]>
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> |
2b188cc1 | 83 | |
c826bd7a DD |
84 | #define CREATE_TRACE_POINTS |
85 | #include <trace/events/io_uring.h> | |
86 | ||
2b188cc1 JA |
87 | #include <uapi/linux/io_uring.h> |
88 | ||
89 | #include "internal.h" | |
561fb04a | 90 | #include "io-wq.h" |
2b188cc1 | 91 | |
5277deaa | 92 | #define IORING_MAX_ENTRIES 32768 |
33a107f0 | 93 | #define IORING_MAX_CQ_ENTRIES (2 * IORING_MAX_ENTRIES) |
65e19f54 JA |
94 | |
95 | /* | |
96 | * Shift of 9 is 512 entries, or exactly one page on 64-bit archs | |
97 | */ | |
98 | #define IORING_FILE_TABLE_SHIFT 9 | |
99 | #define IORING_MAX_FILES_TABLE (1U << IORING_FILE_TABLE_SHIFT) | |
100 | #define IORING_FILE_TABLE_MASK (IORING_MAX_FILES_TABLE - 1) | |
101 | #define IORING_MAX_FIXED_FILES (64 * IORING_MAX_FILES_TABLE) | |
21b55dbc SG |
102 | #define IORING_MAX_RESTRICTIONS (IORING_RESTRICTION_LAST + \ |
103 | IORING_REGISTER_LAST + IORING_OP_LAST) | |
2b188cc1 JA |
104 | |
105 | struct io_uring { | |
106 | u32 head ____cacheline_aligned_in_smp; | |
107 | u32 tail ____cacheline_aligned_in_smp; | |
108 | }; | |
109 | ||
1e84b97b | 110 | /* |
75b28aff HV |
111 | * This data is shared with the application through the mmap at offsets |
112 | * IORING_OFF_SQ_RING and IORING_OFF_CQ_RING. | |
1e84b97b SB |
113 | * |
114 | * The offsets to the member fields are published through struct | |
115 | * io_sqring_offsets when calling io_uring_setup. | |
116 | */ | |
75b28aff | 117 | struct io_rings { |
1e84b97b SB |
118 | /* |
119 | * Head and tail offsets into the ring; the offsets need to be | |
120 | * masked to get valid indices. | |
121 | * | |
75b28aff HV |
122 | * The kernel controls head of the sq ring and the tail of the cq ring, |
123 | * and the application controls tail of the sq ring and the head of the | |
124 | * cq ring. | |
1e84b97b | 125 | */ |
75b28aff | 126 | struct io_uring sq, cq; |
1e84b97b | 127 | /* |
75b28aff | 128 | * Bitmasks to apply to head and tail offsets (constant, equals |
1e84b97b SB |
129 | * ring_entries - 1) |
130 | */ | |
75b28aff HV |
131 | u32 sq_ring_mask, cq_ring_mask; |
132 | /* Ring sizes (constant, power of 2) */ | |
133 | u32 sq_ring_entries, cq_ring_entries; | |
1e84b97b SB |
134 | /* |
135 | * Number of invalid entries dropped by the kernel due to | |
136 | * invalid index stored in array | |
137 | * | |
138 | * Written by the kernel, shouldn't be modified by the | |
139 | * application (i.e. get number of "new events" by comparing to | |
140 | * cached value). | |
141 | * | |
142 | * After a new SQ head value was read by the application this | |
143 | * counter includes all submissions that were dropped reaching | |
144 | * the new SQ head (and possibly more). | |
145 | */ | |
75b28aff | 146 | u32 sq_dropped; |
1e84b97b | 147 | /* |
0d9b5b3a | 148 | * Runtime SQ flags |
1e84b97b SB |
149 | * |
150 | * Written by the kernel, shouldn't be modified by the | |
151 | * application. | |
152 | * | |
153 | * The application needs a full memory barrier before checking | |
154 | * for IORING_SQ_NEED_WAKEUP after updating the sq tail. | |
155 | */ | |
75b28aff | 156 | u32 sq_flags; |
0d9b5b3a SG |
157 | /* |
158 | * Runtime CQ flags | |
159 | * | |
160 | * Written by the application, shouldn't be modified by the | |
161 | * kernel. | |
162 | */ | |
163 | u32 cq_flags; | |
1e84b97b SB |
164 | /* |
165 | * Number of completion events lost because the queue was full; | |
166 | * this should be avoided by the application by making sure | |
0b4295b5 | 167 | * there are not more requests pending than there is space in |
1e84b97b SB |
168 | * the completion queue. |
169 | * | |
170 | * Written by the kernel, shouldn't be modified by the | |
171 | * application (i.e. get number of "new events" by comparing to | |
172 | * cached value). | |
173 | * | |
174 | * As completion events come in out of order this counter is not | |
175 | * ordered with any other data. | |
176 | */ | |
75b28aff | 177 | u32 cq_overflow; |
1e84b97b SB |
178 | /* |
179 | * Ring buffer of completion events. | |
180 | * | |
181 | * The kernel writes completion events fresh every time they are | |
182 | * produced, so the application is allowed to modify pending | |
183 | * entries. | |
184 | */ | |
75b28aff | 185 | struct io_uring_cqe cqes[] ____cacheline_aligned_in_smp; |
2b188cc1 JA |
186 | }; |
187 | ||
edafccee JA |
188 | struct io_mapped_ubuf { |
189 | u64 ubuf; | |
190 | size_t len; | |
191 | struct bio_vec *bvec; | |
192 | unsigned int nr_bvecs; | |
193 | }; | |
194 | ||
65e19f54 JA |
195 | struct fixed_file_table { |
196 | struct file **files; | |
31b51510 JA |
197 | }; |
198 | ||
05589553 XW |
199 | struct fixed_file_ref_node { |
200 | struct percpu_ref refs; | |
201 | struct list_head node; | |
202 | struct list_head file_list; | |
203 | struct fixed_file_data *file_data; | |
4a38aed2 | 204 | struct llist_node llist; |
05589553 XW |
205 | }; |
206 | ||
05f3fb3c JA |
207 | struct fixed_file_data { |
208 | struct fixed_file_table *table; | |
209 | struct io_ring_ctx *ctx; | |
210 | ||
05589553 | 211 | struct percpu_ref *cur_refs; |
05f3fb3c | 212 | struct percpu_ref refs; |
05f3fb3c | 213 | struct completion done; |
05589553 XW |
214 | struct list_head ref_list; |
215 | spinlock_t lock; | |
05f3fb3c JA |
216 | }; |
217 | ||
5a2e745d JA |
218 | struct io_buffer { |
219 | struct list_head list; | |
220 | __u64 addr; | |
221 | __s32 len; | |
222 | __u16 bid; | |
223 | }; | |
224 | ||
21b55dbc SG |
225 | struct io_restriction { |
226 | DECLARE_BITMAP(register_op, IORING_REGISTER_LAST); | |
227 | DECLARE_BITMAP(sqe_op, IORING_OP_LAST); | |
228 | u8 sqe_flags_allowed; | |
229 | u8 sqe_flags_required; | |
7e84e1c7 | 230 | bool registered; |
21b55dbc SG |
231 | }; |
232 | ||
2b188cc1 JA |
233 | struct io_ring_ctx { |
234 | struct { | |
235 | struct percpu_ref refs; | |
236 | } ____cacheline_aligned_in_smp; | |
237 | ||
238 | struct { | |
239 | unsigned int flags; | |
e1d85334 | 240 | unsigned int compat: 1; |
aad5d8da | 241 | unsigned int limit_mem: 1; |
e1d85334 RD |
242 | unsigned int cq_overflow_flushed: 1; |
243 | unsigned int drain_next: 1; | |
244 | unsigned int eventfd_async: 1; | |
21b55dbc | 245 | unsigned int restricted: 1; |
2b188cc1 | 246 | |
75b28aff HV |
247 | /* |
248 | * Ring buffer of indices into array of io_uring_sqe, which is | |
249 | * mmapped by the application using the IORING_OFF_SQES offset. | |
250 | * | |
251 | * This indirection could e.g. be used to assign fixed | |
252 | * io_uring_sqe entries to operations and only submit them to | |
253 | * the queue when needed. | |
254 | * | |
255 | * The kernel modifies neither the indices array nor the entries | |
256 | * array. | |
257 | */ | |
258 | u32 *sq_array; | |
2b188cc1 JA |
259 | unsigned cached_sq_head; |
260 | unsigned sq_entries; | |
261 | unsigned sq_mask; | |
6c271ce2 | 262 | unsigned sq_thread_idle; |
498ccd9e | 263 | unsigned cached_sq_dropped; |
206aefde | 264 | atomic_t cached_cq_overflow; |
ad3eb2c8 | 265 | unsigned long sq_check_overflow; |
de0617e4 JA |
266 | |
267 | struct list_head defer_list; | |
5262f567 | 268 | struct list_head timeout_list; |
1d7bb1d5 | 269 | struct list_head cq_overflow_list; |
fcb323cc JA |
270 | |
271 | wait_queue_head_t inflight_wait; | |
ad3eb2c8 | 272 | struct io_uring_sqe *sq_sqes; |
2b188cc1 JA |
273 | } ____cacheline_aligned_in_smp; |
274 | ||
206aefde JA |
275 | struct io_rings *rings; |
276 | ||
2b188cc1 | 277 | /* IO offload */ |
561fb04a | 278 | struct io_wq *io_wq; |
6c271ce2 | 279 | struct task_struct *sqo_thread; /* if using sq thread polling */ |
2aede0e4 JA |
280 | |
281 | /* | |
282 | * For SQPOLL usage - we hold a reference to the parent task, so we | |
283 | * have access to the ->files | |
284 | */ | |
285 | struct task_struct *sqo_task; | |
286 | ||
287 | /* Only used for accounting purposes */ | |
288 | struct mm_struct *mm_account; | |
289 | ||
6a779382 JA |
290 | struct wait_queue_head *sqo_wait; |
291 | struct wait_queue_head __sqo_wait; | |
292 | struct wait_queue_entry sqo_wait_entry; | |
75b28aff | 293 | |
6b06314c JA |
294 | /* |
295 | * If used, fixed file set. Writers must ensure that ->refs is dead, | |
296 | * readers must ensure that ->refs is alive as long as the file* is | |
297 | * used. Only updated through io_uring_register(2). | |
298 | */ | |
05f3fb3c | 299 | struct fixed_file_data *file_data; |
6b06314c JA |
300 | unsigned nr_user_files; |
301 | ||
edafccee JA |
302 | /* if used, fixed mapped user buffers */ |
303 | unsigned nr_user_bufs; | |
304 | struct io_mapped_ubuf *user_bufs; | |
305 | ||
2b188cc1 JA |
306 | struct user_struct *user; |
307 | ||
0b8c0ec7 | 308 | const struct cred *creds; |
181e448d | 309 | |
0f158b4c JA |
310 | struct completion ref_comp; |
311 | struct completion sq_thread_comp; | |
206aefde | 312 | |
0ddf92e8 JA |
313 | /* if all else fails... */ |
314 | struct io_kiocb *fallback_req; | |
315 | ||
206aefde JA |
316 | #if defined(CONFIG_UNIX) |
317 | struct socket *ring_sock; | |
318 | #endif | |
319 | ||
5a2e745d JA |
320 | struct idr io_buffer_idr; |
321 | ||
071698e1 JA |
322 | struct idr personality_idr; |
323 | ||
206aefde JA |
324 | struct { |
325 | unsigned cached_cq_tail; | |
326 | unsigned cq_entries; | |
327 | unsigned cq_mask; | |
328 | atomic_t cq_timeouts; | |
ad3eb2c8 | 329 | unsigned long cq_check_overflow; |
206aefde JA |
330 | struct wait_queue_head cq_wait; |
331 | struct fasync_struct *cq_fasync; | |
332 | struct eventfd_ctx *cq_ev_fd; | |
333 | } ____cacheline_aligned_in_smp; | |
2b188cc1 JA |
334 | |
335 | struct { | |
336 | struct mutex uring_lock; | |
337 | wait_queue_head_t wait; | |
338 | } ____cacheline_aligned_in_smp; | |
339 | ||
340 | struct { | |
341 | spinlock_t completion_lock; | |
e94f141b | 342 | |
def596e9 | 343 | /* |
540e32a0 | 344 | * ->iopoll_list is protected by the ctx->uring_lock for |
def596e9 JA |
345 | * io_uring instances that don't use IORING_SETUP_SQPOLL. |
346 | * For SQPOLL, only the single threaded io_sq_thread() will | |
347 | * manipulate the list, hence no extra locking is needed there. | |
348 | */ | |
540e32a0 | 349 | struct list_head iopoll_list; |
78076bb6 JA |
350 | struct hlist_head *cancel_hash; |
351 | unsigned cancel_hash_bits; | |
e94f141b | 352 | bool poll_multi_file; |
31b51510 | 353 | |
fcb323cc JA |
354 | spinlock_t inflight_lock; |
355 | struct list_head inflight_list; | |
2b188cc1 | 356 | } ____cacheline_aligned_in_smp; |
85faa7b8 | 357 | |
4a38aed2 JA |
358 | struct delayed_work file_put_work; |
359 | struct llist_head file_put_llist; | |
360 | ||
85faa7b8 | 361 | struct work_struct exit_work; |
21b55dbc | 362 | struct io_restriction restrictions; |
2b188cc1 JA |
363 | }; |
364 | ||
09bb8394 JA |
365 | /* |
366 | * First field must be the file pointer in all the | |
367 | * iocb unions! See also 'struct kiocb' in <linux/fs.h> | |
368 | */ | |
221c5eb2 JA |
369 | struct io_poll_iocb { |
370 | struct file *file; | |
0969e783 JA |
371 | union { |
372 | struct wait_queue_head *head; | |
373 | u64 addr; | |
374 | }; | |
221c5eb2 | 375 | __poll_t events; |
8c838788 | 376 | bool done; |
221c5eb2 | 377 | bool canceled; |
392edb45 | 378 | struct wait_queue_entry wait; |
221c5eb2 JA |
379 | }; |
380 | ||
b5dba59e JA |
381 | struct io_close { |
382 | struct file *file; | |
383 | struct file *put_file; | |
384 | int fd; | |
385 | }; | |
386 | ||
ad8a48ac JA |
387 | struct io_timeout_data { |
388 | struct io_kiocb *req; | |
389 | struct hrtimer timer; | |
390 | struct timespec64 ts; | |
391 | enum hrtimer_mode mode; | |
392 | }; | |
393 | ||
8ed8d3c3 JA |
394 | struct io_accept { |
395 | struct file *file; | |
396 | struct sockaddr __user *addr; | |
397 | int __user *addr_len; | |
398 | int flags; | |
09952e3e | 399 | unsigned long nofile; |
8ed8d3c3 JA |
400 | }; |
401 | ||
402 | struct io_sync { | |
403 | struct file *file; | |
404 | loff_t len; | |
405 | loff_t off; | |
406 | int flags; | |
d63d1b5e | 407 | int mode; |
8ed8d3c3 JA |
408 | }; |
409 | ||
fbf23849 JA |
410 | struct io_cancel { |
411 | struct file *file; | |
412 | u64 addr; | |
413 | }; | |
414 | ||
b29472ee JA |
415 | struct io_timeout { |
416 | struct file *file; | |
417 | u64 addr; | |
418 | int flags; | |
bfe68a22 PB |
419 | u32 off; |
420 | u32 target_seq; | |
135fcde8 | 421 | struct list_head list; |
b29472ee JA |
422 | }; |
423 | ||
9adbd45d JA |
424 | struct io_rw { |
425 | /* NOTE: kiocb has the file as the first member, so don't do it here */ | |
426 | struct kiocb kiocb; | |
427 | u64 addr; | |
428 | u64 len; | |
429 | }; | |
430 | ||
3fbb51c1 JA |
431 | struct io_connect { |
432 | struct file *file; | |
433 | struct sockaddr __user *addr; | |
434 | int addr_len; | |
435 | }; | |
436 | ||
e47293fd JA |
437 | struct io_sr_msg { |
438 | struct file *file; | |
fddaface | 439 | union { |
270a5940 | 440 | struct user_msghdr __user *umsg; |
fddaface JA |
441 | void __user *buf; |
442 | }; | |
e47293fd | 443 | int msg_flags; |
bcda7baa | 444 | int bgid; |
fddaface | 445 | size_t len; |
bcda7baa | 446 | struct io_buffer *kbuf; |
e47293fd JA |
447 | }; |
448 | ||
15b71abe JA |
449 | struct io_open { |
450 | struct file *file; | |
451 | int dfd; | |
15b71abe | 452 | struct filename *filename; |
c12cedf2 | 453 | struct open_how how; |
4022e7af | 454 | unsigned long nofile; |
15b71abe JA |
455 | }; |
456 | ||
05f3fb3c JA |
457 | struct io_files_update { |
458 | struct file *file; | |
459 | u64 arg; | |
460 | u32 nr_args; | |
461 | u32 offset; | |
462 | }; | |
463 | ||
4840e418 JA |
464 | struct io_fadvise { |
465 | struct file *file; | |
466 | u64 offset; | |
467 | u32 len; | |
468 | u32 advice; | |
469 | }; | |
470 | ||
c1ca757b JA |
471 | struct io_madvise { |
472 | struct file *file; | |
473 | u64 addr; | |
474 | u32 len; | |
475 | u32 advice; | |
476 | }; | |
477 | ||
3e4827b0 JA |
478 | struct io_epoll { |
479 | struct file *file; | |
480 | int epfd; | |
481 | int op; | |
482 | int fd; | |
483 | struct epoll_event event; | |
e47293fd JA |
484 | }; |
485 | ||
7d67af2c PB |
486 | struct io_splice { |
487 | struct file *file_out; | |
488 | struct file *file_in; | |
489 | loff_t off_out; | |
490 | loff_t off_in; | |
491 | u64 len; | |
492 | unsigned int flags; | |
493 | }; | |
494 | ||
ddf0322d JA |
495 | struct io_provide_buf { |
496 | struct file *file; | |
497 | __u64 addr; | |
498 | __s32 len; | |
499 | __u32 bgid; | |
500 | __u16 nbufs; | |
501 | __u16 bid; | |
502 | }; | |
503 | ||
1d9e1288 BM |
504 | struct io_statx { |
505 | struct file *file; | |
506 | int dfd; | |
507 | unsigned int mask; | |
508 | unsigned int flags; | |
e62753e4 | 509 | const char __user *filename; |
1d9e1288 BM |
510 | struct statx __user *buffer; |
511 | }; | |
512 | ||
3ca405eb PB |
513 | struct io_completion { |
514 | struct file *file; | |
515 | struct list_head list; | |
0f7e466b | 516 | int cflags; |
3ca405eb PB |
517 | }; |
518 | ||
f499a021 JA |
519 | struct io_async_connect { |
520 | struct sockaddr_storage address; | |
521 | }; | |
522 | ||
03b1230c JA |
523 | struct io_async_msghdr { |
524 | struct iovec fast_iov[UIO_FASTIOV]; | |
525 | struct iovec *iov; | |
526 | struct sockaddr __user *uaddr; | |
527 | struct msghdr msg; | |
b537916c | 528 | struct sockaddr_storage addr; |
03b1230c JA |
529 | }; |
530 | ||
f67676d1 JA |
531 | struct io_async_rw { |
532 | struct iovec fast_iov[UIO_FASTIOV]; | |
ff6165b2 JA |
533 | const struct iovec *free_iovec; |
534 | struct iov_iter iter; | |
227c0c96 | 535 | size_t bytes_done; |
bcf5a063 | 536 | struct wait_page_queue wpq; |
f67676d1 JA |
537 | }; |
538 | ||
1a6b74fc | 539 | struct io_async_ctx { |
f67676d1 JA |
540 | union { |
541 | struct io_async_rw rw; | |
03b1230c | 542 | struct io_async_msghdr msg; |
f499a021 | 543 | struct io_async_connect connect; |
2d28390a | 544 | struct io_timeout_data timeout; |
f67676d1 | 545 | }; |
1a6b74fc JA |
546 | }; |
547 | ||
6b47ee6e PB |
548 | enum { |
549 | REQ_F_FIXED_FILE_BIT = IOSQE_FIXED_FILE_BIT, | |
550 | REQ_F_IO_DRAIN_BIT = IOSQE_IO_DRAIN_BIT, | |
551 | REQ_F_LINK_BIT = IOSQE_IO_LINK_BIT, | |
552 | REQ_F_HARDLINK_BIT = IOSQE_IO_HARDLINK_BIT, | |
553 | REQ_F_FORCE_ASYNC_BIT = IOSQE_ASYNC_BIT, | |
bcda7baa | 554 | REQ_F_BUFFER_SELECT_BIT = IOSQE_BUFFER_SELECT_BIT, |
6b47ee6e | 555 | |
dea3b49c | 556 | REQ_F_LINK_HEAD_BIT, |
6b47ee6e PB |
557 | REQ_F_FAIL_LINK_BIT, |
558 | REQ_F_INFLIGHT_BIT, | |
559 | REQ_F_CUR_POS_BIT, | |
560 | REQ_F_NOWAIT_BIT, | |
6b47ee6e | 561 | REQ_F_LINK_TIMEOUT_BIT, |
6b47ee6e | 562 | REQ_F_ISREG_BIT, |
6b47ee6e | 563 | REQ_F_COMP_LOCKED_BIT, |
99bc4c38 | 564 | REQ_F_NEED_CLEANUP_BIT, |
d7718a9d | 565 | REQ_F_POLLED_BIT, |
bcda7baa | 566 | REQ_F_BUFFER_SELECTED_BIT, |
5b0bbee4 | 567 | REQ_F_NO_FILE_TABLE_BIT, |
7cdaf587 | 568 | REQ_F_WORK_INITIALIZED_BIT, |
84557871 JA |
569 | |
570 | /* not a real bit, just to check we're not overflowing the space */ | |
571 | __REQ_F_LAST_BIT, | |
6b47ee6e PB |
572 | }; |
573 | ||
574 | enum { | |
575 | /* ctx owns file */ | |
576 | REQ_F_FIXED_FILE = BIT(REQ_F_FIXED_FILE_BIT), | |
577 | /* drain existing IO first */ | |
578 | REQ_F_IO_DRAIN = BIT(REQ_F_IO_DRAIN_BIT), | |
579 | /* linked sqes */ | |
580 | REQ_F_LINK = BIT(REQ_F_LINK_BIT), | |
581 | /* doesn't sever on completion < 0 */ | |
582 | REQ_F_HARDLINK = BIT(REQ_F_HARDLINK_BIT), | |
583 | /* IOSQE_ASYNC */ | |
584 | REQ_F_FORCE_ASYNC = BIT(REQ_F_FORCE_ASYNC_BIT), | |
bcda7baa JA |
585 | /* IOSQE_BUFFER_SELECT */ |
586 | REQ_F_BUFFER_SELECT = BIT(REQ_F_BUFFER_SELECT_BIT), | |
6b47ee6e | 587 | |
dea3b49c PB |
588 | /* head of a link */ |
589 | REQ_F_LINK_HEAD = BIT(REQ_F_LINK_HEAD_BIT), | |
6b47ee6e PB |
590 | /* fail rest of links */ |
591 | REQ_F_FAIL_LINK = BIT(REQ_F_FAIL_LINK_BIT), | |
592 | /* on inflight list */ | |
593 | REQ_F_INFLIGHT = BIT(REQ_F_INFLIGHT_BIT), | |
594 | /* read/write uses file position */ | |
595 | REQ_F_CUR_POS = BIT(REQ_F_CUR_POS_BIT), | |
596 | /* must not punt to workers */ | |
597 | REQ_F_NOWAIT = BIT(REQ_F_NOWAIT_BIT), | |
6b47ee6e PB |
598 | /* has linked timeout */ |
599 | REQ_F_LINK_TIMEOUT = BIT(REQ_F_LINK_TIMEOUT_BIT), | |
6b47ee6e PB |
600 | /* regular file */ |
601 | REQ_F_ISREG = BIT(REQ_F_ISREG_BIT), | |
6b47ee6e PB |
602 | /* completion under lock */ |
603 | REQ_F_COMP_LOCKED = BIT(REQ_F_COMP_LOCKED_BIT), | |
99bc4c38 PB |
604 | /* needs cleanup */ |
605 | REQ_F_NEED_CLEANUP = BIT(REQ_F_NEED_CLEANUP_BIT), | |
d7718a9d JA |
606 | /* already went through poll handler */ |
607 | REQ_F_POLLED = BIT(REQ_F_POLLED_BIT), | |
bcda7baa JA |
608 | /* buffer already selected */ |
609 | REQ_F_BUFFER_SELECTED = BIT(REQ_F_BUFFER_SELECTED_BIT), | |
5b0bbee4 JA |
610 | /* doesn't need file table for this request */ |
611 | REQ_F_NO_FILE_TABLE = BIT(REQ_F_NO_FILE_TABLE_BIT), | |
7cdaf587 XW |
612 | /* io_wq_work is initialized */ |
613 | REQ_F_WORK_INITIALIZED = BIT(REQ_F_WORK_INITIALIZED_BIT), | |
d7718a9d JA |
614 | }; |
615 | ||
616 | struct async_poll { | |
617 | struct io_poll_iocb poll; | |
807abcb0 | 618 | struct io_poll_iocb *double_poll; |
6b47ee6e PB |
619 | }; |
620 | ||
09bb8394 JA |
621 | /* |
622 | * NOTE! Each of the iocb union members has the file pointer | |
623 | * as the first entry in their struct definition. So you can | |
624 | * access the file pointer through any of the sub-structs, | |
625 | * or directly as just 'ki_filp' in this struct. | |
626 | */ | |
2b188cc1 | 627 | struct io_kiocb { |
221c5eb2 | 628 | union { |
09bb8394 | 629 | struct file *file; |
9adbd45d | 630 | struct io_rw rw; |
221c5eb2 | 631 | struct io_poll_iocb poll; |
8ed8d3c3 JA |
632 | struct io_accept accept; |
633 | struct io_sync sync; | |
fbf23849 | 634 | struct io_cancel cancel; |
b29472ee | 635 | struct io_timeout timeout; |
3fbb51c1 | 636 | struct io_connect connect; |
e47293fd | 637 | struct io_sr_msg sr_msg; |
15b71abe | 638 | struct io_open open; |
b5dba59e | 639 | struct io_close close; |
05f3fb3c | 640 | struct io_files_update files_update; |
4840e418 | 641 | struct io_fadvise fadvise; |
c1ca757b | 642 | struct io_madvise madvise; |
3e4827b0 | 643 | struct io_epoll epoll; |
7d67af2c | 644 | struct io_splice splice; |
ddf0322d | 645 | struct io_provide_buf pbuf; |
1d9e1288 | 646 | struct io_statx statx; |
3ca405eb PB |
647 | /* use only after cleaning per-op data, see io_clean_op() */ |
648 | struct io_completion compl; | |
221c5eb2 | 649 | }; |
2b188cc1 | 650 | |
1a6b74fc | 651 | struct io_async_ctx *io; |
d625c6ee | 652 | u8 opcode; |
65a6543d XW |
653 | /* polled IO has completed */ |
654 | u8 iopoll_completed; | |
2b188cc1 | 655 | |
4f4eeba8 | 656 | u16 buf_index; |
9cf7c104 | 657 | u32 result; |
4f4eeba8 | 658 | |
010e8e6b PB |
659 | struct io_ring_ctx *ctx; |
660 | unsigned int flags; | |
661 | refcount_t refs; | |
662 | struct task_struct *task; | |
663 | u64 user_data; | |
d7718a9d | 664 | |
010e8e6b | 665 | struct list_head link_list; |
fcb323cc | 666 | |
d21ffe7e PB |
667 | /* |
668 | * 1. used with ctx->iopoll_list with reads/writes | |
669 | * 2. to track reqs with ->files (see io_op_def::file_table) | |
670 | */ | |
010e8e6b PB |
671 | struct list_head inflight_entry; |
672 | ||
673 | struct percpu_ref *fixed_file_refs; | |
674 | struct callback_head task_work; | |
675 | /* for polled requests, i.e. IORING_OP_POLL_ADD and async armed poll */ | |
676 | struct hlist_node hash_node; | |
677 | struct async_poll *apoll; | |
678 | struct io_wq_work work; | |
2b188cc1 | 679 | }; |
05589553 | 680 | |
27dc8338 PB |
681 | struct io_defer_entry { |
682 | struct list_head list; | |
683 | struct io_kiocb *req; | |
9cf7c104 | 684 | u32 seq; |
2b188cc1 JA |
685 | }; |
686 | ||
def596e9 | 687 | #define IO_IOPOLL_BATCH 8 |
2b188cc1 | 688 | |
013538bd JA |
689 | struct io_comp_state { |
690 | unsigned int nr; | |
691 | struct list_head list; | |
692 | struct io_ring_ctx *ctx; | |
693 | }; | |
694 | ||
9a56a232 JA |
695 | struct io_submit_state { |
696 | struct blk_plug plug; | |
697 | ||
2579f913 JA |
698 | /* |
699 | * io_kiocb alloc cache | |
700 | */ | |
701 | void *reqs[IO_IOPOLL_BATCH]; | |
6c8a3134 | 702 | unsigned int free_reqs; |
2579f913 | 703 | |
013538bd JA |
704 | /* |
705 | * Batch completion logic | |
706 | */ | |
707 | struct io_comp_state comp; | |
708 | ||
9a56a232 JA |
709 | /* |
710 | * File reference cache | |
711 | */ | |
712 | struct file *file; | |
713 | unsigned int fd; | |
714 | unsigned int has_refs; | |
9a56a232 JA |
715 | unsigned int ios_left; |
716 | }; | |
717 | ||
d3656344 JA |
718 | struct io_op_def { |
719 | /* needs req->io allocated for deferral/async */ | |
720 | unsigned async_ctx : 1; | |
721 | /* needs current->mm setup, does mm access */ | |
722 | unsigned needs_mm : 1; | |
723 | /* needs req->file assigned */ | |
724 | unsigned needs_file : 1; | |
fd2206e4 JA |
725 | /* don't fail if file grab fails */ |
726 | unsigned needs_file_no_error : 1; | |
d3656344 JA |
727 | /* hash wq insertion if file is a regular file */ |
728 | unsigned hash_reg_file : 1; | |
729 | /* unbound wq insertion if file is a non-regular file */ | |
730 | unsigned unbound_nonreg_file : 1; | |
66f4af93 JA |
731 | /* opcode is not supported by this kernel */ |
732 | unsigned not_supported : 1; | |
f86cd20c JA |
733 | /* needs file table */ |
734 | unsigned file_table : 1; | |
ff002b30 JA |
735 | /* needs ->fs */ |
736 | unsigned needs_fs : 1; | |
8a72758c JA |
737 | /* set if opcode supports polled "wait" */ |
738 | unsigned pollin : 1; | |
739 | unsigned pollout : 1; | |
bcda7baa JA |
740 | /* op supports buffer selection */ |
741 | unsigned buffer_select : 1; | |
57f1a649 | 742 | unsigned needs_fsize : 1; |
d3656344 JA |
743 | }; |
744 | ||
745 | static const struct io_op_def io_op_defs[] = { | |
0463b6c5 PB |
746 | [IORING_OP_NOP] = {}, |
747 | [IORING_OP_READV] = { | |
d3656344 JA |
748 | .async_ctx = 1, |
749 | .needs_mm = 1, | |
750 | .needs_file = 1, | |
751 | .unbound_nonreg_file = 1, | |
8a72758c | 752 | .pollin = 1, |
4d954c25 | 753 | .buffer_select = 1, |
d3656344 | 754 | }, |
0463b6c5 | 755 | [IORING_OP_WRITEV] = { |
d3656344 JA |
756 | .async_ctx = 1, |
757 | .needs_mm = 1, | |
758 | .needs_file = 1, | |
759 | .hash_reg_file = 1, | |
760 | .unbound_nonreg_file = 1, | |
8a72758c | 761 | .pollout = 1, |
57f1a649 | 762 | .needs_fsize = 1, |
d3656344 | 763 | }, |
0463b6c5 | 764 | [IORING_OP_FSYNC] = { |
d3656344 JA |
765 | .needs_file = 1, |
766 | }, | |
0463b6c5 | 767 | [IORING_OP_READ_FIXED] = { |
d3656344 JA |
768 | .needs_file = 1, |
769 | .unbound_nonreg_file = 1, | |
8a72758c | 770 | .pollin = 1, |
d3656344 | 771 | }, |
0463b6c5 | 772 | [IORING_OP_WRITE_FIXED] = { |
d3656344 JA |
773 | .needs_file = 1, |
774 | .hash_reg_file = 1, | |
775 | .unbound_nonreg_file = 1, | |
8a72758c | 776 | .pollout = 1, |
57f1a649 | 777 | .needs_fsize = 1, |
d3656344 | 778 | }, |
0463b6c5 | 779 | [IORING_OP_POLL_ADD] = { |
d3656344 JA |
780 | .needs_file = 1, |
781 | .unbound_nonreg_file = 1, | |
782 | }, | |
0463b6c5 PB |
783 | [IORING_OP_POLL_REMOVE] = {}, |
784 | [IORING_OP_SYNC_FILE_RANGE] = { | |
d3656344 JA |
785 | .needs_file = 1, |
786 | }, | |
0463b6c5 | 787 | [IORING_OP_SENDMSG] = { |
d3656344 JA |
788 | .async_ctx = 1, |
789 | .needs_mm = 1, | |
790 | .needs_file = 1, | |
791 | .unbound_nonreg_file = 1, | |
ff002b30 | 792 | .needs_fs = 1, |
8a72758c | 793 | .pollout = 1, |
d3656344 | 794 | }, |
0463b6c5 | 795 | [IORING_OP_RECVMSG] = { |
d3656344 JA |
796 | .async_ctx = 1, |
797 | .needs_mm = 1, | |
798 | .needs_file = 1, | |
799 | .unbound_nonreg_file = 1, | |
ff002b30 | 800 | .needs_fs = 1, |
8a72758c | 801 | .pollin = 1, |
52de1fe1 | 802 | .buffer_select = 1, |
d3656344 | 803 | }, |
0463b6c5 | 804 | [IORING_OP_TIMEOUT] = { |
d3656344 JA |
805 | .async_ctx = 1, |
806 | .needs_mm = 1, | |
807 | }, | |
0463b6c5 PB |
808 | [IORING_OP_TIMEOUT_REMOVE] = {}, |
809 | [IORING_OP_ACCEPT] = { | |
d3656344 JA |
810 | .needs_mm = 1, |
811 | .needs_file = 1, | |
812 | .unbound_nonreg_file = 1, | |
f86cd20c | 813 | .file_table = 1, |
8a72758c | 814 | .pollin = 1, |
d3656344 | 815 | }, |
0463b6c5 PB |
816 | [IORING_OP_ASYNC_CANCEL] = {}, |
817 | [IORING_OP_LINK_TIMEOUT] = { | |
d3656344 JA |
818 | .async_ctx = 1, |
819 | .needs_mm = 1, | |
820 | }, | |
0463b6c5 | 821 | [IORING_OP_CONNECT] = { |
d3656344 JA |
822 | .async_ctx = 1, |
823 | .needs_mm = 1, | |
824 | .needs_file = 1, | |
825 | .unbound_nonreg_file = 1, | |
8a72758c | 826 | .pollout = 1, |
d3656344 | 827 | }, |
0463b6c5 | 828 | [IORING_OP_FALLOCATE] = { |
d3656344 | 829 | .needs_file = 1, |
57f1a649 | 830 | .needs_fsize = 1, |
d3656344 | 831 | }, |
0463b6c5 | 832 | [IORING_OP_OPENAT] = { |
f86cd20c | 833 | .file_table = 1, |
ff002b30 | 834 | .needs_fs = 1, |
d3656344 | 835 | }, |
0463b6c5 | 836 | [IORING_OP_CLOSE] = { |
fd2206e4 JA |
837 | .needs_file = 1, |
838 | .needs_file_no_error = 1, | |
f86cd20c | 839 | .file_table = 1, |
d3656344 | 840 | }, |
0463b6c5 | 841 | [IORING_OP_FILES_UPDATE] = { |
d3656344 | 842 | .needs_mm = 1, |
f86cd20c | 843 | .file_table = 1, |
d3656344 | 844 | }, |
0463b6c5 | 845 | [IORING_OP_STATX] = { |
d3656344 | 846 | .needs_mm = 1, |
ff002b30 | 847 | .needs_fs = 1, |
5b0bbee4 | 848 | .file_table = 1, |
d3656344 | 849 | }, |
0463b6c5 | 850 | [IORING_OP_READ] = { |
3a6820f2 JA |
851 | .needs_mm = 1, |
852 | .needs_file = 1, | |
853 | .unbound_nonreg_file = 1, | |
8a72758c | 854 | .pollin = 1, |
bcda7baa | 855 | .buffer_select = 1, |
3a6820f2 | 856 | }, |
0463b6c5 | 857 | [IORING_OP_WRITE] = { |
3a6820f2 JA |
858 | .needs_mm = 1, |
859 | .needs_file = 1, | |
860 | .unbound_nonreg_file = 1, | |
8a72758c | 861 | .pollout = 1, |
57f1a649 | 862 | .needs_fsize = 1, |
3a6820f2 | 863 | }, |
0463b6c5 | 864 | [IORING_OP_FADVISE] = { |
4840e418 JA |
865 | .needs_file = 1, |
866 | }, | |
0463b6c5 | 867 | [IORING_OP_MADVISE] = { |
c1ca757b JA |
868 | .needs_mm = 1, |
869 | }, | |
0463b6c5 | 870 | [IORING_OP_SEND] = { |
fddaface JA |
871 | .needs_mm = 1, |
872 | .needs_file = 1, | |
873 | .unbound_nonreg_file = 1, | |
8a72758c | 874 | .pollout = 1, |
fddaface | 875 | }, |
0463b6c5 | 876 | [IORING_OP_RECV] = { |
fddaface JA |
877 | .needs_mm = 1, |
878 | .needs_file = 1, | |
879 | .unbound_nonreg_file = 1, | |
8a72758c | 880 | .pollin = 1, |
bcda7baa | 881 | .buffer_select = 1, |
fddaface | 882 | }, |
0463b6c5 | 883 | [IORING_OP_OPENAT2] = { |
f86cd20c | 884 | .file_table = 1, |
ff002b30 | 885 | .needs_fs = 1, |
cebdb986 | 886 | }, |
3e4827b0 JA |
887 | [IORING_OP_EPOLL_CTL] = { |
888 | .unbound_nonreg_file = 1, | |
889 | .file_table = 1, | |
890 | }, | |
7d67af2c PB |
891 | [IORING_OP_SPLICE] = { |
892 | .needs_file = 1, | |
893 | .hash_reg_file = 1, | |
894 | .unbound_nonreg_file = 1, | |
ddf0322d JA |
895 | }, |
896 | [IORING_OP_PROVIDE_BUFFERS] = {}, | |
067524e9 | 897 | [IORING_OP_REMOVE_BUFFERS] = {}, |
f2a8d5c7 PB |
898 | [IORING_OP_TEE] = { |
899 | .needs_file = 1, | |
900 | .hash_reg_file = 1, | |
901 | .unbound_nonreg_file = 1, | |
902 | }, | |
d3656344 JA |
903 | }; |
904 | ||
2e0464d4 BM |
905 | enum io_mem_account { |
906 | ACCT_LOCKED, | |
907 | ACCT_PINNED, | |
908 | }; | |
909 | ||
81b68a5c PB |
910 | static void __io_complete_rw(struct io_kiocb *req, long res, long res2, |
911 | struct io_comp_state *cs); | |
78e19bbe | 912 | static void io_cqring_fill_event(struct io_kiocb *req, long res); |
ec9c02ad | 913 | static void io_put_req(struct io_kiocb *req); |
c40f6379 | 914 | static void io_double_put_req(struct io_kiocb *req); |
978db57e | 915 | static void __io_double_put_req(struct io_kiocb *req); |
94ae5e77 | 916 | static struct io_kiocb *io_prep_linked_timeout(struct io_kiocb *req); |
7271ef3a | 917 | static void __io_queue_linked_timeout(struct io_kiocb *req); |
94ae5e77 | 918 | static void io_queue_linked_timeout(struct io_kiocb *req); |
05f3fb3c JA |
919 | static int __io_sqe_files_update(struct io_ring_ctx *ctx, |
920 | struct io_uring_files_update *ip, | |
921 | unsigned nr_args); | |
f56040b8 | 922 | static int io_prep_work_files(struct io_kiocb *req); |
3ca405eb | 923 | static void __io_clean_op(struct io_kiocb *req); |
b41e9852 JA |
924 | static int io_file_get(struct io_submit_state *state, struct io_kiocb *req, |
925 | int fd, struct file **out_file, bool fixed); | |
926 | static void __io_queue_sqe(struct io_kiocb *req, | |
f13fad7b JA |
927 | const struct io_uring_sqe *sqe, |
928 | struct io_comp_state *cs); | |
4349f30e | 929 | static void io_file_put_work(struct work_struct *work); |
de0617e4 | 930 | |
b63534c4 JA |
931 | static ssize_t io_import_iovec(int rw, struct io_kiocb *req, |
932 | struct iovec **iovec, struct iov_iter *iter, | |
933 | bool needs_lock); | |
ff6165b2 JA |
934 | static int io_setup_async_rw(struct io_kiocb *req, const struct iovec *iovec, |
935 | const struct iovec *fast_iov, | |
227c0c96 | 936 | struct iov_iter *iter, bool force); |
de0617e4 | 937 | |
2b188cc1 JA |
938 | static struct kmem_cache *req_cachep; |
939 | ||
940 | static const struct file_operations io_uring_fops; | |
941 | ||
942 | struct sock *io_uring_get_socket(struct file *file) | |
943 | { | |
944 | #if defined(CONFIG_UNIX) | |
945 | if (file->f_op == &io_uring_fops) { | |
946 | struct io_ring_ctx *ctx = file->private_data; | |
947 | ||
948 | return ctx->ring_sock->sk; | |
949 | } | |
950 | #endif | |
951 | return NULL; | |
952 | } | |
953 | EXPORT_SYMBOL(io_uring_get_socket); | |
954 | ||
3ca405eb PB |
955 | static inline void io_clean_op(struct io_kiocb *req) |
956 | { | |
bb175342 PB |
957 | if (req->flags & (REQ_F_NEED_CLEANUP | REQ_F_BUFFER_SELECTED | |
958 | REQ_F_INFLIGHT)) | |
3ca405eb PB |
959 | __io_clean_op(req); |
960 | } | |
961 | ||
4349f30e | 962 | static void io_sq_thread_drop_mm(void) |
c40f6379 JA |
963 | { |
964 | struct mm_struct *mm = current->mm; | |
965 | ||
966 | if (mm) { | |
967 | kthread_unuse_mm(mm); | |
968 | mmput(mm); | |
969 | } | |
970 | } | |
971 | ||
972 | static int __io_sq_thread_acquire_mm(struct io_ring_ctx *ctx) | |
973 | { | |
974 | if (!current->mm) { | |
cbcf7214 | 975 | if (unlikely(!(ctx->flags & IORING_SETUP_SQPOLL) || |
2aede0e4 JA |
976 | !ctx->sqo_task->mm || |
977 | !mmget_not_zero(ctx->sqo_task->mm))) | |
c40f6379 | 978 | return -EFAULT; |
2aede0e4 | 979 | kthread_use_mm(ctx->sqo_task->mm); |
c40f6379 JA |
980 | } |
981 | ||
982 | return 0; | |
983 | } | |
984 | ||
985 | static int io_sq_thread_acquire_mm(struct io_ring_ctx *ctx, | |
986 | struct io_kiocb *req) | |
987 | { | |
988 | if (!io_op_defs[req->opcode].needs_mm) | |
989 | return 0; | |
990 | return __io_sq_thread_acquire_mm(ctx); | |
991 | } | |
992 | ||
993 | static inline void req_set_fail_links(struct io_kiocb *req) | |
994 | { | |
995 | if ((req->flags & (REQ_F_LINK | REQ_F_HARDLINK)) == REQ_F_LINK) | |
996 | req->flags |= REQ_F_FAIL_LINK; | |
997 | } | |
4a38aed2 | 998 | |
7cdaf587 XW |
999 | /* |
1000 | * Note: must call io_req_init_async() for the first time you | |
1001 | * touch any members of io_wq_work. | |
1002 | */ | |
1003 | static inline void io_req_init_async(struct io_kiocb *req) | |
1004 | { | |
1005 | if (req->flags & REQ_F_WORK_INITIALIZED) | |
1006 | return; | |
1007 | ||
1008 | memset(&req->work, 0, sizeof(req->work)); | |
1009 | req->flags |= REQ_F_WORK_INITIALIZED; | |
1010 | } | |
1011 | ||
0cdaf760 PB |
1012 | static inline bool io_async_submit(struct io_ring_ctx *ctx) |
1013 | { | |
1014 | return ctx->flags & IORING_SETUP_SQPOLL; | |
1015 | } | |
1016 | ||
2b188cc1 JA |
1017 | static void io_ring_ctx_ref_free(struct percpu_ref *ref) |
1018 | { | |
1019 | struct io_ring_ctx *ctx = container_of(ref, struct io_ring_ctx, refs); | |
1020 | ||
0f158b4c | 1021 | complete(&ctx->ref_comp); |
2b188cc1 JA |
1022 | } |
1023 | ||
8eb7e2d0 PB |
1024 | static inline bool io_is_timeout_noseq(struct io_kiocb *req) |
1025 | { | |
1026 | return !req->timeout.off; | |
1027 | } | |
1028 | ||
2b188cc1 JA |
1029 | static struct io_ring_ctx *io_ring_ctx_alloc(struct io_uring_params *p) |
1030 | { | |
1031 | struct io_ring_ctx *ctx; | |
78076bb6 | 1032 | int hash_bits; |
2b188cc1 JA |
1033 | |
1034 | ctx = kzalloc(sizeof(*ctx), GFP_KERNEL); | |
1035 | if (!ctx) | |
1036 | return NULL; | |
1037 | ||
0ddf92e8 JA |
1038 | ctx->fallback_req = kmem_cache_alloc(req_cachep, GFP_KERNEL); |
1039 | if (!ctx->fallback_req) | |
1040 | goto err; | |
1041 | ||
78076bb6 JA |
1042 | /* |
1043 | * Use 5 bits less than the max cq entries, that should give us around | |
1044 | * 32 entries per hash list if totally full and uniformly spread. | |
1045 | */ | |
1046 | hash_bits = ilog2(p->cq_entries); | |
1047 | hash_bits -= 5; | |
1048 | if (hash_bits <= 0) | |
1049 | hash_bits = 1; | |
1050 | ctx->cancel_hash_bits = hash_bits; | |
1051 | ctx->cancel_hash = kmalloc((1U << hash_bits) * sizeof(struct hlist_head), | |
1052 | GFP_KERNEL); | |
1053 | if (!ctx->cancel_hash) | |
1054 | goto err; | |
1055 | __hash_init(ctx->cancel_hash, 1U << hash_bits); | |
1056 | ||
21482896 | 1057 | if (percpu_ref_init(&ctx->refs, io_ring_ctx_ref_free, |
206aefde JA |
1058 | PERCPU_REF_ALLOW_REINIT, GFP_KERNEL)) |
1059 | goto err; | |
2b188cc1 JA |
1060 | |
1061 | ctx->flags = p->flags; | |
6a779382 JA |
1062 | init_waitqueue_head(&ctx->__sqo_wait); |
1063 | ctx->sqo_wait = &ctx->__sqo_wait; | |
2b188cc1 | 1064 | init_waitqueue_head(&ctx->cq_wait); |
1d7bb1d5 | 1065 | INIT_LIST_HEAD(&ctx->cq_overflow_list); |
0f158b4c JA |
1066 | init_completion(&ctx->ref_comp); |
1067 | init_completion(&ctx->sq_thread_comp); | |
5a2e745d | 1068 | idr_init(&ctx->io_buffer_idr); |
071698e1 | 1069 | idr_init(&ctx->personality_idr); |
2b188cc1 JA |
1070 | mutex_init(&ctx->uring_lock); |
1071 | init_waitqueue_head(&ctx->wait); | |
1072 | spin_lock_init(&ctx->completion_lock); | |
540e32a0 | 1073 | INIT_LIST_HEAD(&ctx->iopoll_list); |
de0617e4 | 1074 | INIT_LIST_HEAD(&ctx->defer_list); |
5262f567 | 1075 | INIT_LIST_HEAD(&ctx->timeout_list); |
fcb323cc JA |
1076 | init_waitqueue_head(&ctx->inflight_wait); |
1077 | spin_lock_init(&ctx->inflight_lock); | |
1078 | INIT_LIST_HEAD(&ctx->inflight_list); | |
4a38aed2 JA |
1079 | INIT_DELAYED_WORK(&ctx->file_put_work, io_file_put_work); |
1080 | init_llist_head(&ctx->file_put_llist); | |
2b188cc1 | 1081 | return ctx; |
206aefde | 1082 | err: |
0ddf92e8 JA |
1083 | if (ctx->fallback_req) |
1084 | kmem_cache_free(req_cachep, ctx->fallback_req); | |
78076bb6 | 1085 | kfree(ctx->cancel_hash); |
206aefde JA |
1086 | kfree(ctx); |
1087 | return NULL; | |
2b188cc1 JA |
1088 | } |
1089 | ||
9cf7c104 | 1090 | static bool req_need_defer(struct io_kiocb *req, u32 seq) |
7adf4eaf | 1091 | { |
2bc9930e JA |
1092 | if (unlikely(req->flags & REQ_F_IO_DRAIN)) { |
1093 | struct io_ring_ctx *ctx = req->ctx; | |
a197f664 | 1094 | |
9cf7c104 | 1095 | return seq != ctx->cached_cq_tail |
31af27c7 | 1096 | + atomic_read(&ctx->cached_cq_overflow); |
2bc9930e | 1097 | } |
de0617e4 | 1098 | |
9d858b21 | 1099 | return false; |
de0617e4 JA |
1100 | } |
1101 | ||
de0617e4 | 1102 | static void __io_commit_cqring(struct io_ring_ctx *ctx) |
2b188cc1 | 1103 | { |
75b28aff | 1104 | struct io_rings *rings = ctx->rings; |
2b188cc1 | 1105 | |
07910158 PB |
1106 | /* order cqe stores with ring update */ |
1107 | smp_store_release(&rings->cq.tail, ctx->cached_cq_tail); | |
2b188cc1 | 1108 | |
07910158 PB |
1109 | if (wq_has_sleeper(&ctx->cq_wait)) { |
1110 | wake_up_interruptible(&ctx->cq_wait); | |
1111 | kill_fasync(&ctx->cq_fasync, SIGIO, POLL_IN); | |
2b188cc1 JA |
1112 | } |
1113 | } | |
1114 | ||
51a4cc11 JA |
1115 | /* |
1116 | * Returns true if we need to defer file table putting. This can only happen | |
1117 | * from the error path with REQ_F_COMP_LOCKED set. | |
1118 | */ | |
1119 | static bool io_req_clean_work(struct io_kiocb *req) | |
18d9be1a | 1120 | { |
7cdaf587 | 1121 | if (!(req->flags & REQ_F_WORK_INITIALIZED)) |
51a4cc11 JA |
1122 | return false; |
1123 | ||
1124 | req->flags &= ~REQ_F_WORK_INITIALIZED; | |
7cdaf587 | 1125 | |
cccf0ee8 JA |
1126 | if (req->work.mm) { |
1127 | mmdrop(req->work.mm); | |
1128 | req->work.mm = NULL; | |
1129 | } | |
1130 | if (req->work.creds) { | |
1131 | put_cred(req->work.creds); | |
1132 | req->work.creds = NULL; | |
1133 | } | |
ff002b30 JA |
1134 | if (req->work.fs) { |
1135 | struct fs_struct *fs = req->work.fs; | |
1136 | ||
51a4cc11 JA |
1137 | if (req->flags & REQ_F_COMP_LOCKED) |
1138 | return true; | |
1139 | ||
ff002b30 JA |
1140 | spin_lock(&req->work.fs->lock); |
1141 | if (--fs->users) | |
1142 | fs = NULL; | |
1143 | spin_unlock(&req->work.fs->lock); | |
1144 | if (fs) | |
1145 | free_fs_struct(fs); | |
b65e0dd6 | 1146 | req->work.fs = NULL; |
ff002b30 | 1147 | } |
51a4cc11 JA |
1148 | |
1149 | return false; | |
561fb04a JA |
1150 | } |
1151 | ||
cbdcb435 | 1152 | static void io_prep_async_work(struct io_kiocb *req) |
18d9be1a | 1153 | { |
d3656344 | 1154 | const struct io_op_def *def = &io_op_defs[req->opcode]; |
54a91f3b | 1155 | |
16d59803 PB |
1156 | io_req_init_async(req); |
1157 | ||
d3656344 | 1158 | if (req->flags & REQ_F_ISREG) { |
eefdf30f | 1159 | if (def->hash_reg_file || (req->ctx->flags & IORING_SETUP_IOPOLL)) |
8766dd51 | 1160 | io_wq_hash_work(&req->work, file_inode(req->file)); |
d3656344 JA |
1161 | } else { |
1162 | if (def->unbound_nonreg_file) | |
3529d8c2 | 1163 | req->work.flags |= IO_WQ_WORK_UNBOUND; |
54a91f3b | 1164 | } |
dca9cf8b PB |
1165 | if (!req->work.mm && def->needs_mm) { |
1166 | mmgrab(current->mm); | |
1167 | req->work.mm = current->mm; | |
1168 | } | |
1169 | if (!req->work.creds) | |
1170 | req->work.creds = get_current_cred(); | |
1171 | if (!req->work.fs && def->needs_fs) { | |
1172 | spin_lock(¤t->fs->lock); | |
1173 | if (!current->fs->in_exec) { | |
1174 | req->work.fs = current->fs; | |
1175 | req->work.fs->users++; | |
1176 | } else { | |
1177 | req->work.flags |= IO_WQ_WORK_CANCEL; | |
1178 | } | |
1179 | spin_unlock(¤t->fs->lock); | |
1180 | } | |
57f1a649 PB |
1181 | if (def->needs_fsize) |
1182 | req->work.fsize = rlimit(RLIMIT_FSIZE); | |
1183 | else | |
1184 | req->work.fsize = RLIM_INFINITY; | |
561fb04a | 1185 | } |
cccf0ee8 | 1186 | |
cbdcb435 | 1187 | static void io_prep_async_link(struct io_kiocb *req) |
561fb04a | 1188 | { |
cbdcb435 | 1189 | struct io_kiocb *cur; |
54a91f3b | 1190 | |
cbdcb435 PB |
1191 | io_prep_async_work(req); |
1192 | if (req->flags & REQ_F_LINK_HEAD) | |
1193 | list_for_each_entry(cur, &req->link_list, link_list) | |
1194 | io_prep_async_work(cur); | |
561fb04a JA |
1195 | } |
1196 | ||
7271ef3a | 1197 | static struct io_kiocb *__io_queue_async_work(struct io_kiocb *req) |
561fb04a | 1198 | { |
a197f664 | 1199 | struct io_ring_ctx *ctx = req->ctx; |
cbdcb435 | 1200 | struct io_kiocb *link = io_prep_linked_timeout(req); |
561fb04a | 1201 | |
8766dd51 PB |
1202 | trace_io_uring_queue_async_work(ctx, io_wq_is_hashed(&req->work), req, |
1203 | &req->work, req->flags); | |
1204 | io_wq_enqueue(ctx->io_wq, &req->work); | |
7271ef3a | 1205 | return link; |
18d9be1a JA |
1206 | } |
1207 | ||
cbdcb435 PB |
1208 | static void io_queue_async_work(struct io_kiocb *req) |
1209 | { | |
7271ef3a JA |
1210 | struct io_kiocb *link; |
1211 | ||
cbdcb435 PB |
1212 | /* init ->work of the whole link before punting */ |
1213 | io_prep_async_link(req); | |
7271ef3a JA |
1214 | link = __io_queue_async_work(req); |
1215 | ||
1216 | if (link) | |
1217 | io_queue_linked_timeout(link); | |
cbdcb435 PB |
1218 | } |
1219 | ||
5262f567 JA |
1220 | static void io_kill_timeout(struct io_kiocb *req) |
1221 | { | |
1222 | int ret; | |
1223 | ||
2d28390a | 1224 | ret = hrtimer_try_to_cancel(&req->io->timeout.timer); |
5262f567 | 1225 | if (ret != -1) { |
01cec8c1 PB |
1226 | atomic_set(&req->ctx->cq_timeouts, |
1227 | atomic_read(&req->ctx->cq_timeouts) + 1); | |
135fcde8 | 1228 | list_del_init(&req->timeout.list); |
f0e20b89 | 1229 | req->flags |= REQ_F_COMP_LOCKED; |
78e19bbe | 1230 | io_cqring_fill_event(req, 0); |
ec9c02ad | 1231 | io_put_req(req); |
5262f567 JA |
1232 | } |
1233 | } | |
1234 | ||
f3606e3a JA |
1235 | static bool io_task_match(struct io_kiocb *req, struct task_struct *tsk) |
1236 | { | |
1237 | struct io_ring_ctx *ctx = req->ctx; | |
1238 | ||
1239 | if (!tsk || req->task == tsk) | |
1240 | return true; | |
1241 | if ((ctx->flags & IORING_SETUP_SQPOLL) && req->task == ctx->sqo_thread) | |
1242 | return true; | |
1243 | return false; | |
1244 | } | |
1245 | ||
76e1b642 JA |
1246 | /* |
1247 | * Returns true if we found and killed one or more timeouts | |
1248 | */ | |
1249 | static bool io_kill_timeouts(struct io_ring_ctx *ctx, struct task_struct *tsk) | |
5262f567 JA |
1250 | { |
1251 | struct io_kiocb *req, *tmp; | |
76e1b642 | 1252 | int canceled = 0; |
5262f567 JA |
1253 | |
1254 | spin_lock_irq(&ctx->completion_lock); | |
f3606e3a | 1255 | list_for_each_entry_safe(req, tmp, &ctx->timeout_list, timeout.list) { |
76e1b642 | 1256 | if (io_task_match(req, tsk)) { |
f3606e3a | 1257 | io_kill_timeout(req); |
76e1b642 JA |
1258 | canceled++; |
1259 | } | |
f3606e3a | 1260 | } |
5262f567 | 1261 | spin_unlock_irq(&ctx->completion_lock); |
76e1b642 | 1262 | return canceled != 0; |
5262f567 JA |
1263 | } |
1264 | ||
04518945 | 1265 | static void __io_queue_deferred(struct io_ring_ctx *ctx) |
de0617e4 | 1266 | { |
04518945 | 1267 | do { |
27dc8338 PB |
1268 | struct io_defer_entry *de = list_first_entry(&ctx->defer_list, |
1269 | struct io_defer_entry, list); | |
7271ef3a | 1270 | struct io_kiocb *link; |
de0617e4 | 1271 | |
9cf7c104 | 1272 | if (req_need_defer(de->req, de->seq)) |
04518945 | 1273 | break; |
27dc8338 | 1274 | list_del_init(&de->list); |
cbdcb435 | 1275 | /* punt-init is done before queueing for defer */ |
7271ef3a JA |
1276 | link = __io_queue_async_work(de->req); |
1277 | if (link) { | |
1278 | __io_queue_linked_timeout(link); | |
1279 | /* drop submission reference */ | |
1280 | link->flags |= REQ_F_COMP_LOCKED; | |
1281 | io_put_req(link); | |
1282 | } | |
27dc8338 | 1283 | kfree(de); |
04518945 PB |
1284 | } while (!list_empty(&ctx->defer_list)); |
1285 | } | |
1286 | ||
360428f8 | 1287 | static void io_flush_timeouts(struct io_ring_ctx *ctx) |
de0617e4 | 1288 | { |
360428f8 PB |
1289 | while (!list_empty(&ctx->timeout_list)) { |
1290 | struct io_kiocb *req = list_first_entry(&ctx->timeout_list, | |
135fcde8 | 1291 | struct io_kiocb, timeout.list); |
de0617e4 | 1292 | |
8eb7e2d0 | 1293 | if (io_is_timeout_noseq(req)) |
360428f8 | 1294 | break; |
bfe68a22 PB |
1295 | if (req->timeout.target_seq != ctx->cached_cq_tail |
1296 | - atomic_read(&ctx->cq_timeouts)) | |
360428f8 | 1297 | break; |
bfe68a22 | 1298 | |
135fcde8 | 1299 | list_del_init(&req->timeout.list); |
5262f567 | 1300 | io_kill_timeout(req); |
360428f8 PB |
1301 | } |
1302 | } | |
5262f567 | 1303 | |
360428f8 PB |
1304 | static void io_commit_cqring(struct io_ring_ctx *ctx) |
1305 | { | |
1306 | io_flush_timeouts(ctx); | |
de0617e4 JA |
1307 | __io_commit_cqring(ctx); |
1308 | ||
04518945 PB |
1309 | if (unlikely(!list_empty(&ctx->defer_list))) |
1310 | __io_queue_deferred(ctx); | |
de0617e4 JA |
1311 | } |
1312 | ||
2b188cc1 JA |
1313 | static struct io_uring_cqe *io_get_cqring(struct io_ring_ctx *ctx) |
1314 | { | |
75b28aff | 1315 | struct io_rings *rings = ctx->rings; |
2b188cc1 JA |
1316 | unsigned tail; |
1317 | ||
1318 | tail = ctx->cached_cq_tail; | |
115e12e5 SB |
1319 | /* |
1320 | * writes to the cq entry need to come after reading head; the | |
1321 | * control dependency is enough as we're using WRITE_ONCE to | |
1322 | * fill the cq entry | |
1323 | */ | |
75b28aff | 1324 | if (tail - READ_ONCE(rings->cq.head) == rings->cq_ring_entries) |
2b188cc1 JA |
1325 | return NULL; |
1326 | ||
1327 | ctx->cached_cq_tail++; | |
75b28aff | 1328 | return &rings->cqes[tail & ctx->cq_mask]; |
2b188cc1 JA |
1329 | } |
1330 | ||
f2842ab5 JA |
1331 | static inline bool io_should_trigger_evfd(struct io_ring_ctx *ctx) |
1332 | { | |
f0b493e6 JA |
1333 | if (!ctx->cq_ev_fd) |
1334 | return false; | |
7e55a19c SG |
1335 | if (READ_ONCE(ctx->rings->cq_flags) & IORING_CQ_EVENTFD_DISABLED) |
1336 | return false; | |
f2842ab5 JA |
1337 | if (!ctx->eventfd_async) |
1338 | return true; | |
b41e9852 | 1339 | return io_wq_current_is_worker(); |
f2842ab5 JA |
1340 | } |
1341 | ||
b41e9852 | 1342 | static void io_cqring_ev_posted(struct io_ring_ctx *ctx) |
1d7bb1d5 JA |
1343 | { |
1344 | if (waitqueue_active(&ctx->wait)) | |
1345 | wake_up(&ctx->wait); | |
6a779382 JA |
1346 | if (waitqueue_active(ctx->sqo_wait)) |
1347 | wake_up(ctx->sqo_wait); | |
b41e9852 | 1348 | if (io_should_trigger_evfd(ctx)) |
1d7bb1d5 JA |
1349 | eventfd_signal(ctx->cq_ev_fd, 1); |
1350 | } | |
1351 | ||
46930143 PB |
1352 | static void io_cqring_mark_overflow(struct io_ring_ctx *ctx) |
1353 | { | |
1354 | if (list_empty(&ctx->cq_overflow_list)) { | |
1355 | clear_bit(0, &ctx->sq_check_overflow); | |
1356 | clear_bit(0, &ctx->cq_check_overflow); | |
1357 | ctx->rings->sq_flags &= ~IORING_SQ_CQ_OVERFLOW; | |
1358 | } | |
1359 | } | |
1360 | ||
e6c8aa9a JA |
1361 | static inline bool io_match_files(struct io_kiocb *req, |
1362 | struct files_struct *files) | |
1363 | { | |
1364 | if (!files) | |
1365 | return true; | |
1366 | if (req->flags & REQ_F_WORK_INITIALIZED) | |
1367 | return req->work.files == files; | |
1368 | return false; | |
1369 | } | |
1370 | ||
c4a2ed72 | 1371 | /* Returns true if there are no backlogged entries after the flush */ |
e6c8aa9a JA |
1372 | static bool io_cqring_overflow_flush(struct io_ring_ctx *ctx, bool force, |
1373 | struct task_struct *tsk, | |
1374 | struct files_struct *files) | |
1d7bb1d5 JA |
1375 | { |
1376 | struct io_rings *rings = ctx->rings; | |
e6c8aa9a | 1377 | struct io_kiocb *req, *tmp; |
1d7bb1d5 | 1378 | struct io_uring_cqe *cqe; |
1d7bb1d5 JA |
1379 | unsigned long flags; |
1380 | LIST_HEAD(list); | |
1381 | ||
1382 | if (!force) { | |
1383 | if (list_empty_careful(&ctx->cq_overflow_list)) | |
c4a2ed72 | 1384 | return true; |
1d7bb1d5 JA |
1385 | if ((ctx->cached_cq_tail - READ_ONCE(rings->cq.head) == |
1386 | rings->cq_ring_entries)) | |
c4a2ed72 | 1387 | return false; |
1d7bb1d5 JA |
1388 | } |
1389 | ||
1390 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
1391 | ||
1392 | /* if force is set, the ring is going away. always drop after that */ | |
1393 | if (force) | |
69b3e546 | 1394 | ctx->cq_overflow_flushed = 1; |
1d7bb1d5 | 1395 | |
c4a2ed72 | 1396 | cqe = NULL; |
e6c8aa9a JA |
1397 | list_for_each_entry_safe(req, tmp, &ctx->cq_overflow_list, compl.list) { |
1398 | if (tsk && req->task != tsk) | |
1399 | continue; | |
1400 | if (!io_match_files(req, files)) | |
1401 | continue; | |
1402 | ||
1d7bb1d5 JA |
1403 | cqe = io_get_cqring(ctx); |
1404 | if (!cqe && !force) | |
1405 | break; | |
1406 | ||
40d8ddd4 | 1407 | list_move(&req->compl.list, &list); |
1d7bb1d5 JA |
1408 | if (cqe) { |
1409 | WRITE_ONCE(cqe->user_data, req->user_data); | |
1410 | WRITE_ONCE(cqe->res, req->result); | |
0f7e466b | 1411 | WRITE_ONCE(cqe->flags, req->compl.cflags); |
1d7bb1d5 JA |
1412 | } else { |
1413 | WRITE_ONCE(ctx->rings->cq_overflow, | |
1414 | atomic_inc_return(&ctx->cached_cq_overflow)); | |
1415 | } | |
1416 | } | |
1417 | ||
1418 | io_commit_cqring(ctx); | |
46930143 PB |
1419 | io_cqring_mark_overflow(ctx); |
1420 | ||
1d7bb1d5 JA |
1421 | spin_unlock_irqrestore(&ctx->completion_lock, flags); |
1422 | io_cqring_ev_posted(ctx); | |
1423 | ||
1424 | while (!list_empty(&list)) { | |
40d8ddd4 PB |
1425 | req = list_first_entry(&list, struct io_kiocb, compl.list); |
1426 | list_del(&req->compl.list); | |
ec9c02ad | 1427 | io_put_req(req); |
1d7bb1d5 | 1428 | } |
c4a2ed72 JA |
1429 | |
1430 | return cqe != NULL; | |
1d7bb1d5 JA |
1431 | } |
1432 | ||
bcda7baa | 1433 | static void __io_cqring_fill_event(struct io_kiocb *req, long res, long cflags) |
2b188cc1 | 1434 | { |
78e19bbe | 1435 | struct io_ring_ctx *ctx = req->ctx; |
2b188cc1 JA |
1436 | struct io_uring_cqe *cqe; |
1437 | ||
78e19bbe | 1438 | trace_io_uring_complete(ctx, req->user_data, res); |
51c3ff62 | 1439 | |
2b188cc1 JA |
1440 | /* |
1441 | * If we can't get a cq entry, userspace overflowed the | |
1442 | * submission (by quite a lot). Increment the overflow count in | |
1443 | * the ring. | |
1444 | */ | |
1445 | cqe = io_get_cqring(ctx); | |
1d7bb1d5 | 1446 | if (likely(cqe)) { |
78e19bbe | 1447 | WRITE_ONCE(cqe->user_data, req->user_data); |
2b188cc1 | 1448 | WRITE_ONCE(cqe->res, res); |
bcda7baa | 1449 | WRITE_ONCE(cqe->flags, cflags); |
0f212204 JA |
1450 | } else if (ctx->cq_overflow_flushed || req->task->io_uring->in_idle) { |
1451 | /* | |
1452 | * If we're in ring overflow flush mode, or in task cancel mode, | |
1453 | * then we cannot store the request for later flushing, we need | |
1454 | * to drop it on the floor. | |
1455 | */ | |
498ccd9e JA |
1456 | WRITE_ONCE(ctx->rings->cq_overflow, |
1457 | atomic_inc_return(&ctx->cached_cq_overflow)); | |
1d7bb1d5 | 1458 | } else { |
ad3eb2c8 JA |
1459 | if (list_empty(&ctx->cq_overflow_list)) { |
1460 | set_bit(0, &ctx->sq_check_overflow); | |
1461 | set_bit(0, &ctx->cq_check_overflow); | |
6d5f9049 | 1462 | ctx->rings->sq_flags |= IORING_SQ_CQ_OVERFLOW; |
ad3eb2c8 | 1463 | } |
40d8ddd4 | 1464 | io_clean_op(req); |
1d7bb1d5 | 1465 | req->result = res; |
0f7e466b | 1466 | req->compl.cflags = cflags; |
40d8ddd4 PB |
1467 | refcount_inc(&req->refs); |
1468 | list_add_tail(&req->compl.list, &ctx->cq_overflow_list); | |
2b188cc1 JA |
1469 | } |
1470 | } | |
1471 | ||
bcda7baa JA |
1472 | static void io_cqring_fill_event(struct io_kiocb *req, long res) |
1473 | { | |
1474 | __io_cqring_fill_event(req, res, 0); | |
1475 | } | |
1476 | ||
e1e16097 | 1477 | static void io_cqring_add_event(struct io_kiocb *req, long res, long cflags) |
2b188cc1 | 1478 | { |
78e19bbe | 1479 | struct io_ring_ctx *ctx = req->ctx; |
2b188cc1 JA |
1480 | unsigned long flags; |
1481 | ||
1482 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
bcda7baa | 1483 | __io_cqring_fill_event(req, res, cflags); |
2b188cc1 JA |
1484 | io_commit_cqring(ctx); |
1485 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
1486 | ||
8c838788 | 1487 | io_cqring_ev_posted(ctx); |
2b188cc1 JA |
1488 | } |
1489 | ||
229a7b63 | 1490 | static void io_submit_flush_completions(struct io_comp_state *cs) |
bcda7baa | 1491 | { |
229a7b63 JA |
1492 | struct io_ring_ctx *ctx = cs->ctx; |
1493 | ||
1494 | spin_lock_irq(&ctx->completion_lock); | |
1495 | while (!list_empty(&cs->list)) { | |
1496 | struct io_kiocb *req; | |
1497 | ||
3ca405eb PB |
1498 | req = list_first_entry(&cs->list, struct io_kiocb, compl.list); |
1499 | list_del(&req->compl.list); | |
0f7e466b | 1500 | __io_cqring_fill_event(req, req->result, req->compl.cflags); |
229a7b63 JA |
1501 | if (!(req->flags & REQ_F_LINK_HEAD)) { |
1502 | req->flags |= REQ_F_COMP_LOCKED; | |
1503 | io_put_req(req); | |
1504 | } else { | |
1505 | spin_unlock_irq(&ctx->completion_lock); | |
1506 | io_put_req(req); | |
1507 | spin_lock_irq(&ctx->completion_lock); | |
1508 | } | |
1509 | } | |
1510 | io_commit_cqring(ctx); | |
1511 | spin_unlock_irq(&ctx->completion_lock); | |
1512 | ||
1513 | io_cqring_ev_posted(ctx); | |
1514 | cs->nr = 0; | |
1515 | } | |
1516 | ||
1517 | static void __io_req_complete(struct io_kiocb *req, long res, unsigned cflags, | |
1518 | struct io_comp_state *cs) | |
1519 | { | |
1520 | if (!cs) { | |
1521 | io_cqring_add_event(req, res, cflags); | |
1522 | io_put_req(req); | |
1523 | } else { | |
3ca405eb | 1524 | io_clean_op(req); |
229a7b63 | 1525 | req->result = res; |
0f7e466b | 1526 | req->compl.cflags = cflags; |
3ca405eb | 1527 | list_add_tail(&req->compl.list, &cs->list); |
229a7b63 JA |
1528 | if (++cs->nr >= 32) |
1529 | io_submit_flush_completions(cs); | |
1530 | } | |
e1e16097 JA |
1531 | } |
1532 | ||
1533 | static void io_req_complete(struct io_kiocb *req, long res) | |
bcda7baa | 1534 | { |
229a7b63 | 1535 | __io_req_complete(req, res, 0, NULL); |
bcda7baa JA |
1536 | } |
1537 | ||
0ddf92e8 JA |
1538 | static inline bool io_is_fallback_req(struct io_kiocb *req) |
1539 | { | |
1540 | return req == (struct io_kiocb *) | |
1541 | ((unsigned long) req->ctx->fallback_req & ~1UL); | |
1542 | } | |
1543 | ||
1544 | static struct io_kiocb *io_get_fallback_req(struct io_ring_ctx *ctx) | |
1545 | { | |
1546 | struct io_kiocb *req; | |
1547 | ||
1548 | req = ctx->fallback_req; | |
dd461af6 | 1549 | if (!test_and_set_bit_lock(0, (unsigned long *) &ctx->fallback_req)) |
0ddf92e8 JA |
1550 | return req; |
1551 | ||
1552 | return NULL; | |
1553 | } | |
1554 | ||
0553b8bd PB |
1555 | static struct io_kiocb *io_alloc_req(struct io_ring_ctx *ctx, |
1556 | struct io_submit_state *state) | |
2b188cc1 | 1557 | { |
fd6fab2c | 1558 | gfp_t gfp = GFP_KERNEL | __GFP_NOWARN; |
2b188cc1 JA |
1559 | struct io_kiocb *req; |
1560 | ||
f6b6c7d6 | 1561 | if (!state->free_reqs) { |
2579f913 JA |
1562 | size_t sz; |
1563 | int ret; | |
1564 | ||
1565 | sz = min_t(size_t, state->ios_left, ARRAY_SIZE(state->reqs)); | |
fd6fab2c JA |
1566 | ret = kmem_cache_alloc_bulk(req_cachep, gfp, sz, state->reqs); |
1567 | ||
1568 | /* | |
1569 | * Bulk alloc is all-or-nothing. If we fail to get a batch, | |
1570 | * retry single alloc to be on the safe side. | |
1571 | */ | |
1572 | if (unlikely(ret <= 0)) { | |
1573 | state->reqs[0] = kmem_cache_alloc(req_cachep, gfp); | |
1574 | if (!state->reqs[0]) | |
0ddf92e8 | 1575 | goto fallback; |
fd6fab2c JA |
1576 | ret = 1; |
1577 | } | |
2579f913 | 1578 | state->free_reqs = ret - 1; |
6c8a3134 | 1579 | req = state->reqs[ret - 1]; |
2579f913 | 1580 | } else { |
2579f913 | 1581 | state->free_reqs--; |
6c8a3134 | 1582 | req = state->reqs[state->free_reqs]; |
2b188cc1 JA |
1583 | } |
1584 | ||
2579f913 | 1585 | return req; |
0ddf92e8 | 1586 | fallback: |
0553b8bd | 1587 | return io_get_fallback_req(ctx); |
2b188cc1 JA |
1588 | } |
1589 | ||
8da11c19 PB |
1590 | static inline void io_put_file(struct io_kiocb *req, struct file *file, |
1591 | bool fixed) | |
1592 | { | |
1593 | if (fixed) | |
05589553 | 1594 | percpu_ref_put(req->fixed_file_refs); |
8da11c19 PB |
1595 | else |
1596 | fput(file); | |
1597 | } | |
1598 | ||
51a4cc11 | 1599 | static bool io_dismantle_req(struct io_kiocb *req) |
2b188cc1 | 1600 | { |
3ca405eb | 1601 | io_clean_op(req); |
929a3af9 | 1602 | |
5acbbc8e JA |
1603 | if (req->io) |
1604 | kfree(req->io); | |
8da11c19 PB |
1605 | if (req->file) |
1606 | io_put_file(req, req->file, (req->flags & REQ_F_FIXED_FILE)); | |
fcb323cc | 1607 | |
51a4cc11 | 1608 | return io_req_clean_work(req); |
e65ef56d JA |
1609 | } |
1610 | ||
51a4cc11 | 1611 | static void __io_free_req_finish(struct io_kiocb *req) |
c6ca97b3 | 1612 | { |
0f212204 | 1613 | struct io_uring_task *tctx = req->task->io_uring; |
51a4cc11 | 1614 | struct io_ring_ctx *ctx = req->ctx; |
c6ca97b3 | 1615 | |
0f212204 JA |
1616 | atomic_long_inc(&tctx->req_complete); |
1617 | if (tctx->in_idle) | |
1618 | wake_up(&tctx->wait); | |
e3bc8e9d JA |
1619 | put_task_struct(req->task); |
1620 | ||
b1e50e54 PB |
1621 | if (likely(!io_is_fallback_req(req))) |
1622 | kmem_cache_free(req_cachep, req); | |
1623 | else | |
ecfc5177 PB |
1624 | clear_bit_unlock(0, (unsigned long *) &ctx->fallback_req); |
1625 | percpu_ref_put(&ctx->refs); | |
e65ef56d JA |
1626 | } |
1627 | ||
51a4cc11 JA |
1628 | static void io_req_task_file_table_put(struct callback_head *cb) |
1629 | { | |
1630 | struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work); | |
1631 | struct fs_struct *fs = req->work.fs; | |
1632 | ||
1633 | spin_lock(&req->work.fs->lock); | |
1634 | if (--fs->users) | |
1635 | fs = NULL; | |
1636 | spin_unlock(&req->work.fs->lock); | |
1637 | if (fs) | |
1638 | free_fs_struct(fs); | |
1639 | req->work.fs = NULL; | |
1640 | __io_free_req_finish(req); | |
1641 | } | |
1642 | ||
1643 | static void __io_free_req(struct io_kiocb *req) | |
1644 | { | |
1645 | if (!io_dismantle_req(req)) { | |
1646 | __io_free_req_finish(req); | |
1647 | } else { | |
1648 | int ret; | |
1649 | ||
1650 | init_task_work(&req->task_work, io_req_task_file_table_put); | |
1651 | ret = task_work_add(req->task, &req->task_work, TWA_RESUME); | |
1652 | if (unlikely(ret)) { | |
1653 | struct task_struct *tsk; | |
1654 | ||
1655 | tsk = io_wq_get_task(req->ctx->io_wq); | |
1656 | task_work_add(tsk, &req->task_work, 0); | |
1657 | } | |
1658 | } | |
1659 | } | |
1660 | ||
a197f664 | 1661 | static bool io_link_cancel_timeout(struct io_kiocb *req) |
2665abfd | 1662 | { |
a197f664 | 1663 | struct io_ring_ctx *ctx = req->ctx; |
2665abfd JA |
1664 | int ret; |
1665 | ||
2d28390a | 1666 | ret = hrtimer_try_to_cancel(&req->io->timeout.timer); |
2665abfd | 1667 | if (ret != -1) { |
78e19bbe | 1668 | io_cqring_fill_event(req, -ECANCELED); |
2665abfd | 1669 | io_commit_cqring(ctx); |
dea3b49c | 1670 | req->flags &= ~REQ_F_LINK_HEAD; |
ec9c02ad | 1671 | io_put_req(req); |
2665abfd JA |
1672 | return true; |
1673 | } | |
1674 | ||
1675 | return false; | |
e65ef56d JA |
1676 | } |
1677 | ||
ab0b6451 | 1678 | static bool __io_kill_linked_timeout(struct io_kiocb *req) |
9e645e11 | 1679 | { |
7c86ffee | 1680 | struct io_kiocb *link; |
ab0b6451 | 1681 | bool wake_ev; |
7c86ffee PB |
1682 | |
1683 | if (list_empty(&req->link_list)) | |
ab0b6451 | 1684 | return false; |
7c86ffee PB |
1685 | link = list_first_entry(&req->link_list, struct io_kiocb, link_list); |
1686 | if (link->opcode != IORING_OP_LINK_TIMEOUT) | |
ab0b6451 | 1687 | return false; |
7c86ffee PB |
1688 | |
1689 | list_del_init(&link->link_list); | |
9b7adba9 | 1690 | link->flags |= REQ_F_COMP_LOCKED; |
7c86ffee PB |
1691 | wake_ev = io_link_cancel_timeout(link); |
1692 | req->flags &= ~REQ_F_LINK_TIMEOUT; | |
ab0b6451 JA |
1693 | return wake_ev; |
1694 | } | |
1695 | ||
1696 | static void io_kill_linked_timeout(struct io_kiocb *req) | |
9e645e11 | 1697 | { |
2665abfd | 1698 | struct io_ring_ctx *ctx = req->ctx; |
ab0b6451 | 1699 | bool wake_ev; |
9e645e11 | 1700 | |
ab0b6451 JA |
1701 | if (!(req->flags & REQ_F_COMP_LOCKED)) { |
1702 | unsigned long flags; | |
1703 | ||
1704 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
1705 | wake_ev = __io_kill_linked_timeout(req); | |
7c86ffee | 1706 | spin_unlock_irqrestore(&ctx->completion_lock, flags); |
ab0b6451 JA |
1707 | } else { |
1708 | wake_ev = __io_kill_linked_timeout(req); | |
1709 | } | |
1710 | ||
7c86ffee PB |
1711 | if (wake_ev) |
1712 | io_cqring_ev_posted(ctx); | |
1713 | } | |
1714 | ||
9b5f7bd9 | 1715 | static struct io_kiocb *io_req_link_next(struct io_kiocb *req) |
7c86ffee PB |
1716 | { |
1717 | struct io_kiocb *nxt; | |
4d7dd462 | 1718 | |
9e645e11 JA |
1719 | /* |
1720 | * The list should never be empty when we are called here. But could | |
1721 | * potentially happen if the chain is messed up, check to be on the | |
1722 | * safe side. | |
1723 | */ | |
7c86ffee | 1724 | if (unlikely(list_empty(&req->link_list))) |
9b5f7bd9 | 1725 | return NULL; |
2665abfd | 1726 | |
7c86ffee PB |
1727 | nxt = list_first_entry(&req->link_list, struct io_kiocb, link_list); |
1728 | list_del_init(&req->link_list); | |
1729 | if (!list_empty(&nxt->link_list)) | |
1730 | nxt->flags |= REQ_F_LINK_HEAD; | |
9b5f7bd9 | 1731 | return nxt; |
9e645e11 JA |
1732 | } |
1733 | ||
1734 | /* | |
dea3b49c | 1735 | * Called if REQ_F_LINK_HEAD is set, and we fail the head request |
9e645e11 | 1736 | */ |
7c86ffee | 1737 | static void __io_fail_links(struct io_kiocb *req) |
9e645e11 | 1738 | { |
2665abfd | 1739 | struct io_ring_ctx *ctx = req->ctx; |
9e645e11 JA |
1740 | |
1741 | while (!list_empty(&req->link_list)) { | |
4493233e PB |
1742 | struct io_kiocb *link = list_first_entry(&req->link_list, |
1743 | struct io_kiocb, link_list); | |
9e645e11 | 1744 | |
4493233e | 1745 | list_del_init(&link->link_list); |
c826bd7a | 1746 | trace_io_uring_fail_link(req, link); |
2665abfd | 1747 | |
7c86ffee | 1748 | io_cqring_fill_event(link, -ECANCELED); |
9b7adba9 | 1749 | link->flags |= REQ_F_COMP_LOCKED; |
7c86ffee | 1750 | __io_double_put_req(link); |
5d960724 | 1751 | req->flags &= ~REQ_F_LINK_TIMEOUT; |
9e645e11 | 1752 | } |
2665abfd JA |
1753 | |
1754 | io_commit_cqring(ctx); | |
2665abfd | 1755 | io_cqring_ev_posted(ctx); |
9e645e11 JA |
1756 | } |
1757 | ||
7c86ffee | 1758 | static void io_fail_links(struct io_kiocb *req) |
9e645e11 | 1759 | { |
7c86ffee | 1760 | struct io_ring_ctx *ctx = req->ctx; |
2665abfd | 1761 | |
7c86ffee | 1762 | if (!(req->flags & REQ_F_COMP_LOCKED)) { |
2665abfd JA |
1763 | unsigned long flags; |
1764 | ||
2665abfd | 1765 | spin_lock_irqsave(&ctx->completion_lock, flags); |
7c86ffee | 1766 | __io_fail_links(req); |
2665abfd JA |
1767 | spin_unlock_irqrestore(&ctx->completion_lock, flags); |
1768 | } else { | |
7c86ffee | 1769 | __io_fail_links(req); |
9e645e11 JA |
1770 | } |
1771 | ||
2665abfd | 1772 | io_cqring_ev_posted(ctx); |
9e645e11 JA |
1773 | } |
1774 | ||
3fa5e0f3 | 1775 | static struct io_kiocb *__io_req_find_next(struct io_kiocb *req) |
c69f8dbe | 1776 | { |
9b0d911a | 1777 | req->flags &= ~REQ_F_LINK_HEAD; |
7c86ffee PB |
1778 | if (req->flags & REQ_F_LINK_TIMEOUT) |
1779 | io_kill_linked_timeout(req); | |
944e58bf | 1780 | |
9e645e11 JA |
1781 | /* |
1782 | * If LINK is set, we have dependent requests in this chain. If we | |
1783 | * didn't fail this request, queue the first one up, moving any other | |
1784 | * dependencies to the next request. In case of failure, fail the rest | |
1785 | * of the chain. | |
1786 | */ | |
9b5f7bd9 PB |
1787 | if (likely(!(req->flags & REQ_F_FAIL_LINK))) |
1788 | return io_req_link_next(req); | |
1789 | io_fail_links(req); | |
1790 | return NULL; | |
4d7dd462 | 1791 | } |
9e645e11 | 1792 | |
3fa5e0f3 PB |
1793 | static struct io_kiocb *io_req_find_next(struct io_kiocb *req) |
1794 | { | |
1795 | if (likely(!(req->flags & REQ_F_LINK_HEAD))) | |
1796 | return NULL; | |
1797 | return __io_req_find_next(req); | |
1798 | } | |
1799 | ||
fd7d6de2 JA |
1800 | static int io_req_task_work_add(struct io_kiocb *req, struct callback_head *cb, |
1801 | bool twa_signal_ok) | |
c2c4c83c JA |
1802 | { |
1803 | struct task_struct *tsk = req->task; | |
1804 | struct io_ring_ctx *ctx = req->ctx; | |
0ba9c9ed | 1805 | int ret, notify; |
c2c4c83c | 1806 | |
6200b0ae JA |
1807 | if (tsk->flags & PF_EXITING) |
1808 | return -ESRCH; | |
1809 | ||
c2c4c83c | 1810 | /* |
0ba9c9ed JA |
1811 | * SQPOLL kernel thread doesn't need notification, just a wakeup. For |
1812 | * all other cases, use TWA_SIGNAL unconditionally to ensure we're | |
1813 | * processing task_work. There's no reliable way to tell if TWA_RESUME | |
1814 | * will do the job. | |
c2c4c83c | 1815 | */ |
0ba9c9ed | 1816 | notify = 0; |
fd7d6de2 | 1817 | if (!(ctx->flags & IORING_SETUP_SQPOLL) && twa_signal_ok) |
c2c4c83c JA |
1818 | notify = TWA_SIGNAL; |
1819 | ||
1820 | ret = task_work_add(tsk, cb, notify); | |
1821 | if (!ret) | |
1822 | wake_up_process(tsk); | |
0ba9c9ed | 1823 | |
c2c4c83c JA |
1824 | return ret; |
1825 | } | |
1826 | ||
c40f6379 JA |
1827 | static void __io_req_task_cancel(struct io_kiocb *req, int error) |
1828 | { | |
1829 | struct io_ring_ctx *ctx = req->ctx; | |
1830 | ||
1831 | spin_lock_irq(&ctx->completion_lock); | |
1832 | io_cqring_fill_event(req, error); | |
1833 | io_commit_cqring(ctx); | |
1834 | spin_unlock_irq(&ctx->completion_lock); | |
1835 | ||
1836 | io_cqring_ev_posted(ctx); | |
1837 | req_set_fail_links(req); | |
1838 | io_double_put_req(req); | |
1839 | } | |
1840 | ||
1841 | static void io_req_task_cancel(struct callback_head *cb) | |
1842 | { | |
1843 | struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work); | |
87ceb6a6 | 1844 | struct io_ring_ctx *ctx = req->ctx; |
c40f6379 JA |
1845 | |
1846 | __io_req_task_cancel(req, -ECANCELED); | |
87ceb6a6 | 1847 | percpu_ref_put(&ctx->refs); |
c40f6379 JA |
1848 | } |
1849 | ||
1850 | static void __io_req_task_submit(struct io_kiocb *req) | |
1851 | { | |
1852 | struct io_ring_ctx *ctx = req->ctx; | |
1853 | ||
c40f6379 JA |
1854 | if (!__io_sq_thread_acquire_mm(ctx)) { |
1855 | mutex_lock(&ctx->uring_lock); | |
1856 | __io_queue_sqe(req, NULL, NULL); | |
1857 | mutex_unlock(&ctx->uring_lock); | |
1858 | } else { | |
1859 | __io_req_task_cancel(req, -EFAULT); | |
1860 | } | |
1861 | } | |
1862 | ||
1863 | static void io_req_task_submit(struct callback_head *cb) | |
1864 | { | |
1865 | struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work); | |
6d816e08 | 1866 | struct io_ring_ctx *ctx = req->ctx; |
c40f6379 JA |
1867 | |
1868 | __io_req_task_submit(req); | |
6d816e08 | 1869 | percpu_ref_put(&ctx->refs); |
c40f6379 JA |
1870 | } |
1871 | ||
1872 | static void io_req_task_queue(struct io_kiocb *req) | |
1873 | { | |
c40f6379 JA |
1874 | int ret; |
1875 | ||
1876 | init_task_work(&req->task_work, io_req_task_submit); | |
6d816e08 | 1877 | percpu_ref_get(&req->ctx->refs); |
c40f6379 | 1878 | |
fd7d6de2 | 1879 | ret = io_req_task_work_add(req, &req->task_work, true); |
c40f6379 | 1880 | if (unlikely(ret)) { |
c2c4c83c JA |
1881 | struct task_struct *tsk; |
1882 | ||
c40f6379 JA |
1883 | init_task_work(&req->task_work, io_req_task_cancel); |
1884 | tsk = io_wq_get_task(req->ctx->io_wq); | |
c2c4c83c JA |
1885 | task_work_add(tsk, &req->task_work, 0); |
1886 | wake_up_process(tsk); | |
c40f6379 | 1887 | } |
c40f6379 JA |
1888 | } |
1889 | ||
c3524383 | 1890 | static void io_queue_next(struct io_kiocb *req) |
c69f8dbe | 1891 | { |
9b5f7bd9 | 1892 | struct io_kiocb *nxt = io_req_find_next(req); |
944e58bf PB |
1893 | |
1894 | if (nxt) | |
906a8c3f | 1895 | io_req_task_queue(nxt); |
c69f8dbe JL |
1896 | } |
1897 | ||
c3524383 | 1898 | static void io_free_req(struct io_kiocb *req) |
7a743e22 | 1899 | { |
c3524383 PB |
1900 | io_queue_next(req); |
1901 | __io_free_req(req); | |
1902 | } | |
8766dd51 | 1903 | |
2d6500d4 PB |
1904 | struct req_batch { |
1905 | void *reqs[IO_IOPOLL_BATCH]; | |
1906 | int to_free; | |
7a743e22 | 1907 | |
5af1d13e PB |
1908 | struct task_struct *task; |
1909 | int task_refs; | |
2d6500d4 PB |
1910 | }; |
1911 | ||
5af1d13e PB |
1912 | static inline void io_init_req_batch(struct req_batch *rb) |
1913 | { | |
1914 | rb->to_free = 0; | |
1915 | rb->task_refs = 0; | |
1916 | rb->task = NULL; | |
1917 | } | |
1918 | ||
2d6500d4 PB |
1919 | static void __io_req_free_batch_flush(struct io_ring_ctx *ctx, |
1920 | struct req_batch *rb) | |
1921 | { | |
1922 | kmem_cache_free_bulk(req_cachep, rb->to_free, rb->reqs); | |
1923 | percpu_ref_put_many(&ctx->refs, rb->to_free); | |
1924 | rb->to_free = 0; | |
1925 | } | |
1926 | ||
1927 | static void io_req_free_batch_finish(struct io_ring_ctx *ctx, | |
1928 | struct req_batch *rb) | |
1929 | { | |
1930 | if (rb->to_free) | |
1931 | __io_req_free_batch_flush(ctx, rb); | |
5af1d13e | 1932 | if (rb->task) { |
0f212204 | 1933 | atomic_long_add(rb->task_refs, &rb->task->io_uring->req_complete); |
5af1d13e PB |
1934 | put_task_struct_many(rb->task, rb->task_refs); |
1935 | rb->task = NULL; | |
1936 | } | |
2d6500d4 PB |
1937 | } |
1938 | ||
1939 | static void io_req_free_batch(struct req_batch *rb, struct io_kiocb *req) | |
1940 | { | |
1941 | if (unlikely(io_is_fallback_req(req))) { | |
1942 | io_free_req(req); | |
1943 | return; | |
1944 | } | |
1945 | if (req->flags & REQ_F_LINK_HEAD) | |
1946 | io_queue_next(req); | |
1947 | ||
e3bc8e9d | 1948 | if (req->task != rb->task) { |
0f212204 JA |
1949 | if (rb->task) { |
1950 | atomic_long_add(rb->task_refs, &rb->task->io_uring->req_complete); | |
e3bc8e9d | 1951 | put_task_struct_many(rb->task, rb->task_refs); |
0f212204 | 1952 | } |
e3bc8e9d JA |
1953 | rb->task = req->task; |
1954 | rb->task_refs = 0; | |
5af1d13e | 1955 | } |
e3bc8e9d | 1956 | rb->task_refs++; |
5af1d13e | 1957 | |
51a4cc11 | 1958 | WARN_ON_ONCE(io_dismantle_req(req)); |
2d6500d4 PB |
1959 | rb->reqs[rb->to_free++] = req; |
1960 | if (unlikely(rb->to_free == ARRAY_SIZE(rb->reqs))) | |
1961 | __io_req_free_batch_flush(req->ctx, rb); | |
7a743e22 PB |
1962 | } |
1963 | ||
ba816ad6 JA |
1964 | /* |
1965 | * Drop reference to request, return next in chain (if there is one) if this | |
1966 | * was the last reference to this request. | |
1967 | */ | |
9b5f7bd9 | 1968 | static struct io_kiocb *io_put_req_find_next(struct io_kiocb *req) |
e65ef56d | 1969 | { |
9b5f7bd9 PB |
1970 | struct io_kiocb *nxt = NULL; |
1971 | ||
2a44f467 | 1972 | if (refcount_dec_and_test(&req->refs)) { |
9b5f7bd9 | 1973 | nxt = io_req_find_next(req); |
4d7dd462 | 1974 | __io_free_req(req); |
2a44f467 | 1975 | } |
9b5f7bd9 | 1976 | return nxt; |
2b188cc1 JA |
1977 | } |
1978 | ||
e65ef56d JA |
1979 | static void io_put_req(struct io_kiocb *req) |
1980 | { | |
1981 | if (refcount_dec_and_test(&req->refs)) | |
1982 | io_free_req(req); | |
2b188cc1 JA |
1983 | } |
1984 | ||
f4db7182 | 1985 | static struct io_wq_work *io_steal_work(struct io_kiocb *req) |
7a743e22 | 1986 | { |
6df1db6b | 1987 | struct io_kiocb *nxt; |
f4db7182 | 1988 | |
7a743e22 | 1989 | /* |
f4db7182 PB |
1990 | * A ref is owned by io-wq in which context we're. So, if that's the |
1991 | * last one, it's safe to steal next work. False negatives are Ok, | |
1992 | * it just will be re-punted async in io_put_work() | |
7a743e22 | 1993 | */ |
f4db7182 PB |
1994 | if (refcount_read(&req->refs) != 1) |
1995 | return NULL; | |
7a743e22 | 1996 | |
9b5f7bd9 | 1997 | nxt = io_req_find_next(req); |
6df1db6b | 1998 | return nxt ? &nxt->work : NULL; |
7a743e22 PB |
1999 | } |
2000 | ||
978db57e JA |
2001 | /* |
2002 | * Must only be used if we don't need to care about links, usually from | |
2003 | * within the completion handling itself. | |
2004 | */ | |
2005 | static void __io_double_put_req(struct io_kiocb *req) | |
78e19bbe JA |
2006 | { |
2007 | /* drop both submit and complete references */ | |
2008 | if (refcount_sub_and_test(2, &req->refs)) | |
2009 | __io_free_req(req); | |
2010 | } | |
2011 | ||
978db57e JA |
2012 | static void io_double_put_req(struct io_kiocb *req) |
2013 | { | |
2014 | /* drop both submit and complete references */ | |
2015 | if (refcount_sub_and_test(2, &req->refs)) | |
2016 | io_free_req(req); | |
2017 | } | |
2018 | ||
1d7bb1d5 | 2019 | static unsigned io_cqring_events(struct io_ring_ctx *ctx, bool noflush) |
a3a0e43f | 2020 | { |
84f97dc2 JA |
2021 | struct io_rings *rings = ctx->rings; |
2022 | ||
ad3eb2c8 JA |
2023 | if (test_bit(0, &ctx->cq_check_overflow)) { |
2024 | /* | |
2025 | * noflush == true is from the waitqueue handler, just ensure | |
2026 | * we wake up the task, and the next invocation will flush the | |
2027 | * entries. We cannot safely to it from here. | |
2028 | */ | |
2029 | if (noflush && !list_empty(&ctx->cq_overflow_list)) | |
2030 | return -1U; | |
1d7bb1d5 | 2031 | |
e6c8aa9a | 2032 | io_cqring_overflow_flush(ctx, false, NULL, NULL); |
ad3eb2c8 | 2033 | } |
1d7bb1d5 | 2034 | |
a3a0e43f JA |
2035 | /* See comment at the top of this file */ |
2036 | smp_rmb(); | |
ad3eb2c8 | 2037 | return ctx->cached_cq_tail - READ_ONCE(rings->cq.head); |
a3a0e43f JA |
2038 | } |
2039 | ||
fb5ccc98 PB |
2040 | static inline unsigned int io_sqring_entries(struct io_ring_ctx *ctx) |
2041 | { | |
2042 | struct io_rings *rings = ctx->rings; | |
2043 | ||
2044 | /* make sure SQ entry isn't read before tail */ | |
2045 | return smp_load_acquire(&rings->sq.tail) - ctx->cached_sq_head; | |
2046 | } | |
2047 | ||
8ff069bf | 2048 | static unsigned int io_put_kbuf(struct io_kiocb *req, struct io_buffer *kbuf) |
e94f141b | 2049 | { |
8ff069bf | 2050 | unsigned int cflags; |
e94f141b | 2051 | |
bcda7baa JA |
2052 | cflags = kbuf->bid << IORING_CQE_BUFFER_SHIFT; |
2053 | cflags |= IORING_CQE_F_BUFFER; | |
0e1b6fe3 | 2054 | req->flags &= ~REQ_F_BUFFER_SELECTED; |
bcda7baa JA |
2055 | kfree(kbuf); |
2056 | return cflags; | |
e94f141b JA |
2057 | } |
2058 | ||
8ff069bf | 2059 | static inline unsigned int io_put_rw_kbuf(struct io_kiocb *req) |
bcda7baa | 2060 | { |
4d954c25 | 2061 | struct io_buffer *kbuf; |
bcda7baa | 2062 | |
4d954c25 | 2063 | kbuf = (struct io_buffer *) (unsigned long) req->rw.addr; |
8ff069bf PB |
2064 | return io_put_kbuf(req, kbuf); |
2065 | } | |
2066 | ||
4c6e277c JA |
2067 | static inline bool io_run_task_work(void) |
2068 | { | |
6200b0ae JA |
2069 | /* |
2070 | * Not safe to run on exiting task, and the task_work handling will | |
2071 | * not add work to such a task. | |
2072 | */ | |
2073 | if (unlikely(current->flags & PF_EXITING)) | |
2074 | return false; | |
4c6e277c JA |
2075 | if (current->task_works) { |
2076 | __set_current_state(TASK_RUNNING); | |
2077 | task_work_run(); | |
2078 | return true; | |
2079 | } | |
2080 | ||
2081 | return false; | |
bcda7baa JA |
2082 | } |
2083 | ||
bbde017a XW |
2084 | static void io_iopoll_queue(struct list_head *again) |
2085 | { | |
2086 | struct io_kiocb *req; | |
2087 | ||
2088 | do { | |
d21ffe7e PB |
2089 | req = list_first_entry(again, struct io_kiocb, inflight_entry); |
2090 | list_del(&req->inflight_entry); | |
81b68a5c | 2091 | __io_complete_rw(req, -EAGAIN, 0, NULL); |
bbde017a XW |
2092 | } while (!list_empty(again)); |
2093 | } | |
2094 | ||
def596e9 JA |
2095 | /* |
2096 | * Find and free completed poll iocbs | |
2097 | */ | |
2098 | static void io_iopoll_complete(struct io_ring_ctx *ctx, unsigned int *nr_events, | |
2099 | struct list_head *done) | |
2100 | { | |
8237e045 | 2101 | struct req_batch rb; |
def596e9 | 2102 | struct io_kiocb *req; |
bbde017a XW |
2103 | LIST_HEAD(again); |
2104 | ||
2105 | /* order with ->result store in io_complete_rw_iopoll() */ | |
2106 | smp_rmb(); | |
def596e9 | 2107 | |
5af1d13e | 2108 | io_init_req_batch(&rb); |
def596e9 | 2109 | while (!list_empty(done)) { |
bcda7baa JA |
2110 | int cflags = 0; |
2111 | ||
d21ffe7e | 2112 | req = list_first_entry(done, struct io_kiocb, inflight_entry); |
bbde017a | 2113 | if (READ_ONCE(req->result) == -EAGAIN) { |
56450c20 | 2114 | req->result = 0; |
bbde017a | 2115 | req->iopoll_completed = 0; |
d21ffe7e | 2116 | list_move_tail(&req->inflight_entry, &again); |
bbde017a XW |
2117 | continue; |
2118 | } | |
d21ffe7e | 2119 | list_del(&req->inflight_entry); |
def596e9 | 2120 | |
bcda7baa | 2121 | if (req->flags & REQ_F_BUFFER_SELECTED) |
8ff069bf | 2122 | cflags = io_put_rw_kbuf(req); |
bcda7baa JA |
2123 | |
2124 | __io_cqring_fill_event(req, req->result, cflags); | |
def596e9 JA |
2125 | (*nr_events)++; |
2126 | ||
c3524383 | 2127 | if (refcount_dec_and_test(&req->refs)) |
2d6500d4 | 2128 | io_req_free_batch(&rb, req); |
def596e9 | 2129 | } |
def596e9 | 2130 | |
09bb8394 | 2131 | io_commit_cqring(ctx); |
32b2244a XW |
2132 | if (ctx->flags & IORING_SETUP_SQPOLL) |
2133 | io_cqring_ev_posted(ctx); | |
2d6500d4 | 2134 | io_req_free_batch_finish(ctx, &rb); |
581f9810 | 2135 | |
bbde017a XW |
2136 | if (!list_empty(&again)) |
2137 | io_iopoll_queue(&again); | |
581f9810 BM |
2138 | } |
2139 | ||
def596e9 JA |
2140 | static int io_do_iopoll(struct io_ring_ctx *ctx, unsigned int *nr_events, |
2141 | long min) | |
2142 | { | |
2143 | struct io_kiocb *req, *tmp; | |
2144 | LIST_HEAD(done); | |
2145 | bool spin; | |
2146 | int ret; | |
2147 | ||
2148 | /* | |
2149 | * Only spin for completions if we don't have multiple devices hanging | |
2150 | * off our complete list, and we're under the requested amount. | |
2151 | */ | |
2152 | spin = !ctx->poll_multi_file && *nr_events < min; | |
2153 | ||
2154 | ret = 0; | |
d21ffe7e | 2155 | list_for_each_entry_safe(req, tmp, &ctx->iopoll_list, inflight_entry) { |
9adbd45d | 2156 | struct kiocb *kiocb = &req->rw.kiocb; |
def596e9 JA |
2157 | |
2158 | /* | |
581f9810 BM |
2159 | * Move completed and retryable entries to our local lists. |
2160 | * If we find a request that requires polling, break out | |
2161 | * and complete those lists first, if we have entries there. | |
def596e9 | 2162 | */ |
65a6543d | 2163 | if (READ_ONCE(req->iopoll_completed)) { |
d21ffe7e | 2164 | list_move_tail(&req->inflight_entry, &done); |
def596e9 JA |
2165 | continue; |
2166 | } | |
2167 | if (!list_empty(&done)) | |
2168 | break; | |
2169 | ||
2170 | ret = kiocb->ki_filp->f_op->iopoll(kiocb, spin); | |
2171 | if (ret < 0) | |
2172 | break; | |
2173 | ||
3aadc23e PB |
2174 | /* iopoll may have completed current req */ |
2175 | if (READ_ONCE(req->iopoll_completed)) | |
d21ffe7e | 2176 | list_move_tail(&req->inflight_entry, &done); |
3aadc23e | 2177 | |
def596e9 JA |
2178 | if (ret && spin) |
2179 | spin = false; | |
2180 | ret = 0; | |
2181 | } | |
2182 | ||
2183 | if (!list_empty(&done)) | |
2184 | io_iopoll_complete(ctx, nr_events, &done); | |
2185 | ||
2186 | return ret; | |
2187 | } | |
2188 | ||
2189 | /* | |
d195a66e | 2190 | * Poll for a minimum of 'min' events. Note that if min == 0 we consider that a |
def596e9 JA |
2191 | * non-spinning poll check - we'll still enter the driver poll loop, but only |
2192 | * as a non-spinning completion check. | |
2193 | */ | |
2194 | static int io_iopoll_getevents(struct io_ring_ctx *ctx, unsigned int *nr_events, | |
2195 | long min) | |
2196 | { | |
540e32a0 | 2197 | while (!list_empty(&ctx->iopoll_list) && !need_resched()) { |
def596e9 JA |
2198 | int ret; |
2199 | ||
2200 | ret = io_do_iopoll(ctx, nr_events, min); | |
2201 | if (ret < 0) | |
2202 | return ret; | |
eba0a4dd | 2203 | if (*nr_events >= min) |
def596e9 JA |
2204 | return 0; |
2205 | } | |
2206 | ||
2207 | return 1; | |
2208 | } | |
2209 | ||
2210 | /* | |
2211 | * We can't just wait for polled events to come to us, we have to actively | |
2212 | * find and complete them. | |
2213 | */ | |
b2edc0a7 | 2214 | static void io_iopoll_try_reap_events(struct io_ring_ctx *ctx) |
def596e9 JA |
2215 | { |
2216 | if (!(ctx->flags & IORING_SETUP_IOPOLL)) | |
2217 | return; | |
2218 | ||
2219 | mutex_lock(&ctx->uring_lock); | |
540e32a0 | 2220 | while (!list_empty(&ctx->iopoll_list)) { |
def596e9 JA |
2221 | unsigned int nr_events = 0; |
2222 | ||
b2edc0a7 | 2223 | io_do_iopoll(ctx, &nr_events, 0); |
08f5439f | 2224 | |
b2edc0a7 PB |
2225 | /* let it sleep and repeat later if can't complete a request */ |
2226 | if (nr_events == 0) | |
2227 | break; | |
08f5439f JA |
2228 | /* |
2229 | * Ensure we allow local-to-the-cpu processing to take place, | |
2230 | * in this case we need to ensure that we reap all events. | |
3fcee5a6 | 2231 | * Also let task_work, etc. to progress by releasing the mutex |
08f5439f | 2232 | */ |
3fcee5a6 PB |
2233 | if (need_resched()) { |
2234 | mutex_unlock(&ctx->uring_lock); | |
2235 | cond_resched(); | |
2236 | mutex_lock(&ctx->uring_lock); | |
2237 | } | |
def596e9 JA |
2238 | } |
2239 | mutex_unlock(&ctx->uring_lock); | |
2240 | } | |
2241 | ||
7668b92a | 2242 | static int io_iopoll_check(struct io_ring_ctx *ctx, long min) |
def596e9 | 2243 | { |
7668b92a | 2244 | unsigned int nr_events = 0; |
2b2ed975 | 2245 | int iters = 0, ret = 0; |
500f9fba | 2246 | |
c7849be9 XW |
2247 | /* |
2248 | * We disallow the app entering submit/complete with polling, but we | |
2249 | * still need to lock the ring to prevent racing with polled issue | |
2250 | * that got punted to a workqueue. | |
2251 | */ | |
2252 | mutex_lock(&ctx->uring_lock); | |
def596e9 | 2253 | do { |
a3a0e43f JA |
2254 | /* |
2255 | * Don't enter poll loop if we already have events pending. | |
2256 | * If we do, we can potentially be spinning for commands that | |
2257 | * already triggered a CQE (eg in error). | |
2258 | */ | |
1d7bb1d5 | 2259 | if (io_cqring_events(ctx, false)) |
a3a0e43f JA |
2260 | break; |
2261 | ||
500f9fba JA |
2262 | /* |
2263 | * If a submit got punted to a workqueue, we can have the | |
2264 | * application entering polling for a command before it gets | |
2265 | * issued. That app will hold the uring_lock for the duration | |
2266 | * of the poll right here, so we need to take a breather every | |
2267 | * now and then to ensure that the issue has a chance to add | |
2268 | * the poll to the issued list. Otherwise we can spin here | |
2269 | * forever, while the workqueue is stuck trying to acquire the | |
2270 | * very same mutex. | |
2271 | */ | |
2272 | if (!(++iters & 7)) { | |
2273 | mutex_unlock(&ctx->uring_lock); | |
4c6e277c | 2274 | io_run_task_work(); |
500f9fba JA |
2275 | mutex_lock(&ctx->uring_lock); |
2276 | } | |
2277 | ||
7668b92a | 2278 | ret = io_iopoll_getevents(ctx, &nr_events, min); |
def596e9 JA |
2279 | if (ret <= 0) |
2280 | break; | |
2281 | ret = 0; | |
7668b92a | 2282 | } while (min && !nr_events && !need_resched()); |
def596e9 | 2283 | |
500f9fba | 2284 | mutex_unlock(&ctx->uring_lock); |
def596e9 JA |
2285 | return ret; |
2286 | } | |
2287 | ||
491381ce | 2288 | static void kiocb_end_write(struct io_kiocb *req) |
2b188cc1 | 2289 | { |
491381ce JA |
2290 | /* |
2291 | * Tell lockdep we inherited freeze protection from submission | |
2292 | * thread. | |
2293 | */ | |
2294 | if (req->flags & REQ_F_ISREG) { | |
2295 | struct inode *inode = file_inode(req->file); | |
2b188cc1 | 2296 | |
491381ce | 2297 | __sb_writers_acquired(inode->i_sb, SB_FREEZE_WRITE); |
2b188cc1 | 2298 | } |
491381ce | 2299 | file_end_write(req->file); |
2b188cc1 JA |
2300 | } |
2301 | ||
a1d7c393 JA |
2302 | static void io_complete_rw_common(struct kiocb *kiocb, long res, |
2303 | struct io_comp_state *cs) | |
2b188cc1 | 2304 | { |
9adbd45d | 2305 | struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb); |
bcda7baa | 2306 | int cflags = 0; |
2b188cc1 | 2307 | |
491381ce JA |
2308 | if (kiocb->ki_flags & IOCB_WRITE) |
2309 | kiocb_end_write(req); | |
2b188cc1 | 2310 | |
4e88d6e7 JA |
2311 | if (res != req->result) |
2312 | req_set_fail_links(req); | |
bcda7baa | 2313 | if (req->flags & REQ_F_BUFFER_SELECTED) |
8ff069bf | 2314 | cflags = io_put_rw_kbuf(req); |
a1d7c393 | 2315 | __io_req_complete(req, res, cflags, cs); |
ba816ad6 JA |
2316 | } |
2317 | ||
b63534c4 JA |
2318 | #ifdef CONFIG_BLOCK |
2319 | static bool io_resubmit_prep(struct io_kiocb *req, int error) | |
2320 | { | |
2321 | struct iovec inline_vecs[UIO_FASTIOV], *iovec = inline_vecs; | |
2322 | ssize_t ret = -ECANCELED; | |
2323 | struct iov_iter iter; | |
2324 | int rw; | |
2325 | ||
2326 | if (error) { | |
2327 | ret = error; | |
2328 | goto end_req; | |
2329 | } | |
2330 | ||
2331 | switch (req->opcode) { | |
2332 | case IORING_OP_READV: | |
2333 | case IORING_OP_READ_FIXED: | |
2334 | case IORING_OP_READ: | |
2335 | rw = READ; | |
2336 | break; | |
2337 | case IORING_OP_WRITEV: | |
2338 | case IORING_OP_WRITE_FIXED: | |
2339 | case IORING_OP_WRITE: | |
2340 | rw = WRITE; | |
2341 | break; | |
2342 | default: | |
2343 | printk_once(KERN_WARNING "io_uring: bad opcode in resubmit %d\n", | |
2344 | req->opcode); | |
2345 | goto end_req; | |
2346 | } | |
2347 | ||
8f3d7496 JA |
2348 | if (!req->io) { |
2349 | ret = io_import_iovec(rw, req, &iovec, &iter, false); | |
2350 | if (ret < 0) | |
2351 | goto end_req; | |
2352 | ret = io_setup_async_rw(req, iovec, inline_vecs, &iter, false); | |
2353 | if (!ret) | |
2354 | return true; | |
2355 | kfree(iovec); | |
2356 | } else { | |
b63534c4 | 2357 | return true; |
8f3d7496 | 2358 | } |
b63534c4 | 2359 | end_req: |
b63534c4 | 2360 | req_set_fail_links(req); |
e1e16097 | 2361 | io_req_complete(req, ret); |
b63534c4 JA |
2362 | return false; |
2363 | } | |
b63534c4 JA |
2364 | #endif |
2365 | ||
2366 | static bool io_rw_reissue(struct io_kiocb *req, long res) | |
2367 | { | |
2368 | #ifdef CONFIG_BLOCK | |
355afaeb | 2369 | umode_t mode = file_inode(req->file)->i_mode; |
b63534c4 JA |
2370 | int ret; |
2371 | ||
355afaeb JA |
2372 | if (!S_ISBLK(mode) && !S_ISREG(mode)) |
2373 | return false; | |
b63534c4 JA |
2374 | if ((res != -EAGAIN && res != -EOPNOTSUPP) || io_wq_current_is_worker()) |
2375 | return false; | |
2376 | ||
fdee946d | 2377 | ret = io_sq_thread_acquire_mm(req->ctx, req); |
6d816e08 | 2378 | |
fdee946d JA |
2379 | if (io_resubmit_prep(req, ret)) { |
2380 | refcount_inc(&req->refs); | |
2381 | io_queue_async_work(req); | |
b63534c4 | 2382 | return true; |
fdee946d JA |
2383 | } |
2384 | ||
b63534c4 JA |
2385 | #endif |
2386 | return false; | |
2387 | } | |
2388 | ||
a1d7c393 JA |
2389 | static void __io_complete_rw(struct io_kiocb *req, long res, long res2, |
2390 | struct io_comp_state *cs) | |
2391 | { | |
2392 | if (!io_rw_reissue(req, res)) | |
2393 | io_complete_rw_common(&req->rw.kiocb, res, cs); | |
ba816ad6 JA |
2394 | } |
2395 | ||
2396 | static void io_complete_rw(struct kiocb *kiocb, long res, long res2) | |
2397 | { | |
9adbd45d | 2398 | struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb); |
ba816ad6 | 2399 | |
a1d7c393 | 2400 | __io_complete_rw(req, res, res2, NULL); |
2b188cc1 JA |
2401 | } |
2402 | ||
def596e9 JA |
2403 | static void io_complete_rw_iopoll(struct kiocb *kiocb, long res, long res2) |
2404 | { | |
9adbd45d | 2405 | struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb); |
def596e9 | 2406 | |
491381ce JA |
2407 | if (kiocb->ki_flags & IOCB_WRITE) |
2408 | kiocb_end_write(req); | |
def596e9 | 2409 | |
2d7d6792 | 2410 | if (res != -EAGAIN && res != req->result) |
4e88d6e7 | 2411 | req_set_fail_links(req); |
bbde017a XW |
2412 | |
2413 | WRITE_ONCE(req->result, res); | |
2414 | /* order with io_poll_complete() checking ->result */ | |
cd664b0e PB |
2415 | smp_wmb(); |
2416 | WRITE_ONCE(req->iopoll_completed, 1); | |
def596e9 JA |
2417 | } |
2418 | ||
2419 | /* | |
2420 | * After the iocb has been issued, it's safe to be found on the poll list. | |
2421 | * Adding the kiocb to the list AFTER submission ensures that we don't | |
2422 | * find it from a io_iopoll_getevents() thread before the issuer is done | |
2423 | * accessing the kiocb cookie. | |
2424 | */ | |
2425 | static void io_iopoll_req_issued(struct io_kiocb *req) | |
2426 | { | |
2427 | struct io_ring_ctx *ctx = req->ctx; | |
2428 | ||
2429 | /* | |
2430 | * Track whether we have multiple files in our lists. This will impact | |
2431 | * how we do polling eventually, not spinning if we're on potentially | |
2432 | * different devices. | |
2433 | */ | |
540e32a0 | 2434 | if (list_empty(&ctx->iopoll_list)) { |
def596e9 JA |
2435 | ctx->poll_multi_file = false; |
2436 | } else if (!ctx->poll_multi_file) { | |
2437 | struct io_kiocb *list_req; | |
2438 | ||
540e32a0 | 2439 | list_req = list_first_entry(&ctx->iopoll_list, struct io_kiocb, |
d21ffe7e | 2440 | inflight_entry); |
9adbd45d | 2441 | if (list_req->file != req->file) |
def596e9 JA |
2442 | ctx->poll_multi_file = true; |
2443 | } | |
2444 | ||
2445 | /* | |
2446 | * For fast devices, IO may have already completed. If it has, add | |
2447 | * it to the front so we find it first. | |
2448 | */ | |
65a6543d | 2449 | if (READ_ONCE(req->iopoll_completed)) |
d21ffe7e | 2450 | list_add(&req->inflight_entry, &ctx->iopoll_list); |
def596e9 | 2451 | else |
d21ffe7e | 2452 | list_add_tail(&req->inflight_entry, &ctx->iopoll_list); |
bdcd3eab | 2453 | |
6a779382 JA |
2454 | if ((ctx->flags & IORING_SETUP_SQPOLL) && wq_has_sleeper(ctx->sqo_wait)) |
2455 | wake_up(ctx->sqo_wait); | |
def596e9 JA |
2456 | } |
2457 | ||
9f13c35b | 2458 | static void __io_state_file_put(struct io_submit_state *state) |
9a56a232 | 2459 | { |
06ef3608 PB |
2460 | if (state->has_refs) |
2461 | fput_many(state->file, state->has_refs); | |
9f13c35b PB |
2462 | state->file = NULL; |
2463 | } | |
2464 | ||
2465 | static inline void io_state_file_put(struct io_submit_state *state) | |
2466 | { | |
2467 | if (state->file) | |
2468 | __io_state_file_put(state); | |
9a56a232 JA |
2469 | } |
2470 | ||
2471 | /* | |
2472 | * Get as many references to a file as we have IOs left in this submission, | |
2473 | * assuming most submissions are for one file, or at least that each file | |
2474 | * has more than one submission. | |
2475 | */ | |
8da11c19 | 2476 | static struct file *__io_file_get(struct io_submit_state *state, int fd) |
9a56a232 JA |
2477 | { |
2478 | if (!state) | |
2479 | return fget(fd); | |
2480 | ||
2481 | if (state->file) { | |
2482 | if (state->fd == fd) { | |
06ef3608 | 2483 | state->has_refs--; |
9a56a232 JA |
2484 | state->ios_left--; |
2485 | return state->file; | |
2486 | } | |
9f13c35b | 2487 | __io_state_file_put(state); |
9a56a232 JA |
2488 | } |
2489 | state->file = fget_many(fd, state->ios_left); | |
2490 | if (!state->file) | |
2491 | return NULL; | |
2492 | ||
2493 | state->fd = fd; | |
9a56a232 | 2494 | state->ios_left--; |
06ef3608 | 2495 | state->has_refs = state->ios_left; |
9a56a232 JA |
2496 | return state->file; |
2497 | } | |
2498 | ||
4503b767 JA |
2499 | static bool io_bdev_nowait(struct block_device *bdev) |
2500 | { | |
2501 | #ifdef CONFIG_BLOCK | |
2502 | return !bdev || queue_is_mq(bdev_get_queue(bdev)); | |
2503 | #else | |
2504 | return true; | |
2505 | #endif | |
2506 | } | |
2507 | ||
2b188cc1 JA |
2508 | /* |
2509 | * If we tracked the file through the SCM inflight mechanism, we could support | |
2510 | * any file. For now, just ensure that anything potentially problematic is done | |
2511 | * inline. | |
2512 | */ | |
af197f50 | 2513 | static bool io_file_supports_async(struct file *file, int rw) |
2b188cc1 JA |
2514 | { |
2515 | umode_t mode = file_inode(file)->i_mode; | |
2516 | ||
4503b767 JA |
2517 | if (S_ISBLK(mode)) { |
2518 | if (io_bdev_nowait(file->f_inode->i_bdev)) | |
2519 | return true; | |
2520 | return false; | |
2521 | } | |
2522 | if (S_ISCHR(mode) || S_ISSOCK(mode)) | |
2b188cc1 | 2523 | return true; |
4503b767 JA |
2524 | if (S_ISREG(mode)) { |
2525 | if (io_bdev_nowait(file->f_inode->i_sb->s_bdev) && | |
2526 | file->f_op != &io_uring_fops) | |
2527 | return true; | |
2528 | return false; | |
2529 | } | |
2b188cc1 | 2530 | |
c5b85625 JA |
2531 | /* any ->read/write should understand O_NONBLOCK */ |
2532 | if (file->f_flags & O_NONBLOCK) | |
2533 | return true; | |
2534 | ||
af197f50 JA |
2535 | if (!(file->f_mode & FMODE_NOWAIT)) |
2536 | return false; | |
2537 | ||
2538 | if (rw == READ) | |
2539 | return file->f_op->read_iter != NULL; | |
2540 | ||
2541 | return file->f_op->write_iter != NULL; | |
2b188cc1 JA |
2542 | } |
2543 | ||
3529d8c2 JA |
2544 | static int io_prep_rw(struct io_kiocb *req, const struct io_uring_sqe *sqe, |
2545 | bool force_nonblock) | |
2b188cc1 | 2546 | { |
def596e9 | 2547 | struct io_ring_ctx *ctx = req->ctx; |
9adbd45d | 2548 | struct kiocb *kiocb = &req->rw.kiocb; |
09bb8394 JA |
2549 | unsigned ioprio; |
2550 | int ret; | |
2b188cc1 | 2551 | |
491381ce JA |
2552 | if (S_ISREG(file_inode(req->file)->i_mode)) |
2553 | req->flags |= REQ_F_ISREG; | |
2554 | ||
2b188cc1 | 2555 | kiocb->ki_pos = READ_ONCE(sqe->off); |
ba04291e JA |
2556 | if (kiocb->ki_pos == -1 && !(req->file->f_mode & FMODE_STREAM)) { |
2557 | req->flags |= REQ_F_CUR_POS; | |
2558 | kiocb->ki_pos = req->file->f_pos; | |
2559 | } | |
2b188cc1 | 2560 | kiocb->ki_hint = ki_hint_validate(file_write_hint(kiocb->ki_filp)); |
3e577dcd PB |
2561 | kiocb->ki_flags = iocb_flags(kiocb->ki_filp); |
2562 | ret = kiocb_set_rw_flags(kiocb, READ_ONCE(sqe->rw_flags)); | |
2563 | if (unlikely(ret)) | |
2564 | return ret; | |
2b188cc1 JA |
2565 | |
2566 | ioprio = READ_ONCE(sqe->ioprio); | |
2567 | if (ioprio) { | |
2568 | ret = ioprio_check_cap(ioprio); | |
2569 | if (ret) | |
09bb8394 | 2570 | return ret; |
2b188cc1 JA |
2571 | |
2572 | kiocb->ki_ioprio = ioprio; | |
2573 | } else | |
2574 | kiocb->ki_ioprio = get_current_ioprio(); | |
2575 | ||
8449eeda | 2576 | /* don't allow async punt if RWF_NOWAIT was requested */ |
c5b85625 | 2577 | if (kiocb->ki_flags & IOCB_NOWAIT) |
8449eeda SB |
2578 | req->flags |= REQ_F_NOWAIT; |
2579 | ||
2580 | if (force_nonblock) | |
2b188cc1 | 2581 | kiocb->ki_flags |= IOCB_NOWAIT; |
8449eeda | 2582 | |
def596e9 | 2583 | if (ctx->flags & IORING_SETUP_IOPOLL) { |
def596e9 JA |
2584 | if (!(kiocb->ki_flags & IOCB_DIRECT) || |
2585 | !kiocb->ki_filp->f_op->iopoll) | |
09bb8394 | 2586 | return -EOPNOTSUPP; |
2b188cc1 | 2587 | |
def596e9 JA |
2588 | kiocb->ki_flags |= IOCB_HIPRI; |
2589 | kiocb->ki_complete = io_complete_rw_iopoll; | |
65a6543d | 2590 | req->iopoll_completed = 0; |
def596e9 | 2591 | } else { |
09bb8394 JA |
2592 | if (kiocb->ki_flags & IOCB_HIPRI) |
2593 | return -EINVAL; | |
def596e9 JA |
2594 | kiocb->ki_complete = io_complete_rw; |
2595 | } | |
9adbd45d | 2596 | |
3529d8c2 JA |
2597 | req->rw.addr = READ_ONCE(sqe->addr); |
2598 | req->rw.len = READ_ONCE(sqe->len); | |
4f4eeba8 | 2599 | req->buf_index = READ_ONCE(sqe->buf_index); |
2b188cc1 | 2600 | return 0; |
2b188cc1 JA |
2601 | } |
2602 | ||
2603 | static inline void io_rw_done(struct kiocb *kiocb, ssize_t ret) | |
2604 | { | |
2605 | switch (ret) { | |
2606 | case -EIOCBQUEUED: | |
2607 | break; | |
2608 | case -ERESTARTSYS: | |
2609 | case -ERESTARTNOINTR: | |
2610 | case -ERESTARTNOHAND: | |
2611 | case -ERESTART_RESTARTBLOCK: | |
2612 | /* | |
2613 | * We can't just restart the syscall, since previously | |
2614 | * submitted sqes may already be in progress. Just fail this | |
2615 | * IO with EINTR. | |
2616 | */ | |
2617 | ret = -EINTR; | |
df561f66 | 2618 | fallthrough; |
2b188cc1 JA |
2619 | default: |
2620 | kiocb->ki_complete(kiocb, ret, 0); | |
2621 | } | |
2622 | } | |
2623 | ||
a1d7c393 JA |
2624 | static void kiocb_done(struct kiocb *kiocb, ssize_t ret, |
2625 | struct io_comp_state *cs) | |
ba816ad6 | 2626 | { |
ba04291e JA |
2627 | struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb); |
2628 | ||
227c0c96 JA |
2629 | /* add previously done IO, if any */ |
2630 | if (req->io && req->io->rw.bytes_done > 0) { | |
2631 | if (ret < 0) | |
2632 | ret = req->io->rw.bytes_done; | |
2633 | else | |
2634 | ret += req->io->rw.bytes_done; | |
2635 | } | |
2636 | ||
ba04291e JA |
2637 | if (req->flags & REQ_F_CUR_POS) |
2638 | req->file->f_pos = kiocb->ki_pos; | |
bcaec089 | 2639 | if (ret >= 0 && kiocb->ki_complete == io_complete_rw) |
a1d7c393 | 2640 | __io_complete_rw(req, ret, 0, cs); |
ba816ad6 JA |
2641 | else |
2642 | io_rw_done(kiocb, ret); | |
2643 | } | |
2644 | ||
9adbd45d | 2645 | static ssize_t io_import_fixed(struct io_kiocb *req, int rw, |
7d009165 | 2646 | struct iov_iter *iter) |
edafccee | 2647 | { |
9adbd45d JA |
2648 | struct io_ring_ctx *ctx = req->ctx; |
2649 | size_t len = req->rw.len; | |
edafccee | 2650 | struct io_mapped_ubuf *imu; |
4f4eeba8 | 2651 | u16 index, buf_index; |
edafccee JA |
2652 | size_t offset; |
2653 | u64 buf_addr; | |
2654 | ||
2655 | /* attempt to use fixed buffers without having provided iovecs */ | |
2656 | if (unlikely(!ctx->user_bufs)) | |
2657 | return -EFAULT; | |
2658 | ||
4f4eeba8 | 2659 | buf_index = req->buf_index; |
edafccee JA |
2660 | if (unlikely(buf_index >= ctx->nr_user_bufs)) |
2661 | return -EFAULT; | |
2662 | ||
2663 | index = array_index_nospec(buf_index, ctx->nr_user_bufs); | |
2664 | imu = &ctx->user_bufs[index]; | |
9adbd45d | 2665 | buf_addr = req->rw.addr; |
edafccee JA |
2666 | |
2667 | /* overflow */ | |
2668 | if (buf_addr + len < buf_addr) | |
2669 | return -EFAULT; | |
2670 | /* not inside the mapped region */ | |
2671 | if (buf_addr < imu->ubuf || buf_addr + len > imu->ubuf + imu->len) | |
2672 | return -EFAULT; | |
2673 | ||
2674 | /* | |
2675 | * May not be a start of buffer, set size appropriately | |
2676 | * and advance us to the beginning. | |
2677 | */ | |
2678 | offset = buf_addr - imu->ubuf; | |
2679 | iov_iter_bvec(iter, rw, imu->bvec, imu->nr_bvecs, offset + len); | |
bd11b3a3 JA |
2680 | |
2681 | if (offset) { | |
2682 | /* | |
2683 | * Don't use iov_iter_advance() here, as it's really slow for | |
2684 | * using the latter parts of a big fixed buffer - it iterates | |
2685 | * over each segment manually. We can cheat a bit here, because | |
2686 | * we know that: | |
2687 | * | |
2688 | * 1) it's a BVEC iter, we set it up | |
2689 | * 2) all bvecs are PAGE_SIZE in size, except potentially the | |
2690 | * first and last bvec | |
2691 | * | |
2692 | * So just find our index, and adjust the iterator afterwards. | |
2693 | * If the offset is within the first bvec (or the whole first | |
2694 | * bvec, just use iov_iter_advance(). This makes it easier | |
2695 | * since we can just skip the first segment, which may not | |
2696 | * be PAGE_SIZE aligned. | |
2697 | */ | |
2698 | const struct bio_vec *bvec = imu->bvec; | |
2699 | ||
2700 | if (offset <= bvec->bv_len) { | |
2701 | iov_iter_advance(iter, offset); | |
2702 | } else { | |
2703 | unsigned long seg_skip; | |
2704 | ||
2705 | /* skip first vec */ | |
2706 | offset -= bvec->bv_len; | |
2707 | seg_skip = 1 + (offset >> PAGE_SHIFT); | |
2708 | ||
2709 | iter->bvec = bvec + seg_skip; | |
2710 | iter->nr_segs -= seg_skip; | |
99c79f66 | 2711 | iter->count -= bvec->bv_len + offset; |
bd11b3a3 | 2712 | iter->iov_offset = offset & ~PAGE_MASK; |
bd11b3a3 JA |
2713 | } |
2714 | } | |
2715 | ||
5e559561 | 2716 | return len; |
edafccee JA |
2717 | } |
2718 | ||
bcda7baa JA |
2719 | static void io_ring_submit_unlock(struct io_ring_ctx *ctx, bool needs_lock) |
2720 | { | |
2721 | if (needs_lock) | |
2722 | mutex_unlock(&ctx->uring_lock); | |
2723 | } | |
2724 | ||
2725 | static void io_ring_submit_lock(struct io_ring_ctx *ctx, bool needs_lock) | |
2726 | { | |
2727 | /* | |
2728 | * "Normal" inline submissions always hold the uring_lock, since we | |
2729 | * grab it from the system call. Same is true for the SQPOLL offload. | |
2730 | * The only exception is when we've detached the request and issue it | |
2731 | * from an async worker thread, grab the lock for that case. | |
2732 | */ | |
2733 | if (needs_lock) | |
2734 | mutex_lock(&ctx->uring_lock); | |
2735 | } | |
2736 | ||
2737 | static struct io_buffer *io_buffer_select(struct io_kiocb *req, size_t *len, | |
2738 | int bgid, struct io_buffer *kbuf, | |
2739 | bool needs_lock) | |
2740 | { | |
2741 | struct io_buffer *head; | |
2742 | ||
2743 | if (req->flags & REQ_F_BUFFER_SELECTED) | |
2744 | return kbuf; | |
2745 | ||
2746 | io_ring_submit_lock(req->ctx, needs_lock); | |
2747 | ||
2748 | lockdep_assert_held(&req->ctx->uring_lock); | |
2749 | ||
2750 | head = idr_find(&req->ctx->io_buffer_idr, bgid); | |
2751 | if (head) { | |
2752 | if (!list_empty(&head->list)) { | |
2753 | kbuf = list_last_entry(&head->list, struct io_buffer, | |
2754 | list); | |
2755 | list_del(&kbuf->list); | |
2756 | } else { | |
2757 | kbuf = head; | |
2758 | idr_remove(&req->ctx->io_buffer_idr, bgid); | |
2759 | } | |
2760 | if (*len > kbuf->len) | |
2761 | *len = kbuf->len; | |
2762 | } else { | |
2763 | kbuf = ERR_PTR(-ENOBUFS); | |
2764 | } | |
2765 | ||
2766 | io_ring_submit_unlock(req->ctx, needs_lock); | |
2767 | ||
2768 | return kbuf; | |
2769 | } | |
2770 | ||
4d954c25 JA |
2771 | static void __user *io_rw_buffer_select(struct io_kiocb *req, size_t *len, |
2772 | bool needs_lock) | |
2773 | { | |
2774 | struct io_buffer *kbuf; | |
4f4eeba8 | 2775 | u16 bgid; |
4d954c25 JA |
2776 | |
2777 | kbuf = (struct io_buffer *) (unsigned long) req->rw.addr; | |
4f4eeba8 | 2778 | bgid = req->buf_index; |
4d954c25 JA |
2779 | kbuf = io_buffer_select(req, len, bgid, kbuf, needs_lock); |
2780 | if (IS_ERR(kbuf)) | |
2781 | return kbuf; | |
2782 | req->rw.addr = (u64) (unsigned long) kbuf; | |
2783 | req->flags |= REQ_F_BUFFER_SELECTED; | |
2784 | return u64_to_user_ptr(kbuf->addr); | |
2785 | } | |
2786 | ||
2787 | #ifdef CONFIG_COMPAT | |
2788 | static ssize_t io_compat_import(struct io_kiocb *req, struct iovec *iov, | |
2789 | bool needs_lock) | |
2790 | { | |
2791 | struct compat_iovec __user *uiov; | |
2792 | compat_ssize_t clen; | |
2793 | void __user *buf; | |
2794 | ssize_t len; | |
2795 | ||
2796 | uiov = u64_to_user_ptr(req->rw.addr); | |
2797 | if (!access_ok(uiov, sizeof(*uiov))) | |
2798 | return -EFAULT; | |
2799 | if (__get_user(clen, &uiov->iov_len)) | |
2800 | return -EFAULT; | |
2801 | if (clen < 0) | |
2802 | return -EINVAL; | |
2803 | ||
2804 | len = clen; | |
2805 | buf = io_rw_buffer_select(req, &len, needs_lock); | |
2806 | if (IS_ERR(buf)) | |
2807 | return PTR_ERR(buf); | |
2808 | iov[0].iov_base = buf; | |
2809 | iov[0].iov_len = (compat_size_t) len; | |
2810 | return 0; | |
2811 | } | |
2812 | #endif | |
2813 | ||
2814 | static ssize_t __io_iov_buffer_select(struct io_kiocb *req, struct iovec *iov, | |
2815 | bool needs_lock) | |
2816 | { | |
2817 | struct iovec __user *uiov = u64_to_user_ptr(req->rw.addr); | |
2818 | void __user *buf; | |
2819 | ssize_t len; | |
2820 | ||
2821 | if (copy_from_user(iov, uiov, sizeof(*uiov))) | |
2822 | return -EFAULT; | |
2823 | ||
2824 | len = iov[0].iov_len; | |
2825 | if (len < 0) | |
2826 | return -EINVAL; | |
2827 | buf = io_rw_buffer_select(req, &len, needs_lock); | |
2828 | if (IS_ERR(buf)) | |
2829 | return PTR_ERR(buf); | |
2830 | iov[0].iov_base = buf; | |
2831 | iov[0].iov_len = len; | |
2832 | return 0; | |
2833 | } | |
2834 | ||
2835 | static ssize_t io_iov_buffer_select(struct io_kiocb *req, struct iovec *iov, | |
2836 | bool needs_lock) | |
2837 | { | |
dddb3e26 JA |
2838 | if (req->flags & REQ_F_BUFFER_SELECTED) { |
2839 | struct io_buffer *kbuf; | |
2840 | ||
2841 | kbuf = (struct io_buffer *) (unsigned long) req->rw.addr; | |
2842 | iov[0].iov_base = u64_to_user_ptr(kbuf->addr); | |
2843 | iov[0].iov_len = kbuf->len; | |
4d954c25 | 2844 | return 0; |
dddb3e26 | 2845 | } |
4d954c25 JA |
2846 | if (!req->rw.len) |
2847 | return 0; | |
2848 | else if (req->rw.len > 1) | |
2849 | return -EINVAL; | |
2850 | ||
2851 | #ifdef CONFIG_COMPAT | |
2852 | if (req->ctx->compat) | |
2853 | return io_compat_import(req, iov, needs_lock); | |
2854 | #endif | |
2855 | ||
2856 | return __io_iov_buffer_select(req, iov, needs_lock); | |
2857 | } | |
2858 | ||
8452fd0c JA |
2859 | static ssize_t __io_import_iovec(int rw, struct io_kiocb *req, |
2860 | struct iovec **iovec, struct iov_iter *iter, | |
2861 | bool needs_lock) | |
2b188cc1 | 2862 | { |
9adbd45d JA |
2863 | void __user *buf = u64_to_user_ptr(req->rw.addr); |
2864 | size_t sqe_len = req->rw.len; | |
4d954c25 | 2865 | ssize_t ret; |
edafccee JA |
2866 | u8 opcode; |
2867 | ||
d625c6ee | 2868 | opcode = req->opcode; |
7d009165 | 2869 | if (opcode == IORING_OP_READ_FIXED || opcode == IORING_OP_WRITE_FIXED) { |
edafccee | 2870 | *iovec = NULL; |
9adbd45d | 2871 | return io_import_fixed(req, rw, iter); |
edafccee | 2872 | } |
2b188cc1 | 2873 | |
bcda7baa | 2874 | /* buffer index only valid with fixed read/write, or buffer select */ |
4f4eeba8 | 2875 | if (req->buf_index && !(req->flags & REQ_F_BUFFER_SELECT)) |
9adbd45d JA |
2876 | return -EINVAL; |
2877 | ||
3a6820f2 | 2878 | if (opcode == IORING_OP_READ || opcode == IORING_OP_WRITE) { |
bcda7baa | 2879 | if (req->flags & REQ_F_BUFFER_SELECT) { |
4d954c25 | 2880 | buf = io_rw_buffer_select(req, &sqe_len, needs_lock); |
867a23ea | 2881 | if (IS_ERR(buf)) |
4d954c25 | 2882 | return PTR_ERR(buf); |
3f9d6441 | 2883 | req->rw.len = sqe_len; |
bcda7baa JA |
2884 | } |
2885 | ||
3a6820f2 JA |
2886 | ret = import_single_range(rw, buf, sqe_len, *iovec, iter); |
2887 | *iovec = NULL; | |
3a901598 | 2888 | return ret < 0 ? ret : sqe_len; |
3a6820f2 JA |
2889 | } |
2890 | ||
4d954c25 JA |
2891 | if (req->flags & REQ_F_BUFFER_SELECT) { |
2892 | ret = io_iov_buffer_select(req, *iovec, needs_lock); | |
3f9d6441 JA |
2893 | if (!ret) { |
2894 | ret = (*iovec)->iov_len; | |
2895 | iov_iter_init(iter, rw, *iovec, 1, ret); | |
2896 | } | |
4d954c25 JA |
2897 | *iovec = NULL; |
2898 | return ret; | |
2899 | } | |
2900 | ||
2b188cc1 | 2901 | #ifdef CONFIG_COMPAT |
cf6fd4bd | 2902 | if (req->ctx->compat) |
2b188cc1 JA |
2903 | return compat_import_iovec(rw, buf, sqe_len, UIO_FASTIOV, |
2904 | iovec, iter); | |
2905 | #endif | |
2906 | ||
2907 | return import_iovec(rw, buf, sqe_len, UIO_FASTIOV, iovec, iter); | |
2908 | } | |
2909 | ||
8452fd0c JA |
2910 | static ssize_t io_import_iovec(int rw, struct io_kiocb *req, |
2911 | struct iovec **iovec, struct iov_iter *iter, | |
2912 | bool needs_lock) | |
2913 | { | |
2914 | if (!req->io) | |
2915 | return __io_import_iovec(rw, req, iovec, iter, needs_lock); | |
2916 | *iovec = NULL; | |
2917 | return iov_iter_count(&req->io->rw.iter); | |
2918 | } | |
2919 | ||
0fef9483 JA |
2920 | static inline loff_t *io_kiocb_ppos(struct kiocb *kiocb) |
2921 | { | |
2922 | return kiocb->ki_filp->f_mode & FMODE_STREAM ? NULL : &kiocb->ki_pos; | |
2923 | } | |
2924 | ||
31b51510 | 2925 | /* |
32960613 JA |
2926 | * For files that don't have ->read_iter() and ->write_iter(), handle them |
2927 | * by looping over ->read() or ->write() manually. | |
31b51510 | 2928 | */ |
32960613 JA |
2929 | static ssize_t loop_rw_iter(int rw, struct file *file, struct kiocb *kiocb, |
2930 | struct iov_iter *iter) | |
2931 | { | |
2932 | ssize_t ret = 0; | |
2933 | ||
2934 | /* | |
2935 | * Don't support polled IO through this interface, and we can't | |
2936 | * support non-blocking either. For the latter, this just causes | |
2937 | * the kiocb to be handled from an async context. | |
2938 | */ | |
2939 | if (kiocb->ki_flags & IOCB_HIPRI) | |
2940 | return -EOPNOTSUPP; | |
2941 | if (kiocb->ki_flags & IOCB_NOWAIT) | |
2942 | return -EAGAIN; | |
2943 | ||
2944 | while (iov_iter_count(iter)) { | |
311ae9e1 | 2945 | struct iovec iovec; |
32960613 JA |
2946 | ssize_t nr; |
2947 | ||
311ae9e1 PB |
2948 | if (!iov_iter_is_bvec(iter)) { |
2949 | iovec = iov_iter_iovec(iter); | |
2950 | } else { | |
2951 | /* fixed buffers import bvec */ | |
2952 | iovec.iov_base = kmap(iter->bvec->bv_page) | |
2953 | + iter->iov_offset; | |
2954 | iovec.iov_len = min(iter->count, | |
2955 | iter->bvec->bv_len - iter->iov_offset); | |
2956 | } | |
2957 | ||
32960613 JA |
2958 | if (rw == READ) { |
2959 | nr = file->f_op->read(file, iovec.iov_base, | |
0fef9483 | 2960 | iovec.iov_len, io_kiocb_ppos(kiocb)); |
32960613 JA |
2961 | } else { |
2962 | nr = file->f_op->write(file, iovec.iov_base, | |
0fef9483 | 2963 | iovec.iov_len, io_kiocb_ppos(kiocb)); |
32960613 JA |
2964 | } |
2965 | ||
311ae9e1 PB |
2966 | if (iov_iter_is_bvec(iter)) |
2967 | kunmap(iter->bvec->bv_page); | |
2968 | ||
32960613 JA |
2969 | if (nr < 0) { |
2970 | if (!ret) | |
2971 | ret = nr; | |
2972 | break; | |
2973 | } | |
2974 | ret += nr; | |
2975 | if (nr != iovec.iov_len) | |
2976 | break; | |
2977 | iov_iter_advance(iter, nr); | |
2978 | } | |
2979 | ||
2980 | return ret; | |
2981 | } | |
2982 | ||
ff6165b2 JA |
2983 | static void io_req_map_rw(struct io_kiocb *req, const struct iovec *iovec, |
2984 | const struct iovec *fast_iov, struct iov_iter *iter) | |
f67676d1 | 2985 | { |
b64e3444 PB |
2986 | struct io_async_rw *rw = &req->io->rw; |
2987 | ||
ff6165b2 JA |
2988 | memcpy(&rw->iter, iter, sizeof(*iter)); |
2989 | rw->free_iovec = NULL; | |
227c0c96 | 2990 | rw->bytes_done = 0; |
ff6165b2 JA |
2991 | /* can only be fixed buffers, no need to do anything */ |
2992 | if (iter->type == ITER_BVEC) | |
2993 | return; | |
b64e3444 | 2994 | if (!iovec) { |
ff6165b2 JA |
2995 | unsigned iov_off = 0; |
2996 | ||
2997 | rw->iter.iov = rw->fast_iov; | |
2998 | if (iter->iov != fast_iov) { | |
2999 | iov_off = iter->iov - fast_iov; | |
3000 | rw->iter.iov += iov_off; | |
3001 | } | |
3002 | if (rw->fast_iov != fast_iov) | |
3003 | memcpy(rw->fast_iov + iov_off, fast_iov + iov_off, | |
45097dae | 3004 | sizeof(struct iovec) * iter->nr_segs); |
99bc4c38 | 3005 | } else { |
ff6165b2 | 3006 | rw->free_iovec = iovec; |
99bc4c38 | 3007 | req->flags |= REQ_F_NEED_CLEANUP; |
f67676d1 JA |
3008 | } |
3009 | } | |
3010 | ||
3d9932a8 XW |
3011 | static inline int __io_alloc_async_ctx(struct io_kiocb *req) |
3012 | { | |
3013 | req->io = kmalloc(sizeof(*req->io), GFP_KERNEL); | |
3014 | return req->io == NULL; | |
3015 | } | |
3016 | ||
b7bb4f7d | 3017 | static int io_alloc_async_ctx(struct io_kiocb *req) |
f67676d1 | 3018 | { |
d3656344 JA |
3019 | if (!io_op_defs[req->opcode].async_ctx) |
3020 | return 0; | |
3d9932a8 XW |
3021 | |
3022 | return __io_alloc_async_ctx(req); | |
b7bb4f7d JA |
3023 | } |
3024 | ||
ff6165b2 JA |
3025 | static int io_setup_async_rw(struct io_kiocb *req, const struct iovec *iovec, |
3026 | const struct iovec *fast_iov, | |
227c0c96 | 3027 | struct iov_iter *iter, bool force) |
b7bb4f7d | 3028 | { |
227c0c96 | 3029 | if (!force && !io_op_defs[req->opcode].async_ctx) |
74566df3 | 3030 | return 0; |
5d204bcf | 3031 | if (!req->io) { |
3d9932a8 | 3032 | if (__io_alloc_async_ctx(req)) |
5d204bcf | 3033 | return -ENOMEM; |
b7bb4f7d | 3034 | |
ff6165b2 | 3035 | io_req_map_rw(req, iovec, fast_iov, iter); |
5d204bcf | 3036 | } |
b7bb4f7d | 3037 | return 0; |
f67676d1 JA |
3038 | } |
3039 | ||
c3e330a4 PB |
3040 | static inline int io_rw_prep_async(struct io_kiocb *req, int rw, |
3041 | bool force_nonblock) | |
3042 | { | |
ff6165b2 | 3043 | struct io_async_rw *iorw = &req->io->rw; |
c183edff | 3044 | struct iovec *iov; |
c3e330a4 PB |
3045 | ssize_t ret; |
3046 | ||
c183edff JA |
3047 | iorw->iter.iov = iov = iorw->fast_iov; |
3048 | ret = __io_import_iovec(rw, req, &iov, &iorw->iter, !force_nonblock); | |
c3e330a4 PB |
3049 | if (unlikely(ret < 0)) |
3050 | return ret; | |
3051 | ||
c183edff | 3052 | iorw->iter.iov = iov; |
ff6165b2 | 3053 | io_req_map_rw(req, iorw->iter.iov, iorw->fast_iov, &iorw->iter); |
c3e330a4 PB |
3054 | return 0; |
3055 | } | |
3056 | ||
3529d8c2 JA |
3057 | static int io_read_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe, |
3058 | bool force_nonblock) | |
f67676d1 JA |
3059 | { |
3060 | ssize_t ret; | |
3061 | ||
3529d8c2 JA |
3062 | ret = io_prep_rw(req, sqe, force_nonblock); |
3063 | if (ret) | |
3064 | return ret; | |
f67676d1 | 3065 | |
3529d8c2 JA |
3066 | if (unlikely(!(req->file->f_mode & FMODE_READ))) |
3067 | return -EBADF; | |
f67676d1 | 3068 | |
5f798bea PB |
3069 | /* either don't need iovec imported or already have it */ |
3070 | if (!req->io || req->flags & REQ_F_NEED_CLEANUP) | |
3529d8c2 | 3071 | return 0; |
c3e330a4 | 3072 | return io_rw_prep_async(req, READ, force_nonblock); |
f67676d1 JA |
3073 | } |
3074 | ||
c1dd91d1 JA |
3075 | /* |
3076 | * This is our waitqueue callback handler, registered through lock_page_async() | |
3077 | * when we initially tried to do the IO with the iocb armed our waitqueue. | |
3078 | * This gets called when the page is unlocked, and we generally expect that to | |
3079 | * happen when the page IO is completed and the page is now uptodate. This will | |
3080 | * queue a task_work based retry of the operation, attempting to copy the data | |
3081 | * again. If the latter fails because the page was NOT uptodate, then we will | |
3082 | * do a thread based blocking retry of the operation. That's the unexpected | |
3083 | * slow path. | |
3084 | */ | |
bcf5a063 JA |
3085 | static int io_async_buf_func(struct wait_queue_entry *wait, unsigned mode, |
3086 | int sync, void *arg) | |
3087 | { | |
3088 | struct wait_page_queue *wpq; | |
3089 | struct io_kiocb *req = wait->private; | |
bcf5a063 | 3090 | struct wait_page_key *key = arg; |
bcf5a063 JA |
3091 | int ret; |
3092 | ||
3093 | wpq = container_of(wait, struct wait_page_queue, wait); | |
3094 | ||
cdc8fcb4 LT |
3095 | if (!wake_page_match(wpq, key)) |
3096 | return 0; | |
3097 | ||
c8d317aa | 3098 | req->rw.kiocb.ki_flags &= ~IOCB_WAITQ; |
bcf5a063 JA |
3099 | list_del_init(&wait->entry); |
3100 | ||
e7375122 | 3101 | init_task_work(&req->task_work, io_req_task_submit); |
6d816e08 JA |
3102 | percpu_ref_get(&req->ctx->refs); |
3103 | ||
bcf5a063 JA |
3104 | /* submit ref gets dropped, acquire a new one */ |
3105 | refcount_inc(&req->refs); | |
fd7d6de2 | 3106 | ret = io_req_task_work_add(req, &req->task_work, true); |
bcf5a063 | 3107 | if (unlikely(ret)) { |
c2c4c83c JA |
3108 | struct task_struct *tsk; |
3109 | ||
bcf5a063 | 3110 | /* queue just for cancelation */ |
e7375122 | 3111 | init_task_work(&req->task_work, io_req_task_cancel); |
bcf5a063 | 3112 | tsk = io_wq_get_task(req->ctx->io_wq); |
e7375122 | 3113 | task_work_add(tsk, &req->task_work, 0); |
c2c4c83c | 3114 | wake_up_process(tsk); |
bcf5a063 | 3115 | } |
bcf5a063 JA |
3116 | return 1; |
3117 | } | |
3118 | ||
c1dd91d1 JA |
3119 | /* |
3120 | * This controls whether a given IO request should be armed for async page | |
3121 | * based retry. If we return false here, the request is handed to the async | |
3122 | * worker threads for retry. If we're doing buffered reads on a regular file, | |
3123 | * we prepare a private wait_page_queue entry and retry the operation. This | |
3124 | * will either succeed because the page is now uptodate and unlocked, or it | |
3125 | * will register a callback when the page is unlocked at IO completion. Through | |
3126 | * that callback, io_uring uses task_work to setup a retry of the operation. | |
3127 | * That retry will attempt the buffered read again. The retry will generally | |
3128 | * succeed, or in rare cases where it fails, we then fall back to using the | |
3129 | * async worker threads for a blocking retry. | |
3130 | */ | |
227c0c96 | 3131 | static bool io_rw_should_retry(struct io_kiocb *req) |
f67676d1 | 3132 | { |
3b2a4439 | 3133 | struct wait_page_queue *wait = &req->io->rw.wpq; |
bcf5a063 | 3134 | struct kiocb *kiocb = &req->rw.kiocb; |
f67676d1 | 3135 | |
bcf5a063 JA |
3136 | /* never retry for NOWAIT, we just complete with -EAGAIN */ |
3137 | if (req->flags & REQ_F_NOWAIT) | |
3138 | return false; | |
f67676d1 | 3139 | |
227c0c96 | 3140 | /* Only for buffered IO */ |
3b2a4439 | 3141 | if (kiocb->ki_flags & (IOCB_DIRECT | IOCB_HIPRI)) |
bcf5a063 | 3142 | return false; |
3b2a4439 | 3143 | |
bcf5a063 JA |
3144 | /* |
3145 | * just use poll if we can, and don't attempt if the fs doesn't | |
3146 | * support callback based unlocks | |
3147 | */ | |
3148 | if (file_can_poll(req->file) || !(req->file->f_mode & FMODE_BUF_RASYNC)) | |
3149 | return false; | |
f67676d1 | 3150 | |
3b2a4439 JA |
3151 | wait->wait.func = io_async_buf_func; |
3152 | wait->wait.private = req; | |
3153 | wait->wait.flags = 0; | |
3154 | INIT_LIST_HEAD(&wait->wait.entry); | |
3155 | kiocb->ki_flags |= IOCB_WAITQ; | |
c8d317aa | 3156 | kiocb->ki_flags &= ~IOCB_NOWAIT; |
3b2a4439 | 3157 | kiocb->ki_waitq = wait; |
3b2a4439 | 3158 | return true; |
bcf5a063 JA |
3159 | } |
3160 | ||
3161 | static int io_iter_do_read(struct io_kiocb *req, struct iov_iter *iter) | |
3162 | { | |
3163 | if (req->file->f_op->read_iter) | |
3164 | return call_read_iter(req->file, &req->rw.kiocb, iter); | |
2dd2111d GH |
3165 | else if (req->file->f_op->read) |
3166 | return loop_rw_iter(READ, req->file, &req->rw.kiocb, iter); | |
3167 | else | |
3168 | return -EINVAL; | |
f67676d1 JA |
3169 | } |
3170 | ||
a1d7c393 JA |
3171 | static int io_read(struct io_kiocb *req, bool force_nonblock, |
3172 | struct io_comp_state *cs) | |
2b188cc1 JA |
3173 | { |
3174 | struct iovec inline_vecs[UIO_FASTIOV], *iovec = inline_vecs; | |
9adbd45d | 3175 | struct kiocb *kiocb = &req->rw.kiocb; |
ff6165b2 | 3176 | struct iov_iter __iter, *iter = &__iter; |
227c0c96 | 3177 | ssize_t io_size, ret, ret2; |
31b51510 | 3178 | size_t iov_count; |
f5cac8b1 | 3179 | bool no_async; |
ff6165b2 JA |
3180 | |
3181 | if (req->io) | |
3182 | iter = &req->io->rw.iter; | |
2b188cc1 | 3183 | |
ff6165b2 | 3184 | ret = io_import_iovec(READ, req, &iovec, iter, !force_nonblock); |
06b76d44 JA |
3185 | if (ret < 0) |
3186 | return ret; | |
eefdf30f | 3187 | iov_count = iov_iter_count(iter); |
fa15bafb PB |
3188 | io_size = ret; |
3189 | req->result = io_size; | |
227c0c96 | 3190 | ret = 0; |
2b188cc1 | 3191 | |
fd6c2e4c JA |
3192 | /* Ensure we clear previously set non-block flag */ |
3193 | if (!force_nonblock) | |
29de5f6a | 3194 | kiocb->ki_flags &= ~IOCB_NOWAIT; |
fd6c2e4c | 3195 | |
24c74678 | 3196 | /* If the file doesn't support async, just async punt */ |
f5cac8b1 JA |
3197 | no_async = force_nonblock && !io_file_supports_async(req->file, READ); |
3198 | if (no_async) | |
f67676d1 | 3199 | goto copy_iov; |
9e645e11 | 3200 | |
0fef9483 | 3201 | ret = rw_verify_area(READ, req->file, io_kiocb_ppos(kiocb), iov_count); |
fa15bafb PB |
3202 | if (unlikely(ret)) |
3203 | goto out_free; | |
2b188cc1 | 3204 | |
227c0c96 | 3205 | ret = io_iter_do_read(req, iter); |
32960613 | 3206 | |
227c0c96 JA |
3207 | if (!ret) { |
3208 | goto done; | |
3209 | } else if (ret == -EIOCBQUEUED) { | |
3210 | ret = 0; | |
3211 | goto out_free; | |
3212 | } else if (ret == -EAGAIN) { | |
eefdf30f JA |
3213 | /* IOPOLL retry should happen for io-wq threads */ |
3214 | if (!force_nonblock && !(req->ctx->flags & IORING_SETUP_IOPOLL)) | |
f91daf56 | 3215 | goto done; |
355afaeb JA |
3216 | /* no retry on NONBLOCK marked file */ |
3217 | if (req->file->f_flags & O_NONBLOCK) | |
3218 | goto done; | |
84216315 JA |
3219 | /* some cases will consume bytes even on error returns */ |
3220 | iov_iter_revert(iter, iov_count - iov_iter_count(iter)); | |
f38c7e3a JA |
3221 | ret = 0; |
3222 | goto copy_iov; | |
227c0c96 | 3223 | } else if (ret < 0) { |
00d23d51 JA |
3224 | /* make sure -ERESTARTSYS -> -EINTR is done */ |
3225 | goto done; | |
227c0c96 JA |
3226 | } |
3227 | ||
3228 | /* read it all, or we did blocking attempt. no retry. */ | |
f91daf56 JA |
3229 | if (!iov_iter_count(iter) || !force_nonblock || |
3230 | (req->file->f_flags & O_NONBLOCK)) | |
227c0c96 JA |
3231 | goto done; |
3232 | ||
3233 | io_size -= ret; | |
3234 | copy_iov: | |
3235 | ret2 = io_setup_async_rw(req, iovec, inline_vecs, iter, true); | |
3236 | if (ret2) { | |
3237 | ret = ret2; | |
3238 | goto out_free; | |
3239 | } | |
f5cac8b1 JA |
3240 | if (no_async) |
3241 | return -EAGAIN; | |
227c0c96 JA |
3242 | /* it's copied and will be cleaned with ->io */ |
3243 | iovec = NULL; | |
3244 | /* now use our persistent iterator, if we aren't already */ | |
3245 | iter = &req->io->rw.iter; | |
3246 | retry: | |
3247 | req->io->rw.bytes_done += ret; | |
3248 | /* if we can retry, do so with the callbacks armed */ | |
3249 | if (!io_rw_should_retry(req)) { | |
fa15bafb PB |
3250 | kiocb->ki_flags &= ~IOCB_WAITQ; |
3251 | return -EAGAIN; | |
2b188cc1 | 3252 | } |
227c0c96 JA |
3253 | |
3254 | /* | |
3255 | * Now retry read with the IOCB_WAITQ parts set in the iocb. If we | |
3256 | * get -EIOCBQUEUED, then we'll get a notification when the desired | |
3257 | * page gets unlocked. We can also get a partial read here, and if we | |
3258 | * do, then just retry at the new offset. | |
3259 | */ | |
3260 | ret = io_iter_do_read(req, iter); | |
3261 | if (ret == -EIOCBQUEUED) { | |
3262 | ret = 0; | |
3263 | goto out_free; | |
3264 | } else if (ret > 0 && ret < io_size) { | |
3265 | /* we got some bytes, but not all. retry. */ | |
3266 | goto retry; | |
3267 | } | |
3268 | done: | |
3269 | kiocb_done(kiocb, ret, cs); | |
3270 | ret = 0; | |
f67676d1 | 3271 | out_free: |
f261c168 | 3272 | /* it's reportedly faster than delegating the null check to kfree() */ |
252917c3 | 3273 | if (iovec) |
6f2cc166 | 3274 | kfree(iovec); |
2b188cc1 JA |
3275 | return ret; |
3276 | } | |
3277 | ||
3529d8c2 JA |
3278 | static int io_write_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe, |
3279 | bool force_nonblock) | |
f67676d1 JA |
3280 | { |
3281 | ssize_t ret; | |
3282 | ||
3529d8c2 JA |
3283 | ret = io_prep_rw(req, sqe, force_nonblock); |
3284 | if (ret) | |
3285 | return ret; | |
f67676d1 | 3286 | |
3529d8c2 JA |
3287 | if (unlikely(!(req->file->f_mode & FMODE_WRITE))) |
3288 | return -EBADF; | |
f67676d1 | 3289 | |
5f798bea PB |
3290 | /* either don't need iovec imported or already have it */ |
3291 | if (!req->io || req->flags & REQ_F_NEED_CLEANUP) | |
3529d8c2 | 3292 | return 0; |
c3e330a4 | 3293 | return io_rw_prep_async(req, WRITE, force_nonblock); |
f67676d1 JA |
3294 | } |
3295 | ||
a1d7c393 JA |
3296 | static int io_write(struct io_kiocb *req, bool force_nonblock, |
3297 | struct io_comp_state *cs) | |
2b188cc1 JA |
3298 | { |
3299 | struct iovec inline_vecs[UIO_FASTIOV], *iovec = inline_vecs; | |
9adbd45d | 3300 | struct kiocb *kiocb = &req->rw.kiocb; |
ff6165b2 | 3301 | struct iov_iter __iter, *iter = &__iter; |
31b51510 | 3302 | size_t iov_count; |
fa15bafb | 3303 | ssize_t ret, ret2, io_size; |
2b188cc1 | 3304 | |
ff6165b2 JA |
3305 | if (req->io) |
3306 | iter = &req->io->rw.iter; | |
3307 | ||
3308 | ret = io_import_iovec(WRITE, req, &iovec, iter, !force_nonblock); | |
06b76d44 JA |
3309 | if (ret < 0) |
3310 | return ret; | |
eefdf30f | 3311 | iov_count = iov_iter_count(iter); |
fa15bafb PB |
3312 | io_size = ret; |
3313 | req->result = io_size; | |
2b188cc1 | 3314 | |
fd6c2e4c JA |
3315 | /* Ensure we clear previously set non-block flag */ |
3316 | if (!force_nonblock) | |
9adbd45d | 3317 | req->rw.kiocb.ki_flags &= ~IOCB_NOWAIT; |
fd6c2e4c | 3318 | |
24c74678 | 3319 | /* If the file doesn't support async, just async punt */ |
af197f50 | 3320 | if (force_nonblock && !io_file_supports_async(req->file, WRITE)) |
f67676d1 | 3321 | goto copy_iov; |
31b51510 | 3322 | |
10d59345 JA |
3323 | /* file path doesn't support NOWAIT for non-direct_IO */ |
3324 | if (force_nonblock && !(kiocb->ki_flags & IOCB_DIRECT) && | |
3325 | (req->flags & REQ_F_ISREG)) | |
f67676d1 | 3326 | goto copy_iov; |
31b51510 | 3327 | |
0fef9483 | 3328 | ret = rw_verify_area(WRITE, req->file, io_kiocb_ppos(kiocb), iov_count); |
fa15bafb PB |
3329 | if (unlikely(ret)) |
3330 | goto out_free; | |
4ed734b0 | 3331 | |
fa15bafb PB |
3332 | /* |
3333 | * Open-code file_start_write here to grab freeze protection, | |
3334 | * which will be released by another thread in | |
3335 | * io_complete_rw(). Fool lockdep by telling it the lock got | |
3336 | * released so that it doesn't complain about the held lock when | |
3337 | * we return to userspace. | |
3338 | */ | |
3339 | if (req->flags & REQ_F_ISREG) { | |
3340 | __sb_start_write(file_inode(req->file)->i_sb, | |
3341 | SB_FREEZE_WRITE, true); | |
3342 | __sb_writers_release(file_inode(req->file)->i_sb, | |
3343 | SB_FREEZE_WRITE); | |
3344 | } | |
3345 | kiocb->ki_flags |= IOCB_WRITE; | |
4ed734b0 | 3346 | |
fa15bafb | 3347 | if (req->file->f_op->write_iter) |
ff6165b2 | 3348 | ret2 = call_write_iter(req->file, kiocb, iter); |
2dd2111d | 3349 | else if (req->file->f_op->write) |
ff6165b2 | 3350 | ret2 = loop_rw_iter(WRITE, req->file, kiocb, iter); |
2dd2111d GH |
3351 | else |
3352 | ret2 = -EINVAL; | |
4ed734b0 | 3353 | |
fa15bafb PB |
3354 | /* |
3355 | * Raw bdev writes will return -EOPNOTSUPP for IOCB_NOWAIT. Just | |
3356 | * retry them without IOCB_NOWAIT. | |
3357 | */ | |
3358 | if (ret2 == -EOPNOTSUPP && (kiocb->ki_flags & IOCB_NOWAIT)) | |
3359 | ret2 = -EAGAIN; | |
355afaeb JA |
3360 | /* no retry on NONBLOCK marked file */ |
3361 | if (ret2 == -EAGAIN && (req->file->f_flags & O_NONBLOCK)) | |
3362 | goto done; | |
fa15bafb | 3363 | if (!force_nonblock || ret2 != -EAGAIN) { |
eefdf30f JA |
3364 | /* IOPOLL retry should happen for io-wq threads */ |
3365 | if ((req->ctx->flags & IORING_SETUP_IOPOLL) && ret2 == -EAGAIN) | |
3366 | goto copy_iov; | |
355afaeb | 3367 | done: |
fa15bafb PB |
3368 | kiocb_done(kiocb, ret2, cs); |
3369 | } else { | |
f67676d1 | 3370 | copy_iov: |
84216315 JA |
3371 | /* some cases will consume bytes even on error returns */ |
3372 | iov_iter_revert(iter, iov_count - iov_iter_count(iter)); | |
227c0c96 | 3373 | ret = io_setup_async_rw(req, iovec, inline_vecs, iter, false); |
ff6165b2 JA |
3374 | if (!ret) |
3375 | return -EAGAIN; | |
2b188cc1 | 3376 | } |
31b51510 | 3377 | out_free: |
f261c168 | 3378 | /* it's reportedly faster than delegating the null check to kfree() */ |
252917c3 | 3379 | if (iovec) |
6f2cc166 | 3380 | kfree(iovec); |
2b188cc1 JA |
3381 | return ret; |
3382 | } | |
3383 | ||
f2a8d5c7 PB |
3384 | static int __io_splice_prep(struct io_kiocb *req, |
3385 | const struct io_uring_sqe *sqe) | |
7d67af2c PB |
3386 | { |
3387 | struct io_splice* sp = &req->splice; | |
3388 | unsigned int valid_flags = SPLICE_F_FD_IN_FIXED | SPLICE_F_ALL; | |
3389 | int ret; | |
3390 | ||
3391 | if (req->flags & REQ_F_NEED_CLEANUP) | |
3392 | return 0; | |
3232dd02 PB |
3393 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
3394 | return -EINVAL; | |
7d67af2c PB |
3395 | |
3396 | sp->file_in = NULL; | |
7d67af2c PB |
3397 | sp->len = READ_ONCE(sqe->len); |
3398 | sp->flags = READ_ONCE(sqe->splice_flags); | |
3399 | ||
3400 | if (unlikely(sp->flags & ~valid_flags)) | |
3401 | return -EINVAL; | |
3402 | ||
3403 | ret = io_file_get(NULL, req, READ_ONCE(sqe->splice_fd_in), &sp->file_in, | |
3404 | (sp->flags & SPLICE_F_FD_IN_FIXED)); | |
3405 | if (ret) | |
3406 | return ret; | |
3407 | req->flags |= REQ_F_NEED_CLEANUP; | |
3408 | ||
7cdaf587 XW |
3409 | if (!S_ISREG(file_inode(sp->file_in)->i_mode)) { |
3410 | /* | |
3411 | * Splice operation will be punted aync, and here need to | |
3412 | * modify io_wq_work.flags, so initialize io_wq_work firstly. | |
3413 | */ | |
3414 | io_req_init_async(req); | |
7d67af2c | 3415 | req->work.flags |= IO_WQ_WORK_UNBOUND; |
7cdaf587 | 3416 | } |
7d67af2c PB |
3417 | |
3418 | return 0; | |
3419 | } | |
3420 | ||
f2a8d5c7 PB |
3421 | static int io_tee_prep(struct io_kiocb *req, |
3422 | const struct io_uring_sqe *sqe) | |
3423 | { | |
3424 | if (READ_ONCE(sqe->splice_off_in) || READ_ONCE(sqe->off)) | |
3425 | return -EINVAL; | |
3426 | return __io_splice_prep(req, sqe); | |
3427 | } | |
3428 | ||
3429 | static int io_tee(struct io_kiocb *req, bool force_nonblock) | |
3430 | { | |
3431 | struct io_splice *sp = &req->splice; | |
3432 | struct file *in = sp->file_in; | |
3433 | struct file *out = sp->file_out; | |
3434 | unsigned int flags = sp->flags & ~SPLICE_F_FD_IN_FIXED; | |
3435 | long ret = 0; | |
3436 | ||
3437 | if (force_nonblock) | |
3438 | return -EAGAIN; | |
3439 | if (sp->len) | |
3440 | ret = do_tee(in, out, sp->len, flags); | |
3441 | ||
3442 | io_put_file(req, in, (sp->flags & SPLICE_F_FD_IN_FIXED)); | |
3443 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
3444 | ||
f2a8d5c7 PB |
3445 | if (ret != sp->len) |
3446 | req_set_fail_links(req); | |
e1e16097 | 3447 | io_req_complete(req, ret); |
f2a8d5c7 PB |
3448 | return 0; |
3449 | } | |
3450 | ||
3451 | static int io_splice_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) | |
3452 | { | |
3453 | struct io_splice* sp = &req->splice; | |
3454 | ||
3455 | sp->off_in = READ_ONCE(sqe->splice_off_in); | |
3456 | sp->off_out = READ_ONCE(sqe->off); | |
3457 | return __io_splice_prep(req, sqe); | |
3458 | } | |
3459 | ||
014db007 | 3460 | static int io_splice(struct io_kiocb *req, bool force_nonblock) |
7d67af2c PB |
3461 | { |
3462 | struct io_splice *sp = &req->splice; | |
3463 | struct file *in = sp->file_in; | |
3464 | struct file *out = sp->file_out; | |
3465 | unsigned int flags = sp->flags & ~SPLICE_F_FD_IN_FIXED; | |
3466 | loff_t *poff_in, *poff_out; | |
c9687426 | 3467 | long ret = 0; |
7d67af2c | 3468 | |
2fb3e822 PB |
3469 | if (force_nonblock) |
3470 | return -EAGAIN; | |
7d67af2c PB |
3471 | |
3472 | poff_in = (sp->off_in == -1) ? NULL : &sp->off_in; | |
3473 | poff_out = (sp->off_out == -1) ? NULL : &sp->off_out; | |
c9687426 | 3474 | |
948a7749 | 3475 | if (sp->len) |
c9687426 | 3476 | ret = do_splice(in, poff_in, out, poff_out, sp->len, flags); |
7d67af2c PB |
3477 | |
3478 | io_put_file(req, in, (sp->flags & SPLICE_F_FD_IN_FIXED)); | |
3479 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
3480 | ||
7d67af2c PB |
3481 | if (ret != sp->len) |
3482 | req_set_fail_links(req); | |
e1e16097 | 3483 | io_req_complete(req, ret); |
7d67af2c PB |
3484 | return 0; |
3485 | } | |
3486 | ||
2b188cc1 JA |
3487 | /* |
3488 | * IORING_OP_NOP just posts a completion event, nothing else. | |
3489 | */ | |
229a7b63 | 3490 | static int io_nop(struct io_kiocb *req, struct io_comp_state *cs) |
2b188cc1 JA |
3491 | { |
3492 | struct io_ring_ctx *ctx = req->ctx; | |
2b188cc1 | 3493 | |
def596e9 JA |
3494 | if (unlikely(ctx->flags & IORING_SETUP_IOPOLL)) |
3495 | return -EINVAL; | |
3496 | ||
229a7b63 | 3497 | __io_req_complete(req, 0, 0, cs); |
2b188cc1 JA |
3498 | return 0; |
3499 | } | |
3500 | ||
3529d8c2 | 3501 | static int io_prep_fsync(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
c992fe29 | 3502 | { |
6b06314c | 3503 | struct io_ring_ctx *ctx = req->ctx; |
c992fe29 | 3504 | |
09bb8394 JA |
3505 | if (!req->file) |
3506 | return -EBADF; | |
c992fe29 | 3507 | |
6b06314c | 3508 | if (unlikely(ctx->flags & IORING_SETUP_IOPOLL)) |
def596e9 | 3509 | return -EINVAL; |
edafccee | 3510 | if (unlikely(sqe->addr || sqe->ioprio || sqe->buf_index)) |
c992fe29 CH |
3511 | return -EINVAL; |
3512 | ||
8ed8d3c3 JA |
3513 | req->sync.flags = READ_ONCE(sqe->fsync_flags); |
3514 | if (unlikely(req->sync.flags & ~IORING_FSYNC_DATASYNC)) | |
3515 | return -EINVAL; | |
3516 | ||
3517 | req->sync.off = READ_ONCE(sqe->off); | |
3518 | req->sync.len = READ_ONCE(sqe->len); | |
c992fe29 CH |
3519 | return 0; |
3520 | } | |
3521 | ||
ac45abc0 | 3522 | static int io_fsync(struct io_kiocb *req, bool force_nonblock) |
8ed8d3c3 | 3523 | { |
8ed8d3c3 | 3524 | loff_t end = req->sync.off + req->sync.len; |
8ed8d3c3 JA |
3525 | int ret; |
3526 | ||
ac45abc0 PB |
3527 | /* fsync always requires a blocking context */ |
3528 | if (force_nonblock) | |
3529 | return -EAGAIN; | |
3530 | ||
9adbd45d | 3531 | ret = vfs_fsync_range(req->file, req->sync.off, |
8ed8d3c3 JA |
3532 | end > 0 ? end : LLONG_MAX, |
3533 | req->sync.flags & IORING_FSYNC_DATASYNC); | |
3534 | if (ret < 0) | |
3535 | req_set_fail_links(req); | |
e1e16097 | 3536 | io_req_complete(req, ret); |
c992fe29 CH |
3537 | return 0; |
3538 | } | |
3539 | ||
d63d1b5e JA |
3540 | static int io_fallocate_prep(struct io_kiocb *req, |
3541 | const struct io_uring_sqe *sqe) | |
3542 | { | |
3543 | if (sqe->ioprio || sqe->buf_index || sqe->rw_flags) | |
3544 | return -EINVAL; | |
3232dd02 PB |
3545 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
3546 | return -EINVAL; | |
d63d1b5e JA |
3547 | |
3548 | req->sync.off = READ_ONCE(sqe->off); | |
3549 | req->sync.len = READ_ONCE(sqe->addr); | |
3550 | req->sync.mode = READ_ONCE(sqe->len); | |
3551 | return 0; | |
3552 | } | |
3553 | ||
014db007 | 3554 | static int io_fallocate(struct io_kiocb *req, bool force_nonblock) |
5d17b4a4 | 3555 | { |
ac45abc0 PB |
3556 | int ret; |
3557 | ||
d63d1b5e | 3558 | /* fallocate always requiring blocking context */ |
ac45abc0 | 3559 | if (force_nonblock) |
5d17b4a4 | 3560 | return -EAGAIN; |
ac45abc0 PB |
3561 | ret = vfs_fallocate(req->file, req->sync.mode, req->sync.off, |
3562 | req->sync.len); | |
ac45abc0 PB |
3563 | if (ret < 0) |
3564 | req_set_fail_links(req); | |
e1e16097 | 3565 | io_req_complete(req, ret); |
5d17b4a4 JA |
3566 | return 0; |
3567 | } | |
3568 | ||
ec65fea5 | 3569 | static int __io_openat_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
b7bb4f7d | 3570 | { |
f8748881 | 3571 | const char __user *fname; |
15b71abe | 3572 | int ret; |
b7bb4f7d | 3573 | |
ec65fea5 | 3574 | if (unlikely(sqe->ioprio || sqe->buf_index)) |
15b71abe | 3575 | return -EINVAL; |
ec65fea5 | 3576 | if (unlikely(req->flags & REQ_F_FIXED_FILE)) |
cf3040ca | 3577 | return -EBADF; |
03b1230c | 3578 | |
ec65fea5 PB |
3579 | /* open.how should be already initialised */ |
3580 | if (!(req->open.how.flags & O_PATH) && force_o_largefile()) | |
08a1d26e | 3581 | req->open.how.flags |= O_LARGEFILE; |
3529d8c2 | 3582 | |
25e72d10 PB |
3583 | req->open.dfd = READ_ONCE(sqe->fd); |
3584 | fname = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
f8748881 | 3585 | req->open.filename = getname(fname); |
15b71abe JA |
3586 | if (IS_ERR(req->open.filename)) { |
3587 | ret = PTR_ERR(req->open.filename); | |
3588 | req->open.filename = NULL; | |
3589 | return ret; | |
3590 | } | |
4022e7af | 3591 | req->open.nofile = rlimit(RLIMIT_NOFILE); |
8fef80bf | 3592 | req->flags |= REQ_F_NEED_CLEANUP; |
15b71abe | 3593 | return 0; |
03b1230c JA |
3594 | } |
3595 | ||
ec65fea5 PB |
3596 | static int io_openat_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
3597 | { | |
3598 | u64 flags, mode; | |
3599 | ||
4eb8dded JA |
3600 | if (unlikely(req->ctx->flags & (IORING_SETUP_IOPOLL|IORING_SETUP_SQPOLL))) |
3601 | return -EINVAL; | |
ec65fea5 PB |
3602 | if (req->flags & REQ_F_NEED_CLEANUP) |
3603 | return 0; | |
3604 | mode = READ_ONCE(sqe->len); | |
3605 | flags = READ_ONCE(sqe->open_flags); | |
3606 | req->open.how = build_open_how(flags, mode); | |
3607 | return __io_openat_prep(req, sqe); | |
3608 | } | |
3609 | ||
cebdb986 | 3610 | static int io_openat2_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
aa1fa28f | 3611 | { |
cebdb986 | 3612 | struct open_how __user *how; |
cebdb986 | 3613 | size_t len; |
0fa03c62 JA |
3614 | int ret; |
3615 | ||
4eb8dded JA |
3616 | if (unlikely(req->ctx->flags & (IORING_SETUP_IOPOLL|IORING_SETUP_SQPOLL))) |
3617 | return -EINVAL; | |
0bdbdd08 PB |
3618 | if (req->flags & REQ_F_NEED_CLEANUP) |
3619 | return 0; | |
cebdb986 JA |
3620 | how = u64_to_user_ptr(READ_ONCE(sqe->addr2)); |
3621 | len = READ_ONCE(sqe->len); | |
cebdb986 JA |
3622 | if (len < OPEN_HOW_SIZE_VER0) |
3623 | return -EINVAL; | |
3529d8c2 | 3624 | |
cebdb986 JA |
3625 | ret = copy_struct_from_user(&req->open.how, sizeof(req->open.how), how, |
3626 | len); | |
3627 | if (ret) | |
3628 | return ret; | |
3529d8c2 | 3629 | |
ec65fea5 | 3630 | return __io_openat_prep(req, sqe); |
cebdb986 JA |
3631 | } |
3632 | ||
014db007 | 3633 | static int io_openat2(struct io_kiocb *req, bool force_nonblock) |
15b71abe JA |
3634 | { |
3635 | struct open_flags op; | |
15b71abe JA |
3636 | struct file *file; |
3637 | int ret; | |
3638 | ||
f86cd20c | 3639 | if (force_nonblock) |
15b71abe | 3640 | return -EAGAIN; |
15b71abe | 3641 | |
cebdb986 | 3642 | ret = build_open_flags(&req->open.how, &op); |
15b71abe JA |
3643 | if (ret) |
3644 | goto err; | |
3645 | ||
4022e7af | 3646 | ret = __get_unused_fd_flags(req->open.how.flags, req->open.nofile); |
15b71abe JA |
3647 | if (ret < 0) |
3648 | goto err; | |
3649 | ||
3650 | file = do_filp_open(req->open.dfd, req->open.filename, &op); | |
3651 | if (IS_ERR(file)) { | |
3652 | put_unused_fd(ret); | |
3653 | ret = PTR_ERR(file); | |
3654 | } else { | |
3655 | fsnotify_open(file); | |
3656 | fd_install(ret, file); | |
3657 | } | |
3658 | err: | |
3659 | putname(req->open.filename); | |
8fef80bf | 3660 | req->flags &= ~REQ_F_NEED_CLEANUP; |
15b71abe JA |
3661 | if (ret < 0) |
3662 | req_set_fail_links(req); | |
e1e16097 | 3663 | io_req_complete(req, ret); |
15b71abe JA |
3664 | return 0; |
3665 | } | |
3666 | ||
014db007 | 3667 | static int io_openat(struct io_kiocb *req, bool force_nonblock) |
cebdb986 | 3668 | { |
014db007 | 3669 | return io_openat2(req, force_nonblock); |
cebdb986 JA |
3670 | } |
3671 | ||
067524e9 JA |
3672 | static int io_remove_buffers_prep(struct io_kiocb *req, |
3673 | const struct io_uring_sqe *sqe) | |
3674 | { | |
3675 | struct io_provide_buf *p = &req->pbuf; | |
3676 | u64 tmp; | |
3677 | ||
3678 | if (sqe->ioprio || sqe->rw_flags || sqe->addr || sqe->len || sqe->off) | |
3679 | return -EINVAL; | |
3680 | ||
3681 | tmp = READ_ONCE(sqe->fd); | |
3682 | if (!tmp || tmp > USHRT_MAX) | |
3683 | return -EINVAL; | |
3684 | ||
3685 | memset(p, 0, sizeof(*p)); | |
3686 | p->nbufs = tmp; | |
3687 | p->bgid = READ_ONCE(sqe->buf_group); | |
3688 | return 0; | |
3689 | } | |
3690 | ||
3691 | static int __io_remove_buffers(struct io_ring_ctx *ctx, struct io_buffer *buf, | |
3692 | int bgid, unsigned nbufs) | |
3693 | { | |
3694 | unsigned i = 0; | |
3695 | ||
3696 | /* shouldn't happen */ | |
3697 | if (!nbufs) | |
3698 | return 0; | |
3699 | ||
3700 | /* the head kbuf is the list itself */ | |
3701 | while (!list_empty(&buf->list)) { | |
3702 | struct io_buffer *nxt; | |
3703 | ||
3704 | nxt = list_first_entry(&buf->list, struct io_buffer, list); | |
3705 | list_del(&nxt->list); | |
3706 | kfree(nxt); | |
3707 | if (++i == nbufs) | |
3708 | return i; | |
3709 | } | |
3710 | i++; | |
3711 | kfree(buf); | |
3712 | idr_remove(&ctx->io_buffer_idr, bgid); | |
3713 | ||
3714 | return i; | |
3715 | } | |
3716 | ||
229a7b63 JA |
3717 | static int io_remove_buffers(struct io_kiocb *req, bool force_nonblock, |
3718 | struct io_comp_state *cs) | |
067524e9 JA |
3719 | { |
3720 | struct io_provide_buf *p = &req->pbuf; | |
3721 | struct io_ring_ctx *ctx = req->ctx; | |
3722 | struct io_buffer *head; | |
3723 | int ret = 0; | |
3724 | ||
3725 | io_ring_submit_lock(ctx, !force_nonblock); | |
3726 | ||
3727 | lockdep_assert_held(&ctx->uring_lock); | |
3728 | ||
3729 | ret = -ENOENT; | |
3730 | head = idr_find(&ctx->io_buffer_idr, p->bgid); | |
3731 | if (head) | |
3732 | ret = __io_remove_buffers(ctx, head, p->bgid, p->nbufs); | |
3733 | ||
3734 | io_ring_submit_lock(ctx, !force_nonblock); | |
3735 | if (ret < 0) | |
3736 | req_set_fail_links(req); | |
229a7b63 | 3737 | __io_req_complete(req, ret, 0, cs); |
067524e9 JA |
3738 | return 0; |
3739 | } | |
3740 | ||
ddf0322d JA |
3741 | static int io_provide_buffers_prep(struct io_kiocb *req, |
3742 | const struct io_uring_sqe *sqe) | |
3743 | { | |
3744 | struct io_provide_buf *p = &req->pbuf; | |
3745 | u64 tmp; | |
3746 | ||
3747 | if (sqe->ioprio || sqe->rw_flags) | |
3748 | return -EINVAL; | |
3749 | ||
3750 | tmp = READ_ONCE(sqe->fd); | |
3751 | if (!tmp || tmp > USHRT_MAX) | |
3752 | return -E2BIG; | |
3753 | p->nbufs = tmp; | |
3754 | p->addr = READ_ONCE(sqe->addr); | |
3755 | p->len = READ_ONCE(sqe->len); | |
3756 | ||
efe68c1c | 3757 | if (!access_ok(u64_to_user_ptr(p->addr), (p->len * p->nbufs))) |
ddf0322d JA |
3758 | return -EFAULT; |
3759 | ||
3760 | p->bgid = READ_ONCE(sqe->buf_group); | |
3761 | tmp = READ_ONCE(sqe->off); | |
3762 | if (tmp > USHRT_MAX) | |
3763 | return -E2BIG; | |
3764 | p->bid = tmp; | |
3765 | return 0; | |
3766 | } | |
3767 | ||
3768 | static int io_add_buffers(struct io_provide_buf *pbuf, struct io_buffer **head) | |
3769 | { | |
3770 | struct io_buffer *buf; | |
3771 | u64 addr = pbuf->addr; | |
3772 | int i, bid = pbuf->bid; | |
3773 | ||
3774 | for (i = 0; i < pbuf->nbufs; i++) { | |
3775 | buf = kmalloc(sizeof(*buf), GFP_KERNEL); | |
3776 | if (!buf) | |
3777 | break; | |
3778 | ||
3779 | buf->addr = addr; | |
3780 | buf->len = pbuf->len; | |
3781 | buf->bid = bid; | |
3782 | addr += pbuf->len; | |
3783 | bid++; | |
3784 | if (!*head) { | |
3785 | INIT_LIST_HEAD(&buf->list); | |
3786 | *head = buf; | |
3787 | } else { | |
3788 | list_add_tail(&buf->list, &(*head)->list); | |
3789 | } | |
3790 | } | |
3791 | ||
3792 | return i ? i : -ENOMEM; | |
3793 | } | |
3794 | ||
229a7b63 JA |
3795 | static int io_provide_buffers(struct io_kiocb *req, bool force_nonblock, |
3796 | struct io_comp_state *cs) | |
ddf0322d JA |
3797 | { |
3798 | struct io_provide_buf *p = &req->pbuf; | |
3799 | struct io_ring_ctx *ctx = req->ctx; | |
3800 | struct io_buffer *head, *list; | |
3801 | int ret = 0; | |
3802 | ||
3803 | io_ring_submit_lock(ctx, !force_nonblock); | |
3804 | ||
3805 | lockdep_assert_held(&ctx->uring_lock); | |
3806 | ||
3807 | list = head = idr_find(&ctx->io_buffer_idr, p->bgid); | |
3808 | ||
3809 | ret = io_add_buffers(p, &head); | |
3810 | if (ret < 0) | |
3811 | goto out; | |
3812 | ||
3813 | if (!list) { | |
3814 | ret = idr_alloc(&ctx->io_buffer_idr, head, p->bgid, p->bgid + 1, | |
3815 | GFP_KERNEL); | |
3816 | if (ret < 0) { | |
067524e9 | 3817 | __io_remove_buffers(ctx, head, p->bgid, -1U); |
ddf0322d JA |
3818 | goto out; |
3819 | } | |
3820 | } | |
3821 | out: | |
3822 | io_ring_submit_unlock(ctx, !force_nonblock); | |
3823 | if (ret < 0) | |
3824 | req_set_fail_links(req); | |
229a7b63 | 3825 | __io_req_complete(req, ret, 0, cs); |
ddf0322d | 3826 | return 0; |
cebdb986 JA |
3827 | } |
3828 | ||
3e4827b0 JA |
3829 | static int io_epoll_ctl_prep(struct io_kiocb *req, |
3830 | const struct io_uring_sqe *sqe) | |
3831 | { | |
3832 | #if defined(CONFIG_EPOLL) | |
3833 | if (sqe->ioprio || sqe->buf_index) | |
3834 | return -EINVAL; | |
6ca56f84 | 3835 | if (unlikely(req->ctx->flags & (IORING_SETUP_IOPOLL | IORING_SETUP_SQPOLL))) |
3232dd02 | 3836 | return -EINVAL; |
3e4827b0 JA |
3837 | |
3838 | req->epoll.epfd = READ_ONCE(sqe->fd); | |
3839 | req->epoll.op = READ_ONCE(sqe->len); | |
3840 | req->epoll.fd = READ_ONCE(sqe->off); | |
3841 | ||
3842 | if (ep_op_has_event(req->epoll.op)) { | |
3843 | struct epoll_event __user *ev; | |
3844 | ||
3845 | ev = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
3846 | if (copy_from_user(&req->epoll.event, ev, sizeof(*ev))) | |
3847 | return -EFAULT; | |
3848 | } | |
3849 | ||
3850 | return 0; | |
3851 | #else | |
3852 | return -EOPNOTSUPP; | |
3853 | #endif | |
3854 | } | |
3855 | ||
229a7b63 JA |
3856 | static int io_epoll_ctl(struct io_kiocb *req, bool force_nonblock, |
3857 | struct io_comp_state *cs) | |
3e4827b0 JA |
3858 | { |
3859 | #if defined(CONFIG_EPOLL) | |
3860 | struct io_epoll *ie = &req->epoll; | |
3861 | int ret; | |
3862 | ||
3863 | ret = do_epoll_ctl(ie->epfd, ie->op, ie->fd, &ie->event, force_nonblock); | |
3864 | if (force_nonblock && ret == -EAGAIN) | |
3865 | return -EAGAIN; | |
3866 | ||
3867 | if (ret < 0) | |
3868 | req_set_fail_links(req); | |
229a7b63 | 3869 | __io_req_complete(req, ret, 0, cs); |
3e4827b0 JA |
3870 | return 0; |
3871 | #else | |
3872 | return -EOPNOTSUPP; | |
3873 | #endif | |
3874 | } | |
3875 | ||
c1ca757b JA |
3876 | static int io_madvise_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
3877 | { | |
3878 | #if defined(CONFIG_ADVISE_SYSCALLS) && defined(CONFIG_MMU) | |
3879 | if (sqe->ioprio || sqe->buf_index || sqe->off) | |
3880 | return -EINVAL; | |
3232dd02 PB |
3881 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
3882 | return -EINVAL; | |
c1ca757b JA |
3883 | |
3884 | req->madvise.addr = READ_ONCE(sqe->addr); | |
3885 | req->madvise.len = READ_ONCE(sqe->len); | |
3886 | req->madvise.advice = READ_ONCE(sqe->fadvise_advice); | |
3887 | return 0; | |
3888 | #else | |
3889 | return -EOPNOTSUPP; | |
3890 | #endif | |
3891 | } | |
3892 | ||
014db007 | 3893 | static int io_madvise(struct io_kiocb *req, bool force_nonblock) |
c1ca757b JA |
3894 | { |
3895 | #if defined(CONFIG_ADVISE_SYSCALLS) && defined(CONFIG_MMU) | |
3896 | struct io_madvise *ma = &req->madvise; | |
3897 | int ret; | |
3898 | ||
3899 | if (force_nonblock) | |
3900 | return -EAGAIN; | |
3901 | ||
3902 | ret = do_madvise(ma->addr, ma->len, ma->advice); | |
3903 | if (ret < 0) | |
3904 | req_set_fail_links(req); | |
e1e16097 | 3905 | io_req_complete(req, ret); |
c1ca757b JA |
3906 | return 0; |
3907 | #else | |
3908 | return -EOPNOTSUPP; | |
3909 | #endif | |
3910 | } | |
3911 | ||
4840e418 JA |
3912 | static int io_fadvise_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
3913 | { | |
3914 | if (sqe->ioprio || sqe->buf_index || sqe->addr) | |
3915 | return -EINVAL; | |
3232dd02 PB |
3916 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
3917 | return -EINVAL; | |
4840e418 JA |
3918 | |
3919 | req->fadvise.offset = READ_ONCE(sqe->off); | |
3920 | req->fadvise.len = READ_ONCE(sqe->len); | |
3921 | req->fadvise.advice = READ_ONCE(sqe->fadvise_advice); | |
3922 | return 0; | |
3923 | } | |
3924 | ||
014db007 | 3925 | static int io_fadvise(struct io_kiocb *req, bool force_nonblock) |
4840e418 JA |
3926 | { |
3927 | struct io_fadvise *fa = &req->fadvise; | |
3928 | int ret; | |
3929 | ||
3e69426d JA |
3930 | if (force_nonblock) { |
3931 | switch (fa->advice) { | |
3932 | case POSIX_FADV_NORMAL: | |
3933 | case POSIX_FADV_RANDOM: | |
3934 | case POSIX_FADV_SEQUENTIAL: | |
3935 | break; | |
3936 | default: | |
3937 | return -EAGAIN; | |
3938 | } | |
3939 | } | |
4840e418 JA |
3940 | |
3941 | ret = vfs_fadvise(req->file, fa->offset, fa->len, fa->advice); | |
3942 | if (ret < 0) | |
3943 | req_set_fail_links(req); | |
e1e16097 | 3944 | io_req_complete(req, ret); |
4840e418 JA |
3945 | return 0; |
3946 | } | |
3947 | ||
eddc7ef5 JA |
3948 | static int io_statx_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
3949 | { | |
6ca56f84 | 3950 | if (unlikely(req->ctx->flags & (IORING_SETUP_IOPOLL | IORING_SETUP_SQPOLL))) |
3232dd02 | 3951 | return -EINVAL; |
eddc7ef5 JA |
3952 | if (sqe->ioprio || sqe->buf_index) |
3953 | return -EINVAL; | |
9c280f90 | 3954 | if (req->flags & REQ_F_FIXED_FILE) |
cf3040ca | 3955 | return -EBADF; |
eddc7ef5 | 3956 | |
1d9e1288 BM |
3957 | req->statx.dfd = READ_ONCE(sqe->fd); |
3958 | req->statx.mask = READ_ONCE(sqe->len); | |
e62753e4 | 3959 | req->statx.filename = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
1d9e1288 BM |
3960 | req->statx.buffer = u64_to_user_ptr(READ_ONCE(sqe->addr2)); |
3961 | req->statx.flags = READ_ONCE(sqe->statx_flags); | |
eddc7ef5 JA |
3962 | |
3963 | return 0; | |
3964 | } | |
3965 | ||
014db007 | 3966 | static int io_statx(struct io_kiocb *req, bool force_nonblock) |
eddc7ef5 | 3967 | { |
1d9e1288 | 3968 | struct io_statx *ctx = &req->statx; |
eddc7ef5 JA |
3969 | int ret; |
3970 | ||
5b0bbee4 JA |
3971 | if (force_nonblock) { |
3972 | /* only need file table for an actual valid fd */ | |
3973 | if (ctx->dfd == -1 || ctx->dfd == AT_FDCWD) | |
3974 | req->flags |= REQ_F_NO_FILE_TABLE; | |
eddc7ef5 | 3975 | return -EAGAIN; |
5b0bbee4 | 3976 | } |
eddc7ef5 | 3977 | |
e62753e4 BM |
3978 | ret = do_statx(ctx->dfd, ctx->filename, ctx->flags, ctx->mask, |
3979 | ctx->buffer); | |
eddc7ef5 | 3980 | |
eddc7ef5 JA |
3981 | if (ret < 0) |
3982 | req_set_fail_links(req); | |
e1e16097 | 3983 | io_req_complete(req, ret); |
eddc7ef5 JA |
3984 | return 0; |
3985 | } | |
3986 | ||
b5dba59e JA |
3987 | static int io_close_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
3988 | { | |
3989 | /* | |
3990 | * If we queue this for async, it must not be cancellable. That would | |
7cdaf587 XW |
3991 | * leave the 'file' in an undeterminate state, and here need to modify |
3992 | * io_wq_work.flags, so initialize io_wq_work firstly. | |
b5dba59e | 3993 | */ |
7cdaf587 | 3994 | io_req_init_async(req); |
b5dba59e JA |
3995 | req->work.flags |= IO_WQ_WORK_NO_CANCEL; |
3996 | ||
3232dd02 PB |
3997 | if (unlikely(req->ctx->flags & (IORING_SETUP_IOPOLL|IORING_SETUP_SQPOLL))) |
3998 | return -EINVAL; | |
b5dba59e JA |
3999 | if (sqe->ioprio || sqe->off || sqe->addr || sqe->len || |
4000 | sqe->rw_flags || sqe->buf_index) | |
4001 | return -EINVAL; | |
9c280f90 | 4002 | if (req->flags & REQ_F_FIXED_FILE) |
cf3040ca | 4003 | return -EBADF; |
b5dba59e JA |
4004 | |
4005 | req->close.fd = READ_ONCE(sqe->fd); | |
0f212204 | 4006 | if ((req->file && req->file->f_op == &io_uring_fops)) |
fd2206e4 | 4007 | return -EBADF; |
b5dba59e | 4008 | |
3af73b28 | 4009 | req->close.put_file = NULL; |
b5dba59e | 4010 | return 0; |
b5dba59e JA |
4011 | } |
4012 | ||
229a7b63 JA |
4013 | static int io_close(struct io_kiocb *req, bool force_nonblock, |
4014 | struct io_comp_state *cs) | |
b5dba59e | 4015 | { |
3af73b28 | 4016 | struct io_close *close = &req->close; |
b5dba59e JA |
4017 | int ret; |
4018 | ||
3af73b28 PB |
4019 | /* might be already done during nonblock submission */ |
4020 | if (!close->put_file) { | |
4021 | ret = __close_fd_get_file(close->fd, &close->put_file); | |
4022 | if (ret < 0) | |
4023 | return (ret == -ENOENT) ? -EBADF : ret; | |
4024 | } | |
b5dba59e JA |
4025 | |
4026 | /* if the file has a flush method, be safe and punt to async */ | |
3af73b28 | 4027 | if (close->put_file->f_op->flush && force_nonblock) { |
24c74678 PB |
4028 | /* was never set, but play safe */ |
4029 | req->flags &= ~REQ_F_NOWAIT; | |
0bf0eefd | 4030 | /* avoid grabbing files - we don't need the files */ |
24c74678 | 4031 | req->flags |= REQ_F_NO_FILE_TABLE; |
0bf0eefd | 4032 | return -EAGAIN; |
a2100672 | 4033 | } |
b5dba59e | 4034 | |
3af73b28 PB |
4035 | /* No ->flush() or already async, safely close from here */ |
4036 | ret = filp_close(close->put_file, req->work.files); | |
4037 | if (ret < 0) | |
4038 | req_set_fail_links(req); | |
3af73b28 PB |
4039 | fput(close->put_file); |
4040 | close->put_file = NULL; | |
229a7b63 | 4041 | __io_req_complete(req, ret, 0, cs); |
1a417f4e | 4042 | return 0; |
b5dba59e JA |
4043 | } |
4044 | ||
3529d8c2 | 4045 | static int io_prep_sfr(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
5d17b4a4 JA |
4046 | { |
4047 | struct io_ring_ctx *ctx = req->ctx; | |
5d17b4a4 JA |
4048 | |
4049 | if (!req->file) | |
4050 | return -EBADF; | |
5d17b4a4 JA |
4051 | |
4052 | if (unlikely(ctx->flags & IORING_SETUP_IOPOLL)) | |
4053 | return -EINVAL; | |
4054 | if (unlikely(sqe->addr || sqe->ioprio || sqe->buf_index)) | |
4055 | return -EINVAL; | |
4056 | ||
8ed8d3c3 JA |
4057 | req->sync.off = READ_ONCE(sqe->off); |
4058 | req->sync.len = READ_ONCE(sqe->len); | |
4059 | req->sync.flags = READ_ONCE(sqe->sync_range_flags); | |
8ed8d3c3 JA |
4060 | return 0; |
4061 | } | |
4062 | ||
ac45abc0 | 4063 | static int io_sync_file_range(struct io_kiocb *req, bool force_nonblock) |
8ed8d3c3 | 4064 | { |
8ed8d3c3 JA |
4065 | int ret; |
4066 | ||
ac45abc0 PB |
4067 | /* sync_file_range always requires a blocking context */ |
4068 | if (force_nonblock) | |
4069 | return -EAGAIN; | |
4070 | ||
9adbd45d | 4071 | ret = sync_file_range(req->file, req->sync.off, req->sync.len, |
8ed8d3c3 JA |
4072 | req->sync.flags); |
4073 | if (ret < 0) | |
4074 | req_set_fail_links(req); | |
e1e16097 | 4075 | io_req_complete(req, ret); |
5d17b4a4 JA |
4076 | return 0; |
4077 | } | |
4078 | ||
469956e8 | 4079 | #if defined(CONFIG_NET) |
02d27d89 PB |
4080 | static int io_setup_async_msg(struct io_kiocb *req, |
4081 | struct io_async_msghdr *kmsg) | |
4082 | { | |
4083 | if (req->io) | |
4084 | return -EAGAIN; | |
4085 | if (io_alloc_async_ctx(req)) { | |
4086 | if (kmsg->iov != kmsg->fast_iov) | |
4087 | kfree(kmsg->iov); | |
4088 | return -ENOMEM; | |
4089 | } | |
4090 | req->flags |= REQ_F_NEED_CLEANUP; | |
4091 | memcpy(&req->io->msg, kmsg, sizeof(*kmsg)); | |
4092 | return -EAGAIN; | |
4093 | } | |
4094 | ||
2ae523ed PB |
4095 | static int io_sendmsg_copy_hdr(struct io_kiocb *req, |
4096 | struct io_async_msghdr *iomsg) | |
4097 | { | |
4098 | iomsg->iov = iomsg->fast_iov; | |
4099 | iomsg->msg.msg_name = &iomsg->addr; | |
4100 | return sendmsg_copy_msghdr(&iomsg->msg, req->sr_msg.umsg, | |
4101 | req->sr_msg.msg_flags, &iomsg->iov); | |
4102 | } | |
4103 | ||
3529d8c2 | 4104 | static int io_sendmsg_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
03b1230c | 4105 | { |
e47293fd | 4106 | struct io_sr_msg *sr = &req->sr_msg; |
3529d8c2 | 4107 | struct io_async_ctx *io = req->io; |
99bc4c38 | 4108 | int ret; |
03b1230c | 4109 | |
d2b6f48b PB |
4110 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4111 | return -EINVAL; | |
4112 | ||
e47293fd | 4113 | sr->msg_flags = READ_ONCE(sqe->msg_flags); |
270a5940 | 4114 | sr->umsg = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
fddaface | 4115 | sr->len = READ_ONCE(sqe->len); |
3529d8c2 | 4116 | |
d8768362 JA |
4117 | #ifdef CONFIG_COMPAT |
4118 | if (req->ctx->compat) | |
4119 | sr->msg_flags |= MSG_CMSG_COMPAT; | |
4120 | #endif | |
4121 | ||
fddaface | 4122 | if (!io || req->opcode == IORING_OP_SEND) |
3529d8c2 | 4123 | return 0; |
5f798bea PB |
4124 | /* iovec is already imported */ |
4125 | if (req->flags & REQ_F_NEED_CLEANUP) | |
4126 | return 0; | |
3529d8c2 | 4127 | |
2ae523ed | 4128 | ret = io_sendmsg_copy_hdr(req, &io->msg); |
99bc4c38 PB |
4129 | if (!ret) |
4130 | req->flags |= REQ_F_NEED_CLEANUP; | |
4131 | return ret; | |
03b1230c JA |
4132 | } |
4133 | ||
229a7b63 JA |
4134 | static int io_sendmsg(struct io_kiocb *req, bool force_nonblock, |
4135 | struct io_comp_state *cs) | |
aa1fa28f | 4136 | { |
6b754c8b | 4137 | struct io_async_msghdr iomsg, *kmsg; |
0fa03c62 | 4138 | struct socket *sock; |
7a7cacba | 4139 | unsigned flags; |
0fa03c62 JA |
4140 | int ret; |
4141 | ||
0fa03c62 | 4142 | sock = sock_from_file(req->file, &ret); |
7a7cacba PB |
4143 | if (unlikely(!sock)) |
4144 | return ret; | |
3529d8c2 | 4145 | |
7a7cacba PB |
4146 | if (req->io) { |
4147 | kmsg = &req->io->msg; | |
4148 | kmsg->msg.msg_name = &req->io->msg.addr; | |
4149 | /* if iov is set, it's allocated already */ | |
4150 | if (!kmsg->iov) | |
4151 | kmsg->iov = kmsg->fast_iov; | |
4152 | kmsg->msg.msg_iter.iov = kmsg->iov; | |
4153 | } else { | |
4154 | ret = io_sendmsg_copy_hdr(req, &iomsg); | |
4155 | if (ret) | |
4156 | return ret; | |
4157 | kmsg = &iomsg; | |
0fa03c62 | 4158 | } |
0fa03c62 | 4159 | |
7a7cacba PB |
4160 | flags = req->sr_msg.msg_flags; |
4161 | if (flags & MSG_DONTWAIT) | |
4162 | req->flags |= REQ_F_NOWAIT; | |
4163 | else if (force_nonblock) | |
4164 | flags |= MSG_DONTWAIT; | |
e47293fd | 4165 | |
7a7cacba PB |
4166 | ret = __sys_sendmsg_sock(sock, &kmsg->msg, flags); |
4167 | if (force_nonblock && ret == -EAGAIN) | |
4168 | return io_setup_async_msg(req, kmsg); | |
4169 | if (ret == -ERESTARTSYS) | |
4170 | ret = -EINTR; | |
0fa03c62 | 4171 | |
6b754c8b | 4172 | if (kmsg->iov != kmsg->fast_iov) |
0b416c3e | 4173 | kfree(kmsg->iov); |
99bc4c38 | 4174 | req->flags &= ~REQ_F_NEED_CLEANUP; |
4e88d6e7 JA |
4175 | if (ret < 0) |
4176 | req_set_fail_links(req); | |
229a7b63 | 4177 | __io_req_complete(req, ret, 0, cs); |
5d17b4a4 | 4178 | return 0; |
03b1230c | 4179 | } |
aa1fa28f | 4180 | |
229a7b63 JA |
4181 | static int io_send(struct io_kiocb *req, bool force_nonblock, |
4182 | struct io_comp_state *cs) | |
fddaface | 4183 | { |
7a7cacba PB |
4184 | struct io_sr_msg *sr = &req->sr_msg; |
4185 | struct msghdr msg; | |
4186 | struct iovec iov; | |
fddaface | 4187 | struct socket *sock; |
7a7cacba | 4188 | unsigned flags; |
fddaface JA |
4189 | int ret; |
4190 | ||
fddaface | 4191 | sock = sock_from_file(req->file, &ret); |
7a7cacba PB |
4192 | if (unlikely(!sock)) |
4193 | return ret; | |
fddaface | 4194 | |
7a7cacba PB |
4195 | ret = import_single_range(WRITE, sr->buf, sr->len, &iov, &msg.msg_iter); |
4196 | if (unlikely(ret)) | |
14c32eee | 4197 | return ret;; |
fddaface | 4198 | |
7a7cacba PB |
4199 | msg.msg_name = NULL; |
4200 | msg.msg_control = NULL; | |
4201 | msg.msg_controllen = 0; | |
4202 | msg.msg_namelen = 0; | |
fddaface | 4203 | |
7a7cacba PB |
4204 | flags = req->sr_msg.msg_flags; |
4205 | if (flags & MSG_DONTWAIT) | |
4206 | req->flags |= REQ_F_NOWAIT; | |
4207 | else if (force_nonblock) | |
4208 | flags |= MSG_DONTWAIT; | |
fddaface | 4209 | |
7a7cacba PB |
4210 | msg.msg_flags = flags; |
4211 | ret = sock_sendmsg(sock, &msg); | |
4212 | if (force_nonblock && ret == -EAGAIN) | |
4213 | return -EAGAIN; | |
4214 | if (ret == -ERESTARTSYS) | |
4215 | ret = -EINTR; | |
fddaface | 4216 | |
fddaface JA |
4217 | if (ret < 0) |
4218 | req_set_fail_links(req); | |
229a7b63 | 4219 | __io_req_complete(req, ret, 0, cs); |
fddaface | 4220 | return 0; |
fddaface JA |
4221 | } |
4222 | ||
1400e697 PB |
4223 | static int __io_recvmsg_copy_hdr(struct io_kiocb *req, |
4224 | struct io_async_msghdr *iomsg) | |
52de1fe1 JA |
4225 | { |
4226 | struct io_sr_msg *sr = &req->sr_msg; | |
4227 | struct iovec __user *uiov; | |
4228 | size_t iov_len; | |
4229 | int ret; | |
4230 | ||
1400e697 PB |
4231 | ret = __copy_msghdr_from_user(&iomsg->msg, sr->umsg, |
4232 | &iomsg->uaddr, &uiov, &iov_len); | |
52de1fe1 JA |
4233 | if (ret) |
4234 | return ret; | |
4235 | ||
4236 | if (req->flags & REQ_F_BUFFER_SELECT) { | |
4237 | if (iov_len > 1) | |
4238 | return -EINVAL; | |
1400e697 | 4239 | if (copy_from_user(iomsg->iov, uiov, sizeof(*uiov))) |
52de1fe1 | 4240 | return -EFAULT; |
1400e697 PB |
4241 | sr->len = iomsg->iov[0].iov_len; |
4242 | iov_iter_init(&iomsg->msg.msg_iter, READ, iomsg->iov, 1, | |
52de1fe1 | 4243 | sr->len); |
1400e697 | 4244 | iomsg->iov = NULL; |
52de1fe1 JA |
4245 | } else { |
4246 | ret = import_iovec(READ, uiov, iov_len, UIO_FASTIOV, | |
1400e697 | 4247 | &iomsg->iov, &iomsg->msg.msg_iter); |
52de1fe1 JA |
4248 | if (ret > 0) |
4249 | ret = 0; | |
4250 | } | |
4251 | ||
4252 | return ret; | |
4253 | } | |
4254 | ||
4255 | #ifdef CONFIG_COMPAT | |
4256 | static int __io_compat_recvmsg_copy_hdr(struct io_kiocb *req, | |
1400e697 | 4257 | struct io_async_msghdr *iomsg) |
52de1fe1 JA |
4258 | { |
4259 | struct compat_msghdr __user *msg_compat; | |
4260 | struct io_sr_msg *sr = &req->sr_msg; | |
4261 | struct compat_iovec __user *uiov; | |
4262 | compat_uptr_t ptr; | |
4263 | compat_size_t len; | |
4264 | int ret; | |
4265 | ||
270a5940 | 4266 | msg_compat = (struct compat_msghdr __user *) sr->umsg; |
1400e697 | 4267 | ret = __get_compat_msghdr(&iomsg->msg, msg_compat, &iomsg->uaddr, |
52de1fe1 JA |
4268 | &ptr, &len); |
4269 | if (ret) | |
4270 | return ret; | |
4271 | ||
4272 | uiov = compat_ptr(ptr); | |
4273 | if (req->flags & REQ_F_BUFFER_SELECT) { | |
4274 | compat_ssize_t clen; | |
4275 | ||
4276 | if (len > 1) | |
4277 | return -EINVAL; | |
4278 | if (!access_ok(uiov, sizeof(*uiov))) | |
4279 | return -EFAULT; | |
4280 | if (__get_user(clen, &uiov->iov_len)) | |
4281 | return -EFAULT; | |
4282 | if (clen < 0) | |
4283 | return -EINVAL; | |
1400e697 PB |
4284 | sr->len = iomsg->iov[0].iov_len; |
4285 | iomsg->iov = NULL; | |
52de1fe1 JA |
4286 | } else { |
4287 | ret = compat_import_iovec(READ, uiov, len, UIO_FASTIOV, | |
1400e697 PB |
4288 | &iomsg->iov, |
4289 | &iomsg->msg.msg_iter); | |
52de1fe1 JA |
4290 | if (ret < 0) |
4291 | return ret; | |
4292 | } | |
4293 | ||
4294 | return 0; | |
4295 | } | |
4296 | #endif | |
4297 | ||
1400e697 PB |
4298 | static int io_recvmsg_copy_hdr(struct io_kiocb *req, |
4299 | struct io_async_msghdr *iomsg) | |
52de1fe1 | 4300 | { |
1400e697 PB |
4301 | iomsg->msg.msg_name = &iomsg->addr; |
4302 | iomsg->iov = iomsg->fast_iov; | |
52de1fe1 JA |
4303 | |
4304 | #ifdef CONFIG_COMPAT | |
4305 | if (req->ctx->compat) | |
1400e697 | 4306 | return __io_compat_recvmsg_copy_hdr(req, iomsg); |
fddaface | 4307 | #endif |
52de1fe1 | 4308 | |
1400e697 | 4309 | return __io_recvmsg_copy_hdr(req, iomsg); |
52de1fe1 JA |
4310 | } |
4311 | ||
bcda7baa | 4312 | static struct io_buffer *io_recv_buffer_select(struct io_kiocb *req, |
7fbb1b54 | 4313 | bool needs_lock) |
bcda7baa JA |
4314 | { |
4315 | struct io_sr_msg *sr = &req->sr_msg; | |
4316 | struct io_buffer *kbuf; | |
4317 | ||
bcda7baa JA |
4318 | kbuf = io_buffer_select(req, &sr->len, sr->bgid, sr->kbuf, needs_lock); |
4319 | if (IS_ERR(kbuf)) | |
4320 | return kbuf; | |
4321 | ||
4322 | sr->kbuf = kbuf; | |
4323 | req->flags |= REQ_F_BUFFER_SELECTED; | |
bcda7baa | 4324 | return kbuf; |
fddaface JA |
4325 | } |
4326 | ||
7fbb1b54 PB |
4327 | static inline unsigned int io_put_recv_kbuf(struct io_kiocb *req) |
4328 | { | |
4329 | return io_put_kbuf(req, req->sr_msg.kbuf); | |
4330 | } | |
4331 | ||
3529d8c2 JA |
4332 | static int io_recvmsg_prep(struct io_kiocb *req, |
4333 | const struct io_uring_sqe *sqe) | |
aa1fa28f | 4334 | { |
e47293fd | 4335 | struct io_sr_msg *sr = &req->sr_msg; |
3529d8c2 | 4336 | struct io_async_ctx *io = req->io; |
99bc4c38 | 4337 | int ret; |
3529d8c2 | 4338 | |
d2b6f48b PB |
4339 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4340 | return -EINVAL; | |
4341 | ||
3529d8c2 | 4342 | sr->msg_flags = READ_ONCE(sqe->msg_flags); |
270a5940 | 4343 | sr->umsg = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
0b7b21e4 | 4344 | sr->len = READ_ONCE(sqe->len); |
bcda7baa | 4345 | sr->bgid = READ_ONCE(sqe->buf_group); |
06b76d44 | 4346 | |
d8768362 JA |
4347 | #ifdef CONFIG_COMPAT |
4348 | if (req->ctx->compat) | |
4349 | sr->msg_flags |= MSG_CMSG_COMPAT; | |
4350 | #endif | |
4351 | ||
fddaface | 4352 | if (!io || req->opcode == IORING_OP_RECV) |
06b76d44 | 4353 | return 0; |
5f798bea PB |
4354 | /* iovec is already imported */ |
4355 | if (req->flags & REQ_F_NEED_CLEANUP) | |
4356 | return 0; | |
03b1230c | 4357 | |
1400e697 | 4358 | ret = io_recvmsg_copy_hdr(req, &io->msg); |
99bc4c38 PB |
4359 | if (!ret) |
4360 | req->flags |= REQ_F_NEED_CLEANUP; | |
4361 | return ret; | |
aa1fa28f JA |
4362 | } |
4363 | ||
229a7b63 JA |
4364 | static int io_recvmsg(struct io_kiocb *req, bool force_nonblock, |
4365 | struct io_comp_state *cs) | |
aa1fa28f | 4366 | { |
6b754c8b | 4367 | struct io_async_msghdr iomsg, *kmsg; |
03b1230c | 4368 | struct socket *sock; |
7fbb1b54 | 4369 | struct io_buffer *kbuf; |
7a7cacba | 4370 | unsigned flags; |
52de1fe1 | 4371 | int ret, cflags = 0; |
03b1230c | 4372 | |
03b1230c | 4373 | sock = sock_from_file(req->file, &ret); |
7a7cacba PB |
4374 | if (unlikely(!sock)) |
4375 | return ret; | |
3529d8c2 | 4376 | |
7a7cacba PB |
4377 | if (req->io) { |
4378 | kmsg = &req->io->msg; | |
4379 | kmsg->msg.msg_name = &req->io->msg.addr; | |
4380 | /* if iov is set, it's allocated already */ | |
4381 | if (!kmsg->iov) | |
4382 | kmsg->iov = kmsg->fast_iov; | |
4383 | kmsg->msg.msg_iter.iov = kmsg->iov; | |
4384 | } else { | |
4385 | ret = io_recvmsg_copy_hdr(req, &iomsg); | |
4386 | if (ret) | |
681fda8d | 4387 | return ret; |
7a7cacba PB |
4388 | kmsg = &iomsg; |
4389 | } | |
03b1230c | 4390 | |
bc02ef33 | 4391 | if (req->flags & REQ_F_BUFFER_SELECT) { |
7fbb1b54 | 4392 | kbuf = io_recv_buffer_select(req, !force_nonblock); |
bc02ef33 | 4393 | if (IS_ERR(kbuf)) |
52de1fe1 | 4394 | return PTR_ERR(kbuf); |
7a7cacba PB |
4395 | kmsg->fast_iov[0].iov_base = u64_to_user_ptr(kbuf->addr); |
4396 | iov_iter_init(&kmsg->msg.msg_iter, READ, kmsg->iov, | |
4397 | 1, req->sr_msg.len); | |
4398 | } | |
52de1fe1 | 4399 | |
7a7cacba PB |
4400 | flags = req->sr_msg.msg_flags; |
4401 | if (flags & MSG_DONTWAIT) | |
4402 | req->flags |= REQ_F_NOWAIT; | |
4403 | else if (force_nonblock) | |
4404 | flags |= MSG_DONTWAIT; | |
e47293fd | 4405 | |
7a7cacba PB |
4406 | ret = __sys_recvmsg_sock(sock, &kmsg->msg, req->sr_msg.umsg, |
4407 | kmsg->uaddr, flags); | |
0e1b6fe3 PB |
4408 | if (force_nonblock && ret == -EAGAIN) |
4409 | return io_setup_async_msg(req, kmsg); | |
7a7cacba PB |
4410 | if (ret == -ERESTARTSYS) |
4411 | ret = -EINTR; | |
03b1230c | 4412 | |
7fbb1b54 PB |
4413 | if (req->flags & REQ_F_BUFFER_SELECTED) |
4414 | cflags = io_put_recv_kbuf(req); | |
6b754c8b | 4415 | if (kmsg->iov != kmsg->fast_iov) |
0b416c3e | 4416 | kfree(kmsg->iov); |
99bc4c38 | 4417 | req->flags &= ~REQ_F_NEED_CLEANUP; |
4e88d6e7 JA |
4418 | if (ret < 0) |
4419 | req_set_fail_links(req); | |
229a7b63 | 4420 | __io_req_complete(req, ret, cflags, cs); |
03b1230c | 4421 | return 0; |
0fa03c62 | 4422 | } |
5d17b4a4 | 4423 | |
229a7b63 JA |
4424 | static int io_recv(struct io_kiocb *req, bool force_nonblock, |
4425 | struct io_comp_state *cs) | |
fddaface | 4426 | { |
6b754c8b | 4427 | struct io_buffer *kbuf; |
7a7cacba PB |
4428 | struct io_sr_msg *sr = &req->sr_msg; |
4429 | struct msghdr msg; | |
4430 | void __user *buf = sr->buf; | |
fddaface | 4431 | struct socket *sock; |
7a7cacba PB |
4432 | struct iovec iov; |
4433 | unsigned flags; | |
bcda7baa | 4434 | int ret, cflags = 0; |
fddaface | 4435 | |
fddaface | 4436 | sock = sock_from_file(req->file, &ret); |
7a7cacba PB |
4437 | if (unlikely(!sock)) |
4438 | return ret; | |
fddaface | 4439 | |
bc02ef33 | 4440 | if (req->flags & REQ_F_BUFFER_SELECT) { |
7fbb1b54 | 4441 | kbuf = io_recv_buffer_select(req, !force_nonblock); |
bcda7baa JA |
4442 | if (IS_ERR(kbuf)) |
4443 | return PTR_ERR(kbuf); | |
7a7cacba | 4444 | buf = u64_to_user_ptr(kbuf->addr); |
bc02ef33 | 4445 | } |
bcda7baa | 4446 | |
7a7cacba | 4447 | ret = import_single_range(READ, buf, sr->len, &iov, &msg.msg_iter); |
14c32eee PB |
4448 | if (unlikely(ret)) |
4449 | goto out_free; | |
fddaface | 4450 | |
7a7cacba PB |
4451 | msg.msg_name = NULL; |
4452 | msg.msg_control = NULL; | |
4453 | msg.msg_controllen = 0; | |
4454 | msg.msg_namelen = 0; | |
4455 | msg.msg_iocb = NULL; | |
4456 | msg.msg_flags = 0; | |
fddaface | 4457 | |
7a7cacba PB |
4458 | flags = req->sr_msg.msg_flags; |
4459 | if (flags & MSG_DONTWAIT) | |
4460 | req->flags |= REQ_F_NOWAIT; | |
4461 | else if (force_nonblock) | |
4462 | flags |= MSG_DONTWAIT; | |
4463 | ||
4464 | ret = sock_recvmsg(sock, &msg, flags); | |
4465 | if (force_nonblock && ret == -EAGAIN) | |
4466 | return -EAGAIN; | |
4467 | if (ret == -ERESTARTSYS) | |
4468 | ret = -EINTR; | |
14c32eee | 4469 | out_free: |
7fbb1b54 PB |
4470 | if (req->flags & REQ_F_BUFFER_SELECTED) |
4471 | cflags = io_put_recv_kbuf(req); | |
fddaface JA |
4472 | if (ret < 0) |
4473 | req_set_fail_links(req); | |
229a7b63 | 4474 | __io_req_complete(req, ret, cflags, cs); |
fddaface | 4475 | return 0; |
fddaface JA |
4476 | } |
4477 | ||
3529d8c2 | 4478 | static int io_accept_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
17f2fe35 | 4479 | { |
8ed8d3c3 JA |
4480 | struct io_accept *accept = &req->accept; |
4481 | ||
17f2fe35 JA |
4482 | if (unlikely(req->ctx->flags & (IORING_SETUP_IOPOLL|IORING_SETUP_SQPOLL))) |
4483 | return -EINVAL; | |
8042d6ce | 4484 | if (sqe->ioprio || sqe->len || sqe->buf_index) |
17f2fe35 JA |
4485 | return -EINVAL; |
4486 | ||
d55e5f5b JA |
4487 | accept->addr = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
4488 | accept->addr_len = u64_to_user_ptr(READ_ONCE(sqe->addr2)); | |
8ed8d3c3 | 4489 | accept->flags = READ_ONCE(sqe->accept_flags); |
09952e3e | 4490 | accept->nofile = rlimit(RLIMIT_NOFILE); |
8ed8d3c3 | 4491 | return 0; |
8ed8d3c3 | 4492 | } |
17f2fe35 | 4493 | |
229a7b63 JA |
4494 | static int io_accept(struct io_kiocb *req, bool force_nonblock, |
4495 | struct io_comp_state *cs) | |
8ed8d3c3 JA |
4496 | { |
4497 | struct io_accept *accept = &req->accept; | |
ac45abc0 | 4498 | unsigned int file_flags = force_nonblock ? O_NONBLOCK : 0; |
8ed8d3c3 JA |
4499 | int ret; |
4500 | ||
e697deed JX |
4501 | if (req->file->f_flags & O_NONBLOCK) |
4502 | req->flags |= REQ_F_NOWAIT; | |
4503 | ||
8ed8d3c3 | 4504 | ret = __sys_accept4_file(req->file, file_flags, accept->addr, |
09952e3e JA |
4505 | accept->addr_len, accept->flags, |
4506 | accept->nofile); | |
8ed8d3c3 | 4507 | if (ret == -EAGAIN && force_nonblock) |
17f2fe35 | 4508 | return -EAGAIN; |
ac45abc0 PB |
4509 | if (ret < 0) { |
4510 | if (ret == -ERESTARTSYS) | |
4511 | ret = -EINTR; | |
4e88d6e7 | 4512 | req_set_fail_links(req); |
ac45abc0 | 4513 | } |
229a7b63 | 4514 | __io_req_complete(req, ret, 0, cs); |
17f2fe35 | 4515 | return 0; |
8ed8d3c3 JA |
4516 | } |
4517 | ||
3529d8c2 | 4518 | static int io_connect_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
f499a021 | 4519 | { |
3529d8c2 JA |
4520 | struct io_connect *conn = &req->connect; |
4521 | struct io_async_ctx *io = req->io; | |
f499a021 | 4522 | |
3fbb51c1 JA |
4523 | if (unlikely(req->ctx->flags & (IORING_SETUP_IOPOLL|IORING_SETUP_SQPOLL))) |
4524 | return -EINVAL; | |
4525 | if (sqe->ioprio || sqe->len || sqe->buf_index || sqe->rw_flags) | |
4526 | return -EINVAL; | |
4527 | ||
3529d8c2 JA |
4528 | conn->addr = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
4529 | conn->addr_len = READ_ONCE(sqe->addr2); | |
4530 | ||
4531 | if (!io) | |
4532 | return 0; | |
4533 | ||
4534 | return move_addr_to_kernel(conn->addr, conn->addr_len, | |
3fbb51c1 | 4535 | &io->connect.address); |
f499a021 JA |
4536 | } |
4537 | ||
229a7b63 JA |
4538 | static int io_connect(struct io_kiocb *req, bool force_nonblock, |
4539 | struct io_comp_state *cs) | |
f8e85cf2 | 4540 | { |
f499a021 | 4541 | struct io_async_ctx __io, *io; |
f8e85cf2 | 4542 | unsigned file_flags; |
3fbb51c1 | 4543 | int ret; |
f8e85cf2 | 4544 | |
f499a021 JA |
4545 | if (req->io) { |
4546 | io = req->io; | |
4547 | } else { | |
3529d8c2 JA |
4548 | ret = move_addr_to_kernel(req->connect.addr, |
4549 | req->connect.addr_len, | |
4550 | &__io.connect.address); | |
f499a021 JA |
4551 | if (ret) |
4552 | goto out; | |
4553 | io = &__io; | |
4554 | } | |
4555 | ||
3fbb51c1 JA |
4556 | file_flags = force_nonblock ? O_NONBLOCK : 0; |
4557 | ||
4558 | ret = __sys_connect_file(req->file, &io->connect.address, | |
4559 | req->connect.addr_len, file_flags); | |
87f80d62 | 4560 | if ((ret == -EAGAIN || ret == -EINPROGRESS) && force_nonblock) { |
b7bb4f7d JA |
4561 | if (req->io) |
4562 | return -EAGAIN; | |
4563 | if (io_alloc_async_ctx(req)) { | |
f499a021 JA |
4564 | ret = -ENOMEM; |
4565 | goto out; | |
4566 | } | |
b7bb4f7d | 4567 | memcpy(&req->io->connect, &__io.connect, sizeof(__io.connect)); |
f8e85cf2 | 4568 | return -EAGAIN; |
f499a021 | 4569 | } |
f8e85cf2 JA |
4570 | if (ret == -ERESTARTSYS) |
4571 | ret = -EINTR; | |
f499a021 | 4572 | out: |
4e88d6e7 JA |
4573 | if (ret < 0) |
4574 | req_set_fail_links(req); | |
229a7b63 | 4575 | __io_req_complete(req, ret, 0, cs); |
f8e85cf2 | 4576 | return 0; |
469956e8 Y |
4577 | } |
4578 | #else /* !CONFIG_NET */ | |
4579 | static int io_sendmsg_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) | |
4580 | { | |
f8e85cf2 | 4581 | return -EOPNOTSUPP; |
f8e85cf2 JA |
4582 | } |
4583 | ||
1e16c2f9 RD |
4584 | static int io_sendmsg(struct io_kiocb *req, bool force_nonblock, |
4585 | struct io_comp_state *cs) | |
469956e8 Y |
4586 | { |
4587 | return -EOPNOTSUPP; | |
4588 | } | |
4589 | ||
1e16c2f9 RD |
4590 | static int io_send(struct io_kiocb *req, bool force_nonblock, |
4591 | struct io_comp_state *cs) | |
469956e8 Y |
4592 | { |
4593 | return -EOPNOTSUPP; | |
4594 | } | |
4595 | ||
4596 | static int io_recvmsg_prep(struct io_kiocb *req, | |
4597 | const struct io_uring_sqe *sqe) | |
4598 | { | |
4599 | return -EOPNOTSUPP; | |
4600 | } | |
4601 | ||
1e16c2f9 RD |
4602 | static int io_recvmsg(struct io_kiocb *req, bool force_nonblock, |
4603 | struct io_comp_state *cs) | |
469956e8 Y |
4604 | { |
4605 | return -EOPNOTSUPP; | |
4606 | } | |
4607 | ||
1e16c2f9 RD |
4608 | static int io_recv(struct io_kiocb *req, bool force_nonblock, |
4609 | struct io_comp_state *cs) | |
469956e8 Y |
4610 | { |
4611 | return -EOPNOTSUPP; | |
4612 | } | |
4613 | ||
4614 | static int io_accept_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) | |
4615 | { | |
4616 | return -EOPNOTSUPP; | |
4617 | } | |
4618 | ||
1e16c2f9 RD |
4619 | static int io_accept(struct io_kiocb *req, bool force_nonblock, |
4620 | struct io_comp_state *cs) | |
469956e8 Y |
4621 | { |
4622 | return -EOPNOTSUPP; | |
4623 | } | |
ce593a6c | 4624 | |
469956e8 Y |
4625 | static int io_connect_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
4626 | { | |
4627 | return -EOPNOTSUPP; | |
4628 | } | |
4629 | ||
1e16c2f9 RD |
4630 | static int io_connect(struct io_kiocb *req, bool force_nonblock, |
4631 | struct io_comp_state *cs) | |
469956e8 | 4632 | { |
f8e85cf2 | 4633 | return -EOPNOTSUPP; |
ce593a6c | 4634 | } |
469956e8 | 4635 | #endif /* CONFIG_NET */ |
f8e85cf2 | 4636 | |
d7718a9d JA |
4637 | struct io_poll_table { |
4638 | struct poll_table_struct pt; | |
4639 | struct io_kiocb *req; | |
4640 | int error; | |
4641 | }; | |
ce593a6c | 4642 | |
d7718a9d JA |
4643 | static int __io_async_wake(struct io_kiocb *req, struct io_poll_iocb *poll, |
4644 | __poll_t mask, task_work_func_t func) | |
4645 | { | |
fd7d6de2 | 4646 | bool twa_signal_ok; |
aa96bf8a | 4647 | int ret; |
d7718a9d JA |
4648 | |
4649 | /* for instances that support it check for an event match first: */ | |
4650 | if (mask && !(mask & poll->events)) | |
4651 | return 0; | |
4652 | ||
4653 | trace_io_uring_task_add(req->ctx, req->opcode, req->user_data, mask); | |
4654 | ||
4655 | list_del_init(&poll->wait.entry); | |
4656 | ||
d7718a9d JA |
4657 | req->result = mask; |
4658 | init_task_work(&req->task_work, func); | |
6d816e08 JA |
4659 | percpu_ref_get(&req->ctx->refs); |
4660 | ||
fd7d6de2 JA |
4661 | /* |
4662 | * If we using the signalfd wait_queue_head for this wakeup, then | |
4663 | * it's not safe to use TWA_SIGNAL as we could be recursing on the | |
4664 | * tsk->sighand->siglock on doing the wakeup. Should not be needed | |
4665 | * either, as the normal wakeup will suffice. | |
4666 | */ | |
4667 | twa_signal_ok = (poll->head != &req->task->sighand->signalfd_wqh); | |
4668 | ||
d7718a9d | 4669 | /* |
e3aabf95 JA |
4670 | * If this fails, then the task is exiting. When a task exits, the |
4671 | * work gets canceled, so just cancel this request as well instead | |
4672 | * of executing it. We can't safely execute it anyway, as we may not | |
4673 | * have the needed state needed for it anyway. | |
d7718a9d | 4674 | */ |
fd7d6de2 | 4675 | ret = io_req_task_work_add(req, &req->task_work, twa_signal_ok); |
aa96bf8a | 4676 | if (unlikely(ret)) { |
c2c4c83c JA |
4677 | struct task_struct *tsk; |
4678 | ||
e3aabf95 | 4679 | WRITE_ONCE(poll->canceled, true); |
aa96bf8a | 4680 | tsk = io_wq_get_task(req->ctx->io_wq); |
ce593a6c JA |
4681 | task_work_add(tsk, &req->task_work, 0); |
4682 | wake_up_process(tsk); | |
aa96bf8a | 4683 | } |
d7718a9d JA |
4684 | return 1; |
4685 | } | |
4686 | ||
74ce6ce4 JA |
4687 | static bool io_poll_rewait(struct io_kiocb *req, struct io_poll_iocb *poll) |
4688 | __acquires(&req->ctx->completion_lock) | |
4689 | { | |
4690 | struct io_ring_ctx *ctx = req->ctx; | |
4691 | ||
4692 | if (!req->result && !READ_ONCE(poll->canceled)) { | |
4693 | struct poll_table_struct pt = { ._key = poll->events }; | |
4694 | ||
4695 | req->result = vfs_poll(req->file, &pt) & poll->events; | |
4696 | } | |
4697 | ||
4698 | spin_lock_irq(&ctx->completion_lock); | |
4699 | if (!req->result && !READ_ONCE(poll->canceled)) { | |
4700 | add_wait_queue(poll->head, &poll->wait); | |
4701 | return true; | |
4702 | } | |
4703 | ||
4704 | return false; | |
4705 | } | |
4706 | ||
d4e7cd36 | 4707 | static struct io_poll_iocb *io_poll_get_double(struct io_kiocb *req) |
18bceab1 | 4708 | { |
d4e7cd36 JA |
4709 | /* pure poll stashes this in ->io, poll driven retry elsewhere */ |
4710 | if (req->opcode == IORING_OP_POLL_ADD) | |
4711 | return (struct io_poll_iocb *) req->io; | |
4712 | return req->apoll->double_poll; | |
4713 | } | |
4714 | ||
4715 | static struct io_poll_iocb *io_poll_get_single(struct io_kiocb *req) | |
4716 | { | |
4717 | if (req->opcode == IORING_OP_POLL_ADD) | |
4718 | return &req->poll; | |
4719 | return &req->apoll->poll; | |
4720 | } | |
4721 | ||
4722 | static void io_poll_remove_double(struct io_kiocb *req) | |
4723 | { | |
4724 | struct io_poll_iocb *poll = io_poll_get_double(req); | |
18bceab1 JA |
4725 | |
4726 | lockdep_assert_held(&req->ctx->completion_lock); | |
4727 | ||
4728 | if (poll && poll->head) { | |
4729 | struct wait_queue_head *head = poll->head; | |
4730 | ||
4731 | spin_lock(&head->lock); | |
4732 | list_del_init(&poll->wait.entry); | |
4733 | if (poll->wait.private) | |
4734 | refcount_dec(&req->refs); | |
4735 | poll->head = NULL; | |
4736 | spin_unlock(&head->lock); | |
4737 | } | |
4738 | } | |
4739 | ||
4740 | static void io_poll_complete(struct io_kiocb *req, __poll_t mask, int error) | |
4741 | { | |
4742 | struct io_ring_ctx *ctx = req->ctx; | |
4743 | ||
d4e7cd36 | 4744 | io_poll_remove_double(req); |
18bceab1 JA |
4745 | req->poll.done = true; |
4746 | io_cqring_fill_event(req, error ? error : mangle_poll(mask)); | |
4747 | io_commit_cqring(ctx); | |
4748 | } | |
4749 | ||
4750 | static void io_poll_task_handler(struct io_kiocb *req, struct io_kiocb **nxt) | |
4751 | { | |
4752 | struct io_ring_ctx *ctx = req->ctx; | |
4753 | ||
4754 | if (io_poll_rewait(req, &req->poll)) { | |
4755 | spin_unlock_irq(&ctx->completion_lock); | |
4756 | return; | |
4757 | } | |
4758 | ||
4759 | hash_del(&req->hash_node); | |
4760 | io_poll_complete(req, req->result, 0); | |
4761 | req->flags |= REQ_F_COMP_LOCKED; | |
9b5f7bd9 | 4762 | *nxt = io_put_req_find_next(req); |
18bceab1 JA |
4763 | spin_unlock_irq(&ctx->completion_lock); |
4764 | ||
4765 | io_cqring_ev_posted(ctx); | |
4766 | } | |
4767 | ||
4768 | static void io_poll_task_func(struct callback_head *cb) | |
4769 | { | |
4770 | struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work); | |
6d816e08 | 4771 | struct io_ring_ctx *ctx = req->ctx; |
18bceab1 JA |
4772 | struct io_kiocb *nxt = NULL; |
4773 | ||
4774 | io_poll_task_handler(req, &nxt); | |
ea1164e5 PB |
4775 | if (nxt) |
4776 | __io_req_task_submit(nxt); | |
6d816e08 | 4777 | percpu_ref_put(&ctx->refs); |
18bceab1 JA |
4778 | } |
4779 | ||
4780 | static int io_poll_double_wake(struct wait_queue_entry *wait, unsigned mode, | |
4781 | int sync, void *key) | |
4782 | { | |
4783 | struct io_kiocb *req = wait->private; | |
d4e7cd36 | 4784 | struct io_poll_iocb *poll = io_poll_get_single(req); |
18bceab1 JA |
4785 | __poll_t mask = key_to_poll(key); |
4786 | ||
4787 | /* for instances that support it check for an event match first: */ | |
4788 | if (mask && !(mask & poll->events)) | |
4789 | return 0; | |
4790 | ||
8706e04e JA |
4791 | list_del_init(&wait->entry); |
4792 | ||
807abcb0 | 4793 | if (poll && poll->head) { |
18bceab1 JA |
4794 | bool done; |
4795 | ||
807abcb0 JA |
4796 | spin_lock(&poll->head->lock); |
4797 | done = list_empty(&poll->wait.entry); | |
18bceab1 | 4798 | if (!done) |
807abcb0 | 4799 | list_del_init(&poll->wait.entry); |
d4e7cd36 JA |
4800 | /* make sure double remove sees this as being gone */ |
4801 | wait->private = NULL; | |
807abcb0 | 4802 | spin_unlock(&poll->head->lock); |
18bceab1 JA |
4803 | if (!done) |
4804 | __io_async_wake(req, poll, mask, io_poll_task_func); | |
4805 | } | |
4806 | refcount_dec(&req->refs); | |
4807 | return 1; | |
4808 | } | |
4809 | ||
4810 | static void io_init_poll_iocb(struct io_poll_iocb *poll, __poll_t events, | |
4811 | wait_queue_func_t wake_func) | |
4812 | { | |
4813 | poll->head = NULL; | |
4814 | poll->done = false; | |
4815 | poll->canceled = false; | |
4816 | poll->events = events; | |
4817 | INIT_LIST_HEAD(&poll->wait.entry); | |
4818 | init_waitqueue_func_entry(&poll->wait, wake_func); | |
4819 | } | |
4820 | ||
4821 | static void __io_queue_proc(struct io_poll_iocb *poll, struct io_poll_table *pt, | |
807abcb0 JA |
4822 | struct wait_queue_head *head, |
4823 | struct io_poll_iocb **poll_ptr) | |
18bceab1 JA |
4824 | { |
4825 | struct io_kiocb *req = pt->req; | |
4826 | ||
4827 | /* | |
4828 | * If poll->head is already set, it's because the file being polled | |
4829 | * uses multiple waitqueues for poll handling (eg one for read, one | |
4830 | * for write). Setup a separate io_poll_iocb if this happens. | |
4831 | */ | |
4832 | if (unlikely(poll->head)) { | |
4833 | /* already have a 2nd entry, fail a third attempt */ | |
807abcb0 | 4834 | if (*poll_ptr) { |
18bceab1 JA |
4835 | pt->error = -EINVAL; |
4836 | return; | |
4837 | } | |
4838 | poll = kmalloc(sizeof(*poll), GFP_ATOMIC); | |
4839 | if (!poll) { | |
4840 | pt->error = -ENOMEM; | |
4841 | return; | |
4842 | } | |
4843 | io_init_poll_iocb(poll, req->poll.events, io_poll_double_wake); | |
4844 | refcount_inc(&req->refs); | |
4845 | poll->wait.private = req; | |
807abcb0 | 4846 | *poll_ptr = poll; |
18bceab1 JA |
4847 | } |
4848 | ||
4849 | pt->error = 0; | |
4850 | poll->head = head; | |
a31eb4a2 JX |
4851 | |
4852 | if (poll->events & EPOLLEXCLUSIVE) | |
4853 | add_wait_queue_exclusive(head, &poll->wait); | |
4854 | else | |
4855 | add_wait_queue(head, &poll->wait); | |
18bceab1 JA |
4856 | } |
4857 | ||
4858 | static void io_async_queue_proc(struct file *file, struct wait_queue_head *head, | |
4859 | struct poll_table_struct *p) | |
4860 | { | |
4861 | struct io_poll_table *pt = container_of(p, struct io_poll_table, pt); | |
807abcb0 | 4862 | struct async_poll *apoll = pt->req->apoll; |
18bceab1 | 4863 | |
807abcb0 | 4864 | __io_queue_proc(&apoll->poll, pt, head, &apoll->double_poll); |
18bceab1 JA |
4865 | } |
4866 | ||
d7718a9d JA |
4867 | static void io_async_task_func(struct callback_head *cb) |
4868 | { | |
4869 | struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work); | |
4870 | struct async_poll *apoll = req->apoll; | |
4871 | struct io_ring_ctx *ctx = req->ctx; | |
4872 | ||
4873 | trace_io_uring_task_run(req->ctx, req->opcode, req->user_data); | |
4874 | ||
74ce6ce4 | 4875 | if (io_poll_rewait(req, &apoll->poll)) { |
d7718a9d | 4876 | spin_unlock_irq(&ctx->completion_lock); |
6d816e08 | 4877 | percpu_ref_put(&ctx->refs); |
74ce6ce4 | 4878 | return; |
d7718a9d JA |
4879 | } |
4880 | ||
31067255 | 4881 | /* If req is still hashed, it cannot have been canceled. Don't check. */ |
0be0b0e3 | 4882 | if (hash_hashed(&req->hash_node)) |
74ce6ce4 | 4883 | hash_del(&req->hash_node); |
2bae047e | 4884 | |
d4e7cd36 | 4885 | io_poll_remove_double(req); |
74ce6ce4 JA |
4886 | spin_unlock_irq(&ctx->completion_lock); |
4887 | ||
0be0b0e3 PB |
4888 | if (!READ_ONCE(apoll->poll.canceled)) |
4889 | __io_req_task_submit(req); | |
4890 | else | |
4891 | __io_req_task_cancel(req, -ECANCELED); | |
aa340845 | 4892 | |
6d816e08 | 4893 | percpu_ref_put(&ctx->refs); |
807abcb0 | 4894 | kfree(apoll->double_poll); |
31067255 | 4895 | kfree(apoll); |
d7718a9d JA |
4896 | } |
4897 | ||
4898 | static int io_async_wake(struct wait_queue_entry *wait, unsigned mode, int sync, | |
4899 | void *key) | |
4900 | { | |
4901 | struct io_kiocb *req = wait->private; | |
4902 | struct io_poll_iocb *poll = &req->apoll->poll; | |
4903 | ||
4904 | trace_io_uring_poll_wake(req->ctx, req->opcode, req->user_data, | |
4905 | key_to_poll(key)); | |
4906 | ||
4907 | return __io_async_wake(req, poll, key_to_poll(key), io_async_task_func); | |
4908 | } | |
4909 | ||
4910 | static void io_poll_req_insert(struct io_kiocb *req) | |
4911 | { | |
4912 | struct io_ring_ctx *ctx = req->ctx; | |
4913 | struct hlist_head *list; | |
4914 | ||
4915 | list = &ctx->cancel_hash[hash_long(req->user_data, ctx->cancel_hash_bits)]; | |
4916 | hlist_add_head(&req->hash_node, list); | |
4917 | } | |
4918 | ||
4919 | static __poll_t __io_arm_poll_handler(struct io_kiocb *req, | |
4920 | struct io_poll_iocb *poll, | |
4921 | struct io_poll_table *ipt, __poll_t mask, | |
4922 | wait_queue_func_t wake_func) | |
4923 | __acquires(&ctx->completion_lock) | |
4924 | { | |
4925 | struct io_ring_ctx *ctx = req->ctx; | |
4926 | bool cancel = false; | |
4927 | ||
18bceab1 | 4928 | io_init_poll_iocb(poll, mask, wake_func); |
b90cd197 | 4929 | poll->file = req->file; |
18bceab1 | 4930 | poll->wait.private = req; |
d7718a9d JA |
4931 | |
4932 | ipt->pt._key = mask; | |
4933 | ipt->req = req; | |
4934 | ipt->error = -EINVAL; | |
4935 | ||
d7718a9d JA |
4936 | mask = vfs_poll(req->file, &ipt->pt) & poll->events; |
4937 | ||
4938 | spin_lock_irq(&ctx->completion_lock); | |
4939 | if (likely(poll->head)) { | |
4940 | spin_lock(&poll->head->lock); | |
4941 | if (unlikely(list_empty(&poll->wait.entry))) { | |
4942 | if (ipt->error) | |
4943 | cancel = true; | |
4944 | ipt->error = 0; | |
4945 | mask = 0; | |
4946 | } | |
4947 | if (mask || ipt->error) | |
4948 | list_del_init(&poll->wait.entry); | |
4949 | else if (cancel) | |
4950 | WRITE_ONCE(poll->canceled, true); | |
4951 | else if (!poll->done) /* actually waiting for an event */ | |
4952 | io_poll_req_insert(req); | |
4953 | spin_unlock(&poll->head->lock); | |
4954 | } | |
4955 | ||
4956 | return mask; | |
4957 | } | |
4958 | ||
4959 | static bool io_arm_poll_handler(struct io_kiocb *req) | |
4960 | { | |
4961 | const struct io_op_def *def = &io_op_defs[req->opcode]; | |
4962 | struct io_ring_ctx *ctx = req->ctx; | |
4963 | struct async_poll *apoll; | |
4964 | struct io_poll_table ipt; | |
4965 | __poll_t mask, ret; | |
9dab14b8 | 4966 | int rw; |
d7718a9d JA |
4967 | |
4968 | if (!req->file || !file_can_poll(req->file)) | |
4969 | return false; | |
24c74678 | 4970 | if (req->flags & REQ_F_POLLED) |
d7718a9d | 4971 | return false; |
9dab14b8 JA |
4972 | if (def->pollin) |
4973 | rw = READ; | |
4974 | else if (def->pollout) | |
4975 | rw = WRITE; | |
4976 | else | |
4977 | return false; | |
4978 | /* if we can't nonblock try, then no point in arming a poll handler */ | |
4979 | if (!io_file_supports_async(req->file, rw)) | |
d7718a9d JA |
4980 | return false; |
4981 | ||
4982 | apoll = kmalloc(sizeof(*apoll), GFP_ATOMIC); | |
4983 | if (unlikely(!apoll)) | |
4984 | return false; | |
807abcb0 | 4985 | apoll->double_poll = NULL; |
d7718a9d JA |
4986 | |
4987 | req->flags |= REQ_F_POLLED; | |
d7718a9d JA |
4988 | req->apoll = apoll; |
4989 | INIT_HLIST_NODE(&req->hash_node); | |
4990 | ||
8755d97a | 4991 | mask = 0; |
d7718a9d | 4992 | if (def->pollin) |
8755d97a | 4993 | mask |= POLLIN | POLLRDNORM; |
d7718a9d JA |
4994 | if (def->pollout) |
4995 | mask |= POLLOUT | POLLWRNORM; | |
4996 | mask |= POLLERR | POLLPRI; | |
4997 | ||
4998 | ipt.pt._qproc = io_async_queue_proc; | |
4999 | ||
5000 | ret = __io_arm_poll_handler(req, &apoll->poll, &ipt, mask, | |
5001 | io_async_wake); | |
a36da65c | 5002 | if (ret || ipt.error) { |
d4e7cd36 | 5003 | io_poll_remove_double(req); |
d7718a9d | 5004 | spin_unlock_irq(&ctx->completion_lock); |
807abcb0 | 5005 | kfree(apoll->double_poll); |
d7718a9d JA |
5006 | kfree(apoll); |
5007 | return false; | |
5008 | } | |
5009 | spin_unlock_irq(&ctx->completion_lock); | |
5010 | trace_io_uring_poll_arm(ctx, req->opcode, req->user_data, mask, | |
5011 | apoll->poll.events); | |
5012 | return true; | |
5013 | } | |
5014 | ||
5015 | static bool __io_poll_remove_one(struct io_kiocb *req, | |
5016 | struct io_poll_iocb *poll) | |
221c5eb2 | 5017 | { |
b41e9852 | 5018 | bool do_complete = false; |
221c5eb2 JA |
5019 | |
5020 | spin_lock(&poll->head->lock); | |
5021 | WRITE_ONCE(poll->canceled, true); | |
392edb45 JA |
5022 | if (!list_empty(&poll->wait.entry)) { |
5023 | list_del_init(&poll->wait.entry); | |
b41e9852 | 5024 | do_complete = true; |
221c5eb2 JA |
5025 | } |
5026 | spin_unlock(&poll->head->lock); | |
3bfa5bcb | 5027 | hash_del(&req->hash_node); |
d7718a9d JA |
5028 | return do_complete; |
5029 | } | |
5030 | ||
5031 | static bool io_poll_remove_one(struct io_kiocb *req) | |
5032 | { | |
5033 | bool do_complete; | |
5034 | ||
d4e7cd36 JA |
5035 | io_poll_remove_double(req); |
5036 | ||
d7718a9d JA |
5037 | if (req->opcode == IORING_OP_POLL_ADD) { |
5038 | do_complete = __io_poll_remove_one(req, &req->poll); | |
5039 | } else { | |
3bfa5bcb JA |
5040 | struct async_poll *apoll = req->apoll; |
5041 | ||
d7718a9d | 5042 | /* non-poll requests have submit ref still */ |
3bfa5bcb JA |
5043 | do_complete = __io_poll_remove_one(req, &apoll->poll); |
5044 | if (do_complete) { | |
d7718a9d | 5045 | io_put_req(req); |
807abcb0 | 5046 | kfree(apoll->double_poll); |
3bfa5bcb JA |
5047 | kfree(apoll); |
5048 | } | |
b1f573bd XW |
5049 | } |
5050 | ||
b41e9852 JA |
5051 | if (do_complete) { |
5052 | io_cqring_fill_event(req, -ECANCELED); | |
5053 | io_commit_cqring(req->ctx); | |
5054 | req->flags |= REQ_F_COMP_LOCKED; | |
f254ac04 | 5055 | req_set_fail_links(req); |
b41e9852 JA |
5056 | io_put_req(req); |
5057 | } | |
5058 | ||
5059 | return do_complete; | |
221c5eb2 JA |
5060 | } |
5061 | ||
76e1b642 JA |
5062 | /* |
5063 | * Returns true if we found and killed one or more poll requests | |
5064 | */ | |
5065 | static bool io_poll_remove_all(struct io_ring_ctx *ctx, struct task_struct *tsk) | |
221c5eb2 | 5066 | { |
78076bb6 | 5067 | struct hlist_node *tmp; |
221c5eb2 | 5068 | struct io_kiocb *req; |
8e2e1faf | 5069 | int posted = 0, i; |
221c5eb2 JA |
5070 | |
5071 | spin_lock_irq(&ctx->completion_lock); | |
78076bb6 JA |
5072 | for (i = 0; i < (1U << ctx->cancel_hash_bits); i++) { |
5073 | struct hlist_head *list; | |
5074 | ||
5075 | list = &ctx->cancel_hash[i]; | |
f3606e3a JA |
5076 | hlist_for_each_entry_safe(req, tmp, list, hash_node) { |
5077 | if (io_task_match(req, tsk)) | |
5078 | posted += io_poll_remove_one(req); | |
5079 | } | |
221c5eb2 JA |
5080 | } |
5081 | spin_unlock_irq(&ctx->completion_lock); | |
b41e9852 | 5082 | |
8e2e1faf JA |
5083 | if (posted) |
5084 | io_cqring_ev_posted(ctx); | |
76e1b642 JA |
5085 | |
5086 | return posted != 0; | |
221c5eb2 JA |
5087 | } |
5088 | ||
47f46768 JA |
5089 | static int io_poll_cancel(struct io_ring_ctx *ctx, __u64 sqe_addr) |
5090 | { | |
78076bb6 | 5091 | struct hlist_head *list; |
47f46768 JA |
5092 | struct io_kiocb *req; |
5093 | ||
78076bb6 JA |
5094 | list = &ctx->cancel_hash[hash_long(sqe_addr, ctx->cancel_hash_bits)]; |
5095 | hlist_for_each_entry(req, list, hash_node) { | |
b41e9852 JA |
5096 | if (sqe_addr != req->user_data) |
5097 | continue; | |
5098 | if (io_poll_remove_one(req)) | |
eac406c6 | 5099 | return 0; |
b41e9852 | 5100 | return -EALREADY; |
47f46768 JA |
5101 | } |
5102 | ||
5103 | return -ENOENT; | |
5104 | } | |
5105 | ||
3529d8c2 JA |
5106 | static int io_poll_remove_prep(struct io_kiocb *req, |
5107 | const struct io_uring_sqe *sqe) | |
0969e783 | 5108 | { |
0969e783 JA |
5109 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
5110 | return -EINVAL; | |
5111 | if (sqe->ioprio || sqe->off || sqe->len || sqe->buf_index || | |
5112 | sqe->poll_events) | |
5113 | return -EINVAL; | |
5114 | ||
5115 | req->poll.addr = READ_ONCE(sqe->addr); | |
0969e783 JA |
5116 | return 0; |
5117 | } | |
5118 | ||
221c5eb2 JA |
5119 | /* |
5120 | * Find a running poll command that matches one specified in sqe->addr, | |
5121 | * and remove it if found. | |
5122 | */ | |
fc4df999 | 5123 | static int io_poll_remove(struct io_kiocb *req) |
221c5eb2 JA |
5124 | { |
5125 | struct io_ring_ctx *ctx = req->ctx; | |
0969e783 | 5126 | u64 addr; |
47f46768 | 5127 | int ret; |
221c5eb2 | 5128 | |
0969e783 | 5129 | addr = req->poll.addr; |
221c5eb2 | 5130 | spin_lock_irq(&ctx->completion_lock); |
0969e783 | 5131 | ret = io_poll_cancel(ctx, addr); |
221c5eb2 JA |
5132 | spin_unlock_irq(&ctx->completion_lock); |
5133 | ||
4e88d6e7 JA |
5134 | if (ret < 0) |
5135 | req_set_fail_links(req); | |
e1e16097 | 5136 | io_req_complete(req, ret); |
221c5eb2 JA |
5137 | return 0; |
5138 | } | |
5139 | ||
221c5eb2 JA |
5140 | static int io_poll_wake(struct wait_queue_entry *wait, unsigned mode, int sync, |
5141 | void *key) | |
5142 | { | |
c2f2eb7d JA |
5143 | struct io_kiocb *req = wait->private; |
5144 | struct io_poll_iocb *poll = &req->poll; | |
221c5eb2 | 5145 | |
d7718a9d | 5146 | return __io_async_wake(req, poll, key_to_poll(key), io_poll_task_func); |
221c5eb2 JA |
5147 | } |
5148 | ||
221c5eb2 JA |
5149 | static void io_poll_queue_proc(struct file *file, struct wait_queue_head *head, |
5150 | struct poll_table_struct *p) | |
5151 | { | |
5152 | struct io_poll_table *pt = container_of(p, struct io_poll_table, pt); | |
5153 | ||
807abcb0 | 5154 | __io_queue_proc(&pt->req->poll, pt, head, (struct io_poll_iocb **) &pt->req->io); |
eac406c6 JA |
5155 | } |
5156 | ||
3529d8c2 | 5157 | static int io_poll_add_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
221c5eb2 JA |
5158 | { |
5159 | struct io_poll_iocb *poll = &req->poll; | |
5769a351 | 5160 | u32 events; |
221c5eb2 JA |
5161 | |
5162 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) | |
5163 | return -EINVAL; | |
5164 | if (sqe->addr || sqe->ioprio || sqe->off || sqe->len || sqe->buf_index) | |
5165 | return -EINVAL; | |
09bb8394 JA |
5166 | if (!poll->file) |
5167 | return -EBADF; | |
221c5eb2 | 5168 | |
5769a351 JX |
5169 | events = READ_ONCE(sqe->poll32_events); |
5170 | #ifdef __BIG_ENDIAN | |
5171 | events = swahw32(events); | |
5172 | #endif | |
a31eb4a2 JX |
5173 | poll->events = demangle_poll(events) | EPOLLERR | EPOLLHUP | |
5174 | (events & EPOLLEXCLUSIVE); | |
0969e783 JA |
5175 | return 0; |
5176 | } | |
5177 | ||
014db007 | 5178 | static int io_poll_add(struct io_kiocb *req) |
0969e783 JA |
5179 | { |
5180 | struct io_poll_iocb *poll = &req->poll; | |
5181 | struct io_ring_ctx *ctx = req->ctx; | |
5182 | struct io_poll_table ipt; | |
0969e783 | 5183 | __poll_t mask; |
0969e783 | 5184 | |
78076bb6 | 5185 | INIT_HLIST_NODE(&req->hash_node); |
d7718a9d | 5186 | ipt.pt._qproc = io_poll_queue_proc; |
36703247 | 5187 | |
d7718a9d JA |
5188 | mask = __io_arm_poll_handler(req, &req->poll, &ipt, poll->events, |
5189 | io_poll_wake); | |
221c5eb2 | 5190 | |
8c838788 | 5191 | if (mask) { /* no async, we'd stolen it */ |
221c5eb2 | 5192 | ipt.error = 0; |
b0dd8a41 | 5193 | io_poll_complete(req, mask, 0); |
221c5eb2 | 5194 | } |
221c5eb2 JA |
5195 | spin_unlock_irq(&ctx->completion_lock); |
5196 | ||
8c838788 JA |
5197 | if (mask) { |
5198 | io_cqring_ev_posted(ctx); | |
014db007 | 5199 | io_put_req(req); |
221c5eb2 | 5200 | } |
8c838788 | 5201 | return ipt.error; |
221c5eb2 JA |
5202 | } |
5203 | ||
5262f567 JA |
5204 | static enum hrtimer_restart io_timeout_fn(struct hrtimer *timer) |
5205 | { | |
ad8a48ac JA |
5206 | struct io_timeout_data *data = container_of(timer, |
5207 | struct io_timeout_data, timer); | |
5208 | struct io_kiocb *req = data->req; | |
5209 | struct io_ring_ctx *ctx = req->ctx; | |
5262f567 JA |
5210 | unsigned long flags; |
5211 | ||
5262f567 | 5212 | spin_lock_irqsave(&ctx->completion_lock, flags); |
01cec8c1 PB |
5213 | atomic_set(&req->ctx->cq_timeouts, |
5214 | atomic_read(&req->ctx->cq_timeouts) + 1); | |
5215 | ||
ef03681a | 5216 | /* |
11365043 JA |
5217 | * We could be racing with timeout deletion. If the list is empty, |
5218 | * then timeout lookup already found it and will be handling it. | |
ef03681a | 5219 | */ |
135fcde8 PB |
5220 | if (!list_empty(&req->timeout.list)) |
5221 | list_del_init(&req->timeout.list); | |
5262f567 | 5222 | |
78e19bbe | 5223 | io_cqring_fill_event(req, -ETIME); |
5262f567 JA |
5224 | io_commit_cqring(ctx); |
5225 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
5226 | ||
5227 | io_cqring_ev_posted(ctx); | |
4e88d6e7 | 5228 | req_set_fail_links(req); |
5262f567 JA |
5229 | io_put_req(req); |
5230 | return HRTIMER_NORESTART; | |
5231 | } | |
5232 | ||
f254ac04 JA |
5233 | static int __io_timeout_cancel(struct io_kiocb *req) |
5234 | { | |
5235 | int ret; | |
5236 | ||
5237 | list_del_init(&req->timeout.list); | |
5238 | ||
5239 | ret = hrtimer_try_to_cancel(&req->io->timeout.timer); | |
5240 | if (ret == -1) | |
5241 | return -EALREADY; | |
5242 | ||
5243 | req_set_fail_links(req); | |
5244 | req->flags |= REQ_F_COMP_LOCKED; | |
5245 | io_cqring_fill_event(req, -ECANCELED); | |
5246 | io_put_req(req); | |
5247 | return 0; | |
5248 | } | |
5249 | ||
47f46768 JA |
5250 | static int io_timeout_cancel(struct io_ring_ctx *ctx, __u64 user_data) |
5251 | { | |
5252 | struct io_kiocb *req; | |
5253 | int ret = -ENOENT; | |
5254 | ||
135fcde8 | 5255 | list_for_each_entry(req, &ctx->timeout_list, timeout.list) { |
47f46768 | 5256 | if (user_data == req->user_data) { |
47f46768 JA |
5257 | ret = 0; |
5258 | break; | |
5259 | } | |
5260 | } | |
5261 | ||
5262 | if (ret == -ENOENT) | |
5263 | return ret; | |
5264 | ||
f254ac04 | 5265 | return __io_timeout_cancel(req); |
47f46768 JA |
5266 | } |
5267 | ||
3529d8c2 JA |
5268 | static int io_timeout_remove_prep(struct io_kiocb *req, |
5269 | const struct io_uring_sqe *sqe) | |
b29472ee | 5270 | { |
b29472ee JA |
5271 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
5272 | return -EINVAL; | |
61710e43 DA |
5273 | if (unlikely(req->flags & (REQ_F_FIXED_FILE | REQ_F_BUFFER_SELECT))) |
5274 | return -EINVAL; | |
5275 | if (sqe->ioprio || sqe->buf_index || sqe->len) | |
b29472ee JA |
5276 | return -EINVAL; |
5277 | ||
5278 | req->timeout.addr = READ_ONCE(sqe->addr); | |
5279 | req->timeout.flags = READ_ONCE(sqe->timeout_flags); | |
5280 | if (req->timeout.flags) | |
5281 | return -EINVAL; | |
5282 | ||
b29472ee JA |
5283 | return 0; |
5284 | } | |
5285 | ||
11365043 JA |
5286 | /* |
5287 | * Remove or update an existing timeout command | |
5288 | */ | |
fc4df999 | 5289 | static int io_timeout_remove(struct io_kiocb *req) |
11365043 JA |
5290 | { |
5291 | struct io_ring_ctx *ctx = req->ctx; | |
47f46768 | 5292 | int ret; |
11365043 | 5293 | |
11365043 | 5294 | spin_lock_irq(&ctx->completion_lock); |
b29472ee | 5295 | ret = io_timeout_cancel(ctx, req->timeout.addr); |
11365043 | 5296 | |
47f46768 | 5297 | io_cqring_fill_event(req, ret); |
11365043 JA |
5298 | io_commit_cqring(ctx); |
5299 | spin_unlock_irq(&ctx->completion_lock); | |
5262f567 | 5300 | io_cqring_ev_posted(ctx); |
4e88d6e7 JA |
5301 | if (ret < 0) |
5302 | req_set_fail_links(req); | |
ec9c02ad | 5303 | io_put_req(req); |
11365043 | 5304 | return 0; |
5262f567 JA |
5305 | } |
5306 | ||
3529d8c2 | 5307 | static int io_timeout_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe, |
2d28390a | 5308 | bool is_timeout_link) |
5262f567 | 5309 | { |
ad8a48ac | 5310 | struct io_timeout_data *data; |
a41525ab | 5311 | unsigned flags; |
56080b02 | 5312 | u32 off = READ_ONCE(sqe->off); |
5262f567 | 5313 | |
ad8a48ac | 5314 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
5262f567 | 5315 | return -EINVAL; |
ad8a48ac | 5316 | if (sqe->ioprio || sqe->buf_index || sqe->len != 1) |
a41525ab | 5317 | return -EINVAL; |
56080b02 | 5318 | if (off && is_timeout_link) |
2d28390a | 5319 | return -EINVAL; |
a41525ab JA |
5320 | flags = READ_ONCE(sqe->timeout_flags); |
5321 | if (flags & ~IORING_TIMEOUT_ABS) | |
5262f567 | 5322 | return -EINVAL; |
bdf20073 | 5323 | |
bfe68a22 | 5324 | req->timeout.off = off; |
26a61679 | 5325 | |
3529d8c2 | 5326 | if (!req->io && io_alloc_async_ctx(req)) |
26a61679 JA |
5327 | return -ENOMEM; |
5328 | ||
5329 | data = &req->io->timeout; | |
ad8a48ac | 5330 | data->req = req; |
ad8a48ac JA |
5331 | |
5332 | if (get_timespec64(&data->ts, u64_to_user_ptr(sqe->addr))) | |
5262f567 JA |
5333 | return -EFAULT; |
5334 | ||
11365043 | 5335 | if (flags & IORING_TIMEOUT_ABS) |
ad8a48ac | 5336 | data->mode = HRTIMER_MODE_ABS; |
11365043 | 5337 | else |
ad8a48ac | 5338 | data->mode = HRTIMER_MODE_REL; |
11365043 | 5339 | |
ad8a48ac JA |
5340 | hrtimer_init(&data->timer, CLOCK_MONOTONIC, data->mode); |
5341 | return 0; | |
5342 | } | |
5343 | ||
fc4df999 | 5344 | static int io_timeout(struct io_kiocb *req) |
ad8a48ac | 5345 | { |
ad8a48ac | 5346 | struct io_ring_ctx *ctx = req->ctx; |
bfe68a22 | 5347 | struct io_timeout_data *data = &req->io->timeout; |
ad8a48ac | 5348 | struct list_head *entry; |
bfe68a22 | 5349 | u32 tail, off = req->timeout.off; |
ad8a48ac | 5350 | |
733f5c95 | 5351 | spin_lock_irq(&ctx->completion_lock); |
93bd25bb | 5352 | |
5262f567 JA |
5353 | /* |
5354 | * sqe->off holds how many events that need to occur for this | |
93bd25bb JA |
5355 | * timeout event to be satisfied. If it isn't set, then this is |
5356 | * a pure timeout request, sequence isn't used. | |
5262f567 | 5357 | */ |
8eb7e2d0 | 5358 | if (io_is_timeout_noseq(req)) { |
93bd25bb JA |
5359 | entry = ctx->timeout_list.prev; |
5360 | goto add; | |
5361 | } | |
5262f567 | 5362 | |
bfe68a22 PB |
5363 | tail = ctx->cached_cq_tail - atomic_read(&ctx->cq_timeouts); |
5364 | req->timeout.target_seq = tail + off; | |
5262f567 JA |
5365 | |
5366 | /* | |
5367 | * Insertion sort, ensuring the first entry in the list is always | |
5368 | * the one we need first. | |
5369 | */ | |
5262f567 | 5370 | list_for_each_prev(entry, &ctx->timeout_list) { |
135fcde8 PB |
5371 | struct io_kiocb *nxt = list_entry(entry, struct io_kiocb, |
5372 | timeout.list); | |
5262f567 | 5373 | |
8eb7e2d0 | 5374 | if (io_is_timeout_noseq(nxt)) |
93bd25bb | 5375 | continue; |
bfe68a22 PB |
5376 | /* nxt.seq is behind @tail, otherwise would've been completed */ |
5377 | if (off >= nxt->timeout.target_seq - tail) | |
5262f567 JA |
5378 | break; |
5379 | } | |
93bd25bb | 5380 | add: |
135fcde8 | 5381 | list_add(&req->timeout.list, entry); |
ad8a48ac JA |
5382 | data->timer.function = io_timeout_fn; |
5383 | hrtimer_start(&data->timer, timespec64_to_ktime(data->ts), data->mode); | |
5262f567 | 5384 | spin_unlock_irq(&ctx->completion_lock); |
5262f567 JA |
5385 | return 0; |
5386 | } | |
5262f567 | 5387 | |
62755e35 JA |
5388 | static bool io_cancel_cb(struct io_wq_work *work, void *data) |
5389 | { | |
5390 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); | |
5391 | ||
5392 | return req->user_data == (unsigned long) data; | |
5393 | } | |
5394 | ||
e977d6d3 | 5395 | static int io_async_cancel_one(struct io_ring_ctx *ctx, void *sqe_addr) |
62755e35 | 5396 | { |
62755e35 | 5397 | enum io_wq_cancel cancel_ret; |
62755e35 JA |
5398 | int ret = 0; |
5399 | ||
4f26bda1 | 5400 | cancel_ret = io_wq_cancel_cb(ctx->io_wq, io_cancel_cb, sqe_addr, false); |
62755e35 JA |
5401 | switch (cancel_ret) { |
5402 | case IO_WQ_CANCEL_OK: | |
5403 | ret = 0; | |
5404 | break; | |
5405 | case IO_WQ_CANCEL_RUNNING: | |
5406 | ret = -EALREADY; | |
5407 | break; | |
5408 | case IO_WQ_CANCEL_NOTFOUND: | |
5409 | ret = -ENOENT; | |
5410 | break; | |
5411 | } | |
5412 | ||
e977d6d3 JA |
5413 | return ret; |
5414 | } | |
5415 | ||
47f46768 JA |
5416 | static void io_async_find_and_cancel(struct io_ring_ctx *ctx, |
5417 | struct io_kiocb *req, __u64 sqe_addr, | |
014db007 | 5418 | int success_ret) |
47f46768 JA |
5419 | { |
5420 | unsigned long flags; | |
5421 | int ret; | |
5422 | ||
5423 | ret = io_async_cancel_one(ctx, (void *) (unsigned long) sqe_addr); | |
5424 | if (ret != -ENOENT) { | |
5425 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
5426 | goto done; | |
5427 | } | |
5428 | ||
5429 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
5430 | ret = io_timeout_cancel(ctx, sqe_addr); | |
5431 | if (ret != -ENOENT) | |
5432 | goto done; | |
5433 | ret = io_poll_cancel(ctx, sqe_addr); | |
5434 | done: | |
b0dd8a41 JA |
5435 | if (!ret) |
5436 | ret = success_ret; | |
47f46768 JA |
5437 | io_cqring_fill_event(req, ret); |
5438 | io_commit_cqring(ctx); | |
5439 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
5440 | io_cqring_ev_posted(ctx); | |
5441 | ||
4e88d6e7 JA |
5442 | if (ret < 0) |
5443 | req_set_fail_links(req); | |
014db007 | 5444 | io_put_req(req); |
47f46768 JA |
5445 | } |
5446 | ||
3529d8c2 JA |
5447 | static int io_async_cancel_prep(struct io_kiocb *req, |
5448 | const struct io_uring_sqe *sqe) | |
e977d6d3 | 5449 | { |
fbf23849 | 5450 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
e977d6d3 | 5451 | return -EINVAL; |
61710e43 DA |
5452 | if (unlikely(req->flags & (REQ_F_FIXED_FILE | REQ_F_BUFFER_SELECT))) |
5453 | return -EINVAL; | |
5454 | if (sqe->ioprio || sqe->off || sqe->len || sqe->cancel_flags) | |
e977d6d3 JA |
5455 | return -EINVAL; |
5456 | ||
fbf23849 JA |
5457 | req->cancel.addr = READ_ONCE(sqe->addr); |
5458 | return 0; | |
5459 | } | |
5460 | ||
014db007 | 5461 | static int io_async_cancel(struct io_kiocb *req) |
fbf23849 JA |
5462 | { |
5463 | struct io_ring_ctx *ctx = req->ctx; | |
fbf23849 | 5464 | |
014db007 | 5465 | io_async_find_and_cancel(ctx, req, req->cancel.addr, 0); |
5262f567 JA |
5466 | return 0; |
5467 | } | |
5468 | ||
05f3fb3c JA |
5469 | static int io_files_update_prep(struct io_kiocb *req, |
5470 | const struct io_uring_sqe *sqe) | |
5471 | { | |
6ca56f84 JA |
5472 | if (unlikely(req->ctx->flags & IORING_SETUP_SQPOLL)) |
5473 | return -EINVAL; | |
61710e43 DA |
5474 | if (unlikely(req->flags & (REQ_F_FIXED_FILE | REQ_F_BUFFER_SELECT))) |
5475 | return -EINVAL; | |
5476 | if (sqe->ioprio || sqe->rw_flags) | |
05f3fb3c JA |
5477 | return -EINVAL; |
5478 | ||
5479 | req->files_update.offset = READ_ONCE(sqe->off); | |
5480 | req->files_update.nr_args = READ_ONCE(sqe->len); | |
5481 | if (!req->files_update.nr_args) | |
5482 | return -EINVAL; | |
5483 | req->files_update.arg = READ_ONCE(sqe->addr); | |
5484 | return 0; | |
5485 | } | |
5486 | ||
229a7b63 JA |
5487 | static int io_files_update(struct io_kiocb *req, bool force_nonblock, |
5488 | struct io_comp_state *cs) | |
fbf23849 JA |
5489 | { |
5490 | struct io_ring_ctx *ctx = req->ctx; | |
05f3fb3c JA |
5491 | struct io_uring_files_update up; |
5492 | int ret; | |
fbf23849 | 5493 | |
f86cd20c | 5494 | if (force_nonblock) |
05f3fb3c | 5495 | return -EAGAIN; |
05f3fb3c JA |
5496 | |
5497 | up.offset = req->files_update.offset; | |
5498 | up.fds = req->files_update.arg; | |
5499 | ||
5500 | mutex_lock(&ctx->uring_lock); | |
5501 | ret = __io_sqe_files_update(ctx, &up, req->files_update.nr_args); | |
5502 | mutex_unlock(&ctx->uring_lock); | |
5503 | ||
5504 | if (ret < 0) | |
5505 | req_set_fail_links(req); | |
229a7b63 | 5506 | __io_req_complete(req, ret, 0, cs); |
5262f567 JA |
5507 | return 0; |
5508 | } | |
5509 | ||
3529d8c2 JA |
5510 | static int io_req_defer_prep(struct io_kiocb *req, |
5511 | const struct io_uring_sqe *sqe) | |
f67676d1 | 5512 | { |
e781573e | 5513 | ssize_t ret = 0; |
f67676d1 | 5514 | |
f1d96a8f PB |
5515 | if (!sqe) |
5516 | return 0; | |
5517 | ||
327d6d96 PB |
5518 | if (io_alloc_async_ctx(req)) |
5519 | return -EAGAIN; | |
f56040b8 PB |
5520 | ret = io_prep_work_files(req); |
5521 | if (unlikely(ret)) | |
5522 | return ret; | |
cccf0ee8 | 5523 | |
202700e1 JA |
5524 | io_prep_async_work(req); |
5525 | ||
d625c6ee | 5526 | switch (req->opcode) { |
e781573e JA |
5527 | case IORING_OP_NOP: |
5528 | break; | |
f67676d1 JA |
5529 | case IORING_OP_READV: |
5530 | case IORING_OP_READ_FIXED: | |
3a6820f2 | 5531 | case IORING_OP_READ: |
3529d8c2 | 5532 | ret = io_read_prep(req, sqe, true); |
f67676d1 JA |
5533 | break; |
5534 | case IORING_OP_WRITEV: | |
5535 | case IORING_OP_WRITE_FIXED: | |
3a6820f2 | 5536 | case IORING_OP_WRITE: |
3529d8c2 | 5537 | ret = io_write_prep(req, sqe, true); |
f67676d1 | 5538 | break; |
0969e783 | 5539 | case IORING_OP_POLL_ADD: |
3529d8c2 | 5540 | ret = io_poll_add_prep(req, sqe); |
0969e783 JA |
5541 | break; |
5542 | case IORING_OP_POLL_REMOVE: | |
3529d8c2 | 5543 | ret = io_poll_remove_prep(req, sqe); |
0969e783 | 5544 | break; |
8ed8d3c3 | 5545 | case IORING_OP_FSYNC: |
3529d8c2 | 5546 | ret = io_prep_fsync(req, sqe); |
8ed8d3c3 JA |
5547 | break; |
5548 | case IORING_OP_SYNC_FILE_RANGE: | |
3529d8c2 | 5549 | ret = io_prep_sfr(req, sqe); |
8ed8d3c3 | 5550 | break; |
03b1230c | 5551 | case IORING_OP_SENDMSG: |
fddaface | 5552 | case IORING_OP_SEND: |
3529d8c2 | 5553 | ret = io_sendmsg_prep(req, sqe); |
03b1230c JA |
5554 | break; |
5555 | case IORING_OP_RECVMSG: | |
fddaface | 5556 | case IORING_OP_RECV: |
3529d8c2 | 5557 | ret = io_recvmsg_prep(req, sqe); |
03b1230c | 5558 | break; |
f499a021 | 5559 | case IORING_OP_CONNECT: |
3529d8c2 | 5560 | ret = io_connect_prep(req, sqe); |
f499a021 | 5561 | break; |
2d28390a | 5562 | case IORING_OP_TIMEOUT: |
3529d8c2 | 5563 | ret = io_timeout_prep(req, sqe, false); |
b7bb4f7d | 5564 | break; |
b29472ee | 5565 | case IORING_OP_TIMEOUT_REMOVE: |
3529d8c2 | 5566 | ret = io_timeout_remove_prep(req, sqe); |
b29472ee | 5567 | break; |
fbf23849 | 5568 | case IORING_OP_ASYNC_CANCEL: |
3529d8c2 | 5569 | ret = io_async_cancel_prep(req, sqe); |
fbf23849 | 5570 | break; |
2d28390a | 5571 | case IORING_OP_LINK_TIMEOUT: |
3529d8c2 | 5572 | ret = io_timeout_prep(req, sqe, true); |
b7bb4f7d | 5573 | break; |
8ed8d3c3 | 5574 | case IORING_OP_ACCEPT: |
3529d8c2 | 5575 | ret = io_accept_prep(req, sqe); |
8ed8d3c3 | 5576 | break; |
d63d1b5e JA |
5577 | case IORING_OP_FALLOCATE: |
5578 | ret = io_fallocate_prep(req, sqe); | |
5579 | break; | |
15b71abe JA |
5580 | case IORING_OP_OPENAT: |
5581 | ret = io_openat_prep(req, sqe); | |
5582 | break; | |
b5dba59e JA |
5583 | case IORING_OP_CLOSE: |
5584 | ret = io_close_prep(req, sqe); | |
5585 | break; | |
05f3fb3c JA |
5586 | case IORING_OP_FILES_UPDATE: |
5587 | ret = io_files_update_prep(req, sqe); | |
5588 | break; | |
eddc7ef5 JA |
5589 | case IORING_OP_STATX: |
5590 | ret = io_statx_prep(req, sqe); | |
5591 | break; | |
4840e418 JA |
5592 | case IORING_OP_FADVISE: |
5593 | ret = io_fadvise_prep(req, sqe); | |
5594 | break; | |
c1ca757b JA |
5595 | case IORING_OP_MADVISE: |
5596 | ret = io_madvise_prep(req, sqe); | |
5597 | break; | |
cebdb986 JA |
5598 | case IORING_OP_OPENAT2: |
5599 | ret = io_openat2_prep(req, sqe); | |
5600 | break; | |
3e4827b0 JA |
5601 | case IORING_OP_EPOLL_CTL: |
5602 | ret = io_epoll_ctl_prep(req, sqe); | |
5603 | break; | |
7d67af2c PB |
5604 | case IORING_OP_SPLICE: |
5605 | ret = io_splice_prep(req, sqe); | |
5606 | break; | |
ddf0322d JA |
5607 | case IORING_OP_PROVIDE_BUFFERS: |
5608 | ret = io_provide_buffers_prep(req, sqe); | |
5609 | break; | |
067524e9 JA |
5610 | case IORING_OP_REMOVE_BUFFERS: |
5611 | ret = io_remove_buffers_prep(req, sqe); | |
5612 | break; | |
f2a8d5c7 PB |
5613 | case IORING_OP_TEE: |
5614 | ret = io_tee_prep(req, sqe); | |
5615 | break; | |
f67676d1 | 5616 | default: |
e781573e JA |
5617 | printk_once(KERN_WARNING "io_uring: unhandled opcode %d\n", |
5618 | req->opcode); | |
5619 | ret = -EINVAL; | |
b7bb4f7d | 5620 | break; |
f67676d1 JA |
5621 | } |
5622 | ||
b7bb4f7d | 5623 | return ret; |
f67676d1 JA |
5624 | } |
5625 | ||
9cf7c104 PB |
5626 | static u32 io_get_sequence(struct io_kiocb *req) |
5627 | { | |
5628 | struct io_kiocb *pos; | |
5629 | struct io_ring_ctx *ctx = req->ctx; | |
5630 | u32 total_submitted, nr_reqs = 1; | |
5631 | ||
5632 | if (req->flags & REQ_F_LINK_HEAD) | |
5633 | list_for_each_entry(pos, &req->link_list, link_list) | |
5634 | nr_reqs++; | |
5635 | ||
5636 | total_submitted = ctx->cached_sq_head - ctx->cached_sq_dropped; | |
5637 | return total_submitted - nr_reqs; | |
5638 | } | |
5639 | ||
3529d8c2 | 5640 | static int io_req_defer(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
de0617e4 | 5641 | { |
a197f664 | 5642 | struct io_ring_ctx *ctx = req->ctx; |
27dc8338 | 5643 | struct io_defer_entry *de; |
f67676d1 | 5644 | int ret; |
9cf7c104 | 5645 | u32 seq; |
de0617e4 | 5646 | |
9d858b21 | 5647 | /* Still need defer if there is pending req in defer list. */ |
9cf7c104 PB |
5648 | if (likely(list_empty_careful(&ctx->defer_list) && |
5649 | !(req->flags & REQ_F_IO_DRAIN))) | |
5650 | return 0; | |
5651 | ||
5652 | seq = io_get_sequence(req); | |
5653 | /* Still a chance to pass the sequence check */ | |
5654 | if (!req_need_defer(req, seq) && list_empty_careful(&ctx->defer_list)) | |
de0617e4 JA |
5655 | return 0; |
5656 | ||
650b5481 | 5657 | if (!req->io) { |
650b5481 | 5658 | ret = io_req_defer_prep(req, sqe); |
327d6d96 | 5659 | if (ret) |
650b5481 PB |
5660 | return ret; |
5661 | } | |
cbdcb435 | 5662 | io_prep_async_link(req); |
27dc8338 PB |
5663 | de = kmalloc(sizeof(*de), GFP_KERNEL); |
5664 | if (!de) | |
5665 | return -ENOMEM; | |
2d28390a | 5666 | |
de0617e4 | 5667 | spin_lock_irq(&ctx->completion_lock); |
9cf7c104 | 5668 | if (!req_need_defer(req, seq) && list_empty(&ctx->defer_list)) { |
de0617e4 | 5669 | spin_unlock_irq(&ctx->completion_lock); |
27dc8338 | 5670 | kfree(de); |
ae34817b PB |
5671 | io_queue_async_work(req); |
5672 | return -EIOCBQUEUED; | |
de0617e4 JA |
5673 | } |
5674 | ||
915967f6 | 5675 | trace_io_uring_defer(ctx, req, req->user_data); |
27dc8338 | 5676 | de->req = req; |
9cf7c104 | 5677 | de->seq = seq; |
27dc8338 | 5678 | list_add_tail(&de->list, &ctx->defer_list); |
de0617e4 JA |
5679 | spin_unlock_irq(&ctx->completion_lock); |
5680 | return -EIOCBQUEUED; | |
5681 | } | |
5682 | ||
f573d384 JA |
5683 | static void io_req_drop_files(struct io_kiocb *req) |
5684 | { | |
5685 | struct io_ring_ctx *ctx = req->ctx; | |
5686 | unsigned long flags; | |
5687 | ||
5688 | spin_lock_irqsave(&ctx->inflight_lock, flags); | |
5689 | list_del(&req->inflight_entry); | |
5690 | if (waitqueue_active(&ctx->inflight_wait)) | |
5691 | wake_up(&ctx->inflight_wait); | |
5692 | spin_unlock_irqrestore(&ctx->inflight_lock, flags); | |
5693 | req->flags &= ~REQ_F_INFLIGHT; | |
0f212204 | 5694 | put_files_struct(req->work.files); |
9b828492 | 5695 | put_nsproxy(req->work.nsproxy); |
f573d384 JA |
5696 | req->work.files = NULL; |
5697 | } | |
5698 | ||
3ca405eb | 5699 | static void __io_clean_op(struct io_kiocb *req) |
99bc4c38 PB |
5700 | { |
5701 | struct io_async_ctx *io = req->io; | |
5702 | ||
0e1b6fe3 PB |
5703 | if (req->flags & REQ_F_BUFFER_SELECTED) { |
5704 | switch (req->opcode) { | |
5705 | case IORING_OP_READV: | |
5706 | case IORING_OP_READ_FIXED: | |
5707 | case IORING_OP_READ: | |
bcda7baa | 5708 | kfree((void *)(unsigned long)req->rw.addr); |
0e1b6fe3 PB |
5709 | break; |
5710 | case IORING_OP_RECVMSG: | |
5711 | case IORING_OP_RECV: | |
bcda7baa | 5712 | kfree(req->sr_msg.kbuf); |
0e1b6fe3 PB |
5713 | break; |
5714 | } | |
5715 | req->flags &= ~REQ_F_BUFFER_SELECTED; | |
99bc4c38 PB |
5716 | } |
5717 | ||
0e1b6fe3 PB |
5718 | if (req->flags & REQ_F_NEED_CLEANUP) { |
5719 | switch (req->opcode) { | |
5720 | case IORING_OP_READV: | |
5721 | case IORING_OP_READ_FIXED: | |
5722 | case IORING_OP_READ: | |
5723 | case IORING_OP_WRITEV: | |
5724 | case IORING_OP_WRITE_FIXED: | |
5725 | case IORING_OP_WRITE: | |
ff6165b2 JA |
5726 | if (io->rw.free_iovec) |
5727 | kfree(io->rw.free_iovec); | |
0e1b6fe3 PB |
5728 | break; |
5729 | case IORING_OP_RECVMSG: | |
5730 | case IORING_OP_SENDMSG: | |
5731 | if (io->msg.iov != io->msg.fast_iov) | |
5732 | kfree(io->msg.iov); | |
5733 | break; | |
5734 | case IORING_OP_SPLICE: | |
5735 | case IORING_OP_TEE: | |
5736 | io_put_file(req, req->splice.file_in, | |
5737 | (req->splice.flags & SPLICE_F_FD_IN_FIXED)); | |
5738 | break; | |
f3cd4850 JA |
5739 | case IORING_OP_OPENAT: |
5740 | case IORING_OP_OPENAT2: | |
5741 | if (req->open.filename) | |
5742 | putname(req->open.filename); | |
5743 | break; | |
0e1b6fe3 PB |
5744 | } |
5745 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
99bc4c38 | 5746 | } |
bb175342 | 5747 | |
f573d384 JA |
5748 | if (req->flags & REQ_F_INFLIGHT) |
5749 | io_req_drop_files(req); | |
99bc4c38 PB |
5750 | } |
5751 | ||
3529d8c2 | 5752 | static int io_issue_sqe(struct io_kiocb *req, const struct io_uring_sqe *sqe, |
f13fad7b | 5753 | bool force_nonblock, struct io_comp_state *cs) |
2b188cc1 | 5754 | { |
a197f664 | 5755 | struct io_ring_ctx *ctx = req->ctx; |
d625c6ee | 5756 | int ret; |
2b188cc1 | 5757 | |
d625c6ee | 5758 | switch (req->opcode) { |
2b188cc1 | 5759 | case IORING_OP_NOP: |
229a7b63 | 5760 | ret = io_nop(req, cs); |
2b188cc1 JA |
5761 | break; |
5762 | case IORING_OP_READV: | |
edafccee | 5763 | case IORING_OP_READ_FIXED: |
3a6820f2 | 5764 | case IORING_OP_READ: |
3529d8c2 JA |
5765 | if (sqe) { |
5766 | ret = io_read_prep(req, sqe, force_nonblock); | |
5767 | if (ret < 0) | |
5768 | break; | |
5769 | } | |
a1d7c393 | 5770 | ret = io_read(req, force_nonblock, cs); |
edafccee | 5771 | break; |
3529d8c2 | 5772 | case IORING_OP_WRITEV: |
edafccee | 5773 | case IORING_OP_WRITE_FIXED: |
3a6820f2 | 5774 | case IORING_OP_WRITE: |
3529d8c2 JA |
5775 | if (sqe) { |
5776 | ret = io_write_prep(req, sqe, force_nonblock); | |
5777 | if (ret < 0) | |
5778 | break; | |
5779 | } | |
a1d7c393 | 5780 | ret = io_write(req, force_nonblock, cs); |
2b188cc1 | 5781 | break; |
c992fe29 | 5782 | case IORING_OP_FSYNC: |
3529d8c2 JA |
5783 | if (sqe) { |
5784 | ret = io_prep_fsync(req, sqe); | |
5785 | if (ret < 0) | |
5786 | break; | |
5787 | } | |
014db007 | 5788 | ret = io_fsync(req, force_nonblock); |
c992fe29 | 5789 | break; |
221c5eb2 | 5790 | case IORING_OP_POLL_ADD: |
3529d8c2 JA |
5791 | if (sqe) { |
5792 | ret = io_poll_add_prep(req, sqe); | |
5793 | if (ret) | |
5794 | break; | |
5795 | } | |
014db007 | 5796 | ret = io_poll_add(req); |
221c5eb2 JA |
5797 | break; |
5798 | case IORING_OP_POLL_REMOVE: | |
3529d8c2 JA |
5799 | if (sqe) { |
5800 | ret = io_poll_remove_prep(req, sqe); | |
5801 | if (ret < 0) | |
5802 | break; | |
5803 | } | |
fc4df999 | 5804 | ret = io_poll_remove(req); |
221c5eb2 | 5805 | break; |
5d17b4a4 | 5806 | case IORING_OP_SYNC_FILE_RANGE: |
3529d8c2 JA |
5807 | if (sqe) { |
5808 | ret = io_prep_sfr(req, sqe); | |
5809 | if (ret < 0) | |
5810 | break; | |
5811 | } | |
014db007 | 5812 | ret = io_sync_file_range(req, force_nonblock); |
5d17b4a4 | 5813 | break; |
0fa03c62 | 5814 | case IORING_OP_SENDMSG: |
fddaface | 5815 | case IORING_OP_SEND: |
3529d8c2 JA |
5816 | if (sqe) { |
5817 | ret = io_sendmsg_prep(req, sqe); | |
5818 | if (ret < 0) | |
5819 | break; | |
5820 | } | |
fddaface | 5821 | if (req->opcode == IORING_OP_SENDMSG) |
229a7b63 | 5822 | ret = io_sendmsg(req, force_nonblock, cs); |
fddaface | 5823 | else |
229a7b63 | 5824 | ret = io_send(req, force_nonblock, cs); |
0fa03c62 | 5825 | break; |
aa1fa28f | 5826 | case IORING_OP_RECVMSG: |
fddaface | 5827 | case IORING_OP_RECV: |
3529d8c2 JA |
5828 | if (sqe) { |
5829 | ret = io_recvmsg_prep(req, sqe); | |
5830 | if (ret) | |
5831 | break; | |
5832 | } | |
fddaface | 5833 | if (req->opcode == IORING_OP_RECVMSG) |
229a7b63 | 5834 | ret = io_recvmsg(req, force_nonblock, cs); |
fddaface | 5835 | else |
229a7b63 | 5836 | ret = io_recv(req, force_nonblock, cs); |
aa1fa28f | 5837 | break; |
5262f567 | 5838 | case IORING_OP_TIMEOUT: |
3529d8c2 JA |
5839 | if (sqe) { |
5840 | ret = io_timeout_prep(req, sqe, false); | |
5841 | if (ret) | |
5842 | break; | |
5843 | } | |
fc4df999 | 5844 | ret = io_timeout(req); |
5262f567 | 5845 | break; |
11365043 | 5846 | case IORING_OP_TIMEOUT_REMOVE: |
3529d8c2 JA |
5847 | if (sqe) { |
5848 | ret = io_timeout_remove_prep(req, sqe); | |
5849 | if (ret) | |
5850 | break; | |
5851 | } | |
fc4df999 | 5852 | ret = io_timeout_remove(req); |
11365043 | 5853 | break; |
17f2fe35 | 5854 | case IORING_OP_ACCEPT: |
3529d8c2 JA |
5855 | if (sqe) { |
5856 | ret = io_accept_prep(req, sqe); | |
5857 | if (ret) | |
5858 | break; | |
5859 | } | |
229a7b63 | 5860 | ret = io_accept(req, force_nonblock, cs); |
17f2fe35 | 5861 | break; |
f8e85cf2 | 5862 | case IORING_OP_CONNECT: |
3529d8c2 JA |
5863 | if (sqe) { |
5864 | ret = io_connect_prep(req, sqe); | |
5865 | if (ret) | |
5866 | break; | |
5867 | } | |
229a7b63 | 5868 | ret = io_connect(req, force_nonblock, cs); |
f8e85cf2 | 5869 | break; |
62755e35 | 5870 | case IORING_OP_ASYNC_CANCEL: |
3529d8c2 JA |
5871 | if (sqe) { |
5872 | ret = io_async_cancel_prep(req, sqe); | |
5873 | if (ret) | |
5874 | break; | |
5875 | } | |
014db007 | 5876 | ret = io_async_cancel(req); |
62755e35 | 5877 | break; |
d63d1b5e JA |
5878 | case IORING_OP_FALLOCATE: |
5879 | if (sqe) { | |
5880 | ret = io_fallocate_prep(req, sqe); | |
5881 | if (ret) | |
5882 | break; | |
5883 | } | |
014db007 | 5884 | ret = io_fallocate(req, force_nonblock); |
d63d1b5e | 5885 | break; |
15b71abe JA |
5886 | case IORING_OP_OPENAT: |
5887 | if (sqe) { | |
5888 | ret = io_openat_prep(req, sqe); | |
5889 | if (ret) | |
5890 | break; | |
5891 | } | |
014db007 | 5892 | ret = io_openat(req, force_nonblock); |
15b71abe | 5893 | break; |
b5dba59e JA |
5894 | case IORING_OP_CLOSE: |
5895 | if (sqe) { | |
5896 | ret = io_close_prep(req, sqe); | |
5897 | if (ret) | |
5898 | break; | |
5899 | } | |
229a7b63 | 5900 | ret = io_close(req, force_nonblock, cs); |
b5dba59e | 5901 | break; |
05f3fb3c JA |
5902 | case IORING_OP_FILES_UPDATE: |
5903 | if (sqe) { | |
5904 | ret = io_files_update_prep(req, sqe); | |
5905 | if (ret) | |
5906 | break; | |
5907 | } | |
229a7b63 | 5908 | ret = io_files_update(req, force_nonblock, cs); |
05f3fb3c | 5909 | break; |
eddc7ef5 JA |
5910 | case IORING_OP_STATX: |
5911 | if (sqe) { | |
5912 | ret = io_statx_prep(req, sqe); | |
5913 | if (ret) | |
5914 | break; | |
5915 | } | |
014db007 | 5916 | ret = io_statx(req, force_nonblock); |
eddc7ef5 | 5917 | break; |
4840e418 JA |
5918 | case IORING_OP_FADVISE: |
5919 | if (sqe) { | |
5920 | ret = io_fadvise_prep(req, sqe); | |
5921 | if (ret) | |
5922 | break; | |
5923 | } | |
014db007 | 5924 | ret = io_fadvise(req, force_nonblock); |
4840e418 | 5925 | break; |
c1ca757b JA |
5926 | case IORING_OP_MADVISE: |
5927 | if (sqe) { | |
5928 | ret = io_madvise_prep(req, sqe); | |
5929 | if (ret) | |
5930 | break; | |
5931 | } | |
014db007 | 5932 | ret = io_madvise(req, force_nonblock); |
c1ca757b | 5933 | break; |
cebdb986 JA |
5934 | case IORING_OP_OPENAT2: |
5935 | if (sqe) { | |
5936 | ret = io_openat2_prep(req, sqe); | |
5937 | if (ret) | |
5938 | break; | |
5939 | } | |
014db007 | 5940 | ret = io_openat2(req, force_nonblock); |
cebdb986 | 5941 | break; |
3e4827b0 JA |
5942 | case IORING_OP_EPOLL_CTL: |
5943 | if (sqe) { | |
5944 | ret = io_epoll_ctl_prep(req, sqe); | |
5945 | if (ret) | |
5946 | break; | |
5947 | } | |
229a7b63 | 5948 | ret = io_epoll_ctl(req, force_nonblock, cs); |
3e4827b0 | 5949 | break; |
7d67af2c PB |
5950 | case IORING_OP_SPLICE: |
5951 | if (sqe) { | |
5952 | ret = io_splice_prep(req, sqe); | |
5953 | if (ret < 0) | |
5954 | break; | |
5955 | } | |
014db007 | 5956 | ret = io_splice(req, force_nonblock); |
7d67af2c | 5957 | break; |
ddf0322d JA |
5958 | case IORING_OP_PROVIDE_BUFFERS: |
5959 | if (sqe) { | |
5960 | ret = io_provide_buffers_prep(req, sqe); | |
5961 | if (ret) | |
5962 | break; | |
5963 | } | |
229a7b63 | 5964 | ret = io_provide_buffers(req, force_nonblock, cs); |
ddf0322d | 5965 | break; |
067524e9 JA |
5966 | case IORING_OP_REMOVE_BUFFERS: |
5967 | if (sqe) { | |
5968 | ret = io_remove_buffers_prep(req, sqe); | |
5969 | if (ret) | |
5970 | break; | |
5971 | } | |
229a7b63 | 5972 | ret = io_remove_buffers(req, force_nonblock, cs); |
3e4827b0 | 5973 | break; |
f2a8d5c7 PB |
5974 | case IORING_OP_TEE: |
5975 | if (sqe) { | |
5976 | ret = io_tee_prep(req, sqe); | |
5977 | if (ret < 0) | |
5978 | break; | |
5979 | } | |
5980 | ret = io_tee(req, force_nonblock); | |
5981 | break; | |
2b188cc1 JA |
5982 | default: |
5983 | ret = -EINVAL; | |
5984 | break; | |
5985 | } | |
5986 | ||
def596e9 JA |
5987 | if (ret) |
5988 | return ret; | |
5989 | ||
b532576e JA |
5990 | /* If the op doesn't have a file, we're not polling for it */ |
5991 | if ((ctx->flags & IORING_SETUP_IOPOLL) && req->file) { | |
11ba820b JA |
5992 | const bool in_async = io_wq_current_is_worker(); |
5993 | ||
11ba820b JA |
5994 | /* workqueue context doesn't hold uring_lock, grab it now */ |
5995 | if (in_async) | |
5996 | mutex_lock(&ctx->uring_lock); | |
5997 | ||
def596e9 | 5998 | io_iopoll_req_issued(req); |
11ba820b JA |
5999 | |
6000 | if (in_async) | |
6001 | mutex_unlock(&ctx->uring_lock); | |
def596e9 JA |
6002 | } |
6003 | ||
6004 | return 0; | |
2b188cc1 JA |
6005 | } |
6006 | ||
f4db7182 | 6007 | static struct io_wq_work *io_wq_submit_work(struct io_wq_work *work) |
2b188cc1 JA |
6008 | { |
6009 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); | |
6df1db6b | 6010 | struct io_kiocb *timeout; |
561fb04a | 6011 | int ret = 0; |
2b188cc1 | 6012 | |
6df1db6b PB |
6013 | timeout = io_prep_linked_timeout(req); |
6014 | if (timeout) | |
6015 | io_queue_linked_timeout(timeout); | |
d4c81f38 | 6016 | |
0c9d5ccd JA |
6017 | /* if NO_CANCEL is set, we must still run the work */ |
6018 | if ((work->flags & (IO_WQ_WORK_CANCEL|IO_WQ_WORK_NO_CANCEL)) == | |
6019 | IO_WQ_WORK_CANCEL) { | |
561fb04a | 6020 | ret = -ECANCELED; |
0c9d5ccd | 6021 | } |
31b51510 | 6022 | |
561fb04a | 6023 | if (!ret) { |
561fb04a | 6024 | do { |
f13fad7b | 6025 | ret = io_issue_sqe(req, NULL, false, NULL); |
561fb04a JA |
6026 | /* |
6027 | * We can get EAGAIN for polled IO even though we're | |
6028 | * forcing a sync submission from here, since we can't | |
6029 | * wait for request slots on the block side. | |
6030 | */ | |
6031 | if (ret != -EAGAIN) | |
6032 | break; | |
6033 | cond_resched(); | |
6034 | } while (1); | |
6035 | } | |
31b51510 | 6036 | |
561fb04a | 6037 | if (ret) { |
4e88d6e7 | 6038 | req_set_fail_links(req); |
e1e16097 | 6039 | io_req_complete(req, ret); |
edafccee | 6040 | } |
2b188cc1 | 6041 | |
f4db7182 | 6042 | return io_steal_work(req); |
2b188cc1 JA |
6043 | } |
6044 | ||
65e19f54 JA |
6045 | static inline struct file *io_file_from_index(struct io_ring_ctx *ctx, |
6046 | int index) | |
6047 | { | |
6048 | struct fixed_file_table *table; | |
6049 | ||
05f3fb3c | 6050 | table = &ctx->file_data->table[index >> IORING_FILE_TABLE_SHIFT]; |
84695089 | 6051 | return table->files[index & IORING_FILE_TABLE_MASK]; |
65e19f54 JA |
6052 | } |
6053 | ||
8da11c19 PB |
6054 | static int io_file_get(struct io_submit_state *state, struct io_kiocb *req, |
6055 | int fd, struct file **out_file, bool fixed) | |
09bb8394 | 6056 | { |
a197f664 | 6057 | struct io_ring_ctx *ctx = req->ctx; |
8da11c19 | 6058 | struct file *file; |
09bb8394 | 6059 | |
8da11c19 | 6060 | if (fixed) { |
05f3fb3c | 6061 | if (unlikely(!ctx->file_data || |
09bb8394 JA |
6062 | (unsigned) fd >= ctx->nr_user_files)) |
6063 | return -EBADF; | |
b7620121 | 6064 | fd = array_index_nospec(fd, ctx->nr_user_files); |
8da11c19 | 6065 | file = io_file_from_index(ctx, fd); |
fd2206e4 JA |
6066 | if (file) { |
6067 | req->fixed_file_refs = ctx->file_data->cur_refs; | |
6068 | percpu_ref_get(req->fixed_file_refs); | |
6069 | } | |
09bb8394 | 6070 | } else { |
c826bd7a | 6071 | trace_io_uring_file_get(ctx, fd); |
8da11c19 | 6072 | file = __io_file_get(state, fd); |
09bb8394 JA |
6073 | } |
6074 | ||
fd2206e4 JA |
6075 | if (file || io_op_defs[req->opcode].needs_file_no_error) { |
6076 | *out_file = file; | |
6077 | return 0; | |
6078 | } | |
6079 | return -EBADF; | |
09bb8394 JA |
6080 | } |
6081 | ||
8da11c19 | 6082 | static int io_req_set_file(struct io_submit_state *state, struct io_kiocb *req, |
63ff8223 | 6083 | int fd) |
8da11c19 | 6084 | { |
8da11c19 PB |
6085 | bool fixed; |
6086 | ||
63ff8223 | 6087 | fixed = (req->flags & REQ_F_FIXED_FILE) != 0; |
0cdaf760 | 6088 | if (unlikely(!fixed && io_async_submit(req->ctx))) |
8da11c19 PB |
6089 | return -EBADF; |
6090 | ||
6091 | return io_file_get(state, req, fd, &req->file, fixed); | |
6092 | } | |
6093 | ||
a197f664 | 6094 | static int io_grab_files(struct io_kiocb *req) |
fcb323cc | 6095 | { |
a197f664 | 6096 | struct io_ring_ctx *ctx = req->ctx; |
fcb323cc | 6097 | |
f56040b8 PB |
6098 | io_req_init_async(req); |
6099 | ||
5b0bbee4 | 6100 | if (req->work.files || (req->flags & REQ_F_NO_FILE_TABLE)) |
f86cd20c | 6101 | return 0; |
b5dba59e | 6102 | |
0f212204 | 6103 | req->work.files = get_files_struct(current); |
9b828492 JA |
6104 | get_nsproxy(current->nsproxy); |
6105 | req->work.nsproxy = current->nsproxy; | |
0f212204 JA |
6106 | req->flags |= REQ_F_INFLIGHT; |
6107 | ||
fcb323cc | 6108 | spin_lock_irq(&ctx->inflight_lock); |
0f212204 | 6109 | list_add(&req->inflight_entry, &ctx->inflight_list); |
fcb323cc | 6110 | spin_unlock_irq(&ctx->inflight_lock); |
0f212204 | 6111 | return 0; |
fcb323cc JA |
6112 | } |
6113 | ||
f56040b8 PB |
6114 | static inline int io_prep_work_files(struct io_kiocb *req) |
6115 | { | |
6116 | if (!io_op_defs[req->opcode].file_table) | |
6117 | return 0; | |
6118 | return io_grab_files(req); | |
6119 | } | |
6120 | ||
2665abfd | 6121 | static enum hrtimer_restart io_link_timeout_fn(struct hrtimer *timer) |
2b188cc1 | 6122 | { |
ad8a48ac JA |
6123 | struct io_timeout_data *data = container_of(timer, |
6124 | struct io_timeout_data, timer); | |
6125 | struct io_kiocb *req = data->req; | |
2665abfd JA |
6126 | struct io_ring_ctx *ctx = req->ctx; |
6127 | struct io_kiocb *prev = NULL; | |
6128 | unsigned long flags; | |
2665abfd JA |
6129 | |
6130 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
6131 | ||
6132 | /* | |
6133 | * We don't expect the list to be empty, that will only happen if we | |
6134 | * race with the completion of the linked work. | |
6135 | */ | |
4493233e PB |
6136 | if (!list_empty(&req->link_list)) { |
6137 | prev = list_entry(req->link_list.prev, struct io_kiocb, | |
6138 | link_list); | |
5d960724 | 6139 | if (refcount_inc_not_zero(&prev->refs)) { |
4493233e | 6140 | list_del_init(&req->link_list); |
5d960724 JA |
6141 | prev->flags &= ~REQ_F_LINK_TIMEOUT; |
6142 | } else | |
76a46e06 | 6143 | prev = NULL; |
2665abfd JA |
6144 | } |
6145 | ||
6146 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
6147 | ||
6148 | if (prev) { | |
4e88d6e7 | 6149 | req_set_fail_links(prev); |
014db007 | 6150 | io_async_find_and_cancel(ctx, req, prev->user_data, -ETIME); |
76a46e06 | 6151 | io_put_req(prev); |
47f46768 | 6152 | } else { |
e1e16097 | 6153 | io_req_complete(req, -ETIME); |
2665abfd | 6154 | } |
2665abfd JA |
6155 | return HRTIMER_NORESTART; |
6156 | } | |
6157 | ||
7271ef3a | 6158 | static void __io_queue_linked_timeout(struct io_kiocb *req) |
2665abfd | 6159 | { |
76a46e06 JA |
6160 | /* |
6161 | * If the list is now empty, then our linked request finished before | |
6162 | * we got a chance to setup the timer | |
6163 | */ | |
4493233e | 6164 | if (!list_empty(&req->link_list)) { |
2d28390a | 6165 | struct io_timeout_data *data = &req->io->timeout; |
94ae5e77 | 6166 | |
ad8a48ac JA |
6167 | data->timer.function = io_link_timeout_fn; |
6168 | hrtimer_start(&data->timer, timespec64_to_ktime(data->ts), | |
6169 | data->mode); | |
2665abfd | 6170 | } |
7271ef3a JA |
6171 | } |
6172 | ||
6173 | static void io_queue_linked_timeout(struct io_kiocb *req) | |
6174 | { | |
6175 | struct io_ring_ctx *ctx = req->ctx; | |
6176 | ||
6177 | spin_lock_irq(&ctx->completion_lock); | |
6178 | __io_queue_linked_timeout(req); | |
76a46e06 | 6179 | spin_unlock_irq(&ctx->completion_lock); |
2665abfd | 6180 | |
2665abfd | 6181 | /* drop submission reference */ |
76a46e06 JA |
6182 | io_put_req(req); |
6183 | } | |
2665abfd | 6184 | |
ad8a48ac | 6185 | static struct io_kiocb *io_prep_linked_timeout(struct io_kiocb *req) |
2665abfd JA |
6186 | { |
6187 | struct io_kiocb *nxt; | |
6188 | ||
dea3b49c | 6189 | if (!(req->flags & REQ_F_LINK_HEAD)) |
2665abfd | 6190 | return NULL; |
6df1db6b | 6191 | if (req->flags & REQ_F_LINK_TIMEOUT) |
d7718a9d | 6192 | return NULL; |
2665abfd | 6193 | |
4493233e PB |
6194 | nxt = list_first_entry_or_null(&req->link_list, struct io_kiocb, |
6195 | link_list); | |
d625c6ee | 6196 | if (!nxt || nxt->opcode != IORING_OP_LINK_TIMEOUT) |
76a46e06 | 6197 | return NULL; |
2665abfd | 6198 | |
76a46e06 | 6199 | req->flags |= REQ_F_LINK_TIMEOUT; |
76a46e06 | 6200 | return nxt; |
2665abfd JA |
6201 | } |
6202 | ||
f13fad7b JA |
6203 | static void __io_queue_sqe(struct io_kiocb *req, const struct io_uring_sqe *sqe, |
6204 | struct io_comp_state *cs) | |
2b188cc1 | 6205 | { |
4a0a7a18 | 6206 | struct io_kiocb *linked_timeout; |
4bc4494e | 6207 | struct io_kiocb *nxt; |
193155c8 | 6208 | const struct cred *old_creds = NULL; |
e0c5c576 | 6209 | int ret; |
2b188cc1 | 6210 | |
4a0a7a18 JA |
6211 | again: |
6212 | linked_timeout = io_prep_linked_timeout(req); | |
6213 | ||
7cdaf587 XW |
6214 | if ((req->flags & REQ_F_WORK_INITIALIZED) && req->work.creds && |
6215 | req->work.creds != current_cred()) { | |
193155c8 JA |
6216 | if (old_creds) |
6217 | revert_creds(old_creds); | |
6218 | if (old_creds == req->work.creds) | |
6219 | old_creds = NULL; /* restored original creds */ | |
6220 | else | |
6221 | old_creds = override_creds(req->work.creds); | |
6222 | } | |
6223 | ||
f13fad7b | 6224 | ret = io_issue_sqe(req, sqe, true, cs); |
491381ce JA |
6225 | |
6226 | /* | |
6227 | * We async punt it if the file wasn't marked NOWAIT, or if the file | |
6228 | * doesn't support non-blocking read/write attempts | |
6229 | */ | |
24c74678 | 6230 | if (ret == -EAGAIN && !(req->flags & REQ_F_NOWAIT)) { |
f063c547 | 6231 | if (!io_arm_poll_handler(req)) { |
86a761f8 | 6232 | punt: |
f063c547 PB |
6233 | ret = io_prep_work_files(req); |
6234 | if (unlikely(ret)) | |
bbad27b2 | 6235 | goto err; |
f063c547 PB |
6236 | /* |
6237 | * Queued up for async execution, worker will release | |
6238 | * submit reference when the iocb is actually submitted. | |
6239 | */ | |
6240 | io_queue_async_work(req); | |
2b188cc1 | 6241 | } |
bbad27b2 | 6242 | |
f063c547 PB |
6243 | if (linked_timeout) |
6244 | io_queue_linked_timeout(linked_timeout); | |
4bc4494e | 6245 | goto exit; |
2b188cc1 | 6246 | } |
e65ef56d | 6247 | |
652532ad | 6248 | if (unlikely(ret)) { |
fcb323cc | 6249 | err: |
652532ad PB |
6250 | /* un-prep timeout, so it'll be killed as any other linked */ |
6251 | req->flags &= ~REQ_F_LINK_TIMEOUT; | |
4e88d6e7 | 6252 | req_set_fail_links(req); |
e65ef56d | 6253 | io_put_req(req); |
e1e16097 | 6254 | io_req_complete(req, ret); |
652532ad | 6255 | goto exit; |
9e645e11 | 6256 | } |
652532ad PB |
6257 | |
6258 | /* drop submission reference */ | |
6259 | nxt = io_put_req_find_next(req); | |
6260 | if (linked_timeout) | |
6261 | io_queue_linked_timeout(linked_timeout); | |
6262 | ||
4a0a7a18 JA |
6263 | if (nxt) { |
6264 | req = nxt; | |
86a761f8 PB |
6265 | |
6266 | if (req->flags & REQ_F_FORCE_ASYNC) | |
6267 | goto punt; | |
4a0a7a18 JA |
6268 | goto again; |
6269 | } | |
4bc4494e | 6270 | exit: |
193155c8 JA |
6271 | if (old_creds) |
6272 | revert_creds(old_creds); | |
2b188cc1 JA |
6273 | } |
6274 | ||
f13fad7b JA |
6275 | static void io_queue_sqe(struct io_kiocb *req, const struct io_uring_sqe *sqe, |
6276 | struct io_comp_state *cs) | |
4fe2c963 JL |
6277 | { |
6278 | int ret; | |
6279 | ||
3529d8c2 | 6280 | ret = io_req_defer(req, sqe); |
4fe2c963 JL |
6281 | if (ret) { |
6282 | if (ret != -EIOCBQUEUED) { | |
1118591a | 6283 | fail_req: |
4e88d6e7 | 6284 | req_set_fail_links(req); |
e1e16097 JA |
6285 | io_put_req(req); |
6286 | io_req_complete(req, ret); | |
4fe2c963 | 6287 | } |
2550878f | 6288 | } else if (req->flags & REQ_F_FORCE_ASYNC) { |
bd2ab18a | 6289 | if (!req->io) { |
bd2ab18a | 6290 | ret = io_req_defer_prep(req, sqe); |
327d6d96 | 6291 | if (unlikely(ret)) |
bd2ab18a PB |
6292 | goto fail_req; |
6293 | } | |
6294 | ||
ce35a47a JA |
6295 | /* |
6296 | * Never try inline submit of IOSQE_ASYNC is set, go straight | |
6297 | * to async execution. | |
6298 | */ | |
3e863ea3 | 6299 | io_req_init_async(req); |
ce35a47a JA |
6300 | req->work.flags |= IO_WQ_WORK_CONCURRENT; |
6301 | io_queue_async_work(req); | |
6302 | } else { | |
f13fad7b | 6303 | __io_queue_sqe(req, sqe, cs); |
ce35a47a | 6304 | } |
4fe2c963 JL |
6305 | } |
6306 | ||
f13fad7b JA |
6307 | static inline void io_queue_link_head(struct io_kiocb *req, |
6308 | struct io_comp_state *cs) | |
4fe2c963 | 6309 | { |
94ae5e77 | 6310 | if (unlikely(req->flags & REQ_F_FAIL_LINK)) { |
e1e16097 JA |
6311 | io_put_req(req); |
6312 | io_req_complete(req, -ECANCELED); | |
1b4a51b6 | 6313 | } else |
f13fad7b | 6314 | io_queue_sqe(req, NULL, cs); |
4fe2c963 JL |
6315 | } |
6316 | ||
1d4240cc | 6317 | static int io_submit_sqe(struct io_kiocb *req, const struct io_uring_sqe *sqe, |
f13fad7b | 6318 | struct io_kiocb **link, struct io_comp_state *cs) |
9e645e11 | 6319 | { |
a197f664 | 6320 | struct io_ring_ctx *ctx = req->ctx; |
ef4ff581 | 6321 | int ret; |
9e645e11 | 6322 | |
9e645e11 JA |
6323 | /* |
6324 | * If we already have a head request, queue this one for async | |
6325 | * submittal once the head completes. If we don't have a head but | |
6326 | * IOSQE_IO_LINK is set in the sqe, start a new head. This one will be | |
6327 | * submitted sync once the chain is complete. If none of those | |
6328 | * conditions are true (normal request), then just queue it. | |
6329 | */ | |
6330 | if (*link) { | |
9d76377f | 6331 | struct io_kiocb *head = *link; |
4e88d6e7 | 6332 | |
8cdf2193 PB |
6333 | /* |
6334 | * Taking sequential execution of a link, draining both sides | |
6335 | * of the link also fullfils IOSQE_IO_DRAIN semantics for all | |
6336 | * requests in the link. So, it drains the head and the | |
6337 | * next after the link request. The last one is done via | |
6338 | * drain_next flag to persist the effect across calls. | |
6339 | */ | |
ef4ff581 | 6340 | if (req->flags & REQ_F_IO_DRAIN) { |
711be031 PB |
6341 | head->flags |= REQ_F_IO_DRAIN; |
6342 | ctx->drain_next = 1; | |
6343 | } | |
3529d8c2 | 6344 | ret = io_req_defer_prep(req, sqe); |
327d6d96 | 6345 | if (unlikely(ret)) { |
4e88d6e7 | 6346 | /* fail even hard links since we don't submit */ |
9d76377f | 6347 | head->flags |= REQ_F_FAIL_LINK; |
1d4240cc | 6348 | return ret; |
2d28390a | 6349 | } |
9d76377f PB |
6350 | trace_io_uring_link(ctx, req, head); |
6351 | list_add_tail(&req->link_list, &head->link_list); | |
32fe525b PB |
6352 | |
6353 | /* last request of a link, enqueue the link */ | |
ef4ff581 | 6354 | if (!(req->flags & (REQ_F_LINK | REQ_F_HARDLINK))) { |
f13fad7b | 6355 | io_queue_link_head(head, cs); |
32fe525b PB |
6356 | *link = NULL; |
6357 | } | |
9e645e11 | 6358 | } else { |
711be031 PB |
6359 | if (unlikely(ctx->drain_next)) { |
6360 | req->flags |= REQ_F_IO_DRAIN; | |
ef4ff581 | 6361 | ctx->drain_next = 0; |
711be031 | 6362 | } |
ef4ff581 | 6363 | if (req->flags & (REQ_F_LINK | REQ_F_HARDLINK)) { |
dea3b49c | 6364 | req->flags |= REQ_F_LINK_HEAD; |
711be031 | 6365 | INIT_LIST_HEAD(&req->link_list); |
f1d96a8f | 6366 | |
711be031 | 6367 | ret = io_req_defer_prep(req, sqe); |
327d6d96 | 6368 | if (unlikely(ret)) |
711be031 PB |
6369 | req->flags |= REQ_F_FAIL_LINK; |
6370 | *link = req; | |
6371 | } else { | |
f13fad7b | 6372 | io_queue_sqe(req, sqe, cs); |
711be031 | 6373 | } |
9e645e11 | 6374 | } |
2e6e1fde | 6375 | |
1d4240cc | 6376 | return 0; |
9e645e11 JA |
6377 | } |
6378 | ||
9a56a232 JA |
6379 | /* |
6380 | * Batched submission is done, ensure local IO is flushed out. | |
6381 | */ | |
6382 | static void io_submit_state_end(struct io_submit_state *state) | |
6383 | { | |
f13fad7b JA |
6384 | if (!list_empty(&state->comp.list)) |
6385 | io_submit_flush_completions(&state->comp); | |
9a56a232 | 6386 | blk_finish_plug(&state->plug); |
9f13c35b | 6387 | io_state_file_put(state); |
2579f913 | 6388 | if (state->free_reqs) |
6c8a3134 | 6389 | kmem_cache_free_bulk(req_cachep, state->free_reqs, state->reqs); |
9a56a232 JA |
6390 | } |
6391 | ||
6392 | /* | |
6393 | * Start submission side cache. | |
6394 | */ | |
6395 | static void io_submit_state_start(struct io_submit_state *state, | |
013538bd | 6396 | struct io_ring_ctx *ctx, unsigned int max_ios) |
9a56a232 JA |
6397 | { |
6398 | blk_start_plug(&state->plug); | |
013538bd JA |
6399 | state->comp.nr = 0; |
6400 | INIT_LIST_HEAD(&state->comp.list); | |
6401 | state->comp.ctx = ctx; | |
2579f913 | 6402 | state->free_reqs = 0; |
9a56a232 JA |
6403 | state->file = NULL; |
6404 | state->ios_left = max_ios; | |
6405 | } | |
6406 | ||
2b188cc1 JA |
6407 | static void io_commit_sqring(struct io_ring_ctx *ctx) |
6408 | { | |
75b28aff | 6409 | struct io_rings *rings = ctx->rings; |
2b188cc1 | 6410 | |
caf582c6 PB |
6411 | /* |
6412 | * Ensure any loads from the SQEs are done at this point, | |
6413 | * since once we write the new head, the application could | |
6414 | * write new data to them. | |
6415 | */ | |
6416 | smp_store_release(&rings->sq.head, ctx->cached_sq_head); | |
2b188cc1 JA |
6417 | } |
6418 | ||
2b188cc1 | 6419 | /* |
3529d8c2 | 6420 | * Fetch an sqe, if one is available. Note that sqe_ptr will point to memory |
2b188cc1 JA |
6421 | * that is mapped by userspace. This means that care needs to be taken to |
6422 | * ensure that reads are stable, as we cannot rely on userspace always | |
6423 | * being a good citizen. If members of the sqe are validated and then later | |
6424 | * used, it's important that those reads are done through READ_ONCE() to | |
6425 | * prevent a re-load down the line. | |
6426 | */ | |
709b302f | 6427 | static const struct io_uring_sqe *io_get_sqe(struct io_ring_ctx *ctx) |
2b188cc1 | 6428 | { |
75b28aff | 6429 | u32 *sq_array = ctx->sq_array; |
2b188cc1 JA |
6430 | unsigned head; |
6431 | ||
6432 | /* | |
6433 | * The cached sq head (or cq tail) serves two purposes: | |
6434 | * | |
6435 | * 1) allows us to batch the cost of updating the user visible | |
6436 | * head updates. | |
6437 | * 2) allows the kernel side to track the head on its own, even | |
6438 | * though the application is the one updating it. | |
6439 | */ | |
ee7d46d9 | 6440 | head = READ_ONCE(sq_array[ctx->cached_sq_head & ctx->sq_mask]); |
709b302f PB |
6441 | if (likely(head < ctx->sq_entries)) |
6442 | return &ctx->sq_sqes[head]; | |
2b188cc1 JA |
6443 | |
6444 | /* drop invalid entries */ | |
498ccd9e | 6445 | ctx->cached_sq_dropped++; |
ee7d46d9 | 6446 | WRITE_ONCE(ctx->rings->sq_dropped, ctx->cached_sq_dropped); |
709b302f PB |
6447 | return NULL; |
6448 | } | |
6449 | ||
6450 | static inline void io_consume_sqe(struct io_ring_ctx *ctx) | |
6451 | { | |
6452 | ctx->cached_sq_head++; | |
2b188cc1 JA |
6453 | } |
6454 | ||
21b55dbc SG |
6455 | /* |
6456 | * Check SQE restrictions (opcode and flags). | |
6457 | * | |
6458 | * Returns 'true' if SQE is allowed, 'false' otherwise. | |
6459 | */ | |
6460 | static inline bool io_check_restriction(struct io_ring_ctx *ctx, | |
6461 | struct io_kiocb *req, | |
6462 | unsigned int sqe_flags) | |
6463 | { | |
6464 | if (!ctx->restricted) | |
6465 | return true; | |
6466 | ||
6467 | if (!test_bit(req->opcode, ctx->restrictions.sqe_op)) | |
6468 | return false; | |
6469 | ||
6470 | if ((sqe_flags & ctx->restrictions.sqe_flags_required) != | |
6471 | ctx->restrictions.sqe_flags_required) | |
6472 | return false; | |
6473 | ||
6474 | if (sqe_flags & ~(ctx->restrictions.sqe_flags_allowed | | |
6475 | ctx->restrictions.sqe_flags_required)) | |
6476 | return false; | |
6477 | ||
6478 | return true; | |
6479 | } | |
6480 | ||
ef4ff581 PB |
6481 | #define SQE_VALID_FLAGS (IOSQE_FIXED_FILE|IOSQE_IO_DRAIN|IOSQE_IO_LINK| \ |
6482 | IOSQE_IO_HARDLINK | IOSQE_ASYNC | \ | |
6483 | IOSQE_BUFFER_SELECT) | |
6484 | ||
6485 | static int io_init_req(struct io_ring_ctx *ctx, struct io_kiocb *req, | |
6486 | const struct io_uring_sqe *sqe, | |
0cdaf760 | 6487 | struct io_submit_state *state) |
0553b8bd | 6488 | { |
ef4ff581 | 6489 | unsigned int sqe_flags; |
63ff8223 | 6490 | int id; |
ef4ff581 | 6491 | |
0553b8bd PB |
6492 | req->opcode = READ_ONCE(sqe->opcode); |
6493 | req->user_data = READ_ONCE(sqe->user_data); | |
6494 | req->io = NULL; | |
6495 | req->file = NULL; | |
6496 | req->ctx = ctx; | |
6497 | req->flags = 0; | |
6498 | /* one is dropped after submission, the other at completion */ | |
6499 | refcount_set(&req->refs, 2); | |
4dd2824d | 6500 | req->task = current; |
e3bc8e9d | 6501 | get_task_struct(req->task); |
0f212204 | 6502 | atomic_long_inc(&req->task->io_uring->req_issue); |
0553b8bd | 6503 | req->result = 0; |
ef4ff581 PB |
6504 | |
6505 | if (unlikely(req->opcode >= IORING_OP_LAST)) | |
6506 | return -EINVAL; | |
6507 | ||
9d8426a0 JA |
6508 | if (unlikely(io_sq_thread_acquire_mm(ctx, req))) |
6509 | return -EFAULT; | |
ef4ff581 PB |
6510 | |
6511 | sqe_flags = READ_ONCE(sqe->flags); | |
6512 | /* enforce forwards compatibility on users */ | |
6513 | if (unlikely(sqe_flags & ~SQE_VALID_FLAGS)) | |
6514 | return -EINVAL; | |
6515 | ||
21b55dbc SG |
6516 | if (unlikely(!io_check_restriction(ctx, req, sqe_flags))) |
6517 | return -EACCES; | |
6518 | ||
ef4ff581 PB |
6519 | if ((sqe_flags & IOSQE_BUFFER_SELECT) && |
6520 | !io_op_defs[req->opcode].buffer_select) | |
6521 | return -EOPNOTSUPP; | |
6522 | ||
6523 | id = READ_ONCE(sqe->personality); | |
6524 | if (id) { | |
7cdaf587 | 6525 | io_req_init_async(req); |
ef4ff581 PB |
6526 | req->work.creds = idr_find(&ctx->personality_idr, id); |
6527 | if (unlikely(!req->work.creds)) | |
6528 | return -EINVAL; | |
6529 | get_cred(req->work.creds); | |
6530 | } | |
6531 | ||
6532 | /* same numerical values with corresponding REQ_F_*, safe to copy */ | |
c11368a5 | 6533 | req->flags |= sqe_flags; |
ef4ff581 | 6534 | |
63ff8223 JA |
6535 | if (!io_op_defs[req->opcode].needs_file) |
6536 | return 0; | |
6537 | ||
6538 | return io_req_set_file(state, req, READ_ONCE(sqe->fd)); | |
0553b8bd PB |
6539 | } |
6540 | ||
0f212204 | 6541 | static int io_submit_sqes(struct io_ring_ctx *ctx, unsigned int nr) |
6c271ce2 | 6542 | { |
ac8691c4 | 6543 | struct io_submit_state state; |
9e645e11 | 6544 | struct io_kiocb *link = NULL; |
9e645e11 | 6545 | int i, submitted = 0; |
6c271ce2 | 6546 | |
c4a2ed72 | 6547 | /* if we have a backlog and couldn't flush it all, return BUSY */ |
ad3eb2c8 JA |
6548 | if (test_bit(0, &ctx->sq_check_overflow)) { |
6549 | if (!list_empty(&ctx->cq_overflow_list) && | |
e6c8aa9a | 6550 | !io_cqring_overflow_flush(ctx, false, NULL, NULL)) |
ad3eb2c8 JA |
6551 | return -EBUSY; |
6552 | } | |
6c271ce2 | 6553 | |
ee7d46d9 PB |
6554 | /* make sure SQ entry isn't read before tail */ |
6555 | nr = min3(nr, ctx->sq_entries, io_sqring_entries(ctx)); | |
9ef4f124 | 6556 | |
2b85edfc PB |
6557 | if (!percpu_ref_tryget_many(&ctx->refs, nr)) |
6558 | return -EAGAIN; | |
6c271ce2 | 6559 | |
013538bd | 6560 | io_submit_state_start(&state, ctx, nr); |
6c271ce2 JA |
6561 | |
6562 | for (i = 0; i < nr; i++) { | |
3529d8c2 | 6563 | const struct io_uring_sqe *sqe; |
196be95c | 6564 | struct io_kiocb *req; |
1cb1edb2 | 6565 | int err; |
fb5ccc98 | 6566 | |
b1e50e54 PB |
6567 | sqe = io_get_sqe(ctx); |
6568 | if (unlikely(!sqe)) { | |
6569 | io_consume_sqe(ctx); | |
6570 | break; | |
6571 | } | |
ac8691c4 | 6572 | req = io_alloc_req(ctx, &state); |
196be95c PB |
6573 | if (unlikely(!req)) { |
6574 | if (!submitted) | |
6575 | submitted = -EAGAIN; | |
fb5ccc98 | 6576 | break; |
196be95c | 6577 | } |
fb5ccc98 | 6578 | |
ac8691c4 | 6579 | err = io_init_req(ctx, req, sqe, &state); |
709b302f | 6580 | io_consume_sqe(ctx); |
d3656344 JA |
6581 | /* will complete beyond this point, count as submitted */ |
6582 | submitted++; | |
6583 | ||
ef4ff581 | 6584 | if (unlikely(err)) { |
1cb1edb2 | 6585 | fail_req: |
e1e16097 JA |
6586 | io_put_req(req); |
6587 | io_req_complete(req, err); | |
196be95c PB |
6588 | break; |
6589 | } | |
fb5ccc98 | 6590 | |
354420f7 | 6591 | trace_io_uring_submit_sqe(ctx, req->opcode, req->user_data, |
0cdaf760 | 6592 | true, io_async_submit(ctx)); |
f13fad7b | 6593 | err = io_submit_sqe(req, sqe, &link, &state.comp); |
1d4240cc PB |
6594 | if (err) |
6595 | goto fail_req; | |
6c271ce2 JA |
6596 | } |
6597 | ||
9466f437 PB |
6598 | if (unlikely(submitted != nr)) { |
6599 | int ref_used = (submitted == -EAGAIN) ? 0 : submitted; | |
6600 | ||
6601 | percpu_ref_put_many(&ctx->refs, nr - ref_used); | |
6602 | } | |
9e645e11 | 6603 | if (link) |
f13fad7b | 6604 | io_queue_link_head(link, &state.comp); |
ac8691c4 | 6605 | io_submit_state_end(&state); |
6c271ce2 | 6606 | |
ae9428ca PB |
6607 | /* Commit SQ ring head once we've consumed and submitted all SQEs */ |
6608 | io_commit_sqring(ctx); | |
6609 | ||
6c271ce2 JA |
6610 | return submitted; |
6611 | } | |
6612 | ||
23b3628e XW |
6613 | static inline void io_ring_set_wakeup_flag(struct io_ring_ctx *ctx) |
6614 | { | |
6615 | /* Tell userspace we may need a wakeup call */ | |
6616 | spin_lock_irq(&ctx->completion_lock); | |
6617 | ctx->rings->sq_flags |= IORING_SQ_NEED_WAKEUP; | |
6618 | spin_unlock_irq(&ctx->completion_lock); | |
6619 | } | |
6620 | ||
6621 | static inline void io_ring_clear_wakeup_flag(struct io_ring_ctx *ctx) | |
6622 | { | |
6623 | spin_lock_irq(&ctx->completion_lock); | |
6624 | ctx->rings->sq_flags &= ~IORING_SQ_NEED_WAKEUP; | |
6625 | spin_unlock_irq(&ctx->completion_lock); | |
6626 | } | |
6627 | ||
6c271ce2 JA |
6628 | static int io_sq_thread(void *data) |
6629 | { | |
6c271ce2 | 6630 | struct io_ring_ctx *ctx = data; |
181e448d | 6631 | const struct cred *old_cred; |
6c271ce2 | 6632 | unsigned long timeout; |
bdcd3eab | 6633 | int ret = 0; |
6c271ce2 | 6634 | |
6a779382 JA |
6635 | init_wait(&ctx->sqo_wait_entry); |
6636 | ||
0f158b4c | 6637 | complete(&ctx->sq_thread_comp); |
a4c0b3de | 6638 | |
181e448d | 6639 | old_cred = override_creds(ctx->creds); |
6c271ce2 | 6640 | |
bdcd3eab | 6641 | timeout = jiffies + ctx->sq_thread_idle; |
2bbcd6d3 | 6642 | while (!kthread_should_park()) { |
fb5ccc98 | 6643 | unsigned int to_submit; |
6c271ce2 | 6644 | |
540e32a0 | 6645 | if (!list_empty(&ctx->iopoll_list)) { |
6c271ce2 JA |
6646 | unsigned nr_events = 0; |
6647 | ||
bdcd3eab | 6648 | mutex_lock(&ctx->uring_lock); |
540e32a0 | 6649 | if (!list_empty(&ctx->iopoll_list) && !need_resched()) |
9dedd563 | 6650 | io_do_iopoll(ctx, &nr_events, 0); |
bdcd3eab | 6651 | else |
6c271ce2 | 6652 | timeout = jiffies + ctx->sq_thread_idle; |
bdcd3eab | 6653 | mutex_unlock(&ctx->uring_lock); |
6c271ce2 JA |
6654 | } |
6655 | ||
fb5ccc98 | 6656 | to_submit = io_sqring_entries(ctx); |
c1edbf5f JA |
6657 | |
6658 | /* | |
6659 | * If submit got -EBUSY, flag us as needing the application | |
6660 | * to enter the kernel to reap and flush events. | |
6661 | */ | |
b772f07a | 6662 | if (!to_submit || ret == -EBUSY || need_resched()) { |
7143b5ac SG |
6663 | /* |
6664 | * Drop cur_mm before scheduling, we can't hold it for | |
6665 | * long periods (or over schedule()). Do this before | |
6666 | * adding ourselves to the waitqueue, as the unuse/drop | |
6667 | * may sleep. | |
6668 | */ | |
4349f30e | 6669 | io_sq_thread_drop_mm(); |
7143b5ac | 6670 | |
6c271ce2 JA |
6671 | /* |
6672 | * We're polling. If we're within the defined idle | |
6673 | * period, then let us spin without work before going | |
c1edbf5f JA |
6674 | * to sleep. The exception is if we got EBUSY doing |
6675 | * more IO, we should wait for the application to | |
6676 | * reap events and wake us up. | |
6c271ce2 | 6677 | */ |
540e32a0 | 6678 | if (!list_empty(&ctx->iopoll_list) || need_resched() || |
df069d80 JA |
6679 | (!time_after(jiffies, timeout) && ret != -EBUSY && |
6680 | !percpu_ref_is_dying(&ctx->refs))) { | |
4c6e277c | 6681 | io_run_task_work(); |
9831a90c | 6682 | cond_resched(); |
6c271ce2 JA |
6683 | continue; |
6684 | } | |
6685 | ||
6a779382 | 6686 | prepare_to_wait(ctx->sqo_wait, &ctx->sqo_wait_entry, |
6c271ce2 JA |
6687 | TASK_INTERRUPTIBLE); |
6688 | ||
bdcd3eab XW |
6689 | /* |
6690 | * While doing polled IO, before going to sleep, we need | |
540e32a0 PB |
6691 | * to check if there are new reqs added to iopoll_list, |
6692 | * it is because reqs may have been punted to io worker | |
6693 | * and will be added to iopoll_list later, hence check | |
6694 | * the iopoll_list again. | |
bdcd3eab XW |
6695 | */ |
6696 | if ((ctx->flags & IORING_SETUP_IOPOLL) && | |
540e32a0 | 6697 | !list_empty_careful(&ctx->iopoll_list)) { |
6a779382 | 6698 | finish_wait(ctx->sqo_wait, &ctx->sqo_wait_entry); |
bdcd3eab XW |
6699 | continue; |
6700 | } | |
6701 | ||
23b3628e | 6702 | io_ring_set_wakeup_flag(ctx); |
6c271ce2 | 6703 | |
fb5ccc98 | 6704 | to_submit = io_sqring_entries(ctx); |
c1edbf5f | 6705 | if (!to_submit || ret == -EBUSY) { |
2bbcd6d3 | 6706 | if (kthread_should_park()) { |
6a779382 | 6707 | finish_wait(ctx->sqo_wait, &ctx->sqo_wait_entry); |
6c271ce2 JA |
6708 | break; |
6709 | } | |
4c6e277c | 6710 | if (io_run_task_work()) { |
6a779382 | 6711 | finish_wait(ctx->sqo_wait, &ctx->sqo_wait_entry); |
23b3628e | 6712 | io_ring_clear_wakeup_flag(ctx); |
b41e9852 JA |
6713 | continue; |
6714 | } | |
6c271ce2 | 6715 | schedule(); |
6a779382 | 6716 | finish_wait(ctx->sqo_wait, &ctx->sqo_wait_entry); |
6c271ce2 | 6717 | |
23b3628e | 6718 | io_ring_clear_wakeup_flag(ctx); |
d4ae271d | 6719 | ret = 0; |
6c271ce2 JA |
6720 | continue; |
6721 | } | |
6a779382 | 6722 | finish_wait(ctx->sqo_wait, &ctx->sqo_wait_entry); |
6c271ce2 | 6723 | |
23b3628e | 6724 | io_ring_clear_wakeup_flag(ctx); |
6c271ce2 JA |
6725 | } |
6726 | ||
8a4955ff | 6727 | mutex_lock(&ctx->uring_lock); |
6b668c9b | 6728 | if (likely(!percpu_ref_is_dying(&ctx->refs))) |
0f212204 | 6729 | ret = io_submit_sqes(ctx, to_submit); |
8a4955ff | 6730 | mutex_unlock(&ctx->uring_lock); |
bdcd3eab | 6731 | timeout = jiffies + ctx->sq_thread_idle; |
6c271ce2 JA |
6732 | } |
6733 | ||
4c6e277c | 6734 | io_run_task_work(); |
b41e9852 | 6735 | |
4349f30e | 6736 | io_sq_thread_drop_mm(); |
181e448d | 6737 | revert_creds(old_cred); |
06058632 | 6738 | |
2bbcd6d3 | 6739 | kthread_parkme(); |
06058632 | 6740 | |
6c271ce2 JA |
6741 | return 0; |
6742 | } | |
6743 | ||
bda52162 JA |
6744 | struct io_wait_queue { |
6745 | struct wait_queue_entry wq; | |
6746 | struct io_ring_ctx *ctx; | |
6747 | unsigned to_wait; | |
6748 | unsigned nr_timeouts; | |
6749 | }; | |
6750 | ||
1d7bb1d5 | 6751 | static inline bool io_should_wake(struct io_wait_queue *iowq, bool noflush) |
bda52162 JA |
6752 | { |
6753 | struct io_ring_ctx *ctx = iowq->ctx; | |
6754 | ||
6755 | /* | |
d195a66e | 6756 | * Wake up if we have enough events, or if a timeout occurred since we |
bda52162 JA |
6757 | * started waiting. For timeouts, we always want to return to userspace, |
6758 | * regardless of event count. | |
6759 | */ | |
1d7bb1d5 | 6760 | return io_cqring_events(ctx, noflush) >= iowq->to_wait || |
bda52162 JA |
6761 | atomic_read(&ctx->cq_timeouts) != iowq->nr_timeouts; |
6762 | } | |
6763 | ||
6764 | static int io_wake_function(struct wait_queue_entry *curr, unsigned int mode, | |
6765 | int wake_flags, void *key) | |
6766 | { | |
6767 | struct io_wait_queue *iowq = container_of(curr, struct io_wait_queue, | |
6768 | wq); | |
6769 | ||
1d7bb1d5 JA |
6770 | /* use noflush == true, as we can't safely rely on locking context */ |
6771 | if (!io_should_wake(iowq, true)) | |
bda52162 JA |
6772 | return -1; |
6773 | ||
6774 | return autoremove_wake_function(curr, mode, wake_flags, key); | |
6775 | } | |
6776 | ||
2b188cc1 JA |
6777 | /* |
6778 | * Wait until events become available, if we don't already have some. The | |
6779 | * application must reap them itself, as they reside on the shared cq ring. | |
6780 | */ | |
6781 | static int io_cqring_wait(struct io_ring_ctx *ctx, int min_events, | |
6782 | const sigset_t __user *sig, size_t sigsz) | |
6783 | { | |
bda52162 JA |
6784 | struct io_wait_queue iowq = { |
6785 | .wq = { | |
6786 | .private = current, | |
6787 | .func = io_wake_function, | |
6788 | .entry = LIST_HEAD_INIT(iowq.wq.entry), | |
6789 | }, | |
6790 | .ctx = ctx, | |
6791 | .to_wait = min_events, | |
6792 | }; | |
75b28aff | 6793 | struct io_rings *rings = ctx->rings; |
e9ffa5c2 | 6794 | int ret = 0; |
2b188cc1 | 6795 | |
b41e9852 JA |
6796 | do { |
6797 | if (io_cqring_events(ctx, false) >= min_events) | |
6798 | return 0; | |
4c6e277c | 6799 | if (!io_run_task_work()) |
b41e9852 | 6800 | break; |
b41e9852 | 6801 | } while (1); |
2b188cc1 JA |
6802 | |
6803 | if (sig) { | |
9e75ad5d AB |
6804 | #ifdef CONFIG_COMPAT |
6805 | if (in_compat_syscall()) | |
6806 | ret = set_compat_user_sigmask((const compat_sigset_t __user *)sig, | |
b772434b | 6807 | sigsz); |
9e75ad5d AB |
6808 | else |
6809 | #endif | |
b772434b | 6810 | ret = set_user_sigmask(sig, sigsz); |
9e75ad5d | 6811 | |
2b188cc1 JA |
6812 | if (ret) |
6813 | return ret; | |
6814 | } | |
6815 | ||
bda52162 | 6816 | iowq.nr_timeouts = atomic_read(&ctx->cq_timeouts); |
c826bd7a | 6817 | trace_io_uring_cqring_wait(ctx, min_events); |
bda52162 JA |
6818 | do { |
6819 | prepare_to_wait_exclusive(&ctx->wait, &iowq.wq, | |
6820 | TASK_INTERRUPTIBLE); | |
ce593a6c | 6821 | /* make sure we run task_work before checking for signals */ |
4c6e277c JA |
6822 | if (io_run_task_work()) |
6823 | continue; | |
bda52162 | 6824 | if (signal_pending(current)) { |
b7db41c9 JA |
6825 | if (current->jobctl & JOBCTL_TASK_WORK) { |
6826 | spin_lock_irq(¤t->sighand->siglock); | |
6827 | current->jobctl &= ~JOBCTL_TASK_WORK; | |
6828 | recalc_sigpending(); | |
6829 | spin_unlock_irq(¤t->sighand->siglock); | |
6830 | continue; | |
6831 | } | |
6832 | ret = -EINTR; | |
bda52162 JA |
6833 | break; |
6834 | } | |
ce593a6c JA |
6835 | if (io_should_wake(&iowq, false)) |
6836 | break; | |
6837 | schedule(); | |
bda52162 JA |
6838 | } while (1); |
6839 | finish_wait(&ctx->wait, &iowq.wq); | |
6840 | ||
b7db41c9 | 6841 | restore_saved_sigmask_unless(ret == -EINTR); |
2b188cc1 | 6842 | |
75b28aff | 6843 | return READ_ONCE(rings->cq.head) == READ_ONCE(rings->cq.tail) ? ret : 0; |
2b188cc1 JA |
6844 | } |
6845 | ||
6b06314c JA |
6846 | static void __io_sqe_files_unregister(struct io_ring_ctx *ctx) |
6847 | { | |
6848 | #if defined(CONFIG_UNIX) | |
6849 | if (ctx->ring_sock) { | |
6850 | struct sock *sock = ctx->ring_sock->sk; | |
6851 | struct sk_buff *skb; | |
6852 | ||
6853 | while ((skb = skb_dequeue(&sock->sk_receive_queue)) != NULL) | |
6854 | kfree_skb(skb); | |
6855 | } | |
6856 | #else | |
6857 | int i; | |
6858 | ||
65e19f54 JA |
6859 | for (i = 0; i < ctx->nr_user_files; i++) { |
6860 | struct file *file; | |
6861 | ||
6862 | file = io_file_from_index(ctx, i); | |
6863 | if (file) | |
6864 | fput(file); | |
6865 | } | |
6b06314c JA |
6866 | #endif |
6867 | } | |
6868 | ||
05f3fb3c JA |
6869 | static void io_file_ref_kill(struct percpu_ref *ref) |
6870 | { | |
6871 | struct fixed_file_data *data; | |
6872 | ||
6873 | data = container_of(ref, struct fixed_file_data, refs); | |
6874 | complete(&data->done); | |
6875 | } | |
6876 | ||
6b06314c JA |
6877 | static int io_sqe_files_unregister(struct io_ring_ctx *ctx) |
6878 | { | |
05f3fb3c | 6879 | struct fixed_file_data *data = ctx->file_data; |
05589553 | 6880 | struct fixed_file_ref_node *ref_node = NULL; |
65e19f54 JA |
6881 | unsigned nr_tables, i; |
6882 | ||
05f3fb3c | 6883 | if (!data) |
6b06314c JA |
6884 | return -ENXIO; |
6885 | ||
6a4d07cd | 6886 | spin_lock(&data->lock); |
05589553 XW |
6887 | if (!list_empty(&data->ref_list)) |
6888 | ref_node = list_first_entry(&data->ref_list, | |
6889 | struct fixed_file_ref_node, node); | |
6a4d07cd | 6890 | spin_unlock(&data->lock); |
05589553 XW |
6891 | if (ref_node) |
6892 | percpu_ref_kill(&ref_node->refs); | |
6893 | ||
6894 | percpu_ref_kill(&data->refs); | |
6895 | ||
6896 | /* wait for all refs nodes to complete */ | |
4a38aed2 | 6897 | flush_delayed_work(&ctx->file_put_work); |
2faf852d | 6898 | wait_for_completion(&data->done); |
05f3fb3c | 6899 | |
6b06314c | 6900 | __io_sqe_files_unregister(ctx); |
65e19f54 JA |
6901 | nr_tables = DIV_ROUND_UP(ctx->nr_user_files, IORING_MAX_FILES_TABLE); |
6902 | for (i = 0; i < nr_tables; i++) | |
05f3fb3c JA |
6903 | kfree(data->table[i].files); |
6904 | kfree(data->table); | |
05589553 XW |
6905 | percpu_ref_exit(&data->refs); |
6906 | kfree(data); | |
05f3fb3c | 6907 | ctx->file_data = NULL; |
6b06314c JA |
6908 | ctx->nr_user_files = 0; |
6909 | return 0; | |
6910 | } | |
6911 | ||
6c271ce2 JA |
6912 | static void io_sq_thread_stop(struct io_ring_ctx *ctx) |
6913 | { | |
6914 | if (ctx->sqo_thread) { | |
7e84e1c7 SG |
6915 | /* |
6916 | * We may arrive here from the error branch in | |
6917 | * io_sq_offload_create() where the kthread is created | |
6918 | * without being waked up, thus wake it up now to make | |
6919 | * sure the wait will complete. | |
6920 | */ | |
6921 | wake_up_process(ctx->sqo_thread); | |
6922 | ||
0f158b4c | 6923 | wait_for_completion(&ctx->sq_thread_comp); |
2bbcd6d3 RP |
6924 | /* |
6925 | * The park is a bit of a work-around, without it we get | |
6926 | * warning spews on shutdown with SQPOLL set and affinity | |
6927 | * set to a single CPU. | |
6928 | */ | |
06058632 | 6929 | kthread_park(ctx->sqo_thread); |
6c271ce2 JA |
6930 | kthread_stop(ctx->sqo_thread); |
6931 | ctx->sqo_thread = NULL; | |
6932 | } | |
6933 | } | |
6934 | ||
6b06314c JA |
6935 | static void io_finish_async(struct io_ring_ctx *ctx) |
6936 | { | |
6c271ce2 JA |
6937 | io_sq_thread_stop(ctx); |
6938 | ||
561fb04a JA |
6939 | if (ctx->io_wq) { |
6940 | io_wq_destroy(ctx->io_wq); | |
6941 | ctx->io_wq = NULL; | |
6b06314c JA |
6942 | } |
6943 | } | |
6944 | ||
6945 | #if defined(CONFIG_UNIX) | |
6b06314c JA |
6946 | /* |
6947 | * Ensure the UNIX gc is aware of our file set, so we are certain that | |
6948 | * the io_uring can be safely unregistered on process exit, even if we have | |
6949 | * loops in the file referencing. | |
6950 | */ | |
6951 | static int __io_sqe_files_scm(struct io_ring_ctx *ctx, int nr, int offset) | |
6952 | { | |
6953 | struct sock *sk = ctx->ring_sock->sk; | |
6954 | struct scm_fp_list *fpl; | |
6955 | struct sk_buff *skb; | |
08a45173 | 6956 | int i, nr_files; |
6b06314c | 6957 | |
6b06314c JA |
6958 | fpl = kzalloc(sizeof(*fpl), GFP_KERNEL); |
6959 | if (!fpl) | |
6960 | return -ENOMEM; | |
6961 | ||
6962 | skb = alloc_skb(0, GFP_KERNEL); | |
6963 | if (!skb) { | |
6964 | kfree(fpl); | |
6965 | return -ENOMEM; | |
6966 | } | |
6967 | ||
6968 | skb->sk = sk; | |
6b06314c | 6969 | |
08a45173 | 6970 | nr_files = 0; |
6b06314c JA |
6971 | fpl->user = get_uid(ctx->user); |
6972 | for (i = 0; i < nr; i++) { | |
65e19f54 JA |
6973 | struct file *file = io_file_from_index(ctx, i + offset); |
6974 | ||
6975 | if (!file) | |
08a45173 | 6976 | continue; |
65e19f54 | 6977 | fpl->fp[nr_files] = get_file(file); |
08a45173 JA |
6978 | unix_inflight(fpl->user, fpl->fp[nr_files]); |
6979 | nr_files++; | |
6b06314c JA |
6980 | } |
6981 | ||
08a45173 JA |
6982 | if (nr_files) { |
6983 | fpl->max = SCM_MAX_FD; | |
6984 | fpl->count = nr_files; | |
6985 | UNIXCB(skb).fp = fpl; | |
05f3fb3c | 6986 | skb->destructor = unix_destruct_scm; |
08a45173 JA |
6987 | refcount_add(skb->truesize, &sk->sk_wmem_alloc); |
6988 | skb_queue_head(&sk->sk_receive_queue, skb); | |
6b06314c | 6989 | |
08a45173 JA |
6990 | for (i = 0; i < nr_files; i++) |
6991 | fput(fpl->fp[i]); | |
6992 | } else { | |
6993 | kfree_skb(skb); | |
6994 | kfree(fpl); | |
6995 | } | |
6b06314c JA |
6996 | |
6997 | return 0; | |
6998 | } | |
6999 | ||
7000 | /* | |
7001 | * If UNIX sockets are enabled, fd passing can cause a reference cycle which | |
7002 | * causes regular reference counting to break down. We rely on the UNIX | |
7003 | * garbage collection to take care of this problem for us. | |
7004 | */ | |
7005 | static int io_sqe_files_scm(struct io_ring_ctx *ctx) | |
7006 | { | |
7007 | unsigned left, total; | |
7008 | int ret = 0; | |
7009 | ||
7010 | total = 0; | |
7011 | left = ctx->nr_user_files; | |
7012 | while (left) { | |
7013 | unsigned this_files = min_t(unsigned, left, SCM_MAX_FD); | |
6b06314c JA |
7014 | |
7015 | ret = __io_sqe_files_scm(ctx, this_files, total); | |
7016 | if (ret) | |
7017 | break; | |
7018 | left -= this_files; | |
7019 | total += this_files; | |
7020 | } | |
7021 | ||
7022 | if (!ret) | |
7023 | return 0; | |
7024 | ||
7025 | while (total < ctx->nr_user_files) { | |
65e19f54 JA |
7026 | struct file *file = io_file_from_index(ctx, total); |
7027 | ||
7028 | if (file) | |
7029 | fput(file); | |
6b06314c JA |
7030 | total++; |
7031 | } | |
7032 | ||
7033 | return ret; | |
7034 | } | |
7035 | #else | |
7036 | static int io_sqe_files_scm(struct io_ring_ctx *ctx) | |
7037 | { | |
7038 | return 0; | |
7039 | } | |
7040 | #endif | |
7041 | ||
65e19f54 JA |
7042 | static int io_sqe_alloc_file_tables(struct io_ring_ctx *ctx, unsigned nr_tables, |
7043 | unsigned nr_files) | |
7044 | { | |
7045 | int i; | |
7046 | ||
7047 | for (i = 0; i < nr_tables; i++) { | |
05f3fb3c | 7048 | struct fixed_file_table *table = &ctx->file_data->table[i]; |
65e19f54 JA |
7049 | unsigned this_files; |
7050 | ||
7051 | this_files = min(nr_files, IORING_MAX_FILES_TABLE); | |
7052 | table->files = kcalloc(this_files, sizeof(struct file *), | |
7053 | GFP_KERNEL); | |
7054 | if (!table->files) | |
7055 | break; | |
7056 | nr_files -= this_files; | |
7057 | } | |
7058 | ||
7059 | if (i == nr_tables) | |
7060 | return 0; | |
7061 | ||
7062 | for (i = 0; i < nr_tables; i++) { | |
05f3fb3c | 7063 | struct fixed_file_table *table = &ctx->file_data->table[i]; |
65e19f54 JA |
7064 | kfree(table->files); |
7065 | } | |
7066 | return 1; | |
7067 | } | |
7068 | ||
05f3fb3c JA |
7069 | static void io_ring_file_put(struct io_ring_ctx *ctx, struct file *file) |
7070 | { | |
7071 | #if defined(CONFIG_UNIX) | |
7072 | struct sock *sock = ctx->ring_sock->sk; | |
7073 | struct sk_buff_head list, *head = &sock->sk_receive_queue; | |
7074 | struct sk_buff *skb; | |
7075 | int i; | |
7076 | ||
7077 | __skb_queue_head_init(&list); | |
7078 | ||
7079 | /* | |
7080 | * Find the skb that holds this file in its SCM_RIGHTS. When found, | |
7081 | * remove this entry and rearrange the file array. | |
7082 | */ | |
7083 | skb = skb_dequeue(head); | |
7084 | while (skb) { | |
7085 | struct scm_fp_list *fp; | |
7086 | ||
7087 | fp = UNIXCB(skb).fp; | |
7088 | for (i = 0; i < fp->count; i++) { | |
7089 | int left; | |
7090 | ||
7091 | if (fp->fp[i] != file) | |
7092 | continue; | |
7093 | ||
7094 | unix_notinflight(fp->user, fp->fp[i]); | |
7095 | left = fp->count - 1 - i; | |
7096 | if (left) { | |
7097 | memmove(&fp->fp[i], &fp->fp[i + 1], | |
7098 | left * sizeof(struct file *)); | |
7099 | } | |
7100 | fp->count--; | |
7101 | if (!fp->count) { | |
7102 | kfree_skb(skb); | |
7103 | skb = NULL; | |
7104 | } else { | |
7105 | __skb_queue_tail(&list, skb); | |
7106 | } | |
7107 | fput(file); | |
7108 | file = NULL; | |
7109 | break; | |
7110 | } | |
7111 | ||
7112 | if (!file) | |
7113 | break; | |
7114 | ||
7115 | __skb_queue_tail(&list, skb); | |
7116 | ||
7117 | skb = skb_dequeue(head); | |
7118 | } | |
7119 | ||
7120 | if (skb_peek(&list)) { | |
7121 | spin_lock_irq(&head->lock); | |
7122 | while ((skb = __skb_dequeue(&list)) != NULL) | |
7123 | __skb_queue_tail(head, skb); | |
7124 | spin_unlock_irq(&head->lock); | |
7125 | } | |
7126 | #else | |
7127 | fput(file); | |
7128 | #endif | |
7129 | } | |
7130 | ||
7131 | struct io_file_put { | |
05589553 | 7132 | struct list_head list; |
05f3fb3c | 7133 | struct file *file; |
05f3fb3c JA |
7134 | }; |
7135 | ||
4a38aed2 | 7136 | static void __io_file_put_work(struct fixed_file_ref_node *ref_node) |
65e19f54 | 7137 | { |
4a38aed2 JA |
7138 | struct fixed_file_data *file_data = ref_node->file_data; |
7139 | struct io_ring_ctx *ctx = file_data->ctx; | |
05f3fb3c | 7140 | struct io_file_put *pfile, *tmp; |
05589553 XW |
7141 | |
7142 | list_for_each_entry_safe(pfile, tmp, &ref_node->file_list, list) { | |
6a4d07cd | 7143 | list_del(&pfile->list); |
05589553 XW |
7144 | io_ring_file_put(ctx, pfile->file); |
7145 | kfree(pfile); | |
65e19f54 | 7146 | } |
05589553 | 7147 | |
6a4d07cd JA |
7148 | spin_lock(&file_data->lock); |
7149 | list_del(&ref_node->node); | |
7150 | spin_unlock(&file_data->lock); | |
05589553 XW |
7151 | |
7152 | percpu_ref_exit(&ref_node->refs); | |
7153 | kfree(ref_node); | |
7154 | percpu_ref_put(&file_data->refs); | |
2faf852d | 7155 | } |
65e19f54 | 7156 | |
4a38aed2 JA |
7157 | static void io_file_put_work(struct work_struct *work) |
7158 | { | |
7159 | struct io_ring_ctx *ctx; | |
7160 | struct llist_node *node; | |
7161 | ||
7162 | ctx = container_of(work, struct io_ring_ctx, file_put_work.work); | |
7163 | node = llist_del_all(&ctx->file_put_llist); | |
7164 | ||
7165 | while (node) { | |
7166 | struct fixed_file_ref_node *ref_node; | |
7167 | struct llist_node *next = node->next; | |
7168 | ||
7169 | ref_node = llist_entry(node, struct fixed_file_ref_node, llist); | |
7170 | __io_file_put_work(ref_node); | |
7171 | node = next; | |
7172 | } | |
7173 | } | |
7174 | ||
05589553 | 7175 | static void io_file_data_ref_zero(struct percpu_ref *ref) |
2faf852d | 7176 | { |
05589553 | 7177 | struct fixed_file_ref_node *ref_node; |
4a38aed2 JA |
7178 | struct io_ring_ctx *ctx; |
7179 | bool first_add; | |
7180 | int delay = HZ; | |
65e19f54 | 7181 | |
05589553 | 7182 | ref_node = container_of(ref, struct fixed_file_ref_node, refs); |
4a38aed2 | 7183 | ctx = ref_node->file_data->ctx; |
05589553 | 7184 | |
4a38aed2 JA |
7185 | if (percpu_ref_is_dying(&ctx->file_data->refs)) |
7186 | delay = 0; | |
05589553 | 7187 | |
4a38aed2 JA |
7188 | first_add = llist_add(&ref_node->llist, &ctx->file_put_llist); |
7189 | if (!delay) | |
7190 | mod_delayed_work(system_wq, &ctx->file_put_work, 0); | |
7191 | else if (first_add) | |
7192 | queue_delayed_work(system_wq, &ctx->file_put_work, delay); | |
05f3fb3c | 7193 | } |
65e19f54 | 7194 | |
05589553 XW |
7195 | static struct fixed_file_ref_node *alloc_fixed_file_ref_node( |
7196 | struct io_ring_ctx *ctx) | |
05f3fb3c | 7197 | { |
05589553 | 7198 | struct fixed_file_ref_node *ref_node; |
05f3fb3c | 7199 | |
05589553 XW |
7200 | ref_node = kzalloc(sizeof(*ref_node), GFP_KERNEL); |
7201 | if (!ref_node) | |
7202 | return ERR_PTR(-ENOMEM); | |
05f3fb3c | 7203 | |
05589553 XW |
7204 | if (percpu_ref_init(&ref_node->refs, io_file_data_ref_zero, |
7205 | 0, GFP_KERNEL)) { | |
7206 | kfree(ref_node); | |
7207 | return ERR_PTR(-ENOMEM); | |
7208 | } | |
7209 | INIT_LIST_HEAD(&ref_node->node); | |
7210 | INIT_LIST_HEAD(&ref_node->file_list); | |
05589553 XW |
7211 | ref_node->file_data = ctx->file_data; |
7212 | return ref_node; | |
05589553 XW |
7213 | } |
7214 | ||
7215 | static void destroy_fixed_file_ref_node(struct fixed_file_ref_node *ref_node) | |
7216 | { | |
7217 | percpu_ref_exit(&ref_node->refs); | |
7218 | kfree(ref_node); | |
65e19f54 JA |
7219 | } |
7220 | ||
6b06314c JA |
7221 | static int io_sqe_files_register(struct io_ring_ctx *ctx, void __user *arg, |
7222 | unsigned nr_args) | |
7223 | { | |
7224 | __s32 __user *fds = (__s32 __user *) arg; | |
65e19f54 | 7225 | unsigned nr_tables; |
05f3fb3c | 7226 | struct file *file; |
6b06314c JA |
7227 | int fd, ret = 0; |
7228 | unsigned i; | |
05589553 | 7229 | struct fixed_file_ref_node *ref_node; |
6b06314c | 7230 | |
05f3fb3c | 7231 | if (ctx->file_data) |
6b06314c JA |
7232 | return -EBUSY; |
7233 | if (!nr_args) | |
7234 | return -EINVAL; | |
7235 | if (nr_args > IORING_MAX_FIXED_FILES) | |
7236 | return -EMFILE; | |
7237 | ||
05f3fb3c JA |
7238 | ctx->file_data = kzalloc(sizeof(*ctx->file_data), GFP_KERNEL); |
7239 | if (!ctx->file_data) | |
7240 | return -ENOMEM; | |
7241 | ctx->file_data->ctx = ctx; | |
7242 | init_completion(&ctx->file_data->done); | |
05589553 | 7243 | INIT_LIST_HEAD(&ctx->file_data->ref_list); |
f7fe9346 | 7244 | spin_lock_init(&ctx->file_data->lock); |
05f3fb3c | 7245 | |
65e19f54 | 7246 | nr_tables = DIV_ROUND_UP(nr_args, IORING_MAX_FILES_TABLE); |
05f3fb3c JA |
7247 | ctx->file_data->table = kcalloc(nr_tables, |
7248 | sizeof(struct fixed_file_table), | |
65e19f54 | 7249 | GFP_KERNEL); |
05f3fb3c JA |
7250 | if (!ctx->file_data->table) { |
7251 | kfree(ctx->file_data); | |
7252 | ctx->file_data = NULL; | |
6b06314c | 7253 | return -ENOMEM; |
05f3fb3c JA |
7254 | } |
7255 | ||
05589553 | 7256 | if (percpu_ref_init(&ctx->file_data->refs, io_file_ref_kill, |
05f3fb3c JA |
7257 | PERCPU_REF_ALLOW_REINIT, GFP_KERNEL)) { |
7258 | kfree(ctx->file_data->table); | |
7259 | kfree(ctx->file_data); | |
7260 | ctx->file_data = NULL; | |
6b06314c | 7261 | return -ENOMEM; |
05f3fb3c | 7262 | } |
6b06314c | 7263 | |
65e19f54 | 7264 | if (io_sqe_alloc_file_tables(ctx, nr_tables, nr_args)) { |
05f3fb3c JA |
7265 | percpu_ref_exit(&ctx->file_data->refs); |
7266 | kfree(ctx->file_data->table); | |
7267 | kfree(ctx->file_data); | |
7268 | ctx->file_data = NULL; | |
65e19f54 JA |
7269 | return -ENOMEM; |
7270 | } | |
7271 | ||
08a45173 | 7272 | for (i = 0; i < nr_args; i++, ctx->nr_user_files++) { |
65e19f54 JA |
7273 | struct fixed_file_table *table; |
7274 | unsigned index; | |
7275 | ||
6b06314c JA |
7276 | ret = -EFAULT; |
7277 | if (copy_from_user(&fd, &fds[i], sizeof(fd))) | |
7278 | break; | |
08a45173 JA |
7279 | /* allow sparse sets */ |
7280 | if (fd == -1) { | |
7281 | ret = 0; | |
7282 | continue; | |
7283 | } | |
6b06314c | 7284 | |
05f3fb3c | 7285 | table = &ctx->file_data->table[i >> IORING_FILE_TABLE_SHIFT]; |
65e19f54 | 7286 | index = i & IORING_FILE_TABLE_MASK; |
05f3fb3c | 7287 | file = fget(fd); |
6b06314c JA |
7288 | |
7289 | ret = -EBADF; | |
05f3fb3c | 7290 | if (!file) |
6b06314c | 7291 | break; |
05f3fb3c | 7292 | |
6b06314c JA |
7293 | /* |
7294 | * Don't allow io_uring instances to be registered. If UNIX | |
7295 | * isn't enabled, then this causes a reference cycle and this | |
7296 | * instance can never get freed. If UNIX is enabled we'll | |
7297 | * handle it just fine, but there's still no point in allowing | |
7298 | * a ring fd as it doesn't support regular read/write anyway. | |
7299 | */ | |
05f3fb3c JA |
7300 | if (file->f_op == &io_uring_fops) { |
7301 | fput(file); | |
6b06314c JA |
7302 | break; |
7303 | } | |
6b06314c | 7304 | ret = 0; |
05f3fb3c | 7305 | table->files[index] = file; |
6b06314c JA |
7306 | } |
7307 | ||
7308 | if (ret) { | |
65e19f54 | 7309 | for (i = 0; i < ctx->nr_user_files; i++) { |
65e19f54 JA |
7310 | file = io_file_from_index(ctx, i); |
7311 | if (file) | |
7312 | fput(file); | |
7313 | } | |
7314 | for (i = 0; i < nr_tables; i++) | |
05f3fb3c | 7315 | kfree(ctx->file_data->table[i].files); |
6b06314c | 7316 | |
667e57da | 7317 | percpu_ref_exit(&ctx->file_data->refs); |
05f3fb3c JA |
7318 | kfree(ctx->file_data->table); |
7319 | kfree(ctx->file_data); | |
7320 | ctx->file_data = NULL; | |
6b06314c JA |
7321 | ctx->nr_user_files = 0; |
7322 | return ret; | |
7323 | } | |
7324 | ||
7325 | ret = io_sqe_files_scm(ctx); | |
05589553 | 7326 | if (ret) { |
6b06314c | 7327 | io_sqe_files_unregister(ctx); |
05589553 XW |
7328 | return ret; |
7329 | } | |
6b06314c | 7330 | |
05589553 XW |
7331 | ref_node = alloc_fixed_file_ref_node(ctx); |
7332 | if (IS_ERR(ref_node)) { | |
7333 | io_sqe_files_unregister(ctx); | |
7334 | return PTR_ERR(ref_node); | |
7335 | } | |
7336 | ||
7337 | ctx->file_data->cur_refs = &ref_node->refs; | |
6a4d07cd | 7338 | spin_lock(&ctx->file_data->lock); |
05589553 | 7339 | list_add(&ref_node->node, &ctx->file_data->ref_list); |
6a4d07cd | 7340 | spin_unlock(&ctx->file_data->lock); |
05589553 | 7341 | percpu_ref_get(&ctx->file_data->refs); |
6b06314c JA |
7342 | return ret; |
7343 | } | |
7344 | ||
c3a31e60 JA |
7345 | static int io_sqe_file_register(struct io_ring_ctx *ctx, struct file *file, |
7346 | int index) | |
7347 | { | |
7348 | #if defined(CONFIG_UNIX) | |
7349 | struct sock *sock = ctx->ring_sock->sk; | |
7350 | struct sk_buff_head *head = &sock->sk_receive_queue; | |
7351 | struct sk_buff *skb; | |
7352 | ||
7353 | /* | |
7354 | * See if we can merge this file into an existing skb SCM_RIGHTS | |
7355 | * file set. If there's no room, fall back to allocating a new skb | |
7356 | * and filling it in. | |
7357 | */ | |
7358 | spin_lock_irq(&head->lock); | |
7359 | skb = skb_peek(head); | |
7360 | if (skb) { | |
7361 | struct scm_fp_list *fpl = UNIXCB(skb).fp; | |
7362 | ||
7363 | if (fpl->count < SCM_MAX_FD) { | |
7364 | __skb_unlink(skb, head); | |
7365 | spin_unlock_irq(&head->lock); | |
7366 | fpl->fp[fpl->count] = get_file(file); | |
7367 | unix_inflight(fpl->user, fpl->fp[fpl->count]); | |
7368 | fpl->count++; | |
7369 | spin_lock_irq(&head->lock); | |
7370 | __skb_queue_head(head, skb); | |
7371 | } else { | |
7372 | skb = NULL; | |
7373 | } | |
7374 | } | |
7375 | spin_unlock_irq(&head->lock); | |
7376 | ||
7377 | if (skb) { | |
7378 | fput(file); | |
7379 | return 0; | |
7380 | } | |
7381 | ||
7382 | return __io_sqe_files_scm(ctx, 1, index); | |
7383 | #else | |
7384 | return 0; | |
7385 | #endif | |
7386 | } | |
7387 | ||
a5318d3c | 7388 | static int io_queue_file_removal(struct fixed_file_data *data, |
05589553 | 7389 | struct file *file) |
05f3fb3c | 7390 | { |
a5318d3c | 7391 | struct io_file_put *pfile; |
05589553 XW |
7392 | struct percpu_ref *refs = data->cur_refs; |
7393 | struct fixed_file_ref_node *ref_node; | |
05f3fb3c | 7394 | |
05f3fb3c | 7395 | pfile = kzalloc(sizeof(*pfile), GFP_KERNEL); |
a5318d3c HD |
7396 | if (!pfile) |
7397 | return -ENOMEM; | |
05f3fb3c | 7398 | |
05589553 | 7399 | ref_node = container_of(refs, struct fixed_file_ref_node, refs); |
05f3fb3c | 7400 | pfile->file = file; |
05589553 XW |
7401 | list_add(&pfile->list, &ref_node->file_list); |
7402 | ||
a5318d3c | 7403 | return 0; |
05f3fb3c JA |
7404 | } |
7405 | ||
7406 | static int __io_sqe_files_update(struct io_ring_ctx *ctx, | |
7407 | struct io_uring_files_update *up, | |
7408 | unsigned nr_args) | |
7409 | { | |
7410 | struct fixed_file_data *data = ctx->file_data; | |
05589553 | 7411 | struct fixed_file_ref_node *ref_node; |
05f3fb3c | 7412 | struct file *file; |
c3a31e60 JA |
7413 | __s32 __user *fds; |
7414 | int fd, i, err; | |
7415 | __u32 done; | |
05589553 | 7416 | bool needs_switch = false; |
c3a31e60 | 7417 | |
05f3fb3c | 7418 | if (check_add_overflow(up->offset, nr_args, &done)) |
c3a31e60 JA |
7419 | return -EOVERFLOW; |
7420 | if (done > ctx->nr_user_files) | |
7421 | return -EINVAL; | |
7422 | ||
05589553 XW |
7423 | ref_node = alloc_fixed_file_ref_node(ctx); |
7424 | if (IS_ERR(ref_node)) | |
7425 | return PTR_ERR(ref_node); | |
7426 | ||
c3a31e60 | 7427 | done = 0; |
05f3fb3c | 7428 | fds = u64_to_user_ptr(up->fds); |
c3a31e60 | 7429 | while (nr_args) { |
65e19f54 JA |
7430 | struct fixed_file_table *table; |
7431 | unsigned index; | |
7432 | ||
c3a31e60 JA |
7433 | err = 0; |
7434 | if (copy_from_user(&fd, &fds[done], sizeof(fd))) { | |
7435 | err = -EFAULT; | |
7436 | break; | |
7437 | } | |
05f3fb3c JA |
7438 | i = array_index_nospec(up->offset, ctx->nr_user_files); |
7439 | table = &ctx->file_data->table[i >> IORING_FILE_TABLE_SHIFT]; | |
65e19f54 JA |
7440 | index = i & IORING_FILE_TABLE_MASK; |
7441 | if (table->files[index]) { | |
98dfd502 | 7442 | file = table->files[index]; |
a5318d3c HD |
7443 | err = io_queue_file_removal(data, file); |
7444 | if (err) | |
7445 | break; | |
65e19f54 | 7446 | table->files[index] = NULL; |
05589553 | 7447 | needs_switch = true; |
c3a31e60 JA |
7448 | } |
7449 | if (fd != -1) { | |
c3a31e60 JA |
7450 | file = fget(fd); |
7451 | if (!file) { | |
7452 | err = -EBADF; | |
7453 | break; | |
7454 | } | |
7455 | /* | |
7456 | * Don't allow io_uring instances to be registered. If | |
7457 | * UNIX isn't enabled, then this causes a reference | |
7458 | * cycle and this instance can never get freed. If UNIX | |
7459 | * is enabled we'll handle it just fine, but there's | |
7460 | * still no point in allowing a ring fd as it doesn't | |
7461 | * support regular read/write anyway. | |
7462 | */ | |
7463 | if (file->f_op == &io_uring_fops) { | |
7464 | fput(file); | |
7465 | err = -EBADF; | |
7466 | break; | |
7467 | } | |
65e19f54 | 7468 | table->files[index] = file; |
c3a31e60 | 7469 | err = io_sqe_file_register(ctx, file, i); |
f3bd9dae | 7470 | if (err) { |
95d1c8e5 | 7471 | table->files[index] = NULL; |
f3bd9dae | 7472 | fput(file); |
c3a31e60 | 7473 | break; |
f3bd9dae | 7474 | } |
c3a31e60 JA |
7475 | } |
7476 | nr_args--; | |
7477 | done++; | |
05f3fb3c JA |
7478 | up->offset++; |
7479 | } | |
7480 | ||
05589553 XW |
7481 | if (needs_switch) { |
7482 | percpu_ref_kill(data->cur_refs); | |
6a4d07cd | 7483 | spin_lock(&data->lock); |
05589553 XW |
7484 | list_add(&ref_node->node, &data->ref_list); |
7485 | data->cur_refs = &ref_node->refs; | |
6a4d07cd | 7486 | spin_unlock(&data->lock); |
05589553 XW |
7487 | percpu_ref_get(&ctx->file_data->refs); |
7488 | } else | |
7489 | destroy_fixed_file_ref_node(ref_node); | |
c3a31e60 JA |
7490 | |
7491 | return done ? done : err; | |
7492 | } | |
05589553 | 7493 | |
05f3fb3c JA |
7494 | static int io_sqe_files_update(struct io_ring_ctx *ctx, void __user *arg, |
7495 | unsigned nr_args) | |
7496 | { | |
7497 | struct io_uring_files_update up; | |
7498 | ||
7499 | if (!ctx->file_data) | |
7500 | return -ENXIO; | |
7501 | if (!nr_args) | |
7502 | return -EINVAL; | |
7503 | if (copy_from_user(&up, arg, sizeof(up))) | |
7504 | return -EFAULT; | |
7505 | if (up.resv) | |
7506 | return -EINVAL; | |
7507 | ||
7508 | return __io_sqe_files_update(ctx, &up, nr_args); | |
7509 | } | |
c3a31e60 | 7510 | |
e9fd9396 | 7511 | static void io_free_work(struct io_wq_work *work) |
7d723065 JA |
7512 | { |
7513 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); | |
7514 | ||
e9fd9396 | 7515 | /* Consider that io_steal_work() relies on this ref */ |
7d723065 JA |
7516 | io_put_req(req); |
7517 | } | |
7518 | ||
24369c2e PB |
7519 | static int io_init_wq_offload(struct io_ring_ctx *ctx, |
7520 | struct io_uring_params *p) | |
7521 | { | |
7522 | struct io_wq_data data; | |
7523 | struct fd f; | |
7524 | struct io_ring_ctx *ctx_attach; | |
7525 | unsigned int concurrency; | |
7526 | int ret = 0; | |
7527 | ||
7528 | data.user = ctx->user; | |
e9fd9396 | 7529 | data.free_work = io_free_work; |
f5fa38c5 | 7530 | data.do_work = io_wq_submit_work; |
24369c2e PB |
7531 | |
7532 | if (!(p->flags & IORING_SETUP_ATTACH_WQ)) { | |
7533 | /* Do QD, or 4 * CPUS, whatever is smallest */ | |
7534 | concurrency = min(ctx->sq_entries, 4 * num_online_cpus()); | |
7535 | ||
7536 | ctx->io_wq = io_wq_create(concurrency, &data); | |
7537 | if (IS_ERR(ctx->io_wq)) { | |
7538 | ret = PTR_ERR(ctx->io_wq); | |
7539 | ctx->io_wq = NULL; | |
7540 | } | |
7541 | return ret; | |
7542 | } | |
7543 | ||
7544 | f = fdget(p->wq_fd); | |
7545 | if (!f.file) | |
7546 | return -EBADF; | |
7547 | ||
7548 | if (f.file->f_op != &io_uring_fops) { | |
7549 | ret = -EINVAL; | |
7550 | goto out_fput; | |
7551 | } | |
7552 | ||
7553 | ctx_attach = f.file->private_data; | |
7554 | /* @io_wq is protected by holding the fd */ | |
7555 | if (!io_wq_get(ctx_attach->io_wq, &data)) { | |
7556 | ret = -EINVAL; | |
7557 | goto out_fput; | |
7558 | } | |
7559 | ||
7560 | ctx->io_wq = ctx_attach->io_wq; | |
7561 | out_fput: | |
7562 | fdput(f); | |
7563 | return ret; | |
7564 | } | |
7565 | ||
0f212204 JA |
7566 | static int io_uring_alloc_task_context(struct task_struct *task) |
7567 | { | |
7568 | struct io_uring_task *tctx; | |
7569 | ||
7570 | tctx = kmalloc(sizeof(*tctx), GFP_KERNEL); | |
7571 | if (unlikely(!tctx)) | |
7572 | return -ENOMEM; | |
7573 | ||
7574 | xa_init(&tctx->xa); | |
7575 | init_waitqueue_head(&tctx->wait); | |
7576 | tctx->last = NULL; | |
7577 | tctx->in_idle = 0; | |
7578 | atomic_long_set(&tctx->req_issue, 0); | |
7579 | atomic_long_set(&tctx->req_complete, 0); | |
7580 | task->io_uring = tctx; | |
7581 | return 0; | |
7582 | } | |
7583 | ||
7584 | void __io_uring_free(struct task_struct *tsk) | |
7585 | { | |
7586 | struct io_uring_task *tctx = tsk->io_uring; | |
7587 | ||
7588 | WARN_ON_ONCE(!xa_empty(&tctx->xa)); | |
7589 | xa_destroy(&tctx->xa); | |
7590 | kfree(tctx); | |
7591 | tsk->io_uring = NULL; | |
7592 | } | |
7593 | ||
7e84e1c7 SG |
7594 | static int io_sq_offload_create(struct io_ring_ctx *ctx, |
7595 | struct io_uring_params *p) | |
2b188cc1 JA |
7596 | { |
7597 | int ret; | |
7598 | ||
6c271ce2 | 7599 | if (ctx->flags & IORING_SETUP_SQPOLL) { |
3ec482d1 JA |
7600 | ret = -EPERM; |
7601 | if (!capable(CAP_SYS_ADMIN)) | |
7602 | goto err; | |
7603 | ||
917257da JA |
7604 | ctx->sq_thread_idle = msecs_to_jiffies(p->sq_thread_idle); |
7605 | if (!ctx->sq_thread_idle) | |
7606 | ctx->sq_thread_idle = HZ; | |
7607 | ||
6c271ce2 | 7608 | if (p->flags & IORING_SETUP_SQ_AFF) { |
44a9bd18 | 7609 | int cpu = p->sq_thread_cpu; |
6c271ce2 | 7610 | |
917257da | 7611 | ret = -EINVAL; |
44a9bd18 JA |
7612 | if (cpu >= nr_cpu_ids) |
7613 | goto err; | |
7889f44d | 7614 | if (!cpu_online(cpu)) |
917257da JA |
7615 | goto err; |
7616 | ||
6c271ce2 JA |
7617 | ctx->sqo_thread = kthread_create_on_cpu(io_sq_thread, |
7618 | ctx, cpu, | |
7619 | "io_uring-sq"); | |
7620 | } else { | |
7621 | ctx->sqo_thread = kthread_create(io_sq_thread, ctx, | |
7622 | "io_uring-sq"); | |
7623 | } | |
7624 | if (IS_ERR(ctx->sqo_thread)) { | |
7625 | ret = PTR_ERR(ctx->sqo_thread); | |
7626 | ctx->sqo_thread = NULL; | |
7627 | goto err; | |
7628 | } | |
0f212204 JA |
7629 | ret = io_uring_alloc_task_context(ctx->sqo_thread); |
7630 | if (ret) | |
7631 | goto err; | |
6c271ce2 JA |
7632 | } else if (p->flags & IORING_SETUP_SQ_AFF) { |
7633 | /* Can't have SQ_AFF without SQPOLL */ | |
7634 | ret = -EINVAL; | |
7635 | goto err; | |
7636 | } | |
7637 | ||
24369c2e PB |
7638 | ret = io_init_wq_offload(ctx, p); |
7639 | if (ret) | |
2b188cc1 | 7640 | goto err; |
2b188cc1 JA |
7641 | |
7642 | return 0; | |
7643 | err: | |
54a91f3b | 7644 | io_finish_async(ctx); |
2b188cc1 JA |
7645 | return ret; |
7646 | } | |
7647 | ||
7e84e1c7 SG |
7648 | static void io_sq_offload_start(struct io_ring_ctx *ctx) |
7649 | { | |
7650 | if ((ctx->flags & IORING_SETUP_SQPOLL) && ctx->sqo_thread) | |
7651 | wake_up_process(ctx->sqo_thread); | |
7652 | } | |
7653 | ||
a087e2b5 BM |
7654 | static inline void __io_unaccount_mem(struct user_struct *user, |
7655 | unsigned long nr_pages) | |
2b188cc1 JA |
7656 | { |
7657 | atomic_long_sub(nr_pages, &user->locked_vm); | |
7658 | } | |
7659 | ||
a087e2b5 BM |
7660 | static inline int __io_account_mem(struct user_struct *user, |
7661 | unsigned long nr_pages) | |
2b188cc1 JA |
7662 | { |
7663 | unsigned long page_limit, cur_pages, new_pages; | |
7664 | ||
7665 | /* Don't allow more pages than we can safely lock */ | |
7666 | page_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT; | |
7667 | ||
7668 | do { | |
7669 | cur_pages = atomic_long_read(&user->locked_vm); | |
7670 | new_pages = cur_pages + nr_pages; | |
7671 | if (new_pages > page_limit) | |
7672 | return -ENOMEM; | |
7673 | } while (atomic_long_cmpxchg(&user->locked_vm, cur_pages, | |
7674 | new_pages) != cur_pages); | |
7675 | ||
7676 | return 0; | |
7677 | } | |
7678 | ||
2e0464d4 BM |
7679 | static void io_unaccount_mem(struct io_ring_ctx *ctx, unsigned long nr_pages, |
7680 | enum io_mem_account acct) | |
a087e2b5 | 7681 | { |
aad5d8da | 7682 | if (ctx->limit_mem) |
a087e2b5 | 7683 | __io_unaccount_mem(ctx->user, nr_pages); |
30975825 | 7684 | |
2aede0e4 | 7685 | if (ctx->mm_account) { |
2e0464d4 | 7686 | if (acct == ACCT_LOCKED) |
2aede0e4 | 7687 | ctx->mm_account->locked_vm -= nr_pages; |
2e0464d4 | 7688 | else if (acct == ACCT_PINNED) |
2aede0e4 | 7689 | atomic64_sub(nr_pages, &ctx->mm_account->pinned_vm); |
2e0464d4 | 7690 | } |
a087e2b5 BM |
7691 | } |
7692 | ||
2e0464d4 BM |
7693 | static int io_account_mem(struct io_ring_ctx *ctx, unsigned long nr_pages, |
7694 | enum io_mem_account acct) | |
a087e2b5 | 7695 | { |
30975825 BM |
7696 | int ret; |
7697 | ||
7698 | if (ctx->limit_mem) { | |
7699 | ret = __io_account_mem(ctx->user, nr_pages); | |
7700 | if (ret) | |
7701 | return ret; | |
7702 | } | |
7703 | ||
2aede0e4 | 7704 | if (ctx->mm_account) { |
2e0464d4 | 7705 | if (acct == ACCT_LOCKED) |
2aede0e4 | 7706 | ctx->mm_account->locked_vm += nr_pages; |
2e0464d4 | 7707 | else if (acct == ACCT_PINNED) |
2aede0e4 | 7708 | atomic64_add(nr_pages, &ctx->mm_account->pinned_vm); |
2e0464d4 | 7709 | } |
a087e2b5 BM |
7710 | |
7711 | return 0; | |
7712 | } | |
7713 | ||
2b188cc1 JA |
7714 | static void io_mem_free(void *ptr) |
7715 | { | |
52e04ef4 MR |
7716 | struct page *page; |
7717 | ||
7718 | if (!ptr) | |
7719 | return; | |
2b188cc1 | 7720 | |
52e04ef4 | 7721 | page = virt_to_head_page(ptr); |
2b188cc1 JA |
7722 | if (put_page_testzero(page)) |
7723 | free_compound_page(page); | |
7724 | } | |
7725 | ||
7726 | static void *io_mem_alloc(size_t size) | |
7727 | { | |
7728 | gfp_t gfp_flags = GFP_KERNEL | __GFP_ZERO | __GFP_NOWARN | __GFP_COMP | | |
7729 | __GFP_NORETRY; | |
7730 | ||
7731 | return (void *) __get_free_pages(gfp_flags, get_order(size)); | |
7732 | } | |
7733 | ||
75b28aff HV |
7734 | static unsigned long rings_size(unsigned sq_entries, unsigned cq_entries, |
7735 | size_t *sq_offset) | |
7736 | { | |
7737 | struct io_rings *rings; | |
7738 | size_t off, sq_array_size; | |
7739 | ||
7740 | off = struct_size(rings, cqes, cq_entries); | |
7741 | if (off == SIZE_MAX) | |
7742 | return SIZE_MAX; | |
7743 | ||
7744 | #ifdef CONFIG_SMP | |
7745 | off = ALIGN(off, SMP_CACHE_BYTES); | |
7746 | if (off == 0) | |
7747 | return SIZE_MAX; | |
7748 | #endif | |
7749 | ||
b36200f5 DV |
7750 | if (sq_offset) |
7751 | *sq_offset = off; | |
7752 | ||
75b28aff HV |
7753 | sq_array_size = array_size(sizeof(u32), sq_entries); |
7754 | if (sq_array_size == SIZE_MAX) | |
7755 | return SIZE_MAX; | |
7756 | ||
7757 | if (check_add_overflow(off, sq_array_size, &off)) | |
7758 | return SIZE_MAX; | |
7759 | ||
75b28aff HV |
7760 | return off; |
7761 | } | |
7762 | ||
2b188cc1 JA |
7763 | static unsigned long ring_pages(unsigned sq_entries, unsigned cq_entries) |
7764 | { | |
75b28aff | 7765 | size_t pages; |
2b188cc1 | 7766 | |
75b28aff HV |
7767 | pages = (size_t)1 << get_order( |
7768 | rings_size(sq_entries, cq_entries, NULL)); | |
7769 | pages += (size_t)1 << get_order( | |
7770 | array_size(sizeof(struct io_uring_sqe), sq_entries)); | |
2b188cc1 | 7771 | |
75b28aff | 7772 | return pages; |
2b188cc1 JA |
7773 | } |
7774 | ||
edafccee JA |
7775 | static int io_sqe_buffer_unregister(struct io_ring_ctx *ctx) |
7776 | { | |
7777 | int i, j; | |
7778 | ||
7779 | if (!ctx->user_bufs) | |
7780 | return -ENXIO; | |
7781 | ||
7782 | for (i = 0; i < ctx->nr_user_bufs; i++) { | |
7783 | struct io_mapped_ubuf *imu = &ctx->user_bufs[i]; | |
7784 | ||
7785 | for (j = 0; j < imu->nr_bvecs; j++) | |
f1f6a7dd | 7786 | unpin_user_page(imu->bvec[j].bv_page); |
edafccee | 7787 | |
2e0464d4 | 7788 | io_unaccount_mem(ctx, imu->nr_bvecs, ACCT_PINNED); |
d4ef6475 | 7789 | kvfree(imu->bvec); |
edafccee JA |
7790 | imu->nr_bvecs = 0; |
7791 | } | |
7792 | ||
7793 | kfree(ctx->user_bufs); | |
7794 | ctx->user_bufs = NULL; | |
7795 | ctx->nr_user_bufs = 0; | |
7796 | return 0; | |
7797 | } | |
7798 | ||
7799 | static int io_copy_iov(struct io_ring_ctx *ctx, struct iovec *dst, | |
7800 | void __user *arg, unsigned index) | |
7801 | { | |
7802 | struct iovec __user *src; | |
7803 | ||
7804 | #ifdef CONFIG_COMPAT | |
7805 | if (ctx->compat) { | |
7806 | struct compat_iovec __user *ciovs; | |
7807 | struct compat_iovec ciov; | |
7808 | ||
7809 | ciovs = (struct compat_iovec __user *) arg; | |
7810 | if (copy_from_user(&ciov, &ciovs[index], sizeof(ciov))) | |
7811 | return -EFAULT; | |
7812 | ||
d55e5f5b | 7813 | dst->iov_base = u64_to_user_ptr((u64)ciov.iov_base); |
edafccee JA |
7814 | dst->iov_len = ciov.iov_len; |
7815 | return 0; | |
7816 | } | |
7817 | #endif | |
7818 | src = (struct iovec __user *) arg; | |
7819 | if (copy_from_user(dst, &src[index], sizeof(*dst))) | |
7820 | return -EFAULT; | |
7821 | return 0; | |
7822 | } | |
7823 | ||
7824 | static int io_sqe_buffer_register(struct io_ring_ctx *ctx, void __user *arg, | |
7825 | unsigned nr_args) | |
7826 | { | |
7827 | struct vm_area_struct **vmas = NULL; | |
7828 | struct page **pages = NULL; | |
7829 | int i, j, got_pages = 0; | |
7830 | int ret = -EINVAL; | |
7831 | ||
7832 | if (ctx->user_bufs) | |
7833 | return -EBUSY; | |
7834 | if (!nr_args || nr_args > UIO_MAXIOV) | |
7835 | return -EINVAL; | |
7836 | ||
7837 | ctx->user_bufs = kcalloc(nr_args, sizeof(struct io_mapped_ubuf), | |
7838 | GFP_KERNEL); | |
7839 | if (!ctx->user_bufs) | |
7840 | return -ENOMEM; | |
7841 | ||
7842 | for (i = 0; i < nr_args; i++) { | |
7843 | struct io_mapped_ubuf *imu = &ctx->user_bufs[i]; | |
7844 | unsigned long off, start, end, ubuf; | |
7845 | int pret, nr_pages; | |
7846 | struct iovec iov; | |
7847 | size_t size; | |
7848 | ||
7849 | ret = io_copy_iov(ctx, &iov, arg, i); | |
7850 | if (ret) | |
a278682d | 7851 | goto err; |
edafccee JA |
7852 | |
7853 | /* | |
7854 | * Don't impose further limits on the size and buffer | |
7855 | * constraints here, we'll -EINVAL later when IO is | |
7856 | * submitted if they are wrong. | |
7857 | */ | |
7858 | ret = -EFAULT; | |
7859 | if (!iov.iov_base || !iov.iov_len) | |
7860 | goto err; | |
7861 | ||
7862 | /* arbitrary limit, but we need something */ | |
7863 | if (iov.iov_len > SZ_1G) | |
7864 | goto err; | |
7865 | ||
7866 | ubuf = (unsigned long) iov.iov_base; | |
7867 | end = (ubuf + iov.iov_len + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
7868 | start = ubuf >> PAGE_SHIFT; | |
7869 | nr_pages = end - start; | |
7870 | ||
2e0464d4 | 7871 | ret = io_account_mem(ctx, nr_pages, ACCT_PINNED); |
a087e2b5 BM |
7872 | if (ret) |
7873 | goto err; | |
edafccee JA |
7874 | |
7875 | ret = 0; | |
7876 | if (!pages || nr_pages > got_pages) { | |
a8c73c1a DE |
7877 | kvfree(vmas); |
7878 | kvfree(pages); | |
d4ef6475 | 7879 | pages = kvmalloc_array(nr_pages, sizeof(struct page *), |
edafccee | 7880 | GFP_KERNEL); |
d4ef6475 | 7881 | vmas = kvmalloc_array(nr_pages, |
edafccee JA |
7882 | sizeof(struct vm_area_struct *), |
7883 | GFP_KERNEL); | |
7884 | if (!pages || !vmas) { | |
7885 | ret = -ENOMEM; | |
2e0464d4 | 7886 | io_unaccount_mem(ctx, nr_pages, ACCT_PINNED); |
edafccee JA |
7887 | goto err; |
7888 | } | |
7889 | got_pages = nr_pages; | |
7890 | } | |
7891 | ||
d4ef6475 | 7892 | imu->bvec = kvmalloc_array(nr_pages, sizeof(struct bio_vec), |
edafccee JA |
7893 | GFP_KERNEL); |
7894 | ret = -ENOMEM; | |
7895 | if (!imu->bvec) { | |
2e0464d4 | 7896 | io_unaccount_mem(ctx, nr_pages, ACCT_PINNED); |
edafccee JA |
7897 | goto err; |
7898 | } | |
7899 | ||
7900 | ret = 0; | |
d8ed45c5 | 7901 | mmap_read_lock(current->mm); |
2113b05d | 7902 | pret = pin_user_pages(ubuf, nr_pages, |
932f4a63 IW |
7903 | FOLL_WRITE | FOLL_LONGTERM, |
7904 | pages, vmas); | |
edafccee JA |
7905 | if (pret == nr_pages) { |
7906 | /* don't support file backed memory */ | |
7907 | for (j = 0; j < nr_pages; j++) { | |
7908 | struct vm_area_struct *vma = vmas[j]; | |
7909 | ||
7910 | if (vma->vm_file && | |
7911 | !is_file_hugepages(vma->vm_file)) { | |
7912 | ret = -EOPNOTSUPP; | |
7913 | break; | |
7914 | } | |
7915 | } | |
7916 | } else { | |
7917 | ret = pret < 0 ? pret : -EFAULT; | |
7918 | } | |
d8ed45c5 | 7919 | mmap_read_unlock(current->mm); |
edafccee JA |
7920 | if (ret) { |
7921 | /* | |
7922 | * if we did partial map, or found file backed vmas, | |
7923 | * release any pages we did get | |
7924 | */ | |
27c4d3a3 | 7925 | if (pret > 0) |
f1f6a7dd | 7926 | unpin_user_pages(pages, pret); |
2e0464d4 | 7927 | io_unaccount_mem(ctx, nr_pages, ACCT_PINNED); |
d4ef6475 | 7928 | kvfree(imu->bvec); |
edafccee JA |
7929 | goto err; |
7930 | } | |
7931 | ||
7932 | off = ubuf & ~PAGE_MASK; | |
7933 | size = iov.iov_len; | |
7934 | for (j = 0; j < nr_pages; j++) { | |
7935 | size_t vec_len; | |
7936 | ||
7937 | vec_len = min_t(size_t, size, PAGE_SIZE - off); | |
7938 | imu->bvec[j].bv_page = pages[j]; | |
7939 | imu->bvec[j].bv_len = vec_len; | |
7940 | imu->bvec[j].bv_offset = off; | |
7941 | off = 0; | |
7942 | size -= vec_len; | |
7943 | } | |
7944 | /* store original address for later verification */ | |
7945 | imu->ubuf = ubuf; | |
7946 | imu->len = iov.iov_len; | |
7947 | imu->nr_bvecs = nr_pages; | |
7948 | ||
7949 | ctx->nr_user_bufs++; | |
7950 | } | |
d4ef6475 MR |
7951 | kvfree(pages); |
7952 | kvfree(vmas); | |
edafccee JA |
7953 | return 0; |
7954 | err: | |
d4ef6475 MR |
7955 | kvfree(pages); |
7956 | kvfree(vmas); | |
edafccee JA |
7957 | io_sqe_buffer_unregister(ctx); |
7958 | return ret; | |
7959 | } | |
7960 | ||
9b402849 JA |
7961 | static int io_eventfd_register(struct io_ring_ctx *ctx, void __user *arg) |
7962 | { | |
7963 | __s32 __user *fds = arg; | |
7964 | int fd; | |
7965 | ||
7966 | if (ctx->cq_ev_fd) | |
7967 | return -EBUSY; | |
7968 | ||
7969 | if (copy_from_user(&fd, fds, sizeof(*fds))) | |
7970 | return -EFAULT; | |
7971 | ||
7972 | ctx->cq_ev_fd = eventfd_ctx_fdget(fd); | |
7973 | if (IS_ERR(ctx->cq_ev_fd)) { | |
7974 | int ret = PTR_ERR(ctx->cq_ev_fd); | |
7975 | ctx->cq_ev_fd = NULL; | |
7976 | return ret; | |
7977 | } | |
7978 | ||
7979 | return 0; | |
7980 | } | |
7981 | ||
7982 | static int io_eventfd_unregister(struct io_ring_ctx *ctx) | |
7983 | { | |
7984 | if (ctx->cq_ev_fd) { | |
7985 | eventfd_ctx_put(ctx->cq_ev_fd); | |
7986 | ctx->cq_ev_fd = NULL; | |
7987 | return 0; | |
7988 | } | |
7989 | ||
7990 | return -ENXIO; | |
7991 | } | |
7992 | ||
5a2e745d JA |
7993 | static int __io_destroy_buffers(int id, void *p, void *data) |
7994 | { | |
7995 | struct io_ring_ctx *ctx = data; | |
7996 | struct io_buffer *buf = p; | |
7997 | ||
067524e9 | 7998 | __io_remove_buffers(ctx, buf, id, -1U); |
5a2e745d JA |
7999 | return 0; |
8000 | } | |
8001 | ||
8002 | static void io_destroy_buffers(struct io_ring_ctx *ctx) | |
8003 | { | |
8004 | idr_for_each(&ctx->io_buffer_idr, __io_destroy_buffers, ctx); | |
8005 | idr_destroy(&ctx->io_buffer_idr); | |
8006 | } | |
8007 | ||
2b188cc1 JA |
8008 | static void io_ring_ctx_free(struct io_ring_ctx *ctx) |
8009 | { | |
6b06314c | 8010 | io_finish_async(ctx); |
5dbcad51 | 8011 | io_sqe_buffer_unregister(ctx); |
2aede0e4 JA |
8012 | |
8013 | if (ctx->sqo_task) { | |
8014 | put_task_struct(ctx->sqo_task); | |
8015 | ctx->sqo_task = NULL; | |
8016 | mmdrop(ctx->mm_account); | |
8017 | ctx->mm_account = NULL; | |
30975825 | 8018 | } |
def596e9 | 8019 | |
6b06314c | 8020 | io_sqe_files_unregister(ctx); |
9b402849 | 8021 | io_eventfd_unregister(ctx); |
5a2e745d | 8022 | io_destroy_buffers(ctx); |
41726c9a | 8023 | idr_destroy(&ctx->personality_idr); |
def596e9 | 8024 | |
2b188cc1 | 8025 | #if defined(CONFIG_UNIX) |
355e8d26 EB |
8026 | if (ctx->ring_sock) { |
8027 | ctx->ring_sock->file = NULL; /* so that iput() is called */ | |
2b188cc1 | 8028 | sock_release(ctx->ring_sock); |
355e8d26 | 8029 | } |
2b188cc1 JA |
8030 | #endif |
8031 | ||
75b28aff | 8032 | io_mem_free(ctx->rings); |
2b188cc1 | 8033 | io_mem_free(ctx->sq_sqes); |
2b188cc1 JA |
8034 | |
8035 | percpu_ref_exit(&ctx->refs); | |
2b188cc1 | 8036 | free_uid(ctx->user); |
181e448d | 8037 | put_cred(ctx->creds); |
78076bb6 | 8038 | kfree(ctx->cancel_hash); |
0ddf92e8 | 8039 | kmem_cache_free(req_cachep, ctx->fallback_req); |
2b188cc1 JA |
8040 | kfree(ctx); |
8041 | } | |
8042 | ||
8043 | static __poll_t io_uring_poll(struct file *file, poll_table *wait) | |
8044 | { | |
8045 | struct io_ring_ctx *ctx = file->private_data; | |
8046 | __poll_t mask = 0; | |
8047 | ||
8048 | poll_wait(file, &ctx->cq_wait, wait); | |
4f7067c3 SB |
8049 | /* |
8050 | * synchronizes with barrier from wq_has_sleeper call in | |
8051 | * io_commit_cqring | |
8052 | */ | |
2b188cc1 | 8053 | smp_rmb(); |
75b28aff HV |
8054 | if (READ_ONCE(ctx->rings->sq.tail) - ctx->cached_sq_head != |
8055 | ctx->rings->sq_ring_entries) | |
2b188cc1 | 8056 | mask |= EPOLLOUT | EPOLLWRNORM; |
63e5d81f | 8057 | if (io_cqring_events(ctx, false)) |
2b188cc1 JA |
8058 | mask |= EPOLLIN | EPOLLRDNORM; |
8059 | ||
8060 | return mask; | |
8061 | } | |
8062 | ||
8063 | static int io_uring_fasync(int fd, struct file *file, int on) | |
8064 | { | |
8065 | struct io_ring_ctx *ctx = file->private_data; | |
8066 | ||
8067 | return fasync_helper(fd, file, on, &ctx->cq_fasync); | |
8068 | } | |
8069 | ||
071698e1 JA |
8070 | static int io_remove_personalities(int id, void *p, void *data) |
8071 | { | |
8072 | struct io_ring_ctx *ctx = data; | |
8073 | const struct cred *cred; | |
8074 | ||
8075 | cred = idr_remove(&ctx->personality_idr, id); | |
8076 | if (cred) | |
8077 | put_cred(cred); | |
8078 | return 0; | |
8079 | } | |
8080 | ||
85faa7b8 JA |
8081 | static void io_ring_exit_work(struct work_struct *work) |
8082 | { | |
b2edc0a7 PB |
8083 | struct io_ring_ctx *ctx = container_of(work, struct io_ring_ctx, |
8084 | exit_work); | |
85faa7b8 | 8085 | |
56952e91 JA |
8086 | /* |
8087 | * If we're doing polled IO and end up having requests being | |
8088 | * submitted async (out-of-line), then completions can come in while | |
8089 | * we're waiting for refs to drop. We need to reap these manually, | |
8090 | * as nobody else will be looking for them. | |
8091 | */ | |
b2edc0a7 | 8092 | do { |
56952e91 | 8093 | if (ctx->rings) |
e6c8aa9a | 8094 | io_cqring_overflow_flush(ctx, true, NULL, NULL); |
b2edc0a7 PB |
8095 | io_iopoll_try_reap_events(ctx); |
8096 | } while (!wait_for_completion_timeout(&ctx->ref_comp, HZ/20)); | |
85faa7b8 JA |
8097 | io_ring_ctx_free(ctx); |
8098 | } | |
8099 | ||
2b188cc1 JA |
8100 | static void io_ring_ctx_wait_and_kill(struct io_ring_ctx *ctx) |
8101 | { | |
8102 | mutex_lock(&ctx->uring_lock); | |
8103 | percpu_ref_kill(&ctx->refs); | |
8104 | mutex_unlock(&ctx->uring_lock); | |
8105 | ||
f3606e3a JA |
8106 | io_kill_timeouts(ctx, NULL); |
8107 | io_poll_remove_all(ctx, NULL); | |
561fb04a JA |
8108 | |
8109 | if (ctx->io_wq) | |
8110 | io_wq_cancel_all(ctx->io_wq); | |
8111 | ||
15dff286 JA |
8112 | /* if we failed setting up the ctx, we might not have any rings */ |
8113 | if (ctx->rings) | |
e6c8aa9a | 8114 | io_cqring_overflow_flush(ctx, true, NULL, NULL); |
b2edc0a7 | 8115 | io_iopoll_try_reap_events(ctx); |
071698e1 | 8116 | idr_for_each(&ctx->personality_idr, io_remove_personalities, ctx); |
309fc03a JA |
8117 | |
8118 | /* | |
8119 | * Do this upfront, so we won't have a grace period where the ring | |
8120 | * is closed but resources aren't reaped yet. This can cause | |
8121 | * spurious failure in setting up a new ring. | |
8122 | */ | |
760618f7 JA |
8123 | io_unaccount_mem(ctx, ring_pages(ctx->sq_entries, ctx->cq_entries), |
8124 | ACCT_LOCKED); | |
309fc03a | 8125 | |
85faa7b8 | 8126 | INIT_WORK(&ctx->exit_work, io_ring_exit_work); |
fc666777 JA |
8127 | /* |
8128 | * Use system_unbound_wq to avoid spawning tons of event kworkers | |
8129 | * if we're exiting a ton of rings at the same time. It just adds | |
8130 | * noise and overhead, there's no discernable change in runtime | |
8131 | * over using system_wq. | |
8132 | */ | |
8133 | queue_work(system_unbound_wq, &ctx->exit_work); | |
2b188cc1 JA |
8134 | } |
8135 | ||
8136 | static int io_uring_release(struct inode *inode, struct file *file) | |
8137 | { | |
8138 | struct io_ring_ctx *ctx = file->private_data; | |
8139 | ||
8140 | file->private_data = NULL; | |
8141 | io_ring_ctx_wait_and_kill(ctx); | |
8142 | return 0; | |
8143 | } | |
8144 | ||
67c4d9e6 PB |
8145 | static bool io_wq_files_match(struct io_wq_work *work, void *data) |
8146 | { | |
8147 | struct files_struct *files = data; | |
8148 | ||
0f212204 | 8149 | return !files || work->files == files; |
67c4d9e6 PB |
8150 | } |
8151 | ||
f254ac04 JA |
8152 | /* |
8153 | * Returns true if 'preq' is the link parent of 'req' | |
8154 | */ | |
8155 | static bool io_match_link(struct io_kiocb *preq, struct io_kiocb *req) | |
8156 | { | |
8157 | struct io_kiocb *link; | |
8158 | ||
8159 | if (!(preq->flags & REQ_F_LINK_HEAD)) | |
8160 | return false; | |
8161 | ||
8162 | list_for_each_entry(link, &preq->link_list, link_list) { | |
8163 | if (link == req) | |
8164 | return true; | |
8165 | } | |
8166 | ||
8167 | return false; | |
8168 | } | |
8169 | ||
c127a2a1 PB |
8170 | static bool io_match_link_files(struct io_kiocb *req, |
8171 | struct files_struct *files) | |
8172 | { | |
8173 | struct io_kiocb *link; | |
8174 | ||
8175 | if (io_match_files(req, files)) | |
8176 | return true; | |
8177 | if (req->flags & REQ_F_LINK_HEAD) { | |
8178 | list_for_each_entry(link, &req->link_list, link_list) { | |
8179 | if (io_match_files(link, files)) | |
8180 | return true; | |
8181 | } | |
8182 | } | |
8183 | return false; | |
8184 | } | |
8185 | ||
f254ac04 JA |
8186 | /* |
8187 | * We're looking to cancel 'req' because it's holding on to our files, but | |
8188 | * 'req' could be a link to another request. See if it is, and cancel that | |
8189 | * parent request if so. | |
8190 | */ | |
8191 | static bool io_poll_remove_link(struct io_ring_ctx *ctx, struct io_kiocb *req) | |
8192 | { | |
8193 | struct hlist_node *tmp; | |
8194 | struct io_kiocb *preq; | |
8195 | bool found = false; | |
8196 | int i; | |
8197 | ||
8198 | spin_lock_irq(&ctx->completion_lock); | |
8199 | for (i = 0; i < (1U << ctx->cancel_hash_bits); i++) { | |
8200 | struct hlist_head *list; | |
8201 | ||
8202 | list = &ctx->cancel_hash[i]; | |
8203 | hlist_for_each_entry_safe(preq, tmp, list, hash_node) { | |
8204 | found = io_match_link(preq, req); | |
8205 | if (found) { | |
8206 | io_poll_remove_one(preq); | |
8207 | break; | |
8208 | } | |
8209 | } | |
8210 | } | |
8211 | spin_unlock_irq(&ctx->completion_lock); | |
8212 | return found; | |
8213 | } | |
8214 | ||
8215 | static bool io_timeout_remove_link(struct io_ring_ctx *ctx, | |
8216 | struct io_kiocb *req) | |
8217 | { | |
8218 | struct io_kiocb *preq; | |
8219 | bool found = false; | |
8220 | ||
8221 | spin_lock_irq(&ctx->completion_lock); | |
8222 | list_for_each_entry(preq, &ctx->timeout_list, timeout.list) { | |
8223 | found = io_match_link(preq, req); | |
8224 | if (found) { | |
8225 | __io_timeout_cancel(preq); | |
8226 | break; | |
8227 | } | |
8228 | } | |
8229 | spin_unlock_irq(&ctx->completion_lock); | |
8230 | return found; | |
8231 | } | |
8232 | ||
b711d4ea JA |
8233 | static bool io_cancel_link_cb(struct io_wq_work *work, void *data) |
8234 | { | |
8235 | return io_match_link(container_of(work, struct io_kiocb, work), data); | |
8236 | } | |
8237 | ||
8238 | static void io_attempt_cancel(struct io_ring_ctx *ctx, struct io_kiocb *req) | |
8239 | { | |
8240 | enum io_wq_cancel cret; | |
8241 | ||
8242 | /* cancel this particular work, if it's running */ | |
8243 | cret = io_wq_cancel_work(ctx->io_wq, &req->work); | |
8244 | if (cret != IO_WQ_CANCEL_NOTFOUND) | |
8245 | return; | |
8246 | ||
8247 | /* find links that hold this pending, cancel those */ | |
8248 | cret = io_wq_cancel_cb(ctx->io_wq, io_cancel_link_cb, req, true); | |
8249 | if (cret != IO_WQ_CANCEL_NOTFOUND) | |
8250 | return; | |
8251 | ||
8252 | /* if we have a poll link holding this pending, cancel that */ | |
8253 | if (io_poll_remove_link(ctx, req)) | |
8254 | return; | |
8255 | ||
8256 | /* final option, timeout link is holding this req pending */ | |
8257 | io_timeout_remove_link(ctx, req); | |
8258 | } | |
8259 | ||
b7ddce3c PB |
8260 | static void io_cancel_defer_files(struct io_ring_ctx *ctx, |
8261 | struct files_struct *files) | |
8262 | { | |
8263 | struct io_defer_entry *de = NULL; | |
8264 | LIST_HEAD(list); | |
8265 | ||
8266 | spin_lock_irq(&ctx->completion_lock); | |
8267 | list_for_each_entry_reverse(de, &ctx->defer_list, list) { | |
c127a2a1 | 8268 | if (io_match_link_files(de->req, files)) { |
b7ddce3c PB |
8269 | list_cut_position(&list, &ctx->defer_list, &de->list); |
8270 | break; | |
8271 | } | |
8272 | } | |
8273 | spin_unlock_irq(&ctx->completion_lock); | |
8274 | ||
8275 | while (!list_empty(&list)) { | |
8276 | de = list_first_entry(&list, struct io_defer_entry, list); | |
8277 | list_del_init(&de->list); | |
8278 | req_set_fail_links(de->req); | |
8279 | io_put_req(de->req); | |
8280 | io_req_complete(de->req, -ECANCELED); | |
8281 | kfree(de); | |
8282 | } | |
8283 | } | |
8284 | ||
76e1b642 JA |
8285 | /* |
8286 | * Returns true if we found and killed one or more files pinning requests | |
8287 | */ | |
8288 | static bool io_uring_cancel_files(struct io_ring_ctx *ctx, | |
fcb323cc JA |
8289 | struct files_struct *files) |
8290 | { | |
67c4d9e6 | 8291 | if (list_empty_careful(&ctx->inflight_list)) |
76e1b642 | 8292 | return false; |
67c4d9e6 | 8293 | |
b7ddce3c | 8294 | io_cancel_defer_files(ctx, files); |
67c4d9e6 PB |
8295 | /* cancel all at once, should be faster than doing it one by one*/ |
8296 | io_wq_cancel_cb(ctx->io_wq, io_wq_files_match, files, true); | |
8297 | ||
fcb323cc | 8298 | while (!list_empty_careful(&ctx->inflight_list)) { |
d8f1b971 XW |
8299 | struct io_kiocb *cancel_req = NULL, *req; |
8300 | DEFINE_WAIT(wait); | |
fcb323cc JA |
8301 | |
8302 | spin_lock_irq(&ctx->inflight_lock); | |
8303 | list_for_each_entry(req, &ctx->inflight_list, inflight_entry) { | |
0f212204 | 8304 | if (files && req->work.files != files) |
768134d4 JA |
8305 | continue; |
8306 | /* req is being completed, ignore */ | |
8307 | if (!refcount_inc_not_zero(&req->refs)) | |
8308 | continue; | |
8309 | cancel_req = req; | |
8310 | break; | |
fcb323cc | 8311 | } |
768134d4 | 8312 | if (cancel_req) |
fcb323cc | 8313 | prepare_to_wait(&ctx->inflight_wait, &wait, |
768134d4 | 8314 | TASK_UNINTERRUPTIBLE); |
fcb323cc JA |
8315 | spin_unlock_irq(&ctx->inflight_lock); |
8316 | ||
768134d4 JA |
8317 | /* We need to keep going until we don't find a matching req */ |
8318 | if (!cancel_req) | |
fcb323cc | 8319 | break; |
bb175342 PB |
8320 | /* cancel this request, or head link requests */ |
8321 | io_attempt_cancel(ctx, cancel_req); | |
8322 | io_put_req(cancel_req); | |
6200b0ae JA |
8323 | /* cancellations _may_ trigger task work */ |
8324 | io_run_task_work(); | |
fcb323cc | 8325 | schedule(); |
d8f1b971 | 8326 | finish_wait(&ctx->inflight_wait, &wait); |
fcb323cc | 8327 | } |
76e1b642 JA |
8328 | |
8329 | return true; | |
fcb323cc JA |
8330 | } |
8331 | ||
801dd57b | 8332 | static bool io_cancel_task_cb(struct io_wq_work *work, void *data) |
44e728b8 | 8333 | { |
801dd57b PB |
8334 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); |
8335 | struct task_struct *task = data; | |
44e728b8 | 8336 | |
f3606e3a | 8337 | return io_task_match(req, task); |
44e728b8 PB |
8338 | } |
8339 | ||
0f212204 JA |
8340 | static bool __io_uring_cancel_task_requests(struct io_ring_ctx *ctx, |
8341 | struct task_struct *task, | |
8342 | struct files_struct *files) | |
8343 | { | |
8344 | bool ret; | |
8345 | ||
8346 | ret = io_uring_cancel_files(ctx, files); | |
8347 | if (!files) { | |
8348 | enum io_wq_cancel cret; | |
8349 | ||
8350 | cret = io_wq_cancel_cb(ctx->io_wq, io_cancel_task_cb, task, true); | |
8351 | if (cret != IO_WQ_CANCEL_NOTFOUND) | |
8352 | ret = true; | |
8353 | ||
8354 | /* SQPOLL thread does its own polling */ | |
8355 | if (!(ctx->flags & IORING_SETUP_SQPOLL)) { | |
8356 | while (!list_empty_careful(&ctx->iopoll_list)) { | |
8357 | io_iopoll_try_reap_events(ctx); | |
8358 | ret = true; | |
8359 | } | |
8360 | } | |
8361 | ||
8362 | ret |= io_poll_remove_all(ctx, task); | |
8363 | ret |= io_kill_timeouts(ctx, task); | |
8364 | } | |
8365 | ||
8366 | return ret; | |
8367 | } | |
8368 | ||
8369 | /* | |
8370 | * We need to iteratively cancel requests, in case a request has dependent | |
8371 | * hard links. These persist even for failure of cancelations, hence keep | |
8372 | * looping until none are found. | |
8373 | */ | |
8374 | static void io_uring_cancel_task_requests(struct io_ring_ctx *ctx, | |
8375 | struct files_struct *files) | |
8376 | { | |
8377 | struct task_struct *task = current; | |
8378 | ||
8379 | if (ctx->flags & IORING_SETUP_SQPOLL) | |
8380 | task = ctx->sqo_thread; | |
8381 | ||
8382 | io_cqring_overflow_flush(ctx, true, task, files); | |
8383 | ||
8384 | while (__io_uring_cancel_task_requests(ctx, task, files)) { | |
8385 | io_run_task_work(); | |
8386 | cond_resched(); | |
8387 | } | |
8388 | } | |
8389 | ||
8390 | /* | |
8391 | * Note that this task has used io_uring. We use it for cancelation purposes. | |
8392 | */ | |
8393 | static int io_uring_add_task_file(struct file *file) | |
8394 | { | |
8395 | if (unlikely(!current->io_uring)) { | |
8396 | int ret; | |
8397 | ||
8398 | ret = io_uring_alloc_task_context(current); | |
8399 | if (unlikely(ret)) | |
8400 | return ret; | |
8401 | } | |
8402 | if (current->io_uring->last != file) { | |
8403 | XA_STATE(xas, ¤t->io_uring->xa, (unsigned long) file); | |
8404 | void *old; | |
8405 | ||
8406 | rcu_read_lock(); | |
8407 | old = xas_load(&xas); | |
8408 | if (old != file) { | |
8409 | get_file(file); | |
8410 | xas_lock(&xas); | |
8411 | xas_store(&xas, file); | |
8412 | xas_unlock(&xas); | |
8413 | } | |
8414 | rcu_read_unlock(); | |
8415 | current->io_uring->last = file; | |
8416 | } | |
8417 | ||
8418 | return 0; | |
8419 | } | |
8420 | ||
8421 | /* | |
8422 | * Remove this io_uring_file -> task mapping. | |
8423 | */ | |
8424 | static void io_uring_del_task_file(struct file *file) | |
8425 | { | |
8426 | struct io_uring_task *tctx = current->io_uring; | |
8427 | XA_STATE(xas, &tctx->xa, (unsigned long) file); | |
8428 | ||
8429 | if (tctx->last == file) | |
8430 | tctx->last = NULL; | |
8431 | ||
8432 | xas_lock(&xas); | |
8433 | file = xas_store(&xas, NULL); | |
8434 | xas_unlock(&xas); | |
8435 | ||
8436 | if (file) | |
8437 | fput(file); | |
8438 | } | |
8439 | ||
8440 | static void __io_uring_attempt_task_drop(struct file *file) | |
8441 | { | |
8442 | XA_STATE(xas, ¤t->io_uring->xa, (unsigned long) file); | |
8443 | struct file *old; | |
8444 | ||
8445 | rcu_read_lock(); | |
8446 | old = xas_load(&xas); | |
8447 | rcu_read_unlock(); | |
8448 | ||
8449 | if (old == file) | |
8450 | io_uring_del_task_file(file); | |
8451 | } | |
8452 | ||
8453 | /* | |
8454 | * Drop task note for this file if we're the only ones that hold it after | |
8455 | * pending fput() | |
8456 | */ | |
8457 | static void io_uring_attempt_task_drop(struct file *file, bool exiting) | |
8458 | { | |
8459 | if (!current->io_uring) | |
8460 | return; | |
8461 | /* | |
8462 | * fput() is pending, will be 2 if the only other ref is our potential | |
8463 | * task file note. If the task is exiting, drop regardless of count. | |
8464 | */ | |
8465 | if (!exiting && atomic_long_read(&file->f_count) != 2) | |
8466 | return; | |
8467 | ||
8468 | __io_uring_attempt_task_drop(file); | |
8469 | } | |
8470 | ||
8471 | void __io_uring_files_cancel(struct files_struct *files) | |
8472 | { | |
8473 | struct io_uring_task *tctx = current->io_uring; | |
8474 | XA_STATE(xas, &tctx->xa, 0); | |
8475 | ||
8476 | /* make sure overflow events are dropped */ | |
8477 | tctx->in_idle = true; | |
8478 | ||
8479 | do { | |
8480 | struct io_ring_ctx *ctx; | |
8481 | struct file *file; | |
8482 | ||
8483 | xas_lock(&xas); | |
8484 | file = xas_next_entry(&xas, ULONG_MAX); | |
8485 | xas_unlock(&xas); | |
8486 | ||
8487 | if (!file) | |
8488 | break; | |
8489 | ||
8490 | ctx = file->private_data; | |
8491 | ||
8492 | io_uring_cancel_task_requests(ctx, files); | |
8493 | if (files) | |
8494 | io_uring_del_task_file(file); | |
8495 | } while (1); | |
8496 | } | |
8497 | ||
8498 | static inline bool io_uring_task_idle(struct io_uring_task *tctx) | |
8499 | { | |
8500 | return atomic_long_read(&tctx->req_issue) == | |
8501 | atomic_long_read(&tctx->req_complete); | |
8502 | } | |
8503 | ||
8504 | /* | |
8505 | * Find any io_uring fd that this task has registered or done IO on, and cancel | |
8506 | * requests. | |
8507 | */ | |
8508 | void __io_uring_task_cancel(void) | |
8509 | { | |
8510 | struct io_uring_task *tctx = current->io_uring; | |
8511 | DEFINE_WAIT(wait); | |
8512 | long completions; | |
8513 | ||
8514 | /* make sure overflow events are dropped */ | |
8515 | tctx->in_idle = true; | |
8516 | ||
8517 | while (!io_uring_task_idle(tctx)) { | |
8518 | /* read completions before cancelations */ | |
8519 | completions = atomic_long_read(&tctx->req_complete); | |
8520 | __io_uring_files_cancel(NULL); | |
8521 | ||
8522 | prepare_to_wait(&tctx->wait, &wait, TASK_UNINTERRUPTIBLE); | |
8523 | ||
8524 | /* | |
8525 | * If we've seen completions, retry. This avoids a race where | |
8526 | * a completion comes in before we did prepare_to_wait(). | |
8527 | */ | |
8528 | if (completions != atomic_long_read(&tctx->req_complete)) | |
8529 | continue; | |
8530 | if (io_uring_task_idle(tctx)) | |
8531 | break; | |
8532 | schedule(); | |
8533 | } | |
8534 | ||
8535 | finish_wait(&tctx->wait, &wait); | |
8536 | tctx->in_idle = false; | |
8537 | } | |
8538 | ||
fcb323cc JA |
8539 | static int io_uring_flush(struct file *file, void *data) |
8540 | { | |
8541 | struct io_ring_ctx *ctx = file->private_data; | |
8542 | ||
6ab23144 JA |
8543 | /* |
8544 | * If the task is going away, cancel work it may have pending | |
8545 | */ | |
801dd57b | 8546 | if (fatal_signal_pending(current) || (current->flags & PF_EXITING)) |
0f212204 | 8547 | data = NULL; |
6ab23144 | 8548 | |
0f212204 JA |
8549 | io_uring_cancel_task_requests(ctx, data); |
8550 | io_uring_attempt_task_drop(file, !data); | |
fcb323cc JA |
8551 | return 0; |
8552 | } | |
8553 | ||
6c5c240e RP |
8554 | static void *io_uring_validate_mmap_request(struct file *file, |
8555 | loff_t pgoff, size_t sz) | |
2b188cc1 | 8556 | { |
2b188cc1 | 8557 | struct io_ring_ctx *ctx = file->private_data; |
6c5c240e | 8558 | loff_t offset = pgoff << PAGE_SHIFT; |
2b188cc1 JA |
8559 | struct page *page; |
8560 | void *ptr; | |
8561 | ||
8562 | switch (offset) { | |
8563 | case IORING_OFF_SQ_RING: | |
75b28aff HV |
8564 | case IORING_OFF_CQ_RING: |
8565 | ptr = ctx->rings; | |
2b188cc1 JA |
8566 | break; |
8567 | case IORING_OFF_SQES: | |
8568 | ptr = ctx->sq_sqes; | |
8569 | break; | |
2b188cc1 | 8570 | default: |
6c5c240e | 8571 | return ERR_PTR(-EINVAL); |
2b188cc1 JA |
8572 | } |
8573 | ||
8574 | page = virt_to_head_page(ptr); | |
a50b854e | 8575 | if (sz > page_size(page)) |
6c5c240e RP |
8576 | return ERR_PTR(-EINVAL); |
8577 | ||
8578 | return ptr; | |
8579 | } | |
8580 | ||
8581 | #ifdef CONFIG_MMU | |
8582 | ||
8583 | static int io_uring_mmap(struct file *file, struct vm_area_struct *vma) | |
8584 | { | |
8585 | size_t sz = vma->vm_end - vma->vm_start; | |
8586 | unsigned long pfn; | |
8587 | void *ptr; | |
8588 | ||
8589 | ptr = io_uring_validate_mmap_request(file, vma->vm_pgoff, sz); | |
8590 | if (IS_ERR(ptr)) | |
8591 | return PTR_ERR(ptr); | |
2b188cc1 JA |
8592 | |
8593 | pfn = virt_to_phys(ptr) >> PAGE_SHIFT; | |
8594 | return remap_pfn_range(vma, vma->vm_start, pfn, sz, vma->vm_page_prot); | |
8595 | } | |
8596 | ||
6c5c240e RP |
8597 | #else /* !CONFIG_MMU */ |
8598 | ||
8599 | static int io_uring_mmap(struct file *file, struct vm_area_struct *vma) | |
8600 | { | |
8601 | return vma->vm_flags & (VM_SHARED | VM_MAYSHARE) ? 0 : -EINVAL; | |
8602 | } | |
8603 | ||
8604 | static unsigned int io_uring_nommu_mmap_capabilities(struct file *file) | |
8605 | { | |
8606 | return NOMMU_MAP_DIRECT | NOMMU_MAP_READ | NOMMU_MAP_WRITE; | |
8607 | } | |
8608 | ||
8609 | static unsigned long io_uring_nommu_get_unmapped_area(struct file *file, | |
8610 | unsigned long addr, unsigned long len, | |
8611 | unsigned long pgoff, unsigned long flags) | |
8612 | { | |
8613 | void *ptr; | |
8614 | ||
8615 | ptr = io_uring_validate_mmap_request(file, pgoff, len); | |
8616 | if (IS_ERR(ptr)) | |
8617 | return PTR_ERR(ptr); | |
8618 | ||
8619 | return (unsigned long) ptr; | |
8620 | } | |
8621 | ||
8622 | #endif /* !CONFIG_MMU */ | |
8623 | ||
2b188cc1 JA |
8624 | SYSCALL_DEFINE6(io_uring_enter, unsigned int, fd, u32, to_submit, |
8625 | u32, min_complete, u32, flags, const sigset_t __user *, sig, | |
8626 | size_t, sigsz) | |
8627 | { | |
8628 | struct io_ring_ctx *ctx; | |
8629 | long ret = -EBADF; | |
8630 | int submitted = 0; | |
8631 | struct fd f; | |
8632 | ||
4c6e277c | 8633 | io_run_task_work(); |
b41e9852 | 8634 | |
6c271ce2 | 8635 | if (flags & ~(IORING_ENTER_GETEVENTS | IORING_ENTER_SQ_WAKEUP)) |
2b188cc1 JA |
8636 | return -EINVAL; |
8637 | ||
8638 | f = fdget(fd); | |
8639 | if (!f.file) | |
8640 | return -EBADF; | |
8641 | ||
8642 | ret = -EOPNOTSUPP; | |
8643 | if (f.file->f_op != &io_uring_fops) | |
8644 | goto out_fput; | |
8645 | ||
8646 | ret = -ENXIO; | |
8647 | ctx = f.file->private_data; | |
8648 | if (!percpu_ref_tryget(&ctx->refs)) | |
8649 | goto out_fput; | |
8650 | ||
7e84e1c7 SG |
8651 | ret = -EBADFD; |
8652 | if (ctx->flags & IORING_SETUP_R_DISABLED) | |
8653 | goto out; | |
8654 | ||
6c271ce2 JA |
8655 | /* |
8656 | * For SQ polling, the thread will do all submissions and completions. | |
8657 | * Just return the requested submit count, and wake the thread if | |
8658 | * we were asked to. | |
8659 | */ | |
b2a9eada | 8660 | ret = 0; |
6c271ce2 | 8661 | if (ctx->flags & IORING_SETUP_SQPOLL) { |
c1edbf5f | 8662 | if (!list_empty_careful(&ctx->cq_overflow_list)) |
e6c8aa9a | 8663 | io_cqring_overflow_flush(ctx, false, NULL, NULL); |
6c271ce2 | 8664 | if (flags & IORING_ENTER_SQ_WAKEUP) |
6a779382 | 8665 | wake_up(ctx->sqo_wait); |
6c271ce2 | 8666 | submitted = to_submit; |
b2a9eada | 8667 | } else if (to_submit) { |
0f212204 JA |
8668 | ret = io_uring_add_task_file(f.file); |
8669 | if (unlikely(ret)) | |
8670 | goto out; | |
2b188cc1 | 8671 | mutex_lock(&ctx->uring_lock); |
0f212204 | 8672 | submitted = io_submit_sqes(ctx, to_submit); |
2b188cc1 | 8673 | mutex_unlock(&ctx->uring_lock); |
7c504e65 PB |
8674 | |
8675 | if (submitted != to_submit) | |
8676 | goto out; | |
2b188cc1 JA |
8677 | } |
8678 | if (flags & IORING_ENTER_GETEVENTS) { | |
8679 | min_complete = min(min_complete, ctx->cq_entries); | |
8680 | ||
32b2244a XW |
8681 | /* |
8682 | * When SETUP_IOPOLL and SETUP_SQPOLL are both enabled, user | |
8683 | * space applications don't need to do io completion events | |
8684 | * polling again, they can rely on io_sq_thread to do polling | |
8685 | * work, which can reduce cpu usage and uring_lock contention. | |
8686 | */ | |
8687 | if (ctx->flags & IORING_SETUP_IOPOLL && | |
8688 | !(ctx->flags & IORING_SETUP_SQPOLL)) { | |
7668b92a | 8689 | ret = io_iopoll_check(ctx, min_complete); |
def596e9 JA |
8690 | } else { |
8691 | ret = io_cqring_wait(ctx, min_complete, sig, sigsz); | |
8692 | } | |
2b188cc1 JA |
8693 | } |
8694 | ||
7c504e65 | 8695 | out: |
6805b32e | 8696 | percpu_ref_put(&ctx->refs); |
2b188cc1 JA |
8697 | out_fput: |
8698 | fdput(f); | |
8699 | return submitted ? submitted : ret; | |
8700 | } | |
8701 | ||
bebdb65e | 8702 | #ifdef CONFIG_PROC_FS |
87ce955b JA |
8703 | static int io_uring_show_cred(int id, void *p, void *data) |
8704 | { | |
8705 | const struct cred *cred = p; | |
8706 | struct seq_file *m = data; | |
8707 | struct user_namespace *uns = seq_user_ns(m); | |
8708 | struct group_info *gi; | |
8709 | kernel_cap_t cap; | |
8710 | unsigned __capi; | |
8711 | int g; | |
8712 | ||
8713 | seq_printf(m, "%5d\n", id); | |
8714 | seq_put_decimal_ull(m, "\tUid:\t", from_kuid_munged(uns, cred->uid)); | |
8715 | seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->euid)); | |
8716 | seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->suid)); | |
8717 | seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->fsuid)); | |
8718 | seq_put_decimal_ull(m, "\n\tGid:\t", from_kgid_munged(uns, cred->gid)); | |
8719 | seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->egid)); | |
8720 | seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->sgid)); | |
8721 | seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->fsgid)); | |
8722 | seq_puts(m, "\n\tGroups:\t"); | |
8723 | gi = cred->group_info; | |
8724 | for (g = 0; g < gi->ngroups; g++) { | |
8725 | seq_put_decimal_ull(m, g ? " " : "", | |
8726 | from_kgid_munged(uns, gi->gid[g])); | |
8727 | } | |
8728 | seq_puts(m, "\n\tCapEff:\t"); | |
8729 | cap = cred->cap_effective; | |
8730 | CAP_FOR_EACH_U32(__capi) | |
8731 | seq_put_hex_ll(m, NULL, cap.cap[CAP_LAST_U32 - __capi], 8); | |
8732 | seq_putc(m, '\n'); | |
8733 | return 0; | |
8734 | } | |
8735 | ||
8736 | static void __io_uring_show_fdinfo(struct io_ring_ctx *ctx, struct seq_file *m) | |
8737 | { | |
fad8e0de | 8738 | bool has_lock; |
87ce955b JA |
8739 | int i; |
8740 | ||
fad8e0de JA |
8741 | /* |
8742 | * Avoid ABBA deadlock between the seq lock and the io_uring mutex, | |
8743 | * since fdinfo case grabs it in the opposite direction of normal use | |
8744 | * cases. If we fail to get the lock, we just don't iterate any | |
8745 | * structures that could be going away outside the io_uring mutex. | |
8746 | */ | |
8747 | has_lock = mutex_trylock(&ctx->uring_lock); | |
8748 | ||
87ce955b | 8749 | seq_printf(m, "UserFiles:\t%u\n", ctx->nr_user_files); |
fad8e0de | 8750 | for (i = 0; has_lock && i < ctx->nr_user_files; i++) { |
87ce955b JA |
8751 | struct fixed_file_table *table; |
8752 | struct file *f; | |
8753 | ||
8754 | table = &ctx->file_data->table[i >> IORING_FILE_TABLE_SHIFT]; | |
8755 | f = table->files[i & IORING_FILE_TABLE_MASK]; | |
8756 | if (f) | |
8757 | seq_printf(m, "%5u: %s\n", i, file_dentry(f)->d_iname); | |
8758 | else | |
8759 | seq_printf(m, "%5u: <none>\n", i); | |
8760 | } | |
8761 | seq_printf(m, "UserBufs:\t%u\n", ctx->nr_user_bufs); | |
fad8e0de | 8762 | for (i = 0; has_lock && i < ctx->nr_user_bufs; i++) { |
87ce955b JA |
8763 | struct io_mapped_ubuf *buf = &ctx->user_bufs[i]; |
8764 | ||
8765 | seq_printf(m, "%5u: 0x%llx/%u\n", i, buf->ubuf, | |
8766 | (unsigned int) buf->len); | |
8767 | } | |
fad8e0de | 8768 | if (has_lock && !idr_is_empty(&ctx->personality_idr)) { |
87ce955b JA |
8769 | seq_printf(m, "Personalities:\n"); |
8770 | idr_for_each(&ctx->personality_idr, io_uring_show_cred, m); | |
8771 | } | |
d7718a9d JA |
8772 | seq_printf(m, "PollList:\n"); |
8773 | spin_lock_irq(&ctx->completion_lock); | |
8774 | for (i = 0; i < (1U << ctx->cancel_hash_bits); i++) { | |
8775 | struct hlist_head *list = &ctx->cancel_hash[i]; | |
8776 | struct io_kiocb *req; | |
8777 | ||
8778 | hlist_for_each_entry(req, list, hash_node) | |
8779 | seq_printf(m, " op=%d, task_works=%d\n", req->opcode, | |
8780 | req->task->task_works != NULL); | |
8781 | } | |
8782 | spin_unlock_irq(&ctx->completion_lock); | |
fad8e0de JA |
8783 | if (has_lock) |
8784 | mutex_unlock(&ctx->uring_lock); | |
87ce955b JA |
8785 | } |
8786 | ||
8787 | static void io_uring_show_fdinfo(struct seq_file *m, struct file *f) | |
8788 | { | |
8789 | struct io_ring_ctx *ctx = f->private_data; | |
8790 | ||
8791 | if (percpu_ref_tryget(&ctx->refs)) { | |
8792 | __io_uring_show_fdinfo(ctx, m); | |
8793 | percpu_ref_put(&ctx->refs); | |
8794 | } | |
8795 | } | |
bebdb65e | 8796 | #endif |
87ce955b | 8797 | |
2b188cc1 JA |
8798 | static const struct file_operations io_uring_fops = { |
8799 | .release = io_uring_release, | |
fcb323cc | 8800 | .flush = io_uring_flush, |
2b188cc1 | 8801 | .mmap = io_uring_mmap, |
6c5c240e RP |
8802 | #ifndef CONFIG_MMU |
8803 | .get_unmapped_area = io_uring_nommu_get_unmapped_area, | |
8804 | .mmap_capabilities = io_uring_nommu_mmap_capabilities, | |
8805 | #endif | |
2b188cc1 JA |
8806 | .poll = io_uring_poll, |
8807 | .fasync = io_uring_fasync, | |
bebdb65e | 8808 | #ifdef CONFIG_PROC_FS |
87ce955b | 8809 | .show_fdinfo = io_uring_show_fdinfo, |
bebdb65e | 8810 | #endif |
2b188cc1 JA |
8811 | }; |
8812 | ||
8813 | static int io_allocate_scq_urings(struct io_ring_ctx *ctx, | |
8814 | struct io_uring_params *p) | |
8815 | { | |
75b28aff HV |
8816 | struct io_rings *rings; |
8817 | size_t size, sq_array_offset; | |
2b188cc1 | 8818 | |
bd740481 JA |
8819 | /* make sure these are sane, as we already accounted them */ |
8820 | ctx->sq_entries = p->sq_entries; | |
8821 | ctx->cq_entries = p->cq_entries; | |
8822 | ||
75b28aff HV |
8823 | size = rings_size(p->sq_entries, p->cq_entries, &sq_array_offset); |
8824 | if (size == SIZE_MAX) | |
8825 | return -EOVERFLOW; | |
8826 | ||
8827 | rings = io_mem_alloc(size); | |
8828 | if (!rings) | |
2b188cc1 JA |
8829 | return -ENOMEM; |
8830 | ||
75b28aff HV |
8831 | ctx->rings = rings; |
8832 | ctx->sq_array = (u32 *)((char *)rings + sq_array_offset); | |
8833 | rings->sq_ring_mask = p->sq_entries - 1; | |
8834 | rings->cq_ring_mask = p->cq_entries - 1; | |
8835 | rings->sq_ring_entries = p->sq_entries; | |
8836 | rings->cq_ring_entries = p->cq_entries; | |
8837 | ctx->sq_mask = rings->sq_ring_mask; | |
8838 | ctx->cq_mask = rings->cq_ring_mask; | |
2b188cc1 JA |
8839 | |
8840 | size = array_size(sizeof(struct io_uring_sqe), p->sq_entries); | |
eb065d30 JA |
8841 | if (size == SIZE_MAX) { |
8842 | io_mem_free(ctx->rings); | |
8843 | ctx->rings = NULL; | |
2b188cc1 | 8844 | return -EOVERFLOW; |
eb065d30 | 8845 | } |
2b188cc1 JA |
8846 | |
8847 | ctx->sq_sqes = io_mem_alloc(size); | |
eb065d30 JA |
8848 | if (!ctx->sq_sqes) { |
8849 | io_mem_free(ctx->rings); | |
8850 | ctx->rings = NULL; | |
2b188cc1 | 8851 | return -ENOMEM; |
eb065d30 | 8852 | } |
2b188cc1 | 8853 | |
2b188cc1 JA |
8854 | return 0; |
8855 | } | |
8856 | ||
8857 | /* | |
8858 | * Allocate an anonymous fd, this is what constitutes the application | |
8859 | * visible backing of an io_uring instance. The application mmaps this | |
8860 | * fd to gain access to the SQ/CQ ring details. If UNIX sockets are enabled, | |
8861 | * we have to tie this fd to a socket for file garbage collection purposes. | |
8862 | */ | |
8863 | static int io_uring_get_fd(struct io_ring_ctx *ctx) | |
8864 | { | |
8865 | struct file *file; | |
8866 | int ret; | |
8867 | ||
8868 | #if defined(CONFIG_UNIX) | |
8869 | ret = sock_create_kern(&init_net, PF_UNIX, SOCK_RAW, IPPROTO_IP, | |
8870 | &ctx->ring_sock); | |
8871 | if (ret) | |
8872 | return ret; | |
8873 | #endif | |
8874 | ||
8875 | ret = get_unused_fd_flags(O_RDWR | O_CLOEXEC); | |
8876 | if (ret < 0) | |
8877 | goto err; | |
8878 | ||
8879 | file = anon_inode_getfile("[io_uring]", &io_uring_fops, ctx, | |
8880 | O_RDWR | O_CLOEXEC); | |
8881 | if (IS_ERR(file)) { | |
0f212204 | 8882 | err_fd: |
2b188cc1 JA |
8883 | put_unused_fd(ret); |
8884 | ret = PTR_ERR(file); | |
8885 | goto err; | |
8886 | } | |
8887 | ||
8888 | #if defined(CONFIG_UNIX) | |
8889 | ctx->ring_sock->file = file; | |
8890 | #endif | |
0f212204 JA |
8891 | if (unlikely(io_uring_add_task_file(file))) { |
8892 | file = ERR_PTR(-ENOMEM); | |
8893 | goto err_fd; | |
8894 | } | |
2b188cc1 JA |
8895 | fd_install(ret, file); |
8896 | return ret; | |
8897 | err: | |
8898 | #if defined(CONFIG_UNIX) | |
8899 | sock_release(ctx->ring_sock); | |
8900 | ctx->ring_sock = NULL; | |
8901 | #endif | |
8902 | return ret; | |
8903 | } | |
8904 | ||
7f13657d XW |
8905 | static int io_uring_create(unsigned entries, struct io_uring_params *p, |
8906 | struct io_uring_params __user *params) | |
2b188cc1 JA |
8907 | { |
8908 | struct user_struct *user = NULL; | |
8909 | struct io_ring_ctx *ctx; | |
aad5d8da | 8910 | bool limit_mem; |
2b188cc1 JA |
8911 | int ret; |
8912 | ||
8110c1a6 | 8913 | if (!entries) |
2b188cc1 | 8914 | return -EINVAL; |
8110c1a6 JA |
8915 | if (entries > IORING_MAX_ENTRIES) { |
8916 | if (!(p->flags & IORING_SETUP_CLAMP)) | |
8917 | return -EINVAL; | |
8918 | entries = IORING_MAX_ENTRIES; | |
8919 | } | |
2b188cc1 JA |
8920 | |
8921 | /* | |
8922 | * Use twice as many entries for the CQ ring. It's possible for the | |
8923 | * application to drive a higher depth than the size of the SQ ring, | |
8924 | * since the sqes are only used at submission time. This allows for | |
33a107f0 JA |
8925 | * some flexibility in overcommitting a bit. If the application has |
8926 | * set IORING_SETUP_CQSIZE, it will have passed in the desired number | |
8927 | * of CQ ring entries manually. | |
2b188cc1 JA |
8928 | */ |
8929 | p->sq_entries = roundup_pow_of_two(entries); | |
33a107f0 JA |
8930 | if (p->flags & IORING_SETUP_CQSIZE) { |
8931 | /* | |
8932 | * If IORING_SETUP_CQSIZE is set, we do the same roundup | |
8933 | * to a power-of-two, if it isn't already. We do NOT impose | |
8934 | * any cq vs sq ring sizing. | |
8935 | */ | |
8110c1a6 | 8936 | if (p->cq_entries < p->sq_entries) |
33a107f0 | 8937 | return -EINVAL; |
8110c1a6 JA |
8938 | if (p->cq_entries > IORING_MAX_CQ_ENTRIES) { |
8939 | if (!(p->flags & IORING_SETUP_CLAMP)) | |
8940 | return -EINVAL; | |
8941 | p->cq_entries = IORING_MAX_CQ_ENTRIES; | |
8942 | } | |
33a107f0 JA |
8943 | p->cq_entries = roundup_pow_of_two(p->cq_entries); |
8944 | } else { | |
8945 | p->cq_entries = 2 * p->sq_entries; | |
8946 | } | |
2b188cc1 JA |
8947 | |
8948 | user = get_uid(current_user()); | |
aad5d8da | 8949 | limit_mem = !capable(CAP_IPC_LOCK); |
2b188cc1 | 8950 | |
aad5d8da | 8951 | if (limit_mem) { |
a087e2b5 | 8952 | ret = __io_account_mem(user, |
2b188cc1 JA |
8953 | ring_pages(p->sq_entries, p->cq_entries)); |
8954 | if (ret) { | |
8955 | free_uid(user); | |
8956 | return ret; | |
8957 | } | |
8958 | } | |
8959 | ||
8960 | ctx = io_ring_ctx_alloc(p); | |
8961 | if (!ctx) { | |
aad5d8da | 8962 | if (limit_mem) |
a087e2b5 | 8963 | __io_unaccount_mem(user, ring_pages(p->sq_entries, |
2b188cc1 JA |
8964 | p->cq_entries)); |
8965 | free_uid(user); | |
8966 | return -ENOMEM; | |
8967 | } | |
8968 | ctx->compat = in_compat_syscall(); | |
2b188cc1 | 8969 | ctx->user = user; |
0b8c0ec7 | 8970 | ctx->creds = get_current_cred(); |
2b188cc1 | 8971 | |
2aede0e4 JA |
8972 | ctx->sqo_task = get_task_struct(current); |
8973 | ||
8974 | /* | |
8975 | * This is just grabbed for accounting purposes. When a process exits, | |
8976 | * the mm is exited and dropped before the files, hence we need to hang | |
8977 | * on to this mm purely for the purposes of being able to unaccount | |
8978 | * memory (locked/pinned vm). It's not used for anything else. | |
8979 | */ | |
6b7898eb | 8980 | mmgrab(current->mm); |
2aede0e4 | 8981 | ctx->mm_account = current->mm; |
6b7898eb | 8982 | |
f74441e6 JA |
8983 | /* |
8984 | * Account memory _before_ installing the file descriptor. Once | |
8985 | * the descriptor is installed, it can get closed at any time. Also | |
8986 | * do this before hitting the general error path, as ring freeing | |
8987 | * will un-account as well. | |
8988 | */ | |
8989 | io_account_mem(ctx, ring_pages(p->sq_entries, p->cq_entries), | |
8990 | ACCT_LOCKED); | |
8991 | ctx->limit_mem = limit_mem; | |
8992 | ||
2b188cc1 JA |
8993 | ret = io_allocate_scq_urings(ctx, p); |
8994 | if (ret) | |
8995 | goto err; | |
8996 | ||
7e84e1c7 | 8997 | ret = io_sq_offload_create(ctx, p); |
2b188cc1 JA |
8998 | if (ret) |
8999 | goto err; | |
9000 | ||
7e84e1c7 SG |
9001 | if (!(p->flags & IORING_SETUP_R_DISABLED)) |
9002 | io_sq_offload_start(ctx); | |
9003 | ||
2b188cc1 | 9004 | memset(&p->sq_off, 0, sizeof(p->sq_off)); |
75b28aff HV |
9005 | p->sq_off.head = offsetof(struct io_rings, sq.head); |
9006 | p->sq_off.tail = offsetof(struct io_rings, sq.tail); | |
9007 | p->sq_off.ring_mask = offsetof(struct io_rings, sq_ring_mask); | |
9008 | p->sq_off.ring_entries = offsetof(struct io_rings, sq_ring_entries); | |
9009 | p->sq_off.flags = offsetof(struct io_rings, sq_flags); | |
9010 | p->sq_off.dropped = offsetof(struct io_rings, sq_dropped); | |
9011 | p->sq_off.array = (char *)ctx->sq_array - (char *)ctx->rings; | |
2b188cc1 JA |
9012 | |
9013 | memset(&p->cq_off, 0, sizeof(p->cq_off)); | |
75b28aff HV |
9014 | p->cq_off.head = offsetof(struct io_rings, cq.head); |
9015 | p->cq_off.tail = offsetof(struct io_rings, cq.tail); | |
9016 | p->cq_off.ring_mask = offsetof(struct io_rings, cq_ring_mask); | |
9017 | p->cq_off.ring_entries = offsetof(struct io_rings, cq_ring_entries); | |
9018 | p->cq_off.overflow = offsetof(struct io_rings, cq_overflow); | |
9019 | p->cq_off.cqes = offsetof(struct io_rings, cqes); | |
0d9b5b3a | 9020 | p->cq_off.flags = offsetof(struct io_rings, cq_flags); |
ac90f249 | 9021 | |
7f13657d XW |
9022 | p->features = IORING_FEAT_SINGLE_MMAP | IORING_FEAT_NODROP | |
9023 | IORING_FEAT_SUBMIT_STABLE | IORING_FEAT_RW_CUR_POS | | |
5769a351 JX |
9024 | IORING_FEAT_CUR_PERSONALITY | IORING_FEAT_FAST_POLL | |
9025 | IORING_FEAT_POLL_32BITS; | |
7f13657d XW |
9026 | |
9027 | if (copy_to_user(params, p, sizeof(*p))) { | |
9028 | ret = -EFAULT; | |
9029 | goto err; | |
9030 | } | |
d1719f70 | 9031 | |
044c1ab3 JA |
9032 | /* |
9033 | * Install ring fd as the very last thing, so we don't risk someone | |
9034 | * having closed it before we finish setup | |
9035 | */ | |
9036 | ret = io_uring_get_fd(ctx); | |
9037 | if (ret < 0) | |
9038 | goto err; | |
9039 | ||
c826bd7a | 9040 | trace_io_uring_create(ret, ctx, p->sq_entries, p->cq_entries, p->flags); |
2b188cc1 JA |
9041 | return ret; |
9042 | err: | |
9043 | io_ring_ctx_wait_and_kill(ctx); | |
9044 | return ret; | |
9045 | } | |
9046 | ||
9047 | /* | |
9048 | * Sets up an aio uring context, and returns the fd. Applications asks for a | |
9049 | * ring size, we return the actual sq/cq ring sizes (among other things) in the | |
9050 | * params structure passed in. | |
9051 | */ | |
9052 | static long io_uring_setup(u32 entries, struct io_uring_params __user *params) | |
9053 | { | |
9054 | struct io_uring_params p; | |
2b188cc1 JA |
9055 | int i; |
9056 | ||
9057 | if (copy_from_user(&p, params, sizeof(p))) | |
9058 | return -EFAULT; | |
9059 | for (i = 0; i < ARRAY_SIZE(p.resv); i++) { | |
9060 | if (p.resv[i]) | |
9061 | return -EINVAL; | |
9062 | } | |
9063 | ||
6c271ce2 | 9064 | if (p.flags & ~(IORING_SETUP_IOPOLL | IORING_SETUP_SQPOLL | |
8110c1a6 | 9065 | IORING_SETUP_SQ_AFF | IORING_SETUP_CQSIZE | |
7e84e1c7 SG |
9066 | IORING_SETUP_CLAMP | IORING_SETUP_ATTACH_WQ | |
9067 | IORING_SETUP_R_DISABLED)) | |
2b188cc1 JA |
9068 | return -EINVAL; |
9069 | ||
7f13657d | 9070 | return io_uring_create(entries, &p, params); |
2b188cc1 JA |
9071 | } |
9072 | ||
9073 | SYSCALL_DEFINE2(io_uring_setup, u32, entries, | |
9074 | struct io_uring_params __user *, params) | |
9075 | { | |
9076 | return io_uring_setup(entries, params); | |
9077 | } | |
9078 | ||
66f4af93 JA |
9079 | static int io_probe(struct io_ring_ctx *ctx, void __user *arg, unsigned nr_args) |
9080 | { | |
9081 | struct io_uring_probe *p; | |
9082 | size_t size; | |
9083 | int i, ret; | |
9084 | ||
9085 | size = struct_size(p, ops, nr_args); | |
9086 | if (size == SIZE_MAX) | |
9087 | return -EOVERFLOW; | |
9088 | p = kzalloc(size, GFP_KERNEL); | |
9089 | if (!p) | |
9090 | return -ENOMEM; | |
9091 | ||
9092 | ret = -EFAULT; | |
9093 | if (copy_from_user(p, arg, size)) | |
9094 | goto out; | |
9095 | ret = -EINVAL; | |
9096 | if (memchr_inv(p, 0, size)) | |
9097 | goto out; | |
9098 | ||
9099 | p->last_op = IORING_OP_LAST - 1; | |
9100 | if (nr_args > IORING_OP_LAST) | |
9101 | nr_args = IORING_OP_LAST; | |
9102 | ||
9103 | for (i = 0; i < nr_args; i++) { | |
9104 | p->ops[i].op = i; | |
9105 | if (!io_op_defs[i].not_supported) | |
9106 | p->ops[i].flags = IO_URING_OP_SUPPORTED; | |
9107 | } | |
9108 | p->ops_len = i; | |
9109 | ||
9110 | ret = 0; | |
9111 | if (copy_to_user(arg, p, size)) | |
9112 | ret = -EFAULT; | |
9113 | out: | |
9114 | kfree(p); | |
9115 | return ret; | |
9116 | } | |
9117 | ||
071698e1 JA |
9118 | static int io_register_personality(struct io_ring_ctx *ctx) |
9119 | { | |
9120 | const struct cred *creds = get_current_cred(); | |
9121 | int id; | |
9122 | ||
9123 | id = idr_alloc_cyclic(&ctx->personality_idr, (void *) creds, 1, | |
9124 | USHRT_MAX, GFP_KERNEL); | |
9125 | if (id < 0) | |
9126 | put_cred(creds); | |
9127 | return id; | |
9128 | } | |
9129 | ||
9130 | static int io_unregister_personality(struct io_ring_ctx *ctx, unsigned id) | |
9131 | { | |
9132 | const struct cred *old_creds; | |
9133 | ||
9134 | old_creds = idr_remove(&ctx->personality_idr, id); | |
9135 | if (old_creds) { | |
9136 | put_cred(old_creds); | |
9137 | return 0; | |
9138 | } | |
9139 | ||
9140 | return -EINVAL; | |
9141 | } | |
9142 | ||
21b55dbc SG |
9143 | static int io_register_restrictions(struct io_ring_ctx *ctx, void __user *arg, |
9144 | unsigned int nr_args) | |
9145 | { | |
9146 | struct io_uring_restriction *res; | |
9147 | size_t size; | |
9148 | int i, ret; | |
9149 | ||
7e84e1c7 SG |
9150 | /* Restrictions allowed only if rings started disabled */ |
9151 | if (!(ctx->flags & IORING_SETUP_R_DISABLED)) | |
9152 | return -EBADFD; | |
9153 | ||
21b55dbc | 9154 | /* We allow only a single restrictions registration */ |
7e84e1c7 | 9155 | if (ctx->restrictions.registered) |
21b55dbc SG |
9156 | return -EBUSY; |
9157 | ||
9158 | if (!arg || nr_args > IORING_MAX_RESTRICTIONS) | |
9159 | return -EINVAL; | |
9160 | ||
9161 | size = array_size(nr_args, sizeof(*res)); | |
9162 | if (size == SIZE_MAX) | |
9163 | return -EOVERFLOW; | |
9164 | ||
9165 | res = memdup_user(arg, size); | |
9166 | if (IS_ERR(res)) | |
9167 | return PTR_ERR(res); | |
9168 | ||
9169 | ret = 0; | |
9170 | ||
9171 | for (i = 0; i < nr_args; i++) { | |
9172 | switch (res[i].opcode) { | |
9173 | case IORING_RESTRICTION_REGISTER_OP: | |
9174 | if (res[i].register_op >= IORING_REGISTER_LAST) { | |
9175 | ret = -EINVAL; | |
9176 | goto out; | |
9177 | } | |
9178 | ||
9179 | __set_bit(res[i].register_op, | |
9180 | ctx->restrictions.register_op); | |
9181 | break; | |
9182 | case IORING_RESTRICTION_SQE_OP: | |
9183 | if (res[i].sqe_op >= IORING_OP_LAST) { | |
9184 | ret = -EINVAL; | |
9185 | goto out; | |
9186 | } | |
9187 | ||
9188 | __set_bit(res[i].sqe_op, ctx->restrictions.sqe_op); | |
9189 | break; | |
9190 | case IORING_RESTRICTION_SQE_FLAGS_ALLOWED: | |
9191 | ctx->restrictions.sqe_flags_allowed = res[i].sqe_flags; | |
9192 | break; | |
9193 | case IORING_RESTRICTION_SQE_FLAGS_REQUIRED: | |
9194 | ctx->restrictions.sqe_flags_required = res[i].sqe_flags; | |
9195 | break; | |
9196 | default: | |
9197 | ret = -EINVAL; | |
9198 | goto out; | |
9199 | } | |
9200 | } | |
9201 | ||
9202 | out: | |
9203 | /* Reset all restrictions if an error happened */ | |
9204 | if (ret != 0) | |
9205 | memset(&ctx->restrictions, 0, sizeof(ctx->restrictions)); | |
9206 | else | |
7e84e1c7 | 9207 | ctx->restrictions.registered = true; |
21b55dbc SG |
9208 | |
9209 | kfree(res); | |
9210 | return ret; | |
9211 | } | |
9212 | ||
7e84e1c7 SG |
9213 | static int io_register_enable_rings(struct io_ring_ctx *ctx) |
9214 | { | |
9215 | if (!(ctx->flags & IORING_SETUP_R_DISABLED)) | |
9216 | return -EBADFD; | |
9217 | ||
9218 | if (ctx->restrictions.registered) | |
9219 | ctx->restricted = 1; | |
9220 | ||
9221 | ctx->flags &= ~IORING_SETUP_R_DISABLED; | |
9222 | ||
9223 | io_sq_offload_start(ctx); | |
9224 | ||
9225 | return 0; | |
9226 | } | |
9227 | ||
071698e1 JA |
9228 | static bool io_register_op_must_quiesce(int op) |
9229 | { | |
9230 | switch (op) { | |
9231 | case IORING_UNREGISTER_FILES: | |
9232 | case IORING_REGISTER_FILES_UPDATE: | |
9233 | case IORING_REGISTER_PROBE: | |
9234 | case IORING_REGISTER_PERSONALITY: | |
9235 | case IORING_UNREGISTER_PERSONALITY: | |
9236 | return false; | |
9237 | default: | |
9238 | return true; | |
9239 | } | |
9240 | } | |
9241 | ||
edafccee JA |
9242 | static int __io_uring_register(struct io_ring_ctx *ctx, unsigned opcode, |
9243 | void __user *arg, unsigned nr_args) | |
b19062a5 JA |
9244 | __releases(ctx->uring_lock) |
9245 | __acquires(ctx->uring_lock) | |
edafccee JA |
9246 | { |
9247 | int ret; | |
9248 | ||
35fa71a0 JA |
9249 | /* |
9250 | * We're inside the ring mutex, if the ref is already dying, then | |
9251 | * someone else killed the ctx or is already going through | |
9252 | * io_uring_register(). | |
9253 | */ | |
9254 | if (percpu_ref_is_dying(&ctx->refs)) | |
9255 | return -ENXIO; | |
9256 | ||
071698e1 | 9257 | if (io_register_op_must_quiesce(opcode)) { |
05f3fb3c | 9258 | percpu_ref_kill(&ctx->refs); |
b19062a5 | 9259 | |
05f3fb3c JA |
9260 | /* |
9261 | * Drop uring mutex before waiting for references to exit. If | |
9262 | * another thread is currently inside io_uring_enter() it might | |
9263 | * need to grab the uring_lock to make progress. If we hold it | |
9264 | * here across the drain wait, then we can deadlock. It's safe | |
9265 | * to drop the mutex here, since no new references will come in | |
9266 | * after we've killed the percpu ref. | |
9267 | */ | |
9268 | mutex_unlock(&ctx->uring_lock); | |
0f158b4c | 9269 | ret = wait_for_completion_interruptible(&ctx->ref_comp); |
05f3fb3c | 9270 | mutex_lock(&ctx->uring_lock); |
c150368b JA |
9271 | if (ret) { |
9272 | percpu_ref_resurrect(&ctx->refs); | |
9273 | ret = -EINTR; | |
21b55dbc SG |
9274 | goto out_quiesce; |
9275 | } | |
9276 | } | |
9277 | ||
9278 | if (ctx->restricted) { | |
9279 | if (opcode >= IORING_REGISTER_LAST) { | |
9280 | ret = -EINVAL; | |
9281 | goto out; | |
9282 | } | |
9283 | ||
9284 | if (!test_bit(opcode, ctx->restrictions.register_op)) { | |
9285 | ret = -EACCES; | |
c150368b JA |
9286 | goto out; |
9287 | } | |
05f3fb3c | 9288 | } |
edafccee JA |
9289 | |
9290 | switch (opcode) { | |
9291 | case IORING_REGISTER_BUFFERS: | |
9292 | ret = io_sqe_buffer_register(ctx, arg, nr_args); | |
9293 | break; | |
9294 | case IORING_UNREGISTER_BUFFERS: | |
9295 | ret = -EINVAL; | |
9296 | if (arg || nr_args) | |
9297 | break; | |
9298 | ret = io_sqe_buffer_unregister(ctx); | |
9299 | break; | |
6b06314c JA |
9300 | case IORING_REGISTER_FILES: |
9301 | ret = io_sqe_files_register(ctx, arg, nr_args); | |
9302 | break; | |
9303 | case IORING_UNREGISTER_FILES: | |
9304 | ret = -EINVAL; | |
9305 | if (arg || nr_args) | |
9306 | break; | |
9307 | ret = io_sqe_files_unregister(ctx); | |
9308 | break; | |
c3a31e60 JA |
9309 | case IORING_REGISTER_FILES_UPDATE: |
9310 | ret = io_sqe_files_update(ctx, arg, nr_args); | |
9311 | break; | |
9b402849 | 9312 | case IORING_REGISTER_EVENTFD: |
f2842ab5 | 9313 | case IORING_REGISTER_EVENTFD_ASYNC: |
9b402849 JA |
9314 | ret = -EINVAL; |
9315 | if (nr_args != 1) | |
9316 | break; | |
9317 | ret = io_eventfd_register(ctx, arg); | |
f2842ab5 JA |
9318 | if (ret) |
9319 | break; | |
9320 | if (opcode == IORING_REGISTER_EVENTFD_ASYNC) | |
9321 | ctx->eventfd_async = 1; | |
9322 | else | |
9323 | ctx->eventfd_async = 0; | |
9b402849 JA |
9324 | break; |
9325 | case IORING_UNREGISTER_EVENTFD: | |
9326 | ret = -EINVAL; | |
9327 | if (arg || nr_args) | |
9328 | break; | |
9329 | ret = io_eventfd_unregister(ctx); | |
9330 | break; | |
66f4af93 JA |
9331 | case IORING_REGISTER_PROBE: |
9332 | ret = -EINVAL; | |
9333 | if (!arg || nr_args > 256) | |
9334 | break; | |
9335 | ret = io_probe(ctx, arg, nr_args); | |
9336 | break; | |
071698e1 JA |
9337 | case IORING_REGISTER_PERSONALITY: |
9338 | ret = -EINVAL; | |
9339 | if (arg || nr_args) | |
9340 | break; | |
9341 | ret = io_register_personality(ctx); | |
9342 | break; | |
9343 | case IORING_UNREGISTER_PERSONALITY: | |
9344 | ret = -EINVAL; | |
9345 | if (arg) | |
9346 | break; | |
9347 | ret = io_unregister_personality(ctx, nr_args); | |
9348 | break; | |
7e84e1c7 SG |
9349 | case IORING_REGISTER_ENABLE_RINGS: |
9350 | ret = -EINVAL; | |
9351 | if (arg || nr_args) | |
9352 | break; | |
9353 | ret = io_register_enable_rings(ctx); | |
9354 | break; | |
21b55dbc SG |
9355 | case IORING_REGISTER_RESTRICTIONS: |
9356 | ret = io_register_restrictions(ctx, arg, nr_args); | |
9357 | break; | |
edafccee JA |
9358 | default: |
9359 | ret = -EINVAL; | |
9360 | break; | |
9361 | } | |
9362 | ||
21b55dbc | 9363 | out: |
071698e1 | 9364 | if (io_register_op_must_quiesce(opcode)) { |
05f3fb3c | 9365 | /* bring the ctx back to life */ |
05f3fb3c | 9366 | percpu_ref_reinit(&ctx->refs); |
21b55dbc | 9367 | out_quiesce: |
0f158b4c | 9368 | reinit_completion(&ctx->ref_comp); |
05f3fb3c | 9369 | } |
edafccee JA |
9370 | return ret; |
9371 | } | |
9372 | ||
9373 | SYSCALL_DEFINE4(io_uring_register, unsigned int, fd, unsigned int, opcode, | |
9374 | void __user *, arg, unsigned int, nr_args) | |
9375 | { | |
9376 | struct io_ring_ctx *ctx; | |
9377 | long ret = -EBADF; | |
9378 | struct fd f; | |
9379 | ||
9380 | f = fdget(fd); | |
9381 | if (!f.file) | |
9382 | return -EBADF; | |
9383 | ||
9384 | ret = -EOPNOTSUPP; | |
9385 | if (f.file->f_op != &io_uring_fops) | |
9386 | goto out_fput; | |
9387 | ||
9388 | ctx = f.file->private_data; | |
9389 | ||
9390 | mutex_lock(&ctx->uring_lock); | |
9391 | ret = __io_uring_register(ctx, opcode, arg, nr_args); | |
9392 | mutex_unlock(&ctx->uring_lock); | |
c826bd7a DD |
9393 | trace_io_uring_register(ctx, opcode, ctx->nr_user_files, ctx->nr_user_bufs, |
9394 | ctx->cq_ev_fd != NULL, ret); | |
edafccee JA |
9395 | out_fput: |
9396 | fdput(f); | |
9397 | return ret; | |
9398 | } | |
9399 | ||
2b188cc1 JA |
9400 | static int __init io_uring_init(void) |
9401 | { | |
d7f62e82 SM |
9402 | #define __BUILD_BUG_VERIFY_ELEMENT(stype, eoffset, etype, ename) do { \ |
9403 | BUILD_BUG_ON(offsetof(stype, ename) != eoffset); \ | |
9404 | BUILD_BUG_ON(sizeof(etype) != sizeof_field(stype, ename)); \ | |
9405 | } while (0) | |
9406 | ||
9407 | #define BUILD_BUG_SQE_ELEM(eoffset, etype, ename) \ | |
9408 | __BUILD_BUG_VERIFY_ELEMENT(struct io_uring_sqe, eoffset, etype, ename) | |
9409 | BUILD_BUG_ON(sizeof(struct io_uring_sqe) != 64); | |
9410 | BUILD_BUG_SQE_ELEM(0, __u8, opcode); | |
9411 | BUILD_BUG_SQE_ELEM(1, __u8, flags); | |
9412 | BUILD_BUG_SQE_ELEM(2, __u16, ioprio); | |
9413 | BUILD_BUG_SQE_ELEM(4, __s32, fd); | |
9414 | BUILD_BUG_SQE_ELEM(8, __u64, off); | |
9415 | BUILD_BUG_SQE_ELEM(8, __u64, addr2); | |
9416 | BUILD_BUG_SQE_ELEM(16, __u64, addr); | |
7d67af2c | 9417 | BUILD_BUG_SQE_ELEM(16, __u64, splice_off_in); |
d7f62e82 SM |
9418 | BUILD_BUG_SQE_ELEM(24, __u32, len); |
9419 | BUILD_BUG_SQE_ELEM(28, __kernel_rwf_t, rw_flags); | |
9420 | BUILD_BUG_SQE_ELEM(28, /* compat */ int, rw_flags); | |
9421 | BUILD_BUG_SQE_ELEM(28, /* compat */ __u32, rw_flags); | |
9422 | BUILD_BUG_SQE_ELEM(28, __u32, fsync_flags); | |
5769a351 JX |
9423 | BUILD_BUG_SQE_ELEM(28, /* compat */ __u16, poll_events); |
9424 | BUILD_BUG_SQE_ELEM(28, __u32, poll32_events); | |
d7f62e82 SM |
9425 | BUILD_BUG_SQE_ELEM(28, __u32, sync_range_flags); |
9426 | BUILD_BUG_SQE_ELEM(28, __u32, msg_flags); | |
9427 | BUILD_BUG_SQE_ELEM(28, __u32, timeout_flags); | |
9428 | BUILD_BUG_SQE_ELEM(28, __u32, accept_flags); | |
9429 | BUILD_BUG_SQE_ELEM(28, __u32, cancel_flags); | |
9430 | BUILD_BUG_SQE_ELEM(28, __u32, open_flags); | |
9431 | BUILD_BUG_SQE_ELEM(28, __u32, statx_flags); | |
9432 | BUILD_BUG_SQE_ELEM(28, __u32, fadvise_advice); | |
7d67af2c | 9433 | BUILD_BUG_SQE_ELEM(28, __u32, splice_flags); |
d7f62e82 SM |
9434 | BUILD_BUG_SQE_ELEM(32, __u64, user_data); |
9435 | BUILD_BUG_SQE_ELEM(40, __u16, buf_index); | |
9436 | BUILD_BUG_SQE_ELEM(42, __u16, personality); | |
7d67af2c | 9437 | BUILD_BUG_SQE_ELEM(44, __s32, splice_fd_in); |
d7f62e82 | 9438 | |
d3656344 | 9439 | BUILD_BUG_ON(ARRAY_SIZE(io_op_defs) != IORING_OP_LAST); |
84557871 | 9440 | BUILD_BUG_ON(__REQ_F_LAST_BIT >= 8 * sizeof(int)); |
2b188cc1 JA |
9441 | req_cachep = KMEM_CACHE(io_kiocb, SLAB_HWCACHE_ALIGN | SLAB_PANIC); |
9442 | return 0; | |
9443 | }; | |
9444 | __initcall(io_uring_init); |