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; |
206aefde | 280 | atomic_t 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, |
84557871 JA |
588 | |
589 | /* not a real bit, just to check we're not overflowing the space */ | |
590 | __REQ_F_LAST_BIT, | |
6b47ee6e PB |
591 | }; |
592 | ||
593 | enum { | |
594 | /* ctx owns file */ | |
595 | REQ_F_FIXED_FILE = BIT(REQ_F_FIXED_FILE_BIT), | |
596 | /* drain existing IO first */ | |
597 | REQ_F_IO_DRAIN = BIT(REQ_F_IO_DRAIN_BIT), | |
598 | /* linked sqes */ | |
599 | REQ_F_LINK = BIT(REQ_F_LINK_BIT), | |
600 | /* doesn't sever on completion < 0 */ | |
601 | REQ_F_HARDLINK = BIT(REQ_F_HARDLINK_BIT), | |
602 | /* IOSQE_ASYNC */ | |
603 | REQ_F_FORCE_ASYNC = BIT(REQ_F_FORCE_ASYNC_BIT), | |
bcda7baa JA |
604 | /* IOSQE_BUFFER_SELECT */ |
605 | REQ_F_BUFFER_SELECT = BIT(REQ_F_BUFFER_SELECT_BIT), | |
6b47ee6e | 606 | |
dea3b49c PB |
607 | /* head of a link */ |
608 | REQ_F_LINK_HEAD = BIT(REQ_F_LINK_HEAD_BIT), | |
6b47ee6e PB |
609 | /* fail rest of links */ |
610 | REQ_F_FAIL_LINK = BIT(REQ_F_FAIL_LINK_BIT), | |
611 | /* on inflight list */ | |
612 | REQ_F_INFLIGHT = BIT(REQ_F_INFLIGHT_BIT), | |
613 | /* read/write uses file position */ | |
614 | REQ_F_CUR_POS = BIT(REQ_F_CUR_POS_BIT), | |
615 | /* must not punt to workers */ | |
616 | REQ_F_NOWAIT = BIT(REQ_F_NOWAIT_BIT), | |
6b47ee6e PB |
617 | /* has linked timeout */ |
618 | REQ_F_LINK_TIMEOUT = BIT(REQ_F_LINK_TIMEOUT_BIT), | |
6b47ee6e PB |
619 | /* regular file */ |
620 | REQ_F_ISREG = BIT(REQ_F_ISREG_BIT), | |
99bc4c38 PB |
621 | /* needs cleanup */ |
622 | REQ_F_NEED_CLEANUP = BIT(REQ_F_NEED_CLEANUP_BIT), | |
d7718a9d JA |
623 | /* already went through poll handler */ |
624 | REQ_F_POLLED = BIT(REQ_F_POLLED_BIT), | |
bcda7baa JA |
625 | /* buffer already selected */ |
626 | REQ_F_BUFFER_SELECTED = BIT(REQ_F_BUFFER_SELECTED_BIT), | |
5b0bbee4 JA |
627 | /* doesn't need file table for this request */ |
628 | REQ_F_NO_FILE_TABLE = BIT(REQ_F_NO_FILE_TABLE_BIT), | |
7cdaf587 XW |
629 | /* io_wq_work is initialized */ |
630 | REQ_F_WORK_INITIALIZED = BIT(REQ_F_WORK_INITIALIZED_BIT), | |
d7718a9d JA |
631 | }; |
632 | ||
633 | struct async_poll { | |
634 | struct io_poll_iocb poll; | |
807abcb0 | 635 | struct io_poll_iocb *double_poll; |
6b47ee6e PB |
636 | }; |
637 | ||
09bb8394 JA |
638 | /* |
639 | * NOTE! Each of the iocb union members has the file pointer | |
640 | * as the first entry in their struct definition. So you can | |
641 | * access the file pointer through any of the sub-structs, | |
642 | * or directly as just 'ki_filp' in this struct. | |
643 | */ | |
2b188cc1 | 644 | struct io_kiocb { |
221c5eb2 | 645 | union { |
09bb8394 | 646 | struct file *file; |
9adbd45d | 647 | struct io_rw rw; |
221c5eb2 | 648 | struct io_poll_iocb poll; |
8ed8d3c3 JA |
649 | struct io_accept accept; |
650 | struct io_sync sync; | |
fbf23849 | 651 | struct io_cancel cancel; |
b29472ee | 652 | struct io_timeout timeout; |
0bdf7a2d | 653 | struct io_timeout_rem timeout_rem; |
3fbb51c1 | 654 | struct io_connect connect; |
e47293fd | 655 | struct io_sr_msg sr_msg; |
15b71abe | 656 | struct io_open open; |
b5dba59e | 657 | struct io_close close; |
05f3fb3c | 658 | struct io_files_update files_update; |
4840e418 | 659 | struct io_fadvise fadvise; |
c1ca757b | 660 | struct io_madvise madvise; |
3e4827b0 | 661 | struct io_epoll epoll; |
7d67af2c | 662 | struct io_splice splice; |
ddf0322d | 663 | struct io_provide_buf pbuf; |
1d9e1288 | 664 | struct io_statx statx; |
3ca405eb PB |
665 | /* use only after cleaning per-op data, see io_clean_op() */ |
666 | struct io_completion compl; | |
221c5eb2 | 667 | }; |
2b188cc1 | 668 | |
e8c2bc1f JA |
669 | /* opcode allocated if it needs to store data for async defer */ |
670 | void *async_data; | |
d625c6ee | 671 | u8 opcode; |
65a6543d XW |
672 | /* polled IO has completed */ |
673 | u8 iopoll_completed; | |
2b188cc1 | 674 | |
4f4eeba8 | 675 | u16 buf_index; |
9cf7c104 | 676 | u32 result; |
4f4eeba8 | 677 | |
010e8e6b PB |
678 | struct io_ring_ctx *ctx; |
679 | unsigned int flags; | |
680 | refcount_t refs; | |
681 | struct task_struct *task; | |
682 | u64 user_data; | |
d7718a9d | 683 | |
010e8e6b | 684 | struct list_head link_list; |
fcb323cc | 685 | |
d21ffe7e PB |
686 | /* |
687 | * 1. used with ctx->iopoll_list with reads/writes | |
688 | * 2. to track reqs with ->files (see io_op_def::file_table) | |
689 | */ | |
010e8e6b PB |
690 | struct list_head inflight_entry; |
691 | ||
692 | struct percpu_ref *fixed_file_refs; | |
693 | struct callback_head task_work; | |
694 | /* for polled requests, i.e. IORING_OP_POLL_ADD and async armed poll */ | |
695 | struct hlist_node hash_node; | |
696 | struct async_poll *apoll; | |
697 | struct io_wq_work work; | |
2b188cc1 | 698 | }; |
05589553 | 699 | |
27dc8338 PB |
700 | struct io_defer_entry { |
701 | struct list_head list; | |
702 | struct io_kiocb *req; | |
9cf7c104 | 703 | u32 seq; |
2b188cc1 JA |
704 | }; |
705 | ||
def596e9 | 706 | #define IO_IOPOLL_BATCH 8 |
2b188cc1 | 707 | |
013538bd JA |
708 | struct io_comp_state { |
709 | unsigned int nr; | |
710 | struct list_head list; | |
711 | struct io_ring_ctx *ctx; | |
712 | }; | |
713 | ||
9a56a232 JA |
714 | struct io_submit_state { |
715 | struct blk_plug plug; | |
716 | ||
2579f913 JA |
717 | /* |
718 | * io_kiocb alloc cache | |
719 | */ | |
720 | void *reqs[IO_IOPOLL_BATCH]; | |
6c8a3134 | 721 | unsigned int free_reqs; |
2579f913 | 722 | |
013538bd JA |
723 | /* |
724 | * Batch completion logic | |
725 | */ | |
726 | struct io_comp_state comp; | |
727 | ||
9a56a232 JA |
728 | /* |
729 | * File reference cache | |
730 | */ | |
731 | struct file *file; | |
732 | unsigned int fd; | |
733 | unsigned int has_refs; | |
9a56a232 JA |
734 | unsigned int ios_left; |
735 | }; | |
736 | ||
d3656344 | 737 | struct io_op_def { |
d3656344 JA |
738 | /* needs req->file assigned */ |
739 | unsigned needs_file : 1; | |
fd2206e4 JA |
740 | /* don't fail if file grab fails */ |
741 | unsigned needs_file_no_error : 1; | |
d3656344 JA |
742 | /* hash wq insertion if file is a regular file */ |
743 | unsigned hash_reg_file : 1; | |
744 | /* unbound wq insertion if file is a non-regular file */ | |
745 | unsigned unbound_nonreg_file : 1; | |
66f4af93 JA |
746 | /* opcode is not supported by this kernel */ |
747 | unsigned not_supported : 1; | |
8a72758c JA |
748 | /* set if opcode supports polled "wait" */ |
749 | unsigned pollin : 1; | |
750 | unsigned pollout : 1; | |
bcda7baa JA |
751 | /* op supports buffer selection */ |
752 | unsigned buffer_select : 1; | |
e8c2bc1f | 753 | /* needs rlimit(RLIMIT_FSIZE) assigned */ |
57f1a649 | 754 | unsigned needs_fsize : 1; |
e8c2bc1f JA |
755 | /* must always have async data allocated */ |
756 | unsigned needs_async_data : 1; | |
757 | /* size of async data needed, if any */ | |
758 | unsigned short async_size; | |
0f203765 | 759 | unsigned work_flags; |
d3656344 JA |
760 | }; |
761 | ||
0918682b | 762 | static const struct io_op_def io_op_defs[] = { |
0463b6c5 PB |
763 | [IORING_OP_NOP] = {}, |
764 | [IORING_OP_READV] = { | |
d3656344 JA |
765 | .needs_file = 1, |
766 | .unbound_nonreg_file = 1, | |
8a72758c | 767 | .pollin = 1, |
4d954c25 | 768 | .buffer_select = 1, |
e8c2bc1f JA |
769 | .needs_async_data = 1, |
770 | .async_size = sizeof(struct io_async_rw), | |
0f203765 | 771 | .work_flags = IO_WQ_WORK_MM | IO_WQ_WORK_BLKCG, |
d3656344 | 772 | }, |
0463b6c5 | 773 | [IORING_OP_WRITEV] = { |
d3656344 JA |
774 | .needs_file = 1, |
775 | .hash_reg_file = 1, | |
776 | .unbound_nonreg_file = 1, | |
8a72758c | 777 | .pollout = 1, |
57f1a649 | 778 | .needs_fsize = 1, |
e8c2bc1f JA |
779 | .needs_async_data = 1, |
780 | .async_size = sizeof(struct io_async_rw), | |
0f203765 | 781 | .work_flags = IO_WQ_WORK_MM | IO_WQ_WORK_BLKCG, |
d3656344 | 782 | }, |
0463b6c5 | 783 | [IORING_OP_FSYNC] = { |
d3656344 | 784 | .needs_file = 1, |
0f203765 | 785 | .work_flags = IO_WQ_WORK_BLKCG, |
d3656344 | 786 | }, |
0463b6c5 | 787 | [IORING_OP_READ_FIXED] = { |
d3656344 JA |
788 | .needs_file = 1, |
789 | .unbound_nonreg_file = 1, | |
8a72758c | 790 | .pollin = 1, |
e8c2bc1f | 791 | .async_size = sizeof(struct io_async_rw), |
0f203765 | 792 | .work_flags = IO_WQ_WORK_BLKCG, |
d3656344 | 793 | }, |
0463b6c5 | 794 | [IORING_OP_WRITE_FIXED] = { |
d3656344 JA |
795 | .needs_file = 1, |
796 | .hash_reg_file = 1, | |
797 | .unbound_nonreg_file = 1, | |
8a72758c | 798 | .pollout = 1, |
57f1a649 | 799 | .needs_fsize = 1, |
e8c2bc1f | 800 | .async_size = sizeof(struct io_async_rw), |
0f203765 | 801 | .work_flags = IO_WQ_WORK_BLKCG, |
d3656344 | 802 | }, |
0463b6c5 | 803 | [IORING_OP_POLL_ADD] = { |
d3656344 JA |
804 | .needs_file = 1, |
805 | .unbound_nonreg_file = 1, | |
806 | }, | |
0463b6c5 PB |
807 | [IORING_OP_POLL_REMOVE] = {}, |
808 | [IORING_OP_SYNC_FILE_RANGE] = { | |
d3656344 | 809 | .needs_file = 1, |
0f203765 | 810 | .work_flags = IO_WQ_WORK_BLKCG, |
d3656344 | 811 | }, |
0463b6c5 | 812 | [IORING_OP_SENDMSG] = { |
d3656344 JA |
813 | .needs_file = 1, |
814 | .unbound_nonreg_file = 1, | |
8a72758c | 815 | .pollout = 1, |
e8c2bc1f JA |
816 | .needs_async_data = 1, |
817 | .async_size = sizeof(struct io_async_msghdr), | |
0f203765 JA |
818 | .work_flags = IO_WQ_WORK_MM | IO_WQ_WORK_BLKCG | |
819 | IO_WQ_WORK_FS, | |
d3656344 | 820 | }, |
0463b6c5 | 821 | [IORING_OP_RECVMSG] = { |
d3656344 JA |
822 | .needs_file = 1, |
823 | .unbound_nonreg_file = 1, | |
8a72758c | 824 | .pollin = 1, |
52de1fe1 | 825 | .buffer_select = 1, |
e8c2bc1f JA |
826 | .needs_async_data = 1, |
827 | .async_size = sizeof(struct io_async_msghdr), | |
0f203765 JA |
828 | .work_flags = IO_WQ_WORK_MM | IO_WQ_WORK_BLKCG | |
829 | IO_WQ_WORK_FS, | |
d3656344 | 830 | }, |
0463b6c5 | 831 | [IORING_OP_TIMEOUT] = { |
e8c2bc1f JA |
832 | .needs_async_data = 1, |
833 | .async_size = sizeof(struct io_timeout_data), | |
0f203765 | 834 | .work_flags = IO_WQ_WORK_MM, |
d3656344 | 835 | }, |
0463b6c5 PB |
836 | [IORING_OP_TIMEOUT_REMOVE] = {}, |
837 | [IORING_OP_ACCEPT] = { | |
d3656344 JA |
838 | .needs_file = 1, |
839 | .unbound_nonreg_file = 1, | |
8a72758c | 840 | .pollin = 1, |
0f203765 | 841 | .work_flags = IO_WQ_WORK_MM | IO_WQ_WORK_FILES, |
d3656344 | 842 | }, |
0463b6c5 PB |
843 | [IORING_OP_ASYNC_CANCEL] = {}, |
844 | [IORING_OP_LINK_TIMEOUT] = { | |
e8c2bc1f JA |
845 | .needs_async_data = 1, |
846 | .async_size = sizeof(struct io_timeout_data), | |
0f203765 | 847 | .work_flags = IO_WQ_WORK_MM, |
d3656344 | 848 | }, |
0463b6c5 | 849 | [IORING_OP_CONNECT] = { |
d3656344 JA |
850 | .needs_file = 1, |
851 | .unbound_nonreg_file = 1, | |
8a72758c | 852 | .pollout = 1, |
e8c2bc1f JA |
853 | .needs_async_data = 1, |
854 | .async_size = sizeof(struct io_async_connect), | |
0f203765 | 855 | .work_flags = IO_WQ_WORK_MM, |
d3656344 | 856 | }, |
0463b6c5 | 857 | [IORING_OP_FALLOCATE] = { |
d3656344 | 858 | .needs_file = 1, |
57f1a649 | 859 | .needs_fsize = 1, |
0f203765 | 860 | .work_flags = IO_WQ_WORK_BLKCG, |
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, |
57f1a649 | 890 | .needs_fsize = 1, |
e8c2bc1f | 891 | .async_size = sizeof(struct io_async_rw), |
0f203765 | 892 | .work_flags = IO_WQ_WORK_MM | IO_WQ_WORK_BLKCG, |
3a6820f2 | 893 | }, |
0463b6c5 | 894 | [IORING_OP_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 | ||
7cdaf587 XW |
1073 | /* |
1074 | * Note: must call io_req_init_async() for the first time you | |
1075 | * touch any members of io_wq_work. | |
1076 | */ | |
1077 | static inline void io_req_init_async(struct io_kiocb *req) | |
1078 | { | |
500a373d JA |
1079 | struct io_uring_task *tctx = current->io_uring; |
1080 | ||
7cdaf587 XW |
1081 | if (req->flags & REQ_F_WORK_INITIALIZED) |
1082 | return; | |
1083 | ||
1084 | memset(&req->work, 0, sizeof(req->work)); | |
1085 | req->flags |= REQ_F_WORK_INITIALIZED; | |
500a373d JA |
1086 | |
1087 | /* Grab a ref if this isn't our static identity */ | |
1088 | req->work.identity = tctx->identity; | |
1089 | if (tctx->identity != &tctx->__identity) | |
1090 | refcount_inc(&req->work.identity->count); | |
7cdaf587 XW |
1091 | } |
1092 | ||
0cdaf760 PB |
1093 | static inline bool io_async_submit(struct io_ring_ctx *ctx) |
1094 | { | |
1095 | return ctx->flags & IORING_SETUP_SQPOLL; | |
1096 | } | |
1097 | ||
2b188cc1 JA |
1098 | static void io_ring_ctx_ref_free(struct percpu_ref *ref) |
1099 | { | |
1100 | struct io_ring_ctx *ctx = container_of(ref, struct io_ring_ctx, refs); | |
1101 | ||
0f158b4c | 1102 | complete(&ctx->ref_comp); |
2b188cc1 JA |
1103 | } |
1104 | ||
8eb7e2d0 PB |
1105 | static inline bool io_is_timeout_noseq(struct io_kiocb *req) |
1106 | { | |
1107 | return !req->timeout.off; | |
1108 | } | |
1109 | ||
2b188cc1 JA |
1110 | static struct io_ring_ctx *io_ring_ctx_alloc(struct io_uring_params *p) |
1111 | { | |
1112 | struct io_ring_ctx *ctx; | |
78076bb6 | 1113 | int hash_bits; |
2b188cc1 JA |
1114 | |
1115 | ctx = kzalloc(sizeof(*ctx), GFP_KERNEL); | |
1116 | if (!ctx) | |
1117 | return NULL; | |
1118 | ||
0ddf92e8 JA |
1119 | ctx->fallback_req = kmem_cache_alloc(req_cachep, GFP_KERNEL); |
1120 | if (!ctx->fallback_req) | |
1121 | goto err; | |
1122 | ||
78076bb6 JA |
1123 | /* |
1124 | * Use 5 bits less than the max cq entries, that should give us around | |
1125 | * 32 entries per hash list if totally full and uniformly spread. | |
1126 | */ | |
1127 | hash_bits = ilog2(p->cq_entries); | |
1128 | hash_bits -= 5; | |
1129 | if (hash_bits <= 0) | |
1130 | hash_bits = 1; | |
1131 | ctx->cancel_hash_bits = hash_bits; | |
1132 | ctx->cancel_hash = kmalloc((1U << hash_bits) * sizeof(struct hlist_head), | |
1133 | GFP_KERNEL); | |
1134 | if (!ctx->cancel_hash) | |
1135 | goto err; | |
1136 | __hash_init(ctx->cancel_hash, 1U << hash_bits); | |
1137 | ||
21482896 | 1138 | if (percpu_ref_init(&ctx->refs, io_ring_ctx_ref_free, |
206aefde JA |
1139 | PERCPU_REF_ALLOW_REINIT, GFP_KERNEL)) |
1140 | goto err; | |
2b188cc1 JA |
1141 | |
1142 | ctx->flags = p->flags; | |
90554200 | 1143 | init_waitqueue_head(&ctx->sqo_sq_wait); |
69fb2131 | 1144 | INIT_LIST_HEAD(&ctx->sqd_list); |
2b188cc1 | 1145 | init_waitqueue_head(&ctx->cq_wait); |
1d7bb1d5 | 1146 | INIT_LIST_HEAD(&ctx->cq_overflow_list); |
0f158b4c JA |
1147 | init_completion(&ctx->ref_comp); |
1148 | init_completion(&ctx->sq_thread_comp); | |
5a2e745d | 1149 | idr_init(&ctx->io_buffer_idr); |
071698e1 | 1150 | idr_init(&ctx->personality_idr); |
2b188cc1 JA |
1151 | mutex_init(&ctx->uring_lock); |
1152 | init_waitqueue_head(&ctx->wait); | |
1153 | spin_lock_init(&ctx->completion_lock); | |
540e32a0 | 1154 | INIT_LIST_HEAD(&ctx->iopoll_list); |
de0617e4 | 1155 | INIT_LIST_HEAD(&ctx->defer_list); |
5262f567 | 1156 | INIT_LIST_HEAD(&ctx->timeout_list); |
fcb323cc JA |
1157 | init_waitqueue_head(&ctx->inflight_wait); |
1158 | spin_lock_init(&ctx->inflight_lock); | |
1159 | INIT_LIST_HEAD(&ctx->inflight_list); | |
4a38aed2 JA |
1160 | INIT_DELAYED_WORK(&ctx->file_put_work, io_file_put_work); |
1161 | init_llist_head(&ctx->file_put_llist); | |
2b188cc1 | 1162 | return ctx; |
206aefde | 1163 | err: |
0ddf92e8 JA |
1164 | if (ctx->fallback_req) |
1165 | kmem_cache_free(req_cachep, ctx->fallback_req); | |
78076bb6 | 1166 | kfree(ctx->cancel_hash); |
206aefde JA |
1167 | kfree(ctx); |
1168 | return NULL; | |
2b188cc1 JA |
1169 | } |
1170 | ||
9cf7c104 | 1171 | static bool req_need_defer(struct io_kiocb *req, u32 seq) |
7adf4eaf | 1172 | { |
2bc9930e JA |
1173 | if (unlikely(req->flags & REQ_F_IO_DRAIN)) { |
1174 | struct io_ring_ctx *ctx = req->ctx; | |
a197f664 | 1175 | |
9cf7c104 | 1176 | return seq != ctx->cached_cq_tail |
31af27c7 | 1177 | + atomic_read(&ctx->cached_cq_overflow); |
2bc9930e | 1178 | } |
de0617e4 | 1179 | |
9d858b21 | 1180 | return false; |
de0617e4 JA |
1181 | } |
1182 | ||
de0617e4 | 1183 | static void __io_commit_cqring(struct io_ring_ctx *ctx) |
2b188cc1 | 1184 | { |
75b28aff | 1185 | struct io_rings *rings = ctx->rings; |
2b188cc1 | 1186 | |
07910158 PB |
1187 | /* order cqe stores with ring update */ |
1188 | smp_store_release(&rings->cq.tail, ctx->cached_cq_tail); | |
2b188cc1 | 1189 | |
07910158 PB |
1190 | if (wq_has_sleeper(&ctx->cq_wait)) { |
1191 | wake_up_interruptible(&ctx->cq_wait); | |
1192 | kill_fasync(&ctx->cq_fasync, SIGIO, POLL_IN); | |
2b188cc1 JA |
1193 | } |
1194 | } | |
1195 | ||
5c3462cf | 1196 | static void io_put_identity(struct io_uring_task *tctx, struct io_kiocb *req) |
1e6fa521 | 1197 | { |
500a373d | 1198 | if (req->work.identity == &tctx->__identity) |
1e6fa521 JA |
1199 | return; |
1200 | if (refcount_dec_and_test(&req->work.identity->count)) | |
1201 | kfree(req->work.identity); | |
1202 | } | |
1203 | ||
4edf20f9 | 1204 | static void io_req_clean_work(struct io_kiocb *req) |
18d9be1a | 1205 | { |
7cdaf587 | 1206 | if (!(req->flags & REQ_F_WORK_INITIALIZED)) |
4edf20f9 | 1207 | return; |
51a4cc11 JA |
1208 | |
1209 | req->flags &= ~REQ_F_WORK_INITIALIZED; | |
7cdaf587 | 1210 | |
dfead8a8 | 1211 | if (req->work.flags & IO_WQ_WORK_MM) { |
98447d65 | 1212 | mmdrop(req->work.identity->mm); |
dfead8a8 | 1213 | req->work.flags &= ~IO_WQ_WORK_MM; |
cccf0ee8 | 1214 | } |
91d8f519 | 1215 | #ifdef CONFIG_BLK_CGROUP |
dfead8a8 | 1216 | if (req->work.flags & IO_WQ_WORK_BLKCG) { |
98447d65 | 1217 | css_put(req->work.identity->blkcg_css); |
dfead8a8 JA |
1218 | req->work.flags &= ~IO_WQ_WORK_BLKCG; |
1219 | } | |
91d8f519 | 1220 | #endif |
dfead8a8 | 1221 | if (req->work.flags & IO_WQ_WORK_CREDS) { |
98447d65 | 1222 | put_cred(req->work.identity->creds); |
dfead8a8 | 1223 | req->work.flags &= ~IO_WQ_WORK_CREDS; |
cccf0ee8 | 1224 | } |
dfead8a8 | 1225 | if (req->work.flags & IO_WQ_WORK_FS) { |
98447d65 | 1226 | struct fs_struct *fs = req->work.identity->fs; |
ff002b30 | 1227 | |
98447d65 | 1228 | spin_lock(&req->work.identity->fs->lock); |
ff002b30 JA |
1229 | if (--fs->users) |
1230 | fs = NULL; | |
98447d65 | 1231 | spin_unlock(&req->work.identity->fs->lock); |
ff002b30 JA |
1232 | if (fs) |
1233 | free_fs_struct(fs); | |
dfead8a8 | 1234 | req->work.flags &= ~IO_WQ_WORK_FS; |
ff002b30 | 1235 | } |
1e6fa521 | 1236 | |
5c3462cf | 1237 | io_put_identity(req->task->io_uring, req); |
561fb04a JA |
1238 | } |
1239 | ||
1e6fa521 JA |
1240 | /* |
1241 | * Create a private copy of io_identity, since some fields don't match | |
1242 | * the current context. | |
1243 | */ | |
1244 | static bool io_identity_cow(struct io_kiocb *req) | |
1245 | { | |
5c3462cf | 1246 | struct io_uring_task *tctx = current->io_uring; |
1e6fa521 JA |
1247 | const struct cred *creds = NULL; |
1248 | struct io_identity *id; | |
1249 | ||
1250 | if (req->work.flags & IO_WQ_WORK_CREDS) | |
1251 | creds = req->work.identity->creds; | |
1252 | ||
1253 | id = kmemdup(req->work.identity, sizeof(*id), GFP_KERNEL); | |
1254 | if (unlikely(!id)) { | |
1255 | req->work.flags |= IO_WQ_WORK_CANCEL; | |
1256 | return false; | |
1257 | } | |
1258 | ||
1259 | /* | |
1260 | * We can safely just re-init the creds we copied Either the field | |
1261 | * matches the current one, or we haven't grabbed it yet. The only | |
1262 | * exception is ->creds, through registered personalities, so handle | |
1263 | * that one separately. | |
1264 | */ | |
1265 | io_init_identity(id); | |
1266 | if (creds) | |
1267 | req->work.identity->creds = creds; | |
1268 | ||
1269 | /* add one for this request */ | |
1270 | refcount_inc(&id->count); | |
1271 | ||
1272 | /* drop old identity, assign new one. one ref for req, one for tctx */ | |
500a373d | 1273 | if (req->work.identity != tctx->identity && |
1e6fa521 JA |
1274 | refcount_sub_and_test(2, &req->work.identity->count)) |
1275 | kfree(req->work.identity); | |
1276 | ||
1277 | req->work.identity = id; | |
500a373d | 1278 | tctx->identity = id; |
1e6fa521 JA |
1279 | return true; |
1280 | } | |
1281 | ||
1282 | static bool io_grab_identity(struct io_kiocb *req) | |
18d9be1a | 1283 | { |
d3656344 | 1284 | const struct io_op_def *def = &io_op_defs[req->opcode]; |
5c3462cf | 1285 | struct io_identity *id = req->work.identity; |
23329513 | 1286 | struct io_ring_ctx *ctx = req->ctx; |
54a91f3b | 1287 | |
1e6fa521 JA |
1288 | if (def->needs_fsize && id->fsize != rlimit(RLIMIT_FSIZE)) |
1289 | return false; | |
16d59803 | 1290 | |
dfead8a8 | 1291 | if (!(req->work.flags & IO_WQ_WORK_FILES) && |
1e6fa521 | 1292 | (def->work_flags & IO_WQ_WORK_FILES) && |
23329513 | 1293 | !(req->flags & REQ_F_NO_FILE_TABLE)) { |
1e6fa521 JA |
1294 | if (id->files != current->files || |
1295 | id->nsproxy != current->nsproxy) | |
1296 | return false; | |
1297 | atomic_inc(&id->files->count); | |
1298 | get_nsproxy(id->nsproxy); | |
23329513 PB |
1299 | req->flags |= REQ_F_INFLIGHT; |
1300 | ||
1301 | spin_lock_irq(&ctx->inflight_lock); | |
1302 | list_add(&req->inflight_entry, &ctx->inflight_list); | |
1303 | spin_unlock_irq(&ctx->inflight_lock); | |
dfead8a8 | 1304 | req->work.flags |= IO_WQ_WORK_FILES; |
23329513 | 1305 | } |
91d8f519 | 1306 | #ifdef CONFIG_BLK_CGROUP |
dfead8a8 JA |
1307 | if (!(req->work.flags & IO_WQ_WORK_BLKCG) && |
1308 | (def->work_flags & IO_WQ_WORK_BLKCG)) { | |
91d8f519 | 1309 | rcu_read_lock(); |
1e6fa521 JA |
1310 | if (id->blkcg_css != blkcg_css()) { |
1311 | rcu_read_unlock(); | |
1312 | return false; | |
1313 | } | |
91d8f519 DZ |
1314 | /* |
1315 | * This should be rare, either the cgroup is dying or the task | |
1316 | * is moving cgroups. Just punt to root for the handful of ios. | |
1317 | */ | |
1e6fa521 | 1318 | if (css_tryget_online(id->blkcg_css)) |
dfead8a8 | 1319 | req->work.flags |= IO_WQ_WORK_BLKCG; |
91d8f519 DZ |
1320 | rcu_read_unlock(); |
1321 | } | |
1322 | #endif | |
dfead8a8 | 1323 | if (!(req->work.flags & IO_WQ_WORK_CREDS)) { |
1e6fa521 JA |
1324 | if (id->creds != current_cred()) |
1325 | return false; | |
1326 | get_cred(id->creds); | |
dfead8a8 JA |
1327 | req->work.flags |= IO_WQ_WORK_CREDS; |
1328 | } | |
4ea33a97 JA |
1329 | #ifdef CONFIG_AUDIT |
1330 | if (!uid_eq(current->loginuid, id->loginuid) || | |
1331 | current->sessionid != id->sessionid) | |
1332 | return false; | |
1333 | #endif | |
dfead8a8 JA |
1334 | if (!(req->work.flags & IO_WQ_WORK_FS) && |
1335 | (def->work_flags & IO_WQ_WORK_FS)) { | |
1e6fa521 JA |
1336 | if (current->fs != id->fs) |
1337 | return false; | |
1338 | spin_lock(&id->fs->lock); | |
1339 | if (!id->fs->in_exec) { | |
1340 | id->fs->users++; | |
dfead8a8 | 1341 | req->work.flags |= IO_WQ_WORK_FS; |
dca9cf8b PB |
1342 | } else { |
1343 | req->work.flags |= IO_WQ_WORK_CANCEL; | |
1344 | } | |
1345 | spin_unlock(¤t->fs->lock); | |
1346 | } | |
1e6fa521 JA |
1347 | |
1348 | return true; | |
1349 | } | |
1350 | ||
1351 | static void io_prep_async_work(struct io_kiocb *req) | |
1352 | { | |
1353 | const struct io_op_def *def = &io_op_defs[req->opcode]; | |
1e6fa521 | 1354 | struct io_ring_ctx *ctx = req->ctx; |
5c3462cf | 1355 | struct io_identity *id; |
1e6fa521 JA |
1356 | |
1357 | io_req_init_async(req); | |
5c3462cf | 1358 | id = req->work.identity; |
1e6fa521 JA |
1359 | |
1360 | if (req->flags & REQ_F_ISREG) { | |
1361 | if (def->hash_reg_file || (ctx->flags & IORING_SETUP_IOPOLL)) | |
1362 | io_wq_hash_work(&req->work, file_inode(req->file)); | |
1363 | } else { | |
1364 | if (def->unbound_nonreg_file) | |
1365 | req->work.flags |= IO_WQ_WORK_UNBOUND; | |
1366 | } | |
1367 | ||
1368 | /* ->mm can never change on us */ | |
1369 | if (!(req->work.flags & IO_WQ_WORK_MM) && | |
1370 | (def->work_flags & IO_WQ_WORK_MM)) { | |
1371 | mmgrab(id->mm); | |
1372 | req->work.flags |= IO_WQ_WORK_MM; | |
1373 | } | |
1374 | ||
1375 | /* if we fail grabbing identity, we must COW, regrab, and retry */ | |
1376 | if (io_grab_identity(req)) | |
1377 | return; | |
1378 | ||
1379 | if (!io_identity_cow(req)) | |
1380 | return; | |
1381 | ||
1382 | /* can't fail at this point */ | |
1383 | if (!io_grab_identity(req)) | |
1384 | WARN_ON(1); | |
561fb04a | 1385 | } |
cccf0ee8 | 1386 | |
cbdcb435 | 1387 | static void io_prep_async_link(struct io_kiocb *req) |
561fb04a | 1388 | { |
cbdcb435 | 1389 | struct io_kiocb *cur; |
54a91f3b | 1390 | |
cbdcb435 PB |
1391 | io_prep_async_work(req); |
1392 | if (req->flags & REQ_F_LINK_HEAD) | |
1393 | list_for_each_entry(cur, &req->link_list, link_list) | |
1394 | io_prep_async_work(cur); | |
561fb04a JA |
1395 | } |
1396 | ||
7271ef3a | 1397 | static struct io_kiocb *__io_queue_async_work(struct io_kiocb *req) |
561fb04a | 1398 | { |
a197f664 | 1399 | struct io_ring_ctx *ctx = req->ctx; |
cbdcb435 | 1400 | struct io_kiocb *link = io_prep_linked_timeout(req); |
561fb04a | 1401 | |
8766dd51 PB |
1402 | trace_io_uring_queue_async_work(ctx, io_wq_is_hashed(&req->work), req, |
1403 | &req->work, req->flags); | |
1404 | io_wq_enqueue(ctx->io_wq, &req->work); | |
7271ef3a | 1405 | return link; |
18d9be1a JA |
1406 | } |
1407 | ||
cbdcb435 PB |
1408 | static void io_queue_async_work(struct io_kiocb *req) |
1409 | { | |
7271ef3a JA |
1410 | struct io_kiocb *link; |
1411 | ||
cbdcb435 PB |
1412 | /* init ->work of the whole link before punting */ |
1413 | io_prep_async_link(req); | |
7271ef3a JA |
1414 | link = __io_queue_async_work(req); |
1415 | ||
1416 | if (link) | |
1417 | io_queue_linked_timeout(link); | |
cbdcb435 PB |
1418 | } |
1419 | ||
5262f567 JA |
1420 | static void io_kill_timeout(struct io_kiocb *req) |
1421 | { | |
e8c2bc1f | 1422 | struct io_timeout_data *io = req->async_data; |
5262f567 JA |
1423 | int ret; |
1424 | ||
e8c2bc1f | 1425 | ret = hrtimer_try_to_cancel(&io->timer); |
5262f567 | 1426 | if (ret != -1) { |
01cec8c1 PB |
1427 | atomic_set(&req->ctx->cq_timeouts, |
1428 | atomic_read(&req->ctx->cq_timeouts) + 1); | |
135fcde8 | 1429 | list_del_init(&req->timeout.list); |
78e19bbe | 1430 | io_cqring_fill_event(req, 0); |
216578e5 | 1431 | io_put_req_deferred(req, 1); |
5262f567 JA |
1432 | } |
1433 | } | |
1434 | ||
f3606e3a JA |
1435 | static bool io_task_match(struct io_kiocb *req, struct task_struct *tsk) |
1436 | { | |
1437 | struct io_ring_ctx *ctx = req->ctx; | |
1438 | ||
1439 | if (!tsk || req->task == tsk) | |
1440 | return true; | |
534ca6d6 JA |
1441 | if (ctx->flags & IORING_SETUP_SQPOLL) { |
1442 | if (ctx->sq_data && req->task == ctx->sq_data->thread) | |
1443 | return true; | |
1444 | } | |
f3606e3a JA |
1445 | return false; |
1446 | } | |
1447 | ||
76e1b642 JA |
1448 | /* |
1449 | * Returns true if we found and killed one or more timeouts | |
1450 | */ | |
1451 | static bool io_kill_timeouts(struct io_ring_ctx *ctx, struct task_struct *tsk) | |
5262f567 JA |
1452 | { |
1453 | struct io_kiocb *req, *tmp; | |
76e1b642 | 1454 | int canceled = 0; |
5262f567 JA |
1455 | |
1456 | spin_lock_irq(&ctx->completion_lock); | |
f3606e3a | 1457 | list_for_each_entry_safe(req, tmp, &ctx->timeout_list, timeout.list) { |
76e1b642 | 1458 | if (io_task_match(req, tsk)) { |
f3606e3a | 1459 | io_kill_timeout(req); |
76e1b642 JA |
1460 | canceled++; |
1461 | } | |
f3606e3a | 1462 | } |
5262f567 | 1463 | spin_unlock_irq(&ctx->completion_lock); |
76e1b642 | 1464 | return canceled != 0; |
5262f567 JA |
1465 | } |
1466 | ||
04518945 | 1467 | static void __io_queue_deferred(struct io_ring_ctx *ctx) |
de0617e4 | 1468 | { |
04518945 | 1469 | do { |
27dc8338 PB |
1470 | struct io_defer_entry *de = list_first_entry(&ctx->defer_list, |
1471 | struct io_defer_entry, list); | |
7271ef3a | 1472 | struct io_kiocb *link; |
de0617e4 | 1473 | |
9cf7c104 | 1474 | if (req_need_defer(de->req, de->seq)) |
04518945 | 1475 | break; |
27dc8338 | 1476 | list_del_init(&de->list); |
cbdcb435 | 1477 | /* punt-init is done before queueing for defer */ |
7271ef3a JA |
1478 | link = __io_queue_async_work(de->req); |
1479 | if (link) { | |
1480 | __io_queue_linked_timeout(link); | |
1481 | /* drop submission reference */ | |
216578e5 | 1482 | io_put_req_deferred(link, 1); |
7271ef3a | 1483 | } |
27dc8338 | 1484 | kfree(de); |
04518945 PB |
1485 | } while (!list_empty(&ctx->defer_list)); |
1486 | } | |
1487 | ||
360428f8 | 1488 | static void io_flush_timeouts(struct io_ring_ctx *ctx) |
de0617e4 | 1489 | { |
360428f8 PB |
1490 | while (!list_empty(&ctx->timeout_list)) { |
1491 | struct io_kiocb *req = list_first_entry(&ctx->timeout_list, | |
135fcde8 | 1492 | struct io_kiocb, timeout.list); |
de0617e4 | 1493 | |
8eb7e2d0 | 1494 | if (io_is_timeout_noseq(req)) |
360428f8 | 1495 | break; |
bfe68a22 PB |
1496 | if (req->timeout.target_seq != ctx->cached_cq_tail |
1497 | - atomic_read(&ctx->cq_timeouts)) | |
360428f8 | 1498 | break; |
bfe68a22 | 1499 | |
135fcde8 | 1500 | list_del_init(&req->timeout.list); |
5262f567 | 1501 | io_kill_timeout(req); |
360428f8 PB |
1502 | } |
1503 | } | |
5262f567 | 1504 | |
360428f8 PB |
1505 | static void io_commit_cqring(struct io_ring_ctx *ctx) |
1506 | { | |
1507 | io_flush_timeouts(ctx); | |
de0617e4 JA |
1508 | __io_commit_cqring(ctx); |
1509 | ||
04518945 PB |
1510 | if (unlikely(!list_empty(&ctx->defer_list))) |
1511 | __io_queue_deferred(ctx); | |
de0617e4 JA |
1512 | } |
1513 | ||
90554200 JA |
1514 | static inline bool io_sqring_full(struct io_ring_ctx *ctx) |
1515 | { | |
1516 | struct io_rings *r = ctx->rings; | |
1517 | ||
1518 | return READ_ONCE(r->sq.tail) - ctx->cached_sq_head == r->sq_ring_entries; | |
1519 | } | |
1520 | ||
2b188cc1 JA |
1521 | static struct io_uring_cqe *io_get_cqring(struct io_ring_ctx *ctx) |
1522 | { | |
75b28aff | 1523 | struct io_rings *rings = ctx->rings; |
2b188cc1 JA |
1524 | unsigned tail; |
1525 | ||
1526 | tail = ctx->cached_cq_tail; | |
115e12e5 SB |
1527 | /* |
1528 | * writes to the cq entry need to come after reading head; the | |
1529 | * control dependency is enough as we're using WRITE_ONCE to | |
1530 | * fill the cq entry | |
1531 | */ | |
75b28aff | 1532 | if (tail - READ_ONCE(rings->cq.head) == rings->cq_ring_entries) |
2b188cc1 JA |
1533 | return NULL; |
1534 | ||
1535 | ctx->cached_cq_tail++; | |
75b28aff | 1536 | return &rings->cqes[tail & ctx->cq_mask]; |
2b188cc1 JA |
1537 | } |
1538 | ||
f2842ab5 JA |
1539 | static inline bool io_should_trigger_evfd(struct io_ring_ctx *ctx) |
1540 | { | |
f0b493e6 JA |
1541 | if (!ctx->cq_ev_fd) |
1542 | return false; | |
7e55a19c SG |
1543 | if (READ_ONCE(ctx->rings->cq_flags) & IORING_CQ_EVENTFD_DISABLED) |
1544 | return false; | |
f2842ab5 JA |
1545 | if (!ctx->eventfd_async) |
1546 | return true; | |
b41e9852 | 1547 | return io_wq_current_is_worker(); |
f2842ab5 JA |
1548 | } |
1549 | ||
b41e9852 | 1550 | static void io_cqring_ev_posted(struct io_ring_ctx *ctx) |
1d7bb1d5 JA |
1551 | { |
1552 | if (waitqueue_active(&ctx->wait)) | |
1553 | wake_up(&ctx->wait); | |
534ca6d6 JA |
1554 | if (ctx->sq_data && waitqueue_active(&ctx->sq_data->wait)) |
1555 | wake_up(&ctx->sq_data->wait); | |
b41e9852 | 1556 | if (io_should_trigger_evfd(ctx)) |
1d7bb1d5 JA |
1557 | eventfd_signal(ctx->cq_ev_fd, 1); |
1558 | } | |
1559 | ||
46930143 PB |
1560 | static void io_cqring_mark_overflow(struct io_ring_ctx *ctx) |
1561 | { | |
1562 | if (list_empty(&ctx->cq_overflow_list)) { | |
1563 | clear_bit(0, &ctx->sq_check_overflow); | |
1564 | clear_bit(0, &ctx->cq_check_overflow); | |
1565 | ctx->rings->sq_flags &= ~IORING_SQ_CQ_OVERFLOW; | |
1566 | } | |
1567 | } | |
1568 | ||
e6c8aa9a JA |
1569 | static inline bool io_match_files(struct io_kiocb *req, |
1570 | struct files_struct *files) | |
1571 | { | |
1572 | if (!files) | |
1573 | return true; | |
dfead8a8 JA |
1574 | if ((req->flags & REQ_F_WORK_INITIALIZED) && |
1575 | (req->work.flags & IO_WQ_WORK_FILES)) | |
98447d65 | 1576 | return req->work.identity->files == files; |
e6c8aa9a JA |
1577 | return false; |
1578 | } | |
1579 | ||
c4a2ed72 | 1580 | /* Returns true if there are no backlogged entries after the flush */ |
e6c8aa9a JA |
1581 | static bool io_cqring_overflow_flush(struct io_ring_ctx *ctx, bool force, |
1582 | struct task_struct *tsk, | |
1583 | struct files_struct *files) | |
1d7bb1d5 JA |
1584 | { |
1585 | struct io_rings *rings = ctx->rings; | |
e6c8aa9a | 1586 | struct io_kiocb *req, *tmp; |
1d7bb1d5 | 1587 | struct io_uring_cqe *cqe; |
1d7bb1d5 JA |
1588 | unsigned long flags; |
1589 | LIST_HEAD(list); | |
1590 | ||
1591 | if (!force) { | |
1592 | if (list_empty_careful(&ctx->cq_overflow_list)) | |
c4a2ed72 | 1593 | return true; |
1d7bb1d5 JA |
1594 | if ((ctx->cached_cq_tail - READ_ONCE(rings->cq.head) == |
1595 | rings->cq_ring_entries)) | |
c4a2ed72 | 1596 | return false; |
1d7bb1d5 JA |
1597 | } |
1598 | ||
1599 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
1600 | ||
1601 | /* if force is set, the ring is going away. always drop after that */ | |
1602 | if (force) | |
69b3e546 | 1603 | ctx->cq_overflow_flushed = 1; |
1d7bb1d5 | 1604 | |
c4a2ed72 | 1605 | cqe = NULL; |
e6c8aa9a JA |
1606 | list_for_each_entry_safe(req, tmp, &ctx->cq_overflow_list, compl.list) { |
1607 | if (tsk && req->task != tsk) | |
1608 | continue; | |
1609 | if (!io_match_files(req, files)) | |
1610 | continue; | |
1611 | ||
1d7bb1d5 JA |
1612 | cqe = io_get_cqring(ctx); |
1613 | if (!cqe && !force) | |
1614 | break; | |
1615 | ||
40d8ddd4 | 1616 | list_move(&req->compl.list, &list); |
1d7bb1d5 JA |
1617 | if (cqe) { |
1618 | WRITE_ONCE(cqe->user_data, req->user_data); | |
1619 | WRITE_ONCE(cqe->res, req->result); | |
0f7e466b | 1620 | WRITE_ONCE(cqe->flags, req->compl.cflags); |
1d7bb1d5 JA |
1621 | } else { |
1622 | WRITE_ONCE(ctx->rings->cq_overflow, | |
1623 | atomic_inc_return(&ctx->cached_cq_overflow)); | |
1624 | } | |
1625 | } | |
1626 | ||
1627 | io_commit_cqring(ctx); | |
46930143 PB |
1628 | io_cqring_mark_overflow(ctx); |
1629 | ||
1d7bb1d5 JA |
1630 | spin_unlock_irqrestore(&ctx->completion_lock, flags); |
1631 | io_cqring_ev_posted(ctx); | |
1632 | ||
1633 | while (!list_empty(&list)) { | |
40d8ddd4 PB |
1634 | req = list_first_entry(&list, struct io_kiocb, compl.list); |
1635 | list_del(&req->compl.list); | |
ec9c02ad | 1636 | io_put_req(req); |
1d7bb1d5 | 1637 | } |
c4a2ed72 JA |
1638 | |
1639 | return cqe != NULL; | |
1d7bb1d5 JA |
1640 | } |
1641 | ||
bcda7baa | 1642 | static void __io_cqring_fill_event(struct io_kiocb *req, long res, long cflags) |
2b188cc1 | 1643 | { |
78e19bbe | 1644 | struct io_ring_ctx *ctx = req->ctx; |
2b188cc1 JA |
1645 | struct io_uring_cqe *cqe; |
1646 | ||
78e19bbe | 1647 | trace_io_uring_complete(ctx, req->user_data, res); |
51c3ff62 | 1648 | |
2b188cc1 JA |
1649 | /* |
1650 | * If we can't get a cq entry, userspace overflowed the | |
1651 | * submission (by quite a lot). Increment the overflow count in | |
1652 | * the ring. | |
1653 | */ | |
1654 | cqe = io_get_cqring(ctx); | |
1d7bb1d5 | 1655 | if (likely(cqe)) { |
78e19bbe | 1656 | WRITE_ONCE(cqe->user_data, req->user_data); |
2b188cc1 | 1657 | WRITE_ONCE(cqe->res, res); |
bcda7baa | 1658 | WRITE_ONCE(cqe->flags, cflags); |
0f212204 JA |
1659 | } else if (ctx->cq_overflow_flushed || req->task->io_uring->in_idle) { |
1660 | /* | |
1661 | * If we're in ring overflow flush mode, or in task cancel mode, | |
1662 | * then we cannot store the request for later flushing, we need | |
1663 | * to drop it on the floor. | |
1664 | */ | |
498ccd9e JA |
1665 | WRITE_ONCE(ctx->rings->cq_overflow, |
1666 | atomic_inc_return(&ctx->cached_cq_overflow)); | |
1d7bb1d5 | 1667 | } else { |
ad3eb2c8 JA |
1668 | if (list_empty(&ctx->cq_overflow_list)) { |
1669 | set_bit(0, &ctx->sq_check_overflow); | |
1670 | set_bit(0, &ctx->cq_check_overflow); | |
6d5f9049 | 1671 | ctx->rings->sq_flags |= IORING_SQ_CQ_OVERFLOW; |
ad3eb2c8 | 1672 | } |
40d8ddd4 | 1673 | io_clean_op(req); |
1d7bb1d5 | 1674 | req->result = res; |
0f7e466b | 1675 | req->compl.cflags = cflags; |
40d8ddd4 PB |
1676 | refcount_inc(&req->refs); |
1677 | list_add_tail(&req->compl.list, &ctx->cq_overflow_list); | |
2b188cc1 JA |
1678 | } |
1679 | } | |
1680 | ||
bcda7baa JA |
1681 | static void io_cqring_fill_event(struct io_kiocb *req, long res) |
1682 | { | |
1683 | __io_cqring_fill_event(req, res, 0); | |
1684 | } | |
1685 | ||
e1e16097 | 1686 | static void io_cqring_add_event(struct io_kiocb *req, long res, long cflags) |
2b188cc1 | 1687 | { |
78e19bbe | 1688 | struct io_ring_ctx *ctx = req->ctx; |
2b188cc1 JA |
1689 | unsigned long flags; |
1690 | ||
1691 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
bcda7baa | 1692 | __io_cqring_fill_event(req, res, cflags); |
2b188cc1 JA |
1693 | io_commit_cqring(ctx); |
1694 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
1695 | ||
8c838788 | 1696 | io_cqring_ev_posted(ctx); |
2b188cc1 JA |
1697 | } |
1698 | ||
229a7b63 | 1699 | static void io_submit_flush_completions(struct io_comp_state *cs) |
bcda7baa | 1700 | { |
229a7b63 JA |
1701 | struct io_ring_ctx *ctx = cs->ctx; |
1702 | ||
1703 | spin_lock_irq(&ctx->completion_lock); | |
1704 | while (!list_empty(&cs->list)) { | |
1705 | struct io_kiocb *req; | |
1706 | ||
3ca405eb PB |
1707 | req = list_first_entry(&cs->list, struct io_kiocb, compl.list); |
1708 | list_del(&req->compl.list); | |
0f7e466b | 1709 | __io_cqring_fill_event(req, req->result, req->compl.cflags); |
216578e5 PB |
1710 | |
1711 | /* | |
1712 | * io_free_req() doesn't care about completion_lock unless one | |
1713 | * of these flags is set. REQ_F_WORK_INITIALIZED is in the list | |
1714 | * because of a potential deadlock with req->work.fs->lock | |
1715 | */ | |
1716 | if (req->flags & (REQ_F_FAIL_LINK|REQ_F_LINK_TIMEOUT | |
1717 | |REQ_F_WORK_INITIALIZED)) { | |
229a7b63 JA |
1718 | spin_unlock_irq(&ctx->completion_lock); |
1719 | io_put_req(req); | |
1720 | spin_lock_irq(&ctx->completion_lock); | |
216578e5 PB |
1721 | } else { |
1722 | io_put_req(req); | |
229a7b63 JA |
1723 | } |
1724 | } | |
1725 | io_commit_cqring(ctx); | |
1726 | spin_unlock_irq(&ctx->completion_lock); | |
1727 | ||
1728 | io_cqring_ev_posted(ctx); | |
1729 | cs->nr = 0; | |
1730 | } | |
1731 | ||
1732 | static void __io_req_complete(struct io_kiocb *req, long res, unsigned cflags, | |
1733 | struct io_comp_state *cs) | |
1734 | { | |
1735 | if (!cs) { | |
1736 | io_cqring_add_event(req, res, cflags); | |
1737 | io_put_req(req); | |
1738 | } else { | |
3ca405eb | 1739 | io_clean_op(req); |
229a7b63 | 1740 | req->result = res; |
0f7e466b | 1741 | req->compl.cflags = cflags; |
3ca405eb | 1742 | list_add_tail(&req->compl.list, &cs->list); |
229a7b63 JA |
1743 | if (++cs->nr >= 32) |
1744 | io_submit_flush_completions(cs); | |
1745 | } | |
e1e16097 JA |
1746 | } |
1747 | ||
1748 | static void io_req_complete(struct io_kiocb *req, long res) | |
bcda7baa | 1749 | { |
229a7b63 | 1750 | __io_req_complete(req, res, 0, NULL); |
bcda7baa JA |
1751 | } |
1752 | ||
0ddf92e8 JA |
1753 | static inline bool io_is_fallback_req(struct io_kiocb *req) |
1754 | { | |
1755 | return req == (struct io_kiocb *) | |
1756 | ((unsigned long) req->ctx->fallback_req & ~1UL); | |
1757 | } | |
1758 | ||
1759 | static struct io_kiocb *io_get_fallback_req(struct io_ring_ctx *ctx) | |
1760 | { | |
1761 | struct io_kiocb *req; | |
1762 | ||
1763 | req = ctx->fallback_req; | |
dd461af6 | 1764 | if (!test_and_set_bit_lock(0, (unsigned long *) &ctx->fallback_req)) |
0ddf92e8 JA |
1765 | return req; |
1766 | ||
1767 | return NULL; | |
1768 | } | |
1769 | ||
0553b8bd PB |
1770 | static struct io_kiocb *io_alloc_req(struct io_ring_ctx *ctx, |
1771 | struct io_submit_state *state) | |
2b188cc1 | 1772 | { |
f6b6c7d6 | 1773 | if (!state->free_reqs) { |
291b2821 | 1774 | gfp_t gfp = GFP_KERNEL | __GFP_NOWARN; |
2579f913 JA |
1775 | size_t sz; |
1776 | int ret; | |
1777 | ||
1778 | sz = min_t(size_t, state->ios_left, ARRAY_SIZE(state->reqs)); | |
fd6fab2c JA |
1779 | ret = kmem_cache_alloc_bulk(req_cachep, gfp, sz, state->reqs); |
1780 | ||
1781 | /* | |
1782 | * Bulk alloc is all-or-nothing. If we fail to get a batch, | |
1783 | * retry single alloc to be on the safe side. | |
1784 | */ | |
1785 | if (unlikely(ret <= 0)) { | |
1786 | state->reqs[0] = kmem_cache_alloc(req_cachep, gfp); | |
1787 | if (!state->reqs[0]) | |
0ddf92e8 | 1788 | goto fallback; |
fd6fab2c JA |
1789 | ret = 1; |
1790 | } | |
291b2821 | 1791 | state->free_reqs = ret; |
2b188cc1 JA |
1792 | } |
1793 | ||
291b2821 PB |
1794 | state->free_reqs--; |
1795 | return state->reqs[state->free_reqs]; | |
0ddf92e8 | 1796 | fallback: |
0553b8bd | 1797 | return io_get_fallback_req(ctx); |
2b188cc1 JA |
1798 | } |
1799 | ||
8da11c19 PB |
1800 | static inline void io_put_file(struct io_kiocb *req, struct file *file, |
1801 | bool fixed) | |
1802 | { | |
1803 | if (fixed) | |
05589553 | 1804 | percpu_ref_put(req->fixed_file_refs); |
8da11c19 PB |
1805 | else |
1806 | fput(file); | |
1807 | } | |
1808 | ||
4edf20f9 | 1809 | static void io_dismantle_req(struct io_kiocb *req) |
2b188cc1 | 1810 | { |
3ca405eb | 1811 | io_clean_op(req); |
929a3af9 | 1812 | |
e8c2bc1f JA |
1813 | if (req->async_data) |
1814 | kfree(req->async_data); | |
8da11c19 PB |
1815 | if (req->file) |
1816 | io_put_file(req, req->file, (req->flags & REQ_F_FIXED_FILE)); | |
fcb323cc | 1817 | |
4edf20f9 | 1818 | io_req_clean_work(req); |
e65ef56d JA |
1819 | } |
1820 | ||
216578e5 | 1821 | static void __io_free_req(struct io_kiocb *req) |
c6ca97b3 | 1822 | { |
1e6fa521 JA |
1823 | struct io_uring_task *tctx = req->task->io_uring; |
1824 | struct io_ring_ctx *ctx = req->ctx; | |
216578e5 PB |
1825 | |
1826 | io_dismantle_req(req); | |
c6ca97b3 | 1827 | |
d8a6df10 | 1828 | percpu_counter_dec(&tctx->inflight); |
0f212204 JA |
1829 | if (tctx->in_idle) |
1830 | wake_up(&tctx->wait); | |
e3bc8e9d JA |
1831 | put_task_struct(req->task); |
1832 | ||
b1e50e54 PB |
1833 | if (likely(!io_is_fallback_req(req))) |
1834 | kmem_cache_free(req_cachep, req); | |
1835 | else | |
ecfc5177 PB |
1836 | clear_bit_unlock(0, (unsigned long *) &ctx->fallback_req); |
1837 | percpu_ref_put(&ctx->refs); | |
e65ef56d JA |
1838 | } |
1839 | ||
a197f664 | 1840 | static bool io_link_cancel_timeout(struct io_kiocb *req) |
2665abfd | 1841 | { |
e8c2bc1f | 1842 | struct io_timeout_data *io = req->async_data; |
a197f664 | 1843 | struct io_ring_ctx *ctx = req->ctx; |
2665abfd JA |
1844 | int ret; |
1845 | ||
e8c2bc1f | 1846 | ret = hrtimer_try_to_cancel(&io->timer); |
2665abfd | 1847 | if (ret != -1) { |
78e19bbe | 1848 | io_cqring_fill_event(req, -ECANCELED); |
2665abfd | 1849 | io_commit_cqring(ctx); |
dea3b49c | 1850 | req->flags &= ~REQ_F_LINK_HEAD; |
216578e5 | 1851 | io_put_req_deferred(req, 1); |
2665abfd JA |
1852 | return true; |
1853 | } | |
1854 | ||
1855 | return false; | |
e65ef56d JA |
1856 | } |
1857 | ||
ab0b6451 | 1858 | static bool __io_kill_linked_timeout(struct io_kiocb *req) |
9e645e11 | 1859 | { |
7c86ffee | 1860 | struct io_kiocb *link; |
ab0b6451 | 1861 | bool wake_ev; |
7c86ffee PB |
1862 | |
1863 | if (list_empty(&req->link_list)) | |
ab0b6451 | 1864 | return false; |
7c86ffee PB |
1865 | link = list_first_entry(&req->link_list, struct io_kiocb, link_list); |
1866 | if (link->opcode != IORING_OP_LINK_TIMEOUT) | |
ab0b6451 | 1867 | return false; |
7c86ffee PB |
1868 | |
1869 | list_del_init(&link->link_list); | |
1870 | wake_ev = io_link_cancel_timeout(link); | |
1871 | req->flags &= ~REQ_F_LINK_TIMEOUT; | |
ab0b6451 JA |
1872 | return wake_ev; |
1873 | } | |
1874 | ||
1875 | static void io_kill_linked_timeout(struct io_kiocb *req) | |
9e645e11 | 1876 | { |
2665abfd | 1877 | struct io_ring_ctx *ctx = req->ctx; |
216578e5 | 1878 | unsigned long flags; |
ab0b6451 | 1879 | bool wake_ev; |
9e645e11 | 1880 | |
216578e5 PB |
1881 | spin_lock_irqsave(&ctx->completion_lock, flags); |
1882 | wake_ev = __io_kill_linked_timeout(req); | |
1883 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
ab0b6451 | 1884 | |
7c86ffee PB |
1885 | if (wake_ev) |
1886 | io_cqring_ev_posted(ctx); | |
1887 | } | |
1888 | ||
9b5f7bd9 | 1889 | static struct io_kiocb *io_req_link_next(struct io_kiocb *req) |
7c86ffee PB |
1890 | { |
1891 | struct io_kiocb *nxt; | |
4d7dd462 | 1892 | |
9e645e11 JA |
1893 | /* |
1894 | * The list should never be empty when we are called here. But could | |
1895 | * potentially happen if the chain is messed up, check to be on the | |
1896 | * safe side. | |
1897 | */ | |
7c86ffee | 1898 | if (unlikely(list_empty(&req->link_list))) |
9b5f7bd9 | 1899 | return NULL; |
2665abfd | 1900 | |
7c86ffee PB |
1901 | nxt = list_first_entry(&req->link_list, struct io_kiocb, link_list); |
1902 | list_del_init(&req->link_list); | |
1903 | if (!list_empty(&nxt->link_list)) | |
1904 | nxt->flags |= REQ_F_LINK_HEAD; | |
9b5f7bd9 | 1905 | return nxt; |
9e645e11 JA |
1906 | } |
1907 | ||
1908 | /* | |
dea3b49c | 1909 | * Called if REQ_F_LINK_HEAD is set, and we fail the head request |
9e645e11 | 1910 | */ |
7c86ffee | 1911 | static void __io_fail_links(struct io_kiocb *req) |
9e645e11 | 1912 | { |
2665abfd | 1913 | struct io_ring_ctx *ctx = req->ctx; |
9e645e11 JA |
1914 | |
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); | |
9e645e11 JA |
1936 | } |
1937 | ||
7c86ffee | 1938 | static void io_fail_links(struct io_kiocb *req) |
9e645e11 | 1939 | { |
7c86ffee | 1940 | struct io_ring_ctx *ctx = req->ctx; |
216578e5 | 1941 | unsigned long flags; |
2665abfd | 1942 | |
216578e5 PB |
1943 | spin_lock_irqsave(&ctx->completion_lock, flags); |
1944 | __io_fail_links(req); | |
1945 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
9e645e11 | 1946 | |
2665abfd | 1947 | io_cqring_ev_posted(ctx); |
9e645e11 JA |
1948 | } |
1949 | ||
3fa5e0f3 | 1950 | static struct io_kiocb *__io_req_find_next(struct io_kiocb *req) |
c69f8dbe | 1951 | { |
9b0d911a | 1952 | req->flags &= ~REQ_F_LINK_HEAD; |
7c86ffee PB |
1953 | if (req->flags & REQ_F_LINK_TIMEOUT) |
1954 | io_kill_linked_timeout(req); | |
944e58bf | 1955 | |
9e645e11 JA |
1956 | /* |
1957 | * If LINK is set, we have dependent requests in this chain. If we | |
1958 | * didn't fail this request, queue the first one up, moving any other | |
1959 | * dependencies to the next request. In case of failure, fail the rest | |
1960 | * of the chain. | |
1961 | */ | |
9b5f7bd9 PB |
1962 | if (likely(!(req->flags & REQ_F_FAIL_LINK))) |
1963 | return io_req_link_next(req); | |
1964 | io_fail_links(req); | |
1965 | return NULL; | |
4d7dd462 | 1966 | } |
9e645e11 | 1967 | |
3fa5e0f3 PB |
1968 | static struct io_kiocb *io_req_find_next(struct io_kiocb *req) |
1969 | { | |
1970 | if (likely(!(req->flags & REQ_F_LINK_HEAD))) | |
1971 | return NULL; | |
1972 | return __io_req_find_next(req); | |
1973 | } | |
1974 | ||
87c4311f | 1975 | static int io_req_task_work_add(struct io_kiocb *req, bool twa_signal_ok) |
c2c4c83c JA |
1976 | { |
1977 | struct task_struct *tsk = req->task; | |
1978 | struct io_ring_ctx *ctx = req->ctx; | |
0ba9c9ed | 1979 | int ret, notify; |
c2c4c83c | 1980 | |
6200b0ae JA |
1981 | if (tsk->flags & PF_EXITING) |
1982 | return -ESRCH; | |
1983 | ||
c2c4c83c | 1984 | /* |
0ba9c9ed JA |
1985 | * SQPOLL kernel thread doesn't need notification, just a wakeup. For |
1986 | * all other cases, use TWA_SIGNAL unconditionally to ensure we're | |
1987 | * processing task_work. There's no reliable way to tell if TWA_RESUME | |
1988 | * will do the job. | |
c2c4c83c | 1989 | */ |
0ba9c9ed | 1990 | notify = 0; |
fd7d6de2 | 1991 | if (!(ctx->flags & IORING_SETUP_SQPOLL) && twa_signal_ok) |
c2c4c83c JA |
1992 | notify = TWA_SIGNAL; |
1993 | ||
87c4311f | 1994 | ret = task_work_add(tsk, &req->task_work, notify); |
c2c4c83c JA |
1995 | if (!ret) |
1996 | wake_up_process(tsk); | |
0ba9c9ed | 1997 | |
c2c4c83c JA |
1998 | return ret; |
1999 | } | |
2000 | ||
c40f6379 JA |
2001 | static void __io_req_task_cancel(struct io_kiocb *req, int error) |
2002 | { | |
2003 | struct io_ring_ctx *ctx = req->ctx; | |
2004 | ||
2005 | spin_lock_irq(&ctx->completion_lock); | |
2006 | io_cqring_fill_event(req, error); | |
2007 | io_commit_cqring(ctx); | |
2008 | spin_unlock_irq(&ctx->completion_lock); | |
2009 | ||
2010 | io_cqring_ev_posted(ctx); | |
2011 | req_set_fail_links(req); | |
2012 | io_double_put_req(req); | |
2013 | } | |
2014 | ||
2015 | static void io_req_task_cancel(struct callback_head *cb) | |
2016 | { | |
2017 | struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work); | |
87ceb6a6 | 2018 | struct io_ring_ctx *ctx = req->ctx; |
c40f6379 JA |
2019 | |
2020 | __io_req_task_cancel(req, -ECANCELED); | |
87ceb6a6 | 2021 | percpu_ref_put(&ctx->refs); |
c40f6379 JA |
2022 | } |
2023 | ||
2024 | static void __io_req_task_submit(struct io_kiocb *req) | |
2025 | { | |
2026 | struct io_ring_ctx *ctx = req->ctx; | |
2027 | ||
c40f6379 JA |
2028 | if (!__io_sq_thread_acquire_mm(ctx)) { |
2029 | mutex_lock(&ctx->uring_lock); | |
c1379e24 | 2030 | __io_queue_sqe(req, NULL); |
c40f6379 JA |
2031 | mutex_unlock(&ctx->uring_lock); |
2032 | } else { | |
2033 | __io_req_task_cancel(req, -EFAULT); | |
2034 | } | |
2035 | } | |
2036 | ||
2037 | static void io_req_task_submit(struct callback_head *cb) | |
2038 | { | |
2039 | struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work); | |
6d816e08 | 2040 | struct io_ring_ctx *ctx = req->ctx; |
c40f6379 JA |
2041 | |
2042 | __io_req_task_submit(req); | |
6d816e08 | 2043 | percpu_ref_put(&ctx->refs); |
c40f6379 JA |
2044 | } |
2045 | ||
2046 | static void io_req_task_queue(struct io_kiocb *req) | |
2047 | { | |
c40f6379 JA |
2048 | int ret; |
2049 | ||
2050 | init_task_work(&req->task_work, io_req_task_submit); | |
6d816e08 | 2051 | percpu_ref_get(&req->ctx->refs); |
c40f6379 | 2052 | |
87c4311f | 2053 | ret = io_req_task_work_add(req, true); |
c40f6379 | 2054 | if (unlikely(ret)) { |
c2c4c83c JA |
2055 | struct task_struct *tsk; |
2056 | ||
c40f6379 JA |
2057 | init_task_work(&req->task_work, io_req_task_cancel); |
2058 | tsk = io_wq_get_task(req->ctx->io_wq); | |
c2c4c83c JA |
2059 | task_work_add(tsk, &req->task_work, 0); |
2060 | wake_up_process(tsk); | |
c40f6379 | 2061 | } |
c40f6379 JA |
2062 | } |
2063 | ||
c3524383 | 2064 | static void io_queue_next(struct io_kiocb *req) |
c69f8dbe | 2065 | { |
9b5f7bd9 | 2066 | struct io_kiocb *nxt = io_req_find_next(req); |
944e58bf PB |
2067 | |
2068 | if (nxt) | |
906a8c3f | 2069 | io_req_task_queue(nxt); |
c69f8dbe JL |
2070 | } |
2071 | ||
c3524383 | 2072 | static void io_free_req(struct io_kiocb *req) |
7a743e22 | 2073 | { |
c3524383 PB |
2074 | io_queue_next(req); |
2075 | __io_free_req(req); | |
2076 | } | |
8766dd51 | 2077 | |
2d6500d4 PB |
2078 | struct req_batch { |
2079 | void *reqs[IO_IOPOLL_BATCH]; | |
2080 | int to_free; | |
7a743e22 | 2081 | |
5af1d13e PB |
2082 | struct task_struct *task; |
2083 | int task_refs; | |
2d6500d4 PB |
2084 | }; |
2085 | ||
5af1d13e PB |
2086 | static inline void io_init_req_batch(struct req_batch *rb) |
2087 | { | |
2088 | rb->to_free = 0; | |
2089 | rb->task_refs = 0; | |
2090 | rb->task = NULL; | |
2091 | } | |
2092 | ||
2d6500d4 PB |
2093 | static void __io_req_free_batch_flush(struct io_ring_ctx *ctx, |
2094 | struct req_batch *rb) | |
2095 | { | |
2096 | kmem_cache_free_bulk(req_cachep, rb->to_free, rb->reqs); | |
2097 | percpu_ref_put_many(&ctx->refs, rb->to_free); | |
2098 | rb->to_free = 0; | |
2099 | } | |
2100 | ||
2101 | static void io_req_free_batch_finish(struct io_ring_ctx *ctx, | |
2102 | struct req_batch *rb) | |
2103 | { | |
2104 | if (rb->to_free) | |
2105 | __io_req_free_batch_flush(ctx, rb); | |
5af1d13e | 2106 | if (rb->task) { |
d8a6df10 JA |
2107 | struct io_uring_task *tctx = rb->task->io_uring; |
2108 | ||
2109 | percpu_counter_sub(&tctx->inflight, rb->task_refs); | |
5af1d13e PB |
2110 | put_task_struct_many(rb->task, rb->task_refs); |
2111 | rb->task = NULL; | |
2112 | } | |
2d6500d4 PB |
2113 | } |
2114 | ||
2115 | static void io_req_free_batch(struct req_batch *rb, struct io_kiocb *req) | |
2116 | { | |
2117 | if (unlikely(io_is_fallback_req(req))) { | |
2118 | io_free_req(req); | |
2119 | return; | |
2120 | } | |
2121 | if (req->flags & REQ_F_LINK_HEAD) | |
2122 | io_queue_next(req); | |
2123 | ||
e3bc8e9d | 2124 | if (req->task != rb->task) { |
0f212204 | 2125 | if (rb->task) { |
d8a6df10 JA |
2126 | struct io_uring_task *tctx = rb->task->io_uring; |
2127 | ||
2128 | percpu_counter_sub(&tctx->inflight, rb->task_refs); | |
e3bc8e9d | 2129 | put_task_struct_many(rb->task, rb->task_refs); |
5af1d13e | 2130 | } |
e3bc8e9d JA |
2131 | rb->task = req->task; |
2132 | rb->task_refs = 0; | |
5af1d13e | 2133 | } |
e3bc8e9d | 2134 | rb->task_refs++; |
5af1d13e | 2135 | |
4edf20f9 | 2136 | io_dismantle_req(req); |
2d6500d4 PB |
2137 | rb->reqs[rb->to_free++] = req; |
2138 | if (unlikely(rb->to_free == ARRAY_SIZE(rb->reqs))) | |
2139 | __io_req_free_batch_flush(req->ctx, rb); | |
7a743e22 PB |
2140 | } |
2141 | ||
ba816ad6 JA |
2142 | /* |
2143 | * Drop reference to request, return next in chain (if there is one) if this | |
2144 | * was the last reference to this request. | |
2145 | */ | |
9b5f7bd9 | 2146 | static struct io_kiocb *io_put_req_find_next(struct io_kiocb *req) |
e65ef56d | 2147 | { |
9b5f7bd9 PB |
2148 | struct io_kiocb *nxt = NULL; |
2149 | ||
2a44f467 | 2150 | if (refcount_dec_and_test(&req->refs)) { |
9b5f7bd9 | 2151 | nxt = io_req_find_next(req); |
4d7dd462 | 2152 | __io_free_req(req); |
2a44f467 | 2153 | } |
9b5f7bd9 | 2154 | return nxt; |
2b188cc1 JA |
2155 | } |
2156 | ||
e65ef56d JA |
2157 | static void io_put_req(struct io_kiocb *req) |
2158 | { | |
2159 | if (refcount_dec_and_test(&req->refs)) | |
2160 | io_free_req(req); | |
2b188cc1 JA |
2161 | } |
2162 | ||
216578e5 PB |
2163 | static void io_put_req_deferred_cb(struct callback_head *cb) |
2164 | { | |
2165 | struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work); | |
2166 | ||
2167 | io_free_req(req); | |
2168 | } | |
2169 | ||
2170 | static void io_free_req_deferred(struct io_kiocb *req) | |
2171 | { | |
2172 | int ret; | |
2173 | ||
2174 | init_task_work(&req->task_work, io_put_req_deferred_cb); | |
2175 | ret = io_req_task_work_add(req, true); | |
2176 | if (unlikely(ret)) { | |
2177 | struct task_struct *tsk; | |
2178 | ||
2179 | tsk = io_wq_get_task(req->ctx->io_wq); | |
2180 | task_work_add(tsk, &req->task_work, 0); | |
2181 | wake_up_process(tsk); | |
2182 | } | |
2183 | } | |
2184 | ||
2185 | static inline void io_put_req_deferred(struct io_kiocb *req, int refs) | |
2186 | { | |
2187 | if (refcount_sub_and_test(refs, &req->refs)) | |
2188 | io_free_req_deferred(req); | |
2189 | } | |
2190 | ||
f4db7182 | 2191 | static struct io_wq_work *io_steal_work(struct io_kiocb *req) |
7a743e22 | 2192 | { |
6df1db6b | 2193 | struct io_kiocb *nxt; |
f4db7182 | 2194 | |
7a743e22 | 2195 | /* |
f4db7182 PB |
2196 | * A ref is owned by io-wq in which context we're. So, if that's the |
2197 | * last one, it's safe to steal next work. False negatives are Ok, | |
2198 | * it just will be re-punted async in io_put_work() | |
7a743e22 | 2199 | */ |
f4db7182 PB |
2200 | if (refcount_read(&req->refs) != 1) |
2201 | return NULL; | |
7a743e22 | 2202 | |
9b5f7bd9 | 2203 | nxt = io_req_find_next(req); |
6df1db6b | 2204 | return nxt ? &nxt->work : NULL; |
7a743e22 PB |
2205 | } |
2206 | ||
978db57e JA |
2207 | static void io_double_put_req(struct io_kiocb *req) |
2208 | { | |
2209 | /* drop both submit and complete references */ | |
2210 | if (refcount_sub_and_test(2, &req->refs)) | |
2211 | io_free_req(req); | |
2212 | } | |
2213 | ||
1d7bb1d5 | 2214 | static unsigned io_cqring_events(struct io_ring_ctx *ctx, bool noflush) |
a3a0e43f | 2215 | { |
84f97dc2 JA |
2216 | struct io_rings *rings = ctx->rings; |
2217 | ||
ad3eb2c8 JA |
2218 | if (test_bit(0, &ctx->cq_check_overflow)) { |
2219 | /* | |
2220 | * noflush == true is from the waitqueue handler, just ensure | |
2221 | * we wake up the task, and the next invocation will flush the | |
2222 | * entries. We cannot safely to it from here. | |
2223 | */ | |
2224 | if (noflush && !list_empty(&ctx->cq_overflow_list)) | |
2225 | return -1U; | |
1d7bb1d5 | 2226 | |
e6c8aa9a | 2227 | io_cqring_overflow_flush(ctx, false, NULL, NULL); |
ad3eb2c8 | 2228 | } |
1d7bb1d5 | 2229 | |
a3a0e43f JA |
2230 | /* See comment at the top of this file */ |
2231 | smp_rmb(); | |
ad3eb2c8 | 2232 | return ctx->cached_cq_tail - READ_ONCE(rings->cq.head); |
a3a0e43f JA |
2233 | } |
2234 | ||
fb5ccc98 PB |
2235 | static inline unsigned int io_sqring_entries(struct io_ring_ctx *ctx) |
2236 | { | |
2237 | struct io_rings *rings = ctx->rings; | |
2238 | ||
2239 | /* make sure SQ entry isn't read before tail */ | |
2240 | return smp_load_acquire(&rings->sq.tail) - ctx->cached_sq_head; | |
2241 | } | |
2242 | ||
8ff069bf | 2243 | static unsigned int io_put_kbuf(struct io_kiocb *req, struct io_buffer *kbuf) |
e94f141b | 2244 | { |
8ff069bf | 2245 | unsigned int cflags; |
e94f141b | 2246 | |
bcda7baa JA |
2247 | cflags = kbuf->bid << IORING_CQE_BUFFER_SHIFT; |
2248 | cflags |= IORING_CQE_F_BUFFER; | |
0e1b6fe3 | 2249 | req->flags &= ~REQ_F_BUFFER_SELECTED; |
bcda7baa JA |
2250 | kfree(kbuf); |
2251 | return cflags; | |
e94f141b JA |
2252 | } |
2253 | ||
8ff069bf | 2254 | static inline unsigned int io_put_rw_kbuf(struct io_kiocb *req) |
bcda7baa | 2255 | { |
4d954c25 | 2256 | struct io_buffer *kbuf; |
bcda7baa | 2257 | |
4d954c25 | 2258 | kbuf = (struct io_buffer *) (unsigned long) req->rw.addr; |
8ff069bf PB |
2259 | return io_put_kbuf(req, kbuf); |
2260 | } | |
2261 | ||
4c6e277c JA |
2262 | static inline bool io_run_task_work(void) |
2263 | { | |
6200b0ae JA |
2264 | /* |
2265 | * Not safe to run on exiting task, and the task_work handling will | |
2266 | * not add work to such a task. | |
2267 | */ | |
2268 | if (unlikely(current->flags & PF_EXITING)) | |
2269 | return false; | |
4c6e277c JA |
2270 | if (current->task_works) { |
2271 | __set_current_state(TASK_RUNNING); | |
2272 | task_work_run(); | |
2273 | return true; | |
2274 | } | |
2275 | ||
2276 | return false; | |
bcda7baa JA |
2277 | } |
2278 | ||
bbde017a XW |
2279 | static void io_iopoll_queue(struct list_head *again) |
2280 | { | |
2281 | struct io_kiocb *req; | |
2282 | ||
2283 | do { | |
d21ffe7e PB |
2284 | req = list_first_entry(again, struct io_kiocb, inflight_entry); |
2285 | list_del(&req->inflight_entry); | |
81b68a5c | 2286 | __io_complete_rw(req, -EAGAIN, 0, NULL); |
bbde017a XW |
2287 | } while (!list_empty(again)); |
2288 | } | |
2289 | ||
def596e9 JA |
2290 | /* |
2291 | * Find and free completed poll iocbs | |
2292 | */ | |
2293 | static void io_iopoll_complete(struct io_ring_ctx *ctx, unsigned int *nr_events, | |
2294 | struct list_head *done) | |
2295 | { | |
8237e045 | 2296 | struct req_batch rb; |
def596e9 | 2297 | struct io_kiocb *req; |
bbde017a XW |
2298 | LIST_HEAD(again); |
2299 | ||
2300 | /* order with ->result store in io_complete_rw_iopoll() */ | |
2301 | smp_rmb(); | |
def596e9 | 2302 | |
5af1d13e | 2303 | io_init_req_batch(&rb); |
def596e9 | 2304 | while (!list_empty(done)) { |
bcda7baa JA |
2305 | int cflags = 0; |
2306 | ||
d21ffe7e | 2307 | req = list_first_entry(done, struct io_kiocb, inflight_entry); |
bbde017a | 2308 | if (READ_ONCE(req->result) == -EAGAIN) { |
56450c20 | 2309 | req->result = 0; |
bbde017a | 2310 | req->iopoll_completed = 0; |
d21ffe7e | 2311 | list_move_tail(&req->inflight_entry, &again); |
bbde017a XW |
2312 | continue; |
2313 | } | |
d21ffe7e | 2314 | list_del(&req->inflight_entry); |
def596e9 | 2315 | |
bcda7baa | 2316 | if (req->flags & REQ_F_BUFFER_SELECTED) |
8ff069bf | 2317 | cflags = io_put_rw_kbuf(req); |
bcda7baa JA |
2318 | |
2319 | __io_cqring_fill_event(req, req->result, cflags); | |
def596e9 JA |
2320 | (*nr_events)++; |
2321 | ||
c3524383 | 2322 | if (refcount_dec_and_test(&req->refs)) |
2d6500d4 | 2323 | io_req_free_batch(&rb, req); |
def596e9 | 2324 | } |
def596e9 | 2325 | |
09bb8394 | 2326 | io_commit_cqring(ctx); |
32b2244a XW |
2327 | if (ctx->flags & IORING_SETUP_SQPOLL) |
2328 | io_cqring_ev_posted(ctx); | |
2d6500d4 | 2329 | io_req_free_batch_finish(ctx, &rb); |
581f9810 | 2330 | |
bbde017a XW |
2331 | if (!list_empty(&again)) |
2332 | io_iopoll_queue(&again); | |
581f9810 BM |
2333 | } |
2334 | ||
def596e9 JA |
2335 | static int io_do_iopoll(struct io_ring_ctx *ctx, unsigned int *nr_events, |
2336 | long min) | |
2337 | { | |
2338 | struct io_kiocb *req, *tmp; | |
2339 | LIST_HEAD(done); | |
2340 | bool spin; | |
2341 | int ret; | |
2342 | ||
2343 | /* | |
2344 | * Only spin for completions if we don't have multiple devices hanging | |
2345 | * off our complete list, and we're under the requested amount. | |
2346 | */ | |
2347 | spin = !ctx->poll_multi_file && *nr_events < min; | |
2348 | ||
2349 | ret = 0; | |
d21ffe7e | 2350 | list_for_each_entry_safe(req, tmp, &ctx->iopoll_list, inflight_entry) { |
9adbd45d | 2351 | struct kiocb *kiocb = &req->rw.kiocb; |
def596e9 JA |
2352 | |
2353 | /* | |
581f9810 BM |
2354 | * Move completed and retryable entries to our local lists. |
2355 | * If we find a request that requires polling, break out | |
2356 | * and complete those lists first, if we have entries there. | |
def596e9 | 2357 | */ |
65a6543d | 2358 | if (READ_ONCE(req->iopoll_completed)) { |
d21ffe7e | 2359 | list_move_tail(&req->inflight_entry, &done); |
def596e9 JA |
2360 | continue; |
2361 | } | |
2362 | if (!list_empty(&done)) | |
2363 | break; | |
2364 | ||
2365 | ret = kiocb->ki_filp->f_op->iopoll(kiocb, spin); | |
2366 | if (ret < 0) | |
2367 | break; | |
2368 | ||
3aadc23e PB |
2369 | /* iopoll may have completed current req */ |
2370 | if (READ_ONCE(req->iopoll_completed)) | |
d21ffe7e | 2371 | list_move_tail(&req->inflight_entry, &done); |
3aadc23e | 2372 | |
def596e9 JA |
2373 | if (ret && spin) |
2374 | spin = false; | |
2375 | ret = 0; | |
2376 | } | |
2377 | ||
2378 | if (!list_empty(&done)) | |
2379 | io_iopoll_complete(ctx, nr_events, &done); | |
2380 | ||
2381 | return ret; | |
2382 | } | |
2383 | ||
2384 | /* | |
d195a66e | 2385 | * Poll for a minimum of 'min' events. Note that if min == 0 we consider that a |
def596e9 JA |
2386 | * non-spinning poll check - we'll still enter the driver poll loop, but only |
2387 | * as a non-spinning completion check. | |
2388 | */ | |
2389 | static int io_iopoll_getevents(struct io_ring_ctx *ctx, unsigned int *nr_events, | |
2390 | long min) | |
2391 | { | |
540e32a0 | 2392 | while (!list_empty(&ctx->iopoll_list) && !need_resched()) { |
def596e9 JA |
2393 | int ret; |
2394 | ||
2395 | ret = io_do_iopoll(ctx, nr_events, min); | |
2396 | if (ret < 0) | |
2397 | return ret; | |
eba0a4dd | 2398 | if (*nr_events >= min) |
def596e9 JA |
2399 | return 0; |
2400 | } | |
2401 | ||
2402 | return 1; | |
2403 | } | |
2404 | ||
2405 | /* | |
2406 | * We can't just wait for polled events to come to us, we have to actively | |
2407 | * find and complete them. | |
2408 | */ | |
b2edc0a7 | 2409 | static void io_iopoll_try_reap_events(struct io_ring_ctx *ctx) |
def596e9 JA |
2410 | { |
2411 | if (!(ctx->flags & IORING_SETUP_IOPOLL)) | |
2412 | return; | |
2413 | ||
2414 | mutex_lock(&ctx->uring_lock); | |
540e32a0 | 2415 | while (!list_empty(&ctx->iopoll_list)) { |
def596e9 JA |
2416 | unsigned int nr_events = 0; |
2417 | ||
b2edc0a7 | 2418 | io_do_iopoll(ctx, &nr_events, 0); |
08f5439f | 2419 | |
b2edc0a7 PB |
2420 | /* let it sleep and repeat later if can't complete a request */ |
2421 | if (nr_events == 0) | |
2422 | break; | |
08f5439f JA |
2423 | /* |
2424 | * Ensure we allow local-to-the-cpu processing to take place, | |
2425 | * in this case we need to ensure that we reap all events. | |
3fcee5a6 | 2426 | * Also let task_work, etc. to progress by releasing the mutex |
08f5439f | 2427 | */ |
3fcee5a6 PB |
2428 | if (need_resched()) { |
2429 | mutex_unlock(&ctx->uring_lock); | |
2430 | cond_resched(); | |
2431 | mutex_lock(&ctx->uring_lock); | |
2432 | } | |
def596e9 JA |
2433 | } |
2434 | mutex_unlock(&ctx->uring_lock); | |
2435 | } | |
2436 | ||
7668b92a | 2437 | static int io_iopoll_check(struct io_ring_ctx *ctx, long min) |
def596e9 | 2438 | { |
7668b92a | 2439 | unsigned int nr_events = 0; |
2b2ed975 | 2440 | int iters = 0, ret = 0; |
500f9fba | 2441 | |
c7849be9 XW |
2442 | /* |
2443 | * We disallow the app entering submit/complete with polling, but we | |
2444 | * still need to lock the ring to prevent racing with polled issue | |
2445 | * that got punted to a workqueue. | |
2446 | */ | |
2447 | mutex_lock(&ctx->uring_lock); | |
def596e9 | 2448 | do { |
a3a0e43f JA |
2449 | /* |
2450 | * Don't enter poll loop if we already have events pending. | |
2451 | * If we do, we can potentially be spinning for commands that | |
2452 | * already triggered a CQE (eg in error). | |
2453 | */ | |
1d7bb1d5 | 2454 | if (io_cqring_events(ctx, false)) |
a3a0e43f JA |
2455 | break; |
2456 | ||
500f9fba JA |
2457 | /* |
2458 | * If a submit got punted to a workqueue, we can have the | |
2459 | * application entering polling for a command before it gets | |
2460 | * issued. That app will hold the uring_lock for the duration | |
2461 | * of the poll right here, so we need to take a breather every | |
2462 | * now and then to ensure that the issue has a chance to add | |
2463 | * the poll to the issued list. Otherwise we can spin here | |
2464 | * forever, while the workqueue is stuck trying to acquire the | |
2465 | * very same mutex. | |
2466 | */ | |
2467 | if (!(++iters & 7)) { | |
2468 | mutex_unlock(&ctx->uring_lock); | |
4c6e277c | 2469 | io_run_task_work(); |
500f9fba JA |
2470 | mutex_lock(&ctx->uring_lock); |
2471 | } | |
2472 | ||
7668b92a | 2473 | ret = io_iopoll_getevents(ctx, &nr_events, min); |
def596e9 JA |
2474 | if (ret <= 0) |
2475 | break; | |
2476 | ret = 0; | |
7668b92a | 2477 | } while (min && !nr_events && !need_resched()); |
def596e9 | 2478 | |
500f9fba | 2479 | mutex_unlock(&ctx->uring_lock); |
def596e9 JA |
2480 | return ret; |
2481 | } | |
2482 | ||
491381ce | 2483 | static void kiocb_end_write(struct io_kiocb *req) |
2b188cc1 | 2484 | { |
491381ce JA |
2485 | /* |
2486 | * Tell lockdep we inherited freeze protection from submission | |
2487 | * thread. | |
2488 | */ | |
2489 | if (req->flags & REQ_F_ISREG) { | |
2490 | struct inode *inode = file_inode(req->file); | |
2b188cc1 | 2491 | |
491381ce | 2492 | __sb_writers_acquired(inode->i_sb, SB_FREEZE_WRITE); |
2b188cc1 | 2493 | } |
491381ce | 2494 | file_end_write(req->file); |
2b188cc1 JA |
2495 | } |
2496 | ||
a1d7c393 JA |
2497 | static void io_complete_rw_common(struct kiocb *kiocb, long res, |
2498 | struct io_comp_state *cs) | |
2b188cc1 | 2499 | { |
9adbd45d | 2500 | struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb); |
bcda7baa | 2501 | int cflags = 0; |
2b188cc1 | 2502 | |
491381ce JA |
2503 | if (kiocb->ki_flags & IOCB_WRITE) |
2504 | kiocb_end_write(req); | |
2b188cc1 | 2505 | |
4e88d6e7 JA |
2506 | if (res != req->result) |
2507 | req_set_fail_links(req); | |
bcda7baa | 2508 | if (req->flags & REQ_F_BUFFER_SELECTED) |
8ff069bf | 2509 | cflags = io_put_rw_kbuf(req); |
a1d7c393 | 2510 | __io_req_complete(req, res, cflags, cs); |
ba816ad6 JA |
2511 | } |
2512 | ||
b63534c4 JA |
2513 | #ifdef CONFIG_BLOCK |
2514 | static bool io_resubmit_prep(struct io_kiocb *req, int error) | |
2515 | { | |
2516 | struct iovec inline_vecs[UIO_FASTIOV], *iovec = inline_vecs; | |
2517 | ssize_t ret = -ECANCELED; | |
2518 | struct iov_iter iter; | |
2519 | int rw; | |
2520 | ||
2521 | if (error) { | |
2522 | ret = error; | |
2523 | goto end_req; | |
2524 | } | |
2525 | ||
2526 | switch (req->opcode) { | |
2527 | case IORING_OP_READV: | |
2528 | case IORING_OP_READ_FIXED: | |
2529 | case IORING_OP_READ: | |
2530 | rw = READ; | |
2531 | break; | |
2532 | case IORING_OP_WRITEV: | |
2533 | case IORING_OP_WRITE_FIXED: | |
2534 | case IORING_OP_WRITE: | |
2535 | rw = WRITE; | |
2536 | break; | |
2537 | default: | |
2538 | printk_once(KERN_WARNING "io_uring: bad opcode in resubmit %d\n", | |
2539 | req->opcode); | |
2540 | goto end_req; | |
2541 | } | |
2542 | ||
e8c2bc1f | 2543 | if (!req->async_data) { |
8f3d7496 JA |
2544 | ret = io_import_iovec(rw, req, &iovec, &iter, false); |
2545 | if (ret < 0) | |
2546 | goto end_req; | |
2547 | ret = io_setup_async_rw(req, iovec, inline_vecs, &iter, false); | |
2548 | if (!ret) | |
2549 | return true; | |
2550 | kfree(iovec); | |
2551 | } else { | |
b63534c4 | 2552 | return true; |
8f3d7496 | 2553 | } |
b63534c4 | 2554 | end_req: |
b63534c4 | 2555 | req_set_fail_links(req); |
e1e16097 | 2556 | io_req_complete(req, ret); |
b63534c4 JA |
2557 | return false; |
2558 | } | |
b63534c4 JA |
2559 | #endif |
2560 | ||
2561 | static bool io_rw_reissue(struct io_kiocb *req, long res) | |
2562 | { | |
2563 | #ifdef CONFIG_BLOCK | |
355afaeb | 2564 | umode_t mode = file_inode(req->file)->i_mode; |
b63534c4 JA |
2565 | int ret; |
2566 | ||
355afaeb JA |
2567 | if (!S_ISBLK(mode) && !S_ISREG(mode)) |
2568 | return false; | |
b63534c4 JA |
2569 | if ((res != -EAGAIN && res != -EOPNOTSUPP) || io_wq_current_is_worker()) |
2570 | return false; | |
2571 | ||
fdee946d | 2572 | ret = io_sq_thread_acquire_mm(req->ctx, req); |
6d816e08 | 2573 | |
fdee946d JA |
2574 | if (io_resubmit_prep(req, ret)) { |
2575 | refcount_inc(&req->refs); | |
2576 | io_queue_async_work(req); | |
b63534c4 | 2577 | return true; |
fdee946d JA |
2578 | } |
2579 | ||
b63534c4 JA |
2580 | #endif |
2581 | return false; | |
2582 | } | |
2583 | ||
a1d7c393 JA |
2584 | static void __io_complete_rw(struct io_kiocb *req, long res, long res2, |
2585 | struct io_comp_state *cs) | |
2586 | { | |
2587 | if (!io_rw_reissue(req, res)) | |
2588 | io_complete_rw_common(&req->rw.kiocb, res, cs); | |
ba816ad6 JA |
2589 | } |
2590 | ||
2591 | static void io_complete_rw(struct kiocb *kiocb, long res, long res2) | |
2592 | { | |
9adbd45d | 2593 | struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb); |
ba816ad6 | 2594 | |
a1d7c393 | 2595 | __io_complete_rw(req, res, res2, NULL); |
2b188cc1 JA |
2596 | } |
2597 | ||
def596e9 JA |
2598 | static void io_complete_rw_iopoll(struct kiocb *kiocb, long res, long res2) |
2599 | { | |
9adbd45d | 2600 | struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb); |
def596e9 | 2601 | |
491381ce JA |
2602 | if (kiocb->ki_flags & IOCB_WRITE) |
2603 | kiocb_end_write(req); | |
def596e9 | 2604 | |
2d7d6792 | 2605 | if (res != -EAGAIN && res != req->result) |
4e88d6e7 | 2606 | req_set_fail_links(req); |
bbde017a XW |
2607 | |
2608 | WRITE_ONCE(req->result, res); | |
2609 | /* order with io_poll_complete() checking ->result */ | |
cd664b0e PB |
2610 | smp_wmb(); |
2611 | WRITE_ONCE(req->iopoll_completed, 1); | |
def596e9 JA |
2612 | } |
2613 | ||
2614 | /* | |
2615 | * After the iocb has been issued, it's safe to be found on the poll list. | |
2616 | * Adding the kiocb to the list AFTER submission ensures that we don't | |
2617 | * find it from a io_iopoll_getevents() thread before the issuer is done | |
2618 | * accessing the kiocb cookie. | |
2619 | */ | |
2620 | static void io_iopoll_req_issued(struct io_kiocb *req) | |
2621 | { | |
2622 | struct io_ring_ctx *ctx = req->ctx; | |
2623 | ||
2624 | /* | |
2625 | * Track whether we have multiple files in our lists. This will impact | |
2626 | * how we do polling eventually, not spinning if we're on potentially | |
2627 | * different devices. | |
2628 | */ | |
540e32a0 | 2629 | if (list_empty(&ctx->iopoll_list)) { |
def596e9 JA |
2630 | ctx->poll_multi_file = false; |
2631 | } else if (!ctx->poll_multi_file) { | |
2632 | struct io_kiocb *list_req; | |
2633 | ||
540e32a0 | 2634 | list_req = list_first_entry(&ctx->iopoll_list, struct io_kiocb, |
d21ffe7e | 2635 | inflight_entry); |
9adbd45d | 2636 | if (list_req->file != req->file) |
def596e9 JA |
2637 | ctx->poll_multi_file = true; |
2638 | } | |
2639 | ||
2640 | /* | |
2641 | * For fast devices, IO may have already completed. If it has, add | |
2642 | * it to the front so we find it first. | |
2643 | */ | |
65a6543d | 2644 | if (READ_ONCE(req->iopoll_completed)) |
d21ffe7e | 2645 | list_add(&req->inflight_entry, &ctx->iopoll_list); |
def596e9 | 2646 | else |
d21ffe7e | 2647 | list_add_tail(&req->inflight_entry, &ctx->iopoll_list); |
bdcd3eab XW |
2648 | |
2649 | if ((ctx->flags & IORING_SETUP_SQPOLL) && | |
534ca6d6 JA |
2650 | wq_has_sleeper(&ctx->sq_data->wait)) |
2651 | wake_up(&ctx->sq_data->wait); | |
def596e9 JA |
2652 | } |
2653 | ||
9f13c35b | 2654 | static void __io_state_file_put(struct io_submit_state *state) |
9a56a232 | 2655 | { |
06ef3608 PB |
2656 | if (state->has_refs) |
2657 | fput_many(state->file, state->has_refs); | |
9f13c35b PB |
2658 | state->file = NULL; |
2659 | } | |
2660 | ||
2661 | static inline void io_state_file_put(struct io_submit_state *state) | |
2662 | { | |
2663 | if (state->file) | |
2664 | __io_state_file_put(state); | |
9a56a232 JA |
2665 | } |
2666 | ||
2667 | /* | |
2668 | * Get as many references to a file as we have IOs left in this submission, | |
2669 | * assuming most submissions are for one file, or at least that each file | |
2670 | * has more than one submission. | |
2671 | */ | |
8da11c19 | 2672 | static struct file *__io_file_get(struct io_submit_state *state, int fd) |
9a56a232 JA |
2673 | { |
2674 | if (!state) | |
2675 | return fget(fd); | |
2676 | ||
2677 | if (state->file) { | |
2678 | if (state->fd == fd) { | |
06ef3608 | 2679 | state->has_refs--; |
9a56a232 JA |
2680 | return state->file; |
2681 | } | |
9f13c35b | 2682 | __io_state_file_put(state); |
9a56a232 JA |
2683 | } |
2684 | state->file = fget_many(fd, state->ios_left); | |
2685 | if (!state->file) | |
2686 | return NULL; | |
2687 | ||
2688 | state->fd = fd; | |
71b547c0 | 2689 | state->has_refs = state->ios_left - 1; |
9a56a232 JA |
2690 | return state->file; |
2691 | } | |
2692 | ||
4503b767 JA |
2693 | static bool io_bdev_nowait(struct block_device *bdev) |
2694 | { | |
2695 | #ifdef CONFIG_BLOCK | |
2696 | return !bdev || queue_is_mq(bdev_get_queue(bdev)); | |
2697 | #else | |
2698 | return true; | |
2699 | #endif | |
2700 | } | |
2701 | ||
2b188cc1 JA |
2702 | /* |
2703 | * If we tracked the file through the SCM inflight mechanism, we could support | |
2704 | * any file. For now, just ensure that anything potentially problematic is done | |
2705 | * inline. | |
2706 | */ | |
af197f50 | 2707 | static bool io_file_supports_async(struct file *file, int rw) |
2b188cc1 JA |
2708 | { |
2709 | umode_t mode = file_inode(file)->i_mode; | |
2710 | ||
4503b767 JA |
2711 | if (S_ISBLK(mode)) { |
2712 | if (io_bdev_nowait(file->f_inode->i_bdev)) | |
2713 | return true; | |
2714 | return false; | |
2715 | } | |
2716 | if (S_ISCHR(mode) || S_ISSOCK(mode)) | |
2b188cc1 | 2717 | return true; |
4503b767 JA |
2718 | if (S_ISREG(mode)) { |
2719 | if (io_bdev_nowait(file->f_inode->i_sb->s_bdev) && | |
2720 | file->f_op != &io_uring_fops) | |
2721 | return true; | |
2722 | return false; | |
2723 | } | |
2b188cc1 | 2724 | |
c5b85625 JA |
2725 | /* any ->read/write should understand O_NONBLOCK */ |
2726 | if (file->f_flags & O_NONBLOCK) | |
2727 | return true; | |
2728 | ||
af197f50 JA |
2729 | if (!(file->f_mode & FMODE_NOWAIT)) |
2730 | return false; | |
2731 | ||
2732 | if (rw == READ) | |
2733 | return file->f_op->read_iter != NULL; | |
2734 | ||
2735 | return file->f_op->write_iter != NULL; | |
2b188cc1 JA |
2736 | } |
2737 | ||
a88fc400 | 2738 | static int io_prep_rw(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
2b188cc1 | 2739 | { |
def596e9 | 2740 | struct io_ring_ctx *ctx = req->ctx; |
9adbd45d | 2741 | struct kiocb *kiocb = &req->rw.kiocb; |
09bb8394 JA |
2742 | unsigned ioprio; |
2743 | int ret; | |
2b188cc1 | 2744 | |
491381ce JA |
2745 | if (S_ISREG(file_inode(req->file)->i_mode)) |
2746 | req->flags |= REQ_F_ISREG; | |
2747 | ||
2b188cc1 | 2748 | kiocb->ki_pos = READ_ONCE(sqe->off); |
ba04291e JA |
2749 | if (kiocb->ki_pos == -1 && !(req->file->f_mode & FMODE_STREAM)) { |
2750 | req->flags |= REQ_F_CUR_POS; | |
2751 | kiocb->ki_pos = req->file->f_pos; | |
2752 | } | |
2b188cc1 | 2753 | kiocb->ki_hint = ki_hint_validate(file_write_hint(kiocb->ki_filp)); |
3e577dcd PB |
2754 | kiocb->ki_flags = iocb_flags(kiocb->ki_filp); |
2755 | ret = kiocb_set_rw_flags(kiocb, READ_ONCE(sqe->rw_flags)); | |
2756 | if (unlikely(ret)) | |
2757 | return ret; | |
2b188cc1 JA |
2758 | |
2759 | ioprio = READ_ONCE(sqe->ioprio); | |
2760 | if (ioprio) { | |
2761 | ret = ioprio_check_cap(ioprio); | |
2762 | if (ret) | |
09bb8394 | 2763 | return ret; |
2b188cc1 JA |
2764 | |
2765 | kiocb->ki_ioprio = ioprio; | |
2766 | } else | |
2767 | kiocb->ki_ioprio = get_current_ioprio(); | |
2768 | ||
8449eeda | 2769 | /* don't allow async punt if RWF_NOWAIT was requested */ |
c5b85625 | 2770 | if (kiocb->ki_flags & IOCB_NOWAIT) |
8449eeda SB |
2771 | req->flags |= REQ_F_NOWAIT; |
2772 | ||
def596e9 | 2773 | if (ctx->flags & IORING_SETUP_IOPOLL) { |
def596e9 JA |
2774 | if (!(kiocb->ki_flags & IOCB_DIRECT) || |
2775 | !kiocb->ki_filp->f_op->iopoll) | |
09bb8394 | 2776 | return -EOPNOTSUPP; |
2b188cc1 | 2777 | |
def596e9 JA |
2778 | kiocb->ki_flags |= IOCB_HIPRI; |
2779 | kiocb->ki_complete = io_complete_rw_iopoll; | |
65a6543d | 2780 | req->iopoll_completed = 0; |
def596e9 | 2781 | } else { |
09bb8394 JA |
2782 | if (kiocb->ki_flags & IOCB_HIPRI) |
2783 | return -EINVAL; | |
def596e9 JA |
2784 | kiocb->ki_complete = io_complete_rw; |
2785 | } | |
9adbd45d | 2786 | |
3529d8c2 JA |
2787 | req->rw.addr = READ_ONCE(sqe->addr); |
2788 | req->rw.len = READ_ONCE(sqe->len); | |
4f4eeba8 | 2789 | req->buf_index = READ_ONCE(sqe->buf_index); |
2b188cc1 | 2790 | return 0; |
2b188cc1 JA |
2791 | } |
2792 | ||
2793 | static inline void io_rw_done(struct kiocb *kiocb, ssize_t ret) | |
2794 | { | |
2795 | switch (ret) { | |
2796 | case -EIOCBQUEUED: | |
2797 | break; | |
2798 | case -ERESTARTSYS: | |
2799 | case -ERESTARTNOINTR: | |
2800 | case -ERESTARTNOHAND: | |
2801 | case -ERESTART_RESTARTBLOCK: | |
2802 | /* | |
2803 | * We can't just restart the syscall, since previously | |
2804 | * submitted sqes may already be in progress. Just fail this | |
2805 | * IO with EINTR. | |
2806 | */ | |
2807 | ret = -EINTR; | |
df561f66 | 2808 | fallthrough; |
2b188cc1 JA |
2809 | default: |
2810 | kiocb->ki_complete(kiocb, ret, 0); | |
2811 | } | |
2812 | } | |
2813 | ||
a1d7c393 JA |
2814 | static void kiocb_done(struct kiocb *kiocb, ssize_t ret, |
2815 | struct io_comp_state *cs) | |
ba816ad6 | 2816 | { |
ba04291e | 2817 | struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb); |
e8c2bc1f | 2818 | struct io_async_rw *io = req->async_data; |
ba04291e | 2819 | |
227c0c96 | 2820 | /* add previously done IO, if any */ |
e8c2bc1f | 2821 | if (io && io->bytes_done > 0) { |
227c0c96 | 2822 | if (ret < 0) |
e8c2bc1f | 2823 | ret = io->bytes_done; |
227c0c96 | 2824 | else |
e8c2bc1f | 2825 | ret += io->bytes_done; |
227c0c96 JA |
2826 | } |
2827 | ||
ba04291e JA |
2828 | if (req->flags & REQ_F_CUR_POS) |
2829 | req->file->f_pos = kiocb->ki_pos; | |
bcaec089 | 2830 | if (ret >= 0 && kiocb->ki_complete == io_complete_rw) |
a1d7c393 | 2831 | __io_complete_rw(req, ret, 0, cs); |
ba816ad6 JA |
2832 | else |
2833 | io_rw_done(kiocb, ret); | |
2834 | } | |
2835 | ||
9adbd45d | 2836 | static ssize_t io_import_fixed(struct io_kiocb *req, int rw, |
7d009165 | 2837 | struct iov_iter *iter) |
edafccee | 2838 | { |
9adbd45d JA |
2839 | struct io_ring_ctx *ctx = req->ctx; |
2840 | size_t len = req->rw.len; | |
edafccee | 2841 | struct io_mapped_ubuf *imu; |
4be1c615 | 2842 | u16 index, buf_index = req->buf_index; |
edafccee JA |
2843 | size_t offset; |
2844 | u64 buf_addr; | |
2845 | ||
edafccee JA |
2846 | if (unlikely(buf_index >= ctx->nr_user_bufs)) |
2847 | return -EFAULT; | |
edafccee JA |
2848 | index = array_index_nospec(buf_index, ctx->nr_user_bufs); |
2849 | imu = &ctx->user_bufs[index]; | |
9adbd45d | 2850 | buf_addr = req->rw.addr; |
edafccee JA |
2851 | |
2852 | /* overflow */ | |
2853 | if (buf_addr + len < buf_addr) | |
2854 | return -EFAULT; | |
2855 | /* not inside the mapped region */ | |
2856 | if (buf_addr < imu->ubuf || buf_addr + len > imu->ubuf + imu->len) | |
2857 | return -EFAULT; | |
2858 | ||
2859 | /* | |
2860 | * May not be a start of buffer, set size appropriately | |
2861 | * and advance us to the beginning. | |
2862 | */ | |
2863 | offset = buf_addr - imu->ubuf; | |
2864 | iov_iter_bvec(iter, rw, imu->bvec, imu->nr_bvecs, offset + len); | |
bd11b3a3 JA |
2865 | |
2866 | if (offset) { | |
2867 | /* | |
2868 | * Don't use iov_iter_advance() here, as it's really slow for | |
2869 | * using the latter parts of a big fixed buffer - it iterates | |
2870 | * over each segment manually. We can cheat a bit here, because | |
2871 | * we know that: | |
2872 | * | |
2873 | * 1) it's a BVEC iter, we set it up | |
2874 | * 2) all bvecs are PAGE_SIZE in size, except potentially the | |
2875 | * first and last bvec | |
2876 | * | |
2877 | * So just find our index, and adjust the iterator afterwards. | |
2878 | * If the offset is within the first bvec (or the whole first | |
2879 | * bvec, just use iov_iter_advance(). This makes it easier | |
2880 | * since we can just skip the first segment, which may not | |
2881 | * be PAGE_SIZE aligned. | |
2882 | */ | |
2883 | const struct bio_vec *bvec = imu->bvec; | |
2884 | ||
2885 | if (offset <= bvec->bv_len) { | |
2886 | iov_iter_advance(iter, offset); | |
2887 | } else { | |
2888 | unsigned long seg_skip; | |
2889 | ||
2890 | /* skip first vec */ | |
2891 | offset -= bvec->bv_len; | |
2892 | seg_skip = 1 + (offset >> PAGE_SHIFT); | |
2893 | ||
2894 | iter->bvec = bvec + seg_skip; | |
2895 | iter->nr_segs -= seg_skip; | |
99c79f66 | 2896 | iter->count -= bvec->bv_len + offset; |
bd11b3a3 | 2897 | iter->iov_offset = offset & ~PAGE_MASK; |
bd11b3a3 JA |
2898 | } |
2899 | } | |
2900 | ||
5e559561 | 2901 | return len; |
edafccee JA |
2902 | } |
2903 | ||
bcda7baa JA |
2904 | static void io_ring_submit_unlock(struct io_ring_ctx *ctx, bool needs_lock) |
2905 | { | |
2906 | if (needs_lock) | |
2907 | mutex_unlock(&ctx->uring_lock); | |
2908 | } | |
2909 | ||
2910 | static void io_ring_submit_lock(struct io_ring_ctx *ctx, bool needs_lock) | |
2911 | { | |
2912 | /* | |
2913 | * "Normal" inline submissions always hold the uring_lock, since we | |
2914 | * grab it from the system call. Same is true for the SQPOLL offload. | |
2915 | * The only exception is when we've detached the request and issue it | |
2916 | * from an async worker thread, grab the lock for that case. | |
2917 | */ | |
2918 | if (needs_lock) | |
2919 | mutex_lock(&ctx->uring_lock); | |
2920 | } | |
2921 | ||
2922 | static struct io_buffer *io_buffer_select(struct io_kiocb *req, size_t *len, | |
2923 | int bgid, struct io_buffer *kbuf, | |
2924 | bool needs_lock) | |
2925 | { | |
2926 | struct io_buffer *head; | |
2927 | ||
2928 | if (req->flags & REQ_F_BUFFER_SELECTED) | |
2929 | return kbuf; | |
2930 | ||
2931 | io_ring_submit_lock(req->ctx, needs_lock); | |
2932 | ||
2933 | lockdep_assert_held(&req->ctx->uring_lock); | |
2934 | ||
2935 | head = idr_find(&req->ctx->io_buffer_idr, bgid); | |
2936 | if (head) { | |
2937 | if (!list_empty(&head->list)) { | |
2938 | kbuf = list_last_entry(&head->list, struct io_buffer, | |
2939 | list); | |
2940 | list_del(&kbuf->list); | |
2941 | } else { | |
2942 | kbuf = head; | |
2943 | idr_remove(&req->ctx->io_buffer_idr, bgid); | |
2944 | } | |
2945 | if (*len > kbuf->len) | |
2946 | *len = kbuf->len; | |
2947 | } else { | |
2948 | kbuf = ERR_PTR(-ENOBUFS); | |
2949 | } | |
2950 | ||
2951 | io_ring_submit_unlock(req->ctx, needs_lock); | |
2952 | ||
2953 | return kbuf; | |
2954 | } | |
2955 | ||
4d954c25 JA |
2956 | static void __user *io_rw_buffer_select(struct io_kiocb *req, size_t *len, |
2957 | bool needs_lock) | |
2958 | { | |
2959 | struct io_buffer *kbuf; | |
4f4eeba8 | 2960 | u16 bgid; |
4d954c25 JA |
2961 | |
2962 | kbuf = (struct io_buffer *) (unsigned long) req->rw.addr; | |
4f4eeba8 | 2963 | bgid = req->buf_index; |
4d954c25 JA |
2964 | kbuf = io_buffer_select(req, len, bgid, kbuf, needs_lock); |
2965 | if (IS_ERR(kbuf)) | |
2966 | return kbuf; | |
2967 | req->rw.addr = (u64) (unsigned long) kbuf; | |
2968 | req->flags |= REQ_F_BUFFER_SELECTED; | |
2969 | return u64_to_user_ptr(kbuf->addr); | |
2970 | } | |
2971 | ||
2972 | #ifdef CONFIG_COMPAT | |
2973 | static ssize_t io_compat_import(struct io_kiocb *req, struct iovec *iov, | |
2974 | bool needs_lock) | |
2975 | { | |
2976 | struct compat_iovec __user *uiov; | |
2977 | compat_ssize_t clen; | |
2978 | void __user *buf; | |
2979 | ssize_t len; | |
2980 | ||
2981 | uiov = u64_to_user_ptr(req->rw.addr); | |
2982 | if (!access_ok(uiov, sizeof(*uiov))) | |
2983 | return -EFAULT; | |
2984 | if (__get_user(clen, &uiov->iov_len)) | |
2985 | return -EFAULT; | |
2986 | if (clen < 0) | |
2987 | return -EINVAL; | |
2988 | ||
2989 | len = clen; | |
2990 | buf = io_rw_buffer_select(req, &len, needs_lock); | |
2991 | if (IS_ERR(buf)) | |
2992 | return PTR_ERR(buf); | |
2993 | iov[0].iov_base = buf; | |
2994 | iov[0].iov_len = (compat_size_t) len; | |
2995 | return 0; | |
2996 | } | |
2997 | #endif | |
2998 | ||
2999 | static ssize_t __io_iov_buffer_select(struct io_kiocb *req, struct iovec *iov, | |
3000 | bool needs_lock) | |
3001 | { | |
3002 | struct iovec __user *uiov = u64_to_user_ptr(req->rw.addr); | |
3003 | void __user *buf; | |
3004 | ssize_t len; | |
3005 | ||
3006 | if (copy_from_user(iov, uiov, sizeof(*uiov))) | |
3007 | return -EFAULT; | |
3008 | ||
3009 | len = iov[0].iov_len; | |
3010 | if (len < 0) | |
3011 | return -EINVAL; | |
3012 | buf = io_rw_buffer_select(req, &len, needs_lock); | |
3013 | if (IS_ERR(buf)) | |
3014 | return PTR_ERR(buf); | |
3015 | iov[0].iov_base = buf; | |
3016 | iov[0].iov_len = len; | |
3017 | return 0; | |
3018 | } | |
3019 | ||
3020 | static ssize_t io_iov_buffer_select(struct io_kiocb *req, struct iovec *iov, | |
3021 | bool needs_lock) | |
3022 | { | |
dddb3e26 JA |
3023 | if (req->flags & REQ_F_BUFFER_SELECTED) { |
3024 | struct io_buffer *kbuf; | |
3025 | ||
3026 | kbuf = (struct io_buffer *) (unsigned long) req->rw.addr; | |
3027 | iov[0].iov_base = u64_to_user_ptr(kbuf->addr); | |
3028 | iov[0].iov_len = kbuf->len; | |
4d954c25 | 3029 | return 0; |
dddb3e26 | 3030 | } |
4d954c25 JA |
3031 | if (!req->rw.len) |
3032 | return 0; | |
3033 | else if (req->rw.len > 1) | |
3034 | return -EINVAL; | |
3035 | ||
3036 | #ifdef CONFIG_COMPAT | |
3037 | if (req->ctx->compat) | |
3038 | return io_compat_import(req, iov, needs_lock); | |
3039 | #endif | |
3040 | ||
3041 | return __io_iov_buffer_select(req, iov, needs_lock); | |
3042 | } | |
3043 | ||
8452fd0c JA |
3044 | static ssize_t __io_import_iovec(int rw, struct io_kiocb *req, |
3045 | struct iovec **iovec, struct iov_iter *iter, | |
3046 | bool needs_lock) | |
2b188cc1 | 3047 | { |
9adbd45d JA |
3048 | void __user *buf = u64_to_user_ptr(req->rw.addr); |
3049 | size_t sqe_len = req->rw.len; | |
4d954c25 | 3050 | ssize_t ret; |
edafccee JA |
3051 | u8 opcode; |
3052 | ||
d625c6ee | 3053 | opcode = req->opcode; |
7d009165 | 3054 | if (opcode == IORING_OP_READ_FIXED || opcode == IORING_OP_WRITE_FIXED) { |
edafccee | 3055 | *iovec = NULL; |
9adbd45d | 3056 | return io_import_fixed(req, rw, iter); |
edafccee | 3057 | } |
2b188cc1 | 3058 | |
bcda7baa | 3059 | /* buffer index only valid with fixed read/write, or buffer select */ |
4f4eeba8 | 3060 | if (req->buf_index && !(req->flags & REQ_F_BUFFER_SELECT)) |
9adbd45d JA |
3061 | return -EINVAL; |
3062 | ||
3a6820f2 | 3063 | if (opcode == IORING_OP_READ || opcode == IORING_OP_WRITE) { |
bcda7baa | 3064 | if (req->flags & REQ_F_BUFFER_SELECT) { |
4d954c25 | 3065 | buf = io_rw_buffer_select(req, &sqe_len, needs_lock); |
867a23ea | 3066 | if (IS_ERR(buf)) |
4d954c25 | 3067 | return PTR_ERR(buf); |
3f9d6441 | 3068 | req->rw.len = sqe_len; |
bcda7baa JA |
3069 | } |
3070 | ||
3a6820f2 JA |
3071 | ret = import_single_range(rw, buf, sqe_len, *iovec, iter); |
3072 | *iovec = NULL; | |
3a901598 | 3073 | return ret < 0 ? ret : sqe_len; |
3a6820f2 JA |
3074 | } |
3075 | ||
4d954c25 JA |
3076 | if (req->flags & REQ_F_BUFFER_SELECT) { |
3077 | ret = io_iov_buffer_select(req, *iovec, needs_lock); | |
3f9d6441 JA |
3078 | if (!ret) { |
3079 | ret = (*iovec)->iov_len; | |
3080 | iov_iter_init(iter, rw, *iovec, 1, ret); | |
3081 | } | |
4d954c25 JA |
3082 | *iovec = NULL; |
3083 | return ret; | |
3084 | } | |
3085 | ||
89cd35c5 CH |
3086 | return __import_iovec(rw, buf, sqe_len, UIO_FASTIOV, iovec, iter, |
3087 | req->ctx->compat); | |
2b188cc1 JA |
3088 | } |
3089 | ||
8452fd0c JA |
3090 | static ssize_t io_import_iovec(int rw, struct io_kiocb *req, |
3091 | struct iovec **iovec, struct iov_iter *iter, | |
3092 | bool needs_lock) | |
3093 | { | |
e8c2bc1f JA |
3094 | struct io_async_rw *iorw = req->async_data; |
3095 | ||
3096 | if (!iorw) | |
8452fd0c JA |
3097 | return __io_import_iovec(rw, req, iovec, iter, needs_lock); |
3098 | *iovec = NULL; | |
e8c2bc1f | 3099 | return iov_iter_count(&iorw->iter); |
8452fd0c JA |
3100 | } |
3101 | ||
0fef9483 JA |
3102 | static inline loff_t *io_kiocb_ppos(struct kiocb *kiocb) |
3103 | { | |
5b09e37e | 3104 | return (kiocb->ki_filp->f_mode & FMODE_STREAM) ? NULL : &kiocb->ki_pos; |
0fef9483 JA |
3105 | } |
3106 | ||
31b51510 | 3107 | /* |
32960613 JA |
3108 | * For files that don't have ->read_iter() and ->write_iter(), handle them |
3109 | * by looping over ->read() or ->write() manually. | |
31b51510 | 3110 | */ |
32960613 JA |
3111 | static ssize_t loop_rw_iter(int rw, struct file *file, struct kiocb *kiocb, |
3112 | struct iov_iter *iter) | |
3113 | { | |
3114 | ssize_t ret = 0; | |
3115 | ||
3116 | /* | |
3117 | * Don't support polled IO through this interface, and we can't | |
3118 | * support non-blocking either. For the latter, this just causes | |
3119 | * the kiocb to be handled from an async context. | |
3120 | */ | |
3121 | if (kiocb->ki_flags & IOCB_HIPRI) | |
3122 | return -EOPNOTSUPP; | |
3123 | if (kiocb->ki_flags & IOCB_NOWAIT) | |
3124 | return -EAGAIN; | |
3125 | ||
3126 | while (iov_iter_count(iter)) { | |
311ae9e1 | 3127 | struct iovec iovec; |
32960613 JA |
3128 | ssize_t nr; |
3129 | ||
311ae9e1 PB |
3130 | if (!iov_iter_is_bvec(iter)) { |
3131 | iovec = iov_iter_iovec(iter); | |
3132 | } else { | |
3133 | /* fixed buffers import bvec */ | |
3134 | iovec.iov_base = kmap(iter->bvec->bv_page) | |
3135 | + iter->iov_offset; | |
3136 | iovec.iov_len = min(iter->count, | |
3137 | iter->bvec->bv_len - iter->iov_offset); | |
3138 | } | |
3139 | ||
32960613 JA |
3140 | if (rw == READ) { |
3141 | nr = file->f_op->read(file, iovec.iov_base, | |
0fef9483 | 3142 | iovec.iov_len, io_kiocb_ppos(kiocb)); |
32960613 JA |
3143 | } else { |
3144 | nr = file->f_op->write(file, iovec.iov_base, | |
0fef9483 | 3145 | iovec.iov_len, io_kiocb_ppos(kiocb)); |
32960613 JA |
3146 | } |
3147 | ||
311ae9e1 PB |
3148 | if (iov_iter_is_bvec(iter)) |
3149 | kunmap(iter->bvec->bv_page); | |
3150 | ||
32960613 JA |
3151 | if (nr < 0) { |
3152 | if (!ret) | |
3153 | ret = nr; | |
3154 | break; | |
3155 | } | |
3156 | ret += nr; | |
3157 | if (nr != iovec.iov_len) | |
3158 | break; | |
3159 | iov_iter_advance(iter, nr); | |
3160 | } | |
3161 | ||
3162 | return ret; | |
3163 | } | |
3164 | ||
ff6165b2 JA |
3165 | static void io_req_map_rw(struct io_kiocb *req, const struct iovec *iovec, |
3166 | const struct iovec *fast_iov, struct iov_iter *iter) | |
f67676d1 | 3167 | { |
e8c2bc1f | 3168 | struct io_async_rw *rw = req->async_data; |
b64e3444 | 3169 | |
ff6165b2 | 3170 | memcpy(&rw->iter, iter, sizeof(*iter)); |
afb87658 | 3171 | rw->free_iovec = iovec; |
227c0c96 | 3172 | rw->bytes_done = 0; |
ff6165b2 JA |
3173 | /* can only be fixed buffers, no need to do anything */ |
3174 | if (iter->type == ITER_BVEC) | |
3175 | return; | |
b64e3444 | 3176 | if (!iovec) { |
ff6165b2 JA |
3177 | unsigned iov_off = 0; |
3178 | ||
3179 | rw->iter.iov = rw->fast_iov; | |
3180 | if (iter->iov != fast_iov) { | |
3181 | iov_off = iter->iov - fast_iov; | |
3182 | rw->iter.iov += iov_off; | |
3183 | } | |
3184 | if (rw->fast_iov != fast_iov) | |
3185 | memcpy(rw->fast_iov + iov_off, fast_iov + iov_off, | |
45097dae | 3186 | sizeof(struct iovec) * iter->nr_segs); |
99bc4c38 PB |
3187 | } else { |
3188 | req->flags |= REQ_F_NEED_CLEANUP; | |
f67676d1 JA |
3189 | } |
3190 | } | |
3191 | ||
e8c2bc1f | 3192 | static inline int __io_alloc_async_data(struct io_kiocb *req) |
3d9932a8 | 3193 | { |
e8c2bc1f JA |
3194 | WARN_ON_ONCE(!io_op_defs[req->opcode].async_size); |
3195 | req->async_data = kmalloc(io_op_defs[req->opcode].async_size, GFP_KERNEL); | |
3196 | return req->async_data == NULL; | |
3d9932a8 XW |
3197 | } |
3198 | ||
e8c2bc1f | 3199 | static int io_alloc_async_data(struct io_kiocb *req) |
f67676d1 | 3200 | { |
e8c2bc1f | 3201 | if (!io_op_defs[req->opcode].needs_async_data) |
d3656344 | 3202 | return 0; |
3d9932a8 | 3203 | |
e8c2bc1f | 3204 | return __io_alloc_async_data(req); |
b7bb4f7d JA |
3205 | } |
3206 | ||
ff6165b2 JA |
3207 | static int io_setup_async_rw(struct io_kiocb *req, const struct iovec *iovec, |
3208 | const struct iovec *fast_iov, | |
227c0c96 | 3209 | struct iov_iter *iter, bool force) |
b7bb4f7d | 3210 | { |
e8c2bc1f | 3211 | if (!force && !io_op_defs[req->opcode].needs_async_data) |
74566df3 | 3212 | return 0; |
e8c2bc1f JA |
3213 | if (!req->async_data) { |
3214 | if (__io_alloc_async_data(req)) | |
5d204bcf | 3215 | return -ENOMEM; |
b7bb4f7d | 3216 | |
ff6165b2 | 3217 | io_req_map_rw(req, iovec, fast_iov, iter); |
5d204bcf | 3218 | } |
b7bb4f7d | 3219 | return 0; |
f67676d1 JA |
3220 | } |
3221 | ||
73debe68 | 3222 | static inline int io_rw_prep_async(struct io_kiocb *req, int rw) |
c3e330a4 | 3223 | { |
e8c2bc1f | 3224 | struct io_async_rw *iorw = req->async_data; |
f4bff104 | 3225 | struct iovec *iov = iorw->fast_iov; |
c3e330a4 PB |
3226 | ssize_t ret; |
3227 | ||
73debe68 | 3228 | ret = __io_import_iovec(rw, req, &iov, &iorw->iter, false); |
c3e330a4 PB |
3229 | if (unlikely(ret < 0)) |
3230 | return ret; | |
3231 | ||
ab0b196c PB |
3232 | iorw->bytes_done = 0; |
3233 | iorw->free_iovec = iov; | |
3234 | if (iov) | |
3235 | req->flags |= REQ_F_NEED_CLEANUP; | |
c3e330a4 PB |
3236 | return 0; |
3237 | } | |
3238 | ||
73debe68 | 3239 | static int io_read_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
f67676d1 JA |
3240 | { |
3241 | ssize_t ret; | |
3242 | ||
a88fc400 | 3243 | ret = io_prep_rw(req, sqe); |
3529d8c2 JA |
3244 | if (ret) |
3245 | return ret; | |
f67676d1 | 3246 | |
3529d8c2 JA |
3247 | if (unlikely(!(req->file->f_mode & FMODE_READ))) |
3248 | return -EBADF; | |
f67676d1 | 3249 | |
5f798bea | 3250 | /* either don't need iovec imported or already have it */ |
2d199895 | 3251 | if (!req->async_data) |
3529d8c2 | 3252 | return 0; |
73debe68 | 3253 | return io_rw_prep_async(req, READ); |
f67676d1 JA |
3254 | } |
3255 | ||
c1dd91d1 JA |
3256 | /* |
3257 | * This is our waitqueue callback handler, registered through lock_page_async() | |
3258 | * when we initially tried to do the IO with the iocb armed our waitqueue. | |
3259 | * This gets called when the page is unlocked, and we generally expect that to | |
3260 | * happen when the page IO is completed and the page is now uptodate. This will | |
3261 | * queue a task_work based retry of the operation, attempting to copy the data | |
3262 | * again. If the latter fails because the page was NOT uptodate, then we will | |
3263 | * do a thread based blocking retry of the operation. That's the unexpected | |
3264 | * slow path. | |
3265 | */ | |
bcf5a063 JA |
3266 | static int io_async_buf_func(struct wait_queue_entry *wait, unsigned mode, |
3267 | int sync, void *arg) | |
3268 | { | |
3269 | struct wait_page_queue *wpq; | |
3270 | struct io_kiocb *req = wait->private; | |
bcf5a063 | 3271 | struct wait_page_key *key = arg; |
bcf5a063 JA |
3272 | int ret; |
3273 | ||
3274 | wpq = container_of(wait, struct wait_page_queue, wait); | |
3275 | ||
cdc8fcb4 LT |
3276 | if (!wake_page_match(wpq, key)) |
3277 | return 0; | |
3278 | ||
c8d317aa | 3279 | req->rw.kiocb.ki_flags &= ~IOCB_WAITQ; |
bcf5a063 JA |
3280 | list_del_init(&wait->entry); |
3281 | ||
e7375122 | 3282 | init_task_work(&req->task_work, io_req_task_submit); |
6d816e08 JA |
3283 | percpu_ref_get(&req->ctx->refs); |
3284 | ||
bcf5a063 JA |
3285 | /* submit ref gets dropped, acquire a new one */ |
3286 | refcount_inc(&req->refs); | |
87c4311f | 3287 | ret = io_req_task_work_add(req, true); |
bcf5a063 | 3288 | if (unlikely(ret)) { |
c2c4c83c JA |
3289 | struct task_struct *tsk; |
3290 | ||
bcf5a063 | 3291 | /* queue just for cancelation */ |
e7375122 | 3292 | init_task_work(&req->task_work, io_req_task_cancel); |
bcf5a063 | 3293 | tsk = io_wq_get_task(req->ctx->io_wq); |
e7375122 | 3294 | task_work_add(tsk, &req->task_work, 0); |
c2c4c83c | 3295 | wake_up_process(tsk); |
bcf5a063 | 3296 | } |
bcf5a063 JA |
3297 | return 1; |
3298 | } | |
3299 | ||
c1dd91d1 JA |
3300 | /* |
3301 | * This controls whether a given IO request should be armed for async page | |
3302 | * based retry. If we return false here, the request is handed to the async | |
3303 | * worker threads for retry. If we're doing buffered reads on a regular file, | |
3304 | * we prepare a private wait_page_queue entry and retry the operation. This | |
3305 | * will either succeed because the page is now uptodate and unlocked, or it | |
3306 | * will register a callback when the page is unlocked at IO completion. Through | |
3307 | * that callback, io_uring uses task_work to setup a retry of the operation. | |
3308 | * That retry will attempt the buffered read again. The retry will generally | |
3309 | * succeed, or in rare cases where it fails, we then fall back to using the | |
3310 | * async worker threads for a blocking retry. | |
3311 | */ | |
227c0c96 | 3312 | static bool io_rw_should_retry(struct io_kiocb *req) |
f67676d1 | 3313 | { |
e8c2bc1f JA |
3314 | struct io_async_rw *rw = req->async_data; |
3315 | struct wait_page_queue *wait = &rw->wpq; | |
bcf5a063 | 3316 | struct kiocb *kiocb = &req->rw.kiocb; |
f67676d1 | 3317 | |
bcf5a063 JA |
3318 | /* never retry for NOWAIT, we just complete with -EAGAIN */ |
3319 | if (req->flags & REQ_F_NOWAIT) | |
3320 | return false; | |
f67676d1 | 3321 | |
227c0c96 | 3322 | /* Only for buffered IO */ |
3b2a4439 | 3323 | if (kiocb->ki_flags & (IOCB_DIRECT | IOCB_HIPRI)) |
bcf5a063 | 3324 | return false; |
3b2a4439 | 3325 | |
bcf5a063 JA |
3326 | /* |
3327 | * just use poll if we can, and don't attempt if the fs doesn't | |
3328 | * support callback based unlocks | |
3329 | */ | |
3330 | if (file_can_poll(req->file) || !(req->file->f_mode & FMODE_BUF_RASYNC)) | |
3331 | return false; | |
f67676d1 | 3332 | |
3b2a4439 JA |
3333 | wait->wait.func = io_async_buf_func; |
3334 | wait->wait.private = req; | |
3335 | wait->wait.flags = 0; | |
3336 | INIT_LIST_HEAD(&wait->wait.entry); | |
3337 | kiocb->ki_flags |= IOCB_WAITQ; | |
c8d317aa | 3338 | kiocb->ki_flags &= ~IOCB_NOWAIT; |
3b2a4439 | 3339 | kiocb->ki_waitq = wait; |
3b2a4439 | 3340 | return true; |
bcf5a063 JA |
3341 | } |
3342 | ||
3343 | static int io_iter_do_read(struct io_kiocb *req, struct iov_iter *iter) | |
3344 | { | |
3345 | if (req->file->f_op->read_iter) | |
3346 | return call_read_iter(req->file, &req->rw.kiocb, iter); | |
2dd2111d GH |
3347 | else if (req->file->f_op->read) |
3348 | return loop_rw_iter(READ, req->file, &req->rw.kiocb, iter); | |
3349 | else | |
3350 | return -EINVAL; | |
f67676d1 JA |
3351 | } |
3352 | ||
a1d7c393 JA |
3353 | static int io_read(struct io_kiocb *req, bool force_nonblock, |
3354 | struct io_comp_state *cs) | |
2b188cc1 JA |
3355 | { |
3356 | struct iovec inline_vecs[UIO_FASTIOV], *iovec = inline_vecs; | |
9adbd45d | 3357 | struct kiocb *kiocb = &req->rw.kiocb; |
ff6165b2 | 3358 | struct iov_iter __iter, *iter = &__iter; |
e8c2bc1f | 3359 | struct io_async_rw *rw = req->async_data; |
227c0c96 | 3360 | ssize_t io_size, ret, ret2; |
31b51510 | 3361 | size_t iov_count; |
f5cac8b1 | 3362 | bool no_async; |
ff6165b2 | 3363 | |
e8c2bc1f JA |
3364 | if (rw) |
3365 | iter = &rw->iter; | |
2b188cc1 | 3366 | |
ff6165b2 | 3367 | ret = io_import_iovec(READ, req, &iovec, iter, !force_nonblock); |
06b76d44 JA |
3368 | if (ret < 0) |
3369 | return ret; | |
eefdf30f | 3370 | iov_count = iov_iter_count(iter); |
fa15bafb PB |
3371 | io_size = ret; |
3372 | req->result = io_size; | |
227c0c96 | 3373 | ret = 0; |
2b188cc1 | 3374 | |
fd6c2e4c JA |
3375 | /* Ensure we clear previously set non-block flag */ |
3376 | if (!force_nonblock) | |
29de5f6a | 3377 | kiocb->ki_flags &= ~IOCB_NOWAIT; |
a88fc400 PB |
3378 | else |
3379 | kiocb->ki_flags |= IOCB_NOWAIT; | |
3380 | ||
fd6c2e4c | 3381 | |
24c74678 | 3382 | /* If the file doesn't support async, just async punt */ |
f5cac8b1 JA |
3383 | no_async = force_nonblock && !io_file_supports_async(req->file, READ); |
3384 | if (no_async) | |
f67676d1 | 3385 | goto copy_iov; |
9e645e11 | 3386 | |
0fef9483 | 3387 | ret = rw_verify_area(READ, req->file, io_kiocb_ppos(kiocb), iov_count); |
fa15bafb PB |
3388 | if (unlikely(ret)) |
3389 | goto out_free; | |
2b188cc1 | 3390 | |
227c0c96 | 3391 | ret = io_iter_do_read(req, iter); |
32960613 | 3392 | |
227c0c96 JA |
3393 | if (!ret) { |
3394 | goto done; | |
3395 | } else if (ret == -EIOCBQUEUED) { | |
3396 | ret = 0; | |
3397 | goto out_free; | |
3398 | } else if (ret == -EAGAIN) { | |
eefdf30f JA |
3399 | /* IOPOLL retry should happen for io-wq threads */ |
3400 | if (!force_nonblock && !(req->ctx->flags & IORING_SETUP_IOPOLL)) | |
f91daf56 | 3401 | goto done; |
355afaeb JA |
3402 | /* no retry on NONBLOCK marked file */ |
3403 | if (req->file->f_flags & O_NONBLOCK) | |
3404 | goto done; | |
84216315 JA |
3405 | /* some cases will consume bytes even on error returns */ |
3406 | iov_iter_revert(iter, iov_count - iov_iter_count(iter)); | |
f38c7e3a JA |
3407 | ret = 0; |
3408 | goto copy_iov; | |
227c0c96 | 3409 | } else if (ret < 0) { |
00d23d51 JA |
3410 | /* make sure -ERESTARTSYS -> -EINTR is done */ |
3411 | goto done; | |
227c0c96 JA |
3412 | } |
3413 | ||
3414 | /* read it all, or we did blocking attempt. no retry. */ | |
f91daf56 JA |
3415 | if (!iov_iter_count(iter) || !force_nonblock || |
3416 | (req->file->f_flags & O_NONBLOCK)) | |
227c0c96 JA |
3417 | goto done; |
3418 | ||
3419 | io_size -= ret; | |
3420 | copy_iov: | |
3421 | ret2 = io_setup_async_rw(req, iovec, inline_vecs, iter, true); | |
3422 | if (ret2) { | |
3423 | ret = ret2; | |
3424 | goto out_free; | |
3425 | } | |
f5cac8b1 JA |
3426 | if (no_async) |
3427 | return -EAGAIN; | |
e8c2bc1f | 3428 | rw = req->async_data; |
227c0c96 JA |
3429 | /* it's copied and will be cleaned with ->io */ |
3430 | iovec = NULL; | |
3431 | /* now use our persistent iterator, if we aren't already */ | |
e8c2bc1f | 3432 | iter = &rw->iter; |
227c0c96 | 3433 | retry: |
e8c2bc1f | 3434 | rw->bytes_done += ret; |
227c0c96 JA |
3435 | /* if we can retry, do so with the callbacks armed */ |
3436 | if (!io_rw_should_retry(req)) { | |
fa15bafb PB |
3437 | kiocb->ki_flags &= ~IOCB_WAITQ; |
3438 | return -EAGAIN; | |
2b188cc1 | 3439 | } |
227c0c96 JA |
3440 | |
3441 | /* | |
3442 | * Now retry read with the IOCB_WAITQ parts set in the iocb. If we | |
3443 | * get -EIOCBQUEUED, then we'll get a notification when the desired | |
3444 | * page gets unlocked. We can also get a partial read here, and if we | |
3445 | * do, then just retry at the new offset. | |
3446 | */ | |
3447 | ret = io_iter_do_read(req, iter); | |
3448 | if (ret == -EIOCBQUEUED) { | |
3449 | ret = 0; | |
3450 | goto out_free; | |
3451 | } else if (ret > 0 && ret < io_size) { | |
3452 | /* we got some bytes, but not all. retry. */ | |
3453 | goto retry; | |
3454 | } | |
3455 | done: | |
3456 | kiocb_done(kiocb, ret, cs); | |
3457 | ret = 0; | |
f67676d1 | 3458 | out_free: |
f261c168 | 3459 | /* it's reportedly faster than delegating the null check to kfree() */ |
252917c3 | 3460 | if (iovec) |
6f2cc166 | 3461 | kfree(iovec); |
2b188cc1 JA |
3462 | return ret; |
3463 | } | |
3464 | ||
73debe68 | 3465 | static int io_write_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
f67676d1 JA |
3466 | { |
3467 | ssize_t ret; | |
3468 | ||
a88fc400 | 3469 | ret = io_prep_rw(req, sqe); |
3529d8c2 JA |
3470 | if (ret) |
3471 | return ret; | |
f67676d1 | 3472 | |
3529d8c2 JA |
3473 | if (unlikely(!(req->file->f_mode & FMODE_WRITE))) |
3474 | return -EBADF; | |
f67676d1 | 3475 | |
5f798bea | 3476 | /* either don't need iovec imported or already have it */ |
2d199895 | 3477 | if (!req->async_data) |
3529d8c2 | 3478 | return 0; |
73debe68 | 3479 | return io_rw_prep_async(req, WRITE); |
f67676d1 JA |
3480 | } |
3481 | ||
a1d7c393 JA |
3482 | static int io_write(struct io_kiocb *req, bool force_nonblock, |
3483 | struct io_comp_state *cs) | |
2b188cc1 JA |
3484 | { |
3485 | struct iovec inline_vecs[UIO_FASTIOV], *iovec = inline_vecs; | |
9adbd45d | 3486 | struct kiocb *kiocb = &req->rw.kiocb; |
ff6165b2 | 3487 | struct iov_iter __iter, *iter = &__iter; |
e8c2bc1f | 3488 | struct io_async_rw *rw = req->async_data; |
31b51510 | 3489 | size_t iov_count; |
fa15bafb | 3490 | ssize_t ret, ret2, io_size; |
2b188cc1 | 3491 | |
e8c2bc1f JA |
3492 | if (rw) |
3493 | iter = &rw->iter; | |
ff6165b2 JA |
3494 | |
3495 | ret = io_import_iovec(WRITE, req, &iovec, iter, !force_nonblock); | |
06b76d44 JA |
3496 | if (ret < 0) |
3497 | return ret; | |
eefdf30f | 3498 | iov_count = iov_iter_count(iter); |
fa15bafb PB |
3499 | io_size = ret; |
3500 | req->result = io_size; | |
2b188cc1 | 3501 | |
fd6c2e4c JA |
3502 | /* Ensure we clear previously set non-block flag */ |
3503 | if (!force_nonblock) | |
a88fc400 PB |
3504 | kiocb->ki_flags &= ~IOCB_NOWAIT; |
3505 | else | |
3506 | kiocb->ki_flags |= IOCB_NOWAIT; | |
fd6c2e4c | 3507 | |
24c74678 | 3508 | /* If the file doesn't support async, just async punt */ |
af197f50 | 3509 | if (force_nonblock && !io_file_supports_async(req->file, WRITE)) |
f67676d1 | 3510 | goto copy_iov; |
31b51510 | 3511 | |
10d59345 JA |
3512 | /* file path doesn't support NOWAIT for non-direct_IO */ |
3513 | if (force_nonblock && !(kiocb->ki_flags & IOCB_DIRECT) && | |
3514 | (req->flags & REQ_F_ISREG)) | |
f67676d1 | 3515 | goto copy_iov; |
31b51510 | 3516 | |
0fef9483 | 3517 | ret = rw_verify_area(WRITE, req->file, io_kiocb_ppos(kiocb), iov_count); |
fa15bafb PB |
3518 | if (unlikely(ret)) |
3519 | goto out_free; | |
4ed734b0 | 3520 | |
fa15bafb PB |
3521 | /* |
3522 | * Open-code file_start_write here to grab freeze protection, | |
3523 | * which will be released by another thread in | |
3524 | * io_complete_rw(). Fool lockdep by telling it the lock got | |
3525 | * released so that it doesn't complain about the held lock when | |
3526 | * we return to userspace. | |
3527 | */ | |
3528 | if (req->flags & REQ_F_ISREG) { | |
3529 | __sb_start_write(file_inode(req->file)->i_sb, | |
3530 | SB_FREEZE_WRITE, true); | |
3531 | __sb_writers_release(file_inode(req->file)->i_sb, | |
3532 | SB_FREEZE_WRITE); | |
3533 | } | |
3534 | kiocb->ki_flags |= IOCB_WRITE; | |
4ed734b0 | 3535 | |
fa15bafb | 3536 | if (req->file->f_op->write_iter) |
ff6165b2 | 3537 | ret2 = call_write_iter(req->file, kiocb, iter); |
2dd2111d | 3538 | else if (req->file->f_op->write) |
ff6165b2 | 3539 | ret2 = loop_rw_iter(WRITE, req->file, kiocb, iter); |
2dd2111d GH |
3540 | else |
3541 | ret2 = -EINVAL; | |
4ed734b0 | 3542 | |
fa15bafb PB |
3543 | /* |
3544 | * Raw bdev writes will return -EOPNOTSUPP for IOCB_NOWAIT. Just | |
3545 | * retry them without IOCB_NOWAIT. | |
3546 | */ | |
3547 | if (ret2 == -EOPNOTSUPP && (kiocb->ki_flags & IOCB_NOWAIT)) | |
3548 | ret2 = -EAGAIN; | |
355afaeb JA |
3549 | /* no retry on NONBLOCK marked file */ |
3550 | if (ret2 == -EAGAIN && (req->file->f_flags & O_NONBLOCK)) | |
3551 | goto done; | |
fa15bafb | 3552 | if (!force_nonblock || ret2 != -EAGAIN) { |
eefdf30f JA |
3553 | /* IOPOLL retry should happen for io-wq threads */ |
3554 | if ((req->ctx->flags & IORING_SETUP_IOPOLL) && ret2 == -EAGAIN) | |
3555 | goto copy_iov; | |
355afaeb | 3556 | done: |
fa15bafb PB |
3557 | kiocb_done(kiocb, ret2, cs); |
3558 | } else { | |
f67676d1 | 3559 | copy_iov: |
84216315 JA |
3560 | /* some cases will consume bytes even on error returns */ |
3561 | iov_iter_revert(iter, iov_count - iov_iter_count(iter)); | |
227c0c96 | 3562 | ret = io_setup_async_rw(req, iovec, inline_vecs, iter, false); |
ff6165b2 JA |
3563 | if (!ret) |
3564 | return -EAGAIN; | |
2b188cc1 | 3565 | } |
31b51510 | 3566 | out_free: |
f261c168 | 3567 | /* it's reportedly faster than delegating the null check to kfree() */ |
252917c3 | 3568 | if (iovec) |
6f2cc166 | 3569 | kfree(iovec); |
2b188cc1 JA |
3570 | return ret; |
3571 | } | |
3572 | ||
f2a8d5c7 PB |
3573 | static int __io_splice_prep(struct io_kiocb *req, |
3574 | const struct io_uring_sqe *sqe) | |
7d67af2c PB |
3575 | { |
3576 | struct io_splice* sp = &req->splice; | |
3577 | unsigned int valid_flags = SPLICE_F_FD_IN_FIXED | SPLICE_F_ALL; | |
7d67af2c | 3578 | |
3232dd02 PB |
3579 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
3580 | return -EINVAL; | |
7d67af2c PB |
3581 | |
3582 | sp->file_in = NULL; | |
7d67af2c PB |
3583 | sp->len = READ_ONCE(sqe->len); |
3584 | sp->flags = READ_ONCE(sqe->splice_flags); | |
3585 | ||
3586 | if (unlikely(sp->flags & ~valid_flags)) | |
3587 | return -EINVAL; | |
3588 | ||
8371adf5 PB |
3589 | sp->file_in = io_file_get(NULL, req, READ_ONCE(sqe->splice_fd_in), |
3590 | (sp->flags & SPLICE_F_FD_IN_FIXED)); | |
3591 | if (!sp->file_in) | |
3592 | return -EBADF; | |
7d67af2c PB |
3593 | req->flags |= REQ_F_NEED_CLEANUP; |
3594 | ||
7cdaf587 XW |
3595 | if (!S_ISREG(file_inode(sp->file_in)->i_mode)) { |
3596 | /* | |
3597 | * Splice operation will be punted aync, and here need to | |
3598 | * modify io_wq_work.flags, so initialize io_wq_work firstly. | |
3599 | */ | |
3600 | io_req_init_async(req); | |
7d67af2c | 3601 | req->work.flags |= IO_WQ_WORK_UNBOUND; |
7cdaf587 | 3602 | } |
7d67af2c PB |
3603 | |
3604 | return 0; | |
3605 | } | |
3606 | ||
f2a8d5c7 PB |
3607 | static int io_tee_prep(struct io_kiocb *req, |
3608 | const struct io_uring_sqe *sqe) | |
3609 | { | |
3610 | if (READ_ONCE(sqe->splice_off_in) || READ_ONCE(sqe->off)) | |
3611 | return -EINVAL; | |
3612 | return __io_splice_prep(req, sqe); | |
3613 | } | |
3614 | ||
3615 | static int io_tee(struct io_kiocb *req, bool force_nonblock) | |
3616 | { | |
3617 | struct io_splice *sp = &req->splice; | |
3618 | struct file *in = sp->file_in; | |
3619 | struct file *out = sp->file_out; | |
3620 | unsigned int flags = sp->flags & ~SPLICE_F_FD_IN_FIXED; | |
3621 | long ret = 0; | |
3622 | ||
3623 | if (force_nonblock) | |
3624 | return -EAGAIN; | |
3625 | if (sp->len) | |
3626 | ret = do_tee(in, out, sp->len, flags); | |
3627 | ||
3628 | io_put_file(req, in, (sp->flags & SPLICE_F_FD_IN_FIXED)); | |
3629 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
3630 | ||
f2a8d5c7 PB |
3631 | if (ret != sp->len) |
3632 | req_set_fail_links(req); | |
e1e16097 | 3633 | io_req_complete(req, ret); |
f2a8d5c7 PB |
3634 | return 0; |
3635 | } | |
3636 | ||
3637 | static int io_splice_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) | |
3638 | { | |
3639 | struct io_splice* sp = &req->splice; | |
3640 | ||
3641 | sp->off_in = READ_ONCE(sqe->splice_off_in); | |
3642 | sp->off_out = READ_ONCE(sqe->off); | |
3643 | return __io_splice_prep(req, sqe); | |
3644 | } | |
3645 | ||
014db007 | 3646 | static int io_splice(struct io_kiocb *req, bool force_nonblock) |
7d67af2c PB |
3647 | { |
3648 | struct io_splice *sp = &req->splice; | |
3649 | struct file *in = sp->file_in; | |
3650 | struct file *out = sp->file_out; | |
3651 | unsigned int flags = sp->flags & ~SPLICE_F_FD_IN_FIXED; | |
3652 | loff_t *poff_in, *poff_out; | |
c9687426 | 3653 | long ret = 0; |
7d67af2c | 3654 | |
2fb3e822 PB |
3655 | if (force_nonblock) |
3656 | return -EAGAIN; | |
7d67af2c PB |
3657 | |
3658 | poff_in = (sp->off_in == -1) ? NULL : &sp->off_in; | |
3659 | poff_out = (sp->off_out == -1) ? NULL : &sp->off_out; | |
c9687426 | 3660 | |
948a7749 | 3661 | if (sp->len) |
c9687426 | 3662 | ret = do_splice(in, poff_in, out, poff_out, sp->len, flags); |
7d67af2c PB |
3663 | |
3664 | io_put_file(req, in, (sp->flags & SPLICE_F_FD_IN_FIXED)); | |
3665 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
3666 | ||
7d67af2c PB |
3667 | if (ret != sp->len) |
3668 | req_set_fail_links(req); | |
e1e16097 | 3669 | io_req_complete(req, ret); |
7d67af2c PB |
3670 | return 0; |
3671 | } | |
3672 | ||
2b188cc1 JA |
3673 | /* |
3674 | * IORING_OP_NOP just posts a completion event, nothing else. | |
3675 | */ | |
229a7b63 | 3676 | static int io_nop(struct io_kiocb *req, struct io_comp_state *cs) |
2b188cc1 JA |
3677 | { |
3678 | struct io_ring_ctx *ctx = req->ctx; | |
2b188cc1 | 3679 | |
def596e9 JA |
3680 | if (unlikely(ctx->flags & IORING_SETUP_IOPOLL)) |
3681 | return -EINVAL; | |
3682 | ||
229a7b63 | 3683 | __io_req_complete(req, 0, 0, cs); |
2b188cc1 JA |
3684 | return 0; |
3685 | } | |
3686 | ||
3529d8c2 | 3687 | static int io_prep_fsync(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
c992fe29 | 3688 | { |
6b06314c | 3689 | struct io_ring_ctx *ctx = req->ctx; |
c992fe29 | 3690 | |
09bb8394 JA |
3691 | if (!req->file) |
3692 | return -EBADF; | |
c992fe29 | 3693 | |
6b06314c | 3694 | if (unlikely(ctx->flags & IORING_SETUP_IOPOLL)) |
def596e9 | 3695 | return -EINVAL; |
edafccee | 3696 | if (unlikely(sqe->addr || sqe->ioprio || sqe->buf_index)) |
c992fe29 CH |
3697 | return -EINVAL; |
3698 | ||
8ed8d3c3 JA |
3699 | req->sync.flags = READ_ONCE(sqe->fsync_flags); |
3700 | if (unlikely(req->sync.flags & ~IORING_FSYNC_DATASYNC)) | |
3701 | return -EINVAL; | |
3702 | ||
3703 | req->sync.off = READ_ONCE(sqe->off); | |
3704 | req->sync.len = READ_ONCE(sqe->len); | |
c992fe29 CH |
3705 | return 0; |
3706 | } | |
3707 | ||
ac45abc0 | 3708 | static int io_fsync(struct io_kiocb *req, bool force_nonblock) |
8ed8d3c3 | 3709 | { |
8ed8d3c3 | 3710 | loff_t end = req->sync.off + req->sync.len; |
8ed8d3c3 JA |
3711 | int ret; |
3712 | ||
ac45abc0 PB |
3713 | /* fsync always requires a blocking context */ |
3714 | if (force_nonblock) | |
3715 | return -EAGAIN; | |
3716 | ||
9adbd45d | 3717 | ret = vfs_fsync_range(req->file, req->sync.off, |
8ed8d3c3 JA |
3718 | end > 0 ? end : LLONG_MAX, |
3719 | req->sync.flags & IORING_FSYNC_DATASYNC); | |
3720 | if (ret < 0) | |
3721 | req_set_fail_links(req); | |
e1e16097 | 3722 | io_req_complete(req, ret); |
c992fe29 CH |
3723 | return 0; |
3724 | } | |
3725 | ||
d63d1b5e JA |
3726 | static int io_fallocate_prep(struct io_kiocb *req, |
3727 | const struct io_uring_sqe *sqe) | |
3728 | { | |
3729 | if (sqe->ioprio || sqe->buf_index || sqe->rw_flags) | |
3730 | return -EINVAL; | |
3232dd02 PB |
3731 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
3732 | return -EINVAL; | |
d63d1b5e JA |
3733 | |
3734 | req->sync.off = READ_ONCE(sqe->off); | |
3735 | req->sync.len = READ_ONCE(sqe->addr); | |
3736 | req->sync.mode = READ_ONCE(sqe->len); | |
3737 | return 0; | |
3738 | } | |
3739 | ||
014db007 | 3740 | static int io_fallocate(struct io_kiocb *req, bool force_nonblock) |
5d17b4a4 | 3741 | { |
ac45abc0 PB |
3742 | int ret; |
3743 | ||
d63d1b5e | 3744 | /* fallocate always requiring blocking context */ |
ac45abc0 | 3745 | if (force_nonblock) |
5d17b4a4 | 3746 | return -EAGAIN; |
ac45abc0 PB |
3747 | ret = vfs_fallocate(req->file, req->sync.mode, req->sync.off, |
3748 | req->sync.len); | |
ac45abc0 PB |
3749 | if (ret < 0) |
3750 | req_set_fail_links(req); | |
e1e16097 | 3751 | io_req_complete(req, ret); |
5d17b4a4 JA |
3752 | return 0; |
3753 | } | |
3754 | ||
ec65fea5 | 3755 | static int __io_openat_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
b7bb4f7d | 3756 | { |
f8748881 | 3757 | const char __user *fname; |
15b71abe | 3758 | int ret; |
b7bb4f7d | 3759 | |
ec65fea5 | 3760 | if (unlikely(sqe->ioprio || sqe->buf_index)) |
15b71abe | 3761 | return -EINVAL; |
ec65fea5 | 3762 | if (unlikely(req->flags & REQ_F_FIXED_FILE)) |
cf3040ca | 3763 | return -EBADF; |
03b1230c | 3764 | |
ec65fea5 PB |
3765 | /* open.how should be already initialised */ |
3766 | if (!(req->open.how.flags & O_PATH) && force_o_largefile()) | |
08a1d26e | 3767 | req->open.how.flags |= O_LARGEFILE; |
3529d8c2 | 3768 | |
25e72d10 PB |
3769 | req->open.dfd = READ_ONCE(sqe->fd); |
3770 | fname = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
f8748881 | 3771 | req->open.filename = getname(fname); |
15b71abe JA |
3772 | if (IS_ERR(req->open.filename)) { |
3773 | ret = PTR_ERR(req->open.filename); | |
3774 | req->open.filename = NULL; | |
3775 | return ret; | |
3776 | } | |
4022e7af | 3777 | req->open.nofile = rlimit(RLIMIT_NOFILE); |
8fef80bf | 3778 | req->flags |= REQ_F_NEED_CLEANUP; |
15b71abe | 3779 | return 0; |
03b1230c JA |
3780 | } |
3781 | ||
ec65fea5 PB |
3782 | static int io_openat_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
3783 | { | |
3784 | u64 flags, mode; | |
3785 | ||
4eb8dded JA |
3786 | if (unlikely(req->ctx->flags & (IORING_SETUP_IOPOLL|IORING_SETUP_SQPOLL))) |
3787 | return -EINVAL; | |
ec65fea5 PB |
3788 | mode = READ_ONCE(sqe->len); |
3789 | flags = READ_ONCE(sqe->open_flags); | |
3790 | req->open.how = build_open_how(flags, mode); | |
3791 | return __io_openat_prep(req, sqe); | |
3792 | } | |
3793 | ||
cebdb986 | 3794 | static int io_openat2_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
aa1fa28f | 3795 | { |
cebdb986 | 3796 | struct open_how __user *how; |
cebdb986 | 3797 | size_t len; |
0fa03c62 JA |
3798 | int ret; |
3799 | ||
4eb8dded JA |
3800 | if (unlikely(req->ctx->flags & (IORING_SETUP_IOPOLL|IORING_SETUP_SQPOLL))) |
3801 | return -EINVAL; | |
cebdb986 JA |
3802 | how = u64_to_user_ptr(READ_ONCE(sqe->addr2)); |
3803 | len = READ_ONCE(sqe->len); | |
cebdb986 JA |
3804 | if (len < OPEN_HOW_SIZE_VER0) |
3805 | return -EINVAL; | |
3529d8c2 | 3806 | |
cebdb986 JA |
3807 | ret = copy_struct_from_user(&req->open.how, sizeof(req->open.how), how, |
3808 | len); | |
3809 | if (ret) | |
3810 | return ret; | |
3529d8c2 | 3811 | |
ec65fea5 | 3812 | return __io_openat_prep(req, sqe); |
cebdb986 JA |
3813 | } |
3814 | ||
014db007 | 3815 | static int io_openat2(struct io_kiocb *req, bool force_nonblock) |
15b71abe JA |
3816 | { |
3817 | struct open_flags op; | |
15b71abe JA |
3818 | struct file *file; |
3819 | int ret; | |
3820 | ||
f86cd20c | 3821 | if (force_nonblock) |
15b71abe | 3822 | return -EAGAIN; |
15b71abe | 3823 | |
cebdb986 | 3824 | ret = build_open_flags(&req->open.how, &op); |
15b71abe JA |
3825 | if (ret) |
3826 | goto err; | |
3827 | ||
4022e7af | 3828 | ret = __get_unused_fd_flags(req->open.how.flags, req->open.nofile); |
15b71abe JA |
3829 | if (ret < 0) |
3830 | goto err; | |
3831 | ||
3832 | file = do_filp_open(req->open.dfd, req->open.filename, &op); | |
3833 | if (IS_ERR(file)) { | |
3834 | put_unused_fd(ret); | |
3835 | ret = PTR_ERR(file); | |
3836 | } else { | |
3837 | fsnotify_open(file); | |
3838 | fd_install(ret, file); | |
3839 | } | |
3840 | err: | |
3841 | putname(req->open.filename); | |
8fef80bf | 3842 | req->flags &= ~REQ_F_NEED_CLEANUP; |
15b71abe JA |
3843 | if (ret < 0) |
3844 | req_set_fail_links(req); | |
e1e16097 | 3845 | io_req_complete(req, ret); |
15b71abe JA |
3846 | return 0; |
3847 | } | |
3848 | ||
014db007 | 3849 | static int io_openat(struct io_kiocb *req, bool force_nonblock) |
cebdb986 | 3850 | { |
014db007 | 3851 | return io_openat2(req, force_nonblock); |
cebdb986 JA |
3852 | } |
3853 | ||
067524e9 JA |
3854 | static int io_remove_buffers_prep(struct io_kiocb *req, |
3855 | const struct io_uring_sqe *sqe) | |
3856 | { | |
3857 | struct io_provide_buf *p = &req->pbuf; | |
3858 | u64 tmp; | |
3859 | ||
3860 | if (sqe->ioprio || sqe->rw_flags || sqe->addr || sqe->len || sqe->off) | |
3861 | return -EINVAL; | |
3862 | ||
3863 | tmp = READ_ONCE(sqe->fd); | |
3864 | if (!tmp || tmp > USHRT_MAX) | |
3865 | return -EINVAL; | |
3866 | ||
3867 | memset(p, 0, sizeof(*p)); | |
3868 | p->nbufs = tmp; | |
3869 | p->bgid = READ_ONCE(sqe->buf_group); | |
3870 | return 0; | |
3871 | } | |
3872 | ||
3873 | static int __io_remove_buffers(struct io_ring_ctx *ctx, struct io_buffer *buf, | |
3874 | int bgid, unsigned nbufs) | |
3875 | { | |
3876 | unsigned i = 0; | |
3877 | ||
3878 | /* shouldn't happen */ | |
3879 | if (!nbufs) | |
3880 | return 0; | |
3881 | ||
3882 | /* the head kbuf is the list itself */ | |
3883 | while (!list_empty(&buf->list)) { | |
3884 | struct io_buffer *nxt; | |
3885 | ||
3886 | nxt = list_first_entry(&buf->list, struct io_buffer, list); | |
3887 | list_del(&nxt->list); | |
3888 | kfree(nxt); | |
3889 | if (++i == nbufs) | |
3890 | return i; | |
3891 | } | |
3892 | i++; | |
3893 | kfree(buf); | |
3894 | idr_remove(&ctx->io_buffer_idr, bgid); | |
3895 | ||
3896 | return i; | |
3897 | } | |
3898 | ||
229a7b63 JA |
3899 | static int io_remove_buffers(struct io_kiocb *req, bool force_nonblock, |
3900 | struct io_comp_state *cs) | |
067524e9 JA |
3901 | { |
3902 | struct io_provide_buf *p = &req->pbuf; | |
3903 | struct io_ring_ctx *ctx = req->ctx; | |
3904 | struct io_buffer *head; | |
3905 | int ret = 0; | |
3906 | ||
3907 | io_ring_submit_lock(ctx, !force_nonblock); | |
3908 | ||
3909 | lockdep_assert_held(&ctx->uring_lock); | |
3910 | ||
3911 | ret = -ENOENT; | |
3912 | head = idr_find(&ctx->io_buffer_idr, p->bgid); | |
3913 | if (head) | |
3914 | ret = __io_remove_buffers(ctx, head, p->bgid, p->nbufs); | |
3915 | ||
3916 | io_ring_submit_lock(ctx, !force_nonblock); | |
3917 | if (ret < 0) | |
3918 | req_set_fail_links(req); | |
229a7b63 | 3919 | __io_req_complete(req, ret, 0, cs); |
067524e9 JA |
3920 | return 0; |
3921 | } | |
3922 | ||
ddf0322d JA |
3923 | static int io_provide_buffers_prep(struct io_kiocb *req, |
3924 | const struct io_uring_sqe *sqe) | |
3925 | { | |
3926 | struct io_provide_buf *p = &req->pbuf; | |
3927 | u64 tmp; | |
3928 | ||
3929 | if (sqe->ioprio || sqe->rw_flags) | |
3930 | return -EINVAL; | |
3931 | ||
3932 | tmp = READ_ONCE(sqe->fd); | |
3933 | if (!tmp || tmp > USHRT_MAX) | |
3934 | return -E2BIG; | |
3935 | p->nbufs = tmp; | |
3936 | p->addr = READ_ONCE(sqe->addr); | |
3937 | p->len = READ_ONCE(sqe->len); | |
3938 | ||
efe68c1c | 3939 | if (!access_ok(u64_to_user_ptr(p->addr), (p->len * p->nbufs))) |
ddf0322d JA |
3940 | return -EFAULT; |
3941 | ||
3942 | p->bgid = READ_ONCE(sqe->buf_group); | |
3943 | tmp = READ_ONCE(sqe->off); | |
3944 | if (tmp > USHRT_MAX) | |
3945 | return -E2BIG; | |
3946 | p->bid = tmp; | |
3947 | return 0; | |
3948 | } | |
3949 | ||
3950 | static int io_add_buffers(struct io_provide_buf *pbuf, struct io_buffer **head) | |
3951 | { | |
3952 | struct io_buffer *buf; | |
3953 | u64 addr = pbuf->addr; | |
3954 | int i, bid = pbuf->bid; | |
3955 | ||
3956 | for (i = 0; i < pbuf->nbufs; i++) { | |
3957 | buf = kmalloc(sizeof(*buf), GFP_KERNEL); | |
3958 | if (!buf) | |
3959 | break; | |
3960 | ||
3961 | buf->addr = addr; | |
3962 | buf->len = pbuf->len; | |
3963 | buf->bid = bid; | |
3964 | addr += pbuf->len; | |
3965 | bid++; | |
3966 | if (!*head) { | |
3967 | INIT_LIST_HEAD(&buf->list); | |
3968 | *head = buf; | |
3969 | } else { | |
3970 | list_add_tail(&buf->list, &(*head)->list); | |
3971 | } | |
3972 | } | |
3973 | ||
3974 | return i ? i : -ENOMEM; | |
3975 | } | |
3976 | ||
229a7b63 JA |
3977 | static int io_provide_buffers(struct io_kiocb *req, bool force_nonblock, |
3978 | struct io_comp_state *cs) | |
ddf0322d JA |
3979 | { |
3980 | struct io_provide_buf *p = &req->pbuf; | |
3981 | struct io_ring_ctx *ctx = req->ctx; | |
3982 | struct io_buffer *head, *list; | |
3983 | int ret = 0; | |
3984 | ||
3985 | io_ring_submit_lock(ctx, !force_nonblock); | |
3986 | ||
3987 | lockdep_assert_held(&ctx->uring_lock); | |
3988 | ||
3989 | list = head = idr_find(&ctx->io_buffer_idr, p->bgid); | |
3990 | ||
3991 | ret = io_add_buffers(p, &head); | |
3992 | if (ret < 0) | |
3993 | goto out; | |
3994 | ||
3995 | if (!list) { | |
3996 | ret = idr_alloc(&ctx->io_buffer_idr, head, p->bgid, p->bgid + 1, | |
3997 | GFP_KERNEL); | |
3998 | if (ret < 0) { | |
067524e9 | 3999 | __io_remove_buffers(ctx, head, p->bgid, -1U); |
ddf0322d JA |
4000 | goto out; |
4001 | } | |
4002 | } | |
4003 | out: | |
4004 | io_ring_submit_unlock(ctx, !force_nonblock); | |
4005 | if (ret < 0) | |
4006 | req_set_fail_links(req); | |
229a7b63 | 4007 | __io_req_complete(req, ret, 0, cs); |
ddf0322d | 4008 | return 0; |
cebdb986 JA |
4009 | } |
4010 | ||
3e4827b0 JA |
4011 | static int io_epoll_ctl_prep(struct io_kiocb *req, |
4012 | const struct io_uring_sqe *sqe) | |
4013 | { | |
4014 | #if defined(CONFIG_EPOLL) | |
4015 | if (sqe->ioprio || sqe->buf_index) | |
4016 | return -EINVAL; | |
6ca56f84 | 4017 | if (unlikely(req->ctx->flags & (IORING_SETUP_IOPOLL | IORING_SETUP_SQPOLL))) |
3232dd02 | 4018 | return -EINVAL; |
3e4827b0 JA |
4019 | |
4020 | req->epoll.epfd = READ_ONCE(sqe->fd); | |
4021 | req->epoll.op = READ_ONCE(sqe->len); | |
4022 | req->epoll.fd = READ_ONCE(sqe->off); | |
4023 | ||
4024 | if (ep_op_has_event(req->epoll.op)) { | |
4025 | struct epoll_event __user *ev; | |
4026 | ||
4027 | ev = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
4028 | if (copy_from_user(&req->epoll.event, ev, sizeof(*ev))) | |
4029 | return -EFAULT; | |
4030 | } | |
4031 | ||
4032 | return 0; | |
4033 | #else | |
4034 | return -EOPNOTSUPP; | |
4035 | #endif | |
4036 | } | |
4037 | ||
229a7b63 JA |
4038 | static int io_epoll_ctl(struct io_kiocb *req, bool force_nonblock, |
4039 | struct io_comp_state *cs) | |
3e4827b0 JA |
4040 | { |
4041 | #if defined(CONFIG_EPOLL) | |
4042 | struct io_epoll *ie = &req->epoll; | |
4043 | int ret; | |
4044 | ||
4045 | ret = do_epoll_ctl(ie->epfd, ie->op, ie->fd, &ie->event, force_nonblock); | |
4046 | if (force_nonblock && ret == -EAGAIN) | |
4047 | return -EAGAIN; | |
4048 | ||
4049 | if (ret < 0) | |
4050 | req_set_fail_links(req); | |
229a7b63 | 4051 | __io_req_complete(req, ret, 0, cs); |
3e4827b0 JA |
4052 | return 0; |
4053 | #else | |
4054 | return -EOPNOTSUPP; | |
4055 | #endif | |
4056 | } | |
4057 | ||
c1ca757b JA |
4058 | static int io_madvise_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
4059 | { | |
4060 | #if defined(CONFIG_ADVISE_SYSCALLS) && defined(CONFIG_MMU) | |
4061 | if (sqe->ioprio || sqe->buf_index || sqe->off) | |
4062 | return -EINVAL; | |
3232dd02 PB |
4063 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4064 | return -EINVAL; | |
c1ca757b JA |
4065 | |
4066 | req->madvise.addr = READ_ONCE(sqe->addr); | |
4067 | req->madvise.len = READ_ONCE(sqe->len); | |
4068 | req->madvise.advice = READ_ONCE(sqe->fadvise_advice); | |
4069 | return 0; | |
4070 | #else | |
4071 | return -EOPNOTSUPP; | |
4072 | #endif | |
4073 | } | |
4074 | ||
014db007 | 4075 | static int io_madvise(struct io_kiocb *req, bool force_nonblock) |
c1ca757b JA |
4076 | { |
4077 | #if defined(CONFIG_ADVISE_SYSCALLS) && defined(CONFIG_MMU) | |
4078 | struct io_madvise *ma = &req->madvise; | |
4079 | int ret; | |
4080 | ||
4081 | if (force_nonblock) | |
4082 | return -EAGAIN; | |
4083 | ||
4084 | ret = do_madvise(ma->addr, ma->len, ma->advice); | |
4085 | if (ret < 0) | |
4086 | req_set_fail_links(req); | |
e1e16097 | 4087 | io_req_complete(req, ret); |
c1ca757b JA |
4088 | return 0; |
4089 | #else | |
4090 | return -EOPNOTSUPP; | |
4091 | #endif | |
4092 | } | |
4093 | ||
4840e418 JA |
4094 | static int io_fadvise_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
4095 | { | |
4096 | if (sqe->ioprio || sqe->buf_index || sqe->addr) | |
4097 | return -EINVAL; | |
3232dd02 PB |
4098 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4099 | return -EINVAL; | |
4840e418 JA |
4100 | |
4101 | req->fadvise.offset = READ_ONCE(sqe->off); | |
4102 | req->fadvise.len = READ_ONCE(sqe->len); | |
4103 | req->fadvise.advice = READ_ONCE(sqe->fadvise_advice); | |
4104 | return 0; | |
4105 | } | |
4106 | ||
014db007 | 4107 | static int io_fadvise(struct io_kiocb *req, bool force_nonblock) |
4840e418 JA |
4108 | { |
4109 | struct io_fadvise *fa = &req->fadvise; | |
4110 | int ret; | |
4111 | ||
3e69426d JA |
4112 | if (force_nonblock) { |
4113 | switch (fa->advice) { | |
4114 | case POSIX_FADV_NORMAL: | |
4115 | case POSIX_FADV_RANDOM: | |
4116 | case POSIX_FADV_SEQUENTIAL: | |
4117 | break; | |
4118 | default: | |
4119 | return -EAGAIN; | |
4120 | } | |
4121 | } | |
4840e418 JA |
4122 | |
4123 | ret = vfs_fadvise(req->file, fa->offset, fa->len, fa->advice); | |
4124 | if (ret < 0) | |
4125 | req_set_fail_links(req); | |
e1e16097 | 4126 | io_req_complete(req, ret); |
4840e418 JA |
4127 | return 0; |
4128 | } | |
4129 | ||
eddc7ef5 JA |
4130 | static int io_statx_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
4131 | { | |
6ca56f84 | 4132 | if (unlikely(req->ctx->flags & (IORING_SETUP_IOPOLL | IORING_SETUP_SQPOLL))) |
3232dd02 | 4133 | return -EINVAL; |
eddc7ef5 JA |
4134 | if (sqe->ioprio || sqe->buf_index) |
4135 | return -EINVAL; | |
9c280f90 | 4136 | if (req->flags & REQ_F_FIXED_FILE) |
cf3040ca | 4137 | return -EBADF; |
eddc7ef5 | 4138 | |
1d9e1288 BM |
4139 | req->statx.dfd = READ_ONCE(sqe->fd); |
4140 | req->statx.mask = READ_ONCE(sqe->len); | |
e62753e4 | 4141 | req->statx.filename = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
1d9e1288 BM |
4142 | req->statx.buffer = u64_to_user_ptr(READ_ONCE(sqe->addr2)); |
4143 | req->statx.flags = READ_ONCE(sqe->statx_flags); | |
eddc7ef5 JA |
4144 | |
4145 | return 0; | |
4146 | } | |
4147 | ||
014db007 | 4148 | static int io_statx(struct io_kiocb *req, bool force_nonblock) |
eddc7ef5 | 4149 | { |
1d9e1288 | 4150 | struct io_statx *ctx = &req->statx; |
eddc7ef5 JA |
4151 | int ret; |
4152 | ||
5b0bbee4 JA |
4153 | if (force_nonblock) { |
4154 | /* only need file table for an actual valid fd */ | |
4155 | if (ctx->dfd == -1 || ctx->dfd == AT_FDCWD) | |
4156 | req->flags |= REQ_F_NO_FILE_TABLE; | |
eddc7ef5 | 4157 | return -EAGAIN; |
5b0bbee4 | 4158 | } |
eddc7ef5 | 4159 | |
e62753e4 BM |
4160 | ret = do_statx(ctx->dfd, ctx->filename, ctx->flags, ctx->mask, |
4161 | ctx->buffer); | |
eddc7ef5 | 4162 | |
eddc7ef5 JA |
4163 | if (ret < 0) |
4164 | req_set_fail_links(req); | |
e1e16097 | 4165 | io_req_complete(req, ret); |
eddc7ef5 JA |
4166 | return 0; |
4167 | } | |
4168 | ||
b5dba59e JA |
4169 | static int io_close_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
4170 | { | |
4171 | /* | |
4172 | * If we queue this for async, it must not be cancellable. That would | |
7cdaf587 XW |
4173 | * leave the 'file' in an undeterminate state, and here need to modify |
4174 | * io_wq_work.flags, so initialize io_wq_work firstly. | |
b5dba59e | 4175 | */ |
7cdaf587 | 4176 | io_req_init_async(req); |
b5dba59e JA |
4177 | req->work.flags |= IO_WQ_WORK_NO_CANCEL; |
4178 | ||
3232dd02 PB |
4179 | if (unlikely(req->ctx->flags & (IORING_SETUP_IOPOLL|IORING_SETUP_SQPOLL))) |
4180 | return -EINVAL; | |
b5dba59e JA |
4181 | if (sqe->ioprio || sqe->off || sqe->addr || sqe->len || |
4182 | sqe->rw_flags || sqe->buf_index) | |
4183 | return -EINVAL; | |
9c280f90 | 4184 | if (req->flags & REQ_F_FIXED_FILE) |
cf3040ca | 4185 | return -EBADF; |
b5dba59e JA |
4186 | |
4187 | req->close.fd = READ_ONCE(sqe->fd); | |
0f212204 | 4188 | if ((req->file && req->file->f_op == &io_uring_fops)) |
fd2206e4 | 4189 | return -EBADF; |
b5dba59e | 4190 | |
3af73b28 | 4191 | req->close.put_file = NULL; |
b5dba59e | 4192 | return 0; |
b5dba59e JA |
4193 | } |
4194 | ||
229a7b63 JA |
4195 | static int io_close(struct io_kiocb *req, bool force_nonblock, |
4196 | struct io_comp_state *cs) | |
b5dba59e | 4197 | { |
3af73b28 | 4198 | struct io_close *close = &req->close; |
b5dba59e JA |
4199 | int ret; |
4200 | ||
3af73b28 PB |
4201 | /* might be already done during nonblock submission */ |
4202 | if (!close->put_file) { | |
4203 | ret = __close_fd_get_file(close->fd, &close->put_file); | |
4204 | if (ret < 0) | |
4205 | return (ret == -ENOENT) ? -EBADF : ret; | |
4206 | } | |
b5dba59e JA |
4207 | |
4208 | /* if the file has a flush method, be safe and punt to async */ | |
3af73b28 | 4209 | if (close->put_file->f_op->flush && force_nonblock) { |
24c74678 PB |
4210 | /* was never set, but play safe */ |
4211 | req->flags &= ~REQ_F_NOWAIT; | |
0bf0eefd | 4212 | /* avoid grabbing files - we don't need the files */ |
24c74678 | 4213 | req->flags |= REQ_F_NO_FILE_TABLE; |
0bf0eefd | 4214 | return -EAGAIN; |
a2100672 | 4215 | } |
b5dba59e | 4216 | |
3af73b28 | 4217 | /* No ->flush() or already async, safely close from here */ |
98447d65 | 4218 | ret = filp_close(close->put_file, req->work.identity->files); |
3af73b28 PB |
4219 | if (ret < 0) |
4220 | req_set_fail_links(req); | |
3af73b28 PB |
4221 | fput(close->put_file); |
4222 | close->put_file = NULL; | |
229a7b63 | 4223 | __io_req_complete(req, ret, 0, cs); |
1a417f4e | 4224 | return 0; |
b5dba59e JA |
4225 | } |
4226 | ||
3529d8c2 | 4227 | static int io_prep_sfr(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
5d17b4a4 JA |
4228 | { |
4229 | struct io_ring_ctx *ctx = req->ctx; | |
5d17b4a4 JA |
4230 | |
4231 | if (!req->file) | |
4232 | return -EBADF; | |
5d17b4a4 JA |
4233 | |
4234 | if (unlikely(ctx->flags & IORING_SETUP_IOPOLL)) | |
4235 | return -EINVAL; | |
4236 | if (unlikely(sqe->addr || sqe->ioprio || sqe->buf_index)) | |
4237 | return -EINVAL; | |
4238 | ||
8ed8d3c3 JA |
4239 | req->sync.off = READ_ONCE(sqe->off); |
4240 | req->sync.len = READ_ONCE(sqe->len); | |
4241 | req->sync.flags = READ_ONCE(sqe->sync_range_flags); | |
8ed8d3c3 JA |
4242 | return 0; |
4243 | } | |
4244 | ||
ac45abc0 | 4245 | static int io_sync_file_range(struct io_kiocb *req, bool force_nonblock) |
8ed8d3c3 | 4246 | { |
8ed8d3c3 JA |
4247 | int ret; |
4248 | ||
ac45abc0 PB |
4249 | /* sync_file_range always requires a blocking context */ |
4250 | if (force_nonblock) | |
4251 | return -EAGAIN; | |
4252 | ||
9adbd45d | 4253 | ret = sync_file_range(req->file, req->sync.off, req->sync.len, |
8ed8d3c3 JA |
4254 | req->sync.flags); |
4255 | if (ret < 0) | |
4256 | req_set_fail_links(req); | |
e1e16097 | 4257 | io_req_complete(req, ret); |
5d17b4a4 JA |
4258 | return 0; |
4259 | } | |
4260 | ||
469956e8 | 4261 | #if defined(CONFIG_NET) |
02d27d89 PB |
4262 | static int io_setup_async_msg(struct io_kiocb *req, |
4263 | struct io_async_msghdr *kmsg) | |
4264 | { | |
e8c2bc1f JA |
4265 | struct io_async_msghdr *async_msg = req->async_data; |
4266 | ||
4267 | if (async_msg) | |
02d27d89 | 4268 | return -EAGAIN; |
e8c2bc1f | 4269 | if (io_alloc_async_data(req)) { |
02d27d89 PB |
4270 | if (kmsg->iov != kmsg->fast_iov) |
4271 | kfree(kmsg->iov); | |
4272 | return -ENOMEM; | |
4273 | } | |
e8c2bc1f | 4274 | async_msg = req->async_data; |
02d27d89 | 4275 | req->flags |= REQ_F_NEED_CLEANUP; |
e8c2bc1f | 4276 | memcpy(async_msg, kmsg, sizeof(*kmsg)); |
02d27d89 PB |
4277 | return -EAGAIN; |
4278 | } | |
4279 | ||
2ae523ed PB |
4280 | static int io_sendmsg_copy_hdr(struct io_kiocb *req, |
4281 | struct io_async_msghdr *iomsg) | |
4282 | { | |
4283 | iomsg->iov = iomsg->fast_iov; | |
4284 | iomsg->msg.msg_name = &iomsg->addr; | |
4285 | return sendmsg_copy_msghdr(&iomsg->msg, req->sr_msg.umsg, | |
4286 | req->sr_msg.msg_flags, &iomsg->iov); | |
4287 | } | |
4288 | ||
3529d8c2 | 4289 | static int io_sendmsg_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
03b1230c | 4290 | { |
e8c2bc1f | 4291 | struct io_async_msghdr *async_msg = req->async_data; |
e47293fd | 4292 | struct io_sr_msg *sr = &req->sr_msg; |
99bc4c38 | 4293 | int ret; |
03b1230c | 4294 | |
d2b6f48b PB |
4295 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4296 | return -EINVAL; | |
4297 | ||
e47293fd | 4298 | sr->msg_flags = READ_ONCE(sqe->msg_flags); |
270a5940 | 4299 | sr->umsg = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
fddaface | 4300 | sr->len = READ_ONCE(sqe->len); |
3529d8c2 | 4301 | |
d8768362 JA |
4302 | #ifdef CONFIG_COMPAT |
4303 | if (req->ctx->compat) | |
4304 | sr->msg_flags |= MSG_CMSG_COMPAT; | |
4305 | #endif | |
4306 | ||
e8c2bc1f | 4307 | if (!async_msg || !io_op_defs[req->opcode].needs_async_data) |
3529d8c2 | 4308 | return 0; |
e8c2bc1f | 4309 | ret = io_sendmsg_copy_hdr(req, async_msg); |
99bc4c38 PB |
4310 | if (!ret) |
4311 | req->flags |= REQ_F_NEED_CLEANUP; | |
4312 | return ret; | |
03b1230c JA |
4313 | } |
4314 | ||
229a7b63 JA |
4315 | static int io_sendmsg(struct io_kiocb *req, bool force_nonblock, |
4316 | struct io_comp_state *cs) | |
aa1fa28f | 4317 | { |
6b754c8b | 4318 | struct io_async_msghdr iomsg, *kmsg; |
0fa03c62 | 4319 | struct socket *sock; |
7a7cacba | 4320 | unsigned flags; |
0fa03c62 JA |
4321 | int ret; |
4322 | ||
0fa03c62 | 4323 | sock = sock_from_file(req->file, &ret); |
7a7cacba PB |
4324 | if (unlikely(!sock)) |
4325 | return ret; | |
3529d8c2 | 4326 | |
e8c2bc1f JA |
4327 | if (req->async_data) { |
4328 | kmsg = req->async_data; | |
4329 | kmsg->msg.msg_name = &kmsg->addr; | |
7a7cacba PB |
4330 | /* if iov is set, it's allocated already */ |
4331 | if (!kmsg->iov) | |
4332 | kmsg->iov = kmsg->fast_iov; | |
4333 | kmsg->msg.msg_iter.iov = kmsg->iov; | |
4334 | } else { | |
4335 | ret = io_sendmsg_copy_hdr(req, &iomsg); | |
4336 | if (ret) | |
4337 | return ret; | |
4338 | kmsg = &iomsg; | |
0fa03c62 | 4339 | } |
0fa03c62 | 4340 | |
7a7cacba PB |
4341 | flags = req->sr_msg.msg_flags; |
4342 | if (flags & MSG_DONTWAIT) | |
4343 | req->flags |= REQ_F_NOWAIT; | |
4344 | else if (force_nonblock) | |
4345 | flags |= MSG_DONTWAIT; | |
e47293fd | 4346 | |
7a7cacba PB |
4347 | ret = __sys_sendmsg_sock(sock, &kmsg->msg, flags); |
4348 | if (force_nonblock && ret == -EAGAIN) | |
4349 | return io_setup_async_msg(req, kmsg); | |
4350 | if (ret == -ERESTARTSYS) | |
4351 | ret = -EINTR; | |
0fa03c62 | 4352 | |
6b754c8b | 4353 | if (kmsg->iov != kmsg->fast_iov) |
0b416c3e | 4354 | kfree(kmsg->iov); |
99bc4c38 | 4355 | req->flags &= ~REQ_F_NEED_CLEANUP; |
4e88d6e7 JA |
4356 | if (ret < 0) |
4357 | req_set_fail_links(req); | |
229a7b63 | 4358 | __io_req_complete(req, ret, 0, cs); |
5d17b4a4 | 4359 | return 0; |
03b1230c | 4360 | } |
aa1fa28f | 4361 | |
229a7b63 JA |
4362 | static int io_send(struct io_kiocb *req, bool force_nonblock, |
4363 | struct io_comp_state *cs) | |
fddaface | 4364 | { |
7a7cacba PB |
4365 | struct io_sr_msg *sr = &req->sr_msg; |
4366 | struct msghdr msg; | |
4367 | struct iovec iov; | |
fddaface | 4368 | struct socket *sock; |
7a7cacba | 4369 | unsigned flags; |
fddaface JA |
4370 | int ret; |
4371 | ||
fddaface | 4372 | sock = sock_from_file(req->file, &ret); |
7a7cacba PB |
4373 | if (unlikely(!sock)) |
4374 | return ret; | |
fddaface | 4375 | |
7a7cacba PB |
4376 | ret = import_single_range(WRITE, sr->buf, sr->len, &iov, &msg.msg_iter); |
4377 | if (unlikely(ret)) | |
14db8411 | 4378 | return ret; |
fddaface | 4379 | |
7a7cacba PB |
4380 | msg.msg_name = NULL; |
4381 | msg.msg_control = NULL; | |
4382 | msg.msg_controllen = 0; | |
4383 | msg.msg_namelen = 0; | |
fddaface | 4384 | |
7a7cacba PB |
4385 | flags = req->sr_msg.msg_flags; |
4386 | if (flags & MSG_DONTWAIT) | |
4387 | req->flags |= REQ_F_NOWAIT; | |
4388 | else if (force_nonblock) | |
4389 | flags |= MSG_DONTWAIT; | |
fddaface | 4390 | |
7a7cacba PB |
4391 | msg.msg_flags = flags; |
4392 | ret = sock_sendmsg(sock, &msg); | |
4393 | if (force_nonblock && ret == -EAGAIN) | |
4394 | return -EAGAIN; | |
4395 | if (ret == -ERESTARTSYS) | |
4396 | ret = -EINTR; | |
fddaface | 4397 | |
fddaface JA |
4398 | if (ret < 0) |
4399 | req_set_fail_links(req); | |
229a7b63 | 4400 | __io_req_complete(req, ret, 0, cs); |
fddaface | 4401 | return 0; |
fddaface JA |
4402 | } |
4403 | ||
1400e697 PB |
4404 | static int __io_recvmsg_copy_hdr(struct io_kiocb *req, |
4405 | struct io_async_msghdr *iomsg) | |
52de1fe1 JA |
4406 | { |
4407 | struct io_sr_msg *sr = &req->sr_msg; | |
4408 | struct iovec __user *uiov; | |
4409 | size_t iov_len; | |
4410 | int ret; | |
4411 | ||
1400e697 PB |
4412 | ret = __copy_msghdr_from_user(&iomsg->msg, sr->umsg, |
4413 | &iomsg->uaddr, &uiov, &iov_len); | |
52de1fe1 JA |
4414 | if (ret) |
4415 | return ret; | |
4416 | ||
4417 | if (req->flags & REQ_F_BUFFER_SELECT) { | |
4418 | if (iov_len > 1) | |
4419 | return -EINVAL; | |
1400e697 | 4420 | if (copy_from_user(iomsg->iov, uiov, sizeof(*uiov))) |
52de1fe1 | 4421 | return -EFAULT; |
1400e697 PB |
4422 | sr->len = iomsg->iov[0].iov_len; |
4423 | iov_iter_init(&iomsg->msg.msg_iter, READ, iomsg->iov, 1, | |
52de1fe1 | 4424 | sr->len); |
1400e697 | 4425 | iomsg->iov = NULL; |
52de1fe1 | 4426 | } else { |
89cd35c5 CH |
4427 | ret = __import_iovec(READ, uiov, iov_len, UIO_FASTIOV, |
4428 | &iomsg->iov, &iomsg->msg.msg_iter, | |
4429 | false); | |
52de1fe1 JA |
4430 | if (ret > 0) |
4431 | ret = 0; | |
4432 | } | |
4433 | ||
4434 | return ret; | |
4435 | } | |
4436 | ||
4437 | #ifdef CONFIG_COMPAT | |
4438 | static int __io_compat_recvmsg_copy_hdr(struct io_kiocb *req, | |
1400e697 | 4439 | struct io_async_msghdr *iomsg) |
52de1fe1 JA |
4440 | { |
4441 | struct compat_msghdr __user *msg_compat; | |
4442 | struct io_sr_msg *sr = &req->sr_msg; | |
4443 | struct compat_iovec __user *uiov; | |
4444 | compat_uptr_t ptr; | |
4445 | compat_size_t len; | |
4446 | int ret; | |
4447 | ||
270a5940 | 4448 | msg_compat = (struct compat_msghdr __user *) sr->umsg; |
1400e697 | 4449 | ret = __get_compat_msghdr(&iomsg->msg, msg_compat, &iomsg->uaddr, |
52de1fe1 JA |
4450 | &ptr, &len); |
4451 | if (ret) | |
4452 | return ret; | |
4453 | ||
4454 | uiov = compat_ptr(ptr); | |
4455 | if (req->flags & REQ_F_BUFFER_SELECT) { | |
4456 | compat_ssize_t clen; | |
4457 | ||
4458 | if (len > 1) | |
4459 | return -EINVAL; | |
4460 | if (!access_ok(uiov, sizeof(*uiov))) | |
4461 | return -EFAULT; | |
4462 | if (__get_user(clen, &uiov->iov_len)) | |
4463 | return -EFAULT; | |
4464 | if (clen < 0) | |
4465 | return -EINVAL; | |
1400e697 PB |
4466 | sr->len = iomsg->iov[0].iov_len; |
4467 | iomsg->iov = NULL; | |
52de1fe1 | 4468 | } else { |
89cd35c5 CH |
4469 | ret = __import_iovec(READ, (struct iovec __user *)uiov, len, |
4470 | UIO_FASTIOV, &iomsg->iov, | |
4471 | &iomsg->msg.msg_iter, true); | |
52de1fe1 JA |
4472 | if (ret < 0) |
4473 | return ret; | |
4474 | } | |
4475 | ||
4476 | return 0; | |
4477 | } | |
4478 | #endif | |
4479 | ||
1400e697 PB |
4480 | static int io_recvmsg_copy_hdr(struct io_kiocb *req, |
4481 | struct io_async_msghdr *iomsg) | |
52de1fe1 | 4482 | { |
1400e697 PB |
4483 | iomsg->msg.msg_name = &iomsg->addr; |
4484 | iomsg->iov = iomsg->fast_iov; | |
52de1fe1 JA |
4485 | |
4486 | #ifdef CONFIG_COMPAT | |
4487 | if (req->ctx->compat) | |
1400e697 | 4488 | return __io_compat_recvmsg_copy_hdr(req, iomsg); |
fddaface | 4489 | #endif |
52de1fe1 | 4490 | |
1400e697 | 4491 | return __io_recvmsg_copy_hdr(req, iomsg); |
52de1fe1 JA |
4492 | } |
4493 | ||
bcda7baa | 4494 | static struct io_buffer *io_recv_buffer_select(struct io_kiocb *req, |
7fbb1b54 | 4495 | bool needs_lock) |
bcda7baa JA |
4496 | { |
4497 | struct io_sr_msg *sr = &req->sr_msg; | |
4498 | struct io_buffer *kbuf; | |
4499 | ||
bcda7baa JA |
4500 | kbuf = io_buffer_select(req, &sr->len, sr->bgid, sr->kbuf, needs_lock); |
4501 | if (IS_ERR(kbuf)) | |
4502 | return kbuf; | |
4503 | ||
4504 | sr->kbuf = kbuf; | |
4505 | req->flags |= REQ_F_BUFFER_SELECTED; | |
bcda7baa | 4506 | return kbuf; |
fddaface JA |
4507 | } |
4508 | ||
7fbb1b54 PB |
4509 | static inline unsigned int io_put_recv_kbuf(struct io_kiocb *req) |
4510 | { | |
4511 | return io_put_kbuf(req, req->sr_msg.kbuf); | |
4512 | } | |
4513 | ||
3529d8c2 JA |
4514 | static int io_recvmsg_prep(struct io_kiocb *req, |
4515 | const struct io_uring_sqe *sqe) | |
aa1fa28f | 4516 | { |
e8c2bc1f | 4517 | struct io_async_msghdr *async_msg = req->async_data; |
e47293fd | 4518 | struct io_sr_msg *sr = &req->sr_msg; |
99bc4c38 | 4519 | int ret; |
3529d8c2 | 4520 | |
d2b6f48b PB |
4521 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4522 | return -EINVAL; | |
4523 | ||
3529d8c2 | 4524 | sr->msg_flags = READ_ONCE(sqe->msg_flags); |
270a5940 | 4525 | sr->umsg = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
0b7b21e4 | 4526 | sr->len = READ_ONCE(sqe->len); |
bcda7baa | 4527 | sr->bgid = READ_ONCE(sqe->buf_group); |
06b76d44 | 4528 | |
d8768362 JA |
4529 | #ifdef CONFIG_COMPAT |
4530 | if (req->ctx->compat) | |
4531 | sr->msg_flags |= MSG_CMSG_COMPAT; | |
4532 | #endif | |
4533 | ||
e8c2bc1f | 4534 | if (!async_msg || !io_op_defs[req->opcode].needs_async_data) |
5f798bea | 4535 | return 0; |
e8c2bc1f | 4536 | ret = io_recvmsg_copy_hdr(req, async_msg); |
99bc4c38 PB |
4537 | if (!ret) |
4538 | req->flags |= REQ_F_NEED_CLEANUP; | |
4539 | return ret; | |
aa1fa28f JA |
4540 | } |
4541 | ||
229a7b63 JA |
4542 | static int io_recvmsg(struct io_kiocb *req, bool force_nonblock, |
4543 | struct io_comp_state *cs) | |
aa1fa28f | 4544 | { |
6b754c8b | 4545 | struct io_async_msghdr iomsg, *kmsg; |
03b1230c | 4546 | struct socket *sock; |
7fbb1b54 | 4547 | struct io_buffer *kbuf; |
7a7cacba | 4548 | unsigned flags; |
52de1fe1 | 4549 | int ret, cflags = 0; |
03b1230c | 4550 | |
03b1230c | 4551 | sock = sock_from_file(req->file, &ret); |
7a7cacba PB |
4552 | if (unlikely(!sock)) |
4553 | return ret; | |
3529d8c2 | 4554 | |
e8c2bc1f JA |
4555 | if (req->async_data) { |
4556 | kmsg = req->async_data; | |
4557 | kmsg->msg.msg_name = &kmsg->addr; | |
7a7cacba PB |
4558 | /* if iov is set, it's allocated already */ |
4559 | if (!kmsg->iov) | |
4560 | kmsg->iov = kmsg->fast_iov; | |
4561 | kmsg->msg.msg_iter.iov = kmsg->iov; | |
4562 | } else { | |
4563 | ret = io_recvmsg_copy_hdr(req, &iomsg); | |
4564 | if (ret) | |
681fda8d | 4565 | return ret; |
7a7cacba PB |
4566 | kmsg = &iomsg; |
4567 | } | |
03b1230c | 4568 | |
bc02ef33 | 4569 | if (req->flags & REQ_F_BUFFER_SELECT) { |
7fbb1b54 | 4570 | kbuf = io_recv_buffer_select(req, !force_nonblock); |
bc02ef33 | 4571 | if (IS_ERR(kbuf)) |
52de1fe1 | 4572 | return PTR_ERR(kbuf); |
7a7cacba PB |
4573 | kmsg->fast_iov[0].iov_base = u64_to_user_ptr(kbuf->addr); |
4574 | iov_iter_init(&kmsg->msg.msg_iter, READ, kmsg->iov, | |
4575 | 1, req->sr_msg.len); | |
4576 | } | |
52de1fe1 | 4577 | |
7a7cacba PB |
4578 | flags = req->sr_msg.msg_flags; |
4579 | if (flags & MSG_DONTWAIT) | |
4580 | req->flags |= REQ_F_NOWAIT; | |
4581 | else if (force_nonblock) | |
4582 | flags |= MSG_DONTWAIT; | |
e47293fd | 4583 | |
7a7cacba PB |
4584 | ret = __sys_recvmsg_sock(sock, &kmsg->msg, req->sr_msg.umsg, |
4585 | kmsg->uaddr, flags); | |
0e1b6fe3 PB |
4586 | if (force_nonblock && ret == -EAGAIN) |
4587 | return io_setup_async_msg(req, kmsg); | |
7a7cacba PB |
4588 | if (ret == -ERESTARTSYS) |
4589 | ret = -EINTR; | |
03b1230c | 4590 | |
7fbb1b54 PB |
4591 | if (req->flags & REQ_F_BUFFER_SELECTED) |
4592 | cflags = io_put_recv_kbuf(req); | |
6b754c8b | 4593 | if (kmsg->iov != kmsg->fast_iov) |
0b416c3e | 4594 | kfree(kmsg->iov); |
99bc4c38 | 4595 | req->flags &= ~REQ_F_NEED_CLEANUP; |
4e88d6e7 JA |
4596 | if (ret < 0) |
4597 | req_set_fail_links(req); | |
229a7b63 | 4598 | __io_req_complete(req, ret, cflags, cs); |
03b1230c | 4599 | return 0; |
0fa03c62 | 4600 | } |
5d17b4a4 | 4601 | |
229a7b63 JA |
4602 | static int io_recv(struct io_kiocb *req, bool force_nonblock, |
4603 | struct io_comp_state *cs) | |
fddaface | 4604 | { |
6b754c8b | 4605 | struct io_buffer *kbuf; |
7a7cacba PB |
4606 | struct io_sr_msg *sr = &req->sr_msg; |
4607 | struct msghdr msg; | |
4608 | void __user *buf = sr->buf; | |
fddaface | 4609 | struct socket *sock; |
7a7cacba PB |
4610 | struct iovec iov; |
4611 | unsigned flags; | |
bcda7baa | 4612 | int ret, cflags = 0; |
fddaface | 4613 | |
fddaface | 4614 | sock = sock_from_file(req->file, &ret); |
7a7cacba PB |
4615 | if (unlikely(!sock)) |
4616 | return ret; | |
fddaface | 4617 | |
bc02ef33 | 4618 | if (req->flags & REQ_F_BUFFER_SELECT) { |
7fbb1b54 | 4619 | kbuf = io_recv_buffer_select(req, !force_nonblock); |
bcda7baa JA |
4620 | if (IS_ERR(kbuf)) |
4621 | return PTR_ERR(kbuf); | |
7a7cacba | 4622 | buf = u64_to_user_ptr(kbuf->addr); |
bc02ef33 | 4623 | } |
bcda7baa | 4624 | |
7a7cacba | 4625 | ret = import_single_range(READ, buf, sr->len, &iov, &msg.msg_iter); |
14c32eee PB |
4626 | if (unlikely(ret)) |
4627 | goto out_free; | |
fddaface | 4628 | |
7a7cacba PB |
4629 | msg.msg_name = NULL; |
4630 | msg.msg_control = NULL; | |
4631 | msg.msg_controllen = 0; | |
4632 | msg.msg_namelen = 0; | |
4633 | msg.msg_iocb = NULL; | |
4634 | msg.msg_flags = 0; | |
fddaface | 4635 | |
7a7cacba PB |
4636 | flags = req->sr_msg.msg_flags; |
4637 | if (flags & MSG_DONTWAIT) | |
4638 | req->flags |= REQ_F_NOWAIT; | |
4639 | else if (force_nonblock) | |
4640 | flags |= MSG_DONTWAIT; | |
4641 | ||
4642 | ret = sock_recvmsg(sock, &msg, flags); | |
4643 | if (force_nonblock && ret == -EAGAIN) | |
4644 | return -EAGAIN; | |
4645 | if (ret == -ERESTARTSYS) | |
4646 | ret = -EINTR; | |
14c32eee | 4647 | out_free: |
7fbb1b54 PB |
4648 | if (req->flags & REQ_F_BUFFER_SELECTED) |
4649 | cflags = io_put_recv_kbuf(req); | |
fddaface JA |
4650 | if (ret < 0) |
4651 | req_set_fail_links(req); | |
229a7b63 | 4652 | __io_req_complete(req, ret, cflags, cs); |
fddaface | 4653 | return 0; |
fddaface JA |
4654 | } |
4655 | ||
3529d8c2 | 4656 | static int io_accept_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
17f2fe35 | 4657 | { |
8ed8d3c3 JA |
4658 | struct io_accept *accept = &req->accept; |
4659 | ||
17f2fe35 JA |
4660 | if (unlikely(req->ctx->flags & (IORING_SETUP_IOPOLL|IORING_SETUP_SQPOLL))) |
4661 | return -EINVAL; | |
8042d6ce | 4662 | if (sqe->ioprio || sqe->len || sqe->buf_index) |
17f2fe35 JA |
4663 | return -EINVAL; |
4664 | ||
d55e5f5b JA |
4665 | accept->addr = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
4666 | accept->addr_len = u64_to_user_ptr(READ_ONCE(sqe->addr2)); | |
8ed8d3c3 | 4667 | accept->flags = READ_ONCE(sqe->accept_flags); |
09952e3e | 4668 | accept->nofile = rlimit(RLIMIT_NOFILE); |
8ed8d3c3 | 4669 | return 0; |
8ed8d3c3 | 4670 | } |
17f2fe35 | 4671 | |
229a7b63 JA |
4672 | static int io_accept(struct io_kiocb *req, bool force_nonblock, |
4673 | struct io_comp_state *cs) | |
8ed8d3c3 JA |
4674 | { |
4675 | struct io_accept *accept = &req->accept; | |
ac45abc0 | 4676 | unsigned int file_flags = force_nonblock ? O_NONBLOCK : 0; |
8ed8d3c3 JA |
4677 | int ret; |
4678 | ||
e697deed JX |
4679 | if (req->file->f_flags & O_NONBLOCK) |
4680 | req->flags |= REQ_F_NOWAIT; | |
4681 | ||
8ed8d3c3 | 4682 | ret = __sys_accept4_file(req->file, file_flags, accept->addr, |
09952e3e JA |
4683 | accept->addr_len, accept->flags, |
4684 | accept->nofile); | |
8ed8d3c3 | 4685 | if (ret == -EAGAIN && force_nonblock) |
17f2fe35 | 4686 | return -EAGAIN; |
ac45abc0 PB |
4687 | if (ret < 0) { |
4688 | if (ret == -ERESTARTSYS) | |
4689 | ret = -EINTR; | |
4e88d6e7 | 4690 | req_set_fail_links(req); |
ac45abc0 | 4691 | } |
229a7b63 | 4692 | __io_req_complete(req, ret, 0, cs); |
17f2fe35 | 4693 | return 0; |
8ed8d3c3 JA |
4694 | } |
4695 | ||
3529d8c2 | 4696 | static int io_connect_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
f499a021 | 4697 | { |
3529d8c2 | 4698 | struct io_connect *conn = &req->connect; |
e8c2bc1f | 4699 | struct io_async_connect *io = req->async_data; |
f499a021 | 4700 | |
3fbb51c1 JA |
4701 | if (unlikely(req->ctx->flags & (IORING_SETUP_IOPOLL|IORING_SETUP_SQPOLL))) |
4702 | return -EINVAL; | |
4703 | if (sqe->ioprio || sqe->len || sqe->buf_index || sqe->rw_flags) | |
4704 | return -EINVAL; | |
4705 | ||
3529d8c2 JA |
4706 | conn->addr = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
4707 | conn->addr_len = READ_ONCE(sqe->addr2); | |
4708 | ||
4709 | if (!io) | |
4710 | return 0; | |
4711 | ||
4712 | return move_addr_to_kernel(conn->addr, conn->addr_len, | |
e8c2bc1f | 4713 | &io->address); |
f499a021 JA |
4714 | } |
4715 | ||
229a7b63 JA |
4716 | static int io_connect(struct io_kiocb *req, bool force_nonblock, |
4717 | struct io_comp_state *cs) | |
f8e85cf2 | 4718 | { |
e8c2bc1f | 4719 | struct io_async_connect __io, *io; |
f8e85cf2 | 4720 | unsigned file_flags; |
3fbb51c1 | 4721 | int ret; |
f8e85cf2 | 4722 | |
e8c2bc1f JA |
4723 | if (req->async_data) { |
4724 | io = req->async_data; | |
f499a021 | 4725 | } else { |
3529d8c2 JA |
4726 | ret = move_addr_to_kernel(req->connect.addr, |
4727 | req->connect.addr_len, | |
e8c2bc1f | 4728 | &__io.address); |
f499a021 JA |
4729 | if (ret) |
4730 | goto out; | |
4731 | io = &__io; | |
4732 | } | |
4733 | ||
3fbb51c1 JA |
4734 | file_flags = force_nonblock ? O_NONBLOCK : 0; |
4735 | ||
e8c2bc1f | 4736 | ret = __sys_connect_file(req->file, &io->address, |
3fbb51c1 | 4737 | req->connect.addr_len, file_flags); |
87f80d62 | 4738 | if ((ret == -EAGAIN || ret == -EINPROGRESS) && force_nonblock) { |
e8c2bc1f | 4739 | if (req->async_data) |
b7bb4f7d | 4740 | return -EAGAIN; |
e8c2bc1f | 4741 | if (io_alloc_async_data(req)) { |
f499a021 JA |
4742 | ret = -ENOMEM; |
4743 | goto out; | |
4744 | } | |
e8c2bc1f JA |
4745 | io = req->async_data; |
4746 | memcpy(req->async_data, &__io, sizeof(__io)); | |
f8e85cf2 | 4747 | return -EAGAIN; |
f499a021 | 4748 | } |
f8e85cf2 JA |
4749 | if (ret == -ERESTARTSYS) |
4750 | ret = -EINTR; | |
f499a021 | 4751 | out: |
4e88d6e7 JA |
4752 | if (ret < 0) |
4753 | req_set_fail_links(req); | |
229a7b63 | 4754 | __io_req_complete(req, ret, 0, cs); |
f8e85cf2 | 4755 | return 0; |
469956e8 Y |
4756 | } |
4757 | #else /* !CONFIG_NET */ | |
4758 | static int io_sendmsg_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) | |
4759 | { | |
f8e85cf2 | 4760 | return -EOPNOTSUPP; |
f8e85cf2 JA |
4761 | } |
4762 | ||
1e16c2f9 RD |
4763 | static int io_sendmsg(struct io_kiocb *req, bool force_nonblock, |
4764 | struct io_comp_state *cs) | |
469956e8 Y |
4765 | { |
4766 | return -EOPNOTSUPP; | |
4767 | } | |
4768 | ||
1e16c2f9 RD |
4769 | static int io_send(struct io_kiocb *req, bool force_nonblock, |
4770 | struct io_comp_state *cs) | |
469956e8 Y |
4771 | { |
4772 | return -EOPNOTSUPP; | |
4773 | } | |
4774 | ||
4775 | static int io_recvmsg_prep(struct io_kiocb *req, | |
4776 | const struct io_uring_sqe *sqe) | |
4777 | { | |
4778 | return -EOPNOTSUPP; | |
4779 | } | |
4780 | ||
1e16c2f9 RD |
4781 | static int io_recvmsg(struct io_kiocb *req, bool force_nonblock, |
4782 | struct io_comp_state *cs) | |
469956e8 Y |
4783 | { |
4784 | return -EOPNOTSUPP; | |
4785 | } | |
4786 | ||
1e16c2f9 RD |
4787 | static int io_recv(struct io_kiocb *req, bool force_nonblock, |
4788 | struct io_comp_state *cs) | |
469956e8 Y |
4789 | { |
4790 | return -EOPNOTSUPP; | |
4791 | } | |
4792 | ||
4793 | static int io_accept_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) | |
4794 | { | |
4795 | return -EOPNOTSUPP; | |
4796 | } | |
4797 | ||
1e16c2f9 RD |
4798 | static int io_accept(struct io_kiocb *req, bool force_nonblock, |
4799 | struct io_comp_state *cs) | |
469956e8 Y |
4800 | { |
4801 | return -EOPNOTSUPP; | |
4802 | } | |
ce593a6c | 4803 | |
469956e8 Y |
4804 | static int io_connect_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
4805 | { | |
4806 | return -EOPNOTSUPP; | |
4807 | } | |
4808 | ||
1e16c2f9 RD |
4809 | static int io_connect(struct io_kiocb *req, bool force_nonblock, |
4810 | struct io_comp_state *cs) | |
469956e8 | 4811 | { |
f8e85cf2 | 4812 | return -EOPNOTSUPP; |
ce593a6c | 4813 | } |
469956e8 | 4814 | #endif /* CONFIG_NET */ |
f8e85cf2 | 4815 | |
d7718a9d JA |
4816 | struct io_poll_table { |
4817 | struct poll_table_struct pt; | |
4818 | struct io_kiocb *req; | |
4819 | int error; | |
4820 | }; | |
ce593a6c | 4821 | |
d7718a9d JA |
4822 | static int __io_async_wake(struct io_kiocb *req, struct io_poll_iocb *poll, |
4823 | __poll_t mask, task_work_func_t func) | |
4824 | { | |
fd7d6de2 | 4825 | bool twa_signal_ok; |
aa96bf8a | 4826 | int ret; |
d7718a9d JA |
4827 | |
4828 | /* for instances that support it check for an event match first: */ | |
4829 | if (mask && !(mask & poll->events)) | |
4830 | return 0; | |
4831 | ||
4832 | trace_io_uring_task_add(req->ctx, req->opcode, req->user_data, mask); | |
4833 | ||
4834 | list_del_init(&poll->wait.entry); | |
4835 | ||
d7718a9d JA |
4836 | req->result = mask; |
4837 | init_task_work(&req->task_work, func); | |
6d816e08 JA |
4838 | percpu_ref_get(&req->ctx->refs); |
4839 | ||
fd7d6de2 JA |
4840 | /* |
4841 | * If we using the signalfd wait_queue_head for this wakeup, then | |
4842 | * it's not safe to use TWA_SIGNAL as we could be recursing on the | |
4843 | * tsk->sighand->siglock on doing the wakeup. Should not be needed | |
4844 | * either, as the normal wakeup will suffice. | |
4845 | */ | |
4846 | twa_signal_ok = (poll->head != &req->task->sighand->signalfd_wqh); | |
4847 | ||
d7718a9d | 4848 | /* |
e3aabf95 JA |
4849 | * If this fails, then the task is exiting. When a task exits, the |
4850 | * work gets canceled, so just cancel this request as well instead | |
4851 | * of executing it. We can't safely execute it anyway, as we may not | |
4852 | * have the needed state needed for it anyway. | |
d7718a9d | 4853 | */ |
87c4311f | 4854 | ret = io_req_task_work_add(req, twa_signal_ok); |
aa96bf8a | 4855 | if (unlikely(ret)) { |
c2c4c83c JA |
4856 | struct task_struct *tsk; |
4857 | ||
e3aabf95 | 4858 | WRITE_ONCE(poll->canceled, true); |
aa96bf8a | 4859 | tsk = io_wq_get_task(req->ctx->io_wq); |
ce593a6c JA |
4860 | task_work_add(tsk, &req->task_work, 0); |
4861 | wake_up_process(tsk); | |
aa96bf8a | 4862 | } |
d7718a9d JA |
4863 | return 1; |
4864 | } | |
4865 | ||
74ce6ce4 JA |
4866 | static bool io_poll_rewait(struct io_kiocb *req, struct io_poll_iocb *poll) |
4867 | __acquires(&req->ctx->completion_lock) | |
4868 | { | |
4869 | struct io_ring_ctx *ctx = req->ctx; | |
4870 | ||
4871 | if (!req->result && !READ_ONCE(poll->canceled)) { | |
4872 | struct poll_table_struct pt = { ._key = poll->events }; | |
4873 | ||
4874 | req->result = vfs_poll(req->file, &pt) & poll->events; | |
4875 | } | |
4876 | ||
4877 | spin_lock_irq(&ctx->completion_lock); | |
4878 | if (!req->result && !READ_ONCE(poll->canceled)) { | |
4879 | add_wait_queue(poll->head, &poll->wait); | |
4880 | return true; | |
4881 | } | |
4882 | ||
4883 | return false; | |
4884 | } | |
4885 | ||
d4e7cd36 | 4886 | static struct io_poll_iocb *io_poll_get_double(struct io_kiocb *req) |
18bceab1 | 4887 | { |
e8c2bc1f | 4888 | /* pure poll stashes this in ->async_data, poll driven retry elsewhere */ |
d4e7cd36 | 4889 | if (req->opcode == IORING_OP_POLL_ADD) |
e8c2bc1f | 4890 | return req->async_data; |
d4e7cd36 JA |
4891 | return req->apoll->double_poll; |
4892 | } | |
4893 | ||
4894 | static struct io_poll_iocb *io_poll_get_single(struct io_kiocb *req) | |
4895 | { | |
4896 | if (req->opcode == IORING_OP_POLL_ADD) | |
4897 | return &req->poll; | |
4898 | return &req->apoll->poll; | |
4899 | } | |
4900 | ||
4901 | static void io_poll_remove_double(struct io_kiocb *req) | |
4902 | { | |
4903 | struct io_poll_iocb *poll = io_poll_get_double(req); | |
18bceab1 JA |
4904 | |
4905 | lockdep_assert_held(&req->ctx->completion_lock); | |
4906 | ||
4907 | if (poll && poll->head) { | |
4908 | struct wait_queue_head *head = poll->head; | |
4909 | ||
4910 | spin_lock(&head->lock); | |
4911 | list_del_init(&poll->wait.entry); | |
4912 | if (poll->wait.private) | |
4913 | refcount_dec(&req->refs); | |
4914 | poll->head = NULL; | |
4915 | spin_unlock(&head->lock); | |
4916 | } | |
4917 | } | |
4918 | ||
4919 | static void io_poll_complete(struct io_kiocb *req, __poll_t mask, int error) | |
4920 | { | |
4921 | struct io_ring_ctx *ctx = req->ctx; | |
4922 | ||
d4e7cd36 | 4923 | io_poll_remove_double(req); |
18bceab1 JA |
4924 | req->poll.done = true; |
4925 | io_cqring_fill_event(req, error ? error : mangle_poll(mask)); | |
4926 | io_commit_cqring(ctx); | |
4927 | } | |
4928 | ||
4929 | static void io_poll_task_handler(struct io_kiocb *req, struct io_kiocb **nxt) | |
4930 | { | |
4931 | struct io_ring_ctx *ctx = req->ctx; | |
4932 | ||
4933 | if (io_poll_rewait(req, &req->poll)) { | |
4934 | spin_unlock_irq(&ctx->completion_lock); | |
4935 | return; | |
4936 | } | |
4937 | ||
4938 | hash_del(&req->hash_node); | |
4939 | io_poll_complete(req, req->result, 0); | |
18bceab1 JA |
4940 | spin_unlock_irq(&ctx->completion_lock); |
4941 | ||
6a0af224 | 4942 | *nxt = io_put_req_find_next(req); |
18bceab1 JA |
4943 | io_cqring_ev_posted(ctx); |
4944 | } | |
4945 | ||
4946 | static void io_poll_task_func(struct callback_head *cb) | |
4947 | { | |
4948 | struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work); | |
6d816e08 | 4949 | struct io_ring_ctx *ctx = req->ctx; |
18bceab1 JA |
4950 | struct io_kiocb *nxt = NULL; |
4951 | ||
4952 | io_poll_task_handler(req, &nxt); | |
ea1164e5 PB |
4953 | if (nxt) |
4954 | __io_req_task_submit(nxt); | |
6d816e08 | 4955 | percpu_ref_put(&ctx->refs); |
18bceab1 JA |
4956 | } |
4957 | ||
4958 | static int io_poll_double_wake(struct wait_queue_entry *wait, unsigned mode, | |
4959 | int sync, void *key) | |
4960 | { | |
4961 | struct io_kiocb *req = wait->private; | |
d4e7cd36 | 4962 | struct io_poll_iocb *poll = io_poll_get_single(req); |
18bceab1 JA |
4963 | __poll_t mask = key_to_poll(key); |
4964 | ||
4965 | /* for instances that support it check for an event match first: */ | |
4966 | if (mask && !(mask & poll->events)) | |
4967 | return 0; | |
4968 | ||
8706e04e JA |
4969 | list_del_init(&wait->entry); |
4970 | ||
807abcb0 | 4971 | if (poll && poll->head) { |
18bceab1 JA |
4972 | bool done; |
4973 | ||
807abcb0 JA |
4974 | spin_lock(&poll->head->lock); |
4975 | done = list_empty(&poll->wait.entry); | |
18bceab1 | 4976 | if (!done) |
807abcb0 | 4977 | list_del_init(&poll->wait.entry); |
d4e7cd36 JA |
4978 | /* make sure double remove sees this as being gone */ |
4979 | wait->private = NULL; | |
807abcb0 | 4980 | spin_unlock(&poll->head->lock); |
18bceab1 JA |
4981 | if (!done) |
4982 | __io_async_wake(req, poll, mask, io_poll_task_func); | |
4983 | } | |
4984 | refcount_dec(&req->refs); | |
4985 | return 1; | |
4986 | } | |
4987 | ||
4988 | static void io_init_poll_iocb(struct io_poll_iocb *poll, __poll_t events, | |
4989 | wait_queue_func_t wake_func) | |
4990 | { | |
4991 | poll->head = NULL; | |
4992 | poll->done = false; | |
4993 | poll->canceled = false; | |
4994 | poll->events = events; | |
4995 | INIT_LIST_HEAD(&poll->wait.entry); | |
4996 | init_waitqueue_func_entry(&poll->wait, wake_func); | |
4997 | } | |
4998 | ||
4999 | static void __io_queue_proc(struct io_poll_iocb *poll, struct io_poll_table *pt, | |
807abcb0 JA |
5000 | struct wait_queue_head *head, |
5001 | struct io_poll_iocb **poll_ptr) | |
18bceab1 JA |
5002 | { |
5003 | struct io_kiocb *req = pt->req; | |
5004 | ||
5005 | /* | |
5006 | * If poll->head is already set, it's because the file being polled | |
5007 | * uses multiple waitqueues for poll handling (eg one for read, one | |
5008 | * for write). Setup a separate io_poll_iocb if this happens. | |
5009 | */ | |
5010 | if (unlikely(poll->head)) { | |
5011 | /* already have a 2nd entry, fail a third attempt */ | |
807abcb0 | 5012 | if (*poll_ptr) { |
18bceab1 JA |
5013 | pt->error = -EINVAL; |
5014 | return; | |
5015 | } | |
5016 | poll = kmalloc(sizeof(*poll), GFP_ATOMIC); | |
5017 | if (!poll) { | |
5018 | pt->error = -ENOMEM; | |
5019 | return; | |
5020 | } | |
5021 | io_init_poll_iocb(poll, req->poll.events, io_poll_double_wake); | |
5022 | refcount_inc(&req->refs); | |
5023 | poll->wait.private = req; | |
807abcb0 | 5024 | *poll_ptr = poll; |
18bceab1 JA |
5025 | } |
5026 | ||
5027 | pt->error = 0; | |
5028 | poll->head = head; | |
a31eb4a2 JX |
5029 | |
5030 | if (poll->events & EPOLLEXCLUSIVE) | |
5031 | add_wait_queue_exclusive(head, &poll->wait); | |
5032 | else | |
5033 | add_wait_queue(head, &poll->wait); | |
18bceab1 JA |
5034 | } |
5035 | ||
5036 | static void io_async_queue_proc(struct file *file, struct wait_queue_head *head, | |
5037 | struct poll_table_struct *p) | |
5038 | { | |
5039 | struct io_poll_table *pt = container_of(p, struct io_poll_table, pt); | |
807abcb0 | 5040 | struct async_poll *apoll = pt->req->apoll; |
18bceab1 | 5041 | |
807abcb0 | 5042 | __io_queue_proc(&apoll->poll, pt, head, &apoll->double_poll); |
18bceab1 JA |
5043 | } |
5044 | ||
d7718a9d JA |
5045 | static void io_async_task_func(struct callback_head *cb) |
5046 | { | |
5047 | struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work); | |
5048 | struct async_poll *apoll = req->apoll; | |
5049 | struct io_ring_ctx *ctx = req->ctx; | |
5050 | ||
5051 | trace_io_uring_task_run(req->ctx, req->opcode, req->user_data); | |
5052 | ||
74ce6ce4 | 5053 | if (io_poll_rewait(req, &apoll->poll)) { |
d7718a9d | 5054 | spin_unlock_irq(&ctx->completion_lock); |
6d816e08 | 5055 | percpu_ref_put(&ctx->refs); |
74ce6ce4 | 5056 | return; |
d7718a9d JA |
5057 | } |
5058 | ||
31067255 | 5059 | /* If req is still hashed, it cannot have been canceled. Don't check. */ |
0be0b0e3 | 5060 | if (hash_hashed(&req->hash_node)) |
74ce6ce4 | 5061 | hash_del(&req->hash_node); |
2bae047e | 5062 | |
d4e7cd36 | 5063 | io_poll_remove_double(req); |
74ce6ce4 JA |
5064 | spin_unlock_irq(&ctx->completion_lock); |
5065 | ||
0be0b0e3 PB |
5066 | if (!READ_ONCE(apoll->poll.canceled)) |
5067 | __io_req_task_submit(req); | |
5068 | else | |
5069 | __io_req_task_cancel(req, -ECANCELED); | |
aa340845 | 5070 | |
6d816e08 | 5071 | percpu_ref_put(&ctx->refs); |
807abcb0 | 5072 | kfree(apoll->double_poll); |
31067255 | 5073 | kfree(apoll); |
d7718a9d JA |
5074 | } |
5075 | ||
5076 | static int io_async_wake(struct wait_queue_entry *wait, unsigned mode, int sync, | |
5077 | void *key) | |
5078 | { | |
5079 | struct io_kiocb *req = wait->private; | |
5080 | struct io_poll_iocb *poll = &req->apoll->poll; | |
5081 | ||
5082 | trace_io_uring_poll_wake(req->ctx, req->opcode, req->user_data, | |
5083 | key_to_poll(key)); | |
5084 | ||
5085 | return __io_async_wake(req, poll, key_to_poll(key), io_async_task_func); | |
5086 | } | |
5087 | ||
5088 | static void io_poll_req_insert(struct io_kiocb *req) | |
5089 | { | |
5090 | struct io_ring_ctx *ctx = req->ctx; | |
5091 | struct hlist_head *list; | |
5092 | ||
5093 | list = &ctx->cancel_hash[hash_long(req->user_data, ctx->cancel_hash_bits)]; | |
5094 | hlist_add_head(&req->hash_node, list); | |
5095 | } | |
5096 | ||
5097 | static __poll_t __io_arm_poll_handler(struct io_kiocb *req, | |
5098 | struct io_poll_iocb *poll, | |
5099 | struct io_poll_table *ipt, __poll_t mask, | |
5100 | wait_queue_func_t wake_func) | |
5101 | __acquires(&ctx->completion_lock) | |
5102 | { | |
5103 | struct io_ring_ctx *ctx = req->ctx; | |
5104 | bool cancel = false; | |
5105 | ||
18bceab1 | 5106 | io_init_poll_iocb(poll, mask, wake_func); |
b90cd197 | 5107 | poll->file = req->file; |
18bceab1 | 5108 | poll->wait.private = req; |
d7718a9d JA |
5109 | |
5110 | ipt->pt._key = mask; | |
5111 | ipt->req = req; | |
5112 | ipt->error = -EINVAL; | |
5113 | ||
d7718a9d JA |
5114 | mask = vfs_poll(req->file, &ipt->pt) & poll->events; |
5115 | ||
5116 | spin_lock_irq(&ctx->completion_lock); | |
5117 | if (likely(poll->head)) { | |
5118 | spin_lock(&poll->head->lock); | |
5119 | if (unlikely(list_empty(&poll->wait.entry))) { | |
5120 | if (ipt->error) | |
5121 | cancel = true; | |
5122 | ipt->error = 0; | |
5123 | mask = 0; | |
5124 | } | |
5125 | if (mask || ipt->error) | |
5126 | list_del_init(&poll->wait.entry); | |
5127 | else if (cancel) | |
5128 | WRITE_ONCE(poll->canceled, true); | |
5129 | else if (!poll->done) /* actually waiting for an event */ | |
5130 | io_poll_req_insert(req); | |
5131 | spin_unlock(&poll->head->lock); | |
5132 | } | |
5133 | ||
5134 | return mask; | |
5135 | } | |
5136 | ||
5137 | static bool io_arm_poll_handler(struct io_kiocb *req) | |
5138 | { | |
5139 | const struct io_op_def *def = &io_op_defs[req->opcode]; | |
5140 | struct io_ring_ctx *ctx = req->ctx; | |
5141 | struct async_poll *apoll; | |
5142 | struct io_poll_table ipt; | |
5143 | __poll_t mask, ret; | |
9dab14b8 | 5144 | int rw; |
d7718a9d JA |
5145 | |
5146 | if (!req->file || !file_can_poll(req->file)) | |
5147 | return false; | |
24c74678 | 5148 | if (req->flags & REQ_F_POLLED) |
d7718a9d | 5149 | return false; |
9dab14b8 JA |
5150 | if (def->pollin) |
5151 | rw = READ; | |
5152 | else if (def->pollout) | |
5153 | rw = WRITE; | |
5154 | else | |
5155 | return false; | |
5156 | /* if we can't nonblock try, then no point in arming a poll handler */ | |
5157 | if (!io_file_supports_async(req->file, rw)) | |
d7718a9d JA |
5158 | return false; |
5159 | ||
5160 | apoll = kmalloc(sizeof(*apoll), GFP_ATOMIC); | |
5161 | if (unlikely(!apoll)) | |
5162 | return false; | |
807abcb0 | 5163 | apoll->double_poll = NULL; |
d7718a9d JA |
5164 | |
5165 | req->flags |= REQ_F_POLLED; | |
d7718a9d JA |
5166 | req->apoll = apoll; |
5167 | INIT_HLIST_NODE(&req->hash_node); | |
5168 | ||
8755d97a | 5169 | mask = 0; |
d7718a9d | 5170 | if (def->pollin) |
8755d97a | 5171 | mask |= POLLIN | POLLRDNORM; |
d7718a9d JA |
5172 | if (def->pollout) |
5173 | mask |= POLLOUT | POLLWRNORM; | |
901341bb LH |
5174 | |
5175 | /* If reading from MSG_ERRQUEUE using recvmsg, ignore POLLIN */ | |
5176 | if ((req->opcode == IORING_OP_RECVMSG) && | |
5177 | (req->sr_msg.msg_flags & MSG_ERRQUEUE)) | |
5178 | mask &= ~POLLIN; | |
5179 | ||
d7718a9d JA |
5180 | mask |= POLLERR | POLLPRI; |
5181 | ||
5182 | ipt.pt._qproc = io_async_queue_proc; | |
5183 | ||
5184 | ret = __io_arm_poll_handler(req, &apoll->poll, &ipt, mask, | |
5185 | io_async_wake); | |
a36da65c | 5186 | if (ret || ipt.error) { |
d4e7cd36 | 5187 | io_poll_remove_double(req); |
d7718a9d | 5188 | spin_unlock_irq(&ctx->completion_lock); |
807abcb0 | 5189 | kfree(apoll->double_poll); |
d7718a9d JA |
5190 | kfree(apoll); |
5191 | return false; | |
5192 | } | |
5193 | spin_unlock_irq(&ctx->completion_lock); | |
5194 | trace_io_uring_poll_arm(ctx, req->opcode, req->user_data, mask, | |
5195 | apoll->poll.events); | |
5196 | return true; | |
5197 | } | |
5198 | ||
5199 | static bool __io_poll_remove_one(struct io_kiocb *req, | |
5200 | struct io_poll_iocb *poll) | |
221c5eb2 | 5201 | { |
b41e9852 | 5202 | bool do_complete = false; |
221c5eb2 JA |
5203 | |
5204 | spin_lock(&poll->head->lock); | |
5205 | WRITE_ONCE(poll->canceled, true); | |
392edb45 JA |
5206 | if (!list_empty(&poll->wait.entry)) { |
5207 | list_del_init(&poll->wait.entry); | |
b41e9852 | 5208 | do_complete = true; |
221c5eb2 JA |
5209 | } |
5210 | spin_unlock(&poll->head->lock); | |
3bfa5bcb | 5211 | hash_del(&req->hash_node); |
d7718a9d JA |
5212 | return do_complete; |
5213 | } | |
5214 | ||
5215 | static bool io_poll_remove_one(struct io_kiocb *req) | |
5216 | { | |
5217 | bool do_complete; | |
5218 | ||
d4e7cd36 JA |
5219 | io_poll_remove_double(req); |
5220 | ||
d7718a9d JA |
5221 | if (req->opcode == IORING_OP_POLL_ADD) { |
5222 | do_complete = __io_poll_remove_one(req, &req->poll); | |
5223 | } else { | |
3bfa5bcb JA |
5224 | struct async_poll *apoll = req->apoll; |
5225 | ||
d7718a9d | 5226 | /* non-poll requests have submit ref still */ |
3bfa5bcb JA |
5227 | do_complete = __io_poll_remove_one(req, &apoll->poll); |
5228 | if (do_complete) { | |
d7718a9d | 5229 | io_put_req(req); |
807abcb0 | 5230 | kfree(apoll->double_poll); |
3bfa5bcb JA |
5231 | kfree(apoll); |
5232 | } | |
b1f573bd XW |
5233 | } |
5234 | ||
b41e9852 JA |
5235 | if (do_complete) { |
5236 | io_cqring_fill_event(req, -ECANCELED); | |
5237 | io_commit_cqring(req->ctx); | |
f254ac04 | 5238 | req_set_fail_links(req); |
216578e5 | 5239 | io_put_req_deferred(req, 1); |
b41e9852 JA |
5240 | } |
5241 | ||
5242 | return do_complete; | |
221c5eb2 JA |
5243 | } |
5244 | ||
76e1b642 JA |
5245 | /* |
5246 | * Returns true if we found and killed one or more poll requests | |
5247 | */ | |
5248 | static bool io_poll_remove_all(struct io_ring_ctx *ctx, struct task_struct *tsk) | |
221c5eb2 | 5249 | { |
78076bb6 | 5250 | struct hlist_node *tmp; |
221c5eb2 | 5251 | struct io_kiocb *req; |
8e2e1faf | 5252 | int posted = 0, i; |
221c5eb2 JA |
5253 | |
5254 | spin_lock_irq(&ctx->completion_lock); | |
78076bb6 JA |
5255 | for (i = 0; i < (1U << ctx->cancel_hash_bits); i++) { |
5256 | struct hlist_head *list; | |
5257 | ||
5258 | list = &ctx->cancel_hash[i]; | |
f3606e3a JA |
5259 | hlist_for_each_entry_safe(req, tmp, list, hash_node) { |
5260 | if (io_task_match(req, tsk)) | |
5261 | posted += io_poll_remove_one(req); | |
5262 | } | |
221c5eb2 JA |
5263 | } |
5264 | spin_unlock_irq(&ctx->completion_lock); | |
b41e9852 | 5265 | |
8e2e1faf JA |
5266 | if (posted) |
5267 | io_cqring_ev_posted(ctx); | |
76e1b642 JA |
5268 | |
5269 | return posted != 0; | |
221c5eb2 JA |
5270 | } |
5271 | ||
47f46768 JA |
5272 | static int io_poll_cancel(struct io_ring_ctx *ctx, __u64 sqe_addr) |
5273 | { | |
78076bb6 | 5274 | struct hlist_head *list; |
47f46768 JA |
5275 | struct io_kiocb *req; |
5276 | ||
78076bb6 JA |
5277 | list = &ctx->cancel_hash[hash_long(sqe_addr, ctx->cancel_hash_bits)]; |
5278 | hlist_for_each_entry(req, list, hash_node) { | |
b41e9852 JA |
5279 | if (sqe_addr != req->user_data) |
5280 | continue; | |
5281 | if (io_poll_remove_one(req)) | |
eac406c6 | 5282 | return 0; |
b41e9852 | 5283 | return -EALREADY; |
47f46768 JA |
5284 | } |
5285 | ||
5286 | return -ENOENT; | |
5287 | } | |
5288 | ||
3529d8c2 JA |
5289 | static int io_poll_remove_prep(struct io_kiocb *req, |
5290 | const struct io_uring_sqe *sqe) | |
0969e783 | 5291 | { |
0969e783 JA |
5292 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
5293 | return -EINVAL; | |
5294 | if (sqe->ioprio || sqe->off || sqe->len || sqe->buf_index || | |
5295 | sqe->poll_events) | |
5296 | return -EINVAL; | |
5297 | ||
5298 | req->poll.addr = READ_ONCE(sqe->addr); | |
0969e783 JA |
5299 | return 0; |
5300 | } | |
5301 | ||
221c5eb2 JA |
5302 | /* |
5303 | * Find a running poll command that matches one specified in sqe->addr, | |
5304 | * and remove it if found. | |
5305 | */ | |
fc4df999 | 5306 | static int io_poll_remove(struct io_kiocb *req) |
221c5eb2 JA |
5307 | { |
5308 | struct io_ring_ctx *ctx = req->ctx; | |
0969e783 | 5309 | u64 addr; |
47f46768 | 5310 | int ret; |
221c5eb2 | 5311 | |
0969e783 | 5312 | addr = req->poll.addr; |
221c5eb2 | 5313 | spin_lock_irq(&ctx->completion_lock); |
0969e783 | 5314 | ret = io_poll_cancel(ctx, addr); |
221c5eb2 JA |
5315 | spin_unlock_irq(&ctx->completion_lock); |
5316 | ||
4e88d6e7 JA |
5317 | if (ret < 0) |
5318 | req_set_fail_links(req); | |
e1e16097 | 5319 | io_req_complete(req, ret); |
221c5eb2 JA |
5320 | return 0; |
5321 | } | |
5322 | ||
221c5eb2 JA |
5323 | static int io_poll_wake(struct wait_queue_entry *wait, unsigned mode, int sync, |
5324 | void *key) | |
5325 | { | |
c2f2eb7d JA |
5326 | struct io_kiocb *req = wait->private; |
5327 | struct io_poll_iocb *poll = &req->poll; | |
221c5eb2 | 5328 | |
d7718a9d | 5329 | return __io_async_wake(req, poll, key_to_poll(key), io_poll_task_func); |
221c5eb2 JA |
5330 | } |
5331 | ||
221c5eb2 JA |
5332 | static void io_poll_queue_proc(struct file *file, struct wait_queue_head *head, |
5333 | struct poll_table_struct *p) | |
5334 | { | |
5335 | struct io_poll_table *pt = container_of(p, struct io_poll_table, pt); | |
5336 | ||
e8c2bc1f | 5337 | __io_queue_proc(&pt->req->poll, pt, head, (struct io_poll_iocb **) &pt->req->async_data); |
eac406c6 JA |
5338 | } |
5339 | ||
3529d8c2 | 5340 | static int io_poll_add_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
221c5eb2 JA |
5341 | { |
5342 | struct io_poll_iocb *poll = &req->poll; | |
5769a351 | 5343 | u32 events; |
221c5eb2 JA |
5344 | |
5345 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) | |
5346 | return -EINVAL; | |
5347 | if (sqe->addr || sqe->ioprio || sqe->off || sqe->len || sqe->buf_index) | |
5348 | return -EINVAL; | |
09bb8394 JA |
5349 | if (!poll->file) |
5350 | return -EBADF; | |
221c5eb2 | 5351 | |
5769a351 JX |
5352 | events = READ_ONCE(sqe->poll32_events); |
5353 | #ifdef __BIG_ENDIAN | |
5354 | events = swahw32(events); | |
5355 | #endif | |
a31eb4a2 JX |
5356 | poll->events = demangle_poll(events) | EPOLLERR | EPOLLHUP | |
5357 | (events & EPOLLEXCLUSIVE); | |
0969e783 JA |
5358 | return 0; |
5359 | } | |
5360 | ||
014db007 | 5361 | static int io_poll_add(struct io_kiocb *req) |
0969e783 JA |
5362 | { |
5363 | struct io_poll_iocb *poll = &req->poll; | |
5364 | struct io_ring_ctx *ctx = req->ctx; | |
5365 | struct io_poll_table ipt; | |
0969e783 | 5366 | __poll_t mask; |
0969e783 | 5367 | |
78076bb6 | 5368 | INIT_HLIST_NODE(&req->hash_node); |
d7718a9d | 5369 | ipt.pt._qproc = io_poll_queue_proc; |
36703247 | 5370 | |
d7718a9d JA |
5371 | mask = __io_arm_poll_handler(req, &req->poll, &ipt, poll->events, |
5372 | io_poll_wake); | |
221c5eb2 | 5373 | |
8c838788 | 5374 | if (mask) { /* no async, we'd stolen it */ |
221c5eb2 | 5375 | ipt.error = 0; |
b0dd8a41 | 5376 | io_poll_complete(req, mask, 0); |
221c5eb2 | 5377 | } |
221c5eb2 JA |
5378 | spin_unlock_irq(&ctx->completion_lock); |
5379 | ||
8c838788 JA |
5380 | if (mask) { |
5381 | io_cqring_ev_posted(ctx); | |
014db007 | 5382 | io_put_req(req); |
221c5eb2 | 5383 | } |
8c838788 | 5384 | return ipt.error; |
221c5eb2 JA |
5385 | } |
5386 | ||
5262f567 JA |
5387 | static enum hrtimer_restart io_timeout_fn(struct hrtimer *timer) |
5388 | { | |
ad8a48ac JA |
5389 | struct io_timeout_data *data = container_of(timer, |
5390 | struct io_timeout_data, timer); | |
5391 | struct io_kiocb *req = data->req; | |
5392 | struct io_ring_ctx *ctx = req->ctx; | |
5262f567 JA |
5393 | unsigned long flags; |
5394 | ||
5262f567 | 5395 | spin_lock_irqsave(&ctx->completion_lock, flags); |
a71976f3 | 5396 | list_del_init(&req->timeout.list); |
01cec8c1 PB |
5397 | atomic_set(&req->ctx->cq_timeouts, |
5398 | atomic_read(&req->ctx->cq_timeouts) + 1); | |
5399 | ||
78e19bbe | 5400 | io_cqring_fill_event(req, -ETIME); |
5262f567 JA |
5401 | io_commit_cqring(ctx); |
5402 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
5403 | ||
5404 | io_cqring_ev_posted(ctx); | |
4e88d6e7 | 5405 | req_set_fail_links(req); |
5262f567 JA |
5406 | io_put_req(req); |
5407 | return HRTIMER_NORESTART; | |
5408 | } | |
5409 | ||
f254ac04 JA |
5410 | static int __io_timeout_cancel(struct io_kiocb *req) |
5411 | { | |
e8c2bc1f | 5412 | struct io_timeout_data *io = req->async_data; |
f254ac04 JA |
5413 | int ret; |
5414 | ||
e8c2bc1f | 5415 | ret = hrtimer_try_to_cancel(&io->timer); |
f254ac04 JA |
5416 | if (ret == -1) |
5417 | return -EALREADY; | |
a71976f3 | 5418 | list_del_init(&req->timeout.list); |
f254ac04 JA |
5419 | |
5420 | req_set_fail_links(req); | |
f254ac04 | 5421 | io_cqring_fill_event(req, -ECANCELED); |
216578e5 | 5422 | io_put_req_deferred(req, 1); |
f254ac04 JA |
5423 | return 0; |
5424 | } | |
5425 | ||
47f46768 JA |
5426 | static int io_timeout_cancel(struct io_ring_ctx *ctx, __u64 user_data) |
5427 | { | |
5428 | struct io_kiocb *req; | |
5429 | int ret = -ENOENT; | |
5430 | ||
135fcde8 | 5431 | list_for_each_entry(req, &ctx->timeout_list, timeout.list) { |
47f46768 | 5432 | if (user_data == req->user_data) { |
47f46768 JA |
5433 | ret = 0; |
5434 | break; | |
5435 | } | |
5436 | } | |
5437 | ||
5438 | if (ret == -ENOENT) | |
5439 | return ret; | |
5440 | ||
f254ac04 | 5441 | return __io_timeout_cancel(req); |
47f46768 JA |
5442 | } |
5443 | ||
3529d8c2 JA |
5444 | static int io_timeout_remove_prep(struct io_kiocb *req, |
5445 | const struct io_uring_sqe *sqe) | |
b29472ee | 5446 | { |
b29472ee JA |
5447 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
5448 | return -EINVAL; | |
61710e43 DA |
5449 | if (unlikely(req->flags & (REQ_F_FIXED_FILE | REQ_F_BUFFER_SELECT))) |
5450 | return -EINVAL; | |
0bdf7a2d | 5451 | if (sqe->ioprio || sqe->buf_index || sqe->len || sqe->timeout_flags) |
b29472ee JA |
5452 | return -EINVAL; |
5453 | ||
0bdf7a2d | 5454 | req->timeout_rem.addr = READ_ONCE(sqe->addr); |
b29472ee JA |
5455 | return 0; |
5456 | } | |
5457 | ||
11365043 JA |
5458 | /* |
5459 | * Remove or update an existing timeout command | |
5460 | */ | |
fc4df999 | 5461 | static int io_timeout_remove(struct io_kiocb *req) |
11365043 JA |
5462 | { |
5463 | struct io_ring_ctx *ctx = req->ctx; | |
47f46768 | 5464 | int ret; |
11365043 | 5465 | |
11365043 | 5466 | spin_lock_irq(&ctx->completion_lock); |
0bdf7a2d | 5467 | ret = io_timeout_cancel(ctx, req->timeout_rem.addr); |
11365043 | 5468 | |
47f46768 | 5469 | io_cqring_fill_event(req, ret); |
11365043 JA |
5470 | io_commit_cqring(ctx); |
5471 | spin_unlock_irq(&ctx->completion_lock); | |
5262f567 | 5472 | io_cqring_ev_posted(ctx); |
4e88d6e7 JA |
5473 | if (ret < 0) |
5474 | req_set_fail_links(req); | |
ec9c02ad | 5475 | io_put_req(req); |
11365043 | 5476 | return 0; |
5262f567 JA |
5477 | } |
5478 | ||
3529d8c2 | 5479 | static int io_timeout_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe, |
2d28390a | 5480 | bool is_timeout_link) |
5262f567 | 5481 | { |
ad8a48ac | 5482 | struct io_timeout_data *data; |
a41525ab | 5483 | unsigned flags; |
56080b02 | 5484 | u32 off = READ_ONCE(sqe->off); |
5262f567 | 5485 | |
ad8a48ac | 5486 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
5262f567 | 5487 | return -EINVAL; |
ad8a48ac | 5488 | if (sqe->ioprio || sqe->buf_index || sqe->len != 1) |
a41525ab | 5489 | return -EINVAL; |
56080b02 | 5490 | if (off && is_timeout_link) |
2d28390a | 5491 | return -EINVAL; |
a41525ab JA |
5492 | flags = READ_ONCE(sqe->timeout_flags); |
5493 | if (flags & ~IORING_TIMEOUT_ABS) | |
5262f567 | 5494 | return -EINVAL; |
bdf20073 | 5495 | |
bfe68a22 | 5496 | req->timeout.off = off; |
26a61679 | 5497 | |
e8c2bc1f | 5498 | if (!req->async_data && io_alloc_async_data(req)) |
26a61679 JA |
5499 | return -ENOMEM; |
5500 | ||
e8c2bc1f | 5501 | data = req->async_data; |
ad8a48ac | 5502 | data->req = req; |
ad8a48ac JA |
5503 | |
5504 | if (get_timespec64(&data->ts, u64_to_user_ptr(sqe->addr))) | |
5262f567 JA |
5505 | return -EFAULT; |
5506 | ||
11365043 | 5507 | if (flags & IORING_TIMEOUT_ABS) |
ad8a48ac | 5508 | data->mode = HRTIMER_MODE_ABS; |
11365043 | 5509 | else |
ad8a48ac | 5510 | data->mode = HRTIMER_MODE_REL; |
11365043 | 5511 | |
ad8a48ac JA |
5512 | hrtimer_init(&data->timer, CLOCK_MONOTONIC, data->mode); |
5513 | return 0; | |
5514 | } | |
5515 | ||
fc4df999 | 5516 | static int io_timeout(struct io_kiocb *req) |
ad8a48ac | 5517 | { |
ad8a48ac | 5518 | struct io_ring_ctx *ctx = req->ctx; |
e8c2bc1f | 5519 | struct io_timeout_data *data = req->async_data; |
ad8a48ac | 5520 | struct list_head *entry; |
bfe68a22 | 5521 | u32 tail, off = req->timeout.off; |
ad8a48ac | 5522 | |
733f5c95 | 5523 | spin_lock_irq(&ctx->completion_lock); |
93bd25bb | 5524 | |
5262f567 JA |
5525 | /* |
5526 | * sqe->off holds how many events that need to occur for this | |
93bd25bb JA |
5527 | * timeout event to be satisfied. If it isn't set, then this is |
5528 | * a pure timeout request, sequence isn't used. | |
5262f567 | 5529 | */ |
8eb7e2d0 | 5530 | if (io_is_timeout_noseq(req)) { |
93bd25bb JA |
5531 | entry = ctx->timeout_list.prev; |
5532 | goto add; | |
5533 | } | |
5262f567 | 5534 | |
bfe68a22 PB |
5535 | tail = ctx->cached_cq_tail - atomic_read(&ctx->cq_timeouts); |
5536 | req->timeout.target_seq = tail + off; | |
5262f567 JA |
5537 | |
5538 | /* | |
5539 | * Insertion sort, ensuring the first entry in the list is always | |
5540 | * the one we need first. | |
5541 | */ | |
5262f567 | 5542 | list_for_each_prev(entry, &ctx->timeout_list) { |
135fcde8 PB |
5543 | struct io_kiocb *nxt = list_entry(entry, struct io_kiocb, |
5544 | timeout.list); | |
5262f567 | 5545 | |
8eb7e2d0 | 5546 | if (io_is_timeout_noseq(nxt)) |
93bd25bb | 5547 | continue; |
bfe68a22 PB |
5548 | /* nxt.seq is behind @tail, otherwise would've been completed */ |
5549 | if (off >= nxt->timeout.target_seq - tail) | |
5262f567 JA |
5550 | break; |
5551 | } | |
93bd25bb | 5552 | add: |
135fcde8 | 5553 | list_add(&req->timeout.list, entry); |
ad8a48ac JA |
5554 | data->timer.function = io_timeout_fn; |
5555 | hrtimer_start(&data->timer, timespec64_to_ktime(data->ts), data->mode); | |
5262f567 | 5556 | spin_unlock_irq(&ctx->completion_lock); |
5262f567 JA |
5557 | return 0; |
5558 | } | |
5262f567 | 5559 | |
62755e35 JA |
5560 | static bool io_cancel_cb(struct io_wq_work *work, void *data) |
5561 | { | |
5562 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); | |
5563 | ||
5564 | return req->user_data == (unsigned long) data; | |
5565 | } | |
5566 | ||
e977d6d3 | 5567 | static int io_async_cancel_one(struct io_ring_ctx *ctx, void *sqe_addr) |
62755e35 | 5568 | { |
62755e35 | 5569 | enum io_wq_cancel cancel_ret; |
62755e35 JA |
5570 | int ret = 0; |
5571 | ||
4f26bda1 | 5572 | cancel_ret = io_wq_cancel_cb(ctx->io_wq, io_cancel_cb, sqe_addr, false); |
62755e35 JA |
5573 | switch (cancel_ret) { |
5574 | case IO_WQ_CANCEL_OK: | |
5575 | ret = 0; | |
5576 | break; | |
5577 | case IO_WQ_CANCEL_RUNNING: | |
5578 | ret = -EALREADY; | |
5579 | break; | |
5580 | case IO_WQ_CANCEL_NOTFOUND: | |
5581 | ret = -ENOENT; | |
5582 | break; | |
5583 | } | |
5584 | ||
e977d6d3 JA |
5585 | return ret; |
5586 | } | |
5587 | ||
47f46768 JA |
5588 | static void io_async_find_and_cancel(struct io_ring_ctx *ctx, |
5589 | struct io_kiocb *req, __u64 sqe_addr, | |
014db007 | 5590 | int success_ret) |
47f46768 JA |
5591 | { |
5592 | unsigned long flags; | |
5593 | int ret; | |
5594 | ||
5595 | ret = io_async_cancel_one(ctx, (void *) (unsigned long) sqe_addr); | |
5596 | if (ret != -ENOENT) { | |
5597 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
5598 | goto done; | |
5599 | } | |
5600 | ||
5601 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
5602 | ret = io_timeout_cancel(ctx, sqe_addr); | |
5603 | if (ret != -ENOENT) | |
5604 | goto done; | |
5605 | ret = io_poll_cancel(ctx, sqe_addr); | |
5606 | done: | |
b0dd8a41 JA |
5607 | if (!ret) |
5608 | ret = success_ret; | |
47f46768 JA |
5609 | io_cqring_fill_event(req, ret); |
5610 | io_commit_cqring(ctx); | |
5611 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
5612 | io_cqring_ev_posted(ctx); | |
5613 | ||
4e88d6e7 JA |
5614 | if (ret < 0) |
5615 | req_set_fail_links(req); | |
014db007 | 5616 | io_put_req(req); |
47f46768 JA |
5617 | } |
5618 | ||
3529d8c2 JA |
5619 | static int io_async_cancel_prep(struct io_kiocb *req, |
5620 | const struct io_uring_sqe *sqe) | |
e977d6d3 | 5621 | { |
fbf23849 | 5622 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
e977d6d3 | 5623 | return -EINVAL; |
61710e43 DA |
5624 | if (unlikely(req->flags & (REQ_F_FIXED_FILE | REQ_F_BUFFER_SELECT))) |
5625 | return -EINVAL; | |
5626 | if (sqe->ioprio || sqe->off || sqe->len || sqe->cancel_flags) | |
e977d6d3 JA |
5627 | return -EINVAL; |
5628 | ||
fbf23849 JA |
5629 | req->cancel.addr = READ_ONCE(sqe->addr); |
5630 | return 0; | |
5631 | } | |
5632 | ||
014db007 | 5633 | static int io_async_cancel(struct io_kiocb *req) |
fbf23849 JA |
5634 | { |
5635 | struct io_ring_ctx *ctx = req->ctx; | |
fbf23849 | 5636 | |
014db007 | 5637 | io_async_find_and_cancel(ctx, req, req->cancel.addr, 0); |
5262f567 JA |
5638 | return 0; |
5639 | } | |
5640 | ||
05f3fb3c JA |
5641 | static int io_files_update_prep(struct io_kiocb *req, |
5642 | const struct io_uring_sqe *sqe) | |
5643 | { | |
6ca56f84 JA |
5644 | if (unlikely(req->ctx->flags & IORING_SETUP_SQPOLL)) |
5645 | return -EINVAL; | |
61710e43 DA |
5646 | if (unlikely(req->flags & (REQ_F_FIXED_FILE | REQ_F_BUFFER_SELECT))) |
5647 | return -EINVAL; | |
5648 | if (sqe->ioprio || sqe->rw_flags) | |
05f3fb3c JA |
5649 | return -EINVAL; |
5650 | ||
5651 | req->files_update.offset = READ_ONCE(sqe->off); | |
5652 | req->files_update.nr_args = READ_ONCE(sqe->len); | |
5653 | if (!req->files_update.nr_args) | |
5654 | return -EINVAL; | |
5655 | req->files_update.arg = READ_ONCE(sqe->addr); | |
5656 | return 0; | |
5657 | } | |
5658 | ||
229a7b63 JA |
5659 | static int io_files_update(struct io_kiocb *req, bool force_nonblock, |
5660 | struct io_comp_state *cs) | |
fbf23849 JA |
5661 | { |
5662 | struct io_ring_ctx *ctx = req->ctx; | |
05f3fb3c JA |
5663 | struct io_uring_files_update up; |
5664 | int ret; | |
fbf23849 | 5665 | |
f86cd20c | 5666 | if (force_nonblock) |
05f3fb3c | 5667 | return -EAGAIN; |
05f3fb3c JA |
5668 | |
5669 | up.offset = req->files_update.offset; | |
5670 | up.fds = req->files_update.arg; | |
5671 | ||
5672 | mutex_lock(&ctx->uring_lock); | |
5673 | ret = __io_sqe_files_update(ctx, &up, req->files_update.nr_args); | |
5674 | mutex_unlock(&ctx->uring_lock); | |
5675 | ||
5676 | if (ret < 0) | |
5677 | req_set_fail_links(req); | |
229a7b63 | 5678 | __io_req_complete(req, ret, 0, cs); |
5262f567 JA |
5679 | return 0; |
5680 | } | |
5681 | ||
bfe76559 | 5682 | static int io_req_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
f67676d1 | 5683 | { |
d625c6ee | 5684 | switch (req->opcode) { |
e781573e | 5685 | case IORING_OP_NOP: |
bfe76559 | 5686 | return 0; |
f67676d1 JA |
5687 | case IORING_OP_READV: |
5688 | case IORING_OP_READ_FIXED: | |
3a6820f2 | 5689 | case IORING_OP_READ: |
bfe76559 | 5690 | return io_read_prep(req, sqe); |
f67676d1 JA |
5691 | case IORING_OP_WRITEV: |
5692 | case IORING_OP_WRITE_FIXED: | |
3a6820f2 | 5693 | case IORING_OP_WRITE: |
bfe76559 | 5694 | return io_write_prep(req, sqe); |
0969e783 | 5695 | case IORING_OP_POLL_ADD: |
bfe76559 | 5696 | return io_poll_add_prep(req, sqe); |
0969e783 | 5697 | case IORING_OP_POLL_REMOVE: |
bfe76559 | 5698 | return io_poll_remove_prep(req, sqe); |
8ed8d3c3 | 5699 | case IORING_OP_FSYNC: |
bfe76559 | 5700 | return io_prep_fsync(req, sqe); |
8ed8d3c3 | 5701 | case IORING_OP_SYNC_FILE_RANGE: |
bfe76559 | 5702 | return io_prep_sfr(req, sqe); |
03b1230c | 5703 | case IORING_OP_SENDMSG: |
fddaface | 5704 | case IORING_OP_SEND: |
bfe76559 | 5705 | return io_sendmsg_prep(req, sqe); |
03b1230c | 5706 | case IORING_OP_RECVMSG: |
fddaface | 5707 | case IORING_OP_RECV: |
bfe76559 | 5708 | return io_recvmsg_prep(req, sqe); |
f499a021 | 5709 | case IORING_OP_CONNECT: |
bfe76559 | 5710 | return io_connect_prep(req, sqe); |
2d28390a | 5711 | case IORING_OP_TIMEOUT: |
bfe76559 | 5712 | return io_timeout_prep(req, sqe, false); |
b29472ee | 5713 | case IORING_OP_TIMEOUT_REMOVE: |
bfe76559 | 5714 | return io_timeout_remove_prep(req, sqe); |
fbf23849 | 5715 | case IORING_OP_ASYNC_CANCEL: |
bfe76559 | 5716 | return io_async_cancel_prep(req, sqe); |
2d28390a | 5717 | case IORING_OP_LINK_TIMEOUT: |
bfe76559 | 5718 | return io_timeout_prep(req, sqe, true); |
8ed8d3c3 | 5719 | case IORING_OP_ACCEPT: |
bfe76559 | 5720 | return io_accept_prep(req, sqe); |
d63d1b5e | 5721 | case IORING_OP_FALLOCATE: |
bfe76559 | 5722 | return io_fallocate_prep(req, sqe); |
15b71abe | 5723 | case IORING_OP_OPENAT: |
bfe76559 | 5724 | return io_openat_prep(req, sqe); |
b5dba59e | 5725 | case IORING_OP_CLOSE: |
bfe76559 | 5726 | return io_close_prep(req, sqe); |
05f3fb3c | 5727 | case IORING_OP_FILES_UPDATE: |
bfe76559 | 5728 | return io_files_update_prep(req, sqe); |
eddc7ef5 | 5729 | case IORING_OP_STATX: |
bfe76559 | 5730 | return io_statx_prep(req, sqe); |
4840e418 | 5731 | case IORING_OP_FADVISE: |
bfe76559 | 5732 | return io_fadvise_prep(req, sqe); |
c1ca757b | 5733 | case IORING_OP_MADVISE: |
bfe76559 | 5734 | return io_madvise_prep(req, sqe); |
cebdb986 | 5735 | case IORING_OP_OPENAT2: |
bfe76559 | 5736 | return io_openat2_prep(req, sqe); |
3e4827b0 | 5737 | case IORING_OP_EPOLL_CTL: |
bfe76559 | 5738 | return io_epoll_ctl_prep(req, sqe); |
7d67af2c | 5739 | case IORING_OP_SPLICE: |
bfe76559 | 5740 | return io_splice_prep(req, sqe); |
ddf0322d | 5741 | case IORING_OP_PROVIDE_BUFFERS: |
bfe76559 | 5742 | return io_provide_buffers_prep(req, sqe); |
067524e9 | 5743 | case IORING_OP_REMOVE_BUFFERS: |
bfe76559 | 5744 | return io_remove_buffers_prep(req, sqe); |
f2a8d5c7 | 5745 | case IORING_OP_TEE: |
bfe76559 | 5746 | return io_tee_prep(req, sqe); |
f67676d1 JA |
5747 | } |
5748 | ||
bfe76559 PB |
5749 | printk_once(KERN_WARNING "io_uring: unhandled opcode %d\n", |
5750 | req->opcode); | |
5751 | return-EINVAL; | |
5752 | } | |
5753 | ||
5754 | static int io_req_defer_prep(struct io_kiocb *req, | |
5755 | const struct io_uring_sqe *sqe) | |
5756 | { | |
bfe76559 PB |
5757 | if (!sqe) |
5758 | return 0; | |
5759 | if (io_alloc_async_data(req)) | |
5760 | return -EAGAIN; | |
bfe76559 | 5761 | return io_req_prep(req, sqe); |
f67676d1 JA |
5762 | } |
5763 | ||
9cf7c104 PB |
5764 | static u32 io_get_sequence(struct io_kiocb *req) |
5765 | { | |
5766 | struct io_kiocb *pos; | |
5767 | struct io_ring_ctx *ctx = req->ctx; | |
5768 | u32 total_submitted, nr_reqs = 1; | |
5769 | ||
5770 | if (req->flags & REQ_F_LINK_HEAD) | |
5771 | list_for_each_entry(pos, &req->link_list, link_list) | |
5772 | nr_reqs++; | |
5773 | ||
5774 | total_submitted = ctx->cached_sq_head - ctx->cached_sq_dropped; | |
5775 | return total_submitted - nr_reqs; | |
5776 | } | |
5777 | ||
3529d8c2 | 5778 | static int io_req_defer(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
de0617e4 | 5779 | { |
a197f664 | 5780 | struct io_ring_ctx *ctx = req->ctx; |
27dc8338 | 5781 | struct io_defer_entry *de; |
f67676d1 | 5782 | int ret; |
9cf7c104 | 5783 | u32 seq; |
de0617e4 | 5784 | |
9d858b21 | 5785 | /* Still need defer if there is pending req in defer list. */ |
9cf7c104 PB |
5786 | if (likely(list_empty_careful(&ctx->defer_list) && |
5787 | !(req->flags & REQ_F_IO_DRAIN))) | |
5788 | return 0; | |
5789 | ||
5790 | seq = io_get_sequence(req); | |
5791 | /* Still a chance to pass the sequence check */ | |
5792 | if (!req_need_defer(req, seq) && list_empty_careful(&ctx->defer_list)) | |
de0617e4 JA |
5793 | return 0; |
5794 | ||
e8c2bc1f | 5795 | if (!req->async_data) { |
650b5481 | 5796 | ret = io_req_defer_prep(req, sqe); |
327d6d96 | 5797 | if (ret) |
650b5481 PB |
5798 | return ret; |
5799 | } | |
cbdcb435 | 5800 | io_prep_async_link(req); |
27dc8338 PB |
5801 | de = kmalloc(sizeof(*de), GFP_KERNEL); |
5802 | if (!de) | |
5803 | return -ENOMEM; | |
2d28390a | 5804 | |
de0617e4 | 5805 | spin_lock_irq(&ctx->completion_lock); |
9cf7c104 | 5806 | if (!req_need_defer(req, seq) && list_empty(&ctx->defer_list)) { |
de0617e4 | 5807 | spin_unlock_irq(&ctx->completion_lock); |
27dc8338 | 5808 | kfree(de); |
ae34817b PB |
5809 | io_queue_async_work(req); |
5810 | return -EIOCBQUEUED; | |
de0617e4 JA |
5811 | } |
5812 | ||
915967f6 | 5813 | trace_io_uring_defer(ctx, req, req->user_data); |
27dc8338 | 5814 | de->req = req; |
9cf7c104 | 5815 | de->seq = seq; |
27dc8338 | 5816 | list_add_tail(&de->list, &ctx->defer_list); |
de0617e4 JA |
5817 | spin_unlock_irq(&ctx->completion_lock); |
5818 | return -EIOCBQUEUED; | |
5819 | } | |
5820 | ||
f573d384 | 5821 | static void io_req_drop_files(struct io_kiocb *req) |
99bc4c38 | 5822 | { |
f573d384 JA |
5823 | struct io_ring_ctx *ctx = req->ctx; |
5824 | unsigned long flags; | |
5825 | ||
5826 | spin_lock_irqsave(&ctx->inflight_lock, flags); | |
5827 | list_del(&req->inflight_entry); | |
5828 | if (waitqueue_active(&ctx->inflight_wait)) | |
5829 | wake_up(&ctx->inflight_wait); | |
5830 | spin_unlock_irqrestore(&ctx->inflight_lock, flags); | |
5831 | req->flags &= ~REQ_F_INFLIGHT; | |
98447d65 JA |
5832 | put_files_struct(req->work.identity->files); |
5833 | put_nsproxy(req->work.identity->nsproxy); | |
dfead8a8 | 5834 | req->work.flags &= ~IO_WQ_WORK_FILES; |
f573d384 | 5835 | } |
99bc4c38 | 5836 | |
3ca405eb | 5837 | static void __io_clean_op(struct io_kiocb *req) |
99bc4c38 | 5838 | { |
0e1b6fe3 PB |
5839 | if (req->flags & REQ_F_BUFFER_SELECTED) { |
5840 | switch (req->opcode) { | |
5841 | case IORING_OP_READV: | |
5842 | case IORING_OP_READ_FIXED: | |
5843 | case IORING_OP_READ: | |
bcda7baa | 5844 | kfree((void *)(unsigned long)req->rw.addr); |
0e1b6fe3 PB |
5845 | break; |
5846 | case IORING_OP_RECVMSG: | |
5847 | case IORING_OP_RECV: | |
bcda7baa | 5848 | kfree(req->sr_msg.kbuf); |
0e1b6fe3 PB |
5849 | break; |
5850 | } | |
5851 | req->flags &= ~REQ_F_BUFFER_SELECTED; | |
99bc4c38 PB |
5852 | } |
5853 | ||
0e1b6fe3 PB |
5854 | if (req->flags & REQ_F_NEED_CLEANUP) { |
5855 | switch (req->opcode) { | |
5856 | case IORING_OP_READV: | |
5857 | case IORING_OP_READ_FIXED: | |
5858 | case IORING_OP_READ: | |
5859 | case IORING_OP_WRITEV: | |
5860 | case IORING_OP_WRITE_FIXED: | |
e8c2bc1f JA |
5861 | case IORING_OP_WRITE: { |
5862 | struct io_async_rw *io = req->async_data; | |
5863 | if (io->free_iovec) | |
5864 | kfree(io->free_iovec); | |
0e1b6fe3 | 5865 | break; |
e8c2bc1f | 5866 | } |
0e1b6fe3 | 5867 | case IORING_OP_RECVMSG: |
e8c2bc1f JA |
5868 | case IORING_OP_SENDMSG: { |
5869 | struct io_async_msghdr *io = req->async_data; | |
5870 | if (io->iov != io->fast_iov) | |
5871 | kfree(io->iov); | |
0e1b6fe3 | 5872 | break; |
e8c2bc1f | 5873 | } |
0e1b6fe3 PB |
5874 | case IORING_OP_SPLICE: |
5875 | case IORING_OP_TEE: | |
5876 | io_put_file(req, req->splice.file_in, | |
5877 | (req->splice.flags & SPLICE_F_FD_IN_FIXED)); | |
5878 | break; | |
f3cd4850 JA |
5879 | case IORING_OP_OPENAT: |
5880 | case IORING_OP_OPENAT2: | |
5881 | if (req->open.filename) | |
5882 | putname(req->open.filename); | |
5883 | break; | |
0e1b6fe3 PB |
5884 | } |
5885 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
99bc4c38 | 5886 | } |
bb175342 | 5887 | |
f573d384 JA |
5888 | if (req->flags & REQ_F_INFLIGHT) |
5889 | io_req_drop_files(req); | |
99bc4c38 PB |
5890 | } |
5891 | ||
c1379e24 PB |
5892 | static int io_issue_sqe(struct io_kiocb *req, bool force_nonblock, |
5893 | struct io_comp_state *cs) | |
2b188cc1 | 5894 | { |
a197f664 | 5895 | struct io_ring_ctx *ctx = req->ctx; |
d625c6ee | 5896 | int ret; |
2b188cc1 | 5897 | |
d625c6ee | 5898 | switch (req->opcode) { |
2b188cc1 | 5899 | case IORING_OP_NOP: |
229a7b63 | 5900 | ret = io_nop(req, cs); |
2b188cc1 JA |
5901 | break; |
5902 | case IORING_OP_READV: | |
edafccee | 5903 | case IORING_OP_READ_FIXED: |
3a6820f2 | 5904 | case IORING_OP_READ: |
a1d7c393 | 5905 | ret = io_read(req, force_nonblock, cs); |
edafccee | 5906 | break; |
3529d8c2 | 5907 | case IORING_OP_WRITEV: |
edafccee | 5908 | case IORING_OP_WRITE_FIXED: |
3a6820f2 | 5909 | case IORING_OP_WRITE: |
a1d7c393 | 5910 | ret = io_write(req, force_nonblock, cs); |
2b188cc1 | 5911 | break; |
c992fe29 | 5912 | case IORING_OP_FSYNC: |
014db007 | 5913 | ret = io_fsync(req, force_nonblock); |
c992fe29 | 5914 | break; |
221c5eb2 | 5915 | case IORING_OP_POLL_ADD: |
014db007 | 5916 | ret = io_poll_add(req); |
221c5eb2 JA |
5917 | break; |
5918 | case IORING_OP_POLL_REMOVE: | |
fc4df999 | 5919 | ret = io_poll_remove(req); |
221c5eb2 | 5920 | break; |
5d17b4a4 | 5921 | case IORING_OP_SYNC_FILE_RANGE: |
014db007 | 5922 | ret = io_sync_file_range(req, force_nonblock); |
5d17b4a4 | 5923 | break; |
0fa03c62 | 5924 | case IORING_OP_SENDMSG: |
062d04d7 PB |
5925 | ret = io_sendmsg(req, force_nonblock, cs); |
5926 | break; | |
fddaface | 5927 | case IORING_OP_SEND: |
062d04d7 | 5928 | ret = io_send(req, force_nonblock, cs); |
0fa03c62 | 5929 | break; |
aa1fa28f | 5930 | case IORING_OP_RECVMSG: |
062d04d7 PB |
5931 | ret = io_recvmsg(req, force_nonblock, cs); |
5932 | break; | |
fddaface | 5933 | case IORING_OP_RECV: |
062d04d7 | 5934 | ret = io_recv(req, force_nonblock, cs); |
aa1fa28f | 5935 | break; |
5262f567 | 5936 | case IORING_OP_TIMEOUT: |
fc4df999 | 5937 | ret = io_timeout(req); |
5262f567 | 5938 | break; |
11365043 | 5939 | case IORING_OP_TIMEOUT_REMOVE: |
fc4df999 | 5940 | ret = io_timeout_remove(req); |
11365043 | 5941 | break; |
17f2fe35 | 5942 | case IORING_OP_ACCEPT: |
229a7b63 | 5943 | ret = io_accept(req, force_nonblock, cs); |
17f2fe35 | 5944 | break; |
f8e85cf2 | 5945 | case IORING_OP_CONNECT: |
229a7b63 | 5946 | ret = io_connect(req, force_nonblock, cs); |
f8e85cf2 | 5947 | break; |
62755e35 | 5948 | case IORING_OP_ASYNC_CANCEL: |
014db007 | 5949 | ret = io_async_cancel(req); |
62755e35 | 5950 | break; |
d63d1b5e | 5951 | case IORING_OP_FALLOCATE: |
014db007 | 5952 | ret = io_fallocate(req, force_nonblock); |
d63d1b5e | 5953 | break; |
15b71abe | 5954 | case IORING_OP_OPENAT: |
014db007 | 5955 | ret = io_openat(req, force_nonblock); |
15b71abe | 5956 | break; |
b5dba59e | 5957 | case IORING_OP_CLOSE: |
229a7b63 | 5958 | ret = io_close(req, force_nonblock, cs); |
b5dba59e | 5959 | break; |
05f3fb3c | 5960 | case IORING_OP_FILES_UPDATE: |
229a7b63 | 5961 | ret = io_files_update(req, force_nonblock, cs); |
05f3fb3c | 5962 | break; |
eddc7ef5 | 5963 | case IORING_OP_STATX: |
014db007 | 5964 | ret = io_statx(req, force_nonblock); |
eddc7ef5 | 5965 | break; |
4840e418 | 5966 | case IORING_OP_FADVISE: |
014db007 | 5967 | ret = io_fadvise(req, force_nonblock); |
4840e418 | 5968 | break; |
c1ca757b | 5969 | case IORING_OP_MADVISE: |
014db007 | 5970 | ret = io_madvise(req, force_nonblock); |
c1ca757b | 5971 | break; |
cebdb986 | 5972 | case IORING_OP_OPENAT2: |
014db007 | 5973 | ret = io_openat2(req, force_nonblock); |
cebdb986 | 5974 | break; |
3e4827b0 | 5975 | case IORING_OP_EPOLL_CTL: |
229a7b63 | 5976 | ret = io_epoll_ctl(req, force_nonblock, cs); |
3e4827b0 | 5977 | break; |
7d67af2c | 5978 | case IORING_OP_SPLICE: |
014db007 | 5979 | ret = io_splice(req, force_nonblock); |
7d67af2c | 5980 | break; |
ddf0322d | 5981 | case IORING_OP_PROVIDE_BUFFERS: |
229a7b63 | 5982 | ret = io_provide_buffers(req, force_nonblock, cs); |
ddf0322d | 5983 | break; |
067524e9 | 5984 | case IORING_OP_REMOVE_BUFFERS: |
229a7b63 | 5985 | ret = io_remove_buffers(req, force_nonblock, cs); |
3e4827b0 | 5986 | break; |
f2a8d5c7 | 5987 | case IORING_OP_TEE: |
f2a8d5c7 PB |
5988 | ret = io_tee(req, force_nonblock); |
5989 | break; | |
2b188cc1 JA |
5990 | default: |
5991 | ret = -EINVAL; | |
5992 | break; | |
5993 | } | |
5994 | ||
def596e9 JA |
5995 | if (ret) |
5996 | return ret; | |
5997 | ||
b532576e JA |
5998 | /* If the op doesn't have a file, we're not polling for it */ |
5999 | if ((ctx->flags & IORING_SETUP_IOPOLL) && req->file) { | |
11ba820b JA |
6000 | const bool in_async = io_wq_current_is_worker(); |
6001 | ||
11ba820b JA |
6002 | /* workqueue context doesn't hold uring_lock, grab it now */ |
6003 | if (in_async) | |
6004 | mutex_lock(&ctx->uring_lock); | |
6005 | ||
def596e9 | 6006 | io_iopoll_req_issued(req); |
11ba820b JA |
6007 | |
6008 | if (in_async) | |
6009 | mutex_unlock(&ctx->uring_lock); | |
def596e9 JA |
6010 | } |
6011 | ||
6012 | return 0; | |
2b188cc1 JA |
6013 | } |
6014 | ||
f4db7182 | 6015 | static struct io_wq_work *io_wq_submit_work(struct io_wq_work *work) |
2b188cc1 JA |
6016 | { |
6017 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); | |
6df1db6b | 6018 | struct io_kiocb *timeout; |
561fb04a | 6019 | int ret = 0; |
2b188cc1 | 6020 | |
6df1db6b PB |
6021 | timeout = io_prep_linked_timeout(req); |
6022 | if (timeout) | |
6023 | io_queue_linked_timeout(timeout); | |
d4c81f38 | 6024 | |
0c9d5ccd JA |
6025 | /* if NO_CANCEL is set, we must still run the work */ |
6026 | if ((work->flags & (IO_WQ_WORK_CANCEL|IO_WQ_WORK_NO_CANCEL)) == | |
6027 | IO_WQ_WORK_CANCEL) { | |
561fb04a | 6028 | ret = -ECANCELED; |
0c9d5ccd | 6029 | } |
31b51510 | 6030 | |
561fb04a | 6031 | if (!ret) { |
561fb04a | 6032 | do { |
c1379e24 | 6033 | ret = io_issue_sqe(req, false, NULL); |
561fb04a JA |
6034 | /* |
6035 | * We can get EAGAIN for polled IO even though we're | |
6036 | * forcing a sync submission from here, since we can't | |
6037 | * wait for request slots on the block side. | |
6038 | */ | |
6039 | if (ret != -EAGAIN) | |
6040 | break; | |
6041 | cond_resched(); | |
6042 | } while (1); | |
6043 | } | |
31b51510 | 6044 | |
561fb04a | 6045 | if (ret) { |
4e88d6e7 | 6046 | req_set_fail_links(req); |
e1e16097 | 6047 | io_req_complete(req, ret); |
edafccee | 6048 | } |
2b188cc1 | 6049 | |
f4db7182 | 6050 | return io_steal_work(req); |
2b188cc1 JA |
6051 | } |
6052 | ||
65e19f54 JA |
6053 | static inline struct file *io_file_from_index(struct io_ring_ctx *ctx, |
6054 | int index) | |
6055 | { | |
6056 | struct fixed_file_table *table; | |
6057 | ||
05f3fb3c | 6058 | table = &ctx->file_data->table[index >> IORING_FILE_TABLE_SHIFT]; |
84695089 | 6059 | return table->files[index & IORING_FILE_TABLE_MASK]; |
65e19f54 JA |
6060 | } |
6061 | ||
8371adf5 PB |
6062 | static struct file *io_file_get(struct io_submit_state *state, |
6063 | struct io_kiocb *req, int fd, bool fixed) | |
09bb8394 | 6064 | { |
a197f664 | 6065 | struct io_ring_ctx *ctx = req->ctx; |
8da11c19 | 6066 | struct file *file; |
09bb8394 | 6067 | |
8da11c19 | 6068 | if (fixed) { |
479f517b | 6069 | if (unlikely((unsigned int)fd >= ctx->nr_user_files)) |
8371adf5 | 6070 | return NULL; |
b7620121 | 6071 | fd = array_index_nospec(fd, ctx->nr_user_files); |
8da11c19 | 6072 | file = io_file_from_index(ctx, fd); |
fd2206e4 | 6073 | if (file) { |
b2e96852 | 6074 | req->fixed_file_refs = &ctx->file_data->node->refs; |
fd2206e4 JA |
6075 | percpu_ref_get(req->fixed_file_refs); |
6076 | } | |
09bb8394 | 6077 | } else { |
c826bd7a | 6078 | trace_io_uring_file_get(ctx, fd); |
8da11c19 | 6079 | file = __io_file_get(state, fd); |
09bb8394 JA |
6080 | } |
6081 | ||
8371adf5 | 6082 | return file; |
09bb8394 JA |
6083 | } |
6084 | ||
8da11c19 | 6085 | static int io_req_set_file(struct io_submit_state *state, struct io_kiocb *req, |
63ff8223 | 6086 | int fd) |
8da11c19 | 6087 | { |
8da11c19 PB |
6088 | bool fixed; |
6089 | ||
63ff8223 | 6090 | fixed = (req->flags & REQ_F_FIXED_FILE) != 0; |
0cdaf760 | 6091 | if (unlikely(!fixed && io_async_submit(req->ctx))) |
8da11c19 PB |
6092 | return -EBADF; |
6093 | ||
8371adf5 PB |
6094 | req->file = io_file_get(state, req, fd, fixed); |
6095 | if (req->file || io_op_defs[req->opcode].needs_file_no_error) | |
f86cd20c | 6096 | return 0; |
8371adf5 | 6097 | return -EBADF; |
f56040b8 PB |
6098 | } |
6099 | ||
2665abfd | 6100 | static enum hrtimer_restart io_link_timeout_fn(struct hrtimer *timer) |
2b188cc1 | 6101 | { |
ad8a48ac JA |
6102 | struct io_timeout_data *data = container_of(timer, |
6103 | struct io_timeout_data, timer); | |
6104 | struct io_kiocb *req = data->req; | |
2665abfd JA |
6105 | struct io_ring_ctx *ctx = req->ctx; |
6106 | struct io_kiocb *prev = NULL; | |
6107 | unsigned long flags; | |
2665abfd JA |
6108 | |
6109 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
6110 | ||
6111 | /* | |
6112 | * We don't expect the list to be empty, that will only happen if we | |
6113 | * race with the completion of the linked work. | |
6114 | */ | |
4493233e PB |
6115 | if (!list_empty(&req->link_list)) { |
6116 | prev = list_entry(req->link_list.prev, struct io_kiocb, | |
6117 | link_list); | |
5d960724 | 6118 | if (refcount_inc_not_zero(&prev->refs)) { |
4493233e | 6119 | list_del_init(&req->link_list); |
5d960724 JA |
6120 | prev->flags &= ~REQ_F_LINK_TIMEOUT; |
6121 | } else | |
76a46e06 | 6122 | prev = NULL; |
2665abfd JA |
6123 | } |
6124 | ||
6125 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
6126 | ||
6127 | if (prev) { | |
4e88d6e7 | 6128 | req_set_fail_links(prev); |
014db007 | 6129 | io_async_find_and_cancel(ctx, req, prev->user_data, -ETIME); |
76a46e06 | 6130 | io_put_req(prev); |
47f46768 | 6131 | } else { |
e1e16097 | 6132 | io_req_complete(req, -ETIME); |
2665abfd | 6133 | } |
2665abfd JA |
6134 | return HRTIMER_NORESTART; |
6135 | } | |
6136 | ||
7271ef3a | 6137 | static void __io_queue_linked_timeout(struct io_kiocb *req) |
2665abfd | 6138 | { |
76a46e06 JA |
6139 | /* |
6140 | * If the list is now empty, then our linked request finished before | |
6141 | * we got a chance to setup the timer | |
6142 | */ | |
4493233e | 6143 | if (!list_empty(&req->link_list)) { |
e8c2bc1f | 6144 | struct io_timeout_data *data = req->async_data; |
94ae5e77 | 6145 | |
ad8a48ac JA |
6146 | data->timer.function = io_link_timeout_fn; |
6147 | hrtimer_start(&data->timer, timespec64_to_ktime(data->ts), | |
6148 | data->mode); | |
2665abfd | 6149 | } |
7271ef3a JA |
6150 | } |
6151 | ||
6152 | static void io_queue_linked_timeout(struct io_kiocb *req) | |
6153 | { | |
6154 | struct io_ring_ctx *ctx = req->ctx; | |
6155 | ||
6156 | spin_lock_irq(&ctx->completion_lock); | |
6157 | __io_queue_linked_timeout(req); | |
76a46e06 | 6158 | spin_unlock_irq(&ctx->completion_lock); |
2665abfd | 6159 | |
2665abfd | 6160 | /* drop submission reference */ |
76a46e06 JA |
6161 | io_put_req(req); |
6162 | } | |
2665abfd | 6163 | |
ad8a48ac | 6164 | static struct io_kiocb *io_prep_linked_timeout(struct io_kiocb *req) |
2665abfd JA |
6165 | { |
6166 | struct io_kiocb *nxt; | |
6167 | ||
dea3b49c | 6168 | if (!(req->flags & REQ_F_LINK_HEAD)) |
2665abfd | 6169 | return NULL; |
6df1db6b | 6170 | if (req->flags & REQ_F_LINK_TIMEOUT) |
d7718a9d | 6171 | return NULL; |
2665abfd | 6172 | |
4493233e PB |
6173 | nxt = list_first_entry_or_null(&req->link_list, struct io_kiocb, |
6174 | link_list); | |
d625c6ee | 6175 | if (!nxt || nxt->opcode != IORING_OP_LINK_TIMEOUT) |
76a46e06 | 6176 | return NULL; |
2665abfd | 6177 | |
76a46e06 | 6178 | req->flags |= REQ_F_LINK_TIMEOUT; |
76a46e06 | 6179 | return nxt; |
2665abfd JA |
6180 | } |
6181 | ||
c1379e24 | 6182 | static void __io_queue_sqe(struct io_kiocb *req, struct io_comp_state *cs) |
2b188cc1 | 6183 | { |
4a0a7a18 | 6184 | struct io_kiocb *linked_timeout; |
4bc4494e | 6185 | struct io_kiocb *nxt; |
193155c8 | 6186 | const struct cred *old_creds = NULL; |
e0c5c576 | 6187 | int ret; |
2b188cc1 | 6188 | |
4a0a7a18 JA |
6189 | again: |
6190 | linked_timeout = io_prep_linked_timeout(req); | |
6191 | ||
98447d65 JA |
6192 | if ((req->flags & REQ_F_WORK_INITIALIZED) && req->work.identity->creds && |
6193 | req->work.identity->creds != current_cred()) { | |
193155c8 JA |
6194 | if (old_creds) |
6195 | revert_creds(old_creds); | |
98447d65 | 6196 | if (old_creds == req->work.identity->creds) |
193155c8 JA |
6197 | old_creds = NULL; /* restored original creds */ |
6198 | else | |
98447d65 | 6199 | old_creds = override_creds(req->work.identity->creds); |
dfead8a8 | 6200 | req->work.flags |= IO_WQ_WORK_CREDS; |
193155c8 JA |
6201 | } |
6202 | ||
c1379e24 | 6203 | ret = io_issue_sqe(req, true, cs); |
491381ce JA |
6204 | |
6205 | /* | |
6206 | * We async punt it if the file wasn't marked NOWAIT, or if the file | |
6207 | * doesn't support non-blocking read/write attempts | |
6208 | */ | |
24c74678 | 6209 | if (ret == -EAGAIN && !(req->flags & REQ_F_NOWAIT)) { |
f063c547 | 6210 | if (!io_arm_poll_handler(req)) { |
86a761f8 | 6211 | punt: |
f063c547 PB |
6212 | /* |
6213 | * Queued up for async execution, worker will release | |
6214 | * submit reference when the iocb is actually submitted. | |
6215 | */ | |
6216 | io_queue_async_work(req); | |
2b188cc1 | 6217 | } |
bbad27b2 | 6218 | |
f063c547 PB |
6219 | if (linked_timeout) |
6220 | io_queue_linked_timeout(linked_timeout); | |
4bc4494e | 6221 | goto exit; |
2b188cc1 | 6222 | } |
e65ef56d | 6223 | |
652532ad | 6224 | if (unlikely(ret)) { |
652532ad PB |
6225 | /* un-prep timeout, so it'll be killed as any other linked */ |
6226 | req->flags &= ~REQ_F_LINK_TIMEOUT; | |
4e88d6e7 | 6227 | req_set_fail_links(req); |
e65ef56d | 6228 | io_put_req(req); |
e1e16097 | 6229 | io_req_complete(req, ret); |
652532ad | 6230 | goto exit; |
9e645e11 | 6231 | } |
652532ad PB |
6232 | |
6233 | /* drop submission reference */ | |
6234 | nxt = io_put_req_find_next(req); | |
6235 | if (linked_timeout) | |
6236 | io_queue_linked_timeout(linked_timeout); | |
6237 | ||
4a0a7a18 JA |
6238 | if (nxt) { |
6239 | req = nxt; | |
86a761f8 PB |
6240 | |
6241 | if (req->flags & REQ_F_FORCE_ASYNC) | |
6242 | goto punt; | |
4a0a7a18 JA |
6243 | goto again; |
6244 | } | |
4bc4494e | 6245 | exit: |
193155c8 JA |
6246 | if (old_creds) |
6247 | revert_creds(old_creds); | |
2b188cc1 JA |
6248 | } |
6249 | ||
f13fad7b JA |
6250 | static void io_queue_sqe(struct io_kiocb *req, const struct io_uring_sqe *sqe, |
6251 | struct io_comp_state *cs) | |
4fe2c963 JL |
6252 | { |
6253 | int ret; | |
6254 | ||
3529d8c2 | 6255 | ret = io_req_defer(req, sqe); |
4fe2c963 JL |
6256 | if (ret) { |
6257 | if (ret != -EIOCBQUEUED) { | |
1118591a | 6258 | fail_req: |
4e88d6e7 | 6259 | req_set_fail_links(req); |
e1e16097 JA |
6260 | io_put_req(req); |
6261 | io_req_complete(req, ret); | |
4fe2c963 | 6262 | } |
2550878f | 6263 | } else if (req->flags & REQ_F_FORCE_ASYNC) { |
e8c2bc1f | 6264 | if (!req->async_data) { |
bd2ab18a | 6265 | ret = io_req_defer_prep(req, sqe); |
327d6d96 | 6266 | if (unlikely(ret)) |
bd2ab18a PB |
6267 | goto fail_req; |
6268 | } | |
6269 | ||
ce35a47a JA |
6270 | /* |
6271 | * Never try inline submit of IOSQE_ASYNC is set, go straight | |
6272 | * to async execution. | |
6273 | */ | |
3e863ea3 | 6274 | io_req_init_async(req); |
ce35a47a JA |
6275 | req->work.flags |= IO_WQ_WORK_CONCURRENT; |
6276 | io_queue_async_work(req); | |
6277 | } else { | |
c1379e24 PB |
6278 | if (sqe) { |
6279 | ret = io_req_prep(req, sqe); | |
6280 | if (unlikely(ret)) | |
6281 | goto fail_req; | |
6282 | } | |
6283 | __io_queue_sqe(req, cs); | |
ce35a47a | 6284 | } |
4fe2c963 JL |
6285 | } |
6286 | ||
f13fad7b JA |
6287 | static inline void io_queue_link_head(struct io_kiocb *req, |
6288 | struct io_comp_state *cs) | |
4fe2c963 | 6289 | { |
94ae5e77 | 6290 | if (unlikely(req->flags & REQ_F_FAIL_LINK)) { |
e1e16097 JA |
6291 | io_put_req(req); |
6292 | io_req_complete(req, -ECANCELED); | |
1b4a51b6 | 6293 | } else |
f13fad7b | 6294 | io_queue_sqe(req, NULL, cs); |
4fe2c963 JL |
6295 | } |
6296 | ||
1d4240cc | 6297 | static int io_submit_sqe(struct io_kiocb *req, const struct io_uring_sqe *sqe, |
f13fad7b | 6298 | struct io_kiocb **link, struct io_comp_state *cs) |
9e645e11 | 6299 | { |
a197f664 | 6300 | struct io_ring_ctx *ctx = req->ctx; |
ef4ff581 | 6301 | int ret; |
9e645e11 | 6302 | |
9e645e11 JA |
6303 | /* |
6304 | * If we already have a head request, queue this one for async | |
6305 | * submittal once the head completes. If we don't have a head but | |
6306 | * IOSQE_IO_LINK is set in the sqe, start a new head. This one will be | |
6307 | * submitted sync once the chain is complete. If none of those | |
6308 | * conditions are true (normal request), then just queue it. | |
6309 | */ | |
6310 | if (*link) { | |
9d76377f | 6311 | struct io_kiocb *head = *link; |
4e88d6e7 | 6312 | |
8cdf2193 PB |
6313 | /* |
6314 | * Taking sequential execution of a link, draining both sides | |
6315 | * of the link also fullfils IOSQE_IO_DRAIN semantics for all | |
6316 | * requests in the link. So, it drains the head and the | |
6317 | * next after the link request. The last one is done via | |
6318 | * drain_next flag to persist the effect across calls. | |
6319 | */ | |
ef4ff581 | 6320 | if (req->flags & REQ_F_IO_DRAIN) { |
711be031 PB |
6321 | head->flags |= REQ_F_IO_DRAIN; |
6322 | ctx->drain_next = 1; | |
6323 | } | |
3529d8c2 | 6324 | ret = io_req_defer_prep(req, sqe); |
327d6d96 | 6325 | if (unlikely(ret)) { |
4e88d6e7 | 6326 | /* fail even hard links since we don't submit */ |
9d76377f | 6327 | head->flags |= REQ_F_FAIL_LINK; |
1d4240cc | 6328 | return ret; |
2d28390a | 6329 | } |
9d76377f PB |
6330 | trace_io_uring_link(ctx, req, head); |
6331 | list_add_tail(&req->link_list, &head->link_list); | |
32fe525b PB |
6332 | |
6333 | /* last request of a link, enqueue the link */ | |
ef4ff581 | 6334 | if (!(req->flags & (REQ_F_LINK | REQ_F_HARDLINK))) { |
f13fad7b | 6335 | io_queue_link_head(head, cs); |
32fe525b PB |
6336 | *link = NULL; |
6337 | } | |
9e645e11 | 6338 | } else { |
711be031 PB |
6339 | if (unlikely(ctx->drain_next)) { |
6340 | req->flags |= REQ_F_IO_DRAIN; | |
ef4ff581 | 6341 | ctx->drain_next = 0; |
711be031 | 6342 | } |
ef4ff581 | 6343 | if (req->flags & (REQ_F_LINK | REQ_F_HARDLINK)) { |
dea3b49c | 6344 | req->flags |= REQ_F_LINK_HEAD; |
711be031 | 6345 | INIT_LIST_HEAD(&req->link_list); |
f1d96a8f | 6346 | |
711be031 | 6347 | ret = io_req_defer_prep(req, sqe); |
327d6d96 | 6348 | if (unlikely(ret)) |
711be031 PB |
6349 | req->flags |= REQ_F_FAIL_LINK; |
6350 | *link = req; | |
6351 | } else { | |
f13fad7b | 6352 | io_queue_sqe(req, sqe, cs); |
711be031 | 6353 | } |
9e645e11 | 6354 | } |
2e6e1fde | 6355 | |
1d4240cc | 6356 | return 0; |
9e645e11 JA |
6357 | } |
6358 | ||
9a56a232 JA |
6359 | /* |
6360 | * Batched submission is done, ensure local IO is flushed out. | |
6361 | */ | |
6362 | static void io_submit_state_end(struct io_submit_state *state) | |
6363 | { | |
f13fad7b JA |
6364 | if (!list_empty(&state->comp.list)) |
6365 | io_submit_flush_completions(&state->comp); | |
9a56a232 | 6366 | blk_finish_plug(&state->plug); |
9f13c35b | 6367 | io_state_file_put(state); |
2579f913 | 6368 | if (state->free_reqs) |
6c8a3134 | 6369 | kmem_cache_free_bulk(req_cachep, state->free_reqs, state->reqs); |
9a56a232 JA |
6370 | } |
6371 | ||
6372 | /* | |
6373 | * Start submission side cache. | |
6374 | */ | |
6375 | static void io_submit_state_start(struct io_submit_state *state, | |
013538bd | 6376 | struct io_ring_ctx *ctx, unsigned int max_ios) |
9a56a232 JA |
6377 | { |
6378 | blk_start_plug(&state->plug); | |
013538bd JA |
6379 | state->comp.nr = 0; |
6380 | INIT_LIST_HEAD(&state->comp.list); | |
6381 | state->comp.ctx = ctx; | |
2579f913 | 6382 | state->free_reqs = 0; |
9a56a232 JA |
6383 | state->file = NULL; |
6384 | state->ios_left = max_ios; | |
6385 | } | |
6386 | ||
2b188cc1 JA |
6387 | static void io_commit_sqring(struct io_ring_ctx *ctx) |
6388 | { | |
75b28aff | 6389 | struct io_rings *rings = ctx->rings; |
2b188cc1 | 6390 | |
caf582c6 PB |
6391 | /* |
6392 | * Ensure any loads from the SQEs are done at this point, | |
6393 | * since once we write the new head, the application could | |
6394 | * write new data to them. | |
6395 | */ | |
6396 | smp_store_release(&rings->sq.head, ctx->cached_sq_head); | |
2b188cc1 JA |
6397 | } |
6398 | ||
2b188cc1 | 6399 | /* |
3529d8c2 | 6400 | * Fetch an sqe, if one is available. Note that sqe_ptr will point to memory |
2b188cc1 JA |
6401 | * that is mapped by userspace. This means that care needs to be taken to |
6402 | * ensure that reads are stable, as we cannot rely on userspace always | |
6403 | * being a good citizen. If members of the sqe are validated and then later | |
6404 | * used, it's important that those reads are done through READ_ONCE() to | |
6405 | * prevent a re-load down the line. | |
6406 | */ | |
709b302f | 6407 | static const struct io_uring_sqe *io_get_sqe(struct io_ring_ctx *ctx) |
2b188cc1 | 6408 | { |
75b28aff | 6409 | u32 *sq_array = ctx->sq_array; |
2b188cc1 JA |
6410 | unsigned head; |
6411 | ||
6412 | /* | |
6413 | * The cached sq head (or cq tail) serves two purposes: | |
6414 | * | |
6415 | * 1) allows us to batch the cost of updating the user visible | |
6416 | * head updates. | |
6417 | * 2) allows the kernel side to track the head on its own, even | |
6418 | * though the application is the one updating it. | |
6419 | */ | |
ee7d46d9 | 6420 | head = READ_ONCE(sq_array[ctx->cached_sq_head & ctx->sq_mask]); |
709b302f PB |
6421 | if (likely(head < ctx->sq_entries)) |
6422 | return &ctx->sq_sqes[head]; | |
2b188cc1 JA |
6423 | |
6424 | /* drop invalid entries */ | |
498ccd9e | 6425 | ctx->cached_sq_dropped++; |
ee7d46d9 | 6426 | WRITE_ONCE(ctx->rings->sq_dropped, ctx->cached_sq_dropped); |
709b302f PB |
6427 | return NULL; |
6428 | } | |
6429 | ||
6430 | static inline void io_consume_sqe(struct io_ring_ctx *ctx) | |
6431 | { | |
6432 | ctx->cached_sq_head++; | |
2b188cc1 JA |
6433 | } |
6434 | ||
21b55dbc SG |
6435 | /* |
6436 | * Check SQE restrictions (opcode and flags). | |
6437 | * | |
6438 | * Returns 'true' if SQE is allowed, 'false' otherwise. | |
6439 | */ | |
6440 | static inline bool io_check_restriction(struct io_ring_ctx *ctx, | |
6441 | struct io_kiocb *req, | |
6442 | unsigned int sqe_flags) | |
6443 | { | |
6444 | if (!ctx->restricted) | |
6445 | return true; | |
6446 | ||
6447 | if (!test_bit(req->opcode, ctx->restrictions.sqe_op)) | |
6448 | return false; | |
6449 | ||
6450 | if ((sqe_flags & ctx->restrictions.sqe_flags_required) != | |
6451 | ctx->restrictions.sqe_flags_required) | |
6452 | return false; | |
6453 | ||
6454 | if (sqe_flags & ~(ctx->restrictions.sqe_flags_allowed | | |
6455 | ctx->restrictions.sqe_flags_required)) | |
6456 | return false; | |
6457 | ||
6458 | return true; | |
6459 | } | |
6460 | ||
ef4ff581 PB |
6461 | #define SQE_VALID_FLAGS (IOSQE_FIXED_FILE|IOSQE_IO_DRAIN|IOSQE_IO_LINK| \ |
6462 | IOSQE_IO_HARDLINK | IOSQE_ASYNC | \ | |
6463 | IOSQE_BUFFER_SELECT) | |
6464 | ||
6465 | static int io_init_req(struct io_ring_ctx *ctx, struct io_kiocb *req, | |
6466 | const struct io_uring_sqe *sqe, | |
0cdaf760 | 6467 | struct io_submit_state *state) |
0553b8bd | 6468 | { |
ef4ff581 | 6469 | unsigned int sqe_flags; |
71b547c0 | 6470 | int id, ret; |
ef4ff581 | 6471 | |
0553b8bd PB |
6472 | req->opcode = READ_ONCE(sqe->opcode); |
6473 | req->user_data = READ_ONCE(sqe->user_data); | |
e8c2bc1f | 6474 | req->async_data = NULL; |
0553b8bd PB |
6475 | req->file = NULL; |
6476 | req->ctx = ctx; | |
6477 | req->flags = 0; | |
6478 | /* one is dropped after submission, the other at completion */ | |
6479 | refcount_set(&req->refs, 2); | |
4dd2824d | 6480 | req->task = current; |
0553b8bd | 6481 | req->result = 0; |
ef4ff581 PB |
6482 | |
6483 | if (unlikely(req->opcode >= IORING_OP_LAST)) | |
6484 | return -EINVAL; | |
6485 | ||
9d8426a0 JA |
6486 | if (unlikely(io_sq_thread_acquire_mm(ctx, req))) |
6487 | return -EFAULT; | |
ef4ff581 PB |
6488 | |
6489 | sqe_flags = READ_ONCE(sqe->flags); | |
6490 | /* enforce forwards compatibility on users */ | |
6491 | if (unlikely(sqe_flags & ~SQE_VALID_FLAGS)) | |
6492 | return -EINVAL; | |
6493 | ||
21b55dbc SG |
6494 | if (unlikely(!io_check_restriction(ctx, req, sqe_flags))) |
6495 | return -EACCES; | |
6496 | ||
ef4ff581 PB |
6497 | if ((sqe_flags & IOSQE_BUFFER_SELECT) && |
6498 | !io_op_defs[req->opcode].buffer_select) | |
6499 | return -EOPNOTSUPP; | |
6500 | ||
6501 | id = READ_ONCE(sqe->personality); | |
6502 | if (id) { | |
1e6fa521 JA |
6503 | struct io_identity *iod; |
6504 | ||
7cdaf587 | 6505 | io_req_init_async(req); |
1e6fa521 JA |
6506 | iod = idr_find(&ctx->personality_idr, id); |
6507 | if (unlikely(!iod)) | |
ef4ff581 | 6508 | return -EINVAL; |
1e6fa521 | 6509 | refcount_inc(&iod->count); |
5c3462cf | 6510 | io_put_identity(current->io_uring, req); |
1e6fa521 JA |
6511 | get_cred(iod->creds); |
6512 | req->work.identity = iod; | |
dfead8a8 | 6513 | req->work.flags |= IO_WQ_WORK_CREDS; |
ef4ff581 PB |
6514 | } |
6515 | ||
6516 | /* same numerical values with corresponding REQ_F_*, safe to copy */ | |
c11368a5 | 6517 | req->flags |= sqe_flags; |
ef4ff581 | 6518 | |
63ff8223 JA |
6519 | if (!io_op_defs[req->opcode].needs_file) |
6520 | return 0; | |
6521 | ||
71b547c0 PB |
6522 | ret = io_req_set_file(state, req, READ_ONCE(sqe->fd)); |
6523 | state->ios_left--; | |
6524 | return ret; | |
0553b8bd PB |
6525 | } |
6526 | ||
0f212204 | 6527 | static int io_submit_sqes(struct io_ring_ctx *ctx, unsigned int nr) |
6c271ce2 | 6528 | { |
ac8691c4 | 6529 | struct io_submit_state state; |
9e645e11 | 6530 | struct io_kiocb *link = NULL; |
9e645e11 | 6531 | int i, submitted = 0; |
6c271ce2 | 6532 | |
c4a2ed72 | 6533 | /* if we have a backlog and couldn't flush it all, return BUSY */ |
ad3eb2c8 JA |
6534 | if (test_bit(0, &ctx->sq_check_overflow)) { |
6535 | if (!list_empty(&ctx->cq_overflow_list) && | |
e6c8aa9a | 6536 | !io_cqring_overflow_flush(ctx, false, NULL, NULL)) |
ad3eb2c8 JA |
6537 | return -EBUSY; |
6538 | } | |
6c271ce2 | 6539 | |
ee7d46d9 PB |
6540 | /* make sure SQ entry isn't read before tail */ |
6541 | nr = min3(nr, ctx->sq_entries, io_sqring_entries(ctx)); | |
9ef4f124 | 6542 | |
2b85edfc PB |
6543 | if (!percpu_ref_tryget_many(&ctx->refs, nr)) |
6544 | return -EAGAIN; | |
6c271ce2 | 6545 | |
d8a6df10 | 6546 | percpu_counter_add(¤t->io_uring->inflight, nr); |
faf7b51c | 6547 | refcount_add(nr, ¤t->usage); |
6c271ce2 | 6548 | |
013538bd | 6549 | io_submit_state_start(&state, ctx, nr); |
b14cca0c | 6550 | |
6c271ce2 | 6551 | for (i = 0; i < nr; i++) { |
3529d8c2 | 6552 | const struct io_uring_sqe *sqe; |
196be95c | 6553 | struct io_kiocb *req; |
1cb1edb2 | 6554 | int err; |
fb5ccc98 | 6555 | |
b1e50e54 PB |
6556 | sqe = io_get_sqe(ctx); |
6557 | if (unlikely(!sqe)) { | |
6558 | io_consume_sqe(ctx); | |
6559 | break; | |
6560 | } | |
ac8691c4 | 6561 | req = io_alloc_req(ctx, &state); |
196be95c PB |
6562 | if (unlikely(!req)) { |
6563 | if (!submitted) | |
6564 | submitted = -EAGAIN; | |
fb5ccc98 | 6565 | break; |
196be95c | 6566 | } |
709b302f | 6567 | io_consume_sqe(ctx); |
d3656344 JA |
6568 | /* will complete beyond this point, count as submitted */ |
6569 | submitted++; | |
6570 | ||
692d8363 | 6571 | err = io_init_req(ctx, req, sqe, &state); |
ef4ff581 | 6572 | if (unlikely(err)) { |
1cb1edb2 | 6573 | fail_req: |
e1e16097 JA |
6574 | io_put_req(req); |
6575 | io_req_complete(req, err); | |
196be95c PB |
6576 | break; |
6577 | } | |
fb5ccc98 | 6578 | |
354420f7 | 6579 | trace_io_uring_submit_sqe(ctx, req->opcode, req->user_data, |
0cdaf760 | 6580 | true, io_async_submit(ctx)); |
f13fad7b | 6581 | err = io_submit_sqe(req, sqe, &link, &state.comp); |
1d4240cc PB |
6582 | if (err) |
6583 | goto fail_req; | |
6c271ce2 JA |
6584 | } |
6585 | ||
9466f437 PB |
6586 | if (unlikely(submitted != nr)) { |
6587 | int ref_used = (submitted == -EAGAIN) ? 0 : submitted; | |
d8a6df10 JA |
6588 | struct io_uring_task *tctx = current->io_uring; |
6589 | int unused = nr - ref_used; | |
9466f437 | 6590 | |
d8a6df10 JA |
6591 | percpu_ref_put_many(&ctx->refs, unused); |
6592 | percpu_counter_sub(&tctx->inflight, unused); | |
6593 | put_task_struct_many(current, unused); | |
9466f437 | 6594 | } |
9e645e11 | 6595 | if (link) |
f13fad7b | 6596 | io_queue_link_head(link, &state.comp); |
ac8691c4 | 6597 | io_submit_state_end(&state); |
6c271ce2 | 6598 | |
ae9428ca PB |
6599 | /* Commit SQ ring head once we've consumed and submitted all SQEs */ |
6600 | io_commit_sqring(ctx); | |
6601 | ||
6c271ce2 JA |
6602 | return submitted; |
6603 | } | |
6604 | ||
23b3628e XW |
6605 | static inline void io_ring_set_wakeup_flag(struct io_ring_ctx *ctx) |
6606 | { | |
6607 | /* Tell userspace we may need a wakeup call */ | |
6608 | spin_lock_irq(&ctx->completion_lock); | |
6609 | ctx->rings->sq_flags |= IORING_SQ_NEED_WAKEUP; | |
6610 | spin_unlock_irq(&ctx->completion_lock); | |
6611 | } | |
6612 | ||
6613 | static inline void io_ring_clear_wakeup_flag(struct io_ring_ctx *ctx) | |
6614 | { | |
6615 | spin_lock_irq(&ctx->completion_lock); | |
6616 | ctx->rings->sq_flags &= ~IORING_SQ_NEED_WAKEUP; | |
6617 | spin_unlock_irq(&ctx->completion_lock); | |
6618 | } | |
6619 | ||
3f0e64d0 JA |
6620 | static int io_sq_wake_function(struct wait_queue_entry *wqe, unsigned mode, |
6621 | int sync, void *key) | |
6c271ce2 | 6622 | { |
3f0e64d0 JA |
6623 | struct io_ring_ctx *ctx = container_of(wqe, struct io_ring_ctx, sqo_wait_entry); |
6624 | int ret; | |
6625 | ||
6626 | ret = autoremove_wake_function(wqe, mode, sync, key); | |
6627 | if (ret) { | |
6628 | unsigned long flags; | |
6629 | ||
6630 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
6631 | ctx->rings->sq_flags &= ~IORING_SQ_NEED_WAKEUP; | |
6632 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
6633 | } | |
6634 | return ret; | |
6635 | } | |
6636 | ||
c8d1ba58 JA |
6637 | enum sq_ret { |
6638 | SQT_IDLE = 1, | |
6639 | SQT_SPIN = 2, | |
6640 | SQT_DID_WORK = 4, | |
6641 | }; | |
6642 | ||
6643 | static enum sq_ret __io_sq_thread(struct io_ring_ctx *ctx, | |
e95eee2d | 6644 | unsigned long start_jiffies, bool cap_entries) |
6c271ce2 | 6645 | { |
c8d1ba58 | 6646 | unsigned long timeout = start_jiffies + ctx->sq_thread_idle; |
534ca6d6 | 6647 | struct io_sq_data *sqd = ctx->sq_data; |
c8d1ba58 | 6648 | unsigned int to_submit; |
bdcd3eab | 6649 | int ret = 0; |
6c271ce2 | 6650 | |
c8d1ba58 JA |
6651 | again: |
6652 | if (!list_empty(&ctx->iopoll_list)) { | |
6653 | unsigned nr_events = 0; | |
a4c0b3de | 6654 | |
c8d1ba58 JA |
6655 | mutex_lock(&ctx->uring_lock); |
6656 | if (!list_empty(&ctx->iopoll_list) && !need_resched()) | |
6657 | io_do_iopoll(ctx, &nr_events, 0); | |
6658 | mutex_unlock(&ctx->uring_lock); | |
6659 | } | |
6c271ce2 | 6660 | |
c8d1ba58 | 6661 | to_submit = io_sqring_entries(ctx); |
6c271ce2 | 6662 | |
c8d1ba58 JA |
6663 | /* |
6664 | * If submit got -EBUSY, flag us as needing the application | |
6665 | * to enter the kernel to reap and flush events. | |
6666 | */ | |
6667 | if (!to_submit || ret == -EBUSY || need_resched()) { | |
6668 | /* | |
6669 | * Drop cur_mm before scheduling, we can't hold it for | |
6670 | * long periods (or over schedule()). Do this before | |
6671 | * adding ourselves to the waitqueue, as the unuse/drop | |
6672 | * may sleep. | |
6673 | */ | |
6674 | io_sq_thread_drop_mm(); | |
6c271ce2 | 6675 | |
c8d1ba58 JA |
6676 | /* |
6677 | * We're polling. If we're within the defined idle | |
6678 | * period, then let us spin without work before going | |
6679 | * to sleep. The exception is if we got EBUSY doing | |
6680 | * more IO, we should wait for the application to | |
6681 | * reap events and wake us up. | |
6682 | */ | |
6683 | if (!list_empty(&ctx->iopoll_list) || need_resched() || | |
6684 | (!time_after(jiffies, timeout) && ret != -EBUSY && | |
6685 | !percpu_ref_is_dying(&ctx->refs))) | |
6686 | return SQT_SPIN; | |
6c271ce2 | 6687 | |
534ca6d6 | 6688 | prepare_to_wait(&sqd->wait, &ctx->sqo_wait_entry, |
c8d1ba58 | 6689 | TASK_INTERRUPTIBLE); |
6c271ce2 | 6690 | |
c8d1ba58 JA |
6691 | /* |
6692 | * While doing polled IO, before going to sleep, we need | |
6693 | * to check if there are new reqs added to iopoll_list, | |
6694 | * it is because reqs may have been punted to io worker | |
6695 | * and will be added to iopoll_list later, hence check | |
6696 | * the iopoll_list again. | |
6697 | */ | |
6698 | if ((ctx->flags & IORING_SETUP_IOPOLL) && | |
6699 | !list_empty_careful(&ctx->iopoll_list)) { | |
534ca6d6 | 6700 | finish_wait(&sqd->wait, &ctx->sqo_wait_entry); |
c8d1ba58 | 6701 | goto again; |
6c271ce2 JA |
6702 | } |
6703 | ||
fb5ccc98 | 6704 | to_submit = io_sqring_entries(ctx); |
c8d1ba58 JA |
6705 | if (!to_submit || ret == -EBUSY) |
6706 | return SQT_IDLE; | |
6707 | } | |
c1edbf5f | 6708 | |
534ca6d6 | 6709 | finish_wait(&sqd->wait, &ctx->sqo_wait_entry); |
c8d1ba58 | 6710 | io_ring_clear_wakeup_flag(ctx); |
7143b5ac | 6711 | |
e95eee2d JA |
6712 | /* if we're handling multiple rings, cap submit size for fairness */ |
6713 | if (cap_entries && to_submit > 8) | |
6714 | to_submit = 8; | |
6715 | ||
c8d1ba58 JA |
6716 | mutex_lock(&ctx->uring_lock); |
6717 | if (likely(!percpu_ref_is_dying(&ctx->refs))) | |
6718 | ret = io_submit_sqes(ctx, to_submit); | |
6719 | mutex_unlock(&ctx->uring_lock); | |
90554200 JA |
6720 | |
6721 | if (!io_sqring_full(ctx) && wq_has_sleeper(&ctx->sqo_sq_wait)) | |
6722 | wake_up(&ctx->sqo_sq_wait); | |
6723 | ||
c8d1ba58 JA |
6724 | return SQT_DID_WORK; |
6725 | } | |
6c271ce2 | 6726 | |
69fb2131 JA |
6727 | static void io_sqd_init_new(struct io_sq_data *sqd) |
6728 | { | |
6729 | struct io_ring_ctx *ctx; | |
6730 | ||
6731 | while (!list_empty(&sqd->ctx_new_list)) { | |
6732 | ctx = list_first_entry(&sqd->ctx_new_list, struct io_ring_ctx, sqd_list); | |
6733 | init_wait(&ctx->sqo_wait_entry); | |
6734 | ctx->sqo_wait_entry.func = io_sq_wake_function; | |
6735 | list_move_tail(&ctx->sqd_list, &sqd->ctx_list); | |
6736 | complete(&ctx->sq_thread_comp); | |
6737 | } | |
6738 | } | |
6739 | ||
c8d1ba58 JA |
6740 | static int io_sq_thread(void *data) |
6741 | { | |
91d8f519 | 6742 | struct cgroup_subsys_state *cur_css = NULL; |
69fb2131 JA |
6743 | const struct cred *old_cred = NULL; |
6744 | struct io_sq_data *sqd = data; | |
6745 | struct io_ring_ctx *ctx; | |
c8d1ba58 | 6746 | unsigned long start_jiffies; |
6c271ce2 | 6747 | |
69fb2131 JA |
6748 | start_jiffies = jiffies; |
6749 | while (!kthread_should_stop()) { | |
6750 | enum sq_ret ret = 0; | |
e95eee2d | 6751 | bool cap_entries; |
c1edbf5f JA |
6752 | |
6753 | /* | |
69fb2131 JA |
6754 | * Any changes to the sqd lists are synchronized through the |
6755 | * kthread parking. This synchronizes the thread vs users, | |
6756 | * the users are synchronized on the sqd->ctx_lock. | |
c1edbf5f | 6757 | */ |
69fb2131 JA |
6758 | if (kthread_should_park()) |
6759 | kthread_parkme(); | |
7143b5ac | 6760 | |
69fb2131 JA |
6761 | if (unlikely(!list_empty(&sqd->ctx_new_list))) |
6762 | io_sqd_init_new(sqd); | |
6c271ce2 | 6763 | |
e95eee2d | 6764 | cap_entries = !list_is_singular(&sqd->ctx_list); |
6c271ce2 | 6765 | |
69fb2131 JA |
6766 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) { |
6767 | if (current->cred != ctx->creds) { | |
6768 | if (old_cred) | |
6769 | revert_creds(old_cred); | |
6770 | old_cred = override_creds(ctx->creds); | |
bdcd3eab | 6771 | } |
91d8f519 | 6772 | io_sq_thread_associate_blkcg(ctx, &cur_css); |
4ea33a97 JA |
6773 | #ifdef CONFIG_AUDIT |
6774 | current->loginuid = ctx->loginuid; | |
6775 | current->sessionid = ctx->sessionid; | |
6776 | #endif | |
bdcd3eab | 6777 | |
e95eee2d | 6778 | ret |= __io_sq_thread(ctx, start_jiffies, cap_entries); |
6c271ce2 | 6779 | |
69fb2131 JA |
6780 | io_sq_thread_drop_mm(); |
6781 | } | |
6c271ce2 | 6782 | |
69fb2131 | 6783 | if (ret & SQT_SPIN) { |
c8d1ba58 JA |
6784 | io_run_task_work(); |
6785 | cond_resched(); | |
69fb2131 JA |
6786 | } else if (ret == SQT_IDLE) { |
6787 | if (kthread_should_park()) | |
6c271ce2 | 6788 | continue; |
69fb2131 JA |
6789 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) |
6790 | io_ring_set_wakeup_flag(ctx); | |
6791 | schedule(); | |
6792 | start_jiffies = jiffies; | |
6793 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) | |
6794 | io_ring_clear_wakeup_flag(ctx); | |
6c271ce2 | 6795 | } |
6c271ce2 JA |
6796 | } |
6797 | ||
4c6e277c | 6798 | io_run_task_work(); |
b41e9852 | 6799 | |
91d8f519 DZ |
6800 | if (cur_css) |
6801 | io_sq_thread_unassociate_blkcg(); | |
69fb2131 JA |
6802 | if (old_cred) |
6803 | revert_creds(old_cred); | |
06058632 | 6804 | |
2bbcd6d3 | 6805 | kthread_parkme(); |
06058632 | 6806 | |
6c271ce2 JA |
6807 | return 0; |
6808 | } | |
6809 | ||
bda52162 JA |
6810 | struct io_wait_queue { |
6811 | struct wait_queue_entry wq; | |
6812 | struct io_ring_ctx *ctx; | |
6813 | unsigned to_wait; | |
6814 | unsigned nr_timeouts; | |
6815 | }; | |
6816 | ||
1d7bb1d5 | 6817 | static inline bool io_should_wake(struct io_wait_queue *iowq, bool noflush) |
bda52162 JA |
6818 | { |
6819 | struct io_ring_ctx *ctx = iowq->ctx; | |
6820 | ||
6821 | /* | |
d195a66e | 6822 | * Wake up if we have enough events, or if a timeout occurred since we |
bda52162 JA |
6823 | * started waiting. For timeouts, we always want to return to userspace, |
6824 | * regardless of event count. | |
6825 | */ | |
1d7bb1d5 | 6826 | return io_cqring_events(ctx, noflush) >= iowq->to_wait || |
bda52162 JA |
6827 | atomic_read(&ctx->cq_timeouts) != iowq->nr_timeouts; |
6828 | } | |
6829 | ||
6830 | static int io_wake_function(struct wait_queue_entry *curr, unsigned int mode, | |
6831 | int wake_flags, void *key) | |
6832 | { | |
6833 | struct io_wait_queue *iowq = container_of(curr, struct io_wait_queue, | |
6834 | wq); | |
6835 | ||
1d7bb1d5 JA |
6836 | /* use noflush == true, as we can't safely rely on locking context */ |
6837 | if (!io_should_wake(iowq, true)) | |
bda52162 JA |
6838 | return -1; |
6839 | ||
6840 | return autoremove_wake_function(curr, mode, wake_flags, key); | |
6841 | } | |
6842 | ||
af9c1a44 JA |
6843 | static int io_run_task_work_sig(void) |
6844 | { | |
6845 | if (io_run_task_work()) | |
6846 | return 1; | |
6847 | if (!signal_pending(current)) | |
6848 | return 0; | |
6849 | if (current->jobctl & JOBCTL_TASK_WORK) { | |
6850 | spin_lock_irq(¤t->sighand->siglock); | |
6851 | current->jobctl &= ~JOBCTL_TASK_WORK; | |
6852 | recalc_sigpending(); | |
6853 | spin_unlock_irq(¤t->sighand->siglock); | |
6854 | return 1; | |
6855 | } | |
6856 | return -EINTR; | |
6857 | } | |
6858 | ||
2b188cc1 JA |
6859 | /* |
6860 | * Wait until events become available, if we don't already have some. The | |
6861 | * application must reap them itself, as they reside on the shared cq ring. | |
6862 | */ | |
6863 | static int io_cqring_wait(struct io_ring_ctx *ctx, int min_events, | |
6864 | const sigset_t __user *sig, size_t sigsz) | |
6865 | { | |
bda52162 JA |
6866 | struct io_wait_queue iowq = { |
6867 | .wq = { | |
6868 | .private = current, | |
6869 | .func = io_wake_function, | |
6870 | .entry = LIST_HEAD_INIT(iowq.wq.entry), | |
6871 | }, | |
6872 | .ctx = ctx, | |
6873 | .to_wait = min_events, | |
6874 | }; | |
75b28aff | 6875 | struct io_rings *rings = ctx->rings; |
e9ffa5c2 | 6876 | int ret = 0; |
2b188cc1 | 6877 | |
b41e9852 JA |
6878 | do { |
6879 | if (io_cqring_events(ctx, false) >= min_events) | |
6880 | return 0; | |
4c6e277c | 6881 | if (!io_run_task_work()) |
b41e9852 | 6882 | break; |
b41e9852 | 6883 | } while (1); |
2b188cc1 JA |
6884 | |
6885 | if (sig) { | |
9e75ad5d AB |
6886 | #ifdef CONFIG_COMPAT |
6887 | if (in_compat_syscall()) | |
6888 | ret = set_compat_user_sigmask((const compat_sigset_t __user *)sig, | |
b772434b | 6889 | sigsz); |
9e75ad5d AB |
6890 | else |
6891 | #endif | |
b772434b | 6892 | ret = set_user_sigmask(sig, sigsz); |
9e75ad5d | 6893 | |
2b188cc1 JA |
6894 | if (ret) |
6895 | return ret; | |
6896 | } | |
6897 | ||
bda52162 | 6898 | iowq.nr_timeouts = atomic_read(&ctx->cq_timeouts); |
c826bd7a | 6899 | trace_io_uring_cqring_wait(ctx, min_events); |
bda52162 JA |
6900 | do { |
6901 | prepare_to_wait_exclusive(&ctx->wait, &iowq.wq, | |
6902 | TASK_INTERRUPTIBLE); | |
ce593a6c | 6903 | /* make sure we run task_work before checking for signals */ |
af9c1a44 JA |
6904 | ret = io_run_task_work_sig(); |
6905 | if (ret > 0) | |
4c6e277c | 6906 | continue; |
af9c1a44 | 6907 | else if (ret < 0) |
bda52162 | 6908 | break; |
ce593a6c JA |
6909 | if (io_should_wake(&iowq, false)) |
6910 | break; | |
6911 | schedule(); | |
bda52162 JA |
6912 | } while (1); |
6913 | finish_wait(&ctx->wait, &iowq.wq); | |
6914 | ||
b7db41c9 | 6915 | restore_saved_sigmask_unless(ret == -EINTR); |
2b188cc1 | 6916 | |
75b28aff | 6917 | return READ_ONCE(rings->cq.head) == READ_ONCE(rings->cq.tail) ? ret : 0; |
2b188cc1 JA |
6918 | } |
6919 | ||
6b06314c JA |
6920 | static void __io_sqe_files_unregister(struct io_ring_ctx *ctx) |
6921 | { | |
6922 | #if defined(CONFIG_UNIX) | |
6923 | if (ctx->ring_sock) { | |
6924 | struct sock *sock = ctx->ring_sock->sk; | |
6925 | struct sk_buff *skb; | |
6926 | ||
6927 | while ((skb = skb_dequeue(&sock->sk_receive_queue)) != NULL) | |
6928 | kfree_skb(skb); | |
6929 | } | |
6930 | #else | |
6931 | int i; | |
6932 | ||
65e19f54 JA |
6933 | for (i = 0; i < ctx->nr_user_files; i++) { |
6934 | struct file *file; | |
6935 | ||
6936 | file = io_file_from_index(ctx, i); | |
6937 | if (file) | |
6938 | fput(file); | |
6939 | } | |
6b06314c JA |
6940 | #endif |
6941 | } | |
6942 | ||
05f3fb3c JA |
6943 | static void io_file_ref_kill(struct percpu_ref *ref) |
6944 | { | |
6945 | struct fixed_file_data *data; | |
6946 | ||
6947 | data = container_of(ref, struct fixed_file_data, refs); | |
6948 | complete(&data->done); | |
6949 | } | |
6950 | ||
6b06314c JA |
6951 | static int io_sqe_files_unregister(struct io_ring_ctx *ctx) |
6952 | { | |
05f3fb3c | 6953 | struct fixed_file_data *data = ctx->file_data; |
05589553 | 6954 | struct fixed_file_ref_node *ref_node = NULL; |
65e19f54 JA |
6955 | unsigned nr_tables, i; |
6956 | ||
05f3fb3c | 6957 | if (!data) |
6b06314c JA |
6958 | return -ENXIO; |
6959 | ||
6a4d07cd | 6960 | spin_lock(&data->lock); |
05589553 XW |
6961 | if (!list_empty(&data->ref_list)) |
6962 | ref_node = list_first_entry(&data->ref_list, | |
6963 | struct fixed_file_ref_node, node); | |
6a4d07cd | 6964 | spin_unlock(&data->lock); |
05589553 XW |
6965 | if (ref_node) |
6966 | percpu_ref_kill(&ref_node->refs); | |
6967 | ||
6968 | percpu_ref_kill(&data->refs); | |
6969 | ||
6970 | /* wait for all refs nodes to complete */ | |
4a38aed2 | 6971 | flush_delayed_work(&ctx->file_put_work); |
2faf852d | 6972 | wait_for_completion(&data->done); |
05f3fb3c | 6973 | |
6b06314c | 6974 | __io_sqe_files_unregister(ctx); |
65e19f54 JA |
6975 | nr_tables = DIV_ROUND_UP(ctx->nr_user_files, IORING_MAX_FILES_TABLE); |
6976 | for (i = 0; i < nr_tables; i++) | |
05f3fb3c JA |
6977 | kfree(data->table[i].files); |
6978 | kfree(data->table); | |
05589553 XW |
6979 | percpu_ref_exit(&data->refs); |
6980 | kfree(data); | |
05f3fb3c | 6981 | ctx->file_data = NULL; |
6b06314c JA |
6982 | ctx->nr_user_files = 0; |
6983 | return 0; | |
6984 | } | |
6985 | ||
534ca6d6 | 6986 | static void io_put_sq_data(struct io_sq_data *sqd) |
6c271ce2 | 6987 | { |
534ca6d6 | 6988 | if (refcount_dec_and_test(&sqd->refs)) { |
2bbcd6d3 RP |
6989 | /* |
6990 | * The park is a bit of a work-around, without it we get | |
6991 | * warning spews on shutdown with SQPOLL set and affinity | |
6992 | * set to a single CPU. | |
6993 | */ | |
534ca6d6 JA |
6994 | if (sqd->thread) { |
6995 | kthread_park(sqd->thread); | |
6996 | kthread_stop(sqd->thread); | |
6997 | } | |
6998 | ||
6999 | kfree(sqd); | |
7000 | } | |
7001 | } | |
7002 | ||
aa06165d JA |
7003 | static struct io_sq_data *io_attach_sq_data(struct io_uring_params *p) |
7004 | { | |
7005 | struct io_ring_ctx *ctx_attach; | |
7006 | struct io_sq_data *sqd; | |
7007 | struct fd f; | |
7008 | ||
7009 | f = fdget(p->wq_fd); | |
7010 | if (!f.file) | |
7011 | return ERR_PTR(-ENXIO); | |
7012 | if (f.file->f_op != &io_uring_fops) { | |
7013 | fdput(f); | |
7014 | return ERR_PTR(-EINVAL); | |
7015 | } | |
7016 | ||
7017 | ctx_attach = f.file->private_data; | |
7018 | sqd = ctx_attach->sq_data; | |
7019 | if (!sqd) { | |
7020 | fdput(f); | |
7021 | return ERR_PTR(-EINVAL); | |
7022 | } | |
7023 | ||
7024 | refcount_inc(&sqd->refs); | |
7025 | fdput(f); | |
7026 | return sqd; | |
7027 | } | |
7028 | ||
534ca6d6 JA |
7029 | static struct io_sq_data *io_get_sq_data(struct io_uring_params *p) |
7030 | { | |
7031 | struct io_sq_data *sqd; | |
7032 | ||
aa06165d JA |
7033 | if (p->flags & IORING_SETUP_ATTACH_WQ) |
7034 | return io_attach_sq_data(p); | |
7035 | ||
534ca6d6 JA |
7036 | sqd = kzalloc(sizeof(*sqd), GFP_KERNEL); |
7037 | if (!sqd) | |
7038 | return ERR_PTR(-ENOMEM); | |
7039 | ||
7040 | refcount_set(&sqd->refs, 1); | |
69fb2131 JA |
7041 | INIT_LIST_HEAD(&sqd->ctx_list); |
7042 | INIT_LIST_HEAD(&sqd->ctx_new_list); | |
7043 | mutex_init(&sqd->ctx_lock); | |
7044 | mutex_init(&sqd->lock); | |
534ca6d6 JA |
7045 | init_waitqueue_head(&sqd->wait); |
7046 | return sqd; | |
7047 | } | |
7048 | ||
69fb2131 JA |
7049 | static void io_sq_thread_unpark(struct io_sq_data *sqd) |
7050 | __releases(&sqd->lock) | |
7051 | { | |
7052 | if (!sqd->thread) | |
7053 | return; | |
7054 | kthread_unpark(sqd->thread); | |
7055 | mutex_unlock(&sqd->lock); | |
7056 | } | |
7057 | ||
7058 | static void io_sq_thread_park(struct io_sq_data *sqd) | |
7059 | __acquires(&sqd->lock) | |
7060 | { | |
7061 | if (!sqd->thread) | |
7062 | return; | |
7063 | mutex_lock(&sqd->lock); | |
7064 | kthread_park(sqd->thread); | |
7065 | } | |
7066 | ||
534ca6d6 JA |
7067 | static void io_sq_thread_stop(struct io_ring_ctx *ctx) |
7068 | { | |
7069 | struct io_sq_data *sqd = ctx->sq_data; | |
7070 | ||
7071 | if (sqd) { | |
7072 | if (sqd->thread) { | |
7073 | /* | |
7074 | * We may arrive here from the error branch in | |
7075 | * io_sq_offload_create() where the kthread is created | |
7076 | * without being waked up, thus wake it up now to make | |
7077 | * sure the wait will complete. | |
7078 | */ | |
7079 | wake_up_process(sqd->thread); | |
7080 | wait_for_completion(&ctx->sq_thread_comp); | |
69fb2131 JA |
7081 | |
7082 | io_sq_thread_park(sqd); | |
7083 | } | |
7084 | ||
7085 | mutex_lock(&sqd->ctx_lock); | |
7086 | list_del(&ctx->sqd_list); | |
7087 | mutex_unlock(&sqd->ctx_lock); | |
7088 | ||
7089 | if (sqd->thread) { | |
7090 | finish_wait(&sqd->wait, &ctx->sqo_wait_entry); | |
7091 | io_sq_thread_unpark(sqd); | |
534ca6d6 JA |
7092 | } |
7093 | ||
7094 | io_put_sq_data(sqd); | |
7095 | ctx->sq_data = NULL; | |
6c271ce2 JA |
7096 | } |
7097 | } | |
7098 | ||
6b06314c JA |
7099 | static void io_finish_async(struct io_ring_ctx *ctx) |
7100 | { | |
6c271ce2 JA |
7101 | io_sq_thread_stop(ctx); |
7102 | ||
561fb04a JA |
7103 | if (ctx->io_wq) { |
7104 | io_wq_destroy(ctx->io_wq); | |
7105 | ctx->io_wq = NULL; | |
6b06314c JA |
7106 | } |
7107 | } | |
7108 | ||
7109 | #if defined(CONFIG_UNIX) | |
6b06314c JA |
7110 | /* |
7111 | * Ensure the UNIX gc is aware of our file set, so we are certain that | |
7112 | * the io_uring can be safely unregistered on process exit, even if we have | |
7113 | * loops in the file referencing. | |
7114 | */ | |
7115 | static int __io_sqe_files_scm(struct io_ring_ctx *ctx, int nr, int offset) | |
7116 | { | |
7117 | struct sock *sk = ctx->ring_sock->sk; | |
7118 | struct scm_fp_list *fpl; | |
7119 | struct sk_buff *skb; | |
08a45173 | 7120 | int i, nr_files; |
6b06314c | 7121 | |
6b06314c JA |
7122 | fpl = kzalloc(sizeof(*fpl), GFP_KERNEL); |
7123 | if (!fpl) | |
7124 | return -ENOMEM; | |
7125 | ||
7126 | skb = alloc_skb(0, GFP_KERNEL); | |
7127 | if (!skb) { | |
7128 | kfree(fpl); | |
7129 | return -ENOMEM; | |
7130 | } | |
7131 | ||
7132 | skb->sk = sk; | |
6b06314c | 7133 | |
08a45173 | 7134 | nr_files = 0; |
6b06314c JA |
7135 | fpl->user = get_uid(ctx->user); |
7136 | for (i = 0; i < nr; i++) { | |
65e19f54 JA |
7137 | struct file *file = io_file_from_index(ctx, i + offset); |
7138 | ||
7139 | if (!file) | |
08a45173 | 7140 | continue; |
65e19f54 | 7141 | fpl->fp[nr_files] = get_file(file); |
08a45173 JA |
7142 | unix_inflight(fpl->user, fpl->fp[nr_files]); |
7143 | nr_files++; | |
6b06314c JA |
7144 | } |
7145 | ||
08a45173 JA |
7146 | if (nr_files) { |
7147 | fpl->max = SCM_MAX_FD; | |
7148 | fpl->count = nr_files; | |
7149 | UNIXCB(skb).fp = fpl; | |
05f3fb3c | 7150 | skb->destructor = unix_destruct_scm; |
08a45173 JA |
7151 | refcount_add(skb->truesize, &sk->sk_wmem_alloc); |
7152 | skb_queue_head(&sk->sk_receive_queue, skb); | |
6b06314c | 7153 | |
08a45173 JA |
7154 | for (i = 0; i < nr_files; i++) |
7155 | fput(fpl->fp[i]); | |
7156 | } else { | |
7157 | kfree_skb(skb); | |
7158 | kfree(fpl); | |
7159 | } | |
6b06314c JA |
7160 | |
7161 | return 0; | |
7162 | } | |
7163 | ||
7164 | /* | |
7165 | * If UNIX sockets are enabled, fd passing can cause a reference cycle which | |
7166 | * causes regular reference counting to break down. We rely on the UNIX | |
7167 | * garbage collection to take care of this problem for us. | |
7168 | */ | |
7169 | static int io_sqe_files_scm(struct io_ring_ctx *ctx) | |
7170 | { | |
7171 | unsigned left, total; | |
7172 | int ret = 0; | |
7173 | ||
7174 | total = 0; | |
7175 | left = ctx->nr_user_files; | |
7176 | while (left) { | |
7177 | unsigned this_files = min_t(unsigned, left, SCM_MAX_FD); | |
6b06314c JA |
7178 | |
7179 | ret = __io_sqe_files_scm(ctx, this_files, total); | |
7180 | if (ret) | |
7181 | break; | |
7182 | left -= this_files; | |
7183 | total += this_files; | |
7184 | } | |
7185 | ||
7186 | if (!ret) | |
7187 | return 0; | |
7188 | ||
7189 | while (total < ctx->nr_user_files) { | |
65e19f54 JA |
7190 | struct file *file = io_file_from_index(ctx, total); |
7191 | ||
7192 | if (file) | |
7193 | fput(file); | |
6b06314c JA |
7194 | total++; |
7195 | } | |
7196 | ||
7197 | return ret; | |
7198 | } | |
7199 | #else | |
7200 | static int io_sqe_files_scm(struct io_ring_ctx *ctx) | |
7201 | { | |
7202 | return 0; | |
7203 | } | |
7204 | #endif | |
7205 | ||
5398ae69 PB |
7206 | static int io_sqe_alloc_file_tables(struct fixed_file_data *file_data, |
7207 | unsigned nr_tables, unsigned nr_files) | |
65e19f54 JA |
7208 | { |
7209 | int i; | |
7210 | ||
7211 | for (i = 0; i < nr_tables; i++) { | |
5398ae69 | 7212 | struct fixed_file_table *table = &file_data->table[i]; |
65e19f54 JA |
7213 | unsigned this_files; |
7214 | ||
7215 | this_files = min(nr_files, IORING_MAX_FILES_TABLE); | |
7216 | table->files = kcalloc(this_files, sizeof(struct file *), | |
7217 | GFP_KERNEL); | |
7218 | if (!table->files) | |
7219 | break; | |
7220 | nr_files -= this_files; | |
7221 | } | |
7222 | ||
7223 | if (i == nr_tables) | |
7224 | return 0; | |
7225 | ||
7226 | for (i = 0; i < nr_tables; i++) { | |
5398ae69 | 7227 | struct fixed_file_table *table = &file_data->table[i]; |
65e19f54 JA |
7228 | kfree(table->files); |
7229 | } | |
7230 | return 1; | |
7231 | } | |
7232 | ||
05f3fb3c JA |
7233 | static void io_ring_file_put(struct io_ring_ctx *ctx, struct file *file) |
7234 | { | |
7235 | #if defined(CONFIG_UNIX) | |
7236 | struct sock *sock = ctx->ring_sock->sk; | |
7237 | struct sk_buff_head list, *head = &sock->sk_receive_queue; | |
7238 | struct sk_buff *skb; | |
7239 | int i; | |
7240 | ||
7241 | __skb_queue_head_init(&list); | |
7242 | ||
7243 | /* | |
7244 | * Find the skb that holds this file in its SCM_RIGHTS. When found, | |
7245 | * remove this entry and rearrange the file array. | |
7246 | */ | |
7247 | skb = skb_dequeue(head); | |
7248 | while (skb) { | |
7249 | struct scm_fp_list *fp; | |
7250 | ||
7251 | fp = UNIXCB(skb).fp; | |
7252 | for (i = 0; i < fp->count; i++) { | |
7253 | int left; | |
7254 | ||
7255 | if (fp->fp[i] != file) | |
7256 | continue; | |
7257 | ||
7258 | unix_notinflight(fp->user, fp->fp[i]); | |
7259 | left = fp->count - 1 - i; | |
7260 | if (left) { | |
7261 | memmove(&fp->fp[i], &fp->fp[i + 1], | |
7262 | left * sizeof(struct file *)); | |
7263 | } | |
7264 | fp->count--; | |
7265 | if (!fp->count) { | |
7266 | kfree_skb(skb); | |
7267 | skb = NULL; | |
7268 | } else { | |
7269 | __skb_queue_tail(&list, skb); | |
7270 | } | |
7271 | fput(file); | |
7272 | file = NULL; | |
7273 | break; | |
7274 | } | |
7275 | ||
7276 | if (!file) | |
7277 | break; | |
7278 | ||
7279 | __skb_queue_tail(&list, skb); | |
7280 | ||
7281 | skb = skb_dequeue(head); | |
7282 | } | |
7283 | ||
7284 | if (skb_peek(&list)) { | |
7285 | spin_lock_irq(&head->lock); | |
7286 | while ((skb = __skb_dequeue(&list)) != NULL) | |
7287 | __skb_queue_tail(head, skb); | |
7288 | spin_unlock_irq(&head->lock); | |
7289 | } | |
7290 | #else | |
7291 | fput(file); | |
7292 | #endif | |
7293 | } | |
7294 | ||
7295 | struct io_file_put { | |
05589553 | 7296 | struct list_head list; |
05f3fb3c | 7297 | struct file *file; |
05f3fb3c JA |
7298 | }; |
7299 | ||
4a38aed2 | 7300 | static void __io_file_put_work(struct fixed_file_ref_node *ref_node) |
65e19f54 | 7301 | { |
4a38aed2 JA |
7302 | struct fixed_file_data *file_data = ref_node->file_data; |
7303 | struct io_ring_ctx *ctx = file_data->ctx; | |
05f3fb3c | 7304 | struct io_file_put *pfile, *tmp; |
05589553 XW |
7305 | |
7306 | list_for_each_entry_safe(pfile, tmp, &ref_node->file_list, list) { | |
6a4d07cd | 7307 | list_del(&pfile->list); |
05589553 XW |
7308 | io_ring_file_put(ctx, pfile->file); |
7309 | kfree(pfile); | |
65e19f54 | 7310 | } |
05589553 | 7311 | |
6a4d07cd JA |
7312 | spin_lock(&file_data->lock); |
7313 | list_del(&ref_node->node); | |
7314 | spin_unlock(&file_data->lock); | |
05589553 XW |
7315 | |
7316 | percpu_ref_exit(&ref_node->refs); | |
7317 | kfree(ref_node); | |
7318 | percpu_ref_put(&file_data->refs); | |
2faf852d | 7319 | } |
65e19f54 | 7320 | |
4a38aed2 JA |
7321 | static void io_file_put_work(struct work_struct *work) |
7322 | { | |
7323 | struct io_ring_ctx *ctx; | |
7324 | struct llist_node *node; | |
7325 | ||
7326 | ctx = container_of(work, struct io_ring_ctx, file_put_work.work); | |
7327 | node = llist_del_all(&ctx->file_put_llist); | |
7328 | ||
7329 | while (node) { | |
7330 | struct fixed_file_ref_node *ref_node; | |
7331 | struct llist_node *next = node->next; | |
7332 | ||
7333 | ref_node = llist_entry(node, struct fixed_file_ref_node, llist); | |
7334 | __io_file_put_work(ref_node); | |
7335 | node = next; | |
7336 | } | |
7337 | } | |
7338 | ||
05589553 | 7339 | static void io_file_data_ref_zero(struct percpu_ref *ref) |
2faf852d | 7340 | { |
05589553 | 7341 | struct fixed_file_ref_node *ref_node; |
4a38aed2 JA |
7342 | struct io_ring_ctx *ctx; |
7343 | bool first_add; | |
7344 | int delay = HZ; | |
65e19f54 | 7345 | |
05589553 | 7346 | ref_node = container_of(ref, struct fixed_file_ref_node, refs); |
4a38aed2 | 7347 | ctx = ref_node->file_data->ctx; |
05589553 | 7348 | |
4a38aed2 JA |
7349 | if (percpu_ref_is_dying(&ctx->file_data->refs)) |
7350 | delay = 0; | |
05589553 | 7351 | |
4a38aed2 JA |
7352 | first_add = llist_add(&ref_node->llist, &ctx->file_put_llist); |
7353 | if (!delay) | |
7354 | mod_delayed_work(system_wq, &ctx->file_put_work, 0); | |
7355 | else if (first_add) | |
7356 | queue_delayed_work(system_wq, &ctx->file_put_work, delay); | |
05f3fb3c | 7357 | } |
65e19f54 | 7358 | |
05589553 XW |
7359 | static struct fixed_file_ref_node *alloc_fixed_file_ref_node( |
7360 | struct io_ring_ctx *ctx) | |
05f3fb3c | 7361 | { |
05589553 | 7362 | struct fixed_file_ref_node *ref_node; |
05f3fb3c | 7363 | |
05589553 XW |
7364 | ref_node = kzalloc(sizeof(*ref_node), GFP_KERNEL); |
7365 | if (!ref_node) | |
7366 | return ERR_PTR(-ENOMEM); | |
05f3fb3c | 7367 | |
05589553 XW |
7368 | if (percpu_ref_init(&ref_node->refs, io_file_data_ref_zero, |
7369 | 0, GFP_KERNEL)) { | |
7370 | kfree(ref_node); | |
7371 | return ERR_PTR(-ENOMEM); | |
7372 | } | |
7373 | INIT_LIST_HEAD(&ref_node->node); | |
7374 | INIT_LIST_HEAD(&ref_node->file_list); | |
05589553 XW |
7375 | ref_node->file_data = ctx->file_data; |
7376 | return ref_node; | |
05589553 XW |
7377 | } |
7378 | ||
7379 | static void destroy_fixed_file_ref_node(struct fixed_file_ref_node *ref_node) | |
7380 | { | |
7381 | percpu_ref_exit(&ref_node->refs); | |
7382 | kfree(ref_node); | |
65e19f54 JA |
7383 | } |
7384 | ||
6b06314c JA |
7385 | static int io_sqe_files_register(struct io_ring_ctx *ctx, void __user *arg, |
7386 | unsigned nr_args) | |
7387 | { | |
7388 | __s32 __user *fds = (__s32 __user *) arg; | |
600cf3f8 | 7389 | unsigned nr_tables, i; |
05f3fb3c | 7390 | struct file *file; |
600cf3f8 | 7391 | int fd, ret = -ENOMEM; |
05589553 | 7392 | struct fixed_file_ref_node *ref_node; |
5398ae69 | 7393 | struct fixed_file_data *file_data; |
6b06314c | 7394 | |
05f3fb3c | 7395 | if (ctx->file_data) |
6b06314c JA |
7396 | return -EBUSY; |
7397 | if (!nr_args) | |
7398 | return -EINVAL; | |
7399 | if (nr_args > IORING_MAX_FIXED_FILES) | |
7400 | return -EMFILE; | |
7401 | ||
5398ae69 PB |
7402 | file_data = kzalloc(sizeof(*ctx->file_data), GFP_KERNEL); |
7403 | if (!file_data) | |
05f3fb3c | 7404 | return -ENOMEM; |
5398ae69 PB |
7405 | file_data->ctx = ctx; |
7406 | init_completion(&file_data->done); | |
7407 | INIT_LIST_HEAD(&file_data->ref_list); | |
7408 | spin_lock_init(&file_data->lock); | |
05f3fb3c | 7409 | |
65e19f54 | 7410 | nr_tables = DIV_ROUND_UP(nr_args, IORING_MAX_FILES_TABLE); |
035fbafc | 7411 | file_data->table = kcalloc(nr_tables, sizeof(*file_data->table), |
5398ae69 | 7412 | GFP_KERNEL); |
600cf3f8 PB |
7413 | if (!file_data->table) |
7414 | goto out_free; | |
05f3fb3c | 7415 | |
5398ae69 | 7416 | if (percpu_ref_init(&file_data->refs, io_file_ref_kill, |
600cf3f8 PB |
7417 | PERCPU_REF_ALLOW_REINIT, GFP_KERNEL)) |
7418 | goto out_free; | |
6b06314c | 7419 | |
600cf3f8 PB |
7420 | if (io_sqe_alloc_file_tables(file_data, nr_tables, nr_args)) |
7421 | goto out_ref; | |
55cbc256 | 7422 | ctx->file_data = file_data; |
65e19f54 | 7423 | |
08a45173 | 7424 | for (i = 0; i < nr_args; i++, ctx->nr_user_files++) { |
65e19f54 JA |
7425 | struct fixed_file_table *table; |
7426 | unsigned index; | |
7427 | ||
600cf3f8 PB |
7428 | if (copy_from_user(&fd, &fds[i], sizeof(fd))) { |
7429 | ret = -EFAULT; | |
7430 | goto out_fput; | |
7431 | } | |
08a45173 | 7432 | /* allow sparse sets */ |
600cf3f8 | 7433 | if (fd == -1) |
08a45173 | 7434 | continue; |
6b06314c | 7435 | |
05f3fb3c | 7436 | file = fget(fd); |
6b06314c | 7437 | ret = -EBADF; |
05f3fb3c | 7438 | if (!file) |
600cf3f8 | 7439 | goto out_fput; |
05f3fb3c | 7440 | |
6b06314c JA |
7441 | /* |
7442 | * Don't allow io_uring instances to be registered. If UNIX | |
7443 | * isn't enabled, then this causes a reference cycle and this | |
7444 | * instance can never get freed. If UNIX is enabled we'll | |
7445 | * handle it just fine, but there's still no point in allowing | |
7446 | * a ring fd as it doesn't support regular read/write anyway. | |
7447 | */ | |
05f3fb3c JA |
7448 | if (file->f_op == &io_uring_fops) { |
7449 | fput(file); | |
600cf3f8 | 7450 | goto out_fput; |
6b06314c | 7451 | } |
600cf3f8 PB |
7452 | table = &file_data->table[i >> IORING_FILE_TABLE_SHIFT]; |
7453 | index = i & IORING_FILE_TABLE_MASK; | |
05f3fb3c | 7454 | table->files[index] = file; |
6b06314c JA |
7455 | } |
7456 | ||
6b06314c | 7457 | ret = io_sqe_files_scm(ctx); |
05589553 | 7458 | if (ret) { |
6b06314c | 7459 | io_sqe_files_unregister(ctx); |
05589553 XW |
7460 | return ret; |
7461 | } | |
6b06314c | 7462 | |
05589553 XW |
7463 | ref_node = alloc_fixed_file_ref_node(ctx); |
7464 | if (IS_ERR(ref_node)) { | |
7465 | io_sqe_files_unregister(ctx); | |
7466 | return PTR_ERR(ref_node); | |
7467 | } | |
7468 | ||
b2e96852 | 7469 | file_data->node = ref_node; |
5398ae69 PB |
7470 | spin_lock(&file_data->lock); |
7471 | list_add(&ref_node->node, &file_data->ref_list); | |
7472 | spin_unlock(&file_data->lock); | |
7473 | percpu_ref_get(&file_data->refs); | |
6b06314c | 7474 | return ret; |
600cf3f8 PB |
7475 | out_fput: |
7476 | for (i = 0; i < ctx->nr_user_files; i++) { | |
7477 | file = io_file_from_index(ctx, i); | |
7478 | if (file) | |
7479 | fput(file); | |
7480 | } | |
7481 | for (i = 0; i < nr_tables; i++) | |
7482 | kfree(file_data->table[i].files); | |
7483 | ctx->nr_user_files = 0; | |
7484 | out_ref: | |
7485 | percpu_ref_exit(&file_data->refs); | |
7486 | out_free: | |
7487 | kfree(file_data->table); | |
7488 | kfree(file_data); | |
55cbc256 | 7489 | ctx->file_data = NULL; |
6b06314c JA |
7490 | return ret; |
7491 | } | |
7492 | ||
c3a31e60 JA |
7493 | static int io_sqe_file_register(struct io_ring_ctx *ctx, struct file *file, |
7494 | int index) | |
7495 | { | |
7496 | #if defined(CONFIG_UNIX) | |
7497 | struct sock *sock = ctx->ring_sock->sk; | |
7498 | struct sk_buff_head *head = &sock->sk_receive_queue; | |
7499 | struct sk_buff *skb; | |
7500 | ||
7501 | /* | |
7502 | * See if we can merge this file into an existing skb SCM_RIGHTS | |
7503 | * file set. If there's no room, fall back to allocating a new skb | |
7504 | * and filling it in. | |
7505 | */ | |
7506 | spin_lock_irq(&head->lock); | |
7507 | skb = skb_peek(head); | |
7508 | if (skb) { | |
7509 | struct scm_fp_list *fpl = UNIXCB(skb).fp; | |
7510 | ||
7511 | if (fpl->count < SCM_MAX_FD) { | |
7512 | __skb_unlink(skb, head); | |
7513 | spin_unlock_irq(&head->lock); | |
7514 | fpl->fp[fpl->count] = get_file(file); | |
7515 | unix_inflight(fpl->user, fpl->fp[fpl->count]); | |
7516 | fpl->count++; | |
7517 | spin_lock_irq(&head->lock); | |
7518 | __skb_queue_head(head, skb); | |
7519 | } else { | |
7520 | skb = NULL; | |
7521 | } | |
7522 | } | |
7523 | spin_unlock_irq(&head->lock); | |
7524 | ||
7525 | if (skb) { | |
7526 | fput(file); | |
7527 | return 0; | |
7528 | } | |
7529 | ||
7530 | return __io_sqe_files_scm(ctx, 1, index); | |
7531 | #else | |
7532 | return 0; | |
7533 | #endif | |
7534 | } | |
7535 | ||
a5318d3c | 7536 | static int io_queue_file_removal(struct fixed_file_data *data, |
05589553 | 7537 | struct file *file) |
05f3fb3c | 7538 | { |
a5318d3c | 7539 | struct io_file_put *pfile; |
b2e96852 | 7540 | struct fixed_file_ref_node *ref_node = data->node; |
05f3fb3c | 7541 | |
05f3fb3c | 7542 | pfile = kzalloc(sizeof(*pfile), GFP_KERNEL); |
a5318d3c HD |
7543 | if (!pfile) |
7544 | return -ENOMEM; | |
05f3fb3c JA |
7545 | |
7546 | pfile->file = file; | |
05589553 XW |
7547 | list_add(&pfile->list, &ref_node->file_list); |
7548 | ||
a5318d3c | 7549 | return 0; |
05f3fb3c JA |
7550 | } |
7551 | ||
7552 | static int __io_sqe_files_update(struct io_ring_ctx *ctx, | |
7553 | struct io_uring_files_update *up, | |
7554 | unsigned nr_args) | |
7555 | { | |
7556 | struct fixed_file_data *data = ctx->file_data; | |
05589553 | 7557 | struct fixed_file_ref_node *ref_node; |
05f3fb3c | 7558 | struct file *file; |
c3a31e60 JA |
7559 | __s32 __user *fds; |
7560 | int fd, i, err; | |
7561 | __u32 done; | |
05589553 | 7562 | bool needs_switch = false; |
c3a31e60 | 7563 | |
05f3fb3c | 7564 | if (check_add_overflow(up->offset, nr_args, &done)) |
c3a31e60 JA |
7565 | return -EOVERFLOW; |
7566 | if (done > ctx->nr_user_files) | |
7567 | return -EINVAL; | |
7568 | ||
05589553 XW |
7569 | ref_node = alloc_fixed_file_ref_node(ctx); |
7570 | if (IS_ERR(ref_node)) | |
7571 | return PTR_ERR(ref_node); | |
7572 | ||
c3a31e60 | 7573 | done = 0; |
05f3fb3c | 7574 | fds = u64_to_user_ptr(up->fds); |
c3a31e60 | 7575 | while (nr_args) { |
65e19f54 JA |
7576 | struct fixed_file_table *table; |
7577 | unsigned index; | |
7578 | ||
c3a31e60 JA |
7579 | err = 0; |
7580 | if (copy_from_user(&fd, &fds[done], sizeof(fd))) { | |
7581 | err = -EFAULT; | |
7582 | break; | |
7583 | } | |
05f3fb3c JA |
7584 | i = array_index_nospec(up->offset, ctx->nr_user_files); |
7585 | table = &ctx->file_data->table[i >> IORING_FILE_TABLE_SHIFT]; | |
65e19f54 JA |
7586 | index = i & IORING_FILE_TABLE_MASK; |
7587 | if (table->files[index]) { | |
98dfd502 | 7588 | file = table->files[index]; |
a5318d3c HD |
7589 | err = io_queue_file_removal(data, file); |
7590 | if (err) | |
7591 | break; | |
65e19f54 | 7592 | table->files[index] = NULL; |
05589553 | 7593 | needs_switch = true; |
c3a31e60 JA |
7594 | } |
7595 | if (fd != -1) { | |
c3a31e60 JA |
7596 | file = fget(fd); |
7597 | if (!file) { | |
7598 | err = -EBADF; | |
7599 | break; | |
7600 | } | |
7601 | /* | |
7602 | * Don't allow io_uring instances to be registered. If | |
7603 | * UNIX isn't enabled, then this causes a reference | |
7604 | * cycle and this instance can never get freed. If UNIX | |
7605 | * is enabled we'll handle it just fine, but there's | |
7606 | * still no point in allowing a ring fd as it doesn't | |
7607 | * support regular read/write anyway. | |
7608 | */ | |
7609 | if (file->f_op == &io_uring_fops) { | |
7610 | fput(file); | |
7611 | err = -EBADF; | |
7612 | break; | |
7613 | } | |
65e19f54 | 7614 | table->files[index] = file; |
c3a31e60 | 7615 | err = io_sqe_file_register(ctx, file, i); |
f3bd9dae | 7616 | if (err) { |
95d1c8e5 | 7617 | table->files[index] = NULL; |
f3bd9dae | 7618 | fput(file); |
c3a31e60 | 7619 | break; |
f3bd9dae | 7620 | } |
c3a31e60 JA |
7621 | } |
7622 | nr_args--; | |
7623 | done++; | |
05f3fb3c JA |
7624 | up->offset++; |
7625 | } | |
7626 | ||
05589553 | 7627 | if (needs_switch) { |
b2e96852 | 7628 | percpu_ref_kill(&data->node->refs); |
6a4d07cd | 7629 | spin_lock(&data->lock); |
05589553 | 7630 | list_add(&ref_node->node, &data->ref_list); |
b2e96852 | 7631 | data->node = ref_node; |
6a4d07cd | 7632 | spin_unlock(&data->lock); |
05589553 XW |
7633 | percpu_ref_get(&ctx->file_data->refs); |
7634 | } else | |
7635 | destroy_fixed_file_ref_node(ref_node); | |
c3a31e60 JA |
7636 | |
7637 | return done ? done : err; | |
7638 | } | |
05589553 | 7639 | |
05f3fb3c JA |
7640 | static int io_sqe_files_update(struct io_ring_ctx *ctx, void __user *arg, |
7641 | unsigned nr_args) | |
7642 | { | |
7643 | struct io_uring_files_update up; | |
7644 | ||
7645 | if (!ctx->file_data) | |
7646 | return -ENXIO; | |
7647 | if (!nr_args) | |
7648 | return -EINVAL; | |
7649 | if (copy_from_user(&up, arg, sizeof(up))) | |
7650 | return -EFAULT; | |
7651 | if (up.resv) | |
7652 | return -EINVAL; | |
7653 | ||
7654 | return __io_sqe_files_update(ctx, &up, nr_args); | |
7655 | } | |
c3a31e60 | 7656 | |
e9fd9396 | 7657 | static void io_free_work(struct io_wq_work *work) |
7d723065 JA |
7658 | { |
7659 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); | |
7660 | ||
e9fd9396 | 7661 | /* Consider that io_steal_work() relies on this ref */ |
7d723065 JA |
7662 | io_put_req(req); |
7663 | } | |
7664 | ||
24369c2e PB |
7665 | static int io_init_wq_offload(struct io_ring_ctx *ctx, |
7666 | struct io_uring_params *p) | |
7667 | { | |
7668 | struct io_wq_data data; | |
7669 | struct fd f; | |
7670 | struct io_ring_ctx *ctx_attach; | |
7671 | unsigned int concurrency; | |
7672 | int ret = 0; | |
7673 | ||
7674 | data.user = ctx->user; | |
e9fd9396 | 7675 | data.free_work = io_free_work; |
f5fa38c5 | 7676 | data.do_work = io_wq_submit_work; |
24369c2e PB |
7677 | |
7678 | if (!(p->flags & IORING_SETUP_ATTACH_WQ)) { | |
7679 | /* Do QD, or 4 * CPUS, whatever is smallest */ | |
7680 | concurrency = min(ctx->sq_entries, 4 * num_online_cpus()); | |
7681 | ||
7682 | ctx->io_wq = io_wq_create(concurrency, &data); | |
7683 | if (IS_ERR(ctx->io_wq)) { | |
7684 | ret = PTR_ERR(ctx->io_wq); | |
7685 | ctx->io_wq = NULL; | |
7686 | } | |
7687 | return ret; | |
7688 | } | |
7689 | ||
7690 | f = fdget(p->wq_fd); | |
7691 | if (!f.file) | |
7692 | return -EBADF; | |
7693 | ||
7694 | if (f.file->f_op != &io_uring_fops) { | |
7695 | ret = -EINVAL; | |
7696 | goto out_fput; | |
7697 | } | |
7698 | ||
7699 | ctx_attach = f.file->private_data; | |
7700 | /* @io_wq is protected by holding the fd */ | |
7701 | if (!io_wq_get(ctx_attach->io_wq, &data)) { | |
7702 | ret = -EINVAL; | |
7703 | goto out_fput; | |
7704 | } | |
7705 | ||
7706 | ctx->io_wq = ctx_attach->io_wq; | |
7707 | out_fput: | |
7708 | fdput(f); | |
7709 | return ret; | |
7710 | } | |
7711 | ||
0f212204 JA |
7712 | static int io_uring_alloc_task_context(struct task_struct *task) |
7713 | { | |
7714 | struct io_uring_task *tctx; | |
d8a6df10 | 7715 | int ret; |
0f212204 JA |
7716 | |
7717 | tctx = kmalloc(sizeof(*tctx), GFP_KERNEL); | |
7718 | if (unlikely(!tctx)) | |
7719 | return -ENOMEM; | |
7720 | ||
d8a6df10 JA |
7721 | ret = percpu_counter_init(&tctx->inflight, 0, GFP_KERNEL); |
7722 | if (unlikely(ret)) { | |
7723 | kfree(tctx); | |
7724 | return ret; | |
7725 | } | |
7726 | ||
0f212204 JA |
7727 | xa_init(&tctx->xa); |
7728 | init_waitqueue_head(&tctx->wait); | |
7729 | tctx->last = NULL; | |
7730 | tctx->in_idle = 0; | |
500a373d JA |
7731 | io_init_identity(&tctx->__identity); |
7732 | tctx->identity = &tctx->__identity; | |
0f212204 JA |
7733 | task->io_uring = tctx; |
7734 | return 0; | |
7735 | } | |
7736 | ||
7737 | void __io_uring_free(struct task_struct *tsk) | |
7738 | { | |
7739 | struct io_uring_task *tctx = tsk->io_uring; | |
7740 | ||
7741 | WARN_ON_ONCE(!xa_empty(&tctx->xa)); | |
500a373d JA |
7742 | WARN_ON_ONCE(refcount_read(&tctx->identity->count) != 1); |
7743 | if (tctx->identity != &tctx->__identity) | |
7744 | kfree(tctx->identity); | |
d8a6df10 | 7745 | percpu_counter_destroy(&tctx->inflight); |
0f212204 JA |
7746 | kfree(tctx); |
7747 | tsk->io_uring = NULL; | |
7748 | } | |
7749 | ||
7e84e1c7 SG |
7750 | static int io_sq_offload_create(struct io_ring_ctx *ctx, |
7751 | struct io_uring_params *p) | |
2b188cc1 JA |
7752 | { |
7753 | int ret; | |
7754 | ||
6c271ce2 | 7755 | if (ctx->flags & IORING_SETUP_SQPOLL) { |
534ca6d6 JA |
7756 | struct io_sq_data *sqd; |
7757 | ||
3ec482d1 JA |
7758 | ret = -EPERM; |
7759 | if (!capable(CAP_SYS_ADMIN)) | |
7760 | goto err; | |
7761 | ||
534ca6d6 JA |
7762 | sqd = io_get_sq_data(p); |
7763 | if (IS_ERR(sqd)) { | |
7764 | ret = PTR_ERR(sqd); | |
7765 | goto err; | |
7766 | } | |
69fb2131 | 7767 | |
534ca6d6 | 7768 | ctx->sq_data = sqd; |
69fb2131 JA |
7769 | io_sq_thread_park(sqd); |
7770 | mutex_lock(&sqd->ctx_lock); | |
7771 | list_add(&ctx->sqd_list, &sqd->ctx_new_list); | |
7772 | mutex_unlock(&sqd->ctx_lock); | |
7773 | io_sq_thread_unpark(sqd); | |
534ca6d6 | 7774 | |
917257da JA |
7775 | ctx->sq_thread_idle = msecs_to_jiffies(p->sq_thread_idle); |
7776 | if (!ctx->sq_thread_idle) | |
7777 | ctx->sq_thread_idle = HZ; | |
7778 | ||
aa06165d JA |
7779 | if (sqd->thread) |
7780 | goto done; | |
7781 | ||
6c271ce2 | 7782 | if (p->flags & IORING_SETUP_SQ_AFF) { |
44a9bd18 | 7783 | int cpu = p->sq_thread_cpu; |
6c271ce2 | 7784 | |
917257da | 7785 | ret = -EINVAL; |
44a9bd18 JA |
7786 | if (cpu >= nr_cpu_ids) |
7787 | goto err; | |
7889f44d | 7788 | if (!cpu_online(cpu)) |
917257da JA |
7789 | goto err; |
7790 | ||
69fb2131 | 7791 | sqd->thread = kthread_create_on_cpu(io_sq_thread, sqd, |
534ca6d6 | 7792 | cpu, "io_uring-sq"); |
6c271ce2 | 7793 | } else { |
69fb2131 | 7794 | sqd->thread = kthread_create(io_sq_thread, sqd, |
6c271ce2 JA |
7795 | "io_uring-sq"); |
7796 | } | |
534ca6d6 JA |
7797 | if (IS_ERR(sqd->thread)) { |
7798 | ret = PTR_ERR(sqd->thread); | |
7799 | sqd->thread = NULL; | |
6c271ce2 JA |
7800 | goto err; |
7801 | } | |
534ca6d6 | 7802 | ret = io_uring_alloc_task_context(sqd->thread); |
0f212204 JA |
7803 | if (ret) |
7804 | goto err; | |
6c271ce2 JA |
7805 | } else if (p->flags & IORING_SETUP_SQ_AFF) { |
7806 | /* Can't have SQ_AFF without SQPOLL */ | |
7807 | ret = -EINVAL; | |
7808 | goto err; | |
7809 | } | |
7810 | ||
aa06165d | 7811 | done: |
24369c2e PB |
7812 | ret = io_init_wq_offload(ctx, p); |
7813 | if (ret) | |
2b188cc1 | 7814 | goto err; |
2b188cc1 JA |
7815 | |
7816 | return 0; | |
7817 | err: | |
54a91f3b | 7818 | io_finish_async(ctx); |
2b188cc1 JA |
7819 | return ret; |
7820 | } | |
7821 | ||
7e84e1c7 SG |
7822 | static void io_sq_offload_start(struct io_ring_ctx *ctx) |
7823 | { | |
534ca6d6 JA |
7824 | struct io_sq_data *sqd = ctx->sq_data; |
7825 | ||
7826 | if ((ctx->flags & IORING_SETUP_SQPOLL) && sqd->thread) | |
7827 | wake_up_process(sqd->thread); | |
7e84e1c7 SG |
7828 | } |
7829 | ||
a087e2b5 BM |
7830 | static inline void __io_unaccount_mem(struct user_struct *user, |
7831 | unsigned long nr_pages) | |
2b188cc1 JA |
7832 | { |
7833 | atomic_long_sub(nr_pages, &user->locked_vm); | |
7834 | } | |
7835 | ||
a087e2b5 BM |
7836 | static inline int __io_account_mem(struct user_struct *user, |
7837 | unsigned long nr_pages) | |
2b188cc1 JA |
7838 | { |
7839 | unsigned long page_limit, cur_pages, new_pages; | |
7840 | ||
7841 | /* Don't allow more pages than we can safely lock */ | |
7842 | page_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT; | |
7843 | ||
7844 | do { | |
7845 | cur_pages = atomic_long_read(&user->locked_vm); | |
7846 | new_pages = cur_pages + nr_pages; | |
7847 | if (new_pages > page_limit) | |
7848 | return -ENOMEM; | |
7849 | } while (atomic_long_cmpxchg(&user->locked_vm, cur_pages, | |
7850 | new_pages) != cur_pages); | |
7851 | ||
7852 | return 0; | |
7853 | } | |
7854 | ||
2e0464d4 BM |
7855 | static void io_unaccount_mem(struct io_ring_ctx *ctx, unsigned long nr_pages, |
7856 | enum io_mem_account acct) | |
a087e2b5 | 7857 | { |
aad5d8da | 7858 | if (ctx->limit_mem) |
a087e2b5 | 7859 | __io_unaccount_mem(ctx->user, nr_pages); |
30975825 | 7860 | |
2aede0e4 | 7861 | if (ctx->mm_account) { |
2e0464d4 | 7862 | if (acct == ACCT_LOCKED) |
2aede0e4 | 7863 | ctx->mm_account->locked_vm -= nr_pages; |
2e0464d4 | 7864 | else if (acct == ACCT_PINNED) |
2aede0e4 | 7865 | atomic64_sub(nr_pages, &ctx->mm_account->pinned_vm); |
2e0464d4 | 7866 | } |
a087e2b5 BM |
7867 | } |
7868 | ||
2e0464d4 BM |
7869 | static int io_account_mem(struct io_ring_ctx *ctx, unsigned long nr_pages, |
7870 | enum io_mem_account acct) | |
a087e2b5 | 7871 | { |
30975825 BM |
7872 | int ret; |
7873 | ||
7874 | if (ctx->limit_mem) { | |
7875 | ret = __io_account_mem(ctx->user, nr_pages); | |
7876 | if (ret) | |
7877 | return ret; | |
7878 | } | |
7879 | ||
2aede0e4 | 7880 | if (ctx->mm_account) { |
2e0464d4 | 7881 | if (acct == ACCT_LOCKED) |
2aede0e4 | 7882 | ctx->mm_account->locked_vm += nr_pages; |
2e0464d4 | 7883 | else if (acct == ACCT_PINNED) |
2aede0e4 | 7884 | atomic64_add(nr_pages, &ctx->mm_account->pinned_vm); |
2e0464d4 | 7885 | } |
a087e2b5 BM |
7886 | |
7887 | return 0; | |
7888 | } | |
7889 | ||
2b188cc1 JA |
7890 | static void io_mem_free(void *ptr) |
7891 | { | |
52e04ef4 MR |
7892 | struct page *page; |
7893 | ||
7894 | if (!ptr) | |
7895 | return; | |
2b188cc1 | 7896 | |
52e04ef4 | 7897 | page = virt_to_head_page(ptr); |
2b188cc1 JA |
7898 | if (put_page_testzero(page)) |
7899 | free_compound_page(page); | |
7900 | } | |
7901 | ||
7902 | static void *io_mem_alloc(size_t size) | |
7903 | { | |
7904 | gfp_t gfp_flags = GFP_KERNEL | __GFP_ZERO | __GFP_NOWARN | __GFP_COMP | | |
7905 | __GFP_NORETRY; | |
7906 | ||
7907 | return (void *) __get_free_pages(gfp_flags, get_order(size)); | |
7908 | } | |
7909 | ||
75b28aff HV |
7910 | static unsigned long rings_size(unsigned sq_entries, unsigned cq_entries, |
7911 | size_t *sq_offset) | |
7912 | { | |
7913 | struct io_rings *rings; | |
7914 | size_t off, sq_array_size; | |
7915 | ||
7916 | off = struct_size(rings, cqes, cq_entries); | |
7917 | if (off == SIZE_MAX) | |
7918 | return SIZE_MAX; | |
7919 | ||
7920 | #ifdef CONFIG_SMP | |
7921 | off = ALIGN(off, SMP_CACHE_BYTES); | |
7922 | if (off == 0) | |
7923 | return SIZE_MAX; | |
7924 | #endif | |
7925 | ||
b36200f5 DV |
7926 | if (sq_offset) |
7927 | *sq_offset = off; | |
7928 | ||
75b28aff HV |
7929 | sq_array_size = array_size(sizeof(u32), sq_entries); |
7930 | if (sq_array_size == SIZE_MAX) | |
7931 | return SIZE_MAX; | |
7932 | ||
7933 | if (check_add_overflow(off, sq_array_size, &off)) | |
7934 | return SIZE_MAX; | |
7935 | ||
75b28aff HV |
7936 | return off; |
7937 | } | |
7938 | ||
2b188cc1 JA |
7939 | static unsigned long ring_pages(unsigned sq_entries, unsigned cq_entries) |
7940 | { | |
75b28aff | 7941 | size_t pages; |
2b188cc1 | 7942 | |
75b28aff HV |
7943 | pages = (size_t)1 << get_order( |
7944 | rings_size(sq_entries, cq_entries, NULL)); | |
7945 | pages += (size_t)1 << get_order( | |
7946 | array_size(sizeof(struct io_uring_sqe), sq_entries)); | |
2b188cc1 | 7947 | |
75b28aff | 7948 | return pages; |
2b188cc1 JA |
7949 | } |
7950 | ||
edafccee JA |
7951 | static int io_sqe_buffer_unregister(struct io_ring_ctx *ctx) |
7952 | { | |
7953 | int i, j; | |
7954 | ||
7955 | if (!ctx->user_bufs) | |
7956 | return -ENXIO; | |
7957 | ||
7958 | for (i = 0; i < ctx->nr_user_bufs; i++) { | |
7959 | struct io_mapped_ubuf *imu = &ctx->user_bufs[i]; | |
7960 | ||
7961 | for (j = 0; j < imu->nr_bvecs; j++) | |
f1f6a7dd | 7962 | unpin_user_page(imu->bvec[j].bv_page); |
edafccee | 7963 | |
de293938 JA |
7964 | if (imu->acct_pages) |
7965 | io_unaccount_mem(ctx, imu->acct_pages, ACCT_PINNED); | |
d4ef6475 | 7966 | kvfree(imu->bvec); |
edafccee JA |
7967 | imu->nr_bvecs = 0; |
7968 | } | |
7969 | ||
7970 | kfree(ctx->user_bufs); | |
7971 | ctx->user_bufs = NULL; | |
7972 | ctx->nr_user_bufs = 0; | |
7973 | return 0; | |
7974 | } | |
7975 | ||
7976 | static int io_copy_iov(struct io_ring_ctx *ctx, struct iovec *dst, | |
7977 | void __user *arg, unsigned index) | |
7978 | { | |
7979 | struct iovec __user *src; | |
7980 | ||
7981 | #ifdef CONFIG_COMPAT | |
7982 | if (ctx->compat) { | |
7983 | struct compat_iovec __user *ciovs; | |
7984 | struct compat_iovec ciov; | |
7985 | ||
7986 | ciovs = (struct compat_iovec __user *) arg; | |
7987 | if (copy_from_user(&ciov, &ciovs[index], sizeof(ciov))) | |
7988 | return -EFAULT; | |
7989 | ||
d55e5f5b | 7990 | dst->iov_base = u64_to_user_ptr((u64)ciov.iov_base); |
edafccee JA |
7991 | dst->iov_len = ciov.iov_len; |
7992 | return 0; | |
7993 | } | |
7994 | #endif | |
7995 | src = (struct iovec __user *) arg; | |
7996 | if (copy_from_user(dst, &src[index], sizeof(*dst))) | |
7997 | return -EFAULT; | |
7998 | return 0; | |
7999 | } | |
8000 | ||
de293938 JA |
8001 | /* |
8002 | * Not super efficient, but this is just a registration time. And we do cache | |
8003 | * the last compound head, so generally we'll only do a full search if we don't | |
8004 | * match that one. | |
8005 | * | |
8006 | * We check if the given compound head page has already been accounted, to | |
8007 | * avoid double accounting it. This allows us to account the full size of the | |
8008 | * page, not just the constituent pages of a huge page. | |
8009 | */ | |
8010 | static bool headpage_already_acct(struct io_ring_ctx *ctx, struct page **pages, | |
8011 | int nr_pages, struct page *hpage) | |
8012 | { | |
8013 | int i, j; | |
8014 | ||
8015 | /* check current page array */ | |
8016 | for (i = 0; i < nr_pages; i++) { | |
8017 | if (!PageCompound(pages[i])) | |
8018 | continue; | |
8019 | if (compound_head(pages[i]) == hpage) | |
8020 | return true; | |
8021 | } | |
8022 | ||
8023 | /* check previously registered pages */ | |
8024 | for (i = 0; i < ctx->nr_user_bufs; i++) { | |
8025 | struct io_mapped_ubuf *imu = &ctx->user_bufs[i]; | |
8026 | ||
8027 | for (j = 0; j < imu->nr_bvecs; j++) { | |
8028 | if (!PageCompound(imu->bvec[j].bv_page)) | |
8029 | continue; | |
8030 | if (compound_head(imu->bvec[j].bv_page) == hpage) | |
8031 | return true; | |
8032 | } | |
8033 | } | |
8034 | ||
8035 | return false; | |
8036 | } | |
8037 | ||
8038 | static int io_buffer_account_pin(struct io_ring_ctx *ctx, struct page **pages, | |
8039 | int nr_pages, struct io_mapped_ubuf *imu, | |
8040 | struct page **last_hpage) | |
8041 | { | |
8042 | int i, ret; | |
8043 | ||
8044 | for (i = 0; i < nr_pages; i++) { | |
8045 | if (!PageCompound(pages[i])) { | |
8046 | imu->acct_pages++; | |
8047 | } else { | |
8048 | struct page *hpage; | |
8049 | ||
8050 | hpage = compound_head(pages[i]); | |
8051 | if (hpage == *last_hpage) | |
8052 | continue; | |
8053 | *last_hpage = hpage; | |
8054 | if (headpage_already_acct(ctx, pages, i, hpage)) | |
8055 | continue; | |
8056 | imu->acct_pages += page_size(hpage) >> PAGE_SHIFT; | |
8057 | } | |
8058 | } | |
8059 | ||
8060 | if (!imu->acct_pages) | |
8061 | return 0; | |
8062 | ||
8063 | ret = io_account_mem(ctx, imu->acct_pages, ACCT_PINNED); | |
8064 | if (ret) | |
8065 | imu->acct_pages = 0; | |
8066 | return ret; | |
8067 | } | |
8068 | ||
edafccee JA |
8069 | static int io_sqe_buffer_register(struct io_ring_ctx *ctx, void __user *arg, |
8070 | unsigned nr_args) | |
8071 | { | |
8072 | struct vm_area_struct **vmas = NULL; | |
8073 | struct page **pages = NULL; | |
de293938 | 8074 | struct page *last_hpage = NULL; |
edafccee JA |
8075 | int i, j, got_pages = 0; |
8076 | int ret = -EINVAL; | |
8077 | ||
8078 | if (ctx->user_bufs) | |
8079 | return -EBUSY; | |
8080 | if (!nr_args || nr_args > UIO_MAXIOV) | |
8081 | return -EINVAL; | |
8082 | ||
8083 | ctx->user_bufs = kcalloc(nr_args, sizeof(struct io_mapped_ubuf), | |
8084 | GFP_KERNEL); | |
8085 | if (!ctx->user_bufs) | |
8086 | return -ENOMEM; | |
8087 | ||
8088 | for (i = 0; i < nr_args; i++) { | |
8089 | struct io_mapped_ubuf *imu = &ctx->user_bufs[i]; | |
8090 | unsigned long off, start, end, ubuf; | |
8091 | int pret, nr_pages; | |
8092 | struct iovec iov; | |
8093 | size_t size; | |
8094 | ||
8095 | ret = io_copy_iov(ctx, &iov, arg, i); | |
8096 | if (ret) | |
a278682d | 8097 | goto err; |
edafccee JA |
8098 | |
8099 | /* | |
8100 | * Don't impose further limits on the size and buffer | |
8101 | * constraints here, we'll -EINVAL later when IO is | |
8102 | * submitted if they are wrong. | |
8103 | */ | |
8104 | ret = -EFAULT; | |
8105 | if (!iov.iov_base || !iov.iov_len) | |
8106 | goto err; | |
8107 | ||
8108 | /* arbitrary limit, but we need something */ | |
8109 | if (iov.iov_len > SZ_1G) | |
8110 | goto err; | |
8111 | ||
8112 | ubuf = (unsigned long) iov.iov_base; | |
8113 | end = (ubuf + iov.iov_len + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
8114 | start = ubuf >> PAGE_SHIFT; | |
8115 | nr_pages = end - start; | |
8116 | ||
edafccee JA |
8117 | ret = 0; |
8118 | if (!pages || nr_pages > got_pages) { | |
a8c73c1a DE |
8119 | kvfree(vmas); |
8120 | kvfree(pages); | |
d4ef6475 | 8121 | pages = kvmalloc_array(nr_pages, sizeof(struct page *), |
edafccee | 8122 | GFP_KERNEL); |
d4ef6475 | 8123 | vmas = kvmalloc_array(nr_pages, |
edafccee JA |
8124 | sizeof(struct vm_area_struct *), |
8125 | GFP_KERNEL); | |
8126 | if (!pages || !vmas) { | |
8127 | ret = -ENOMEM; | |
edafccee JA |
8128 | goto err; |
8129 | } | |
8130 | got_pages = nr_pages; | |
8131 | } | |
8132 | ||
d4ef6475 | 8133 | imu->bvec = kvmalloc_array(nr_pages, sizeof(struct bio_vec), |
edafccee JA |
8134 | GFP_KERNEL); |
8135 | ret = -ENOMEM; | |
de293938 | 8136 | if (!imu->bvec) |
edafccee | 8137 | goto err; |
edafccee JA |
8138 | |
8139 | ret = 0; | |
d8ed45c5 | 8140 | mmap_read_lock(current->mm); |
2113b05d | 8141 | pret = pin_user_pages(ubuf, nr_pages, |
932f4a63 IW |
8142 | FOLL_WRITE | FOLL_LONGTERM, |
8143 | pages, vmas); | |
edafccee JA |
8144 | if (pret == nr_pages) { |
8145 | /* don't support file backed memory */ | |
8146 | for (j = 0; j < nr_pages; j++) { | |
8147 | struct vm_area_struct *vma = vmas[j]; | |
8148 | ||
8149 | if (vma->vm_file && | |
8150 | !is_file_hugepages(vma->vm_file)) { | |
8151 | ret = -EOPNOTSUPP; | |
8152 | break; | |
8153 | } | |
8154 | } | |
8155 | } else { | |
8156 | ret = pret < 0 ? pret : -EFAULT; | |
8157 | } | |
d8ed45c5 | 8158 | mmap_read_unlock(current->mm); |
edafccee JA |
8159 | if (ret) { |
8160 | /* | |
8161 | * if we did partial map, or found file backed vmas, | |
8162 | * release any pages we did get | |
8163 | */ | |
27c4d3a3 | 8164 | if (pret > 0) |
f1f6a7dd | 8165 | unpin_user_pages(pages, pret); |
de293938 JA |
8166 | kvfree(imu->bvec); |
8167 | goto err; | |
8168 | } | |
8169 | ||
8170 | ret = io_buffer_account_pin(ctx, pages, pret, imu, &last_hpage); | |
8171 | if (ret) { | |
8172 | unpin_user_pages(pages, pret); | |
d4ef6475 | 8173 | kvfree(imu->bvec); |
edafccee JA |
8174 | goto err; |
8175 | } | |
8176 | ||
8177 | off = ubuf & ~PAGE_MASK; | |
8178 | size = iov.iov_len; | |
8179 | for (j = 0; j < nr_pages; j++) { | |
8180 | size_t vec_len; | |
8181 | ||
8182 | vec_len = min_t(size_t, size, PAGE_SIZE - off); | |
8183 | imu->bvec[j].bv_page = pages[j]; | |
8184 | imu->bvec[j].bv_len = vec_len; | |
8185 | imu->bvec[j].bv_offset = off; | |
8186 | off = 0; | |
8187 | size -= vec_len; | |
8188 | } | |
8189 | /* store original address for later verification */ | |
8190 | imu->ubuf = ubuf; | |
8191 | imu->len = iov.iov_len; | |
8192 | imu->nr_bvecs = nr_pages; | |
8193 | ||
8194 | ctx->nr_user_bufs++; | |
8195 | } | |
d4ef6475 MR |
8196 | kvfree(pages); |
8197 | kvfree(vmas); | |
edafccee JA |
8198 | return 0; |
8199 | err: | |
d4ef6475 MR |
8200 | kvfree(pages); |
8201 | kvfree(vmas); | |
edafccee JA |
8202 | io_sqe_buffer_unregister(ctx); |
8203 | return ret; | |
8204 | } | |
8205 | ||
9b402849 JA |
8206 | static int io_eventfd_register(struct io_ring_ctx *ctx, void __user *arg) |
8207 | { | |
8208 | __s32 __user *fds = arg; | |
8209 | int fd; | |
8210 | ||
8211 | if (ctx->cq_ev_fd) | |
8212 | return -EBUSY; | |
8213 | ||
8214 | if (copy_from_user(&fd, fds, sizeof(*fds))) | |
8215 | return -EFAULT; | |
8216 | ||
8217 | ctx->cq_ev_fd = eventfd_ctx_fdget(fd); | |
8218 | if (IS_ERR(ctx->cq_ev_fd)) { | |
8219 | int ret = PTR_ERR(ctx->cq_ev_fd); | |
8220 | ctx->cq_ev_fd = NULL; | |
8221 | return ret; | |
8222 | } | |
8223 | ||
8224 | return 0; | |
8225 | } | |
8226 | ||
8227 | static int io_eventfd_unregister(struct io_ring_ctx *ctx) | |
8228 | { | |
8229 | if (ctx->cq_ev_fd) { | |
8230 | eventfd_ctx_put(ctx->cq_ev_fd); | |
8231 | ctx->cq_ev_fd = NULL; | |
8232 | return 0; | |
8233 | } | |
8234 | ||
8235 | return -ENXIO; | |
8236 | } | |
8237 | ||
5a2e745d JA |
8238 | static int __io_destroy_buffers(int id, void *p, void *data) |
8239 | { | |
8240 | struct io_ring_ctx *ctx = data; | |
8241 | struct io_buffer *buf = p; | |
8242 | ||
067524e9 | 8243 | __io_remove_buffers(ctx, buf, id, -1U); |
5a2e745d JA |
8244 | return 0; |
8245 | } | |
8246 | ||
8247 | static void io_destroy_buffers(struct io_ring_ctx *ctx) | |
8248 | { | |
8249 | idr_for_each(&ctx->io_buffer_idr, __io_destroy_buffers, ctx); | |
8250 | idr_destroy(&ctx->io_buffer_idr); | |
8251 | } | |
8252 | ||
2b188cc1 JA |
8253 | static void io_ring_ctx_free(struct io_ring_ctx *ctx) |
8254 | { | |
6b06314c | 8255 | io_finish_async(ctx); |
5dbcad51 | 8256 | io_sqe_buffer_unregister(ctx); |
2aede0e4 JA |
8257 | |
8258 | if (ctx->sqo_task) { | |
8259 | put_task_struct(ctx->sqo_task); | |
8260 | ctx->sqo_task = NULL; | |
8261 | mmdrop(ctx->mm_account); | |
8262 | ctx->mm_account = NULL; | |
30975825 | 8263 | } |
def596e9 | 8264 | |
91d8f519 DZ |
8265 | #ifdef CONFIG_BLK_CGROUP |
8266 | if (ctx->sqo_blkcg_css) | |
8267 | css_put(ctx->sqo_blkcg_css); | |
8268 | #endif | |
8269 | ||
6b06314c | 8270 | io_sqe_files_unregister(ctx); |
9b402849 | 8271 | io_eventfd_unregister(ctx); |
5a2e745d | 8272 | io_destroy_buffers(ctx); |
41726c9a | 8273 | idr_destroy(&ctx->personality_idr); |
def596e9 | 8274 | |
2b188cc1 | 8275 | #if defined(CONFIG_UNIX) |
355e8d26 EB |
8276 | if (ctx->ring_sock) { |
8277 | ctx->ring_sock->file = NULL; /* so that iput() is called */ | |
2b188cc1 | 8278 | sock_release(ctx->ring_sock); |
355e8d26 | 8279 | } |
2b188cc1 JA |
8280 | #endif |
8281 | ||
75b28aff | 8282 | io_mem_free(ctx->rings); |
2b188cc1 | 8283 | io_mem_free(ctx->sq_sqes); |
2b188cc1 JA |
8284 | |
8285 | percpu_ref_exit(&ctx->refs); | |
2b188cc1 | 8286 | free_uid(ctx->user); |
181e448d | 8287 | put_cred(ctx->creds); |
78076bb6 | 8288 | kfree(ctx->cancel_hash); |
0ddf92e8 | 8289 | kmem_cache_free(req_cachep, ctx->fallback_req); |
2b188cc1 JA |
8290 | kfree(ctx); |
8291 | } | |
8292 | ||
8293 | static __poll_t io_uring_poll(struct file *file, poll_table *wait) | |
8294 | { | |
8295 | struct io_ring_ctx *ctx = file->private_data; | |
8296 | __poll_t mask = 0; | |
8297 | ||
8298 | poll_wait(file, &ctx->cq_wait, wait); | |
4f7067c3 SB |
8299 | /* |
8300 | * synchronizes with barrier from wq_has_sleeper call in | |
8301 | * io_commit_cqring | |
8302 | */ | |
2b188cc1 | 8303 | smp_rmb(); |
90554200 | 8304 | if (!io_sqring_full(ctx)) |
2b188cc1 | 8305 | mask |= EPOLLOUT | EPOLLWRNORM; |
63e5d81f | 8306 | if (io_cqring_events(ctx, false)) |
2b188cc1 JA |
8307 | mask |= EPOLLIN | EPOLLRDNORM; |
8308 | ||
8309 | return mask; | |
8310 | } | |
8311 | ||
8312 | static int io_uring_fasync(int fd, struct file *file, int on) | |
8313 | { | |
8314 | struct io_ring_ctx *ctx = file->private_data; | |
8315 | ||
8316 | return fasync_helper(fd, file, on, &ctx->cq_fasync); | |
8317 | } | |
8318 | ||
071698e1 JA |
8319 | static int io_remove_personalities(int id, void *p, void *data) |
8320 | { | |
8321 | struct io_ring_ctx *ctx = data; | |
1e6fa521 | 8322 | struct io_identity *iod; |
071698e1 | 8323 | |
1e6fa521 JA |
8324 | iod = idr_remove(&ctx->personality_idr, id); |
8325 | if (iod) { | |
8326 | put_cred(iod->creds); | |
8327 | if (refcount_dec_and_test(&iod->count)) | |
8328 | kfree(iod); | |
8329 | } | |
071698e1 JA |
8330 | return 0; |
8331 | } | |
8332 | ||
85faa7b8 JA |
8333 | static void io_ring_exit_work(struct work_struct *work) |
8334 | { | |
b2edc0a7 PB |
8335 | struct io_ring_ctx *ctx = container_of(work, struct io_ring_ctx, |
8336 | exit_work); | |
85faa7b8 | 8337 | |
56952e91 JA |
8338 | /* |
8339 | * If we're doing polled IO and end up having requests being | |
8340 | * submitted async (out-of-line), then completions can come in while | |
8341 | * we're waiting for refs to drop. We need to reap these manually, | |
8342 | * as nobody else will be looking for them. | |
8343 | */ | |
b2edc0a7 | 8344 | do { |
56952e91 | 8345 | if (ctx->rings) |
e6c8aa9a | 8346 | io_cqring_overflow_flush(ctx, true, NULL, NULL); |
b2edc0a7 PB |
8347 | io_iopoll_try_reap_events(ctx); |
8348 | } while (!wait_for_completion_timeout(&ctx->ref_comp, HZ/20)); | |
85faa7b8 JA |
8349 | io_ring_ctx_free(ctx); |
8350 | } | |
8351 | ||
2b188cc1 JA |
8352 | static void io_ring_ctx_wait_and_kill(struct io_ring_ctx *ctx) |
8353 | { | |
8354 | mutex_lock(&ctx->uring_lock); | |
8355 | percpu_ref_kill(&ctx->refs); | |
8356 | mutex_unlock(&ctx->uring_lock); | |
8357 | ||
f3606e3a JA |
8358 | io_kill_timeouts(ctx, NULL); |
8359 | io_poll_remove_all(ctx, NULL); | |
561fb04a JA |
8360 | |
8361 | if (ctx->io_wq) | |
8362 | io_wq_cancel_all(ctx->io_wq); | |
8363 | ||
15dff286 JA |
8364 | /* if we failed setting up the ctx, we might not have any rings */ |
8365 | if (ctx->rings) | |
e6c8aa9a | 8366 | io_cqring_overflow_flush(ctx, true, NULL, NULL); |
b2edc0a7 | 8367 | io_iopoll_try_reap_events(ctx); |
071698e1 | 8368 | idr_for_each(&ctx->personality_idr, io_remove_personalities, ctx); |
309fc03a JA |
8369 | |
8370 | /* | |
8371 | * Do this upfront, so we won't have a grace period where the ring | |
8372 | * is closed but resources aren't reaped yet. This can cause | |
8373 | * spurious failure in setting up a new ring. | |
8374 | */ | |
760618f7 JA |
8375 | io_unaccount_mem(ctx, ring_pages(ctx->sq_entries, ctx->cq_entries), |
8376 | ACCT_LOCKED); | |
309fc03a | 8377 | |
85faa7b8 | 8378 | INIT_WORK(&ctx->exit_work, io_ring_exit_work); |
fc666777 JA |
8379 | /* |
8380 | * Use system_unbound_wq to avoid spawning tons of event kworkers | |
8381 | * if we're exiting a ton of rings at the same time. It just adds | |
8382 | * noise and overhead, there's no discernable change in runtime | |
8383 | * over using system_wq. | |
8384 | */ | |
8385 | queue_work(system_unbound_wq, &ctx->exit_work); | |
2b188cc1 JA |
8386 | } |
8387 | ||
8388 | static int io_uring_release(struct inode *inode, struct file *file) | |
8389 | { | |
8390 | struct io_ring_ctx *ctx = file->private_data; | |
8391 | ||
8392 | file->private_data = NULL; | |
8393 | io_ring_ctx_wait_and_kill(ctx); | |
8394 | return 0; | |
8395 | } | |
8396 | ||
67c4d9e6 PB |
8397 | static bool io_wq_files_match(struct io_wq_work *work, void *data) |
8398 | { | |
8399 | struct files_struct *files = data; | |
8400 | ||
dfead8a8 | 8401 | return !files || ((work->flags & IO_WQ_WORK_FILES) && |
98447d65 | 8402 | work->identity->files == files); |
67c4d9e6 PB |
8403 | } |
8404 | ||
f254ac04 JA |
8405 | /* |
8406 | * Returns true if 'preq' is the link parent of 'req' | |
8407 | */ | |
8408 | static bool io_match_link(struct io_kiocb *preq, struct io_kiocb *req) | |
8409 | { | |
8410 | struct io_kiocb *link; | |
8411 | ||
8412 | if (!(preq->flags & REQ_F_LINK_HEAD)) | |
8413 | return false; | |
8414 | ||
8415 | list_for_each_entry(link, &preq->link_list, link_list) { | |
8416 | if (link == req) | |
8417 | return true; | |
8418 | } | |
8419 | ||
8420 | return false; | |
8421 | } | |
8422 | ||
c127a2a1 PB |
8423 | static bool io_match_link_files(struct io_kiocb *req, |
8424 | struct files_struct *files) | |
8425 | { | |
8426 | struct io_kiocb *link; | |
8427 | ||
8428 | if (io_match_files(req, files)) | |
8429 | return true; | |
8430 | if (req->flags & REQ_F_LINK_HEAD) { | |
8431 | list_for_each_entry(link, &req->link_list, link_list) { | |
8432 | if (io_match_files(link, files)) | |
8433 | return true; | |
8434 | } | |
8435 | } | |
8436 | return false; | |
8437 | } | |
8438 | ||
f254ac04 JA |
8439 | /* |
8440 | * We're looking to cancel 'req' because it's holding on to our files, but | |
8441 | * 'req' could be a link to another request. See if it is, and cancel that | |
8442 | * parent request if so. | |
8443 | */ | |
8444 | static bool io_poll_remove_link(struct io_ring_ctx *ctx, struct io_kiocb *req) | |
8445 | { | |
8446 | struct hlist_node *tmp; | |
8447 | struct io_kiocb *preq; | |
8448 | bool found = false; | |
8449 | int i; | |
8450 | ||
8451 | spin_lock_irq(&ctx->completion_lock); | |
8452 | for (i = 0; i < (1U << ctx->cancel_hash_bits); i++) { | |
8453 | struct hlist_head *list; | |
8454 | ||
8455 | list = &ctx->cancel_hash[i]; | |
8456 | hlist_for_each_entry_safe(preq, tmp, list, hash_node) { | |
8457 | found = io_match_link(preq, req); | |
8458 | if (found) { | |
8459 | io_poll_remove_one(preq); | |
8460 | break; | |
8461 | } | |
8462 | } | |
8463 | } | |
8464 | spin_unlock_irq(&ctx->completion_lock); | |
8465 | return found; | |
8466 | } | |
8467 | ||
8468 | static bool io_timeout_remove_link(struct io_ring_ctx *ctx, | |
8469 | struct io_kiocb *req) | |
8470 | { | |
8471 | struct io_kiocb *preq; | |
8472 | bool found = false; | |
8473 | ||
8474 | spin_lock_irq(&ctx->completion_lock); | |
8475 | list_for_each_entry(preq, &ctx->timeout_list, timeout.list) { | |
8476 | found = io_match_link(preq, req); | |
8477 | if (found) { | |
8478 | __io_timeout_cancel(preq); | |
8479 | break; | |
8480 | } | |
8481 | } | |
8482 | spin_unlock_irq(&ctx->completion_lock); | |
8483 | return found; | |
8484 | } | |
8485 | ||
b711d4ea JA |
8486 | static bool io_cancel_link_cb(struct io_wq_work *work, void *data) |
8487 | { | |
8488 | return io_match_link(container_of(work, struct io_kiocb, work), data); | |
8489 | } | |
8490 | ||
8491 | static void io_attempt_cancel(struct io_ring_ctx *ctx, struct io_kiocb *req) | |
8492 | { | |
8493 | enum io_wq_cancel cret; | |
8494 | ||
8495 | /* cancel this particular work, if it's running */ | |
8496 | cret = io_wq_cancel_work(ctx->io_wq, &req->work); | |
8497 | if (cret != IO_WQ_CANCEL_NOTFOUND) | |
8498 | return; | |
8499 | ||
8500 | /* find links that hold this pending, cancel those */ | |
8501 | cret = io_wq_cancel_cb(ctx->io_wq, io_cancel_link_cb, req, true); | |
8502 | if (cret != IO_WQ_CANCEL_NOTFOUND) | |
8503 | return; | |
8504 | ||
8505 | /* if we have a poll link holding this pending, cancel that */ | |
8506 | if (io_poll_remove_link(ctx, req)) | |
8507 | return; | |
8508 | ||
8509 | /* final option, timeout link is holding this req pending */ | |
8510 | io_timeout_remove_link(ctx, req); | |
8511 | } | |
8512 | ||
b7ddce3c PB |
8513 | static void io_cancel_defer_files(struct io_ring_ctx *ctx, |
8514 | struct files_struct *files) | |
8515 | { | |
8516 | struct io_defer_entry *de = NULL; | |
8517 | LIST_HEAD(list); | |
8518 | ||
8519 | spin_lock_irq(&ctx->completion_lock); | |
8520 | list_for_each_entry_reverse(de, &ctx->defer_list, list) { | |
c127a2a1 | 8521 | if (io_match_link_files(de->req, files)) { |
b7ddce3c PB |
8522 | list_cut_position(&list, &ctx->defer_list, &de->list); |
8523 | break; | |
8524 | } | |
8525 | } | |
8526 | spin_unlock_irq(&ctx->completion_lock); | |
8527 | ||
8528 | while (!list_empty(&list)) { | |
8529 | de = list_first_entry(&list, struct io_defer_entry, list); | |
8530 | list_del_init(&de->list); | |
8531 | req_set_fail_links(de->req); | |
8532 | io_put_req(de->req); | |
8533 | io_req_complete(de->req, -ECANCELED); | |
8534 | kfree(de); | |
8535 | } | |
8536 | } | |
8537 | ||
76e1b642 JA |
8538 | /* |
8539 | * Returns true if we found and killed one or more files pinning requests | |
8540 | */ | |
8541 | static bool io_uring_cancel_files(struct io_ring_ctx *ctx, | |
fcb323cc JA |
8542 | struct files_struct *files) |
8543 | { | |
67c4d9e6 | 8544 | if (list_empty_careful(&ctx->inflight_list)) |
76e1b642 | 8545 | return false; |
67c4d9e6 | 8546 | |
b7ddce3c | 8547 | io_cancel_defer_files(ctx, files); |
67c4d9e6 PB |
8548 | /* cancel all at once, should be faster than doing it one by one*/ |
8549 | io_wq_cancel_cb(ctx->io_wq, io_wq_files_match, files, true); | |
8550 | ||
fcb323cc | 8551 | while (!list_empty_careful(&ctx->inflight_list)) { |
d8f1b971 XW |
8552 | struct io_kiocb *cancel_req = NULL, *req; |
8553 | DEFINE_WAIT(wait); | |
fcb323cc JA |
8554 | |
8555 | spin_lock_irq(&ctx->inflight_lock); | |
8556 | list_for_each_entry(req, &ctx->inflight_list, inflight_entry) { | |
dfead8a8 | 8557 | if (files && (req->work.flags & IO_WQ_WORK_FILES) && |
98447d65 | 8558 | req->work.identity->files != files) |
768134d4 JA |
8559 | continue; |
8560 | /* req is being completed, ignore */ | |
8561 | if (!refcount_inc_not_zero(&req->refs)) | |
8562 | continue; | |
8563 | cancel_req = req; | |
8564 | break; | |
fcb323cc | 8565 | } |
768134d4 | 8566 | if (cancel_req) |
fcb323cc | 8567 | prepare_to_wait(&ctx->inflight_wait, &wait, |
768134d4 | 8568 | TASK_UNINTERRUPTIBLE); |
fcb323cc JA |
8569 | spin_unlock_irq(&ctx->inflight_lock); |
8570 | ||
768134d4 JA |
8571 | /* We need to keep going until we don't find a matching req */ |
8572 | if (!cancel_req) | |
fcb323cc | 8573 | break; |
bb175342 PB |
8574 | /* cancel this request, or head link requests */ |
8575 | io_attempt_cancel(ctx, cancel_req); | |
8576 | io_put_req(cancel_req); | |
6200b0ae JA |
8577 | /* cancellations _may_ trigger task work */ |
8578 | io_run_task_work(); | |
fcb323cc | 8579 | schedule(); |
d8f1b971 | 8580 | finish_wait(&ctx->inflight_wait, &wait); |
fcb323cc | 8581 | } |
76e1b642 JA |
8582 | |
8583 | return true; | |
fcb323cc JA |
8584 | } |
8585 | ||
801dd57b | 8586 | static bool io_cancel_task_cb(struct io_wq_work *work, void *data) |
44e728b8 | 8587 | { |
801dd57b PB |
8588 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); |
8589 | struct task_struct *task = data; | |
44e728b8 | 8590 | |
f3606e3a | 8591 | return io_task_match(req, task); |
44e728b8 PB |
8592 | } |
8593 | ||
0f212204 JA |
8594 | static bool __io_uring_cancel_task_requests(struct io_ring_ctx *ctx, |
8595 | struct task_struct *task, | |
8596 | struct files_struct *files) | |
8597 | { | |
8598 | bool ret; | |
8599 | ||
8600 | ret = io_uring_cancel_files(ctx, files); | |
8601 | if (!files) { | |
8602 | enum io_wq_cancel cret; | |
8603 | ||
8604 | cret = io_wq_cancel_cb(ctx->io_wq, io_cancel_task_cb, task, true); | |
8605 | if (cret != IO_WQ_CANCEL_NOTFOUND) | |
8606 | ret = true; | |
8607 | ||
8608 | /* SQPOLL thread does its own polling */ | |
8609 | if (!(ctx->flags & IORING_SETUP_SQPOLL)) { | |
8610 | while (!list_empty_careful(&ctx->iopoll_list)) { | |
8611 | io_iopoll_try_reap_events(ctx); | |
8612 | ret = true; | |
8613 | } | |
8614 | } | |
8615 | ||
8616 | ret |= io_poll_remove_all(ctx, task); | |
8617 | ret |= io_kill_timeouts(ctx, task); | |
8618 | } | |
8619 | ||
8620 | return ret; | |
8621 | } | |
8622 | ||
8623 | /* | |
8624 | * We need to iteratively cancel requests, in case a request has dependent | |
8625 | * hard links. These persist even for failure of cancelations, hence keep | |
8626 | * looping until none are found. | |
8627 | */ | |
8628 | static void io_uring_cancel_task_requests(struct io_ring_ctx *ctx, | |
8629 | struct files_struct *files) | |
8630 | { | |
8631 | struct task_struct *task = current; | |
8632 | ||
534ca6d6 JA |
8633 | if ((ctx->flags & IORING_SETUP_SQPOLL) && ctx->sq_data) |
8634 | task = ctx->sq_data->thread; | |
0f212204 JA |
8635 | |
8636 | io_cqring_overflow_flush(ctx, true, task, files); | |
8637 | ||
8638 | while (__io_uring_cancel_task_requests(ctx, task, files)) { | |
8639 | io_run_task_work(); | |
8640 | cond_resched(); | |
8641 | } | |
8642 | } | |
8643 | ||
8644 | /* | |
8645 | * Note that this task has used io_uring. We use it for cancelation purposes. | |
8646 | */ | |
8647 | static int io_uring_add_task_file(struct file *file) | |
8648 | { | |
236434c3 MWO |
8649 | struct io_uring_task *tctx = current->io_uring; |
8650 | ||
8651 | if (unlikely(!tctx)) { | |
0f212204 JA |
8652 | int ret; |
8653 | ||
8654 | ret = io_uring_alloc_task_context(current); | |
8655 | if (unlikely(ret)) | |
8656 | return ret; | |
236434c3 | 8657 | tctx = current->io_uring; |
0f212204 | 8658 | } |
236434c3 MWO |
8659 | if (tctx->last != file) { |
8660 | void *old = xa_load(&tctx->xa, (unsigned long)file); | |
0f212204 | 8661 | |
236434c3 | 8662 | if (!old) { |
0f212204 | 8663 | get_file(file); |
236434c3 | 8664 | xa_store(&tctx->xa, (unsigned long)file, file, GFP_KERNEL); |
0f212204 | 8665 | } |
236434c3 | 8666 | tctx->last = file; |
0f212204 JA |
8667 | } |
8668 | ||
8669 | return 0; | |
8670 | } | |
8671 | ||
8672 | /* | |
8673 | * Remove this io_uring_file -> task mapping. | |
8674 | */ | |
8675 | static void io_uring_del_task_file(struct file *file) | |
8676 | { | |
8677 | struct io_uring_task *tctx = current->io_uring; | |
0f212204 JA |
8678 | |
8679 | if (tctx->last == file) | |
8680 | tctx->last = NULL; | |
5e2ed8c4 | 8681 | file = xa_erase(&tctx->xa, (unsigned long)file); |
0f212204 JA |
8682 | if (file) |
8683 | fput(file); | |
8684 | } | |
8685 | ||
8686 | static void __io_uring_attempt_task_drop(struct file *file) | |
8687 | { | |
5e2ed8c4 | 8688 | struct file *old = xa_load(¤t->io_uring->xa, (unsigned long)file); |
0f212204 JA |
8689 | |
8690 | if (old == file) | |
8691 | io_uring_del_task_file(file); | |
8692 | } | |
8693 | ||
8694 | /* | |
8695 | * Drop task note for this file if we're the only ones that hold it after | |
8696 | * pending fput() | |
8697 | */ | |
8698 | static void io_uring_attempt_task_drop(struct file *file, bool exiting) | |
8699 | { | |
8700 | if (!current->io_uring) | |
8701 | return; | |
8702 | /* | |
8703 | * fput() is pending, will be 2 if the only other ref is our potential | |
8704 | * task file note. If the task is exiting, drop regardless of count. | |
8705 | */ | |
8706 | if (!exiting && atomic_long_read(&file->f_count) != 2) | |
8707 | return; | |
8708 | ||
8709 | __io_uring_attempt_task_drop(file); | |
8710 | } | |
8711 | ||
8712 | void __io_uring_files_cancel(struct files_struct *files) | |
8713 | { | |
8714 | struct io_uring_task *tctx = current->io_uring; | |
ce765372 MWO |
8715 | struct file *file; |
8716 | unsigned long index; | |
0f212204 JA |
8717 | |
8718 | /* make sure overflow events are dropped */ | |
8719 | tctx->in_idle = true; | |
8720 | ||
ce765372 MWO |
8721 | xa_for_each(&tctx->xa, index, file) { |
8722 | struct io_ring_ctx *ctx = file->private_data; | |
0f212204 JA |
8723 | |
8724 | io_uring_cancel_task_requests(ctx, files); | |
8725 | if (files) | |
8726 | io_uring_del_task_file(file); | |
ce765372 | 8727 | } |
0f212204 JA |
8728 | } |
8729 | ||
0f212204 JA |
8730 | /* |
8731 | * Find any io_uring fd that this task has registered or done IO on, and cancel | |
8732 | * requests. | |
8733 | */ | |
8734 | void __io_uring_task_cancel(void) | |
8735 | { | |
8736 | struct io_uring_task *tctx = current->io_uring; | |
8737 | DEFINE_WAIT(wait); | |
d8a6df10 | 8738 | s64 inflight; |
0f212204 JA |
8739 | |
8740 | /* make sure overflow events are dropped */ | |
8741 | tctx->in_idle = true; | |
8742 | ||
d8a6df10 | 8743 | do { |
0f212204 | 8744 | /* read completions before cancelations */ |
d8a6df10 JA |
8745 | inflight = percpu_counter_sum(&tctx->inflight); |
8746 | if (!inflight) | |
8747 | break; | |
0f212204 JA |
8748 | __io_uring_files_cancel(NULL); |
8749 | ||
8750 | prepare_to_wait(&tctx->wait, &wait, TASK_UNINTERRUPTIBLE); | |
8751 | ||
8752 | /* | |
8753 | * If we've seen completions, retry. This avoids a race where | |
8754 | * a completion comes in before we did prepare_to_wait(). | |
8755 | */ | |
d8a6df10 | 8756 | if (inflight != percpu_counter_sum(&tctx->inflight)) |
0f212204 | 8757 | continue; |
0f212204 | 8758 | schedule(); |
d8a6df10 | 8759 | } while (1); |
0f212204 JA |
8760 | |
8761 | finish_wait(&tctx->wait, &wait); | |
8762 | tctx->in_idle = false; | |
44e728b8 PB |
8763 | } |
8764 | ||
fcb323cc JA |
8765 | static int io_uring_flush(struct file *file, void *data) |
8766 | { | |
8767 | struct io_ring_ctx *ctx = file->private_data; | |
8768 | ||
6ab23144 JA |
8769 | /* |
8770 | * If the task is going away, cancel work it may have pending | |
8771 | */ | |
801dd57b | 8772 | if (fatal_signal_pending(current) || (current->flags & PF_EXITING)) |
0f212204 | 8773 | data = NULL; |
6ab23144 | 8774 | |
0f212204 JA |
8775 | io_uring_cancel_task_requests(ctx, data); |
8776 | io_uring_attempt_task_drop(file, !data); | |
fcb323cc JA |
8777 | return 0; |
8778 | } | |
8779 | ||
6c5c240e RP |
8780 | static void *io_uring_validate_mmap_request(struct file *file, |
8781 | loff_t pgoff, size_t sz) | |
2b188cc1 | 8782 | { |
2b188cc1 | 8783 | struct io_ring_ctx *ctx = file->private_data; |
6c5c240e | 8784 | loff_t offset = pgoff << PAGE_SHIFT; |
2b188cc1 JA |
8785 | struct page *page; |
8786 | void *ptr; | |
8787 | ||
8788 | switch (offset) { | |
8789 | case IORING_OFF_SQ_RING: | |
75b28aff HV |
8790 | case IORING_OFF_CQ_RING: |
8791 | ptr = ctx->rings; | |
2b188cc1 JA |
8792 | break; |
8793 | case IORING_OFF_SQES: | |
8794 | ptr = ctx->sq_sqes; | |
8795 | break; | |
2b188cc1 | 8796 | default: |
6c5c240e | 8797 | return ERR_PTR(-EINVAL); |
2b188cc1 JA |
8798 | } |
8799 | ||
8800 | page = virt_to_head_page(ptr); | |
a50b854e | 8801 | if (sz > page_size(page)) |
6c5c240e RP |
8802 | return ERR_PTR(-EINVAL); |
8803 | ||
8804 | return ptr; | |
8805 | } | |
8806 | ||
8807 | #ifdef CONFIG_MMU | |
8808 | ||
8809 | static int io_uring_mmap(struct file *file, struct vm_area_struct *vma) | |
8810 | { | |
8811 | size_t sz = vma->vm_end - vma->vm_start; | |
8812 | unsigned long pfn; | |
8813 | void *ptr; | |
8814 | ||
8815 | ptr = io_uring_validate_mmap_request(file, vma->vm_pgoff, sz); | |
8816 | if (IS_ERR(ptr)) | |
8817 | return PTR_ERR(ptr); | |
2b188cc1 JA |
8818 | |
8819 | pfn = virt_to_phys(ptr) >> PAGE_SHIFT; | |
8820 | return remap_pfn_range(vma, vma->vm_start, pfn, sz, vma->vm_page_prot); | |
8821 | } | |
8822 | ||
6c5c240e RP |
8823 | #else /* !CONFIG_MMU */ |
8824 | ||
8825 | static int io_uring_mmap(struct file *file, struct vm_area_struct *vma) | |
8826 | { | |
8827 | return vma->vm_flags & (VM_SHARED | VM_MAYSHARE) ? 0 : -EINVAL; | |
8828 | } | |
8829 | ||
8830 | static unsigned int io_uring_nommu_mmap_capabilities(struct file *file) | |
8831 | { | |
8832 | return NOMMU_MAP_DIRECT | NOMMU_MAP_READ | NOMMU_MAP_WRITE; | |
8833 | } | |
8834 | ||
8835 | static unsigned long io_uring_nommu_get_unmapped_area(struct file *file, | |
8836 | unsigned long addr, unsigned long len, | |
8837 | unsigned long pgoff, unsigned long flags) | |
8838 | { | |
8839 | void *ptr; | |
8840 | ||
8841 | ptr = io_uring_validate_mmap_request(file, pgoff, len); | |
8842 | if (IS_ERR(ptr)) | |
8843 | return PTR_ERR(ptr); | |
8844 | ||
8845 | return (unsigned long) ptr; | |
8846 | } | |
8847 | ||
8848 | #endif /* !CONFIG_MMU */ | |
8849 | ||
90554200 JA |
8850 | static void io_sqpoll_wait_sq(struct io_ring_ctx *ctx) |
8851 | { | |
8852 | DEFINE_WAIT(wait); | |
8853 | ||
8854 | do { | |
8855 | if (!io_sqring_full(ctx)) | |
8856 | break; | |
8857 | ||
8858 | prepare_to_wait(&ctx->sqo_sq_wait, &wait, TASK_INTERRUPTIBLE); | |
8859 | ||
8860 | if (!io_sqring_full(ctx)) | |
8861 | break; | |
8862 | ||
8863 | schedule(); | |
8864 | } while (!signal_pending(current)); | |
8865 | ||
8866 | finish_wait(&ctx->sqo_sq_wait, &wait); | |
8867 | } | |
8868 | ||
2b188cc1 JA |
8869 | SYSCALL_DEFINE6(io_uring_enter, unsigned int, fd, u32, to_submit, |
8870 | u32, min_complete, u32, flags, const sigset_t __user *, sig, | |
8871 | size_t, sigsz) | |
8872 | { | |
8873 | struct io_ring_ctx *ctx; | |
8874 | long ret = -EBADF; | |
8875 | int submitted = 0; | |
8876 | struct fd f; | |
8877 | ||
4c6e277c | 8878 | io_run_task_work(); |
b41e9852 | 8879 | |
90554200 JA |
8880 | if (flags & ~(IORING_ENTER_GETEVENTS | IORING_ENTER_SQ_WAKEUP | |
8881 | IORING_ENTER_SQ_WAIT)) | |
2b188cc1 JA |
8882 | return -EINVAL; |
8883 | ||
8884 | f = fdget(fd); | |
8885 | if (!f.file) | |
8886 | return -EBADF; | |
8887 | ||
8888 | ret = -EOPNOTSUPP; | |
8889 | if (f.file->f_op != &io_uring_fops) | |
8890 | goto out_fput; | |
8891 | ||
8892 | ret = -ENXIO; | |
8893 | ctx = f.file->private_data; | |
8894 | if (!percpu_ref_tryget(&ctx->refs)) | |
8895 | goto out_fput; | |
8896 | ||
7e84e1c7 SG |
8897 | ret = -EBADFD; |
8898 | if (ctx->flags & IORING_SETUP_R_DISABLED) | |
8899 | goto out; | |
8900 | ||
6c271ce2 JA |
8901 | /* |
8902 | * For SQ polling, the thread will do all submissions and completions. | |
8903 | * Just return the requested submit count, and wake the thread if | |
8904 | * we were asked to. | |
8905 | */ | |
b2a9eada | 8906 | ret = 0; |
6c271ce2 | 8907 | if (ctx->flags & IORING_SETUP_SQPOLL) { |
c1edbf5f | 8908 | if (!list_empty_careful(&ctx->cq_overflow_list)) |
e6c8aa9a | 8909 | io_cqring_overflow_flush(ctx, false, NULL, NULL); |
6c271ce2 | 8910 | if (flags & IORING_ENTER_SQ_WAKEUP) |
534ca6d6 | 8911 | wake_up(&ctx->sq_data->wait); |
90554200 JA |
8912 | if (flags & IORING_ENTER_SQ_WAIT) |
8913 | io_sqpoll_wait_sq(ctx); | |
6c271ce2 | 8914 | submitted = to_submit; |
b2a9eada | 8915 | } else if (to_submit) { |
0f212204 JA |
8916 | ret = io_uring_add_task_file(f.file); |
8917 | if (unlikely(ret)) | |
8918 | goto out; | |
2b188cc1 | 8919 | mutex_lock(&ctx->uring_lock); |
0f212204 | 8920 | submitted = io_submit_sqes(ctx, to_submit); |
2b188cc1 | 8921 | mutex_unlock(&ctx->uring_lock); |
7c504e65 PB |
8922 | |
8923 | if (submitted != to_submit) | |
8924 | goto out; | |
2b188cc1 JA |
8925 | } |
8926 | if (flags & IORING_ENTER_GETEVENTS) { | |
8927 | min_complete = min(min_complete, ctx->cq_entries); | |
8928 | ||
32b2244a XW |
8929 | /* |
8930 | * When SETUP_IOPOLL and SETUP_SQPOLL are both enabled, user | |
8931 | * space applications don't need to do io completion events | |
8932 | * polling again, they can rely on io_sq_thread to do polling | |
8933 | * work, which can reduce cpu usage and uring_lock contention. | |
8934 | */ | |
8935 | if (ctx->flags & IORING_SETUP_IOPOLL && | |
8936 | !(ctx->flags & IORING_SETUP_SQPOLL)) { | |
7668b92a | 8937 | ret = io_iopoll_check(ctx, min_complete); |
def596e9 JA |
8938 | } else { |
8939 | ret = io_cqring_wait(ctx, min_complete, sig, sigsz); | |
8940 | } | |
2b188cc1 JA |
8941 | } |
8942 | ||
7c504e65 | 8943 | out: |
6805b32e | 8944 | percpu_ref_put(&ctx->refs); |
2b188cc1 JA |
8945 | out_fput: |
8946 | fdput(f); | |
8947 | return submitted ? submitted : ret; | |
8948 | } | |
8949 | ||
bebdb65e | 8950 | #ifdef CONFIG_PROC_FS |
87ce955b JA |
8951 | static int io_uring_show_cred(int id, void *p, void *data) |
8952 | { | |
8953 | const struct cred *cred = p; | |
8954 | struct seq_file *m = data; | |
8955 | struct user_namespace *uns = seq_user_ns(m); | |
8956 | struct group_info *gi; | |
8957 | kernel_cap_t cap; | |
8958 | unsigned __capi; | |
8959 | int g; | |
8960 | ||
8961 | seq_printf(m, "%5d\n", id); | |
8962 | seq_put_decimal_ull(m, "\tUid:\t", from_kuid_munged(uns, cred->uid)); | |
8963 | seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->euid)); | |
8964 | seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->suid)); | |
8965 | seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->fsuid)); | |
8966 | seq_put_decimal_ull(m, "\n\tGid:\t", from_kgid_munged(uns, cred->gid)); | |
8967 | seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->egid)); | |
8968 | seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->sgid)); | |
8969 | seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->fsgid)); | |
8970 | seq_puts(m, "\n\tGroups:\t"); | |
8971 | gi = cred->group_info; | |
8972 | for (g = 0; g < gi->ngroups; g++) { | |
8973 | seq_put_decimal_ull(m, g ? " " : "", | |
8974 | from_kgid_munged(uns, gi->gid[g])); | |
8975 | } | |
8976 | seq_puts(m, "\n\tCapEff:\t"); | |
8977 | cap = cred->cap_effective; | |
8978 | CAP_FOR_EACH_U32(__capi) | |
8979 | seq_put_hex_ll(m, NULL, cap.cap[CAP_LAST_U32 - __capi], 8); | |
8980 | seq_putc(m, '\n'); | |
8981 | return 0; | |
8982 | } | |
8983 | ||
8984 | static void __io_uring_show_fdinfo(struct io_ring_ctx *ctx, struct seq_file *m) | |
8985 | { | |
dbbe9c64 | 8986 | struct io_sq_data *sq = NULL; |
fad8e0de | 8987 | bool has_lock; |
87ce955b JA |
8988 | int i; |
8989 | ||
fad8e0de JA |
8990 | /* |
8991 | * Avoid ABBA deadlock between the seq lock and the io_uring mutex, | |
8992 | * since fdinfo case grabs it in the opposite direction of normal use | |
8993 | * cases. If we fail to get the lock, we just don't iterate any | |
8994 | * structures that could be going away outside the io_uring mutex. | |
8995 | */ | |
8996 | has_lock = mutex_trylock(&ctx->uring_lock); | |
8997 | ||
dbbe9c64 JQ |
8998 | if (has_lock && (ctx->flags & IORING_SETUP_SQPOLL)) |
8999 | sq = ctx->sq_data; | |
9000 | ||
9001 | seq_printf(m, "SqThread:\t%d\n", sq ? task_pid_nr(sq->thread) : -1); | |
9002 | seq_printf(m, "SqThreadCpu:\t%d\n", sq ? task_cpu(sq->thread) : -1); | |
87ce955b | 9003 | seq_printf(m, "UserFiles:\t%u\n", ctx->nr_user_files); |
fad8e0de | 9004 | for (i = 0; has_lock && i < ctx->nr_user_files; i++) { |
87ce955b JA |
9005 | struct fixed_file_table *table; |
9006 | struct file *f; | |
9007 | ||
9008 | table = &ctx->file_data->table[i >> IORING_FILE_TABLE_SHIFT]; | |
9009 | f = table->files[i & IORING_FILE_TABLE_MASK]; | |
9010 | if (f) | |
9011 | seq_printf(m, "%5u: %s\n", i, file_dentry(f)->d_iname); | |
9012 | else | |
9013 | seq_printf(m, "%5u: <none>\n", i); | |
9014 | } | |
9015 | seq_printf(m, "UserBufs:\t%u\n", ctx->nr_user_bufs); | |
fad8e0de | 9016 | for (i = 0; has_lock && i < ctx->nr_user_bufs; i++) { |
87ce955b JA |
9017 | struct io_mapped_ubuf *buf = &ctx->user_bufs[i]; |
9018 | ||
9019 | seq_printf(m, "%5u: 0x%llx/%u\n", i, buf->ubuf, | |
9020 | (unsigned int) buf->len); | |
9021 | } | |
fad8e0de | 9022 | if (has_lock && !idr_is_empty(&ctx->personality_idr)) { |
87ce955b JA |
9023 | seq_printf(m, "Personalities:\n"); |
9024 | idr_for_each(&ctx->personality_idr, io_uring_show_cred, m); | |
9025 | } | |
d7718a9d JA |
9026 | seq_printf(m, "PollList:\n"); |
9027 | spin_lock_irq(&ctx->completion_lock); | |
9028 | for (i = 0; i < (1U << ctx->cancel_hash_bits); i++) { | |
9029 | struct hlist_head *list = &ctx->cancel_hash[i]; | |
9030 | struct io_kiocb *req; | |
9031 | ||
9032 | hlist_for_each_entry(req, list, hash_node) | |
9033 | seq_printf(m, " op=%d, task_works=%d\n", req->opcode, | |
9034 | req->task->task_works != NULL); | |
9035 | } | |
9036 | spin_unlock_irq(&ctx->completion_lock); | |
fad8e0de JA |
9037 | if (has_lock) |
9038 | mutex_unlock(&ctx->uring_lock); | |
87ce955b JA |
9039 | } |
9040 | ||
9041 | static void io_uring_show_fdinfo(struct seq_file *m, struct file *f) | |
9042 | { | |
9043 | struct io_ring_ctx *ctx = f->private_data; | |
9044 | ||
9045 | if (percpu_ref_tryget(&ctx->refs)) { | |
9046 | __io_uring_show_fdinfo(ctx, m); | |
9047 | percpu_ref_put(&ctx->refs); | |
9048 | } | |
9049 | } | |
bebdb65e | 9050 | #endif |
87ce955b | 9051 | |
2b188cc1 JA |
9052 | static const struct file_operations io_uring_fops = { |
9053 | .release = io_uring_release, | |
fcb323cc | 9054 | .flush = io_uring_flush, |
2b188cc1 | 9055 | .mmap = io_uring_mmap, |
6c5c240e RP |
9056 | #ifndef CONFIG_MMU |
9057 | .get_unmapped_area = io_uring_nommu_get_unmapped_area, | |
9058 | .mmap_capabilities = io_uring_nommu_mmap_capabilities, | |
9059 | #endif | |
2b188cc1 JA |
9060 | .poll = io_uring_poll, |
9061 | .fasync = io_uring_fasync, | |
bebdb65e | 9062 | #ifdef CONFIG_PROC_FS |
87ce955b | 9063 | .show_fdinfo = io_uring_show_fdinfo, |
bebdb65e | 9064 | #endif |
2b188cc1 JA |
9065 | }; |
9066 | ||
9067 | static int io_allocate_scq_urings(struct io_ring_ctx *ctx, | |
9068 | struct io_uring_params *p) | |
9069 | { | |
75b28aff HV |
9070 | struct io_rings *rings; |
9071 | size_t size, sq_array_offset; | |
2b188cc1 | 9072 | |
bd740481 JA |
9073 | /* make sure these are sane, as we already accounted them */ |
9074 | ctx->sq_entries = p->sq_entries; | |
9075 | ctx->cq_entries = p->cq_entries; | |
9076 | ||
75b28aff HV |
9077 | size = rings_size(p->sq_entries, p->cq_entries, &sq_array_offset); |
9078 | if (size == SIZE_MAX) | |
9079 | return -EOVERFLOW; | |
9080 | ||
9081 | rings = io_mem_alloc(size); | |
9082 | if (!rings) | |
2b188cc1 JA |
9083 | return -ENOMEM; |
9084 | ||
75b28aff HV |
9085 | ctx->rings = rings; |
9086 | ctx->sq_array = (u32 *)((char *)rings + sq_array_offset); | |
9087 | rings->sq_ring_mask = p->sq_entries - 1; | |
9088 | rings->cq_ring_mask = p->cq_entries - 1; | |
9089 | rings->sq_ring_entries = p->sq_entries; | |
9090 | rings->cq_ring_entries = p->cq_entries; | |
9091 | ctx->sq_mask = rings->sq_ring_mask; | |
9092 | ctx->cq_mask = rings->cq_ring_mask; | |
2b188cc1 JA |
9093 | |
9094 | size = array_size(sizeof(struct io_uring_sqe), p->sq_entries); | |
eb065d30 JA |
9095 | if (size == SIZE_MAX) { |
9096 | io_mem_free(ctx->rings); | |
9097 | ctx->rings = NULL; | |
2b188cc1 | 9098 | return -EOVERFLOW; |
eb065d30 | 9099 | } |
2b188cc1 JA |
9100 | |
9101 | ctx->sq_sqes = io_mem_alloc(size); | |
eb065d30 JA |
9102 | if (!ctx->sq_sqes) { |
9103 | io_mem_free(ctx->rings); | |
9104 | ctx->rings = NULL; | |
2b188cc1 | 9105 | return -ENOMEM; |
eb065d30 | 9106 | } |
2b188cc1 | 9107 | |
2b188cc1 JA |
9108 | return 0; |
9109 | } | |
9110 | ||
9111 | /* | |
9112 | * Allocate an anonymous fd, this is what constitutes the application | |
9113 | * visible backing of an io_uring instance. The application mmaps this | |
9114 | * fd to gain access to the SQ/CQ ring details. If UNIX sockets are enabled, | |
9115 | * we have to tie this fd to a socket for file garbage collection purposes. | |
9116 | */ | |
9117 | static int io_uring_get_fd(struct io_ring_ctx *ctx) | |
9118 | { | |
9119 | struct file *file; | |
9120 | int ret; | |
9121 | ||
9122 | #if defined(CONFIG_UNIX) | |
9123 | ret = sock_create_kern(&init_net, PF_UNIX, SOCK_RAW, IPPROTO_IP, | |
9124 | &ctx->ring_sock); | |
9125 | if (ret) | |
9126 | return ret; | |
9127 | #endif | |
9128 | ||
9129 | ret = get_unused_fd_flags(O_RDWR | O_CLOEXEC); | |
9130 | if (ret < 0) | |
9131 | goto err; | |
9132 | ||
9133 | file = anon_inode_getfile("[io_uring]", &io_uring_fops, ctx, | |
9134 | O_RDWR | O_CLOEXEC); | |
9135 | if (IS_ERR(file)) { | |
0f212204 | 9136 | err_fd: |
2b188cc1 JA |
9137 | put_unused_fd(ret); |
9138 | ret = PTR_ERR(file); | |
9139 | goto err; | |
9140 | } | |
9141 | ||
9142 | #if defined(CONFIG_UNIX) | |
9143 | ctx->ring_sock->file = file; | |
9144 | #endif | |
0f212204 JA |
9145 | if (unlikely(io_uring_add_task_file(file))) { |
9146 | file = ERR_PTR(-ENOMEM); | |
9147 | goto err_fd; | |
9148 | } | |
2b188cc1 JA |
9149 | fd_install(ret, file); |
9150 | return ret; | |
9151 | err: | |
9152 | #if defined(CONFIG_UNIX) | |
9153 | sock_release(ctx->ring_sock); | |
9154 | ctx->ring_sock = NULL; | |
9155 | #endif | |
9156 | return ret; | |
9157 | } | |
9158 | ||
7f13657d XW |
9159 | static int io_uring_create(unsigned entries, struct io_uring_params *p, |
9160 | struct io_uring_params __user *params) | |
2b188cc1 JA |
9161 | { |
9162 | struct user_struct *user = NULL; | |
9163 | struct io_ring_ctx *ctx; | |
aad5d8da | 9164 | bool limit_mem; |
2b188cc1 JA |
9165 | int ret; |
9166 | ||
8110c1a6 | 9167 | if (!entries) |
2b188cc1 | 9168 | return -EINVAL; |
8110c1a6 JA |
9169 | if (entries > IORING_MAX_ENTRIES) { |
9170 | if (!(p->flags & IORING_SETUP_CLAMP)) | |
9171 | return -EINVAL; | |
9172 | entries = IORING_MAX_ENTRIES; | |
9173 | } | |
2b188cc1 JA |
9174 | |
9175 | /* | |
9176 | * Use twice as many entries for the CQ ring. It's possible for the | |
9177 | * application to drive a higher depth than the size of the SQ ring, | |
9178 | * since the sqes are only used at submission time. This allows for | |
33a107f0 JA |
9179 | * some flexibility in overcommitting a bit. If the application has |
9180 | * set IORING_SETUP_CQSIZE, it will have passed in the desired number | |
9181 | * of CQ ring entries manually. | |
2b188cc1 JA |
9182 | */ |
9183 | p->sq_entries = roundup_pow_of_two(entries); | |
33a107f0 JA |
9184 | if (p->flags & IORING_SETUP_CQSIZE) { |
9185 | /* | |
9186 | * If IORING_SETUP_CQSIZE is set, we do the same roundup | |
9187 | * to a power-of-two, if it isn't already. We do NOT impose | |
9188 | * any cq vs sq ring sizing. | |
9189 | */ | |
8110c1a6 | 9190 | if (p->cq_entries < p->sq_entries) |
33a107f0 | 9191 | return -EINVAL; |
8110c1a6 JA |
9192 | if (p->cq_entries > IORING_MAX_CQ_ENTRIES) { |
9193 | if (!(p->flags & IORING_SETUP_CLAMP)) | |
9194 | return -EINVAL; | |
9195 | p->cq_entries = IORING_MAX_CQ_ENTRIES; | |
9196 | } | |
33a107f0 JA |
9197 | p->cq_entries = roundup_pow_of_two(p->cq_entries); |
9198 | } else { | |
9199 | p->cq_entries = 2 * p->sq_entries; | |
9200 | } | |
2b188cc1 JA |
9201 | |
9202 | user = get_uid(current_user()); | |
aad5d8da | 9203 | limit_mem = !capable(CAP_IPC_LOCK); |
2b188cc1 | 9204 | |
aad5d8da | 9205 | if (limit_mem) { |
a087e2b5 | 9206 | ret = __io_account_mem(user, |
2b188cc1 JA |
9207 | ring_pages(p->sq_entries, p->cq_entries)); |
9208 | if (ret) { | |
9209 | free_uid(user); | |
9210 | return ret; | |
9211 | } | |
9212 | } | |
9213 | ||
9214 | ctx = io_ring_ctx_alloc(p); | |
9215 | if (!ctx) { | |
aad5d8da | 9216 | if (limit_mem) |
a087e2b5 | 9217 | __io_unaccount_mem(user, ring_pages(p->sq_entries, |
2b188cc1 JA |
9218 | p->cq_entries)); |
9219 | free_uid(user); | |
9220 | return -ENOMEM; | |
9221 | } | |
9222 | ctx->compat = in_compat_syscall(); | |
2b188cc1 | 9223 | ctx->user = user; |
0b8c0ec7 | 9224 | ctx->creds = get_current_cred(); |
4ea33a97 JA |
9225 | #ifdef CONFIG_AUDIT |
9226 | ctx->loginuid = current->loginuid; | |
9227 | ctx->sessionid = current->sessionid; | |
9228 | #endif | |
2aede0e4 JA |
9229 | ctx->sqo_task = get_task_struct(current); |
9230 | ||
9231 | /* | |
9232 | * This is just grabbed for accounting purposes. When a process exits, | |
9233 | * the mm is exited and dropped before the files, hence we need to hang | |
9234 | * on to this mm purely for the purposes of being able to unaccount | |
9235 | * memory (locked/pinned vm). It's not used for anything else. | |
9236 | */ | |
6b7898eb | 9237 | mmgrab(current->mm); |
2aede0e4 | 9238 | ctx->mm_account = current->mm; |
6b7898eb | 9239 | |
91d8f519 DZ |
9240 | #ifdef CONFIG_BLK_CGROUP |
9241 | /* | |
9242 | * The sq thread will belong to the original cgroup it was inited in. | |
9243 | * If the cgroup goes offline (e.g. disabling the io controller), then | |
9244 | * issued bios will be associated with the closest cgroup later in the | |
9245 | * block layer. | |
9246 | */ | |
9247 | rcu_read_lock(); | |
9248 | ctx->sqo_blkcg_css = blkcg_css(); | |
9249 | ret = css_tryget_online(ctx->sqo_blkcg_css); | |
9250 | rcu_read_unlock(); | |
9251 | if (!ret) { | |
9252 | /* don't init against a dying cgroup, have the user try again */ | |
9253 | ctx->sqo_blkcg_css = NULL; | |
9254 | ret = -ENODEV; | |
9255 | goto err; | |
9256 | } | |
9257 | #endif | |
6b7898eb | 9258 | |
f74441e6 JA |
9259 | /* |
9260 | * Account memory _before_ installing the file descriptor. Once | |
9261 | * the descriptor is installed, it can get closed at any time. Also | |
9262 | * do this before hitting the general error path, as ring freeing | |
9263 | * will un-account as well. | |
9264 | */ | |
9265 | io_account_mem(ctx, ring_pages(p->sq_entries, p->cq_entries), | |
9266 | ACCT_LOCKED); | |
9267 | ctx->limit_mem = limit_mem; | |
9268 | ||
2b188cc1 JA |
9269 | ret = io_allocate_scq_urings(ctx, p); |
9270 | if (ret) | |
9271 | goto err; | |
9272 | ||
7e84e1c7 | 9273 | ret = io_sq_offload_create(ctx, p); |
2b188cc1 JA |
9274 | if (ret) |
9275 | goto err; | |
9276 | ||
7e84e1c7 SG |
9277 | if (!(p->flags & IORING_SETUP_R_DISABLED)) |
9278 | io_sq_offload_start(ctx); | |
9279 | ||
2b188cc1 | 9280 | memset(&p->sq_off, 0, sizeof(p->sq_off)); |
75b28aff HV |
9281 | p->sq_off.head = offsetof(struct io_rings, sq.head); |
9282 | p->sq_off.tail = offsetof(struct io_rings, sq.tail); | |
9283 | p->sq_off.ring_mask = offsetof(struct io_rings, sq_ring_mask); | |
9284 | p->sq_off.ring_entries = offsetof(struct io_rings, sq_ring_entries); | |
9285 | p->sq_off.flags = offsetof(struct io_rings, sq_flags); | |
9286 | p->sq_off.dropped = offsetof(struct io_rings, sq_dropped); | |
9287 | p->sq_off.array = (char *)ctx->sq_array - (char *)ctx->rings; | |
2b188cc1 JA |
9288 | |
9289 | memset(&p->cq_off, 0, sizeof(p->cq_off)); | |
75b28aff HV |
9290 | p->cq_off.head = offsetof(struct io_rings, cq.head); |
9291 | p->cq_off.tail = offsetof(struct io_rings, cq.tail); | |
9292 | p->cq_off.ring_mask = offsetof(struct io_rings, cq_ring_mask); | |
9293 | p->cq_off.ring_entries = offsetof(struct io_rings, cq_ring_entries); | |
9294 | p->cq_off.overflow = offsetof(struct io_rings, cq_overflow); | |
9295 | p->cq_off.cqes = offsetof(struct io_rings, cqes); | |
0d9b5b3a | 9296 | p->cq_off.flags = offsetof(struct io_rings, cq_flags); |
ac90f249 | 9297 | |
7f13657d XW |
9298 | p->features = IORING_FEAT_SINGLE_MMAP | IORING_FEAT_NODROP | |
9299 | IORING_FEAT_SUBMIT_STABLE | IORING_FEAT_RW_CUR_POS | | |
5769a351 JX |
9300 | IORING_FEAT_CUR_PERSONALITY | IORING_FEAT_FAST_POLL | |
9301 | IORING_FEAT_POLL_32BITS; | |
7f13657d XW |
9302 | |
9303 | if (copy_to_user(params, p, sizeof(*p))) { | |
9304 | ret = -EFAULT; | |
9305 | goto err; | |
9306 | } | |
d1719f70 | 9307 | |
044c1ab3 JA |
9308 | /* |
9309 | * Install ring fd as the very last thing, so we don't risk someone | |
9310 | * having closed it before we finish setup | |
9311 | */ | |
9312 | ret = io_uring_get_fd(ctx); | |
9313 | if (ret < 0) | |
9314 | goto err; | |
9315 | ||
c826bd7a | 9316 | trace_io_uring_create(ret, ctx, p->sq_entries, p->cq_entries, p->flags); |
2b188cc1 JA |
9317 | return ret; |
9318 | err: | |
9319 | io_ring_ctx_wait_and_kill(ctx); | |
9320 | return ret; | |
9321 | } | |
9322 | ||
9323 | /* | |
9324 | * Sets up an aio uring context, and returns the fd. Applications asks for a | |
9325 | * ring size, we return the actual sq/cq ring sizes (among other things) in the | |
9326 | * params structure passed in. | |
9327 | */ | |
9328 | static long io_uring_setup(u32 entries, struct io_uring_params __user *params) | |
9329 | { | |
9330 | struct io_uring_params p; | |
2b188cc1 JA |
9331 | int i; |
9332 | ||
9333 | if (copy_from_user(&p, params, sizeof(p))) | |
9334 | return -EFAULT; | |
9335 | for (i = 0; i < ARRAY_SIZE(p.resv); i++) { | |
9336 | if (p.resv[i]) | |
9337 | return -EINVAL; | |
9338 | } | |
9339 | ||
6c271ce2 | 9340 | if (p.flags & ~(IORING_SETUP_IOPOLL | IORING_SETUP_SQPOLL | |
8110c1a6 | 9341 | IORING_SETUP_SQ_AFF | IORING_SETUP_CQSIZE | |
7e84e1c7 SG |
9342 | IORING_SETUP_CLAMP | IORING_SETUP_ATTACH_WQ | |
9343 | IORING_SETUP_R_DISABLED)) | |
2b188cc1 JA |
9344 | return -EINVAL; |
9345 | ||
7f13657d | 9346 | return io_uring_create(entries, &p, params); |
2b188cc1 JA |
9347 | } |
9348 | ||
9349 | SYSCALL_DEFINE2(io_uring_setup, u32, entries, | |
9350 | struct io_uring_params __user *, params) | |
9351 | { | |
9352 | return io_uring_setup(entries, params); | |
9353 | } | |
9354 | ||
66f4af93 JA |
9355 | static int io_probe(struct io_ring_ctx *ctx, void __user *arg, unsigned nr_args) |
9356 | { | |
9357 | struct io_uring_probe *p; | |
9358 | size_t size; | |
9359 | int i, ret; | |
9360 | ||
9361 | size = struct_size(p, ops, nr_args); | |
9362 | if (size == SIZE_MAX) | |
9363 | return -EOVERFLOW; | |
9364 | p = kzalloc(size, GFP_KERNEL); | |
9365 | if (!p) | |
9366 | return -ENOMEM; | |
9367 | ||
9368 | ret = -EFAULT; | |
9369 | if (copy_from_user(p, arg, size)) | |
9370 | goto out; | |
9371 | ret = -EINVAL; | |
9372 | if (memchr_inv(p, 0, size)) | |
9373 | goto out; | |
9374 | ||
9375 | p->last_op = IORING_OP_LAST - 1; | |
9376 | if (nr_args > IORING_OP_LAST) | |
9377 | nr_args = IORING_OP_LAST; | |
9378 | ||
9379 | for (i = 0; i < nr_args; i++) { | |
9380 | p->ops[i].op = i; | |
9381 | if (!io_op_defs[i].not_supported) | |
9382 | p->ops[i].flags = IO_URING_OP_SUPPORTED; | |
9383 | } | |
9384 | p->ops_len = i; | |
9385 | ||
9386 | ret = 0; | |
9387 | if (copy_to_user(arg, p, size)) | |
9388 | ret = -EFAULT; | |
9389 | out: | |
9390 | kfree(p); | |
9391 | return ret; | |
9392 | } | |
9393 | ||
071698e1 JA |
9394 | static int io_register_personality(struct io_ring_ctx *ctx) |
9395 | { | |
1e6fa521 JA |
9396 | struct io_identity *id; |
9397 | int ret; | |
071698e1 | 9398 | |
1e6fa521 JA |
9399 | id = kmalloc(sizeof(*id), GFP_KERNEL); |
9400 | if (unlikely(!id)) | |
9401 | return -ENOMEM; | |
9402 | ||
9403 | io_init_identity(id); | |
9404 | id->creds = get_current_cred(); | |
9405 | ||
9406 | ret = idr_alloc_cyclic(&ctx->personality_idr, id, 1, USHRT_MAX, GFP_KERNEL); | |
9407 | if (ret < 0) { | |
9408 | put_cred(id->creds); | |
9409 | kfree(id); | |
9410 | } | |
9411 | return ret; | |
071698e1 JA |
9412 | } |
9413 | ||
9414 | static int io_unregister_personality(struct io_ring_ctx *ctx, unsigned id) | |
9415 | { | |
1e6fa521 | 9416 | struct io_identity *iod; |
071698e1 | 9417 | |
1e6fa521 JA |
9418 | iod = idr_remove(&ctx->personality_idr, id); |
9419 | if (iod) { | |
9420 | put_cred(iod->creds); | |
9421 | if (refcount_dec_and_test(&iod->count)) | |
9422 | kfree(iod); | |
071698e1 JA |
9423 | return 0; |
9424 | } | |
9425 | ||
9426 | return -EINVAL; | |
9427 | } | |
9428 | ||
21b55dbc SG |
9429 | static int io_register_restrictions(struct io_ring_ctx *ctx, void __user *arg, |
9430 | unsigned int nr_args) | |
9431 | { | |
9432 | struct io_uring_restriction *res; | |
9433 | size_t size; | |
9434 | int i, ret; | |
9435 | ||
7e84e1c7 SG |
9436 | /* Restrictions allowed only if rings started disabled */ |
9437 | if (!(ctx->flags & IORING_SETUP_R_DISABLED)) | |
9438 | return -EBADFD; | |
9439 | ||
21b55dbc | 9440 | /* We allow only a single restrictions registration */ |
7e84e1c7 | 9441 | if (ctx->restrictions.registered) |
21b55dbc SG |
9442 | return -EBUSY; |
9443 | ||
9444 | if (!arg || nr_args > IORING_MAX_RESTRICTIONS) | |
9445 | return -EINVAL; | |
9446 | ||
9447 | size = array_size(nr_args, sizeof(*res)); | |
9448 | if (size == SIZE_MAX) | |
9449 | return -EOVERFLOW; | |
9450 | ||
9451 | res = memdup_user(arg, size); | |
9452 | if (IS_ERR(res)) | |
9453 | return PTR_ERR(res); | |
9454 | ||
9455 | ret = 0; | |
9456 | ||
9457 | for (i = 0; i < nr_args; i++) { | |
9458 | switch (res[i].opcode) { | |
9459 | case IORING_RESTRICTION_REGISTER_OP: | |
9460 | if (res[i].register_op >= IORING_REGISTER_LAST) { | |
9461 | ret = -EINVAL; | |
9462 | goto out; | |
9463 | } | |
9464 | ||
9465 | __set_bit(res[i].register_op, | |
9466 | ctx->restrictions.register_op); | |
9467 | break; | |
9468 | case IORING_RESTRICTION_SQE_OP: | |
9469 | if (res[i].sqe_op >= IORING_OP_LAST) { | |
9470 | ret = -EINVAL; | |
9471 | goto out; | |
9472 | } | |
9473 | ||
9474 | __set_bit(res[i].sqe_op, ctx->restrictions.sqe_op); | |
9475 | break; | |
9476 | case IORING_RESTRICTION_SQE_FLAGS_ALLOWED: | |
9477 | ctx->restrictions.sqe_flags_allowed = res[i].sqe_flags; | |
9478 | break; | |
9479 | case IORING_RESTRICTION_SQE_FLAGS_REQUIRED: | |
9480 | ctx->restrictions.sqe_flags_required = res[i].sqe_flags; | |
9481 | break; | |
9482 | default: | |
9483 | ret = -EINVAL; | |
9484 | goto out; | |
9485 | } | |
9486 | } | |
9487 | ||
9488 | out: | |
9489 | /* Reset all restrictions if an error happened */ | |
9490 | if (ret != 0) | |
9491 | memset(&ctx->restrictions, 0, sizeof(ctx->restrictions)); | |
9492 | else | |
7e84e1c7 | 9493 | ctx->restrictions.registered = true; |
21b55dbc SG |
9494 | |
9495 | kfree(res); | |
9496 | return ret; | |
9497 | } | |
9498 | ||
7e84e1c7 SG |
9499 | static int io_register_enable_rings(struct io_ring_ctx *ctx) |
9500 | { | |
9501 | if (!(ctx->flags & IORING_SETUP_R_DISABLED)) | |
9502 | return -EBADFD; | |
9503 | ||
9504 | if (ctx->restrictions.registered) | |
9505 | ctx->restricted = 1; | |
9506 | ||
9507 | ctx->flags &= ~IORING_SETUP_R_DISABLED; | |
9508 | ||
9509 | io_sq_offload_start(ctx); | |
9510 | ||
9511 | return 0; | |
9512 | } | |
9513 | ||
071698e1 JA |
9514 | static bool io_register_op_must_quiesce(int op) |
9515 | { | |
9516 | switch (op) { | |
9517 | case IORING_UNREGISTER_FILES: | |
9518 | case IORING_REGISTER_FILES_UPDATE: | |
9519 | case IORING_REGISTER_PROBE: | |
9520 | case IORING_REGISTER_PERSONALITY: | |
9521 | case IORING_UNREGISTER_PERSONALITY: | |
9522 | return false; | |
9523 | default: | |
9524 | return true; | |
9525 | } | |
9526 | } | |
9527 | ||
edafccee JA |
9528 | static int __io_uring_register(struct io_ring_ctx *ctx, unsigned opcode, |
9529 | void __user *arg, unsigned nr_args) | |
b19062a5 JA |
9530 | __releases(ctx->uring_lock) |
9531 | __acquires(ctx->uring_lock) | |
edafccee JA |
9532 | { |
9533 | int ret; | |
9534 | ||
35fa71a0 JA |
9535 | /* |
9536 | * We're inside the ring mutex, if the ref is already dying, then | |
9537 | * someone else killed the ctx or is already going through | |
9538 | * io_uring_register(). | |
9539 | */ | |
9540 | if (percpu_ref_is_dying(&ctx->refs)) | |
9541 | return -ENXIO; | |
9542 | ||
071698e1 | 9543 | if (io_register_op_must_quiesce(opcode)) { |
05f3fb3c | 9544 | percpu_ref_kill(&ctx->refs); |
b19062a5 | 9545 | |
05f3fb3c JA |
9546 | /* |
9547 | * Drop uring mutex before waiting for references to exit. If | |
9548 | * another thread is currently inside io_uring_enter() it might | |
9549 | * need to grab the uring_lock to make progress. If we hold it | |
9550 | * here across the drain wait, then we can deadlock. It's safe | |
9551 | * to drop the mutex here, since no new references will come in | |
9552 | * after we've killed the percpu ref. | |
9553 | */ | |
9554 | mutex_unlock(&ctx->uring_lock); | |
af9c1a44 JA |
9555 | do { |
9556 | ret = wait_for_completion_interruptible(&ctx->ref_comp); | |
9557 | if (!ret) | |
9558 | break; | |
ed6930c9 JA |
9559 | ret = io_run_task_work_sig(); |
9560 | if (ret < 0) | |
9561 | break; | |
af9c1a44 JA |
9562 | } while (1); |
9563 | ||
05f3fb3c | 9564 | mutex_lock(&ctx->uring_lock); |
af9c1a44 | 9565 | |
c150368b JA |
9566 | if (ret) { |
9567 | percpu_ref_resurrect(&ctx->refs); | |
21b55dbc SG |
9568 | goto out_quiesce; |
9569 | } | |
9570 | } | |
9571 | ||
9572 | if (ctx->restricted) { | |
9573 | if (opcode >= IORING_REGISTER_LAST) { | |
9574 | ret = -EINVAL; | |
9575 | goto out; | |
9576 | } | |
9577 | ||
9578 | if (!test_bit(opcode, ctx->restrictions.register_op)) { | |
9579 | ret = -EACCES; | |
c150368b JA |
9580 | goto out; |
9581 | } | |
05f3fb3c | 9582 | } |
edafccee JA |
9583 | |
9584 | switch (opcode) { | |
9585 | case IORING_REGISTER_BUFFERS: | |
9586 | ret = io_sqe_buffer_register(ctx, arg, nr_args); | |
9587 | break; | |
9588 | case IORING_UNREGISTER_BUFFERS: | |
9589 | ret = -EINVAL; | |
9590 | if (arg || nr_args) | |
9591 | break; | |
9592 | ret = io_sqe_buffer_unregister(ctx); | |
9593 | break; | |
6b06314c JA |
9594 | case IORING_REGISTER_FILES: |
9595 | ret = io_sqe_files_register(ctx, arg, nr_args); | |
9596 | break; | |
9597 | case IORING_UNREGISTER_FILES: | |
9598 | ret = -EINVAL; | |
9599 | if (arg || nr_args) | |
9600 | break; | |
9601 | ret = io_sqe_files_unregister(ctx); | |
9602 | break; | |
c3a31e60 JA |
9603 | case IORING_REGISTER_FILES_UPDATE: |
9604 | ret = io_sqe_files_update(ctx, arg, nr_args); | |
9605 | break; | |
9b402849 | 9606 | case IORING_REGISTER_EVENTFD: |
f2842ab5 | 9607 | case IORING_REGISTER_EVENTFD_ASYNC: |
9b402849 JA |
9608 | ret = -EINVAL; |
9609 | if (nr_args != 1) | |
9610 | break; | |
9611 | ret = io_eventfd_register(ctx, arg); | |
f2842ab5 JA |
9612 | if (ret) |
9613 | break; | |
9614 | if (opcode == IORING_REGISTER_EVENTFD_ASYNC) | |
9615 | ctx->eventfd_async = 1; | |
9616 | else | |
9617 | ctx->eventfd_async = 0; | |
9b402849 JA |
9618 | break; |
9619 | case IORING_UNREGISTER_EVENTFD: | |
9620 | ret = -EINVAL; | |
9621 | if (arg || nr_args) | |
9622 | break; | |
9623 | ret = io_eventfd_unregister(ctx); | |
9624 | break; | |
66f4af93 JA |
9625 | case IORING_REGISTER_PROBE: |
9626 | ret = -EINVAL; | |
9627 | if (!arg || nr_args > 256) | |
9628 | break; | |
9629 | ret = io_probe(ctx, arg, nr_args); | |
9630 | break; | |
071698e1 JA |
9631 | case IORING_REGISTER_PERSONALITY: |
9632 | ret = -EINVAL; | |
9633 | if (arg || nr_args) | |
9634 | break; | |
9635 | ret = io_register_personality(ctx); | |
9636 | break; | |
9637 | case IORING_UNREGISTER_PERSONALITY: | |
9638 | ret = -EINVAL; | |
9639 | if (arg) | |
9640 | break; | |
9641 | ret = io_unregister_personality(ctx, nr_args); | |
9642 | break; | |
7e84e1c7 SG |
9643 | case IORING_REGISTER_ENABLE_RINGS: |
9644 | ret = -EINVAL; | |
9645 | if (arg || nr_args) | |
9646 | break; | |
9647 | ret = io_register_enable_rings(ctx); | |
9648 | break; | |
21b55dbc SG |
9649 | case IORING_REGISTER_RESTRICTIONS: |
9650 | ret = io_register_restrictions(ctx, arg, nr_args); | |
9651 | break; | |
edafccee JA |
9652 | default: |
9653 | ret = -EINVAL; | |
9654 | break; | |
9655 | } | |
9656 | ||
21b55dbc | 9657 | out: |
071698e1 | 9658 | if (io_register_op_must_quiesce(opcode)) { |
05f3fb3c | 9659 | /* bring the ctx back to life */ |
05f3fb3c | 9660 | percpu_ref_reinit(&ctx->refs); |
21b55dbc | 9661 | out_quiesce: |
0f158b4c | 9662 | reinit_completion(&ctx->ref_comp); |
05f3fb3c | 9663 | } |
edafccee JA |
9664 | return ret; |
9665 | } | |
9666 | ||
9667 | SYSCALL_DEFINE4(io_uring_register, unsigned int, fd, unsigned int, opcode, | |
9668 | void __user *, arg, unsigned int, nr_args) | |
9669 | { | |
9670 | struct io_ring_ctx *ctx; | |
9671 | long ret = -EBADF; | |
9672 | struct fd f; | |
9673 | ||
9674 | f = fdget(fd); | |
9675 | if (!f.file) | |
9676 | return -EBADF; | |
9677 | ||
9678 | ret = -EOPNOTSUPP; | |
9679 | if (f.file->f_op != &io_uring_fops) | |
9680 | goto out_fput; | |
9681 | ||
9682 | ctx = f.file->private_data; | |
9683 | ||
9684 | mutex_lock(&ctx->uring_lock); | |
9685 | ret = __io_uring_register(ctx, opcode, arg, nr_args); | |
9686 | mutex_unlock(&ctx->uring_lock); | |
c826bd7a DD |
9687 | trace_io_uring_register(ctx, opcode, ctx->nr_user_files, ctx->nr_user_bufs, |
9688 | ctx->cq_ev_fd != NULL, ret); | |
edafccee JA |
9689 | out_fput: |
9690 | fdput(f); | |
9691 | return ret; | |
9692 | } | |
9693 | ||
2b188cc1 JA |
9694 | static int __init io_uring_init(void) |
9695 | { | |
d7f62e82 SM |
9696 | #define __BUILD_BUG_VERIFY_ELEMENT(stype, eoffset, etype, ename) do { \ |
9697 | BUILD_BUG_ON(offsetof(stype, ename) != eoffset); \ | |
9698 | BUILD_BUG_ON(sizeof(etype) != sizeof_field(stype, ename)); \ | |
9699 | } while (0) | |
9700 | ||
9701 | #define BUILD_BUG_SQE_ELEM(eoffset, etype, ename) \ | |
9702 | __BUILD_BUG_VERIFY_ELEMENT(struct io_uring_sqe, eoffset, etype, ename) | |
9703 | BUILD_BUG_ON(sizeof(struct io_uring_sqe) != 64); | |
9704 | BUILD_BUG_SQE_ELEM(0, __u8, opcode); | |
9705 | BUILD_BUG_SQE_ELEM(1, __u8, flags); | |
9706 | BUILD_BUG_SQE_ELEM(2, __u16, ioprio); | |
9707 | BUILD_BUG_SQE_ELEM(4, __s32, fd); | |
9708 | BUILD_BUG_SQE_ELEM(8, __u64, off); | |
9709 | BUILD_BUG_SQE_ELEM(8, __u64, addr2); | |
9710 | BUILD_BUG_SQE_ELEM(16, __u64, addr); | |
7d67af2c | 9711 | BUILD_BUG_SQE_ELEM(16, __u64, splice_off_in); |
d7f62e82 SM |
9712 | BUILD_BUG_SQE_ELEM(24, __u32, len); |
9713 | BUILD_BUG_SQE_ELEM(28, __kernel_rwf_t, rw_flags); | |
9714 | BUILD_BUG_SQE_ELEM(28, /* compat */ int, rw_flags); | |
9715 | BUILD_BUG_SQE_ELEM(28, /* compat */ __u32, rw_flags); | |
9716 | BUILD_BUG_SQE_ELEM(28, __u32, fsync_flags); | |
5769a351 JX |
9717 | BUILD_BUG_SQE_ELEM(28, /* compat */ __u16, poll_events); |
9718 | BUILD_BUG_SQE_ELEM(28, __u32, poll32_events); | |
d7f62e82 SM |
9719 | BUILD_BUG_SQE_ELEM(28, __u32, sync_range_flags); |
9720 | BUILD_BUG_SQE_ELEM(28, __u32, msg_flags); | |
9721 | BUILD_BUG_SQE_ELEM(28, __u32, timeout_flags); | |
9722 | BUILD_BUG_SQE_ELEM(28, __u32, accept_flags); | |
9723 | BUILD_BUG_SQE_ELEM(28, __u32, cancel_flags); | |
9724 | BUILD_BUG_SQE_ELEM(28, __u32, open_flags); | |
9725 | BUILD_BUG_SQE_ELEM(28, __u32, statx_flags); | |
9726 | BUILD_BUG_SQE_ELEM(28, __u32, fadvise_advice); | |
7d67af2c | 9727 | BUILD_BUG_SQE_ELEM(28, __u32, splice_flags); |
d7f62e82 SM |
9728 | BUILD_BUG_SQE_ELEM(32, __u64, user_data); |
9729 | BUILD_BUG_SQE_ELEM(40, __u16, buf_index); | |
9730 | BUILD_BUG_SQE_ELEM(42, __u16, personality); | |
7d67af2c | 9731 | BUILD_BUG_SQE_ELEM(44, __s32, splice_fd_in); |
d7f62e82 | 9732 | |
d3656344 | 9733 | BUILD_BUG_ON(ARRAY_SIZE(io_op_defs) != IORING_OP_LAST); |
84557871 | 9734 | BUILD_BUG_ON(__REQ_F_LAST_BIT >= 8 * sizeof(int)); |
2b188cc1 JA |
9735 | req_cachep = KMEM_CACHE(io_kiocb, SLAB_HWCACHE_ALIGN | SLAB_PANIC); |
9736 | return 0; | |
9737 | }; | |
9738 | __initcall(io_uring_init); |