |
3225 | return -EBADF; | |
f67676d1 | 3226 | |
5f798bea PB |
3227 | /* either don't need iovec imported or already have it */ |
3228 | if (!req->io || req->flags & REQ_F_NEED_CLEANUP) | |
3529d8c2 | 3229 | return 0; |
c3e330a4 | 3230 | return io_rw_prep_async(req, WRITE, force_nonblock); |
f67676d1 JA |
3231 | } |
3232 | ||
a1d7c393 JA |
3233 | static int io_write(struct io_kiocb *req, bool force_nonblock, |
3234 | struct io_comp_state *cs) | |
2b188cc1 JA |
3235 | { |
3236 | struct iovec inline_vecs[UIO_FASTIOV], *iovec = inline_vecs; | |
9adbd45d | 3237 | struct kiocb *kiocb = &req->rw.kiocb; |
ff6165b2 | 3238 | struct iov_iter __iter, *iter = &__iter; |
31b51510 | 3239 | size_t iov_count; |
fa15bafb | 3240 | ssize_t ret, ret2, io_size; |
2b188cc1 | 3241 | |
ff6165b2 JA |
3242 | if (req->io) |
3243 | iter = &req->io->rw.iter; | |
3244 | ||
3245 | ret = io_import_iovec(WRITE, req, &iovec, iter, !force_nonblock); | |
06b76d44 JA |
3246 | if (ret < 0) |
3247 | return ret; | |
eefdf30f | 3248 | iov_count = iov_iter_count(iter); |
fa15bafb PB |
3249 | io_size = ret; |
3250 | req->result = io_size; | |
2b188cc1 | 3251 | |
fd6c2e4c JA |
3252 | /* Ensure we clear previously set non-block flag */ |
3253 | if (!force_nonblock) | |
9adbd45d | 3254 | req->rw.kiocb.ki_flags &= ~IOCB_NOWAIT; |
fd6c2e4c | 3255 | |
24c74678 | 3256 | /* If the file doesn't support async, just async punt */ |
af197f50 | 3257 | if (force_nonblock && !io_file_supports_async(req->file, WRITE)) |
f67676d1 | 3258 | goto copy_iov; |
31b51510 | 3259 | |
10d59345 JA |
3260 | /* file path doesn't support NOWAIT for non-direct_IO */ |
3261 | if (force_nonblock && !(kiocb->ki_flags & IOCB_DIRECT) && | |
3262 | (req->flags & REQ_F_ISREG)) | |
f67676d1 | 3263 | goto copy_iov; |
31b51510 | 3264 | |
0fef9483 | 3265 | ret = rw_verify_area(WRITE, req->file, io_kiocb_ppos(kiocb), iov_count); |
fa15bafb PB |
3266 | if (unlikely(ret)) |
3267 | goto out_free; | |
4ed734b0 | 3268 | |
fa15bafb PB |
3269 | /* |
3270 | * Open-code file_start_write here to grab freeze protection, | |
3271 | * which will be released by another thread in | |
3272 | * io_complete_rw(). Fool lockdep by telling it the lock got | |
3273 | * released so that it doesn't complain about the held lock when | |
3274 | * we return to userspace. | |
3275 | */ | |
3276 | if (req->flags & REQ_F_ISREG) { | |
3277 | __sb_start_write(file_inode(req->file)->i_sb, | |
3278 | SB_FREEZE_WRITE, true); | |
3279 | __sb_writers_release(file_inode(req->file)->i_sb, | |
3280 | SB_FREEZE_WRITE); | |
3281 | } | |
3282 | kiocb->ki_flags |= IOCB_WRITE; | |
4ed734b0 | 3283 | |
fa15bafb | 3284 | if (req->file->f_op->write_iter) |
ff6165b2 | 3285 | ret2 = call_write_iter(req->file, kiocb, iter); |
2dd2111d | 3286 | else if (req->file->f_op->write) |
ff6165b2 | 3287 | ret2 = loop_rw_iter(WRITE, req->file, kiocb, iter); |
2dd2111d GH |
3288 | else |
3289 | ret2 = -EINVAL; | |
4ed734b0 | 3290 | |
fa15bafb PB |
3291 | /* |
3292 | * Raw bdev writes will return -EOPNOTSUPP for IOCB_NOWAIT. Just | |
3293 | * retry them without IOCB_NOWAIT. | |
3294 | */ | |
3295 | if (ret2 == -EOPNOTSUPP && (kiocb->ki_flags & IOCB_NOWAIT)) | |
3296 | ret2 = -EAGAIN; | |
355afaeb JA |
3297 | /* no retry on NONBLOCK marked file */ |
3298 | if (ret2 == -EAGAIN && (req->file->f_flags & O_NONBLOCK)) | |
3299 | goto done; | |
fa15bafb | 3300 | if (!force_nonblock || ret2 != -EAGAIN) { |
eefdf30f JA |
3301 | /* IOPOLL retry should happen for io-wq threads */ |
3302 | if ((req->ctx->flags & IORING_SETUP_IOPOLL) && ret2 == -EAGAIN) | |
3303 | goto copy_iov; | |
355afaeb | 3304 | done: |
fa15bafb PB |
3305 | kiocb_done(kiocb, ret2, cs); |
3306 | } else { | |
eefdf30f | 3307 | copy_iov: |
84216315 JA |
3308 | /* some cases will consume bytes even on error returns */ |
3309 | iov_iter_revert(iter, iov_count - iov_iter_count(iter)); | |
227c0c96 | 3310 | ret = io_setup_async_rw(req, iovec, inline_vecs, iter, false); |
ff6165b2 JA |
3311 | if (!ret) |
3312 | return -EAGAIN; | |
2b188cc1 | 3313 | } |
31b51510 | 3314 | out_free: |
f261c168 | 3315 | /* it's reportedly faster than delegating the null check to kfree() */ |
252917c3 | 3316 | if (iovec) |
6f2cc166 | 3317 | kfree(iovec); |
2b188cc1 JA |
3318 | return ret; |
3319 | } | |
3320 | ||
f2a8d5c7 PB |
3321 | static int __io_splice_prep(struct io_kiocb *req, |
3322 | const struct io_uring_sqe *sqe) | |
7d67af2c PB |
3323 | { |
3324 | struct io_splice* sp = &req->splice; | |
3325 | unsigned int valid_flags = SPLICE_F_FD_IN_FIXED | SPLICE_F_ALL; | |
3326 | int ret; | |
3327 | ||
3328 | if (req->flags & REQ_F_NEED_CLEANUP) | |
3329 | return 0; | |
3232dd02 PB |
3330 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
3331 | return -EINVAL; | |
7d67af2c PB |
3332 | |
3333 | sp->file_in = NULL; | |
7d67af2c PB |
3334 | sp->len = READ_ONCE(sqe->len); |
3335 | sp->flags = READ_ONCE(sqe->splice_flags); | |
3336 | ||
3337 | if (unlikely(sp->flags & ~valid_flags)) | |
3338 | return -EINVAL; | |
3339 | ||
3340 | ret = io_file_get(NULL, req, READ_ONCE(sqe->splice_fd_in), &sp->file_in, | |
3341 | (sp->flags & SPLICE_F_FD_IN_FIXED)); | |
3342 | if (ret) | |
3343 | return ret; | |
3344 | req->flags |= REQ_F_NEED_CLEANUP; | |
3345 | ||
7cdaf587 XW |
3346 | if (!S_ISREG(file_inode(sp->file_in)->i_mode)) { |
3347 | /* | |
3348 | * Splice operation will be punted aync, and here need to | |
3349 | * modify io_wq_work.flags, so initialize io_wq_work firstly. | |
3350 | */ | |
3351 | io_req_init_async(req); | |
7d67af2c | 3352 | req->work.flags |= IO_WQ_WORK_UNBOUND; |
7cdaf587 | 3353 | } |
7d67af2c PB |
3354 | |
3355 | return 0; | |
3356 | } | |
3357 | ||
f2a8d5c7 PB |
3358 | static int io_tee_prep(struct io_kiocb *req, |
3359 | const struct io_uring_sqe *sqe) | |
3360 | { | |
3361 | if (READ_ONCE(sqe->splice_off_in) || READ_ONCE(sqe->off)) | |
3362 | return -EINVAL; | |
3363 | return __io_splice_prep(req, sqe); | |
3364 | } | |
3365 | ||
3366 | static int io_tee(struct io_kiocb *req, bool force_nonblock) | |
3367 | { | |
3368 | struct io_splice *sp = &req->splice; | |
3369 | struct file *in = sp->file_in; | |
3370 | struct file *out = sp->file_out; | |
3371 | unsigned int flags = sp->flags & ~SPLICE_F_FD_IN_FIXED; | |
3372 | long ret = 0; | |
3373 | ||
3374 | if (force_nonblock) | |
3375 | return -EAGAIN; | |
3376 | if (sp->len) | |
3377 | ret = do_tee(in, out, sp->len, flags); | |
3378 | ||
3379 | io_put_file(req, in, (sp->flags & SPLICE_F_FD_IN_FIXED)); | |
3380 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
3381 | ||
f2a8d5c7 PB |
3382 | if (ret != sp->len) |
3383 | req_set_fail_links(req); | |
e1e16097 | 3384 | io_req_complete(req, ret); |
f2a8d5c7 PB |
3385 | return 0; |
3386 | } | |
3387 | ||
3388 | static int io_splice_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) | |
3389 | { | |
3390 | struct io_splice* sp = &req->splice; | |
3391 | ||
3392 | sp->off_in = READ_ONCE(sqe->splice_off_in); | |
3393 | sp->off_out = READ_ONCE(sqe->off); | |
3394 | return __io_splice_prep(req, sqe); | |
3395 | } | |
3396 | ||
014db007 | 3397 | static int io_splice(struct io_kiocb *req, bool force_nonblock) |
7d67af2c PB |
3398 | { |
3399 | struct io_splice *sp = &req->splice; | |
3400 | struct file *in = sp->file_in; | |
3401 | struct file *out = sp->file_out; | |
3402 | unsigned int flags = sp->flags & ~SPLICE_F_FD_IN_FIXED; | |
3403 | loff_t *poff_in, *poff_out; | |
c9687426 | 3404 | long ret = 0; |
7d67af2c | 3405 | |
2fb3e822 PB |
3406 | if (force_nonblock) |
3407 | return -EAGAIN; | |
7d67af2c PB |
3408 | |
3409 | poff_in = (sp->off_in == -1) ? NULL : &sp->off_in; | |
3410 | poff_out = (sp->off_out == -1) ? NULL : &sp->off_out; | |
c9687426 | 3411 | |
948a7749 | 3412 | if (sp->len) |
c9687426 | 3413 | ret = do_splice(in, poff_in, out, poff_out, sp->len, flags); |
7d67af2c PB |
3414 | |
3415 | io_put_file(req, in, (sp->flags & SPLICE_F_FD_IN_FIXED)); | |
3416 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
3417 | ||
7d67af2c PB |
3418 | if (ret != sp->len) |
3419 | req_set_fail_links(req); | |
e1e16097 | 3420 | io_req_complete(req, ret); |
7d67af2c PB |
3421 | return 0; |
3422 | } | |
3423 | ||
2b188cc1 JA |
3424 | /* |
3425 | * IORING_OP_NOP just posts a completion event, nothing else. | |
3426 | */ | |
229a7b63 | 3427 | static int io_nop(struct io_kiocb *req, struct io_comp_state *cs) |
2b188cc1 JA |
3428 | { |
3429 | struct io_ring_ctx *ctx = req->ctx; | |
2b188cc1 | 3430 | |
def596e9 JA |
3431 | if (unlikely(ctx->flags & IORING_SETUP_IOPOLL)) |
3432 | return -EINVAL; | |
3433 | ||
229a7b63 | 3434 | __io_req_complete(req, 0, 0, cs); |
2b188cc1 JA |
3435 | return 0; |
3436 | } | |
3437 | ||
3529d8c2 | 3438 | static int io_prep_fsync(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
c992fe29 | 3439 | { |
6b06314c | 3440 | struct io_ring_ctx *ctx = req->ctx; |
c992fe29 | 3441 | |
09bb8394 JA |
3442 | if (!req->file) |
3443 | return -EBADF; | |
c992fe29 | 3444 | |
6b06314c | 3445 | if (unlikely(ctx->flags & IORING_SETUP_IOPOLL)) |
def596e9 | 3446 | return -EINVAL; |
edafccee | 3447 | if (unlikely(sqe->addr || sqe->ioprio || sqe->buf_index)) |
c992fe29 CH |
3448 | return -EINVAL; |
3449 | ||
8ed8d3c3 JA |
3450 | req->sync.flags = READ_ONCE(sqe->fsync_flags); |
3451 | if (unlikely(req->sync.flags & ~IORING_FSYNC_DATASYNC)) | |
3452 | return -EINVAL; | |
3453 | ||
3454 | req->sync.off = READ_ONCE(sqe->off); | |
3455 | req->sync.len = READ_ONCE(sqe->len); | |
c992fe29 CH |
3456 | return 0; |
3457 | } | |
3458 | ||
ac45abc0 | 3459 | static int io_fsync(struct io_kiocb *req, bool force_nonblock) |
8ed8d3c3 | 3460 | { |
8ed8d3c3 | 3461 | loff_t end = req->sync.off + req->sync.len; |
8ed8d3c3 JA |
3462 | int ret; |
3463 | ||
ac45abc0 PB |
3464 | /* fsync always requires a blocking context */ |
3465 | if (force_nonblock) | |
3466 | return -EAGAIN; | |
3467 | ||
9adbd45d | 3468 | ret = vfs_fsync_range(req->file, req->sync.off, |
8ed8d3c3 JA |
3469 | end > 0 ? end : LLONG_MAX, |
3470 | req->sync.flags & IORING_FSYNC_DATASYNC); | |
3471 | if (ret < 0) | |
3472 | req_set_fail_links(req); | |
e1e16097 | 3473 | io_req_complete(req, ret); |
c992fe29 CH |
3474 | return 0; |
3475 | } | |
3476 | ||
d63d1b5e JA |
3477 | static int io_fallocate_prep(struct io_kiocb *req, |
3478 | const struct io_uring_sqe *sqe) | |
3479 | { | |
3480 | if (sqe->ioprio || sqe->buf_index || sqe->rw_flags) | |
3481 | return -EINVAL; | |
3232dd02 PB |
3482 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
3483 | return -EINVAL; | |
d63d1b5e JA |
3484 | |
3485 | req->sync.off = READ_ONCE(sqe->off); | |
3486 | req->sync.len = READ_ONCE(sqe->addr); | |
3487 | req->sync.mode = READ_ONCE(sqe->len); | |
3488 | return 0; | |
3489 | } | |
3490 | ||
014db007 | 3491 | static int io_fallocate(struct io_kiocb *req, bool force_nonblock) |
5d17b4a4 | 3492 | { |
ac45abc0 PB |
3493 | int ret; |
3494 | ||
d63d1b5e | 3495 | /* fallocate always requiring blocking context */ |
ac45abc0 | 3496 | if (force_nonblock) |
5d17b4a4 | 3497 | return -EAGAIN; |
ac45abc0 PB |
3498 | ret = vfs_fallocate(req->file, req->sync.mode, req->sync.off, |
3499 | req->sync.len); | |
ac45abc0 PB |
3500 | if (ret < 0) |
3501 | req_set_fail_links(req); | |
e1e16097 | 3502 | io_req_complete(req, ret); |
5d17b4a4 JA |
3503 | return 0; |
3504 | } | |
3505 | ||
ec65fea5 | 3506 | static int __io_openat_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
b7bb4f7d | 3507 | { |
f8748881 | 3508 | const char __user *fname; |
15b71abe | 3509 | int ret; |
b7bb4f7d | 3510 | |
3232dd02 | 3511 | if (unlikely(req->ctx->flags & (IORING_SETUP_IOPOLL|IORING_SETUP_SQPOLL))) |
15b71abe | 3512 | return -EINVAL; |
ec65fea5 | 3513 | if (unlikely(sqe->ioprio || sqe->buf_index)) |
15b71abe | 3514 | return -EINVAL; |
ec65fea5 | 3515 | if (unlikely(req->flags & REQ_F_FIXED_FILE)) |
cf3040ca | 3516 | return -EBADF; |
03b1230c | 3517 | |
ec65fea5 PB |
3518 | /* open.how should be already initialised */ |
3519 | if (!(req->open.how.flags & O_PATH) && force_o_largefile()) | |
08a1d26e | 3520 | req->open.how.flags |= O_LARGEFILE; |
3529d8c2 | 3521 | |
25e72d10 PB |
3522 | req->open.dfd = READ_ONCE(sqe->fd); |
3523 | fname = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
f8748881 | 3524 | req->open.filename = getname(fname); |
15b71abe JA |
3525 | if (IS_ERR(req->open.filename)) { |
3526 | ret = PTR_ERR(req->open.filename); | |
3527 | req->open.filename = NULL; | |
3528 | return ret; | |
3529 | } | |
4022e7af | 3530 | req->open.nofile = rlimit(RLIMIT_NOFILE); |
8fef80bf | 3531 | req->flags |= REQ_F_NEED_CLEANUP; |
15b71abe | 3532 | return 0; |
03b1230c JA |
3533 | } |
3534 | ||
ec65fea5 PB |
3535 | static int io_openat_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
3536 | { | |
3537 | u64 flags, mode; | |
3538 | ||
3539 | if (req->flags & REQ_F_NEED_CLEANUP) | |
3540 | return 0; | |
3541 | mode = READ_ONCE(sqe->len); | |
3542 | flags = READ_ONCE(sqe->open_flags); | |
3543 | req->open.how = build_open_how(flags, mode); | |
3544 | return __io_openat_prep(req, sqe); | |
3545 | } | |
3546 | ||
cebdb986 | 3547 | static int io_openat2_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
aa1fa28f | 3548 | { |
cebdb986 | 3549 | struct open_how __user *how; |
cebdb986 | 3550 | size_t len; |
0fa03c62 JA |
3551 | int ret; |
3552 | ||
0bdbdd08 PB |
3553 | if (req->flags & REQ_F_NEED_CLEANUP) |
3554 | return 0; | |
cebdb986 JA |
3555 | how = u64_to_user_ptr(READ_ONCE(sqe->addr2)); |
3556 | len = READ_ONCE(sqe->len); | |
cebdb986 JA |
3557 | if (len < OPEN_HOW_SIZE_VER0) |
3558 | return -EINVAL; | |
3529d8c2 | 3559 | |
cebdb986 JA |
3560 | ret = copy_struct_from_user(&req->open.how, sizeof(req->open.how), how, |
3561 | len); | |
3562 | if (ret) | |
3563 | return ret; | |
3529d8c2 | 3564 | |
ec65fea5 | 3565 | return __io_openat_prep(req, sqe); |
cebdb986 JA |
3566 | } |
3567 | ||
014db007 | 3568 | static int io_openat2(struct io_kiocb *req, bool force_nonblock) |
15b71abe JA |
3569 | { |
3570 | struct open_flags op; | |
15b71abe JA |
3571 | struct file *file; |
3572 | int ret; | |
3573 | ||
f86cd20c | 3574 | if (force_nonblock) |
15b71abe | 3575 | return -EAGAIN; |
15b71abe | 3576 | |
cebdb986 | 3577 | ret = build_open_flags(&req->open.how, &op); |
15b71abe JA |
3578 | if (ret) |
3579 | goto err; | |
3580 | ||
4022e7af | 3581 | ret = __get_unused_fd_flags(req->open.how.flags, req->open.nofile); |
15b71abe JA |
3582 | if (ret < 0) |
3583 | goto err; | |
3584 | ||
3585 | file = do_filp_open(req->open.dfd, req->open.filename, &op); | |
3586 | if (IS_ERR(file)) { | |
3587 | put_unused_fd(ret); | |
3588 | ret = PTR_ERR(file); | |
3589 | } else { | |
3590 | fsnotify_open(file); | |
3591 | fd_install(ret, file); | |
3592 | } | |
3593 | err: | |
3594 | putname(req->open.filename); | |
8fef80bf | 3595 | req->flags &= ~REQ_F_NEED_CLEANUP; |
15b71abe JA |
3596 | if (ret < 0) |
3597 | req_set_fail_links(req); | |
e1e16097 | 3598 | io_req_complete(req, ret); |
15b71abe JA |
3599 | return 0; |
3600 | } | |
3601 | ||
014db007 | 3602 | static int io_openat(struct io_kiocb *req, bool force_nonblock) |
cebdb986 | 3603 | { |
014db007 | 3604 | return io_openat2(req, force_nonblock); |
cebdb986 JA |
3605 | } |
3606 | ||
067524e9 JA |
3607 | static int io_remove_buffers_prep(struct io_kiocb *req, |
3608 | const struct io_uring_sqe *sqe) | |
3609 | { | |
3610 | struct io_provide_buf *p = &req->pbuf; | |
3611 | u64 tmp; | |
3612 | ||
3613 | if (sqe->ioprio || sqe->rw_flags || sqe->addr || sqe->len || sqe->off) | |
3614 | return -EINVAL; | |
3615 | ||
3616 | tmp = READ_ONCE(sqe->fd); | |
3617 | if (!tmp || tmp > USHRT_MAX) | |
3618 | return -EINVAL; | |
3619 | ||
3620 | memset(p, 0, sizeof(*p)); | |
3621 | p->nbufs = tmp; | |
3622 | p->bgid = READ_ONCE(sqe->buf_group); | |
3623 | return 0; | |
3624 | } | |
3625 | ||
3626 | static int __io_remove_buffers(struct io_ring_ctx *ctx, struct io_buffer *buf, | |
3627 | int bgid, unsigned nbufs) | |
3628 | { | |
3629 | unsigned i = 0; | |
3630 | ||
3631 | /* shouldn't happen */ | |
3632 | if (!nbufs) | |
3633 | return 0; | |
3634 | ||
3635 | /* the head kbuf is the list itself */ | |
3636 | while (!list_empty(&buf->list)) { | |
3637 | struct io_buffer *nxt; | |
3638 | ||
3639 | nxt = list_first_entry(&buf->list, struct io_buffer, list); | |
3640 | list_del(&nxt->list); | |
3641 | kfree(nxt); | |
3642 | if (++i == nbufs) | |
3643 | return i; | |
3644 | } | |
3645 | i++; | |
3646 | kfree(buf); | |
3647 | idr_remove(&ctx->io_buffer_idr, bgid); | |
3648 | ||
3649 | return i; | |
3650 | } | |
3651 | ||
229a7b63 JA |
3652 | static int io_remove_buffers(struct io_kiocb *req, bool force_nonblock, |
3653 | struct io_comp_state *cs) | |
067524e9 JA |
3654 | { |
3655 | struct io_provide_buf *p = &req->pbuf; | |
3656 | struct io_ring_ctx *ctx = req->ctx; | |
3657 | struct io_buffer *head; | |
3658 | int ret = 0; | |
3659 | ||
3660 | io_ring_submit_lock(ctx, !force_nonblock); | |
3661 | ||
3662 | lockdep_assert_held(&ctx->uring_lock); | |
3663 | ||
3664 | ret = -ENOENT; | |
3665 | head = idr_find(&ctx->io_buffer_idr, p->bgid); | |
3666 | if (head) | |
3667 | ret = __io_remove_buffers(ctx, head, p->bgid, p->nbufs); | |
3668 | ||
3669 | io_ring_submit_lock(ctx, !force_nonblock); | |
3670 | if (ret < 0) | |
3671 | req_set_fail_links(req); | |
229a7b63 | 3672 | __io_req_complete(req, ret, 0, cs); |
067524e9 JA |
3673 | return 0; |
3674 | } | |
3675 | ||
ddf0322d JA |
3676 | static int io_provide_buffers_prep(struct io_kiocb *req, |
3677 | const struct io_uring_sqe *sqe) | |
3678 | { | |
3679 | struct io_provide_buf *p = &req->pbuf; | |
3680 | u64 tmp; | |
3681 | ||
3682 | if (sqe->ioprio || sqe->rw_flags) | |
3683 | return -EINVAL; | |
3684 | ||
3685 | tmp = READ_ONCE(sqe->fd); | |
3686 | if (!tmp || tmp > USHRT_MAX) | |
3687 | return -E2BIG; | |
3688 | p->nbufs = tmp; | |
3689 | p->addr = READ_ONCE(sqe->addr); | |
3690 | p->len = READ_ONCE(sqe->len); | |
3691 | ||
efe68c1c | 3692 | if (!access_ok(u64_to_user_ptr(p->addr), (p->len * p->nbufs))) |
ddf0322d JA |
3693 | return -EFAULT; |
3694 | ||
3695 | p->bgid = READ_ONCE(sqe->buf_group); | |
3696 | tmp = READ_ONCE(sqe->off); | |
3697 | if (tmp > USHRT_MAX) | |
3698 | return -E2BIG; | |
3699 | p->bid = tmp; | |
3700 | return 0; | |
3701 | } | |
3702 | ||
3703 | static int io_add_buffers(struct io_provide_buf *pbuf, struct io_buffer **head) | |
3704 | { | |
3705 | struct io_buffer *buf; | |
3706 | u64 addr = pbuf->addr; | |
3707 | int i, bid = pbuf->bid; | |
3708 | ||
3709 | for (i = 0; i < pbuf->nbufs; i++) { | |
3710 | buf = kmalloc(sizeof(*buf), GFP_KERNEL); | |
3711 | if (!buf) | |
3712 | break; | |
3713 | ||
3714 | buf->addr = addr; | |
3715 | buf->len = pbuf->len; | |
3716 | buf->bid = bid; | |
3717 | addr += pbuf->len; | |
3718 | bid++; | |
3719 | if (!*head) { | |
3720 | INIT_LIST_HEAD(&buf->list); | |
3721 | *head = buf; | |
3722 | } else { | |
3723 | list_add_tail(&buf->list, &(*head)->list); | |
3724 | } | |
3725 | } | |
3726 | ||
3727 | return i ? i : -ENOMEM; | |
3728 | } | |
3729 | ||
229a7b63 JA |
3730 | static int io_provide_buffers(struct io_kiocb *req, bool force_nonblock, |
3731 | struct io_comp_state *cs) | |
ddf0322d JA |
3732 | { |
3733 | struct io_provide_buf *p = &req->pbuf; | |
3734 | struct io_ring_ctx *ctx = req->ctx; | |
3735 | struct io_buffer *head, *list; | |
3736 | int ret = 0; | |
3737 | ||
3738 | io_ring_submit_lock(ctx, !force_nonblock); | |
3739 | ||
3740 | lockdep_assert_held(&ctx->uring_lock); | |
3741 | ||
3742 | list = head = idr_find(&ctx->io_buffer_idr, p->bgid); | |
3743 | ||
3744 | ret = io_add_buffers(p, &head); | |
3745 | if (ret < 0) | |
3746 | goto out; | |
3747 | ||
3748 | if (!list) { | |
3749 | ret = idr_alloc(&ctx->io_buffer_idr, head, p->bgid, p->bgid + 1, | |
3750 | GFP_KERNEL); | |
3751 | if (ret < 0) { | |
067524e9 | 3752 | __io_remove_buffers(ctx, head, p->bgid, -1U); |
ddf0322d JA |
3753 | goto out; |
3754 | } | |
3755 | } | |
3756 | out: | |
3757 | io_ring_submit_unlock(ctx, !force_nonblock); | |
3758 | if (ret < 0) | |
3759 | req_set_fail_links(req); | |
229a7b63 | 3760 | __io_req_complete(req, ret, 0, cs); |
ddf0322d | 3761 | return 0; |
cebdb986 JA |
3762 | } |
3763 | ||
3e4827b0 JA |
3764 | static int io_epoll_ctl_prep(struct io_kiocb *req, |
3765 | const struct io_uring_sqe *sqe) | |
3766 | { | |
3767 | #if defined(CONFIG_EPOLL) | |
3768 | if (sqe->ioprio || sqe->buf_index) | |
3769 | return -EINVAL; | |
3232dd02 PB |
3770 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
3771 | return -EINVAL; | |
3e4827b0 JA |
3772 | |
3773 | req->epoll.epfd = READ_ONCE(sqe->fd); | |
3774 | req->epoll.op = READ_ONCE(sqe->len); | |
3775 | req->epoll.fd = READ_ONCE(sqe->off); | |
3776 | ||
3777 | if (ep_op_has_event(req->epoll.op)) { | |
3778 | struct epoll_event __user *ev; | |
3779 | ||
3780 | ev = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
3781 | if (copy_from_user(&req->epoll.event, ev, sizeof(*ev))) | |
3782 | return -EFAULT; | |
3783 | } | |
3784 | ||
3785 | return 0; | |
3786 | #else | |
3787 | return -EOPNOTSUPP; | |
3788 | #endif | |
3789 | } | |
3790 | ||
229a7b63 JA |
3791 | static int io_epoll_ctl(struct io_kiocb *req, bool force_nonblock, |
3792 | struct io_comp_state *cs) | |
3e4827b0 JA |
3793 | { |
3794 | #if defined(CONFIG_EPOLL) | |
3795 | struct io_epoll *ie = &req->epoll; | |
3796 | int ret; | |
3797 | ||
3798 | ret = do_epoll_ctl(ie->epfd, ie->op, ie->fd, &ie->event, force_nonblock); | |
3799 | if (force_nonblock && ret == -EAGAIN) | |
3800 | return -EAGAIN; | |
3801 | ||
3802 | if (ret < 0) | |
3803 | req_set_fail_links(req); | |
229a7b63 | 3804 | __io_req_complete(req, ret, 0, cs); |
3e4827b0 JA |
3805 | return 0; |
3806 | #else | |
3807 | return -EOPNOTSUPP; | |
3808 | #endif | |
3809 | } | |
3810 | ||
c1ca757b JA |
3811 | static int io_madvise_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
3812 | { | |
3813 | #if defined(CONFIG_ADVISE_SYSCALLS) && defined(CONFIG_MMU) | |
3814 | if (sqe->ioprio || sqe->buf_index || sqe->off) | |
3815 | return -EINVAL; | |
3232dd02 PB |
3816 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
3817 | return -EINVAL; | |
c1ca757b JA |
3818 | |
3819 | req->madvise.addr = READ_ONCE(sqe->addr); | |
3820 | req->madvise.len = READ_ONCE(sqe->len); | |
3821 | req->madvise.advice = READ_ONCE(sqe->fadvise_advice); | |
3822 | return 0; | |
3823 | #else | |
3824 | return -EOPNOTSUPP; | |
3825 | #endif | |
3826 | } | |
3827 | ||
014db007 | 3828 | static int io_madvise(struct io_kiocb *req, bool force_nonblock) |
c1ca757b JA |
3829 | { |
3830 | #if defined(CONFIG_ADVISE_SYSCALLS) && defined(CONFIG_MMU) | |
3831 | struct io_madvise *ma = &req->madvise; | |
3832 | int ret; | |
3833 | ||
3834 | if (force_nonblock) | |
3835 | return -EAGAIN; | |
3836 | ||
3837 | ret = do_madvise(ma->addr, ma->len, ma->advice); | |
3838 | if (ret < 0) | |
3839 | req_set_fail_links(req); | |
e1e16097 | 3840 | io_req_complete(req, ret); |
c1ca757b JA |
3841 | return 0; |
3842 | #else | |
3843 | return -EOPNOTSUPP; | |
3844 | #endif | |
3845 | } | |
3846 | ||
4840e418 JA |
3847 | static int io_fadvise_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
3848 | { | |
3849 | if (sqe->ioprio || sqe->buf_index || sqe->addr) | |
3850 | return -EINVAL; | |
3232dd02 PB |
3851 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
3852 | return -EINVAL; | |
4840e418 JA |
3853 | |
3854 | req->fadvise.offset = READ_ONCE(sqe->off); | |
3855 | req->fadvise.len = READ_ONCE(sqe->len); | |
3856 | req->fadvise.advice = READ_ONCE(sqe->fadvise_advice); | |
3857 | return 0; | |
3858 | } | |
3859 | ||
014db007 | 3860 | static int io_fadvise(struct io_kiocb *req, bool force_nonblock) |
4840e418 JA |
3861 | { |
3862 | struct io_fadvise *fa = &req->fadvise; | |
3863 | int ret; | |
3864 | ||
3e69426d JA |
3865 | if (force_nonblock) { |
3866 | switch (fa->advice) { | |
3867 | case POSIX_FADV_NORMAL: | |
3868 | case POSIX_FADV_RANDOM: | |
3869 | case POSIX_FADV_SEQUENTIAL: | |
3870 | break; | |
3871 | default: | |
3872 | return -EAGAIN; | |
3873 | } | |
3874 | } | |
4840e418 JA |
3875 | |
3876 | ret = vfs_fadvise(req->file, fa->offset, fa->len, fa->advice); | |
3877 | if (ret < 0) | |
3878 | req_set_fail_links(req); | |
e1e16097 | 3879 | io_req_complete(req, ret); |
4840e418 JA |
3880 | return 0; |
3881 | } | |
3882 | ||
eddc7ef5 JA |
3883 | static int io_statx_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
3884 | { | |
3232dd02 PB |
3885 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
3886 | return -EINVAL; | |
eddc7ef5 JA |
3887 | if (sqe->ioprio || sqe->buf_index) |
3888 | return -EINVAL; | |
9c280f90 | 3889 | if (req->flags & REQ_F_FIXED_FILE) |
cf3040ca | 3890 | return -EBADF; |
eddc7ef5 | 3891 | |
1d9e1288 BM |
3892 | req->statx.dfd = READ_ONCE(sqe->fd); |
3893 | req->statx.mask = READ_ONCE(sqe->len); | |
e62753e4 | 3894 | req->statx.filename = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
1d9e1288 BM |
3895 | req->statx.buffer = u64_to_user_ptr(READ_ONCE(sqe->addr2)); |
3896 | req->statx.flags = READ_ONCE(sqe->statx_flags); | |
eddc7ef5 JA |
3897 | |
3898 | return 0; | |
3899 | } | |
3900 | ||
014db007 | 3901 | static int io_statx(struct io_kiocb *req, bool force_nonblock) |
eddc7ef5 | 3902 | { |
1d9e1288 | 3903 | struct io_statx *ctx = &req->statx; |
eddc7ef5 JA |
3904 | int ret; |
3905 | ||
5b0bbee4 JA |
3906 | if (force_nonblock) { |
3907 | /* only need file table for an actual valid fd */ | |
3908 | if (ctx->dfd == -1 || ctx->dfd == AT_FDCWD) | |
3909 | req->flags |= REQ_F_NO_FILE_TABLE; | |
eddc7ef5 | 3910 | return -EAGAIN; |
5b0bbee4 | 3911 | } |
eddc7ef5 | 3912 | |
e62753e4 BM |
3913 | ret = do_statx(ctx->dfd, ctx->filename, ctx->flags, ctx->mask, |
3914 | ctx->buffer); | |
eddc7ef5 | 3915 | |
eddc7ef5 JA |
3916 | if (ret < 0) |
3917 | req_set_fail_links(req); | |
e1e16097 | 3918 | io_req_complete(req, ret); |
eddc7ef5 JA |
3919 | return 0; |
3920 | } | |
3921 | ||
b5dba59e JA |
3922 | static int io_close_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
3923 | { | |
3924 | /* | |
3925 | * If we queue this for async, it must not be cancellable. That would | |
7cdaf587 XW |
3926 | * leave the 'file' in an undeterminate state, and here need to modify |
3927 | * io_wq_work.flags, so initialize io_wq_work firstly. | |
b5dba59e | 3928 | */ |
7cdaf587 | 3929 | io_req_init_async(req); |
b5dba59e JA |
3930 | req->work.flags |= IO_WQ_WORK_NO_CANCEL; |
3931 | ||
3232dd02 PB |
3932 | if (unlikely(req->ctx->flags & (IORING_SETUP_IOPOLL|IORING_SETUP_SQPOLL))) |
3933 | return -EINVAL; | |
b5dba59e JA |
3934 | if (sqe->ioprio || sqe->off || sqe->addr || sqe->len || |
3935 | sqe->rw_flags || sqe->buf_index) | |
3936 | return -EINVAL; | |
9c280f90 | 3937 | if (req->flags & REQ_F_FIXED_FILE) |
cf3040ca | 3938 | return -EBADF; |
b5dba59e JA |
3939 | |
3940 | req->close.fd = READ_ONCE(sqe->fd); | |
fd2206e4 JA |
3941 | if ((req->file && req->file->f_op == &io_uring_fops) || |
3942 | req->close.fd == req->ctx->ring_fd) | |
3943 | return -EBADF; | |
b5dba59e | 3944 | |
3af73b28 | 3945 | req->close.put_file = NULL; |
b5dba59e | 3946 | return 0; |
b5dba59e JA |
3947 | } |
3948 | ||
229a7b63 JA |
3949 | static int io_close(struct io_kiocb *req, bool force_nonblock, |
3950 | struct io_comp_state *cs) | |
b5dba59e | 3951 | { |
3af73b28 | 3952 | struct io_close *close = &req->close; |
b5dba59e JA |
3953 | int ret; |
3954 | ||
3af73b28 PB |
3955 | /* might be already done during nonblock submission */ |
3956 | if (!close->put_file) { | |
3957 | ret = __close_fd_get_file(close->fd, &close->put_file); | |
3958 | if (ret < 0) | |
3959 | return (ret == -ENOENT) ? -EBADF : ret; | |
3960 | } | |
b5dba59e JA |
3961 | |
3962 | /* if the file has a flush method, be safe and punt to async */ | |
3af73b28 | 3963 | if (close->put_file->f_op->flush && force_nonblock) { |
24c74678 PB |
3964 | /* was never set, but play safe */ |
3965 | req->flags &= ~REQ_F_NOWAIT; | |
0bf0eefd | 3966 | /* avoid grabbing files - we don't need the files */ |
24c74678 | 3967 | req->flags |= REQ_F_NO_FILE_TABLE; |
0bf0eefd | 3968 | return -EAGAIN; |
a2100672 | 3969 | } |
b5dba59e | 3970 | |
3af73b28 PB |
3971 | /* No ->flush() or already async, safely close from here */ |
3972 | ret = filp_close(close->put_file, req->work.files); | |
3973 | if (ret < 0) | |
3974 | req_set_fail_links(req); | |
3af73b28 PB |
3975 | fput(close->put_file); |
3976 | close->put_file = NULL; | |
229a7b63 | 3977 | __io_req_complete(req, ret, 0, cs); |
1a417f4e | 3978 | return 0; |
b5dba59e JA |
3979 | } |
3980 | ||
3529d8c2 | 3981 | static int io_prep_sfr(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
5d17b4a4 JA |
3982 | { |
3983 | struct io_ring_ctx *ctx = req->ctx; | |
5d17b4a4 JA |
3984 | |
3985 | if (!req->file) | |
3986 | return -EBADF; | |
5d17b4a4 JA |
3987 | |
3988 | if (unlikely(ctx->flags & IORING_SETUP_IOPOLL)) | |
3989 | return -EINVAL; | |
3990 | if (unlikely(sqe->addr || sqe->ioprio || sqe->buf_index)) | |
3991 | return -EINVAL; | |
3992 | ||
8ed8d3c3 JA |
3993 | req->sync.off = READ_ONCE(sqe->off); |
3994 | req->sync.len = READ_ONCE(sqe->len); | |
3995 | req->sync.flags = READ_ONCE(sqe->sync_range_flags); | |
8ed8d3c3 JA |
3996 | return 0; |
3997 | } | |
3998 | ||
ac45abc0 | 3999 | static int io_sync_file_range(struct io_kiocb *req, bool force_nonblock) |
8ed8d3c3 | 4000 | { |
8ed8d3c3 JA |
4001 | int ret; |
4002 | ||
ac45abc0 PB |
4003 | /* sync_file_range always requires a blocking context */ |
4004 | if (force_nonblock) | |
4005 | return -EAGAIN; | |
4006 | ||
9adbd45d | 4007 | ret = sync_file_range(req->file, req->sync.off, req->sync.len, |
8ed8d3c3 JA |
4008 | req->sync.flags); |
4009 | if (ret < 0) | |
4010 | req_set_fail_links(req); | |
e1e16097 | 4011 | io_req_complete(req, ret); |
5d17b4a4 JA |
4012 | return 0; |
4013 | } | |
4014 | ||
469956e8 | 4015 | #if defined(CONFIG_NET) |
02d27d89 PB |
4016 | static int io_setup_async_msg(struct io_kiocb *req, |
4017 | struct io_async_msghdr *kmsg) | |
4018 | { | |
4019 | if (req->io) | |
4020 | return -EAGAIN; | |
4021 | if (io_alloc_async_ctx(req)) { | |
4022 | if (kmsg->iov != kmsg->fast_iov) | |
4023 | kfree(kmsg->iov); | |
4024 | return -ENOMEM; | |
4025 | } | |
4026 | req->flags |= REQ_F_NEED_CLEANUP; | |
4027 | memcpy(&req->io->msg, kmsg, sizeof(*kmsg)); | |
4028 | return -EAGAIN; | |
4029 | } | |
4030 | ||
2ae523ed PB |
4031 | static int io_sendmsg_copy_hdr(struct io_kiocb *req, |
4032 | struct io_async_msghdr *iomsg) | |
4033 | { | |
4034 | iomsg->iov = iomsg->fast_iov; | |
4035 | iomsg->msg.msg_name = &iomsg->addr; | |
4036 | return sendmsg_copy_msghdr(&iomsg->msg, req->sr_msg.umsg, | |
4037 | req->sr_msg.msg_flags, &iomsg->iov); | |
4038 | } | |
4039 | ||
3529d8c2 | 4040 | static int io_sendmsg_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
03b1230c | 4041 | { |
e47293fd | 4042 | struct io_sr_msg *sr = &req->sr_msg; |
3529d8c2 | 4043 | struct io_async_ctx *io = req->io; |
99bc4c38 | 4044 | int ret; |
03b1230c | 4045 | |
d2b6f48b PB |
4046 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4047 | return -EINVAL; | |
4048 | ||
e47293fd | 4049 | sr->msg_flags = READ_ONCE(sqe->msg_flags); |
270a5940 | 4050 | sr->umsg = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
fddaface | 4051 | sr->len = READ_ONCE(sqe->len); |
3529d8c2 | 4052 | |
d8768362 JA |
4053 | #ifdef CONFIG_COMPAT |
4054 | if (req->ctx->compat) | |
4055 | sr->msg_flags |= MSG_CMSG_COMPAT; | |
4056 | #endif | |
4057 | ||
fddaface | 4058 | if (!io || req->opcode == IORING_OP_SEND) |
3529d8c2 | 4059 | return 0; |
5f798bea PB |
4060 | /* iovec is already imported */ |
4061 | if (req->flags & REQ_F_NEED_CLEANUP) | |
4062 | return 0; | |
3529d8c2 | 4063 | |
2ae523ed | 4064 | ret = io_sendmsg_copy_hdr(req, &io->msg); |
99bc4c38 PB |
4065 | if (!ret) |
4066 | req->flags |= REQ_F_NEED_CLEANUP; | |
4067 | return ret; | |
03b1230c JA |
4068 | } |
4069 | ||
229a7b63 JA |
4070 | static int io_sendmsg(struct io_kiocb *req, bool force_nonblock, |
4071 | struct io_comp_state *cs) | |
aa1fa28f | 4072 | { |
6b754c8b | 4073 | struct io_async_msghdr iomsg, *kmsg; |
0fa03c62 | 4074 | struct socket *sock; |
7a7cacba | 4075 | unsigned flags; |
0fa03c62 JA |
4076 | int ret; |
4077 | ||
0fa03c62 | 4078 | sock = sock_from_file(req->file, &ret); |
7a7cacba PB |
4079 | if (unlikely(!sock)) |
4080 | return ret; | |
3529d8c2 | 4081 | |
7a7cacba PB |
4082 | if (req->io) { |
4083 | kmsg = &req->io->msg; | |
4084 | kmsg->msg.msg_name = &req->io->msg.addr; | |
4085 | /* if iov is set, it's allocated already */ | |
4086 | if (!kmsg->iov) | |
4087 | kmsg->iov = kmsg->fast_iov; | |
4088 | kmsg->msg.msg_iter.iov = kmsg->iov; | |
4089 | } else { | |
4090 | ret = io_sendmsg_copy_hdr(req, &iomsg); | |
4091 | if (ret) | |
4092 | return ret; | |
4093 | kmsg = &iomsg; | |
0fa03c62 | 4094 | } |
0fa03c62 | 4095 | |
7a7cacba PB |
4096 | flags = req->sr_msg.msg_flags; |
4097 | if (flags & MSG_DONTWAIT) | |
4098 | req->flags |= REQ_F_NOWAIT; | |
4099 | else if (force_nonblock) | |
4100 | flags |= MSG_DONTWAIT; | |
e47293fd | 4101 | |
7a7cacba PB |
4102 | ret = __sys_sendmsg_sock(sock, &kmsg->msg, flags); |
4103 | if (force_nonblock && ret == -EAGAIN) | |
4104 | return io_setup_async_msg(req, kmsg); | |
4105 | if (ret == -ERESTARTSYS) | |
4106 | ret = -EINTR; | |
0fa03c62 | 4107 | |
6b754c8b | 4108 | if (kmsg->iov != kmsg->fast_iov) |
0b416c3e | 4109 | kfree(kmsg->iov); |
99bc4c38 | 4110 | req->flags &= ~REQ_F_NEED_CLEANUP; |
4e88d6e7 JA |
4111 | if (ret < 0) |
4112 | req_set_fail_links(req); | |
229a7b63 | 4113 | __io_req_complete(req, ret, 0, cs); |
5d17b4a4 | 4114 | return 0; |
03b1230c | 4115 | } |
aa1fa28f | 4116 | |
229a7b63 JA |
4117 | static int io_send(struct io_kiocb *req, bool force_nonblock, |
4118 | struct io_comp_state *cs) | |
fddaface | 4119 | { |
7a7cacba PB |
4120 | struct io_sr_msg *sr = &req->sr_msg; |
4121 | struct msghdr msg; | |
4122 | struct iovec iov; | |
fddaface | 4123 | struct socket *sock; |
7a7cacba | 4124 | unsigned flags; |
fddaface JA |
4125 | int ret; |
4126 | ||
fddaface | 4127 | sock = sock_from_file(req->file, &ret); |
7a7cacba PB |
4128 | if (unlikely(!sock)) |
4129 | return ret; | |
fddaface | 4130 | |
7a7cacba PB |
4131 | ret = import_single_range(WRITE, sr->buf, sr->len, &iov, &msg.msg_iter); |
4132 | if (unlikely(ret)) | |
14c32eee | 4133 | return ret;; |
fddaface | 4134 | |
7a7cacba PB |
4135 | msg.msg_name = NULL; |
4136 | msg.msg_control = NULL; | |
4137 | msg.msg_controllen = 0; | |
4138 | msg.msg_namelen = 0; | |
fddaface | 4139 | |
7a7cacba PB |
4140 | flags = req->sr_msg.msg_flags; |
4141 | if (flags & MSG_DONTWAIT) | |
4142 | req->flags |= REQ_F_NOWAIT; | |
4143 | else if (force_nonblock) | |
4144 | flags |= MSG_DONTWAIT; | |
fddaface | 4145 | |
7a7cacba PB |
4146 | msg.msg_flags = flags; |
4147 | ret = sock_sendmsg(sock, &msg); | |
4148 | if (force_nonblock && ret == -EAGAIN) | |
4149 | return -EAGAIN; | |
4150 | if (ret == -ERESTARTSYS) | |
4151 | ret = -EINTR; | |
fddaface | 4152 | |
fddaface JA |
4153 | if (ret < 0) |
4154 | req_set_fail_links(req); | |
229a7b63 | 4155 | __io_req_complete(req, ret, 0, cs); |
fddaface | 4156 | return 0; |
fddaface JA |
4157 | } |
4158 | ||
1400e697 PB |
4159 | static int __io_recvmsg_copy_hdr(struct io_kiocb *req, |
4160 | struct io_async_msghdr *iomsg) | |
52de1fe1 JA |
4161 | { |
4162 | struct io_sr_msg *sr = &req->sr_msg; | |
4163 | struct iovec __user *uiov; | |
4164 | size_t iov_len; | |
4165 | int ret; | |
4166 | ||
1400e697 PB |
4167 | ret = __copy_msghdr_from_user(&iomsg->msg, sr->umsg, |
4168 | &iomsg->uaddr, &uiov, &iov_len); | |
52de1fe1 JA |
4169 | if (ret) |
4170 | return ret; | |
4171 | ||
4172 | if (req->flags & REQ_F_BUFFER_SELECT) { | |
4173 | if (iov_len > 1) | |
4174 | return -EINVAL; | |
1400e697 | 4175 | if (copy_from_user(iomsg->iov, uiov, sizeof(*uiov))) |
52de1fe1 | 4176 | return -EFAULT; |
1400e697 PB |
4177 | sr->len = iomsg->iov[0].iov_len; |
4178 | iov_iter_init(&iomsg->msg.msg_iter, READ, iomsg->iov, 1, | |
52de1fe1 | 4179 | sr->len); |
1400e697 | 4180 | iomsg->iov = NULL; |
52de1fe1 JA |
4181 | } else { |
4182 | ret = import_iovec(READ, uiov, iov_len, UIO_FASTIOV, | |
1400e697 | 4183 | &iomsg->iov, &iomsg->msg.msg_iter); |
52de1fe1 JA |
4184 | if (ret > 0) |
4185 | ret = 0; | |
4186 | } | |
4187 | ||
4188 | return ret; | |
4189 | } | |
4190 | ||
4191 | #ifdef CONFIG_COMPAT | |
4192 | static int __io_compat_recvmsg_copy_hdr(struct io_kiocb *req, | |
1400e697 | 4193 | struct io_async_msghdr *iomsg) |
52de1fe1 JA |
4194 | { |
4195 | struct compat_msghdr __user *msg_compat; | |
4196 | struct io_sr_msg *sr = &req->sr_msg; | |
4197 | struct compat_iovec __user *uiov; | |
4198 | compat_uptr_t ptr; | |
4199 | compat_size_t len; | |
4200 | int ret; | |
4201 | ||
270a5940 | 4202 | msg_compat = (struct compat_msghdr __user *) sr->umsg; |
1400e697 | 4203 | ret = __get_compat_msghdr(&iomsg->msg, msg_compat, &iomsg->uaddr, |
52de1fe1 JA |
4204 | &ptr, &len); |
4205 | if (ret) | |
4206 | return ret; | |
4207 | ||
4208 | uiov = compat_ptr(ptr); | |
4209 | if (req->flags & REQ_F_BUFFER_SELECT) { | |
4210 | compat_ssize_t clen; | |
4211 | ||
4212 | if (len > 1) | |
4213 | return -EINVAL; | |
4214 | if (!access_ok(uiov, sizeof(*uiov))) | |
4215 | return -EFAULT; | |
4216 | if (__get_user(clen, &uiov->iov_len)) | |
4217 | return -EFAULT; | |
4218 | if (clen < 0) | |
4219 | return -EINVAL; | |
1400e697 PB |
4220 | sr->len = iomsg->iov[0].iov_len; |
4221 | iomsg->iov = NULL; | |
52de1fe1 JA |
4222 | } else { |
4223 | ret = compat_import_iovec(READ, uiov, len, UIO_FASTIOV, | |
1400e697 PB |
4224 | &iomsg->iov, |
4225 | &iomsg->msg.msg_iter); | |
52de1fe1 JA |
4226 | if (ret < 0) |
4227 | return ret; | |
4228 | } | |
4229 | ||
4230 | return 0; | |
4231 | } | |
4232 | #endif | |
4233 | ||
1400e697 PB |
4234 | static int io_recvmsg_copy_hdr(struct io_kiocb *req, |
4235 | struct io_async_msghdr *iomsg) | |
52de1fe1 | 4236 | { |
1400e697 PB |
4237 | iomsg->msg.msg_name = &iomsg->addr; |
4238 | iomsg->iov = iomsg->fast_iov; | |
52de1fe1 JA |
4239 | |
4240 | #ifdef CONFIG_COMPAT | |
4241 | if (req->ctx->compat) | |
1400e697 | 4242 | return __io_compat_recvmsg_copy_hdr(req, iomsg); |
fddaface | 4243 | #endif |
52de1fe1 | 4244 | |
1400e697 | 4245 | return __io_recvmsg_copy_hdr(req, iomsg); |
52de1fe1 JA |
4246 | } |
4247 | ||
bcda7baa | 4248 | static struct io_buffer *io_recv_buffer_select(struct io_kiocb *req, |
7fbb1b54 | 4249 | bool needs_lock) |
bcda7baa JA |
4250 | { |
4251 | struct io_sr_msg *sr = &req->sr_msg; | |
4252 | struct io_buffer *kbuf; | |
4253 | ||
bcda7baa JA |
4254 | kbuf = io_buffer_select(req, &sr->len, sr->bgid, sr->kbuf, needs_lock); |
4255 | if (IS_ERR(kbuf)) | |
4256 | return kbuf; | |
4257 | ||
4258 | sr->kbuf = kbuf; | |
4259 | req->flags |= REQ_F_BUFFER_SELECTED; | |
bcda7baa | 4260 | return kbuf; |
fddaface JA |
4261 | } |
4262 | ||
7fbb1b54 PB |
4263 | static inline unsigned int io_put_recv_kbuf(struct io_kiocb *req) |
4264 | { | |
4265 | return io_put_kbuf(req, req->sr_msg.kbuf); | |
4266 | } | |
4267 | ||
3529d8c2 JA |
4268 | static int io_recvmsg_prep(struct io_kiocb *req, |
4269 | const struct io_uring_sqe *sqe) | |
aa1fa28f | 4270 | { |
e47293fd | 4271 | struct io_sr_msg *sr = &req->sr_msg; |
3529d8c2 | 4272 | struct io_async_ctx *io = req->io; |
99bc4c38 | 4273 | int ret; |
3529d8c2 | 4274 | |
d2b6f48b PB |
4275 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4276 | return -EINVAL; | |
4277 | ||
3529d8c2 | 4278 | sr->msg_flags = READ_ONCE(sqe->msg_flags); |
270a5940 | 4279 | sr->umsg = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
0b7b21e4 | 4280 | sr->len = READ_ONCE(sqe->len); |
bcda7baa | 4281 | sr->bgid = READ_ONCE(sqe->buf_group); |
06b76d44 | 4282 | |
d8768362 JA |
4283 | #ifdef CONFIG_COMPAT |
4284 | if (req->ctx->compat) | |
4285 | sr->msg_flags |= MSG_CMSG_COMPAT; | |
4286 | #endif | |
4287 | ||
fddaface | 4288 | if (!io || req->opcode == IORING_OP_RECV) |
06b76d44 | 4289 | return 0; |
5f798bea PB |
4290 | /* iovec is already imported */ |
4291 | if (req->flags & REQ_F_NEED_CLEANUP) | |
4292 | return 0; | |
03b1230c | 4293 | |
1400e697 | 4294 | ret = io_recvmsg_copy_hdr(req, &io->msg); |
99bc4c38 PB |
4295 | if (!ret) |
4296 | req->flags |= REQ_F_NEED_CLEANUP; | |
4297 | return ret; | |
aa1fa28f JA |
4298 | } |
4299 | ||
229a7b63 JA |
4300 | static int io_recvmsg(struct io_kiocb *req, bool force_nonblock, |
4301 | struct io_comp_state *cs) | |
aa1fa28f | 4302 | { |
6b754c8b | 4303 | struct io_async_msghdr iomsg, *kmsg; |
03b1230c | 4304 | struct socket *sock; |
7fbb1b54 | 4305 | struct io_buffer *kbuf; |
7a7cacba | 4306 | unsigned flags; |
52de1fe1 | 4307 | int ret, cflags = 0; |
03b1230c | 4308 | |
03b1230c | 4309 | sock = sock_from_file(req->file, &ret); |
7a7cacba PB |
4310 | if (unlikely(!sock)) |
4311 | return ret; | |
3529d8c2 | 4312 | |
7a7cacba PB |
4313 | if (req->io) { |
4314 | kmsg = &req->io->msg; | |
4315 | kmsg->msg.msg_name = &req->io->msg.addr; | |
4316 | /* if iov is set, it's allocated already */ | |
4317 | if (!kmsg->iov) | |
4318 | kmsg->iov = kmsg->fast_iov; | |
4319 | kmsg->msg.msg_iter.iov = kmsg->iov; | |
4320 | } else { | |
4321 | ret = io_recvmsg_copy_hdr(req, &iomsg); | |
4322 | if (ret) | |
681fda8d | 4323 | return ret; |
7a7cacba PB |
4324 | kmsg = &iomsg; |
4325 | } | |
03b1230c | 4326 | |
bc02ef33 | 4327 | if (req->flags & REQ_F_BUFFER_SELECT) { |
7fbb1b54 | 4328 | kbuf = io_recv_buffer_select(req, !force_nonblock); |
bc02ef33 | 4329 | if (IS_ERR(kbuf)) |
52de1fe1 | 4330 | return PTR_ERR(kbuf); |
7a7cacba PB |
4331 | kmsg->fast_iov[0].iov_base = u64_to_user_ptr(kbuf->addr); |
4332 | iov_iter_init(&kmsg->msg.msg_iter, READ, kmsg->iov, | |
4333 | 1, req->sr_msg.len); | |
4334 | } | |
52de1fe1 | 4335 | |
7a7cacba PB |
4336 | flags = req->sr_msg.msg_flags; |
4337 | if (flags & MSG_DONTWAIT) | |
4338 | req->flags |= REQ_F_NOWAIT; | |
4339 | else if (force_nonblock) | |
4340 | flags |= MSG_DONTWAIT; | |
e47293fd | 4341 | |
7a7cacba PB |
4342 | ret = __sys_recvmsg_sock(sock, &kmsg->msg, req->sr_msg.umsg, |
4343 | kmsg->uaddr, flags); | |
0e1b6fe3 PB |
4344 | if (force_nonblock && ret == -EAGAIN) |
4345 | return io_setup_async_msg(req, kmsg); | |
7a7cacba PB |
4346 | if (ret == -ERESTARTSYS) |
4347 | ret = -EINTR; | |
03b1230c | 4348 | |
7fbb1b54 PB |
4349 | if (req->flags & REQ_F_BUFFER_SELECTED) |
4350 | cflags = io_put_recv_kbuf(req); | |
6b754c8b | 4351 | if (kmsg->iov != kmsg->fast_iov) |
0b416c3e | 4352 | kfree(kmsg->iov); |
99bc4c38 | 4353 | req->flags &= ~REQ_F_NEED_CLEANUP; |
4e88d6e7 JA |
4354 | if (ret < 0) |
4355 | req_set_fail_links(req); | |
229a7b63 | 4356 | __io_req_complete(req, ret, cflags, cs); |
03b1230c | 4357 | return 0; |
0fa03c62 | 4358 | } |
5d17b4a4 | 4359 | |
229a7b63 JA |
4360 | static int io_recv(struct io_kiocb *req, bool force_nonblock, |
4361 | struct io_comp_state *cs) | |
fddaface | 4362 | { |
6b754c8b | 4363 | struct io_buffer *kbuf; |
7a7cacba PB |
4364 | struct io_sr_msg *sr = &req->sr_msg; |
4365 | struct msghdr msg; | |
4366 | void __user *buf = sr->buf; | |
fddaface | 4367 | struct socket *sock; |
7a7cacba PB |
4368 | struct iovec iov; |
4369 | unsigned flags; | |
bcda7baa | 4370 | int ret, cflags = 0; |
fddaface | 4371 | |
fddaface | 4372 | sock = sock_from_file(req->file, &ret); |
7a7cacba PB |
4373 | if (unlikely(!sock)) |
4374 | return ret; | |
fddaface | 4375 | |
bc02ef33 | 4376 | if (req->flags & REQ_F_BUFFER_SELECT) { |
7fbb1b54 | 4377 | kbuf = io_recv_buffer_select(req, !force_nonblock); |
bcda7baa JA |
4378 | if (IS_ERR(kbuf)) |
4379 | return PTR_ERR(kbuf); | |
7a7cacba | 4380 | buf = u64_to_user_ptr(kbuf->addr); |
bc02ef33 | 4381 | } |
bcda7baa | 4382 | |
7a7cacba | 4383 | ret = import_single_range(READ, buf, sr->len, &iov, &msg.msg_iter); |
14c32eee PB |
4384 | if (unlikely(ret)) |
4385 | goto out_free; | |
fddaface | 4386 | |
7a7cacba PB |
4387 | msg.msg_name = NULL; |
4388 | msg.msg_control = NULL; | |
4389 | msg.msg_controllen = 0; | |
4390 | msg.msg_namelen = 0; | |
4391 | msg.msg_iocb = NULL; | |
4392 | msg.msg_flags = 0; | |
fddaface | 4393 | |
7a7cacba PB |
4394 | flags = req->sr_msg.msg_flags; |
4395 | if (flags & MSG_DONTWAIT) | |
4396 | req->flags |= REQ_F_NOWAIT; | |
4397 | else if (force_nonblock) | |
4398 | flags |= MSG_DONTWAIT; | |
4399 | ||
4400 | ret = sock_recvmsg(sock, &msg, flags); | |
4401 | if (force_nonblock && ret == -EAGAIN) | |
4402 | return -EAGAIN; | |
4403 | if (ret == -ERESTARTSYS) | |
4404 | ret = -EINTR; | |
14c32eee | 4405 | out_free: |
7fbb1b54 PB |
4406 | if (req->flags & REQ_F_BUFFER_SELECTED) |
4407 | cflags = io_put_recv_kbuf(req); | |
fddaface JA |
4408 | if (ret < 0) |
4409 | req_set_fail_links(req); | |
229a7b63 | 4410 | __io_req_complete(req, ret, cflags, cs); |
fddaface | 4411 | return 0; |
fddaface JA |
4412 | } |
4413 | ||
3529d8c2 | 4414 | static int io_accept_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
17f2fe35 | 4415 | { |
8ed8d3c3 JA |
4416 | struct io_accept *accept = &req->accept; |
4417 | ||
17f2fe35 JA |
4418 | if (unlikely(req->ctx->flags & (IORING_SETUP_IOPOLL|IORING_SETUP_SQPOLL))) |
4419 | return -EINVAL; | |
8042d6ce | 4420 | if (sqe->ioprio || sqe->len || sqe->buf_index) |
17f2fe35 JA |
4421 | return -EINVAL; |
4422 | ||
d55e5f5b JA |
4423 | accept->addr = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
4424 | accept->addr_len = u64_to_user_ptr(READ_ONCE(sqe->addr2)); | |
8ed8d3c3 | 4425 | accept->flags = READ_ONCE(sqe->accept_flags); |
09952e3e | 4426 | accept->nofile = rlimit(RLIMIT_NOFILE); |
8ed8d3c3 | 4427 | return 0; |
8ed8d3c3 | 4428 | } |
17f2fe35 | 4429 | |
229a7b63 JA |
4430 | static int io_accept(struct io_kiocb *req, bool force_nonblock, |
4431 | struct io_comp_state *cs) | |
8ed8d3c3 JA |
4432 | { |
4433 | struct io_accept *accept = &req->accept; | |
ac45abc0 | 4434 | unsigned int file_flags = force_nonblock ? O_NONBLOCK : 0; |
8ed8d3c3 JA |
4435 | int ret; |
4436 | ||
e697deed JX |
4437 | if (req->file->f_flags & O_NONBLOCK) |
4438 | req->flags |= REQ_F_NOWAIT; | |
4439 | ||
8ed8d3c3 | 4440 | ret = __sys_accept4_file(req->file, file_flags, accept->addr, |
09952e3e JA |
4441 | accept->addr_len, accept->flags, |
4442 | accept->nofile); | |
8ed8d3c3 | 4443 | if (ret == -EAGAIN && force_nonblock) |
17f2fe35 | 4444 | return -EAGAIN; |
ac45abc0 PB |
4445 | if (ret < 0) { |
4446 | if (ret == -ERESTARTSYS) | |
4447 | ret = -EINTR; | |
4e88d6e7 | 4448 | req_set_fail_links(req); |
ac45abc0 | 4449 | } |
229a7b63 | 4450 | __io_req_complete(req, ret, 0, cs); |
17f2fe35 | 4451 | return 0; |
8ed8d3c3 JA |
4452 | } |
4453 | ||
3529d8c2 | 4454 | static int io_connect_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
f499a021 | 4455 | { |
3529d8c2 JA |
4456 | struct io_connect *conn = &req->connect; |
4457 | struct io_async_ctx *io = req->io; | |
f499a021 | 4458 | |
3fbb51c1 JA |
4459 | if (unlikely(req->ctx->flags & (IORING_SETUP_IOPOLL|IORING_SETUP_SQPOLL))) |
4460 | return -EINVAL; | |
4461 | if (sqe->ioprio || sqe->len || sqe->buf_index || sqe->rw_flags) | |
4462 | return -EINVAL; | |
4463 | ||
3529d8c2 JA |
4464 | conn->addr = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
4465 | conn->addr_len = READ_ONCE(sqe->addr2); | |
4466 | ||
4467 | if (!io) | |
4468 | return 0; | |
4469 | ||
4470 | return move_addr_to_kernel(conn->addr, conn->addr_len, | |
3fbb51c1 | 4471 | &io->connect.address); |
f499a021 JA |
4472 | } |
4473 | ||
229a7b63 JA |
4474 | static int io_connect(struct io_kiocb *req, bool force_nonblock, |
4475 | struct io_comp_state *cs) | |
f8e85cf2 | 4476 | { |
f499a021 | 4477 | struct io_async_ctx __io, *io; |
f8e85cf2 | 4478 | unsigned file_flags; |
3fbb51c1 | 4479 | int ret; |
f8e85cf2 | 4480 | |
f499a021 JA |
4481 | if (req->io) { |
4482 | io = req->io; | |
4483 | } else { | |
3529d8c2 JA |
4484 | ret = move_addr_to_kernel(req->connect.addr, |
4485 | req->connect.addr_len, | |
4486 | &__io.connect.address); | |
f499a021 JA |
4487 | if (ret) |
4488 | goto out; | |
4489 | io = &__io; | |
4490 | } | |
4491 | ||
3fbb51c1 JA |
4492 | file_flags = force_nonblock ? O_NONBLOCK : 0; |
4493 | ||
4494 | ret = __sys_connect_file(req->file, &io->connect.address, | |
4495 | req->connect.addr_len, file_flags); | |
87f80d62 | 4496 | if ((ret == -EAGAIN || ret == -EINPROGRESS) && force_nonblock) { |
b7bb4f7d JA |
4497 | if (req->io) |
4498 | return -EAGAIN; | |
4499 | if (io_alloc_async_ctx(req)) { | |
f499a021 JA |
4500 | ret = -ENOMEM; |
4501 | goto out; | |
4502 | } | |
b7bb4f7d | 4503 | memcpy(&req->io->connect, &__io.connect, sizeof(__io.connect)); |
f8e85cf2 | 4504 | return -EAGAIN; |
f499a021 | 4505 | } |
f8e85cf2 JA |
4506 | if (ret == -ERESTARTSYS) |
4507 | ret = -EINTR; | |
f499a021 | 4508 | out: |
4e88d6e7 JA |
4509 | if (ret < 0) |
4510 | req_set_fail_links(req); | |
229a7b63 | 4511 | __io_req_complete(req, ret, 0, cs); |
f8e85cf2 | 4512 | return 0; |
469956e8 Y |
4513 | } |
4514 | #else /* !CONFIG_NET */ | |
4515 | static int io_sendmsg_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) | |
4516 | { | |
f8e85cf2 | 4517 | return -EOPNOTSUPP; |
f8e85cf2 JA |
4518 | } |
4519 | ||
1e16c2f9 RD |
4520 | static int io_sendmsg(struct io_kiocb *req, bool force_nonblock, |
4521 | struct io_comp_state *cs) | |
469956e8 Y |
4522 | { |
4523 | return -EOPNOTSUPP; | |
4524 | } | |
4525 | ||
1e16c2f9 RD |
4526 | static int io_send(struct io_kiocb *req, bool force_nonblock, |
4527 | struct io_comp_state *cs) | |
469956e8 Y |
4528 | { |
4529 | return -EOPNOTSUPP; | |
4530 | } | |
4531 | ||
4532 | static int io_recvmsg_prep(struct io_kiocb *req, | |
4533 | const struct io_uring_sqe *sqe) | |
4534 | { | |
4535 | return -EOPNOTSUPP; | |
4536 | } | |
4537 | ||
1e16c2f9 RD |
4538 | static int io_recvmsg(struct io_kiocb *req, bool force_nonblock, |
4539 | struct io_comp_state *cs) | |
469956e8 Y |
4540 | { |
4541 | return -EOPNOTSUPP; | |
4542 | } | |
4543 | ||
1e16c2f9 RD |
4544 | static int io_recv(struct io_kiocb *req, bool force_nonblock, |
4545 | struct io_comp_state *cs) | |
469956e8 Y |
4546 | { |
4547 | return -EOPNOTSUPP; | |
4548 | } | |
4549 | ||
4550 | static int io_accept_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) | |
4551 | { | |
4552 | return -EOPNOTSUPP; | |
4553 | } | |
4554 | ||
1e16c2f9 RD |
4555 | static int io_accept(struct io_kiocb *req, bool force_nonblock, |
4556 | struct io_comp_state *cs) | |
469956e8 Y |
4557 | { |
4558 | return -EOPNOTSUPP; | |
4559 | } | |
ce593a6c | 4560 | |
469956e8 Y |
4561 | static int io_connect_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
4562 | { | |
4563 | return -EOPNOTSUPP; | |
4564 | } | |
4565 | ||
1e16c2f9 RD |
4566 | static int io_connect(struct io_kiocb *req, bool force_nonblock, |
4567 | struct io_comp_state *cs) | |
469956e8 | 4568 | { |
f8e85cf2 | 4569 | return -EOPNOTSUPP; |
ce593a6c | 4570 | } |
469956e8 | 4571 | #endif /* CONFIG_NET */ |
f8e85cf2 | 4572 | |
d7718a9d JA |
4573 | struct io_poll_table { |
4574 | struct poll_table_struct pt; | |
4575 | struct io_kiocb *req; | |
4576 | int error; | |
4577 | }; | |
ce593a6c | 4578 | |
d7718a9d JA |
4579 | static int __io_async_wake(struct io_kiocb *req, struct io_poll_iocb *poll, |
4580 | __poll_t mask, task_work_func_t func) | |
4581 | { | |
fd7d6de2 | 4582 | bool twa_signal_ok; |
aa96bf8a | 4583 | int ret; |
d7718a9d JA |
4584 | |
4585 | /* for instances that support it check for an event match first: */ | |
4586 | if (mask && !(mask & poll->events)) | |
4587 | return 0; | |
4588 | ||
4589 | trace_io_uring_task_add(req->ctx, req->opcode, req->user_data, mask); | |
4590 | ||
4591 | list_del_init(&poll->wait.entry); | |
4592 | ||
d7718a9d JA |
4593 | req->result = mask; |
4594 | init_task_work(&req->task_work, func); | |
6d816e08 JA |
4595 | percpu_ref_get(&req->ctx->refs); |
4596 | ||
fd7d6de2 JA |
4597 | /* |
4598 | * If we using the signalfd wait_queue_head for this wakeup, then | |
4599 | * it's not safe to use TWA_SIGNAL as we could be recursing on the | |
4600 | * tsk->sighand->siglock on doing the wakeup. Should not be needed | |
4601 | * either, as the normal wakeup will suffice. | |
4602 | */ | |
4603 | twa_signal_ok = (poll->head != &req->task->sighand->signalfd_wqh); | |
4604 | ||
d7718a9d | 4605 | /* |
e3aabf95 JA |
4606 | * If this fails, then the task is exiting. When a task exits, the |
4607 | * work gets canceled, so just cancel this request as well instead | |
4608 | * of executing it. We can't safely execute it anyway, as we may not | |
4609 | * have the needed state needed for it anyway. | |
d7718a9d | 4610 | */ |
fd7d6de2 | 4611 | ret = io_req_task_work_add(req, &req->task_work, twa_signal_ok); |
aa96bf8a | 4612 | if (unlikely(ret)) { |
c2c4c83c JA |
4613 | struct task_struct *tsk; |
4614 | ||
e3aabf95 | 4615 | WRITE_ONCE(poll->canceled, true); |
aa96bf8a | 4616 | tsk = io_wq_get_task(req->ctx->io_wq); |
ce593a6c JA |
4617 | task_work_add(tsk, &req->task_work, 0); |
4618 | wake_up_process(tsk); | |
aa96bf8a | 4619 | } |
d7718a9d JA |
4620 | return 1; |
4621 | } | |
4622 | ||
74ce6ce4 JA |
4623 | static bool io_poll_rewait(struct io_kiocb *req, struct io_poll_iocb *poll) |
4624 | __acquires(&req->ctx->completion_lock) | |
4625 | { | |
4626 | struct io_ring_ctx *ctx = req->ctx; | |
4627 | ||
4628 | if (!req->result && !READ_ONCE(poll->canceled)) { | |
4629 | struct poll_table_struct pt = { ._key = poll->events }; | |
4630 | ||
4631 | req->result = vfs_poll(req->file, &pt) & poll->events; | |
4632 | } | |
4633 | ||
4634 | spin_lock_irq(&ctx->completion_lock); | |
4635 | if (!req->result && !READ_ONCE(poll->canceled)) { | |
4636 | add_wait_queue(poll->head, &poll->wait); | |
4637 | return true; | |
4638 | } | |
4639 | ||
4640 | return false; | |
4641 | } | |
4642 | ||
d4e7cd36 | 4643 | static struct io_poll_iocb *io_poll_get_double(struct io_kiocb *req) |
18bceab1 | 4644 | { |
d4e7cd36 JA |
4645 | /* pure poll stashes this in ->io, poll driven retry elsewhere */ |
4646 | if (req->opcode == IORING_OP_POLL_ADD) | |
4647 | return (struct io_poll_iocb *) req->io; | |
4648 | return req->apoll->double_poll; | |
4649 | } | |
4650 | ||
4651 | static struct io_poll_iocb *io_poll_get_single(struct io_kiocb *req) | |
4652 | { | |
4653 | if (req->opcode == IORING_OP_POLL_ADD) | |
4654 | return &req->poll; | |
4655 | return &req->apoll->poll; | |
4656 | } | |
4657 | ||
4658 | static void io_poll_remove_double(struct io_kiocb *req) | |
4659 | { | |
4660 | struct io_poll_iocb *poll = io_poll_get_double(req); | |
18bceab1 JA |
4661 | |
4662 | lockdep_assert_held(&req->ctx->completion_lock); | |
4663 | ||
4664 | if (poll && poll->head) { | |
4665 | struct wait_queue_head *head = poll->head; | |
4666 | ||
4667 | spin_lock(&head->lock); | |
4668 | list_del_init(&poll->wait.entry); | |
4669 | if (poll->wait.private) | |
4670 | refcount_dec(&req->refs); | |
4671 | poll->head = NULL; | |
4672 | spin_unlock(&head->lock); | |
4673 | } | |
4674 | } | |
4675 | ||
4676 | static void io_poll_complete(struct io_kiocb *req, __poll_t mask, int error) | |
4677 | { | |
4678 | struct io_ring_ctx *ctx = req->ctx; | |
4679 | ||
d4e7cd36 | 4680 | io_poll_remove_double(req); |
18bceab1 JA |
4681 | req->poll.done = true; |
4682 | io_cqring_fill_event(req, error ? error : mangle_poll(mask)); | |
4683 | io_commit_cqring(ctx); | |
4684 | } | |
4685 | ||
4686 | static void io_poll_task_handler(struct io_kiocb *req, struct io_kiocb **nxt) | |
4687 | { | |
4688 | struct io_ring_ctx *ctx = req->ctx; | |
4689 | ||
4690 | if (io_poll_rewait(req, &req->poll)) { | |
4691 | spin_unlock_irq(&ctx->completion_lock); | |
4692 | return; | |
4693 | } | |
4694 | ||
4695 | hash_del(&req->hash_node); | |
4696 | io_poll_complete(req, req->result, 0); | |
4697 | req->flags |= REQ_F_COMP_LOCKED; | |
9b5f7bd9 | 4698 | *nxt = io_put_req_find_next(req); |
18bceab1 JA |
4699 | spin_unlock_irq(&ctx->completion_lock); |
4700 | ||
4701 | io_cqring_ev_posted(ctx); | |
4702 | } | |
4703 | ||
4704 | static void io_poll_task_func(struct callback_head *cb) | |
4705 | { | |
4706 | struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work); | |
6d816e08 | 4707 | struct io_ring_ctx *ctx = req->ctx; |
18bceab1 JA |
4708 | struct io_kiocb *nxt = NULL; |
4709 | ||
4710 | io_poll_task_handler(req, &nxt); | |
ea1164e5 PB |
4711 | if (nxt) |
4712 | __io_req_task_submit(nxt); | |
6d816e08 | 4713 | percpu_ref_put(&ctx->refs); |
18bceab1 JA |
4714 | } |
4715 | ||
4716 | static int io_poll_double_wake(struct wait_queue_entry *wait, unsigned mode, | |
4717 | int sync, void *key) | |
4718 | { | |
4719 | struct io_kiocb *req = wait->private; | |
d4e7cd36 | 4720 | struct io_poll_iocb *poll = io_poll_get_single(req); |
18bceab1 JA |
4721 | __poll_t mask = key_to_poll(key); |
4722 | ||
4723 | /* for instances that support it check for an event match first: */ | |
4724 | if (mask && !(mask & poll->events)) | |
4725 | return 0; | |
4726 | ||
807abcb0 | 4727 | if (poll && poll->head) { |
18bceab1 JA |
4728 | bool done; |
4729 | ||
807abcb0 JA |
4730 | spin_lock(&poll->head->lock); |
4731 | done = list_empty(&poll->wait.entry); | |
18bceab1 | 4732 | if (!done) |
807abcb0 | 4733 | list_del_init(&poll->wait.entry); |
d4e7cd36 JA |
4734 | /* make sure double remove sees this as being gone */ |
4735 | wait->private = NULL; | |
807abcb0 | 4736 | spin_unlock(&poll->head->lock); |
18bceab1 JA |
4737 | if (!done) |
4738 | __io_async_wake(req, poll, mask, io_poll_task_func); | |
4739 | } | |
4740 | refcount_dec(&req->refs); | |
4741 | return 1; | |
4742 | } | |
4743 | ||
4744 | static void io_init_poll_iocb(struct io_poll_iocb *poll, __poll_t events, | |
4745 | wait_queue_func_t wake_func) | |
4746 | { | |
4747 | poll->head = NULL; | |
4748 | poll->done = false; | |
4749 | poll->canceled = false; | |
4750 | poll->events = events; | |
4751 | INIT_LIST_HEAD(&poll->wait.entry); | |
4752 | init_waitqueue_func_entry(&poll->wait, wake_func); | |
4753 | } | |
4754 | ||
4755 | static void __io_queue_proc(struct io_poll_iocb *poll, struct io_poll_table *pt, | |
807abcb0 JA |
4756 | struct wait_queue_head *head, |
4757 | struct io_poll_iocb **poll_ptr) | |
18bceab1 JA |
4758 | { |
4759 | struct io_kiocb *req = pt->req; | |
4760 | ||
4761 | /* | |
4762 | * If poll->head is already set, it's because the file being polled | |
4763 | * uses multiple waitqueues for poll handling (eg one for read, one | |
4764 | * for write). Setup a separate io_poll_iocb if this happens. | |
4765 | */ | |
4766 | if (unlikely(poll->head)) { | |
4767 | /* already have a 2nd entry, fail a third attempt */ | |
807abcb0 | 4768 | if (*poll_ptr) { |
18bceab1 JA |
4769 | pt->error = -EINVAL; |
4770 | return; | |
4771 | } | |
4772 | poll = kmalloc(sizeof(*poll), GFP_ATOMIC); | |
4773 | if (!poll) { | |
4774 | pt->error = -ENOMEM; | |
4775 | return; | |
4776 | } | |
4777 | io_init_poll_iocb(poll, req->poll.events, io_poll_double_wake); | |
4778 | refcount_inc(&req->refs); | |
4779 | poll->wait.private = req; | |
807abcb0 | 4780 | *poll_ptr = poll; |
18bceab1 JA |
4781 | } |
4782 | ||
4783 | pt->error = 0; | |
4784 | poll->head = head; | |
a31eb4a2 JX |
4785 | |
4786 | if (poll->events & EPOLLEXCLUSIVE) | |
4787 | add_wait_queue_exclusive(head, &poll->wait); | |
4788 | else | |
4789 | add_wait_queue(head, &poll->wait); | |
18bceab1 JA |
4790 | } |
4791 | ||
4792 | static void io_async_queue_proc(struct file *file, struct wait_queue_head *head, | |
4793 | struct poll_table_struct *p) | |
4794 | { | |
4795 | struct io_poll_table *pt = container_of(p, struct io_poll_table, pt); | |
807abcb0 | 4796 | struct async_poll *apoll = pt->req->apoll; |
18bceab1 | 4797 | |
807abcb0 | 4798 | __io_queue_proc(&apoll->poll, pt, head, &apoll->double_poll); |
18bceab1 JA |
4799 | } |
4800 | ||
d7718a9d JA |
4801 | static void io_async_task_func(struct callback_head *cb) |
4802 | { | |
4803 | struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work); | |
4804 | struct async_poll *apoll = req->apoll; | |
4805 | struct io_ring_ctx *ctx = req->ctx; | |
4806 | ||
4807 | trace_io_uring_task_run(req->ctx, req->opcode, req->user_data); | |
4808 | ||
74ce6ce4 | 4809 | if (io_poll_rewait(req, &apoll->poll)) { |
d7718a9d | 4810 | spin_unlock_irq(&ctx->completion_lock); |
6d816e08 | 4811 | percpu_ref_put(&ctx->refs); |
74ce6ce4 | 4812 | return; |
d7718a9d JA |
4813 | } |
4814 | ||
31067255 | 4815 | /* If req is still hashed, it cannot have been canceled. Don't check. */ |
0be0b0e3 | 4816 | if (hash_hashed(&req->hash_node)) |
74ce6ce4 | 4817 | hash_del(&req->hash_node); |
2bae047e | 4818 | |
d4e7cd36 | 4819 | io_poll_remove_double(req); |
74ce6ce4 JA |
4820 | spin_unlock_irq(&ctx->completion_lock); |
4821 | ||
0be0b0e3 PB |
4822 | if (!READ_ONCE(apoll->poll.canceled)) |
4823 | __io_req_task_submit(req); | |
4824 | else | |
4825 | __io_req_task_cancel(req, -ECANCELED); | |
aa340845 | 4826 | |
6d816e08 | 4827 | percpu_ref_put(&ctx->refs); |
807abcb0 | 4828 | kfree(apoll->double_poll); |
31067255 | 4829 | kfree(apoll); |
d7718a9d JA |
4830 | } |
4831 | ||
4832 | static int io_async_wake(struct wait_queue_entry *wait, unsigned mode, int sync, | |
4833 | void *key) | |
4834 | { | |
4835 | struct io_kiocb *req = wait->private; | |
4836 | struct io_poll_iocb *poll = &req->apoll->poll; | |
4837 | ||
4838 | trace_io_uring_poll_wake(req->ctx, req->opcode, req->user_data, | |
4839 | key_to_poll(key)); | |
4840 | ||
4841 | return __io_async_wake(req, poll, key_to_poll(key), io_async_task_func); | |
4842 | } | |
4843 | ||
4844 | static void io_poll_req_insert(struct io_kiocb *req) | |
4845 | { | |
4846 | struct io_ring_ctx *ctx = req->ctx; | |
4847 | struct hlist_head *list; | |
4848 | ||
4849 | list = &ctx->cancel_hash[hash_long(req->user_data, ctx->cancel_hash_bits)]; | |
4850 | hlist_add_head(&req->hash_node, list); | |
4851 | } | |
4852 | ||
4853 | static __poll_t __io_arm_poll_handler(struct io_kiocb *req, | |
4854 | struct io_poll_iocb *poll, | |
4855 | struct io_poll_table *ipt, __poll_t mask, | |
4856 | wait_queue_func_t wake_func) | |
4857 | __acquires(&ctx->completion_lock) | |
4858 | { | |
4859 | struct io_ring_ctx *ctx = req->ctx; | |
4860 | bool cancel = false; | |
4861 | ||
18bceab1 | 4862 | io_init_poll_iocb(poll, mask, wake_func); |
b90cd197 | 4863 | poll->file = req->file; |
18bceab1 | 4864 | poll->wait.private = req; |
d7718a9d JA |
4865 | |
4866 | ipt->pt._key = mask; | |
4867 | ipt->req = req; | |
4868 | ipt->error = -EINVAL; | |
4869 | ||
d7718a9d JA |
4870 | mask = vfs_poll(req->file, &ipt->pt) & poll->events; |
4871 | ||
4872 | spin_lock_irq(&ctx->completion_lock); | |
4873 | if (likely(poll->head)) { | |
4874 | spin_lock(&poll->head->lock); | |
4875 | if (unlikely(list_empty(&poll->wait.entry))) { | |
4876 | if (ipt->error) | |
4877 | cancel = true; | |
4878 | ipt->error = 0; | |
4879 | mask = 0; | |
4880 | } | |
4881 | if (mask || ipt->error) | |
4882 | list_del_init(&poll->wait.entry); | |
4883 | else if (cancel) | |
4884 | WRITE_ONCE(poll->canceled, true); | |
4885 | else if (!poll->done) /* actually waiting for an event */ | |
4886 | io_poll_req_insert(req); | |
4887 | spin_unlock(&poll->head->lock); | |
4888 | } | |
4889 | ||
4890 | return mask; | |
4891 | } | |
4892 | ||
4893 | static bool io_arm_poll_handler(struct io_kiocb *req) | |
4894 | { | |
4895 | const struct io_op_def *def = &io_op_defs[req->opcode]; | |
4896 | struct io_ring_ctx *ctx = req->ctx; | |
4897 | struct async_poll *apoll; | |
4898 | struct io_poll_table ipt; | |
4899 | __poll_t mask, ret; | |
9dab14b8 | 4900 | int rw; |
d7718a9d JA |
4901 | |
4902 | if (!req->file || !file_can_poll(req->file)) | |
4903 | return false; | |
24c74678 | 4904 | if (req->flags & REQ_F_POLLED) |
d7718a9d | 4905 | return false; |
9dab14b8 JA |
4906 | if (def->pollin) |
4907 | rw = READ; | |
4908 | else if (def->pollout) | |
4909 | rw = WRITE; | |
4910 | else | |
4911 | return false; | |
4912 | /* if we can't nonblock try, then no point in arming a poll handler */ | |
4913 | if (!io_file_supports_async(req->file, rw)) | |
d7718a9d JA |
4914 | return false; |
4915 | ||
4916 | apoll = kmalloc(sizeof(*apoll), GFP_ATOMIC); | |
4917 | if (unlikely(!apoll)) | |
4918 | return false; | |
807abcb0 | 4919 | apoll->double_poll = NULL; |
d7718a9d JA |
4920 | |
4921 | req->flags |= REQ_F_POLLED; | |
4dd2824d | 4922 | io_get_req_task(req); |
d7718a9d JA |
4923 | req->apoll = apoll; |
4924 | INIT_HLIST_NODE(&req->hash_node); | |
4925 | ||
8755d97a | 4926 | mask = 0; |
d7718a9d | 4927 | if (def->pollin) |
8755d97a | 4928 | mask |= POLLIN | POLLRDNORM; |
d7718a9d JA |
4929 | if (def->pollout) |
4930 | mask |= POLLOUT | POLLWRNORM; | |
4931 | mask |= POLLERR | POLLPRI; | |
4932 | ||
4933 | ipt.pt._qproc = io_async_queue_proc; | |
4934 | ||
4935 | ret = __io_arm_poll_handler(req, &apoll->poll, &ipt, mask, | |
4936 | io_async_wake); | |
a36da65c | 4937 | if (ret || ipt.error) { |
d4e7cd36 | 4938 | io_poll_remove_double(req); |
d7718a9d | 4939 | spin_unlock_irq(&ctx->completion_lock); |
807abcb0 | 4940 | kfree(apoll->double_poll); |
d7718a9d JA |
4941 | kfree(apoll); |
4942 | return false; | |
4943 | } | |
4944 | spin_unlock_irq(&ctx->completion_lock); | |
4945 | trace_io_uring_poll_arm(ctx, req->opcode, req->user_data, mask, | |
4946 | apoll->poll.events); | |
4947 | return true; | |
4948 | } | |
4949 | ||
4950 | static bool __io_poll_remove_one(struct io_kiocb *req, | |
4951 | struct io_poll_iocb *poll) | |
221c5eb2 | 4952 | { |
b41e9852 | 4953 | bool do_complete = false; |
221c5eb2 JA |
4954 | |
4955 | spin_lock(&poll->head->lock); | |
4956 | WRITE_ONCE(poll->canceled, true); | |
392edb45 JA |
4957 | if (!list_empty(&poll->wait.entry)) { |
4958 | list_del_init(&poll->wait.entry); | |
b41e9852 | 4959 | do_complete = true; |
221c5eb2 JA |
4960 | } |
4961 | spin_unlock(&poll->head->lock); | |
3bfa5bcb | 4962 | hash_del(&req->hash_node); |
d7718a9d JA |
4963 | return do_complete; |
4964 | } | |
4965 | ||
4966 | static bool io_poll_remove_one(struct io_kiocb *req) | |
4967 | { | |
4968 | bool do_complete; | |
4969 | ||
d4e7cd36 JA |
4970 | io_poll_remove_double(req); |
4971 | ||
d7718a9d JA |
4972 | if (req->opcode == IORING_OP_POLL_ADD) { |
4973 | do_complete = __io_poll_remove_one(req, &req->poll); | |
4974 | } else { | |
3bfa5bcb JA |
4975 | struct async_poll *apoll = req->apoll; |
4976 | ||
d7718a9d | 4977 | /* non-poll requests have submit ref still */ |
3bfa5bcb JA |
4978 | do_complete = __io_poll_remove_one(req, &apoll->poll); |
4979 | if (do_complete) { | |
d7718a9d | 4980 | io_put_req(req); |
807abcb0 | 4981 | kfree(apoll->double_poll); |
3bfa5bcb JA |
4982 | kfree(apoll); |
4983 | } | |
b1f573bd XW |
4984 | } |
4985 | ||
b41e9852 JA |
4986 | if (do_complete) { |
4987 | io_cqring_fill_event(req, -ECANCELED); | |
4988 | io_commit_cqring(req->ctx); | |
4989 | req->flags |= REQ_F_COMP_LOCKED; | |
f254ac04 | 4990 | req_set_fail_links(req); |
b41e9852 JA |
4991 | io_put_req(req); |
4992 | } | |
4993 | ||
4994 | return do_complete; | |
221c5eb2 JA |
4995 | } |
4996 | ||
4997 | static void io_poll_remove_all(struct io_ring_ctx *ctx) | |
4998 | { | |
78076bb6 | 4999 | struct hlist_node *tmp; |
221c5eb2 | 5000 | struct io_kiocb *req; |
8e2e1faf | 5001 | int posted = 0, i; |
221c5eb2 JA |
5002 | |
5003 | spin_lock_irq(&ctx->completion_lock); | |
78076bb6 JA |
5004 | for (i = 0; i < (1U << ctx->cancel_hash_bits); i++) { |
5005 | struct hlist_head *list; | |
5006 | ||
5007 | list = &ctx->cancel_hash[i]; | |
5008 | hlist_for_each_entry_safe(req, tmp, list, hash_node) | |
8e2e1faf | 5009 | posted += io_poll_remove_one(req); |
221c5eb2 JA |
5010 | } |
5011 | spin_unlock_irq(&ctx->completion_lock); | |
b41e9852 | 5012 | |
8e2e1faf JA |
5013 | if (posted) |
5014 | io_cqring_ev_posted(ctx); | |
221c5eb2 JA |
5015 | } |
5016 | ||
47f46768 JA |
5017 | static int io_poll_cancel(struct io_ring_ctx *ctx, __u64 sqe_addr) |
5018 | { | |
78076bb6 | 5019 | struct hlist_head *list; |
47f46768 JA |
5020 | struct io_kiocb *req; |
5021 | ||
78076bb6 JA |
5022 | list = &ctx->cancel_hash[hash_long(sqe_addr, ctx->cancel_hash_bits)]; |
5023 | hlist_for_each_entry(req, list, hash_node) { | |
b41e9852 JA |
5024 | if (sqe_addr != req->user_data) |
5025 | continue; | |
5026 | if (io_poll_remove_one(req)) | |
eac406c6 | 5027 | return 0; |
b41e9852 | 5028 | return -EALREADY; |
47f46768 JA |
5029 | } |
5030 | ||
5031 | return -ENOENT; | |
5032 | } | |
5033 | ||
3529d8c2 JA |
5034 | static int io_poll_remove_prep(struct io_kiocb *req, |
5035 | const struct io_uring_sqe *sqe) | |
0969e783 | 5036 | { |
0969e783 JA |
5037 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
5038 | return -EINVAL; | |
5039 | if (sqe->ioprio || sqe->off || sqe->len || sqe->buf_index || | |
5040 | sqe->poll_events) | |
5041 | return -EINVAL; | |
5042 | ||
5043 | req->poll.addr = READ_ONCE(sqe->addr); | |
0969e783 JA |
5044 | return 0; |
5045 | } | |
5046 | ||
221c5eb2 JA |
5047 | /* |
5048 | * Find a running poll command that matches one specified in sqe->addr, | |
5049 | * and remove it if found. | |
5050 | */ | |
fc4df999 | 5051 | static int io_poll_remove(struct io_kiocb *req) |
221c5eb2 JA |
5052 | { |
5053 | struct io_ring_ctx *ctx = req->ctx; | |
0969e783 | 5054 | u64 addr; |
47f46768 | 5055 | int ret; |
221c5eb2 | 5056 | |
0969e783 | 5057 | addr = req->poll.addr; |
221c5eb2 | 5058 | spin_lock_irq(&ctx->completion_lock); |
0969e783 | 5059 | ret = io_poll_cancel(ctx, addr); |
221c5eb2 JA |
5060 | spin_unlock_irq(&ctx->completion_lock); |
5061 | ||
4e88d6e7 JA |
5062 | if (ret < 0) |
5063 | req_set_fail_links(req); | |
e1e16097 | 5064 | io_req_complete(req, ret); |
221c5eb2 JA |
5065 | return 0; |
5066 | } | |
5067 | ||
221c5eb2 JA |
5068 | static int io_poll_wake(struct wait_queue_entry *wait, unsigned mode, int sync, |
5069 | void *key) | |
5070 | { | |
c2f2eb7d JA |
5071 | struct io_kiocb *req = wait->private; |
5072 | struct io_poll_iocb *poll = &req->poll; | |
221c5eb2 | 5073 | |
d7718a9d | 5074 | return __io_async_wake(req, poll, key_to_poll(key), io_poll_task_func); |
221c5eb2 JA |
5075 | } |
5076 | ||
221c5eb2 JA |
5077 | static void io_poll_queue_proc(struct file *file, struct wait_queue_head *head, |
5078 | struct poll_table_struct *p) | |
5079 | { | |
5080 | struct io_poll_table *pt = container_of(p, struct io_poll_table, pt); | |
5081 | ||
807abcb0 | 5082 | __io_queue_proc(&pt->req->poll, pt, head, (struct io_poll_iocb **) &pt->req->io); |
eac406c6 JA |
5083 | } |
5084 | ||
3529d8c2 | 5085 | static int io_poll_add_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
221c5eb2 JA |
5086 | { |
5087 | struct io_poll_iocb *poll = &req->poll; | |
5769a351 | 5088 | u32 events; |
221c5eb2 JA |
5089 | |
5090 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) | |
5091 | return -EINVAL; | |
5092 | if (sqe->addr || sqe->ioprio || sqe->off || sqe->len || sqe->buf_index) | |
5093 | return -EINVAL; | |
09bb8394 JA |
5094 | if (!poll->file) |
5095 | return -EBADF; | |
221c5eb2 | 5096 | |
5769a351 JX |
5097 | events = READ_ONCE(sqe->poll32_events); |
5098 | #ifdef __BIG_ENDIAN | |
5099 | events = swahw32(events); | |
5100 | #endif | |
a31eb4a2 JX |
5101 | poll->events = demangle_poll(events) | EPOLLERR | EPOLLHUP | |
5102 | (events & EPOLLEXCLUSIVE); | |
b41e9852 | 5103 | |
4dd2824d | 5104 | io_get_req_task(req); |
0969e783 JA |
5105 | return 0; |
5106 | } | |
5107 | ||
014db007 | 5108 | static int io_poll_add(struct io_kiocb *req) |
0969e783 JA |
5109 | { |
5110 | struct io_poll_iocb *poll = &req->poll; | |
5111 | struct io_ring_ctx *ctx = req->ctx; | |
5112 | struct io_poll_table ipt; | |
0969e783 | 5113 | __poll_t mask; |
0969e783 | 5114 | |
78076bb6 | 5115 | INIT_HLIST_NODE(&req->hash_node); |
d7718a9d | 5116 | ipt.pt._qproc = io_poll_queue_proc; |
36703247 | 5117 | |
d7718a9d JA |
5118 | mask = __io_arm_poll_handler(req, &req->poll, &ipt, poll->events, |
5119 | io_poll_wake); | |
221c5eb2 | 5120 | |
8c838788 | 5121 | if (mask) { /* no async, we'd stolen it */ |
221c5eb2 | 5122 | ipt.error = 0; |
b0dd8a41 | 5123 | io_poll_complete(req, mask, 0); |
221c5eb2 | 5124 | } |
221c5eb2 JA |
5125 | spin_unlock_irq(&ctx->completion_lock); |
5126 | ||
8c838788 JA |
5127 | if (mask) { |
5128 | io_cqring_ev_posted(ctx); | |
014db007 | 5129 | io_put_req(req); |
221c5eb2 | 5130 | } |
8c838788 | 5131 | return ipt.error; |
221c5eb2 JA |
5132 | } |
5133 | ||
5262f567 JA |
5134 | static enum hrtimer_restart io_timeout_fn(struct hrtimer *timer) |
5135 | { | |
ad8a48ac JA |
5136 | struct io_timeout_data *data = container_of(timer, |
5137 | struct io_timeout_data, timer); | |
5138 | struct io_kiocb *req = data->req; | |
5139 | struct io_ring_ctx *ctx = req->ctx; | |
5262f567 JA |
5140 | unsigned long flags; |
5141 | ||
5262f567 | 5142 | spin_lock_irqsave(&ctx->completion_lock, flags); |
01cec8c1 PB |
5143 | atomic_set(&req->ctx->cq_timeouts, |
5144 | atomic_read(&req->ctx->cq_timeouts) + 1); | |
5145 | ||
ef03681a | 5146 | /* |
11365043 JA |
5147 | * We could be racing with timeout deletion. If the list is empty, |
5148 | * then timeout lookup already found it and will be handling it. | |
ef03681a | 5149 | */ |
135fcde8 PB |
5150 | if (!list_empty(&req->timeout.list)) |
5151 | list_del_init(&req->timeout.list); | |
5262f567 | 5152 | |
78e19bbe | 5153 | io_cqring_fill_event(req, -ETIME); |
5262f567 JA |
5154 | io_commit_cqring(ctx); |
5155 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
5156 | ||
5157 | io_cqring_ev_posted(ctx); | |
4e88d6e7 | 5158 | req_set_fail_links(req); |
5262f567 JA |
5159 | io_put_req(req); |
5160 | return HRTIMER_NORESTART; | |
5161 | } | |
5162 | ||
f254ac04 JA |
5163 | static int __io_timeout_cancel(struct io_kiocb *req) |
5164 | { | |
5165 | int ret; | |
5166 | ||
5167 | list_del_init(&req->timeout.list); | |
5168 | ||
5169 | ret = hrtimer_try_to_cancel(&req->io->timeout.timer); | |
5170 | if (ret == -1) | |
5171 | return -EALREADY; | |
5172 | ||
5173 | req_set_fail_links(req); | |
5174 | req->flags |= REQ_F_COMP_LOCKED; | |
5175 | io_cqring_fill_event(req, -ECANCELED); | |
5176 | io_put_req(req); | |
5177 | return 0; | |
5178 | } | |
5179 | ||
47f46768 JA |
5180 | static int io_timeout_cancel(struct io_ring_ctx *ctx, __u64 user_data) |
5181 | { | |
5182 | struct io_kiocb *req; | |
5183 | int ret = -ENOENT; | |
5184 | ||
135fcde8 | 5185 | list_for_each_entry(req, &ctx->timeout_list, timeout.list) { |
47f46768 | 5186 | if (user_data == req->user_data) { |
47f46768 JA |
5187 | ret = 0; |
5188 | break; | |
5189 | } | |
5190 | } | |
5191 | ||
5192 | if (ret == -ENOENT) | |
5193 | return ret; | |
5194 | ||
f254ac04 | 5195 | return __io_timeout_cancel(req); |
47f46768 JA |
5196 | } |
5197 | ||
3529d8c2 JA |
5198 | static int io_timeout_remove_prep(struct io_kiocb *req, |
5199 | const struct io_uring_sqe *sqe) | |
b29472ee | 5200 | { |
b29472ee JA |
5201 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
5202 | return -EINVAL; | |
61710e43 DA |
5203 | if (unlikely(req->flags & (REQ_F_FIXED_FILE | REQ_F_BUFFER_SELECT))) |
5204 | return -EINVAL; | |
5205 | if (sqe->ioprio || sqe->buf_index || sqe->len) | |
b29472ee JA |
5206 | return -EINVAL; |
5207 | ||
5208 | req->timeout.addr = READ_ONCE(sqe->addr); | |
5209 | req->timeout.flags = READ_ONCE(sqe->timeout_flags); | |
5210 | if (req->timeout.flags) | |
5211 | return -EINVAL; | |
5212 | ||
b29472ee JA |
5213 | return 0; |
5214 | } | |
5215 | ||
11365043 JA |
5216 | /* |
5217 | * Remove or update an existing timeout command | |
5218 | */ | |
fc4df999 | 5219 | static int io_timeout_remove(struct io_kiocb *req) |
11365043 JA |
5220 | { |
5221 | struct io_ring_ctx *ctx = req->ctx; | |
47f46768 | 5222 | int ret; |
11365043 | 5223 | |
11365043 | 5224 | spin_lock_irq(&ctx->completion_lock); |
b29472ee | 5225 | ret = io_timeout_cancel(ctx, req->timeout.addr); |
11365043 | 5226 | |
47f46768 | 5227 | io_cqring_fill_event(req, ret); |
11365043 JA |
5228 | io_commit_cqring(ctx); |
5229 | spin_unlock_irq(&ctx->completion_lock); | |
5262f567 | 5230 | io_cqring_ev_posted(ctx); |
4e88d6e7 JA |
5231 | if (ret < 0) |
5232 | req_set_fail_links(req); | |
ec9c02ad | 5233 | io_put_req(req); |
11365043 | 5234 | return 0; |
5262f567 JA |
5235 | } |
5236 | ||
3529d8c2 | 5237 | static int io_timeout_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe, |
2d28390a | 5238 | bool is_timeout_link) |
5262f567 | 5239 | { |
ad8a48ac | 5240 | struct io_timeout_data *data; |
a41525ab | 5241 | unsigned flags; |
56080b02 | 5242 | u32 off = READ_ONCE(sqe->off); |
5262f567 | 5243 | |
ad8a48ac | 5244 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
5262f567 | 5245 | return -EINVAL; |
ad8a48ac | 5246 | if (sqe->ioprio || sqe->buf_index || sqe->len != 1) |
a41525ab | 5247 | return -EINVAL; |
56080b02 | 5248 | if (off && is_timeout_link) |
2d28390a | 5249 | return -EINVAL; |
a41525ab JA |
5250 | flags = READ_ONCE(sqe->timeout_flags); |
5251 | if (flags & ~IORING_TIMEOUT_ABS) | |
5262f567 | 5252 | return -EINVAL; |
bdf20073 | 5253 | |
bfe68a22 | 5254 | req->timeout.off = off; |
26a61679 | 5255 | |
3529d8c2 | 5256 | if (!req->io && io_alloc_async_ctx(req)) |
26a61679 JA |
5257 | return -ENOMEM; |
5258 | ||
5259 | data = &req->io->timeout; | |
ad8a48ac | 5260 | data->req = req; |
ad8a48ac JA |
5261 | |
5262 | if (get_timespec64(&data->ts, u64_to_user_ptr(sqe->addr))) | |
5262f567 JA |
5263 | return -EFAULT; |
5264 | ||
11365043 | 5265 | if (flags & IORING_TIMEOUT_ABS) |
ad8a48ac | 5266 | data->mode = HRTIMER_MODE_ABS; |
11365043 | 5267 | else |
ad8a48ac | 5268 | data->mode = HRTIMER_MODE_REL; |
11365043 | 5269 | |
ad8a48ac JA |
5270 | hrtimer_init(&data->timer, CLOCK_MONOTONIC, data->mode); |
5271 | return 0; | |
5272 | } | |
5273 | ||
fc4df999 | 5274 | static int io_timeout(struct io_kiocb *req) |
ad8a48ac | 5275 | { |
ad8a48ac | 5276 | struct io_ring_ctx *ctx = req->ctx; |
bfe68a22 | 5277 | struct io_timeout_data *data = &req->io->timeout; |
ad8a48ac | 5278 | struct list_head *entry; |
bfe68a22 | 5279 | u32 tail, off = req->timeout.off; |
ad8a48ac | 5280 | |
733f5c95 | 5281 | spin_lock_irq(&ctx->completion_lock); |
93bd25bb | 5282 | |
5262f567 JA |
5283 | /* |
5284 | * sqe->off holds how many events that need to occur for this | |
93bd25bb JA |
5285 | * timeout event to be satisfied. If it isn't set, then this is |
5286 | * a pure timeout request, sequence isn't used. | |
5262f567 | 5287 | */ |
8eb7e2d0 | 5288 | if (io_is_timeout_noseq(req)) { |
93bd25bb JA |
5289 | entry = ctx->timeout_list.prev; |
5290 | goto add; | |
5291 | } | |
5262f567 | 5292 | |
bfe68a22 PB |
5293 | tail = ctx->cached_cq_tail - atomic_read(&ctx->cq_timeouts); |
5294 | req->timeout.target_seq = tail + off; | |
5262f567 JA |
5295 | |
5296 | /* | |
5297 | * Insertion sort, ensuring the first entry in the list is always | |
5298 | * the one we need first. | |
5299 | */ | |
5262f567 | 5300 | list_for_each_prev(entry, &ctx->timeout_list) { |
135fcde8 PB |
5301 | struct io_kiocb *nxt = list_entry(entry, struct io_kiocb, |
5302 | timeout.list); | |
5262f567 | 5303 | |
8eb7e2d0 | 5304 | if (io_is_timeout_noseq(nxt)) |
93bd25bb | 5305 | continue; |
bfe68a22 PB |
5306 | /* nxt.seq is behind @tail, otherwise would've been completed */ |
5307 | if (off >= nxt->timeout.target_seq - tail) | |
5262f567 JA |
5308 | break; |
5309 | } | |
93bd25bb | 5310 | add: |
135fcde8 | 5311 | list_add(&req->timeout.list, entry); |
ad8a48ac JA |
5312 | data->timer.function = io_timeout_fn; |
5313 | hrtimer_start(&data->timer, timespec64_to_ktime(data->ts), data->mode); | |
5262f567 | 5314 | spin_unlock_irq(&ctx->completion_lock); |
5262f567 JA |
5315 | return 0; |
5316 | } | |
5262f567 | 5317 | |
62755e35 JA |
5318 | static bool io_cancel_cb(struct io_wq_work *work, void *data) |
5319 | { | |
5320 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); | |
5321 | ||
5322 | return req->user_data == (unsigned long) data; | |
5323 | } | |
5324 | ||
e977d6d3 | 5325 | static int io_async_cancel_one(struct io_ring_ctx *ctx, void *sqe_addr) |
62755e35 | 5326 | { |
62755e35 | 5327 | enum io_wq_cancel cancel_ret; |
62755e35 JA |
5328 | int ret = 0; |
5329 | ||
4f26bda1 | 5330 | cancel_ret = io_wq_cancel_cb(ctx->io_wq, io_cancel_cb, sqe_addr, false); |
62755e35 JA |
5331 | switch (cancel_ret) { |
5332 | case IO_WQ_CANCEL_OK: | |
5333 | ret = 0; | |
5334 | break; | |
5335 | case IO_WQ_CANCEL_RUNNING: | |
5336 | ret = -EALREADY; | |
5337 | break; | |
5338 | case IO_WQ_CANCEL_NOTFOUND: | |
5339 | ret = -ENOENT; | |
5340 | break; | |
5341 | } | |
5342 | ||
e977d6d3 JA |
5343 | return ret; |
5344 | } | |
5345 | ||
47f46768 JA |
5346 | static void io_async_find_and_cancel(struct io_ring_ctx *ctx, |
5347 | struct io_kiocb *req, __u64 sqe_addr, | |
014db007 | 5348 | int success_ret) |
47f46768 JA |
5349 | { |
5350 | unsigned long flags; | |
5351 | int ret; | |
5352 | ||
5353 | ret = io_async_cancel_one(ctx, (void *) (unsigned long) sqe_addr); | |
5354 | if (ret != -ENOENT) { | |
5355 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
5356 | goto done; | |
5357 | } | |
5358 | ||
5359 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
5360 | ret = io_timeout_cancel(ctx, sqe_addr); | |
5361 | if (ret != -ENOENT) | |
5362 | goto done; | |
5363 | ret = io_poll_cancel(ctx, sqe_addr); | |
5364 | done: | |
b0dd8a41 JA |
5365 | if (!ret) |
5366 | ret = success_ret; | |
47f46768 JA |
5367 | io_cqring_fill_event(req, ret); |
5368 | io_commit_cqring(ctx); | |
5369 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
5370 | io_cqring_ev_posted(ctx); | |
5371 | ||
4e88d6e7 JA |
5372 | if (ret < 0) |
5373 | req_set_fail_links(req); | |
014db007 | 5374 | io_put_req(req); |
47f46768 JA |
5375 | } |
5376 | ||
3529d8c2 JA |
5377 | static int io_async_cancel_prep(struct io_kiocb *req, |
5378 | const struct io_uring_sqe *sqe) | |
e977d6d3 | 5379 | { |
fbf23849 | 5380 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
e977d6d3 | 5381 | return -EINVAL; |
61710e43 DA |
5382 | if (unlikely(req->flags & (REQ_F_FIXED_FILE | REQ_F_BUFFER_SELECT))) |
5383 | return -EINVAL; | |
5384 | if (sqe->ioprio || sqe->off || sqe->len || sqe->cancel_flags) | |
e977d6d3 JA |
5385 | return -EINVAL; |
5386 | ||
fbf23849 JA |
5387 | req->cancel.addr = READ_ONCE(sqe->addr); |
5388 | return 0; | |
5389 | } | |
5390 | ||
014db007 | 5391 | static int io_async_cancel(struct io_kiocb *req) |
fbf23849 JA |
5392 | { |
5393 | struct io_ring_ctx *ctx = req->ctx; | |
fbf23849 | 5394 | |
014db007 | 5395 | io_async_find_and_cancel(ctx, req, req->cancel.addr, 0); |
5262f567 JA |
5396 | return 0; |
5397 | } | |
5398 | ||
05f3fb3c JA |
5399 | static int io_files_update_prep(struct io_kiocb *req, |
5400 | const struct io_uring_sqe *sqe) | |
5401 | { | |
61710e43 DA |
5402 | if (unlikely(req->flags & (REQ_F_FIXED_FILE | REQ_F_BUFFER_SELECT))) |
5403 | return -EINVAL; | |
5404 | if (sqe->ioprio || sqe->rw_flags) | |
05f3fb3c JA |
5405 | return -EINVAL; |
5406 | ||
5407 | req->files_update.offset = READ_ONCE(sqe->off); | |
5408 | req->files_update.nr_args = READ_ONCE(sqe->len); | |
5409 | if (!req->files_update.nr_args) | |
5410 | return -EINVAL; | |
5411 | req->files_update.arg = READ_ONCE(sqe->addr); | |
5412 | return 0; | |
5413 | } | |
5414 | ||
229a7b63 JA |
5415 | static int io_files_update(struct io_kiocb *req, bool force_nonblock, |
5416 | struct io_comp_state *cs) | |
fbf23849 JA |
5417 | { |
5418 | struct io_ring_ctx *ctx = req->ctx; | |
05f3fb3c JA |
5419 | struct io_uring_files_update up; |
5420 | int ret; | |
fbf23849 | 5421 | |
f86cd20c | 5422 | if (force_nonblock) |
05f3fb3c | 5423 | return -EAGAIN; |
05f3fb3c JA |
5424 | |
5425 | up.offset = req->files_update.offset; | |
5426 | up.fds = req->files_update.arg; | |
5427 | ||
5428 | mutex_lock(&ctx->uring_lock); | |
5429 | ret = __io_sqe_files_update(ctx, &up, req->files_update.nr_args); | |
5430 | mutex_unlock(&ctx->uring_lock); | |
5431 | ||
5432 | if (ret < 0) | |
5433 | req_set_fail_links(req); | |
229a7b63 | 5434 | __io_req_complete(req, ret, 0, cs); |
5262f567 JA |
5435 | return 0; |
5436 | } | |
5437 | ||
3529d8c2 JA |
5438 | static int io_req_defer_prep(struct io_kiocb *req, |
5439 | const struct io_uring_sqe *sqe) | |
f67676d1 | 5440 | { |
e781573e | 5441 | ssize_t ret = 0; |
f67676d1 | 5442 | |
f1d96a8f PB |
5443 | if (!sqe) |
5444 | return 0; | |
5445 | ||
327d6d96 PB |
5446 | if (io_alloc_async_ctx(req)) |
5447 | return -EAGAIN; | |
f56040b8 PB |
5448 | ret = io_prep_work_files(req); |
5449 | if (unlikely(ret)) | |
5450 | return ret; | |
cccf0ee8 | 5451 | |
d625c6ee | 5452 | switch (req->opcode) { |
e781573e JA |
5453 | case IORING_OP_NOP: |
5454 | break; | |
f67676d1 JA |
5455 | case IORING_OP_READV: |
5456 | case IORING_OP_READ_FIXED: | |
3a6820f2 | 5457 | case IORING_OP_READ: |
3529d8c2 | 5458 | ret = io_read_prep(req, sqe, true); |
f67676d1 JA |
5459 | break; |
5460 | case IORING_OP_WRITEV: | |
5461 | case IORING_OP_WRITE_FIXED: | |
3a6820f2 | 5462 | case IORING_OP_WRITE: |
3529d8c2 | 5463 | ret = io_write_prep(req, sqe, true); |
f67676d1 | 5464 | break; |
0969e783 | 5465 | case IORING_OP_POLL_ADD: |
3529d8c2 | 5466 | ret = io_poll_add_prep(req, sqe); |
0969e783 JA |
5467 | break; |
5468 | case IORING_OP_POLL_REMOVE: | |
3529d8c2 | 5469 | ret = io_poll_remove_prep(req, sqe); |
0969e783 | 5470 | break; |
8ed8d3c3 | 5471 | case IORING_OP_FSYNC: |
3529d8c2 | 5472 | ret = io_prep_fsync(req, sqe); |
8ed8d3c3 JA |
5473 | break; |
5474 | case IORING_OP_SYNC_FILE_RANGE: | |
3529d8c2 | 5475 | ret = io_prep_sfr(req, sqe); |
8ed8d3c3 | 5476 | break; |
03b1230c | 5477 | case IORING_OP_SENDMSG: |
fddaface | 5478 | case IORING_OP_SEND: |
3529d8c2 | 5479 | ret = io_sendmsg_prep(req, sqe); |
03b1230c JA |
5480 | break; |
5481 | case IORING_OP_RECVMSG: | |
fddaface | 5482 | case IORING_OP_RECV: |
3529d8c2 | 5483 | ret = io_recvmsg_prep(req, sqe); |
03b1230c | 5484 | break; |
f499a021 | 5485 | case IORING_OP_CONNECT: |
3529d8c2 | 5486 | ret = io_connect_prep(req, sqe); |
f499a021 | 5487 | break; |
2d28390a | 5488 | case IORING_OP_TIMEOUT: |
3529d8c2 | 5489 | ret = io_timeout_prep(req, sqe, false); |
b7bb4f7d | 5490 | break; |
b29472ee | 5491 | case IORING_OP_TIMEOUT_REMOVE: |
3529d8c2 | 5492 | ret = io_timeout_remove_prep(req, sqe); |
b29472ee | 5493 | break; |
fbf23849 | 5494 | case IORING_OP_ASYNC_CANCEL: |
3529d8c2 | 5495 | ret = io_async_cancel_prep(req, sqe); |
fbf23849 | 5496 | break; |
2d28390a | 5497 | case IORING_OP_LINK_TIMEOUT: |
3529d8c2 | 5498 | ret = io_timeout_prep(req, sqe, true); |
b7bb4f7d | 5499 | break; |
8ed8d3c3 | 5500 | case IORING_OP_ACCEPT: |
3529d8c2 | 5501 | ret = io_accept_prep(req, sqe); |
8ed8d3c3 | 5502 | break; |
d63d1b5e JA |
5503 | case IORING_OP_FALLOCATE: |
5504 | ret = io_fallocate_prep(req, sqe); | |
5505 | break; | |
15b71abe JA |
5506 | case IORING_OP_OPENAT: |
5507 | ret = io_openat_prep(req, sqe); | |
5508 | break; | |
b5dba59e JA |
5509 | case IORING_OP_CLOSE: |
5510 | ret = io_close_prep(req, sqe); | |
5511 | break; | |
05f3fb3c JA |
5512 | case IORING_OP_FILES_UPDATE: |
5513 | ret = io_files_update_prep(req, sqe); | |
5514 | break; | |
eddc7ef5 JA |
5515 | case IORING_OP_STATX: |
5516 | ret = io_statx_prep(req, sqe); | |
5517 | break; | |
4840e418 JA |
5518 | case IORING_OP_FADVISE: |
5519 | ret = io_fadvise_prep(req, sqe); | |
5520 | break; | |
c1ca757b JA |
5521 | case IORING_OP_MADVISE: |
5522 | ret = io_madvise_prep(req, sqe); | |
5523 | break; | |
cebdb986 JA |
5524 | case IORING_OP_OPENAT2: |
5525 | ret = io_openat2_prep(req, sqe); | |
5526 | break; | |
3e4827b0 JA |
5527 | case IORING_OP_EPOLL_CTL: |
5528 | ret = io_epoll_ctl_prep(req, sqe); | |
5529 | break; | |
7d67af2c PB |
5530 | case IORING_OP_SPLICE: |
5531 | ret = io_splice_prep(req, sqe); | |
5532 | break; | |
ddf0322d JA |
5533 | case IORING_OP_PROVIDE_BUFFERS: |
5534 | ret = io_provide_buffers_prep(req, sqe); | |
5535 | break; | |
067524e9 JA |
5536 | case IORING_OP_REMOVE_BUFFERS: |
5537 | ret = io_remove_buffers_prep(req, sqe); | |
5538 | break; | |
f2a8d5c7 PB |
5539 | case IORING_OP_TEE: |
5540 | ret = io_tee_prep(req, sqe); | |
5541 | break; | |
f67676d1 | 5542 | default: |
e781573e JA |
5543 | printk_once(KERN_WARNING "io_uring: unhandled opcode %d\n", |
5544 | req->opcode); | |
5545 | ret = -EINVAL; | |
b7bb4f7d | 5546 | break; |
f67676d1 JA |
5547 | } |
5548 | ||
b7bb4f7d | 5549 | return ret; |
f67676d1 JA |
5550 | } |
5551 | ||
9cf7c104 PB |
5552 | static u32 io_get_sequence(struct io_kiocb *req) |
5553 | { | |
5554 | struct io_kiocb *pos; | |
5555 | struct io_ring_ctx *ctx = req->ctx; | |
5556 | u32 total_submitted, nr_reqs = 1; | |
5557 | ||
5558 | if (req->flags & REQ_F_LINK_HEAD) | |
5559 | list_for_each_entry(pos, &req->link_list, link_list) | |
5560 | nr_reqs++; | |
5561 | ||
5562 | total_submitted = ctx->cached_sq_head - ctx->cached_sq_dropped; | |
5563 | return total_submitted - nr_reqs; | |
5564 | } | |
5565 | ||
3529d8c2 | 5566 | static int io_req_defer(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
de0617e4 | 5567 | { |
a197f664 | 5568 | struct io_ring_ctx *ctx = req->ctx; |
27dc8338 | 5569 | struct io_defer_entry *de; |
f67676d1 | 5570 | int ret; |
9cf7c104 | 5571 | u32 seq; |
de0617e4 | 5572 | |
9d858b21 | 5573 | /* Still need defer if there is pending req in defer list. */ |
9cf7c104 PB |
5574 | if (likely(list_empty_careful(&ctx->defer_list) && |
5575 | !(req->flags & REQ_F_IO_DRAIN))) | |
5576 | return 0; | |
5577 | ||
5578 | seq = io_get_sequence(req); | |
5579 | /* Still a chance to pass the sequence check */ | |
5580 | if (!req_need_defer(req, seq) && list_empty_careful(&ctx->defer_list)) | |
de0617e4 JA |
5581 | return 0; |
5582 | ||
650b5481 | 5583 | if (!req->io) { |
650b5481 | 5584 | ret = io_req_defer_prep(req, sqe); |
327d6d96 | 5585 | if (ret) |
650b5481 PB |
5586 | return ret; |
5587 | } | |
cbdcb435 | 5588 | io_prep_async_link(req); |
27dc8338 PB |
5589 | de = kmalloc(sizeof(*de), GFP_KERNEL); |
5590 | if (!de) | |
5591 | return -ENOMEM; | |
2d28390a | 5592 | |
de0617e4 | 5593 | spin_lock_irq(&ctx->completion_lock); |
9cf7c104 | 5594 | if (!req_need_defer(req, seq) && list_empty(&ctx->defer_list)) { |
de0617e4 | 5595 | spin_unlock_irq(&ctx->completion_lock); |
27dc8338 | 5596 | kfree(de); |
ae34817b PB |
5597 | io_queue_async_work(req); |
5598 | return -EIOCBQUEUED; | |
de0617e4 JA |
5599 | } |
5600 | ||
915967f6 | 5601 | trace_io_uring_defer(ctx, req, req->user_data); |
27dc8338 | 5602 | de->req = req; |
9cf7c104 | 5603 | de->seq = seq; |
27dc8338 | 5604 | list_add_tail(&de->list, &ctx->defer_list); |
de0617e4 JA |
5605 | spin_unlock_irq(&ctx->completion_lock); |
5606 | return -EIOCBQUEUED; | |
5607 | } | |
5608 | ||
3ca405eb | 5609 | static void __io_clean_op(struct io_kiocb *req) |
99bc4c38 PB |
5610 | { |
5611 | struct io_async_ctx *io = req->io; | |
5612 | ||
0e1b6fe3 PB |
5613 | if (req->flags & REQ_F_BUFFER_SELECTED) { |
5614 | switch (req->opcode) { | |
5615 | case IORING_OP_READV: | |
5616 | case IORING_OP_READ_FIXED: | |
5617 | case IORING_OP_READ: | |
bcda7baa | 5618 | kfree((void *)(unsigned long)req->rw.addr); |
0e1b6fe3 PB |
5619 | break; |
5620 | case IORING_OP_RECVMSG: | |
5621 | case IORING_OP_RECV: | |
bcda7baa | 5622 | kfree(req->sr_msg.kbuf); |
0e1b6fe3 PB |
5623 | break; |
5624 | } | |
5625 | req->flags &= ~REQ_F_BUFFER_SELECTED; | |
99bc4c38 PB |
5626 | } |
5627 | ||
0e1b6fe3 PB |
5628 | if (req->flags & REQ_F_NEED_CLEANUP) { |
5629 | switch (req->opcode) { | |
5630 | case IORING_OP_READV: | |
5631 | case IORING_OP_READ_FIXED: | |
5632 | case IORING_OP_READ: | |
5633 | case IORING_OP_WRITEV: | |
5634 | case IORING_OP_WRITE_FIXED: | |
5635 | case IORING_OP_WRITE: | |
ff6165b2 JA |
5636 | if (io->rw.free_iovec) |
5637 | kfree(io->rw.free_iovec); | |
0e1b6fe3 PB |
5638 | break; |
5639 | case IORING_OP_RECVMSG: | |
5640 | case IORING_OP_SENDMSG: | |
5641 | if (io->msg.iov != io->msg.fast_iov) | |
5642 | kfree(io->msg.iov); | |
5643 | break; | |
5644 | case IORING_OP_SPLICE: | |
5645 | case IORING_OP_TEE: | |
5646 | io_put_file(req, req->splice.file_in, | |
5647 | (req->splice.flags & SPLICE_F_FD_IN_FIXED)); | |
5648 | break; | |
5649 | } | |
5650 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
99bc4c38 | 5651 | } |
bb175342 PB |
5652 | |
5653 | if (req->flags & REQ_F_INFLIGHT) { | |
5654 | struct io_ring_ctx *ctx = req->ctx; | |
5655 | unsigned long flags; | |
5656 | ||
5657 | spin_lock_irqsave(&ctx->inflight_lock, flags); | |
5658 | list_del(&req->inflight_entry); | |
5659 | if (waitqueue_active(&ctx->inflight_wait)) | |
5660 | wake_up(&ctx->inflight_wait); | |
5661 | spin_unlock_irqrestore(&ctx->inflight_lock, flags); | |
5662 | req->flags &= ~REQ_F_INFLIGHT; | |
5663 | } | |
99bc4c38 PB |
5664 | } |
5665 | ||
3529d8c2 | 5666 | static int io_issue_sqe(struct io_kiocb *req, const struct io_uring_sqe *sqe, |
f13fad7b | 5667 | bool force_nonblock, struct io_comp_state *cs) |
2b188cc1 | 5668 | { |
a197f664 | 5669 | struct io_ring_ctx *ctx = req->ctx; |
d625c6ee | 5670 | int ret; |
2b188cc1 | 5671 | |
d625c6ee | 5672 | switch (req->opcode) { |
2b188cc1 | 5673 | case IORING_OP_NOP: |
229a7b63 | 5674 | ret = io_nop(req, cs); |
2b188cc1 JA |
5675 | break; |
5676 | case IORING_OP_READV: | |
edafccee | 5677 | case IORING_OP_READ_FIXED: |
3a6820f2 | 5678 | case IORING_OP_READ: |
3529d8c2 JA |
5679 | if (sqe) { |
5680 | ret = io_read_prep(req, sqe, force_nonblock); | |
5681 | if (ret < 0) | |
5682 | break; | |
5683 | } | |
a1d7c393 | 5684 | ret = io_read(req, force_nonblock, cs); |
edafccee | 5685 | break; |
3529d8c2 | 5686 | case IORING_OP_WRITEV: |
edafccee | 5687 | case IORING_OP_WRITE_FIXED: |
3a6820f2 | 5688 | case IORING_OP_WRITE: |
3529d8c2 JA |
5689 | if (sqe) { |
5690 | ret = io_write_prep(req, sqe, force_nonblock); | |
5691 | if (ret < 0) | |
5692 | break; | |
5693 | } | |
a1d7c393 | 5694 | ret = io_write(req, force_nonblock, cs); |
2b188cc1 | 5695 | break; |
c992fe29 | 5696 | case IORING_OP_FSYNC: |
3529d8c2 JA |
5697 | if (sqe) { |
5698 | ret = io_prep_fsync(req, sqe); | |
5699 | if (ret < 0) | |
5700 | break; | |
5701 | } | |
014db007 | 5702 | ret = io_fsync(req, force_nonblock); |
c992fe29 | 5703 | break; |
221c5eb2 | 5704 | case IORING_OP_POLL_ADD: |
3529d8c2 JA |
5705 | if (sqe) { |
5706 | ret = io_poll_add_prep(req, sqe); | |
5707 | if (ret) | |
5708 | break; | |
5709 | } | |
014db007 | 5710 | ret = io_poll_add(req); |
221c5eb2 JA |
5711 | break; |
5712 | case IORING_OP_POLL_REMOVE: | |
3529d8c2 JA |
5713 | if (sqe) { |
5714 | ret = io_poll_remove_prep(req, sqe); | |
5715 | if (ret < 0) | |
5716 | break; | |
5717 | } | |
fc4df999 | 5718 | ret = io_poll_remove(req); |
221c5eb2 | 5719 | break; |
5d17b4a4 | 5720 | case IORING_OP_SYNC_FILE_RANGE: |
3529d8c2 JA |
5721 | if (sqe) { |
5722 | ret = io_prep_sfr(req, sqe); | |
5723 | if (ret < 0) | |
5724 | break; | |
5725 | } | |
014db007 | 5726 | ret = io_sync_file_range(req, force_nonblock); |
5d17b4a4 | 5727 | break; |
0fa03c62 | 5728 | case IORING_OP_SENDMSG: |
fddaface | 5729 | case IORING_OP_SEND: |
3529d8c2 JA |
5730 | if (sqe) { |
5731 | ret = io_sendmsg_prep(req, sqe); | |
5732 | if (ret < 0) | |
5733 | break; | |
5734 | } | |
fddaface | 5735 | if (req->opcode == IORING_OP_SENDMSG) |
229a7b63 | 5736 | ret = io_sendmsg(req, force_nonblock, cs); |
fddaface | 5737 | else |
229a7b63 | 5738 | ret = io_send(req, force_nonblock, cs); |
0fa03c62 | 5739 | break; |
aa1fa28f | 5740 | case IORING_OP_RECVMSG: |
fddaface | 5741 | case IORING_OP_RECV: |
3529d8c2 JA |
5742 | if (sqe) { |
5743 | ret = io_recvmsg_prep(req, sqe); | |
5744 | if (ret) | |
5745 | break; | |
5746 | } | |
fddaface | 5747 | if (req->opcode == IORING_OP_RECVMSG) |
229a7b63 | 5748 | ret = io_recvmsg(req, force_nonblock, cs); |
fddaface | 5749 | else |
229a7b63 | 5750 | ret = io_recv(req, force_nonblock, cs); |
aa1fa28f | 5751 | break; |
5262f567 | 5752 | case IORING_OP_TIMEOUT: |
3529d8c2 JA |
5753 | if (sqe) { |
5754 | ret = io_timeout_prep(req, sqe, false); | |
5755 | if (ret) | |
5756 | break; | |
5757 | } | |
fc4df999 | 5758 | ret = io_timeout(req); |
5262f567 | 5759 | break; |
11365043 | 5760 | case IORING_OP_TIMEOUT_REMOVE: |
3529d8c2 JA |
5761 | if (sqe) { |
5762 | ret = io_timeout_remove_prep(req, sqe); | |
5763 | if (ret) | |
5764 | break; | |
5765 | } | |
fc4df999 | 5766 | ret = io_timeout_remove(req); |
11365043 | 5767 | break; |
17f2fe35 | 5768 | case IORING_OP_ACCEPT: |
3529d8c2 JA |
5769 | if (sqe) { |
5770 | ret = io_accept_prep(req, sqe); | |
5771 | if (ret) | |
5772 | break; | |
5773 | } | |
229a7b63 | 5774 | ret = io_accept(req, force_nonblock, cs); |
17f2fe35 | 5775 | break; |
f8e85cf2 | 5776 | case IORING_OP_CONNECT: |
3529d8c2 JA |
5777 | if (sqe) { |
5778 | ret = io_connect_prep(req, sqe); | |
5779 | if (ret) | |
5780 | break; | |
5781 | } | |
229a7b63 | 5782 | ret = io_connect(req, force_nonblock, cs); |
f8e85cf2 | 5783 | break; |
62755e35 | 5784 | case IORING_OP_ASYNC_CANCEL: |
3529d8c2 JA |
5785 | if (sqe) { |
5786 | ret = io_async_cancel_prep(req, sqe); | |
5787 | if (ret) | |
5788 | break; | |
5789 | } | |
014db007 | 5790 | ret = io_async_cancel(req); |
62755e35 | 5791 | break; |
d63d1b5e JA |
5792 | case IORING_OP_FALLOCATE: |
5793 | if (sqe) { | |
5794 | ret = io_fallocate_prep(req, sqe); | |
5795 | if (ret) | |
5796 | break; | |
5797 | } | |
014db007 | 5798 | ret = io_fallocate(req, force_nonblock); |
d63d1b5e | 5799 | break; |
15b71abe JA |
5800 | case IORING_OP_OPENAT: |
5801 | if (sqe) { | |
5802 | ret = io_openat_prep(req, sqe); | |
5803 | if (ret) | |
5804 | break; | |
5805 | } | |
014db007 | 5806 | ret = io_openat(req, force_nonblock); |
15b71abe | 5807 | break; |
b5dba59e JA |
5808 | case IORING_OP_CLOSE: |
5809 | if (sqe) { | |
5810 | ret = io_close_prep(req, sqe); | |
5811 | if (ret) | |
5812 | break; | |
5813 | } | |
229a7b63 | 5814 | ret = io_close(req, force_nonblock, cs); |
b5dba59e | 5815 | break; |
05f3fb3c JA |
5816 | case IORING_OP_FILES_UPDATE: |
5817 | if (sqe) { | |
5818 | ret = io_files_update_prep(req, sqe); | |
5819 | if (ret) | |
5820 | break; | |
5821 | } | |
229a7b63 | 5822 | ret = io_files_update(req, force_nonblock, cs); |
05f3fb3c | 5823 | break; |
eddc7ef5 JA |
5824 | case IORING_OP_STATX: |
5825 | if (sqe) { | |
5826 | ret = io_statx_prep(req, sqe); | |
5827 | if (ret) | |
5828 | break; | |
5829 | } | |
014db007 | 5830 | ret = io_statx(req, force_nonblock); |
eddc7ef5 | 5831 | break; |
4840e418 JA |
5832 | case IORING_OP_FADVISE: |
5833 | if (sqe) { | |
5834 | ret = io_fadvise_prep(req, sqe); | |
5835 | if (ret) | |
5836 | break; | |
5837 | } | |
014db007 | 5838 | ret = io_fadvise(req, force_nonblock); |
4840e418 | 5839 | break; |
c1ca757b JA |
5840 | case IORING_OP_MADVISE: |
5841 | if (sqe) { | |
5842 | ret = io_madvise_prep(req, sqe); | |
5843 | if (ret) | |
5844 | break; | |
5845 | } | |
014db007 | 5846 | ret = io_madvise(req, force_nonblock); |
c1ca757b | 5847 | break; |
cebdb986 JA |
5848 | case IORING_OP_OPENAT2: |
5849 | if (sqe) { | |
5850 | ret = io_openat2_prep(req, sqe); | |
5851 | if (ret) | |
5852 | break; | |
5853 | } | |
014db007 | 5854 | ret = io_openat2(req, force_nonblock); |
cebdb986 | 5855 | break; |
3e4827b0 JA |
5856 | case IORING_OP_EPOLL_CTL: |
5857 | if (sqe) { | |
5858 | ret = io_epoll_ctl_prep(req, sqe); | |
5859 | if (ret) | |
5860 | break; | |
5861 | } | |
229a7b63 | 5862 | ret = io_epoll_ctl(req, force_nonblock, cs); |
3e4827b0 | 5863 | break; |
7d67af2c PB |
5864 | case IORING_OP_SPLICE: |
5865 | if (sqe) { | |
5866 | ret = io_splice_prep(req, sqe); | |
5867 | if (ret < 0) | |
5868 | break; | |
5869 | } | |
014db007 | 5870 | ret = io_splice(req, force_nonblock); |
7d67af2c | 5871 | break; |
ddf0322d JA |
5872 | case IORING_OP_PROVIDE_BUFFERS: |
5873 | if (sqe) { | |
5874 | ret = io_provide_buffers_prep(req, sqe); | |
5875 | if (ret) | |
5876 | break; | |
5877 | } | |
229a7b63 | 5878 | ret = io_provide_buffers(req, force_nonblock, cs); |
ddf0322d | 5879 | break; |
067524e9 JA |
5880 | case IORING_OP_REMOVE_BUFFERS: |
5881 | if (sqe) { | |
5882 | ret = io_remove_buffers_prep(req, sqe); | |
5883 | if (ret) | |
5884 | break; | |
5885 | } | |
229a7b63 | 5886 | ret = io_remove_buffers(req, force_nonblock, cs); |
3e4827b0 | 5887 | break; |
f2a8d5c7 PB |
5888 | case IORING_OP_TEE: |
5889 | if (sqe) { | |
5890 | ret = io_tee_prep(req, sqe); | |
5891 | if (ret < 0) | |
5892 | break; | |
5893 | } | |
5894 | ret = io_tee(req, force_nonblock); | |
5895 | break; | |
2b188cc1 JA |
5896 | default: |
5897 | ret = -EINVAL; | |
5898 | break; | |
5899 | } | |
5900 | ||
def596e9 JA |
5901 | if (ret) |
5902 | return ret; | |
5903 | ||
b532576e JA |
5904 | /* If the op doesn't have a file, we're not polling for it */ |
5905 | if ((ctx->flags & IORING_SETUP_IOPOLL) && req->file) { | |
11ba820b JA |
5906 | const bool in_async = io_wq_current_is_worker(); |
5907 | ||
11ba820b JA |
5908 | /* workqueue context doesn't hold uring_lock, grab it now */ |
5909 | if (in_async) | |
5910 | mutex_lock(&ctx->uring_lock); | |
5911 | ||
def596e9 | 5912 | io_iopoll_req_issued(req); |
11ba820b JA |
5913 | |
5914 | if (in_async) | |
5915 | mutex_unlock(&ctx->uring_lock); | |
def596e9 JA |
5916 | } |
5917 | ||
5918 | return 0; | |
2b188cc1 JA |
5919 | } |
5920 | ||
f4db7182 | 5921 | static struct io_wq_work *io_wq_submit_work(struct io_wq_work *work) |
2b188cc1 JA |
5922 | { |
5923 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); | |
6df1db6b | 5924 | struct io_kiocb *timeout; |
561fb04a | 5925 | int ret = 0; |
2b188cc1 | 5926 | |
6df1db6b PB |
5927 | timeout = io_prep_linked_timeout(req); |
5928 | if (timeout) | |
5929 | io_queue_linked_timeout(timeout); | |
d4c81f38 | 5930 | |
0c9d5ccd JA |
5931 | /* if NO_CANCEL is set, we must still run the work */ |
5932 | if ((work->flags & (IO_WQ_WORK_CANCEL|IO_WQ_WORK_NO_CANCEL)) == | |
5933 | IO_WQ_WORK_CANCEL) { | |
561fb04a | 5934 | ret = -ECANCELED; |
0c9d5ccd | 5935 | } |
31b51510 | 5936 | |
561fb04a | 5937 | if (!ret) { |
561fb04a | 5938 | do { |
f13fad7b | 5939 | ret = io_issue_sqe(req, NULL, false, NULL); |
561fb04a JA |
5940 | /* |
5941 | * We can get EAGAIN for polled IO even though we're | |
5942 | * forcing a sync submission from here, since we can't | |
5943 | * wait for request slots on the block side. | |
5944 | */ | |
5945 | if (ret != -EAGAIN) | |
5946 | break; | |
5947 | cond_resched(); | |
5948 | } while (1); | |
5949 | } | |
31b51510 | 5950 | |
561fb04a | 5951 | if (ret) { |
4e88d6e7 | 5952 | req_set_fail_links(req); |
e1e16097 | 5953 | io_req_complete(req, ret); |
edafccee | 5954 | } |
2b188cc1 | 5955 | |
f4db7182 | 5956 | return io_steal_work(req); |
2b188cc1 JA |
5957 | } |
5958 | ||
65e19f54 JA |
5959 | static inline struct file *io_file_from_index(struct io_ring_ctx *ctx, |
5960 | int index) | |
5961 | { | |
5962 | struct fixed_file_table *table; | |
5963 | ||
05f3fb3c | 5964 | table = &ctx->file_data->table[index >> IORING_FILE_TABLE_SHIFT]; |
84695089 | 5965 | return table->files[index & IORING_FILE_TABLE_MASK]; |
65e19f54 JA |
5966 | } |
5967 | ||
8da11c19 PB |
5968 | static int io_file_get(struct io_submit_state *state, struct io_kiocb *req, |
5969 | int fd, struct file **out_file, bool fixed) | |
09bb8394 | 5970 | { |
a197f664 | 5971 | struct io_ring_ctx *ctx = req->ctx; |
8da11c19 | 5972 | struct file *file; |
09bb8394 | 5973 | |
8da11c19 | 5974 | if (fixed) { |
05f3fb3c | 5975 | if (unlikely(!ctx->file_data || |
09bb8394 JA |
5976 | (unsigned) fd >= ctx->nr_user_files)) |
5977 | return -EBADF; | |
b7620121 | 5978 | fd = array_index_nospec(fd, ctx->nr_user_files); |
8da11c19 | 5979 | file = io_file_from_index(ctx, fd); |
fd2206e4 JA |
5980 | if (file) { |
5981 | req->fixed_file_refs = ctx->file_data->cur_refs; | |
5982 | percpu_ref_get(req->fixed_file_refs); | |
5983 | } | |
09bb8394 | 5984 | } else { |
c826bd7a | 5985 | trace_io_uring_file_get(ctx, fd); |
8da11c19 | 5986 | file = __io_file_get(state, fd); |
09bb8394 JA |
5987 | } |
5988 | ||
fd2206e4 JA |
5989 | if (file || io_op_defs[req->opcode].needs_file_no_error) { |
5990 | *out_file = file; | |
5991 | return 0; | |
5992 | } | |
5993 | return -EBADF; | |
09bb8394 JA |
5994 | } |
5995 | ||
8da11c19 | 5996 | static int io_req_set_file(struct io_submit_state *state, struct io_kiocb *req, |
63ff8223 | 5997 | int fd) |
8da11c19 | 5998 | { |
8da11c19 PB |
5999 | bool fixed; |
6000 | ||
63ff8223 | 6001 | fixed = (req->flags & REQ_F_FIXED_FILE) != 0; |
0cdaf760 | 6002 | if (unlikely(!fixed && io_async_submit(req->ctx))) |
8da11c19 PB |
6003 | return -EBADF; |
6004 | ||
6005 | return io_file_get(state, req, fd, &req->file, fixed); | |
6006 | } | |
6007 | ||
a197f664 | 6008 | static int io_grab_files(struct io_kiocb *req) |
fcb323cc JA |
6009 | { |
6010 | int ret = -EBADF; | |
a197f664 | 6011 | struct io_ring_ctx *ctx = req->ctx; |
fcb323cc | 6012 | |
f56040b8 PB |
6013 | io_req_init_async(req); |
6014 | ||
5b0bbee4 | 6015 | if (req->work.files || (req->flags & REQ_F_NO_FILE_TABLE)) |
f86cd20c | 6016 | return 0; |
b14cca0c | 6017 | if (!ctx->ring_file) |
b5dba59e JA |
6018 | return -EBADF; |
6019 | ||
fcb323cc JA |
6020 | rcu_read_lock(); |
6021 | spin_lock_irq(&ctx->inflight_lock); | |
6022 | /* | |
6023 | * We use the f_ops->flush() handler to ensure that we can flush | |
6024 | * out work accessing these files if the fd is closed. Check if | |
6025 | * the fd has changed since we started down this path, and disallow | |
6026 | * this operation if it has. | |
6027 | */ | |
b14cca0c | 6028 | if (fcheck(ctx->ring_fd) == ctx->ring_file) { |
fcb323cc JA |
6029 | list_add(&req->inflight_entry, &ctx->inflight_list); |
6030 | req->flags |= REQ_F_INFLIGHT; | |
6031 | req->work.files = current->files; | |
6032 | ret = 0; | |
6033 | } | |
6034 | spin_unlock_irq(&ctx->inflight_lock); | |
6035 | rcu_read_unlock(); | |
6036 | ||
6037 | return ret; | |
6038 | } | |
6039 | ||
f56040b8 PB |
6040 | static inline int io_prep_work_files(struct io_kiocb *req) |
6041 | { | |
6042 | if (!io_op_defs[req->opcode].file_table) | |
6043 | return 0; | |
6044 | return io_grab_files(req); | |
6045 | } | |
6046 | ||
2665abfd | 6047 | static enum hrtimer_restart io_link_timeout_fn(struct hrtimer *timer) |
2b188cc1 | 6048 | { |
ad8a48ac JA |
6049 | struct io_timeout_data *data = container_of(timer, |
6050 | struct io_timeout_data, timer); | |
6051 | struct io_kiocb *req = data->req; | |
2665abfd JA |
6052 | struct io_ring_ctx *ctx = req->ctx; |
6053 | struct io_kiocb *prev = NULL; | |
6054 | unsigned long flags; | |
2665abfd JA |
6055 | |
6056 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
6057 | ||
6058 | /* | |
6059 | * We don't expect the list to be empty, that will only happen if we | |
6060 | * race with the completion of the linked work. | |
6061 | */ | |
4493233e PB |
6062 | if (!list_empty(&req->link_list)) { |
6063 | prev = list_entry(req->link_list.prev, struct io_kiocb, | |
6064 | link_list); | |
5d960724 | 6065 | if (refcount_inc_not_zero(&prev->refs)) { |
4493233e | 6066 | list_del_init(&req->link_list); |
5d960724 JA |
6067 | prev->flags &= ~REQ_F_LINK_TIMEOUT; |
6068 | } else | |
76a46e06 | 6069 | prev = NULL; |
2665abfd JA |
6070 | } |
6071 | ||
6072 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
6073 | ||
6074 | if (prev) { | |
4e88d6e7 | 6075 | req_set_fail_links(prev); |
014db007 | 6076 | io_async_find_and_cancel(ctx, req, prev->user_data, -ETIME); |
76a46e06 | 6077 | io_put_req(prev); |
47f46768 | 6078 | } else { |
e1e16097 | 6079 | io_req_complete(req, -ETIME); |
2665abfd | 6080 | } |
2665abfd JA |
6081 | return HRTIMER_NORESTART; |
6082 | } | |
6083 | ||
7271ef3a | 6084 | static void __io_queue_linked_timeout(struct io_kiocb *req) |
2665abfd | 6085 | { |
76a46e06 JA |
6086 | /* |
6087 | * If the list is now empty, then our linked request finished before | |
6088 | * we got a chance to setup the timer | |
6089 | */ | |
4493233e | 6090 | if (!list_empty(&req->link_list)) { |
2d28390a | 6091 | struct io_timeout_data *data = &req->io->timeout; |
94ae5e77 | 6092 | |
ad8a48ac JA |
6093 | data->timer.function = io_link_timeout_fn; |
6094 | hrtimer_start(&data->timer, timespec64_to_ktime(data->ts), | |
6095 | data->mode); | |
2665abfd | 6096 | } |
7271ef3a JA |
6097 | } |
6098 | ||
6099 | static void io_queue_linked_timeout(struct io_kiocb *req) | |
6100 | { | |
6101 | struct io_ring_ctx *ctx = req->ctx; | |
6102 | ||
6103 | spin_lock_irq(&ctx->completion_lock); | |
6104 | __io_queue_linked_timeout(req); | |
76a46e06 | 6105 | spin_unlock_irq(&ctx->completion_lock); |
2665abfd | 6106 | |
2665abfd | 6107 | /* drop submission reference */ |
76a46e06 JA |
6108 | io_put_req(req); |
6109 | } | |
2665abfd | 6110 | |
ad8a48ac | 6111 | static struct io_kiocb *io_prep_linked_timeout(struct io_kiocb *req) |
2665abfd JA |
6112 | { |
6113 | struct io_kiocb *nxt; | |
6114 | ||
dea3b49c | 6115 | if (!(req->flags & REQ_F_LINK_HEAD)) |
2665abfd | 6116 | return NULL; |
6df1db6b | 6117 | if (req->flags & REQ_F_LINK_TIMEOUT) |
d7718a9d | 6118 | return NULL; |
2665abfd | 6119 | |
4493233e PB |
6120 | nxt = list_first_entry_or_null(&req->link_list, struct io_kiocb, |
6121 | link_list); | |
d625c6ee | 6122 | if (!nxt || nxt->opcode != IORING_OP_LINK_TIMEOUT) |
76a46e06 | 6123 | return NULL; |
2665abfd | 6124 | |
76a46e06 | 6125 | req->flags |= REQ_F_LINK_TIMEOUT; |
76a46e06 | 6126 | return nxt; |
2665abfd JA |
6127 | } |
6128 | ||
f13fad7b JA |
6129 | static void __io_queue_sqe(struct io_kiocb *req, const struct io_uring_sqe *sqe, |
6130 | struct io_comp_state *cs) | |
2b188cc1 | 6131 | { |
4a0a7a18 | 6132 | struct io_kiocb *linked_timeout; |
4bc4494e | 6133 | struct io_kiocb *nxt; |
193155c8 | 6134 | const struct cred *old_creds = NULL; |
e0c5c576 | 6135 | int ret; |
2b188cc1 | 6136 | |
4a0a7a18 JA |
6137 | again: |
6138 | linked_timeout = io_prep_linked_timeout(req); | |
6139 | ||
7cdaf587 XW |
6140 | if ((req->flags & REQ_F_WORK_INITIALIZED) && req->work.creds && |
6141 | req->work.creds != current_cred()) { | |
193155c8 JA |
6142 | if (old_creds) |
6143 | revert_creds(old_creds); | |
6144 | if (old_creds == req->work.creds) | |
6145 | old_creds = NULL; /* restored original creds */ | |
6146 | else | |
6147 | old_creds = override_creds(req->work.creds); | |
6148 | } | |
6149 | ||
f13fad7b | 6150 | ret = io_issue_sqe(req, sqe, true, cs); |
491381ce JA |
6151 | |
6152 | /* | |
6153 | * We async punt it if the file wasn't marked NOWAIT, or if the file | |
6154 | * doesn't support non-blocking read/write attempts | |
6155 | */ | |
24c74678 | 6156 | if (ret == -EAGAIN && !(req->flags & REQ_F_NOWAIT)) { |
f063c547 | 6157 | if (!io_arm_poll_handler(req)) { |
86a761f8 | 6158 | punt: |
f063c547 PB |
6159 | ret = io_prep_work_files(req); |
6160 | if (unlikely(ret)) | |
bbad27b2 | 6161 | goto err; |
f063c547 PB |
6162 | /* |
6163 | * Queued up for async execution, worker will release | |
6164 | * submit reference when the iocb is actually submitted. | |
6165 | */ | |
6166 | io_queue_async_work(req); | |
2b188cc1 | 6167 | } |
bbad27b2 | 6168 | |
f063c547 PB |
6169 | if (linked_timeout) |
6170 | io_queue_linked_timeout(linked_timeout); | |
4bc4494e | 6171 | goto exit; |
2b188cc1 | 6172 | } |
e65ef56d | 6173 | |
652532ad | 6174 | if (unlikely(ret)) { |
fcb323cc | 6175 | err: |
652532ad PB |
6176 | /* un-prep timeout, so it'll be killed as any other linked */ |
6177 | req->flags &= ~REQ_F_LINK_TIMEOUT; | |
4e88d6e7 | 6178 | req_set_fail_links(req); |
e65ef56d | 6179 | io_put_req(req); |
e1e16097 | 6180 | io_req_complete(req, ret); |
652532ad | 6181 | goto exit; |
9e645e11 | 6182 | } |
652532ad PB |
6183 | |
6184 | /* drop submission reference */ | |
6185 | nxt = io_put_req_find_next(req); | |
6186 | if (linked_timeout) | |
6187 | io_queue_linked_timeout(linked_timeout); | |
6188 | ||
4a0a7a18 JA |
6189 | if (nxt) { |
6190 | req = nxt; | |
86a761f8 PB |
6191 | |
6192 | if (req->flags & REQ_F_FORCE_ASYNC) | |
6193 | goto punt; | |
4a0a7a18 JA |
6194 | goto again; |
6195 | } | |
4bc4494e | 6196 | exit: |
193155c8 JA |
6197 | if (old_creds) |
6198 | revert_creds(old_creds); | |
2b188cc1 JA |
6199 | } |
6200 | ||
f13fad7b JA |
6201 | static void io_queue_sqe(struct io_kiocb *req, const struct io_uring_sqe *sqe, |
6202 | struct io_comp_state *cs) | |
4fe2c963 JL |
6203 | { |
6204 | int ret; | |
6205 | ||
3529d8c2 | 6206 | ret = io_req_defer(req, sqe); |
4fe2c963 JL |
6207 | if (ret) { |
6208 | if (ret != -EIOCBQUEUED) { | |
1118591a | 6209 | fail_req: |
4e88d6e7 | 6210 | req_set_fail_links(req); |
e1e16097 JA |
6211 | io_put_req(req); |
6212 | io_req_complete(req, ret); | |
4fe2c963 | 6213 | } |
2550878f | 6214 | } else if (req->flags & REQ_F_FORCE_ASYNC) { |
bd2ab18a | 6215 | if (!req->io) { |
bd2ab18a | 6216 | ret = io_req_defer_prep(req, sqe); |
327d6d96 | 6217 | if (unlikely(ret)) |
bd2ab18a PB |
6218 | goto fail_req; |
6219 | } | |
6220 | ||
ce35a47a JA |
6221 | /* |
6222 | * Never try inline submit of IOSQE_ASYNC is set, go straight | |
6223 | * to async execution. | |
6224 | */ | |
3e863ea3 | 6225 | io_req_init_async(req); |
ce35a47a JA |
6226 | req->work.flags |= IO_WQ_WORK_CONCURRENT; |
6227 | io_queue_async_work(req); | |
6228 | } else { | |
f13fad7b | 6229 | __io_queue_sqe(req, sqe, cs); |
ce35a47a | 6230 | } |
4fe2c963 JL |
6231 | } |
6232 | ||
f13fad7b JA |
6233 | static inline void io_queue_link_head(struct io_kiocb *req, |
6234 | struct io_comp_state *cs) | |
4fe2c963 | 6235 | { |
94ae5e77 | 6236 | if (unlikely(req->flags & REQ_F_FAIL_LINK)) { |
e1e16097 JA |
6237 | io_put_req(req); |
6238 | io_req_complete(req, -ECANCELED); | |
1b4a51b6 | 6239 | } else |
f13fad7b | 6240 | io_queue_sqe(req, NULL, cs); |
4fe2c963 JL |
6241 | } |
6242 | ||
1d4240cc | 6243 | static int io_submit_sqe(struct io_kiocb *req, const struct io_uring_sqe *sqe, |
f13fad7b | 6244 | struct io_kiocb **link, struct io_comp_state *cs) |
9e645e11 | 6245 | { |
a197f664 | 6246 | struct io_ring_ctx *ctx = req->ctx; |
ef4ff581 | 6247 | int ret; |
9e645e11 | 6248 | |
9e645e11 JA |
6249 | /* |
6250 | * If we already have a head request, queue this one for async | |
6251 | * submittal once the head completes. If we don't have a head but | |
6252 | * IOSQE_IO_LINK is set in the sqe, start a new head. This one will be | |
6253 | * submitted sync once the chain is complete. If none of those | |
6254 | * conditions are true (normal request), then just queue it. | |
6255 | */ | |
6256 | if (*link) { | |
9d76377f | 6257 | struct io_kiocb *head = *link; |
4e88d6e7 | 6258 | |
8cdf2193 PB |
6259 | /* |
6260 | * Taking sequential execution of a link, draining both sides | |
6261 | * of the link also fullfils IOSQE_IO_DRAIN semantics for all | |
6262 | * requests in the link. So, it drains the head and the | |
6263 | * next after the link request. The last one is done via | |
6264 | * drain_next flag to persist the effect across calls. | |
6265 | */ | |
ef4ff581 | 6266 | if (req->flags & REQ_F_IO_DRAIN) { |
711be031 PB |
6267 | head->flags |= REQ_F_IO_DRAIN; |
6268 | ctx->drain_next = 1; | |
6269 | } | |
3529d8c2 | 6270 | ret = io_req_defer_prep(req, sqe); |
327d6d96 | 6271 | if (unlikely(ret)) { |
4e88d6e7 | 6272 | /* fail even hard links since we don't submit */ |
9d76377f | 6273 | head->flags |= REQ_F_FAIL_LINK; |
1d4240cc | 6274 | return ret; |
2d28390a | 6275 | } |
9d76377f | 6276 | trace_io_uring_link(ctx, req, head); |
c40f6379 | 6277 | io_get_req_task(req); |
9d76377f | 6278 | list_add_tail(&req->link_list, &head->link_list); |
32fe525b PB |
6279 | |
6280 | /* last request of a link, enqueue the link */ | |
ef4ff581 | 6281 | if (!(req->flags & (REQ_F_LINK | REQ_F_HARDLINK))) { |
f13fad7b | 6282 | io_queue_link_head(head, cs); |
32fe525b PB |
6283 | *link = NULL; |
6284 | } | |
9e645e11 | 6285 | } else { |
711be031 PB |
6286 | if (unlikely(ctx->drain_next)) { |
6287 | req->flags |= REQ_F_IO_DRAIN; | |
ef4ff581 | 6288 | ctx->drain_next = 0; |
711be031 | 6289 | } |
ef4ff581 | 6290 | if (req->flags & (REQ_F_LINK | REQ_F_HARDLINK)) { |
dea3b49c | 6291 | req->flags |= REQ_F_LINK_HEAD; |
711be031 | 6292 | INIT_LIST_HEAD(&req->link_list); |
f1d96a8f | 6293 | |
711be031 | 6294 | ret = io_req_defer_prep(req, sqe); |
327d6d96 | 6295 | if (unlikely(ret)) |
711be031 PB |
6296 | req->flags |= REQ_F_FAIL_LINK; |
6297 | *link = req; | |
6298 | } else { | |
f13fad7b | 6299 | io_queue_sqe(req, sqe, cs); |
711be031 | 6300 | } |
9e645e11 | 6301 | } |
2e6e1fde | 6302 | |
1d4240cc | 6303 | return 0; |
9e645e11 JA |
6304 | } |
6305 | ||
9a56a232 JA |
6306 | /* |
6307 | * Batched submission is done, ensure local IO is flushed out. | |
6308 | */ | |
6309 | static void io_submit_state_end(struct io_submit_state *state) | |
6310 | { | |
f13fad7b JA |
6311 | if (!list_empty(&state->comp.list)) |
6312 | io_submit_flush_completions(&state->comp); | |
9a56a232 | 6313 | blk_finish_plug(&state->plug); |
9f13c35b | 6314 | io_state_file_put(state); |
2579f913 | 6315 | if (state->free_reqs) |
6c8a3134 | 6316 | kmem_cache_free_bulk(req_cachep, state->free_reqs, state->reqs); |
9a56a232 JA |
6317 | } |
6318 | ||
6319 | /* | |
6320 | * Start submission side cache. | |
6321 | */ | |
6322 | static void io_submit_state_start(struct io_submit_state *state, | |
013538bd | 6323 | struct io_ring_ctx *ctx, unsigned int max_ios) |
9a56a232 JA |
6324 | { |
6325 | blk_start_plug(&state->plug); | |
b63534c4 JA |
6326 | #ifdef CONFIG_BLOCK |
6327 | state->plug.nowait = true; | |
6328 | #endif | |
013538bd JA |
6329 | state->comp.nr = 0; |
6330 | INIT_LIST_HEAD(&state->comp.list); | |
6331 | state->comp.ctx = ctx; | |
2579f913 | 6332 | state->free_reqs = 0; |
9a56a232 JA |
6333 | state->file = NULL; |
6334 | state->ios_left = max_ios; | |
6335 | } | |
6336 | ||
2b188cc1 JA |
6337 | static void io_commit_sqring(struct io_ring_ctx *ctx) |
6338 | { | |
75b28aff | 6339 | struct io_rings *rings = ctx->rings; |
2b188cc1 | 6340 | |
caf582c6 PB |
6341 | /* |
6342 | * Ensure any loads from the SQEs are done at this point, | |
6343 | * since once we write the new head, the application could | |
6344 | * write new data to them. | |
6345 | */ | |
6346 | smp_store_release(&rings->sq.head, ctx->cached_sq_head); | |
2b188cc1 JA |
6347 | } |
6348 | ||
2b188cc1 | 6349 | /* |
3529d8c2 | 6350 | * Fetch an sqe, if one is available. Note that sqe_ptr will point to memory |
2b188cc1 JA |
6351 | * that is mapped by userspace. This means that care needs to be taken to |
6352 | * ensure that reads are stable, as we cannot rely on userspace always | |
6353 | * being a good citizen. If members of the sqe are validated and then later | |
6354 | * used, it's important that those reads are done through READ_ONCE() to | |
6355 | * prevent a re-load down the line. | |
6356 | */ | |
709b302f | 6357 | static const struct io_uring_sqe *io_get_sqe(struct io_ring_ctx *ctx) |
2b188cc1 | 6358 | { |
75b28aff | 6359 | u32 *sq_array = ctx->sq_array; |
2b188cc1 JA |
6360 | unsigned head; |
6361 | ||
6362 | /* | |
6363 | * The cached sq head (or cq tail) serves two purposes: | |
6364 | * | |
6365 | * 1) allows us to batch the cost of updating the user visible | |
6366 | * head updates. | |
6367 | * 2) allows the kernel side to track the head on its own, even | |
6368 | * though the application is the one updating it. | |
6369 | */ | |
ee7d46d9 | 6370 | head = READ_ONCE(sq_array[ctx->cached_sq_head & ctx->sq_mask]); |
709b302f PB |
6371 | if (likely(head < ctx->sq_entries)) |
6372 | return &ctx->sq_sqes[head]; | |
2b188cc1 JA |
6373 | |
6374 | /* drop invalid entries */ | |
498ccd9e | 6375 | ctx->cached_sq_dropped++; |
ee7d46d9 | 6376 | WRITE_ONCE(ctx->rings->sq_dropped, ctx->cached_sq_dropped); |
709b302f PB |
6377 | return NULL; |
6378 | } | |
6379 | ||
6380 | static inline void io_consume_sqe(struct io_ring_ctx *ctx) | |
6381 | { | |
6382 | ctx->cached_sq_head++; | |
2b188cc1 JA |
6383 | } |
6384 | ||
ef4ff581 PB |
6385 | #define SQE_VALID_FLAGS (IOSQE_FIXED_FILE|IOSQE_IO_DRAIN|IOSQE_IO_LINK| \ |
6386 | IOSQE_IO_HARDLINK | IOSQE_ASYNC | \ | |
6387 | IOSQE_BUFFER_SELECT) | |
6388 | ||
6389 | static int io_init_req(struct io_ring_ctx *ctx, struct io_kiocb *req, | |
6390 | const struct io_uring_sqe *sqe, | |
0cdaf760 | 6391 | struct io_submit_state *state) |
0553b8bd | 6392 | { |
ef4ff581 | 6393 | unsigned int sqe_flags; |
63ff8223 | 6394 | int id; |
ef4ff581 | 6395 | |
0553b8bd PB |
6396 | req->opcode = READ_ONCE(sqe->opcode); |
6397 | req->user_data = READ_ONCE(sqe->user_data); | |
6398 | req->io = NULL; | |
6399 | req->file = NULL; | |
6400 | req->ctx = ctx; | |
6401 | req->flags = 0; | |
6402 | /* one is dropped after submission, the other at completion */ | |
6403 | refcount_set(&req->refs, 2); | |
4dd2824d | 6404 | req->task = current; |
0553b8bd | 6405 | req->result = 0; |
ef4ff581 PB |
6406 | |
6407 | if (unlikely(req->opcode >= IORING_OP_LAST)) | |
6408 | return -EINVAL; | |
6409 | ||
9d8426a0 JA |
6410 | if (unlikely(io_sq_thread_acquire_mm(ctx, req))) |
6411 | return -EFAULT; | |
ef4ff581 PB |
6412 | |
6413 | sqe_flags = READ_ONCE(sqe->flags); | |
6414 | /* enforce forwards compatibility on users */ | |
6415 | if (unlikely(sqe_flags & ~SQE_VALID_FLAGS)) | |
6416 | return -EINVAL; | |
6417 | ||
6418 | if ((sqe_flags & IOSQE_BUFFER_SELECT) && | |
6419 | !io_op_defs[req->opcode].buffer_select) | |
6420 | return -EOPNOTSUPP; | |
6421 | ||
6422 | id = READ_ONCE(sqe->personality); | |
6423 | if (id) { | |
7cdaf587 | 6424 | io_req_init_async(req); |
ef4ff581 PB |
6425 | req->work.creds = idr_find(&ctx->personality_idr, id); |
6426 | if (unlikely(!req->work.creds)) | |
6427 | return -EINVAL; | |
6428 | get_cred(req->work.creds); | |
6429 | } | |
6430 | ||
6431 | /* same numerical values with corresponding REQ_F_*, safe to copy */ | |
c11368a5 | 6432 | req->flags |= sqe_flags; |
ef4ff581 | 6433 | |
63ff8223 JA |
6434 | if (!io_op_defs[req->opcode].needs_file) |
6435 | return 0; | |
6436 | ||
6437 | return io_req_set_file(state, req, READ_ONCE(sqe->fd)); | |
0553b8bd PB |
6438 | } |
6439 | ||
fb5ccc98 | 6440 | static int io_submit_sqes(struct io_ring_ctx *ctx, unsigned int nr, |
0cdaf760 | 6441 | struct file *ring_file, int ring_fd) |
6c271ce2 | 6442 | { |
ac8691c4 | 6443 | struct io_submit_state state; |
9e645e11 | 6444 | struct io_kiocb *link = NULL; |
9e645e11 | 6445 | int i, submitted = 0; |
6c271ce2 | 6446 | |
c4a2ed72 | 6447 | /* if we have a backlog and couldn't flush it all, return BUSY */ |
ad3eb2c8 JA |
6448 | if (test_bit(0, &ctx->sq_check_overflow)) { |
6449 | if (!list_empty(&ctx->cq_overflow_list) && | |
6450 | !io_cqring_overflow_flush(ctx, false)) | |
6451 | return -EBUSY; | |
6452 | } | |
6c271ce2 | 6453 | |
ee7d46d9 PB |
6454 | /* make sure SQ entry isn't read before tail */ |
6455 | nr = min3(nr, ctx->sq_entries, io_sqring_entries(ctx)); | |
9ef4f124 | 6456 | |
2b85edfc PB |
6457 | if (!percpu_ref_tryget_many(&ctx->refs, nr)) |
6458 | return -EAGAIN; | |
6c271ce2 | 6459 | |
013538bd | 6460 | io_submit_state_start(&state, ctx, nr); |
6c271ce2 | 6461 | |
b14cca0c PB |
6462 | ctx->ring_fd = ring_fd; |
6463 | ctx->ring_file = ring_file; | |
6464 | ||
6c271ce2 | 6465 | for (i = 0; i < nr; i++) { |
3529d8c2 | 6466 | const struct io_uring_sqe *sqe; |
196be95c | 6467 | struct io_kiocb *req; |
1cb1edb2 | 6468 | int err; |
fb5ccc98 | 6469 | |
b1e50e54 PB |
6470 | sqe = io_get_sqe(ctx); |
6471 | if (unlikely(!sqe)) { | |
6472 | io_consume_sqe(ctx); | |
6473 | break; | |
6474 | } | |
ac8691c4 | 6475 | req = io_alloc_req(ctx, &state); |
196be95c PB |
6476 | if (unlikely(!req)) { |
6477 | if (!submitted) | |
6478 | submitted = -EAGAIN; | |
fb5ccc98 | 6479 | break; |
196be95c | 6480 | } |
fb5ccc98 | 6481 | |
ac8691c4 | 6482 | err = io_init_req(ctx, req, sqe, &state); |
709b302f | 6483 | io_consume_sqe(ctx); |
d3656344 JA |
6484 | /* will complete beyond this point, count as submitted */ |
6485 | submitted++; | |
6486 | ||
ef4ff581 | 6487 | if (unlikely(err)) { |
1cb1edb2 | 6488 | fail_req: |
e1e16097 JA |
6489 | io_put_req(req); |
6490 | io_req_complete(req, err); | |
196be95c PB |
6491 | break; |
6492 | } | |
fb5ccc98 | 6493 | |
354420f7 | 6494 | trace_io_uring_submit_sqe(ctx, req->opcode, req->user_data, |
0cdaf760 | 6495 | true, io_async_submit(ctx)); |
f13fad7b | 6496 | err = io_submit_sqe(req, sqe, &link, &state.comp); |
1d4240cc PB |
6497 | if (err) |
6498 | goto fail_req; | |
6c271ce2 JA |
6499 | } |
6500 | ||
9466f437 PB |
6501 | if (unlikely(submitted != nr)) { |
6502 | int ref_used = (submitted == -EAGAIN) ? 0 : submitted; | |
6503 | ||
6504 | percpu_ref_put_many(&ctx->refs, nr - ref_used); | |
6505 | } | |
9e645e11 | 6506 | if (link) |
f13fad7b | 6507 | io_queue_link_head(link, &state.comp); |
ac8691c4 | 6508 | io_submit_state_end(&state); |
6c271ce2 | 6509 | |
ae9428ca PB |
6510 | /* Commit SQ ring head once we've consumed and submitted all SQEs */ |
6511 | io_commit_sqring(ctx); | |
6512 | ||
6c271ce2 JA |
6513 | return submitted; |
6514 | } | |
6515 | ||
23b3628e XW |
6516 | static inline void io_ring_set_wakeup_flag(struct io_ring_ctx *ctx) |
6517 | { | |
6518 | /* Tell userspace we may need a wakeup call */ | |
6519 | spin_lock_irq(&ctx->completion_lock); | |
6520 | ctx->rings->sq_flags |= IORING_SQ_NEED_WAKEUP; | |
6521 | spin_unlock_irq(&ctx->completion_lock); | |
6522 | } | |
6523 | ||
6524 | static inline void io_ring_clear_wakeup_flag(struct io_ring_ctx *ctx) | |
6525 | { | |
6526 | spin_lock_irq(&ctx->completion_lock); | |
6527 | ctx->rings->sq_flags &= ~IORING_SQ_NEED_WAKEUP; | |
6528 | spin_unlock_irq(&ctx->completion_lock); | |
6529 | } | |
6530 | ||
6c271ce2 JA |
6531 | static int io_sq_thread(void *data) |
6532 | { | |
6c271ce2 | 6533 | struct io_ring_ctx *ctx = data; |
181e448d | 6534 | const struct cred *old_cred; |
6c271ce2 | 6535 | DEFINE_WAIT(wait); |
6c271ce2 | 6536 | unsigned long timeout; |
bdcd3eab | 6537 | int ret = 0; |
6c271ce2 | 6538 | |
0f158b4c | 6539 | complete(&ctx->sq_thread_comp); |
a4c0b3de | 6540 | |
181e448d | 6541 | old_cred = override_creds(ctx->creds); |
6c271ce2 | 6542 | |
bdcd3eab | 6543 | timeout = jiffies + ctx->sq_thread_idle; |
2bbcd6d3 | 6544 | while (!kthread_should_park()) { |
fb5ccc98 | 6545 | unsigned int to_submit; |
6c271ce2 | 6546 | |
540e32a0 | 6547 | if (!list_empty(&ctx->iopoll_list)) { |
6c271ce2 JA |
6548 | unsigned nr_events = 0; |
6549 | ||
bdcd3eab | 6550 | mutex_lock(&ctx->uring_lock); |
540e32a0 | 6551 | if (!list_empty(&ctx->iopoll_list) && !need_resched()) |
9dedd563 | 6552 | io_do_iopoll(ctx, &nr_events, 0); |
bdcd3eab | 6553 | else |
6c271ce2 | 6554 | timeout = jiffies + ctx->sq_thread_idle; |
bdcd3eab | 6555 | mutex_unlock(&ctx->uring_lock); |
6c271ce2 JA |
6556 | } |
6557 | ||
fb5ccc98 | 6558 | to_submit = io_sqring_entries(ctx); |
c1edbf5f JA |
6559 | |
6560 | /* | |
6561 | * If submit got -EBUSY, flag us as needing the application | |
6562 | * to enter the kernel to reap and flush events. | |
6563 | */ | |
b772f07a | 6564 | if (!to_submit || ret == -EBUSY || need_resched()) { |
7143b5ac SG |
6565 | /* |
6566 | * Drop cur_mm before scheduling, we can't hold it for | |
6567 | * long periods (or over schedule()). Do this before | |
6568 | * adding ourselves to the waitqueue, as the unuse/drop | |
6569 | * may sleep. | |
6570 | */ | |
4349f30e | 6571 | io_sq_thread_drop_mm(); |
7143b5ac | 6572 | |
6c271ce2 JA |
6573 | /* |
6574 | * We're polling. If we're within the defined idle | |
6575 | * period, then let us spin without work before going | |
c1edbf5f JA |
6576 | * to sleep. The exception is if we got EBUSY doing |
6577 | * more IO, we should wait for the application to | |
6578 | * reap events and wake us up. | |
6c271ce2 | 6579 | */ |
540e32a0 | 6580 | if (!list_empty(&ctx->iopoll_list) || need_resched() || |
df069d80 JA |
6581 | (!time_after(jiffies, timeout) && ret != -EBUSY && |
6582 | !percpu_ref_is_dying(&ctx->refs))) { | |
4c6e277c | 6583 | io_run_task_work(); |
9831a90c | 6584 | cond_resched(); |
6c271ce2 JA |
6585 | continue; |
6586 | } | |
6587 | ||
6c271ce2 JA |
6588 | prepare_to_wait(&ctx->sqo_wait, &wait, |
6589 | TASK_INTERRUPTIBLE); | |
6590 | ||
bdcd3eab XW |
6591 | /* |
6592 | * While doing polled IO, before going to sleep, we need | |
540e32a0 PB |
6593 | * to check if there are new reqs added to iopoll_list, |
6594 | * it is because reqs may have been punted to io worker | |
6595 | * and will be added to iopoll_list later, hence check | |
6596 | * the iopoll_list again. | |
bdcd3eab XW |
6597 | */ |
6598 | if ((ctx->flags & IORING_SETUP_IOPOLL) && | |
540e32a0 | 6599 | !list_empty_careful(&ctx->iopoll_list)) { |
bdcd3eab XW |
6600 | finish_wait(&ctx->sqo_wait, &wait); |
6601 | continue; | |
6602 | } | |
6603 | ||
23b3628e | 6604 | io_ring_set_wakeup_flag(ctx); |
6c271ce2 | 6605 | |
fb5ccc98 | 6606 | to_submit = io_sqring_entries(ctx); |
c1edbf5f | 6607 | if (!to_submit || ret == -EBUSY) { |
2bbcd6d3 | 6608 | if (kthread_should_park()) { |
6c271ce2 JA |
6609 | finish_wait(&ctx->sqo_wait, &wait); |
6610 | break; | |
6611 | } | |
4c6e277c | 6612 | if (io_run_task_work()) { |
10bea96d | 6613 | finish_wait(&ctx->sqo_wait, &wait); |
23b3628e | 6614 | io_ring_clear_wakeup_flag(ctx); |
b41e9852 JA |
6615 | continue; |
6616 | } | |
6c271ce2 JA |
6617 | if (signal_pending(current)) |
6618 | flush_signals(current); | |
6619 | schedule(); | |
6620 | finish_wait(&ctx->sqo_wait, &wait); | |
6621 | ||
23b3628e | 6622 | io_ring_clear_wakeup_flag(ctx); |
d4ae271d | 6623 | ret = 0; |
6c271ce2 JA |
6624 | continue; |
6625 | } | |
6626 | finish_wait(&ctx->sqo_wait, &wait); | |
6627 | ||
23b3628e | 6628 | io_ring_clear_wakeup_flag(ctx); |
6c271ce2 JA |
6629 | } |
6630 | ||
8a4955ff | 6631 | mutex_lock(&ctx->uring_lock); |
6b668c9b XW |
6632 | if (likely(!percpu_ref_is_dying(&ctx->refs))) |
6633 | ret = io_submit_sqes(ctx, to_submit, NULL, -1); | |
8a4955ff | 6634 | mutex_unlock(&ctx->uring_lock); |
bdcd3eab | 6635 | timeout = jiffies + ctx->sq_thread_idle; |
6c271ce2 JA |
6636 | } |
6637 | ||
4c6e277c | 6638 | io_run_task_work(); |
b41e9852 | 6639 | |
4349f30e | 6640 | io_sq_thread_drop_mm(); |
181e448d | 6641 | revert_creds(old_cred); |
06058632 | 6642 | |
2bbcd6d3 | 6643 | kthread_parkme(); |
06058632 | 6644 | |
6c271ce2 JA |
6645 | return 0; |
6646 | } | |
6647 | ||
bda52162 JA |
6648 | struct io_wait_queue { |
6649 | struct wait_queue_entry wq; | |
6650 | struct io_ring_ctx *ctx; | |
6651 | unsigned to_wait; | |
6652 | unsigned nr_timeouts; | |
6653 | }; | |
6654 | ||
1d7bb1d5 | 6655 | static inline bool io_should_wake(struct io_wait_queue *iowq, bool noflush) |
bda52162 JA |
6656 | { |
6657 | struct io_ring_ctx *ctx = iowq->ctx; | |
6658 | ||
6659 | /* | |
d195a66e | 6660 | * Wake up if we have enough events, or if a timeout occurred since we |
bda52162 JA |
6661 | * started waiting. For timeouts, we always want to return to userspace, |
6662 | * regardless of event count. | |
6663 | */ | |
1d7bb1d5 | 6664 | return io_cqring_events(ctx, noflush) >= iowq->to_wait || |
bda52162 JA |
6665 | atomic_read(&ctx->cq_timeouts) != iowq->nr_timeouts; |
6666 | } | |
6667 | ||
6668 | static int io_wake_function(struct wait_queue_entry *curr, unsigned int mode, | |
6669 | int wake_flags, void *key) | |
6670 | { | |
6671 | struct io_wait_queue *iowq = container_of(curr, struct io_wait_queue, | |
6672 | wq); | |
6673 | ||
1d7bb1d5 JA |
6674 | /* use noflush == true, as we can't safely rely on locking context */ |
6675 | if (!io_should_wake(iowq, true)) | |
bda52162 JA |
6676 | return -1; |
6677 | ||
6678 | return autoremove_wake_function(curr, mode, wake_flags, key); | |
6679 | } | |
6680 | ||
2b188cc1 JA |
6681 | /* |
6682 | * Wait until events become available, if we don't already have some. The | |
6683 | * application must reap them itself, as they reside on the shared cq ring. | |
6684 | */ | |
6685 | static int io_cqring_wait(struct io_ring_ctx *ctx, int min_events, | |
6686 | const sigset_t __user *sig, size_t sigsz) | |
6687 | { | |
bda52162 JA |
6688 | struct io_wait_queue iowq = { |
6689 | .wq = { | |
6690 | .private = current, | |
6691 | .func = io_wake_function, | |
6692 | .entry = LIST_HEAD_INIT(iowq.wq.entry), | |
6693 | }, | |
6694 | .ctx = ctx, | |
6695 | .to_wait = min_events, | |
6696 | }; | |
75b28aff | 6697 | struct io_rings *rings = ctx->rings; |
e9ffa5c2 | 6698 | int ret = 0; |
2b188cc1 | 6699 | |
b41e9852 JA |
6700 | do { |
6701 | if (io_cqring_events(ctx, false) >= min_events) | |
6702 | return 0; | |
4c6e277c | 6703 | if (!io_run_task_work()) |
b41e9852 | 6704 | break; |
b41e9852 | 6705 | } while (1); |
2b188cc1 JA |
6706 | |
6707 | if (sig) { | |
9e75ad5d AB |
6708 | #ifdef CONFIG_COMPAT |
6709 | if (in_compat_syscall()) | |
6710 | ret = set_compat_user_sigmask((const compat_sigset_t __user *)sig, | |
b772434b | 6711 | sigsz); |
9e75ad5d AB |
6712 | else |
6713 | #endif | |
b772434b | 6714 | ret = set_user_sigmask(sig, sigsz); |
9e75ad5d | 6715 | |
2b188cc1 JA |
6716 | if (ret) |
6717 | return ret; | |
6718 | } | |
6719 | ||
bda52162 | 6720 | iowq.nr_timeouts = atomic_read(&ctx->cq_timeouts); |
c826bd7a | 6721 | trace_io_uring_cqring_wait(ctx, min_events); |
bda52162 JA |
6722 | do { |
6723 | prepare_to_wait_exclusive(&ctx->wait, &iowq.wq, | |
6724 | TASK_INTERRUPTIBLE); | |
ce593a6c | 6725 | /* make sure we run task_work before checking for signals */ |
4c6e277c JA |
6726 | if (io_run_task_work()) |
6727 | continue; | |
bda52162 | 6728 | if (signal_pending(current)) { |
b7db41c9 JA |
6729 | if (current->jobctl & JOBCTL_TASK_WORK) { |
6730 | spin_lock_irq(¤t->sighand->siglock); | |
6731 | current->jobctl &= ~JOBCTL_TASK_WORK; | |
6732 | recalc_sigpending(); | |
6733 | spin_unlock_irq(¤t->sighand->siglock); | |
6734 | continue; | |
6735 | } | |
6736 | ret = -EINTR; | |
bda52162 JA |
6737 | break; |
6738 | } | |
ce593a6c JA |
6739 | if (io_should_wake(&iowq, false)) |
6740 | break; | |
6741 | schedule(); | |
bda52162 JA |
6742 | } while (1); |
6743 | finish_wait(&ctx->wait, &iowq.wq); | |
6744 | ||
b7db41c9 | 6745 | restore_saved_sigmask_unless(ret == -EINTR); |
2b188cc1 | 6746 | |
75b28aff | 6747 | return READ_ONCE(rings->cq.head) == READ_ONCE(rings->cq.tail) ? ret : 0; |
2b188cc1 JA |
6748 | } |
6749 | ||
6b06314c JA |
6750 | static void __io_sqe_files_unregister(struct io_ring_ctx *ctx) |
6751 | { | |
6752 | #if defined(CONFIG_UNIX) | |
6753 | if (ctx->ring_sock) { | |
6754 | struct sock *sock = ctx->ring_sock->sk; | |
6755 | struct sk_buff *skb; | |
6756 | ||
6757 | while ((skb = skb_dequeue(&sock->sk_receive_queue)) != NULL) | |
6758 | kfree_skb(skb); | |
6759 | } | |
6760 | #else | |
6761 | int i; | |
6762 | ||
65e19f54 JA |
6763 | for (i = 0; i < ctx->nr_user_files; i++) { |
6764 | struct file *file; | |
6765 | ||
6766 | file = io_file_from_index(ctx, i); | |
6767 | if (file) | |
6768 | fput(file); | |
6769 | } | |
6b06314c JA |
6770 | #endif |
6771 | } | |
6772 | ||
05f3fb3c JA |
6773 | static void io_file_ref_kill(struct percpu_ref *ref) |
6774 | { | |
6775 | struct fixed_file_data *data; | |
6776 | ||
6777 | data = container_of(ref, struct fixed_file_data, refs); | |
6778 | complete(&data->done); | |
6779 | } | |
6780 | ||
6b06314c JA |
6781 | static int io_sqe_files_unregister(struct io_ring_ctx *ctx) |
6782 | { | |
05f3fb3c | 6783 | struct fixed_file_data *data = ctx->file_data; |
05589553 | 6784 | struct fixed_file_ref_node *ref_node = NULL; |
65e19f54 JA |
6785 | unsigned nr_tables, i; |
6786 | ||
05f3fb3c | 6787 | if (!data) |
6b06314c JA |
6788 | return -ENXIO; |
6789 | ||
6a4d07cd | 6790 | spin_lock(&data->lock); |
05589553 XW |
6791 | if (!list_empty(&data->ref_list)) |
6792 | ref_node = list_first_entry(&data->ref_list, | |
6793 | struct fixed_file_ref_node, node); | |
6a4d07cd | 6794 | spin_unlock(&data->lock); |
05589553 XW |
6795 | if (ref_node) |
6796 | percpu_ref_kill(&ref_node->refs); | |
6797 | ||
6798 | percpu_ref_kill(&data->refs); | |
6799 | ||
6800 | /* wait for all refs nodes to complete */ | |
4a38aed2 | 6801 | flush_delayed_work(&ctx->file_put_work); |
2faf852d | 6802 | wait_for_completion(&data->done); |
05f3fb3c | 6803 | |
6b06314c | 6804 | __io_sqe_files_unregister(ctx); |
65e19f54 JA |
6805 | nr_tables = DIV_ROUND_UP(ctx->nr_user_files, IORING_MAX_FILES_TABLE); |
6806 | for (i = 0; i < nr_tables; i++) | |
05f3fb3c JA |
6807 | kfree(data->table[i].files); |
6808 | kfree(data->table); | |
05589553 XW |
6809 | percpu_ref_exit(&data->refs); |
6810 | kfree(data); | |
05f3fb3c | 6811 | ctx->file_data = NULL; |
6b06314c JA |
6812 | ctx->nr_user_files = 0; |
6813 | return 0; | |
6814 | } | |
6815 | ||
6c271ce2 JA |
6816 | static void io_sq_thread_stop(struct io_ring_ctx *ctx) |
6817 | { | |
6818 | if (ctx->sqo_thread) { | |
0f158b4c | 6819 | wait_for_completion(&ctx->sq_thread_comp); |
2bbcd6d3 RP |
6820 | /* |
6821 | * The park is a bit of a work-around, without it we get | |
6822 | * warning spews on shutdown with SQPOLL set and affinity | |
6823 | * set to a single CPU. | |
6824 | */ | |
06058632 | 6825 | kthread_park(ctx->sqo_thread); |
6c271ce2 JA |
6826 | kthread_stop(ctx->sqo_thread); |
6827 | ctx->sqo_thread = NULL; | |
6828 | } | |
6829 | } | |
6830 | ||
6b06314c JA |
6831 | static void io_finish_async(struct io_ring_ctx *ctx) |
6832 | { | |
6c271ce2 JA |
6833 | io_sq_thread_stop(ctx); |
6834 | ||
561fb04a JA |
6835 | if (ctx->io_wq) { |
6836 | io_wq_destroy(ctx->io_wq); | |
6837 | ctx->io_wq = NULL; | |
6b06314c JA |
6838 | } |
6839 | } | |
6840 | ||
6841 | #if defined(CONFIG_UNIX) | |
6b06314c JA |
6842 | /* |
6843 | * Ensure the UNIX gc is aware of our file set, so we are certain that | |
6844 | * the io_uring can be safely unregistered on process exit, even if we have | |
6845 | * loops in the file referencing. | |
6846 | */ | |
6847 | static int __io_sqe_files_scm(struct io_ring_ctx *ctx, int nr, int offset) | |
6848 | { | |
6849 | struct sock *sk = ctx->ring_sock->sk; | |
6850 | struct scm_fp_list *fpl; | |
6851 | struct sk_buff *skb; | |
08a45173 | 6852 | int i, nr_files; |
6b06314c | 6853 | |
6b06314c JA |
6854 | fpl = kzalloc(sizeof(*fpl), GFP_KERNEL); |
6855 | if (!fpl) | |
6856 | return -ENOMEM; | |
6857 | ||
6858 | skb = alloc_skb(0, GFP_KERNEL); | |
6859 | if (!skb) { | |
6860 | kfree(fpl); | |
6861 | return -ENOMEM; | |
6862 | } | |
6863 | ||
6864 | skb->sk = sk; | |
6b06314c | 6865 | |
08a45173 | 6866 | nr_files = 0; |
6b06314c JA |
6867 | fpl->user = get_uid(ctx->user); |
6868 | for (i = 0; i < nr; i++) { | |
65e19f54 JA |
6869 | struct file *file = io_file_from_index(ctx, i + offset); |
6870 | ||
6871 | if (!file) | |
08a45173 | 6872 | continue; |
65e19f54 | 6873 | fpl->fp[nr_files] = get_file(file); |
08a45173 JA |
6874 | unix_inflight(fpl->user, fpl->fp[nr_files]); |
6875 | nr_files++; | |
6b06314c JA |
6876 | } |
6877 | ||
08a45173 JA |
6878 | if (nr_files) { |
6879 | fpl->max = SCM_MAX_FD; | |
6880 | fpl->count = nr_files; | |
6881 | UNIXCB(skb).fp = fpl; | |
05f3fb3c | 6882 | skb->destructor = unix_destruct_scm; |
08a45173 JA |
6883 | refcount_add(skb->truesize, &sk->sk_wmem_alloc); |
6884 | skb_queue_head(&sk->sk_receive_queue, skb); | |
6b06314c | 6885 | |
08a45173 JA |
6886 | for (i = 0; i < nr_files; i++) |
6887 | fput(fpl->fp[i]); | |
6888 | } else { | |
6889 | kfree_skb(skb); | |
6890 | kfree(fpl); | |
6891 | } | |
6b06314c JA |
6892 | |
6893 | return 0; | |
6894 | } | |
6895 | ||
6896 | /* | |
6897 | * If UNIX sockets are enabled, fd passing can cause a reference cycle which | |
6898 | * causes regular reference counting to break down. We rely on the UNIX | |
6899 | * garbage collection to take care of this problem for us. | |
6900 | */ | |
6901 | static int io_sqe_files_scm(struct io_ring_ctx *ctx) | |
6902 | { | |
6903 | unsigned left, total; | |
6904 | int ret = 0; | |
6905 | ||
6906 | total = 0; | |
6907 | left = ctx->nr_user_files; | |
6908 | while (left) { | |
6909 | unsigned this_files = min_t(unsigned, left, SCM_MAX_FD); | |
6b06314c JA |
6910 | |
6911 | ret = __io_sqe_files_scm(ctx, this_files, total); | |
6912 | if (ret) | |
6913 | break; | |
6914 | left -= this_files; | |
6915 | total += this_files; | |
6916 | } | |
6917 | ||
6918 | if (!ret) | |
6919 | return 0; | |
6920 | ||
6921 | while (total < ctx->nr_user_files) { | |
65e19f54 JA |
6922 | struct file *file = io_file_from_index(ctx, total); |
6923 | ||
6924 | if (file) | |
6925 | fput(file); | |
6b06314c JA |
6926 | total++; |
6927 | } | |
6928 | ||
6929 | return ret; | |
6930 | } | |
6931 | #else | |
6932 | static int io_sqe_files_scm(struct io_ring_ctx *ctx) | |
6933 | { | |
6934 | return 0; | |
6935 | } | |
6936 | #endif | |
6937 | ||
65e19f54 JA |
6938 | static int io_sqe_alloc_file_tables(struct io_ring_ctx *ctx, unsigned nr_tables, |
6939 | unsigned nr_files) | |
6940 | { | |
6941 | int i; | |
6942 | ||
6943 | for (i = 0; i < nr_tables; i++) { | |
05f3fb3c | 6944 | struct fixed_file_table *table = &ctx->file_data->table[i]; |
65e19f54 JA |
6945 | unsigned this_files; |
6946 | ||
6947 | this_files = min(nr_files, IORING_MAX_FILES_TABLE); | |
6948 | table->files = kcalloc(this_files, sizeof(struct file *), | |
6949 | GFP_KERNEL); | |
6950 | if (!table->files) | |
6951 | break; | |
6952 | nr_files -= this_files; | |
6953 | } | |
6954 | ||
6955 | if (i == nr_tables) | |
6956 | return 0; | |
6957 | ||
6958 | for (i = 0; i < nr_tables; i++) { | |
05f3fb3c | 6959 | struct fixed_file_table *table = &ctx->file_data->table[i]; |
65e19f54 JA |
6960 | kfree(table->files); |
6961 | } | |
6962 | return 1; | |
6963 | } | |
6964 | ||
05f3fb3c JA |
6965 | static void io_ring_file_put(struct io_ring_ctx *ctx, struct file *file) |
6966 | { | |
6967 | #if defined(CONFIG_UNIX) | |
6968 | struct sock *sock = ctx->ring_sock->sk; | |
6969 | struct sk_buff_head list, *head = &sock->sk_receive_queue; | |
6970 | struct sk_buff *skb; | |
6971 | int i; | |
6972 | ||
6973 | __skb_queue_head_init(&list); | |
6974 | ||
6975 | /* | |
6976 | * Find the skb that holds this file in its SCM_RIGHTS. When found, | |
6977 | * remove this entry and rearrange the file array. | |
6978 | */ | |
6979 | skb = skb_dequeue(head); | |
6980 | while (skb) { | |
6981 | struct scm_fp_list *fp; | |
6982 | ||
6983 | fp = UNIXCB(skb).fp; | |
6984 | for (i = 0; i < fp->count; i++) { | |
6985 | int left; | |
6986 | ||
6987 | if (fp->fp[i] != file) | |
6988 | continue; | |
6989 | ||
6990 | unix_notinflight(fp->user, fp->fp[i]); | |
6991 | left = fp->count - 1 - i; | |
6992 | if (left) { | |
6993 | memmove(&fp->fp[i], &fp->fp[i + 1], | |
6994 | left * sizeof(struct file *)); | |
6995 | } | |
6996 | fp->count--; | |
6997 | if (!fp->count) { | |
6998 | kfree_skb(skb); | |
6999 | skb = NULL; | |
7000 | } else { | |
7001 | __skb_queue_tail(&list, skb); | |
7002 | } | |
7003 | fput(file); | |
7004 | file = NULL; | |
7005 | break; | |
7006 | } | |
7007 | ||
7008 | if (!file) | |
7009 | break; | |
7010 | ||
7011 | __skb_queue_tail(&list, skb); | |
7012 | ||
7013 | skb = skb_dequeue(head); | |
7014 | } | |
7015 | ||
7016 | if (skb_peek(&list)) { | |
7017 | spin_lock_irq(&head->lock); | |
7018 | while ((skb = __skb_dequeue(&list)) != NULL) | |
7019 | __skb_queue_tail(head, skb); | |
7020 | spin_unlock_irq(&head->lock); | |
7021 | } | |
7022 | #else | |
7023 | fput(file); | |
7024 | #endif | |
7025 | } | |
7026 | ||
7027 | struct io_file_put { | |
05589553 | 7028 | struct list_head list; |
05f3fb3c | 7029 | struct file *file; |
05f3fb3c JA |
7030 | }; |
7031 | ||
4a38aed2 | 7032 | static void __io_file_put_work(struct fixed_file_ref_node *ref_node) |
65e19f54 | 7033 | { |
4a38aed2 JA |
7034 | struct fixed_file_data *file_data = ref_node->file_data; |
7035 | struct io_ring_ctx *ctx = file_data->ctx; | |
05f3fb3c | 7036 | struct io_file_put *pfile, *tmp; |
05589553 XW |
7037 | |
7038 | list_for_each_entry_safe(pfile, tmp, &ref_node->file_list, list) { | |
6a4d07cd | 7039 | list_del(&pfile->list); |
05589553 XW |
7040 | io_ring_file_put(ctx, pfile->file); |
7041 | kfree(pfile); | |
65e19f54 | 7042 | } |
05589553 | 7043 | |
6a4d07cd JA |
7044 | spin_lock(&file_data->lock); |
7045 | list_del(&ref_node->node); | |
7046 | spin_unlock(&file_data->lock); | |
05589553 XW |
7047 | |
7048 | percpu_ref_exit(&ref_node->refs); | |
7049 | kfree(ref_node); | |
7050 | percpu_ref_put(&file_data->refs); | |
2faf852d | 7051 | } |
65e19f54 | 7052 | |
4a38aed2 JA |
7053 | static void io_file_put_work(struct work_struct *work) |
7054 | { | |
7055 | struct io_ring_ctx *ctx; | |
7056 | struct llist_node *node; | |
7057 | ||
7058 | ctx = container_of(work, struct io_ring_ctx, file_put_work.work); | |
7059 | node = llist_del_all(&ctx->file_put_llist); | |
7060 | ||
7061 | while (node) { | |
7062 | struct fixed_file_ref_node *ref_node; | |
7063 | struct llist_node *next = node->next; | |
7064 | ||
7065 | ref_node = llist_entry(node, struct fixed_file_ref_node, llist); | |
7066 | __io_file_put_work(ref_node); | |
7067 | node = next; | |
7068 | } | |
7069 | } | |
7070 | ||
05589553 | 7071 | static void io_file_data_ref_zero(struct percpu_ref *ref) |
2faf852d | 7072 | { |
05589553 | 7073 | struct fixed_file_ref_node *ref_node; |
4a38aed2 JA |
7074 | struct io_ring_ctx *ctx; |
7075 | bool first_add; | |
7076 | int delay = HZ; | |
65e19f54 | 7077 | |
05589553 | 7078 | ref_node = container_of(ref, struct fixed_file_ref_node, refs); |
4a38aed2 | 7079 | ctx = ref_node->file_data->ctx; |
05589553 | 7080 | |
4a38aed2 JA |
7081 | if (percpu_ref_is_dying(&ctx->file_data->refs)) |
7082 | delay = 0; | |
05589553 | 7083 | |
4a38aed2 JA |
7084 | first_add = llist_add(&ref_node->llist, &ctx->file_put_llist); |
7085 | if (!delay) | |
7086 | mod_delayed_work(system_wq, &ctx->file_put_work, 0); | |
7087 | else if (first_add) | |
7088 | queue_delayed_work(system_wq, &ctx->file_put_work, delay); | |
05f3fb3c | 7089 | } |
65e19f54 | 7090 | |
05589553 XW |
7091 | static struct fixed_file_ref_node *alloc_fixed_file_ref_node( |
7092 | struct io_ring_ctx *ctx) | |
05f3fb3c | 7093 | { |
05589553 | 7094 | struct fixed_file_ref_node *ref_node; |
05f3fb3c | 7095 | |
05589553 XW |
7096 | ref_node = kzalloc(sizeof(*ref_node), GFP_KERNEL); |
7097 | if (!ref_node) | |
7098 | return ERR_PTR(-ENOMEM); | |
05f3fb3c | 7099 | |
05589553 XW |
7100 | if (percpu_ref_init(&ref_node->refs, io_file_data_ref_zero, |
7101 | 0, GFP_KERNEL)) { | |
7102 | kfree(ref_node); | |
7103 | return ERR_PTR(-ENOMEM); | |
7104 | } | |
7105 | INIT_LIST_HEAD(&ref_node->node); | |
7106 | INIT_LIST_HEAD(&ref_node->file_list); | |
05589553 XW |
7107 | ref_node->file_data = ctx->file_data; |
7108 | return ref_node; | |
05589553 XW |
7109 | } |
7110 | ||
7111 | static void destroy_fixed_file_ref_node(struct fixed_file_ref_node *ref_node) | |
7112 | { | |
7113 | percpu_ref_exit(&ref_node->refs); | |
7114 | kfree(ref_node); | |
65e19f54 JA |
7115 | } |
7116 | ||
6b06314c JA |
7117 | static int io_sqe_files_register(struct io_ring_ctx *ctx, void __user *arg, |
7118 | unsigned nr_args) | |
7119 | { | |
7120 | __s32 __user *fds = (__s32 __user *) arg; | |
65e19f54 | 7121 | unsigned nr_tables; |
05f3fb3c | 7122 | struct file *file; |
6b06314c JA |
7123 | int fd, ret = 0; |
7124 | unsigned i; | |
05589553 | 7125 | struct fixed_file_ref_node *ref_node; |
6b06314c | 7126 | |
05f3fb3c | 7127 | if (ctx->file_data) |
6b06314c JA |
7128 | return -EBUSY; |
7129 | if (!nr_args) | |
7130 | return -EINVAL; | |
7131 | if (nr_args > IORING_MAX_FIXED_FILES) | |
7132 | return -EMFILE; | |
7133 | ||
05f3fb3c JA |
7134 | ctx->file_data = kzalloc(sizeof(*ctx->file_data), GFP_KERNEL); |
7135 | if (!ctx->file_data) | |
7136 | return -ENOMEM; | |
7137 | ctx->file_data->ctx = ctx; | |
7138 | init_completion(&ctx->file_data->done); | |
05589553 | 7139 | INIT_LIST_HEAD(&ctx->file_data->ref_list); |
f7fe9346 | 7140 | spin_lock_init(&ctx->file_data->lock); |
05f3fb3c | 7141 | |
65e19f54 | 7142 | nr_tables = DIV_ROUND_UP(nr_args, IORING_MAX_FILES_TABLE); |
05f3fb3c JA |
7143 | ctx->file_data->table = kcalloc(nr_tables, |
7144 | sizeof(struct fixed_file_table), | |
65e19f54 | 7145 | GFP_KERNEL); |
05f3fb3c JA |
7146 | if (!ctx->file_data->table) { |
7147 | kfree(ctx->file_data); | |
7148 | ctx->file_data = NULL; | |
6b06314c | 7149 | return -ENOMEM; |
05f3fb3c JA |
7150 | } |
7151 | ||
05589553 | 7152 | if (percpu_ref_init(&ctx->file_data->refs, io_file_ref_kill, |
05f3fb3c JA |
7153 | PERCPU_REF_ALLOW_REINIT, GFP_KERNEL)) { |
7154 | kfree(ctx->file_data->table); | |
7155 | kfree(ctx->file_data); | |
7156 | ctx->file_data = NULL; | |
6b06314c | 7157 | return -ENOMEM; |
05f3fb3c | 7158 | } |
6b06314c | 7159 | |
65e19f54 | 7160 | if (io_sqe_alloc_file_tables(ctx, nr_tables, nr_args)) { |
05f3fb3c JA |
7161 | percpu_ref_exit(&ctx->file_data->refs); |
7162 | kfree(ctx->file_data->table); | |
7163 | kfree(ctx->file_data); | |
7164 | ctx->file_data = NULL; | |
65e19f54 JA |
7165 | return -ENOMEM; |
7166 | } | |
7167 | ||
08a45173 | 7168 | for (i = 0; i < nr_args; i++, ctx->nr_user_files++) { |
65e19f54 JA |
7169 | struct fixed_file_table *table; |
7170 | unsigned index; | |
7171 | ||
6b06314c JA |
7172 | ret = -EFAULT; |
7173 | if (copy_from_user(&fd, &fds[i], sizeof(fd))) | |
7174 | break; | |
08a45173 JA |
7175 | /* allow sparse sets */ |
7176 | if (fd == -1) { | |
7177 | ret = 0; | |
7178 | continue; | |
7179 | } | |
6b06314c | 7180 | |
05f3fb3c | 7181 | table = &ctx->file_data->table[i >> IORING_FILE_TABLE_SHIFT]; |
65e19f54 | 7182 | index = i & IORING_FILE_TABLE_MASK; |
05f3fb3c | 7183 | file = fget(fd); |
6b06314c JA |
7184 | |
7185 | ret = -EBADF; | |
05f3fb3c | 7186 | if (!file) |
6b06314c | 7187 | break; |
05f3fb3c | 7188 | |
6b06314c JA |
7189 | /* |
7190 | * Don't allow io_uring instances to be registered. If UNIX | |
7191 | * isn't enabled, then this causes a reference cycle and this | |
7192 | * instance can never get freed. If UNIX is enabled we'll | |
7193 | * handle it just fine, but there's still no point in allowing | |
7194 | * a ring fd as it doesn't support regular read/write anyway. | |
7195 | */ | |
05f3fb3c JA |
7196 | if (file->f_op == &io_uring_fops) { |
7197 | fput(file); | |
6b06314c JA |
7198 | break; |
7199 | } | |
6b06314c | 7200 | ret = 0; |
05f3fb3c | 7201 | table->files[index] = file; |
6b06314c JA |
7202 | } |
7203 | ||
7204 | if (ret) { | |
65e19f54 | 7205 | for (i = 0; i < ctx->nr_user_files; i++) { |
65e19f54 JA |
7206 | file = io_file_from_index(ctx, i); |
7207 | if (file) | |
7208 | fput(file); | |
7209 | } | |
7210 | for (i = 0; i < nr_tables; i++) | |
05f3fb3c | 7211 | kfree(ctx->file_data->table[i].files); |
6b06314c | 7212 | |
667e57da | 7213 | percpu_ref_exit(&ctx->file_data->refs); |
05f3fb3c JA |
7214 | kfree(ctx->file_data->table); |
7215 | kfree(ctx->file_data); | |
7216 | ctx->file_data = NULL; | |
6b06314c JA |
7217 | ctx->nr_user_files = 0; |
7218 | return ret; | |
7219 | } | |
7220 | ||
7221 | ret = io_sqe_files_scm(ctx); | |
05589553 | 7222 | if (ret) { |
6b06314c | 7223 | io_sqe_files_unregister(ctx); |
05589553 XW |
7224 | return ret; |
7225 | } | |
6b06314c | 7226 | |
05589553 XW |
7227 | ref_node = alloc_fixed_file_ref_node(ctx); |
7228 | if (IS_ERR(ref_node)) { | |
7229 | io_sqe_files_unregister(ctx); | |
7230 | return PTR_ERR(ref_node); | |
7231 | } | |
7232 | ||
7233 | ctx->file_data->cur_refs = &ref_node->refs; | |
6a4d07cd | 7234 | spin_lock(&ctx->file_data->lock); |
05589553 | 7235 | list_add(&ref_node->node, &ctx->file_data->ref_list); |
6a4d07cd | 7236 | spin_unlock(&ctx->file_data->lock); |
05589553 | 7237 | percpu_ref_get(&ctx->file_data->refs); |
6b06314c JA |
7238 | return ret; |
7239 | } | |
7240 | ||
c3a31e60 JA |
7241 | static int io_sqe_file_register(struct io_ring_ctx *ctx, struct file *file, |
7242 | int index) | |
7243 | { | |
7244 | #if defined(CONFIG_UNIX) | |
7245 | struct sock *sock = ctx->ring_sock->sk; | |
7246 | struct sk_buff_head *head = &sock->sk_receive_queue; | |
7247 | struct sk_buff *skb; | |
7248 | ||
7249 | /* | |
7250 | * See if we can merge this file into an existing skb SCM_RIGHTS | |
7251 | * file set. If there's no room, fall back to allocating a new skb | |
7252 | * and filling it in. | |
7253 | */ | |
7254 | spin_lock_irq(&head->lock); | |
7255 | skb = skb_peek(head); | |
7256 | if (skb) { | |
7257 | struct scm_fp_list *fpl = UNIXCB(skb).fp; | |
7258 | ||
7259 | if (fpl->count < SCM_MAX_FD) { | |
7260 | __skb_unlink(skb, head); | |
7261 | spin_unlock_irq(&head->lock); | |
7262 | fpl->fp[fpl->count] = get_file(file); | |
7263 | unix_inflight(fpl->user, fpl->fp[fpl->count]); | |
7264 | fpl->count++; | |
7265 | spin_lock_irq(&head->lock); | |
7266 | __skb_queue_head(head, skb); | |
7267 | } else { | |
7268 | skb = NULL; | |
7269 | } | |
7270 | } | |
7271 | spin_unlock_irq(&head->lock); | |
7272 | ||
7273 | if (skb) { | |
7274 | fput(file); | |
7275 | return 0; | |
7276 | } | |
7277 | ||
7278 | return __io_sqe_files_scm(ctx, 1, index); | |
7279 | #else | |
7280 | return 0; | |
7281 | #endif | |
7282 | } | |
7283 | ||
a5318d3c | 7284 | static int io_queue_file_removal(struct fixed_file_data *data, |
05589553 | 7285 | struct file *file) |
05f3fb3c | 7286 | { |
a5318d3c | 7287 | struct io_file_put *pfile; |
05589553 XW |
7288 | struct percpu_ref *refs = data->cur_refs; |
7289 | struct fixed_file_ref_node *ref_node; | |
05f3fb3c | 7290 | |
05f3fb3c | 7291 | pfile = kzalloc(sizeof(*pfile), GFP_KERNEL); |
a5318d3c HD |
7292 | if (!pfile) |
7293 | return -ENOMEM; | |
05f3fb3c | 7294 | |
05589553 | 7295 | ref_node = container_of(refs, struct fixed_file_ref_node, refs); |
05f3fb3c | 7296 | pfile->file = file; |
05589553 XW |
7297 | list_add(&pfile->list, &ref_node->file_list); |
7298 | ||
a5318d3c | 7299 | return 0; |
05f3fb3c JA |
7300 | } |
7301 | ||
7302 | static int __io_sqe_files_update(struct io_ring_ctx *ctx, | |
7303 | struct io_uring_files_update *up, | |
7304 | unsigned nr_args) | |
7305 | { | |
7306 | struct fixed_file_data *data = ctx->file_data; | |
05589553 | 7307 | struct fixed_file_ref_node *ref_node; |
05f3fb3c | 7308 | struct file *file; |
c3a31e60 JA |
7309 | __s32 __user *fds; |
7310 | int fd, i, err; | |
7311 | __u32 done; | |
05589553 | 7312 | bool needs_switch = false; |
c3a31e60 | 7313 | |
05f3fb3c | 7314 | if (check_add_overflow(up->offset, nr_args, &done)) |
c3a31e60 JA |
7315 | return -EOVERFLOW; |
7316 | if (done > ctx->nr_user_files) | |
7317 | return -EINVAL; | |
7318 | ||
05589553 XW |
7319 | ref_node = alloc_fixed_file_ref_node(ctx); |
7320 | if (IS_ERR(ref_node)) | |
7321 | return PTR_ERR(ref_node); | |
7322 | ||
c3a31e60 | 7323 | done = 0; |
05f3fb3c | 7324 | fds = u64_to_user_ptr(up->fds); |
c3a31e60 | 7325 | while (nr_args) { |
65e19f54 JA |
7326 | struct fixed_file_table *table; |
7327 | unsigned index; | |
7328 | ||
c3a31e60 JA |
7329 | err = 0; |
7330 | if (copy_from_user(&fd, &fds[done], sizeof(fd))) { | |
7331 | err = -EFAULT; | |
7332 | break; | |
7333 | } | |
05f3fb3c JA |
7334 | i = array_index_nospec(up->offset, ctx->nr_user_files); |
7335 | table = &ctx->file_data->table[i >> IORING_FILE_TABLE_SHIFT]; | |
65e19f54 JA |
7336 | index = i & IORING_FILE_TABLE_MASK; |
7337 | if (table->files[index]) { | |
98dfd502 | 7338 | file = table->files[index]; |
a5318d3c HD |
7339 | err = io_queue_file_removal(data, file); |
7340 | if (err) | |
7341 | break; | |
65e19f54 | 7342 | table->files[index] = NULL; |
05589553 | 7343 | needs_switch = true; |
c3a31e60 JA |
7344 | } |
7345 | if (fd != -1) { | |
c3a31e60 JA |
7346 | file = fget(fd); |
7347 | if (!file) { | |
7348 | err = -EBADF; | |
7349 | break; | |
7350 | } | |
7351 | /* | |
7352 | * Don't allow io_uring instances to be registered. If | |
7353 | * UNIX isn't enabled, then this causes a reference | |
7354 | * cycle and this instance can never get freed. If UNIX | |
7355 | * is enabled we'll handle it just fine, but there's | |
7356 | * still no point in allowing a ring fd as it doesn't | |
7357 | * support regular read/write anyway. | |
7358 | */ | |
7359 | if (file->f_op == &io_uring_fops) { | |
7360 | fput(file); | |
7361 | err = -EBADF; | |
7362 | break; | |
7363 | } | |
65e19f54 | 7364 | table->files[index] = file; |
c3a31e60 | 7365 | err = io_sqe_file_register(ctx, file, i); |
f3bd9dae | 7366 | if (err) { |
95d1c8e5 | 7367 | table->files[index] = NULL; |
f3bd9dae | 7368 | fput(file); |
c3a31e60 | 7369 | break; |
f3bd9dae | 7370 | } |
c3a31e60 JA |
7371 | } |
7372 | nr_args--; | |
7373 | done++; | |
05f3fb3c JA |
7374 | up->offset++; |
7375 | } | |
7376 | ||
05589553 XW |
7377 | if (needs_switch) { |
7378 | percpu_ref_kill(data->cur_refs); | |
6a4d07cd | 7379 | spin_lock(&data->lock); |
05589553 XW |
7380 | list_add(&ref_node->node, &data->ref_list); |
7381 | data->cur_refs = &ref_node->refs; | |
6a4d07cd | 7382 | spin_unlock(&data->lock); |
05589553 XW |
7383 | percpu_ref_get(&ctx->file_data->refs); |
7384 | } else | |
7385 | destroy_fixed_file_ref_node(ref_node); | |
c3a31e60 JA |
7386 | |
7387 | return done ? done : err; | |
7388 | } | |
05589553 | 7389 | |
05f3fb3c JA |
7390 | static int io_sqe_files_update(struct io_ring_ctx *ctx, void __user *arg, |
7391 | unsigned nr_args) | |
7392 | { | |
7393 | struct io_uring_files_update up; | |
7394 | ||
7395 | if (!ctx->file_data) | |
7396 | return -ENXIO; | |
7397 | if (!nr_args) | |
7398 | return -EINVAL; | |
7399 | if (copy_from_user(&up, arg, sizeof(up))) | |
7400 | return -EFAULT; | |
7401 | if (up.resv) | |
7402 | return -EINVAL; | |
7403 | ||
7404 | return __io_sqe_files_update(ctx, &up, nr_args); | |
7405 | } | |
c3a31e60 | 7406 | |
e9fd9396 | 7407 | static void io_free_work(struct io_wq_work *work) |
7d723065 JA |
7408 | { |
7409 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); | |
7410 | ||
e9fd9396 | 7411 | /* Consider that io_steal_work() relies on this ref */ |
7d723065 JA |
7412 | io_put_req(req); |
7413 | } | |
7414 | ||
24369c2e PB |
7415 | static int io_init_wq_offload(struct io_ring_ctx *ctx, |
7416 | struct io_uring_params *p) | |
7417 | { | |
7418 | struct io_wq_data data; | |
7419 | struct fd f; | |
7420 | struct io_ring_ctx *ctx_attach; | |
7421 | unsigned int concurrency; | |
7422 | int ret = 0; | |
7423 | ||
7424 | data.user = ctx->user; | |
e9fd9396 | 7425 | data.free_work = io_free_work; |
f5fa38c5 | 7426 | data.do_work = io_wq_submit_work; |
24369c2e PB |
7427 | |
7428 | if (!(p->flags & IORING_SETUP_ATTACH_WQ)) { | |
7429 | /* Do QD, or 4 * CPUS, whatever is smallest */ | |
7430 | concurrency = min(ctx->sq_entries, 4 * num_online_cpus()); | |
7431 | ||
7432 | ctx->io_wq = io_wq_create(concurrency, &data); | |
7433 | if (IS_ERR(ctx->io_wq)) { | |
7434 | ret = PTR_ERR(ctx->io_wq); | |
7435 | ctx->io_wq = NULL; | |
7436 | } | |
7437 | return ret; | |
7438 | } | |
7439 | ||
7440 | f = fdget(p->wq_fd); | |
7441 | if (!f.file) | |
7442 | return -EBADF; | |
7443 | ||
7444 | if (f.file->f_op != &io_uring_fops) { | |
7445 | ret = -EINVAL; | |
7446 | goto out_fput; | |
7447 | } | |
7448 | ||
7449 | ctx_attach = f.file->private_data; | |
7450 | /* @io_wq is protected by holding the fd */ | |
7451 | if (!io_wq_get(ctx_attach->io_wq, &data)) { | |
7452 | ret = -EINVAL; | |
7453 | goto out_fput; | |
7454 | } | |
7455 | ||
7456 | ctx->io_wq = ctx_attach->io_wq; | |
7457 | out_fput: | |
7458 | fdput(f); | |
7459 | return ret; | |
7460 | } | |
7461 | ||
6c271ce2 JA |
7462 | static int io_sq_offload_start(struct io_ring_ctx *ctx, |
7463 | struct io_uring_params *p) | |
2b188cc1 JA |
7464 | { |
7465 | int ret; | |
7466 | ||
6c271ce2 | 7467 | if (ctx->flags & IORING_SETUP_SQPOLL) { |
3ec482d1 JA |
7468 | ret = -EPERM; |
7469 | if (!capable(CAP_SYS_ADMIN)) | |
7470 | goto err; | |
7471 | ||
917257da JA |
7472 | ctx->sq_thread_idle = msecs_to_jiffies(p->sq_thread_idle); |
7473 | if (!ctx->sq_thread_idle) | |
7474 | ctx->sq_thread_idle = HZ; | |
7475 | ||
6c271ce2 | 7476 | if (p->flags & IORING_SETUP_SQ_AFF) { |
44a9bd18 | 7477 | int cpu = p->sq_thread_cpu; |
6c271ce2 | 7478 | |
917257da | 7479 | ret = -EINVAL; |
44a9bd18 JA |
7480 | if (cpu >= nr_cpu_ids) |
7481 | goto err; | |
7889f44d | 7482 | if (!cpu_online(cpu)) |
917257da JA |
7483 | goto err; |
7484 | ||
6c271ce2 JA |
7485 | ctx->sqo_thread = kthread_create_on_cpu(io_sq_thread, |
7486 | ctx, cpu, | |
7487 | "io_uring-sq"); | |
7488 | } else { | |
7489 | ctx->sqo_thread = kthread_create(io_sq_thread, ctx, | |
7490 | "io_uring-sq"); | |
7491 | } | |
7492 | if (IS_ERR(ctx->sqo_thread)) { | |
7493 | ret = PTR_ERR(ctx->sqo_thread); | |
7494 | ctx->sqo_thread = NULL; | |
7495 | goto err; | |
7496 | } | |
7497 | wake_up_process(ctx->sqo_thread); | |
7498 | } else if (p->flags & IORING_SETUP_SQ_AFF) { | |
7499 | /* Can't have SQ_AFF without SQPOLL */ | |
7500 | ret = -EINVAL; | |
7501 | goto err; | |
7502 | } | |
7503 | ||
24369c2e PB |
7504 | ret = io_init_wq_offload(ctx, p); |
7505 | if (ret) | |
2b188cc1 | 7506 | goto err; |
2b188cc1 JA |
7507 | |
7508 | return 0; | |
7509 | err: | |
54a91f3b | 7510 | io_finish_async(ctx); |
2b188cc1 JA |
7511 | return ret; |
7512 | } | |
7513 | ||
a087e2b5 BM |
7514 | static inline void __io_unaccount_mem(struct user_struct *user, |
7515 | unsigned long nr_pages) | |
2b188cc1 JA |
7516 | { |
7517 | atomic_long_sub(nr_pages, &user->locked_vm); | |
7518 | } | |
7519 | ||
a087e2b5 BM |
7520 | static inline int __io_account_mem(struct user_struct *user, |
7521 | unsigned long nr_pages) | |
2b188cc1 JA |
7522 | { |
7523 | unsigned long page_limit, cur_pages, new_pages; | |
7524 | ||
7525 | /* Don't allow more pages than we can safely lock */ | |
7526 | page_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT; | |
7527 | ||
7528 | do { | |
7529 | cur_pages = atomic_long_read(&user->locked_vm); | |
7530 | new_pages = cur_pages + nr_pages; | |
7531 | if (new_pages > page_limit) | |
7532 | return -ENOMEM; | |
7533 | } while (atomic_long_cmpxchg(&user->locked_vm, cur_pages, | |
7534 | new_pages) != cur_pages); | |
7535 | ||
7536 | return 0; | |
7537 | } | |
7538 | ||
2e0464d4 BM |
7539 | static void io_unaccount_mem(struct io_ring_ctx *ctx, unsigned long nr_pages, |
7540 | enum io_mem_account acct) | |
a087e2b5 | 7541 | { |
aad5d8da | 7542 | if (ctx->limit_mem) |
a087e2b5 | 7543 | __io_unaccount_mem(ctx->user, nr_pages); |
30975825 | 7544 | |
2e0464d4 BM |
7545 | if (ctx->sqo_mm) { |
7546 | if (acct == ACCT_LOCKED) | |
7547 | ctx->sqo_mm->locked_vm -= nr_pages; | |
7548 | else if (acct == ACCT_PINNED) | |
7549 | atomic64_sub(nr_pages, &ctx->sqo_mm->pinned_vm); | |
7550 | } | |
a087e2b5 BM |
7551 | } |
7552 | ||
2e0464d4 BM |
7553 | static int io_account_mem(struct io_ring_ctx *ctx, unsigned long nr_pages, |
7554 | enum io_mem_account acct) | |
a087e2b5 | 7555 | { |
30975825 BM |
7556 | int ret; |
7557 | ||
7558 | if (ctx->limit_mem) { | |
7559 | ret = __io_account_mem(ctx->user, nr_pages); | |
7560 | if (ret) | |
7561 | return ret; | |
7562 | } | |
7563 | ||
2e0464d4 BM |
7564 | if (ctx->sqo_mm) { |
7565 | if (acct == ACCT_LOCKED) | |
7566 | ctx->sqo_mm->locked_vm += nr_pages; | |
7567 | else if (acct == ACCT_PINNED) | |
7568 | atomic64_add(nr_pages, &ctx->sqo_mm->pinned_vm); | |
7569 | } | |
a087e2b5 BM |
7570 | |
7571 | return 0; | |
7572 | } | |
7573 | ||
2b188cc1 JA |
7574 | static void io_mem_free(void *ptr) |
7575 | { | |
52e04ef4 MR |
7576 | struct page *page; |
7577 | ||
7578 | if (!ptr) | |
7579 | return; | |
2b188cc1 | 7580 | |
52e04ef4 | 7581 | page = virt_to_head_page(ptr); |
2b188cc1 JA |
7582 | if (put_page_testzero(page)) |
7583 | free_compound_page(page); | |
7584 | } | |
7585 | ||
7586 | static void *io_mem_alloc(size_t size) | |
7587 | { | |
7588 | gfp_t gfp_flags = GFP_KERNEL | __GFP_ZERO | __GFP_NOWARN | __GFP_COMP | | |
7589 | __GFP_NORETRY; | |
7590 | ||
7591 | return (void *) __get_free_pages(gfp_flags, get_order(size)); | |
7592 | } | |
7593 | ||
75b28aff HV |
7594 | static unsigned long rings_size(unsigned sq_entries, unsigned cq_entries, |
7595 | size_t *sq_offset) | |
7596 | { | |
7597 | struct io_rings *rings; | |
7598 | size_t off, sq_array_size; | |
7599 | ||
7600 | off = struct_size(rings, cqes, cq_entries); | |
7601 | if (off == SIZE_MAX) | |
7602 | return SIZE_MAX; | |
7603 | ||
7604 | #ifdef CONFIG_SMP | |
7605 | off = ALIGN(off, SMP_CACHE_BYTES); | |
7606 | if (off == 0) | |
7607 | return SIZE_MAX; | |
7608 | #endif | |
7609 | ||
b36200f5 DV |
7610 | if (sq_offset) |
7611 | *sq_offset = off; | |
7612 | ||
75b28aff HV |
7613 | sq_array_size = array_size(sizeof(u32), sq_entries); |
7614 | if (sq_array_size == SIZE_MAX) | |
7615 | return SIZE_MAX; | |
7616 | ||
7617 | if (check_add_overflow(off, sq_array_size, &off)) | |
7618 | return SIZE_MAX; | |
7619 | ||
75b28aff HV |
7620 | return off; |
7621 | } | |
7622 | ||
2b188cc1 JA |
7623 | static unsigned long ring_pages(unsigned sq_entries, unsigned cq_entries) |
7624 | { | |
75b28aff | 7625 | size_t pages; |
2b188cc1 | 7626 | |
75b28aff HV |
7627 | pages = (size_t)1 << get_order( |
7628 | rings_size(sq_entries, cq_entries, NULL)); | |
7629 | pages += (size_t)1 << get_order( | |
7630 | array_size(sizeof(struct io_uring_sqe), sq_entries)); | |
2b188cc1 | 7631 | |
75b28aff | 7632 | return pages; |
2b188cc1 JA |
7633 | } |
7634 | ||
edafccee JA |
7635 | static int io_sqe_buffer_unregister(struct io_ring_ctx *ctx) |
7636 | { | |
7637 | int i, j; | |
7638 | ||
7639 | if (!ctx->user_bufs) | |
7640 | return -ENXIO; | |
7641 | ||
7642 | for (i = 0; i < ctx->nr_user_bufs; i++) { | |
7643 | struct io_mapped_ubuf *imu = &ctx->user_bufs[i]; | |
7644 | ||
7645 | for (j = 0; j < imu->nr_bvecs; j++) | |
f1f6a7dd | 7646 | unpin_user_page(imu->bvec[j].bv_page); |
edafccee | 7647 | |
2e0464d4 | 7648 | io_unaccount_mem(ctx, imu->nr_bvecs, ACCT_PINNED); |
d4ef6475 | 7649 | kvfree(imu->bvec); |
edafccee JA |
7650 | imu->nr_bvecs = 0; |
7651 | } | |
7652 | ||
7653 | kfree(ctx->user_bufs); | |
7654 | ctx->user_bufs = NULL; | |
7655 | ctx->nr_user_bufs = 0; | |
7656 | return 0; | |
7657 | } | |
7658 | ||
7659 | static int io_copy_iov(struct io_ring_ctx *ctx, struct iovec *dst, | |
7660 | void __user *arg, unsigned index) | |
7661 | { | |
7662 | struct iovec __user *src; | |
7663 | ||
7664 | #ifdef CONFIG_COMPAT | |
7665 | if (ctx->compat) { | |
7666 | struct compat_iovec __user *ciovs; | |
7667 | struct compat_iovec ciov; | |
7668 | ||
7669 | ciovs = (struct compat_iovec __user *) arg; | |
7670 | if (copy_from_user(&ciov, &ciovs[index], sizeof(ciov))) | |
7671 | return -EFAULT; | |
7672 | ||
d55e5f5b | 7673 | dst->iov_base = u64_to_user_ptr((u64)ciov.iov_base); |
edafccee JA |
7674 | dst->iov_len = ciov.iov_len; |
7675 | return 0; | |
7676 | } | |
7677 | #endif | |
7678 | src = (struct iovec __user *) arg; | |
7679 | if (copy_from_user(dst, &src[index], sizeof(*dst))) | |
7680 | return -EFAULT; | |
7681 | return 0; | |
7682 | } | |
7683 | ||
7684 | static int io_sqe_buffer_register(struct io_ring_ctx *ctx, void __user *arg, | |
7685 | unsigned nr_args) | |
7686 | { | |
7687 | struct vm_area_struct **vmas = NULL; | |
7688 | struct page **pages = NULL; | |
7689 | int i, j, got_pages = 0; | |
7690 | int ret = -EINVAL; | |
7691 | ||
7692 | if (ctx->user_bufs) | |
7693 | return -EBUSY; | |
7694 | if (!nr_args || nr_args > UIO_MAXIOV) | |
7695 | return -EINVAL; | |
7696 | ||
7697 | ctx->user_bufs = kcalloc(nr_args, sizeof(struct io_mapped_ubuf), | |
7698 | GFP_KERNEL); | |
7699 | if (!ctx->user_bufs) | |
7700 | return -ENOMEM; | |
7701 | ||
7702 | for (i = 0; i < nr_args; i++) { | |
7703 | struct io_mapped_ubuf *imu = &ctx->user_bufs[i]; | |
7704 | unsigned long off, start, end, ubuf; | |
7705 | int pret, nr_pages; | |
7706 | struct iovec iov; | |
7707 | size_t size; | |
7708 | ||
7709 | ret = io_copy_iov(ctx, &iov, arg, i); | |
7710 | if (ret) | |
a278682d | 7711 | goto err; |
edafccee JA |
7712 | |
7713 | /* | |
7714 | * Don't impose further limits on the size and buffer | |
7715 | * constraints here, we'll -EINVAL later when IO is | |
7716 | * submitted if they are wrong. | |
7717 | */ | |
7718 | ret = -EFAULT; | |
7719 | if (!iov.iov_base || !iov.iov_len) | |
7720 | goto err; | |
7721 | ||
7722 | /* arbitrary limit, but we need something */ | |
7723 | if (iov.iov_len > SZ_1G) | |
7724 | goto err; | |
7725 | ||
7726 | ubuf = (unsigned long) iov.iov_base; | |
7727 | end = (ubuf + iov.iov_len + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
7728 | start = ubuf >> PAGE_SHIFT; | |
7729 | nr_pages = end - start; | |
7730 | ||
2e0464d4 | 7731 | ret = io_account_mem(ctx, nr_pages, ACCT_PINNED); |
a087e2b5 BM |
7732 | if (ret) |
7733 | goto err; | |
edafccee JA |
7734 | |
7735 | ret = 0; | |
7736 | if (!pages || nr_pages > got_pages) { | |
a8c73c1a DE |
7737 | kvfree(vmas); |
7738 | kvfree(pages); | |
d4ef6475 | 7739 | pages = kvmalloc_array(nr_pages, sizeof(struct page *), |
edafccee | 7740 | GFP_KERNEL); |
d4ef6475 | 7741 | vmas = kvmalloc_array(nr_pages, |
edafccee JA |
7742 | sizeof(struct vm_area_struct *), |
7743 | GFP_KERNEL); | |
7744 | if (!pages || !vmas) { | |
7745 | ret = -ENOMEM; | |
2e0464d4 | 7746 | io_unaccount_mem(ctx, nr_pages, ACCT_PINNED); |
edafccee JA |
7747 | goto err; |
7748 | } | |
7749 | got_pages = nr_pages; | |
7750 | } | |
7751 | ||
d4ef6475 | 7752 | imu->bvec = kvmalloc_array(nr_pages, sizeof(struct bio_vec), |
edafccee JA |
7753 | GFP_KERNEL); |
7754 | ret = -ENOMEM; | |
7755 | if (!imu->bvec) { | |
2e0464d4 | 7756 | io_unaccount_mem(ctx, nr_pages, ACCT_PINNED); |
edafccee JA |
7757 | goto err; |
7758 | } | |
7759 | ||
7760 | ret = 0; | |
d8ed45c5 | 7761 | mmap_read_lock(current->mm); |
2113b05d | 7762 | pret = pin_user_pages(ubuf, nr_pages, |
932f4a63 IW |
7763 | FOLL_WRITE | FOLL_LONGTERM, |
7764 | pages, vmas); | |
edafccee JA |
7765 | if (pret == nr_pages) { |
7766 | /* don't support file backed memory */ | |
7767 | for (j = 0; j < nr_pages; j++) { | |
7768 | struct vm_area_struct *vma = vmas[j]; | |
7769 | ||
7770 | if (vma->vm_file && | |
7771 | !is_file_hugepages(vma->vm_file)) { | |
7772 | ret = -EOPNOTSUPP; | |
7773 | break; | |
7774 | } | |
7775 | } | |
7776 | } else { | |
7777 | ret = pret < 0 ? pret : -EFAULT; | |
7778 | } | |
d8ed45c5 | 7779 | mmap_read_unlock(current->mm); |
edafccee JA |
7780 | if (ret) { |
7781 | /* | |
7782 | * if we did partial map, or found file backed vmas, | |
7783 | * release any pages we did get | |
7784 | */ | |
27c4d3a3 | 7785 | if (pret > 0) |
f1f6a7dd | 7786 | unpin_user_pages(pages, pret); |
2e0464d4 | 7787 | io_unaccount_mem(ctx, nr_pages, ACCT_PINNED); |
d4ef6475 | 7788 | kvfree(imu->bvec); |
edafccee JA |
7789 | goto err; |
7790 | } | |
7791 | ||
7792 | off = ubuf & ~PAGE_MASK; | |
7793 | size = iov.iov_len; | |
7794 | for (j = 0; j < nr_pages; j++) { | |
7795 | size_t vec_len; | |
7796 | ||
7797 | vec_len = min_t(size_t, size, PAGE_SIZE - off); | |
7798 | imu->bvec[j].bv_page = pages[j]; | |
7799 | imu->bvec[j].bv_len = vec_len; | |
7800 | imu->bvec[j].bv_offset = off; | |
7801 | off = 0; | |
7802 | size -= vec_len; | |
7803 | } | |
7804 | /* store original address for later verification */ | |
7805 | imu->ubuf = ubuf; | |
7806 | imu->len = iov.iov_len; | |
7807 | imu->nr_bvecs = nr_pages; | |
7808 | ||
7809 | ctx->nr_user_bufs++; | |
7810 | } | |
d4ef6475 MR |
7811 | kvfree(pages); |
7812 | kvfree(vmas); | |
edafccee JA |
7813 | return 0; |
7814 | err: | |
d4ef6475 MR |
7815 | kvfree(pages); |
7816 | kvfree(vmas); | |
edafccee JA |
7817 | io_sqe_buffer_unregister(ctx); |
7818 | return ret; | |
7819 | } | |
7820 | ||
9b402849 JA |
7821 | static int io_eventfd_register(struct io_ring_ctx *ctx, void __user *arg) |
7822 | { | |
7823 | __s32 __user *fds = arg; | |
7824 | int fd; | |
7825 | ||
7826 | if (ctx->cq_ev_fd) | |
7827 | return -EBUSY; | |
7828 | ||
7829 | if (copy_from_user(&fd, fds, sizeof(*fds))) | |
7830 | return -EFAULT; | |
7831 | ||
7832 | ctx->cq_ev_fd = eventfd_ctx_fdget(fd); | |
7833 | if (IS_ERR(ctx->cq_ev_fd)) { | |
7834 | int ret = PTR_ERR(ctx->cq_ev_fd); | |
7835 | ctx->cq_ev_fd = NULL; | |
7836 | return ret; | |
7837 | } | |
7838 | ||
7839 | return 0; | |
7840 | } | |
7841 | ||
7842 | static int io_eventfd_unregister(struct io_ring_ctx *ctx) | |
7843 | { | |
7844 | if (ctx->cq_ev_fd) { | |
7845 | eventfd_ctx_put(ctx->cq_ev_fd); | |
7846 | ctx->cq_ev_fd = NULL; | |
7847 | return 0; | |
7848 | } | |
7849 | ||
7850 | return -ENXIO; | |
7851 | } | |
7852 | ||
5a2e745d JA |
7853 | static int __io_destroy_buffers(int id, void *p, void *data) |
7854 | { | |
7855 | struct io_ring_ctx *ctx = data; | |
7856 | struct io_buffer *buf = p; | |
7857 | ||
067524e9 | 7858 | __io_remove_buffers(ctx, buf, id, -1U); |
5a2e745d JA |
7859 | return 0; |
7860 | } | |
7861 | ||
7862 | static void io_destroy_buffers(struct io_ring_ctx *ctx) | |
7863 | { | |
7864 | idr_for_each(&ctx->io_buffer_idr, __io_destroy_buffers, ctx); | |
7865 | idr_destroy(&ctx->io_buffer_idr); | |
7866 | } | |
7867 | ||
2b188cc1 JA |
7868 | static void io_ring_ctx_free(struct io_ring_ctx *ctx) |
7869 | { | |
6b06314c | 7870 | io_finish_async(ctx); |
5dbcad51 | 7871 | io_sqe_buffer_unregister(ctx); |
30975825 | 7872 | if (ctx->sqo_mm) { |
2b188cc1 | 7873 | mmdrop(ctx->sqo_mm); |
30975825 BM |
7874 | ctx->sqo_mm = NULL; |
7875 | } | |
def596e9 | 7876 | |
6b06314c | 7877 | io_sqe_files_unregister(ctx); |
9b402849 | 7878 | io_eventfd_unregister(ctx); |
5a2e745d | 7879 | io_destroy_buffers(ctx); |
41726c9a | 7880 | idr_destroy(&ctx->personality_idr); |
def596e9 | 7881 | |
2b188cc1 | 7882 | #if defined(CONFIG_UNIX) |
355e8d26 EB |
7883 | if (ctx->ring_sock) { |
7884 | ctx->ring_sock->file = NULL; /* so that iput() is called */ | |
2b188cc1 | 7885 | sock_release(ctx->ring_sock); |
355e8d26 | 7886 | } |
2b188cc1 JA |
7887 | #endif |
7888 | ||
75b28aff | 7889 | io_mem_free(ctx->rings); |
2b188cc1 | 7890 | io_mem_free(ctx->sq_sqes); |
2b188cc1 JA |
7891 | |
7892 | percpu_ref_exit(&ctx->refs); | |
2b188cc1 | 7893 | free_uid(ctx->user); |
181e448d | 7894 | put_cred(ctx->creds); |
78076bb6 | 7895 | kfree(ctx->cancel_hash); |
0ddf92e8 | 7896 | kmem_cache_free(req_cachep, ctx->fallback_req); |
2b188cc1 JA |
7897 | kfree(ctx); |
7898 | } | |
7899 | ||
7900 | static __poll_t io_uring_poll(struct file *file, poll_table *wait) | |
7901 | { | |
7902 | struct io_ring_ctx *ctx = file->private_data; | |
7903 | __poll_t mask = 0; | |
7904 | ||
7905 | poll_wait(file, &ctx->cq_wait, wait); | |
4f7067c3 SB |
7906 | /* |
7907 | * synchronizes with barrier from wq_has_sleeper call in | |
7908 | * io_commit_cqring | |
7909 | */ | |
2b188cc1 | 7910 | smp_rmb(); |
75b28aff HV |
7911 | if (READ_ONCE(ctx->rings->sq.tail) - ctx->cached_sq_head != |
7912 | ctx->rings->sq_ring_entries) | |
2b188cc1 | 7913 | mask |= EPOLLOUT | EPOLLWRNORM; |
63e5d81f | 7914 | if (io_cqring_events(ctx, false)) |
2b188cc1 JA |
7915 | mask |= EPOLLIN | EPOLLRDNORM; |
7916 | ||
7917 | return mask; | |
7918 | } | |
7919 | ||
7920 | static int io_uring_fasync(int fd, struct file *file, int on) | |
7921 | { | |
7922 | struct io_ring_ctx *ctx = file->private_data; | |
7923 | ||
7924 | return fasync_helper(fd, file, on, &ctx->cq_fasync); | |
7925 | } | |
7926 | ||
071698e1 JA |
7927 | static int io_remove_personalities(int id, void *p, void *data) |
7928 | { | |
7929 | struct io_ring_ctx *ctx = data; | |
7930 | const struct cred *cred; | |
7931 | ||
7932 | cred = idr_remove(&ctx->personality_idr, id); | |
7933 | if (cred) | |
7934 | put_cred(cred); | |
7935 | return 0; | |
7936 | } | |
7937 | ||
85faa7b8 JA |
7938 | static void io_ring_exit_work(struct work_struct *work) |
7939 | { | |
b2edc0a7 PB |
7940 | struct io_ring_ctx *ctx = container_of(work, struct io_ring_ctx, |
7941 | exit_work); | |
85faa7b8 | 7942 | |
56952e91 JA |
7943 | /* |
7944 | * If we're doing polled IO and end up having requests being | |
7945 | * submitted async (out-of-line), then completions can come in while | |
7946 | * we're waiting for refs to drop. We need to reap these manually, | |
7947 | * as nobody else will be looking for them. | |
7948 | */ | |
b2edc0a7 | 7949 | do { |
56952e91 JA |
7950 | if (ctx->rings) |
7951 | io_cqring_overflow_flush(ctx, true); | |
b2edc0a7 PB |
7952 | io_iopoll_try_reap_events(ctx); |
7953 | } while (!wait_for_completion_timeout(&ctx->ref_comp, HZ/20)); | |
85faa7b8 JA |
7954 | io_ring_ctx_free(ctx); |
7955 | } | |
7956 | ||
2b188cc1 JA |
7957 | static void io_ring_ctx_wait_and_kill(struct io_ring_ctx *ctx) |
7958 | { | |
7959 | mutex_lock(&ctx->uring_lock); | |
7960 | percpu_ref_kill(&ctx->refs); | |
7961 | mutex_unlock(&ctx->uring_lock); | |
7962 | ||
5262f567 | 7963 | io_kill_timeouts(ctx); |
221c5eb2 | 7964 | io_poll_remove_all(ctx); |
561fb04a JA |
7965 | |
7966 | if (ctx->io_wq) | |
7967 | io_wq_cancel_all(ctx->io_wq); | |
7968 | ||
15dff286 JA |
7969 | /* if we failed setting up the ctx, we might not have any rings */ |
7970 | if (ctx->rings) | |
7971 | io_cqring_overflow_flush(ctx, true); | |
b2edc0a7 | 7972 | io_iopoll_try_reap_events(ctx); |
071698e1 | 7973 | idr_for_each(&ctx->personality_idr, io_remove_personalities, ctx); |
309fc03a JA |
7974 | |
7975 | /* | |
7976 | * Do this upfront, so we won't have a grace period where the ring | |
7977 | * is closed but resources aren't reaped yet. This can cause | |
7978 | * spurious failure in setting up a new ring. | |
7979 | */ | |
760618f7 JA |
7980 | io_unaccount_mem(ctx, ring_pages(ctx->sq_entries, ctx->cq_entries), |
7981 | ACCT_LOCKED); | |
309fc03a | 7982 | |
85faa7b8 | 7983 | INIT_WORK(&ctx->exit_work, io_ring_exit_work); |
fc666777 JA |
7984 | /* |
7985 | * Use system_unbound_wq to avoid spawning tons of event kworkers | |
7986 | * if we're exiting a ton of rings at the same time. It just adds | |
7987 | * noise and overhead, there's no discernable change in runtime | |
7988 | * over using system_wq. | |
7989 | */ | |
7990 | queue_work(system_unbound_wq, &ctx->exit_work); | |
2b188cc1 JA |
7991 | } |
7992 | ||
7993 | static int io_uring_release(struct inode *inode, struct file *file) | |
7994 | { | |
7995 | struct io_ring_ctx *ctx = file->private_data; | |
7996 | ||
7997 | file->private_data = NULL; | |
7998 | io_ring_ctx_wait_and_kill(ctx); | |
7999 | return 0; | |
8000 | } | |
8001 | ||
67c4d9e6 PB |
8002 | static bool io_wq_files_match(struct io_wq_work *work, void *data) |
8003 | { | |
8004 | struct files_struct *files = data; | |
8005 | ||
8006 | return work->files == files; | |
8007 | } | |
8008 | ||
f254ac04 JA |
8009 | /* |
8010 | * Returns true if 'preq' is the link parent of 'req' | |
8011 | */ | |
8012 | static bool io_match_link(struct io_kiocb *preq, struct io_kiocb *req) | |
8013 | { | |
8014 | struct io_kiocb *link; | |
8015 | ||
8016 | if (!(preq->flags & REQ_F_LINK_HEAD)) | |
8017 | return false; | |
8018 | ||
8019 | list_for_each_entry(link, &preq->link_list, link_list) { | |
8020 | if (link == req) | |
8021 | return true; | |
8022 | } | |
8023 | ||
8024 | return false; | |
8025 | } | |
8026 | ||
c127a2a1 PB |
8027 | static inline bool io_match_files(struct io_kiocb *req, |
8028 | struct files_struct *files) | |
8029 | { | |
8030 | return (req->flags & REQ_F_WORK_INITIALIZED) && req->work.files == files; | |
8031 | } | |
8032 | ||
8033 | static bool io_match_link_files(struct io_kiocb *req, | |
8034 | struct files_struct *files) | |
8035 | { | |
8036 | struct io_kiocb *link; | |
8037 | ||
8038 | if (io_match_files(req, files)) | |
8039 | return true; | |
8040 | if (req->flags & REQ_F_LINK_HEAD) { | |
8041 | list_for_each_entry(link, &req->link_list, link_list) { | |
8042 | if (io_match_files(link, files)) | |
8043 | return true; | |
8044 | } | |
8045 | } | |
8046 | return false; | |
8047 | } | |
8048 | ||
f254ac04 JA |
8049 | /* |
8050 | * We're looking to cancel 'req' because it's holding on to our files, but | |
8051 | * 'req' could be a link to another request. See if it is, and cancel that | |
8052 | * parent request if so. | |
8053 | */ | |
8054 | static bool io_poll_remove_link(struct io_ring_ctx *ctx, struct io_kiocb *req) | |
8055 | { | |
8056 | struct hlist_node *tmp; | |
8057 | struct io_kiocb *preq; | |
8058 | bool found = false; | |
8059 | int i; | |
8060 | ||
8061 | spin_lock_irq(&ctx->completion_lock); | |
8062 | for (i = 0; i < (1U << ctx->cancel_hash_bits); i++) { | |
8063 | struct hlist_head *list; | |
8064 | ||
8065 | list = &ctx->cancel_hash[i]; | |
8066 | hlist_for_each_entry_safe(preq, tmp, list, hash_node) { | |
8067 | found = io_match_link(preq, req); | |
8068 | if (found) { | |
8069 | io_poll_remove_one(preq); | |
8070 | break; | |
8071 | } | |
8072 | } | |
8073 | } | |
8074 | spin_unlock_irq(&ctx->completion_lock); | |
8075 | return found; | |
8076 | } | |
8077 | ||
8078 | static bool io_timeout_remove_link(struct io_ring_ctx *ctx, | |
8079 | struct io_kiocb *req) | |
8080 | { | |
8081 | struct io_kiocb *preq; | |
8082 | bool found = false; | |
8083 | ||
8084 | spin_lock_irq(&ctx->completion_lock); | |
8085 | list_for_each_entry(preq, &ctx->timeout_list, timeout.list) { | |
8086 | found = io_match_link(preq, req); | |
8087 | if (found) { | |
8088 | __io_timeout_cancel(preq); | |
8089 | break; | |
8090 | } | |
8091 | } | |
8092 | spin_unlock_irq(&ctx->completion_lock); | |
8093 | return found; | |
8094 | } | |
8095 | ||
b711d4ea JA |
8096 | static bool io_cancel_link_cb(struct io_wq_work *work, void *data) |
8097 | { | |
8098 | return io_match_link(container_of(work, struct io_kiocb, work), data); | |
8099 | } | |
8100 | ||
8101 | static void io_attempt_cancel(struct io_ring_ctx *ctx, struct io_kiocb *req) | |
8102 | { | |
8103 | enum io_wq_cancel cret; | |
8104 | ||
8105 | /* cancel this particular work, if it's running */ | |
8106 | cret = io_wq_cancel_work(ctx->io_wq, &req->work); | |
8107 | if (cret != IO_WQ_CANCEL_NOTFOUND) | |
8108 | return; | |
8109 | ||
8110 | /* find links that hold this pending, cancel those */ | |
8111 | cret = io_wq_cancel_cb(ctx->io_wq, io_cancel_link_cb, req, true); | |
8112 | if (cret != IO_WQ_CANCEL_NOTFOUND) | |
8113 | return; | |
8114 | ||
8115 | /* if we have a poll link holding this pending, cancel that */ | |
8116 | if (io_poll_remove_link(ctx, req)) | |
8117 | return; | |
8118 | ||
8119 | /* final option, timeout link is holding this req pending */ | |
8120 | io_timeout_remove_link(ctx, req); | |
8121 | } | |
8122 | ||
b7ddce3c PB |
8123 | static void io_cancel_defer_files(struct io_ring_ctx *ctx, |
8124 | struct files_struct *files) | |
8125 | { | |
8126 | struct io_defer_entry *de = NULL; | |
8127 | LIST_HEAD(list); | |
8128 | ||
8129 | spin_lock_irq(&ctx->completion_lock); | |
8130 | list_for_each_entry_reverse(de, &ctx->defer_list, list) { | |
c127a2a1 | 8131 | if (io_match_link_files(de->req, files)) { |
b7ddce3c PB |
8132 | list_cut_position(&list, &ctx->defer_list, &de->list); |
8133 | break; | |
8134 | } | |
8135 | } | |
8136 | spin_unlock_irq(&ctx->completion_lock); | |
8137 | ||
8138 | while (!list_empty(&list)) { | |
8139 | de = list_first_entry(&list, struct io_defer_entry, list); | |
8140 | list_del_init(&de->list); | |
8141 | req_set_fail_links(de->req); | |
8142 | io_put_req(de->req); | |
8143 | io_req_complete(de->req, -ECANCELED); | |
8144 | kfree(de); | |
8145 | } | |
8146 | } | |
8147 | ||
fcb323cc JA |
8148 | static void io_uring_cancel_files(struct io_ring_ctx *ctx, |
8149 | struct files_struct *files) | |
8150 | { | |
67c4d9e6 PB |
8151 | if (list_empty_careful(&ctx->inflight_list)) |
8152 | return; | |
8153 | ||
b7ddce3c | 8154 | io_cancel_defer_files(ctx, files); |
67c4d9e6 PB |
8155 | /* cancel all at once, should be faster than doing it one by one*/ |
8156 | io_wq_cancel_cb(ctx->io_wq, io_wq_files_match, files, true); | |
8157 | ||
fcb323cc | 8158 | while (!list_empty_careful(&ctx->inflight_list)) { |
d8f1b971 XW |
8159 | struct io_kiocb *cancel_req = NULL, *req; |
8160 | DEFINE_WAIT(wait); | |
fcb323cc JA |
8161 | |
8162 | spin_lock_irq(&ctx->inflight_lock); | |
8163 | list_for_each_entry(req, &ctx->inflight_list, inflight_entry) { | |
768134d4 JA |
8164 | if (req->work.files != files) |
8165 | continue; | |
8166 | /* req is being completed, ignore */ | |
8167 | if (!refcount_inc_not_zero(&req->refs)) | |
8168 | continue; | |
8169 | cancel_req = req; | |
8170 | break; | |
fcb323cc | 8171 | } |
768134d4 | 8172 | if (cancel_req) |
fcb323cc | 8173 | prepare_to_wait(&ctx->inflight_wait, &wait, |
768134d4 | 8174 | TASK_UNINTERRUPTIBLE); |
fcb323cc JA |
8175 | spin_unlock_irq(&ctx->inflight_lock); |
8176 | ||
768134d4 JA |
8177 | /* We need to keep going until we don't find a matching req */ |
8178 | if (!cancel_req) | |
fcb323cc | 8179 | break; |
bb175342 PB |
8180 | /* cancel this request, or head link requests */ |
8181 | io_attempt_cancel(ctx, cancel_req); | |
8182 | io_put_req(cancel_req); | |
fcb323cc | 8183 | schedule(); |
d8f1b971 | 8184 | finish_wait(&ctx->inflight_wait, &wait); |
fcb323cc JA |
8185 | } |
8186 | } | |
8187 | ||
801dd57b | 8188 | static bool io_cancel_task_cb(struct io_wq_work *work, void *data) |
44e728b8 | 8189 | { |
801dd57b PB |
8190 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); |
8191 | struct task_struct *task = data; | |
44e728b8 | 8192 | |
801dd57b | 8193 | return req->task == task; |
44e728b8 PB |
8194 | } |
8195 | ||
fcb323cc JA |
8196 | static int io_uring_flush(struct file *file, void *data) |
8197 | { | |
8198 | struct io_ring_ctx *ctx = file->private_data; | |
8199 | ||
8200 | io_uring_cancel_files(ctx, data); | |
6ab23144 JA |
8201 | |
8202 | /* | |
8203 | * If the task is going away, cancel work it may have pending | |
8204 | */ | |
801dd57b PB |
8205 | if (fatal_signal_pending(current) || (current->flags & PF_EXITING)) |
8206 | io_wq_cancel_cb(ctx->io_wq, io_cancel_task_cb, current, true); | |
6ab23144 | 8207 | |
fcb323cc JA |
8208 | return 0; |
8209 | } | |
8210 | ||
6c5c240e RP |
8211 | static void *io_uring_validate_mmap_request(struct file *file, |
8212 | loff_t pgoff, size_t sz) | |
2b188cc1 | 8213 | { |
2b188cc1 | 8214 | struct io_ring_ctx *ctx = file->private_data; |
6c5c240e | 8215 | loff_t offset = pgoff << PAGE_SHIFT; |
2b188cc1 JA |
8216 | struct page *page; |
8217 | void *ptr; | |
8218 | ||
8219 | switch (offset) { | |
8220 | case IORING_OFF_SQ_RING: | |
75b28aff HV |
8221 | case IORING_OFF_CQ_RING: |
8222 | ptr = ctx->rings; | |
2b188cc1 JA |
8223 | break; |
8224 | case IORING_OFF_SQES: | |
8225 | ptr = ctx->sq_sqes; | |
8226 | break; | |
2b188cc1 | 8227 | default: |
6c5c240e | 8228 | return ERR_PTR(-EINVAL); |
2b188cc1 JA |
8229 | } |
8230 | ||
8231 | page = virt_to_head_page(ptr); | |
a50b854e | 8232 | if (sz > page_size(page)) |
6c5c240e RP |
8233 | return ERR_PTR(-EINVAL); |
8234 | ||
8235 | return ptr; | |
8236 | } | |
8237 | ||
8238 | #ifdef CONFIG_MMU | |
8239 | ||
8240 | static int io_uring_mmap(struct file *file, struct vm_area_struct *vma) | |
8241 | { | |
8242 | size_t sz = vma->vm_end - vma->vm_start; | |
8243 | unsigned long pfn; | |
8244 | void *ptr; | |
8245 | ||
8246 | ptr = io_uring_validate_mmap_request(file, vma->vm_pgoff, sz); | |
8247 | if (IS_ERR(ptr)) | |
8248 | return PTR_ERR(ptr); | |
2b188cc1 JA |
8249 | |
8250 | pfn = virt_to_phys(ptr) >> PAGE_SHIFT; | |
8251 | return remap_pfn_range(vma, vma->vm_start, pfn, sz, vma->vm_page_prot); | |
8252 | } | |
8253 | ||
6c5c240e RP |
8254 | #else /* !CONFIG_MMU */ |
8255 | ||
8256 | static int io_uring_mmap(struct file *file, struct vm_area_struct *vma) | |
8257 | { | |
8258 | return vma->vm_flags & (VM_SHARED | VM_MAYSHARE) ? 0 : -EINVAL; | |
8259 | } | |
8260 | ||
8261 | static unsigned int io_uring_nommu_mmap_capabilities(struct file *file) | |
8262 | { | |
8263 | return NOMMU_MAP_DIRECT | NOMMU_MAP_READ | NOMMU_MAP_WRITE; | |
8264 | } | |
8265 | ||
8266 | static unsigned long io_uring_nommu_get_unmapped_area(struct file *file, | |
8267 | unsigned long addr, unsigned long len, | |
8268 | unsigned long pgoff, unsigned long flags) | |
8269 | { | |
8270 | void *ptr; | |
8271 | ||
8272 | ptr = io_uring_validate_mmap_request(file, pgoff, len); | |
8273 | if (IS_ERR(ptr)) | |
8274 | return PTR_ERR(ptr); | |
8275 | ||
8276 | return (unsigned long) ptr; | |
8277 | } | |
8278 | ||
8279 | #endif /* !CONFIG_MMU */ | |
8280 | ||
2b188cc1 JA |
8281 | SYSCALL_DEFINE6(io_uring_enter, unsigned int, fd, u32, to_submit, |
8282 | u32, min_complete, u32, flags, const sigset_t __user *, sig, | |
8283 | size_t, sigsz) | |
8284 | { | |
8285 | struct io_ring_ctx *ctx; | |
8286 | long ret = -EBADF; | |
8287 | int submitted = 0; | |
8288 | struct fd f; | |
8289 | ||
4c6e277c | 8290 | io_run_task_work(); |
b41e9852 | 8291 | |
6c271ce2 | 8292 | if (flags & ~(IORING_ENTER_GETEVENTS | IORING_ENTER_SQ_WAKEUP)) |
2b188cc1 JA |
8293 | return -EINVAL; |
8294 | ||
8295 | f = fdget(fd); | |
8296 | if (!f.file) | |
8297 | return -EBADF; | |
8298 | ||
8299 | ret = -EOPNOTSUPP; | |
8300 | if (f.file->f_op != &io_uring_fops) | |
8301 | goto out_fput; | |
8302 | ||
8303 | ret = -ENXIO; | |
8304 | ctx = f.file->private_data; | |
8305 | if (!percpu_ref_tryget(&ctx->refs)) | |
8306 | goto out_fput; | |
8307 | ||
6c271ce2 JA |
8308 | /* |
8309 | * For SQ polling, the thread will do all submissions and completions. | |
8310 | * Just return the requested submit count, and wake the thread if | |
8311 | * we were asked to. | |
8312 | */ | |
b2a9eada | 8313 | ret = 0; |
6c271ce2 | 8314 | if (ctx->flags & IORING_SETUP_SQPOLL) { |
c1edbf5f JA |
8315 | if (!list_empty_careful(&ctx->cq_overflow_list)) |
8316 | io_cqring_overflow_flush(ctx, false); | |
6c271ce2 JA |
8317 | if (flags & IORING_ENTER_SQ_WAKEUP) |
8318 | wake_up(&ctx->sqo_wait); | |
8319 | submitted = to_submit; | |
b2a9eada | 8320 | } else if (to_submit) { |
2b188cc1 | 8321 | mutex_lock(&ctx->uring_lock); |
0cdaf760 | 8322 | submitted = io_submit_sqes(ctx, to_submit, f.file, fd); |
2b188cc1 | 8323 | mutex_unlock(&ctx->uring_lock); |
7c504e65 PB |
8324 | |
8325 | if (submitted != to_submit) | |
8326 | goto out; | |
2b188cc1 JA |
8327 | } |
8328 | if (flags & IORING_ENTER_GETEVENTS) { | |
8329 | min_complete = min(min_complete, ctx->cq_entries); | |
8330 | ||
32b2244a XW |
8331 | /* |
8332 | * When SETUP_IOPOLL and SETUP_SQPOLL are both enabled, user | |
8333 | * space applications don't need to do io completion events | |
8334 | * polling again, they can rely on io_sq_thread to do polling | |
8335 | * work, which can reduce cpu usage and uring_lock contention. | |
8336 | */ | |
8337 | if (ctx->flags & IORING_SETUP_IOPOLL && | |
8338 | !(ctx->flags & IORING_SETUP_SQPOLL)) { | |
7668b92a | 8339 | ret = io_iopoll_check(ctx, min_complete); |
def596e9 JA |
8340 | } else { |
8341 | ret = io_cqring_wait(ctx, min_complete, sig, sigsz); | |
8342 | } | |
2b188cc1 JA |
8343 | } |
8344 | ||
7c504e65 | 8345 | out: |
6805b32e | 8346 | percpu_ref_put(&ctx->refs); |
2b188cc1 JA |
8347 | out_fput: |
8348 | fdput(f); | |
8349 | return submitted ? submitted : ret; | |
8350 | } | |
8351 | ||
bebdb65e | 8352 | #ifdef CONFIG_PROC_FS |
87ce955b JA |
8353 | static int io_uring_show_cred(int id, void *p, void *data) |
8354 | { | |
8355 | const struct cred *cred = p; | |
8356 | struct seq_file *m = data; | |
8357 | struct user_namespace *uns = seq_user_ns(m); | |
8358 | struct group_info *gi; | |
8359 | kernel_cap_t cap; | |
8360 | unsigned __capi; | |
8361 | int g; | |
8362 | ||
8363 | seq_printf(m, "%5d\n", id); | |
8364 | seq_put_decimal_ull(m, "\tUid:\t", from_kuid_munged(uns, cred->uid)); | |
8365 | seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->euid)); | |
8366 | seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->suid)); | |
8367 | seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->fsuid)); | |
8368 | seq_put_decimal_ull(m, "\n\tGid:\t", from_kgid_munged(uns, cred->gid)); | |
8369 | seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->egid)); | |
8370 | seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->sgid)); | |
8371 | seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->fsgid)); | |
8372 | seq_puts(m, "\n\tGroups:\t"); | |
8373 | gi = cred->group_info; | |
8374 | for (g = 0; g < gi->ngroups; g++) { | |
8375 | seq_put_decimal_ull(m, g ? " " : "", | |
8376 | from_kgid_munged(uns, gi->gid[g])); | |
8377 | } | |
8378 | seq_puts(m, "\n\tCapEff:\t"); | |
8379 | cap = cred->cap_effective; | |
8380 | CAP_FOR_EACH_U32(__capi) | |
8381 | seq_put_hex_ll(m, NULL, cap.cap[CAP_LAST_U32 - __capi], 8); | |
8382 | seq_putc(m, '\n'); | |
8383 | return 0; | |
8384 | } | |
8385 | ||
8386 | static void __io_uring_show_fdinfo(struct io_ring_ctx *ctx, struct seq_file *m) | |
8387 | { | |
8388 | int i; | |
8389 | ||
8390 | mutex_lock(&ctx->uring_lock); | |
8391 | seq_printf(m, "UserFiles:\t%u\n", ctx->nr_user_files); | |
8392 | for (i = 0; i < ctx->nr_user_files; i++) { | |
8393 | struct fixed_file_table *table; | |
8394 | struct file *f; | |
8395 | ||
8396 | table = &ctx->file_data->table[i >> IORING_FILE_TABLE_SHIFT]; | |
8397 | f = table->files[i & IORING_FILE_TABLE_MASK]; | |
8398 | if (f) | |
8399 | seq_printf(m, "%5u: %s\n", i, file_dentry(f)->d_iname); | |
8400 | else | |
8401 | seq_printf(m, "%5u: <none>\n", i); | |
8402 | } | |
8403 | seq_printf(m, "UserBufs:\t%u\n", ctx->nr_user_bufs); | |
8404 | for (i = 0; i < ctx->nr_user_bufs; i++) { | |
8405 | struct io_mapped_ubuf *buf = &ctx->user_bufs[i]; | |
8406 | ||
8407 | seq_printf(m, "%5u: 0x%llx/%u\n", i, buf->ubuf, | |
8408 | (unsigned int) buf->len); | |
8409 | } | |
8410 | if (!idr_is_empty(&ctx->personality_idr)) { | |
8411 | seq_printf(m, "Personalities:\n"); | |
8412 | idr_for_each(&ctx->personality_idr, io_uring_show_cred, m); | |
8413 | } | |
d7718a9d JA |
8414 | seq_printf(m, "PollList:\n"); |
8415 | spin_lock_irq(&ctx->completion_lock); | |
8416 | for (i = 0; i < (1U << ctx->cancel_hash_bits); i++) { | |
8417 | struct hlist_head *list = &ctx->cancel_hash[i]; | |
8418 | struct io_kiocb *req; | |
8419 | ||
8420 | hlist_for_each_entry(req, list, hash_node) | |
8421 | seq_printf(m, " op=%d, task_works=%d\n", req->opcode, | |
8422 | req->task->task_works != NULL); | |
8423 | } | |
8424 | spin_unlock_irq(&ctx->completion_lock); | |
87ce955b JA |
8425 | mutex_unlock(&ctx->uring_lock); |
8426 | } | |
8427 | ||
8428 | static void io_uring_show_fdinfo(struct seq_file *m, struct file *f) | |
8429 | { | |
8430 | struct io_ring_ctx *ctx = f->private_data; | |
8431 | ||
8432 | if (percpu_ref_tryget(&ctx->refs)) { | |
8433 | __io_uring_show_fdinfo(ctx, m); | |
8434 | percpu_ref_put(&ctx->refs); | |
8435 | } | |
8436 | } | |
bebdb65e | 8437 | #endif |
87ce955b | 8438 | |
2b188cc1 JA |
8439 | static const struct file_operations io_uring_fops = { |
8440 | .release = io_uring_release, | |
fcb323cc | 8441 | .flush = io_uring_flush, |
2b188cc1 | 8442 | .mmap = io_uring_mmap, |
6c5c240e RP |
8443 | #ifndef CONFIG_MMU |
8444 | .get_unmapped_area = io_uring_nommu_get_unmapped_area, | |
8445 | .mmap_capabilities = io_uring_nommu_mmap_capabilities, | |
8446 | #endif | |
2b188cc1 JA |
8447 | .poll = io_uring_poll, |
8448 | .fasync = io_uring_fasync, | |
bebdb65e | 8449 | #ifdef CONFIG_PROC_FS |
87ce955b | 8450 | .show_fdinfo = io_uring_show_fdinfo, |
bebdb65e | 8451 | #endif |
2b188cc1 JA |
8452 | }; |
8453 | ||
8454 | static int io_allocate_scq_urings(struct io_ring_ctx *ctx, | |
8455 | struct io_uring_params *p) | |
8456 | { | |
75b28aff HV |
8457 | struct io_rings *rings; |
8458 | size_t size, sq_array_offset; | |
2b188cc1 | 8459 | |
bd740481 JA |
8460 | /* make sure these are sane, as we already accounted them */ |
8461 | ctx->sq_entries = p->sq_entries; | |
8462 | ctx->cq_entries = p->cq_entries; | |
8463 | ||
75b28aff HV |
8464 | size = rings_size(p->sq_entries, p->cq_entries, &sq_array_offset); |
8465 | if (size == SIZE_MAX) | |
8466 | return -EOVERFLOW; | |
8467 | ||
8468 | rings = io_mem_alloc(size); | |
8469 | if (!rings) | |
2b188cc1 JA |
8470 | return -ENOMEM; |
8471 | ||
75b28aff HV |
8472 | ctx->rings = rings; |
8473 | ctx->sq_array = (u32 *)((char *)rings + sq_array_offset); | |
8474 | rings->sq_ring_mask = p->sq_entries - 1; | |
8475 | rings->cq_ring_mask = p->cq_entries - 1; | |
8476 | rings->sq_ring_entries = p->sq_entries; | |
8477 | rings->cq_ring_entries = p->cq_entries; | |
8478 | ctx->sq_mask = rings->sq_ring_mask; | |
8479 | ctx->cq_mask = rings->cq_ring_mask; | |
2b188cc1 JA |
8480 | |
8481 | size = array_size(sizeof(struct io_uring_sqe), p->sq_entries); | |
eb065d30 JA |
8482 | if (size == SIZE_MAX) { |
8483 | io_mem_free(ctx->rings); | |
8484 | ctx->rings = NULL; | |
2b188cc1 | 8485 | return -EOVERFLOW; |
eb065d30 | 8486 | } |
2b188cc1 JA |
8487 | |
8488 | ctx->sq_sqes = io_mem_alloc(size); | |
eb065d30 JA |
8489 | if (!ctx->sq_sqes) { |
8490 | io_mem_free(ctx->rings); | |
8491 | ctx->rings = NULL; | |
2b188cc1 | 8492 | return -ENOMEM; |
eb065d30 | 8493 | } |
2b188cc1 | 8494 | |
2b188cc1 JA |
8495 | return 0; |
8496 | } | |
8497 | ||
8498 | /* | |
8499 | * Allocate an anonymous fd, this is what constitutes the application | |
8500 | * visible backing of an io_uring instance. The application mmaps this | |
8501 | * fd to gain access to the SQ/CQ ring details. If UNIX sockets are enabled, | |
8502 | * we have to tie this fd to a socket for file garbage collection purposes. | |
8503 | */ | |
8504 | static int io_uring_get_fd(struct io_ring_ctx *ctx) | |
8505 | { | |
8506 | struct file *file; | |
8507 | int ret; | |
8508 | ||
8509 | #if defined(CONFIG_UNIX) | |
8510 | ret = sock_create_kern(&init_net, PF_UNIX, SOCK_RAW, IPPROTO_IP, | |
8511 | &ctx->ring_sock); | |
8512 | if (ret) | |
8513 | return ret; | |
8514 | #endif | |
8515 | ||
8516 | ret = get_unused_fd_flags(O_RDWR | O_CLOEXEC); | |
8517 | if (ret < 0) | |
8518 | goto err; | |
8519 | ||
8520 | file = anon_inode_getfile("[io_uring]", &io_uring_fops, ctx, | |
8521 | O_RDWR | O_CLOEXEC); | |
8522 | if (IS_ERR(file)) { | |
8523 | put_unused_fd(ret); | |
8524 | ret = PTR_ERR(file); | |
8525 | goto err; | |
8526 | } | |
8527 | ||
8528 | #if defined(CONFIG_UNIX) | |
8529 | ctx->ring_sock->file = file; | |
8530 | #endif | |
8531 | fd_install(ret, file); | |
8532 | return ret; | |
8533 | err: | |
8534 | #if defined(CONFIG_UNIX) | |
8535 | sock_release(ctx->ring_sock); | |
8536 | ctx->ring_sock = NULL; | |
8537 | #endif | |
8538 | return ret; | |
8539 | } | |
8540 | ||
7f13657d XW |
8541 | static int io_uring_create(unsigned entries, struct io_uring_params *p, |
8542 | struct io_uring_params __user *params) | |
2b188cc1 JA |
8543 | { |
8544 | struct user_struct *user = NULL; | |
8545 | struct io_ring_ctx *ctx; | |
aad5d8da | 8546 | bool limit_mem; |
2b188cc1 JA |
8547 | int ret; |
8548 | ||
8110c1a6 | 8549 | if (!entries) |
2b188cc1 | 8550 | return -EINVAL; |
8110c1a6 JA |
8551 | if (entries > IORING_MAX_ENTRIES) { |
8552 | if (!(p->flags & IORING_SETUP_CLAMP)) | |
8553 | return -EINVAL; | |
8554 | entries = IORING_MAX_ENTRIES; | |
8555 | } | |
2b188cc1 JA |
8556 | |
8557 | /* | |
8558 | * Use twice as many entries for the CQ ring. It's possible for the | |
8559 | * application to drive a higher depth than the size of the SQ ring, | |
8560 | * since the sqes are only used at submission time. This allows for | |
33a107f0 JA |
8561 | * some flexibility in overcommitting a bit. If the application has |
8562 | * set IORING_SETUP_CQSIZE, it will have passed in the desired number | |
8563 | * of CQ ring entries manually. | |
2b188cc1 JA |
8564 | */ |
8565 | p->sq_entries = roundup_pow_of_two(entries); | |
33a107f0 JA |
8566 | if (p->flags & IORING_SETUP_CQSIZE) { |
8567 | /* | |
8568 | * If IORING_SETUP_CQSIZE is set, we do the same roundup | |
8569 | * to a power-of-two, if it isn't already. We do NOT impose | |
8570 | * any cq vs sq ring sizing. | |
8571 | */ | |
8110c1a6 | 8572 | if (p->cq_entries < p->sq_entries) |
33a107f0 | 8573 | return -EINVAL; |
8110c1a6 JA |
8574 | if (p->cq_entries > IORING_MAX_CQ_ENTRIES) { |
8575 | if (!(p->flags & IORING_SETUP_CLAMP)) | |
8576 | return -EINVAL; | |
8577 | p->cq_entries = IORING_MAX_CQ_ENTRIES; | |
8578 | } | |
33a107f0 JA |
8579 | p->cq_entries = roundup_pow_of_two(p->cq_entries); |
8580 | } else { | |
8581 | p->cq_entries = 2 * p->sq_entries; | |
8582 | } | |
2b188cc1 JA |
8583 | |
8584 | user = get_uid(current_user()); | |
aad5d8da | 8585 | limit_mem = !capable(CAP_IPC_LOCK); |
2b188cc1 | 8586 | |
aad5d8da | 8587 | if (limit_mem) { |
a087e2b5 | 8588 | ret = __io_account_mem(user, |
2b188cc1 JA |
8589 | ring_pages(p->sq_entries, p->cq_entries)); |
8590 | if (ret) { | |
8591 | free_uid(user); | |
8592 | return ret; | |
8593 | } | |
8594 | } | |
8595 | ||
8596 | ctx = io_ring_ctx_alloc(p); | |
8597 | if (!ctx) { | |
aad5d8da | 8598 | if (limit_mem) |
a087e2b5 | 8599 | __io_unaccount_mem(user, ring_pages(p->sq_entries, |
2b188cc1 JA |
8600 | p->cq_entries)); |
8601 | free_uid(user); | |
8602 | return -ENOMEM; | |
8603 | } | |
8604 | ctx->compat = in_compat_syscall(); | |
2b188cc1 | 8605 | ctx->user = user; |
0b8c0ec7 | 8606 | ctx->creds = get_current_cred(); |
2b188cc1 | 8607 | |
6b7898eb JA |
8608 | mmgrab(current->mm); |
8609 | ctx->sqo_mm = current->mm; | |
8610 | ||
f74441e6 JA |
8611 | /* |
8612 | * Account memory _before_ installing the file descriptor. Once | |
8613 | * the descriptor is installed, it can get closed at any time. Also | |
8614 | * do this before hitting the general error path, as ring freeing | |
8615 | * will un-account as well. | |
8616 | */ | |
8617 | io_account_mem(ctx, ring_pages(p->sq_entries, p->cq_entries), | |
8618 | ACCT_LOCKED); | |
8619 | ctx->limit_mem = limit_mem; | |
8620 | ||
2b188cc1 JA |
8621 | ret = io_allocate_scq_urings(ctx, p); |
8622 | if (ret) | |
8623 | goto err; | |
8624 | ||
6c271ce2 | 8625 | ret = io_sq_offload_start(ctx, p); |
2b188cc1 JA |
8626 | if (ret) |
8627 | goto err; | |
8628 | ||
2b188cc1 | 8629 | memset(&p->sq_off, 0, sizeof(p->sq_off)); |
75b28aff HV |
8630 | p->sq_off.head = offsetof(struct io_rings, sq.head); |
8631 | p->sq_off.tail = offsetof(struct io_rings, sq.tail); | |
8632 | p->sq_off.ring_mask = offsetof(struct io_rings, sq_ring_mask); | |
8633 | p->sq_off.ring_entries = offsetof(struct io_rings, sq_ring_entries); | |
8634 | p->sq_off.flags = offsetof(struct io_rings, sq_flags); | |
8635 | p->sq_off.dropped = offsetof(struct io_rings, sq_dropped); | |
8636 | p->sq_off.array = (char *)ctx->sq_array - (char *)ctx->rings; | |
2b188cc1 JA |
8637 | |
8638 | memset(&p->cq_off, 0, sizeof(p->cq_off)); | |
75b28aff HV |
8639 | p->cq_off.head = offsetof(struct io_rings, cq.head); |
8640 | p->cq_off.tail = offsetof(struct io_rings, cq.tail); | |
8641 | p->cq_off.ring_mask = offsetof(struct io_rings, cq_ring_mask); | |
8642 | p->cq_off.ring_entries = offsetof(struct io_rings, cq_ring_entries); | |
8643 | p->cq_off.overflow = offsetof(struct io_rings, cq_overflow); | |
8644 | p->cq_off.cqes = offsetof(struct io_rings, cqes); | |
0d9b5b3a | 8645 | p->cq_off.flags = offsetof(struct io_rings, cq_flags); |
ac90f249 | 8646 | |
7f13657d XW |
8647 | p->features = IORING_FEAT_SINGLE_MMAP | IORING_FEAT_NODROP | |
8648 | IORING_FEAT_SUBMIT_STABLE | IORING_FEAT_RW_CUR_POS | | |
5769a351 JX |
8649 | IORING_FEAT_CUR_PERSONALITY | IORING_FEAT_FAST_POLL | |
8650 | IORING_FEAT_POLL_32BITS; | |
7f13657d XW |
8651 | |
8652 | if (copy_to_user(params, p, sizeof(*p))) { | |
8653 | ret = -EFAULT; | |
8654 | goto err; | |
8655 | } | |
d1719f70 | 8656 | |
044c1ab3 JA |
8657 | /* |
8658 | * Install ring fd as the very last thing, so we don't risk someone | |
8659 | * having closed it before we finish setup | |
8660 | */ | |
8661 | ret = io_uring_get_fd(ctx); | |
8662 | if (ret < 0) | |
8663 | goto err; | |
8664 | ||
c826bd7a | 8665 | trace_io_uring_create(ret, ctx, p->sq_entries, p->cq_entries, p->flags); |
2b188cc1 JA |
8666 | return ret; |
8667 | err: | |
8668 | io_ring_ctx_wait_and_kill(ctx); | |
8669 | return ret; | |
8670 | } | |
8671 | ||
8672 | /* | |
8673 | * Sets up an aio uring context, and returns the fd. Applications asks for a | |
8674 | * ring size, we return the actual sq/cq ring sizes (among other things) in the | |
8675 | * params structure passed in. | |
8676 | */ | |
8677 | static long io_uring_setup(u32 entries, struct io_uring_params __user *params) | |
8678 | { | |
8679 | struct io_uring_params p; | |
2b188cc1 JA |
8680 | int i; |
8681 | ||
8682 | if (copy_from_user(&p, params, sizeof(p))) | |
8683 | return -EFAULT; | |
8684 | for (i = 0; i < ARRAY_SIZE(p.resv); i++) { | |
8685 | if (p.resv[i]) | |
8686 | return -EINVAL; | |
8687 | } | |
8688 | ||
6c271ce2 | 8689 | if (p.flags & ~(IORING_SETUP_IOPOLL | IORING_SETUP_SQPOLL | |
8110c1a6 | 8690 | IORING_SETUP_SQ_AFF | IORING_SETUP_CQSIZE | |
24369c2e | 8691 | IORING_SETUP_CLAMP | IORING_SETUP_ATTACH_WQ)) |
2b188cc1 JA |
8692 | return -EINVAL; |
8693 | ||
7f13657d | 8694 | return io_uring_create(entries, &p, params); |
2b188cc1 JA |
8695 | } |
8696 | ||
8697 | SYSCALL_DEFINE2(io_uring_setup, u32, entries, | |
8698 | struct io_uring_params __user *, params) | |
8699 | { | |
8700 | return io_uring_setup(entries, params); | |
8701 | } | |
8702 | ||
66f4af93 JA |
8703 | static int io_probe(struct io_ring_ctx *ctx, void __user *arg, unsigned nr_args) |
8704 | { | |
8705 | struct io_uring_probe *p; | |
8706 | size_t size; | |
8707 | int i, ret; | |
8708 | ||
8709 | size = struct_size(p, ops, nr_args); | |
8710 | if (size == SIZE_MAX) | |
8711 | return -EOVERFLOW; | |
8712 | p = kzalloc(size, GFP_KERNEL); | |
8713 | if (!p) | |
8714 | return -ENOMEM; | |
8715 | ||
8716 | ret = -EFAULT; | |
8717 | if (copy_from_user(p, arg, size)) | |
8718 | goto out; | |
8719 | ret = -EINVAL; | |
8720 | if (memchr_inv(p, 0, size)) | |
8721 | goto out; | |
8722 | ||
8723 | p->last_op = IORING_OP_LAST - 1; | |
8724 | if (nr_args > IORING_OP_LAST) | |
8725 | nr_args = IORING_OP_LAST; | |
8726 | ||
8727 | for (i = 0; i < nr_args; i++) { | |
8728 | p->ops[i].op = i; | |
8729 | if (!io_op_defs[i].not_supported) | |
8730 | p->ops[i].flags = IO_URING_OP_SUPPORTED; | |
8731 | } | |
8732 | p->ops_len = i; | |
8733 | ||
8734 | ret = 0; | |
8735 | if (copy_to_user(arg, p, size)) | |
8736 | ret = -EFAULT; | |
8737 | out: | |
8738 | kfree(p); | |
8739 | return ret; | |
8740 | } | |
8741 | ||
071698e1 JA |
8742 | static int io_register_personality(struct io_ring_ctx *ctx) |
8743 | { | |
8744 | const struct cred *creds = get_current_cred(); | |
8745 | int id; | |
8746 | ||
8747 | id = idr_alloc_cyclic(&ctx->personality_idr, (void *) creds, 1, | |
8748 | USHRT_MAX, GFP_KERNEL); | |
8749 | if (id < 0) | |
8750 | put_cred(creds); | |
8751 | return id; | |
8752 | } | |
8753 | ||
8754 | static int io_unregister_personality(struct io_ring_ctx *ctx, unsigned id) | |
8755 | { | |
8756 | const struct cred *old_creds; | |
8757 | ||
8758 | old_creds = idr_remove(&ctx->personality_idr, id); | |
8759 | if (old_creds) { | |
8760 | put_cred(old_creds); | |
8761 | return 0; | |
8762 | } | |
8763 | ||
8764 | return -EINVAL; | |
8765 | } | |
8766 | ||
8767 | static bool io_register_op_must_quiesce(int op) | |
8768 | { | |
8769 | switch (op) { | |
8770 | case IORING_UNREGISTER_FILES: | |
8771 | case IORING_REGISTER_FILES_UPDATE: | |
8772 | case IORING_REGISTER_PROBE: | |
8773 | case IORING_REGISTER_PERSONALITY: | |
8774 | case IORING_UNREGISTER_PERSONALITY: | |
8775 | return false; | |
8776 | default: | |
8777 | return true; | |
8778 | } | |
8779 | } | |
8780 | ||
edafccee JA |
8781 | static int __io_uring_register(struct io_ring_ctx *ctx, unsigned opcode, |
8782 | void __user *arg, unsigned nr_args) | |
b19062a5 JA |
8783 | __releases(ctx->uring_lock) |
8784 | __acquires(ctx->uring_lock) | |
edafccee JA |
8785 | { |
8786 | int ret; | |
8787 | ||
35fa71a0 JA |
8788 | /* |
8789 | * We're inside the ring mutex, if the ref is already dying, then | |
8790 | * someone else killed the ctx or is already going through | |
8791 | * io_uring_register(). | |
8792 | */ | |
8793 | if (percpu_ref_is_dying(&ctx->refs)) | |
8794 | return -ENXIO; | |
8795 | ||
071698e1 | 8796 | if (io_register_op_must_quiesce(opcode)) { |
05f3fb3c | 8797 | percpu_ref_kill(&ctx->refs); |
b19062a5 | 8798 | |
05f3fb3c JA |
8799 | /* |
8800 | * Drop uring mutex before waiting for references to exit. If | |
8801 | * another thread is currently inside io_uring_enter() it might | |
8802 | * need to grab the uring_lock to make progress. If we hold it | |
8803 | * here across the drain wait, then we can deadlock. It's safe | |
8804 | * to drop the mutex here, since no new references will come in | |
8805 | * after we've killed the percpu ref. | |
8806 | */ | |
8807 | mutex_unlock(&ctx->uring_lock); | |
0f158b4c | 8808 | ret = wait_for_completion_interruptible(&ctx->ref_comp); |
05f3fb3c | 8809 | mutex_lock(&ctx->uring_lock); |
c150368b JA |
8810 | if (ret) { |
8811 | percpu_ref_resurrect(&ctx->refs); | |
8812 | ret = -EINTR; | |
8813 | goto out; | |
8814 | } | |
05f3fb3c | 8815 | } |
edafccee JA |
8816 | |
8817 | switch (opcode) { | |
8818 | case IORING_REGISTER_BUFFERS: | |
8819 | ret = io_sqe_buffer_register(ctx, arg, nr_args); | |
8820 | break; | |
8821 | case IORING_UNREGISTER_BUFFERS: | |
8822 | ret = -EINVAL; | |
8823 | if (arg || nr_args) | |
8824 | break; | |
8825 | ret = io_sqe_buffer_unregister(ctx); | |
8826 | break; | |
6b06314c JA |
8827 | case IORING_REGISTER_FILES: |
8828 | ret = io_sqe_files_register(ctx, arg, nr_args); | |
8829 | break; | |
8830 | case IORING_UNREGISTER_FILES: | |
8831 | ret = -EINVAL; | |
8832 | if (arg || nr_args) | |
8833 | break; | |
8834 | ret = io_sqe_files_unregister(ctx); | |
8835 | break; | |
c3a31e60 JA |
8836 | case IORING_REGISTER_FILES_UPDATE: |
8837 | ret = io_sqe_files_update(ctx, arg, nr_args); | |
8838 | break; | |
9b402849 | 8839 | case IORING_REGISTER_EVENTFD: |
f2842ab5 | 8840 | case IORING_REGISTER_EVENTFD_ASYNC: |
9b402849 JA |
8841 | ret = -EINVAL; |
8842 | if (nr_args != 1) | |
8843 | break; | |
8844 | ret = io_eventfd_register(ctx, arg); | |
f2842ab5 JA |
8845 | if (ret) |
8846 | break; | |
8847 | if (opcode == IORING_REGISTER_EVENTFD_ASYNC) | |
8848 | ctx->eventfd_async = 1; | |
8849 | else | |
8850 | ctx->eventfd_async = 0; | |
9b402849 JA |
8851 | break; |
8852 | case IORING_UNREGISTER_EVENTFD: | |
8853 | ret = -EINVAL; | |
8854 | if (arg || nr_args) | |
8855 | break; | |
8856 | ret = io_eventfd_unregister(ctx); | |
8857 | break; | |
66f4af93 JA |
8858 | case IORING_REGISTER_PROBE: |
8859 | ret = -EINVAL; | |
8860 | if (!arg || nr_args > 256) | |
8861 | break; | |
8862 | ret = io_probe(ctx, arg, nr_args); | |
8863 | break; | |
071698e1 JA |
8864 | case IORING_REGISTER_PERSONALITY: |
8865 | ret = -EINVAL; | |
8866 | if (arg || nr_args) | |
8867 | break; | |
8868 | ret = io_register_personality(ctx); | |
8869 | break; | |
8870 | case IORING_UNREGISTER_PERSONALITY: | |
8871 | ret = -EINVAL; | |
8872 | if (arg) | |
8873 | break; | |
8874 | ret = io_unregister_personality(ctx, nr_args); | |
8875 | break; | |
edafccee JA |
8876 | default: |
8877 | ret = -EINVAL; | |
8878 | break; | |
8879 | } | |
8880 | ||
071698e1 | 8881 | if (io_register_op_must_quiesce(opcode)) { |
05f3fb3c | 8882 | /* bring the ctx back to life */ |
05f3fb3c | 8883 | percpu_ref_reinit(&ctx->refs); |
c150368b | 8884 | out: |
0f158b4c | 8885 | reinit_completion(&ctx->ref_comp); |
05f3fb3c | 8886 | } |
edafccee JA |
8887 | return ret; |
8888 | } | |
8889 | ||
8890 | SYSCALL_DEFINE4(io_uring_register, unsigned int, fd, unsigned int, opcode, | |
8891 | void __user *, arg, unsigned int, nr_args) | |
8892 | { | |
8893 | struct io_ring_ctx *ctx; | |
8894 | long ret = -EBADF; | |
8895 | struct fd f; | |
8896 | ||
8897 | f = fdget(fd); | |
8898 | if (!f.file) | |
8899 | return -EBADF; | |
8900 | ||
8901 | ret = -EOPNOTSUPP; | |
8902 | if (f.file->f_op != &io_uring_fops) | |
8903 | goto out_fput; | |
8904 | ||
8905 | ctx = f.file->private_data; | |
8906 | ||
8907 | mutex_lock(&ctx->uring_lock); | |
8908 | ret = __io_uring_register(ctx, opcode, arg, nr_args); | |
8909 | mutex_unlock(&ctx->uring_lock); | |
c826bd7a DD |
8910 | trace_io_uring_register(ctx, opcode, ctx->nr_user_files, ctx->nr_user_bufs, |
8911 | ctx->cq_ev_fd != NULL, ret); | |
edafccee JA |
8912 | out_fput: |
8913 | fdput(f); | |
8914 | return ret; | |
8915 | } | |
8916 | ||
2b188cc1 JA |
8917 | static int __init io_uring_init(void) |
8918 | { | |
d7f62e82 SM |
8919 | #define __BUILD_BUG_VERIFY_ELEMENT(stype, eoffset, etype, ename) do { \ |
8920 | BUILD_BUG_ON(offsetof(stype, ename) != eoffset); \ | |
8921 | BUILD_BUG_ON(sizeof(etype) != sizeof_field(stype, ename)); \ | |
8922 | } while (0) | |
8923 | ||
8924 | #define BUILD_BUG_SQE_ELEM(eoffset, etype, ename) \ | |
8925 | __BUILD_BUG_VERIFY_ELEMENT(struct io_uring_sqe, eoffset, etype, ename) | |
8926 | BUILD_BUG_ON(sizeof(struct io_uring_sqe) != 64); | |
8927 | BUILD_BUG_SQE_ELEM(0, __u8, opcode); | |
8928 | BUILD_BUG_SQE_ELEM(1, __u8, flags); | |
8929 | BUILD_BUG_SQE_ELEM(2, __u16, ioprio); | |
8930 | BUILD_BUG_SQE_ELEM(4, __s32, fd); | |
8931 | BUILD_BUG_SQE_ELEM(8, __u64, off); | |
8932 | BUILD_BUG_SQE_ELEM(8, __u64, addr2); | |
8933 | BUILD_BUG_SQE_ELEM(16, __u64, addr); | |
7d67af2c | 8934 | BUILD_BUG_SQE_ELEM(16, __u64, splice_off_in); |
d7f62e82 SM |
8935 | BUILD_BUG_SQE_ELEM(24, __u32, len); |
8936 | BUILD_BUG_SQE_ELEM(28, __kernel_rwf_t, rw_flags); | |
8937 | BUILD_BUG_SQE_ELEM(28, /* compat */ int, rw_flags); | |
8938 | BUILD_BUG_SQE_ELEM(28, /* compat */ __u32, rw_flags); | |
8939 | BUILD_BUG_SQE_ELEM(28, __u32, fsync_flags); | |
5769a351 JX |
8940 | BUILD_BUG_SQE_ELEM(28, /* compat */ __u16, poll_events); |
8941 | BUILD_BUG_SQE_ELEM(28, __u32, poll32_events); | |
d7f62e82 SM |
8942 | BUILD_BUG_SQE_ELEM(28, __u32, sync_range_flags); |
8943 | BUILD_BUG_SQE_ELEM(28, __u32, msg_flags); | |
8944 | BUILD_BUG_SQE_ELEM(28, __u32, timeout_flags); | |
8945 | BUILD_BUG_SQE_ELEM(28, __u32, accept_flags); | |
8946 | BUILD_BUG_SQE_ELEM(28, __u32, cancel_flags); | |
8947 | BUILD_BUG_SQE_ELEM(28, __u32, open_flags); | |
8948 | BUILD_BUG_SQE_ELEM(28, __u32, statx_flags); | |
8949 | BUILD_BUG_SQE_ELEM(28, __u32, fadvise_advice); | |
7d67af2c | 8950 | BUILD_BUG_SQE_ELEM(28, __u32, splice_flags); |
d7f62e82 SM |
8951 | BUILD_BUG_SQE_ELEM(32, __u64, user_data); |
8952 | BUILD_BUG_SQE_ELEM(40, __u16, buf_index); | |
8953 | BUILD_BUG_SQE_ELEM(42, __u16, personality); | |
7d67af2c | 8954 | BUILD_BUG_SQE_ELEM(44, __s32, splice_fd_in); |
d7f62e82 | 8955 | |
d3656344 | 8956 | BUILD_BUG_ON(ARRAY_SIZE(io_op_defs) != IORING_OP_LAST); |
84557871 | 8957 | BUILD_BUG_ON(__REQ_F_LAST_BIT >= 8 * sizeof(int)); |
2b188cc1 JA |
8958 | req_cachep = KMEM_CACHE(io_kiocb, SLAB_HWCACHE_ALIGN | SLAB_PANIC); |
8959 | return 0; | |
8960 | }; | |
8961 | __initcall(io_uring_init); |