]>
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; | |
08369246 XW |
248 | |
249 | unsigned sq_thread_idle; | |
534ca6d6 JA |
250 | }; |
251 | ||
2b188cc1 JA |
252 | struct io_ring_ctx { |
253 | struct { | |
254 | struct percpu_ref refs; | |
255 | } ____cacheline_aligned_in_smp; | |
256 | ||
257 | struct { | |
258 | unsigned int flags; | |
e1d85334 | 259 | unsigned int compat: 1; |
aad5d8da | 260 | unsigned int limit_mem: 1; |
e1d85334 RD |
261 | unsigned int cq_overflow_flushed: 1; |
262 | unsigned int drain_next: 1; | |
263 | unsigned int eventfd_async: 1; | |
21b55dbc | 264 | unsigned int restricted: 1; |
d9d05217 | 265 | unsigned int sqo_dead: 1; |
2b188cc1 | 266 | |
75b28aff HV |
267 | /* |
268 | * Ring buffer of indices into array of io_uring_sqe, which is | |
269 | * mmapped by the application using the IORING_OFF_SQES offset. | |
270 | * | |
271 | * This indirection could e.g. be used to assign fixed | |
272 | * io_uring_sqe entries to operations and only submit them to | |
273 | * the queue when needed. | |
274 | * | |
275 | * The kernel modifies neither the indices array nor the entries | |
276 | * array. | |
277 | */ | |
278 | u32 *sq_array; | |
2b188cc1 JA |
279 | unsigned cached_sq_head; |
280 | unsigned sq_entries; | |
281 | unsigned sq_mask; | |
6c271ce2 | 282 | unsigned sq_thread_idle; |
498ccd9e | 283 | unsigned cached_sq_dropped; |
2c3bac6d | 284 | unsigned cached_cq_overflow; |
ad3eb2c8 | 285 | unsigned long sq_check_overflow; |
de0617e4 JA |
286 | |
287 | struct list_head defer_list; | |
5262f567 | 288 | struct list_head timeout_list; |
1d7bb1d5 | 289 | struct list_head cq_overflow_list; |
fcb323cc | 290 | |
ad3eb2c8 | 291 | struct io_uring_sqe *sq_sqes; |
2b188cc1 JA |
292 | } ____cacheline_aligned_in_smp; |
293 | ||
206aefde JA |
294 | struct io_rings *rings; |
295 | ||
2b188cc1 | 296 | /* IO offload */ |
561fb04a | 297 | struct io_wq *io_wq; |
2aede0e4 JA |
298 | |
299 | /* | |
300 | * For SQPOLL usage - we hold a reference to the parent task, so we | |
301 | * have access to the ->files | |
302 | */ | |
303 | struct task_struct *sqo_task; | |
304 | ||
305 | /* Only used for accounting purposes */ | |
306 | struct mm_struct *mm_account; | |
307 | ||
91d8f519 DZ |
308 | #ifdef CONFIG_BLK_CGROUP |
309 | struct cgroup_subsys_state *sqo_blkcg_css; | |
310 | #endif | |
311 | ||
534ca6d6 JA |
312 | struct io_sq_data *sq_data; /* if using sq thread polling */ |
313 | ||
90554200 | 314 | struct wait_queue_head sqo_sq_wait; |
69fb2131 | 315 | struct list_head sqd_list; |
75b28aff | 316 | |
6b06314c JA |
317 | /* |
318 | * If used, fixed file set. Writers must ensure that ->refs is dead, | |
319 | * readers must ensure that ->refs is alive as long as the file* is | |
320 | * used. Only updated through io_uring_register(2). | |
321 | */ | |
05f3fb3c | 322 | struct fixed_file_data *file_data; |
6b06314c JA |
323 | unsigned nr_user_files; |
324 | ||
edafccee JA |
325 | /* if used, fixed mapped user buffers */ |
326 | unsigned nr_user_bufs; | |
327 | struct io_mapped_ubuf *user_bufs; | |
328 | ||
2b188cc1 JA |
329 | struct user_struct *user; |
330 | ||
0b8c0ec7 | 331 | const struct cred *creds; |
181e448d | 332 | |
4ea33a97 JA |
333 | #ifdef CONFIG_AUDIT |
334 | kuid_t loginuid; | |
335 | unsigned int sessionid; | |
336 | #endif | |
337 | ||
0f158b4c JA |
338 | struct completion ref_comp; |
339 | struct completion sq_thread_comp; | |
206aefde | 340 | |
0ddf92e8 JA |
341 | /* if all else fails... */ |
342 | struct io_kiocb *fallback_req; | |
343 | ||
206aefde JA |
344 | #if defined(CONFIG_UNIX) |
345 | struct socket *ring_sock; | |
346 | #endif | |
347 | ||
5a2e745d JA |
348 | struct idr io_buffer_idr; |
349 | ||
071698e1 JA |
350 | struct idr personality_idr; |
351 | ||
206aefde JA |
352 | struct { |
353 | unsigned cached_cq_tail; | |
354 | unsigned cq_entries; | |
355 | unsigned cq_mask; | |
356 | atomic_t cq_timeouts; | |
f010505b | 357 | unsigned cq_last_tm_flush; |
ad3eb2c8 | 358 | unsigned long cq_check_overflow; |
206aefde JA |
359 | struct wait_queue_head cq_wait; |
360 | struct fasync_struct *cq_fasync; | |
361 | struct eventfd_ctx *cq_ev_fd; | |
362 | } ____cacheline_aligned_in_smp; | |
2b188cc1 JA |
363 | |
364 | struct { | |
365 | struct mutex uring_lock; | |
366 | wait_queue_head_t wait; | |
367 | } ____cacheline_aligned_in_smp; | |
368 | ||
369 | struct { | |
370 | spinlock_t completion_lock; | |
e94f141b | 371 | |
def596e9 | 372 | /* |
540e32a0 | 373 | * ->iopoll_list is protected by the ctx->uring_lock for |
def596e9 JA |
374 | * io_uring instances that don't use IORING_SETUP_SQPOLL. |
375 | * For SQPOLL, only the single threaded io_sq_thread() will | |
376 | * manipulate the list, hence no extra locking is needed there. | |
377 | */ | |
540e32a0 | 378 | struct list_head iopoll_list; |
78076bb6 JA |
379 | struct hlist_head *cancel_hash; |
380 | unsigned cancel_hash_bits; | |
e94f141b | 381 | bool poll_multi_file; |
31b51510 | 382 | |
fcb323cc JA |
383 | spinlock_t inflight_lock; |
384 | struct list_head inflight_list; | |
2b188cc1 | 385 | } ____cacheline_aligned_in_smp; |
85faa7b8 | 386 | |
4a38aed2 JA |
387 | struct delayed_work file_put_work; |
388 | struct llist_head file_put_llist; | |
389 | ||
85faa7b8 | 390 | struct work_struct exit_work; |
21b55dbc | 391 | struct io_restriction restrictions; |
2b188cc1 JA |
392 | }; |
393 | ||
09bb8394 JA |
394 | /* |
395 | * First field must be the file pointer in all the | |
396 | * iocb unions! See also 'struct kiocb' in <linux/fs.h> | |
397 | */ | |
221c5eb2 JA |
398 | struct io_poll_iocb { |
399 | struct file *file; | |
018043be | 400 | struct wait_queue_head *head; |
221c5eb2 | 401 | __poll_t events; |
8c838788 | 402 | bool done; |
221c5eb2 | 403 | bool canceled; |
392edb45 | 404 | struct wait_queue_entry wait; |
221c5eb2 JA |
405 | }; |
406 | ||
018043be PB |
407 | struct io_poll_remove { |
408 | struct file *file; | |
409 | u64 addr; | |
410 | }; | |
411 | ||
b5dba59e JA |
412 | struct io_close { |
413 | struct file *file; | |
414 | struct file *put_file; | |
415 | int fd; | |
416 | }; | |
417 | ||
ad8a48ac JA |
418 | struct io_timeout_data { |
419 | struct io_kiocb *req; | |
420 | struct hrtimer timer; | |
421 | struct timespec64 ts; | |
422 | enum hrtimer_mode mode; | |
423 | }; | |
424 | ||
8ed8d3c3 JA |
425 | struct io_accept { |
426 | struct file *file; | |
427 | struct sockaddr __user *addr; | |
428 | int __user *addr_len; | |
429 | int flags; | |
09952e3e | 430 | unsigned long nofile; |
8ed8d3c3 JA |
431 | }; |
432 | ||
433 | struct io_sync { | |
434 | struct file *file; | |
435 | loff_t len; | |
436 | loff_t off; | |
437 | int flags; | |
d63d1b5e | 438 | int mode; |
8ed8d3c3 JA |
439 | }; |
440 | ||
fbf23849 JA |
441 | struct io_cancel { |
442 | struct file *file; | |
443 | u64 addr; | |
444 | }; | |
445 | ||
b29472ee JA |
446 | struct io_timeout { |
447 | struct file *file; | |
bfe68a22 PB |
448 | u32 off; |
449 | u32 target_seq; | |
135fcde8 | 450 | struct list_head list; |
90cd7e42 PB |
451 | /* head of the link, used by linked timeouts only */ |
452 | struct io_kiocb *head; | |
b29472ee JA |
453 | }; |
454 | ||
0bdf7a2d PB |
455 | struct io_timeout_rem { |
456 | struct file *file; | |
457 | u64 addr; | |
9c8e11b3 PB |
458 | |
459 | /* timeout update */ | |
460 | struct timespec64 ts; | |
461 | u32 flags; | |
0bdf7a2d PB |
462 | }; |
463 | ||
9adbd45d JA |
464 | struct io_rw { |
465 | /* NOTE: kiocb has the file as the first member, so don't do it here */ | |
466 | struct kiocb kiocb; | |
467 | u64 addr; | |
468 | u64 len; | |
469 | }; | |
470 | ||
3fbb51c1 JA |
471 | struct io_connect { |
472 | struct file *file; | |
473 | struct sockaddr __user *addr; | |
474 | int addr_len; | |
475 | }; | |
476 | ||
e47293fd JA |
477 | struct io_sr_msg { |
478 | struct file *file; | |
fddaface | 479 | union { |
270a5940 | 480 | struct user_msghdr __user *umsg; |
fddaface JA |
481 | void __user *buf; |
482 | }; | |
e47293fd | 483 | int msg_flags; |
bcda7baa | 484 | int bgid; |
fddaface | 485 | size_t len; |
bcda7baa | 486 | struct io_buffer *kbuf; |
e47293fd JA |
487 | }; |
488 | ||
15b71abe JA |
489 | struct io_open { |
490 | struct file *file; | |
491 | int dfd; | |
944d1444 | 492 | bool ignore_nonblock; |
15b71abe | 493 | struct filename *filename; |
c12cedf2 | 494 | struct open_how how; |
4022e7af | 495 | unsigned long nofile; |
15b71abe JA |
496 | }; |
497 | ||
05f3fb3c JA |
498 | struct io_files_update { |
499 | struct file *file; | |
500 | u64 arg; | |
501 | u32 nr_args; | |
502 | u32 offset; | |
503 | }; | |
504 | ||
4840e418 JA |
505 | struct io_fadvise { |
506 | struct file *file; | |
507 | u64 offset; | |
508 | u32 len; | |
509 | u32 advice; | |
510 | }; | |
511 | ||
c1ca757b JA |
512 | struct io_madvise { |
513 | struct file *file; | |
514 | u64 addr; | |
515 | u32 len; | |
516 | u32 advice; | |
517 | }; | |
518 | ||
3e4827b0 JA |
519 | struct io_epoll { |
520 | struct file *file; | |
521 | int epfd; | |
522 | int op; | |
523 | int fd; | |
524 | struct epoll_event event; | |
e47293fd JA |
525 | }; |
526 | ||
7d67af2c PB |
527 | struct io_splice { |
528 | struct file *file_out; | |
529 | struct file *file_in; | |
530 | loff_t off_out; | |
531 | loff_t off_in; | |
532 | u64 len; | |
533 | unsigned int flags; | |
534 | }; | |
535 | ||
ddf0322d JA |
536 | struct io_provide_buf { |
537 | struct file *file; | |
538 | __u64 addr; | |
539 | __s32 len; | |
540 | __u32 bgid; | |
541 | __u16 nbufs; | |
542 | __u16 bid; | |
543 | }; | |
544 | ||
1d9e1288 BM |
545 | struct io_statx { |
546 | struct file *file; | |
547 | int dfd; | |
548 | unsigned int mask; | |
549 | unsigned int flags; | |
e62753e4 | 550 | const char __user *filename; |
1d9e1288 BM |
551 | struct statx __user *buffer; |
552 | }; | |
553 | ||
36f4fa68 JA |
554 | struct io_shutdown { |
555 | struct file *file; | |
556 | int how; | |
557 | }; | |
558 | ||
80a261fd JA |
559 | struct io_rename { |
560 | struct file *file; | |
561 | int old_dfd; | |
562 | int new_dfd; | |
563 | struct filename *oldpath; | |
564 | struct filename *newpath; | |
565 | int flags; | |
566 | }; | |
567 | ||
14a1143b JA |
568 | struct io_unlink { |
569 | struct file *file; | |
570 | int dfd; | |
571 | int flags; | |
572 | struct filename *filename; | |
573 | }; | |
574 | ||
3ca405eb PB |
575 | struct io_completion { |
576 | struct file *file; | |
577 | struct list_head list; | |
0f7e466b | 578 | int cflags; |
3ca405eb PB |
579 | }; |
580 | ||
f499a021 JA |
581 | struct io_async_connect { |
582 | struct sockaddr_storage address; | |
583 | }; | |
584 | ||
03b1230c JA |
585 | struct io_async_msghdr { |
586 | struct iovec fast_iov[UIO_FASTIOV]; | |
587 | struct iovec *iov; | |
588 | struct sockaddr __user *uaddr; | |
589 | struct msghdr msg; | |
b537916c | 590 | struct sockaddr_storage addr; |
03b1230c JA |
591 | }; |
592 | ||
f67676d1 JA |
593 | struct io_async_rw { |
594 | struct iovec fast_iov[UIO_FASTIOV]; | |
ff6165b2 JA |
595 | const struct iovec *free_iovec; |
596 | struct iov_iter iter; | |
227c0c96 | 597 | size_t bytes_done; |
bcf5a063 | 598 | struct wait_page_queue wpq; |
f67676d1 JA |
599 | }; |
600 | ||
6b47ee6e PB |
601 | enum { |
602 | REQ_F_FIXED_FILE_BIT = IOSQE_FIXED_FILE_BIT, | |
603 | REQ_F_IO_DRAIN_BIT = IOSQE_IO_DRAIN_BIT, | |
604 | REQ_F_LINK_BIT = IOSQE_IO_LINK_BIT, | |
605 | REQ_F_HARDLINK_BIT = IOSQE_IO_HARDLINK_BIT, | |
606 | REQ_F_FORCE_ASYNC_BIT = IOSQE_ASYNC_BIT, | |
bcda7baa | 607 | REQ_F_BUFFER_SELECT_BIT = IOSQE_BUFFER_SELECT_BIT, |
6b47ee6e | 608 | |
6b47ee6e PB |
609 | REQ_F_FAIL_LINK_BIT, |
610 | REQ_F_INFLIGHT_BIT, | |
611 | REQ_F_CUR_POS_BIT, | |
612 | REQ_F_NOWAIT_BIT, | |
6b47ee6e | 613 | REQ_F_LINK_TIMEOUT_BIT, |
6b47ee6e | 614 | REQ_F_ISREG_BIT, |
99bc4c38 | 615 | REQ_F_NEED_CLEANUP_BIT, |
d7718a9d | 616 | REQ_F_POLLED_BIT, |
bcda7baa | 617 | REQ_F_BUFFER_SELECTED_BIT, |
5b0bbee4 | 618 | REQ_F_NO_FILE_TABLE_BIT, |
7cdaf587 | 619 | REQ_F_WORK_INITIALIZED_BIT, |
900fad45 | 620 | REQ_F_LTIMEOUT_ACTIVE_BIT, |
84557871 JA |
621 | |
622 | /* not a real bit, just to check we're not overflowing the space */ | |
623 | __REQ_F_LAST_BIT, | |
6b47ee6e PB |
624 | }; |
625 | ||
626 | enum { | |
627 | /* ctx owns file */ | |
628 | REQ_F_FIXED_FILE = BIT(REQ_F_FIXED_FILE_BIT), | |
629 | /* drain existing IO first */ | |
630 | REQ_F_IO_DRAIN = BIT(REQ_F_IO_DRAIN_BIT), | |
631 | /* linked sqes */ | |
632 | REQ_F_LINK = BIT(REQ_F_LINK_BIT), | |
633 | /* doesn't sever on completion < 0 */ | |
634 | REQ_F_HARDLINK = BIT(REQ_F_HARDLINK_BIT), | |
635 | /* IOSQE_ASYNC */ | |
636 | REQ_F_FORCE_ASYNC = BIT(REQ_F_FORCE_ASYNC_BIT), | |
bcda7baa JA |
637 | /* IOSQE_BUFFER_SELECT */ |
638 | REQ_F_BUFFER_SELECT = BIT(REQ_F_BUFFER_SELECT_BIT), | |
6b47ee6e | 639 | |
6b47ee6e PB |
640 | /* fail rest of links */ |
641 | REQ_F_FAIL_LINK = BIT(REQ_F_FAIL_LINK_BIT), | |
642 | /* on inflight list */ | |
643 | REQ_F_INFLIGHT = BIT(REQ_F_INFLIGHT_BIT), | |
644 | /* read/write uses file position */ | |
645 | REQ_F_CUR_POS = BIT(REQ_F_CUR_POS_BIT), | |
646 | /* must not punt to workers */ | |
647 | REQ_F_NOWAIT = BIT(REQ_F_NOWAIT_BIT), | |
900fad45 | 648 | /* has or had linked timeout */ |
6b47ee6e | 649 | REQ_F_LINK_TIMEOUT = BIT(REQ_F_LINK_TIMEOUT_BIT), |
6b47ee6e PB |
650 | /* regular file */ |
651 | REQ_F_ISREG = BIT(REQ_F_ISREG_BIT), | |
99bc4c38 PB |
652 | /* needs cleanup */ |
653 | REQ_F_NEED_CLEANUP = BIT(REQ_F_NEED_CLEANUP_BIT), | |
d7718a9d JA |
654 | /* already went through poll handler */ |
655 | REQ_F_POLLED = BIT(REQ_F_POLLED_BIT), | |
bcda7baa JA |
656 | /* buffer already selected */ |
657 | REQ_F_BUFFER_SELECTED = BIT(REQ_F_BUFFER_SELECTED_BIT), | |
5b0bbee4 JA |
658 | /* doesn't need file table for this request */ |
659 | REQ_F_NO_FILE_TABLE = BIT(REQ_F_NO_FILE_TABLE_BIT), | |
7cdaf587 XW |
660 | /* io_wq_work is initialized */ |
661 | REQ_F_WORK_INITIALIZED = BIT(REQ_F_WORK_INITIALIZED_BIT), | |
900fad45 PB |
662 | /* linked timeout is active, i.e. prepared by link's head */ |
663 | REQ_F_LTIMEOUT_ACTIVE = BIT(REQ_F_LTIMEOUT_ACTIVE_BIT), | |
d7718a9d JA |
664 | }; |
665 | ||
666 | struct async_poll { | |
667 | struct io_poll_iocb poll; | |
807abcb0 | 668 | struct io_poll_iocb *double_poll; |
6b47ee6e PB |
669 | }; |
670 | ||
09bb8394 JA |
671 | /* |
672 | * NOTE! Each of the iocb union members has the file pointer | |
673 | * as the first entry in their struct definition. So you can | |
674 | * access the file pointer through any of the sub-structs, | |
675 | * or directly as just 'ki_filp' in this struct. | |
676 | */ | |
2b188cc1 | 677 | struct io_kiocb { |
221c5eb2 | 678 | union { |
09bb8394 | 679 | struct file *file; |
9adbd45d | 680 | struct io_rw rw; |
221c5eb2 | 681 | struct io_poll_iocb poll; |
018043be | 682 | struct io_poll_remove poll_remove; |
8ed8d3c3 JA |
683 | struct io_accept accept; |
684 | struct io_sync sync; | |
fbf23849 | 685 | struct io_cancel cancel; |
b29472ee | 686 | struct io_timeout timeout; |
0bdf7a2d | 687 | struct io_timeout_rem timeout_rem; |
3fbb51c1 | 688 | struct io_connect connect; |
e47293fd | 689 | struct io_sr_msg sr_msg; |
15b71abe | 690 | struct io_open open; |
b5dba59e | 691 | struct io_close close; |
05f3fb3c | 692 | struct io_files_update files_update; |
4840e418 | 693 | struct io_fadvise fadvise; |
c1ca757b | 694 | struct io_madvise madvise; |
3e4827b0 | 695 | struct io_epoll epoll; |
7d67af2c | 696 | struct io_splice splice; |
ddf0322d | 697 | struct io_provide_buf pbuf; |
1d9e1288 | 698 | struct io_statx statx; |
36f4fa68 | 699 | struct io_shutdown shutdown; |
80a261fd | 700 | struct io_rename rename; |
14a1143b | 701 | struct io_unlink unlink; |
3ca405eb PB |
702 | /* use only after cleaning per-op data, see io_clean_op() */ |
703 | struct io_completion compl; | |
221c5eb2 | 704 | }; |
2b188cc1 | 705 | |
e8c2bc1f JA |
706 | /* opcode allocated if it needs to store data for async defer */ |
707 | void *async_data; | |
d625c6ee | 708 | u8 opcode; |
65a6543d XW |
709 | /* polled IO has completed */ |
710 | u8 iopoll_completed; | |
2b188cc1 | 711 | |
4f4eeba8 | 712 | u16 buf_index; |
9cf7c104 | 713 | u32 result; |
4f4eeba8 | 714 | |
010e8e6b PB |
715 | struct io_ring_ctx *ctx; |
716 | unsigned int flags; | |
717 | refcount_t refs; | |
718 | struct task_struct *task; | |
719 | u64 user_data; | |
d7718a9d | 720 | |
f2f87370 | 721 | struct io_kiocb *link; |
0415767e | 722 | struct percpu_ref *fixed_file_refs; |
fcb323cc | 723 | |
d21ffe7e PB |
724 | /* |
725 | * 1. used with ctx->iopoll_list with reads/writes | |
726 | * 2. to track reqs with ->files (see io_op_def::file_table) | |
727 | */ | |
010e8e6b | 728 | struct list_head inflight_entry; |
010e8e6b PB |
729 | struct callback_head task_work; |
730 | /* for polled requests, i.e. IORING_OP_POLL_ADD and async armed poll */ | |
731 | struct hlist_node hash_node; | |
732 | struct async_poll *apoll; | |
733 | struct io_wq_work work; | |
2b188cc1 | 734 | }; |
05589553 | 735 | |
27dc8338 PB |
736 | struct io_defer_entry { |
737 | struct list_head list; | |
738 | struct io_kiocb *req; | |
9cf7c104 | 739 | u32 seq; |
2b188cc1 JA |
740 | }; |
741 | ||
def596e9 | 742 | #define IO_IOPOLL_BATCH 8 |
2b188cc1 | 743 | |
013538bd JA |
744 | struct io_comp_state { |
745 | unsigned int nr; | |
746 | struct list_head list; | |
747 | struct io_ring_ctx *ctx; | |
748 | }; | |
749 | ||
9a56a232 JA |
750 | struct io_submit_state { |
751 | struct blk_plug plug; | |
752 | ||
2579f913 JA |
753 | /* |
754 | * io_kiocb alloc cache | |
755 | */ | |
756 | void *reqs[IO_IOPOLL_BATCH]; | |
6c8a3134 | 757 | unsigned int free_reqs; |
2579f913 | 758 | |
27926b68 JA |
759 | bool plug_started; |
760 | ||
013538bd JA |
761 | /* |
762 | * Batch completion logic | |
763 | */ | |
764 | struct io_comp_state comp; | |
765 | ||
9a56a232 JA |
766 | /* |
767 | * File reference cache | |
768 | */ | |
769 | struct file *file; | |
770 | unsigned int fd; | |
6e1271e6 | 771 | unsigned int file_refs; |
9a56a232 JA |
772 | unsigned int ios_left; |
773 | }; | |
774 | ||
d3656344 | 775 | struct io_op_def { |
d3656344 JA |
776 | /* needs req->file assigned */ |
777 | unsigned needs_file : 1; | |
fd2206e4 JA |
778 | /* don't fail if file grab fails */ |
779 | unsigned needs_file_no_error : 1; | |
d3656344 JA |
780 | /* hash wq insertion if file is a regular file */ |
781 | unsigned hash_reg_file : 1; | |
782 | /* unbound wq insertion if file is a non-regular file */ | |
783 | unsigned unbound_nonreg_file : 1; | |
66f4af93 JA |
784 | /* opcode is not supported by this kernel */ |
785 | unsigned not_supported : 1; | |
8a72758c JA |
786 | /* set if opcode supports polled "wait" */ |
787 | unsigned pollin : 1; | |
788 | unsigned pollout : 1; | |
bcda7baa JA |
789 | /* op supports buffer selection */ |
790 | unsigned buffer_select : 1; | |
e8c2bc1f JA |
791 | /* must always have async data allocated */ |
792 | unsigned needs_async_data : 1; | |
27926b68 JA |
793 | /* should block plug */ |
794 | unsigned plug : 1; | |
e8c2bc1f JA |
795 | /* size of async data needed, if any */ |
796 | unsigned short async_size; | |
0f203765 | 797 | unsigned work_flags; |
d3656344 JA |
798 | }; |
799 | ||
0918682b | 800 | static const struct io_op_def io_op_defs[] = { |
0463b6c5 PB |
801 | [IORING_OP_NOP] = {}, |
802 | [IORING_OP_READV] = { | |
d3656344 JA |
803 | .needs_file = 1, |
804 | .unbound_nonreg_file = 1, | |
8a72758c | 805 | .pollin = 1, |
4d954c25 | 806 | .buffer_select = 1, |
e8c2bc1f | 807 | .needs_async_data = 1, |
27926b68 | 808 | .plug = 1, |
e8c2bc1f | 809 | .async_size = sizeof(struct io_async_rw), |
0f203765 | 810 | .work_flags = IO_WQ_WORK_MM | IO_WQ_WORK_BLKCG, |
d3656344 | 811 | }, |
0463b6c5 | 812 | [IORING_OP_WRITEV] = { |
d3656344 JA |
813 | .needs_file = 1, |
814 | .hash_reg_file = 1, | |
815 | .unbound_nonreg_file = 1, | |
8a72758c | 816 | .pollout = 1, |
e8c2bc1f | 817 | .needs_async_data = 1, |
27926b68 | 818 | .plug = 1, |
e8c2bc1f | 819 | .async_size = sizeof(struct io_async_rw), |
69228338 JA |
820 | .work_flags = IO_WQ_WORK_MM | IO_WQ_WORK_BLKCG | |
821 | IO_WQ_WORK_FSIZE, | |
d3656344 | 822 | }, |
0463b6c5 | 823 | [IORING_OP_FSYNC] = { |
d3656344 | 824 | .needs_file = 1, |
0f203765 | 825 | .work_flags = IO_WQ_WORK_BLKCG, |
d3656344 | 826 | }, |
0463b6c5 | 827 | [IORING_OP_READ_FIXED] = { |
d3656344 JA |
828 | .needs_file = 1, |
829 | .unbound_nonreg_file = 1, | |
8a72758c | 830 | .pollin = 1, |
27926b68 | 831 | .plug = 1, |
e8c2bc1f | 832 | .async_size = sizeof(struct io_async_rw), |
4017eb91 | 833 | .work_flags = IO_WQ_WORK_BLKCG | IO_WQ_WORK_MM, |
d3656344 | 834 | }, |
0463b6c5 | 835 | [IORING_OP_WRITE_FIXED] = { |
d3656344 JA |
836 | .needs_file = 1, |
837 | .hash_reg_file = 1, | |
838 | .unbound_nonreg_file = 1, | |
8a72758c | 839 | .pollout = 1, |
27926b68 | 840 | .plug = 1, |
e8c2bc1f | 841 | .async_size = sizeof(struct io_async_rw), |
4017eb91 JA |
842 | .work_flags = IO_WQ_WORK_BLKCG | IO_WQ_WORK_FSIZE | |
843 | IO_WQ_WORK_MM, | |
d3656344 | 844 | }, |
0463b6c5 | 845 | [IORING_OP_POLL_ADD] = { |
d3656344 JA |
846 | .needs_file = 1, |
847 | .unbound_nonreg_file = 1, | |
848 | }, | |
0463b6c5 PB |
849 | [IORING_OP_POLL_REMOVE] = {}, |
850 | [IORING_OP_SYNC_FILE_RANGE] = { | |
d3656344 | 851 | .needs_file = 1, |
0f203765 | 852 | .work_flags = IO_WQ_WORK_BLKCG, |
d3656344 | 853 | }, |
0463b6c5 | 854 | [IORING_OP_SENDMSG] = { |
d3656344 JA |
855 | .needs_file = 1, |
856 | .unbound_nonreg_file = 1, | |
8a72758c | 857 | .pollout = 1, |
e8c2bc1f JA |
858 | .needs_async_data = 1, |
859 | .async_size = sizeof(struct io_async_msghdr), | |
10cad2c4 | 860 | .work_flags = IO_WQ_WORK_MM | IO_WQ_WORK_BLKCG, |
d3656344 | 861 | }, |
0463b6c5 | 862 | [IORING_OP_RECVMSG] = { |
d3656344 JA |
863 | .needs_file = 1, |
864 | .unbound_nonreg_file = 1, | |
8a72758c | 865 | .pollin = 1, |
52de1fe1 | 866 | .buffer_select = 1, |
e8c2bc1f JA |
867 | .needs_async_data = 1, |
868 | .async_size = sizeof(struct io_async_msghdr), | |
10cad2c4 | 869 | .work_flags = IO_WQ_WORK_MM | IO_WQ_WORK_BLKCG, |
d3656344 | 870 | }, |
0463b6c5 | 871 | [IORING_OP_TIMEOUT] = { |
e8c2bc1f JA |
872 | .needs_async_data = 1, |
873 | .async_size = sizeof(struct io_timeout_data), | |
0f203765 | 874 | .work_flags = IO_WQ_WORK_MM, |
d3656344 | 875 | }, |
9c8e11b3 PB |
876 | [IORING_OP_TIMEOUT_REMOVE] = { |
877 | /* used by timeout updates' prep() */ | |
878 | .work_flags = IO_WQ_WORK_MM, | |
879 | }, | |
0463b6c5 | 880 | [IORING_OP_ACCEPT] = { |
d3656344 JA |
881 | .needs_file = 1, |
882 | .unbound_nonreg_file = 1, | |
8a72758c | 883 | .pollin = 1, |
0f203765 | 884 | .work_flags = IO_WQ_WORK_MM | IO_WQ_WORK_FILES, |
d3656344 | 885 | }, |
0463b6c5 PB |
886 | [IORING_OP_ASYNC_CANCEL] = {}, |
887 | [IORING_OP_LINK_TIMEOUT] = { | |
e8c2bc1f JA |
888 | .needs_async_data = 1, |
889 | .async_size = sizeof(struct io_timeout_data), | |
0f203765 | 890 | .work_flags = IO_WQ_WORK_MM, |
d3656344 | 891 | }, |
0463b6c5 | 892 | [IORING_OP_CONNECT] = { |
d3656344 JA |
893 | .needs_file = 1, |
894 | .unbound_nonreg_file = 1, | |
8a72758c | 895 | .pollout = 1, |
e8c2bc1f JA |
896 | .needs_async_data = 1, |
897 | .async_size = sizeof(struct io_async_connect), | |
0f203765 | 898 | .work_flags = IO_WQ_WORK_MM, |
d3656344 | 899 | }, |
0463b6c5 | 900 | [IORING_OP_FALLOCATE] = { |
d3656344 | 901 | .needs_file = 1, |
69228338 | 902 | .work_flags = IO_WQ_WORK_BLKCG | IO_WQ_WORK_FSIZE, |
d3656344 | 903 | }, |
0463b6c5 | 904 | [IORING_OP_OPENAT] = { |
0f203765 | 905 | .work_flags = IO_WQ_WORK_FILES | IO_WQ_WORK_BLKCG | |
14587a46 | 906 | IO_WQ_WORK_FS | IO_WQ_WORK_MM, |
d3656344 | 907 | }, |
0463b6c5 | 908 | [IORING_OP_CLOSE] = { |
fd2206e4 JA |
909 | .needs_file = 1, |
910 | .needs_file_no_error = 1, | |
0f203765 | 911 | .work_flags = IO_WQ_WORK_FILES | IO_WQ_WORK_BLKCG, |
d3656344 | 912 | }, |
0463b6c5 | 913 | [IORING_OP_FILES_UPDATE] = { |
0f203765 | 914 | .work_flags = IO_WQ_WORK_FILES | IO_WQ_WORK_MM, |
d3656344 | 915 | }, |
0463b6c5 | 916 | [IORING_OP_STATX] = { |
0f203765 JA |
917 | .work_flags = IO_WQ_WORK_FILES | IO_WQ_WORK_MM | |
918 | IO_WQ_WORK_FS | IO_WQ_WORK_BLKCG, | |
d3656344 | 919 | }, |
0463b6c5 | 920 | [IORING_OP_READ] = { |
3a6820f2 JA |
921 | .needs_file = 1, |
922 | .unbound_nonreg_file = 1, | |
8a72758c | 923 | .pollin = 1, |
bcda7baa | 924 | .buffer_select = 1, |
27926b68 | 925 | .plug = 1, |
e8c2bc1f | 926 | .async_size = sizeof(struct io_async_rw), |
0f203765 | 927 | .work_flags = IO_WQ_WORK_MM | IO_WQ_WORK_BLKCG, |
3a6820f2 | 928 | }, |
0463b6c5 | 929 | [IORING_OP_WRITE] = { |
3a6820f2 JA |
930 | .needs_file = 1, |
931 | .unbound_nonreg_file = 1, | |
8a72758c | 932 | .pollout = 1, |
27926b68 | 933 | .plug = 1, |
e8c2bc1f | 934 | .async_size = sizeof(struct io_async_rw), |
69228338 JA |
935 | .work_flags = IO_WQ_WORK_MM | IO_WQ_WORK_BLKCG | |
936 | IO_WQ_WORK_FSIZE, | |
3a6820f2 | 937 | }, |
0463b6c5 | 938 | [IORING_OP_FADVISE] = { |
4840e418 | 939 | .needs_file = 1, |
0f203765 | 940 | .work_flags = IO_WQ_WORK_BLKCG, |
4840e418 | 941 | }, |
0463b6c5 | 942 | [IORING_OP_MADVISE] = { |
0f203765 | 943 | .work_flags = IO_WQ_WORK_MM | IO_WQ_WORK_BLKCG, |
c1ca757b | 944 | }, |
0463b6c5 | 945 | [IORING_OP_SEND] = { |
fddaface JA |
946 | .needs_file = 1, |
947 | .unbound_nonreg_file = 1, | |
8a72758c | 948 | .pollout = 1, |
0f203765 | 949 | .work_flags = IO_WQ_WORK_MM | IO_WQ_WORK_BLKCG, |
fddaface | 950 | }, |
0463b6c5 | 951 | [IORING_OP_RECV] = { |
fddaface JA |
952 | .needs_file = 1, |
953 | .unbound_nonreg_file = 1, | |
8a72758c | 954 | .pollin = 1, |
bcda7baa | 955 | .buffer_select = 1, |
0f203765 | 956 | .work_flags = IO_WQ_WORK_MM | IO_WQ_WORK_BLKCG, |
fddaface | 957 | }, |
0463b6c5 | 958 | [IORING_OP_OPENAT2] = { |
0f203765 | 959 | .work_flags = IO_WQ_WORK_FILES | IO_WQ_WORK_FS | |
14587a46 | 960 | IO_WQ_WORK_BLKCG | IO_WQ_WORK_MM, |
cebdb986 | 961 | }, |
3e4827b0 JA |
962 | [IORING_OP_EPOLL_CTL] = { |
963 | .unbound_nonreg_file = 1, | |
0f203765 | 964 | .work_flags = IO_WQ_WORK_FILES, |
3e4827b0 | 965 | }, |
7d67af2c PB |
966 | [IORING_OP_SPLICE] = { |
967 | .needs_file = 1, | |
968 | .hash_reg_file = 1, | |
969 | .unbound_nonreg_file = 1, | |
0f203765 | 970 | .work_flags = IO_WQ_WORK_BLKCG, |
ddf0322d JA |
971 | }, |
972 | [IORING_OP_PROVIDE_BUFFERS] = {}, | |
067524e9 | 973 | [IORING_OP_REMOVE_BUFFERS] = {}, |
f2a8d5c7 PB |
974 | [IORING_OP_TEE] = { |
975 | .needs_file = 1, | |
976 | .hash_reg_file = 1, | |
977 | .unbound_nonreg_file = 1, | |
978 | }, | |
36f4fa68 JA |
979 | [IORING_OP_SHUTDOWN] = { |
980 | .needs_file = 1, | |
981 | }, | |
80a261fd JA |
982 | [IORING_OP_RENAMEAT] = { |
983 | .work_flags = IO_WQ_WORK_MM | IO_WQ_WORK_FILES | | |
984 | IO_WQ_WORK_FS | IO_WQ_WORK_BLKCG, | |
985 | }, | |
14a1143b JA |
986 | [IORING_OP_UNLINKAT] = { |
987 | .work_flags = IO_WQ_WORK_MM | IO_WQ_WORK_FILES | | |
988 | IO_WQ_WORK_FS | IO_WQ_WORK_BLKCG, | |
989 | }, | |
d3656344 JA |
990 | }; |
991 | ||
2e0464d4 BM |
992 | enum io_mem_account { |
993 | ACCT_LOCKED, | |
994 | ACCT_PINNED, | |
995 | }; | |
996 | ||
90df0853 PB |
997 | static void __io_uring_cancel_task_requests(struct io_ring_ctx *ctx, |
998 | struct task_struct *task); | |
999 | ||
1ffc5422 PB |
1000 | static void destroy_fixed_file_ref_node(struct fixed_file_ref_node *ref_node); |
1001 | static struct fixed_file_ref_node *alloc_fixed_file_ref_node( | |
1002 | struct io_ring_ctx *ctx); | |
1003 | ||
81b68a5c PB |
1004 | static void __io_complete_rw(struct io_kiocb *req, long res, long res2, |
1005 | struct io_comp_state *cs); | |
78e19bbe | 1006 | static void io_cqring_fill_event(struct io_kiocb *req, long res); |
ec9c02ad | 1007 | static void io_put_req(struct io_kiocb *req); |
216578e5 | 1008 | static void io_put_req_deferred(struct io_kiocb *req, int nr); |
c40f6379 | 1009 | static void io_double_put_req(struct io_kiocb *req); |
94ae5e77 | 1010 | static struct io_kiocb *io_prep_linked_timeout(struct io_kiocb *req); |
7271ef3a | 1011 | static void __io_queue_linked_timeout(struct io_kiocb *req); |
94ae5e77 | 1012 | static void io_queue_linked_timeout(struct io_kiocb *req); |
05f3fb3c JA |
1013 | static int __io_sqe_files_update(struct io_ring_ctx *ctx, |
1014 | struct io_uring_files_update *ip, | |
1015 | unsigned nr_args); | |
3ca405eb | 1016 | static void __io_clean_op(struct io_kiocb *req); |
8371adf5 PB |
1017 | static struct file *io_file_get(struct io_submit_state *state, |
1018 | struct io_kiocb *req, int fd, bool fixed); | |
c1379e24 | 1019 | static void __io_queue_sqe(struct io_kiocb *req, struct io_comp_state *cs); |
4349f30e | 1020 | static void io_file_put_work(struct work_struct *work); |
de0617e4 | 1021 | |
b63534c4 JA |
1022 | static ssize_t io_import_iovec(int rw, struct io_kiocb *req, |
1023 | struct iovec **iovec, struct iov_iter *iter, | |
1024 | bool needs_lock); | |
ff6165b2 JA |
1025 | static int io_setup_async_rw(struct io_kiocb *req, const struct iovec *iovec, |
1026 | const struct iovec *fast_iov, | |
227c0c96 | 1027 | struct iov_iter *iter, bool force); |
de0617e4 | 1028 | |
2b188cc1 JA |
1029 | static struct kmem_cache *req_cachep; |
1030 | ||
0918682b | 1031 | static const struct file_operations io_uring_fops; |
2b188cc1 JA |
1032 | |
1033 | struct sock *io_uring_get_socket(struct file *file) | |
1034 | { | |
1035 | #if defined(CONFIG_UNIX) | |
1036 | if (file->f_op == &io_uring_fops) { | |
1037 | struct io_ring_ctx *ctx = file->private_data; | |
1038 | ||
1039 | return ctx->ring_sock->sk; | |
1040 | } | |
1041 | #endif | |
1042 | return NULL; | |
1043 | } | |
1044 | EXPORT_SYMBOL(io_uring_get_socket); | |
1045 | ||
f2f87370 PB |
1046 | #define io_for_each_link(pos, head) \ |
1047 | for (pos = (head); pos; pos = pos->link) | |
1048 | ||
3ca405eb PB |
1049 | static inline void io_clean_op(struct io_kiocb *req) |
1050 | { | |
bb175342 PB |
1051 | if (req->flags & (REQ_F_NEED_CLEANUP | REQ_F_BUFFER_SELECTED | |
1052 | REQ_F_INFLIGHT)) | |
3ca405eb PB |
1053 | __io_clean_op(req); |
1054 | } | |
1055 | ||
36f72fe2 PB |
1056 | static inline void io_set_resource_node(struct io_kiocb *req) |
1057 | { | |
1058 | struct io_ring_ctx *ctx = req->ctx; | |
1059 | ||
1060 | if (!req->fixed_file_refs) { | |
1061 | req->fixed_file_refs = &ctx->file_data->node->refs; | |
1062 | percpu_ref_get(req->fixed_file_refs); | |
1063 | } | |
1064 | } | |
1065 | ||
08d23634 PB |
1066 | static bool io_match_task(struct io_kiocb *head, |
1067 | struct task_struct *task, | |
1068 | struct files_struct *files) | |
1069 | { | |
1070 | struct io_kiocb *req; | |
1071 | ||
1072 | if (task && head->task != task) | |
1073 | return false; | |
1074 | if (!files) | |
1075 | return true; | |
1076 | ||
1077 | io_for_each_link(req, head) { | |
1078 | if ((req->flags & REQ_F_WORK_INITIALIZED) && | |
1079 | (req->work.flags & IO_WQ_WORK_FILES) && | |
1080 | req->work.identity->files == files) | |
1081 | return true; | |
1082 | } | |
1083 | return false; | |
1084 | } | |
1085 | ||
28cea78a | 1086 | static void io_sq_thread_drop_mm_files(void) |
c40f6379 | 1087 | { |
28cea78a | 1088 | struct files_struct *files = current->files; |
c40f6379 JA |
1089 | struct mm_struct *mm = current->mm; |
1090 | ||
1091 | if (mm) { | |
1092 | kthread_unuse_mm(mm); | |
1093 | mmput(mm); | |
4b70cf9d | 1094 | current->mm = NULL; |
c40f6379 | 1095 | } |
28cea78a JA |
1096 | if (files) { |
1097 | struct nsproxy *nsproxy = current->nsproxy; | |
1098 | ||
1099 | task_lock(current); | |
1100 | current->files = NULL; | |
1101 | current->nsproxy = NULL; | |
1102 | task_unlock(current); | |
1103 | put_files_struct(files); | |
1104 | put_nsproxy(nsproxy); | |
1105 | } | |
1106 | } | |
1107 | ||
1a38ffc9 | 1108 | static int __io_sq_thread_acquire_files(struct io_ring_ctx *ctx) |
28cea78a | 1109 | { |
621fadc2 PB |
1110 | if (current->flags & PF_EXITING) |
1111 | return -EFAULT; | |
1112 | ||
28cea78a JA |
1113 | if (!current->files) { |
1114 | struct files_struct *files; | |
1115 | struct nsproxy *nsproxy; | |
1116 | ||
1117 | task_lock(ctx->sqo_task); | |
1118 | files = ctx->sqo_task->files; | |
1119 | if (!files) { | |
1120 | task_unlock(ctx->sqo_task); | |
1a38ffc9 | 1121 | return -EOWNERDEAD; |
28cea78a JA |
1122 | } |
1123 | atomic_inc(&files->count); | |
1124 | get_nsproxy(ctx->sqo_task->nsproxy); | |
1125 | nsproxy = ctx->sqo_task->nsproxy; | |
1126 | task_unlock(ctx->sqo_task); | |
1127 | ||
1128 | task_lock(current); | |
1129 | current->files = files; | |
1130 | current->nsproxy = nsproxy; | |
1131 | task_unlock(current); | |
1132 | } | |
1a38ffc9 | 1133 | return 0; |
c40f6379 JA |
1134 | } |
1135 | ||
1136 | static int __io_sq_thread_acquire_mm(struct io_ring_ctx *ctx) | |
1137 | { | |
4b70cf9d JA |
1138 | struct mm_struct *mm; |
1139 | ||
621fadc2 PB |
1140 | if (current->flags & PF_EXITING) |
1141 | return -EFAULT; | |
4b70cf9d JA |
1142 | if (current->mm) |
1143 | return 0; | |
1144 | ||
1145 | /* Should never happen */ | |
1146 | if (unlikely(!(ctx->flags & IORING_SETUP_SQPOLL))) | |
1147 | return -EFAULT; | |
1148 | ||
1149 | task_lock(ctx->sqo_task); | |
1150 | mm = ctx->sqo_task->mm; | |
1151 | if (unlikely(!mm || !mmget_not_zero(mm))) | |
1152 | mm = NULL; | |
1153 | task_unlock(ctx->sqo_task); | |
1154 | ||
1155 | if (mm) { | |
1156 | kthread_use_mm(mm); | |
1157 | return 0; | |
c40f6379 JA |
1158 | } |
1159 | ||
4b70cf9d | 1160 | return -EFAULT; |
c40f6379 JA |
1161 | } |
1162 | ||
28cea78a JA |
1163 | static int io_sq_thread_acquire_mm_files(struct io_ring_ctx *ctx, |
1164 | struct io_kiocb *req) | |
c40f6379 | 1165 | { |
28cea78a | 1166 | const struct io_op_def *def = &io_op_defs[req->opcode]; |
1a38ffc9 | 1167 | int ret; |
28cea78a JA |
1168 | |
1169 | if (def->work_flags & IO_WQ_WORK_MM) { | |
1a38ffc9 | 1170 | ret = __io_sq_thread_acquire_mm(ctx); |
28cea78a JA |
1171 | if (unlikely(ret)) |
1172 | return ret; | |
1173 | } | |
1174 | ||
1a38ffc9 PB |
1175 | if (def->needs_file || (def->work_flags & IO_WQ_WORK_FILES)) { |
1176 | ret = __io_sq_thread_acquire_files(ctx); | |
1177 | if (unlikely(ret)) | |
1178 | return ret; | |
1179 | } | |
28cea78a JA |
1180 | |
1181 | return 0; | |
c40f6379 JA |
1182 | } |
1183 | ||
91d8f519 DZ |
1184 | static void io_sq_thread_associate_blkcg(struct io_ring_ctx *ctx, |
1185 | struct cgroup_subsys_state **cur_css) | |
1186 | ||
1187 | { | |
1188 | #ifdef CONFIG_BLK_CGROUP | |
1189 | /* puts the old one when swapping */ | |
1190 | if (*cur_css != ctx->sqo_blkcg_css) { | |
1191 | kthread_associate_blkcg(ctx->sqo_blkcg_css); | |
1192 | *cur_css = ctx->sqo_blkcg_css; | |
1193 | } | |
1194 | #endif | |
1195 | } | |
1196 | ||
1197 | static void io_sq_thread_unassociate_blkcg(void) | |
1198 | { | |
1199 | #ifdef CONFIG_BLK_CGROUP | |
1200 | kthread_associate_blkcg(NULL); | |
1201 | #endif | |
1202 | } | |
1203 | ||
c40f6379 JA |
1204 | static inline void req_set_fail_links(struct io_kiocb *req) |
1205 | { | |
1206 | if ((req->flags & (REQ_F_LINK | REQ_F_HARDLINK)) == REQ_F_LINK) | |
1207 | req->flags |= REQ_F_FAIL_LINK; | |
1208 | } | |
4a38aed2 | 1209 | |
1e6fa521 JA |
1210 | /* |
1211 | * None of these are dereferenced, they are simply used to check if any of | |
1212 | * them have changed. If we're under current and check they are still the | |
1213 | * same, we're fine to grab references to them for actual out-of-line use. | |
1214 | */ | |
1215 | static void io_init_identity(struct io_identity *id) | |
1216 | { | |
1217 | id->files = current->files; | |
1218 | id->mm = current->mm; | |
1219 | #ifdef CONFIG_BLK_CGROUP | |
1220 | rcu_read_lock(); | |
1221 | id->blkcg_css = blkcg_css(); | |
1222 | rcu_read_unlock(); | |
1223 | #endif | |
1224 | id->creds = current_cred(); | |
1225 | id->nsproxy = current->nsproxy; | |
1226 | id->fs = current->fs; | |
1227 | id->fsize = rlimit(RLIMIT_FSIZE); | |
4ea33a97 JA |
1228 | #ifdef CONFIG_AUDIT |
1229 | id->loginuid = current->loginuid; | |
1230 | id->sessionid = current->sessionid; | |
1231 | #endif | |
1e6fa521 JA |
1232 | refcount_set(&id->count, 1); |
1233 | } | |
1234 | ||
ec99ca6c PB |
1235 | static inline void __io_req_init_async(struct io_kiocb *req) |
1236 | { | |
1237 | memset(&req->work, 0, sizeof(req->work)); | |
1238 | req->flags |= REQ_F_WORK_INITIALIZED; | |
1239 | } | |
1240 | ||
7cdaf587 XW |
1241 | /* |
1242 | * Note: must call io_req_init_async() for the first time you | |
1243 | * touch any members of io_wq_work. | |
1244 | */ | |
1245 | static inline void io_req_init_async(struct io_kiocb *req) | |
1246 | { | |
500a373d JA |
1247 | struct io_uring_task *tctx = current->io_uring; |
1248 | ||
7cdaf587 XW |
1249 | if (req->flags & REQ_F_WORK_INITIALIZED) |
1250 | return; | |
1251 | ||
ec99ca6c | 1252 | __io_req_init_async(req); |
500a373d JA |
1253 | |
1254 | /* Grab a ref if this isn't our static identity */ | |
1255 | req->work.identity = tctx->identity; | |
1256 | if (tctx->identity != &tctx->__identity) | |
1257 | refcount_inc(&req->work.identity->count); | |
7cdaf587 XW |
1258 | } |
1259 | ||
0cdaf760 PB |
1260 | static inline bool io_async_submit(struct io_ring_ctx *ctx) |
1261 | { | |
1262 | return ctx->flags & IORING_SETUP_SQPOLL; | |
1263 | } | |
1264 | ||
2b188cc1 JA |
1265 | static void io_ring_ctx_ref_free(struct percpu_ref *ref) |
1266 | { | |
1267 | struct io_ring_ctx *ctx = container_of(ref, struct io_ring_ctx, refs); | |
1268 | ||
0f158b4c | 1269 | complete(&ctx->ref_comp); |
2b188cc1 JA |
1270 | } |
1271 | ||
8eb7e2d0 PB |
1272 | static inline bool io_is_timeout_noseq(struct io_kiocb *req) |
1273 | { | |
1274 | return !req->timeout.off; | |
1275 | } | |
1276 | ||
2b188cc1 JA |
1277 | static struct io_ring_ctx *io_ring_ctx_alloc(struct io_uring_params *p) |
1278 | { | |
1279 | struct io_ring_ctx *ctx; | |
78076bb6 | 1280 | int hash_bits; |
2b188cc1 JA |
1281 | |
1282 | ctx = kzalloc(sizeof(*ctx), GFP_KERNEL); | |
1283 | if (!ctx) | |
1284 | return NULL; | |
1285 | ||
0ddf92e8 JA |
1286 | ctx->fallback_req = kmem_cache_alloc(req_cachep, GFP_KERNEL); |
1287 | if (!ctx->fallback_req) | |
1288 | goto err; | |
1289 | ||
78076bb6 JA |
1290 | /* |
1291 | * Use 5 bits less than the max cq entries, that should give us around | |
1292 | * 32 entries per hash list if totally full and uniformly spread. | |
1293 | */ | |
1294 | hash_bits = ilog2(p->cq_entries); | |
1295 | hash_bits -= 5; | |
1296 | if (hash_bits <= 0) | |
1297 | hash_bits = 1; | |
1298 | ctx->cancel_hash_bits = hash_bits; | |
1299 | ctx->cancel_hash = kmalloc((1U << hash_bits) * sizeof(struct hlist_head), | |
1300 | GFP_KERNEL); | |
1301 | if (!ctx->cancel_hash) | |
1302 | goto err; | |
1303 | __hash_init(ctx->cancel_hash, 1U << hash_bits); | |
1304 | ||
21482896 | 1305 | if (percpu_ref_init(&ctx->refs, io_ring_ctx_ref_free, |
206aefde JA |
1306 | PERCPU_REF_ALLOW_REINIT, GFP_KERNEL)) |
1307 | goto err; | |
2b188cc1 JA |
1308 | |
1309 | ctx->flags = p->flags; | |
90554200 | 1310 | init_waitqueue_head(&ctx->sqo_sq_wait); |
69fb2131 | 1311 | INIT_LIST_HEAD(&ctx->sqd_list); |
2b188cc1 | 1312 | init_waitqueue_head(&ctx->cq_wait); |
1d7bb1d5 | 1313 | INIT_LIST_HEAD(&ctx->cq_overflow_list); |
0f158b4c JA |
1314 | init_completion(&ctx->ref_comp); |
1315 | init_completion(&ctx->sq_thread_comp); | |
5a2e745d | 1316 | idr_init(&ctx->io_buffer_idr); |
071698e1 | 1317 | idr_init(&ctx->personality_idr); |
2b188cc1 JA |
1318 | mutex_init(&ctx->uring_lock); |
1319 | init_waitqueue_head(&ctx->wait); | |
1320 | spin_lock_init(&ctx->completion_lock); | |
540e32a0 | 1321 | INIT_LIST_HEAD(&ctx->iopoll_list); |
de0617e4 | 1322 | INIT_LIST_HEAD(&ctx->defer_list); |
5262f567 | 1323 | INIT_LIST_HEAD(&ctx->timeout_list); |
fcb323cc JA |
1324 | spin_lock_init(&ctx->inflight_lock); |
1325 | INIT_LIST_HEAD(&ctx->inflight_list); | |
4a38aed2 JA |
1326 | INIT_DELAYED_WORK(&ctx->file_put_work, io_file_put_work); |
1327 | init_llist_head(&ctx->file_put_llist); | |
2b188cc1 | 1328 | return ctx; |
206aefde | 1329 | err: |
0ddf92e8 JA |
1330 | if (ctx->fallback_req) |
1331 | kmem_cache_free(req_cachep, ctx->fallback_req); | |
78076bb6 | 1332 | kfree(ctx->cancel_hash); |
206aefde JA |
1333 | kfree(ctx); |
1334 | return NULL; | |
2b188cc1 JA |
1335 | } |
1336 | ||
9cf7c104 | 1337 | static bool req_need_defer(struct io_kiocb *req, u32 seq) |
7adf4eaf | 1338 | { |
2bc9930e JA |
1339 | if (unlikely(req->flags & REQ_F_IO_DRAIN)) { |
1340 | struct io_ring_ctx *ctx = req->ctx; | |
a197f664 | 1341 | |
9cf7c104 | 1342 | return seq != ctx->cached_cq_tail |
2c3bac6d | 1343 | + READ_ONCE(ctx->cached_cq_overflow); |
2bc9930e | 1344 | } |
de0617e4 | 1345 | |
9d858b21 | 1346 | return false; |
de0617e4 JA |
1347 | } |
1348 | ||
de0617e4 | 1349 | static void __io_commit_cqring(struct io_ring_ctx *ctx) |
2b188cc1 | 1350 | { |
75b28aff | 1351 | struct io_rings *rings = ctx->rings; |
2b188cc1 | 1352 | |
07910158 PB |
1353 | /* order cqe stores with ring update */ |
1354 | smp_store_release(&rings->cq.tail, ctx->cached_cq_tail); | |
2b188cc1 JA |
1355 | } |
1356 | ||
5c3462cf | 1357 | static void io_put_identity(struct io_uring_task *tctx, struct io_kiocb *req) |
1e6fa521 | 1358 | { |
500a373d | 1359 | if (req->work.identity == &tctx->__identity) |
1e6fa521 JA |
1360 | return; |
1361 | if (refcount_dec_and_test(&req->work.identity->count)) | |
1362 | kfree(req->work.identity); | |
1363 | } | |
1364 | ||
4edf20f9 | 1365 | static void io_req_clean_work(struct io_kiocb *req) |
18d9be1a | 1366 | { |
7cdaf587 | 1367 | if (!(req->flags & REQ_F_WORK_INITIALIZED)) |
4edf20f9 | 1368 | return; |
51a4cc11 JA |
1369 | |
1370 | req->flags &= ~REQ_F_WORK_INITIALIZED; | |
7cdaf587 | 1371 | |
dfead8a8 | 1372 | if (req->work.flags & IO_WQ_WORK_MM) { |
98447d65 | 1373 | mmdrop(req->work.identity->mm); |
dfead8a8 | 1374 | req->work.flags &= ~IO_WQ_WORK_MM; |
cccf0ee8 | 1375 | } |
91d8f519 | 1376 | #ifdef CONFIG_BLK_CGROUP |
dfead8a8 | 1377 | if (req->work.flags & IO_WQ_WORK_BLKCG) { |
98447d65 | 1378 | css_put(req->work.identity->blkcg_css); |
dfead8a8 JA |
1379 | req->work.flags &= ~IO_WQ_WORK_BLKCG; |
1380 | } | |
91d8f519 | 1381 | #endif |
dfead8a8 | 1382 | if (req->work.flags & IO_WQ_WORK_CREDS) { |
98447d65 | 1383 | put_cred(req->work.identity->creds); |
dfead8a8 | 1384 | req->work.flags &= ~IO_WQ_WORK_CREDS; |
cccf0ee8 | 1385 | } |
dfead8a8 | 1386 | if (req->work.flags & IO_WQ_WORK_FS) { |
98447d65 | 1387 | struct fs_struct *fs = req->work.identity->fs; |
51a4cc11 | 1388 | |
98447d65 | 1389 | spin_lock(&req->work.identity->fs->lock); |
ff002b30 JA |
1390 | if (--fs->users) |
1391 | fs = NULL; | |
98447d65 | 1392 | spin_unlock(&req->work.identity->fs->lock); |
ff002b30 JA |
1393 | if (fs) |
1394 | free_fs_struct(fs); | |
dfead8a8 | 1395 | req->work.flags &= ~IO_WQ_WORK_FS; |
ff002b30 | 1396 | } |
51a4cc11 | 1397 | |
5c3462cf | 1398 | io_put_identity(req->task->io_uring, req); |
561fb04a JA |
1399 | } |
1400 | ||
1e6fa521 JA |
1401 | /* |
1402 | * Create a private copy of io_identity, since some fields don't match | |
1403 | * the current context. | |
1404 | */ | |
1405 | static bool io_identity_cow(struct io_kiocb *req) | |
1406 | { | |
5c3462cf | 1407 | struct io_uring_task *tctx = current->io_uring; |
1e6fa521 JA |
1408 | const struct cred *creds = NULL; |
1409 | struct io_identity *id; | |
1410 | ||
1411 | if (req->work.flags & IO_WQ_WORK_CREDS) | |
1412 | creds = req->work.identity->creds; | |
1413 | ||
1414 | id = kmemdup(req->work.identity, sizeof(*id), GFP_KERNEL); | |
1415 | if (unlikely(!id)) { | |
1416 | req->work.flags |= IO_WQ_WORK_CANCEL; | |
1417 | return false; | |
1418 | } | |
1419 | ||
1420 | /* | |
1421 | * We can safely just re-init the creds we copied Either the field | |
1422 | * matches the current one, or we haven't grabbed it yet. The only | |
1423 | * exception is ->creds, through registered personalities, so handle | |
1424 | * that one separately. | |
1425 | */ | |
1426 | io_init_identity(id); | |
1427 | if (creds) | |
e8c954df | 1428 | id->creds = creds; |
1e6fa521 JA |
1429 | |
1430 | /* add one for this request */ | |
1431 | refcount_inc(&id->count); | |
1432 | ||
cb8a8ae3 JA |
1433 | /* drop tctx and req identity references, if needed */ |
1434 | if (tctx->identity != &tctx->__identity && | |
1435 | refcount_dec_and_test(&tctx->identity->count)) | |
1436 | kfree(tctx->identity); | |
1437 | if (req->work.identity != &tctx->__identity && | |
1438 | refcount_dec_and_test(&req->work.identity->count)) | |
1e6fa521 JA |
1439 | kfree(req->work.identity); |
1440 | ||
1441 | req->work.identity = id; | |
500a373d | 1442 | tctx->identity = id; |
1e6fa521 JA |
1443 | return true; |
1444 | } | |
1445 | ||
1446 | static bool io_grab_identity(struct io_kiocb *req) | |
18d9be1a | 1447 | { |
d3656344 | 1448 | const struct io_op_def *def = &io_op_defs[req->opcode]; |
5c3462cf | 1449 | struct io_identity *id = req->work.identity; |
23329513 | 1450 | struct io_ring_ctx *ctx = req->ctx; |
54a91f3b | 1451 | |
69228338 JA |
1452 | if (def->work_flags & IO_WQ_WORK_FSIZE) { |
1453 | if (id->fsize != rlimit(RLIMIT_FSIZE)) | |
1454 | return false; | |
1455 | req->work.flags |= IO_WQ_WORK_FSIZE; | |
1456 | } | |
91d8f519 | 1457 | #ifdef CONFIG_BLK_CGROUP |
dfead8a8 JA |
1458 | if (!(req->work.flags & IO_WQ_WORK_BLKCG) && |
1459 | (def->work_flags & IO_WQ_WORK_BLKCG)) { | |
91d8f519 | 1460 | rcu_read_lock(); |
1e6fa521 JA |
1461 | if (id->blkcg_css != blkcg_css()) { |
1462 | rcu_read_unlock(); | |
1463 | return false; | |
1464 | } | |
91d8f519 DZ |
1465 | /* |
1466 | * This should be rare, either the cgroup is dying or the task | |
1467 | * is moving cgroups. Just punt to root for the handful of ios. | |
1468 | */ | |
1e6fa521 | 1469 | if (css_tryget_online(id->blkcg_css)) |
dfead8a8 | 1470 | req->work.flags |= IO_WQ_WORK_BLKCG; |
91d8f519 DZ |
1471 | rcu_read_unlock(); |
1472 | } | |
1473 | #endif | |
dfead8a8 | 1474 | if (!(req->work.flags & IO_WQ_WORK_CREDS)) { |
1e6fa521 JA |
1475 | if (id->creds != current_cred()) |
1476 | return false; | |
1477 | get_cred(id->creds); | |
dfead8a8 JA |
1478 | req->work.flags |= IO_WQ_WORK_CREDS; |
1479 | } | |
4ea33a97 JA |
1480 | #ifdef CONFIG_AUDIT |
1481 | if (!uid_eq(current->loginuid, id->loginuid) || | |
1482 | current->sessionid != id->sessionid) | |
1483 | return false; | |
1484 | #endif | |
dfead8a8 JA |
1485 | if (!(req->work.flags & IO_WQ_WORK_FS) && |
1486 | (def->work_flags & IO_WQ_WORK_FS)) { | |
1e6fa521 JA |
1487 | if (current->fs != id->fs) |
1488 | return false; | |
1489 | spin_lock(&id->fs->lock); | |
1490 | if (!id->fs->in_exec) { | |
1491 | id->fs->users++; | |
dfead8a8 | 1492 | req->work.flags |= IO_WQ_WORK_FS; |
dca9cf8b PB |
1493 | } else { |
1494 | req->work.flags |= IO_WQ_WORK_CANCEL; | |
1495 | } | |
1496 | spin_unlock(¤t->fs->lock); | |
1497 | } | |
af604703 PB |
1498 | if (!(req->work.flags & IO_WQ_WORK_FILES) && |
1499 | (def->work_flags & IO_WQ_WORK_FILES) && | |
1500 | !(req->flags & REQ_F_NO_FILE_TABLE)) { | |
1501 | if (id->files != current->files || | |
1502 | id->nsproxy != current->nsproxy) | |
1503 | return false; | |
1504 | atomic_inc(&id->files->count); | |
1505 | get_nsproxy(id->nsproxy); | |
1506 | req->flags |= REQ_F_INFLIGHT; | |
1507 | ||
1508 | spin_lock_irq(&ctx->inflight_lock); | |
1509 | list_add(&req->inflight_entry, &ctx->inflight_list); | |
1510 | spin_unlock_irq(&ctx->inflight_lock); | |
1511 | req->work.flags |= IO_WQ_WORK_FILES; | |
1512 | } | |
77788775 JA |
1513 | if (!(req->work.flags & IO_WQ_WORK_MM) && |
1514 | (def->work_flags & IO_WQ_WORK_MM)) { | |
1515 | if (id->mm != current->mm) | |
1516 | return false; | |
1517 | mmgrab(id->mm); | |
1518 | req->work.flags |= IO_WQ_WORK_MM; | |
1519 | } | |
1e6fa521 JA |
1520 | |
1521 | return true; | |
1522 | } | |
1523 | ||
1524 | static void io_prep_async_work(struct io_kiocb *req) | |
1525 | { | |
1526 | const struct io_op_def *def = &io_op_defs[req->opcode]; | |
1e6fa521 JA |
1527 | struct io_ring_ctx *ctx = req->ctx; |
1528 | ||
1529 | io_req_init_async(req); | |
1530 | ||
feaadc4f PB |
1531 | if (req->flags & REQ_F_FORCE_ASYNC) |
1532 | req->work.flags |= IO_WQ_WORK_CONCURRENT; | |
1533 | ||
1e6fa521 JA |
1534 | if (req->flags & REQ_F_ISREG) { |
1535 | if (def->hash_reg_file || (ctx->flags & IORING_SETUP_IOPOLL)) | |
1536 | io_wq_hash_work(&req->work, file_inode(req->file)); | |
1537 | } else { | |
1538 | if (def->unbound_nonreg_file) | |
1539 | req->work.flags |= IO_WQ_WORK_UNBOUND; | |
1540 | } | |
1541 | ||
1e6fa521 JA |
1542 | /* if we fail grabbing identity, we must COW, regrab, and retry */ |
1543 | if (io_grab_identity(req)) | |
1544 | return; | |
1545 | ||
1546 | if (!io_identity_cow(req)) | |
1547 | return; | |
1548 | ||
1549 | /* can't fail at this point */ | |
1550 | if (!io_grab_identity(req)) | |
1551 | WARN_ON(1); | |
561fb04a | 1552 | } |
cccf0ee8 | 1553 | |
cbdcb435 | 1554 | static void io_prep_async_link(struct io_kiocb *req) |
561fb04a | 1555 | { |
cbdcb435 | 1556 | struct io_kiocb *cur; |
54a91f3b | 1557 | |
f2f87370 PB |
1558 | io_for_each_link(cur, req) |
1559 | io_prep_async_work(cur); | |
561fb04a JA |
1560 | } |
1561 | ||
7271ef3a | 1562 | static struct io_kiocb *__io_queue_async_work(struct io_kiocb *req) |
561fb04a | 1563 | { |
a197f664 | 1564 | struct io_ring_ctx *ctx = req->ctx; |
cbdcb435 | 1565 | struct io_kiocb *link = io_prep_linked_timeout(req); |
561fb04a | 1566 | |
8766dd51 PB |
1567 | trace_io_uring_queue_async_work(ctx, io_wq_is_hashed(&req->work), req, |
1568 | &req->work, req->flags); | |
1569 | io_wq_enqueue(ctx->io_wq, &req->work); | |
7271ef3a | 1570 | return link; |
18d9be1a JA |
1571 | } |
1572 | ||
cbdcb435 PB |
1573 | static void io_queue_async_work(struct io_kiocb *req) |
1574 | { | |
7271ef3a JA |
1575 | struct io_kiocb *link; |
1576 | ||
cbdcb435 PB |
1577 | /* init ->work of the whole link before punting */ |
1578 | io_prep_async_link(req); | |
7271ef3a JA |
1579 | link = __io_queue_async_work(req); |
1580 | ||
1581 | if (link) | |
1582 | io_queue_linked_timeout(link); | |
cbdcb435 PB |
1583 | } |
1584 | ||
5262f567 JA |
1585 | static void io_kill_timeout(struct io_kiocb *req) |
1586 | { | |
e8c2bc1f | 1587 | struct io_timeout_data *io = req->async_data; |
5262f567 JA |
1588 | int ret; |
1589 | ||
e8c2bc1f | 1590 | ret = hrtimer_try_to_cancel(&io->timer); |
5262f567 | 1591 | if (ret != -1) { |
01cec8c1 PB |
1592 | atomic_set(&req->ctx->cq_timeouts, |
1593 | atomic_read(&req->ctx->cq_timeouts) + 1); | |
135fcde8 | 1594 | list_del_init(&req->timeout.list); |
78e19bbe | 1595 | io_cqring_fill_event(req, 0); |
216578e5 | 1596 | io_put_req_deferred(req, 1); |
5262f567 JA |
1597 | } |
1598 | } | |
1599 | ||
76e1b642 JA |
1600 | /* |
1601 | * Returns true if we found and killed one or more timeouts | |
1602 | */ | |
6b81928d PB |
1603 | static bool io_kill_timeouts(struct io_ring_ctx *ctx, struct task_struct *tsk, |
1604 | struct files_struct *files) | |
5262f567 JA |
1605 | { |
1606 | struct io_kiocb *req, *tmp; | |
76e1b642 | 1607 | int canceled = 0; |
5262f567 JA |
1608 | |
1609 | spin_lock_irq(&ctx->completion_lock); | |
f3606e3a | 1610 | list_for_each_entry_safe(req, tmp, &ctx->timeout_list, timeout.list) { |
6b81928d | 1611 | if (io_match_task(req, tsk, files)) { |
f3606e3a | 1612 | io_kill_timeout(req); |
76e1b642 JA |
1613 | canceled++; |
1614 | } | |
f3606e3a | 1615 | } |
5262f567 | 1616 | spin_unlock_irq(&ctx->completion_lock); |
76e1b642 | 1617 | return canceled != 0; |
5262f567 JA |
1618 | } |
1619 | ||
04518945 | 1620 | static void __io_queue_deferred(struct io_ring_ctx *ctx) |
de0617e4 | 1621 | { |
04518945 | 1622 | do { |
27dc8338 PB |
1623 | struct io_defer_entry *de = list_first_entry(&ctx->defer_list, |
1624 | struct io_defer_entry, list); | |
7271ef3a | 1625 | struct io_kiocb *link; |
de0617e4 | 1626 | |
9cf7c104 | 1627 | if (req_need_defer(de->req, de->seq)) |
04518945 | 1628 | break; |
27dc8338 | 1629 | list_del_init(&de->list); |
cbdcb435 | 1630 | /* punt-init is done before queueing for defer */ |
7271ef3a JA |
1631 | link = __io_queue_async_work(de->req); |
1632 | if (link) { | |
1633 | __io_queue_linked_timeout(link); | |
1634 | /* drop submission reference */ | |
216578e5 | 1635 | io_put_req_deferred(link, 1); |
7271ef3a | 1636 | } |
27dc8338 | 1637 | kfree(de); |
04518945 PB |
1638 | } while (!list_empty(&ctx->defer_list)); |
1639 | } | |
1640 | ||
360428f8 | 1641 | static void io_flush_timeouts(struct io_ring_ctx *ctx) |
de0617e4 | 1642 | { |
f010505b MDG |
1643 | u32 seq; |
1644 | ||
1645 | if (list_empty(&ctx->timeout_list)) | |
1646 | return; | |
1647 | ||
1648 | seq = ctx->cached_cq_tail - atomic_read(&ctx->cq_timeouts); | |
1649 | ||
1650 | do { | |
1651 | u32 events_needed, events_got; | |
360428f8 | 1652 | struct io_kiocb *req = list_first_entry(&ctx->timeout_list, |
135fcde8 | 1653 | struct io_kiocb, timeout.list); |
de0617e4 | 1654 | |
8eb7e2d0 | 1655 | if (io_is_timeout_noseq(req)) |
360428f8 | 1656 | break; |
f010505b MDG |
1657 | |
1658 | /* | |
1659 | * Since seq can easily wrap around over time, subtract | |
1660 | * the last seq at which timeouts were flushed before comparing. | |
1661 | * Assuming not more than 2^31-1 events have happened since, | |
1662 | * these subtractions won't have wrapped, so we can check if | |
1663 | * target is in [last_seq, current_seq] by comparing the two. | |
1664 | */ | |
1665 | events_needed = req->timeout.target_seq - ctx->cq_last_tm_flush; | |
1666 | events_got = seq - ctx->cq_last_tm_flush; | |
1667 | if (events_got < events_needed) | |
360428f8 | 1668 | break; |
bfe68a22 | 1669 | |
135fcde8 | 1670 | list_del_init(&req->timeout.list); |
5262f567 | 1671 | io_kill_timeout(req); |
f010505b MDG |
1672 | } while (!list_empty(&ctx->timeout_list)); |
1673 | ||
1674 | ctx->cq_last_tm_flush = seq; | |
360428f8 | 1675 | } |
5262f567 | 1676 | |
360428f8 PB |
1677 | static void io_commit_cqring(struct io_ring_ctx *ctx) |
1678 | { | |
1679 | io_flush_timeouts(ctx); | |
de0617e4 JA |
1680 | __io_commit_cqring(ctx); |
1681 | ||
04518945 PB |
1682 | if (unlikely(!list_empty(&ctx->defer_list))) |
1683 | __io_queue_deferred(ctx); | |
de0617e4 JA |
1684 | } |
1685 | ||
90554200 JA |
1686 | static inline bool io_sqring_full(struct io_ring_ctx *ctx) |
1687 | { | |
1688 | struct io_rings *r = ctx->rings; | |
1689 | ||
1690 | return READ_ONCE(r->sq.tail) - ctx->cached_sq_head == r->sq_ring_entries; | |
1691 | } | |
1692 | ||
2b188cc1 JA |
1693 | static struct io_uring_cqe *io_get_cqring(struct io_ring_ctx *ctx) |
1694 | { | |
75b28aff | 1695 | struct io_rings *rings = ctx->rings; |
2b188cc1 JA |
1696 | unsigned tail; |
1697 | ||
1698 | tail = ctx->cached_cq_tail; | |
115e12e5 SB |
1699 | /* |
1700 | * writes to the cq entry need to come after reading head; the | |
1701 | * control dependency is enough as we're using WRITE_ONCE to | |
1702 | * fill the cq entry | |
1703 | */ | |
75b28aff | 1704 | if (tail - READ_ONCE(rings->cq.head) == rings->cq_ring_entries) |
2b188cc1 JA |
1705 | return NULL; |
1706 | ||
1707 | ctx->cached_cq_tail++; | |
75b28aff | 1708 | return &rings->cqes[tail & ctx->cq_mask]; |
2b188cc1 JA |
1709 | } |
1710 | ||
f2842ab5 JA |
1711 | static inline bool io_should_trigger_evfd(struct io_ring_ctx *ctx) |
1712 | { | |
f0b493e6 JA |
1713 | if (!ctx->cq_ev_fd) |
1714 | return false; | |
7e55a19c SG |
1715 | if (READ_ONCE(ctx->rings->cq_flags) & IORING_CQ_EVENTFD_DISABLED) |
1716 | return false; | |
f2842ab5 JA |
1717 | if (!ctx->eventfd_async) |
1718 | return true; | |
b41e9852 | 1719 | return io_wq_current_is_worker(); |
f2842ab5 JA |
1720 | } |
1721 | ||
e23de15f PB |
1722 | static inline unsigned __io_cqring_events(struct io_ring_ctx *ctx) |
1723 | { | |
1724 | return ctx->cached_cq_tail - READ_ONCE(ctx->rings->cq.head); | |
1725 | } | |
1726 | ||
b41e9852 | 1727 | static void io_cqring_ev_posted(struct io_ring_ctx *ctx) |
1d7bb1d5 | 1728 | { |
b1445e59 PB |
1729 | /* see waitqueue_active() comment */ |
1730 | smp_mb(); | |
1731 | ||
1d7bb1d5 JA |
1732 | if (waitqueue_active(&ctx->wait)) |
1733 | wake_up(&ctx->wait); | |
534ca6d6 JA |
1734 | if (ctx->sq_data && waitqueue_active(&ctx->sq_data->wait)) |
1735 | wake_up(&ctx->sq_data->wait); | |
b41e9852 | 1736 | if (io_should_trigger_evfd(ctx)) |
1d7bb1d5 | 1737 | eventfd_signal(ctx->cq_ev_fd, 1); |
b1445e59 | 1738 | if (waitqueue_active(&ctx->cq_wait)) { |
4aa84f2f PB |
1739 | wake_up_interruptible(&ctx->cq_wait); |
1740 | kill_fasync(&ctx->cq_fasync, SIGIO, POLL_IN); | |
1741 | } | |
1d7bb1d5 JA |
1742 | } |
1743 | ||
80c18e4a PB |
1744 | static void io_cqring_ev_posted_iopoll(struct io_ring_ctx *ctx) |
1745 | { | |
b1445e59 PB |
1746 | /* see waitqueue_active() comment */ |
1747 | smp_mb(); | |
1748 | ||
80c18e4a PB |
1749 | if (ctx->flags & IORING_SETUP_SQPOLL) { |
1750 | if (waitqueue_active(&ctx->wait)) | |
1751 | wake_up(&ctx->wait); | |
1752 | } | |
1753 | if (io_should_trigger_evfd(ctx)) | |
1754 | eventfd_signal(ctx->cq_ev_fd, 1); | |
b1445e59 | 1755 | if (waitqueue_active(&ctx->cq_wait)) { |
4aa84f2f PB |
1756 | wake_up_interruptible(&ctx->cq_wait); |
1757 | kill_fasync(&ctx->cq_fasync, SIGIO, POLL_IN); | |
1758 | } | |
80c18e4a PB |
1759 | } |
1760 | ||
c4a2ed72 | 1761 | /* Returns true if there are no backlogged entries after the flush */ |
6c503150 PB |
1762 | static bool __io_cqring_overflow_flush(struct io_ring_ctx *ctx, bool force, |
1763 | struct task_struct *tsk, | |
1764 | struct files_struct *files) | |
1d7bb1d5 JA |
1765 | { |
1766 | struct io_rings *rings = ctx->rings; | |
e6c8aa9a | 1767 | struct io_kiocb *req, *tmp; |
1d7bb1d5 | 1768 | struct io_uring_cqe *cqe; |
1d7bb1d5 | 1769 | unsigned long flags; |
09e88404 | 1770 | bool all_flushed; |
1d7bb1d5 JA |
1771 | LIST_HEAD(list); |
1772 | ||
e23de15f PB |
1773 | if (!force && __io_cqring_events(ctx) == rings->cq_ring_entries) |
1774 | return false; | |
1d7bb1d5 JA |
1775 | |
1776 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
e6c8aa9a | 1777 | list_for_each_entry_safe(req, tmp, &ctx->cq_overflow_list, compl.list) { |
08d23634 | 1778 | if (!io_match_task(req, tsk, files)) |
e6c8aa9a JA |
1779 | continue; |
1780 | ||
1d7bb1d5 JA |
1781 | cqe = io_get_cqring(ctx); |
1782 | if (!cqe && !force) | |
1783 | break; | |
1784 | ||
40d8ddd4 | 1785 | list_move(&req->compl.list, &list); |
1d7bb1d5 JA |
1786 | if (cqe) { |
1787 | WRITE_ONCE(cqe->user_data, req->user_data); | |
1788 | WRITE_ONCE(cqe->res, req->result); | |
0f7e466b | 1789 | WRITE_ONCE(cqe->flags, req->compl.cflags); |
1d7bb1d5 | 1790 | } else { |
2c3bac6d | 1791 | ctx->cached_cq_overflow++; |
1d7bb1d5 | 1792 | WRITE_ONCE(ctx->rings->cq_overflow, |
2c3bac6d | 1793 | ctx->cached_cq_overflow); |
1d7bb1d5 JA |
1794 | } |
1795 | } | |
1796 | ||
09e88404 PB |
1797 | all_flushed = list_empty(&ctx->cq_overflow_list); |
1798 | if (all_flushed) { | |
1799 | clear_bit(0, &ctx->sq_check_overflow); | |
1800 | clear_bit(0, &ctx->cq_check_overflow); | |
1801 | ctx->rings->sq_flags &= ~IORING_SQ_CQ_OVERFLOW; | |
1802 | } | |
46930143 | 1803 | |
09e88404 | 1804 | io_commit_cqring(ctx); |
1d7bb1d5 JA |
1805 | spin_unlock_irqrestore(&ctx->completion_lock, flags); |
1806 | io_cqring_ev_posted(ctx); | |
1807 | ||
1808 | while (!list_empty(&list)) { | |
40d8ddd4 PB |
1809 | req = list_first_entry(&list, struct io_kiocb, compl.list); |
1810 | list_del(&req->compl.list); | |
ec9c02ad | 1811 | io_put_req(req); |
1d7bb1d5 | 1812 | } |
c4a2ed72 | 1813 | |
09e88404 | 1814 | return all_flushed; |
1d7bb1d5 JA |
1815 | } |
1816 | ||
6c503150 PB |
1817 | static void io_cqring_overflow_flush(struct io_ring_ctx *ctx, bool force, |
1818 | struct task_struct *tsk, | |
1819 | struct files_struct *files) | |
1820 | { | |
1821 | if (test_bit(0, &ctx->cq_check_overflow)) { | |
1822 | /* iopoll syncs against uring_lock, not completion_lock */ | |
1823 | if (ctx->flags & IORING_SETUP_IOPOLL) | |
1824 | mutex_lock(&ctx->uring_lock); | |
1825 | __io_cqring_overflow_flush(ctx, force, tsk, files); | |
1826 | if (ctx->flags & IORING_SETUP_IOPOLL) | |
1827 | mutex_unlock(&ctx->uring_lock); | |
1828 | } | |
1829 | } | |
1830 | ||
bcda7baa | 1831 | static void __io_cqring_fill_event(struct io_kiocb *req, long res, long cflags) |
2b188cc1 | 1832 | { |
78e19bbe | 1833 | struct io_ring_ctx *ctx = req->ctx; |
2b188cc1 JA |
1834 | struct io_uring_cqe *cqe; |
1835 | ||
78e19bbe | 1836 | trace_io_uring_complete(ctx, req->user_data, res); |
51c3ff62 | 1837 | |
2b188cc1 JA |
1838 | /* |
1839 | * If we can't get a cq entry, userspace overflowed the | |
1840 | * submission (by quite a lot). Increment the overflow count in | |
1841 | * the ring. | |
1842 | */ | |
1843 | cqe = io_get_cqring(ctx); | |
1d7bb1d5 | 1844 | if (likely(cqe)) { |
78e19bbe | 1845 | WRITE_ONCE(cqe->user_data, req->user_data); |
2b188cc1 | 1846 | WRITE_ONCE(cqe->res, res); |
bcda7baa | 1847 | WRITE_ONCE(cqe->flags, cflags); |
fdaf083c JA |
1848 | } else if (ctx->cq_overflow_flushed || |
1849 | atomic_read(&req->task->io_uring->in_idle)) { | |
0f212204 JA |
1850 | /* |
1851 | * If we're in ring overflow flush mode, or in task cancel mode, | |
1852 | * then we cannot store the request for later flushing, we need | |
1853 | * to drop it on the floor. | |
1854 | */ | |
2c3bac6d PB |
1855 | ctx->cached_cq_overflow++; |
1856 | WRITE_ONCE(ctx->rings->cq_overflow, ctx->cached_cq_overflow); | |
1d7bb1d5 | 1857 | } else { |
ad3eb2c8 JA |
1858 | if (list_empty(&ctx->cq_overflow_list)) { |
1859 | set_bit(0, &ctx->sq_check_overflow); | |
1860 | set_bit(0, &ctx->cq_check_overflow); | |
6d5f9049 | 1861 | ctx->rings->sq_flags |= IORING_SQ_CQ_OVERFLOW; |
ad3eb2c8 | 1862 | } |
40d8ddd4 | 1863 | io_clean_op(req); |
1d7bb1d5 | 1864 | req->result = res; |
0f7e466b | 1865 | req->compl.cflags = cflags; |
40d8ddd4 PB |
1866 | refcount_inc(&req->refs); |
1867 | list_add_tail(&req->compl.list, &ctx->cq_overflow_list); | |
2b188cc1 JA |
1868 | } |
1869 | } | |
1870 | ||
bcda7baa JA |
1871 | static void io_cqring_fill_event(struct io_kiocb *req, long res) |
1872 | { | |
1873 | __io_cqring_fill_event(req, res, 0); | |
1874 | } | |
1875 | ||
e1e16097 | 1876 | static void io_cqring_add_event(struct io_kiocb *req, long res, long cflags) |
2b188cc1 | 1877 | { |
78e19bbe | 1878 | struct io_ring_ctx *ctx = req->ctx; |
2b188cc1 JA |
1879 | unsigned long flags; |
1880 | ||
1881 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
bcda7baa | 1882 | __io_cqring_fill_event(req, res, cflags); |
2b188cc1 JA |
1883 | io_commit_cqring(ctx); |
1884 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
1885 | ||
8c838788 | 1886 | io_cqring_ev_posted(ctx); |
2b188cc1 JA |
1887 | } |
1888 | ||
229a7b63 | 1889 | static void io_submit_flush_completions(struct io_comp_state *cs) |
bcda7baa | 1890 | { |
229a7b63 JA |
1891 | struct io_ring_ctx *ctx = cs->ctx; |
1892 | ||
1893 | spin_lock_irq(&ctx->completion_lock); | |
1894 | while (!list_empty(&cs->list)) { | |
1895 | struct io_kiocb *req; | |
1896 | ||
3ca405eb PB |
1897 | req = list_first_entry(&cs->list, struct io_kiocb, compl.list); |
1898 | list_del(&req->compl.list); | |
0f7e466b | 1899 | __io_cqring_fill_event(req, req->result, req->compl.cflags); |
216578e5 PB |
1900 | |
1901 | /* | |
1902 | * io_free_req() doesn't care about completion_lock unless one | |
1903 | * of these flags is set. REQ_F_WORK_INITIALIZED is in the list | |
1904 | * because of a potential deadlock with req->work.fs->lock | |
1905 | */ | |
1906 | if (req->flags & (REQ_F_FAIL_LINK|REQ_F_LINK_TIMEOUT | |
1907 | |REQ_F_WORK_INITIALIZED)) { | |
229a7b63 JA |
1908 | spin_unlock_irq(&ctx->completion_lock); |
1909 | io_put_req(req); | |
1910 | spin_lock_irq(&ctx->completion_lock); | |
216578e5 PB |
1911 | } else { |
1912 | io_put_req(req); | |
229a7b63 JA |
1913 | } |
1914 | } | |
1915 | io_commit_cqring(ctx); | |
1916 | spin_unlock_irq(&ctx->completion_lock); | |
1917 | ||
1918 | io_cqring_ev_posted(ctx); | |
1919 | cs->nr = 0; | |
1920 | } | |
1921 | ||
1922 | static void __io_req_complete(struct io_kiocb *req, long res, unsigned cflags, | |
1923 | struct io_comp_state *cs) | |
1924 | { | |
1925 | if (!cs) { | |
1926 | io_cqring_add_event(req, res, cflags); | |
1927 | io_put_req(req); | |
1928 | } else { | |
3ca405eb | 1929 | io_clean_op(req); |
229a7b63 | 1930 | req->result = res; |
0f7e466b | 1931 | req->compl.cflags = cflags; |
3ca405eb | 1932 | list_add_tail(&req->compl.list, &cs->list); |
229a7b63 JA |
1933 | if (++cs->nr >= 32) |
1934 | io_submit_flush_completions(cs); | |
1935 | } | |
e1e16097 JA |
1936 | } |
1937 | ||
1938 | static void io_req_complete(struct io_kiocb *req, long res) | |
bcda7baa | 1939 | { |
229a7b63 | 1940 | __io_req_complete(req, res, 0, NULL); |
bcda7baa JA |
1941 | } |
1942 | ||
0ddf92e8 JA |
1943 | static inline bool io_is_fallback_req(struct io_kiocb *req) |
1944 | { | |
1945 | return req == (struct io_kiocb *) | |
1946 | ((unsigned long) req->ctx->fallback_req & ~1UL); | |
1947 | } | |
1948 | ||
1949 | static struct io_kiocb *io_get_fallback_req(struct io_ring_ctx *ctx) | |
1950 | { | |
1951 | struct io_kiocb *req; | |
1952 | ||
1953 | req = ctx->fallback_req; | |
dd461af6 | 1954 | if (!test_and_set_bit_lock(0, (unsigned long *) &ctx->fallback_req)) |
0ddf92e8 JA |
1955 | return req; |
1956 | ||
1957 | return NULL; | |
1958 | } | |
1959 | ||
0553b8bd PB |
1960 | static struct io_kiocb *io_alloc_req(struct io_ring_ctx *ctx, |
1961 | struct io_submit_state *state) | |
2b188cc1 | 1962 | { |
f6b6c7d6 | 1963 | if (!state->free_reqs) { |
291b2821 | 1964 | gfp_t gfp = GFP_KERNEL | __GFP_NOWARN; |
2579f913 JA |
1965 | size_t sz; |
1966 | int ret; | |
1967 | ||
1968 | sz = min_t(size_t, state->ios_left, ARRAY_SIZE(state->reqs)); | |
fd6fab2c JA |
1969 | ret = kmem_cache_alloc_bulk(req_cachep, gfp, sz, state->reqs); |
1970 | ||
1971 | /* | |
1972 | * Bulk alloc is all-or-nothing. If we fail to get a batch, | |
1973 | * retry single alloc to be on the safe side. | |
1974 | */ | |
1975 | if (unlikely(ret <= 0)) { | |
1976 | state->reqs[0] = kmem_cache_alloc(req_cachep, gfp); | |
1977 | if (!state->reqs[0]) | |
0ddf92e8 | 1978 | goto fallback; |
fd6fab2c JA |
1979 | ret = 1; |
1980 | } | |
291b2821 | 1981 | state->free_reqs = ret; |
2b188cc1 JA |
1982 | } |
1983 | ||
291b2821 PB |
1984 | state->free_reqs--; |
1985 | return state->reqs[state->free_reqs]; | |
0ddf92e8 | 1986 | fallback: |
0553b8bd | 1987 | return io_get_fallback_req(ctx); |
2b188cc1 JA |
1988 | } |
1989 | ||
8da11c19 PB |
1990 | static inline void io_put_file(struct io_kiocb *req, struct file *file, |
1991 | bool fixed) | |
1992 | { | |
36f72fe2 | 1993 | if (!fixed) |
8da11c19 PB |
1994 | fput(file); |
1995 | } | |
1996 | ||
4edf20f9 | 1997 | static void io_dismantle_req(struct io_kiocb *req) |
2b188cc1 | 1998 | { |
3ca405eb | 1999 | io_clean_op(req); |
929a3af9 | 2000 | |
e8c2bc1f JA |
2001 | if (req->async_data) |
2002 | kfree(req->async_data); | |
8da11c19 PB |
2003 | if (req->file) |
2004 | io_put_file(req, req->file, (req->flags & REQ_F_FIXED_FILE)); | |
36f72fe2 PB |
2005 | if (req->fixed_file_refs) |
2006 | percpu_ref_put(req->fixed_file_refs); | |
4edf20f9 | 2007 | io_req_clean_work(req); |
e65ef56d JA |
2008 | } |
2009 | ||
216578e5 | 2010 | static void __io_free_req(struct io_kiocb *req) |
c6ca97b3 | 2011 | { |
0f212204 | 2012 | struct io_uring_task *tctx = req->task->io_uring; |
51a4cc11 | 2013 | struct io_ring_ctx *ctx = req->ctx; |
c6ca97b3 | 2014 | |
216578e5 | 2015 | io_dismantle_req(req); |
c6ca97b3 | 2016 | |
d8a6df10 | 2017 | percpu_counter_dec(&tctx->inflight); |
fdaf083c | 2018 | if (atomic_read(&tctx->in_idle)) |
0f212204 | 2019 | wake_up(&tctx->wait); |
e3bc8e9d JA |
2020 | put_task_struct(req->task); |
2021 | ||
b1e50e54 PB |
2022 | if (likely(!io_is_fallback_req(req))) |
2023 | kmem_cache_free(req_cachep, req); | |
2024 | else | |
ecfc5177 PB |
2025 | clear_bit_unlock(0, (unsigned long *) &ctx->fallback_req); |
2026 | percpu_ref_put(&ctx->refs); | |
e65ef56d JA |
2027 | } |
2028 | ||
f2f87370 PB |
2029 | static inline void io_remove_next_linked(struct io_kiocb *req) |
2030 | { | |
2031 | struct io_kiocb *nxt = req->link; | |
2032 | ||
2033 | req->link = nxt->link; | |
2034 | nxt->link = NULL; | |
2035 | } | |
2036 | ||
c9abd7ad | 2037 | static void io_kill_linked_timeout(struct io_kiocb *req) |
2665abfd | 2038 | { |
a197f664 | 2039 | struct io_ring_ctx *ctx = req->ctx; |
7c86ffee | 2040 | struct io_kiocb *link; |
c9abd7ad PB |
2041 | bool cancelled = false; |
2042 | unsigned long flags; | |
7c86ffee | 2043 | |
c9abd7ad | 2044 | spin_lock_irqsave(&ctx->completion_lock, flags); |
f2f87370 PB |
2045 | link = req->link; |
2046 | ||
900fad45 PB |
2047 | /* |
2048 | * Can happen if a linked timeout fired and link had been like | |
2049 | * req -> link t-out -> link t-out [-> ...] | |
2050 | */ | |
c9abd7ad PB |
2051 | if (link && (link->flags & REQ_F_LTIMEOUT_ACTIVE)) { |
2052 | struct io_timeout_data *io = link->async_data; | |
2053 | int ret; | |
7c86ffee | 2054 | |
f2f87370 | 2055 | io_remove_next_linked(req); |
90cd7e42 | 2056 | link->timeout.head = NULL; |
c9abd7ad PB |
2057 | ret = hrtimer_try_to_cancel(&io->timer); |
2058 | if (ret != -1) { | |
2059 | io_cqring_fill_event(link, -ECANCELED); | |
2060 | io_commit_cqring(ctx); | |
2061 | cancelled = true; | |
2062 | } | |
2063 | } | |
7c86ffee | 2064 | req->flags &= ~REQ_F_LINK_TIMEOUT; |
216578e5 | 2065 | spin_unlock_irqrestore(&ctx->completion_lock, flags); |
ab0b6451 | 2066 | |
c9abd7ad | 2067 | if (cancelled) { |
7c86ffee | 2068 | io_cqring_ev_posted(ctx); |
c9abd7ad PB |
2069 | io_put_req(link); |
2070 | } | |
7c86ffee PB |
2071 | } |
2072 | ||
9e645e11 | 2073 | |
d148ca4b | 2074 | static void io_fail_links(struct io_kiocb *req) |
9e645e11 | 2075 | { |
f2f87370 | 2076 | struct io_kiocb *link, *nxt; |
2665abfd | 2077 | struct io_ring_ctx *ctx = req->ctx; |
d148ca4b | 2078 | unsigned long flags; |
9e645e11 | 2079 | |
d148ca4b | 2080 | spin_lock_irqsave(&ctx->completion_lock, flags); |
f2f87370 PB |
2081 | link = req->link; |
2082 | req->link = NULL; | |
9e645e11 | 2083 | |
f2f87370 PB |
2084 | while (link) { |
2085 | nxt = link->link; | |
2086 | link->link = NULL; | |
2665abfd | 2087 | |
f2f87370 | 2088 | trace_io_uring_fail_link(req, link); |
7c86ffee | 2089 | io_cqring_fill_event(link, -ECANCELED); |
216578e5 PB |
2090 | |
2091 | /* | |
2092 | * It's ok to free under spinlock as they're not linked anymore, | |
2093 | * but avoid REQ_F_WORK_INITIALIZED because it may deadlock on | |
2094 | * work.fs->lock. | |
2095 | */ | |
2096 | if (link->flags & REQ_F_WORK_INITIALIZED) | |
2097 | io_put_req_deferred(link, 2); | |
2098 | else | |
2099 | io_double_put_req(link); | |
f2f87370 | 2100 | link = nxt; |
9e645e11 | 2101 | } |
2665abfd | 2102 | io_commit_cqring(ctx); |
216578e5 | 2103 | spin_unlock_irqrestore(&ctx->completion_lock, flags); |
9e645e11 | 2104 | |
2665abfd | 2105 | io_cqring_ev_posted(ctx); |
9e645e11 JA |
2106 | } |
2107 | ||
3fa5e0f3 | 2108 | static struct io_kiocb *__io_req_find_next(struct io_kiocb *req) |
c69f8dbe | 2109 | { |
7c86ffee PB |
2110 | if (req->flags & REQ_F_LINK_TIMEOUT) |
2111 | io_kill_linked_timeout(req); | |
944e58bf | 2112 | |
9e645e11 JA |
2113 | /* |
2114 | * If LINK is set, we have dependent requests in this chain. If we | |
2115 | * didn't fail this request, queue the first one up, moving any other | |
2116 | * dependencies to the next request. In case of failure, fail the rest | |
2117 | * of the chain. | |
2118 | */ | |
f2f87370 PB |
2119 | if (likely(!(req->flags & REQ_F_FAIL_LINK))) { |
2120 | struct io_kiocb *nxt = req->link; | |
2121 | ||
2122 | req->link = NULL; | |
2123 | return nxt; | |
2124 | } | |
9b5f7bd9 PB |
2125 | io_fail_links(req); |
2126 | return NULL; | |
4d7dd462 | 2127 | } |
9e645e11 | 2128 | |
f2f87370 | 2129 | static inline struct io_kiocb *io_req_find_next(struct io_kiocb *req) |
3fa5e0f3 | 2130 | { |
f2f87370 | 2131 | if (likely(!(req->link) && !(req->flags & REQ_F_LINK_TIMEOUT))) |
3fa5e0f3 PB |
2132 | return NULL; |
2133 | return __io_req_find_next(req); | |
2134 | } | |
2135 | ||
355fb9e2 | 2136 | static int io_req_task_work_add(struct io_kiocb *req) |
c2c4c83c JA |
2137 | { |
2138 | struct task_struct *tsk = req->task; | |
2139 | struct io_ring_ctx *ctx = req->ctx; | |
91989c70 JA |
2140 | enum task_work_notify_mode notify; |
2141 | int ret; | |
c2c4c83c | 2142 | |
6200b0ae JA |
2143 | if (tsk->flags & PF_EXITING) |
2144 | return -ESRCH; | |
2145 | ||
c2c4c83c | 2146 | /* |
0ba9c9ed JA |
2147 | * SQPOLL kernel thread doesn't need notification, just a wakeup. For |
2148 | * all other cases, use TWA_SIGNAL unconditionally to ensure we're | |
2149 | * processing task_work. There's no reliable way to tell if TWA_RESUME | |
2150 | * will do the job. | |
c2c4c83c | 2151 | */ |
91989c70 | 2152 | notify = TWA_NONE; |
355fb9e2 | 2153 | if (!(ctx->flags & IORING_SETUP_SQPOLL)) |
c2c4c83c JA |
2154 | notify = TWA_SIGNAL; |
2155 | ||
87c4311f | 2156 | ret = task_work_add(tsk, &req->task_work, notify); |
c2c4c83c JA |
2157 | if (!ret) |
2158 | wake_up_process(tsk); | |
0ba9c9ed | 2159 | |
c2c4c83c JA |
2160 | return ret; |
2161 | } | |
2162 | ||
c40f6379 JA |
2163 | static void __io_req_task_cancel(struct io_kiocb *req, int error) |
2164 | { | |
2165 | struct io_ring_ctx *ctx = req->ctx; | |
2166 | ||
2167 | spin_lock_irq(&ctx->completion_lock); | |
2168 | io_cqring_fill_event(req, error); | |
2169 | io_commit_cqring(ctx); | |
2170 | spin_unlock_irq(&ctx->completion_lock); | |
2171 | ||
2172 | io_cqring_ev_posted(ctx); | |
2173 | req_set_fail_links(req); | |
2174 | io_double_put_req(req); | |
2175 | } | |
2176 | ||
2177 | static void io_req_task_cancel(struct callback_head *cb) | |
2178 | { | |
2179 | struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work); | |
87ceb6a6 | 2180 | struct io_ring_ctx *ctx = req->ctx; |
c40f6379 JA |
2181 | |
2182 | __io_req_task_cancel(req, -ECANCELED); | |
87ceb6a6 | 2183 | percpu_ref_put(&ctx->refs); |
c40f6379 JA |
2184 | } |
2185 | ||
2186 | static void __io_req_task_submit(struct io_kiocb *req) | |
2187 | { | |
2188 | struct io_ring_ctx *ctx = req->ctx; | |
2189 | ||
81b6d05c | 2190 | mutex_lock(&ctx->uring_lock); |
d9d05217 PB |
2191 | if (!ctx->sqo_dead && |
2192 | !__io_sq_thread_acquire_mm(ctx) && | |
2193 | !__io_sq_thread_acquire_files(ctx)) | |
c1379e24 | 2194 | __io_queue_sqe(req, NULL); |
81b6d05c | 2195 | else |
c40f6379 | 2196 | __io_req_task_cancel(req, -EFAULT); |
81b6d05c | 2197 | mutex_unlock(&ctx->uring_lock); |
c40f6379 JA |
2198 | } |
2199 | ||
2200 | static void io_req_task_submit(struct callback_head *cb) | |
2201 | { | |
2202 | struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work); | |
6d816e08 | 2203 | struct io_ring_ctx *ctx = req->ctx; |
c40f6379 JA |
2204 | |
2205 | __io_req_task_submit(req); | |
6d816e08 | 2206 | percpu_ref_put(&ctx->refs); |
c40f6379 JA |
2207 | } |
2208 | ||
2209 | static void io_req_task_queue(struct io_kiocb *req) | |
2210 | { | |
c40f6379 JA |
2211 | int ret; |
2212 | ||
2213 | init_task_work(&req->task_work, io_req_task_submit); | |
6d816e08 | 2214 | percpu_ref_get(&req->ctx->refs); |
c40f6379 | 2215 | |
355fb9e2 | 2216 | ret = io_req_task_work_add(req); |
c40f6379 | 2217 | if (unlikely(ret)) { |
c2c4c83c JA |
2218 | struct task_struct *tsk; |
2219 | ||
c40f6379 JA |
2220 | init_task_work(&req->task_work, io_req_task_cancel); |
2221 | tsk = io_wq_get_task(req->ctx->io_wq); | |
91989c70 | 2222 | task_work_add(tsk, &req->task_work, TWA_NONE); |
c2c4c83c | 2223 | wake_up_process(tsk); |
c40f6379 | 2224 | } |
c40f6379 JA |
2225 | } |
2226 | ||
f2f87370 | 2227 | static inline void io_queue_next(struct io_kiocb *req) |
c69f8dbe | 2228 | { |
9b5f7bd9 | 2229 | struct io_kiocb *nxt = io_req_find_next(req); |
944e58bf PB |
2230 | |
2231 | if (nxt) | |
906a8c3f | 2232 | io_req_task_queue(nxt); |
c69f8dbe JL |
2233 | } |
2234 | ||
c3524383 | 2235 | static void io_free_req(struct io_kiocb *req) |
7a743e22 | 2236 | { |
c3524383 PB |
2237 | io_queue_next(req); |
2238 | __io_free_req(req); | |
2239 | } | |
8766dd51 | 2240 | |
2d6500d4 PB |
2241 | struct req_batch { |
2242 | void *reqs[IO_IOPOLL_BATCH]; | |
2243 | int to_free; | |
7a743e22 | 2244 | |
5af1d13e PB |
2245 | struct task_struct *task; |
2246 | int task_refs; | |
2d6500d4 PB |
2247 | }; |
2248 | ||
5af1d13e PB |
2249 | static inline void io_init_req_batch(struct req_batch *rb) |
2250 | { | |
2251 | rb->to_free = 0; | |
2252 | rb->task_refs = 0; | |
2253 | rb->task = NULL; | |
2254 | } | |
2255 | ||
2d6500d4 PB |
2256 | static void __io_req_free_batch_flush(struct io_ring_ctx *ctx, |
2257 | struct req_batch *rb) | |
2258 | { | |
2259 | kmem_cache_free_bulk(req_cachep, rb->to_free, rb->reqs); | |
2260 | percpu_ref_put_many(&ctx->refs, rb->to_free); | |
2261 | rb->to_free = 0; | |
2262 | } | |
2263 | ||
2264 | static void io_req_free_batch_finish(struct io_ring_ctx *ctx, | |
2265 | struct req_batch *rb) | |
2266 | { | |
2267 | if (rb->to_free) | |
2268 | __io_req_free_batch_flush(ctx, rb); | |
5af1d13e | 2269 | if (rb->task) { |
d8a6df10 JA |
2270 | struct io_uring_task *tctx = rb->task->io_uring; |
2271 | ||
2272 | percpu_counter_sub(&tctx->inflight, rb->task_refs); | |
c93cc9e1 JA |
2273 | if (atomic_read(&tctx->in_idle)) |
2274 | wake_up(&tctx->wait); | |
5af1d13e PB |
2275 | put_task_struct_many(rb->task, rb->task_refs); |
2276 | rb->task = NULL; | |
2277 | } | |
2d6500d4 PB |
2278 | } |
2279 | ||
2280 | static void io_req_free_batch(struct req_batch *rb, struct io_kiocb *req) | |
2281 | { | |
2282 | if (unlikely(io_is_fallback_req(req))) { | |
2283 | io_free_req(req); | |
2284 | return; | |
2285 | } | |
f2f87370 | 2286 | io_queue_next(req); |
2d6500d4 | 2287 | |
e3bc8e9d | 2288 | if (req->task != rb->task) { |
0f212204 | 2289 | if (rb->task) { |
d8a6df10 JA |
2290 | struct io_uring_task *tctx = rb->task->io_uring; |
2291 | ||
2292 | percpu_counter_sub(&tctx->inflight, rb->task_refs); | |
c93cc9e1 JA |
2293 | if (atomic_read(&tctx->in_idle)) |
2294 | wake_up(&tctx->wait); | |
e3bc8e9d | 2295 | put_task_struct_many(rb->task, rb->task_refs); |
5af1d13e | 2296 | } |
e3bc8e9d JA |
2297 | rb->task = req->task; |
2298 | rb->task_refs = 0; | |
5af1d13e | 2299 | } |
e3bc8e9d | 2300 | rb->task_refs++; |
5af1d13e | 2301 | |
4edf20f9 | 2302 | io_dismantle_req(req); |
2d6500d4 PB |
2303 | rb->reqs[rb->to_free++] = req; |
2304 | if (unlikely(rb->to_free == ARRAY_SIZE(rb->reqs))) | |
2305 | __io_req_free_batch_flush(req->ctx, rb); | |
7a743e22 PB |
2306 | } |
2307 | ||
ba816ad6 JA |
2308 | /* |
2309 | * Drop reference to request, return next in chain (if there is one) if this | |
2310 | * was the last reference to this request. | |
2311 | */ | |
9b5f7bd9 | 2312 | static struct io_kiocb *io_put_req_find_next(struct io_kiocb *req) |
e65ef56d | 2313 | { |
9b5f7bd9 PB |
2314 | struct io_kiocb *nxt = NULL; |
2315 | ||
2a44f467 | 2316 | if (refcount_dec_and_test(&req->refs)) { |
9b5f7bd9 | 2317 | nxt = io_req_find_next(req); |
4d7dd462 | 2318 | __io_free_req(req); |
2a44f467 | 2319 | } |
9b5f7bd9 | 2320 | return nxt; |
2b188cc1 JA |
2321 | } |
2322 | ||
e65ef56d JA |
2323 | static void io_put_req(struct io_kiocb *req) |
2324 | { | |
2325 | if (refcount_dec_and_test(&req->refs)) | |
2326 | io_free_req(req); | |
2b188cc1 JA |
2327 | } |
2328 | ||
216578e5 PB |
2329 | static void io_put_req_deferred_cb(struct callback_head *cb) |
2330 | { | |
2331 | struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work); | |
2332 | ||
2333 | io_free_req(req); | |
2334 | } | |
2335 | ||
2336 | static void io_free_req_deferred(struct io_kiocb *req) | |
2337 | { | |
2338 | int ret; | |
2339 | ||
2340 | init_task_work(&req->task_work, io_put_req_deferred_cb); | |
355fb9e2 | 2341 | ret = io_req_task_work_add(req); |
216578e5 PB |
2342 | if (unlikely(ret)) { |
2343 | struct task_struct *tsk; | |
2344 | ||
2345 | tsk = io_wq_get_task(req->ctx->io_wq); | |
91989c70 | 2346 | task_work_add(tsk, &req->task_work, TWA_NONE); |
216578e5 PB |
2347 | wake_up_process(tsk); |
2348 | } | |
2349 | } | |
2350 | ||
2351 | static inline void io_put_req_deferred(struct io_kiocb *req, int refs) | |
2352 | { | |
2353 | if (refcount_sub_and_test(refs, &req->refs)) | |
2354 | io_free_req_deferred(req); | |
2355 | } | |
2356 | ||
f4db7182 | 2357 | static struct io_wq_work *io_steal_work(struct io_kiocb *req) |
7a743e22 | 2358 | { |
6df1db6b | 2359 | struct io_kiocb *nxt; |
f4db7182 | 2360 | |
7a743e22 | 2361 | /* |
f4db7182 PB |
2362 | * A ref is owned by io-wq in which context we're. So, if that's the |
2363 | * last one, it's safe to steal next work. False negatives are Ok, | |
2364 | * it just will be re-punted async in io_put_work() | |
7a743e22 | 2365 | */ |
f4db7182 PB |
2366 | if (refcount_read(&req->refs) != 1) |
2367 | return NULL; | |
7a743e22 | 2368 | |
9b5f7bd9 | 2369 | nxt = io_req_find_next(req); |
6df1db6b | 2370 | return nxt ? &nxt->work : NULL; |
7a743e22 PB |
2371 | } |
2372 | ||
978db57e JA |
2373 | static void io_double_put_req(struct io_kiocb *req) |
2374 | { | |
2375 | /* drop both submit and complete references */ | |
2376 | if (refcount_sub_and_test(2, &req->refs)) | |
2377 | io_free_req(req); | |
2378 | } | |
2379 | ||
6c503150 | 2380 | static unsigned io_cqring_events(struct io_ring_ctx *ctx) |
a3a0e43f JA |
2381 | { |
2382 | /* See comment at the top of this file */ | |
2383 | smp_rmb(); | |
e23de15f | 2384 | return __io_cqring_events(ctx); |
a3a0e43f JA |
2385 | } |
2386 | ||
fb5ccc98 PB |
2387 | static inline unsigned int io_sqring_entries(struct io_ring_ctx *ctx) |
2388 | { | |
2389 | struct io_rings *rings = ctx->rings; | |
2390 | ||
2391 | /* make sure SQ entry isn't read before tail */ | |
2392 | return smp_load_acquire(&rings->sq.tail) - ctx->cached_sq_head; | |
2393 | } | |
2394 | ||
8ff069bf | 2395 | static unsigned int io_put_kbuf(struct io_kiocb *req, struct io_buffer *kbuf) |
e94f141b | 2396 | { |
8ff069bf | 2397 | unsigned int cflags; |
e94f141b | 2398 | |
bcda7baa JA |
2399 | cflags = kbuf->bid << IORING_CQE_BUFFER_SHIFT; |
2400 | cflags |= IORING_CQE_F_BUFFER; | |
0e1b6fe3 | 2401 | req->flags &= ~REQ_F_BUFFER_SELECTED; |
bcda7baa JA |
2402 | kfree(kbuf); |
2403 | return cflags; | |
e94f141b JA |
2404 | } |
2405 | ||
8ff069bf | 2406 | static inline unsigned int io_put_rw_kbuf(struct io_kiocb *req) |
bcda7baa | 2407 | { |
4d954c25 | 2408 | struct io_buffer *kbuf; |
bcda7baa | 2409 | |
4d954c25 | 2410 | kbuf = (struct io_buffer *) (unsigned long) req->rw.addr; |
8ff069bf PB |
2411 | return io_put_kbuf(req, kbuf); |
2412 | } | |
2413 | ||
4c6e277c JA |
2414 | static inline bool io_run_task_work(void) |
2415 | { | |
6200b0ae JA |
2416 | /* |
2417 | * Not safe to run on exiting task, and the task_work handling will | |
2418 | * not add work to such a task. | |
2419 | */ | |
2420 | if (unlikely(current->flags & PF_EXITING)) | |
2421 | return false; | |
4c6e277c JA |
2422 | if (current->task_works) { |
2423 | __set_current_state(TASK_RUNNING); | |
2424 | task_work_run(); | |
2425 | return true; | |
2426 | } | |
2427 | ||
2428 | return false; | |
bcda7baa JA |
2429 | } |
2430 | ||
bbde017a XW |
2431 | static void io_iopoll_queue(struct list_head *again) |
2432 | { | |
2433 | struct io_kiocb *req; | |
2434 | ||
2435 | do { | |
d21ffe7e PB |
2436 | req = list_first_entry(again, struct io_kiocb, inflight_entry); |
2437 | list_del(&req->inflight_entry); | |
81b68a5c | 2438 | __io_complete_rw(req, -EAGAIN, 0, NULL); |
bbde017a XW |
2439 | } while (!list_empty(again)); |
2440 | } | |
2441 | ||
def596e9 JA |
2442 | /* |
2443 | * Find and free completed poll iocbs | |
2444 | */ | |
2445 | static void io_iopoll_complete(struct io_ring_ctx *ctx, unsigned int *nr_events, | |
2446 | struct list_head *done) | |
2447 | { | |
8237e045 | 2448 | struct req_batch rb; |
def596e9 | 2449 | struct io_kiocb *req; |
bbde017a XW |
2450 | LIST_HEAD(again); |
2451 | ||
2452 | /* order with ->result store in io_complete_rw_iopoll() */ | |
2453 | smp_rmb(); | |
def596e9 | 2454 | |
5af1d13e | 2455 | io_init_req_batch(&rb); |
def596e9 | 2456 | while (!list_empty(done)) { |
bcda7baa JA |
2457 | int cflags = 0; |
2458 | ||
d21ffe7e | 2459 | req = list_first_entry(done, struct io_kiocb, inflight_entry); |
bbde017a | 2460 | if (READ_ONCE(req->result) == -EAGAIN) { |
56450c20 | 2461 | req->result = 0; |
bbde017a | 2462 | req->iopoll_completed = 0; |
d21ffe7e | 2463 | list_move_tail(&req->inflight_entry, &again); |
bbde017a XW |
2464 | continue; |
2465 | } | |
d21ffe7e | 2466 | list_del(&req->inflight_entry); |
def596e9 | 2467 | |
bcda7baa | 2468 | if (req->flags & REQ_F_BUFFER_SELECTED) |
8ff069bf | 2469 | cflags = io_put_rw_kbuf(req); |
bcda7baa JA |
2470 | |
2471 | __io_cqring_fill_event(req, req->result, cflags); | |
def596e9 JA |
2472 | (*nr_events)++; |
2473 | ||
c3524383 | 2474 | if (refcount_dec_and_test(&req->refs)) |
2d6500d4 | 2475 | io_req_free_batch(&rb, req); |
def596e9 | 2476 | } |
def596e9 | 2477 | |
09bb8394 | 2478 | io_commit_cqring(ctx); |
80c18e4a | 2479 | io_cqring_ev_posted_iopoll(ctx); |
2d6500d4 | 2480 | io_req_free_batch_finish(ctx, &rb); |
581f9810 | 2481 | |
bbde017a XW |
2482 | if (!list_empty(&again)) |
2483 | io_iopoll_queue(&again); | |
581f9810 BM |
2484 | } |
2485 | ||
def596e9 JA |
2486 | static int io_do_iopoll(struct io_ring_ctx *ctx, unsigned int *nr_events, |
2487 | long min) | |
2488 | { | |
2489 | struct io_kiocb *req, *tmp; | |
2490 | LIST_HEAD(done); | |
2491 | bool spin; | |
2492 | int ret; | |
2493 | ||
2494 | /* | |
2495 | * Only spin for completions if we don't have multiple devices hanging | |
2496 | * off our complete list, and we're under the requested amount. | |
2497 | */ | |
2498 | spin = !ctx->poll_multi_file && *nr_events < min; | |
2499 | ||
2500 | ret = 0; | |
d21ffe7e | 2501 | list_for_each_entry_safe(req, tmp, &ctx->iopoll_list, inflight_entry) { |
9adbd45d | 2502 | struct kiocb *kiocb = &req->rw.kiocb; |
def596e9 JA |
2503 | |
2504 | /* | |
581f9810 BM |
2505 | * Move completed and retryable entries to our local lists. |
2506 | * If we find a request that requires polling, break out | |
2507 | * and complete those lists first, if we have entries there. | |
def596e9 | 2508 | */ |
65a6543d | 2509 | if (READ_ONCE(req->iopoll_completed)) { |
d21ffe7e | 2510 | list_move_tail(&req->inflight_entry, &done); |
def596e9 JA |
2511 | continue; |
2512 | } | |
2513 | if (!list_empty(&done)) | |
2514 | break; | |
2515 | ||
2516 | ret = kiocb->ki_filp->f_op->iopoll(kiocb, spin); | |
2517 | if (ret < 0) | |
2518 | break; | |
2519 | ||
3aadc23e PB |
2520 | /* iopoll may have completed current req */ |
2521 | if (READ_ONCE(req->iopoll_completed)) | |
d21ffe7e | 2522 | list_move_tail(&req->inflight_entry, &done); |
3aadc23e | 2523 | |
def596e9 JA |
2524 | if (ret && spin) |
2525 | spin = false; | |
2526 | ret = 0; | |
2527 | } | |
2528 | ||
2529 | if (!list_empty(&done)) | |
2530 | io_iopoll_complete(ctx, nr_events, &done); | |
2531 | ||
2532 | return ret; | |
2533 | } | |
2534 | ||
2535 | /* | |
d195a66e | 2536 | * Poll for a minimum of 'min' events. Note that if min == 0 we consider that a |
def596e9 JA |
2537 | * non-spinning poll check - we'll still enter the driver poll loop, but only |
2538 | * as a non-spinning completion check. | |
2539 | */ | |
2540 | static int io_iopoll_getevents(struct io_ring_ctx *ctx, unsigned int *nr_events, | |
2541 | long min) | |
2542 | { | |
540e32a0 | 2543 | while (!list_empty(&ctx->iopoll_list) && !need_resched()) { |
def596e9 JA |
2544 | int ret; |
2545 | ||
2546 | ret = io_do_iopoll(ctx, nr_events, min); | |
2547 | if (ret < 0) | |
2548 | return ret; | |
eba0a4dd | 2549 | if (*nr_events >= min) |
def596e9 JA |
2550 | return 0; |
2551 | } | |
2552 | ||
2553 | return 1; | |
2554 | } | |
2555 | ||
2556 | /* | |
2557 | * We can't just wait for polled events to come to us, we have to actively | |
2558 | * find and complete them. | |
2559 | */ | |
b2edc0a7 | 2560 | static void io_iopoll_try_reap_events(struct io_ring_ctx *ctx) |
def596e9 JA |
2561 | { |
2562 | if (!(ctx->flags & IORING_SETUP_IOPOLL)) | |
2563 | return; | |
2564 | ||
2565 | mutex_lock(&ctx->uring_lock); | |
540e32a0 | 2566 | while (!list_empty(&ctx->iopoll_list)) { |
def596e9 JA |
2567 | unsigned int nr_events = 0; |
2568 | ||
b2edc0a7 | 2569 | io_do_iopoll(ctx, &nr_events, 0); |
08f5439f | 2570 | |
b2edc0a7 PB |
2571 | /* let it sleep and repeat later if can't complete a request */ |
2572 | if (nr_events == 0) | |
2573 | break; | |
08f5439f JA |
2574 | /* |
2575 | * Ensure we allow local-to-the-cpu processing to take place, | |
2576 | * in this case we need to ensure that we reap all events. | |
3fcee5a6 | 2577 | * Also let task_work, etc. to progress by releasing the mutex |
08f5439f | 2578 | */ |
3fcee5a6 PB |
2579 | if (need_resched()) { |
2580 | mutex_unlock(&ctx->uring_lock); | |
2581 | cond_resched(); | |
2582 | mutex_lock(&ctx->uring_lock); | |
2583 | } | |
def596e9 JA |
2584 | } |
2585 | mutex_unlock(&ctx->uring_lock); | |
2586 | } | |
2587 | ||
7668b92a | 2588 | static int io_iopoll_check(struct io_ring_ctx *ctx, long min) |
def596e9 | 2589 | { |
7668b92a | 2590 | unsigned int nr_events = 0; |
2b2ed975 | 2591 | int iters = 0, ret = 0; |
500f9fba | 2592 | |
c7849be9 XW |
2593 | /* |
2594 | * We disallow the app entering submit/complete with polling, but we | |
2595 | * still need to lock the ring to prevent racing with polled issue | |
2596 | * that got punted to a workqueue. | |
2597 | */ | |
2598 | mutex_lock(&ctx->uring_lock); | |
def596e9 | 2599 | do { |
a3a0e43f JA |
2600 | /* |
2601 | * Don't enter poll loop if we already have events pending. | |
2602 | * If we do, we can potentially be spinning for commands that | |
2603 | * already triggered a CQE (eg in error). | |
2604 | */ | |
6c503150 PB |
2605 | if (test_bit(0, &ctx->cq_check_overflow)) |
2606 | __io_cqring_overflow_flush(ctx, false, NULL, NULL); | |
2607 | if (io_cqring_events(ctx)) | |
a3a0e43f JA |
2608 | break; |
2609 | ||
500f9fba JA |
2610 | /* |
2611 | * If a submit got punted to a workqueue, we can have the | |
2612 | * application entering polling for a command before it gets | |
2613 | * issued. That app will hold the uring_lock for the duration | |
2614 | * of the poll right here, so we need to take a breather every | |
2615 | * now and then to ensure that the issue has a chance to add | |
2616 | * the poll to the issued list. Otherwise we can spin here | |
2617 | * forever, while the workqueue is stuck trying to acquire the | |
2618 | * very same mutex. | |
2619 | */ | |
2620 | if (!(++iters & 7)) { | |
2621 | mutex_unlock(&ctx->uring_lock); | |
4c6e277c | 2622 | io_run_task_work(); |
500f9fba JA |
2623 | mutex_lock(&ctx->uring_lock); |
2624 | } | |
2625 | ||
7668b92a | 2626 | ret = io_iopoll_getevents(ctx, &nr_events, min); |
def596e9 JA |
2627 | if (ret <= 0) |
2628 | break; | |
2629 | ret = 0; | |
7668b92a | 2630 | } while (min && !nr_events && !need_resched()); |
def596e9 | 2631 | |
500f9fba | 2632 | mutex_unlock(&ctx->uring_lock); |
def596e9 JA |
2633 | return ret; |
2634 | } | |
2635 | ||
491381ce | 2636 | static void kiocb_end_write(struct io_kiocb *req) |
2b188cc1 | 2637 | { |
491381ce JA |
2638 | /* |
2639 | * Tell lockdep we inherited freeze protection from submission | |
2640 | * thread. | |
2641 | */ | |
2642 | if (req->flags & REQ_F_ISREG) { | |
2643 | struct inode *inode = file_inode(req->file); | |
2b188cc1 | 2644 | |
491381ce | 2645 | __sb_writers_acquired(inode->i_sb, SB_FREEZE_WRITE); |
2b188cc1 | 2646 | } |
491381ce | 2647 | file_end_write(req->file); |
2b188cc1 JA |
2648 | } |
2649 | ||
a1d7c393 JA |
2650 | static void io_complete_rw_common(struct kiocb *kiocb, long res, |
2651 | struct io_comp_state *cs) | |
2b188cc1 | 2652 | { |
9adbd45d | 2653 | struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb); |
bcda7baa | 2654 | int cflags = 0; |
2b188cc1 | 2655 | |
491381ce JA |
2656 | if (kiocb->ki_flags & IOCB_WRITE) |
2657 | kiocb_end_write(req); | |
2b188cc1 | 2658 | |
4e88d6e7 JA |
2659 | if (res != req->result) |
2660 | req_set_fail_links(req); | |
bcda7baa | 2661 | if (req->flags & REQ_F_BUFFER_SELECTED) |
8ff069bf | 2662 | cflags = io_put_rw_kbuf(req); |
a1d7c393 | 2663 | __io_req_complete(req, res, cflags, cs); |
ba816ad6 JA |
2664 | } |
2665 | ||
b63534c4 JA |
2666 | #ifdef CONFIG_BLOCK |
2667 | static bool io_resubmit_prep(struct io_kiocb *req, int error) | |
2668 | { | |
2669 | struct iovec inline_vecs[UIO_FASTIOV], *iovec = inline_vecs; | |
2670 | ssize_t ret = -ECANCELED; | |
2671 | struct iov_iter iter; | |
2672 | int rw; | |
2673 | ||
2674 | if (error) { | |
2675 | ret = error; | |
2676 | goto end_req; | |
2677 | } | |
2678 | ||
2679 | switch (req->opcode) { | |
2680 | case IORING_OP_READV: | |
2681 | case IORING_OP_READ_FIXED: | |
2682 | case IORING_OP_READ: | |
2683 | rw = READ; | |
2684 | break; | |
2685 | case IORING_OP_WRITEV: | |
2686 | case IORING_OP_WRITE_FIXED: | |
2687 | case IORING_OP_WRITE: | |
2688 | rw = WRITE; | |
2689 | break; | |
2690 | default: | |
2691 | printk_once(KERN_WARNING "io_uring: bad opcode in resubmit %d\n", | |
2692 | req->opcode); | |
2693 | goto end_req; | |
2694 | } | |
2695 | ||
e8c2bc1f | 2696 | if (!req->async_data) { |
8f3d7496 JA |
2697 | ret = io_import_iovec(rw, req, &iovec, &iter, false); |
2698 | if (ret < 0) | |
2699 | goto end_req; | |
2700 | ret = io_setup_async_rw(req, iovec, inline_vecs, &iter, false); | |
2701 | if (!ret) | |
2702 | return true; | |
2703 | kfree(iovec); | |
2704 | } else { | |
b63534c4 | 2705 | return true; |
8f3d7496 | 2706 | } |
b63534c4 | 2707 | end_req: |
b63534c4 | 2708 | req_set_fail_links(req); |
b63534c4 JA |
2709 | return false; |
2710 | } | |
b63534c4 JA |
2711 | #endif |
2712 | ||
2713 | static bool io_rw_reissue(struct io_kiocb *req, long res) | |
2714 | { | |
2715 | #ifdef CONFIG_BLOCK | |
355afaeb | 2716 | umode_t mode = file_inode(req->file)->i_mode; |
b63534c4 JA |
2717 | int ret; |
2718 | ||
355afaeb JA |
2719 | if (!S_ISBLK(mode) && !S_ISREG(mode)) |
2720 | return false; | |
b63534c4 JA |
2721 | if ((res != -EAGAIN && res != -EOPNOTSUPP) || io_wq_current_is_worker()) |
2722 | return false; | |
2723 | ||
55e6ac1e PB |
2724 | lockdep_assert_held(&req->ctx->uring_lock); |
2725 | ||
28cea78a | 2726 | ret = io_sq_thread_acquire_mm_files(req->ctx, req); |
6d816e08 | 2727 | |
fdee946d JA |
2728 | if (io_resubmit_prep(req, ret)) { |
2729 | refcount_inc(&req->refs); | |
2730 | io_queue_async_work(req); | |
b63534c4 | 2731 | return true; |
fdee946d JA |
2732 | } |
2733 | ||
b63534c4 JA |
2734 | #endif |
2735 | return false; | |
2736 | } | |
2737 | ||
a1d7c393 JA |
2738 | static void __io_complete_rw(struct io_kiocb *req, long res, long res2, |
2739 | struct io_comp_state *cs) | |
2740 | { | |
2741 | if (!io_rw_reissue(req, res)) | |
2742 | io_complete_rw_common(&req->rw.kiocb, res, cs); | |
ba816ad6 JA |
2743 | } |
2744 | ||
2745 | static void io_complete_rw(struct kiocb *kiocb, long res, long res2) | |
2746 | { | |
9adbd45d | 2747 | struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb); |
ba816ad6 | 2748 | |
a1d7c393 | 2749 | __io_complete_rw(req, res, res2, NULL); |
2b188cc1 JA |
2750 | } |
2751 | ||
def596e9 JA |
2752 | static void io_complete_rw_iopoll(struct kiocb *kiocb, long res, long res2) |
2753 | { | |
9adbd45d | 2754 | struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb); |
def596e9 | 2755 | |
491381ce JA |
2756 | if (kiocb->ki_flags & IOCB_WRITE) |
2757 | kiocb_end_write(req); | |
def596e9 | 2758 | |
2d7d6792 | 2759 | if (res != -EAGAIN && res != req->result) |
4e88d6e7 | 2760 | req_set_fail_links(req); |
bbde017a XW |
2761 | |
2762 | WRITE_ONCE(req->result, res); | |
2763 | /* order with io_poll_complete() checking ->result */ | |
cd664b0e PB |
2764 | smp_wmb(); |
2765 | WRITE_ONCE(req->iopoll_completed, 1); | |
def596e9 JA |
2766 | } |
2767 | ||
2768 | /* | |
2769 | * After the iocb has been issued, it's safe to be found on the poll list. | |
2770 | * Adding the kiocb to the list AFTER submission ensures that we don't | |
2771 | * find it from a io_iopoll_getevents() thread before the issuer is done | |
2772 | * accessing the kiocb cookie. | |
2773 | */ | |
2e9dbe90 | 2774 | static void io_iopoll_req_issued(struct io_kiocb *req, bool in_async) |
def596e9 JA |
2775 | { |
2776 | struct io_ring_ctx *ctx = req->ctx; | |
2777 | ||
2778 | /* | |
2779 | * Track whether we have multiple files in our lists. This will impact | |
2780 | * how we do polling eventually, not spinning if we're on potentially | |
2781 | * different devices. | |
2782 | */ | |
540e32a0 | 2783 | if (list_empty(&ctx->iopoll_list)) { |
def596e9 JA |
2784 | ctx->poll_multi_file = false; |
2785 | } else if (!ctx->poll_multi_file) { | |
2786 | struct io_kiocb *list_req; | |
2787 | ||
540e32a0 | 2788 | list_req = list_first_entry(&ctx->iopoll_list, struct io_kiocb, |
d21ffe7e | 2789 | inflight_entry); |
9adbd45d | 2790 | if (list_req->file != req->file) |
def596e9 JA |
2791 | ctx->poll_multi_file = true; |
2792 | } | |
2793 | ||
2794 | /* | |
2795 | * For fast devices, IO may have already completed. If it has, add | |
2796 | * it to the front so we find it first. | |
2797 | */ | |
65a6543d | 2798 | if (READ_ONCE(req->iopoll_completed)) |
d21ffe7e | 2799 | list_add(&req->inflight_entry, &ctx->iopoll_list); |
def596e9 | 2800 | else |
d21ffe7e | 2801 | list_add_tail(&req->inflight_entry, &ctx->iopoll_list); |
bdcd3eab | 2802 | |
2e9dbe90 XW |
2803 | /* |
2804 | * If IORING_SETUP_SQPOLL is enabled, sqes are either handled in sq thread | |
2805 | * task context or in io worker task context. If current task context is | |
2806 | * sq thread, we don't need to check whether should wake up sq thread. | |
2807 | */ | |
2808 | if (in_async && (ctx->flags & IORING_SETUP_SQPOLL) && | |
534ca6d6 JA |
2809 | wq_has_sleeper(&ctx->sq_data->wait)) |
2810 | wake_up(&ctx->sq_data->wait); | |
def596e9 JA |
2811 | } |
2812 | ||
6e1271e6 | 2813 | static inline void __io_state_file_put(struct io_submit_state *state) |
9a56a232 | 2814 | { |
6e1271e6 PB |
2815 | fput_many(state->file, state->file_refs); |
2816 | state->file_refs = 0; | |
9f13c35b PB |
2817 | } |
2818 | ||
2819 | static inline void io_state_file_put(struct io_submit_state *state) | |
2820 | { | |
6e1271e6 | 2821 | if (state->file_refs) |
9f13c35b | 2822 | __io_state_file_put(state); |
9a56a232 JA |
2823 | } |
2824 | ||
2825 | /* | |
2826 | * Get as many references to a file as we have IOs left in this submission, | |
2827 | * assuming most submissions are for one file, or at least that each file | |
2828 | * has more than one submission. | |
2829 | */ | |
8da11c19 | 2830 | static struct file *__io_file_get(struct io_submit_state *state, int fd) |
9a56a232 JA |
2831 | { |
2832 | if (!state) | |
2833 | return fget(fd); | |
2834 | ||
6e1271e6 | 2835 | if (state->file_refs) { |
9a56a232 | 2836 | if (state->fd == fd) { |
6e1271e6 | 2837 | state->file_refs--; |
9a56a232 JA |
2838 | return state->file; |
2839 | } | |
9f13c35b | 2840 | __io_state_file_put(state); |
9a56a232 JA |
2841 | } |
2842 | state->file = fget_many(fd, state->ios_left); | |
6e1271e6 | 2843 | if (unlikely(!state->file)) |
9a56a232 JA |
2844 | return NULL; |
2845 | ||
2846 | state->fd = fd; | |
6e1271e6 | 2847 | state->file_refs = state->ios_left - 1; |
9a56a232 JA |
2848 | return state->file; |
2849 | } | |
2850 | ||
4503b767 JA |
2851 | static bool io_bdev_nowait(struct block_device *bdev) |
2852 | { | |
9ba0d0c8 | 2853 | return !bdev || blk_queue_nowait(bdev_get_queue(bdev)); |
4503b767 JA |
2854 | } |
2855 | ||
2b188cc1 JA |
2856 | /* |
2857 | * If we tracked the file through the SCM inflight mechanism, we could support | |
2858 | * any file. For now, just ensure that anything potentially problematic is done | |
2859 | * inline. | |
2860 | */ | |
af197f50 | 2861 | static bool io_file_supports_async(struct file *file, int rw) |
2b188cc1 JA |
2862 | { |
2863 | umode_t mode = file_inode(file)->i_mode; | |
2864 | ||
4503b767 | 2865 | if (S_ISBLK(mode)) { |
4e7b5671 CH |
2866 | if (IS_ENABLED(CONFIG_BLOCK) && |
2867 | io_bdev_nowait(I_BDEV(file->f_mapping->host))) | |
4503b767 JA |
2868 | return true; |
2869 | return false; | |
2870 | } | |
2871 | if (S_ISCHR(mode) || S_ISSOCK(mode)) | |
2b188cc1 | 2872 | return true; |
4503b767 | 2873 | if (S_ISREG(mode)) { |
4e7b5671 CH |
2874 | if (IS_ENABLED(CONFIG_BLOCK) && |
2875 | io_bdev_nowait(file->f_inode->i_sb->s_bdev) && | |
4503b767 JA |
2876 | file->f_op != &io_uring_fops) |
2877 | return true; | |
2878 | return false; | |
2879 | } | |
2b188cc1 | 2880 | |
c5b85625 JA |
2881 | /* any ->read/write should understand O_NONBLOCK */ |
2882 | if (file->f_flags & O_NONBLOCK) | |
2883 | return true; | |
2884 | ||
af197f50 JA |
2885 | if (!(file->f_mode & FMODE_NOWAIT)) |
2886 | return false; | |
2887 | ||
2888 | if (rw == READ) | |
2889 | return file->f_op->read_iter != NULL; | |
2890 | ||
2891 | return file->f_op->write_iter != NULL; | |
2b188cc1 JA |
2892 | } |
2893 | ||
a88fc400 | 2894 | static int io_prep_rw(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
2b188cc1 | 2895 | { |
def596e9 | 2896 | struct io_ring_ctx *ctx = req->ctx; |
9adbd45d | 2897 | struct kiocb *kiocb = &req->rw.kiocb; |
09bb8394 JA |
2898 | unsigned ioprio; |
2899 | int ret; | |
2b188cc1 | 2900 | |
491381ce JA |
2901 | if (S_ISREG(file_inode(req->file)->i_mode)) |
2902 | req->flags |= REQ_F_ISREG; | |
2903 | ||
2b188cc1 | 2904 | kiocb->ki_pos = READ_ONCE(sqe->off); |
ba04291e JA |
2905 | if (kiocb->ki_pos == -1 && !(req->file->f_mode & FMODE_STREAM)) { |
2906 | req->flags |= REQ_F_CUR_POS; | |
2907 | kiocb->ki_pos = req->file->f_pos; | |
2908 | } | |
2b188cc1 | 2909 | kiocb->ki_hint = ki_hint_validate(file_write_hint(kiocb->ki_filp)); |
3e577dcd PB |
2910 | kiocb->ki_flags = iocb_flags(kiocb->ki_filp); |
2911 | ret = kiocb_set_rw_flags(kiocb, READ_ONCE(sqe->rw_flags)); | |
2912 | if (unlikely(ret)) | |
2913 | return ret; | |
2b188cc1 JA |
2914 | |
2915 | ioprio = READ_ONCE(sqe->ioprio); | |
2916 | if (ioprio) { | |
2917 | ret = ioprio_check_cap(ioprio); | |
2918 | if (ret) | |
09bb8394 | 2919 | return ret; |
2b188cc1 JA |
2920 | |
2921 | kiocb->ki_ioprio = ioprio; | |
2922 | } else | |
2923 | kiocb->ki_ioprio = get_current_ioprio(); | |
2924 | ||
8449eeda | 2925 | /* don't allow async punt if RWF_NOWAIT was requested */ |
c5b85625 | 2926 | if (kiocb->ki_flags & IOCB_NOWAIT) |
8449eeda SB |
2927 | req->flags |= REQ_F_NOWAIT; |
2928 | ||
def596e9 | 2929 | if (ctx->flags & IORING_SETUP_IOPOLL) { |
def596e9 JA |
2930 | if (!(kiocb->ki_flags & IOCB_DIRECT) || |
2931 | !kiocb->ki_filp->f_op->iopoll) | |
09bb8394 | 2932 | return -EOPNOTSUPP; |
2b188cc1 | 2933 | |
def596e9 JA |
2934 | kiocb->ki_flags |= IOCB_HIPRI; |
2935 | kiocb->ki_complete = io_complete_rw_iopoll; | |
65a6543d | 2936 | req->iopoll_completed = 0; |
def596e9 | 2937 | } else { |
09bb8394 JA |
2938 | if (kiocb->ki_flags & IOCB_HIPRI) |
2939 | return -EINVAL; | |
def596e9 JA |
2940 | kiocb->ki_complete = io_complete_rw; |
2941 | } | |
9adbd45d | 2942 | |
3529d8c2 JA |
2943 | req->rw.addr = READ_ONCE(sqe->addr); |
2944 | req->rw.len = READ_ONCE(sqe->len); | |
4f4eeba8 | 2945 | req->buf_index = READ_ONCE(sqe->buf_index); |
2b188cc1 | 2946 | return 0; |
2b188cc1 JA |
2947 | } |
2948 | ||
2949 | static inline void io_rw_done(struct kiocb *kiocb, ssize_t ret) | |
2950 | { | |
2951 | switch (ret) { | |
2952 | case -EIOCBQUEUED: | |
2953 | break; | |
2954 | case -ERESTARTSYS: | |
2955 | case -ERESTARTNOINTR: | |
2956 | case -ERESTARTNOHAND: | |
2957 | case -ERESTART_RESTARTBLOCK: | |
2958 | /* | |
2959 | * We can't just restart the syscall, since previously | |
2960 | * submitted sqes may already be in progress. Just fail this | |
2961 | * IO with EINTR. | |
2962 | */ | |
2963 | ret = -EINTR; | |
df561f66 | 2964 | fallthrough; |
2b188cc1 JA |
2965 | default: |
2966 | kiocb->ki_complete(kiocb, ret, 0); | |
2967 | } | |
2968 | } | |
2969 | ||
a1d7c393 JA |
2970 | static void kiocb_done(struct kiocb *kiocb, ssize_t ret, |
2971 | struct io_comp_state *cs) | |
ba816ad6 | 2972 | { |
ba04291e | 2973 | struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb); |
e8c2bc1f | 2974 | struct io_async_rw *io = req->async_data; |
ba04291e | 2975 | |
227c0c96 | 2976 | /* add previously done IO, if any */ |
e8c2bc1f | 2977 | if (io && io->bytes_done > 0) { |
227c0c96 | 2978 | if (ret < 0) |
e8c2bc1f | 2979 | ret = io->bytes_done; |
227c0c96 | 2980 | else |
e8c2bc1f | 2981 | ret += io->bytes_done; |
227c0c96 JA |
2982 | } |
2983 | ||
ba04291e JA |
2984 | if (req->flags & REQ_F_CUR_POS) |
2985 | req->file->f_pos = kiocb->ki_pos; | |
bcaec089 | 2986 | if (ret >= 0 && kiocb->ki_complete == io_complete_rw) |
a1d7c393 | 2987 | __io_complete_rw(req, ret, 0, cs); |
ba816ad6 JA |
2988 | else |
2989 | io_rw_done(kiocb, ret); | |
2990 | } | |
2991 | ||
9adbd45d | 2992 | static ssize_t io_import_fixed(struct io_kiocb *req, int rw, |
7d009165 | 2993 | struct iov_iter *iter) |
edafccee | 2994 | { |
9adbd45d JA |
2995 | struct io_ring_ctx *ctx = req->ctx; |
2996 | size_t len = req->rw.len; | |
edafccee | 2997 | struct io_mapped_ubuf *imu; |
4be1c615 | 2998 | u16 index, buf_index = req->buf_index; |
edafccee JA |
2999 | size_t offset; |
3000 | u64 buf_addr; | |
3001 | ||
edafccee JA |
3002 | if (unlikely(buf_index >= ctx->nr_user_bufs)) |
3003 | return -EFAULT; | |
edafccee JA |
3004 | index = array_index_nospec(buf_index, ctx->nr_user_bufs); |
3005 | imu = &ctx->user_bufs[index]; | |
9adbd45d | 3006 | buf_addr = req->rw.addr; |
edafccee JA |
3007 | |
3008 | /* overflow */ | |
3009 | if (buf_addr + len < buf_addr) | |
3010 | return -EFAULT; | |
3011 | /* not inside the mapped region */ | |
3012 | if (buf_addr < imu->ubuf || buf_addr + len > imu->ubuf + imu->len) | |
3013 | return -EFAULT; | |
3014 | ||
3015 | /* | |
3016 | * May not be a start of buffer, set size appropriately | |
3017 | * and advance us to the beginning. | |
3018 | */ | |
3019 | offset = buf_addr - imu->ubuf; | |
3020 | iov_iter_bvec(iter, rw, imu->bvec, imu->nr_bvecs, offset + len); | |
bd11b3a3 JA |
3021 | |
3022 | if (offset) { | |
3023 | /* | |
3024 | * Don't use iov_iter_advance() here, as it's really slow for | |
3025 | * using the latter parts of a big fixed buffer - it iterates | |
3026 | * over each segment manually. We can cheat a bit here, because | |
3027 | * we know that: | |
3028 | * | |
3029 | * 1) it's a BVEC iter, we set it up | |
3030 | * 2) all bvecs are PAGE_SIZE in size, except potentially the | |
3031 | * first and last bvec | |
3032 | * | |
3033 | * So just find our index, and adjust the iterator afterwards. | |
3034 | * If the offset is within the first bvec (or the whole first | |
3035 | * bvec, just use iov_iter_advance(). This makes it easier | |
3036 | * since we can just skip the first segment, which may not | |
3037 | * be PAGE_SIZE aligned. | |
3038 | */ | |
3039 | const struct bio_vec *bvec = imu->bvec; | |
3040 | ||
3041 | if (offset <= bvec->bv_len) { | |
3042 | iov_iter_advance(iter, offset); | |
3043 | } else { | |
3044 | unsigned long seg_skip; | |
3045 | ||
3046 | /* skip first vec */ | |
3047 | offset -= bvec->bv_len; | |
3048 | seg_skip = 1 + (offset >> PAGE_SHIFT); | |
3049 | ||
3050 | iter->bvec = bvec + seg_skip; | |
3051 | iter->nr_segs -= seg_skip; | |
99c79f66 | 3052 | iter->count -= bvec->bv_len + offset; |
bd11b3a3 | 3053 | iter->iov_offset = offset & ~PAGE_MASK; |
bd11b3a3 JA |
3054 | } |
3055 | } | |
3056 | ||
5e559561 | 3057 | return len; |
edafccee JA |
3058 | } |
3059 | ||
bcda7baa JA |
3060 | static void io_ring_submit_unlock(struct io_ring_ctx *ctx, bool needs_lock) |
3061 | { | |
3062 | if (needs_lock) | |
3063 | mutex_unlock(&ctx->uring_lock); | |
3064 | } | |
3065 | ||
3066 | static void io_ring_submit_lock(struct io_ring_ctx *ctx, bool needs_lock) | |
3067 | { | |
3068 | /* | |
3069 | * "Normal" inline submissions always hold the uring_lock, since we | |
3070 | * grab it from the system call. Same is true for the SQPOLL offload. | |
3071 | * The only exception is when we've detached the request and issue it | |
3072 | * from an async worker thread, grab the lock for that case. | |
3073 | */ | |
3074 | if (needs_lock) | |
3075 | mutex_lock(&ctx->uring_lock); | |
3076 | } | |
3077 | ||
3078 | static struct io_buffer *io_buffer_select(struct io_kiocb *req, size_t *len, | |
3079 | int bgid, struct io_buffer *kbuf, | |
3080 | bool needs_lock) | |
3081 | { | |
3082 | struct io_buffer *head; | |
3083 | ||
3084 | if (req->flags & REQ_F_BUFFER_SELECTED) | |
3085 | return kbuf; | |
3086 | ||
3087 | io_ring_submit_lock(req->ctx, needs_lock); | |
3088 | ||
3089 | lockdep_assert_held(&req->ctx->uring_lock); | |
3090 | ||
3091 | head = idr_find(&req->ctx->io_buffer_idr, bgid); | |
3092 | if (head) { | |
3093 | if (!list_empty(&head->list)) { | |
3094 | kbuf = list_last_entry(&head->list, struct io_buffer, | |
3095 | list); | |
3096 | list_del(&kbuf->list); | |
3097 | } else { | |
3098 | kbuf = head; | |
3099 | idr_remove(&req->ctx->io_buffer_idr, bgid); | |
3100 | } | |
3101 | if (*len > kbuf->len) | |
3102 | *len = kbuf->len; | |
3103 | } else { | |
3104 | kbuf = ERR_PTR(-ENOBUFS); | |
3105 | } | |
3106 | ||
3107 | io_ring_submit_unlock(req->ctx, needs_lock); | |
3108 | ||
3109 | return kbuf; | |
3110 | } | |
3111 | ||
4d954c25 JA |
3112 | static void __user *io_rw_buffer_select(struct io_kiocb *req, size_t *len, |
3113 | bool needs_lock) | |
3114 | { | |
3115 | struct io_buffer *kbuf; | |
4f4eeba8 | 3116 | u16 bgid; |
4d954c25 JA |
3117 | |
3118 | kbuf = (struct io_buffer *) (unsigned long) req->rw.addr; | |
4f4eeba8 | 3119 | bgid = req->buf_index; |
4d954c25 JA |
3120 | kbuf = io_buffer_select(req, len, bgid, kbuf, needs_lock); |
3121 | if (IS_ERR(kbuf)) | |
3122 | return kbuf; | |
3123 | req->rw.addr = (u64) (unsigned long) kbuf; | |
3124 | req->flags |= REQ_F_BUFFER_SELECTED; | |
3125 | return u64_to_user_ptr(kbuf->addr); | |
3126 | } | |
3127 | ||
3128 | #ifdef CONFIG_COMPAT | |
3129 | static ssize_t io_compat_import(struct io_kiocb *req, struct iovec *iov, | |
3130 | bool needs_lock) | |
3131 | { | |
3132 | struct compat_iovec __user *uiov; | |
3133 | compat_ssize_t clen; | |
3134 | void __user *buf; | |
3135 | ssize_t len; | |
3136 | ||
3137 | uiov = u64_to_user_ptr(req->rw.addr); | |
3138 | if (!access_ok(uiov, sizeof(*uiov))) | |
3139 | return -EFAULT; | |
3140 | if (__get_user(clen, &uiov->iov_len)) | |
3141 | return -EFAULT; | |
3142 | if (clen < 0) | |
3143 | return -EINVAL; | |
3144 | ||
3145 | len = clen; | |
3146 | buf = io_rw_buffer_select(req, &len, needs_lock); | |
3147 | if (IS_ERR(buf)) | |
3148 | return PTR_ERR(buf); | |
3149 | iov[0].iov_base = buf; | |
3150 | iov[0].iov_len = (compat_size_t) len; | |
3151 | return 0; | |
3152 | } | |
3153 | #endif | |
3154 | ||
3155 | static ssize_t __io_iov_buffer_select(struct io_kiocb *req, struct iovec *iov, | |
3156 | bool needs_lock) | |
3157 | { | |
3158 | struct iovec __user *uiov = u64_to_user_ptr(req->rw.addr); | |
3159 | void __user *buf; | |
3160 | ssize_t len; | |
3161 | ||
3162 | if (copy_from_user(iov, uiov, sizeof(*uiov))) | |
3163 | return -EFAULT; | |
3164 | ||
3165 | len = iov[0].iov_len; | |
3166 | if (len < 0) | |
3167 | return -EINVAL; | |
3168 | buf = io_rw_buffer_select(req, &len, needs_lock); | |
3169 | if (IS_ERR(buf)) | |
3170 | return PTR_ERR(buf); | |
3171 | iov[0].iov_base = buf; | |
3172 | iov[0].iov_len = len; | |
3173 | return 0; | |
3174 | } | |
3175 | ||
3176 | static ssize_t io_iov_buffer_select(struct io_kiocb *req, struct iovec *iov, | |
3177 | bool needs_lock) | |
3178 | { | |
dddb3e26 JA |
3179 | if (req->flags & REQ_F_BUFFER_SELECTED) { |
3180 | struct io_buffer *kbuf; | |
3181 | ||
3182 | kbuf = (struct io_buffer *) (unsigned long) req->rw.addr; | |
3183 | iov[0].iov_base = u64_to_user_ptr(kbuf->addr); | |
3184 | iov[0].iov_len = kbuf->len; | |
4d954c25 | 3185 | return 0; |
dddb3e26 | 3186 | } |
dd201662 | 3187 | if (req->rw.len != 1) |
4d954c25 JA |
3188 | return -EINVAL; |
3189 | ||
3190 | #ifdef CONFIG_COMPAT | |
3191 | if (req->ctx->compat) | |
3192 | return io_compat_import(req, iov, needs_lock); | |
3193 | #endif | |
3194 | ||
3195 | return __io_iov_buffer_select(req, iov, needs_lock); | |
3196 | } | |
3197 | ||
2846c481 | 3198 | static ssize_t io_import_iovec(int rw, struct io_kiocb *req, |
8452fd0c JA |
3199 | struct iovec **iovec, struct iov_iter *iter, |
3200 | bool needs_lock) | |
2b188cc1 | 3201 | { |
9adbd45d JA |
3202 | void __user *buf = u64_to_user_ptr(req->rw.addr); |
3203 | size_t sqe_len = req->rw.len; | |
4d954c25 | 3204 | ssize_t ret; |
edafccee JA |
3205 | u8 opcode; |
3206 | ||
d625c6ee | 3207 | opcode = req->opcode; |
7d009165 | 3208 | if (opcode == IORING_OP_READ_FIXED || opcode == IORING_OP_WRITE_FIXED) { |
edafccee | 3209 | *iovec = NULL; |
9adbd45d | 3210 | return io_import_fixed(req, rw, iter); |
edafccee | 3211 | } |
2b188cc1 | 3212 | |
bcda7baa | 3213 | /* buffer index only valid with fixed read/write, or buffer select */ |
4f4eeba8 | 3214 | if (req->buf_index && !(req->flags & REQ_F_BUFFER_SELECT)) |
9adbd45d JA |
3215 | return -EINVAL; |
3216 | ||
3a6820f2 | 3217 | if (opcode == IORING_OP_READ || opcode == IORING_OP_WRITE) { |
bcda7baa | 3218 | if (req->flags & REQ_F_BUFFER_SELECT) { |
4d954c25 | 3219 | buf = io_rw_buffer_select(req, &sqe_len, needs_lock); |
867a23ea | 3220 | if (IS_ERR(buf)) |
4d954c25 | 3221 | return PTR_ERR(buf); |
3f9d6441 | 3222 | req->rw.len = sqe_len; |
bcda7baa JA |
3223 | } |
3224 | ||
3a6820f2 JA |
3225 | ret = import_single_range(rw, buf, sqe_len, *iovec, iter); |
3226 | *iovec = NULL; | |
10fc72e4 | 3227 | return ret; |
3a6820f2 JA |
3228 | } |
3229 | ||
4d954c25 JA |
3230 | if (req->flags & REQ_F_BUFFER_SELECT) { |
3231 | ret = io_iov_buffer_select(req, *iovec, needs_lock); | |
3f9d6441 JA |
3232 | if (!ret) { |
3233 | ret = (*iovec)->iov_len; | |
3234 | iov_iter_init(iter, rw, *iovec, 1, ret); | |
3235 | } | |
4d954c25 JA |
3236 | *iovec = NULL; |
3237 | return ret; | |
3238 | } | |
3239 | ||
89cd35c5 CH |
3240 | return __import_iovec(rw, buf, sqe_len, UIO_FASTIOV, iovec, iter, |
3241 | req->ctx->compat); | |
2b188cc1 JA |
3242 | } |
3243 | ||
0fef9483 JA |
3244 | static inline loff_t *io_kiocb_ppos(struct kiocb *kiocb) |
3245 | { | |
5b09e37e | 3246 | return (kiocb->ki_filp->f_mode & FMODE_STREAM) ? NULL : &kiocb->ki_pos; |
0fef9483 JA |
3247 | } |
3248 | ||
31b51510 | 3249 | /* |
32960613 JA |
3250 | * For files that don't have ->read_iter() and ->write_iter(), handle them |
3251 | * by looping over ->read() or ->write() manually. | |
31b51510 | 3252 | */ |
4017eb91 | 3253 | static ssize_t loop_rw_iter(int rw, struct io_kiocb *req, struct iov_iter *iter) |
32960613 | 3254 | { |
4017eb91 JA |
3255 | struct kiocb *kiocb = &req->rw.kiocb; |
3256 | struct file *file = req->file; | |
32960613 JA |
3257 | ssize_t ret = 0; |
3258 | ||
3259 | /* | |
3260 | * Don't support polled IO through this interface, and we can't | |
3261 | * support non-blocking either. For the latter, this just causes | |
3262 | * the kiocb to be handled from an async context. | |
3263 | */ | |
3264 | if (kiocb->ki_flags & IOCB_HIPRI) | |
3265 | return -EOPNOTSUPP; | |
3266 | if (kiocb->ki_flags & IOCB_NOWAIT) | |
3267 | return -EAGAIN; | |
3268 | ||
3269 | while (iov_iter_count(iter)) { | |
311ae9e1 | 3270 | struct iovec iovec; |
32960613 JA |
3271 | ssize_t nr; |
3272 | ||
311ae9e1 PB |
3273 | if (!iov_iter_is_bvec(iter)) { |
3274 | iovec = iov_iter_iovec(iter); | |
3275 | } else { | |
4017eb91 JA |
3276 | iovec.iov_base = u64_to_user_ptr(req->rw.addr); |
3277 | iovec.iov_len = req->rw.len; | |
311ae9e1 PB |
3278 | } |
3279 | ||
32960613 JA |
3280 | if (rw == READ) { |
3281 | nr = file->f_op->read(file, iovec.iov_base, | |
0fef9483 | 3282 | iovec.iov_len, io_kiocb_ppos(kiocb)); |
32960613 JA |
3283 | } else { |
3284 | nr = file->f_op->write(file, iovec.iov_base, | |
0fef9483 | 3285 | iovec.iov_len, io_kiocb_ppos(kiocb)); |
32960613 JA |
3286 | } |
3287 | ||
3288 | if (nr < 0) { | |
3289 | if (!ret) | |
3290 | ret = nr; | |
3291 | break; | |
3292 | } | |
3293 | ret += nr; | |
3294 | if (nr != iovec.iov_len) | |
3295 | break; | |
4017eb91 JA |
3296 | req->rw.len -= nr; |
3297 | req->rw.addr += nr; | |
32960613 JA |
3298 | iov_iter_advance(iter, nr); |
3299 | } | |
3300 | ||
3301 | return ret; | |
3302 | } | |
3303 | ||
ff6165b2 JA |
3304 | static void io_req_map_rw(struct io_kiocb *req, const struct iovec *iovec, |
3305 | const struct iovec *fast_iov, struct iov_iter *iter) | |
f67676d1 | 3306 | { |
e8c2bc1f | 3307 | struct io_async_rw *rw = req->async_data; |
b64e3444 | 3308 | |
ff6165b2 | 3309 | memcpy(&rw->iter, iter, sizeof(*iter)); |
afb87658 | 3310 | rw->free_iovec = iovec; |
227c0c96 | 3311 | rw->bytes_done = 0; |
ff6165b2 | 3312 | /* can only be fixed buffers, no need to do anything */ |
9c3a205c | 3313 | if (iov_iter_is_bvec(iter)) |
ff6165b2 | 3314 | return; |
b64e3444 | 3315 | if (!iovec) { |
ff6165b2 JA |
3316 | unsigned iov_off = 0; |
3317 | ||
3318 | rw->iter.iov = rw->fast_iov; | |
3319 | if (iter->iov != fast_iov) { | |
3320 | iov_off = iter->iov - fast_iov; | |
3321 | rw->iter.iov += iov_off; | |
3322 | } | |
3323 | if (rw->fast_iov != fast_iov) | |
3324 | memcpy(rw->fast_iov + iov_off, fast_iov + iov_off, | |
45097dae | 3325 | sizeof(struct iovec) * iter->nr_segs); |
99bc4c38 PB |
3326 | } else { |
3327 | req->flags |= REQ_F_NEED_CLEANUP; | |
f67676d1 JA |
3328 | } |
3329 | } | |
3330 | ||
e8c2bc1f | 3331 | static inline int __io_alloc_async_data(struct io_kiocb *req) |
3d9932a8 | 3332 | { |
e8c2bc1f JA |
3333 | WARN_ON_ONCE(!io_op_defs[req->opcode].async_size); |
3334 | req->async_data = kmalloc(io_op_defs[req->opcode].async_size, GFP_KERNEL); | |
3335 | return req->async_data == NULL; | |
3d9932a8 XW |
3336 | } |
3337 | ||
e8c2bc1f | 3338 | static int io_alloc_async_data(struct io_kiocb *req) |
f67676d1 | 3339 | { |
e8c2bc1f | 3340 | if (!io_op_defs[req->opcode].needs_async_data) |
d3656344 | 3341 | return 0; |
3d9932a8 | 3342 | |
e8c2bc1f | 3343 | return __io_alloc_async_data(req); |
b7bb4f7d JA |
3344 | } |
3345 | ||
ff6165b2 JA |
3346 | static int io_setup_async_rw(struct io_kiocb *req, const struct iovec *iovec, |
3347 | const struct iovec *fast_iov, | |
227c0c96 | 3348 | struct iov_iter *iter, bool force) |
b7bb4f7d | 3349 | { |
e8c2bc1f | 3350 | if (!force && !io_op_defs[req->opcode].needs_async_data) |
74566df3 | 3351 | return 0; |
e8c2bc1f JA |
3352 | if (!req->async_data) { |
3353 | if (__io_alloc_async_data(req)) | |
5d204bcf | 3354 | return -ENOMEM; |
b7bb4f7d | 3355 | |
ff6165b2 | 3356 | io_req_map_rw(req, iovec, fast_iov, iter); |
5d204bcf | 3357 | } |
b7bb4f7d | 3358 | return 0; |
f67676d1 JA |
3359 | } |
3360 | ||
73debe68 | 3361 | static inline int io_rw_prep_async(struct io_kiocb *req, int rw) |
c3e330a4 | 3362 | { |
e8c2bc1f | 3363 | struct io_async_rw *iorw = req->async_data; |
f4bff104 | 3364 | struct iovec *iov = iorw->fast_iov; |
c3e330a4 PB |
3365 | ssize_t ret; |
3366 | ||
2846c481 | 3367 | ret = io_import_iovec(rw, req, &iov, &iorw->iter, false); |
c3e330a4 PB |
3368 | if (unlikely(ret < 0)) |
3369 | return ret; | |
3370 | ||
ab0b196c PB |
3371 | iorw->bytes_done = 0; |
3372 | iorw->free_iovec = iov; | |
3373 | if (iov) | |
3374 | req->flags |= REQ_F_NEED_CLEANUP; | |
c3e330a4 PB |
3375 | return 0; |
3376 | } | |
3377 | ||
73debe68 | 3378 | static int io_read_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
f67676d1 JA |
3379 | { |
3380 | ssize_t ret; | |
3381 | ||
a88fc400 | 3382 | ret = io_prep_rw(req, sqe); |
3529d8c2 JA |
3383 | if (ret) |
3384 | return ret; | |
f67676d1 | 3385 | |
3529d8c2 JA |
3386 | if (unlikely(!(req->file->f_mode & FMODE_READ))) |
3387 | return -EBADF; | |
f67676d1 | 3388 | |
5f798bea | 3389 | /* either don't need iovec imported or already have it */ |
2d199895 | 3390 | if (!req->async_data) |
3529d8c2 | 3391 | return 0; |
73debe68 | 3392 | return io_rw_prep_async(req, READ); |
f67676d1 JA |
3393 | } |
3394 | ||
c1dd91d1 JA |
3395 | /* |
3396 | * This is our waitqueue callback handler, registered through lock_page_async() | |
3397 | * when we initially tried to do the IO with the iocb armed our waitqueue. | |
3398 | * This gets called when the page is unlocked, and we generally expect that to | |
3399 | * happen when the page IO is completed and the page is now uptodate. This will | |
3400 | * queue a task_work based retry of the operation, attempting to copy the data | |
3401 | * again. If the latter fails because the page was NOT uptodate, then we will | |
3402 | * do a thread based blocking retry of the operation. That's the unexpected | |
3403 | * slow path. | |
3404 | */ | |
bcf5a063 JA |
3405 | static int io_async_buf_func(struct wait_queue_entry *wait, unsigned mode, |
3406 | int sync, void *arg) | |
3407 | { | |
3408 | struct wait_page_queue *wpq; | |
3409 | struct io_kiocb *req = wait->private; | |
bcf5a063 | 3410 | struct wait_page_key *key = arg; |
bcf5a063 JA |
3411 | int ret; |
3412 | ||
3413 | wpq = container_of(wait, struct wait_page_queue, wait); | |
3414 | ||
cdc8fcb4 LT |
3415 | if (!wake_page_match(wpq, key)) |
3416 | return 0; | |
3417 | ||
c8d317aa | 3418 | req->rw.kiocb.ki_flags &= ~IOCB_WAITQ; |
bcf5a063 JA |
3419 | list_del_init(&wait->entry); |
3420 | ||
e7375122 | 3421 | init_task_work(&req->task_work, io_req_task_submit); |
6d816e08 JA |
3422 | percpu_ref_get(&req->ctx->refs); |
3423 | ||
bcf5a063 JA |
3424 | /* submit ref gets dropped, acquire a new one */ |
3425 | refcount_inc(&req->refs); | |
355fb9e2 | 3426 | ret = io_req_task_work_add(req); |
bcf5a063 | 3427 | if (unlikely(ret)) { |
c2c4c83c JA |
3428 | struct task_struct *tsk; |
3429 | ||
bcf5a063 | 3430 | /* queue just for cancelation */ |
e7375122 | 3431 | init_task_work(&req->task_work, io_req_task_cancel); |
bcf5a063 | 3432 | tsk = io_wq_get_task(req->ctx->io_wq); |
91989c70 | 3433 | task_work_add(tsk, &req->task_work, TWA_NONE); |
c2c4c83c | 3434 | wake_up_process(tsk); |
bcf5a063 | 3435 | } |
bcf5a063 JA |
3436 | return 1; |
3437 | } | |
3438 | ||
c1dd91d1 JA |
3439 | /* |
3440 | * This controls whether a given IO request should be armed for async page | |
3441 | * based retry. If we return false here, the request is handed to the async | |
3442 | * worker threads for retry. If we're doing buffered reads on a regular file, | |
3443 | * we prepare a private wait_page_queue entry and retry the operation. This | |
3444 | * will either succeed because the page is now uptodate and unlocked, or it | |
3445 | * will register a callback when the page is unlocked at IO completion. Through | |
3446 | * that callback, io_uring uses task_work to setup a retry of the operation. | |
3447 | * That retry will attempt the buffered read again. The retry will generally | |
3448 | * succeed, or in rare cases where it fails, we then fall back to using the | |
3449 | * async worker threads for a blocking retry. | |
3450 | */ | |
227c0c96 | 3451 | static bool io_rw_should_retry(struct io_kiocb *req) |
f67676d1 | 3452 | { |
e8c2bc1f JA |
3453 | struct io_async_rw *rw = req->async_data; |
3454 | struct wait_page_queue *wait = &rw->wpq; | |
bcf5a063 | 3455 | struct kiocb *kiocb = &req->rw.kiocb; |
f67676d1 | 3456 | |
bcf5a063 JA |
3457 | /* never retry for NOWAIT, we just complete with -EAGAIN */ |
3458 | if (req->flags & REQ_F_NOWAIT) | |
3459 | return false; | |
f67676d1 | 3460 | |
227c0c96 | 3461 | /* Only for buffered IO */ |
3b2a4439 | 3462 | if (kiocb->ki_flags & (IOCB_DIRECT | IOCB_HIPRI)) |
bcf5a063 | 3463 | return false; |
3b2a4439 | 3464 | |
bcf5a063 JA |
3465 | /* |
3466 | * just use poll if we can, and don't attempt if the fs doesn't | |
3467 | * support callback based unlocks | |
3468 | */ | |
3469 | if (file_can_poll(req->file) || !(req->file->f_mode & FMODE_BUF_RASYNC)) | |
3470 | return false; | |
f67676d1 | 3471 | |
3b2a4439 JA |
3472 | wait->wait.func = io_async_buf_func; |
3473 | wait->wait.private = req; | |
3474 | wait->wait.flags = 0; | |
3475 | INIT_LIST_HEAD(&wait->wait.entry); | |
3476 | kiocb->ki_flags |= IOCB_WAITQ; | |
c8d317aa | 3477 | kiocb->ki_flags &= ~IOCB_NOWAIT; |
3b2a4439 | 3478 | kiocb->ki_waitq = wait; |
3b2a4439 | 3479 | return true; |
bcf5a063 JA |
3480 | } |
3481 | ||
3482 | static int io_iter_do_read(struct io_kiocb *req, struct iov_iter *iter) | |
3483 | { | |
3484 | if (req->file->f_op->read_iter) | |
3485 | return call_read_iter(req->file, &req->rw.kiocb, iter); | |
2dd2111d | 3486 | else if (req->file->f_op->read) |
4017eb91 | 3487 | return loop_rw_iter(READ, req, iter); |
2dd2111d GH |
3488 | else |
3489 | return -EINVAL; | |
f67676d1 JA |
3490 | } |
3491 | ||
a1d7c393 JA |
3492 | static int io_read(struct io_kiocb *req, bool force_nonblock, |
3493 | struct io_comp_state *cs) | |
2b188cc1 JA |
3494 | { |
3495 | struct iovec inline_vecs[UIO_FASTIOV], *iovec = inline_vecs; | |
9adbd45d | 3496 | struct kiocb *kiocb = &req->rw.kiocb; |
ff6165b2 | 3497 | struct iov_iter __iter, *iter = &__iter; |
e8c2bc1f | 3498 | struct io_async_rw *rw = req->async_data; |
227c0c96 | 3499 | ssize_t io_size, ret, ret2; |
f5cac8b1 | 3500 | bool no_async; |
ff6165b2 | 3501 | |
2846c481 | 3502 | if (rw) { |
e8c2bc1f | 3503 | iter = &rw->iter; |
2846c481 PB |
3504 | iovec = NULL; |
3505 | } else { | |
3506 | ret = io_import_iovec(READ, req, &iovec, iter, !force_nonblock); | |
3507 | if (ret < 0) | |
3508 | return ret; | |
3509 | } | |
632546c4 | 3510 | io_size = iov_iter_count(iter); |
fa15bafb | 3511 | req->result = io_size; |
227c0c96 | 3512 | ret = 0; |
2b188cc1 | 3513 | |
fd6c2e4c JA |
3514 | /* Ensure we clear previously set non-block flag */ |
3515 | if (!force_nonblock) | |
29de5f6a | 3516 | kiocb->ki_flags &= ~IOCB_NOWAIT; |
a88fc400 PB |
3517 | else |
3518 | kiocb->ki_flags |= IOCB_NOWAIT; | |
3519 | ||
fd6c2e4c | 3520 | |
24c74678 | 3521 | /* If the file doesn't support async, just async punt */ |
f5cac8b1 JA |
3522 | no_async = force_nonblock && !io_file_supports_async(req->file, READ); |
3523 | if (no_async) | |
f67676d1 | 3524 | goto copy_iov; |
9e645e11 | 3525 | |
632546c4 | 3526 | ret = rw_verify_area(READ, req->file, io_kiocb_ppos(kiocb), io_size); |
fa15bafb PB |
3527 | if (unlikely(ret)) |
3528 | goto out_free; | |
2b188cc1 | 3529 | |
227c0c96 | 3530 | ret = io_iter_do_read(req, iter); |
32960613 | 3531 | |
227c0c96 JA |
3532 | if (!ret) { |
3533 | goto done; | |
3534 | } else if (ret == -EIOCBQUEUED) { | |
3535 | ret = 0; | |
3536 | goto out_free; | |
3537 | } else if (ret == -EAGAIN) { | |
eefdf30f JA |
3538 | /* IOPOLL retry should happen for io-wq threads */ |
3539 | if (!force_nonblock && !(req->ctx->flags & IORING_SETUP_IOPOLL)) | |
f91daf56 | 3540 | goto done; |
355afaeb JA |
3541 | /* no retry on NONBLOCK marked file */ |
3542 | if (req->file->f_flags & O_NONBLOCK) | |
3543 | goto done; | |
84216315 | 3544 | /* some cases will consume bytes even on error returns */ |
632546c4 | 3545 | iov_iter_revert(iter, io_size - iov_iter_count(iter)); |
f38c7e3a JA |
3546 | ret = 0; |
3547 | goto copy_iov; | |
227c0c96 | 3548 | } else if (ret < 0) { |
00d23d51 JA |
3549 | /* make sure -ERESTARTSYS -> -EINTR is done */ |
3550 | goto done; | |
227c0c96 JA |
3551 | } |
3552 | ||
3553 | /* read it all, or we did blocking attempt. no retry. */ | |
f91daf56 JA |
3554 | if (!iov_iter_count(iter) || !force_nonblock || |
3555 | (req->file->f_flags & O_NONBLOCK)) | |
227c0c96 JA |
3556 | goto done; |
3557 | ||
3558 | io_size -= ret; | |
3559 | copy_iov: | |
3560 | ret2 = io_setup_async_rw(req, iovec, inline_vecs, iter, true); | |
3561 | if (ret2) { | |
3562 | ret = ret2; | |
3563 | goto out_free; | |
3564 | } | |
f5cac8b1 JA |
3565 | if (no_async) |
3566 | return -EAGAIN; | |
e8c2bc1f | 3567 | rw = req->async_data; |
227c0c96 JA |
3568 | /* it's copied and will be cleaned with ->io */ |
3569 | iovec = NULL; | |
3570 | /* now use our persistent iterator, if we aren't already */ | |
e8c2bc1f | 3571 | iter = &rw->iter; |
227c0c96 | 3572 | retry: |
e8c2bc1f | 3573 | rw->bytes_done += ret; |
227c0c96 JA |
3574 | /* if we can retry, do so with the callbacks armed */ |
3575 | if (!io_rw_should_retry(req)) { | |
fa15bafb PB |
3576 | kiocb->ki_flags &= ~IOCB_WAITQ; |
3577 | return -EAGAIN; | |
2b188cc1 | 3578 | } |
227c0c96 JA |
3579 | |
3580 | /* | |
3581 | * Now retry read with the IOCB_WAITQ parts set in the iocb. If we | |
3582 | * get -EIOCBQUEUED, then we'll get a notification when the desired | |
3583 | * page gets unlocked. We can also get a partial read here, and if we | |
3584 | * do, then just retry at the new offset. | |
3585 | */ | |
3586 | ret = io_iter_do_read(req, iter); | |
3587 | if (ret == -EIOCBQUEUED) { | |
3588 | ret = 0; | |
3589 | goto out_free; | |
3590 | } else if (ret > 0 && ret < io_size) { | |
3591 | /* we got some bytes, but not all. retry. */ | |
3592 | goto retry; | |
3593 | } | |
3594 | done: | |
3595 | kiocb_done(kiocb, ret, cs); | |
3596 | ret = 0; | |
f67676d1 | 3597 | out_free: |
f261c168 | 3598 | /* it's reportedly faster than delegating the null check to kfree() */ |
252917c3 | 3599 | if (iovec) |
6f2cc166 | 3600 | kfree(iovec); |
2b188cc1 JA |
3601 | return ret; |
3602 | } | |
3603 | ||
73debe68 | 3604 | static int io_write_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
f67676d1 JA |
3605 | { |
3606 | ssize_t ret; | |
3607 | ||
a88fc400 | 3608 | ret = io_prep_rw(req, sqe); |
3529d8c2 JA |
3609 | if (ret) |
3610 | return ret; | |
f67676d1 | 3611 | |
3529d8c2 JA |
3612 | if (unlikely(!(req->file->f_mode & FMODE_WRITE))) |
3613 | return -EBADF; | |
f67676d1 | 3614 | |
5f798bea | 3615 | /* either don't need iovec imported or already have it */ |
2d199895 | 3616 | if (!req->async_data) |
3529d8c2 | 3617 | return 0; |
73debe68 | 3618 | return io_rw_prep_async(req, WRITE); |
f67676d1 JA |
3619 | } |
3620 | ||
a1d7c393 JA |
3621 | static int io_write(struct io_kiocb *req, bool force_nonblock, |
3622 | struct io_comp_state *cs) | |
2b188cc1 JA |
3623 | { |
3624 | struct iovec inline_vecs[UIO_FASTIOV], *iovec = inline_vecs; | |
9adbd45d | 3625 | struct kiocb *kiocb = &req->rw.kiocb; |
ff6165b2 | 3626 | struct iov_iter __iter, *iter = &__iter; |
e8c2bc1f | 3627 | struct io_async_rw *rw = req->async_data; |
fa15bafb | 3628 | ssize_t ret, ret2, io_size; |
2b188cc1 | 3629 | |
2846c481 | 3630 | if (rw) { |
e8c2bc1f | 3631 | iter = &rw->iter; |
2846c481 PB |
3632 | iovec = NULL; |
3633 | } else { | |
3634 | ret = io_import_iovec(WRITE, req, &iovec, iter, !force_nonblock); | |
3635 | if (ret < 0) | |
3636 | return ret; | |
3637 | } | |
632546c4 | 3638 | io_size = iov_iter_count(iter); |
fa15bafb | 3639 | req->result = io_size; |
2b188cc1 | 3640 | |
fd6c2e4c JA |
3641 | /* Ensure we clear previously set non-block flag */ |
3642 | if (!force_nonblock) | |
a88fc400 PB |
3643 | kiocb->ki_flags &= ~IOCB_NOWAIT; |
3644 | else | |
3645 | kiocb->ki_flags |= IOCB_NOWAIT; | |
fd6c2e4c | 3646 | |
24c74678 | 3647 | /* If the file doesn't support async, just async punt */ |
af197f50 | 3648 | if (force_nonblock && !io_file_supports_async(req->file, WRITE)) |
f67676d1 | 3649 | goto copy_iov; |
31b51510 | 3650 | |
10d59345 JA |
3651 | /* file path doesn't support NOWAIT for non-direct_IO */ |
3652 | if (force_nonblock && !(kiocb->ki_flags & IOCB_DIRECT) && | |
3653 | (req->flags & REQ_F_ISREG)) | |
f67676d1 | 3654 | goto copy_iov; |
31b51510 | 3655 | |
632546c4 | 3656 | ret = rw_verify_area(WRITE, req->file, io_kiocb_ppos(kiocb), io_size); |
fa15bafb PB |
3657 | if (unlikely(ret)) |
3658 | goto out_free; | |
4ed734b0 | 3659 | |
fa15bafb PB |
3660 | /* |
3661 | * Open-code file_start_write here to grab freeze protection, | |
3662 | * which will be released by another thread in | |
3663 | * io_complete_rw(). Fool lockdep by telling it the lock got | |
3664 | * released so that it doesn't complain about the held lock when | |
3665 | * we return to userspace. | |
3666 | */ | |
3667 | if (req->flags & REQ_F_ISREG) { | |
8a3c84b6 | 3668 | sb_start_write(file_inode(req->file)->i_sb); |
fa15bafb PB |
3669 | __sb_writers_release(file_inode(req->file)->i_sb, |
3670 | SB_FREEZE_WRITE); | |
3671 | } | |
3672 | kiocb->ki_flags |= IOCB_WRITE; | |
4ed734b0 | 3673 | |
fa15bafb | 3674 | if (req->file->f_op->write_iter) |
ff6165b2 | 3675 | ret2 = call_write_iter(req->file, kiocb, iter); |
2dd2111d | 3676 | else if (req->file->f_op->write) |
4017eb91 | 3677 | ret2 = loop_rw_iter(WRITE, req, iter); |
2dd2111d GH |
3678 | else |
3679 | ret2 = -EINVAL; | |
4ed734b0 | 3680 | |
fa15bafb PB |
3681 | /* |
3682 | * Raw bdev writes will return -EOPNOTSUPP for IOCB_NOWAIT. Just | |
3683 | * retry them without IOCB_NOWAIT. | |
3684 | */ | |
3685 | if (ret2 == -EOPNOTSUPP && (kiocb->ki_flags & IOCB_NOWAIT)) | |
3686 | ret2 = -EAGAIN; | |
355afaeb JA |
3687 | /* no retry on NONBLOCK marked file */ |
3688 | if (ret2 == -EAGAIN && (req->file->f_flags & O_NONBLOCK)) | |
3689 | goto done; | |
fa15bafb | 3690 | if (!force_nonblock || ret2 != -EAGAIN) { |
eefdf30f JA |
3691 | /* IOPOLL retry should happen for io-wq threads */ |
3692 | if ((req->ctx->flags & IORING_SETUP_IOPOLL) && ret2 == -EAGAIN) | |
3693 | goto copy_iov; | |
355afaeb | 3694 | done: |
fa15bafb PB |
3695 | kiocb_done(kiocb, ret2, cs); |
3696 | } else { | |
f67676d1 | 3697 | copy_iov: |
84216315 | 3698 | /* some cases will consume bytes even on error returns */ |
632546c4 | 3699 | iov_iter_revert(iter, io_size - iov_iter_count(iter)); |
227c0c96 | 3700 | ret = io_setup_async_rw(req, iovec, inline_vecs, iter, false); |
ff6165b2 JA |
3701 | if (!ret) |
3702 | return -EAGAIN; | |
2b188cc1 | 3703 | } |
31b51510 | 3704 | out_free: |
f261c168 | 3705 | /* it's reportedly faster than delegating the null check to kfree() */ |
252917c3 | 3706 | if (iovec) |
6f2cc166 | 3707 | kfree(iovec); |
2b188cc1 JA |
3708 | return ret; |
3709 | } | |
3710 | ||
80a261fd JA |
3711 | static int io_renameat_prep(struct io_kiocb *req, |
3712 | const struct io_uring_sqe *sqe) | |
3713 | { | |
3714 | struct io_rename *ren = &req->rename; | |
3715 | const char __user *oldf, *newf; | |
3716 | ||
3717 | if (unlikely(req->flags & REQ_F_FIXED_FILE)) | |
3718 | return -EBADF; | |
3719 | ||
3720 | ren->old_dfd = READ_ONCE(sqe->fd); | |
3721 | oldf = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
3722 | newf = u64_to_user_ptr(READ_ONCE(sqe->addr2)); | |
3723 | ren->new_dfd = READ_ONCE(sqe->len); | |
3724 | ren->flags = READ_ONCE(sqe->rename_flags); | |
3725 | ||
3726 | ren->oldpath = getname(oldf); | |
3727 | if (IS_ERR(ren->oldpath)) | |
3728 | return PTR_ERR(ren->oldpath); | |
3729 | ||
3730 | ren->newpath = getname(newf); | |
3731 | if (IS_ERR(ren->newpath)) { | |
3732 | putname(ren->oldpath); | |
3733 | return PTR_ERR(ren->newpath); | |
3734 | } | |
3735 | ||
3736 | req->flags |= REQ_F_NEED_CLEANUP; | |
3737 | return 0; | |
3738 | } | |
3739 | ||
3740 | static int io_renameat(struct io_kiocb *req, bool force_nonblock) | |
3741 | { | |
3742 | struct io_rename *ren = &req->rename; | |
3743 | int ret; | |
3744 | ||
3745 | if (force_nonblock) | |
3746 | return -EAGAIN; | |
3747 | ||
3748 | ret = do_renameat2(ren->old_dfd, ren->oldpath, ren->new_dfd, | |
3749 | ren->newpath, ren->flags); | |
3750 | ||
3751 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
3752 | if (ret < 0) | |
3753 | req_set_fail_links(req); | |
3754 | io_req_complete(req, ret); | |
3755 | return 0; | |
3756 | } | |
3757 | ||
14a1143b JA |
3758 | static int io_unlinkat_prep(struct io_kiocb *req, |
3759 | const struct io_uring_sqe *sqe) | |
3760 | { | |
3761 | struct io_unlink *un = &req->unlink; | |
3762 | const char __user *fname; | |
3763 | ||
3764 | if (unlikely(req->flags & REQ_F_FIXED_FILE)) | |
3765 | return -EBADF; | |
3766 | ||
3767 | un->dfd = READ_ONCE(sqe->fd); | |
3768 | ||
3769 | un->flags = READ_ONCE(sqe->unlink_flags); | |
3770 | if (un->flags & ~AT_REMOVEDIR) | |
3771 | return -EINVAL; | |
3772 | ||
3773 | fname = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
3774 | un->filename = getname(fname); | |
3775 | if (IS_ERR(un->filename)) | |
3776 | return PTR_ERR(un->filename); | |
3777 | ||
3778 | req->flags |= REQ_F_NEED_CLEANUP; | |
3779 | return 0; | |
3780 | } | |
3781 | ||
3782 | static int io_unlinkat(struct io_kiocb *req, bool force_nonblock) | |
3783 | { | |
3784 | struct io_unlink *un = &req->unlink; | |
3785 | int ret; | |
3786 | ||
3787 | if (force_nonblock) | |
3788 | return -EAGAIN; | |
3789 | ||
3790 | if (un->flags & AT_REMOVEDIR) | |
3791 | ret = do_rmdir(un->dfd, un->filename); | |
3792 | else | |
3793 | ret = do_unlinkat(un->dfd, un->filename); | |
3794 | ||
3795 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
3796 | if (ret < 0) | |
3797 | req_set_fail_links(req); | |
3798 | io_req_complete(req, ret); | |
3799 | return 0; | |
3800 | } | |
3801 | ||
36f4fa68 JA |
3802 | static int io_shutdown_prep(struct io_kiocb *req, |
3803 | const struct io_uring_sqe *sqe) | |
3804 | { | |
3805 | #if defined(CONFIG_NET) | |
3806 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) | |
3807 | return -EINVAL; | |
3808 | if (sqe->ioprio || sqe->off || sqe->addr || sqe->rw_flags || | |
3809 | sqe->buf_index) | |
3810 | return -EINVAL; | |
3811 | ||
3812 | req->shutdown.how = READ_ONCE(sqe->len); | |
3813 | return 0; | |
3814 | #else | |
3815 | return -EOPNOTSUPP; | |
3816 | #endif | |
3817 | } | |
3818 | ||
3819 | static int io_shutdown(struct io_kiocb *req, bool force_nonblock) | |
3820 | { | |
3821 | #if defined(CONFIG_NET) | |
3822 | struct socket *sock; | |
3823 | int ret; | |
3824 | ||
3825 | if (force_nonblock) | |
3826 | return -EAGAIN; | |
3827 | ||
48aba79b | 3828 | sock = sock_from_file(req->file); |
36f4fa68 | 3829 | if (unlikely(!sock)) |
48aba79b | 3830 | return -ENOTSOCK; |
36f4fa68 JA |
3831 | |
3832 | ret = __sys_shutdown_sock(sock, req->shutdown.how); | |
a146468d JA |
3833 | if (ret < 0) |
3834 | req_set_fail_links(req); | |
36f4fa68 JA |
3835 | io_req_complete(req, ret); |
3836 | return 0; | |
3837 | #else | |
3838 | return -EOPNOTSUPP; | |
3839 | #endif | |
3840 | } | |
3841 | ||
f2a8d5c7 PB |
3842 | static int __io_splice_prep(struct io_kiocb *req, |
3843 | const struct io_uring_sqe *sqe) | |
7d67af2c PB |
3844 | { |
3845 | struct io_splice* sp = &req->splice; | |
3846 | unsigned int valid_flags = SPLICE_F_FD_IN_FIXED | SPLICE_F_ALL; | |
7d67af2c | 3847 | |
3232dd02 PB |
3848 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
3849 | return -EINVAL; | |
7d67af2c PB |
3850 | |
3851 | sp->file_in = NULL; | |
7d67af2c PB |
3852 | sp->len = READ_ONCE(sqe->len); |
3853 | sp->flags = READ_ONCE(sqe->splice_flags); | |
3854 | ||
3855 | if (unlikely(sp->flags & ~valid_flags)) | |
3856 | return -EINVAL; | |
3857 | ||
8371adf5 PB |
3858 | sp->file_in = io_file_get(NULL, req, READ_ONCE(sqe->splice_fd_in), |
3859 | (sp->flags & SPLICE_F_FD_IN_FIXED)); | |
3860 | if (!sp->file_in) | |
3861 | return -EBADF; | |
7d67af2c PB |
3862 | req->flags |= REQ_F_NEED_CLEANUP; |
3863 | ||
7cdaf587 XW |
3864 | if (!S_ISREG(file_inode(sp->file_in)->i_mode)) { |
3865 | /* | |
3866 | * Splice operation will be punted aync, and here need to | |
3867 | * modify io_wq_work.flags, so initialize io_wq_work firstly. | |
3868 | */ | |
3869 | io_req_init_async(req); | |
7d67af2c | 3870 | req->work.flags |= IO_WQ_WORK_UNBOUND; |
7cdaf587 | 3871 | } |
7d67af2c PB |
3872 | |
3873 | return 0; | |
3874 | } | |
3875 | ||
f2a8d5c7 PB |
3876 | static int io_tee_prep(struct io_kiocb *req, |
3877 | const struct io_uring_sqe *sqe) | |
3878 | { | |
3879 | if (READ_ONCE(sqe->splice_off_in) || READ_ONCE(sqe->off)) | |
3880 | return -EINVAL; | |
3881 | return __io_splice_prep(req, sqe); | |
3882 | } | |
3883 | ||
3884 | static int io_tee(struct io_kiocb *req, bool force_nonblock) | |
3885 | { | |
3886 | struct io_splice *sp = &req->splice; | |
3887 | struct file *in = sp->file_in; | |
3888 | struct file *out = sp->file_out; | |
3889 | unsigned int flags = sp->flags & ~SPLICE_F_FD_IN_FIXED; | |
3890 | long ret = 0; | |
3891 | ||
3892 | if (force_nonblock) | |
3893 | return -EAGAIN; | |
3894 | if (sp->len) | |
3895 | ret = do_tee(in, out, sp->len, flags); | |
3896 | ||
3897 | io_put_file(req, in, (sp->flags & SPLICE_F_FD_IN_FIXED)); | |
3898 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
3899 | ||
f2a8d5c7 PB |
3900 | if (ret != sp->len) |
3901 | req_set_fail_links(req); | |
e1e16097 | 3902 | io_req_complete(req, ret); |
f2a8d5c7 PB |
3903 | return 0; |
3904 | } | |
3905 | ||
3906 | static int io_splice_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) | |
3907 | { | |
3908 | struct io_splice* sp = &req->splice; | |
3909 | ||
3910 | sp->off_in = READ_ONCE(sqe->splice_off_in); | |
3911 | sp->off_out = READ_ONCE(sqe->off); | |
3912 | return __io_splice_prep(req, sqe); | |
3913 | } | |
3914 | ||
014db007 | 3915 | static int io_splice(struct io_kiocb *req, bool force_nonblock) |
7d67af2c PB |
3916 | { |
3917 | struct io_splice *sp = &req->splice; | |
3918 | struct file *in = sp->file_in; | |
3919 | struct file *out = sp->file_out; | |
3920 | unsigned int flags = sp->flags & ~SPLICE_F_FD_IN_FIXED; | |
3921 | loff_t *poff_in, *poff_out; | |
c9687426 | 3922 | long ret = 0; |
7d67af2c | 3923 | |
2fb3e822 PB |
3924 | if (force_nonblock) |
3925 | return -EAGAIN; | |
7d67af2c PB |
3926 | |
3927 | poff_in = (sp->off_in == -1) ? NULL : &sp->off_in; | |
3928 | poff_out = (sp->off_out == -1) ? NULL : &sp->off_out; | |
c9687426 | 3929 | |
948a7749 | 3930 | if (sp->len) |
c9687426 | 3931 | ret = do_splice(in, poff_in, out, poff_out, sp->len, flags); |
7d67af2c PB |
3932 | |
3933 | io_put_file(req, in, (sp->flags & SPLICE_F_FD_IN_FIXED)); | |
3934 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
3935 | ||
7d67af2c PB |
3936 | if (ret != sp->len) |
3937 | req_set_fail_links(req); | |
e1e16097 | 3938 | io_req_complete(req, ret); |
7d67af2c PB |
3939 | return 0; |
3940 | } | |
3941 | ||
2b188cc1 JA |
3942 | /* |
3943 | * IORING_OP_NOP just posts a completion event, nothing else. | |
3944 | */ | |
229a7b63 | 3945 | static int io_nop(struct io_kiocb *req, struct io_comp_state *cs) |
2b188cc1 JA |
3946 | { |
3947 | struct io_ring_ctx *ctx = req->ctx; | |
2b188cc1 | 3948 | |
def596e9 JA |
3949 | if (unlikely(ctx->flags & IORING_SETUP_IOPOLL)) |
3950 | return -EINVAL; | |
3951 | ||
229a7b63 | 3952 | __io_req_complete(req, 0, 0, cs); |
2b188cc1 JA |
3953 | return 0; |
3954 | } | |
3955 | ||
3529d8c2 | 3956 | static int io_prep_fsync(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
c992fe29 | 3957 | { |
6b06314c | 3958 | struct io_ring_ctx *ctx = req->ctx; |
c992fe29 | 3959 | |
09bb8394 JA |
3960 | if (!req->file) |
3961 | return -EBADF; | |
c992fe29 | 3962 | |
6b06314c | 3963 | if (unlikely(ctx->flags & IORING_SETUP_IOPOLL)) |
def596e9 | 3964 | return -EINVAL; |
edafccee | 3965 | if (unlikely(sqe->addr || sqe->ioprio || sqe->buf_index)) |
c992fe29 CH |
3966 | return -EINVAL; |
3967 | ||
8ed8d3c3 JA |
3968 | req->sync.flags = READ_ONCE(sqe->fsync_flags); |
3969 | if (unlikely(req->sync.flags & ~IORING_FSYNC_DATASYNC)) | |
3970 | return -EINVAL; | |
3971 | ||
3972 | req->sync.off = READ_ONCE(sqe->off); | |
3973 | req->sync.len = READ_ONCE(sqe->len); | |
c992fe29 CH |
3974 | return 0; |
3975 | } | |
3976 | ||
ac45abc0 | 3977 | static int io_fsync(struct io_kiocb *req, bool force_nonblock) |
8ed8d3c3 | 3978 | { |
8ed8d3c3 | 3979 | loff_t end = req->sync.off + req->sync.len; |
8ed8d3c3 JA |
3980 | int ret; |
3981 | ||
ac45abc0 PB |
3982 | /* fsync always requires a blocking context */ |
3983 | if (force_nonblock) | |
3984 | return -EAGAIN; | |
3985 | ||
9adbd45d | 3986 | ret = vfs_fsync_range(req->file, req->sync.off, |
8ed8d3c3 JA |
3987 | end > 0 ? end : LLONG_MAX, |
3988 | req->sync.flags & IORING_FSYNC_DATASYNC); | |
3989 | if (ret < 0) | |
3990 | req_set_fail_links(req); | |
e1e16097 | 3991 | io_req_complete(req, ret); |
c992fe29 CH |
3992 | return 0; |
3993 | } | |
3994 | ||
d63d1b5e JA |
3995 | static int io_fallocate_prep(struct io_kiocb *req, |
3996 | const struct io_uring_sqe *sqe) | |
3997 | { | |
3998 | if (sqe->ioprio || sqe->buf_index || sqe->rw_flags) | |
3999 | return -EINVAL; | |
3232dd02 PB |
4000 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4001 | return -EINVAL; | |
d63d1b5e JA |
4002 | |
4003 | req->sync.off = READ_ONCE(sqe->off); | |
4004 | req->sync.len = READ_ONCE(sqe->addr); | |
4005 | req->sync.mode = READ_ONCE(sqe->len); | |
4006 | return 0; | |
4007 | } | |
4008 | ||
014db007 | 4009 | static int io_fallocate(struct io_kiocb *req, bool force_nonblock) |
5d17b4a4 | 4010 | { |
ac45abc0 PB |
4011 | int ret; |
4012 | ||
d63d1b5e | 4013 | /* fallocate always requiring blocking context */ |
ac45abc0 | 4014 | if (force_nonblock) |
5d17b4a4 | 4015 | return -EAGAIN; |
ac45abc0 PB |
4016 | ret = vfs_fallocate(req->file, req->sync.mode, req->sync.off, |
4017 | req->sync.len); | |
ac45abc0 PB |
4018 | if (ret < 0) |
4019 | req_set_fail_links(req); | |
e1e16097 | 4020 | io_req_complete(req, ret); |
5d17b4a4 JA |
4021 | return 0; |
4022 | } | |
4023 | ||
ec65fea5 | 4024 | static int __io_openat_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
b7bb4f7d | 4025 | { |
f8748881 | 4026 | const char __user *fname; |
15b71abe | 4027 | int ret; |
b7bb4f7d | 4028 | |
ec65fea5 | 4029 | if (unlikely(sqe->ioprio || sqe->buf_index)) |
15b71abe | 4030 | return -EINVAL; |
ec65fea5 | 4031 | if (unlikely(req->flags & REQ_F_FIXED_FILE)) |
cf3040ca | 4032 | return -EBADF; |
03b1230c | 4033 | |
ec65fea5 PB |
4034 | /* open.how should be already initialised */ |
4035 | if (!(req->open.how.flags & O_PATH) && force_o_largefile()) | |
08a1d26e | 4036 | req->open.how.flags |= O_LARGEFILE; |
3529d8c2 | 4037 | |
25e72d10 PB |
4038 | req->open.dfd = READ_ONCE(sqe->fd); |
4039 | fname = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
f8748881 | 4040 | req->open.filename = getname(fname); |
15b71abe JA |
4041 | if (IS_ERR(req->open.filename)) { |
4042 | ret = PTR_ERR(req->open.filename); | |
4043 | req->open.filename = NULL; | |
4044 | return ret; | |
4045 | } | |
4022e7af | 4046 | req->open.nofile = rlimit(RLIMIT_NOFILE); |
944d1444 | 4047 | req->open.ignore_nonblock = false; |
8fef80bf | 4048 | req->flags |= REQ_F_NEED_CLEANUP; |
15b71abe | 4049 | return 0; |
03b1230c JA |
4050 | } |
4051 | ||
ec65fea5 PB |
4052 | static int io_openat_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
4053 | { | |
4054 | u64 flags, mode; | |
4055 | ||
14587a46 | 4056 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4eb8dded | 4057 | return -EINVAL; |
ec65fea5 PB |
4058 | mode = READ_ONCE(sqe->len); |
4059 | flags = READ_ONCE(sqe->open_flags); | |
4060 | req->open.how = build_open_how(flags, mode); | |
4061 | return __io_openat_prep(req, sqe); | |
4062 | } | |
4063 | ||
cebdb986 | 4064 | static int io_openat2_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
aa1fa28f | 4065 | { |
cebdb986 | 4066 | struct open_how __user *how; |
cebdb986 | 4067 | size_t len; |
0fa03c62 JA |
4068 | int ret; |
4069 | ||
14587a46 | 4070 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4eb8dded | 4071 | return -EINVAL; |
cebdb986 JA |
4072 | how = u64_to_user_ptr(READ_ONCE(sqe->addr2)); |
4073 | len = READ_ONCE(sqe->len); | |
cebdb986 JA |
4074 | if (len < OPEN_HOW_SIZE_VER0) |
4075 | return -EINVAL; | |
3529d8c2 | 4076 | |
cebdb986 JA |
4077 | ret = copy_struct_from_user(&req->open.how, sizeof(req->open.how), how, |
4078 | len); | |
4079 | if (ret) | |
4080 | return ret; | |
3529d8c2 | 4081 | |
ec65fea5 | 4082 | return __io_openat_prep(req, sqe); |
cebdb986 JA |
4083 | } |
4084 | ||
014db007 | 4085 | static int io_openat2(struct io_kiocb *req, bool force_nonblock) |
15b71abe JA |
4086 | { |
4087 | struct open_flags op; | |
15b71abe JA |
4088 | struct file *file; |
4089 | int ret; | |
4090 | ||
944d1444 | 4091 | if (force_nonblock && !req->open.ignore_nonblock) |
15b71abe | 4092 | return -EAGAIN; |
15b71abe | 4093 | |
cebdb986 | 4094 | ret = build_open_flags(&req->open.how, &op); |
15b71abe JA |
4095 | if (ret) |
4096 | goto err; | |
4097 | ||
4022e7af | 4098 | ret = __get_unused_fd_flags(req->open.how.flags, req->open.nofile); |
15b71abe JA |
4099 | if (ret < 0) |
4100 | goto err; | |
4101 | ||
4102 | file = do_filp_open(req->open.dfd, req->open.filename, &op); | |
4103 | if (IS_ERR(file)) { | |
4104 | put_unused_fd(ret); | |
4105 | ret = PTR_ERR(file); | |
944d1444 JA |
4106 | /* |
4107 | * A work-around to ensure that /proc/self works that way | |
4108 | * that it should - if we get -EOPNOTSUPP back, then assume | |
4109 | * that proc_self_get_link() failed us because we're in async | |
4110 | * context. We should be safe to retry this from the task | |
4111 | * itself with force_nonblock == false set, as it should not | |
4112 | * block on lookup. Would be nice to know this upfront and | |
4113 | * avoid the async dance, but doesn't seem feasible. | |
4114 | */ | |
4115 | if (ret == -EOPNOTSUPP && io_wq_current_is_worker()) { | |
4116 | req->open.ignore_nonblock = true; | |
4117 | refcount_inc(&req->refs); | |
4118 | io_req_task_queue(req); | |
4119 | return 0; | |
4120 | } | |
15b71abe JA |
4121 | } else { |
4122 | fsnotify_open(file); | |
4123 | fd_install(ret, file); | |
4124 | } | |
4125 | err: | |
4126 | putname(req->open.filename); | |
8fef80bf | 4127 | req->flags &= ~REQ_F_NEED_CLEANUP; |
15b71abe JA |
4128 | if (ret < 0) |
4129 | req_set_fail_links(req); | |
e1e16097 | 4130 | io_req_complete(req, ret); |
15b71abe JA |
4131 | return 0; |
4132 | } | |
4133 | ||
014db007 | 4134 | static int io_openat(struct io_kiocb *req, bool force_nonblock) |
cebdb986 | 4135 | { |
014db007 | 4136 | return io_openat2(req, force_nonblock); |
cebdb986 JA |
4137 | } |
4138 | ||
067524e9 JA |
4139 | static int io_remove_buffers_prep(struct io_kiocb *req, |
4140 | const struct io_uring_sqe *sqe) | |
4141 | { | |
4142 | struct io_provide_buf *p = &req->pbuf; | |
4143 | u64 tmp; | |
4144 | ||
4145 | if (sqe->ioprio || sqe->rw_flags || sqe->addr || sqe->len || sqe->off) | |
4146 | return -EINVAL; | |
4147 | ||
4148 | tmp = READ_ONCE(sqe->fd); | |
4149 | if (!tmp || tmp > USHRT_MAX) | |
4150 | return -EINVAL; | |
4151 | ||
4152 | memset(p, 0, sizeof(*p)); | |
4153 | p->nbufs = tmp; | |
4154 | p->bgid = READ_ONCE(sqe->buf_group); | |
4155 | return 0; | |
4156 | } | |
4157 | ||
4158 | static int __io_remove_buffers(struct io_ring_ctx *ctx, struct io_buffer *buf, | |
4159 | int bgid, unsigned nbufs) | |
4160 | { | |
4161 | unsigned i = 0; | |
4162 | ||
4163 | /* shouldn't happen */ | |
4164 | if (!nbufs) | |
4165 | return 0; | |
4166 | ||
4167 | /* the head kbuf is the list itself */ | |
4168 | while (!list_empty(&buf->list)) { | |
4169 | struct io_buffer *nxt; | |
4170 | ||
4171 | nxt = list_first_entry(&buf->list, struct io_buffer, list); | |
4172 | list_del(&nxt->list); | |
4173 | kfree(nxt); | |
4174 | if (++i == nbufs) | |
4175 | return i; | |
4176 | } | |
4177 | i++; | |
4178 | kfree(buf); | |
4179 | idr_remove(&ctx->io_buffer_idr, bgid); | |
4180 | ||
4181 | return i; | |
4182 | } | |
4183 | ||
229a7b63 JA |
4184 | static int io_remove_buffers(struct io_kiocb *req, bool force_nonblock, |
4185 | struct io_comp_state *cs) | |
067524e9 JA |
4186 | { |
4187 | struct io_provide_buf *p = &req->pbuf; | |
4188 | struct io_ring_ctx *ctx = req->ctx; | |
4189 | struct io_buffer *head; | |
4190 | int ret = 0; | |
4191 | ||
4192 | io_ring_submit_lock(ctx, !force_nonblock); | |
4193 | ||
4194 | lockdep_assert_held(&ctx->uring_lock); | |
4195 | ||
4196 | ret = -ENOENT; | |
4197 | head = idr_find(&ctx->io_buffer_idr, p->bgid); | |
4198 | if (head) | |
4199 | ret = __io_remove_buffers(ctx, head, p->bgid, p->nbufs); | |
067524e9 JA |
4200 | if (ret < 0) |
4201 | req_set_fail_links(req); | |
067524e9 | 4202 | |
31bff9a5 PB |
4203 | /* need to hold the lock to complete IOPOLL requests */ |
4204 | if (ctx->flags & IORING_SETUP_IOPOLL) { | |
4205 | __io_req_complete(req, ret, 0, cs); | |
4206 | io_ring_submit_unlock(ctx, !force_nonblock); | |
4207 | } else { | |
4208 | io_ring_submit_unlock(ctx, !force_nonblock); | |
4209 | __io_req_complete(req, ret, 0, cs); | |
4210 | } | |
067524e9 JA |
4211 | return 0; |
4212 | } | |
4213 | ||
ddf0322d JA |
4214 | static int io_provide_buffers_prep(struct io_kiocb *req, |
4215 | const struct io_uring_sqe *sqe) | |
4216 | { | |
4217 | struct io_provide_buf *p = &req->pbuf; | |
4218 | u64 tmp; | |
4219 | ||
4220 | if (sqe->ioprio || sqe->rw_flags) | |
4221 | return -EINVAL; | |
4222 | ||
4223 | tmp = READ_ONCE(sqe->fd); | |
4224 | if (!tmp || tmp > USHRT_MAX) | |
4225 | return -E2BIG; | |
4226 | p->nbufs = tmp; | |
4227 | p->addr = READ_ONCE(sqe->addr); | |
4228 | p->len = READ_ONCE(sqe->len); | |
4229 | ||
efe68c1c | 4230 | if (!access_ok(u64_to_user_ptr(p->addr), (p->len * p->nbufs))) |
ddf0322d JA |
4231 | return -EFAULT; |
4232 | ||
4233 | p->bgid = READ_ONCE(sqe->buf_group); | |
4234 | tmp = READ_ONCE(sqe->off); | |
4235 | if (tmp > USHRT_MAX) | |
4236 | return -E2BIG; | |
4237 | p->bid = tmp; | |
4238 | return 0; | |
4239 | } | |
4240 | ||
4241 | static int io_add_buffers(struct io_provide_buf *pbuf, struct io_buffer **head) | |
4242 | { | |
4243 | struct io_buffer *buf; | |
4244 | u64 addr = pbuf->addr; | |
4245 | int i, bid = pbuf->bid; | |
4246 | ||
4247 | for (i = 0; i < pbuf->nbufs; i++) { | |
4248 | buf = kmalloc(sizeof(*buf), GFP_KERNEL); | |
4249 | if (!buf) | |
4250 | break; | |
4251 | ||
4252 | buf->addr = addr; | |
4253 | buf->len = pbuf->len; | |
4254 | buf->bid = bid; | |
4255 | addr += pbuf->len; | |
4256 | bid++; | |
4257 | if (!*head) { | |
4258 | INIT_LIST_HEAD(&buf->list); | |
4259 | *head = buf; | |
4260 | } else { | |
4261 | list_add_tail(&buf->list, &(*head)->list); | |
4262 | } | |
4263 | } | |
4264 | ||
4265 | return i ? i : -ENOMEM; | |
4266 | } | |
4267 | ||
229a7b63 JA |
4268 | static int io_provide_buffers(struct io_kiocb *req, bool force_nonblock, |
4269 | struct io_comp_state *cs) | |
ddf0322d JA |
4270 | { |
4271 | struct io_provide_buf *p = &req->pbuf; | |
4272 | struct io_ring_ctx *ctx = req->ctx; | |
4273 | struct io_buffer *head, *list; | |
4274 | int ret = 0; | |
4275 | ||
4276 | io_ring_submit_lock(ctx, !force_nonblock); | |
4277 | ||
4278 | lockdep_assert_held(&ctx->uring_lock); | |
4279 | ||
4280 | list = head = idr_find(&ctx->io_buffer_idr, p->bgid); | |
4281 | ||
4282 | ret = io_add_buffers(p, &head); | |
4283 | if (ret < 0) | |
4284 | goto out; | |
4285 | ||
4286 | if (!list) { | |
4287 | ret = idr_alloc(&ctx->io_buffer_idr, head, p->bgid, p->bgid + 1, | |
4288 | GFP_KERNEL); | |
4289 | if (ret < 0) { | |
067524e9 | 4290 | __io_remove_buffers(ctx, head, p->bgid, -1U); |
ddf0322d JA |
4291 | goto out; |
4292 | } | |
4293 | } | |
4294 | out: | |
ddf0322d JA |
4295 | if (ret < 0) |
4296 | req_set_fail_links(req); | |
31bff9a5 PB |
4297 | |
4298 | /* need to hold the lock to complete IOPOLL requests */ | |
4299 | if (ctx->flags & IORING_SETUP_IOPOLL) { | |
4300 | __io_req_complete(req, ret, 0, cs); | |
4301 | io_ring_submit_unlock(ctx, !force_nonblock); | |
4302 | } else { | |
4303 | io_ring_submit_unlock(ctx, !force_nonblock); | |
4304 | __io_req_complete(req, ret, 0, cs); | |
4305 | } | |
ddf0322d | 4306 | return 0; |
cebdb986 JA |
4307 | } |
4308 | ||
3e4827b0 JA |
4309 | static int io_epoll_ctl_prep(struct io_kiocb *req, |
4310 | const struct io_uring_sqe *sqe) | |
4311 | { | |
4312 | #if defined(CONFIG_EPOLL) | |
4313 | if (sqe->ioprio || sqe->buf_index) | |
4314 | return -EINVAL; | |
6ca56f84 | 4315 | if (unlikely(req->ctx->flags & (IORING_SETUP_IOPOLL | IORING_SETUP_SQPOLL))) |
3232dd02 | 4316 | return -EINVAL; |
3e4827b0 JA |
4317 | |
4318 | req->epoll.epfd = READ_ONCE(sqe->fd); | |
4319 | req->epoll.op = READ_ONCE(sqe->len); | |
4320 | req->epoll.fd = READ_ONCE(sqe->off); | |
4321 | ||
4322 | if (ep_op_has_event(req->epoll.op)) { | |
4323 | struct epoll_event __user *ev; | |
4324 | ||
4325 | ev = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
4326 | if (copy_from_user(&req->epoll.event, ev, sizeof(*ev))) | |
4327 | return -EFAULT; | |
4328 | } | |
4329 | ||
4330 | return 0; | |
4331 | #else | |
4332 | return -EOPNOTSUPP; | |
4333 | #endif | |
4334 | } | |
4335 | ||
229a7b63 JA |
4336 | static int io_epoll_ctl(struct io_kiocb *req, bool force_nonblock, |
4337 | struct io_comp_state *cs) | |
3e4827b0 JA |
4338 | { |
4339 | #if defined(CONFIG_EPOLL) | |
4340 | struct io_epoll *ie = &req->epoll; | |
4341 | int ret; | |
4342 | ||
4343 | ret = do_epoll_ctl(ie->epfd, ie->op, ie->fd, &ie->event, force_nonblock); | |
4344 | if (force_nonblock && ret == -EAGAIN) | |
4345 | return -EAGAIN; | |
4346 | ||
4347 | if (ret < 0) | |
4348 | req_set_fail_links(req); | |
229a7b63 | 4349 | __io_req_complete(req, ret, 0, cs); |
3e4827b0 JA |
4350 | return 0; |
4351 | #else | |
4352 | return -EOPNOTSUPP; | |
4353 | #endif | |
4354 | } | |
4355 | ||
c1ca757b JA |
4356 | static int io_madvise_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
4357 | { | |
4358 | #if defined(CONFIG_ADVISE_SYSCALLS) && defined(CONFIG_MMU) | |
4359 | if (sqe->ioprio || sqe->buf_index || sqe->off) | |
4360 | return -EINVAL; | |
3232dd02 PB |
4361 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4362 | return -EINVAL; | |
c1ca757b JA |
4363 | |
4364 | req->madvise.addr = READ_ONCE(sqe->addr); | |
4365 | req->madvise.len = READ_ONCE(sqe->len); | |
4366 | req->madvise.advice = READ_ONCE(sqe->fadvise_advice); | |
4367 | return 0; | |
4368 | #else | |
4369 | return -EOPNOTSUPP; | |
4370 | #endif | |
4371 | } | |
4372 | ||
014db007 | 4373 | static int io_madvise(struct io_kiocb *req, bool force_nonblock) |
c1ca757b JA |
4374 | { |
4375 | #if defined(CONFIG_ADVISE_SYSCALLS) && defined(CONFIG_MMU) | |
4376 | struct io_madvise *ma = &req->madvise; | |
4377 | int ret; | |
4378 | ||
4379 | if (force_nonblock) | |
4380 | return -EAGAIN; | |
4381 | ||
0726b01e | 4382 | ret = do_madvise(current->mm, ma->addr, ma->len, ma->advice); |
c1ca757b JA |
4383 | if (ret < 0) |
4384 | req_set_fail_links(req); | |
e1e16097 | 4385 | io_req_complete(req, ret); |
c1ca757b JA |
4386 | return 0; |
4387 | #else | |
4388 | return -EOPNOTSUPP; | |
4389 | #endif | |
4390 | } | |
4391 | ||
4840e418 JA |
4392 | static int io_fadvise_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
4393 | { | |
4394 | if (sqe->ioprio || sqe->buf_index || sqe->addr) | |
4395 | return -EINVAL; | |
3232dd02 PB |
4396 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4397 | return -EINVAL; | |
4840e418 JA |
4398 | |
4399 | req->fadvise.offset = READ_ONCE(sqe->off); | |
4400 | req->fadvise.len = READ_ONCE(sqe->len); | |
4401 | req->fadvise.advice = READ_ONCE(sqe->fadvise_advice); | |
4402 | return 0; | |
4403 | } | |
4404 | ||
014db007 | 4405 | static int io_fadvise(struct io_kiocb *req, bool force_nonblock) |
4840e418 JA |
4406 | { |
4407 | struct io_fadvise *fa = &req->fadvise; | |
4408 | int ret; | |
4409 | ||
3e69426d JA |
4410 | if (force_nonblock) { |
4411 | switch (fa->advice) { | |
4412 | case POSIX_FADV_NORMAL: | |
4413 | case POSIX_FADV_RANDOM: | |
4414 | case POSIX_FADV_SEQUENTIAL: | |
4415 | break; | |
4416 | default: | |
4417 | return -EAGAIN; | |
4418 | } | |
4419 | } | |
4840e418 JA |
4420 | |
4421 | ret = vfs_fadvise(req->file, fa->offset, fa->len, fa->advice); | |
4422 | if (ret < 0) | |
4423 | req_set_fail_links(req); | |
e1e16097 | 4424 | io_req_complete(req, ret); |
4840e418 JA |
4425 | return 0; |
4426 | } | |
4427 | ||
eddc7ef5 JA |
4428 | static int io_statx_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
4429 | { | |
6ca56f84 | 4430 | if (unlikely(req->ctx->flags & (IORING_SETUP_IOPOLL | IORING_SETUP_SQPOLL))) |
3232dd02 | 4431 | return -EINVAL; |
eddc7ef5 JA |
4432 | if (sqe->ioprio || sqe->buf_index) |
4433 | return -EINVAL; | |
9c280f90 | 4434 | if (req->flags & REQ_F_FIXED_FILE) |
cf3040ca | 4435 | return -EBADF; |
eddc7ef5 | 4436 | |
1d9e1288 BM |
4437 | req->statx.dfd = READ_ONCE(sqe->fd); |
4438 | req->statx.mask = READ_ONCE(sqe->len); | |
e62753e4 | 4439 | req->statx.filename = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
1d9e1288 BM |
4440 | req->statx.buffer = u64_to_user_ptr(READ_ONCE(sqe->addr2)); |
4441 | req->statx.flags = READ_ONCE(sqe->statx_flags); | |
eddc7ef5 JA |
4442 | |
4443 | return 0; | |
4444 | } | |
4445 | ||
014db007 | 4446 | static int io_statx(struct io_kiocb *req, bool force_nonblock) |
eddc7ef5 | 4447 | { |
1d9e1288 | 4448 | struct io_statx *ctx = &req->statx; |
eddc7ef5 JA |
4449 | int ret; |
4450 | ||
5b0bbee4 JA |
4451 | if (force_nonblock) { |
4452 | /* only need file table for an actual valid fd */ | |
4453 | if (ctx->dfd == -1 || ctx->dfd == AT_FDCWD) | |
4454 | req->flags |= REQ_F_NO_FILE_TABLE; | |
eddc7ef5 | 4455 | return -EAGAIN; |
5b0bbee4 | 4456 | } |
eddc7ef5 | 4457 | |
e62753e4 BM |
4458 | ret = do_statx(ctx->dfd, ctx->filename, ctx->flags, ctx->mask, |
4459 | ctx->buffer); | |
eddc7ef5 | 4460 | |
eddc7ef5 JA |
4461 | if (ret < 0) |
4462 | req_set_fail_links(req); | |
e1e16097 | 4463 | io_req_complete(req, ret); |
eddc7ef5 JA |
4464 | return 0; |
4465 | } | |
4466 | ||
b5dba59e JA |
4467 | static int io_close_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
4468 | { | |
4469 | /* | |
4470 | * If we queue this for async, it must not be cancellable. That would | |
7cdaf587 XW |
4471 | * leave the 'file' in an undeterminate state, and here need to modify |
4472 | * io_wq_work.flags, so initialize io_wq_work firstly. | |
b5dba59e | 4473 | */ |
7cdaf587 | 4474 | io_req_init_async(req); |
b5dba59e JA |
4475 | req->work.flags |= IO_WQ_WORK_NO_CANCEL; |
4476 | ||
14587a46 | 4477 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
3232dd02 | 4478 | return -EINVAL; |
b5dba59e JA |
4479 | if (sqe->ioprio || sqe->off || sqe->addr || sqe->len || |
4480 | sqe->rw_flags || sqe->buf_index) | |
4481 | return -EINVAL; | |
9c280f90 | 4482 | if (req->flags & REQ_F_FIXED_FILE) |
cf3040ca | 4483 | return -EBADF; |
b5dba59e JA |
4484 | |
4485 | req->close.fd = READ_ONCE(sqe->fd); | |
0f212204 | 4486 | if ((req->file && req->file->f_op == &io_uring_fops)) |
fd2206e4 | 4487 | return -EBADF; |
b5dba59e | 4488 | |
3af73b28 | 4489 | req->close.put_file = NULL; |
b5dba59e | 4490 | return 0; |
b5dba59e JA |
4491 | } |
4492 | ||
229a7b63 JA |
4493 | static int io_close(struct io_kiocb *req, bool force_nonblock, |
4494 | struct io_comp_state *cs) | |
b5dba59e | 4495 | { |
3af73b28 | 4496 | struct io_close *close = &req->close; |
b5dba59e JA |
4497 | int ret; |
4498 | ||
3af73b28 PB |
4499 | /* might be already done during nonblock submission */ |
4500 | if (!close->put_file) { | |
9fe83c43 | 4501 | ret = close_fd_get_file(close->fd, &close->put_file); |
3af73b28 PB |
4502 | if (ret < 0) |
4503 | return (ret == -ENOENT) ? -EBADF : ret; | |
4504 | } | |
b5dba59e JA |
4505 | |
4506 | /* if the file has a flush method, be safe and punt to async */ | |
3af73b28 | 4507 | if (close->put_file->f_op->flush && force_nonblock) { |
24c74678 PB |
4508 | /* was never set, but play safe */ |
4509 | req->flags &= ~REQ_F_NOWAIT; | |
0bf0eefd | 4510 | /* avoid grabbing files - we don't need the files */ |
24c74678 | 4511 | req->flags |= REQ_F_NO_FILE_TABLE; |
0bf0eefd | 4512 | return -EAGAIN; |
a2100672 | 4513 | } |
b5dba59e | 4514 | |
3af73b28 | 4515 | /* No ->flush() or already async, safely close from here */ |
98447d65 | 4516 | ret = filp_close(close->put_file, req->work.identity->files); |
3af73b28 PB |
4517 | if (ret < 0) |
4518 | req_set_fail_links(req); | |
3af73b28 PB |
4519 | fput(close->put_file); |
4520 | close->put_file = NULL; | |
229a7b63 | 4521 | __io_req_complete(req, ret, 0, cs); |
1a417f4e | 4522 | return 0; |
b5dba59e JA |
4523 | } |
4524 | ||
3529d8c2 | 4525 | static int io_prep_sfr(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
5d17b4a4 JA |
4526 | { |
4527 | struct io_ring_ctx *ctx = req->ctx; | |
5d17b4a4 JA |
4528 | |
4529 | if (!req->file) | |
4530 | return -EBADF; | |
5d17b4a4 JA |
4531 | |
4532 | if (unlikely(ctx->flags & IORING_SETUP_IOPOLL)) | |
4533 | return -EINVAL; | |
4534 | if (unlikely(sqe->addr || sqe->ioprio || sqe->buf_index)) | |
4535 | return -EINVAL; | |
4536 | ||
8ed8d3c3 JA |
4537 | req->sync.off = READ_ONCE(sqe->off); |
4538 | req->sync.len = READ_ONCE(sqe->len); | |
4539 | req->sync.flags = READ_ONCE(sqe->sync_range_flags); | |
8ed8d3c3 JA |
4540 | return 0; |
4541 | } | |
4542 | ||
ac45abc0 | 4543 | static int io_sync_file_range(struct io_kiocb *req, bool force_nonblock) |
8ed8d3c3 | 4544 | { |
8ed8d3c3 JA |
4545 | int ret; |
4546 | ||
ac45abc0 PB |
4547 | /* sync_file_range always requires a blocking context */ |
4548 | if (force_nonblock) | |
4549 | return -EAGAIN; | |
4550 | ||
9adbd45d | 4551 | ret = sync_file_range(req->file, req->sync.off, req->sync.len, |
8ed8d3c3 JA |
4552 | req->sync.flags); |
4553 | if (ret < 0) | |
4554 | req_set_fail_links(req); | |
e1e16097 | 4555 | io_req_complete(req, ret); |
5d17b4a4 JA |
4556 | return 0; |
4557 | } | |
4558 | ||
469956e8 | 4559 | #if defined(CONFIG_NET) |
02d27d89 PB |
4560 | static int io_setup_async_msg(struct io_kiocb *req, |
4561 | struct io_async_msghdr *kmsg) | |
4562 | { | |
e8c2bc1f JA |
4563 | struct io_async_msghdr *async_msg = req->async_data; |
4564 | ||
4565 | if (async_msg) | |
02d27d89 | 4566 | return -EAGAIN; |
e8c2bc1f | 4567 | if (io_alloc_async_data(req)) { |
02d27d89 PB |
4568 | if (kmsg->iov != kmsg->fast_iov) |
4569 | kfree(kmsg->iov); | |
4570 | return -ENOMEM; | |
4571 | } | |
e8c2bc1f | 4572 | async_msg = req->async_data; |
02d27d89 | 4573 | req->flags |= REQ_F_NEED_CLEANUP; |
e8c2bc1f | 4574 | memcpy(async_msg, kmsg, sizeof(*kmsg)); |
02d27d89 PB |
4575 | return -EAGAIN; |
4576 | } | |
4577 | ||
2ae523ed PB |
4578 | static int io_sendmsg_copy_hdr(struct io_kiocb *req, |
4579 | struct io_async_msghdr *iomsg) | |
4580 | { | |
4581 | iomsg->iov = iomsg->fast_iov; | |
4582 | iomsg->msg.msg_name = &iomsg->addr; | |
4583 | return sendmsg_copy_msghdr(&iomsg->msg, req->sr_msg.umsg, | |
4584 | req->sr_msg.msg_flags, &iomsg->iov); | |
4585 | } | |
4586 | ||
3529d8c2 | 4587 | static int io_sendmsg_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
03b1230c | 4588 | { |
e8c2bc1f | 4589 | struct io_async_msghdr *async_msg = req->async_data; |
e47293fd | 4590 | struct io_sr_msg *sr = &req->sr_msg; |
99bc4c38 | 4591 | int ret; |
03b1230c | 4592 | |
d2b6f48b PB |
4593 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4594 | return -EINVAL; | |
4595 | ||
e47293fd | 4596 | sr->msg_flags = READ_ONCE(sqe->msg_flags); |
270a5940 | 4597 | sr->umsg = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
fddaface | 4598 | sr->len = READ_ONCE(sqe->len); |
3529d8c2 | 4599 | |
d8768362 JA |
4600 | #ifdef CONFIG_COMPAT |
4601 | if (req->ctx->compat) | |
4602 | sr->msg_flags |= MSG_CMSG_COMPAT; | |
4603 | #endif | |
4604 | ||
e8c2bc1f | 4605 | if (!async_msg || !io_op_defs[req->opcode].needs_async_data) |
3529d8c2 | 4606 | return 0; |
e8c2bc1f | 4607 | ret = io_sendmsg_copy_hdr(req, async_msg); |
99bc4c38 PB |
4608 | if (!ret) |
4609 | req->flags |= REQ_F_NEED_CLEANUP; | |
4610 | return ret; | |
03b1230c JA |
4611 | } |
4612 | ||
229a7b63 JA |
4613 | static int io_sendmsg(struct io_kiocb *req, bool force_nonblock, |
4614 | struct io_comp_state *cs) | |
aa1fa28f | 4615 | { |
6b754c8b | 4616 | struct io_async_msghdr iomsg, *kmsg; |
0fa03c62 | 4617 | struct socket *sock; |
7a7cacba | 4618 | unsigned flags; |
0fa03c62 JA |
4619 | int ret; |
4620 | ||
dba4a925 | 4621 | sock = sock_from_file(req->file); |
7a7cacba | 4622 | if (unlikely(!sock)) |
dba4a925 | 4623 | return -ENOTSOCK; |
3529d8c2 | 4624 | |
e8c2bc1f JA |
4625 | if (req->async_data) { |
4626 | kmsg = req->async_data; | |
4627 | kmsg->msg.msg_name = &kmsg->addr; | |
7a7cacba PB |
4628 | /* if iov is set, it's allocated already */ |
4629 | if (!kmsg->iov) | |
4630 | kmsg->iov = kmsg->fast_iov; | |
4631 | kmsg->msg.msg_iter.iov = kmsg->iov; | |
4632 | } else { | |
4633 | ret = io_sendmsg_copy_hdr(req, &iomsg); | |
4634 | if (ret) | |
4635 | return ret; | |
4636 | kmsg = &iomsg; | |
0fa03c62 | 4637 | } |
0fa03c62 | 4638 | |
7a7cacba PB |
4639 | flags = req->sr_msg.msg_flags; |
4640 | if (flags & MSG_DONTWAIT) | |
4641 | req->flags |= REQ_F_NOWAIT; | |
4642 | else if (force_nonblock) | |
4643 | flags |= MSG_DONTWAIT; | |
e47293fd | 4644 | |
7a7cacba PB |
4645 | ret = __sys_sendmsg_sock(sock, &kmsg->msg, flags); |
4646 | if (force_nonblock && ret == -EAGAIN) | |
4647 | return io_setup_async_msg(req, kmsg); | |
4648 | if (ret == -ERESTARTSYS) | |
4649 | ret = -EINTR; | |
0fa03c62 | 4650 | |
6b754c8b | 4651 | if (kmsg->iov != kmsg->fast_iov) |
0b416c3e | 4652 | kfree(kmsg->iov); |
99bc4c38 | 4653 | req->flags &= ~REQ_F_NEED_CLEANUP; |
4e88d6e7 JA |
4654 | if (ret < 0) |
4655 | req_set_fail_links(req); | |
229a7b63 | 4656 | __io_req_complete(req, ret, 0, cs); |
5d17b4a4 | 4657 | return 0; |
03b1230c | 4658 | } |
aa1fa28f | 4659 | |
229a7b63 JA |
4660 | static int io_send(struct io_kiocb *req, bool force_nonblock, |
4661 | struct io_comp_state *cs) | |
fddaface | 4662 | { |
7a7cacba PB |
4663 | struct io_sr_msg *sr = &req->sr_msg; |
4664 | struct msghdr msg; | |
4665 | struct iovec iov; | |
fddaface | 4666 | struct socket *sock; |
7a7cacba | 4667 | unsigned flags; |
fddaface JA |
4668 | int ret; |
4669 | ||
dba4a925 | 4670 | sock = sock_from_file(req->file); |
7a7cacba | 4671 | if (unlikely(!sock)) |
dba4a925 | 4672 | return -ENOTSOCK; |
fddaface | 4673 | |
7a7cacba PB |
4674 | ret = import_single_range(WRITE, sr->buf, sr->len, &iov, &msg.msg_iter); |
4675 | if (unlikely(ret)) | |
14db8411 | 4676 | return ret; |
fddaface | 4677 | |
7a7cacba PB |
4678 | msg.msg_name = NULL; |
4679 | msg.msg_control = NULL; | |
4680 | msg.msg_controllen = 0; | |
4681 | msg.msg_namelen = 0; | |
fddaface | 4682 | |
7a7cacba PB |
4683 | flags = req->sr_msg.msg_flags; |
4684 | if (flags & MSG_DONTWAIT) | |
4685 | req->flags |= REQ_F_NOWAIT; | |
4686 | else if (force_nonblock) | |
4687 | flags |= MSG_DONTWAIT; | |
fddaface | 4688 | |
7a7cacba PB |
4689 | msg.msg_flags = flags; |
4690 | ret = sock_sendmsg(sock, &msg); | |
4691 | if (force_nonblock && ret == -EAGAIN) | |
4692 | return -EAGAIN; | |
4693 | if (ret == -ERESTARTSYS) | |
4694 | ret = -EINTR; | |
fddaface | 4695 | |
fddaface JA |
4696 | if (ret < 0) |
4697 | req_set_fail_links(req); | |
229a7b63 | 4698 | __io_req_complete(req, ret, 0, cs); |
fddaface | 4699 | return 0; |
fddaface JA |
4700 | } |
4701 | ||
1400e697 PB |
4702 | static int __io_recvmsg_copy_hdr(struct io_kiocb *req, |
4703 | struct io_async_msghdr *iomsg) | |
52de1fe1 JA |
4704 | { |
4705 | struct io_sr_msg *sr = &req->sr_msg; | |
4706 | struct iovec __user *uiov; | |
4707 | size_t iov_len; | |
4708 | int ret; | |
4709 | ||
1400e697 PB |
4710 | ret = __copy_msghdr_from_user(&iomsg->msg, sr->umsg, |
4711 | &iomsg->uaddr, &uiov, &iov_len); | |
52de1fe1 JA |
4712 | if (ret) |
4713 | return ret; | |
4714 | ||
4715 | if (req->flags & REQ_F_BUFFER_SELECT) { | |
4716 | if (iov_len > 1) | |
4717 | return -EINVAL; | |
1400e697 | 4718 | if (copy_from_user(iomsg->iov, uiov, sizeof(*uiov))) |
52de1fe1 | 4719 | return -EFAULT; |
1400e697 PB |
4720 | sr->len = iomsg->iov[0].iov_len; |
4721 | iov_iter_init(&iomsg->msg.msg_iter, READ, iomsg->iov, 1, | |
52de1fe1 | 4722 | sr->len); |
1400e697 | 4723 | iomsg->iov = NULL; |
52de1fe1 | 4724 | } else { |
89cd35c5 CH |
4725 | ret = __import_iovec(READ, uiov, iov_len, UIO_FASTIOV, |
4726 | &iomsg->iov, &iomsg->msg.msg_iter, | |
4727 | false); | |
52de1fe1 JA |
4728 | if (ret > 0) |
4729 | ret = 0; | |
4730 | } | |
4731 | ||
4732 | return ret; | |
4733 | } | |
4734 | ||
4735 | #ifdef CONFIG_COMPAT | |
4736 | static int __io_compat_recvmsg_copy_hdr(struct io_kiocb *req, | |
1400e697 | 4737 | struct io_async_msghdr *iomsg) |
52de1fe1 JA |
4738 | { |
4739 | struct compat_msghdr __user *msg_compat; | |
4740 | struct io_sr_msg *sr = &req->sr_msg; | |
4741 | struct compat_iovec __user *uiov; | |
4742 | compat_uptr_t ptr; | |
4743 | compat_size_t len; | |
4744 | int ret; | |
4745 | ||
270a5940 | 4746 | msg_compat = (struct compat_msghdr __user *) sr->umsg; |
1400e697 | 4747 | ret = __get_compat_msghdr(&iomsg->msg, msg_compat, &iomsg->uaddr, |
52de1fe1 JA |
4748 | &ptr, &len); |
4749 | if (ret) | |
4750 | return ret; | |
4751 | ||
4752 | uiov = compat_ptr(ptr); | |
4753 | if (req->flags & REQ_F_BUFFER_SELECT) { | |
4754 | compat_ssize_t clen; | |
4755 | ||
4756 | if (len > 1) | |
4757 | return -EINVAL; | |
4758 | if (!access_ok(uiov, sizeof(*uiov))) | |
4759 | return -EFAULT; | |
4760 | if (__get_user(clen, &uiov->iov_len)) | |
4761 | return -EFAULT; | |
4762 | if (clen < 0) | |
4763 | return -EINVAL; | |
2d280bc8 PB |
4764 | sr->len = clen; |
4765 | iomsg->iov[0].iov_len = clen; | |
1400e697 | 4766 | iomsg->iov = NULL; |
52de1fe1 | 4767 | } else { |
89cd35c5 CH |
4768 | ret = __import_iovec(READ, (struct iovec __user *)uiov, len, |
4769 | UIO_FASTIOV, &iomsg->iov, | |
4770 | &iomsg->msg.msg_iter, true); | |
52de1fe1 JA |
4771 | if (ret < 0) |
4772 | return ret; | |
4773 | } | |
4774 | ||
4775 | return 0; | |
4776 | } | |
4777 | #endif | |
4778 | ||
1400e697 PB |
4779 | static int io_recvmsg_copy_hdr(struct io_kiocb *req, |
4780 | struct io_async_msghdr *iomsg) | |
52de1fe1 | 4781 | { |
1400e697 PB |
4782 | iomsg->msg.msg_name = &iomsg->addr; |
4783 | iomsg->iov = iomsg->fast_iov; | |
52de1fe1 JA |
4784 | |
4785 | #ifdef CONFIG_COMPAT | |
4786 | if (req->ctx->compat) | |
1400e697 | 4787 | return __io_compat_recvmsg_copy_hdr(req, iomsg); |
fddaface | 4788 | #endif |
52de1fe1 | 4789 | |
1400e697 | 4790 | return __io_recvmsg_copy_hdr(req, iomsg); |
52de1fe1 JA |
4791 | } |
4792 | ||
bcda7baa | 4793 | static struct io_buffer *io_recv_buffer_select(struct io_kiocb *req, |
7fbb1b54 | 4794 | bool needs_lock) |
bcda7baa JA |
4795 | { |
4796 | struct io_sr_msg *sr = &req->sr_msg; | |
4797 | struct io_buffer *kbuf; | |
4798 | ||
bcda7baa JA |
4799 | kbuf = io_buffer_select(req, &sr->len, sr->bgid, sr->kbuf, needs_lock); |
4800 | if (IS_ERR(kbuf)) | |
4801 | return kbuf; | |
4802 | ||
4803 | sr->kbuf = kbuf; | |
4804 | req->flags |= REQ_F_BUFFER_SELECTED; | |
bcda7baa | 4805 | return kbuf; |
fddaface JA |
4806 | } |
4807 | ||
7fbb1b54 PB |
4808 | static inline unsigned int io_put_recv_kbuf(struct io_kiocb *req) |
4809 | { | |
4810 | return io_put_kbuf(req, req->sr_msg.kbuf); | |
4811 | } | |
4812 | ||
3529d8c2 JA |
4813 | static int io_recvmsg_prep(struct io_kiocb *req, |
4814 | const struct io_uring_sqe *sqe) | |
aa1fa28f | 4815 | { |
e8c2bc1f | 4816 | struct io_async_msghdr *async_msg = req->async_data; |
e47293fd | 4817 | struct io_sr_msg *sr = &req->sr_msg; |
99bc4c38 | 4818 | int ret; |
3529d8c2 | 4819 | |
d2b6f48b PB |
4820 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4821 | return -EINVAL; | |
4822 | ||
3529d8c2 | 4823 | sr->msg_flags = READ_ONCE(sqe->msg_flags); |
270a5940 | 4824 | sr->umsg = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
0b7b21e4 | 4825 | sr->len = READ_ONCE(sqe->len); |
bcda7baa | 4826 | sr->bgid = READ_ONCE(sqe->buf_group); |
06b76d44 | 4827 | |
d8768362 JA |
4828 | #ifdef CONFIG_COMPAT |
4829 | if (req->ctx->compat) | |
4830 | sr->msg_flags |= MSG_CMSG_COMPAT; | |
4831 | #endif | |
4832 | ||
e8c2bc1f | 4833 | if (!async_msg || !io_op_defs[req->opcode].needs_async_data) |
5f798bea | 4834 | return 0; |
e8c2bc1f | 4835 | ret = io_recvmsg_copy_hdr(req, async_msg); |
99bc4c38 PB |
4836 | if (!ret) |
4837 | req->flags |= REQ_F_NEED_CLEANUP; | |
4838 | return ret; | |
aa1fa28f JA |
4839 | } |
4840 | ||
229a7b63 JA |
4841 | static int io_recvmsg(struct io_kiocb *req, bool force_nonblock, |
4842 | struct io_comp_state *cs) | |
aa1fa28f | 4843 | { |
6b754c8b | 4844 | struct io_async_msghdr iomsg, *kmsg; |
03b1230c | 4845 | struct socket *sock; |
7fbb1b54 | 4846 | struct io_buffer *kbuf; |
7a7cacba | 4847 | unsigned flags; |
52de1fe1 | 4848 | int ret, cflags = 0; |
03b1230c | 4849 | |
dba4a925 | 4850 | sock = sock_from_file(req->file); |
7a7cacba | 4851 | if (unlikely(!sock)) |
dba4a925 | 4852 | return -ENOTSOCK; |
3529d8c2 | 4853 | |
e8c2bc1f JA |
4854 | if (req->async_data) { |
4855 | kmsg = req->async_data; | |
4856 | kmsg->msg.msg_name = &kmsg->addr; | |
7a7cacba PB |
4857 | /* if iov is set, it's allocated already */ |
4858 | if (!kmsg->iov) | |
4859 | kmsg->iov = kmsg->fast_iov; | |
4860 | kmsg->msg.msg_iter.iov = kmsg->iov; | |
4861 | } else { | |
4862 | ret = io_recvmsg_copy_hdr(req, &iomsg); | |
4863 | if (ret) | |
681fda8d | 4864 | return ret; |
7a7cacba PB |
4865 | kmsg = &iomsg; |
4866 | } | |
03b1230c | 4867 | |
bc02ef33 | 4868 | if (req->flags & REQ_F_BUFFER_SELECT) { |
7fbb1b54 | 4869 | kbuf = io_recv_buffer_select(req, !force_nonblock); |
bc02ef33 | 4870 | if (IS_ERR(kbuf)) |
52de1fe1 | 4871 | return PTR_ERR(kbuf); |
7a7cacba PB |
4872 | kmsg->fast_iov[0].iov_base = u64_to_user_ptr(kbuf->addr); |
4873 | iov_iter_init(&kmsg->msg.msg_iter, READ, kmsg->iov, | |
4874 | 1, req->sr_msg.len); | |
4875 | } | |
52de1fe1 | 4876 | |
7a7cacba PB |
4877 | flags = req->sr_msg.msg_flags; |
4878 | if (flags & MSG_DONTWAIT) | |
4879 | req->flags |= REQ_F_NOWAIT; | |
4880 | else if (force_nonblock) | |
4881 | flags |= MSG_DONTWAIT; | |
e47293fd | 4882 | |
7a7cacba PB |
4883 | ret = __sys_recvmsg_sock(sock, &kmsg->msg, req->sr_msg.umsg, |
4884 | kmsg->uaddr, flags); | |
0e1b6fe3 PB |
4885 | if (force_nonblock && ret == -EAGAIN) |
4886 | return io_setup_async_msg(req, kmsg); | |
7a7cacba PB |
4887 | if (ret == -ERESTARTSYS) |
4888 | ret = -EINTR; | |
03b1230c | 4889 | |
7fbb1b54 PB |
4890 | if (req->flags & REQ_F_BUFFER_SELECTED) |
4891 | cflags = io_put_recv_kbuf(req); | |
6b754c8b | 4892 | if (kmsg->iov != kmsg->fast_iov) |
0b416c3e | 4893 | kfree(kmsg->iov); |
99bc4c38 | 4894 | req->flags &= ~REQ_F_NEED_CLEANUP; |
4e88d6e7 JA |
4895 | if (ret < 0) |
4896 | req_set_fail_links(req); | |
229a7b63 | 4897 | __io_req_complete(req, ret, cflags, cs); |
03b1230c | 4898 | return 0; |
0fa03c62 | 4899 | } |
5d17b4a4 | 4900 | |
229a7b63 JA |
4901 | static int io_recv(struct io_kiocb *req, bool force_nonblock, |
4902 | struct io_comp_state *cs) | |
fddaface | 4903 | { |
6b754c8b | 4904 | struct io_buffer *kbuf; |
7a7cacba PB |
4905 | struct io_sr_msg *sr = &req->sr_msg; |
4906 | struct msghdr msg; | |
4907 | void __user *buf = sr->buf; | |
fddaface | 4908 | struct socket *sock; |
7a7cacba PB |
4909 | struct iovec iov; |
4910 | unsigned flags; | |
bcda7baa | 4911 | int ret, cflags = 0; |
fddaface | 4912 | |
dba4a925 | 4913 | sock = sock_from_file(req->file); |
7a7cacba | 4914 | if (unlikely(!sock)) |
dba4a925 | 4915 | return -ENOTSOCK; |
fddaface | 4916 | |
bc02ef33 | 4917 | if (req->flags & REQ_F_BUFFER_SELECT) { |
7fbb1b54 | 4918 | kbuf = io_recv_buffer_select(req, !force_nonblock); |
bcda7baa JA |
4919 | if (IS_ERR(kbuf)) |
4920 | return PTR_ERR(kbuf); | |
7a7cacba | 4921 | buf = u64_to_user_ptr(kbuf->addr); |
bc02ef33 | 4922 | } |
bcda7baa | 4923 | |
7a7cacba | 4924 | ret = import_single_range(READ, buf, sr->len, &iov, &msg.msg_iter); |
14c32eee PB |
4925 | if (unlikely(ret)) |
4926 | goto out_free; | |
fddaface | 4927 | |
7a7cacba PB |
4928 | msg.msg_name = NULL; |
4929 | msg.msg_control = NULL; | |
4930 | msg.msg_controllen = 0; | |
4931 | msg.msg_namelen = 0; | |
4932 | msg.msg_iocb = NULL; | |
4933 | msg.msg_flags = 0; | |
fddaface | 4934 | |
7a7cacba PB |
4935 | flags = req->sr_msg.msg_flags; |
4936 | if (flags & MSG_DONTWAIT) | |
4937 | req->flags |= REQ_F_NOWAIT; | |
4938 | else if (force_nonblock) | |
4939 | flags |= MSG_DONTWAIT; | |
4940 | ||
4941 | ret = sock_recvmsg(sock, &msg, flags); | |
4942 | if (force_nonblock && ret == -EAGAIN) | |
4943 | return -EAGAIN; | |
4944 | if (ret == -ERESTARTSYS) | |
4945 | ret = -EINTR; | |
14c32eee | 4946 | out_free: |
7fbb1b54 PB |
4947 | if (req->flags & REQ_F_BUFFER_SELECTED) |
4948 | cflags = io_put_recv_kbuf(req); | |
fddaface JA |
4949 | if (ret < 0) |
4950 | req_set_fail_links(req); | |
229a7b63 | 4951 | __io_req_complete(req, ret, cflags, cs); |
fddaface | 4952 | return 0; |
fddaface JA |
4953 | } |
4954 | ||
3529d8c2 | 4955 | static int io_accept_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
17f2fe35 | 4956 | { |
8ed8d3c3 JA |
4957 | struct io_accept *accept = &req->accept; |
4958 | ||
14587a46 | 4959 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
17f2fe35 | 4960 | return -EINVAL; |
8042d6ce | 4961 | if (sqe->ioprio || sqe->len || sqe->buf_index) |
17f2fe35 JA |
4962 | return -EINVAL; |
4963 | ||
d55e5f5b JA |
4964 | accept->addr = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
4965 | accept->addr_len = u64_to_user_ptr(READ_ONCE(sqe->addr2)); | |
8ed8d3c3 | 4966 | accept->flags = READ_ONCE(sqe->accept_flags); |
09952e3e | 4967 | accept->nofile = rlimit(RLIMIT_NOFILE); |
8ed8d3c3 | 4968 | return 0; |
8ed8d3c3 | 4969 | } |
17f2fe35 | 4970 | |
229a7b63 JA |
4971 | static int io_accept(struct io_kiocb *req, bool force_nonblock, |
4972 | struct io_comp_state *cs) | |
8ed8d3c3 JA |
4973 | { |
4974 | struct io_accept *accept = &req->accept; | |
ac45abc0 | 4975 | unsigned int file_flags = force_nonblock ? O_NONBLOCK : 0; |
8ed8d3c3 JA |
4976 | int ret; |
4977 | ||
e697deed JX |
4978 | if (req->file->f_flags & O_NONBLOCK) |
4979 | req->flags |= REQ_F_NOWAIT; | |
4980 | ||
8ed8d3c3 | 4981 | ret = __sys_accept4_file(req->file, file_flags, accept->addr, |
09952e3e JA |
4982 | accept->addr_len, accept->flags, |
4983 | accept->nofile); | |
8ed8d3c3 | 4984 | if (ret == -EAGAIN && force_nonblock) |
17f2fe35 | 4985 | return -EAGAIN; |
ac45abc0 PB |
4986 | if (ret < 0) { |
4987 | if (ret == -ERESTARTSYS) | |
4988 | ret = -EINTR; | |
4e88d6e7 | 4989 | req_set_fail_links(req); |
ac45abc0 | 4990 | } |
229a7b63 | 4991 | __io_req_complete(req, ret, 0, cs); |
17f2fe35 | 4992 | return 0; |
8ed8d3c3 JA |
4993 | } |
4994 | ||
3529d8c2 | 4995 | static int io_connect_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
f499a021 | 4996 | { |
3529d8c2 | 4997 | struct io_connect *conn = &req->connect; |
e8c2bc1f | 4998 | struct io_async_connect *io = req->async_data; |
f499a021 | 4999 | |
14587a46 | 5000 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
3fbb51c1 JA |
5001 | return -EINVAL; |
5002 | if (sqe->ioprio || sqe->len || sqe->buf_index || sqe->rw_flags) | |
5003 | return -EINVAL; | |
5004 | ||
3529d8c2 JA |
5005 | conn->addr = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
5006 | conn->addr_len = READ_ONCE(sqe->addr2); | |
5007 | ||
5008 | if (!io) | |
5009 | return 0; | |
5010 | ||
5011 | return move_addr_to_kernel(conn->addr, conn->addr_len, | |
e8c2bc1f | 5012 | &io->address); |
f499a021 JA |
5013 | } |
5014 | ||
229a7b63 JA |
5015 | static int io_connect(struct io_kiocb *req, bool force_nonblock, |
5016 | struct io_comp_state *cs) | |
f8e85cf2 | 5017 | { |
e8c2bc1f | 5018 | struct io_async_connect __io, *io; |
f8e85cf2 | 5019 | unsigned file_flags; |
3fbb51c1 | 5020 | int ret; |
f8e85cf2 | 5021 | |
e8c2bc1f JA |
5022 | if (req->async_data) { |
5023 | io = req->async_data; | |
f499a021 | 5024 | } else { |
3529d8c2 JA |
5025 | ret = move_addr_to_kernel(req->connect.addr, |
5026 | req->connect.addr_len, | |
e8c2bc1f | 5027 | &__io.address); |
f499a021 JA |
5028 | if (ret) |
5029 | goto out; | |
5030 | io = &__io; | |
5031 | } | |
5032 | ||
3fbb51c1 JA |
5033 | file_flags = force_nonblock ? O_NONBLOCK : 0; |
5034 | ||
e8c2bc1f | 5035 | ret = __sys_connect_file(req->file, &io->address, |
3fbb51c1 | 5036 | req->connect.addr_len, file_flags); |
87f80d62 | 5037 | if ((ret == -EAGAIN || ret == -EINPROGRESS) && force_nonblock) { |
e8c2bc1f | 5038 | if (req->async_data) |
b7bb4f7d | 5039 | return -EAGAIN; |
e8c2bc1f | 5040 | if (io_alloc_async_data(req)) { |
f499a021 JA |
5041 | ret = -ENOMEM; |
5042 | goto out; | |
5043 | } | |
e8c2bc1f JA |
5044 | io = req->async_data; |
5045 | memcpy(req->async_data, &__io, sizeof(__io)); | |
f8e85cf2 | 5046 | return -EAGAIN; |
f499a021 | 5047 | } |
f8e85cf2 JA |
5048 | if (ret == -ERESTARTSYS) |
5049 | ret = -EINTR; | |
f499a021 | 5050 | out: |
4e88d6e7 JA |
5051 | if (ret < 0) |
5052 | req_set_fail_links(req); | |
229a7b63 | 5053 | __io_req_complete(req, ret, 0, cs); |
f8e85cf2 | 5054 | return 0; |
469956e8 Y |
5055 | } |
5056 | #else /* !CONFIG_NET */ | |
5057 | static int io_sendmsg_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) | |
5058 | { | |
f8e85cf2 | 5059 | return -EOPNOTSUPP; |
f8e85cf2 JA |
5060 | } |
5061 | ||
1e16c2f9 RD |
5062 | static int io_sendmsg(struct io_kiocb *req, bool force_nonblock, |
5063 | struct io_comp_state *cs) | |
469956e8 Y |
5064 | { |
5065 | return -EOPNOTSUPP; | |
5066 | } | |
5067 | ||
1e16c2f9 RD |
5068 | static int io_send(struct io_kiocb *req, bool force_nonblock, |
5069 | struct io_comp_state *cs) | |
469956e8 Y |
5070 | { |
5071 | return -EOPNOTSUPP; | |
5072 | } | |
5073 | ||
5074 | static int io_recvmsg_prep(struct io_kiocb *req, | |
5075 | const struct io_uring_sqe *sqe) | |
5076 | { | |
5077 | return -EOPNOTSUPP; | |
5078 | } | |
5079 | ||
1e16c2f9 RD |
5080 | static int io_recvmsg(struct io_kiocb *req, bool force_nonblock, |
5081 | struct io_comp_state *cs) | |
469956e8 Y |
5082 | { |
5083 | return -EOPNOTSUPP; | |
5084 | } | |
5085 | ||
1e16c2f9 RD |
5086 | static int io_recv(struct io_kiocb *req, bool force_nonblock, |
5087 | struct io_comp_state *cs) | |
469956e8 Y |
5088 | { |
5089 | return -EOPNOTSUPP; | |
5090 | } | |
5091 | ||
5092 | static int io_accept_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) | |
5093 | { | |
5094 | return -EOPNOTSUPP; | |
5095 | } | |
5096 | ||
1e16c2f9 RD |
5097 | static int io_accept(struct io_kiocb *req, bool force_nonblock, |
5098 | struct io_comp_state *cs) | |
469956e8 Y |
5099 | { |
5100 | return -EOPNOTSUPP; | |
5101 | } | |
ce593a6c | 5102 | |
469956e8 Y |
5103 | static int io_connect_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
5104 | { | |
5105 | return -EOPNOTSUPP; | |
5106 | } | |
5107 | ||
1e16c2f9 RD |
5108 | static int io_connect(struct io_kiocb *req, bool force_nonblock, |
5109 | struct io_comp_state *cs) | |
469956e8 | 5110 | { |
f8e85cf2 | 5111 | return -EOPNOTSUPP; |
ce593a6c | 5112 | } |
469956e8 | 5113 | #endif /* CONFIG_NET */ |
f8e85cf2 | 5114 | |
d7718a9d JA |
5115 | struct io_poll_table { |
5116 | struct poll_table_struct pt; | |
5117 | struct io_kiocb *req; | |
5118 | int error; | |
5119 | }; | |
ce593a6c | 5120 | |
d7718a9d JA |
5121 | static int __io_async_wake(struct io_kiocb *req, struct io_poll_iocb *poll, |
5122 | __poll_t mask, task_work_func_t func) | |
5123 | { | |
aa96bf8a | 5124 | int ret; |
d7718a9d JA |
5125 | |
5126 | /* for instances that support it check for an event match first: */ | |
5127 | if (mask && !(mask & poll->events)) | |
5128 | return 0; | |
5129 | ||
5130 | trace_io_uring_task_add(req->ctx, req->opcode, req->user_data, mask); | |
5131 | ||
5132 | list_del_init(&poll->wait.entry); | |
5133 | ||
d7718a9d JA |
5134 | req->result = mask; |
5135 | init_task_work(&req->task_work, func); | |
6d816e08 JA |
5136 | percpu_ref_get(&req->ctx->refs); |
5137 | ||
d7718a9d | 5138 | /* |
e3aabf95 JA |
5139 | * If this fails, then the task is exiting. When a task exits, the |
5140 | * work gets canceled, so just cancel this request as well instead | |
5141 | * of executing it. We can't safely execute it anyway, as we may not | |
5142 | * have the needed state needed for it anyway. | |
d7718a9d | 5143 | */ |
355fb9e2 | 5144 | ret = io_req_task_work_add(req); |
aa96bf8a | 5145 | if (unlikely(ret)) { |
c2c4c83c JA |
5146 | struct task_struct *tsk; |
5147 | ||
e3aabf95 | 5148 | WRITE_ONCE(poll->canceled, true); |
aa96bf8a | 5149 | tsk = io_wq_get_task(req->ctx->io_wq); |
91989c70 | 5150 | task_work_add(tsk, &req->task_work, TWA_NONE); |
ce593a6c | 5151 | wake_up_process(tsk); |
aa96bf8a | 5152 | } |
d7718a9d JA |
5153 | return 1; |
5154 | } | |
5155 | ||
74ce6ce4 JA |
5156 | static bool io_poll_rewait(struct io_kiocb *req, struct io_poll_iocb *poll) |
5157 | __acquires(&req->ctx->completion_lock) | |
5158 | { | |
5159 | struct io_ring_ctx *ctx = req->ctx; | |
5160 | ||
5161 | if (!req->result && !READ_ONCE(poll->canceled)) { | |
5162 | struct poll_table_struct pt = { ._key = poll->events }; | |
5163 | ||
5164 | req->result = vfs_poll(req->file, &pt) & poll->events; | |
5165 | } | |
5166 | ||
5167 | spin_lock_irq(&ctx->completion_lock); | |
5168 | if (!req->result && !READ_ONCE(poll->canceled)) { | |
5169 | add_wait_queue(poll->head, &poll->wait); | |
5170 | return true; | |
5171 | } | |
5172 | ||
5173 | return false; | |
5174 | } | |
5175 | ||
d4e7cd36 | 5176 | static struct io_poll_iocb *io_poll_get_double(struct io_kiocb *req) |
18bceab1 | 5177 | { |
e8c2bc1f | 5178 | /* pure poll stashes this in ->async_data, poll driven retry elsewhere */ |
d4e7cd36 | 5179 | if (req->opcode == IORING_OP_POLL_ADD) |
e8c2bc1f | 5180 | return req->async_data; |
d4e7cd36 JA |
5181 | return req->apoll->double_poll; |
5182 | } | |
5183 | ||
5184 | static struct io_poll_iocb *io_poll_get_single(struct io_kiocb *req) | |
5185 | { | |
5186 | if (req->opcode == IORING_OP_POLL_ADD) | |
5187 | return &req->poll; | |
5188 | return &req->apoll->poll; | |
5189 | } | |
5190 | ||
5191 | static void io_poll_remove_double(struct io_kiocb *req) | |
5192 | { | |
5193 | struct io_poll_iocb *poll = io_poll_get_double(req); | |
18bceab1 JA |
5194 | |
5195 | lockdep_assert_held(&req->ctx->completion_lock); | |
5196 | ||
5197 | if (poll && poll->head) { | |
5198 | struct wait_queue_head *head = poll->head; | |
5199 | ||
5200 | spin_lock(&head->lock); | |
5201 | list_del_init(&poll->wait.entry); | |
5202 | if (poll->wait.private) | |
5203 | refcount_dec(&req->refs); | |
5204 | poll->head = NULL; | |
5205 | spin_unlock(&head->lock); | |
5206 | } | |
5207 | } | |
5208 | ||
5209 | static void io_poll_complete(struct io_kiocb *req, __poll_t mask, int error) | |
5210 | { | |
5211 | struct io_ring_ctx *ctx = req->ctx; | |
5212 | ||
d4e7cd36 | 5213 | io_poll_remove_double(req); |
18bceab1 JA |
5214 | req->poll.done = true; |
5215 | io_cqring_fill_event(req, error ? error : mangle_poll(mask)); | |
5216 | io_commit_cqring(ctx); | |
5217 | } | |
5218 | ||
dd221f46 | 5219 | static void io_poll_task_func(struct callback_head *cb) |
18bceab1 | 5220 | { |
dd221f46 | 5221 | struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work); |
18bceab1 | 5222 | struct io_ring_ctx *ctx = req->ctx; |
dd221f46 | 5223 | struct io_kiocb *nxt; |
18bceab1 JA |
5224 | |
5225 | if (io_poll_rewait(req, &req->poll)) { | |
5226 | spin_unlock_irq(&ctx->completion_lock); | |
dd221f46 PB |
5227 | } else { |
5228 | hash_del(&req->hash_node); | |
5229 | io_poll_complete(req, req->result, 0); | |
5230 | spin_unlock_irq(&ctx->completion_lock); | |
18bceab1 | 5231 | |
dd221f46 PB |
5232 | nxt = io_put_req_find_next(req); |
5233 | io_cqring_ev_posted(ctx); | |
5234 | if (nxt) | |
5235 | __io_req_task_submit(nxt); | |
5236 | } | |
18bceab1 | 5237 | |
6d816e08 | 5238 | percpu_ref_put(&ctx->refs); |
18bceab1 JA |
5239 | } |
5240 | ||
5241 | static int io_poll_double_wake(struct wait_queue_entry *wait, unsigned mode, | |
5242 | int sync, void *key) | |
5243 | { | |
5244 | struct io_kiocb *req = wait->private; | |
d4e7cd36 | 5245 | struct io_poll_iocb *poll = io_poll_get_single(req); |
18bceab1 JA |
5246 | __poll_t mask = key_to_poll(key); |
5247 | ||
5248 | /* for instances that support it check for an event match first: */ | |
5249 | if (mask && !(mask & poll->events)) | |
5250 | return 0; | |
5251 | ||
8706e04e JA |
5252 | list_del_init(&wait->entry); |
5253 | ||
807abcb0 | 5254 | if (poll && poll->head) { |
18bceab1 JA |
5255 | bool done; |
5256 | ||
807abcb0 JA |
5257 | spin_lock(&poll->head->lock); |
5258 | done = list_empty(&poll->wait.entry); | |
18bceab1 | 5259 | if (!done) |
807abcb0 | 5260 | list_del_init(&poll->wait.entry); |
d4e7cd36 JA |
5261 | /* make sure double remove sees this as being gone */ |
5262 | wait->private = NULL; | |
807abcb0 | 5263 | spin_unlock(&poll->head->lock); |
c8b5e260 JA |
5264 | if (!done) { |
5265 | /* use wait func handler, so it matches the rq type */ | |
5266 | poll->wait.func(&poll->wait, mode, sync, key); | |
5267 | } | |
18bceab1 JA |
5268 | } |
5269 | refcount_dec(&req->refs); | |
5270 | return 1; | |
5271 | } | |
5272 | ||
5273 | static void io_init_poll_iocb(struct io_poll_iocb *poll, __poll_t events, | |
5274 | wait_queue_func_t wake_func) | |
5275 | { | |
5276 | poll->head = NULL; | |
5277 | poll->done = false; | |
5278 | poll->canceled = false; | |
5279 | poll->events = events; | |
5280 | INIT_LIST_HEAD(&poll->wait.entry); | |
5281 | init_waitqueue_func_entry(&poll->wait, wake_func); | |
5282 | } | |
5283 | ||
5284 | static void __io_queue_proc(struct io_poll_iocb *poll, struct io_poll_table *pt, | |
807abcb0 JA |
5285 | struct wait_queue_head *head, |
5286 | struct io_poll_iocb **poll_ptr) | |
18bceab1 JA |
5287 | { |
5288 | struct io_kiocb *req = pt->req; | |
5289 | ||
5290 | /* | |
5291 | * If poll->head is already set, it's because the file being polled | |
5292 | * uses multiple waitqueues for poll handling (eg one for read, one | |
5293 | * for write). Setup a separate io_poll_iocb if this happens. | |
5294 | */ | |
5295 | if (unlikely(poll->head)) { | |
58852d4d PB |
5296 | struct io_poll_iocb *poll_one = poll; |
5297 | ||
18bceab1 | 5298 | /* already have a 2nd entry, fail a third attempt */ |
807abcb0 | 5299 | if (*poll_ptr) { |
18bceab1 JA |
5300 | pt->error = -EINVAL; |
5301 | return; | |
5302 | } | |
5303 | poll = kmalloc(sizeof(*poll), GFP_ATOMIC); | |
5304 | if (!poll) { | |
5305 | pt->error = -ENOMEM; | |
5306 | return; | |
5307 | } | |
58852d4d | 5308 | io_init_poll_iocb(poll, poll_one->events, io_poll_double_wake); |
18bceab1 JA |
5309 | refcount_inc(&req->refs); |
5310 | poll->wait.private = req; | |
807abcb0 | 5311 | *poll_ptr = poll; |
18bceab1 JA |
5312 | } |
5313 | ||
5314 | pt->error = 0; | |
5315 | poll->head = head; | |
a31eb4a2 JX |
5316 | |
5317 | if (poll->events & EPOLLEXCLUSIVE) | |
5318 | add_wait_queue_exclusive(head, &poll->wait); | |
5319 | else | |
5320 | add_wait_queue(head, &poll->wait); | |
18bceab1 JA |
5321 | } |
5322 | ||
5323 | static void io_async_queue_proc(struct file *file, struct wait_queue_head *head, | |
5324 | struct poll_table_struct *p) | |
5325 | { | |
5326 | struct io_poll_table *pt = container_of(p, struct io_poll_table, pt); | |
807abcb0 | 5327 | struct async_poll *apoll = pt->req->apoll; |
18bceab1 | 5328 | |
807abcb0 | 5329 | __io_queue_proc(&apoll->poll, pt, head, &apoll->double_poll); |
18bceab1 JA |
5330 | } |
5331 | ||
d7718a9d JA |
5332 | static void io_async_task_func(struct callback_head *cb) |
5333 | { | |
5334 | struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work); | |
5335 | struct async_poll *apoll = req->apoll; | |
5336 | struct io_ring_ctx *ctx = req->ctx; | |
5337 | ||
5338 | trace_io_uring_task_run(req->ctx, req->opcode, req->user_data); | |
5339 | ||
74ce6ce4 | 5340 | if (io_poll_rewait(req, &apoll->poll)) { |
d7718a9d | 5341 | spin_unlock_irq(&ctx->completion_lock); |
6d816e08 | 5342 | percpu_ref_put(&ctx->refs); |
74ce6ce4 | 5343 | return; |
d7718a9d JA |
5344 | } |
5345 | ||
31067255 | 5346 | /* If req is still hashed, it cannot have been canceled. Don't check. */ |
0be0b0e3 | 5347 | if (hash_hashed(&req->hash_node)) |
74ce6ce4 | 5348 | hash_del(&req->hash_node); |
2bae047e | 5349 | |
d4e7cd36 | 5350 | io_poll_remove_double(req); |
74ce6ce4 JA |
5351 | spin_unlock_irq(&ctx->completion_lock); |
5352 | ||
0be0b0e3 PB |
5353 | if (!READ_ONCE(apoll->poll.canceled)) |
5354 | __io_req_task_submit(req); | |
5355 | else | |
5356 | __io_req_task_cancel(req, -ECANCELED); | |
aa340845 | 5357 | |
6d816e08 | 5358 | percpu_ref_put(&ctx->refs); |
807abcb0 | 5359 | kfree(apoll->double_poll); |
31067255 | 5360 | kfree(apoll); |
d7718a9d JA |
5361 | } |
5362 | ||
5363 | static int io_async_wake(struct wait_queue_entry *wait, unsigned mode, int sync, | |
5364 | void *key) | |
5365 | { | |
5366 | struct io_kiocb *req = wait->private; | |
5367 | struct io_poll_iocb *poll = &req->apoll->poll; | |
5368 | ||
5369 | trace_io_uring_poll_wake(req->ctx, req->opcode, req->user_data, | |
5370 | key_to_poll(key)); | |
5371 | ||
5372 | return __io_async_wake(req, poll, key_to_poll(key), io_async_task_func); | |
5373 | } | |
5374 | ||
5375 | static void io_poll_req_insert(struct io_kiocb *req) | |
5376 | { | |
5377 | struct io_ring_ctx *ctx = req->ctx; | |
5378 | struct hlist_head *list; | |
5379 | ||
5380 | list = &ctx->cancel_hash[hash_long(req->user_data, ctx->cancel_hash_bits)]; | |
5381 | hlist_add_head(&req->hash_node, list); | |
5382 | } | |
5383 | ||
5384 | static __poll_t __io_arm_poll_handler(struct io_kiocb *req, | |
5385 | struct io_poll_iocb *poll, | |
5386 | struct io_poll_table *ipt, __poll_t mask, | |
5387 | wait_queue_func_t wake_func) | |
5388 | __acquires(&ctx->completion_lock) | |
5389 | { | |
5390 | struct io_ring_ctx *ctx = req->ctx; | |
5391 | bool cancel = false; | |
5392 | ||
4d52f338 | 5393 | INIT_HLIST_NODE(&req->hash_node); |
18bceab1 | 5394 | io_init_poll_iocb(poll, mask, wake_func); |
b90cd197 | 5395 | poll->file = req->file; |
18bceab1 | 5396 | poll->wait.private = req; |
d7718a9d JA |
5397 | |
5398 | ipt->pt._key = mask; | |
5399 | ipt->req = req; | |
5400 | ipt->error = -EINVAL; | |
5401 | ||
d7718a9d JA |
5402 | mask = vfs_poll(req->file, &ipt->pt) & poll->events; |
5403 | ||
5404 | spin_lock_irq(&ctx->completion_lock); | |
5405 | if (likely(poll->head)) { | |
5406 | spin_lock(&poll->head->lock); | |
5407 | if (unlikely(list_empty(&poll->wait.entry))) { | |
5408 | if (ipt->error) | |
5409 | cancel = true; | |
5410 | ipt->error = 0; | |
5411 | mask = 0; | |
5412 | } | |
5413 | if (mask || ipt->error) | |
5414 | list_del_init(&poll->wait.entry); | |
5415 | else if (cancel) | |
5416 | WRITE_ONCE(poll->canceled, true); | |
5417 | else if (!poll->done) /* actually waiting for an event */ | |
5418 | io_poll_req_insert(req); | |
5419 | spin_unlock(&poll->head->lock); | |
5420 | } | |
5421 | ||
5422 | return mask; | |
5423 | } | |
5424 | ||
5425 | static bool io_arm_poll_handler(struct io_kiocb *req) | |
5426 | { | |
5427 | const struct io_op_def *def = &io_op_defs[req->opcode]; | |
5428 | struct io_ring_ctx *ctx = req->ctx; | |
5429 | struct async_poll *apoll; | |
5430 | struct io_poll_table ipt; | |
5431 | __poll_t mask, ret; | |
9dab14b8 | 5432 | int rw; |
d7718a9d JA |
5433 | |
5434 | if (!req->file || !file_can_poll(req->file)) | |
5435 | return false; | |
24c74678 | 5436 | if (req->flags & REQ_F_POLLED) |
d7718a9d | 5437 | return false; |
9dab14b8 JA |
5438 | if (def->pollin) |
5439 | rw = READ; | |
5440 | else if (def->pollout) | |
5441 | rw = WRITE; | |
5442 | else | |
5443 | return false; | |
5444 | /* if we can't nonblock try, then no point in arming a poll handler */ | |
5445 | if (!io_file_supports_async(req->file, rw)) | |
d7718a9d JA |
5446 | return false; |
5447 | ||
5448 | apoll = kmalloc(sizeof(*apoll), GFP_ATOMIC); | |
5449 | if (unlikely(!apoll)) | |
5450 | return false; | |
807abcb0 | 5451 | apoll->double_poll = NULL; |
d7718a9d JA |
5452 | |
5453 | req->flags |= REQ_F_POLLED; | |
d7718a9d | 5454 | req->apoll = apoll; |
d7718a9d | 5455 | |
8755d97a | 5456 | mask = 0; |
d7718a9d | 5457 | if (def->pollin) |
8755d97a | 5458 | mask |= POLLIN | POLLRDNORM; |
d7718a9d JA |
5459 | if (def->pollout) |
5460 | mask |= POLLOUT | POLLWRNORM; | |
901341bb LH |
5461 | |
5462 | /* If reading from MSG_ERRQUEUE using recvmsg, ignore POLLIN */ | |
5463 | if ((req->opcode == IORING_OP_RECVMSG) && | |
5464 | (req->sr_msg.msg_flags & MSG_ERRQUEUE)) | |
5465 | mask &= ~POLLIN; | |
5466 | ||
d7718a9d JA |
5467 | mask |= POLLERR | POLLPRI; |
5468 | ||
5469 | ipt.pt._qproc = io_async_queue_proc; | |
5470 | ||
5471 | ret = __io_arm_poll_handler(req, &apoll->poll, &ipt, mask, | |
5472 | io_async_wake); | |
a36da65c | 5473 | if (ret || ipt.error) { |
d4e7cd36 | 5474 | io_poll_remove_double(req); |
d7718a9d | 5475 | spin_unlock_irq(&ctx->completion_lock); |
807abcb0 | 5476 | kfree(apoll->double_poll); |
d7718a9d JA |
5477 | kfree(apoll); |
5478 | return false; | |
5479 | } | |
5480 | spin_unlock_irq(&ctx->completion_lock); | |
5481 | trace_io_uring_poll_arm(ctx, req->opcode, req->user_data, mask, | |
5482 | apoll->poll.events); | |
5483 | return true; | |
5484 | } | |
5485 | ||
5486 | static bool __io_poll_remove_one(struct io_kiocb *req, | |
5487 | struct io_poll_iocb *poll) | |
221c5eb2 | 5488 | { |
b41e9852 | 5489 | bool do_complete = false; |
221c5eb2 JA |
5490 | |
5491 | spin_lock(&poll->head->lock); | |
5492 | WRITE_ONCE(poll->canceled, true); | |
392edb45 JA |
5493 | if (!list_empty(&poll->wait.entry)) { |
5494 | list_del_init(&poll->wait.entry); | |
b41e9852 | 5495 | do_complete = true; |
221c5eb2 JA |
5496 | } |
5497 | spin_unlock(&poll->head->lock); | |
3bfa5bcb | 5498 | hash_del(&req->hash_node); |
d7718a9d JA |
5499 | return do_complete; |
5500 | } | |
5501 | ||
5502 | static bool io_poll_remove_one(struct io_kiocb *req) | |
5503 | { | |
5504 | bool do_complete; | |
5505 | ||
d4e7cd36 JA |
5506 | io_poll_remove_double(req); |
5507 | ||
d7718a9d JA |
5508 | if (req->opcode == IORING_OP_POLL_ADD) { |
5509 | do_complete = __io_poll_remove_one(req, &req->poll); | |
5510 | } else { | |
3bfa5bcb JA |
5511 | struct async_poll *apoll = req->apoll; |
5512 | ||
d7718a9d | 5513 | /* non-poll requests have submit ref still */ |
3bfa5bcb JA |
5514 | do_complete = __io_poll_remove_one(req, &apoll->poll); |
5515 | if (do_complete) { | |
d7718a9d | 5516 | io_put_req(req); |
807abcb0 | 5517 | kfree(apoll->double_poll); |
3bfa5bcb JA |
5518 | kfree(apoll); |
5519 | } | |
b1f573bd XW |
5520 | } |
5521 | ||
b41e9852 JA |
5522 | if (do_complete) { |
5523 | io_cqring_fill_event(req, -ECANCELED); | |
5524 | io_commit_cqring(req->ctx); | |
f254ac04 | 5525 | req_set_fail_links(req); |
216578e5 | 5526 | io_put_req_deferred(req, 1); |
b41e9852 JA |
5527 | } |
5528 | ||
5529 | return do_complete; | |
221c5eb2 JA |
5530 | } |
5531 | ||
76e1b642 JA |
5532 | /* |
5533 | * Returns true if we found and killed one or more poll requests | |
5534 | */ | |
6b81928d PB |
5535 | static bool io_poll_remove_all(struct io_ring_ctx *ctx, struct task_struct *tsk, |
5536 | struct files_struct *files) | |
221c5eb2 | 5537 | { |
78076bb6 | 5538 | struct hlist_node *tmp; |
221c5eb2 | 5539 | struct io_kiocb *req; |
8e2e1faf | 5540 | int posted = 0, i; |
221c5eb2 JA |
5541 | |
5542 | spin_lock_irq(&ctx->completion_lock); | |
78076bb6 JA |
5543 | for (i = 0; i < (1U << ctx->cancel_hash_bits); i++) { |
5544 | struct hlist_head *list; | |
5545 | ||
5546 | list = &ctx->cancel_hash[i]; | |
f3606e3a | 5547 | hlist_for_each_entry_safe(req, tmp, list, hash_node) { |
6b81928d | 5548 | if (io_match_task(req, tsk, files)) |
f3606e3a JA |
5549 | posted += io_poll_remove_one(req); |
5550 | } | |
221c5eb2 JA |
5551 | } |
5552 | spin_unlock_irq(&ctx->completion_lock); | |
b41e9852 | 5553 | |
8e2e1faf JA |
5554 | if (posted) |
5555 | io_cqring_ev_posted(ctx); | |
76e1b642 JA |
5556 | |
5557 | return posted != 0; | |
221c5eb2 JA |
5558 | } |
5559 | ||
47f46768 JA |
5560 | static int io_poll_cancel(struct io_ring_ctx *ctx, __u64 sqe_addr) |
5561 | { | |
78076bb6 | 5562 | struct hlist_head *list; |
47f46768 JA |
5563 | struct io_kiocb *req; |
5564 | ||
78076bb6 JA |
5565 | list = &ctx->cancel_hash[hash_long(sqe_addr, ctx->cancel_hash_bits)]; |
5566 | hlist_for_each_entry(req, list, hash_node) { | |
b41e9852 JA |
5567 | if (sqe_addr != req->user_data) |
5568 | continue; | |
5569 | if (io_poll_remove_one(req)) | |
eac406c6 | 5570 | return 0; |
b41e9852 | 5571 | return -EALREADY; |
47f46768 JA |
5572 | } |
5573 | ||
5574 | return -ENOENT; | |
5575 | } | |
5576 | ||
3529d8c2 JA |
5577 | static int io_poll_remove_prep(struct io_kiocb *req, |
5578 | const struct io_uring_sqe *sqe) | |
0969e783 | 5579 | { |
0969e783 JA |
5580 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
5581 | return -EINVAL; | |
5582 | if (sqe->ioprio || sqe->off || sqe->len || sqe->buf_index || | |
5583 | sqe->poll_events) | |
5584 | return -EINVAL; | |
5585 | ||
018043be | 5586 | req->poll_remove.addr = READ_ONCE(sqe->addr); |
0969e783 JA |
5587 | return 0; |
5588 | } | |
5589 | ||
221c5eb2 JA |
5590 | /* |
5591 | * Find a running poll command that matches one specified in sqe->addr, | |
5592 | * and remove it if found. | |
5593 | */ | |
fc4df999 | 5594 | static int io_poll_remove(struct io_kiocb *req) |
221c5eb2 JA |
5595 | { |
5596 | struct io_ring_ctx *ctx = req->ctx; | |
47f46768 | 5597 | int ret; |
221c5eb2 | 5598 | |
221c5eb2 | 5599 | spin_lock_irq(&ctx->completion_lock); |
018043be | 5600 | ret = io_poll_cancel(ctx, req->poll_remove.addr); |
221c5eb2 JA |
5601 | spin_unlock_irq(&ctx->completion_lock); |
5602 | ||
4e88d6e7 JA |
5603 | if (ret < 0) |
5604 | req_set_fail_links(req); | |
e1e16097 | 5605 | io_req_complete(req, ret); |
221c5eb2 JA |
5606 | return 0; |
5607 | } | |
5608 | ||
221c5eb2 JA |
5609 | static int io_poll_wake(struct wait_queue_entry *wait, unsigned mode, int sync, |
5610 | void *key) | |
5611 | { | |
c2f2eb7d JA |
5612 | struct io_kiocb *req = wait->private; |
5613 | struct io_poll_iocb *poll = &req->poll; | |
221c5eb2 | 5614 | |
d7718a9d | 5615 | return __io_async_wake(req, poll, key_to_poll(key), io_poll_task_func); |
221c5eb2 JA |
5616 | } |
5617 | ||
221c5eb2 JA |
5618 | static void io_poll_queue_proc(struct file *file, struct wait_queue_head *head, |
5619 | struct poll_table_struct *p) | |
5620 | { | |
5621 | struct io_poll_table *pt = container_of(p, struct io_poll_table, pt); | |
5622 | ||
e8c2bc1f | 5623 | __io_queue_proc(&pt->req->poll, pt, head, (struct io_poll_iocb **) &pt->req->async_data); |
eac406c6 JA |
5624 | } |
5625 | ||
3529d8c2 | 5626 | static int io_poll_add_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
221c5eb2 JA |
5627 | { |
5628 | struct io_poll_iocb *poll = &req->poll; | |
5769a351 | 5629 | u32 events; |
221c5eb2 JA |
5630 | |
5631 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) | |
5632 | return -EINVAL; | |
5633 | if (sqe->addr || sqe->ioprio || sqe->off || sqe->len || sqe->buf_index) | |
5634 | return -EINVAL; | |
5635 | ||
5769a351 JX |
5636 | events = READ_ONCE(sqe->poll32_events); |
5637 | #ifdef __BIG_ENDIAN | |
5638 | events = swahw32(events); | |
5639 | #endif | |
a31eb4a2 JX |
5640 | poll->events = demangle_poll(events) | EPOLLERR | EPOLLHUP | |
5641 | (events & EPOLLEXCLUSIVE); | |
0969e783 JA |
5642 | return 0; |
5643 | } | |
5644 | ||
014db007 | 5645 | static int io_poll_add(struct io_kiocb *req) |
0969e783 JA |
5646 | { |
5647 | struct io_poll_iocb *poll = &req->poll; | |
5648 | struct io_ring_ctx *ctx = req->ctx; | |
5649 | struct io_poll_table ipt; | |
0969e783 | 5650 | __poll_t mask; |
0969e783 | 5651 | |
d7718a9d | 5652 | ipt.pt._qproc = io_poll_queue_proc; |
36703247 | 5653 | |
d7718a9d JA |
5654 | mask = __io_arm_poll_handler(req, &req->poll, &ipt, poll->events, |
5655 | io_poll_wake); | |
221c5eb2 | 5656 | |
8c838788 | 5657 | if (mask) { /* no async, we'd stolen it */ |
221c5eb2 | 5658 | ipt.error = 0; |
b0dd8a41 | 5659 | io_poll_complete(req, mask, 0); |
221c5eb2 | 5660 | } |
221c5eb2 JA |
5661 | spin_unlock_irq(&ctx->completion_lock); |
5662 | ||
8c838788 JA |
5663 | if (mask) { |
5664 | io_cqring_ev_posted(ctx); | |
014db007 | 5665 | io_put_req(req); |
221c5eb2 | 5666 | } |
8c838788 | 5667 | return ipt.error; |
221c5eb2 JA |
5668 | } |
5669 | ||
5262f567 JA |
5670 | static enum hrtimer_restart io_timeout_fn(struct hrtimer *timer) |
5671 | { | |
ad8a48ac JA |
5672 | struct io_timeout_data *data = container_of(timer, |
5673 | struct io_timeout_data, timer); | |
5674 | struct io_kiocb *req = data->req; | |
5675 | struct io_ring_ctx *ctx = req->ctx; | |
5262f567 JA |
5676 | unsigned long flags; |
5677 | ||
5262f567 | 5678 | spin_lock_irqsave(&ctx->completion_lock, flags); |
a71976f3 | 5679 | list_del_init(&req->timeout.list); |
01cec8c1 PB |
5680 | atomic_set(&req->ctx->cq_timeouts, |
5681 | atomic_read(&req->ctx->cq_timeouts) + 1); | |
5682 | ||
78e19bbe | 5683 | io_cqring_fill_event(req, -ETIME); |
5262f567 JA |
5684 | io_commit_cqring(ctx); |
5685 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
5686 | ||
5687 | io_cqring_ev_posted(ctx); | |
4e88d6e7 | 5688 | req_set_fail_links(req); |
5262f567 JA |
5689 | io_put_req(req); |
5690 | return HRTIMER_NORESTART; | |
5691 | } | |
5692 | ||
fbd15848 PB |
5693 | static struct io_kiocb *io_timeout_extract(struct io_ring_ctx *ctx, |
5694 | __u64 user_data) | |
f254ac04 | 5695 | { |
fbd15848 | 5696 | struct io_timeout_data *io; |
47f46768 JA |
5697 | struct io_kiocb *req; |
5698 | int ret = -ENOENT; | |
f254ac04 | 5699 | |
135fcde8 | 5700 | list_for_each_entry(req, &ctx->timeout_list, timeout.list) { |
47f46768 | 5701 | if (user_data == req->user_data) { |
47f46768 JA |
5702 | ret = 0; |
5703 | break; | |
5704 | } | |
5705 | } | |
5706 | ||
5707 | if (ret == -ENOENT) | |
fbd15848 PB |
5708 | return ERR_PTR(ret); |
5709 | ||
5710 | io = req->async_data; | |
e8c2bc1f | 5711 | ret = hrtimer_try_to_cancel(&io->timer); |
f254ac04 | 5712 | if (ret == -1) |
fbd15848 | 5713 | return ERR_PTR(-EALREADY); |
a71976f3 | 5714 | list_del_init(&req->timeout.list); |
fbd15848 PB |
5715 | return req; |
5716 | } | |
47f46768 | 5717 | |
fbd15848 PB |
5718 | static int io_timeout_cancel(struct io_ring_ctx *ctx, __u64 user_data) |
5719 | { | |
5720 | struct io_kiocb *req = io_timeout_extract(ctx, user_data); | |
5721 | ||
5722 | if (IS_ERR(req)) | |
5723 | return PTR_ERR(req); | |
f254ac04 JA |
5724 | |
5725 | req_set_fail_links(req); | |
f254ac04 | 5726 | io_cqring_fill_event(req, -ECANCELED); |
216578e5 | 5727 | io_put_req_deferred(req, 1); |
f254ac04 JA |
5728 | return 0; |
5729 | } | |
5730 | ||
9c8e11b3 PB |
5731 | static int io_timeout_update(struct io_ring_ctx *ctx, __u64 user_data, |
5732 | struct timespec64 *ts, enum hrtimer_mode mode) | |
47f46768 | 5733 | { |
9c8e11b3 PB |
5734 | struct io_kiocb *req = io_timeout_extract(ctx, user_data); |
5735 | struct io_timeout_data *data; | |
47f46768 | 5736 | |
9c8e11b3 PB |
5737 | if (IS_ERR(req)) |
5738 | return PTR_ERR(req); | |
47f46768 | 5739 | |
9c8e11b3 PB |
5740 | req->timeout.off = 0; /* noseq */ |
5741 | data = req->async_data; | |
5742 | list_add_tail(&req->timeout.list, &ctx->timeout_list); | |
5743 | hrtimer_init(&data->timer, CLOCK_MONOTONIC, mode); | |
5744 | data->timer.function = io_timeout_fn; | |
5745 | hrtimer_start(&data->timer, timespec64_to_ktime(*ts), mode); | |
5746 | return 0; | |
47f46768 JA |
5747 | } |
5748 | ||
3529d8c2 JA |
5749 | static int io_timeout_remove_prep(struct io_kiocb *req, |
5750 | const struct io_uring_sqe *sqe) | |
b29472ee | 5751 | { |
9c8e11b3 PB |
5752 | struct io_timeout_rem *tr = &req->timeout_rem; |
5753 | ||
b29472ee JA |
5754 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
5755 | return -EINVAL; | |
61710e43 DA |
5756 | if (unlikely(req->flags & (REQ_F_FIXED_FILE | REQ_F_BUFFER_SELECT))) |
5757 | return -EINVAL; | |
9c8e11b3 | 5758 | if (sqe->ioprio || sqe->buf_index || sqe->len) |
b29472ee JA |
5759 | return -EINVAL; |
5760 | ||
9c8e11b3 PB |
5761 | tr->addr = READ_ONCE(sqe->addr); |
5762 | tr->flags = READ_ONCE(sqe->timeout_flags); | |
5763 | if (tr->flags & IORING_TIMEOUT_UPDATE) { | |
5764 | if (tr->flags & ~(IORING_TIMEOUT_UPDATE|IORING_TIMEOUT_ABS)) | |
5765 | return -EINVAL; | |
5766 | if (get_timespec64(&tr->ts, u64_to_user_ptr(sqe->addr2))) | |
5767 | return -EFAULT; | |
5768 | } else if (tr->flags) { | |
5769 | /* timeout removal doesn't support flags */ | |
b29472ee | 5770 | return -EINVAL; |
9c8e11b3 | 5771 | } |
b29472ee | 5772 | |
b29472ee JA |
5773 | return 0; |
5774 | } | |
5775 | ||
11365043 JA |
5776 | /* |
5777 | * Remove or update an existing timeout command | |
5778 | */ | |
fc4df999 | 5779 | static int io_timeout_remove(struct io_kiocb *req) |
11365043 | 5780 | { |
9c8e11b3 | 5781 | struct io_timeout_rem *tr = &req->timeout_rem; |
11365043 | 5782 | struct io_ring_ctx *ctx = req->ctx; |
47f46768 | 5783 | int ret; |
11365043 | 5784 | |
11365043 | 5785 | spin_lock_irq(&ctx->completion_lock); |
9c8e11b3 PB |
5786 | if (req->timeout_rem.flags & IORING_TIMEOUT_UPDATE) { |
5787 | enum hrtimer_mode mode = (tr->flags & IORING_TIMEOUT_ABS) | |
5788 | ? HRTIMER_MODE_ABS : HRTIMER_MODE_REL; | |
5789 | ||
5790 | ret = io_timeout_update(ctx, tr->addr, &tr->ts, mode); | |
5791 | } else { | |
5792 | ret = io_timeout_cancel(ctx, tr->addr); | |
5793 | } | |
11365043 | 5794 | |
47f46768 | 5795 | io_cqring_fill_event(req, ret); |
11365043 JA |
5796 | io_commit_cqring(ctx); |
5797 | spin_unlock_irq(&ctx->completion_lock); | |
5262f567 | 5798 | io_cqring_ev_posted(ctx); |
4e88d6e7 JA |
5799 | if (ret < 0) |
5800 | req_set_fail_links(req); | |
ec9c02ad | 5801 | io_put_req(req); |
11365043 | 5802 | return 0; |
5262f567 JA |
5803 | } |
5804 | ||
3529d8c2 | 5805 | static int io_timeout_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe, |
2d28390a | 5806 | bool is_timeout_link) |
5262f567 | 5807 | { |
ad8a48ac | 5808 | struct io_timeout_data *data; |
a41525ab | 5809 | unsigned flags; |
56080b02 | 5810 | u32 off = READ_ONCE(sqe->off); |
5262f567 | 5811 | |
ad8a48ac | 5812 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
5262f567 | 5813 | return -EINVAL; |
ad8a48ac | 5814 | if (sqe->ioprio || sqe->buf_index || sqe->len != 1) |
a41525ab | 5815 | return -EINVAL; |
56080b02 | 5816 | if (off && is_timeout_link) |
2d28390a | 5817 | return -EINVAL; |
a41525ab JA |
5818 | flags = READ_ONCE(sqe->timeout_flags); |
5819 | if (flags & ~IORING_TIMEOUT_ABS) | |
5262f567 | 5820 | return -EINVAL; |
bdf20073 | 5821 | |
bfe68a22 | 5822 | req->timeout.off = off; |
26a61679 | 5823 | |
e8c2bc1f | 5824 | if (!req->async_data && io_alloc_async_data(req)) |
26a61679 JA |
5825 | return -ENOMEM; |
5826 | ||
e8c2bc1f | 5827 | data = req->async_data; |
ad8a48ac | 5828 | data->req = req; |
ad8a48ac JA |
5829 | |
5830 | if (get_timespec64(&data->ts, u64_to_user_ptr(sqe->addr))) | |
5262f567 JA |
5831 | return -EFAULT; |
5832 | ||
11365043 | 5833 | if (flags & IORING_TIMEOUT_ABS) |
ad8a48ac | 5834 | data->mode = HRTIMER_MODE_ABS; |
11365043 | 5835 | else |
ad8a48ac | 5836 | data->mode = HRTIMER_MODE_REL; |
11365043 | 5837 | |
ad8a48ac JA |
5838 | hrtimer_init(&data->timer, CLOCK_MONOTONIC, data->mode); |
5839 | return 0; | |
5840 | } | |
5841 | ||
fc4df999 | 5842 | static int io_timeout(struct io_kiocb *req) |
ad8a48ac | 5843 | { |
ad8a48ac | 5844 | struct io_ring_ctx *ctx = req->ctx; |
e8c2bc1f | 5845 | struct io_timeout_data *data = req->async_data; |
ad8a48ac | 5846 | struct list_head *entry; |
bfe68a22 | 5847 | u32 tail, off = req->timeout.off; |
ad8a48ac | 5848 | |
733f5c95 | 5849 | spin_lock_irq(&ctx->completion_lock); |
93bd25bb | 5850 | |
5262f567 JA |
5851 | /* |
5852 | * sqe->off holds how many events that need to occur for this | |
93bd25bb JA |
5853 | * timeout event to be satisfied. If it isn't set, then this is |
5854 | * a pure timeout request, sequence isn't used. | |
5262f567 | 5855 | */ |
8eb7e2d0 | 5856 | if (io_is_timeout_noseq(req)) { |
93bd25bb JA |
5857 | entry = ctx->timeout_list.prev; |
5858 | goto add; | |
5859 | } | |
5262f567 | 5860 | |
bfe68a22 PB |
5861 | tail = ctx->cached_cq_tail - atomic_read(&ctx->cq_timeouts); |
5862 | req->timeout.target_seq = tail + off; | |
5262f567 | 5863 | |
f010505b MDG |
5864 | /* Update the last seq here in case io_flush_timeouts() hasn't. |
5865 | * This is safe because ->completion_lock is held, and submissions | |
5866 | * and completions are never mixed in the same ->completion_lock section. | |
5867 | */ | |
5868 | ctx->cq_last_tm_flush = tail; | |
5869 | ||
5262f567 JA |
5870 | /* |
5871 | * Insertion sort, ensuring the first entry in the list is always | |
5872 | * the one we need first. | |
5873 | */ | |
5262f567 | 5874 | list_for_each_prev(entry, &ctx->timeout_list) { |
135fcde8 PB |
5875 | struct io_kiocb *nxt = list_entry(entry, struct io_kiocb, |
5876 | timeout.list); | |
5262f567 | 5877 | |
8eb7e2d0 | 5878 | if (io_is_timeout_noseq(nxt)) |
93bd25bb | 5879 | continue; |
bfe68a22 PB |
5880 | /* nxt.seq is behind @tail, otherwise would've been completed */ |
5881 | if (off >= nxt->timeout.target_seq - tail) | |
5262f567 JA |
5882 | break; |
5883 | } | |
93bd25bb | 5884 | add: |
135fcde8 | 5885 | list_add(&req->timeout.list, entry); |
ad8a48ac JA |
5886 | data->timer.function = io_timeout_fn; |
5887 | hrtimer_start(&data->timer, timespec64_to_ktime(data->ts), data->mode); | |
5262f567 | 5888 | spin_unlock_irq(&ctx->completion_lock); |
5262f567 JA |
5889 | return 0; |
5890 | } | |
5262f567 | 5891 | |
62755e35 JA |
5892 | static bool io_cancel_cb(struct io_wq_work *work, void *data) |
5893 | { | |
5894 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); | |
5895 | ||
5896 | return req->user_data == (unsigned long) data; | |
5897 | } | |
5898 | ||
e977d6d3 | 5899 | static int io_async_cancel_one(struct io_ring_ctx *ctx, void *sqe_addr) |
62755e35 | 5900 | { |
62755e35 | 5901 | enum io_wq_cancel cancel_ret; |
62755e35 JA |
5902 | int ret = 0; |
5903 | ||
4f26bda1 | 5904 | cancel_ret = io_wq_cancel_cb(ctx->io_wq, io_cancel_cb, sqe_addr, false); |
62755e35 JA |
5905 | switch (cancel_ret) { |
5906 | case IO_WQ_CANCEL_OK: | |
5907 | ret = 0; | |
5908 | break; | |
5909 | case IO_WQ_CANCEL_RUNNING: | |
5910 | ret = -EALREADY; | |
5911 | break; | |
5912 | case IO_WQ_CANCEL_NOTFOUND: | |
5913 | ret = -ENOENT; | |
5914 | break; | |
5915 | } | |
5916 | ||
e977d6d3 JA |
5917 | return ret; |
5918 | } | |
5919 | ||
47f46768 JA |
5920 | static void io_async_find_and_cancel(struct io_ring_ctx *ctx, |
5921 | struct io_kiocb *req, __u64 sqe_addr, | |
014db007 | 5922 | int success_ret) |
47f46768 JA |
5923 | { |
5924 | unsigned long flags; | |
5925 | int ret; | |
5926 | ||
5927 | ret = io_async_cancel_one(ctx, (void *) (unsigned long) sqe_addr); | |
5928 | if (ret != -ENOENT) { | |
5929 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
5930 | goto done; | |
5931 | } | |
5932 | ||
5933 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
5934 | ret = io_timeout_cancel(ctx, sqe_addr); | |
5935 | if (ret != -ENOENT) | |
5936 | goto done; | |
5937 | ret = io_poll_cancel(ctx, sqe_addr); | |
5938 | done: | |
b0dd8a41 JA |
5939 | if (!ret) |
5940 | ret = success_ret; | |
47f46768 JA |
5941 | io_cqring_fill_event(req, ret); |
5942 | io_commit_cqring(ctx); | |
5943 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
5944 | io_cqring_ev_posted(ctx); | |
5945 | ||
4e88d6e7 JA |
5946 | if (ret < 0) |
5947 | req_set_fail_links(req); | |
014db007 | 5948 | io_put_req(req); |
47f46768 JA |
5949 | } |
5950 | ||
3529d8c2 JA |
5951 | static int io_async_cancel_prep(struct io_kiocb *req, |
5952 | const struct io_uring_sqe *sqe) | |
e977d6d3 | 5953 | { |
fbf23849 | 5954 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
e977d6d3 | 5955 | return -EINVAL; |
61710e43 DA |
5956 | if (unlikely(req->flags & (REQ_F_FIXED_FILE | REQ_F_BUFFER_SELECT))) |
5957 | return -EINVAL; | |
5958 | if (sqe->ioprio || sqe->off || sqe->len || sqe->cancel_flags) | |
e977d6d3 JA |
5959 | return -EINVAL; |
5960 | ||
fbf23849 JA |
5961 | req->cancel.addr = READ_ONCE(sqe->addr); |
5962 | return 0; | |
5963 | } | |
5964 | ||
014db007 | 5965 | static int io_async_cancel(struct io_kiocb *req) |
fbf23849 JA |
5966 | { |
5967 | struct io_ring_ctx *ctx = req->ctx; | |
fbf23849 | 5968 | |
014db007 | 5969 | io_async_find_and_cancel(ctx, req, req->cancel.addr, 0); |
5262f567 JA |
5970 | return 0; |
5971 | } | |
5972 | ||
05f3fb3c JA |
5973 | static int io_files_update_prep(struct io_kiocb *req, |
5974 | const struct io_uring_sqe *sqe) | |
5975 | { | |
6ca56f84 JA |
5976 | if (unlikely(req->ctx->flags & IORING_SETUP_SQPOLL)) |
5977 | return -EINVAL; | |
61710e43 DA |
5978 | if (unlikely(req->flags & (REQ_F_FIXED_FILE | REQ_F_BUFFER_SELECT))) |
5979 | return -EINVAL; | |
5980 | if (sqe->ioprio || sqe->rw_flags) | |
05f3fb3c JA |
5981 | return -EINVAL; |
5982 | ||
5983 | req->files_update.offset = READ_ONCE(sqe->off); | |
5984 | req->files_update.nr_args = READ_ONCE(sqe->len); | |
5985 | if (!req->files_update.nr_args) | |
5986 | return -EINVAL; | |
5987 | req->files_update.arg = READ_ONCE(sqe->addr); | |
5988 | return 0; | |
5989 | } | |
5990 | ||
229a7b63 JA |
5991 | static int io_files_update(struct io_kiocb *req, bool force_nonblock, |
5992 | struct io_comp_state *cs) | |
fbf23849 JA |
5993 | { |
5994 | struct io_ring_ctx *ctx = req->ctx; | |
05f3fb3c JA |
5995 | struct io_uring_files_update up; |
5996 | int ret; | |
fbf23849 | 5997 | |
f86cd20c | 5998 | if (force_nonblock) |
05f3fb3c | 5999 | return -EAGAIN; |
05f3fb3c JA |
6000 | |
6001 | up.offset = req->files_update.offset; | |
6002 | up.fds = req->files_update.arg; | |
6003 | ||
6004 | mutex_lock(&ctx->uring_lock); | |
6005 | ret = __io_sqe_files_update(ctx, &up, req->files_update.nr_args); | |
6006 | mutex_unlock(&ctx->uring_lock); | |
6007 | ||
6008 | if (ret < 0) | |
6009 | req_set_fail_links(req); | |
229a7b63 | 6010 | __io_req_complete(req, ret, 0, cs); |
5262f567 JA |
6011 | return 0; |
6012 | } | |
6013 | ||
bfe76559 | 6014 | static int io_req_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
f67676d1 | 6015 | { |
d625c6ee | 6016 | switch (req->opcode) { |
e781573e | 6017 | case IORING_OP_NOP: |
bfe76559 | 6018 | return 0; |
f67676d1 JA |
6019 | case IORING_OP_READV: |
6020 | case IORING_OP_READ_FIXED: | |
3a6820f2 | 6021 | case IORING_OP_READ: |
bfe76559 | 6022 | return io_read_prep(req, sqe); |
f67676d1 JA |
6023 | case IORING_OP_WRITEV: |
6024 | case IORING_OP_WRITE_FIXED: | |
3a6820f2 | 6025 | case IORING_OP_WRITE: |
bfe76559 | 6026 | return io_write_prep(req, sqe); |
0969e783 | 6027 | case IORING_OP_POLL_ADD: |
bfe76559 | 6028 | return io_poll_add_prep(req, sqe); |
0969e783 | 6029 | case IORING_OP_POLL_REMOVE: |
bfe76559 | 6030 | return io_poll_remove_prep(req, sqe); |
8ed8d3c3 | 6031 | case IORING_OP_FSYNC: |
bfe76559 | 6032 | return io_prep_fsync(req, sqe); |
8ed8d3c3 | 6033 | case IORING_OP_SYNC_FILE_RANGE: |
bfe76559 | 6034 | return io_prep_sfr(req, sqe); |
03b1230c | 6035 | case IORING_OP_SENDMSG: |
fddaface | 6036 | case IORING_OP_SEND: |
bfe76559 | 6037 | return io_sendmsg_prep(req, sqe); |
03b1230c | 6038 | case IORING_OP_RECVMSG: |
fddaface | 6039 | case IORING_OP_RECV: |
bfe76559 | 6040 | return io_recvmsg_prep(req, sqe); |
f499a021 | 6041 | case IORING_OP_CONNECT: |
bfe76559 | 6042 | return io_connect_prep(req, sqe); |
2d28390a | 6043 | case IORING_OP_TIMEOUT: |
bfe76559 | 6044 | return io_timeout_prep(req, sqe, false); |
b29472ee | 6045 | case IORING_OP_TIMEOUT_REMOVE: |
bfe76559 | 6046 | return io_timeout_remove_prep(req, sqe); |
fbf23849 | 6047 | case IORING_OP_ASYNC_CANCEL: |
bfe76559 | 6048 | return io_async_cancel_prep(req, sqe); |
2d28390a | 6049 | case IORING_OP_LINK_TIMEOUT: |
bfe76559 | 6050 | return io_timeout_prep(req, sqe, true); |
8ed8d3c3 | 6051 | case IORING_OP_ACCEPT: |
bfe76559 | 6052 | return io_accept_prep(req, sqe); |
d63d1b5e | 6053 | case IORING_OP_FALLOCATE: |
bfe76559 | 6054 | return io_fallocate_prep(req, sqe); |
15b71abe | 6055 | case IORING_OP_OPENAT: |
bfe76559 | 6056 | return io_openat_prep(req, sqe); |
b5dba59e | 6057 | case IORING_OP_CLOSE: |
bfe76559 | 6058 | return io_close_prep(req, sqe); |
05f3fb3c | 6059 | case IORING_OP_FILES_UPDATE: |
bfe76559 | 6060 | return io_files_update_prep(req, sqe); |
eddc7ef5 | 6061 | case IORING_OP_STATX: |
bfe76559 | 6062 | return io_statx_prep(req, sqe); |
4840e418 | 6063 | case IORING_OP_FADVISE: |
bfe76559 | 6064 | return io_fadvise_prep(req, sqe); |
c1ca757b | 6065 | case IORING_OP_MADVISE: |
bfe76559 | 6066 | return io_madvise_prep(req, sqe); |
cebdb986 | 6067 | case IORING_OP_OPENAT2: |
bfe76559 | 6068 | return io_openat2_prep(req, sqe); |
3e4827b0 | 6069 | case IORING_OP_EPOLL_CTL: |
bfe76559 | 6070 | return io_epoll_ctl_prep(req, sqe); |
7d67af2c | 6071 | case IORING_OP_SPLICE: |
bfe76559 | 6072 | return io_splice_prep(req, sqe); |
ddf0322d | 6073 | case IORING_OP_PROVIDE_BUFFERS: |
bfe76559 | 6074 | return io_provide_buffers_prep(req, sqe); |
067524e9 | 6075 | case IORING_OP_REMOVE_BUFFERS: |
bfe76559 | 6076 | return io_remove_buffers_prep(req, sqe); |
f2a8d5c7 | 6077 | case IORING_OP_TEE: |
bfe76559 | 6078 | return io_tee_prep(req, sqe); |
36f4fa68 JA |
6079 | case IORING_OP_SHUTDOWN: |
6080 | return io_shutdown_prep(req, sqe); | |
80a261fd JA |
6081 | case IORING_OP_RENAMEAT: |
6082 | return io_renameat_prep(req, sqe); | |
14a1143b JA |
6083 | case IORING_OP_UNLINKAT: |
6084 | return io_unlinkat_prep(req, sqe); | |
f67676d1 JA |
6085 | } |
6086 | ||
bfe76559 PB |
6087 | printk_once(KERN_WARNING "io_uring: unhandled opcode %d\n", |
6088 | req->opcode); | |
6089 | return-EINVAL; | |
6090 | } | |
6091 | ||
6092 | static int io_req_defer_prep(struct io_kiocb *req, | |
6093 | const struct io_uring_sqe *sqe) | |
6094 | { | |
bfe76559 PB |
6095 | if (!sqe) |
6096 | return 0; | |
6097 | if (io_alloc_async_data(req)) | |
6098 | return -EAGAIN; | |
bfe76559 | 6099 | return io_req_prep(req, sqe); |
f67676d1 JA |
6100 | } |
6101 | ||
9cf7c104 PB |
6102 | static u32 io_get_sequence(struct io_kiocb *req) |
6103 | { | |
6104 | struct io_kiocb *pos; | |
6105 | struct io_ring_ctx *ctx = req->ctx; | |
f2f87370 | 6106 | u32 total_submitted, nr_reqs = 0; |
9cf7c104 | 6107 | |
f2f87370 PB |
6108 | io_for_each_link(pos, req) |
6109 | nr_reqs++; | |
9cf7c104 PB |
6110 | |
6111 | total_submitted = ctx->cached_sq_head - ctx->cached_sq_dropped; | |
6112 | return total_submitted - nr_reqs; | |
6113 | } | |
6114 | ||
3529d8c2 | 6115 | static int io_req_defer(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
de0617e4 | 6116 | { |
a197f664 | 6117 | struct io_ring_ctx *ctx = req->ctx; |
27dc8338 | 6118 | struct io_defer_entry *de; |
f67676d1 | 6119 | int ret; |
9cf7c104 | 6120 | u32 seq; |
de0617e4 | 6121 | |
9d858b21 | 6122 | /* Still need defer if there is pending req in defer list. */ |
9cf7c104 PB |
6123 | if (likely(list_empty_careful(&ctx->defer_list) && |
6124 | !(req->flags & REQ_F_IO_DRAIN))) | |
6125 | return 0; | |
6126 | ||
6127 | seq = io_get_sequence(req); | |
6128 | /* Still a chance to pass the sequence check */ | |
6129 | if (!req_need_defer(req, seq) && list_empty_careful(&ctx->defer_list)) | |
de0617e4 JA |
6130 | return 0; |
6131 | ||
e8c2bc1f | 6132 | if (!req->async_data) { |
650b5481 | 6133 | ret = io_req_defer_prep(req, sqe); |
327d6d96 | 6134 | if (ret) |
650b5481 PB |
6135 | return ret; |
6136 | } | |
cbdcb435 | 6137 | io_prep_async_link(req); |
27dc8338 PB |
6138 | de = kmalloc(sizeof(*de), GFP_KERNEL); |
6139 | if (!de) | |
6140 | return -ENOMEM; | |
2d28390a | 6141 | |
de0617e4 | 6142 | spin_lock_irq(&ctx->completion_lock); |
9cf7c104 | 6143 | if (!req_need_defer(req, seq) && list_empty(&ctx->defer_list)) { |
de0617e4 | 6144 | spin_unlock_irq(&ctx->completion_lock); |
27dc8338 | 6145 | kfree(de); |
ae34817b PB |
6146 | io_queue_async_work(req); |
6147 | return -EIOCBQUEUED; | |
de0617e4 JA |
6148 | } |
6149 | ||
915967f6 | 6150 | trace_io_uring_defer(ctx, req, req->user_data); |
27dc8338 | 6151 | de->req = req; |
9cf7c104 | 6152 | de->seq = seq; |
27dc8338 | 6153 | list_add_tail(&de->list, &ctx->defer_list); |
de0617e4 JA |
6154 | spin_unlock_irq(&ctx->completion_lock); |
6155 | return -EIOCBQUEUED; | |
6156 | } | |
6157 | ||
f573d384 | 6158 | static void io_req_drop_files(struct io_kiocb *req) |
99bc4c38 | 6159 | { |
f573d384 | 6160 | struct io_ring_ctx *ctx = req->ctx; |
c98de08c | 6161 | struct io_uring_task *tctx = req->task->io_uring; |
f573d384 JA |
6162 | unsigned long flags; |
6163 | ||
dfea9fce PB |
6164 | put_files_struct(req->work.identity->files); |
6165 | put_nsproxy(req->work.identity->nsproxy); | |
f573d384 JA |
6166 | spin_lock_irqsave(&ctx->inflight_lock, flags); |
6167 | list_del(&req->inflight_entry); | |
f573d384 JA |
6168 | spin_unlock_irqrestore(&ctx->inflight_lock, flags); |
6169 | req->flags &= ~REQ_F_INFLIGHT; | |
dfead8a8 | 6170 | req->work.flags &= ~IO_WQ_WORK_FILES; |
dfea9fce PB |
6171 | if (atomic_read(&tctx->in_idle)) |
6172 | wake_up(&tctx->wait); | |
f573d384 | 6173 | } |
99bc4c38 | 6174 | |
3ca405eb | 6175 | static void __io_clean_op(struct io_kiocb *req) |
99bc4c38 | 6176 | { |
0e1b6fe3 PB |
6177 | if (req->flags & REQ_F_BUFFER_SELECTED) { |
6178 | switch (req->opcode) { | |
6179 | case IORING_OP_READV: | |
6180 | case IORING_OP_READ_FIXED: | |
6181 | case IORING_OP_READ: | |
bcda7baa | 6182 | kfree((void *)(unsigned long)req->rw.addr); |
0e1b6fe3 PB |
6183 | break; |
6184 | case IORING_OP_RECVMSG: | |
6185 | case IORING_OP_RECV: | |
bcda7baa | 6186 | kfree(req->sr_msg.kbuf); |
0e1b6fe3 PB |
6187 | break; |
6188 | } | |
6189 | req->flags &= ~REQ_F_BUFFER_SELECTED; | |
99bc4c38 PB |
6190 | } |
6191 | ||
0e1b6fe3 PB |
6192 | if (req->flags & REQ_F_NEED_CLEANUP) { |
6193 | switch (req->opcode) { | |
6194 | case IORING_OP_READV: | |
6195 | case IORING_OP_READ_FIXED: | |
6196 | case IORING_OP_READ: | |
6197 | case IORING_OP_WRITEV: | |
6198 | case IORING_OP_WRITE_FIXED: | |
e8c2bc1f JA |
6199 | case IORING_OP_WRITE: { |
6200 | struct io_async_rw *io = req->async_data; | |
6201 | if (io->free_iovec) | |
6202 | kfree(io->free_iovec); | |
0e1b6fe3 | 6203 | break; |
e8c2bc1f | 6204 | } |
0e1b6fe3 | 6205 | case IORING_OP_RECVMSG: |
e8c2bc1f JA |
6206 | case IORING_OP_SENDMSG: { |
6207 | struct io_async_msghdr *io = req->async_data; | |
6208 | if (io->iov != io->fast_iov) | |
6209 | kfree(io->iov); | |
0e1b6fe3 | 6210 | break; |
e8c2bc1f | 6211 | } |
0e1b6fe3 PB |
6212 | case IORING_OP_SPLICE: |
6213 | case IORING_OP_TEE: | |
6214 | io_put_file(req, req->splice.file_in, | |
6215 | (req->splice.flags & SPLICE_F_FD_IN_FIXED)); | |
6216 | break; | |
f3cd4850 JA |
6217 | case IORING_OP_OPENAT: |
6218 | case IORING_OP_OPENAT2: | |
6219 | if (req->open.filename) | |
6220 | putname(req->open.filename); | |
6221 | break; | |
80a261fd JA |
6222 | case IORING_OP_RENAMEAT: |
6223 | putname(req->rename.oldpath); | |
6224 | putname(req->rename.newpath); | |
6225 | break; | |
14a1143b JA |
6226 | case IORING_OP_UNLINKAT: |
6227 | putname(req->unlink.filename); | |
6228 | break; | |
0e1b6fe3 PB |
6229 | } |
6230 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
99bc4c38 | 6231 | } |
bb175342 | 6232 | |
f573d384 JA |
6233 | if (req->flags & REQ_F_INFLIGHT) |
6234 | io_req_drop_files(req); | |
99bc4c38 PB |
6235 | } |
6236 | ||
c1379e24 PB |
6237 | static int io_issue_sqe(struct io_kiocb *req, bool force_nonblock, |
6238 | struct io_comp_state *cs) | |
2b188cc1 | 6239 | { |
a197f664 | 6240 | struct io_ring_ctx *ctx = req->ctx; |
d625c6ee | 6241 | int ret; |
2b188cc1 | 6242 | |
d625c6ee | 6243 | switch (req->opcode) { |
2b188cc1 | 6244 | case IORING_OP_NOP: |
229a7b63 | 6245 | ret = io_nop(req, cs); |
2b188cc1 JA |
6246 | break; |
6247 | case IORING_OP_READV: | |
edafccee | 6248 | case IORING_OP_READ_FIXED: |
3a6820f2 | 6249 | case IORING_OP_READ: |
a1d7c393 | 6250 | ret = io_read(req, force_nonblock, cs); |
edafccee | 6251 | break; |
3529d8c2 | 6252 | case IORING_OP_WRITEV: |
edafccee | 6253 | case IORING_OP_WRITE_FIXED: |
3a6820f2 | 6254 | case IORING_OP_WRITE: |
a1d7c393 | 6255 | ret = io_write(req, force_nonblock, cs); |
2b188cc1 | 6256 | break; |
c992fe29 | 6257 | case IORING_OP_FSYNC: |
014db007 | 6258 | ret = io_fsync(req, force_nonblock); |
c992fe29 | 6259 | break; |
221c5eb2 | 6260 | case IORING_OP_POLL_ADD: |
014db007 | 6261 | ret = io_poll_add(req); |
221c5eb2 JA |
6262 | break; |
6263 | case IORING_OP_POLL_REMOVE: | |
fc4df999 | 6264 | ret = io_poll_remove(req); |
221c5eb2 | 6265 | break; |
5d17b4a4 | 6266 | case IORING_OP_SYNC_FILE_RANGE: |
014db007 | 6267 | ret = io_sync_file_range(req, force_nonblock); |
5d17b4a4 | 6268 | break; |
0fa03c62 | 6269 | case IORING_OP_SENDMSG: |
062d04d7 PB |
6270 | ret = io_sendmsg(req, force_nonblock, cs); |
6271 | break; | |
fddaface | 6272 | case IORING_OP_SEND: |
062d04d7 | 6273 | ret = io_send(req, force_nonblock, cs); |
0fa03c62 | 6274 | break; |
aa1fa28f | 6275 | case IORING_OP_RECVMSG: |
062d04d7 PB |
6276 | ret = io_recvmsg(req, force_nonblock, cs); |
6277 | break; | |
fddaface | 6278 | case IORING_OP_RECV: |
062d04d7 | 6279 | ret = io_recv(req, force_nonblock, cs); |
aa1fa28f | 6280 | break; |
5262f567 | 6281 | case IORING_OP_TIMEOUT: |
fc4df999 | 6282 | ret = io_timeout(req); |
5262f567 | 6283 | break; |
11365043 | 6284 | case IORING_OP_TIMEOUT_REMOVE: |
fc4df999 | 6285 | ret = io_timeout_remove(req); |
11365043 | 6286 | break; |
17f2fe35 | 6287 | case IORING_OP_ACCEPT: |
229a7b63 | 6288 | ret = io_accept(req, force_nonblock, cs); |
17f2fe35 | 6289 | break; |
f8e85cf2 | 6290 | case IORING_OP_CONNECT: |
229a7b63 | 6291 | ret = io_connect(req, force_nonblock, cs); |
f8e85cf2 | 6292 | break; |
62755e35 | 6293 | case IORING_OP_ASYNC_CANCEL: |
014db007 | 6294 | ret = io_async_cancel(req); |
62755e35 | 6295 | break; |
d63d1b5e | 6296 | case IORING_OP_FALLOCATE: |
014db007 | 6297 | ret = io_fallocate(req, force_nonblock); |
d63d1b5e | 6298 | break; |
15b71abe | 6299 | case IORING_OP_OPENAT: |
014db007 | 6300 | ret = io_openat(req, force_nonblock); |
15b71abe | 6301 | break; |
b5dba59e | 6302 | case IORING_OP_CLOSE: |
229a7b63 | 6303 | ret = io_close(req, force_nonblock, cs); |
b5dba59e | 6304 | break; |
05f3fb3c | 6305 | case IORING_OP_FILES_UPDATE: |
229a7b63 | 6306 | ret = io_files_update(req, force_nonblock, cs); |
05f3fb3c | 6307 | break; |
eddc7ef5 | 6308 | case IORING_OP_STATX: |
014db007 | 6309 | ret = io_statx(req, force_nonblock); |
eddc7ef5 | 6310 | break; |
4840e418 | 6311 | case IORING_OP_FADVISE: |
014db007 | 6312 | ret = io_fadvise(req, force_nonblock); |
4840e418 | 6313 | break; |
c1ca757b | 6314 | case IORING_OP_MADVISE: |
014db007 | 6315 | ret = io_madvise(req, force_nonblock); |
c1ca757b | 6316 | break; |
cebdb986 | 6317 | case IORING_OP_OPENAT2: |
014db007 | 6318 | ret = io_openat2(req, force_nonblock); |
cebdb986 | 6319 | break; |
3e4827b0 | 6320 | case IORING_OP_EPOLL_CTL: |
229a7b63 | 6321 | ret = io_epoll_ctl(req, force_nonblock, cs); |
3e4827b0 | 6322 | break; |
7d67af2c | 6323 | case IORING_OP_SPLICE: |
014db007 | 6324 | ret = io_splice(req, force_nonblock); |
7d67af2c | 6325 | break; |
ddf0322d | 6326 | case IORING_OP_PROVIDE_BUFFERS: |
229a7b63 | 6327 | ret = io_provide_buffers(req, force_nonblock, cs); |
ddf0322d | 6328 | break; |
067524e9 | 6329 | case IORING_OP_REMOVE_BUFFERS: |
229a7b63 | 6330 | ret = io_remove_buffers(req, force_nonblock, cs); |
3e4827b0 | 6331 | break; |
f2a8d5c7 | 6332 | case IORING_OP_TEE: |
f2a8d5c7 PB |
6333 | ret = io_tee(req, force_nonblock); |
6334 | break; | |
36f4fa68 JA |
6335 | case IORING_OP_SHUTDOWN: |
6336 | ret = io_shutdown(req, force_nonblock); | |
6337 | break; | |
80a261fd JA |
6338 | case IORING_OP_RENAMEAT: |
6339 | ret = io_renameat(req, force_nonblock); | |
6340 | break; | |
14a1143b JA |
6341 | case IORING_OP_UNLINKAT: |
6342 | ret = io_unlinkat(req, force_nonblock); | |
6343 | break; | |
2b188cc1 JA |
6344 | default: |
6345 | ret = -EINVAL; | |
6346 | break; | |
6347 | } | |
6348 | ||
def596e9 JA |
6349 | if (ret) |
6350 | return ret; | |
6351 | ||
b532576e JA |
6352 | /* If the op doesn't have a file, we're not polling for it */ |
6353 | if ((ctx->flags & IORING_SETUP_IOPOLL) && req->file) { | |
11ba820b JA |
6354 | const bool in_async = io_wq_current_is_worker(); |
6355 | ||
11ba820b JA |
6356 | /* workqueue context doesn't hold uring_lock, grab it now */ |
6357 | if (in_async) | |
6358 | mutex_lock(&ctx->uring_lock); | |
6359 | ||
2e9dbe90 | 6360 | io_iopoll_req_issued(req, in_async); |
11ba820b JA |
6361 | |
6362 | if (in_async) | |
6363 | mutex_unlock(&ctx->uring_lock); | |
def596e9 JA |
6364 | } |
6365 | ||
6366 | return 0; | |
2b188cc1 JA |
6367 | } |
6368 | ||
f4db7182 | 6369 | static struct io_wq_work *io_wq_submit_work(struct io_wq_work *work) |
2b188cc1 JA |
6370 | { |
6371 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); | |
6df1db6b | 6372 | struct io_kiocb *timeout; |
561fb04a | 6373 | int ret = 0; |
2b188cc1 | 6374 | |
6df1db6b PB |
6375 | timeout = io_prep_linked_timeout(req); |
6376 | if (timeout) | |
6377 | io_queue_linked_timeout(timeout); | |
d4c81f38 | 6378 | |
0c9d5ccd JA |
6379 | /* if NO_CANCEL is set, we must still run the work */ |
6380 | if ((work->flags & (IO_WQ_WORK_CANCEL|IO_WQ_WORK_NO_CANCEL)) == | |
6381 | IO_WQ_WORK_CANCEL) { | |
561fb04a | 6382 | ret = -ECANCELED; |
0c9d5ccd | 6383 | } |
31b51510 | 6384 | |
561fb04a | 6385 | if (!ret) { |
561fb04a | 6386 | do { |
c1379e24 | 6387 | ret = io_issue_sqe(req, false, NULL); |
561fb04a JA |
6388 | /* |
6389 | * We can get EAGAIN for polled IO even though we're | |
6390 | * forcing a sync submission from here, since we can't | |
6391 | * wait for request slots on the block side. | |
6392 | */ | |
6393 | if (ret != -EAGAIN) | |
6394 | break; | |
6395 | cond_resched(); | |
6396 | } while (1); | |
6397 | } | |
31b51510 | 6398 | |
561fb04a | 6399 | if (ret) { |
c07e6719 XW |
6400 | struct io_ring_ctx *lock_ctx = NULL; |
6401 | ||
6402 | if (req->ctx->flags & IORING_SETUP_IOPOLL) | |
6403 | lock_ctx = req->ctx; | |
6404 | ||
dad1b124 | 6405 | /* |
c07e6719 XW |
6406 | * io_iopoll_complete() does not hold completion_lock to |
6407 | * complete polled io, so here for polled io, we can not call | |
6408 | * io_req_complete() directly, otherwise there maybe concurrent | |
6409 | * access to cqring, defer_list, etc, which is not safe. Given | |
6410 | * that io_iopoll_complete() is always called under uring_lock, | |
6411 | * so here for polled io, we also get uring_lock to complete | |
6412 | * it. | |
dad1b124 | 6413 | */ |
c07e6719 XW |
6414 | if (lock_ctx) |
6415 | mutex_lock(&lock_ctx->uring_lock); | |
dad1b124 | 6416 | |
c07e6719 XW |
6417 | req_set_fail_links(req); |
6418 | io_req_complete(req, ret); | |
6419 | ||
6420 | if (lock_ctx) | |
6421 | mutex_unlock(&lock_ctx->uring_lock); | |
edafccee | 6422 | } |
2b188cc1 | 6423 | |
f4db7182 | 6424 | return io_steal_work(req); |
2b188cc1 JA |
6425 | } |
6426 | ||
65e19f54 JA |
6427 | static inline struct file *io_file_from_index(struct io_ring_ctx *ctx, |
6428 | int index) | |
6429 | { | |
6430 | struct fixed_file_table *table; | |
6431 | ||
05f3fb3c | 6432 | table = &ctx->file_data->table[index >> IORING_FILE_TABLE_SHIFT]; |
84695089 | 6433 | return table->files[index & IORING_FILE_TABLE_MASK]; |
65e19f54 JA |
6434 | } |
6435 | ||
8371adf5 PB |
6436 | static struct file *io_file_get(struct io_submit_state *state, |
6437 | struct io_kiocb *req, int fd, bool fixed) | |
09bb8394 | 6438 | { |
a197f664 | 6439 | struct io_ring_ctx *ctx = req->ctx; |
8da11c19 | 6440 | struct file *file; |
09bb8394 | 6441 | |
8da11c19 | 6442 | if (fixed) { |
479f517b | 6443 | if (unlikely((unsigned int)fd >= ctx->nr_user_files)) |
8371adf5 | 6444 | return NULL; |
b7620121 | 6445 | fd = array_index_nospec(fd, ctx->nr_user_files); |
8da11c19 | 6446 | file = io_file_from_index(ctx, fd); |
36f72fe2 | 6447 | io_set_resource_node(req); |
09bb8394 | 6448 | } else { |
c826bd7a | 6449 | trace_io_uring_file_get(ctx, fd); |
8da11c19 | 6450 | file = __io_file_get(state, fd); |
09bb8394 JA |
6451 | } |
6452 | ||
8371adf5 | 6453 | return file; |
09bb8394 JA |
6454 | } |
6455 | ||
2665abfd | 6456 | static enum hrtimer_restart io_link_timeout_fn(struct hrtimer *timer) |
2b188cc1 | 6457 | { |
ad8a48ac JA |
6458 | struct io_timeout_data *data = container_of(timer, |
6459 | struct io_timeout_data, timer); | |
90cd7e42 | 6460 | struct io_kiocb *prev, *req = data->req; |
2665abfd | 6461 | struct io_ring_ctx *ctx = req->ctx; |
2665abfd | 6462 | unsigned long flags; |
2665abfd JA |
6463 | |
6464 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
90cd7e42 PB |
6465 | prev = req->timeout.head; |
6466 | req->timeout.head = NULL; | |
2665abfd JA |
6467 | |
6468 | /* | |
6469 | * We don't expect the list to be empty, that will only happen if we | |
6470 | * race with the completion of the linked work. | |
6471 | */ | |
90cd7e42 | 6472 | if (prev && refcount_inc_not_zero(&prev->refs)) |
f2f87370 | 6473 | io_remove_next_linked(prev); |
90cd7e42 PB |
6474 | else |
6475 | prev = NULL; | |
2665abfd JA |
6476 | spin_unlock_irqrestore(&ctx->completion_lock, flags); |
6477 | ||
6478 | if (prev) { | |
4e88d6e7 | 6479 | req_set_fail_links(prev); |
014db007 | 6480 | io_async_find_and_cancel(ctx, req, prev->user_data, -ETIME); |
76a46e06 | 6481 | io_put_req(prev); |
47f46768 | 6482 | } else { |
e1e16097 | 6483 | io_req_complete(req, -ETIME); |
2665abfd | 6484 | } |
2665abfd JA |
6485 | return HRTIMER_NORESTART; |
6486 | } | |
6487 | ||
7271ef3a | 6488 | static void __io_queue_linked_timeout(struct io_kiocb *req) |
2665abfd | 6489 | { |
76a46e06 | 6490 | /* |
f2f87370 PB |
6491 | * If the back reference is NULL, then our linked request finished |
6492 | * before we got a chance to setup the timer | |
76a46e06 | 6493 | */ |
90cd7e42 | 6494 | if (req->timeout.head) { |
e8c2bc1f | 6495 | struct io_timeout_data *data = req->async_data; |
94ae5e77 | 6496 | |
ad8a48ac JA |
6497 | data->timer.function = io_link_timeout_fn; |
6498 | hrtimer_start(&data->timer, timespec64_to_ktime(data->ts), | |
6499 | data->mode); | |
2665abfd | 6500 | } |
7271ef3a JA |
6501 | } |
6502 | ||
6503 | static void io_queue_linked_timeout(struct io_kiocb *req) | |
6504 | { | |
6505 | struct io_ring_ctx *ctx = req->ctx; | |
6506 | ||
6507 | spin_lock_irq(&ctx->completion_lock); | |
6508 | __io_queue_linked_timeout(req); | |
76a46e06 | 6509 | spin_unlock_irq(&ctx->completion_lock); |
2665abfd | 6510 | |
2665abfd | 6511 | /* drop submission reference */ |
76a46e06 JA |
6512 | io_put_req(req); |
6513 | } | |
2665abfd | 6514 | |
ad8a48ac | 6515 | static struct io_kiocb *io_prep_linked_timeout(struct io_kiocb *req) |
2665abfd | 6516 | { |
f2f87370 | 6517 | struct io_kiocb *nxt = req->link; |
2665abfd | 6518 | |
f2f87370 PB |
6519 | if (!nxt || (req->flags & REQ_F_LINK_TIMEOUT) || |
6520 | nxt->opcode != IORING_OP_LINK_TIMEOUT) | |
76a46e06 | 6521 | return NULL; |
2665abfd | 6522 | |
90cd7e42 | 6523 | nxt->timeout.head = req; |
900fad45 | 6524 | nxt->flags |= REQ_F_LTIMEOUT_ACTIVE; |
76a46e06 | 6525 | req->flags |= REQ_F_LINK_TIMEOUT; |
76a46e06 | 6526 | return nxt; |
2665abfd JA |
6527 | } |
6528 | ||
c1379e24 | 6529 | static void __io_queue_sqe(struct io_kiocb *req, struct io_comp_state *cs) |
2b188cc1 | 6530 | { |
4a0a7a18 | 6531 | struct io_kiocb *linked_timeout; |
193155c8 | 6532 | const struct cred *old_creds = NULL; |
e0c5c576 | 6533 | int ret; |
2b188cc1 | 6534 | |
4a0a7a18 JA |
6535 | again: |
6536 | linked_timeout = io_prep_linked_timeout(req); | |
6537 | ||
2e5aa6cb PB |
6538 | if ((req->flags & REQ_F_WORK_INITIALIZED) && |
6539 | (req->work.flags & IO_WQ_WORK_CREDS) && | |
98447d65 | 6540 | req->work.identity->creds != current_cred()) { |
193155c8 JA |
6541 | if (old_creds) |
6542 | revert_creds(old_creds); | |
98447d65 | 6543 | if (old_creds == req->work.identity->creds) |
193155c8 JA |
6544 | old_creds = NULL; /* restored original creds */ |
6545 | else | |
98447d65 | 6546 | old_creds = override_creds(req->work.identity->creds); |
193155c8 JA |
6547 | } |
6548 | ||
c1379e24 | 6549 | ret = io_issue_sqe(req, true, cs); |
491381ce JA |
6550 | |
6551 | /* | |
6552 | * We async punt it if the file wasn't marked NOWAIT, or if the file | |
6553 | * doesn't support non-blocking read/write attempts | |
6554 | */ | |
24c74678 | 6555 | if (ret == -EAGAIN && !(req->flags & REQ_F_NOWAIT)) { |
f063c547 | 6556 | if (!io_arm_poll_handler(req)) { |
f063c547 PB |
6557 | /* |
6558 | * Queued up for async execution, worker will release | |
6559 | * submit reference when the iocb is actually submitted. | |
6560 | */ | |
6561 | io_queue_async_work(req); | |
2b188cc1 | 6562 | } |
bbad27b2 | 6563 | |
f063c547 PB |
6564 | if (linked_timeout) |
6565 | io_queue_linked_timeout(linked_timeout); | |
0d63c148 PB |
6566 | } else if (likely(!ret)) { |
6567 | /* drop submission reference */ | |
6568 | req = io_put_req_find_next(req); | |
6569 | if (linked_timeout) | |
6570 | io_queue_linked_timeout(linked_timeout); | |
e65ef56d | 6571 | |
0d63c148 PB |
6572 | if (req) { |
6573 | if (!(req->flags & REQ_F_FORCE_ASYNC)) | |
6574 | goto again; | |
6575 | io_queue_async_work(req); | |
6576 | } | |
6577 | } else { | |
652532ad PB |
6578 | /* un-prep timeout, so it'll be killed as any other linked */ |
6579 | req->flags &= ~REQ_F_LINK_TIMEOUT; | |
4e88d6e7 | 6580 | req_set_fail_links(req); |
e65ef56d | 6581 | io_put_req(req); |
e1e16097 | 6582 | io_req_complete(req, ret); |
9e645e11 | 6583 | } |
652532ad | 6584 | |
193155c8 JA |
6585 | if (old_creds) |
6586 | revert_creds(old_creds); | |
2b188cc1 JA |
6587 | } |
6588 | ||
f13fad7b JA |
6589 | static void io_queue_sqe(struct io_kiocb *req, const struct io_uring_sqe *sqe, |
6590 | struct io_comp_state *cs) | |
4fe2c963 JL |
6591 | { |
6592 | int ret; | |
6593 | ||
3529d8c2 | 6594 | ret = io_req_defer(req, sqe); |
4fe2c963 JL |
6595 | if (ret) { |
6596 | if (ret != -EIOCBQUEUED) { | |
1118591a | 6597 | fail_req: |
4e88d6e7 | 6598 | req_set_fail_links(req); |
e1e16097 JA |
6599 | io_put_req(req); |
6600 | io_req_complete(req, ret); | |
4fe2c963 | 6601 | } |
2550878f | 6602 | } else if (req->flags & REQ_F_FORCE_ASYNC) { |
e8c2bc1f | 6603 | if (!req->async_data) { |
bd2ab18a | 6604 | ret = io_req_defer_prep(req, sqe); |
327d6d96 | 6605 | if (unlikely(ret)) |
bd2ab18a PB |
6606 | goto fail_req; |
6607 | } | |
ce35a47a JA |
6608 | io_queue_async_work(req); |
6609 | } else { | |
c1379e24 PB |
6610 | if (sqe) { |
6611 | ret = io_req_prep(req, sqe); | |
6612 | if (unlikely(ret)) | |
6613 | goto fail_req; | |
6614 | } | |
6615 | __io_queue_sqe(req, cs); | |
ce35a47a | 6616 | } |
4fe2c963 JL |
6617 | } |
6618 | ||
f13fad7b JA |
6619 | static inline void io_queue_link_head(struct io_kiocb *req, |
6620 | struct io_comp_state *cs) | |
4fe2c963 | 6621 | { |
94ae5e77 | 6622 | if (unlikely(req->flags & REQ_F_FAIL_LINK)) { |
e1e16097 JA |
6623 | io_put_req(req); |
6624 | io_req_complete(req, -ECANCELED); | |
1b4a51b6 | 6625 | } else |
f13fad7b | 6626 | io_queue_sqe(req, NULL, cs); |
4fe2c963 JL |
6627 | } |
6628 | ||
863e0560 PB |
6629 | struct io_submit_link { |
6630 | struct io_kiocb *head; | |
6631 | struct io_kiocb *last; | |
6632 | }; | |
6633 | ||
1d4240cc | 6634 | static int io_submit_sqe(struct io_kiocb *req, const struct io_uring_sqe *sqe, |
863e0560 | 6635 | struct io_submit_link *link, struct io_comp_state *cs) |
9e645e11 | 6636 | { |
a197f664 | 6637 | struct io_ring_ctx *ctx = req->ctx; |
ef4ff581 | 6638 | int ret; |
9e645e11 | 6639 | |
9e645e11 JA |
6640 | /* |
6641 | * If we already have a head request, queue this one for async | |
6642 | * submittal once the head completes. If we don't have a head but | |
6643 | * IOSQE_IO_LINK is set in the sqe, start a new head. This one will be | |
6644 | * submitted sync once the chain is complete. If none of those | |
6645 | * conditions are true (normal request), then just queue it. | |
6646 | */ | |
863e0560 PB |
6647 | if (link->head) { |
6648 | struct io_kiocb *head = link->head; | |
4e88d6e7 | 6649 | |
8cdf2193 PB |
6650 | /* |
6651 | * Taking sequential execution of a link, draining both sides | |
6652 | * of the link also fullfils IOSQE_IO_DRAIN semantics for all | |
6653 | * requests in the link. So, it drains the head and the | |
6654 | * next after the link request. The last one is done via | |
6655 | * drain_next flag to persist the effect across calls. | |
6656 | */ | |
ef4ff581 | 6657 | if (req->flags & REQ_F_IO_DRAIN) { |
711be031 PB |
6658 | head->flags |= REQ_F_IO_DRAIN; |
6659 | ctx->drain_next = 1; | |
6660 | } | |
3529d8c2 | 6661 | ret = io_req_defer_prep(req, sqe); |
327d6d96 | 6662 | if (unlikely(ret)) { |
4e88d6e7 | 6663 | /* fail even hard links since we don't submit */ |
9d76377f | 6664 | head->flags |= REQ_F_FAIL_LINK; |
1d4240cc | 6665 | return ret; |
2d28390a | 6666 | } |
9d76377f | 6667 | trace_io_uring_link(ctx, req, head); |
f2f87370 | 6668 | link->last->link = req; |
863e0560 | 6669 | link->last = req; |
32fe525b PB |
6670 | |
6671 | /* last request of a link, enqueue the link */ | |
ef4ff581 | 6672 | if (!(req->flags & (REQ_F_LINK | REQ_F_HARDLINK))) { |
f13fad7b | 6673 | io_queue_link_head(head, cs); |
863e0560 | 6674 | link->head = NULL; |
32fe525b | 6675 | } |
9e645e11 | 6676 | } else { |
711be031 PB |
6677 | if (unlikely(ctx->drain_next)) { |
6678 | req->flags |= REQ_F_IO_DRAIN; | |
ef4ff581 | 6679 | ctx->drain_next = 0; |
711be031 | 6680 | } |
ef4ff581 | 6681 | if (req->flags & (REQ_F_LINK | REQ_F_HARDLINK)) { |
711be031 | 6682 | ret = io_req_defer_prep(req, sqe); |
327d6d96 | 6683 | if (unlikely(ret)) |
711be031 | 6684 | req->flags |= REQ_F_FAIL_LINK; |
863e0560 PB |
6685 | link->head = req; |
6686 | link->last = req; | |
711be031 | 6687 | } else { |
f13fad7b | 6688 | io_queue_sqe(req, sqe, cs); |
711be031 | 6689 | } |
9e645e11 | 6690 | } |
2e6e1fde | 6691 | |
1d4240cc | 6692 | return 0; |
9e645e11 JA |
6693 | } |
6694 | ||
9a56a232 JA |
6695 | /* |
6696 | * Batched submission is done, ensure local IO is flushed out. | |
6697 | */ | |
6698 | static void io_submit_state_end(struct io_submit_state *state) | |
6699 | { | |
f13fad7b JA |
6700 | if (!list_empty(&state->comp.list)) |
6701 | io_submit_flush_completions(&state->comp); | |
27926b68 JA |
6702 | if (state->plug_started) |
6703 | blk_finish_plug(&state->plug); | |
9f13c35b | 6704 | io_state_file_put(state); |
2579f913 | 6705 | if (state->free_reqs) |
6c8a3134 | 6706 | kmem_cache_free_bulk(req_cachep, state->free_reqs, state->reqs); |
9a56a232 JA |
6707 | } |
6708 | ||
6709 | /* | |
6710 | * Start submission side cache. | |
6711 | */ | |
6712 | static void io_submit_state_start(struct io_submit_state *state, | |
013538bd | 6713 | struct io_ring_ctx *ctx, unsigned int max_ios) |
9a56a232 | 6714 | { |
27926b68 | 6715 | state->plug_started = false; |
013538bd JA |
6716 | state->comp.nr = 0; |
6717 | INIT_LIST_HEAD(&state->comp.list); | |
6718 | state->comp.ctx = ctx; | |
2579f913 | 6719 | state->free_reqs = 0; |
6e1271e6 | 6720 | state->file_refs = 0; |
9a56a232 JA |
6721 | state->ios_left = max_ios; |
6722 | } | |
6723 | ||
2b188cc1 JA |
6724 | static void io_commit_sqring(struct io_ring_ctx *ctx) |
6725 | { | |
75b28aff | 6726 | struct io_rings *rings = ctx->rings; |
2b188cc1 | 6727 | |
caf582c6 PB |
6728 | /* |
6729 | * Ensure any loads from the SQEs are done at this point, | |
6730 | * since once we write the new head, the application could | |
6731 | * write new data to them. | |
6732 | */ | |
6733 | smp_store_release(&rings->sq.head, ctx->cached_sq_head); | |
2b188cc1 JA |
6734 | } |
6735 | ||
2b188cc1 | 6736 | /* |
3529d8c2 | 6737 | * Fetch an sqe, if one is available. Note that sqe_ptr will point to memory |
2b188cc1 JA |
6738 | * that is mapped by userspace. This means that care needs to be taken to |
6739 | * ensure that reads are stable, as we cannot rely on userspace always | |
6740 | * being a good citizen. If members of the sqe are validated and then later | |
6741 | * used, it's important that those reads are done through READ_ONCE() to | |
6742 | * prevent a re-load down the line. | |
6743 | */ | |
709b302f | 6744 | static const struct io_uring_sqe *io_get_sqe(struct io_ring_ctx *ctx) |
2b188cc1 | 6745 | { |
75b28aff | 6746 | u32 *sq_array = ctx->sq_array; |
2b188cc1 JA |
6747 | unsigned head; |
6748 | ||
6749 | /* | |
6750 | * The cached sq head (or cq tail) serves two purposes: | |
6751 | * | |
6752 | * 1) allows us to batch the cost of updating the user visible | |
6753 | * head updates. | |
6754 | * 2) allows the kernel side to track the head on its own, even | |
6755 | * though the application is the one updating it. | |
6756 | */ | |
ee7d46d9 | 6757 | head = READ_ONCE(sq_array[ctx->cached_sq_head & ctx->sq_mask]); |
709b302f PB |
6758 | if (likely(head < ctx->sq_entries)) |
6759 | return &ctx->sq_sqes[head]; | |
2b188cc1 JA |
6760 | |
6761 | /* drop invalid entries */ | |
498ccd9e | 6762 | ctx->cached_sq_dropped++; |
ee7d46d9 | 6763 | WRITE_ONCE(ctx->rings->sq_dropped, ctx->cached_sq_dropped); |
709b302f PB |
6764 | return NULL; |
6765 | } | |
6766 | ||
6767 | static inline void io_consume_sqe(struct io_ring_ctx *ctx) | |
6768 | { | |
6769 | ctx->cached_sq_head++; | |
2b188cc1 JA |
6770 | } |
6771 | ||
21b55dbc SG |
6772 | /* |
6773 | * Check SQE restrictions (opcode and flags). | |
6774 | * | |
6775 | * Returns 'true' if SQE is allowed, 'false' otherwise. | |
6776 | */ | |
6777 | static inline bool io_check_restriction(struct io_ring_ctx *ctx, | |
6778 | struct io_kiocb *req, | |
6779 | unsigned int sqe_flags) | |
6780 | { | |
6781 | if (!ctx->restricted) | |
6782 | return true; | |
6783 | ||
6784 | if (!test_bit(req->opcode, ctx->restrictions.sqe_op)) | |
6785 | return false; | |
6786 | ||
6787 | if ((sqe_flags & ctx->restrictions.sqe_flags_required) != | |
6788 | ctx->restrictions.sqe_flags_required) | |
6789 | return false; | |
6790 | ||
6791 | if (sqe_flags & ~(ctx->restrictions.sqe_flags_allowed | | |
6792 | ctx->restrictions.sqe_flags_required)) | |
6793 | return false; | |
6794 | ||
6795 | return true; | |
6796 | } | |
6797 | ||
ef4ff581 PB |
6798 | #define SQE_VALID_FLAGS (IOSQE_FIXED_FILE|IOSQE_IO_DRAIN|IOSQE_IO_LINK| \ |
6799 | IOSQE_IO_HARDLINK | IOSQE_ASYNC | \ | |
6800 | IOSQE_BUFFER_SELECT) | |
6801 | ||
6802 | static int io_init_req(struct io_ring_ctx *ctx, struct io_kiocb *req, | |
6803 | const struct io_uring_sqe *sqe, | |
0cdaf760 | 6804 | struct io_submit_state *state) |
0553b8bd | 6805 | { |
ef4ff581 | 6806 | unsigned int sqe_flags; |
71b547c0 | 6807 | int id, ret; |
ef4ff581 | 6808 | |
0553b8bd PB |
6809 | req->opcode = READ_ONCE(sqe->opcode); |
6810 | req->user_data = READ_ONCE(sqe->user_data); | |
e8c2bc1f | 6811 | req->async_data = NULL; |
0553b8bd PB |
6812 | req->file = NULL; |
6813 | req->ctx = ctx; | |
6814 | req->flags = 0; | |
f2f87370 | 6815 | req->link = NULL; |
36f72fe2 | 6816 | req->fixed_file_refs = NULL; |
0553b8bd PB |
6817 | /* one is dropped after submission, the other at completion */ |
6818 | refcount_set(&req->refs, 2); | |
4dd2824d | 6819 | req->task = current; |
0553b8bd | 6820 | req->result = 0; |
ef4ff581 PB |
6821 | |
6822 | if (unlikely(req->opcode >= IORING_OP_LAST)) | |
6823 | return -EINVAL; | |
6824 | ||
28cea78a | 6825 | if (unlikely(io_sq_thread_acquire_mm_files(ctx, req))) |
9d8426a0 | 6826 | return -EFAULT; |
ef4ff581 PB |
6827 | |
6828 | sqe_flags = READ_ONCE(sqe->flags); | |
6829 | /* enforce forwards compatibility on users */ | |
6830 | if (unlikely(sqe_flags & ~SQE_VALID_FLAGS)) | |
6831 | return -EINVAL; | |
6832 | ||
21b55dbc SG |
6833 | if (unlikely(!io_check_restriction(ctx, req, sqe_flags))) |
6834 | return -EACCES; | |
6835 | ||
ef4ff581 PB |
6836 | if ((sqe_flags & IOSQE_BUFFER_SELECT) && |
6837 | !io_op_defs[req->opcode].buffer_select) | |
6838 | return -EOPNOTSUPP; | |
6839 | ||
6840 | id = READ_ONCE(sqe->personality); | |
6841 | if (id) { | |
1e6fa521 JA |
6842 | struct io_identity *iod; |
6843 | ||
1e6fa521 JA |
6844 | iod = idr_find(&ctx->personality_idr, id); |
6845 | if (unlikely(!iod)) | |
ef4ff581 | 6846 | return -EINVAL; |
1e6fa521 | 6847 | refcount_inc(&iod->count); |
ec99ca6c PB |
6848 | |
6849 | __io_req_init_async(req); | |
1e6fa521 JA |
6850 | get_cred(iod->creds); |
6851 | req->work.identity = iod; | |
dfead8a8 | 6852 | req->work.flags |= IO_WQ_WORK_CREDS; |
ef4ff581 PB |
6853 | } |
6854 | ||
6855 | /* same numerical values with corresponding REQ_F_*, safe to copy */ | |
c11368a5 | 6856 | req->flags |= sqe_flags; |
ef4ff581 | 6857 | |
27926b68 JA |
6858 | /* |
6859 | * Plug now if we have more than 1 IO left after this, and the target | |
6860 | * is potentially a read/write to block based storage. | |
6861 | */ | |
6862 | if (!state->plug_started && state->ios_left > 1 && | |
6863 | io_op_defs[req->opcode].plug) { | |
6864 | blk_start_plug(&state->plug); | |
6865 | state->plug_started = true; | |
6866 | } | |
6867 | ||
bd5bbda7 PB |
6868 | ret = 0; |
6869 | if (io_op_defs[req->opcode].needs_file) { | |
6870 | bool fixed = req->flags & REQ_F_FIXED_FILE; | |
6871 | ||
6872 | req->file = io_file_get(state, req, READ_ONCE(sqe->fd), fixed); | |
6873 | if (unlikely(!req->file && | |
6874 | !io_op_defs[req->opcode].needs_file_no_error)) | |
6875 | ret = -EBADF; | |
6876 | } | |
63ff8223 | 6877 | |
71b547c0 PB |
6878 | state->ios_left--; |
6879 | return ret; | |
0553b8bd PB |
6880 | } |
6881 | ||
0f212204 | 6882 | static int io_submit_sqes(struct io_ring_ctx *ctx, unsigned int nr) |
6c271ce2 | 6883 | { |
ac8691c4 | 6884 | struct io_submit_state state; |
863e0560 | 6885 | struct io_submit_link link; |
9e645e11 | 6886 | int i, submitted = 0; |
6c271ce2 | 6887 | |
c4a2ed72 | 6888 | /* if we have a backlog and couldn't flush it all, return BUSY */ |
ad3eb2c8 | 6889 | if (test_bit(0, &ctx->sq_check_overflow)) { |
6c503150 | 6890 | if (!__io_cqring_overflow_flush(ctx, false, NULL, NULL)) |
ad3eb2c8 JA |
6891 | return -EBUSY; |
6892 | } | |
6c271ce2 | 6893 | |
ee7d46d9 PB |
6894 | /* make sure SQ entry isn't read before tail */ |
6895 | nr = min3(nr, ctx->sq_entries, io_sqring_entries(ctx)); | |
9ef4f124 | 6896 | |
2b85edfc PB |
6897 | if (!percpu_ref_tryget_many(&ctx->refs, nr)) |
6898 | return -EAGAIN; | |
6c271ce2 | 6899 | |
d8a6df10 | 6900 | percpu_counter_add(¤t->io_uring->inflight, nr); |
faf7b51c | 6901 | refcount_add(nr, ¤t->usage); |
6c271ce2 | 6902 | |
013538bd | 6903 | io_submit_state_start(&state, ctx, nr); |
863e0560 | 6904 | link.head = NULL; |
b14cca0c | 6905 | |
6c271ce2 | 6906 | for (i = 0; i < nr; i++) { |
3529d8c2 | 6907 | const struct io_uring_sqe *sqe; |
196be95c | 6908 | struct io_kiocb *req; |
1cb1edb2 | 6909 | int err; |
fb5ccc98 | 6910 | |
b1e50e54 PB |
6911 | sqe = io_get_sqe(ctx); |
6912 | if (unlikely(!sqe)) { | |
6913 | io_consume_sqe(ctx); | |
6914 | break; | |
6915 | } | |
ac8691c4 | 6916 | req = io_alloc_req(ctx, &state); |
196be95c PB |
6917 | if (unlikely(!req)) { |
6918 | if (!submitted) | |
6919 | submitted = -EAGAIN; | |
fb5ccc98 | 6920 | break; |
196be95c | 6921 | } |
709b302f | 6922 | io_consume_sqe(ctx); |
d3656344 JA |
6923 | /* will complete beyond this point, count as submitted */ |
6924 | submitted++; | |
6925 | ||
692d8363 | 6926 | err = io_init_req(ctx, req, sqe, &state); |
ef4ff581 | 6927 | if (unlikely(err)) { |
1cb1edb2 | 6928 | fail_req: |
e1e16097 JA |
6929 | io_put_req(req); |
6930 | io_req_complete(req, err); | |
196be95c PB |
6931 | break; |
6932 | } | |
fb5ccc98 | 6933 | |
354420f7 | 6934 | trace_io_uring_submit_sqe(ctx, req->opcode, req->user_data, |
0cdaf760 | 6935 | true, io_async_submit(ctx)); |
f13fad7b | 6936 | err = io_submit_sqe(req, sqe, &link, &state.comp); |
1d4240cc PB |
6937 | if (err) |
6938 | goto fail_req; | |
6c271ce2 JA |
6939 | } |
6940 | ||
9466f437 PB |
6941 | if (unlikely(submitted != nr)) { |
6942 | int ref_used = (submitted == -EAGAIN) ? 0 : submitted; | |
d8a6df10 JA |
6943 | struct io_uring_task *tctx = current->io_uring; |
6944 | int unused = nr - ref_used; | |
9466f437 | 6945 | |
d8a6df10 JA |
6946 | percpu_ref_put_many(&ctx->refs, unused); |
6947 | percpu_counter_sub(&tctx->inflight, unused); | |
6948 | put_task_struct_many(current, unused); | |
9466f437 | 6949 | } |
863e0560 PB |
6950 | if (link.head) |
6951 | io_queue_link_head(link.head, &state.comp); | |
ac8691c4 | 6952 | io_submit_state_end(&state); |
6c271ce2 | 6953 | |
ae9428ca PB |
6954 | /* Commit SQ ring head once we've consumed and submitted all SQEs */ |
6955 | io_commit_sqring(ctx); | |
6956 | ||
6c271ce2 JA |
6957 | return submitted; |
6958 | } | |
6959 | ||
23b3628e XW |
6960 | static inline void io_ring_set_wakeup_flag(struct io_ring_ctx *ctx) |
6961 | { | |
6962 | /* Tell userspace we may need a wakeup call */ | |
6963 | spin_lock_irq(&ctx->completion_lock); | |
6964 | ctx->rings->sq_flags |= IORING_SQ_NEED_WAKEUP; | |
6965 | spin_unlock_irq(&ctx->completion_lock); | |
6966 | } | |
6967 | ||
6968 | static inline void io_ring_clear_wakeup_flag(struct io_ring_ctx *ctx) | |
6969 | { | |
6970 | spin_lock_irq(&ctx->completion_lock); | |
6971 | ctx->rings->sq_flags &= ~IORING_SQ_NEED_WAKEUP; | |
6972 | spin_unlock_irq(&ctx->completion_lock); | |
6973 | } | |
6974 | ||
08369246 | 6975 | static int __io_sq_thread(struct io_ring_ctx *ctx, bool cap_entries) |
6c271ce2 | 6976 | { |
c8d1ba58 | 6977 | unsigned int to_submit; |
bdcd3eab | 6978 | int ret = 0; |
6c271ce2 | 6979 | |
c8d1ba58 | 6980 | to_submit = io_sqring_entries(ctx); |
e95eee2d JA |
6981 | /* if we're handling multiple rings, cap submit size for fairness */ |
6982 | if (cap_entries && to_submit > 8) | |
6983 | to_submit = 8; | |
6984 | ||
906a3c6f | 6985 | if (!list_empty(&ctx->iopoll_list) || to_submit) { |
c8d1ba58 | 6986 | unsigned nr_events = 0; |
a4c0b3de | 6987 | |
c8d1ba58 | 6988 | mutex_lock(&ctx->uring_lock); |
906a3c6f | 6989 | if (!list_empty(&ctx->iopoll_list)) |
c8d1ba58 | 6990 | io_do_iopoll(ctx, &nr_events, 0); |
906a3c6f | 6991 | |
d9d05217 PB |
6992 | if (to_submit && !ctx->sqo_dead && |
6993 | likely(!percpu_ref_is_dying(&ctx->refs))) | |
08369246 | 6994 | ret = io_submit_sqes(ctx, to_submit); |
c8d1ba58 JA |
6995 | mutex_unlock(&ctx->uring_lock); |
6996 | } | |
6c271ce2 | 6997 | |
90554200 JA |
6998 | if (!io_sqring_full(ctx) && wq_has_sleeper(&ctx->sqo_sq_wait)) |
6999 | wake_up(&ctx->sqo_sq_wait); | |
6c271ce2 | 7000 | |
08369246 XW |
7001 | return ret; |
7002 | } | |
6c271ce2 | 7003 | |
08369246 XW |
7004 | static void io_sqd_update_thread_idle(struct io_sq_data *sqd) |
7005 | { | |
7006 | struct io_ring_ctx *ctx; | |
7007 | unsigned sq_thread_idle = 0; | |
6c271ce2 | 7008 | |
08369246 XW |
7009 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) { |
7010 | if (sq_thread_idle < ctx->sq_thread_idle) | |
7011 | sq_thread_idle = ctx->sq_thread_idle; | |
c8d1ba58 | 7012 | } |
c1edbf5f | 7013 | |
08369246 | 7014 | sqd->sq_thread_idle = sq_thread_idle; |
c8d1ba58 | 7015 | } |
6c271ce2 | 7016 | |
69fb2131 JA |
7017 | static void io_sqd_init_new(struct io_sq_data *sqd) |
7018 | { | |
7019 | struct io_ring_ctx *ctx; | |
7020 | ||
7021 | while (!list_empty(&sqd->ctx_new_list)) { | |
7022 | ctx = list_first_entry(&sqd->ctx_new_list, struct io_ring_ctx, sqd_list); | |
69fb2131 JA |
7023 | list_move_tail(&ctx->sqd_list, &sqd->ctx_list); |
7024 | complete(&ctx->sq_thread_comp); | |
7025 | } | |
08369246 XW |
7026 | |
7027 | io_sqd_update_thread_idle(sqd); | |
69fb2131 JA |
7028 | } |
7029 | ||
c8d1ba58 JA |
7030 | static int io_sq_thread(void *data) |
7031 | { | |
91d8f519 | 7032 | struct cgroup_subsys_state *cur_css = NULL; |
28cea78a JA |
7033 | struct files_struct *old_files = current->files; |
7034 | struct nsproxy *old_nsproxy = current->nsproxy; | |
69fb2131 JA |
7035 | const struct cred *old_cred = NULL; |
7036 | struct io_sq_data *sqd = data; | |
7037 | struct io_ring_ctx *ctx; | |
a0d9205f | 7038 | unsigned long timeout = 0; |
08369246 | 7039 | DEFINE_WAIT(wait); |
6c271ce2 | 7040 | |
28cea78a JA |
7041 | task_lock(current); |
7042 | current->files = NULL; | |
7043 | current->nsproxy = NULL; | |
7044 | task_unlock(current); | |
6c271ce2 | 7045 | |
69fb2131 | 7046 | while (!kthread_should_stop()) { |
08369246 XW |
7047 | int ret; |
7048 | bool cap_entries, sqt_spin, needs_sched; | |
c1edbf5f JA |
7049 | |
7050 | /* | |
69fb2131 JA |
7051 | * Any changes to the sqd lists are synchronized through the |
7052 | * kthread parking. This synchronizes the thread vs users, | |
7053 | * the users are synchronized on the sqd->ctx_lock. | |
c1edbf5f | 7054 | */ |
65b2b213 | 7055 | if (kthread_should_park()) { |
69fb2131 | 7056 | kthread_parkme(); |
65b2b213 XW |
7057 | /* |
7058 | * When sq thread is unparked, in case the previous park operation | |
7059 | * comes from io_put_sq_data(), which means that sq thread is going | |
7060 | * to be stopped, so here needs to have a check. | |
7061 | */ | |
7062 | if (kthread_should_stop()) | |
7063 | break; | |
7064 | } | |
7143b5ac | 7065 | |
08369246 | 7066 | if (unlikely(!list_empty(&sqd->ctx_new_list))) { |
69fb2131 | 7067 | io_sqd_init_new(sqd); |
08369246 XW |
7068 | timeout = jiffies + sqd->sq_thread_idle; |
7069 | } | |
6c271ce2 | 7070 | |
08369246 | 7071 | sqt_spin = false; |
e95eee2d | 7072 | cap_entries = !list_is_singular(&sqd->ctx_list); |
69fb2131 JA |
7073 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) { |
7074 | if (current->cred != ctx->creds) { | |
7075 | if (old_cred) | |
7076 | revert_creds(old_cred); | |
7077 | old_cred = override_creds(ctx->creds); | |
bdcd3eab | 7078 | } |
91d8f519 | 7079 | io_sq_thread_associate_blkcg(ctx, &cur_css); |
4ea33a97 JA |
7080 | #ifdef CONFIG_AUDIT |
7081 | current->loginuid = ctx->loginuid; | |
7082 | current->sessionid = ctx->sessionid; | |
7083 | #endif | |
bdcd3eab | 7084 | |
08369246 XW |
7085 | ret = __io_sq_thread(ctx, cap_entries); |
7086 | if (!sqt_spin && (ret > 0 || !list_empty(&ctx->iopoll_list))) | |
7087 | sqt_spin = true; | |
6c271ce2 | 7088 | |
28cea78a | 7089 | io_sq_thread_drop_mm_files(); |
69fb2131 | 7090 | } |
6c271ce2 | 7091 | |
08369246 | 7092 | if (sqt_spin || !time_after(jiffies, timeout)) { |
c8d1ba58 | 7093 | io_run_task_work(); |
d434ab6d | 7094 | io_sq_thread_drop_mm_files(); |
c8d1ba58 | 7095 | cond_resched(); |
08369246 XW |
7096 | if (sqt_spin) |
7097 | timeout = jiffies + sqd->sq_thread_idle; | |
7098 | continue; | |
7099 | } | |
7100 | ||
7101 | if (kthread_should_park()) | |
7102 | continue; | |
7103 | ||
7104 | needs_sched = true; | |
7105 | prepare_to_wait(&sqd->wait, &wait, TASK_INTERRUPTIBLE); | |
7106 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) { | |
7107 | if ((ctx->flags & IORING_SETUP_IOPOLL) && | |
7108 | !list_empty_careful(&ctx->iopoll_list)) { | |
7109 | needs_sched = false; | |
7110 | break; | |
7111 | } | |
7112 | if (io_sqring_entries(ctx)) { | |
7113 | needs_sched = false; | |
7114 | break; | |
7115 | } | |
7116 | } | |
7117 | ||
7118 | if (needs_sched) { | |
69fb2131 JA |
7119 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) |
7120 | io_ring_set_wakeup_flag(ctx); | |
08369246 | 7121 | |
69fb2131 | 7122 | schedule(); |
69fb2131 JA |
7123 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) |
7124 | io_ring_clear_wakeup_flag(ctx); | |
6c271ce2 | 7125 | } |
08369246 XW |
7126 | |
7127 | finish_wait(&sqd->wait, &wait); | |
7128 | timeout = jiffies + sqd->sq_thread_idle; | |
6c271ce2 JA |
7129 | } |
7130 | ||
4c6e277c | 7131 | io_run_task_work(); |
d434ab6d | 7132 | io_sq_thread_drop_mm_files(); |
b41e9852 | 7133 | |
91d8f519 DZ |
7134 | if (cur_css) |
7135 | io_sq_thread_unassociate_blkcg(); | |
69fb2131 JA |
7136 | if (old_cred) |
7137 | revert_creds(old_cred); | |
06058632 | 7138 | |
28cea78a JA |
7139 | task_lock(current); |
7140 | current->files = old_files; | |
7141 | current->nsproxy = old_nsproxy; | |
7142 | task_unlock(current); | |
7143 | ||
2bbcd6d3 | 7144 | kthread_parkme(); |
06058632 | 7145 | |
6c271ce2 JA |
7146 | return 0; |
7147 | } | |
7148 | ||
bda52162 JA |
7149 | struct io_wait_queue { |
7150 | struct wait_queue_entry wq; | |
7151 | struct io_ring_ctx *ctx; | |
7152 | unsigned to_wait; | |
7153 | unsigned nr_timeouts; | |
7154 | }; | |
7155 | ||
6c503150 | 7156 | static inline bool io_should_wake(struct io_wait_queue *iowq) |
bda52162 JA |
7157 | { |
7158 | struct io_ring_ctx *ctx = iowq->ctx; | |
7159 | ||
7160 | /* | |
d195a66e | 7161 | * Wake up if we have enough events, or if a timeout occurred since we |
bda52162 JA |
7162 | * started waiting. For timeouts, we always want to return to userspace, |
7163 | * regardless of event count. | |
7164 | */ | |
6c503150 | 7165 | return io_cqring_events(ctx) >= iowq->to_wait || |
bda52162 JA |
7166 | atomic_read(&ctx->cq_timeouts) != iowq->nr_timeouts; |
7167 | } | |
7168 | ||
7169 | static int io_wake_function(struct wait_queue_entry *curr, unsigned int mode, | |
7170 | int wake_flags, void *key) | |
7171 | { | |
7172 | struct io_wait_queue *iowq = container_of(curr, struct io_wait_queue, | |
7173 | wq); | |
7174 | ||
6c503150 PB |
7175 | /* |
7176 | * Cannot safely flush overflowed CQEs from here, ensure we wake up | |
7177 | * the task, and the next invocation will do it. | |
7178 | */ | |
7179 | if (io_should_wake(iowq) || test_bit(0, &iowq->ctx->cq_check_overflow)) | |
7180 | return autoremove_wake_function(curr, mode, wake_flags, key); | |
7181 | return -1; | |
bda52162 JA |
7182 | } |
7183 | ||
af9c1a44 JA |
7184 | static int io_run_task_work_sig(void) |
7185 | { | |
7186 | if (io_run_task_work()) | |
7187 | return 1; | |
7188 | if (!signal_pending(current)) | |
7189 | return 0; | |
792ee0f6 JA |
7190 | if (test_tsk_thread_flag(current, TIF_NOTIFY_SIGNAL)) |
7191 | return -ERESTARTSYS; | |
af9c1a44 JA |
7192 | return -EINTR; |
7193 | } | |
7194 | ||
2b188cc1 JA |
7195 | /* |
7196 | * Wait until events become available, if we don't already have some. The | |
7197 | * application must reap them itself, as they reside on the shared cq ring. | |
7198 | */ | |
7199 | static int io_cqring_wait(struct io_ring_ctx *ctx, int min_events, | |
c73ebb68 HX |
7200 | const sigset_t __user *sig, size_t sigsz, |
7201 | struct __kernel_timespec __user *uts) | |
2b188cc1 | 7202 | { |
bda52162 JA |
7203 | struct io_wait_queue iowq = { |
7204 | .wq = { | |
7205 | .private = current, | |
7206 | .func = io_wake_function, | |
7207 | .entry = LIST_HEAD_INIT(iowq.wq.entry), | |
7208 | }, | |
7209 | .ctx = ctx, | |
7210 | .to_wait = min_events, | |
7211 | }; | |
75b28aff | 7212 | struct io_rings *rings = ctx->rings; |
c73ebb68 HX |
7213 | struct timespec64 ts; |
7214 | signed long timeout = 0; | |
e9ffa5c2 | 7215 | int ret = 0; |
2b188cc1 | 7216 | |
b41e9852 | 7217 | do { |
6c503150 PB |
7218 | io_cqring_overflow_flush(ctx, false, NULL, NULL); |
7219 | if (io_cqring_events(ctx) >= min_events) | |
b41e9852 | 7220 | return 0; |
4c6e277c | 7221 | if (!io_run_task_work()) |
b41e9852 | 7222 | break; |
b41e9852 | 7223 | } while (1); |
2b188cc1 JA |
7224 | |
7225 | if (sig) { | |
9e75ad5d AB |
7226 | #ifdef CONFIG_COMPAT |
7227 | if (in_compat_syscall()) | |
7228 | ret = set_compat_user_sigmask((const compat_sigset_t __user *)sig, | |
b772434b | 7229 | sigsz); |
9e75ad5d AB |
7230 | else |
7231 | #endif | |
b772434b | 7232 | ret = set_user_sigmask(sig, sigsz); |
9e75ad5d | 7233 | |
2b188cc1 JA |
7234 | if (ret) |
7235 | return ret; | |
7236 | } | |
7237 | ||
c73ebb68 HX |
7238 | if (uts) { |
7239 | if (get_timespec64(&ts, uts)) | |
7240 | return -EFAULT; | |
7241 | timeout = timespec64_to_jiffies(&ts); | |
7242 | } | |
7243 | ||
bda52162 | 7244 | iowq.nr_timeouts = atomic_read(&ctx->cq_timeouts); |
c826bd7a | 7245 | trace_io_uring_cqring_wait(ctx, min_events); |
bda52162 | 7246 | do { |
6c503150 | 7247 | io_cqring_overflow_flush(ctx, false, NULL, NULL); |
bda52162 JA |
7248 | prepare_to_wait_exclusive(&ctx->wait, &iowq.wq, |
7249 | TASK_INTERRUPTIBLE); | |
ce593a6c | 7250 | /* make sure we run task_work before checking for signals */ |
af9c1a44 JA |
7251 | ret = io_run_task_work_sig(); |
7252 | if (ret > 0) | |
4c6e277c | 7253 | continue; |
af9c1a44 | 7254 | else if (ret < 0) |
bda52162 | 7255 | break; |
6c503150 | 7256 | if (io_should_wake(&iowq)) |
ce593a6c | 7257 | break; |
6c503150 PB |
7258 | if (test_bit(0, &ctx->cq_check_overflow)) |
7259 | continue; | |
c73ebb68 HX |
7260 | if (uts) { |
7261 | timeout = schedule_timeout(timeout); | |
7262 | if (timeout == 0) { | |
7263 | ret = -ETIME; | |
7264 | break; | |
7265 | } | |
7266 | } else { | |
7267 | schedule(); | |
7268 | } | |
bda52162 JA |
7269 | } while (1); |
7270 | finish_wait(&ctx->wait, &iowq.wq); | |
7271 | ||
b7db41c9 | 7272 | restore_saved_sigmask_unless(ret == -EINTR); |
2b188cc1 | 7273 | |
75b28aff | 7274 | return READ_ONCE(rings->cq.head) == READ_ONCE(rings->cq.tail) ? ret : 0; |
2b188cc1 JA |
7275 | } |
7276 | ||
6b06314c JA |
7277 | static void __io_sqe_files_unregister(struct io_ring_ctx *ctx) |
7278 | { | |
7279 | #if defined(CONFIG_UNIX) | |
7280 | if (ctx->ring_sock) { | |
7281 | struct sock *sock = ctx->ring_sock->sk; | |
7282 | struct sk_buff *skb; | |
7283 | ||
7284 | while ((skb = skb_dequeue(&sock->sk_receive_queue)) != NULL) | |
7285 | kfree_skb(skb); | |
7286 | } | |
7287 | #else | |
7288 | int i; | |
7289 | ||
65e19f54 JA |
7290 | for (i = 0; i < ctx->nr_user_files; i++) { |
7291 | struct file *file; | |
7292 | ||
7293 | file = io_file_from_index(ctx, i); | |
7294 | if (file) | |
7295 | fput(file); | |
7296 | } | |
6b06314c JA |
7297 | #endif |
7298 | } | |
7299 | ||
05f3fb3c JA |
7300 | static void io_file_ref_kill(struct percpu_ref *ref) |
7301 | { | |
7302 | struct fixed_file_data *data; | |
7303 | ||
7304 | data = container_of(ref, struct fixed_file_data, refs); | |
7305 | complete(&data->done); | |
7306 | } | |
7307 | ||
1642b445 PB |
7308 | static void io_sqe_files_set_node(struct fixed_file_data *file_data, |
7309 | struct fixed_file_ref_node *ref_node) | |
7310 | { | |
7311 | spin_lock_bh(&file_data->lock); | |
7312 | file_data->node = ref_node; | |
7313 | list_add_tail(&ref_node->node, &file_data->ref_list); | |
7314 | spin_unlock_bh(&file_data->lock); | |
7315 | percpu_ref_get(&file_data->refs); | |
7316 | } | |
7317 | ||
6b06314c JA |
7318 | static int io_sqe_files_unregister(struct io_ring_ctx *ctx) |
7319 | { | |
05f3fb3c | 7320 | struct fixed_file_data *data = ctx->file_data; |
1ffc5422 | 7321 | struct fixed_file_ref_node *backup_node, *ref_node = NULL; |
65e19f54 | 7322 | unsigned nr_tables, i; |
1ffc5422 | 7323 | int ret; |
65e19f54 | 7324 | |
05f3fb3c | 7325 | if (!data) |
6b06314c | 7326 | return -ENXIO; |
1ffc5422 PB |
7327 | backup_node = alloc_fixed_file_ref_node(ctx); |
7328 | if (!backup_node) | |
7329 | return -ENOMEM; | |
6b06314c | 7330 | |
ac0648a5 | 7331 | spin_lock_bh(&data->lock); |
1e5d770b | 7332 | ref_node = data->node; |
ac0648a5 | 7333 | spin_unlock_bh(&data->lock); |
05589553 XW |
7334 | if (ref_node) |
7335 | percpu_ref_kill(&ref_node->refs); | |
7336 | ||
7337 | percpu_ref_kill(&data->refs); | |
7338 | ||
7339 | /* wait for all refs nodes to complete */ | |
4a38aed2 | 7340 | flush_delayed_work(&ctx->file_put_work); |
1ffc5422 PB |
7341 | do { |
7342 | ret = wait_for_completion_interruptible(&data->done); | |
7343 | if (!ret) | |
7344 | break; | |
7345 | ret = io_run_task_work_sig(); | |
7346 | if (ret < 0) { | |
7347 | percpu_ref_resurrect(&data->refs); | |
7348 | reinit_completion(&data->done); | |
7349 | io_sqe_files_set_node(data, backup_node); | |
7350 | return ret; | |
7351 | } | |
7352 | } while (1); | |
05f3fb3c | 7353 | |
6b06314c | 7354 | __io_sqe_files_unregister(ctx); |
65e19f54 JA |
7355 | nr_tables = DIV_ROUND_UP(ctx->nr_user_files, IORING_MAX_FILES_TABLE); |
7356 | for (i = 0; i < nr_tables; i++) | |
05f3fb3c JA |
7357 | kfree(data->table[i].files); |
7358 | kfree(data->table); | |
05589553 XW |
7359 | percpu_ref_exit(&data->refs); |
7360 | kfree(data); | |
05f3fb3c | 7361 | ctx->file_data = NULL; |
6b06314c | 7362 | ctx->nr_user_files = 0; |
1ffc5422 | 7363 | destroy_fixed_file_ref_node(backup_node); |
6b06314c JA |
7364 | return 0; |
7365 | } | |
7366 | ||
534ca6d6 | 7367 | static void io_put_sq_data(struct io_sq_data *sqd) |
6c271ce2 | 7368 | { |
534ca6d6 | 7369 | if (refcount_dec_and_test(&sqd->refs)) { |
2bbcd6d3 RP |
7370 | /* |
7371 | * The park is a bit of a work-around, without it we get | |
7372 | * warning spews on shutdown with SQPOLL set and affinity | |
7373 | * set to a single CPU. | |
7374 | */ | |
534ca6d6 JA |
7375 | if (sqd->thread) { |
7376 | kthread_park(sqd->thread); | |
7377 | kthread_stop(sqd->thread); | |
7378 | } | |
7379 | ||
7380 | kfree(sqd); | |
7381 | } | |
7382 | } | |
7383 | ||
aa06165d JA |
7384 | static struct io_sq_data *io_attach_sq_data(struct io_uring_params *p) |
7385 | { | |
7386 | struct io_ring_ctx *ctx_attach; | |
7387 | struct io_sq_data *sqd; | |
7388 | struct fd f; | |
7389 | ||
7390 | f = fdget(p->wq_fd); | |
7391 | if (!f.file) | |
7392 | return ERR_PTR(-ENXIO); | |
7393 | if (f.file->f_op != &io_uring_fops) { | |
7394 | fdput(f); | |
7395 | return ERR_PTR(-EINVAL); | |
7396 | } | |
7397 | ||
7398 | ctx_attach = f.file->private_data; | |
7399 | sqd = ctx_attach->sq_data; | |
7400 | if (!sqd) { | |
7401 | fdput(f); | |
7402 | return ERR_PTR(-EINVAL); | |
7403 | } | |
7404 | ||
7405 | refcount_inc(&sqd->refs); | |
7406 | fdput(f); | |
7407 | return sqd; | |
7408 | } | |
7409 | ||
534ca6d6 JA |
7410 | static struct io_sq_data *io_get_sq_data(struct io_uring_params *p) |
7411 | { | |
7412 | struct io_sq_data *sqd; | |
7413 | ||
aa06165d JA |
7414 | if (p->flags & IORING_SETUP_ATTACH_WQ) |
7415 | return io_attach_sq_data(p); | |
7416 | ||
534ca6d6 JA |
7417 | sqd = kzalloc(sizeof(*sqd), GFP_KERNEL); |
7418 | if (!sqd) | |
7419 | return ERR_PTR(-ENOMEM); | |
7420 | ||
7421 | refcount_set(&sqd->refs, 1); | |
69fb2131 JA |
7422 | INIT_LIST_HEAD(&sqd->ctx_list); |
7423 | INIT_LIST_HEAD(&sqd->ctx_new_list); | |
7424 | mutex_init(&sqd->ctx_lock); | |
7425 | mutex_init(&sqd->lock); | |
534ca6d6 JA |
7426 | init_waitqueue_head(&sqd->wait); |
7427 | return sqd; | |
7428 | } | |
7429 | ||
69fb2131 JA |
7430 | static void io_sq_thread_unpark(struct io_sq_data *sqd) |
7431 | __releases(&sqd->lock) | |
7432 | { | |
7433 | if (!sqd->thread) | |
7434 | return; | |
7435 | kthread_unpark(sqd->thread); | |
7436 | mutex_unlock(&sqd->lock); | |
7437 | } | |
7438 | ||
7439 | static void io_sq_thread_park(struct io_sq_data *sqd) | |
7440 | __acquires(&sqd->lock) | |
7441 | { | |
7442 | if (!sqd->thread) | |
7443 | return; | |
7444 | mutex_lock(&sqd->lock); | |
7445 | kthread_park(sqd->thread); | |
7446 | } | |
7447 | ||
534ca6d6 JA |
7448 | static void io_sq_thread_stop(struct io_ring_ctx *ctx) |
7449 | { | |
7450 | struct io_sq_data *sqd = ctx->sq_data; | |
7451 | ||
7452 | if (sqd) { | |
7453 | if (sqd->thread) { | |
7454 | /* | |
7455 | * We may arrive here from the error branch in | |
7456 | * io_sq_offload_create() where the kthread is created | |
7457 | * without being waked up, thus wake it up now to make | |
7458 | * sure the wait will complete. | |
7459 | */ | |
7460 | wake_up_process(sqd->thread); | |
7461 | wait_for_completion(&ctx->sq_thread_comp); | |
69fb2131 JA |
7462 | |
7463 | io_sq_thread_park(sqd); | |
7464 | } | |
7465 | ||
7466 | mutex_lock(&sqd->ctx_lock); | |
7467 | list_del(&ctx->sqd_list); | |
08369246 | 7468 | io_sqd_update_thread_idle(sqd); |
69fb2131 JA |
7469 | mutex_unlock(&sqd->ctx_lock); |
7470 | ||
08369246 | 7471 | if (sqd->thread) |
69fb2131 | 7472 | io_sq_thread_unpark(sqd); |
534ca6d6 JA |
7473 | |
7474 | io_put_sq_data(sqd); | |
7475 | ctx->sq_data = NULL; | |
6c271ce2 JA |
7476 | } |
7477 | } | |
7478 | ||
6b06314c JA |
7479 | static void io_finish_async(struct io_ring_ctx *ctx) |
7480 | { | |
6c271ce2 JA |
7481 | io_sq_thread_stop(ctx); |
7482 | ||
561fb04a JA |
7483 | if (ctx->io_wq) { |
7484 | io_wq_destroy(ctx->io_wq); | |
7485 | ctx->io_wq = NULL; | |
6b06314c JA |
7486 | } |
7487 | } | |
7488 | ||
7489 | #if defined(CONFIG_UNIX) | |
6b06314c JA |
7490 | /* |
7491 | * Ensure the UNIX gc is aware of our file set, so we are certain that | |
7492 | * the io_uring can be safely unregistered on process exit, even if we have | |
7493 | * loops in the file referencing. | |
7494 | */ | |
7495 | static int __io_sqe_files_scm(struct io_ring_ctx *ctx, int nr, int offset) | |
7496 | { | |
7497 | struct sock *sk = ctx->ring_sock->sk; | |
7498 | struct scm_fp_list *fpl; | |
7499 | struct sk_buff *skb; | |
08a45173 | 7500 | int i, nr_files; |
6b06314c | 7501 | |
6b06314c JA |
7502 | fpl = kzalloc(sizeof(*fpl), GFP_KERNEL); |
7503 | if (!fpl) | |
7504 | return -ENOMEM; | |
7505 | ||
7506 | skb = alloc_skb(0, GFP_KERNEL); | |
7507 | if (!skb) { | |
7508 | kfree(fpl); | |
7509 | return -ENOMEM; | |
7510 | } | |
7511 | ||
7512 | skb->sk = sk; | |
6b06314c | 7513 | |
08a45173 | 7514 | nr_files = 0; |
6b06314c JA |
7515 | fpl->user = get_uid(ctx->user); |
7516 | for (i = 0; i < nr; i++) { | |
65e19f54 JA |
7517 | struct file *file = io_file_from_index(ctx, i + offset); |
7518 | ||
7519 | if (!file) | |
08a45173 | 7520 | continue; |
65e19f54 | 7521 | fpl->fp[nr_files] = get_file(file); |
08a45173 JA |
7522 | unix_inflight(fpl->user, fpl->fp[nr_files]); |
7523 | nr_files++; | |
6b06314c JA |
7524 | } |
7525 | ||
08a45173 JA |
7526 | if (nr_files) { |
7527 | fpl->max = SCM_MAX_FD; | |
7528 | fpl->count = nr_files; | |
7529 | UNIXCB(skb).fp = fpl; | |
05f3fb3c | 7530 | skb->destructor = unix_destruct_scm; |
08a45173 JA |
7531 | refcount_add(skb->truesize, &sk->sk_wmem_alloc); |
7532 | skb_queue_head(&sk->sk_receive_queue, skb); | |
6b06314c | 7533 | |
08a45173 JA |
7534 | for (i = 0; i < nr_files; i++) |
7535 | fput(fpl->fp[i]); | |
7536 | } else { | |
7537 | kfree_skb(skb); | |
7538 | kfree(fpl); | |
7539 | } | |
6b06314c JA |
7540 | |
7541 | return 0; | |
7542 | } | |
7543 | ||
7544 | /* | |
7545 | * If UNIX sockets are enabled, fd passing can cause a reference cycle which | |
7546 | * causes regular reference counting to break down. We rely on the UNIX | |
7547 | * garbage collection to take care of this problem for us. | |
7548 | */ | |
7549 | static int io_sqe_files_scm(struct io_ring_ctx *ctx) | |
7550 | { | |
7551 | unsigned left, total; | |
7552 | int ret = 0; | |
7553 | ||
7554 | total = 0; | |
7555 | left = ctx->nr_user_files; | |
7556 | while (left) { | |
7557 | unsigned this_files = min_t(unsigned, left, SCM_MAX_FD); | |
6b06314c JA |
7558 | |
7559 | ret = __io_sqe_files_scm(ctx, this_files, total); | |
7560 | if (ret) | |
7561 | break; | |
7562 | left -= this_files; | |
7563 | total += this_files; | |
7564 | } | |
7565 | ||
7566 | if (!ret) | |
7567 | return 0; | |
7568 | ||
7569 | while (total < ctx->nr_user_files) { | |
65e19f54 JA |
7570 | struct file *file = io_file_from_index(ctx, total); |
7571 | ||
7572 | if (file) | |
7573 | fput(file); | |
6b06314c JA |
7574 | total++; |
7575 | } | |
7576 | ||
7577 | return ret; | |
7578 | } | |
7579 | #else | |
7580 | static int io_sqe_files_scm(struct io_ring_ctx *ctx) | |
7581 | { | |
7582 | return 0; | |
7583 | } | |
7584 | #endif | |
7585 | ||
5398ae69 PB |
7586 | static int io_sqe_alloc_file_tables(struct fixed_file_data *file_data, |
7587 | unsigned nr_tables, unsigned nr_files) | |
65e19f54 JA |
7588 | { |
7589 | int i; | |
7590 | ||
7591 | for (i = 0; i < nr_tables; i++) { | |
5398ae69 | 7592 | struct fixed_file_table *table = &file_data->table[i]; |
65e19f54 JA |
7593 | unsigned this_files; |
7594 | ||
7595 | this_files = min(nr_files, IORING_MAX_FILES_TABLE); | |
7596 | table->files = kcalloc(this_files, sizeof(struct file *), | |
7597 | GFP_KERNEL); | |
7598 | if (!table->files) | |
7599 | break; | |
7600 | nr_files -= this_files; | |
7601 | } | |
7602 | ||
7603 | if (i == nr_tables) | |
7604 | return 0; | |
7605 | ||
7606 | for (i = 0; i < nr_tables; i++) { | |
5398ae69 | 7607 | struct fixed_file_table *table = &file_data->table[i]; |
65e19f54 JA |
7608 | kfree(table->files); |
7609 | } | |
7610 | return 1; | |
7611 | } | |
7612 | ||
05f3fb3c JA |
7613 | static void io_ring_file_put(struct io_ring_ctx *ctx, struct file *file) |
7614 | { | |
7615 | #if defined(CONFIG_UNIX) | |
7616 | struct sock *sock = ctx->ring_sock->sk; | |
7617 | struct sk_buff_head list, *head = &sock->sk_receive_queue; | |
7618 | struct sk_buff *skb; | |
7619 | int i; | |
7620 | ||
7621 | __skb_queue_head_init(&list); | |
7622 | ||
7623 | /* | |
7624 | * Find the skb that holds this file in its SCM_RIGHTS. When found, | |
7625 | * remove this entry and rearrange the file array. | |
7626 | */ | |
7627 | skb = skb_dequeue(head); | |
7628 | while (skb) { | |
7629 | struct scm_fp_list *fp; | |
7630 | ||
7631 | fp = UNIXCB(skb).fp; | |
7632 | for (i = 0; i < fp->count; i++) { | |
7633 | int left; | |
7634 | ||
7635 | if (fp->fp[i] != file) | |
7636 | continue; | |
7637 | ||
7638 | unix_notinflight(fp->user, fp->fp[i]); | |
7639 | left = fp->count - 1 - i; | |
7640 | if (left) { | |
7641 | memmove(&fp->fp[i], &fp->fp[i + 1], | |
7642 | left * sizeof(struct file *)); | |
7643 | } | |
7644 | fp->count--; | |
7645 | if (!fp->count) { | |
7646 | kfree_skb(skb); | |
7647 | skb = NULL; | |
7648 | } else { | |
7649 | __skb_queue_tail(&list, skb); | |
7650 | } | |
7651 | fput(file); | |
7652 | file = NULL; | |
7653 | break; | |
7654 | } | |
7655 | ||
7656 | if (!file) | |
7657 | break; | |
7658 | ||
7659 | __skb_queue_tail(&list, skb); | |
7660 | ||
7661 | skb = skb_dequeue(head); | |
7662 | } | |
7663 | ||
7664 | if (skb_peek(&list)) { | |
7665 | spin_lock_irq(&head->lock); | |
7666 | while ((skb = __skb_dequeue(&list)) != NULL) | |
7667 | __skb_queue_tail(head, skb); | |
7668 | spin_unlock_irq(&head->lock); | |
7669 | } | |
7670 | #else | |
7671 | fput(file); | |
7672 | #endif | |
7673 | } | |
7674 | ||
7675 | struct io_file_put { | |
05589553 | 7676 | struct list_head list; |
05f3fb3c | 7677 | struct file *file; |
05f3fb3c JA |
7678 | }; |
7679 | ||
4a38aed2 | 7680 | static void __io_file_put_work(struct fixed_file_ref_node *ref_node) |
65e19f54 | 7681 | { |
4a38aed2 JA |
7682 | struct fixed_file_data *file_data = ref_node->file_data; |
7683 | struct io_ring_ctx *ctx = file_data->ctx; | |
05f3fb3c | 7684 | struct io_file_put *pfile, *tmp; |
05589553 XW |
7685 | |
7686 | list_for_each_entry_safe(pfile, tmp, &ref_node->file_list, list) { | |
6a4d07cd | 7687 | list_del(&pfile->list); |
05589553 XW |
7688 | io_ring_file_put(ctx, pfile->file); |
7689 | kfree(pfile); | |
65e19f54 | 7690 | } |
05589553 | 7691 | |
05589553 XW |
7692 | percpu_ref_exit(&ref_node->refs); |
7693 | kfree(ref_node); | |
7694 | percpu_ref_put(&file_data->refs); | |
2faf852d | 7695 | } |
65e19f54 | 7696 | |
4a38aed2 JA |
7697 | static void io_file_put_work(struct work_struct *work) |
7698 | { | |
7699 | struct io_ring_ctx *ctx; | |
7700 | struct llist_node *node; | |
7701 | ||
7702 | ctx = container_of(work, struct io_ring_ctx, file_put_work.work); | |
7703 | node = llist_del_all(&ctx->file_put_llist); | |
7704 | ||
7705 | while (node) { | |
7706 | struct fixed_file_ref_node *ref_node; | |
7707 | struct llist_node *next = node->next; | |
7708 | ||
7709 | ref_node = llist_entry(node, struct fixed_file_ref_node, llist); | |
7710 | __io_file_put_work(ref_node); | |
7711 | node = next; | |
7712 | } | |
7713 | } | |
7714 | ||
05589553 | 7715 | static void io_file_data_ref_zero(struct percpu_ref *ref) |
2faf852d | 7716 | { |
05589553 | 7717 | struct fixed_file_ref_node *ref_node; |
e297822b | 7718 | struct fixed_file_data *data; |
4a38aed2 | 7719 | struct io_ring_ctx *ctx; |
e297822b | 7720 | bool first_add = false; |
4a38aed2 | 7721 | int delay = HZ; |
65e19f54 | 7722 | |
05589553 | 7723 | ref_node = container_of(ref, struct fixed_file_ref_node, refs); |
e297822b PB |
7724 | data = ref_node->file_data; |
7725 | ctx = data->ctx; | |
7726 | ||
ac0648a5 | 7727 | spin_lock_bh(&data->lock); |
e297822b PB |
7728 | ref_node->done = true; |
7729 | ||
7730 | while (!list_empty(&data->ref_list)) { | |
7731 | ref_node = list_first_entry(&data->ref_list, | |
7732 | struct fixed_file_ref_node, node); | |
7733 | /* recycle ref nodes in order */ | |
7734 | if (!ref_node->done) | |
7735 | break; | |
7736 | list_del(&ref_node->node); | |
7737 | first_add |= llist_add(&ref_node->llist, &ctx->file_put_llist); | |
7738 | } | |
ac0648a5 | 7739 | spin_unlock_bh(&data->lock); |
05589553 | 7740 | |
e297822b | 7741 | if (percpu_ref_is_dying(&data->refs)) |
4a38aed2 | 7742 | delay = 0; |
05589553 | 7743 | |
4a38aed2 JA |
7744 | if (!delay) |
7745 | mod_delayed_work(system_wq, &ctx->file_put_work, 0); | |
7746 | else if (first_add) | |
7747 | queue_delayed_work(system_wq, &ctx->file_put_work, delay); | |
05f3fb3c | 7748 | } |
65e19f54 | 7749 | |
05589553 XW |
7750 | static struct fixed_file_ref_node *alloc_fixed_file_ref_node( |
7751 | struct io_ring_ctx *ctx) | |
05f3fb3c | 7752 | { |
05589553 | 7753 | struct fixed_file_ref_node *ref_node; |
05f3fb3c | 7754 | |
05589553 XW |
7755 | ref_node = kzalloc(sizeof(*ref_node), GFP_KERNEL); |
7756 | if (!ref_node) | |
3e2224c5 | 7757 | return NULL; |
05f3fb3c | 7758 | |
05589553 XW |
7759 | if (percpu_ref_init(&ref_node->refs, io_file_data_ref_zero, |
7760 | 0, GFP_KERNEL)) { | |
7761 | kfree(ref_node); | |
3e2224c5 | 7762 | return NULL; |
05589553 XW |
7763 | } |
7764 | INIT_LIST_HEAD(&ref_node->node); | |
7765 | INIT_LIST_HEAD(&ref_node->file_list); | |
05589553 | 7766 | ref_node->file_data = ctx->file_data; |
e297822b | 7767 | ref_node->done = false; |
05589553 | 7768 | return ref_node; |
05589553 XW |
7769 | } |
7770 | ||
7771 | static void destroy_fixed_file_ref_node(struct fixed_file_ref_node *ref_node) | |
7772 | { | |
7773 | percpu_ref_exit(&ref_node->refs); | |
7774 | kfree(ref_node); | |
65e19f54 JA |
7775 | } |
7776 | ||
6b06314c JA |
7777 | static int io_sqe_files_register(struct io_ring_ctx *ctx, void __user *arg, |
7778 | unsigned nr_args) | |
7779 | { | |
7780 | __s32 __user *fds = (__s32 __user *) arg; | |
600cf3f8 | 7781 | unsigned nr_tables, i; |
05f3fb3c | 7782 | struct file *file; |
600cf3f8 | 7783 | int fd, ret = -ENOMEM; |
05589553 | 7784 | struct fixed_file_ref_node *ref_node; |
5398ae69 | 7785 | struct fixed_file_data *file_data; |
6b06314c | 7786 | |
05f3fb3c | 7787 | if (ctx->file_data) |
6b06314c JA |
7788 | return -EBUSY; |
7789 | if (!nr_args) | |
7790 | return -EINVAL; | |
7791 | if (nr_args > IORING_MAX_FIXED_FILES) | |
7792 | return -EMFILE; | |
7793 | ||
5398ae69 PB |
7794 | file_data = kzalloc(sizeof(*ctx->file_data), GFP_KERNEL); |
7795 | if (!file_data) | |
05f3fb3c | 7796 | return -ENOMEM; |
5398ae69 PB |
7797 | file_data->ctx = ctx; |
7798 | init_completion(&file_data->done); | |
7799 | INIT_LIST_HEAD(&file_data->ref_list); | |
7800 | spin_lock_init(&file_data->lock); | |
05f3fb3c | 7801 | |
65e19f54 | 7802 | nr_tables = DIV_ROUND_UP(nr_args, IORING_MAX_FILES_TABLE); |
035fbafc | 7803 | file_data->table = kcalloc(nr_tables, sizeof(*file_data->table), |
5398ae69 | 7804 | GFP_KERNEL); |
600cf3f8 PB |
7805 | if (!file_data->table) |
7806 | goto out_free; | |
05f3fb3c | 7807 | |
5398ae69 | 7808 | if (percpu_ref_init(&file_data->refs, io_file_ref_kill, |
600cf3f8 PB |
7809 | PERCPU_REF_ALLOW_REINIT, GFP_KERNEL)) |
7810 | goto out_free; | |
6b06314c | 7811 | |
600cf3f8 PB |
7812 | if (io_sqe_alloc_file_tables(file_data, nr_tables, nr_args)) |
7813 | goto out_ref; | |
55cbc256 | 7814 | ctx->file_data = file_data; |
65e19f54 | 7815 | |
08a45173 | 7816 | for (i = 0; i < nr_args; i++, ctx->nr_user_files++) { |
65e19f54 JA |
7817 | struct fixed_file_table *table; |
7818 | unsigned index; | |
7819 | ||
600cf3f8 PB |
7820 | if (copy_from_user(&fd, &fds[i], sizeof(fd))) { |
7821 | ret = -EFAULT; | |
7822 | goto out_fput; | |
7823 | } | |
08a45173 | 7824 | /* allow sparse sets */ |
600cf3f8 | 7825 | if (fd == -1) |
08a45173 | 7826 | continue; |
6b06314c | 7827 | |
05f3fb3c | 7828 | file = fget(fd); |
6b06314c | 7829 | ret = -EBADF; |
05f3fb3c | 7830 | if (!file) |
600cf3f8 | 7831 | goto out_fput; |
05f3fb3c | 7832 | |
6b06314c JA |
7833 | /* |
7834 | * Don't allow io_uring instances to be registered. If UNIX | |
7835 | * isn't enabled, then this causes a reference cycle and this | |
7836 | * instance can never get freed. If UNIX is enabled we'll | |
7837 | * handle it just fine, but there's still no point in allowing | |
7838 | * a ring fd as it doesn't support regular read/write anyway. | |
7839 | */ | |
05f3fb3c JA |
7840 | if (file->f_op == &io_uring_fops) { |
7841 | fput(file); | |
600cf3f8 | 7842 | goto out_fput; |
6b06314c | 7843 | } |
600cf3f8 PB |
7844 | table = &file_data->table[i >> IORING_FILE_TABLE_SHIFT]; |
7845 | index = i & IORING_FILE_TABLE_MASK; | |
05f3fb3c | 7846 | table->files[index] = file; |
6b06314c JA |
7847 | } |
7848 | ||
6b06314c | 7849 | ret = io_sqe_files_scm(ctx); |
05589553 | 7850 | if (ret) { |
6b06314c | 7851 | io_sqe_files_unregister(ctx); |
05589553 XW |
7852 | return ret; |
7853 | } | |
6b06314c | 7854 | |
05589553 | 7855 | ref_node = alloc_fixed_file_ref_node(ctx); |
3e2224c5 | 7856 | if (!ref_node) { |
05589553 | 7857 | io_sqe_files_unregister(ctx); |
3e2224c5 | 7858 | return -ENOMEM; |
05589553 XW |
7859 | } |
7860 | ||
1642b445 | 7861 | io_sqe_files_set_node(file_data, ref_node); |
6b06314c | 7862 | return ret; |
600cf3f8 PB |
7863 | out_fput: |
7864 | for (i = 0; i < ctx->nr_user_files; i++) { | |
7865 | file = io_file_from_index(ctx, i); | |
7866 | if (file) | |
7867 | fput(file); | |
7868 | } | |
7869 | for (i = 0; i < nr_tables; i++) | |
7870 | kfree(file_data->table[i].files); | |
7871 | ctx->nr_user_files = 0; | |
7872 | out_ref: | |
7873 | percpu_ref_exit(&file_data->refs); | |
7874 | out_free: | |
7875 | kfree(file_data->table); | |
7876 | kfree(file_data); | |
55cbc256 | 7877 | ctx->file_data = NULL; |
6b06314c JA |
7878 | return ret; |
7879 | } | |
7880 | ||
c3a31e60 JA |
7881 | static int io_sqe_file_register(struct io_ring_ctx *ctx, struct file *file, |
7882 | int index) | |
7883 | { | |
7884 | #if defined(CONFIG_UNIX) | |
7885 | struct sock *sock = ctx->ring_sock->sk; | |
7886 | struct sk_buff_head *head = &sock->sk_receive_queue; | |
7887 | struct sk_buff *skb; | |
7888 | ||
7889 | /* | |
7890 | * See if we can merge this file into an existing skb SCM_RIGHTS | |
7891 | * file set. If there's no room, fall back to allocating a new skb | |
7892 | * and filling it in. | |
7893 | */ | |
7894 | spin_lock_irq(&head->lock); | |
7895 | skb = skb_peek(head); | |
7896 | if (skb) { | |
7897 | struct scm_fp_list *fpl = UNIXCB(skb).fp; | |
7898 | ||
7899 | if (fpl->count < SCM_MAX_FD) { | |
7900 | __skb_unlink(skb, head); | |
7901 | spin_unlock_irq(&head->lock); | |
7902 | fpl->fp[fpl->count] = get_file(file); | |
7903 | unix_inflight(fpl->user, fpl->fp[fpl->count]); | |
7904 | fpl->count++; | |
7905 | spin_lock_irq(&head->lock); | |
7906 | __skb_queue_head(head, skb); | |
7907 | } else { | |
7908 | skb = NULL; | |
7909 | } | |
7910 | } | |
7911 | spin_unlock_irq(&head->lock); | |
7912 | ||
7913 | if (skb) { | |
7914 | fput(file); | |
7915 | return 0; | |
7916 | } | |
7917 | ||
7918 | return __io_sqe_files_scm(ctx, 1, index); | |
7919 | #else | |
7920 | return 0; | |
7921 | #endif | |
7922 | } | |
7923 | ||
a5318d3c | 7924 | static int io_queue_file_removal(struct fixed_file_data *data, |
05589553 | 7925 | struct file *file) |
05f3fb3c | 7926 | { |
a5318d3c | 7927 | struct io_file_put *pfile; |
b2e96852 | 7928 | struct fixed_file_ref_node *ref_node = data->node; |
05f3fb3c | 7929 | |
05f3fb3c | 7930 | pfile = kzalloc(sizeof(*pfile), GFP_KERNEL); |
a5318d3c HD |
7931 | if (!pfile) |
7932 | return -ENOMEM; | |
05f3fb3c JA |
7933 | |
7934 | pfile->file = file; | |
05589553 XW |
7935 | list_add(&pfile->list, &ref_node->file_list); |
7936 | ||
a5318d3c | 7937 | return 0; |
05f3fb3c JA |
7938 | } |
7939 | ||
7940 | static int __io_sqe_files_update(struct io_ring_ctx *ctx, | |
7941 | struct io_uring_files_update *up, | |
7942 | unsigned nr_args) | |
7943 | { | |
7944 | struct fixed_file_data *data = ctx->file_data; | |
05589553 | 7945 | struct fixed_file_ref_node *ref_node; |
05f3fb3c | 7946 | struct file *file; |
c3a31e60 JA |
7947 | __s32 __user *fds; |
7948 | int fd, i, err; | |
7949 | __u32 done; | |
05589553 | 7950 | bool needs_switch = false; |
c3a31e60 | 7951 | |
05f3fb3c | 7952 | if (check_add_overflow(up->offset, nr_args, &done)) |
c3a31e60 JA |
7953 | return -EOVERFLOW; |
7954 | if (done > ctx->nr_user_files) | |
7955 | return -EINVAL; | |
7956 | ||
05589553 | 7957 | ref_node = alloc_fixed_file_ref_node(ctx); |
3e2224c5 MWO |
7958 | if (!ref_node) |
7959 | return -ENOMEM; | |
05589553 | 7960 | |
c3a31e60 | 7961 | done = 0; |
05f3fb3c | 7962 | fds = u64_to_user_ptr(up->fds); |
c3a31e60 | 7963 | while (nr_args) { |
65e19f54 JA |
7964 | struct fixed_file_table *table; |
7965 | unsigned index; | |
7966 | ||
c3a31e60 JA |
7967 | err = 0; |
7968 | if (copy_from_user(&fd, &fds[done], sizeof(fd))) { | |
7969 | err = -EFAULT; | |
7970 | break; | |
7971 | } | |
05f3fb3c JA |
7972 | i = array_index_nospec(up->offset, ctx->nr_user_files); |
7973 | table = &ctx->file_data->table[i >> IORING_FILE_TABLE_SHIFT]; | |
65e19f54 JA |
7974 | index = i & IORING_FILE_TABLE_MASK; |
7975 | if (table->files[index]) { | |
98dfd502 | 7976 | file = table->files[index]; |
a5318d3c HD |
7977 | err = io_queue_file_removal(data, file); |
7978 | if (err) | |
7979 | break; | |
65e19f54 | 7980 | table->files[index] = NULL; |
05589553 | 7981 | needs_switch = true; |
c3a31e60 JA |
7982 | } |
7983 | if (fd != -1) { | |
c3a31e60 JA |
7984 | file = fget(fd); |
7985 | if (!file) { | |
7986 | err = -EBADF; | |
7987 | break; | |
7988 | } | |
7989 | /* | |
7990 | * Don't allow io_uring instances to be registered. If | |
7991 | * UNIX isn't enabled, then this causes a reference | |
7992 | * cycle and this instance can never get freed. If UNIX | |
7993 | * is enabled we'll handle it just fine, but there's | |
7994 | * still no point in allowing a ring fd as it doesn't | |
7995 | * support regular read/write anyway. | |
7996 | */ | |
7997 | if (file->f_op == &io_uring_fops) { | |
7998 | fput(file); | |
7999 | err = -EBADF; | |
8000 | break; | |
8001 | } | |
65e19f54 | 8002 | table->files[index] = file; |
c3a31e60 | 8003 | err = io_sqe_file_register(ctx, file, i); |
f3bd9dae | 8004 | if (err) { |
95d1c8e5 | 8005 | table->files[index] = NULL; |
f3bd9dae | 8006 | fput(file); |
c3a31e60 | 8007 | break; |
f3bd9dae | 8008 | } |
c3a31e60 JA |
8009 | } |
8010 | nr_args--; | |
8011 | done++; | |
05f3fb3c JA |
8012 | up->offset++; |
8013 | } | |
8014 | ||
05589553 | 8015 | if (needs_switch) { |
b2e96852 | 8016 | percpu_ref_kill(&data->node->refs); |
1642b445 | 8017 | io_sqe_files_set_node(data, ref_node); |
05589553 XW |
8018 | } else |
8019 | destroy_fixed_file_ref_node(ref_node); | |
c3a31e60 JA |
8020 | |
8021 | return done ? done : err; | |
8022 | } | |
05589553 | 8023 | |
05f3fb3c JA |
8024 | static int io_sqe_files_update(struct io_ring_ctx *ctx, void __user *arg, |
8025 | unsigned nr_args) | |
8026 | { | |
8027 | struct io_uring_files_update up; | |
8028 | ||
8029 | if (!ctx->file_data) | |
8030 | return -ENXIO; | |
8031 | if (!nr_args) | |
8032 | return -EINVAL; | |
8033 | if (copy_from_user(&up, arg, sizeof(up))) | |
8034 | return -EFAULT; | |
8035 | if (up.resv) | |
8036 | return -EINVAL; | |
8037 | ||
8038 | return __io_sqe_files_update(ctx, &up, nr_args); | |
8039 | } | |
c3a31e60 | 8040 | |
e9fd9396 | 8041 | static void io_free_work(struct io_wq_work *work) |
7d723065 JA |
8042 | { |
8043 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); | |
8044 | ||
e9fd9396 | 8045 | /* Consider that io_steal_work() relies on this ref */ |
7d723065 JA |
8046 | io_put_req(req); |
8047 | } | |
8048 | ||
24369c2e PB |
8049 | static int io_init_wq_offload(struct io_ring_ctx *ctx, |
8050 | struct io_uring_params *p) | |
8051 | { | |
8052 | struct io_wq_data data; | |
8053 | struct fd f; | |
8054 | struct io_ring_ctx *ctx_attach; | |
8055 | unsigned int concurrency; | |
8056 | int ret = 0; | |
8057 | ||
8058 | data.user = ctx->user; | |
e9fd9396 | 8059 | data.free_work = io_free_work; |
f5fa38c5 | 8060 | data.do_work = io_wq_submit_work; |
24369c2e PB |
8061 | |
8062 | if (!(p->flags & IORING_SETUP_ATTACH_WQ)) { | |
8063 | /* Do QD, or 4 * CPUS, whatever is smallest */ | |
8064 | concurrency = min(ctx->sq_entries, 4 * num_online_cpus()); | |
8065 | ||
8066 | ctx->io_wq = io_wq_create(concurrency, &data); | |
8067 | if (IS_ERR(ctx->io_wq)) { | |
8068 | ret = PTR_ERR(ctx->io_wq); | |
8069 | ctx->io_wq = NULL; | |
8070 | } | |
8071 | return ret; | |
8072 | } | |
8073 | ||
8074 | f = fdget(p->wq_fd); | |
8075 | if (!f.file) | |
8076 | return -EBADF; | |
8077 | ||
8078 | if (f.file->f_op != &io_uring_fops) { | |
8079 | ret = -EINVAL; | |
8080 | goto out_fput; | |
8081 | } | |
8082 | ||
8083 | ctx_attach = f.file->private_data; | |
8084 | /* @io_wq is protected by holding the fd */ | |
8085 | if (!io_wq_get(ctx_attach->io_wq, &data)) { | |
8086 | ret = -EINVAL; | |
8087 | goto out_fput; | |
8088 | } | |
8089 | ||
8090 | ctx->io_wq = ctx_attach->io_wq; | |
8091 | out_fput: | |
8092 | fdput(f); | |
8093 | return ret; | |
8094 | } | |
8095 | ||
0f212204 JA |
8096 | static int io_uring_alloc_task_context(struct task_struct *task) |
8097 | { | |
8098 | struct io_uring_task *tctx; | |
d8a6df10 | 8099 | int ret; |
0f212204 JA |
8100 | |
8101 | tctx = kmalloc(sizeof(*tctx), GFP_KERNEL); | |
8102 | if (unlikely(!tctx)) | |
8103 | return -ENOMEM; | |
8104 | ||
d8a6df10 JA |
8105 | ret = percpu_counter_init(&tctx->inflight, 0, GFP_KERNEL); |
8106 | if (unlikely(ret)) { | |
8107 | kfree(tctx); | |
8108 | return ret; | |
8109 | } | |
8110 | ||
0f212204 JA |
8111 | xa_init(&tctx->xa); |
8112 | init_waitqueue_head(&tctx->wait); | |
8113 | tctx->last = NULL; | |
fdaf083c JA |
8114 | atomic_set(&tctx->in_idle, 0); |
8115 | tctx->sqpoll = false; | |
500a373d JA |
8116 | io_init_identity(&tctx->__identity); |
8117 | tctx->identity = &tctx->__identity; | |
0f212204 JA |
8118 | task->io_uring = tctx; |
8119 | return 0; | |
8120 | } | |
8121 | ||
8122 | void __io_uring_free(struct task_struct *tsk) | |
8123 | { | |
8124 | struct io_uring_task *tctx = tsk->io_uring; | |
8125 | ||
8126 | WARN_ON_ONCE(!xa_empty(&tctx->xa)); | |
500a373d JA |
8127 | WARN_ON_ONCE(refcount_read(&tctx->identity->count) != 1); |
8128 | if (tctx->identity != &tctx->__identity) | |
8129 | kfree(tctx->identity); | |
d8a6df10 | 8130 | percpu_counter_destroy(&tctx->inflight); |
0f212204 JA |
8131 | kfree(tctx); |
8132 | tsk->io_uring = NULL; | |
8133 | } | |
8134 | ||
7e84e1c7 SG |
8135 | static int io_sq_offload_create(struct io_ring_ctx *ctx, |
8136 | struct io_uring_params *p) | |
2b188cc1 JA |
8137 | { |
8138 | int ret; | |
8139 | ||
6c271ce2 | 8140 | if (ctx->flags & IORING_SETUP_SQPOLL) { |
534ca6d6 JA |
8141 | struct io_sq_data *sqd; |
8142 | ||
3ec482d1 | 8143 | ret = -EPERM; |
ce59fc69 | 8144 | if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_NICE)) |
3ec482d1 JA |
8145 | goto err; |
8146 | ||
534ca6d6 JA |
8147 | sqd = io_get_sq_data(p); |
8148 | if (IS_ERR(sqd)) { | |
8149 | ret = PTR_ERR(sqd); | |
8150 | goto err; | |
8151 | } | |
69fb2131 | 8152 | |
534ca6d6 | 8153 | ctx->sq_data = sqd; |
69fb2131 JA |
8154 | io_sq_thread_park(sqd); |
8155 | mutex_lock(&sqd->ctx_lock); | |
8156 | list_add(&ctx->sqd_list, &sqd->ctx_new_list); | |
8157 | mutex_unlock(&sqd->ctx_lock); | |
8158 | io_sq_thread_unpark(sqd); | |
534ca6d6 | 8159 | |
917257da JA |
8160 | ctx->sq_thread_idle = msecs_to_jiffies(p->sq_thread_idle); |
8161 | if (!ctx->sq_thread_idle) | |
8162 | ctx->sq_thread_idle = HZ; | |
8163 | ||
aa06165d JA |
8164 | if (sqd->thread) |
8165 | goto done; | |
8166 | ||
6c271ce2 | 8167 | if (p->flags & IORING_SETUP_SQ_AFF) { |
44a9bd18 | 8168 | int cpu = p->sq_thread_cpu; |
6c271ce2 | 8169 | |
917257da | 8170 | ret = -EINVAL; |
44a9bd18 JA |
8171 | if (cpu >= nr_cpu_ids) |
8172 | goto err; | |
7889f44d | 8173 | if (!cpu_online(cpu)) |
917257da JA |
8174 | goto err; |
8175 | ||
69fb2131 | 8176 | sqd->thread = kthread_create_on_cpu(io_sq_thread, sqd, |
534ca6d6 | 8177 | cpu, "io_uring-sq"); |
6c271ce2 | 8178 | } else { |
69fb2131 | 8179 | sqd->thread = kthread_create(io_sq_thread, sqd, |
6c271ce2 JA |
8180 | "io_uring-sq"); |
8181 | } | |
534ca6d6 JA |
8182 | if (IS_ERR(sqd->thread)) { |
8183 | ret = PTR_ERR(sqd->thread); | |
8184 | sqd->thread = NULL; | |
6c271ce2 JA |
8185 | goto err; |
8186 | } | |
534ca6d6 | 8187 | ret = io_uring_alloc_task_context(sqd->thread); |
0f212204 JA |
8188 | if (ret) |
8189 | goto err; | |
6c271ce2 JA |
8190 | } else if (p->flags & IORING_SETUP_SQ_AFF) { |
8191 | /* Can't have SQ_AFF without SQPOLL */ | |
8192 | ret = -EINVAL; | |
8193 | goto err; | |
8194 | } | |
8195 | ||
aa06165d | 8196 | done: |
24369c2e PB |
8197 | ret = io_init_wq_offload(ctx, p); |
8198 | if (ret) | |
2b188cc1 | 8199 | goto err; |
2b188cc1 JA |
8200 | |
8201 | return 0; | |
8202 | err: | |
54a91f3b | 8203 | io_finish_async(ctx); |
2b188cc1 JA |
8204 | return ret; |
8205 | } | |
8206 | ||
7e84e1c7 SG |
8207 | static void io_sq_offload_start(struct io_ring_ctx *ctx) |
8208 | { | |
534ca6d6 JA |
8209 | struct io_sq_data *sqd = ctx->sq_data; |
8210 | ||
8211 | if ((ctx->flags & IORING_SETUP_SQPOLL) && sqd->thread) | |
8212 | wake_up_process(sqd->thread); | |
7e84e1c7 SG |
8213 | } |
8214 | ||
a087e2b5 BM |
8215 | static inline void __io_unaccount_mem(struct user_struct *user, |
8216 | unsigned long nr_pages) | |
2b188cc1 JA |
8217 | { |
8218 | atomic_long_sub(nr_pages, &user->locked_vm); | |
8219 | } | |
8220 | ||
a087e2b5 BM |
8221 | static inline int __io_account_mem(struct user_struct *user, |
8222 | unsigned long nr_pages) | |
2b188cc1 JA |
8223 | { |
8224 | unsigned long page_limit, cur_pages, new_pages; | |
8225 | ||
8226 | /* Don't allow more pages than we can safely lock */ | |
8227 | page_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT; | |
8228 | ||
8229 | do { | |
8230 | cur_pages = atomic_long_read(&user->locked_vm); | |
8231 | new_pages = cur_pages + nr_pages; | |
8232 | if (new_pages > page_limit) | |
8233 | return -ENOMEM; | |
8234 | } while (atomic_long_cmpxchg(&user->locked_vm, cur_pages, | |
8235 | new_pages) != cur_pages); | |
8236 | ||
8237 | return 0; | |
8238 | } | |
8239 | ||
2e0464d4 BM |
8240 | static void io_unaccount_mem(struct io_ring_ctx *ctx, unsigned long nr_pages, |
8241 | enum io_mem_account acct) | |
a087e2b5 | 8242 | { |
aad5d8da | 8243 | if (ctx->limit_mem) |
a087e2b5 | 8244 | __io_unaccount_mem(ctx->user, nr_pages); |
30975825 | 8245 | |
2aede0e4 | 8246 | if (ctx->mm_account) { |
4bc4a912 JA |
8247 | if (acct == ACCT_LOCKED) { |
8248 | mmap_write_lock(ctx->mm_account); | |
2aede0e4 | 8249 | ctx->mm_account->locked_vm -= nr_pages; |
4bc4a912 JA |
8250 | mmap_write_unlock(ctx->mm_account); |
8251 | }else if (acct == ACCT_PINNED) { | |
2aede0e4 | 8252 | atomic64_sub(nr_pages, &ctx->mm_account->pinned_vm); |
4bc4a912 | 8253 | } |
2e0464d4 | 8254 | } |
a087e2b5 BM |
8255 | } |
8256 | ||
2e0464d4 BM |
8257 | static int io_account_mem(struct io_ring_ctx *ctx, unsigned long nr_pages, |
8258 | enum io_mem_account acct) | |
a087e2b5 | 8259 | { |
30975825 BM |
8260 | int ret; |
8261 | ||
8262 | if (ctx->limit_mem) { | |
8263 | ret = __io_account_mem(ctx->user, nr_pages); | |
8264 | if (ret) | |
8265 | return ret; | |
8266 | } | |
8267 | ||
2aede0e4 | 8268 | if (ctx->mm_account) { |
4bc4a912 JA |
8269 | if (acct == ACCT_LOCKED) { |
8270 | mmap_write_lock(ctx->mm_account); | |
2aede0e4 | 8271 | ctx->mm_account->locked_vm += nr_pages; |
4bc4a912 JA |
8272 | mmap_write_unlock(ctx->mm_account); |
8273 | } else if (acct == ACCT_PINNED) { | |
2aede0e4 | 8274 | atomic64_add(nr_pages, &ctx->mm_account->pinned_vm); |
4bc4a912 | 8275 | } |
2e0464d4 | 8276 | } |
a087e2b5 BM |
8277 | |
8278 | return 0; | |
8279 | } | |
8280 | ||
2b188cc1 JA |
8281 | static void io_mem_free(void *ptr) |
8282 | { | |
52e04ef4 MR |
8283 | struct page *page; |
8284 | ||
8285 | if (!ptr) | |
8286 | return; | |
2b188cc1 | 8287 | |
52e04ef4 | 8288 | page = virt_to_head_page(ptr); |
2b188cc1 JA |
8289 | if (put_page_testzero(page)) |
8290 | free_compound_page(page); | |
8291 | } | |
8292 | ||
8293 | static void *io_mem_alloc(size_t size) | |
8294 | { | |
8295 | gfp_t gfp_flags = GFP_KERNEL | __GFP_ZERO | __GFP_NOWARN | __GFP_COMP | | |
8296 | __GFP_NORETRY; | |
8297 | ||
8298 | return (void *) __get_free_pages(gfp_flags, get_order(size)); | |
8299 | } | |
8300 | ||
75b28aff HV |
8301 | static unsigned long rings_size(unsigned sq_entries, unsigned cq_entries, |
8302 | size_t *sq_offset) | |
8303 | { | |
8304 | struct io_rings *rings; | |
8305 | size_t off, sq_array_size; | |
8306 | ||
8307 | off = struct_size(rings, cqes, cq_entries); | |
8308 | if (off == SIZE_MAX) | |
8309 | return SIZE_MAX; | |
8310 | ||
8311 | #ifdef CONFIG_SMP | |
8312 | off = ALIGN(off, SMP_CACHE_BYTES); | |
8313 | if (off == 0) | |
8314 | return SIZE_MAX; | |
8315 | #endif | |
8316 | ||
b36200f5 DV |
8317 | if (sq_offset) |
8318 | *sq_offset = off; | |
8319 | ||
75b28aff HV |
8320 | sq_array_size = array_size(sizeof(u32), sq_entries); |
8321 | if (sq_array_size == SIZE_MAX) | |
8322 | return SIZE_MAX; | |
8323 | ||
8324 | if (check_add_overflow(off, sq_array_size, &off)) | |
8325 | return SIZE_MAX; | |
8326 | ||
75b28aff HV |
8327 | return off; |
8328 | } | |
8329 | ||
2b188cc1 JA |
8330 | static unsigned long ring_pages(unsigned sq_entries, unsigned cq_entries) |
8331 | { | |
75b28aff | 8332 | size_t pages; |
2b188cc1 | 8333 | |
75b28aff HV |
8334 | pages = (size_t)1 << get_order( |
8335 | rings_size(sq_entries, cq_entries, NULL)); | |
8336 | pages += (size_t)1 << get_order( | |
8337 | array_size(sizeof(struct io_uring_sqe), sq_entries)); | |
2b188cc1 | 8338 | |
75b28aff | 8339 | return pages; |
2b188cc1 JA |
8340 | } |
8341 | ||
edafccee JA |
8342 | static int io_sqe_buffer_unregister(struct io_ring_ctx *ctx) |
8343 | { | |
8344 | int i, j; | |
8345 | ||
8346 | if (!ctx->user_bufs) | |
8347 | return -ENXIO; | |
8348 | ||
8349 | for (i = 0; i < ctx->nr_user_bufs; i++) { | |
8350 | struct io_mapped_ubuf *imu = &ctx->user_bufs[i]; | |
8351 | ||
8352 | for (j = 0; j < imu->nr_bvecs; j++) | |
f1f6a7dd | 8353 | unpin_user_page(imu->bvec[j].bv_page); |
edafccee | 8354 | |
de293938 JA |
8355 | if (imu->acct_pages) |
8356 | io_unaccount_mem(ctx, imu->acct_pages, ACCT_PINNED); | |
d4ef6475 | 8357 | kvfree(imu->bvec); |
edafccee JA |
8358 | imu->nr_bvecs = 0; |
8359 | } | |
8360 | ||
8361 | kfree(ctx->user_bufs); | |
8362 | ctx->user_bufs = NULL; | |
8363 | ctx->nr_user_bufs = 0; | |
8364 | return 0; | |
8365 | } | |
8366 | ||
8367 | static int io_copy_iov(struct io_ring_ctx *ctx, struct iovec *dst, | |
8368 | void __user *arg, unsigned index) | |
8369 | { | |
8370 | struct iovec __user *src; | |
8371 | ||
8372 | #ifdef CONFIG_COMPAT | |
8373 | if (ctx->compat) { | |
8374 | struct compat_iovec __user *ciovs; | |
8375 | struct compat_iovec ciov; | |
8376 | ||
8377 | ciovs = (struct compat_iovec __user *) arg; | |
8378 | if (copy_from_user(&ciov, &ciovs[index], sizeof(ciov))) | |
8379 | return -EFAULT; | |
8380 | ||
d55e5f5b | 8381 | dst->iov_base = u64_to_user_ptr((u64)ciov.iov_base); |
edafccee JA |
8382 | dst->iov_len = ciov.iov_len; |
8383 | return 0; | |
8384 | } | |
8385 | #endif | |
8386 | src = (struct iovec __user *) arg; | |
8387 | if (copy_from_user(dst, &src[index], sizeof(*dst))) | |
8388 | return -EFAULT; | |
8389 | return 0; | |
8390 | } | |
8391 | ||
de293938 JA |
8392 | /* |
8393 | * Not super efficient, but this is just a registration time. And we do cache | |
8394 | * the last compound head, so generally we'll only do a full search if we don't | |
8395 | * match that one. | |
8396 | * | |
8397 | * We check if the given compound head page has already been accounted, to | |
8398 | * avoid double accounting it. This allows us to account the full size of the | |
8399 | * page, not just the constituent pages of a huge page. | |
8400 | */ | |
8401 | static bool headpage_already_acct(struct io_ring_ctx *ctx, struct page **pages, | |
8402 | int nr_pages, struct page *hpage) | |
8403 | { | |
8404 | int i, j; | |
8405 | ||
8406 | /* check current page array */ | |
8407 | for (i = 0; i < nr_pages; i++) { | |
8408 | if (!PageCompound(pages[i])) | |
8409 | continue; | |
8410 | if (compound_head(pages[i]) == hpage) | |
8411 | return true; | |
8412 | } | |
8413 | ||
8414 | /* check previously registered pages */ | |
8415 | for (i = 0; i < ctx->nr_user_bufs; i++) { | |
8416 | struct io_mapped_ubuf *imu = &ctx->user_bufs[i]; | |
8417 | ||
8418 | for (j = 0; j < imu->nr_bvecs; j++) { | |
8419 | if (!PageCompound(imu->bvec[j].bv_page)) | |
8420 | continue; | |
8421 | if (compound_head(imu->bvec[j].bv_page) == hpage) | |
8422 | return true; | |
8423 | } | |
8424 | } | |
8425 | ||
8426 | return false; | |
8427 | } | |
8428 | ||
8429 | static int io_buffer_account_pin(struct io_ring_ctx *ctx, struct page **pages, | |
8430 | int nr_pages, struct io_mapped_ubuf *imu, | |
8431 | struct page **last_hpage) | |
8432 | { | |
8433 | int i, ret; | |
8434 | ||
8435 | for (i = 0; i < nr_pages; i++) { | |
8436 | if (!PageCompound(pages[i])) { | |
8437 | imu->acct_pages++; | |
8438 | } else { | |
8439 | struct page *hpage; | |
8440 | ||
8441 | hpage = compound_head(pages[i]); | |
8442 | if (hpage == *last_hpage) | |
8443 | continue; | |
8444 | *last_hpage = hpage; | |
8445 | if (headpage_already_acct(ctx, pages, i, hpage)) | |
8446 | continue; | |
8447 | imu->acct_pages += page_size(hpage) >> PAGE_SHIFT; | |
8448 | } | |
8449 | } | |
8450 | ||
8451 | if (!imu->acct_pages) | |
8452 | return 0; | |
8453 | ||
8454 | ret = io_account_mem(ctx, imu->acct_pages, ACCT_PINNED); | |
8455 | if (ret) | |
8456 | imu->acct_pages = 0; | |
8457 | return ret; | |
8458 | } | |
8459 | ||
edafccee JA |
8460 | static int io_sqe_buffer_register(struct io_ring_ctx *ctx, void __user *arg, |
8461 | unsigned nr_args) | |
8462 | { | |
8463 | struct vm_area_struct **vmas = NULL; | |
8464 | struct page **pages = NULL; | |
de293938 | 8465 | struct page *last_hpage = NULL; |
edafccee JA |
8466 | int i, j, got_pages = 0; |
8467 | int ret = -EINVAL; | |
8468 | ||
8469 | if (ctx->user_bufs) | |
8470 | return -EBUSY; | |
8471 | if (!nr_args || nr_args > UIO_MAXIOV) | |
8472 | return -EINVAL; | |
8473 | ||
8474 | ctx->user_bufs = kcalloc(nr_args, sizeof(struct io_mapped_ubuf), | |
8475 | GFP_KERNEL); | |
8476 | if (!ctx->user_bufs) | |
8477 | return -ENOMEM; | |
8478 | ||
8479 | for (i = 0; i < nr_args; i++) { | |
8480 | struct io_mapped_ubuf *imu = &ctx->user_bufs[i]; | |
8481 | unsigned long off, start, end, ubuf; | |
8482 | int pret, nr_pages; | |
8483 | struct iovec iov; | |
8484 | size_t size; | |
8485 | ||
8486 | ret = io_copy_iov(ctx, &iov, arg, i); | |
8487 | if (ret) | |
a278682d | 8488 | goto err; |
edafccee JA |
8489 | |
8490 | /* | |
8491 | * Don't impose further limits on the size and buffer | |
8492 | * constraints here, we'll -EINVAL later when IO is | |
8493 | * submitted if they are wrong. | |
8494 | */ | |
8495 | ret = -EFAULT; | |
8496 | if (!iov.iov_base || !iov.iov_len) | |
8497 | goto err; | |
8498 | ||
8499 | /* arbitrary limit, but we need something */ | |
8500 | if (iov.iov_len > SZ_1G) | |
8501 | goto err; | |
8502 | ||
8503 | ubuf = (unsigned long) iov.iov_base; | |
8504 | end = (ubuf + iov.iov_len + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
8505 | start = ubuf >> PAGE_SHIFT; | |
8506 | nr_pages = end - start; | |
8507 | ||
edafccee JA |
8508 | ret = 0; |
8509 | if (!pages || nr_pages > got_pages) { | |
a8c73c1a DE |
8510 | kvfree(vmas); |
8511 | kvfree(pages); | |
d4ef6475 | 8512 | pages = kvmalloc_array(nr_pages, sizeof(struct page *), |
edafccee | 8513 | GFP_KERNEL); |
d4ef6475 | 8514 | vmas = kvmalloc_array(nr_pages, |
edafccee JA |
8515 | sizeof(struct vm_area_struct *), |
8516 | GFP_KERNEL); | |
8517 | if (!pages || !vmas) { | |
8518 | ret = -ENOMEM; | |
edafccee JA |
8519 | goto err; |
8520 | } | |
8521 | got_pages = nr_pages; | |
8522 | } | |
8523 | ||
d4ef6475 | 8524 | imu->bvec = kvmalloc_array(nr_pages, sizeof(struct bio_vec), |
edafccee JA |
8525 | GFP_KERNEL); |
8526 | ret = -ENOMEM; | |
de293938 | 8527 | if (!imu->bvec) |
edafccee | 8528 | goto err; |
edafccee JA |
8529 | |
8530 | ret = 0; | |
d8ed45c5 | 8531 | mmap_read_lock(current->mm); |
2113b05d | 8532 | pret = pin_user_pages(ubuf, nr_pages, |
932f4a63 IW |
8533 | FOLL_WRITE | FOLL_LONGTERM, |
8534 | pages, vmas); | |
edafccee JA |
8535 | if (pret == nr_pages) { |
8536 | /* don't support file backed memory */ | |
8537 | for (j = 0; j < nr_pages; j++) { | |
8538 | struct vm_area_struct *vma = vmas[j]; | |
8539 | ||
8540 | if (vma->vm_file && | |
8541 | !is_file_hugepages(vma->vm_file)) { | |
8542 | ret = -EOPNOTSUPP; | |
8543 | break; | |
8544 | } | |
8545 | } | |
8546 | } else { | |
8547 | ret = pret < 0 ? pret : -EFAULT; | |
8548 | } | |
d8ed45c5 | 8549 | mmap_read_unlock(current->mm); |
edafccee JA |
8550 | if (ret) { |
8551 | /* | |
8552 | * if we did partial map, or found file backed vmas, | |
8553 | * release any pages we did get | |
8554 | */ | |
27c4d3a3 | 8555 | if (pret > 0) |
f1f6a7dd | 8556 | unpin_user_pages(pages, pret); |
de293938 JA |
8557 | kvfree(imu->bvec); |
8558 | goto err; | |
8559 | } | |
8560 | ||
8561 | ret = io_buffer_account_pin(ctx, pages, pret, imu, &last_hpage); | |
8562 | if (ret) { | |
8563 | unpin_user_pages(pages, pret); | |
d4ef6475 | 8564 | kvfree(imu->bvec); |
edafccee JA |
8565 | goto err; |
8566 | } | |
8567 | ||
8568 | off = ubuf & ~PAGE_MASK; | |
8569 | size = iov.iov_len; | |
8570 | for (j = 0; j < nr_pages; j++) { | |
8571 | size_t vec_len; | |
8572 | ||
8573 | vec_len = min_t(size_t, size, PAGE_SIZE - off); | |
8574 | imu->bvec[j].bv_page = pages[j]; | |
8575 | imu->bvec[j].bv_len = vec_len; | |
8576 | imu->bvec[j].bv_offset = off; | |
8577 | off = 0; | |
8578 | size -= vec_len; | |
8579 | } | |
8580 | /* store original address for later verification */ | |
8581 | imu->ubuf = ubuf; | |
8582 | imu->len = iov.iov_len; | |
8583 | imu->nr_bvecs = nr_pages; | |
8584 | ||
8585 | ctx->nr_user_bufs++; | |
8586 | } | |
d4ef6475 MR |
8587 | kvfree(pages); |
8588 | kvfree(vmas); | |
edafccee JA |
8589 | return 0; |
8590 | err: | |
d4ef6475 MR |
8591 | kvfree(pages); |
8592 | kvfree(vmas); | |
edafccee JA |
8593 | io_sqe_buffer_unregister(ctx); |
8594 | return ret; | |
8595 | } | |
8596 | ||
9b402849 JA |
8597 | static int io_eventfd_register(struct io_ring_ctx *ctx, void __user *arg) |
8598 | { | |
8599 | __s32 __user *fds = arg; | |
8600 | int fd; | |
8601 | ||
8602 | if (ctx->cq_ev_fd) | |
8603 | return -EBUSY; | |
8604 | ||
8605 | if (copy_from_user(&fd, fds, sizeof(*fds))) | |
8606 | return -EFAULT; | |
8607 | ||
8608 | ctx->cq_ev_fd = eventfd_ctx_fdget(fd); | |
8609 | if (IS_ERR(ctx->cq_ev_fd)) { | |
8610 | int ret = PTR_ERR(ctx->cq_ev_fd); | |
8611 | ctx->cq_ev_fd = NULL; | |
8612 | return ret; | |
8613 | } | |
8614 | ||
8615 | return 0; | |
8616 | } | |
8617 | ||
8618 | static int io_eventfd_unregister(struct io_ring_ctx *ctx) | |
8619 | { | |
8620 | if (ctx->cq_ev_fd) { | |
8621 | eventfd_ctx_put(ctx->cq_ev_fd); | |
8622 | ctx->cq_ev_fd = NULL; | |
8623 | return 0; | |
8624 | } | |
8625 | ||
8626 | return -ENXIO; | |
8627 | } | |
8628 | ||
5a2e745d JA |
8629 | static int __io_destroy_buffers(int id, void *p, void *data) |
8630 | { | |
8631 | struct io_ring_ctx *ctx = data; | |
8632 | struct io_buffer *buf = p; | |
8633 | ||
067524e9 | 8634 | __io_remove_buffers(ctx, buf, id, -1U); |
5a2e745d JA |
8635 | return 0; |
8636 | } | |
8637 | ||
8638 | static void io_destroy_buffers(struct io_ring_ctx *ctx) | |
8639 | { | |
8640 | idr_for_each(&ctx->io_buffer_idr, __io_destroy_buffers, ctx); | |
8641 | idr_destroy(&ctx->io_buffer_idr); | |
8642 | } | |
8643 | ||
2b188cc1 JA |
8644 | static void io_ring_ctx_free(struct io_ring_ctx *ctx) |
8645 | { | |
6b06314c | 8646 | io_finish_async(ctx); |
5dbcad51 | 8647 | io_sqe_buffer_unregister(ctx); |
2aede0e4 JA |
8648 | |
8649 | if (ctx->sqo_task) { | |
8650 | put_task_struct(ctx->sqo_task); | |
8651 | ctx->sqo_task = NULL; | |
8652 | mmdrop(ctx->mm_account); | |
8653 | ctx->mm_account = NULL; | |
30975825 | 8654 | } |
def596e9 | 8655 | |
91d8f519 DZ |
8656 | #ifdef CONFIG_BLK_CGROUP |
8657 | if (ctx->sqo_blkcg_css) | |
8658 | css_put(ctx->sqo_blkcg_css); | |
8659 | #endif | |
8660 | ||
6b06314c | 8661 | io_sqe_files_unregister(ctx); |
9b402849 | 8662 | io_eventfd_unregister(ctx); |
5a2e745d | 8663 | io_destroy_buffers(ctx); |
41726c9a | 8664 | idr_destroy(&ctx->personality_idr); |
def596e9 | 8665 | |
2b188cc1 | 8666 | #if defined(CONFIG_UNIX) |
355e8d26 EB |
8667 | if (ctx->ring_sock) { |
8668 | ctx->ring_sock->file = NULL; /* so that iput() is called */ | |
2b188cc1 | 8669 | sock_release(ctx->ring_sock); |
355e8d26 | 8670 | } |
2b188cc1 JA |
8671 | #endif |
8672 | ||
75b28aff | 8673 | io_mem_free(ctx->rings); |
2b188cc1 | 8674 | io_mem_free(ctx->sq_sqes); |
2b188cc1 JA |
8675 | |
8676 | percpu_ref_exit(&ctx->refs); | |
2b188cc1 | 8677 | free_uid(ctx->user); |
181e448d | 8678 | put_cred(ctx->creds); |
78076bb6 | 8679 | kfree(ctx->cancel_hash); |
0ddf92e8 | 8680 | kmem_cache_free(req_cachep, ctx->fallback_req); |
2b188cc1 JA |
8681 | kfree(ctx); |
8682 | } | |
8683 | ||
8684 | static __poll_t io_uring_poll(struct file *file, poll_table *wait) | |
8685 | { | |
8686 | struct io_ring_ctx *ctx = file->private_data; | |
8687 | __poll_t mask = 0; | |
8688 | ||
8689 | poll_wait(file, &ctx->cq_wait, wait); | |
4f7067c3 SB |
8690 | /* |
8691 | * synchronizes with barrier from wq_has_sleeper call in | |
8692 | * io_commit_cqring | |
8693 | */ | |
2b188cc1 | 8694 | smp_rmb(); |
90554200 | 8695 | if (!io_sqring_full(ctx)) |
2b188cc1 | 8696 | mask |= EPOLLOUT | EPOLLWRNORM; |
6c503150 PB |
8697 | io_cqring_overflow_flush(ctx, false, NULL, NULL); |
8698 | if (io_cqring_events(ctx)) | |
2b188cc1 JA |
8699 | mask |= EPOLLIN | EPOLLRDNORM; |
8700 | ||
8701 | return mask; | |
8702 | } | |
8703 | ||
8704 | static int io_uring_fasync(int fd, struct file *file, int on) | |
8705 | { | |
8706 | struct io_ring_ctx *ctx = file->private_data; | |
8707 | ||
8708 | return fasync_helper(fd, file, on, &ctx->cq_fasync); | |
8709 | } | |
8710 | ||
071698e1 JA |
8711 | static int io_remove_personalities(int id, void *p, void *data) |
8712 | { | |
8713 | struct io_ring_ctx *ctx = data; | |
1e6fa521 | 8714 | struct io_identity *iod; |
071698e1 | 8715 | |
1e6fa521 JA |
8716 | iod = idr_remove(&ctx->personality_idr, id); |
8717 | if (iod) { | |
8718 | put_cred(iod->creds); | |
8719 | if (refcount_dec_and_test(&iod->count)) | |
8720 | kfree(iod); | |
8721 | } | |
071698e1 JA |
8722 | return 0; |
8723 | } | |
8724 | ||
85faa7b8 JA |
8725 | static void io_ring_exit_work(struct work_struct *work) |
8726 | { | |
b2edc0a7 PB |
8727 | struct io_ring_ctx *ctx = container_of(work, struct io_ring_ctx, |
8728 | exit_work); | |
85faa7b8 | 8729 | |
56952e91 JA |
8730 | /* |
8731 | * If we're doing polled IO and end up having requests being | |
8732 | * submitted async (out-of-line), then completions can come in while | |
8733 | * we're waiting for refs to drop. We need to reap these manually, | |
8734 | * as nobody else will be looking for them. | |
8735 | */ | |
b2edc0a7 | 8736 | do { |
90df0853 | 8737 | __io_uring_cancel_task_requests(ctx, NULL); |
b2edc0a7 | 8738 | } while (!wait_for_completion_timeout(&ctx->ref_comp, HZ/20)); |
85faa7b8 JA |
8739 | io_ring_ctx_free(ctx); |
8740 | } | |
8741 | ||
00c18640 JA |
8742 | static bool io_cancel_ctx_cb(struct io_wq_work *work, void *data) |
8743 | { | |
8744 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); | |
8745 | ||
8746 | return req->ctx == data; | |
8747 | } | |
8748 | ||
2b188cc1 JA |
8749 | static void io_ring_ctx_wait_and_kill(struct io_ring_ctx *ctx) |
8750 | { | |
8751 | mutex_lock(&ctx->uring_lock); | |
8752 | percpu_ref_kill(&ctx->refs); | |
d9d05217 PB |
8753 | |
8754 | if (WARN_ON_ONCE((ctx->flags & IORING_SETUP_SQPOLL) && !ctx->sqo_dead)) | |
8755 | ctx->sqo_dead = 1; | |
8756 | ||
cda286f0 PB |
8757 | /* if force is set, the ring is going away. always drop after that */ |
8758 | ctx->cq_overflow_flushed = 1; | |
634578f8 | 8759 | if (ctx->rings) |
6c503150 | 8760 | __io_cqring_overflow_flush(ctx, true, NULL, NULL); |
2b188cc1 JA |
8761 | mutex_unlock(&ctx->uring_lock); |
8762 | ||
6b81928d PB |
8763 | io_kill_timeouts(ctx, NULL, NULL); |
8764 | io_poll_remove_all(ctx, NULL, NULL); | |
561fb04a JA |
8765 | |
8766 | if (ctx->io_wq) | |
00c18640 | 8767 | io_wq_cancel_cb(ctx->io_wq, io_cancel_ctx_cb, ctx, true); |
561fb04a | 8768 | |
15dff286 | 8769 | /* if we failed setting up the ctx, we might not have any rings */ |
b2edc0a7 | 8770 | io_iopoll_try_reap_events(ctx); |
071698e1 | 8771 | idr_for_each(&ctx->personality_idr, io_remove_personalities, ctx); |
309fc03a JA |
8772 | |
8773 | /* | |
8774 | * Do this upfront, so we won't have a grace period where the ring | |
8775 | * is closed but resources aren't reaped yet. This can cause | |
8776 | * spurious failure in setting up a new ring. | |
8777 | */ | |
760618f7 JA |
8778 | io_unaccount_mem(ctx, ring_pages(ctx->sq_entries, ctx->cq_entries), |
8779 | ACCT_LOCKED); | |
309fc03a | 8780 | |
85faa7b8 | 8781 | INIT_WORK(&ctx->exit_work, io_ring_exit_work); |
fc666777 JA |
8782 | /* |
8783 | * Use system_unbound_wq to avoid spawning tons of event kworkers | |
8784 | * if we're exiting a ton of rings at the same time. It just adds | |
8785 | * noise and overhead, there's no discernable change in runtime | |
8786 | * over using system_wq. | |
8787 | */ | |
8788 | queue_work(system_unbound_wq, &ctx->exit_work); | |
2b188cc1 JA |
8789 | } |
8790 | ||
8791 | static int io_uring_release(struct inode *inode, struct file *file) | |
8792 | { | |
8793 | struct io_ring_ctx *ctx = file->private_data; | |
8794 | ||
8795 | file->private_data = NULL; | |
8796 | io_ring_ctx_wait_and_kill(ctx); | |
8797 | return 0; | |
8798 | } | |
8799 | ||
f6edbabb PB |
8800 | struct io_task_cancel { |
8801 | struct task_struct *task; | |
8802 | struct files_struct *files; | |
8803 | }; | |
f254ac04 | 8804 | |
f6edbabb | 8805 | static bool io_cancel_task_cb(struct io_wq_work *work, void *data) |
b711d4ea | 8806 | { |
9a472ef7 | 8807 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); |
f6edbabb | 8808 | struct io_task_cancel *cancel = data; |
9a472ef7 PB |
8809 | bool ret; |
8810 | ||
f6edbabb | 8811 | if (cancel->files && (req->flags & REQ_F_LINK_TIMEOUT)) { |
9a472ef7 PB |
8812 | unsigned long flags; |
8813 | struct io_ring_ctx *ctx = req->ctx; | |
8814 | ||
8815 | /* protect against races with linked timeouts */ | |
8816 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
f6edbabb | 8817 | ret = io_match_task(req, cancel->task, cancel->files); |
9a472ef7 PB |
8818 | spin_unlock_irqrestore(&ctx->completion_lock, flags); |
8819 | } else { | |
f6edbabb | 8820 | ret = io_match_task(req, cancel->task, cancel->files); |
9a472ef7 PB |
8821 | } |
8822 | return ret; | |
b711d4ea JA |
8823 | } |
8824 | ||
b7ddce3c | 8825 | static void io_cancel_defer_files(struct io_ring_ctx *ctx, |
ef9865a4 | 8826 | struct task_struct *task, |
b7ddce3c PB |
8827 | struct files_struct *files) |
8828 | { | |
8829 | struct io_defer_entry *de = NULL; | |
8830 | LIST_HEAD(list); | |
8831 | ||
8832 | spin_lock_irq(&ctx->completion_lock); | |
8833 | list_for_each_entry_reverse(de, &ctx->defer_list, list) { | |
08d23634 | 8834 | if (io_match_task(de->req, task, files)) { |
b7ddce3c PB |
8835 | list_cut_position(&list, &ctx->defer_list, &de->list); |
8836 | break; | |
8837 | } | |
8838 | } | |
8839 | spin_unlock_irq(&ctx->completion_lock); | |
8840 | ||
8841 | while (!list_empty(&list)) { | |
8842 | de = list_first_entry(&list, struct io_defer_entry, list); | |
8843 | list_del_init(&de->list); | |
8844 | req_set_fail_links(de->req); | |
8845 | io_put_req(de->req); | |
8846 | io_req_complete(de->req, -ECANCELED); | |
8847 | kfree(de); | |
8848 | } | |
8849 | } | |
8850 | ||
b52fda00 | 8851 | static void io_uring_cancel_files(struct io_ring_ctx *ctx, |
df9923f9 | 8852 | struct task_struct *task, |
fcb323cc JA |
8853 | struct files_struct *files) |
8854 | { | |
fcb323cc | 8855 | while (!list_empty_careful(&ctx->inflight_list)) { |
bee749b1 | 8856 | struct io_task_cancel cancel = { .task = task, .files = files }; |
f6edbabb | 8857 | struct io_kiocb *req; |
d8f1b971 | 8858 | DEFINE_WAIT(wait); |
f6edbabb | 8859 | bool found = false; |
fcb323cc JA |
8860 | |
8861 | spin_lock_irq(&ctx->inflight_lock); | |
8862 | list_for_each_entry(req, &ctx->inflight_list, inflight_entry) { | |
bee749b1 | 8863 | if (req->task != task || |
98447d65 | 8864 | req->work.identity->files != files) |
768134d4 | 8865 | continue; |
f6edbabb | 8866 | found = true; |
768134d4 | 8867 | break; |
fcb323cc | 8868 | } |
f6edbabb | 8869 | if (found) |
c98de08c PB |
8870 | prepare_to_wait(&task->io_uring->wait, &wait, |
8871 | TASK_UNINTERRUPTIBLE); | |
fcb323cc JA |
8872 | spin_unlock_irq(&ctx->inflight_lock); |
8873 | ||
768134d4 | 8874 | /* We need to keep going until we don't find a matching req */ |
f6edbabb | 8875 | if (!found) |
fcb323cc | 8876 | break; |
f6edbabb PB |
8877 | |
8878 | io_wq_cancel_cb(ctx->io_wq, io_cancel_task_cb, &cancel, true); | |
8879 | io_poll_remove_all(ctx, task, files); | |
8880 | io_kill_timeouts(ctx, task, files); | |
6200b0ae JA |
8881 | /* cancellations _may_ trigger task work */ |
8882 | io_run_task_work(); | |
fcb323cc | 8883 | schedule(); |
c98de08c | 8884 | finish_wait(&task->io_uring->wait, &wait); |
fcb323cc | 8885 | } |
44e728b8 PB |
8886 | } |
8887 | ||
b52fda00 PB |
8888 | static void __io_uring_cancel_task_requests(struct io_ring_ctx *ctx, |
8889 | struct task_struct *task) | |
0f212204 | 8890 | { |
b52fda00 | 8891 | while (1) { |
f6edbabb | 8892 | struct io_task_cancel cancel = { .task = task, .files = NULL, }; |
0f212204 | 8893 | enum io_wq_cancel cret; |
b52fda00 | 8894 | bool ret = false; |
0f212204 | 8895 | |
90df0853 PB |
8896 | if (ctx->io_wq) { |
8897 | cret = io_wq_cancel_cb(ctx->io_wq, io_cancel_task_cb, | |
8898 | &cancel, true); | |
8899 | ret |= (cret != IO_WQ_CANCEL_NOTFOUND); | |
8900 | } | |
0f212204 JA |
8901 | |
8902 | /* SQPOLL thread does its own polling */ | |
8903 | if (!(ctx->flags & IORING_SETUP_SQPOLL)) { | |
8904 | while (!list_empty_careful(&ctx->iopoll_list)) { | |
8905 | io_iopoll_try_reap_events(ctx); | |
8906 | ret = true; | |
8907 | } | |
8908 | } | |
8909 | ||
6b81928d PB |
8910 | ret |= io_poll_remove_all(ctx, task, NULL); |
8911 | ret |= io_kill_timeouts(ctx, task, NULL); | |
55583d72 | 8912 | ret |= io_run_task_work(); |
b52fda00 PB |
8913 | if (!ret) |
8914 | break; | |
b52fda00 | 8915 | cond_resched(); |
0f212204 | 8916 | } |
0f212204 JA |
8917 | } |
8918 | ||
d9d05217 PB |
8919 | static void io_disable_sqo_submit(struct io_ring_ctx *ctx) |
8920 | { | |
8921 | WARN_ON_ONCE(ctx->sqo_task != current); | |
8922 | ||
8923 | mutex_lock(&ctx->uring_lock); | |
8924 | ctx->sqo_dead = 1; | |
8925 | mutex_unlock(&ctx->uring_lock); | |
8926 | ||
8927 | /* make sure callers enter the ring to get error */ | |
b4411616 PB |
8928 | if (ctx->rings) |
8929 | io_ring_set_wakeup_flag(ctx); | |
d9d05217 PB |
8930 | } |
8931 | ||
0f212204 JA |
8932 | /* |
8933 | * We need to iteratively cancel requests, in case a request has dependent | |
8934 | * hard links. These persist even for failure of cancelations, hence keep | |
8935 | * looping until none are found. | |
8936 | */ | |
8937 | static void io_uring_cancel_task_requests(struct io_ring_ctx *ctx, | |
8938 | struct files_struct *files) | |
8939 | { | |
8940 | struct task_struct *task = current; | |
8941 | ||
fdaf083c | 8942 | if ((ctx->flags & IORING_SETUP_SQPOLL) && ctx->sq_data) { |
d9d05217 PB |
8943 | /* for SQPOLL only sqo_task has task notes */ |
8944 | io_disable_sqo_submit(ctx); | |
534ca6d6 | 8945 | task = ctx->sq_data->thread; |
fdaf083c JA |
8946 | atomic_inc(&task->io_uring->in_idle); |
8947 | io_sq_thread_park(ctx->sq_data); | |
8948 | } | |
0f212204 | 8949 | |
df9923f9 | 8950 | io_cancel_defer_files(ctx, task, files); |
0f212204 JA |
8951 | io_cqring_overflow_flush(ctx, true, task, files); |
8952 | ||
b52fda00 PB |
8953 | if (!files) |
8954 | __io_uring_cancel_task_requests(ctx, task); | |
bee749b1 PB |
8955 | else |
8956 | io_uring_cancel_files(ctx, task, files); | |
fdaf083c JA |
8957 | |
8958 | if ((ctx->flags & IORING_SETUP_SQPOLL) && ctx->sq_data) { | |
8959 | atomic_dec(&task->io_uring->in_idle); | |
8960 | /* | |
8961 | * If the files that are going away are the ones in the thread | |
8962 | * identity, clear them out. | |
8963 | */ | |
8964 | if (task->io_uring->identity->files == files) | |
8965 | task->io_uring->identity->files = NULL; | |
8966 | io_sq_thread_unpark(ctx->sq_data); | |
8967 | } | |
0f212204 JA |
8968 | } |
8969 | ||
8970 | /* | |
8971 | * Note that this task has used io_uring. We use it for cancelation purposes. | |
8972 | */ | |
fdaf083c | 8973 | static int io_uring_add_task_file(struct io_ring_ctx *ctx, struct file *file) |
0f212204 | 8974 | { |
236434c3 | 8975 | struct io_uring_task *tctx = current->io_uring; |
a528b04e | 8976 | int ret; |
236434c3 MWO |
8977 | |
8978 | if (unlikely(!tctx)) { | |
0f212204 JA |
8979 | ret = io_uring_alloc_task_context(current); |
8980 | if (unlikely(ret)) | |
8981 | return ret; | |
236434c3 | 8982 | tctx = current->io_uring; |
0f212204 | 8983 | } |
236434c3 MWO |
8984 | if (tctx->last != file) { |
8985 | void *old = xa_load(&tctx->xa, (unsigned long)file); | |
0f212204 | 8986 | |
236434c3 | 8987 | if (!old) { |
0f212204 | 8988 | get_file(file); |
a528b04e PB |
8989 | ret = xa_err(xa_store(&tctx->xa, (unsigned long)file, |
8990 | file, GFP_KERNEL)); | |
8991 | if (ret) { | |
8992 | fput(file); | |
8993 | return ret; | |
8994 | } | |
0f212204 | 8995 | } |
236434c3 | 8996 | tctx->last = file; |
0f212204 JA |
8997 | } |
8998 | ||
fdaf083c JA |
8999 | /* |
9000 | * This is race safe in that the task itself is doing this, hence it | |
9001 | * cannot be going through the exit/cancel paths at the same time. | |
9002 | * This cannot be modified while exit/cancel is running. | |
9003 | */ | |
9004 | if (!tctx->sqpoll && (ctx->flags & IORING_SETUP_SQPOLL)) | |
9005 | tctx->sqpoll = true; | |
9006 | ||
0f212204 JA |
9007 | return 0; |
9008 | } | |
9009 | ||
9010 | /* | |
9011 | * Remove this io_uring_file -> task mapping. | |
9012 | */ | |
9013 | static void io_uring_del_task_file(struct file *file) | |
9014 | { | |
9015 | struct io_uring_task *tctx = current->io_uring; | |
0f212204 JA |
9016 | |
9017 | if (tctx->last == file) | |
9018 | tctx->last = NULL; | |
5e2ed8c4 | 9019 | file = xa_erase(&tctx->xa, (unsigned long)file); |
0f212204 JA |
9020 | if (file) |
9021 | fput(file); | |
9022 | } | |
9023 | ||
de7f1d9e PB |
9024 | static void io_uring_remove_task_files(struct io_uring_task *tctx) |
9025 | { | |
9026 | struct file *file; | |
9027 | unsigned long index; | |
9028 | ||
9029 | xa_for_each(&tctx->xa, index, file) | |
9030 | io_uring_del_task_file(file); | |
9031 | } | |
9032 | ||
0f212204 JA |
9033 | void __io_uring_files_cancel(struct files_struct *files) |
9034 | { | |
9035 | struct io_uring_task *tctx = current->io_uring; | |
ce765372 MWO |
9036 | struct file *file; |
9037 | unsigned long index; | |
0f212204 JA |
9038 | |
9039 | /* make sure overflow events are dropped */ | |
fdaf083c | 9040 | atomic_inc(&tctx->in_idle); |
de7f1d9e PB |
9041 | xa_for_each(&tctx->xa, index, file) |
9042 | io_uring_cancel_task_requests(file->private_data, files); | |
fdaf083c | 9043 | atomic_dec(&tctx->in_idle); |
de7f1d9e PB |
9044 | |
9045 | if (files) | |
9046 | io_uring_remove_task_files(tctx); | |
fdaf083c JA |
9047 | } |
9048 | ||
9049 | static s64 tctx_inflight(struct io_uring_task *tctx) | |
9050 | { | |
9051 | unsigned long index; | |
9052 | struct file *file; | |
9053 | s64 inflight; | |
9054 | ||
9055 | inflight = percpu_counter_sum(&tctx->inflight); | |
9056 | if (!tctx->sqpoll) | |
9057 | return inflight; | |
9058 | ||
9059 | /* | |
9060 | * If we have SQPOLL rings, then we need to iterate and find them, and | |
9061 | * add the pending count for those. | |
9062 | */ | |
9063 | xa_for_each(&tctx->xa, index, file) { | |
9064 | struct io_ring_ctx *ctx = file->private_data; | |
9065 | ||
9066 | if (ctx->flags & IORING_SETUP_SQPOLL) { | |
9067 | struct io_uring_task *__tctx = ctx->sqo_task->io_uring; | |
9068 | ||
9069 | inflight += percpu_counter_sum(&__tctx->inflight); | |
9070 | } | |
9071 | } | |
9072 | ||
9073 | return inflight; | |
0f212204 JA |
9074 | } |
9075 | ||
0f212204 JA |
9076 | /* |
9077 | * Find any io_uring fd that this task has registered or done IO on, and cancel | |
9078 | * requests. | |
9079 | */ | |
9080 | void __io_uring_task_cancel(void) | |
9081 | { | |
9082 | struct io_uring_task *tctx = current->io_uring; | |
9083 | DEFINE_WAIT(wait); | |
d8a6df10 | 9084 | s64 inflight; |
0f212204 JA |
9085 | |
9086 | /* make sure overflow events are dropped */ | |
fdaf083c | 9087 | atomic_inc(&tctx->in_idle); |
0f212204 | 9088 | |
d8a6df10 | 9089 | do { |
0f212204 | 9090 | /* read completions before cancelations */ |
fdaf083c | 9091 | inflight = tctx_inflight(tctx); |
d8a6df10 JA |
9092 | if (!inflight) |
9093 | break; | |
0f212204 JA |
9094 | __io_uring_files_cancel(NULL); |
9095 | ||
9096 | prepare_to_wait(&tctx->wait, &wait, TASK_UNINTERRUPTIBLE); | |
9097 | ||
9098 | /* | |
9099 | * If we've seen completions, retry. This avoids a race where | |
9100 | * a completion comes in before we did prepare_to_wait(). | |
9101 | */ | |
fdaf083c | 9102 | if (inflight != tctx_inflight(tctx)) |
0f212204 | 9103 | continue; |
0f212204 | 9104 | schedule(); |
f57555ed | 9105 | finish_wait(&tctx->wait, &wait); |
d8a6df10 | 9106 | } while (1); |
0f212204 | 9107 | |
a8d13dbc | 9108 | finish_wait(&tctx->wait, &wait); |
fdaf083c | 9109 | atomic_dec(&tctx->in_idle); |
de7f1d9e PB |
9110 | |
9111 | io_uring_remove_task_files(tctx); | |
44e728b8 PB |
9112 | } |
9113 | ||
fcb323cc JA |
9114 | static int io_uring_flush(struct file *file, void *data) |
9115 | { | |
6b5733eb | 9116 | struct io_uring_task *tctx = current->io_uring; |
d9d05217 | 9117 | struct io_ring_ctx *ctx = file->private_data; |
6b5733eb PB |
9118 | |
9119 | if (!tctx) | |
4f793dc4 PB |
9120 | return 0; |
9121 | ||
6b5733eb PB |
9122 | /* we should have cancelled and erased it before PF_EXITING */ |
9123 | WARN_ON_ONCE((current->flags & PF_EXITING) && | |
9124 | xa_load(&tctx->xa, (unsigned long)file)); | |
9125 | ||
4f793dc4 PB |
9126 | /* |
9127 | * fput() is pending, will be 2 if the only other ref is our potential | |
9128 | * task file note. If the task is exiting, drop regardless of count. | |
9129 | */ | |
6b5733eb PB |
9130 | if (atomic_long_read(&file->f_count) != 2) |
9131 | return 0; | |
4f793dc4 | 9132 | |
d9d05217 PB |
9133 | if (ctx->flags & IORING_SETUP_SQPOLL) { |
9134 | /* there is only one file note, which is owned by sqo_task */ | |
9135 | WARN_ON_ONCE((ctx->sqo_task == current) == | |
9136 | !xa_load(&tctx->xa, (unsigned long)file)); | |
9137 | ||
9138 | io_disable_sqo_submit(ctx); | |
9139 | } | |
9140 | ||
9141 | if (!(ctx->flags & IORING_SETUP_SQPOLL) || ctx->sqo_task == current) | |
9142 | io_uring_del_task_file(file); | |
fcb323cc JA |
9143 | return 0; |
9144 | } | |
9145 | ||
6c5c240e RP |
9146 | static void *io_uring_validate_mmap_request(struct file *file, |
9147 | loff_t pgoff, size_t sz) | |
2b188cc1 | 9148 | { |
2b188cc1 | 9149 | struct io_ring_ctx *ctx = file->private_data; |
6c5c240e | 9150 | loff_t offset = pgoff << PAGE_SHIFT; |
2b188cc1 JA |
9151 | struct page *page; |
9152 | void *ptr; | |
9153 | ||
9154 | switch (offset) { | |
9155 | case IORING_OFF_SQ_RING: | |
75b28aff HV |
9156 | case IORING_OFF_CQ_RING: |
9157 | ptr = ctx->rings; | |
2b188cc1 JA |
9158 | break; |
9159 | case IORING_OFF_SQES: | |
9160 | ptr = ctx->sq_sqes; | |
9161 | break; | |
2b188cc1 | 9162 | default: |
6c5c240e | 9163 | return ERR_PTR(-EINVAL); |
2b188cc1 JA |
9164 | } |
9165 | ||
9166 | page = virt_to_head_page(ptr); | |
a50b854e | 9167 | if (sz > page_size(page)) |
6c5c240e RP |
9168 | return ERR_PTR(-EINVAL); |
9169 | ||
9170 | return ptr; | |
9171 | } | |
9172 | ||
9173 | #ifdef CONFIG_MMU | |
9174 | ||
9175 | static int io_uring_mmap(struct file *file, struct vm_area_struct *vma) | |
9176 | { | |
9177 | size_t sz = vma->vm_end - vma->vm_start; | |
9178 | unsigned long pfn; | |
9179 | void *ptr; | |
9180 | ||
9181 | ptr = io_uring_validate_mmap_request(file, vma->vm_pgoff, sz); | |
9182 | if (IS_ERR(ptr)) | |
9183 | return PTR_ERR(ptr); | |
2b188cc1 JA |
9184 | |
9185 | pfn = virt_to_phys(ptr) >> PAGE_SHIFT; | |
9186 | return remap_pfn_range(vma, vma->vm_start, pfn, sz, vma->vm_page_prot); | |
9187 | } | |
9188 | ||
6c5c240e RP |
9189 | #else /* !CONFIG_MMU */ |
9190 | ||
9191 | static int io_uring_mmap(struct file *file, struct vm_area_struct *vma) | |
9192 | { | |
9193 | return vma->vm_flags & (VM_SHARED | VM_MAYSHARE) ? 0 : -EINVAL; | |
9194 | } | |
9195 | ||
9196 | static unsigned int io_uring_nommu_mmap_capabilities(struct file *file) | |
9197 | { | |
9198 | return NOMMU_MAP_DIRECT | NOMMU_MAP_READ | NOMMU_MAP_WRITE; | |
9199 | } | |
9200 | ||
9201 | static unsigned long io_uring_nommu_get_unmapped_area(struct file *file, | |
9202 | unsigned long addr, unsigned long len, | |
9203 | unsigned long pgoff, unsigned long flags) | |
9204 | { | |
9205 | void *ptr; | |
9206 | ||
9207 | ptr = io_uring_validate_mmap_request(file, pgoff, len); | |
9208 | if (IS_ERR(ptr)) | |
9209 | return PTR_ERR(ptr); | |
9210 | ||
9211 | return (unsigned long) ptr; | |
9212 | } | |
9213 | ||
9214 | #endif /* !CONFIG_MMU */ | |
9215 | ||
d9d05217 | 9216 | static int io_sqpoll_wait_sq(struct io_ring_ctx *ctx) |
90554200 | 9217 | { |
d9d05217 | 9218 | int ret = 0; |
90554200 JA |
9219 | DEFINE_WAIT(wait); |
9220 | ||
9221 | do { | |
9222 | if (!io_sqring_full(ctx)) | |
9223 | break; | |
9224 | ||
9225 | prepare_to_wait(&ctx->sqo_sq_wait, &wait, TASK_INTERRUPTIBLE); | |
9226 | ||
d9d05217 PB |
9227 | if (unlikely(ctx->sqo_dead)) { |
9228 | ret = -EOWNERDEAD; | |
9229 | goto out; | |
9230 | } | |
9231 | ||
90554200 JA |
9232 | if (!io_sqring_full(ctx)) |
9233 | break; | |
9234 | ||
9235 | schedule(); | |
9236 | } while (!signal_pending(current)); | |
9237 | ||
9238 | finish_wait(&ctx->sqo_sq_wait, &wait); | |
d9d05217 PB |
9239 | out: |
9240 | return ret; | |
90554200 JA |
9241 | } |
9242 | ||
c73ebb68 HX |
9243 | static int io_get_ext_arg(unsigned flags, const void __user *argp, size_t *argsz, |
9244 | struct __kernel_timespec __user **ts, | |
9245 | const sigset_t __user **sig) | |
9246 | { | |
9247 | struct io_uring_getevents_arg arg; | |
9248 | ||
9249 | /* | |
9250 | * If EXT_ARG isn't set, then we have no timespec and the argp pointer | |
9251 | * is just a pointer to the sigset_t. | |
9252 | */ | |
9253 | if (!(flags & IORING_ENTER_EXT_ARG)) { | |
9254 | *sig = (const sigset_t __user *) argp; | |
9255 | *ts = NULL; | |
9256 | return 0; | |
9257 | } | |
9258 | ||
9259 | /* | |
9260 | * EXT_ARG is set - ensure we agree on the size of it and copy in our | |
9261 | * timespec and sigset_t pointers if good. | |
9262 | */ | |
9263 | if (*argsz != sizeof(arg)) | |
9264 | return -EINVAL; | |
9265 | if (copy_from_user(&arg, argp, sizeof(arg))) | |
9266 | return -EFAULT; | |
9267 | *sig = u64_to_user_ptr(arg.sigmask); | |
9268 | *argsz = arg.sigmask_sz; | |
9269 | *ts = u64_to_user_ptr(arg.ts); | |
9270 | return 0; | |
9271 | } | |
9272 | ||
2b188cc1 | 9273 | SYSCALL_DEFINE6(io_uring_enter, unsigned int, fd, u32, to_submit, |
c73ebb68 HX |
9274 | u32, min_complete, u32, flags, const void __user *, argp, |
9275 | size_t, argsz) | |
2b188cc1 JA |
9276 | { |
9277 | struct io_ring_ctx *ctx; | |
9278 | long ret = -EBADF; | |
9279 | int submitted = 0; | |
9280 | struct fd f; | |
9281 | ||
4c6e277c | 9282 | io_run_task_work(); |
b41e9852 | 9283 | |
90554200 | 9284 | if (flags & ~(IORING_ENTER_GETEVENTS | IORING_ENTER_SQ_WAKEUP | |
c73ebb68 | 9285 | IORING_ENTER_SQ_WAIT | IORING_ENTER_EXT_ARG)) |
2b188cc1 JA |
9286 | return -EINVAL; |
9287 | ||
9288 | f = fdget(fd); | |
9289 | if (!f.file) | |
9290 | return -EBADF; | |
9291 | ||
9292 | ret = -EOPNOTSUPP; | |
9293 | if (f.file->f_op != &io_uring_fops) | |
9294 | goto out_fput; | |
9295 | ||
9296 | ret = -ENXIO; | |
9297 | ctx = f.file->private_data; | |
9298 | if (!percpu_ref_tryget(&ctx->refs)) | |
9299 | goto out_fput; | |
9300 | ||
7e84e1c7 SG |
9301 | ret = -EBADFD; |
9302 | if (ctx->flags & IORING_SETUP_R_DISABLED) | |
9303 | goto out; | |
9304 | ||
6c271ce2 JA |
9305 | /* |
9306 | * For SQ polling, the thread will do all submissions and completions. | |
9307 | * Just return the requested submit count, and wake the thread if | |
9308 | * we were asked to. | |
9309 | */ | |
b2a9eada | 9310 | ret = 0; |
6c271ce2 | 9311 | if (ctx->flags & IORING_SETUP_SQPOLL) { |
6c503150 | 9312 | io_cqring_overflow_flush(ctx, false, NULL, NULL); |
89448c47 | 9313 | |
d9d05217 PB |
9314 | ret = -EOWNERDEAD; |
9315 | if (unlikely(ctx->sqo_dead)) | |
9316 | goto out; | |
6c271ce2 | 9317 | if (flags & IORING_ENTER_SQ_WAKEUP) |
534ca6d6 | 9318 | wake_up(&ctx->sq_data->wait); |
d9d05217 PB |
9319 | if (flags & IORING_ENTER_SQ_WAIT) { |
9320 | ret = io_sqpoll_wait_sq(ctx); | |
9321 | if (ret) | |
9322 | goto out; | |
9323 | } | |
6c271ce2 | 9324 | submitted = to_submit; |
b2a9eada | 9325 | } else if (to_submit) { |
fdaf083c | 9326 | ret = io_uring_add_task_file(ctx, f.file); |
0f212204 JA |
9327 | if (unlikely(ret)) |
9328 | goto out; | |
2b188cc1 | 9329 | mutex_lock(&ctx->uring_lock); |
0f212204 | 9330 | submitted = io_submit_sqes(ctx, to_submit); |
2b188cc1 | 9331 | mutex_unlock(&ctx->uring_lock); |
7c504e65 PB |
9332 | |
9333 | if (submitted != to_submit) | |
9334 | goto out; | |
2b188cc1 JA |
9335 | } |
9336 | if (flags & IORING_ENTER_GETEVENTS) { | |
c73ebb68 HX |
9337 | const sigset_t __user *sig; |
9338 | struct __kernel_timespec __user *ts; | |
9339 | ||
9340 | ret = io_get_ext_arg(flags, argp, &argsz, &ts, &sig); | |
9341 | if (unlikely(ret)) | |
9342 | goto out; | |
9343 | ||
2b188cc1 JA |
9344 | min_complete = min(min_complete, ctx->cq_entries); |
9345 | ||
32b2244a XW |
9346 | /* |
9347 | * When SETUP_IOPOLL and SETUP_SQPOLL are both enabled, user | |
9348 | * space applications don't need to do io completion events | |
9349 | * polling again, they can rely on io_sq_thread to do polling | |
9350 | * work, which can reduce cpu usage and uring_lock contention. | |
9351 | */ | |
9352 | if (ctx->flags & IORING_SETUP_IOPOLL && | |
9353 | !(ctx->flags & IORING_SETUP_SQPOLL)) { | |
7668b92a | 9354 | ret = io_iopoll_check(ctx, min_complete); |
def596e9 | 9355 | } else { |
c73ebb68 | 9356 | ret = io_cqring_wait(ctx, min_complete, sig, argsz, ts); |
def596e9 | 9357 | } |
2b188cc1 JA |
9358 | } |
9359 | ||
7c504e65 | 9360 | out: |
6805b32e | 9361 | percpu_ref_put(&ctx->refs); |
2b188cc1 JA |
9362 | out_fput: |
9363 | fdput(f); | |
9364 | return submitted ? submitted : ret; | |
9365 | } | |
9366 | ||
bebdb65e | 9367 | #ifdef CONFIG_PROC_FS |
87ce955b JA |
9368 | static int io_uring_show_cred(int id, void *p, void *data) |
9369 | { | |
6b47ab81 JA |
9370 | struct io_identity *iod = p; |
9371 | const struct cred *cred = iod->creds; | |
87ce955b JA |
9372 | struct seq_file *m = data; |
9373 | struct user_namespace *uns = seq_user_ns(m); | |
9374 | struct group_info *gi; | |
9375 | kernel_cap_t cap; | |
9376 | unsigned __capi; | |
9377 | int g; | |
9378 | ||
9379 | seq_printf(m, "%5d\n", id); | |
9380 | seq_put_decimal_ull(m, "\tUid:\t", from_kuid_munged(uns, cred->uid)); | |
9381 | seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->euid)); | |
9382 | seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->suid)); | |
9383 | seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->fsuid)); | |
9384 | seq_put_decimal_ull(m, "\n\tGid:\t", from_kgid_munged(uns, cred->gid)); | |
9385 | seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->egid)); | |
9386 | seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->sgid)); | |
9387 | seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->fsgid)); | |
9388 | seq_puts(m, "\n\tGroups:\t"); | |
9389 | gi = cred->group_info; | |
9390 | for (g = 0; g < gi->ngroups; g++) { | |
9391 | seq_put_decimal_ull(m, g ? " " : "", | |
9392 | from_kgid_munged(uns, gi->gid[g])); | |
9393 | } | |
9394 | seq_puts(m, "\n\tCapEff:\t"); | |
9395 | cap = cred->cap_effective; | |
9396 | CAP_FOR_EACH_U32(__capi) | |
9397 | seq_put_hex_ll(m, NULL, cap.cap[CAP_LAST_U32 - __capi], 8); | |
9398 | seq_putc(m, '\n'); | |
9399 | return 0; | |
9400 | } | |
9401 | ||
9402 | static void __io_uring_show_fdinfo(struct io_ring_ctx *ctx, struct seq_file *m) | |
9403 | { | |
dbbe9c64 | 9404 | struct io_sq_data *sq = NULL; |
fad8e0de | 9405 | bool has_lock; |
87ce955b JA |
9406 | int i; |
9407 | ||
fad8e0de JA |
9408 | /* |
9409 | * Avoid ABBA deadlock between the seq lock and the io_uring mutex, | |
9410 | * since fdinfo case grabs it in the opposite direction of normal use | |
9411 | * cases. If we fail to get the lock, we just don't iterate any | |
9412 | * structures that could be going away outside the io_uring mutex. | |
9413 | */ | |
9414 | has_lock = mutex_trylock(&ctx->uring_lock); | |
9415 | ||
dbbe9c64 JQ |
9416 | if (has_lock && (ctx->flags & IORING_SETUP_SQPOLL)) |
9417 | sq = ctx->sq_data; | |
9418 | ||
9419 | seq_printf(m, "SqThread:\t%d\n", sq ? task_pid_nr(sq->thread) : -1); | |
9420 | seq_printf(m, "SqThreadCpu:\t%d\n", sq ? task_cpu(sq->thread) : -1); | |
87ce955b | 9421 | seq_printf(m, "UserFiles:\t%u\n", ctx->nr_user_files); |
fad8e0de | 9422 | for (i = 0; has_lock && i < ctx->nr_user_files; i++) { |
87ce955b JA |
9423 | struct fixed_file_table *table; |
9424 | struct file *f; | |
9425 | ||
9426 | table = &ctx->file_data->table[i >> IORING_FILE_TABLE_SHIFT]; | |
9427 | f = table->files[i & IORING_FILE_TABLE_MASK]; | |
9428 | if (f) | |
9429 | seq_printf(m, "%5u: %s\n", i, file_dentry(f)->d_iname); | |
9430 | else | |
9431 | seq_printf(m, "%5u: <none>\n", i); | |
9432 | } | |
9433 | seq_printf(m, "UserBufs:\t%u\n", ctx->nr_user_bufs); | |
fad8e0de | 9434 | for (i = 0; has_lock && i < ctx->nr_user_bufs; i++) { |
87ce955b JA |
9435 | struct io_mapped_ubuf *buf = &ctx->user_bufs[i]; |
9436 | ||
9437 | seq_printf(m, "%5u: 0x%llx/%u\n", i, buf->ubuf, | |
9438 | (unsigned int) buf->len); | |
9439 | } | |
fad8e0de | 9440 | if (has_lock && !idr_is_empty(&ctx->personality_idr)) { |
87ce955b JA |
9441 | seq_printf(m, "Personalities:\n"); |
9442 | idr_for_each(&ctx->personality_idr, io_uring_show_cred, m); | |
9443 | } | |
d7718a9d JA |
9444 | seq_printf(m, "PollList:\n"); |
9445 | spin_lock_irq(&ctx->completion_lock); | |
9446 | for (i = 0; i < (1U << ctx->cancel_hash_bits); i++) { | |
9447 | struct hlist_head *list = &ctx->cancel_hash[i]; | |
9448 | struct io_kiocb *req; | |
9449 | ||
9450 | hlist_for_each_entry(req, list, hash_node) | |
9451 | seq_printf(m, " op=%d, task_works=%d\n", req->opcode, | |
9452 | req->task->task_works != NULL); | |
9453 | } | |
9454 | spin_unlock_irq(&ctx->completion_lock); | |
fad8e0de JA |
9455 | if (has_lock) |
9456 | mutex_unlock(&ctx->uring_lock); | |
87ce955b JA |
9457 | } |
9458 | ||
9459 | static void io_uring_show_fdinfo(struct seq_file *m, struct file *f) | |
9460 | { | |
9461 | struct io_ring_ctx *ctx = f->private_data; | |
9462 | ||
9463 | if (percpu_ref_tryget(&ctx->refs)) { | |
9464 | __io_uring_show_fdinfo(ctx, m); | |
9465 | percpu_ref_put(&ctx->refs); | |
9466 | } | |
9467 | } | |
bebdb65e | 9468 | #endif |
87ce955b | 9469 | |
2b188cc1 JA |
9470 | static const struct file_operations io_uring_fops = { |
9471 | .release = io_uring_release, | |
fcb323cc | 9472 | .flush = io_uring_flush, |
2b188cc1 | 9473 | .mmap = io_uring_mmap, |
6c5c240e RP |
9474 | #ifndef CONFIG_MMU |
9475 | .get_unmapped_area = io_uring_nommu_get_unmapped_area, | |
9476 | .mmap_capabilities = io_uring_nommu_mmap_capabilities, | |
9477 | #endif | |
2b188cc1 JA |
9478 | .poll = io_uring_poll, |
9479 | .fasync = io_uring_fasync, | |
bebdb65e | 9480 | #ifdef CONFIG_PROC_FS |
87ce955b | 9481 | .show_fdinfo = io_uring_show_fdinfo, |
bebdb65e | 9482 | #endif |
2b188cc1 JA |
9483 | }; |
9484 | ||
9485 | static int io_allocate_scq_urings(struct io_ring_ctx *ctx, | |
9486 | struct io_uring_params *p) | |
9487 | { | |
75b28aff HV |
9488 | struct io_rings *rings; |
9489 | size_t size, sq_array_offset; | |
2b188cc1 | 9490 | |
bd740481 JA |
9491 | /* make sure these are sane, as we already accounted them */ |
9492 | ctx->sq_entries = p->sq_entries; | |
9493 | ctx->cq_entries = p->cq_entries; | |
9494 | ||
75b28aff HV |
9495 | size = rings_size(p->sq_entries, p->cq_entries, &sq_array_offset); |
9496 | if (size == SIZE_MAX) | |
9497 | return -EOVERFLOW; | |
9498 | ||
9499 | rings = io_mem_alloc(size); | |
9500 | if (!rings) | |
2b188cc1 JA |
9501 | return -ENOMEM; |
9502 | ||
75b28aff HV |
9503 | ctx->rings = rings; |
9504 | ctx->sq_array = (u32 *)((char *)rings + sq_array_offset); | |
9505 | rings->sq_ring_mask = p->sq_entries - 1; | |
9506 | rings->cq_ring_mask = p->cq_entries - 1; | |
9507 | rings->sq_ring_entries = p->sq_entries; | |
9508 | rings->cq_ring_entries = p->cq_entries; | |
9509 | ctx->sq_mask = rings->sq_ring_mask; | |
9510 | ctx->cq_mask = rings->cq_ring_mask; | |
2b188cc1 JA |
9511 | |
9512 | size = array_size(sizeof(struct io_uring_sqe), p->sq_entries); | |
eb065d30 JA |
9513 | if (size == SIZE_MAX) { |
9514 | io_mem_free(ctx->rings); | |
9515 | ctx->rings = NULL; | |
2b188cc1 | 9516 | return -EOVERFLOW; |
eb065d30 | 9517 | } |
2b188cc1 JA |
9518 | |
9519 | ctx->sq_sqes = io_mem_alloc(size); | |
eb065d30 JA |
9520 | if (!ctx->sq_sqes) { |
9521 | io_mem_free(ctx->rings); | |
9522 | ctx->rings = NULL; | |
2b188cc1 | 9523 | return -ENOMEM; |
eb065d30 | 9524 | } |
2b188cc1 | 9525 | |
2b188cc1 JA |
9526 | return 0; |
9527 | } | |
9528 | ||
9faadcc8 PB |
9529 | static int io_uring_install_fd(struct io_ring_ctx *ctx, struct file *file) |
9530 | { | |
9531 | int ret, fd; | |
9532 | ||
9533 | fd = get_unused_fd_flags(O_RDWR | O_CLOEXEC); | |
9534 | if (fd < 0) | |
9535 | return fd; | |
9536 | ||
9537 | ret = io_uring_add_task_file(ctx, file); | |
9538 | if (ret) { | |
9539 | put_unused_fd(fd); | |
9540 | return ret; | |
9541 | } | |
9542 | fd_install(fd, file); | |
9543 | return fd; | |
9544 | } | |
9545 | ||
2b188cc1 JA |
9546 | /* |
9547 | * Allocate an anonymous fd, this is what constitutes the application | |
9548 | * visible backing of an io_uring instance. The application mmaps this | |
9549 | * fd to gain access to the SQ/CQ ring details. If UNIX sockets are enabled, | |
9550 | * we have to tie this fd to a socket for file garbage collection purposes. | |
9551 | */ | |
9faadcc8 | 9552 | static struct file *io_uring_get_file(struct io_ring_ctx *ctx) |
2b188cc1 JA |
9553 | { |
9554 | struct file *file; | |
9faadcc8 | 9555 | #if defined(CONFIG_UNIX) |
2b188cc1 JA |
9556 | int ret; |
9557 | ||
2b188cc1 JA |
9558 | ret = sock_create_kern(&init_net, PF_UNIX, SOCK_RAW, IPPROTO_IP, |
9559 | &ctx->ring_sock); | |
9560 | if (ret) | |
9faadcc8 | 9561 | return ERR_PTR(ret); |
2b188cc1 JA |
9562 | #endif |
9563 | ||
2b188cc1 JA |
9564 | file = anon_inode_getfile("[io_uring]", &io_uring_fops, ctx, |
9565 | O_RDWR | O_CLOEXEC); | |
2b188cc1 | 9566 | #if defined(CONFIG_UNIX) |
9faadcc8 PB |
9567 | if (IS_ERR(file)) { |
9568 | sock_release(ctx->ring_sock); | |
9569 | ctx->ring_sock = NULL; | |
9570 | } else { | |
9571 | ctx->ring_sock->file = file; | |
0f212204 | 9572 | } |
2b188cc1 | 9573 | #endif |
9faadcc8 | 9574 | return file; |
2b188cc1 JA |
9575 | } |
9576 | ||
7f13657d XW |
9577 | static int io_uring_create(unsigned entries, struct io_uring_params *p, |
9578 | struct io_uring_params __user *params) | |
2b188cc1 JA |
9579 | { |
9580 | struct user_struct *user = NULL; | |
9581 | struct io_ring_ctx *ctx; | |
9faadcc8 | 9582 | struct file *file; |
aad5d8da | 9583 | bool limit_mem; |
2b188cc1 JA |
9584 | int ret; |
9585 | ||
8110c1a6 | 9586 | if (!entries) |
2b188cc1 | 9587 | return -EINVAL; |
8110c1a6 JA |
9588 | if (entries > IORING_MAX_ENTRIES) { |
9589 | if (!(p->flags & IORING_SETUP_CLAMP)) | |
9590 | return -EINVAL; | |
9591 | entries = IORING_MAX_ENTRIES; | |
9592 | } | |
2b188cc1 JA |
9593 | |
9594 | /* | |
9595 | * Use twice as many entries for the CQ ring. It's possible for the | |
9596 | * application to drive a higher depth than the size of the SQ ring, | |
9597 | * since the sqes are only used at submission time. This allows for | |
33a107f0 JA |
9598 | * some flexibility in overcommitting a bit. If the application has |
9599 | * set IORING_SETUP_CQSIZE, it will have passed in the desired number | |
9600 | * of CQ ring entries manually. | |
2b188cc1 JA |
9601 | */ |
9602 | p->sq_entries = roundup_pow_of_two(entries); | |
33a107f0 JA |
9603 | if (p->flags & IORING_SETUP_CQSIZE) { |
9604 | /* | |
9605 | * If IORING_SETUP_CQSIZE is set, we do the same roundup | |
9606 | * to a power-of-two, if it isn't already. We do NOT impose | |
9607 | * any cq vs sq ring sizing. | |
9608 | */ | |
eb2667b3 | 9609 | if (!p->cq_entries) |
33a107f0 | 9610 | return -EINVAL; |
8110c1a6 JA |
9611 | if (p->cq_entries > IORING_MAX_CQ_ENTRIES) { |
9612 | if (!(p->flags & IORING_SETUP_CLAMP)) | |
9613 | return -EINVAL; | |
9614 | p->cq_entries = IORING_MAX_CQ_ENTRIES; | |
9615 | } | |
eb2667b3 JQ |
9616 | p->cq_entries = roundup_pow_of_two(p->cq_entries); |
9617 | if (p->cq_entries < p->sq_entries) | |
9618 | return -EINVAL; | |
33a107f0 JA |
9619 | } else { |
9620 | p->cq_entries = 2 * p->sq_entries; | |
9621 | } | |
2b188cc1 JA |
9622 | |
9623 | user = get_uid(current_user()); | |
aad5d8da | 9624 | limit_mem = !capable(CAP_IPC_LOCK); |
2b188cc1 | 9625 | |
aad5d8da | 9626 | if (limit_mem) { |
a087e2b5 | 9627 | ret = __io_account_mem(user, |
2b188cc1 JA |
9628 | ring_pages(p->sq_entries, p->cq_entries)); |
9629 | if (ret) { | |
9630 | free_uid(user); | |
9631 | return ret; | |
9632 | } | |
9633 | } | |
9634 | ||
9635 | ctx = io_ring_ctx_alloc(p); | |
9636 | if (!ctx) { | |
aad5d8da | 9637 | if (limit_mem) |
a087e2b5 | 9638 | __io_unaccount_mem(user, ring_pages(p->sq_entries, |
2b188cc1 JA |
9639 | p->cq_entries)); |
9640 | free_uid(user); | |
9641 | return -ENOMEM; | |
9642 | } | |
9643 | ctx->compat = in_compat_syscall(); | |
2b188cc1 | 9644 | ctx->user = user; |
0b8c0ec7 | 9645 | ctx->creds = get_current_cred(); |
4ea33a97 JA |
9646 | #ifdef CONFIG_AUDIT |
9647 | ctx->loginuid = current->loginuid; | |
9648 | ctx->sessionid = current->sessionid; | |
9649 | #endif | |
2aede0e4 JA |
9650 | ctx->sqo_task = get_task_struct(current); |
9651 | ||
9652 | /* | |
9653 | * This is just grabbed for accounting purposes. When a process exits, | |
9654 | * the mm is exited and dropped before the files, hence we need to hang | |
9655 | * on to this mm purely for the purposes of being able to unaccount | |
9656 | * memory (locked/pinned vm). It's not used for anything else. | |
9657 | */ | |
6b7898eb | 9658 | mmgrab(current->mm); |
2aede0e4 | 9659 | ctx->mm_account = current->mm; |
6b7898eb | 9660 | |
91d8f519 DZ |
9661 | #ifdef CONFIG_BLK_CGROUP |
9662 | /* | |
9663 | * The sq thread will belong to the original cgroup it was inited in. | |
9664 | * If the cgroup goes offline (e.g. disabling the io controller), then | |
9665 | * issued bios will be associated with the closest cgroup later in the | |
9666 | * block layer. | |
9667 | */ | |
9668 | rcu_read_lock(); | |
9669 | ctx->sqo_blkcg_css = blkcg_css(); | |
9670 | ret = css_tryget_online(ctx->sqo_blkcg_css); | |
9671 | rcu_read_unlock(); | |
9672 | if (!ret) { | |
9673 | /* don't init against a dying cgroup, have the user try again */ | |
9674 | ctx->sqo_blkcg_css = NULL; | |
9675 | ret = -ENODEV; | |
9676 | goto err; | |
9677 | } | |
9678 | #endif | |
6b7898eb | 9679 | |
f74441e6 JA |
9680 | /* |
9681 | * Account memory _before_ installing the file descriptor. Once | |
9682 | * the descriptor is installed, it can get closed at any time. Also | |
9683 | * do this before hitting the general error path, as ring freeing | |
9684 | * will un-account as well. | |
9685 | */ | |
9686 | io_account_mem(ctx, ring_pages(p->sq_entries, p->cq_entries), | |
9687 | ACCT_LOCKED); | |
9688 | ctx->limit_mem = limit_mem; | |
9689 | ||
2b188cc1 JA |
9690 | ret = io_allocate_scq_urings(ctx, p); |
9691 | if (ret) | |
9692 | goto err; | |
9693 | ||
7e84e1c7 | 9694 | ret = io_sq_offload_create(ctx, p); |
2b188cc1 JA |
9695 | if (ret) |
9696 | goto err; | |
9697 | ||
7e84e1c7 SG |
9698 | if (!(p->flags & IORING_SETUP_R_DISABLED)) |
9699 | io_sq_offload_start(ctx); | |
9700 | ||
2b188cc1 | 9701 | memset(&p->sq_off, 0, sizeof(p->sq_off)); |
75b28aff HV |
9702 | p->sq_off.head = offsetof(struct io_rings, sq.head); |
9703 | p->sq_off.tail = offsetof(struct io_rings, sq.tail); | |
9704 | p->sq_off.ring_mask = offsetof(struct io_rings, sq_ring_mask); | |
9705 | p->sq_off.ring_entries = offsetof(struct io_rings, sq_ring_entries); | |
9706 | p->sq_off.flags = offsetof(struct io_rings, sq_flags); | |
9707 | p->sq_off.dropped = offsetof(struct io_rings, sq_dropped); | |
9708 | p->sq_off.array = (char *)ctx->sq_array - (char *)ctx->rings; | |
2b188cc1 JA |
9709 | |
9710 | memset(&p->cq_off, 0, sizeof(p->cq_off)); | |
75b28aff HV |
9711 | p->cq_off.head = offsetof(struct io_rings, cq.head); |
9712 | p->cq_off.tail = offsetof(struct io_rings, cq.tail); | |
9713 | p->cq_off.ring_mask = offsetof(struct io_rings, cq_ring_mask); | |
9714 | p->cq_off.ring_entries = offsetof(struct io_rings, cq_ring_entries); | |
9715 | p->cq_off.overflow = offsetof(struct io_rings, cq_overflow); | |
9716 | p->cq_off.cqes = offsetof(struct io_rings, cqes); | |
0d9b5b3a | 9717 | p->cq_off.flags = offsetof(struct io_rings, cq_flags); |
ac90f249 | 9718 | |
7f13657d XW |
9719 | p->features = IORING_FEAT_SINGLE_MMAP | IORING_FEAT_NODROP | |
9720 | IORING_FEAT_SUBMIT_STABLE | IORING_FEAT_RW_CUR_POS | | |
5769a351 | 9721 | IORING_FEAT_CUR_PERSONALITY | IORING_FEAT_FAST_POLL | |
c73ebb68 HX |
9722 | IORING_FEAT_POLL_32BITS | IORING_FEAT_SQPOLL_NONFIXED | |
9723 | IORING_FEAT_EXT_ARG; | |
7f13657d XW |
9724 | |
9725 | if (copy_to_user(params, p, sizeof(*p))) { | |
9726 | ret = -EFAULT; | |
9727 | goto err; | |
9728 | } | |
d1719f70 | 9729 | |
9faadcc8 PB |
9730 | file = io_uring_get_file(ctx); |
9731 | if (IS_ERR(file)) { | |
9732 | ret = PTR_ERR(file); | |
9733 | goto err; | |
9734 | } | |
9735 | ||
044c1ab3 JA |
9736 | /* |
9737 | * Install ring fd as the very last thing, so we don't risk someone | |
9738 | * having closed it before we finish setup | |
9739 | */ | |
9faadcc8 PB |
9740 | ret = io_uring_install_fd(ctx, file); |
9741 | if (ret < 0) { | |
06585c49 | 9742 | io_disable_sqo_submit(ctx); |
9faadcc8 PB |
9743 | /* fput will clean it up */ |
9744 | fput(file); | |
9745 | return ret; | |
9746 | } | |
044c1ab3 | 9747 | |
c826bd7a | 9748 | trace_io_uring_create(ret, ctx, p->sq_entries, p->cq_entries, p->flags); |
2b188cc1 JA |
9749 | return ret; |
9750 | err: | |
d9d05217 | 9751 | io_disable_sqo_submit(ctx); |
2b188cc1 JA |
9752 | io_ring_ctx_wait_and_kill(ctx); |
9753 | return ret; | |
9754 | } | |
9755 | ||
9756 | /* | |
9757 | * Sets up an aio uring context, and returns the fd. Applications asks for a | |
9758 | * ring size, we return the actual sq/cq ring sizes (among other things) in the | |
9759 | * params structure passed in. | |
9760 | */ | |
9761 | static long io_uring_setup(u32 entries, struct io_uring_params __user *params) | |
9762 | { | |
9763 | struct io_uring_params p; | |
2b188cc1 JA |
9764 | int i; |
9765 | ||
9766 | if (copy_from_user(&p, params, sizeof(p))) | |
9767 | return -EFAULT; | |
9768 | for (i = 0; i < ARRAY_SIZE(p.resv); i++) { | |
9769 | if (p.resv[i]) | |
9770 | return -EINVAL; | |
9771 | } | |
9772 | ||
6c271ce2 | 9773 | if (p.flags & ~(IORING_SETUP_IOPOLL | IORING_SETUP_SQPOLL | |
8110c1a6 | 9774 | IORING_SETUP_SQ_AFF | IORING_SETUP_CQSIZE | |
7e84e1c7 SG |
9775 | IORING_SETUP_CLAMP | IORING_SETUP_ATTACH_WQ | |
9776 | IORING_SETUP_R_DISABLED)) | |
2b188cc1 JA |
9777 | return -EINVAL; |
9778 | ||
7f13657d | 9779 | return io_uring_create(entries, &p, params); |
2b188cc1 JA |
9780 | } |
9781 | ||
9782 | SYSCALL_DEFINE2(io_uring_setup, u32, entries, | |
9783 | struct io_uring_params __user *, params) | |
9784 | { | |
9785 | return io_uring_setup(entries, params); | |
9786 | } | |
9787 | ||
66f4af93 JA |
9788 | static int io_probe(struct io_ring_ctx *ctx, void __user *arg, unsigned nr_args) |
9789 | { | |
9790 | struct io_uring_probe *p; | |
9791 | size_t size; | |
9792 | int i, ret; | |
9793 | ||
9794 | size = struct_size(p, ops, nr_args); | |
9795 | if (size == SIZE_MAX) | |
9796 | return -EOVERFLOW; | |
9797 | p = kzalloc(size, GFP_KERNEL); | |
9798 | if (!p) | |
9799 | return -ENOMEM; | |
9800 | ||
9801 | ret = -EFAULT; | |
9802 | if (copy_from_user(p, arg, size)) | |
9803 | goto out; | |
9804 | ret = -EINVAL; | |
9805 | if (memchr_inv(p, 0, size)) | |
9806 | goto out; | |
9807 | ||
9808 | p->last_op = IORING_OP_LAST - 1; | |
9809 | if (nr_args > IORING_OP_LAST) | |
9810 | nr_args = IORING_OP_LAST; | |
9811 | ||
9812 | for (i = 0; i < nr_args; i++) { | |
9813 | p->ops[i].op = i; | |
9814 | if (!io_op_defs[i].not_supported) | |
9815 | p->ops[i].flags = IO_URING_OP_SUPPORTED; | |
9816 | } | |
9817 | p->ops_len = i; | |
9818 | ||
9819 | ret = 0; | |
9820 | if (copy_to_user(arg, p, size)) | |
9821 | ret = -EFAULT; | |
9822 | out: | |
9823 | kfree(p); | |
9824 | return ret; | |
9825 | } | |
9826 | ||
071698e1 JA |
9827 | static int io_register_personality(struct io_ring_ctx *ctx) |
9828 | { | |
1e6fa521 JA |
9829 | struct io_identity *id; |
9830 | int ret; | |
071698e1 | 9831 | |
1e6fa521 JA |
9832 | id = kmalloc(sizeof(*id), GFP_KERNEL); |
9833 | if (unlikely(!id)) | |
9834 | return -ENOMEM; | |
9835 | ||
9836 | io_init_identity(id); | |
9837 | id->creds = get_current_cred(); | |
9838 | ||
9839 | ret = idr_alloc_cyclic(&ctx->personality_idr, id, 1, USHRT_MAX, GFP_KERNEL); | |
9840 | if (ret < 0) { | |
9841 | put_cred(id->creds); | |
9842 | kfree(id); | |
9843 | } | |
9844 | return ret; | |
071698e1 JA |
9845 | } |
9846 | ||
9847 | static int io_unregister_personality(struct io_ring_ctx *ctx, unsigned id) | |
9848 | { | |
1e6fa521 | 9849 | struct io_identity *iod; |
071698e1 | 9850 | |
1e6fa521 JA |
9851 | iod = idr_remove(&ctx->personality_idr, id); |
9852 | if (iod) { | |
9853 | put_cred(iod->creds); | |
9854 | if (refcount_dec_and_test(&iod->count)) | |
9855 | kfree(iod); | |
071698e1 JA |
9856 | return 0; |
9857 | } | |
9858 | ||
9859 | return -EINVAL; | |
9860 | } | |
9861 | ||
21b55dbc SG |
9862 | static int io_register_restrictions(struct io_ring_ctx *ctx, void __user *arg, |
9863 | unsigned int nr_args) | |
9864 | { | |
9865 | struct io_uring_restriction *res; | |
9866 | size_t size; | |
9867 | int i, ret; | |
9868 | ||
7e84e1c7 SG |
9869 | /* Restrictions allowed only if rings started disabled */ |
9870 | if (!(ctx->flags & IORING_SETUP_R_DISABLED)) | |
9871 | return -EBADFD; | |
9872 | ||
21b55dbc | 9873 | /* We allow only a single restrictions registration */ |
7e84e1c7 | 9874 | if (ctx->restrictions.registered) |
21b55dbc SG |
9875 | return -EBUSY; |
9876 | ||
9877 | if (!arg || nr_args > IORING_MAX_RESTRICTIONS) | |
9878 | return -EINVAL; | |
9879 | ||
9880 | size = array_size(nr_args, sizeof(*res)); | |
9881 | if (size == SIZE_MAX) | |
9882 | return -EOVERFLOW; | |
9883 | ||
9884 | res = memdup_user(arg, size); | |
9885 | if (IS_ERR(res)) | |
9886 | return PTR_ERR(res); | |
9887 | ||
9888 | ret = 0; | |
9889 | ||
9890 | for (i = 0; i < nr_args; i++) { | |
9891 | switch (res[i].opcode) { | |
9892 | case IORING_RESTRICTION_REGISTER_OP: | |
9893 | if (res[i].register_op >= IORING_REGISTER_LAST) { | |
9894 | ret = -EINVAL; | |
9895 | goto out; | |
9896 | } | |
9897 | ||
9898 | __set_bit(res[i].register_op, | |
9899 | ctx->restrictions.register_op); | |
9900 | break; | |
9901 | case IORING_RESTRICTION_SQE_OP: | |
9902 | if (res[i].sqe_op >= IORING_OP_LAST) { | |
9903 | ret = -EINVAL; | |
9904 | goto out; | |
9905 | } | |
9906 | ||
9907 | __set_bit(res[i].sqe_op, ctx->restrictions.sqe_op); | |
9908 | break; | |
9909 | case IORING_RESTRICTION_SQE_FLAGS_ALLOWED: | |
9910 | ctx->restrictions.sqe_flags_allowed = res[i].sqe_flags; | |
9911 | break; | |
9912 | case IORING_RESTRICTION_SQE_FLAGS_REQUIRED: | |
9913 | ctx->restrictions.sqe_flags_required = res[i].sqe_flags; | |
9914 | break; | |
9915 | default: | |
9916 | ret = -EINVAL; | |
9917 | goto out; | |
9918 | } | |
9919 | } | |
9920 | ||
9921 | out: | |
9922 | /* Reset all restrictions if an error happened */ | |
9923 | if (ret != 0) | |
9924 | memset(&ctx->restrictions, 0, sizeof(ctx->restrictions)); | |
9925 | else | |
7e84e1c7 | 9926 | ctx->restrictions.registered = true; |
21b55dbc SG |
9927 | |
9928 | kfree(res); | |
9929 | return ret; | |
9930 | } | |
9931 | ||
7e84e1c7 SG |
9932 | static int io_register_enable_rings(struct io_ring_ctx *ctx) |
9933 | { | |
9934 | if (!(ctx->flags & IORING_SETUP_R_DISABLED)) | |
9935 | return -EBADFD; | |
9936 | ||
9937 | if (ctx->restrictions.registered) | |
9938 | ctx->restricted = 1; | |
9939 | ||
9940 | ctx->flags &= ~IORING_SETUP_R_DISABLED; | |
9941 | ||
9942 | io_sq_offload_start(ctx); | |
9943 | ||
9944 | return 0; | |
9945 | } | |
9946 | ||
071698e1 JA |
9947 | static bool io_register_op_must_quiesce(int op) |
9948 | { | |
9949 | switch (op) { | |
9950 | case IORING_UNREGISTER_FILES: | |
9951 | case IORING_REGISTER_FILES_UPDATE: | |
9952 | case IORING_REGISTER_PROBE: | |
9953 | case IORING_REGISTER_PERSONALITY: | |
9954 | case IORING_UNREGISTER_PERSONALITY: | |
9955 | return false; | |
9956 | default: | |
9957 | return true; | |
9958 | } | |
9959 | } | |
9960 | ||
edafccee JA |
9961 | static int __io_uring_register(struct io_ring_ctx *ctx, unsigned opcode, |
9962 | void __user *arg, unsigned nr_args) | |
b19062a5 JA |
9963 | __releases(ctx->uring_lock) |
9964 | __acquires(ctx->uring_lock) | |
edafccee JA |
9965 | { |
9966 | int ret; | |
9967 | ||
35fa71a0 JA |
9968 | /* |
9969 | * We're inside the ring mutex, if the ref is already dying, then | |
9970 | * someone else killed the ctx or is already going through | |
9971 | * io_uring_register(). | |
9972 | */ | |
9973 | if (percpu_ref_is_dying(&ctx->refs)) | |
9974 | return -ENXIO; | |
9975 | ||
071698e1 | 9976 | if (io_register_op_must_quiesce(opcode)) { |
05f3fb3c | 9977 | percpu_ref_kill(&ctx->refs); |
b19062a5 | 9978 | |
05f3fb3c JA |
9979 | /* |
9980 | * Drop uring mutex before waiting for references to exit. If | |
9981 | * another thread is currently inside io_uring_enter() it might | |
9982 | * need to grab the uring_lock to make progress. If we hold it | |
9983 | * here across the drain wait, then we can deadlock. It's safe | |
9984 | * to drop the mutex here, since no new references will come in | |
9985 | * after we've killed the percpu ref. | |
9986 | */ | |
9987 | mutex_unlock(&ctx->uring_lock); | |
af9c1a44 JA |
9988 | do { |
9989 | ret = wait_for_completion_interruptible(&ctx->ref_comp); | |
9990 | if (!ret) | |
9991 | break; | |
ed6930c9 JA |
9992 | ret = io_run_task_work_sig(); |
9993 | if (ret < 0) | |
9994 | break; | |
af9c1a44 JA |
9995 | } while (1); |
9996 | ||
05f3fb3c | 9997 | mutex_lock(&ctx->uring_lock); |
af9c1a44 | 9998 | |
c150368b JA |
9999 | if (ret) { |
10000 | percpu_ref_resurrect(&ctx->refs); | |
21b55dbc SG |
10001 | goto out_quiesce; |
10002 | } | |
10003 | } | |
10004 | ||
10005 | if (ctx->restricted) { | |
10006 | if (opcode >= IORING_REGISTER_LAST) { | |
10007 | ret = -EINVAL; | |
10008 | goto out; | |
10009 | } | |
10010 | ||
10011 | if (!test_bit(opcode, ctx->restrictions.register_op)) { | |
10012 | ret = -EACCES; | |
c150368b JA |
10013 | goto out; |
10014 | } | |
05f3fb3c | 10015 | } |
edafccee JA |
10016 | |
10017 | switch (opcode) { | |
10018 | case IORING_REGISTER_BUFFERS: | |
10019 | ret = io_sqe_buffer_register(ctx, arg, nr_args); | |
10020 | break; | |
10021 | case IORING_UNREGISTER_BUFFERS: | |
10022 | ret = -EINVAL; | |
10023 | if (arg || nr_args) | |
10024 | break; | |
10025 | ret = io_sqe_buffer_unregister(ctx); | |
10026 | break; | |
6b06314c JA |
10027 | case IORING_REGISTER_FILES: |
10028 | ret = io_sqe_files_register(ctx, arg, nr_args); | |
10029 | break; | |
10030 | case IORING_UNREGISTER_FILES: | |
10031 | ret = -EINVAL; | |
10032 | if (arg || nr_args) | |
10033 | break; | |
10034 | ret = io_sqe_files_unregister(ctx); | |
10035 | break; | |
c3a31e60 JA |
10036 | case IORING_REGISTER_FILES_UPDATE: |
10037 | ret = io_sqe_files_update(ctx, arg, nr_args); | |
10038 | break; | |
9b402849 | 10039 | case IORING_REGISTER_EVENTFD: |
f2842ab5 | 10040 | case IORING_REGISTER_EVENTFD_ASYNC: |
9b402849 JA |
10041 | ret = -EINVAL; |
10042 | if (nr_args != 1) | |
10043 | break; | |
10044 | ret = io_eventfd_register(ctx, arg); | |
f2842ab5 JA |
10045 | if (ret) |
10046 | break; | |
10047 | if (opcode == IORING_REGISTER_EVENTFD_ASYNC) | |
10048 | ctx->eventfd_async = 1; | |
10049 | else | |
10050 | ctx->eventfd_async = 0; | |
9b402849 JA |
10051 | break; |
10052 | case IORING_UNREGISTER_EVENTFD: | |
10053 | ret = -EINVAL; | |
10054 | if (arg || nr_args) | |
10055 | break; | |
10056 | ret = io_eventfd_unregister(ctx); | |
10057 | break; | |
66f4af93 JA |
10058 | case IORING_REGISTER_PROBE: |
10059 | ret = -EINVAL; | |
10060 | if (!arg || nr_args > 256) | |
10061 | break; | |
10062 | ret = io_probe(ctx, arg, nr_args); | |
10063 | break; | |
071698e1 JA |
10064 | case IORING_REGISTER_PERSONALITY: |
10065 | ret = -EINVAL; | |
10066 | if (arg || nr_args) | |
10067 | break; | |
10068 | ret = io_register_personality(ctx); | |
10069 | break; | |
10070 | case IORING_UNREGISTER_PERSONALITY: | |
10071 | ret = -EINVAL; | |
10072 | if (arg) | |
10073 | break; | |
10074 | ret = io_unregister_personality(ctx, nr_args); | |
10075 | break; | |
7e84e1c7 SG |
10076 | case IORING_REGISTER_ENABLE_RINGS: |
10077 | ret = -EINVAL; | |
10078 | if (arg || nr_args) | |
10079 | break; | |
10080 | ret = io_register_enable_rings(ctx); | |
10081 | break; | |
21b55dbc SG |
10082 | case IORING_REGISTER_RESTRICTIONS: |
10083 | ret = io_register_restrictions(ctx, arg, nr_args); | |
10084 | break; | |
edafccee JA |
10085 | default: |
10086 | ret = -EINVAL; | |
10087 | break; | |
10088 | } | |
10089 | ||
21b55dbc | 10090 | out: |
071698e1 | 10091 | if (io_register_op_must_quiesce(opcode)) { |
05f3fb3c | 10092 | /* bring the ctx back to life */ |
05f3fb3c | 10093 | percpu_ref_reinit(&ctx->refs); |
21b55dbc | 10094 | out_quiesce: |
0f158b4c | 10095 | reinit_completion(&ctx->ref_comp); |
05f3fb3c | 10096 | } |
edafccee JA |
10097 | return ret; |
10098 | } | |
10099 | ||
10100 | SYSCALL_DEFINE4(io_uring_register, unsigned int, fd, unsigned int, opcode, | |
10101 | void __user *, arg, unsigned int, nr_args) | |
10102 | { | |
10103 | struct io_ring_ctx *ctx; | |
10104 | long ret = -EBADF; | |
10105 | struct fd f; | |
10106 | ||
10107 | f = fdget(fd); | |
10108 | if (!f.file) | |
10109 | return -EBADF; | |
10110 | ||
10111 | ret = -EOPNOTSUPP; | |
10112 | if (f.file->f_op != &io_uring_fops) | |
10113 | goto out_fput; | |
10114 | ||
10115 | ctx = f.file->private_data; | |
10116 | ||
10117 | mutex_lock(&ctx->uring_lock); | |
10118 | ret = __io_uring_register(ctx, opcode, arg, nr_args); | |
10119 | mutex_unlock(&ctx->uring_lock); | |
c826bd7a DD |
10120 | trace_io_uring_register(ctx, opcode, ctx->nr_user_files, ctx->nr_user_bufs, |
10121 | ctx->cq_ev_fd != NULL, ret); | |
edafccee JA |
10122 | out_fput: |
10123 | fdput(f); | |
10124 | return ret; | |
10125 | } | |
10126 | ||
2b188cc1 JA |
10127 | static int __init io_uring_init(void) |
10128 | { | |
d7f62e82 SM |
10129 | #define __BUILD_BUG_VERIFY_ELEMENT(stype, eoffset, etype, ename) do { \ |
10130 | BUILD_BUG_ON(offsetof(stype, ename) != eoffset); \ | |
10131 | BUILD_BUG_ON(sizeof(etype) != sizeof_field(stype, ename)); \ | |
10132 | } while (0) | |
10133 | ||
10134 | #define BUILD_BUG_SQE_ELEM(eoffset, etype, ename) \ | |
10135 | __BUILD_BUG_VERIFY_ELEMENT(struct io_uring_sqe, eoffset, etype, ename) | |
10136 | BUILD_BUG_ON(sizeof(struct io_uring_sqe) != 64); | |
10137 | BUILD_BUG_SQE_ELEM(0, __u8, opcode); | |
10138 | BUILD_BUG_SQE_ELEM(1, __u8, flags); | |
10139 | BUILD_BUG_SQE_ELEM(2, __u16, ioprio); | |
10140 | BUILD_BUG_SQE_ELEM(4, __s32, fd); | |
10141 | BUILD_BUG_SQE_ELEM(8, __u64, off); | |
10142 | BUILD_BUG_SQE_ELEM(8, __u64, addr2); | |
10143 | BUILD_BUG_SQE_ELEM(16, __u64, addr); | |
7d67af2c | 10144 | BUILD_BUG_SQE_ELEM(16, __u64, splice_off_in); |
d7f62e82 SM |
10145 | BUILD_BUG_SQE_ELEM(24, __u32, len); |
10146 | BUILD_BUG_SQE_ELEM(28, __kernel_rwf_t, rw_flags); | |
10147 | BUILD_BUG_SQE_ELEM(28, /* compat */ int, rw_flags); | |
10148 | BUILD_BUG_SQE_ELEM(28, /* compat */ __u32, rw_flags); | |
10149 | BUILD_BUG_SQE_ELEM(28, __u32, fsync_flags); | |
5769a351 JX |
10150 | BUILD_BUG_SQE_ELEM(28, /* compat */ __u16, poll_events); |
10151 | BUILD_BUG_SQE_ELEM(28, __u32, poll32_events); | |
d7f62e82 SM |
10152 | BUILD_BUG_SQE_ELEM(28, __u32, sync_range_flags); |
10153 | BUILD_BUG_SQE_ELEM(28, __u32, msg_flags); | |
10154 | BUILD_BUG_SQE_ELEM(28, __u32, timeout_flags); | |
10155 | BUILD_BUG_SQE_ELEM(28, __u32, accept_flags); | |
10156 | BUILD_BUG_SQE_ELEM(28, __u32, cancel_flags); | |
10157 | BUILD_BUG_SQE_ELEM(28, __u32, open_flags); | |
10158 | BUILD_BUG_SQE_ELEM(28, __u32, statx_flags); | |
10159 | BUILD_BUG_SQE_ELEM(28, __u32, fadvise_advice); | |
7d67af2c | 10160 | BUILD_BUG_SQE_ELEM(28, __u32, splice_flags); |
d7f62e82 SM |
10161 | BUILD_BUG_SQE_ELEM(32, __u64, user_data); |
10162 | BUILD_BUG_SQE_ELEM(40, __u16, buf_index); | |
10163 | BUILD_BUG_SQE_ELEM(42, __u16, personality); | |
7d67af2c | 10164 | BUILD_BUG_SQE_ELEM(44, __s32, splice_fd_in); |
d7f62e82 | 10165 | |
d3656344 | 10166 | BUILD_BUG_ON(ARRAY_SIZE(io_op_defs) != IORING_OP_LAST); |
84557871 | 10167 | BUILD_BUG_ON(__REQ_F_LAST_BIT >= 8 * sizeof(int)); |
2b188cc1 JA |
10168 | req_cachep = KMEM_CACHE(io_kiocb, SLAB_HWCACHE_ALIGN | SLAB_PANIC); |
10169 | return 0; | |
10170 | }; | |
10171 | __initcall(io_uring_init); |