]>
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; |
05589553 XW |
208 | }; |
209 | ||
05f3fb3c JA |
210 | struct fixed_file_data { |
211 | struct fixed_file_table *table; | |
212 | struct io_ring_ctx *ctx; | |
213 | ||
b2e96852 | 214 | struct fixed_file_ref_node *node; |
05f3fb3c | 215 | struct percpu_ref refs; |
05f3fb3c | 216 | struct completion done; |
05589553 XW |
217 | struct list_head ref_list; |
218 | spinlock_t lock; | |
05f3fb3c JA |
219 | }; |
220 | ||
5a2e745d JA |
221 | struct io_buffer { |
222 | struct list_head list; | |
223 | __u64 addr; | |
224 | __s32 len; | |
225 | __u16 bid; | |
226 | }; | |
227 | ||
21b55dbc SG |
228 | struct io_restriction { |
229 | DECLARE_BITMAP(register_op, IORING_REGISTER_LAST); | |
230 | DECLARE_BITMAP(sqe_op, IORING_OP_LAST); | |
231 | u8 sqe_flags_allowed; | |
232 | u8 sqe_flags_required; | |
7e84e1c7 | 233 | bool registered; |
21b55dbc SG |
234 | }; |
235 | ||
534ca6d6 JA |
236 | struct io_sq_data { |
237 | refcount_t refs; | |
69fb2131 JA |
238 | struct mutex lock; |
239 | ||
240 | /* ctx's that are using this sqd */ | |
241 | struct list_head ctx_list; | |
242 | struct list_head ctx_new_list; | |
243 | struct mutex ctx_lock; | |
244 | ||
534ca6d6 JA |
245 | struct task_struct *thread; |
246 | struct wait_queue_head wait; | |
247 | }; | |
248 | ||
2b188cc1 JA |
249 | struct io_ring_ctx { |
250 | struct { | |
251 | struct percpu_ref refs; | |
252 | } ____cacheline_aligned_in_smp; | |
253 | ||
254 | struct { | |
255 | unsigned int flags; | |
e1d85334 | 256 | unsigned int compat: 1; |
aad5d8da | 257 | unsigned int limit_mem: 1; |
e1d85334 RD |
258 | unsigned int cq_overflow_flushed: 1; |
259 | unsigned int drain_next: 1; | |
260 | unsigned int eventfd_async: 1; | |
21b55dbc | 261 | unsigned int restricted: 1; |
2b188cc1 | 262 | |
75b28aff HV |
263 | /* |
264 | * Ring buffer of indices into array of io_uring_sqe, which is | |
265 | * mmapped by the application using the IORING_OFF_SQES offset. | |
266 | * | |
267 | * This indirection could e.g. be used to assign fixed | |
268 | * io_uring_sqe entries to operations and only submit them to | |
269 | * the queue when needed. | |
270 | * | |
271 | * The kernel modifies neither the indices array nor the entries | |
272 | * array. | |
273 | */ | |
274 | u32 *sq_array; | |
2b188cc1 JA |
275 | unsigned cached_sq_head; |
276 | unsigned sq_entries; | |
277 | unsigned sq_mask; | |
6c271ce2 | 278 | unsigned sq_thread_idle; |
498ccd9e | 279 | unsigned cached_sq_dropped; |
2c3bac6d | 280 | unsigned cached_cq_overflow; |
ad3eb2c8 | 281 | unsigned long sq_check_overflow; |
de0617e4 JA |
282 | |
283 | struct list_head defer_list; | |
5262f567 | 284 | struct list_head timeout_list; |
1d7bb1d5 | 285 | struct list_head cq_overflow_list; |
fcb323cc JA |
286 | |
287 | wait_queue_head_t inflight_wait; | |
ad3eb2c8 | 288 | struct io_uring_sqe *sq_sqes; |
2b188cc1 JA |
289 | } ____cacheline_aligned_in_smp; |
290 | ||
206aefde JA |
291 | struct io_rings *rings; |
292 | ||
2b188cc1 | 293 | /* IO offload */ |
561fb04a | 294 | struct io_wq *io_wq; |
2aede0e4 JA |
295 | |
296 | /* | |
297 | * For SQPOLL usage - we hold a reference to the parent task, so we | |
298 | * have access to the ->files | |
299 | */ | |
300 | struct task_struct *sqo_task; | |
301 | ||
302 | /* Only used for accounting purposes */ | |
303 | struct mm_struct *mm_account; | |
304 | ||
91d8f519 DZ |
305 | #ifdef CONFIG_BLK_CGROUP |
306 | struct cgroup_subsys_state *sqo_blkcg_css; | |
307 | #endif | |
308 | ||
534ca6d6 JA |
309 | struct io_sq_data *sq_data; /* if using sq thread polling */ |
310 | ||
90554200 | 311 | struct wait_queue_head sqo_sq_wait; |
6a779382 | 312 | struct wait_queue_entry sqo_wait_entry; |
69fb2131 | 313 | struct list_head sqd_list; |
75b28aff | 314 | |
6b06314c JA |
315 | /* |
316 | * If used, fixed file set. Writers must ensure that ->refs is dead, | |
317 | * readers must ensure that ->refs is alive as long as the file* is | |
318 | * used. Only updated through io_uring_register(2). | |
319 | */ | |
05f3fb3c | 320 | struct fixed_file_data *file_data; |
6b06314c JA |
321 | unsigned nr_user_files; |
322 | ||
edafccee JA |
323 | /* if used, fixed mapped user buffers */ |
324 | unsigned nr_user_bufs; | |
325 | struct io_mapped_ubuf *user_bufs; | |
326 | ||
2b188cc1 JA |
327 | struct user_struct *user; |
328 | ||
0b8c0ec7 | 329 | const struct cred *creds; |
181e448d | 330 | |
4ea33a97 JA |
331 | #ifdef CONFIG_AUDIT |
332 | kuid_t loginuid; | |
333 | unsigned int sessionid; | |
334 | #endif | |
335 | ||
0f158b4c JA |
336 | struct completion ref_comp; |
337 | struct completion sq_thread_comp; | |
206aefde | 338 | |
0ddf92e8 JA |
339 | /* if all else fails... */ |
340 | struct io_kiocb *fallback_req; | |
341 | ||
206aefde JA |
342 | #if defined(CONFIG_UNIX) |
343 | struct socket *ring_sock; | |
344 | #endif | |
345 | ||
5a2e745d JA |
346 | struct idr io_buffer_idr; |
347 | ||
071698e1 JA |
348 | struct idr personality_idr; |
349 | ||
206aefde JA |
350 | struct { |
351 | unsigned cached_cq_tail; | |
352 | unsigned cq_entries; | |
353 | unsigned cq_mask; | |
354 | atomic_t cq_timeouts; | |
ad3eb2c8 | 355 | unsigned long cq_check_overflow; |
206aefde JA |
356 | struct wait_queue_head cq_wait; |
357 | struct fasync_struct *cq_fasync; | |
358 | struct eventfd_ctx *cq_ev_fd; | |
359 | } ____cacheline_aligned_in_smp; | |
2b188cc1 JA |
360 | |
361 | struct { | |
362 | struct mutex uring_lock; | |
363 | wait_queue_head_t wait; | |
364 | } ____cacheline_aligned_in_smp; | |
365 | ||
366 | struct { | |
367 | spinlock_t completion_lock; | |
e94f141b | 368 | |
def596e9 | 369 | /* |
540e32a0 | 370 | * ->iopoll_list is protected by the ctx->uring_lock for |
def596e9 JA |
371 | * io_uring instances that don't use IORING_SETUP_SQPOLL. |
372 | * For SQPOLL, only the single threaded io_sq_thread() will | |
373 | * manipulate the list, hence no extra locking is needed there. | |
374 | */ | |
540e32a0 | 375 | struct list_head iopoll_list; |
78076bb6 JA |
376 | struct hlist_head *cancel_hash; |
377 | unsigned cancel_hash_bits; | |
e94f141b | 378 | bool poll_multi_file; |
31b51510 | 379 | |
fcb323cc JA |
380 | spinlock_t inflight_lock; |
381 | struct list_head inflight_list; | |
2b188cc1 | 382 | } ____cacheline_aligned_in_smp; |
85faa7b8 | 383 | |
4a38aed2 JA |
384 | struct delayed_work file_put_work; |
385 | struct llist_head file_put_llist; | |
386 | ||
85faa7b8 | 387 | struct work_struct exit_work; |
21b55dbc | 388 | struct io_restriction restrictions; |
2b188cc1 JA |
389 | }; |
390 | ||
09bb8394 JA |
391 | /* |
392 | * First field must be the file pointer in all the | |
393 | * iocb unions! See also 'struct kiocb' in <linux/fs.h> | |
394 | */ | |
221c5eb2 JA |
395 | struct io_poll_iocb { |
396 | struct file *file; | |
0969e783 JA |
397 | union { |
398 | struct wait_queue_head *head; | |
399 | u64 addr; | |
400 | }; | |
221c5eb2 | 401 | __poll_t events; |
8c838788 | 402 | bool done; |
221c5eb2 | 403 | bool canceled; |
392edb45 | 404 | struct wait_queue_entry wait; |
221c5eb2 JA |
405 | }; |
406 | ||
b5dba59e JA |
407 | struct io_close { |
408 | struct file *file; | |
409 | struct file *put_file; | |
410 | int fd; | |
411 | }; | |
412 | ||
ad8a48ac JA |
413 | struct io_timeout_data { |
414 | struct io_kiocb *req; | |
415 | struct hrtimer timer; | |
416 | struct timespec64 ts; | |
417 | enum hrtimer_mode mode; | |
418 | }; | |
419 | ||
8ed8d3c3 JA |
420 | struct io_accept { |
421 | struct file *file; | |
422 | struct sockaddr __user *addr; | |
423 | int __user *addr_len; | |
424 | int flags; | |
09952e3e | 425 | unsigned long nofile; |
8ed8d3c3 JA |
426 | }; |
427 | ||
428 | struct io_sync { | |
429 | struct file *file; | |
430 | loff_t len; | |
431 | loff_t off; | |
432 | int flags; | |
d63d1b5e | 433 | int mode; |
8ed8d3c3 JA |
434 | }; |
435 | ||
fbf23849 JA |
436 | struct io_cancel { |
437 | struct file *file; | |
438 | u64 addr; | |
439 | }; | |
440 | ||
b29472ee JA |
441 | struct io_timeout { |
442 | struct file *file; | |
bfe68a22 PB |
443 | u32 off; |
444 | u32 target_seq; | |
135fcde8 | 445 | struct list_head list; |
b29472ee JA |
446 | }; |
447 | ||
0bdf7a2d PB |
448 | struct io_timeout_rem { |
449 | struct file *file; | |
450 | u64 addr; | |
451 | }; | |
452 | ||
9adbd45d JA |
453 | struct io_rw { |
454 | /* NOTE: kiocb has the file as the first member, so don't do it here */ | |
455 | struct kiocb kiocb; | |
456 | u64 addr; | |
457 | u64 len; | |
458 | }; | |
459 | ||
3fbb51c1 JA |
460 | struct io_connect { |
461 | struct file *file; | |
462 | struct sockaddr __user *addr; | |
463 | int addr_len; | |
464 | }; | |
465 | ||
e47293fd JA |
466 | struct io_sr_msg { |
467 | struct file *file; | |
fddaface | 468 | union { |
270a5940 | 469 | struct user_msghdr __user *umsg; |
fddaface JA |
470 | void __user *buf; |
471 | }; | |
e47293fd | 472 | int msg_flags; |
bcda7baa | 473 | int bgid; |
fddaface | 474 | size_t len; |
bcda7baa | 475 | struct io_buffer *kbuf; |
e47293fd JA |
476 | }; |
477 | ||
15b71abe JA |
478 | struct io_open { |
479 | struct file *file; | |
480 | int dfd; | |
15b71abe | 481 | struct filename *filename; |
c12cedf2 | 482 | struct open_how how; |
4022e7af | 483 | unsigned long nofile; |
15b71abe JA |
484 | }; |
485 | ||
05f3fb3c JA |
486 | struct io_files_update { |
487 | struct file *file; | |
488 | u64 arg; | |
489 | u32 nr_args; | |
490 | u32 offset; | |
491 | }; | |
492 | ||
4840e418 JA |
493 | struct io_fadvise { |
494 | struct file *file; | |
495 | u64 offset; | |
496 | u32 len; | |
497 | u32 advice; | |
498 | }; | |
499 | ||
c1ca757b JA |
500 | struct io_madvise { |
501 | struct file *file; | |
502 | u64 addr; | |
503 | u32 len; | |
504 | u32 advice; | |
505 | }; | |
506 | ||
3e4827b0 JA |
507 | struct io_epoll { |
508 | struct file *file; | |
509 | int epfd; | |
510 | int op; | |
511 | int fd; | |
512 | struct epoll_event event; | |
e47293fd JA |
513 | }; |
514 | ||
7d67af2c PB |
515 | struct io_splice { |
516 | struct file *file_out; | |
517 | struct file *file_in; | |
518 | loff_t off_out; | |
519 | loff_t off_in; | |
520 | u64 len; | |
521 | unsigned int flags; | |
522 | }; | |
523 | ||
ddf0322d JA |
524 | struct io_provide_buf { |
525 | struct file *file; | |
526 | __u64 addr; | |
527 | __s32 len; | |
528 | __u32 bgid; | |
529 | __u16 nbufs; | |
530 | __u16 bid; | |
531 | }; | |
532 | ||
1d9e1288 BM |
533 | struct io_statx { |
534 | struct file *file; | |
535 | int dfd; | |
536 | unsigned int mask; | |
537 | unsigned int flags; | |
e62753e4 | 538 | const char __user *filename; |
1d9e1288 BM |
539 | struct statx __user *buffer; |
540 | }; | |
541 | ||
3ca405eb PB |
542 | struct io_completion { |
543 | struct file *file; | |
544 | struct list_head list; | |
0f7e466b | 545 | int cflags; |
3ca405eb PB |
546 | }; |
547 | ||
f499a021 JA |
548 | struct io_async_connect { |
549 | struct sockaddr_storage address; | |
550 | }; | |
551 | ||
03b1230c JA |
552 | struct io_async_msghdr { |
553 | struct iovec fast_iov[UIO_FASTIOV]; | |
554 | struct iovec *iov; | |
555 | struct sockaddr __user *uaddr; | |
556 | struct msghdr msg; | |
b537916c | 557 | struct sockaddr_storage addr; |
03b1230c JA |
558 | }; |
559 | ||
f67676d1 JA |
560 | struct io_async_rw { |
561 | struct iovec fast_iov[UIO_FASTIOV]; | |
ff6165b2 JA |
562 | const struct iovec *free_iovec; |
563 | struct iov_iter iter; | |
227c0c96 | 564 | size_t bytes_done; |
bcf5a063 | 565 | struct wait_page_queue wpq; |
f67676d1 JA |
566 | }; |
567 | ||
6b47ee6e PB |
568 | enum { |
569 | REQ_F_FIXED_FILE_BIT = IOSQE_FIXED_FILE_BIT, | |
570 | REQ_F_IO_DRAIN_BIT = IOSQE_IO_DRAIN_BIT, | |
571 | REQ_F_LINK_BIT = IOSQE_IO_LINK_BIT, | |
572 | REQ_F_HARDLINK_BIT = IOSQE_IO_HARDLINK_BIT, | |
573 | REQ_F_FORCE_ASYNC_BIT = IOSQE_ASYNC_BIT, | |
bcda7baa | 574 | REQ_F_BUFFER_SELECT_BIT = IOSQE_BUFFER_SELECT_BIT, |
6b47ee6e | 575 | |
dea3b49c | 576 | REQ_F_LINK_HEAD_BIT, |
6b47ee6e PB |
577 | REQ_F_FAIL_LINK_BIT, |
578 | REQ_F_INFLIGHT_BIT, | |
579 | REQ_F_CUR_POS_BIT, | |
580 | REQ_F_NOWAIT_BIT, | |
6b47ee6e | 581 | REQ_F_LINK_TIMEOUT_BIT, |
6b47ee6e | 582 | REQ_F_ISREG_BIT, |
99bc4c38 | 583 | REQ_F_NEED_CLEANUP_BIT, |
d7718a9d | 584 | REQ_F_POLLED_BIT, |
bcda7baa | 585 | REQ_F_BUFFER_SELECTED_BIT, |
5b0bbee4 | 586 | REQ_F_NO_FILE_TABLE_BIT, |
7cdaf587 | 587 | REQ_F_WORK_INITIALIZED_BIT, |
900fad45 | 588 | REQ_F_LTIMEOUT_ACTIVE_BIT, |
84557871 JA |
589 | |
590 | /* not a real bit, just to check we're not overflowing the space */ | |
591 | __REQ_F_LAST_BIT, | |
6b47ee6e PB |
592 | }; |
593 | ||
594 | enum { | |
595 | /* ctx owns file */ | |
596 | REQ_F_FIXED_FILE = BIT(REQ_F_FIXED_FILE_BIT), | |
597 | /* drain existing IO first */ | |
598 | REQ_F_IO_DRAIN = BIT(REQ_F_IO_DRAIN_BIT), | |
599 | /* linked sqes */ | |
600 | REQ_F_LINK = BIT(REQ_F_LINK_BIT), | |
601 | /* doesn't sever on completion < 0 */ | |
602 | REQ_F_HARDLINK = BIT(REQ_F_HARDLINK_BIT), | |
603 | /* IOSQE_ASYNC */ | |
604 | REQ_F_FORCE_ASYNC = BIT(REQ_F_FORCE_ASYNC_BIT), | |
bcda7baa JA |
605 | /* IOSQE_BUFFER_SELECT */ |
606 | REQ_F_BUFFER_SELECT = BIT(REQ_F_BUFFER_SELECT_BIT), | |
6b47ee6e | 607 | |
dea3b49c PB |
608 | /* head of a link */ |
609 | REQ_F_LINK_HEAD = BIT(REQ_F_LINK_HEAD_BIT), | |
6b47ee6e PB |
610 | /* fail rest of links */ |
611 | REQ_F_FAIL_LINK = BIT(REQ_F_FAIL_LINK_BIT), | |
612 | /* on inflight list */ | |
613 | REQ_F_INFLIGHT = BIT(REQ_F_INFLIGHT_BIT), | |
614 | /* read/write uses file position */ | |
615 | REQ_F_CUR_POS = BIT(REQ_F_CUR_POS_BIT), | |
616 | /* must not punt to workers */ | |
617 | REQ_F_NOWAIT = BIT(REQ_F_NOWAIT_BIT), | |
900fad45 | 618 | /* has or had linked timeout */ |
6b47ee6e | 619 | REQ_F_LINK_TIMEOUT = BIT(REQ_F_LINK_TIMEOUT_BIT), |
6b47ee6e PB |
620 | /* regular file */ |
621 | REQ_F_ISREG = BIT(REQ_F_ISREG_BIT), | |
99bc4c38 PB |
622 | /* needs cleanup */ |
623 | REQ_F_NEED_CLEANUP = BIT(REQ_F_NEED_CLEANUP_BIT), | |
d7718a9d JA |
624 | /* already went through poll handler */ |
625 | REQ_F_POLLED = BIT(REQ_F_POLLED_BIT), | |
bcda7baa JA |
626 | /* buffer already selected */ |
627 | REQ_F_BUFFER_SELECTED = BIT(REQ_F_BUFFER_SELECTED_BIT), | |
5b0bbee4 JA |
628 | /* doesn't need file table for this request */ |
629 | REQ_F_NO_FILE_TABLE = BIT(REQ_F_NO_FILE_TABLE_BIT), | |
7cdaf587 XW |
630 | /* io_wq_work is initialized */ |
631 | REQ_F_WORK_INITIALIZED = BIT(REQ_F_WORK_INITIALIZED_BIT), | |
900fad45 PB |
632 | /* linked timeout is active, i.e. prepared by link's head */ |
633 | REQ_F_LTIMEOUT_ACTIVE = BIT(REQ_F_LTIMEOUT_ACTIVE_BIT), | |
d7718a9d JA |
634 | }; |
635 | ||
636 | struct async_poll { | |
637 | struct io_poll_iocb poll; | |
807abcb0 | 638 | struct io_poll_iocb *double_poll; |
6b47ee6e PB |
639 | }; |
640 | ||
09bb8394 JA |
641 | /* |
642 | * NOTE! Each of the iocb union members has the file pointer | |
643 | * as the first entry in their struct definition. So you can | |
644 | * access the file pointer through any of the sub-structs, | |
645 | * or directly as just 'ki_filp' in this struct. | |
646 | */ | |
2b188cc1 | 647 | struct io_kiocb { |
221c5eb2 | 648 | union { |
09bb8394 | 649 | struct file *file; |
9adbd45d | 650 | struct io_rw rw; |
221c5eb2 | 651 | struct io_poll_iocb poll; |
8ed8d3c3 JA |
652 | struct io_accept accept; |
653 | struct io_sync sync; | |
fbf23849 | 654 | struct io_cancel cancel; |
b29472ee | 655 | struct io_timeout timeout; |
0bdf7a2d | 656 | struct io_timeout_rem timeout_rem; |
3fbb51c1 | 657 | struct io_connect connect; |
e47293fd | 658 | struct io_sr_msg sr_msg; |
15b71abe | 659 | struct io_open open; |
b5dba59e | 660 | struct io_close close; |
05f3fb3c | 661 | struct io_files_update files_update; |
4840e418 | 662 | struct io_fadvise fadvise; |
c1ca757b | 663 | struct io_madvise madvise; |
3e4827b0 | 664 | struct io_epoll epoll; |
7d67af2c | 665 | struct io_splice splice; |
ddf0322d | 666 | struct io_provide_buf pbuf; |
1d9e1288 | 667 | struct io_statx statx; |
3ca405eb PB |
668 | /* use only after cleaning per-op data, see io_clean_op() */ |
669 | struct io_completion compl; | |
221c5eb2 | 670 | }; |
2b188cc1 | 671 | |
e8c2bc1f JA |
672 | /* opcode allocated if it needs to store data for async defer */ |
673 | void *async_data; | |
d625c6ee | 674 | u8 opcode; |
65a6543d XW |
675 | /* polled IO has completed */ |
676 | u8 iopoll_completed; | |
2b188cc1 | 677 | |
4f4eeba8 | 678 | u16 buf_index; |
9cf7c104 | 679 | u32 result; |
4f4eeba8 | 680 | |
010e8e6b PB |
681 | struct io_ring_ctx *ctx; |
682 | unsigned int flags; | |
683 | refcount_t refs; | |
684 | struct task_struct *task; | |
685 | u64 user_data; | |
d7718a9d | 686 | |
010e8e6b | 687 | struct list_head link_list; |
fcb323cc | 688 | |
d21ffe7e PB |
689 | /* |
690 | * 1. used with ctx->iopoll_list with reads/writes | |
691 | * 2. to track reqs with ->files (see io_op_def::file_table) | |
692 | */ | |
010e8e6b PB |
693 | struct list_head inflight_entry; |
694 | ||
695 | struct percpu_ref *fixed_file_refs; | |
696 | struct callback_head task_work; | |
697 | /* for polled requests, i.e. IORING_OP_POLL_ADD and async armed poll */ | |
698 | struct hlist_node hash_node; | |
699 | struct async_poll *apoll; | |
700 | struct io_wq_work work; | |
2b188cc1 | 701 | }; |
05589553 | 702 | |
27dc8338 PB |
703 | struct io_defer_entry { |
704 | struct list_head list; | |
705 | struct io_kiocb *req; | |
9cf7c104 | 706 | u32 seq; |
2b188cc1 JA |
707 | }; |
708 | ||
def596e9 | 709 | #define IO_IOPOLL_BATCH 8 |
2b188cc1 | 710 | |
013538bd JA |
711 | struct io_comp_state { |
712 | unsigned int nr; | |
713 | struct list_head list; | |
714 | struct io_ring_ctx *ctx; | |
715 | }; | |
716 | ||
9a56a232 JA |
717 | struct io_submit_state { |
718 | struct blk_plug plug; | |
719 | ||
2579f913 JA |
720 | /* |
721 | * io_kiocb alloc cache | |
722 | */ | |
723 | void *reqs[IO_IOPOLL_BATCH]; | |
6c8a3134 | 724 | unsigned int free_reqs; |
2579f913 | 725 | |
013538bd JA |
726 | /* |
727 | * Batch completion logic | |
728 | */ | |
729 | struct io_comp_state comp; | |
730 | ||
9a56a232 JA |
731 | /* |
732 | * File reference cache | |
733 | */ | |
734 | struct file *file; | |
735 | unsigned int fd; | |
736 | unsigned int has_refs; | |
9a56a232 JA |
737 | unsigned int ios_left; |
738 | }; | |
739 | ||
d3656344 | 740 | struct io_op_def { |
d3656344 JA |
741 | /* needs req->file assigned */ |
742 | unsigned needs_file : 1; | |
fd2206e4 JA |
743 | /* don't fail if file grab fails */ |
744 | unsigned needs_file_no_error : 1; | |
d3656344 JA |
745 | /* hash wq insertion if file is a regular file */ |
746 | unsigned hash_reg_file : 1; | |
747 | /* unbound wq insertion if file is a non-regular file */ | |
748 | unsigned unbound_nonreg_file : 1; | |
66f4af93 JA |
749 | /* opcode is not supported by this kernel */ |
750 | unsigned not_supported : 1; | |
8a72758c JA |
751 | /* set if opcode supports polled "wait" */ |
752 | unsigned pollin : 1; | |
753 | unsigned pollout : 1; | |
bcda7baa JA |
754 | /* op supports buffer selection */ |
755 | unsigned buffer_select : 1; | |
e8c2bc1f JA |
756 | /* must always have async data allocated */ |
757 | unsigned needs_async_data : 1; | |
758 | /* size of async data needed, if any */ | |
759 | unsigned short async_size; | |
0f203765 | 760 | unsigned work_flags; |
d3656344 JA |
761 | }; |
762 | ||
0918682b | 763 | static const struct io_op_def io_op_defs[] = { |
0463b6c5 PB |
764 | [IORING_OP_NOP] = {}, |
765 | [IORING_OP_READV] = { | |
d3656344 JA |
766 | .needs_file = 1, |
767 | .unbound_nonreg_file = 1, | |
8a72758c | 768 | .pollin = 1, |
4d954c25 | 769 | .buffer_select = 1, |
e8c2bc1f JA |
770 | .needs_async_data = 1, |
771 | .async_size = sizeof(struct io_async_rw), | |
0f203765 | 772 | .work_flags = IO_WQ_WORK_MM | IO_WQ_WORK_BLKCG, |
d3656344 | 773 | }, |
0463b6c5 | 774 | [IORING_OP_WRITEV] = { |
d3656344 JA |
775 | .needs_file = 1, |
776 | .hash_reg_file = 1, | |
777 | .unbound_nonreg_file = 1, | |
8a72758c | 778 | .pollout = 1, |
e8c2bc1f JA |
779 | .needs_async_data = 1, |
780 | .async_size = sizeof(struct io_async_rw), | |
69228338 JA |
781 | .work_flags = IO_WQ_WORK_MM | IO_WQ_WORK_BLKCG | |
782 | IO_WQ_WORK_FSIZE, | |
d3656344 | 783 | }, |
0463b6c5 | 784 | [IORING_OP_FSYNC] = { |
d3656344 | 785 | .needs_file = 1, |
0f203765 | 786 | .work_flags = IO_WQ_WORK_BLKCG, |
d3656344 | 787 | }, |
0463b6c5 | 788 | [IORING_OP_READ_FIXED] = { |
d3656344 JA |
789 | .needs_file = 1, |
790 | .unbound_nonreg_file = 1, | |
8a72758c | 791 | .pollin = 1, |
e8c2bc1f | 792 | .async_size = sizeof(struct io_async_rw), |
4017eb91 | 793 | .work_flags = IO_WQ_WORK_BLKCG | IO_WQ_WORK_MM, |
d3656344 | 794 | }, |
0463b6c5 | 795 | [IORING_OP_WRITE_FIXED] = { |
d3656344 JA |
796 | .needs_file = 1, |
797 | .hash_reg_file = 1, | |
798 | .unbound_nonreg_file = 1, | |
8a72758c | 799 | .pollout = 1, |
e8c2bc1f | 800 | .async_size = sizeof(struct io_async_rw), |
4017eb91 JA |
801 | .work_flags = IO_WQ_WORK_BLKCG | IO_WQ_WORK_FSIZE | |
802 | IO_WQ_WORK_MM, | |
d3656344 | 803 | }, |
0463b6c5 | 804 | [IORING_OP_POLL_ADD] = { |
d3656344 JA |
805 | .needs_file = 1, |
806 | .unbound_nonreg_file = 1, | |
807 | }, | |
0463b6c5 PB |
808 | [IORING_OP_POLL_REMOVE] = {}, |
809 | [IORING_OP_SYNC_FILE_RANGE] = { | |
d3656344 | 810 | .needs_file = 1, |
0f203765 | 811 | .work_flags = IO_WQ_WORK_BLKCG, |
d3656344 | 812 | }, |
0463b6c5 | 813 | [IORING_OP_SENDMSG] = { |
d3656344 JA |
814 | .needs_file = 1, |
815 | .unbound_nonreg_file = 1, | |
8a72758c | 816 | .pollout = 1, |
e8c2bc1f JA |
817 | .needs_async_data = 1, |
818 | .async_size = sizeof(struct io_async_msghdr), | |
0f203765 JA |
819 | .work_flags = IO_WQ_WORK_MM | IO_WQ_WORK_BLKCG | |
820 | IO_WQ_WORK_FS, | |
d3656344 | 821 | }, |
0463b6c5 | 822 | [IORING_OP_RECVMSG] = { |
d3656344 JA |
823 | .needs_file = 1, |
824 | .unbound_nonreg_file = 1, | |
8a72758c | 825 | .pollin = 1, |
52de1fe1 | 826 | .buffer_select = 1, |
e8c2bc1f JA |
827 | .needs_async_data = 1, |
828 | .async_size = sizeof(struct io_async_msghdr), | |
0f203765 JA |
829 | .work_flags = IO_WQ_WORK_MM | IO_WQ_WORK_BLKCG | |
830 | IO_WQ_WORK_FS, | |
d3656344 | 831 | }, |
0463b6c5 | 832 | [IORING_OP_TIMEOUT] = { |
e8c2bc1f JA |
833 | .needs_async_data = 1, |
834 | .async_size = sizeof(struct io_timeout_data), | |
0f203765 | 835 | .work_flags = IO_WQ_WORK_MM, |
d3656344 | 836 | }, |
0463b6c5 PB |
837 | [IORING_OP_TIMEOUT_REMOVE] = {}, |
838 | [IORING_OP_ACCEPT] = { | |
d3656344 JA |
839 | .needs_file = 1, |
840 | .unbound_nonreg_file = 1, | |
8a72758c | 841 | .pollin = 1, |
0f203765 | 842 | .work_flags = IO_WQ_WORK_MM | IO_WQ_WORK_FILES, |
d3656344 | 843 | }, |
0463b6c5 PB |
844 | [IORING_OP_ASYNC_CANCEL] = {}, |
845 | [IORING_OP_LINK_TIMEOUT] = { | |
e8c2bc1f JA |
846 | .needs_async_data = 1, |
847 | .async_size = sizeof(struct io_timeout_data), | |
0f203765 | 848 | .work_flags = IO_WQ_WORK_MM, |
d3656344 | 849 | }, |
0463b6c5 | 850 | [IORING_OP_CONNECT] = { |
d3656344 JA |
851 | .needs_file = 1, |
852 | .unbound_nonreg_file = 1, | |
8a72758c | 853 | .pollout = 1, |
e8c2bc1f JA |
854 | .needs_async_data = 1, |
855 | .async_size = sizeof(struct io_async_connect), | |
0f203765 | 856 | .work_flags = IO_WQ_WORK_MM, |
d3656344 | 857 | }, |
0463b6c5 | 858 | [IORING_OP_FALLOCATE] = { |
d3656344 | 859 | .needs_file = 1, |
69228338 | 860 | .work_flags = IO_WQ_WORK_BLKCG | IO_WQ_WORK_FSIZE, |
d3656344 | 861 | }, |
0463b6c5 | 862 | [IORING_OP_OPENAT] = { |
0f203765 JA |
863 | .work_flags = IO_WQ_WORK_FILES | IO_WQ_WORK_BLKCG | |
864 | IO_WQ_WORK_FS, | |
d3656344 | 865 | }, |
0463b6c5 | 866 | [IORING_OP_CLOSE] = { |
fd2206e4 JA |
867 | .needs_file = 1, |
868 | .needs_file_no_error = 1, | |
0f203765 | 869 | .work_flags = IO_WQ_WORK_FILES | IO_WQ_WORK_BLKCG, |
d3656344 | 870 | }, |
0463b6c5 | 871 | [IORING_OP_FILES_UPDATE] = { |
0f203765 | 872 | .work_flags = IO_WQ_WORK_FILES | IO_WQ_WORK_MM, |
d3656344 | 873 | }, |
0463b6c5 | 874 | [IORING_OP_STATX] = { |
0f203765 JA |
875 | .work_flags = IO_WQ_WORK_FILES | IO_WQ_WORK_MM | |
876 | IO_WQ_WORK_FS | IO_WQ_WORK_BLKCG, | |
d3656344 | 877 | }, |
0463b6c5 | 878 | [IORING_OP_READ] = { |
3a6820f2 JA |
879 | .needs_file = 1, |
880 | .unbound_nonreg_file = 1, | |
8a72758c | 881 | .pollin = 1, |
bcda7baa | 882 | .buffer_select = 1, |
e8c2bc1f | 883 | .async_size = sizeof(struct io_async_rw), |
0f203765 | 884 | .work_flags = IO_WQ_WORK_MM | IO_WQ_WORK_BLKCG, |
3a6820f2 | 885 | }, |
0463b6c5 | 886 | [IORING_OP_WRITE] = { |
3a6820f2 JA |
887 | .needs_file = 1, |
888 | .unbound_nonreg_file = 1, | |
8a72758c | 889 | .pollout = 1, |
e8c2bc1f | 890 | .async_size = sizeof(struct io_async_rw), |
69228338 JA |
891 | .work_flags = IO_WQ_WORK_MM | IO_WQ_WORK_BLKCG | |
892 | IO_WQ_WORK_FSIZE, | |
3a6820f2 | 893 | }, |
0463b6c5 | 894 | [IORING_OP_FADVISE] = { |
4840e418 | 895 | .needs_file = 1, |
0f203765 | 896 | .work_flags = IO_WQ_WORK_BLKCG, |
4840e418 | 897 | }, |
0463b6c5 | 898 | [IORING_OP_MADVISE] = { |
0f203765 | 899 | .work_flags = IO_WQ_WORK_MM | IO_WQ_WORK_BLKCG, |
c1ca757b | 900 | }, |
0463b6c5 | 901 | [IORING_OP_SEND] = { |
fddaface JA |
902 | .needs_file = 1, |
903 | .unbound_nonreg_file = 1, | |
8a72758c | 904 | .pollout = 1, |
0f203765 | 905 | .work_flags = IO_WQ_WORK_MM | IO_WQ_WORK_BLKCG, |
fddaface | 906 | }, |
0463b6c5 | 907 | [IORING_OP_RECV] = { |
fddaface JA |
908 | .needs_file = 1, |
909 | .unbound_nonreg_file = 1, | |
8a72758c | 910 | .pollin = 1, |
bcda7baa | 911 | .buffer_select = 1, |
0f203765 | 912 | .work_flags = IO_WQ_WORK_MM | IO_WQ_WORK_BLKCG, |
fddaface | 913 | }, |
0463b6c5 | 914 | [IORING_OP_OPENAT2] = { |
0f203765 JA |
915 | .work_flags = IO_WQ_WORK_FILES | IO_WQ_WORK_FS | |
916 | IO_WQ_WORK_BLKCG, | |
cebdb986 | 917 | }, |
3e4827b0 JA |
918 | [IORING_OP_EPOLL_CTL] = { |
919 | .unbound_nonreg_file = 1, | |
0f203765 | 920 | .work_flags = IO_WQ_WORK_FILES, |
3e4827b0 | 921 | }, |
7d67af2c PB |
922 | [IORING_OP_SPLICE] = { |
923 | .needs_file = 1, | |
924 | .hash_reg_file = 1, | |
925 | .unbound_nonreg_file = 1, | |
0f203765 | 926 | .work_flags = IO_WQ_WORK_BLKCG, |
ddf0322d JA |
927 | }, |
928 | [IORING_OP_PROVIDE_BUFFERS] = {}, | |
067524e9 | 929 | [IORING_OP_REMOVE_BUFFERS] = {}, |
f2a8d5c7 PB |
930 | [IORING_OP_TEE] = { |
931 | .needs_file = 1, | |
932 | .hash_reg_file = 1, | |
933 | .unbound_nonreg_file = 1, | |
934 | }, | |
d3656344 JA |
935 | }; |
936 | ||
2e0464d4 BM |
937 | enum io_mem_account { |
938 | ACCT_LOCKED, | |
939 | ACCT_PINNED, | |
940 | }; | |
941 | ||
81b68a5c PB |
942 | static void __io_complete_rw(struct io_kiocb *req, long res, long res2, |
943 | struct io_comp_state *cs); | |
78e19bbe | 944 | static void io_cqring_fill_event(struct io_kiocb *req, long res); |
ec9c02ad | 945 | static void io_put_req(struct io_kiocb *req); |
216578e5 | 946 | static void io_put_req_deferred(struct io_kiocb *req, int nr); |
c40f6379 | 947 | static void io_double_put_req(struct io_kiocb *req); |
94ae5e77 | 948 | static struct io_kiocb *io_prep_linked_timeout(struct io_kiocb *req); |
7271ef3a | 949 | static void __io_queue_linked_timeout(struct io_kiocb *req); |
94ae5e77 | 950 | static void io_queue_linked_timeout(struct io_kiocb *req); |
05f3fb3c JA |
951 | static int __io_sqe_files_update(struct io_ring_ctx *ctx, |
952 | struct io_uring_files_update *ip, | |
953 | unsigned nr_args); | |
3ca405eb | 954 | static void __io_clean_op(struct io_kiocb *req); |
8371adf5 PB |
955 | static struct file *io_file_get(struct io_submit_state *state, |
956 | struct io_kiocb *req, int fd, bool fixed); | |
c1379e24 | 957 | static void __io_queue_sqe(struct io_kiocb *req, struct io_comp_state *cs); |
4349f30e | 958 | static void io_file_put_work(struct work_struct *work); |
de0617e4 | 959 | |
b63534c4 JA |
960 | static ssize_t io_import_iovec(int rw, struct io_kiocb *req, |
961 | struct iovec **iovec, struct iov_iter *iter, | |
962 | bool needs_lock); | |
ff6165b2 JA |
963 | static int io_setup_async_rw(struct io_kiocb *req, const struct iovec *iovec, |
964 | const struct iovec *fast_iov, | |
227c0c96 | 965 | struct iov_iter *iter, bool force); |
de0617e4 | 966 | |
2b188cc1 JA |
967 | static struct kmem_cache *req_cachep; |
968 | ||
0918682b | 969 | static const struct file_operations io_uring_fops; |
2b188cc1 JA |
970 | |
971 | struct sock *io_uring_get_socket(struct file *file) | |
972 | { | |
973 | #if defined(CONFIG_UNIX) | |
974 | if (file->f_op == &io_uring_fops) { | |
975 | struct io_ring_ctx *ctx = file->private_data; | |
976 | ||
977 | return ctx->ring_sock->sk; | |
978 | } | |
979 | #endif | |
980 | return NULL; | |
981 | } | |
982 | EXPORT_SYMBOL(io_uring_get_socket); | |
983 | ||
3ca405eb PB |
984 | static inline void io_clean_op(struct io_kiocb *req) |
985 | { | |
bb175342 PB |
986 | if (req->flags & (REQ_F_NEED_CLEANUP | REQ_F_BUFFER_SELECTED | |
987 | REQ_F_INFLIGHT)) | |
3ca405eb PB |
988 | __io_clean_op(req); |
989 | } | |
990 | ||
4349f30e | 991 | static void io_sq_thread_drop_mm(void) |
c40f6379 JA |
992 | { |
993 | struct mm_struct *mm = current->mm; | |
994 | ||
995 | if (mm) { | |
996 | kthread_unuse_mm(mm); | |
997 | mmput(mm); | |
998 | } | |
999 | } | |
1000 | ||
1001 | static int __io_sq_thread_acquire_mm(struct io_ring_ctx *ctx) | |
1002 | { | |
1003 | if (!current->mm) { | |
cbcf7214 | 1004 | if (unlikely(!(ctx->flags & IORING_SETUP_SQPOLL) || |
2aede0e4 JA |
1005 | !ctx->sqo_task->mm || |
1006 | !mmget_not_zero(ctx->sqo_task->mm))) | |
c40f6379 | 1007 | return -EFAULT; |
2aede0e4 | 1008 | kthread_use_mm(ctx->sqo_task->mm); |
c40f6379 JA |
1009 | } |
1010 | ||
1011 | return 0; | |
1012 | } | |
1013 | ||
1014 | static int io_sq_thread_acquire_mm(struct io_ring_ctx *ctx, | |
1015 | struct io_kiocb *req) | |
1016 | { | |
0f203765 | 1017 | if (!(io_op_defs[req->opcode].work_flags & IO_WQ_WORK_MM)) |
c40f6379 JA |
1018 | return 0; |
1019 | return __io_sq_thread_acquire_mm(ctx); | |
1020 | } | |
1021 | ||
91d8f519 DZ |
1022 | static void io_sq_thread_associate_blkcg(struct io_ring_ctx *ctx, |
1023 | struct cgroup_subsys_state **cur_css) | |
1024 | ||
1025 | { | |
1026 | #ifdef CONFIG_BLK_CGROUP | |
1027 | /* puts the old one when swapping */ | |
1028 | if (*cur_css != ctx->sqo_blkcg_css) { | |
1029 | kthread_associate_blkcg(ctx->sqo_blkcg_css); | |
1030 | *cur_css = ctx->sqo_blkcg_css; | |
1031 | } | |
1032 | #endif | |
1033 | } | |
1034 | ||
1035 | static void io_sq_thread_unassociate_blkcg(void) | |
1036 | { | |
1037 | #ifdef CONFIG_BLK_CGROUP | |
1038 | kthread_associate_blkcg(NULL); | |
1039 | #endif | |
1040 | } | |
1041 | ||
c40f6379 JA |
1042 | static inline void req_set_fail_links(struct io_kiocb *req) |
1043 | { | |
1044 | if ((req->flags & (REQ_F_LINK | REQ_F_HARDLINK)) == REQ_F_LINK) | |
1045 | req->flags |= REQ_F_FAIL_LINK; | |
1046 | } | |
4a38aed2 | 1047 | |
1e6fa521 JA |
1048 | /* |
1049 | * None of these are dereferenced, they are simply used to check if any of | |
1050 | * them have changed. If we're under current and check they are still the | |
1051 | * same, we're fine to grab references to them for actual out-of-line use. | |
1052 | */ | |
1053 | static void io_init_identity(struct io_identity *id) | |
1054 | { | |
1055 | id->files = current->files; | |
1056 | id->mm = current->mm; | |
1057 | #ifdef CONFIG_BLK_CGROUP | |
1058 | rcu_read_lock(); | |
1059 | id->blkcg_css = blkcg_css(); | |
1060 | rcu_read_unlock(); | |
1061 | #endif | |
1062 | id->creds = current_cred(); | |
1063 | id->nsproxy = current->nsproxy; | |
1064 | id->fs = current->fs; | |
1065 | id->fsize = rlimit(RLIMIT_FSIZE); | |
4ea33a97 JA |
1066 | #ifdef CONFIG_AUDIT |
1067 | id->loginuid = current->loginuid; | |
1068 | id->sessionid = current->sessionid; | |
1069 | #endif | |
1e6fa521 JA |
1070 | refcount_set(&id->count, 1); |
1071 | } | |
1072 | ||
ec99ca6c PB |
1073 | static inline void __io_req_init_async(struct io_kiocb *req) |
1074 | { | |
1075 | memset(&req->work, 0, sizeof(req->work)); | |
1076 | req->flags |= REQ_F_WORK_INITIALIZED; | |
1077 | } | |
1078 | ||
7cdaf587 XW |
1079 | /* |
1080 | * Note: must call io_req_init_async() for the first time you | |
1081 | * touch any members of io_wq_work. | |
1082 | */ | |
1083 | static inline void io_req_init_async(struct io_kiocb *req) | |
1084 | { | |
500a373d JA |
1085 | struct io_uring_task *tctx = current->io_uring; |
1086 | ||
7cdaf587 XW |
1087 | if (req->flags & REQ_F_WORK_INITIALIZED) |
1088 | return; | |
1089 | ||
ec99ca6c | 1090 | __io_req_init_async(req); |
500a373d JA |
1091 | |
1092 | /* Grab a ref if this isn't our static identity */ | |
1093 | req->work.identity = tctx->identity; | |
1094 | if (tctx->identity != &tctx->__identity) | |
1095 | refcount_inc(&req->work.identity->count); | |
7cdaf587 XW |
1096 | } |
1097 | ||
0cdaf760 PB |
1098 | static inline bool io_async_submit(struct io_ring_ctx *ctx) |
1099 | { | |
1100 | return ctx->flags & IORING_SETUP_SQPOLL; | |
1101 | } | |
1102 | ||
2b188cc1 JA |
1103 | static void io_ring_ctx_ref_free(struct percpu_ref *ref) |
1104 | { | |
1105 | struct io_ring_ctx *ctx = container_of(ref, struct io_ring_ctx, refs); | |
1106 | ||
0f158b4c | 1107 | complete(&ctx->ref_comp); |
2b188cc1 JA |
1108 | } |
1109 | ||
8eb7e2d0 PB |
1110 | static inline bool io_is_timeout_noseq(struct io_kiocb *req) |
1111 | { | |
1112 | return !req->timeout.off; | |
1113 | } | |
1114 | ||
2b188cc1 JA |
1115 | static struct io_ring_ctx *io_ring_ctx_alloc(struct io_uring_params *p) |
1116 | { | |
1117 | struct io_ring_ctx *ctx; | |
78076bb6 | 1118 | int hash_bits; |
2b188cc1 JA |
1119 | |
1120 | ctx = kzalloc(sizeof(*ctx), GFP_KERNEL); | |
1121 | if (!ctx) | |
1122 | return NULL; | |
1123 | ||
0ddf92e8 JA |
1124 | ctx->fallback_req = kmem_cache_alloc(req_cachep, GFP_KERNEL); |
1125 | if (!ctx->fallback_req) | |
1126 | goto err; | |
1127 | ||
78076bb6 JA |
1128 | /* |
1129 | * Use 5 bits less than the max cq entries, that should give us around | |
1130 | * 32 entries per hash list if totally full and uniformly spread. | |
1131 | */ | |
1132 | hash_bits = ilog2(p->cq_entries); | |
1133 | hash_bits -= 5; | |
1134 | if (hash_bits <= 0) | |
1135 | hash_bits = 1; | |
1136 | ctx->cancel_hash_bits = hash_bits; | |
1137 | ctx->cancel_hash = kmalloc((1U << hash_bits) * sizeof(struct hlist_head), | |
1138 | GFP_KERNEL); | |
1139 | if (!ctx->cancel_hash) | |
1140 | goto err; | |
1141 | __hash_init(ctx->cancel_hash, 1U << hash_bits); | |
1142 | ||
21482896 | 1143 | if (percpu_ref_init(&ctx->refs, io_ring_ctx_ref_free, |
206aefde JA |
1144 | PERCPU_REF_ALLOW_REINIT, GFP_KERNEL)) |
1145 | goto err; | |
2b188cc1 JA |
1146 | |
1147 | ctx->flags = p->flags; | |
90554200 | 1148 | init_waitqueue_head(&ctx->sqo_sq_wait); |
69fb2131 | 1149 | INIT_LIST_HEAD(&ctx->sqd_list); |
2b188cc1 | 1150 | init_waitqueue_head(&ctx->cq_wait); |
1d7bb1d5 | 1151 | INIT_LIST_HEAD(&ctx->cq_overflow_list); |
0f158b4c JA |
1152 | init_completion(&ctx->ref_comp); |
1153 | init_completion(&ctx->sq_thread_comp); | |
5a2e745d | 1154 | idr_init(&ctx->io_buffer_idr); |
071698e1 | 1155 | idr_init(&ctx->personality_idr); |
2b188cc1 JA |
1156 | mutex_init(&ctx->uring_lock); |
1157 | init_waitqueue_head(&ctx->wait); | |
1158 | spin_lock_init(&ctx->completion_lock); | |
540e32a0 | 1159 | INIT_LIST_HEAD(&ctx->iopoll_list); |
de0617e4 | 1160 | INIT_LIST_HEAD(&ctx->defer_list); |
5262f567 | 1161 | INIT_LIST_HEAD(&ctx->timeout_list); |
fcb323cc JA |
1162 | init_waitqueue_head(&ctx->inflight_wait); |
1163 | spin_lock_init(&ctx->inflight_lock); | |
1164 | INIT_LIST_HEAD(&ctx->inflight_list); | |
4a38aed2 JA |
1165 | INIT_DELAYED_WORK(&ctx->file_put_work, io_file_put_work); |
1166 | init_llist_head(&ctx->file_put_llist); | |
2b188cc1 | 1167 | return ctx; |
206aefde | 1168 | err: |
0ddf92e8 JA |
1169 | if (ctx->fallback_req) |
1170 | kmem_cache_free(req_cachep, ctx->fallback_req); | |
78076bb6 | 1171 | kfree(ctx->cancel_hash); |
206aefde JA |
1172 | kfree(ctx); |
1173 | return NULL; | |
2b188cc1 JA |
1174 | } |
1175 | ||
9cf7c104 | 1176 | static bool req_need_defer(struct io_kiocb *req, u32 seq) |
7adf4eaf | 1177 | { |
2bc9930e JA |
1178 | if (unlikely(req->flags & REQ_F_IO_DRAIN)) { |
1179 | struct io_ring_ctx *ctx = req->ctx; | |
a197f664 | 1180 | |
9cf7c104 | 1181 | return seq != ctx->cached_cq_tail |
2c3bac6d | 1182 | + READ_ONCE(ctx->cached_cq_overflow); |
2bc9930e | 1183 | } |
de0617e4 | 1184 | |
9d858b21 | 1185 | return false; |
de0617e4 JA |
1186 | } |
1187 | ||
de0617e4 | 1188 | static void __io_commit_cqring(struct io_ring_ctx *ctx) |
2b188cc1 | 1189 | { |
75b28aff | 1190 | struct io_rings *rings = ctx->rings; |
2b188cc1 | 1191 | |
07910158 PB |
1192 | /* order cqe stores with ring update */ |
1193 | smp_store_release(&rings->cq.tail, ctx->cached_cq_tail); | |
2b188cc1 | 1194 | |
07910158 PB |
1195 | if (wq_has_sleeper(&ctx->cq_wait)) { |
1196 | wake_up_interruptible(&ctx->cq_wait); | |
1197 | kill_fasync(&ctx->cq_fasync, SIGIO, POLL_IN); | |
2b188cc1 JA |
1198 | } |
1199 | } | |
1200 | ||
5c3462cf | 1201 | static void io_put_identity(struct io_uring_task *tctx, struct io_kiocb *req) |
1e6fa521 | 1202 | { |
500a373d | 1203 | if (req->work.identity == &tctx->__identity) |
1e6fa521 JA |
1204 | return; |
1205 | if (refcount_dec_and_test(&req->work.identity->count)) | |
1206 | kfree(req->work.identity); | |
1207 | } | |
1208 | ||
4edf20f9 | 1209 | static void io_req_clean_work(struct io_kiocb *req) |
18d9be1a | 1210 | { |
7cdaf587 | 1211 | if (!(req->flags & REQ_F_WORK_INITIALIZED)) |
4edf20f9 | 1212 | return; |
51a4cc11 JA |
1213 | |
1214 | req->flags &= ~REQ_F_WORK_INITIALIZED; | |
7cdaf587 | 1215 | |
dfead8a8 | 1216 | if (req->work.flags & IO_WQ_WORK_MM) { |
98447d65 | 1217 | mmdrop(req->work.identity->mm); |
dfead8a8 | 1218 | req->work.flags &= ~IO_WQ_WORK_MM; |
cccf0ee8 | 1219 | } |
91d8f519 | 1220 | #ifdef CONFIG_BLK_CGROUP |
dfead8a8 | 1221 | if (req->work.flags & IO_WQ_WORK_BLKCG) { |
98447d65 | 1222 | css_put(req->work.identity->blkcg_css); |
dfead8a8 JA |
1223 | req->work.flags &= ~IO_WQ_WORK_BLKCG; |
1224 | } | |
91d8f519 | 1225 | #endif |
dfead8a8 | 1226 | if (req->work.flags & IO_WQ_WORK_CREDS) { |
98447d65 | 1227 | put_cred(req->work.identity->creds); |
dfead8a8 | 1228 | req->work.flags &= ~IO_WQ_WORK_CREDS; |
cccf0ee8 | 1229 | } |
dfead8a8 | 1230 | if (req->work.flags & IO_WQ_WORK_FS) { |
98447d65 | 1231 | struct fs_struct *fs = req->work.identity->fs; |
51a4cc11 | 1232 | |
98447d65 | 1233 | spin_lock(&req->work.identity->fs->lock); |
ff002b30 JA |
1234 | if (--fs->users) |
1235 | fs = NULL; | |
98447d65 | 1236 | spin_unlock(&req->work.identity->fs->lock); |
ff002b30 JA |
1237 | if (fs) |
1238 | free_fs_struct(fs); | |
dfead8a8 | 1239 | req->work.flags &= ~IO_WQ_WORK_FS; |
ff002b30 | 1240 | } |
51a4cc11 | 1241 | |
5c3462cf | 1242 | io_put_identity(req->task->io_uring, req); |
561fb04a JA |
1243 | } |
1244 | ||
1e6fa521 JA |
1245 | /* |
1246 | * Create a private copy of io_identity, since some fields don't match | |
1247 | * the current context. | |
1248 | */ | |
1249 | static bool io_identity_cow(struct io_kiocb *req) | |
1250 | { | |
5c3462cf | 1251 | struct io_uring_task *tctx = current->io_uring; |
1e6fa521 JA |
1252 | const struct cred *creds = NULL; |
1253 | struct io_identity *id; | |
1254 | ||
1255 | if (req->work.flags & IO_WQ_WORK_CREDS) | |
1256 | creds = req->work.identity->creds; | |
1257 | ||
1258 | id = kmemdup(req->work.identity, sizeof(*id), GFP_KERNEL); | |
1259 | if (unlikely(!id)) { | |
1260 | req->work.flags |= IO_WQ_WORK_CANCEL; | |
1261 | return false; | |
1262 | } | |
1263 | ||
1264 | /* | |
1265 | * We can safely just re-init the creds we copied Either the field | |
1266 | * matches the current one, or we haven't grabbed it yet. The only | |
1267 | * exception is ->creds, through registered personalities, so handle | |
1268 | * that one separately. | |
1269 | */ | |
1270 | io_init_identity(id); | |
1271 | if (creds) | |
1272 | req->work.identity->creds = creds; | |
1273 | ||
1274 | /* add one for this request */ | |
1275 | refcount_inc(&id->count); | |
1276 | ||
1277 | /* drop old identity, assign new one. one ref for req, one for tctx */ | |
500a373d | 1278 | if (req->work.identity != tctx->identity && |
1e6fa521 JA |
1279 | refcount_sub_and_test(2, &req->work.identity->count)) |
1280 | kfree(req->work.identity); | |
1281 | ||
1282 | req->work.identity = id; | |
500a373d | 1283 | tctx->identity = id; |
1e6fa521 JA |
1284 | return true; |
1285 | } | |
1286 | ||
1287 | static bool io_grab_identity(struct io_kiocb *req) | |
18d9be1a | 1288 | { |
d3656344 | 1289 | const struct io_op_def *def = &io_op_defs[req->opcode]; |
5c3462cf | 1290 | struct io_identity *id = req->work.identity; |
23329513 | 1291 | struct io_ring_ctx *ctx = req->ctx; |
54a91f3b | 1292 | |
69228338 JA |
1293 | if (def->work_flags & IO_WQ_WORK_FSIZE) { |
1294 | if (id->fsize != rlimit(RLIMIT_FSIZE)) | |
1295 | return false; | |
1296 | req->work.flags |= IO_WQ_WORK_FSIZE; | |
1297 | } | |
16d59803 | 1298 | |
dfead8a8 | 1299 | if (!(req->work.flags & IO_WQ_WORK_FILES) && |
1e6fa521 | 1300 | (def->work_flags & IO_WQ_WORK_FILES) && |
23329513 | 1301 | !(req->flags & REQ_F_NO_FILE_TABLE)) { |
1e6fa521 JA |
1302 | if (id->files != current->files || |
1303 | id->nsproxy != current->nsproxy) | |
1304 | return false; | |
1305 | atomic_inc(&id->files->count); | |
1306 | get_nsproxy(id->nsproxy); | |
23329513 PB |
1307 | req->flags |= REQ_F_INFLIGHT; |
1308 | ||
1309 | spin_lock_irq(&ctx->inflight_lock); | |
1310 | list_add(&req->inflight_entry, &ctx->inflight_list); | |
1311 | spin_unlock_irq(&ctx->inflight_lock); | |
dfead8a8 | 1312 | req->work.flags |= IO_WQ_WORK_FILES; |
dca9cf8b | 1313 | } |
91d8f519 | 1314 | #ifdef CONFIG_BLK_CGROUP |
dfead8a8 JA |
1315 | if (!(req->work.flags & IO_WQ_WORK_BLKCG) && |
1316 | (def->work_flags & IO_WQ_WORK_BLKCG)) { | |
91d8f519 | 1317 | rcu_read_lock(); |
1e6fa521 JA |
1318 | if (id->blkcg_css != blkcg_css()) { |
1319 | rcu_read_unlock(); | |
1320 | return false; | |
1321 | } | |
91d8f519 DZ |
1322 | /* |
1323 | * This should be rare, either the cgroup is dying or the task | |
1324 | * is moving cgroups. Just punt to root for the handful of ios. | |
1325 | */ | |
1e6fa521 | 1326 | if (css_tryget_online(id->blkcg_css)) |
dfead8a8 | 1327 | req->work.flags |= IO_WQ_WORK_BLKCG; |
91d8f519 DZ |
1328 | rcu_read_unlock(); |
1329 | } | |
1330 | #endif | |
dfead8a8 | 1331 | if (!(req->work.flags & IO_WQ_WORK_CREDS)) { |
1e6fa521 JA |
1332 | if (id->creds != current_cred()) |
1333 | return false; | |
1334 | get_cred(id->creds); | |
dfead8a8 JA |
1335 | req->work.flags |= IO_WQ_WORK_CREDS; |
1336 | } | |
4ea33a97 JA |
1337 | #ifdef CONFIG_AUDIT |
1338 | if (!uid_eq(current->loginuid, id->loginuid) || | |
1339 | current->sessionid != id->sessionid) | |
1340 | return false; | |
1341 | #endif | |
dfead8a8 JA |
1342 | if (!(req->work.flags & IO_WQ_WORK_FS) && |
1343 | (def->work_flags & IO_WQ_WORK_FS)) { | |
1e6fa521 JA |
1344 | if (current->fs != id->fs) |
1345 | return false; | |
1346 | spin_lock(&id->fs->lock); | |
1347 | if (!id->fs->in_exec) { | |
1348 | id->fs->users++; | |
dfead8a8 | 1349 | req->work.flags |= IO_WQ_WORK_FS; |
dca9cf8b PB |
1350 | } else { |
1351 | req->work.flags |= IO_WQ_WORK_CANCEL; | |
1352 | } | |
1353 | spin_unlock(¤t->fs->lock); | |
1354 | } | |
1e6fa521 JA |
1355 | |
1356 | return true; | |
1357 | } | |
1358 | ||
1359 | static void io_prep_async_work(struct io_kiocb *req) | |
1360 | { | |
1361 | const struct io_op_def *def = &io_op_defs[req->opcode]; | |
1e6fa521 | 1362 | struct io_ring_ctx *ctx = req->ctx; |
5c3462cf | 1363 | struct io_identity *id; |
1e6fa521 JA |
1364 | |
1365 | io_req_init_async(req); | |
5c3462cf | 1366 | id = req->work.identity; |
1e6fa521 | 1367 | |
feaadc4f PB |
1368 | if (req->flags & REQ_F_FORCE_ASYNC) |
1369 | req->work.flags |= IO_WQ_WORK_CONCURRENT; | |
1370 | ||
1e6fa521 JA |
1371 | if (req->flags & REQ_F_ISREG) { |
1372 | if (def->hash_reg_file || (ctx->flags & IORING_SETUP_IOPOLL)) | |
1373 | io_wq_hash_work(&req->work, file_inode(req->file)); | |
1374 | } else { | |
1375 | if (def->unbound_nonreg_file) | |
1376 | req->work.flags |= IO_WQ_WORK_UNBOUND; | |
1377 | } | |
1378 | ||
1379 | /* ->mm can never change on us */ | |
1380 | if (!(req->work.flags & IO_WQ_WORK_MM) && | |
1381 | (def->work_flags & IO_WQ_WORK_MM)) { | |
1382 | mmgrab(id->mm); | |
1383 | req->work.flags |= IO_WQ_WORK_MM; | |
1384 | } | |
1385 | ||
1386 | /* if we fail grabbing identity, we must COW, regrab, and retry */ | |
1387 | if (io_grab_identity(req)) | |
1388 | return; | |
1389 | ||
1390 | if (!io_identity_cow(req)) | |
1391 | return; | |
1392 | ||
1393 | /* can't fail at this point */ | |
1394 | if (!io_grab_identity(req)) | |
1395 | WARN_ON(1); | |
561fb04a | 1396 | } |
cccf0ee8 | 1397 | |
cbdcb435 | 1398 | static void io_prep_async_link(struct io_kiocb *req) |
561fb04a | 1399 | { |
cbdcb435 | 1400 | struct io_kiocb *cur; |
54a91f3b | 1401 | |
cbdcb435 PB |
1402 | io_prep_async_work(req); |
1403 | if (req->flags & REQ_F_LINK_HEAD) | |
1404 | list_for_each_entry(cur, &req->link_list, link_list) | |
1405 | io_prep_async_work(cur); | |
561fb04a JA |
1406 | } |
1407 | ||
7271ef3a | 1408 | static struct io_kiocb *__io_queue_async_work(struct io_kiocb *req) |
561fb04a | 1409 | { |
a197f664 | 1410 | struct io_ring_ctx *ctx = req->ctx; |
cbdcb435 | 1411 | struct io_kiocb *link = io_prep_linked_timeout(req); |
561fb04a | 1412 | |
8766dd51 PB |
1413 | trace_io_uring_queue_async_work(ctx, io_wq_is_hashed(&req->work), req, |
1414 | &req->work, req->flags); | |
1415 | io_wq_enqueue(ctx->io_wq, &req->work); | |
7271ef3a | 1416 | return link; |
18d9be1a JA |
1417 | } |
1418 | ||
cbdcb435 PB |
1419 | static void io_queue_async_work(struct io_kiocb *req) |
1420 | { | |
7271ef3a JA |
1421 | struct io_kiocb *link; |
1422 | ||
cbdcb435 PB |
1423 | /* init ->work of the whole link before punting */ |
1424 | io_prep_async_link(req); | |
7271ef3a JA |
1425 | link = __io_queue_async_work(req); |
1426 | ||
1427 | if (link) | |
1428 | io_queue_linked_timeout(link); | |
cbdcb435 PB |
1429 | } |
1430 | ||
5262f567 JA |
1431 | static void io_kill_timeout(struct io_kiocb *req) |
1432 | { | |
e8c2bc1f | 1433 | struct io_timeout_data *io = req->async_data; |
5262f567 JA |
1434 | int ret; |
1435 | ||
e8c2bc1f | 1436 | ret = hrtimer_try_to_cancel(&io->timer); |
5262f567 | 1437 | if (ret != -1) { |
01cec8c1 PB |
1438 | atomic_set(&req->ctx->cq_timeouts, |
1439 | atomic_read(&req->ctx->cq_timeouts) + 1); | |
135fcde8 | 1440 | list_del_init(&req->timeout.list); |
78e19bbe | 1441 | io_cqring_fill_event(req, 0); |
216578e5 | 1442 | io_put_req_deferred(req, 1); |
5262f567 JA |
1443 | } |
1444 | } | |
1445 | ||
f3606e3a JA |
1446 | static bool io_task_match(struct io_kiocb *req, struct task_struct *tsk) |
1447 | { | |
1448 | struct io_ring_ctx *ctx = req->ctx; | |
1449 | ||
1450 | if (!tsk || req->task == tsk) | |
1451 | return true; | |
534ca6d6 JA |
1452 | if (ctx->flags & IORING_SETUP_SQPOLL) { |
1453 | if (ctx->sq_data && req->task == ctx->sq_data->thread) | |
1454 | return true; | |
1455 | } | |
f3606e3a JA |
1456 | return false; |
1457 | } | |
1458 | ||
76e1b642 JA |
1459 | /* |
1460 | * Returns true if we found and killed one or more timeouts | |
1461 | */ | |
1462 | static bool io_kill_timeouts(struct io_ring_ctx *ctx, struct task_struct *tsk) | |
5262f567 JA |
1463 | { |
1464 | struct io_kiocb *req, *tmp; | |
76e1b642 | 1465 | int canceled = 0; |
5262f567 JA |
1466 | |
1467 | spin_lock_irq(&ctx->completion_lock); | |
f3606e3a | 1468 | list_for_each_entry_safe(req, tmp, &ctx->timeout_list, timeout.list) { |
76e1b642 | 1469 | if (io_task_match(req, tsk)) { |
f3606e3a | 1470 | io_kill_timeout(req); |
76e1b642 JA |
1471 | canceled++; |
1472 | } | |
f3606e3a | 1473 | } |
5262f567 | 1474 | spin_unlock_irq(&ctx->completion_lock); |
76e1b642 | 1475 | return canceled != 0; |
5262f567 JA |
1476 | } |
1477 | ||
04518945 | 1478 | static void __io_queue_deferred(struct io_ring_ctx *ctx) |
de0617e4 | 1479 | { |
04518945 | 1480 | do { |
27dc8338 PB |
1481 | struct io_defer_entry *de = list_first_entry(&ctx->defer_list, |
1482 | struct io_defer_entry, list); | |
7271ef3a | 1483 | struct io_kiocb *link; |
de0617e4 | 1484 | |
9cf7c104 | 1485 | if (req_need_defer(de->req, de->seq)) |
04518945 | 1486 | break; |
27dc8338 | 1487 | list_del_init(&de->list); |
cbdcb435 | 1488 | /* punt-init is done before queueing for defer */ |
7271ef3a JA |
1489 | link = __io_queue_async_work(de->req); |
1490 | if (link) { | |
1491 | __io_queue_linked_timeout(link); | |
1492 | /* drop submission reference */ | |
216578e5 | 1493 | io_put_req_deferred(link, 1); |
7271ef3a | 1494 | } |
27dc8338 | 1495 | kfree(de); |
04518945 PB |
1496 | } while (!list_empty(&ctx->defer_list)); |
1497 | } | |
1498 | ||
360428f8 | 1499 | static void io_flush_timeouts(struct io_ring_ctx *ctx) |
de0617e4 | 1500 | { |
360428f8 PB |
1501 | while (!list_empty(&ctx->timeout_list)) { |
1502 | struct io_kiocb *req = list_first_entry(&ctx->timeout_list, | |
135fcde8 | 1503 | struct io_kiocb, timeout.list); |
de0617e4 | 1504 | |
8eb7e2d0 | 1505 | if (io_is_timeout_noseq(req)) |
360428f8 | 1506 | break; |
bfe68a22 PB |
1507 | if (req->timeout.target_seq != ctx->cached_cq_tail |
1508 | - atomic_read(&ctx->cq_timeouts)) | |
360428f8 | 1509 | break; |
bfe68a22 | 1510 | |
135fcde8 | 1511 | list_del_init(&req->timeout.list); |
5262f567 | 1512 | io_kill_timeout(req); |
360428f8 PB |
1513 | } |
1514 | } | |
5262f567 | 1515 | |
360428f8 PB |
1516 | static void io_commit_cqring(struct io_ring_ctx *ctx) |
1517 | { | |
1518 | io_flush_timeouts(ctx); | |
de0617e4 JA |
1519 | __io_commit_cqring(ctx); |
1520 | ||
04518945 PB |
1521 | if (unlikely(!list_empty(&ctx->defer_list))) |
1522 | __io_queue_deferred(ctx); | |
de0617e4 JA |
1523 | } |
1524 | ||
90554200 JA |
1525 | static inline bool io_sqring_full(struct io_ring_ctx *ctx) |
1526 | { | |
1527 | struct io_rings *r = ctx->rings; | |
1528 | ||
1529 | return READ_ONCE(r->sq.tail) - ctx->cached_sq_head == r->sq_ring_entries; | |
1530 | } | |
1531 | ||
2b188cc1 JA |
1532 | static struct io_uring_cqe *io_get_cqring(struct io_ring_ctx *ctx) |
1533 | { | |
75b28aff | 1534 | struct io_rings *rings = ctx->rings; |
2b188cc1 JA |
1535 | unsigned tail; |
1536 | ||
1537 | tail = ctx->cached_cq_tail; | |
115e12e5 SB |
1538 | /* |
1539 | * writes to the cq entry need to come after reading head; the | |
1540 | * control dependency is enough as we're using WRITE_ONCE to | |
1541 | * fill the cq entry | |
1542 | */ | |
75b28aff | 1543 | if (tail - READ_ONCE(rings->cq.head) == rings->cq_ring_entries) |
2b188cc1 JA |
1544 | return NULL; |
1545 | ||
1546 | ctx->cached_cq_tail++; | |
75b28aff | 1547 | return &rings->cqes[tail & ctx->cq_mask]; |
2b188cc1 JA |
1548 | } |
1549 | ||
f2842ab5 JA |
1550 | static inline bool io_should_trigger_evfd(struct io_ring_ctx *ctx) |
1551 | { | |
f0b493e6 JA |
1552 | if (!ctx->cq_ev_fd) |
1553 | return false; | |
7e55a19c SG |
1554 | if (READ_ONCE(ctx->rings->cq_flags) & IORING_CQ_EVENTFD_DISABLED) |
1555 | return false; | |
f2842ab5 JA |
1556 | if (!ctx->eventfd_async) |
1557 | return true; | |
b41e9852 | 1558 | return io_wq_current_is_worker(); |
f2842ab5 JA |
1559 | } |
1560 | ||
b41e9852 | 1561 | static void io_cqring_ev_posted(struct io_ring_ctx *ctx) |
1d7bb1d5 JA |
1562 | { |
1563 | if (waitqueue_active(&ctx->wait)) | |
1564 | wake_up(&ctx->wait); | |
534ca6d6 JA |
1565 | if (ctx->sq_data && waitqueue_active(&ctx->sq_data->wait)) |
1566 | wake_up(&ctx->sq_data->wait); | |
b41e9852 | 1567 | if (io_should_trigger_evfd(ctx)) |
1d7bb1d5 JA |
1568 | eventfd_signal(ctx->cq_ev_fd, 1); |
1569 | } | |
1570 | ||
46930143 PB |
1571 | static void io_cqring_mark_overflow(struct io_ring_ctx *ctx) |
1572 | { | |
1573 | if (list_empty(&ctx->cq_overflow_list)) { | |
1574 | clear_bit(0, &ctx->sq_check_overflow); | |
1575 | clear_bit(0, &ctx->cq_check_overflow); | |
1576 | ctx->rings->sq_flags &= ~IORING_SQ_CQ_OVERFLOW; | |
1577 | } | |
1578 | } | |
1579 | ||
e6c8aa9a JA |
1580 | static inline bool io_match_files(struct io_kiocb *req, |
1581 | struct files_struct *files) | |
1582 | { | |
1583 | if (!files) | |
1584 | return true; | |
dfead8a8 JA |
1585 | if ((req->flags & REQ_F_WORK_INITIALIZED) && |
1586 | (req->work.flags & IO_WQ_WORK_FILES)) | |
98447d65 | 1587 | return req->work.identity->files == files; |
e6c8aa9a JA |
1588 | return false; |
1589 | } | |
1590 | ||
c4a2ed72 | 1591 | /* Returns true if there are no backlogged entries after the flush */ |
e6c8aa9a JA |
1592 | static bool io_cqring_overflow_flush(struct io_ring_ctx *ctx, bool force, |
1593 | struct task_struct *tsk, | |
1594 | struct files_struct *files) | |
1d7bb1d5 JA |
1595 | { |
1596 | struct io_rings *rings = ctx->rings; | |
e6c8aa9a | 1597 | struct io_kiocb *req, *tmp; |
1d7bb1d5 | 1598 | struct io_uring_cqe *cqe; |
1d7bb1d5 JA |
1599 | unsigned long flags; |
1600 | LIST_HEAD(list); | |
1601 | ||
1602 | if (!force) { | |
1603 | if (list_empty_careful(&ctx->cq_overflow_list)) | |
c4a2ed72 | 1604 | return true; |
1d7bb1d5 JA |
1605 | if ((ctx->cached_cq_tail - READ_ONCE(rings->cq.head) == |
1606 | rings->cq_ring_entries)) | |
c4a2ed72 | 1607 | return false; |
1d7bb1d5 JA |
1608 | } |
1609 | ||
1610 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
1611 | ||
1612 | /* if force is set, the ring is going away. always drop after that */ | |
1613 | if (force) | |
69b3e546 | 1614 | ctx->cq_overflow_flushed = 1; |
1d7bb1d5 | 1615 | |
c4a2ed72 | 1616 | cqe = NULL; |
e6c8aa9a JA |
1617 | list_for_each_entry_safe(req, tmp, &ctx->cq_overflow_list, compl.list) { |
1618 | if (tsk && req->task != tsk) | |
1619 | continue; | |
1620 | if (!io_match_files(req, files)) | |
1621 | continue; | |
1622 | ||
1d7bb1d5 JA |
1623 | cqe = io_get_cqring(ctx); |
1624 | if (!cqe && !force) | |
1625 | break; | |
1626 | ||
40d8ddd4 | 1627 | list_move(&req->compl.list, &list); |
1d7bb1d5 JA |
1628 | if (cqe) { |
1629 | WRITE_ONCE(cqe->user_data, req->user_data); | |
1630 | WRITE_ONCE(cqe->res, req->result); | |
0f7e466b | 1631 | WRITE_ONCE(cqe->flags, req->compl.cflags); |
1d7bb1d5 | 1632 | } else { |
2c3bac6d | 1633 | ctx->cached_cq_overflow++; |
1d7bb1d5 | 1634 | WRITE_ONCE(ctx->rings->cq_overflow, |
2c3bac6d | 1635 | ctx->cached_cq_overflow); |
1d7bb1d5 JA |
1636 | } |
1637 | } | |
1638 | ||
1639 | io_commit_cqring(ctx); | |
46930143 PB |
1640 | io_cqring_mark_overflow(ctx); |
1641 | ||
1d7bb1d5 JA |
1642 | spin_unlock_irqrestore(&ctx->completion_lock, flags); |
1643 | io_cqring_ev_posted(ctx); | |
1644 | ||
1645 | while (!list_empty(&list)) { | |
40d8ddd4 PB |
1646 | req = list_first_entry(&list, struct io_kiocb, compl.list); |
1647 | list_del(&req->compl.list); | |
ec9c02ad | 1648 | io_put_req(req); |
1d7bb1d5 | 1649 | } |
c4a2ed72 JA |
1650 | |
1651 | return cqe != NULL; | |
1d7bb1d5 JA |
1652 | } |
1653 | ||
bcda7baa | 1654 | static void __io_cqring_fill_event(struct io_kiocb *req, long res, long cflags) |
2b188cc1 | 1655 | { |
78e19bbe | 1656 | struct io_ring_ctx *ctx = req->ctx; |
2b188cc1 JA |
1657 | struct io_uring_cqe *cqe; |
1658 | ||
78e19bbe | 1659 | trace_io_uring_complete(ctx, req->user_data, res); |
51c3ff62 | 1660 | |
2b188cc1 JA |
1661 | /* |
1662 | * If we can't get a cq entry, userspace overflowed the | |
1663 | * submission (by quite a lot). Increment the overflow count in | |
1664 | * the ring. | |
1665 | */ | |
1666 | cqe = io_get_cqring(ctx); | |
1d7bb1d5 | 1667 | if (likely(cqe)) { |
78e19bbe | 1668 | WRITE_ONCE(cqe->user_data, req->user_data); |
2b188cc1 | 1669 | WRITE_ONCE(cqe->res, res); |
bcda7baa | 1670 | WRITE_ONCE(cqe->flags, cflags); |
0f212204 JA |
1671 | } else if (ctx->cq_overflow_flushed || req->task->io_uring->in_idle) { |
1672 | /* | |
1673 | * If we're in ring overflow flush mode, or in task cancel mode, | |
1674 | * then we cannot store the request for later flushing, we need | |
1675 | * to drop it on the floor. | |
1676 | */ | |
2c3bac6d PB |
1677 | ctx->cached_cq_overflow++; |
1678 | WRITE_ONCE(ctx->rings->cq_overflow, ctx->cached_cq_overflow); | |
1d7bb1d5 | 1679 | } else { |
ad3eb2c8 JA |
1680 | if (list_empty(&ctx->cq_overflow_list)) { |
1681 | set_bit(0, &ctx->sq_check_overflow); | |
1682 | set_bit(0, &ctx->cq_check_overflow); | |
6d5f9049 | 1683 | ctx->rings->sq_flags |= IORING_SQ_CQ_OVERFLOW; |
ad3eb2c8 | 1684 | } |
40d8ddd4 | 1685 | io_clean_op(req); |
1d7bb1d5 | 1686 | req->result = res; |
0f7e466b | 1687 | req->compl.cflags = cflags; |
40d8ddd4 PB |
1688 | refcount_inc(&req->refs); |
1689 | list_add_tail(&req->compl.list, &ctx->cq_overflow_list); | |
2b188cc1 JA |
1690 | } |
1691 | } | |
1692 | ||
bcda7baa JA |
1693 | static void io_cqring_fill_event(struct io_kiocb *req, long res) |
1694 | { | |
1695 | __io_cqring_fill_event(req, res, 0); | |
1696 | } | |
1697 | ||
e1e16097 | 1698 | static void io_cqring_add_event(struct io_kiocb *req, long res, long cflags) |
2b188cc1 | 1699 | { |
78e19bbe | 1700 | struct io_ring_ctx *ctx = req->ctx; |
2b188cc1 JA |
1701 | unsigned long flags; |
1702 | ||
1703 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
bcda7baa | 1704 | __io_cqring_fill_event(req, res, cflags); |
2b188cc1 JA |
1705 | io_commit_cqring(ctx); |
1706 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
1707 | ||
8c838788 | 1708 | io_cqring_ev_posted(ctx); |
2b188cc1 JA |
1709 | } |
1710 | ||
229a7b63 | 1711 | static void io_submit_flush_completions(struct io_comp_state *cs) |
bcda7baa | 1712 | { |
229a7b63 JA |
1713 | struct io_ring_ctx *ctx = cs->ctx; |
1714 | ||
1715 | spin_lock_irq(&ctx->completion_lock); | |
1716 | while (!list_empty(&cs->list)) { | |
1717 | struct io_kiocb *req; | |
1718 | ||
3ca405eb PB |
1719 | req = list_first_entry(&cs->list, struct io_kiocb, compl.list); |
1720 | list_del(&req->compl.list); | |
0f7e466b | 1721 | __io_cqring_fill_event(req, req->result, req->compl.cflags); |
216578e5 PB |
1722 | |
1723 | /* | |
1724 | * io_free_req() doesn't care about completion_lock unless one | |
1725 | * of these flags is set. REQ_F_WORK_INITIALIZED is in the list | |
1726 | * because of a potential deadlock with req->work.fs->lock | |
1727 | */ | |
1728 | if (req->flags & (REQ_F_FAIL_LINK|REQ_F_LINK_TIMEOUT | |
1729 | |REQ_F_WORK_INITIALIZED)) { | |
229a7b63 JA |
1730 | spin_unlock_irq(&ctx->completion_lock); |
1731 | io_put_req(req); | |
1732 | spin_lock_irq(&ctx->completion_lock); | |
216578e5 PB |
1733 | } else { |
1734 | io_put_req(req); | |
229a7b63 JA |
1735 | } |
1736 | } | |
1737 | io_commit_cqring(ctx); | |
1738 | spin_unlock_irq(&ctx->completion_lock); | |
1739 | ||
1740 | io_cqring_ev_posted(ctx); | |
1741 | cs->nr = 0; | |
1742 | } | |
1743 | ||
1744 | static void __io_req_complete(struct io_kiocb *req, long res, unsigned cflags, | |
1745 | struct io_comp_state *cs) | |
1746 | { | |
1747 | if (!cs) { | |
1748 | io_cqring_add_event(req, res, cflags); | |
1749 | io_put_req(req); | |
1750 | } else { | |
3ca405eb | 1751 | io_clean_op(req); |
229a7b63 | 1752 | req->result = res; |
0f7e466b | 1753 | req->compl.cflags = cflags; |
3ca405eb | 1754 | list_add_tail(&req->compl.list, &cs->list); |
229a7b63 JA |
1755 | if (++cs->nr >= 32) |
1756 | io_submit_flush_completions(cs); | |
1757 | } | |
e1e16097 JA |
1758 | } |
1759 | ||
1760 | static void io_req_complete(struct io_kiocb *req, long res) | |
bcda7baa | 1761 | { |
229a7b63 | 1762 | __io_req_complete(req, res, 0, NULL); |
bcda7baa JA |
1763 | } |
1764 | ||
0ddf92e8 JA |
1765 | static inline bool io_is_fallback_req(struct io_kiocb *req) |
1766 | { | |
1767 | return req == (struct io_kiocb *) | |
1768 | ((unsigned long) req->ctx->fallback_req & ~1UL); | |
1769 | } | |
1770 | ||
1771 | static struct io_kiocb *io_get_fallback_req(struct io_ring_ctx *ctx) | |
1772 | { | |
1773 | struct io_kiocb *req; | |
1774 | ||
1775 | req = ctx->fallback_req; | |
dd461af6 | 1776 | if (!test_and_set_bit_lock(0, (unsigned long *) &ctx->fallback_req)) |
0ddf92e8 JA |
1777 | return req; |
1778 | ||
1779 | return NULL; | |
1780 | } | |
1781 | ||
0553b8bd PB |
1782 | static struct io_kiocb *io_alloc_req(struct io_ring_ctx *ctx, |
1783 | struct io_submit_state *state) | |
2b188cc1 | 1784 | { |
f6b6c7d6 | 1785 | if (!state->free_reqs) { |
291b2821 | 1786 | gfp_t gfp = GFP_KERNEL | __GFP_NOWARN; |
2579f913 JA |
1787 | size_t sz; |
1788 | int ret; | |
1789 | ||
1790 | sz = min_t(size_t, state->ios_left, ARRAY_SIZE(state->reqs)); | |
fd6fab2c JA |
1791 | ret = kmem_cache_alloc_bulk(req_cachep, gfp, sz, state->reqs); |
1792 | ||
1793 | /* | |
1794 | * Bulk alloc is all-or-nothing. If we fail to get a batch, | |
1795 | * retry single alloc to be on the safe side. | |
1796 | */ | |
1797 | if (unlikely(ret <= 0)) { | |
1798 | state->reqs[0] = kmem_cache_alloc(req_cachep, gfp); | |
1799 | if (!state->reqs[0]) | |
0ddf92e8 | 1800 | goto fallback; |
fd6fab2c JA |
1801 | ret = 1; |
1802 | } | |
291b2821 | 1803 | state->free_reqs = ret; |
2b188cc1 JA |
1804 | } |
1805 | ||
291b2821 PB |
1806 | state->free_reqs--; |
1807 | return state->reqs[state->free_reqs]; | |
0ddf92e8 | 1808 | fallback: |
0553b8bd | 1809 | return io_get_fallback_req(ctx); |
2b188cc1 JA |
1810 | } |
1811 | ||
8da11c19 PB |
1812 | static inline void io_put_file(struct io_kiocb *req, struct file *file, |
1813 | bool fixed) | |
1814 | { | |
1815 | if (fixed) | |
05589553 | 1816 | percpu_ref_put(req->fixed_file_refs); |
8da11c19 PB |
1817 | else |
1818 | fput(file); | |
1819 | } | |
1820 | ||
4edf20f9 | 1821 | static void io_dismantle_req(struct io_kiocb *req) |
2b188cc1 | 1822 | { |
3ca405eb | 1823 | io_clean_op(req); |
929a3af9 | 1824 | |
e8c2bc1f JA |
1825 | if (req->async_data) |
1826 | kfree(req->async_data); | |
8da11c19 PB |
1827 | if (req->file) |
1828 | io_put_file(req, req->file, (req->flags & REQ_F_FIXED_FILE)); | |
fcb323cc | 1829 | |
4edf20f9 | 1830 | io_req_clean_work(req); |
e65ef56d JA |
1831 | } |
1832 | ||
216578e5 | 1833 | static void __io_free_req(struct io_kiocb *req) |
c6ca97b3 | 1834 | { |
0f212204 | 1835 | struct io_uring_task *tctx = req->task->io_uring; |
51a4cc11 | 1836 | struct io_ring_ctx *ctx = req->ctx; |
c6ca97b3 | 1837 | |
216578e5 | 1838 | io_dismantle_req(req); |
c6ca97b3 | 1839 | |
d8a6df10 | 1840 | percpu_counter_dec(&tctx->inflight); |
0f212204 JA |
1841 | if (tctx->in_idle) |
1842 | wake_up(&tctx->wait); | |
e3bc8e9d JA |
1843 | put_task_struct(req->task); |
1844 | ||
b1e50e54 PB |
1845 | if (likely(!io_is_fallback_req(req))) |
1846 | kmem_cache_free(req_cachep, req); | |
1847 | else | |
ecfc5177 PB |
1848 | clear_bit_unlock(0, (unsigned long *) &ctx->fallback_req); |
1849 | percpu_ref_put(&ctx->refs); | |
e65ef56d JA |
1850 | } |
1851 | ||
c9abd7ad | 1852 | static void io_kill_linked_timeout(struct io_kiocb *req) |
2665abfd | 1853 | { |
a197f664 | 1854 | struct io_ring_ctx *ctx = req->ctx; |
7c86ffee | 1855 | struct io_kiocb *link; |
c9abd7ad PB |
1856 | bool cancelled = false; |
1857 | unsigned long flags; | |
7c86ffee | 1858 | |
c9abd7ad PB |
1859 | spin_lock_irqsave(&ctx->completion_lock, flags); |
1860 | link = list_first_entry_or_null(&req->link_list, struct io_kiocb, | |
1861 | link_list); | |
900fad45 PB |
1862 | /* |
1863 | * Can happen if a linked timeout fired and link had been like | |
1864 | * req -> link t-out -> link t-out [-> ...] | |
1865 | */ | |
c9abd7ad PB |
1866 | if (link && (link->flags & REQ_F_LTIMEOUT_ACTIVE)) { |
1867 | struct io_timeout_data *io = link->async_data; | |
1868 | int ret; | |
7c86ffee | 1869 | |
c9abd7ad PB |
1870 | list_del_init(&link->link_list); |
1871 | ret = hrtimer_try_to_cancel(&io->timer); | |
1872 | if (ret != -1) { | |
1873 | io_cqring_fill_event(link, -ECANCELED); | |
1874 | io_commit_cqring(ctx); | |
1875 | cancelled = true; | |
1876 | } | |
1877 | } | |
7c86ffee | 1878 | req->flags &= ~REQ_F_LINK_TIMEOUT; |
216578e5 | 1879 | spin_unlock_irqrestore(&ctx->completion_lock, flags); |
ab0b6451 | 1880 | |
c9abd7ad | 1881 | if (cancelled) { |
7c86ffee | 1882 | io_cqring_ev_posted(ctx); |
c9abd7ad PB |
1883 | io_put_req(link); |
1884 | } | |
7c86ffee PB |
1885 | } |
1886 | ||
9b5f7bd9 | 1887 | static struct io_kiocb *io_req_link_next(struct io_kiocb *req) |
7c86ffee PB |
1888 | { |
1889 | struct io_kiocb *nxt; | |
4d7dd462 | 1890 | |
9e645e11 JA |
1891 | /* |
1892 | * The list should never be empty when we are called here. But could | |
1893 | * potentially happen if the chain is messed up, check to be on the | |
1894 | * safe side. | |
1895 | */ | |
7c86ffee | 1896 | if (unlikely(list_empty(&req->link_list))) |
9b5f7bd9 | 1897 | return NULL; |
2665abfd | 1898 | |
7c86ffee PB |
1899 | nxt = list_first_entry(&req->link_list, struct io_kiocb, link_list); |
1900 | list_del_init(&req->link_list); | |
1901 | if (!list_empty(&nxt->link_list)) | |
1902 | nxt->flags |= REQ_F_LINK_HEAD; | |
9b5f7bd9 | 1903 | return nxt; |
9e645e11 JA |
1904 | } |
1905 | ||
1906 | /* | |
dea3b49c | 1907 | * Called if REQ_F_LINK_HEAD is set, and we fail the head request |
9e645e11 | 1908 | */ |
d148ca4b | 1909 | static void io_fail_links(struct io_kiocb *req) |
9e645e11 | 1910 | { |
2665abfd | 1911 | struct io_ring_ctx *ctx = req->ctx; |
d148ca4b | 1912 | unsigned long flags; |
9e645e11 | 1913 | |
d148ca4b | 1914 | spin_lock_irqsave(&ctx->completion_lock, flags); |
9e645e11 | 1915 | while (!list_empty(&req->link_list)) { |
4493233e PB |
1916 | struct io_kiocb *link = list_first_entry(&req->link_list, |
1917 | struct io_kiocb, link_list); | |
9e645e11 | 1918 | |
4493233e | 1919 | list_del_init(&link->link_list); |
c826bd7a | 1920 | trace_io_uring_fail_link(req, link); |
2665abfd | 1921 | |
7c86ffee | 1922 | io_cqring_fill_event(link, -ECANCELED); |
216578e5 PB |
1923 | |
1924 | /* | |
1925 | * It's ok to free under spinlock as they're not linked anymore, | |
1926 | * but avoid REQ_F_WORK_INITIALIZED because it may deadlock on | |
1927 | * work.fs->lock. | |
1928 | */ | |
1929 | if (link->flags & REQ_F_WORK_INITIALIZED) | |
1930 | io_put_req_deferred(link, 2); | |
1931 | else | |
1932 | io_double_put_req(link); | |
9e645e11 | 1933 | } |
2665abfd JA |
1934 | |
1935 | io_commit_cqring(ctx); | |
216578e5 | 1936 | spin_unlock_irqrestore(&ctx->completion_lock, flags); |
9e645e11 | 1937 | |
2665abfd | 1938 | io_cqring_ev_posted(ctx); |
9e645e11 JA |
1939 | } |
1940 | ||
3fa5e0f3 | 1941 | static struct io_kiocb *__io_req_find_next(struct io_kiocb *req) |
c69f8dbe | 1942 | { |
9b0d911a | 1943 | req->flags &= ~REQ_F_LINK_HEAD; |
7c86ffee PB |
1944 | if (req->flags & REQ_F_LINK_TIMEOUT) |
1945 | io_kill_linked_timeout(req); | |
944e58bf | 1946 | |
9e645e11 JA |
1947 | /* |
1948 | * If LINK is set, we have dependent requests in this chain. If we | |
1949 | * didn't fail this request, queue the first one up, moving any other | |
1950 | * dependencies to the next request. In case of failure, fail the rest | |
1951 | * of the chain. | |
1952 | */ | |
9b5f7bd9 PB |
1953 | if (likely(!(req->flags & REQ_F_FAIL_LINK))) |
1954 | return io_req_link_next(req); | |
1955 | io_fail_links(req); | |
1956 | return NULL; | |
4d7dd462 | 1957 | } |
9e645e11 | 1958 | |
3fa5e0f3 PB |
1959 | static struct io_kiocb *io_req_find_next(struct io_kiocb *req) |
1960 | { | |
1961 | if (likely(!(req->flags & REQ_F_LINK_HEAD))) | |
1962 | return NULL; | |
1963 | return __io_req_find_next(req); | |
1964 | } | |
1965 | ||
87c4311f | 1966 | static int io_req_task_work_add(struct io_kiocb *req, bool twa_signal_ok) |
c2c4c83c JA |
1967 | { |
1968 | struct task_struct *tsk = req->task; | |
1969 | struct io_ring_ctx *ctx = req->ctx; | |
91989c70 JA |
1970 | enum task_work_notify_mode notify; |
1971 | int ret; | |
c2c4c83c | 1972 | |
6200b0ae JA |
1973 | if (tsk->flags & PF_EXITING) |
1974 | return -ESRCH; | |
1975 | ||
c2c4c83c | 1976 | /* |
0ba9c9ed JA |
1977 | * SQPOLL kernel thread doesn't need notification, just a wakeup. For |
1978 | * all other cases, use TWA_SIGNAL unconditionally to ensure we're | |
1979 | * processing task_work. There's no reliable way to tell if TWA_RESUME | |
1980 | * will do the job. | |
c2c4c83c | 1981 | */ |
91989c70 | 1982 | notify = TWA_NONE; |
fd7d6de2 | 1983 | if (!(ctx->flags & IORING_SETUP_SQPOLL) && twa_signal_ok) |
c2c4c83c JA |
1984 | notify = TWA_SIGNAL; |
1985 | ||
87c4311f | 1986 | ret = task_work_add(tsk, &req->task_work, notify); |
c2c4c83c JA |
1987 | if (!ret) |
1988 | wake_up_process(tsk); | |
0ba9c9ed | 1989 | |
c2c4c83c JA |
1990 | return ret; |
1991 | } | |
1992 | ||
c40f6379 JA |
1993 | static void __io_req_task_cancel(struct io_kiocb *req, int error) |
1994 | { | |
1995 | struct io_ring_ctx *ctx = req->ctx; | |
1996 | ||
1997 | spin_lock_irq(&ctx->completion_lock); | |
1998 | io_cqring_fill_event(req, error); | |
1999 | io_commit_cqring(ctx); | |
2000 | spin_unlock_irq(&ctx->completion_lock); | |
2001 | ||
2002 | io_cqring_ev_posted(ctx); | |
2003 | req_set_fail_links(req); | |
2004 | io_double_put_req(req); | |
2005 | } | |
2006 | ||
2007 | static void io_req_task_cancel(struct callback_head *cb) | |
2008 | { | |
2009 | struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work); | |
87ceb6a6 | 2010 | struct io_ring_ctx *ctx = req->ctx; |
c40f6379 JA |
2011 | |
2012 | __io_req_task_cancel(req, -ECANCELED); | |
87ceb6a6 | 2013 | percpu_ref_put(&ctx->refs); |
c40f6379 JA |
2014 | } |
2015 | ||
2016 | static void __io_req_task_submit(struct io_kiocb *req) | |
2017 | { | |
2018 | struct io_ring_ctx *ctx = req->ctx; | |
2019 | ||
c40f6379 JA |
2020 | if (!__io_sq_thread_acquire_mm(ctx)) { |
2021 | mutex_lock(&ctx->uring_lock); | |
c1379e24 | 2022 | __io_queue_sqe(req, NULL); |
c40f6379 JA |
2023 | mutex_unlock(&ctx->uring_lock); |
2024 | } else { | |
2025 | __io_req_task_cancel(req, -EFAULT); | |
2026 | } | |
2027 | } | |
2028 | ||
2029 | static void io_req_task_submit(struct callback_head *cb) | |
2030 | { | |
2031 | struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work); | |
6d816e08 | 2032 | struct io_ring_ctx *ctx = req->ctx; |
c40f6379 JA |
2033 | |
2034 | __io_req_task_submit(req); | |
6d816e08 | 2035 | percpu_ref_put(&ctx->refs); |
c40f6379 JA |
2036 | } |
2037 | ||
2038 | static void io_req_task_queue(struct io_kiocb *req) | |
2039 | { | |
c40f6379 JA |
2040 | int ret; |
2041 | ||
2042 | init_task_work(&req->task_work, io_req_task_submit); | |
6d816e08 | 2043 | percpu_ref_get(&req->ctx->refs); |
c40f6379 | 2044 | |
87c4311f | 2045 | ret = io_req_task_work_add(req, true); |
c40f6379 | 2046 | if (unlikely(ret)) { |
c2c4c83c JA |
2047 | struct task_struct *tsk; |
2048 | ||
c40f6379 JA |
2049 | init_task_work(&req->task_work, io_req_task_cancel); |
2050 | tsk = io_wq_get_task(req->ctx->io_wq); | |
91989c70 | 2051 | task_work_add(tsk, &req->task_work, TWA_NONE); |
c2c4c83c | 2052 | wake_up_process(tsk); |
c40f6379 | 2053 | } |
c40f6379 JA |
2054 | } |
2055 | ||
c3524383 | 2056 | static void io_queue_next(struct io_kiocb *req) |
c69f8dbe | 2057 | { |
9b5f7bd9 | 2058 | struct io_kiocb *nxt = io_req_find_next(req); |
944e58bf PB |
2059 | |
2060 | if (nxt) | |
906a8c3f | 2061 | io_req_task_queue(nxt); |
c69f8dbe JL |
2062 | } |
2063 | ||
c3524383 | 2064 | static void io_free_req(struct io_kiocb *req) |
7a743e22 | 2065 | { |
c3524383 PB |
2066 | io_queue_next(req); |
2067 | __io_free_req(req); | |
2068 | } | |
8766dd51 | 2069 | |
2d6500d4 PB |
2070 | struct req_batch { |
2071 | void *reqs[IO_IOPOLL_BATCH]; | |
2072 | int to_free; | |
7a743e22 | 2073 | |
5af1d13e PB |
2074 | struct task_struct *task; |
2075 | int task_refs; | |
2d6500d4 PB |
2076 | }; |
2077 | ||
5af1d13e PB |
2078 | static inline void io_init_req_batch(struct req_batch *rb) |
2079 | { | |
2080 | rb->to_free = 0; | |
2081 | rb->task_refs = 0; | |
2082 | rb->task = NULL; | |
2083 | } | |
2084 | ||
2d6500d4 PB |
2085 | static void __io_req_free_batch_flush(struct io_ring_ctx *ctx, |
2086 | struct req_batch *rb) | |
2087 | { | |
2088 | kmem_cache_free_bulk(req_cachep, rb->to_free, rb->reqs); | |
2089 | percpu_ref_put_many(&ctx->refs, rb->to_free); | |
2090 | rb->to_free = 0; | |
2091 | } | |
2092 | ||
2093 | static void io_req_free_batch_finish(struct io_ring_ctx *ctx, | |
2094 | struct req_batch *rb) | |
2095 | { | |
2096 | if (rb->to_free) | |
2097 | __io_req_free_batch_flush(ctx, rb); | |
5af1d13e | 2098 | if (rb->task) { |
d8a6df10 JA |
2099 | struct io_uring_task *tctx = rb->task->io_uring; |
2100 | ||
2101 | percpu_counter_sub(&tctx->inflight, rb->task_refs); | |
5af1d13e PB |
2102 | put_task_struct_many(rb->task, rb->task_refs); |
2103 | rb->task = NULL; | |
2104 | } | |
2d6500d4 PB |
2105 | } |
2106 | ||
2107 | static void io_req_free_batch(struct req_batch *rb, struct io_kiocb *req) | |
2108 | { | |
2109 | if (unlikely(io_is_fallback_req(req))) { | |
2110 | io_free_req(req); | |
2111 | return; | |
2112 | } | |
2113 | if (req->flags & REQ_F_LINK_HEAD) | |
2114 | io_queue_next(req); | |
2115 | ||
e3bc8e9d | 2116 | if (req->task != rb->task) { |
0f212204 | 2117 | if (rb->task) { |
d8a6df10 JA |
2118 | struct io_uring_task *tctx = rb->task->io_uring; |
2119 | ||
2120 | percpu_counter_sub(&tctx->inflight, rb->task_refs); | |
e3bc8e9d | 2121 | put_task_struct_many(rb->task, rb->task_refs); |
5af1d13e | 2122 | } |
e3bc8e9d JA |
2123 | rb->task = req->task; |
2124 | rb->task_refs = 0; | |
5af1d13e | 2125 | } |
e3bc8e9d | 2126 | rb->task_refs++; |
5af1d13e | 2127 | |
4edf20f9 | 2128 | io_dismantle_req(req); |
2d6500d4 PB |
2129 | rb->reqs[rb->to_free++] = req; |
2130 | if (unlikely(rb->to_free == ARRAY_SIZE(rb->reqs))) | |
2131 | __io_req_free_batch_flush(req->ctx, rb); | |
7a743e22 PB |
2132 | } |
2133 | ||
ba816ad6 JA |
2134 | /* |
2135 | * Drop reference to request, return next in chain (if there is one) if this | |
2136 | * was the last reference to this request. | |
2137 | */ | |
9b5f7bd9 | 2138 | static struct io_kiocb *io_put_req_find_next(struct io_kiocb *req) |
e65ef56d | 2139 | { |
9b5f7bd9 PB |
2140 | struct io_kiocb *nxt = NULL; |
2141 | ||
2a44f467 | 2142 | if (refcount_dec_and_test(&req->refs)) { |
9b5f7bd9 | 2143 | nxt = io_req_find_next(req); |
4d7dd462 | 2144 | __io_free_req(req); |
2a44f467 | 2145 | } |
9b5f7bd9 | 2146 | return nxt; |
2b188cc1 JA |
2147 | } |
2148 | ||
e65ef56d JA |
2149 | static void io_put_req(struct io_kiocb *req) |
2150 | { | |
2151 | if (refcount_dec_and_test(&req->refs)) | |
2152 | io_free_req(req); | |
2b188cc1 JA |
2153 | } |
2154 | ||
216578e5 PB |
2155 | static void io_put_req_deferred_cb(struct callback_head *cb) |
2156 | { | |
2157 | struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work); | |
2158 | ||
2159 | io_free_req(req); | |
2160 | } | |
2161 | ||
2162 | static void io_free_req_deferred(struct io_kiocb *req) | |
2163 | { | |
2164 | int ret; | |
2165 | ||
2166 | init_task_work(&req->task_work, io_put_req_deferred_cb); | |
2167 | ret = io_req_task_work_add(req, true); | |
2168 | if (unlikely(ret)) { | |
2169 | struct task_struct *tsk; | |
2170 | ||
2171 | tsk = io_wq_get_task(req->ctx->io_wq); | |
91989c70 | 2172 | task_work_add(tsk, &req->task_work, TWA_NONE); |
216578e5 PB |
2173 | wake_up_process(tsk); |
2174 | } | |
2175 | } | |
2176 | ||
2177 | static inline void io_put_req_deferred(struct io_kiocb *req, int refs) | |
2178 | { | |
2179 | if (refcount_sub_and_test(refs, &req->refs)) | |
2180 | io_free_req_deferred(req); | |
2181 | } | |
2182 | ||
f4db7182 | 2183 | static struct io_wq_work *io_steal_work(struct io_kiocb *req) |
7a743e22 | 2184 | { |
6df1db6b | 2185 | struct io_kiocb *nxt; |
f4db7182 | 2186 | |
7a743e22 | 2187 | /* |
f4db7182 PB |
2188 | * A ref is owned by io-wq in which context we're. So, if that's the |
2189 | * last one, it's safe to steal next work. False negatives are Ok, | |
2190 | * it just will be re-punted async in io_put_work() | |
7a743e22 | 2191 | */ |
f4db7182 PB |
2192 | if (refcount_read(&req->refs) != 1) |
2193 | return NULL; | |
7a743e22 | 2194 | |
9b5f7bd9 | 2195 | nxt = io_req_find_next(req); |
6df1db6b | 2196 | return nxt ? &nxt->work : NULL; |
7a743e22 PB |
2197 | } |
2198 | ||
978db57e JA |
2199 | static void io_double_put_req(struct io_kiocb *req) |
2200 | { | |
2201 | /* drop both submit and complete references */ | |
2202 | if (refcount_sub_and_test(2, &req->refs)) | |
2203 | io_free_req(req); | |
2204 | } | |
2205 | ||
1d7bb1d5 | 2206 | static unsigned io_cqring_events(struct io_ring_ctx *ctx, bool noflush) |
a3a0e43f | 2207 | { |
84f97dc2 JA |
2208 | struct io_rings *rings = ctx->rings; |
2209 | ||
ad3eb2c8 JA |
2210 | if (test_bit(0, &ctx->cq_check_overflow)) { |
2211 | /* | |
2212 | * noflush == true is from the waitqueue handler, just ensure | |
2213 | * we wake up the task, and the next invocation will flush the | |
2214 | * entries. We cannot safely to it from here. | |
2215 | */ | |
2216 | if (noflush && !list_empty(&ctx->cq_overflow_list)) | |
2217 | return -1U; | |
1d7bb1d5 | 2218 | |
e6c8aa9a | 2219 | io_cqring_overflow_flush(ctx, false, NULL, NULL); |
ad3eb2c8 | 2220 | } |
1d7bb1d5 | 2221 | |
a3a0e43f JA |
2222 | /* See comment at the top of this file */ |
2223 | smp_rmb(); | |
ad3eb2c8 | 2224 | return ctx->cached_cq_tail - READ_ONCE(rings->cq.head); |
a3a0e43f JA |
2225 | } |
2226 | ||
fb5ccc98 PB |
2227 | static inline unsigned int io_sqring_entries(struct io_ring_ctx *ctx) |
2228 | { | |
2229 | struct io_rings *rings = ctx->rings; | |
2230 | ||
2231 | /* make sure SQ entry isn't read before tail */ | |
2232 | return smp_load_acquire(&rings->sq.tail) - ctx->cached_sq_head; | |
2233 | } | |
2234 | ||
8ff069bf | 2235 | static unsigned int io_put_kbuf(struct io_kiocb *req, struct io_buffer *kbuf) |
e94f141b | 2236 | { |
8ff069bf | 2237 | unsigned int cflags; |
e94f141b | 2238 | |
bcda7baa JA |
2239 | cflags = kbuf->bid << IORING_CQE_BUFFER_SHIFT; |
2240 | cflags |= IORING_CQE_F_BUFFER; | |
0e1b6fe3 | 2241 | req->flags &= ~REQ_F_BUFFER_SELECTED; |
bcda7baa JA |
2242 | kfree(kbuf); |
2243 | return cflags; | |
e94f141b JA |
2244 | } |
2245 | ||
8ff069bf | 2246 | static inline unsigned int io_put_rw_kbuf(struct io_kiocb *req) |
bcda7baa | 2247 | { |
4d954c25 | 2248 | struct io_buffer *kbuf; |
bcda7baa | 2249 | |
4d954c25 | 2250 | kbuf = (struct io_buffer *) (unsigned long) req->rw.addr; |
8ff069bf PB |
2251 | return io_put_kbuf(req, kbuf); |
2252 | } | |
2253 | ||
4c6e277c JA |
2254 | static inline bool io_run_task_work(void) |
2255 | { | |
6200b0ae JA |
2256 | /* |
2257 | * Not safe to run on exiting task, and the task_work handling will | |
2258 | * not add work to such a task. | |
2259 | */ | |
2260 | if (unlikely(current->flags & PF_EXITING)) | |
2261 | return false; | |
4c6e277c JA |
2262 | if (current->task_works) { |
2263 | __set_current_state(TASK_RUNNING); | |
2264 | task_work_run(); | |
2265 | return true; | |
2266 | } | |
2267 | ||
2268 | return false; | |
bcda7baa JA |
2269 | } |
2270 | ||
bbde017a XW |
2271 | static void io_iopoll_queue(struct list_head *again) |
2272 | { | |
2273 | struct io_kiocb *req; | |
2274 | ||
2275 | do { | |
d21ffe7e PB |
2276 | req = list_first_entry(again, struct io_kiocb, inflight_entry); |
2277 | list_del(&req->inflight_entry); | |
81b68a5c | 2278 | __io_complete_rw(req, -EAGAIN, 0, NULL); |
bbde017a XW |
2279 | } while (!list_empty(again)); |
2280 | } | |
2281 | ||
def596e9 JA |
2282 | /* |
2283 | * Find and free completed poll iocbs | |
2284 | */ | |
2285 | static void io_iopoll_complete(struct io_ring_ctx *ctx, unsigned int *nr_events, | |
2286 | struct list_head *done) | |
2287 | { | |
8237e045 | 2288 | struct req_batch rb; |
def596e9 | 2289 | struct io_kiocb *req; |
bbde017a XW |
2290 | LIST_HEAD(again); |
2291 | ||
2292 | /* order with ->result store in io_complete_rw_iopoll() */ | |
2293 | smp_rmb(); | |
def596e9 | 2294 | |
5af1d13e | 2295 | io_init_req_batch(&rb); |
def596e9 | 2296 | while (!list_empty(done)) { |
bcda7baa JA |
2297 | int cflags = 0; |
2298 | ||
d21ffe7e | 2299 | req = list_first_entry(done, struct io_kiocb, inflight_entry); |
bbde017a | 2300 | if (READ_ONCE(req->result) == -EAGAIN) { |
56450c20 | 2301 | req->result = 0; |
bbde017a | 2302 | req->iopoll_completed = 0; |
d21ffe7e | 2303 | list_move_tail(&req->inflight_entry, &again); |
bbde017a XW |
2304 | continue; |
2305 | } | |
d21ffe7e | 2306 | list_del(&req->inflight_entry); |
def596e9 | 2307 | |
bcda7baa | 2308 | if (req->flags & REQ_F_BUFFER_SELECTED) |
8ff069bf | 2309 | cflags = io_put_rw_kbuf(req); |
bcda7baa JA |
2310 | |
2311 | __io_cqring_fill_event(req, req->result, cflags); | |
def596e9 JA |
2312 | (*nr_events)++; |
2313 | ||
c3524383 | 2314 | if (refcount_dec_and_test(&req->refs)) |
2d6500d4 | 2315 | io_req_free_batch(&rb, req); |
def596e9 | 2316 | } |
def596e9 | 2317 | |
09bb8394 | 2318 | io_commit_cqring(ctx); |
32b2244a XW |
2319 | if (ctx->flags & IORING_SETUP_SQPOLL) |
2320 | io_cqring_ev_posted(ctx); | |
2d6500d4 | 2321 | io_req_free_batch_finish(ctx, &rb); |
581f9810 | 2322 | |
bbde017a XW |
2323 | if (!list_empty(&again)) |
2324 | io_iopoll_queue(&again); | |
581f9810 BM |
2325 | } |
2326 | ||
def596e9 JA |
2327 | static int io_do_iopoll(struct io_ring_ctx *ctx, unsigned int *nr_events, |
2328 | long min) | |
2329 | { | |
2330 | struct io_kiocb *req, *tmp; | |
2331 | LIST_HEAD(done); | |
2332 | bool spin; | |
2333 | int ret; | |
2334 | ||
2335 | /* | |
2336 | * Only spin for completions if we don't have multiple devices hanging | |
2337 | * off our complete list, and we're under the requested amount. | |
2338 | */ | |
2339 | spin = !ctx->poll_multi_file && *nr_events < min; | |
2340 | ||
2341 | ret = 0; | |
d21ffe7e | 2342 | list_for_each_entry_safe(req, tmp, &ctx->iopoll_list, inflight_entry) { |
9adbd45d | 2343 | struct kiocb *kiocb = &req->rw.kiocb; |
def596e9 JA |
2344 | |
2345 | /* | |
581f9810 BM |
2346 | * Move completed and retryable entries to our local lists. |
2347 | * If we find a request that requires polling, break out | |
2348 | * and complete those lists first, if we have entries there. | |
def596e9 | 2349 | */ |
65a6543d | 2350 | if (READ_ONCE(req->iopoll_completed)) { |
d21ffe7e | 2351 | list_move_tail(&req->inflight_entry, &done); |
def596e9 JA |
2352 | continue; |
2353 | } | |
2354 | if (!list_empty(&done)) | |
2355 | break; | |
2356 | ||
2357 | ret = kiocb->ki_filp->f_op->iopoll(kiocb, spin); | |
2358 | if (ret < 0) | |
2359 | break; | |
2360 | ||
3aadc23e PB |
2361 | /* iopoll may have completed current req */ |
2362 | if (READ_ONCE(req->iopoll_completed)) | |
d21ffe7e | 2363 | list_move_tail(&req->inflight_entry, &done); |
3aadc23e | 2364 | |
def596e9 JA |
2365 | if (ret && spin) |
2366 | spin = false; | |
2367 | ret = 0; | |
2368 | } | |
2369 | ||
2370 | if (!list_empty(&done)) | |
2371 | io_iopoll_complete(ctx, nr_events, &done); | |
2372 | ||
2373 | return ret; | |
2374 | } | |
2375 | ||
2376 | /* | |
d195a66e | 2377 | * Poll for a minimum of 'min' events. Note that if min == 0 we consider that a |
def596e9 JA |
2378 | * non-spinning poll check - we'll still enter the driver poll loop, but only |
2379 | * as a non-spinning completion check. | |
2380 | */ | |
2381 | static int io_iopoll_getevents(struct io_ring_ctx *ctx, unsigned int *nr_events, | |
2382 | long min) | |
2383 | { | |
540e32a0 | 2384 | while (!list_empty(&ctx->iopoll_list) && !need_resched()) { |
def596e9 JA |
2385 | int ret; |
2386 | ||
2387 | ret = io_do_iopoll(ctx, nr_events, min); | |
2388 | if (ret < 0) | |
2389 | return ret; | |
eba0a4dd | 2390 | if (*nr_events >= min) |
def596e9 JA |
2391 | return 0; |
2392 | } | |
2393 | ||
2394 | return 1; | |
2395 | } | |
2396 | ||
2397 | /* | |
2398 | * We can't just wait for polled events to come to us, we have to actively | |
2399 | * find and complete them. | |
2400 | */ | |
b2edc0a7 | 2401 | static void io_iopoll_try_reap_events(struct io_ring_ctx *ctx) |
def596e9 JA |
2402 | { |
2403 | if (!(ctx->flags & IORING_SETUP_IOPOLL)) | |
2404 | return; | |
2405 | ||
2406 | mutex_lock(&ctx->uring_lock); | |
540e32a0 | 2407 | while (!list_empty(&ctx->iopoll_list)) { |
def596e9 JA |
2408 | unsigned int nr_events = 0; |
2409 | ||
b2edc0a7 | 2410 | io_do_iopoll(ctx, &nr_events, 0); |
08f5439f | 2411 | |
b2edc0a7 PB |
2412 | /* let it sleep and repeat later if can't complete a request */ |
2413 | if (nr_events == 0) | |
2414 | break; | |
08f5439f JA |
2415 | /* |
2416 | * Ensure we allow local-to-the-cpu processing to take place, | |
2417 | * in this case we need to ensure that we reap all events. | |
3fcee5a6 | 2418 | * Also let task_work, etc. to progress by releasing the mutex |
08f5439f | 2419 | */ |
3fcee5a6 PB |
2420 | if (need_resched()) { |
2421 | mutex_unlock(&ctx->uring_lock); | |
2422 | cond_resched(); | |
2423 | mutex_lock(&ctx->uring_lock); | |
2424 | } | |
def596e9 JA |
2425 | } |
2426 | mutex_unlock(&ctx->uring_lock); | |
2427 | } | |
2428 | ||
7668b92a | 2429 | static int io_iopoll_check(struct io_ring_ctx *ctx, long min) |
def596e9 | 2430 | { |
7668b92a | 2431 | unsigned int nr_events = 0; |
2b2ed975 | 2432 | int iters = 0, ret = 0; |
500f9fba | 2433 | |
c7849be9 XW |
2434 | /* |
2435 | * We disallow the app entering submit/complete with polling, but we | |
2436 | * still need to lock the ring to prevent racing with polled issue | |
2437 | * that got punted to a workqueue. | |
2438 | */ | |
2439 | mutex_lock(&ctx->uring_lock); | |
def596e9 | 2440 | do { |
a3a0e43f JA |
2441 | /* |
2442 | * Don't enter poll loop if we already have events pending. | |
2443 | * If we do, we can potentially be spinning for commands that | |
2444 | * already triggered a CQE (eg in error). | |
2445 | */ | |
1d7bb1d5 | 2446 | if (io_cqring_events(ctx, false)) |
a3a0e43f JA |
2447 | break; |
2448 | ||
500f9fba JA |
2449 | /* |
2450 | * If a submit got punted to a workqueue, we can have the | |
2451 | * application entering polling for a command before it gets | |
2452 | * issued. That app will hold the uring_lock for the duration | |
2453 | * of the poll right here, so we need to take a breather every | |
2454 | * now and then to ensure that the issue has a chance to add | |
2455 | * the poll to the issued list. Otherwise we can spin here | |
2456 | * forever, while the workqueue is stuck trying to acquire the | |
2457 | * very same mutex. | |
2458 | */ | |
2459 | if (!(++iters & 7)) { | |
2460 | mutex_unlock(&ctx->uring_lock); | |
4c6e277c | 2461 | io_run_task_work(); |
500f9fba JA |
2462 | mutex_lock(&ctx->uring_lock); |
2463 | } | |
2464 | ||
7668b92a | 2465 | ret = io_iopoll_getevents(ctx, &nr_events, min); |
def596e9 JA |
2466 | if (ret <= 0) |
2467 | break; | |
2468 | ret = 0; | |
7668b92a | 2469 | } while (min && !nr_events && !need_resched()); |
def596e9 | 2470 | |
500f9fba | 2471 | mutex_unlock(&ctx->uring_lock); |
def596e9 JA |
2472 | return ret; |
2473 | } | |
2474 | ||
491381ce | 2475 | static void kiocb_end_write(struct io_kiocb *req) |
2b188cc1 | 2476 | { |
491381ce JA |
2477 | /* |
2478 | * Tell lockdep we inherited freeze protection from submission | |
2479 | * thread. | |
2480 | */ | |
2481 | if (req->flags & REQ_F_ISREG) { | |
2482 | struct inode *inode = file_inode(req->file); | |
2b188cc1 | 2483 | |
491381ce | 2484 | __sb_writers_acquired(inode->i_sb, SB_FREEZE_WRITE); |
2b188cc1 | 2485 | } |
491381ce | 2486 | file_end_write(req->file); |
2b188cc1 JA |
2487 | } |
2488 | ||
a1d7c393 JA |
2489 | static void io_complete_rw_common(struct kiocb *kiocb, long res, |
2490 | struct io_comp_state *cs) | |
2b188cc1 | 2491 | { |
9adbd45d | 2492 | struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb); |
bcda7baa | 2493 | int cflags = 0; |
2b188cc1 | 2494 | |
491381ce JA |
2495 | if (kiocb->ki_flags & IOCB_WRITE) |
2496 | kiocb_end_write(req); | |
2b188cc1 | 2497 | |
4e88d6e7 JA |
2498 | if (res != req->result) |
2499 | req_set_fail_links(req); | |
bcda7baa | 2500 | if (req->flags & REQ_F_BUFFER_SELECTED) |
8ff069bf | 2501 | cflags = io_put_rw_kbuf(req); |
a1d7c393 | 2502 | __io_req_complete(req, res, cflags, cs); |
ba816ad6 JA |
2503 | } |
2504 | ||
b63534c4 JA |
2505 | #ifdef CONFIG_BLOCK |
2506 | static bool io_resubmit_prep(struct io_kiocb *req, int error) | |
2507 | { | |
2508 | struct iovec inline_vecs[UIO_FASTIOV], *iovec = inline_vecs; | |
2509 | ssize_t ret = -ECANCELED; | |
2510 | struct iov_iter iter; | |
2511 | int rw; | |
2512 | ||
2513 | if (error) { | |
2514 | ret = error; | |
2515 | goto end_req; | |
2516 | } | |
2517 | ||
2518 | switch (req->opcode) { | |
2519 | case IORING_OP_READV: | |
2520 | case IORING_OP_READ_FIXED: | |
2521 | case IORING_OP_READ: | |
2522 | rw = READ; | |
2523 | break; | |
2524 | case IORING_OP_WRITEV: | |
2525 | case IORING_OP_WRITE_FIXED: | |
2526 | case IORING_OP_WRITE: | |
2527 | rw = WRITE; | |
2528 | break; | |
2529 | default: | |
2530 | printk_once(KERN_WARNING "io_uring: bad opcode in resubmit %d\n", | |
2531 | req->opcode); | |
2532 | goto end_req; | |
2533 | } | |
2534 | ||
e8c2bc1f | 2535 | if (!req->async_data) { |
8f3d7496 JA |
2536 | ret = io_import_iovec(rw, req, &iovec, &iter, false); |
2537 | if (ret < 0) | |
2538 | goto end_req; | |
2539 | ret = io_setup_async_rw(req, iovec, inline_vecs, &iter, false); | |
2540 | if (!ret) | |
2541 | return true; | |
2542 | kfree(iovec); | |
2543 | } else { | |
b63534c4 | 2544 | return true; |
8f3d7496 | 2545 | } |
b63534c4 | 2546 | end_req: |
b63534c4 | 2547 | req_set_fail_links(req); |
e1e16097 | 2548 | io_req_complete(req, ret); |
b63534c4 JA |
2549 | return false; |
2550 | } | |
b63534c4 JA |
2551 | #endif |
2552 | ||
2553 | static bool io_rw_reissue(struct io_kiocb *req, long res) | |
2554 | { | |
2555 | #ifdef CONFIG_BLOCK | |
355afaeb | 2556 | umode_t mode = file_inode(req->file)->i_mode; |
b63534c4 JA |
2557 | int ret; |
2558 | ||
355afaeb JA |
2559 | if (!S_ISBLK(mode) && !S_ISREG(mode)) |
2560 | return false; | |
b63534c4 JA |
2561 | if ((res != -EAGAIN && res != -EOPNOTSUPP) || io_wq_current_is_worker()) |
2562 | return false; | |
2563 | ||
fdee946d | 2564 | ret = io_sq_thread_acquire_mm(req->ctx, req); |
6d816e08 | 2565 | |
fdee946d JA |
2566 | if (io_resubmit_prep(req, ret)) { |
2567 | refcount_inc(&req->refs); | |
2568 | io_queue_async_work(req); | |
b63534c4 | 2569 | return true; |
fdee946d JA |
2570 | } |
2571 | ||
b63534c4 JA |
2572 | #endif |
2573 | return false; | |
2574 | } | |
2575 | ||
a1d7c393 JA |
2576 | static void __io_complete_rw(struct io_kiocb *req, long res, long res2, |
2577 | struct io_comp_state *cs) | |
2578 | { | |
2579 | if (!io_rw_reissue(req, res)) | |
2580 | io_complete_rw_common(&req->rw.kiocb, res, cs); | |
ba816ad6 JA |
2581 | } |
2582 | ||
2583 | static void io_complete_rw(struct kiocb *kiocb, long res, long res2) | |
2584 | { | |
9adbd45d | 2585 | struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb); |
ba816ad6 | 2586 | |
a1d7c393 | 2587 | __io_complete_rw(req, res, res2, NULL); |
2b188cc1 JA |
2588 | } |
2589 | ||
def596e9 JA |
2590 | static void io_complete_rw_iopoll(struct kiocb *kiocb, long res, long res2) |
2591 | { | |
9adbd45d | 2592 | struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb); |
def596e9 | 2593 | |
491381ce JA |
2594 | if (kiocb->ki_flags & IOCB_WRITE) |
2595 | kiocb_end_write(req); | |
def596e9 | 2596 | |
2d7d6792 | 2597 | if (res != -EAGAIN && res != req->result) |
4e88d6e7 | 2598 | req_set_fail_links(req); |
bbde017a XW |
2599 | |
2600 | WRITE_ONCE(req->result, res); | |
2601 | /* order with io_poll_complete() checking ->result */ | |
cd664b0e PB |
2602 | smp_wmb(); |
2603 | WRITE_ONCE(req->iopoll_completed, 1); | |
def596e9 JA |
2604 | } |
2605 | ||
2606 | /* | |
2607 | * After the iocb has been issued, it's safe to be found on the poll list. | |
2608 | * Adding the kiocb to the list AFTER submission ensures that we don't | |
2609 | * find it from a io_iopoll_getevents() thread before the issuer is done | |
2610 | * accessing the kiocb cookie. | |
2611 | */ | |
2612 | static void io_iopoll_req_issued(struct io_kiocb *req) | |
2613 | { | |
2614 | struct io_ring_ctx *ctx = req->ctx; | |
2615 | ||
2616 | /* | |
2617 | * Track whether we have multiple files in our lists. This will impact | |
2618 | * how we do polling eventually, not spinning if we're on potentially | |
2619 | * different devices. | |
2620 | */ | |
540e32a0 | 2621 | if (list_empty(&ctx->iopoll_list)) { |
def596e9 JA |
2622 | ctx->poll_multi_file = false; |
2623 | } else if (!ctx->poll_multi_file) { | |
2624 | struct io_kiocb *list_req; | |
2625 | ||
540e32a0 | 2626 | list_req = list_first_entry(&ctx->iopoll_list, struct io_kiocb, |
d21ffe7e | 2627 | inflight_entry); |
9adbd45d | 2628 | if (list_req->file != req->file) |
def596e9 JA |
2629 | ctx->poll_multi_file = true; |
2630 | } | |
2631 | ||
2632 | /* | |
2633 | * For fast devices, IO may have already completed. If it has, add | |
2634 | * it to the front so we find it first. | |
2635 | */ | |
65a6543d | 2636 | if (READ_ONCE(req->iopoll_completed)) |
d21ffe7e | 2637 | list_add(&req->inflight_entry, &ctx->iopoll_list); |
def596e9 | 2638 | else |
d21ffe7e | 2639 | list_add_tail(&req->inflight_entry, &ctx->iopoll_list); |
bdcd3eab XW |
2640 | |
2641 | if ((ctx->flags & IORING_SETUP_SQPOLL) && | |
534ca6d6 JA |
2642 | wq_has_sleeper(&ctx->sq_data->wait)) |
2643 | wake_up(&ctx->sq_data->wait); | |
def596e9 JA |
2644 | } |
2645 | ||
9f13c35b | 2646 | static void __io_state_file_put(struct io_submit_state *state) |
9a56a232 | 2647 | { |
06ef3608 PB |
2648 | if (state->has_refs) |
2649 | fput_many(state->file, state->has_refs); | |
9f13c35b PB |
2650 | state->file = NULL; |
2651 | } | |
2652 | ||
2653 | static inline void io_state_file_put(struct io_submit_state *state) | |
2654 | { | |
2655 | if (state->file) | |
2656 | __io_state_file_put(state); | |
9a56a232 JA |
2657 | } |
2658 | ||
2659 | /* | |
2660 | * Get as many references to a file as we have IOs left in this submission, | |
2661 | * assuming most submissions are for one file, or at least that each file | |
2662 | * has more than one submission. | |
2663 | */ | |
8da11c19 | 2664 | static struct file *__io_file_get(struct io_submit_state *state, int fd) |
9a56a232 JA |
2665 | { |
2666 | if (!state) | |
2667 | return fget(fd); | |
2668 | ||
2669 | if (state->file) { | |
2670 | if (state->fd == fd) { | |
06ef3608 | 2671 | state->has_refs--; |
9a56a232 JA |
2672 | return state->file; |
2673 | } | |
9f13c35b | 2674 | __io_state_file_put(state); |
9a56a232 JA |
2675 | } |
2676 | state->file = fget_many(fd, state->ios_left); | |
2677 | if (!state->file) | |
2678 | return NULL; | |
2679 | ||
2680 | state->fd = fd; | |
71b547c0 | 2681 | state->has_refs = state->ios_left - 1; |
9a56a232 JA |
2682 | return state->file; |
2683 | } | |
2684 | ||
4503b767 JA |
2685 | static bool io_bdev_nowait(struct block_device *bdev) |
2686 | { | |
2687 | #ifdef CONFIG_BLOCK | |
9ba0d0c8 | 2688 | return !bdev || blk_queue_nowait(bdev_get_queue(bdev)); |
4503b767 JA |
2689 | #else |
2690 | return true; | |
2691 | #endif | |
2692 | } | |
2693 | ||
2b188cc1 JA |
2694 | /* |
2695 | * If we tracked the file through the SCM inflight mechanism, we could support | |
2696 | * any file. For now, just ensure that anything potentially problematic is done | |
2697 | * inline. | |
2698 | */ | |
af197f50 | 2699 | static bool io_file_supports_async(struct file *file, int rw) |
2b188cc1 JA |
2700 | { |
2701 | umode_t mode = file_inode(file)->i_mode; | |
2702 | ||
4503b767 JA |
2703 | if (S_ISBLK(mode)) { |
2704 | if (io_bdev_nowait(file->f_inode->i_bdev)) | |
2705 | return true; | |
2706 | return false; | |
2707 | } | |
2708 | if (S_ISCHR(mode) || S_ISSOCK(mode)) | |
2b188cc1 | 2709 | return true; |
4503b767 JA |
2710 | if (S_ISREG(mode)) { |
2711 | if (io_bdev_nowait(file->f_inode->i_sb->s_bdev) && | |
2712 | file->f_op != &io_uring_fops) | |
2713 | return true; | |
2714 | return false; | |
2715 | } | |
2b188cc1 | 2716 | |
c5b85625 JA |
2717 | /* any ->read/write should understand O_NONBLOCK */ |
2718 | if (file->f_flags & O_NONBLOCK) | |
2719 | return true; | |
2720 | ||
af197f50 JA |
2721 | if (!(file->f_mode & FMODE_NOWAIT)) |
2722 | return false; | |
2723 | ||
2724 | if (rw == READ) | |
2725 | return file->f_op->read_iter != NULL; | |
2726 | ||
2727 | return file->f_op->write_iter != NULL; | |
2b188cc1 JA |
2728 | } |
2729 | ||
a88fc400 | 2730 | static int io_prep_rw(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
2b188cc1 | 2731 | { |
def596e9 | 2732 | struct io_ring_ctx *ctx = req->ctx; |
9adbd45d | 2733 | struct kiocb *kiocb = &req->rw.kiocb; |
09bb8394 JA |
2734 | unsigned ioprio; |
2735 | int ret; | |
2b188cc1 | 2736 | |
491381ce JA |
2737 | if (S_ISREG(file_inode(req->file)->i_mode)) |
2738 | req->flags |= REQ_F_ISREG; | |
2739 | ||
2b188cc1 | 2740 | kiocb->ki_pos = READ_ONCE(sqe->off); |
ba04291e JA |
2741 | if (kiocb->ki_pos == -1 && !(req->file->f_mode & FMODE_STREAM)) { |
2742 | req->flags |= REQ_F_CUR_POS; | |
2743 | kiocb->ki_pos = req->file->f_pos; | |
2744 | } | |
2b188cc1 | 2745 | kiocb->ki_hint = ki_hint_validate(file_write_hint(kiocb->ki_filp)); |
3e577dcd PB |
2746 | kiocb->ki_flags = iocb_flags(kiocb->ki_filp); |
2747 | ret = kiocb_set_rw_flags(kiocb, READ_ONCE(sqe->rw_flags)); | |
2748 | if (unlikely(ret)) | |
2749 | return ret; | |
2b188cc1 JA |
2750 | |
2751 | ioprio = READ_ONCE(sqe->ioprio); | |
2752 | if (ioprio) { | |
2753 | ret = ioprio_check_cap(ioprio); | |
2754 | if (ret) | |
09bb8394 | 2755 | return ret; |
2b188cc1 JA |
2756 | |
2757 | kiocb->ki_ioprio = ioprio; | |
2758 | } else | |
2759 | kiocb->ki_ioprio = get_current_ioprio(); | |
2760 | ||
8449eeda | 2761 | /* don't allow async punt if RWF_NOWAIT was requested */ |
c5b85625 | 2762 | if (kiocb->ki_flags & IOCB_NOWAIT) |
8449eeda SB |
2763 | req->flags |= REQ_F_NOWAIT; |
2764 | ||
def596e9 | 2765 | if (ctx->flags & IORING_SETUP_IOPOLL) { |
def596e9 JA |
2766 | if (!(kiocb->ki_flags & IOCB_DIRECT) || |
2767 | !kiocb->ki_filp->f_op->iopoll) | |
09bb8394 | 2768 | return -EOPNOTSUPP; |
2b188cc1 | 2769 | |
def596e9 JA |
2770 | kiocb->ki_flags |= IOCB_HIPRI; |
2771 | kiocb->ki_complete = io_complete_rw_iopoll; | |
65a6543d | 2772 | req->iopoll_completed = 0; |
def596e9 | 2773 | } else { |
09bb8394 JA |
2774 | if (kiocb->ki_flags & IOCB_HIPRI) |
2775 | return -EINVAL; | |
def596e9 JA |
2776 | kiocb->ki_complete = io_complete_rw; |
2777 | } | |
9adbd45d | 2778 | |
3529d8c2 JA |
2779 | req->rw.addr = READ_ONCE(sqe->addr); |
2780 | req->rw.len = READ_ONCE(sqe->len); | |
4f4eeba8 | 2781 | req->buf_index = READ_ONCE(sqe->buf_index); |
2b188cc1 | 2782 | return 0; |
2b188cc1 JA |
2783 | } |
2784 | ||
2785 | static inline void io_rw_done(struct kiocb *kiocb, ssize_t ret) | |
2786 | { | |
2787 | switch (ret) { | |
2788 | case -EIOCBQUEUED: | |
2789 | break; | |
2790 | case -ERESTARTSYS: | |
2791 | case -ERESTARTNOINTR: | |
2792 | case -ERESTARTNOHAND: | |
2793 | case -ERESTART_RESTARTBLOCK: | |
2794 | /* | |
2795 | * We can't just restart the syscall, since previously | |
2796 | * submitted sqes may already be in progress. Just fail this | |
2797 | * IO with EINTR. | |
2798 | */ | |
2799 | ret = -EINTR; | |
df561f66 | 2800 | fallthrough; |
2b188cc1 JA |
2801 | default: |
2802 | kiocb->ki_complete(kiocb, ret, 0); | |
2803 | } | |
2804 | } | |
2805 | ||
a1d7c393 JA |
2806 | static void kiocb_done(struct kiocb *kiocb, ssize_t ret, |
2807 | struct io_comp_state *cs) | |
ba816ad6 | 2808 | { |
ba04291e | 2809 | struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb); |
e8c2bc1f | 2810 | struct io_async_rw *io = req->async_data; |
ba04291e | 2811 | |
227c0c96 | 2812 | /* add previously done IO, if any */ |
e8c2bc1f | 2813 | if (io && io->bytes_done > 0) { |
227c0c96 | 2814 | if (ret < 0) |
e8c2bc1f | 2815 | ret = io->bytes_done; |
227c0c96 | 2816 | else |
e8c2bc1f | 2817 | ret += io->bytes_done; |
227c0c96 JA |
2818 | } |
2819 | ||
ba04291e JA |
2820 | if (req->flags & REQ_F_CUR_POS) |
2821 | req->file->f_pos = kiocb->ki_pos; | |
bcaec089 | 2822 | if (ret >= 0 && kiocb->ki_complete == io_complete_rw) |
a1d7c393 | 2823 | __io_complete_rw(req, ret, 0, cs); |
ba816ad6 JA |
2824 | else |
2825 | io_rw_done(kiocb, ret); | |
2826 | } | |
2827 | ||
9adbd45d | 2828 | static ssize_t io_import_fixed(struct io_kiocb *req, int rw, |
7d009165 | 2829 | struct iov_iter *iter) |
edafccee | 2830 | { |
9adbd45d JA |
2831 | struct io_ring_ctx *ctx = req->ctx; |
2832 | size_t len = req->rw.len; | |
edafccee | 2833 | struct io_mapped_ubuf *imu; |
4be1c615 | 2834 | u16 index, buf_index = req->buf_index; |
edafccee JA |
2835 | size_t offset; |
2836 | u64 buf_addr; | |
2837 | ||
edafccee JA |
2838 | if (unlikely(buf_index >= ctx->nr_user_bufs)) |
2839 | return -EFAULT; | |
edafccee JA |
2840 | index = array_index_nospec(buf_index, ctx->nr_user_bufs); |
2841 | imu = &ctx->user_bufs[index]; | |
9adbd45d | 2842 | buf_addr = req->rw.addr; |
edafccee JA |
2843 | |
2844 | /* overflow */ | |
2845 | if (buf_addr + len < buf_addr) | |
2846 | return -EFAULT; | |
2847 | /* not inside the mapped region */ | |
2848 | if (buf_addr < imu->ubuf || buf_addr + len > imu->ubuf + imu->len) | |
2849 | return -EFAULT; | |
2850 | ||
2851 | /* | |
2852 | * May not be a start of buffer, set size appropriately | |
2853 | * and advance us to the beginning. | |
2854 | */ | |
2855 | offset = buf_addr - imu->ubuf; | |
2856 | iov_iter_bvec(iter, rw, imu->bvec, imu->nr_bvecs, offset + len); | |
bd11b3a3 JA |
2857 | |
2858 | if (offset) { | |
2859 | /* | |
2860 | * Don't use iov_iter_advance() here, as it's really slow for | |
2861 | * using the latter parts of a big fixed buffer - it iterates | |
2862 | * over each segment manually. We can cheat a bit here, because | |
2863 | * we know that: | |
2864 | * | |
2865 | * 1) it's a BVEC iter, we set it up | |
2866 | * 2) all bvecs are PAGE_SIZE in size, except potentially the | |
2867 | * first and last bvec | |
2868 | * | |
2869 | * So just find our index, and adjust the iterator afterwards. | |
2870 | * If the offset is within the first bvec (or the whole first | |
2871 | * bvec, just use iov_iter_advance(). This makes it easier | |
2872 | * since we can just skip the first segment, which may not | |
2873 | * be PAGE_SIZE aligned. | |
2874 | */ | |
2875 | const struct bio_vec *bvec = imu->bvec; | |
2876 | ||
2877 | if (offset <= bvec->bv_len) { | |
2878 | iov_iter_advance(iter, offset); | |
2879 | } else { | |
2880 | unsigned long seg_skip; | |
2881 | ||
2882 | /* skip first vec */ | |
2883 | offset -= bvec->bv_len; | |
2884 | seg_skip = 1 + (offset >> PAGE_SHIFT); | |
2885 | ||
2886 | iter->bvec = bvec + seg_skip; | |
2887 | iter->nr_segs -= seg_skip; | |
99c79f66 | 2888 | iter->count -= bvec->bv_len + offset; |
bd11b3a3 | 2889 | iter->iov_offset = offset & ~PAGE_MASK; |
bd11b3a3 JA |
2890 | } |
2891 | } | |
2892 | ||
5e559561 | 2893 | return len; |
edafccee JA |
2894 | } |
2895 | ||
bcda7baa JA |
2896 | static void io_ring_submit_unlock(struct io_ring_ctx *ctx, bool needs_lock) |
2897 | { | |
2898 | if (needs_lock) | |
2899 | mutex_unlock(&ctx->uring_lock); | |
2900 | } | |
2901 | ||
2902 | static void io_ring_submit_lock(struct io_ring_ctx *ctx, bool needs_lock) | |
2903 | { | |
2904 | /* | |
2905 | * "Normal" inline submissions always hold the uring_lock, since we | |
2906 | * grab it from the system call. Same is true for the SQPOLL offload. | |
2907 | * The only exception is when we've detached the request and issue it | |
2908 | * from an async worker thread, grab the lock for that case. | |
2909 | */ | |
2910 | if (needs_lock) | |
2911 | mutex_lock(&ctx->uring_lock); | |
2912 | } | |
2913 | ||
2914 | static struct io_buffer *io_buffer_select(struct io_kiocb *req, size_t *len, | |
2915 | int bgid, struct io_buffer *kbuf, | |
2916 | bool needs_lock) | |
2917 | { | |
2918 | struct io_buffer *head; | |
2919 | ||
2920 | if (req->flags & REQ_F_BUFFER_SELECTED) | |
2921 | return kbuf; | |
2922 | ||
2923 | io_ring_submit_lock(req->ctx, needs_lock); | |
2924 | ||
2925 | lockdep_assert_held(&req->ctx->uring_lock); | |
2926 | ||
2927 | head = idr_find(&req->ctx->io_buffer_idr, bgid); | |
2928 | if (head) { | |
2929 | if (!list_empty(&head->list)) { | |
2930 | kbuf = list_last_entry(&head->list, struct io_buffer, | |
2931 | list); | |
2932 | list_del(&kbuf->list); | |
2933 | } else { | |
2934 | kbuf = head; | |
2935 | idr_remove(&req->ctx->io_buffer_idr, bgid); | |
2936 | } | |
2937 | if (*len > kbuf->len) | |
2938 | *len = kbuf->len; | |
2939 | } else { | |
2940 | kbuf = ERR_PTR(-ENOBUFS); | |
2941 | } | |
2942 | ||
2943 | io_ring_submit_unlock(req->ctx, needs_lock); | |
2944 | ||
2945 | return kbuf; | |
2946 | } | |
2947 | ||
4d954c25 JA |
2948 | static void __user *io_rw_buffer_select(struct io_kiocb *req, size_t *len, |
2949 | bool needs_lock) | |
2950 | { | |
2951 | struct io_buffer *kbuf; | |
4f4eeba8 | 2952 | u16 bgid; |
4d954c25 JA |
2953 | |
2954 | kbuf = (struct io_buffer *) (unsigned long) req->rw.addr; | |
4f4eeba8 | 2955 | bgid = req->buf_index; |
4d954c25 JA |
2956 | kbuf = io_buffer_select(req, len, bgid, kbuf, needs_lock); |
2957 | if (IS_ERR(kbuf)) | |
2958 | return kbuf; | |
2959 | req->rw.addr = (u64) (unsigned long) kbuf; | |
2960 | req->flags |= REQ_F_BUFFER_SELECTED; | |
2961 | return u64_to_user_ptr(kbuf->addr); | |
2962 | } | |
2963 | ||
2964 | #ifdef CONFIG_COMPAT | |
2965 | static ssize_t io_compat_import(struct io_kiocb *req, struct iovec *iov, | |
2966 | bool needs_lock) | |
2967 | { | |
2968 | struct compat_iovec __user *uiov; | |
2969 | compat_ssize_t clen; | |
2970 | void __user *buf; | |
2971 | ssize_t len; | |
2972 | ||
2973 | uiov = u64_to_user_ptr(req->rw.addr); | |
2974 | if (!access_ok(uiov, sizeof(*uiov))) | |
2975 | return -EFAULT; | |
2976 | if (__get_user(clen, &uiov->iov_len)) | |
2977 | return -EFAULT; | |
2978 | if (clen < 0) | |
2979 | return -EINVAL; | |
2980 | ||
2981 | len = clen; | |
2982 | buf = io_rw_buffer_select(req, &len, needs_lock); | |
2983 | if (IS_ERR(buf)) | |
2984 | return PTR_ERR(buf); | |
2985 | iov[0].iov_base = buf; | |
2986 | iov[0].iov_len = (compat_size_t) len; | |
2987 | return 0; | |
2988 | } | |
2989 | #endif | |
2990 | ||
2991 | static ssize_t __io_iov_buffer_select(struct io_kiocb *req, struct iovec *iov, | |
2992 | bool needs_lock) | |
2993 | { | |
2994 | struct iovec __user *uiov = u64_to_user_ptr(req->rw.addr); | |
2995 | void __user *buf; | |
2996 | ssize_t len; | |
2997 | ||
2998 | if (copy_from_user(iov, uiov, sizeof(*uiov))) | |
2999 | return -EFAULT; | |
3000 | ||
3001 | len = iov[0].iov_len; | |
3002 | if (len < 0) | |
3003 | return -EINVAL; | |
3004 | buf = io_rw_buffer_select(req, &len, needs_lock); | |
3005 | if (IS_ERR(buf)) | |
3006 | return PTR_ERR(buf); | |
3007 | iov[0].iov_base = buf; | |
3008 | iov[0].iov_len = len; | |
3009 | return 0; | |
3010 | } | |
3011 | ||
3012 | static ssize_t io_iov_buffer_select(struct io_kiocb *req, struct iovec *iov, | |
3013 | bool needs_lock) | |
3014 | { | |
dddb3e26 JA |
3015 | if (req->flags & REQ_F_BUFFER_SELECTED) { |
3016 | struct io_buffer *kbuf; | |
3017 | ||
3018 | kbuf = (struct io_buffer *) (unsigned long) req->rw.addr; | |
3019 | iov[0].iov_base = u64_to_user_ptr(kbuf->addr); | |
3020 | iov[0].iov_len = kbuf->len; | |
4d954c25 | 3021 | return 0; |
dddb3e26 | 3022 | } |
4d954c25 JA |
3023 | if (!req->rw.len) |
3024 | return 0; | |
3025 | else if (req->rw.len > 1) | |
3026 | return -EINVAL; | |
3027 | ||
3028 | #ifdef CONFIG_COMPAT | |
3029 | if (req->ctx->compat) | |
3030 | return io_compat_import(req, iov, needs_lock); | |
3031 | #endif | |
3032 | ||
3033 | return __io_iov_buffer_select(req, iov, needs_lock); | |
3034 | } | |
3035 | ||
8452fd0c JA |
3036 | static ssize_t __io_import_iovec(int rw, struct io_kiocb *req, |
3037 | struct iovec **iovec, struct iov_iter *iter, | |
3038 | bool needs_lock) | |
2b188cc1 | 3039 | { |
9adbd45d JA |
3040 | void __user *buf = u64_to_user_ptr(req->rw.addr); |
3041 | size_t sqe_len = req->rw.len; | |
4d954c25 | 3042 | ssize_t ret; |
edafccee JA |
3043 | u8 opcode; |
3044 | ||
d625c6ee | 3045 | opcode = req->opcode; |
7d009165 | 3046 | if (opcode == IORING_OP_READ_FIXED || opcode == IORING_OP_WRITE_FIXED) { |
edafccee | 3047 | *iovec = NULL; |
9adbd45d | 3048 | return io_import_fixed(req, rw, iter); |
edafccee | 3049 | } |
2b188cc1 | 3050 | |
bcda7baa | 3051 | /* buffer index only valid with fixed read/write, or buffer select */ |
4f4eeba8 | 3052 | if (req->buf_index && !(req->flags & REQ_F_BUFFER_SELECT)) |
9adbd45d JA |
3053 | return -EINVAL; |
3054 | ||
3a6820f2 | 3055 | if (opcode == IORING_OP_READ || opcode == IORING_OP_WRITE) { |
bcda7baa | 3056 | if (req->flags & REQ_F_BUFFER_SELECT) { |
4d954c25 | 3057 | buf = io_rw_buffer_select(req, &sqe_len, needs_lock); |
867a23ea | 3058 | if (IS_ERR(buf)) |
4d954c25 | 3059 | return PTR_ERR(buf); |
3f9d6441 | 3060 | req->rw.len = sqe_len; |
bcda7baa JA |
3061 | } |
3062 | ||
3a6820f2 JA |
3063 | ret = import_single_range(rw, buf, sqe_len, *iovec, iter); |
3064 | *iovec = NULL; | |
3a901598 | 3065 | return ret < 0 ? ret : sqe_len; |
3a6820f2 JA |
3066 | } |
3067 | ||
4d954c25 JA |
3068 | if (req->flags & REQ_F_BUFFER_SELECT) { |
3069 | ret = io_iov_buffer_select(req, *iovec, needs_lock); | |
3f9d6441 JA |
3070 | if (!ret) { |
3071 | ret = (*iovec)->iov_len; | |
3072 | iov_iter_init(iter, rw, *iovec, 1, ret); | |
3073 | } | |
4d954c25 JA |
3074 | *iovec = NULL; |
3075 | return ret; | |
3076 | } | |
3077 | ||
89cd35c5 CH |
3078 | return __import_iovec(rw, buf, sqe_len, UIO_FASTIOV, iovec, iter, |
3079 | req->ctx->compat); | |
2b188cc1 JA |
3080 | } |
3081 | ||
8452fd0c JA |
3082 | static ssize_t io_import_iovec(int rw, struct io_kiocb *req, |
3083 | struct iovec **iovec, struct iov_iter *iter, | |
3084 | bool needs_lock) | |
3085 | { | |
e8c2bc1f JA |
3086 | struct io_async_rw *iorw = req->async_data; |
3087 | ||
3088 | if (!iorw) | |
8452fd0c JA |
3089 | return __io_import_iovec(rw, req, iovec, iter, needs_lock); |
3090 | *iovec = NULL; | |
e8c2bc1f | 3091 | return iov_iter_count(&iorw->iter); |
8452fd0c JA |
3092 | } |
3093 | ||
0fef9483 JA |
3094 | static inline loff_t *io_kiocb_ppos(struct kiocb *kiocb) |
3095 | { | |
5b09e37e | 3096 | return (kiocb->ki_filp->f_mode & FMODE_STREAM) ? NULL : &kiocb->ki_pos; |
0fef9483 JA |
3097 | } |
3098 | ||
31b51510 | 3099 | /* |
32960613 JA |
3100 | * For files that don't have ->read_iter() and ->write_iter(), handle them |
3101 | * by looping over ->read() or ->write() manually. | |
31b51510 | 3102 | */ |
4017eb91 | 3103 | static ssize_t loop_rw_iter(int rw, struct io_kiocb *req, struct iov_iter *iter) |
32960613 | 3104 | { |
4017eb91 JA |
3105 | struct kiocb *kiocb = &req->rw.kiocb; |
3106 | struct file *file = req->file; | |
32960613 JA |
3107 | ssize_t ret = 0; |
3108 | ||
3109 | /* | |
3110 | * Don't support polled IO through this interface, and we can't | |
3111 | * support non-blocking either. For the latter, this just causes | |
3112 | * the kiocb to be handled from an async context. | |
3113 | */ | |
3114 | if (kiocb->ki_flags & IOCB_HIPRI) | |
3115 | return -EOPNOTSUPP; | |
3116 | if (kiocb->ki_flags & IOCB_NOWAIT) | |
3117 | return -EAGAIN; | |
3118 | ||
3119 | while (iov_iter_count(iter)) { | |
311ae9e1 | 3120 | struct iovec iovec; |
32960613 JA |
3121 | ssize_t nr; |
3122 | ||
311ae9e1 PB |
3123 | if (!iov_iter_is_bvec(iter)) { |
3124 | iovec = iov_iter_iovec(iter); | |
3125 | } else { | |
4017eb91 JA |
3126 | iovec.iov_base = u64_to_user_ptr(req->rw.addr); |
3127 | iovec.iov_len = req->rw.len; | |
311ae9e1 PB |
3128 | } |
3129 | ||
32960613 JA |
3130 | if (rw == READ) { |
3131 | nr = file->f_op->read(file, iovec.iov_base, | |
0fef9483 | 3132 | iovec.iov_len, io_kiocb_ppos(kiocb)); |
32960613 JA |
3133 | } else { |
3134 | nr = file->f_op->write(file, iovec.iov_base, | |
0fef9483 | 3135 | iovec.iov_len, io_kiocb_ppos(kiocb)); |
32960613 JA |
3136 | } |
3137 | ||
3138 | if (nr < 0) { | |
3139 | if (!ret) | |
3140 | ret = nr; | |
3141 | break; | |
3142 | } | |
3143 | ret += nr; | |
3144 | if (nr != iovec.iov_len) | |
3145 | break; | |
4017eb91 JA |
3146 | req->rw.len -= nr; |
3147 | req->rw.addr += nr; | |
32960613 JA |
3148 | iov_iter_advance(iter, nr); |
3149 | } | |
3150 | ||
3151 | return ret; | |
3152 | } | |
3153 | ||
ff6165b2 JA |
3154 | static void io_req_map_rw(struct io_kiocb *req, const struct iovec *iovec, |
3155 | const struct iovec *fast_iov, struct iov_iter *iter) | |
f67676d1 | 3156 | { |
e8c2bc1f | 3157 | struct io_async_rw *rw = req->async_data; |
b64e3444 | 3158 | |
ff6165b2 | 3159 | memcpy(&rw->iter, iter, sizeof(*iter)); |
afb87658 | 3160 | rw->free_iovec = iovec; |
227c0c96 | 3161 | rw->bytes_done = 0; |
ff6165b2 JA |
3162 | /* can only be fixed buffers, no need to do anything */ |
3163 | if (iter->type == ITER_BVEC) | |
3164 | return; | |
b64e3444 | 3165 | if (!iovec) { |
ff6165b2 JA |
3166 | unsigned iov_off = 0; |
3167 | ||
3168 | rw->iter.iov = rw->fast_iov; | |
3169 | if (iter->iov != fast_iov) { | |
3170 | iov_off = iter->iov - fast_iov; | |
3171 | rw->iter.iov += iov_off; | |
3172 | } | |
3173 | if (rw->fast_iov != fast_iov) | |
3174 | memcpy(rw->fast_iov + iov_off, fast_iov + iov_off, | |
45097dae | 3175 | sizeof(struct iovec) * iter->nr_segs); |
99bc4c38 PB |
3176 | } else { |
3177 | req->flags |= REQ_F_NEED_CLEANUP; | |
f67676d1 JA |
3178 | } |
3179 | } | |
3180 | ||
e8c2bc1f | 3181 | static inline int __io_alloc_async_data(struct io_kiocb *req) |
3d9932a8 | 3182 | { |
e8c2bc1f JA |
3183 | WARN_ON_ONCE(!io_op_defs[req->opcode].async_size); |
3184 | req->async_data = kmalloc(io_op_defs[req->opcode].async_size, GFP_KERNEL); | |
3185 | return req->async_data == NULL; | |
3d9932a8 XW |
3186 | } |
3187 | ||
e8c2bc1f | 3188 | static int io_alloc_async_data(struct io_kiocb *req) |
f67676d1 | 3189 | { |
e8c2bc1f | 3190 | if (!io_op_defs[req->opcode].needs_async_data) |
d3656344 | 3191 | return 0; |
3d9932a8 | 3192 | |
e8c2bc1f | 3193 | return __io_alloc_async_data(req); |
b7bb4f7d JA |
3194 | } |
3195 | ||
ff6165b2 JA |
3196 | static int io_setup_async_rw(struct io_kiocb *req, const struct iovec *iovec, |
3197 | const struct iovec *fast_iov, | |
227c0c96 | 3198 | struct iov_iter *iter, bool force) |
b7bb4f7d | 3199 | { |
e8c2bc1f | 3200 | if (!force && !io_op_defs[req->opcode].needs_async_data) |
74566df3 | 3201 | return 0; |
e8c2bc1f JA |
3202 | if (!req->async_data) { |
3203 | if (__io_alloc_async_data(req)) | |
5d204bcf | 3204 | return -ENOMEM; |
b7bb4f7d | 3205 | |
ff6165b2 | 3206 | io_req_map_rw(req, iovec, fast_iov, iter); |
5d204bcf | 3207 | } |
b7bb4f7d | 3208 | return 0; |
f67676d1 JA |
3209 | } |
3210 | ||
73debe68 | 3211 | static inline int io_rw_prep_async(struct io_kiocb *req, int rw) |
c3e330a4 | 3212 | { |
e8c2bc1f | 3213 | struct io_async_rw *iorw = req->async_data; |
f4bff104 | 3214 | struct iovec *iov = iorw->fast_iov; |
c3e330a4 PB |
3215 | ssize_t ret; |
3216 | ||
73debe68 | 3217 | ret = __io_import_iovec(rw, req, &iov, &iorw->iter, false); |
c3e330a4 PB |
3218 | if (unlikely(ret < 0)) |
3219 | return ret; | |
3220 | ||
ab0b196c PB |
3221 | iorw->bytes_done = 0; |
3222 | iorw->free_iovec = iov; | |
3223 | if (iov) | |
3224 | req->flags |= REQ_F_NEED_CLEANUP; | |
c3e330a4 PB |
3225 | return 0; |
3226 | } | |
3227 | ||
73debe68 | 3228 | static int io_read_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
f67676d1 JA |
3229 | { |
3230 | ssize_t ret; | |
3231 | ||
a88fc400 | 3232 | ret = io_prep_rw(req, sqe); |
3529d8c2 JA |
3233 | if (ret) |
3234 | return ret; | |
f67676d1 | 3235 | |
3529d8c2 JA |
3236 | if (unlikely(!(req->file->f_mode & FMODE_READ))) |
3237 | return -EBADF; | |
f67676d1 | 3238 | |
5f798bea | 3239 | /* either don't need iovec imported or already have it */ |
2d199895 | 3240 | if (!req->async_data) |
3529d8c2 | 3241 | return 0; |
73debe68 | 3242 | return io_rw_prep_async(req, READ); |
f67676d1 JA |
3243 | } |
3244 | ||
c1dd91d1 JA |
3245 | /* |
3246 | * This is our waitqueue callback handler, registered through lock_page_async() | |
3247 | * when we initially tried to do the IO with the iocb armed our waitqueue. | |
3248 | * This gets called when the page is unlocked, and we generally expect that to | |
3249 | * happen when the page IO is completed and the page is now uptodate. This will | |
3250 | * queue a task_work based retry of the operation, attempting to copy the data | |
3251 | * again. If the latter fails because the page was NOT uptodate, then we will | |
3252 | * do a thread based blocking retry of the operation. That's the unexpected | |
3253 | * slow path. | |
3254 | */ | |
bcf5a063 JA |
3255 | static int io_async_buf_func(struct wait_queue_entry *wait, unsigned mode, |
3256 | int sync, void *arg) | |
3257 | { | |
3258 | struct wait_page_queue *wpq; | |
3259 | struct io_kiocb *req = wait->private; | |
bcf5a063 | 3260 | struct wait_page_key *key = arg; |
bcf5a063 JA |
3261 | int ret; |
3262 | ||
3263 | wpq = container_of(wait, struct wait_page_queue, wait); | |
3264 | ||
cdc8fcb4 LT |
3265 | if (!wake_page_match(wpq, key)) |
3266 | return 0; | |
3267 | ||
c8d317aa | 3268 | req->rw.kiocb.ki_flags &= ~IOCB_WAITQ; |
bcf5a063 JA |
3269 | list_del_init(&wait->entry); |
3270 | ||
e7375122 | 3271 | init_task_work(&req->task_work, io_req_task_submit); |
6d816e08 JA |
3272 | percpu_ref_get(&req->ctx->refs); |
3273 | ||
bcf5a063 JA |
3274 | /* submit ref gets dropped, acquire a new one */ |
3275 | refcount_inc(&req->refs); | |
87c4311f | 3276 | ret = io_req_task_work_add(req, true); |
bcf5a063 | 3277 | if (unlikely(ret)) { |
c2c4c83c JA |
3278 | struct task_struct *tsk; |
3279 | ||
bcf5a063 | 3280 | /* queue just for cancelation */ |
e7375122 | 3281 | init_task_work(&req->task_work, io_req_task_cancel); |
bcf5a063 | 3282 | tsk = io_wq_get_task(req->ctx->io_wq); |
91989c70 | 3283 | task_work_add(tsk, &req->task_work, TWA_NONE); |
c2c4c83c | 3284 | wake_up_process(tsk); |
bcf5a063 | 3285 | } |
bcf5a063 JA |
3286 | return 1; |
3287 | } | |
3288 | ||
c1dd91d1 JA |
3289 | /* |
3290 | * This controls whether a given IO request should be armed for async page | |
3291 | * based retry. If we return false here, the request is handed to the async | |
3292 | * worker threads for retry. If we're doing buffered reads on a regular file, | |
3293 | * we prepare a private wait_page_queue entry and retry the operation. This | |
3294 | * will either succeed because the page is now uptodate and unlocked, or it | |
3295 | * will register a callback when the page is unlocked at IO completion. Through | |
3296 | * that callback, io_uring uses task_work to setup a retry of the operation. | |
3297 | * That retry will attempt the buffered read again. The retry will generally | |
3298 | * succeed, or in rare cases where it fails, we then fall back to using the | |
3299 | * async worker threads for a blocking retry. | |
3300 | */ | |
227c0c96 | 3301 | static bool io_rw_should_retry(struct io_kiocb *req) |
f67676d1 | 3302 | { |
e8c2bc1f JA |
3303 | struct io_async_rw *rw = req->async_data; |
3304 | struct wait_page_queue *wait = &rw->wpq; | |
bcf5a063 | 3305 | struct kiocb *kiocb = &req->rw.kiocb; |
f67676d1 | 3306 | |
bcf5a063 JA |
3307 | /* never retry for NOWAIT, we just complete with -EAGAIN */ |
3308 | if (req->flags & REQ_F_NOWAIT) | |
3309 | return false; | |
f67676d1 | 3310 | |
227c0c96 | 3311 | /* Only for buffered IO */ |
3b2a4439 | 3312 | if (kiocb->ki_flags & (IOCB_DIRECT | IOCB_HIPRI)) |
bcf5a063 | 3313 | return false; |
3b2a4439 | 3314 | |
bcf5a063 JA |
3315 | /* |
3316 | * just use poll if we can, and don't attempt if the fs doesn't | |
3317 | * support callback based unlocks | |
3318 | */ | |
3319 | if (file_can_poll(req->file) || !(req->file->f_mode & FMODE_BUF_RASYNC)) | |
3320 | return false; | |
f67676d1 | 3321 | |
3b2a4439 JA |
3322 | wait->wait.func = io_async_buf_func; |
3323 | wait->wait.private = req; | |
3324 | wait->wait.flags = 0; | |
3325 | INIT_LIST_HEAD(&wait->wait.entry); | |
3326 | kiocb->ki_flags |= IOCB_WAITQ; | |
c8d317aa | 3327 | kiocb->ki_flags &= ~IOCB_NOWAIT; |
3b2a4439 | 3328 | kiocb->ki_waitq = wait; |
3b2a4439 | 3329 | return true; |
bcf5a063 JA |
3330 | } |
3331 | ||
3332 | static int io_iter_do_read(struct io_kiocb *req, struct iov_iter *iter) | |
3333 | { | |
3334 | if (req->file->f_op->read_iter) | |
3335 | return call_read_iter(req->file, &req->rw.kiocb, iter); | |
2dd2111d | 3336 | else if (req->file->f_op->read) |
4017eb91 | 3337 | return loop_rw_iter(READ, req, iter); |
2dd2111d GH |
3338 | else |
3339 | return -EINVAL; | |
f67676d1 JA |
3340 | } |
3341 | ||
a1d7c393 JA |
3342 | static int io_read(struct io_kiocb *req, bool force_nonblock, |
3343 | struct io_comp_state *cs) | |
2b188cc1 JA |
3344 | { |
3345 | struct iovec inline_vecs[UIO_FASTIOV], *iovec = inline_vecs; | |
9adbd45d | 3346 | struct kiocb *kiocb = &req->rw.kiocb; |
ff6165b2 | 3347 | struct iov_iter __iter, *iter = &__iter; |
e8c2bc1f | 3348 | struct io_async_rw *rw = req->async_data; |
227c0c96 | 3349 | ssize_t io_size, ret, ret2; |
31b51510 | 3350 | size_t iov_count; |
f5cac8b1 | 3351 | bool no_async; |
ff6165b2 | 3352 | |
e8c2bc1f JA |
3353 | if (rw) |
3354 | iter = &rw->iter; | |
2b188cc1 | 3355 | |
ff6165b2 | 3356 | ret = io_import_iovec(READ, req, &iovec, iter, !force_nonblock); |
06b76d44 JA |
3357 | if (ret < 0) |
3358 | return ret; | |
eefdf30f | 3359 | iov_count = iov_iter_count(iter); |
fa15bafb PB |
3360 | io_size = ret; |
3361 | req->result = io_size; | |
227c0c96 | 3362 | ret = 0; |
2b188cc1 | 3363 | |
fd6c2e4c JA |
3364 | /* Ensure we clear previously set non-block flag */ |
3365 | if (!force_nonblock) | |
29de5f6a | 3366 | kiocb->ki_flags &= ~IOCB_NOWAIT; |
a88fc400 PB |
3367 | else |
3368 | kiocb->ki_flags |= IOCB_NOWAIT; | |
3369 | ||
fd6c2e4c | 3370 | |
24c74678 | 3371 | /* If the file doesn't support async, just async punt */ |
f5cac8b1 JA |
3372 | no_async = force_nonblock && !io_file_supports_async(req->file, READ); |
3373 | if (no_async) | |
f67676d1 | 3374 | goto copy_iov; |
9e645e11 | 3375 | |
0fef9483 | 3376 | ret = rw_verify_area(READ, req->file, io_kiocb_ppos(kiocb), iov_count); |
fa15bafb PB |
3377 | if (unlikely(ret)) |
3378 | goto out_free; | |
2b188cc1 | 3379 | |
227c0c96 | 3380 | ret = io_iter_do_read(req, iter); |
32960613 | 3381 | |
227c0c96 JA |
3382 | if (!ret) { |
3383 | goto done; | |
3384 | } else if (ret == -EIOCBQUEUED) { | |
3385 | ret = 0; | |
3386 | goto out_free; | |
3387 | } else if (ret == -EAGAIN) { | |
eefdf30f JA |
3388 | /* IOPOLL retry should happen for io-wq threads */ |
3389 | if (!force_nonblock && !(req->ctx->flags & IORING_SETUP_IOPOLL)) | |
f91daf56 | 3390 | goto done; |
355afaeb JA |
3391 | /* no retry on NONBLOCK marked file */ |
3392 | if (req->file->f_flags & O_NONBLOCK) | |
3393 | goto done; | |
84216315 JA |
3394 | /* some cases will consume bytes even on error returns */ |
3395 | iov_iter_revert(iter, iov_count - iov_iter_count(iter)); | |
f38c7e3a JA |
3396 | ret = 0; |
3397 | goto copy_iov; | |
227c0c96 | 3398 | } else if (ret < 0) { |
00d23d51 JA |
3399 | /* make sure -ERESTARTSYS -> -EINTR is done */ |
3400 | goto done; | |
227c0c96 JA |
3401 | } |
3402 | ||
3403 | /* read it all, or we did blocking attempt. no retry. */ | |
f91daf56 JA |
3404 | if (!iov_iter_count(iter) || !force_nonblock || |
3405 | (req->file->f_flags & O_NONBLOCK)) | |
227c0c96 JA |
3406 | goto done; |
3407 | ||
3408 | io_size -= ret; | |
3409 | copy_iov: | |
3410 | ret2 = io_setup_async_rw(req, iovec, inline_vecs, iter, true); | |
3411 | if (ret2) { | |
3412 | ret = ret2; | |
3413 | goto out_free; | |
3414 | } | |
f5cac8b1 JA |
3415 | if (no_async) |
3416 | return -EAGAIN; | |
e8c2bc1f | 3417 | rw = req->async_data; |
227c0c96 JA |
3418 | /* it's copied and will be cleaned with ->io */ |
3419 | iovec = NULL; | |
3420 | /* now use our persistent iterator, if we aren't already */ | |
e8c2bc1f | 3421 | iter = &rw->iter; |
227c0c96 | 3422 | retry: |
e8c2bc1f | 3423 | rw->bytes_done += ret; |
227c0c96 JA |
3424 | /* if we can retry, do so with the callbacks armed */ |
3425 | if (!io_rw_should_retry(req)) { | |
fa15bafb PB |
3426 | kiocb->ki_flags &= ~IOCB_WAITQ; |
3427 | return -EAGAIN; | |
2b188cc1 | 3428 | } |
227c0c96 JA |
3429 | |
3430 | /* | |
3431 | * Now retry read with the IOCB_WAITQ parts set in the iocb. If we | |
3432 | * get -EIOCBQUEUED, then we'll get a notification when the desired | |
3433 | * page gets unlocked. We can also get a partial read here, and if we | |
3434 | * do, then just retry at the new offset. | |
3435 | */ | |
3436 | ret = io_iter_do_read(req, iter); | |
3437 | if (ret == -EIOCBQUEUED) { | |
3438 | ret = 0; | |
3439 | goto out_free; | |
3440 | } else if (ret > 0 && ret < io_size) { | |
3441 | /* we got some bytes, but not all. retry. */ | |
3442 | goto retry; | |
3443 | } | |
3444 | done: | |
3445 | kiocb_done(kiocb, ret, cs); | |
3446 | ret = 0; | |
f67676d1 | 3447 | out_free: |
f261c168 | 3448 | /* it's reportedly faster than delegating the null check to kfree() */ |
252917c3 | 3449 | if (iovec) |
6f2cc166 | 3450 | kfree(iovec); |
2b188cc1 JA |
3451 | return ret; |
3452 | } | |
3453 | ||
73debe68 | 3454 | static int io_write_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
f67676d1 JA |
3455 | { |
3456 | ssize_t ret; | |
3457 | ||
a88fc400 | 3458 | ret = io_prep_rw(req, sqe); |
3529d8c2 JA |
3459 | if (ret) |
3460 | return ret; | |
f67676d1 | 3461 | |
3529d8c2 JA |
3462 | if (unlikely(!(req->file->f_mode & FMODE_WRITE))) |
3463 | return -EBADF; | |
f67676d1 | 3464 | |
5f798bea | 3465 | /* either don't need iovec imported or already have it */ |
2d199895 | 3466 | if (!req->async_data) |
3529d8c2 | 3467 | return 0; |
73debe68 | 3468 | return io_rw_prep_async(req, WRITE); |
f67676d1 JA |
3469 | } |
3470 | ||
a1d7c393 JA |
3471 | static int io_write(struct io_kiocb *req, bool force_nonblock, |
3472 | struct io_comp_state *cs) | |
2b188cc1 JA |
3473 | { |
3474 | struct iovec inline_vecs[UIO_FASTIOV], *iovec = inline_vecs; | |
9adbd45d | 3475 | struct kiocb *kiocb = &req->rw.kiocb; |
ff6165b2 | 3476 | struct iov_iter __iter, *iter = &__iter; |
e8c2bc1f | 3477 | struct io_async_rw *rw = req->async_data; |
31b51510 | 3478 | size_t iov_count; |
fa15bafb | 3479 | ssize_t ret, ret2, io_size; |
2b188cc1 | 3480 | |
e8c2bc1f JA |
3481 | if (rw) |
3482 | iter = &rw->iter; | |
ff6165b2 JA |
3483 | |
3484 | ret = io_import_iovec(WRITE, req, &iovec, iter, !force_nonblock); | |
06b76d44 JA |
3485 | if (ret < 0) |
3486 | return ret; | |
eefdf30f | 3487 | iov_count = iov_iter_count(iter); |
fa15bafb PB |
3488 | io_size = ret; |
3489 | req->result = io_size; | |
2b188cc1 | 3490 | |
fd6c2e4c JA |
3491 | /* Ensure we clear previously set non-block flag */ |
3492 | if (!force_nonblock) | |
a88fc400 PB |
3493 | kiocb->ki_flags &= ~IOCB_NOWAIT; |
3494 | else | |
3495 | kiocb->ki_flags |= IOCB_NOWAIT; | |
fd6c2e4c | 3496 | |
24c74678 | 3497 | /* If the file doesn't support async, just async punt */ |
af197f50 | 3498 | if (force_nonblock && !io_file_supports_async(req->file, WRITE)) |
f67676d1 | 3499 | goto copy_iov; |
31b51510 | 3500 | |
10d59345 JA |
3501 | /* file path doesn't support NOWAIT for non-direct_IO */ |
3502 | if (force_nonblock && !(kiocb->ki_flags & IOCB_DIRECT) && | |
3503 | (req->flags & REQ_F_ISREG)) | |
f67676d1 | 3504 | goto copy_iov; |
31b51510 | 3505 | |
0fef9483 | 3506 | ret = rw_verify_area(WRITE, req->file, io_kiocb_ppos(kiocb), iov_count); |
fa15bafb PB |
3507 | if (unlikely(ret)) |
3508 | goto out_free; | |
4ed734b0 | 3509 | |
fa15bafb PB |
3510 | /* |
3511 | * Open-code file_start_write here to grab freeze protection, | |
3512 | * which will be released by another thread in | |
3513 | * io_complete_rw(). Fool lockdep by telling it the lock got | |
3514 | * released so that it doesn't complain about the held lock when | |
3515 | * we return to userspace. | |
3516 | */ | |
3517 | if (req->flags & REQ_F_ISREG) { | |
3518 | __sb_start_write(file_inode(req->file)->i_sb, | |
3519 | SB_FREEZE_WRITE, true); | |
3520 | __sb_writers_release(file_inode(req->file)->i_sb, | |
3521 | SB_FREEZE_WRITE); | |
3522 | } | |
3523 | kiocb->ki_flags |= IOCB_WRITE; | |
4ed734b0 | 3524 | |
fa15bafb | 3525 | if (req->file->f_op->write_iter) |
ff6165b2 | 3526 | ret2 = call_write_iter(req->file, kiocb, iter); |
2dd2111d | 3527 | else if (req->file->f_op->write) |
4017eb91 | 3528 | ret2 = loop_rw_iter(WRITE, req, iter); |
2dd2111d GH |
3529 | else |
3530 | ret2 = -EINVAL; | |
4ed734b0 | 3531 | |
fa15bafb PB |
3532 | /* |
3533 | * Raw bdev writes will return -EOPNOTSUPP for IOCB_NOWAIT. Just | |
3534 | * retry them without IOCB_NOWAIT. | |
3535 | */ | |
3536 | if (ret2 == -EOPNOTSUPP && (kiocb->ki_flags & IOCB_NOWAIT)) | |
3537 | ret2 = -EAGAIN; | |
355afaeb JA |
3538 | /* no retry on NONBLOCK marked file */ |
3539 | if (ret2 == -EAGAIN && (req->file->f_flags & O_NONBLOCK)) | |
3540 | goto done; | |
fa15bafb | 3541 | if (!force_nonblock || ret2 != -EAGAIN) { |
eefdf30f JA |
3542 | /* IOPOLL retry should happen for io-wq threads */ |
3543 | if ((req->ctx->flags & IORING_SETUP_IOPOLL) && ret2 == -EAGAIN) | |
3544 | goto copy_iov; | |
355afaeb | 3545 | done: |
fa15bafb PB |
3546 | kiocb_done(kiocb, ret2, cs); |
3547 | } else { | |
f67676d1 | 3548 | copy_iov: |
84216315 JA |
3549 | /* some cases will consume bytes even on error returns */ |
3550 | iov_iter_revert(iter, iov_count - iov_iter_count(iter)); | |
227c0c96 | 3551 | ret = io_setup_async_rw(req, iovec, inline_vecs, iter, false); |
ff6165b2 JA |
3552 | if (!ret) |
3553 | return -EAGAIN; | |
2b188cc1 | 3554 | } |
31b51510 | 3555 | out_free: |
f261c168 | 3556 | /* it's reportedly faster than delegating the null check to kfree() */ |
252917c3 | 3557 | if (iovec) |
6f2cc166 | 3558 | kfree(iovec); |
2b188cc1 JA |
3559 | return ret; |
3560 | } | |
3561 | ||
f2a8d5c7 PB |
3562 | static int __io_splice_prep(struct io_kiocb *req, |
3563 | const struct io_uring_sqe *sqe) | |
7d67af2c PB |
3564 | { |
3565 | struct io_splice* sp = &req->splice; | |
3566 | unsigned int valid_flags = SPLICE_F_FD_IN_FIXED | SPLICE_F_ALL; | |
7d67af2c | 3567 | |
3232dd02 PB |
3568 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
3569 | return -EINVAL; | |
7d67af2c PB |
3570 | |
3571 | sp->file_in = NULL; | |
7d67af2c PB |
3572 | sp->len = READ_ONCE(sqe->len); |
3573 | sp->flags = READ_ONCE(sqe->splice_flags); | |
3574 | ||
3575 | if (unlikely(sp->flags & ~valid_flags)) | |
3576 | return -EINVAL; | |
3577 | ||
8371adf5 PB |
3578 | sp->file_in = io_file_get(NULL, req, READ_ONCE(sqe->splice_fd_in), |
3579 | (sp->flags & SPLICE_F_FD_IN_FIXED)); | |
3580 | if (!sp->file_in) | |
3581 | return -EBADF; | |
7d67af2c PB |
3582 | req->flags |= REQ_F_NEED_CLEANUP; |
3583 | ||
7cdaf587 XW |
3584 | if (!S_ISREG(file_inode(sp->file_in)->i_mode)) { |
3585 | /* | |
3586 | * Splice operation will be punted aync, and here need to | |
3587 | * modify io_wq_work.flags, so initialize io_wq_work firstly. | |
3588 | */ | |
3589 | io_req_init_async(req); | |
7d67af2c | 3590 | req->work.flags |= IO_WQ_WORK_UNBOUND; |
7cdaf587 | 3591 | } |
7d67af2c PB |
3592 | |
3593 | return 0; | |
3594 | } | |
3595 | ||
f2a8d5c7 PB |
3596 | static int io_tee_prep(struct io_kiocb *req, |
3597 | const struct io_uring_sqe *sqe) | |
3598 | { | |
3599 | if (READ_ONCE(sqe->splice_off_in) || READ_ONCE(sqe->off)) | |
3600 | return -EINVAL; | |
3601 | return __io_splice_prep(req, sqe); | |
3602 | } | |
3603 | ||
3604 | static int io_tee(struct io_kiocb *req, bool force_nonblock) | |
3605 | { | |
3606 | struct io_splice *sp = &req->splice; | |
3607 | struct file *in = sp->file_in; | |
3608 | struct file *out = sp->file_out; | |
3609 | unsigned int flags = sp->flags & ~SPLICE_F_FD_IN_FIXED; | |
3610 | long ret = 0; | |
3611 | ||
3612 | if (force_nonblock) | |
3613 | return -EAGAIN; | |
3614 | if (sp->len) | |
3615 | ret = do_tee(in, out, sp->len, flags); | |
3616 | ||
3617 | io_put_file(req, in, (sp->flags & SPLICE_F_FD_IN_FIXED)); | |
3618 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
3619 | ||
f2a8d5c7 PB |
3620 | if (ret != sp->len) |
3621 | req_set_fail_links(req); | |
e1e16097 | 3622 | io_req_complete(req, ret); |
f2a8d5c7 PB |
3623 | return 0; |
3624 | } | |
3625 | ||
3626 | static int io_splice_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) | |
3627 | { | |
3628 | struct io_splice* sp = &req->splice; | |
3629 | ||
3630 | sp->off_in = READ_ONCE(sqe->splice_off_in); | |
3631 | sp->off_out = READ_ONCE(sqe->off); | |
3632 | return __io_splice_prep(req, sqe); | |
3633 | } | |
3634 | ||
014db007 | 3635 | static int io_splice(struct io_kiocb *req, bool force_nonblock) |
7d67af2c PB |
3636 | { |
3637 | struct io_splice *sp = &req->splice; | |
3638 | struct file *in = sp->file_in; | |
3639 | struct file *out = sp->file_out; | |
3640 | unsigned int flags = sp->flags & ~SPLICE_F_FD_IN_FIXED; | |
3641 | loff_t *poff_in, *poff_out; | |
c9687426 | 3642 | long ret = 0; |
7d67af2c | 3643 | |
2fb3e822 PB |
3644 | if (force_nonblock) |
3645 | return -EAGAIN; | |
7d67af2c PB |
3646 | |
3647 | poff_in = (sp->off_in == -1) ? NULL : &sp->off_in; | |
3648 | poff_out = (sp->off_out == -1) ? NULL : &sp->off_out; | |
c9687426 | 3649 | |
948a7749 | 3650 | if (sp->len) |
c9687426 | 3651 | ret = do_splice(in, poff_in, out, poff_out, sp->len, flags); |
7d67af2c PB |
3652 | |
3653 | io_put_file(req, in, (sp->flags & SPLICE_F_FD_IN_FIXED)); | |
3654 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
3655 | ||
7d67af2c PB |
3656 | if (ret != sp->len) |
3657 | req_set_fail_links(req); | |
e1e16097 | 3658 | io_req_complete(req, ret); |
7d67af2c PB |
3659 | return 0; |
3660 | } | |
3661 | ||
2b188cc1 JA |
3662 | /* |
3663 | * IORING_OP_NOP just posts a completion event, nothing else. | |
3664 | */ | |
229a7b63 | 3665 | static int io_nop(struct io_kiocb *req, struct io_comp_state *cs) |
2b188cc1 JA |
3666 | { |
3667 | struct io_ring_ctx *ctx = req->ctx; | |
2b188cc1 | 3668 | |
def596e9 JA |
3669 | if (unlikely(ctx->flags & IORING_SETUP_IOPOLL)) |
3670 | return -EINVAL; | |
3671 | ||
229a7b63 | 3672 | __io_req_complete(req, 0, 0, cs); |
2b188cc1 JA |
3673 | return 0; |
3674 | } | |
3675 | ||
3529d8c2 | 3676 | static int io_prep_fsync(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
c992fe29 | 3677 | { |
6b06314c | 3678 | struct io_ring_ctx *ctx = req->ctx; |
c992fe29 | 3679 | |
09bb8394 JA |
3680 | if (!req->file) |
3681 | return -EBADF; | |
c992fe29 | 3682 | |
6b06314c | 3683 | if (unlikely(ctx->flags & IORING_SETUP_IOPOLL)) |
def596e9 | 3684 | return -EINVAL; |
edafccee | 3685 | if (unlikely(sqe->addr || sqe->ioprio || sqe->buf_index)) |
c992fe29 CH |
3686 | return -EINVAL; |
3687 | ||
8ed8d3c3 JA |
3688 | req->sync.flags = READ_ONCE(sqe->fsync_flags); |
3689 | if (unlikely(req->sync.flags & ~IORING_FSYNC_DATASYNC)) | |
3690 | return -EINVAL; | |
3691 | ||
3692 | req->sync.off = READ_ONCE(sqe->off); | |
3693 | req->sync.len = READ_ONCE(sqe->len); | |
c992fe29 CH |
3694 | return 0; |
3695 | } | |
3696 | ||
ac45abc0 | 3697 | static int io_fsync(struct io_kiocb *req, bool force_nonblock) |
8ed8d3c3 | 3698 | { |
8ed8d3c3 | 3699 | loff_t end = req->sync.off + req->sync.len; |
8ed8d3c3 JA |
3700 | int ret; |
3701 | ||
ac45abc0 PB |
3702 | /* fsync always requires a blocking context */ |
3703 | if (force_nonblock) | |
3704 | return -EAGAIN; | |
3705 | ||
9adbd45d | 3706 | ret = vfs_fsync_range(req->file, req->sync.off, |
8ed8d3c3 JA |
3707 | end > 0 ? end : LLONG_MAX, |
3708 | req->sync.flags & IORING_FSYNC_DATASYNC); | |
3709 | if (ret < 0) | |
3710 | req_set_fail_links(req); | |
e1e16097 | 3711 | io_req_complete(req, ret); |
c992fe29 CH |
3712 | return 0; |
3713 | } | |
3714 | ||
d63d1b5e JA |
3715 | static int io_fallocate_prep(struct io_kiocb *req, |
3716 | const struct io_uring_sqe *sqe) | |
3717 | { | |
3718 | if (sqe->ioprio || sqe->buf_index || sqe->rw_flags) | |
3719 | return -EINVAL; | |
3232dd02 PB |
3720 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
3721 | return -EINVAL; | |
d63d1b5e JA |
3722 | |
3723 | req->sync.off = READ_ONCE(sqe->off); | |
3724 | req->sync.len = READ_ONCE(sqe->addr); | |
3725 | req->sync.mode = READ_ONCE(sqe->len); | |
3726 | return 0; | |
3727 | } | |
3728 | ||
014db007 | 3729 | static int io_fallocate(struct io_kiocb *req, bool force_nonblock) |
5d17b4a4 | 3730 | { |
ac45abc0 PB |
3731 | int ret; |
3732 | ||
d63d1b5e | 3733 | /* fallocate always requiring blocking context */ |
ac45abc0 | 3734 | if (force_nonblock) |
5d17b4a4 | 3735 | return -EAGAIN; |
ac45abc0 PB |
3736 | ret = vfs_fallocate(req->file, req->sync.mode, req->sync.off, |
3737 | req->sync.len); | |
ac45abc0 PB |
3738 | if (ret < 0) |
3739 | req_set_fail_links(req); | |
e1e16097 | 3740 | io_req_complete(req, ret); |
5d17b4a4 JA |
3741 | return 0; |
3742 | } | |
3743 | ||
ec65fea5 | 3744 | static int __io_openat_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
b7bb4f7d | 3745 | { |
f8748881 | 3746 | const char __user *fname; |
15b71abe | 3747 | int ret; |
b7bb4f7d | 3748 | |
ec65fea5 | 3749 | if (unlikely(sqe->ioprio || sqe->buf_index)) |
15b71abe | 3750 | return -EINVAL; |
ec65fea5 | 3751 | if (unlikely(req->flags & REQ_F_FIXED_FILE)) |
cf3040ca | 3752 | return -EBADF; |
03b1230c | 3753 | |
ec65fea5 PB |
3754 | /* open.how should be already initialised */ |
3755 | if (!(req->open.how.flags & O_PATH) && force_o_largefile()) | |
08a1d26e | 3756 | req->open.how.flags |= O_LARGEFILE; |
3529d8c2 | 3757 | |
25e72d10 PB |
3758 | req->open.dfd = READ_ONCE(sqe->fd); |
3759 | fname = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
f8748881 | 3760 | req->open.filename = getname(fname); |
15b71abe JA |
3761 | if (IS_ERR(req->open.filename)) { |
3762 | ret = PTR_ERR(req->open.filename); | |
3763 | req->open.filename = NULL; | |
3764 | return ret; | |
3765 | } | |
4022e7af | 3766 | req->open.nofile = rlimit(RLIMIT_NOFILE); |
8fef80bf | 3767 | req->flags |= REQ_F_NEED_CLEANUP; |
15b71abe | 3768 | return 0; |
03b1230c JA |
3769 | } |
3770 | ||
ec65fea5 PB |
3771 | static int io_openat_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
3772 | { | |
3773 | u64 flags, mode; | |
3774 | ||
4eb8dded JA |
3775 | if (unlikely(req->ctx->flags & (IORING_SETUP_IOPOLL|IORING_SETUP_SQPOLL))) |
3776 | return -EINVAL; | |
ec65fea5 PB |
3777 | mode = READ_ONCE(sqe->len); |
3778 | flags = READ_ONCE(sqe->open_flags); | |
3779 | req->open.how = build_open_how(flags, mode); | |
3780 | return __io_openat_prep(req, sqe); | |
3781 | } | |
3782 | ||
cebdb986 | 3783 | static int io_openat2_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
aa1fa28f | 3784 | { |
cebdb986 | 3785 | struct open_how __user *how; |
cebdb986 | 3786 | size_t len; |
0fa03c62 JA |
3787 | int ret; |
3788 | ||
4eb8dded JA |
3789 | if (unlikely(req->ctx->flags & (IORING_SETUP_IOPOLL|IORING_SETUP_SQPOLL))) |
3790 | return -EINVAL; | |
cebdb986 JA |
3791 | how = u64_to_user_ptr(READ_ONCE(sqe->addr2)); |
3792 | len = READ_ONCE(sqe->len); | |
cebdb986 JA |
3793 | if (len < OPEN_HOW_SIZE_VER0) |
3794 | return -EINVAL; | |
3529d8c2 | 3795 | |
cebdb986 JA |
3796 | ret = copy_struct_from_user(&req->open.how, sizeof(req->open.how), how, |
3797 | len); | |
3798 | if (ret) | |
3799 | return ret; | |
3529d8c2 | 3800 | |
ec65fea5 | 3801 | return __io_openat_prep(req, sqe); |
cebdb986 JA |
3802 | } |
3803 | ||
014db007 | 3804 | static int io_openat2(struct io_kiocb *req, bool force_nonblock) |
15b71abe JA |
3805 | { |
3806 | struct open_flags op; | |
15b71abe JA |
3807 | struct file *file; |
3808 | int ret; | |
3809 | ||
f86cd20c | 3810 | if (force_nonblock) |
15b71abe | 3811 | return -EAGAIN; |
15b71abe | 3812 | |
cebdb986 | 3813 | ret = build_open_flags(&req->open.how, &op); |
15b71abe JA |
3814 | if (ret) |
3815 | goto err; | |
3816 | ||
4022e7af | 3817 | ret = __get_unused_fd_flags(req->open.how.flags, req->open.nofile); |
15b71abe JA |
3818 | if (ret < 0) |
3819 | goto err; | |
3820 | ||
3821 | file = do_filp_open(req->open.dfd, req->open.filename, &op); | |
3822 | if (IS_ERR(file)) { | |
3823 | put_unused_fd(ret); | |
3824 | ret = PTR_ERR(file); | |
3825 | } else { | |
3826 | fsnotify_open(file); | |
3827 | fd_install(ret, file); | |
3828 | } | |
3829 | err: | |
3830 | putname(req->open.filename); | |
8fef80bf | 3831 | req->flags &= ~REQ_F_NEED_CLEANUP; |
15b71abe JA |
3832 | if (ret < 0) |
3833 | req_set_fail_links(req); | |
e1e16097 | 3834 | io_req_complete(req, ret); |
15b71abe JA |
3835 | return 0; |
3836 | } | |
3837 | ||
014db007 | 3838 | static int io_openat(struct io_kiocb *req, bool force_nonblock) |
cebdb986 | 3839 | { |
014db007 | 3840 | return io_openat2(req, force_nonblock); |
cebdb986 JA |
3841 | } |
3842 | ||
067524e9 JA |
3843 | static int io_remove_buffers_prep(struct io_kiocb *req, |
3844 | const struct io_uring_sqe *sqe) | |
3845 | { | |
3846 | struct io_provide_buf *p = &req->pbuf; | |
3847 | u64 tmp; | |
3848 | ||
3849 | if (sqe->ioprio || sqe->rw_flags || sqe->addr || sqe->len || sqe->off) | |
3850 | return -EINVAL; | |
3851 | ||
3852 | tmp = READ_ONCE(sqe->fd); | |
3853 | if (!tmp || tmp > USHRT_MAX) | |
3854 | return -EINVAL; | |
3855 | ||
3856 | memset(p, 0, sizeof(*p)); | |
3857 | p->nbufs = tmp; | |
3858 | p->bgid = READ_ONCE(sqe->buf_group); | |
3859 | return 0; | |
3860 | } | |
3861 | ||
3862 | static int __io_remove_buffers(struct io_ring_ctx *ctx, struct io_buffer *buf, | |
3863 | int bgid, unsigned nbufs) | |
3864 | { | |
3865 | unsigned i = 0; | |
3866 | ||
3867 | /* shouldn't happen */ | |
3868 | if (!nbufs) | |
3869 | return 0; | |
3870 | ||
3871 | /* the head kbuf is the list itself */ | |
3872 | while (!list_empty(&buf->list)) { | |
3873 | struct io_buffer *nxt; | |
3874 | ||
3875 | nxt = list_first_entry(&buf->list, struct io_buffer, list); | |
3876 | list_del(&nxt->list); | |
3877 | kfree(nxt); | |
3878 | if (++i == nbufs) | |
3879 | return i; | |
3880 | } | |
3881 | i++; | |
3882 | kfree(buf); | |
3883 | idr_remove(&ctx->io_buffer_idr, bgid); | |
3884 | ||
3885 | return i; | |
3886 | } | |
3887 | ||
229a7b63 JA |
3888 | static int io_remove_buffers(struct io_kiocb *req, bool force_nonblock, |
3889 | struct io_comp_state *cs) | |
067524e9 JA |
3890 | { |
3891 | struct io_provide_buf *p = &req->pbuf; | |
3892 | struct io_ring_ctx *ctx = req->ctx; | |
3893 | struct io_buffer *head; | |
3894 | int ret = 0; | |
3895 | ||
3896 | io_ring_submit_lock(ctx, !force_nonblock); | |
3897 | ||
3898 | lockdep_assert_held(&ctx->uring_lock); | |
3899 | ||
3900 | ret = -ENOENT; | |
3901 | head = idr_find(&ctx->io_buffer_idr, p->bgid); | |
3902 | if (head) | |
3903 | ret = __io_remove_buffers(ctx, head, p->bgid, p->nbufs); | |
3904 | ||
3905 | io_ring_submit_lock(ctx, !force_nonblock); | |
3906 | if (ret < 0) | |
3907 | req_set_fail_links(req); | |
229a7b63 | 3908 | __io_req_complete(req, ret, 0, cs); |
067524e9 JA |
3909 | return 0; |
3910 | } | |
3911 | ||
ddf0322d JA |
3912 | static int io_provide_buffers_prep(struct io_kiocb *req, |
3913 | const struct io_uring_sqe *sqe) | |
3914 | { | |
3915 | struct io_provide_buf *p = &req->pbuf; | |
3916 | u64 tmp; | |
3917 | ||
3918 | if (sqe->ioprio || sqe->rw_flags) | |
3919 | return -EINVAL; | |
3920 | ||
3921 | tmp = READ_ONCE(sqe->fd); | |
3922 | if (!tmp || tmp > USHRT_MAX) | |
3923 | return -E2BIG; | |
3924 | p->nbufs = tmp; | |
3925 | p->addr = READ_ONCE(sqe->addr); | |
3926 | p->len = READ_ONCE(sqe->len); | |
3927 | ||
efe68c1c | 3928 | if (!access_ok(u64_to_user_ptr(p->addr), (p->len * p->nbufs))) |
ddf0322d JA |
3929 | return -EFAULT; |
3930 | ||
3931 | p->bgid = READ_ONCE(sqe->buf_group); | |
3932 | tmp = READ_ONCE(sqe->off); | |
3933 | if (tmp > USHRT_MAX) | |
3934 | return -E2BIG; | |
3935 | p->bid = tmp; | |
3936 | return 0; | |
3937 | } | |
3938 | ||
3939 | static int io_add_buffers(struct io_provide_buf *pbuf, struct io_buffer **head) | |
3940 | { | |
3941 | struct io_buffer *buf; | |
3942 | u64 addr = pbuf->addr; | |
3943 | int i, bid = pbuf->bid; | |
3944 | ||
3945 | for (i = 0; i < pbuf->nbufs; i++) { | |
3946 | buf = kmalloc(sizeof(*buf), GFP_KERNEL); | |
3947 | if (!buf) | |
3948 | break; | |
3949 | ||
3950 | buf->addr = addr; | |
3951 | buf->len = pbuf->len; | |
3952 | buf->bid = bid; | |
3953 | addr += pbuf->len; | |
3954 | bid++; | |
3955 | if (!*head) { | |
3956 | INIT_LIST_HEAD(&buf->list); | |
3957 | *head = buf; | |
3958 | } else { | |
3959 | list_add_tail(&buf->list, &(*head)->list); | |
3960 | } | |
3961 | } | |
3962 | ||
3963 | return i ? i : -ENOMEM; | |
3964 | } | |
3965 | ||
229a7b63 JA |
3966 | static int io_provide_buffers(struct io_kiocb *req, bool force_nonblock, |
3967 | struct io_comp_state *cs) | |
ddf0322d JA |
3968 | { |
3969 | struct io_provide_buf *p = &req->pbuf; | |
3970 | struct io_ring_ctx *ctx = req->ctx; | |
3971 | struct io_buffer *head, *list; | |
3972 | int ret = 0; | |
3973 | ||
3974 | io_ring_submit_lock(ctx, !force_nonblock); | |
3975 | ||
3976 | lockdep_assert_held(&ctx->uring_lock); | |
3977 | ||
3978 | list = head = idr_find(&ctx->io_buffer_idr, p->bgid); | |
3979 | ||
3980 | ret = io_add_buffers(p, &head); | |
3981 | if (ret < 0) | |
3982 | goto out; | |
3983 | ||
3984 | if (!list) { | |
3985 | ret = idr_alloc(&ctx->io_buffer_idr, head, p->bgid, p->bgid + 1, | |
3986 | GFP_KERNEL); | |
3987 | if (ret < 0) { | |
067524e9 | 3988 | __io_remove_buffers(ctx, head, p->bgid, -1U); |
ddf0322d JA |
3989 | goto out; |
3990 | } | |
3991 | } | |
3992 | out: | |
3993 | io_ring_submit_unlock(ctx, !force_nonblock); | |
3994 | if (ret < 0) | |
3995 | req_set_fail_links(req); | |
229a7b63 | 3996 | __io_req_complete(req, ret, 0, cs); |
ddf0322d | 3997 | return 0; |
cebdb986 JA |
3998 | } |
3999 | ||
3e4827b0 JA |
4000 | static int io_epoll_ctl_prep(struct io_kiocb *req, |
4001 | const struct io_uring_sqe *sqe) | |
4002 | { | |
4003 | #if defined(CONFIG_EPOLL) | |
4004 | if (sqe->ioprio || sqe->buf_index) | |
4005 | return -EINVAL; | |
6ca56f84 | 4006 | if (unlikely(req->ctx->flags & (IORING_SETUP_IOPOLL | IORING_SETUP_SQPOLL))) |
3232dd02 | 4007 | return -EINVAL; |
3e4827b0 JA |
4008 | |
4009 | req->epoll.epfd = READ_ONCE(sqe->fd); | |
4010 | req->epoll.op = READ_ONCE(sqe->len); | |
4011 | req->epoll.fd = READ_ONCE(sqe->off); | |
4012 | ||
4013 | if (ep_op_has_event(req->epoll.op)) { | |
4014 | struct epoll_event __user *ev; | |
4015 | ||
4016 | ev = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
4017 | if (copy_from_user(&req->epoll.event, ev, sizeof(*ev))) | |
4018 | return -EFAULT; | |
4019 | } | |
4020 | ||
4021 | return 0; | |
4022 | #else | |
4023 | return -EOPNOTSUPP; | |
4024 | #endif | |
4025 | } | |
4026 | ||
229a7b63 JA |
4027 | static int io_epoll_ctl(struct io_kiocb *req, bool force_nonblock, |
4028 | struct io_comp_state *cs) | |
3e4827b0 JA |
4029 | { |
4030 | #if defined(CONFIG_EPOLL) | |
4031 | struct io_epoll *ie = &req->epoll; | |
4032 | int ret; | |
4033 | ||
4034 | ret = do_epoll_ctl(ie->epfd, ie->op, ie->fd, &ie->event, force_nonblock); | |
4035 | if (force_nonblock && ret == -EAGAIN) | |
4036 | return -EAGAIN; | |
4037 | ||
4038 | if (ret < 0) | |
4039 | req_set_fail_links(req); | |
229a7b63 | 4040 | __io_req_complete(req, ret, 0, cs); |
3e4827b0 JA |
4041 | return 0; |
4042 | #else | |
4043 | return -EOPNOTSUPP; | |
4044 | #endif | |
4045 | } | |
4046 | ||
c1ca757b JA |
4047 | static int io_madvise_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
4048 | { | |
4049 | #if defined(CONFIG_ADVISE_SYSCALLS) && defined(CONFIG_MMU) | |
4050 | if (sqe->ioprio || sqe->buf_index || sqe->off) | |
4051 | return -EINVAL; | |
3232dd02 PB |
4052 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4053 | return -EINVAL; | |
c1ca757b JA |
4054 | |
4055 | req->madvise.addr = READ_ONCE(sqe->addr); | |
4056 | req->madvise.len = READ_ONCE(sqe->len); | |
4057 | req->madvise.advice = READ_ONCE(sqe->fadvise_advice); | |
4058 | return 0; | |
4059 | #else | |
4060 | return -EOPNOTSUPP; | |
4061 | #endif | |
4062 | } | |
4063 | ||
014db007 | 4064 | static int io_madvise(struct io_kiocb *req, bool force_nonblock) |
c1ca757b JA |
4065 | { |
4066 | #if defined(CONFIG_ADVISE_SYSCALLS) && defined(CONFIG_MMU) | |
4067 | struct io_madvise *ma = &req->madvise; | |
4068 | int ret; | |
4069 | ||
4070 | if (force_nonblock) | |
4071 | return -EAGAIN; | |
4072 | ||
0726b01e | 4073 | ret = do_madvise(current->mm, ma->addr, ma->len, ma->advice); |
c1ca757b JA |
4074 | if (ret < 0) |
4075 | req_set_fail_links(req); | |
e1e16097 | 4076 | io_req_complete(req, ret); |
c1ca757b JA |
4077 | return 0; |
4078 | #else | |
4079 | return -EOPNOTSUPP; | |
4080 | #endif | |
4081 | } | |
4082 | ||
4840e418 JA |
4083 | static int io_fadvise_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
4084 | { | |
4085 | if (sqe->ioprio || sqe->buf_index || sqe->addr) | |
4086 | return -EINVAL; | |
3232dd02 PB |
4087 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4088 | return -EINVAL; | |
4840e418 JA |
4089 | |
4090 | req->fadvise.offset = READ_ONCE(sqe->off); | |
4091 | req->fadvise.len = READ_ONCE(sqe->len); | |
4092 | req->fadvise.advice = READ_ONCE(sqe->fadvise_advice); | |
4093 | return 0; | |
4094 | } | |
4095 | ||
014db007 | 4096 | static int io_fadvise(struct io_kiocb *req, bool force_nonblock) |
4840e418 JA |
4097 | { |
4098 | struct io_fadvise *fa = &req->fadvise; | |
4099 | int ret; | |
4100 | ||
3e69426d JA |
4101 | if (force_nonblock) { |
4102 | switch (fa->advice) { | |
4103 | case POSIX_FADV_NORMAL: | |
4104 | case POSIX_FADV_RANDOM: | |
4105 | case POSIX_FADV_SEQUENTIAL: | |
4106 | break; | |
4107 | default: | |
4108 | return -EAGAIN; | |
4109 | } | |
4110 | } | |
4840e418 JA |
4111 | |
4112 | ret = vfs_fadvise(req->file, fa->offset, fa->len, fa->advice); | |
4113 | if (ret < 0) | |
4114 | req_set_fail_links(req); | |
e1e16097 | 4115 | io_req_complete(req, ret); |
4840e418 JA |
4116 | return 0; |
4117 | } | |
4118 | ||
eddc7ef5 JA |
4119 | static int io_statx_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
4120 | { | |
6ca56f84 | 4121 | if (unlikely(req->ctx->flags & (IORING_SETUP_IOPOLL | IORING_SETUP_SQPOLL))) |
3232dd02 | 4122 | return -EINVAL; |
eddc7ef5 JA |
4123 | if (sqe->ioprio || sqe->buf_index) |
4124 | return -EINVAL; | |
9c280f90 | 4125 | if (req->flags & REQ_F_FIXED_FILE) |
cf3040ca | 4126 | return -EBADF; |
eddc7ef5 | 4127 | |
1d9e1288 BM |
4128 | req->statx.dfd = READ_ONCE(sqe->fd); |
4129 | req->statx.mask = READ_ONCE(sqe->len); | |
e62753e4 | 4130 | req->statx.filename = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
1d9e1288 BM |
4131 | req->statx.buffer = u64_to_user_ptr(READ_ONCE(sqe->addr2)); |
4132 | req->statx.flags = READ_ONCE(sqe->statx_flags); | |
eddc7ef5 JA |
4133 | |
4134 | return 0; | |
4135 | } | |
4136 | ||
014db007 | 4137 | static int io_statx(struct io_kiocb *req, bool force_nonblock) |
eddc7ef5 | 4138 | { |
1d9e1288 | 4139 | struct io_statx *ctx = &req->statx; |
eddc7ef5 JA |
4140 | int ret; |
4141 | ||
5b0bbee4 JA |
4142 | if (force_nonblock) { |
4143 | /* only need file table for an actual valid fd */ | |
4144 | if (ctx->dfd == -1 || ctx->dfd == AT_FDCWD) | |
4145 | req->flags |= REQ_F_NO_FILE_TABLE; | |
eddc7ef5 | 4146 | return -EAGAIN; |
5b0bbee4 | 4147 | } |
eddc7ef5 | 4148 | |
e62753e4 BM |
4149 | ret = do_statx(ctx->dfd, ctx->filename, ctx->flags, ctx->mask, |
4150 | ctx->buffer); | |
eddc7ef5 | 4151 | |
eddc7ef5 JA |
4152 | if (ret < 0) |
4153 | req_set_fail_links(req); | |
e1e16097 | 4154 | io_req_complete(req, ret); |
eddc7ef5 JA |
4155 | return 0; |
4156 | } | |
4157 | ||
b5dba59e JA |
4158 | static int io_close_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
4159 | { | |
4160 | /* | |
4161 | * If we queue this for async, it must not be cancellable. That would | |
7cdaf587 XW |
4162 | * leave the 'file' in an undeterminate state, and here need to modify |
4163 | * io_wq_work.flags, so initialize io_wq_work firstly. | |
b5dba59e | 4164 | */ |
7cdaf587 | 4165 | io_req_init_async(req); |
b5dba59e JA |
4166 | req->work.flags |= IO_WQ_WORK_NO_CANCEL; |
4167 | ||
3232dd02 PB |
4168 | if (unlikely(req->ctx->flags & (IORING_SETUP_IOPOLL|IORING_SETUP_SQPOLL))) |
4169 | return -EINVAL; | |
b5dba59e JA |
4170 | if (sqe->ioprio || sqe->off || sqe->addr || sqe->len || |
4171 | sqe->rw_flags || sqe->buf_index) | |
4172 | return -EINVAL; | |
9c280f90 | 4173 | if (req->flags & REQ_F_FIXED_FILE) |
cf3040ca | 4174 | return -EBADF; |
b5dba59e JA |
4175 | |
4176 | req->close.fd = READ_ONCE(sqe->fd); | |
0f212204 | 4177 | if ((req->file && req->file->f_op == &io_uring_fops)) |
fd2206e4 | 4178 | return -EBADF; |
b5dba59e | 4179 | |
3af73b28 | 4180 | req->close.put_file = NULL; |
b5dba59e | 4181 | return 0; |
b5dba59e JA |
4182 | } |
4183 | ||
229a7b63 JA |
4184 | static int io_close(struct io_kiocb *req, bool force_nonblock, |
4185 | struct io_comp_state *cs) | |
b5dba59e | 4186 | { |
3af73b28 | 4187 | struct io_close *close = &req->close; |
b5dba59e JA |
4188 | int ret; |
4189 | ||
3af73b28 PB |
4190 | /* might be already done during nonblock submission */ |
4191 | if (!close->put_file) { | |
4192 | ret = __close_fd_get_file(close->fd, &close->put_file); | |
4193 | if (ret < 0) | |
4194 | return (ret == -ENOENT) ? -EBADF : ret; | |
4195 | } | |
b5dba59e JA |
4196 | |
4197 | /* if the file has a flush method, be safe and punt to async */ | |
3af73b28 | 4198 | if (close->put_file->f_op->flush && force_nonblock) { |
24c74678 PB |
4199 | /* was never set, but play safe */ |
4200 | req->flags &= ~REQ_F_NOWAIT; | |
0bf0eefd | 4201 | /* avoid grabbing files - we don't need the files */ |
24c74678 | 4202 | req->flags |= REQ_F_NO_FILE_TABLE; |
0bf0eefd | 4203 | return -EAGAIN; |
a2100672 | 4204 | } |
b5dba59e | 4205 | |
3af73b28 | 4206 | /* No ->flush() or already async, safely close from here */ |
98447d65 | 4207 | ret = filp_close(close->put_file, req->work.identity->files); |
3af73b28 PB |
4208 | if (ret < 0) |
4209 | req_set_fail_links(req); | |
3af73b28 PB |
4210 | fput(close->put_file); |
4211 | close->put_file = NULL; | |
229a7b63 | 4212 | __io_req_complete(req, ret, 0, cs); |
1a417f4e | 4213 | return 0; |
b5dba59e JA |
4214 | } |
4215 | ||
3529d8c2 | 4216 | static int io_prep_sfr(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
5d17b4a4 JA |
4217 | { |
4218 | struct io_ring_ctx *ctx = req->ctx; | |
5d17b4a4 JA |
4219 | |
4220 | if (!req->file) | |
4221 | return -EBADF; | |
5d17b4a4 JA |
4222 | |
4223 | if (unlikely(ctx->flags & IORING_SETUP_IOPOLL)) | |
4224 | return -EINVAL; | |
4225 | if (unlikely(sqe->addr || sqe->ioprio || sqe->buf_index)) | |
4226 | return -EINVAL; | |
4227 | ||
8ed8d3c3 JA |
4228 | req->sync.off = READ_ONCE(sqe->off); |
4229 | req->sync.len = READ_ONCE(sqe->len); | |
4230 | req->sync.flags = READ_ONCE(sqe->sync_range_flags); | |
8ed8d3c3 JA |
4231 | return 0; |
4232 | } | |
4233 | ||
ac45abc0 | 4234 | static int io_sync_file_range(struct io_kiocb *req, bool force_nonblock) |
8ed8d3c3 | 4235 | { |
8ed8d3c3 JA |
4236 | int ret; |
4237 | ||
ac45abc0 PB |
4238 | /* sync_file_range always requires a blocking context */ |
4239 | if (force_nonblock) | |
4240 | return -EAGAIN; | |
4241 | ||
9adbd45d | 4242 | ret = sync_file_range(req->file, req->sync.off, req->sync.len, |
8ed8d3c3 JA |
4243 | req->sync.flags); |
4244 | if (ret < 0) | |
4245 | req_set_fail_links(req); | |
e1e16097 | 4246 | io_req_complete(req, ret); |
5d17b4a4 JA |
4247 | return 0; |
4248 | } | |
4249 | ||
469956e8 | 4250 | #if defined(CONFIG_NET) |
02d27d89 PB |
4251 | static int io_setup_async_msg(struct io_kiocb *req, |
4252 | struct io_async_msghdr *kmsg) | |
4253 | { | |
e8c2bc1f JA |
4254 | struct io_async_msghdr *async_msg = req->async_data; |
4255 | ||
4256 | if (async_msg) | |
02d27d89 | 4257 | return -EAGAIN; |
e8c2bc1f | 4258 | if (io_alloc_async_data(req)) { |
02d27d89 PB |
4259 | if (kmsg->iov != kmsg->fast_iov) |
4260 | kfree(kmsg->iov); | |
4261 | return -ENOMEM; | |
4262 | } | |
e8c2bc1f | 4263 | async_msg = req->async_data; |
02d27d89 | 4264 | req->flags |= REQ_F_NEED_CLEANUP; |
e8c2bc1f | 4265 | memcpy(async_msg, kmsg, sizeof(*kmsg)); |
02d27d89 PB |
4266 | return -EAGAIN; |
4267 | } | |
4268 | ||
2ae523ed PB |
4269 | static int io_sendmsg_copy_hdr(struct io_kiocb *req, |
4270 | struct io_async_msghdr *iomsg) | |
4271 | { | |
4272 | iomsg->iov = iomsg->fast_iov; | |
4273 | iomsg->msg.msg_name = &iomsg->addr; | |
4274 | return sendmsg_copy_msghdr(&iomsg->msg, req->sr_msg.umsg, | |
4275 | req->sr_msg.msg_flags, &iomsg->iov); | |
4276 | } | |
4277 | ||
3529d8c2 | 4278 | static int io_sendmsg_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
03b1230c | 4279 | { |
e8c2bc1f | 4280 | struct io_async_msghdr *async_msg = req->async_data; |
e47293fd | 4281 | struct io_sr_msg *sr = &req->sr_msg; |
99bc4c38 | 4282 | int ret; |
03b1230c | 4283 | |
d2b6f48b PB |
4284 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4285 | return -EINVAL; | |
4286 | ||
e47293fd | 4287 | sr->msg_flags = READ_ONCE(sqe->msg_flags); |
270a5940 | 4288 | sr->umsg = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
fddaface | 4289 | sr->len = READ_ONCE(sqe->len); |
3529d8c2 | 4290 | |
d8768362 JA |
4291 | #ifdef CONFIG_COMPAT |
4292 | if (req->ctx->compat) | |
4293 | sr->msg_flags |= MSG_CMSG_COMPAT; | |
4294 | #endif | |
4295 | ||
e8c2bc1f | 4296 | if (!async_msg || !io_op_defs[req->opcode].needs_async_data) |
3529d8c2 | 4297 | return 0; |
e8c2bc1f | 4298 | ret = io_sendmsg_copy_hdr(req, async_msg); |
99bc4c38 PB |
4299 | if (!ret) |
4300 | req->flags |= REQ_F_NEED_CLEANUP; | |
4301 | return ret; | |
03b1230c JA |
4302 | } |
4303 | ||
229a7b63 JA |
4304 | static int io_sendmsg(struct io_kiocb *req, bool force_nonblock, |
4305 | struct io_comp_state *cs) | |
aa1fa28f | 4306 | { |
6b754c8b | 4307 | struct io_async_msghdr iomsg, *kmsg; |
0fa03c62 | 4308 | struct socket *sock; |
7a7cacba | 4309 | unsigned flags; |
0fa03c62 JA |
4310 | int ret; |
4311 | ||
0fa03c62 | 4312 | sock = sock_from_file(req->file, &ret); |
7a7cacba PB |
4313 | if (unlikely(!sock)) |
4314 | return ret; | |
3529d8c2 | 4315 | |
e8c2bc1f JA |
4316 | if (req->async_data) { |
4317 | kmsg = req->async_data; | |
4318 | kmsg->msg.msg_name = &kmsg->addr; | |
7a7cacba PB |
4319 | /* if iov is set, it's allocated already */ |
4320 | if (!kmsg->iov) | |
4321 | kmsg->iov = kmsg->fast_iov; | |
4322 | kmsg->msg.msg_iter.iov = kmsg->iov; | |
4323 | } else { | |
4324 | ret = io_sendmsg_copy_hdr(req, &iomsg); | |
4325 | if (ret) | |
4326 | return ret; | |
4327 | kmsg = &iomsg; | |
0fa03c62 | 4328 | } |
0fa03c62 | 4329 | |
7a7cacba PB |
4330 | flags = req->sr_msg.msg_flags; |
4331 | if (flags & MSG_DONTWAIT) | |
4332 | req->flags |= REQ_F_NOWAIT; | |
4333 | else if (force_nonblock) | |
4334 | flags |= MSG_DONTWAIT; | |
e47293fd | 4335 | |
7a7cacba PB |
4336 | ret = __sys_sendmsg_sock(sock, &kmsg->msg, flags); |
4337 | if (force_nonblock && ret == -EAGAIN) | |
4338 | return io_setup_async_msg(req, kmsg); | |
4339 | if (ret == -ERESTARTSYS) | |
4340 | ret = -EINTR; | |
0fa03c62 | 4341 | |
6b754c8b | 4342 | if (kmsg->iov != kmsg->fast_iov) |
0b416c3e | 4343 | kfree(kmsg->iov); |
99bc4c38 | 4344 | req->flags &= ~REQ_F_NEED_CLEANUP; |
4e88d6e7 JA |
4345 | if (ret < 0) |
4346 | req_set_fail_links(req); | |
229a7b63 | 4347 | __io_req_complete(req, ret, 0, cs); |
5d17b4a4 | 4348 | return 0; |
03b1230c | 4349 | } |
aa1fa28f | 4350 | |
229a7b63 JA |
4351 | static int io_send(struct io_kiocb *req, bool force_nonblock, |
4352 | struct io_comp_state *cs) | |
fddaface | 4353 | { |
7a7cacba PB |
4354 | struct io_sr_msg *sr = &req->sr_msg; |
4355 | struct msghdr msg; | |
4356 | struct iovec iov; | |
fddaface | 4357 | struct socket *sock; |
7a7cacba | 4358 | unsigned flags; |
fddaface JA |
4359 | int ret; |
4360 | ||
fddaface | 4361 | sock = sock_from_file(req->file, &ret); |
7a7cacba PB |
4362 | if (unlikely(!sock)) |
4363 | return ret; | |
fddaface | 4364 | |
7a7cacba PB |
4365 | ret = import_single_range(WRITE, sr->buf, sr->len, &iov, &msg.msg_iter); |
4366 | if (unlikely(ret)) | |
14db8411 | 4367 | return ret; |
fddaface | 4368 | |
7a7cacba PB |
4369 | msg.msg_name = NULL; |
4370 | msg.msg_control = NULL; | |
4371 | msg.msg_controllen = 0; | |
4372 | msg.msg_namelen = 0; | |
fddaface | 4373 | |
7a7cacba PB |
4374 | flags = req->sr_msg.msg_flags; |
4375 | if (flags & MSG_DONTWAIT) | |
4376 | req->flags |= REQ_F_NOWAIT; | |
4377 | else if (force_nonblock) | |
4378 | flags |= MSG_DONTWAIT; | |
fddaface | 4379 | |
7a7cacba PB |
4380 | msg.msg_flags = flags; |
4381 | ret = sock_sendmsg(sock, &msg); | |
4382 | if (force_nonblock && ret == -EAGAIN) | |
4383 | return -EAGAIN; | |
4384 | if (ret == -ERESTARTSYS) | |
4385 | ret = -EINTR; | |
fddaface | 4386 | |
fddaface JA |
4387 | if (ret < 0) |
4388 | req_set_fail_links(req); | |
229a7b63 | 4389 | __io_req_complete(req, ret, 0, cs); |
fddaface | 4390 | return 0; |
fddaface JA |
4391 | } |
4392 | ||
1400e697 PB |
4393 | static int __io_recvmsg_copy_hdr(struct io_kiocb *req, |
4394 | struct io_async_msghdr *iomsg) | |
52de1fe1 JA |
4395 | { |
4396 | struct io_sr_msg *sr = &req->sr_msg; | |
4397 | struct iovec __user *uiov; | |
4398 | size_t iov_len; | |
4399 | int ret; | |
4400 | ||
1400e697 PB |
4401 | ret = __copy_msghdr_from_user(&iomsg->msg, sr->umsg, |
4402 | &iomsg->uaddr, &uiov, &iov_len); | |
52de1fe1 JA |
4403 | if (ret) |
4404 | return ret; | |
4405 | ||
4406 | if (req->flags & REQ_F_BUFFER_SELECT) { | |
4407 | if (iov_len > 1) | |
4408 | return -EINVAL; | |
1400e697 | 4409 | if (copy_from_user(iomsg->iov, uiov, sizeof(*uiov))) |
52de1fe1 | 4410 | return -EFAULT; |
1400e697 PB |
4411 | sr->len = iomsg->iov[0].iov_len; |
4412 | iov_iter_init(&iomsg->msg.msg_iter, READ, iomsg->iov, 1, | |
52de1fe1 | 4413 | sr->len); |
1400e697 | 4414 | iomsg->iov = NULL; |
52de1fe1 | 4415 | } else { |
89cd35c5 CH |
4416 | ret = __import_iovec(READ, uiov, iov_len, UIO_FASTIOV, |
4417 | &iomsg->iov, &iomsg->msg.msg_iter, | |
4418 | false); | |
52de1fe1 JA |
4419 | if (ret > 0) |
4420 | ret = 0; | |
4421 | } | |
4422 | ||
4423 | return ret; | |
4424 | } | |
4425 | ||
4426 | #ifdef CONFIG_COMPAT | |
4427 | static int __io_compat_recvmsg_copy_hdr(struct io_kiocb *req, | |
1400e697 | 4428 | struct io_async_msghdr *iomsg) |
52de1fe1 JA |
4429 | { |
4430 | struct compat_msghdr __user *msg_compat; | |
4431 | struct io_sr_msg *sr = &req->sr_msg; | |
4432 | struct compat_iovec __user *uiov; | |
4433 | compat_uptr_t ptr; | |
4434 | compat_size_t len; | |
4435 | int ret; | |
4436 | ||
270a5940 | 4437 | msg_compat = (struct compat_msghdr __user *) sr->umsg; |
1400e697 | 4438 | ret = __get_compat_msghdr(&iomsg->msg, msg_compat, &iomsg->uaddr, |
52de1fe1 JA |
4439 | &ptr, &len); |
4440 | if (ret) | |
4441 | return ret; | |
4442 | ||
4443 | uiov = compat_ptr(ptr); | |
4444 | if (req->flags & REQ_F_BUFFER_SELECT) { | |
4445 | compat_ssize_t clen; | |
4446 | ||
4447 | if (len > 1) | |
4448 | return -EINVAL; | |
4449 | if (!access_ok(uiov, sizeof(*uiov))) | |
4450 | return -EFAULT; | |
4451 | if (__get_user(clen, &uiov->iov_len)) | |
4452 | return -EFAULT; | |
4453 | if (clen < 0) | |
4454 | return -EINVAL; | |
1400e697 PB |
4455 | sr->len = iomsg->iov[0].iov_len; |
4456 | iomsg->iov = NULL; | |
52de1fe1 | 4457 | } else { |
89cd35c5 CH |
4458 | ret = __import_iovec(READ, (struct iovec __user *)uiov, len, |
4459 | UIO_FASTIOV, &iomsg->iov, | |
4460 | &iomsg->msg.msg_iter, true); | |
52de1fe1 JA |
4461 | if (ret < 0) |
4462 | return ret; | |
4463 | } | |
4464 | ||
4465 | return 0; | |
4466 | } | |
4467 | #endif | |
4468 | ||
1400e697 PB |
4469 | static int io_recvmsg_copy_hdr(struct io_kiocb *req, |
4470 | struct io_async_msghdr *iomsg) | |
52de1fe1 | 4471 | { |
1400e697 PB |
4472 | iomsg->msg.msg_name = &iomsg->addr; |
4473 | iomsg->iov = iomsg->fast_iov; | |
52de1fe1 JA |
4474 | |
4475 | #ifdef CONFIG_COMPAT | |
4476 | if (req->ctx->compat) | |
1400e697 | 4477 | return __io_compat_recvmsg_copy_hdr(req, iomsg); |
fddaface | 4478 | #endif |
52de1fe1 | 4479 | |
1400e697 | 4480 | return __io_recvmsg_copy_hdr(req, iomsg); |
52de1fe1 JA |
4481 | } |
4482 | ||
bcda7baa | 4483 | static struct io_buffer *io_recv_buffer_select(struct io_kiocb *req, |
7fbb1b54 | 4484 | bool needs_lock) |
bcda7baa JA |
4485 | { |
4486 | struct io_sr_msg *sr = &req->sr_msg; | |
4487 | struct io_buffer *kbuf; | |
4488 | ||
bcda7baa JA |
4489 | kbuf = io_buffer_select(req, &sr->len, sr->bgid, sr->kbuf, needs_lock); |
4490 | if (IS_ERR(kbuf)) | |
4491 | return kbuf; | |
4492 | ||
4493 | sr->kbuf = kbuf; | |
4494 | req->flags |= REQ_F_BUFFER_SELECTED; | |
bcda7baa | 4495 | return kbuf; |
fddaface JA |
4496 | } |
4497 | ||
7fbb1b54 PB |
4498 | static inline unsigned int io_put_recv_kbuf(struct io_kiocb *req) |
4499 | { | |
4500 | return io_put_kbuf(req, req->sr_msg.kbuf); | |
4501 | } | |
4502 | ||
3529d8c2 JA |
4503 | static int io_recvmsg_prep(struct io_kiocb *req, |
4504 | const struct io_uring_sqe *sqe) | |
aa1fa28f | 4505 | { |
e8c2bc1f | 4506 | struct io_async_msghdr *async_msg = req->async_data; |
e47293fd | 4507 | struct io_sr_msg *sr = &req->sr_msg; |
99bc4c38 | 4508 | int ret; |
3529d8c2 | 4509 | |
d2b6f48b PB |
4510 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4511 | return -EINVAL; | |
4512 | ||
3529d8c2 | 4513 | sr->msg_flags = READ_ONCE(sqe->msg_flags); |
270a5940 | 4514 | sr->umsg = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
0b7b21e4 | 4515 | sr->len = READ_ONCE(sqe->len); |
bcda7baa | 4516 | sr->bgid = READ_ONCE(sqe->buf_group); |
06b76d44 | 4517 | |
d8768362 JA |
4518 | #ifdef CONFIG_COMPAT |
4519 | if (req->ctx->compat) | |
4520 | sr->msg_flags |= MSG_CMSG_COMPAT; | |
4521 | #endif | |
4522 | ||
e8c2bc1f | 4523 | if (!async_msg || !io_op_defs[req->opcode].needs_async_data) |
5f798bea | 4524 | return 0; |
e8c2bc1f | 4525 | ret = io_recvmsg_copy_hdr(req, async_msg); |
99bc4c38 PB |
4526 | if (!ret) |
4527 | req->flags |= REQ_F_NEED_CLEANUP; | |
4528 | return ret; | |
aa1fa28f JA |
4529 | } |
4530 | ||
229a7b63 JA |
4531 | static int io_recvmsg(struct io_kiocb *req, bool force_nonblock, |
4532 | struct io_comp_state *cs) | |
aa1fa28f | 4533 | { |
6b754c8b | 4534 | struct io_async_msghdr iomsg, *kmsg; |
03b1230c | 4535 | struct socket *sock; |
7fbb1b54 | 4536 | struct io_buffer *kbuf; |
7a7cacba | 4537 | unsigned flags; |
52de1fe1 | 4538 | int ret, cflags = 0; |
03b1230c | 4539 | |
03b1230c | 4540 | sock = sock_from_file(req->file, &ret); |
7a7cacba PB |
4541 | if (unlikely(!sock)) |
4542 | return ret; | |
3529d8c2 | 4543 | |
e8c2bc1f JA |
4544 | if (req->async_data) { |
4545 | kmsg = req->async_data; | |
4546 | kmsg->msg.msg_name = &kmsg->addr; | |
7a7cacba PB |
4547 | /* if iov is set, it's allocated already */ |
4548 | if (!kmsg->iov) | |
4549 | kmsg->iov = kmsg->fast_iov; | |
4550 | kmsg->msg.msg_iter.iov = kmsg->iov; | |
4551 | } else { | |
4552 | ret = io_recvmsg_copy_hdr(req, &iomsg); | |
4553 | if (ret) | |
681fda8d | 4554 | return ret; |
7a7cacba PB |
4555 | kmsg = &iomsg; |
4556 | } | |
03b1230c | 4557 | |
bc02ef33 | 4558 | if (req->flags & REQ_F_BUFFER_SELECT) { |
7fbb1b54 | 4559 | kbuf = io_recv_buffer_select(req, !force_nonblock); |
bc02ef33 | 4560 | if (IS_ERR(kbuf)) |
52de1fe1 | 4561 | return PTR_ERR(kbuf); |
7a7cacba PB |
4562 | kmsg->fast_iov[0].iov_base = u64_to_user_ptr(kbuf->addr); |
4563 | iov_iter_init(&kmsg->msg.msg_iter, READ, kmsg->iov, | |
4564 | 1, req->sr_msg.len); | |
4565 | } | |
52de1fe1 | 4566 | |
7a7cacba PB |
4567 | flags = req->sr_msg.msg_flags; |
4568 | if (flags & MSG_DONTWAIT) | |
4569 | req->flags |= REQ_F_NOWAIT; | |
4570 | else if (force_nonblock) | |
4571 | flags |= MSG_DONTWAIT; | |
e47293fd | 4572 | |
7a7cacba PB |
4573 | ret = __sys_recvmsg_sock(sock, &kmsg->msg, req->sr_msg.umsg, |
4574 | kmsg->uaddr, flags); | |
0e1b6fe3 PB |
4575 | if (force_nonblock && ret == -EAGAIN) |
4576 | return io_setup_async_msg(req, kmsg); | |
7a7cacba PB |
4577 | if (ret == -ERESTARTSYS) |
4578 | ret = -EINTR; | |
03b1230c | 4579 | |
7fbb1b54 PB |
4580 | if (req->flags & REQ_F_BUFFER_SELECTED) |
4581 | cflags = io_put_recv_kbuf(req); | |
6b754c8b | 4582 | if (kmsg->iov != kmsg->fast_iov) |
0b416c3e | 4583 | kfree(kmsg->iov); |
99bc4c38 | 4584 | req->flags &= ~REQ_F_NEED_CLEANUP; |
4e88d6e7 JA |
4585 | if (ret < 0) |
4586 | req_set_fail_links(req); | |
229a7b63 | 4587 | __io_req_complete(req, ret, cflags, cs); |
03b1230c | 4588 | return 0; |
0fa03c62 | 4589 | } |
5d17b4a4 | 4590 | |
229a7b63 JA |
4591 | static int io_recv(struct io_kiocb *req, bool force_nonblock, |
4592 | struct io_comp_state *cs) | |
fddaface | 4593 | { |
6b754c8b | 4594 | struct io_buffer *kbuf; |
7a7cacba PB |
4595 | struct io_sr_msg *sr = &req->sr_msg; |
4596 | struct msghdr msg; | |
4597 | void __user *buf = sr->buf; | |
fddaface | 4598 | struct socket *sock; |
7a7cacba PB |
4599 | struct iovec iov; |
4600 | unsigned flags; | |
bcda7baa | 4601 | int ret, cflags = 0; |
fddaface | 4602 | |
fddaface | 4603 | sock = sock_from_file(req->file, &ret); |
7a7cacba PB |
4604 | if (unlikely(!sock)) |
4605 | return ret; | |
fddaface | 4606 | |
bc02ef33 | 4607 | if (req->flags & REQ_F_BUFFER_SELECT) { |
7fbb1b54 | 4608 | kbuf = io_recv_buffer_select(req, !force_nonblock); |
bcda7baa JA |
4609 | if (IS_ERR(kbuf)) |
4610 | return PTR_ERR(kbuf); | |
7a7cacba | 4611 | buf = u64_to_user_ptr(kbuf->addr); |
bc02ef33 | 4612 | } |
bcda7baa | 4613 | |
7a7cacba | 4614 | ret = import_single_range(READ, buf, sr->len, &iov, &msg.msg_iter); |
14c32eee PB |
4615 | if (unlikely(ret)) |
4616 | goto out_free; | |
fddaface | 4617 | |
7a7cacba PB |
4618 | msg.msg_name = NULL; |
4619 | msg.msg_control = NULL; | |
4620 | msg.msg_controllen = 0; | |
4621 | msg.msg_namelen = 0; | |
4622 | msg.msg_iocb = NULL; | |
4623 | msg.msg_flags = 0; | |
fddaface | 4624 | |
7a7cacba PB |
4625 | flags = req->sr_msg.msg_flags; |
4626 | if (flags & MSG_DONTWAIT) | |
4627 | req->flags |= REQ_F_NOWAIT; | |
4628 | else if (force_nonblock) | |
4629 | flags |= MSG_DONTWAIT; | |
4630 | ||
4631 | ret = sock_recvmsg(sock, &msg, flags); | |
4632 | if (force_nonblock && ret == -EAGAIN) | |
4633 | return -EAGAIN; | |
4634 | if (ret == -ERESTARTSYS) | |
4635 | ret = -EINTR; | |
14c32eee | 4636 | out_free: |
7fbb1b54 PB |
4637 | if (req->flags & REQ_F_BUFFER_SELECTED) |
4638 | cflags = io_put_recv_kbuf(req); | |
fddaface JA |
4639 | if (ret < 0) |
4640 | req_set_fail_links(req); | |
229a7b63 | 4641 | __io_req_complete(req, ret, cflags, cs); |
fddaface | 4642 | return 0; |
fddaface JA |
4643 | } |
4644 | ||
3529d8c2 | 4645 | static int io_accept_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
17f2fe35 | 4646 | { |
8ed8d3c3 JA |
4647 | struct io_accept *accept = &req->accept; |
4648 | ||
17f2fe35 JA |
4649 | if (unlikely(req->ctx->flags & (IORING_SETUP_IOPOLL|IORING_SETUP_SQPOLL))) |
4650 | return -EINVAL; | |
8042d6ce | 4651 | if (sqe->ioprio || sqe->len || sqe->buf_index) |
17f2fe35 JA |
4652 | return -EINVAL; |
4653 | ||
d55e5f5b JA |
4654 | accept->addr = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
4655 | accept->addr_len = u64_to_user_ptr(READ_ONCE(sqe->addr2)); | |
8ed8d3c3 | 4656 | accept->flags = READ_ONCE(sqe->accept_flags); |
09952e3e | 4657 | accept->nofile = rlimit(RLIMIT_NOFILE); |
8ed8d3c3 | 4658 | return 0; |
8ed8d3c3 | 4659 | } |
17f2fe35 | 4660 | |
229a7b63 JA |
4661 | static int io_accept(struct io_kiocb *req, bool force_nonblock, |
4662 | struct io_comp_state *cs) | |
8ed8d3c3 JA |
4663 | { |
4664 | struct io_accept *accept = &req->accept; | |
ac45abc0 | 4665 | unsigned int file_flags = force_nonblock ? O_NONBLOCK : 0; |
8ed8d3c3 JA |
4666 | int ret; |
4667 | ||
e697deed JX |
4668 | if (req->file->f_flags & O_NONBLOCK) |
4669 | req->flags |= REQ_F_NOWAIT; | |
4670 | ||
8ed8d3c3 | 4671 | ret = __sys_accept4_file(req->file, file_flags, accept->addr, |
09952e3e JA |
4672 | accept->addr_len, accept->flags, |
4673 | accept->nofile); | |
8ed8d3c3 | 4674 | if (ret == -EAGAIN && force_nonblock) |
17f2fe35 | 4675 | return -EAGAIN; |
ac45abc0 PB |
4676 | if (ret < 0) { |
4677 | if (ret == -ERESTARTSYS) | |
4678 | ret = -EINTR; | |
4e88d6e7 | 4679 | req_set_fail_links(req); |
ac45abc0 | 4680 | } |
229a7b63 | 4681 | __io_req_complete(req, ret, 0, cs); |
17f2fe35 | 4682 | return 0; |
8ed8d3c3 JA |
4683 | } |
4684 | ||
3529d8c2 | 4685 | static int io_connect_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
f499a021 | 4686 | { |
3529d8c2 | 4687 | struct io_connect *conn = &req->connect; |
e8c2bc1f | 4688 | struct io_async_connect *io = req->async_data; |
f499a021 | 4689 | |
3fbb51c1 JA |
4690 | if (unlikely(req->ctx->flags & (IORING_SETUP_IOPOLL|IORING_SETUP_SQPOLL))) |
4691 | return -EINVAL; | |
4692 | if (sqe->ioprio || sqe->len || sqe->buf_index || sqe->rw_flags) | |
4693 | return -EINVAL; | |
4694 | ||
3529d8c2 JA |
4695 | conn->addr = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
4696 | conn->addr_len = READ_ONCE(sqe->addr2); | |
4697 | ||
4698 | if (!io) | |
4699 | return 0; | |
4700 | ||
4701 | return move_addr_to_kernel(conn->addr, conn->addr_len, | |
e8c2bc1f | 4702 | &io->address); |
f499a021 JA |
4703 | } |
4704 | ||
229a7b63 JA |
4705 | static int io_connect(struct io_kiocb *req, bool force_nonblock, |
4706 | struct io_comp_state *cs) | |
f8e85cf2 | 4707 | { |
e8c2bc1f | 4708 | struct io_async_connect __io, *io; |
f8e85cf2 | 4709 | unsigned file_flags; |
3fbb51c1 | 4710 | int ret; |
f8e85cf2 | 4711 | |
e8c2bc1f JA |
4712 | if (req->async_data) { |
4713 | io = req->async_data; | |
f499a021 | 4714 | } else { |
3529d8c2 JA |
4715 | ret = move_addr_to_kernel(req->connect.addr, |
4716 | req->connect.addr_len, | |
e8c2bc1f | 4717 | &__io.address); |
f499a021 JA |
4718 | if (ret) |
4719 | goto out; | |
4720 | io = &__io; | |
4721 | } | |
4722 | ||
3fbb51c1 JA |
4723 | file_flags = force_nonblock ? O_NONBLOCK : 0; |
4724 | ||
e8c2bc1f | 4725 | ret = __sys_connect_file(req->file, &io->address, |
3fbb51c1 | 4726 | req->connect.addr_len, file_flags); |
87f80d62 | 4727 | if ((ret == -EAGAIN || ret == -EINPROGRESS) && force_nonblock) { |
e8c2bc1f | 4728 | if (req->async_data) |
b7bb4f7d | 4729 | return -EAGAIN; |
e8c2bc1f | 4730 | if (io_alloc_async_data(req)) { |
f499a021 JA |
4731 | ret = -ENOMEM; |
4732 | goto out; | |
4733 | } | |
e8c2bc1f JA |
4734 | io = req->async_data; |
4735 | memcpy(req->async_data, &__io, sizeof(__io)); | |
f8e85cf2 | 4736 | return -EAGAIN; |
f499a021 | 4737 | } |
f8e85cf2 JA |
4738 | if (ret == -ERESTARTSYS) |
4739 | ret = -EINTR; | |
f499a021 | 4740 | out: |
4e88d6e7 JA |
4741 | if (ret < 0) |
4742 | req_set_fail_links(req); | |
229a7b63 | 4743 | __io_req_complete(req, ret, 0, cs); |
f8e85cf2 | 4744 | return 0; |
469956e8 Y |
4745 | } |
4746 | #else /* !CONFIG_NET */ | |
4747 | static int io_sendmsg_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) | |
4748 | { | |
f8e85cf2 | 4749 | return -EOPNOTSUPP; |
f8e85cf2 JA |
4750 | } |
4751 | ||
1e16c2f9 RD |
4752 | static int io_sendmsg(struct io_kiocb *req, bool force_nonblock, |
4753 | struct io_comp_state *cs) | |
469956e8 Y |
4754 | { |
4755 | return -EOPNOTSUPP; | |
4756 | } | |
4757 | ||
1e16c2f9 RD |
4758 | static int io_send(struct io_kiocb *req, bool force_nonblock, |
4759 | struct io_comp_state *cs) | |
469956e8 Y |
4760 | { |
4761 | return -EOPNOTSUPP; | |
4762 | } | |
4763 | ||
4764 | static int io_recvmsg_prep(struct io_kiocb *req, | |
4765 | const struct io_uring_sqe *sqe) | |
4766 | { | |
4767 | return -EOPNOTSUPP; | |
4768 | } | |
4769 | ||
1e16c2f9 RD |
4770 | static int io_recvmsg(struct io_kiocb *req, bool force_nonblock, |
4771 | struct io_comp_state *cs) | |
469956e8 Y |
4772 | { |
4773 | return -EOPNOTSUPP; | |
4774 | } | |
4775 | ||
1e16c2f9 RD |
4776 | static int io_recv(struct io_kiocb *req, bool force_nonblock, |
4777 | struct io_comp_state *cs) | |
469956e8 Y |
4778 | { |
4779 | return -EOPNOTSUPP; | |
4780 | } | |
4781 | ||
4782 | static int io_accept_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) | |
4783 | { | |
4784 | return -EOPNOTSUPP; | |
4785 | } | |
4786 | ||
1e16c2f9 RD |
4787 | static int io_accept(struct io_kiocb *req, bool force_nonblock, |
4788 | struct io_comp_state *cs) | |
469956e8 Y |
4789 | { |
4790 | return -EOPNOTSUPP; | |
4791 | } | |
ce593a6c | 4792 | |
469956e8 Y |
4793 | static int io_connect_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
4794 | { | |
4795 | return -EOPNOTSUPP; | |
4796 | } | |
4797 | ||
1e16c2f9 RD |
4798 | static int io_connect(struct io_kiocb *req, bool force_nonblock, |
4799 | struct io_comp_state *cs) | |
469956e8 | 4800 | { |
f8e85cf2 | 4801 | return -EOPNOTSUPP; |
ce593a6c | 4802 | } |
469956e8 | 4803 | #endif /* CONFIG_NET */ |
f8e85cf2 | 4804 | |
d7718a9d JA |
4805 | struct io_poll_table { |
4806 | struct poll_table_struct pt; | |
4807 | struct io_kiocb *req; | |
4808 | int error; | |
4809 | }; | |
ce593a6c | 4810 | |
d7718a9d JA |
4811 | static int __io_async_wake(struct io_kiocb *req, struct io_poll_iocb *poll, |
4812 | __poll_t mask, task_work_func_t func) | |
4813 | { | |
fd7d6de2 | 4814 | bool twa_signal_ok; |
aa96bf8a | 4815 | int ret; |
d7718a9d JA |
4816 | |
4817 | /* for instances that support it check for an event match first: */ | |
4818 | if (mask && !(mask & poll->events)) | |
4819 | return 0; | |
4820 | ||
4821 | trace_io_uring_task_add(req->ctx, req->opcode, req->user_data, mask); | |
4822 | ||
4823 | list_del_init(&poll->wait.entry); | |
4824 | ||
d7718a9d JA |
4825 | req->result = mask; |
4826 | init_task_work(&req->task_work, func); | |
6d816e08 JA |
4827 | percpu_ref_get(&req->ctx->refs); |
4828 | ||
fd7d6de2 JA |
4829 | /* |
4830 | * If we using the signalfd wait_queue_head for this wakeup, then | |
4831 | * it's not safe to use TWA_SIGNAL as we could be recursing on the | |
4832 | * tsk->sighand->siglock on doing the wakeup. Should not be needed | |
4833 | * either, as the normal wakeup will suffice. | |
4834 | */ | |
4835 | twa_signal_ok = (poll->head != &req->task->sighand->signalfd_wqh); | |
4836 | ||
d7718a9d | 4837 | /* |
e3aabf95 JA |
4838 | * If this fails, then the task is exiting. When a task exits, the |
4839 | * work gets canceled, so just cancel this request as well instead | |
4840 | * of executing it. We can't safely execute it anyway, as we may not | |
4841 | * have the needed state needed for it anyway. | |
d7718a9d | 4842 | */ |
87c4311f | 4843 | ret = io_req_task_work_add(req, twa_signal_ok); |
aa96bf8a | 4844 | if (unlikely(ret)) { |
c2c4c83c JA |
4845 | struct task_struct *tsk; |
4846 | ||
e3aabf95 | 4847 | WRITE_ONCE(poll->canceled, true); |
aa96bf8a | 4848 | tsk = io_wq_get_task(req->ctx->io_wq); |
91989c70 | 4849 | task_work_add(tsk, &req->task_work, TWA_NONE); |
ce593a6c | 4850 | wake_up_process(tsk); |
aa96bf8a | 4851 | } |
d7718a9d JA |
4852 | return 1; |
4853 | } | |
4854 | ||
74ce6ce4 JA |
4855 | static bool io_poll_rewait(struct io_kiocb *req, struct io_poll_iocb *poll) |
4856 | __acquires(&req->ctx->completion_lock) | |
4857 | { | |
4858 | struct io_ring_ctx *ctx = req->ctx; | |
4859 | ||
4860 | if (!req->result && !READ_ONCE(poll->canceled)) { | |
4861 | struct poll_table_struct pt = { ._key = poll->events }; | |
4862 | ||
4863 | req->result = vfs_poll(req->file, &pt) & poll->events; | |
4864 | } | |
4865 | ||
4866 | spin_lock_irq(&ctx->completion_lock); | |
4867 | if (!req->result && !READ_ONCE(poll->canceled)) { | |
4868 | add_wait_queue(poll->head, &poll->wait); | |
4869 | return true; | |
4870 | } | |
4871 | ||
4872 | return false; | |
4873 | } | |
4874 | ||
d4e7cd36 | 4875 | static struct io_poll_iocb *io_poll_get_double(struct io_kiocb *req) |
18bceab1 | 4876 | { |
e8c2bc1f | 4877 | /* pure poll stashes this in ->async_data, poll driven retry elsewhere */ |
d4e7cd36 | 4878 | if (req->opcode == IORING_OP_POLL_ADD) |
e8c2bc1f | 4879 | return req->async_data; |
d4e7cd36 JA |
4880 | return req->apoll->double_poll; |
4881 | } | |
4882 | ||
4883 | static struct io_poll_iocb *io_poll_get_single(struct io_kiocb *req) | |
4884 | { | |
4885 | if (req->opcode == IORING_OP_POLL_ADD) | |
4886 | return &req->poll; | |
4887 | return &req->apoll->poll; | |
4888 | } | |
4889 | ||
4890 | static void io_poll_remove_double(struct io_kiocb *req) | |
4891 | { | |
4892 | struct io_poll_iocb *poll = io_poll_get_double(req); | |
18bceab1 JA |
4893 | |
4894 | lockdep_assert_held(&req->ctx->completion_lock); | |
4895 | ||
4896 | if (poll && poll->head) { | |
4897 | struct wait_queue_head *head = poll->head; | |
4898 | ||
4899 | spin_lock(&head->lock); | |
4900 | list_del_init(&poll->wait.entry); | |
4901 | if (poll->wait.private) | |
4902 | refcount_dec(&req->refs); | |
4903 | poll->head = NULL; | |
4904 | spin_unlock(&head->lock); | |
4905 | } | |
4906 | } | |
4907 | ||
4908 | static void io_poll_complete(struct io_kiocb *req, __poll_t mask, int error) | |
4909 | { | |
4910 | struct io_ring_ctx *ctx = req->ctx; | |
4911 | ||
d4e7cd36 | 4912 | io_poll_remove_double(req); |
18bceab1 JA |
4913 | req->poll.done = true; |
4914 | io_cqring_fill_event(req, error ? error : mangle_poll(mask)); | |
4915 | io_commit_cqring(ctx); | |
4916 | } | |
4917 | ||
dd221f46 | 4918 | static void io_poll_task_func(struct callback_head *cb) |
18bceab1 | 4919 | { |
dd221f46 | 4920 | struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work); |
18bceab1 | 4921 | struct io_ring_ctx *ctx = req->ctx; |
dd221f46 | 4922 | struct io_kiocb *nxt; |
18bceab1 JA |
4923 | |
4924 | if (io_poll_rewait(req, &req->poll)) { | |
4925 | spin_unlock_irq(&ctx->completion_lock); | |
dd221f46 PB |
4926 | } else { |
4927 | hash_del(&req->hash_node); | |
4928 | io_poll_complete(req, req->result, 0); | |
4929 | spin_unlock_irq(&ctx->completion_lock); | |
18bceab1 | 4930 | |
dd221f46 PB |
4931 | nxt = io_put_req_find_next(req); |
4932 | io_cqring_ev_posted(ctx); | |
4933 | if (nxt) | |
4934 | __io_req_task_submit(nxt); | |
4935 | } | |
18bceab1 | 4936 | |
6d816e08 | 4937 | percpu_ref_put(&ctx->refs); |
18bceab1 JA |
4938 | } |
4939 | ||
4940 | static int io_poll_double_wake(struct wait_queue_entry *wait, unsigned mode, | |
4941 | int sync, void *key) | |
4942 | { | |
4943 | struct io_kiocb *req = wait->private; | |
d4e7cd36 | 4944 | struct io_poll_iocb *poll = io_poll_get_single(req); |
18bceab1 JA |
4945 | __poll_t mask = key_to_poll(key); |
4946 | ||
4947 | /* for instances that support it check for an event match first: */ | |
4948 | if (mask && !(mask & poll->events)) | |
4949 | return 0; | |
4950 | ||
8706e04e JA |
4951 | list_del_init(&wait->entry); |
4952 | ||
807abcb0 | 4953 | if (poll && poll->head) { |
18bceab1 JA |
4954 | bool done; |
4955 | ||
807abcb0 JA |
4956 | spin_lock(&poll->head->lock); |
4957 | done = list_empty(&poll->wait.entry); | |
18bceab1 | 4958 | if (!done) |
807abcb0 | 4959 | list_del_init(&poll->wait.entry); |
d4e7cd36 JA |
4960 | /* make sure double remove sees this as being gone */ |
4961 | wait->private = NULL; | |
807abcb0 | 4962 | spin_unlock(&poll->head->lock); |
c8b5e260 JA |
4963 | if (!done) { |
4964 | /* use wait func handler, so it matches the rq type */ | |
4965 | poll->wait.func(&poll->wait, mode, sync, key); | |
4966 | } | |
18bceab1 JA |
4967 | } |
4968 | refcount_dec(&req->refs); | |
4969 | return 1; | |
4970 | } | |
4971 | ||
4972 | static void io_init_poll_iocb(struct io_poll_iocb *poll, __poll_t events, | |
4973 | wait_queue_func_t wake_func) | |
4974 | { | |
4975 | poll->head = NULL; | |
4976 | poll->done = false; | |
4977 | poll->canceled = false; | |
4978 | poll->events = events; | |
4979 | INIT_LIST_HEAD(&poll->wait.entry); | |
4980 | init_waitqueue_func_entry(&poll->wait, wake_func); | |
4981 | } | |
4982 | ||
4983 | static void __io_queue_proc(struct io_poll_iocb *poll, struct io_poll_table *pt, | |
807abcb0 JA |
4984 | struct wait_queue_head *head, |
4985 | struct io_poll_iocb **poll_ptr) | |
18bceab1 JA |
4986 | { |
4987 | struct io_kiocb *req = pt->req; | |
4988 | ||
4989 | /* | |
4990 | * If poll->head is already set, it's because the file being polled | |
4991 | * uses multiple waitqueues for poll handling (eg one for read, one | |
4992 | * for write). Setup a separate io_poll_iocb if this happens. | |
4993 | */ | |
4994 | if (unlikely(poll->head)) { | |
58852d4d PB |
4995 | struct io_poll_iocb *poll_one = poll; |
4996 | ||
18bceab1 | 4997 | /* already have a 2nd entry, fail a third attempt */ |
807abcb0 | 4998 | if (*poll_ptr) { |
18bceab1 JA |
4999 | pt->error = -EINVAL; |
5000 | return; | |
5001 | } | |
5002 | poll = kmalloc(sizeof(*poll), GFP_ATOMIC); | |
5003 | if (!poll) { | |
5004 | pt->error = -ENOMEM; | |
5005 | return; | |
5006 | } | |
58852d4d | 5007 | io_init_poll_iocb(poll, poll_one->events, io_poll_double_wake); |
18bceab1 JA |
5008 | refcount_inc(&req->refs); |
5009 | poll->wait.private = req; | |
807abcb0 | 5010 | *poll_ptr = poll; |
18bceab1 JA |
5011 | } |
5012 | ||
5013 | pt->error = 0; | |
5014 | poll->head = head; | |
a31eb4a2 JX |
5015 | |
5016 | if (poll->events & EPOLLEXCLUSIVE) | |
5017 | add_wait_queue_exclusive(head, &poll->wait); | |
5018 | else | |
5019 | add_wait_queue(head, &poll->wait); | |
18bceab1 JA |
5020 | } |
5021 | ||
5022 | static void io_async_queue_proc(struct file *file, struct wait_queue_head *head, | |
5023 | struct poll_table_struct *p) | |
5024 | { | |
5025 | struct io_poll_table *pt = container_of(p, struct io_poll_table, pt); | |
807abcb0 | 5026 | struct async_poll *apoll = pt->req->apoll; |
18bceab1 | 5027 | |
807abcb0 | 5028 | __io_queue_proc(&apoll->poll, pt, head, &apoll->double_poll); |
18bceab1 JA |
5029 | } |
5030 | ||
d7718a9d JA |
5031 | static void io_async_task_func(struct callback_head *cb) |
5032 | { | |
5033 | struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work); | |
5034 | struct async_poll *apoll = req->apoll; | |
5035 | struct io_ring_ctx *ctx = req->ctx; | |
5036 | ||
5037 | trace_io_uring_task_run(req->ctx, req->opcode, req->user_data); | |
5038 | ||
74ce6ce4 | 5039 | if (io_poll_rewait(req, &apoll->poll)) { |
d7718a9d | 5040 | spin_unlock_irq(&ctx->completion_lock); |
6d816e08 | 5041 | percpu_ref_put(&ctx->refs); |
74ce6ce4 | 5042 | return; |
d7718a9d JA |
5043 | } |
5044 | ||
31067255 | 5045 | /* If req is still hashed, it cannot have been canceled. Don't check. */ |
0be0b0e3 | 5046 | if (hash_hashed(&req->hash_node)) |
74ce6ce4 | 5047 | hash_del(&req->hash_node); |
2bae047e | 5048 | |
d4e7cd36 | 5049 | io_poll_remove_double(req); |
74ce6ce4 JA |
5050 | spin_unlock_irq(&ctx->completion_lock); |
5051 | ||
0be0b0e3 PB |
5052 | if (!READ_ONCE(apoll->poll.canceled)) |
5053 | __io_req_task_submit(req); | |
5054 | else | |
5055 | __io_req_task_cancel(req, -ECANCELED); | |
aa340845 | 5056 | |
6d816e08 | 5057 | percpu_ref_put(&ctx->refs); |
807abcb0 | 5058 | kfree(apoll->double_poll); |
31067255 | 5059 | kfree(apoll); |
d7718a9d JA |
5060 | } |
5061 | ||
5062 | static int io_async_wake(struct wait_queue_entry *wait, unsigned mode, int sync, | |
5063 | void *key) | |
5064 | { | |
5065 | struct io_kiocb *req = wait->private; | |
5066 | struct io_poll_iocb *poll = &req->apoll->poll; | |
5067 | ||
5068 | trace_io_uring_poll_wake(req->ctx, req->opcode, req->user_data, | |
5069 | key_to_poll(key)); | |
5070 | ||
5071 | return __io_async_wake(req, poll, key_to_poll(key), io_async_task_func); | |
5072 | } | |
5073 | ||
5074 | static void io_poll_req_insert(struct io_kiocb *req) | |
5075 | { | |
5076 | struct io_ring_ctx *ctx = req->ctx; | |
5077 | struct hlist_head *list; | |
5078 | ||
5079 | list = &ctx->cancel_hash[hash_long(req->user_data, ctx->cancel_hash_bits)]; | |
5080 | hlist_add_head(&req->hash_node, list); | |
5081 | } | |
5082 | ||
5083 | static __poll_t __io_arm_poll_handler(struct io_kiocb *req, | |
5084 | struct io_poll_iocb *poll, | |
5085 | struct io_poll_table *ipt, __poll_t mask, | |
5086 | wait_queue_func_t wake_func) | |
5087 | __acquires(&ctx->completion_lock) | |
5088 | { | |
5089 | struct io_ring_ctx *ctx = req->ctx; | |
5090 | bool cancel = false; | |
5091 | ||
4d52f338 | 5092 | INIT_HLIST_NODE(&req->hash_node); |
18bceab1 | 5093 | io_init_poll_iocb(poll, mask, wake_func); |
b90cd197 | 5094 | poll->file = req->file; |
18bceab1 | 5095 | poll->wait.private = req; |
d7718a9d JA |
5096 | |
5097 | ipt->pt._key = mask; | |
5098 | ipt->req = req; | |
5099 | ipt->error = -EINVAL; | |
5100 | ||
d7718a9d JA |
5101 | mask = vfs_poll(req->file, &ipt->pt) & poll->events; |
5102 | ||
5103 | spin_lock_irq(&ctx->completion_lock); | |
5104 | if (likely(poll->head)) { | |
5105 | spin_lock(&poll->head->lock); | |
5106 | if (unlikely(list_empty(&poll->wait.entry))) { | |
5107 | if (ipt->error) | |
5108 | cancel = true; | |
5109 | ipt->error = 0; | |
5110 | mask = 0; | |
5111 | } | |
5112 | if (mask || ipt->error) | |
5113 | list_del_init(&poll->wait.entry); | |
5114 | else if (cancel) | |
5115 | WRITE_ONCE(poll->canceled, true); | |
5116 | else if (!poll->done) /* actually waiting for an event */ | |
5117 | io_poll_req_insert(req); | |
5118 | spin_unlock(&poll->head->lock); | |
5119 | } | |
5120 | ||
5121 | return mask; | |
5122 | } | |
5123 | ||
5124 | static bool io_arm_poll_handler(struct io_kiocb *req) | |
5125 | { | |
5126 | const struct io_op_def *def = &io_op_defs[req->opcode]; | |
5127 | struct io_ring_ctx *ctx = req->ctx; | |
5128 | struct async_poll *apoll; | |
5129 | struct io_poll_table ipt; | |
5130 | __poll_t mask, ret; | |
9dab14b8 | 5131 | int rw; |
d7718a9d JA |
5132 | |
5133 | if (!req->file || !file_can_poll(req->file)) | |
5134 | return false; | |
24c74678 | 5135 | if (req->flags & REQ_F_POLLED) |
d7718a9d | 5136 | return false; |
9dab14b8 JA |
5137 | if (def->pollin) |
5138 | rw = READ; | |
5139 | else if (def->pollout) | |
5140 | rw = WRITE; | |
5141 | else | |
5142 | return false; | |
5143 | /* if we can't nonblock try, then no point in arming a poll handler */ | |
5144 | if (!io_file_supports_async(req->file, rw)) | |
d7718a9d JA |
5145 | return false; |
5146 | ||
5147 | apoll = kmalloc(sizeof(*apoll), GFP_ATOMIC); | |
5148 | if (unlikely(!apoll)) | |
5149 | return false; | |
807abcb0 | 5150 | apoll->double_poll = NULL; |
d7718a9d JA |
5151 | |
5152 | req->flags |= REQ_F_POLLED; | |
d7718a9d | 5153 | req->apoll = apoll; |
d7718a9d | 5154 | |
8755d97a | 5155 | mask = 0; |
d7718a9d | 5156 | if (def->pollin) |
8755d97a | 5157 | mask |= POLLIN | POLLRDNORM; |
d7718a9d JA |
5158 | if (def->pollout) |
5159 | mask |= POLLOUT | POLLWRNORM; | |
901341bb LH |
5160 | |
5161 | /* If reading from MSG_ERRQUEUE using recvmsg, ignore POLLIN */ | |
5162 | if ((req->opcode == IORING_OP_RECVMSG) && | |
5163 | (req->sr_msg.msg_flags & MSG_ERRQUEUE)) | |
5164 | mask &= ~POLLIN; | |
5165 | ||
d7718a9d JA |
5166 | mask |= POLLERR | POLLPRI; |
5167 | ||
5168 | ipt.pt._qproc = io_async_queue_proc; | |
5169 | ||
5170 | ret = __io_arm_poll_handler(req, &apoll->poll, &ipt, mask, | |
5171 | io_async_wake); | |
a36da65c | 5172 | if (ret || ipt.error) { |
d4e7cd36 | 5173 | io_poll_remove_double(req); |
d7718a9d | 5174 | spin_unlock_irq(&ctx->completion_lock); |
807abcb0 | 5175 | kfree(apoll->double_poll); |
d7718a9d JA |
5176 | kfree(apoll); |
5177 | return false; | |
5178 | } | |
5179 | spin_unlock_irq(&ctx->completion_lock); | |
5180 | trace_io_uring_poll_arm(ctx, req->opcode, req->user_data, mask, | |
5181 | apoll->poll.events); | |
5182 | return true; | |
5183 | } | |
5184 | ||
5185 | static bool __io_poll_remove_one(struct io_kiocb *req, | |
5186 | struct io_poll_iocb *poll) | |
221c5eb2 | 5187 | { |
b41e9852 | 5188 | bool do_complete = false; |
221c5eb2 JA |
5189 | |
5190 | spin_lock(&poll->head->lock); | |
5191 | WRITE_ONCE(poll->canceled, true); | |
392edb45 JA |
5192 | if (!list_empty(&poll->wait.entry)) { |
5193 | list_del_init(&poll->wait.entry); | |
b41e9852 | 5194 | do_complete = true; |
221c5eb2 JA |
5195 | } |
5196 | spin_unlock(&poll->head->lock); | |
3bfa5bcb | 5197 | hash_del(&req->hash_node); |
d7718a9d JA |
5198 | return do_complete; |
5199 | } | |
5200 | ||
5201 | static bool io_poll_remove_one(struct io_kiocb *req) | |
5202 | { | |
5203 | bool do_complete; | |
5204 | ||
d4e7cd36 JA |
5205 | io_poll_remove_double(req); |
5206 | ||
d7718a9d JA |
5207 | if (req->opcode == IORING_OP_POLL_ADD) { |
5208 | do_complete = __io_poll_remove_one(req, &req->poll); | |
5209 | } else { | |
3bfa5bcb JA |
5210 | struct async_poll *apoll = req->apoll; |
5211 | ||
d7718a9d | 5212 | /* non-poll requests have submit ref still */ |
3bfa5bcb JA |
5213 | do_complete = __io_poll_remove_one(req, &apoll->poll); |
5214 | if (do_complete) { | |
d7718a9d | 5215 | io_put_req(req); |
807abcb0 | 5216 | kfree(apoll->double_poll); |
3bfa5bcb JA |
5217 | kfree(apoll); |
5218 | } | |
b1f573bd XW |
5219 | } |
5220 | ||
b41e9852 JA |
5221 | if (do_complete) { |
5222 | io_cqring_fill_event(req, -ECANCELED); | |
5223 | io_commit_cqring(req->ctx); | |
f254ac04 | 5224 | req_set_fail_links(req); |
216578e5 | 5225 | io_put_req_deferred(req, 1); |
b41e9852 JA |
5226 | } |
5227 | ||
5228 | return do_complete; | |
221c5eb2 JA |
5229 | } |
5230 | ||
76e1b642 JA |
5231 | /* |
5232 | * Returns true if we found and killed one or more poll requests | |
5233 | */ | |
5234 | static bool io_poll_remove_all(struct io_ring_ctx *ctx, struct task_struct *tsk) | |
221c5eb2 | 5235 | { |
78076bb6 | 5236 | struct hlist_node *tmp; |
221c5eb2 | 5237 | struct io_kiocb *req; |
8e2e1faf | 5238 | int posted = 0, i; |
221c5eb2 JA |
5239 | |
5240 | spin_lock_irq(&ctx->completion_lock); | |
78076bb6 JA |
5241 | for (i = 0; i < (1U << ctx->cancel_hash_bits); i++) { |
5242 | struct hlist_head *list; | |
5243 | ||
5244 | list = &ctx->cancel_hash[i]; | |
f3606e3a JA |
5245 | hlist_for_each_entry_safe(req, tmp, list, hash_node) { |
5246 | if (io_task_match(req, tsk)) | |
5247 | posted += io_poll_remove_one(req); | |
5248 | } | |
221c5eb2 JA |
5249 | } |
5250 | spin_unlock_irq(&ctx->completion_lock); | |
b41e9852 | 5251 | |
8e2e1faf JA |
5252 | if (posted) |
5253 | io_cqring_ev_posted(ctx); | |
76e1b642 JA |
5254 | |
5255 | return posted != 0; | |
221c5eb2 JA |
5256 | } |
5257 | ||
47f46768 JA |
5258 | static int io_poll_cancel(struct io_ring_ctx *ctx, __u64 sqe_addr) |
5259 | { | |
78076bb6 | 5260 | struct hlist_head *list; |
47f46768 JA |
5261 | struct io_kiocb *req; |
5262 | ||
78076bb6 JA |
5263 | list = &ctx->cancel_hash[hash_long(sqe_addr, ctx->cancel_hash_bits)]; |
5264 | hlist_for_each_entry(req, list, hash_node) { | |
b41e9852 JA |
5265 | if (sqe_addr != req->user_data) |
5266 | continue; | |
5267 | if (io_poll_remove_one(req)) | |
eac406c6 | 5268 | return 0; |
b41e9852 | 5269 | return -EALREADY; |
47f46768 JA |
5270 | } |
5271 | ||
5272 | return -ENOENT; | |
5273 | } | |
5274 | ||
3529d8c2 JA |
5275 | static int io_poll_remove_prep(struct io_kiocb *req, |
5276 | const struct io_uring_sqe *sqe) | |
0969e783 | 5277 | { |
0969e783 JA |
5278 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
5279 | return -EINVAL; | |
5280 | if (sqe->ioprio || sqe->off || sqe->len || sqe->buf_index || | |
5281 | sqe->poll_events) | |
5282 | return -EINVAL; | |
5283 | ||
5284 | req->poll.addr = READ_ONCE(sqe->addr); | |
0969e783 JA |
5285 | return 0; |
5286 | } | |
5287 | ||
221c5eb2 JA |
5288 | /* |
5289 | * Find a running poll command that matches one specified in sqe->addr, | |
5290 | * and remove it if found. | |
5291 | */ | |
fc4df999 | 5292 | static int io_poll_remove(struct io_kiocb *req) |
221c5eb2 JA |
5293 | { |
5294 | struct io_ring_ctx *ctx = req->ctx; | |
0969e783 | 5295 | u64 addr; |
47f46768 | 5296 | int ret; |
221c5eb2 | 5297 | |
0969e783 | 5298 | addr = req->poll.addr; |
221c5eb2 | 5299 | spin_lock_irq(&ctx->completion_lock); |
0969e783 | 5300 | ret = io_poll_cancel(ctx, addr); |
221c5eb2 JA |
5301 | spin_unlock_irq(&ctx->completion_lock); |
5302 | ||
4e88d6e7 JA |
5303 | if (ret < 0) |
5304 | req_set_fail_links(req); | |
e1e16097 | 5305 | io_req_complete(req, ret); |
221c5eb2 JA |
5306 | return 0; |
5307 | } | |
5308 | ||
221c5eb2 JA |
5309 | static int io_poll_wake(struct wait_queue_entry *wait, unsigned mode, int sync, |
5310 | void *key) | |
5311 | { | |
c2f2eb7d JA |
5312 | struct io_kiocb *req = wait->private; |
5313 | struct io_poll_iocb *poll = &req->poll; | |
221c5eb2 | 5314 | |
d7718a9d | 5315 | return __io_async_wake(req, poll, key_to_poll(key), io_poll_task_func); |
221c5eb2 JA |
5316 | } |
5317 | ||
221c5eb2 JA |
5318 | static void io_poll_queue_proc(struct file *file, struct wait_queue_head *head, |
5319 | struct poll_table_struct *p) | |
5320 | { | |
5321 | struct io_poll_table *pt = container_of(p, struct io_poll_table, pt); | |
5322 | ||
e8c2bc1f | 5323 | __io_queue_proc(&pt->req->poll, pt, head, (struct io_poll_iocb **) &pt->req->async_data); |
eac406c6 JA |
5324 | } |
5325 | ||
3529d8c2 | 5326 | static int io_poll_add_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
221c5eb2 JA |
5327 | { |
5328 | struct io_poll_iocb *poll = &req->poll; | |
5769a351 | 5329 | u32 events; |
221c5eb2 JA |
5330 | |
5331 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) | |
5332 | return -EINVAL; | |
5333 | if (sqe->addr || sqe->ioprio || sqe->off || sqe->len || sqe->buf_index) | |
5334 | return -EINVAL; | |
5335 | ||
5769a351 JX |
5336 | events = READ_ONCE(sqe->poll32_events); |
5337 | #ifdef __BIG_ENDIAN | |
5338 | events = swahw32(events); | |
5339 | #endif | |
a31eb4a2 JX |
5340 | poll->events = demangle_poll(events) | EPOLLERR | EPOLLHUP | |
5341 | (events & EPOLLEXCLUSIVE); | |
0969e783 JA |
5342 | return 0; |
5343 | } | |
5344 | ||
014db007 | 5345 | static int io_poll_add(struct io_kiocb *req) |
0969e783 JA |
5346 | { |
5347 | struct io_poll_iocb *poll = &req->poll; | |
5348 | struct io_ring_ctx *ctx = req->ctx; | |
5349 | struct io_poll_table ipt; | |
0969e783 | 5350 | __poll_t mask; |
0969e783 | 5351 | |
d7718a9d | 5352 | ipt.pt._qproc = io_poll_queue_proc; |
36703247 | 5353 | |
d7718a9d JA |
5354 | mask = __io_arm_poll_handler(req, &req->poll, &ipt, poll->events, |
5355 | io_poll_wake); | |
221c5eb2 | 5356 | |
8c838788 | 5357 | if (mask) { /* no async, we'd stolen it */ |
221c5eb2 | 5358 | ipt.error = 0; |
b0dd8a41 | 5359 | io_poll_complete(req, mask, 0); |
221c5eb2 | 5360 | } |
221c5eb2 JA |
5361 | spin_unlock_irq(&ctx->completion_lock); |
5362 | ||
8c838788 JA |
5363 | if (mask) { |
5364 | io_cqring_ev_posted(ctx); | |
014db007 | 5365 | io_put_req(req); |
221c5eb2 | 5366 | } |
8c838788 | 5367 | return ipt.error; |
221c5eb2 JA |
5368 | } |
5369 | ||
5262f567 JA |
5370 | static enum hrtimer_restart io_timeout_fn(struct hrtimer *timer) |
5371 | { | |
ad8a48ac JA |
5372 | struct io_timeout_data *data = container_of(timer, |
5373 | struct io_timeout_data, timer); | |
5374 | struct io_kiocb *req = data->req; | |
5375 | struct io_ring_ctx *ctx = req->ctx; | |
5262f567 JA |
5376 | unsigned long flags; |
5377 | ||
5262f567 | 5378 | spin_lock_irqsave(&ctx->completion_lock, flags); |
a71976f3 | 5379 | list_del_init(&req->timeout.list); |
01cec8c1 PB |
5380 | atomic_set(&req->ctx->cq_timeouts, |
5381 | atomic_read(&req->ctx->cq_timeouts) + 1); | |
5382 | ||
78e19bbe | 5383 | io_cqring_fill_event(req, -ETIME); |
5262f567 JA |
5384 | io_commit_cqring(ctx); |
5385 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
5386 | ||
5387 | io_cqring_ev_posted(ctx); | |
4e88d6e7 | 5388 | req_set_fail_links(req); |
5262f567 JA |
5389 | io_put_req(req); |
5390 | return HRTIMER_NORESTART; | |
5391 | } | |
5392 | ||
f254ac04 JA |
5393 | static int __io_timeout_cancel(struct io_kiocb *req) |
5394 | { | |
e8c2bc1f | 5395 | struct io_timeout_data *io = req->async_data; |
f254ac04 JA |
5396 | int ret; |
5397 | ||
e8c2bc1f | 5398 | ret = hrtimer_try_to_cancel(&io->timer); |
f254ac04 JA |
5399 | if (ret == -1) |
5400 | return -EALREADY; | |
a71976f3 | 5401 | list_del_init(&req->timeout.list); |
f254ac04 JA |
5402 | |
5403 | req_set_fail_links(req); | |
f254ac04 | 5404 | io_cqring_fill_event(req, -ECANCELED); |
216578e5 | 5405 | io_put_req_deferred(req, 1); |
f254ac04 JA |
5406 | return 0; |
5407 | } | |
5408 | ||
47f46768 JA |
5409 | static int io_timeout_cancel(struct io_ring_ctx *ctx, __u64 user_data) |
5410 | { | |
5411 | struct io_kiocb *req; | |
5412 | int ret = -ENOENT; | |
5413 | ||
135fcde8 | 5414 | list_for_each_entry(req, &ctx->timeout_list, timeout.list) { |
47f46768 | 5415 | if (user_data == req->user_data) { |
47f46768 JA |
5416 | ret = 0; |
5417 | break; | |
5418 | } | |
5419 | } | |
5420 | ||
5421 | if (ret == -ENOENT) | |
5422 | return ret; | |
5423 | ||
f254ac04 | 5424 | return __io_timeout_cancel(req); |
47f46768 JA |
5425 | } |
5426 | ||
3529d8c2 JA |
5427 | static int io_timeout_remove_prep(struct io_kiocb *req, |
5428 | const struct io_uring_sqe *sqe) | |
b29472ee | 5429 | { |
b29472ee JA |
5430 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
5431 | return -EINVAL; | |
61710e43 DA |
5432 | if (unlikely(req->flags & (REQ_F_FIXED_FILE | REQ_F_BUFFER_SELECT))) |
5433 | return -EINVAL; | |
0bdf7a2d | 5434 | if (sqe->ioprio || sqe->buf_index || sqe->len || sqe->timeout_flags) |
b29472ee JA |
5435 | return -EINVAL; |
5436 | ||
0bdf7a2d | 5437 | req->timeout_rem.addr = READ_ONCE(sqe->addr); |
b29472ee JA |
5438 | return 0; |
5439 | } | |
5440 | ||
11365043 JA |
5441 | /* |
5442 | * Remove or update an existing timeout command | |
5443 | */ | |
fc4df999 | 5444 | static int io_timeout_remove(struct io_kiocb *req) |
11365043 JA |
5445 | { |
5446 | struct io_ring_ctx *ctx = req->ctx; | |
47f46768 | 5447 | int ret; |
11365043 | 5448 | |
11365043 | 5449 | spin_lock_irq(&ctx->completion_lock); |
0bdf7a2d | 5450 | ret = io_timeout_cancel(ctx, req->timeout_rem.addr); |
11365043 | 5451 | |
47f46768 | 5452 | io_cqring_fill_event(req, ret); |
11365043 JA |
5453 | io_commit_cqring(ctx); |
5454 | spin_unlock_irq(&ctx->completion_lock); | |
5262f567 | 5455 | io_cqring_ev_posted(ctx); |
4e88d6e7 JA |
5456 | if (ret < 0) |
5457 | req_set_fail_links(req); | |
ec9c02ad | 5458 | io_put_req(req); |
11365043 | 5459 | return 0; |
5262f567 JA |
5460 | } |
5461 | ||
3529d8c2 | 5462 | static int io_timeout_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe, |
2d28390a | 5463 | bool is_timeout_link) |
5262f567 | 5464 | { |
ad8a48ac | 5465 | struct io_timeout_data *data; |
a41525ab | 5466 | unsigned flags; |
56080b02 | 5467 | u32 off = READ_ONCE(sqe->off); |
5262f567 | 5468 | |
ad8a48ac | 5469 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
5262f567 | 5470 | return -EINVAL; |
ad8a48ac | 5471 | if (sqe->ioprio || sqe->buf_index || sqe->len != 1) |
a41525ab | 5472 | return -EINVAL; |
56080b02 | 5473 | if (off && is_timeout_link) |
2d28390a | 5474 | return -EINVAL; |
a41525ab JA |
5475 | flags = READ_ONCE(sqe->timeout_flags); |
5476 | if (flags & ~IORING_TIMEOUT_ABS) | |
5262f567 | 5477 | return -EINVAL; |
bdf20073 | 5478 | |
bfe68a22 | 5479 | req->timeout.off = off; |
26a61679 | 5480 | |
e8c2bc1f | 5481 | if (!req->async_data && io_alloc_async_data(req)) |
26a61679 JA |
5482 | return -ENOMEM; |
5483 | ||
e8c2bc1f | 5484 | data = req->async_data; |
ad8a48ac | 5485 | data->req = req; |
ad8a48ac JA |
5486 | |
5487 | if (get_timespec64(&data->ts, u64_to_user_ptr(sqe->addr))) | |
5262f567 JA |
5488 | return -EFAULT; |
5489 | ||
11365043 | 5490 | if (flags & IORING_TIMEOUT_ABS) |
ad8a48ac | 5491 | data->mode = HRTIMER_MODE_ABS; |
11365043 | 5492 | else |
ad8a48ac | 5493 | data->mode = HRTIMER_MODE_REL; |
11365043 | 5494 | |
ad8a48ac JA |
5495 | hrtimer_init(&data->timer, CLOCK_MONOTONIC, data->mode); |
5496 | return 0; | |
5497 | } | |
5498 | ||
fc4df999 | 5499 | static int io_timeout(struct io_kiocb *req) |
ad8a48ac | 5500 | { |
ad8a48ac | 5501 | struct io_ring_ctx *ctx = req->ctx; |
e8c2bc1f | 5502 | struct io_timeout_data *data = req->async_data; |
ad8a48ac | 5503 | struct list_head *entry; |
bfe68a22 | 5504 | u32 tail, off = req->timeout.off; |
ad8a48ac | 5505 | |
733f5c95 | 5506 | spin_lock_irq(&ctx->completion_lock); |
93bd25bb | 5507 | |
5262f567 JA |
5508 | /* |
5509 | * sqe->off holds how many events that need to occur for this | |
93bd25bb JA |
5510 | * timeout event to be satisfied. If it isn't set, then this is |
5511 | * a pure timeout request, sequence isn't used. | |
5262f567 | 5512 | */ |
8eb7e2d0 | 5513 | if (io_is_timeout_noseq(req)) { |
93bd25bb JA |
5514 | entry = ctx->timeout_list.prev; |
5515 | goto add; | |
5516 | } | |
5262f567 | 5517 | |
bfe68a22 PB |
5518 | tail = ctx->cached_cq_tail - atomic_read(&ctx->cq_timeouts); |
5519 | req->timeout.target_seq = tail + off; | |
5262f567 JA |
5520 | |
5521 | /* | |
5522 | * Insertion sort, ensuring the first entry in the list is always | |
5523 | * the one we need first. | |
5524 | */ | |
5262f567 | 5525 | list_for_each_prev(entry, &ctx->timeout_list) { |
135fcde8 PB |
5526 | struct io_kiocb *nxt = list_entry(entry, struct io_kiocb, |
5527 | timeout.list); | |
5262f567 | 5528 | |
8eb7e2d0 | 5529 | if (io_is_timeout_noseq(nxt)) |
93bd25bb | 5530 | continue; |
bfe68a22 PB |
5531 | /* nxt.seq is behind @tail, otherwise would've been completed */ |
5532 | if (off >= nxt->timeout.target_seq - tail) | |
5262f567 JA |
5533 | break; |
5534 | } | |
93bd25bb | 5535 | add: |
135fcde8 | 5536 | list_add(&req->timeout.list, entry); |
ad8a48ac JA |
5537 | data->timer.function = io_timeout_fn; |
5538 | hrtimer_start(&data->timer, timespec64_to_ktime(data->ts), data->mode); | |
5262f567 | 5539 | spin_unlock_irq(&ctx->completion_lock); |
5262f567 JA |
5540 | return 0; |
5541 | } | |
5262f567 | 5542 | |
62755e35 JA |
5543 | static bool io_cancel_cb(struct io_wq_work *work, void *data) |
5544 | { | |
5545 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); | |
5546 | ||
5547 | return req->user_data == (unsigned long) data; | |
5548 | } | |
5549 | ||
e977d6d3 | 5550 | static int io_async_cancel_one(struct io_ring_ctx *ctx, void *sqe_addr) |
62755e35 | 5551 | { |
62755e35 | 5552 | enum io_wq_cancel cancel_ret; |
62755e35 JA |
5553 | int ret = 0; |
5554 | ||
4f26bda1 | 5555 | cancel_ret = io_wq_cancel_cb(ctx->io_wq, io_cancel_cb, sqe_addr, false); |
62755e35 JA |
5556 | switch (cancel_ret) { |
5557 | case IO_WQ_CANCEL_OK: | |
5558 | ret = 0; | |
5559 | break; | |
5560 | case IO_WQ_CANCEL_RUNNING: | |
5561 | ret = -EALREADY; | |
5562 | break; | |
5563 | case IO_WQ_CANCEL_NOTFOUND: | |
5564 | ret = -ENOENT; | |
5565 | break; | |
5566 | } | |
5567 | ||
e977d6d3 JA |
5568 | return ret; |
5569 | } | |
5570 | ||
47f46768 JA |
5571 | static void io_async_find_and_cancel(struct io_ring_ctx *ctx, |
5572 | struct io_kiocb *req, __u64 sqe_addr, | |
014db007 | 5573 | int success_ret) |
47f46768 JA |
5574 | { |
5575 | unsigned long flags; | |
5576 | int ret; | |
5577 | ||
5578 | ret = io_async_cancel_one(ctx, (void *) (unsigned long) sqe_addr); | |
5579 | if (ret != -ENOENT) { | |
5580 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
5581 | goto done; | |
5582 | } | |
5583 | ||
5584 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
5585 | ret = io_timeout_cancel(ctx, sqe_addr); | |
5586 | if (ret != -ENOENT) | |
5587 | goto done; | |
5588 | ret = io_poll_cancel(ctx, sqe_addr); | |
5589 | done: | |
b0dd8a41 JA |
5590 | if (!ret) |
5591 | ret = success_ret; | |
47f46768 JA |
5592 | io_cqring_fill_event(req, ret); |
5593 | io_commit_cqring(ctx); | |
5594 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
5595 | io_cqring_ev_posted(ctx); | |
5596 | ||
4e88d6e7 JA |
5597 | if (ret < 0) |
5598 | req_set_fail_links(req); | |
014db007 | 5599 | io_put_req(req); |
47f46768 JA |
5600 | } |
5601 | ||
3529d8c2 JA |
5602 | static int io_async_cancel_prep(struct io_kiocb *req, |
5603 | const struct io_uring_sqe *sqe) | |
e977d6d3 | 5604 | { |
fbf23849 | 5605 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
e977d6d3 | 5606 | return -EINVAL; |
61710e43 DA |
5607 | if (unlikely(req->flags & (REQ_F_FIXED_FILE | REQ_F_BUFFER_SELECT))) |
5608 | return -EINVAL; | |
5609 | if (sqe->ioprio || sqe->off || sqe->len || sqe->cancel_flags) | |
e977d6d3 JA |
5610 | return -EINVAL; |
5611 | ||
fbf23849 JA |
5612 | req->cancel.addr = READ_ONCE(sqe->addr); |
5613 | return 0; | |
5614 | } | |
5615 | ||
014db007 | 5616 | static int io_async_cancel(struct io_kiocb *req) |
fbf23849 JA |
5617 | { |
5618 | struct io_ring_ctx *ctx = req->ctx; | |
fbf23849 | 5619 | |
014db007 | 5620 | io_async_find_and_cancel(ctx, req, req->cancel.addr, 0); |
5262f567 JA |
5621 | return 0; |
5622 | } | |
5623 | ||
05f3fb3c JA |
5624 | static int io_files_update_prep(struct io_kiocb *req, |
5625 | const struct io_uring_sqe *sqe) | |
5626 | { | |
6ca56f84 JA |
5627 | if (unlikely(req->ctx->flags & IORING_SETUP_SQPOLL)) |
5628 | return -EINVAL; | |
61710e43 DA |
5629 | if (unlikely(req->flags & (REQ_F_FIXED_FILE | REQ_F_BUFFER_SELECT))) |
5630 | return -EINVAL; | |
5631 | if (sqe->ioprio || sqe->rw_flags) | |
05f3fb3c JA |
5632 | return -EINVAL; |
5633 | ||
5634 | req->files_update.offset = READ_ONCE(sqe->off); | |
5635 | req->files_update.nr_args = READ_ONCE(sqe->len); | |
5636 | if (!req->files_update.nr_args) | |
5637 | return -EINVAL; | |
5638 | req->files_update.arg = READ_ONCE(sqe->addr); | |
5639 | return 0; | |
5640 | } | |
5641 | ||
229a7b63 JA |
5642 | static int io_files_update(struct io_kiocb *req, bool force_nonblock, |
5643 | struct io_comp_state *cs) | |
fbf23849 JA |
5644 | { |
5645 | struct io_ring_ctx *ctx = req->ctx; | |
05f3fb3c JA |
5646 | struct io_uring_files_update up; |
5647 | int ret; | |
fbf23849 | 5648 | |
f86cd20c | 5649 | if (force_nonblock) |
05f3fb3c | 5650 | return -EAGAIN; |
05f3fb3c JA |
5651 | |
5652 | up.offset = req->files_update.offset; | |
5653 | up.fds = req->files_update.arg; | |
5654 | ||
5655 | mutex_lock(&ctx->uring_lock); | |
5656 | ret = __io_sqe_files_update(ctx, &up, req->files_update.nr_args); | |
5657 | mutex_unlock(&ctx->uring_lock); | |
5658 | ||
5659 | if (ret < 0) | |
5660 | req_set_fail_links(req); | |
229a7b63 | 5661 | __io_req_complete(req, ret, 0, cs); |
5262f567 JA |
5662 | return 0; |
5663 | } | |
5664 | ||
bfe76559 | 5665 | static int io_req_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
f67676d1 | 5666 | { |
d625c6ee | 5667 | switch (req->opcode) { |
e781573e | 5668 | case IORING_OP_NOP: |
bfe76559 | 5669 | return 0; |
f67676d1 JA |
5670 | case IORING_OP_READV: |
5671 | case IORING_OP_READ_FIXED: | |
3a6820f2 | 5672 | case IORING_OP_READ: |
bfe76559 | 5673 | return io_read_prep(req, sqe); |
f67676d1 JA |
5674 | case IORING_OP_WRITEV: |
5675 | case IORING_OP_WRITE_FIXED: | |
3a6820f2 | 5676 | case IORING_OP_WRITE: |
bfe76559 | 5677 | return io_write_prep(req, sqe); |
0969e783 | 5678 | case IORING_OP_POLL_ADD: |
bfe76559 | 5679 | return io_poll_add_prep(req, sqe); |
0969e783 | 5680 | case IORING_OP_POLL_REMOVE: |
bfe76559 | 5681 | return io_poll_remove_prep(req, sqe); |
8ed8d3c3 | 5682 | case IORING_OP_FSYNC: |
bfe76559 | 5683 | return io_prep_fsync(req, sqe); |
8ed8d3c3 | 5684 | case IORING_OP_SYNC_FILE_RANGE: |
bfe76559 | 5685 | return io_prep_sfr(req, sqe); |
03b1230c | 5686 | case IORING_OP_SENDMSG: |
fddaface | 5687 | case IORING_OP_SEND: |
bfe76559 | 5688 | return io_sendmsg_prep(req, sqe); |
03b1230c | 5689 | case IORING_OP_RECVMSG: |
fddaface | 5690 | case IORING_OP_RECV: |
bfe76559 | 5691 | return io_recvmsg_prep(req, sqe); |
f499a021 | 5692 | case IORING_OP_CONNECT: |
bfe76559 | 5693 | return io_connect_prep(req, sqe); |
2d28390a | 5694 | case IORING_OP_TIMEOUT: |
bfe76559 | 5695 | return io_timeout_prep(req, sqe, false); |
b29472ee | 5696 | case IORING_OP_TIMEOUT_REMOVE: |
bfe76559 | 5697 | return io_timeout_remove_prep(req, sqe); |
fbf23849 | 5698 | case IORING_OP_ASYNC_CANCEL: |
bfe76559 | 5699 | return io_async_cancel_prep(req, sqe); |
2d28390a | 5700 | case IORING_OP_LINK_TIMEOUT: |
bfe76559 | 5701 | return io_timeout_prep(req, sqe, true); |
8ed8d3c3 | 5702 | case IORING_OP_ACCEPT: |
bfe76559 | 5703 | return io_accept_prep(req, sqe); |
d63d1b5e | 5704 | case IORING_OP_FALLOCATE: |
bfe76559 | 5705 | return io_fallocate_prep(req, sqe); |
15b71abe | 5706 | case IORING_OP_OPENAT: |
bfe76559 | 5707 | return io_openat_prep(req, sqe); |
b5dba59e | 5708 | case IORING_OP_CLOSE: |
bfe76559 | 5709 | return io_close_prep(req, sqe); |
05f3fb3c | 5710 | case IORING_OP_FILES_UPDATE: |
bfe76559 | 5711 | return io_files_update_prep(req, sqe); |
eddc7ef5 | 5712 | case IORING_OP_STATX: |
bfe76559 | 5713 | return io_statx_prep(req, sqe); |
4840e418 | 5714 | case IORING_OP_FADVISE: |
bfe76559 | 5715 | return io_fadvise_prep(req, sqe); |
c1ca757b | 5716 | case IORING_OP_MADVISE: |
bfe76559 | 5717 | return io_madvise_prep(req, sqe); |
cebdb986 | 5718 | case IORING_OP_OPENAT2: |
bfe76559 | 5719 | return io_openat2_prep(req, sqe); |
3e4827b0 | 5720 | case IORING_OP_EPOLL_CTL: |
bfe76559 | 5721 | return io_epoll_ctl_prep(req, sqe); |
7d67af2c | 5722 | case IORING_OP_SPLICE: |
bfe76559 | 5723 | return io_splice_prep(req, sqe); |
ddf0322d | 5724 | case IORING_OP_PROVIDE_BUFFERS: |
bfe76559 | 5725 | return io_provide_buffers_prep(req, sqe); |
067524e9 | 5726 | case IORING_OP_REMOVE_BUFFERS: |
bfe76559 | 5727 | return io_remove_buffers_prep(req, sqe); |
f2a8d5c7 | 5728 | case IORING_OP_TEE: |
bfe76559 | 5729 | return io_tee_prep(req, sqe); |
f67676d1 JA |
5730 | } |
5731 | ||
bfe76559 PB |
5732 | printk_once(KERN_WARNING "io_uring: unhandled opcode %d\n", |
5733 | req->opcode); | |
5734 | return-EINVAL; | |
5735 | } | |
5736 | ||
5737 | static int io_req_defer_prep(struct io_kiocb *req, | |
5738 | const struct io_uring_sqe *sqe) | |
5739 | { | |
bfe76559 PB |
5740 | if (!sqe) |
5741 | return 0; | |
5742 | if (io_alloc_async_data(req)) | |
5743 | return -EAGAIN; | |
bfe76559 | 5744 | return io_req_prep(req, sqe); |
f67676d1 JA |
5745 | } |
5746 | ||
9cf7c104 PB |
5747 | static u32 io_get_sequence(struct io_kiocb *req) |
5748 | { | |
5749 | struct io_kiocb *pos; | |
5750 | struct io_ring_ctx *ctx = req->ctx; | |
5751 | u32 total_submitted, nr_reqs = 1; | |
5752 | ||
5753 | if (req->flags & REQ_F_LINK_HEAD) | |
5754 | list_for_each_entry(pos, &req->link_list, link_list) | |
5755 | nr_reqs++; | |
5756 | ||
5757 | total_submitted = ctx->cached_sq_head - ctx->cached_sq_dropped; | |
5758 | return total_submitted - nr_reqs; | |
5759 | } | |
5760 | ||
3529d8c2 | 5761 | static int io_req_defer(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
de0617e4 | 5762 | { |
a197f664 | 5763 | struct io_ring_ctx *ctx = req->ctx; |
27dc8338 | 5764 | struct io_defer_entry *de; |
f67676d1 | 5765 | int ret; |
9cf7c104 | 5766 | u32 seq; |
de0617e4 | 5767 | |
9d858b21 | 5768 | /* Still need defer if there is pending req in defer list. */ |
9cf7c104 PB |
5769 | if (likely(list_empty_careful(&ctx->defer_list) && |
5770 | !(req->flags & REQ_F_IO_DRAIN))) | |
5771 | return 0; | |
5772 | ||
5773 | seq = io_get_sequence(req); | |
5774 | /* Still a chance to pass the sequence check */ | |
5775 | if (!req_need_defer(req, seq) && list_empty_careful(&ctx->defer_list)) | |
de0617e4 JA |
5776 | return 0; |
5777 | ||
e8c2bc1f | 5778 | if (!req->async_data) { |
650b5481 | 5779 | ret = io_req_defer_prep(req, sqe); |
327d6d96 | 5780 | if (ret) |
650b5481 PB |
5781 | return ret; |
5782 | } | |
cbdcb435 | 5783 | io_prep_async_link(req); |
27dc8338 PB |
5784 | de = kmalloc(sizeof(*de), GFP_KERNEL); |
5785 | if (!de) | |
5786 | return -ENOMEM; | |
2d28390a | 5787 | |
de0617e4 | 5788 | spin_lock_irq(&ctx->completion_lock); |
9cf7c104 | 5789 | if (!req_need_defer(req, seq) && list_empty(&ctx->defer_list)) { |
de0617e4 | 5790 | spin_unlock_irq(&ctx->completion_lock); |
27dc8338 | 5791 | kfree(de); |
ae34817b PB |
5792 | io_queue_async_work(req); |
5793 | return -EIOCBQUEUED; | |
de0617e4 JA |
5794 | } |
5795 | ||
915967f6 | 5796 | trace_io_uring_defer(ctx, req, req->user_data); |
27dc8338 | 5797 | de->req = req; |
9cf7c104 | 5798 | de->seq = seq; |
27dc8338 | 5799 | list_add_tail(&de->list, &ctx->defer_list); |
de0617e4 JA |
5800 | spin_unlock_irq(&ctx->completion_lock); |
5801 | return -EIOCBQUEUED; | |
5802 | } | |
5803 | ||
f573d384 | 5804 | static void io_req_drop_files(struct io_kiocb *req) |
99bc4c38 | 5805 | { |
f573d384 JA |
5806 | struct io_ring_ctx *ctx = req->ctx; |
5807 | unsigned long flags; | |
5808 | ||
5809 | spin_lock_irqsave(&ctx->inflight_lock, flags); | |
5810 | list_del(&req->inflight_entry); | |
5811 | if (waitqueue_active(&ctx->inflight_wait)) | |
5812 | wake_up(&ctx->inflight_wait); | |
5813 | spin_unlock_irqrestore(&ctx->inflight_lock, flags); | |
5814 | req->flags &= ~REQ_F_INFLIGHT; | |
98447d65 JA |
5815 | put_files_struct(req->work.identity->files); |
5816 | put_nsproxy(req->work.identity->nsproxy); | |
dfead8a8 | 5817 | req->work.flags &= ~IO_WQ_WORK_FILES; |
f573d384 | 5818 | } |
99bc4c38 | 5819 | |
3ca405eb | 5820 | static void __io_clean_op(struct io_kiocb *req) |
99bc4c38 | 5821 | { |
0e1b6fe3 PB |
5822 | if (req->flags & REQ_F_BUFFER_SELECTED) { |
5823 | switch (req->opcode) { | |
5824 | case IORING_OP_READV: | |
5825 | case IORING_OP_READ_FIXED: | |
5826 | case IORING_OP_READ: | |
bcda7baa | 5827 | kfree((void *)(unsigned long)req->rw.addr); |
0e1b6fe3 PB |
5828 | break; |
5829 | case IORING_OP_RECVMSG: | |
5830 | case IORING_OP_RECV: | |
bcda7baa | 5831 | kfree(req->sr_msg.kbuf); |
0e1b6fe3 PB |
5832 | break; |
5833 | } | |
5834 | req->flags &= ~REQ_F_BUFFER_SELECTED; | |
99bc4c38 PB |
5835 | } |
5836 | ||
0e1b6fe3 PB |
5837 | if (req->flags & REQ_F_NEED_CLEANUP) { |
5838 | switch (req->opcode) { | |
5839 | case IORING_OP_READV: | |
5840 | case IORING_OP_READ_FIXED: | |
5841 | case IORING_OP_READ: | |
5842 | case IORING_OP_WRITEV: | |
5843 | case IORING_OP_WRITE_FIXED: | |
e8c2bc1f JA |
5844 | case IORING_OP_WRITE: { |
5845 | struct io_async_rw *io = req->async_data; | |
5846 | if (io->free_iovec) | |
5847 | kfree(io->free_iovec); | |
0e1b6fe3 | 5848 | break; |
e8c2bc1f | 5849 | } |
0e1b6fe3 | 5850 | case IORING_OP_RECVMSG: |
e8c2bc1f JA |
5851 | case IORING_OP_SENDMSG: { |
5852 | struct io_async_msghdr *io = req->async_data; | |
5853 | if (io->iov != io->fast_iov) | |
5854 | kfree(io->iov); | |
0e1b6fe3 | 5855 | break; |
e8c2bc1f | 5856 | } |
0e1b6fe3 PB |
5857 | case IORING_OP_SPLICE: |
5858 | case IORING_OP_TEE: | |
5859 | io_put_file(req, req->splice.file_in, | |
5860 | (req->splice.flags & SPLICE_F_FD_IN_FIXED)); | |
5861 | break; | |
f3cd4850 JA |
5862 | case IORING_OP_OPENAT: |
5863 | case IORING_OP_OPENAT2: | |
5864 | if (req->open.filename) | |
5865 | putname(req->open.filename); | |
5866 | break; | |
0e1b6fe3 PB |
5867 | } |
5868 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
99bc4c38 | 5869 | } |
bb175342 | 5870 | |
f573d384 JA |
5871 | if (req->flags & REQ_F_INFLIGHT) |
5872 | io_req_drop_files(req); | |
99bc4c38 PB |
5873 | } |
5874 | ||
c1379e24 PB |
5875 | static int io_issue_sqe(struct io_kiocb *req, bool force_nonblock, |
5876 | struct io_comp_state *cs) | |
2b188cc1 | 5877 | { |
a197f664 | 5878 | struct io_ring_ctx *ctx = req->ctx; |
d625c6ee | 5879 | int ret; |
2b188cc1 | 5880 | |
d625c6ee | 5881 | switch (req->opcode) { |
2b188cc1 | 5882 | case IORING_OP_NOP: |
229a7b63 | 5883 | ret = io_nop(req, cs); |
2b188cc1 JA |
5884 | break; |
5885 | case IORING_OP_READV: | |
edafccee | 5886 | case IORING_OP_READ_FIXED: |
3a6820f2 | 5887 | case IORING_OP_READ: |
a1d7c393 | 5888 | ret = io_read(req, force_nonblock, cs); |
edafccee | 5889 | break; |
3529d8c2 | 5890 | case IORING_OP_WRITEV: |
edafccee | 5891 | case IORING_OP_WRITE_FIXED: |
3a6820f2 | 5892 | case IORING_OP_WRITE: |
a1d7c393 | 5893 | ret = io_write(req, force_nonblock, cs); |
2b188cc1 | 5894 | break; |
c992fe29 | 5895 | case IORING_OP_FSYNC: |
014db007 | 5896 | ret = io_fsync(req, force_nonblock); |
c992fe29 | 5897 | break; |
221c5eb2 | 5898 | case IORING_OP_POLL_ADD: |
014db007 | 5899 | ret = io_poll_add(req); |
221c5eb2 JA |
5900 | break; |
5901 | case IORING_OP_POLL_REMOVE: | |
fc4df999 | 5902 | ret = io_poll_remove(req); |
221c5eb2 | 5903 | break; |
5d17b4a4 | 5904 | case IORING_OP_SYNC_FILE_RANGE: |
014db007 | 5905 | ret = io_sync_file_range(req, force_nonblock); |
5d17b4a4 | 5906 | break; |
0fa03c62 | 5907 | case IORING_OP_SENDMSG: |
062d04d7 PB |
5908 | ret = io_sendmsg(req, force_nonblock, cs); |
5909 | break; | |
fddaface | 5910 | case IORING_OP_SEND: |
062d04d7 | 5911 | ret = io_send(req, force_nonblock, cs); |
0fa03c62 | 5912 | break; |
aa1fa28f | 5913 | case IORING_OP_RECVMSG: |
062d04d7 PB |
5914 | ret = io_recvmsg(req, force_nonblock, cs); |
5915 | break; | |
fddaface | 5916 | case IORING_OP_RECV: |
062d04d7 | 5917 | ret = io_recv(req, force_nonblock, cs); |
aa1fa28f | 5918 | break; |
5262f567 | 5919 | case IORING_OP_TIMEOUT: |
fc4df999 | 5920 | ret = io_timeout(req); |
5262f567 | 5921 | break; |
11365043 | 5922 | case IORING_OP_TIMEOUT_REMOVE: |
fc4df999 | 5923 | ret = io_timeout_remove(req); |
11365043 | 5924 | break; |
17f2fe35 | 5925 | case IORING_OP_ACCEPT: |
229a7b63 | 5926 | ret = io_accept(req, force_nonblock, cs); |
17f2fe35 | 5927 | break; |
f8e85cf2 | 5928 | case IORING_OP_CONNECT: |
229a7b63 | 5929 | ret = io_connect(req, force_nonblock, cs); |
f8e85cf2 | 5930 | break; |
62755e35 | 5931 | case IORING_OP_ASYNC_CANCEL: |
014db007 | 5932 | ret = io_async_cancel(req); |
62755e35 | 5933 | break; |
d63d1b5e | 5934 | case IORING_OP_FALLOCATE: |
014db007 | 5935 | ret = io_fallocate(req, force_nonblock); |
d63d1b5e | 5936 | break; |
15b71abe | 5937 | case IORING_OP_OPENAT: |
014db007 | 5938 | ret = io_openat(req, force_nonblock); |
15b71abe | 5939 | break; |
b5dba59e | 5940 | case IORING_OP_CLOSE: |
229a7b63 | 5941 | ret = io_close(req, force_nonblock, cs); |
b5dba59e | 5942 | break; |
05f3fb3c | 5943 | case IORING_OP_FILES_UPDATE: |
229a7b63 | 5944 | ret = io_files_update(req, force_nonblock, cs); |
05f3fb3c | 5945 | break; |
eddc7ef5 | 5946 | case IORING_OP_STATX: |
014db007 | 5947 | ret = io_statx(req, force_nonblock); |
eddc7ef5 | 5948 | break; |
4840e418 | 5949 | case IORING_OP_FADVISE: |
014db007 | 5950 | ret = io_fadvise(req, force_nonblock); |
4840e418 | 5951 | break; |
c1ca757b | 5952 | case IORING_OP_MADVISE: |
014db007 | 5953 | ret = io_madvise(req, force_nonblock); |
c1ca757b | 5954 | break; |
cebdb986 | 5955 | case IORING_OP_OPENAT2: |
014db007 | 5956 | ret = io_openat2(req, force_nonblock); |
cebdb986 | 5957 | break; |
3e4827b0 | 5958 | case IORING_OP_EPOLL_CTL: |
229a7b63 | 5959 | ret = io_epoll_ctl(req, force_nonblock, cs); |
3e4827b0 | 5960 | break; |
7d67af2c | 5961 | case IORING_OP_SPLICE: |
014db007 | 5962 | ret = io_splice(req, force_nonblock); |
7d67af2c | 5963 | break; |
ddf0322d | 5964 | case IORING_OP_PROVIDE_BUFFERS: |
229a7b63 | 5965 | ret = io_provide_buffers(req, force_nonblock, cs); |
ddf0322d | 5966 | break; |
067524e9 | 5967 | case IORING_OP_REMOVE_BUFFERS: |
229a7b63 | 5968 | ret = io_remove_buffers(req, force_nonblock, cs); |
3e4827b0 | 5969 | break; |
f2a8d5c7 | 5970 | case IORING_OP_TEE: |
f2a8d5c7 PB |
5971 | ret = io_tee(req, force_nonblock); |
5972 | break; | |
2b188cc1 JA |
5973 | default: |
5974 | ret = -EINVAL; | |
5975 | break; | |
5976 | } | |
5977 | ||
def596e9 JA |
5978 | if (ret) |
5979 | return ret; | |
5980 | ||
b532576e JA |
5981 | /* If the op doesn't have a file, we're not polling for it */ |
5982 | if ((ctx->flags & IORING_SETUP_IOPOLL) && req->file) { | |
11ba820b JA |
5983 | const bool in_async = io_wq_current_is_worker(); |
5984 | ||
11ba820b JA |
5985 | /* workqueue context doesn't hold uring_lock, grab it now */ |
5986 | if (in_async) | |
5987 | mutex_lock(&ctx->uring_lock); | |
5988 | ||
def596e9 | 5989 | io_iopoll_req_issued(req); |
11ba820b JA |
5990 | |
5991 | if (in_async) | |
5992 | mutex_unlock(&ctx->uring_lock); | |
def596e9 JA |
5993 | } |
5994 | ||
5995 | return 0; | |
2b188cc1 JA |
5996 | } |
5997 | ||
f4db7182 | 5998 | static struct io_wq_work *io_wq_submit_work(struct io_wq_work *work) |
2b188cc1 JA |
5999 | { |
6000 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); | |
6df1db6b | 6001 | struct io_kiocb *timeout; |
561fb04a | 6002 | int ret = 0; |
2b188cc1 | 6003 | |
6df1db6b PB |
6004 | timeout = io_prep_linked_timeout(req); |
6005 | if (timeout) | |
6006 | io_queue_linked_timeout(timeout); | |
d4c81f38 | 6007 | |
0c9d5ccd JA |
6008 | /* if NO_CANCEL is set, we must still run the work */ |
6009 | if ((work->flags & (IO_WQ_WORK_CANCEL|IO_WQ_WORK_NO_CANCEL)) == | |
6010 | IO_WQ_WORK_CANCEL) { | |
561fb04a | 6011 | ret = -ECANCELED; |
0c9d5ccd | 6012 | } |
31b51510 | 6013 | |
561fb04a | 6014 | if (!ret) { |
561fb04a | 6015 | do { |
c1379e24 | 6016 | ret = io_issue_sqe(req, false, NULL); |
561fb04a JA |
6017 | /* |
6018 | * We can get EAGAIN for polled IO even though we're | |
6019 | * forcing a sync submission from here, since we can't | |
6020 | * wait for request slots on the block side. | |
6021 | */ | |
6022 | if (ret != -EAGAIN) | |
6023 | break; | |
6024 | cond_resched(); | |
6025 | } while (1); | |
6026 | } | |
31b51510 | 6027 | |
561fb04a | 6028 | if (ret) { |
4e88d6e7 | 6029 | req_set_fail_links(req); |
e1e16097 | 6030 | io_req_complete(req, ret); |
edafccee | 6031 | } |
2b188cc1 | 6032 | |
f4db7182 | 6033 | return io_steal_work(req); |
2b188cc1 JA |
6034 | } |
6035 | ||
65e19f54 JA |
6036 | static inline struct file *io_file_from_index(struct io_ring_ctx *ctx, |
6037 | int index) | |
6038 | { | |
6039 | struct fixed_file_table *table; | |
6040 | ||
05f3fb3c | 6041 | table = &ctx->file_data->table[index >> IORING_FILE_TABLE_SHIFT]; |
84695089 | 6042 | return table->files[index & IORING_FILE_TABLE_MASK]; |
65e19f54 JA |
6043 | } |
6044 | ||
8371adf5 PB |
6045 | static struct file *io_file_get(struct io_submit_state *state, |
6046 | struct io_kiocb *req, int fd, bool fixed) | |
09bb8394 | 6047 | { |
a197f664 | 6048 | struct io_ring_ctx *ctx = req->ctx; |
8da11c19 | 6049 | struct file *file; |
09bb8394 | 6050 | |
8da11c19 | 6051 | if (fixed) { |
479f517b | 6052 | if (unlikely((unsigned int)fd >= ctx->nr_user_files)) |
8371adf5 | 6053 | return NULL; |
b7620121 | 6054 | fd = array_index_nospec(fd, ctx->nr_user_files); |
8da11c19 | 6055 | file = io_file_from_index(ctx, fd); |
fd2206e4 | 6056 | if (file) { |
b2e96852 | 6057 | req->fixed_file_refs = &ctx->file_data->node->refs; |
fd2206e4 JA |
6058 | percpu_ref_get(req->fixed_file_refs); |
6059 | } | |
09bb8394 | 6060 | } else { |
c826bd7a | 6061 | trace_io_uring_file_get(ctx, fd); |
8da11c19 | 6062 | file = __io_file_get(state, fd); |
09bb8394 JA |
6063 | } |
6064 | ||
8371adf5 | 6065 | return file; |
09bb8394 JA |
6066 | } |
6067 | ||
8da11c19 | 6068 | static int io_req_set_file(struct io_submit_state *state, struct io_kiocb *req, |
63ff8223 | 6069 | int fd) |
8da11c19 | 6070 | { |
8da11c19 PB |
6071 | bool fixed; |
6072 | ||
63ff8223 | 6073 | fixed = (req->flags & REQ_F_FIXED_FILE) != 0; |
0cdaf760 | 6074 | if (unlikely(!fixed && io_async_submit(req->ctx))) |
8da11c19 PB |
6075 | return -EBADF; |
6076 | ||
8371adf5 PB |
6077 | req->file = io_file_get(state, req, fd, fixed); |
6078 | if (req->file || io_op_defs[req->opcode].needs_file_no_error) | |
f86cd20c | 6079 | return 0; |
8371adf5 | 6080 | return -EBADF; |
f56040b8 PB |
6081 | } |
6082 | ||
2665abfd | 6083 | static enum hrtimer_restart io_link_timeout_fn(struct hrtimer *timer) |
2b188cc1 | 6084 | { |
ad8a48ac JA |
6085 | struct io_timeout_data *data = container_of(timer, |
6086 | struct io_timeout_data, timer); | |
6087 | struct io_kiocb *req = data->req; | |
2665abfd JA |
6088 | struct io_ring_ctx *ctx = req->ctx; |
6089 | struct io_kiocb *prev = NULL; | |
6090 | unsigned long flags; | |
2665abfd JA |
6091 | |
6092 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
6093 | ||
6094 | /* | |
6095 | * We don't expect the list to be empty, that will only happen if we | |
6096 | * race with the completion of the linked work. | |
6097 | */ | |
4493233e PB |
6098 | if (!list_empty(&req->link_list)) { |
6099 | prev = list_entry(req->link_list.prev, struct io_kiocb, | |
6100 | link_list); | |
900fad45 | 6101 | if (refcount_inc_not_zero(&prev->refs)) |
4493233e | 6102 | list_del_init(&req->link_list); |
900fad45 | 6103 | else |
76a46e06 | 6104 | prev = NULL; |
2665abfd JA |
6105 | } |
6106 | ||
6107 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
6108 | ||
6109 | if (prev) { | |
4e88d6e7 | 6110 | req_set_fail_links(prev); |
014db007 | 6111 | io_async_find_and_cancel(ctx, req, prev->user_data, -ETIME); |
76a46e06 | 6112 | io_put_req(prev); |
47f46768 | 6113 | } else { |
e1e16097 | 6114 | io_req_complete(req, -ETIME); |
2665abfd | 6115 | } |
2665abfd JA |
6116 | return HRTIMER_NORESTART; |
6117 | } | |
6118 | ||
7271ef3a | 6119 | static void __io_queue_linked_timeout(struct io_kiocb *req) |
2665abfd | 6120 | { |
76a46e06 JA |
6121 | /* |
6122 | * If the list is now empty, then our linked request finished before | |
6123 | * we got a chance to setup the timer | |
6124 | */ | |
4493233e | 6125 | if (!list_empty(&req->link_list)) { |
e8c2bc1f | 6126 | struct io_timeout_data *data = req->async_data; |
94ae5e77 | 6127 | |
ad8a48ac JA |
6128 | data->timer.function = io_link_timeout_fn; |
6129 | hrtimer_start(&data->timer, timespec64_to_ktime(data->ts), | |
6130 | data->mode); | |
2665abfd | 6131 | } |
7271ef3a JA |
6132 | } |
6133 | ||
6134 | static void io_queue_linked_timeout(struct io_kiocb *req) | |
6135 | { | |
6136 | struct io_ring_ctx *ctx = req->ctx; | |
6137 | ||
6138 | spin_lock_irq(&ctx->completion_lock); | |
6139 | __io_queue_linked_timeout(req); | |
76a46e06 | 6140 | spin_unlock_irq(&ctx->completion_lock); |
2665abfd | 6141 | |
2665abfd | 6142 | /* drop submission reference */ |
76a46e06 JA |
6143 | io_put_req(req); |
6144 | } | |
2665abfd | 6145 | |
ad8a48ac | 6146 | static struct io_kiocb *io_prep_linked_timeout(struct io_kiocb *req) |
2665abfd JA |
6147 | { |
6148 | struct io_kiocb *nxt; | |
6149 | ||
dea3b49c | 6150 | if (!(req->flags & REQ_F_LINK_HEAD)) |
2665abfd | 6151 | return NULL; |
6df1db6b | 6152 | if (req->flags & REQ_F_LINK_TIMEOUT) |
d7718a9d | 6153 | return NULL; |
2665abfd | 6154 | |
4493233e PB |
6155 | nxt = list_first_entry_or_null(&req->link_list, struct io_kiocb, |
6156 | link_list); | |
d625c6ee | 6157 | if (!nxt || nxt->opcode != IORING_OP_LINK_TIMEOUT) |
76a46e06 | 6158 | return NULL; |
2665abfd | 6159 | |
900fad45 | 6160 | nxt->flags |= REQ_F_LTIMEOUT_ACTIVE; |
76a46e06 | 6161 | req->flags |= REQ_F_LINK_TIMEOUT; |
76a46e06 | 6162 | return nxt; |
2665abfd JA |
6163 | } |
6164 | ||
c1379e24 | 6165 | static void __io_queue_sqe(struct io_kiocb *req, struct io_comp_state *cs) |
2b188cc1 | 6166 | { |
4a0a7a18 | 6167 | struct io_kiocb *linked_timeout; |
193155c8 | 6168 | const struct cred *old_creds = NULL; |
e0c5c576 | 6169 | int ret; |
2b188cc1 | 6170 | |
4a0a7a18 JA |
6171 | again: |
6172 | linked_timeout = io_prep_linked_timeout(req); | |
6173 | ||
2e5aa6cb PB |
6174 | if ((req->flags & REQ_F_WORK_INITIALIZED) && |
6175 | (req->work.flags & IO_WQ_WORK_CREDS) && | |
98447d65 | 6176 | req->work.identity->creds != current_cred()) { |
193155c8 JA |
6177 | if (old_creds) |
6178 | revert_creds(old_creds); | |
98447d65 | 6179 | if (old_creds == req->work.identity->creds) |
193155c8 JA |
6180 | old_creds = NULL; /* restored original creds */ |
6181 | else | |
98447d65 | 6182 | old_creds = override_creds(req->work.identity->creds); |
193155c8 JA |
6183 | } |
6184 | ||
c1379e24 | 6185 | ret = io_issue_sqe(req, true, cs); |
491381ce JA |
6186 | |
6187 | /* | |
6188 | * We async punt it if the file wasn't marked NOWAIT, or if the file | |
6189 | * doesn't support non-blocking read/write attempts | |
6190 | */ | |
24c74678 | 6191 | if (ret == -EAGAIN && !(req->flags & REQ_F_NOWAIT)) { |
f063c547 | 6192 | if (!io_arm_poll_handler(req)) { |
f063c547 PB |
6193 | /* |
6194 | * Queued up for async execution, worker will release | |
6195 | * submit reference when the iocb is actually submitted. | |
6196 | */ | |
6197 | io_queue_async_work(req); | |
2b188cc1 | 6198 | } |
bbad27b2 | 6199 | |
f063c547 PB |
6200 | if (linked_timeout) |
6201 | io_queue_linked_timeout(linked_timeout); | |
0d63c148 PB |
6202 | } else if (likely(!ret)) { |
6203 | /* drop submission reference */ | |
6204 | req = io_put_req_find_next(req); | |
6205 | if (linked_timeout) | |
6206 | io_queue_linked_timeout(linked_timeout); | |
e65ef56d | 6207 | |
0d63c148 PB |
6208 | if (req) { |
6209 | if (!(req->flags & REQ_F_FORCE_ASYNC)) | |
6210 | goto again; | |
6211 | io_queue_async_work(req); | |
6212 | } | |
6213 | } else { | |
652532ad PB |
6214 | /* un-prep timeout, so it'll be killed as any other linked */ |
6215 | req->flags &= ~REQ_F_LINK_TIMEOUT; | |
4e88d6e7 | 6216 | req_set_fail_links(req); |
e65ef56d | 6217 | io_put_req(req); |
e1e16097 | 6218 | io_req_complete(req, ret); |
9e645e11 | 6219 | } |
652532ad | 6220 | |
193155c8 JA |
6221 | if (old_creds) |
6222 | revert_creds(old_creds); | |
2b188cc1 JA |
6223 | } |
6224 | ||
f13fad7b JA |
6225 | static void io_queue_sqe(struct io_kiocb *req, const struct io_uring_sqe *sqe, |
6226 | struct io_comp_state *cs) | |
4fe2c963 JL |
6227 | { |
6228 | int ret; | |
6229 | ||
3529d8c2 | 6230 | ret = io_req_defer(req, sqe); |
4fe2c963 JL |
6231 | if (ret) { |
6232 | if (ret != -EIOCBQUEUED) { | |
1118591a | 6233 | fail_req: |
4e88d6e7 | 6234 | req_set_fail_links(req); |
e1e16097 JA |
6235 | io_put_req(req); |
6236 | io_req_complete(req, ret); | |
4fe2c963 | 6237 | } |
2550878f | 6238 | } else if (req->flags & REQ_F_FORCE_ASYNC) { |
e8c2bc1f | 6239 | if (!req->async_data) { |
bd2ab18a | 6240 | ret = io_req_defer_prep(req, sqe); |
327d6d96 | 6241 | if (unlikely(ret)) |
bd2ab18a PB |
6242 | goto fail_req; |
6243 | } | |
ce35a47a JA |
6244 | io_queue_async_work(req); |
6245 | } else { | |
c1379e24 PB |
6246 | if (sqe) { |
6247 | ret = io_req_prep(req, sqe); | |
6248 | if (unlikely(ret)) | |
6249 | goto fail_req; | |
6250 | } | |
6251 | __io_queue_sqe(req, cs); | |
ce35a47a | 6252 | } |
4fe2c963 JL |
6253 | } |
6254 | ||
f13fad7b JA |
6255 | static inline void io_queue_link_head(struct io_kiocb *req, |
6256 | struct io_comp_state *cs) | |
4fe2c963 | 6257 | { |
94ae5e77 | 6258 | if (unlikely(req->flags & REQ_F_FAIL_LINK)) { |
e1e16097 JA |
6259 | io_put_req(req); |
6260 | io_req_complete(req, -ECANCELED); | |
1b4a51b6 | 6261 | } else |
f13fad7b | 6262 | io_queue_sqe(req, NULL, cs); |
4fe2c963 JL |
6263 | } |
6264 | ||
1d4240cc | 6265 | static int io_submit_sqe(struct io_kiocb *req, const struct io_uring_sqe *sqe, |
f13fad7b | 6266 | struct io_kiocb **link, struct io_comp_state *cs) |
9e645e11 | 6267 | { |
a197f664 | 6268 | struct io_ring_ctx *ctx = req->ctx; |
ef4ff581 | 6269 | int ret; |
9e645e11 | 6270 | |
9e645e11 JA |
6271 | /* |
6272 | * If we already have a head request, queue this one for async | |
6273 | * submittal once the head completes. If we don't have a head but | |
6274 | * IOSQE_IO_LINK is set in the sqe, start a new head. This one will be | |
6275 | * submitted sync once the chain is complete. If none of those | |
6276 | * conditions are true (normal request), then just queue it. | |
6277 | */ | |
6278 | if (*link) { | |
9d76377f | 6279 | struct io_kiocb *head = *link; |
4e88d6e7 | 6280 | |
8cdf2193 PB |
6281 | /* |
6282 | * Taking sequential execution of a link, draining both sides | |
6283 | * of the link also fullfils IOSQE_IO_DRAIN semantics for all | |
6284 | * requests in the link. So, it drains the head and the | |
6285 | * next after the link request. The last one is done via | |
6286 | * drain_next flag to persist the effect across calls. | |
6287 | */ | |
ef4ff581 | 6288 | if (req->flags & REQ_F_IO_DRAIN) { |
711be031 PB |
6289 | head->flags |= REQ_F_IO_DRAIN; |
6290 | ctx->drain_next = 1; | |
6291 | } | |
3529d8c2 | 6292 | ret = io_req_defer_prep(req, sqe); |
327d6d96 | 6293 | if (unlikely(ret)) { |
4e88d6e7 | 6294 | /* fail even hard links since we don't submit */ |
9d76377f | 6295 | head->flags |= REQ_F_FAIL_LINK; |
1d4240cc | 6296 | return ret; |
2d28390a | 6297 | } |
9d76377f PB |
6298 | trace_io_uring_link(ctx, req, head); |
6299 | list_add_tail(&req->link_list, &head->link_list); | |
32fe525b PB |
6300 | |
6301 | /* last request of a link, enqueue the link */ | |
ef4ff581 | 6302 | if (!(req->flags & (REQ_F_LINK | REQ_F_HARDLINK))) { |
f13fad7b | 6303 | io_queue_link_head(head, cs); |
32fe525b PB |
6304 | *link = NULL; |
6305 | } | |
9e645e11 | 6306 | } else { |
711be031 PB |
6307 | if (unlikely(ctx->drain_next)) { |
6308 | req->flags |= REQ_F_IO_DRAIN; | |
ef4ff581 | 6309 | ctx->drain_next = 0; |
711be031 | 6310 | } |
ef4ff581 | 6311 | if (req->flags & (REQ_F_LINK | REQ_F_HARDLINK)) { |
dea3b49c | 6312 | req->flags |= REQ_F_LINK_HEAD; |
711be031 | 6313 | INIT_LIST_HEAD(&req->link_list); |
f1d96a8f | 6314 | |
711be031 | 6315 | ret = io_req_defer_prep(req, sqe); |
327d6d96 | 6316 | if (unlikely(ret)) |
711be031 PB |
6317 | req->flags |= REQ_F_FAIL_LINK; |
6318 | *link = req; | |
6319 | } else { | |
f13fad7b | 6320 | io_queue_sqe(req, sqe, cs); |
711be031 | 6321 | } |
9e645e11 | 6322 | } |
2e6e1fde | 6323 | |
1d4240cc | 6324 | return 0; |
9e645e11 JA |
6325 | } |
6326 | ||
9a56a232 JA |
6327 | /* |
6328 | * Batched submission is done, ensure local IO is flushed out. | |
6329 | */ | |
6330 | static void io_submit_state_end(struct io_submit_state *state) | |
6331 | { | |
f13fad7b JA |
6332 | if (!list_empty(&state->comp.list)) |
6333 | io_submit_flush_completions(&state->comp); | |
9a56a232 | 6334 | blk_finish_plug(&state->plug); |
9f13c35b | 6335 | io_state_file_put(state); |
2579f913 | 6336 | if (state->free_reqs) |
6c8a3134 | 6337 | kmem_cache_free_bulk(req_cachep, state->free_reqs, state->reqs); |
9a56a232 JA |
6338 | } |
6339 | ||
6340 | /* | |
6341 | * Start submission side cache. | |
6342 | */ | |
6343 | static void io_submit_state_start(struct io_submit_state *state, | |
013538bd | 6344 | struct io_ring_ctx *ctx, unsigned int max_ios) |
9a56a232 JA |
6345 | { |
6346 | blk_start_plug(&state->plug); | |
013538bd JA |
6347 | state->comp.nr = 0; |
6348 | INIT_LIST_HEAD(&state->comp.list); | |
6349 | state->comp.ctx = ctx; | |
2579f913 | 6350 | state->free_reqs = 0; |
9a56a232 JA |
6351 | state->file = NULL; |
6352 | state->ios_left = max_ios; | |
6353 | } | |
6354 | ||
2b188cc1 JA |
6355 | static void io_commit_sqring(struct io_ring_ctx *ctx) |
6356 | { | |
75b28aff | 6357 | struct io_rings *rings = ctx->rings; |
2b188cc1 | 6358 | |
caf582c6 PB |
6359 | /* |
6360 | * Ensure any loads from the SQEs are done at this point, | |
6361 | * since once we write the new head, the application could | |
6362 | * write new data to them. | |
6363 | */ | |
6364 | smp_store_release(&rings->sq.head, ctx->cached_sq_head); | |
2b188cc1 JA |
6365 | } |
6366 | ||
2b188cc1 | 6367 | /* |
3529d8c2 | 6368 | * Fetch an sqe, if one is available. Note that sqe_ptr will point to memory |
2b188cc1 JA |
6369 | * that is mapped by userspace. This means that care needs to be taken to |
6370 | * ensure that reads are stable, as we cannot rely on userspace always | |
6371 | * being a good citizen. If members of the sqe are validated and then later | |
6372 | * used, it's important that those reads are done through READ_ONCE() to | |
6373 | * prevent a re-load down the line. | |
6374 | */ | |
709b302f | 6375 | static const struct io_uring_sqe *io_get_sqe(struct io_ring_ctx *ctx) |
2b188cc1 | 6376 | { |
75b28aff | 6377 | u32 *sq_array = ctx->sq_array; |
2b188cc1 JA |
6378 | unsigned head; |
6379 | ||
6380 | /* | |
6381 | * The cached sq head (or cq tail) serves two purposes: | |
6382 | * | |
6383 | * 1) allows us to batch the cost of updating the user visible | |
6384 | * head updates. | |
6385 | * 2) allows the kernel side to track the head on its own, even | |
6386 | * though the application is the one updating it. | |
6387 | */ | |
ee7d46d9 | 6388 | head = READ_ONCE(sq_array[ctx->cached_sq_head & ctx->sq_mask]); |
709b302f PB |
6389 | if (likely(head < ctx->sq_entries)) |
6390 | return &ctx->sq_sqes[head]; | |
2b188cc1 JA |
6391 | |
6392 | /* drop invalid entries */ | |
498ccd9e | 6393 | ctx->cached_sq_dropped++; |
ee7d46d9 | 6394 | WRITE_ONCE(ctx->rings->sq_dropped, ctx->cached_sq_dropped); |
709b302f PB |
6395 | return NULL; |
6396 | } | |
6397 | ||
6398 | static inline void io_consume_sqe(struct io_ring_ctx *ctx) | |
6399 | { | |
6400 | ctx->cached_sq_head++; | |
2b188cc1 JA |
6401 | } |
6402 | ||
21b55dbc SG |
6403 | /* |
6404 | * Check SQE restrictions (opcode and flags). | |
6405 | * | |
6406 | * Returns 'true' if SQE is allowed, 'false' otherwise. | |
6407 | */ | |
6408 | static inline bool io_check_restriction(struct io_ring_ctx *ctx, | |
6409 | struct io_kiocb *req, | |
6410 | unsigned int sqe_flags) | |
6411 | { | |
6412 | if (!ctx->restricted) | |
6413 | return true; | |
6414 | ||
6415 | if (!test_bit(req->opcode, ctx->restrictions.sqe_op)) | |
6416 | return false; | |
6417 | ||
6418 | if ((sqe_flags & ctx->restrictions.sqe_flags_required) != | |
6419 | ctx->restrictions.sqe_flags_required) | |
6420 | return false; | |
6421 | ||
6422 | if (sqe_flags & ~(ctx->restrictions.sqe_flags_allowed | | |
6423 | ctx->restrictions.sqe_flags_required)) | |
6424 | return false; | |
6425 | ||
6426 | return true; | |
6427 | } | |
6428 | ||
ef4ff581 PB |
6429 | #define SQE_VALID_FLAGS (IOSQE_FIXED_FILE|IOSQE_IO_DRAIN|IOSQE_IO_LINK| \ |
6430 | IOSQE_IO_HARDLINK | IOSQE_ASYNC | \ | |
6431 | IOSQE_BUFFER_SELECT) | |
6432 | ||
6433 | static int io_init_req(struct io_ring_ctx *ctx, struct io_kiocb *req, | |
6434 | const struct io_uring_sqe *sqe, | |
0cdaf760 | 6435 | struct io_submit_state *state) |
0553b8bd | 6436 | { |
ef4ff581 | 6437 | unsigned int sqe_flags; |
71b547c0 | 6438 | int id, ret; |
ef4ff581 | 6439 | |
0553b8bd PB |
6440 | req->opcode = READ_ONCE(sqe->opcode); |
6441 | req->user_data = READ_ONCE(sqe->user_data); | |
e8c2bc1f | 6442 | req->async_data = NULL; |
0553b8bd PB |
6443 | req->file = NULL; |
6444 | req->ctx = ctx; | |
6445 | req->flags = 0; | |
6446 | /* one is dropped after submission, the other at completion */ | |
6447 | refcount_set(&req->refs, 2); | |
4dd2824d | 6448 | req->task = current; |
0553b8bd | 6449 | req->result = 0; |
ef4ff581 PB |
6450 | |
6451 | if (unlikely(req->opcode >= IORING_OP_LAST)) | |
6452 | return -EINVAL; | |
6453 | ||
9d8426a0 JA |
6454 | if (unlikely(io_sq_thread_acquire_mm(ctx, req))) |
6455 | return -EFAULT; | |
ef4ff581 PB |
6456 | |
6457 | sqe_flags = READ_ONCE(sqe->flags); | |
6458 | /* enforce forwards compatibility on users */ | |
6459 | if (unlikely(sqe_flags & ~SQE_VALID_FLAGS)) | |
6460 | return -EINVAL; | |
6461 | ||
21b55dbc SG |
6462 | if (unlikely(!io_check_restriction(ctx, req, sqe_flags))) |
6463 | return -EACCES; | |
6464 | ||
ef4ff581 PB |
6465 | if ((sqe_flags & IOSQE_BUFFER_SELECT) && |
6466 | !io_op_defs[req->opcode].buffer_select) | |
6467 | return -EOPNOTSUPP; | |
6468 | ||
6469 | id = READ_ONCE(sqe->personality); | |
6470 | if (id) { | |
1e6fa521 JA |
6471 | struct io_identity *iod; |
6472 | ||
1e6fa521 JA |
6473 | iod = idr_find(&ctx->personality_idr, id); |
6474 | if (unlikely(!iod)) | |
ef4ff581 | 6475 | return -EINVAL; |
1e6fa521 | 6476 | refcount_inc(&iod->count); |
ec99ca6c PB |
6477 | |
6478 | __io_req_init_async(req); | |
1e6fa521 JA |
6479 | get_cred(iod->creds); |
6480 | req->work.identity = iod; | |
dfead8a8 | 6481 | req->work.flags |= IO_WQ_WORK_CREDS; |
ef4ff581 PB |
6482 | } |
6483 | ||
6484 | /* same numerical values with corresponding REQ_F_*, safe to copy */ | |
c11368a5 | 6485 | req->flags |= sqe_flags; |
ef4ff581 | 6486 | |
63ff8223 JA |
6487 | if (!io_op_defs[req->opcode].needs_file) |
6488 | return 0; | |
6489 | ||
71b547c0 PB |
6490 | ret = io_req_set_file(state, req, READ_ONCE(sqe->fd)); |
6491 | state->ios_left--; | |
6492 | return ret; | |
0553b8bd PB |
6493 | } |
6494 | ||
0f212204 | 6495 | static int io_submit_sqes(struct io_ring_ctx *ctx, unsigned int nr) |
6c271ce2 | 6496 | { |
ac8691c4 | 6497 | struct io_submit_state state; |
9e645e11 | 6498 | struct io_kiocb *link = NULL; |
9e645e11 | 6499 | int i, submitted = 0; |
6c271ce2 | 6500 | |
c4a2ed72 | 6501 | /* if we have a backlog and couldn't flush it all, return BUSY */ |
ad3eb2c8 JA |
6502 | if (test_bit(0, &ctx->sq_check_overflow)) { |
6503 | if (!list_empty(&ctx->cq_overflow_list) && | |
e6c8aa9a | 6504 | !io_cqring_overflow_flush(ctx, false, NULL, NULL)) |
ad3eb2c8 JA |
6505 | return -EBUSY; |
6506 | } | |
6c271ce2 | 6507 | |
ee7d46d9 PB |
6508 | /* make sure SQ entry isn't read before tail */ |
6509 | nr = min3(nr, ctx->sq_entries, io_sqring_entries(ctx)); | |
9ef4f124 | 6510 | |
2b85edfc PB |
6511 | if (!percpu_ref_tryget_many(&ctx->refs, nr)) |
6512 | return -EAGAIN; | |
6c271ce2 | 6513 | |
d8a6df10 | 6514 | percpu_counter_add(¤t->io_uring->inflight, nr); |
faf7b51c | 6515 | refcount_add(nr, ¤t->usage); |
6c271ce2 | 6516 | |
013538bd | 6517 | io_submit_state_start(&state, ctx, nr); |
b14cca0c | 6518 | |
6c271ce2 | 6519 | for (i = 0; i < nr; i++) { |
3529d8c2 | 6520 | const struct io_uring_sqe *sqe; |
196be95c | 6521 | struct io_kiocb *req; |
1cb1edb2 | 6522 | int err; |
fb5ccc98 | 6523 | |
b1e50e54 PB |
6524 | sqe = io_get_sqe(ctx); |
6525 | if (unlikely(!sqe)) { | |
6526 | io_consume_sqe(ctx); | |
6527 | break; | |
6528 | } | |
ac8691c4 | 6529 | req = io_alloc_req(ctx, &state); |
196be95c PB |
6530 | if (unlikely(!req)) { |
6531 | if (!submitted) | |
6532 | submitted = -EAGAIN; | |
fb5ccc98 | 6533 | break; |
196be95c | 6534 | } |
709b302f | 6535 | io_consume_sqe(ctx); |
d3656344 JA |
6536 | /* will complete beyond this point, count as submitted */ |
6537 | submitted++; | |
6538 | ||
692d8363 | 6539 | err = io_init_req(ctx, req, sqe, &state); |
ef4ff581 | 6540 | if (unlikely(err)) { |
1cb1edb2 | 6541 | fail_req: |
e1e16097 JA |
6542 | io_put_req(req); |
6543 | io_req_complete(req, err); | |
196be95c PB |
6544 | break; |
6545 | } | |
fb5ccc98 | 6546 | |
354420f7 | 6547 | trace_io_uring_submit_sqe(ctx, req->opcode, req->user_data, |
0cdaf760 | 6548 | true, io_async_submit(ctx)); |
f13fad7b | 6549 | err = io_submit_sqe(req, sqe, &link, &state.comp); |
1d4240cc PB |
6550 | if (err) |
6551 | goto fail_req; | |
6c271ce2 JA |
6552 | } |
6553 | ||
9466f437 PB |
6554 | if (unlikely(submitted != nr)) { |
6555 | int ref_used = (submitted == -EAGAIN) ? 0 : submitted; | |
d8a6df10 JA |
6556 | struct io_uring_task *tctx = current->io_uring; |
6557 | int unused = nr - ref_used; | |
9466f437 | 6558 | |
d8a6df10 JA |
6559 | percpu_ref_put_many(&ctx->refs, unused); |
6560 | percpu_counter_sub(&tctx->inflight, unused); | |
6561 | put_task_struct_many(current, unused); | |
9466f437 | 6562 | } |
9e645e11 | 6563 | if (link) |
f13fad7b | 6564 | io_queue_link_head(link, &state.comp); |
ac8691c4 | 6565 | io_submit_state_end(&state); |
6c271ce2 | 6566 | |
ae9428ca PB |
6567 | /* Commit SQ ring head once we've consumed and submitted all SQEs */ |
6568 | io_commit_sqring(ctx); | |
6569 | ||
6c271ce2 JA |
6570 | return submitted; |
6571 | } | |
6572 | ||
23b3628e XW |
6573 | static inline void io_ring_set_wakeup_flag(struct io_ring_ctx *ctx) |
6574 | { | |
6575 | /* Tell userspace we may need a wakeup call */ | |
6576 | spin_lock_irq(&ctx->completion_lock); | |
6577 | ctx->rings->sq_flags |= IORING_SQ_NEED_WAKEUP; | |
6578 | spin_unlock_irq(&ctx->completion_lock); | |
6579 | } | |
6580 | ||
6581 | static inline void io_ring_clear_wakeup_flag(struct io_ring_ctx *ctx) | |
6582 | { | |
6583 | spin_lock_irq(&ctx->completion_lock); | |
6584 | ctx->rings->sq_flags &= ~IORING_SQ_NEED_WAKEUP; | |
6585 | spin_unlock_irq(&ctx->completion_lock); | |
6586 | } | |
6587 | ||
3f0e64d0 JA |
6588 | static int io_sq_wake_function(struct wait_queue_entry *wqe, unsigned mode, |
6589 | int sync, void *key) | |
6c271ce2 | 6590 | { |
3f0e64d0 JA |
6591 | struct io_ring_ctx *ctx = container_of(wqe, struct io_ring_ctx, sqo_wait_entry); |
6592 | int ret; | |
6593 | ||
6594 | ret = autoremove_wake_function(wqe, mode, sync, key); | |
6595 | if (ret) { | |
6596 | unsigned long flags; | |
6597 | ||
6598 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
6599 | ctx->rings->sq_flags &= ~IORING_SQ_NEED_WAKEUP; | |
6600 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
6601 | } | |
6602 | return ret; | |
6603 | } | |
6604 | ||
c8d1ba58 JA |
6605 | enum sq_ret { |
6606 | SQT_IDLE = 1, | |
6607 | SQT_SPIN = 2, | |
6608 | SQT_DID_WORK = 4, | |
6609 | }; | |
6610 | ||
6611 | static enum sq_ret __io_sq_thread(struct io_ring_ctx *ctx, | |
e95eee2d | 6612 | unsigned long start_jiffies, bool cap_entries) |
6c271ce2 | 6613 | { |
c8d1ba58 | 6614 | unsigned long timeout = start_jiffies + ctx->sq_thread_idle; |
534ca6d6 | 6615 | struct io_sq_data *sqd = ctx->sq_data; |
c8d1ba58 | 6616 | unsigned int to_submit; |
bdcd3eab | 6617 | int ret = 0; |
6c271ce2 | 6618 | |
c8d1ba58 JA |
6619 | again: |
6620 | if (!list_empty(&ctx->iopoll_list)) { | |
6621 | unsigned nr_events = 0; | |
a4c0b3de | 6622 | |
c8d1ba58 JA |
6623 | mutex_lock(&ctx->uring_lock); |
6624 | if (!list_empty(&ctx->iopoll_list) && !need_resched()) | |
6625 | io_do_iopoll(ctx, &nr_events, 0); | |
6626 | mutex_unlock(&ctx->uring_lock); | |
6627 | } | |
6c271ce2 | 6628 | |
c8d1ba58 | 6629 | to_submit = io_sqring_entries(ctx); |
6c271ce2 | 6630 | |
c8d1ba58 JA |
6631 | /* |
6632 | * If submit got -EBUSY, flag us as needing the application | |
6633 | * to enter the kernel to reap and flush events. | |
6634 | */ | |
6635 | if (!to_submit || ret == -EBUSY || need_resched()) { | |
6636 | /* | |
6637 | * Drop cur_mm before scheduling, we can't hold it for | |
6638 | * long periods (or over schedule()). Do this before | |
6639 | * adding ourselves to the waitqueue, as the unuse/drop | |
6640 | * may sleep. | |
6641 | */ | |
6642 | io_sq_thread_drop_mm(); | |
6c271ce2 | 6643 | |
c8d1ba58 JA |
6644 | /* |
6645 | * We're polling. If we're within the defined idle | |
6646 | * period, then let us spin without work before going | |
6647 | * to sleep. The exception is if we got EBUSY doing | |
6648 | * more IO, we should wait for the application to | |
6649 | * reap events and wake us up. | |
6650 | */ | |
6651 | if (!list_empty(&ctx->iopoll_list) || need_resched() || | |
6652 | (!time_after(jiffies, timeout) && ret != -EBUSY && | |
6653 | !percpu_ref_is_dying(&ctx->refs))) | |
6654 | return SQT_SPIN; | |
6c271ce2 | 6655 | |
534ca6d6 | 6656 | prepare_to_wait(&sqd->wait, &ctx->sqo_wait_entry, |
c8d1ba58 | 6657 | TASK_INTERRUPTIBLE); |
6c271ce2 | 6658 | |
c8d1ba58 JA |
6659 | /* |
6660 | * While doing polled IO, before going to sleep, we need | |
6661 | * to check if there are new reqs added to iopoll_list, | |
6662 | * it is because reqs may have been punted to io worker | |
6663 | * and will be added to iopoll_list later, hence check | |
6664 | * the iopoll_list again. | |
6665 | */ | |
6666 | if ((ctx->flags & IORING_SETUP_IOPOLL) && | |
6667 | !list_empty_careful(&ctx->iopoll_list)) { | |
534ca6d6 | 6668 | finish_wait(&sqd->wait, &ctx->sqo_wait_entry); |
c8d1ba58 | 6669 | goto again; |
6c271ce2 JA |
6670 | } |
6671 | ||
fb5ccc98 | 6672 | to_submit = io_sqring_entries(ctx); |
c8d1ba58 JA |
6673 | if (!to_submit || ret == -EBUSY) |
6674 | return SQT_IDLE; | |
6675 | } | |
c1edbf5f | 6676 | |
534ca6d6 | 6677 | finish_wait(&sqd->wait, &ctx->sqo_wait_entry); |
c8d1ba58 | 6678 | io_ring_clear_wakeup_flag(ctx); |
7143b5ac | 6679 | |
e95eee2d JA |
6680 | /* if we're handling multiple rings, cap submit size for fairness */ |
6681 | if (cap_entries && to_submit > 8) | |
6682 | to_submit = 8; | |
6683 | ||
c8d1ba58 JA |
6684 | mutex_lock(&ctx->uring_lock); |
6685 | if (likely(!percpu_ref_is_dying(&ctx->refs))) | |
6686 | ret = io_submit_sqes(ctx, to_submit); | |
6687 | mutex_unlock(&ctx->uring_lock); | |
90554200 JA |
6688 | |
6689 | if (!io_sqring_full(ctx) && wq_has_sleeper(&ctx->sqo_sq_wait)) | |
6690 | wake_up(&ctx->sqo_sq_wait); | |
6691 | ||
c8d1ba58 JA |
6692 | return SQT_DID_WORK; |
6693 | } | |
6c271ce2 | 6694 | |
69fb2131 JA |
6695 | static void io_sqd_init_new(struct io_sq_data *sqd) |
6696 | { | |
6697 | struct io_ring_ctx *ctx; | |
6698 | ||
6699 | while (!list_empty(&sqd->ctx_new_list)) { | |
6700 | ctx = list_first_entry(&sqd->ctx_new_list, struct io_ring_ctx, sqd_list); | |
6701 | init_wait(&ctx->sqo_wait_entry); | |
6702 | ctx->sqo_wait_entry.func = io_sq_wake_function; | |
6703 | list_move_tail(&ctx->sqd_list, &sqd->ctx_list); | |
6704 | complete(&ctx->sq_thread_comp); | |
6705 | } | |
6706 | } | |
6707 | ||
c8d1ba58 JA |
6708 | static int io_sq_thread(void *data) |
6709 | { | |
91d8f519 | 6710 | struct cgroup_subsys_state *cur_css = NULL; |
69fb2131 JA |
6711 | const struct cred *old_cred = NULL; |
6712 | struct io_sq_data *sqd = data; | |
6713 | struct io_ring_ctx *ctx; | |
c8d1ba58 | 6714 | unsigned long start_jiffies; |
6c271ce2 | 6715 | |
69fb2131 JA |
6716 | start_jiffies = jiffies; |
6717 | while (!kthread_should_stop()) { | |
6718 | enum sq_ret ret = 0; | |
e95eee2d | 6719 | bool cap_entries; |
c1edbf5f JA |
6720 | |
6721 | /* | |
69fb2131 JA |
6722 | * Any changes to the sqd lists are synchronized through the |
6723 | * kthread parking. This synchronizes the thread vs users, | |
6724 | * the users are synchronized on the sqd->ctx_lock. | |
c1edbf5f | 6725 | */ |
69fb2131 JA |
6726 | if (kthread_should_park()) |
6727 | kthread_parkme(); | |
7143b5ac | 6728 | |
69fb2131 JA |
6729 | if (unlikely(!list_empty(&sqd->ctx_new_list))) |
6730 | io_sqd_init_new(sqd); | |
6c271ce2 | 6731 | |
e95eee2d | 6732 | cap_entries = !list_is_singular(&sqd->ctx_list); |
6c271ce2 | 6733 | |
69fb2131 JA |
6734 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) { |
6735 | if (current->cred != ctx->creds) { | |
6736 | if (old_cred) | |
6737 | revert_creds(old_cred); | |
6738 | old_cred = override_creds(ctx->creds); | |
bdcd3eab | 6739 | } |
91d8f519 | 6740 | io_sq_thread_associate_blkcg(ctx, &cur_css); |
4ea33a97 JA |
6741 | #ifdef CONFIG_AUDIT |
6742 | current->loginuid = ctx->loginuid; | |
6743 | current->sessionid = ctx->sessionid; | |
6744 | #endif | |
bdcd3eab | 6745 | |
e95eee2d | 6746 | ret |= __io_sq_thread(ctx, start_jiffies, cap_entries); |
6c271ce2 | 6747 | |
69fb2131 JA |
6748 | io_sq_thread_drop_mm(); |
6749 | } | |
6c271ce2 | 6750 | |
69fb2131 | 6751 | if (ret & SQT_SPIN) { |
c8d1ba58 JA |
6752 | io_run_task_work(); |
6753 | cond_resched(); | |
69fb2131 JA |
6754 | } else if (ret == SQT_IDLE) { |
6755 | if (kthread_should_park()) | |
6c271ce2 | 6756 | continue; |
69fb2131 JA |
6757 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) |
6758 | io_ring_set_wakeup_flag(ctx); | |
6759 | schedule(); | |
6760 | start_jiffies = jiffies; | |
6761 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) | |
6762 | io_ring_clear_wakeup_flag(ctx); | |
6c271ce2 | 6763 | } |
6c271ce2 JA |
6764 | } |
6765 | ||
4c6e277c | 6766 | io_run_task_work(); |
b41e9852 | 6767 | |
91d8f519 DZ |
6768 | if (cur_css) |
6769 | io_sq_thread_unassociate_blkcg(); | |
69fb2131 JA |
6770 | if (old_cred) |
6771 | revert_creds(old_cred); | |
06058632 | 6772 | |
2bbcd6d3 | 6773 | kthread_parkme(); |
06058632 | 6774 | |
6c271ce2 JA |
6775 | return 0; |
6776 | } | |
6777 | ||
bda52162 JA |
6778 | struct io_wait_queue { |
6779 | struct wait_queue_entry wq; | |
6780 | struct io_ring_ctx *ctx; | |
6781 | unsigned to_wait; | |
6782 | unsigned nr_timeouts; | |
6783 | }; | |
6784 | ||
1d7bb1d5 | 6785 | static inline bool io_should_wake(struct io_wait_queue *iowq, bool noflush) |
bda52162 JA |
6786 | { |
6787 | struct io_ring_ctx *ctx = iowq->ctx; | |
6788 | ||
6789 | /* | |
d195a66e | 6790 | * Wake up if we have enough events, or if a timeout occurred since we |
bda52162 JA |
6791 | * started waiting. For timeouts, we always want to return to userspace, |
6792 | * regardless of event count. | |
6793 | */ | |
1d7bb1d5 | 6794 | return io_cqring_events(ctx, noflush) >= iowq->to_wait || |
bda52162 JA |
6795 | atomic_read(&ctx->cq_timeouts) != iowq->nr_timeouts; |
6796 | } | |
6797 | ||
6798 | static int io_wake_function(struct wait_queue_entry *curr, unsigned int mode, | |
6799 | int wake_flags, void *key) | |
6800 | { | |
6801 | struct io_wait_queue *iowq = container_of(curr, struct io_wait_queue, | |
6802 | wq); | |
6803 | ||
1d7bb1d5 JA |
6804 | /* use noflush == true, as we can't safely rely on locking context */ |
6805 | if (!io_should_wake(iowq, true)) | |
bda52162 JA |
6806 | return -1; |
6807 | ||
6808 | return autoremove_wake_function(curr, mode, wake_flags, key); | |
6809 | } | |
6810 | ||
af9c1a44 JA |
6811 | static int io_run_task_work_sig(void) |
6812 | { | |
6813 | if (io_run_task_work()) | |
6814 | return 1; | |
6815 | if (!signal_pending(current)) | |
6816 | return 0; | |
6817 | if (current->jobctl & JOBCTL_TASK_WORK) { | |
6818 | spin_lock_irq(¤t->sighand->siglock); | |
6819 | current->jobctl &= ~JOBCTL_TASK_WORK; | |
6820 | recalc_sigpending(); | |
6821 | spin_unlock_irq(¤t->sighand->siglock); | |
6822 | return 1; | |
6823 | } | |
6824 | return -EINTR; | |
6825 | } | |
6826 | ||
2b188cc1 JA |
6827 | /* |
6828 | * Wait until events become available, if we don't already have some. The | |
6829 | * application must reap them itself, as they reside on the shared cq ring. | |
6830 | */ | |
6831 | static int io_cqring_wait(struct io_ring_ctx *ctx, int min_events, | |
6832 | const sigset_t __user *sig, size_t sigsz) | |
6833 | { | |
bda52162 JA |
6834 | struct io_wait_queue iowq = { |
6835 | .wq = { | |
6836 | .private = current, | |
6837 | .func = io_wake_function, | |
6838 | .entry = LIST_HEAD_INIT(iowq.wq.entry), | |
6839 | }, | |
6840 | .ctx = ctx, | |
6841 | .to_wait = min_events, | |
6842 | }; | |
75b28aff | 6843 | struct io_rings *rings = ctx->rings; |
e9ffa5c2 | 6844 | int ret = 0; |
2b188cc1 | 6845 | |
b41e9852 JA |
6846 | do { |
6847 | if (io_cqring_events(ctx, false) >= min_events) | |
6848 | return 0; | |
4c6e277c | 6849 | if (!io_run_task_work()) |
b41e9852 | 6850 | break; |
b41e9852 | 6851 | } while (1); |
2b188cc1 JA |
6852 | |
6853 | if (sig) { | |
9e75ad5d AB |
6854 | #ifdef CONFIG_COMPAT |
6855 | if (in_compat_syscall()) | |
6856 | ret = set_compat_user_sigmask((const compat_sigset_t __user *)sig, | |
b772434b | 6857 | sigsz); |
9e75ad5d AB |
6858 | else |
6859 | #endif | |
b772434b | 6860 | ret = set_user_sigmask(sig, sigsz); |
9e75ad5d | 6861 | |
2b188cc1 JA |
6862 | if (ret) |
6863 | return ret; | |
6864 | } | |
6865 | ||
bda52162 | 6866 | iowq.nr_timeouts = atomic_read(&ctx->cq_timeouts); |
c826bd7a | 6867 | trace_io_uring_cqring_wait(ctx, min_events); |
bda52162 JA |
6868 | do { |
6869 | prepare_to_wait_exclusive(&ctx->wait, &iowq.wq, | |
6870 | TASK_INTERRUPTIBLE); | |
ce593a6c | 6871 | /* make sure we run task_work before checking for signals */ |
af9c1a44 JA |
6872 | ret = io_run_task_work_sig(); |
6873 | if (ret > 0) | |
4c6e277c | 6874 | continue; |
af9c1a44 | 6875 | else if (ret < 0) |
bda52162 | 6876 | break; |
ce593a6c JA |
6877 | if (io_should_wake(&iowq, false)) |
6878 | break; | |
6879 | schedule(); | |
bda52162 JA |
6880 | } while (1); |
6881 | finish_wait(&ctx->wait, &iowq.wq); | |
6882 | ||
b7db41c9 | 6883 | restore_saved_sigmask_unless(ret == -EINTR); |
2b188cc1 | 6884 | |
75b28aff | 6885 | return READ_ONCE(rings->cq.head) == READ_ONCE(rings->cq.tail) ? ret : 0; |
2b188cc1 JA |
6886 | } |
6887 | ||
6b06314c JA |
6888 | static void __io_sqe_files_unregister(struct io_ring_ctx *ctx) |
6889 | { | |
6890 | #if defined(CONFIG_UNIX) | |
6891 | if (ctx->ring_sock) { | |
6892 | struct sock *sock = ctx->ring_sock->sk; | |
6893 | struct sk_buff *skb; | |
6894 | ||
6895 | while ((skb = skb_dequeue(&sock->sk_receive_queue)) != NULL) | |
6896 | kfree_skb(skb); | |
6897 | } | |
6898 | #else | |
6899 | int i; | |
6900 | ||
65e19f54 JA |
6901 | for (i = 0; i < ctx->nr_user_files; i++) { |
6902 | struct file *file; | |
6903 | ||
6904 | file = io_file_from_index(ctx, i); | |
6905 | if (file) | |
6906 | fput(file); | |
6907 | } | |
6b06314c JA |
6908 | #endif |
6909 | } | |
6910 | ||
05f3fb3c JA |
6911 | static void io_file_ref_kill(struct percpu_ref *ref) |
6912 | { | |
6913 | struct fixed_file_data *data; | |
6914 | ||
6915 | data = container_of(ref, struct fixed_file_data, refs); | |
6916 | complete(&data->done); | |
6917 | } | |
6918 | ||
6b06314c JA |
6919 | static int io_sqe_files_unregister(struct io_ring_ctx *ctx) |
6920 | { | |
05f3fb3c | 6921 | struct fixed_file_data *data = ctx->file_data; |
05589553 | 6922 | struct fixed_file_ref_node *ref_node = NULL; |
65e19f54 JA |
6923 | unsigned nr_tables, i; |
6924 | ||
05f3fb3c | 6925 | if (!data) |
6b06314c JA |
6926 | return -ENXIO; |
6927 | ||
6a4d07cd | 6928 | spin_lock(&data->lock); |
05589553 XW |
6929 | if (!list_empty(&data->ref_list)) |
6930 | ref_node = list_first_entry(&data->ref_list, | |
6931 | struct fixed_file_ref_node, node); | |
6a4d07cd | 6932 | spin_unlock(&data->lock); |
05589553 XW |
6933 | if (ref_node) |
6934 | percpu_ref_kill(&ref_node->refs); | |
6935 | ||
6936 | percpu_ref_kill(&data->refs); | |
6937 | ||
6938 | /* wait for all refs nodes to complete */ | |
4a38aed2 | 6939 | flush_delayed_work(&ctx->file_put_work); |
2faf852d | 6940 | wait_for_completion(&data->done); |
05f3fb3c | 6941 | |
6b06314c | 6942 | __io_sqe_files_unregister(ctx); |
65e19f54 JA |
6943 | nr_tables = DIV_ROUND_UP(ctx->nr_user_files, IORING_MAX_FILES_TABLE); |
6944 | for (i = 0; i < nr_tables; i++) | |
05f3fb3c JA |
6945 | kfree(data->table[i].files); |
6946 | kfree(data->table); | |
05589553 XW |
6947 | percpu_ref_exit(&data->refs); |
6948 | kfree(data); | |
05f3fb3c | 6949 | ctx->file_data = NULL; |
6b06314c JA |
6950 | ctx->nr_user_files = 0; |
6951 | return 0; | |
6952 | } | |
6953 | ||
534ca6d6 | 6954 | static void io_put_sq_data(struct io_sq_data *sqd) |
6c271ce2 | 6955 | { |
534ca6d6 | 6956 | if (refcount_dec_and_test(&sqd->refs)) { |
2bbcd6d3 RP |
6957 | /* |
6958 | * The park is a bit of a work-around, without it we get | |
6959 | * warning spews on shutdown with SQPOLL set and affinity | |
6960 | * set to a single CPU. | |
6961 | */ | |
534ca6d6 JA |
6962 | if (sqd->thread) { |
6963 | kthread_park(sqd->thread); | |
6964 | kthread_stop(sqd->thread); | |
6965 | } | |
6966 | ||
6967 | kfree(sqd); | |
6968 | } | |
6969 | } | |
6970 | ||
aa06165d JA |
6971 | static struct io_sq_data *io_attach_sq_data(struct io_uring_params *p) |
6972 | { | |
6973 | struct io_ring_ctx *ctx_attach; | |
6974 | struct io_sq_data *sqd; | |
6975 | struct fd f; | |
6976 | ||
6977 | f = fdget(p->wq_fd); | |
6978 | if (!f.file) | |
6979 | return ERR_PTR(-ENXIO); | |
6980 | if (f.file->f_op != &io_uring_fops) { | |
6981 | fdput(f); | |
6982 | return ERR_PTR(-EINVAL); | |
6983 | } | |
6984 | ||
6985 | ctx_attach = f.file->private_data; | |
6986 | sqd = ctx_attach->sq_data; | |
6987 | if (!sqd) { | |
6988 | fdput(f); | |
6989 | return ERR_PTR(-EINVAL); | |
6990 | } | |
6991 | ||
6992 | refcount_inc(&sqd->refs); | |
6993 | fdput(f); | |
6994 | return sqd; | |
6995 | } | |
6996 | ||
534ca6d6 JA |
6997 | static struct io_sq_data *io_get_sq_data(struct io_uring_params *p) |
6998 | { | |
6999 | struct io_sq_data *sqd; | |
7000 | ||
aa06165d JA |
7001 | if (p->flags & IORING_SETUP_ATTACH_WQ) |
7002 | return io_attach_sq_data(p); | |
7003 | ||
534ca6d6 JA |
7004 | sqd = kzalloc(sizeof(*sqd), GFP_KERNEL); |
7005 | if (!sqd) | |
7006 | return ERR_PTR(-ENOMEM); | |
7007 | ||
7008 | refcount_set(&sqd->refs, 1); | |
69fb2131 JA |
7009 | INIT_LIST_HEAD(&sqd->ctx_list); |
7010 | INIT_LIST_HEAD(&sqd->ctx_new_list); | |
7011 | mutex_init(&sqd->ctx_lock); | |
7012 | mutex_init(&sqd->lock); | |
534ca6d6 JA |
7013 | init_waitqueue_head(&sqd->wait); |
7014 | return sqd; | |
7015 | } | |
7016 | ||
69fb2131 JA |
7017 | static void io_sq_thread_unpark(struct io_sq_data *sqd) |
7018 | __releases(&sqd->lock) | |
7019 | { | |
7020 | if (!sqd->thread) | |
7021 | return; | |
7022 | kthread_unpark(sqd->thread); | |
7023 | mutex_unlock(&sqd->lock); | |
7024 | } | |
7025 | ||
7026 | static void io_sq_thread_park(struct io_sq_data *sqd) | |
7027 | __acquires(&sqd->lock) | |
7028 | { | |
7029 | if (!sqd->thread) | |
7030 | return; | |
7031 | mutex_lock(&sqd->lock); | |
7032 | kthread_park(sqd->thread); | |
7033 | } | |
7034 | ||
534ca6d6 JA |
7035 | static void io_sq_thread_stop(struct io_ring_ctx *ctx) |
7036 | { | |
7037 | struct io_sq_data *sqd = ctx->sq_data; | |
7038 | ||
7039 | if (sqd) { | |
7040 | if (sqd->thread) { | |
7041 | /* | |
7042 | * We may arrive here from the error branch in | |
7043 | * io_sq_offload_create() where the kthread is created | |
7044 | * without being waked up, thus wake it up now to make | |
7045 | * sure the wait will complete. | |
7046 | */ | |
7047 | wake_up_process(sqd->thread); | |
7048 | wait_for_completion(&ctx->sq_thread_comp); | |
69fb2131 JA |
7049 | |
7050 | io_sq_thread_park(sqd); | |
7051 | } | |
7052 | ||
7053 | mutex_lock(&sqd->ctx_lock); | |
7054 | list_del(&ctx->sqd_list); | |
7055 | mutex_unlock(&sqd->ctx_lock); | |
7056 | ||
7057 | if (sqd->thread) { | |
7058 | finish_wait(&sqd->wait, &ctx->sqo_wait_entry); | |
7059 | io_sq_thread_unpark(sqd); | |
534ca6d6 JA |
7060 | } |
7061 | ||
7062 | io_put_sq_data(sqd); | |
7063 | ctx->sq_data = NULL; | |
6c271ce2 JA |
7064 | } |
7065 | } | |
7066 | ||
6b06314c JA |
7067 | static void io_finish_async(struct io_ring_ctx *ctx) |
7068 | { | |
6c271ce2 JA |
7069 | io_sq_thread_stop(ctx); |
7070 | ||
561fb04a JA |
7071 | if (ctx->io_wq) { |
7072 | io_wq_destroy(ctx->io_wq); | |
7073 | ctx->io_wq = NULL; | |
6b06314c JA |
7074 | } |
7075 | } | |
7076 | ||
7077 | #if defined(CONFIG_UNIX) | |
6b06314c JA |
7078 | /* |
7079 | * Ensure the UNIX gc is aware of our file set, so we are certain that | |
7080 | * the io_uring can be safely unregistered on process exit, even if we have | |
7081 | * loops in the file referencing. | |
7082 | */ | |
7083 | static int __io_sqe_files_scm(struct io_ring_ctx *ctx, int nr, int offset) | |
7084 | { | |
7085 | struct sock *sk = ctx->ring_sock->sk; | |
7086 | struct scm_fp_list *fpl; | |
7087 | struct sk_buff *skb; | |
08a45173 | 7088 | int i, nr_files; |
6b06314c | 7089 | |
6b06314c JA |
7090 | fpl = kzalloc(sizeof(*fpl), GFP_KERNEL); |
7091 | if (!fpl) | |
7092 | return -ENOMEM; | |
7093 | ||
7094 | skb = alloc_skb(0, GFP_KERNEL); | |
7095 | if (!skb) { | |
7096 | kfree(fpl); | |
7097 | return -ENOMEM; | |
7098 | } | |
7099 | ||
7100 | skb->sk = sk; | |
6b06314c | 7101 | |
08a45173 | 7102 | nr_files = 0; |
6b06314c JA |
7103 | fpl->user = get_uid(ctx->user); |
7104 | for (i = 0; i < nr; i++) { | |
65e19f54 JA |
7105 | struct file *file = io_file_from_index(ctx, i + offset); |
7106 | ||
7107 | if (!file) | |
08a45173 | 7108 | continue; |
65e19f54 | 7109 | fpl->fp[nr_files] = get_file(file); |
08a45173 JA |
7110 | unix_inflight(fpl->user, fpl->fp[nr_files]); |
7111 | nr_files++; | |
6b06314c JA |
7112 | } |
7113 | ||
08a45173 JA |
7114 | if (nr_files) { |
7115 | fpl->max = SCM_MAX_FD; | |
7116 | fpl->count = nr_files; | |
7117 | UNIXCB(skb).fp = fpl; | |
05f3fb3c | 7118 | skb->destructor = unix_destruct_scm; |
08a45173 JA |
7119 | refcount_add(skb->truesize, &sk->sk_wmem_alloc); |
7120 | skb_queue_head(&sk->sk_receive_queue, skb); | |
6b06314c | 7121 | |
08a45173 JA |
7122 | for (i = 0; i < nr_files; i++) |
7123 | fput(fpl->fp[i]); | |
7124 | } else { | |
7125 | kfree_skb(skb); | |
7126 | kfree(fpl); | |
7127 | } | |
6b06314c JA |
7128 | |
7129 | return 0; | |
7130 | } | |
7131 | ||
7132 | /* | |
7133 | * If UNIX sockets are enabled, fd passing can cause a reference cycle which | |
7134 | * causes regular reference counting to break down. We rely on the UNIX | |
7135 | * garbage collection to take care of this problem for us. | |
7136 | */ | |
7137 | static int io_sqe_files_scm(struct io_ring_ctx *ctx) | |
7138 | { | |
7139 | unsigned left, total; | |
7140 | int ret = 0; | |
7141 | ||
7142 | total = 0; | |
7143 | left = ctx->nr_user_files; | |
7144 | while (left) { | |
7145 | unsigned this_files = min_t(unsigned, left, SCM_MAX_FD); | |
6b06314c JA |
7146 | |
7147 | ret = __io_sqe_files_scm(ctx, this_files, total); | |
7148 | if (ret) | |
7149 | break; | |
7150 | left -= this_files; | |
7151 | total += this_files; | |
7152 | } | |
7153 | ||
7154 | if (!ret) | |
7155 | return 0; | |
7156 | ||
7157 | while (total < ctx->nr_user_files) { | |
65e19f54 JA |
7158 | struct file *file = io_file_from_index(ctx, total); |
7159 | ||
7160 | if (file) | |
7161 | fput(file); | |
6b06314c JA |
7162 | total++; |
7163 | } | |
7164 | ||
7165 | return ret; | |
7166 | } | |
7167 | #else | |
7168 | static int io_sqe_files_scm(struct io_ring_ctx *ctx) | |
7169 | { | |
7170 | return 0; | |
7171 | } | |
7172 | #endif | |
7173 | ||
5398ae69 PB |
7174 | static int io_sqe_alloc_file_tables(struct fixed_file_data *file_data, |
7175 | unsigned nr_tables, unsigned nr_files) | |
65e19f54 JA |
7176 | { |
7177 | int i; | |
7178 | ||
7179 | for (i = 0; i < nr_tables; i++) { | |
5398ae69 | 7180 | struct fixed_file_table *table = &file_data->table[i]; |
65e19f54 JA |
7181 | unsigned this_files; |
7182 | ||
7183 | this_files = min(nr_files, IORING_MAX_FILES_TABLE); | |
7184 | table->files = kcalloc(this_files, sizeof(struct file *), | |
7185 | GFP_KERNEL); | |
7186 | if (!table->files) | |
7187 | break; | |
7188 | nr_files -= this_files; | |
7189 | } | |
7190 | ||
7191 | if (i == nr_tables) | |
7192 | return 0; | |
7193 | ||
7194 | for (i = 0; i < nr_tables; i++) { | |
5398ae69 | 7195 | struct fixed_file_table *table = &file_data->table[i]; |
65e19f54 JA |
7196 | kfree(table->files); |
7197 | } | |
7198 | return 1; | |
7199 | } | |
7200 | ||
05f3fb3c JA |
7201 | static void io_ring_file_put(struct io_ring_ctx *ctx, struct file *file) |
7202 | { | |
7203 | #if defined(CONFIG_UNIX) | |
7204 | struct sock *sock = ctx->ring_sock->sk; | |
7205 | struct sk_buff_head list, *head = &sock->sk_receive_queue; | |
7206 | struct sk_buff *skb; | |
7207 | int i; | |
7208 | ||
7209 | __skb_queue_head_init(&list); | |
7210 | ||
7211 | /* | |
7212 | * Find the skb that holds this file in its SCM_RIGHTS. When found, | |
7213 | * remove this entry and rearrange the file array. | |
7214 | */ | |
7215 | skb = skb_dequeue(head); | |
7216 | while (skb) { | |
7217 | struct scm_fp_list *fp; | |
7218 | ||
7219 | fp = UNIXCB(skb).fp; | |
7220 | for (i = 0; i < fp->count; i++) { | |
7221 | int left; | |
7222 | ||
7223 | if (fp->fp[i] != file) | |
7224 | continue; | |
7225 | ||
7226 | unix_notinflight(fp->user, fp->fp[i]); | |
7227 | left = fp->count - 1 - i; | |
7228 | if (left) { | |
7229 | memmove(&fp->fp[i], &fp->fp[i + 1], | |
7230 | left * sizeof(struct file *)); | |
7231 | } | |
7232 | fp->count--; | |
7233 | if (!fp->count) { | |
7234 | kfree_skb(skb); | |
7235 | skb = NULL; | |
7236 | } else { | |
7237 | __skb_queue_tail(&list, skb); | |
7238 | } | |
7239 | fput(file); | |
7240 | file = NULL; | |
7241 | break; | |
7242 | } | |
7243 | ||
7244 | if (!file) | |
7245 | break; | |
7246 | ||
7247 | __skb_queue_tail(&list, skb); | |
7248 | ||
7249 | skb = skb_dequeue(head); | |
7250 | } | |
7251 | ||
7252 | if (skb_peek(&list)) { | |
7253 | spin_lock_irq(&head->lock); | |
7254 | while ((skb = __skb_dequeue(&list)) != NULL) | |
7255 | __skb_queue_tail(head, skb); | |
7256 | spin_unlock_irq(&head->lock); | |
7257 | } | |
7258 | #else | |
7259 | fput(file); | |
7260 | #endif | |
7261 | } | |
7262 | ||
7263 | struct io_file_put { | |
05589553 | 7264 | struct list_head list; |
05f3fb3c | 7265 | struct file *file; |
05f3fb3c JA |
7266 | }; |
7267 | ||
4a38aed2 | 7268 | static void __io_file_put_work(struct fixed_file_ref_node *ref_node) |
65e19f54 | 7269 | { |
4a38aed2 JA |
7270 | struct fixed_file_data *file_data = ref_node->file_data; |
7271 | struct io_ring_ctx *ctx = file_data->ctx; | |
05f3fb3c | 7272 | struct io_file_put *pfile, *tmp; |
05589553 XW |
7273 | |
7274 | list_for_each_entry_safe(pfile, tmp, &ref_node->file_list, list) { | |
6a4d07cd | 7275 | list_del(&pfile->list); |
05589553 XW |
7276 | io_ring_file_put(ctx, pfile->file); |
7277 | kfree(pfile); | |
65e19f54 | 7278 | } |
05589553 | 7279 | |
6a4d07cd JA |
7280 | spin_lock(&file_data->lock); |
7281 | list_del(&ref_node->node); | |
7282 | spin_unlock(&file_data->lock); | |
05589553 XW |
7283 | |
7284 | percpu_ref_exit(&ref_node->refs); | |
7285 | kfree(ref_node); | |
7286 | percpu_ref_put(&file_data->refs); | |
2faf852d | 7287 | } |
65e19f54 | 7288 | |
4a38aed2 JA |
7289 | static void io_file_put_work(struct work_struct *work) |
7290 | { | |
7291 | struct io_ring_ctx *ctx; | |
7292 | struct llist_node *node; | |
7293 | ||
7294 | ctx = container_of(work, struct io_ring_ctx, file_put_work.work); | |
7295 | node = llist_del_all(&ctx->file_put_llist); | |
7296 | ||
7297 | while (node) { | |
7298 | struct fixed_file_ref_node *ref_node; | |
7299 | struct llist_node *next = node->next; | |
7300 | ||
7301 | ref_node = llist_entry(node, struct fixed_file_ref_node, llist); | |
7302 | __io_file_put_work(ref_node); | |
7303 | node = next; | |
7304 | } | |
7305 | } | |
7306 | ||
05589553 | 7307 | static void io_file_data_ref_zero(struct percpu_ref *ref) |
2faf852d | 7308 | { |
05589553 | 7309 | struct fixed_file_ref_node *ref_node; |
4a38aed2 JA |
7310 | struct io_ring_ctx *ctx; |
7311 | bool first_add; | |
7312 | int delay = HZ; | |
65e19f54 | 7313 | |
05589553 | 7314 | ref_node = container_of(ref, struct fixed_file_ref_node, refs); |
4a38aed2 | 7315 | ctx = ref_node->file_data->ctx; |
05589553 | 7316 | |
4a38aed2 JA |
7317 | if (percpu_ref_is_dying(&ctx->file_data->refs)) |
7318 | delay = 0; | |
05589553 | 7319 | |
4a38aed2 JA |
7320 | first_add = llist_add(&ref_node->llist, &ctx->file_put_llist); |
7321 | if (!delay) | |
7322 | mod_delayed_work(system_wq, &ctx->file_put_work, 0); | |
7323 | else if (first_add) | |
7324 | queue_delayed_work(system_wq, &ctx->file_put_work, delay); | |
05f3fb3c | 7325 | } |
65e19f54 | 7326 | |
05589553 XW |
7327 | static struct fixed_file_ref_node *alloc_fixed_file_ref_node( |
7328 | struct io_ring_ctx *ctx) | |
05f3fb3c | 7329 | { |
05589553 | 7330 | struct fixed_file_ref_node *ref_node; |
05f3fb3c | 7331 | |
05589553 XW |
7332 | ref_node = kzalloc(sizeof(*ref_node), GFP_KERNEL); |
7333 | if (!ref_node) | |
7334 | return ERR_PTR(-ENOMEM); | |
05f3fb3c | 7335 | |
05589553 XW |
7336 | if (percpu_ref_init(&ref_node->refs, io_file_data_ref_zero, |
7337 | 0, GFP_KERNEL)) { | |
7338 | kfree(ref_node); | |
7339 | return ERR_PTR(-ENOMEM); | |
7340 | } | |
7341 | INIT_LIST_HEAD(&ref_node->node); | |
7342 | INIT_LIST_HEAD(&ref_node->file_list); | |
05589553 XW |
7343 | ref_node->file_data = ctx->file_data; |
7344 | return ref_node; | |
05589553 XW |
7345 | } |
7346 | ||
7347 | static void destroy_fixed_file_ref_node(struct fixed_file_ref_node *ref_node) | |
7348 | { | |
7349 | percpu_ref_exit(&ref_node->refs); | |
7350 | kfree(ref_node); | |
65e19f54 JA |
7351 | } |
7352 | ||
6b06314c JA |
7353 | static int io_sqe_files_register(struct io_ring_ctx *ctx, void __user *arg, |
7354 | unsigned nr_args) | |
7355 | { | |
7356 | __s32 __user *fds = (__s32 __user *) arg; | |
600cf3f8 | 7357 | unsigned nr_tables, i; |
05f3fb3c | 7358 | struct file *file; |
600cf3f8 | 7359 | int fd, ret = -ENOMEM; |
05589553 | 7360 | struct fixed_file_ref_node *ref_node; |
5398ae69 | 7361 | struct fixed_file_data *file_data; |
6b06314c | 7362 | |
05f3fb3c | 7363 | if (ctx->file_data) |
6b06314c JA |
7364 | return -EBUSY; |
7365 | if (!nr_args) | |
7366 | return -EINVAL; | |
7367 | if (nr_args > IORING_MAX_FIXED_FILES) | |
7368 | return -EMFILE; | |
7369 | ||
5398ae69 PB |
7370 | file_data = kzalloc(sizeof(*ctx->file_data), GFP_KERNEL); |
7371 | if (!file_data) | |
05f3fb3c | 7372 | return -ENOMEM; |
5398ae69 PB |
7373 | file_data->ctx = ctx; |
7374 | init_completion(&file_data->done); | |
7375 | INIT_LIST_HEAD(&file_data->ref_list); | |
7376 | spin_lock_init(&file_data->lock); | |
05f3fb3c | 7377 | |
65e19f54 | 7378 | nr_tables = DIV_ROUND_UP(nr_args, IORING_MAX_FILES_TABLE); |
035fbafc | 7379 | file_data->table = kcalloc(nr_tables, sizeof(*file_data->table), |
5398ae69 | 7380 | GFP_KERNEL); |
600cf3f8 PB |
7381 | if (!file_data->table) |
7382 | goto out_free; | |
05f3fb3c | 7383 | |
5398ae69 | 7384 | if (percpu_ref_init(&file_data->refs, io_file_ref_kill, |
600cf3f8 PB |
7385 | PERCPU_REF_ALLOW_REINIT, GFP_KERNEL)) |
7386 | goto out_free; | |
6b06314c | 7387 | |
600cf3f8 PB |
7388 | if (io_sqe_alloc_file_tables(file_data, nr_tables, nr_args)) |
7389 | goto out_ref; | |
55cbc256 | 7390 | ctx->file_data = file_data; |
65e19f54 | 7391 | |
08a45173 | 7392 | for (i = 0; i < nr_args; i++, ctx->nr_user_files++) { |
65e19f54 JA |
7393 | struct fixed_file_table *table; |
7394 | unsigned index; | |
7395 | ||
600cf3f8 PB |
7396 | if (copy_from_user(&fd, &fds[i], sizeof(fd))) { |
7397 | ret = -EFAULT; | |
7398 | goto out_fput; | |
7399 | } | |
08a45173 | 7400 | /* allow sparse sets */ |
600cf3f8 | 7401 | if (fd == -1) |
08a45173 | 7402 | continue; |
6b06314c | 7403 | |
05f3fb3c | 7404 | file = fget(fd); |
6b06314c | 7405 | ret = -EBADF; |
05f3fb3c | 7406 | if (!file) |
600cf3f8 | 7407 | goto out_fput; |
05f3fb3c | 7408 | |
6b06314c JA |
7409 | /* |
7410 | * Don't allow io_uring instances to be registered. If UNIX | |
7411 | * isn't enabled, then this causes a reference cycle and this | |
7412 | * instance can never get freed. If UNIX is enabled we'll | |
7413 | * handle it just fine, but there's still no point in allowing | |
7414 | * a ring fd as it doesn't support regular read/write anyway. | |
7415 | */ | |
05f3fb3c JA |
7416 | if (file->f_op == &io_uring_fops) { |
7417 | fput(file); | |
600cf3f8 | 7418 | goto out_fput; |
6b06314c | 7419 | } |
600cf3f8 PB |
7420 | table = &file_data->table[i >> IORING_FILE_TABLE_SHIFT]; |
7421 | index = i & IORING_FILE_TABLE_MASK; | |
05f3fb3c | 7422 | table->files[index] = file; |
6b06314c JA |
7423 | } |
7424 | ||
6b06314c | 7425 | ret = io_sqe_files_scm(ctx); |
05589553 | 7426 | if (ret) { |
6b06314c | 7427 | io_sqe_files_unregister(ctx); |
05589553 XW |
7428 | return ret; |
7429 | } | |
6b06314c | 7430 | |
05589553 XW |
7431 | ref_node = alloc_fixed_file_ref_node(ctx); |
7432 | if (IS_ERR(ref_node)) { | |
7433 | io_sqe_files_unregister(ctx); | |
7434 | return PTR_ERR(ref_node); | |
7435 | } | |
7436 | ||
b2e96852 | 7437 | file_data->node = ref_node; |
5398ae69 PB |
7438 | spin_lock(&file_data->lock); |
7439 | list_add(&ref_node->node, &file_data->ref_list); | |
7440 | spin_unlock(&file_data->lock); | |
7441 | percpu_ref_get(&file_data->refs); | |
6b06314c | 7442 | return ret; |
600cf3f8 PB |
7443 | out_fput: |
7444 | for (i = 0; i < ctx->nr_user_files; i++) { | |
7445 | file = io_file_from_index(ctx, i); | |
7446 | if (file) | |
7447 | fput(file); | |
7448 | } | |
7449 | for (i = 0; i < nr_tables; i++) | |
7450 | kfree(file_data->table[i].files); | |
7451 | ctx->nr_user_files = 0; | |
7452 | out_ref: | |
7453 | percpu_ref_exit(&file_data->refs); | |
7454 | out_free: | |
7455 | kfree(file_data->table); | |
7456 | kfree(file_data); | |
55cbc256 | 7457 | ctx->file_data = NULL; |
6b06314c JA |
7458 | return ret; |
7459 | } | |
7460 | ||
c3a31e60 JA |
7461 | static int io_sqe_file_register(struct io_ring_ctx *ctx, struct file *file, |
7462 | int index) | |
7463 | { | |
7464 | #if defined(CONFIG_UNIX) | |
7465 | struct sock *sock = ctx->ring_sock->sk; | |
7466 | struct sk_buff_head *head = &sock->sk_receive_queue; | |
7467 | struct sk_buff *skb; | |
7468 | ||
7469 | /* | |
7470 | * See if we can merge this file into an existing skb SCM_RIGHTS | |
7471 | * file set. If there's no room, fall back to allocating a new skb | |
7472 | * and filling it in. | |
7473 | */ | |
7474 | spin_lock_irq(&head->lock); | |
7475 | skb = skb_peek(head); | |
7476 | if (skb) { | |
7477 | struct scm_fp_list *fpl = UNIXCB(skb).fp; | |
7478 | ||
7479 | if (fpl->count < SCM_MAX_FD) { | |
7480 | __skb_unlink(skb, head); | |
7481 | spin_unlock_irq(&head->lock); | |
7482 | fpl->fp[fpl->count] = get_file(file); | |
7483 | unix_inflight(fpl->user, fpl->fp[fpl->count]); | |
7484 | fpl->count++; | |
7485 | spin_lock_irq(&head->lock); | |
7486 | __skb_queue_head(head, skb); | |
7487 | } else { | |
7488 | skb = NULL; | |
7489 | } | |
7490 | } | |
7491 | spin_unlock_irq(&head->lock); | |
7492 | ||
7493 | if (skb) { | |
7494 | fput(file); | |
7495 | return 0; | |
7496 | } | |
7497 | ||
7498 | return __io_sqe_files_scm(ctx, 1, index); | |
7499 | #else | |
7500 | return 0; | |
7501 | #endif | |
7502 | } | |
7503 | ||
a5318d3c | 7504 | static int io_queue_file_removal(struct fixed_file_data *data, |
05589553 | 7505 | struct file *file) |
05f3fb3c | 7506 | { |
a5318d3c | 7507 | struct io_file_put *pfile; |
b2e96852 | 7508 | struct fixed_file_ref_node *ref_node = data->node; |
05f3fb3c | 7509 | |
05f3fb3c | 7510 | pfile = kzalloc(sizeof(*pfile), GFP_KERNEL); |
a5318d3c HD |
7511 | if (!pfile) |
7512 | return -ENOMEM; | |
05f3fb3c JA |
7513 | |
7514 | pfile->file = file; | |
05589553 XW |
7515 | list_add(&pfile->list, &ref_node->file_list); |
7516 | ||
a5318d3c | 7517 | return 0; |
05f3fb3c JA |
7518 | } |
7519 | ||
7520 | static int __io_sqe_files_update(struct io_ring_ctx *ctx, | |
7521 | struct io_uring_files_update *up, | |
7522 | unsigned nr_args) | |
7523 | { | |
7524 | struct fixed_file_data *data = ctx->file_data; | |
05589553 | 7525 | struct fixed_file_ref_node *ref_node; |
05f3fb3c | 7526 | struct file *file; |
c3a31e60 JA |
7527 | __s32 __user *fds; |
7528 | int fd, i, err; | |
7529 | __u32 done; | |
05589553 | 7530 | bool needs_switch = false; |
c3a31e60 | 7531 | |
05f3fb3c | 7532 | if (check_add_overflow(up->offset, nr_args, &done)) |
c3a31e60 JA |
7533 | return -EOVERFLOW; |
7534 | if (done > ctx->nr_user_files) | |
7535 | return -EINVAL; | |
7536 | ||
05589553 XW |
7537 | ref_node = alloc_fixed_file_ref_node(ctx); |
7538 | if (IS_ERR(ref_node)) | |
7539 | return PTR_ERR(ref_node); | |
7540 | ||
c3a31e60 | 7541 | done = 0; |
05f3fb3c | 7542 | fds = u64_to_user_ptr(up->fds); |
c3a31e60 | 7543 | while (nr_args) { |
65e19f54 JA |
7544 | struct fixed_file_table *table; |
7545 | unsigned index; | |
7546 | ||
c3a31e60 JA |
7547 | err = 0; |
7548 | if (copy_from_user(&fd, &fds[done], sizeof(fd))) { | |
7549 | err = -EFAULT; | |
7550 | break; | |
7551 | } | |
05f3fb3c JA |
7552 | i = array_index_nospec(up->offset, ctx->nr_user_files); |
7553 | table = &ctx->file_data->table[i >> IORING_FILE_TABLE_SHIFT]; | |
65e19f54 JA |
7554 | index = i & IORING_FILE_TABLE_MASK; |
7555 | if (table->files[index]) { | |
98dfd502 | 7556 | file = table->files[index]; |
a5318d3c HD |
7557 | err = io_queue_file_removal(data, file); |
7558 | if (err) | |
7559 | break; | |
65e19f54 | 7560 | table->files[index] = NULL; |
05589553 | 7561 | needs_switch = true; |
c3a31e60 JA |
7562 | } |
7563 | if (fd != -1) { | |
c3a31e60 JA |
7564 | file = fget(fd); |
7565 | if (!file) { | |
7566 | err = -EBADF; | |
7567 | break; | |
7568 | } | |
7569 | /* | |
7570 | * Don't allow io_uring instances to be registered. If | |
7571 | * UNIX isn't enabled, then this causes a reference | |
7572 | * cycle and this instance can never get freed. If UNIX | |
7573 | * is enabled we'll handle it just fine, but there's | |
7574 | * still no point in allowing a ring fd as it doesn't | |
7575 | * support regular read/write anyway. | |
7576 | */ | |
7577 | if (file->f_op == &io_uring_fops) { | |
7578 | fput(file); | |
7579 | err = -EBADF; | |
7580 | break; | |
7581 | } | |
65e19f54 | 7582 | table->files[index] = file; |
c3a31e60 | 7583 | err = io_sqe_file_register(ctx, file, i); |
f3bd9dae | 7584 | if (err) { |
95d1c8e5 | 7585 | table->files[index] = NULL; |
f3bd9dae | 7586 | fput(file); |
c3a31e60 | 7587 | break; |
f3bd9dae | 7588 | } |
c3a31e60 JA |
7589 | } |
7590 | nr_args--; | |
7591 | done++; | |
05f3fb3c JA |
7592 | up->offset++; |
7593 | } | |
7594 | ||
05589553 | 7595 | if (needs_switch) { |
b2e96852 | 7596 | percpu_ref_kill(&data->node->refs); |
6a4d07cd | 7597 | spin_lock(&data->lock); |
05589553 | 7598 | list_add(&ref_node->node, &data->ref_list); |
b2e96852 | 7599 | data->node = ref_node; |
6a4d07cd | 7600 | spin_unlock(&data->lock); |
05589553 XW |
7601 | percpu_ref_get(&ctx->file_data->refs); |
7602 | } else | |
7603 | destroy_fixed_file_ref_node(ref_node); | |
c3a31e60 JA |
7604 | |
7605 | return done ? done : err; | |
7606 | } | |
05589553 | 7607 | |
05f3fb3c JA |
7608 | static int io_sqe_files_update(struct io_ring_ctx *ctx, void __user *arg, |
7609 | unsigned nr_args) | |
7610 | { | |
7611 | struct io_uring_files_update up; | |
7612 | ||
7613 | if (!ctx->file_data) | |
7614 | return -ENXIO; | |
7615 | if (!nr_args) | |
7616 | return -EINVAL; | |
7617 | if (copy_from_user(&up, arg, sizeof(up))) | |
7618 | return -EFAULT; | |
7619 | if (up.resv) | |
7620 | return -EINVAL; | |
7621 | ||
7622 | return __io_sqe_files_update(ctx, &up, nr_args); | |
7623 | } | |
c3a31e60 | 7624 | |
e9fd9396 | 7625 | static void io_free_work(struct io_wq_work *work) |
7d723065 JA |
7626 | { |
7627 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); | |
7628 | ||
e9fd9396 | 7629 | /* Consider that io_steal_work() relies on this ref */ |
7d723065 JA |
7630 | io_put_req(req); |
7631 | } | |
7632 | ||
24369c2e PB |
7633 | static int io_init_wq_offload(struct io_ring_ctx *ctx, |
7634 | struct io_uring_params *p) | |
7635 | { | |
7636 | struct io_wq_data data; | |
7637 | struct fd f; | |
7638 | struct io_ring_ctx *ctx_attach; | |
7639 | unsigned int concurrency; | |
7640 | int ret = 0; | |
7641 | ||
7642 | data.user = ctx->user; | |
e9fd9396 | 7643 | data.free_work = io_free_work; |
f5fa38c5 | 7644 | data.do_work = io_wq_submit_work; |
24369c2e PB |
7645 | |
7646 | if (!(p->flags & IORING_SETUP_ATTACH_WQ)) { | |
7647 | /* Do QD, or 4 * CPUS, whatever is smallest */ | |
7648 | concurrency = min(ctx->sq_entries, 4 * num_online_cpus()); | |
7649 | ||
7650 | ctx->io_wq = io_wq_create(concurrency, &data); | |
7651 | if (IS_ERR(ctx->io_wq)) { | |
7652 | ret = PTR_ERR(ctx->io_wq); | |
7653 | ctx->io_wq = NULL; | |
7654 | } | |
7655 | return ret; | |
7656 | } | |
7657 | ||
7658 | f = fdget(p->wq_fd); | |
7659 | if (!f.file) | |
7660 | return -EBADF; | |
7661 | ||
7662 | if (f.file->f_op != &io_uring_fops) { | |
7663 | ret = -EINVAL; | |
7664 | goto out_fput; | |
7665 | } | |
7666 | ||
7667 | ctx_attach = f.file->private_data; | |
7668 | /* @io_wq is protected by holding the fd */ | |
7669 | if (!io_wq_get(ctx_attach->io_wq, &data)) { | |
7670 | ret = -EINVAL; | |
7671 | goto out_fput; | |
7672 | } | |
7673 | ||
7674 | ctx->io_wq = ctx_attach->io_wq; | |
7675 | out_fput: | |
7676 | fdput(f); | |
7677 | return ret; | |
7678 | } | |
7679 | ||
0f212204 JA |
7680 | static int io_uring_alloc_task_context(struct task_struct *task) |
7681 | { | |
7682 | struct io_uring_task *tctx; | |
d8a6df10 | 7683 | int ret; |
0f212204 JA |
7684 | |
7685 | tctx = kmalloc(sizeof(*tctx), GFP_KERNEL); | |
7686 | if (unlikely(!tctx)) | |
7687 | return -ENOMEM; | |
7688 | ||
d8a6df10 JA |
7689 | ret = percpu_counter_init(&tctx->inflight, 0, GFP_KERNEL); |
7690 | if (unlikely(ret)) { | |
7691 | kfree(tctx); | |
7692 | return ret; | |
7693 | } | |
7694 | ||
0f212204 JA |
7695 | xa_init(&tctx->xa); |
7696 | init_waitqueue_head(&tctx->wait); | |
7697 | tctx->last = NULL; | |
7698 | tctx->in_idle = 0; | |
500a373d JA |
7699 | io_init_identity(&tctx->__identity); |
7700 | tctx->identity = &tctx->__identity; | |
0f212204 JA |
7701 | task->io_uring = tctx; |
7702 | return 0; | |
7703 | } | |
7704 | ||
7705 | void __io_uring_free(struct task_struct *tsk) | |
7706 | { | |
7707 | struct io_uring_task *tctx = tsk->io_uring; | |
7708 | ||
7709 | WARN_ON_ONCE(!xa_empty(&tctx->xa)); | |
500a373d JA |
7710 | WARN_ON_ONCE(refcount_read(&tctx->identity->count) != 1); |
7711 | if (tctx->identity != &tctx->__identity) | |
7712 | kfree(tctx->identity); | |
d8a6df10 | 7713 | percpu_counter_destroy(&tctx->inflight); |
0f212204 JA |
7714 | kfree(tctx); |
7715 | tsk->io_uring = NULL; | |
7716 | } | |
7717 | ||
7e84e1c7 SG |
7718 | static int io_sq_offload_create(struct io_ring_ctx *ctx, |
7719 | struct io_uring_params *p) | |
2b188cc1 JA |
7720 | { |
7721 | int ret; | |
7722 | ||
6c271ce2 | 7723 | if (ctx->flags & IORING_SETUP_SQPOLL) { |
534ca6d6 JA |
7724 | struct io_sq_data *sqd; |
7725 | ||
3ec482d1 JA |
7726 | ret = -EPERM; |
7727 | if (!capable(CAP_SYS_ADMIN)) | |
7728 | goto err; | |
7729 | ||
534ca6d6 JA |
7730 | sqd = io_get_sq_data(p); |
7731 | if (IS_ERR(sqd)) { | |
7732 | ret = PTR_ERR(sqd); | |
7733 | goto err; | |
7734 | } | |
69fb2131 | 7735 | |
534ca6d6 | 7736 | ctx->sq_data = sqd; |
69fb2131 JA |
7737 | io_sq_thread_park(sqd); |
7738 | mutex_lock(&sqd->ctx_lock); | |
7739 | list_add(&ctx->sqd_list, &sqd->ctx_new_list); | |
7740 | mutex_unlock(&sqd->ctx_lock); | |
7741 | io_sq_thread_unpark(sqd); | |
534ca6d6 | 7742 | |
917257da JA |
7743 | ctx->sq_thread_idle = msecs_to_jiffies(p->sq_thread_idle); |
7744 | if (!ctx->sq_thread_idle) | |
7745 | ctx->sq_thread_idle = HZ; | |
7746 | ||
aa06165d JA |
7747 | if (sqd->thread) |
7748 | goto done; | |
7749 | ||
6c271ce2 | 7750 | if (p->flags & IORING_SETUP_SQ_AFF) { |
44a9bd18 | 7751 | int cpu = p->sq_thread_cpu; |
6c271ce2 | 7752 | |
917257da | 7753 | ret = -EINVAL; |
44a9bd18 JA |
7754 | if (cpu >= nr_cpu_ids) |
7755 | goto err; | |
7889f44d | 7756 | if (!cpu_online(cpu)) |
917257da JA |
7757 | goto err; |
7758 | ||
69fb2131 | 7759 | sqd->thread = kthread_create_on_cpu(io_sq_thread, sqd, |
534ca6d6 | 7760 | cpu, "io_uring-sq"); |
6c271ce2 | 7761 | } else { |
69fb2131 | 7762 | sqd->thread = kthread_create(io_sq_thread, sqd, |
6c271ce2 JA |
7763 | "io_uring-sq"); |
7764 | } | |
534ca6d6 JA |
7765 | if (IS_ERR(sqd->thread)) { |
7766 | ret = PTR_ERR(sqd->thread); | |
7767 | sqd->thread = NULL; | |
6c271ce2 JA |
7768 | goto err; |
7769 | } | |
534ca6d6 | 7770 | ret = io_uring_alloc_task_context(sqd->thread); |
0f212204 JA |
7771 | if (ret) |
7772 | goto err; | |
6c271ce2 JA |
7773 | } else if (p->flags & IORING_SETUP_SQ_AFF) { |
7774 | /* Can't have SQ_AFF without SQPOLL */ | |
7775 | ret = -EINVAL; | |
7776 | goto err; | |
7777 | } | |
7778 | ||
aa06165d | 7779 | done: |
24369c2e PB |
7780 | ret = io_init_wq_offload(ctx, p); |
7781 | if (ret) | |
2b188cc1 | 7782 | goto err; |
2b188cc1 JA |
7783 | |
7784 | return 0; | |
7785 | err: | |
54a91f3b | 7786 | io_finish_async(ctx); |
2b188cc1 JA |
7787 | return ret; |
7788 | } | |
7789 | ||
7e84e1c7 SG |
7790 | static void io_sq_offload_start(struct io_ring_ctx *ctx) |
7791 | { | |
534ca6d6 JA |
7792 | struct io_sq_data *sqd = ctx->sq_data; |
7793 | ||
7794 | if ((ctx->flags & IORING_SETUP_SQPOLL) && sqd->thread) | |
7795 | wake_up_process(sqd->thread); | |
7e84e1c7 SG |
7796 | } |
7797 | ||
a087e2b5 BM |
7798 | static inline void __io_unaccount_mem(struct user_struct *user, |
7799 | unsigned long nr_pages) | |
2b188cc1 JA |
7800 | { |
7801 | atomic_long_sub(nr_pages, &user->locked_vm); | |
7802 | } | |
7803 | ||
a087e2b5 BM |
7804 | static inline int __io_account_mem(struct user_struct *user, |
7805 | unsigned long nr_pages) | |
2b188cc1 JA |
7806 | { |
7807 | unsigned long page_limit, cur_pages, new_pages; | |
7808 | ||
7809 | /* Don't allow more pages than we can safely lock */ | |
7810 | page_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT; | |
7811 | ||
7812 | do { | |
7813 | cur_pages = atomic_long_read(&user->locked_vm); | |
7814 | new_pages = cur_pages + nr_pages; | |
7815 | if (new_pages > page_limit) | |
7816 | return -ENOMEM; | |
7817 | } while (atomic_long_cmpxchg(&user->locked_vm, cur_pages, | |
7818 | new_pages) != cur_pages); | |
7819 | ||
7820 | return 0; | |
7821 | } | |
7822 | ||
2e0464d4 BM |
7823 | static void io_unaccount_mem(struct io_ring_ctx *ctx, unsigned long nr_pages, |
7824 | enum io_mem_account acct) | |
a087e2b5 | 7825 | { |
aad5d8da | 7826 | if (ctx->limit_mem) |
a087e2b5 | 7827 | __io_unaccount_mem(ctx->user, nr_pages); |
30975825 | 7828 | |
2aede0e4 | 7829 | if (ctx->mm_account) { |
2e0464d4 | 7830 | if (acct == ACCT_LOCKED) |
2aede0e4 | 7831 | ctx->mm_account->locked_vm -= nr_pages; |
2e0464d4 | 7832 | else if (acct == ACCT_PINNED) |
2aede0e4 | 7833 | atomic64_sub(nr_pages, &ctx->mm_account->pinned_vm); |
2e0464d4 | 7834 | } |
a087e2b5 BM |
7835 | } |
7836 | ||
2e0464d4 BM |
7837 | static int io_account_mem(struct io_ring_ctx *ctx, unsigned long nr_pages, |
7838 | enum io_mem_account acct) | |
a087e2b5 | 7839 | { |
30975825 BM |
7840 | int ret; |
7841 | ||
7842 | if (ctx->limit_mem) { | |
7843 | ret = __io_account_mem(ctx->user, nr_pages); | |
7844 | if (ret) | |
7845 | return ret; | |
7846 | } | |
7847 | ||
2aede0e4 | 7848 | if (ctx->mm_account) { |
2e0464d4 | 7849 | if (acct == ACCT_LOCKED) |
2aede0e4 | 7850 | ctx->mm_account->locked_vm += nr_pages; |
2e0464d4 | 7851 | else if (acct == ACCT_PINNED) |
2aede0e4 | 7852 | atomic64_add(nr_pages, &ctx->mm_account->pinned_vm); |
2e0464d4 | 7853 | } |
a087e2b5 BM |
7854 | |
7855 | return 0; | |
7856 | } | |
7857 | ||
2b188cc1 JA |
7858 | static void io_mem_free(void *ptr) |
7859 | { | |
52e04ef4 MR |
7860 | struct page *page; |
7861 | ||
7862 | if (!ptr) | |
7863 | return; | |
2b188cc1 | 7864 | |
52e04ef4 | 7865 | page = virt_to_head_page(ptr); |
2b188cc1 JA |
7866 | if (put_page_testzero(page)) |
7867 | free_compound_page(page); | |
7868 | } | |
7869 | ||
7870 | static void *io_mem_alloc(size_t size) | |
7871 | { | |
7872 | gfp_t gfp_flags = GFP_KERNEL | __GFP_ZERO | __GFP_NOWARN | __GFP_COMP | | |
7873 | __GFP_NORETRY; | |
7874 | ||
7875 | return (void *) __get_free_pages(gfp_flags, get_order(size)); | |
7876 | } | |
7877 | ||
75b28aff HV |
7878 | static unsigned long rings_size(unsigned sq_entries, unsigned cq_entries, |
7879 | size_t *sq_offset) | |
7880 | { | |
7881 | struct io_rings *rings; | |
7882 | size_t off, sq_array_size; | |
7883 | ||
7884 | off = struct_size(rings, cqes, cq_entries); | |
7885 | if (off == SIZE_MAX) | |
7886 | return SIZE_MAX; | |
7887 | ||
7888 | #ifdef CONFIG_SMP | |
7889 | off = ALIGN(off, SMP_CACHE_BYTES); | |
7890 | if (off == 0) | |
7891 | return SIZE_MAX; | |
7892 | #endif | |
7893 | ||
b36200f5 DV |
7894 | if (sq_offset) |
7895 | *sq_offset = off; | |
7896 | ||
75b28aff HV |
7897 | sq_array_size = array_size(sizeof(u32), sq_entries); |
7898 | if (sq_array_size == SIZE_MAX) | |
7899 | return SIZE_MAX; | |
7900 | ||
7901 | if (check_add_overflow(off, sq_array_size, &off)) | |
7902 | return SIZE_MAX; | |
7903 | ||
75b28aff HV |
7904 | return off; |
7905 | } | |
7906 | ||
2b188cc1 JA |
7907 | static unsigned long ring_pages(unsigned sq_entries, unsigned cq_entries) |
7908 | { | |
75b28aff | 7909 | size_t pages; |
2b188cc1 | 7910 | |
75b28aff HV |
7911 | pages = (size_t)1 << get_order( |
7912 | rings_size(sq_entries, cq_entries, NULL)); | |
7913 | pages += (size_t)1 << get_order( | |
7914 | array_size(sizeof(struct io_uring_sqe), sq_entries)); | |
2b188cc1 | 7915 | |
75b28aff | 7916 | return pages; |
2b188cc1 JA |
7917 | } |
7918 | ||
edafccee JA |
7919 | static int io_sqe_buffer_unregister(struct io_ring_ctx *ctx) |
7920 | { | |
7921 | int i, j; | |
7922 | ||
7923 | if (!ctx->user_bufs) | |
7924 | return -ENXIO; | |
7925 | ||
7926 | for (i = 0; i < ctx->nr_user_bufs; i++) { | |
7927 | struct io_mapped_ubuf *imu = &ctx->user_bufs[i]; | |
7928 | ||
7929 | for (j = 0; j < imu->nr_bvecs; j++) | |
f1f6a7dd | 7930 | unpin_user_page(imu->bvec[j].bv_page); |
edafccee | 7931 | |
de293938 JA |
7932 | if (imu->acct_pages) |
7933 | io_unaccount_mem(ctx, imu->acct_pages, ACCT_PINNED); | |
d4ef6475 | 7934 | kvfree(imu->bvec); |
edafccee JA |
7935 | imu->nr_bvecs = 0; |
7936 | } | |
7937 | ||
7938 | kfree(ctx->user_bufs); | |
7939 | ctx->user_bufs = NULL; | |
7940 | ctx->nr_user_bufs = 0; | |
7941 | return 0; | |
7942 | } | |
7943 | ||
7944 | static int io_copy_iov(struct io_ring_ctx *ctx, struct iovec *dst, | |
7945 | void __user *arg, unsigned index) | |
7946 | { | |
7947 | struct iovec __user *src; | |
7948 | ||
7949 | #ifdef CONFIG_COMPAT | |
7950 | if (ctx->compat) { | |
7951 | struct compat_iovec __user *ciovs; | |
7952 | struct compat_iovec ciov; | |
7953 | ||
7954 | ciovs = (struct compat_iovec __user *) arg; | |
7955 | if (copy_from_user(&ciov, &ciovs[index], sizeof(ciov))) | |
7956 | return -EFAULT; | |
7957 | ||
d55e5f5b | 7958 | dst->iov_base = u64_to_user_ptr((u64)ciov.iov_base); |
edafccee JA |
7959 | dst->iov_len = ciov.iov_len; |
7960 | return 0; | |
7961 | } | |
7962 | #endif | |
7963 | src = (struct iovec __user *) arg; | |
7964 | if (copy_from_user(dst, &src[index], sizeof(*dst))) | |
7965 | return -EFAULT; | |
7966 | return 0; | |
7967 | } | |
7968 | ||
de293938 JA |
7969 | /* |
7970 | * Not super efficient, but this is just a registration time. And we do cache | |
7971 | * the last compound head, so generally we'll only do a full search if we don't | |
7972 | * match that one. | |
7973 | * | |
7974 | * We check if the given compound head page has already been accounted, to | |
7975 | * avoid double accounting it. This allows us to account the full size of the | |
7976 | * page, not just the constituent pages of a huge page. | |
7977 | */ | |
7978 | static bool headpage_already_acct(struct io_ring_ctx *ctx, struct page **pages, | |
7979 | int nr_pages, struct page *hpage) | |
7980 | { | |
7981 | int i, j; | |
7982 | ||
7983 | /* check current page array */ | |
7984 | for (i = 0; i < nr_pages; i++) { | |
7985 | if (!PageCompound(pages[i])) | |
7986 | continue; | |
7987 | if (compound_head(pages[i]) == hpage) | |
7988 | return true; | |
7989 | } | |
7990 | ||
7991 | /* check previously registered pages */ | |
7992 | for (i = 0; i < ctx->nr_user_bufs; i++) { | |
7993 | struct io_mapped_ubuf *imu = &ctx->user_bufs[i]; | |
7994 | ||
7995 | for (j = 0; j < imu->nr_bvecs; j++) { | |
7996 | if (!PageCompound(imu->bvec[j].bv_page)) | |
7997 | continue; | |
7998 | if (compound_head(imu->bvec[j].bv_page) == hpage) | |
7999 | return true; | |
8000 | } | |
8001 | } | |
8002 | ||
8003 | return false; | |
8004 | } | |
8005 | ||
8006 | static int io_buffer_account_pin(struct io_ring_ctx *ctx, struct page **pages, | |
8007 | int nr_pages, struct io_mapped_ubuf *imu, | |
8008 | struct page **last_hpage) | |
8009 | { | |
8010 | int i, ret; | |
8011 | ||
8012 | for (i = 0; i < nr_pages; i++) { | |
8013 | if (!PageCompound(pages[i])) { | |
8014 | imu->acct_pages++; | |
8015 | } else { | |
8016 | struct page *hpage; | |
8017 | ||
8018 | hpage = compound_head(pages[i]); | |
8019 | if (hpage == *last_hpage) | |
8020 | continue; | |
8021 | *last_hpage = hpage; | |
8022 | if (headpage_already_acct(ctx, pages, i, hpage)) | |
8023 | continue; | |
8024 | imu->acct_pages += page_size(hpage) >> PAGE_SHIFT; | |
8025 | } | |
8026 | } | |
8027 | ||
8028 | if (!imu->acct_pages) | |
8029 | return 0; | |
8030 | ||
8031 | ret = io_account_mem(ctx, imu->acct_pages, ACCT_PINNED); | |
8032 | if (ret) | |
8033 | imu->acct_pages = 0; | |
8034 | return ret; | |
8035 | } | |
8036 | ||
edafccee JA |
8037 | static int io_sqe_buffer_register(struct io_ring_ctx *ctx, void __user *arg, |
8038 | unsigned nr_args) | |
8039 | { | |
8040 | struct vm_area_struct **vmas = NULL; | |
8041 | struct page **pages = NULL; | |
de293938 | 8042 | struct page *last_hpage = NULL; |
edafccee JA |
8043 | int i, j, got_pages = 0; |
8044 | int ret = -EINVAL; | |
8045 | ||
8046 | if (ctx->user_bufs) | |
8047 | return -EBUSY; | |
8048 | if (!nr_args || nr_args > UIO_MAXIOV) | |
8049 | return -EINVAL; | |
8050 | ||
8051 | ctx->user_bufs = kcalloc(nr_args, sizeof(struct io_mapped_ubuf), | |
8052 | GFP_KERNEL); | |
8053 | if (!ctx->user_bufs) | |
8054 | return -ENOMEM; | |
8055 | ||
8056 | for (i = 0; i < nr_args; i++) { | |
8057 | struct io_mapped_ubuf *imu = &ctx->user_bufs[i]; | |
8058 | unsigned long off, start, end, ubuf; | |
8059 | int pret, nr_pages; | |
8060 | struct iovec iov; | |
8061 | size_t size; | |
8062 | ||
8063 | ret = io_copy_iov(ctx, &iov, arg, i); | |
8064 | if (ret) | |
a278682d | 8065 | goto err; |
edafccee JA |
8066 | |
8067 | /* | |
8068 | * Don't impose further limits on the size and buffer | |
8069 | * constraints here, we'll -EINVAL later when IO is | |
8070 | * submitted if they are wrong. | |
8071 | */ | |
8072 | ret = -EFAULT; | |
8073 | if (!iov.iov_base || !iov.iov_len) | |
8074 | goto err; | |
8075 | ||
8076 | /* arbitrary limit, but we need something */ | |
8077 | if (iov.iov_len > SZ_1G) | |
8078 | goto err; | |
8079 | ||
8080 | ubuf = (unsigned long) iov.iov_base; | |
8081 | end = (ubuf + iov.iov_len + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
8082 | start = ubuf >> PAGE_SHIFT; | |
8083 | nr_pages = end - start; | |
8084 | ||
edafccee JA |
8085 | ret = 0; |
8086 | if (!pages || nr_pages > got_pages) { | |
a8c73c1a DE |
8087 | kvfree(vmas); |
8088 | kvfree(pages); | |
d4ef6475 | 8089 | pages = kvmalloc_array(nr_pages, sizeof(struct page *), |
edafccee | 8090 | GFP_KERNEL); |
d4ef6475 | 8091 | vmas = kvmalloc_array(nr_pages, |
edafccee JA |
8092 | sizeof(struct vm_area_struct *), |
8093 | GFP_KERNEL); | |
8094 | if (!pages || !vmas) { | |
8095 | ret = -ENOMEM; | |
edafccee JA |
8096 | goto err; |
8097 | } | |
8098 | got_pages = nr_pages; | |
8099 | } | |
8100 | ||
d4ef6475 | 8101 | imu->bvec = kvmalloc_array(nr_pages, sizeof(struct bio_vec), |
edafccee JA |
8102 | GFP_KERNEL); |
8103 | ret = -ENOMEM; | |
de293938 | 8104 | if (!imu->bvec) |
edafccee | 8105 | goto err; |
edafccee JA |
8106 | |
8107 | ret = 0; | |
d8ed45c5 | 8108 | mmap_read_lock(current->mm); |
2113b05d | 8109 | pret = pin_user_pages(ubuf, nr_pages, |
932f4a63 IW |
8110 | FOLL_WRITE | FOLL_LONGTERM, |
8111 | pages, vmas); | |
edafccee JA |
8112 | if (pret == nr_pages) { |
8113 | /* don't support file backed memory */ | |
8114 | for (j = 0; j < nr_pages; j++) { | |
8115 | struct vm_area_struct *vma = vmas[j]; | |
8116 | ||
8117 | if (vma->vm_file && | |
8118 | !is_file_hugepages(vma->vm_file)) { | |
8119 | ret = -EOPNOTSUPP; | |
8120 | break; | |
8121 | } | |
8122 | } | |
8123 | } else { | |
8124 | ret = pret < 0 ? pret : -EFAULT; | |
8125 | } | |
d8ed45c5 | 8126 | mmap_read_unlock(current->mm); |
edafccee JA |
8127 | if (ret) { |
8128 | /* | |
8129 | * if we did partial map, or found file backed vmas, | |
8130 | * release any pages we did get | |
8131 | */ | |
27c4d3a3 | 8132 | if (pret > 0) |
f1f6a7dd | 8133 | unpin_user_pages(pages, pret); |
de293938 JA |
8134 | kvfree(imu->bvec); |
8135 | goto err; | |
8136 | } | |
8137 | ||
8138 | ret = io_buffer_account_pin(ctx, pages, pret, imu, &last_hpage); | |
8139 | if (ret) { | |
8140 | unpin_user_pages(pages, pret); | |
d4ef6475 | 8141 | kvfree(imu->bvec); |
edafccee JA |
8142 | goto err; |
8143 | } | |
8144 | ||
8145 | off = ubuf & ~PAGE_MASK; | |
8146 | size = iov.iov_len; | |
8147 | for (j = 0; j < nr_pages; j++) { | |
8148 | size_t vec_len; | |
8149 | ||
8150 | vec_len = min_t(size_t, size, PAGE_SIZE - off); | |
8151 | imu->bvec[j].bv_page = pages[j]; | |
8152 | imu->bvec[j].bv_len = vec_len; | |
8153 | imu->bvec[j].bv_offset = off; | |
8154 | off = 0; | |
8155 | size -= vec_len; | |
8156 | } | |
8157 | /* store original address for later verification */ | |
8158 | imu->ubuf = ubuf; | |
8159 | imu->len = iov.iov_len; | |
8160 | imu->nr_bvecs = nr_pages; | |
8161 | ||
8162 | ctx->nr_user_bufs++; | |
8163 | } | |
d4ef6475 MR |
8164 | kvfree(pages); |
8165 | kvfree(vmas); | |
edafccee JA |
8166 | return 0; |
8167 | err: | |
d4ef6475 MR |
8168 | kvfree(pages); |
8169 | kvfree(vmas); | |
edafccee JA |
8170 | io_sqe_buffer_unregister(ctx); |
8171 | return ret; | |
8172 | } | |
8173 | ||
9b402849 JA |
8174 | static int io_eventfd_register(struct io_ring_ctx *ctx, void __user *arg) |
8175 | { | |
8176 | __s32 __user *fds = arg; | |
8177 | int fd; | |
8178 | ||
8179 | if (ctx->cq_ev_fd) | |
8180 | return -EBUSY; | |
8181 | ||
8182 | if (copy_from_user(&fd, fds, sizeof(*fds))) | |
8183 | return -EFAULT; | |
8184 | ||
8185 | ctx->cq_ev_fd = eventfd_ctx_fdget(fd); | |
8186 | if (IS_ERR(ctx->cq_ev_fd)) { | |
8187 | int ret = PTR_ERR(ctx->cq_ev_fd); | |
8188 | ctx->cq_ev_fd = NULL; | |
8189 | return ret; | |
8190 | } | |
8191 | ||
8192 | return 0; | |
8193 | } | |
8194 | ||
8195 | static int io_eventfd_unregister(struct io_ring_ctx *ctx) | |
8196 | { | |
8197 | if (ctx->cq_ev_fd) { | |
8198 | eventfd_ctx_put(ctx->cq_ev_fd); | |
8199 | ctx->cq_ev_fd = NULL; | |
8200 | return 0; | |
8201 | } | |
8202 | ||
8203 | return -ENXIO; | |
8204 | } | |
8205 | ||
5a2e745d JA |
8206 | static int __io_destroy_buffers(int id, void *p, void *data) |
8207 | { | |
8208 | struct io_ring_ctx *ctx = data; | |
8209 | struct io_buffer *buf = p; | |
8210 | ||
067524e9 | 8211 | __io_remove_buffers(ctx, buf, id, -1U); |
5a2e745d JA |
8212 | return 0; |
8213 | } | |
8214 | ||
8215 | static void io_destroy_buffers(struct io_ring_ctx *ctx) | |
8216 | { | |
8217 | idr_for_each(&ctx->io_buffer_idr, __io_destroy_buffers, ctx); | |
8218 | idr_destroy(&ctx->io_buffer_idr); | |
8219 | } | |
8220 | ||
2b188cc1 JA |
8221 | static void io_ring_ctx_free(struct io_ring_ctx *ctx) |
8222 | { | |
6b06314c | 8223 | io_finish_async(ctx); |
5dbcad51 | 8224 | io_sqe_buffer_unregister(ctx); |
2aede0e4 JA |
8225 | |
8226 | if (ctx->sqo_task) { | |
8227 | put_task_struct(ctx->sqo_task); | |
8228 | ctx->sqo_task = NULL; | |
8229 | mmdrop(ctx->mm_account); | |
8230 | ctx->mm_account = NULL; | |
30975825 | 8231 | } |
def596e9 | 8232 | |
91d8f519 DZ |
8233 | #ifdef CONFIG_BLK_CGROUP |
8234 | if (ctx->sqo_blkcg_css) | |
8235 | css_put(ctx->sqo_blkcg_css); | |
8236 | #endif | |
8237 | ||
6b06314c | 8238 | io_sqe_files_unregister(ctx); |
9b402849 | 8239 | io_eventfd_unregister(ctx); |
5a2e745d | 8240 | io_destroy_buffers(ctx); |
41726c9a | 8241 | idr_destroy(&ctx->personality_idr); |
def596e9 | 8242 | |
2b188cc1 | 8243 | #if defined(CONFIG_UNIX) |
355e8d26 EB |
8244 | if (ctx->ring_sock) { |
8245 | ctx->ring_sock->file = NULL; /* so that iput() is called */ | |
2b188cc1 | 8246 | sock_release(ctx->ring_sock); |
355e8d26 | 8247 | } |
2b188cc1 JA |
8248 | #endif |
8249 | ||
75b28aff | 8250 | io_mem_free(ctx->rings); |
2b188cc1 | 8251 | io_mem_free(ctx->sq_sqes); |
2b188cc1 JA |
8252 | |
8253 | percpu_ref_exit(&ctx->refs); | |
2b188cc1 | 8254 | free_uid(ctx->user); |
181e448d | 8255 | put_cred(ctx->creds); |
78076bb6 | 8256 | kfree(ctx->cancel_hash); |
0ddf92e8 | 8257 | kmem_cache_free(req_cachep, ctx->fallback_req); |
2b188cc1 JA |
8258 | kfree(ctx); |
8259 | } | |
8260 | ||
8261 | static __poll_t io_uring_poll(struct file *file, poll_table *wait) | |
8262 | { | |
8263 | struct io_ring_ctx *ctx = file->private_data; | |
8264 | __poll_t mask = 0; | |
8265 | ||
8266 | poll_wait(file, &ctx->cq_wait, wait); | |
4f7067c3 SB |
8267 | /* |
8268 | * synchronizes with barrier from wq_has_sleeper call in | |
8269 | * io_commit_cqring | |
8270 | */ | |
2b188cc1 | 8271 | smp_rmb(); |
90554200 | 8272 | if (!io_sqring_full(ctx)) |
2b188cc1 | 8273 | mask |= EPOLLOUT | EPOLLWRNORM; |
63e5d81f | 8274 | if (io_cqring_events(ctx, false)) |
2b188cc1 JA |
8275 | mask |= EPOLLIN | EPOLLRDNORM; |
8276 | ||
8277 | return mask; | |
8278 | } | |
8279 | ||
8280 | static int io_uring_fasync(int fd, struct file *file, int on) | |
8281 | { | |
8282 | struct io_ring_ctx *ctx = file->private_data; | |
8283 | ||
8284 | return fasync_helper(fd, file, on, &ctx->cq_fasync); | |
8285 | } | |
8286 | ||
071698e1 JA |
8287 | static int io_remove_personalities(int id, void *p, void *data) |
8288 | { | |
8289 | struct io_ring_ctx *ctx = data; | |
1e6fa521 | 8290 | struct io_identity *iod; |
071698e1 | 8291 | |
1e6fa521 JA |
8292 | iod = idr_remove(&ctx->personality_idr, id); |
8293 | if (iod) { | |
8294 | put_cred(iod->creds); | |
8295 | if (refcount_dec_and_test(&iod->count)) | |
8296 | kfree(iod); | |
8297 | } | |
071698e1 JA |
8298 | return 0; |
8299 | } | |
8300 | ||
85faa7b8 JA |
8301 | static void io_ring_exit_work(struct work_struct *work) |
8302 | { | |
b2edc0a7 PB |
8303 | struct io_ring_ctx *ctx = container_of(work, struct io_ring_ctx, |
8304 | exit_work); | |
85faa7b8 | 8305 | |
56952e91 JA |
8306 | /* |
8307 | * If we're doing polled IO and end up having requests being | |
8308 | * submitted async (out-of-line), then completions can come in while | |
8309 | * we're waiting for refs to drop. We need to reap these manually, | |
8310 | * as nobody else will be looking for them. | |
8311 | */ | |
b2edc0a7 | 8312 | do { |
56952e91 | 8313 | if (ctx->rings) |
e6c8aa9a | 8314 | io_cqring_overflow_flush(ctx, true, NULL, NULL); |
b2edc0a7 PB |
8315 | io_iopoll_try_reap_events(ctx); |
8316 | } while (!wait_for_completion_timeout(&ctx->ref_comp, HZ/20)); | |
85faa7b8 JA |
8317 | io_ring_ctx_free(ctx); |
8318 | } | |
8319 | ||
2b188cc1 JA |
8320 | static void io_ring_ctx_wait_and_kill(struct io_ring_ctx *ctx) |
8321 | { | |
8322 | mutex_lock(&ctx->uring_lock); | |
8323 | percpu_ref_kill(&ctx->refs); | |
8324 | mutex_unlock(&ctx->uring_lock); | |
8325 | ||
f3606e3a JA |
8326 | io_kill_timeouts(ctx, NULL); |
8327 | io_poll_remove_all(ctx, NULL); | |
561fb04a JA |
8328 | |
8329 | if (ctx->io_wq) | |
8330 | io_wq_cancel_all(ctx->io_wq); | |
8331 | ||
15dff286 JA |
8332 | /* if we failed setting up the ctx, we might not have any rings */ |
8333 | if (ctx->rings) | |
e6c8aa9a | 8334 | io_cqring_overflow_flush(ctx, true, NULL, NULL); |
b2edc0a7 | 8335 | io_iopoll_try_reap_events(ctx); |
071698e1 | 8336 | idr_for_each(&ctx->personality_idr, io_remove_personalities, ctx); |
309fc03a JA |
8337 | |
8338 | /* | |
8339 | * Do this upfront, so we won't have a grace period where the ring | |
8340 | * is closed but resources aren't reaped yet. This can cause | |
8341 | * spurious failure in setting up a new ring. | |
8342 | */ | |
760618f7 JA |
8343 | io_unaccount_mem(ctx, ring_pages(ctx->sq_entries, ctx->cq_entries), |
8344 | ACCT_LOCKED); | |
309fc03a | 8345 | |
85faa7b8 | 8346 | INIT_WORK(&ctx->exit_work, io_ring_exit_work); |
fc666777 JA |
8347 | /* |
8348 | * Use system_unbound_wq to avoid spawning tons of event kworkers | |
8349 | * if we're exiting a ton of rings at the same time. It just adds | |
8350 | * noise and overhead, there's no discernable change in runtime | |
8351 | * over using system_wq. | |
8352 | */ | |
8353 | queue_work(system_unbound_wq, &ctx->exit_work); | |
2b188cc1 JA |
8354 | } |
8355 | ||
8356 | static int io_uring_release(struct inode *inode, struct file *file) | |
8357 | { | |
8358 | struct io_ring_ctx *ctx = file->private_data; | |
8359 | ||
8360 | file->private_data = NULL; | |
8361 | io_ring_ctx_wait_and_kill(ctx); | |
8362 | return 0; | |
8363 | } | |
8364 | ||
67c4d9e6 PB |
8365 | static bool io_wq_files_match(struct io_wq_work *work, void *data) |
8366 | { | |
8367 | struct files_struct *files = data; | |
8368 | ||
dfead8a8 | 8369 | return !files || ((work->flags & IO_WQ_WORK_FILES) && |
98447d65 | 8370 | work->identity->files == files); |
67c4d9e6 PB |
8371 | } |
8372 | ||
f254ac04 JA |
8373 | /* |
8374 | * Returns true if 'preq' is the link parent of 'req' | |
8375 | */ | |
8376 | static bool io_match_link(struct io_kiocb *preq, struct io_kiocb *req) | |
8377 | { | |
8378 | struct io_kiocb *link; | |
8379 | ||
8380 | if (!(preq->flags & REQ_F_LINK_HEAD)) | |
8381 | return false; | |
8382 | ||
8383 | list_for_each_entry(link, &preq->link_list, link_list) { | |
8384 | if (link == req) | |
8385 | return true; | |
8386 | } | |
8387 | ||
8388 | return false; | |
8389 | } | |
8390 | ||
c127a2a1 PB |
8391 | static bool io_match_link_files(struct io_kiocb *req, |
8392 | struct files_struct *files) | |
8393 | { | |
8394 | struct io_kiocb *link; | |
8395 | ||
8396 | if (io_match_files(req, files)) | |
8397 | return true; | |
8398 | if (req->flags & REQ_F_LINK_HEAD) { | |
8399 | list_for_each_entry(link, &req->link_list, link_list) { | |
8400 | if (io_match_files(link, files)) | |
8401 | return true; | |
8402 | } | |
8403 | } | |
8404 | return false; | |
8405 | } | |
8406 | ||
f254ac04 JA |
8407 | /* |
8408 | * We're looking to cancel 'req' because it's holding on to our files, but | |
8409 | * 'req' could be a link to another request. See if it is, and cancel that | |
8410 | * parent request if so. | |
8411 | */ | |
8412 | static bool io_poll_remove_link(struct io_ring_ctx *ctx, struct io_kiocb *req) | |
8413 | { | |
8414 | struct hlist_node *tmp; | |
8415 | struct io_kiocb *preq; | |
8416 | bool found = false; | |
8417 | int i; | |
8418 | ||
8419 | spin_lock_irq(&ctx->completion_lock); | |
8420 | for (i = 0; i < (1U << ctx->cancel_hash_bits); i++) { | |
8421 | struct hlist_head *list; | |
8422 | ||
8423 | list = &ctx->cancel_hash[i]; | |
8424 | hlist_for_each_entry_safe(preq, tmp, list, hash_node) { | |
8425 | found = io_match_link(preq, req); | |
8426 | if (found) { | |
8427 | io_poll_remove_one(preq); | |
8428 | break; | |
8429 | } | |
8430 | } | |
8431 | } | |
8432 | spin_unlock_irq(&ctx->completion_lock); | |
8433 | return found; | |
8434 | } | |
8435 | ||
8436 | static bool io_timeout_remove_link(struct io_ring_ctx *ctx, | |
8437 | struct io_kiocb *req) | |
8438 | { | |
8439 | struct io_kiocb *preq; | |
8440 | bool found = false; | |
8441 | ||
8442 | spin_lock_irq(&ctx->completion_lock); | |
8443 | list_for_each_entry(preq, &ctx->timeout_list, timeout.list) { | |
8444 | found = io_match_link(preq, req); | |
8445 | if (found) { | |
8446 | __io_timeout_cancel(preq); | |
8447 | break; | |
8448 | } | |
8449 | } | |
8450 | spin_unlock_irq(&ctx->completion_lock); | |
8451 | return found; | |
8452 | } | |
8453 | ||
b711d4ea JA |
8454 | static bool io_cancel_link_cb(struct io_wq_work *work, void *data) |
8455 | { | |
8456 | return io_match_link(container_of(work, struct io_kiocb, work), data); | |
8457 | } | |
8458 | ||
8459 | static void io_attempt_cancel(struct io_ring_ctx *ctx, struct io_kiocb *req) | |
8460 | { | |
8461 | enum io_wq_cancel cret; | |
8462 | ||
8463 | /* cancel this particular work, if it's running */ | |
8464 | cret = io_wq_cancel_work(ctx->io_wq, &req->work); | |
8465 | if (cret != IO_WQ_CANCEL_NOTFOUND) | |
8466 | return; | |
8467 | ||
8468 | /* find links that hold this pending, cancel those */ | |
8469 | cret = io_wq_cancel_cb(ctx->io_wq, io_cancel_link_cb, req, true); | |
8470 | if (cret != IO_WQ_CANCEL_NOTFOUND) | |
8471 | return; | |
8472 | ||
8473 | /* if we have a poll link holding this pending, cancel that */ | |
8474 | if (io_poll_remove_link(ctx, req)) | |
8475 | return; | |
8476 | ||
8477 | /* final option, timeout link is holding this req pending */ | |
8478 | io_timeout_remove_link(ctx, req); | |
8479 | } | |
8480 | ||
b7ddce3c PB |
8481 | static void io_cancel_defer_files(struct io_ring_ctx *ctx, |
8482 | struct files_struct *files) | |
8483 | { | |
8484 | struct io_defer_entry *de = NULL; | |
8485 | LIST_HEAD(list); | |
8486 | ||
8487 | spin_lock_irq(&ctx->completion_lock); | |
8488 | list_for_each_entry_reverse(de, &ctx->defer_list, list) { | |
c127a2a1 | 8489 | if (io_match_link_files(de->req, files)) { |
b7ddce3c PB |
8490 | list_cut_position(&list, &ctx->defer_list, &de->list); |
8491 | break; | |
8492 | } | |
8493 | } | |
8494 | spin_unlock_irq(&ctx->completion_lock); | |
8495 | ||
8496 | while (!list_empty(&list)) { | |
8497 | de = list_first_entry(&list, struct io_defer_entry, list); | |
8498 | list_del_init(&de->list); | |
8499 | req_set_fail_links(de->req); | |
8500 | io_put_req(de->req); | |
8501 | io_req_complete(de->req, -ECANCELED); | |
8502 | kfree(de); | |
8503 | } | |
8504 | } | |
8505 | ||
76e1b642 JA |
8506 | /* |
8507 | * Returns true if we found and killed one or more files pinning requests | |
8508 | */ | |
8509 | static bool io_uring_cancel_files(struct io_ring_ctx *ctx, | |
fcb323cc JA |
8510 | struct files_struct *files) |
8511 | { | |
67c4d9e6 | 8512 | if (list_empty_careful(&ctx->inflight_list)) |
76e1b642 | 8513 | return false; |
67c4d9e6 | 8514 | |
b7ddce3c | 8515 | io_cancel_defer_files(ctx, files); |
67c4d9e6 PB |
8516 | /* cancel all at once, should be faster than doing it one by one*/ |
8517 | io_wq_cancel_cb(ctx->io_wq, io_wq_files_match, files, true); | |
8518 | ||
fcb323cc | 8519 | while (!list_empty_careful(&ctx->inflight_list)) { |
d8f1b971 XW |
8520 | struct io_kiocb *cancel_req = NULL, *req; |
8521 | DEFINE_WAIT(wait); | |
fcb323cc JA |
8522 | |
8523 | spin_lock_irq(&ctx->inflight_lock); | |
8524 | list_for_each_entry(req, &ctx->inflight_list, inflight_entry) { | |
dfead8a8 | 8525 | if (files && (req->work.flags & IO_WQ_WORK_FILES) && |
98447d65 | 8526 | req->work.identity->files != files) |
768134d4 JA |
8527 | continue; |
8528 | /* req is being completed, ignore */ | |
8529 | if (!refcount_inc_not_zero(&req->refs)) | |
8530 | continue; | |
8531 | cancel_req = req; | |
8532 | break; | |
fcb323cc | 8533 | } |
768134d4 | 8534 | if (cancel_req) |
fcb323cc | 8535 | prepare_to_wait(&ctx->inflight_wait, &wait, |
768134d4 | 8536 | TASK_UNINTERRUPTIBLE); |
fcb323cc JA |
8537 | spin_unlock_irq(&ctx->inflight_lock); |
8538 | ||
768134d4 JA |
8539 | /* We need to keep going until we don't find a matching req */ |
8540 | if (!cancel_req) | |
fcb323cc | 8541 | break; |
bb175342 PB |
8542 | /* cancel this request, or head link requests */ |
8543 | io_attempt_cancel(ctx, cancel_req); | |
8544 | io_put_req(cancel_req); | |
6200b0ae JA |
8545 | /* cancellations _may_ trigger task work */ |
8546 | io_run_task_work(); | |
fcb323cc | 8547 | schedule(); |
d8f1b971 | 8548 | finish_wait(&ctx->inflight_wait, &wait); |
fcb323cc | 8549 | } |
76e1b642 JA |
8550 | |
8551 | return true; | |
fcb323cc JA |
8552 | } |
8553 | ||
801dd57b | 8554 | static bool io_cancel_task_cb(struct io_wq_work *work, void *data) |
44e728b8 | 8555 | { |
801dd57b PB |
8556 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); |
8557 | struct task_struct *task = data; | |
44e728b8 | 8558 | |
f3606e3a | 8559 | return io_task_match(req, task); |
44e728b8 PB |
8560 | } |
8561 | ||
0f212204 JA |
8562 | static bool __io_uring_cancel_task_requests(struct io_ring_ctx *ctx, |
8563 | struct task_struct *task, | |
8564 | struct files_struct *files) | |
8565 | { | |
8566 | bool ret; | |
8567 | ||
8568 | ret = io_uring_cancel_files(ctx, files); | |
8569 | if (!files) { | |
8570 | enum io_wq_cancel cret; | |
8571 | ||
8572 | cret = io_wq_cancel_cb(ctx->io_wq, io_cancel_task_cb, task, true); | |
8573 | if (cret != IO_WQ_CANCEL_NOTFOUND) | |
8574 | ret = true; | |
8575 | ||
8576 | /* SQPOLL thread does its own polling */ | |
8577 | if (!(ctx->flags & IORING_SETUP_SQPOLL)) { | |
8578 | while (!list_empty_careful(&ctx->iopoll_list)) { | |
8579 | io_iopoll_try_reap_events(ctx); | |
8580 | ret = true; | |
8581 | } | |
8582 | } | |
8583 | ||
8584 | ret |= io_poll_remove_all(ctx, task); | |
8585 | ret |= io_kill_timeouts(ctx, task); | |
8586 | } | |
8587 | ||
8588 | return ret; | |
8589 | } | |
8590 | ||
8591 | /* | |
8592 | * We need to iteratively cancel requests, in case a request has dependent | |
8593 | * hard links. These persist even for failure of cancelations, hence keep | |
8594 | * looping until none are found. | |
8595 | */ | |
8596 | static void io_uring_cancel_task_requests(struct io_ring_ctx *ctx, | |
8597 | struct files_struct *files) | |
8598 | { | |
8599 | struct task_struct *task = current; | |
8600 | ||
534ca6d6 JA |
8601 | if ((ctx->flags & IORING_SETUP_SQPOLL) && ctx->sq_data) |
8602 | task = ctx->sq_data->thread; | |
0f212204 JA |
8603 | |
8604 | io_cqring_overflow_flush(ctx, true, task, files); | |
8605 | ||
8606 | while (__io_uring_cancel_task_requests(ctx, task, files)) { | |
8607 | io_run_task_work(); | |
8608 | cond_resched(); | |
8609 | } | |
8610 | } | |
8611 | ||
8612 | /* | |
8613 | * Note that this task has used io_uring. We use it for cancelation purposes. | |
8614 | */ | |
8615 | static int io_uring_add_task_file(struct file *file) | |
8616 | { | |
236434c3 MWO |
8617 | struct io_uring_task *tctx = current->io_uring; |
8618 | ||
8619 | if (unlikely(!tctx)) { | |
0f212204 JA |
8620 | int ret; |
8621 | ||
8622 | ret = io_uring_alloc_task_context(current); | |
8623 | if (unlikely(ret)) | |
8624 | return ret; | |
236434c3 | 8625 | tctx = current->io_uring; |
0f212204 | 8626 | } |
236434c3 MWO |
8627 | if (tctx->last != file) { |
8628 | void *old = xa_load(&tctx->xa, (unsigned long)file); | |
0f212204 | 8629 | |
236434c3 | 8630 | if (!old) { |
0f212204 | 8631 | get_file(file); |
236434c3 | 8632 | xa_store(&tctx->xa, (unsigned long)file, file, GFP_KERNEL); |
0f212204 | 8633 | } |
236434c3 | 8634 | tctx->last = file; |
0f212204 JA |
8635 | } |
8636 | ||
8637 | return 0; | |
8638 | } | |
8639 | ||
8640 | /* | |
8641 | * Remove this io_uring_file -> task mapping. | |
8642 | */ | |
8643 | static void io_uring_del_task_file(struct file *file) | |
8644 | { | |
8645 | struct io_uring_task *tctx = current->io_uring; | |
0f212204 JA |
8646 | |
8647 | if (tctx->last == file) | |
8648 | tctx->last = NULL; | |
5e2ed8c4 | 8649 | file = xa_erase(&tctx->xa, (unsigned long)file); |
0f212204 JA |
8650 | if (file) |
8651 | fput(file); | |
8652 | } | |
8653 | ||
0f212204 JA |
8654 | /* |
8655 | * Drop task note for this file if we're the only ones that hold it after | |
8656 | * pending fput() | |
8657 | */ | |
c8fb20b5 | 8658 | static void io_uring_attempt_task_drop(struct file *file) |
0f212204 JA |
8659 | { |
8660 | if (!current->io_uring) | |
8661 | return; | |
8662 | /* | |
8663 | * fput() is pending, will be 2 if the only other ref is our potential | |
8664 | * task file note. If the task is exiting, drop regardless of count. | |
8665 | */ | |
c8fb20b5 PB |
8666 | if (fatal_signal_pending(current) || (current->flags & PF_EXITING) || |
8667 | atomic_long_read(&file->f_count) == 2) | |
8668 | io_uring_del_task_file(file); | |
0f212204 JA |
8669 | } |
8670 | ||
8671 | void __io_uring_files_cancel(struct files_struct *files) | |
8672 | { | |
8673 | struct io_uring_task *tctx = current->io_uring; | |
ce765372 MWO |
8674 | struct file *file; |
8675 | unsigned long index; | |
0f212204 JA |
8676 | |
8677 | /* make sure overflow events are dropped */ | |
8678 | tctx->in_idle = true; | |
8679 | ||
ce765372 MWO |
8680 | xa_for_each(&tctx->xa, index, file) { |
8681 | struct io_ring_ctx *ctx = file->private_data; | |
0f212204 JA |
8682 | |
8683 | io_uring_cancel_task_requests(ctx, files); | |
8684 | if (files) | |
8685 | io_uring_del_task_file(file); | |
ce765372 | 8686 | } |
0f212204 JA |
8687 | } |
8688 | ||
0f212204 JA |
8689 | /* |
8690 | * Find any io_uring fd that this task has registered or done IO on, and cancel | |
8691 | * requests. | |
8692 | */ | |
8693 | void __io_uring_task_cancel(void) | |
8694 | { | |
8695 | struct io_uring_task *tctx = current->io_uring; | |
8696 | DEFINE_WAIT(wait); | |
d8a6df10 | 8697 | s64 inflight; |
0f212204 JA |
8698 | |
8699 | /* make sure overflow events are dropped */ | |
8700 | tctx->in_idle = true; | |
8701 | ||
d8a6df10 | 8702 | do { |
0f212204 | 8703 | /* read completions before cancelations */ |
d8a6df10 JA |
8704 | inflight = percpu_counter_sum(&tctx->inflight); |
8705 | if (!inflight) | |
8706 | break; | |
0f212204 JA |
8707 | __io_uring_files_cancel(NULL); |
8708 | ||
8709 | prepare_to_wait(&tctx->wait, &wait, TASK_UNINTERRUPTIBLE); | |
8710 | ||
8711 | /* | |
8712 | * If we've seen completions, retry. This avoids a race where | |
8713 | * a completion comes in before we did prepare_to_wait(). | |
8714 | */ | |
d8a6df10 | 8715 | if (inflight != percpu_counter_sum(&tctx->inflight)) |
0f212204 | 8716 | continue; |
0f212204 | 8717 | schedule(); |
d8a6df10 | 8718 | } while (1); |
0f212204 JA |
8719 | |
8720 | finish_wait(&tctx->wait, &wait); | |
8721 | tctx->in_idle = false; | |
44e728b8 PB |
8722 | } |
8723 | ||
fcb323cc JA |
8724 | static int io_uring_flush(struct file *file, void *data) |
8725 | { | |
c8fb20b5 | 8726 | io_uring_attempt_task_drop(file); |
fcb323cc JA |
8727 | return 0; |
8728 | } | |
8729 | ||
6c5c240e RP |
8730 | static void *io_uring_validate_mmap_request(struct file *file, |
8731 | loff_t pgoff, size_t sz) | |
2b188cc1 | 8732 | { |
2b188cc1 | 8733 | struct io_ring_ctx *ctx = file->private_data; |
6c5c240e | 8734 | loff_t offset = pgoff << PAGE_SHIFT; |
2b188cc1 JA |
8735 | struct page *page; |
8736 | void *ptr; | |
8737 | ||
8738 | switch (offset) { | |
8739 | case IORING_OFF_SQ_RING: | |
75b28aff HV |
8740 | case IORING_OFF_CQ_RING: |
8741 | ptr = ctx->rings; | |
2b188cc1 JA |
8742 | break; |
8743 | case IORING_OFF_SQES: | |
8744 | ptr = ctx->sq_sqes; | |
8745 | break; | |
2b188cc1 | 8746 | default: |
6c5c240e | 8747 | return ERR_PTR(-EINVAL); |
2b188cc1 JA |
8748 | } |
8749 | ||
8750 | page = virt_to_head_page(ptr); | |
a50b854e | 8751 | if (sz > page_size(page)) |
6c5c240e RP |
8752 | return ERR_PTR(-EINVAL); |
8753 | ||
8754 | return ptr; | |
8755 | } | |
8756 | ||
8757 | #ifdef CONFIG_MMU | |
8758 | ||
8759 | static int io_uring_mmap(struct file *file, struct vm_area_struct *vma) | |
8760 | { | |
8761 | size_t sz = vma->vm_end - vma->vm_start; | |
8762 | unsigned long pfn; | |
8763 | void *ptr; | |
8764 | ||
8765 | ptr = io_uring_validate_mmap_request(file, vma->vm_pgoff, sz); | |
8766 | if (IS_ERR(ptr)) | |
8767 | return PTR_ERR(ptr); | |
2b188cc1 JA |
8768 | |
8769 | pfn = virt_to_phys(ptr) >> PAGE_SHIFT; | |
8770 | return remap_pfn_range(vma, vma->vm_start, pfn, sz, vma->vm_page_prot); | |
8771 | } | |
8772 | ||
6c5c240e RP |
8773 | #else /* !CONFIG_MMU */ |
8774 | ||
8775 | static int io_uring_mmap(struct file *file, struct vm_area_struct *vma) | |
8776 | { | |
8777 | return vma->vm_flags & (VM_SHARED | VM_MAYSHARE) ? 0 : -EINVAL; | |
8778 | } | |
8779 | ||
8780 | static unsigned int io_uring_nommu_mmap_capabilities(struct file *file) | |
8781 | { | |
8782 | return NOMMU_MAP_DIRECT | NOMMU_MAP_READ | NOMMU_MAP_WRITE; | |
8783 | } | |
8784 | ||
8785 | static unsigned long io_uring_nommu_get_unmapped_area(struct file *file, | |
8786 | unsigned long addr, unsigned long len, | |
8787 | unsigned long pgoff, unsigned long flags) | |
8788 | { | |
8789 | void *ptr; | |
8790 | ||
8791 | ptr = io_uring_validate_mmap_request(file, pgoff, len); | |
8792 | if (IS_ERR(ptr)) | |
8793 | return PTR_ERR(ptr); | |
8794 | ||
8795 | return (unsigned long) ptr; | |
8796 | } | |
8797 | ||
8798 | #endif /* !CONFIG_MMU */ | |
8799 | ||
90554200 JA |
8800 | static void io_sqpoll_wait_sq(struct io_ring_ctx *ctx) |
8801 | { | |
8802 | DEFINE_WAIT(wait); | |
8803 | ||
8804 | do { | |
8805 | if (!io_sqring_full(ctx)) | |
8806 | break; | |
8807 | ||
8808 | prepare_to_wait(&ctx->sqo_sq_wait, &wait, TASK_INTERRUPTIBLE); | |
8809 | ||
8810 | if (!io_sqring_full(ctx)) | |
8811 | break; | |
8812 | ||
8813 | schedule(); | |
8814 | } while (!signal_pending(current)); | |
8815 | ||
8816 | finish_wait(&ctx->sqo_sq_wait, &wait); | |
8817 | } | |
8818 | ||
2b188cc1 JA |
8819 | SYSCALL_DEFINE6(io_uring_enter, unsigned int, fd, u32, to_submit, |
8820 | u32, min_complete, u32, flags, const sigset_t __user *, sig, | |
8821 | size_t, sigsz) | |
8822 | { | |
8823 | struct io_ring_ctx *ctx; | |
8824 | long ret = -EBADF; | |
8825 | int submitted = 0; | |
8826 | struct fd f; | |
8827 | ||
4c6e277c | 8828 | io_run_task_work(); |
b41e9852 | 8829 | |
90554200 JA |
8830 | if (flags & ~(IORING_ENTER_GETEVENTS | IORING_ENTER_SQ_WAKEUP | |
8831 | IORING_ENTER_SQ_WAIT)) | |
2b188cc1 JA |
8832 | return -EINVAL; |
8833 | ||
8834 | f = fdget(fd); | |
8835 | if (!f.file) | |
8836 | return -EBADF; | |
8837 | ||
8838 | ret = -EOPNOTSUPP; | |
8839 | if (f.file->f_op != &io_uring_fops) | |
8840 | goto out_fput; | |
8841 | ||
8842 | ret = -ENXIO; | |
8843 | ctx = f.file->private_data; | |
8844 | if (!percpu_ref_tryget(&ctx->refs)) | |
8845 | goto out_fput; | |
8846 | ||
7e84e1c7 SG |
8847 | ret = -EBADFD; |
8848 | if (ctx->flags & IORING_SETUP_R_DISABLED) | |
8849 | goto out; | |
8850 | ||
6c271ce2 JA |
8851 | /* |
8852 | * For SQ polling, the thread will do all submissions and completions. | |
8853 | * Just return the requested submit count, and wake the thread if | |
8854 | * we were asked to. | |
8855 | */ | |
b2a9eada | 8856 | ret = 0; |
6c271ce2 | 8857 | if (ctx->flags & IORING_SETUP_SQPOLL) { |
c1edbf5f | 8858 | if (!list_empty_careful(&ctx->cq_overflow_list)) |
e6c8aa9a | 8859 | io_cqring_overflow_flush(ctx, false, NULL, NULL); |
6c271ce2 | 8860 | if (flags & IORING_ENTER_SQ_WAKEUP) |
534ca6d6 | 8861 | wake_up(&ctx->sq_data->wait); |
90554200 JA |
8862 | if (flags & IORING_ENTER_SQ_WAIT) |
8863 | io_sqpoll_wait_sq(ctx); | |
6c271ce2 | 8864 | submitted = to_submit; |
b2a9eada | 8865 | } else if (to_submit) { |
0f212204 JA |
8866 | ret = io_uring_add_task_file(f.file); |
8867 | if (unlikely(ret)) | |
8868 | goto out; | |
2b188cc1 | 8869 | mutex_lock(&ctx->uring_lock); |
0f212204 | 8870 | submitted = io_submit_sqes(ctx, to_submit); |
2b188cc1 | 8871 | mutex_unlock(&ctx->uring_lock); |
7c504e65 PB |
8872 | |
8873 | if (submitted != to_submit) | |
8874 | goto out; | |
2b188cc1 JA |
8875 | } |
8876 | if (flags & IORING_ENTER_GETEVENTS) { | |
8877 | min_complete = min(min_complete, ctx->cq_entries); | |
8878 | ||
32b2244a XW |
8879 | /* |
8880 | * When SETUP_IOPOLL and SETUP_SQPOLL are both enabled, user | |
8881 | * space applications don't need to do io completion events | |
8882 | * polling again, they can rely on io_sq_thread to do polling | |
8883 | * work, which can reduce cpu usage and uring_lock contention. | |
8884 | */ | |
8885 | if (ctx->flags & IORING_SETUP_IOPOLL && | |
8886 | !(ctx->flags & IORING_SETUP_SQPOLL)) { | |
7668b92a | 8887 | ret = io_iopoll_check(ctx, min_complete); |
def596e9 JA |
8888 | } else { |
8889 | ret = io_cqring_wait(ctx, min_complete, sig, sigsz); | |
8890 | } | |
2b188cc1 JA |
8891 | } |
8892 | ||
7c504e65 | 8893 | out: |
6805b32e | 8894 | percpu_ref_put(&ctx->refs); |
2b188cc1 JA |
8895 | out_fput: |
8896 | fdput(f); | |
8897 | return submitted ? submitted : ret; | |
8898 | } | |
8899 | ||
bebdb65e | 8900 | #ifdef CONFIG_PROC_FS |
87ce955b JA |
8901 | static int io_uring_show_cred(int id, void *p, void *data) |
8902 | { | |
8903 | const struct cred *cred = p; | |
8904 | struct seq_file *m = data; | |
8905 | struct user_namespace *uns = seq_user_ns(m); | |
8906 | struct group_info *gi; | |
8907 | kernel_cap_t cap; | |
8908 | unsigned __capi; | |
8909 | int g; | |
8910 | ||
8911 | seq_printf(m, "%5d\n", id); | |
8912 | seq_put_decimal_ull(m, "\tUid:\t", from_kuid_munged(uns, cred->uid)); | |
8913 | seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->euid)); | |
8914 | seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->suid)); | |
8915 | seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->fsuid)); | |
8916 | seq_put_decimal_ull(m, "\n\tGid:\t", from_kgid_munged(uns, cred->gid)); | |
8917 | seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->egid)); | |
8918 | seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->sgid)); | |
8919 | seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->fsgid)); | |
8920 | seq_puts(m, "\n\tGroups:\t"); | |
8921 | gi = cred->group_info; | |
8922 | for (g = 0; g < gi->ngroups; g++) { | |
8923 | seq_put_decimal_ull(m, g ? " " : "", | |
8924 | from_kgid_munged(uns, gi->gid[g])); | |
8925 | } | |
8926 | seq_puts(m, "\n\tCapEff:\t"); | |
8927 | cap = cred->cap_effective; | |
8928 | CAP_FOR_EACH_U32(__capi) | |
8929 | seq_put_hex_ll(m, NULL, cap.cap[CAP_LAST_U32 - __capi], 8); | |
8930 | seq_putc(m, '\n'); | |
8931 | return 0; | |
8932 | } | |
8933 | ||
8934 | static void __io_uring_show_fdinfo(struct io_ring_ctx *ctx, struct seq_file *m) | |
8935 | { | |
dbbe9c64 | 8936 | struct io_sq_data *sq = NULL; |
fad8e0de | 8937 | bool has_lock; |
87ce955b JA |
8938 | int i; |
8939 | ||
fad8e0de JA |
8940 | /* |
8941 | * Avoid ABBA deadlock between the seq lock and the io_uring mutex, | |
8942 | * since fdinfo case grabs it in the opposite direction of normal use | |
8943 | * cases. If we fail to get the lock, we just don't iterate any | |
8944 | * structures that could be going away outside the io_uring mutex. | |
8945 | */ | |
8946 | has_lock = mutex_trylock(&ctx->uring_lock); | |
8947 | ||
dbbe9c64 JQ |
8948 | if (has_lock && (ctx->flags & IORING_SETUP_SQPOLL)) |
8949 | sq = ctx->sq_data; | |
8950 | ||
8951 | seq_printf(m, "SqThread:\t%d\n", sq ? task_pid_nr(sq->thread) : -1); | |
8952 | seq_printf(m, "SqThreadCpu:\t%d\n", sq ? task_cpu(sq->thread) : -1); | |
87ce955b | 8953 | seq_printf(m, "UserFiles:\t%u\n", ctx->nr_user_files); |
fad8e0de | 8954 | for (i = 0; has_lock && i < ctx->nr_user_files; i++) { |
87ce955b JA |
8955 | struct fixed_file_table *table; |
8956 | struct file *f; | |
8957 | ||
8958 | table = &ctx->file_data->table[i >> IORING_FILE_TABLE_SHIFT]; | |
8959 | f = table->files[i & IORING_FILE_TABLE_MASK]; | |
8960 | if (f) | |
8961 | seq_printf(m, "%5u: %s\n", i, file_dentry(f)->d_iname); | |
8962 | else | |
8963 | seq_printf(m, "%5u: <none>\n", i); | |
8964 | } | |
8965 | seq_printf(m, "UserBufs:\t%u\n", ctx->nr_user_bufs); | |
fad8e0de | 8966 | for (i = 0; has_lock && i < ctx->nr_user_bufs; i++) { |
87ce955b JA |
8967 | struct io_mapped_ubuf *buf = &ctx->user_bufs[i]; |
8968 | ||
8969 | seq_printf(m, "%5u: 0x%llx/%u\n", i, buf->ubuf, | |
8970 | (unsigned int) buf->len); | |
8971 | } | |
fad8e0de | 8972 | if (has_lock && !idr_is_empty(&ctx->personality_idr)) { |
87ce955b JA |
8973 | seq_printf(m, "Personalities:\n"); |
8974 | idr_for_each(&ctx->personality_idr, io_uring_show_cred, m); | |
8975 | } | |
d7718a9d JA |
8976 | seq_printf(m, "PollList:\n"); |
8977 | spin_lock_irq(&ctx->completion_lock); | |
8978 | for (i = 0; i < (1U << ctx->cancel_hash_bits); i++) { | |
8979 | struct hlist_head *list = &ctx->cancel_hash[i]; | |
8980 | struct io_kiocb *req; | |
8981 | ||
8982 | hlist_for_each_entry(req, list, hash_node) | |
8983 | seq_printf(m, " op=%d, task_works=%d\n", req->opcode, | |
8984 | req->task->task_works != NULL); | |
8985 | } | |
8986 | spin_unlock_irq(&ctx->completion_lock); | |
fad8e0de JA |
8987 | if (has_lock) |
8988 | mutex_unlock(&ctx->uring_lock); | |
87ce955b JA |
8989 | } |
8990 | ||
8991 | static void io_uring_show_fdinfo(struct seq_file *m, struct file *f) | |
8992 | { | |
8993 | struct io_ring_ctx *ctx = f->private_data; | |
8994 | ||
8995 | if (percpu_ref_tryget(&ctx->refs)) { | |
8996 | __io_uring_show_fdinfo(ctx, m); | |
8997 | percpu_ref_put(&ctx->refs); | |
8998 | } | |
8999 | } | |
bebdb65e | 9000 | #endif |
87ce955b | 9001 | |
2b188cc1 JA |
9002 | static const struct file_operations io_uring_fops = { |
9003 | .release = io_uring_release, | |
fcb323cc | 9004 | .flush = io_uring_flush, |
2b188cc1 | 9005 | .mmap = io_uring_mmap, |
6c5c240e RP |
9006 | #ifndef CONFIG_MMU |
9007 | .get_unmapped_area = io_uring_nommu_get_unmapped_area, | |
9008 | .mmap_capabilities = io_uring_nommu_mmap_capabilities, | |
9009 | #endif | |
2b188cc1 JA |
9010 | .poll = io_uring_poll, |
9011 | .fasync = io_uring_fasync, | |
bebdb65e | 9012 | #ifdef CONFIG_PROC_FS |
87ce955b | 9013 | .show_fdinfo = io_uring_show_fdinfo, |
bebdb65e | 9014 | #endif |
2b188cc1 JA |
9015 | }; |
9016 | ||
9017 | static int io_allocate_scq_urings(struct io_ring_ctx *ctx, | |
9018 | struct io_uring_params *p) | |
9019 | { | |
75b28aff HV |
9020 | struct io_rings *rings; |
9021 | size_t size, sq_array_offset; | |
2b188cc1 | 9022 | |
bd740481 JA |
9023 | /* make sure these are sane, as we already accounted them */ |
9024 | ctx->sq_entries = p->sq_entries; | |
9025 | ctx->cq_entries = p->cq_entries; | |
9026 | ||
75b28aff HV |
9027 | size = rings_size(p->sq_entries, p->cq_entries, &sq_array_offset); |
9028 | if (size == SIZE_MAX) | |
9029 | return -EOVERFLOW; | |
9030 | ||
9031 | rings = io_mem_alloc(size); | |
9032 | if (!rings) | |
2b188cc1 JA |
9033 | return -ENOMEM; |
9034 | ||
75b28aff HV |
9035 | ctx->rings = rings; |
9036 | ctx->sq_array = (u32 *)((char *)rings + sq_array_offset); | |
9037 | rings->sq_ring_mask = p->sq_entries - 1; | |
9038 | rings->cq_ring_mask = p->cq_entries - 1; | |
9039 | rings->sq_ring_entries = p->sq_entries; | |
9040 | rings->cq_ring_entries = p->cq_entries; | |
9041 | ctx->sq_mask = rings->sq_ring_mask; | |
9042 | ctx->cq_mask = rings->cq_ring_mask; | |
2b188cc1 JA |
9043 | |
9044 | size = array_size(sizeof(struct io_uring_sqe), p->sq_entries); | |
eb065d30 JA |
9045 | if (size == SIZE_MAX) { |
9046 | io_mem_free(ctx->rings); | |
9047 | ctx->rings = NULL; | |
2b188cc1 | 9048 | return -EOVERFLOW; |
eb065d30 | 9049 | } |
2b188cc1 JA |
9050 | |
9051 | ctx->sq_sqes = io_mem_alloc(size); | |
eb065d30 JA |
9052 | if (!ctx->sq_sqes) { |
9053 | io_mem_free(ctx->rings); | |
9054 | ctx->rings = NULL; | |
2b188cc1 | 9055 | return -ENOMEM; |
eb065d30 | 9056 | } |
2b188cc1 | 9057 | |
2b188cc1 JA |
9058 | return 0; |
9059 | } | |
9060 | ||
9061 | /* | |
9062 | * Allocate an anonymous fd, this is what constitutes the application | |
9063 | * visible backing of an io_uring instance. The application mmaps this | |
9064 | * fd to gain access to the SQ/CQ ring details. If UNIX sockets are enabled, | |
9065 | * we have to tie this fd to a socket for file garbage collection purposes. | |
9066 | */ | |
9067 | static int io_uring_get_fd(struct io_ring_ctx *ctx) | |
9068 | { | |
9069 | struct file *file; | |
9070 | int ret; | |
9071 | ||
9072 | #if defined(CONFIG_UNIX) | |
9073 | ret = sock_create_kern(&init_net, PF_UNIX, SOCK_RAW, IPPROTO_IP, | |
9074 | &ctx->ring_sock); | |
9075 | if (ret) | |
9076 | return ret; | |
9077 | #endif | |
9078 | ||
9079 | ret = get_unused_fd_flags(O_RDWR | O_CLOEXEC); | |
9080 | if (ret < 0) | |
9081 | goto err; | |
9082 | ||
9083 | file = anon_inode_getfile("[io_uring]", &io_uring_fops, ctx, | |
9084 | O_RDWR | O_CLOEXEC); | |
9085 | if (IS_ERR(file)) { | |
0f212204 | 9086 | err_fd: |
2b188cc1 JA |
9087 | put_unused_fd(ret); |
9088 | ret = PTR_ERR(file); | |
9089 | goto err; | |
9090 | } | |
9091 | ||
9092 | #if defined(CONFIG_UNIX) | |
9093 | ctx->ring_sock->file = file; | |
9094 | #endif | |
0f212204 JA |
9095 | if (unlikely(io_uring_add_task_file(file))) { |
9096 | file = ERR_PTR(-ENOMEM); | |
9097 | goto err_fd; | |
9098 | } | |
2b188cc1 JA |
9099 | fd_install(ret, file); |
9100 | return ret; | |
9101 | err: | |
9102 | #if defined(CONFIG_UNIX) | |
9103 | sock_release(ctx->ring_sock); | |
9104 | ctx->ring_sock = NULL; | |
9105 | #endif | |
9106 | return ret; | |
9107 | } | |
9108 | ||
7f13657d XW |
9109 | static int io_uring_create(unsigned entries, struct io_uring_params *p, |
9110 | struct io_uring_params __user *params) | |
2b188cc1 JA |
9111 | { |
9112 | struct user_struct *user = NULL; | |
9113 | struct io_ring_ctx *ctx; | |
aad5d8da | 9114 | bool limit_mem; |
2b188cc1 JA |
9115 | int ret; |
9116 | ||
8110c1a6 | 9117 | if (!entries) |
2b188cc1 | 9118 | return -EINVAL; |
8110c1a6 JA |
9119 | if (entries > IORING_MAX_ENTRIES) { |
9120 | if (!(p->flags & IORING_SETUP_CLAMP)) | |
9121 | return -EINVAL; | |
9122 | entries = IORING_MAX_ENTRIES; | |
9123 | } | |
2b188cc1 JA |
9124 | |
9125 | /* | |
9126 | * Use twice as many entries for the CQ ring. It's possible for the | |
9127 | * application to drive a higher depth than the size of the SQ ring, | |
9128 | * since the sqes are only used at submission time. This allows for | |
33a107f0 JA |
9129 | * some flexibility in overcommitting a bit. If the application has |
9130 | * set IORING_SETUP_CQSIZE, it will have passed in the desired number | |
9131 | * of CQ ring entries manually. | |
2b188cc1 JA |
9132 | */ |
9133 | p->sq_entries = roundup_pow_of_two(entries); | |
33a107f0 JA |
9134 | if (p->flags & IORING_SETUP_CQSIZE) { |
9135 | /* | |
9136 | * If IORING_SETUP_CQSIZE is set, we do the same roundup | |
9137 | * to a power-of-two, if it isn't already. We do NOT impose | |
9138 | * any cq vs sq ring sizing. | |
9139 | */ | |
8110c1a6 | 9140 | if (p->cq_entries < p->sq_entries) |
33a107f0 | 9141 | return -EINVAL; |
8110c1a6 JA |
9142 | if (p->cq_entries > IORING_MAX_CQ_ENTRIES) { |
9143 | if (!(p->flags & IORING_SETUP_CLAMP)) | |
9144 | return -EINVAL; | |
9145 | p->cq_entries = IORING_MAX_CQ_ENTRIES; | |
9146 | } | |
33a107f0 JA |
9147 | p->cq_entries = roundup_pow_of_two(p->cq_entries); |
9148 | } else { | |
9149 | p->cq_entries = 2 * p->sq_entries; | |
9150 | } | |
2b188cc1 JA |
9151 | |
9152 | user = get_uid(current_user()); | |
aad5d8da | 9153 | limit_mem = !capable(CAP_IPC_LOCK); |
2b188cc1 | 9154 | |
aad5d8da | 9155 | if (limit_mem) { |
a087e2b5 | 9156 | ret = __io_account_mem(user, |
2b188cc1 JA |
9157 | ring_pages(p->sq_entries, p->cq_entries)); |
9158 | if (ret) { | |
9159 | free_uid(user); | |
9160 | return ret; | |
9161 | } | |
9162 | } | |
9163 | ||
9164 | ctx = io_ring_ctx_alloc(p); | |
9165 | if (!ctx) { | |
aad5d8da | 9166 | if (limit_mem) |
a087e2b5 | 9167 | __io_unaccount_mem(user, ring_pages(p->sq_entries, |
2b188cc1 JA |
9168 | p->cq_entries)); |
9169 | free_uid(user); | |
9170 | return -ENOMEM; | |
9171 | } | |
9172 | ctx->compat = in_compat_syscall(); | |
2b188cc1 | 9173 | ctx->user = user; |
0b8c0ec7 | 9174 | ctx->creds = get_current_cred(); |
4ea33a97 JA |
9175 | #ifdef CONFIG_AUDIT |
9176 | ctx->loginuid = current->loginuid; | |
9177 | ctx->sessionid = current->sessionid; | |
9178 | #endif | |
2aede0e4 JA |
9179 | ctx->sqo_task = get_task_struct(current); |
9180 | ||
9181 | /* | |
9182 | * This is just grabbed for accounting purposes. When a process exits, | |
9183 | * the mm is exited and dropped before the files, hence we need to hang | |
9184 | * on to this mm purely for the purposes of being able to unaccount | |
9185 | * memory (locked/pinned vm). It's not used for anything else. | |
9186 | */ | |
6b7898eb | 9187 | mmgrab(current->mm); |
2aede0e4 | 9188 | ctx->mm_account = current->mm; |
6b7898eb | 9189 | |
91d8f519 DZ |
9190 | #ifdef CONFIG_BLK_CGROUP |
9191 | /* | |
9192 | * The sq thread will belong to the original cgroup it was inited in. | |
9193 | * If the cgroup goes offline (e.g. disabling the io controller), then | |
9194 | * issued bios will be associated with the closest cgroup later in the | |
9195 | * block layer. | |
9196 | */ | |
9197 | rcu_read_lock(); | |
9198 | ctx->sqo_blkcg_css = blkcg_css(); | |
9199 | ret = css_tryget_online(ctx->sqo_blkcg_css); | |
9200 | rcu_read_unlock(); | |
9201 | if (!ret) { | |
9202 | /* don't init against a dying cgroup, have the user try again */ | |
9203 | ctx->sqo_blkcg_css = NULL; | |
9204 | ret = -ENODEV; | |
9205 | goto err; | |
9206 | } | |
9207 | #endif | |
6b7898eb | 9208 | |
f74441e6 JA |
9209 | /* |
9210 | * Account memory _before_ installing the file descriptor. Once | |
9211 | * the descriptor is installed, it can get closed at any time. Also | |
9212 | * do this before hitting the general error path, as ring freeing | |
9213 | * will un-account as well. | |
9214 | */ | |
9215 | io_account_mem(ctx, ring_pages(p->sq_entries, p->cq_entries), | |
9216 | ACCT_LOCKED); | |
9217 | ctx->limit_mem = limit_mem; | |
9218 | ||
2b188cc1 JA |
9219 | ret = io_allocate_scq_urings(ctx, p); |
9220 | if (ret) | |
9221 | goto err; | |
9222 | ||
7e84e1c7 | 9223 | ret = io_sq_offload_create(ctx, p); |
2b188cc1 JA |
9224 | if (ret) |
9225 | goto err; | |
9226 | ||
7e84e1c7 SG |
9227 | if (!(p->flags & IORING_SETUP_R_DISABLED)) |
9228 | io_sq_offload_start(ctx); | |
9229 | ||
2b188cc1 | 9230 | memset(&p->sq_off, 0, sizeof(p->sq_off)); |
75b28aff HV |
9231 | p->sq_off.head = offsetof(struct io_rings, sq.head); |
9232 | p->sq_off.tail = offsetof(struct io_rings, sq.tail); | |
9233 | p->sq_off.ring_mask = offsetof(struct io_rings, sq_ring_mask); | |
9234 | p->sq_off.ring_entries = offsetof(struct io_rings, sq_ring_entries); | |
9235 | p->sq_off.flags = offsetof(struct io_rings, sq_flags); | |
9236 | p->sq_off.dropped = offsetof(struct io_rings, sq_dropped); | |
9237 | p->sq_off.array = (char *)ctx->sq_array - (char *)ctx->rings; | |
2b188cc1 JA |
9238 | |
9239 | memset(&p->cq_off, 0, sizeof(p->cq_off)); | |
75b28aff HV |
9240 | p->cq_off.head = offsetof(struct io_rings, cq.head); |
9241 | p->cq_off.tail = offsetof(struct io_rings, cq.tail); | |
9242 | p->cq_off.ring_mask = offsetof(struct io_rings, cq_ring_mask); | |
9243 | p->cq_off.ring_entries = offsetof(struct io_rings, cq_ring_entries); | |
9244 | p->cq_off.overflow = offsetof(struct io_rings, cq_overflow); | |
9245 | p->cq_off.cqes = offsetof(struct io_rings, cqes); | |
0d9b5b3a | 9246 | p->cq_off.flags = offsetof(struct io_rings, cq_flags); |
ac90f249 | 9247 | |
7f13657d XW |
9248 | p->features = IORING_FEAT_SINGLE_MMAP | IORING_FEAT_NODROP | |
9249 | IORING_FEAT_SUBMIT_STABLE | IORING_FEAT_RW_CUR_POS | | |
5769a351 JX |
9250 | IORING_FEAT_CUR_PERSONALITY | IORING_FEAT_FAST_POLL | |
9251 | IORING_FEAT_POLL_32BITS; | |
7f13657d XW |
9252 | |
9253 | if (copy_to_user(params, p, sizeof(*p))) { | |
9254 | ret = -EFAULT; | |
9255 | goto err; | |
9256 | } | |
d1719f70 | 9257 | |
044c1ab3 JA |
9258 | /* |
9259 | * Install ring fd as the very last thing, so we don't risk someone | |
9260 | * having closed it before we finish setup | |
9261 | */ | |
9262 | ret = io_uring_get_fd(ctx); | |
9263 | if (ret < 0) | |
9264 | goto err; | |
9265 | ||
c826bd7a | 9266 | trace_io_uring_create(ret, ctx, p->sq_entries, p->cq_entries, p->flags); |
2b188cc1 JA |
9267 | return ret; |
9268 | err: | |
9269 | io_ring_ctx_wait_and_kill(ctx); | |
9270 | return ret; | |
9271 | } | |
9272 | ||
9273 | /* | |
9274 | * Sets up an aio uring context, and returns the fd. Applications asks for a | |
9275 | * ring size, we return the actual sq/cq ring sizes (among other things) in the | |
9276 | * params structure passed in. | |
9277 | */ | |
9278 | static long io_uring_setup(u32 entries, struct io_uring_params __user *params) | |
9279 | { | |
9280 | struct io_uring_params p; | |
2b188cc1 JA |
9281 | int i; |
9282 | ||
9283 | if (copy_from_user(&p, params, sizeof(p))) | |
9284 | return -EFAULT; | |
9285 | for (i = 0; i < ARRAY_SIZE(p.resv); i++) { | |
9286 | if (p.resv[i]) | |
9287 | return -EINVAL; | |
9288 | } | |
9289 | ||
6c271ce2 | 9290 | if (p.flags & ~(IORING_SETUP_IOPOLL | IORING_SETUP_SQPOLL | |
8110c1a6 | 9291 | IORING_SETUP_SQ_AFF | IORING_SETUP_CQSIZE | |
7e84e1c7 SG |
9292 | IORING_SETUP_CLAMP | IORING_SETUP_ATTACH_WQ | |
9293 | IORING_SETUP_R_DISABLED)) | |
2b188cc1 JA |
9294 | return -EINVAL; |
9295 | ||
7f13657d | 9296 | return io_uring_create(entries, &p, params); |
2b188cc1 JA |
9297 | } |
9298 | ||
9299 | SYSCALL_DEFINE2(io_uring_setup, u32, entries, | |
9300 | struct io_uring_params __user *, params) | |
9301 | { | |
9302 | return io_uring_setup(entries, params); | |
9303 | } | |
9304 | ||
66f4af93 JA |
9305 | static int io_probe(struct io_ring_ctx *ctx, void __user *arg, unsigned nr_args) |
9306 | { | |
9307 | struct io_uring_probe *p; | |
9308 | size_t size; | |
9309 | int i, ret; | |
9310 | ||
9311 | size = struct_size(p, ops, nr_args); | |
9312 | if (size == SIZE_MAX) | |
9313 | return -EOVERFLOW; | |
9314 | p = kzalloc(size, GFP_KERNEL); | |
9315 | if (!p) | |
9316 | return -ENOMEM; | |
9317 | ||
9318 | ret = -EFAULT; | |
9319 | if (copy_from_user(p, arg, size)) | |
9320 | goto out; | |
9321 | ret = -EINVAL; | |
9322 | if (memchr_inv(p, 0, size)) | |
9323 | goto out; | |
9324 | ||
9325 | p->last_op = IORING_OP_LAST - 1; | |
9326 | if (nr_args > IORING_OP_LAST) | |
9327 | nr_args = IORING_OP_LAST; | |
9328 | ||
9329 | for (i = 0; i < nr_args; i++) { | |
9330 | p->ops[i].op = i; | |
9331 | if (!io_op_defs[i].not_supported) | |
9332 | p->ops[i].flags = IO_URING_OP_SUPPORTED; | |
9333 | } | |
9334 | p->ops_len = i; | |
9335 | ||
9336 | ret = 0; | |
9337 | if (copy_to_user(arg, p, size)) | |
9338 | ret = -EFAULT; | |
9339 | out: | |
9340 | kfree(p); | |
9341 | return ret; | |
9342 | } | |
9343 | ||
071698e1 JA |
9344 | static int io_register_personality(struct io_ring_ctx *ctx) |
9345 | { | |
1e6fa521 JA |
9346 | struct io_identity *id; |
9347 | int ret; | |
071698e1 | 9348 | |
1e6fa521 JA |
9349 | id = kmalloc(sizeof(*id), GFP_KERNEL); |
9350 | if (unlikely(!id)) | |
9351 | return -ENOMEM; | |
9352 | ||
9353 | io_init_identity(id); | |
9354 | id->creds = get_current_cred(); | |
9355 | ||
9356 | ret = idr_alloc_cyclic(&ctx->personality_idr, id, 1, USHRT_MAX, GFP_KERNEL); | |
9357 | if (ret < 0) { | |
9358 | put_cred(id->creds); | |
9359 | kfree(id); | |
9360 | } | |
9361 | return ret; | |
071698e1 JA |
9362 | } |
9363 | ||
9364 | static int io_unregister_personality(struct io_ring_ctx *ctx, unsigned id) | |
9365 | { | |
1e6fa521 | 9366 | struct io_identity *iod; |
071698e1 | 9367 | |
1e6fa521 JA |
9368 | iod = idr_remove(&ctx->personality_idr, id); |
9369 | if (iod) { | |
9370 | put_cred(iod->creds); | |
9371 | if (refcount_dec_and_test(&iod->count)) | |
9372 | kfree(iod); | |
071698e1 JA |
9373 | return 0; |
9374 | } | |
9375 | ||
9376 | return -EINVAL; | |
9377 | } | |
9378 | ||
21b55dbc SG |
9379 | static int io_register_restrictions(struct io_ring_ctx *ctx, void __user *arg, |
9380 | unsigned int nr_args) | |
9381 | { | |
9382 | struct io_uring_restriction *res; | |
9383 | size_t size; | |
9384 | int i, ret; | |
9385 | ||
7e84e1c7 SG |
9386 | /* Restrictions allowed only if rings started disabled */ |
9387 | if (!(ctx->flags & IORING_SETUP_R_DISABLED)) | |
9388 | return -EBADFD; | |
9389 | ||
21b55dbc | 9390 | /* We allow only a single restrictions registration */ |
7e84e1c7 | 9391 | if (ctx->restrictions.registered) |
21b55dbc SG |
9392 | return -EBUSY; |
9393 | ||
9394 | if (!arg || nr_args > IORING_MAX_RESTRICTIONS) | |
9395 | return -EINVAL; | |
9396 | ||
9397 | size = array_size(nr_args, sizeof(*res)); | |
9398 | if (size == SIZE_MAX) | |
9399 | return -EOVERFLOW; | |
9400 | ||
9401 | res = memdup_user(arg, size); | |
9402 | if (IS_ERR(res)) | |
9403 | return PTR_ERR(res); | |
9404 | ||
9405 | ret = 0; | |
9406 | ||
9407 | for (i = 0; i < nr_args; i++) { | |
9408 | switch (res[i].opcode) { | |
9409 | case IORING_RESTRICTION_REGISTER_OP: | |
9410 | if (res[i].register_op >= IORING_REGISTER_LAST) { | |
9411 | ret = -EINVAL; | |
9412 | goto out; | |
9413 | } | |
9414 | ||
9415 | __set_bit(res[i].register_op, | |
9416 | ctx->restrictions.register_op); | |
9417 | break; | |
9418 | case IORING_RESTRICTION_SQE_OP: | |
9419 | if (res[i].sqe_op >= IORING_OP_LAST) { | |
9420 | ret = -EINVAL; | |
9421 | goto out; | |
9422 | } | |
9423 | ||
9424 | __set_bit(res[i].sqe_op, ctx->restrictions.sqe_op); | |
9425 | break; | |
9426 | case IORING_RESTRICTION_SQE_FLAGS_ALLOWED: | |
9427 | ctx->restrictions.sqe_flags_allowed = res[i].sqe_flags; | |
9428 | break; | |
9429 | case IORING_RESTRICTION_SQE_FLAGS_REQUIRED: | |
9430 | ctx->restrictions.sqe_flags_required = res[i].sqe_flags; | |
9431 | break; | |
9432 | default: | |
9433 | ret = -EINVAL; | |
9434 | goto out; | |
9435 | } | |
9436 | } | |
9437 | ||
9438 | out: | |
9439 | /* Reset all restrictions if an error happened */ | |
9440 | if (ret != 0) | |
9441 | memset(&ctx->restrictions, 0, sizeof(ctx->restrictions)); | |
9442 | else | |
7e84e1c7 | 9443 | ctx->restrictions.registered = true; |
21b55dbc SG |
9444 | |
9445 | kfree(res); | |
9446 | return ret; | |
9447 | } | |
9448 | ||
7e84e1c7 SG |
9449 | static int io_register_enable_rings(struct io_ring_ctx *ctx) |
9450 | { | |
9451 | if (!(ctx->flags & IORING_SETUP_R_DISABLED)) | |
9452 | return -EBADFD; | |
9453 | ||
9454 | if (ctx->restrictions.registered) | |
9455 | ctx->restricted = 1; | |
9456 | ||
9457 | ctx->flags &= ~IORING_SETUP_R_DISABLED; | |
9458 | ||
9459 | io_sq_offload_start(ctx); | |
9460 | ||
9461 | return 0; | |
9462 | } | |
9463 | ||
071698e1 JA |
9464 | static bool io_register_op_must_quiesce(int op) |
9465 | { | |
9466 | switch (op) { | |
9467 | case IORING_UNREGISTER_FILES: | |
9468 | case IORING_REGISTER_FILES_UPDATE: | |
9469 | case IORING_REGISTER_PROBE: | |
9470 | case IORING_REGISTER_PERSONALITY: | |
9471 | case IORING_UNREGISTER_PERSONALITY: | |
9472 | return false; | |
9473 | default: | |
9474 | return true; | |
9475 | } | |
9476 | } | |
9477 | ||
edafccee JA |
9478 | static int __io_uring_register(struct io_ring_ctx *ctx, unsigned opcode, |
9479 | void __user *arg, unsigned nr_args) | |
b19062a5 JA |
9480 | __releases(ctx->uring_lock) |
9481 | __acquires(ctx->uring_lock) | |
edafccee JA |
9482 | { |
9483 | int ret; | |
9484 | ||
35fa71a0 JA |
9485 | /* |
9486 | * We're inside the ring mutex, if the ref is already dying, then | |
9487 | * someone else killed the ctx or is already going through | |
9488 | * io_uring_register(). | |
9489 | */ | |
9490 | if (percpu_ref_is_dying(&ctx->refs)) | |
9491 | return -ENXIO; | |
9492 | ||
071698e1 | 9493 | if (io_register_op_must_quiesce(opcode)) { |
05f3fb3c | 9494 | percpu_ref_kill(&ctx->refs); |
b19062a5 | 9495 | |
05f3fb3c JA |
9496 | /* |
9497 | * Drop uring mutex before waiting for references to exit. If | |
9498 | * another thread is currently inside io_uring_enter() it might | |
9499 | * need to grab the uring_lock to make progress. If we hold it | |
9500 | * here across the drain wait, then we can deadlock. It's safe | |
9501 | * to drop the mutex here, since no new references will come in | |
9502 | * after we've killed the percpu ref. | |
9503 | */ | |
9504 | mutex_unlock(&ctx->uring_lock); | |
af9c1a44 JA |
9505 | do { |
9506 | ret = wait_for_completion_interruptible(&ctx->ref_comp); | |
9507 | if (!ret) | |
9508 | break; | |
ed6930c9 JA |
9509 | ret = io_run_task_work_sig(); |
9510 | if (ret < 0) | |
9511 | break; | |
af9c1a44 JA |
9512 | } while (1); |
9513 | ||
05f3fb3c | 9514 | mutex_lock(&ctx->uring_lock); |
af9c1a44 | 9515 | |
c150368b JA |
9516 | if (ret) { |
9517 | percpu_ref_resurrect(&ctx->refs); | |
21b55dbc SG |
9518 | goto out_quiesce; |
9519 | } | |
9520 | } | |
9521 | ||
9522 | if (ctx->restricted) { | |
9523 | if (opcode >= IORING_REGISTER_LAST) { | |
9524 | ret = -EINVAL; | |
9525 | goto out; | |
9526 | } | |
9527 | ||
9528 | if (!test_bit(opcode, ctx->restrictions.register_op)) { | |
9529 | ret = -EACCES; | |
c150368b JA |
9530 | goto out; |
9531 | } | |
05f3fb3c | 9532 | } |
edafccee JA |
9533 | |
9534 | switch (opcode) { | |
9535 | case IORING_REGISTER_BUFFERS: | |
9536 | ret = io_sqe_buffer_register(ctx, arg, nr_args); | |
9537 | break; | |
9538 | case IORING_UNREGISTER_BUFFERS: | |
9539 | ret = -EINVAL; | |
9540 | if (arg || nr_args) | |
9541 | break; | |
9542 | ret = io_sqe_buffer_unregister(ctx); | |
9543 | break; | |
6b06314c JA |
9544 | case IORING_REGISTER_FILES: |
9545 | ret = io_sqe_files_register(ctx, arg, nr_args); | |
9546 | break; | |
9547 | case IORING_UNREGISTER_FILES: | |
9548 | ret = -EINVAL; | |
9549 | if (arg || nr_args) | |
9550 | break; | |
9551 | ret = io_sqe_files_unregister(ctx); | |
9552 | break; | |
c3a31e60 JA |
9553 | case IORING_REGISTER_FILES_UPDATE: |
9554 | ret = io_sqe_files_update(ctx, arg, nr_args); | |
9555 | break; | |
9b402849 | 9556 | case IORING_REGISTER_EVENTFD: |
f2842ab5 | 9557 | case IORING_REGISTER_EVENTFD_ASYNC: |
9b402849 JA |
9558 | ret = -EINVAL; |
9559 | if (nr_args != 1) | |
9560 | break; | |
9561 | ret = io_eventfd_register(ctx, arg); | |
f2842ab5 JA |
9562 | if (ret) |
9563 | break; | |
9564 | if (opcode == IORING_REGISTER_EVENTFD_ASYNC) | |
9565 | ctx->eventfd_async = 1; | |
9566 | else | |
9567 | ctx->eventfd_async = 0; | |
9b402849 JA |
9568 | break; |
9569 | case IORING_UNREGISTER_EVENTFD: | |
9570 | ret = -EINVAL; | |
9571 | if (arg || nr_args) | |
9572 | break; | |
9573 | ret = io_eventfd_unregister(ctx); | |
9574 | break; | |
66f4af93 JA |
9575 | case IORING_REGISTER_PROBE: |
9576 | ret = -EINVAL; | |
9577 | if (!arg || nr_args > 256) | |
9578 | break; | |
9579 | ret = io_probe(ctx, arg, nr_args); | |
9580 | break; | |
071698e1 JA |
9581 | case IORING_REGISTER_PERSONALITY: |
9582 | ret = -EINVAL; | |
9583 | if (arg || nr_args) | |
9584 | break; | |
9585 | ret = io_register_personality(ctx); | |
9586 | break; | |
9587 | case IORING_UNREGISTER_PERSONALITY: | |
9588 | ret = -EINVAL; | |
9589 | if (arg) | |
9590 | break; | |
9591 | ret = io_unregister_personality(ctx, nr_args); | |
9592 | break; | |
7e84e1c7 SG |
9593 | case IORING_REGISTER_ENABLE_RINGS: |
9594 | ret = -EINVAL; | |
9595 | if (arg || nr_args) | |
9596 | break; | |
9597 | ret = io_register_enable_rings(ctx); | |
9598 | break; | |
21b55dbc SG |
9599 | case IORING_REGISTER_RESTRICTIONS: |
9600 | ret = io_register_restrictions(ctx, arg, nr_args); | |
9601 | break; | |
edafccee JA |
9602 | default: |
9603 | ret = -EINVAL; | |
9604 | break; | |
9605 | } | |
9606 | ||
21b55dbc | 9607 | out: |
071698e1 | 9608 | if (io_register_op_must_quiesce(opcode)) { |
05f3fb3c | 9609 | /* bring the ctx back to life */ |
05f3fb3c | 9610 | percpu_ref_reinit(&ctx->refs); |
21b55dbc | 9611 | out_quiesce: |
0f158b4c | 9612 | reinit_completion(&ctx->ref_comp); |
05f3fb3c | 9613 | } |
edafccee JA |
9614 | return ret; |
9615 | } | |
9616 | ||
9617 | SYSCALL_DEFINE4(io_uring_register, unsigned int, fd, unsigned int, opcode, | |
9618 | void __user *, arg, unsigned int, nr_args) | |
9619 | { | |
9620 | struct io_ring_ctx *ctx; | |
9621 | long ret = -EBADF; | |
9622 | struct fd f; | |
9623 | ||
9624 | f = fdget(fd); | |
9625 | if (!f.file) | |
9626 | return -EBADF; | |
9627 | ||
9628 | ret = -EOPNOTSUPP; | |
9629 | if (f.file->f_op != &io_uring_fops) | |
9630 | goto out_fput; | |
9631 | ||
9632 | ctx = f.file->private_data; | |
9633 | ||
9634 | mutex_lock(&ctx->uring_lock); | |
9635 | ret = __io_uring_register(ctx, opcode, arg, nr_args); | |
9636 | mutex_unlock(&ctx->uring_lock); | |
c826bd7a DD |
9637 | trace_io_uring_register(ctx, opcode, ctx->nr_user_files, ctx->nr_user_bufs, |
9638 | ctx->cq_ev_fd != NULL, ret); | |
edafccee JA |
9639 | out_fput: |
9640 | fdput(f); | |
9641 | return ret; | |
9642 | } | |
9643 | ||
2b188cc1 JA |
9644 | static int __init io_uring_init(void) |
9645 | { | |
d7f62e82 SM |
9646 | #define __BUILD_BUG_VERIFY_ELEMENT(stype, eoffset, etype, ename) do { \ |
9647 | BUILD_BUG_ON(offsetof(stype, ename) != eoffset); \ | |
9648 | BUILD_BUG_ON(sizeof(etype) != sizeof_field(stype, ename)); \ | |
9649 | } while (0) | |
9650 | ||
9651 | #define BUILD_BUG_SQE_ELEM(eoffset, etype, ename) \ | |
9652 | __BUILD_BUG_VERIFY_ELEMENT(struct io_uring_sqe, eoffset, etype, ename) | |
9653 | BUILD_BUG_ON(sizeof(struct io_uring_sqe) != 64); | |
9654 | BUILD_BUG_SQE_ELEM(0, __u8, opcode); | |
9655 | BUILD_BUG_SQE_ELEM(1, __u8, flags); | |
9656 | BUILD_BUG_SQE_ELEM(2, __u16, ioprio); | |
9657 | BUILD_BUG_SQE_ELEM(4, __s32, fd); | |
9658 | BUILD_BUG_SQE_ELEM(8, __u64, off); | |
9659 | BUILD_BUG_SQE_ELEM(8, __u64, addr2); | |
9660 | BUILD_BUG_SQE_ELEM(16, __u64, addr); | |
7d67af2c | 9661 | BUILD_BUG_SQE_ELEM(16, __u64, splice_off_in); |
d7f62e82 SM |
9662 | BUILD_BUG_SQE_ELEM(24, __u32, len); |
9663 | BUILD_BUG_SQE_ELEM(28, __kernel_rwf_t, rw_flags); | |
9664 | BUILD_BUG_SQE_ELEM(28, /* compat */ int, rw_flags); | |
9665 | BUILD_BUG_SQE_ELEM(28, /* compat */ __u32, rw_flags); | |
9666 | BUILD_BUG_SQE_ELEM(28, __u32, fsync_flags); | |
5769a351 JX |
9667 | BUILD_BUG_SQE_ELEM(28, /* compat */ __u16, poll_events); |
9668 | BUILD_BUG_SQE_ELEM(28, __u32, poll32_events); | |
d7f62e82 SM |
9669 | BUILD_BUG_SQE_ELEM(28, __u32, sync_range_flags); |
9670 | BUILD_BUG_SQE_ELEM(28, __u32, msg_flags); | |
9671 | BUILD_BUG_SQE_ELEM(28, __u32, timeout_flags); | |
9672 | BUILD_BUG_SQE_ELEM(28, __u32, accept_flags); | |
9673 | BUILD_BUG_SQE_ELEM(28, __u32, cancel_flags); | |
9674 | BUILD_BUG_SQE_ELEM(28, __u32, open_flags); | |
9675 | BUILD_BUG_SQE_ELEM(28, __u32, statx_flags); | |
9676 | BUILD_BUG_SQE_ELEM(28, __u32, fadvise_advice); | |
7d67af2c | 9677 | BUILD_BUG_SQE_ELEM(28, __u32, splice_flags); |
d7f62e82 SM |
9678 | BUILD_BUG_SQE_ELEM(32, __u64, user_data); |
9679 | BUILD_BUG_SQE_ELEM(40, __u16, buf_index); | |
9680 | BUILD_BUG_SQE_ELEM(42, __u16, personality); | |
7d67af2c | 9681 | BUILD_BUG_SQE_ELEM(44, __s32, splice_fd_in); |
d7f62e82 | 9682 | |
d3656344 | 9683 | BUILD_BUG_ON(ARRAY_SIZE(io_op_defs) != IORING_OP_LAST); |
84557871 | 9684 | BUILD_BUG_ON(__REQ_F_LAST_BIT >= 8 * sizeof(int)); |
2b188cc1 JA |
9685 | req_cachep = KMEM_CACHE(io_kiocb, SLAB_HWCACHE_ALIGN | SLAB_PANIC); |
9686 | return 0; | |
9687 | }; | |
9688 | __initcall(io_uring_init); |