]>
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> |
0f212204 | 82 | #include <linux/io_uring.h> |
91d8f519 | 83 | #include <linux/blk-cgroup.h> |
4ea33a97 | 84 | #include <linux/audit.h> |
2b188cc1 | 85 | |
c826bd7a DD |
86 | #define CREATE_TRACE_POINTS |
87 | #include <trace/events/io_uring.h> | |
88 | ||
2b188cc1 JA |
89 | #include <uapi/linux/io_uring.h> |
90 | ||
91 | #include "internal.h" | |
561fb04a | 92 | #include "io-wq.h" |
2b188cc1 | 93 | |
5277deaa | 94 | #define IORING_MAX_ENTRIES 32768 |
33a107f0 | 95 | #define IORING_MAX_CQ_ENTRIES (2 * IORING_MAX_ENTRIES) |
65e19f54 JA |
96 | |
97 | /* | |
98 | * Shift of 9 is 512 entries, or exactly one page on 64-bit archs | |
99 | */ | |
100 | #define IORING_FILE_TABLE_SHIFT 9 | |
101 | #define IORING_MAX_FILES_TABLE (1U << IORING_FILE_TABLE_SHIFT) | |
102 | #define IORING_FILE_TABLE_MASK (IORING_MAX_FILES_TABLE - 1) | |
103 | #define IORING_MAX_FIXED_FILES (64 * IORING_MAX_FILES_TABLE) | |
21b55dbc SG |
104 | #define IORING_MAX_RESTRICTIONS (IORING_RESTRICTION_LAST + \ |
105 | IORING_REGISTER_LAST + IORING_OP_LAST) | |
2b188cc1 JA |
106 | |
107 | struct io_uring { | |
108 | u32 head ____cacheline_aligned_in_smp; | |
109 | u32 tail ____cacheline_aligned_in_smp; | |
110 | }; | |
111 | ||
1e84b97b | 112 | /* |
75b28aff HV |
113 | * This data is shared with the application through the mmap at offsets |
114 | * IORING_OFF_SQ_RING and IORING_OFF_CQ_RING. | |
1e84b97b SB |
115 | * |
116 | * The offsets to the member fields are published through struct | |
117 | * io_sqring_offsets when calling io_uring_setup. | |
118 | */ | |
75b28aff | 119 | struct io_rings { |
1e84b97b SB |
120 | /* |
121 | * Head and tail offsets into the ring; the offsets need to be | |
122 | * masked to get valid indices. | |
123 | * | |
75b28aff HV |
124 | * The kernel controls head of the sq ring and the tail of the cq ring, |
125 | * and the application controls tail of the sq ring and the head of the | |
126 | * cq ring. | |
1e84b97b | 127 | */ |
75b28aff | 128 | struct io_uring sq, cq; |
1e84b97b | 129 | /* |
75b28aff | 130 | * Bitmasks to apply to head and tail offsets (constant, equals |
1e84b97b SB |
131 | * ring_entries - 1) |
132 | */ | |
75b28aff HV |
133 | u32 sq_ring_mask, cq_ring_mask; |
134 | /* Ring sizes (constant, power of 2) */ | |
135 | u32 sq_ring_entries, cq_ring_entries; | |
1e84b97b SB |
136 | /* |
137 | * Number of invalid entries dropped by the kernel due to | |
138 | * invalid index stored in array | |
139 | * | |
140 | * Written by the kernel, shouldn't be modified by the | |
141 | * application (i.e. get number of "new events" by comparing to | |
142 | * cached value). | |
143 | * | |
144 | * After a new SQ head value was read by the application this | |
145 | * counter includes all submissions that were dropped reaching | |
146 | * the new SQ head (and possibly more). | |
147 | */ | |
75b28aff | 148 | u32 sq_dropped; |
1e84b97b | 149 | /* |
0d9b5b3a | 150 | * Runtime SQ flags |
1e84b97b SB |
151 | * |
152 | * Written by the kernel, shouldn't be modified by the | |
153 | * application. | |
154 | * | |
155 | * The application needs a full memory barrier before checking | |
156 | * for IORING_SQ_NEED_WAKEUP after updating the sq tail. | |
157 | */ | |
75b28aff | 158 | u32 sq_flags; |
0d9b5b3a SG |
159 | /* |
160 | * Runtime CQ flags | |
161 | * | |
162 | * Written by the application, shouldn't be modified by the | |
163 | * kernel. | |
164 | */ | |
165 | u32 cq_flags; | |
1e84b97b SB |
166 | /* |
167 | * Number of completion events lost because the queue was full; | |
168 | * this should be avoided by the application by making sure | |
0b4295b5 | 169 | * there are not more requests pending than there is space in |
1e84b97b SB |
170 | * the completion queue. |
171 | * | |
172 | * Written by the kernel, shouldn't be modified by the | |
173 | * application (i.e. get number of "new events" by comparing to | |
174 | * cached value). | |
175 | * | |
176 | * As completion events come in out of order this counter is not | |
177 | * ordered with any other data. | |
178 | */ | |
75b28aff | 179 | u32 cq_overflow; |
1e84b97b SB |
180 | /* |
181 | * Ring buffer of completion events. | |
182 | * | |
183 | * The kernel writes completion events fresh every time they are | |
184 | * produced, so the application is allowed to modify pending | |
185 | * entries. | |
186 | */ | |
75b28aff | 187 | struct io_uring_cqe cqes[] ____cacheline_aligned_in_smp; |
2b188cc1 JA |
188 | }; |
189 | ||
edafccee JA |
190 | struct io_mapped_ubuf { |
191 | u64 ubuf; | |
192 | size_t len; | |
193 | struct bio_vec *bvec; | |
194 | unsigned int nr_bvecs; | |
de293938 | 195 | unsigned long acct_pages; |
edafccee JA |
196 | }; |
197 | ||
65e19f54 JA |
198 | struct fixed_file_table { |
199 | struct file **files; | |
31b51510 JA |
200 | }; |
201 | ||
05589553 XW |
202 | struct fixed_file_ref_node { |
203 | struct percpu_ref refs; | |
204 | struct list_head node; | |
205 | struct list_head file_list; | |
206 | struct fixed_file_data *file_data; | |
4a38aed2 | 207 | struct llist_node llist; |
e297822b | 208 | bool done; |
05589553 XW |
209 | }; |
210 | ||
05f3fb3c JA |
211 | struct fixed_file_data { |
212 | struct fixed_file_table *table; | |
213 | struct io_ring_ctx *ctx; | |
214 | ||
b2e96852 | 215 | struct fixed_file_ref_node *node; |
05f3fb3c | 216 | struct percpu_ref refs; |
05f3fb3c | 217 | struct completion done; |
05589553 XW |
218 | struct list_head ref_list; |
219 | spinlock_t lock; | |
05f3fb3c JA |
220 | }; |
221 | ||
5a2e745d JA |
222 | struct io_buffer { |
223 | struct list_head list; | |
224 | __u64 addr; | |
225 | __s32 len; | |
226 | __u16 bid; | |
227 | }; | |
228 | ||
21b55dbc SG |
229 | struct io_restriction { |
230 | DECLARE_BITMAP(register_op, IORING_REGISTER_LAST); | |
231 | DECLARE_BITMAP(sqe_op, IORING_OP_LAST); | |
232 | u8 sqe_flags_allowed; | |
233 | u8 sqe_flags_required; | |
7e84e1c7 | 234 | bool registered; |
21b55dbc SG |
235 | }; |
236 | ||
534ca6d6 JA |
237 | struct io_sq_data { |
238 | refcount_t refs; | |
69fb2131 JA |
239 | struct mutex lock; |
240 | ||
241 | /* ctx's that are using this sqd */ | |
242 | struct list_head ctx_list; | |
243 | struct list_head ctx_new_list; | |
244 | struct mutex ctx_lock; | |
245 | ||
534ca6d6 JA |
246 | struct task_struct *thread; |
247 | struct wait_queue_head wait; | |
248 | }; | |
249 | ||
2b188cc1 JA |
250 | struct io_ring_ctx { |
251 | struct { | |
252 | struct percpu_ref refs; | |
253 | } ____cacheline_aligned_in_smp; | |
254 | ||
255 | struct { | |
256 | unsigned int flags; | |
e1d85334 | 257 | unsigned int compat: 1; |
aad5d8da | 258 | unsigned int limit_mem: 1; |
e1d85334 RD |
259 | unsigned int cq_overflow_flushed: 1; |
260 | unsigned int drain_next: 1; | |
261 | unsigned int eventfd_async: 1; | |
21b55dbc | 262 | unsigned int restricted: 1; |
2b188cc1 | 263 | |
75b28aff HV |
264 | /* |
265 | * Ring buffer of indices into array of io_uring_sqe, which is | |
266 | * mmapped by the application using the IORING_OFF_SQES offset. | |
267 | * | |
268 | * This indirection could e.g. be used to assign fixed | |
269 | * io_uring_sqe entries to operations and only submit them to | |
270 | * the queue when needed. | |
271 | * | |
272 | * The kernel modifies neither the indices array nor the entries | |
273 | * array. | |
274 | */ | |
275 | u32 *sq_array; | |
2b188cc1 JA |
276 | unsigned cached_sq_head; |
277 | unsigned sq_entries; | |
278 | unsigned sq_mask; | |
6c271ce2 | 279 | unsigned sq_thread_idle; |
498ccd9e | 280 | unsigned cached_sq_dropped; |
2c3bac6d | 281 | unsigned cached_cq_overflow; |
ad3eb2c8 | 282 | unsigned long sq_check_overflow; |
de0617e4 JA |
283 | |
284 | struct list_head defer_list; | |
5262f567 | 285 | struct list_head timeout_list; |
1d7bb1d5 | 286 | struct list_head cq_overflow_list; |
fcb323cc JA |
287 | |
288 | wait_queue_head_t inflight_wait; | |
ad3eb2c8 | 289 | struct io_uring_sqe *sq_sqes; |
2b188cc1 JA |
290 | } ____cacheline_aligned_in_smp; |
291 | ||
206aefde JA |
292 | struct io_rings *rings; |
293 | ||
2b188cc1 | 294 | /* IO offload */ |
561fb04a | 295 | struct io_wq *io_wq; |
2aede0e4 JA |
296 | |
297 | /* | |
298 | * For SQPOLL usage - we hold a reference to the parent task, so we | |
299 | * have access to the ->files | |
300 | */ | |
301 | struct task_struct *sqo_task; | |
302 | ||
303 | /* Only used for accounting purposes */ | |
304 | struct mm_struct *mm_account; | |
305 | ||
91d8f519 DZ |
306 | #ifdef CONFIG_BLK_CGROUP |
307 | struct cgroup_subsys_state *sqo_blkcg_css; | |
308 | #endif | |
309 | ||
534ca6d6 JA |
310 | struct io_sq_data *sq_data; /* if using sq thread polling */ |
311 | ||
90554200 | 312 | struct wait_queue_head sqo_sq_wait; |
6a779382 | 313 | struct wait_queue_entry sqo_wait_entry; |
69fb2131 | 314 | struct list_head sqd_list; |
75b28aff | 315 | |
6b06314c JA |
316 | /* |
317 | * If used, fixed file set. Writers must ensure that ->refs is dead, | |
318 | * readers must ensure that ->refs is alive as long as the file* is | |
319 | * used. Only updated through io_uring_register(2). | |
320 | */ | |
05f3fb3c | 321 | struct fixed_file_data *file_data; |
6b06314c JA |
322 | unsigned nr_user_files; |
323 | ||
edafccee JA |
324 | /* if used, fixed mapped user buffers */ |
325 | unsigned nr_user_bufs; | |
326 | struct io_mapped_ubuf *user_bufs; | |
327 | ||
2b188cc1 JA |
328 | struct user_struct *user; |
329 | ||
0b8c0ec7 | 330 | const struct cred *creds; |
181e448d | 331 | |
4ea33a97 JA |
332 | #ifdef CONFIG_AUDIT |
333 | kuid_t loginuid; | |
334 | unsigned int sessionid; | |
335 | #endif | |
336 | ||
0f158b4c JA |
337 | struct completion ref_comp; |
338 | struct completion sq_thread_comp; | |
206aefde | 339 | |
0ddf92e8 JA |
340 | /* if all else fails... */ |
341 | struct io_kiocb *fallback_req; | |
342 | ||
206aefde JA |
343 | #if defined(CONFIG_UNIX) |
344 | struct socket *ring_sock; | |
345 | #endif | |
346 | ||
5a2e745d JA |
347 | struct idr io_buffer_idr; |
348 | ||
071698e1 JA |
349 | struct idr personality_idr; |
350 | ||
206aefde JA |
351 | struct { |
352 | unsigned cached_cq_tail; | |
353 | unsigned cq_entries; | |
354 | unsigned cq_mask; | |
355 | atomic_t cq_timeouts; | |
ad3eb2c8 | 356 | unsigned long cq_check_overflow; |
206aefde JA |
357 | struct wait_queue_head cq_wait; |
358 | struct fasync_struct *cq_fasync; | |
359 | struct eventfd_ctx *cq_ev_fd; | |
360 | } ____cacheline_aligned_in_smp; | |
2b188cc1 JA |
361 | |
362 | struct { | |
363 | struct mutex uring_lock; | |
364 | wait_queue_head_t wait; | |
365 | } ____cacheline_aligned_in_smp; | |
366 | ||
367 | struct { | |
368 | spinlock_t completion_lock; | |
e94f141b | 369 | |
def596e9 | 370 | /* |
540e32a0 | 371 | * ->iopoll_list is protected by the ctx->uring_lock for |
def596e9 JA |
372 | * io_uring instances that don't use IORING_SETUP_SQPOLL. |
373 | * For SQPOLL, only the single threaded io_sq_thread() will | |
374 | * manipulate the list, hence no extra locking is needed there. | |
375 | */ | |
540e32a0 | 376 | struct list_head iopoll_list; |
78076bb6 JA |
377 | struct hlist_head *cancel_hash; |
378 | unsigned cancel_hash_bits; | |
e94f141b | 379 | bool poll_multi_file; |
31b51510 | 380 | |
fcb323cc JA |
381 | spinlock_t inflight_lock; |
382 | struct list_head inflight_list; | |
2b188cc1 | 383 | } ____cacheline_aligned_in_smp; |
85faa7b8 | 384 | |
4a38aed2 JA |
385 | struct delayed_work file_put_work; |
386 | struct llist_head file_put_llist; | |
387 | ||
85faa7b8 | 388 | struct work_struct exit_work; |
21b55dbc | 389 | struct io_restriction restrictions; |
2b188cc1 JA |
390 | }; |
391 | ||
09bb8394 JA |
392 | /* |
393 | * First field must be the file pointer in all the | |
394 | * iocb unions! See also 'struct kiocb' in <linux/fs.h> | |
395 | */ | |
221c5eb2 JA |
396 | struct io_poll_iocb { |
397 | struct file *file; | |
0969e783 JA |
398 | union { |
399 | struct wait_queue_head *head; | |
400 | u64 addr; | |
401 | }; | |
221c5eb2 | 402 | __poll_t events; |
8c838788 | 403 | bool done; |
221c5eb2 | 404 | bool canceled; |
392edb45 | 405 | struct wait_queue_entry wait; |
221c5eb2 JA |
406 | }; |
407 | ||
b5dba59e JA |
408 | struct io_close { |
409 | struct file *file; | |
410 | struct file *put_file; | |
411 | int fd; | |
412 | }; | |
413 | ||
ad8a48ac JA |
414 | struct io_timeout_data { |
415 | struct io_kiocb *req; | |
416 | struct hrtimer timer; | |
417 | struct timespec64 ts; | |
418 | enum hrtimer_mode mode; | |
419 | }; | |
420 | ||
8ed8d3c3 JA |
421 | struct io_accept { |
422 | struct file *file; | |
423 | struct sockaddr __user *addr; | |
424 | int __user *addr_len; | |
425 | int flags; | |
09952e3e | 426 | unsigned long nofile; |
8ed8d3c3 JA |
427 | }; |
428 | ||
429 | struct io_sync { | |
430 | struct file *file; | |
431 | loff_t len; | |
432 | loff_t off; | |
433 | int flags; | |
d63d1b5e | 434 | int mode; |
8ed8d3c3 JA |
435 | }; |
436 | ||
fbf23849 JA |
437 | struct io_cancel { |
438 | struct file *file; | |
439 | u64 addr; | |
440 | }; | |
441 | ||
b29472ee JA |
442 | struct io_timeout { |
443 | struct file *file; | |
bfe68a22 PB |
444 | u32 off; |
445 | u32 target_seq; | |
135fcde8 | 446 | struct list_head list; |
b29472ee JA |
447 | }; |
448 | ||
0bdf7a2d PB |
449 | struct io_timeout_rem { |
450 | struct file *file; | |
451 | u64 addr; | |
452 | }; | |
453 | ||
9adbd45d JA |
454 | struct io_rw { |
455 | /* NOTE: kiocb has the file as the first member, so don't do it here */ | |
456 | struct kiocb kiocb; | |
457 | u64 addr; | |
458 | u64 len; | |
459 | }; | |
460 | ||
3fbb51c1 JA |
461 | struct io_connect { |
462 | struct file *file; | |
463 | struct sockaddr __user *addr; | |
464 | int addr_len; | |
465 | }; | |
466 | ||
e47293fd JA |
467 | struct io_sr_msg { |
468 | struct file *file; | |
fddaface | 469 | union { |
270a5940 | 470 | struct user_msghdr __user *umsg; |
fddaface JA |
471 | void __user *buf; |
472 | }; | |
e47293fd | 473 | int msg_flags; |
bcda7baa | 474 | int bgid; |
fddaface | 475 | size_t len; |
bcda7baa | 476 | struct io_buffer *kbuf; |
e47293fd JA |
477 | }; |
478 | ||
15b71abe JA |
479 | struct io_open { |
480 | struct file *file; | |
481 | int dfd; | |
944d1444 | 482 | bool ignore_nonblock; |
15b71abe | 483 | struct filename *filename; |
c12cedf2 | 484 | struct open_how how; |
4022e7af | 485 | unsigned long nofile; |
15b71abe JA |
486 | }; |
487 | ||
05f3fb3c JA |
488 | struct io_files_update { |
489 | struct file *file; | |
490 | u64 arg; | |
491 | u32 nr_args; | |
492 | u32 offset; | |
493 | }; | |
494 | ||
4840e418 JA |
495 | struct io_fadvise { |
496 | struct file *file; | |
497 | u64 offset; | |
498 | u32 len; | |
499 | u32 advice; | |
500 | }; | |
501 | ||
c1ca757b JA |
502 | struct io_madvise { |
503 | struct file *file; | |
504 | u64 addr; | |
505 | u32 len; | |
506 | u32 advice; | |
507 | }; | |
508 | ||
3e4827b0 JA |
509 | struct io_epoll { |
510 | struct file *file; | |
511 | int epfd; | |
512 | int op; | |
513 | int fd; | |
514 | struct epoll_event event; | |
e47293fd JA |
515 | }; |
516 | ||
7d67af2c PB |
517 | struct io_splice { |
518 | struct file *file_out; | |
519 | struct file *file_in; | |
520 | loff_t off_out; | |
521 | loff_t off_in; | |
522 | u64 len; | |
523 | unsigned int flags; | |
524 | }; | |
525 | ||
ddf0322d JA |
526 | struct io_provide_buf { |
527 | struct file *file; | |
528 | __u64 addr; | |
529 | __s32 len; | |
530 | __u32 bgid; | |
531 | __u16 nbufs; | |
532 | __u16 bid; | |
533 | }; | |
534 | ||
1d9e1288 BM |
535 | struct io_statx { |
536 | struct file *file; | |
537 | int dfd; | |
538 | unsigned int mask; | |
539 | unsigned int flags; | |
e62753e4 | 540 | const char __user *filename; |
1d9e1288 BM |
541 | struct statx __user *buffer; |
542 | }; | |
543 | ||
36f4fa68 JA |
544 | struct io_shutdown { |
545 | struct file *file; | |
546 | int how; | |
547 | }; | |
548 | ||
3ca405eb PB |
549 | struct io_completion { |
550 | struct file *file; | |
551 | struct list_head list; | |
0f7e466b | 552 | int cflags; |
3ca405eb PB |
553 | }; |
554 | ||
f499a021 JA |
555 | struct io_async_connect { |
556 | struct sockaddr_storage address; | |
557 | }; | |
558 | ||
03b1230c JA |
559 | struct io_async_msghdr { |
560 | struct iovec fast_iov[UIO_FASTIOV]; | |
561 | struct iovec *iov; | |
562 | struct sockaddr __user *uaddr; | |
563 | struct msghdr msg; | |
b537916c | 564 | struct sockaddr_storage addr; |
03b1230c JA |
565 | }; |
566 | ||
f67676d1 JA |
567 | struct io_async_rw { |
568 | struct iovec fast_iov[UIO_FASTIOV]; | |
ff6165b2 JA |
569 | const struct iovec *free_iovec; |
570 | struct iov_iter iter; | |
227c0c96 | 571 | size_t bytes_done; |
bcf5a063 | 572 | struct wait_page_queue wpq; |
f67676d1 JA |
573 | }; |
574 | ||
6b47ee6e PB |
575 | enum { |
576 | REQ_F_FIXED_FILE_BIT = IOSQE_FIXED_FILE_BIT, | |
577 | REQ_F_IO_DRAIN_BIT = IOSQE_IO_DRAIN_BIT, | |
578 | REQ_F_LINK_BIT = IOSQE_IO_LINK_BIT, | |
579 | REQ_F_HARDLINK_BIT = IOSQE_IO_HARDLINK_BIT, | |
580 | REQ_F_FORCE_ASYNC_BIT = IOSQE_ASYNC_BIT, | |
bcda7baa | 581 | REQ_F_BUFFER_SELECT_BIT = IOSQE_BUFFER_SELECT_BIT, |
6b47ee6e | 582 | |
dea3b49c | 583 | REQ_F_LINK_HEAD_BIT, |
6b47ee6e PB |
584 | REQ_F_FAIL_LINK_BIT, |
585 | REQ_F_INFLIGHT_BIT, | |
586 | REQ_F_CUR_POS_BIT, | |
587 | REQ_F_NOWAIT_BIT, | |
6b47ee6e | 588 | REQ_F_LINK_TIMEOUT_BIT, |
6b47ee6e | 589 | REQ_F_ISREG_BIT, |
99bc4c38 | 590 | REQ_F_NEED_CLEANUP_BIT, |
d7718a9d | 591 | REQ_F_POLLED_BIT, |
bcda7baa | 592 | REQ_F_BUFFER_SELECTED_BIT, |
5b0bbee4 | 593 | REQ_F_NO_FILE_TABLE_BIT, |
7cdaf587 | 594 | REQ_F_WORK_INITIALIZED_BIT, |
900fad45 | 595 | REQ_F_LTIMEOUT_ACTIVE_BIT, |
84557871 JA |
596 | |
597 | /* not a real bit, just to check we're not overflowing the space */ | |
598 | __REQ_F_LAST_BIT, | |
6b47ee6e PB |
599 | }; |
600 | ||
601 | enum { | |
602 | /* ctx owns file */ | |
603 | REQ_F_FIXED_FILE = BIT(REQ_F_FIXED_FILE_BIT), | |
604 | /* drain existing IO first */ | |
605 | REQ_F_IO_DRAIN = BIT(REQ_F_IO_DRAIN_BIT), | |
606 | /* linked sqes */ | |
607 | REQ_F_LINK = BIT(REQ_F_LINK_BIT), | |
608 | /* doesn't sever on completion < 0 */ | |
609 | REQ_F_HARDLINK = BIT(REQ_F_HARDLINK_BIT), | |
610 | /* IOSQE_ASYNC */ | |
611 | REQ_F_FORCE_ASYNC = BIT(REQ_F_FORCE_ASYNC_BIT), | |
bcda7baa JA |
612 | /* IOSQE_BUFFER_SELECT */ |
613 | REQ_F_BUFFER_SELECT = BIT(REQ_F_BUFFER_SELECT_BIT), | |
6b47ee6e | 614 | |
dea3b49c PB |
615 | /* head of a link */ |
616 | REQ_F_LINK_HEAD = BIT(REQ_F_LINK_HEAD_BIT), | |
6b47ee6e PB |
617 | /* fail rest of links */ |
618 | REQ_F_FAIL_LINK = BIT(REQ_F_FAIL_LINK_BIT), | |
619 | /* on inflight list */ | |
620 | REQ_F_INFLIGHT = BIT(REQ_F_INFLIGHT_BIT), | |
621 | /* read/write uses file position */ | |
622 | REQ_F_CUR_POS = BIT(REQ_F_CUR_POS_BIT), | |
623 | /* must not punt to workers */ | |
624 | REQ_F_NOWAIT = BIT(REQ_F_NOWAIT_BIT), | |
900fad45 | 625 | /* has or had linked timeout */ |
6b47ee6e | 626 | REQ_F_LINK_TIMEOUT = BIT(REQ_F_LINK_TIMEOUT_BIT), |
6b47ee6e PB |
627 | /* regular file */ |
628 | REQ_F_ISREG = BIT(REQ_F_ISREG_BIT), | |
99bc4c38 PB |
629 | /* needs cleanup */ |
630 | REQ_F_NEED_CLEANUP = BIT(REQ_F_NEED_CLEANUP_BIT), | |
d7718a9d JA |
631 | /* already went through poll handler */ |
632 | REQ_F_POLLED = BIT(REQ_F_POLLED_BIT), | |
bcda7baa JA |
633 | /* buffer already selected */ |
634 | REQ_F_BUFFER_SELECTED = BIT(REQ_F_BUFFER_SELECTED_BIT), | |
5b0bbee4 JA |
635 | /* doesn't need file table for this request */ |
636 | REQ_F_NO_FILE_TABLE = BIT(REQ_F_NO_FILE_TABLE_BIT), | |
7cdaf587 XW |
637 | /* io_wq_work is initialized */ |
638 | REQ_F_WORK_INITIALIZED = BIT(REQ_F_WORK_INITIALIZED_BIT), | |
900fad45 PB |
639 | /* linked timeout is active, i.e. prepared by link's head */ |
640 | REQ_F_LTIMEOUT_ACTIVE = BIT(REQ_F_LTIMEOUT_ACTIVE_BIT), | |
d7718a9d JA |
641 | }; |
642 | ||
643 | struct async_poll { | |
644 | struct io_poll_iocb poll; | |
807abcb0 | 645 | struct io_poll_iocb *double_poll; |
6b47ee6e PB |
646 | }; |
647 | ||
09bb8394 JA |
648 | /* |
649 | * NOTE! Each of the iocb union members has the file pointer | |
650 | * as the first entry in their struct definition. So you can | |
651 | * access the file pointer through any of the sub-structs, | |
652 | * or directly as just 'ki_filp' in this struct. | |
653 | */ | |
2b188cc1 | 654 | struct io_kiocb { |
221c5eb2 | 655 | union { |
09bb8394 | 656 | struct file *file; |
9adbd45d | 657 | struct io_rw rw; |
221c5eb2 | 658 | struct io_poll_iocb poll; |
8ed8d3c3 JA |
659 | struct io_accept accept; |
660 | struct io_sync sync; | |
fbf23849 | 661 | struct io_cancel cancel; |
b29472ee | 662 | struct io_timeout timeout; |
0bdf7a2d | 663 | struct io_timeout_rem timeout_rem; |
3fbb51c1 | 664 | struct io_connect connect; |
e47293fd | 665 | struct io_sr_msg sr_msg; |
15b71abe | 666 | struct io_open open; |
b5dba59e | 667 | struct io_close close; |
05f3fb3c | 668 | struct io_files_update files_update; |
4840e418 | 669 | struct io_fadvise fadvise; |
c1ca757b | 670 | struct io_madvise madvise; |
3e4827b0 | 671 | struct io_epoll epoll; |
7d67af2c | 672 | struct io_splice splice; |
ddf0322d | 673 | struct io_provide_buf pbuf; |
1d9e1288 | 674 | struct io_statx statx; |
36f4fa68 | 675 | struct io_shutdown shutdown; |
3ca405eb PB |
676 | /* use only after cleaning per-op data, see io_clean_op() */ |
677 | struct io_completion compl; | |
221c5eb2 | 678 | }; |
2b188cc1 | 679 | |
e8c2bc1f JA |
680 | /* opcode allocated if it needs to store data for async defer */ |
681 | void *async_data; | |
d625c6ee | 682 | u8 opcode; |
65a6543d XW |
683 | /* polled IO has completed */ |
684 | u8 iopoll_completed; | |
2b188cc1 | 685 | |
4f4eeba8 | 686 | u16 buf_index; |
9cf7c104 | 687 | u32 result; |
4f4eeba8 | 688 | |
010e8e6b PB |
689 | struct io_ring_ctx *ctx; |
690 | unsigned int flags; | |
691 | refcount_t refs; | |
692 | struct task_struct *task; | |
693 | u64 user_data; | |
d7718a9d | 694 | |
010e8e6b | 695 | struct list_head link_list; |
fcb323cc | 696 | |
d21ffe7e PB |
697 | /* |
698 | * 1. used with ctx->iopoll_list with reads/writes | |
699 | * 2. to track reqs with ->files (see io_op_def::file_table) | |
700 | */ | |
010e8e6b PB |
701 | struct list_head inflight_entry; |
702 | ||
703 | struct percpu_ref *fixed_file_refs; | |
704 | struct callback_head task_work; | |
705 | /* for polled requests, i.e. IORING_OP_POLL_ADD and async armed poll */ | |
706 | struct hlist_node hash_node; | |
707 | struct async_poll *apoll; | |
708 | struct io_wq_work work; | |
2b188cc1 | 709 | }; |
05589553 | 710 | |
27dc8338 PB |
711 | struct io_defer_entry { |
712 | struct list_head list; | |
713 | struct io_kiocb *req; | |
9cf7c104 | 714 | u32 seq; |
2b188cc1 JA |
715 | }; |
716 | ||
def596e9 | 717 | #define IO_IOPOLL_BATCH 8 |
2b188cc1 | 718 | |
013538bd JA |
719 | struct io_comp_state { |
720 | unsigned int nr; | |
721 | struct list_head list; | |
722 | struct io_ring_ctx *ctx; | |
723 | }; | |
724 | ||
9a56a232 JA |
725 | struct io_submit_state { |
726 | struct blk_plug plug; | |
727 | ||
2579f913 JA |
728 | /* |
729 | * io_kiocb alloc cache | |
730 | */ | |
731 | void *reqs[IO_IOPOLL_BATCH]; | |
6c8a3134 | 732 | unsigned int free_reqs; |
2579f913 | 733 | |
013538bd JA |
734 | /* |
735 | * Batch completion logic | |
736 | */ | |
737 | struct io_comp_state comp; | |
738 | ||
9a56a232 JA |
739 | /* |
740 | * File reference cache | |
741 | */ | |
742 | struct file *file; | |
743 | unsigned int fd; | |
744 | unsigned int has_refs; | |
9a56a232 JA |
745 | unsigned int ios_left; |
746 | }; | |
747 | ||
d3656344 | 748 | struct io_op_def { |
d3656344 JA |
749 | /* needs req->file assigned */ |
750 | unsigned needs_file : 1; | |
fd2206e4 JA |
751 | /* don't fail if file grab fails */ |
752 | unsigned needs_file_no_error : 1; | |
d3656344 JA |
753 | /* hash wq insertion if file is a regular file */ |
754 | unsigned hash_reg_file : 1; | |
755 | /* unbound wq insertion if file is a non-regular file */ | |
756 | unsigned unbound_nonreg_file : 1; | |
66f4af93 JA |
757 | /* opcode is not supported by this kernel */ |
758 | unsigned not_supported : 1; | |
8a72758c JA |
759 | /* set if opcode supports polled "wait" */ |
760 | unsigned pollin : 1; | |
761 | unsigned pollout : 1; | |
bcda7baa JA |
762 | /* op supports buffer selection */ |
763 | unsigned buffer_select : 1; | |
e8c2bc1f JA |
764 | /* must always have async data allocated */ |
765 | unsigned needs_async_data : 1; | |
766 | /* size of async data needed, if any */ | |
767 | unsigned short async_size; | |
0f203765 | 768 | unsigned work_flags; |
d3656344 JA |
769 | }; |
770 | ||
0918682b | 771 | static const struct io_op_def io_op_defs[] = { |
0463b6c5 PB |
772 | [IORING_OP_NOP] = {}, |
773 | [IORING_OP_READV] = { | |
d3656344 JA |
774 | .needs_file = 1, |
775 | .unbound_nonreg_file = 1, | |
8a72758c | 776 | .pollin = 1, |
4d954c25 | 777 | .buffer_select = 1, |
e8c2bc1f JA |
778 | .needs_async_data = 1, |
779 | .async_size = sizeof(struct io_async_rw), | |
0f203765 | 780 | .work_flags = IO_WQ_WORK_MM | IO_WQ_WORK_BLKCG, |
d3656344 | 781 | }, |
0463b6c5 | 782 | [IORING_OP_WRITEV] = { |
d3656344 JA |
783 | .needs_file = 1, |
784 | .hash_reg_file = 1, | |
785 | .unbound_nonreg_file = 1, | |
8a72758c | 786 | .pollout = 1, |
e8c2bc1f JA |
787 | .needs_async_data = 1, |
788 | .async_size = sizeof(struct io_async_rw), | |
69228338 JA |
789 | .work_flags = IO_WQ_WORK_MM | IO_WQ_WORK_BLKCG | |
790 | IO_WQ_WORK_FSIZE, | |
d3656344 | 791 | }, |
0463b6c5 | 792 | [IORING_OP_FSYNC] = { |
d3656344 | 793 | .needs_file = 1, |
0f203765 | 794 | .work_flags = IO_WQ_WORK_BLKCG, |
d3656344 | 795 | }, |
0463b6c5 | 796 | [IORING_OP_READ_FIXED] = { |
d3656344 JA |
797 | .needs_file = 1, |
798 | .unbound_nonreg_file = 1, | |
8a72758c | 799 | .pollin = 1, |
e8c2bc1f | 800 | .async_size = sizeof(struct io_async_rw), |
4017eb91 | 801 | .work_flags = IO_WQ_WORK_BLKCG | IO_WQ_WORK_MM, |
d3656344 | 802 | }, |
0463b6c5 | 803 | [IORING_OP_WRITE_FIXED] = { |
d3656344 JA |
804 | .needs_file = 1, |
805 | .hash_reg_file = 1, | |
806 | .unbound_nonreg_file = 1, | |
8a72758c | 807 | .pollout = 1, |
e8c2bc1f | 808 | .async_size = sizeof(struct io_async_rw), |
4017eb91 JA |
809 | .work_flags = IO_WQ_WORK_BLKCG | IO_WQ_WORK_FSIZE | |
810 | IO_WQ_WORK_MM, | |
d3656344 | 811 | }, |
0463b6c5 | 812 | [IORING_OP_POLL_ADD] = { |
d3656344 JA |
813 | .needs_file = 1, |
814 | .unbound_nonreg_file = 1, | |
815 | }, | |
0463b6c5 PB |
816 | [IORING_OP_POLL_REMOVE] = {}, |
817 | [IORING_OP_SYNC_FILE_RANGE] = { | |
d3656344 | 818 | .needs_file = 1, |
0f203765 | 819 | .work_flags = IO_WQ_WORK_BLKCG, |
d3656344 | 820 | }, |
0463b6c5 | 821 | [IORING_OP_SENDMSG] = { |
d3656344 JA |
822 | .needs_file = 1, |
823 | .unbound_nonreg_file = 1, | |
8a72758c | 824 | .pollout = 1, |
e8c2bc1f JA |
825 | .needs_async_data = 1, |
826 | .async_size = sizeof(struct io_async_msghdr), | |
0f203765 JA |
827 | .work_flags = IO_WQ_WORK_MM | IO_WQ_WORK_BLKCG | |
828 | IO_WQ_WORK_FS, | |
d3656344 | 829 | }, |
0463b6c5 | 830 | [IORING_OP_RECVMSG] = { |
d3656344 JA |
831 | .needs_file = 1, |
832 | .unbound_nonreg_file = 1, | |
8a72758c | 833 | .pollin = 1, |
52de1fe1 | 834 | .buffer_select = 1, |
e8c2bc1f JA |
835 | .needs_async_data = 1, |
836 | .async_size = sizeof(struct io_async_msghdr), | |
0f203765 JA |
837 | .work_flags = IO_WQ_WORK_MM | IO_WQ_WORK_BLKCG | |
838 | IO_WQ_WORK_FS, | |
d3656344 | 839 | }, |
0463b6c5 | 840 | [IORING_OP_TIMEOUT] = { |
e8c2bc1f JA |
841 | .needs_async_data = 1, |
842 | .async_size = sizeof(struct io_timeout_data), | |
0f203765 | 843 | .work_flags = IO_WQ_WORK_MM, |
d3656344 | 844 | }, |
0463b6c5 PB |
845 | [IORING_OP_TIMEOUT_REMOVE] = {}, |
846 | [IORING_OP_ACCEPT] = { | |
d3656344 JA |
847 | .needs_file = 1, |
848 | .unbound_nonreg_file = 1, | |
8a72758c | 849 | .pollin = 1, |
0f203765 | 850 | .work_flags = IO_WQ_WORK_MM | IO_WQ_WORK_FILES, |
d3656344 | 851 | }, |
0463b6c5 PB |
852 | [IORING_OP_ASYNC_CANCEL] = {}, |
853 | [IORING_OP_LINK_TIMEOUT] = { | |
e8c2bc1f JA |
854 | .needs_async_data = 1, |
855 | .async_size = sizeof(struct io_timeout_data), | |
0f203765 | 856 | .work_flags = IO_WQ_WORK_MM, |
d3656344 | 857 | }, |
0463b6c5 | 858 | [IORING_OP_CONNECT] = { |
d3656344 JA |
859 | .needs_file = 1, |
860 | .unbound_nonreg_file = 1, | |
8a72758c | 861 | .pollout = 1, |
e8c2bc1f JA |
862 | .needs_async_data = 1, |
863 | .async_size = sizeof(struct io_async_connect), | |
0f203765 | 864 | .work_flags = IO_WQ_WORK_MM, |
d3656344 | 865 | }, |
0463b6c5 | 866 | [IORING_OP_FALLOCATE] = { |
d3656344 | 867 | .needs_file = 1, |
69228338 | 868 | .work_flags = IO_WQ_WORK_BLKCG | IO_WQ_WORK_FSIZE, |
d3656344 | 869 | }, |
0463b6c5 | 870 | [IORING_OP_OPENAT] = { |
0f203765 JA |
871 | .work_flags = IO_WQ_WORK_FILES | IO_WQ_WORK_BLKCG | |
872 | IO_WQ_WORK_FS, | |
d3656344 | 873 | }, |
0463b6c5 | 874 | [IORING_OP_CLOSE] = { |
fd2206e4 JA |
875 | .needs_file = 1, |
876 | .needs_file_no_error = 1, | |
0f203765 | 877 | .work_flags = IO_WQ_WORK_FILES | IO_WQ_WORK_BLKCG, |
d3656344 | 878 | }, |
0463b6c5 | 879 | [IORING_OP_FILES_UPDATE] = { |
0f203765 | 880 | .work_flags = IO_WQ_WORK_FILES | IO_WQ_WORK_MM, |
d3656344 | 881 | }, |
0463b6c5 | 882 | [IORING_OP_STATX] = { |
0f203765 JA |
883 | .work_flags = IO_WQ_WORK_FILES | IO_WQ_WORK_MM | |
884 | IO_WQ_WORK_FS | IO_WQ_WORK_BLKCG, | |
d3656344 | 885 | }, |
0463b6c5 | 886 | [IORING_OP_READ] = { |
3a6820f2 JA |
887 | .needs_file = 1, |
888 | .unbound_nonreg_file = 1, | |
8a72758c | 889 | .pollin = 1, |
bcda7baa | 890 | .buffer_select = 1, |
e8c2bc1f | 891 | .async_size = sizeof(struct io_async_rw), |
0f203765 | 892 | .work_flags = IO_WQ_WORK_MM | IO_WQ_WORK_BLKCG, |
3a6820f2 | 893 | }, |
0463b6c5 | 894 | [IORING_OP_WRITE] = { |
3a6820f2 JA |
895 | .needs_file = 1, |
896 | .unbound_nonreg_file = 1, | |
8a72758c | 897 | .pollout = 1, |
e8c2bc1f | 898 | .async_size = sizeof(struct io_async_rw), |
69228338 JA |
899 | .work_flags = IO_WQ_WORK_MM | IO_WQ_WORK_BLKCG | |
900 | IO_WQ_WORK_FSIZE, | |
3a6820f2 | 901 | }, |
0463b6c5 | 902 | [IORING_OP_FADVISE] = { |
4840e418 | 903 | .needs_file = 1, |
0f203765 | 904 | .work_flags = IO_WQ_WORK_BLKCG, |
4840e418 | 905 | }, |
0463b6c5 | 906 | [IORING_OP_MADVISE] = { |
0f203765 | 907 | .work_flags = IO_WQ_WORK_MM | IO_WQ_WORK_BLKCG, |
c1ca757b | 908 | }, |
0463b6c5 | 909 | [IORING_OP_SEND] = { |
fddaface JA |
910 | .needs_file = 1, |
911 | .unbound_nonreg_file = 1, | |
8a72758c | 912 | .pollout = 1, |
0f203765 | 913 | .work_flags = IO_WQ_WORK_MM | IO_WQ_WORK_BLKCG, |
fddaface | 914 | }, |
0463b6c5 | 915 | [IORING_OP_RECV] = { |
fddaface JA |
916 | .needs_file = 1, |
917 | .unbound_nonreg_file = 1, | |
8a72758c | 918 | .pollin = 1, |
bcda7baa | 919 | .buffer_select = 1, |
0f203765 | 920 | .work_flags = IO_WQ_WORK_MM | IO_WQ_WORK_BLKCG, |
fddaface | 921 | }, |
0463b6c5 | 922 | [IORING_OP_OPENAT2] = { |
0f203765 JA |
923 | .work_flags = IO_WQ_WORK_FILES | IO_WQ_WORK_FS | |
924 | IO_WQ_WORK_BLKCG, | |
cebdb986 | 925 | }, |
3e4827b0 JA |
926 | [IORING_OP_EPOLL_CTL] = { |
927 | .unbound_nonreg_file = 1, | |
0f203765 | 928 | .work_flags = IO_WQ_WORK_FILES, |
3e4827b0 | 929 | }, |
7d67af2c PB |
930 | [IORING_OP_SPLICE] = { |
931 | .needs_file = 1, | |
932 | .hash_reg_file = 1, | |
933 | .unbound_nonreg_file = 1, | |
0f203765 | 934 | .work_flags = IO_WQ_WORK_BLKCG, |
ddf0322d JA |
935 | }, |
936 | [IORING_OP_PROVIDE_BUFFERS] = {}, | |
067524e9 | 937 | [IORING_OP_REMOVE_BUFFERS] = {}, |
f2a8d5c7 PB |
938 | [IORING_OP_TEE] = { |
939 | .needs_file = 1, | |
940 | .hash_reg_file = 1, | |
941 | .unbound_nonreg_file = 1, | |
942 | }, | |
36f4fa68 JA |
943 | [IORING_OP_SHUTDOWN] = { |
944 | .needs_file = 1, | |
945 | }, | |
d3656344 JA |
946 | }; |
947 | ||
2e0464d4 BM |
948 | enum io_mem_account { |
949 | ACCT_LOCKED, | |
950 | ACCT_PINNED, | |
951 | }; | |
952 | ||
81b68a5c PB |
953 | static void __io_complete_rw(struct io_kiocb *req, long res, long res2, |
954 | struct io_comp_state *cs); | |
78e19bbe | 955 | static void io_cqring_fill_event(struct io_kiocb *req, long res); |
ec9c02ad | 956 | static void io_put_req(struct io_kiocb *req); |
216578e5 | 957 | static void io_put_req_deferred(struct io_kiocb *req, int nr); |
c40f6379 | 958 | static void io_double_put_req(struct io_kiocb *req); |
94ae5e77 | 959 | static struct io_kiocb *io_prep_linked_timeout(struct io_kiocb *req); |
7271ef3a | 960 | static void __io_queue_linked_timeout(struct io_kiocb *req); |
94ae5e77 | 961 | static void io_queue_linked_timeout(struct io_kiocb *req); |
05f3fb3c JA |
962 | static int __io_sqe_files_update(struct io_ring_ctx *ctx, |
963 | struct io_uring_files_update *ip, | |
964 | unsigned nr_args); | |
3ca405eb | 965 | static void __io_clean_op(struct io_kiocb *req); |
8371adf5 PB |
966 | static struct file *io_file_get(struct io_submit_state *state, |
967 | struct io_kiocb *req, int fd, bool fixed); | |
c1379e24 | 968 | static void __io_queue_sqe(struct io_kiocb *req, struct io_comp_state *cs); |
4349f30e | 969 | static void io_file_put_work(struct work_struct *work); |
de0617e4 | 970 | |
b63534c4 JA |
971 | static ssize_t io_import_iovec(int rw, struct io_kiocb *req, |
972 | struct iovec **iovec, struct iov_iter *iter, | |
973 | bool needs_lock); | |
ff6165b2 JA |
974 | static int io_setup_async_rw(struct io_kiocb *req, const struct iovec *iovec, |
975 | const struct iovec *fast_iov, | |
227c0c96 | 976 | struct iov_iter *iter, bool force); |
de0617e4 | 977 | |
2b188cc1 JA |
978 | static struct kmem_cache *req_cachep; |
979 | ||
0918682b | 980 | static const struct file_operations io_uring_fops; |
2b188cc1 JA |
981 | |
982 | struct sock *io_uring_get_socket(struct file *file) | |
983 | { | |
984 | #if defined(CONFIG_UNIX) | |
985 | if (file->f_op == &io_uring_fops) { | |
986 | struct io_ring_ctx *ctx = file->private_data; | |
987 | ||
988 | return ctx->ring_sock->sk; | |
989 | } | |
990 | #endif | |
991 | return NULL; | |
992 | } | |
993 | EXPORT_SYMBOL(io_uring_get_socket); | |
994 | ||
3ca405eb PB |
995 | static inline void io_clean_op(struct io_kiocb *req) |
996 | { | |
bb175342 PB |
997 | if (req->flags & (REQ_F_NEED_CLEANUP | REQ_F_BUFFER_SELECTED | |
998 | REQ_F_INFLIGHT)) | |
3ca405eb PB |
999 | __io_clean_op(req); |
1000 | } | |
1001 | ||
4349f30e | 1002 | static void io_sq_thread_drop_mm(void) |
c40f6379 JA |
1003 | { |
1004 | struct mm_struct *mm = current->mm; | |
1005 | ||
1006 | if (mm) { | |
1007 | kthread_unuse_mm(mm); | |
1008 | mmput(mm); | |
4b70cf9d | 1009 | current->mm = NULL; |
c40f6379 JA |
1010 | } |
1011 | } | |
1012 | ||
1013 | static int __io_sq_thread_acquire_mm(struct io_ring_ctx *ctx) | |
1014 | { | |
4b70cf9d JA |
1015 | struct mm_struct *mm; |
1016 | ||
1017 | if (current->mm) | |
1018 | return 0; | |
1019 | ||
1020 | /* Should never happen */ | |
1021 | if (unlikely(!(ctx->flags & IORING_SETUP_SQPOLL))) | |
1022 | return -EFAULT; | |
1023 | ||
1024 | task_lock(ctx->sqo_task); | |
1025 | mm = ctx->sqo_task->mm; | |
1026 | if (unlikely(!mm || !mmget_not_zero(mm))) | |
1027 | mm = NULL; | |
1028 | task_unlock(ctx->sqo_task); | |
1029 | ||
1030 | if (mm) { | |
1031 | kthread_use_mm(mm); | |
1032 | return 0; | |
c40f6379 JA |
1033 | } |
1034 | ||
4b70cf9d | 1035 | return -EFAULT; |
c40f6379 JA |
1036 | } |
1037 | ||
1038 | static int io_sq_thread_acquire_mm(struct io_ring_ctx *ctx, | |
1039 | struct io_kiocb *req) | |
1040 | { | |
0f203765 | 1041 | if (!(io_op_defs[req->opcode].work_flags & IO_WQ_WORK_MM)) |
c40f6379 JA |
1042 | return 0; |
1043 | return __io_sq_thread_acquire_mm(ctx); | |
1044 | } | |
1045 | ||
91d8f519 DZ |
1046 | static void io_sq_thread_associate_blkcg(struct io_ring_ctx *ctx, |
1047 | struct cgroup_subsys_state **cur_css) | |
1048 | ||
1049 | { | |
1050 | #ifdef CONFIG_BLK_CGROUP | |
1051 | /* puts the old one when swapping */ | |
1052 | if (*cur_css != ctx->sqo_blkcg_css) { | |
1053 | kthread_associate_blkcg(ctx->sqo_blkcg_css); | |
1054 | *cur_css = ctx->sqo_blkcg_css; | |
1055 | } | |
1056 | #endif | |
1057 | } | |
1058 | ||
1059 | static void io_sq_thread_unassociate_blkcg(void) | |
1060 | { | |
1061 | #ifdef CONFIG_BLK_CGROUP | |
1062 | kthread_associate_blkcg(NULL); | |
1063 | #endif | |
1064 | } | |
1065 | ||
c40f6379 JA |
1066 | static inline void req_set_fail_links(struct io_kiocb *req) |
1067 | { | |
1068 | if ((req->flags & (REQ_F_LINK | REQ_F_HARDLINK)) == REQ_F_LINK) | |
1069 | req->flags |= REQ_F_FAIL_LINK; | |
1070 | } | |
4a38aed2 | 1071 | |
1e6fa521 JA |
1072 | /* |
1073 | * None of these are dereferenced, they are simply used to check if any of | |
1074 | * them have changed. If we're under current and check they are still the | |
1075 | * same, we're fine to grab references to them for actual out-of-line use. | |
1076 | */ | |
1077 | static void io_init_identity(struct io_identity *id) | |
1078 | { | |
1079 | id->files = current->files; | |
1080 | id->mm = current->mm; | |
1081 | #ifdef CONFIG_BLK_CGROUP | |
1082 | rcu_read_lock(); | |
1083 | id->blkcg_css = blkcg_css(); | |
1084 | rcu_read_unlock(); | |
1085 | #endif | |
1086 | id->creds = current_cred(); | |
1087 | id->nsproxy = current->nsproxy; | |
1088 | id->fs = current->fs; | |
1089 | id->fsize = rlimit(RLIMIT_FSIZE); | |
4ea33a97 JA |
1090 | #ifdef CONFIG_AUDIT |
1091 | id->loginuid = current->loginuid; | |
1092 | id->sessionid = current->sessionid; | |
1093 | #endif | |
1e6fa521 JA |
1094 | refcount_set(&id->count, 1); |
1095 | } | |
1096 | ||
ec99ca6c PB |
1097 | static inline void __io_req_init_async(struct io_kiocb *req) |
1098 | { | |
1099 | memset(&req->work, 0, sizeof(req->work)); | |
1100 | req->flags |= REQ_F_WORK_INITIALIZED; | |
1101 | } | |
1102 | ||
7cdaf587 XW |
1103 | /* |
1104 | * Note: must call io_req_init_async() for the first time you | |
1105 | * touch any members of io_wq_work. | |
1106 | */ | |
1107 | static inline void io_req_init_async(struct io_kiocb *req) | |
1108 | { | |
500a373d JA |
1109 | struct io_uring_task *tctx = current->io_uring; |
1110 | ||
7cdaf587 XW |
1111 | if (req->flags & REQ_F_WORK_INITIALIZED) |
1112 | return; | |
1113 | ||
ec99ca6c | 1114 | __io_req_init_async(req); |
500a373d JA |
1115 | |
1116 | /* Grab a ref if this isn't our static identity */ | |
1117 | req->work.identity = tctx->identity; | |
1118 | if (tctx->identity != &tctx->__identity) | |
1119 | refcount_inc(&req->work.identity->count); | |
7cdaf587 XW |
1120 | } |
1121 | ||
0cdaf760 PB |
1122 | static inline bool io_async_submit(struct io_ring_ctx *ctx) |
1123 | { | |
1124 | return ctx->flags & IORING_SETUP_SQPOLL; | |
1125 | } | |
1126 | ||
2b188cc1 JA |
1127 | static void io_ring_ctx_ref_free(struct percpu_ref *ref) |
1128 | { | |
1129 | struct io_ring_ctx *ctx = container_of(ref, struct io_ring_ctx, refs); | |
1130 | ||
0f158b4c | 1131 | complete(&ctx->ref_comp); |
2b188cc1 JA |
1132 | } |
1133 | ||
8eb7e2d0 PB |
1134 | static inline bool io_is_timeout_noseq(struct io_kiocb *req) |
1135 | { | |
1136 | return !req->timeout.off; | |
1137 | } | |
1138 | ||
2b188cc1 JA |
1139 | static struct io_ring_ctx *io_ring_ctx_alloc(struct io_uring_params *p) |
1140 | { | |
1141 | struct io_ring_ctx *ctx; | |
78076bb6 | 1142 | int hash_bits; |
2b188cc1 JA |
1143 | |
1144 | ctx = kzalloc(sizeof(*ctx), GFP_KERNEL); | |
1145 | if (!ctx) | |
1146 | return NULL; | |
1147 | ||
0ddf92e8 JA |
1148 | ctx->fallback_req = kmem_cache_alloc(req_cachep, GFP_KERNEL); |
1149 | if (!ctx->fallback_req) | |
1150 | goto err; | |
1151 | ||
78076bb6 JA |
1152 | /* |
1153 | * Use 5 bits less than the max cq entries, that should give us around | |
1154 | * 32 entries per hash list if totally full and uniformly spread. | |
1155 | */ | |
1156 | hash_bits = ilog2(p->cq_entries); | |
1157 | hash_bits -= 5; | |
1158 | if (hash_bits <= 0) | |
1159 | hash_bits = 1; | |
1160 | ctx->cancel_hash_bits = hash_bits; | |
1161 | ctx->cancel_hash = kmalloc((1U << hash_bits) * sizeof(struct hlist_head), | |
1162 | GFP_KERNEL); | |
1163 | if (!ctx->cancel_hash) | |
1164 | goto err; | |
1165 | __hash_init(ctx->cancel_hash, 1U << hash_bits); | |
1166 | ||
21482896 | 1167 | if (percpu_ref_init(&ctx->refs, io_ring_ctx_ref_free, |
206aefde JA |
1168 | PERCPU_REF_ALLOW_REINIT, GFP_KERNEL)) |
1169 | goto err; | |
2b188cc1 JA |
1170 | |
1171 | ctx->flags = p->flags; | |
90554200 | 1172 | init_waitqueue_head(&ctx->sqo_sq_wait); |
69fb2131 | 1173 | INIT_LIST_HEAD(&ctx->sqd_list); |
2b188cc1 | 1174 | init_waitqueue_head(&ctx->cq_wait); |
1d7bb1d5 | 1175 | INIT_LIST_HEAD(&ctx->cq_overflow_list); |
0f158b4c JA |
1176 | init_completion(&ctx->ref_comp); |
1177 | init_completion(&ctx->sq_thread_comp); | |
5a2e745d | 1178 | idr_init(&ctx->io_buffer_idr); |
071698e1 | 1179 | idr_init(&ctx->personality_idr); |
2b188cc1 JA |
1180 | mutex_init(&ctx->uring_lock); |
1181 | init_waitqueue_head(&ctx->wait); | |
1182 | spin_lock_init(&ctx->completion_lock); | |
540e32a0 | 1183 | INIT_LIST_HEAD(&ctx->iopoll_list); |
de0617e4 | 1184 | INIT_LIST_HEAD(&ctx->defer_list); |
5262f567 | 1185 | INIT_LIST_HEAD(&ctx->timeout_list); |
fcb323cc JA |
1186 | init_waitqueue_head(&ctx->inflight_wait); |
1187 | spin_lock_init(&ctx->inflight_lock); | |
1188 | INIT_LIST_HEAD(&ctx->inflight_list); | |
4a38aed2 JA |
1189 | INIT_DELAYED_WORK(&ctx->file_put_work, io_file_put_work); |
1190 | init_llist_head(&ctx->file_put_llist); | |
2b188cc1 | 1191 | return ctx; |
206aefde | 1192 | err: |
0ddf92e8 JA |
1193 | if (ctx->fallback_req) |
1194 | kmem_cache_free(req_cachep, ctx->fallback_req); | |
78076bb6 | 1195 | kfree(ctx->cancel_hash); |
206aefde JA |
1196 | kfree(ctx); |
1197 | return NULL; | |
2b188cc1 JA |
1198 | } |
1199 | ||
9cf7c104 | 1200 | static bool req_need_defer(struct io_kiocb *req, u32 seq) |
7adf4eaf | 1201 | { |
2bc9930e JA |
1202 | if (unlikely(req->flags & REQ_F_IO_DRAIN)) { |
1203 | struct io_ring_ctx *ctx = req->ctx; | |
a197f664 | 1204 | |
9cf7c104 | 1205 | return seq != ctx->cached_cq_tail |
2c3bac6d | 1206 | + READ_ONCE(ctx->cached_cq_overflow); |
2bc9930e | 1207 | } |
de0617e4 | 1208 | |
9d858b21 | 1209 | return false; |
de0617e4 JA |
1210 | } |
1211 | ||
de0617e4 | 1212 | static void __io_commit_cqring(struct io_ring_ctx *ctx) |
2b188cc1 | 1213 | { |
75b28aff | 1214 | struct io_rings *rings = ctx->rings; |
2b188cc1 | 1215 | |
07910158 PB |
1216 | /* order cqe stores with ring update */ |
1217 | smp_store_release(&rings->cq.tail, ctx->cached_cq_tail); | |
2b188cc1 | 1218 | |
07910158 PB |
1219 | if (wq_has_sleeper(&ctx->cq_wait)) { |
1220 | wake_up_interruptible(&ctx->cq_wait); | |
1221 | kill_fasync(&ctx->cq_fasync, SIGIO, POLL_IN); | |
2b188cc1 JA |
1222 | } |
1223 | } | |
1224 | ||
5c3462cf | 1225 | static void io_put_identity(struct io_uring_task *tctx, struct io_kiocb *req) |
1e6fa521 | 1226 | { |
500a373d | 1227 | if (req->work.identity == &tctx->__identity) |
1e6fa521 JA |
1228 | return; |
1229 | if (refcount_dec_and_test(&req->work.identity->count)) | |
1230 | kfree(req->work.identity); | |
1231 | } | |
1232 | ||
4edf20f9 | 1233 | static void io_req_clean_work(struct io_kiocb *req) |
18d9be1a | 1234 | { |
7cdaf587 | 1235 | if (!(req->flags & REQ_F_WORK_INITIALIZED)) |
4edf20f9 | 1236 | return; |
51a4cc11 JA |
1237 | |
1238 | req->flags &= ~REQ_F_WORK_INITIALIZED; | |
7cdaf587 | 1239 | |
dfead8a8 | 1240 | if (req->work.flags & IO_WQ_WORK_MM) { |
98447d65 | 1241 | mmdrop(req->work.identity->mm); |
dfead8a8 | 1242 | req->work.flags &= ~IO_WQ_WORK_MM; |
cccf0ee8 | 1243 | } |
91d8f519 | 1244 | #ifdef CONFIG_BLK_CGROUP |
dfead8a8 | 1245 | if (req->work.flags & IO_WQ_WORK_BLKCG) { |
98447d65 | 1246 | css_put(req->work.identity->blkcg_css); |
dfead8a8 JA |
1247 | req->work.flags &= ~IO_WQ_WORK_BLKCG; |
1248 | } | |
91d8f519 | 1249 | #endif |
dfead8a8 | 1250 | if (req->work.flags & IO_WQ_WORK_CREDS) { |
98447d65 | 1251 | put_cred(req->work.identity->creds); |
dfead8a8 | 1252 | req->work.flags &= ~IO_WQ_WORK_CREDS; |
cccf0ee8 | 1253 | } |
dfead8a8 | 1254 | if (req->work.flags & IO_WQ_WORK_FS) { |
98447d65 | 1255 | struct fs_struct *fs = req->work.identity->fs; |
51a4cc11 | 1256 | |
98447d65 | 1257 | spin_lock(&req->work.identity->fs->lock); |
ff002b30 JA |
1258 | if (--fs->users) |
1259 | fs = NULL; | |
98447d65 | 1260 | spin_unlock(&req->work.identity->fs->lock); |
ff002b30 JA |
1261 | if (fs) |
1262 | free_fs_struct(fs); | |
dfead8a8 | 1263 | req->work.flags &= ~IO_WQ_WORK_FS; |
ff002b30 | 1264 | } |
51a4cc11 | 1265 | |
5c3462cf | 1266 | io_put_identity(req->task->io_uring, req); |
561fb04a JA |
1267 | } |
1268 | ||
1e6fa521 JA |
1269 | /* |
1270 | * Create a private copy of io_identity, since some fields don't match | |
1271 | * the current context. | |
1272 | */ | |
1273 | static bool io_identity_cow(struct io_kiocb *req) | |
1274 | { | |
5c3462cf | 1275 | struct io_uring_task *tctx = current->io_uring; |
1e6fa521 JA |
1276 | const struct cred *creds = NULL; |
1277 | struct io_identity *id; | |
1278 | ||
1279 | if (req->work.flags & IO_WQ_WORK_CREDS) | |
1280 | creds = req->work.identity->creds; | |
1281 | ||
1282 | id = kmemdup(req->work.identity, sizeof(*id), GFP_KERNEL); | |
1283 | if (unlikely(!id)) { | |
1284 | req->work.flags |= IO_WQ_WORK_CANCEL; | |
1285 | return false; | |
1286 | } | |
1287 | ||
1288 | /* | |
1289 | * We can safely just re-init the creds we copied Either the field | |
1290 | * matches the current one, or we haven't grabbed it yet. The only | |
1291 | * exception is ->creds, through registered personalities, so handle | |
1292 | * that one separately. | |
1293 | */ | |
1294 | io_init_identity(id); | |
1295 | if (creds) | |
1296 | req->work.identity->creds = creds; | |
1297 | ||
1298 | /* add one for this request */ | |
1299 | refcount_inc(&id->count); | |
1300 | ||
cb8a8ae3 JA |
1301 | /* drop tctx and req identity references, if needed */ |
1302 | if (tctx->identity != &tctx->__identity && | |
1303 | refcount_dec_and_test(&tctx->identity->count)) | |
1304 | kfree(tctx->identity); | |
1305 | if (req->work.identity != &tctx->__identity && | |
1306 | refcount_dec_and_test(&req->work.identity->count)) | |
1e6fa521 JA |
1307 | kfree(req->work.identity); |
1308 | ||
1309 | req->work.identity = id; | |
500a373d | 1310 | tctx->identity = id; |
1e6fa521 JA |
1311 | return true; |
1312 | } | |
1313 | ||
1314 | static bool io_grab_identity(struct io_kiocb *req) | |
18d9be1a | 1315 | { |
d3656344 | 1316 | const struct io_op_def *def = &io_op_defs[req->opcode]; |
5c3462cf | 1317 | struct io_identity *id = req->work.identity; |
23329513 | 1318 | struct io_ring_ctx *ctx = req->ctx; |
54a91f3b | 1319 | |
69228338 JA |
1320 | if (def->work_flags & IO_WQ_WORK_FSIZE) { |
1321 | if (id->fsize != rlimit(RLIMIT_FSIZE)) | |
1322 | return false; | |
1323 | req->work.flags |= IO_WQ_WORK_FSIZE; | |
1324 | } | |
16d59803 | 1325 | |
dfead8a8 | 1326 | if (!(req->work.flags & IO_WQ_WORK_FILES) && |
1e6fa521 | 1327 | (def->work_flags & IO_WQ_WORK_FILES) && |
23329513 | 1328 | !(req->flags & REQ_F_NO_FILE_TABLE)) { |
1e6fa521 JA |
1329 | if (id->files != current->files || |
1330 | id->nsproxy != current->nsproxy) | |
1331 | return false; | |
1332 | atomic_inc(&id->files->count); | |
1333 | get_nsproxy(id->nsproxy); | |
23329513 PB |
1334 | req->flags |= REQ_F_INFLIGHT; |
1335 | ||
1336 | spin_lock_irq(&ctx->inflight_lock); | |
1337 | list_add(&req->inflight_entry, &ctx->inflight_list); | |
1338 | spin_unlock_irq(&ctx->inflight_lock); | |
dfead8a8 | 1339 | req->work.flags |= IO_WQ_WORK_FILES; |
dca9cf8b | 1340 | } |
91d8f519 | 1341 | #ifdef CONFIG_BLK_CGROUP |
dfead8a8 JA |
1342 | if (!(req->work.flags & IO_WQ_WORK_BLKCG) && |
1343 | (def->work_flags & IO_WQ_WORK_BLKCG)) { | |
91d8f519 | 1344 | rcu_read_lock(); |
1e6fa521 JA |
1345 | if (id->blkcg_css != blkcg_css()) { |
1346 | rcu_read_unlock(); | |
1347 | return false; | |
1348 | } | |
91d8f519 DZ |
1349 | /* |
1350 | * This should be rare, either the cgroup is dying or the task | |
1351 | * is moving cgroups. Just punt to root for the handful of ios. | |
1352 | */ | |
1e6fa521 | 1353 | if (css_tryget_online(id->blkcg_css)) |
dfead8a8 | 1354 | req->work.flags |= IO_WQ_WORK_BLKCG; |
91d8f519 DZ |
1355 | rcu_read_unlock(); |
1356 | } | |
1357 | #endif | |
dfead8a8 | 1358 | if (!(req->work.flags & IO_WQ_WORK_CREDS)) { |
1e6fa521 JA |
1359 | if (id->creds != current_cred()) |
1360 | return false; | |
1361 | get_cred(id->creds); | |
dfead8a8 JA |
1362 | req->work.flags |= IO_WQ_WORK_CREDS; |
1363 | } | |
4ea33a97 JA |
1364 | #ifdef CONFIG_AUDIT |
1365 | if (!uid_eq(current->loginuid, id->loginuid) || | |
1366 | current->sessionid != id->sessionid) | |
1367 | return false; | |
1368 | #endif | |
dfead8a8 JA |
1369 | if (!(req->work.flags & IO_WQ_WORK_FS) && |
1370 | (def->work_flags & IO_WQ_WORK_FS)) { | |
1e6fa521 JA |
1371 | if (current->fs != id->fs) |
1372 | return false; | |
1373 | spin_lock(&id->fs->lock); | |
1374 | if (!id->fs->in_exec) { | |
1375 | id->fs->users++; | |
dfead8a8 | 1376 | req->work.flags |= IO_WQ_WORK_FS; |
dca9cf8b PB |
1377 | } else { |
1378 | req->work.flags |= IO_WQ_WORK_CANCEL; | |
1379 | } | |
1380 | spin_unlock(¤t->fs->lock); | |
1381 | } | |
1e6fa521 JA |
1382 | |
1383 | return true; | |
1384 | } | |
1385 | ||
1386 | static void io_prep_async_work(struct io_kiocb *req) | |
1387 | { | |
1388 | const struct io_op_def *def = &io_op_defs[req->opcode]; | |
1e6fa521 | 1389 | struct io_ring_ctx *ctx = req->ctx; |
5c3462cf | 1390 | struct io_identity *id; |
1e6fa521 JA |
1391 | |
1392 | io_req_init_async(req); | |
5c3462cf | 1393 | id = req->work.identity; |
1e6fa521 | 1394 | |
feaadc4f PB |
1395 | if (req->flags & REQ_F_FORCE_ASYNC) |
1396 | req->work.flags |= IO_WQ_WORK_CONCURRENT; | |
1397 | ||
1e6fa521 JA |
1398 | if (req->flags & REQ_F_ISREG) { |
1399 | if (def->hash_reg_file || (ctx->flags & IORING_SETUP_IOPOLL)) | |
1400 | io_wq_hash_work(&req->work, file_inode(req->file)); | |
1401 | } else { | |
1402 | if (def->unbound_nonreg_file) | |
1403 | req->work.flags |= IO_WQ_WORK_UNBOUND; | |
1404 | } | |
1405 | ||
1406 | /* ->mm can never change on us */ | |
1407 | if (!(req->work.flags & IO_WQ_WORK_MM) && | |
1408 | (def->work_flags & IO_WQ_WORK_MM)) { | |
1409 | mmgrab(id->mm); | |
1410 | req->work.flags |= IO_WQ_WORK_MM; | |
1411 | } | |
1412 | ||
1413 | /* if we fail grabbing identity, we must COW, regrab, and retry */ | |
1414 | if (io_grab_identity(req)) | |
1415 | return; | |
1416 | ||
1417 | if (!io_identity_cow(req)) | |
1418 | return; | |
1419 | ||
1420 | /* can't fail at this point */ | |
1421 | if (!io_grab_identity(req)) | |
1422 | WARN_ON(1); | |
561fb04a | 1423 | } |
cccf0ee8 | 1424 | |
cbdcb435 | 1425 | static void io_prep_async_link(struct io_kiocb *req) |
561fb04a | 1426 | { |
cbdcb435 | 1427 | struct io_kiocb *cur; |
54a91f3b | 1428 | |
cbdcb435 PB |
1429 | io_prep_async_work(req); |
1430 | if (req->flags & REQ_F_LINK_HEAD) | |
1431 | list_for_each_entry(cur, &req->link_list, link_list) | |
1432 | io_prep_async_work(cur); | |
561fb04a JA |
1433 | } |
1434 | ||
7271ef3a | 1435 | static struct io_kiocb *__io_queue_async_work(struct io_kiocb *req) |
561fb04a | 1436 | { |
a197f664 | 1437 | struct io_ring_ctx *ctx = req->ctx; |
cbdcb435 | 1438 | struct io_kiocb *link = io_prep_linked_timeout(req); |
561fb04a | 1439 | |
8766dd51 PB |
1440 | trace_io_uring_queue_async_work(ctx, io_wq_is_hashed(&req->work), req, |
1441 | &req->work, req->flags); | |
1442 | io_wq_enqueue(ctx->io_wq, &req->work); | |
7271ef3a | 1443 | return link; |
18d9be1a JA |
1444 | } |
1445 | ||
cbdcb435 PB |
1446 | static void io_queue_async_work(struct io_kiocb *req) |
1447 | { | |
7271ef3a JA |
1448 | struct io_kiocb *link; |
1449 | ||
cbdcb435 PB |
1450 | /* init ->work of the whole link before punting */ |
1451 | io_prep_async_link(req); | |
7271ef3a JA |
1452 | link = __io_queue_async_work(req); |
1453 | ||
1454 | if (link) | |
1455 | io_queue_linked_timeout(link); | |
cbdcb435 PB |
1456 | } |
1457 | ||
5262f567 JA |
1458 | static void io_kill_timeout(struct io_kiocb *req) |
1459 | { | |
e8c2bc1f | 1460 | struct io_timeout_data *io = req->async_data; |
5262f567 JA |
1461 | int ret; |
1462 | ||
e8c2bc1f | 1463 | ret = hrtimer_try_to_cancel(&io->timer); |
5262f567 | 1464 | if (ret != -1) { |
01cec8c1 PB |
1465 | atomic_set(&req->ctx->cq_timeouts, |
1466 | atomic_read(&req->ctx->cq_timeouts) + 1); | |
135fcde8 | 1467 | list_del_init(&req->timeout.list); |
78e19bbe | 1468 | io_cqring_fill_event(req, 0); |
216578e5 | 1469 | io_put_req_deferred(req, 1); |
5262f567 JA |
1470 | } |
1471 | } | |
1472 | ||
f3606e3a JA |
1473 | static bool io_task_match(struct io_kiocb *req, struct task_struct *tsk) |
1474 | { | |
1475 | struct io_ring_ctx *ctx = req->ctx; | |
1476 | ||
1477 | if (!tsk || req->task == tsk) | |
1478 | return true; | |
534ca6d6 JA |
1479 | if (ctx->flags & IORING_SETUP_SQPOLL) { |
1480 | if (ctx->sq_data && req->task == ctx->sq_data->thread) | |
1481 | return true; | |
1482 | } | |
f3606e3a JA |
1483 | return false; |
1484 | } | |
1485 | ||
76e1b642 JA |
1486 | /* |
1487 | * Returns true if we found and killed one or more timeouts | |
1488 | */ | |
1489 | static bool io_kill_timeouts(struct io_ring_ctx *ctx, struct task_struct *tsk) | |
5262f567 JA |
1490 | { |
1491 | struct io_kiocb *req, *tmp; | |
76e1b642 | 1492 | int canceled = 0; |
5262f567 JA |
1493 | |
1494 | spin_lock_irq(&ctx->completion_lock); | |
f3606e3a | 1495 | list_for_each_entry_safe(req, tmp, &ctx->timeout_list, timeout.list) { |
76e1b642 | 1496 | if (io_task_match(req, tsk)) { |
f3606e3a | 1497 | io_kill_timeout(req); |
76e1b642 JA |
1498 | canceled++; |
1499 | } | |
f3606e3a | 1500 | } |
5262f567 | 1501 | spin_unlock_irq(&ctx->completion_lock); |
76e1b642 | 1502 | return canceled != 0; |
5262f567 JA |
1503 | } |
1504 | ||
04518945 | 1505 | static void __io_queue_deferred(struct io_ring_ctx *ctx) |
de0617e4 | 1506 | { |
04518945 | 1507 | do { |
27dc8338 PB |
1508 | struct io_defer_entry *de = list_first_entry(&ctx->defer_list, |
1509 | struct io_defer_entry, list); | |
7271ef3a | 1510 | struct io_kiocb *link; |
de0617e4 | 1511 | |
9cf7c104 | 1512 | if (req_need_defer(de->req, de->seq)) |
04518945 | 1513 | break; |
27dc8338 | 1514 | list_del_init(&de->list); |
cbdcb435 | 1515 | /* punt-init is done before queueing for defer */ |
7271ef3a JA |
1516 | link = __io_queue_async_work(de->req); |
1517 | if (link) { | |
1518 | __io_queue_linked_timeout(link); | |
1519 | /* drop submission reference */ | |
216578e5 | 1520 | io_put_req_deferred(link, 1); |
7271ef3a | 1521 | } |
27dc8338 | 1522 | kfree(de); |
04518945 PB |
1523 | } while (!list_empty(&ctx->defer_list)); |
1524 | } | |
1525 | ||
360428f8 | 1526 | static void io_flush_timeouts(struct io_ring_ctx *ctx) |
de0617e4 | 1527 | { |
360428f8 PB |
1528 | while (!list_empty(&ctx->timeout_list)) { |
1529 | struct io_kiocb *req = list_first_entry(&ctx->timeout_list, | |
135fcde8 | 1530 | struct io_kiocb, timeout.list); |
de0617e4 | 1531 | |
8eb7e2d0 | 1532 | if (io_is_timeout_noseq(req)) |
360428f8 | 1533 | break; |
bfe68a22 PB |
1534 | if (req->timeout.target_seq != ctx->cached_cq_tail |
1535 | - atomic_read(&ctx->cq_timeouts)) | |
360428f8 | 1536 | break; |
bfe68a22 | 1537 | |
135fcde8 | 1538 | list_del_init(&req->timeout.list); |
5262f567 | 1539 | io_kill_timeout(req); |
360428f8 PB |
1540 | } |
1541 | } | |
5262f567 | 1542 | |
360428f8 PB |
1543 | static void io_commit_cqring(struct io_ring_ctx *ctx) |
1544 | { | |
1545 | io_flush_timeouts(ctx); | |
de0617e4 JA |
1546 | __io_commit_cqring(ctx); |
1547 | ||
04518945 PB |
1548 | if (unlikely(!list_empty(&ctx->defer_list))) |
1549 | __io_queue_deferred(ctx); | |
de0617e4 JA |
1550 | } |
1551 | ||
90554200 JA |
1552 | static inline bool io_sqring_full(struct io_ring_ctx *ctx) |
1553 | { | |
1554 | struct io_rings *r = ctx->rings; | |
1555 | ||
1556 | return READ_ONCE(r->sq.tail) - ctx->cached_sq_head == r->sq_ring_entries; | |
1557 | } | |
1558 | ||
2b188cc1 JA |
1559 | static struct io_uring_cqe *io_get_cqring(struct io_ring_ctx *ctx) |
1560 | { | |
75b28aff | 1561 | struct io_rings *rings = ctx->rings; |
2b188cc1 JA |
1562 | unsigned tail; |
1563 | ||
1564 | tail = ctx->cached_cq_tail; | |
115e12e5 SB |
1565 | /* |
1566 | * writes to the cq entry need to come after reading head; the | |
1567 | * control dependency is enough as we're using WRITE_ONCE to | |
1568 | * fill the cq entry | |
1569 | */ | |
75b28aff | 1570 | if (tail - READ_ONCE(rings->cq.head) == rings->cq_ring_entries) |
2b188cc1 JA |
1571 | return NULL; |
1572 | ||
1573 | ctx->cached_cq_tail++; | |
75b28aff | 1574 | return &rings->cqes[tail & ctx->cq_mask]; |
2b188cc1 JA |
1575 | } |
1576 | ||
f2842ab5 JA |
1577 | static inline bool io_should_trigger_evfd(struct io_ring_ctx *ctx) |
1578 | { | |
f0b493e6 JA |
1579 | if (!ctx->cq_ev_fd) |
1580 | return false; | |
7e55a19c SG |
1581 | if (READ_ONCE(ctx->rings->cq_flags) & IORING_CQ_EVENTFD_DISABLED) |
1582 | return false; | |
f2842ab5 JA |
1583 | if (!ctx->eventfd_async) |
1584 | return true; | |
b41e9852 | 1585 | return io_wq_current_is_worker(); |
f2842ab5 JA |
1586 | } |
1587 | ||
b41e9852 | 1588 | static void io_cqring_ev_posted(struct io_ring_ctx *ctx) |
1d7bb1d5 JA |
1589 | { |
1590 | if (waitqueue_active(&ctx->wait)) | |
1591 | wake_up(&ctx->wait); | |
534ca6d6 JA |
1592 | if (ctx->sq_data && waitqueue_active(&ctx->sq_data->wait)) |
1593 | wake_up(&ctx->sq_data->wait); | |
b41e9852 | 1594 | if (io_should_trigger_evfd(ctx)) |
1d7bb1d5 JA |
1595 | eventfd_signal(ctx->cq_ev_fd, 1); |
1596 | } | |
1597 | ||
46930143 PB |
1598 | static void io_cqring_mark_overflow(struct io_ring_ctx *ctx) |
1599 | { | |
1600 | if (list_empty(&ctx->cq_overflow_list)) { | |
1601 | clear_bit(0, &ctx->sq_check_overflow); | |
1602 | clear_bit(0, &ctx->cq_check_overflow); | |
1603 | ctx->rings->sq_flags &= ~IORING_SQ_CQ_OVERFLOW; | |
1604 | } | |
1605 | } | |
1606 | ||
99b32808 PB |
1607 | static inline bool __io_match_files(struct io_kiocb *req, |
1608 | struct files_struct *files) | |
e6c8aa9a | 1609 | { |
99b32808 PB |
1610 | return ((req->flags & REQ_F_WORK_INITIALIZED) && |
1611 | (req->work.flags & IO_WQ_WORK_FILES)) && | |
1612 | req->work.identity->files == files; | |
1613 | } | |
1614 | ||
1615 | static bool io_match_files(struct io_kiocb *req, | |
1616 | struct files_struct *files) | |
1617 | { | |
1618 | struct io_kiocb *link; | |
1619 | ||
e6c8aa9a JA |
1620 | if (!files) |
1621 | return true; | |
99b32808 PB |
1622 | if (__io_match_files(req, files)) |
1623 | return true; | |
1624 | if (req->flags & REQ_F_LINK_HEAD) { | |
1625 | list_for_each_entry(link, &req->link_list, link_list) { | |
1626 | if (__io_match_files(link, files)) | |
1627 | return true; | |
1628 | } | |
1629 | } | |
e6c8aa9a JA |
1630 | return false; |
1631 | } | |
1632 | ||
c4a2ed72 | 1633 | /* Returns true if there are no backlogged entries after the flush */ |
e6c8aa9a JA |
1634 | static bool io_cqring_overflow_flush(struct io_ring_ctx *ctx, bool force, |
1635 | struct task_struct *tsk, | |
1636 | struct files_struct *files) | |
1d7bb1d5 JA |
1637 | { |
1638 | struct io_rings *rings = ctx->rings; | |
e6c8aa9a | 1639 | struct io_kiocb *req, *tmp; |
1d7bb1d5 | 1640 | struct io_uring_cqe *cqe; |
1d7bb1d5 JA |
1641 | unsigned long flags; |
1642 | LIST_HEAD(list); | |
1643 | ||
1644 | if (!force) { | |
1645 | if (list_empty_careful(&ctx->cq_overflow_list)) | |
c4a2ed72 | 1646 | return true; |
1d7bb1d5 JA |
1647 | if ((ctx->cached_cq_tail - READ_ONCE(rings->cq.head) == |
1648 | rings->cq_ring_entries)) | |
c4a2ed72 | 1649 | return false; |
1d7bb1d5 JA |
1650 | } |
1651 | ||
1652 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
1653 | ||
1654 | /* if force is set, the ring is going away. always drop after that */ | |
1655 | if (force) | |
69b3e546 | 1656 | ctx->cq_overflow_flushed = 1; |
1d7bb1d5 | 1657 | |
c4a2ed72 | 1658 | cqe = NULL; |
e6c8aa9a JA |
1659 | list_for_each_entry_safe(req, tmp, &ctx->cq_overflow_list, compl.list) { |
1660 | if (tsk && req->task != tsk) | |
1661 | continue; | |
1662 | if (!io_match_files(req, files)) | |
1663 | continue; | |
1664 | ||
1d7bb1d5 JA |
1665 | cqe = io_get_cqring(ctx); |
1666 | if (!cqe && !force) | |
1667 | break; | |
1668 | ||
40d8ddd4 | 1669 | list_move(&req->compl.list, &list); |
1d7bb1d5 JA |
1670 | if (cqe) { |
1671 | WRITE_ONCE(cqe->user_data, req->user_data); | |
1672 | WRITE_ONCE(cqe->res, req->result); | |
0f7e466b | 1673 | WRITE_ONCE(cqe->flags, req->compl.cflags); |
1d7bb1d5 | 1674 | } else { |
2c3bac6d | 1675 | ctx->cached_cq_overflow++; |
1d7bb1d5 | 1676 | WRITE_ONCE(ctx->rings->cq_overflow, |
2c3bac6d | 1677 | ctx->cached_cq_overflow); |
1d7bb1d5 JA |
1678 | } |
1679 | } | |
1680 | ||
1681 | io_commit_cqring(ctx); | |
46930143 PB |
1682 | io_cqring_mark_overflow(ctx); |
1683 | ||
1d7bb1d5 JA |
1684 | spin_unlock_irqrestore(&ctx->completion_lock, flags); |
1685 | io_cqring_ev_posted(ctx); | |
1686 | ||
1687 | while (!list_empty(&list)) { | |
40d8ddd4 PB |
1688 | req = list_first_entry(&list, struct io_kiocb, compl.list); |
1689 | list_del(&req->compl.list); | |
ec9c02ad | 1690 | io_put_req(req); |
1d7bb1d5 | 1691 | } |
c4a2ed72 JA |
1692 | |
1693 | return cqe != NULL; | |
1d7bb1d5 JA |
1694 | } |
1695 | ||
bcda7baa | 1696 | static void __io_cqring_fill_event(struct io_kiocb *req, long res, long cflags) |
2b188cc1 | 1697 | { |
78e19bbe | 1698 | struct io_ring_ctx *ctx = req->ctx; |
2b188cc1 JA |
1699 | struct io_uring_cqe *cqe; |
1700 | ||
78e19bbe | 1701 | trace_io_uring_complete(ctx, req->user_data, res); |
51c3ff62 | 1702 | |
2b188cc1 JA |
1703 | /* |
1704 | * If we can't get a cq entry, userspace overflowed the | |
1705 | * submission (by quite a lot). Increment the overflow count in | |
1706 | * the ring. | |
1707 | */ | |
1708 | cqe = io_get_cqring(ctx); | |
1d7bb1d5 | 1709 | if (likely(cqe)) { |
78e19bbe | 1710 | WRITE_ONCE(cqe->user_data, req->user_data); |
2b188cc1 | 1711 | WRITE_ONCE(cqe->res, res); |
bcda7baa | 1712 | WRITE_ONCE(cqe->flags, cflags); |
fdaf083c JA |
1713 | } else if (ctx->cq_overflow_flushed || |
1714 | atomic_read(&req->task->io_uring->in_idle)) { | |
0f212204 JA |
1715 | /* |
1716 | * If we're in ring overflow flush mode, or in task cancel mode, | |
1717 | * then we cannot store the request for later flushing, we need | |
1718 | * to drop it on the floor. | |
1719 | */ | |
2c3bac6d PB |
1720 | ctx->cached_cq_overflow++; |
1721 | WRITE_ONCE(ctx->rings->cq_overflow, ctx->cached_cq_overflow); | |
1d7bb1d5 | 1722 | } else { |
ad3eb2c8 JA |
1723 | if (list_empty(&ctx->cq_overflow_list)) { |
1724 | set_bit(0, &ctx->sq_check_overflow); | |
1725 | set_bit(0, &ctx->cq_check_overflow); | |
6d5f9049 | 1726 | ctx->rings->sq_flags |= IORING_SQ_CQ_OVERFLOW; |
ad3eb2c8 | 1727 | } |
40d8ddd4 | 1728 | io_clean_op(req); |
1d7bb1d5 | 1729 | req->result = res; |
0f7e466b | 1730 | req->compl.cflags = cflags; |
40d8ddd4 PB |
1731 | refcount_inc(&req->refs); |
1732 | list_add_tail(&req->compl.list, &ctx->cq_overflow_list); | |
2b188cc1 JA |
1733 | } |
1734 | } | |
1735 | ||
bcda7baa JA |
1736 | static void io_cqring_fill_event(struct io_kiocb *req, long res) |
1737 | { | |
1738 | __io_cqring_fill_event(req, res, 0); | |
1739 | } | |
1740 | ||
e1e16097 | 1741 | static void io_cqring_add_event(struct io_kiocb *req, long res, long cflags) |
2b188cc1 | 1742 | { |
78e19bbe | 1743 | struct io_ring_ctx *ctx = req->ctx; |
2b188cc1 JA |
1744 | unsigned long flags; |
1745 | ||
1746 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
bcda7baa | 1747 | __io_cqring_fill_event(req, res, cflags); |
2b188cc1 JA |
1748 | io_commit_cqring(ctx); |
1749 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
1750 | ||
8c838788 | 1751 | io_cqring_ev_posted(ctx); |
2b188cc1 JA |
1752 | } |
1753 | ||
229a7b63 | 1754 | static void io_submit_flush_completions(struct io_comp_state *cs) |
bcda7baa | 1755 | { |
229a7b63 JA |
1756 | struct io_ring_ctx *ctx = cs->ctx; |
1757 | ||
1758 | spin_lock_irq(&ctx->completion_lock); | |
1759 | while (!list_empty(&cs->list)) { | |
1760 | struct io_kiocb *req; | |
1761 | ||
3ca405eb PB |
1762 | req = list_first_entry(&cs->list, struct io_kiocb, compl.list); |
1763 | list_del(&req->compl.list); | |
0f7e466b | 1764 | __io_cqring_fill_event(req, req->result, req->compl.cflags); |
216578e5 PB |
1765 | |
1766 | /* | |
1767 | * io_free_req() doesn't care about completion_lock unless one | |
1768 | * of these flags is set. REQ_F_WORK_INITIALIZED is in the list | |
1769 | * because of a potential deadlock with req->work.fs->lock | |
1770 | */ | |
1771 | if (req->flags & (REQ_F_FAIL_LINK|REQ_F_LINK_TIMEOUT | |
1772 | |REQ_F_WORK_INITIALIZED)) { | |
229a7b63 JA |
1773 | spin_unlock_irq(&ctx->completion_lock); |
1774 | io_put_req(req); | |
1775 | spin_lock_irq(&ctx->completion_lock); | |
216578e5 PB |
1776 | } else { |
1777 | io_put_req(req); | |
229a7b63 JA |
1778 | } |
1779 | } | |
1780 | io_commit_cqring(ctx); | |
1781 | spin_unlock_irq(&ctx->completion_lock); | |
1782 | ||
1783 | io_cqring_ev_posted(ctx); | |
1784 | cs->nr = 0; | |
1785 | } | |
1786 | ||
1787 | static void __io_req_complete(struct io_kiocb *req, long res, unsigned cflags, | |
1788 | struct io_comp_state *cs) | |
1789 | { | |
1790 | if (!cs) { | |
1791 | io_cqring_add_event(req, res, cflags); | |
1792 | io_put_req(req); | |
1793 | } else { | |
3ca405eb | 1794 | io_clean_op(req); |
229a7b63 | 1795 | req->result = res; |
0f7e466b | 1796 | req->compl.cflags = cflags; |
3ca405eb | 1797 | list_add_tail(&req->compl.list, &cs->list); |
229a7b63 JA |
1798 | if (++cs->nr >= 32) |
1799 | io_submit_flush_completions(cs); | |
1800 | } | |
e1e16097 JA |
1801 | } |
1802 | ||
1803 | static void io_req_complete(struct io_kiocb *req, long res) | |
bcda7baa | 1804 | { |
229a7b63 | 1805 | __io_req_complete(req, res, 0, NULL); |
bcda7baa JA |
1806 | } |
1807 | ||
0ddf92e8 JA |
1808 | static inline bool io_is_fallback_req(struct io_kiocb *req) |
1809 | { | |
1810 | return req == (struct io_kiocb *) | |
1811 | ((unsigned long) req->ctx->fallback_req & ~1UL); | |
1812 | } | |
1813 | ||
1814 | static struct io_kiocb *io_get_fallback_req(struct io_ring_ctx *ctx) | |
1815 | { | |
1816 | struct io_kiocb *req; | |
1817 | ||
1818 | req = ctx->fallback_req; | |
dd461af6 | 1819 | if (!test_and_set_bit_lock(0, (unsigned long *) &ctx->fallback_req)) |
0ddf92e8 JA |
1820 | return req; |
1821 | ||
1822 | return NULL; | |
1823 | } | |
1824 | ||
0553b8bd PB |
1825 | static struct io_kiocb *io_alloc_req(struct io_ring_ctx *ctx, |
1826 | struct io_submit_state *state) | |
2b188cc1 | 1827 | { |
f6b6c7d6 | 1828 | if (!state->free_reqs) { |
291b2821 | 1829 | gfp_t gfp = GFP_KERNEL | __GFP_NOWARN; |
2579f913 JA |
1830 | size_t sz; |
1831 | int ret; | |
1832 | ||
1833 | sz = min_t(size_t, state->ios_left, ARRAY_SIZE(state->reqs)); | |
fd6fab2c JA |
1834 | ret = kmem_cache_alloc_bulk(req_cachep, gfp, sz, state->reqs); |
1835 | ||
1836 | /* | |
1837 | * Bulk alloc is all-or-nothing. If we fail to get a batch, | |
1838 | * retry single alloc to be on the safe side. | |
1839 | */ | |
1840 | if (unlikely(ret <= 0)) { | |
1841 | state->reqs[0] = kmem_cache_alloc(req_cachep, gfp); | |
1842 | if (!state->reqs[0]) | |
0ddf92e8 | 1843 | goto fallback; |
fd6fab2c JA |
1844 | ret = 1; |
1845 | } | |
291b2821 | 1846 | state->free_reqs = ret; |
2b188cc1 JA |
1847 | } |
1848 | ||
291b2821 PB |
1849 | state->free_reqs--; |
1850 | return state->reqs[state->free_reqs]; | |
0ddf92e8 | 1851 | fallback: |
0553b8bd | 1852 | return io_get_fallback_req(ctx); |
2b188cc1 JA |
1853 | } |
1854 | ||
8da11c19 PB |
1855 | static inline void io_put_file(struct io_kiocb *req, struct file *file, |
1856 | bool fixed) | |
1857 | { | |
1858 | if (fixed) | |
05589553 | 1859 | percpu_ref_put(req->fixed_file_refs); |
8da11c19 PB |
1860 | else |
1861 | fput(file); | |
1862 | } | |
1863 | ||
4edf20f9 | 1864 | static void io_dismantle_req(struct io_kiocb *req) |
2b188cc1 | 1865 | { |
3ca405eb | 1866 | io_clean_op(req); |
929a3af9 | 1867 | |
e8c2bc1f JA |
1868 | if (req->async_data) |
1869 | kfree(req->async_data); | |
8da11c19 PB |
1870 | if (req->file) |
1871 | io_put_file(req, req->file, (req->flags & REQ_F_FIXED_FILE)); | |
fcb323cc | 1872 | |
4edf20f9 | 1873 | io_req_clean_work(req); |
e65ef56d JA |
1874 | } |
1875 | ||
216578e5 | 1876 | static void __io_free_req(struct io_kiocb *req) |
c6ca97b3 | 1877 | { |
0f212204 | 1878 | struct io_uring_task *tctx = req->task->io_uring; |
51a4cc11 | 1879 | struct io_ring_ctx *ctx = req->ctx; |
c6ca97b3 | 1880 | |
216578e5 | 1881 | io_dismantle_req(req); |
c6ca97b3 | 1882 | |
d8a6df10 | 1883 | percpu_counter_dec(&tctx->inflight); |
fdaf083c | 1884 | if (atomic_read(&tctx->in_idle)) |
0f212204 | 1885 | wake_up(&tctx->wait); |
e3bc8e9d JA |
1886 | put_task_struct(req->task); |
1887 | ||
b1e50e54 PB |
1888 | if (likely(!io_is_fallback_req(req))) |
1889 | kmem_cache_free(req_cachep, req); | |
1890 | else | |
ecfc5177 PB |
1891 | clear_bit_unlock(0, (unsigned long *) &ctx->fallback_req); |
1892 | percpu_ref_put(&ctx->refs); | |
e65ef56d JA |
1893 | } |
1894 | ||
c9abd7ad | 1895 | static void io_kill_linked_timeout(struct io_kiocb *req) |
2665abfd | 1896 | { |
a197f664 | 1897 | struct io_ring_ctx *ctx = req->ctx; |
7c86ffee | 1898 | struct io_kiocb *link; |
c9abd7ad PB |
1899 | bool cancelled = false; |
1900 | unsigned long flags; | |
7c86ffee | 1901 | |
c9abd7ad PB |
1902 | spin_lock_irqsave(&ctx->completion_lock, flags); |
1903 | link = list_first_entry_or_null(&req->link_list, struct io_kiocb, | |
1904 | link_list); | |
900fad45 PB |
1905 | /* |
1906 | * Can happen if a linked timeout fired and link had been like | |
1907 | * req -> link t-out -> link t-out [-> ...] | |
1908 | */ | |
c9abd7ad PB |
1909 | if (link && (link->flags & REQ_F_LTIMEOUT_ACTIVE)) { |
1910 | struct io_timeout_data *io = link->async_data; | |
1911 | int ret; | |
7c86ffee | 1912 | |
c9abd7ad PB |
1913 | list_del_init(&link->link_list); |
1914 | ret = hrtimer_try_to_cancel(&io->timer); | |
1915 | if (ret != -1) { | |
1916 | io_cqring_fill_event(link, -ECANCELED); | |
1917 | io_commit_cqring(ctx); | |
1918 | cancelled = true; | |
1919 | } | |
1920 | } | |
7c86ffee | 1921 | req->flags &= ~REQ_F_LINK_TIMEOUT; |
216578e5 | 1922 | spin_unlock_irqrestore(&ctx->completion_lock, flags); |
ab0b6451 | 1923 | |
c9abd7ad | 1924 | if (cancelled) { |
7c86ffee | 1925 | io_cqring_ev_posted(ctx); |
c9abd7ad PB |
1926 | io_put_req(link); |
1927 | } | |
7c86ffee PB |
1928 | } |
1929 | ||
9b5f7bd9 | 1930 | static struct io_kiocb *io_req_link_next(struct io_kiocb *req) |
7c86ffee PB |
1931 | { |
1932 | struct io_kiocb *nxt; | |
4d7dd462 | 1933 | |
9e645e11 JA |
1934 | /* |
1935 | * The list should never be empty when we are called here. But could | |
1936 | * potentially happen if the chain is messed up, check to be on the | |
1937 | * safe side. | |
1938 | */ | |
7c86ffee | 1939 | if (unlikely(list_empty(&req->link_list))) |
9b5f7bd9 | 1940 | return NULL; |
2665abfd | 1941 | |
7c86ffee PB |
1942 | nxt = list_first_entry(&req->link_list, struct io_kiocb, link_list); |
1943 | list_del_init(&req->link_list); | |
1944 | if (!list_empty(&nxt->link_list)) | |
1945 | nxt->flags |= REQ_F_LINK_HEAD; | |
9b5f7bd9 | 1946 | return nxt; |
9e645e11 JA |
1947 | } |
1948 | ||
1949 | /* | |
dea3b49c | 1950 | * Called if REQ_F_LINK_HEAD is set, and we fail the head request |
9e645e11 | 1951 | */ |
d148ca4b | 1952 | static void io_fail_links(struct io_kiocb *req) |
9e645e11 | 1953 | { |
2665abfd | 1954 | struct io_ring_ctx *ctx = req->ctx; |
d148ca4b | 1955 | unsigned long flags; |
9e645e11 | 1956 | |
d148ca4b | 1957 | spin_lock_irqsave(&ctx->completion_lock, flags); |
9e645e11 | 1958 | while (!list_empty(&req->link_list)) { |
4493233e PB |
1959 | struct io_kiocb *link = list_first_entry(&req->link_list, |
1960 | struct io_kiocb, link_list); | |
9e645e11 | 1961 | |
4493233e | 1962 | list_del_init(&link->link_list); |
c826bd7a | 1963 | trace_io_uring_fail_link(req, link); |
2665abfd | 1964 | |
7c86ffee | 1965 | io_cqring_fill_event(link, -ECANCELED); |
216578e5 PB |
1966 | |
1967 | /* | |
1968 | * It's ok to free under spinlock as they're not linked anymore, | |
1969 | * but avoid REQ_F_WORK_INITIALIZED because it may deadlock on | |
1970 | * work.fs->lock. | |
1971 | */ | |
1972 | if (link->flags & REQ_F_WORK_INITIALIZED) | |
1973 | io_put_req_deferred(link, 2); | |
1974 | else | |
1975 | io_double_put_req(link); | |
9e645e11 | 1976 | } |
2665abfd JA |
1977 | |
1978 | io_commit_cqring(ctx); | |
216578e5 | 1979 | spin_unlock_irqrestore(&ctx->completion_lock, flags); |
9e645e11 | 1980 | |
2665abfd | 1981 | io_cqring_ev_posted(ctx); |
9e645e11 JA |
1982 | } |
1983 | ||
3fa5e0f3 | 1984 | static struct io_kiocb *__io_req_find_next(struct io_kiocb *req) |
c69f8dbe | 1985 | { |
9b0d911a | 1986 | req->flags &= ~REQ_F_LINK_HEAD; |
7c86ffee PB |
1987 | if (req->flags & REQ_F_LINK_TIMEOUT) |
1988 | io_kill_linked_timeout(req); | |
944e58bf | 1989 | |
9e645e11 JA |
1990 | /* |
1991 | * If LINK is set, we have dependent requests in this chain. If we | |
1992 | * didn't fail this request, queue the first one up, moving any other | |
1993 | * dependencies to the next request. In case of failure, fail the rest | |
1994 | * of the chain. | |
1995 | */ | |
9b5f7bd9 PB |
1996 | if (likely(!(req->flags & REQ_F_FAIL_LINK))) |
1997 | return io_req_link_next(req); | |
1998 | io_fail_links(req); | |
1999 | return NULL; | |
4d7dd462 | 2000 | } |
9e645e11 | 2001 | |
3fa5e0f3 PB |
2002 | static struct io_kiocb *io_req_find_next(struct io_kiocb *req) |
2003 | { | |
2004 | if (likely(!(req->flags & REQ_F_LINK_HEAD))) | |
2005 | return NULL; | |
2006 | return __io_req_find_next(req); | |
2007 | } | |
2008 | ||
87c4311f | 2009 | static int io_req_task_work_add(struct io_kiocb *req, bool twa_signal_ok) |
c2c4c83c JA |
2010 | { |
2011 | struct task_struct *tsk = req->task; | |
2012 | struct io_ring_ctx *ctx = req->ctx; | |
91989c70 JA |
2013 | enum task_work_notify_mode notify; |
2014 | int ret; | |
c2c4c83c | 2015 | |
6200b0ae JA |
2016 | if (tsk->flags & PF_EXITING) |
2017 | return -ESRCH; | |
2018 | ||
c2c4c83c | 2019 | /* |
0ba9c9ed JA |
2020 | * SQPOLL kernel thread doesn't need notification, just a wakeup. For |
2021 | * all other cases, use TWA_SIGNAL unconditionally to ensure we're | |
2022 | * processing task_work. There's no reliable way to tell if TWA_RESUME | |
2023 | * will do the job. | |
c2c4c83c | 2024 | */ |
91989c70 | 2025 | notify = TWA_NONE; |
fd7d6de2 | 2026 | if (!(ctx->flags & IORING_SETUP_SQPOLL) && twa_signal_ok) |
c2c4c83c JA |
2027 | notify = TWA_SIGNAL; |
2028 | ||
87c4311f | 2029 | ret = task_work_add(tsk, &req->task_work, notify); |
c2c4c83c JA |
2030 | if (!ret) |
2031 | wake_up_process(tsk); | |
0ba9c9ed | 2032 | |
c2c4c83c JA |
2033 | return ret; |
2034 | } | |
2035 | ||
c40f6379 JA |
2036 | static void __io_req_task_cancel(struct io_kiocb *req, int error) |
2037 | { | |
2038 | struct io_ring_ctx *ctx = req->ctx; | |
2039 | ||
2040 | spin_lock_irq(&ctx->completion_lock); | |
2041 | io_cqring_fill_event(req, error); | |
2042 | io_commit_cqring(ctx); | |
2043 | spin_unlock_irq(&ctx->completion_lock); | |
2044 | ||
2045 | io_cqring_ev_posted(ctx); | |
2046 | req_set_fail_links(req); | |
2047 | io_double_put_req(req); | |
2048 | } | |
2049 | ||
2050 | static void io_req_task_cancel(struct callback_head *cb) | |
2051 | { | |
2052 | struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work); | |
87ceb6a6 | 2053 | struct io_ring_ctx *ctx = req->ctx; |
c40f6379 JA |
2054 | |
2055 | __io_req_task_cancel(req, -ECANCELED); | |
87ceb6a6 | 2056 | percpu_ref_put(&ctx->refs); |
c40f6379 JA |
2057 | } |
2058 | ||
2059 | static void __io_req_task_submit(struct io_kiocb *req) | |
2060 | { | |
2061 | struct io_ring_ctx *ctx = req->ctx; | |
2062 | ||
c40f6379 JA |
2063 | if (!__io_sq_thread_acquire_mm(ctx)) { |
2064 | mutex_lock(&ctx->uring_lock); | |
c1379e24 | 2065 | __io_queue_sqe(req, NULL); |
c40f6379 JA |
2066 | mutex_unlock(&ctx->uring_lock); |
2067 | } else { | |
2068 | __io_req_task_cancel(req, -EFAULT); | |
2069 | } | |
2070 | } | |
2071 | ||
2072 | static void io_req_task_submit(struct callback_head *cb) | |
2073 | { | |
2074 | struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work); | |
6d816e08 | 2075 | struct io_ring_ctx *ctx = req->ctx; |
c40f6379 JA |
2076 | |
2077 | __io_req_task_submit(req); | |
6d816e08 | 2078 | percpu_ref_put(&ctx->refs); |
c40f6379 JA |
2079 | } |
2080 | ||
2081 | static void io_req_task_queue(struct io_kiocb *req) | |
2082 | { | |
c40f6379 JA |
2083 | int ret; |
2084 | ||
2085 | init_task_work(&req->task_work, io_req_task_submit); | |
6d816e08 | 2086 | percpu_ref_get(&req->ctx->refs); |
c40f6379 | 2087 | |
87c4311f | 2088 | ret = io_req_task_work_add(req, true); |
c40f6379 | 2089 | if (unlikely(ret)) { |
c2c4c83c JA |
2090 | struct task_struct *tsk; |
2091 | ||
c40f6379 JA |
2092 | init_task_work(&req->task_work, io_req_task_cancel); |
2093 | tsk = io_wq_get_task(req->ctx->io_wq); | |
91989c70 | 2094 | task_work_add(tsk, &req->task_work, TWA_NONE); |
c2c4c83c | 2095 | wake_up_process(tsk); |
c40f6379 | 2096 | } |
c40f6379 JA |
2097 | } |
2098 | ||
c3524383 | 2099 | static void io_queue_next(struct io_kiocb *req) |
c69f8dbe | 2100 | { |
9b5f7bd9 | 2101 | struct io_kiocb *nxt = io_req_find_next(req); |
944e58bf PB |
2102 | |
2103 | if (nxt) | |
906a8c3f | 2104 | io_req_task_queue(nxt); |
c69f8dbe JL |
2105 | } |
2106 | ||
c3524383 | 2107 | static void io_free_req(struct io_kiocb *req) |
7a743e22 | 2108 | { |
c3524383 PB |
2109 | io_queue_next(req); |
2110 | __io_free_req(req); | |
2111 | } | |
8766dd51 | 2112 | |
2d6500d4 PB |
2113 | struct req_batch { |
2114 | void *reqs[IO_IOPOLL_BATCH]; | |
2115 | int to_free; | |
7a743e22 | 2116 | |
5af1d13e PB |
2117 | struct task_struct *task; |
2118 | int task_refs; | |
2d6500d4 PB |
2119 | }; |
2120 | ||
5af1d13e PB |
2121 | static inline void io_init_req_batch(struct req_batch *rb) |
2122 | { | |
2123 | rb->to_free = 0; | |
2124 | rb->task_refs = 0; | |
2125 | rb->task = NULL; | |
2126 | } | |
2127 | ||
2d6500d4 PB |
2128 | static void __io_req_free_batch_flush(struct io_ring_ctx *ctx, |
2129 | struct req_batch *rb) | |
2130 | { | |
2131 | kmem_cache_free_bulk(req_cachep, rb->to_free, rb->reqs); | |
2132 | percpu_ref_put_many(&ctx->refs, rb->to_free); | |
2133 | rb->to_free = 0; | |
2134 | } | |
2135 | ||
2136 | static void io_req_free_batch_finish(struct io_ring_ctx *ctx, | |
2137 | struct req_batch *rb) | |
2138 | { | |
2139 | if (rb->to_free) | |
2140 | __io_req_free_batch_flush(ctx, rb); | |
5af1d13e | 2141 | if (rb->task) { |
d8a6df10 JA |
2142 | struct io_uring_task *tctx = rb->task->io_uring; |
2143 | ||
2144 | percpu_counter_sub(&tctx->inflight, rb->task_refs); | |
5af1d13e PB |
2145 | put_task_struct_many(rb->task, rb->task_refs); |
2146 | rb->task = NULL; | |
2147 | } | |
2d6500d4 PB |
2148 | } |
2149 | ||
2150 | static void io_req_free_batch(struct req_batch *rb, struct io_kiocb *req) | |
2151 | { | |
2152 | if (unlikely(io_is_fallback_req(req))) { | |
2153 | io_free_req(req); | |
2154 | return; | |
2155 | } | |
2156 | if (req->flags & REQ_F_LINK_HEAD) | |
2157 | io_queue_next(req); | |
2158 | ||
e3bc8e9d | 2159 | if (req->task != rb->task) { |
0f212204 | 2160 | if (rb->task) { |
d8a6df10 JA |
2161 | struct io_uring_task *tctx = rb->task->io_uring; |
2162 | ||
2163 | percpu_counter_sub(&tctx->inflight, rb->task_refs); | |
e3bc8e9d | 2164 | put_task_struct_many(rb->task, rb->task_refs); |
5af1d13e | 2165 | } |
e3bc8e9d JA |
2166 | rb->task = req->task; |
2167 | rb->task_refs = 0; | |
5af1d13e | 2168 | } |
e3bc8e9d | 2169 | rb->task_refs++; |
5af1d13e | 2170 | |
4edf20f9 | 2171 | io_dismantle_req(req); |
2d6500d4 PB |
2172 | rb->reqs[rb->to_free++] = req; |
2173 | if (unlikely(rb->to_free == ARRAY_SIZE(rb->reqs))) | |
2174 | __io_req_free_batch_flush(req->ctx, rb); | |
7a743e22 PB |
2175 | } |
2176 | ||
ba816ad6 JA |
2177 | /* |
2178 | * Drop reference to request, return next in chain (if there is one) if this | |
2179 | * was the last reference to this request. | |
2180 | */ | |
9b5f7bd9 | 2181 | static struct io_kiocb *io_put_req_find_next(struct io_kiocb *req) |
e65ef56d | 2182 | { |
9b5f7bd9 PB |
2183 | struct io_kiocb *nxt = NULL; |
2184 | ||
2a44f467 | 2185 | if (refcount_dec_and_test(&req->refs)) { |
9b5f7bd9 | 2186 | nxt = io_req_find_next(req); |
4d7dd462 | 2187 | __io_free_req(req); |
2a44f467 | 2188 | } |
9b5f7bd9 | 2189 | return nxt; |
2b188cc1 JA |
2190 | } |
2191 | ||
e65ef56d JA |
2192 | static void io_put_req(struct io_kiocb *req) |
2193 | { | |
2194 | if (refcount_dec_and_test(&req->refs)) | |
2195 | io_free_req(req); | |
2b188cc1 JA |
2196 | } |
2197 | ||
216578e5 PB |
2198 | static void io_put_req_deferred_cb(struct callback_head *cb) |
2199 | { | |
2200 | struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work); | |
2201 | ||
2202 | io_free_req(req); | |
2203 | } | |
2204 | ||
2205 | static void io_free_req_deferred(struct io_kiocb *req) | |
2206 | { | |
2207 | int ret; | |
2208 | ||
2209 | init_task_work(&req->task_work, io_put_req_deferred_cb); | |
2210 | ret = io_req_task_work_add(req, true); | |
2211 | if (unlikely(ret)) { | |
2212 | struct task_struct *tsk; | |
2213 | ||
2214 | tsk = io_wq_get_task(req->ctx->io_wq); | |
91989c70 | 2215 | task_work_add(tsk, &req->task_work, TWA_NONE); |
216578e5 PB |
2216 | wake_up_process(tsk); |
2217 | } | |
2218 | } | |
2219 | ||
2220 | static inline void io_put_req_deferred(struct io_kiocb *req, int refs) | |
2221 | { | |
2222 | if (refcount_sub_and_test(refs, &req->refs)) | |
2223 | io_free_req_deferred(req); | |
2224 | } | |
2225 | ||
f4db7182 | 2226 | static struct io_wq_work *io_steal_work(struct io_kiocb *req) |
7a743e22 | 2227 | { |
6df1db6b | 2228 | struct io_kiocb *nxt; |
f4db7182 | 2229 | |
7a743e22 | 2230 | /* |
f4db7182 PB |
2231 | * A ref is owned by io-wq in which context we're. So, if that's the |
2232 | * last one, it's safe to steal next work. False negatives are Ok, | |
2233 | * it just will be re-punted async in io_put_work() | |
7a743e22 | 2234 | */ |
f4db7182 PB |
2235 | if (refcount_read(&req->refs) != 1) |
2236 | return NULL; | |
7a743e22 | 2237 | |
9b5f7bd9 | 2238 | nxt = io_req_find_next(req); |
6df1db6b | 2239 | return nxt ? &nxt->work : NULL; |
7a743e22 PB |
2240 | } |
2241 | ||
978db57e JA |
2242 | static void io_double_put_req(struct io_kiocb *req) |
2243 | { | |
2244 | /* drop both submit and complete references */ | |
2245 | if (refcount_sub_and_test(2, &req->refs)) | |
2246 | io_free_req(req); | |
2247 | } | |
2248 | ||
1d7bb1d5 | 2249 | static unsigned io_cqring_events(struct io_ring_ctx *ctx, bool noflush) |
a3a0e43f | 2250 | { |
84f97dc2 JA |
2251 | struct io_rings *rings = ctx->rings; |
2252 | ||
ad3eb2c8 JA |
2253 | if (test_bit(0, &ctx->cq_check_overflow)) { |
2254 | /* | |
2255 | * noflush == true is from the waitqueue handler, just ensure | |
2256 | * we wake up the task, and the next invocation will flush the | |
2257 | * entries. We cannot safely to it from here. | |
2258 | */ | |
2259 | if (noflush && !list_empty(&ctx->cq_overflow_list)) | |
2260 | return -1U; | |
1d7bb1d5 | 2261 | |
e6c8aa9a | 2262 | io_cqring_overflow_flush(ctx, false, NULL, NULL); |
ad3eb2c8 | 2263 | } |
1d7bb1d5 | 2264 | |
a3a0e43f JA |
2265 | /* See comment at the top of this file */ |
2266 | smp_rmb(); | |
ad3eb2c8 | 2267 | return ctx->cached_cq_tail - READ_ONCE(rings->cq.head); |
a3a0e43f JA |
2268 | } |
2269 | ||
fb5ccc98 PB |
2270 | static inline unsigned int io_sqring_entries(struct io_ring_ctx *ctx) |
2271 | { | |
2272 | struct io_rings *rings = ctx->rings; | |
2273 | ||
2274 | /* make sure SQ entry isn't read before tail */ | |
2275 | return smp_load_acquire(&rings->sq.tail) - ctx->cached_sq_head; | |
2276 | } | |
2277 | ||
8ff069bf | 2278 | static unsigned int io_put_kbuf(struct io_kiocb *req, struct io_buffer *kbuf) |
e94f141b | 2279 | { |
8ff069bf | 2280 | unsigned int cflags; |
e94f141b | 2281 | |
bcda7baa JA |
2282 | cflags = kbuf->bid << IORING_CQE_BUFFER_SHIFT; |
2283 | cflags |= IORING_CQE_F_BUFFER; | |
0e1b6fe3 | 2284 | req->flags &= ~REQ_F_BUFFER_SELECTED; |
bcda7baa JA |
2285 | kfree(kbuf); |
2286 | return cflags; | |
e94f141b JA |
2287 | } |
2288 | ||
8ff069bf | 2289 | static inline unsigned int io_put_rw_kbuf(struct io_kiocb *req) |
bcda7baa | 2290 | { |
4d954c25 | 2291 | struct io_buffer *kbuf; |
bcda7baa | 2292 | |
4d954c25 | 2293 | kbuf = (struct io_buffer *) (unsigned long) req->rw.addr; |
8ff069bf PB |
2294 | return io_put_kbuf(req, kbuf); |
2295 | } | |
2296 | ||
4c6e277c JA |
2297 | static inline bool io_run_task_work(void) |
2298 | { | |
6200b0ae JA |
2299 | /* |
2300 | * Not safe to run on exiting task, and the task_work handling will | |
2301 | * not add work to such a task. | |
2302 | */ | |
2303 | if (unlikely(current->flags & PF_EXITING)) | |
2304 | return false; | |
4c6e277c JA |
2305 | if (current->task_works) { |
2306 | __set_current_state(TASK_RUNNING); | |
2307 | task_work_run(); | |
2308 | return true; | |
2309 | } | |
2310 | ||
2311 | return false; | |
bcda7baa JA |
2312 | } |
2313 | ||
bbde017a XW |
2314 | static void io_iopoll_queue(struct list_head *again) |
2315 | { | |
2316 | struct io_kiocb *req; | |
2317 | ||
2318 | do { | |
d21ffe7e PB |
2319 | req = list_first_entry(again, struct io_kiocb, inflight_entry); |
2320 | list_del(&req->inflight_entry); | |
81b68a5c | 2321 | __io_complete_rw(req, -EAGAIN, 0, NULL); |
bbde017a XW |
2322 | } while (!list_empty(again)); |
2323 | } | |
2324 | ||
def596e9 JA |
2325 | /* |
2326 | * Find and free completed poll iocbs | |
2327 | */ | |
2328 | static void io_iopoll_complete(struct io_ring_ctx *ctx, unsigned int *nr_events, | |
2329 | struct list_head *done) | |
2330 | { | |
8237e045 | 2331 | struct req_batch rb; |
def596e9 | 2332 | struct io_kiocb *req; |
bbde017a XW |
2333 | LIST_HEAD(again); |
2334 | ||
2335 | /* order with ->result store in io_complete_rw_iopoll() */ | |
2336 | smp_rmb(); | |
def596e9 | 2337 | |
5af1d13e | 2338 | io_init_req_batch(&rb); |
def596e9 | 2339 | while (!list_empty(done)) { |
bcda7baa JA |
2340 | int cflags = 0; |
2341 | ||
d21ffe7e | 2342 | req = list_first_entry(done, struct io_kiocb, inflight_entry); |
bbde017a | 2343 | if (READ_ONCE(req->result) == -EAGAIN) { |
56450c20 | 2344 | req->result = 0; |
bbde017a | 2345 | req->iopoll_completed = 0; |
d21ffe7e | 2346 | list_move_tail(&req->inflight_entry, &again); |
bbde017a XW |
2347 | continue; |
2348 | } | |
d21ffe7e | 2349 | list_del(&req->inflight_entry); |
def596e9 | 2350 | |
bcda7baa | 2351 | if (req->flags & REQ_F_BUFFER_SELECTED) |
8ff069bf | 2352 | cflags = io_put_rw_kbuf(req); |
bcda7baa JA |
2353 | |
2354 | __io_cqring_fill_event(req, req->result, cflags); | |
def596e9 JA |
2355 | (*nr_events)++; |
2356 | ||
c3524383 | 2357 | if (refcount_dec_and_test(&req->refs)) |
2d6500d4 | 2358 | io_req_free_batch(&rb, req); |
def596e9 | 2359 | } |
def596e9 | 2360 | |
09bb8394 | 2361 | io_commit_cqring(ctx); |
32b2244a XW |
2362 | if (ctx->flags & IORING_SETUP_SQPOLL) |
2363 | io_cqring_ev_posted(ctx); | |
2d6500d4 | 2364 | io_req_free_batch_finish(ctx, &rb); |
581f9810 | 2365 | |
bbde017a XW |
2366 | if (!list_empty(&again)) |
2367 | io_iopoll_queue(&again); | |
581f9810 BM |
2368 | } |
2369 | ||
def596e9 JA |
2370 | static int io_do_iopoll(struct io_ring_ctx *ctx, unsigned int *nr_events, |
2371 | long min) | |
2372 | { | |
2373 | struct io_kiocb *req, *tmp; | |
2374 | LIST_HEAD(done); | |
2375 | bool spin; | |
2376 | int ret; | |
2377 | ||
2378 | /* | |
2379 | * Only spin for completions if we don't have multiple devices hanging | |
2380 | * off our complete list, and we're under the requested amount. | |
2381 | */ | |
2382 | spin = !ctx->poll_multi_file && *nr_events < min; | |
2383 | ||
2384 | ret = 0; | |
d21ffe7e | 2385 | list_for_each_entry_safe(req, tmp, &ctx->iopoll_list, inflight_entry) { |
9adbd45d | 2386 | struct kiocb *kiocb = &req->rw.kiocb; |
def596e9 JA |
2387 | |
2388 | /* | |
581f9810 BM |
2389 | * Move completed and retryable entries to our local lists. |
2390 | * If we find a request that requires polling, break out | |
2391 | * and complete those lists first, if we have entries there. | |
def596e9 | 2392 | */ |
65a6543d | 2393 | if (READ_ONCE(req->iopoll_completed)) { |
d21ffe7e | 2394 | list_move_tail(&req->inflight_entry, &done); |
def596e9 JA |
2395 | continue; |
2396 | } | |
2397 | if (!list_empty(&done)) | |
2398 | break; | |
2399 | ||
2400 | ret = kiocb->ki_filp->f_op->iopoll(kiocb, spin); | |
2401 | if (ret < 0) | |
2402 | break; | |
2403 | ||
3aadc23e PB |
2404 | /* iopoll may have completed current req */ |
2405 | if (READ_ONCE(req->iopoll_completed)) | |
d21ffe7e | 2406 | list_move_tail(&req->inflight_entry, &done); |
3aadc23e | 2407 | |
def596e9 JA |
2408 | if (ret && spin) |
2409 | spin = false; | |
2410 | ret = 0; | |
2411 | } | |
2412 | ||
2413 | if (!list_empty(&done)) | |
2414 | io_iopoll_complete(ctx, nr_events, &done); | |
2415 | ||
2416 | return ret; | |
2417 | } | |
2418 | ||
2419 | /* | |
d195a66e | 2420 | * Poll for a minimum of 'min' events. Note that if min == 0 we consider that a |
def596e9 JA |
2421 | * non-spinning poll check - we'll still enter the driver poll loop, but only |
2422 | * as a non-spinning completion check. | |
2423 | */ | |
2424 | static int io_iopoll_getevents(struct io_ring_ctx *ctx, unsigned int *nr_events, | |
2425 | long min) | |
2426 | { | |
540e32a0 | 2427 | while (!list_empty(&ctx->iopoll_list) && !need_resched()) { |
def596e9 JA |
2428 | int ret; |
2429 | ||
2430 | ret = io_do_iopoll(ctx, nr_events, min); | |
2431 | if (ret < 0) | |
2432 | return ret; | |
eba0a4dd | 2433 | if (*nr_events >= min) |
def596e9 JA |
2434 | return 0; |
2435 | } | |
2436 | ||
2437 | return 1; | |
2438 | } | |
2439 | ||
2440 | /* | |
2441 | * We can't just wait for polled events to come to us, we have to actively | |
2442 | * find and complete them. | |
2443 | */ | |
b2edc0a7 | 2444 | static void io_iopoll_try_reap_events(struct io_ring_ctx *ctx) |
def596e9 JA |
2445 | { |
2446 | if (!(ctx->flags & IORING_SETUP_IOPOLL)) | |
2447 | return; | |
2448 | ||
2449 | mutex_lock(&ctx->uring_lock); | |
540e32a0 | 2450 | while (!list_empty(&ctx->iopoll_list)) { |
def596e9 JA |
2451 | unsigned int nr_events = 0; |
2452 | ||
b2edc0a7 | 2453 | io_do_iopoll(ctx, &nr_events, 0); |
08f5439f | 2454 | |
b2edc0a7 PB |
2455 | /* let it sleep and repeat later if can't complete a request */ |
2456 | if (nr_events == 0) | |
2457 | break; | |
08f5439f JA |
2458 | /* |
2459 | * Ensure we allow local-to-the-cpu processing to take place, | |
2460 | * in this case we need to ensure that we reap all events. | |
3fcee5a6 | 2461 | * Also let task_work, etc. to progress by releasing the mutex |
08f5439f | 2462 | */ |
3fcee5a6 PB |
2463 | if (need_resched()) { |
2464 | mutex_unlock(&ctx->uring_lock); | |
2465 | cond_resched(); | |
2466 | mutex_lock(&ctx->uring_lock); | |
2467 | } | |
def596e9 JA |
2468 | } |
2469 | mutex_unlock(&ctx->uring_lock); | |
2470 | } | |
2471 | ||
7668b92a | 2472 | static int io_iopoll_check(struct io_ring_ctx *ctx, long min) |
def596e9 | 2473 | { |
7668b92a | 2474 | unsigned int nr_events = 0; |
2b2ed975 | 2475 | int iters = 0, ret = 0; |
500f9fba | 2476 | |
c7849be9 XW |
2477 | /* |
2478 | * We disallow the app entering submit/complete with polling, but we | |
2479 | * still need to lock the ring to prevent racing with polled issue | |
2480 | * that got punted to a workqueue. | |
2481 | */ | |
2482 | mutex_lock(&ctx->uring_lock); | |
def596e9 | 2483 | do { |
a3a0e43f JA |
2484 | /* |
2485 | * Don't enter poll loop if we already have events pending. | |
2486 | * If we do, we can potentially be spinning for commands that | |
2487 | * already triggered a CQE (eg in error). | |
2488 | */ | |
1d7bb1d5 | 2489 | if (io_cqring_events(ctx, false)) |
a3a0e43f JA |
2490 | break; |
2491 | ||
500f9fba JA |
2492 | /* |
2493 | * If a submit got punted to a workqueue, we can have the | |
2494 | * application entering polling for a command before it gets | |
2495 | * issued. That app will hold the uring_lock for the duration | |
2496 | * of the poll right here, so we need to take a breather every | |
2497 | * now and then to ensure that the issue has a chance to add | |
2498 | * the poll to the issued list. Otherwise we can spin here | |
2499 | * forever, while the workqueue is stuck trying to acquire the | |
2500 | * very same mutex. | |
2501 | */ | |
2502 | if (!(++iters & 7)) { | |
2503 | mutex_unlock(&ctx->uring_lock); | |
4c6e277c | 2504 | io_run_task_work(); |
500f9fba JA |
2505 | mutex_lock(&ctx->uring_lock); |
2506 | } | |
2507 | ||
7668b92a | 2508 | ret = io_iopoll_getevents(ctx, &nr_events, min); |
def596e9 JA |
2509 | if (ret <= 0) |
2510 | break; | |
2511 | ret = 0; | |
7668b92a | 2512 | } while (min && !nr_events && !need_resched()); |
def596e9 | 2513 | |
500f9fba | 2514 | mutex_unlock(&ctx->uring_lock); |
def596e9 JA |
2515 | return ret; |
2516 | } | |
2517 | ||
491381ce | 2518 | static void kiocb_end_write(struct io_kiocb *req) |
2b188cc1 | 2519 | { |
491381ce JA |
2520 | /* |
2521 | * Tell lockdep we inherited freeze protection from submission | |
2522 | * thread. | |
2523 | */ | |
2524 | if (req->flags & REQ_F_ISREG) { | |
2525 | struct inode *inode = file_inode(req->file); | |
2b188cc1 | 2526 | |
491381ce | 2527 | __sb_writers_acquired(inode->i_sb, SB_FREEZE_WRITE); |
2b188cc1 | 2528 | } |
491381ce | 2529 | file_end_write(req->file); |
2b188cc1 JA |
2530 | } |
2531 | ||
a1d7c393 JA |
2532 | static void io_complete_rw_common(struct kiocb *kiocb, long res, |
2533 | struct io_comp_state *cs) | |
2b188cc1 | 2534 | { |
9adbd45d | 2535 | struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb); |
bcda7baa | 2536 | int cflags = 0; |
2b188cc1 | 2537 | |
491381ce JA |
2538 | if (kiocb->ki_flags & IOCB_WRITE) |
2539 | kiocb_end_write(req); | |
2b188cc1 | 2540 | |
4e88d6e7 JA |
2541 | if (res != req->result) |
2542 | req_set_fail_links(req); | |
bcda7baa | 2543 | if (req->flags & REQ_F_BUFFER_SELECTED) |
8ff069bf | 2544 | cflags = io_put_rw_kbuf(req); |
a1d7c393 | 2545 | __io_req_complete(req, res, cflags, cs); |
ba816ad6 JA |
2546 | } |
2547 | ||
b63534c4 JA |
2548 | #ifdef CONFIG_BLOCK |
2549 | static bool io_resubmit_prep(struct io_kiocb *req, int error) | |
2550 | { | |
2551 | struct iovec inline_vecs[UIO_FASTIOV], *iovec = inline_vecs; | |
2552 | ssize_t ret = -ECANCELED; | |
2553 | struct iov_iter iter; | |
2554 | int rw; | |
2555 | ||
2556 | if (error) { | |
2557 | ret = error; | |
2558 | goto end_req; | |
2559 | } | |
2560 | ||
2561 | switch (req->opcode) { | |
2562 | case IORING_OP_READV: | |
2563 | case IORING_OP_READ_FIXED: | |
2564 | case IORING_OP_READ: | |
2565 | rw = READ; | |
2566 | break; | |
2567 | case IORING_OP_WRITEV: | |
2568 | case IORING_OP_WRITE_FIXED: | |
2569 | case IORING_OP_WRITE: | |
2570 | rw = WRITE; | |
2571 | break; | |
2572 | default: | |
2573 | printk_once(KERN_WARNING "io_uring: bad opcode in resubmit %d\n", | |
2574 | req->opcode); | |
2575 | goto end_req; | |
2576 | } | |
2577 | ||
e8c2bc1f | 2578 | if (!req->async_data) { |
8f3d7496 JA |
2579 | ret = io_import_iovec(rw, req, &iovec, &iter, false); |
2580 | if (ret < 0) | |
2581 | goto end_req; | |
2582 | ret = io_setup_async_rw(req, iovec, inline_vecs, &iter, false); | |
2583 | if (!ret) | |
2584 | return true; | |
2585 | kfree(iovec); | |
2586 | } else { | |
b63534c4 | 2587 | return true; |
8f3d7496 | 2588 | } |
b63534c4 | 2589 | end_req: |
b63534c4 | 2590 | req_set_fail_links(req); |
b63534c4 JA |
2591 | return false; |
2592 | } | |
b63534c4 JA |
2593 | #endif |
2594 | ||
2595 | static bool io_rw_reissue(struct io_kiocb *req, long res) | |
2596 | { | |
2597 | #ifdef CONFIG_BLOCK | |
355afaeb | 2598 | umode_t mode = file_inode(req->file)->i_mode; |
b63534c4 JA |
2599 | int ret; |
2600 | ||
355afaeb JA |
2601 | if (!S_ISBLK(mode) && !S_ISREG(mode)) |
2602 | return false; | |
b63534c4 JA |
2603 | if ((res != -EAGAIN && res != -EOPNOTSUPP) || io_wq_current_is_worker()) |
2604 | return false; | |
2605 | ||
fdee946d | 2606 | ret = io_sq_thread_acquire_mm(req->ctx, req); |
6d816e08 | 2607 | |
fdee946d JA |
2608 | if (io_resubmit_prep(req, ret)) { |
2609 | refcount_inc(&req->refs); | |
2610 | io_queue_async_work(req); | |
b63534c4 | 2611 | return true; |
fdee946d JA |
2612 | } |
2613 | ||
b63534c4 JA |
2614 | #endif |
2615 | return false; | |
2616 | } | |
2617 | ||
a1d7c393 JA |
2618 | static void __io_complete_rw(struct io_kiocb *req, long res, long res2, |
2619 | struct io_comp_state *cs) | |
2620 | { | |
2621 | if (!io_rw_reissue(req, res)) | |
2622 | io_complete_rw_common(&req->rw.kiocb, res, cs); | |
ba816ad6 JA |
2623 | } |
2624 | ||
2625 | static void io_complete_rw(struct kiocb *kiocb, long res, long res2) | |
2626 | { | |
9adbd45d | 2627 | struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb); |
ba816ad6 | 2628 | |
a1d7c393 | 2629 | __io_complete_rw(req, res, res2, NULL); |
2b188cc1 JA |
2630 | } |
2631 | ||
def596e9 JA |
2632 | static void io_complete_rw_iopoll(struct kiocb *kiocb, long res, long res2) |
2633 | { | |
9adbd45d | 2634 | struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb); |
def596e9 | 2635 | |
491381ce JA |
2636 | if (kiocb->ki_flags & IOCB_WRITE) |
2637 | kiocb_end_write(req); | |
def596e9 | 2638 | |
2d7d6792 | 2639 | if (res != -EAGAIN && res != req->result) |
4e88d6e7 | 2640 | req_set_fail_links(req); |
bbde017a XW |
2641 | |
2642 | WRITE_ONCE(req->result, res); | |
2643 | /* order with io_poll_complete() checking ->result */ | |
cd664b0e PB |
2644 | smp_wmb(); |
2645 | WRITE_ONCE(req->iopoll_completed, 1); | |
def596e9 JA |
2646 | } |
2647 | ||
2648 | /* | |
2649 | * After the iocb has been issued, it's safe to be found on the poll list. | |
2650 | * Adding the kiocb to the list AFTER submission ensures that we don't | |
2651 | * find it from a io_iopoll_getevents() thread before the issuer is done | |
2652 | * accessing the kiocb cookie. | |
2653 | */ | |
2654 | static void io_iopoll_req_issued(struct io_kiocb *req) | |
2655 | { | |
2656 | struct io_ring_ctx *ctx = req->ctx; | |
2657 | ||
2658 | /* | |
2659 | * Track whether we have multiple files in our lists. This will impact | |
2660 | * how we do polling eventually, not spinning if we're on potentially | |
2661 | * different devices. | |
2662 | */ | |
540e32a0 | 2663 | if (list_empty(&ctx->iopoll_list)) { |
def596e9 JA |
2664 | ctx->poll_multi_file = false; |
2665 | } else if (!ctx->poll_multi_file) { | |
2666 | struct io_kiocb *list_req; | |
2667 | ||
540e32a0 | 2668 | list_req = list_first_entry(&ctx->iopoll_list, struct io_kiocb, |
d21ffe7e | 2669 | inflight_entry); |
9adbd45d | 2670 | if (list_req->file != req->file) |
def596e9 JA |
2671 | ctx->poll_multi_file = true; |
2672 | } | |
2673 | ||
2674 | /* | |
2675 | * For fast devices, IO may have already completed. If it has, add | |
2676 | * it to the front so we find it first. | |
2677 | */ | |
65a6543d | 2678 | if (READ_ONCE(req->iopoll_completed)) |
d21ffe7e | 2679 | list_add(&req->inflight_entry, &ctx->iopoll_list); |
def596e9 | 2680 | else |
d21ffe7e | 2681 | list_add_tail(&req->inflight_entry, &ctx->iopoll_list); |
bdcd3eab XW |
2682 | |
2683 | if ((ctx->flags & IORING_SETUP_SQPOLL) && | |
534ca6d6 JA |
2684 | wq_has_sleeper(&ctx->sq_data->wait)) |
2685 | wake_up(&ctx->sq_data->wait); | |
def596e9 JA |
2686 | } |
2687 | ||
9f13c35b | 2688 | static void __io_state_file_put(struct io_submit_state *state) |
9a56a232 | 2689 | { |
06ef3608 PB |
2690 | if (state->has_refs) |
2691 | fput_many(state->file, state->has_refs); | |
9f13c35b PB |
2692 | state->file = NULL; |
2693 | } | |
2694 | ||
2695 | static inline void io_state_file_put(struct io_submit_state *state) | |
2696 | { | |
2697 | if (state->file) | |
2698 | __io_state_file_put(state); | |
9a56a232 JA |
2699 | } |
2700 | ||
2701 | /* | |
2702 | * Get as many references to a file as we have IOs left in this submission, | |
2703 | * assuming most submissions are for one file, or at least that each file | |
2704 | * has more than one submission. | |
2705 | */ | |
8da11c19 | 2706 | static struct file *__io_file_get(struct io_submit_state *state, int fd) |
9a56a232 JA |
2707 | { |
2708 | if (!state) | |
2709 | return fget(fd); | |
2710 | ||
2711 | if (state->file) { | |
2712 | if (state->fd == fd) { | |
06ef3608 | 2713 | state->has_refs--; |
9a56a232 JA |
2714 | return state->file; |
2715 | } | |
9f13c35b | 2716 | __io_state_file_put(state); |
9a56a232 JA |
2717 | } |
2718 | state->file = fget_many(fd, state->ios_left); | |
2719 | if (!state->file) | |
2720 | return NULL; | |
2721 | ||
2722 | state->fd = fd; | |
71b547c0 | 2723 | state->has_refs = state->ios_left - 1; |
9a56a232 JA |
2724 | return state->file; |
2725 | } | |
2726 | ||
4503b767 JA |
2727 | static bool io_bdev_nowait(struct block_device *bdev) |
2728 | { | |
2729 | #ifdef CONFIG_BLOCK | |
9ba0d0c8 | 2730 | return !bdev || blk_queue_nowait(bdev_get_queue(bdev)); |
4503b767 JA |
2731 | #else |
2732 | return true; | |
2733 | #endif | |
2734 | } | |
2735 | ||
2b188cc1 JA |
2736 | /* |
2737 | * If we tracked the file through the SCM inflight mechanism, we could support | |
2738 | * any file. For now, just ensure that anything potentially problematic is done | |
2739 | * inline. | |
2740 | */ | |
af197f50 | 2741 | static bool io_file_supports_async(struct file *file, int rw) |
2b188cc1 JA |
2742 | { |
2743 | umode_t mode = file_inode(file)->i_mode; | |
2744 | ||
4503b767 JA |
2745 | if (S_ISBLK(mode)) { |
2746 | if (io_bdev_nowait(file->f_inode->i_bdev)) | |
2747 | return true; | |
2748 | return false; | |
2749 | } | |
2750 | if (S_ISCHR(mode) || S_ISSOCK(mode)) | |
2b188cc1 | 2751 | return true; |
4503b767 JA |
2752 | if (S_ISREG(mode)) { |
2753 | if (io_bdev_nowait(file->f_inode->i_sb->s_bdev) && | |
2754 | file->f_op != &io_uring_fops) | |
2755 | return true; | |
2756 | return false; | |
2757 | } | |
2b188cc1 | 2758 | |
c5b85625 JA |
2759 | /* any ->read/write should understand O_NONBLOCK */ |
2760 | if (file->f_flags & O_NONBLOCK) | |
2761 | return true; | |
2762 | ||
af197f50 JA |
2763 | if (!(file->f_mode & FMODE_NOWAIT)) |
2764 | return false; | |
2765 | ||
2766 | if (rw == READ) | |
2767 | return file->f_op->read_iter != NULL; | |
2768 | ||
2769 | return file->f_op->write_iter != NULL; | |
2b188cc1 JA |
2770 | } |
2771 | ||
a88fc400 | 2772 | static int io_prep_rw(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
2b188cc1 | 2773 | { |
def596e9 | 2774 | struct io_ring_ctx *ctx = req->ctx; |
9adbd45d | 2775 | struct kiocb *kiocb = &req->rw.kiocb; |
09bb8394 JA |
2776 | unsigned ioprio; |
2777 | int ret; | |
2b188cc1 | 2778 | |
491381ce JA |
2779 | if (S_ISREG(file_inode(req->file)->i_mode)) |
2780 | req->flags |= REQ_F_ISREG; | |
2781 | ||
2b188cc1 | 2782 | kiocb->ki_pos = READ_ONCE(sqe->off); |
ba04291e JA |
2783 | if (kiocb->ki_pos == -1 && !(req->file->f_mode & FMODE_STREAM)) { |
2784 | req->flags |= REQ_F_CUR_POS; | |
2785 | kiocb->ki_pos = req->file->f_pos; | |
2786 | } | |
2b188cc1 | 2787 | kiocb->ki_hint = ki_hint_validate(file_write_hint(kiocb->ki_filp)); |
3e577dcd PB |
2788 | kiocb->ki_flags = iocb_flags(kiocb->ki_filp); |
2789 | ret = kiocb_set_rw_flags(kiocb, READ_ONCE(sqe->rw_flags)); | |
2790 | if (unlikely(ret)) | |
2791 | return ret; | |
2b188cc1 JA |
2792 | |
2793 | ioprio = READ_ONCE(sqe->ioprio); | |
2794 | if (ioprio) { | |
2795 | ret = ioprio_check_cap(ioprio); | |
2796 | if (ret) | |
09bb8394 | 2797 | return ret; |
2b188cc1 JA |
2798 | |
2799 | kiocb->ki_ioprio = ioprio; | |
2800 | } else | |
2801 | kiocb->ki_ioprio = get_current_ioprio(); | |
2802 | ||
8449eeda | 2803 | /* don't allow async punt if RWF_NOWAIT was requested */ |
c5b85625 | 2804 | if (kiocb->ki_flags & IOCB_NOWAIT) |
8449eeda SB |
2805 | req->flags |= REQ_F_NOWAIT; |
2806 | ||
def596e9 | 2807 | if (ctx->flags & IORING_SETUP_IOPOLL) { |
def596e9 JA |
2808 | if (!(kiocb->ki_flags & IOCB_DIRECT) || |
2809 | !kiocb->ki_filp->f_op->iopoll) | |
09bb8394 | 2810 | return -EOPNOTSUPP; |
2b188cc1 | 2811 | |
def596e9 JA |
2812 | kiocb->ki_flags |= IOCB_HIPRI; |
2813 | kiocb->ki_complete = io_complete_rw_iopoll; | |
65a6543d | 2814 | req->iopoll_completed = 0; |
def596e9 | 2815 | } else { |
09bb8394 JA |
2816 | if (kiocb->ki_flags & IOCB_HIPRI) |
2817 | return -EINVAL; | |
def596e9 JA |
2818 | kiocb->ki_complete = io_complete_rw; |
2819 | } | |
9adbd45d | 2820 | |
3529d8c2 JA |
2821 | req->rw.addr = READ_ONCE(sqe->addr); |
2822 | req->rw.len = READ_ONCE(sqe->len); | |
4f4eeba8 | 2823 | req->buf_index = READ_ONCE(sqe->buf_index); |
2b188cc1 | 2824 | return 0; |
2b188cc1 JA |
2825 | } |
2826 | ||
2827 | static inline void io_rw_done(struct kiocb *kiocb, ssize_t ret) | |
2828 | { | |
2829 | switch (ret) { | |
2830 | case -EIOCBQUEUED: | |
2831 | break; | |
2832 | case -ERESTARTSYS: | |
2833 | case -ERESTARTNOINTR: | |
2834 | case -ERESTARTNOHAND: | |
2835 | case -ERESTART_RESTARTBLOCK: | |
2836 | /* | |
2837 | * We can't just restart the syscall, since previously | |
2838 | * submitted sqes may already be in progress. Just fail this | |
2839 | * IO with EINTR. | |
2840 | */ | |
2841 | ret = -EINTR; | |
df561f66 | 2842 | fallthrough; |
2b188cc1 JA |
2843 | default: |
2844 | kiocb->ki_complete(kiocb, ret, 0); | |
2845 | } | |
2846 | } | |
2847 | ||
a1d7c393 JA |
2848 | static void kiocb_done(struct kiocb *kiocb, ssize_t ret, |
2849 | struct io_comp_state *cs) | |
ba816ad6 | 2850 | { |
ba04291e | 2851 | struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb); |
e8c2bc1f | 2852 | struct io_async_rw *io = req->async_data; |
ba04291e | 2853 | |
227c0c96 | 2854 | /* add previously done IO, if any */ |
e8c2bc1f | 2855 | if (io && io->bytes_done > 0) { |
227c0c96 | 2856 | if (ret < 0) |
e8c2bc1f | 2857 | ret = io->bytes_done; |
227c0c96 | 2858 | else |
e8c2bc1f | 2859 | ret += io->bytes_done; |
227c0c96 JA |
2860 | } |
2861 | ||
ba04291e JA |
2862 | if (req->flags & REQ_F_CUR_POS) |
2863 | req->file->f_pos = kiocb->ki_pos; | |
bcaec089 | 2864 | if (ret >= 0 && kiocb->ki_complete == io_complete_rw) |
a1d7c393 | 2865 | __io_complete_rw(req, ret, 0, cs); |
ba816ad6 JA |
2866 | else |
2867 | io_rw_done(kiocb, ret); | |
2868 | } | |
2869 | ||
9adbd45d | 2870 | static ssize_t io_import_fixed(struct io_kiocb *req, int rw, |
7d009165 | 2871 | struct iov_iter *iter) |
edafccee | 2872 | { |
9adbd45d JA |
2873 | struct io_ring_ctx *ctx = req->ctx; |
2874 | size_t len = req->rw.len; | |
edafccee | 2875 | struct io_mapped_ubuf *imu; |
4be1c615 | 2876 | u16 index, buf_index = req->buf_index; |
edafccee JA |
2877 | size_t offset; |
2878 | u64 buf_addr; | |
2879 | ||
edafccee JA |
2880 | if (unlikely(buf_index >= ctx->nr_user_bufs)) |
2881 | return -EFAULT; | |
edafccee JA |
2882 | index = array_index_nospec(buf_index, ctx->nr_user_bufs); |
2883 | imu = &ctx->user_bufs[index]; | |
9adbd45d | 2884 | buf_addr = req->rw.addr; |
edafccee JA |
2885 | |
2886 | /* overflow */ | |
2887 | if (buf_addr + len < buf_addr) | |
2888 | return -EFAULT; | |
2889 | /* not inside the mapped region */ | |
2890 | if (buf_addr < imu->ubuf || buf_addr + len > imu->ubuf + imu->len) | |
2891 | return -EFAULT; | |
2892 | ||
2893 | /* | |
2894 | * May not be a start of buffer, set size appropriately | |
2895 | * and advance us to the beginning. | |
2896 | */ | |
2897 | offset = buf_addr - imu->ubuf; | |
2898 | iov_iter_bvec(iter, rw, imu->bvec, imu->nr_bvecs, offset + len); | |
bd11b3a3 JA |
2899 | |
2900 | if (offset) { | |
2901 | /* | |
2902 | * Don't use iov_iter_advance() here, as it's really slow for | |
2903 | * using the latter parts of a big fixed buffer - it iterates | |
2904 | * over each segment manually. We can cheat a bit here, because | |
2905 | * we know that: | |
2906 | * | |
2907 | * 1) it's a BVEC iter, we set it up | |
2908 | * 2) all bvecs are PAGE_SIZE in size, except potentially the | |
2909 | * first and last bvec | |
2910 | * | |
2911 | * So just find our index, and adjust the iterator afterwards. | |
2912 | * If the offset is within the first bvec (or the whole first | |
2913 | * bvec, just use iov_iter_advance(). This makes it easier | |
2914 | * since we can just skip the first segment, which may not | |
2915 | * be PAGE_SIZE aligned. | |
2916 | */ | |
2917 | const struct bio_vec *bvec = imu->bvec; | |
2918 | ||
2919 | if (offset <= bvec->bv_len) { | |
2920 | iov_iter_advance(iter, offset); | |
2921 | } else { | |
2922 | unsigned long seg_skip; | |
2923 | ||
2924 | /* skip first vec */ | |
2925 | offset -= bvec->bv_len; | |
2926 | seg_skip = 1 + (offset >> PAGE_SHIFT); | |
2927 | ||
2928 | iter->bvec = bvec + seg_skip; | |
2929 | iter->nr_segs -= seg_skip; | |
99c79f66 | 2930 | iter->count -= bvec->bv_len + offset; |
bd11b3a3 | 2931 | iter->iov_offset = offset & ~PAGE_MASK; |
bd11b3a3 JA |
2932 | } |
2933 | } | |
2934 | ||
5e559561 | 2935 | return len; |
edafccee JA |
2936 | } |
2937 | ||
bcda7baa JA |
2938 | static void io_ring_submit_unlock(struct io_ring_ctx *ctx, bool needs_lock) |
2939 | { | |
2940 | if (needs_lock) | |
2941 | mutex_unlock(&ctx->uring_lock); | |
2942 | } | |
2943 | ||
2944 | static void io_ring_submit_lock(struct io_ring_ctx *ctx, bool needs_lock) | |
2945 | { | |
2946 | /* | |
2947 | * "Normal" inline submissions always hold the uring_lock, since we | |
2948 | * grab it from the system call. Same is true for the SQPOLL offload. | |
2949 | * The only exception is when we've detached the request and issue it | |
2950 | * from an async worker thread, grab the lock for that case. | |
2951 | */ | |
2952 | if (needs_lock) | |
2953 | mutex_lock(&ctx->uring_lock); | |
2954 | } | |
2955 | ||
2956 | static struct io_buffer *io_buffer_select(struct io_kiocb *req, size_t *len, | |
2957 | int bgid, struct io_buffer *kbuf, | |
2958 | bool needs_lock) | |
2959 | { | |
2960 | struct io_buffer *head; | |
2961 | ||
2962 | if (req->flags & REQ_F_BUFFER_SELECTED) | |
2963 | return kbuf; | |
2964 | ||
2965 | io_ring_submit_lock(req->ctx, needs_lock); | |
2966 | ||
2967 | lockdep_assert_held(&req->ctx->uring_lock); | |
2968 | ||
2969 | head = idr_find(&req->ctx->io_buffer_idr, bgid); | |
2970 | if (head) { | |
2971 | if (!list_empty(&head->list)) { | |
2972 | kbuf = list_last_entry(&head->list, struct io_buffer, | |
2973 | list); | |
2974 | list_del(&kbuf->list); | |
2975 | } else { | |
2976 | kbuf = head; | |
2977 | idr_remove(&req->ctx->io_buffer_idr, bgid); | |
2978 | } | |
2979 | if (*len > kbuf->len) | |
2980 | *len = kbuf->len; | |
2981 | } else { | |
2982 | kbuf = ERR_PTR(-ENOBUFS); | |
2983 | } | |
2984 | ||
2985 | io_ring_submit_unlock(req->ctx, needs_lock); | |
2986 | ||
2987 | return kbuf; | |
2988 | } | |
2989 | ||
4d954c25 JA |
2990 | static void __user *io_rw_buffer_select(struct io_kiocb *req, size_t *len, |
2991 | bool needs_lock) | |
2992 | { | |
2993 | struct io_buffer *kbuf; | |
4f4eeba8 | 2994 | u16 bgid; |
4d954c25 JA |
2995 | |
2996 | kbuf = (struct io_buffer *) (unsigned long) req->rw.addr; | |
4f4eeba8 | 2997 | bgid = req->buf_index; |
4d954c25 JA |
2998 | kbuf = io_buffer_select(req, len, bgid, kbuf, needs_lock); |
2999 | if (IS_ERR(kbuf)) | |
3000 | return kbuf; | |
3001 | req->rw.addr = (u64) (unsigned long) kbuf; | |
3002 | req->flags |= REQ_F_BUFFER_SELECTED; | |
3003 | return u64_to_user_ptr(kbuf->addr); | |
3004 | } | |
3005 | ||
3006 | #ifdef CONFIG_COMPAT | |
3007 | static ssize_t io_compat_import(struct io_kiocb *req, struct iovec *iov, | |
3008 | bool needs_lock) | |
3009 | { | |
3010 | struct compat_iovec __user *uiov; | |
3011 | compat_ssize_t clen; | |
3012 | void __user *buf; | |
3013 | ssize_t len; | |
3014 | ||
3015 | uiov = u64_to_user_ptr(req->rw.addr); | |
3016 | if (!access_ok(uiov, sizeof(*uiov))) | |
3017 | return -EFAULT; | |
3018 | if (__get_user(clen, &uiov->iov_len)) | |
3019 | return -EFAULT; | |
3020 | if (clen < 0) | |
3021 | return -EINVAL; | |
3022 | ||
3023 | len = clen; | |
3024 | buf = io_rw_buffer_select(req, &len, needs_lock); | |
3025 | if (IS_ERR(buf)) | |
3026 | return PTR_ERR(buf); | |
3027 | iov[0].iov_base = buf; | |
3028 | iov[0].iov_len = (compat_size_t) len; | |
3029 | return 0; | |
3030 | } | |
3031 | #endif | |
3032 | ||
3033 | static ssize_t __io_iov_buffer_select(struct io_kiocb *req, struct iovec *iov, | |
3034 | bool needs_lock) | |
3035 | { | |
3036 | struct iovec __user *uiov = u64_to_user_ptr(req->rw.addr); | |
3037 | void __user *buf; | |
3038 | ssize_t len; | |
3039 | ||
3040 | if (copy_from_user(iov, uiov, sizeof(*uiov))) | |
3041 | return -EFAULT; | |
3042 | ||
3043 | len = iov[0].iov_len; | |
3044 | if (len < 0) | |
3045 | return -EINVAL; | |
3046 | buf = io_rw_buffer_select(req, &len, needs_lock); | |
3047 | if (IS_ERR(buf)) | |
3048 | return PTR_ERR(buf); | |
3049 | iov[0].iov_base = buf; | |
3050 | iov[0].iov_len = len; | |
3051 | return 0; | |
3052 | } | |
3053 | ||
3054 | static ssize_t io_iov_buffer_select(struct io_kiocb *req, struct iovec *iov, | |
3055 | bool needs_lock) | |
3056 | { | |
dddb3e26 JA |
3057 | if (req->flags & REQ_F_BUFFER_SELECTED) { |
3058 | struct io_buffer *kbuf; | |
3059 | ||
3060 | kbuf = (struct io_buffer *) (unsigned long) req->rw.addr; | |
3061 | iov[0].iov_base = u64_to_user_ptr(kbuf->addr); | |
3062 | iov[0].iov_len = kbuf->len; | |
4d954c25 | 3063 | return 0; |
dddb3e26 | 3064 | } |
4d954c25 JA |
3065 | if (!req->rw.len) |
3066 | return 0; | |
3067 | else if (req->rw.len > 1) | |
3068 | return -EINVAL; | |
3069 | ||
3070 | #ifdef CONFIG_COMPAT | |
3071 | if (req->ctx->compat) | |
3072 | return io_compat_import(req, iov, needs_lock); | |
3073 | #endif | |
3074 | ||
3075 | return __io_iov_buffer_select(req, iov, needs_lock); | |
3076 | } | |
3077 | ||
8452fd0c JA |
3078 | static ssize_t __io_import_iovec(int rw, struct io_kiocb *req, |
3079 | struct iovec **iovec, struct iov_iter *iter, | |
3080 | bool needs_lock) | |
2b188cc1 | 3081 | { |
9adbd45d JA |
3082 | void __user *buf = u64_to_user_ptr(req->rw.addr); |
3083 | size_t sqe_len = req->rw.len; | |
4d954c25 | 3084 | ssize_t ret; |
edafccee JA |
3085 | u8 opcode; |
3086 | ||
d625c6ee | 3087 | opcode = req->opcode; |
7d009165 | 3088 | if (opcode == IORING_OP_READ_FIXED || opcode == IORING_OP_WRITE_FIXED) { |
edafccee | 3089 | *iovec = NULL; |
9adbd45d | 3090 | return io_import_fixed(req, rw, iter); |
edafccee | 3091 | } |
2b188cc1 | 3092 | |
bcda7baa | 3093 | /* buffer index only valid with fixed read/write, or buffer select */ |
4f4eeba8 | 3094 | if (req->buf_index && !(req->flags & REQ_F_BUFFER_SELECT)) |
9adbd45d JA |
3095 | return -EINVAL; |
3096 | ||
3a6820f2 | 3097 | if (opcode == IORING_OP_READ || opcode == IORING_OP_WRITE) { |
bcda7baa | 3098 | if (req->flags & REQ_F_BUFFER_SELECT) { |
4d954c25 | 3099 | buf = io_rw_buffer_select(req, &sqe_len, needs_lock); |
867a23ea | 3100 | if (IS_ERR(buf)) |
4d954c25 | 3101 | return PTR_ERR(buf); |
3f9d6441 | 3102 | req->rw.len = sqe_len; |
bcda7baa JA |
3103 | } |
3104 | ||
3a6820f2 JA |
3105 | ret = import_single_range(rw, buf, sqe_len, *iovec, iter); |
3106 | *iovec = NULL; | |
3a901598 | 3107 | return ret < 0 ? ret : sqe_len; |
3a6820f2 JA |
3108 | } |
3109 | ||
4d954c25 JA |
3110 | if (req->flags & REQ_F_BUFFER_SELECT) { |
3111 | ret = io_iov_buffer_select(req, *iovec, needs_lock); | |
3f9d6441 JA |
3112 | if (!ret) { |
3113 | ret = (*iovec)->iov_len; | |
3114 | iov_iter_init(iter, rw, *iovec, 1, ret); | |
3115 | } | |
4d954c25 JA |
3116 | *iovec = NULL; |
3117 | return ret; | |
3118 | } | |
3119 | ||
89cd35c5 CH |
3120 | return __import_iovec(rw, buf, sqe_len, UIO_FASTIOV, iovec, iter, |
3121 | req->ctx->compat); | |
2b188cc1 JA |
3122 | } |
3123 | ||
8452fd0c JA |
3124 | static ssize_t io_import_iovec(int rw, struct io_kiocb *req, |
3125 | struct iovec **iovec, struct iov_iter *iter, | |
3126 | bool needs_lock) | |
3127 | { | |
e8c2bc1f JA |
3128 | struct io_async_rw *iorw = req->async_data; |
3129 | ||
3130 | if (!iorw) | |
8452fd0c JA |
3131 | return __io_import_iovec(rw, req, iovec, iter, needs_lock); |
3132 | *iovec = NULL; | |
e8c2bc1f | 3133 | return iov_iter_count(&iorw->iter); |
8452fd0c JA |
3134 | } |
3135 | ||
0fef9483 JA |
3136 | static inline loff_t *io_kiocb_ppos(struct kiocb *kiocb) |
3137 | { | |
5b09e37e | 3138 | return (kiocb->ki_filp->f_mode & FMODE_STREAM) ? NULL : &kiocb->ki_pos; |
0fef9483 JA |
3139 | } |
3140 | ||
31b51510 | 3141 | /* |
32960613 JA |
3142 | * For files that don't have ->read_iter() and ->write_iter(), handle them |
3143 | * by looping over ->read() or ->write() manually. | |
31b51510 | 3144 | */ |
4017eb91 | 3145 | static ssize_t loop_rw_iter(int rw, struct io_kiocb *req, struct iov_iter *iter) |
32960613 | 3146 | { |
4017eb91 JA |
3147 | struct kiocb *kiocb = &req->rw.kiocb; |
3148 | struct file *file = req->file; | |
32960613 JA |
3149 | ssize_t ret = 0; |
3150 | ||
3151 | /* | |
3152 | * Don't support polled IO through this interface, and we can't | |
3153 | * support non-blocking either. For the latter, this just causes | |
3154 | * the kiocb to be handled from an async context. | |
3155 | */ | |
3156 | if (kiocb->ki_flags & IOCB_HIPRI) | |
3157 | return -EOPNOTSUPP; | |
3158 | if (kiocb->ki_flags & IOCB_NOWAIT) | |
3159 | return -EAGAIN; | |
3160 | ||
3161 | while (iov_iter_count(iter)) { | |
311ae9e1 | 3162 | struct iovec iovec; |
32960613 JA |
3163 | ssize_t nr; |
3164 | ||
311ae9e1 PB |
3165 | if (!iov_iter_is_bvec(iter)) { |
3166 | iovec = iov_iter_iovec(iter); | |
3167 | } else { | |
4017eb91 JA |
3168 | iovec.iov_base = u64_to_user_ptr(req->rw.addr); |
3169 | iovec.iov_len = req->rw.len; | |
311ae9e1 PB |
3170 | } |
3171 | ||
32960613 JA |
3172 | if (rw == READ) { |
3173 | nr = file->f_op->read(file, iovec.iov_base, | |
0fef9483 | 3174 | iovec.iov_len, io_kiocb_ppos(kiocb)); |
32960613 JA |
3175 | } else { |
3176 | nr = file->f_op->write(file, iovec.iov_base, | |
0fef9483 | 3177 | iovec.iov_len, io_kiocb_ppos(kiocb)); |
32960613 JA |
3178 | } |
3179 | ||
3180 | if (nr < 0) { | |
3181 | if (!ret) | |
3182 | ret = nr; | |
3183 | break; | |
3184 | } | |
3185 | ret += nr; | |
3186 | if (nr != iovec.iov_len) | |
3187 | break; | |
4017eb91 JA |
3188 | req->rw.len -= nr; |
3189 | req->rw.addr += nr; | |
32960613 JA |
3190 | iov_iter_advance(iter, nr); |
3191 | } | |
3192 | ||
3193 | return ret; | |
3194 | } | |
3195 | ||
ff6165b2 JA |
3196 | static void io_req_map_rw(struct io_kiocb *req, const struct iovec *iovec, |
3197 | const struct iovec *fast_iov, struct iov_iter *iter) | |
f67676d1 | 3198 | { |
e8c2bc1f | 3199 | struct io_async_rw *rw = req->async_data; |
b64e3444 | 3200 | |
ff6165b2 | 3201 | memcpy(&rw->iter, iter, sizeof(*iter)); |
afb87658 | 3202 | rw->free_iovec = iovec; |
227c0c96 | 3203 | rw->bytes_done = 0; |
ff6165b2 JA |
3204 | /* can only be fixed buffers, no need to do anything */ |
3205 | if (iter->type == ITER_BVEC) | |
3206 | return; | |
b64e3444 | 3207 | if (!iovec) { |
ff6165b2 JA |
3208 | unsigned iov_off = 0; |
3209 | ||
3210 | rw->iter.iov = rw->fast_iov; | |
3211 | if (iter->iov != fast_iov) { | |
3212 | iov_off = iter->iov - fast_iov; | |
3213 | rw->iter.iov += iov_off; | |
3214 | } | |
3215 | if (rw->fast_iov != fast_iov) | |
3216 | memcpy(rw->fast_iov + iov_off, fast_iov + iov_off, | |
45097dae | 3217 | sizeof(struct iovec) * iter->nr_segs); |
99bc4c38 PB |
3218 | } else { |
3219 | req->flags |= REQ_F_NEED_CLEANUP; | |
f67676d1 JA |
3220 | } |
3221 | } | |
3222 | ||
e8c2bc1f | 3223 | static inline int __io_alloc_async_data(struct io_kiocb *req) |
3d9932a8 | 3224 | { |
e8c2bc1f JA |
3225 | WARN_ON_ONCE(!io_op_defs[req->opcode].async_size); |
3226 | req->async_data = kmalloc(io_op_defs[req->opcode].async_size, GFP_KERNEL); | |
3227 | return req->async_data == NULL; | |
3d9932a8 XW |
3228 | } |
3229 | ||
e8c2bc1f | 3230 | static int io_alloc_async_data(struct io_kiocb *req) |
f67676d1 | 3231 | { |
e8c2bc1f | 3232 | if (!io_op_defs[req->opcode].needs_async_data) |
d3656344 | 3233 | return 0; |
3d9932a8 | 3234 | |
e8c2bc1f | 3235 | return __io_alloc_async_data(req); |
b7bb4f7d JA |
3236 | } |
3237 | ||
ff6165b2 JA |
3238 | static int io_setup_async_rw(struct io_kiocb *req, const struct iovec *iovec, |
3239 | const struct iovec *fast_iov, | |
227c0c96 | 3240 | struct iov_iter *iter, bool force) |
b7bb4f7d | 3241 | { |
e8c2bc1f | 3242 | if (!force && !io_op_defs[req->opcode].needs_async_data) |
74566df3 | 3243 | return 0; |
e8c2bc1f JA |
3244 | if (!req->async_data) { |
3245 | if (__io_alloc_async_data(req)) | |
5d204bcf | 3246 | return -ENOMEM; |
b7bb4f7d | 3247 | |
ff6165b2 | 3248 | io_req_map_rw(req, iovec, fast_iov, iter); |
5d204bcf | 3249 | } |
b7bb4f7d | 3250 | return 0; |
f67676d1 JA |
3251 | } |
3252 | ||
73debe68 | 3253 | static inline int io_rw_prep_async(struct io_kiocb *req, int rw) |
c3e330a4 | 3254 | { |
e8c2bc1f | 3255 | struct io_async_rw *iorw = req->async_data; |
f4bff104 | 3256 | struct iovec *iov = iorw->fast_iov; |
c3e330a4 PB |
3257 | ssize_t ret; |
3258 | ||
73debe68 | 3259 | ret = __io_import_iovec(rw, req, &iov, &iorw->iter, false); |
c3e330a4 PB |
3260 | if (unlikely(ret < 0)) |
3261 | return ret; | |
3262 | ||
ab0b196c PB |
3263 | iorw->bytes_done = 0; |
3264 | iorw->free_iovec = iov; | |
3265 | if (iov) | |
3266 | req->flags |= REQ_F_NEED_CLEANUP; | |
c3e330a4 PB |
3267 | return 0; |
3268 | } | |
3269 | ||
73debe68 | 3270 | static int io_read_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
f67676d1 JA |
3271 | { |
3272 | ssize_t ret; | |
3273 | ||
a88fc400 | 3274 | ret = io_prep_rw(req, sqe); |
3529d8c2 JA |
3275 | if (ret) |
3276 | return ret; | |
f67676d1 | 3277 | |
3529d8c2 JA |
3278 | if (unlikely(!(req->file->f_mode & FMODE_READ))) |
3279 | return -EBADF; | |
f67676d1 | 3280 | |
5f798bea | 3281 | /* either don't need iovec imported or already have it */ |
2d199895 | 3282 | if (!req->async_data) |
3529d8c2 | 3283 | return 0; |
73debe68 | 3284 | return io_rw_prep_async(req, READ); |
f67676d1 JA |
3285 | } |
3286 | ||
c1dd91d1 JA |
3287 | /* |
3288 | * This is our waitqueue callback handler, registered through lock_page_async() | |
3289 | * when we initially tried to do the IO with the iocb armed our waitqueue. | |
3290 | * This gets called when the page is unlocked, and we generally expect that to | |
3291 | * happen when the page IO is completed and the page is now uptodate. This will | |
3292 | * queue a task_work based retry of the operation, attempting to copy the data | |
3293 | * again. If the latter fails because the page was NOT uptodate, then we will | |
3294 | * do a thread based blocking retry of the operation. That's the unexpected | |
3295 | * slow path. | |
3296 | */ | |
bcf5a063 JA |
3297 | static int io_async_buf_func(struct wait_queue_entry *wait, unsigned mode, |
3298 | int sync, void *arg) | |
3299 | { | |
3300 | struct wait_page_queue *wpq; | |
3301 | struct io_kiocb *req = wait->private; | |
bcf5a063 | 3302 | struct wait_page_key *key = arg; |
bcf5a063 JA |
3303 | int ret; |
3304 | ||
3305 | wpq = container_of(wait, struct wait_page_queue, wait); | |
3306 | ||
cdc8fcb4 LT |
3307 | if (!wake_page_match(wpq, key)) |
3308 | return 0; | |
3309 | ||
c8d317aa | 3310 | req->rw.kiocb.ki_flags &= ~IOCB_WAITQ; |
bcf5a063 JA |
3311 | list_del_init(&wait->entry); |
3312 | ||
e7375122 | 3313 | init_task_work(&req->task_work, io_req_task_submit); |
6d816e08 JA |
3314 | percpu_ref_get(&req->ctx->refs); |
3315 | ||
bcf5a063 JA |
3316 | /* submit ref gets dropped, acquire a new one */ |
3317 | refcount_inc(&req->refs); | |
87c4311f | 3318 | ret = io_req_task_work_add(req, true); |
bcf5a063 | 3319 | if (unlikely(ret)) { |
c2c4c83c JA |
3320 | struct task_struct *tsk; |
3321 | ||
bcf5a063 | 3322 | /* queue just for cancelation */ |
e7375122 | 3323 | init_task_work(&req->task_work, io_req_task_cancel); |
bcf5a063 | 3324 | tsk = io_wq_get_task(req->ctx->io_wq); |
91989c70 | 3325 | task_work_add(tsk, &req->task_work, TWA_NONE); |
c2c4c83c | 3326 | wake_up_process(tsk); |
bcf5a063 | 3327 | } |
bcf5a063 JA |
3328 | return 1; |
3329 | } | |
3330 | ||
c1dd91d1 JA |
3331 | /* |
3332 | * This controls whether a given IO request should be armed for async page | |
3333 | * based retry. If we return false here, the request is handed to the async | |
3334 | * worker threads for retry. If we're doing buffered reads on a regular file, | |
3335 | * we prepare a private wait_page_queue entry and retry the operation. This | |
3336 | * will either succeed because the page is now uptodate and unlocked, or it | |
3337 | * will register a callback when the page is unlocked at IO completion. Through | |
3338 | * that callback, io_uring uses task_work to setup a retry of the operation. | |
3339 | * That retry will attempt the buffered read again. The retry will generally | |
3340 | * succeed, or in rare cases where it fails, we then fall back to using the | |
3341 | * async worker threads for a blocking retry. | |
3342 | */ | |
227c0c96 | 3343 | static bool io_rw_should_retry(struct io_kiocb *req) |
f67676d1 | 3344 | { |
e8c2bc1f JA |
3345 | struct io_async_rw *rw = req->async_data; |
3346 | struct wait_page_queue *wait = &rw->wpq; | |
bcf5a063 | 3347 | struct kiocb *kiocb = &req->rw.kiocb; |
f67676d1 | 3348 | |
bcf5a063 JA |
3349 | /* never retry for NOWAIT, we just complete with -EAGAIN */ |
3350 | if (req->flags & REQ_F_NOWAIT) | |
3351 | return false; | |
f67676d1 | 3352 | |
227c0c96 | 3353 | /* Only for buffered IO */ |
3b2a4439 | 3354 | if (kiocb->ki_flags & (IOCB_DIRECT | IOCB_HIPRI)) |
bcf5a063 | 3355 | return false; |
3b2a4439 | 3356 | |
bcf5a063 JA |
3357 | /* |
3358 | * just use poll if we can, and don't attempt if the fs doesn't | |
3359 | * support callback based unlocks | |
3360 | */ | |
3361 | if (file_can_poll(req->file) || !(req->file->f_mode & FMODE_BUF_RASYNC)) | |
3362 | return false; | |
f67676d1 | 3363 | |
3b2a4439 JA |
3364 | wait->wait.func = io_async_buf_func; |
3365 | wait->wait.private = req; | |
3366 | wait->wait.flags = 0; | |
3367 | INIT_LIST_HEAD(&wait->wait.entry); | |
3368 | kiocb->ki_flags |= IOCB_WAITQ; | |
c8d317aa | 3369 | kiocb->ki_flags &= ~IOCB_NOWAIT; |
3b2a4439 | 3370 | kiocb->ki_waitq = wait; |
3b2a4439 | 3371 | return true; |
bcf5a063 JA |
3372 | } |
3373 | ||
3374 | static int io_iter_do_read(struct io_kiocb *req, struct iov_iter *iter) | |
3375 | { | |
3376 | if (req->file->f_op->read_iter) | |
3377 | return call_read_iter(req->file, &req->rw.kiocb, iter); | |
2dd2111d | 3378 | else if (req->file->f_op->read) |
4017eb91 | 3379 | return loop_rw_iter(READ, req, iter); |
2dd2111d GH |
3380 | else |
3381 | return -EINVAL; | |
f67676d1 JA |
3382 | } |
3383 | ||
a1d7c393 JA |
3384 | static int io_read(struct io_kiocb *req, bool force_nonblock, |
3385 | struct io_comp_state *cs) | |
2b188cc1 JA |
3386 | { |
3387 | struct iovec inline_vecs[UIO_FASTIOV], *iovec = inline_vecs; | |
9adbd45d | 3388 | struct kiocb *kiocb = &req->rw.kiocb; |
ff6165b2 | 3389 | struct iov_iter __iter, *iter = &__iter; |
e8c2bc1f | 3390 | struct io_async_rw *rw = req->async_data; |
227c0c96 | 3391 | ssize_t io_size, ret, ret2; |
31b51510 | 3392 | size_t iov_count; |
f5cac8b1 | 3393 | bool no_async; |
ff6165b2 | 3394 | |
e8c2bc1f JA |
3395 | if (rw) |
3396 | iter = &rw->iter; | |
2b188cc1 | 3397 | |
ff6165b2 | 3398 | ret = io_import_iovec(READ, req, &iovec, iter, !force_nonblock); |
06b76d44 JA |
3399 | if (ret < 0) |
3400 | return ret; | |
eefdf30f | 3401 | iov_count = iov_iter_count(iter); |
fa15bafb PB |
3402 | io_size = ret; |
3403 | req->result = io_size; | |
227c0c96 | 3404 | ret = 0; |
2b188cc1 | 3405 | |
fd6c2e4c JA |
3406 | /* Ensure we clear previously set non-block flag */ |
3407 | if (!force_nonblock) | |
29de5f6a | 3408 | kiocb->ki_flags &= ~IOCB_NOWAIT; |
a88fc400 PB |
3409 | else |
3410 | kiocb->ki_flags |= IOCB_NOWAIT; | |
3411 | ||
fd6c2e4c | 3412 | |
24c74678 | 3413 | /* If the file doesn't support async, just async punt */ |
f5cac8b1 JA |
3414 | no_async = force_nonblock && !io_file_supports_async(req->file, READ); |
3415 | if (no_async) | |
f67676d1 | 3416 | goto copy_iov; |
9e645e11 | 3417 | |
0fef9483 | 3418 | ret = rw_verify_area(READ, req->file, io_kiocb_ppos(kiocb), iov_count); |
fa15bafb PB |
3419 | if (unlikely(ret)) |
3420 | goto out_free; | |
2b188cc1 | 3421 | |
227c0c96 | 3422 | ret = io_iter_do_read(req, iter); |
32960613 | 3423 | |
227c0c96 JA |
3424 | if (!ret) { |
3425 | goto done; | |
3426 | } else if (ret == -EIOCBQUEUED) { | |
3427 | ret = 0; | |
3428 | goto out_free; | |
3429 | } else if (ret == -EAGAIN) { | |
eefdf30f JA |
3430 | /* IOPOLL retry should happen for io-wq threads */ |
3431 | if (!force_nonblock && !(req->ctx->flags & IORING_SETUP_IOPOLL)) | |
f91daf56 | 3432 | goto done; |
355afaeb JA |
3433 | /* no retry on NONBLOCK marked file */ |
3434 | if (req->file->f_flags & O_NONBLOCK) | |
3435 | goto done; | |
84216315 JA |
3436 | /* some cases will consume bytes even on error returns */ |
3437 | iov_iter_revert(iter, iov_count - iov_iter_count(iter)); | |
f38c7e3a JA |
3438 | ret = 0; |
3439 | goto copy_iov; | |
227c0c96 | 3440 | } else if (ret < 0) { |
00d23d51 JA |
3441 | /* make sure -ERESTARTSYS -> -EINTR is done */ |
3442 | goto done; | |
227c0c96 JA |
3443 | } |
3444 | ||
3445 | /* read it all, or we did blocking attempt. no retry. */ | |
f91daf56 JA |
3446 | if (!iov_iter_count(iter) || !force_nonblock || |
3447 | (req->file->f_flags & O_NONBLOCK)) | |
227c0c96 JA |
3448 | goto done; |
3449 | ||
3450 | io_size -= ret; | |
3451 | copy_iov: | |
3452 | ret2 = io_setup_async_rw(req, iovec, inline_vecs, iter, true); | |
3453 | if (ret2) { | |
3454 | ret = ret2; | |
3455 | goto out_free; | |
3456 | } | |
f5cac8b1 JA |
3457 | if (no_async) |
3458 | return -EAGAIN; | |
e8c2bc1f | 3459 | rw = req->async_data; |
227c0c96 JA |
3460 | /* it's copied and will be cleaned with ->io */ |
3461 | iovec = NULL; | |
3462 | /* now use our persistent iterator, if we aren't already */ | |
e8c2bc1f | 3463 | iter = &rw->iter; |
227c0c96 | 3464 | retry: |
e8c2bc1f | 3465 | rw->bytes_done += ret; |
227c0c96 JA |
3466 | /* if we can retry, do so with the callbacks armed */ |
3467 | if (!io_rw_should_retry(req)) { | |
fa15bafb PB |
3468 | kiocb->ki_flags &= ~IOCB_WAITQ; |
3469 | return -EAGAIN; | |
2b188cc1 | 3470 | } |
227c0c96 JA |
3471 | |
3472 | /* | |
3473 | * Now retry read with the IOCB_WAITQ parts set in the iocb. If we | |
3474 | * get -EIOCBQUEUED, then we'll get a notification when the desired | |
3475 | * page gets unlocked. We can also get a partial read here, and if we | |
3476 | * do, then just retry at the new offset. | |
3477 | */ | |
3478 | ret = io_iter_do_read(req, iter); | |
3479 | if (ret == -EIOCBQUEUED) { | |
3480 | ret = 0; | |
3481 | goto out_free; | |
3482 | } else if (ret > 0 && ret < io_size) { | |
3483 | /* we got some bytes, but not all. retry. */ | |
3484 | goto retry; | |
3485 | } | |
3486 | done: | |
3487 | kiocb_done(kiocb, ret, cs); | |
3488 | ret = 0; | |
f67676d1 | 3489 | out_free: |
f261c168 | 3490 | /* it's reportedly faster than delegating the null check to kfree() */ |
252917c3 | 3491 | if (iovec) |
6f2cc166 | 3492 | kfree(iovec); |
2b188cc1 JA |
3493 | return ret; |
3494 | } | |
3495 | ||
73debe68 | 3496 | static int io_write_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
f67676d1 JA |
3497 | { |
3498 | ssize_t ret; | |
3499 | ||
a88fc400 | 3500 | ret = io_prep_rw(req, sqe); |
3529d8c2 JA |
3501 | if (ret) |
3502 | return ret; | |
f67676d1 | 3503 | |
3529d8c2 JA |
3504 | if (unlikely(!(req->file->f_mode & FMODE_WRITE))) |
3505 | return -EBADF; | |
f67676d1 | 3506 | |
5f798bea | 3507 | /* either don't need iovec imported or already have it */ |
2d199895 | 3508 | if (!req->async_data) |
3529d8c2 | 3509 | return 0; |
73debe68 | 3510 | return io_rw_prep_async(req, WRITE); |
f67676d1 JA |
3511 | } |
3512 | ||
a1d7c393 JA |
3513 | static int io_write(struct io_kiocb *req, bool force_nonblock, |
3514 | struct io_comp_state *cs) | |
2b188cc1 JA |
3515 | { |
3516 | struct iovec inline_vecs[UIO_FASTIOV], *iovec = inline_vecs; | |
9adbd45d | 3517 | struct kiocb *kiocb = &req->rw.kiocb; |
ff6165b2 | 3518 | struct iov_iter __iter, *iter = &__iter; |
e8c2bc1f | 3519 | struct io_async_rw *rw = req->async_data; |
31b51510 | 3520 | size_t iov_count; |
fa15bafb | 3521 | ssize_t ret, ret2, io_size; |
2b188cc1 | 3522 | |
e8c2bc1f JA |
3523 | if (rw) |
3524 | iter = &rw->iter; | |
ff6165b2 JA |
3525 | |
3526 | ret = io_import_iovec(WRITE, req, &iovec, iter, !force_nonblock); | |
06b76d44 JA |
3527 | if (ret < 0) |
3528 | return ret; | |
eefdf30f | 3529 | iov_count = iov_iter_count(iter); |
fa15bafb PB |
3530 | io_size = ret; |
3531 | req->result = io_size; | |
2b188cc1 | 3532 | |
fd6c2e4c JA |
3533 | /* Ensure we clear previously set non-block flag */ |
3534 | if (!force_nonblock) | |
a88fc400 PB |
3535 | kiocb->ki_flags &= ~IOCB_NOWAIT; |
3536 | else | |
3537 | kiocb->ki_flags |= IOCB_NOWAIT; | |
fd6c2e4c | 3538 | |
24c74678 | 3539 | /* If the file doesn't support async, just async punt */ |
af197f50 | 3540 | if (force_nonblock && !io_file_supports_async(req->file, WRITE)) |
f67676d1 | 3541 | goto copy_iov; |
31b51510 | 3542 | |
10d59345 JA |
3543 | /* file path doesn't support NOWAIT for non-direct_IO */ |
3544 | if (force_nonblock && !(kiocb->ki_flags & IOCB_DIRECT) && | |
3545 | (req->flags & REQ_F_ISREG)) | |
f67676d1 | 3546 | goto copy_iov; |
31b51510 | 3547 | |
0fef9483 | 3548 | ret = rw_verify_area(WRITE, req->file, io_kiocb_ppos(kiocb), iov_count); |
fa15bafb PB |
3549 | if (unlikely(ret)) |
3550 | goto out_free; | |
4ed734b0 | 3551 | |
fa15bafb PB |
3552 | /* |
3553 | * Open-code file_start_write here to grab freeze protection, | |
3554 | * which will be released by another thread in | |
3555 | * io_complete_rw(). Fool lockdep by telling it the lock got | |
3556 | * released so that it doesn't complain about the held lock when | |
3557 | * we return to userspace. | |
3558 | */ | |
3559 | if (req->flags & REQ_F_ISREG) { | |
8a3c84b6 | 3560 | sb_start_write(file_inode(req->file)->i_sb); |
fa15bafb PB |
3561 | __sb_writers_release(file_inode(req->file)->i_sb, |
3562 | SB_FREEZE_WRITE); | |
3563 | } | |
3564 | kiocb->ki_flags |= IOCB_WRITE; | |
4ed734b0 | 3565 | |
fa15bafb | 3566 | if (req->file->f_op->write_iter) |
ff6165b2 | 3567 | ret2 = call_write_iter(req->file, kiocb, iter); |
2dd2111d | 3568 | else if (req->file->f_op->write) |
4017eb91 | 3569 | ret2 = loop_rw_iter(WRITE, req, iter); |
2dd2111d GH |
3570 | else |
3571 | ret2 = -EINVAL; | |
4ed734b0 | 3572 | |
fa15bafb PB |
3573 | /* |
3574 | * Raw bdev writes will return -EOPNOTSUPP for IOCB_NOWAIT. Just | |
3575 | * retry them without IOCB_NOWAIT. | |
3576 | */ | |
3577 | if (ret2 == -EOPNOTSUPP && (kiocb->ki_flags & IOCB_NOWAIT)) | |
3578 | ret2 = -EAGAIN; | |
355afaeb JA |
3579 | /* no retry on NONBLOCK marked file */ |
3580 | if (ret2 == -EAGAIN && (req->file->f_flags & O_NONBLOCK)) | |
3581 | goto done; | |
fa15bafb | 3582 | if (!force_nonblock || ret2 != -EAGAIN) { |
eefdf30f JA |
3583 | /* IOPOLL retry should happen for io-wq threads */ |
3584 | if ((req->ctx->flags & IORING_SETUP_IOPOLL) && ret2 == -EAGAIN) | |
3585 | goto copy_iov; | |
355afaeb | 3586 | done: |
fa15bafb PB |
3587 | kiocb_done(kiocb, ret2, cs); |
3588 | } else { | |
f67676d1 | 3589 | copy_iov: |
84216315 JA |
3590 | /* some cases will consume bytes even on error returns */ |
3591 | iov_iter_revert(iter, iov_count - iov_iter_count(iter)); | |
227c0c96 | 3592 | ret = io_setup_async_rw(req, iovec, inline_vecs, iter, false); |
ff6165b2 JA |
3593 | if (!ret) |
3594 | return -EAGAIN; | |
2b188cc1 | 3595 | } |
31b51510 | 3596 | out_free: |
f261c168 | 3597 | /* it's reportedly faster than delegating the null check to kfree() */ |
252917c3 | 3598 | if (iovec) |
6f2cc166 | 3599 | kfree(iovec); |
2b188cc1 JA |
3600 | return ret; |
3601 | } | |
3602 | ||
36f4fa68 JA |
3603 | static int io_shutdown_prep(struct io_kiocb *req, |
3604 | const struct io_uring_sqe *sqe) | |
3605 | { | |
3606 | #if defined(CONFIG_NET) | |
3607 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) | |
3608 | return -EINVAL; | |
3609 | if (sqe->ioprio || sqe->off || sqe->addr || sqe->rw_flags || | |
3610 | sqe->buf_index) | |
3611 | return -EINVAL; | |
3612 | ||
3613 | req->shutdown.how = READ_ONCE(sqe->len); | |
3614 | return 0; | |
3615 | #else | |
3616 | return -EOPNOTSUPP; | |
3617 | #endif | |
3618 | } | |
3619 | ||
3620 | static int io_shutdown(struct io_kiocb *req, bool force_nonblock) | |
3621 | { | |
3622 | #if defined(CONFIG_NET) | |
3623 | struct socket *sock; | |
3624 | int ret; | |
3625 | ||
3626 | if (force_nonblock) | |
3627 | return -EAGAIN; | |
3628 | ||
3629 | sock = sock_from_file(req->file, &ret); | |
3630 | if (unlikely(!sock)) | |
3631 | return ret; | |
3632 | ||
3633 | ret = __sys_shutdown_sock(sock, req->shutdown.how); | |
3634 | io_req_complete(req, ret); | |
3635 | return 0; | |
3636 | #else | |
3637 | return -EOPNOTSUPP; | |
3638 | #endif | |
3639 | } | |
3640 | ||
f2a8d5c7 PB |
3641 | static int __io_splice_prep(struct io_kiocb *req, |
3642 | const struct io_uring_sqe *sqe) | |
7d67af2c PB |
3643 | { |
3644 | struct io_splice* sp = &req->splice; | |
3645 | unsigned int valid_flags = SPLICE_F_FD_IN_FIXED | SPLICE_F_ALL; | |
7d67af2c | 3646 | |
3232dd02 PB |
3647 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
3648 | return -EINVAL; | |
7d67af2c PB |
3649 | |
3650 | sp->file_in = NULL; | |
7d67af2c PB |
3651 | sp->len = READ_ONCE(sqe->len); |
3652 | sp->flags = READ_ONCE(sqe->splice_flags); | |
3653 | ||
3654 | if (unlikely(sp->flags & ~valid_flags)) | |
3655 | return -EINVAL; | |
3656 | ||
8371adf5 PB |
3657 | sp->file_in = io_file_get(NULL, req, READ_ONCE(sqe->splice_fd_in), |
3658 | (sp->flags & SPLICE_F_FD_IN_FIXED)); | |
3659 | if (!sp->file_in) | |
3660 | return -EBADF; | |
7d67af2c PB |
3661 | req->flags |= REQ_F_NEED_CLEANUP; |
3662 | ||
7cdaf587 XW |
3663 | if (!S_ISREG(file_inode(sp->file_in)->i_mode)) { |
3664 | /* | |
3665 | * Splice operation will be punted aync, and here need to | |
3666 | * modify io_wq_work.flags, so initialize io_wq_work firstly. | |
3667 | */ | |
3668 | io_req_init_async(req); | |
7d67af2c | 3669 | req->work.flags |= IO_WQ_WORK_UNBOUND; |
7cdaf587 | 3670 | } |
7d67af2c PB |
3671 | |
3672 | return 0; | |
3673 | } | |
3674 | ||
f2a8d5c7 PB |
3675 | static int io_tee_prep(struct io_kiocb *req, |
3676 | const struct io_uring_sqe *sqe) | |
3677 | { | |
3678 | if (READ_ONCE(sqe->splice_off_in) || READ_ONCE(sqe->off)) | |
3679 | return -EINVAL; | |
3680 | return __io_splice_prep(req, sqe); | |
3681 | } | |
3682 | ||
3683 | static int io_tee(struct io_kiocb *req, bool force_nonblock) | |
3684 | { | |
3685 | struct io_splice *sp = &req->splice; | |
3686 | struct file *in = sp->file_in; | |
3687 | struct file *out = sp->file_out; | |
3688 | unsigned int flags = sp->flags & ~SPLICE_F_FD_IN_FIXED; | |
3689 | long ret = 0; | |
3690 | ||
3691 | if (force_nonblock) | |
3692 | return -EAGAIN; | |
3693 | if (sp->len) | |
3694 | ret = do_tee(in, out, sp->len, flags); | |
3695 | ||
3696 | io_put_file(req, in, (sp->flags & SPLICE_F_FD_IN_FIXED)); | |
3697 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
3698 | ||
f2a8d5c7 PB |
3699 | if (ret != sp->len) |
3700 | req_set_fail_links(req); | |
e1e16097 | 3701 | io_req_complete(req, ret); |
f2a8d5c7 PB |
3702 | return 0; |
3703 | } | |
3704 | ||
3705 | static int io_splice_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) | |
3706 | { | |
3707 | struct io_splice* sp = &req->splice; | |
3708 | ||
3709 | sp->off_in = READ_ONCE(sqe->splice_off_in); | |
3710 | sp->off_out = READ_ONCE(sqe->off); | |
3711 | return __io_splice_prep(req, sqe); | |
3712 | } | |
3713 | ||
014db007 | 3714 | static int io_splice(struct io_kiocb *req, bool force_nonblock) |
7d67af2c PB |
3715 | { |
3716 | struct io_splice *sp = &req->splice; | |
3717 | struct file *in = sp->file_in; | |
3718 | struct file *out = sp->file_out; | |
3719 | unsigned int flags = sp->flags & ~SPLICE_F_FD_IN_FIXED; | |
3720 | loff_t *poff_in, *poff_out; | |
c9687426 | 3721 | long ret = 0; |
7d67af2c | 3722 | |
2fb3e822 PB |
3723 | if (force_nonblock) |
3724 | return -EAGAIN; | |
7d67af2c PB |
3725 | |
3726 | poff_in = (sp->off_in == -1) ? NULL : &sp->off_in; | |
3727 | poff_out = (sp->off_out == -1) ? NULL : &sp->off_out; | |
c9687426 | 3728 | |
948a7749 | 3729 | if (sp->len) |
c9687426 | 3730 | ret = do_splice(in, poff_in, out, poff_out, sp->len, flags); |
7d67af2c PB |
3731 | |
3732 | io_put_file(req, in, (sp->flags & SPLICE_F_FD_IN_FIXED)); | |
3733 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
3734 | ||
7d67af2c PB |
3735 | if (ret != sp->len) |
3736 | req_set_fail_links(req); | |
e1e16097 | 3737 | io_req_complete(req, ret); |
7d67af2c PB |
3738 | return 0; |
3739 | } | |
3740 | ||
2b188cc1 JA |
3741 | /* |
3742 | * IORING_OP_NOP just posts a completion event, nothing else. | |
3743 | */ | |
229a7b63 | 3744 | static int io_nop(struct io_kiocb *req, struct io_comp_state *cs) |
2b188cc1 JA |
3745 | { |
3746 | struct io_ring_ctx *ctx = req->ctx; | |
2b188cc1 | 3747 | |
def596e9 JA |
3748 | if (unlikely(ctx->flags & IORING_SETUP_IOPOLL)) |
3749 | return -EINVAL; | |
3750 | ||
229a7b63 | 3751 | __io_req_complete(req, 0, 0, cs); |
2b188cc1 JA |
3752 | return 0; |
3753 | } | |
3754 | ||
3529d8c2 | 3755 | static int io_prep_fsync(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
c992fe29 | 3756 | { |
6b06314c | 3757 | struct io_ring_ctx *ctx = req->ctx; |
c992fe29 | 3758 | |
09bb8394 JA |
3759 | if (!req->file) |
3760 | return -EBADF; | |
c992fe29 | 3761 | |
6b06314c | 3762 | if (unlikely(ctx->flags & IORING_SETUP_IOPOLL)) |
def596e9 | 3763 | return -EINVAL; |
edafccee | 3764 | if (unlikely(sqe->addr || sqe->ioprio || sqe->buf_index)) |
c992fe29 CH |
3765 | return -EINVAL; |
3766 | ||
8ed8d3c3 JA |
3767 | req->sync.flags = READ_ONCE(sqe->fsync_flags); |
3768 | if (unlikely(req->sync.flags & ~IORING_FSYNC_DATASYNC)) | |
3769 | return -EINVAL; | |
3770 | ||
3771 | req->sync.off = READ_ONCE(sqe->off); | |
3772 | req->sync.len = READ_ONCE(sqe->len); | |
c992fe29 CH |
3773 | return 0; |
3774 | } | |
3775 | ||
ac45abc0 | 3776 | static int io_fsync(struct io_kiocb *req, bool force_nonblock) |
8ed8d3c3 | 3777 | { |
8ed8d3c3 | 3778 | loff_t end = req->sync.off + req->sync.len; |
8ed8d3c3 JA |
3779 | int ret; |
3780 | ||
ac45abc0 PB |
3781 | /* fsync always requires a blocking context */ |
3782 | if (force_nonblock) | |
3783 | return -EAGAIN; | |
3784 | ||
9adbd45d | 3785 | ret = vfs_fsync_range(req->file, req->sync.off, |
8ed8d3c3 JA |
3786 | end > 0 ? end : LLONG_MAX, |
3787 | req->sync.flags & IORING_FSYNC_DATASYNC); | |
3788 | if (ret < 0) | |
3789 | req_set_fail_links(req); | |
e1e16097 | 3790 | io_req_complete(req, ret); |
c992fe29 CH |
3791 | return 0; |
3792 | } | |
3793 | ||
d63d1b5e JA |
3794 | static int io_fallocate_prep(struct io_kiocb *req, |
3795 | const struct io_uring_sqe *sqe) | |
3796 | { | |
3797 | if (sqe->ioprio || sqe->buf_index || sqe->rw_flags) | |
3798 | return -EINVAL; | |
3232dd02 PB |
3799 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
3800 | return -EINVAL; | |
d63d1b5e JA |
3801 | |
3802 | req->sync.off = READ_ONCE(sqe->off); | |
3803 | req->sync.len = READ_ONCE(sqe->addr); | |
3804 | req->sync.mode = READ_ONCE(sqe->len); | |
3805 | return 0; | |
3806 | } | |
3807 | ||
014db007 | 3808 | static int io_fallocate(struct io_kiocb *req, bool force_nonblock) |
5d17b4a4 | 3809 | { |
ac45abc0 PB |
3810 | int ret; |
3811 | ||
d63d1b5e | 3812 | /* fallocate always requiring blocking context */ |
ac45abc0 | 3813 | if (force_nonblock) |
5d17b4a4 | 3814 | return -EAGAIN; |
ac45abc0 PB |
3815 | ret = vfs_fallocate(req->file, req->sync.mode, req->sync.off, |
3816 | req->sync.len); | |
ac45abc0 PB |
3817 | if (ret < 0) |
3818 | req_set_fail_links(req); | |
e1e16097 | 3819 | io_req_complete(req, ret); |
5d17b4a4 JA |
3820 | return 0; |
3821 | } | |
3822 | ||
ec65fea5 | 3823 | static int __io_openat_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
b7bb4f7d | 3824 | { |
f8748881 | 3825 | const char __user *fname; |
15b71abe | 3826 | int ret; |
b7bb4f7d | 3827 | |
ec65fea5 | 3828 | if (unlikely(sqe->ioprio || sqe->buf_index)) |
15b71abe | 3829 | return -EINVAL; |
ec65fea5 | 3830 | if (unlikely(req->flags & REQ_F_FIXED_FILE)) |
cf3040ca | 3831 | return -EBADF; |
03b1230c | 3832 | |
ec65fea5 PB |
3833 | /* open.how should be already initialised */ |
3834 | if (!(req->open.how.flags & O_PATH) && force_o_largefile()) | |
08a1d26e | 3835 | req->open.how.flags |= O_LARGEFILE; |
3529d8c2 | 3836 | |
25e72d10 PB |
3837 | req->open.dfd = READ_ONCE(sqe->fd); |
3838 | fname = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
f8748881 | 3839 | req->open.filename = getname(fname); |
15b71abe JA |
3840 | if (IS_ERR(req->open.filename)) { |
3841 | ret = PTR_ERR(req->open.filename); | |
3842 | req->open.filename = NULL; | |
3843 | return ret; | |
3844 | } | |
4022e7af | 3845 | req->open.nofile = rlimit(RLIMIT_NOFILE); |
944d1444 | 3846 | req->open.ignore_nonblock = false; |
8fef80bf | 3847 | req->flags |= REQ_F_NEED_CLEANUP; |
15b71abe | 3848 | return 0; |
03b1230c JA |
3849 | } |
3850 | ||
ec65fea5 PB |
3851 | static int io_openat_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
3852 | { | |
3853 | u64 flags, mode; | |
3854 | ||
4eb8dded JA |
3855 | if (unlikely(req->ctx->flags & (IORING_SETUP_IOPOLL|IORING_SETUP_SQPOLL))) |
3856 | return -EINVAL; | |
ec65fea5 PB |
3857 | mode = READ_ONCE(sqe->len); |
3858 | flags = READ_ONCE(sqe->open_flags); | |
3859 | req->open.how = build_open_how(flags, mode); | |
3860 | return __io_openat_prep(req, sqe); | |
3861 | } | |
3862 | ||
cebdb986 | 3863 | static int io_openat2_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
aa1fa28f | 3864 | { |
cebdb986 | 3865 | struct open_how __user *how; |
cebdb986 | 3866 | size_t len; |
0fa03c62 JA |
3867 | int ret; |
3868 | ||
4eb8dded JA |
3869 | if (unlikely(req->ctx->flags & (IORING_SETUP_IOPOLL|IORING_SETUP_SQPOLL))) |
3870 | return -EINVAL; | |
cebdb986 JA |
3871 | how = u64_to_user_ptr(READ_ONCE(sqe->addr2)); |
3872 | len = READ_ONCE(sqe->len); | |
cebdb986 JA |
3873 | if (len < OPEN_HOW_SIZE_VER0) |
3874 | return -EINVAL; | |
3529d8c2 | 3875 | |
cebdb986 JA |
3876 | ret = copy_struct_from_user(&req->open.how, sizeof(req->open.how), how, |
3877 | len); | |
3878 | if (ret) | |
3879 | return ret; | |
3529d8c2 | 3880 | |
ec65fea5 | 3881 | return __io_openat_prep(req, sqe); |
cebdb986 JA |
3882 | } |
3883 | ||
014db007 | 3884 | static int io_openat2(struct io_kiocb *req, bool force_nonblock) |
15b71abe JA |
3885 | { |
3886 | struct open_flags op; | |
15b71abe JA |
3887 | struct file *file; |
3888 | int ret; | |
3889 | ||
944d1444 | 3890 | if (force_nonblock && !req->open.ignore_nonblock) |
15b71abe | 3891 | return -EAGAIN; |
15b71abe | 3892 | |
cebdb986 | 3893 | ret = build_open_flags(&req->open.how, &op); |
15b71abe JA |
3894 | if (ret) |
3895 | goto err; | |
3896 | ||
4022e7af | 3897 | ret = __get_unused_fd_flags(req->open.how.flags, req->open.nofile); |
15b71abe JA |
3898 | if (ret < 0) |
3899 | goto err; | |
3900 | ||
3901 | file = do_filp_open(req->open.dfd, req->open.filename, &op); | |
3902 | if (IS_ERR(file)) { | |
3903 | put_unused_fd(ret); | |
3904 | ret = PTR_ERR(file); | |
944d1444 JA |
3905 | /* |
3906 | * A work-around to ensure that /proc/self works that way | |
3907 | * that it should - if we get -EOPNOTSUPP back, then assume | |
3908 | * that proc_self_get_link() failed us because we're in async | |
3909 | * context. We should be safe to retry this from the task | |
3910 | * itself with force_nonblock == false set, as it should not | |
3911 | * block on lookup. Would be nice to know this upfront and | |
3912 | * avoid the async dance, but doesn't seem feasible. | |
3913 | */ | |
3914 | if (ret == -EOPNOTSUPP && io_wq_current_is_worker()) { | |
3915 | req->open.ignore_nonblock = true; | |
3916 | refcount_inc(&req->refs); | |
3917 | io_req_task_queue(req); | |
3918 | return 0; | |
3919 | } | |
15b71abe JA |
3920 | } else { |
3921 | fsnotify_open(file); | |
3922 | fd_install(ret, file); | |
3923 | } | |
3924 | err: | |
3925 | putname(req->open.filename); | |
8fef80bf | 3926 | req->flags &= ~REQ_F_NEED_CLEANUP; |
15b71abe JA |
3927 | if (ret < 0) |
3928 | req_set_fail_links(req); | |
e1e16097 | 3929 | io_req_complete(req, ret); |
15b71abe JA |
3930 | return 0; |
3931 | } | |
3932 | ||
014db007 | 3933 | static int io_openat(struct io_kiocb *req, bool force_nonblock) |
cebdb986 | 3934 | { |
014db007 | 3935 | return io_openat2(req, force_nonblock); |
cebdb986 JA |
3936 | } |
3937 | ||
067524e9 JA |
3938 | static int io_remove_buffers_prep(struct io_kiocb *req, |
3939 | const struct io_uring_sqe *sqe) | |
3940 | { | |
3941 | struct io_provide_buf *p = &req->pbuf; | |
3942 | u64 tmp; | |
3943 | ||
3944 | if (sqe->ioprio || sqe->rw_flags || sqe->addr || sqe->len || sqe->off) | |
3945 | return -EINVAL; | |
3946 | ||
3947 | tmp = READ_ONCE(sqe->fd); | |
3948 | if (!tmp || tmp > USHRT_MAX) | |
3949 | return -EINVAL; | |
3950 | ||
3951 | memset(p, 0, sizeof(*p)); | |
3952 | p->nbufs = tmp; | |
3953 | p->bgid = READ_ONCE(sqe->buf_group); | |
3954 | return 0; | |
3955 | } | |
3956 | ||
3957 | static int __io_remove_buffers(struct io_ring_ctx *ctx, struct io_buffer *buf, | |
3958 | int bgid, unsigned nbufs) | |
3959 | { | |
3960 | unsigned i = 0; | |
3961 | ||
3962 | /* shouldn't happen */ | |
3963 | if (!nbufs) | |
3964 | return 0; | |
3965 | ||
3966 | /* the head kbuf is the list itself */ | |
3967 | while (!list_empty(&buf->list)) { | |
3968 | struct io_buffer *nxt; | |
3969 | ||
3970 | nxt = list_first_entry(&buf->list, struct io_buffer, list); | |
3971 | list_del(&nxt->list); | |
3972 | kfree(nxt); | |
3973 | if (++i == nbufs) | |
3974 | return i; | |
3975 | } | |
3976 | i++; | |
3977 | kfree(buf); | |
3978 | idr_remove(&ctx->io_buffer_idr, bgid); | |
3979 | ||
3980 | return i; | |
3981 | } | |
3982 | ||
229a7b63 JA |
3983 | static int io_remove_buffers(struct io_kiocb *req, bool force_nonblock, |
3984 | struct io_comp_state *cs) | |
067524e9 JA |
3985 | { |
3986 | struct io_provide_buf *p = &req->pbuf; | |
3987 | struct io_ring_ctx *ctx = req->ctx; | |
3988 | struct io_buffer *head; | |
3989 | int ret = 0; | |
3990 | ||
3991 | io_ring_submit_lock(ctx, !force_nonblock); | |
3992 | ||
3993 | lockdep_assert_held(&ctx->uring_lock); | |
3994 | ||
3995 | ret = -ENOENT; | |
3996 | head = idr_find(&ctx->io_buffer_idr, p->bgid); | |
3997 | if (head) | |
3998 | ret = __io_remove_buffers(ctx, head, p->bgid, p->nbufs); | |
3999 | ||
4000 | io_ring_submit_lock(ctx, !force_nonblock); | |
4001 | if (ret < 0) | |
4002 | req_set_fail_links(req); | |
229a7b63 | 4003 | __io_req_complete(req, ret, 0, cs); |
067524e9 JA |
4004 | return 0; |
4005 | } | |
4006 | ||
ddf0322d JA |
4007 | static int io_provide_buffers_prep(struct io_kiocb *req, |
4008 | const struct io_uring_sqe *sqe) | |
4009 | { | |
4010 | struct io_provide_buf *p = &req->pbuf; | |
4011 | u64 tmp; | |
4012 | ||
4013 | if (sqe->ioprio || sqe->rw_flags) | |
4014 | return -EINVAL; | |
4015 | ||
4016 | tmp = READ_ONCE(sqe->fd); | |
4017 | if (!tmp || tmp > USHRT_MAX) | |
4018 | return -E2BIG; | |
4019 | p->nbufs = tmp; | |
4020 | p->addr = READ_ONCE(sqe->addr); | |
4021 | p->len = READ_ONCE(sqe->len); | |
4022 | ||
efe68c1c | 4023 | if (!access_ok(u64_to_user_ptr(p->addr), (p->len * p->nbufs))) |
ddf0322d JA |
4024 | return -EFAULT; |
4025 | ||
4026 | p->bgid = READ_ONCE(sqe->buf_group); | |
4027 | tmp = READ_ONCE(sqe->off); | |
4028 | if (tmp > USHRT_MAX) | |
4029 | return -E2BIG; | |
4030 | p->bid = tmp; | |
4031 | return 0; | |
4032 | } | |
4033 | ||
4034 | static int io_add_buffers(struct io_provide_buf *pbuf, struct io_buffer **head) | |
4035 | { | |
4036 | struct io_buffer *buf; | |
4037 | u64 addr = pbuf->addr; | |
4038 | int i, bid = pbuf->bid; | |
4039 | ||
4040 | for (i = 0; i < pbuf->nbufs; i++) { | |
4041 | buf = kmalloc(sizeof(*buf), GFP_KERNEL); | |
4042 | if (!buf) | |
4043 | break; | |
4044 | ||
4045 | buf->addr = addr; | |
4046 | buf->len = pbuf->len; | |
4047 | buf->bid = bid; | |
4048 | addr += pbuf->len; | |
4049 | bid++; | |
4050 | if (!*head) { | |
4051 | INIT_LIST_HEAD(&buf->list); | |
4052 | *head = buf; | |
4053 | } else { | |
4054 | list_add_tail(&buf->list, &(*head)->list); | |
4055 | } | |
4056 | } | |
4057 | ||
4058 | return i ? i : -ENOMEM; | |
4059 | } | |
4060 | ||
229a7b63 JA |
4061 | static int io_provide_buffers(struct io_kiocb *req, bool force_nonblock, |
4062 | struct io_comp_state *cs) | |
ddf0322d JA |
4063 | { |
4064 | struct io_provide_buf *p = &req->pbuf; | |
4065 | struct io_ring_ctx *ctx = req->ctx; | |
4066 | struct io_buffer *head, *list; | |
4067 | int ret = 0; | |
4068 | ||
4069 | io_ring_submit_lock(ctx, !force_nonblock); | |
4070 | ||
4071 | lockdep_assert_held(&ctx->uring_lock); | |
4072 | ||
4073 | list = head = idr_find(&ctx->io_buffer_idr, p->bgid); | |
4074 | ||
4075 | ret = io_add_buffers(p, &head); | |
4076 | if (ret < 0) | |
4077 | goto out; | |
4078 | ||
4079 | if (!list) { | |
4080 | ret = idr_alloc(&ctx->io_buffer_idr, head, p->bgid, p->bgid + 1, | |
4081 | GFP_KERNEL); | |
4082 | if (ret < 0) { | |
067524e9 | 4083 | __io_remove_buffers(ctx, head, p->bgid, -1U); |
ddf0322d JA |
4084 | goto out; |
4085 | } | |
4086 | } | |
4087 | out: | |
4088 | io_ring_submit_unlock(ctx, !force_nonblock); | |
4089 | if (ret < 0) | |
4090 | req_set_fail_links(req); | |
229a7b63 | 4091 | __io_req_complete(req, ret, 0, cs); |
ddf0322d | 4092 | return 0; |
cebdb986 JA |
4093 | } |
4094 | ||
3e4827b0 JA |
4095 | static int io_epoll_ctl_prep(struct io_kiocb *req, |
4096 | const struct io_uring_sqe *sqe) | |
4097 | { | |
4098 | #if defined(CONFIG_EPOLL) | |
4099 | if (sqe->ioprio || sqe->buf_index) | |
4100 | return -EINVAL; | |
6ca56f84 | 4101 | if (unlikely(req->ctx->flags & (IORING_SETUP_IOPOLL | IORING_SETUP_SQPOLL))) |
3232dd02 | 4102 | return -EINVAL; |
3e4827b0 JA |
4103 | |
4104 | req->epoll.epfd = READ_ONCE(sqe->fd); | |
4105 | req->epoll.op = READ_ONCE(sqe->len); | |
4106 | req->epoll.fd = READ_ONCE(sqe->off); | |
4107 | ||
4108 | if (ep_op_has_event(req->epoll.op)) { | |
4109 | struct epoll_event __user *ev; | |
4110 | ||
4111 | ev = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
4112 | if (copy_from_user(&req->epoll.event, ev, sizeof(*ev))) | |
4113 | return -EFAULT; | |
4114 | } | |
4115 | ||
4116 | return 0; | |
4117 | #else | |
4118 | return -EOPNOTSUPP; | |
4119 | #endif | |
4120 | } | |
4121 | ||
229a7b63 JA |
4122 | static int io_epoll_ctl(struct io_kiocb *req, bool force_nonblock, |
4123 | struct io_comp_state *cs) | |
3e4827b0 JA |
4124 | { |
4125 | #if defined(CONFIG_EPOLL) | |
4126 | struct io_epoll *ie = &req->epoll; | |
4127 | int ret; | |
4128 | ||
4129 | ret = do_epoll_ctl(ie->epfd, ie->op, ie->fd, &ie->event, force_nonblock); | |
4130 | if (force_nonblock && ret == -EAGAIN) | |
4131 | return -EAGAIN; | |
4132 | ||
4133 | if (ret < 0) | |
4134 | req_set_fail_links(req); | |
229a7b63 | 4135 | __io_req_complete(req, ret, 0, cs); |
3e4827b0 JA |
4136 | return 0; |
4137 | #else | |
4138 | return -EOPNOTSUPP; | |
4139 | #endif | |
4140 | } | |
4141 | ||
c1ca757b JA |
4142 | static int io_madvise_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
4143 | { | |
4144 | #if defined(CONFIG_ADVISE_SYSCALLS) && defined(CONFIG_MMU) | |
4145 | if (sqe->ioprio || sqe->buf_index || sqe->off) | |
4146 | return -EINVAL; | |
3232dd02 PB |
4147 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4148 | return -EINVAL; | |
c1ca757b JA |
4149 | |
4150 | req->madvise.addr = READ_ONCE(sqe->addr); | |
4151 | req->madvise.len = READ_ONCE(sqe->len); | |
4152 | req->madvise.advice = READ_ONCE(sqe->fadvise_advice); | |
4153 | return 0; | |
4154 | #else | |
4155 | return -EOPNOTSUPP; | |
4156 | #endif | |
4157 | } | |
4158 | ||
014db007 | 4159 | static int io_madvise(struct io_kiocb *req, bool force_nonblock) |
c1ca757b JA |
4160 | { |
4161 | #if defined(CONFIG_ADVISE_SYSCALLS) && defined(CONFIG_MMU) | |
4162 | struct io_madvise *ma = &req->madvise; | |
4163 | int ret; | |
4164 | ||
4165 | if (force_nonblock) | |
4166 | return -EAGAIN; | |
4167 | ||
0726b01e | 4168 | ret = do_madvise(current->mm, ma->addr, ma->len, ma->advice); |
c1ca757b JA |
4169 | if (ret < 0) |
4170 | req_set_fail_links(req); | |
e1e16097 | 4171 | io_req_complete(req, ret); |
c1ca757b JA |
4172 | return 0; |
4173 | #else | |
4174 | return -EOPNOTSUPP; | |
4175 | #endif | |
4176 | } | |
4177 | ||
4840e418 JA |
4178 | static int io_fadvise_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
4179 | { | |
4180 | if (sqe->ioprio || sqe->buf_index || sqe->addr) | |
4181 | return -EINVAL; | |
3232dd02 PB |
4182 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4183 | return -EINVAL; | |
4840e418 JA |
4184 | |
4185 | req->fadvise.offset = READ_ONCE(sqe->off); | |
4186 | req->fadvise.len = READ_ONCE(sqe->len); | |
4187 | req->fadvise.advice = READ_ONCE(sqe->fadvise_advice); | |
4188 | return 0; | |
4189 | } | |
4190 | ||
014db007 | 4191 | static int io_fadvise(struct io_kiocb *req, bool force_nonblock) |
4840e418 JA |
4192 | { |
4193 | struct io_fadvise *fa = &req->fadvise; | |
4194 | int ret; | |
4195 | ||
3e69426d JA |
4196 | if (force_nonblock) { |
4197 | switch (fa->advice) { | |
4198 | case POSIX_FADV_NORMAL: | |
4199 | case POSIX_FADV_RANDOM: | |
4200 | case POSIX_FADV_SEQUENTIAL: | |
4201 | break; | |
4202 | default: | |
4203 | return -EAGAIN; | |
4204 | } | |
4205 | } | |
4840e418 JA |
4206 | |
4207 | ret = vfs_fadvise(req->file, fa->offset, fa->len, fa->advice); | |
4208 | if (ret < 0) | |
4209 | req_set_fail_links(req); | |
e1e16097 | 4210 | io_req_complete(req, ret); |
4840e418 JA |
4211 | return 0; |
4212 | } | |
4213 | ||
eddc7ef5 JA |
4214 | static int io_statx_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
4215 | { | |
6ca56f84 | 4216 | if (unlikely(req->ctx->flags & (IORING_SETUP_IOPOLL | IORING_SETUP_SQPOLL))) |
3232dd02 | 4217 | return -EINVAL; |
eddc7ef5 JA |
4218 | if (sqe->ioprio || sqe->buf_index) |
4219 | return -EINVAL; | |
9c280f90 | 4220 | if (req->flags & REQ_F_FIXED_FILE) |
cf3040ca | 4221 | return -EBADF; |
eddc7ef5 | 4222 | |
1d9e1288 BM |
4223 | req->statx.dfd = READ_ONCE(sqe->fd); |
4224 | req->statx.mask = READ_ONCE(sqe->len); | |
e62753e4 | 4225 | req->statx.filename = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
1d9e1288 BM |
4226 | req->statx.buffer = u64_to_user_ptr(READ_ONCE(sqe->addr2)); |
4227 | req->statx.flags = READ_ONCE(sqe->statx_flags); | |
eddc7ef5 JA |
4228 | |
4229 | return 0; | |
4230 | } | |
4231 | ||
014db007 | 4232 | static int io_statx(struct io_kiocb *req, bool force_nonblock) |
eddc7ef5 | 4233 | { |
1d9e1288 | 4234 | struct io_statx *ctx = &req->statx; |
eddc7ef5 JA |
4235 | int ret; |
4236 | ||
5b0bbee4 JA |
4237 | if (force_nonblock) { |
4238 | /* only need file table for an actual valid fd */ | |
4239 | if (ctx->dfd == -1 || ctx->dfd == AT_FDCWD) | |
4240 | req->flags |= REQ_F_NO_FILE_TABLE; | |
eddc7ef5 | 4241 | return -EAGAIN; |
5b0bbee4 | 4242 | } |
eddc7ef5 | 4243 | |
e62753e4 BM |
4244 | ret = do_statx(ctx->dfd, ctx->filename, ctx->flags, ctx->mask, |
4245 | ctx->buffer); | |
eddc7ef5 | 4246 | |
eddc7ef5 JA |
4247 | if (ret < 0) |
4248 | req_set_fail_links(req); | |
e1e16097 | 4249 | io_req_complete(req, ret); |
eddc7ef5 JA |
4250 | return 0; |
4251 | } | |
4252 | ||
b5dba59e JA |
4253 | static int io_close_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
4254 | { | |
4255 | /* | |
4256 | * If we queue this for async, it must not be cancellable. That would | |
7cdaf587 XW |
4257 | * leave the 'file' in an undeterminate state, and here need to modify |
4258 | * io_wq_work.flags, so initialize io_wq_work firstly. | |
b5dba59e | 4259 | */ |
7cdaf587 | 4260 | io_req_init_async(req); |
b5dba59e JA |
4261 | req->work.flags |= IO_WQ_WORK_NO_CANCEL; |
4262 | ||
3232dd02 PB |
4263 | if (unlikely(req->ctx->flags & (IORING_SETUP_IOPOLL|IORING_SETUP_SQPOLL))) |
4264 | return -EINVAL; | |
b5dba59e JA |
4265 | if (sqe->ioprio || sqe->off || sqe->addr || sqe->len || |
4266 | sqe->rw_flags || sqe->buf_index) | |
4267 | return -EINVAL; | |
9c280f90 | 4268 | if (req->flags & REQ_F_FIXED_FILE) |
cf3040ca | 4269 | return -EBADF; |
b5dba59e JA |
4270 | |
4271 | req->close.fd = READ_ONCE(sqe->fd); | |
0f212204 | 4272 | if ((req->file && req->file->f_op == &io_uring_fops)) |
fd2206e4 | 4273 | return -EBADF; |
b5dba59e | 4274 | |
3af73b28 | 4275 | req->close.put_file = NULL; |
b5dba59e | 4276 | return 0; |
b5dba59e JA |
4277 | } |
4278 | ||
229a7b63 JA |
4279 | static int io_close(struct io_kiocb *req, bool force_nonblock, |
4280 | struct io_comp_state *cs) | |
b5dba59e | 4281 | { |
3af73b28 | 4282 | struct io_close *close = &req->close; |
b5dba59e JA |
4283 | int ret; |
4284 | ||
3af73b28 PB |
4285 | /* might be already done during nonblock submission */ |
4286 | if (!close->put_file) { | |
4287 | ret = __close_fd_get_file(close->fd, &close->put_file); | |
4288 | if (ret < 0) | |
4289 | return (ret == -ENOENT) ? -EBADF : ret; | |
4290 | } | |
b5dba59e JA |
4291 | |
4292 | /* if the file has a flush method, be safe and punt to async */ | |
3af73b28 | 4293 | if (close->put_file->f_op->flush && force_nonblock) { |
24c74678 PB |
4294 | /* was never set, but play safe */ |
4295 | req->flags &= ~REQ_F_NOWAIT; | |
0bf0eefd | 4296 | /* avoid grabbing files - we don't need the files */ |
24c74678 | 4297 | req->flags |= REQ_F_NO_FILE_TABLE; |
0bf0eefd | 4298 | return -EAGAIN; |
a2100672 | 4299 | } |
b5dba59e | 4300 | |
3af73b28 | 4301 | /* No ->flush() or already async, safely close from here */ |
98447d65 | 4302 | ret = filp_close(close->put_file, req->work.identity->files); |
3af73b28 PB |
4303 | if (ret < 0) |
4304 | req_set_fail_links(req); | |
3af73b28 PB |
4305 | fput(close->put_file); |
4306 | close->put_file = NULL; | |
229a7b63 | 4307 | __io_req_complete(req, ret, 0, cs); |
1a417f4e | 4308 | return 0; |
b5dba59e JA |
4309 | } |
4310 | ||
3529d8c2 | 4311 | static int io_prep_sfr(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
5d17b4a4 JA |
4312 | { |
4313 | struct io_ring_ctx *ctx = req->ctx; | |
5d17b4a4 JA |
4314 | |
4315 | if (!req->file) | |
4316 | return -EBADF; | |
5d17b4a4 JA |
4317 | |
4318 | if (unlikely(ctx->flags & IORING_SETUP_IOPOLL)) | |
4319 | return -EINVAL; | |
4320 | if (unlikely(sqe->addr || sqe->ioprio || sqe->buf_index)) | |
4321 | return -EINVAL; | |
4322 | ||
8ed8d3c3 JA |
4323 | req->sync.off = READ_ONCE(sqe->off); |
4324 | req->sync.len = READ_ONCE(sqe->len); | |
4325 | req->sync.flags = READ_ONCE(sqe->sync_range_flags); | |
8ed8d3c3 JA |
4326 | return 0; |
4327 | } | |
4328 | ||
ac45abc0 | 4329 | static int io_sync_file_range(struct io_kiocb *req, bool force_nonblock) |
8ed8d3c3 | 4330 | { |
8ed8d3c3 JA |
4331 | int ret; |
4332 | ||
ac45abc0 PB |
4333 | /* sync_file_range always requires a blocking context */ |
4334 | if (force_nonblock) | |
4335 | return -EAGAIN; | |
4336 | ||
9adbd45d | 4337 | ret = sync_file_range(req->file, req->sync.off, req->sync.len, |
8ed8d3c3 JA |
4338 | req->sync.flags); |
4339 | if (ret < 0) | |
4340 | req_set_fail_links(req); | |
e1e16097 | 4341 | io_req_complete(req, ret); |
5d17b4a4 JA |
4342 | return 0; |
4343 | } | |
4344 | ||
469956e8 | 4345 | #if defined(CONFIG_NET) |
02d27d89 PB |
4346 | static int io_setup_async_msg(struct io_kiocb *req, |
4347 | struct io_async_msghdr *kmsg) | |
4348 | { | |
e8c2bc1f JA |
4349 | struct io_async_msghdr *async_msg = req->async_data; |
4350 | ||
4351 | if (async_msg) | |
02d27d89 | 4352 | return -EAGAIN; |
e8c2bc1f | 4353 | if (io_alloc_async_data(req)) { |
02d27d89 PB |
4354 | if (kmsg->iov != kmsg->fast_iov) |
4355 | kfree(kmsg->iov); | |
4356 | return -ENOMEM; | |
4357 | } | |
e8c2bc1f | 4358 | async_msg = req->async_data; |
02d27d89 | 4359 | req->flags |= REQ_F_NEED_CLEANUP; |
e8c2bc1f | 4360 | memcpy(async_msg, kmsg, sizeof(*kmsg)); |
02d27d89 PB |
4361 | return -EAGAIN; |
4362 | } | |
4363 | ||
2ae523ed PB |
4364 | static int io_sendmsg_copy_hdr(struct io_kiocb *req, |
4365 | struct io_async_msghdr *iomsg) | |
4366 | { | |
4367 | iomsg->iov = iomsg->fast_iov; | |
4368 | iomsg->msg.msg_name = &iomsg->addr; | |
4369 | return sendmsg_copy_msghdr(&iomsg->msg, req->sr_msg.umsg, | |
4370 | req->sr_msg.msg_flags, &iomsg->iov); | |
4371 | } | |
4372 | ||
3529d8c2 | 4373 | static int io_sendmsg_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
03b1230c | 4374 | { |
e8c2bc1f | 4375 | struct io_async_msghdr *async_msg = req->async_data; |
e47293fd | 4376 | struct io_sr_msg *sr = &req->sr_msg; |
99bc4c38 | 4377 | int ret; |
03b1230c | 4378 | |
d2b6f48b PB |
4379 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4380 | return -EINVAL; | |
4381 | ||
e47293fd | 4382 | sr->msg_flags = READ_ONCE(sqe->msg_flags); |
270a5940 | 4383 | sr->umsg = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
fddaface | 4384 | sr->len = READ_ONCE(sqe->len); |
3529d8c2 | 4385 | |
d8768362 JA |
4386 | #ifdef CONFIG_COMPAT |
4387 | if (req->ctx->compat) | |
4388 | sr->msg_flags |= MSG_CMSG_COMPAT; | |
4389 | #endif | |
4390 | ||
e8c2bc1f | 4391 | if (!async_msg || !io_op_defs[req->opcode].needs_async_data) |
3529d8c2 | 4392 | return 0; |
e8c2bc1f | 4393 | ret = io_sendmsg_copy_hdr(req, async_msg); |
99bc4c38 PB |
4394 | if (!ret) |
4395 | req->flags |= REQ_F_NEED_CLEANUP; | |
4396 | return ret; | |
03b1230c JA |
4397 | } |
4398 | ||
229a7b63 JA |
4399 | static int io_sendmsg(struct io_kiocb *req, bool force_nonblock, |
4400 | struct io_comp_state *cs) | |
aa1fa28f | 4401 | { |
6b754c8b | 4402 | struct io_async_msghdr iomsg, *kmsg; |
0fa03c62 | 4403 | struct socket *sock; |
7a7cacba | 4404 | unsigned flags; |
0fa03c62 JA |
4405 | int ret; |
4406 | ||
0fa03c62 | 4407 | sock = sock_from_file(req->file, &ret); |
7a7cacba PB |
4408 | if (unlikely(!sock)) |
4409 | return ret; | |
3529d8c2 | 4410 | |
e8c2bc1f JA |
4411 | if (req->async_data) { |
4412 | kmsg = req->async_data; | |
4413 | kmsg->msg.msg_name = &kmsg->addr; | |
7a7cacba PB |
4414 | /* if iov is set, it's allocated already */ |
4415 | if (!kmsg->iov) | |
4416 | kmsg->iov = kmsg->fast_iov; | |
4417 | kmsg->msg.msg_iter.iov = kmsg->iov; | |
4418 | } else { | |
4419 | ret = io_sendmsg_copy_hdr(req, &iomsg); | |
4420 | if (ret) | |
4421 | return ret; | |
4422 | kmsg = &iomsg; | |
0fa03c62 | 4423 | } |
0fa03c62 | 4424 | |
7a7cacba PB |
4425 | flags = req->sr_msg.msg_flags; |
4426 | if (flags & MSG_DONTWAIT) | |
4427 | req->flags |= REQ_F_NOWAIT; | |
4428 | else if (force_nonblock) | |
4429 | flags |= MSG_DONTWAIT; | |
e47293fd | 4430 | |
7a7cacba PB |
4431 | ret = __sys_sendmsg_sock(sock, &kmsg->msg, flags); |
4432 | if (force_nonblock && ret == -EAGAIN) | |
4433 | return io_setup_async_msg(req, kmsg); | |
4434 | if (ret == -ERESTARTSYS) | |
4435 | ret = -EINTR; | |
0fa03c62 | 4436 | |
6b754c8b | 4437 | if (kmsg->iov != kmsg->fast_iov) |
0b416c3e | 4438 | kfree(kmsg->iov); |
99bc4c38 | 4439 | req->flags &= ~REQ_F_NEED_CLEANUP; |
4e88d6e7 JA |
4440 | if (ret < 0) |
4441 | req_set_fail_links(req); | |
229a7b63 | 4442 | __io_req_complete(req, ret, 0, cs); |
5d17b4a4 | 4443 | return 0; |
03b1230c | 4444 | } |
aa1fa28f | 4445 | |
229a7b63 JA |
4446 | static int io_send(struct io_kiocb *req, bool force_nonblock, |
4447 | struct io_comp_state *cs) | |
fddaface | 4448 | { |
7a7cacba PB |
4449 | struct io_sr_msg *sr = &req->sr_msg; |
4450 | struct msghdr msg; | |
4451 | struct iovec iov; | |
fddaface | 4452 | struct socket *sock; |
7a7cacba | 4453 | unsigned flags; |
fddaface JA |
4454 | int ret; |
4455 | ||
fddaface | 4456 | sock = sock_from_file(req->file, &ret); |
7a7cacba PB |
4457 | if (unlikely(!sock)) |
4458 | return ret; | |
fddaface | 4459 | |
7a7cacba PB |
4460 | ret = import_single_range(WRITE, sr->buf, sr->len, &iov, &msg.msg_iter); |
4461 | if (unlikely(ret)) | |
14db8411 | 4462 | return ret; |
fddaface | 4463 | |
7a7cacba PB |
4464 | msg.msg_name = NULL; |
4465 | msg.msg_control = NULL; | |
4466 | msg.msg_controllen = 0; | |
4467 | msg.msg_namelen = 0; | |
fddaface | 4468 | |
7a7cacba PB |
4469 | flags = req->sr_msg.msg_flags; |
4470 | if (flags & MSG_DONTWAIT) | |
4471 | req->flags |= REQ_F_NOWAIT; | |
4472 | else if (force_nonblock) | |
4473 | flags |= MSG_DONTWAIT; | |
fddaface | 4474 | |
7a7cacba PB |
4475 | msg.msg_flags = flags; |
4476 | ret = sock_sendmsg(sock, &msg); | |
4477 | if (force_nonblock && ret == -EAGAIN) | |
4478 | return -EAGAIN; | |
4479 | if (ret == -ERESTARTSYS) | |
4480 | ret = -EINTR; | |
fddaface | 4481 | |
fddaface JA |
4482 | if (ret < 0) |
4483 | req_set_fail_links(req); | |
229a7b63 | 4484 | __io_req_complete(req, ret, 0, cs); |
fddaface | 4485 | return 0; |
fddaface JA |
4486 | } |
4487 | ||
1400e697 PB |
4488 | static int __io_recvmsg_copy_hdr(struct io_kiocb *req, |
4489 | struct io_async_msghdr *iomsg) | |
52de1fe1 JA |
4490 | { |
4491 | struct io_sr_msg *sr = &req->sr_msg; | |
4492 | struct iovec __user *uiov; | |
4493 | size_t iov_len; | |
4494 | int ret; | |
4495 | ||
1400e697 PB |
4496 | ret = __copy_msghdr_from_user(&iomsg->msg, sr->umsg, |
4497 | &iomsg->uaddr, &uiov, &iov_len); | |
52de1fe1 JA |
4498 | if (ret) |
4499 | return ret; | |
4500 | ||
4501 | if (req->flags & REQ_F_BUFFER_SELECT) { | |
4502 | if (iov_len > 1) | |
4503 | return -EINVAL; | |
1400e697 | 4504 | if (copy_from_user(iomsg->iov, uiov, sizeof(*uiov))) |
52de1fe1 | 4505 | return -EFAULT; |
1400e697 PB |
4506 | sr->len = iomsg->iov[0].iov_len; |
4507 | iov_iter_init(&iomsg->msg.msg_iter, READ, iomsg->iov, 1, | |
52de1fe1 | 4508 | sr->len); |
1400e697 | 4509 | iomsg->iov = NULL; |
52de1fe1 | 4510 | } else { |
89cd35c5 CH |
4511 | ret = __import_iovec(READ, uiov, iov_len, UIO_FASTIOV, |
4512 | &iomsg->iov, &iomsg->msg.msg_iter, | |
4513 | false); | |
52de1fe1 JA |
4514 | if (ret > 0) |
4515 | ret = 0; | |
4516 | } | |
4517 | ||
4518 | return ret; | |
4519 | } | |
4520 | ||
4521 | #ifdef CONFIG_COMPAT | |
4522 | static int __io_compat_recvmsg_copy_hdr(struct io_kiocb *req, | |
1400e697 | 4523 | struct io_async_msghdr *iomsg) |
52de1fe1 JA |
4524 | { |
4525 | struct compat_msghdr __user *msg_compat; | |
4526 | struct io_sr_msg *sr = &req->sr_msg; | |
4527 | struct compat_iovec __user *uiov; | |
4528 | compat_uptr_t ptr; | |
4529 | compat_size_t len; | |
4530 | int ret; | |
4531 | ||
270a5940 | 4532 | msg_compat = (struct compat_msghdr __user *) sr->umsg; |
1400e697 | 4533 | ret = __get_compat_msghdr(&iomsg->msg, msg_compat, &iomsg->uaddr, |
52de1fe1 JA |
4534 | &ptr, &len); |
4535 | if (ret) | |
4536 | return ret; | |
4537 | ||
4538 | uiov = compat_ptr(ptr); | |
4539 | if (req->flags & REQ_F_BUFFER_SELECT) { | |
4540 | compat_ssize_t clen; | |
4541 | ||
4542 | if (len > 1) | |
4543 | return -EINVAL; | |
4544 | if (!access_ok(uiov, sizeof(*uiov))) | |
4545 | return -EFAULT; | |
4546 | if (__get_user(clen, &uiov->iov_len)) | |
4547 | return -EFAULT; | |
4548 | if (clen < 0) | |
4549 | return -EINVAL; | |
1400e697 PB |
4550 | sr->len = iomsg->iov[0].iov_len; |
4551 | iomsg->iov = NULL; | |
52de1fe1 | 4552 | } else { |
89cd35c5 CH |
4553 | ret = __import_iovec(READ, (struct iovec __user *)uiov, len, |
4554 | UIO_FASTIOV, &iomsg->iov, | |
4555 | &iomsg->msg.msg_iter, true); | |
52de1fe1 JA |
4556 | if (ret < 0) |
4557 | return ret; | |
4558 | } | |
4559 | ||
4560 | return 0; | |
4561 | } | |
4562 | #endif | |
4563 | ||
1400e697 PB |
4564 | static int io_recvmsg_copy_hdr(struct io_kiocb *req, |
4565 | struct io_async_msghdr *iomsg) | |
52de1fe1 | 4566 | { |
1400e697 PB |
4567 | iomsg->msg.msg_name = &iomsg->addr; |
4568 | iomsg->iov = iomsg->fast_iov; | |
52de1fe1 JA |
4569 | |
4570 | #ifdef CONFIG_COMPAT | |
4571 | if (req->ctx->compat) | |
1400e697 | 4572 | return __io_compat_recvmsg_copy_hdr(req, iomsg); |
fddaface | 4573 | #endif |
52de1fe1 | 4574 | |
1400e697 | 4575 | return __io_recvmsg_copy_hdr(req, iomsg); |
52de1fe1 JA |
4576 | } |
4577 | ||
bcda7baa | 4578 | static struct io_buffer *io_recv_buffer_select(struct io_kiocb *req, |
7fbb1b54 | 4579 | bool needs_lock) |
bcda7baa JA |
4580 | { |
4581 | struct io_sr_msg *sr = &req->sr_msg; | |
4582 | struct io_buffer *kbuf; | |
4583 | ||
bcda7baa JA |
4584 | kbuf = io_buffer_select(req, &sr->len, sr->bgid, sr->kbuf, needs_lock); |
4585 | if (IS_ERR(kbuf)) | |
4586 | return kbuf; | |
4587 | ||
4588 | sr->kbuf = kbuf; | |
4589 | req->flags |= REQ_F_BUFFER_SELECTED; | |
bcda7baa | 4590 | return kbuf; |
fddaface JA |
4591 | } |
4592 | ||
7fbb1b54 PB |
4593 | static inline unsigned int io_put_recv_kbuf(struct io_kiocb *req) |
4594 | { | |
4595 | return io_put_kbuf(req, req->sr_msg.kbuf); | |
4596 | } | |
4597 | ||
3529d8c2 JA |
4598 | static int io_recvmsg_prep(struct io_kiocb *req, |
4599 | const struct io_uring_sqe *sqe) | |
aa1fa28f | 4600 | { |
e8c2bc1f | 4601 | struct io_async_msghdr *async_msg = req->async_data; |
e47293fd | 4602 | struct io_sr_msg *sr = &req->sr_msg; |
99bc4c38 | 4603 | int ret; |
3529d8c2 | 4604 | |
d2b6f48b PB |
4605 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4606 | return -EINVAL; | |
4607 | ||
3529d8c2 | 4608 | sr->msg_flags = READ_ONCE(sqe->msg_flags); |
270a5940 | 4609 | sr->umsg = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
0b7b21e4 | 4610 | sr->len = READ_ONCE(sqe->len); |
bcda7baa | 4611 | sr->bgid = READ_ONCE(sqe->buf_group); |
06b76d44 | 4612 | |
d8768362 JA |
4613 | #ifdef CONFIG_COMPAT |
4614 | if (req->ctx->compat) | |
4615 | sr->msg_flags |= MSG_CMSG_COMPAT; | |
4616 | #endif | |
4617 | ||
e8c2bc1f | 4618 | if (!async_msg || !io_op_defs[req->opcode].needs_async_data) |
5f798bea | 4619 | return 0; |
e8c2bc1f | 4620 | ret = io_recvmsg_copy_hdr(req, async_msg); |
99bc4c38 PB |
4621 | if (!ret) |
4622 | req->flags |= REQ_F_NEED_CLEANUP; | |
4623 | return ret; | |
aa1fa28f JA |
4624 | } |
4625 | ||
229a7b63 JA |
4626 | static int io_recvmsg(struct io_kiocb *req, bool force_nonblock, |
4627 | struct io_comp_state *cs) | |
aa1fa28f | 4628 | { |
6b754c8b | 4629 | struct io_async_msghdr iomsg, *kmsg; |
03b1230c | 4630 | struct socket *sock; |
7fbb1b54 | 4631 | struct io_buffer *kbuf; |
7a7cacba | 4632 | unsigned flags; |
52de1fe1 | 4633 | int ret, cflags = 0; |
03b1230c | 4634 | |
03b1230c | 4635 | sock = sock_from_file(req->file, &ret); |
7a7cacba PB |
4636 | if (unlikely(!sock)) |
4637 | return ret; | |
3529d8c2 | 4638 | |
e8c2bc1f JA |
4639 | if (req->async_data) { |
4640 | kmsg = req->async_data; | |
4641 | kmsg->msg.msg_name = &kmsg->addr; | |
7a7cacba PB |
4642 | /* if iov is set, it's allocated already */ |
4643 | if (!kmsg->iov) | |
4644 | kmsg->iov = kmsg->fast_iov; | |
4645 | kmsg->msg.msg_iter.iov = kmsg->iov; | |
4646 | } else { | |
4647 | ret = io_recvmsg_copy_hdr(req, &iomsg); | |
4648 | if (ret) | |
681fda8d | 4649 | return ret; |
7a7cacba PB |
4650 | kmsg = &iomsg; |
4651 | } | |
03b1230c | 4652 | |
bc02ef33 | 4653 | if (req->flags & REQ_F_BUFFER_SELECT) { |
7fbb1b54 | 4654 | kbuf = io_recv_buffer_select(req, !force_nonblock); |
bc02ef33 | 4655 | if (IS_ERR(kbuf)) |
52de1fe1 | 4656 | return PTR_ERR(kbuf); |
7a7cacba PB |
4657 | kmsg->fast_iov[0].iov_base = u64_to_user_ptr(kbuf->addr); |
4658 | iov_iter_init(&kmsg->msg.msg_iter, READ, kmsg->iov, | |
4659 | 1, req->sr_msg.len); | |
4660 | } | |
52de1fe1 | 4661 | |
7a7cacba PB |
4662 | flags = req->sr_msg.msg_flags; |
4663 | if (flags & MSG_DONTWAIT) | |
4664 | req->flags |= REQ_F_NOWAIT; | |
4665 | else if (force_nonblock) | |
4666 | flags |= MSG_DONTWAIT; | |
e47293fd | 4667 | |
7a7cacba PB |
4668 | ret = __sys_recvmsg_sock(sock, &kmsg->msg, req->sr_msg.umsg, |
4669 | kmsg->uaddr, flags); | |
0e1b6fe3 PB |
4670 | if (force_nonblock && ret == -EAGAIN) |
4671 | return io_setup_async_msg(req, kmsg); | |
7a7cacba PB |
4672 | if (ret == -ERESTARTSYS) |
4673 | ret = -EINTR; | |
03b1230c | 4674 | |
7fbb1b54 PB |
4675 | if (req->flags & REQ_F_BUFFER_SELECTED) |
4676 | cflags = io_put_recv_kbuf(req); | |
6b754c8b | 4677 | if (kmsg->iov != kmsg->fast_iov) |
0b416c3e | 4678 | kfree(kmsg->iov); |
99bc4c38 | 4679 | req->flags &= ~REQ_F_NEED_CLEANUP; |
4e88d6e7 JA |
4680 | if (ret < 0) |
4681 | req_set_fail_links(req); | |
229a7b63 | 4682 | __io_req_complete(req, ret, cflags, cs); |
03b1230c | 4683 | return 0; |
0fa03c62 | 4684 | } |
5d17b4a4 | 4685 | |
229a7b63 JA |
4686 | static int io_recv(struct io_kiocb *req, bool force_nonblock, |
4687 | struct io_comp_state *cs) | |
fddaface | 4688 | { |
6b754c8b | 4689 | struct io_buffer *kbuf; |
7a7cacba PB |
4690 | struct io_sr_msg *sr = &req->sr_msg; |
4691 | struct msghdr msg; | |
4692 | void __user *buf = sr->buf; | |
fddaface | 4693 | struct socket *sock; |
7a7cacba PB |
4694 | struct iovec iov; |
4695 | unsigned flags; | |
bcda7baa | 4696 | int ret, cflags = 0; |
fddaface | 4697 | |
fddaface | 4698 | sock = sock_from_file(req->file, &ret); |
7a7cacba PB |
4699 | if (unlikely(!sock)) |
4700 | return ret; | |
fddaface | 4701 | |
bc02ef33 | 4702 | if (req->flags & REQ_F_BUFFER_SELECT) { |
7fbb1b54 | 4703 | kbuf = io_recv_buffer_select(req, !force_nonblock); |
bcda7baa JA |
4704 | if (IS_ERR(kbuf)) |
4705 | return PTR_ERR(kbuf); | |
7a7cacba | 4706 | buf = u64_to_user_ptr(kbuf->addr); |
bc02ef33 | 4707 | } |
bcda7baa | 4708 | |
7a7cacba | 4709 | ret = import_single_range(READ, buf, sr->len, &iov, &msg.msg_iter); |
14c32eee PB |
4710 | if (unlikely(ret)) |
4711 | goto out_free; | |
fddaface | 4712 | |
7a7cacba PB |
4713 | msg.msg_name = NULL; |
4714 | msg.msg_control = NULL; | |
4715 | msg.msg_controllen = 0; | |
4716 | msg.msg_namelen = 0; | |
4717 | msg.msg_iocb = NULL; | |
4718 | msg.msg_flags = 0; | |
fddaface | 4719 | |
7a7cacba PB |
4720 | flags = req->sr_msg.msg_flags; |
4721 | if (flags & MSG_DONTWAIT) | |
4722 | req->flags |= REQ_F_NOWAIT; | |
4723 | else if (force_nonblock) | |
4724 | flags |= MSG_DONTWAIT; | |
4725 | ||
4726 | ret = sock_recvmsg(sock, &msg, flags); | |
4727 | if (force_nonblock && ret == -EAGAIN) | |
4728 | return -EAGAIN; | |
4729 | if (ret == -ERESTARTSYS) | |
4730 | ret = -EINTR; | |
14c32eee | 4731 | out_free: |
7fbb1b54 PB |
4732 | if (req->flags & REQ_F_BUFFER_SELECTED) |
4733 | cflags = io_put_recv_kbuf(req); | |
fddaface JA |
4734 | if (ret < 0) |
4735 | req_set_fail_links(req); | |
229a7b63 | 4736 | __io_req_complete(req, ret, cflags, cs); |
fddaface | 4737 | return 0; |
fddaface JA |
4738 | } |
4739 | ||
3529d8c2 | 4740 | static int io_accept_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
17f2fe35 | 4741 | { |
8ed8d3c3 JA |
4742 | struct io_accept *accept = &req->accept; |
4743 | ||
17f2fe35 JA |
4744 | if (unlikely(req->ctx->flags & (IORING_SETUP_IOPOLL|IORING_SETUP_SQPOLL))) |
4745 | return -EINVAL; | |
8042d6ce | 4746 | if (sqe->ioprio || sqe->len || sqe->buf_index) |
17f2fe35 JA |
4747 | return -EINVAL; |
4748 | ||
d55e5f5b JA |
4749 | accept->addr = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
4750 | accept->addr_len = u64_to_user_ptr(READ_ONCE(sqe->addr2)); | |
8ed8d3c3 | 4751 | accept->flags = READ_ONCE(sqe->accept_flags); |
09952e3e | 4752 | accept->nofile = rlimit(RLIMIT_NOFILE); |
8ed8d3c3 | 4753 | return 0; |
8ed8d3c3 | 4754 | } |
17f2fe35 | 4755 | |
229a7b63 JA |
4756 | static int io_accept(struct io_kiocb *req, bool force_nonblock, |
4757 | struct io_comp_state *cs) | |
8ed8d3c3 JA |
4758 | { |
4759 | struct io_accept *accept = &req->accept; | |
ac45abc0 | 4760 | unsigned int file_flags = force_nonblock ? O_NONBLOCK : 0; |
8ed8d3c3 JA |
4761 | int ret; |
4762 | ||
e697deed JX |
4763 | if (req->file->f_flags & O_NONBLOCK) |
4764 | req->flags |= REQ_F_NOWAIT; | |
4765 | ||
8ed8d3c3 | 4766 | ret = __sys_accept4_file(req->file, file_flags, accept->addr, |
09952e3e JA |
4767 | accept->addr_len, accept->flags, |
4768 | accept->nofile); | |
8ed8d3c3 | 4769 | if (ret == -EAGAIN && force_nonblock) |
17f2fe35 | 4770 | return -EAGAIN; |
ac45abc0 PB |
4771 | if (ret < 0) { |
4772 | if (ret == -ERESTARTSYS) | |
4773 | ret = -EINTR; | |
4e88d6e7 | 4774 | req_set_fail_links(req); |
ac45abc0 | 4775 | } |
229a7b63 | 4776 | __io_req_complete(req, ret, 0, cs); |
17f2fe35 | 4777 | return 0; |
8ed8d3c3 JA |
4778 | } |
4779 | ||
3529d8c2 | 4780 | static int io_connect_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
f499a021 | 4781 | { |
3529d8c2 | 4782 | struct io_connect *conn = &req->connect; |
e8c2bc1f | 4783 | struct io_async_connect *io = req->async_data; |
f499a021 | 4784 | |
3fbb51c1 JA |
4785 | if (unlikely(req->ctx->flags & (IORING_SETUP_IOPOLL|IORING_SETUP_SQPOLL))) |
4786 | return -EINVAL; | |
4787 | if (sqe->ioprio || sqe->len || sqe->buf_index || sqe->rw_flags) | |
4788 | return -EINVAL; | |
4789 | ||
3529d8c2 JA |
4790 | conn->addr = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
4791 | conn->addr_len = READ_ONCE(sqe->addr2); | |
4792 | ||
4793 | if (!io) | |
4794 | return 0; | |
4795 | ||
4796 | return move_addr_to_kernel(conn->addr, conn->addr_len, | |
e8c2bc1f | 4797 | &io->address); |
f499a021 JA |
4798 | } |
4799 | ||
229a7b63 JA |
4800 | static int io_connect(struct io_kiocb *req, bool force_nonblock, |
4801 | struct io_comp_state *cs) | |
f8e85cf2 | 4802 | { |
e8c2bc1f | 4803 | struct io_async_connect __io, *io; |
f8e85cf2 | 4804 | unsigned file_flags; |
3fbb51c1 | 4805 | int ret; |
f8e85cf2 | 4806 | |
e8c2bc1f JA |
4807 | if (req->async_data) { |
4808 | io = req->async_data; | |
f499a021 | 4809 | } else { |
3529d8c2 JA |
4810 | ret = move_addr_to_kernel(req->connect.addr, |
4811 | req->connect.addr_len, | |
e8c2bc1f | 4812 | &__io.address); |
f499a021 JA |
4813 | if (ret) |
4814 | goto out; | |
4815 | io = &__io; | |
4816 | } | |
4817 | ||
3fbb51c1 JA |
4818 | file_flags = force_nonblock ? O_NONBLOCK : 0; |
4819 | ||
e8c2bc1f | 4820 | ret = __sys_connect_file(req->file, &io->address, |
3fbb51c1 | 4821 | req->connect.addr_len, file_flags); |
87f80d62 | 4822 | if ((ret == -EAGAIN || ret == -EINPROGRESS) && force_nonblock) { |
e8c2bc1f | 4823 | if (req->async_data) |
b7bb4f7d | 4824 | return -EAGAIN; |
e8c2bc1f | 4825 | if (io_alloc_async_data(req)) { |
f499a021 JA |
4826 | ret = -ENOMEM; |
4827 | goto out; | |
4828 | } | |
e8c2bc1f JA |
4829 | io = req->async_data; |
4830 | memcpy(req->async_data, &__io, sizeof(__io)); | |
f8e85cf2 | 4831 | return -EAGAIN; |
f499a021 | 4832 | } |
f8e85cf2 JA |
4833 | if (ret == -ERESTARTSYS) |
4834 | ret = -EINTR; | |
f499a021 | 4835 | out: |
4e88d6e7 JA |
4836 | if (ret < 0) |
4837 | req_set_fail_links(req); | |
229a7b63 | 4838 | __io_req_complete(req, ret, 0, cs); |
f8e85cf2 | 4839 | return 0; |
469956e8 Y |
4840 | } |
4841 | #else /* !CONFIG_NET */ | |
4842 | static int io_sendmsg_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) | |
4843 | { | |
f8e85cf2 | 4844 | return -EOPNOTSUPP; |
f8e85cf2 JA |
4845 | } |
4846 | ||
1e16c2f9 RD |
4847 | static int io_sendmsg(struct io_kiocb *req, bool force_nonblock, |
4848 | struct io_comp_state *cs) | |
469956e8 Y |
4849 | { |
4850 | return -EOPNOTSUPP; | |
4851 | } | |
4852 | ||
1e16c2f9 RD |
4853 | static int io_send(struct io_kiocb *req, bool force_nonblock, |
4854 | struct io_comp_state *cs) | |
469956e8 Y |
4855 | { |
4856 | return -EOPNOTSUPP; | |
4857 | } | |
4858 | ||
4859 | static int io_recvmsg_prep(struct io_kiocb *req, | |
4860 | const struct io_uring_sqe *sqe) | |
4861 | { | |
4862 | return -EOPNOTSUPP; | |
4863 | } | |
4864 | ||
1e16c2f9 RD |
4865 | static int io_recvmsg(struct io_kiocb *req, bool force_nonblock, |
4866 | struct io_comp_state *cs) | |
469956e8 Y |
4867 | { |
4868 | return -EOPNOTSUPP; | |
4869 | } | |
4870 | ||
1e16c2f9 RD |
4871 | static int io_recv(struct io_kiocb *req, bool force_nonblock, |
4872 | struct io_comp_state *cs) | |
469956e8 Y |
4873 | { |
4874 | return -EOPNOTSUPP; | |
4875 | } | |
4876 | ||
4877 | static int io_accept_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) | |
4878 | { | |
4879 | return -EOPNOTSUPP; | |
4880 | } | |
4881 | ||
1e16c2f9 RD |
4882 | static int io_accept(struct io_kiocb *req, bool force_nonblock, |
4883 | struct io_comp_state *cs) | |
469956e8 Y |
4884 | { |
4885 | return -EOPNOTSUPP; | |
4886 | } | |
ce593a6c | 4887 | |
469956e8 Y |
4888 | static int io_connect_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
4889 | { | |
4890 | return -EOPNOTSUPP; | |
4891 | } | |
4892 | ||
1e16c2f9 RD |
4893 | static int io_connect(struct io_kiocb *req, bool force_nonblock, |
4894 | struct io_comp_state *cs) | |
469956e8 | 4895 | { |
f8e85cf2 | 4896 | return -EOPNOTSUPP; |
ce593a6c | 4897 | } |
469956e8 | 4898 | #endif /* CONFIG_NET */ |
f8e85cf2 | 4899 | |
d7718a9d JA |
4900 | struct io_poll_table { |
4901 | struct poll_table_struct pt; | |
4902 | struct io_kiocb *req; | |
4903 | int error; | |
4904 | }; | |
ce593a6c | 4905 | |
d7718a9d JA |
4906 | static int __io_async_wake(struct io_kiocb *req, struct io_poll_iocb *poll, |
4907 | __poll_t mask, task_work_func_t func) | |
4908 | { | |
fd7d6de2 | 4909 | bool twa_signal_ok; |
aa96bf8a | 4910 | int ret; |
d7718a9d JA |
4911 | |
4912 | /* for instances that support it check for an event match first: */ | |
4913 | if (mask && !(mask & poll->events)) | |
4914 | return 0; | |
4915 | ||
4916 | trace_io_uring_task_add(req->ctx, req->opcode, req->user_data, mask); | |
4917 | ||
4918 | list_del_init(&poll->wait.entry); | |
4919 | ||
d7718a9d JA |
4920 | req->result = mask; |
4921 | init_task_work(&req->task_work, func); | |
6d816e08 JA |
4922 | percpu_ref_get(&req->ctx->refs); |
4923 | ||
fd7d6de2 JA |
4924 | /* |
4925 | * If we using the signalfd wait_queue_head for this wakeup, then | |
4926 | * it's not safe to use TWA_SIGNAL as we could be recursing on the | |
4927 | * tsk->sighand->siglock on doing the wakeup. Should not be needed | |
4928 | * either, as the normal wakeup will suffice. | |
4929 | */ | |
4930 | twa_signal_ok = (poll->head != &req->task->sighand->signalfd_wqh); | |
4931 | ||
d7718a9d | 4932 | /* |
e3aabf95 JA |
4933 | * If this fails, then the task is exiting. When a task exits, the |
4934 | * work gets canceled, so just cancel this request as well instead | |
4935 | * of executing it. We can't safely execute it anyway, as we may not | |
4936 | * have the needed state needed for it anyway. | |
d7718a9d | 4937 | */ |
87c4311f | 4938 | ret = io_req_task_work_add(req, twa_signal_ok); |
aa96bf8a | 4939 | if (unlikely(ret)) { |
c2c4c83c JA |
4940 | struct task_struct *tsk; |
4941 | ||
e3aabf95 | 4942 | WRITE_ONCE(poll->canceled, true); |
aa96bf8a | 4943 | tsk = io_wq_get_task(req->ctx->io_wq); |
91989c70 | 4944 | task_work_add(tsk, &req->task_work, TWA_NONE); |
ce593a6c | 4945 | wake_up_process(tsk); |
aa96bf8a | 4946 | } |
d7718a9d JA |
4947 | return 1; |
4948 | } | |
4949 | ||
74ce6ce4 JA |
4950 | static bool io_poll_rewait(struct io_kiocb *req, struct io_poll_iocb *poll) |
4951 | __acquires(&req->ctx->completion_lock) | |
4952 | { | |
4953 | struct io_ring_ctx *ctx = req->ctx; | |
4954 | ||
4955 | if (!req->result && !READ_ONCE(poll->canceled)) { | |
4956 | struct poll_table_struct pt = { ._key = poll->events }; | |
4957 | ||
4958 | req->result = vfs_poll(req->file, &pt) & poll->events; | |
4959 | } | |
4960 | ||
4961 | spin_lock_irq(&ctx->completion_lock); | |
4962 | if (!req->result && !READ_ONCE(poll->canceled)) { | |
4963 | add_wait_queue(poll->head, &poll->wait); | |
4964 | return true; | |
4965 | } | |
4966 | ||
4967 | return false; | |
4968 | } | |
4969 | ||
d4e7cd36 | 4970 | static struct io_poll_iocb *io_poll_get_double(struct io_kiocb *req) |
18bceab1 | 4971 | { |
e8c2bc1f | 4972 | /* pure poll stashes this in ->async_data, poll driven retry elsewhere */ |
d4e7cd36 | 4973 | if (req->opcode == IORING_OP_POLL_ADD) |
e8c2bc1f | 4974 | return req->async_data; |
d4e7cd36 JA |
4975 | return req->apoll->double_poll; |
4976 | } | |
4977 | ||
4978 | static struct io_poll_iocb *io_poll_get_single(struct io_kiocb *req) | |
4979 | { | |
4980 | if (req->opcode == IORING_OP_POLL_ADD) | |
4981 | return &req->poll; | |
4982 | return &req->apoll->poll; | |
4983 | } | |
4984 | ||
4985 | static void io_poll_remove_double(struct io_kiocb *req) | |
4986 | { | |
4987 | struct io_poll_iocb *poll = io_poll_get_double(req); | |
18bceab1 JA |
4988 | |
4989 | lockdep_assert_held(&req->ctx->completion_lock); | |
4990 | ||
4991 | if (poll && poll->head) { | |
4992 | struct wait_queue_head *head = poll->head; | |
4993 | ||
4994 | spin_lock(&head->lock); | |
4995 | list_del_init(&poll->wait.entry); | |
4996 | if (poll->wait.private) | |
4997 | refcount_dec(&req->refs); | |
4998 | poll->head = NULL; | |
4999 | spin_unlock(&head->lock); | |
5000 | } | |
5001 | } | |
5002 | ||
5003 | static void io_poll_complete(struct io_kiocb *req, __poll_t mask, int error) | |
5004 | { | |
5005 | struct io_ring_ctx *ctx = req->ctx; | |
5006 | ||
d4e7cd36 | 5007 | io_poll_remove_double(req); |
18bceab1 JA |
5008 | req->poll.done = true; |
5009 | io_cqring_fill_event(req, error ? error : mangle_poll(mask)); | |
5010 | io_commit_cqring(ctx); | |
5011 | } | |
5012 | ||
dd221f46 | 5013 | static void io_poll_task_func(struct callback_head *cb) |
18bceab1 | 5014 | { |
dd221f46 | 5015 | struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work); |
18bceab1 | 5016 | struct io_ring_ctx *ctx = req->ctx; |
dd221f46 | 5017 | struct io_kiocb *nxt; |
18bceab1 JA |
5018 | |
5019 | if (io_poll_rewait(req, &req->poll)) { | |
5020 | spin_unlock_irq(&ctx->completion_lock); | |
dd221f46 PB |
5021 | } else { |
5022 | hash_del(&req->hash_node); | |
5023 | io_poll_complete(req, req->result, 0); | |
5024 | spin_unlock_irq(&ctx->completion_lock); | |
18bceab1 | 5025 | |
dd221f46 PB |
5026 | nxt = io_put_req_find_next(req); |
5027 | io_cqring_ev_posted(ctx); | |
5028 | if (nxt) | |
5029 | __io_req_task_submit(nxt); | |
5030 | } | |
18bceab1 | 5031 | |
6d816e08 | 5032 | percpu_ref_put(&ctx->refs); |
18bceab1 JA |
5033 | } |
5034 | ||
5035 | static int io_poll_double_wake(struct wait_queue_entry *wait, unsigned mode, | |
5036 | int sync, void *key) | |
5037 | { | |
5038 | struct io_kiocb *req = wait->private; | |
d4e7cd36 | 5039 | struct io_poll_iocb *poll = io_poll_get_single(req); |
18bceab1 JA |
5040 | __poll_t mask = key_to_poll(key); |
5041 | ||
5042 | /* for instances that support it check for an event match first: */ | |
5043 | if (mask && !(mask & poll->events)) | |
5044 | return 0; | |
5045 | ||
8706e04e JA |
5046 | list_del_init(&wait->entry); |
5047 | ||
807abcb0 | 5048 | if (poll && poll->head) { |
18bceab1 JA |
5049 | bool done; |
5050 | ||
807abcb0 JA |
5051 | spin_lock(&poll->head->lock); |
5052 | done = list_empty(&poll->wait.entry); | |
18bceab1 | 5053 | if (!done) |
807abcb0 | 5054 | list_del_init(&poll->wait.entry); |
d4e7cd36 JA |
5055 | /* make sure double remove sees this as being gone */ |
5056 | wait->private = NULL; | |
807abcb0 | 5057 | spin_unlock(&poll->head->lock); |
c8b5e260 JA |
5058 | if (!done) { |
5059 | /* use wait func handler, so it matches the rq type */ | |
5060 | poll->wait.func(&poll->wait, mode, sync, key); | |
5061 | } | |
18bceab1 JA |
5062 | } |
5063 | refcount_dec(&req->refs); | |
5064 | return 1; | |
5065 | } | |
5066 | ||
5067 | static void io_init_poll_iocb(struct io_poll_iocb *poll, __poll_t events, | |
5068 | wait_queue_func_t wake_func) | |
5069 | { | |
5070 | poll->head = NULL; | |
5071 | poll->done = false; | |
5072 | poll->canceled = false; | |
5073 | poll->events = events; | |
5074 | INIT_LIST_HEAD(&poll->wait.entry); | |
5075 | init_waitqueue_func_entry(&poll->wait, wake_func); | |
5076 | } | |
5077 | ||
5078 | static void __io_queue_proc(struct io_poll_iocb *poll, struct io_poll_table *pt, | |
807abcb0 JA |
5079 | struct wait_queue_head *head, |
5080 | struct io_poll_iocb **poll_ptr) | |
18bceab1 JA |
5081 | { |
5082 | struct io_kiocb *req = pt->req; | |
5083 | ||
5084 | /* | |
5085 | * If poll->head is already set, it's because the file being polled | |
5086 | * uses multiple waitqueues for poll handling (eg one for read, one | |
5087 | * for write). Setup a separate io_poll_iocb if this happens. | |
5088 | */ | |
5089 | if (unlikely(poll->head)) { | |
58852d4d PB |
5090 | struct io_poll_iocb *poll_one = poll; |
5091 | ||
18bceab1 | 5092 | /* already have a 2nd entry, fail a third attempt */ |
807abcb0 | 5093 | if (*poll_ptr) { |
18bceab1 JA |
5094 | pt->error = -EINVAL; |
5095 | return; | |
5096 | } | |
5097 | poll = kmalloc(sizeof(*poll), GFP_ATOMIC); | |
5098 | if (!poll) { | |
5099 | pt->error = -ENOMEM; | |
5100 | return; | |
5101 | } | |
58852d4d | 5102 | io_init_poll_iocb(poll, poll_one->events, io_poll_double_wake); |
18bceab1 JA |
5103 | refcount_inc(&req->refs); |
5104 | poll->wait.private = req; | |
807abcb0 | 5105 | *poll_ptr = poll; |
18bceab1 JA |
5106 | } |
5107 | ||
5108 | pt->error = 0; | |
5109 | poll->head = head; | |
a31eb4a2 JX |
5110 | |
5111 | if (poll->events & EPOLLEXCLUSIVE) | |
5112 | add_wait_queue_exclusive(head, &poll->wait); | |
5113 | else | |
5114 | add_wait_queue(head, &poll->wait); | |
18bceab1 JA |
5115 | } |
5116 | ||
5117 | static void io_async_queue_proc(struct file *file, struct wait_queue_head *head, | |
5118 | struct poll_table_struct *p) | |
5119 | { | |
5120 | struct io_poll_table *pt = container_of(p, struct io_poll_table, pt); | |
807abcb0 | 5121 | struct async_poll *apoll = pt->req->apoll; |
18bceab1 | 5122 | |
807abcb0 | 5123 | __io_queue_proc(&apoll->poll, pt, head, &apoll->double_poll); |
18bceab1 JA |
5124 | } |
5125 | ||
d7718a9d JA |
5126 | static void io_async_task_func(struct callback_head *cb) |
5127 | { | |
5128 | struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work); | |
5129 | struct async_poll *apoll = req->apoll; | |
5130 | struct io_ring_ctx *ctx = req->ctx; | |
5131 | ||
5132 | trace_io_uring_task_run(req->ctx, req->opcode, req->user_data); | |
5133 | ||
74ce6ce4 | 5134 | if (io_poll_rewait(req, &apoll->poll)) { |
d7718a9d | 5135 | spin_unlock_irq(&ctx->completion_lock); |
6d816e08 | 5136 | percpu_ref_put(&ctx->refs); |
74ce6ce4 | 5137 | return; |
d7718a9d JA |
5138 | } |
5139 | ||
31067255 | 5140 | /* If req is still hashed, it cannot have been canceled. Don't check. */ |
0be0b0e3 | 5141 | if (hash_hashed(&req->hash_node)) |
74ce6ce4 | 5142 | hash_del(&req->hash_node); |
2bae047e | 5143 | |
d4e7cd36 | 5144 | io_poll_remove_double(req); |
74ce6ce4 JA |
5145 | spin_unlock_irq(&ctx->completion_lock); |
5146 | ||
0be0b0e3 PB |
5147 | if (!READ_ONCE(apoll->poll.canceled)) |
5148 | __io_req_task_submit(req); | |
5149 | else | |
5150 | __io_req_task_cancel(req, -ECANCELED); | |
aa340845 | 5151 | |
6d816e08 | 5152 | percpu_ref_put(&ctx->refs); |
807abcb0 | 5153 | kfree(apoll->double_poll); |
31067255 | 5154 | kfree(apoll); |
d7718a9d JA |
5155 | } |
5156 | ||
5157 | static int io_async_wake(struct wait_queue_entry *wait, unsigned mode, int sync, | |
5158 | void *key) | |
5159 | { | |
5160 | struct io_kiocb *req = wait->private; | |
5161 | struct io_poll_iocb *poll = &req->apoll->poll; | |
5162 | ||
5163 | trace_io_uring_poll_wake(req->ctx, req->opcode, req->user_data, | |
5164 | key_to_poll(key)); | |
5165 | ||
5166 | return __io_async_wake(req, poll, key_to_poll(key), io_async_task_func); | |
5167 | } | |
5168 | ||
5169 | static void io_poll_req_insert(struct io_kiocb *req) | |
5170 | { | |
5171 | struct io_ring_ctx *ctx = req->ctx; | |
5172 | struct hlist_head *list; | |
5173 | ||
5174 | list = &ctx->cancel_hash[hash_long(req->user_data, ctx->cancel_hash_bits)]; | |
5175 | hlist_add_head(&req->hash_node, list); | |
5176 | } | |
5177 | ||
5178 | static __poll_t __io_arm_poll_handler(struct io_kiocb *req, | |
5179 | struct io_poll_iocb *poll, | |
5180 | struct io_poll_table *ipt, __poll_t mask, | |
5181 | wait_queue_func_t wake_func) | |
5182 | __acquires(&ctx->completion_lock) | |
5183 | { | |
5184 | struct io_ring_ctx *ctx = req->ctx; | |
5185 | bool cancel = false; | |
5186 | ||
4d52f338 | 5187 | INIT_HLIST_NODE(&req->hash_node); |
18bceab1 | 5188 | io_init_poll_iocb(poll, mask, wake_func); |
b90cd197 | 5189 | poll->file = req->file; |
18bceab1 | 5190 | poll->wait.private = req; |
d7718a9d JA |
5191 | |
5192 | ipt->pt._key = mask; | |
5193 | ipt->req = req; | |
5194 | ipt->error = -EINVAL; | |
5195 | ||
d7718a9d JA |
5196 | mask = vfs_poll(req->file, &ipt->pt) & poll->events; |
5197 | ||
5198 | spin_lock_irq(&ctx->completion_lock); | |
5199 | if (likely(poll->head)) { | |
5200 | spin_lock(&poll->head->lock); | |
5201 | if (unlikely(list_empty(&poll->wait.entry))) { | |
5202 | if (ipt->error) | |
5203 | cancel = true; | |
5204 | ipt->error = 0; | |
5205 | mask = 0; | |
5206 | } | |
5207 | if (mask || ipt->error) | |
5208 | list_del_init(&poll->wait.entry); | |
5209 | else if (cancel) | |
5210 | WRITE_ONCE(poll->canceled, true); | |
5211 | else if (!poll->done) /* actually waiting for an event */ | |
5212 | io_poll_req_insert(req); | |
5213 | spin_unlock(&poll->head->lock); | |
5214 | } | |
5215 | ||
5216 | return mask; | |
5217 | } | |
5218 | ||
5219 | static bool io_arm_poll_handler(struct io_kiocb *req) | |
5220 | { | |
5221 | const struct io_op_def *def = &io_op_defs[req->opcode]; | |
5222 | struct io_ring_ctx *ctx = req->ctx; | |
5223 | struct async_poll *apoll; | |
5224 | struct io_poll_table ipt; | |
5225 | __poll_t mask, ret; | |
9dab14b8 | 5226 | int rw; |
d7718a9d JA |
5227 | |
5228 | if (!req->file || !file_can_poll(req->file)) | |
5229 | return false; | |
24c74678 | 5230 | if (req->flags & REQ_F_POLLED) |
d7718a9d | 5231 | return false; |
9dab14b8 JA |
5232 | if (def->pollin) |
5233 | rw = READ; | |
5234 | else if (def->pollout) | |
5235 | rw = WRITE; | |
5236 | else | |
5237 | return false; | |
5238 | /* if we can't nonblock try, then no point in arming a poll handler */ | |
5239 | if (!io_file_supports_async(req->file, rw)) | |
d7718a9d JA |
5240 | return false; |
5241 | ||
5242 | apoll = kmalloc(sizeof(*apoll), GFP_ATOMIC); | |
5243 | if (unlikely(!apoll)) | |
5244 | return false; | |
807abcb0 | 5245 | apoll->double_poll = NULL; |
d7718a9d JA |
5246 | |
5247 | req->flags |= REQ_F_POLLED; | |
d7718a9d | 5248 | req->apoll = apoll; |
d7718a9d | 5249 | |
8755d97a | 5250 | mask = 0; |
d7718a9d | 5251 | if (def->pollin) |
8755d97a | 5252 | mask |= POLLIN | POLLRDNORM; |
d7718a9d JA |
5253 | if (def->pollout) |
5254 | mask |= POLLOUT | POLLWRNORM; | |
901341bb LH |
5255 | |
5256 | /* If reading from MSG_ERRQUEUE using recvmsg, ignore POLLIN */ | |
5257 | if ((req->opcode == IORING_OP_RECVMSG) && | |
5258 | (req->sr_msg.msg_flags & MSG_ERRQUEUE)) | |
5259 | mask &= ~POLLIN; | |
5260 | ||
d7718a9d JA |
5261 | mask |= POLLERR | POLLPRI; |
5262 | ||
5263 | ipt.pt._qproc = io_async_queue_proc; | |
5264 | ||
5265 | ret = __io_arm_poll_handler(req, &apoll->poll, &ipt, mask, | |
5266 | io_async_wake); | |
a36da65c | 5267 | if (ret || ipt.error) { |
d4e7cd36 | 5268 | io_poll_remove_double(req); |
d7718a9d | 5269 | spin_unlock_irq(&ctx->completion_lock); |
807abcb0 | 5270 | kfree(apoll->double_poll); |
d7718a9d JA |
5271 | kfree(apoll); |
5272 | return false; | |
5273 | } | |
5274 | spin_unlock_irq(&ctx->completion_lock); | |
5275 | trace_io_uring_poll_arm(ctx, req->opcode, req->user_data, mask, | |
5276 | apoll->poll.events); | |
5277 | return true; | |
5278 | } | |
5279 | ||
5280 | static bool __io_poll_remove_one(struct io_kiocb *req, | |
5281 | struct io_poll_iocb *poll) | |
221c5eb2 | 5282 | { |
b41e9852 | 5283 | bool do_complete = false; |
221c5eb2 JA |
5284 | |
5285 | spin_lock(&poll->head->lock); | |
5286 | WRITE_ONCE(poll->canceled, true); | |
392edb45 JA |
5287 | if (!list_empty(&poll->wait.entry)) { |
5288 | list_del_init(&poll->wait.entry); | |
b41e9852 | 5289 | do_complete = true; |
221c5eb2 JA |
5290 | } |
5291 | spin_unlock(&poll->head->lock); | |
3bfa5bcb | 5292 | hash_del(&req->hash_node); |
d7718a9d JA |
5293 | return do_complete; |
5294 | } | |
5295 | ||
5296 | static bool io_poll_remove_one(struct io_kiocb *req) | |
5297 | { | |
5298 | bool do_complete; | |
5299 | ||
d4e7cd36 JA |
5300 | io_poll_remove_double(req); |
5301 | ||
d7718a9d JA |
5302 | if (req->opcode == IORING_OP_POLL_ADD) { |
5303 | do_complete = __io_poll_remove_one(req, &req->poll); | |
5304 | } else { | |
3bfa5bcb JA |
5305 | struct async_poll *apoll = req->apoll; |
5306 | ||
d7718a9d | 5307 | /* non-poll requests have submit ref still */ |
3bfa5bcb JA |
5308 | do_complete = __io_poll_remove_one(req, &apoll->poll); |
5309 | if (do_complete) { | |
d7718a9d | 5310 | io_put_req(req); |
807abcb0 | 5311 | kfree(apoll->double_poll); |
3bfa5bcb JA |
5312 | kfree(apoll); |
5313 | } | |
b1f573bd XW |
5314 | } |
5315 | ||
b41e9852 JA |
5316 | if (do_complete) { |
5317 | io_cqring_fill_event(req, -ECANCELED); | |
5318 | io_commit_cqring(req->ctx); | |
f254ac04 | 5319 | req_set_fail_links(req); |
216578e5 | 5320 | io_put_req_deferred(req, 1); |
b41e9852 JA |
5321 | } |
5322 | ||
5323 | return do_complete; | |
221c5eb2 JA |
5324 | } |
5325 | ||
76e1b642 JA |
5326 | /* |
5327 | * Returns true if we found and killed one or more poll requests | |
5328 | */ | |
5329 | static bool io_poll_remove_all(struct io_ring_ctx *ctx, struct task_struct *tsk) | |
221c5eb2 | 5330 | { |
78076bb6 | 5331 | struct hlist_node *tmp; |
221c5eb2 | 5332 | struct io_kiocb *req; |
8e2e1faf | 5333 | int posted = 0, i; |
221c5eb2 JA |
5334 | |
5335 | spin_lock_irq(&ctx->completion_lock); | |
78076bb6 JA |
5336 | for (i = 0; i < (1U << ctx->cancel_hash_bits); i++) { |
5337 | struct hlist_head *list; | |
5338 | ||
5339 | list = &ctx->cancel_hash[i]; | |
f3606e3a JA |
5340 | hlist_for_each_entry_safe(req, tmp, list, hash_node) { |
5341 | if (io_task_match(req, tsk)) | |
5342 | posted += io_poll_remove_one(req); | |
5343 | } | |
221c5eb2 JA |
5344 | } |
5345 | spin_unlock_irq(&ctx->completion_lock); | |
b41e9852 | 5346 | |
8e2e1faf JA |
5347 | if (posted) |
5348 | io_cqring_ev_posted(ctx); | |
76e1b642 JA |
5349 | |
5350 | return posted != 0; | |
221c5eb2 JA |
5351 | } |
5352 | ||
47f46768 JA |
5353 | static int io_poll_cancel(struct io_ring_ctx *ctx, __u64 sqe_addr) |
5354 | { | |
78076bb6 | 5355 | struct hlist_head *list; |
47f46768 JA |
5356 | struct io_kiocb *req; |
5357 | ||
78076bb6 JA |
5358 | list = &ctx->cancel_hash[hash_long(sqe_addr, ctx->cancel_hash_bits)]; |
5359 | hlist_for_each_entry(req, list, hash_node) { | |
b41e9852 JA |
5360 | if (sqe_addr != req->user_data) |
5361 | continue; | |
5362 | if (io_poll_remove_one(req)) | |
eac406c6 | 5363 | return 0; |
b41e9852 | 5364 | return -EALREADY; |
47f46768 JA |
5365 | } |
5366 | ||
5367 | return -ENOENT; | |
5368 | } | |
5369 | ||
3529d8c2 JA |
5370 | static int io_poll_remove_prep(struct io_kiocb *req, |
5371 | const struct io_uring_sqe *sqe) | |
0969e783 | 5372 | { |
0969e783 JA |
5373 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
5374 | return -EINVAL; | |
5375 | if (sqe->ioprio || sqe->off || sqe->len || sqe->buf_index || | |
5376 | sqe->poll_events) | |
5377 | return -EINVAL; | |
5378 | ||
5379 | req->poll.addr = READ_ONCE(sqe->addr); | |
0969e783 JA |
5380 | return 0; |
5381 | } | |
5382 | ||
221c5eb2 JA |
5383 | /* |
5384 | * Find a running poll command that matches one specified in sqe->addr, | |
5385 | * and remove it if found. | |
5386 | */ | |
fc4df999 | 5387 | static int io_poll_remove(struct io_kiocb *req) |
221c5eb2 JA |
5388 | { |
5389 | struct io_ring_ctx *ctx = req->ctx; | |
0969e783 | 5390 | u64 addr; |
47f46768 | 5391 | int ret; |
221c5eb2 | 5392 | |
0969e783 | 5393 | addr = req->poll.addr; |
221c5eb2 | 5394 | spin_lock_irq(&ctx->completion_lock); |
0969e783 | 5395 | ret = io_poll_cancel(ctx, addr); |
221c5eb2 JA |
5396 | spin_unlock_irq(&ctx->completion_lock); |
5397 | ||
4e88d6e7 JA |
5398 | if (ret < 0) |
5399 | req_set_fail_links(req); | |
e1e16097 | 5400 | io_req_complete(req, ret); |
221c5eb2 JA |
5401 | return 0; |
5402 | } | |
5403 | ||
221c5eb2 JA |
5404 | static int io_poll_wake(struct wait_queue_entry *wait, unsigned mode, int sync, |
5405 | void *key) | |
5406 | { | |
c2f2eb7d JA |
5407 | struct io_kiocb *req = wait->private; |
5408 | struct io_poll_iocb *poll = &req->poll; | |
221c5eb2 | 5409 | |
d7718a9d | 5410 | return __io_async_wake(req, poll, key_to_poll(key), io_poll_task_func); |
221c5eb2 JA |
5411 | } |
5412 | ||
221c5eb2 JA |
5413 | static void io_poll_queue_proc(struct file *file, struct wait_queue_head *head, |
5414 | struct poll_table_struct *p) | |
5415 | { | |
5416 | struct io_poll_table *pt = container_of(p, struct io_poll_table, pt); | |
5417 | ||
e8c2bc1f | 5418 | __io_queue_proc(&pt->req->poll, pt, head, (struct io_poll_iocb **) &pt->req->async_data); |
eac406c6 JA |
5419 | } |
5420 | ||
3529d8c2 | 5421 | static int io_poll_add_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
221c5eb2 JA |
5422 | { |
5423 | struct io_poll_iocb *poll = &req->poll; | |
5769a351 | 5424 | u32 events; |
221c5eb2 JA |
5425 | |
5426 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) | |
5427 | return -EINVAL; | |
5428 | if (sqe->addr || sqe->ioprio || sqe->off || sqe->len || sqe->buf_index) | |
5429 | return -EINVAL; | |
5430 | ||
5769a351 JX |
5431 | events = READ_ONCE(sqe->poll32_events); |
5432 | #ifdef __BIG_ENDIAN | |
5433 | events = swahw32(events); | |
5434 | #endif | |
a31eb4a2 JX |
5435 | poll->events = demangle_poll(events) | EPOLLERR | EPOLLHUP | |
5436 | (events & EPOLLEXCLUSIVE); | |
0969e783 JA |
5437 | return 0; |
5438 | } | |
5439 | ||
014db007 | 5440 | static int io_poll_add(struct io_kiocb *req) |
0969e783 JA |
5441 | { |
5442 | struct io_poll_iocb *poll = &req->poll; | |
5443 | struct io_ring_ctx *ctx = req->ctx; | |
5444 | struct io_poll_table ipt; | |
0969e783 | 5445 | __poll_t mask; |
0969e783 | 5446 | |
d7718a9d | 5447 | ipt.pt._qproc = io_poll_queue_proc; |
36703247 | 5448 | |
d7718a9d JA |
5449 | mask = __io_arm_poll_handler(req, &req->poll, &ipt, poll->events, |
5450 | io_poll_wake); | |
221c5eb2 | 5451 | |
8c838788 | 5452 | if (mask) { /* no async, we'd stolen it */ |
221c5eb2 | 5453 | ipt.error = 0; |
b0dd8a41 | 5454 | io_poll_complete(req, mask, 0); |
221c5eb2 | 5455 | } |
221c5eb2 JA |
5456 | spin_unlock_irq(&ctx->completion_lock); |
5457 | ||
8c838788 JA |
5458 | if (mask) { |
5459 | io_cqring_ev_posted(ctx); | |
014db007 | 5460 | io_put_req(req); |
221c5eb2 | 5461 | } |
8c838788 | 5462 | return ipt.error; |
221c5eb2 JA |
5463 | } |
5464 | ||
5262f567 JA |
5465 | static enum hrtimer_restart io_timeout_fn(struct hrtimer *timer) |
5466 | { | |
ad8a48ac JA |
5467 | struct io_timeout_data *data = container_of(timer, |
5468 | struct io_timeout_data, timer); | |
5469 | struct io_kiocb *req = data->req; | |
5470 | struct io_ring_ctx *ctx = req->ctx; | |
5262f567 JA |
5471 | unsigned long flags; |
5472 | ||
5262f567 | 5473 | spin_lock_irqsave(&ctx->completion_lock, flags); |
a71976f3 | 5474 | list_del_init(&req->timeout.list); |
01cec8c1 PB |
5475 | atomic_set(&req->ctx->cq_timeouts, |
5476 | atomic_read(&req->ctx->cq_timeouts) + 1); | |
5477 | ||
78e19bbe | 5478 | io_cqring_fill_event(req, -ETIME); |
5262f567 JA |
5479 | io_commit_cqring(ctx); |
5480 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
5481 | ||
5482 | io_cqring_ev_posted(ctx); | |
4e88d6e7 | 5483 | req_set_fail_links(req); |
5262f567 JA |
5484 | io_put_req(req); |
5485 | return HRTIMER_NORESTART; | |
5486 | } | |
5487 | ||
f254ac04 JA |
5488 | static int __io_timeout_cancel(struct io_kiocb *req) |
5489 | { | |
e8c2bc1f | 5490 | struct io_timeout_data *io = req->async_data; |
f254ac04 JA |
5491 | int ret; |
5492 | ||
e8c2bc1f | 5493 | ret = hrtimer_try_to_cancel(&io->timer); |
f254ac04 JA |
5494 | if (ret == -1) |
5495 | return -EALREADY; | |
a71976f3 | 5496 | list_del_init(&req->timeout.list); |
f254ac04 JA |
5497 | |
5498 | req_set_fail_links(req); | |
f254ac04 | 5499 | io_cqring_fill_event(req, -ECANCELED); |
216578e5 | 5500 | io_put_req_deferred(req, 1); |
f254ac04 JA |
5501 | return 0; |
5502 | } | |
5503 | ||
47f46768 JA |
5504 | static int io_timeout_cancel(struct io_ring_ctx *ctx, __u64 user_data) |
5505 | { | |
5506 | struct io_kiocb *req; | |
5507 | int ret = -ENOENT; | |
5508 | ||
135fcde8 | 5509 | list_for_each_entry(req, &ctx->timeout_list, timeout.list) { |
47f46768 | 5510 | if (user_data == req->user_data) { |
47f46768 JA |
5511 | ret = 0; |
5512 | break; | |
5513 | } | |
5514 | } | |
5515 | ||
5516 | if (ret == -ENOENT) | |
5517 | return ret; | |
5518 | ||
f254ac04 | 5519 | return __io_timeout_cancel(req); |
47f46768 JA |
5520 | } |
5521 | ||
3529d8c2 JA |
5522 | static int io_timeout_remove_prep(struct io_kiocb *req, |
5523 | const struct io_uring_sqe *sqe) | |
b29472ee | 5524 | { |
b29472ee JA |
5525 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
5526 | return -EINVAL; | |
61710e43 DA |
5527 | if (unlikely(req->flags & (REQ_F_FIXED_FILE | REQ_F_BUFFER_SELECT))) |
5528 | return -EINVAL; | |
0bdf7a2d | 5529 | if (sqe->ioprio || sqe->buf_index || sqe->len || sqe->timeout_flags) |
b29472ee JA |
5530 | return -EINVAL; |
5531 | ||
0bdf7a2d | 5532 | req->timeout_rem.addr = READ_ONCE(sqe->addr); |
b29472ee JA |
5533 | return 0; |
5534 | } | |
5535 | ||
11365043 JA |
5536 | /* |
5537 | * Remove or update an existing timeout command | |
5538 | */ | |
fc4df999 | 5539 | static int io_timeout_remove(struct io_kiocb *req) |
11365043 JA |
5540 | { |
5541 | struct io_ring_ctx *ctx = req->ctx; | |
47f46768 | 5542 | int ret; |
11365043 | 5543 | |
11365043 | 5544 | spin_lock_irq(&ctx->completion_lock); |
0bdf7a2d | 5545 | ret = io_timeout_cancel(ctx, req->timeout_rem.addr); |
11365043 | 5546 | |
47f46768 | 5547 | io_cqring_fill_event(req, ret); |
11365043 JA |
5548 | io_commit_cqring(ctx); |
5549 | spin_unlock_irq(&ctx->completion_lock); | |
5262f567 | 5550 | io_cqring_ev_posted(ctx); |
4e88d6e7 JA |
5551 | if (ret < 0) |
5552 | req_set_fail_links(req); | |
ec9c02ad | 5553 | io_put_req(req); |
11365043 | 5554 | return 0; |
5262f567 JA |
5555 | } |
5556 | ||
3529d8c2 | 5557 | static int io_timeout_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe, |
2d28390a | 5558 | bool is_timeout_link) |
5262f567 | 5559 | { |
ad8a48ac | 5560 | struct io_timeout_data *data; |
a41525ab | 5561 | unsigned flags; |
56080b02 | 5562 | u32 off = READ_ONCE(sqe->off); |
5262f567 | 5563 | |
ad8a48ac | 5564 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
5262f567 | 5565 | return -EINVAL; |
ad8a48ac | 5566 | if (sqe->ioprio || sqe->buf_index || sqe->len != 1) |
a41525ab | 5567 | return -EINVAL; |
56080b02 | 5568 | if (off && is_timeout_link) |
2d28390a | 5569 | return -EINVAL; |
a41525ab JA |
5570 | flags = READ_ONCE(sqe->timeout_flags); |
5571 | if (flags & ~IORING_TIMEOUT_ABS) | |
5262f567 | 5572 | return -EINVAL; |
bdf20073 | 5573 | |
bfe68a22 | 5574 | req->timeout.off = off; |
26a61679 | 5575 | |
e8c2bc1f | 5576 | if (!req->async_data && io_alloc_async_data(req)) |
26a61679 JA |
5577 | return -ENOMEM; |
5578 | ||
e8c2bc1f | 5579 | data = req->async_data; |
ad8a48ac | 5580 | data->req = req; |
ad8a48ac JA |
5581 | |
5582 | if (get_timespec64(&data->ts, u64_to_user_ptr(sqe->addr))) | |
5262f567 JA |
5583 | return -EFAULT; |
5584 | ||
11365043 | 5585 | if (flags & IORING_TIMEOUT_ABS) |
ad8a48ac | 5586 | data->mode = HRTIMER_MODE_ABS; |
11365043 | 5587 | else |
ad8a48ac | 5588 | data->mode = HRTIMER_MODE_REL; |
11365043 | 5589 | |
ad8a48ac JA |
5590 | hrtimer_init(&data->timer, CLOCK_MONOTONIC, data->mode); |
5591 | return 0; | |
5592 | } | |
5593 | ||
fc4df999 | 5594 | static int io_timeout(struct io_kiocb *req) |
ad8a48ac | 5595 | { |
ad8a48ac | 5596 | struct io_ring_ctx *ctx = req->ctx; |
e8c2bc1f | 5597 | struct io_timeout_data *data = req->async_data; |
ad8a48ac | 5598 | struct list_head *entry; |
bfe68a22 | 5599 | u32 tail, off = req->timeout.off; |
ad8a48ac | 5600 | |
733f5c95 | 5601 | spin_lock_irq(&ctx->completion_lock); |
93bd25bb | 5602 | |
5262f567 JA |
5603 | /* |
5604 | * sqe->off holds how many events that need to occur for this | |
93bd25bb JA |
5605 | * timeout event to be satisfied. If it isn't set, then this is |
5606 | * a pure timeout request, sequence isn't used. | |
5262f567 | 5607 | */ |
8eb7e2d0 | 5608 | if (io_is_timeout_noseq(req)) { |
93bd25bb JA |
5609 | entry = ctx->timeout_list.prev; |
5610 | goto add; | |
5611 | } | |
5262f567 | 5612 | |
bfe68a22 PB |
5613 | tail = ctx->cached_cq_tail - atomic_read(&ctx->cq_timeouts); |
5614 | req->timeout.target_seq = tail + off; | |
5262f567 JA |
5615 | |
5616 | /* | |
5617 | * Insertion sort, ensuring the first entry in the list is always | |
5618 | * the one we need first. | |
5619 | */ | |
5262f567 | 5620 | list_for_each_prev(entry, &ctx->timeout_list) { |
135fcde8 PB |
5621 | struct io_kiocb *nxt = list_entry(entry, struct io_kiocb, |
5622 | timeout.list); | |
5262f567 | 5623 | |
8eb7e2d0 | 5624 | if (io_is_timeout_noseq(nxt)) |
93bd25bb | 5625 | continue; |
bfe68a22 PB |
5626 | /* nxt.seq is behind @tail, otherwise would've been completed */ |
5627 | if (off >= nxt->timeout.target_seq - tail) | |
5262f567 JA |
5628 | break; |
5629 | } | |
93bd25bb | 5630 | add: |
135fcde8 | 5631 | list_add(&req->timeout.list, entry); |
ad8a48ac JA |
5632 | data->timer.function = io_timeout_fn; |
5633 | hrtimer_start(&data->timer, timespec64_to_ktime(data->ts), data->mode); | |
5262f567 | 5634 | spin_unlock_irq(&ctx->completion_lock); |
5262f567 JA |
5635 | return 0; |
5636 | } | |
5262f567 | 5637 | |
62755e35 JA |
5638 | static bool io_cancel_cb(struct io_wq_work *work, void *data) |
5639 | { | |
5640 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); | |
5641 | ||
5642 | return req->user_data == (unsigned long) data; | |
5643 | } | |
5644 | ||
e977d6d3 | 5645 | static int io_async_cancel_one(struct io_ring_ctx *ctx, void *sqe_addr) |
62755e35 | 5646 | { |
62755e35 | 5647 | enum io_wq_cancel cancel_ret; |
62755e35 JA |
5648 | int ret = 0; |
5649 | ||
4f26bda1 | 5650 | cancel_ret = io_wq_cancel_cb(ctx->io_wq, io_cancel_cb, sqe_addr, false); |
62755e35 JA |
5651 | switch (cancel_ret) { |
5652 | case IO_WQ_CANCEL_OK: | |
5653 | ret = 0; | |
5654 | break; | |
5655 | case IO_WQ_CANCEL_RUNNING: | |
5656 | ret = -EALREADY; | |
5657 | break; | |
5658 | case IO_WQ_CANCEL_NOTFOUND: | |
5659 | ret = -ENOENT; | |
5660 | break; | |
5661 | } | |
5662 | ||
e977d6d3 JA |
5663 | return ret; |
5664 | } | |
5665 | ||
47f46768 JA |
5666 | static void io_async_find_and_cancel(struct io_ring_ctx *ctx, |
5667 | struct io_kiocb *req, __u64 sqe_addr, | |
014db007 | 5668 | int success_ret) |
47f46768 JA |
5669 | { |
5670 | unsigned long flags; | |
5671 | int ret; | |
5672 | ||
5673 | ret = io_async_cancel_one(ctx, (void *) (unsigned long) sqe_addr); | |
5674 | if (ret != -ENOENT) { | |
5675 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
5676 | goto done; | |
5677 | } | |
5678 | ||
5679 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
5680 | ret = io_timeout_cancel(ctx, sqe_addr); | |
5681 | if (ret != -ENOENT) | |
5682 | goto done; | |
5683 | ret = io_poll_cancel(ctx, sqe_addr); | |
5684 | done: | |
b0dd8a41 JA |
5685 | if (!ret) |
5686 | ret = success_ret; | |
47f46768 JA |
5687 | io_cqring_fill_event(req, ret); |
5688 | io_commit_cqring(ctx); | |
5689 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
5690 | io_cqring_ev_posted(ctx); | |
5691 | ||
4e88d6e7 JA |
5692 | if (ret < 0) |
5693 | req_set_fail_links(req); | |
014db007 | 5694 | io_put_req(req); |
47f46768 JA |
5695 | } |
5696 | ||
3529d8c2 JA |
5697 | static int io_async_cancel_prep(struct io_kiocb *req, |
5698 | const struct io_uring_sqe *sqe) | |
e977d6d3 | 5699 | { |
fbf23849 | 5700 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
e977d6d3 | 5701 | return -EINVAL; |
61710e43 DA |
5702 | if (unlikely(req->flags & (REQ_F_FIXED_FILE | REQ_F_BUFFER_SELECT))) |
5703 | return -EINVAL; | |
5704 | if (sqe->ioprio || sqe->off || sqe->len || sqe->cancel_flags) | |
e977d6d3 JA |
5705 | return -EINVAL; |
5706 | ||
fbf23849 JA |
5707 | req->cancel.addr = READ_ONCE(sqe->addr); |
5708 | return 0; | |
5709 | } | |
5710 | ||
014db007 | 5711 | static int io_async_cancel(struct io_kiocb *req) |
fbf23849 JA |
5712 | { |
5713 | struct io_ring_ctx *ctx = req->ctx; | |
fbf23849 | 5714 | |
014db007 | 5715 | io_async_find_and_cancel(ctx, req, req->cancel.addr, 0); |
5262f567 JA |
5716 | return 0; |
5717 | } | |
5718 | ||
05f3fb3c JA |
5719 | static int io_files_update_prep(struct io_kiocb *req, |
5720 | const struct io_uring_sqe *sqe) | |
5721 | { | |
6ca56f84 JA |
5722 | if (unlikely(req->ctx->flags & IORING_SETUP_SQPOLL)) |
5723 | return -EINVAL; | |
61710e43 DA |
5724 | if (unlikely(req->flags & (REQ_F_FIXED_FILE | REQ_F_BUFFER_SELECT))) |
5725 | return -EINVAL; | |
5726 | if (sqe->ioprio || sqe->rw_flags) | |
05f3fb3c JA |
5727 | return -EINVAL; |
5728 | ||
5729 | req->files_update.offset = READ_ONCE(sqe->off); | |
5730 | req->files_update.nr_args = READ_ONCE(sqe->len); | |
5731 | if (!req->files_update.nr_args) | |
5732 | return -EINVAL; | |
5733 | req->files_update.arg = READ_ONCE(sqe->addr); | |
5734 | return 0; | |
5735 | } | |
5736 | ||
229a7b63 JA |
5737 | static int io_files_update(struct io_kiocb *req, bool force_nonblock, |
5738 | struct io_comp_state *cs) | |
fbf23849 JA |
5739 | { |
5740 | struct io_ring_ctx *ctx = req->ctx; | |
05f3fb3c JA |
5741 | struct io_uring_files_update up; |
5742 | int ret; | |
fbf23849 | 5743 | |
f86cd20c | 5744 | if (force_nonblock) |
05f3fb3c | 5745 | return -EAGAIN; |
05f3fb3c JA |
5746 | |
5747 | up.offset = req->files_update.offset; | |
5748 | up.fds = req->files_update.arg; | |
5749 | ||
5750 | mutex_lock(&ctx->uring_lock); | |
5751 | ret = __io_sqe_files_update(ctx, &up, req->files_update.nr_args); | |
5752 | mutex_unlock(&ctx->uring_lock); | |
5753 | ||
5754 | if (ret < 0) | |
5755 | req_set_fail_links(req); | |
229a7b63 | 5756 | __io_req_complete(req, ret, 0, cs); |
5262f567 JA |
5757 | return 0; |
5758 | } | |
5759 | ||
bfe76559 | 5760 | static int io_req_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
f67676d1 | 5761 | { |
d625c6ee | 5762 | switch (req->opcode) { |
e781573e | 5763 | case IORING_OP_NOP: |
bfe76559 | 5764 | return 0; |
f67676d1 JA |
5765 | case IORING_OP_READV: |
5766 | case IORING_OP_READ_FIXED: | |
3a6820f2 | 5767 | case IORING_OP_READ: |
bfe76559 | 5768 | return io_read_prep(req, sqe); |
f67676d1 JA |
5769 | case IORING_OP_WRITEV: |
5770 | case IORING_OP_WRITE_FIXED: | |
3a6820f2 | 5771 | case IORING_OP_WRITE: |
bfe76559 | 5772 | return io_write_prep(req, sqe); |
0969e783 | 5773 | case IORING_OP_POLL_ADD: |
bfe76559 | 5774 | return io_poll_add_prep(req, sqe); |
0969e783 | 5775 | case IORING_OP_POLL_REMOVE: |
bfe76559 | 5776 | return io_poll_remove_prep(req, sqe); |
8ed8d3c3 | 5777 | case IORING_OP_FSYNC: |
bfe76559 | 5778 | return io_prep_fsync(req, sqe); |
8ed8d3c3 | 5779 | case IORING_OP_SYNC_FILE_RANGE: |
bfe76559 | 5780 | return io_prep_sfr(req, sqe); |
03b1230c | 5781 | case IORING_OP_SENDMSG: |
fddaface | 5782 | case IORING_OP_SEND: |
bfe76559 | 5783 | return io_sendmsg_prep(req, sqe); |
03b1230c | 5784 | case IORING_OP_RECVMSG: |
fddaface | 5785 | case IORING_OP_RECV: |
bfe76559 | 5786 | return io_recvmsg_prep(req, sqe); |
f499a021 | 5787 | case IORING_OP_CONNECT: |
bfe76559 | 5788 | return io_connect_prep(req, sqe); |
2d28390a | 5789 | case IORING_OP_TIMEOUT: |
bfe76559 | 5790 | return io_timeout_prep(req, sqe, false); |
b29472ee | 5791 | case IORING_OP_TIMEOUT_REMOVE: |
bfe76559 | 5792 | return io_timeout_remove_prep(req, sqe); |
fbf23849 | 5793 | case IORING_OP_ASYNC_CANCEL: |
bfe76559 | 5794 | return io_async_cancel_prep(req, sqe); |
2d28390a | 5795 | case IORING_OP_LINK_TIMEOUT: |
bfe76559 | 5796 | return io_timeout_prep(req, sqe, true); |
8ed8d3c3 | 5797 | case IORING_OP_ACCEPT: |
bfe76559 | 5798 | return io_accept_prep(req, sqe); |
d63d1b5e | 5799 | case IORING_OP_FALLOCATE: |
bfe76559 | 5800 | return io_fallocate_prep(req, sqe); |
15b71abe | 5801 | case IORING_OP_OPENAT: |
bfe76559 | 5802 | return io_openat_prep(req, sqe); |
b5dba59e | 5803 | case IORING_OP_CLOSE: |
bfe76559 | 5804 | return io_close_prep(req, sqe); |
05f3fb3c | 5805 | case IORING_OP_FILES_UPDATE: |
bfe76559 | 5806 | return io_files_update_prep(req, sqe); |
eddc7ef5 | 5807 | case IORING_OP_STATX: |
bfe76559 | 5808 | return io_statx_prep(req, sqe); |
4840e418 | 5809 | case IORING_OP_FADVISE: |
bfe76559 | 5810 | return io_fadvise_prep(req, sqe); |
c1ca757b | 5811 | case IORING_OP_MADVISE: |
bfe76559 | 5812 | return io_madvise_prep(req, sqe); |
cebdb986 | 5813 | case IORING_OP_OPENAT2: |
bfe76559 | 5814 | return io_openat2_prep(req, sqe); |
3e4827b0 | 5815 | case IORING_OP_EPOLL_CTL: |
bfe76559 | 5816 | return io_epoll_ctl_prep(req, sqe); |
7d67af2c | 5817 | case IORING_OP_SPLICE: |
bfe76559 | 5818 | return io_splice_prep(req, sqe); |
ddf0322d | 5819 | case IORING_OP_PROVIDE_BUFFERS: |
bfe76559 | 5820 | return io_provide_buffers_prep(req, sqe); |
067524e9 | 5821 | case IORING_OP_REMOVE_BUFFERS: |
bfe76559 | 5822 | return io_remove_buffers_prep(req, sqe); |
f2a8d5c7 | 5823 | case IORING_OP_TEE: |
bfe76559 | 5824 | return io_tee_prep(req, sqe); |
36f4fa68 JA |
5825 | case IORING_OP_SHUTDOWN: |
5826 | return io_shutdown_prep(req, sqe); | |
f67676d1 JA |
5827 | } |
5828 | ||
bfe76559 PB |
5829 | printk_once(KERN_WARNING "io_uring: unhandled opcode %d\n", |
5830 | req->opcode); | |
5831 | return-EINVAL; | |
5832 | } | |
5833 | ||
5834 | static int io_req_defer_prep(struct io_kiocb *req, | |
5835 | const struct io_uring_sqe *sqe) | |
5836 | { | |
bfe76559 PB |
5837 | if (!sqe) |
5838 | return 0; | |
5839 | if (io_alloc_async_data(req)) | |
5840 | return -EAGAIN; | |
bfe76559 | 5841 | return io_req_prep(req, sqe); |
f67676d1 JA |
5842 | } |
5843 | ||
9cf7c104 PB |
5844 | static u32 io_get_sequence(struct io_kiocb *req) |
5845 | { | |
5846 | struct io_kiocb *pos; | |
5847 | struct io_ring_ctx *ctx = req->ctx; | |
5848 | u32 total_submitted, nr_reqs = 1; | |
5849 | ||
5850 | if (req->flags & REQ_F_LINK_HEAD) | |
5851 | list_for_each_entry(pos, &req->link_list, link_list) | |
5852 | nr_reqs++; | |
5853 | ||
5854 | total_submitted = ctx->cached_sq_head - ctx->cached_sq_dropped; | |
5855 | return total_submitted - nr_reqs; | |
5856 | } | |
5857 | ||
3529d8c2 | 5858 | static int io_req_defer(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
de0617e4 | 5859 | { |
a197f664 | 5860 | struct io_ring_ctx *ctx = req->ctx; |
27dc8338 | 5861 | struct io_defer_entry *de; |
f67676d1 | 5862 | int ret; |
9cf7c104 | 5863 | u32 seq; |
de0617e4 | 5864 | |
9d858b21 | 5865 | /* Still need defer if there is pending req in defer list. */ |
9cf7c104 PB |
5866 | if (likely(list_empty_careful(&ctx->defer_list) && |
5867 | !(req->flags & REQ_F_IO_DRAIN))) | |
5868 | return 0; | |
5869 | ||
5870 | seq = io_get_sequence(req); | |
5871 | /* Still a chance to pass the sequence check */ | |
5872 | if (!req_need_defer(req, seq) && list_empty_careful(&ctx->defer_list)) | |
de0617e4 JA |
5873 | return 0; |
5874 | ||
e8c2bc1f | 5875 | if (!req->async_data) { |
650b5481 | 5876 | ret = io_req_defer_prep(req, sqe); |
327d6d96 | 5877 | if (ret) |
650b5481 PB |
5878 | return ret; |
5879 | } | |
cbdcb435 | 5880 | io_prep_async_link(req); |
27dc8338 PB |
5881 | de = kmalloc(sizeof(*de), GFP_KERNEL); |
5882 | if (!de) | |
5883 | return -ENOMEM; | |
2d28390a | 5884 | |
de0617e4 | 5885 | spin_lock_irq(&ctx->completion_lock); |
9cf7c104 | 5886 | if (!req_need_defer(req, seq) && list_empty(&ctx->defer_list)) { |
de0617e4 | 5887 | spin_unlock_irq(&ctx->completion_lock); |
27dc8338 | 5888 | kfree(de); |
ae34817b PB |
5889 | io_queue_async_work(req); |
5890 | return -EIOCBQUEUED; | |
de0617e4 JA |
5891 | } |
5892 | ||
915967f6 | 5893 | trace_io_uring_defer(ctx, req, req->user_data); |
27dc8338 | 5894 | de->req = req; |
9cf7c104 | 5895 | de->seq = seq; |
27dc8338 | 5896 | list_add_tail(&de->list, &ctx->defer_list); |
de0617e4 JA |
5897 | spin_unlock_irq(&ctx->completion_lock); |
5898 | return -EIOCBQUEUED; | |
5899 | } | |
5900 | ||
f573d384 | 5901 | static void io_req_drop_files(struct io_kiocb *req) |
99bc4c38 | 5902 | { |
f573d384 JA |
5903 | struct io_ring_ctx *ctx = req->ctx; |
5904 | unsigned long flags; | |
5905 | ||
5906 | spin_lock_irqsave(&ctx->inflight_lock, flags); | |
5907 | list_del(&req->inflight_entry); | |
5908 | if (waitqueue_active(&ctx->inflight_wait)) | |
5909 | wake_up(&ctx->inflight_wait); | |
5910 | spin_unlock_irqrestore(&ctx->inflight_lock, flags); | |
5911 | req->flags &= ~REQ_F_INFLIGHT; | |
98447d65 JA |
5912 | put_files_struct(req->work.identity->files); |
5913 | put_nsproxy(req->work.identity->nsproxy); | |
dfead8a8 | 5914 | req->work.flags &= ~IO_WQ_WORK_FILES; |
f573d384 | 5915 | } |
99bc4c38 | 5916 | |
3ca405eb | 5917 | static void __io_clean_op(struct io_kiocb *req) |
99bc4c38 | 5918 | { |
0e1b6fe3 PB |
5919 | if (req->flags & REQ_F_BUFFER_SELECTED) { |
5920 | switch (req->opcode) { | |
5921 | case IORING_OP_READV: | |
5922 | case IORING_OP_READ_FIXED: | |
5923 | case IORING_OP_READ: | |
bcda7baa | 5924 | kfree((void *)(unsigned long)req->rw.addr); |
0e1b6fe3 PB |
5925 | break; |
5926 | case IORING_OP_RECVMSG: | |
5927 | case IORING_OP_RECV: | |
bcda7baa | 5928 | kfree(req->sr_msg.kbuf); |
0e1b6fe3 PB |
5929 | break; |
5930 | } | |
5931 | req->flags &= ~REQ_F_BUFFER_SELECTED; | |
99bc4c38 PB |
5932 | } |
5933 | ||
0e1b6fe3 PB |
5934 | if (req->flags & REQ_F_NEED_CLEANUP) { |
5935 | switch (req->opcode) { | |
5936 | case IORING_OP_READV: | |
5937 | case IORING_OP_READ_FIXED: | |
5938 | case IORING_OP_READ: | |
5939 | case IORING_OP_WRITEV: | |
5940 | case IORING_OP_WRITE_FIXED: | |
e8c2bc1f JA |
5941 | case IORING_OP_WRITE: { |
5942 | struct io_async_rw *io = req->async_data; | |
5943 | if (io->free_iovec) | |
5944 | kfree(io->free_iovec); | |
0e1b6fe3 | 5945 | break; |
e8c2bc1f | 5946 | } |
0e1b6fe3 | 5947 | case IORING_OP_RECVMSG: |
e8c2bc1f JA |
5948 | case IORING_OP_SENDMSG: { |
5949 | struct io_async_msghdr *io = req->async_data; | |
5950 | if (io->iov != io->fast_iov) | |
5951 | kfree(io->iov); | |
0e1b6fe3 | 5952 | break; |
e8c2bc1f | 5953 | } |
0e1b6fe3 PB |
5954 | case IORING_OP_SPLICE: |
5955 | case IORING_OP_TEE: | |
5956 | io_put_file(req, req->splice.file_in, | |
5957 | (req->splice.flags & SPLICE_F_FD_IN_FIXED)); | |
5958 | break; | |
f3cd4850 JA |
5959 | case IORING_OP_OPENAT: |
5960 | case IORING_OP_OPENAT2: | |
5961 | if (req->open.filename) | |
5962 | putname(req->open.filename); | |
5963 | break; | |
0e1b6fe3 PB |
5964 | } |
5965 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
99bc4c38 | 5966 | } |
bb175342 | 5967 | |
f573d384 JA |
5968 | if (req->flags & REQ_F_INFLIGHT) |
5969 | io_req_drop_files(req); | |
99bc4c38 PB |
5970 | } |
5971 | ||
c1379e24 PB |
5972 | static int io_issue_sqe(struct io_kiocb *req, bool force_nonblock, |
5973 | struct io_comp_state *cs) | |
2b188cc1 | 5974 | { |
a197f664 | 5975 | struct io_ring_ctx *ctx = req->ctx; |
d625c6ee | 5976 | int ret; |
2b188cc1 | 5977 | |
d625c6ee | 5978 | switch (req->opcode) { |
2b188cc1 | 5979 | case IORING_OP_NOP: |
229a7b63 | 5980 | ret = io_nop(req, cs); |
2b188cc1 JA |
5981 | break; |
5982 | case IORING_OP_READV: | |
edafccee | 5983 | case IORING_OP_READ_FIXED: |
3a6820f2 | 5984 | case IORING_OP_READ: |
a1d7c393 | 5985 | ret = io_read(req, force_nonblock, cs); |
edafccee | 5986 | break; |
3529d8c2 | 5987 | case IORING_OP_WRITEV: |
edafccee | 5988 | case IORING_OP_WRITE_FIXED: |
3a6820f2 | 5989 | case IORING_OP_WRITE: |
a1d7c393 | 5990 | ret = io_write(req, force_nonblock, cs); |
2b188cc1 | 5991 | break; |
c992fe29 | 5992 | case IORING_OP_FSYNC: |
014db007 | 5993 | ret = io_fsync(req, force_nonblock); |
c992fe29 | 5994 | break; |
221c5eb2 | 5995 | case IORING_OP_POLL_ADD: |
014db007 | 5996 | ret = io_poll_add(req); |
221c5eb2 JA |
5997 | break; |
5998 | case IORING_OP_POLL_REMOVE: | |
fc4df999 | 5999 | ret = io_poll_remove(req); |
221c5eb2 | 6000 | break; |
5d17b4a4 | 6001 | case IORING_OP_SYNC_FILE_RANGE: |
014db007 | 6002 | ret = io_sync_file_range(req, force_nonblock); |
5d17b4a4 | 6003 | break; |
0fa03c62 | 6004 | case IORING_OP_SENDMSG: |
062d04d7 PB |
6005 | ret = io_sendmsg(req, force_nonblock, cs); |
6006 | break; | |
fddaface | 6007 | case IORING_OP_SEND: |
062d04d7 | 6008 | ret = io_send(req, force_nonblock, cs); |
0fa03c62 | 6009 | break; |
aa1fa28f | 6010 | case IORING_OP_RECVMSG: |
062d04d7 PB |
6011 | ret = io_recvmsg(req, force_nonblock, cs); |
6012 | break; | |
fddaface | 6013 | case IORING_OP_RECV: |
062d04d7 | 6014 | ret = io_recv(req, force_nonblock, cs); |
aa1fa28f | 6015 | break; |
5262f567 | 6016 | case IORING_OP_TIMEOUT: |
fc4df999 | 6017 | ret = io_timeout(req); |
5262f567 | 6018 | break; |
11365043 | 6019 | case IORING_OP_TIMEOUT_REMOVE: |
fc4df999 | 6020 | ret = io_timeout_remove(req); |
11365043 | 6021 | break; |
17f2fe35 | 6022 | case IORING_OP_ACCEPT: |
229a7b63 | 6023 | ret = io_accept(req, force_nonblock, cs); |
17f2fe35 | 6024 | break; |
f8e85cf2 | 6025 | case IORING_OP_CONNECT: |
229a7b63 | 6026 | ret = io_connect(req, force_nonblock, cs); |
f8e85cf2 | 6027 | break; |
62755e35 | 6028 | case IORING_OP_ASYNC_CANCEL: |
014db007 | 6029 | ret = io_async_cancel(req); |
62755e35 | 6030 | break; |
d63d1b5e | 6031 | case IORING_OP_FALLOCATE: |
014db007 | 6032 | ret = io_fallocate(req, force_nonblock); |
d63d1b5e | 6033 | break; |
15b71abe | 6034 | case IORING_OP_OPENAT: |
014db007 | 6035 | ret = io_openat(req, force_nonblock); |
15b71abe | 6036 | break; |
b5dba59e | 6037 | case IORING_OP_CLOSE: |
229a7b63 | 6038 | ret = io_close(req, force_nonblock, cs); |
b5dba59e | 6039 | break; |
05f3fb3c | 6040 | case IORING_OP_FILES_UPDATE: |
229a7b63 | 6041 | ret = io_files_update(req, force_nonblock, cs); |
05f3fb3c | 6042 | break; |
eddc7ef5 | 6043 | case IORING_OP_STATX: |
014db007 | 6044 | ret = io_statx(req, force_nonblock); |
eddc7ef5 | 6045 | break; |
4840e418 | 6046 | case IORING_OP_FADVISE: |
014db007 | 6047 | ret = io_fadvise(req, force_nonblock); |
4840e418 | 6048 | break; |
c1ca757b | 6049 | case IORING_OP_MADVISE: |
014db007 | 6050 | ret = io_madvise(req, force_nonblock); |
c1ca757b | 6051 | break; |
cebdb986 | 6052 | case IORING_OP_OPENAT2: |
014db007 | 6053 | ret = io_openat2(req, force_nonblock); |
cebdb986 | 6054 | break; |
3e4827b0 | 6055 | case IORING_OP_EPOLL_CTL: |
229a7b63 | 6056 | ret = io_epoll_ctl(req, force_nonblock, cs); |
3e4827b0 | 6057 | break; |
7d67af2c | 6058 | case IORING_OP_SPLICE: |
014db007 | 6059 | ret = io_splice(req, force_nonblock); |
7d67af2c | 6060 | break; |
ddf0322d | 6061 | case IORING_OP_PROVIDE_BUFFERS: |
229a7b63 | 6062 | ret = io_provide_buffers(req, force_nonblock, cs); |
ddf0322d | 6063 | break; |
067524e9 | 6064 | case IORING_OP_REMOVE_BUFFERS: |
229a7b63 | 6065 | ret = io_remove_buffers(req, force_nonblock, cs); |
3e4827b0 | 6066 | break; |
f2a8d5c7 | 6067 | case IORING_OP_TEE: |
f2a8d5c7 PB |
6068 | ret = io_tee(req, force_nonblock); |
6069 | break; | |
36f4fa68 JA |
6070 | case IORING_OP_SHUTDOWN: |
6071 | ret = io_shutdown(req, force_nonblock); | |
6072 | break; | |
2b188cc1 JA |
6073 | default: |
6074 | ret = -EINVAL; | |
6075 | break; | |
6076 | } | |
6077 | ||
def596e9 JA |
6078 | if (ret) |
6079 | return ret; | |
6080 | ||
b532576e JA |
6081 | /* If the op doesn't have a file, we're not polling for it */ |
6082 | if ((ctx->flags & IORING_SETUP_IOPOLL) && req->file) { | |
11ba820b JA |
6083 | const bool in_async = io_wq_current_is_worker(); |
6084 | ||
11ba820b JA |
6085 | /* workqueue context doesn't hold uring_lock, grab it now */ |
6086 | if (in_async) | |
6087 | mutex_lock(&ctx->uring_lock); | |
6088 | ||
def596e9 | 6089 | io_iopoll_req_issued(req); |
11ba820b JA |
6090 | |
6091 | if (in_async) | |
6092 | mutex_unlock(&ctx->uring_lock); | |
def596e9 JA |
6093 | } |
6094 | ||
6095 | return 0; | |
2b188cc1 JA |
6096 | } |
6097 | ||
f4db7182 | 6098 | static struct io_wq_work *io_wq_submit_work(struct io_wq_work *work) |
2b188cc1 JA |
6099 | { |
6100 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); | |
6df1db6b | 6101 | struct io_kiocb *timeout; |
561fb04a | 6102 | int ret = 0; |
2b188cc1 | 6103 | |
6df1db6b PB |
6104 | timeout = io_prep_linked_timeout(req); |
6105 | if (timeout) | |
6106 | io_queue_linked_timeout(timeout); | |
d4c81f38 | 6107 | |
0c9d5ccd JA |
6108 | /* if NO_CANCEL is set, we must still run the work */ |
6109 | if ((work->flags & (IO_WQ_WORK_CANCEL|IO_WQ_WORK_NO_CANCEL)) == | |
6110 | IO_WQ_WORK_CANCEL) { | |
561fb04a | 6111 | ret = -ECANCELED; |
0c9d5ccd | 6112 | } |
31b51510 | 6113 | |
561fb04a | 6114 | if (!ret) { |
561fb04a | 6115 | do { |
c1379e24 | 6116 | ret = io_issue_sqe(req, false, NULL); |
561fb04a JA |
6117 | /* |
6118 | * We can get EAGAIN for polled IO even though we're | |
6119 | * forcing a sync submission from here, since we can't | |
6120 | * wait for request slots on the block side. | |
6121 | */ | |
6122 | if (ret != -EAGAIN) | |
6123 | break; | |
6124 | cond_resched(); | |
6125 | } while (1); | |
6126 | } | |
31b51510 | 6127 | |
561fb04a | 6128 | if (ret) { |
4e88d6e7 | 6129 | req_set_fail_links(req); |
e1e16097 | 6130 | io_req_complete(req, ret); |
edafccee | 6131 | } |
2b188cc1 | 6132 | |
f4db7182 | 6133 | return io_steal_work(req); |
2b188cc1 JA |
6134 | } |
6135 | ||
65e19f54 JA |
6136 | static inline struct file *io_file_from_index(struct io_ring_ctx *ctx, |
6137 | int index) | |
6138 | { | |
6139 | struct fixed_file_table *table; | |
6140 | ||
05f3fb3c | 6141 | table = &ctx->file_data->table[index >> IORING_FILE_TABLE_SHIFT]; |
84695089 | 6142 | return table->files[index & IORING_FILE_TABLE_MASK]; |
65e19f54 JA |
6143 | } |
6144 | ||
8371adf5 PB |
6145 | static struct file *io_file_get(struct io_submit_state *state, |
6146 | struct io_kiocb *req, int fd, bool fixed) | |
09bb8394 | 6147 | { |
a197f664 | 6148 | struct io_ring_ctx *ctx = req->ctx; |
8da11c19 | 6149 | struct file *file; |
09bb8394 | 6150 | |
8da11c19 | 6151 | if (fixed) { |
479f517b | 6152 | if (unlikely((unsigned int)fd >= ctx->nr_user_files)) |
8371adf5 | 6153 | return NULL; |
b7620121 | 6154 | fd = array_index_nospec(fd, ctx->nr_user_files); |
8da11c19 | 6155 | file = io_file_from_index(ctx, fd); |
fd2206e4 | 6156 | if (file) { |
b2e96852 | 6157 | req->fixed_file_refs = &ctx->file_data->node->refs; |
fd2206e4 JA |
6158 | percpu_ref_get(req->fixed_file_refs); |
6159 | } | |
09bb8394 | 6160 | } else { |
c826bd7a | 6161 | trace_io_uring_file_get(ctx, fd); |
8da11c19 | 6162 | file = __io_file_get(state, fd); |
09bb8394 JA |
6163 | } |
6164 | ||
8371adf5 | 6165 | return file; |
09bb8394 JA |
6166 | } |
6167 | ||
8da11c19 | 6168 | static int io_req_set_file(struct io_submit_state *state, struct io_kiocb *req, |
63ff8223 | 6169 | int fd) |
8da11c19 | 6170 | { |
8da11c19 PB |
6171 | bool fixed; |
6172 | ||
63ff8223 | 6173 | fixed = (req->flags & REQ_F_FIXED_FILE) != 0; |
0cdaf760 | 6174 | if (unlikely(!fixed && io_async_submit(req->ctx))) |
8da11c19 PB |
6175 | return -EBADF; |
6176 | ||
8371adf5 PB |
6177 | req->file = io_file_get(state, req, fd, fixed); |
6178 | if (req->file || io_op_defs[req->opcode].needs_file_no_error) | |
f86cd20c | 6179 | return 0; |
8371adf5 | 6180 | return -EBADF; |
f56040b8 PB |
6181 | } |
6182 | ||
2665abfd | 6183 | static enum hrtimer_restart io_link_timeout_fn(struct hrtimer *timer) |
2b188cc1 | 6184 | { |
ad8a48ac JA |
6185 | struct io_timeout_data *data = container_of(timer, |
6186 | struct io_timeout_data, timer); | |
6187 | struct io_kiocb *req = data->req; | |
2665abfd JA |
6188 | struct io_ring_ctx *ctx = req->ctx; |
6189 | struct io_kiocb *prev = NULL; | |
6190 | unsigned long flags; | |
2665abfd JA |
6191 | |
6192 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
6193 | ||
6194 | /* | |
6195 | * We don't expect the list to be empty, that will only happen if we | |
6196 | * race with the completion of the linked work. | |
6197 | */ | |
4493233e PB |
6198 | if (!list_empty(&req->link_list)) { |
6199 | prev = list_entry(req->link_list.prev, struct io_kiocb, | |
6200 | link_list); | |
900fad45 | 6201 | if (refcount_inc_not_zero(&prev->refs)) |
4493233e | 6202 | list_del_init(&req->link_list); |
900fad45 | 6203 | else |
76a46e06 | 6204 | prev = NULL; |
2665abfd JA |
6205 | } |
6206 | ||
6207 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
6208 | ||
6209 | if (prev) { | |
4e88d6e7 | 6210 | req_set_fail_links(prev); |
014db007 | 6211 | io_async_find_and_cancel(ctx, req, prev->user_data, -ETIME); |
76a46e06 | 6212 | io_put_req(prev); |
47f46768 | 6213 | } else { |
e1e16097 | 6214 | io_req_complete(req, -ETIME); |
2665abfd | 6215 | } |
2665abfd JA |
6216 | return HRTIMER_NORESTART; |
6217 | } | |
6218 | ||
7271ef3a | 6219 | static void __io_queue_linked_timeout(struct io_kiocb *req) |
2665abfd | 6220 | { |
76a46e06 JA |
6221 | /* |
6222 | * If the list is now empty, then our linked request finished before | |
6223 | * we got a chance to setup the timer | |
6224 | */ | |
4493233e | 6225 | if (!list_empty(&req->link_list)) { |
e8c2bc1f | 6226 | struct io_timeout_data *data = req->async_data; |
94ae5e77 | 6227 | |
ad8a48ac JA |
6228 | data->timer.function = io_link_timeout_fn; |
6229 | hrtimer_start(&data->timer, timespec64_to_ktime(data->ts), | |
6230 | data->mode); | |
2665abfd | 6231 | } |
7271ef3a JA |
6232 | } |
6233 | ||
6234 | static void io_queue_linked_timeout(struct io_kiocb *req) | |
6235 | { | |
6236 | struct io_ring_ctx *ctx = req->ctx; | |
6237 | ||
6238 | spin_lock_irq(&ctx->completion_lock); | |
6239 | __io_queue_linked_timeout(req); | |
76a46e06 | 6240 | spin_unlock_irq(&ctx->completion_lock); |
2665abfd | 6241 | |
2665abfd | 6242 | /* drop submission reference */ |
76a46e06 JA |
6243 | io_put_req(req); |
6244 | } | |
2665abfd | 6245 | |
ad8a48ac | 6246 | static struct io_kiocb *io_prep_linked_timeout(struct io_kiocb *req) |
2665abfd JA |
6247 | { |
6248 | struct io_kiocb *nxt; | |
6249 | ||
dea3b49c | 6250 | if (!(req->flags & REQ_F_LINK_HEAD)) |
2665abfd | 6251 | return NULL; |
6df1db6b | 6252 | if (req->flags & REQ_F_LINK_TIMEOUT) |
d7718a9d | 6253 | return NULL; |
2665abfd | 6254 | |
4493233e PB |
6255 | nxt = list_first_entry_or_null(&req->link_list, struct io_kiocb, |
6256 | link_list); | |
d625c6ee | 6257 | if (!nxt || nxt->opcode != IORING_OP_LINK_TIMEOUT) |
76a46e06 | 6258 | return NULL; |
2665abfd | 6259 | |
900fad45 | 6260 | nxt->flags |= REQ_F_LTIMEOUT_ACTIVE; |
76a46e06 | 6261 | req->flags |= REQ_F_LINK_TIMEOUT; |
76a46e06 | 6262 | return nxt; |
2665abfd JA |
6263 | } |
6264 | ||
c1379e24 | 6265 | static void __io_queue_sqe(struct io_kiocb *req, struct io_comp_state *cs) |
2b188cc1 | 6266 | { |
4a0a7a18 | 6267 | struct io_kiocb *linked_timeout; |
193155c8 | 6268 | const struct cred *old_creds = NULL; |
e0c5c576 | 6269 | int ret; |
2b188cc1 | 6270 | |
4a0a7a18 JA |
6271 | again: |
6272 | linked_timeout = io_prep_linked_timeout(req); | |
6273 | ||
2e5aa6cb PB |
6274 | if ((req->flags & REQ_F_WORK_INITIALIZED) && |
6275 | (req->work.flags & IO_WQ_WORK_CREDS) && | |
98447d65 | 6276 | req->work.identity->creds != current_cred()) { |
193155c8 JA |
6277 | if (old_creds) |
6278 | revert_creds(old_creds); | |
98447d65 | 6279 | if (old_creds == req->work.identity->creds) |
193155c8 JA |
6280 | old_creds = NULL; /* restored original creds */ |
6281 | else | |
98447d65 | 6282 | old_creds = override_creds(req->work.identity->creds); |
193155c8 JA |
6283 | } |
6284 | ||
c1379e24 | 6285 | ret = io_issue_sqe(req, true, cs); |
491381ce JA |
6286 | |
6287 | /* | |
6288 | * We async punt it if the file wasn't marked NOWAIT, or if the file | |
6289 | * doesn't support non-blocking read/write attempts | |
6290 | */ | |
24c74678 | 6291 | if (ret == -EAGAIN && !(req->flags & REQ_F_NOWAIT)) { |
f063c547 | 6292 | if (!io_arm_poll_handler(req)) { |
f063c547 PB |
6293 | /* |
6294 | * Queued up for async execution, worker will release | |
6295 | * submit reference when the iocb is actually submitted. | |
6296 | */ | |
6297 | io_queue_async_work(req); | |
2b188cc1 | 6298 | } |
bbad27b2 | 6299 | |
f063c547 PB |
6300 | if (linked_timeout) |
6301 | io_queue_linked_timeout(linked_timeout); | |
0d63c148 PB |
6302 | } else if (likely(!ret)) { |
6303 | /* drop submission reference */ | |
6304 | req = io_put_req_find_next(req); | |
6305 | if (linked_timeout) | |
6306 | io_queue_linked_timeout(linked_timeout); | |
e65ef56d | 6307 | |
0d63c148 PB |
6308 | if (req) { |
6309 | if (!(req->flags & REQ_F_FORCE_ASYNC)) | |
6310 | goto again; | |
6311 | io_queue_async_work(req); | |
6312 | } | |
6313 | } else { | |
652532ad PB |
6314 | /* un-prep timeout, so it'll be killed as any other linked */ |
6315 | req->flags &= ~REQ_F_LINK_TIMEOUT; | |
4e88d6e7 | 6316 | req_set_fail_links(req); |
e65ef56d | 6317 | io_put_req(req); |
e1e16097 | 6318 | io_req_complete(req, ret); |
9e645e11 | 6319 | } |
652532ad | 6320 | |
193155c8 JA |
6321 | if (old_creds) |
6322 | revert_creds(old_creds); | |
2b188cc1 JA |
6323 | } |
6324 | ||
f13fad7b JA |
6325 | static void io_queue_sqe(struct io_kiocb *req, const struct io_uring_sqe *sqe, |
6326 | struct io_comp_state *cs) | |
4fe2c963 JL |
6327 | { |
6328 | int ret; | |
6329 | ||
3529d8c2 | 6330 | ret = io_req_defer(req, sqe); |
4fe2c963 JL |
6331 | if (ret) { |
6332 | if (ret != -EIOCBQUEUED) { | |
1118591a | 6333 | fail_req: |
4e88d6e7 | 6334 | req_set_fail_links(req); |
e1e16097 JA |
6335 | io_put_req(req); |
6336 | io_req_complete(req, ret); | |
4fe2c963 | 6337 | } |
2550878f | 6338 | } else if (req->flags & REQ_F_FORCE_ASYNC) { |
e8c2bc1f | 6339 | if (!req->async_data) { |
bd2ab18a | 6340 | ret = io_req_defer_prep(req, sqe); |
327d6d96 | 6341 | if (unlikely(ret)) |
bd2ab18a PB |
6342 | goto fail_req; |
6343 | } | |
ce35a47a JA |
6344 | io_queue_async_work(req); |
6345 | } else { | |
c1379e24 PB |
6346 | if (sqe) { |
6347 | ret = io_req_prep(req, sqe); | |
6348 | if (unlikely(ret)) | |
6349 | goto fail_req; | |
6350 | } | |
6351 | __io_queue_sqe(req, cs); | |
ce35a47a | 6352 | } |
4fe2c963 JL |
6353 | } |
6354 | ||
f13fad7b JA |
6355 | static inline void io_queue_link_head(struct io_kiocb *req, |
6356 | struct io_comp_state *cs) | |
4fe2c963 | 6357 | { |
94ae5e77 | 6358 | if (unlikely(req->flags & REQ_F_FAIL_LINK)) { |
e1e16097 JA |
6359 | io_put_req(req); |
6360 | io_req_complete(req, -ECANCELED); | |
1b4a51b6 | 6361 | } else |
f13fad7b | 6362 | io_queue_sqe(req, NULL, cs); |
4fe2c963 JL |
6363 | } |
6364 | ||
1d4240cc | 6365 | static int io_submit_sqe(struct io_kiocb *req, const struct io_uring_sqe *sqe, |
f13fad7b | 6366 | struct io_kiocb **link, struct io_comp_state *cs) |
9e645e11 | 6367 | { |
a197f664 | 6368 | struct io_ring_ctx *ctx = req->ctx; |
ef4ff581 | 6369 | int ret; |
9e645e11 | 6370 | |
9e645e11 JA |
6371 | /* |
6372 | * If we already have a head request, queue this one for async | |
6373 | * submittal once the head completes. If we don't have a head but | |
6374 | * IOSQE_IO_LINK is set in the sqe, start a new head. This one will be | |
6375 | * submitted sync once the chain is complete. If none of those | |
6376 | * conditions are true (normal request), then just queue it. | |
6377 | */ | |
6378 | if (*link) { | |
9d76377f | 6379 | struct io_kiocb *head = *link; |
4e88d6e7 | 6380 | |
8cdf2193 PB |
6381 | /* |
6382 | * Taking sequential execution of a link, draining both sides | |
6383 | * of the link also fullfils IOSQE_IO_DRAIN semantics for all | |
6384 | * requests in the link. So, it drains the head and the | |
6385 | * next after the link request. The last one is done via | |
6386 | * drain_next flag to persist the effect across calls. | |
6387 | */ | |
ef4ff581 | 6388 | if (req->flags & REQ_F_IO_DRAIN) { |
711be031 PB |
6389 | head->flags |= REQ_F_IO_DRAIN; |
6390 | ctx->drain_next = 1; | |
6391 | } | |
3529d8c2 | 6392 | ret = io_req_defer_prep(req, sqe); |
327d6d96 | 6393 | if (unlikely(ret)) { |
4e88d6e7 | 6394 | /* fail even hard links since we don't submit */ |
9d76377f | 6395 | head->flags |= REQ_F_FAIL_LINK; |
1d4240cc | 6396 | return ret; |
2d28390a | 6397 | } |
9d76377f PB |
6398 | trace_io_uring_link(ctx, req, head); |
6399 | list_add_tail(&req->link_list, &head->link_list); | |
32fe525b PB |
6400 | |
6401 | /* last request of a link, enqueue the link */ | |
ef4ff581 | 6402 | if (!(req->flags & (REQ_F_LINK | REQ_F_HARDLINK))) { |
f13fad7b | 6403 | io_queue_link_head(head, cs); |
32fe525b PB |
6404 | *link = NULL; |
6405 | } | |
9e645e11 | 6406 | } else { |
711be031 PB |
6407 | if (unlikely(ctx->drain_next)) { |
6408 | req->flags |= REQ_F_IO_DRAIN; | |
ef4ff581 | 6409 | ctx->drain_next = 0; |
711be031 | 6410 | } |
ef4ff581 | 6411 | if (req->flags & (REQ_F_LINK | REQ_F_HARDLINK)) { |
dea3b49c | 6412 | req->flags |= REQ_F_LINK_HEAD; |
711be031 | 6413 | INIT_LIST_HEAD(&req->link_list); |
f1d96a8f | 6414 | |
711be031 | 6415 | ret = io_req_defer_prep(req, sqe); |
327d6d96 | 6416 | if (unlikely(ret)) |
711be031 PB |
6417 | req->flags |= REQ_F_FAIL_LINK; |
6418 | *link = req; | |
6419 | } else { | |
f13fad7b | 6420 | io_queue_sqe(req, sqe, cs); |
711be031 | 6421 | } |
9e645e11 | 6422 | } |
2e6e1fde | 6423 | |
1d4240cc | 6424 | return 0; |
9e645e11 JA |
6425 | } |
6426 | ||
9a56a232 JA |
6427 | /* |
6428 | * Batched submission is done, ensure local IO is flushed out. | |
6429 | */ | |
6430 | static void io_submit_state_end(struct io_submit_state *state) | |
6431 | { | |
f13fad7b JA |
6432 | if (!list_empty(&state->comp.list)) |
6433 | io_submit_flush_completions(&state->comp); | |
9a56a232 | 6434 | blk_finish_plug(&state->plug); |
9f13c35b | 6435 | io_state_file_put(state); |
2579f913 | 6436 | if (state->free_reqs) |
6c8a3134 | 6437 | kmem_cache_free_bulk(req_cachep, state->free_reqs, state->reqs); |
9a56a232 JA |
6438 | } |
6439 | ||
6440 | /* | |
6441 | * Start submission side cache. | |
6442 | */ | |
6443 | static void io_submit_state_start(struct io_submit_state *state, | |
013538bd | 6444 | struct io_ring_ctx *ctx, unsigned int max_ios) |
9a56a232 JA |
6445 | { |
6446 | blk_start_plug(&state->plug); | |
013538bd JA |
6447 | state->comp.nr = 0; |
6448 | INIT_LIST_HEAD(&state->comp.list); | |
6449 | state->comp.ctx = ctx; | |
2579f913 | 6450 | state->free_reqs = 0; |
9a56a232 JA |
6451 | state->file = NULL; |
6452 | state->ios_left = max_ios; | |
6453 | } | |
6454 | ||
2b188cc1 JA |
6455 | static void io_commit_sqring(struct io_ring_ctx *ctx) |
6456 | { | |
75b28aff | 6457 | struct io_rings *rings = ctx->rings; |
2b188cc1 | 6458 | |
caf582c6 PB |
6459 | /* |
6460 | * Ensure any loads from the SQEs are done at this point, | |
6461 | * since once we write the new head, the application could | |
6462 | * write new data to them. | |
6463 | */ | |
6464 | smp_store_release(&rings->sq.head, ctx->cached_sq_head); | |
2b188cc1 JA |
6465 | } |
6466 | ||
2b188cc1 | 6467 | /* |
3529d8c2 | 6468 | * Fetch an sqe, if one is available. Note that sqe_ptr will point to memory |
2b188cc1 JA |
6469 | * that is mapped by userspace. This means that care needs to be taken to |
6470 | * ensure that reads are stable, as we cannot rely on userspace always | |
6471 | * being a good citizen. If members of the sqe are validated and then later | |
6472 | * used, it's important that those reads are done through READ_ONCE() to | |
6473 | * prevent a re-load down the line. | |
6474 | */ | |
709b302f | 6475 | static const struct io_uring_sqe *io_get_sqe(struct io_ring_ctx *ctx) |
2b188cc1 | 6476 | { |
75b28aff | 6477 | u32 *sq_array = ctx->sq_array; |
2b188cc1 JA |
6478 | unsigned head; |
6479 | ||
6480 | /* | |
6481 | * The cached sq head (or cq tail) serves two purposes: | |
6482 | * | |
6483 | * 1) allows us to batch the cost of updating the user visible | |
6484 | * head updates. | |
6485 | * 2) allows the kernel side to track the head on its own, even | |
6486 | * though the application is the one updating it. | |
6487 | */ | |
ee7d46d9 | 6488 | head = READ_ONCE(sq_array[ctx->cached_sq_head & ctx->sq_mask]); |
709b302f PB |
6489 | if (likely(head < ctx->sq_entries)) |
6490 | return &ctx->sq_sqes[head]; | |
2b188cc1 JA |
6491 | |
6492 | /* drop invalid entries */ | |
498ccd9e | 6493 | ctx->cached_sq_dropped++; |
ee7d46d9 | 6494 | WRITE_ONCE(ctx->rings->sq_dropped, ctx->cached_sq_dropped); |
709b302f PB |
6495 | return NULL; |
6496 | } | |
6497 | ||
6498 | static inline void io_consume_sqe(struct io_ring_ctx *ctx) | |
6499 | { | |
6500 | ctx->cached_sq_head++; | |
2b188cc1 JA |
6501 | } |
6502 | ||
21b55dbc SG |
6503 | /* |
6504 | * Check SQE restrictions (opcode and flags). | |
6505 | * | |
6506 | * Returns 'true' if SQE is allowed, 'false' otherwise. | |
6507 | */ | |
6508 | static inline bool io_check_restriction(struct io_ring_ctx *ctx, | |
6509 | struct io_kiocb *req, | |
6510 | unsigned int sqe_flags) | |
6511 | { | |
6512 | if (!ctx->restricted) | |
6513 | return true; | |
6514 | ||
6515 | if (!test_bit(req->opcode, ctx->restrictions.sqe_op)) | |
6516 | return false; | |
6517 | ||
6518 | if ((sqe_flags & ctx->restrictions.sqe_flags_required) != | |
6519 | ctx->restrictions.sqe_flags_required) | |
6520 | return false; | |
6521 | ||
6522 | if (sqe_flags & ~(ctx->restrictions.sqe_flags_allowed | | |
6523 | ctx->restrictions.sqe_flags_required)) | |
6524 | return false; | |
6525 | ||
6526 | return true; | |
6527 | } | |
6528 | ||
ef4ff581 PB |
6529 | #define SQE_VALID_FLAGS (IOSQE_FIXED_FILE|IOSQE_IO_DRAIN|IOSQE_IO_LINK| \ |
6530 | IOSQE_IO_HARDLINK | IOSQE_ASYNC | \ | |
6531 | IOSQE_BUFFER_SELECT) | |
6532 | ||
6533 | static int io_init_req(struct io_ring_ctx *ctx, struct io_kiocb *req, | |
6534 | const struct io_uring_sqe *sqe, | |
0cdaf760 | 6535 | struct io_submit_state *state) |
0553b8bd | 6536 | { |
ef4ff581 | 6537 | unsigned int sqe_flags; |
71b547c0 | 6538 | int id, ret; |
ef4ff581 | 6539 | |
0553b8bd PB |
6540 | req->opcode = READ_ONCE(sqe->opcode); |
6541 | req->user_data = READ_ONCE(sqe->user_data); | |
e8c2bc1f | 6542 | req->async_data = NULL; |
0553b8bd PB |
6543 | req->file = NULL; |
6544 | req->ctx = ctx; | |
6545 | req->flags = 0; | |
6546 | /* one is dropped after submission, the other at completion */ | |
6547 | refcount_set(&req->refs, 2); | |
4dd2824d | 6548 | req->task = current; |
0553b8bd | 6549 | req->result = 0; |
ef4ff581 PB |
6550 | |
6551 | if (unlikely(req->opcode >= IORING_OP_LAST)) | |
6552 | return -EINVAL; | |
6553 | ||
9d8426a0 JA |
6554 | if (unlikely(io_sq_thread_acquire_mm(ctx, req))) |
6555 | return -EFAULT; | |
ef4ff581 PB |
6556 | |
6557 | sqe_flags = READ_ONCE(sqe->flags); | |
6558 | /* enforce forwards compatibility on users */ | |
6559 | if (unlikely(sqe_flags & ~SQE_VALID_FLAGS)) | |
6560 | return -EINVAL; | |
6561 | ||
21b55dbc SG |
6562 | if (unlikely(!io_check_restriction(ctx, req, sqe_flags))) |
6563 | return -EACCES; | |
6564 | ||
ef4ff581 PB |
6565 | if ((sqe_flags & IOSQE_BUFFER_SELECT) && |
6566 | !io_op_defs[req->opcode].buffer_select) | |
6567 | return -EOPNOTSUPP; | |
6568 | ||
6569 | id = READ_ONCE(sqe->personality); | |
6570 | if (id) { | |
1e6fa521 JA |
6571 | struct io_identity *iod; |
6572 | ||
1e6fa521 JA |
6573 | iod = idr_find(&ctx->personality_idr, id); |
6574 | if (unlikely(!iod)) | |
ef4ff581 | 6575 | return -EINVAL; |
1e6fa521 | 6576 | refcount_inc(&iod->count); |
ec99ca6c PB |
6577 | |
6578 | __io_req_init_async(req); | |
1e6fa521 JA |
6579 | get_cred(iod->creds); |
6580 | req->work.identity = iod; | |
dfead8a8 | 6581 | req->work.flags |= IO_WQ_WORK_CREDS; |
ef4ff581 PB |
6582 | } |
6583 | ||
6584 | /* same numerical values with corresponding REQ_F_*, safe to copy */ | |
c11368a5 | 6585 | req->flags |= sqe_flags; |
ef4ff581 | 6586 | |
63ff8223 JA |
6587 | if (!io_op_defs[req->opcode].needs_file) |
6588 | return 0; | |
6589 | ||
71b547c0 PB |
6590 | ret = io_req_set_file(state, req, READ_ONCE(sqe->fd)); |
6591 | state->ios_left--; | |
6592 | return ret; | |
0553b8bd PB |
6593 | } |
6594 | ||
0f212204 | 6595 | static int io_submit_sqes(struct io_ring_ctx *ctx, unsigned int nr) |
6c271ce2 | 6596 | { |
ac8691c4 | 6597 | struct io_submit_state state; |
9e645e11 | 6598 | struct io_kiocb *link = NULL; |
9e645e11 | 6599 | int i, submitted = 0; |
6c271ce2 | 6600 | |
c4a2ed72 | 6601 | /* if we have a backlog and couldn't flush it all, return BUSY */ |
ad3eb2c8 JA |
6602 | if (test_bit(0, &ctx->sq_check_overflow)) { |
6603 | if (!list_empty(&ctx->cq_overflow_list) && | |
e6c8aa9a | 6604 | !io_cqring_overflow_flush(ctx, false, NULL, NULL)) |
ad3eb2c8 JA |
6605 | return -EBUSY; |
6606 | } | |
6c271ce2 | 6607 | |
ee7d46d9 PB |
6608 | /* make sure SQ entry isn't read before tail */ |
6609 | nr = min3(nr, ctx->sq_entries, io_sqring_entries(ctx)); | |
9ef4f124 | 6610 | |
2b85edfc PB |
6611 | if (!percpu_ref_tryget_many(&ctx->refs, nr)) |
6612 | return -EAGAIN; | |
6c271ce2 | 6613 | |
d8a6df10 | 6614 | percpu_counter_add(¤t->io_uring->inflight, nr); |
faf7b51c | 6615 | refcount_add(nr, ¤t->usage); |
6c271ce2 | 6616 | |
013538bd | 6617 | io_submit_state_start(&state, ctx, nr); |
b14cca0c | 6618 | |
6c271ce2 | 6619 | for (i = 0; i < nr; i++) { |
3529d8c2 | 6620 | const struct io_uring_sqe *sqe; |
196be95c | 6621 | struct io_kiocb *req; |
1cb1edb2 | 6622 | int err; |
fb5ccc98 | 6623 | |
b1e50e54 PB |
6624 | sqe = io_get_sqe(ctx); |
6625 | if (unlikely(!sqe)) { | |
6626 | io_consume_sqe(ctx); | |
6627 | break; | |
6628 | } | |
ac8691c4 | 6629 | req = io_alloc_req(ctx, &state); |
196be95c PB |
6630 | if (unlikely(!req)) { |
6631 | if (!submitted) | |
6632 | submitted = -EAGAIN; | |
fb5ccc98 | 6633 | break; |
196be95c | 6634 | } |
709b302f | 6635 | io_consume_sqe(ctx); |
d3656344 JA |
6636 | /* will complete beyond this point, count as submitted */ |
6637 | submitted++; | |
6638 | ||
692d8363 | 6639 | err = io_init_req(ctx, req, sqe, &state); |
ef4ff581 | 6640 | if (unlikely(err)) { |
1cb1edb2 | 6641 | fail_req: |
e1e16097 JA |
6642 | io_put_req(req); |
6643 | io_req_complete(req, err); | |
196be95c PB |
6644 | break; |
6645 | } | |
fb5ccc98 | 6646 | |
354420f7 | 6647 | trace_io_uring_submit_sqe(ctx, req->opcode, req->user_data, |
0cdaf760 | 6648 | true, io_async_submit(ctx)); |
f13fad7b | 6649 | err = io_submit_sqe(req, sqe, &link, &state.comp); |
1d4240cc PB |
6650 | if (err) |
6651 | goto fail_req; | |
6c271ce2 JA |
6652 | } |
6653 | ||
9466f437 PB |
6654 | if (unlikely(submitted != nr)) { |
6655 | int ref_used = (submitted == -EAGAIN) ? 0 : submitted; | |
d8a6df10 JA |
6656 | struct io_uring_task *tctx = current->io_uring; |
6657 | int unused = nr - ref_used; | |
9466f437 | 6658 | |
d8a6df10 JA |
6659 | percpu_ref_put_many(&ctx->refs, unused); |
6660 | percpu_counter_sub(&tctx->inflight, unused); | |
6661 | put_task_struct_many(current, unused); | |
9466f437 | 6662 | } |
9e645e11 | 6663 | if (link) |
f13fad7b | 6664 | io_queue_link_head(link, &state.comp); |
ac8691c4 | 6665 | io_submit_state_end(&state); |
6c271ce2 | 6666 | |
ae9428ca PB |
6667 | /* Commit SQ ring head once we've consumed and submitted all SQEs */ |
6668 | io_commit_sqring(ctx); | |
6669 | ||
6c271ce2 JA |
6670 | return submitted; |
6671 | } | |
6672 | ||
23b3628e XW |
6673 | static inline void io_ring_set_wakeup_flag(struct io_ring_ctx *ctx) |
6674 | { | |
6675 | /* Tell userspace we may need a wakeup call */ | |
6676 | spin_lock_irq(&ctx->completion_lock); | |
6677 | ctx->rings->sq_flags |= IORING_SQ_NEED_WAKEUP; | |
6678 | spin_unlock_irq(&ctx->completion_lock); | |
6679 | } | |
6680 | ||
6681 | static inline void io_ring_clear_wakeup_flag(struct io_ring_ctx *ctx) | |
6682 | { | |
6683 | spin_lock_irq(&ctx->completion_lock); | |
6684 | ctx->rings->sq_flags &= ~IORING_SQ_NEED_WAKEUP; | |
6685 | spin_unlock_irq(&ctx->completion_lock); | |
6686 | } | |
6687 | ||
3f0e64d0 JA |
6688 | static int io_sq_wake_function(struct wait_queue_entry *wqe, unsigned mode, |
6689 | int sync, void *key) | |
6c271ce2 | 6690 | { |
3f0e64d0 JA |
6691 | struct io_ring_ctx *ctx = container_of(wqe, struct io_ring_ctx, sqo_wait_entry); |
6692 | int ret; | |
6693 | ||
6694 | ret = autoremove_wake_function(wqe, mode, sync, key); | |
6695 | if (ret) { | |
6696 | unsigned long flags; | |
6697 | ||
6698 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
6699 | ctx->rings->sq_flags &= ~IORING_SQ_NEED_WAKEUP; | |
6700 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
6701 | } | |
6702 | return ret; | |
6703 | } | |
6704 | ||
c8d1ba58 JA |
6705 | enum sq_ret { |
6706 | SQT_IDLE = 1, | |
6707 | SQT_SPIN = 2, | |
6708 | SQT_DID_WORK = 4, | |
6709 | }; | |
6710 | ||
6711 | static enum sq_ret __io_sq_thread(struct io_ring_ctx *ctx, | |
e95eee2d | 6712 | unsigned long start_jiffies, bool cap_entries) |
6c271ce2 | 6713 | { |
c8d1ba58 | 6714 | unsigned long timeout = start_jiffies + ctx->sq_thread_idle; |
534ca6d6 | 6715 | struct io_sq_data *sqd = ctx->sq_data; |
c8d1ba58 | 6716 | unsigned int to_submit; |
bdcd3eab | 6717 | int ret = 0; |
6c271ce2 | 6718 | |
c8d1ba58 JA |
6719 | again: |
6720 | if (!list_empty(&ctx->iopoll_list)) { | |
6721 | unsigned nr_events = 0; | |
a4c0b3de | 6722 | |
c8d1ba58 JA |
6723 | mutex_lock(&ctx->uring_lock); |
6724 | if (!list_empty(&ctx->iopoll_list) && !need_resched()) | |
6725 | io_do_iopoll(ctx, &nr_events, 0); | |
6726 | mutex_unlock(&ctx->uring_lock); | |
6727 | } | |
6c271ce2 | 6728 | |
c8d1ba58 | 6729 | to_submit = io_sqring_entries(ctx); |
6c271ce2 | 6730 | |
c8d1ba58 JA |
6731 | /* |
6732 | * If submit got -EBUSY, flag us as needing the application | |
6733 | * to enter the kernel to reap and flush events. | |
6734 | */ | |
6735 | if (!to_submit || ret == -EBUSY || need_resched()) { | |
6736 | /* | |
6737 | * Drop cur_mm before scheduling, we can't hold it for | |
6738 | * long periods (or over schedule()). Do this before | |
6739 | * adding ourselves to the waitqueue, as the unuse/drop | |
6740 | * may sleep. | |
6741 | */ | |
6742 | io_sq_thread_drop_mm(); | |
6c271ce2 | 6743 | |
c8d1ba58 JA |
6744 | /* |
6745 | * We're polling. If we're within the defined idle | |
6746 | * period, then let us spin without work before going | |
6747 | * to sleep. The exception is if we got EBUSY doing | |
6748 | * more IO, we should wait for the application to | |
6749 | * reap events and wake us up. | |
6750 | */ | |
6751 | if (!list_empty(&ctx->iopoll_list) || need_resched() || | |
6752 | (!time_after(jiffies, timeout) && ret != -EBUSY && | |
6753 | !percpu_ref_is_dying(&ctx->refs))) | |
6754 | return SQT_SPIN; | |
6c271ce2 | 6755 | |
534ca6d6 | 6756 | prepare_to_wait(&sqd->wait, &ctx->sqo_wait_entry, |
c8d1ba58 | 6757 | TASK_INTERRUPTIBLE); |
6c271ce2 | 6758 | |
c8d1ba58 JA |
6759 | /* |
6760 | * While doing polled IO, before going to sleep, we need | |
6761 | * to check if there are new reqs added to iopoll_list, | |
6762 | * it is because reqs may have been punted to io worker | |
6763 | * and will be added to iopoll_list later, hence check | |
6764 | * the iopoll_list again. | |
6765 | */ | |
6766 | if ((ctx->flags & IORING_SETUP_IOPOLL) && | |
6767 | !list_empty_careful(&ctx->iopoll_list)) { | |
534ca6d6 | 6768 | finish_wait(&sqd->wait, &ctx->sqo_wait_entry); |
c8d1ba58 | 6769 | goto again; |
6c271ce2 JA |
6770 | } |
6771 | ||
fb5ccc98 | 6772 | to_submit = io_sqring_entries(ctx); |
c8d1ba58 JA |
6773 | if (!to_submit || ret == -EBUSY) |
6774 | return SQT_IDLE; | |
6775 | } | |
c1edbf5f | 6776 | |
534ca6d6 | 6777 | finish_wait(&sqd->wait, &ctx->sqo_wait_entry); |
c8d1ba58 | 6778 | io_ring_clear_wakeup_flag(ctx); |
7143b5ac | 6779 | |
e95eee2d JA |
6780 | /* if we're handling multiple rings, cap submit size for fairness */ |
6781 | if (cap_entries && to_submit > 8) | |
6782 | to_submit = 8; | |
6783 | ||
c8d1ba58 JA |
6784 | mutex_lock(&ctx->uring_lock); |
6785 | if (likely(!percpu_ref_is_dying(&ctx->refs))) | |
6786 | ret = io_submit_sqes(ctx, to_submit); | |
6787 | mutex_unlock(&ctx->uring_lock); | |
90554200 JA |
6788 | |
6789 | if (!io_sqring_full(ctx) && wq_has_sleeper(&ctx->sqo_sq_wait)) | |
6790 | wake_up(&ctx->sqo_sq_wait); | |
6791 | ||
c8d1ba58 JA |
6792 | return SQT_DID_WORK; |
6793 | } | |
6c271ce2 | 6794 | |
69fb2131 JA |
6795 | static void io_sqd_init_new(struct io_sq_data *sqd) |
6796 | { | |
6797 | struct io_ring_ctx *ctx; | |
6798 | ||
6799 | while (!list_empty(&sqd->ctx_new_list)) { | |
6800 | ctx = list_first_entry(&sqd->ctx_new_list, struct io_ring_ctx, sqd_list); | |
6801 | init_wait(&ctx->sqo_wait_entry); | |
6802 | ctx->sqo_wait_entry.func = io_sq_wake_function; | |
6803 | list_move_tail(&ctx->sqd_list, &sqd->ctx_list); | |
6804 | complete(&ctx->sq_thread_comp); | |
6805 | } | |
6806 | } | |
6807 | ||
c8d1ba58 JA |
6808 | static int io_sq_thread(void *data) |
6809 | { | |
91d8f519 | 6810 | struct cgroup_subsys_state *cur_css = NULL; |
69fb2131 JA |
6811 | const struct cred *old_cred = NULL; |
6812 | struct io_sq_data *sqd = data; | |
6813 | struct io_ring_ctx *ctx; | |
c8d1ba58 | 6814 | unsigned long start_jiffies; |
6c271ce2 | 6815 | |
69fb2131 JA |
6816 | start_jiffies = jiffies; |
6817 | while (!kthread_should_stop()) { | |
6818 | enum sq_ret ret = 0; | |
e95eee2d | 6819 | bool cap_entries; |
c1edbf5f JA |
6820 | |
6821 | /* | |
69fb2131 JA |
6822 | * Any changes to the sqd lists are synchronized through the |
6823 | * kthread parking. This synchronizes the thread vs users, | |
6824 | * the users are synchronized on the sqd->ctx_lock. | |
c1edbf5f | 6825 | */ |
69fb2131 JA |
6826 | if (kthread_should_park()) |
6827 | kthread_parkme(); | |
7143b5ac | 6828 | |
69fb2131 JA |
6829 | if (unlikely(!list_empty(&sqd->ctx_new_list))) |
6830 | io_sqd_init_new(sqd); | |
6c271ce2 | 6831 | |
e95eee2d | 6832 | cap_entries = !list_is_singular(&sqd->ctx_list); |
6c271ce2 | 6833 | |
69fb2131 JA |
6834 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) { |
6835 | if (current->cred != ctx->creds) { | |
6836 | if (old_cred) | |
6837 | revert_creds(old_cred); | |
6838 | old_cred = override_creds(ctx->creds); | |
bdcd3eab | 6839 | } |
91d8f519 | 6840 | io_sq_thread_associate_blkcg(ctx, &cur_css); |
4ea33a97 JA |
6841 | #ifdef CONFIG_AUDIT |
6842 | current->loginuid = ctx->loginuid; | |
6843 | current->sessionid = ctx->sessionid; | |
6844 | #endif | |
bdcd3eab | 6845 | |
e95eee2d | 6846 | ret |= __io_sq_thread(ctx, start_jiffies, cap_entries); |
6c271ce2 | 6847 | |
69fb2131 JA |
6848 | io_sq_thread_drop_mm(); |
6849 | } | |
6c271ce2 | 6850 | |
69fb2131 | 6851 | if (ret & SQT_SPIN) { |
c8d1ba58 JA |
6852 | io_run_task_work(); |
6853 | cond_resched(); | |
69fb2131 JA |
6854 | } else if (ret == SQT_IDLE) { |
6855 | if (kthread_should_park()) | |
6c271ce2 | 6856 | continue; |
69fb2131 JA |
6857 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) |
6858 | io_ring_set_wakeup_flag(ctx); | |
6859 | schedule(); | |
6860 | start_jiffies = jiffies; | |
6861 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) | |
6862 | io_ring_clear_wakeup_flag(ctx); | |
6c271ce2 | 6863 | } |
6c271ce2 JA |
6864 | } |
6865 | ||
4c6e277c | 6866 | io_run_task_work(); |
b41e9852 | 6867 | |
91d8f519 DZ |
6868 | if (cur_css) |
6869 | io_sq_thread_unassociate_blkcg(); | |
69fb2131 JA |
6870 | if (old_cred) |
6871 | revert_creds(old_cred); | |
06058632 | 6872 | |
2bbcd6d3 | 6873 | kthread_parkme(); |
06058632 | 6874 | |
6c271ce2 JA |
6875 | return 0; |
6876 | } | |
6877 | ||
bda52162 JA |
6878 | struct io_wait_queue { |
6879 | struct wait_queue_entry wq; | |
6880 | struct io_ring_ctx *ctx; | |
6881 | unsigned to_wait; | |
6882 | unsigned nr_timeouts; | |
6883 | }; | |
6884 | ||
1d7bb1d5 | 6885 | static inline bool io_should_wake(struct io_wait_queue *iowq, bool noflush) |
bda52162 JA |
6886 | { |
6887 | struct io_ring_ctx *ctx = iowq->ctx; | |
6888 | ||
6889 | /* | |
d195a66e | 6890 | * Wake up if we have enough events, or if a timeout occurred since we |
bda52162 JA |
6891 | * started waiting. For timeouts, we always want to return to userspace, |
6892 | * regardless of event count. | |
6893 | */ | |
1d7bb1d5 | 6894 | return io_cqring_events(ctx, noflush) >= iowq->to_wait || |
bda52162 JA |
6895 | atomic_read(&ctx->cq_timeouts) != iowq->nr_timeouts; |
6896 | } | |
6897 | ||
6898 | static int io_wake_function(struct wait_queue_entry *curr, unsigned int mode, | |
6899 | int wake_flags, void *key) | |
6900 | { | |
6901 | struct io_wait_queue *iowq = container_of(curr, struct io_wait_queue, | |
6902 | wq); | |
6903 | ||
1d7bb1d5 JA |
6904 | /* use noflush == true, as we can't safely rely on locking context */ |
6905 | if (!io_should_wake(iowq, true)) | |
bda52162 JA |
6906 | return -1; |
6907 | ||
6908 | return autoremove_wake_function(curr, mode, wake_flags, key); | |
6909 | } | |
6910 | ||
af9c1a44 JA |
6911 | static int io_run_task_work_sig(void) |
6912 | { | |
6913 | if (io_run_task_work()) | |
6914 | return 1; | |
6915 | if (!signal_pending(current)) | |
6916 | return 0; | |
6917 | if (current->jobctl & JOBCTL_TASK_WORK) { | |
6918 | spin_lock_irq(¤t->sighand->siglock); | |
6919 | current->jobctl &= ~JOBCTL_TASK_WORK; | |
6920 | recalc_sigpending(); | |
6921 | spin_unlock_irq(¤t->sighand->siglock); | |
6922 | return 1; | |
6923 | } | |
6924 | return -EINTR; | |
6925 | } | |
6926 | ||
2b188cc1 JA |
6927 | /* |
6928 | * Wait until events become available, if we don't already have some. The | |
6929 | * application must reap them itself, as they reside on the shared cq ring. | |
6930 | */ | |
6931 | static int io_cqring_wait(struct io_ring_ctx *ctx, int min_events, | |
6932 | const sigset_t __user *sig, size_t sigsz) | |
6933 | { | |
bda52162 JA |
6934 | struct io_wait_queue iowq = { |
6935 | .wq = { | |
6936 | .private = current, | |
6937 | .func = io_wake_function, | |
6938 | .entry = LIST_HEAD_INIT(iowq.wq.entry), | |
6939 | }, | |
6940 | .ctx = ctx, | |
6941 | .to_wait = min_events, | |
6942 | }; | |
75b28aff | 6943 | struct io_rings *rings = ctx->rings; |
e9ffa5c2 | 6944 | int ret = 0; |
2b188cc1 | 6945 | |
b41e9852 JA |
6946 | do { |
6947 | if (io_cqring_events(ctx, false) >= min_events) | |
6948 | return 0; | |
4c6e277c | 6949 | if (!io_run_task_work()) |
b41e9852 | 6950 | break; |
b41e9852 | 6951 | } while (1); |
2b188cc1 JA |
6952 | |
6953 | if (sig) { | |
9e75ad5d AB |
6954 | #ifdef CONFIG_COMPAT |
6955 | if (in_compat_syscall()) | |
6956 | ret = set_compat_user_sigmask((const compat_sigset_t __user *)sig, | |
b772434b | 6957 | sigsz); |
9e75ad5d AB |
6958 | else |
6959 | #endif | |
b772434b | 6960 | ret = set_user_sigmask(sig, sigsz); |
9e75ad5d | 6961 | |
2b188cc1 JA |
6962 | if (ret) |
6963 | return ret; | |
6964 | } | |
6965 | ||
bda52162 | 6966 | iowq.nr_timeouts = atomic_read(&ctx->cq_timeouts); |
c826bd7a | 6967 | trace_io_uring_cqring_wait(ctx, min_events); |
bda52162 JA |
6968 | do { |
6969 | prepare_to_wait_exclusive(&ctx->wait, &iowq.wq, | |
6970 | TASK_INTERRUPTIBLE); | |
ce593a6c | 6971 | /* make sure we run task_work before checking for signals */ |
af9c1a44 JA |
6972 | ret = io_run_task_work_sig(); |
6973 | if (ret > 0) | |
4c6e277c | 6974 | continue; |
af9c1a44 | 6975 | else if (ret < 0) |
bda52162 | 6976 | break; |
ce593a6c JA |
6977 | if (io_should_wake(&iowq, false)) |
6978 | break; | |
6979 | schedule(); | |
bda52162 JA |
6980 | } while (1); |
6981 | finish_wait(&ctx->wait, &iowq.wq); | |
6982 | ||
b7db41c9 | 6983 | restore_saved_sigmask_unless(ret == -EINTR); |
2b188cc1 | 6984 | |
75b28aff | 6985 | return READ_ONCE(rings->cq.head) == READ_ONCE(rings->cq.tail) ? ret : 0; |
2b188cc1 JA |
6986 | } |
6987 | ||
6b06314c JA |
6988 | static void __io_sqe_files_unregister(struct io_ring_ctx *ctx) |
6989 | { | |
6990 | #if defined(CONFIG_UNIX) | |
6991 | if (ctx->ring_sock) { | |
6992 | struct sock *sock = ctx->ring_sock->sk; | |
6993 | struct sk_buff *skb; | |
6994 | ||
6995 | while ((skb = skb_dequeue(&sock->sk_receive_queue)) != NULL) | |
6996 | kfree_skb(skb); | |
6997 | } | |
6998 | #else | |
6999 | int i; | |
7000 | ||
65e19f54 JA |
7001 | for (i = 0; i < ctx->nr_user_files; i++) { |
7002 | struct file *file; | |
7003 | ||
7004 | file = io_file_from_index(ctx, i); | |
7005 | if (file) | |
7006 | fput(file); | |
7007 | } | |
6b06314c JA |
7008 | #endif |
7009 | } | |
7010 | ||
05f3fb3c JA |
7011 | static void io_file_ref_kill(struct percpu_ref *ref) |
7012 | { | |
7013 | struct fixed_file_data *data; | |
7014 | ||
7015 | data = container_of(ref, struct fixed_file_data, refs); | |
7016 | complete(&data->done); | |
7017 | } | |
7018 | ||
6b06314c JA |
7019 | static int io_sqe_files_unregister(struct io_ring_ctx *ctx) |
7020 | { | |
05f3fb3c | 7021 | struct fixed_file_data *data = ctx->file_data; |
05589553 | 7022 | struct fixed_file_ref_node *ref_node = NULL; |
65e19f54 JA |
7023 | unsigned nr_tables, i; |
7024 | ||
05f3fb3c | 7025 | if (!data) |
6b06314c JA |
7026 | return -ENXIO; |
7027 | ||
6a4d07cd | 7028 | spin_lock(&data->lock); |
1e5d770b | 7029 | ref_node = data->node; |
6a4d07cd | 7030 | spin_unlock(&data->lock); |
05589553 XW |
7031 | if (ref_node) |
7032 | percpu_ref_kill(&ref_node->refs); | |
7033 | ||
7034 | percpu_ref_kill(&data->refs); | |
7035 | ||
7036 | /* wait for all refs nodes to complete */ | |
4a38aed2 | 7037 | flush_delayed_work(&ctx->file_put_work); |
2faf852d | 7038 | wait_for_completion(&data->done); |
05f3fb3c | 7039 | |
6b06314c | 7040 | __io_sqe_files_unregister(ctx); |
65e19f54 JA |
7041 | nr_tables = DIV_ROUND_UP(ctx->nr_user_files, IORING_MAX_FILES_TABLE); |
7042 | for (i = 0; i < nr_tables; i++) | |
05f3fb3c JA |
7043 | kfree(data->table[i].files); |
7044 | kfree(data->table); | |
05589553 XW |
7045 | percpu_ref_exit(&data->refs); |
7046 | kfree(data); | |
05f3fb3c | 7047 | ctx->file_data = NULL; |
6b06314c JA |
7048 | ctx->nr_user_files = 0; |
7049 | return 0; | |
7050 | } | |
7051 | ||
534ca6d6 | 7052 | static void io_put_sq_data(struct io_sq_data *sqd) |
6c271ce2 | 7053 | { |
534ca6d6 | 7054 | if (refcount_dec_and_test(&sqd->refs)) { |
2bbcd6d3 RP |
7055 | /* |
7056 | * The park is a bit of a work-around, without it we get | |
7057 | * warning spews on shutdown with SQPOLL set and affinity | |
7058 | * set to a single CPU. | |
7059 | */ | |
534ca6d6 JA |
7060 | if (sqd->thread) { |
7061 | kthread_park(sqd->thread); | |
7062 | kthread_stop(sqd->thread); | |
7063 | } | |
7064 | ||
7065 | kfree(sqd); | |
7066 | } | |
7067 | } | |
7068 | ||
aa06165d JA |
7069 | static struct io_sq_data *io_attach_sq_data(struct io_uring_params *p) |
7070 | { | |
7071 | struct io_ring_ctx *ctx_attach; | |
7072 | struct io_sq_data *sqd; | |
7073 | struct fd f; | |
7074 | ||
7075 | f = fdget(p->wq_fd); | |
7076 | if (!f.file) | |
7077 | return ERR_PTR(-ENXIO); | |
7078 | if (f.file->f_op != &io_uring_fops) { | |
7079 | fdput(f); | |
7080 | return ERR_PTR(-EINVAL); | |
7081 | } | |
7082 | ||
7083 | ctx_attach = f.file->private_data; | |
7084 | sqd = ctx_attach->sq_data; | |
7085 | if (!sqd) { | |
7086 | fdput(f); | |
7087 | return ERR_PTR(-EINVAL); | |
7088 | } | |
7089 | ||
7090 | refcount_inc(&sqd->refs); | |
7091 | fdput(f); | |
7092 | return sqd; | |
7093 | } | |
7094 | ||
534ca6d6 JA |
7095 | static struct io_sq_data *io_get_sq_data(struct io_uring_params *p) |
7096 | { | |
7097 | struct io_sq_data *sqd; | |
7098 | ||
aa06165d JA |
7099 | if (p->flags & IORING_SETUP_ATTACH_WQ) |
7100 | return io_attach_sq_data(p); | |
7101 | ||
534ca6d6 JA |
7102 | sqd = kzalloc(sizeof(*sqd), GFP_KERNEL); |
7103 | if (!sqd) | |
7104 | return ERR_PTR(-ENOMEM); | |
7105 | ||
7106 | refcount_set(&sqd->refs, 1); | |
69fb2131 JA |
7107 | INIT_LIST_HEAD(&sqd->ctx_list); |
7108 | INIT_LIST_HEAD(&sqd->ctx_new_list); | |
7109 | mutex_init(&sqd->ctx_lock); | |
7110 | mutex_init(&sqd->lock); | |
534ca6d6 JA |
7111 | init_waitqueue_head(&sqd->wait); |
7112 | return sqd; | |
7113 | } | |
7114 | ||
69fb2131 JA |
7115 | static void io_sq_thread_unpark(struct io_sq_data *sqd) |
7116 | __releases(&sqd->lock) | |
7117 | { | |
7118 | if (!sqd->thread) | |
7119 | return; | |
7120 | kthread_unpark(sqd->thread); | |
7121 | mutex_unlock(&sqd->lock); | |
7122 | } | |
7123 | ||
7124 | static void io_sq_thread_park(struct io_sq_data *sqd) | |
7125 | __acquires(&sqd->lock) | |
7126 | { | |
7127 | if (!sqd->thread) | |
7128 | return; | |
7129 | mutex_lock(&sqd->lock); | |
7130 | kthread_park(sqd->thread); | |
7131 | } | |
7132 | ||
534ca6d6 JA |
7133 | static void io_sq_thread_stop(struct io_ring_ctx *ctx) |
7134 | { | |
7135 | struct io_sq_data *sqd = ctx->sq_data; | |
7136 | ||
7137 | if (sqd) { | |
7138 | if (sqd->thread) { | |
7139 | /* | |
7140 | * We may arrive here from the error branch in | |
7141 | * io_sq_offload_create() where the kthread is created | |
7142 | * without being waked up, thus wake it up now to make | |
7143 | * sure the wait will complete. | |
7144 | */ | |
7145 | wake_up_process(sqd->thread); | |
7146 | wait_for_completion(&ctx->sq_thread_comp); | |
69fb2131 JA |
7147 | |
7148 | io_sq_thread_park(sqd); | |
7149 | } | |
7150 | ||
7151 | mutex_lock(&sqd->ctx_lock); | |
7152 | list_del(&ctx->sqd_list); | |
7153 | mutex_unlock(&sqd->ctx_lock); | |
7154 | ||
7155 | if (sqd->thread) { | |
7156 | finish_wait(&sqd->wait, &ctx->sqo_wait_entry); | |
7157 | io_sq_thread_unpark(sqd); | |
534ca6d6 JA |
7158 | } |
7159 | ||
7160 | io_put_sq_data(sqd); | |
7161 | ctx->sq_data = NULL; | |
6c271ce2 JA |
7162 | } |
7163 | } | |
7164 | ||
6b06314c JA |
7165 | static void io_finish_async(struct io_ring_ctx *ctx) |
7166 | { | |
6c271ce2 JA |
7167 | io_sq_thread_stop(ctx); |
7168 | ||
561fb04a JA |
7169 | if (ctx->io_wq) { |
7170 | io_wq_destroy(ctx->io_wq); | |
7171 | ctx->io_wq = NULL; | |
6b06314c JA |
7172 | } |
7173 | } | |
7174 | ||
7175 | #if defined(CONFIG_UNIX) | |
6b06314c JA |
7176 | /* |
7177 | * Ensure the UNIX gc is aware of our file set, so we are certain that | |
7178 | * the io_uring can be safely unregistered on process exit, even if we have | |
7179 | * loops in the file referencing. | |
7180 | */ | |
7181 | static int __io_sqe_files_scm(struct io_ring_ctx *ctx, int nr, int offset) | |
7182 | { | |
7183 | struct sock *sk = ctx->ring_sock->sk; | |
7184 | struct scm_fp_list *fpl; | |
7185 | struct sk_buff *skb; | |
08a45173 | 7186 | int i, nr_files; |
6b06314c | 7187 | |
6b06314c JA |
7188 | fpl = kzalloc(sizeof(*fpl), GFP_KERNEL); |
7189 | if (!fpl) | |
7190 | return -ENOMEM; | |
7191 | ||
7192 | skb = alloc_skb(0, GFP_KERNEL); | |
7193 | if (!skb) { | |
7194 | kfree(fpl); | |
7195 | return -ENOMEM; | |
7196 | } | |
7197 | ||
7198 | skb->sk = sk; | |
6b06314c | 7199 | |
08a45173 | 7200 | nr_files = 0; |
6b06314c JA |
7201 | fpl->user = get_uid(ctx->user); |
7202 | for (i = 0; i < nr; i++) { | |
65e19f54 JA |
7203 | struct file *file = io_file_from_index(ctx, i + offset); |
7204 | ||
7205 | if (!file) | |
08a45173 | 7206 | continue; |
65e19f54 | 7207 | fpl->fp[nr_files] = get_file(file); |
08a45173 JA |
7208 | unix_inflight(fpl->user, fpl->fp[nr_files]); |
7209 | nr_files++; | |
6b06314c JA |
7210 | } |
7211 | ||
08a45173 JA |
7212 | if (nr_files) { |
7213 | fpl->max = SCM_MAX_FD; | |
7214 | fpl->count = nr_files; | |
7215 | UNIXCB(skb).fp = fpl; | |
05f3fb3c | 7216 | skb->destructor = unix_destruct_scm; |
08a45173 JA |
7217 | refcount_add(skb->truesize, &sk->sk_wmem_alloc); |
7218 | skb_queue_head(&sk->sk_receive_queue, skb); | |
6b06314c | 7219 | |
08a45173 JA |
7220 | for (i = 0; i < nr_files; i++) |
7221 | fput(fpl->fp[i]); | |
7222 | } else { | |
7223 | kfree_skb(skb); | |
7224 | kfree(fpl); | |
7225 | } | |
6b06314c JA |
7226 | |
7227 | return 0; | |
7228 | } | |
7229 | ||
7230 | /* | |
7231 | * If UNIX sockets are enabled, fd passing can cause a reference cycle which | |
7232 | * causes regular reference counting to break down. We rely on the UNIX | |
7233 | * garbage collection to take care of this problem for us. | |
7234 | */ | |
7235 | static int io_sqe_files_scm(struct io_ring_ctx *ctx) | |
7236 | { | |
7237 | unsigned left, total; | |
7238 | int ret = 0; | |
7239 | ||
7240 | total = 0; | |
7241 | left = ctx->nr_user_files; | |
7242 | while (left) { | |
7243 | unsigned this_files = min_t(unsigned, left, SCM_MAX_FD); | |
6b06314c JA |
7244 | |
7245 | ret = __io_sqe_files_scm(ctx, this_files, total); | |
7246 | if (ret) | |
7247 | break; | |
7248 | left -= this_files; | |
7249 | total += this_files; | |
7250 | } | |
7251 | ||
7252 | if (!ret) | |
7253 | return 0; | |
7254 | ||
7255 | while (total < ctx->nr_user_files) { | |
65e19f54 JA |
7256 | struct file *file = io_file_from_index(ctx, total); |
7257 | ||
7258 | if (file) | |
7259 | fput(file); | |
6b06314c JA |
7260 | total++; |
7261 | } | |
7262 | ||
7263 | return ret; | |
7264 | } | |
7265 | #else | |
7266 | static int io_sqe_files_scm(struct io_ring_ctx *ctx) | |
7267 | { | |
7268 | return 0; | |
7269 | } | |
7270 | #endif | |
7271 | ||
5398ae69 PB |
7272 | static int io_sqe_alloc_file_tables(struct fixed_file_data *file_data, |
7273 | unsigned nr_tables, unsigned nr_files) | |
65e19f54 JA |
7274 | { |
7275 | int i; | |
7276 | ||
7277 | for (i = 0; i < nr_tables; i++) { | |
5398ae69 | 7278 | struct fixed_file_table *table = &file_data->table[i]; |
65e19f54 JA |
7279 | unsigned this_files; |
7280 | ||
7281 | this_files = min(nr_files, IORING_MAX_FILES_TABLE); | |
7282 | table->files = kcalloc(this_files, sizeof(struct file *), | |
7283 | GFP_KERNEL); | |
7284 | if (!table->files) | |
7285 | break; | |
7286 | nr_files -= this_files; | |
7287 | } | |
7288 | ||
7289 | if (i == nr_tables) | |
7290 | return 0; | |
7291 | ||
7292 | for (i = 0; i < nr_tables; i++) { | |
5398ae69 | 7293 | struct fixed_file_table *table = &file_data->table[i]; |
65e19f54 JA |
7294 | kfree(table->files); |
7295 | } | |
7296 | return 1; | |
7297 | } | |
7298 | ||
05f3fb3c JA |
7299 | static void io_ring_file_put(struct io_ring_ctx *ctx, struct file *file) |
7300 | { | |
7301 | #if defined(CONFIG_UNIX) | |
7302 | struct sock *sock = ctx->ring_sock->sk; | |
7303 | struct sk_buff_head list, *head = &sock->sk_receive_queue; | |
7304 | struct sk_buff *skb; | |
7305 | int i; | |
7306 | ||
7307 | __skb_queue_head_init(&list); | |
7308 | ||
7309 | /* | |
7310 | * Find the skb that holds this file in its SCM_RIGHTS. When found, | |
7311 | * remove this entry and rearrange the file array. | |
7312 | */ | |
7313 | skb = skb_dequeue(head); | |
7314 | while (skb) { | |
7315 | struct scm_fp_list *fp; | |
7316 | ||
7317 | fp = UNIXCB(skb).fp; | |
7318 | for (i = 0; i < fp->count; i++) { | |
7319 | int left; | |
7320 | ||
7321 | if (fp->fp[i] != file) | |
7322 | continue; | |
7323 | ||
7324 | unix_notinflight(fp->user, fp->fp[i]); | |
7325 | left = fp->count - 1 - i; | |
7326 | if (left) { | |
7327 | memmove(&fp->fp[i], &fp->fp[i + 1], | |
7328 | left * sizeof(struct file *)); | |
7329 | } | |
7330 | fp->count--; | |
7331 | if (!fp->count) { | |
7332 | kfree_skb(skb); | |
7333 | skb = NULL; | |
7334 | } else { | |
7335 | __skb_queue_tail(&list, skb); | |
7336 | } | |
7337 | fput(file); | |
7338 | file = NULL; | |
7339 | break; | |
7340 | } | |
7341 | ||
7342 | if (!file) | |
7343 | break; | |
7344 | ||
7345 | __skb_queue_tail(&list, skb); | |
7346 | ||
7347 | skb = skb_dequeue(head); | |
7348 | } | |
7349 | ||
7350 | if (skb_peek(&list)) { | |
7351 | spin_lock_irq(&head->lock); | |
7352 | while ((skb = __skb_dequeue(&list)) != NULL) | |
7353 | __skb_queue_tail(head, skb); | |
7354 | spin_unlock_irq(&head->lock); | |
7355 | } | |
7356 | #else | |
7357 | fput(file); | |
7358 | #endif | |
7359 | } | |
7360 | ||
7361 | struct io_file_put { | |
05589553 | 7362 | struct list_head list; |
05f3fb3c | 7363 | struct file *file; |
05f3fb3c JA |
7364 | }; |
7365 | ||
4a38aed2 | 7366 | static void __io_file_put_work(struct fixed_file_ref_node *ref_node) |
65e19f54 | 7367 | { |
4a38aed2 JA |
7368 | struct fixed_file_data *file_data = ref_node->file_data; |
7369 | struct io_ring_ctx *ctx = file_data->ctx; | |
05f3fb3c | 7370 | struct io_file_put *pfile, *tmp; |
05589553 XW |
7371 | |
7372 | list_for_each_entry_safe(pfile, tmp, &ref_node->file_list, list) { | |
6a4d07cd | 7373 | list_del(&pfile->list); |
05589553 XW |
7374 | io_ring_file_put(ctx, pfile->file); |
7375 | kfree(pfile); | |
65e19f54 | 7376 | } |
05589553 | 7377 | |
05589553 XW |
7378 | percpu_ref_exit(&ref_node->refs); |
7379 | kfree(ref_node); | |
7380 | percpu_ref_put(&file_data->refs); | |
2faf852d | 7381 | } |
65e19f54 | 7382 | |
4a38aed2 JA |
7383 | static void io_file_put_work(struct work_struct *work) |
7384 | { | |
7385 | struct io_ring_ctx *ctx; | |
7386 | struct llist_node *node; | |
7387 | ||
7388 | ctx = container_of(work, struct io_ring_ctx, file_put_work.work); | |
7389 | node = llist_del_all(&ctx->file_put_llist); | |
7390 | ||
7391 | while (node) { | |
7392 | struct fixed_file_ref_node *ref_node; | |
7393 | struct llist_node *next = node->next; | |
7394 | ||
7395 | ref_node = llist_entry(node, struct fixed_file_ref_node, llist); | |
7396 | __io_file_put_work(ref_node); | |
7397 | node = next; | |
7398 | } | |
7399 | } | |
7400 | ||
05589553 | 7401 | static void io_file_data_ref_zero(struct percpu_ref *ref) |
2faf852d | 7402 | { |
05589553 | 7403 | struct fixed_file_ref_node *ref_node; |
e297822b | 7404 | struct fixed_file_data *data; |
4a38aed2 | 7405 | struct io_ring_ctx *ctx; |
e297822b | 7406 | bool first_add = false; |
4a38aed2 | 7407 | int delay = HZ; |
65e19f54 | 7408 | |
05589553 | 7409 | ref_node = container_of(ref, struct fixed_file_ref_node, refs); |
e297822b PB |
7410 | data = ref_node->file_data; |
7411 | ctx = data->ctx; | |
7412 | ||
7413 | spin_lock(&data->lock); | |
7414 | ref_node->done = true; | |
7415 | ||
7416 | while (!list_empty(&data->ref_list)) { | |
7417 | ref_node = list_first_entry(&data->ref_list, | |
7418 | struct fixed_file_ref_node, node); | |
7419 | /* recycle ref nodes in order */ | |
7420 | if (!ref_node->done) | |
7421 | break; | |
7422 | list_del(&ref_node->node); | |
7423 | first_add |= llist_add(&ref_node->llist, &ctx->file_put_llist); | |
7424 | } | |
7425 | spin_unlock(&data->lock); | |
05589553 | 7426 | |
e297822b | 7427 | if (percpu_ref_is_dying(&data->refs)) |
4a38aed2 | 7428 | delay = 0; |
05589553 | 7429 | |
4a38aed2 JA |
7430 | if (!delay) |
7431 | mod_delayed_work(system_wq, &ctx->file_put_work, 0); | |
7432 | else if (first_add) | |
7433 | queue_delayed_work(system_wq, &ctx->file_put_work, delay); | |
05f3fb3c | 7434 | } |
65e19f54 | 7435 | |
05589553 XW |
7436 | static struct fixed_file_ref_node *alloc_fixed_file_ref_node( |
7437 | struct io_ring_ctx *ctx) | |
05f3fb3c | 7438 | { |
05589553 | 7439 | struct fixed_file_ref_node *ref_node; |
05f3fb3c | 7440 | |
05589553 XW |
7441 | ref_node = kzalloc(sizeof(*ref_node), GFP_KERNEL); |
7442 | if (!ref_node) | |
7443 | return ERR_PTR(-ENOMEM); | |
05f3fb3c | 7444 | |
05589553 XW |
7445 | if (percpu_ref_init(&ref_node->refs, io_file_data_ref_zero, |
7446 | 0, GFP_KERNEL)) { | |
7447 | kfree(ref_node); | |
7448 | return ERR_PTR(-ENOMEM); | |
7449 | } | |
7450 | INIT_LIST_HEAD(&ref_node->node); | |
7451 | INIT_LIST_HEAD(&ref_node->file_list); | |
05589553 | 7452 | ref_node->file_data = ctx->file_data; |
e297822b | 7453 | ref_node->done = false; |
05589553 | 7454 | return ref_node; |
05589553 XW |
7455 | } |
7456 | ||
7457 | static void destroy_fixed_file_ref_node(struct fixed_file_ref_node *ref_node) | |
7458 | { | |
7459 | percpu_ref_exit(&ref_node->refs); | |
7460 | kfree(ref_node); | |
65e19f54 JA |
7461 | } |
7462 | ||
6b06314c JA |
7463 | static int io_sqe_files_register(struct io_ring_ctx *ctx, void __user *arg, |
7464 | unsigned nr_args) | |
7465 | { | |
7466 | __s32 __user *fds = (__s32 __user *) arg; | |
600cf3f8 | 7467 | unsigned nr_tables, i; |
05f3fb3c | 7468 | struct file *file; |
600cf3f8 | 7469 | int fd, ret = -ENOMEM; |
05589553 | 7470 | struct fixed_file_ref_node *ref_node; |
5398ae69 | 7471 | struct fixed_file_data *file_data; |
6b06314c | 7472 | |
05f3fb3c | 7473 | if (ctx->file_data) |
6b06314c JA |
7474 | return -EBUSY; |
7475 | if (!nr_args) | |
7476 | return -EINVAL; | |
7477 | if (nr_args > IORING_MAX_FIXED_FILES) | |
7478 | return -EMFILE; | |
7479 | ||
5398ae69 PB |
7480 | file_data = kzalloc(sizeof(*ctx->file_data), GFP_KERNEL); |
7481 | if (!file_data) | |
05f3fb3c | 7482 | return -ENOMEM; |
5398ae69 PB |
7483 | file_data->ctx = ctx; |
7484 | init_completion(&file_data->done); | |
7485 | INIT_LIST_HEAD(&file_data->ref_list); | |
7486 | spin_lock_init(&file_data->lock); | |
05f3fb3c | 7487 | |
65e19f54 | 7488 | nr_tables = DIV_ROUND_UP(nr_args, IORING_MAX_FILES_TABLE); |
035fbafc | 7489 | file_data->table = kcalloc(nr_tables, sizeof(*file_data->table), |
5398ae69 | 7490 | GFP_KERNEL); |
600cf3f8 PB |
7491 | if (!file_data->table) |
7492 | goto out_free; | |
05f3fb3c | 7493 | |
5398ae69 | 7494 | if (percpu_ref_init(&file_data->refs, io_file_ref_kill, |
600cf3f8 PB |
7495 | PERCPU_REF_ALLOW_REINIT, GFP_KERNEL)) |
7496 | goto out_free; | |
6b06314c | 7497 | |
600cf3f8 PB |
7498 | if (io_sqe_alloc_file_tables(file_data, nr_tables, nr_args)) |
7499 | goto out_ref; | |
55cbc256 | 7500 | ctx->file_data = file_data; |
65e19f54 | 7501 | |
08a45173 | 7502 | for (i = 0; i < nr_args; i++, ctx->nr_user_files++) { |
65e19f54 JA |
7503 | struct fixed_file_table *table; |
7504 | unsigned index; | |
7505 | ||
600cf3f8 PB |
7506 | if (copy_from_user(&fd, &fds[i], sizeof(fd))) { |
7507 | ret = -EFAULT; | |
7508 | goto out_fput; | |
7509 | } | |
08a45173 | 7510 | /* allow sparse sets */ |
600cf3f8 | 7511 | if (fd == -1) |
08a45173 | 7512 | continue; |
6b06314c | 7513 | |
05f3fb3c | 7514 | file = fget(fd); |
6b06314c | 7515 | ret = -EBADF; |
05f3fb3c | 7516 | if (!file) |
600cf3f8 | 7517 | goto out_fput; |
05f3fb3c | 7518 | |
6b06314c JA |
7519 | /* |
7520 | * Don't allow io_uring instances to be registered. If UNIX | |
7521 | * isn't enabled, then this causes a reference cycle and this | |
7522 | * instance can never get freed. If UNIX is enabled we'll | |
7523 | * handle it just fine, but there's still no point in allowing | |
7524 | * a ring fd as it doesn't support regular read/write anyway. | |
7525 | */ | |
05f3fb3c JA |
7526 | if (file->f_op == &io_uring_fops) { |
7527 | fput(file); | |
600cf3f8 | 7528 | goto out_fput; |
6b06314c | 7529 | } |
600cf3f8 PB |
7530 | table = &file_data->table[i >> IORING_FILE_TABLE_SHIFT]; |
7531 | index = i & IORING_FILE_TABLE_MASK; | |
05f3fb3c | 7532 | table->files[index] = file; |
6b06314c JA |
7533 | } |
7534 | ||
6b06314c | 7535 | ret = io_sqe_files_scm(ctx); |
05589553 | 7536 | if (ret) { |
6b06314c | 7537 | io_sqe_files_unregister(ctx); |
05589553 XW |
7538 | return ret; |
7539 | } | |
6b06314c | 7540 | |
05589553 XW |
7541 | ref_node = alloc_fixed_file_ref_node(ctx); |
7542 | if (IS_ERR(ref_node)) { | |
7543 | io_sqe_files_unregister(ctx); | |
7544 | return PTR_ERR(ref_node); | |
7545 | } | |
7546 | ||
b2e96852 | 7547 | file_data->node = ref_node; |
5398ae69 | 7548 | spin_lock(&file_data->lock); |
e297822b | 7549 | list_add_tail(&ref_node->node, &file_data->ref_list); |
5398ae69 PB |
7550 | spin_unlock(&file_data->lock); |
7551 | percpu_ref_get(&file_data->refs); | |
6b06314c | 7552 | return ret; |
600cf3f8 PB |
7553 | out_fput: |
7554 | for (i = 0; i < ctx->nr_user_files; i++) { | |
7555 | file = io_file_from_index(ctx, i); | |
7556 | if (file) | |
7557 | fput(file); | |
7558 | } | |
7559 | for (i = 0; i < nr_tables; i++) | |
7560 | kfree(file_data->table[i].files); | |
7561 | ctx->nr_user_files = 0; | |
7562 | out_ref: | |
7563 | percpu_ref_exit(&file_data->refs); | |
7564 | out_free: | |
7565 | kfree(file_data->table); | |
7566 | kfree(file_data); | |
55cbc256 | 7567 | ctx->file_data = NULL; |
6b06314c JA |
7568 | return ret; |
7569 | } | |
7570 | ||
c3a31e60 JA |
7571 | static int io_sqe_file_register(struct io_ring_ctx *ctx, struct file *file, |
7572 | int index) | |
7573 | { | |
7574 | #if defined(CONFIG_UNIX) | |
7575 | struct sock *sock = ctx->ring_sock->sk; | |
7576 | struct sk_buff_head *head = &sock->sk_receive_queue; | |
7577 | struct sk_buff *skb; | |
7578 | ||
7579 | /* | |
7580 | * See if we can merge this file into an existing skb SCM_RIGHTS | |
7581 | * file set. If there's no room, fall back to allocating a new skb | |
7582 | * and filling it in. | |
7583 | */ | |
7584 | spin_lock_irq(&head->lock); | |
7585 | skb = skb_peek(head); | |
7586 | if (skb) { | |
7587 | struct scm_fp_list *fpl = UNIXCB(skb).fp; | |
7588 | ||
7589 | if (fpl->count < SCM_MAX_FD) { | |
7590 | __skb_unlink(skb, head); | |
7591 | spin_unlock_irq(&head->lock); | |
7592 | fpl->fp[fpl->count] = get_file(file); | |
7593 | unix_inflight(fpl->user, fpl->fp[fpl->count]); | |
7594 | fpl->count++; | |
7595 | spin_lock_irq(&head->lock); | |
7596 | __skb_queue_head(head, skb); | |
7597 | } else { | |
7598 | skb = NULL; | |
7599 | } | |
7600 | } | |
7601 | spin_unlock_irq(&head->lock); | |
7602 | ||
7603 | if (skb) { | |
7604 | fput(file); | |
7605 | return 0; | |
7606 | } | |
7607 | ||
7608 | return __io_sqe_files_scm(ctx, 1, index); | |
7609 | #else | |
7610 | return 0; | |
7611 | #endif | |
7612 | } | |
7613 | ||
a5318d3c | 7614 | static int io_queue_file_removal(struct fixed_file_data *data, |
05589553 | 7615 | struct file *file) |
05f3fb3c | 7616 | { |
a5318d3c | 7617 | struct io_file_put *pfile; |
b2e96852 | 7618 | struct fixed_file_ref_node *ref_node = data->node; |
05f3fb3c | 7619 | |
05f3fb3c | 7620 | pfile = kzalloc(sizeof(*pfile), GFP_KERNEL); |
a5318d3c HD |
7621 | if (!pfile) |
7622 | return -ENOMEM; | |
05f3fb3c JA |
7623 | |
7624 | pfile->file = file; | |
05589553 XW |
7625 | list_add(&pfile->list, &ref_node->file_list); |
7626 | ||
a5318d3c | 7627 | return 0; |
05f3fb3c JA |
7628 | } |
7629 | ||
7630 | static int __io_sqe_files_update(struct io_ring_ctx *ctx, | |
7631 | struct io_uring_files_update *up, | |
7632 | unsigned nr_args) | |
7633 | { | |
7634 | struct fixed_file_data *data = ctx->file_data; | |
05589553 | 7635 | struct fixed_file_ref_node *ref_node; |
05f3fb3c | 7636 | struct file *file; |
c3a31e60 JA |
7637 | __s32 __user *fds; |
7638 | int fd, i, err; | |
7639 | __u32 done; | |
05589553 | 7640 | bool needs_switch = false; |
c3a31e60 | 7641 | |
05f3fb3c | 7642 | if (check_add_overflow(up->offset, nr_args, &done)) |
c3a31e60 JA |
7643 | return -EOVERFLOW; |
7644 | if (done > ctx->nr_user_files) | |
7645 | return -EINVAL; | |
7646 | ||
05589553 XW |
7647 | ref_node = alloc_fixed_file_ref_node(ctx); |
7648 | if (IS_ERR(ref_node)) | |
7649 | return PTR_ERR(ref_node); | |
7650 | ||
c3a31e60 | 7651 | done = 0; |
05f3fb3c | 7652 | fds = u64_to_user_ptr(up->fds); |
c3a31e60 | 7653 | while (nr_args) { |
65e19f54 JA |
7654 | struct fixed_file_table *table; |
7655 | unsigned index; | |
7656 | ||
c3a31e60 JA |
7657 | err = 0; |
7658 | if (copy_from_user(&fd, &fds[done], sizeof(fd))) { | |
7659 | err = -EFAULT; | |
7660 | break; | |
7661 | } | |
05f3fb3c JA |
7662 | i = array_index_nospec(up->offset, ctx->nr_user_files); |
7663 | table = &ctx->file_data->table[i >> IORING_FILE_TABLE_SHIFT]; | |
65e19f54 JA |
7664 | index = i & IORING_FILE_TABLE_MASK; |
7665 | if (table->files[index]) { | |
98dfd502 | 7666 | file = table->files[index]; |
a5318d3c HD |
7667 | err = io_queue_file_removal(data, file); |
7668 | if (err) | |
7669 | break; | |
65e19f54 | 7670 | table->files[index] = NULL; |
05589553 | 7671 | needs_switch = true; |
c3a31e60 JA |
7672 | } |
7673 | if (fd != -1) { | |
c3a31e60 JA |
7674 | file = fget(fd); |
7675 | if (!file) { | |
7676 | err = -EBADF; | |
7677 | break; | |
7678 | } | |
7679 | /* | |
7680 | * Don't allow io_uring instances to be registered. If | |
7681 | * UNIX isn't enabled, then this causes a reference | |
7682 | * cycle and this instance can never get freed. If UNIX | |
7683 | * is enabled we'll handle it just fine, but there's | |
7684 | * still no point in allowing a ring fd as it doesn't | |
7685 | * support regular read/write anyway. | |
7686 | */ | |
7687 | if (file->f_op == &io_uring_fops) { | |
7688 | fput(file); | |
7689 | err = -EBADF; | |
7690 | break; | |
7691 | } | |
65e19f54 | 7692 | table->files[index] = file; |
c3a31e60 | 7693 | err = io_sqe_file_register(ctx, file, i); |
f3bd9dae | 7694 | if (err) { |
95d1c8e5 | 7695 | table->files[index] = NULL; |
f3bd9dae | 7696 | fput(file); |
c3a31e60 | 7697 | break; |
f3bd9dae | 7698 | } |
c3a31e60 JA |
7699 | } |
7700 | nr_args--; | |
7701 | done++; | |
05f3fb3c JA |
7702 | up->offset++; |
7703 | } | |
7704 | ||
05589553 | 7705 | if (needs_switch) { |
b2e96852 | 7706 | percpu_ref_kill(&data->node->refs); |
6a4d07cd | 7707 | spin_lock(&data->lock); |
e297822b | 7708 | list_add_tail(&ref_node->node, &data->ref_list); |
b2e96852 | 7709 | data->node = ref_node; |
6a4d07cd | 7710 | spin_unlock(&data->lock); |
05589553 XW |
7711 | percpu_ref_get(&ctx->file_data->refs); |
7712 | } else | |
7713 | destroy_fixed_file_ref_node(ref_node); | |
c3a31e60 JA |
7714 | |
7715 | return done ? done : err; | |
7716 | } | |
05589553 | 7717 | |
05f3fb3c JA |
7718 | static int io_sqe_files_update(struct io_ring_ctx *ctx, void __user *arg, |
7719 | unsigned nr_args) | |
7720 | { | |
7721 | struct io_uring_files_update up; | |
7722 | ||
7723 | if (!ctx->file_data) | |
7724 | return -ENXIO; | |
7725 | if (!nr_args) | |
7726 | return -EINVAL; | |
7727 | if (copy_from_user(&up, arg, sizeof(up))) | |
7728 | return -EFAULT; | |
7729 | if (up.resv) | |
7730 | return -EINVAL; | |
7731 | ||
7732 | return __io_sqe_files_update(ctx, &up, nr_args); | |
7733 | } | |
c3a31e60 | 7734 | |
e9fd9396 | 7735 | static void io_free_work(struct io_wq_work *work) |
7d723065 JA |
7736 | { |
7737 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); | |
7738 | ||
e9fd9396 | 7739 | /* Consider that io_steal_work() relies on this ref */ |
7d723065 JA |
7740 | io_put_req(req); |
7741 | } | |
7742 | ||
24369c2e PB |
7743 | static int io_init_wq_offload(struct io_ring_ctx *ctx, |
7744 | struct io_uring_params *p) | |
7745 | { | |
7746 | struct io_wq_data data; | |
7747 | struct fd f; | |
7748 | struct io_ring_ctx *ctx_attach; | |
7749 | unsigned int concurrency; | |
7750 | int ret = 0; | |
7751 | ||
7752 | data.user = ctx->user; | |
e9fd9396 | 7753 | data.free_work = io_free_work; |
f5fa38c5 | 7754 | data.do_work = io_wq_submit_work; |
24369c2e PB |
7755 | |
7756 | if (!(p->flags & IORING_SETUP_ATTACH_WQ)) { | |
7757 | /* Do QD, or 4 * CPUS, whatever is smallest */ | |
7758 | concurrency = min(ctx->sq_entries, 4 * num_online_cpus()); | |
7759 | ||
7760 | ctx->io_wq = io_wq_create(concurrency, &data); | |
7761 | if (IS_ERR(ctx->io_wq)) { | |
7762 | ret = PTR_ERR(ctx->io_wq); | |
7763 | ctx->io_wq = NULL; | |
7764 | } | |
7765 | return ret; | |
7766 | } | |
7767 | ||
7768 | f = fdget(p->wq_fd); | |
7769 | if (!f.file) | |
7770 | return -EBADF; | |
7771 | ||
7772 | if (f.file->f_op != &io_uring_fops) { | |
7773 | ret = -EINVAL; | |
7774 | goto out_fput; | |
7775 | } | |
7776 | ||
7777 | ctx_attach = f.file->private_data; | |
7778 | /* @io_wq is protected by holding the fd */ | |
7779 | if (!io_wq_get(ctx_attach->io_wq, &data)) { | |
7780 | ret = -EINVAL; | |
7781 | goto out_fput; | |
7782 | } | |
7783 | ||
7784 | ctx->io_wq = ctx_attach->io_wq; | |
7785 | out_fput: | |
7786 | fdput(f); | |
7787 | return ret; | |
7788 | } | |
7789 | ||
0f212204 JA |
7790 | static int io_uring_alloc_task_context(struct task_struct *task) |
7791 | { | |
7792 | struct io_uring_task *tctx; | |
d8a6df10 | 7793 | int ret; |
0f212204 JA |
7794 | |
7795 | tctx = kmalloc(sizeof(*tctx), GFP_KERNEL); | |
7796 | if (unlikely(!tctx)) | |
7797 | return -ENOMEM; | |
7798 | ||
d8a6df10 JA |
7799 | ret = percpu_counter_init(&tctx->inflight, 0, GFP_KERNEL); |
7800 | if (unlikely(ret)) { | |
7801 | kfree(tctx); | |
7802 | return ret; | |
7803 | } | |
7804 | ||
0f212204 JA |
7805 | xa_init(&tctx->xa); |
7806 | init_waitqueue_head(&tctx->wait); | |
7807 | tctx->last = NULL; | |
fdaf083c JA |
7808 | atomic_set(&tctx->in_idle, 0); |
7809 | tctx->sqpoll = false; | |
500a373d JA |
7810 | io_init_identity(&tctx->__identity); |
7811 | tctx->identity = &tctx->__identity; | |
0f212204 JA |
7812 | task->io_uring = tctx; |
7813 | return 0; | |
7814 | } | |
7815 | ||
7816 | void __io_uring_free(struct task_struct *tsk) | |
7817 | { | |
7818 | struct io_uring_task *tctx = tsk->io_uring; | |
7819 | ||
7820 | WARN_ON_ONCE(!xa_empty(&tctx->xa)); | |
500a373d JA |
7821 | WARN_ON_ONCE(refcount_read(&tctx->identity->count) != 1); |
7822 | if (tctx->identity != &tctx->__identity) | |
7823 | kfree(tctx->identity); | |
d8a6df10 | 7824 | percpu_counter_destroy(&tctx->inflight); |
0f212204 JA |
7825 | kfree(tctx); |
7826 | tsk->io_uring = NULL; | |
7827 | } | |
7828 | ||
7e84e1c7 SG |
7829 | static int io_sq_offload_create(struct io_ring_ctx *ctx, |
7830 | struct io_uring_params *p) | |
2b188cc1 JA |
7831 | { |
7832 | int ret; | |
7833 | ||
6c271ce2 | 7834 | if (ctx->flags & IORING_SETUP_SQPOLL) { |
534ca6d6 JA |
7835 | struct io_sq_data *sqd; |
7836 | ||
3ec482d1 | 7837 | ret = -EPERM; |
ce59fc69 | 7838 | if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_NICE)) |
3ec482d1 JA |
7839 | goto err; |
7840 | ||
534ca6d6 JA |
7841 | sqd = io_get_sq_data(p); |
7842 | if (IS_ERR(sqd)) { | |
7843 | ret = PTR_ERR(sqd); | |
7844 | goto err; | |
7845 | } | |
69fb2131 | 7846 | |
534ca6d6 | 7847 | ctx->sq_data = sqd; |
69fb2131 JA |
7848 | io_sq_thread_park(sqd); |
7849 | mutex_lock(&sqd->ctx_lock); | |
7850 | list_add(&ctx->sqd_list, &sqd->ctx_new_list); | |
7851 | mutex_unlock(&sqd->ctx_lock); | |
7852 | io_sq_thread_unpark(sqd); | |
534ca6d6 | 7853 | |
917257da JA |
7854 | ctx->sq_thread_idle = msecs_to_jiffies(p->sq_thread_idle); |
7855 | if (!ctx->sq_thread_idle) | |
7856 | ctx->sq_thread_idle = HZ; | |
7857 | ||
aa06165d JA |
7858 | if (sqd->thread) |
7859 | goto done; | |
7860 | ||
6c271ce2 | 7861 | if (p->flags & IORING_SETUP_SQ_AFF) { |
44a9bd18 | 7862 | int cpu = p->sq_thread_cpu; |
6c271ce2 | 7863 | |
917257da | 7864 | ret = -EINVAL; |
44a9bd18 JA |
7865 | if (cpu >= nr_cpu_ids) |
7866 | goto err; | |
7889f44d | 7867 | if (!cpu_online(cpu)) |
917257da JA |
7868 | goto err; |
7869 | ||
69fb2131 | 7870 | sqd->thread = kthread_create_on_cpu(io_sq_thread, sqd, |
534ca6d6 | 7871 | cpu, "io_uring-sq"); |
6c271ce2 | 7872 | } else { |
69fb2131 | 7873 | sqd->thread = kthread_create(io_sq_thread, sqd, |
6c271ce2 JA |
7874 | "io_uring-sq"); |
7875 | } | |
534ca6d6 JA |
7876 | if (IS_ERR(sqd->thread)) { |
7877 | ret = PTR_ERR(sqd->thread); | |
7878 | sqd->thread = NULL; | |
6c271ce2 JA |
7879 | goto err; |
7880 | } | |
534ca6d6 | 7881 | ret = io_uring_alloc_task_context(sqd->thread); |
0f212204 JA |
7882 | if (ret) |
7883 | goto err; | |
6c271ce2 JA |
7884 | } else if (p->flags & IORING_SETUP_SQ_AFF) { |
7885 | /* Can't have SQ_AFF without SQPOLL */ | |
7886 | ret = -EINVAL; | |
7887 | goto err; | |
7888 | } | |
7889 | ||
aa06165d | 7890 | done: |
24369c2e PB |
7891 | ret = io_init_wq_offload(ctx, p); |
7892 | if (ret) | |
2b188cc1 | 7893 | goto err; |
2b188cc1 JA |
7894 | |
7895 | return 0; | |
7896 | err: | |
54a91f3b | 7897 | io_finish_async(ctx); |
2b188cc1 JA |
7898 | return ret; |
7899 | } | |
7900 | ||
7e84e1c7 SG |
7901 | static void io_sq_offload_start(struct io_ring_ctx *ctx) |
7902 | { | |
534ca6d6 JA |
7903 | struct io_sq_data *sqd = ctx->sq_data; |
7904 | ||
7905 | if ((ctx->flags & IORING_SETUP_SQPOLL) && sqd->thread) | |
7906 | wake_up_process(sqd->thread); | |
7e84e1c7 SG |
7907 | } |
7908 | ||
a087e2b5 BM |
7909 | static inline void __io_unaccount_mem(struct user_struct *user, |
7910 | unsigned long nr_pages) | |
2b188cc1 JA |
7911 | { |
7912 | atomic_long_sub(nr_pages, &user->locked_vm); | |
7913 | } | |
7914 | ||
a087e2b5 BM |
7915 | static inline int __io_account_mem(struct user_struct *user, |
7916 | unsigned long nr_pages) | |
2b188cc1 JA |
7917 | { |
7918 | unsigned long page_limit, cur_pages, new_pages; | |
7919 | ||
7920 | /* Don't allow more pages than we can safely lock */ | |
7921 | page_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT; | |
7922 | ||
7923 | do { | |
7924 | cur_pages = atomic_long_read(&user->locked_vm); | |
7925 | new_pages = cur_pages + nr_pages; | |
7926 | if (new_pages > page_limit) | |
7927 | return -ENOMEM; | |
7928 | } while (atomic_long_cmpxchg(&user->locked_vm, cur_pages, | |
7929 | new_pages) != cur_pages); | |
7930 | ||
7931 | return 0; | |
7932 | } | |
7933 | ||
2e0464d4 BM |
7934 | static void io_unaccount_mem(struct io_ring_ctx *ctx, unsigned long nr_pages, |
7935 | enum io_mem_account acct) | |
a087e2b5 | 7936 | { |
aad5d8da | 7937 | if (ctx->limit_mem) |
a087e2b5 | 7938 | __io_unaccount_mem(ctx->user, nr_pages); |
30975825 | 7939 | |
2aede0e4 | 7940 | if (ctx->mm_account) { |
2e0464d4 | 7941 | if (acct == ACCT_LOCKED) |
2aede0e4 | 7942 | ctx->mm_account->locked_vm -= nr_pages; |
2e0464d4 | 7943 | else if (acct == ACCT_PINNED) |
2aede0e4 | 7944 | atomic64_sub(nr_pages, &ctx->mm_account->pinned_vm); |
2e0464d4 | 7945 | } |
a087e2b5 BM |
7946 | } |
7947 | ||
2e0464d4 BM |
7948 | static int io_account_mem(struct io_ring_ctx *ctx, unsigned long nr_pages, |
7949 | enum io_mem_account acct) | |
a087e2b5 | 7950 | { |
30975825 BM |
7951 | int ret; |
7952 | ||
7953 | if (ctx->limit_mem) { | |
7954 | ret = __io_account_mem(ctx->user, nr_pages); | |
7955 | if (ret) | |
7956 | return ret; | |
7957 | } | |
7958 | ||
2aede0e4 | 7959 | if (ctx->mm_account) { |
2e0464d4 | 7960 | if (acct == ACCT_LOCKED) |
2aede0e4 | 7961 | ctx->mm_account->locked_vm += nr_pages; |
2e0464d4 | 7962 | else if (acct == ACCT_PINNED) |
2aede0e4 | 7963 | atomic64_add(nr_pages, &ctx->mm_account->pinned_vm); |
2e0464d4 | 7964 | } |
a087e2b5 BM |
7965 | |
7966 | return 0; | |
7967 | } | |
7968 | ||
2b188cc1 JA |
7969 | static void io_mem_free(void *ptr) |
7970 | { | |
52e04ef4 MR |
7971 | struct page *page; |
7972 | ||
7973 | if (!ptr) | |
7974 | return; | |
2b188cc1 | 7975 | |
52e04ef4 | 7976 | page = virt_to_head_page(ptr); |
2b188cc1 JA |
7977 | if (put_page_testzero(page)) |
7978 | free_compound_page(page); | |
7979 | } | |
7980 | ||
7981 | static void *io_mem_alloc(size_t size) | |
7982 | { | |
7983 | gfp_t gfp_flags = GFP_KERNEL | __GFP_ZERO | __GFP_NOWARN | __GFP_COMP | | |
7984 | __GFP_NORETRY; | |
7985 | ||
7986 | return (void *) __get_free_pages(gfp_flags, get_order(size)); | |
7987 | } | |
7988 | ||
75b28aff HV |
7989 | static unsigned long rings_size(unsigned sq_entries, unsigned cq_entries, |
7990 | size_t *sq_offset) | |
7991 | { | |
7992 | struct io_rings *rings; | |
7993 | size_t off, sq_array_size; | |
7994 | ||
7995 | off = struct_size(rings, cqes, cq_entries); | |
7996 | if (off == SIZE_MAX) | |
7997 | return SIZE_MAX; | |
7998 | ||
7999 | #ifdef CONFIG_SMP | |
8000 | off = ALIGN(off, SMP_CACHE_BYTES); | |
8001 | if (off == 0) | |
8002 | return SIZE_MAX; | |
8003 | #endif | |
8004 | ||
b36200f5 DV |
8005 | if (sq_offset) |
8006 | *sq_offset = off; | |
8007 | ||
75b28aff HV |
8008 | sq_array_size = array_size(sizeof(u32), sq_entries); |
8009 | if (sq_array_size == SIZE_MAX) | |
8010 | return SIZE_MAX; | |
8011 | ||
8012 | if (check_add_overflow(off, sq_array_size, &off)) | |
8013 | return SIZE_MAX; | |
8014 | ||
75b28aff HV |
8015 | return off; |
8016 | } | |
8017 | ||
2b188cc1 JA |
8018 | static unsigned long ring_pages(unsigned sq_entries, unsigned cq_entries) |
8019 | { | |
75b28aff | 8020 | size_t pages; |
2b188cc1 | 8021 | |
75b28aff HV |
8022 | pages = (size_t)1 << get_order( |
8023 | rings_size(sq_entries, cq_entries, NULL)); | |
8024 | pages += (size_t)1 << get_order( | |
8025 | array_size(sizeof(struct io_uring_sqe), sq_entries)); | |
2b188cc1 | 8026 | |
75b28aff | 8027 | return pages; |
2b188cc1 JA |
8028 | } |
8029 | ||
edafccee JA |
8030 | static int io_sqe_buffer_unregister(struct io_ring_ctx *ctx) |
8031 | { | |
8032 | int i, j; | |
8033 | ||
8034 | if (!ctx->user_bufs) | |
8035 | return -ENXIO; | |
8036 | ||
8037 | for (i = 0; i < ctx->nr_user_bufs; i++) { | |
8038 | struct io_mapped_ubuf *imu = &ctx->user_bufs[i]; | |
8039 | ||
8040 | for (j = 0; j < imu->nr_bvecs; j++) | |
f1f6a7dd | 8041 | unpin_user_page(imu->bvec[j].bv_page); |
edafccee | 8042 | |
de293938 JA |
8043 | if (imu->acct_pages) |
8044 | io_unaccount_mem(ctx, imu->acct_pages, ACCT_PINNED); | |
d4ef6475 | 8045 | kvfree(imu->bvec); |
edafccee JA |
8046 | imu->nr_bvecs = 0; |
8047 | } | |
8048 | ||
8049 | kfree(ctx->user_bufs); | |
8050 | ctx->user_bufs = NULL; | |
8051 | ctx->nr_user_bufs = 0; | |
8052 | return 0; | |
8053 | } | |
8054 | ||
8055 | static int io_copy_iov(struct io_ring_ctx *ctx, struct iovec *dst, | |
8056 | void __user *arg, unsigned index) | |
8057 | { | |
8058 | struct iovec __user *src; | |
8059 | ||
8060 | #ifdef CONFIG_COMPAT | |
8061 | if (ctx->compat) { | |
8062 | struct compat_iovec __user *ciovs; | |
8063 | struct compat_iovec ciov; | |
8064 | ||
8065 | ciovs = (struct compat_iovec __user *) arg; | |
8066 | if (copy_from_user(&ciov, &ciovs[index], sizeof(ciov))) | |
8067 | return -EFAULT; | |
8068 | ||
d55e5f5b | 8069 | dst->iov_base = u64_to_user_ptr((u64)ciov.iov_base); |
edafccee JA |
8070 | dst->iov_len = ciov.iov_len; |
8071 | return 0; | |
8072 | } | |
8073 | #endif | |
8074 | src = (struct iovec __user *) arg; | |
8075 | if (copy_from_user(dst, &src[index], sizeof(*dst))) | |
8076 | return -EFAULT; | |
8077 | return 0; | |
8078 | } | |
8079 | ||
de293938 JA |
8080 | /* |
8081 | * Not super efficient, but this is just a registration time. And we do cache | |
8082 | * the last compound head, so generally we'll only do a full search if we don't | |
8083 | * match that one. | |
8084 | * | |
8085 | * We check if the given compound head page has already been accounted, to | |
8086 | * avoid double accounting it. This allows us to account the full size of the | |
8087 | * page, not just the constituent pages of a huge page. | |
8088 | */ | |
8089 | static bool headpage_already_acct(struct io_ring_ctx *ctx, struct page **pages, | |
8090 | int nr_pages, struct page *hpage) | |
8091 | { | |
8092 | int i, j; | |
8093 | ||
8094 | /* check current page array */ | |
8095 | for (i = 0; i < nr_pages; i++) { | |
8096 | if (!PageCompound(pages[i])) | |
8097 | continue; | |
8098 | if (compound_head(pages[i]) == hpage) | |
8099 | return true; | |
8100 | } | |
8101 | ||
8102 | /* check previously registered pages */ | |
8103 | for (i = 0; i < ctx->nr_user_bufs; i++) { | |
8104 | struct io_mapped_ubuf *imu = &ctx->user_bufs[i]; | |
8105 | ||
8106 | for (j = 0; j < imu->nr_bvecs; j++) { | |
8107 | if (!PageCompound(imu->bvec[j].bv_page)) | |
8108 | continue; | |
8109 | if (compound_head(imu->bvec[j].bv_page) == hpage) | |
8110 | return true; | |
8111 | } | |
8112 | } | |
8113 | ||
8114 | return false; | |
8115 | } | |
8116 | ||
8117 | static int io_buffer_account_pin(struct io_ring_ctx *ctx, struct page **pages, | |
8118 | int nr_pages, struct io_mapped_ubuf *imu, | |
8119 | struct page **last_hpage) | |
8120 | { | |
8121 | int i, ret; | |
8122 | ||
8123 | for (i = 0; i < nr_pages; i++) { | |
8124 | if (!PageCompound(pages[i])) { | |
8125 | imu->acct_pages++; | |
8126 | } else { | |
8127 | struct page *hpage; | |
8128 | ||
8129 | hpage = compound_head(pages[i]); | |
8130 | if (hpage == *last_hpage) | |
8131 | continue; | |
8132 | *last_hpage = hpage; | |
8133 | if (headpage_already_acct(ctx, pages, i, hpage)) | |
8134 | continue; | |
8135 | imu->acct_pages += page_size(hpage) >> PAGE_SHIFT; | |
8136 | } | |
8137 | } | |
8138 | ||
8139 | if (!imu->acct_pages) | |
8140 | return 0; | |
8141 | ||
8142 | ret = io_account_mem(ctx, imu->acct_pages, ACCT_PINNED); | |
8143 | if (ret) | |
8144 | imu->acct_pages = 0; | |
8145 | return ret; | |
8146 | } | |
8147 | ||
edafccee JA |
8148 | static int io_sqe_buffer_register(struct io_ring_ctx *ctx, void __user *arg, |
8149 | unsigned nr_args) | |
8150 | { | |
8151 | struct vm_area_struct **vmas = NULL; | |
8152 | struct page **pages = NULL; | |
de293938 | 8153 | struct page *last_hpage = NULL; |
edafccee JA |
8154 | int i, j, got_pages = 0; |
8155 | int ret = -EINVAL; | |
8156 | ||
8157 | if (ctx->user_bufs) | |
8158 | return -EBUSY; | |
8159 | if (!nr_args || nr_args > UIO_MAXIOV) | |
8160 | return -EINVAL; | |
8161 | ||
8162 | ctx->user_bufs = kcalloc(nr_args, sizeof(struct io_mapped_ubuf), | |
8163 | GFP_KERNEL); | |
8164 | if (!ctx->user_bufs) | |
8165 | return -ENOMEM; | |
8166 | ||
8167 | for (i = 0; i < nr_args; i++) { | |
8168 | struct io_mapped_ubuf *imu = &ctx->user_bufs[i]; | |
8169 | unsigned long off, start, end, ubuf; | |
8170 | int pret, nr_pages; | |
8171 | struct iovec iov; | |
8172 | size_t size; | |
8173 | ||
8174 | ret = io_copy_iov(ctx, &iov, arg, i); | |
8175 | if (ret) | |
a278682d | 8176 | goto err; |
edafccee JA |
8177 | |
8178 | /* | |
8179 | * Don't impose further limits on the size and buffer | |
8180 | * constraints here, we'll -EINVAL later when IO is | |
8181 | * submitted if they are wrong. | |
8182 | */ | |
8183 | ret = -EFAULT; | |
8184 | if (!iov.iov_base || !iov.iov_len) | |
8185 | goto err; | |
8186 | ||
8187 | /* arbitrary limit, but we need something */ | |
8188 | if (iov.iov_len > SZ_1G) | |
8189 | goto err; | |
8190 | ||
8191 | ubuf = (unsigned long) iov.iov_base; | |
8192 | end = (ubuf + iov.iov_len + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
8193 | start = ubuf >> PAGE_SHIFT; | |
8194 | nr_pages = end - start; | |
8195 | ||
edafccee JA |
8196 | ret = 0; |
8197 | if (!pages || nr_pages > got_pages) { | |
a8c73c1a DE |
8198 | kvfree(vmas); |
8199 | kvfree(pages); | |
d4ef6475 | 8200 | pages = kvmalloc_array(nr_pages, sizeof(struct page *), |
edafccee | 8201 | GFP_KERNEL); |
d4ef6475 | 8202 | vmas = kvmalloc_array(nr_pages, |
edafccee JA |
8203 | sizeof(struct vm_area_struct *), |
8204 | GFP_KERNEL); | |
8205 | if (!pages || !vmas) { | |
8206 | ret = -ENOMEM; | |
edafccee JA |
8207 | goto err; |
8208 | } | |
8209 | got_pages = nr_pages; | |
8210 | } | |
8211 | ||
d4ef6475 | 8212 | imu->bvec = kvmalloc_array(nr_pages, sizeof(struct bio_vec), |
edafccee JA |
8213 | GFP_KERNEL); |
8214 | ret = -ENOMEM; | |
de293938 | 8215 | if (!imu->bvec) |
edafccee | 8216 | goto err; |
edafccee JA |
8217 | |
8218 | ret = 0; | |
d8ed45c5 | 8219 | mmap_read_lock(current->mm); |
2113b05d | 8220 | pret = pin_user_pages(ubuf, nr_pages, |
932f4a63 IW |
8221 | FOLL_WRITE | FOLL_LONGTERM, |
8222 | pages, vmas); | |
edafccee JA |
8223 | if (pret == nr_pages) { |
8224 | /* don't support file backed memory */ | |
8225 | for (j = 0; j < nr_pages; j++) { | |
8226 | struct vm_area_struct *vma = vmas[j]; | |
8227 | ||
8228 | if (vma->vm_file && | |
8229 | !is_file_hugepages(vma->vm_file)) { | |
8230 | ret = -EOPNOTSUPP; | |
8231 | break; | |
8232 | } | |
8233 | } | |
8234 | } else { | |
8235 | ret = pret < 0 ? pret : -EFAULT; | |
8236 | } | |
d8ed45c5 | 8237 | mmap_read_unlock(current->mm); |
edafccee JA |
8238 | if (ret) { |
8239 | /* | |
8240 | * if we did partial map, or found file backed vmas, | |
8241 | * release any pages we did get | |
8242 | */ | |
27c4d3a3 | 8243 | if (pret > 0) |
f1f6a7dd | 8244 | unpin_user_pages(pages, pret); |
de293938 JA |
8245 | kvfree(imu->bvec); |
8246 | goto err; | |
8247 | } | |
8248 | ||
8249 | ret = io_buffer_account_pin(ctx, pages, pret, imu, &last_hpage); | |
8250 | if (ret) { | |
8251 | unpin_user_pages(pages, pret); | |
d4ef6475 | 8252 | kvfree(imu->bvec); |
edafccee JA |
8253 | goto err; |
8254 | } | |
8255 | ||
8256 | off = ubuf & ~PAGE_MASK; | |
8257 | size = iov.iov_len; | |
8258 | for (j = 0; j < nr_pages; j++) { | |
8259 | size_t vec_len; | |
8260 | ||
8261 | vec_len = min_t(size_t, size, PAGE_SIZE - off); | |
8262 | imu->bvec[j].bv_page = pages[j]; | |
8263 | imu->bvec[j].bv_len = vec_len; | |
8264 | imu->bvec[j].bv_offset = off; | |
8265 | off = 0; | |
8266 | size -= vec_len; | |
8267 | } | |
8268 | /* store original address for later verification */ | |
8269 | imu->ubuf = ubuf; | |
8270 | imu->len = iov.iov_len; | |
8271 | imu->nr_bvecs = nr_pages; | |
8272 | ||
8273 | ctx->nr_user_bufs++; | |
8274 | } | |
d4ef6475 MR |
8275 | kvfree(pages); |
8276 | kvfree(vmas); | |
edafccee JA |
8277 | return 0; |
8278 | err: | |
d4ef6475 MR |
8279 | kvfree(pages); |
8280 | kvfree(vmas); | |
edafccee JA |
8281 | io_sqe_buffer_unregister(ctx); |
8282 | return ret; | |
8283 | } | |
8284 | ||
9b402849 JA |
8285 | static int io_eventfd_register(struct io_ring_ctx *ctx, void __user *arg) |
8286 | { | |
8287 | __s32 __user *fds = arg; | |
8288 | int fd; | |
8289 | ||
8290 | if (ctx->cq_ev_fd) | |
8291 | return -EBUSY; | |
8292 | ||
8293 | if (copy_from_user(&fd, fds, sizeof(*fds))) | |
8294 | return -EFAULT; | |
8295 | ||
8296 | ctx->cq_ev_fd = eventfd_ctx_fdget(fd); | |
8297 | if (IS_ERR(ctx->cq_ev_fd)) { | |
8298 | int ret = PTR_ERR(ctx->cq_ev_fd); | |
8299 | ctx->cq_ev_fd = NULL; | |
8300 | return ret; | |
8301 | } | |
8302 | ||
8303 | return 0; | |
8304 | } | |
8305 | ||
8306 | static int io_eventfd_unregister(struct io_ring_ctx *ctx) | |
8307 | { | |
8308 | if (ctx->cq_ev_fd) { | |
8309 | eventfd_ctx_put(ctx->cq_ev_fd); | |
8310 | ctx->cq_ev_fd = NULL; | |
8311 | return 0; | |
8312 | } | |
8313 | ||
8314 | return -ENXIO; | |
8315 | } | |
8316 | ||
5a2e745d JA |
8317 | static int __io_destroy_buffers(int id, void *p, void *data) |
8318 | { | |
8319 | struct io_ring_ctx *ctx = data; | |
8320 | struct io_buffer *buf = p; | |
8321 | ||
067524e9 | 8322 | __io_remove_buffers(ctx, buf, id, -1U); |
5a2e745d JA |
8323 | return 0; |
8324 | } | |
8325 | ||
8326 | static void io_destroy_buffers(struct io_ring_ctx *ctx) | |
8327 | { | |
8328 | idr_for_each(&ctx->io_buffer_idr, __io_destroy_buffers, ctx); | |
8329 | idr_destroy(&ctx->io_buffer_idr); | |
8330 | } | |
8331 | ||
2b188cc1 JA |
8332 | static void io_ring_ctx_free(struct io_ring_ctx *ctx) |
8333 | { | |
6b06314c | 8334 | io_finish_async(ctx); |
5dbcad51 | 8335 | io_sqe_buffer_unregister(ctx); |
2aede0e4 JA |
8336 | |
8337 | if (ctx->sqo_task) { | |
8338 | put_task_struct(ctx->sqo_task); | |
8339 | ctx->sqo_task = NULL; | |
8340 | mmdrop(ctx->mm_account); | |
8341 | ctx->mm_account = NULL; | |
30975825 | 8342 | } |
def596e9 | 8343 | |
91d8f519 DZ |
8344 | #ifdef CONFIG_BLK_CGROUP |
8345 | if (ctx->sqo_blkcg_css) | |
8346 | css_put(ctx->sqo_blkcg_css); | |
8347 | #endif | |
8348 | ||
6b06314c | 8349 | io_sqe_files_unregister(ctx); |
9b402849 | 8350 | io_eventfd_unregister(ctx); |
5a2e745d | 8351 | io_destroy_buffers(ctx); |
41726c9a | 8352 | idr_destroy(&ctx->personality_idr); |
def596e9 | 8353 | |
2b188cc1 | 8354 | #if defined(CONFIG_UNIX) |
355e8d26 EB |
8355 | if (ctx->ring_sock) { |
8356 | ctx->ring_sock->file = NULL; /* so that iput() is called */ | |
2b188cc1 | 8357 | sock_release(ctx->ring_sock); |
355e8d26 | 8358 | } |
2b188cc1 JA |
8359 | #endif |
8360 | ||
75b28aff | 8361 | io_mem_free(ctx->rings); |
2b188cc1 | 8362 | io_mem_free(ctx->sq_sqes); |
2b188cc1 JA |
8363 | |
8364 | percpu_ref_exit(&ctx->refs); | |
2b188cc1 | 8365 | free_uid(ctx->user); |
181e448d | 8366 | put_cred(ctx->creds); |
78076bb6 | 8367 | kfree(ctx->cancel_hash); |
0ddf92e8 | 8368 | kmem_cache_free(req_cachep, ctx->fallback_req); |
2b188cc1 JA |
8369 | kfree(ctx); |
8370 | } | |
8371 | ||
8372 | static __poll_t io_uring_poll(struct file *file, poll_table *wait) | |
8373 | { | |
8374 | struct io_ring_ctx *ctx = file->private_data; | |
8375 | __poll_t mask = 0; | |
8376 | ||
8377 | poll_wait(file, &ctx->cq_wait, wait); | |
4f7067c3 SB |
8378 | /* |
8379 | * synchronizes with barrier from wq_has_sleeper call in | |
8380 | * io_commit_cqring | |
8381 | */ | |
2b188cc1 | 8382 | smp_rmb(); |
90554200 | 8383 | if (!io_sqring_full(ctx)) |
2b188cc1 | 8384 | mask |= EPOLLOUT | EPOLLWRNORM; |
63e5d81f | 8385 | if (io_cqring_events(ctx, false)) |
2b188cc1 JA |
8386 | mask |= EPOLLIN | EPOLLRDNORM; |
8387 | ||
8388 | return mask; | |
8389 | } | |
8390 | ||
8391 | static int io_uring_fasync(int fd, struct file *file, int on) | |
8392 | { | |
8393 | struct io_ring_ctx *ctx = file->private_data; | |
8394 | ||
8395 | return fasync_helper(fd, file, on, &ctx->cq_fasync); | |
8396 | } | |
8397 | ||
071698e1 JA |
8398 | static int io_remove_personalities(int id, void *p, void *data) |
8399 | { | |
8400 | struct io_ring_ctx *ctx = data; | |
1e6fa521 | 8401 | struct io_identity *iod; |
071698e1 | 8402 | |
1e6fa521 JA |
8403 | iod = idr_remove(&ctx->personality_idr, id); |
8404 | if (iod) { | |
8405 | put_cred(iod->creds); | |
8406 | if (refcount_dec_and_test(&iod->count)) | |
8407 | kfree(iod); | |
8408 | } | |
071698e1 JA |
8409 | return 0; |
8410 | } | |
8411 | ||
85faa7b8 JA |
8412 | static void io_ring_exit_work(struct work_struct *work) |
8413 | { | |
b2edc0a7 PB |
8414 | struct io_ring_ctx *ctx = container_of(work, struct io_ring_ctx, |
8415 | exit_work); | |
85faa7b8 | 8416 | |
56952e91 JA |
8417 | /* |
8418 | * If we're doing polled IO and end up having requests being | |
8419 | * submitted async (out-of-line), then completions can come in while | |
8420 | * we're waiting for refs to drop. We need to reap these manually, | |
8421 | * as nobody else will be looking for them. | |
8422 | */ | |
b2edc0a7 | 8423 | do { |
56952e91 | 8424 | if (ctx->rings) |
e6c8aa9a | 8425 | io_cqring_overflow_flush(ctx, true, NULL, NULL); |
b2edc0a7 PB |
8426 | io_iopoll_try_reap_events(ctx); |
8427 | } while (!wait_for_completion_timeout(&ctx->ref_comp, HZ/20)); | |
85faa7b8 JA |
8428 | io_ring_ctx_free(ctx); |
8429 | } | |
8430 | ||
2b188cc1 JA |
8431 | static void io_ring_ctx_wait_and_kill(struct io_ring_ctx *ctx) |
8432 | { | |
8433 | mutex_lock(&ctx->uring_lock); | |
8434 | percpu_ref_kill(&ctx->refs); | |
8435 | mutex_unlock(&ctx->uring_lock); | |
8436 | ||
f3606e3a JA |
8437 | io_kill_timeouts(ctx, NULL); |
8438 | io_poll_remove_all(ctx, NULL); | |
561fb04a JA |
8439 | |
8440 | if (ctx->io_wq) | |
8441 | io_wq_cancel_all(ctx->io_wq); | |
8442 | ||
15dff286 JA |
8443 | /* if we failed setting up the ctx, we might not have any rings */ |
8444 | if (ctx->rings) | |
e6c8aa9a | 8445 | io_cqring_overflow_flush(ctx, true, NULL, NULL); |
b2edc0a7 | 8446 | io_iopoll_try_reap_events(ctx); |
071698e1 | 8447 | idr_for_each(&ctx->personality_idr, io_remove_personalities, ctx); |
309fc03a JA |
8448 | |
8449 | /* | |
8450 | * Do this upfront, so we won't have a grace period where the ring | |
8451 | * is closed but resources aren't reaped yet. This can cause | |
8452 | * spurious failure in setting up a new ring. | |
8453 | */ | |
760618f7 JA |
8454 | io_unaccount_mem(ctx, ring_pages(ctx->sq_entries, ctx->cq_entries), |
8455 | ACCT_LOCKED); | |
309fc03a | 8456 | |
85faa7b8 | 8457 | INIT_WORK(&ctx->exit_work, io_ring_exit_work); |
fc666777 JA |
8458 | /* |
8459 | * Use system_unbound_wq to avoid spawning tons of event kworkers | |
8460 | * if we're exiting a ton of rings at the same time. It just adds | |
8461 | * noise and overhead, there's no discernable change in runtime | |
8462 | * over using system_wq. | |
8463 | */ | |
8464 | queue_work(system_unbound_wq, &ctx->exit_work); | |
2b188cc1 JA |
8465 | } |
8466 | ||
8467 | static int io_uring_release(struct inode *inode, struct file *file) | |
8468 | { | |
8469 | struct io_ring_ctx *ctx = file->private_data; | |
8470 | ||
8471 | file->private_data = NULL; | |
8472 | io_ring_ctx_wait_and_kill(ctx); | |
8473 | return 0; | |
8474 | } | |
8475 | ||
67c4d9e6 PB |
8476 | static bool io_wq_files_match(struct io_wq_work *work, void *data) |
8477 | { | |
8478 | struct files_struct *files = data; | |
8479 | ||
dfead8a8 | 8480 | return !files || ((work->flags & IO_WQ_WORK_FILES) && |
98447d65 | 8481 | work->identity->files == files); |
67c4d9e6 PB |
8482 | } |
8483 | ||
f254ac04 JA |
8484 | /* |
8485 | * Returns true if 'preq' is the link parent of 'req' | |
8486 | */ | |
8487 | static bool io_match_link(struct io_kiocb *preq, struct io_kiocb *req) | |
8488 | { | |
8489 | struct io_kiocb *link; | |
8490 | ||
8491 | if (!(preq->flags & REQ_F_LINK_HEAD)) | |
8492 | return false; | |
8493 | ||
8494 | list_for_each_entry(link, &preq->link_list, link_list) { | |
8495 | if (link == req) | |
8496 | return true; | |
8497 | } | |
8498 | ||
8499 | return false; | |
8500 | } | |
8501 | ||
8502 | /* | |
8503 | * We're looking to cancel 'req' because it's holding on to our files, but | |
8504 | * 'req' could be a link to another request. See if it is, and cancel that | |
8505 | * parent request if so. | |
8506 | */ | |
8507 | static bool io_poll_remove_link(struct io_ring_ctx *ctx, struct io_kiocb *req) | |
8508 | { | |
8509 | struct hlist_node *tmp; | |
8510 | struct io_kiocb *preq; | |
8511 | bool found = false; | |
8512 | int i; | |
8513 | ||
8514 | spin_lock_irq(&ctx->completion_lock); | |
8515 | for (i = 0; i < (1U << ctx->cancel_hash_bits); i++) { | |
8516 | struct hlist_head *list; | |
8517 | ||
8518 | list = &ctx->cancel_hash[i]; | |
8519 | hlist_for_each_entry_safe(preq, tmp, list, hash_node) { | |
8520 | found = io_match_link(preq, req); | |
8521 | if (found) { | |
8522 | io_poll_remove_one(preq); | |
8523 | break; | |
8524 | } | |
8525 | } | |
8526 | } | |
8527 | spin_unlock_irq(&ctx->completion_lock); | |
8528 | return found; | |
8529 | } | |
8530 | ||
8531 | static bool io_timeout_remove_link(struct io_ring_ctx *ctx, | |
8532 | struct io_kiocb *req) | |
8533 | { | |
8534 | struct io_kiocb *preq; | |
8535 | bool found = false; | |
8536 | ||
8537 | spin_lock_irq(&ctx->completion_lock); | |
8538 | list_for_each_entry(preq, &ctx->timeout_list, timeout.list) { | |
8539 | found = io_match_link(preq, req); | |
8540 | if (found) { | |
8541 | __io_timeout_cancel(preq); | |
8542 | break; | |
8543 | } | |
8544 | } | |
8545 | spin_unlock_irq(&ctx->completion_lock); | |
8546 | return found; | |
8547 | } | |
8548 | ||
b711d4ea JA |
8549 | static bool io_cancel_link_cb(struct io_wq_work *work, void *data) |
8550 | { | |
9a472ef7 PB |
8551 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); |
8552 | bool ret; | |
8553 | ||
8554 | if (req->flags & REQ_F_LINK_TIMEOUT) { | |
8555 | unsigned long flags; | |
8556 | struct io_ring_ctx *ctx = req->ctx; | |
8557 | ||
8558 | /* protect against races with linked timeouts */ | |
8559 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
8560 | ret = io_match_link(req, data); | |
8561 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
8562 | } else { | |
8563 | ret = io_match_link(req, data); | |
8564 | } | |
8565 | return ret; | |
b711d4ea JA |
8566 | } |
8567 | ||
8568 | static void io_attempt_cancel(struct io_ring_ctx *ctx, struct io_kiocb *req) | |
8569 | { | |
8570 | enum io_wq_cancel cret; | |
8571 | ||
8572 | /* cancel this particular work, if it's running */ | |
8573 | cret = io_wq_cancel_work(ctx->io_wq, &req->work); | |
8574 | if (cret != IO_WQ_CANCEL_NOTFOUND) | |
8575 | return; | |
8576 | ||
8577 | /* find links that hold this pending, cancel those */ | |
8578 | cret = io_wq_cancel_cb(ctx->io_wq, io_cancel_link_cb, req, true); | |
8579 | if (cret != IO_WQ_CANCEL_NOTFOUND) | |
8580 | return; | |
8581 | ||
8582 | /* if we have a poll link holding this pending, cancel that */ | |
8583 | if (io_poll_remove_link(ctx, req)) | |
8584 | return; | |
8585 | ||
8586 | /* final option, timeout link is holding this req pending */ | |
8587 | io_timeout_remove_link(ctx, req); | |
8588 | } | |
8589 | ||
b7ddce3c | 8590 | static void io_cancel_defer_files(struct io_ring_ctx *ctx, |
ef9865a4 | 8591 | struct task_struct *task, |
b7ddce3c PB |
8592 | struct files_struct *files) |
8593 | { | |
8594 | struct io_defer_entry *de = NULL; | |
8595 | LIST_HEAD(list); | |
8596 | ||
8597 | spin_lock_irq(&ctx->completion_lock); | |
8598 | list_for_each_entry_reverse(de, &ctx->defer_list, list) { | |
ef9865a4 PB |
8599 | if (io_task_match(de->req, task) && |
8600 | io_match_files(de->req, files)) { | |
b7ddce3c PB |
8601 | list_cut_position(&list, &ctx->defer_list, &de->list); |
8602 | break; | |
8603 | } | |
8604 | } | |
8605 | spin_unlock_irq(&ctx->completion_lock); | |
8606 | ||
8607 | while (!list_empty(&list)) { | |
8608 | de = list_first_entry(&list, struct io_defer_entry, list); | |
8609 | list_del_init(&de->list); | |
8610 | req_set_fail_links(de->req); | |
8611 | io_put_req(de->req); | |
8612 | io_req_complete(de->req, -ECANCELED); | |
8613 | kfree(de); | |
8614 | } | |
8615 | } | |
8616 | ||
76e1b642 JA |
8617 | /* |
8618 | * Returns true if we found and killed one or more files pinning requests | |
8619 | */ | |
8620 | static bool io_uring_cancel_files(struct io_ring_ctx *ctx, | |
fcb323cc JA |
8621 | struct files_struct *files) |
8622 | { | |
67c4d9e6 | 8623 | if (list_empty_careful(&ctx->inflight_list)) |
76e1b642 | 8624 | return false; |
67c4d9e6 PB |
8625 | |
8626 | /* cancel all at once, should be faster than doing it one by one*/ | |
8627 | io_wq_cancel_cb(ctx->io_wq, io_wq_files_match, files, true); | |
8628 | ||
fcb323cc | 8629 | while (!list_empty_careful(&ctx->inflight_list)) { |
d8f1b971 XW |
8630 | struct io_kiocb *cancel_req = NULL, *req; |
8631 | DEFINE_WAIT(wait); | |
fcb323cc JA |
8632 | |
8633 | spin_lock_irq(&ctx->inflight_lock); | |
8634 | list_for_each_entry(req, &ctx->inflight_list, inflight_entry) { | |
dfead8a8 | 8635 | if (files && (req->work.flags & IO_WQ_WORK_FILES) && |
98447d65 | 8636 | req->work.identity->files != files) |
768134d4 JA |
8637 | continue; |
8638 | /* req is being completed, ignore */ | |
8639 | if (!refcount_inc_not_zero(&req->refs)) | |
8640 | continue; | |
8641 | cancel_req = req; | |
8642 | break; | |
fcb323cc | 8643 | } |
768134d4 | 8644 | if (cancel_req) |
fcb323cc | 8645 | prepare_to_wait(&ctx->inflight_wait, &wait, |
768134d4 | 8646 | TASK_UNINTERRUPTIBLE); |
fcb323cc JA |
8647 | spin_unlock_irq(&ctx->inflight_lock); |
8648 | ||
768134d4 JA |
8649 | /* We need to keep going until we don't find a matching req */ |
8650 | if (!cancel_req) | |
fcb323cc | 8651 | break; |
bb175342 PB |
8652 | /* cancel this request, or head link requests */ |
8653 | io_attempt_cancel(ctx, cancel_req); | |
8654 | io_put_req(cancel_req); | |
6200b0ae JA |
8655 | /* cancellations _may_ trigger task work */ |
8656 | io_run_task_work(); | |
fcb323cc | 8657 | schedule(); |
d8f1b971 | 8658 | finish_wait(&ctx->inflight_wait, &wait); |
fcb323cc | 8659 | } |
76e1b642 JA |
8660 | |
8661 | return true; | |
fcb323cc JA |
8662 | } |
8663 | ||
801dd57b | 8664 | static bool io_cancel_task_cb(struct io_wq_work *work, void *data) |
44e728b8 | 8665 | { |
801dd57b PB |
8666 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); |
8667 | struct task_struct *task = data; | |
44e728b8 | 8668 | |
f3606e3a | 8669 | return io_task_match(req, task); |
44e728b8 PB |
8670 | } |
8671 | ||
0f212204 JA |
8672 | static bool __io_uring_cancel_task_requests(struct io_ring_ctx *ctx, |
8673 | struct task_struct *task, | |
8674 | struct files_struct *files) | |
8675 | { | |
8676 | bool ret; | |
8677 | ||
8678 | ret = io_uring_cancel_files(ctx, files); | |
8679 | if (!files) { | |
8680 | enum io_wq_cancel cret; | |
8681 | ||
8682 | cret = io_wq_cancel_cb(ctx->io_wq, io_cancel_task_cb, task, true); | |
8683 | if (cret != IO_WQ_CANCEL_NOTFOUND) | |
8684 | ret = true; | |
8685 | ||
8686 | /* SQPOLL thread does its own polling */ | |
8687 | if (!(ctx->flags & IORING_SETUP_SQPOLL)) { | |
8688 | while (!list_empty_careful(&ctx->iopoll_list)) { | |
8689 | io_iopoll_try_reap_events(ctx); | |
8690 | ret = true; | |
8691 | } | |
8692 | } | |
8693 | ||
8694 | ret |= io_poll_remove_all(ctx, task); | |
8695 | ret |= io_kill_timeouts(ctx, task); | |
8696 | } | |
8697 | ||
8698 | return ret; | |
8699 | } | |
8700 | ||
8701 | /* | |
8702 | * We need to iteratively cancel requests, in case a request has dependent | |
8703 | * hard links. These persist even for failure of cancelations, hence keep | |
8704 | * looping until none are found. | |
8705 | */ | |
8706 | static void io_uring_cancel_task_requests(struct io_ring_ctx *ctx, | |
8707 | struct files_struct *files) | |
8708 | { | |
8709 | struct task_struct *task = current; | |
8710 | ||
fdaf083c | 8711 | if ((ctx->flags & IORING_SETUP_SQPOLL) && ctx->sq_data) { |
534ca6d6 | 8712 | task = ctx->sq_data->thread; |
fdaf083c JA |
8713 | atomic_inc(&task->io_uring->in_idle); |
8714 | io_sq_thread_park(ctx->sq_data); | |
8715 | } | |
0f212204 | 8716 | |
ef9865a4 PB |
8717 | if (files) |
8718 | io_cancel_defer_files(ctx, NULL, files); | |
8719 | else | |
8720 | io_cancel_defer_files(ctx, task, NULL); | |
8721 | ||
0f212204 JA |
8722 | io_cqring_overflow_flush(ctx, true, task, files); |
8723 | ||
8724 | while (__io_uring_cancel_task_requests(ctx, task, files)) { | |
8725 | io_run_task_work(); | |
8726 | cond_resched(); | |
8727 | } | |
fdaf083c JA |
8728 | |
8729 | if ((ctx->flags & IORING_SETUP_SQPOLL) && ctx->sq_data) { | |
8730 | atomic_dec(&task->io_uring->in_idle); | |
8731 | /* | |
8732 | * If the files that are going away are the ones in the thread | |
8733 | * identity, clear them out. | |
8734 | */ | |
8735 | if (task->io_uring->identity->files == files) | |
8736 | task->io_uring->identity->files = NULL; | |
8737 | io_sq_thread_unpark(ctx->sq_data); | |
8738 | } | |
0f212204 JA |
8739 | } |
8740 | ||
8741 | /* | |
8742 | * Note that this task has used io_uring. We use it for cancelation purposes. | |
8743 | */ | |
fdaf083c | 8744 | static int io_uring_add_task_file(struct io_ring_ctx *ctx, struct file *file) |
0f212204 | 8745 | { |
236434c3 MWO |
8746 | struct io_uring_task *tctx = current->io_uring; |
8747 | ||
8748 | if (unlikely(!tctx)) { | |
0f212204 JA |
8749 | int ret; |
8750 | ||
8751 | ret = io_uring_alloc_task_context(current); | |
8752 | if (unlikely(ret)) | |
8753 | return ret; | |
236434c3 | 8754 | tctx = current->io_uring; |
0f212204 | 8755 | } |
236434c3 MWO |
8756 | if (tctx->last != file) { |
8757 | void *old = xa_load(&tctx->xa, (unsigned long)file); | |
0f212204 | 8758 | |
236434c3 | 8759 | if (!old) { |
0f212204 | 8760 | get_file(file); |
236434c3 | 8761 | xa_store(&tctx->xa, (unsigned long)file, file, GFP_KERNEL); |
0f212204 | 8762 | } |
236434c3 | 8763 | tctx->last = file; |
0f212204 JA |
8764 | } |
8765 | ||
fdaf083c JA |
8766 | /* |
8767 | * This is race safe in that the task itself is doing this, hence it | |
8768 | * cannot be going through the exit/cancel paths at the same time. | |
8769 | * This cannot be modified while exit/cancel is running. | |
8770 | */ | |
8771 | if (!tctx->sqpoll && (ctx->flags & IORING_SETUP_SQPOLL)) | |
8772 | tctx->sqpoll = true; | |
8773 | ||
0f212204 JA |
8774 | return 0; |
8775 | } | |
8776 | ||
8777 | /* | |
8778 | * Remove this io_uring_file -> task mapping. | |
8779 | */ | |
8780 | static void io_uring_del_task_file(struct file *file) | |
8781 | { | |
8782 | struct io_uring_task *tctx = current->io_uring; | |
0f212204 JA |
8783 | |
8784 | if (tctx->last == file) | |
8785 | tctx->last = NULL; | |
5e2ed8c4 | 8786 | file = xa_erase(&tctx->xa, (unsigned long)file); |
0f212204 JA |
8787 | if (file) |
8788 | fput(file); | |
8789 | } | |
8790 | ||
0f212204 JA |
8791 | /* |
8792 | * Drop task note for this file if we're the only ones that hold it after | |
8793 | * pending fput() | |
8794 | */ | |
c8fb20b5 | 8795 | static void io_uring_attempt_task_drop(struct file *file) |
0f212204 JA |
8796 | { |
8797 | if (!current->io_uring) | |
8798 | return; | |
8799 | /* | |
8800 | * fput() is pending, will be 2 if the only other ref is our potential | |
8801 | * task file note. If the task is exiting, drop regardless of count. | |
8802 | */ | |
c8fb20b5 PB |
8803 | if (fatal_signal_pending(current) || (current->flags & PF_EXITING) || |
8804 | atomic_long_read(&file->f_count) == 2) | |
8805 | io_uring_del_task_file(file); | |
0f212204 JA |
8806 | } |
8807 | ||
8808 | void __io_uring_files_cancel(struct files_struct *files) | |
8809 | { | |
8810 | struct io_uring_task *tctx = current->io_uring; | |
ce765372 MWO |
8811 | struct file *file; |
8812 | unsigned long index; | |
0f212204 JA |
8813 | |
8814 | /* make sure overflow events are dropped */ | |
fdaf083c | 8815 | atomic_inc(&tctx->in_idle); |
0f212204 | 8816 | |
ce765372 MWO |
8817 | xa_for_each(&tctx->xa, index, file) { |
8818 | struct io_ring_ctx *ctx = file->private_data; | |
0f212204 JA |
8819 | |
8820 | io_uring_cancel_task_requests(ctx, files); | |
8821 | if (files) | |
8822 | io_uring_del_task_file(file); | |
ce765372 | 8823 | } |
fdaf083c JA |
8824 | |
8825 | atomic_dec(&tctx->in_idle); | |
8826 | } | |
8827 | ||
8828 | static s64 tctx_inflight(struct io_uring_task *tctx) | |
8829 | { | |
8830 | unsigned long index; | |
8831 | struct file *file; | |
8832 | s64 inflight; | |
8833 | ||
8834 | inflight = percpu_counter_sum(&tctx->inflight); | |
8835 | if (!tctx->sqpoll) | |
8836 | return inflight; | |
8837 | ||
8838 | /* | |
8839 | * If we have SQPOLL rings, then we need to iterate and find them, and | |
8840 | * add the pending count for those. | |
8841 | */ | |
8842 | xa_for_each(&tctx->xa, index, file) { | |
8843 | struct io_ring_ctx *ctx = file->private_data; | |
8844 | ||
8845 | if (ctx->flags & IORING_SETUP_SQPOLL) { | |
8846 | struct io_uring_task *__tctx = ctx->sqo_task->io_uring; | |
8847 | ||
8848 | inflight += percpu_counter_sum(&__tctx->inflight); | |
8849 | } | |
8850 | } | |
8851 | ||
8852 | return inflight; | |
0f212204 JA |
8853 | } |
8854 | ||
0f212204 JA |
8855 | /* |
8856 | * Find any io_uring fd that this task has registered or done IO on, and cancel | |
8857 | * requests. | |
8858 | */ | |
8859 | void __io_uring_task_cancel(void) | |
8860 | { | |
8861 | struct io_uring_task *tctx = current->io_uring; | |
8862 | DEFINE_WAIT(wait); | |
d8a6df10 | 8863 | s64 inflight; |
0f212204 JA |
8864 | |
8865 | /* make sure overflow events are dropped */ | |
fdaf083c | 8866 | atomic_inc(&tctx->in_idle); |
0f212204 | 8867 | |
d8a6df10 | 8868 | do { |
0f212204 | 8869 | /* read completions before cancelations */ |
fdaf083c | 8870 | inflight = tctx_inflight(tctx); |
d8a6df10 JA |
8871 | if (!inflight) |
8872 | break; | |
0f212204 JA |
8873 | __io_uring_files_cancel(NULL); |
8874 | ||
8875 | prepare_to_wait(&tctx->wait, &wait, TASK_UNINTERRUPTIBLE); | |
8876 | ||
8877 | /* | |
8878 | * If we've seen completions, retry. This avoids a race where | |
8879 | * a completion comes in before we did prepare_to_wait(). | |
8880 | */ | |
fdaf083c | 8881 | if (inflight != tctx_inflight(tctx)) |
0f212204 | 8882 | continue; |
0f212204 | 8883 | schedule(); |
d8a6df10 | 8884 | } while (1); |
0f212204 JA |
8885 | |
8886 | finish_wait(&tctx->wait, &wait); | |
fdaf083c | 8887 | atomic_dec(&tctx->in_idle); |
44e728b8 PB |
8888 | } |
8889 | ||
fcb323cc JA |
8890 | static int io_uring_flush(struct file *file, void *data) |
8891 | { | |
c8fb20b5 | 8892 | io_uring_attempt_task_drop(file); |
fcb323cc JA |
8893 | return 0; |
8894 | } | |
8895 | ||
6c5c240e RP |
8896 | static void *io_uring_validate_mmap_request(struct file *file, |
8897 | loff_t pgoff, size_t sz) | |
2b188cc1 | 8898 | { |
2b188cc1 | 8899 | struct io_ring_ctx *ctx = file->private_data; |
6c5c240e | 8900 | loff_t offset = pgoff << PAGE_SHIFT; |
2b188cc1 JA |
8901 | struct page *page; |
8902 | void *ptr; | |
8903 | ||
8904 | switch (offset) { | |
8905 | case IORING_OFF_SQ_RING: | |
75b28aff HV |
8906 | case IORING_OFF_CQ_RING: |
8907 | ptr = ctx->rings; | |
2b188cc1 JA |
8908 | break; |
8909 | case IORING_OFF_SQES: | |
8910 | ptr = ctx->sq_sqes; | |
8911 | break; | |
2b188cc1 | 8912 | default: |
6c5c240e | 8913 | return ERR_PTR(-EINVAL); |
2b188cc1 JA |
8914 | } |
8915 | ||
8916 | page = virt_to_head_page(ptr); | |
a50b854e | 8917 | if (sz > page_size(page)) |
6c5c240e RP |
8918 | return ERR_PTR(-EINVAL); |
8919 | ||
8920 | return ptr; | |
8921 | } | |
8922 | ||
8923 | #ifdef CONFIG_MMU | |
8924 | ||
8925 | static int io_uring_mmap(struct file *file, struct vm_area_struct *vma) | |
8926 | { | |
8927 | size_t sz = vma->vm_end - vma->vm_start; | |
8928 | unsigned long pfn; | |
8929 | void *ptr; | |
8930 | ||
8931 | ptr = io_uring_validate_mmap_request(file, vma->vm_pgoff, sz); | |
8932 | if (IS_ERR(ptr)) | |
8933 | return PTR_ERR(ptr); | |
2b188cc1 JA |
8934 | |
8935 | pfn = virt_to_phys(ptr) >> PAGE_SHIFT; | |
8936 | return remap_pfn_range(vma, vma->vm_start, pfn, sz, vma->vm_page_prot); | |
8937 | } | |
8938 | ||
6c5c240e RP |
8939 | #else /* !CONFIG_MMU */ |
8940 | ||
8941 | static int io_uring_mmap(struct file *file, struct vm_area_struct *vma) | |
8942 | { | |
8943 | return vma->vm_flags & (VM_SHARED | VM_MAYSHARE) ? 0 : -EINVAL; | |
8944 | } | |
8945 | ||
8946 | static unsigned int io_uring_nommu_mmap_capabilities(struct file *file) | |
8947 | { | |
8948 | return NOMMU_MAP_DIRECT | NOMMU_MAP_READ | NOMMU_MAP_WRITE; | |
8949 | } | |
8950 | ||
8951 | static unsigned long io_uring_nommu_get_unmapped_area(struct file *file, | |
8952 | unsigned long addr, unsigned long len, | |
8953 | unsigned long pgoff, unsigned long flags) | |
8954 | { | |
8955 | void *ptr; | |
8956 | ||
8957 | ptr = io_uring_validate_mmap_request(file, pgoff, len); | |
8958 | if (IS_ERR(ptr)) | |
8959 | return PTR_ERR(ptr); | |
8960 | ||
8961 | return (unsigned long) ptr; | |
8962 | } | |
8963 | ||
8964 | #endif /* !CONFIG_MMU */ | |
8965 | ||
90554200 JA |
8966 | static void io_sqpoll_wait_sq(struct io_ring_ctx *ctx) |
8967 | { | |
8968 | DEFINE_WAIT(wait); | |
8969 | ||
8970 | do { | |
8971 | if (!io_sqring_full(ctx)) | |
8972 | break; | |
8973 | ||
8974 | prepare_to_wait(&ctx->sqo_sq_wait, &wait, TASK_INTERRUPTIBLE); | |
8975 | ||
8976 | if (!io_sqring_full(ctx)) | |
8977 | break; | |
8978 | ||
8979 | schedule(); | |
8980 | } while (!signal_pending(current)); | |
8981 | ||
8982 | finish_wait(&ctx->sqo_sq_wait, &wait); | |
8983 | } | |
8984 | ||
2b188cc1 JA |
8985 | SYSCALL_DEFINE6(io_uring_enter, unsigned int, fd, u32, to_submit, |
8986 | u32, min_complete, u32, flags, const sigset_t __user *, sig, | |
8987 | size_t, sigsz) | |
8988 | { | |
8989 | struct io_ring_ctx *ctx; | |
8990 | long ret = -EBADF; | |
8991 | int submitted = 0; | |
8992 | struct fd f; | |
8993 | ||
4c6e277c | 8994 | io_run_task_work(); |
b41e9852 | 8995 | |
90554200 JA |
8996 | if (flags & ~(IORING_ENTER_GETEVENTS | IORING_ENTER_SQ_WAKEUP | |
8997 | IORING_ENTER_SQ_WAIT)) | |
2b188cc1 JA |
8998 | return -EINVAL; |
8999 | ||
9000 | f = fdget(fd); | |
9001 | if (!f.file) | |
9002 | return -EBADF; | |
9003 | ||
9004 | ret = -EOPNOTSUPP; | |
9005 | if (f.file->f_op != &io_uring_fops) | |
9006 | goto out_fput; | |
9007 | ||
9008 | ret = -ENXIO; | |
9009 | ctx = f.file->private_data; | |
9010 | if (!percpu_ref_tryget(&ctx->refs)) | |
9011 | goto out_fput; | |
9012 | ||
7e84e1c7 SG |
9013 | ret = -EBADFD; |
9014 | if (ctx->flags & IORING_SETUP_R_DISABLED) | |
9015 | goto out; | |
9016 | ||
6c271ce2 JA |
9017 | /* |
9018 | * For SQ polling, the thread will do all submissions and completions. | |
9019 | * Just return the requested submit count, and wake the thread if | |
9020 | * we were asked to. | |
9021 | */ | |
b2a9eada | 9022 | ret = 0; |
6c271ce2 | 9023 | if (ctx->flags & IORING_SETUP_SQPOLL) { |
c1edbf5f | 9024 | if (!list_empty_careful(&ctx->cq_overflow_list)) |
e6c8aa9a | 9025 | io_cqring_overflow_flush(ctx, false, NULL, NULL); |
6c271ce2 | 9026 | if (flags & IORING_ENTER_SQ_WAKEUP) |
534ca6d6 | 9027 | wake_up(&ctx->sq_data->wait); |
90554200 JA |
9028 | if (flags & IORING_ENTER_SQ_WAIT) |
9029 | io_sqpoll_wait_sq(ctx); | |
6c271ce2 | 9030 | submitted = to_submit; |
b2a9eada | 9031 | } else if (to_submit) { |
fdaf083c | 9032 | ret = io_uring_add_task_file(ctx, f.file); |
0f212204 JA |
9033 | if (unlikely(ret)) |
9034 | goto out; | |
2b188cc1 | 9035 | mutex_lock(&ctx->uring_lock); |
0f212204 | 9036 | submitted = io_submit_sqes(ctx, to_submit); |
2b188cc1 | 9037 | mutex_unlock(&ctx->uring_lock); |
7c504e65 PB |
9038 | |
9039 | if (submitted != to_submit) | |
9040 | goto out; | |
2b188cc1 JA |
9041 | } |
9042 | if (flags & IORING_ENTER_GETEVENTS) { | |
9043 | min_complete = min(min_complete, ctx->cq_entries); | |
9044 | ||
32b2244a XW |
9045 | /* |
9046 | * When SETUP_IOPOLL and SETUP_SQPOLL are both enabled, user | |
9047 | * space applications don't need to do io completion events | |
9048 | * polling again, they can rely on io_sq_thread to do polling | |
9049 | * work, which can reduce cpu usage and uring_lock contention. | |
9050 | */ | |
9051 | if (ctx->flags & IORING_SETUP_IOPOLL && | |
9052 | !(ctx->flags & IORING_SETUP_SQPOLL)) { | |
7668b92a | 9053 | ret = io_iopoll_check(ctx, min_complete); |
def596e9 JA |
9054 | } else { |
9055 | ret = io_cqring_wait(ctx, min_complete, sig, sigsz); | |
9056 | } | |
2b188cc1 JA |
9057 | } |
9058 | ||
7c504e65 | 9059 | out: |
6805b32e | 9060 | percpu_ref_put(&ctx->refs); |
2b188cc1 JA |
9061 | out_fput: |
9062 | fdput(f); | |
9063 | return submitted ? submitted : ret; | |
9064 | } | |
9065 | ||
bebdb65e | 9066 | #ifdef CONFIG_PROC_FS |
87ce955b JA |
9067 | static int io_uring_show_cred(int id, void *p, void *data) |
9068 | { | |
6b47ab81 JA |
9069 | struct io_identity *iod = p; |
9070 | const struct cred *cred = iod->creds; | |
87ce955b JA |
9071 | struct seq_file *m = data; |
9072 | struct user_namespace *uns = seq_user_ns(m); | |
9073 | struct group_info *gi; | |
9074 | kernel_cap_t cap; | |
9075 | unsigned __capi; | |
9076 | int g; | |
9077 | ||
9078 | seq_printf(m, "%5d\n", id); | |
9079 | seq_put_decimal_ull(m, "\tUid:\t", from_kuid_munged(uns, cred->uid)); | |
9080 | seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->euid)); | |
9081 | seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->suid)); | |
9082 | seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->fsuid)); | |
9083 | seq_put_decimal_ull(m, "\n\tGid:\t", from_kgid_munged(uns, cred->gid)); | |
9084 | seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->egid)); | |
9085 | seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->sgid)); | |
9086 | seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->fsgid)); | |
9087 | seq_puts(m, "\n\tGroups:\t"); | |
9088 | gi = cred->group_info; | |
9089 | for (g = 0; g < gi->ngroups; g++) { | |
9090 | seq_put_decimal_ull(m, g ? " " : "", | |
9091 | from_kgid_munged(uns, gi->gid[g])); | |
9092 | } | |
9093 | seq_puts(m, "\n\tCapEff:\t"); | |
9094 | cap = cred->cap_effective; | |
9095 | CAP_FOR_EACH_U32(__capi) | |
9096 | seq_put_hex_ll(m, NULL, cap.cap[CAP_LAST_U32 - __capi], 8); | |
9097 | seq_putc(m, '\n'); | |
9098 | return 0; | |
9099 | } | |
9100 | ||
9101 | static void __io_uring_show_fdinfo(struct io_ring_ctx *ctx, struct seq_file *m) | |
9102 | { | |
dbbe9c64 | 9103 | struct io_sq_data *sq = NULL; |
fad8e0de | 9104 | bool has_lock; |
87ce955b JA |
9105 | int i; |
9106 | ||
fad8e0de JA |
9107 | /* |
9108 | * Avoid ABBA deadlock between the seq lock and the io_uring mutex, | |
9109 | * since fdinfo case grabs it in the opposite direction of normal use | |
9110 | * cases. If we fail to get the lock, we just don't iterate any | |
9111 | * structures that could be going away outside the io_uring mutex. | |
9112 | */ | |
9113 | has_lock = mutex_trylock(&ctx->uring_lock); | |
9114 | ||
dbbe9c64 JQ |
9115 | if (has_lock && (ctx->flags & IORING_SETUP_SQPOLL)) |
9116 | sq = ctx->sq_data; | |
9117 | ||
9118 | seq_printf(m, "SqThread:\t%d\n", sq ? task_pid_nr(sq->thread) : -1); | |
9119 | seq_printf(m, "SqThreadCpu:\t%d\n", sq ? task_cpu(sq->thread) : -1); | |
87ce955b | 9120 | seq_printf(m, "UserFiles:\t%u\n", ctx->nr_user_files); |
fad8e0de | 9121 | for (i = 0; has_lock && i < ctx->nr_user_files; i++) { |
87ce955b JA |
9122 | struct fixed_file_table *table; |
9123 | struct file *f; | |
9124 | ||
9125 | table = &ctx->file_data->table[i >> IORING_FILE_TABLE_SHIFT]; | |
9126 | f = table->files[i & IORING_FILE_TABLE_MASK]; | |
9127 | if (f) | |
9128 | seq_printf(m, "%5u: %s\n", i, file_dentry(f)->d_iname); | |
9129 | else | |
9130 | seq_printf(m, "%5u: <none>\n", i); | |
9131 | } | |
9132 | seq_printf(m, "UserBufs:\t%u\n", ctx->nr_user_bufs); | |
fad8e0de | 9133 | for (i = 0; has_lock && i < ctx->nr_user_bufs; i++) { |
87ce955b JA |
9134 | struct io_mapped_ubuf *buf = &ctx->user_bufs[i]; |
9135 | ||
9136 | seq_printf(m, "%5u: 0x%llx/%u\n", i, buf->ubuf, | |
9137 | (unsigned int) buf->len); | |
9138 | } | |
fad8e0de | 9139 | if (has_lock && !idr_is_empty(&ctx->personality_idr)) { |
87ce955b JA |
9140 | seq_printf(m, "Personalities:\n"); |
9141 | idr_for_each(&ctx->personality_idr, io_uring_show_cred, m); | |
9142 | } | |
d7718a9d JA |
9143 | seq_printf(m, "PollList:\n"); |
9144 | spin_lock_irq(&ctx->completion_lock); | |
9145 | for (i = 0; i < (1U << ctx->cancel_hash_bits); i++) { | |
9146 | struct hlist_head *list = &ctx->cancel_hash[i]; | |
9147 | struct io_kiocb *req; | |
9148 | ||
9149 | hlist_for_each_entry(req, list, hash_node) | |
9150 | seq_printf(m, " op=%d, task_works=%d\n", req->opcode, | |
9151 | req->task->task_works != NULL); | |
9152 | } | |
9153 | spin_unlock_irq(&ctx->completion_lock); | |
fad8e0de JA |
9154 | if (has_lock) |
9155 | mutex_unlock(&ctx->uring_lock); | |
87ce955b JA |
9156 | } |
9157 | ||
9158 | static void io_uring_show_fdinfo(struct seq_file *m, struct file *f) | |
9159 | { | |
9160 | struct io_ring_ctx *ctx = f->private_data; | |
9161 | ||
9162 | if (percpu_ref_tryget(&ctx->refs)) { | |
9163 | __io_uring_show_fdinfo(ctx, m); | |
9164 | percpu_ref_put(&ctx->refs); | |
9165 | } | |
9166 | } | |
bebdb65e | 9167 | #endif |
87ce955b | 9168 | |
2b188cc1 JA |
9169 | static const struct file_operations io_uring_fops = { |
9170 | .release = io_uring_release, | |
fcb323cc | 9171 | .flush = io_uring_flush, |
2b188cc1 | 9172 | .mmap = io_uring_mmap, |
6c5c240e RP |
9173 | #ifndef CONFIG_MMU |
9174 | .get_unmapped_area = io_uring_nommu_get_unmapped_area, | |
9175 | .mmap_capabilities = io_uring_nommu_mmap_capabilities, | |
9176 | #endif | |
2b188cc1 JA |
9177 | .poll = io_uring_poll, |
9178 | .fasync = io_uring_fasync, | |
bebdb65e | 9179 | #ifdef CONFIG_PROC_FS |
87ce955b | 9180 | .show_fdinfo = io_uring_show_fdinfo, |
bebdb65e | 9181 | #endif |
2b188cc1 JA |
9182 | }; |
9183 | ||
9184 | static int io_allocate_scq_urings(struct io_ring_ctx *ctx, | |
9185 | struct io_uring_params *p) | |
9186 | { | |
75b28aff HV |
9187 | struct io_rings *rings; |
9188 | size_t size, sq_array_offset; | |
2b188cc1 | 9189 | |
bd740481 JA |
9190 | /* make sure these are sane, as we already accounted them */ |
9191 | ctx->sq_entries = p->sq_entries; | |
9192 | ctx->cq_entries = p->cq_entries; | |
9193 | ||
75b28aff HV |
9194 | size = rings_size(p->sq_entries, p->cq_entries, &sq_array_offset); |
9195 | if (size == SIZE_MAX) | |
9196 | return -EOVERFLOW; | |
9197 | ||
9198 | rings = io_mem_alloc(size); | |
9199 | if (!rings) | |
2b188cc1 JA |
9200 | return -ENOMEM; |
9201 | ||
75b28aff HV |
9202 | ctx->rings = rings; |
9203 | ctx->sq_array = (u32 *)((char *)rings + sq_array_offset); | |
9204 | rings->sq_ring_mask = p->sq_entries - 1; | |
9205 | rings->cq_ring_mask = p->cq_entries - 1; | |
9206 | rings->sq_ring_entries = p->sq_entries; | |
9207 | rings->cq_ring_entries = p->cq_entries; | |
9208 | ctx->sq_mask = rings->sq_ring_mask; | |
9209 | ctx->cq_mask = rings->cq_ring_mask; | |
2b188cc1 JA |
9210 | |
9211 | size = array_size(sizeof(struct io_uring_sqe), p->sq_entries); | |
eb065d30 JA |
9212 | if (size == SIZE_MAX) { |
9213 | io_mem_free(ctx->rings); | |
9214 | ctx->rings = NULL; | |
2b188cc1 | 9215 | return -EOVERFLOW; |
eb065d30 | 9216 | } |
2b188cc1 JA |
9217 | |
9218 | ctx->sq_sqes = io_mem_alloc(size); | |
eb065d30 JA |
9219 | if (!ctx->sq_sqes) { |
9220 | io_mem_free(ctx->rings); | |
9221 | ctx->rings = NULL; | |
2b188cc1 | 9222 | return -ENOMEM; |
eb065d30 | 9223 | } |
2b188cc1 | 9224 | |
2b188cc1 JA |
9225 | return 0; |
9226 | } | |
9227 | ||
9228 | /* | |
9229 | * Allocate an anonymous fd, this is what constitutes the application | |
9230 | * visible backing of an io_uring instance. The application mmaps this | |
9231 | * fd to gain access to the SQ/CQ ring details. If UNIX sockets are enabled, | |
9232 | * we have to tie this fd to a socket for file garbage collection purposes. | |
9233 | */ | |
9234 | static int io_uring_get_fd(struct io_ring_ctx *ctx) | |
9235 | { | |
9236 | struct file *file; | |
9237 | int ret; | |
9238 | ||
9239 | #if defined(CONFIG_UNIX) | |
9240 | ret = sock_create_kern(&init_net, PF_UNIX, SOCK_RAW, IPPROTO_IP, | |
9241 | &ctx->ring_sock); | |
9242 | if (ret) | |
9243 | return ret; | |
9244 | #endif | |
9245 | ||
9246 | ret = get_unused_fd_flags(O_RDWR | O_CLOEXEC); | |
9247 | if (ret < 0) | |
9248 | goto err; | |
9249 | ||
9250 | file = anon_inode_getfile("[io_uring]", &io_uring_fops, ctx, | |
9251 | O_RDWR | O_CLOEXEC); | |
9252 | if (IS_ERR(file)) { | |
0f212204 | 9253 | err_fd: |
2b188cc1 JA |
9254 | put_unused_fd(ret); |
9255 | ret = PTR_ERR(file); | |
9256 | goto err; | |
9257 | } | |
9258 | ||
9259 | #if defined(CONFIG_UNIX) | |
9260 | ctx->ring_sock->file = file; | |
9261 | #endif | |
fdaf083c | 9262 | if (unlikely(io_uring_add_task_file(ctx, file))) { |
0f212204 JA |
9263 | file = ERR_PTR(-ENOMEM); |
9264 | goto err_fd; | |
9265 | } | |
2b188cc1 JA |
9266 | fd_install(ret, file); |
9267 | return ret; | |
9268 | err: | |
9269 | #if defined(CONFIG_UNIX) | |
9270 | sock_release(ctx->ring_sock); | |
9271 | ctx->ring_sock = NULL; | |
9272 | #endif | |
9273 | return ret; | |
9274 | } | |
9275 | ||
7f13657d XW |
9276 | static int io_uring_create(unsigned entries, struct io_uring_params *p, |
9277 | struct io_uring_params __user *params) | |
2b188cc1 JA |
9278 | { |
9279 | struct user_struct *user = NULL; | |
9280 | struct io_ring_ctx *ctx; | |
aad5d8da | 9281 | bool limit_mem; |
2b188cc1 JA |
9282 | int ret; |
9283 | ||
8110c1a6 | 9284 | if (!entries) |
2b188cc1 | 9285 | return -EINVAL; |
8110c1a6 JA |
9286 | if (entries > IORING_MAX_ENTRIES) { |
9287 | if (!(p->flags & IORING_SETUP_CLAMP)) | |
9288 | return -EINVAL; | |
9289 | entries = IORING_MAX_ENTRIES; | |
9290 | } | |
2b188cc1 JA |
9291 | |
9292 | /* | |
9293 | * Use twice as many entries for the CQ ring. It's possible for the | |
9294 | * application to drive a higher depth than the size of the SQ ring, | |
9295 | * since the sqes are only used at submission time. This allows for | |
33a107f0 JA |
9296 | * some flexibility in overcommitting a bit. If the application has |
9297 | * set IORING_SETUP_CQSIZE, it will have passed in the desired number | |
9298 | * of CQ ring entries manually. | |
2b188cc1 JA |
9299 | */ |
9300 | p->sq_entries = roundup_pow_of_two(entries); | |
33a107f0 JA |
9301 | if (p->flags & IORING_SETUP_CQSIZE) { |
9302 | /* | |
9303 | * If IORING_SETUP_CQSIZE is set, we do the same roundup | |
9304 | * to a power-of-two, if it isn't already. We do NOT impose | |
9305 | * any cq vs sq ring sizing. | |
9306 | */ | |
88ec3211 | 9307 | p->cq_entries = roundup_pow_of_two(p->cq_entries); |
8110c1a6 | 9308 | if (p->cq_entries < p->sq_entries) |
33a107f0 | 9309 | return -EINVAL; |
8110c1a6 JA |
9310 | if (p->cq_entries > IORING_MAX_CQ_ENTRIES) { |
9311 | if (!(p->flags & IORING_SETUP_CLAMP)) | |
9312 | return -EINVAL; | |
9313 | p->cq_entries = IORING_MAX_CQ_ENTRIES; | |
9314 | } | |
33a107f0 JA |
9315 | } else { |
9316 | p->cq_entries = 2 * p->sq_entries; | |
9317 | } | |
2b188cc1 JA |
9318 | |
9319 | user = get_uid(current_user()); | |
aad5d8da | 9320 | limit_mem = !capable(CAP_IPC_LOCK); |
2b188cc1 | 9321 | |
aad5d8da | 9322 | if (limit_mem) { |
a087e2b5 | 9323 | ret = __io_account_mem(user, |
2b188cc1 JA |
9324 | ring_pages(p->sq_entries, p->cq_entries)); |
9325 | if (ret) { | |
9326 | free_uid(user); | |
9327 | return ret; | |
9328 | } | |
9329 | } | |
9330 | ||
9331 | ctx = io_ring_ctx_alloc(p); | |
9332 | if (!ctx) { | |
aad5d8da | 9333 | if (limit_mem) |
a087e2b5 | 9334 | __io_unaccount_mem(user, ring_pages(p->sq_entries, |
2b188cc1 JA |
9335 | p->cq_entries)); |
9336 | free_uid(user); | |
9337 | return -ENOMEM; | |
9338 | } | |
9339 | ctx->compat = in_compat_syscall(); | |
2b188cc1 | 9340 | ctx->user = user; |
0b8c0ec7 | 9341 | ctx->creds = get_current_cred(); |
4ea33a97 JA |
9342 | #ifdef CONFIG_AUDIT |
9343 | ctx->loginuid = current->loginuid; | |
9344 | ctx->sessionid = current->sessionid; | |
9345 | #endif | |
2aede0e4 JA |
9346 | ctx->sqo_task = get_task_struct(current); |
9347 | ||
9348 | /* | |
9349 | * This is just grabbed for accounting purposes. When a process exits, | |
9350 | * the mm is exited and dropped before the files, hence we need to hang | |
9351 | * on to this mm purely for the purposes of being able to unaccount | |
9352 | * memory (locked/pinned vm). It's not used for anything else. | |
9353 | */ | |
6b7898eb | 9354 | mmgrab(current->mm); |
2aede0e4 | 9355 | ctx->mm_account = current->mm; |
6b7898eb | 9356 | |
91d8f519 DZ |
9357 | #ifdef CONFIG_BLK_CGROUP |
9358 | /* | |
9359 | * The sq thread will belong to the original cgroup it was inited in. | |
9360 | * If the cgroup goes offline (e.g. disabling the io controller), then | |
9361 | * issued bios will be associated with the closest cgroup later in the | |
9362 | * block layer. | |
9363 | */ | |
9364 | rcu_read_lock(); | |
9365 | ctx->sqo_blkcg_css = blkcg_css(); | |
9366 | ret = css_tryget_online(ctx->sqo_blkcg_css); | |
9367 | rcu_read_unlock(); | |
9368 | if (!ret) { | |
9369 | /* don't init against a dying cgroup, have the user try again */ | |
9370 | ctx->sqo_blkcg_css = NULL; | |
9371 | ret = -ENODEV; | |
9372 | goto err; | |
9373 | } | |
9374 | #endif | |
6b7898eb | 9375 | |
f74441e6 JA |
9376 | /* |
9377 | * Account memory _before_ installing the file descriptor. Once | |
9378 | * the descriptor is installed, it can get closed at any time. Also | |
9379 | * do this before hitting the general error path, as ring freeing | |
9380 | * will un-account as well. | |
9381 | */ | |
9382 | io_account_mem(ctx, ring_pages(p->sq_entries, p->cq_entries), | |
9383 | ACCT_LOCKED); | |
9384 | ctx->limit_mem = limit_mem; | |
9385 | ||
2b188cc1 JA |
9386 | ret = io_allocate_scq_urings(ctx, p); |
9387 | if (ret) | |
9388 | goto err; | |
9389 | ||
7e84e1c7 | 9390 | ret = io_sq_offload_create(ctx, p); |
2b188cc1 JA |
9391 | if (ret) |
9392 | goto err; | |
9393 | ||
7e84e1c7 SG |
9394 | if (!(p->flags & IORING_SETUP_R_DISABLED)) |
9395 | io_sq_offload_start(ctx); | |
9396 | ||
2b188cc1 | 9397 | memset(&p->sq_off, 0, sizeof(p->sq_off)); |
75b28aff HV |
9398 | p->sq_off.head = offsetof(struct io_rings, sq.head); |
9399 | p->sq_off.tail = offsetof(struct io_rings, sq.tail); | |
9400 | p->sq_off.ring_mask = offsetof(struct io_rings, sq_ring_mask); | |
9401 | p->sq_off.ring_entries = offsetof(struct io_rings, sq_ring_entries); | |
9402 | p->sq_off.flags = offsetof(struct io_rings, sq_flags); | |
9403 | p->sq_off.dropped = offsetof(struct io_rings, sq_dropped); | |
9404 | p->sq_off.array = (char *)ctx->sq_array - (char *)ctx->rings; | |
2b188cc1 JA |
9405 | |
9406 | memset(&p->cq_off, 0, sizeof(p->cq_off)); | |
75b28aff HV |
9407 | p->cq_off.head = offsetof(struct io_rings, cq.head); |
9408 | p->cq_off.tail = offsetof(struct io_rings, cq.tail); | |
9409 | p->cq_off.ring_mask = offsetof(struct io_rings, cq_ring_mask); | |
9410 | p->cq_off.ring_entries = offsetof(struct io_rings, cq_ring_entries); | |
9411 | p->cq_off.overflow = offsetof(struct io_rings, cq_overflow); | |
9412 | p->cq_off.cqes = offsetof(struct io_rings, cqes); | |
0d9b5b3a | 9413 | p->cq_off.flags = offsetof(struct io_rings, cq_flags); |
ac90f249 | 9414 | |
7f13657d XW |
9415 | p->features = IORING_FEAT_SINGLE_MMAP | IORING_FEAT_NODROP | |
9416 | IORING_FEAT_SUBMIT_STABLE | IORING_FEAT_RW_CUR_POS | | |
5769a351 JX |
9417 | IORING_FEAT_CUR_PERSONALITY | IORING_FEAT_FAST_POLL | |
9418 | IORING_FEAT_POLL_32BITS; | |
7f13657d XW |
9419 | |
9420 | if (copy_to_user(params, p, sizeof(*p))) { | |
9421 | ret = -EFAULT; | |
9422 | goto err; | |
9423 | } | |
d1719f70 | 9424 | |
044c1ab3 JA |
9425 | /* |
9426 | * Install ring fd as the very last thing, so we don't risk someone | |
9427 | * having closed it before we finish setup | |
9428 | */ | |
9429 | ret = io_uring_get_fd(ctx); | |
9430 | if (ret < 0) | |
9431 | goto err; | |
9432 | ||
c826bd7a | 9433 | trace_io_uring_create(ret, ctx, p->sq_entries, p->cq_entries, p->flags); |
2b188cc1 JA |
9434 | return ret; |
9435 | err: | |
9436 | io_ring_ctx_wait_and_kill(ctx); | |
9437 | return ret; | |
9438 | } | |
9439 | ||
9440 | /* | |
9441 | * Sets up an aio uring context, and returns the fd. Applications asks for a | |
9442 | * ring size, we return the actual sq/cq ring sizes (among other things) in the | |
9443 | * params structure passed in. | |
9444 | */ | |
9445 | static long io_uring_setup(u32 entries, struct io_uring_params __user *params) | |
9446 | { | |
9447 | struct io_uring_params p; | |
2b188cc1 JA |
9448 | int i; |
9449 | ||
9450 | if (copy_from_user(&p, params, sizeof(p))) | |
9451 | return -EFAULT; | |
9452 | for (i = 0; i < ARRAY_SIZE(p.resv); i++) { | |
9453 | if (p.resv[i]) | |
9454 | return -EINVAL; | |
9455 | } | |
9456 | ||
6c271ce2 | 9457 | if (p.flags & ~(IORING_SETUP_IOPOLL | IORING_SETUP_SQPOLL | |
8110c1a6 | 9458 | IORING_SETUP_SQ_AFF | IORING_SETUP_CQSIZE | |
7e84e1c7 SG |
9459 | IORING_SETUP_CLAMP | IORING_SETUP_ATTACH_WQ | |
9460 | IORING_SETUP_R_DISABLED)) | |
2b188cc1 JA |
9461 | return -EINVAL; |
9462 | ||
7f13657d | 9463 | return io_uring_create(entries, &p, params); |
2b188cc1 JA |
9464 | } |
9465 | ||
9466 | SYSCALL_DEFINE2(io_uring_setup, u32, entries, | |
9467 | struct io_uring_params __user *, params) | |
9468 | { | |
9469 | return io_uring_setup(entries, params); | |
9470 | } | |
9471 | ||
66f4af93 JA |
9472 | static int io_probe(struct io_ring_ctx *ctx, void __user *arg, unsigned nr_args) |
9473 | { | |
9474 | struct io_uring_probe *p; | |
9475 | size_t size; | |
9476 | int i, ret; | |
9477 | ||
9478 | size = struct_size(p, ops, nr_args); | |
9479 | if (size == SIZE_MAX) | |
9480 | return -EOVERFLOW; | |
9481 | p = kzalloc(size, GFP_KERNEL); | |
9482 | if (!p) | |
9483 | return -ENOMEM; | |
9484 | ||
9485 | ret = -EFAULT; | |
9486 | if (copy_from_user(p, arg, size)) | |
9487 | goto out; | |
9488 | ret = -EINVAL; | |
9489 | if (memchr_inv(p, 0, size)) | |
9490 | goto out; | |
9491 | ||
9492 | p->last_op = IORING_OP_LAST - 1; | |
9493 | if (nr_args > IORING_OP_LAST) | |
9494 | nr_args = IORING_OP_LAST; | |
9495 | ||
9496 | for (i = 0; i < nr_args; i++) { | |
9497 | p->ops[i].op = i; | |
9498 | if (!io_op_defs[i].not_supported) | |
9499 | p->ops[i].flags = IO_URING_OP_SUPPORTED; | |
9500 | } | |
9501 | p->ops_len = i; | |
9502 | ||
9503 | ret = 0; | |
9504 | if (copy_to_user(arg, p, size)) | |
9505 | ret = -EFAULT; | |
9506 | out: | |
9507 | kfree(p); | |
9508 | return ret; | |
9509 | } | |
9510 | ||
071698e1 JA |
9511 | static int io_register_personality(struct io_ring_ctx *ctx) |
9512 | { | |
1e6fa521 JA |
9513 | struct io_identity *id; |
9514 | int ret; | |
071698e1 | 9515 | |
1e6fa521 JA |
9516 | id = kmalloc(sizeof(*id), GFP_KERNEL); |
9517 | if (unlikely(!id)) | |
9518 | return -ENOMEM; | |
9519 | ||
9520 | io_init_identity(id); | |
9521 | id->creds = get_current_cred(); | |
9522 | ||
9523 | ret = idr_alloc_cyclic(&ctx->personality_idr, id, 1, USHRT_MAX, GFP_KERNEL); | |
9524 | if (ret < 0) { | |
9525 | put_cred(id->creds); | |
9526 | kfree(id); | |
9527 | } | |
9528 | return ret; | |
071698e1 JA |
9529 | } |
9530 | ||
9531 | static int io_unregister_personality(struct io_ring_ctx *ctx, unsigned id) | |
9532 | { | |
1e6fa521 | 9533 | struct io_identity *iod; |
071698e1 | 9534 | |
1e6fa521 JA |
9535 | iod = idr_remove(&ctx->personality_idr, id); |
9536 | if (iod) { | |
9537 | put_cred(iod->creds); | |
9538 | if (refcount_dec_and_test(&iod->count)) | |
9539 | kfree(iod); | |
071698e1 JA |
9540 | return 0; |
9541 | } | |
9542 | ||
9543 | return -EINVAL; | |
9544 | } | |
9545 | ||
21b55dbc SG |
9546 | static int io_register_restrictions(struct io_ring_ctx *ctx, void __user *arg, |
9547 | unsigned int nr_args) | |
9548 | { | |
9549 | struct io_uring_restriction *res; | |
9550 | size_t size; | |
9551 | int i, ret; | |
9552 | ||
7e84e1c7 SG |
9553 | /* Restrictions allowed only if rings started disabled */ |
9554 | if (!(ctx->flags & IORING_SETUP_R_DISABLED)) | |
9555 | return -EBADFD; | |
9556 | ||
21b55dbc | 9557 | /* We allow only a single restrictions registration */ |
7e84e1c7 | 9558 | if (ctx->restrictions.registered) |
21b55dbc SG |
9559 | return -EBUSY; |
9560 | ||
9561 | if (!arg || nr_args > IORING_MAX_RESTRICTIONS) | |
9562 | return -EINVAL; | |
9563 | ||
9564 | size = array_size(nr_args, sizeof(*res)); | |
9565 | if (size == SIZE_MAX) | |
9566 | return -EOVERFLOW; | |
9567 | ||
9568 | res = memdup_user(arg, size); | |
9569 | if (IS_ERR(res)) | |
9570 | return PTR_ERR(res); | |
9571 | ||
9572 | ret = 0; | |
9573 | ||
9574 | for (i = 0; i < nr_args; i++) { | |
9575 | switch (res[i].opcode) { | |
9576 | case IORING_RESTRICTION_REGISTER_OP: | |
9577 | if (res[i].register_op >= IORING_REGISTER_LAST) { | |
9578 | ret = -EINVAL; | |
9579 | goto out; | |
9580 | } | |
9581 | ||
9582 | __set_bit(res[i].register_op, | |
9583 | ctx->restrictions.register_op); | |
9584 | break; | |
9585 | case IORING_RESTRICTION_SQE_OP: | |
9586 | if (res[i].sqe_op >= IORING_OP_LAST) { | |
9587 | ret = -EINVAL; | |
9588 | goto out; | |
9589 | } | |
9590 | ||
9591 | __set_bit(res[i].sqe_op, ctx->restrictions.sqe_op); | |
9592 | break; | |
9593 | case IORING_RESTRICTION_SQE_FLAGS_ALLOWED: | |
9594 | ctx->restrictions.sqe_flags_allowed = res[i].sqe_flags; | |
9595 | break; | |
9596 | case IORING_RESTRICTION_SQE_FLAGS_REQUIRED: | |
9597 | ctx->restrictions.sqe_flags_required = res[i].sqe_flags; | |
9598 | break; | |
9599 | default: | |
9600 | ret = -EINVAL; | |
9601 | goto out; | |
9602 | } | |
9603 | } | |
9604 | ||
9605 | out: | |
9606 | /* Reset all restrictions if an error happened */ | |
9607 | if (ret != 0) | |
9608 | memset(&ctx->restrictions, 0, sizeof(ctx->restrictions)); | |
9609 | else | |
7e84e1c7 | 9610 | ctx->restrictions.registered = true; |
21b55dbc SG |
9611 | |
9612 | kfree(res); | |
9613 | return ret; | |
9614 | } | |
9615 | ||
7e84e1c7 SG |
9616 | static int io_register_enable_rings(struct io_ring_ctx *ctx) |
9617 | { | |
9618 | if (!(ctx->flags & IORING_SETUP_R_DISABLED)) | |
9619 | return -EBADFD; | |
9620 | ||
9621 | if (ctx->restrictions.registered) | |
9622 | ctx->restricted = 1; | |
9623 | ||
9624 | ctx->flags &= ~IORING_SETUP_R_DISABLED; | |
9625 | ||
9626 | io_sq_offload_start(ctx); | |
9627 | ||
9628 | return 0; | |
9629 | } | |
9630 | ||
071698e1 JA |
9631 | static bool io_register_op_must_quiesce(int op) |
9632 | { | |
9633 | switch (op) { | |
9634 | case IORING_UNREGISTER_FILES: | |
9635 | case IORING_REGISTER_FILES_UPDATE: | |
9636 | case IORING_REGISTER_PROBE: | |
9637 | case IORING_REGISTER_PERSONALITY: | |
9638 | case IORING_UNREGISTER_PERSONALITY: | |
9639 | return false; | |
9640 | default: | |
9641 | return true; | |
9642 | } | |
9643 | } | |
9644 | ||
edafccee JA |
9645 | static int __io_uring_register(struct io_ring_ctx *ctx, unsigned opcode, |
9646 | void __user *arg, unsigned nr_args) | |
b19062a5 JA |
9647 | __releases(ctx->uring_lock) |
9648 | __acquires(ctx->uring_lock) | |
edafccee JA |
9649 | { |
9650 | int ret; | |
9651 | ||
35fa71a0 JA |
9652 | /* |
9653 | * We're inside the ring mutex, if the ref is already dying, then | |
9654 | * someone else killed the ctx or is already going through | |
9655 | * io_uring_register(). | |
9656 | */ | |
9657 | if (percpu_ref_is_dying(&ctx->refs)) | |
9658 | return -ENXIO; | |
9659 | ||
071698e1 | 9660 | if (io_register_op_must_quiesce(opcode)) { |
05f3fb3c | 9661 | percpu_ref_kill(&ctx->refs); |
b19062a5 | 9662 | |
05f3fb3c JA |
9663 | /* |
9664 | * Drop uring mutex before waiting for references to exit. If | |
9665 | * another thread is currently inside io_uring_enter() it might | |
9666 | * need to grab the uring_lock to make progress. If we hold it | |
9667 | * here across the drain wait, then we can deadlock. It's safe | |
9668 | * to drop the mutex here, since no new references will come in | |
9669 | * after we've killed the percpu ref. | |
9670 | */ | |
9671 | mutex_unlock(&ctx->uring_lock); | |
af9c1a44 JA |
9672 | do { |
9673 | ret = wait_for_completion_interruptible(&ctx->ref_comp); | |
9674 | if (!ret) | |
9675 | break; | |
ed6930c9 JA |
9676 | ret = io_run_task_work_sig(); |
9677 | if (ret < 0) | |
9678 | break; | |
af9c1a44 JA |
9679 | } while (1); |
9680 | ||
05f3fb3c | 9681 | mutex_lock(&ctx->uring_lock); |
af9c1a44 | 9682 | |
c150368b JA |
9683 | if (ret) { |
9684 | percpu_ref_resurrect(&ctx->refs); | |
21b55dbc SG |
9685 | goto out_quiesce; |
9686 | } | |
9687 | } | |
9688 | ||
9689 | if (ctx->restricted) { | |
9690 | if (opcode >= IORING_REGISTER_LAST) { | |
9691 | ret = -EINVAL; | |
9692 | goto out; | |
9693 | } | |
9694 | ||
9695 | if (!test_bit(opcode, ctx->restrictions.register_op)) { | |
9696 | ret = -EACCES; | |
c150368b JA |
9697 | goto out; |
9698 | } | |
05f3fb3c | 9699 | } |
edafccee JA |
9700 | |
9701 | switch (opcode) { | |
9702 | case IORING_REGISTER_BUFFERS: | |
9703 | ret = io_sqe_buffer_register(ctx, arg, nr_args); | |
9704 | break; | |
9705 | case IORING_UNREGISTER_BUFFERS: | |
9706 | ret = -EINVAL; | |
9707 | if (arg || nr_args) | |
9708 | break; | |
9709 | ret = io_sqe_buffer_unregister(ctx); | |
9710 | break; | |
6b06314c JA |
9711 | case IORING_REGISTER_FILES: |
9712 | ret = io_sqe_files_register(ctx, arg, nr_args); | |
9713 | break; | |
9714 | case IORING_UNREGISTER_FILES: | |
9715 | ret = -EINVAL; | |
9716 | if (arg || nr_args) | |
9717 | break; | |
9718 | ret = io_sqe_files_unregister(ctx); | |
9719 | break; | |
c3a31e60 JA |
9720 | case IORING_REGISTER_FILES_UPDATE: |
9721 | ret = io_sqe_files_update(ctx, arg, nr_args); | |
9722 | break; | |
9b402849 | 9723 | case IORING_REGISTER_EVENTFD: |
f2842ab5 | 9724 | case IORING_REGISTER_EVENTFD_ASYNC: |
9b402849 JA |
9725 | ret = -EINVAL; |
9726 | if (nr_args != 1) | |
9727 | break; | |
9728 | ret = io_eventfd_register(ctx, arg); | |
f2842ab5 JA |
9729 | if (ret) |
9730 | break; | |
9731 | if (opcode == IORING_REGISTER_EVENTFD_ASYNC) | |
9732 | ctx->eventfd_async = 1; | |
9733 | else | |
9734 | ctx->eventfd_async = 0; | |
9b402849 JA |
9735 | break; |
9736 | case IORING_UNREGISTER_EVENTFD: | |
9737 | ret = -EINVAL; | |
9738 | if (arg || nr_args) | |
9739 | break; | |
9740 | ret = io_eventfd_unregister(ctx); | |
9741 | break; | |
66f4af93 JA |
9742 | case IORING_REGISTER_PROBE: |
9743 | ret = -EINVAL; | |
9744 | if (!arg || nr_args > 256) | |
9745 | break; | |
9746 | ret = io_probe(ctx, arg, nr_args); | |
9747 | break; | |
071698e1 JA |
9748 | case IORING_REGISTER_PERSONALITY: |
9749 | ret = -EINVAL; | |
9750 | if (arg || nr_args) | |
9751 | break; | |
9752 | ret = io_register_personality(ctx); | |
9753 | break; | |
9754 | case IORING_UNREGISTER_PERSONALITY: | |
9755 | ret = -EINVAL; | |
9756 | if (arg) | |
9757 | break; | |
9758 | ret = io_unregister_personality(ctx, nr_args); | |
9759 | break; | |
7e84e1c7 SG |
9760 | case IORING_REGISTER_ENABLE_RINGS: |
9761 | ret = -EINVAL; | |
9762 | if (arg || nr_args) | |
9763 | break; | |
9764 | ret = io_register_enable_rings(ctx); | |
9765 | break; | |
21b55dbc SG |
9766 | case IORING_REGISTER_RESTRICTIONS: |
9767 | ret = io_register_restrictions(ctx, arg, nr_args); | |
9768 | break; | |
edafccee JA |
9769 | default: |
9770 | ret = -EINVAL; | |
9771 | break; | |
9772 | } | |
9773 | ||
21b55dbc | 9774 | out: |
071698e1 | 9775 | if (io_register_op_must_quiesce(opcode)) { |
05f3fb3c | 9776 | /* bring the ctx back to life */ |
05f3fb3c | 9777 | percpu_ref_reinit(&ctx->refs); |
21b55dbc | 9778 | out_quiesce: |
0f158b4c | 9779 | reinit_completion(&ctx->ref_comp); |
05f3fb3c | 9780 | } |
edafccee JA |
9781 | return ret; |
9782 | } | |
9783 | ||
9784 | SYSCALL_DEFINE4(io_uring_register, unsigned int, fd, unsigned int, opcode, | |
9785 | void __user *, arg, unsigned int, nr_args) | |
9786 | { | |
9787 | struct io_ring_ctx *ctx; | |
9788 | long ret = -EBADF; | |
9789 | struct fd f; | |
9790 | ||
9791 | f = fdget(fd); | |
9792 | if (!f.file) | |
9793 | return -EBADF; | |
9794 | ||
9795 | ret = -EOPNOTSUPP; | |
9796 | if (f.file->f_op != &io_uring_fops) | |
9797 | goto out_fput; | |
9798 | ||
9799 | ctx = f.file->private_data; | |
9800 | ||
9801 | mutex_lock(&ctx->uring_lock); | |
9802 | ret = __io_uring_register(ctx, opcode, arg, nr_args); | |
9803 | mutex_unlock(&ctx->uring_lock); | |
c826bd7a DD |
9804 | trace_io_uring_register(ctx, opcode, ctx->nr_user_files, ctx->nr_user_bufs, |
9805 | ctx->cq_ev_fd != NULL, ret); | |
edafccee JA |
9806 | out_fput: |
9807 | fdput(f); | |
9808 | return ret; | |
9809 | } | |
9810 | ||
2b188cc1 JA |
9811 | static int __init io_uring_init(void) |
9812 | { | |
d7f62e82 SM |
9813 | #define __BUILD_BUG_VERIFY_ELEMENT(stype, eoffset, etype, ename) do { \ |
9814 | BUILD_BUG_ON(offsetof(stype, ename) != eoffset); \ | |
9815 | BUILD_BUG_ON(sizeof(etype) != sizeof_field(stype, ename)); \ | |
9816 | } while (0) | |
9817 | ||
9818 | #define BUILD_BUG_SQE_ELEM(eoffset, etype, ename) \ | |
9819 | __BUILD_BUG_VERIFY_ELEMENT(struct io_uring_sqe, eoffset, etype, ename) | |
9820 | BUILD_BUG_ON(sizeof(struct io_uring_sqe) != 64); | |
9821 | BUILD_BUG_SQE_ELEM(0, __u8, opcode); | |
9822 | BUILD_BUG_SQE_ELEM(1, __u8, flags); | |
9823 | BUILD_BUG_SQE_ELEM(2, __u16, ioprio); | |
9824 | BUILD_BUG_SQE_ELEM(4, __s32, fd); | |
9825 | BUILD_BUG_SQE_ELEM(8, __u64, off); | |
9826 | BUILD_BUG_SQE_ELEM(8, __u64, addr2); | |
9827 | BUILD_BUG_SQE_ELEM(16, __u64, addr); | |
7d67af2c | 9828 | BUILD_BUG_SQE_ELEM(16, __u64, splice_off_in); |
d7f62e82 SM |
9829 | BUILD_BUG_SQE_ELEM(24, __u32, len); |
9830 | BUILD_BUG_SQE_ELEM(28, __kernel_rwf_t, rw_flags); | |
9831 | BUILD_BUG_SQE_ELEM(28, /* compat */ int, rw_flags); | |
9832 | BUILD_BUG_SQE_ELEM(28, /* compat */ __u32, rw_flags); | |
9833 | BUILD_BUG_SQE_ELEM(28, __u32, fsync_flags); | |
5769a351 JX |
9834 | BUILD_BUG_SQE_ELEM(28, /* compat */ __u16, poll_events); |
9835 | BUILD_BUG_SQE_ELEM(28, __u32, poll32_events); | |
d7f62e82 SM |
9836 | BUILD_BUG_SQE_ELEM(28, __u32, sync_range_flags); |
9837 | BUILD_BUG_SQE_ELEM(28, __u32, msg_flags); | |
9838 | BUILD_BUG_SQE_ELEM(28, __u32, timeout_flags); | |
9839 | BUILD_BUG_SQE_ELEM(28, __u32, accept_flags); | |
9840 | BUILD_BUG_SQE_ELEM(28, __u32, cancel_flags); | |
9841 | BUILD_BUG_SQE_ELEM(28, __u32, open_flags); | |
9842 | BUILD_BUG_SQE_ELEM(28, __u32, statx_flags); | |
9843 | BUILD_BUG_SQE_ELEM(28, __u32, fadvise_advice); | |
7d67af2c | 9844 | BUILD_BUG_SQE_ELEM(28, __u32, splice_flags); |
d7f62e82 SM |
9845 | BUILD_BUG_SQE_ELEM(32, __u64, user_data); |
9846 | BUILD_BUG_SQE_ELEM(40, __u16, buf_index); | |
9847 | BUILD_BUG_SQE_ELEM(42, __u16, personality); | |
7d67af2c | 9848 | BUILD_BUG_SQE_ELEM(44, __s32, splice_fd_in); |
d7f62e82 | 9849 | |
d3656344 | 9850 | BUILD_BUG_ON(ARRAY_SIZE(io_op_defs) != IORING_OP_LAST); |
84557871 | 9851 | BUILD_BUG_ON(__REQ_F_LAST_BIT >= 8 * sizeof(int)); |
2b188cc1 JA |
9852 | req_cachep = KMEM_CACHE(io_kiocb, SLAB_HWCACHE_ALIGN | SLAB_PANIC); |
9853 | return 0; | |
9854 | }; | |
9855 | __initcall(io_uring_init); |