Commit | Line | Data |
---|---|---|
2b188cc1 JA |
1 | // SPDX-License-Identifier: GPL-2.0 |
2 | /* | |
3 | * Shared application/kernel submission and completion ring pairs, for | |
4 | * supporting fast/efficient IO. | |
5 | * | |
6 | * A note on the read/write ordering memory barriers that are matched between | |
1e84b97b SB |
7 | * the application and kernel side. |
8 | * | |
9 | * After the application reads the CQ ring tail, it must use an | |
10 | * appropriate smp_rmb() to pair with the smp_wmb() the kernel uses | |
11 | * before writing the tail (using smp_load_acquire to read the tail will | |
12 | * do). It also needs a smp_mb() before updating CQ head (ordering the | |
13 | * entry load(s) with the head store), pairing with an implicit barrier | |
14 | * through a control-dependency in io_get_cqring (smp_store_release to | |
15 | * store head will do). Failure to do so could lead to reading invalid | |
16 | * CQ entries. | |
17 | * | |
18 | * Likewise, the application must use an appropriate smp_wmb() before | |
19 | * writing the SQ tail (ordering SQ entry stores with the tail store), | |
20 | * which pairs with smp_load_acquire in io_get_sqring (smp_store_release | |
21 | * to store the tail will do). And it needs a barrier ordering the SQ | |
22 | * head load before writing new SQ entries (smp_load_acquire to read | |
23 | * head will do). | |
24 | * | |
25 | * When using the SQ poll thread (IORING_SETUP_SQPOLL), the application | |
26 | * needs to check the SQ flags for IORING_SQ_NEED_WAKEUP *after* | |
27 | * updating the SQ tail; a full memory barrier smp_mb() is needed | |
28 | * between. | |
2b188cc1 JA |
29 | * |
30 | * Also see the examples in the liburing library: | |
31 | * | |
32 | * git://git.kernel.dk/liburing | |
33 | * | |
34 | * io_uring also uses READ/WRITE_ONCE() for _any_ store or load that happens | |
35 | * from data shared between the kernel and application. This is done both | |
36 | * for ordering purposes, but also to ensure that once a value is loaded from | |
37 | * data that the application could potentially modify, it remains stable. | |
38 | * | |
39 | * Copyright (C) 2018-2019 Jens Axboe | |
c992fe29 | 40 | * Copyright (c) 2018-2019 Christoph Hellwig |
2b188cc1 JA |
41 | */ |
42 | #include <linux/kernel.h> | |
43 | #include <linux/init.h> | |
44 | #include <linux/errno.h> | |
45 | #include <linux/syscalls.h> | |
46 | #include <linux/compat.h> | |
52de1fe1 | 47 | #include <net/compat.h> |
2b188cc1 JA |
48 | #include <linux/refcount.h> |
49 | #include <linux/uio.h> | |
6b47ee6e | 50 | #include <linux/bits.h> |
2b188cc1 JA |
51 | |
52 | #include <linux/sched/signal.h> | |
53 | #include <linux/fs.h> | |
54 | #include <linux/file.h> | |
55 | #include <linux/fdtable.h> | |
56 | #include <linux/mm.h> | |
57 | #include <linux/mman.h> | |
2b188cc1 JA |
58 | #include <linux/percpu.h> |
59 | #include <linux/slab.h> | |
6c271ce2 | 60 | #include <linux/kthread.h> |
2b188cc1 | 61 | #include <linux/blkdev.h> |
edafccee | 62 | #include <linux/bvec.h> |
2b188cc1 JA |
63 | #include <linux/net.h> |
64 | #include <net/sock.h> | |
65 | #include <net/af_unix.h> | |
6b06314c | 66 | #include <net/scm.h> |
2b188cc1 JA |
67 | #include <linux/anon_inodes.h> |
68 | #include <linux/sched/mm.h> | |
69 | #include <linux/uaccess.h> | |
70 | #include <linux/nospec.h> | |
edafccee JA |
71 | #include <linux/sizes.h> |
72 | #include <linux/hugetlb.h> | |
aa4c3967 | 73 | #include <linux/highmem.h> |
15b71abe JA |
74 | #include <linux/namei.h> |
75 | #include <linux/fsnotify.h> | |
4840e418 | 76 | #include <linux/fadvise.h> |
3e4827b0 | 77 | #include <linux/eventpoll.h> |
ff002b30 | 78 | #include <linux/fs_struct.h> |
7d67af2c | 79 | #include <linux/splice.h> |
b41e9852 | 80 | #include <linux/task_work.h> |
bcf5a063 | 81 | #include <linux/pagemap.h> |
0f212204 | 82 | #include <linux/io_uring.h> |
91d8f519 | 83 | #include <linux/blk-cgroup.h> |
2b188cc1 | 84 | |
c826bd7a DD |
85 | #define CREATE_TRACE_POINTS |
86 | #include <trace/events/io_uring.h> | |
87 | ||
2b188cc1 JA |
88 | #include <uapi/linux/io_uring.h> |
89 | ||
90 | #include "internal.h" | |
561fb04a | 91 | #include "io-wq.h" |
2b188cc1 | 92 | |
5277deaa | 93 | #define IORING_MAX_ENTRIES 32768 |
33a107f0 | 94 | #define IORING_MAX_CQ_ENTRIES (2 * IORING_MAX_ENTRIES) |
65e19f54 JA |
95 | |
96 | /* | |
97 | * Shift of 9 is 512 entries, or exactly one page on 64-bit archs | |
98 | */ | |
99 | #define IORING_FILE_TABLE_SHIFT 9 | |
100 | #define IORING_MAX_FILES_TABLE (1U << IORING_FILE_TABLE_SHIFT) | |
101 | #define IORING_FILE_TABLE_MASK (IORING_MAX_FILES_TABLE - 1) | |
102 | #define IORING_MAX_FIXED_FILES (64 * IORING_MAX_FILES_TABLE) | |
21b55dbc SG |
103 | #define IORING_MAX_RESTRICTIONS (IORING_RESTRICTION_LAST + \ |
104 | IORING_REGISTER_LAST + IORING_OP_LAST) | |
2b188cc1 JA |
105 | |
106 | struct io_uring { | |
107 | u32 head ____cacheline_aligned_in_smp; | |
108 | u32 tail ____cacheline_aligned_in_smp; | |
109 | }; | |
110 | ||
1e84b97b | 111 | /* |
75b28aff HV |
112 | * This data is shared with the application through the mmap at offsets |
113 | * IORING_OFF_SQ_RING and IORING_OFF_CQ_RING. | |
1e84b97b SB |
114 | * |
115 | * The offsets to the member fields are published through struct | |
116 | * io_sqring_offsets when calling io_uring_setup. | |
117 | */ | |
75b28aff | 118 | struct io_rings { |
1e84b97b SB |
119 | /* |
120 | * Head and tail offsets into the ring; the offsets need to be | |
121 | * masked to get valid indices. | |
122 | * | |
75b28aff HV |
123 | * The kernel controls head of the sq ring and the tail of the cq ring, |
124 | * and the application controls tail of the sq ring and the head of the | |
125 | * cq ring. | |
1e84b97b | 126 | */ |
75b28aff | 127 | struct io_uring sq, cq; |
1e84b97b | 128 | /* |
75b28aff | 129 | * Bitmasks to apply to head and tail offsets (constant, equals |
1e84b97b SB |
130 | * ring_entries - 1) |
131 | */ | |
75b28aff HV |
132 | u32 sq_ring_mask, cq_ring_mask; |
133 | /* Ring sizes (constant, power of 2) */ | |
134 | u32 sq_ring_entries, cq_ring_entries; | |
1e84b97b SB |
135 | /* |
136 | * Number of invalid entries dropped by the kernel due to | |
137 | * invalid index stored in array | |
138 | * | |
139 | * Written by the kernel, shouldn't be modified by the | |
140 | * application (i.e. get number of "new events" by comparing to | |
141 | * cached value). | |
142 | * | |
143 | * After a new SQ head value was read by the application this | |
144 | * counter includes all submissions that were dropped reaching | |
145 | * the new SQ head (and possibly more). | |
146 | */ | |
75b28aff | 147 | u32 sq_dropped; |
1e84b97b | 148 | /* |
0d9b5b3a | 149 | * Runtime SQ flags |
1e84b97b SB |
150 | * |
151 | * Written by the kernel, shouldn't be modified by the | |
152 | * application. | |
153 | * | |
154 | * The application needs a full memory barrier before checking | |
155 | * for IORING_SQ_NEED_WAKEUP after updating the sq tail. | |
156 | */ | |
75b28aff | 157 | u32 sq_flags; |
0d9b5b3a SG |
158 | /* |
159 | * Runtime CQ flags | |
160 | * | |
161 | * Written by the application, shouldn't be modified by the | |
162 | * kernel. | |
163 | */ | |
164 | u32 cq_flags; | |
1e84b97b SB |
165 | /* |
166 | * Number of completion events lost because the queue was full; | |
167 | * this should be avoided by the application by making sure | |
0b4295b5 | 168 | * there are not more requests pending than there is space in |
1e84b97b SB |
169 | * the completion queue. |
170 | * | |
171 | * Written by the kernel, shouldn't be modified by the | |
172 | * application (i.e. get number of "new events" by comparing to | |
173 | * cached value). | |
174 | * | |
175 | * As completion events come in out of order this counter is not | |
176 | * ordered with any other data. | |
177 | */ | |
75b28aff | 178 | u32 cq_overflow; |
1e84b97b SB |
179 | /* |
180 | * Ring buffer of completion events. | |
181 | * | |
182 | * The kernel writes completion events fresh every time they are | |
183 | * produced, so the application is allowed to modify pending | |
184 | * entries. | |
185 | */ | |
75b28aff | 186 | struct io_uring_cqe cqes[] ____cacheline_aligned_in_smp; |
2b188cc1 JA |
187 | }; |
188 | ||
edafccee JA |
189 | struct io_mapped_ubuf { |
190 | u64 ubuf; | |
191 | size_t len; | |
192 | struct bio_vec *bvec; | |
193 | unsigned int nr_bvecs; | |
de293938 | 194 | unsigned long acct_pages; |
edafccee JA |
195 | }; |
196 | ||
65e19f54 JA |
197 | struct fixed_file_table { |
198 | struct file **files; | |
31b51510 JA |
199 | }; |
200 | ||
05589553 XW |
201 | struct fixed_file_ref_node { |
202 | struct percpu_ref refs; | |
203 | struct list_head node; | |
204 | struct list_head file_list; | |
205 | struct fixed_file_data *file_data; | |
4a38aed2 | 206 | struct llist_node llist; |
05589553 XW |
207 | }; |
208 | ||
05f3fb3c JA |
209 | struct fixed_file_data { |
210 | struct fixed_file_table *table; | |
211 | struct io_ring_ctx *ctx; | |
212 | ||
05589553 | 213 | struct percpu_ref *cur_refs; |
05f3fb3c | 214 | struct percpu_ref refs; |
05f3fb3c | 215 | struct completion done; |
05589553 XW |
216 | struct list_head ref_list; |
217 | spinlock_t lock; | |
05f3fb3c JA |
218 | }; |
219 | ||
5a2e745d JA |
220 | struct io_buffer { |
221 | struct list_head list; | |
222 | __u64 addr; | |
223 | __s32 len; | |
224 | __u16 bid; | |
225 | }; | |
226 | ||
21b55dbc SG |
227 | struct io_restriction { |
228 | DECLARE_BITMAP(register_op, IORING_REGISTER_LAST); | |
229 | DECLARE_BITMAP(sqe_op, IORING_OP_LAST); | |
230 | u8 sqe_flags_allowed; | |
231 | u8 sqe_flags_required; | |
7e84e1c7 | 232 | bool registered; |
21b55dbc SG |
233 | }; |
234 | ||
534ca6d6 JA |
235 | struct io_sq_data { |
236 | refcount_t refs; | |
69fb2131 JA |
237 | struct mutex lock; |
238 | ||
239 | /* ctx's that are using this sqd */ | |
240 | struct list_head ctx_list; | |
241 | struct list_head ctx_new_list; | |
242 | struct mutex ctx_lock; | |
243 | ||
534ca6d6 JA |
244 | struct task_struct *thread; |
245 | struct wait_queue_head wait; | |
246 | }; | |
247 | ||
2b188cc1 JA |
248 | struct io_ring_ctx { |
249 | struct { | |
250 | struct percpu_ref refs; | |
251 | } ____cacheline_aligned_in_smp; | |
252 | ||
253 | struct { | |
254 | unsigned int flags; | |
e1d85334 | 255 | unsigned int compat: 1; |
aad5d8da | 256 | unsigned int limit_mem: 1; |
e1d85334 RD |
257 | unsigned int cq_overflow_flushed: 1; |
258 | unsigned int drain_next: 1; | |
259 | unsigned int eventfd_async: 1; | |
21b55dbc | 260 | unsigned int restricted: 1; |
2b188cc1 | 261 | |
75b28aff HV |
262 | /* |
263 | * Ring buffer of indices into array of io_uring_sqe, which is | |
264 | * mmapped by the application using the IORING_OFF_SQES offset. | |
265 | * | |
266 | * This indirection could e.g. be used to assign fixed | |
267 | * io_uring_sqe entries to operations and only submit them to | |
268 | * the queue when needed. | |
269 | * | |
270 | * The kernel modifies neither the indices array nor the entries | |
271 | * array. | |
272 | */ | |
273 | u32 *sq_array; | |
2b188cc1 JA |
274 | unsigned cached_sq_head; |
275 | unsigned sq_entries; | |
276 | unsigned sq_mask; | |
6c271ce2 | 277 | unsigned sq_thread_idle; |
498ccd9e | 278 | unsigned cached_sq_dropped; |
206aefde | 279 | atomic_t cached_cq_overflow; |
ad3eb2c8 | 280 | unsigned long sq_check_overflow; |
de0617e4 JA |
281 | |
282 | struct list_head defer_list; | |
5262f567 | 283 | struct list_head timeout_list; |
1d7bb1d5 | 284 | struct list_head cq_overflow_list; |
fcb323cc JA |
285 | |
286 | wait_queue_head_t inflight_wait; | |
ad3eb2c8 | 287 | struct io_uring_sqe *sq_sqes; |
2b188cc1 JA |
288 | } ____cacheline_aligned_in_smp; |
289 | ||
206aefde JA |
290 | struct io_rings *rings; |
291 | ||
2b188cc1 | 292 | /* IO offload */ |
561fb04a | 293 | struct io_wq *io_wq; |
2aede0e4 JA |
294 | |
295 | /* | |
296 | * For SQPOLL usage - we hold a reference to the parent task, so we | |
297 | * have access to the ->files | |
298 | */ | |
299 | struct task_struct *sqo_task; | |
300 | ||
301 | /* Only used for accounting purposes */ | |
302 | struct mm_struct *mm_account; | |
303 | ||
91d8f519 DZ |
304 | #ifdef CONFIG_BLK_CGROUP |
305 | struct cgroup_subsys_state *sqo_blkcg_css; | |
306 | #endif | |
307 | ||
534ca6d6 JA |
308 | struct io_sq_data *sq_data; /* if using sq thread polling */ |
309 | ||
90554200 | 310 | struct wait_queue_head sqo_sq_wait; |
6a779382 | 311 | struct wait_queue_entry sqo_wait_entry; |
69fb2131 | 312 | struct list_head sqd_list; |
75b28aff | 313 | |
6b06314c JA |
314 | /* |
315 | * If used, fixed file set. Writers must ensure that ->refs is dead, | |
316 | * readers must ensure that ->refs is alive as long as the file* is | |
317 | * used. Only updated through io_uring_register(2). | |
318 | */ | |
05f3fb3c | 319 | struct fixed_file_data *file_data; |
6b06314c JA |
320 | unsigned nr_user_files; |
321 | ||
edafccee JA |
322 | /* if used, fixed mapped user buffers */ |
323 | unsigned nr_user_bufs; | |
324 | struct io_mapped_ubuf *user_bufs; | |
325 | ||
2b188cc1 JA |
326 | struct user_struct *user; |
327 | ||
0b8c0ec7 | 328 | const struct cred *creds; |
181e448d | 329 | |
0f158b4c JA |
330 | struct completion ref_comp; |
331 | struct completion sq_thread_comp; | |
206aefde | 332 | |
0ddf92e8 JA |
333 | /* if all else fails... */ |
334 | struct io_kiocb *fallback_req; | |
335 | ||
206aefde JA |
336 | #if defined(CONFIG_UNIX) |
337 | struct socket *ring_sock; | |
338 | #endif | |
339 | ||
5a2e745d JA |
340 | struct idr io_buffer_idr; |
341 | ||
071698e1 JA |
342 | struct idr personality_idr; |
343 | ||
206aefde JA |
344 | struct { |
345 | unsigned cached_cq_tail; | |
346 | unsigned cq_entries; | |
347 | unsigned cq_mask; | |
348 | atomic_t cq_timeouts; | |
ad3eb2c8 | 349 | unsigned long cq_check_overflow; |
206aefde JA |
350 | struct wait_queue_head cq_wait; |
351 | struct fasync_struct *cq_fasync; | |
352 | struct eventfd_ctx *cq_ev_fd; | |
353 | } ____cacheline_aligned_in_smp; | |
2b188cc1 JA |
354 | |
355 | struct { | |
356 | struct mutex uring_lock; | |
357 | wait_queue_head_t wait; | |
358 | } ____cacheline_aligned_in_smp; | |
359 | ||
360 | struct { | |
361 | spinlock_t completion_lock; | |
e94f141b | 362 | |
def596e9 | 363 | /* |
540e32a0 | 364 | * ->iopoll_list is protected by the ctx->uring_lock for |
def596e9 JA |
365 | * io_uring instances that don't use IORING_SETUP_SQPOLL. |
366 | * For SQPOLL, only the single threaded io_sq_thread() will | |
367 | * manipulate the list, hence no extra locking is needed there. | |
368 | */ | |
540e32a0 | 369 | struct list_head iopoll_list; |
78076bb6 JA |
370 | struct hlist_head *cancel_hash; |
371 | unsigned cancel_hash_bits; | |
e94f141b | 372 | bool poll_multi_file; |
31b51510 | 373 | |
fcb323cc JA |
374 | spinlock_t inflight_lock; |
375 | struct list_head inflight_list; | |
2b188cc1 | 376 | } ____cacheline_aligned_in_smp; |
85faa7b8 | 377 | |
4a38aed2 JA |
378 | struct delayed_work file_put_work; |
379 | struct llist_head file_put_llist; | |
380 | ||
85faa7b8 | 381 | struct work_struct exit_work; |
21b55dbc | 382 | struct io_restriction restrictions; |
2b188cc1 JA |
383 | }; |
384 | ||
09bb8394 JA |
385 | /* |
386 | * First field must be the file pointer in all the | |
387 | * iocb unions! See also 'struct kiocb' in <linux/fs.h> | |
388 | */ | |
221c5eb2 JA |
389 | struct io_poll_iocb { |
390 | struct file *file; | |
0969e783 JA |
391 | union { |
392 | struct wait_queue_head *head; | |
393 | u64 addr; | |
394 | }; | |
221c5eb2 | 395 | __poll_t events; |
8c838788 | 396 | bool done; |
221c5eb2 | 397 | bool canceled; |
392edb45 | 398 | struct wait_queue_entry wait; |
221c5eb2 JA |
399 | }; |
400 | ||
b5dba59e JA |
401 | struct io_close { |
402 | struct file *file; | |
403 | struct file *put_file; | |
404 | int fd; | |
405 | }; | |
406 | ||
ad8a48ac JA |
407 | struct io_timeout_data { |
408 | struct io_kiocb *req; | |
409 | struct hrtimer timer; | |
410 | struct timespec64 ts; | |
411 | enum hrtimer_mode mode; | |
412 | }; | |
413 | ||
8ed8d3c3 JA |
414 | struct io_accept { |
415 | struct file *file; | |
416 | struct sockaddr __user *addr; | |
417 | int __user *addr_len; | |
418 | int flags; | |
09952e3e | 419 | unsigned long nofile; |
8ed8d3c3 JA |
420 | }; |
421 | ||
422 | struct io_sync { | |
423 | struct file *file; | |
424 | loff_t len; | |
425 | loff_t off; | |
426 | int flags; | |
d63d1b5e | 427 | int mode; |
8ed8d3c3 JA |
428 | }; |
429 | ||
fbf23849 JA |
430 | struct io_cancel { |
431 | struct file *file; | |
432 | u64 addr; | |
433 | }; | |
434 | ||
b29472ee JA |
435 | struct io_timeout { |
436 | struct file *file; | |
437 | u64 addr; | |
438 | int flags; | |
bfe68a22 PB |
439 | u32 off; |
440 | u32 target_seq; | |
135fcde8 | 441 | struct list_head list; |
b29472ee JA |
442 | }; |
443 | ||
9adbd45d JA |
444 | struct io_rw { |
445 | /* NOTE: kiocb has the file as the first member, so don't do it here */ | |
446 | struct kiocb kiocb; | |
447 | u64 addr; | |
448 | u64 len; | |
449 | }; | |
450 | ||
3fbb51c1 JA |
451 | struct io_connect { |
452 | struct file *file; | |
453 | struct sockaddr __user *addr; | |
454 | int addr_len; | |
455 | }; | |
456 | ||
e47293fd JA |
457 | struct io_sr_msg { |
458 | struct file *file; | |
fddaface | 459 | union { |
270a5940 | 460 | struct user_msghdr __user *umsg; |
fddaface JA |
461 | void __user *buf; |
462 | }; | |
e47293fd | 463 | int msg_flags; |
bcda7baa | 464 | int bgid; |
fddaface | 465 | size_t len; |
bcda7baa | 466 | struct io_buffer *kbuf; |
e47293fd JA |
467 | }; |
468 | ||
15b71abe JA |
469 | struct io_open { |
470 | struct file *file; | |
471 | int dfd; | |
15b71abe | 472 | struct filename *filename; |
c12cedf2 | 473 | struct open_how how; |
4022e7af | 474 | unsigned long nofile; |
15b71abe JA |
475 | }; |
476 | ||
05f3fb3c JA |
477 | struct io_files_update { |
478 | struct file *file; | |
479 | u64 arg; | |
480 | u32 nr_args; | |
481 | u32 offset; | |
482 | }; | |
483 | ||
4840e418 JA |
484 | struct io_fadvise { |
485 | struct file *file; | |
486 | u64 offset; | |
487 | u32 len; | |
488 | u32 advice; | |
489 | }; | |
490 | ||
c1ca757b JA |
491 | struct io_madvise { |
492 | struct file *file; | |
493 | u64 addr; | |
494 | u32 len; | |
495 | u32 advice; | |
496 | }; | |
497 | ||
3e4827b0 JA |
498 | struct io_epoll { |
499 | struct file *file; | |
500 | int epfd; | |
501 | int op; | |
502 | int fd; | |
503 | struct epoll_event event; | |
e47293fd JA |
504 | }; |
505 | ||
7d67af2c PB |
506 | struct io_splice { |
507 | struct file *file_out; | |
508 | struct file *file_in; | |
509 | loff_t off_out; | |
510 | loff_t off_in; | |
511 | u64 len; | |
512 | unsigned int flags; | |
513 | }; | |
514 | ||
ddf0322d JA |
515 | struct io_provide_buf { |
516 | struct file *file; | |
517 | __u64 addr; | |
518 | __s32 len; | |
519 | __u32 bgid; | |
520 | __u16 nbufs; | |
521 | __u16 bid; | |
522 | }; | |
523 | ||
1d9e1288 BM |
524 | struct io_statx { |
525 | struct file *file; | |
526 | int dfd; | |
527 | unsigned int mask; | |
528 | unsigned int flags; | |
e62753e4 | 529 | const char __user *filename; |
1d9e1288 BM |
530 | struct statx __user *buffer; |
531 | }; | |
532 | ||
3ca405eb PB |
533 | struct io_completion { |
534 | struct file *file; | |
535 | struct list_head list; | |
0f7e466b | 536 | int cflags; |
3ca405eb PB |
537 | }; |
538 | ||
f499a021 JA |
539 | struct io_async_connect { |
540 | struct sockaddr_storage address; | |
541 | }; | |
542 | ||
03b1230c JA |
543 | struct io_async_msghdr { |
544 | struct iovec fast_iov[UIO_FASTIOV]; | |
545 | struct iovec *iov; | |
546 | struct sockaddr __user *uaddr; | |
547 | struct msghdr msg; | |
b537916c | 548 | struct sockaddr_storage addr; |
03b1230c JA |
549 | }; |
550 | ||
f67676d1 JA |
551 | struct io_async_rw { |
552 | struct iovec fast_iov[UIO_FASTIOV]; | |
ff6165b2 JA |
553 | const struct iovec *free_iovec; |
554 | struct iov_iter iter; | |
227c0c96 | 555 | size_t bytes_done; |
bcf5a063 | 556 | struct wait_page_queue wpq; |
f67676d1 JA |
557 | }; |
558 | ||
6b47ee6e PB |
559 | enum { |
560 | REQ_F_FIXED_FILE_BIT = IOSQE_FIXED_FILE_BIT, | |
561 | REQ_F_IO_DRAIN_BIT = IOSQE_IO_DRAIN_BIT, | |
562 | REQ_F_LINK_BIT = IOSQE_IO_LINK_BIT, | |
563 | REQ_F_HARDLINK_BIT = IOSQE_IO_HARDLINK_BIT, | |
564 | REQ_F_FORCE_ASYNC_BIT = IOSQE_ASYNC_BIT, | |
bcda7baa | 565 | REQ_F_BUFFER_SELECT_BIT = IOSQE_BUFFER_SELECT_BIT, |
6b47ee6e | 566 | |
dea3b49c | 567 | REQ_F_LINK_HEAD_BIT, |
6b47ee6e PB |
568 | REQ_F_FAIL_LINK_BIT, |
569 | REQ_F_INFLIGHT_BIT, | |
570 | REQ_F_CUR_POS_BIT, | |
571 | REQ_F_NOWAIT_BIT, | |
6b47ee6e | 572 | REQ_F_LINK_TIMEOUT_BIT, |
6b47ee6e | 573 | REQ_F_ISREG_BIT, |
6b47ee6e | 574 | REQ_F_COMP_LOCKED_BIT, |
99bc4c38 | 575 | REQ_F_NEED_CLEANUP_BIT, |
d7718a9d | 576 | REQ_F_POLLED_BIT, |
bcda7baa | 577 | REQ_F_BUFFER_SELECTED_BIT, |
5b0bbee4 | 578 | REQ_F_NO_FILE_TABLE_BIT, |
7cdaf587 | 579 | REQ_F_WORK_INITIALIZED_BIT, |
84557871 JA |
580 | |
581 | /* not a real bit, just to check we're not overflowing the space */ | |
582 | __REQ_F_LAST_BIT, | |
6b47ee6e PB |
583 | }; |
584 | ||
585 | enum { | |
586 | /* ctx owns file */ | |
587 | REQ_F_FIXED_FILE = BIT(REQ_F_FIXED_FILE_BIT), | |
588 | /* drain existing IO first */ | |
589 | REQ_F_IO_DRAIN = BIT(REQ_F_IO_DRAIN_BIT), | |
590 | /* linked sqes */ | |
591 | REQ_F_LINK = BIT(REQ_F_LINK_BIT), | |
592 | /* doesn't sever on completion < 0 */ | |
593 | REQ_F_HARDLINK = BIT(REQ_F_HARDLINK_BIT), | |
594 | /* IOSQE_ASYNC */ | |
595 | REQ_F_FORCE_ASYNC = BIT(REQ_F_FORCE_ASYNC_BIT), | |
bcda7baa JA |
596 | /* IOSQE_BUFFER_SELECT */ |
597 | REQ_F_BUFFER_SELECT = BIT(REQ_F_BUFFER_SELECT_BIT), | |
6b47ee6e | 598 | |
dea3b49c PB |
599 | /* head of a link */ |
600 | REQ_F_LINK_HEAD = BIT(REQ_F_LINK_HEAD_BIT), | |
6b47ee6e PB |
601 | /* fail rest of links */ |
602 | REQ_F_FAIL_LINK = BIT(REQ_F_FAIL_LINK_BIT), | |
603 | /* on inflight list */ | |
604 | REQ_F_INFLIGHT = BIT(REQ_F_INFLIGHT_BIT), | |
605 | /* read/write uses file position */ | |
606 | REQ_F_CUR_POS = BIT(REQ_F_CUR_POS_BIT), | |
607 | /* must not punt to workers */ | |
608 | REQ_F_NOWAIT = BIT(REQ_F_NOWAIT_BIT), | |
6b47ee6e PB |
609 | /* has linked timeout */ |
610 | REQ_F_LINK_TIMEOUT = BIT(REQ_F_LINK_TIMEOUT_BIT), | |
6b47ee6e PB |
611 | /* regular file */ |
612 | REQ_F_ISREG = BIT(REQ_F_ISREG_BIT), | |
6b47ee6e PB |
613 | /* completion under lock */ |
614 | REQ_F_COMP_LOCKED = BIT(REQ_F_COMP_LOCKED_BIT), | |
99bc4c38 PB |
615 | /* needs cleanup */ |
616 | REQ_F_NEED_CLEANUP = BIT(REQ_F_NEED_CLEANUP_BIT), | |
d7718a9d JA |
617 | /* already went through poll handler */ |
618 | REQ_F_POLLED = BIT(REQ_F_POLLED_BIT), | |
bcda7baa JA |
619 | /* buffer already selected */ |
620 | REQ_F_BUFFER_SELECTED = BIT(REQ_F_BUFFER_SELECTED_BIT), | |
5b0bbee4 JA |
621 | /* doesn't need file table for this request */ |
622 | REQ_F_NO_FILE_TABLE = BIT(REQ_F_NO_FILE_TABLE_BIT), | |
7cdaf587 XW |
623 | /* io_wq_work is initialized */ |
624 | REQ_F_WORK_INITIALIZED = BIT(REQ_F_WORK_INITIALIZED_BIT), | |
d7718a9d JA |
625 | }; |
626 | ||
627 | struct async_poll { | |
628 | struct io_poll_iocb poll; | |
807abcb0 | 629 | struct io_poll_iocb *double_poll; |
6b47ee6e PB |
630 | }; |
631 | ||
09bb8394 JA |
632 | /* |
633 | * NOTE! Each of the iocb union members has the file pointer | |
634 | * as the first entry in their struct definition. So you can | |
635 | * access the file pointer through any of the sub-structs, | |
636 | * or directly as just 'ki_filp' in this struct. | |
637 | */ | |
2b188cc1 | 638 | struct io_kiocb { |
221c5eb2 | 639 | union { |
09bb8394 | 640 | struct file *file; |
9adbd45d | 641 | struct io_rw rw; |
221c5eb2 | 642 | struct io_poll_iocb poll; |
8ed8d3c3 JA |
643 | struct io_accept accept; |
644 | struct io_sync sync; | |
fbf23849 | 645 | struct io_cancel cancel; |
b29472ee | 646 | struct io_timeout timeout; |
3fbb51c1 | 647 | struct io_connect connect; |
e47293fd | 648 | struct io_sr_msg sr_msg; |
15b71abe | 649 | struct io_open open; |
b5dba59e | 650 | struct io_close close; |
05f3fb3c | 651 | struct io_files_update files_update; |
4840e418 | 652 | struct io_fadvise fadvise; |
c1ca757b | 653 | struct io_madvise madvise; |
3e4827b0 | 654 | struct io_epoll epoll; |
7d67af2c | 655 | struct io_splice splice; |
ddf0322d | 656 | struct io_provide_buf pbuf; |
1d9e1288 | 657 | struct io_statx statx; |
3ca405eb PB |
658 | /* use only after cleaning per-op data, see io_clean_op() */ |
659 | struct io_completion compl; | |
221c5eb2 | 660 | }; |
2b188cc1 | 661 | |
e8c2bc1f JA |
662 | /* opcode allocated if it needs to store data for async defer */ |
663 | void *async_data; | |
d625c6ee | 664 | u8 opcode; |
65a6543d XW |
665 | /* polled IO has completed */ |
666 | u8 iopoll_completed; | |
2b188cc1 | 667 | |
4f4eeba8 | 668 | u16 buf_index; |
9cf7c104 | 669 | u32 result; |
4f4eeba8 | 670 | |
010e8e6b PB |
671 | struct io_ring_ctx *ctx; |
672 | unsigned int flags; | |
673 | refcount_t refs; | |
674 | struct task_struct *task; | |
675 | u64 user_data; | |
d7718a9d | 676 | |
010e8e6b | 677 | struct list_head link_list; |
fcb323cc | 678 | |
d21ffe7e PB |
679 | /* |
680 | * 1. used with ctx->iopoll_list with reads/writes | |
681 | * 2. to track reqs with ->files (see io_op_def::file_table) | |
682 | */ | |
010e8e6b PB |
683 | struct list_head inflight_entry; |
684 | ||
685 | struct percpu_ref *fixed_file_refs; | |
686 | struct callback_head task_work; | |
687 | /* for polled requests, i.e. IORING_OP_POLL_ADD and async armed poll */ | |
688 | struct hlist_node hash_node; | |
689 | struct async_poll *apoll; | |
690 | struct io_wq_work work; | |
2b188cc1 | 691 | }; |
05589553 | 692 | |
27dc8338 PB |
693 | struct io_defer_entry { |
694 | struct list_head list; | |
695 | struct io_kiocb *req; | |
9cf7c104 | 696 | u32 seq; |
2b188cc1 JA |
697 | }; |
698 | ||
def596e9 | 699 | #define IO_IOPOLL_BATCH 8 |
2b188cc1 | 700 | |
013538bd JA |
701 | struct io_comp_state { |
702 | unsigned int nr; | |
703 | struct list_head list; | |
704 | struct io_ring_ctx *ctx; | |
705 | }; | |
706 | ||
9a56a232 JA |
707 | struct io_submit_state { |
708 | struct blk_plug plug; | |
709 | ||
2579f913 JA |
710 | /* |
711 | * io_kiocb alloc cache | |
712 | */ | |
713 | void *reqs[IO_IOPOLL_BATCH]; | |
6c8a3134 | 714 | unsigned int free_reqs; |
2579f913 | 715 | |
013538bd JA |
716 | /* |
717 | * Batch completion logic | |
718 | */ | |
719 | struct io_comp_state comp; | |
720 | ||
9a56a232 JA |
721 | /* |
722 | * File reference cache | |
723 | */ | |
724 | struct file *file; | |
725 | unsigned int fd; | |
726 | unsigned int has_refs; | |
9a56a232 JA |
727 | unsigned int ios_left; |
728 | }; | |
729 | ||
d3656344 | 730 | struct io_op_def { |
d3656344 JA |
731 | /* needs current->mm setup, does mm access */ |
732 | unsigned needs_mm : 1; | |
733 | /* needs req->file assigned */ | |
734 | unsigned needs_file : 1; | |
fd2206e4 JA |
735 | /* don't fail if file grab fails */ |
736 | unsigned needs_file_no_error : 1; | |
d3656344 JA |
737 | /* hash wq insertion if file is a regular file */ |
738 | unsigned hash_reg_file : 1; | |
739 | /* unbound wq insertion if file is a non-regular file */ | |
740 | unsigned unbound_nonreg_file : 1; | |
66f4af93 JA |
741 | /* opcode is not supported by this kernel */ |
742 | unsigned not_supported : 1; | |
f86cd20c JA |
743 | /* needs file table */ |
744 | unsigned file_table : 1; | |
ff002b30 JA |
745 | /* needs ->fs */ |
746 | unsigned needs_fs : 1; | |
8a72758c JA |
747 | /* set if opcode supports polled "wait" */ |
748 | unsigned pollin : 1; | |
749 | unsigned pollout : 1; | |
bcda7baa JA |
750 | /* op supports buffer selection */ |
751 | unsigned buffer_select : 1; | |
e8c2bc1f | 752 | /* needs rlimit(RLIMIT_FSIZE) assigned */ |
57f1a649 | 753 | unsigned needs_fsize : 1; |
e8c2bc1f JA |
754 | /* must always have async data allocated */ |
755 | unsigned needs_async_data : 1; | |
91d8f519 DZ |
756 | /* needs blkcg context, issues async io potentially */ |
757 | unsigned needs_blkcg : 1; | |
e8c2bc1f JA |
758 | /* size of async data needed, if any */ |
759 | unsigned short async_size; | |
d3656344 JA |
760 | }; |
761 | ||
738277ad | 762 | static const struct io_op_def io_op_defs[] __read_mostly = { |
0463b6c5 PB |
763 | [IORING_OP_NOP] = {}, |
764 | [IORING_OP_READV] = { | |
d3656344 JA |
765 | .needs_mm = 1, |
766 | .needs_file = 1, | |
767 | .unbound_nonreg_file = 1, | |
8a72758c | 768 | .pollin = 1, |
4d954c25 | 769 | .buffer_select = 1, |
e8c2bc1f | 770 | .needs_async_data = 1, |
91d8f519 | 771 | .needs_blkcg = 1, |
e8c2bc1f | 772 | .async_size = sizeof(struct io_async_rw), |
d3656344 | 773 | }, |
0463b6c5 | 774 | [IORING_OP_WRITEV] = { |
d3656344 JA |
775 | .needs_mm = 1, |
776 | .needs_file = 1, | |
777 | .hash_reg_file = 1, | |
778 | .unbound_nonreg_file = 1, | |
8a72758c | 779 | .pollout = 1, |
57f1a649 | 780 | .needs_fsize = 1, |
e8c2bc1f | 781 | .needs_async_data = 1, |
91d8f519 | 782 | .needs_blkcg = 1, |
e8c2bc1f | 783 | .async_size = sizeof(struct io_async_rw), |
d3656344 | 784 | }, |
0463b6c5 | 785 | [IORING_OP_FSYNC] = { |
d3656344 | 786 | .needs_file = 1, |
91d8f519 | 787 | .needs_blkcg = 1, |
d3656344 | 788 | }, |
0463b6c5 | 789 | [IORING_OP_READ_FIXED] = { |
d3656344 JA |
790 | .needs_file = 1, |
791 | .unbound_nonreg_file = 1, | |
8a72758c | 792 | .pollin = 1, |
91d8f519 | 793 | .needs_blkcg = 1, |
e8c2bc1f | 794 | .async_size = sizeof(struct io_async_rw), |
d3656344 | 795 | }, |
0463b6c5 | 796 | [IORING_OP_WRITE_FIXED] = { |
d3656344 JA |
797 | .needs_file = 1, |
798 | .hash_reg_file = 1, | |
799 | .unbound_nonreg_file = 1, | |
8a72758c | 800 | .pollout = 1, |
57f1a649 | 801 | .needs_fsize = 1, |
91d8f519 | 802 | .needs_blkcg = 1, |
e8c2bc1f | 803 | .async_size = sizeof(struct io_async_rw), |
d3656344 | 804 | }, |
0463b6c5 | 805 | [IORING_OP_POLL_ADD] = { |
d3656344 JA |
806 | .needs_file = 1, |
807 | .unbound_nonreg_file = 1, | |
808 | }, | |
0463b6c5 PB |
809 | [IORING_OP_POLL_REMOVE] = {}, |
810 | [IORING_OP_SYNC_FILE_RANGE] = { | |
d3656344 | 811 | .needs_file = 1, |
91d8f519 | 812 | .needs_blkcg = 1, |
d3656344 | 813 | }, |
0463b6c5 | 814 | [IORING_OP_SENDMSG] = { |
d3656344 JA |
815 | .needs_mm = 1, |
816 | .needs_file = 1, | |
817 | .unbound_nonreg_file = 1, | |
ff002b30 | 818 | .needs_fs = 1, |
8a72758c | 819 | .pollout = 1, |
e8c2bc1f | 820 | .needs_async_data = 1, |
91d8f519 | 821 | .needs_blkcg = 1, |
e8c2bc1f | 822 | .async_size = sizeof(struct io_async_msghdr), |
d3656344 | 823 | }, |
0463b6c5 | 824 | [IORING_OP_RECVMSG] = { |
d3656344 JA |
825 | .needs_mm = 1, |
826 | .needs_file = 1, | |
827 | .unbound_nonreg_file = 1, | |
ff002b30 | 828 | .needs_fs = 1, |
8a72758c | 829 | .pollin = 1, |
52de1fe1 | 830 | .buffer_select = 1, |
e8c2bc1f | 831 | .needs_async_data = 1, |
91d8f519 | 832 | .needs_blkcg = 1, |
e8c2bc1f | 833 | .async_size = sizeof(struct io_async_msghdr), |
d3656344 | 834 | }, |
0463b6c5 | 835 | [IORING_OP_TIMEOUT] = { |
d3656344 | 836 | .needs_mm = 1, |
e8c2bc1f JA |
837 | .needs_async_data = 1, |
838 | .async_size = sizeof(struct io_timeout_data), | |
d3656344 | 839 | }, |
0463b6c5 PB |
840 | [IORING_OP_TIMEOUT_REMOVE] = {}, |
841 | [IORING_OP_ACCEPT] = { | |
d3656344 JA |
842 | .needs_mm = 1, |
843 | .needs_file = 1, | |
844 | .unbound_nonreg_file = 1, | |
f86cd20c | 845 | .file_table = 1, |
8a72758c | 846 | .pollin = 1, |
d3656344 | 847 | }, |
0463b6c5 PB |
848 | [IORING_OP_ASYNC_CANCEL] = {}, |
849 | [IORING_OP_LINK_TIMEOUT] = { | |
d3656344 | 850 | .needs_mm = 1, |
e8c2bc1f JA |
851 | .needs_async_data = 1, |
852 | .async_size = sizeof(struct io_timeout_data), | |
d3656344 | 853 | }, |
0463b6c5 | 854 | [IORING_OP_CONNECT] = { |
d3656344 JA |
855 | .needs_mm = 1, |
856 | .needs_file = 1, | |
857 | .unbound_nonreg_file = 1, | |
8a72758c | 858 | .pollout = 1, |
e8c2bc1f JA |
859 | .needs_async_data = 1, |
860 | .async_size = sizeof(struct io_async_connect), | |
d3656344 | 861 | }, |
0463b6c5 | 862 | [IORING_OP_FALLOCATE] = { |
d3656344 | 863 | .needs_file = 1, |
57f1a649 | 864 | .needs_fsize = 1, |
91d8f519 | 865 | .needs_blkcg = 1, |
d3656344 | 866 | }, |
0463b6c5 | 867 | [IORING_OP_OPENAT] = { |
f86cd20c | 868 | .file_table = 1, |
ff002b30 | 869 | .needs_fs = 1, |
91d8f519 | 870 | .needs_blkcg = 1, |
d3656344 | 871 | }, |
0463b6c5 | 872 | [IORING_OP_CLOSE] = { |
fd2206e4 JA |
873 | .needs_file = 1, |
874 | .needs_file_no_error = 1, | |
f86cd20c | 875 | .file_table = 1, |
91d8f519 | 876 | .needs_blkcg = 1, |
d3656344 | 877 | }, |
0463b6c5 | 878 | [IORING_OP_FILES_UPDATE] = { |
d3656344 | 879 | .needs_mm = 1, |
f86cd20c | 880 | .file_table = 1, |
d3656344 | 881 | }, |
0463b6c5 | 882 | [IORING_OP_STATX] = { |
d3656344 | 883 | .needs_mm = 1, |
ff002b30 | 884 | .needs_fs = 1, |
5b0bbee4 | 885 | .file_table = 1, |
91d8f519 | 886 | .needs_blkcg = 1, |
d3656344 | 887 | }, |
0463b6c5 | 888 | [IORING_OP_READ] = { |
3a6820f2 JA |
889 | .needs_mm = 1, |
890 | .needs_file = 1, | |
891 | .unbound_nonreg_file = 1, | |
8a72758c | 892 | .pollin = 1, |
bcda7baa | 893 | .buffer_select = 1, |
91d8f519 | 894 | .needs_blkcg = 1, |
e8c2bc1f | 895 | .async_size = sizeof(struct io_async_rw), |
3a6820f2 | 896 | }, |
0463b6c5 | 897 | [IORING_OP_WRITE] = { |
3a6820f2 JA |
898 | .needs_mm = 1, |
899 | .needs_file = 1, | |
900 | .unbound_nonreg_file = 1, | |
8a72758c | 901 | .pollout = 1, |
57f1a649 | 902 | .needs_fsize = 1, |
91d8f519 | 903 | .needs_blkcg = 1, |
e8c2bc1f | 904 | .async_size = sizeof(struct io_async_rw), |
3a6820f2 | 905 | }, |
0463b6c5 | 906 | [IORING_OP_FADVISE] = { |
4840e418 | 907 | .needs_file = 1, |
91d8f519 | 908 | .needs_blkcg = 1, |
4840e418 | 909 | }, |
0463b6c5 | 910 | [IORING_OP_MADVISE] = { |
c1ca757b | 911 | .needs_mm = 1, |
91d8f519 | 912 | .needs_blkcg = 1, |
c1ca757b | 913 | }, |
0463b6c5 | 914 | [IORING_OP_SEND] = { |
fddaface JA |
915 | .needs_mm = 1, |
916 | .needs_file = 1, | |
917 | .unbound_nonreg_file = 1, | |
8a72758c | 918 | .pollout = 1, |
91d8f519 | 919 | .needs_blkcg = 1, |
fddaface | 920 | }, |
0463b6c5 | 921 | [IORING_OP_RECV] = { |
fddaface JA |
922 | .needs_mm = 1, |
923 | .needs_file = 1, | |
924 | .unbound_nonreg_file = 1, | |
8a72758c | 925 | .pollin = 1, |
bcda7baa | 926 | .buffer_select = 1, |
91d8f519 | 927 | .needs_blkcg = 1, |
fddaface | 928 | }, |
0463b6c5 | 929 | [IORING_OP_OPENAT2] = { |
f86cd20c | 930 | .file_table = 1, |
ff002b30 | 931 | .needs_fs = 1, |
91d8f519 | 932 | .needs_blkcg = 1, |
cebdb986 | 933 | }, |
3e4827b0 JA |
934 | [IORING_OP_EPOLL_CTL] = { |
935 | .unbound_nonreg_file = 1, | |
936 | .file_table = 1, | |
937 | }, | |
7d67af2c PB |
938 | [IORING_OP_SPLICE] = { |
939 | .needs_file = 1, | |
940 | .hash_reg_file = 1, | |
941 | .unbound_nonreg_file = 1, | |
91d8f519 | 942 | .needs_blkcg = 1, |
ddf0322d JA |
943 | }, |
944 | [IORING_OP_PROVIDE_BUFFERS] = {}, | |
067524e9 | 945 | [IORING_OP_REMOVE_BUFFERS] = {}, |
f2a8d5c7 PB |
946 | [IORING_OP_TEE] = { |
947 | .needs_file = 1, | |
948 | .hash_reg_file = 1, | |
949 | .unbound_nonreg_file = 1, | |
950 | }, | |
d3656344 JA |
951 | }; |
952 | ||
2e0464d4 BM |
953 | enum io_mem_account { |
954 | ACCT_LOCKED, | |
955 | ACCT_PINNED, | |
956 | }; | |
957 | ||
81b68a5c PB |
958 | static void __io_complete_rw(struct io_kiocb *req, long res, long res2, |
959 | struct io_comp_state *cs); | |
78e19bbe | 960 | static void io_cqring_fill_event(struct io_kiocb *req, long res); |
ec9c02ad | 961 | static void io_put_req(struct io_kiocb *req); |
c40f6379 | 962 | static void io_double_put_req(struct io_kiocb *req); |
978db57e | 963 | static void __io_double_put_req(struct io_kiocb *req); |
94ae5e77 | 964 | static struct io_kiocb *io_prep_linked_timeout(struct io_kiocb *req); |
7271ef3a | 965 | static void __io_queue_linked_timeout(struct io_kiocb *req); |
94ae5e77 | 966 | static void io_queue_linked_timeout(struct io_kiocb *req); |
05f3fb3c JA |
967 | static int __io_sqe_files_update(struct io_ring_ctx *ctx, |
968 | struct io_uring_files_update *ip, | |
969 | unsigned nr_args); | |
f56040b8 | 970 | static int io_prep_work_files(struct io_kiocb *req); |
3ca405eb | 971 | static void __io_clean_op(struct io_kiocb *req); |
b41e9852 JA |
972 | static int io_file_get(struct io_submit_state *state, struct io_kiocb *req, |
973 | int fd, struct file **out_file, bool fixed); | |
c1379e24 | 974 | static void __io_queue_sqe(struct io_kiocb *req, struct io_comp_state *cs); |
4349f30e | 975 | static void io_file_put_work(struct work_struct *work); |
de0617e4 | 976 | |
b63534c4 JA |
977 | static ssize_t io_import_iovec(int rw, struct io_kiocb *req, |
978 | struct iovec **iovec, struct iov_iter *iter, | |
979 | bool needs_lock); | |
ff6165b2 JA |
980 | static int io_setup_async_rw(struct io_kiocb *req, const struct iovec *iovec, |
981 | const struct iovec *fast_iov, | |
227c0c96 | 982 | struct iov_iter *iter, bool force); |
de0617e4 | 983 | |
2b188cc1 JA |
984 | static struct kmem_cache *req_cachep; |
985 | ||
738277ad | 986 | static const struct file_operations io_uring_fops __read_mostly; |
2b188cc1 JA |
987 | |
988 | struct sock *io_uring_get_socket(struct file *file) | |
989 | { | |
990 | #if defined(CONFIG_UNIX) | |
991 | if (file->f_op == &io_uring_fops) { | |
992 | struct io_ring_ctx *ctx = file->private_data; | |
993 | ||
994 | return ctx->ring_sock->sk; | |
995 | } | |
996 | #endif | |
997 | return NULL; | |
998 | } | |
999 | EXPORT_SYMBOL(io_uring_get_socket); | |
1000 | ||
3ca405eb PB |
1001 | static inline void io_clean_op(struct io_kiocb *req) |
1002 | { | |
bb175342 PB |
1003 | if (req->flags & (REQ_F_NEED_CLEANUP | REQ_F_BUFFER_SELECTED | |
1004 | REQ_F_INFLIGHT)) | |
3ca405eb PB |
1005 | __io_clean_op(req); |
1006 | } | |
1007 | ||
4349f30e | 1008 | static void io_sq_thread_drop_mm(void) |
c40f6379 JA |
1009 | { |
1010 | struct mm_struct *mm = current->mm; | |
1011 | ||
1012 | if (mm) { | |
1013 | kthread_unuse_mm(mm); | |
1014 | mmput(mm); | |
1015 | } | |
1016 | } | |
1017 | ||
1018 | static int __io_sq_thread_acquire_mm(struct io_ring_ctx *ctx) | |
1019 | { | |
1020 | if (!current->mm) { | |
cbcf7214 | 1021 | if (unlikely(!(ctx->flags & IORING_SETUP_SQPOLL) || |
2aede0e4 JA |
1022 | !ctx->sqo_task->mm || |
1023 | !mmget_not_zero(ctx->sqo_task->mm))) | |
c40f6379 | 1024 | return -EFAULT; |
2aede0e4 | 1025 | kthread_use_mm(ctx->sqo_task->mm); |
c40f6379 JA |
1026 | } |
1027 | ||
1028 | return 0; | |
1029 | } | |
1030 | ||
1031 | static int io_sq_thread_acquire_mm(struct io_ring_ctx *ctx, | |
1032 | struct io_kiocb *req) | |
1033 | { | |
1034 | if (!io_op_defs[req->opcode].needs_mm) | |
1035 | return 0; | |
1036 | return __io_sq_thread_acquire_mm(ctx); | |
1037 | } | |
1038 | ||
91d8f519 DZ |
1039 | static void io_sq_thread_associate_blkcg(struct io_ring_ctx *ctx, |
1040 | struct cgroup_subsys_state **cur_css) | |
1041 | ||
1042 | { | |
1043 | #ifdef CONFIG_BLK_CGROUP | |
1044 | /* puts the old one when swapping */ | |
1045 | if (*cur_css != ctx->sqo_blkcg_css) { | |
1046 | kthread_associate_blkcg(ctx->sqo_blkcg_css); | |
1047 | *cur_css = ctx->sqo_blkcg_css; | |
1048 | } | |
1049 | #endif | |
1050 | } | |
1051 | ||
1052 | static void io_sq_thread_unassociate_blkcg(void) | |
1053 | { | |
1054 | #ifdef CONFIG_BLK_CGROUP | |
1055 | kthread_associate_blkcg(NULL); | |
1056 | #endif | |
1057 | } | |
1058 | ||
c40f6379 JA |
1059 | static inline void req_set_fail_links(struct io_kiocb *req) |
1060 | { | |
1061 | if ((req->flags & (REQ_F_LINK | REQ_F_HARDLINK)) == REQ_F_LINK) | |
1062 | req->flags |= REQ_F_FAIL_LINK; | |
1063 | } | |
4a38aed2 | 1064 | |
7cdaf587 XW |
1065 | /* |
1066 | * Note: must call io_req_init_async() for the first time you | |
1067 | * touch any members of io_wq_work. | |
1068 | */ | |
1069 | static inline void io_req_init_async(struct io_kiocb *req) | |
1070 | { | |
1071 | if (req->flags & REQ_F_WORK_INITIALIZED) | |
1072 | return; | |
1073 | ||
1074 | memset(&req->work, 0, sizeof(req->work)); | |
1075 | req->flags |= REQ_F_WORK_INITIALIZED; | |
1076 | } | |
1077 | ||
0cdaf760 PB |
1078 | static inline bool io_async_submit(struct io_ring_ctx *ctx) |
1079 | { | |
1080 | return ctx->flags & IORING_SETUP_SQPOLL; | |
1081 | } | |
1082 | ||
2b188cc1 JA |
1083 | static void io_ring_ctx_ref_free(struct percpu_ref *ref) |
1084 | { | |
1085 | struct io_ring_ctx *ctx = container_of(ref, struct io_ring_ctx, refs); | |
1086 | ||
0f158b4c | 1087 | complete(&ctx->ref_comp); |
2b188cc1 JA |
1088 | } |
1089 | ||
8eb7e2d0 PB |
1090 | static inline bool io_is_timeout_noseq(struct io_kiocb *req) |
1091 | { | |
1092 | return !req->timeout.off; | |
1093 | } | |
1094 | ||
2b188cc1 JA |
1095 | static struct io_ring_ctx *io_ring_ctx_alloc(struct io_uring_params *p) |
1096 | { | |
1097 | struct io_ring_ctx *ctx; | |
78076bb6 | 1098 | int hash_bits; |
2b188cc1 JA |
1099 | |
1100 | ctx = kzalloc(sizeof(*ctx), GFP_KERNEL); | |
1101 | if (!ctx) | |
1102 | return NULL; | |
1103 | ||
0ddf92e8 JA |
1104 | ctx->fallback_req = kmem_cache_alloc(req_cachep, GFP_KERNEL); |
1105 | if (!ctx->fallback_req) | |
1106 | goto err; | |
1107 | ||
78076bb6 JA |
1108 | /* |
1109 | * Use 5 bits less than the max cq entries, that should give us around | |
1110 | * 32 entries per hash list if totally full and uniformly spread. | |
1111 | */ | |
1112 | hash_bits = ilog2(p->cq_entries); | |
1113 | hash_bits -= 5; | |
1114 | if (hash_bits <= 0) | |
1115 | hash_bits = 1; | |
1116 | ctx->cancel_hash_bits = hash_bits; | |
1117 | ctx->cancel_hash = kmalloc((1U << hash_bits) * sizeof(struct hlist_head), | |
1118 | GFP_KERNEL); | |
1119 | if (!ctx->cancel_hash) | |
1120 | goto err; | |
1121 | __hash_init(ctx->cancel_hash, 1U << hash_bits); | |
1122 | ||
21482896 | 1123 | if (percpu_ref_init(&ctx->refs, io_ring_ctx_ref_free, |
206aefde JA |
1124 | PERCPU_REF_ALLOW_REINIT, GFP_KERNEL)) |
1125 | goto err; | |
2b188cc1 JA |
1126 | |
1127 | ctx->flags = p->flags; | |
90554200 | 1128 | init_waitqueue_head(&ctx->sqo_sq_wait); |
69fb2131 | 1129 | INIT_LIST_HEAD(&ctx->sqd_list); |
2b188cc1 | 1130 | init_waitqueue_head(&ctx->cq_wait); |
1d7bb1d5 | 1131 | INIT_LIST_HEAD(&ctx->cq_overflow_list); |
0f158b4c JA |
1132 | init_completion(&ctx->ref_comp); |
1133 | init_completion(&ctx->sq_thread_comp); | |
5a2e745d | 1134 | idr_init(&ctx->io_buffer_idr); |
071698e1 | 1135 | idr_init(&ctx->personality_idr); |
2b188cc1 JA |
1136 | mutex_init(&ctx->uring_lock); |
1137 | init_waitqueue_head(&ctx->wait); | |
1138 | spin_lock_init(&ctx->completion_lock); | |
540e32a0 | 1139 | INIT_LIST_HEAD(&ctx->iopoll_list); |
de0617e4 | 1140 | INIT_LIST_HEAD(&ctx->defer_list); |
5262f567 | 1141 | INIT_LIST_HEAD(&ctx->timeout_list); |
fcb323cc JA |
1142 | init_waitqueue_head(&ctx->inflight_wait); |
1143 | spin_lock_init(&ctx->inflight_lock); | |
1144 | INIT_LIST_HEAD(&ctx->inflight_list); | |
4a38aed2 JA |
1145 | INIT_DELAYED_WORK(&ctx->file_put_work, io_file_put_work); |
1146 | init_llist_head(&ctx->file_put_llist); | |
2b188cc1 | 1147 | return ctx; |
206aefde | 1148 | err: |
0ddf92e8 JA |
1149 | if (ctx->fallback_req) |
1150 | kmem_cache_free(req_cachep, ctx->fallback_req); | |
78076bb6 | 1151 | kfree(ctx->cancel_hash); |
206aefde JA |
1152 | kfree(ctx); |
1153 | return NULL; | |
2b188cc1 JA |
1154 | } |
1155 | ||
9cf7c104 | 1156 | static bool req_need_defer(struct io_kiocb *req, u32 seq) |
7adf4eaf | 1157 | { |
2bc9930e JA |
1158 | if (unlikely(req->flags & REQ_F_IO_DRAIN)) { |
1159 | struct io_ring_ctx *ctx = req->ctx; | |
a197f664 | 1160 | |
9cf7c104 | 1161 | return seq != ctx->cached_cq_tail |
31af27c7 | 1162 | + atomic_read(&ctx->cached_cq_overflow); |
2bc9930e | 1163 | } |
de0617e4 | 1164 | |
9d858b21 | 1165 | return false; |
de0617e4 JA |
1166 | } |
1167 | ||
de0617e4 | 1168 | static void __io_commit_cqring(struct io_ring_ctx *ctx) |
2b188cc1 | 1169 | { |
75b28aff | 1170 | struct io_rings *rings = ctx->rings; |
2b188cc1 | 1171 | |
07910158 PB |
1172 | /* order cqe stores with ring update */ |
1173 | smp_store_release(&rings->cq.tail, ctx->cached_cq_tail); | |
2b188cc1 | 1174 | |
07910158 PB |
1175 | if (wq_has_sleeper(&ctx->cq_wait)) { |
1176 | wake_up_interruptible(&ctx->cq_wait); | |
1177 | kill_fasync(&ctx->cq_fasync, SIGIO, POLL_IN); | |
2b188cc1 JA |
1178 | } |
1179 | } | |
1180 | ||
51a4cc11 JA |
1181 | /* |
1182 | * Returns true if we need to defer file table putting. This can only happen | |
1183 | * from the error path with REQ_F_COMP_LOCKED set. | |
1184 | */ | |
1185 | static bool io_req_clean_work(struct io_kiocb *req) | |
18d9be1a | 1186 | { |
7cdaf587 | 1187 | if (!(req->flags & REQ_F_WORK_INITIALIZED)) |
51a4cc11 JA |
1188 | return false; |
1189 | ||
1190 | req->flags &= ~REQ_F_WORK_INITIALIZED; | |
7cdaf587 | 1191 | |
cccf0ee8 JA |
1192 | if (req->work.mm) { |
1193 | mmdrop(req->work.mm); | |
1194 | req->work.mm = NULL; | |
1195 | } | |
91d8f519 DZ |
1196 | #ifdef CONFIG_BLK_CGROUP |
1197 | if (req->work.blkcg_css) | |
1198 | css_put(req->work.blkcg_css); | |
1199 | #endif | |
cccf0ee8 JA |
1200 | if (req->work.creds) { |
1201 | put_cred(req->work.creds); | |
1202 | req->work.creds = NULL; | |
1203 | } | |
ff002b30 JA |
1204 | if (req->work.fs) { |
1205 | struct fs_struct *fs = req->work.fs; | |
1206 | ||
51a4cc11 JA |
1207 | if (req->flags & REQ_F_COMP_LOCKED) |
1208 | return true; | |
1209 | ||
ff002b30 JA |
1210 | spin_lock(&req->work.fs->lock); |
1211 | if (--fs->users) | |
1212 | fs = NULL; | |
1213 | spin_unlock(&req->work.fs->lock); | |
1214 | if (fs) | |
1215 | free_fs_struct(fs); | |
b65e0dd6 | 1216 | req->work.fs = NULL; |
ff002b30 | 1217 | } |
51a4cc11 JA |
1218 | |
1219 | return false; | |
561fb04a JA |
1220 | } |
1221 | ||
cbdcb435 | 1222 | static void io_prep_async_work(struct io_kiocb *req) |
18d9be1a | 1223 | { |
d3656344 | 1224 | const struct io_op_def *def = &io_op_defs[req->opcode]; |
54a91f3b | 1225 | |
16d59803 PB |
1226 | io_req_init_async(req); |
1227 | ||
d3656344 | 1228 | if (req->flags & REQ_F_ISREG) { |
eefdf30f | 1229 | if (def->hash_reg_file || (req->ctx->flags & IORING_SETUP_IOPOLL)) |
8766dd51 | 1230 | io_wq_hash_work(&req->work, file_inode(req->file)); |
d3656344 JA |
1231 | } else { |
1232 | if (def->unbound_nonreg_file) | |
3529d8c2 | 1233 | req->work.flags |= IO_WQ_WORK_UNBOUND; |
54a91f3b | 1234 | } |
dca9cf8b PB |
1235 | if (!req->work.mm && def->needs_mm) { |
1236 | mmgrab(current->mm); | |
1237 | req->work.mm = current->mm; | |
1238 | } | |
91d8f519 DZ |
1239 | #ifdef CONFIG_BLK_CGROUP |
1240 | if (!req->work.blkcg_css && def->needs_blkcg) { | |
1241 | rcu_read_lock(); | |
1242 | req->work.blkcg_css = blkcg_css(); | |
1243 | /* | |
1244 | * This should be rare, either the cgroup is dying or the task | |
1245 | * is moving cgroups. Just punt to root for the handful of ios. | |
1246 | */ | |
1247 | if (!css_tryget_online(req->work.blkcg_css)) | |
1248 | req->work.blkcg_css = NULL; | |
1249 | rcu_read_unlock(); | |
1250 | } | |
1251 | #endif | |
dca9cf8b PB |
1252 | if (!req->work.creds) |
1253 | req->work.creds = get_current_cred(); | |
1254 | if (!req->work.fs && def->needs_fs) { | |
1255 | spin_lock(¤t->fs->lock); | |
1256 | if (!current->fs->in_exec) { | |
1257 | req->work.fs = current->fs; | |
1258 | req->work.fs->users++; | |
1259 | } else { | |
1260 | req->work.flags |= IO_WQ_WORK_CANCEL; | |
1261 | } | |
1262 | spin_unlock(¤t->fs->lock); | |
1263 | } | |
57f1a649 PB |
1264 | if (def->needs_fsize) |
1265 | req->work.fsize = rlimit(RLIMIT_FSIZE); | |
1266 | else | |
1267 | req->work.fsize = RLIM_INFINITY; | |
561fb04a | 1268 | } |
cccf0ee8 | 1269 | |
cbdcb435 | 1270 | static void io_prep_async_link(struct io_kiocb *req) |
561fb04a | 1271 | { |
cbdcb435 | 1272 | struct io_kiocb *cur; |
54a91f3b | 1273 | |
cbdcb435 PB |
1274 | io_prep_async_work(req); |
1275 | if (req->flags & REQ_F_LINK_HEAD) | |
1276 | list_for_each_entry(cur, &req->link_list, link_list) | |
1277 | io_prep_async_work(cur); | |
561fb04a JA |
1278 | } |
1279 | ||
7271ef3a | 1280 | static struct io_kiocb *__io_queue_async_work(struct io_kiocb *req) |
561fb04a | 1281 | { |
a197f664 | 1282 | struct io_ring_ctx *ctx = req->ctx; |
cbdcb435 | 1283 | struct io_kiocb *link = io_prep_linked_timeout(req); |
561fb04a | 1284 | |
8766dd51 PB |
1285 | trace_io_uring_queue_async_work(ctx, io_wq_is_hashed(&req->work), req, |
1286 | &req->work, req->flags); | |
1287 | io_wq_enqueue(ctx->io_wq, &req->work); | |
7271ef3a | 1288 | return link; |
18d9be1a JA |
1289 | } |
1290 | ||
cbdcb435 PB |
1291 | static void io_queue_async_work(struct io_kiocb *req) |
1292 | { | |
7271ef3a JA |
1293 | struct io_kiocb *link; |
1294 | ||
cbdcb435 PB |
1295 | /* init ->work of the whole link before punting */ |
1296 | io_prep_async_link(req); | |
7271ef3a JA |
1297 | link = __io_queue_async_work(req); |
1298 | ||
1299 | if (link) | |
1300 | io_queue_linked_timeout(link); | |
cbdcb435 PB |
1301 | } |
1302 | ||
5262f567 JA |
1303 | static void io_kill_timeout(struct io_kiocb *req) |
1304 | { | |
e8c2bc1f | 1305 | struct io_timeout_data *io = req->async_data; |
5262f567 JA |
1306 | int ret; |
1307 | ||
e8c2bc1f | 1308 | ret = hrtimer_try_to_cancel(&io->timer); |
5262f567 | 1309 | if (ret != -1) { |
01cec8c1 PB |
1310 | atomic_set(&req->ctx->cq_timeouts, |
1311 | atomic_read(&req->ctx->cq_timeouts) + 1); | |
135fcde8 | 1312 | list_del_init(&req->timeout.list); |
f0e20b89 | 1313 | req->flags |= REQ_F_COMP_LOCKED; |
78e19bbe | 1314 | io_cqring_fill_event(req, 0); |
ec9c02ad | 1315 | io_put_req(req); |
5262f567 JA |
1316 | } |
1317 | } | |
1318 | ||
f3606e3a JA |
1319 | static bool io_task_match(struct io_kiocb *req, struct task_struct *tsk) |
1320 | { | |
1321 | struct io_ring_ctx *ctx = req->ctx; | |
1322 | ||
1323 | if (!tsk || req->task == tsk) | |
1324 | return true; | |
534ca6d6 JA |
1325 | if (ctx->flags & IORING_SETUP_SQPOLL) { |
1326 | if (ctx->sq_data && req->task == ctx->sq_data->thread) | |
1327 | return true; | |
1328 | } | |
f3606e3a JA |
1329 | return false; |
1330 | } | |
1331 | ||
76e1b642 JA |
1332 | /* |
1333 | * Returns true if we found and killed one or more timeouts | |
1334 | */ | |
1335 | static bool io_kill_timeouts(struct io_ring_ctx *ctx, struct task_struct *tsk) | |
5262f567 JA |
1336 | { |
1337 | struct io_kiocb *req, *tmp; | |
76e1b642 | 1338 | int canceled = 0; |
5262f567 JA |
1339 | |
1340 | spin_lock_irq(&ctx->completion_lock); | |
f3606e3a | 1341 | list_for_each_entry_safe(req, tmp, &ctx->timeout_list, timeout.list) { |
76e1b642 | 1342 | if (io_task_match(req, tsk)) { |
f3606e3a | 1343 | io_kill_timeout(req); |
76e1b642 JA |
1344 | canceled++; |
1345 | } | |
f3606e3a | 1346 | } |
5262f567 | 1347 | spin_unlock_irq(&ctx->completion_lock); |
76e1b642 | 1348 | return canceled != 0; |
5262f567 JA |
1349 | } |
1350 | ||
04518945 | 1351 | static void __io_queue_deferred(struct io_ring_ctx *ctx) |
de0617e4 | 1352 | { |
04518945 | 1353 | do { |
27dc8338 PB |
1354 | struct io_defer_entry *de = list_first_entry(&ctx->defer_list, |
1355 | struct io_defer_entry, list); | |
7271ef3a | 1356 | struct io_kiocb *link; |
de0617e4 | 1357 | |
9cf7c104 | 1358 | if (req_need_defer(de->req, de->seq)) |
04518945 | 1359 | break; |
27dc8338 | 1360 | list_del_init(&de->list); |
cbdcb435 | 1361 | /* punt-init is done before queueing for defer */ |
7271ef3a JA |
1362 | link = __io_queue_async_work(de->req); |
1363 | if (link) { | |
1364 | __io_queue_linked_timeout(link); | |
1365 | /* drop submission reference */ | |
1366 | link->flags |= REQ_F_COMP_LOCKED; | |
1367 | io_put_req(link); | |
1368 | } | |
27dc8338 | 1369 | kfree(de); |
04518945 PB |
1370 | } while (!list_empty(&ctx->defer_list)); |
1371 | } | |
1372 | ||
360428f8 | 1373 | static void io_flush_timeouts(struct io_ring_ctx *ctx) |
de0617e4 | 1374 | { |
360428f8 PB |
1375 | while (!list_empty(&ctx->timeout_list)) { |
1376 | struct io_kiocb *req = list_first_entry(&ctx->timeout_list, | |
135fcde8 | 1377 | struct io_kiocb, timeout.list); |
de0617e4 | 1378 | |
8eb7e2d0 | 1379 | if (io_is_timeout_noseq(req)) |
360428f8 | 1380 | break; |
bfe68a22 PB |
1381 | if (req->timeout.target_seq != ctx->cached_cq_tail |
1382 | - atomic_read(&ctx->cq_timeouts)) | |
360428f8 | 1383 | break; |
bfe68a22 | 1384 | |
135fcde8 | 1385 | list_del_init(&req->timeout.list); |
5262f567 | 1386 | io_kill_timeout(req); |
360428f8 PB |
1387 | } |
1388 | } | |
5262f567 | 1389 | |
360428f8 PB |
1390 | static void io_commit_cqring(struct io_ring_ctx *ctx) |
1391 | { | |
1392 | io_flush_timeouts(ctx); | |
de0617e4 JA |
1393 | __io_commit_cqring(ctx); |
1394 | ||
04518945 PB |
1395 | if (unlikely(!list_empty(&ctx->defer_list))) |
1396 | __io_queue_deferred(ctx); | |
de0617e4 JA |
1397 | } |
1398 | ||
90554200 JA |
1399 | static inline bool io_sqring_full(struct io_ring_ctx *ctx) |
1400 | { | |
1401 | struct io_rings *r = ctx->rings; | |
1402 | ||
1403 | return READ_ONCE(r->sq.tail) - ctx->cached_sq_head == r->sq_ring_entries; | |
1404 | } | |
1405 | ||
2b188cc1 JA |
1406 | static struct io_uring_cqe *io_get_cqring(struct io_ring_ctx *ctx) |
1407 | { | |
75b28aff | 1408 | struct io_rings *rings = ctx->rings; |
2b188cc1 JA |
1409 | unsigned tail; |
1410 | ||
1411 | tail = ctx->cached_cq_tail; | |
115e12e5 SB |
1412 | /* |
1413 | * writes to the cq entry need to come after reading head; the | |
1414 | * control dependency is enough as we're using WRITE_ONCE to | |
1415 | * fill the cq entry | |
1416 | */ | |
75b28aff | 1417 | if (tail - READ_ONCE(rings->cq.head) == rings->cq_ring_entries) |
2b188cc1 JA |
1418 | return NULL; |
1419 | ||
1420 | ctx->cached_cq_tail++; | |
75b28aff | 1421 | return &rings->cqes[tail & ctx->cq_mask]; |
2b188cc1 JA |
1422 | } |
1423 | ||
f2842ab5 JA |
1424 | static inline bool io_should_trigger_evfd(struct io_ring_ctx *ctx) |
1425 | { | |
f0b493e6 JA |
1426 | if (!ctx->cq_ev_fd) |
1427 | return false; | |
7e55a19c SG |
1428 | if (READ_ONCE(ctx->rings->cq_flags) & IORING_CQ_EVENTFD_DISABLED) |
1429 | return false; | |
f2842ab5 JA |
1430 | if (!ctx->eventfd_async) |
1431 | return true; | |
b41e9852 | 1432 | return io_wq_current_is_worker(); |
f2842ab5 JA |
1433 | } |
1434 | ||
b41e9852 | 1435 | static void io_cqring_ev_posted(struct io_ring_ctx *ctx) |
1d7bb1d5 JA |
1436 | { |
1437 | if (waitqueue_active(&ctx->wait)) | |
1438 | wake_up(&ctx->wait); | |
534ca6d6 JA |
1439 | if (ctx->sq_data && waitqueue_active(&ctx->sq_data->wait)) |
1440 | wake_up(&ctx->sq_data->wait); | |
b41e9852 | 1441 | if (io_should_trigger_evfd(ctx)) |
1d7bb1d5 JA |
1442 | eventfd_signal(ctx->cq_ev_fd, 1); |
1443 | } | |
1444 | ||
46930143 PB |
1445 | static void io_cqring_mark_overflow(struct io_ring_ctx *ctx) |
1446 | { | |
1447 | if (list_empty(&ctx->cq_overflow_list)) { | |
1448 | clear_bit(0, &ctx->sq_check_overflow); | |
1449 | clear_bit(0, &ctx->cq_check_overflow); | |
1450 | ctx->rings->sq_flags &= ~IORING_SQ_CQ_OVERFLOW; | |
1451 | } | |
1452 | } | |
1453 | ||
e6c8aa9a JA |
1454 | static inline bool io_match_files(struct io_kiocb *req, |
1455 | struct files_struct *files) | |
1456 | { | |
1457 | if (!files) | |
1458 | return true; | |
1459 | if (req->flags & REQ_F_WORK_INITIALIZED) | |
1460 | return req->work.files == files; | |
1461 | return false; | |
1462 | } | |
1463 | ||
c4a2ed72 | 1464 | /* Returns true if there are no backlogged entries after the flush */ |
e6c8aa9a JA |
1465 | static bool io_cqring_overflow_flush(struct io_ring_ctx *ctx, bool force, |
1466 | struct task_struct *tsk, | |
1467 | struct files_struct *files) | |
1d7bb1d5 JA |
1468 | { |
1469 | struct io_rings *rings = ctx->rings; | |
e6c8aa9a | 1470 | struct io_kiocb *req, *tmp; |
1d7bb1d5 | 1471 | struct io_uring_cqe *cqe; |
1d7bb1d5 JA |
1472 | unsigned long flags; |
1473 | LIST_HEAD(list); | |
1474 | ||
1475 | if (!force) { | |
1476 | if (list_empty_careful(&ctx->cq_overflow_list)) | |
c4a2ed72 | 1477 | return true; |
1d7bb1d5 JA |
1478 | if ((ctx->cached_cq_tail - READ_ONCE(rings->cq.head) == |
1479 | rings->cq_ring_entries)) | |
c4a2ed72 | 1480 | return false; |
1d7bb1d5 JA |
1481 | } |
1482 | ||
1483 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
1484 | ||
1485 | /* if force is set, the ring is going away. always drop after that */ | |
1486 | if (force) | |
69b3e546 | 1487 | ctx->cq_overflow_flushed = 1; |
1d7bb1d5 | 1488 | |
c4a2ed72 | 1489 | cqe = NULL; |
e6c8aa9a JA |
1490 | list_for_each_entry_safe(req, tmp, &ctx->cq_overflow_list, compl.list) { |
1491 | if (tsk && req->task != tsk) | |
1492 | continue; | |
1493 | if (!io_match_files(req, files)) | |
1494 | continue; | |
1495 | ||
1d7bb1d5 JA |
1496 | cqe = io_get_cqring(ctx); |
1497 | if (!cqe && !force) | |
1498 | break; | |
1499 | ||
40d8ddd4 | 1500 | list_move(&req->compl.list, &list); |
1d7bb1d5 JA |
1501 | if (cqe) { |
1502 | WRITE_ONCE(cqe->user_data, req->user_data); | |
1503 | WRITE_ONCE(cqe->res, req->result); | |
0f7e466b | 1504 | WRITE_ONCE(cqe->flags, req->compl.cflags); |
1d7bb1d5 JA |
1505 | } else { |
1506 | WRITE_ONCE(ctx->rings->cq_overflow, | |
1507 | atomic_inc_return(&ctx->cached_cq_overflow)); | |
1508 | } | |
1509 | } | |
1510 | ||
1511 | io_commit_cqring(ctx); | |
46930143 PB |
1512 | io_cqring_mark_overflow(ctx); |
1513 | ||
1d7bb1d5 JA |
1514 | spin_unlock_irqrestore(&ctx->completion_lock, flags); |
1515 | io_cqring_ev_posted(ctx); | |
1516 | ||
1517 | while (!list_empty(&list)) { | |
40d8ddd4 PB |
1518 | req = list_first_entry(&list, struct io_kiocb, compl.list); |
1519 | list_del(&req->compl.list); | |
ec9c02ad | 1520 | io_put_req(req); |
1d7bb1d5 | 1521 | } |
c4a2ed72 JA |
1522 | |
1523 | return cqe != NULL; | |
1d7bb1d5 JA |
1524 | } |
1525 | ||
bcda7baa | 1526 | static void __io_cqring_fill_event(struct io_kiocb *req, long res, long cflags) |
2b188cc1 | 1527 | { |
78e19bbe | 1528 | struct io_ring_ctx *ctx = req->ctx; |
2b188cc1 JA |
1529 | struct io_uring_cqe *cqe; |
1530 | ||
78e19bbe | 1531 | trace_io_uring_complete(ctx, req->user_data, res); |
51c3ff62 | 1532 | |
2b188cc1 JA |
1533 | /* |
1534 | * If we can't get a cq entry, userspace overflowed the | |
1535 | * submission (by quite a lot). Increment the overflow count in | |
1536 | * the ring. | |
1537 | */ | |
1538 | cqe = io_get_cqring(ctx); | |
1d7bb1d5 | 1539 | if (likely(cqe)) { |
78e19bbe | 1540 | WRITE_ONCE(cqe->user_data, req->user_data); |
2b188cc1 | 1541 | WRITE_ONCE(cqe->res, res); |
bcda7baa | 1542 | WRITE_ONCE(cqe->flags, cflags); |
0f212204 JA |
1543 | } else if (ctx->cq_overflow_flushed || req->task->io_uring->in_idle) { |
1544 | /* | |
1545 | * If we're in ring overflow flush mode, or in task cancel mode, | |
1546 | * then we cannot store the request for later flushing, we need | |
1547 | * to drop it on the floor. | |
1548 | */ | |
498ccd9e JA |
1549 | WRITE_ONCE(ctx->rings->cq_overflow, |
1550 | atomic_inc_return(&ctx->cached_cq_overflow)); | |
1d7bb1d5 | 1551 | } else { |
ad3eb2c8 JA |
1552 | if (list_empty(&ctx->cq_overflow_list)) { |
1553 | set_bit(0, &ctx->sq_check_overflow); | |
1554 | set_bit(0, &ctx->cq_check_overflow); | |
6d5f9049 | 1555 | ctx->rings->sq_flags |= IORING_SQ_CQ_OVERFLOW; |
ad3eb2c8 | 1556 | } |
40d8ddd4 | 1557 | io_clean_op(req); |
1d7bb1d5 | 1558 | req->result = res; |
0f7e466b | 1559 | req->compl.cflags = cflags; |
40d8ddd4 PB |
1560 | refcount_inc(&req->refs); |
1561 | list_add_tail(&req->compl.list, &ctx->cq_overflow_list); | |
2b188cc1 JA |
1562 | } |
1563 | } | |
1564 | ||
bcda7baa JA |
1565 | static void io_cqring_fill_event(struct io_kiocb *req, long res) |
1566 | { | |
1567 | __io_cqring_fill_event(req, res, 0); | |
1568 | } | |
1569 | ||
e1e16097 | 1570 | static void io_cqring_add_event(struct io_kiocb *req, long res, long cflags) |
2b188cc1 | 1571 | { |
78e19bbe | 1572 | struct io_ring_ctx *ctx = req->ctx; |
2b188cc1 JA |
1573 | unsigned long flags; |
1574 | ||
1575 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
bcda7baa | 1576 | __io_cqring_fill_event(req, res, cflags); |
2b188cc1 JA |
1577 | io_commit_cqring(ctx); |
1578 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
1579 | ||
8c838788 | 1580 | io_cqring_ev_posted(ctx); |
2b188cc1 JA |
1581 | } |
1582 | ||
229a7b63 | 1583 | static void io_submit_flush_completions(struct io_comp_state *cs) |
bcda7baa | 1584 | { |
229a7b63 JA |
1585 | struct io_ring_ctx *ctx = cs->ctx; |
1586 | ||
1587 | spin_lock_irq(&ctx->completion_lock); | |
1588 | while (!list_empty(&cs->list)) { | |
1589 | struct io_kiocb *req; | |
1590 | ||
3ca405eb PB |
1591 | req = list_first_entry(&cs->list, struct io_kiocb, compl.list); |
1592 | list_del(&req->compl.list); | |
0f7e466b | 1593 | __io_cqring_fill_event(req, req->result, req->compl.cflags); |
229a7b63 JA |
1594 | if (!(req->flags & REQ_F_LINK_HEAD)) { |
1595 | req->flags |= REQ_F_COMP_LOCKED; | |
1596 | io_put_req(req); | |
1597 | } else { | |
1598 | spin_unlock_irq(&ctx->completion_lock); | |
1599 | io_put_req(req); | |
1600 | spin_lock_irq(&ctx->completion_lock); | |
1601 | } | |
1602 | } | |
1603 | io_commit_cqring(ctx); | |
1604 | spin_unlock_irq(&ctx->completion_lock); | |
1605 | ||
1606 | io_cqring_ev_posted(ctx); | |
1607 | cs->nr = 0; | |
1608 | } | |
1609 | ||
1610 | static void __io_req_complete(struct io_kiocb *req, long res, unsigned cflags, | |
1611 | struct io_comp_state *cs) | |
1612 | { | |
1613 | if (!cs) { | |
1614 | io_cqring_add_event(req, res, cflags); | |
1615 | io_put_req(req); | |
1616 | } else { | |
3ca405eb | 1617 | io_clean_op(req); |
229a7b63 | 1618 | req->result = res; |
0f7e466b | 1619 | req->compl.cflags = cflags; |
3ca405eb | 1620 | list_add_tail(&req->compl.list, &cs->list); |
229a7b63 JA |
1621 | if (++cs->nr >= 32) |
1622 | io_submit_flush_completions(cs); | |
1623 | } | |
e1e16097 JA |
1624 | } |
1625 | ||
1626 | static void io_req_complete(struct io_kiocb *req, long res) | |
bcda7baa | 1627 | { |
229a7b63 | 1628 | __io_req_complete(req, res, 0, NULL); |
bcda7baa JA |
1629 | } |
1630 | ||
0ddf92e8 JA |
1631 | static inline bool io_is_fallback_req(struct io_kiocb *req) |
1632 | { | |
1633 | return req == (struct io_kiocb *) | |
1634 | ((unsigned long) req->ctx->fallback_req & ~1UL); | |
1635 | } | |
1636 | ||
1637 | static struct io_kiocb *io_get_fallback_req(struct io_ring_ctx *ctx) | |
1638 | { | |
1639 | struct io_kiocb *req; | |
1640 | ||
1641 | req = ctx->fallback_req; | |
dd461af6 | 1642 | if (!test_and_set_bit_lock(0, (unsigned long *) &ctx->fallback_req)) |
0ddf92e8 JA |
1643 | return req; |
1644 | ||
1645 | return NULL; | |
1646 | } | |
1647 | ||
0553b8bd PB |
1648 | static struct io_kiocb *io_alloc_req(struct io_ring_ctx *ctx, |
1649 | struct io_submit_state *state) | |
2b188cc1 | 1650 | { |
f6b6c7d6 | 1651 | if (!state->free_reqs) { |
291b2821 | 1652 | gfp_t gfp = GFP_KERNEL | __GFP_NOWARN; |
2579f913 JA |
1653 | size_t sz; |
1654 | int ret; | |
1655 | ||
1656 | sz = min_t(size_t, state->ios_left, ARRAY_SIZE(state->reqs)); | |
fd6fab2c JA |
1657 | ret = kmem_cache_alloc_bulk(req_cachep, gfp, sz, state->reqs); |
1658 | ||
1659 | /* | |
1660 | * Bulk alloc is all-or-nothing. If we fail to get a batch, | |
1661 | * retry single alloc to be on the safe side. | |
1662 | */ | |
1663 | if (unlikely(ret <= 0)) { | |
1664 | state->reqs[0] = kmem_cache_alloc(req_cachep, gfp); | |
1665 | if (!state->reqs[0]) | |
0ddf92e8 | 1666 | goto fallback; |
fd6fab2c JA |
1667 | ret = 1; |
1668 | } | |
291b2821 | 1669 | state->free_reqs = ret; |
2b188cc1 JA |
1670 | } |
1671 | ||
291b2821 PB |
1672 | state->free_reqs--; |
1673 | return state->reqs[state->free_reqs]; | |
0ddf92e8 | 1674 | fallback: |
0553b8bd | 1675 | return io_get_fallback_req(ctx); |
2b188cc1 JA |
1676 | } |
1677 | ||
8da11c19 PB |
1678 | static inline void io_put_file(struct io_kiocb *req, struct file *file, |
1679 | bool fixed) | |
1680 | { | |
1681 | if (fixed) | |
05589553 | 1682 | percpu_ref_put(req->fixed_file_refs); |
8da11c19 PB |
1683 | else |
1684 | fput(file); | |
1685 | } | |
1686 | ||
51a4cc11 | 1687 | static bool io_dismantle_req(struct io_kiocb *req) |
2b188cc1 | 1688 | { |
3ca405eb | 1689 | io_clean_op(req); |
929a3af9 | 1690 | |
e8c2bc1f JA |
1691 | if (req->async_data) |
1692 | kfree(req->async_data); | |
8da11c19 PB |
1693 | if (req->file) |
1694 | io_put_file(req, req->file, (req->flags & REQ_F_FIXED_FILE)); | |
fcb323cc | 1695 | |
51a4cc11 | 1696 | return io_req_clean_work(req); |
e65ef56d JA |
1697 | } |
1698 | ||
51a4cc11 | 1699 | static void __io_free_req_finish(struct io_kiocb *req) |
c6ca97b3 | 1700 | { |
0f212204 | 1701 | struct io_uring_task *tctx = req->task->io_uring; |
51a4cc11 | 1702 | struct io_ring_ctx *ctx = req->ctx; |
c6ca97b3 | 1703 | |
0f212204 JA |
1704 | atomic_long_inc(&tctx->req_complete); |
1705 | if (tctx->in_idle) | |
1706 | wake_up(&tctx->wait); | |
e3bc8e9d JA |
1707 | put_task_struct(req->task); |
1708 | ||
b1e50e54 PB |
1709 | if (likely(!io_is_fallback_req(req))) |
1710 | kmem_cache_free(req_cachep, req); | |
1711 | else | |
ecfc5177 PB |
1712 | clear_bit_unlock(0, (unsigned long *) &ctx->fallback_req); |
1713 | percpu_ref_put(&ctx->refs); | |
e65ef56d JA |
1714 | } |
1715 | ||
51a4cc11 JA |
1716 | static void io_req_task_file_table_put(struct callback_head *cb) |
1717 | { | |
1718 | struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work); | |
1719 | struct fs_struct *fs = req->work.fs; | |
1720 | ||
1721 | spin_lock(&req->work.fs->lock); | |
1722 | if (--fs->users) | |
1723 | fs = NULL; | |
1724 | spin_unlock(&req->work.fs->lock); | |
1725 | if (fs) | |
1726 | free_fs_struct(fs); | |
1727 | req->work.fs = NULL; | |
1728 | __io_free_req_finish(req); | |
1729 | } | |
1730 | ||
1731 | static void __io_free_req(struct io_kiocb *req) | |
1732 | { | |
1733 | if (!io_dismantle_req(req)) { | |
1734 | __io_free_req_finish(req); | |
1735 | } else { | |
1736 | int ret; | |
1737 | ||
1738 | init_task_work(&req->task_work, io_req_task_file_table_put); | |
1739 | ret = task_work_add(req->task, &req->task_work, TWA_RESUME); | |
1740 | if (unlikely(ret)) { | |
1741 | struct task_struct *tsk; | |
1742 | ||
1743 | tsk = io_wq_get_task(req->ctx->io_wq); | |
1744 | task_work_add(tsk, &req->task_work, 0); | |
1745 | } | |
1746 | } | |
1747 | } | |
1748 | ||
a197f664 | 1749 | static bool io_link_cancel_timeout(struct io_kiocb *req) |
2665abfd | 1750 | { |
e8c2bc1f | 1751 | struct io_timeout_data *io = req->async_data; |
a197f664 | 1752 | struct io_ring_ctx *ctx = req->ctx; |
2665abfd JA |
1753 | int ret; |
1754 | ||
e8c2bc1f | 1755 | ret = hrtimer_try_to_cancel(&io->timer); |
2665abfd | 1756 | if (ret != -1) { |
78e19bbe | 1757 | io_cqring_fill_event(req, -ECANCELED); |
2665abfd | 1758 | io_commit_cqring(ctx); |
dea3b49c | 1759 | req->flags &= ~REQ_F_LINK_HEAD; |
ec9c02ad | 1760 | io_put_req(req); |
2665abfd JA |
1761 | return true; |
1762 | } | |
1763 | ||
1764 | return false; | |
e65ef56d JA |
1765 | } |
1766 | ||
ab0b6451 | 1767 | static bool __io_kill_linked_timeout(struct io_kiocb *req) |
9e645e11 | 1768 | { |
7c86ffee | 1769 | struct io_kiocb *link; |
ab0b6451 | 1770 | bool wake_ev; |
7c86ffee PB |
1771 | |
1772 | if (list_empty(&req->link_list)) | |
ab0b6451 | 1773 | return false; |
7c86ffee PB |
1774 | link = list_first_entry(&req->link_list, struct io_kiocb, link_list); |
1775 | if (link->opcode != IORING_OP_LINK_TIMEOUT) | |
ab0b6451 | 1776 | return false; |
7c86ffee PB |
1777 | |
1778 | list_del_init(&link->link_list); | |
9b7adba9 | 1779 | link->flags |= REQ_F_COMP_LOCKED; |
7c86ffee PB |
1780 | wake_ev = io_link_cancel_timeout(link); |
1781 | req->flags &= ~REQ_F_LINK_TIMEOUT; | |
ab0b6451 JA |
1782 | return wake_ev; |
1783 | } | |
1784 | ||
1785 | static void io_kill_linked_timeout(struct io_kiocb *req) | |
9e645e11 | 1786 | { |
2665abfd | 1787 | struct io_ring_ctx *ctx = req->ctx; |
ab0b6451 | 1788 | bool wake_ev; |
9e645e11 | 1789 | |
ab0b6451 JA |
1790 | if (!(req->flags & REQ_F_COMP_LOCKED)) { |
1791 | unsigned long flags; | |
1792 | ||
1793 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
1794 | wake_ev = __io_kill_linked_timeout(req); | |
7c86ffee | 1795 | spin_unlock_irqrestore(&ctx->completion_lock, flags); |
ab0b6451 JA |
1796 | } else { |
1797 | wake_ev = __io_kill_linked_timeout(req); | |
1798 | } | |
1799 | ||
7c86ffee PB |
1800 | if (wake_ev) |
1801 | io_cqring_ev_posted(ctx); | |
1802 | } | |
1803 | ||
9b5f7bd9 | 1804 | static struct io_kiocb *io_req_link_next(struct io_kiocb *req) |
7c86ffee PB |
1805 | { |
1806 | struct io_kiocb *nxt; | |
4d7dd462 | 1807 | |
9e645e11 JA |
1808 | /* |
1809 | * The list should never be empty when we are called here. But could | |
1810 | * potentially happen if the chain is messed up, check to be on the | |
1811 | * safe side. | |
1812 | */ | |
7c86ffee | 1813 | if (unlikely(list_empty(&req->link_list))) |
9b5f7bd9 | 1814 | return NULL; |
2665abfd | 1815 | |
7c86ffee PB |
1816 | nxt = list_first_entry(&req->link_list, struct io_kiocb, link_list); |
1817 | list_del_init(&req->link_list); | |
1818 | if (!list_empty(&nxt->link_list)) | |
1819 | nxt->flags |= REQ_F_LINK_HEAD; | |
9b5f7bd9 | 1820 | return nxt; |
9e645e11 JA |
1821 | } |
1822 | ||
1823 | /* | |
dea3b49c | 1824 | * Called if REQ_F_LINK_HEAD is set, and we fail the head request |
9e645e11 | 1825 | */ |
7c86ffee | 1826 | static void __io_fail_links(struct io_kiocb *req) |
9e645e11 | 1827 | { |
2665abfd | 1828 | struct io_ring_ctx *ctx = req->ctx; |
9e645e11 JA |
1829 | |
1830 | while (!list_empty(&req->link_list)) { | |
4493233e PB |
1831 | struct io_kiocb *link = list_first_entry(&req->link_list, |
1832 | struct io_kiocb, link_list); | |
9e645e11 | 1833 | |
4493233e | 1834 | list_del_init(&link->link_list); |
c826bd7a | 1835 | trace_io_uring_fail_link(req, link); |
2665abfd | 1836 | |
7c86ffee | 1837 | io_cqring_fill_event(link, -ECANCELED); |
9b7adba9 | 1838 | link->flags |= REQ_F_COMP_LOCKED; |
7c86ffee | 1839 | __io_double_put_req(link); |
5d960724 | 1840 | req->flags &= ~REQ_F_LINK_TIMEOUT; |
9e645e11 | 1841 | } |
2665abfd JA |
1842 | |
1843 | io_commit_cqring(ctx); | |
2665abfd | 1844 | io_cqring_ev_posted(ctx); |
9e645e11 JA |
1845 | } |
1846 | ||
7c86ffee | 1847 | static void io_fail_links(struct io_kiocb *req) |
9e645e11 | 1848 | { |
7c86ffee | 1849 | struct io_ring_ctx *ctx = req->ctx; |
2665abfd | 1850 | |
7c86ffee | 1851 | if (!(req->flags & REQ_F_COMP_LOCKED)) { |
2665abfd JA |
1852 | unsigned long flags; |
1853 | ||
2665abfd | 1854 | spin_lock_irqsave(&ctx->completion_lock, flags); |
7c86ffee | 1855 | __io_fail_links(req); |
2665abfd JA |
1856 | spin_unlock_irqrestore(&ctx->completion_lock, flags); |
1857 | } else { | |
7c86ffee | 1858 | __io_fail_links(req); |
9e645e11 JA |
1859 | } |
1860 | ||
2665abfd | 1861 | io_cqring_ev_posted(ctx); |
9e645e11 JA |
1862 | } |
1863 | ||
3fa5e0f3 | 1864 | static struct io_kiocb *__io_req_find_next(struct io_kiocb *req) |
c69f8dbe | 1865 | { |
9b0d911a | 1866 | req->flags &= ~REQ_F_LINK_HEAD; |
7c86ffee PB |
1867 | if (req->flags & REQ_F_LINK_TIMEOUT) |
1868 | io_kill_linked_timeout(req); | |
944e58bf | 1869 | |
9e645e11 JA |
1870 | /* |
1871 | * If LINK is set, we have dependent requests in this chain. If we | |
1872 | * didn't fail this request, queue the first one up, moving any other | |
1873 | * dependencies to the next request. In case of failure, fail the rest | |
1874 | * of the chain. | |
1875 | */ | |
9b5f7bd9 PB |
1876 | if (likely(!(req->flags & REQ_F_FAIL_LINK))) |
1877 | return io_req_link_next(req); | |
1878 | io_fail_links(req); | |
1879 | return NULL; | |
4d7dd462 | 1880 | } |
9e645e11 | 1881 | |
3fa5e0f3 PB |
1882 | static struct io_kiocb *io_req_find_next(struct io_kiocb *req) |
1883 | { | |
1884 | if (likely(!(req->flags & REQ_F_LINK_HEAD))) | |
1885 | return NULL; | |
1886 | return __io_req_find_next(req); | |
1887 | } | |
1888 | ||
87c4311f | 1889 | static int io_req_task_work_add(struct io_kiocb *req, bool twa_signal_ok) |
c2c4c83c JA |
1890 | { |
1891 | struct task_struct *tsk = req->task; | |
1892 | struct io_ring_ctx *ctx = req->ctx; | |
0ba9c9ed | 1893 | int ret, notify; |
c2c4c83c | 1894 | |
6200b0ae JA |
1895 | if (tsk->flags & PF_EXITING) |
1896 | return -ESRCH; | |
1897 | ||
c2c4c83c | 1898 | /* |
0ba9c9ed JA |
1899 | * SQPOLL kernel thread doesn't need notification, just a wakeup. For |
1900 | * all other cases, use TWA_SIGNAL unconditionally to ensure we're | |
1901 | * processing task_work. There's no reliable way to tell if TWA_RESUME | |
1902 | * will do the job. | |
c2c4c83c | 1903 | */ |
0ba9c9ed | 1904 | notify = 0; |
fd7d6de2 | 1905 | if (!(ctx->flags & IORING_SETUP_SQPOLL) && twa_signal_ok) |
c2c4c83c JA |
1906 | notify = TWA_SIGNAL; |
1907 | ||
87c4311f | 1908 | ret = task_work_add(tsk, &req->task_work, notify); |
c2c4c83c JA |
1909 | if (!ret) |
1910 | wake_up_process(tsk); | |
0ba9c9ed | 1911 | |
c2c4c83c JA |
1912 | return ret; |
1913 | } | |
1914 | ||
c40f6379 JA |
1915 | static void __io_req_task_cancel(struct io_kiocb *req, int error) |
1916 | { | |
1917 | struct io_ring_ctx *ctx = req->ctx; | |
1918 | ||
1919 | spin_lock_irq(&ctx->completion_lock); | |
1920 | io_cqring_fill_event(req, error); | |
1921 | io_commit_cqring(ctx); | |
1922 | spin_unlock_irq(&ctx->completion_lock); | |
1923 | ||
1924 | io_cqring_ev_posted(ctx); | |
1925 | req_set_fail_links(req); | |
1926 | io_double_put_req(req); | |
1927 | } | |
1928 | ||
1929 | static void io_req_task_cancel(struct callback_head *cb) | |
1930 | { | |
1931 | struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work); | |
87ceb6a6 | 1932 | struct io_ring_ctx *ctx = req->ctx; |
c40f6379 JA |
1933 | |
1934 | __io_req_task_cancel(req, -ECANCELED); | |
87ceb6a6 | 1935 | percpu_ref_put(&ctx->refs); |
c40f6379 JA |
1936 | } |
1937 | ||
1938 | static void __io_req_task_submit(struct io_kiocb *req) | |
1939 | { | |
1940 | struct io_ring_ctx *ctx = req->ctx; | |
1941 | ||
c40f6379 JA |
1942 | if (!__io_sq_thread_acquire_mm(ctx)) { |
1943 | mutex_lock(&ctx->uring_lock); | |
c1379e24 | 1944 | __io_queue_sqe(req, NULL); |
c40f6379 JA |
1945 | mutex_unlock(&ctx->uring_lock); |
1946 | } else { | |
1947 | __io_req_task_cancel(req, -EFAULT); | |
1948 | } | |
1949 | } | |
1950 | ||
1951 | static void io_req_task_submit(struct callback_head *cb) | |
1952 | { | |
1953 | struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work); | |
6d816e08 | 1954 | struct io_ring_ctx *ctx = req->ctx; |
c40f6379 JA |
1955 | |
1956 | __io_req_task_submit(req); | |
6d816e08 | 1957 | percpu_ref_put(&ctx->refs); |
c40f6379 JA |
1958 | } |
1959 | ||
1960 | static void io_req_task_queue(struct io_kiocb *req) | |
1961 | { | |
c40f6379 JA |
1962 | int ret; |
1963 | ||
1964 | init_task_work(&req->task_work, io_req_task_submit); | |
6d816e08 | 1965 | percpu_ref_get(&req->ctx->refs); |
c40f6379 | 1966 | |
87c4311f | 1967 | ret = io_req_task_work_add(req, true); |
c40f6379 | 1968 | if (unlikely(ret)) { |
c2c4c83c JA |
1969 | struct task_struct *tsk; |
1970 | ||
c40f6379 JA |
1971 | init_task_work(&req->task_work, io_req_task_cancel); |
1972 | tsk = io_wq_get_task(req->ctx->io_wq); | |
c2c4c83c JA |
1973 | task_work_add(tsk, &req->task_work, 0); |
1974 | wake_up_process(tsk); | |
c40f6379 | 1975 | } |
c40f6379 JA |
1976 | } |
1977 | ||
c3524383 | 1978 | static void io_queue_next(struct io_kiocb *req) |
c69f8dbe | 1979 | { |
9b5f7bd9 | 1980 | struct io_kiocb *nxt = io_req_find_next(req); |
944e58bf PB |
1981 | |
1982 | if (nxt) | |
906a8c3f | 1983 | io_req_task_queue(nxt); |
c69f8dbe JL |
1984 | } |
1985 | ||
c3524383 | 1986 | static void io_free_req(struct io_kiocb *req) |
7a743e22 | 1987 | { |
c3524383 PB |
1988 | io_queue_next(req); |
1989 | __io_free_req(req); | |
1990 | } | |
8766dd51 | 1991 | |
2d6500d4 PB |
1992 | struct req_batch { |
1993 | void *reqs[IO_IOPOLL_BATCH]; | |
1994 | int to_free; | |
7a743e22 | 1995 | |
5af1d13e PB |
1996 | struct task_struct *task; |
1997 | int task_refs; | |
2d6500d4 PB |
1998 | }; |
1999 | ||
5af1d13e PB |
2000 | static inline void io_init_req_batch(struct req_batch *rb) |
2001 | { | |
2002 | rb->to_free = 0; | |
2003 | rb->task_refs = 0; | |
2004 | rb->task = NULL; | |
2005 | } | |
2006 | ||
2d6500d4 PB |
2007 | static void __io_req_free_batch_flush(struct io_ring_ctx *ctx, |
2008 | struct req_batch *rb) | |
2009 | { | |
2010 | kmem_cache_free_bulk(req_cachep, rb->to_free, rb->reqs); | |
2011 | percpu_ref_put_many(&ctx->refs, rb->to_free); | |
2012 | rb->to_free = 0; | |
2013 | } | |
2014 | ||
2015 | static void io_req_free_batch_finish(struct io_ring_ctx *ctx, | |
2016 | struct req_batch *rb) | |
2017 | { | |
2018 | if (rb->to_free) | |
2019 | __io_req_free_batch_flush(ctx, rb); | |
5af1d13e | 2020 | if (rb->task) { |
0f212204 | 2021 | atomic_long_add(rb->task_refs, &rb->task->io_uring->req_complete); |
5af1d13e PB |
2022 | put_task_struct_many(rb->task, rb->task_refs); |
2023 | rb->task = NULL; | |
2024 | } | |
2d6500d4 PB |
2025 | } |
2026 | ||
2027 | static void io_req_free_batch(struct req_batch *rb, struct io_kiocb *req) | |
2028 | { | |
2029 | if (unlikely(io_is_fallback_req(req))) { | |
2030 | io_free_req(req); | |
2031 | return; | |
2032 | } | |
2033 | if (req->flags & REQ_F_LINK_HEAD) | |
2034 | io_queue_next(req); | |
2035 | ||
e3bc8e9d | 2036 | if (req->task != rb->task) { |
0f212204 JA |
2037 | if (rb->task) { |
2038 | atomic_long_add(rb->task_refs, &rb->task->io_uring->req_complete); | |
e3bc8e9d | 2039 | put_task_struct_many(rb->task, rb->task_refs); |
0f212204 | 2040 | } |
e3bc8e9d JA |
2041 | rb->task = req->task; |
2042 | rb->task_refs = 0; | |
5af1d13e | 2043 | } |
e3bc8e9d | 2044 | rb->task_refs++; |
5af1d13e | 2045 | |
51a4cc11 | 2046 | WARN_ON_ONCE(io_dismantle_req(req)); |
2d6500d4 PB |
2047 | rb->reqs[rb->to_free++] = req; |
2048 | if (unlikely(rb->to_free == ARRAY_SIZE(rb->reqs))) | |
2049 | __io_req_free_batch_flush(req->ctx, rb); | |
7a743e22 PB |
2050 | } |
2051 | ||
ba816ad6 JA |
2052 | /* |
2053 | * Drop reference to request, return next in chain (if there is one) if this | |
2054 | * was the last reference to this request. | |
2055 | */ | |
9b5f7bd9 | 2056 | static struct io_kiocb *io_put_req_find_next(struct io_kiocb *req) |
e65ef56d | 2057 | { |
9b5f7bd9 PB |
2058 | struct io_kiocb *nxt = NULL; |
2059 | ||
2a44f467 | 2060 | if (refcount_dec_and_test(&req->refs)) { |
9b5f7bd9 | 2061 | nxt = io_req_find_next(req); |
4d7dd462 | 2062 | __io_free_req(req); |
2a44f467 | 2063 | } |
9b5f7bd9 | 2064 | return nxt; |
2b188cc1 JA |
2065 | } |
2066 | ||
e65ef56d JA |
2067 | static void io_put_req(struct io_kiocb *req) |
2068 | { | |
2069 | if (refcount_dec_and_test(&req->refs)) | |
2070 | io_free_req(req); | |
2b188cc1 JA |
2071 | } |
2072 | ||
f4db7182 | 2073 | static struct io_wq_work *io_steal_work(struct io_kiocb *req) |
7a743e22 | 2074 | { |
6df1db6b | 2075 | struct io_kiocb *nxt; |
f4db7182 | 2076 | |
7a743e22 | 2077 | /* |
f4db7182 PB |
2078 | * A ref is owned by io-wq in which context we're. So, if that's the |
2079 | * last one, it's safe to steal next work. False negatives are Ok, | |
2080 | * it just will be re-punted async in io_put_work() | |
7a743e22 | 2081 | */ |
f4db7182 PB |
2082 | if (refcount_read(&req->refs) != 1) |
2083 | return NULL; | |
7a743e22 | 2084 | |
9b5f7bd9 | 2085 | nxt = io_req_find_next(req); |
6df1db6b | 2086 | return nxt ? &nxt->work : NULL; |
7a743e22 PB |
2087 | } |
2088 | ||
978db57e JA |
2089 | /* |
2090 | * Must only be used if we don't need to care about links, usually from | |
2091 | * within the completion handling itself. | |
2092 | */ | |
2093 | static void __io_double_put_req(struct io_kiocb *req) | |
78e19bbe JA |
2094 | { |
2095 | /* drop both submit and complete references */ | |
2096 | if (refcount_sub_and_test(2, &req->refs)) | |
2097 | __io_free_req(req); | |
2098 | } | |
2099 | ||
978db57e JA |
2100 | static void io_double_put_req(struct io_kiocb *req) |
2101 | { | |
2102 | /* drop both submit and complete references */ | |
2103 | if (refcount_sub_and_test(2, &req->refs)) | |
2104 | io_free_req(req); | |
2105 | } | |
2106 | ||
1d7bb1d5 | 2107 | static unsigned io_cqring_events(struct io_ring_ctx *ctx, bool noflush) |
a3a0e43f | 2108 | { |
84f97dc2 JA |
2109 | struct io_rings *rings = ctx->rings; |
2110 | ||
ad3eb2c8 JA |
2111 | if (test_bit(0, &ctx->cq_check_overflow)) { |
2112 | /* | |
2113 | * noflush == true is from the waitqueue handler, just ensure | |
2114 | * we wake up the task, and the next invocation will flush the | |
2115 | * entries. We cannot safely to it from here. | |
2116 | */ | |
2117 | if (noflush && !list_empty(&ctx->cq_overflow_list)) | |
2118 | return -1U; | |
1d7bb1d5 | 2119 | |
e6c8aa9a | 2120 | io_cqring_overflow_flush(ctx, false, NULL, NULL); |
ad3eb2c8 | 2121 | } |
1d7bb1d5 | 2122 | |
a3a0e43f JA |
2123 | /* See comment at the top of this file */ |
2124 | smp_rmb(); | |
ad3eb2c8 | 2125 | return ctx->cached_cq_tail - READ_ONCE(rings->cq.head); |
a3a0e43f JA |
2126 | } |
2127 | ||
fb5ccc98 PB |
2128 | static inline unsigned int io_sqring_entries(struct io_ring_ctx *ctx) |
2129 | { | |
2130 | struct io_rings *rings = ctx->rings; | |
2131 | ||
2132 | /* make sure SQ entry isn't read before tail */ | |
2133 | return smp_load_acquire(&rings->sq.tail) - ctx->cached_sq_head; | |
2134 | } | |
2135 | ||
8ff069bf | 2136 | static unsigned int io_put_kbuf(struct io_kiocb *req, struct io_buffer *kbuf) |
e94f141b | 2137 | { |
8ff069bf | 2138 | unsigned int cflags; |
e94f141b | 2139 | |
bcda7baa JA |
2140 | cflags = kbuf->bid << IORING_CQE_BUFFER_SHIFT; |
2141 | cflags |= IORING_CQE_F_BUFFER; | |
0e1b6fe3 | 2142 | req->flags &= ~REQ_F_BUFFER_SELECTED; |
bcda7baa JA |
2143 | kfree(kbuf); |
2144 | return cflags; | |
e94f141b JA |
2145 | } |
2146 | ||
8ff069bf | 2147 | static inline unsigned int io_put_rw_kbuf(struct io_kiocb *req) |
bcda7baa | 2148 | { |
4d954c25 | 2149 | struct io_buffer *kbuf; |
bcda7baa | 2150 | |
4d954c25 | 2151 | kbuf = (struct io_buffer *) (unsigned long) req->rw.addr; |
8ff069bf PB |
2152 | return io_put_kbuf(req, kbuf); |
2153 | } | |
2154 | ||
4c6e277c JA |
2155 | static inline bool io_run_task_work(void) |
2156 | { | |
6200b0ae JA |
2157 | /* |
2158 | * Not safe to run on exiting task, and the task_work handling will | |
2159 | * not add work to such a task. | |
2160 | */ | |
2161 | if (unlikely(current->flags & PF_EXITING)) | |
2162 | return false; | |
4c6e277c JA |
2163 | if (current->task_works) { |
2164 | __set_current_state(TASK_RUNNING); | |
2165 | task_work_run(); | |
2166 | return true; | |
2167 | } | |
2168 | ||
2169 | return false; | |
bcda7baa JA |
2170 | } |
2171 | ||
bbde017a XW |
2172 | static void io_iopoll_queue(struct list_head *again) |
2173 | { | |
2174 | struct io_kiocb *req; | |
2175 | ||
2176 | do { | |
d21ffe7e PB |
2177 | req = list_first_entry(again, struct io_kiocb, inflight_entry); |
2178 | list_del(&req->inflight_entry); | |
81b68a5c | 2179 | __io_complete_rw(req, -EAGAIN, 0, NULL); |
bbde017a XW |
2180 | } while (!list_empty(again)); |
2181 | } | |
2182 | ||
def596e9 JA |
2183 | /* |
2184 | * Find and free completed poll iocbs | |
2185 | */ | |
2186 | static void io_iopoll_complete(struct io_ring_ctx *ctx, unsigned int *nr_events, | |
2187 | struct list_head *done) | |
2188 | { | |
8237e045 | 2189 | struct req_batch rb; |
def596e9 | 2190 | struct io_kiocb *req; |
bbde017a XW |
2191 | LIST_HEAD(again); |
2192 | ||
2193 | /* order with ->result store in io_complete_rw_iopoll() */ | |
2194 | smp_rmb(); | |
def596e9 | 2195 | |
5af1d13e | 2196 | io_init_req_batch(&rb); |
def596e9 | 2197 | while (!list_empty(done)) { |
bcda7baa JA |
2198 | int cflags = 0; |
2199 | ||
d21ffe7e | 2200 | req = list_first_entry(done, struct io_kiocb, inflight_entry); |
bbde017a | 2201 | if (READ_ONCE(req->result) == -EAGAIN) { |
56450c20 | 2202 | req->result = 0; |
bbde017a | 2203 | req->iopoll_completed = 0; |
d21ffe7e | 2204 | list_move_tail(&req->inflight_entry, &again); |
bbde017a XW |
2205 | continue; |
2206 | } | |
d21ffe7e | 2207 | list_del(&req->inflight_entry); |
def596e9 | 2208 | |
bcda7baa | 2209 | if (req->flags & REQ_F_BUFFER_SELECTED) |
8ff069bf | 2210 | cflags = io_put_rw_kbuf(req); |
bcda7baa JA |
2211 | |
2212 | __io_cqring_fill_event(req, req->result, cflags); | |
def596e9 JA |
2213 | (*nr_events)++; |
2214 | ||
c3524383 | 2215 | if (refcount_dec_and_test(&req->refs)) |
2d6500d4 | 2216 | io_req_free_batch(&rb, req); |
def596e9 | 2217 | } |
def596e9 | 2218 | |
09bb8394 | 2219 | io_commit_cqring(ctx); |
32b2244a XW |
2220 | if (ctx->flags & IORING_SETUP_SQPOLL) |
2221 | io_cqring_ev_posted(ctx); | |
2d6500d4 | 2222 | io_req_free_batch_finish(ctx, &rb); |
581f9810 | 2223 | |
bbde017a XW |
2224 | if (!list_empty(&again)) |
2225 | io_iopoll_queue(&again); | |
581f9810 BM |
2226 | } |
2227 | ||
def596e9 JA |
2228 | static int io_do_iopoll(struct io_ring_ctx *ctx, unsigned int *nr_events, |
2229 | long min) | |
2230 | { | |
2231 | struct io_kiocb *req, *tmp; | |
2232 | LIST_HEAD(done); | |
2233 | bool spin; | |
2234 | int ret; | |
2235 | ||
2236 | /* | |
2237 | * Only spin for completions if we don't have multiple devices hanging | |
2238 | * off our complete list, and we're under the requested amount. | |
2239 | */ | |
2240 | spin = !ctx->poll_multi_file && *nr_events < min; | |
2241 | ||
2242 | ret = 0; | |
d21ffe7e | 2243 | list_for_each_entry_safe(req, tmp, &ctx->iopoll_list, inflight_entry) { |
9adbd45d | 2244 | struct kiocb *kiocb = &req->rw.kiocb; |
def596e9 JA |
2245 | |
2246 | /* | |
581f9810 BM |
2247 | * Move completed and retryable entries to our local lists. |
2248 | * If we find a request that requires polling, break out | |
2249 | * and complete those lists first, if we have entries there. | |
def596e9 | 2250 | */ |
65a6543d | 2251 | if (READ_ONCE(req->iopoll_completed)) { |
d21ffe7e | 2252 | list_move_tail(&req->inflight_entry, &done); |
def596e9 JA |
2253 | continue; |
2254 | } | |
2255 | if (!list_empty(&done)) | |
2256 | break; | |
2257 | ||
2258 | ret = kiocb->ki_filp->f_op->iopoll(kiocb, spin); | |
2259 | if (ret < 0) | |
2260 | break; | |
2261 | ||
3aadc23e PB |
2262 | /* iopoll may have completed current req */ |
2263 | if (READ_ONCE(req->iopoll_completed)) | |
d21ffe7e | 2264 | list_move_tail(&req->inflight_entry, &done); |
3aadc23e | 2265 | |
def596e9 JA |
2266 | if (ret && spin) |
2267 | spin = false; | |
2268 | ret = 0; | |
2269 | } | |
2270 | ||
2271 | if (!list_empty(&done)) | |
2272 | io_iopoll_complete(ctx, nr_events, &done); | |
2273 | ||
2274 | return ret; | |
2275 | } | |
2276 | ||
2277 | /* | |
d195a66e | 2278 | * Poll for a minimum of 'min' events. Note that if min == 0 we consider that a |
def596e9 JA |
2279 | * non-spinning poll check - we'll still enter the driver poll loop, but only |
2280 | * as a non-spinning completion check. | |
2281 | */ | |
2282 | static int io_iopoll_getevents(struct io_ring_ctx *ctx, unsigned int *nr_events, | |
2283 | long min) | |
2284 | { | |
540e32a0 | 2285 | while (!list_empty(&ctx->iopoll_list) && !need_resched()) { |
def596e9 JA |
2286 | int ret; |
2287 | ||
2288 | ret = io_do_iopoll(ctx, nr_events, min); | |
2289 | if (ret < 0) | |
2290 | return ret; | |
eba0a4dd | 2291 | if (*nr_events >= min) |
def596e9 JA |
2292 | return 0; |
2293 | } | |
2294 | ||
2295 | return 1; | |
2296 | } | |
2297 | ||
2298 | /* | |
2299 | * We can't just wait for polled events to come to us, we have to actively | |
2300 | * find and complete them. | |
2301 | */ | |
b2edc0a7 | 2302 | static void io_iopoll_try_reap_events(struct io_ring_ctx *ctx) |
def596e9 JA |
2303 | { |
2304 | if (!(ctx->flags & IORING_SETUP_IOPOLL)) | |
2305 | return; | |
2306 | ||
2307 | mutex_lock(&ctx->uring_lock); | |
540e32a0 | 2308 | while (!list_empty(&ctx->iopoll_list)) { |
def596e9 JA |
2309 | unsigned int nr_events = 0; |
2310 | ||
b2edc0a7 | 2311 | io_do_iopoll(ctx, &nr_events, 0); |
08f5439f | 2312 | |
b2edc0a7 PB |
2313 | /* let it sleep and repeat later if can't complete a request */ |
2314 | if (nr_events == 0) | |
2315 | break; | |
08f5439f JA |
2316 | /* |
2317 | * Ensure we allow local-to-the-cpu processing to take place, | |
2318 | * in this case we need to ensure that we reap all events. | |
3fcee5a6 | 2319 | * Also let task_work, etc. to progress by releasing the mutex |
08f5439f | 2320 | */ |
3fcee5a6 PB |
2321 | if (need_resched()) { |
2322 | mutex_unlock(&ctx->uring_lock); | |
2323 | cond_resched(); | |
2324 | mutex_lock(&ctx->uring_lock); | |
2325 | } | |
def596e9 JA |
2326 | } |
2327 | mutex_unlock(&ctx->uring_lock); | |
2328 | } | |
2329 | ||
7668b92a | 2330 | static int io_iopoll_check(struct io_ring_ctx *ctx, long min) |
def596e9 | 2331 | { |
7668b92a | 2332 | unsigned int nr_events = 0; |
2b2ed975 | 2333 | int iters = 0, ret = 0; |
500f9fba | 2334 | |
c7849be9 XW |
2335 | /* |
2336 | * We disallow the app entering submit/complete with polling, but we | |
2337 | * still need to lock the ring to prevent racing with polled issue | |
2338 | * that got punted to a workqueue. | |
2339 | */ | |
2340 | mutex_lock(&ctx->uring_lock); | |
def596e9 | 2341 | do { |
a3a0e43f JA |
2342 | /* |
2343 | * Don't enter poll loop if we already have events pending. | |
2344 | * If we do, we can potentially be spinning for commands that | |
2345 | * already triggered a CQE (eg in error). | |
2346 | */ | |
1d7bb1d5 | 2347 | if (io_cqring_events(ctx, false)) |
a3a0e43f JA |
2348 | break; |
2349 | ||
500f9fba JA |
2350 | /* |
2351 | * If a submit got punted to a workqueue, we can have the | |
2352 | * application entering polling for a command before it gets | |
2353 | * issued. That app will hold the uring_lock for the duration | |
2354 | * of the poll right here, so we need to take a breather every | |
2355 | * now and then to ensure that the issue has a chance to add | |
2356 | * the poll to the issued list. Otherwise we can spin here | |
2357 | * forever, while the workqueue is stuck trying to acquire the | |
2358 | * very same mutex. | |
2359 | */ | |
2360 | if (!(++iters & 7)) { | |
2361 | mutex_unlock(&ctx->uring_lock); | |
4c6e277c | 2362 | io_run_task_work(); |
500f9fba JA |
2363 | mutex_lock(&ctx->uring_lock); |
2364 | } | |
2365 | ||
7668b92a | 2366 | ret = io_iopoll_getevents(ctx, &nr_events, min); |
def596e9 JA |
2367 | if (ret <= 0) |
2368 | break; | |
2369 | ret = 0; | |
7668b92a | 2370 | } while (min && !nr_events && !need_resched()); |
def596e9 | 2371 | |
500f9fba | 2372 | mutex_unlock(&ctx->uring_lock); |
def596e9 JA |
2373 | return ret; |
2374 | } | |
2375 | ||
491381ce | 2376 | static void kiocb_end_write(struct io_kiocb *req) |
2b188cc1 | 2377 | { |
491381ce JA |
2378 | /* |
2379 | * Tell lockdep we inherited freeze protection from submission | |
2380 | * thread. | |
2381 | */ | |
2382 | if (req->flags & REQ_F_ISREG) { | |
2383 | struct inode *inode = file_inode(req->file); | |
2b188cc1 | 2384 | |
491381ce | 2385 | __sb_writers_acquired(inode->i_sb, SB_FREEZE_WRITE); |
2b188cc1 | 2386 | } |
491381ce | 2387 | file_end_write(req->file); |
2b188cc1 JA |
2388 | } |
2389 | ||
a1d7c393 JA |
2390 | static void io_complete_rw_common(struct kiocb *kiocb, long res, |
2391 | struct io_comp_state *cs) | |
2b188cc1 | 2392 | { |
9adbd45d | 2393 | struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb); |
bcda7baa | 2394 | int cflags = 0; |
2b188cc1 | 2395 | |
491381ce JA |
2396 | if (kiocb->ki_flags & IOCB_WRITE) |
2397 | kiocb_end_write(req); | |
2b188cc1 | 2398 | |
4e88d6e7 JA |
2399 | if (res != req->result) |
2400 | req_set_fail_links(req); | |
bcda7baa | 2401 | if (req->flags & REQ_F_BUFFER_SELECTED) |
8ff069bf | 2402 | cflags = io_put_rw_kbuf(req); |
a1d7c393 | 2403 | __io_req_complete(req, res, cflags, cs); |
ba816ad6 JA |
2404 | } |
2405 | ||
b63534c4 JA |
2406 | #ifdef CONFIG_BLOCK |
2407 | static bool io_resubmit_prep(struct io_kiocb *req, int error) | |
2408 | { | |
2409 | struct iovec inline_vecs[UIO_FASTIOV], *iovec = inline_vecs; | |
2410 | ssize_t ret = -ECANCELED; | |
2411 | struct iov_iter iter; | |
2412 | int rw; | |
2413 | ||
2414 | if (error) { | |
2415 | ret = error; | |
2416 | goto end_req; | |
2417 | } | |
2418 | ||
2419 | switch (req->opcode) { | |
2420 | case IORING_OP_READV: | |
2421 | case IORING_OP_READ_FIXED: | |
2422 | case IORING_OP_READ: | |
2423 | rw = READ; | |
2424 | break; | |
2425 | case IORING_OP_WRITEV: | |
2426 | case IORING_OP_WRITE_FIXED: | |
2427 | case IORING_OP_WRITE: | |
2428 | rw = WRITE; | |
2429 | break; | |
2430 | default: | |
2431 | printk_once(KERN_WARNING "io_uring: bad opcode in resubmit %d\n", | |
2432 | req->opcode); | |
2433 | goto end_req; | |
2434 | } | |
2435 | ||
e8c2bc1f | 2436 | if (!req->async_data) { |
8f3d7496 JA |
2437 | ret = io_import_iovec(rw, req, &iovec, &iter, false); |
2438 | if (ret < 0) | |
2439 | goto end_req; | |
2440 | ret = io_setup_async_rw(req, iovec, inline_vecs, &iter, false); | |
2441 | if (!ret) | |
2442 | return true; | |
2443 | kfree(iovec); | |
2444 | } else { | |
b63534c4 | 2445 | return true; |
8f3d7496 | 2446 | } |
b63534c4 | 2447 | end_req: |
b63534c4 | 2448 | req_set_fail_links(req); |
e1e16097 | 2449 | io_req_complete(req, ret); |
b63534c4 JA |
2450 | return false; |
2451 | } | |
b63534c4 JA |
2452 | #endif |
2453 | ||
2454 | static bool io_rw_reissue(struct io_kiocb *req, long res) | |
2455 | { | |
2456 | #ifdef CONFIG_BLOCK | |
355afaeb | 2457 | umode_t mode = file_inode(req->file)->i_mode; |
b63534c4 JA |
2458 | int ret; |
2459 | ||
355afaeb JA |
2460 | if (!S_ISBLK(mode) && !S_ISREG(mode)) |
2461 | return false; | |
b63534c4 JA |
2462 | if ((res != -EAGAIN && res != -EOPNOTSUPP) || io_wq_current_is_worker()) |
2463 | return false; | |
2464 | ||
fdee946d | 2465 | ret = io_sq_thread_acquire_mm(req->ctx, req); |
6d816e08 | 2466 | |
fdee946d JA |
2467 | if (io_resubmit_prep(req, ret)) { |
2468 | refcount_inc(&req->refs); | |
2469 | io_queue_async_work(req); | |
b63534c4 | 2470 | return true; |
fdee946d JA |
2471 | } |
2472 | ||
b63534c4 JA |
2473 | #endif |
2474 | return false; | |
2475 | } | |
2476 | ||
a1d7c393 JA |
2477 | static void __io_complete_rw(struct io_kiocb *req, long res, long res2, |
2478 | struct io_comp_state *cs) | |
2479 | { | |
2480 | if (!io_rw_reissue(req, res)) | |
2481 | io_complete_rw_common(&req->rw.kiocb, res, cs); | |
ba816ad6 JA |
2482 | } |
2483 | ||
2484 | static void io_complete_rw(struct kiocb *kiocb, long res, long res2) | |
2485 | { | |
9adbd45d | 2486 | struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb); |
ba816ad6 | 2487 | |
a1d7c393 | 2488 | __io_complete_rw(req, res, res2, NULL); |
2b188cc1 JA |
2489 | } |
2490 | ||
def596e9 JA |
2491 | static void io_complete_rw_iopoll(struct kiocb *kiocb, long res, long res2) |
2492 | { | |
9adbd45d | 2493 | struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb); |
def596e9 | 2494 | |
491381ce JA |
2495 | if (kiocb->ki_flags & IOCB_WRITE) |
2496 | kiocb_end_write(req); | |
def596e9 | 2497 | |
2d7d6792 | 2498 | if (res != -EAGAIN && res != req->result) |
4e88d6e7 | 2499 | req_set_fail_links(req); |
bbde017a XW |
2500 | |
2501 | WRITE_ONCE(req->result, res); | |
2502 | /* order with io_poll_complete() checking ->result */ | |
cd664b0e PB |
2503 | smp_wmb(); |
2504 | WRITE_ONCE(req->iopoll_completed, 1); | |
def596e9 JA |
2505 | } |
2506 | ||
2507 | /* | |
2508 | * After the iocb has been issued, it's safe to be found on the poll list. | |
2509 | * Adding the kiocb to the list AFTER submission ensures that we don't | |
2510 | * find it from a io_iopoll_getevents() thread before the issuer is done | |
2511 | * accessing the kiocb cookie. | |
2512 | */ | |
2513 | static void io_iopoll_req_issued(struct io_kiocb *req) | |
2514 | { | |
2515 | struct io_ring_ctx *ctx = req->ctx; | |
2516 | ||
2517 | /* | |
2518 | * Track whether we have multiple files in our lists. This will impact | |
2519 | * how we do polling eventually, not spinning if we're on potentially | |
2520 | * different devices. | |
2521 | */ | |
540e32a0 | 2522 | if (list_empty(&ctx->iopoll_list)) { |
def596e9 JA |
2523 | ctx->poll_multi_file = false; |
2524 | } else if (!ctx->poll_multi_file) { | |
2525 | struct io_kiocb *list_req; | |
2526 | ||
540e32a0 | 2527 | list_req = list_first_entry(&ctx->iopoll_list, struct io_kiocb, |
d21ffe7e | 2528 | inflight_entry); |
9adbd45d | 2529 | if (list_req->file != req->file) |
def596e9 JA |
2530 | ctx->poll_multi_file = true; |
2531 | } | |
2532 | ||
2533 | /* | |
2534 | * For fast devices, IO may have already completed. If it has, add | |
2535 | * it to the front so we find it first. | |
2536 | */ | |
65a6543d | 2537 | if (READ_ONCE(req->iopoll_completed)) |
d21ffe7e | 2538 | list_add(&req->inflight_entry, &ctx->iopoll_list); |
def596e9 | 2539 | else |
d21ffe7e | 2540 | list_add_tail(&req->inflight_entry, &ctx->iopoll_list); |
bdcd3eab | 2541 | |
534ca6d6 JA |
2542 | if ((ctx->flags & IORING_SETUP_SQPOLL) && |
2543 | wq_has_sleeper(&ctx->sq_data->wait)) | |
2544 | wake_up(&ctx->sq_data->wait); | |
def596e9 JA |
2545 | } |
2546 | ||
9f13c35b | 2547 | static void __io_state_file_put(struct io_submit_state *state) |
9a56a232 | 2548 | { |
06ef3608 PB |
2549 | if (state->has_refs) |
2550 | fput_many(state->file, state->has_refs); | |
9f13c35b PB |
2551 | state->file = NULL; |
2552 | } | |
2553 | ||
2554 | static inline void io_state_file_put(struct io_submit_state *state) | |
2555 | { | |
2556 | if (state->file) | |
2557 | __io_state_file_put(state); | |
9a56a232 JA |
2558 | } |
2559 | ||
2560 | /* | |
2561 | * Get as many references to a file as we have IOs left in this submission, | |
2562 | * assuming most submissions are for one file, or at least that each file | |
2563 | * has more than one submission. | |
2564 | */ | |
8da11c19 | 2565 | static struct file *__io_file_get(struct io_submit_state *state, int fd) |
9a56a232 JA |
2566 | { |
2567 | if (!state) | |
2568 | return fget(fd); | |
2569 | ||
2570 | if (state->file) { | |
2571 | if (state->fd == fd) { | |
06ef3608 | 2572 | state->has_refs--; |
9a56a232 JA |
2573 | state->ios_left--; |
2574 | return state->file; | |
2575 | } | |
9f13c35b | 2576 | __io_state_file_put(state); |
9a56a232 JA |
2577 | } |
2578 | state->file = fget_many(fd, state->ios_left); | |
2579 | if (!state->file) | |
2580 | return NULL; | |
2581 | ||
2582 | state->fd = fd; | |
9a56a232 | 2583 | state->ios_left--; |
06ef3608 | 2584 | state->has_refs = state->ios_left; |
9a56a232 JA |
2585 | return state->file; |
2586 | } | |
2587 | ||
4503b767 JA |
2588 | static bool io_bdev_nowait(struct block_device *bdev) |
2589 | { | |
2590 | #ifdef CONFIG_BLOCK | |
2591 | return !bdev || queue_is_mq(bdev_get_queue(bdev)); | |
2592 | #else | |
2593 | return true; | |
2594 | #endif | |
2595 | } | |
2596 | ||
2b188cc1 JA |
2597 | /* |
2598 | * If we tracked the file through the SCM inflight mechanism, we could support | |
2599 | * any file. For now, just ensure that anything potentially problematic is done | |
2600 | * inline. | |
2601 | */ | |
af197f50 | 2602 | static bool io_file_supports_async(struct file *file, int rw) |
2b188cc1 JA |
2603 | { |
2604 | umode_t mode = file_inode(file)->i_mode; | |
2605 | ||
4503b767 JA |
2606 | if (S_ISBLK(mode)) { |
2607 | if (io_bdev_nowait(file->f_inode->i_bdev)) | |
2608 | return true; | |
2609 | return false; | |
2610 | } | |
2611 | if (S_ISCHR(mode) || S_ISSOCK(mode)) | |
2b188cc1 | 2612 | return true; |
4503b767 JA |
2613 | if (S_ISREG(mode)) { |
2614 | if (io_bdev_nowait(file->f_inode->i_sb->s_bdev) && | |
2615 | file->f_op != &io_uring_fops) | |
2616 | return true; | |
2617 | return false; | |
2618 | } | |
2b188cc1 | 2619 | |
c5b85625 JA |
2620 | /* any ->read/write should understand O_NONBLOCK */ |
2621 | if (file->f_flags & O_NONBLOCK) | |
2622 | return true; | |
2623 | ||
af197f50 JA |
2624 | if (!(file->f_mode & FMODE_NOWAIT)) |
2625 | return false; | |
2626 | ||
2627 | if (rw == READ) | |
2628 | return file->f_op->read_iter != NULL; | |
2629 | ||
2630 | return file->f_op->write_iter != NULL; | |
2b188cc1 JA |
2631 | } |
2632 | ||
a88fc400 | 2633 | static int io_prep_rw(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
2b188cc1 | 2634 | { |
def596e9 | 2635 | struct io_ring_ctx *ctx = req->ctx; |
9adbd45d | 2636 | struct kiocb *kiocb = &req->rw.kiocb; |
09bb8394 JA |
2637 | unsigned ioprio; |
2638 | int ret; | |
2b188cc1 | 2639 | |
491381ce JA |
2640 | if (S_ISREG(file_inode(req->file)->i_mode)) |
2641 | req->flags |= REQ_F_ISREG; | |
2642 | ||
2b188cc1 | 2643 | kiocb->ki_pos = READ_ONCE(sqe->off); |
ba04291e JA |
2644 | if (kiocb->ki_pos == -1 && !(req->file->f_mode & FMODE_STREAM)) { |
2645 | req->flags |= REQ_F_CUR_POS; | |
2646 | kiocb->ki_pos = req->file->f_pos; | |
2647 | } | |
2b188cc1 | 2648 | kiocb->ki_hint = ki_hint_validate(file_write_hint(kiocb->ki_filp)); |
3e577dcd PB |
2649 | kiocb->ki_flags = iocb_flags(kiocb->ki_filp); |
2650 | ret = kiocb_set_rw_flags(kiocb, READ_ONCE(sqe->rw_flags)); | |
2651 | if (unlikely(ret)) | |
2652 | return ret; | |
2b188cc1 JA |
2653 | |
2654 | ioprio = READ_ONCE(sqe->ioprio); | |
2655 | if (ioprio) { | |
2656 | ret = ioprio_check_cap(ioprio); | |
2657 | if (ret) | |
09bb8394 | 2658 | return ret; |
2b188cc1 JA |
2659 | |
2660 | kiocb->ki_ioprio = ioprio; | |
2661 | } else | |
2662 | kiocb->ki_ioprio = get_current_ioprio(); | |
2663 | ||
8449eeda | 2664 | /* don't allow async punt if RWF_NOWAIT was requested */ |
c5b85625 | 2665 | if (kiocb->ki_flags & IOCB_NOWAIT) |
8449eeda SB |
2666 | req->flags |= REQ_F_NOWAIT; |
2667 | ||
def596e9 | 2668 | if (ctx->flags & IORING_SETUP_IOPOLL) { |
def596e9 JA |
2669 | if (!(kiocb->ki_flags & IOCB_DIRECT) || |
2670 | !kiocb->ki_filp->f_op->iopoll) | |
09bb8394 | 2671 | return -EOPNOTSUPP; |
2b188cc1 | 2672 | |
def596e9 JA |
2673 | kiocb->ki_flags |= IOCB_HIPRI; |
2674 | kiocb->ki_complete = io_complete_rw_iopoll; | |
65a6543d | 2675 | req->iopoll_completed = 0; |
def596e9 | 2676 | } else { |
09bb8394 JA |
2677 | if (kiocb->ki_flags & IOCB_HIPRI) |
2678 | return -EINVAL; | |
def596e9 JA |
2679 | kiocb->ki_complete = io_complete_rw; |
2680 | } | |
9adbd45d | 2681 | |
3529d8c2 JA |
2682 | req->rw.addr = READ_ONCE(sqe->addr); |
2683 | req->rw.len = READ_ONCE(sqe->len); | |
4f4eeba8 | 2684 | req->buf_index = READ_ONCE(sqe->buf_index); |
2b188cc1 | 2685 | return 0; |
2b188cc1 JA |
2686 | } |
2687 | ||
2688 | static inline void io_rw_done(struct kiocb *kiocb, ssize_t ret) | |
2689 | { | |
2690 | switch (ret) { | |
2691 | case -EIOCBQUEUED: | |
2692 | break; | |
2693 | case -ERESTARTSYS: | |
2694 | case -ERESTARTNOINTR: | |
2695 | case -ERESTARTNOHAND: | |
2696 | case -ERESTART_RESTARTBLOCK: | |
2697 | /* | |
2698 | * We can't just restart the syscall, since previously | |
2699 | * submitted sqes may already be in progress. Just fail this | |
2700 | * IO with EINTR. | |
2701 | */ | |
2702 | ret = -EINTR; | |
df561f66 | 2703 | fallthrough; |
2b188cc1 JA |
2704 | default: |
2705 | kiocb->ki_complete(kiocb, ret, 0); | |
2706 | } | |
2707 | } | |
2708 | ||
a1d7c393 JA |
2709 | static void kiocb_done(struct kiocb *kiocb, ssize_t ret, |
2710 | struct io_comp_state *cs) | |
ba816ad6 | 2711 | { |
ba04291e | 2712 | struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb); |
e8c2bc1f | 2713 | struct io_async_rw *io = req->async_data; |
ba04291e | 2714 | |
227c0c96 | 2715 | /* add previously done IO, if any */ |
e8c2bc1f | 2716 | if (io && io->bytes_done > 0) { |
227c0c96 | 2717 | if (ret < 0) |
e8c2bc1f | 2718 | ret = io->bytes_done; |
227c0c96 | 2719 | else |
e8c2bc1f | 2720 | ret += io->bytes_done; |
227c0c96 JA |
2721 | } |
2722 | ||
ba04291e JA |
2723 | if (req->flags & REQ_F_CUR_POS) |
2724 | req->file->f_pos = kiocb->ki_pos; | |
bcaec089 | 2725 | if (ret >= 0 && kiocb->ki_complete == io_complete_rw) |
a1d7c393 | 2726 | __io_complete_rw(req, ret, 0, cs); |
ba816ad6 JA |
2727 | else |
2728 | io_rw_done(kiocb, ret); | |
2729 | } | |
2730 | ||
9adbd45d | 2731 | static ssize_t io_import_fixed(struct io_kiocb *req, int rw, |
7d009165 | 2732 | struct iov_iter *iter) |
edafccee | 2733 | { |
9adbd45d JA |
2734 | struct io_ring_ctx *ctx = req->ctx; |
2735 | size_t len = req->rw.len; | |
edafccee | 2736 | struct io_mapped_ubuf *imu; |
4be1c615 | 2737 | u16 index, buf_index = req->buf_index; |
edafccee JA |
2738 | size_t offset; |
2739 | u64 buf_addr; | |
2740 | ||
edafccee JA |
2741 | if (unlikely(buf_index >= ctx->nr_user_bufs)) |
2742 | return -EFAULT; | |
edafccee JA |
2743 | index = array_index_nospec(buf_index, ctx->nr_user_bufs); |
2744 | imu = &ctx->user_bufs[index]; | |
9adbd45d | 2745 | buf_addr = req->rw.addr; |
edafccee JA |
2746 | |
2747 | /* overflow */ | |
2748 | if (buf_addr + len < buf_addr) | |
2749 | return -EFAULT; | |
2750 | /* not inside the mapped region */ | |
2751 | if (buf_addr < imu->ubuf || buf_addr + len > imu->ubuf + imu->len) | |
2752 | return -EFAULT; | |
2753 | ||
2754 | /* | |
2755 | * May not be a start of buffer, set size appropriately | |
2756 | * and advance us to the beginning. | |
2757 | */ | |
2758 | offset = buf_addr - imu->ubuf; | |
2759 | iov_iter_bvec(iter, rw, imu->bvec, imu->nr_bvecs, offset + len); | |
bd11b3a3 JA |
2760 | |
2761 | if (offset) { | |
2762 | /* | |
2763 | * Don't use iov_iter_advance() here, as it's really slow for | |
2764 | * using the latter parts of a big fixed buffer - it iterates | |
2765 | * over each segment manually. We can cheat a bit here, because | |
2766 | * we know that: | |
2767 | * | |
2768 | * 1) it's a BVEC iter, we set it up | |
2769 | * 2) all bvecs are PAGE_SIZE in size, except potentially the | |
2770 | * first and last bvec | |
2771 | * | |
2772 | * So just find our index, and adjust the iterator afterwards. | |
2773 | * If the offset is within the first bvec (or the whole first | |
2774 | * bvec, just use iov_iter_advance(). This makes it easier | |
2775 | * since we can just skip the first segment, which may not | |
2776 | * be PAGE_SIZE aligned. | |
2777 | */ | |
2778 | const struct bio_vec *bvec = imu->bvec; | |
2779 | ||
2780 | if (offset <= bvec->bv_len) { | |
2781 | iov_iter_advance(iter, offset); | |
2782 | } else { | |
2783 | unsigned long seg_skip; | |
2784 | ||
2785 | /* skip first vec */ | |
2786 | offset -= bvec->bv_len; | |
2787 | seg_skip = 1 + (offset >> PAGE_SHIFT); | |
2788 | ||
2789 | iter->bvec = bvec + seg_skip; | |
2790 | iter->nr_segs -= seg_skip; | |
99c79f66 | 2791 | iter->count -= bvec->bv_len + offset; |
bd11b3a3 | 2792 | iter->iov_offset = offset & ~PAGE_MASK; |
bd11b3a3 JA |
2793 | } |
2794 | } | |
2795 | ||
5e559561 | 2796 | return len; |
edafccee JA |
2797 | } |
2798 | ||
bcda7baa JA |
2799 | static void io_ring_submit_unlock(struct io_ring_ctx *ctx, bool needs_lock) |
2800 | { | |
2801 | if (needs_lock) | |
2802 | mutex_unlock(&ctx->uring_lock); | |
2803 | } | |
2804 | ||
2805 | static void io_ring_submit_lock(struct io_ring_ctx *ctx, bool needs_lock) | |
2806 | { | |
2807 | /* | |
2808 | * "Normal" inline submissions always hold the uring_lock, since we | |
2809 | * grab it from the system call. Same is true for the SQPOLL offload. | |
2810 | * The only exception is when we've detached the request and issue it | |
2811 | * from an async worker thread, grab the lock for that case. | |
2812 | */ | |
2813 | if (needs_lock) | |
2814 | mutex_lock(&ctx->uring_lock); | |
2815 | } | |
2816 | ||
2817 | static struct io_buffer *io_buffer_select(struct io_kiocb *req, size_t *len, | |
2818 | int bgid, struct io_buffer *kbuf, | |
2819 | bool needs_lock) | |
2820 | { | |
2821 | struct io_buffer *head; | |
2822 | ||
2823 | if (req->flags & REQ_F_BUFFER_SELECTED) | |
2824 | return kbuf; | |
2825 | ||
2826 | io_ring_submit_lock(req->ctx, needs_lock); | |
2827 | ||
2828 | lockdep_assert_held(&req->ctx->uring_lock); | |
2829 | ||
2830 | head = idr_find(&req->ctx->io_buffer_idr, bgid); | |
2831 | if (head) { | |
2832 | if (!list_empty(&head->list)) { | |
2833 | kbuf = list_last_entry(&head->list, struct io_buffer, | |
2834 | list); | |
2835 | list_del(&kbuf->list); | |
2836 | } else { | |
2837 | kbuf = head; | |
2838 | idr_remove(&req->ctx->io_buffer_idr, bgid); | |
2839 | } | |
2840 | if (*len > kbuf->len) | |
2841 | *len = kbuf->len; | |
2842 | } else { | |
2843 | kbuf = ERR_PTR(-ENOBUFS); | |
2844 | } | |
2845 | ||
2846 | io_ring_submit_unlock(req->ctx, needs_lock); | |
2847 | ||
2848 | return kbuf; | |
2849 | } | |
2850 | ||
4d954c25 JA |
2851 | static void __user *io_rw_buffer_select(struct io_kiocb *req, size_t *len, |
2852 | bool needs_lock) | |
2853 | { | |
2854 | struct io_buffer *kbuf; | |
4f4eeba8 | 2855 | u16 bgid; |
4d954c25 JA |
2856 | |
2857 | kbuf = (struct io_buffer *) (unsigned long) req->rw.addr; | |
4f4eeba8 | 2858 | bgid = req->buf_index; |
4d954c25 JA |
2859 | kbuf = io_buffer_select(req, len, bgid, kbuf, needs_lock); |
2860 | if (IS_ERR(kbuf)) | |
2861 | return kbuf; | |
2862 | req->rw.addr = (u64) (unsigned long) kbuf; | |
2863 | req->flags |= REQ_F_BUFFER_SELECTED; | |
2864 | return u64_to_user_ptr(kbuf->addr); | |
2865 | } | |
2866 | ||
2867 | #ifdef CONFIG_COMPAT | |
2868 | static ssize_t io_compat_import(struct io_kiocb *req, struct iovec *iov, | |
2869 | bool needs_lock) | |
2870 | { | |
2871 | struct compat_iovec __user *uiov; | |
2872 | compat_ssize_t clen; | |
2873 | void __user *buf; | |
2874 | ssize_t len; | |
2875 | ||
2876 | uiov = u64_to_user_ptr(req->rw.addr); | |
2877 | if (!access_ok(uiov, sizeof(*uiov))) | |
2878 | return -EFAULT; | |
2879 | if (__get_user(clen, &uiov->iov_len)) | |
2880 | return -EFAULT; | |
2881 | if (clen < 0) | |
2882 | return -EINVAL; | |
2883 | ||
2884 | len = clen; | |
2885 | buf = io_rw_buffer_select(req, &len, needs_lock); | |
2886 | if (IS_ERR(buf)) | |
2887 | return PTR_ERR(buf); | |
2888 | iov[0].iov_base = buf; | |
2889 | iov[0].iov_len = (compat_size_t) len; | |
2890 | return 0; | |
2891 | } | |
2892 | #endif | |
2893 | ||
2894 | static ssize_t __io_iov_buffer_select(struct io_kiocb *req, struct iovec *iov, | |
2895 | bool needs_lock) | |
2896 | { | |
2897 | struct iovec __user *uiov = u64_to_user_ptr(req->rw.addr); | |
2898 | void __user *buf; | |
2899 | ssize_t len; | |
2900 | ||
2901 | if (copy_from_user(iov, uiov, sizeof(*uiov))) | |
2902 | return -EFAULT; | |
2903 | ||
2904 | len = iov[0].iov_len; | |
2905 | if (len < 0) | |
2906 | return -EINVAL; | |
2907 | buf = io_rw_buffer_select(req, &len, needs_lock); | |
2908 | if (IS_ERR(buf)) | |
2909 | return PTR_ERR(buf); | |
2910 | iov[0].iov_base = buf; | |
2911 | iov[0].iov_len = len; | |
2912 | return 0; | |
2913 | } | |
2914 | ||
2915 | static ssize_t io_iov_buffer_select(struct io_kiocb *req, struct iovec *iov, | |
2916 | bool needs_lock) | |
2917 | { | |
dddb3e26 JA |
2918 | if (req->flags & REQ_F_BUFFER_SELECTED) { |
2919 | struct io_buffer *kbuf; | |
2920 | ||
2921 | kbuf = (struct io_buffer *) (unsigned long) req->rw.addr; | |
2922 | iov[0].iov_base = u64_to_user_ptr(kbuf->addr); | |
2923 | iov[0].iov_len = kbuf->len; | |
4d954c25 | 2924 | return 0; |
dddb3e26 | 2925 | } |
4d954c25 JA |
2926 | if (!req->rw.len) |
2927 | return 0; | |
2928 | else if (req->rw.len > 1) | |
2929 | return -EINVAL; | |
2930 | ||
2931 | #ifdef CONFIG_COMPAT | |
2932 | if (req->ctx->compat) | |
2933 | return io_compat_import(req, iov, needs_lock); | |
2934 | #endif | |
2935 | ||
2936 | return __io_iov_buffer_select(req, iov, needs_lock); | |
2937 | } | |
2938 | ||
8452fd0c JA |
2939 | static ssize_t __io_import_iovec(int rw, struct io_kiocb *req, |
2940 | struct iovec **iovec, struct iov_iter *iter, | |
2941 | bool needs_lock) | |
2b188cc1 | 2942 | { |
9adbd45d JA |
2943 | void __user *buf = u64_to_user_ptr(req->rw.addr); |
2944 | size_t sqe_len = req->rw.len; | |
4d954c25 | 2945 | ssize_t ret; |
edafccee JA |
2946 | u8 opcode; |
2947 | ||
d625c6ee | 2948 | opcode = req->opcode; |
7d009165 | 2949 | if (opcode == IORING_OP_READ_FIXED || opcode == IORING_OP_WRITE_FIXED) { |
edafccee | 2950 | *iovec = NULL; |
9adbd45d | 2951 | return io_import_fixed(req, rw, iter); |
edafccee | 2952 | } |
2b188cc1 | 2953 | |
bcda7baa | 2954 | /* buffer index only valid with fixed read/write, or buffer select */ |
4f4eeba8 | 2955 | if (req->buf_index && !(req->flags & REQ_F_BUFFER_SELECT)) |
9adbd45d JA |
2956 | return -EINVAL; |
2957 | ||
3a6820f2 | 2958 | if (opcode == IORING_OP_READ || opcode == IORING_OP_WRITE) { |
bcda7baa | 2959 | if (req->flags & REQ_F_BUFFER_SELECT) { |
4d954c25 | 2960 | buf = io_rw_buffer_select(req, &sqe_len, needs_lock); |
867a23ea | 2961 | if (IS_ERR(buf)) |
4d954c25 | 2962 | return PTR_ERR(buf); |
3f9d6441 | 2963 | req->rw.len = sqe_len; |
bcda7baa JA |
2964 | } |
2965 | ||
3a6820f2 JA |
2966 | ret = import_single_range(rw, buf, sqe_len, *iovec, iter); |
2967 | *iovec = NULL; | |
3a901598 | 2968 | return ret < 0 ? ret : sqe_len; |
3a6820f2 JA |
2969 | } |
2970 | ||
4d954c25 JA |
2971 | if (req->flags & REQ_F_BUFFER_SELECT) { |
2972 | ret = io_iov_buffer_select(req, *iovec, needs_lock); | |
3f9d6441 JA |
2973 | if (!ret) { |
2974 | ret = (*iovec)->iov_len; | |
2975 | iov_iter_init(iter, rw, *iovec, 1, ret); | |
2976 | } | |
4d954c25 JA |
2977 | *iovec = NULL; |
2978 | return ret; | |
2979 | } | |
2980 | ||
2b188cc1 | 2981 | #ifdef CONFIG_COMPAT |
cf6fd4bd | 2982 | if (req->ctx->compat) |
2b188cc1 JA |
2983 | return compat_import_iovec(rw, buf, sqe_len, UIO_FASTIOV, |
2984 | iovec, iter); | |
2985 | #endif | |
2986 | ||
2987 | return import_iovec(rw, buf, sqe_len, UIO_FASTIOV, iovec, iter); | |
2988 | } | |
2989 | ||
8452fd0c JA |
2990 | static ssize_t io_import_iovec(int rw, struct io_kiocb *req, |
2991 | struct iovec **iovec, struct iov_iter *iter, | |
2992 | bool needs_lock) | |
2993 | { | |
e8c2bc1f JA |
2994 | struct io_async_rw *iorw = req->async_data; |
2995 | ||
2996 | if (!iorw) | |
8452fd0c JA |
2997 | return __io_import_iovec(rw, req, iovec, iter, needs_lock); |
2998 | *iovec = NULL; | |
e8c2bc1f | 2999 | return iov_iter_count(&iorw->iter); |
8452fd0c JA |
3000 | } |
3001 | ||
0fef9483 JA |
3002 | static inline loff_t *io_kiocb_ppos(struct kiocb *kiocb) |
3003 | { | |
5b09e37e | 3004 | return (kiocb->ki_filp->f_mode & FMODE_STREAM) ? NULL : &kiocb->ki_pos; |
0fef9483 JA |
3005 | } |
3006 | ||
31b51510 | 3007 | /* |
32960613 JA |
3008 | * For files that don't have ->read_iter() and ->write_iter(), handle them |
3009 | * by looping over ->read() or ->write() manually. | |
31b51510 | 3010 | */ |
32960613 JA |
3011 | static ssize_t loop_rw_iter(int rw, struct file *file, struct kiocb *kiocb, |
3012 | struct iov_iter *iter) | |
3013 | { | |
3014 | ssize_t ret = 0; | |
3015 | ||
3016 | /* | |
3017 | * Don't support polled IO through this interface, and we can't | |
3018 | * support non-blocking either. For the latter, this just causes | |
3019 | * the kiocb to be handled from an async context. | |
3020 | */ | |
3021 | if (kiocb->ki_flags & IOCB_HIPRI) | |
3022 | return -EOPNOTSUPP; | |
3023 | if (kiocb->ki_flags & IOCB_NOWAIT) | |
3024 | return -EAGAIN; | |
3025 | ||
3026 | while (iov_iter_count(iter)) { | |
311ae9e1 | 3027 | struct iovec iovec; |
32960613 JA |
3028 | ssize_t nr; |
3029 | ||
311ae9e1 PB |
3030 | if (!iov_iter_is_bvec(iter)) { |
3031 | iovec = iov_iter_iovec(iter); | |
3032 | } else { | |
3033 | /* fixed buffers import bvec */ | |
3034 | iovec.iov_base = kmap(iter->bvec->bv_page) | |
3035 | + iter->iov_offset; | |
3036 | iovec.iov_len = min(iter->count, | |
3037 | iter->bvec->bv_len - iter->iov_offset); | |
3038 | } | |
3039 | ||
32960613 JA |
3040 | if (rw == READ) { |
3041 | nr = file->f_op->read(file, iovec.iov_base, | |
0fef9483 | 3042 | iovec.iov_len, io_kiocb_ppos(kiocb)); |
32960613 JA |
3043 | } else { |
3044 | nr = file->f_op->write(file, iovec.iov_base, | |
0fef9483 | 3045 | iovec.iov_len, io_kiocb_ppos(kiocb)); |
32960613 JA |
3046 | } |
3047 | ||
311ae9e1 PB |
3048 | if (iov_iter_is_bvec(iter)) |
3049 | kunmap(iter->bvec->bv_page); | |
3050 | ||
32960613 JA |
3051 | if (nr < 0) { |
3052 | if (!ret) | |
3053 | ret = nr; | |
3054 | break; | |
3055 | } | |
3056 | ret += nr; | |
3057 | if (nr != iovec.iov_len) | |
3058 | break; | |
3059 | iov_iter_advance(iter, nr); | |
3060 | } | |
3061 | ||
3062 | return ret; | |
3063 | } | |
3064 | ||
ff6165b2 JA |
3065 | static void io_req_map_rw(struct io_kiocb *req, const struct iovec *iovec, |
3066 | const struct iovec *fast_iov, struct iov_iter *iter) | |
f67676d1 | 3067 | { |
e8c2bc1f | 3068 | struct io_async_rw *rw = req->async_data; |
b64e3444 | 3069 | |
ff6165b2 | 3070 | memcpy(&rw->iter, iter, sizeof(*iter)); |
afb87658 | 3071 | rw->free_iovec = iovec; |
227c0c96 | 3072 | rw->bytes_done = 0; |
ff6165b2 JA |
3073 | /* can only be fixed buffers, no need to do anything */ |
3074 | if (iter->type == ITER_BVEC) | |
3075 | return; | |
b64e3444 | 3076 | if (!iovec) { |
ff6165b2 JA |
3077 | unsigned iov_off = 0; |
3078 | ||
3079 | rw->iter.iov = rw->fast_iov; | |
3080 | if (iter->iov != fast_iov) { | |
3081 | iov_off = iter->iov - fast_iov; | |
3082 | rw->iter.iov += iov_off; | |
3083 | } | |
3084 | if (rw->fast_iov != fast_iov) | |
3085 | memcpy(rw->fast_iov + iov_off, fast_iov + iov_off, | |
45097dae | 3086 | sizeof(struct iovec) * iter->nr_segs); |
99bc4c38 PB |
3087 | } else { |
3088 | req->flags |= REQ_F_NEED_CLEANUP; | |
f67676d1 JA |
3089 | } |
3090 | } | |
3091 | ||
e8c2bc1f | 3092 | static inline int __io_alloc_async_data(struct io_kiocb *req) |
3d9932a8 | 3093 | { |
e8c2bc1f JA |
3094 | WARN_ON_ONCE(!io_op_defs[req->opcode].async_size); |
3095 | req->async_data = kmalloc(io_op_defs[req->opcode].async_size, GFP_KERNEL); | |
3096 | return req->async_data == NULL; | |
3d9932a8 XW |
3097 | } |
3098 | ||
e8c2bc1f | 3099 | static int io_alloc_async_data(struct io_kiocb *req) |
f67676d1 | 3100 | { |
e8c2bc1f | 3101 | if (!io_op_defs[req->opcode].needs_async_data) |
d3656344 | 3102 | return 0; |
3d9932a8 | 3103 | |
e8c2bc1f | 3104 | return __io_alloc_async_data(req); |
b7bb4f7d JA |
3105 | } |
3106 | ||
ff6165b2 JA |
3107 | static int io_setup_async_rw(struct io_kiocb *req, const struct iovec *iovec, |
3108 | const struct iovec *fast_iov, | |
227c0c96 | 3109 | struct iov_iter *iter, bool force) |
b7bb4f7d | 3110 | { |
e8c2bc1f | 3111 | if (!force && !io_op_defs[req->opcode].needs_async_data) |
74566df3 | 3112 | return 0; |
e8c2bc1f JA |
3113 | if (!req->async_data) { |
3114 | if (__io_alloc_async_data(req)) | |
5d204bcf | 3115 | return -ENOMEM; |
b7bb4f7d | 3116 | |
ff6165b2 | 3117 | io_req_map_rw(req, iovec, fast_iov, iter); |
5d204bcf | 3118 | } |
b7bb4f7d | 3119 | return 0; |
f67676d1 JA |
3120 | } |
3121 | ||
73debe68 | 3122 | static inline int io_rw_prep_async(struct io_kiocb *req, int rw) |
c3e330a4 | 3123 | { |
e8c2bc1f | 3124 | struct io_async_rw *iorw = req->async_data; |
f4bff104 | 3125 | struct iovec *iov = iorw->fast_iov; |
c3e330a4 PB |
3126 | ssize_t ret; |
3127 | ||
73debe68 | 3128 | ret = __io_import_iovec(rw, req, &iov, &iorw->iter, false); |
c3e330a4 PB |
3129 | if (unlikely(ret < 0)) |
3130 | return ret; | |
3131 | ||
ab0b196c PB |
3132 | iorw->bytes_done = 0; |
3133 | iorw->free_iovec = iov; | |
3134 | if (iov) | |
3135 | req->flags |= REQ_F_NEED_CLEANUP; | |
c3e330a4 PB |
3136 | return 0; |
3137 | } | |
3138 | ||
73debe68 | 3139 | static int io_read_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
f67676d1 JA |
3140 | { |
3141 | ssize_t ret; | |
3142 | ||
a88fc400 | 3143 | ret = io_prep_rw(req, sqe); |
3529d8c2 JA |
3144 | if (ret) |
3145 | return ret; | |
f67676d1 | 3146 | |
3529d8c2 JA |
3147 | if (unlikely(!(req->file->f_mode & FMODE_READ))) |
3148 | return -EBADF; | |
f67676d1 | 3149 | |
5f798bea | 3150 | /* either don't need iovec imported or already have it */ |
2d199895 | 3151 | if (!req->async_data) |
3529d8c2 | 3152 | return 0; |
73debe68 | 3153 | return io_rw_prep_async(req, READ); |
f67676d1 JA |
3154 | } |
3155 | ||
c1dd91d1 JA |
3156 | /* |
3157 | * This is our waitqueue callback handler, registered through lock_page_async() | |
3158 | * when we initially tried to do the IO with the iocb armed our waitqueue. | |
3159 | * This gets called when the page is unlocked, and we generally expect that to | |
3160 | * happen when the page IO is completed and the page is now uptodate. This will | |
3161 | * queue a task_work based retry of the operation, attempting to copy the data | |
3162 | * again. If the latter fails because the page was NOT uptodate, then we will | |
3163 | * do a thread based blocking retry of the operation. That's the unexpected | |
3164 | * slow path. | |
3165 | */ | |
bcf5a063 JA |
3166 | static int io_async_buf_func(struct wait_queue_entry *wait, unsigned mode, |
3167 | int sync, void *arg) | |
3168 | { | |
3169 | struct wait_page_queue *wpq; | |
3170 | struct io_kiocb *req = wait->private; | |
bcf5a063 | 3171 | struct wait_page_key *key = arg; |
bcf5a063 JA |
3172 | int ret; |
3173 | ||
3174 | wpq = container_of(wait, struct wait_page_queue, wait); | |
3175 | ||
cdc8fcb4 LT |
3176 | if (!wake_page_match(wpq, key)) |
3177 | return 0; | |
3178 | ||
c8d317aa | 3179 | req->rw.kiocb.ki_flags &= ~IOCB_WAITQ; |
bcf5a063 JA |
3180 | list_del_init(&wait->entry); |
3181 | ||
e7375122 | 3182 | init_task_work(&req->task_work, io_req_task_submit); |
6d816e08 JA |
3183 | percpu_ref_get(&req->ctx->refs); |
3184 | ||
bcf5a063 JA |
3185 | /* submit ref gets dropped, acquire a new one */ |
3186 | refcount_inc(&req->refs); | |
87c4311f | 3187 | ret = io_req_task_work_add(req, true); |
bcf5a063 | 3188 | if (unlikely(ret)) { |
c2c4c83c JA |
3189 | struct task_struct *tsk; |
3190 | ||
bcf5a063 | 3191 | /* queue just for cancelation */ |
e7375122 | 3192 | init_task_work(&req->task_work, io_req_task_cancel); |
bcf5a063 | 3193 | tsk = io_wq_get_task(req->ctx->io_wq); |
e7375122 | 3194 | task_work_add(tsk, &req->task_work, 0); |
c2c4c83c | 3195 | wake_up_process(tsk); |
bcf5a063 | 3196 | } |
bcf5a063 JA |
3197 | return 1; |
3198 | } | |
3199 | ||
c1dd91d1 JA |
3200 | /* |
3201 | * This controls whether a given IO request should be armed for async page | |
3202 | * based retry. If we return false here, the request is handed to the async | |
3203 | * worker threads for retry. If we're doing buffered reads on a regular file, | |
3204 | * we prepare a private wait_page_queue entry and retry the operation. This | |
3205 | * will either succeed because the page is now uptodate and unlocked, or it | |
3206 | * will register a callback when the page is unlocked at IO completion. Through | |
3207 | * that callback, io_uring uses task_work to setup a retry of the operation. | |
3208 | * That retry will attempt the buffered read again. The retry will generally | |
3209 | * succeed, or in rare cases where it fails, we then fall back to using the | |
3210 | * async worker threads for a blocking retry. | |
3211 | */ | |
227c0c96 | 3212 | static bool io_rw_should_retry(struct io_kiocb *req) |
f67676d1 | 3213 | { |
e8c2bc1f JA |
3214 | struct io_async_rw *rw = req->async_data; |
3215 | struct wait_page_queue *wait = &rw->wpq; | |
bcf5a063 | 3216 | struct kiocb *kiocb = &req->rw.kiocb; |
f67676d1 | 3217 | |
bcf5a063 JA |
3218 | /* never retry for NOWAIT, we just complete with -EAGAIN */ |
3219 | if (req->flags & REQ_F_NOWAIT) | |
3220 | return false; | |
f67676d1 | 3221 | |
227c0c96 | 3222 | /* Only for buffered IO */ |
3b2a4439 | 3223 | if (kiocb->ki_flags & (IOCB_DIRECT | IOCB_HIPRI)) |
bcf5a063 | 3224 | return false; |
3b2a4439 | 3225 | |
bcf5a063 JA |
3226 | /* |
3227 | * just use poll if we can, and don't attempt if the fs doesn't | |
3228 | * support callback based unlocks | |
3229 | */ | |
3230 | if (file_can_poll(req->file) || !(req->file->f_mode & FMODE_BUF_RASYNC)) | |
3231 | return false; | |
f67676d1 | 3232 | |
3b2a4439 JA |
3233 | wait->wait.func = io_async_buf_func; |
3234 | wait->wait.private = req; | |
3235 | wait->wait.flags = 0; | |
3236 | INIT_LIST_HEAD(&wait->wait.entry); | |
3237 | kiocb->ki_flags |= IOCB_WAITQ; | |
c8d317aa | 3238 | kiocb->ki_flags &= ~IOCB_NOWAIT; |
3b2a4439 | 3239 | kiocb->ki_waitq = wait; |
3b2a4439 | 3240 | return true; |
bcf5a063 JA |
3241 | } |
3242 | ||
3243 | static int io_iter_do_read(struct io_kiocb *req, struct iov_iter *iter) | |
3244 | { | |
3245 | if (req->file->f_op->read_iter) | |
3246 | return call_read_iter(req->file, &req->rw.kiocb, iter); | |
2dd2111d GH |
3247 | else if (req->file->f_op->read) |
3248 | return loop_rw_iter(READ, req->file, &req->rw.kiocb, iter); | |
3249 | else | |
3250 | return -EINVAL; | |
f67676d1 JA |
3251 | } |
3252 | ||
a1d7c393 JA |
3253 | static int io_read(struct io_kiocb *req, bool force_nonblock, |
3254 | struct io_comp_state *cs) | |
2b188cc1 JA |
3255 | { |
3256 | struct iovec inline_vecs[UIO_FASTIOV], *iovec = inline_vecs; | |
9adbd45d | 3257 | struct kiocb *kiocb = &req->rw.kiocb; |
ff6165b2 | 3258 | struct iov_iter __iter, *iter = &__iter; |
e8c2bc1f | 3259 | struct io_async_rw *rw = req->async_data; |
227c0c96 | 3260 | ssize_t io_size, ret, ret2; |
31b51510 | 3261 | size_t iov_count; |
f5cac8b1 | 3262 | bool no_async; |
ff6165b2 | 3263 | |
e8c2bc1f JA |
3264 | if (rw) |
3265 | iter = &rw->iter; | |
2b188cc1 | 3266 | |
ff6165b2 | 3267 | ret = io_import_iovec(READ, req, &iovec, iter, !force_nonblock); |
06b76d44 JA |
3268 | if (ret < 0) |
3269 | return ret; | |
eefdf30f | 3270 | iov_count = iov_iter_count(iter); |
fa15bafb PB |
3271 | io_size = ret; |
3272 | req->result = io_size; | |
227c0c96 | 3273 | ret = 0; |
2b188cc1 | 3274 | |
fd6c2e4c JA |
3275 | /* Ensure we clear previously set non-block flag */ |
3276 | if (!force_nonblock) | |
29de5f6a | 3277 | kiocb->ki_flags &= ~IOCB_NOWAIT; |
a88fc400 PB |
3278 | else |
3279 | kiocb->ki_flags |= IOCB_NOWAIT; | |
3280 | ||
fd6c2e4c | 3281 | |
24c74678 | 3282 | /* If the file doesn't support async, just async punt */ |
f5cac8b1 JA |
3283 | no_async = force_nonblock && !io_file_supports_async(req->file, READ); |
3284 | if (no_async) | |
f67676d1 | 3285 | goto copy_iov; |
9e645e11 | 3286 | |
0fef9483 | 3287 | ret = rw_verify_area(READ, req->file, io_kiocb_ppos(kiocb), iov_count); |
fa15bafb PB |
3288 | if (unlikely(ret)) |
3289 | goto out_free; | |
2b188cc1 | 3290 | |
227c0c96 | 3291 | ret = io_iter_do_read(req, iter); |
32960613 | 3292 | |
227c0c96 JA |
3293 | if (!ret) { |
3294 | goto done; | |
3295 | } else if (ret == -EIOCBQUEUED) { | |
3296 | ret = 0; | |
3297 | goto out_free; | |
3298 | } else if (ret == -EAGAIN) { | |
eefdf30f JA |
3299 | /* IOPOLL retry should happen for io-wq threads */ |
3300 | if (!force_nonblock && !(req->ctx->flags & IORING_SETUP_IOPOLL)) | |
f91daf56 | 3301 | goto done; |
355afaeb JA |
3302 | /* no retry on NONBLOCK marked file */ |
3303 | if (req->file->f_flags & O_NONBLOCK) | |
3304 | goto done; | |
84216315 JA |
3305 | /* some cases will consume bytes even on error returns */ |
3306 | iov_iter_revert(iter, iov_count - iov_iter_count(iter)); | |
f38c7e3a JA |
3307 | ret = 0; |
3308 | goto copy_iov; | |
227c0c96 | 3309 | } else if (ret < 0) { |
00d23d51 JA |
3310 | /* make sure -ERESTARTSYS -> -EINTR is done */ |
3311 | goto done; | |
227c0c96 JA |
3312 | } |
3313 | ||
3314 | /* read it all, or we did blocking attempt. no retry. */ | |
f91daf56 JA |
3315 | if (!iov_iter_count(iter) || !force_nonblock || |
3316 | (req->file->f_flags & O_NONBLOCK)) | |
227c0c96 JA |
3317 | goto done; |
3318 | ||
3319 | io_size -= ret; | |
3320 | copy_iov: | |
3321 | ret2 = io_setup_async_rw(req, iovec, inline_vecs, iter, true); | |
3322 | if (ret2) { | |
3323 | ret = ret2; | |
3324 | goto out_free; | |
3325 | } | |
f5cac8b1 JA |
3326 | if (no_async) |
3327 | return -EAGAIN; | |
e8c2bc1f | 3328 | rw = req->async_data; |
227c0c96 JA |
3329 | /* it's copied and will be cleaned with ->io */ |
3330 | iovec = NULL; | |
3331 | /* now use our persistent iterator, if we aren't already */ | |
e8c2bc1f | 3332 | iter = &rw->iter; |
227c0c96 | 3333 | retry: |
e8c2bc1f | 3334 | rw->bytes_done += ret; |
227c0c96 JA |
3335 | /* if we can retry, do so with the callbacks armed */ |
3336 | if (!io_rw_should_retry(req)) { | |
fa15bafb PB |
3337 | kiocb->ki_flags &= ~IOCB_WAITQ; |
3338 | return -EAGAIN; | |
2b188cc1 | 3339 | } |
227c0c96 JA |
3340 | |
3341 | /* | |
3342 | * Now retry read with the IOCB_WAITQ parts set in the iocb. If we | |
3343 | * get -EIOCBQUEUED, then we'll get a notification when the desired | |
3344 | * page gets unlocked. We can also get a partial read here, and if we | |
3345 | * do, then just retry at the new offset. | |
3346 | */ | |
3347 | ret = io_iter_do_read(req, iter); | |
3348 | if (ret == -EIOCBQUEUED) { | |
3349 | ret = 0; | |
3350 | goto out_free; | |
3351 | } else if (ret > 0 && ret < io_size) { | |
3352 | /* we got some bytes, but not all. retry. */ | |
3353 | goto retry; | |
3354 | } | |
3355 | done: | |
3356 | kiocb_done(kiocb, ret, cs); | |
3357 | ret = 0; | |
f67676d1 | 3358 | out_free: |
f261c168 | 3359 | /* it's reportedly faster than delegating the null check to kfree() */ |
252917c3 | 3360 | if (iovec) |
6f2cc166 | 3361 | kfree(iovec); |
2b188cc1 JA |
3362 | return ret; |
3363 | } | |
3364 | ||
73debe68 | 3365 | static int io_write_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
f67676d1 JA |
3366 | { |
3367 | ssize_t ret; | |
3368 | ||
a88fc400 | 3369 | ret = io_prep_rw(req, sqe); |
3529d8c2 JA |
3370 | if (ret) |
3371 | return ret; | |
f67676d1 | 3372 | |
3529d8c2 JA |
3373 | if (unlikely(!(req->file->f_mode & FMODE_WRITE))) |
3374 | return -EBADF; | |
f67676d1 | 3375 | |
5f798bea | 3376 | /* either don't need iovec imported or already have it */ |
2d199895 | 3377 | if (!req->async_data) |
3529d8c2 | 3378 | return 0; |
73debe68 | 3379 | return io_rw_prep_async(req, WRITE); |
f67676d1 JA |
3380 | } |
3381 | ||
a1d7c393 JA |
3382 | static int io_write(struct io_kiocb *req, bool force_nonblock, |
3383 | struct io_comp_state *cs) | |
2b188cc1 JA |
3384 | { |
3385 | struct iovec inline_vecs[UIO_FASTIOV], *iovec = inline_vecs; | |
9adbd45d | 3386 | struct kiocb *kiocb = &req->rw.kiocb; |
ff6165b2 | 3387 | struct iov_iter __iter, *iter = &__iter; |
e8c2bc1f | 3388 | struct io_async_rw *rw = req->async_data; |
31b51510 | 3389 | size_t iov_count; |
fa15bafb | 3390 | ssize_t ret, ret2, io_size; |
2b188cc1 | 3391 | |
e8c2bc1f JA |
3392 | if (rw) |
3393 | iter = &rw->iter; | |
ff6165b2 JA |
3394 | |
3395 | ret = io_import_iovec(WRITE, req, &iovec, iter, !force_nonblock); | |
06b76d44 JA |
3396 | if (ret < 0) |
3397 | return ret; | |
eefdf30f | 3398 | iov_count = iov_iter_count(iter); |
fa15bafb PB |
3399 | io_size = ret; |
3400 | req->result = io_size; | |
2b188cc1 | 3401 | |
fd6c2e4c JA |
3402 | /* Ensure we clear previously set non-block flag */ |
3403 | if (!force_nonblock) | |
a88fc400 PB |
3404 | kiocb->ki_flags &= ~IOCB_NOWAIT; |
3405 | else | |
3406 | kiocb->ki_flags |= IOCB_NOWAIT; | |
fd6c2e4c | 3407 | |
24c74678 | 3408 | /* If the file doesn't support async, just async punt */ |
af197f50 | 3409 | if (force_nonblock && !io_file_supports_async(req->file, WRITE)) |
f67676d1 | 3410 | goto copy_iov; |
31b51510 | 3411 | |
10d59345 JA |
3412 | /* file path doesn't support NOWAIT for non-direct_IO */ |
3413 | if (force_nonblock && !(kiocb->ki_flags & IOCB_DIRECT) && | |
3414 | (req->flags & REQ_F_ISREG)) | |
f67676d1 | 3415 | goto copy_iov; |
31b51510 | 3416 | |
0fef9483 | 3417 | ret = rw_verify_area(WRITE, req->file, io_kiocb_ppos(kiocb), iov_count); |
fa15bafb PB |
3418 | if (unlikely(ret)) |
3419 | goto out_free; | |
4ed734b0 | 3420 | |
fa15bafb PB |
3421 | /* |
3422 | * Open-code file_start_write here to grab freeze protection, | |
3423 | * which will be released by another thread in | |
3424 | * io_complete_rw(). Fool lockdep by telling it the lock got | |
3425 | * released so that it doesn't complain about the held lock when | |
3426 | * we return to userspace. | |
3427 | */ | |
3428 | if (req->flags & REQ_F_ISREG) { | |
3429 | __sb_start_write(file_inode(req->file)->i_sb, | |
3430 | SB_FREEZE_WRITE, true); | |
3431 | __sb_writers_release(file_inode(req->file)->i_sb, | |
3432 | SB_FREEZE_WRITE); | |
3433 | } | |
3434 | kiocb->ki_flags |= IOCB_WRITE; | |
4ed734b0 | 3435 | |
fa15bafb | 3436 | if (req->file->f_op->write_iter) |
ff6165b2 | 3437 | ret2 = call_write_iter(req->file, kiocb, iter); |
2dd2111d | 3438 | else if (req->file->f_op->write) |
ff6165b2 | 3439 | ret2 = loop_rw_iter(WRITE, req->file, kiocb, iter); |
2dd2111d GH |
3440 | else |
3441 | ret2 = -EINVAL; | |
4ed734b0 | 3442 | |
fa15bafb PB |
3443 | /* |
3444 | * Raw bdev writes will return -EOPNOTSUPP for IOCB_NOWAIT. Just | |
3445 | * retry them without IOCB_NOWAIT. | |
3446 | */ | |
3447 | if (ret2 == -EOPNOTSUPP && (kiocb->ki_flags & IOCB_NOWAIT)) | |
3448 | ret2 = -EAGAIN; | |
355afaeb JA |
3449 | /* no retry on NONBLOCK marked file */ |
3450 | if (ret2 == -EAGAIN && (req->file->f_flags & O_NONBLOCK)) | |
3451 | goto done; | |
fa15bafb | 3452 | if (!force_nonblock || ret2 != -EAGAIN) { |
eefdf30f JA |
3453 | /* IOPOLL retry should happen for io-wq threads */ |
3454 | if ((req->ctx->flags & IORING_SETUP_IOPOLL) && ret2 == -EAGAIN) | |
3455 | goto copy_iov; | |
355afaeb | 3456 | done: |
fa15bafb PB |
3457 | kiocb_done(kiocb, ret2, cs); |
3458 | } else { | |
f67676d1 | 3459 | copy_iov: |
84216315 JA |
3460 | /* some cases will consume bytes even on error returns */ |
3461 | iov_iter_revert(iter, iov_count - iov_iter_count(iter)); | |
227c0c96 | 3462 | ret = io_setup_async_rw(req, iovec, inline_vecs, iter, false); |
ff6165b2 JA |
3463 | if (!ret) |
3464 | return -EAGAIN; | |
2b188cc1 | 3465 | } |
31b51510 | 3466 | out_free: |
f261c168 | 3467 | /* it's reportedly faster than delegating the null check to kfree() */ |
252917c3 | 3468 | if (iovec) |
6f2cc166 | 3469 | kfree(iovec); |
2b188cc1 JA |
3470 | return ret; |
3471 | } | |
3472 | ||
f2a8d5c7 PB |
3473 | static int __io_splice_prep(struct io_kiocb *req, |
3474 | const struct io_uring_sqe *sqe) | |
7d67af2c PB |
3475 | { |
3476 | struct io_splice* sp = &req->splice; | |
3477 | unsigned int valid_flags = SPLICE_F_FD_IN_FIXED | SPLICE_F_ALL; | |
3478 | int ret; | |
3479 | ||
3232dd02 PB |
3480 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
3481 | return -EINVAL; | |
7d67af2c PB |
3482 | |
3483 | sp->file_in = NULL; | |
7d67af2c PB |
3484 | sp->len = READ_ONCE(sqe->len); |
3485 | sp->flags = READ_ONCE(sqe->splice_flags); | |
3486 | ||
3487 | if (unlikely(sp->flags & ~valid_flags)) | |
3488 | return -EINVAL; | |
3489 | ||
3490 | ret = io_file_get(NULL, req, READ_ONCE(sqe->splice_fd_in), &sp->file_in, | |
3491 | (sp->flags & SPLICE_F_FD_IN_FIXED)); | |
3492 | if (ret) | |
3493 | return ret; | |
3494 | req->flags |= REQ_F_NEED_CLEANUP; | |
3495 | ||
7cdaf587 XW |
3496 | if (!S_ISREG(file_inode(sp->file_in)->i_mode)) { |
3497 | /* | |
3498 | * Splice operation will be punted aync, and here need to | |
3499 | * modify io_wq_work.flags, so initialize io_wq_work firstly. | |
3500 | */ | |
3501 | io_req_init_async(req); | |
7d67af2c | 3502 | req->work.flags |= IO_WQ_WORK_UNBOUND; |
7cdaf587 | 3503 | } |
7d67af2c PB |
3504 | |
3505 | return 0; | |
3506 | } | |
3507 | ||
f2a8d5c7 PB |
3508 | static int io_tee_prep(struct io_kiocb *req, |
3509 | const struct io_uring_sqe *sqe) | |
3510 | { | |
3511 | if (READ_ONCE(sqe->splice_off_in) || READ_ONCE(sqe->off)) | |
3512 | return -EINVAL; | |
3513 | return __io_splice_prep(req, sqe); | |
3514 | } | |
3515 | ||
3516 | static int io_tee(struct io_kiocb *req, bool force_nonblock) | |
3517 | { | |
3518 | struct io_splice *sp = &req->splice; | |
3519 | struct file *in = sp->file_in; | |
3520 | struct file *out = sp->file_out; | |
3521 | unsigned int flags = sp->flags & ~SPLICE_F_FD_IN_FIXED; | |
3522 | long ret = 0; | |
3523 | ||
3524 | if (force_nonblock) | |
3525 | return -EAGAIN; | |
3526 | if (sp->len) | |
3527 | ret = do_tee(in, out, sp->len, flags); | |
3528 | ||
3529 | io_put_file(req, in, (sp->flags & SPLICE_F_FD_IN_FIXED)); | |
3530 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
3531 | ||
f2a8d5c7 PB |
3532 | if (ret != sp->len) |
3533 | req_set_fail_links(req); | |
e1e16097 | 3534 | io_req_complete(req, ret); |
f2a8d5c7 PB |
3535 | return 0; |
3536 | } | |
3537 | ||
3538 | static int io_splice_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) | |
3539 | { | |
3540 | struct io_splice* sp = &req->splice; | |
3541 | ||
3542 | sp->off_in = READ_ONCE(sqe->splice_off_in); | |
3543 | sp->off_out = READ_ONCE(sqe->off); | |
3544 | return __io_splice_prep(req, sqe); | |
3545 | } | |
3546 | ||
014db007 | 3547 | static int io_splice(struct io_kiocb *req, bool force_nonblock) |
7d67af2c PB |
3548 | { |
3549 | struct io_splice *sp = &req->splice; | |
3550 | struct file *in = sp->file_in; | |
3551 | struct file *out = sp->file_out; | |
3552 | unsigned int flags = sp->flags & ~SPLICE_F_FD_IN_FIXED; | |
3553 | loff_t *poff_in, *poff_out; | |
c9687426 | 3554 | long ret = 0; |
7d67af2c | 3555 | |
2fb3e822 PB |
3556 | if (force_nonblock) |
3557 | return -EAGAIN; | |
7d67af2c PB |
3558 | |
3559 | poff_in = (sp->off_in == -1) ? NULL : &sp->off_in; | |
3560 | poff_out = (sp->off_out == -1) ? NULL : &sp->off_out; | |
c9687426 | 3561 | |
948a7749 | 3562 | if (sp->len) |
c9687426 | 3563 | ret = do_splice(in, poff_in, out, poff_out, sp->len, flags); |
7d67af2c PB |
3564 | |
3565 | io_put_file(req, in, (sp->flags & SPLICE_F_FD_IN_FIXED)); | |
3566 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
3567 | ||
7d67af2c PB |
3568 | if (ret != sp->len) |
3569 | req_set_fail_links(req); | |
e1e16097 | 3570 | io_req_complete(req, ret); |
7d67af2c PB |
3571 | return 0; |
3572 | } | |
3573 | ||
2b188cc1 JA |
3574 | /* |
3575 | * IORING_OP_NOP just posts a completion event, nothing else. | |
3576 | */ | |
229a7b63 | 3577 | static int io_nop(struct io_kiocb *req, struct io_comp_state *cs) |
2b188cc1 JA |
3578 | { |
3579 | struct io_ring_ctx *ctx = req->ctx; | |
2b188cc1 | 3580 | |
def596e9 JA |
3581 | if (unlikely(ctx->flags & IORING_SETUP_IOPOLL)) |
3582 | return -EINVAL; | |
3583 | ||
229a7b63 | 3584 | __io_req_complete(req, 0, 0, cs); |
2b188cc1 JA |
3585 | return 0; |
3586 | } | |
3587 | ||
3529d8c2 | 3588 | static int io_prep_fsync(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
c992fe29 | 3589 | { |
6b06314c | 3590 | struct io_ring_ctx *ctx = req->ctx; |
c992fe29 | 3591 | |
09bb8394 JA |
3592 | if (!req->file) |
3593 | return -EBADF; | |
c992fe29 | 3594 | |
6b06314c | 3595 | if (unlikely(ctx->flags & IORING_SETUP_IOPOLL)) |
def596e9 | 3596 | return -EINVAL; |
edafccee | 3597 | if (unlikely(sqe->addr || sqe->ioprio || sqe->buf_index)) |
c992fe29 CH |
3598 | return -EINVAL; |
3599 | ||
8ed8d3c3 JA |
3600 | req->sync.flags = READ_ONCE(sqe->fsync_flags); |
3601 | if (unlikely(req->sync.flags & ~IORING_FSYNC_DATASYNC)) | |
3602 | return -EINVAL; | |
3603 | ||
3604 | req->sync.off = READ_ONCE(sqe->off); | |
3605 | req->sync.len = READ_ONCE(sqe->len); | |
c992fe29 CH |
3606 | return 0; |
3607 | } | |
3608 | ||
ac45abc0 | 3609 | static int io_fsync(struct io_kiocb *req, bool force_nonblock) |
8ed8d3c3 | 3610 | { |
8ed8d3c3 | 3611 | loff_t end = req->sync.off + req->sync.len; |
8ed8d3c3 JA |
3612 | int ret; |
3613 | ||
ac45abc0 PB |
3614 | /* fsync always requires a blocking context */ |
3615 | if (force_nonblock) | |
3616 | return -EAGAIN; | |
3617 | ||
9adbd45d | 3618 | ret = vfs_fsync_range(req->file, req->sync.off, |
8ed8d3c3 JA |
3619 | end > 0 ? end : LLONG_MAX, |
3620 | req->sync.flags & IORING_FSYNC_DATASYNC); | |
3621 | if (ret < 0) | |
3622 | req_set_fail_links(req); | |
e1e16097 | 3623 | io_req_complete(req, ret); |
c992fe29 CH |
3624 | return 0; |
3625 | } | |
3626 | ||
d63d1b5e JA |
3627 | static int io_fallocate_prep(struct io_kiocb *req, |
3628 | const struct io_uring_sqe *sqe) | |
3629 | { | |
3630 | if (sqe->ioprio || sqe->buf_index || sqe->rw_flags) | |
3631 | return -EINVAL; | |
3232dd02 PB |
3632 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
3633 | return -EINVAL; | |
d63d1b5e JA |
3634 | |
3635 | req->sync.off = READ_ONCE(sqe->off); | |
3636 | req->sync.len = READ_ONCE(sqe->addr); | |
3637 | req->sync.mode = READ_ONCE(sqe->len); | |
3638 | return 0; | |
3639 | } | |
3640 | ||
014db007 | 3641 | static int io_fallocate(struct io_kiocb *req, bool force_nonblock) |
5d17b4a4 | 3642 | { |
ac45abc0 PB |
3643 | int ret; |
3644 | ||
d63d1b5e | 3645 | /* fallocate always requiring blocking context */ |
ac45abc0 | 3646 | if (force_nonblock) |
5d17b4a4 | 3647 | return -EAGAIN; |
ac45abc0 PB |
3648 | ret = vfs_fallocate(req->file, req->sync.mode, req->sync.off, |
3649 | req->sync.len); | |
ac45abc0 PB |
3650 | if (ret < 0) |
3651 | req_set_fail_links(req); | |
e1e16097 | 3652 | io_req_complete(req, ret); |
5d17b4a4 JA |
3653 | return 0; |
3654 | } | |
3655 | ||
ec65fea5 | 3656 | static int __io_openat_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
b7bb4f7d | 3657 | { |
f8748881 | 3658 | const char __user *fname; |
15b71abe | 3659 | int ret; |
b7bb4f7d | 3660 | |
ec65fea5 | 3661 | if (unlikely(sqe->ioprio || sqe->buf_index)) |
15b71abe | 3662 | return -EINVAL; |
ec65fea5 | 3663 | if (unlikely(req->flags & REQ_F_FIXED_FILE)) |
cf3040ca | 3664 | return -EBADF; |
03b1230c | 3665 | |
ec65fea5 PB |
3666 | /* open.how should be already initialised */ |
3667 | if (!(req->open.how.flags & O_PATH) && force_o_largefile()) | |
08a1d26e | 3668 | req->open.how.flags |= O_LARGEFILE; |
3529d8c2 | 3669 | |
25e72d10 PB |
3670 | req->open.dfd = READ_ONCE(sqe->fd); |
3671 | fname = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
f8748881 | 3672 | req->open.filename = getname(fname); |
15b71abe JA |
3673 | if (IS_ERR(req->open.filename)) { |
3674 | ret = PTR_ERR(req->open.filename); | |
3675 | req->open.filename = NULL; | |
3676 | return ret; | |
3677 | } | |
4022e7af | 3678 | req->open.nofile = rlimit(RLIMIT_NOFILE); |
8fef80bf | 3679 | req->flags |= REQ_F_NEED_CLEANUP; |
15b71abe | 3680 | return 0; |
03b1230c JA |
3681 | } |
3682 | ||
ec65fea5 PB |
3683 | static int io_openat_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
3684 | { | |
3685 | u64 flags, mode; | |
3686 | ||
4eb8dded JA |
3687 | if (unlikely(req->ctx->flags & (IORING_SETUP_IOPOLL|IORING_SETUP_SQPOLL))) |
3688 | return -EINVAL; | |
ec65fea5 PB |
3689 | mode = READ_ONCE(sqe->len); |
3690 | flags = READ_ONCE(sqe->open_flags); | |
3691 | req->open.how = build_open_how(flags, mode); | |
3692 | return __io_openat_prep(req, sqe); | |
3693 | } | |
3694 | ||
cebdb986 | 3695 | static int io_openat2_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
aa1fa28f | 3696 | { |
cebdb986 | 3697 | struct open_how __user *how; |
cebdb986 | 3698 | size_t len; |
0fa03c62 JA |
3699 | int ret; |
3700 | ||
4eb8dded JA |
3701 | if (unlikely(req->ctx->flags & (IORING_SETUP_IOPOLL|IORING_SETUP_SQPOLL))) |
3702 | return -EINVAL; | |
cebdb986 JA |
3703 | how = u64_to_user_ptr(READ_ONCE(sqe->addr2)); |
3704 | len = READ_ONCE(sqe->len); | |
cebdb986 JA |
3705 | if (len < OPEN_HOW_SIZE_VER0) |
3706 | return -EINVAL; | |
3529d8c2 | 3707 | |
cebdb986 JA |
3708 | ret = copy_struct_from_user(&req->open.how, sizeof(req->open.how), how, |
3709 | len); | |
3710 | if (ret) | |
3711 | return ret; | |
3529d8c2 | 3712 | |
ec65fea5 | 3713 | return __io_openat_prep(req, sqe); |
cebdb986 JA |
3714 | } |
3715 | ||
014db007 | 3716 | static int io_openat2(struct io_kiocb *req, bool force_nonblock) |
15b71abe JA |
3717 | { |
3718 | struct open_flags op; | |
15b71abe JA |
3719 | struct file *file; |
3720 | int ret; | |
3721 | ||
f86cd20c | 3722 | if (force_nonblock) |
15b71abe | 3723 | return -EAGAIN; |
15b71abe | 3724 | |
cebdb986 | 3725 | ret = build_open_flags(&req->open.how, &op); |
15b71abe JA |
3726 | if (ret) |
3727 | goto err; | |
3728 | ||
4022e7af | 3729 | ret = __get_unused_fd_flags(req->open.how.flags, req->open.nofile); |
15b71abe JA |
3730 | if (ret < 0) |
3731 | goto err; | |
3732 | ||
3733 | file = do_filp_open(req->open.dfd, req->open.filename, &op); | |
3734 | if (IS_ERR(file)) { | |
3735 | put_unused_fd(ret); | |
3736 | ret = PTR_ERR(file); | |
3737 | } else { | |
3738 | fsnotify_open(file); | |
3739 | fd_install(ret, file); | |
3740 | } | |
3741 | err: | |
3742 | putname(req->open.filename); | |
8fef80bf | 3743 | req->flags &= ~REQ_F_NEED_CLEANUP; |
15b71abe JA |
3744 | if (ret < 0) |
3745 | req_set_fail_links(req); | |
e1e16097 | 3746 | io_req_complete(req, ret); |
15b71abe JA |
3747 | return 0; |
3748 | } | |
3749 | ||
014db007 | 3750 | static int io_openat(struct io_kiocb *req, bool force_nonblock) |
cebdb986 | 3751 | { |
014db007 | 3752 | return io_openat2(req, force_nonblock); |
cebdb986 JA |
3753 | } |
3754 | ||
067524e9 JA |
3755 | static int io_remove_buffers_prep(struct io_kiocb *req, |
3756 | const struct io_uring_sqe *sqe) | |
3757 | { | |
3758 | struct io_provide_buf *p = &req->pbuf; | |
3759 | u64 tmp; | |
3760 | ||
3761 | if (sqe->ioprio || sqe->rw_flags || sqe->addr || sqe->len || sqe->off) | |
3762 | return -EINVAL; | |
3763 | ||
3764 | tmp = READ_ONCE(sqe->fd); | |
3765 | if (!tmp || tmp > USHRT_MAX) | |
3766 | return -EINVAL; | |
3767 | ||
3768 | memset(p, 0, sizeof(*p)); | |
3769 | p->nbufs = tmp; | |
3770 | p->bgid = READ_ONCE(sqe->buf_group); | |
3771 | return 0; | |
3772 | } | |
3773 | ||
3774 | static int __io_remove_buffers(struct io_ring_ctx *ctx, struct io_buffer *buf, | |
3775 | int bgid, unsigned nbufs) | |
3776 | { | |
3777 | unsigned i = 0; | |
3778 | ||
3779 | /* shouldn't happen */ | |
3780 | if (!nbufs) | |
3781 | return 0; | |
3782 | ||
3783 | /* the head kbuf is the list itself */ | |
3784 | while (!list_empty(&buf->list)) { | |
3785 | struct io_buffer *nxt; | |
3786 | ||
3787 | nxt = list_first_entry(&buf->list, struct io_buffer, list); | |
3788 | list_del(&nxt->list); | |
3789 | kfree(nxt); | |
3790 | if (++i == nbufs) | |
3791 | return i; | |
3792 | } | |
3793 | i++; | |
3794 | kfree(buf); | |
3795 | idr_remove(&ctx->io_buffer_idr, bgid); | |
3796 | ||
3797 | return i; | |
3798 | } | |
3799 | ||
229a7b63 JA |
3800 | static int io_remove_buffers(struct io_kiocb *req, bool force_nonblock, |
3801 | struct io_comp_state *cs) | |
067524e9 JA |
3802 | { |
3803 | struct io_provide_buf *p = &req->pbuf; | |
3804 | struct io_ring_ctx *ctx = req->ctx; | |
3805 | struct io_buffer *head; | |
3806 | int ret = 0; | |
3807 | ||
3808 | io_ring_submit_lock(ctx, !force_nonblock); | |
3809 | ||
3810 | lockdep_assert_held(&ctx->uring_lock); | |
3811 | ||
3812 | ret = -ENOENT; | |
3813 | head = idr_find(&ctx->io_buffer_idr, p->bgid); | |
3814 | if (head) | |
3815 | ret = __io_remove_buffers(ctx, head, p->bgid, p->nbufs); | |
3816 | ||
3817 | io_ring_submit_lock(ctx, !force_nonblock); | |
3818 | if (ret < 0) | |
3819 | req_set_fail_links(req); | |
229a7b63 | 3820 | __io_req_complete(req, ret, 0, cs); |
067524e9 JA |
3821 | return 0; |
3822 | } | |
3823 | ||
ddf0322d JA |
3824 | static int io_provide_buffers_prep(struct io_kiocb *req, |
3825 | const struct io_uring_sqe *sqe) | |
3826 | { | |
3827 | struct io_provide_buf *p = &req->pbuf; | |
3828 | u64 tmp; | |
3829 | ||
3830 | if (sqe->ioprio || sqe->rw_flags) | |
3831 | return -EINVAL; | |
3832 | ||
3833 | tmp = READ_ONCE(sqe->fd); | |
3834 | if (!tmp || tmp > USHRT_MAX) | |
3835 | return -E2BIG; | |
3836 | p->nbufs = tmp; | |
3837 | p->addr = READ_ONCE(sqe->addr); | |
3838 | p->len = READ_ONCE(sqe->len); | |
3839 | ||
efe68c1c | 3840 | if (!access_ok(u64_to_user_ptr(p->addr), (p->len * p->nbufs))) |
ddf0322d JA |
3841 | return -EFAULT; |
3842 | ||
3843 | p->bgid = READ_ONCE(sqe->buf_group); | |
3844 | tmp = READ_ONCE(sqe->off); | |
3845 | if (tmp > USHRT_MAX) | |
3846 | return -E2BIG; | |
3847 | p->bid = tmp; | |
3848 | return 0; | |
3849 | } | |
3850 | ||
3851 | static int io_add_buffers(struct io_provide_buf *pbuf, struct io_buffer **head) | |
3852 | { | |
3853 | struct io_buffer *buf; | |
3854 | u64 addr = pbuf->addr; | |
3855 | int i, bid = pbuf->bid; | |
3856 | ||
3857 | for (i = 0; i < pbuf->nbufs; i++) { | |
3858 | buf = kmalloc(sizeof(*buf), GFP_KERNEL); | |
3859 | if (!buf) | |
3860 | break; | |
3861 | ||
3862 | buf->addr = addr; | |
3863 | buf->len = pbuf->len; | |
3864 | buf->bid = bid; | |
3865 | addr += pbuf->len; | |
3866 | bid++; | |
3867 | if (!*head) { | |
3868 | INIT_LIST_HEAD(&buf->list); | |
3869 | *head = buf; | |
3870 | } else { | |
3871 | list_add_tail(&buf->list, &(*head)->list); | |
3872 | } | |
3873 | } | |
3874 | ||
3875 | return i ? i : -ENOMEM; | |
3876 | } | |
3877 | ||
229a7b63 JA |
3878 | static int io_provide_buffers(struct io_kiocb *req, bool force_nonblock, |
3879 | struct io_comp_state *cs) | |
ddf0322d JA |
3880 | { |
3881 | struct io_provide_buf *p = &req->pbuf; | |
3882 | struct io_ring_ctx *ctx = req->ctx; | |
3883 | struct io_buffer *head, *list; | |
3884 | int ret = 0; | |
3885 | ||
3886 | io_ring_submit_lock(ctx, !force_nonblock); | |
3887 | ||
3888 | lockdep_assert_held(&ctx->uring_lock); | |
3889 | ||
3890 | list = head = idr_find(&ctx->io_buffer_idr, p->bgid); | |
3891 | ||
3892 | ret = io_add_buffers(p, &head); | |
3893 | if (ret < 0) | |
3894 | goto out; | |
3895 | ||
3896 | if (!list) { | |
3897 | ret = idr_alloc(&ctx->io_buffer_idr, head, p->bgid, p->bgid + 1, | |
3898 | GFP_KERNEL); | |
3899 | if (ret < 0) { | |
067524e9 | 3900 | __io_remove_buffers(ctx, head, p->bgid, -1U); |
ddf0322d JA |
3901 | goto out; |
3902 | } | |
3903 | } | |
3904 | out: | |
3905 | io_ring_submit_unlock(ctx, !force_nonblock); | |
3906 | if (ret < 0) | |
3907 | req_set_fail_links(req); | |
229a7b63 | 3908 | __io_req_complete(req, ret, 0, cs); |
ddf0322d | 3909 | return 0; |
cebdb986 JA |
3910 | } |
3911 | ||
3e4827b0 JA |
3912 | static int io_epoll_ctl_prep(struct io_kiocb *req, |
3913 | const struct io_uring_sqe *sqe) | |
3914 | { | |
3915 | #if defined(CONFIG_EPOLL) | |
3916 | if (sqe->ioprio || sqe->buf_index) | |
3917 | return -EINVAL; | |
6ca56f84 | 3918 | if (unlikely(req->ctx->flags & (IORING_SETUP_IOPOLL | IORING_SETUP_SQPOLL))) |
3232dd02 | 3919 | return -EINVAL; |
3e4827b0 JA |
3920 | |
3921 | req->epoll.epfd = READ_ONCE(sqe->fd); | |
3922 | req->epoll.op = READ_ONCE(sqe->len); | |
3923 | req->epoll.fd = READ_ONCE(sqe->off); | |
3924 | ||
3925 | if (ep_op_has_event(req->epoll.op)) { | |
3926 | struct epoll_event __user *ev; | |
3927 | ||
3928 | ev = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
3929 | if (copy_from_user(&req->epoll.event, ev, sizeof(*ev))) | |
3930 | return -EFAULT; | |
3931 | } | |
3932 | ||
3933 | return 0; | |
3934 | #else | |
3935 | return -EOPNOTSUPP; | |
3936 | #endif | |
3937 | } | |
3938 | ||
229a7b63 JA |
3939 | static int io_epoll_ctl(struct io_kiocb *req, bool force_nonblock, |
3940 | struct io_comp_state *cs) | |
3e4827b0 JA |
3941 | { |
3942 | #if defined(CONFIG_EPOLL) | |
3943 | struct io_epoll *ie = &req->epoll; | |
3944 | int ret; | |
3945 | ||
3946 | ret = do_epoll_ctl(ie->epfd, ie->op, ie->fd, &ie->event, force_nonblock); | |
3947 | if (force_nonblock && ret == -EAGAIN) | |
3948 | return -EAGAIN; | |
3949 | ||
3950 | if (ret < 0) | |
3951 | req_set_fail_links(req); | |
229a7b63 | 3952 | __io_req_complete(req, ret, 0, cs); |
3e4827b0 JA |
3953 | return 0; |
3954 | #else | |
3955 | return -EOPNOTSUPP; | |
3956 | #endif | |
3957 | } | |
3958 | ||
c1ca757b JA |
3959 | static int io_madvise_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
3960 | { | |
3961 | #if defined(CONFIG_ADVISE_SYSCALLS) && defined(CONFIG_MMU) | |
3962 | if (sqe->ioprio || sqe->buf_index || sqe->off) | |
3963 | return -EINVAL; | |
3232dd02 PB |
3964 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
3965 | return -EINVAL; | |
c1ca757b JA |
3966 | |
3967 | req->madvise.addr = READ_ONCE(sqe->addr); | |
3968 | req->madvise.len = READ_ONCE(sqe->len); | |
3969 | req->madvise.advice = READ_ONCE(sqe->fadvise_advice); | |
3970 | return 0; | |
3971 | #else | |
3972 | return -EOPNOTSUPP; | |
3973 | #endif | |
3974 | } | |
3975 | ||
014db007 | 3976 | static int io_madvise(struct io_kiocb *req, bool force_nonblock) |
c1ca757b JA |
3977 | { |
3978 | #if defined(CONFIG_ADVISE_SYSCALLS) && defined(CONFIG_MMU) | |
3979 | struct io_madvise *ma = &req->madvise; | |
3980 | int ret; | |
3981 | ||
3982 | if (force_nonblock) | |
3983 | return -EAGAIN; | |
3984 | ||
3985 | ret = do_madvise(ma->addr, ma->len, ma->advice); | |
3986 | if (ret < 0) | |
3987 | req_set_fail_links(req); | |
e1e16097 | 3988 | io_req_complete(req, ret); |
c1ca757b JA |
3989 | return 0; |
3990 | #else | |
3991 | return -EOPNOTSUPP; | |
3992 | #endif | |
3993 | } | |
3994 | ||
4840e418 JA |
3995 | static int io_fadvise_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
3996 | { | |
3997 | if (sqe->ioprio || sqe->buf_index || sqe->addr) | |
3998 | return -EINVAL; | |
3232dd02 PB |
3999 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4000 | return -EINVAL; | |
4840e418 JA |
4001 | |
4002 | req->fadvise.offset = READ_ONCE(sqe->off); | |
4003 | req->fadvise.len = READ_ONCE(sqe->len); | |
4004 | req->fadvise.advice = READ_ONCE(sqe->fadvise_advice); | |
4005 | return 0; | |
4006 | } | |
4007 | ||
014db007 | 4008 | static int io_fadvise(struct io_kiocb *req, bool force_nonblock) |
4840e418 JA |
4009 | { |
4010 | struct io_fadvise *fa = &req->fadvise; | |
4011 | int ret; | |
4012 | ||
3e69426d JA |
4013 | if (force_nonblock) { |
4014 | switch (fa->advice) { | |
4015 | case POSIX_FADV_NORMAL: | |
4016 | case POSIX_FADV_RANDOM: | |
4017 | case POSIX_FADV_SEQUENTIAL: | |
4018 | break; | |
4019 | default: | |
4020 | return -EAGAIN; | |
4021 | } | |
4022 | } | |
4840e418 JA |
4023 | |
4024 | ret = vfs_fadvise(req->file, fa->offset, fa->len, fa->advice); | |
4025 | if (ret < 0) | |
4026 | req_set_fail_links(req); | |
e1e16097 | 4027 | io_req_complete(req, ret); |
4840e418 JA |
4028 | return 0; |
4029 | } | |
4030 | ||
eddc7ef5 JA |
4031 | static int io_statx_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
4032 | { | |
6ca56f84 | 4033 | if (unlikely(req->ctx->flags & (IORING_SETUP_IOPOLL | IORING_SETUP_SQPOLL))) |
3232dd02 | 4034 | return -EINVAL; |
eddc7ef5 JA |
4035 | if (sqe->ioprio || sqe->buf_index) |
4036 | return -EINVAL; | |
9c280f90 | 4037 | if (req->flags & REQ_F_FIXED_FILE) |
cf3040ca | 4038 | return -EBADF; |
eddc7ef5 | 4039 | |
1d9e1288 BM |
4040 | req->statx.dfd = READ_ONCE(sqe->fd); |
4041 | req->statx.mask = READ_ONCE(sqe->len); | |
e62753e4 | 4042 | req->statx.filename = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
1d9e1288 BM |
4043 | req->statx.buffer = u64_to_user_ptr(READ_ONCE(sqe->addr2)); |
4044 | req->statx.flags = READ_ONCE(sqe->statx_flags); | |
eddc7ef5 JA |
4045 | |
4046 | return 0; | |
4047 | } | |
4048 | ||
014db007 | 4049 | static int io_statx(struct io_kiocb *req, bool force_nonblock) |
eddc7ef5 | 4050 | { |
1d9e1288 | 4051 | struct io_statx *ctx = &req->statx; |
eddc7ef5 JA |
4052 | int ret; |
4053 | ||
5b0bbee4 JA |
4054 | if (force_nonblock) { |
4055 | /* only need file table for an actual valid fd */ | |
4056 | if (ctx->dfd == -1 || ctx->dfd == AT_FDCWD) | |
4057 | req->flags |= REQ_F_NO_FILE_TABLE; | |
eddc7ef5 | 4058 | return -EAGAIN; |
5b0bbee4 | 4059 | } |
eddc7ef5 | 4060 | |
e62753e4 BM |
4061 | ret = do_statx(ctx->dfd, ctx->filename, ctx->flags, ctx->mask, |
4062 | ctx->buffer); | |
eddc7ef5 | 4063 | |
eddc7ef5 JA |
4064 | if (ret < 0) |
4065 | req_set_fail_links(req); | |
e1e16097 | 4066 | io_req_complete(req, ret); |
eddc7ef5 JA |
4067 | return 0; |
4068 | } | |
4069 | ||
b5dba59e JA |
4070 | static int io_close_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
4071 | { | |
4072 | /* | |
4073 | * If we queue this for async, it must not be cancellable. That would | |
7cdaf587 XW |
4074 | * leave the 'file' in an undeterminate state, and here need to modify |
4075 | * io_wq_work.flags, so initialize io_wq_work firstly. | |
b5dba59e | 4076 | */ |
7cdaf587 | 4077 | io_req_init_async(req); |
b5dba59e JA |
4078 | req->work.flags |= IO_WQ_WORK_NO_CANCEL; |
4079 | ||
3232dd02 PB |
4080 | if (unlikely(req->ctx->flags & (IORING_SETUP_IOPOLL|IORING_SETUP_SQPOLL))) |
4081 | return -EINVAL; | |
b5dba59e JA |
4082 | if (sqe->ioprio || sqe->off || sqe->addr || sqe->len || |
4083 | sqe->rw_flags || sqe->buf_index) | |
4084 | return -EINVAL; | |
9c280f90 | 4085 | if (req->flags & REQ_F_FIXED_FILE) |
cf3040ca | 4086 | return -EBADF; |
b5dba59e JA |
4087 | |
4088 | req->close.fd = READ_ONCE(sqe->fd); | |
0f212204 | 4089 | if ((req->file && req->file->f_op == &io_uring_fops)) |
fd2206e4 | 4090 | return -EBADF; |
b5dba59e | 4091 | |
3af73b28 | 4092 | req->close.put_file = NULL; |
b5dba59e | 4093 | return 0; |
b5dba59e JA |
4094 | } |
4095 | ||
229a7b63 JA |
4096 | static int io_close(struct io_kiocb *req, bool force_nonblock, |
4097 | struct io_comp_state *cs) | |
b5dba59e | 4098 | { |
3af73b28 | 4099 | struct io_close *close = &req->close; |
b5dba59e JA |
4100 | int ret; |
4101 | ||
3af73b28 PB |
4102 | /* might be already done during nonblock submission */ |
4103 | if (!close->put_file) { | |
4104 | ret = __close_fd_get_file(close->fd, &close->put_file); | |
4105 | if (ret < 0) | |
4106 | return (ret == -ENOENT) ? -EBADF : ret; | |
4107 | } | |
b5dba59e JA |
4108 | |
4109 | /* if the file has a flush method, be safe and punt to async */ | |
3af73b28 | 4110 | if (close->put_file->f_op->flush && force_nonblock) { |
24c74678 PB |
4111 | /* was never set, but play safe */ |
4112 | req->flags &= ~REQ_F_NOWAIT; | |
0bf0eefd | 4113 | /* avoid grabbing files - we don't need the files */ |
24c74678 | 4114 | req->flags |= REQ_F_NO_FILE_TABLE; |
0bf0eefd | 4115 | return -EAGAIN; |
a2100672 | 4116 | } |
b5dba59e | 4117 | |
3af73b28 PB |
4118 | /* No ->flush() or already async, safely close from here */ |
4119 | ret = filp_close(close->put_file, req->work.files); | |
4120 | if (ret < 0) | |
4121 | req_set_fail_links(req); | |
3af73b28 PB |
4122 | fput(close->put_file); |
4123 | close->put_file = NULL; | |
229a7b63 | 4124 | __io_req_complete(req, ret, 0, cs); |
1a417f4e | 4125 | return 0; |
b5dba59e JA |
4126 | } |
4127 | ||
3529d8c2 | 4128 | static int io_prep_sfr(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
5d17b4a4 JA |
4129 | { |
4130 | struct io_ring_ctx *ctx = req->ctx; | |
5d17b4a4 JA |
4131 | |
4132 | if (!req->file) | |
4133 | return -EBADF; | |
5d17b4a4 JA |
4134 | |
4135 | if (unlikely(ctx->flags & IORING_SETUP_IOPOLL)) | |
4136 | return -EINVAL; | |
4137 | if (unlikely(sqe->addr || sqe->ioprio || sqe->buf_index)) | |
4138 | return -EINVAL; | |
4139 | ||
8ed8d3c3 JA |
4140 | req->sync.off = READ_ONCE(sqe->off); |
4141 | req->sync.len = READ_ONCE(sqe->len); | |
4142 | req->sync.flags = READ_ONCE(sqe->sync_range_flags); | |
8ed8d3c3 JA |
4143 | return 0; |
4144 | } | |
4145 | ||
ac45abc0 | 4146 | static int io_sync_file_range(struct io_kiocb *req, bool force_nonblock) |
8ed8d3c3 | 4147 | { |
8ed8d3c3 JA |
4148 | int ret; |
4149 | ||
ac45abc0 PB |
4150 | /* sync_file_range always requires a blocking context */ |
4151 | if (force_nonblock) | |
4152 | return -EAGAIN; | |
4153 | ||
9adbd45d | 4154 | ret = sync_file_range(req->file, req->sync.off, req->sync.len, |
8ed8d3c3 JA |
4155 | req->sync.flags); |
4156 | if (ret < 0) | |
4157 | req_set_fail_links(req); | |
e1e16097 | 4158 | io_req_complete(req, ret); |
5d17b4a4 JA |
4159 | return 0; |
4160 | } | |
4161 | ||
469956e8 | 4162 | #if defined(CONFIG_NET) |
02d27d89 PB |
4163 | static int io_setup_async_msg(struct io_kiocb *req, |
4164 | struct io_async_msghdr *kmsg) | |
4165 | { | |
e8c2bc1f JA |
4166 | struct io_async_msghdr *async_msg = req->async_data; |
4167 | ||
4168 | if (async_msg) | |
02d27d89 | 4169 | return -EAGAIN; |
e8c2bc1f | 4170 | if (io_alloc_async_data(req)) { |
02d27d89 PB |
4171 | if (kmsg->iov != kmsg->fast_iov) |
4172 | kfree(kmsg->iov); | |
4173 | return -ENOMEM; | |
4174 | } | |
e8c2bc1f | 4175 | async_msg = req->async_data; |
02d27d89 | 4176 | req->flags |= REQ_F_NEED_CLEANUP; |
e8c2bc1f | 4177 | memcpy(async_msg, kmsg, sizeof(*kmsg)); |
02d27d89 PB |
4178 | return -EAGAIN; |
4179 | } | |
4180 | ||
2ae523ed PB |
4181 | static int io_sendmsg_copy_hdr(struct io_kiocb *req, |
4182 | struct io_async_msghdr *iomsg) | |
4183 | { | |
4184 | iomsg->iov = iomsg->fast_iov; | |
4185 | iomsg->msg.msg_name = &iomsg->addr; | |
4186 | return sendmsg_copy_msghdr(&iomsg->msg, req->sr_msg.umsg, | |
4187 | req->sr_msg.msg_flags, &iomsg->iov); | |
4188 | } | |
4189 | ||
3529d8c2 | 4190 | static int io_sendmsg_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
03b1230c | 4191 | { |
e8c2bc1f | 4192 | struct io_async_msghdr *async_msg = req->async_data; |
e47293fd | 4193 | struct io_sr_msg *sr = &req->sr_msg; |
99bc4c38 | 4194 | int ret; |
03b1230c | 4195 | |
d2b6f48b PB |
4196 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4197 | return -EINVAL; | |
4198 | ||
e47293fd | 4199 | sr->msg_flags = READ_ONCE(sqe->msg_flags); |
270a5940 | 4200 | sr->umsg = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
fddaface | 4201 | sr->len = READ_ONCE(sqe->len); |
3529d8c2 | 4202 | |
d8768362 JA |
4203 | #ifdef CONFIG_COMPAT |
4204 | if (req->ctx->compat) | |
4205 | sr->msg_flags |= MSG_CMSG_COMPAT; | |
4206 | #endif | |
4207 | ||
e8c2bc1f | 4208 | if (!async_msg || !io_op_defs[req->opcode].needs_async_data) |
3529d8c2 | 4209 | return 0; |
e8c2bc1f | 4210 | ret = io_sendmsg_copy_hdr(req, async_msg); |
99bc4c38 PB |
4211 | if (!ret) |
4212 | req->flags |= REQ_F_NEED_CLEANUP; | |
4213 | return ret; | |
03b1230c JA |
4214 | } |
4215 | ||
229a7b63 JA |
4216 | static int io_sendmsg(struct io_kiocb *req, bool force_nonblock, |
4217 | struct io_comp_state *cs) | |
aa1fa28f | 4218 | { |
6b754c8b | 4219 | struct io_async_msghdr iomsg, *kmsg; |
0fa03c62 | 4220 | struct socket *sock; |
7a7cacba | 4221 | unsigned flags; |
0fa03c62 JA |
4222 | int ret; |
4223 | ||
0fa03c62 | 4224 | sock = sock_from_file(req->file, &ret); |
7a7cacba PB |
4225 | if (unlikely(!sock)) |
4226 | return ret; | |
3529d8c2 | 4227 | |
e8c2bc1f JA |
4228 | if (req->async_data) { |
4229 | kmsg = req->async_data; | |
4230 | kmsg->msg.msg_name = &kmsg->addr; | |
7a7cacba PB |
4231 | /* if iov is set, it's allocated already */ |
4232 | if (!kmsg->iov) | |
4233 | kmsg->iov = kmsg->fast_iov; | |
4234 | kmsg->msg.msg_iter.iov = kmsg->iov; | |
4235 | } else { | |
4236 | ret = io_sendmsg_copy_hdr(req, &iomsg); | |
4237 | if (ret) | |
4238 | return ret; | |
4239 | kmsg = &iomsg; | |
0fa03c62 | 4240 | } |
0fa03c62 | 4241 | |
7a7cacba PB |
4242 | flags = req->sr_msg.msg_flags; |
4243 | if (flags & MSG_DONTWAIT) | |
4244 | req->flags |= REQ_F_NOWAIT; | |
4245 | else if (force_nonblock) | |
4246 | flags |= MSG_DONTWAIT; | |
e47293fd | 4247 | |
7a7cacba PB |
4248 | ret = __sys_sendmsg_sock(sock, &kmsg->msg, flags); |
4249 | if (force_nonblock && ret == -EAGAIN) | |
4250 | return io_setup_async_msg(req, kmsg); | |
4251 | if (ret == -ERESTARTSYS) | |
4252 | ret = -EINTR; | |
0fa03c62 | 4253 | |
6b754c8b | 4254 | if (kmsg->iov != kmsg->fast_iov) |
0b416c3e | 4255 | kfree(kmsg->iov); |
99bc4c38 | 4256 | req->flags &= ~REQ_F_NEED_CLEANUP; |
4e88d6e7 JA |
4257 | if (ret < 0) |
4258 | req_set_fail_links(req); | |
229a7b63 | 4259 | __io_req_complete(req, ret, 0, cs); |
5d17b4a4 | 4260 | return 0; |
03b1230c | 4261 | } |
aa1fa28f | 4262 | |
229a7b63 JA |
4263 | static int io_send(struct io_kiocb *req, bool force_nonblock, |
4264 | struct io_comp_state *cs) | |
fddaface | 4265 | { |
7a7cacba PB |
4266 | struct io_sr_msg *sr = &req->sr_msg; |
4267 | struct msghdr msg; | |
4268 | struct iovec iov; | |
fddaface | 4269 | struct socket *sock; |
7a7cacba | 4270 | unsigned flags; |
fddaface JA |
4271 | int ret; |
4272 | ||
fddaface | 4273 | sock = sock_from_file(req->file, &ret); |
7a7cacba PB |
4274 | if (unlikely(!sock)) |
4275 | return ret; | |
fddaface | 4276 | |
7a7cacba PB |
4277 | ret = import_single_range(WRITE, sr->buf, sr->len, &iov, &msg.msg_iter); |
4278 | if (unlikely(ret)) | |
14db8411 | 4279 | return ret; |
fddaface | 4280 | |
7a7cacba PB |
4281 | msg.msg_name = NULL; |
4282 | msg.msg_control = NULL; | |
4283 | msg.msg_controllen = 0; | |
4284 | msg.msg_namelen = 0; | |
fddaface | 4285 | |
7a7cacba PB |
4286 | flags = req->sr_msg.msg_flags; |
4287 | if (flags & MSG_DONTWAIT) | |
4288 | req->flags |= REQ_F_NOWAIT; | |
4289 | else if (force_nonblock) | |
4290 | flags |= MSG_DONTWAIT; | |
fddaface | 4291 | |
7a7cacba PB |
4292 | msg.msg_flags = flags; |
4293 | ret = sock_sendmsg(sock, &msg); | |
4294 | if (force_nonblock && ret == -EAGAIN) | |
4295 | return -EAGAIN; | |
4296 | if (ret == -ERESTARTSYS) | |
4297 | ret = -EINTR; | |
fddaface | 4298 | |
fddaface JA |
4299 | if (ret < 0) |
4300 | req_set_fail_links(req); | |
229a7b63 | 4301 | __io_req_complete(req, ret, 0, cs); |
fddaface | 4302 | return 0; |
fddaface JA |
4303 | } |
4304 | ||
1400e697 PB |
4305 | static int __io_recvmsg_copy_hdr(struct io_kiocb *req, |
4306 | struct io_async_msghdr *iomsg) | |
52de1fe1 JA |
4307 | { |
4308 | struct io_sr_msg *sr = &req->sr_msg; | |
4309 | struct iovec __user *uiov; | |
4310 | size_t iov_len; | |
4311 | int ret; | |
4312 | ||
1400e697 PB |
4313 | ret = __copy_msghdr_from_user(&iomsg->msg, sr->umsg, |
4314 | &iomsg->uaddr, &uiov, &iov_len); | |
52de1fe1 JA |
4315 | if (ret) |
4316 | return ret; | |
4317 | ||
4318 | if (req->flags & REQ_F_BUFFER_SELECT) { | |
4319 | if (iov_len > 1) | |
4320 | return -EINVAL; | |
1400e697 | 4321 | if (copy_from_user(iomsg->iov, uiov, sizeof(*uiov))) |
52de1fe1 | 4322 | return -EFAULT; |
1400e697 PB |
4323 | sr->len = iomsg->iov[0].iov_len; |
4324 | iov_iter_init(&iomsg->msg.msg_iter, READ, iomsg->iov, 1, | |
52de1fe1 | 4325 | sr->len); |
1400e697 | 4326 | iomsg->iov = NULL; |
52de1fe1 JA |
4327 | } else { |
4328 | ret = import_iovec(READ, uiov, iov_len, UIO_FASTIOV, | |
1400e697 | 4329 | &iomsg->iov, &iomsg->msg.msg_iter); |
52de1fe1 JA |
4330 | if (ret > 0) |
4331 | ret = 0; | |
4332 | } | |
4333 | ||
4334 | return ret; | |
4335 | } | |
4336 | ||
4337 | #ifdef CONFIG_COMPAT | |
4338 | static int __io_compat_recvmsg_copy_hdr(struct io_kiocb *req, | |
1400e697 | 4339 | struct io_async_msghdr *iomsg) |
52de1fe1 JA |
4340 | { |
4341 | struct compat_msghdr __user *msg_compat; | |
4342 | struct io_sr_msg *sr = &req->sr_msg; | |
4343 | struct compat_iovec __user *uiov; | |
4344 | compat_uptr_t ptr; | |
4345 | compat_size_t len; | |
4346 | int ret; | |
4347 | ||
270a5940 | 4348 | msg_compat = (struct compat_msghdr __user *) sr->umsg; |
1400e697 | 4349 | ret = __get_compat_msghdr(&iomsg->msg, msg_compat, &iomsg->uaddr, |
52de1fe1 JA |
4350 | &ptr, &len); |
4351 | if (ret) | |
4352 | return ret; | |
4353 | ||
4354 | uiov = compat_ptr(ptr); | |
4355 | if (req->flags & REQ_F_BUFFER_SELECT) { | |
4356 | compat_ssize_t clen; | |
4357 | ||
4358 | if (len > 1) | |
4359 | return -EINVAL; | |
4360 | if (!access_ok(uiov, sizeof(*uiov))) | |
4361 | return -EFAULT; | |
4362 | if (__get_user(clen, &uiov->iov_len)) | |
4363 | return -EFAULT; | |
4364 | if (clen < 0) | |
4365 | return -EINVAL; | |
1400e697 PB |
4366 | sr->len = iomsg->iov[0].iov_len; |
4367 | iomsg->iov = NULL; | |
52de1fe1 JA |
4368 | } else { |
4369 | ret = compat_import_iovec(READ, uiov, len, UIO_FASTIOV, | |
1400e697 PB |
4370 | &iomsg->iov, |
4371 | &iomsg->msg.msg_iter); | |
52de1fe1 JA |
4372 | if (ret < 0) |
4373 | return ret; | |
4374 | } | |
4375 | ||
4376 | return 0; | |
4377 | } | |
4378 | #endif | |
4379 | ||
1400e697 PB |
4380 | static int io_recvmsg_copy_hdr(struct io_kiocb *req, |
4381 | struct io_async_msghdr *iomsg) | |
52de1fe1 | 4382 | { |
1400e697 PB |
4383 | iomsg->msg.msg_name = &iomsg->addr; |
4384 | iomsg->iov = iomsg->fast_iov; | |
52de1fe1 JA |
4385 | |
4386 | #ifdef CONFIG_COMPAT | |
4387 | if (req->ctx->compat) | |
1400e697 | 4388 | return __io_compat_recvmsg_copy_hdr(req, iomsg); |
fddaface | 4389 | #endif |
52de1fe1 | 4390 | |
1400e697 | 4391 | return __io_recvmsg_copy_hdr(req, iomsg); |
52de1fe1 JA |
4392 | } |
4393 | ||
bcda7baa | 4394 | static struct io_buffer *io_recv_buffer_select(struct io_kiocb *req, |
7fbb1b54 | 4395 | bool needs_lock) |
bcda7baa JA |
4396 | { |
4397 | struct io_sr_msg *sr = &req->sr_msg; | |
4398 | struct io_buffer *kbuf; | |
4399 | ||
bcda7baa JA |
4400 | kbuf = io_buffer_select(req, &sr->len, sr->bgid, sr->kbuf, needs_lock); |
4401 | if (IS_ERR(kbuf)) | |
4402 | return kbuf; | |
4403 | ||
4404 | sr->kbuf = kbuf; | |
4405 | req->flags |= REQ_F_BUFFER_SELECTED; | |
bcda7baa | 4406 | return kbuf; |
fddaface JA |
4407 | } |
4408 | ||
7fbb1b54 PB |
4409 | static inline unsigned int io_put_recv_kbuf(struct io_kiocb *req) |
4410 | { | |
4411 | return io_put_kbuf(req, req->sr_msg.kbuf); | |
4412 | } | |
4413 | ||
3529d8c2 JA |
4414 | static int io_recvmsg_prep(struct io_kiocb *req, |
4415 | const struct io_uring_sqe *sqe) | |
aa1fa28f | 4416 | { |
e8c2bc1f | 4417 | struct io_async_msghdr *async_msg = req->async_data; |
e47293fd | 4418 | struct io_sr_msg *sr = &req->sr_msg; |
99bc4c38 | 4419 | int ret; |
3529d8c2 | 4420 | |
d2b6f48b PB |
4421 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4422 | return -EINVAL; | |
4423 | ||
3529d8c2 | 4424 | sr->msg_flags = READ_ONCE(sqe->msg_flags); |
270a5940 | 4425 | sr->umsg = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
0b7b21e4 | 4426 | sr->len = READ_ONCE(sqe->len); |
bcda7baa | 4427 | sr->bgid = READ_ONCE(sqe->buf_group); |
06b76d44 | 4428 | |
d8768362 JA |
4429 | #ifdef CONFIG_COMPAT |
4430 | if (req->ctx->compat) | |
4431 | sr->msg_flags |= MSG_CMSG_COMPAT; | |
4432 | #endif | |
4433 | ||
e8c2bc1f | 4434 | if (!async_msg || !io_op_defs[req->opcode].needs_async_data) |
06b76d44 | 4435 | return 0; |
e8c2bc1f | 4436 | ret = io_recvmsg_copy_hdr(req, async_msg); |
99bc4c38 PB |
4437 | if (!ret) |
4438 | req->flags |= REQ_F_NEED_CLEANUP; | |
4439 | return ret; | |
aa1fa28f JA |
4440 | } |
4441 | ||
229a7b63 JA |
4442 | static int io_recvmsg(struct io_kiocb *req, bool force_nonblock, |
4443 | struct io_comp_state *cs) | |
aa1fa28f | 4444 | { |
6b754c8b | 4445 | struct io_async_msghdr iomsg, *kmsg; |
03b1230c | 4446 | struct socket *sock; |
7fbb1b54 | 4447 | struct io_buffer *kbuf; |
7a7cacba | 4448 | unsigned flags; |
52de1fe1 | 4449 | int ret, cflags = 0; |
03b1230c | 4450 | |
03b1230c | 4451 | sock = sock_from_file(req->file, &ret); |
7a7cacba PB |
4452 | if (unlikely(!sock)) |
4453 | return ret; | |
3529d8c2 | 4454 | |
e8c2bc1f JA |
4455 | if (req->async_data) { |
4456 | kmsg = req->async_data; | |
4457 | kmsg->msg.msg_name = &kmsg->addr; | |
7a7cacba PB |
4458 | /* if iov is set, it's allocated already */ |
4459 | if (!kmsg->iov) | |
4460 | kmsg->iov = kmsg->fast_iov; | |
4461 | kmsg->msg.msg_iter.iov = kmsg->iov; | |
4462 | } else { | |
4463 | ret = io_recvmsg_copy_hdr(req, &iomsg); | |
4464 | if (ret) | |
681fda8d | 4465 | return ret; |
7a7cacba PB |
4466 | kmsg = &iomsg; |
4467 | } | |
03b1230c | 4468 | |
bc02ef33 | 4469 | if (req->flags & REQ_F_BUFFER_SELECT) { |
7fbb1b54 | 4470 | kbuf = io_recv_buffer_select(req, !force_nonblock); |
bc02ef33 | 4471 | if (IS_ERR(kbuf)) |
52de1fe1 | 4472 | return PTR_ERR(kbuf); |
7a7cacba PB |
4473 | kmsg->fast_iov[0].iov_base = u64_to_user_ptr(kbuf->addr); |
4474 | iov_iter_init(&kmsg->msg.msg_iter, READ, kmsg->iov, | |
4475 | 1, req->sr_msg.len); | |
4476 | } | |
52de1fe1 | 4477 | |
7a7cacba PB |
4478 | flags = req->sr_msg.msg_flags; |
4479 | if (flags & MSG_DONTWAIT) | |
4480 | req->flags |= REQ_F_NOWAIT; | |
4481 | else if (force_nonblock) | |
4482 | flags |= MSG_DONTWAIT; | |
e47293fd | 4483 | |
7a7cacba PB |
4484 | ret = __sys_recvmsg_sock(sock, &kmsg->msg, req->sr_msg.umsg, |
4485 | kmsg->uaddr, flags); | |
0e1b6fe3 PB |
4486 | if (force_nonblock && ret == -EAGAIN) |
4487 | return io_setup_async_msg(req, kmsg); | |
7a7cacba PB |
4488 | if (ret == -ERESTARTSYS) |
4489 | ret = -EINTR; | |
03b1230c | 4490 | |
7fbb1b54 PB |
4491 | if (req->flags & REQ_F_BUFFER_SELECTED) |
4492 | cflags = io_put_recv_kbuf(req); | |
6b754c8b | 4493 | if (kmsg->iov != kmsg->fast_iov) |
0b416c3e | 4494 | kfree(kmsg->iov); |
99bc4c38 | 4495 | req->flags &= ~REQ_F_NEED_CLEANUP; |
4e88d6e7 JA |
4496 | if (ret < 0) |
4497 | req_set_fail_links(req); | |
229a7b63 | 4498 | __io_req_complete(req, ret, cflags, cs); |
03b1230c | 4499 | return 0; |
0fa03c62 | 4500 | } |
5d17b4a4 | 4501 | |
229a7b63 JA |
4502 | static int io_recv(struct io_kiocb *req, bool force_nonblock, |
4503 | struct io_comp_state *cs) | |
fddaface | 4504 | { |
6b754c8b | 4505 | struct io_buffer *kbuf; |
7a7cacba PB |
4506 | struct io_sr_msg *sr = &req->sr_msg; |
4507 | struct msghdr msg; | |
4508 | void __user *buf = sr->buf; | |
fddaface | 4509 | struct socket *sock; |
7a7cacba PB |
4510 | struct iovec iov; |
4511 | unsigned flags; | |
bcda7baa | 4512 | int ret, cflags = 0; |
fddaface | 4513 | |
fddaface | 4514 | sock = sock_from_file(req->file, &ret); |
7a7cacba PB |
4515 | if (unlikely(!sock)) |
4516 | return ret; | |
fddaface | 4517 | |
bc02ef33 | 4518 | if (req->flags & REQ_F_BUFFER_SELECT) { |
7fbb1b54 | 4519 | kbuf = io_recv_buffer_select(req, !force_nonblock); |
bcda7baa JA |
4520 | if (IS_ERR(kbuf)) |
4521 | return PTR_ERR(kbuf); | |
7a7cacba | 4522 | buf = u64_to_user_ptr(kbuf->addr); |
bc02ef33 | 4523 | } |
bcda7baa | 4524 | |
7a7cacba | 4525 | ret = import_single_range(READ, buf, sr->len, &iov, &msg.msg_iter); |
14c32eee PB |
4526 | if (unlikely(ret)) |
4527 | goto out_free; | |
fddaface | 4528 | |
7a7cacba PB |
4529 | msg.msg_name = NULL; |
4530 | msg.msg_control = NULL; | |
4531 | msg.msg_controllen = 0; | |
4532 | msg.msg_namelen = 0; | |
4533 | msg.msg_iocb = NULL; | |
4534 | msg.msg_flags = 0; | |
fddaface | 4535 | |
7a7cacba PB |
4536 | flags = req->sr_msg.msg_flags; |
4537 | if (flags & MSG_DONTWAIT) | |
4538 | req->flags |= REQ_F_NOWAIT; | |
4539 | else if (force_nonblock) | |
4540 | flags |= MSG_DONTWAIT; | |
4541 | ||
4542 | ret = sock_recvmsg(sock, &msg, flags); | |
4543 | if (force_nonblock && ret == -EAGAIN) | |
4544 | return -EAGAIN; | |
4545 | if (ret == -ERESTARTSYS) | |
4546 | ret = -EINTR; | |
14c32eee | 4547 | out_free: |
7fbb1b54 PB |
4548 | if (req->flags & REQ_F_BUFFER_SELECTED) |
4549 | cflags = io_put_recv_kbuf(req); | |
fddaface JA |
4550 | if (ret < 0) |
4551 | req_set_fail_links(req); | |
229a7b63 | 4552 | __io_req_complete(req, ret, cflags, cs); |
fddaface | 4553 | return 0; |
fddaface JA |
4554 | } |
4555 | ||
3529d8c2 | 4556 | static int io_accept_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
17f2fe35 | 4557 | { |
8ed8d3c3 JA |
4558 | struct io_accept *accept = &req->accept; |
4559 | ||
17f2fe35 JA |
4560 | if (unlikely(req->ctx->flags & (IORING_SETUP_IOPOLL|IORING_SETUP_SQPOLL))) |
4561 | return -EINVAL; | |
8042d6ce | 4562 | if (sqe->ioprio || sqe->len || sqe->buf_index) |
17f2fe35 JA |
4563 | return -EINVAL; |
4564 | ||
d55e5f5b JA |
4565 | accept->addr = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
4566 | accept->addr_len = u64_to_user_ptr(READ_ONCE(sqe->addr2)); | |
8ed8d3c3 | 4567 | accept->flags = READ_ONCE(sqe->accept_flags); |
09952e3e | 4568 | accept->nofile = rlimit(RLIMIT_NOFILE); |
8ed8d3c3 | 4569 | return 0; |
8ed8d3c3 | 4570 | } |
17f2fe35 | 4571 | |
229a7b63 JA |
4572 | static int io_accept(struct io_kiocb *req, bool force_nonblock, |
4573 | struct io_comp_state *cs) | |
8ed8d3c3 JA |
4574 | { |
4575 | struct io_accept *accept = &req->accept; | |
ac45abc0 | 4576 | unsigned int file_flags = force_nonblock ? O_NONBLOCK : 0; |
8ed8d3c3 JA |
4577 | int ret; |
4578 | ||
e697deed JX |
4579 | if (req->file->f_flags & O_NONBLOCK) |
4580 | req->flags |= REQ_F_NOWAIT; | |
4581 | ||
8ed8d3c3 | 4582 | ret = __sys_accept4_file(req->file, file_flags, accept->addr, |
09952e3e JA |
4583 | accept->addr_len, accept->flags, |
4584 | accept->nofile); | |
8ed8d3c3 | 4585 | if (ret == -EAGAIN && force_nonblock) |
17f2fe35 | 4586 | return -EAGAIN; |
ac45abc0 PB |
4587 | if (ret < 0) { |
4588 | if (ret == -ERESTARTSYS) | |
4589 | ret = -EINTR; | |
4e88d6e7 | 4590 | req_set_fail_links(req); |
ac45abc0 | 4591 | } |
229a7b63 | 4592 | __io_req_complete(req, ret, 0, cs); |
17f2fe35 | 4593 | return 0; |
8ed8d3c3 JA |
4594 | } |
4595 | ||
3529d8c2 | 4596 | static int io_connect_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
f499a021 | 4597 | { |
3529d8c2 | 4598 | struct io_connect *conn = &req->connect; |
e8c2bc1f | 4599 | struct io_async_connect *io = req->async_data; |
f499a021 | 4600 | |
3fbb51c1 JA |
4601 | if (unlikely(req->ctx->flags & (IORING_SETUP_IOPOLL|IORING_SETUP_SQPOLL))) |
4602 | return -EINVAL; | |
4603 | if (sqe->ioprio || sqe->len || sqe->buf_index || sqe->rw_flags) | |
4604 | return -EINVAL; | |
4605 | ||
3529d8c2 JA |
4606 | conn->addr = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
4607 | conn->addr_len = READ_ONCE(sqe->addr2); | |
4608 | ||
4609 | if (!io) | |
4610 | return 0; | |
4611 | ||
4612 | return move_addr_to_kernel(conn->addr, conn->addr_len, | |
e8c2bc1f | 4613 | &io->address); |
f499a021 JA |
4614 | } |
4615 | ||
229a7b63 JA |
4616 | static int io_connect(struct io_kiocb *req, bool force_nonblock, |
4617 | struct io_comp_state *cs) | |
f8e85cf2 | 4618 | { |
e8c2bc1f | 4619 | struct io_async_connect __io, *io; |
f8e85cf2 | 4620 | unsigned file_flags; |
3fbb51c1 | 4621 | int ret; |
f8e85cf2 | 4622 | |
e8c2bc1f JA |
4623 | if (req->async_data) { |
4624 | io = req->async_data; | |
f499a021 | 4625 | } else { |
3529d8c2 JA |
4626 | ret = move_addr_to_kernel(req->connect.addr, |
4627 | req->connect.addr_len, | |
e8c2bc1f | 4628 | &__io.address); |
f499a021 JA |
4629 | if (ret) |
4630 | goto out; | |
4631 | io = &__io; | |
4632 | } | |
4633 | ||
3fbb51c1 JA |
4634 | file_flags = force_nonblock ? O_NONBLOCK : 0; |
4635 | ||
e8c2bc1f | 4636 | ret = __sys_connect_file(req->file, &io->address, |
3fbb51c1 | 4637 | req->connect.addr_len, file_flags); |
87f80d62 | 4638 | if ((ret == -EAGAIN || ret == -EINPROGRESS) && force_nonblock) { |
e8c2bc1f | 4639 | if (req->async_data) |
b7bb4f7d | 4640 | return -EAGAIN; |
e8c2bc1f | 4641 | if (io_alloc_async_data(req)) { |
f499a021 JA |
4642 | ret = -ENOMEM; |
4643 | goto out; | |
4644 | } | |
e8c2bc1f JA |
4645 | io = req->async_data; |
4646 | memcpy(req->async_data, &__io, sizeof(__io)); | |
f8e85cf2 | 4647 | return -EAGAIN; |
f499a021 | 4648 | } |
f8e85cf2 JA |
4649 | if (ret == -ERESTARTSYS) |
4650 | ret = -EINTR; | |
f499a021 | 4651 | out: |
4e88d6e7 JA |
4652 | if (ret < 0) |
4653 | req_set_fail_links(req); | |
229a7b63 | 4654 | __io_req_complete(req, ret, 0, cs); |
f8e85cf2 | 4655 | return 0; |
469956e8 Y |
4656 | } |
4657 | #else /* !CONFIG_NET */ | |
4658 | static int io_sendmsg_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) | |
4659 | { | |
f8e85cf2 | 4660 | return -EOPNOTSUPP; |
f8e85cf2 JA |
4661 | } |
4662 | ||
1e16c2f9 RD |
4663 | static int io_sendmsg(struct io_kiocb *req, bool force_nonblock, |
4664 | struct io_comp_state *cs) | |
469956e8 Y |
4665 | { |
4666 | return -EOPNOTSUPP; | |
4667 | } | |
4668 | ||
1e16c2f9 RD |
4669 | static int io_send(struct io_kiocb *req, bool force_nonblock, |
4670 | struct io_comp_state *cs) | |
469956e8 Y |
4671 | { |
4672 | return -EOPNOTSUPP; | |
4673 | } | |
4674 | ||
4675 | static int io_recvmsg_prep(struct io_kiocb *req, | |
4676 | const struct io_uring_sqe *sqe) | |
4677 | { | |
4678 | return -EOPNOTSUPP; | |
4679 | } | |
4680 | ||
1e16c2f9 RD |
4681 | static int io_recvmsg(struct io_kiocb *req, bool force_nonblock, |
4682 | struct io_comp_state *cs) | |
469956e8 Y |
4683 | { |
4684 | return -EOPNOTSUPP; | |
4685 | } | |
4686 | ||
1e16c2f9 RD |
4687 | static int io_recv(struct io_kiocb *req, bool force_nonblock, |
4688 | struct io_comp_state *cs) | |
469956e8 Y |
4689 | { |
4690 | return -EOPNOTSUPP; | |
4691 | } | |
4692 | ||
4693 | static int io_accept_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) | |
4694 | { | |
4695 | return -EOPNOTSUPP; | |
4696 | } | |
4697 | ||
1e16c2f9 RD |
4698 | static int io_accept(struct io_kiocb *req, bool force_nonblock, |
4699 | struct io_comp_state *cs) | |
469956e8 Y |
4700 | { |
4701 | return -EOPNOTSUPP; | |
4702 | } | |
ce593a6c | 4703 | |
469956e8 Y |
4704 | static int io_connect_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
4705 | { | |
4706 | return -EOPNOTSUPP; | |
4707 | } | |
4708 | ||
1e16c2f9 RD |
4709 | static int io_connect(struct io_kiocb *req, bool force_nonblock, |
4710 | struct io_comp_state *cs) | |
469956e8 | 4711 | { |
f8e85cf2 | 4712 | return -EOPNOTSUPP; |
ce593a6c | 4713 | } |
469956e8 | 4714 | #endif /* CONFIG_NET */ |
f8e85cf2 | 4715 | |
d7718a9d JA |
4716 | struct io_poll_table { |
4717 | struct poll_table_struct pt; | |
4718 | struct io_kiocb *req; | |
4719 | int error; | |
4720 | }; | |
ce593a6c | 4721 | |
d7718a9d JA |
4722 | static int __io_async_wake(struct io_kiocb *req, struct io_poll_iocb *poll, |
4723 | __poll_t mask, task_work_func_t func) | |
4724 | { | |
fd7d6de2 | 4725 | bool twa_signal_ok; |
aa96bf8a | 4726 | int ret; |
d7718a9d JA |
4727 | |
4728 | /* for instances that support it check for an event match first: */ | |
4729 | if (mask && !(mask & poll->events)) | |
4730 | return 0; | |
4731 | ||
4732 | trace_io_uring_task_add(req->ctx, req->opcode, req->user_data, mask); | |
4733 | ||
4734 | list_del_init(&poll->wait.entry); | |
4735 | ||
d7718a9d JA |
4736 | req->result = mask; |
4737 | init_task_work(&req->task_work, func); | |
6d816e08 JA |
4738 | percpu_ref_get(&req->ctx->refs); |
4739 | ||
fd7d6de2 JA |
4740 | /* |
4741 | * If we using the signalfd wait_queue_head for this wakeup, then | |
4742 | * it's not safe to use TWA_SIGNAL as we could be recursing on the | |
4743 | * tsk->sighand->siglock on doing the wakeup. Should not be needed | |
4744 | * either, as the normal wakeup will suffice. | |
4745 | */ | |
4746 | twa_signal_ok = (poll->head != &req->task->sighand->signalfd_wqh); | |
4747 | ||
d7718a9d | 4748 | /* |
e3aabf95 JA |
4749 | * If this fails, then the task is exiting. When a task exits, the |
4750 | * work gets canceled, so just cancel this request as well instead | |
4751 | * of executing it. We can't safely execute it anyway, as we may not | |
4752 | * have the needed state needed for it anyway. | |
d7718a9d | 4753 | */ |
87c4311f | 4754 | ret = io_req_task_work_add(req, twa_signal_ok); |
aa96bf8a | 4755 | if (unlikely(ret)) { |
c2c4c83c JA |
4756 | struct task_struct *tsk; |
4757 | ||
e3aabf95 | 4758 | WRITE_ONCE(poll->canceled, true); |
aa96bf8a | 4759 | tsk = io_wq_get_task(req->ctx->io_wq); |
ce593a6c JA |
4760 | task_work_add(tsk, &req->task_work, 0); |
4761 | wake_up_process(tsk); | |
aa96bf8a | 4762 | } |
d7718a9d JA |
4763 | return 1; |
4764 | } | |
4765 | ||
74ce6ce4 JA |
4766 | static bool io_poll_rewait(struct io_kiocb *req, struct io_poll_iocb *poll) |
4767 | __acquires(&req->ctx->completion_lock) | |
4768 | { | |
4769 | struct io_ring_ctx *ctx = req->ctx; | |
4770 | ||
4771 | if (!req->result && !READ_ONCE(poll->canceled)) { | |
4772 | struct poll_table_struct pt = { ._key = poll->events }; | |
4773 | ||
4774 | req->result = vfs_poll(req->file, &pt) & poll->events; | |
4775 | } | |
4776 | ||
4777 | spin_lock_irq(&ctx->completion_lock); | |
4778 | if (!req->result && !READ_ONCE(poll->canceled)) { | |
4779 | add_wait_queue(poll->head, &poll->wait); | |
4780 | return true; | |
4781 | } | |
4782 | ||
4783 | return false; | |
4784 | } | |
4785 | ||
d4e7cd36 | 4786 | static struct io_poll_iocb *io_poll_get_double(struct io_kiocb *req) |
18bceab1 | 4787 | { |
e8c2bc1f | 4788 | /* pure poll stashes this in ->async_data, poll driven retry elsewhere */ |
d4e7cd36 | 4789 | if (req->opcode == IORING_OP_POLL_ADD) |
e8c2bc1f | 4790 | return req->async_data; |
d4e7cd36 JA |
4791 | return req->apoll->double_poll; |
4792 | } | |
4793 | ||
4794 | static struct io_poll_iocb *io_poll_get_single(struct io_kiocb *req) | |
4795 | { | |
4796 | if (req->opcode == IORING_OP_POLL_ADD) | |
4797 | return &req->poll; | |
4798 | return &req->apoll->poll; | |
4799 | } | |
4800 | ||
4801 | static void io_poll_remove_double(struct io_kiocb *req) | |
4802 | { | |
4803 | struct io_poll_iocb *poll = io_poll_get_double(req); | |
18bceab1 JA |
4804 | |
4805 | lockdep_assert_held(&req->ctx->completion_lock); | |
4806 | ||
4807 | if (poll && poll->head) { | |
4808 | struct wait_queue_head *head = poll->head; | |
4809 | ||
4810 | spin_lock(&head->lock); | |
4811 | list_del_init(&poll->wait.entry); | |
4812 | if (poll->wait.private) | |
4813 | refcount_dec(&req->refs); | |
4814 | poll->head = NULL; | |
4815 | spin_unlock(&head->lock); | |
4816 | } | |
4817 | } | |
4818 | ||
4819 | static void io_poll_complete(struct io_kiocb *req, __poll_t mask, int error) | |
4820 | { | |
4821 | struct io_ring_ctx *ctx = req->ctx; | |
4822 | ||
d4e7cd36 | 4823 | io_poll_remove_double(req); |
18bceab1 JA |
4824 | req->poll.done = true; |
4825 | io_cqring_fill_event(req, error ? error : mangle_poll(mask)); | |
4826 | io_commit_cqring(ctx); | |
4827 | } | |
4828 | ||
4829 | static void io_poll_task_handler(struct io_kiocb *req, struct io_kiocb **nxt) | |
4830 | { | |
4831 | struct io_ring_ctx *ctx = req->ctx; | |
4832 | ||
4833 | if (io_poll_rewait(req, &req->poll)) { | |
4834 | spin_unlock_irq(&ctx->completion_lock); | |
4835 | return; | |
4836 | } | |
4837 | ||
4838 | hash_del(&req->hash_node); | |
4839 | io_poll_complete(req, req->result, 0); | |
4840 | req->flags |= REQ_F_COMP_LOCKED; | |
9b5f7bd9 | 4841 | *nxt = io_put_req_find_next(req); |
18bceab1 JA |
4842 | spin_unlock_irq(&ctx->completion_lock); |
4843 | ||
4844 | io_cqring_ev_posted(ctx); | |
4845 | } | |
4846 | ||
4847 | static void io_poll_task_func(struct callback_head *cb) | |
4848 | { | |
4849 | struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work); | |
6d816e08 | 4850 | struct io_ring_ctx *ctx = req->ctx; |
18bceab1 JA |
4851 | struct io_kiocb *nxt = NULL; |
4852 | ||
4853 | io_poll_task_handler(req, &nxt); | |
ea1164e5 PB |
4854 | if (nxt) |
4855 | __io_req_task_submit(nxt); | |
6d816e08 | 4856 | percpu_ref_put(&ctx->refs); |
18bceab1 JA |
4857 | } |
4858 | ||
4859 | static int io_poll_double_wake(struct wait_queue_entry *wait, unsigned mode, | |
4860 | int sync, void *key) | |
4861 | { | |
4862 | struct io_kiocb *req = wait->private; | |
d4e7cd36 | 4863 | struct io_poll_iocb *poll = io_poll_get_single(req); |
18bceab1 JA |
4864 | __poll_t mask = key_to_poll(key); |
4865 | ||
4866 | /* for instances that support it check for an event match first: */ | |
4867 | if (mask && !(mask & poll->events)) | |
4868 | return 0; | |
4869 | ||
8706e04e JA |
4870 | list_del_init(&wait->entry); |
4871 | ||
807abcb0 | 4872 | if (poll && poll->head) { |
18bceab1 JA |
4873 | bool done; |
4874 | ||
807abcb0 JA |
4875 | spin_lock(&poll->head->lock); |
4876 | done = list_empty(&poll->wait.entry); | |
18bceab1 | 4877 | if (!done) |
807abcb0 | 4878 | list_del_init(&poll->wait.entry); |
d4e7cd36 JA |
4879 | /* make sure double remove sees this as being gone */ |
4880 | wait->private = NULL; | |
807abcb0 | 4881 | spin_unlock(&poll->head->lock); |
18bceab1 JA |
4882 | if (!done) |
4883 | __io_async_wake(req, poll, mask, io_poll_task_func); | |
4884 | } | |
4885 | refcount_dec(&req->refs); | |
4886 | return 1; | |
4887 | } | |
4888 | ||
4889 | static void io_init_poll_iocb(struct io_poll_iocb *poll, __poll_t events, | |
4890 | wait_queue_func_t wake_func) | |
4891 | { | |
4892 | poll->head = NULL; | |
4893 | poll->done = false; | |
4894 | poll->canceled = false; | |
4895 | poll->events = events; | |
4896 | INIT_LIST_HEAD(&poll->wait.entry); | |
4897 | init_waitqueue_func_entry(&poll->wait, wake_func); | |
4898 | } | |
4899 | ||
4900 | static void __io_queue_proc(struct io_poll_iocb *poll, struct io_poll_table *pt, | |
807abcb0 JA |
4901 | struct wait_queue_head *head, |
4902 | struct io_poll_iocb **poll_ptr) | |
18bceab1 JA |
4903 | { |
4904 | struct io_kiocb *req = pt->req; | |
4905 | ||
4906 | /* | |
4907 | * If poll->head is already set, it's because the file being polled | |
4908 | * uses multiple waitqueues for poll handling (eg one for read, one | |
4909 | * for write). Setup a separate io_poll_iocb if this happens. | |
4910 | */ | |
4911 | if (unlikely(poll->head)) { | |
4912 | /* already have a 2nd entry, fail a third attempt */ | |
807abcb0 | 4913 | if (*poll_ptr) { |
18bceab1 JA |
4914 | pt->error = -EINVAL; |
4915 | return; | |
4916 | } | |
4917 | poll = kmalloc(sizeof(*poll), GFP_ATOMIC); | |
4918 | if (!poll) { | |
4919 | pt->error = -ENOMEM; | |
4920 | return; | |
4921 | } | |
4922 | io_init_poll_iocb(poll, req->poll.events, io_poll_double_wake); | |
4923 | refcount_inc(&req->refs); | |
4924 | poll->wait.private = req; | |
807abcb0 | 4925 | *poll_ptr = poll; |
18bceab1 JA |
4926 | } |
4927 | ||
4928 | pt->error = 0; | |
4929 | poll->head = head; | |
a31eb4a2 JX |
4930 | |
4931 | if (poll->events & EPOLLEXCLUSIVE) | |
4932 | add_wait_queue_exclusive(head, &poll->wait); | |
4933 | else | |
4934 | add_wait_queue(head, &poll->wait); | |
18bceab1 JA |
4935 | } |
4936 | ||
4937 | static void io_async_queue_proc(struct file *file, struct wait_queue_head *head, | |
4938 | struct poll_table_struct *p) | |
4939 | { | |
4940 | struct io_poll_table *pt = container_of(p, struct io_poll_table, pt); | |
807abcb0 | 4941 | struct async_poll *apoll = pt->req->apoll; |
18bceab1 | 4942 | |
807abcb0 | 4943 | __io_queue_proc(&apoll->poll, pt, head, &apoll->double_poll); |
18bceab1 JA |
4944 | } |
4945 | ||
d7718a9d JA |
4946 | static void io_async_task_func(struct callback_head *cb) |
4947 | { | |
4948 | struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work); | |
4949 | struct async_poll *apoll = req->apoll; | |
4950 | struct io_ring_ctx *ctx = req->ctx; | |
4951 | ||
4952 | trace_io_uring_task_run(req->ctx, req->opcode, req->user_data); | |
4953 | ||
74ce6ce4 | 4954 | if (io_poll_rewait(req, &apoll->poll)) { |
d7718a9d | 4955 | spin_unlock_irq(&ctx->completion_lock); |
6d816e08 | 4956 | percpu_ref_put(&ctx->refs); |
74ce6ce4 | 4957 | return; |
d7718a9d JA |
4958 | } |
4959 | ||
31067255 | 4960 | /* If req is still hashed, it cannot have been canceled. Don't check. */ |
0be0b0e3 | 4961 | if (hash_hashed(&req->hash_node)) |
74ce6ce4 | 4962 | hash_del(&req->hash_node); |
2bae047e | 4963 | |
d4e7cd36 | 4964 | io_poll_remove_double(req); |
74ce6ce4 JA |
4965 | spin_unlock_irq(&ctx->completion_lock); |
4966 | ||
0be0b0e3 PB |
4967 | if (!READ_ONCE(apoll->poll.canceled)) |
4968 | __io_req_task_submit(req); | |
4969 | else | |
4970 | __io_req_task_cancel(req, -ECANCELED); | |
aa340845 | 4971 | |
6d816e08 | 4972 | percpu_ref_put(&ctx->refs); |
807abcb0 | 4973 | kfree(apoll->double_poll); |
31067255 | 4974 | kfree(apoll); |
d7718a9d JA |
4975 | } |
4976 | ||
4977 | static int io_async_wake(struct wait_queue_entry *wait, unsigned mode, int sync, | |
4978 | void *key) | |
4979 | { | |
4980 | struct io_kiocb *req = wait->private; | |
4981 | struct io_poll_iocb *poll = &req->apoll->poll; | |
4982 | ||
4983 | trace_io_uring_poll_wake(req->ctx, req->opcode, req->user_data, | |
4984 | key_to_poll(key)); | |
4985 | ||
4986 | return __io_async_wake(req, poll, key_to_poll(key), io_async_task_func); | |
4987 | } | |
4988 | ||
4989 | static void io_poll_req_insert(struct io_kiocb *req) | |
4990 | { | |
4991 | struct io_ring_ctx *ctx = req->ctx; | |
4992 | struct hlist_head *list; | |
4993 | ||
4994 | list = &ctx->cancel_hash[hash_long(req->user_data, ctx->cancel_hash_bits)]; | |
4995 | hlist_add_head(&req->hash_node, list); | |
4996 | } | |
4997 | ||
4998 | static __poll_t __io_arm_poll_handler(struct io_kiocb *req, | |
4999 | struct io_poll_iocb *poll, | |
5000 | struct io_poll_table *ipt, __poll_t mask, | |
5001 | wait_queue_func_t wake_func) | |
5002 | __acquires(&ctx->completion_lock) | |
5003 | { | |
5004 | struct io_ring_ctx *ctx = req->ctx; | |
5005 | bool cancel = false; | |
5006 | ||
18bceab1 | 5007 | io_init_poll_iocb(poll, mask, wake_func); |
b90cd197 | 5008 | poll->file = req->file; |
18bceab1 | 5009 | poll->wait.private = req; |
d7718a9d JA |
5010 | |
5011 | ipt->pt._key = mask; | |
5012 | ipt->req = req; | |
5013 | ipt->error = -EINVAL; | |
5014 | ||
d7718a9d JA |
5015 | mask = vfs_poll(req->file, &ipt->pt) & poll->events; |
5016 | ||
5017 | spin_lock_irq(&ctx->completion_lock); | |
5018 | if (likely(poll->head)) { | |
5019 | spin_lock(&poll->head->lock); | |
5020 | if (unlikely(list_empty(&poll->wait.entry))) { | |
5021 | if (ipt->error) | |
5022 | cancel = true; | |
5023 | ipt->error = 0; | |
5024 | mask = 0; | |
5025 | } | |
5026 | if (mask || ipt->error) | |
5027 | list_del_init(&poll->wait.entry); | |
5028 | else if (cancel) | |
5029 | WRITE_ONCE(poll->canceled, true); | |
5030 | else if (!poll->done) /* actually waiting for an event */ | |
5031 | io_poll_req_insert(req); | |
5032 | spin_unlock(&poll->head->lock); | |
5033 | } | |
5034 | ||
5035 | return mask; | |
5036 | } | |
5037 | ||
5038 | static bool io_arm_poll_handler(struct io_kiocb *req) | |
5039 | { | |
5040 | const struct io_op_def *def = &io_op_defs[req->opcode]; | |
5041 | struct io_ring_ctx *ctx = req->ctx; | |
5042 | struct async_poll *apoll; | |
5043 | struct io_poll_table ipt; | |
5044 | __poll_t mask, ret; | |
9dab14b8 | 5045 | int rw; |
d7718a9d JA |
5046 | |
5047 | if (!req->file || !file_can_poll(req->file)) | |
5048 | return false; | |
24c74678 | 5049 | if (req->flags & REQ_F_POLLED) |
d7718a9d | 5050 | return false; |
9dab14b8 JA |
5051 | if (def->pollin) |
5052 | rw = READ; | |
5053 | else if (def->pollout) | |
5054 | rw = WRITE; | |
5055 | else | |
5056 | return false; | |
5057 | /* if we can't nonblock try, then no point in arming a poll handler */ | |
5058 | if (!io_file_supports_async(req->file, rw)) | |
d7718a9d JA |
5059 | return false; |
5060 | ||
5061 | apoll = kmalloc(sizeof(*apoll), GFP_ATOMIC); | |
5062 | if (unlikely(!apoll)) | |
5063 | return false; | |
807abcb0 | 5064 | apoll->double_poll = NULL; |
d7718a9d JA |
5065 | |
5066 | req->flags |= REQ_F_POLLED; | |
d7718a9d JA |
5067 | req->apoll = apoll; |
5068 | INIT_HLIST_NODE(&req->hash_node); | |
5069 | ||
8755d97a | 5070 | mask = 0; |
d7718a9d | 5071 | if (def->pollin) |
8755d97a | 5072 | mask |= POLLIN | POLLRDNORM; |
d7718a9d JA |
5073 | if (def->pollout) |
5074 | mask |= POLLOUT | POLLWRNORM; | |
5075 | mask |= POLLERR | POLLPRI; | |
5076 | ||
5077 | ipt.pt._qproc = io_async_queue_proc; | |
5078 | ||
5079 | ret = __io_arm_poll_handler(req, &apoll->poll, &ipt, mask, | |
5080 | io_async_wake); | |
a36da65c | 5081 | if (ret || ipt.error) { |
d4e7cd36 | 5082 | io_poll_remove_double(req); |
d7718a9d | 5083 | spin_unlock_irq(&ctx->completion_lock); |
807abcb0 | 5084 | kfree(apoll->double_poll); |
d7718a9d JA |
5085 | kfree(apoll); |
5086 | return false; | |
5087 | } | |
5088 | spin_unlock_irq(&ctx->completion_lock); | |
5089 | trace_io_uring_poll_arm(ctx, req->opcode, req->user_data, mask, | |
5090 | apoll->poll.events); | |
5091 | return true; | |
5092 | } | |
5093 | ||
5094 | static bool __io_poll_remove_one(struct io_kiocb *req, | |
5095 | struct io_poll_iocb *poll) | |
221c5eb2 | 5096 | { |
b41e9852 | 5097 | bool do_complete = false; |
221c5eb2 JA |
5098 | |
5099 | spin_lock(&poll->head->lock); | |
5100 | WRITE_ONCE(poll->canceled, true); | |
392edb45 JA |
5101 | if (!list_empty(&poll->wait.entry)) { |
5102 | list_del_init(&poll->wait.entry); | |
b41e9852 | 5103 | do_complete = true; |
221c5eb2 JA |
5104 | } |
5105 | spin_unlock(&poll->head->lock); | |
3bfa5bcb | 5106 | hash_del(&req->hash_node); |
d7718a9d JA |
5107 | return do_complete; |
5108 | } | |
5109 | ||
5110 | static bool io_poll_remove_one(struct io_kiocb *req) | |
5111 | { | |
5112 | bool do_complete; | |
5113 | ||
d4e7cd36 JA |
5114 | io_poll_remove_double(req); |
5115 | ||
d7718a9d JA |
5116 | if (req->opcode == IORING_OP_POLL_ADD) { |
5117 | do_complete = __io_poll_remove_one(req, &req->poll); | |
5118 | } else { | |
3bfa5bcb JA |
5119 | struct async_poll *apoll = req->apoll; |
5120 | ||
d7718a9d | 5121 | /* non-poll requests have submit ref still */ |
3bfa5bcb JA |
5122 | do_complete = __io_poll_remove_one(req, &apoll->poll); |
5123 | if (do_complete) { | |
d7718a9d | 5124 | io_put_req(req); |
807abcb0 | 5125 | kfree(apoll->double_poll); |
3bfa5bcb JA |
5126 | kfree(apoll); |
5127 | } | |
b1f573bd XW |
5128 | } |
5129 | ||
b41e9852 JA |
5130 | if (do_complete) { |
5131 | io_cqring_fill_event(req, -ECANCELED); | |
5132 | io_commit_cqring(req->ctx); | |
5133 | req->flags |= REQ_F_COMP_LOCKED; | |
f254ac04 | 5134 | req_set_fail_links(req); |
b41e9852 JA |
5135 | io_put_req(req); |
5136 | } | |
5137 | ||
5138 | return do_complete; | |
221c5eb2 JA |
5139 | } |
5140 | ||
76e1b642 JA |
5141 | /* |
5142 | * Returns true if we found and killed one or more poll requests | |
5143 | */ | |
5144 | static bool io_poll_remove_all(struct io_ring_ctx *ctx, struct task_struct *tsk) | |
221c5eb2 | 5145 | { |
78076bb6 | 5146 | struct hlist_node *tmp; |
221c5eb2 | 5147 | struct io_kiocb *req; |
8e2e1faf | 5148 | int posted = 0, i; |
221c5eb2 JA |
5149 | |
5150 | spin_lock_irq(&ctx->completion_lock); | |
78076bb6 JA |
5151 | for (i = 0; i < (1U << ctx->cancel_hash_bits); i++) { |
5152 | struct hlist_head *list; | |
5153 | ||
5154 | list = &ctx->cancel_hash[i]; | |
f3606e3a JA |
5155 | hlist_for_each_entry_safe(req, tmp, list, hash_node) { |
5156 | if (io_task_match(req, tsk)) | |
5157 | posted += io_poll_remove_one(req); | |
5158 | } | |
221c5eb2 JA |
5159 | } |
5160 | spin_unlock_irq(&ctx->completion_lock); | |
b41e9852 | 5161 | |
8e2e1faf JA |
5162 | if (posted) |
5163 | io_cqring_ev_posted(ctx); | |
76e1b642 JA |
5164 | |
5165 | return posted != 0; | |
221c5eb2 JA |
5166 | } |
5167 | ||
47f46768 JA |
5168 | static int io_poll_cancel(struct io_ring_ctx *ctx, __u64 sqe_addr) |
5169 | { | |
78076bb6 | 5170 | struct hlist_head *list; |
47f46768 JA |
5171 | struct io_kiocb *req; |
5172 | ||
78076bb6 JA |
5173 | list = &ctx->cancel_hash[hash_long(sqe_addr, ctx->cancel_hash_bits)]; |
5174 | hlist_for_each_entry(req, list, hash_node) { | |
b41e9852 JA |
5175 | if (sqe_addr != req->user_data) |
5176 | continue; | |
5177 | if (io_poll_remove_one(req)) | |
eac406c6 | 5178 | return 0; |
b41e9852 | 5179 | return -EALREADY; |
47f46768 JA |
5180 | } |
5181 | ||
5182 | return -ENOENT; | |
5183 | } | |
5184 | ||
3529d8c2 JA |
5185 | static int io_poll_remove_prep(struct io_kiocb *req, |
5186 | const struct io_uring_sqe *sqe) | |
0969e783 | 5187 | { |
0969e783 JA |
5188 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
5189 | return -EINVAL; | |
5190 | if (sqe->ioprio || sqe->off || sqe->len || sqe->buf_index || | |
5191 | sqe->poll_events) | |
5192 | return -EINVAL; | |
5193 | ||
5194 | req->poll.addr = READ_ONCE(sqe->addr); | |
0969e783 JA |
5195 | return 0; |
5196 | } | |
5197 | ||
221c5eb2 JA |
5198 | /* |
5199 | * Find a running poll command that matches one specified in sqe->addr, | |
5200 | * and remove it if found. | |
5201 | */ | |
fc4df999 | 5202 | static int io_poll_remove(struct io_kiocb *req) |
221c5eb2 JA |
5203 | { |
5204 | struct io_ring_ctx *ctx = req->ctx; | |
0969e783 | 5205 | u64 addr; |
47f46768 | 5206 | int ret; |
221c5eb2 | 5207 | |
0969e783 | 5208 | addr = req->poll.addr; |
221c5eb2 | 5209 | spin_lock_irq(&ctx->completion_lock); |
0969e783 | 5210 | ret = io_poll_cancel(ctx, addr); |
221c5eb2 JA |
5211 | spin_unlock_irq(&ctx->completion_lock); |
5212 | ||
4e88d6e7 JA |
5213 | if (ret < 0) |
5214 | req_set_fail_links(req); | |
e1e16097 | 5215 | io_req_complete(req, ret); |
221c5eb2 JA |
5216 | return 0; |
5217 | } | |
5218 | ||
221c5eb2 JA |
5219 | static int io_poll_wake(struct wait_queue_entry *wait, unsigned mode, int sync, |
5220 | void *key) | |
5221 | { | |
c2f2eb7d JA |
5222 | struct io_kiocb *req = wait->private; |
5223 | struct io_poll_iocb *poll = &req->poll; | |
221c5eb2 | 5224 | |
d7718a9d | 5225 | return __io_async_wake(req, poll, key_to_poll(key), io_poll_task_func); |
221c5eb2 JA |
5226 | } |
5227 | ||
221c5eb2 JA |
5228 | static void io_poll_queue_proc(struct file *file, struct wait_queue_head *head, |
5229 | struct poll_table_struct *p) | |
5230 | { | |
5231 | struct io_poll_table *pt = container_of(p, struct io_poll_table, pt); | |
5232 | ||
e8c2bc1f | 5233 | __io_queue_proc(&pt->req->poll, pt, head, (struct io_poll_iocb **) &pt->req->async_data); |
eac406c6 JA |
5234 | } |
5235 | ||
3529d8c2 | 5236 | static int io_poll_add_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
221c5eb2 JA |
5237 | { |
5238 | struct io_poll_iocb *poll = &req->poll; | |
5769a351 | 5239 | u32 events; |
221c5eb2 JA |
5240 | |
5241 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) | |
5242 | return -EINVAL; | |
5243 | if (sqe->addr || sqe->ioprio || sqe->off || sqe->len || sqe->buf_index) | |
5244 | return -EINVAL; | |
09bb8394 JA |
5245 | if (!poll->file) |
5246 | return -EBADF; | |
221c5eb2 | 5247 | |
5769a351 JX |
5248 | events = READ_ONCE(sqe->poll32_events); |
5249 | #ifdef __BIG_ENDIAN | |
5250 | events = swahw32(events); | |
5251 | #endif | |
a31eb4a2 JX |
5252 | poll->events = demangle_poll(events) | EPOLLERR | EPOLLHUP | |
5253 | (events & EPOLLEXCLUSIVE); | |
0969e783 JA |
5254 | return 0; |
5255 | } | |
5256 | ||
014db007 | 5257 | static int io_poll_add(struct io_kiocb *req) |
0969e783 JA |
5258 | { |
5259 | struct io_poll_iocb *poll = &req->poll; | |
5260 | struct io_ring_ctx *ctx = req->ctx; | |
5261 | struct io_poll_table ipt; | |
0969e783 | 5262 | __poll_t mask; |
0969e783 | 5263 | |
78076bb6 | 5264 | INIT_HLIST_NODE(&req->hash_node); |
d7718a9d | 5265 | ipt.pt._qproc = io_poll_queue_proc; |
36703247 | 5266 | |
d7718a9d JA |
5267 | mask = __io_arm_poll_handler(req, &req->poll, &ipt, poll->events, |
5268 | io_poll_wake); | |
221c5eb2 | 5269 | |
8c838788 | 5270 | if (mask) { /* no async, we'd stolen it */ |
221c5eb2 | 5271 | ipt.error = 0; |
b0dd8a41 | 5272 | io_poll_complete(req, mask, 0); |
221c5eb2 | 5273 | } |
221c5eb2 JA |
5274 | spin_unlock_irq(&ctx->completion_lock); |
5275 | ||
8c838788 JA |
5276 | if (mask) { |
5277 | io_cqring_ev_posted(ctx); | |
014db007 | 5278 | io_put_req(req); |
221c5eb2 | 5279 | } |
8c838788 | 5280 | return ipt.error; |
221c5eb2 JA |
5281 | } |
5282 | ||
5262f567 JA |
5283 | static enum hrtimer_restart io_timeout_fn(struct hrtimer *timer) |
5284 | { | |
ad8a48ac JA |
5285 | struct io_timeout_data *data = container_of(timer, |
5286 | struct io_timeout_data, timer); | |
5287 | struct io_kiocb *req = data->req; | |
5288 | struct io_ring_ctx *ctx = req->ctx; | |
5262f567 JA |
5289 | unsigned long flags; |
5290 | ||
5262f567 | 5291 | spin_lock_irqsave(&ctx->completion_lock, flags); |
01cec8c1 PB |
5292 | atomic_set(&req->ctx->cq_timeouts, |
5293 | atomic_read(&req->ctx->cq_timeouts) + 1); | |
5294 | ||
ef03681a | 5295 | /* |
11365043 JA |
5296 | * We could be racing with timeout deletion. If the list is empty, |
5297 | * then timeout lookup already found it and will be handling it. | |
ef03681a | 5298 | */ |
135fcde8 PB |
5299 | if (!list_empty(&req->timeout.list)) |
5300 | list_del_init(&req->timeout.list); | |
5262f567 | 5301 | |
78e19bbe | 5302 | io_cqring_fill_event(req, -ETIME); |
5262f567 JA |
5303 | io_commit_cqring(ctx); |
5304 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
5305 | ||
5306 | io_cqring_ev_posted(ctx); | |
4e88d6e7 | 5307 | req_set_fail_links(req); |
5262f567 JA |
5308 | io_put_req(req); |
5309 | return HRTIMER_NORESTART; | |
5310 | } | |
5311 | ||
f254ac04 JA |
5312 | static int __io_timeout_cancel(struct io_kiocb *req) |
5313 | { | |
e8c2bc1f | 5314 | struct io_timeout_data *io = req->async_data; |
f254ac04 JA |
5315 | int ret; |
5316 | ||
5317 | list_del_init(&req->timeout.list); | |
5318 | ||
e8c2bc1f | 5319 | ret = hrtimer_try_to_cancel(&io->timer); |
f254ac04 JA |
5320 | if (ret == -1) |
5321 | return -EALREADY; | |
5322 | ||
5323 | req_set_fail_links(req); | |
5324 | req->flags |= REQ_F_COMP_LOCKED; | |
5325 | io_cqring_fill_event(req, -ECANCELED); | |
5326 | io_put_req(req); | |
5327 | return 0; | |
5328 | } | |
5329 | ||
47f46768 JA |
5330 | static int io_timeout_cancel(struct io_ring_ctx *ctx, __u64 user_data) |
5331 | { | |
5332 | struct io_kiocb *req; | |
5333 | int ret = -ENOENT; | |
5334 | ||
135fcde8 | 5335 | list_for_each_entry(req, &ctx->timeout_list, timeout.list) { |
47f46768 | 5336 | if (user_data == req->user_data) { |
47f46768 JA |
5337 | ret = 0; |
5338 | break; | |
5339 | } | |
5340 | } | |
5341 | ||
5342 | if (ret == -ENOENT) | |
5343 | return ret; | |
5344 | ||
f254ac04 | 5345 | return __io_timeout_cancel(req); |
47f46768 JA |
5346 | } |
5347 | ||
3529d8c2 JA |
5348 | static int io_timeout_remove_prep(struct io_kiocb *req, |
5349 | const struct io_uring_sqe *sqe) | |
b29472ee | 5350 | { |
b29472ee JA |
5351 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
5352 | return -EINVAL; | |
61710e43 DA |
5353 | if (unlikely(req->flags & (REQ_F_FIXED_FILE | REQ_F_BUFFER_SELECT))) |
5354 | return -EINVAL; | |
5355 | if (sqe->ioprio || sqe->buf_index || sqe->len) | |
b29472ee JA |
5356 | return -EINVAL; |
5357 | ||
5358 | req->timeout.addr = READ_ONCE(sqe->addr); | |
5359 | req->timeout.flags = READ_ONCE(sqe->timeout_flags); | |
5360 | if (req->timeout.flags) | |
5361 | return -EINVAL; | |
5362 | ||
b29472ee JA |
5363 | return 0; |
5364 | } | |
5365 | ||
11365043 JA |
5366 | /* |
5367 | * Remove or update an existing timeout command | |
5368 | */ | |
fc4df999 | 5369 | static int io_timeout_remove(struct io_kiocb *req) |
11365043 JA |
5370 | { |
5371 | struct io_ring_ctx *ctx = req->ctx; | |
47f46768 | 5372 | int ret; |
11365043 | 5373 | |
11365043 | 5374 | spin_lock_irq(&ctx->completion_lock); |
b29472ee | 5375 | ret = io_timeout_cancel(ctx, req->timeout.addr); |
11365043 | 5376 | |
47f46768 | 5377 | io_cqring_fill_event(req, ret); |
11365043 JA |
5378 | io_commit_cqring(ctx); |
5379 | spin_unlock_irq(&ctx->completion_lock); | |
5262f567 | 5380 | io_cqring_ev_posted(ctx); |
4e88d6e7 JA |
5381 | if (ret < 0) |
5382 | req_set_fail_links(req); | |
ec9c02ad | 5383 | io_put_req(req); |
11365043 | 5384 | return 0; |
5262f567 JA |
5385 | } |
5386 | ||
3529d8c2 | 5387 | static int io_timeout_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe, |
2d28390a | 5388 | bool is_timeout_link) |
5262f567 | 5389 | { |
ad8a48ac | 5390 | struct io_timeout_data *data; |
a41525ab | 5391 | unsigned flags; |
56080b02 | 5392 | u32 off = READ_ONCE(sqe->off); |
5262f567 | 5393 | |
ad8a48ac | 5394 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
5262f567 | 5395 | return -EINVAL; |
ad8a48ac | 5396 | if (sqe->ioprio || sqe->buf_index || sqe->len != 1) |
a41525ab | 5397 | return -EINVAL; |
56080b02 | 5398 | if (off && is_timeout_link) |
2d28390a | 5399 | return -EINVAL; |
a41525ab JA |
5400 | flags = READ_ONCE(sqe->timeout_flags); |
5401 | if (flags & ~IORING_TIMEOUT_ABS) | |
5262f567 | 5402 | return -EINVAL; |
bdf20073 | 5403 | |
bfe68a22 | 5404 | req->timeout.off = off; |
26a61679 | 5405 | |
e8c2bc1f | 5406 | if (!req->async_data && io_alloc_async_data(req)) |
26a61679 JA |
5407 | return -ENOMEM; |
5408 | ||
e8c2bc1f | 5409 | data = req->async_data; |
ad8a48ac | 5410 | data->req = req; |
ad8a48ac JA |
5411 | |
5412 | if (get_timespec64(&data->ts, u64_to_user_ptr(sqe->addr))) | |
5262f567 JA |
5413 | return -EFAULT; |
5414 | ||
11365043 | 5415 | if (flags & IORING_TIMEOUT_ABS) |
ad8a48ac | 5416 | data->mode = HRTIMER_MODE_ABS; |
11365043 | 5417 | else |
ad8a48ac | 5418 | data->mode = HRTIMER_MODE_REL; |
11365043 | 5419 | |
ad8a48ac JA |
5420 | hrtimer_init(&data->timer, CLOCK_MONOTONIC, data->mode); |
5421 | return 0; | |
5422 | } | |
5423 | ||
fc4df999 | 5424 | static int io_timeout(struct io_kiocb *req) |
ad8a48ac | 5425 | { |
ad8a48ac | 5426 | struct io_ring_ctx *ctx = req->ctx; |
e8c2bc1f | 5427 | struct io_timeout_data *data = req->async_data; |
ad8a48ac | 5428 | struct list_head *entry; |
bfe68a22 | 5429 | u32 tail, off = req->timeout.off; |
ad8a48ac | 5430 | |
733f5c95 | 5431 | spin_lock_irq(&ctx->completion_lock); |
93bd25bb | 5432 | |
5262f567 JA |
5433 | /* |
5434 | * sqe->off holds how many events that need to occur for this | |
93bd25bb JA |
5435 | * timeout event to be satisfied. If it isn't set, then this is |
5436 | * a pure timeout request, sequence isn't used. | |
5262f567 | 5437 | */ |
8eb7e2d0 | 5438 | if (io_is_timeout_noseq(req)) { |
93bd25bb JA |
5439 | entry = ctx->timeout_list.prev; |
5440 | goto add; | |
5441 | } | |
5262f567 | 5442 | |
bfe68a22 PB |
5443 | tail = ctx->cached_cq_tail - atomic_read(&ctx->cq_timeouts); |
5444 | req->timeout.target_seq = tail + off; | |
5262f567 JA |
5445 | |
5446 | /* | |
5447 | * Insertion sort, ensuring the first entry in the list is always | |
5448 | * the one we need first. | |
5449 | */ | |
5262f567 | 5450 | list_for_each_prev(entry, &ctx->timeout_list) { |
135fcde8 PB |
5451 | struct io_kiocb *nxt = list_entry(entry, struct io_kiocb, |
5452 | timeout.list); | |
5262f567 | 5453 | |
8eb7e2d0 | 5454 | if (io_is_timeout_noseq(nxt)) |
93bd25bb | 5455 | continue; |
bfe68a22 PB |
5456 | /* nxt.seq is behind @tail, otherwise would've been completed */ |
5457 | if (off >= nxt->timeout.target_seq - tail) | |
5262f567 JA |
5458 | break; |
5459 | } | |
93bd25bb | 5460 | add: |
135fcde8 | 5461 | list_add(&req->timeout.list, entry); |
ad8a48ac JA |
5462 | data->timer.function = io_timeout_fn; |
5463 | hrtimer_start(&data->timer, timespec64_to_ktime(data->ts), data->mode); | |
5262f567 | 5464 | spin_unlock_irq(&ctx->completion_lock); |
5262f567 JA |
5465 | return 0; |
5466 | } | |
5262f567 | 5467 | |
62755e35 JA |
5468 | static bool io_cancel_cb(struct io_wq_work *work, void *data) |
5469 | { | |
5470 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); | |
5471 | ||
5472 | return req->user_data == (unsigned long) data; | |
5473 | } | |
5474 | ||
e977d6d3 | 5475 | static int io_async_cancel_one(struct io_ring_ctx *ctx, void *sqe_addr) |
62755e35 | 5476 | { |
62755e35 | 5477 | enum io_wq_cancel cancel_ret; |
62755e35 JA |
5478 | int ret = 0; |
5479 | ||
4f26bda1 | 5480 | cancel_ret = io_wq_cancel_cb(ctx->io_wq, io_cancel_cb, sqe_addr, false); |
62755e35 JA |
5481 | switch (cancel_ret) { |
5482 | case IO_WQ_CANCEL_OK: | |
5483 | ret = 0; | |
5484 | break; | |
5485 | case IO_WQ_CANCEL_RUNNING: | |
5486 | ret = -EALREADY; | |
5487 | break; | |
5488 | case IO_WQ_CANCEL_NOTFOUND: | |
5489 | ret = -ENOENT; | |
5490 | break; | |
5491 | } | |
5492 | ||
e977d6d3 JA |
5493 | return ret; |
5494 | } | |
5495 | ||
47f46768 JA |
5496 | static void io_async_find_and_cancel(struct io_ring_ctx *ctx, |
5497 | struct io_kiocb *req, __u64 sqe_addr, | |
014db007 | 5498 | int success_ret) |
47f46768 JA |
5499 | { |
5500 | unsigned long flags; | |
5501 | int ret; | |
5502 | ||
5503 | ret = io_async_cancel_one(ctx, (void *) (unsigned long) sqe_addr); | |
5504 | if (ret != -ENOENT) { | |
5505 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
5506 | goto done; | |
5507 | } | |
5508 | ||
5509 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
5510 | ret = io_timeout_cancel(ctx, sqe_addr); | |
5511 | if (ret != -ENOENT) | |
5512 | goto done; | |
5513 | ret = io_poll_cancel(ctx, sqe_addr); | |
5514 | done: | |
b0dd8a41 JA |
5515 | if (!ret) |
5516 | ret = success_ret; | |
47f46768 JA |
5517 | io_cqring_fill_event(req, ret); |
5518 | io_commit_cqring(ctx); | |
5519 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
5520 | io_cqring_ev_posted(ctx); | |
5521 | ||
4e88d6e7 JA |
5522 | if (ret < 0) |
5523 | req_set_fail_links(req); | |
014db007 | 5524 | io_put_req(req); |
47f46768 JA |
5525 | } |
5526 | ||
3529d8c2 JA |
5527 | static int io_async_cancel_prep(struct io_kiocb *req, |
5528 | const struct io_uring_sqe *sqe) | |
e977d6d3 | 5529 | { |
fbf23849 | 5530 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
e977d6d3 | 5531 | return -EINVAL; |
61710e43 DA |
5532 | if (unlikely(req->flags & (REQ_F_FIXED_FILE | REQ_F_BUFFER_SELECT))) |
5533 | return -EINVAL; | |
5534 | if (sqe->ioprio || sqe->off || sqe->len || sqe->cancel_flags) | |
e977d6d3 JA |
5535 | return -EINVAL; |
5536 | ||
fbf23849 JA |
5537 | req->cancel.addr = READ_ONCE(sqe->addr); |
5538 | return 0; | |
5539 | } | |
5540 | ||
014db007 | 5541 | static int io_async_cancel(struct io_kiocb *req) |
fbf23849 JA |
5542 | { |
5543 | struct io_ring_ctx *ctx = req->ctx; | |
fbf23849 | 5544 | |
014db007 | 5545 | io_async_find_and_cancel(ctx, req, req->cancel.addr, 0); |
5262f567 JA |
5546 | return 0; |
5547 | } | |
5548 | ||
05f3fb3c JA |
5549 | static int io_files_update_prep(struct io_kiocb *req, |
5550 | const struct io_uring_sqe *sqe) | |
5551 | { | |
6ca56f84 JA |
5552 | if (unlikely(req->ctx->flags & IORING_SETUP_SQPOLL)) |
5553 | return -EINVAL; | |
61710e43 DA |
5554 | if (unlikely(req->flags & (REQ_F_FIXED_FILE | REQ_F_BUFFER_SELECT))) |
5555 | return -EINVAL; | |
5556 | if (sqe->ioprio || sqe->rw_flags) | |
05f3fb3c JA |
5557 | return -EINVAL; |
5558 | ||
5559 | req->files_update.offset = READ_ONCE(sqe->off); | |
5560 | req->files_update.nr_args = READ_ONCE(sqe->len); | |
5561 | if (!req->files_update.nr_args) | |
5562 | return -EINVAL; | |
5563 | req->files_update.arg = READ_ONCE(sqe->addr); | |
5564 | return 0; | |
5565 | } | |
5566 | ||
229a7b63 JA |
5567 | static int io_files_update(struct io_kiocb *req, bool force_nonblock, |
5568 | struct io_comp_state *cs) | |
fbf23849 JA |
5569 | { |
5570 | struct io_ring_ctx *ctx = req->ctx; | |
05f3fb3c JA |
5571 | struct io_uring_files_update up; |
5572 | int ret; | |
fbf23849 | 5573 | |
f86cd20c | 5574 | if (force_nonblock) |
05f3fb3c | 5575 | return -EAGAIN; |
05f3fb3c JA |
5576 | |
5577 | up.offset = req->files_update.offset; | |
5578 | up.fds = req->files_update.arg; | |
5579 | ||
5580 | mutex_lock(&ctx->uring_lock); | |
5581 | ret = __io_sqe_files_update(ctx, &up, req->files_update.nr_args); | |
5582 | mutex_unlock(&ctx->uring_lock); | |
5583 | ||
5584 | if (ret < 0) | |
5585 | req_set_fail_links(req); | |
229a7b63 | 5586 | __io_req_complete(req, ret, 0, cs); |
5262f567 JA |
5587 | return 0; |
5588 | } | |
5589 | ||
bfe76559 | 5590 | static int io_req_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
f67676d1 | 5591 | { |
d625c6ee | 5592 | switch (req->opcode) { |
e781573e | 5593 | case IORING_OP_NOP: |
bfe76559 | 5594 | return 0; |
f67676d1 JA |
5595 | case IORING_OP_READV: |
5596 | case IORING_OP_READ_FIXED: | |
3a6820f2 | 5597 | case IORING_OP_READ: |
bfe76559 | 5598 | return io_read_prep(req, sqe); |
f67676d1 JA |
5599 | case IORING_OP_WRITEV: |
5600 | case IORING_OP_WRITE_FIXED: | |
3a6820f2 | 5601 | case IORING_OP_WRITE: |
bfe76559 | 5602 | return io_write_prep(req, sqe); |
0969e783 | 5603 | case IORING_OP_POLL_ADD: |
bfe76559 | 5604 | return io_poll_add_prep(req, sqe); |
0969e783 | 5605 | case IORING_OP_POLL_REMOVE: |
bfe76559 | 5606 | return io_poll_remove_prep(req, sqe); |
8ed8d3c3 | 5607 | case IORING_OP_FSYNC: |
bfe76559 | 5608 | return io_prep_fsync(req, sqe); |
8ed8d3c3 | 5609 | case IORING_OP_SYNC_FILE_RANGE: |
bfe76559 | 5610 | return io_prep_sfr(req, sqe); |
03b1230c | 5611 | case IORING_OP_SENDMSG: |
fddaface | 5612 | case IORING_OP_SEND: |
bfe76559 | 5613 | return io_sendmsg_prep(req, sqe); |
03b1230c | 5614 | case IORING_OP_RECVMSG: |
fddaface | 5615 | case IORING_OP_RECV: |
bfe76559 | 5616 | return io_recvmsg_prep(req, sqe); |
f499a021 | 5617 | case IORING_OP_CONNECT: |
bfe76559 | 5618 | return io_connect_prep(req, sqe); |
2d28390a | 5619 | case IORING_OP_TIMEOUT: |
bfe76559 | 5620 | return io_timeout_prep(req, sqe, false); |
b29472ee | 5621 | case IORING_OP_TIMEOUT_REMOVE: |
bfe76559 | 5622 | return io_timeout_remove_prep(req, sqe); |
fbf23849 | 5623 | case IORING_OP_ASYNC_CANCEL: |
bfe76559 | 5624 | return io_async_cancel_prep(req, sqe); |
2d28390a | 5625 | case IORING_OP_LINK_TIMEOUT: |
bfe76559 | 5626 | return io_timeout_prep(req, sqe, true); |
8ed8d3c3 | 5627 | case IORING_OP_ACCEPT: |
bfe76559 | 5628 | return io_accept_prep(req, sqe); |
d63d1b5e | 5629 | case IORING_OP_FALLOCATE: |
bfe76559 | 5630 | return io_fallocate_prep(req, sqe); |
15b71abe | 5631 | case IORING_OP_OPENAT: |
bfe76559 | 5632 | return io_openat_prep(req, sqe); |
b5dba59e | 5633 | case IORING_OP_CLOSE: |
bfe76559 | 5634 | return io_close_prep(req, sqe); |
05f3fb3c | 5635 | case IORING_OP_FILES_UPDATE: |
bfe76559 | 5636 | return io_files_update_prep(req, sqe); |
eddc7ef5 | 5637 | case IORING_OP_STATX: |
bfe76559 | 5638 | return io_statx_prep(req, sqe); |
4840e418 | 5639 | case IORING_OP_FADVISE: |
bfe76559 | 5640 | return io_fadvise_prep(req, sqe); |
c1ca757b | 5641 | case IORING_OP_MADVISE: |
bfe76559 | 5642 | return io_madvise_prep(req, sqe); |
cebdb986 | 5643 | case IORING_OP_OPENAT2: |
bfe76559 | 5644 | return io_openat2_prep(req, sqe); |
3e4827b0 | 5645 | case IORING_OP_EPOLL_CTL: |
bfe76559 | 5646 | return io_epoll_ctl_prep(req, sqe); |
7d67af2c | 5647 | case IORING_OP_SPLICE: |
bfe76559 | 5648 | return io_splice_prep(req, sqe); |
ddf0322d | 5649 | case IORING_OP_PROVIDE_BUFFERS: |
bfe76559 | 5650 | return io_provide_buffers_prep(req, sqe); |
067524e9 | 5651 | case IORING_OP_REMOVE_BUFFERS: |
bfe76559 | 5652 | return io_remove_buffers_prep(req, sqe); |
f2a8d5c7 | 5653 | case IORING_OP_TEE: |
bfe76559 | 5654 | return io_tee_prep(req, sqe); |
f67676d1 JA |
5655 | } |
5656 | ||
bfe76559 PB |
5657 | printk_once(KERN_WARNING "io_uring: unhandled opcode %d\n", |
5658 | req->opcode); | |
5659 | return-EINVAL; | |
5660 | } | |
5661 | ||
5662 | static int io_req_defer_prep(struct io_kiocb *req, | |
5663 | const struct io_uring_sqe *sqe) | |
5664 | { | |
5665 | int ret; | |
5666 | ||
5667 | if (!sqe) | |
5668 | return 0; | |
5669 | if (io_alloc_async_data(req)) | |
5670 | return -EAGAIN; | |
5671 | ||
5672 | ret = io_prep_work_files(req); | |
5673 | if (unlikely(ret)) | |
5674 | return ret; | |
5675 | ||
5676 | io_prep_async_work(req); | |
5677 | ||
5678 | return io_req_prep(req, sqe); | |
f67676d1 JA |
5679 | } |
5680 | ||
9cf7c104 PB |
5681 | static u32 io_get_sequence(struct io_kiocb *req) |
5682 | { | |
5683 | struct io_kiocb *pos; | |
5684 | struct io_ring_ctx *ctx = req->ctx; | |
5685 | u32 total_submitted, nr_reqs = 1; | |
5686 | ||
5687 | if (req->flags & REQ_F_LINK_HEAD) | |
5688 | list_for_each_entry(pos, &req->link_list, link_list) | |
5689 | nr_reqs++; | |
5690 | ||
5691 | total_submitted = ctx->cached_sq_head - ctx->cached_sq_dropped; | |
5692 | return total_submitted - nr_reqs; | |
5693 | } | |
5694 | ||
3529d8c2 | 5695 | static int io_req_defer(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
de0617e4 | 5696 | { |
a197f664 | 5697 | struct io_ring_ctx *ctx = req->ctx; |
27dc8338 | 5698 | struct io_defer_entry *de; |
f67676d1 | 5699 | int ret; |
9cf7c104 | 5700 | u32 seq; |
de0617e4 | 5701 | |
9d858b21 | 5702 | /* Still need defer if there is pending req in defer list. */ |
9cf7c104 PB |
5703 | if (likely(list_empty_careful(&ctx->defer_list) && |
5704 | !(req->flags & REQ_F_IO_DRAIN))) | |
5705 | return 0; | |
5706 | ||
5707 | seq = io_get_sequence(req); | |
5708 | /* Still a chance to pass the sequence check */ | |
5709 | if (!req_need_defer(req, seq) && list_empty_careful(&ctx->defer_list)) | |
de0617e4 JA |
5710 | return 0; |
5711 | ||
e8c2bc1f | 5712 | if (!req->async_data) { |
650b5481 | 5713 | ret = io_req_defer_prep(req, sqe); |
327d6d96 | 5714 | if (ret) |
650b5481 PB |
5715 | return ret; |
5716 | } | |
cbdcb435 | 5717 | io_prep_async_link(req); |
27dc8338 PB |
5718 | de = kmalloc(sizeof(*de), GFP_KERNEL); |
5719 | if (!de) | |
5720 | return -ENOMEM; | |
2d28390a | 5721 | |
de0617e4 | 5722 | spin_lock_irq(&ctx->completion_lock); |
9cf7c104 | 5723 | if (!req_need_defer(req, seq) && list_empty(&ctx->defer_list)) { |
de0617e4 | 5724 | spin_unlock_irq(&ctx->completion_lock); |
27dc8338 | 5725 | kfree(de); |
ae34817b PB |
5726 | io_queue_async_work(req); |
5727 | return -EIOCBQUEUED; | |
de0617e4 JA |
5728 | } |
5729 | ||
915967f6 | 5730 | trace_io_uring_defer(ctx, req, req->user_data); |
27dc8338 | 5731 | de->req = req; |
9cf7c104 | 5732 | de->seq = seq; |
27dc8338 | 5733 | list_add_tail(&de->list, &ctx->defer_list); |
de0617e4 JA |
5734 | spin_unlock_irq(&ctx->completion_lock); |
5735 | return -EIOCBQUEUED; | |
5736 | } | |
5737 | ||
f573d384 JA |
5738 | static void io_req_drop_files(struct io_kiocb *req) |
5739 | { | |
5740 | struct io_ring_ctx *ctx = req->ctx; | |
5741 | unsigned long flags; | |
5742 | ||
5743 | spin_lock_irqsave(&ctx->inflight_lock, flags); | |
5744 | list_del(&req->inflight_entry); | |
5745 | if (waitqueue_active(&ctx->inflight_wait)) | |
5746 | wake_up(&ctx->inflight_wait); | |
5747 | spin_unlock_irqrestore(&ctx->inflight_lock, flags); | |
5748 | req->flags &= ~REQ_F_INFLIGHT; | |
0f212204 | 5749 | put_files_struct(req->work.files); |
9b828492 | 5750 | put_nsproxy(req->work.nsproxy); |
f573d384 JA |
5751 | req->work.files = NULL; |
5752 | } | |
5753 | ||
3ca405eb | 5754 | static void __io_clean_op(struct io_kiocb *req) |
99bc4c38 | 5755 | { |
0e1b6fe3 PB |
5756 | if (req->flags & REQ_F_BUFFER_SELECTED) { |
5757 | switch (req->opcode) { | |
5758 | case IORING_OP_READV: | |
5759 | case IORING_OP_READ_FIXED: | |
5760 | case IORING_OP_READ: | |
bcda7baa | 5761 | kfree((void *)(unsigned long)req->rw.addr); |
0e1b6fe3 PB |
5762 | break; |
5763 | case IORING_OP_RECVMSG: | |
5764 | case IORING_OP_RECV: | |
bcda7baa | 5765 | kfree(req->sr_msg.kbuf); |
0e1b6fe3 PB |
5766 | break; |
5767 | } | |
5768 | req->flags &= ~REQ_F_BUFFER_SELECTED; | |
99bc4c38 PB |
5769 | } |
5770 | ||
0e1b6fe3 PB |
5771 | if (req->flags & REQ_F_NEED_CLEANUP) { |
5772 | switch (req->opcode) { | |
5773 | case IORING_OP_READV: | |
5774 | case IORING_OP_READ_FIXED: | |
5775 | case IORING_OP_READ: | |
5776 | case IORING_OP_WRITEV: | |
5777 | case IORING_OP_WRITE_FIXED: | |
e8c2bc1f JA |
5778 | case IORING_OP_WRITE: { |
5779 | struct io_async_rw *io = req->async_data; | |
5780 | if (io->free_iovec) | |
5781 | kfree(io->free_iovec); | |
0e1b6fe3 | 5782 | break; |
e8c2bc1f | 5783 | } |
0e1b6fe3 | 5784 | case IORING_OP_RECVMSG: |
e8c2bc1f JA |
5785 | case IORING_OP_SENDMSG: { |
5786 | struct io_async_msghdr *io = req->async_data; | |
5787 | if (io->iov != io->fast_iov) | |
5788 | kfree(io->iov); | |
0e1b6fe3 | 5789 | break; |
e8c2bc1f | 5790 | } |
0e1b6fe3 PB |
5791 | case IORING_OP_SPLICE: |
5792 | case IORING_OP_TEE: | |
5793 | io_put_file(req, req->splice.file_in, | |
5794 | (req->splice.flags & SPLICE_F_FD_IN_FIXED)); | |
5795 | break; | |
f3cd4850 JA |
5796 | case IORING_OP_OPENAT: |
5797 | case IORING_OP_OPENAT2: | |
5798 | if (req->open.filename) | |
5799 | putname(req->open.filename); | |
5800 | break; | |
0e1b6fe3 PB |
5801 | } |
5802 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
99bc4c38 | 5803 | } |
bb175342 | 5804 | |
f573d384 JA |
5805 | if (req->flags & REQ_F_INFLIGHT) |
5806 | io_req_drop_files(req); | |
99bc4c38 PB |
5807 | } |
5808 | ||
c1379e24 PB |
5809 | static int io_issue_sqe(struct io_kiocb *req, bool force_nonblock, |
5810 | struct io_comp_state *cs) | |
2b188cc1 | 5811 | { |
a197f664 | 5812 | struct io_ring_ctx *ctx = req->ctx; |
d625c6ee | 5813 | int ret; |
2b188cc1 | 5814 | |
d625c6ee | 5815 | switch (req->opcode) { |
2b188cc1 | 5816 | case IORING_OP_NOP: |
229a7b63 | 5817 | ret = io_nop(req, cs); |
2b188cc1 JA |
5818 | break; |
5819 | case IORING_OP_READV: | |
edafccee | 5820 | case IORING_OP_READ_FIXED: |
3a6820f2 | 5821 | case IORING_OP_READ: |
a1d7c393 | 5822 | ret = io_read(req, force_nonblock, cs); |
edafccee | 5823 | break; |
3529d8c2 | 5824 | case IORING_OP_WRITEV: |
edafccee | 5825 | case IORING_OP_WRITE_FIXED: |
3a6820f2 | 5826 | case IORING_OP_WRITE: |
a1d7c393 | 5827 | ret = io_write(req, force_nonblock, cs); |
2b188cc1 | 5828 | break; |
c992fe29 | 5829 | case IORING_OP_FSYNC: |
014db007 | 5830 | ret = io_fsync(req, force_nonblock); |
c992fe29 | 5831 | break; |
221c5eb2 | 5832 | case IORING_OP_POLL_ADD: |
014db007 | 5833 | ret = io_poll_add(req); |
221c5eb2 JA |
5834 | break; |
5835 | case IORING_OP_POLL_REMOVE: | |
fc4df999 | 5836 | ret = io_poll_remove(req); |
221c5eb2 | 5837 | break; |
5d17b4a4 | 5838 | case IORING_OP_SYNC_FILE_RANGE: |
014db007 | 5839 | ret = io_sync_file_range(req, force_nonblock); |
5d17b4a4 | 5840 | break; |
0fa03c62 | 5841 | case IORING_OP_SENDMSG: |
fddaface | 5842 | case IORING_OP_SEND: |
fddaface | 5843 | if (req->opcode == IORING_OP_SENDMSG) |
229a7b63 | 5844 | ret = io_sendmsg(req, force_nonblock, cs); |
fddaface | 5845 | else |
229a7b63 | 5846 | ret = io_send(req, force_nonblock, cs); |
0fa03c62 | 5847 | break; |
aa1fa28f | 5848 | case IORING_OP_RECVMSG: |
fddaface | 5849 | case IORING_OP_RECV: |
fddaface | 5850 | if (req->opcode == IORING_OP_RECVMSG) |
229a7b63 | 5851 | ret = io_recvmsg(req, force_nonblock, cs); |
fddaface | 5852 | else |
229a7b63 | 5853 | ret = io_recv(req, force_nonblock, cs); |
aa1fa28f | 5854 | break; |
5262f567 | 5855 | case IORING_OP_TIMEOUT: |
fc4df999 | 5856 | ret = io_timeout(req); |
5262f567 | 5857 | break; |
11365043 | 5858 | case IORING_OP_TIMEOUT_REMOVE: |
fc4df999 | 5859 | ret = io_timeout_remove(req); |
11365043 | 5860 | break; |
17f2fe35 | 5861 | case IORING_OP_ACCEPT: |
229a7b63 | 5862 | ret = io_accept(req, force_nonblock, cs); |
17f2fe35 | 5863 | break; |
f8e85cf2 | 5864 | case IORING_OP_CONNECT: |
229a7b63 | 5865 | ret = io_connect(req, force_nonblock, cs); |
f8e85cf2 | 5866 | break; |
62755e35 | 5867 | case IORING_OP_ASYNC_CANCEL: |
014db007 | 5868 | ret = io_async_cancel(req); |
62755e35 | 5869 | break; |
d63d1b5e | 5870 | case IORING_OP_FALLOCATE: |
014db007 | 5871 | ret = io_fallocate(req, force_nonblock); |
d63d1b5e | 5872 | break; |
15b71abe | 5873 | case IORING_OP_OPENAT: |
014db007 | 5874 | ret = io_openat(req, force_nonblock); |
15b71abe | 5875 | break; |
b5dba59e | 5876 | case IORING_OP_CLOSE: |
229a7b63 | 5877 | ret = io_close(req, force_nonblock, cs); |
b5dba59e | 5878 | break; |
05f3fb3c | 5879 | case IORING_OP_FILES_UPDATE: |
229a7b63 | 5880 | ret = io_files_update(req, force_nonblock, cs); |
05f3fb3c | 5881 | break; |
eddc7ef5 | 5882 | case IORING_OP_STATX: |
014db007 | 5883 | ret = io_statx(req, force_nonblock); |
eddc7ef5 | 5884 | break; |
4840e418 | 5885 | case IORING_OP_FADVISE: |
014db007 | 5886 | ret = io_fadvise(req, force_nonblock); |
4840e418 | 5887 | break; |
c1ca757b | 5888 | case IORING_OP_MADVISE: |
014db007 | 5889 | ret = io_madvise(req, force_nonblock); |
c1ca757b | 5890 | break; |
cebdb986 | 5891 | case IORING_OP_OPENAT2: |
014db007 | 5892 | ret = io_openat2(req, force_nonblock); |
cebdb986 | 5893 | break; |
3e4827b0 | 5894 | case IORING_OP_EPOLL_CTL: |
229a7b63 | 5895 | ret = io_epoll_ctl(req, force_nonblock, cs); |
3e4827b0 | 5896 | break; |
7d67af2c | 5897 | case IORING_OP_SPLICE: |
014db007 | 5898 | ret = io_splice(req, force_nonblock); |
7d67af2c | 5899 | break; |
ddf0322d | 5900 | case IORING_OP_PROVIDE_BUFFERS: |
229a7b63 | 5901 | ret = io_provide_buffers(req, force_nonblock, cs); |
ddf0322d | 5902 | break; |
067524e9 | 5903 | case IORING_OP_REMOVE_BUFFERS: |
229a7b63 | 5904 | ret = io_remove_buffers(req, force_nonblock, cs); |
3e4827b0 | 5905 | break; |
f2a8d5c7 | 5906 | case IORING_OP_TEE: |
f2a8d5c7 PB |
5907 | ret = io_tee(req, force_nonblock); |
5908 | break; | |
2b188cc1 JA |
5909 | default: |
5910 | ret = -EINVAL; | |
5911 | break; | |
5912 | } | |
5913 | ||
def596e9 JA |
5914 | if (ret) |
5915 | return ret; | |
5916 | ||
b532576e JA |
5917 | /* If the op doesn't have a file, we're not polling for it */ |
5918 | if ((ctx->flags & IORING_SETUP_IOPOLL) && req->file) { | |
11ba820b JA |
5919 | const bool in_async = io_wq_current_is_worker(); |
5920 | ||
11ba820b JA |
5921 | /* workqueue context doesn't hold uring_lock, grab it now */ |
5922 | if (in_async) | |
5923 | mutex_lock(&ctx->uring_lock); | |
5924 | ||
def596e9 | 5925 | io_iopoll_req_issued(req); |
11ba820b JA |
5926 | |
5927 | if (in_async) | |
5928 | mutex_unlock(&ctx->uring_lock); | |
def596e9 JA |
5929 | } |
5930 | ||
5931 | return 0; | |
2b188cc1 JA |
5932 | } |
5933 | ||
f4db7182 | 5934 | static struct io_wq_work *io_wq_submit_work(struct io_wq_work *work) |
2b188cc1 JA |
5935 | { |
5936 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); | |
6df1db6b | 5937 | struct io_kiocb *timeout; |
561fb04a | 5938 | int ret = 0; |
2b188cc1 | 5939 | |
6df1db6b PB |
5940 | timeout = io_prep_linked_timeout(req); |
5941 | if (timeout) | |
5942 | io_queue_linked_timeout(timeout); | |
d4c81f38 | 5943 | |
0c9d5ccd JA |
5944 | /* if NO_CANCEL is set, we must still run the work */ |
5945 | if ((work->flags & (IO_WQ_WORK_CANCEL|IO_WQ_WORK_NO_CANCEL)) == | |
5946 | IO_WQ_WORK_CANCEL) { | |
561fb04a | 5947 | ret = -ECANCELED; |
0c9d5ccd | 5948 | } |
31b51510 | 5949 | |
561fb04a | 5950 | if (!ret) { |
561fb04a | 5951 | do { |
c1379e24 | 5952 | ret = io_issue_sqe(req, false, NULL); |
561fb04a JA |
5953 | /* |
5954 | * We can get EAGAIN for polled IO even though we're | |
5955 | * forcing a sync submission from here, since we can't | |
5956 | * wait for request slots on the block side. | |
5957 | */ | |
5958 | if (ret != -EAGAIN) | |
5959 | break; | |
5960 | cond_resched(); | |
5961 | } while (1); | |
5962 | } | |
31b51510 | 5963 | |
561fb04a | 5964 | if (ret) { |
4e88d6e7 | 5965 | req_set_fail_links(req); |
e1e16097 | 5966 | io_req_complete(req, ret); |
edafccee | 5967 | } |
2b188cc1 | 5968 | |
f4db7182 | 5969 | return io_steal_work(req); |
2b188cc1 JA |
5970 | } |
5971 | ||
65e19f54 JA |
5972 | static inline struct file *io_file_from_index(struct io_ring_ctx *ctx, |
5973 | int index) | |
5974 | { | |
5975 | struct fixed_file_table *table; | |
5976 | ||
05f3fb3c | 5977 | table = &ctx->file_data->table[index >> IORING_FILE_TABLE_SHIFT]; |
84695089 | 5978 | return table->files[index & IORING_FILE_TABLE_MASK]; |
65e19f54 JA |
5979 | } |
5980 | ||
8da11c19 PB |
5981 | static int io_file_get(struct io_submit_state *state, struct io_kiocb *req, |
5982 | int fd, struct file **out_file, bool fixed) | |
09bb8394 | 5983 | { |
a197f664 | 5984 | struct io_ring_ctx *ctx = req->ctx; |
8da11c19 | 5985 | struct file *file; |
09bb8394 | 5986 | |
8da11c19 | 5987 | if (fixed) { |
05f3fb3c | 5988 | if (unlikely(!ctx->file_data || |
09bb8394 JA |
5989 | (unsigned) fd >= ctx->nr_user_files)) |
5990 | return -EBADF; | |
b7620121 | 5991 | fd = array_index_nospec(fd, ctx->nr_user_files); |
8da11c19 | 5992 | file = io_file_from_index(ctx, fd); |
fd2206e4 JA |
5993 | if (file) { |
5994 | req->fixed_file_refs = ctx->file_data->cur_refs; | |
5995 | percpu_ref_get(req->fixed_file_refs); | |
5996 | } | |
09bb8394 | 5997 | } else { |
c826bd7a | 5998 | trace_io_uring_file_get(ctx, fd); |
8da11c19 | 5999 | file = __io_file_get(state, fd); |
09bb8394 JA |
6000 | } |
6001 | ||
fd2206e4 JA |
6002 | if (file || io_op_defs[req->opcode].needs_file_no_error) { |
6003 | *out_file = file; | |
6004 | return 0; | |
6005 | } | |
6006 | return -EBADF; | |
09bb8394 JA |
6007 | } |
6008 | ||
8da11c19 | 6009 | static int io_req_set_file(struct io_submit_state *state, struct io_kiocb *req, |
63ff8223 | 6010 | int fd) |
8da11c19 | 6011 | { |
8da11c19 PB |
6012 | bool fixed; |
6013 | ||
63ff8223 | 6014 | fixed = (req->flags & REQ_F_FIXED_FILE) != 0; |
0cdaf760 | 6015 | if (unlikely(!fixed && io_async_submit(req->ctx))) |
8da11c19 PB |
6016 | return -EBADF; |
6017 | ||
6018 | return io_file_get(state, req, fd, &req->file, fixed); | |
6019 | } | |
6020 | ||
a197f664 | 6021 | static int io_grab_files(struct io_kiocb *req) |
fcb323cc | 6022 | { |
a197f664 | 6023 | struct io_ring_ctx *ctx = req->ctx; |
fcb323cc | 6024 | |
f56040b8 PB |
6025 | io_req_init_async(req); |
6026 | ||
5b0bbee4 | 6027 | if (req->work.files || (req->flags & REQ_F_NO_FILE_TABLE)) |
f86cd20c | 6028 | return 0; |
b5dba59e | 6029 | |
0f212204 | 6030 | req->work.files = get_files_struct(current); |
9b828492 JA |
6031 | get_nsproxy(current->nsproxy); |
6032 | req->work.nsproxy = current->nsproxy; | |
0f212204 JA |
6033 | req->flags |= REQ_F_INFLIGHT; |
6034 | ||
fcb323cc | 6035 | spin_lock_irq(&ctx->inflight_lock); |
0f212204 | 6036 | list_add(&req->inflight_entry, &ctx->inflight_list); |
fcb323cc | 6037 | spin_unlock_irq(&ctx->inflight_lock); |
0f212204 | 6038 | return 0; |
fcb323cc JA |
6039 | } |
6040 | ||
f56040b8 PB |
6041 | static inline int io_prep_work_files(struct io_kiocb *req) |
6042 | { | |
6043 | if (!io_op_defs[req->opcode].file_table) | |
6044 | return 0; | |
6045 | return io_grab_files(req); | |
6046 | } | |
6047 | ||
2665abfd | 6048 | static enum hrtimer_restart io_link_timeout_fn(struct hrtimer *timer) |
2b188cc1 | 6049 | { |
ad8a48ac JA |
6050 | struct io_timeout_data *data = container_of(timer, |
6051 | struct io_timeout_data, timer); | |
6052 | struct io_kiocb *req = data->req; | |
2665abfd JA |
6053 | struct io_ring_ctx *ctx = req->ctx; |
6054 | struct io_kiocb *prev = NULL; | |
6055 | unsigned long flags; | |
2665abfd JA |
6056 | |
6057 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
6058 | ||
6059 | /* | |
6060 | * We don't expect the list to be empty, that will only happen if we | |
6061 | * race with the completion of the linked work. | |
6062 | */ | |
4493233e PB |
6063 | if (!list_empty(&req->link_list)) { |
6064 | prev = list_entry(req->link_list.prev, struct io_kiocb, | |
6065 | link_list); | |
5d960724 | 6066 | if (refcount_inc_not_zero(&prev->refs)) { |
4493233e | 6067 | list_del_init(&req->link_list); |
5d960724 JA |
6068 | prev->flags &= ~REQ_F_LINK_TIMEOUT; |
6069 | } else | |
76a46e06 | 6070 | prev = NULL; |
2665abfd JA |
6071 | } |
6072 | ||
6073 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
6074 | ||
6075 | if (prev) { | |
4e88d6e7 | 6076 | req_set_fail_links(prev); |
014db007 | 6077 | io_async_find_and_cancel(ctx, req, prev->user_data, -ETIME); |
76a46e06 | 6078 | io_put_req(prev); |
47f46768 | 6079 | } else { |
e1e16097 | 6080 | io_req_complete(req, -ETIME); |
2665abfd | 6081 | } |
2665abfd JA |
6082 | return HRTIMER_NORESTART; |
6083 | } | |
6084 | ||
7271ef3a | 6085 | static void __io_queue_linked_timeout(struct io_kiocb *req) |
2665abfd | 6086 | { |
76a46e06 JA |
6087 | /* |
6088 | * If the list is now empty, then our linked request finished before | |
6089 | * we got a chance to setup the timer | |
6090 | */ | |
4493233e | 6091 | if (!list_empty(&req->link_list)) { |
e8c2bc1f | 6092 | struct io_timeout_data *data = req->async_data; |
94ae5e77 | 6093 | |
ad8a48ac JA |
6094 | data->timer.function = io_link_timeout_fn; |
6095 | hrtimer_start(&data->timer, timespec64_to_ktime(data->ts), | |
6096 | data->mode); | |
2665abfd | 6097 | } |
7271ef3a JA |
6098 | } |
6099 | ||
6100 | static void io_queue_linked_timeout(struct io_kiocb *req) | |
6101 | { | |
6102 | struct io_ring_ctx *ctx = req->ctx; | |
6103 | ||
6104 | spin_lock_irq(&ctx->completion_lock); | |
6105 | __io_queue_linked_timeout(req); | |
76a46e06 | 6106 | spin_unlock_irq(&ctx->completion_lock); |
2665abfd | 6107 | |
2665abfd | 6108 | /* drop submission reference */ |
76a46e06 JA |
6109 | io_put_req(req); |
6110 | } | |
2665abfd | 6111 | |
ad8a48ac | 6112 | static struct io_kiocb *io_prep_linked_timeout(struct io_kiocb *req) |
2665abfd JA |
6113 | { |
6114 | struct io_kiocb *nxt; | |
6115 | ||
dea3b49c | 6116 | if (!(req->flags & REQ_F_LINK_HEAD)) |
2665abfd | 6117 | return NULL; |
6df1db6b | 6118 | if (req->flags & REQ_F_LINK_TIMEOUT) |
d7718a9d | 6119 | return NULL; |
2665abfd | 6120 | |
4493233e PB |
6121 | nxt = list_first_entry_or_null(&req->link_list, struct io_kiocb, |
6122 | link_list); | |
d625c6ee | 6123 | if (!nxt || nxt->opcode != IORING_OP_LINK_TIMEOUT) |
76a46e06 | 6124 | return NULL; |
2665abfd | 6125 | |
76a46e06 | 6126 | req->flags |= REQ_F_LINK_TIMEOUT; |
76a46e06 | 6127 | return nxt; |
2665abfd JA |
6128 | } |
6129 | ||
c1379e24 | 6130 | static void __io_queue_sqe(struct io_kiocb *req, struct io_comp_state *cs) |
2b188cc1 | 6131 | { |
4a0a7a18 | 6132 | struct io_kiocb *linked_timeout; |
4bc4494e | 6133 | struct io_kiocb *nxt; |
193155c8 | 6134 | const struct cred *old_creds = NULL; |
e0c5c576 | 6135 | int ret; |
2b188cc1 | 6136 | |
4a0a7a18 JA |
6137 | again: |
6138 | linked_timeout = io_prep_linked_timeout(req); | |
6139 | ||
7cdaf587 XW |
6140 | if ((req->flags & REQ_F_WORK_INITIALIZED) && req->work.creds && |
6141 | req->work.creds != current_cred()) { | |
193155c8 JA |
6142 | if (old_creds) |
6143 | revert_creds(old_creds); | |
6144 | if (old_creds == req->work.creds) | |
6145 | old_creds = NULL; /* restored original creds */ | |
6146 | else | |
6147 | old_creds = override_creds(req->work.creds); | |
6148 | } | |
6149 | ||
c1379e24 | 6150 | ret = io_issue_sqe(req, true, cs); |
491381ce JA |
6151 | |
6152 | /* | |
6153 | * We async punt it if the file wasn't marked NOWAIT, or if the file | |
6154 | * doesn't support non-blocking read/write attempts | |
6155 | */ | |
24c74678 | 6156 | if (ret == -EAGAIN && !(req->flags & REQ_F_NOWAIT)) { |
f063c547 | 6157 | if (!io_arm_poll_handler(req)) { |
86a761f8 | 6158 | punt: |
f063c547 PB |
6159 | ret = io_prep_work_files(req); |
6160 | if (unlikely(ret)) | |
bbad27b2 | 6161 | goto err; |
f063c547 PB |
6162 | /* |
6163 | * Queued up for async execution, worker will release | |
6164 | * submit reference when the iocb is actually submitted. | |
6165 | */ | |
6166 | io_queue_async_work(req); | |
2b188cc1 | 6167 | } |
bbad27b2 | 6168 | |
f063c547 PB |
6169 | if (linked_timeout) |
6170 | io_queue_linked_timeout(linked_timeout); | |
4bc4494e | 6171 | goto exit; |
2b188cc1 | 6172 | } |
e65ef56d | 6173 | |
652532ad | 6174 | if (unlikely(ret)) { |
fcb323cc | 6175 | err: |
652532ad PB |
6176 | /* un-prep timeout, so it'll be killed as any other linked */ |
6177 | req->flags &= ~REQ_F_LINK_TIMEOUT; | |
4e88d6e7 | 6178 | req_set_fail_links(req); |
e65ef56d | 6179 | io_put_req(req); |
e1e16097 | 6180 | io_req_complete(req, ret); |
652532ad | 6181 | goto exit; |
9e645e11 | 6182 | } |
652532ad PB |
6183 | |
6184 | /* drop submission reference */ | |
6185 | nxt = io_put_req_find_next(req); | |
6186 | if (linked_timeout) | |
6187 | io_queue_linked_timeout(linked_timeout); | |
6188 | ||
4a0a7a18 JA |
6189 | if (nxt) { |
6190 | req = nxt; | |
86a761f8 PB |
6191 | |
6192 | if (req->flags & REQ_F_FORCE_ASYNC) | |
6193 | goto punt; | |
4a0a7a18 JA |
6194 | goto again; |
6195 | } | |
4bc4494e | 6196 | exit: |
193155c8 JA |
6197 | if (old_creds) |
6198 | revert_creds(old_creds); | |
2b188cc1 JA |
6199 | } |
6200 | ||
f13fad7b JA |
6201 | static void io_queue_sqe(struct io_kiocb *req, const struct io_uring_sqe *sqe, |
6202 | struct io_comp_state *cs) | |
4fe2c963 JL |
6203 | { |
6204 | int ret; | |
6205 | ||
3529d8c2 | 6206 | ret = io_req_defer(req, sqe); |
4fe2c963 JL |
6207 | if (ret) { |
6208 | if (ret != -EIOCBQUEUED) { | |
1118591a | 6209 | fail_req: |
4e88d6e7 | 6210 | req_set_fail_links(req); |
e1e16097 JA |
6211 | io_put_req(req); |
6212 | io_req_complete(req, ret); | |
4fe2c963 | 6213 | } |
2550878f | 6214 | } else if (req->flags & REQ_F_FORCE_ASYNC) { |
e8c2bc1f | 6215 | if (!req->async_data) { |
bd2ab18a | 6216 | ret = io_req_defer_prep(req, sqe); |
327d6d96 | 6217 | if (unlikely(ret)) |
bd2ab18a PB |
6218 | goto fail_req; |
6219 | } | |
6220 | ||
ce35a47a JA |
6221 | /* |
6222 | * Never try inline submit of IOSQE_ASYNC is set, go straight | |
6223 | * to async execution. | |
6224 | */ | |
3e863ea3 | 6225 | io_req_init_async(req); |
ce35a47a JA |
6226 | req->work.flags |= IO_WQ_WORK_CONCURRENT; |
6227 | io_queue_async_work(req); | |
6228 | } else { | |
c1379e24 PB |
6229 | if (sqe) { |
6230 | ret = io_req_prep(req, sqe); | |
6231 | if (unlikely(ret)) | |
6232 | goto fail_req; | |
6233 | } | |
6234 | __io_queue_sqe(req, cs); | |
ce35a47a | 6235 | } |
4fe2c963 JL |
6236 | } |
6237 | ||
f13fad7b JA |
6238 | static inline void io_queue_link_head(struct io_kiocb *req, |
6239 | struct io_comp_state *cs) | |
4fe2c963 | 6240 | { |
94ae5e77 | 6241 | if (unlikely(req->flags & REQ_F_FAIL_LINK)) { |
e1e16097 JA |
6242 | io_put_req(req); |
6243 | io_req_complete(req, -ECANCELED); | |
1b4a51b6 | 6244 | } else |
f13fad7b | 6245 | io_queue_sqe(req, NULL, cs); |
4fe2c963 JL |
6246 | } |
6247 | ||
1d4240cc | 6248 | static int io_submit_sqe(struct io_kiocb *req, const struct io_uring_sqe *sqe, |
f13fad7b | 6249 | struct io_kiocb **link, struct io_comp_state *cs) |
9e645e11 | 6250 | { |
a197f664 | 6251 | struct io_ring_ctx *ctx = req->ctx; |
ef4ff581 | 6252 | int ret; |
9e645e11 | 6253 | |
9e645e11 JA |
6254 | /* |
6255 | * If we already have a head request, queue this one for async | |
6256 | * submittal once the head completes. If we don't have a head but | |
6257 | * IOSQE_IO_LINK is set in the sqe, start a new head. This one will be | |
6258 | * submitted sync once the chain is complete. If none of those | |
6259 | * conditions are true (normal request), then just queue it. | |
6260 | */ | |
6261 | if (*link) { | |
9d76377f | 6262 | struct io_kiocb *head = *link; |
4e88d6e7 | 6263 | |
8cdf2193 PB |
6264 | /* |
6265 | * Taking sequential execution of a link, draining both sides | |
6266 | * of the link also fullfils IOSQE_IO_DRAIN semantics for all | |
6267 | * requests in the link. So, it drains the head and the | |
6268 | * next after the link request. The last one is done via | |
6269 | * drain_next flag to persist the effect across calls. | |
6270 | */ | |
ef4ff581 | 6271 | if (req->flags & REQ_F_IO_DRAIN) { |
711be031 PB |
6272 | head->flags |= REQ_F_IO_DRAIN; |
6273 | ctx->drain_next = 1; | |
6274 | } | |
3529d8c2 | 6275 | ret = io_req_defer_prep(req, sqe); |
327d6d96 | 6276 | if (unlikely(ret)) { |
4e88d6e7 | 6277 | /* fail even hard links since we don't submit */ |
9d76377f | 6278 | head->flags |= REQ_F_FAIL_LINK; |
1d4240cc | 6279 | return ret; |
2d28390a | 6280 | } |
9d76377f PB |
6281 | trace_io_uring_link(ctx, req, head); |
6282 | list_add_tail(&req->link_list, &head->link_list); | |
32fe525b PB |
6283 | |
6284 | /* last request of a link, enqueue the link */ | |
ef4ff581 | 6285 | if (!(req->flags & (REQ_F_LINK | REQ_F_HARDLINK))) { |
f13fad7b | 6286 | io_queue_link_head(head, cs); |
32fe525b PB |
6287 | *link = NULL; |
6288 | } | |
9e645e11 | 6289 | } else { |
711be031 PB |
6290 | if (unlikely(ctx->drain_next)) { |
6291 | req->flags |= REQ_F_IO_DRAIN; | |
ef4ff581 | 6292 | ctx->drain_next = 0; |
711be031 | 6293 | } |
ef4ff581 | 6294 | if (req->flags & (REQ_F_LINK | REQ_F_HARDLINK)) { |
dea3b49c | 6295 | req->flags |= REQ_F_LINK_HEAD; |
711be031 | 6296 | INIT_LIST_HEAD(&req->link_list); |
f1d96a8f | 6297 | |
711be031 | 6298 | ret = io_req_defer_prep(req, sqe); |
327d6d96 | 6299 | if (unlikely(ret)) |
711be031 PB |
6300 | req->flags |= REQ_F_FAIL_LINK; |
6301 | *link = req; | |
6302 | } else { | |
f13fad7b | 6303 | io_queue_sqe(req, sqe, cs); |
711be031 | 6304 | } |
9e645e11 | 6305 | } |
2e6e1fde | 6306 | |
1d4240cc | 6307 | return 0; |
9e645e11 JA |
6308 | } |
6309 | ||
9a56a232 JA |
6310 | /* |
6311 | * Batched submission is done, ensure local IO is flushed out. | |
6312 | */ | |
6313 | static void io_submit_state_end(struct io_submit_state *state) | |
6314 | { | |
f13fad7b JA |
6315 | if (!list_empty(&state->comp.list)) |
6316 | io_submit_flush_completions(&state->comp); | |
9a56a232 | 6317 | blk_finish_plug(&state->plug); |
9f13c35b | 6318 | io_state_file_put(state); |
2579f913 | 6319 | if (state->free_reqs) |
6c8a3134 | 6320 | kmem_cache_free_bulk(req_cachep, state->free_reqs, state->reqs); |
9a56a232 JA |
6321 | } |
6322 | ||
6323 | /* | |
6324 | * Start submission side cache. | |
6325 | */ | |
6326 | static void io_submit_state_start(struct io_submit_state *state, | |
013538bd | 6327 | struct io_ring_ctx *ctx, unsigned int max_ios) |
9a56a232 JA |
6328 | { |
6329 | blk_start_plug(&state->plug); | |
013538bd JA |
6330 | state->comp.nr = 0; |
6331 | INIT_LIST_HEAD(&state->comp.list); | |
6332 | state->comp.ctx = ctx; | |
2579f913 | 6333 | state->free_reqs = 0; |
9a56a232 JA |
6334 | state->file = NULL; |
6335 | state->ios_left = max_ios; | |
6336 | } | |
6337 | ||
2b188cc1 JA |
6338 | static void io_commit_sqring(struct io_ring_ctx *ctx) |
6339 | { | |
75b28aff | 6340 | struct io_rings *rings = ctx->rings; |
2b188cc1 | 6341 | |
caf582c6 PB |
6342 | /* |
6343 | * Ensure any loads from the SQEs are done at this point, | |
6344 | * since once we write the new head, the application could | |
6345 | * write new data to them. | |
6346 | */ | |
6347 | smp_store_release(&rings->sq.head, ctx->cached_sq_head); | |
2b188cc1 JA |
6348 | } |
6349 | ||
2b188cc1 | 6350 | /* |
3529d8c2 | 6351 | * Fetch an sqe, if one is available. Note that sqe_ptr will point to memory |
2b188cc1 JA |
6352 | * that is mapped by userspace. This means that care needs to be taken to |
6353 | * ensure that reads are stable, as we cannot rely on userspace always | |
6354 | * being a good citizen. If members of the sqe are validated and then later | |
6355 | * used, it's important that those reads are done through READ_ONCE() to | |
6356 | * prevent a re-load down the line. | |
6357 | */ | |
709b302f | 6358 | static const struct io_uring_sqe *io_get_sqe(struct io_ring_ctx *ctx) |
2b188cc1 | 6359 | { |
75b28aff | 6360 | u32 *sq_array = ctx->sq_array; |
2b188cc1 JA |
6361 | unsigned head; |
6362 | ||
6363 | /* | |
6364 | * The cached sq head (or cq tail) serves two purposes: | |
6365 | * | |
6366 | * 1) allows us to batch the cost of updating the user visible | |
6367 | * head updates. | |
6368 | * 2) allows the kernel side to track the head on its own, even | |
6369 | * though the application is the one updating it. | |
6370 | */ | |
ee7d46d9 | 6371 | head = READ_ONCE(sq_array[ctx->cached_sq_head & ctx->sq_mask]); |
709b302f PB |
6372 | if (likely(head < ctx->sq_entries)) |
6373 | return &ctx->sq_sqes[head]; | |
2b188cc1 JA |
6374 | |
6375 | /* drop invalid entries */ | |
498ccd9e | 6376 | ctx->cached_sq_dropped++; |
ee7d46d9 | 6377 | WRITE_ONCE(ctx->rings->sq_dropped, ctx->cached_sq_dropped); |
709b302f PB |
6378 | return NULL; |
6379 | } | |
6380 | ||
6381 | static inline void io_consume_sqe(struct io_ring_ctx *ctx) | |
6382 | { | |
6383 | ctx->cached_sq_head++; | |
2b188cc1 JA |
6384 | } |
6385 | ||
21b55dbc SG |
6386 | /* |
6387 | * Check SQE restrictions (opcode and flags). | |
6388 | * | |
6389 | * Returns 'true' if SQE is allowed, 'false' otherwise. | |
6390 | */ | |
6391 | static inline bool io_check_restriction(struct io_ring_ctx *ctx, | |
6392 | struct io_kiocb *req, | |
6393 | unsigned int sqe_flags) | |
6394 | { | |
6395 | if (!ctx->restricted) | |
6396 | return true; | |
6397 | ||
6398 | if (!test_bit(req->opcode, ctx->restrictions.sqe_op)) | |
6399 | return false; | |
6400 | ||
6401 | if ((sqe_flags & ctx->restrictions.sqe_flags_required) != | |
6402 | ctx->restrictions.sqe_flags_required) | |
6403 | return false; | |
6404 | ||
6405 | if (sqe_flags & ~(ctx->restrictions.sqe_flags_allowed | | |
6406 | ctx->restrictions.sqe_flags_required)) | |
6407 | return false; | |
6408 | ||
6409 | return true; | |
6410 | } | |
6411 | ||
ef4ff581 PB |
6412 | #define SQE_VALID_FLAGS (IOSQE_FIXED_FILE|IOSQE_IO_DRAIN|IOSQE_IO_LINK| \ |
6413 | IOSQE_IO_HARDLINK | IOSQE_ASYNC | \ | |
6414 | IOSQE_BUFFER_SELECT) | |
6415 | ||
6416 | static int io_init_req(struct io_ring_ctx *ctx, struct io_kiocb *req, | |
6417 | const struct io_uring_sqe *sqe, | |
0cdaf760 | 6418 | struct io_submit_state *state) |
0553b8bd | 6419 | { |
ef4ff581 | 6420 | unsigned int sqe_flags; |
63ff8223 | 6421 | int id; |
ef4ff581 | 6422 | |
0553b8bd PB |
6423 | req->opcode = READ_ONCE(sqe->opcode); |
6424 | req->user_data = READ_ONCE(sqe->user_data); | |
e8c2bc1f | 6425 | req->async_data = NULL; |
0553b8bd PB |
6426 | req->file = NULL; |
6427 | req->ctx = ctx; | |
6428 | req->flags = 0; | |
6429 | /* one is dropped after submission, the other at completion */ | |
6430 | refcount_set(&req->refs, 2); | |
4dd2824d | 6431 | req->task = current; |
0553b8bd | 6432 | req->result = 0; |
ef4ff581 PB |
6433 | |
6434 | if (unlikely(req->opcode >= IORING_OP_LAST)) | |
6435 | return -EINVAL; | |
6436 | ||
9d8426a0 JA |
6437 | if (unlikely(io_sq_thread_acquire_mm(ctx, req))) |
6438 | return -EFAULT; | |
ef4ff581 PB |
6439 | |
6440 | sqe_flags = READ_ONCE(sqe->flags); | |
6441 | /* enforce forwards compatibility on users */ | |
6442 | if (unlikely(sqe_flags & ~SQE_VALID_FLAGS)) | |
6443 | return -EINVAL; | |
6444 | ||
21b55dbc SG |
6445 | if (unlikely(!io_check_restriction(ctx, req, sqe_flags))) |
6446 | return -EACCES; | |
6447 | ||
ef4ff581 PB |
6448 | if ((sqe_flags & IOSQE_BUFFER_SELECT) && |
6449 | !io_op_defs[req->opcode].buffer_select) | |
6450 | return -EOPNOTSUPP; | |
6451 | ||
6452 | id = READ_ONCE(sqe->personality); | |
6453 | if (id) { | |
7cdaf587 | 6454 | io_req_init_async(req); |
ef4ff581 PB |
6455 | req->work.creds = idr_find(&ctx->personality_idr, id); |
6456 | if (unlikely(!req->work.creds)) | |
6457 | return -EINVAL; | |
6458 | get_cred(req->work.creds); | |
6459 | } | |
6460 | ||
6461 | /* same numerical values with corresponding REQ_F_*, safe to copy */ | |
c11368a5 | 6462 | req->flags |= sqe_flags; |
ef4ff581 | 6463 | |
63ff8223 JA |
6464 | if (!io_op_defs[req->opcode].needs_file) |
6465 | return 0; | |
6466 | ||
6467 | return io_req_set_file(state, req, READ_ONCE(sqe->fd)); | |
0553b8bd PB |
6468 | } |
6469 | ||
0f212204 | 6470 | static int io_submit_sqes(struct io_ring_ctx *ctx, unsigned int nr) |
6c271ce2 | 6471 | { |
ac8691c4 | 6472 | struct io_submit_state state; |
9e645e11 | 6473 | struct io_kiocb *link = NULL; |
9e645e11 | 6474 | int i, submitted = 0; |
6c271ce2 | 6475 | |
c4a2ed72 | 6476 | /* if we have a backlog and couldn't flush it all, return BUSY */ |
ad3eb2c8 JA |
6477 | if (test_bit(0, &ctx->sq_check_overflow)) { |
6478 | if (!list_empty(&ctx->cq_overflow_list) && | |
e6c8aa9a | 6479 | !io_cqring_overflow_flush(ctx, false, NULL, NULL)) |
ad3eb2c8 JA |
6480 | return -EBUSY; |
6481 | } | |
6c271ce2 | 6482 | |
ee7d46d9 PB |
6483 | /* make sure SQ entry isn't read before tail */ |
6484 | nr = min3(nr, ctx->sq_entries, io_sqring_entries(ctx)); | |
9ef4f124 | 6485 | |
2b85edfc PB |
6486 | if (!percpu_ref_tryget_many(&ctx->refs, nr)) |
6487 | return -EAGAIN; | |
6c271ce2 | 6488 | |
faf7b51c JA |
6489 | atomic_long_add(nr, ¤t->io_uring->req_issue); |
6490 | refcount_add(nr, ¤t->usage); | |
6491 | ||
013538bd | 6492 | io_submit_state_start(&state, ctx, nr); |
6c271ce2 JA |
6493 | |
6494 | for (i = 0; i < nr; i++) { | |
3529d8c2 | 6495 | const struct io_uring_sqe *sqe; |
196be95c | 6496 | struct io_kiocb *req; |
1cb1edb2 | 6497 | int err; |
fb5ccc98 | 6498 | |
b1e50e54 PB |
6499 | sqe = io_get_sqe(ctx); |
6500 | if (unlikely(!sqe)) { | |
6501 | io_consume_sqe(ctx); | |
6502 | break; | |
6503 | } | |
ac8691c4 | 6504 | req = io_alloc_req(ctx, &state); |
196be95c PB |
6505 | if (unlikely(!req)) { |
6506 | if (!submitted) | |
6507 | submitted = -EAGAIN; | |
fb5ccc98 | 6508 | break; |
196be95c | 6509 | } |
fb5ccc98 | 6510 | |
ac8691c4 | 6511 | err = io_init_req(ctx, req, sqe, &state); |
709b302f | 6512 | io_consume_sqe(ctx); |
d3656344 JA |
6513 | /* will complete beyond this point, count as submitted */ |
6514 | submitted++; | |
6515 | ||
ef4ff581 | 6516 | if (unlikely(err)) { |
1cb1edb2 | 6517 | fail_req: |
e1e16097 JA |
6518 | io_put_req(req); |
6519 | io_req_complete(req, err); | |
196be95c PB |
6520 | break; |
6521 | } | |
fb5ccc98 | 6522 | |
354420f7 | 6523 | trace_io_uring_submit_sqe(ctx, req->opcode, req->user_data, |
0cdaf760 | 6524 | true, io_async_submit(ctx)); |
f13fad7b | 6525 | err = io_submit_sqe(req, sqe, &link, &state.comp); |
1d4240cc PB |
6526 | if (err) |
6527 | goto fail_req; | |
6c271ce2 JA |
6528 | } |
6529 | ||
9466f437 PB |
6530 | if (unlikely(submitted != nr)) { |
6531 | int ref_used = (submitted == -EAGAIN) ? 0 : submitted; | |
6532 | ||
6533 | percpu_ref_put_many(&ctx->refs, nr - ref_used); | |
faf7b51c JA |
6534 | atomic_long_sub(nr - ref_used, ¤t->io_uring->req_issue); |
6535 | put_task_struct_many(current, nr - ref_used); | |
9466f437 | 6536 | } |
9e645e11 | 6537 | if (link) |
f13fad7b | 6538 | io_queue_link_head(link, &state.comp); |
ac8691c4 | 6539 | io_submit_state_end(&state); |
6c271ce2 | 6540 | |
ae9428ca PB |
6541 | /* Commit SQ ring head once we've consumed and submitted all SQEs */ |
6542 | io_commit_sqring(ctx); | |
6543 | ||
6c271ce2 JA |
6544 | return submitted; |
6545 | } | |
6546 | ||
23b3628e XW |
6547 | static inline void io_ring_set_wakeup_flag(struct io_ring_ctx *ctx) |
6548 | { | |
6549 | /* Tell userspace we may need a wakeup call */ | |
6550 | spin_lock_irq(&ctx->completion_lock); | |
6551 | ctx->rings->sq_flags |= IORING_SQ_NEED_WAKEUP; | |
6552 | spin_unlock_irq(&ctx->completion_lock); | |
6553 | } | |
6554 | ||
6555 | static inline void io_ring_clear_wakeup_flag(struct io_ring_ctx *ctx) | |
6556 | { | |
6557 | spin_lock_irq(&ctx->completion_lock); | |
6558 | ctx->rings->sq_flags &= ~IORING_SQ_NEED_WAKEUP; | |
6559 | spin_unlock_irq(&ctx->completion_lock); | |
6560 | } | |
6561 | ||
3f0e64d0 JA |
6562 | static int io_sq_wake_function(struct wait_queue_entry *wqe, unsigned mode, |
6563 | int sync, void *key) | |
6564 | { | |
6565 | struct io_ring_ctx *ctx = container_of(wqe, struct io_ring_ctx, sqo_wait_entry); | |
6566 | int ret; | |
6567 | ||
6568 | ret = autoremove_wake_function(wqe, mode, sync, key); | |
6569 | if (ret) { | |
6570 | unsigned long flags; | |
6571 | ||
6572 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
6573 | ctx->rings->sq_flags &= ~IORING_SQ_NEED_WAKEUP; | |
6574 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
6575 | } | |
6576 | return ret; | |
6577 | } | |
6578 | ||
c8d1ba58 JA |
6579 | enum sq_ret { |
6580 | SQT_IDLE = 1, | |
6581 | SQT_SPIN = 2, | |
6582 | SQT_DID_WORK = 4, | |
6583 | }; | |
6584 | ||
6585 | static enum sq_ret __io_sq_thread(struct io_ring_ctx *ctx, | |
e95eee2d | 6586 | unsigned long start_jiffies, bool cap_entries) |
6c271ce2 | 6587 | { |
c8d1ba58 | 6588 | unsigned long timeout = start_jiffies + ctx->sq_thread_idle; |
534ca6d6 | 6589 | struct io_sq_data *sqd = ctx->sq_data; |
c8d1ba58 | 6590 | unsigned int to_submit; |
bdcd3eab | 6591 | int ret = 0; |
6c271ce2 | 6592 | |
c8d1ba58 JA |
6593 | again: |
6594 | if (!list_empty(&ctx->iopoll_list)) { | |
6595 | unsigned nr_events = 0; | |
6a779382 | 6596 | |
c8d1ba58 JA |
6597 | mutex_lock(&ctx->uring_lock); |
6598 | if (!list_empty(&ctx->iopoll_list) && !need_resched()) | |
6599 | io_do_iopoll(ctx, &nr_events, 0); | |
6600 | mutex_unlock(&ctx->uring_lock); | |
6601 | } | |
a4c0b3de | 6602 | |
c8d1ba58 | 6603 | to_submit = io_sqring_entries(ctx); |
6c271ce2 | 6604 | |
c8d1ba58 JA |
6605 | /* |
6606 | * If submit got -EBUSY, flag us as needing the application | |
6607 | * to enter the kernel to reap and flush events. | |
6608 | */ | |
6609 | if (!to_submit || ret == -EBUSY || need_resched()) { | |
6610 | /* | |
6611 | * Drop cur_mm before scheduling, we can't hold it for | |
6612 | * long periods (or over schedule()). Do this before | |
6613 | * adding ourselves to the waitqueue, as the unuse/drop | |
6614 | * may sleep. | |
6615 | */ | |
6616 | io_sq_thread_drop_mm(); | |
6617 | ||
6618 | /* | |
6619 | * We're polling. If we're within the defined idle | |
6620 | * period, then let us spin without work before going | |
6621 | * to sleep. The exception is if we got EBUSY doing | |
6622 | * more IO, we should wait for the application to | |
6623 | * reap events and wake us up. | |
6624 | */ | |
6625 | if (!list_empty(&ctx->iopoll_list) || need_resched() || | |
6626 | (!time_after(jiffies, timeout) && ret != -EBUSY && | |
6627 | !percpu_ref_is_dying(&ctx->refs))) | |
6628 | return SQT_SPIN; | |
6c271ce2 | 6629 | |
534ca6d6 | 6630 | prepare_to_wait(&sqd->wait, &ctx->sqo_wait_entry, |
c8d1ba58 | 6631 | TASK_INTERRUPTIBLE); |
6c271ce2 | 6632 | |
c8d1ba58 JA |
6633 | /* |
6634 | * While doing polled IO, before going to sleep, we need | |
6635 | * to check if there are new reqs added to iopoll_list, | |
6636 | * it is because reqs may have been punted to io worker | |
6637 | * and will be added to iopoll_list later, hence check | |
6638 | * the iopoll_list again. | |
6639 | */ | |
6640 | if ((ctx->flags & IORING_SETUP_IOPOLL) && | |
6641 | !list_empty_careful(&ctx->iopoll_list)) { | |
534ca6d6 | 6642 | finish_wait(&sqd->wait, &ctx->sqo_wait_entry); |
c8d1ba58 | 6643 | goto again; |
6c271ce2 JA |
6644 | } |
6645 | ||
fb5ccc98 | 6646 | to_submit = io_sqring_entries(ctx); |
c8d1ba58 JA |
6647 | if (!to_submit || ret == -EBUSY) |
6648 | return SQT_IDLE; | |
6649 | } | |
c1edbf5f | 6650 | |
534ca6d6 | 6651 | finish_wait(&sqd->wait, &ctx->sqo_wait_entry); |
c8d1ba58 | 6652 | io_ring_clear_wakeup_flag(ctx); |
7143b5ac | 6653 | |
e95eee2d JA |
6654 | /* if we're handling multiple rings, cap submit size for fairness */ |
6655 | if (cap_entries && to_submit > 8) | |
6656 | to_submit = 8; | |
6657 | ||
c8d1ba58 JA |
6658 | mutex_lock(&ctx->uring_lock); |
6659 | if (likely(!percpu_ref_is_dying(&ctx->refs))) | |
6660 | ret = io_submit_sqes(ctx, to_submit); | |
6661 | mutex_unlock(&ctx->uring_lock); | |
90554200 JA |
6662 | |
6663 | if (!io_sqring_full(ctx) && wq_has_sleeper(&ctx->sqo_sq_wait)) | |
6664 | wake_up(&ctx->sqo_sq_wait); | |
6665 | ||
c8d1ba58 JA |
6666 | return SQT_DID_WORK; |
6667 | } | |
6c271ce2 | 6668 | |
69fb2131 JA |
6669 | static void io_sqd_init_new(struct io_sq_data *sqd) |
6670 | { | |
6671 | struct io_ring_ctx *ctx; | |
6672 | ||
6673 | while (!list_empty(&sqd->ctx_new_list)) { | |
6674 | ctx = list_first_entry(&sqd->ctx_new_list, struct io_ring_ctx, sqd_list); | |
6675 | init_wait(&ctx->sqo_wait_entry); | |
6676 | ctx->sqo_wait_entry.func = io_sq_wake_function; | |
6677 | list_move_tail(&ctx->sqd_list, &sqd->ctx_list); | |
6678 | complete(&ctx->sq_thread_comp); | |
6679 | } | |
6680 | } | |
6681 | ||
c8d1ba58 JA |
6682 | static int io_sq_thread(void *data) |
6683 | { | |
91d8f519 | 6684 | struct cgroup_subsys_state *cur_css = NULL; |
69fb2131 JA |
6685 | const struct cred *old_cred = NULL; |
6686 | struct io_sq_data *sqd = data; | |
6687 | struct io_ring_ctx *ctx; | |
c8d1ba58 | 6688 | unsigned long start_jiffies; |
6c271ce2 | 6689 | |
69fb2131 JA |
6690 | start_jiffies = jiffies; |
6691 | while (!kthread_should_stop()) { | |
6692 | enum sq_ret ret = 0; | |
e95eee2d | 6693 | bool cap_entries; |
6c271ce2 | 6694 | |
69fb2131 JA |
6695 | /* |
6696 | * Any changes to the sqd lists are synchronized through the | |
6697 | * kthread parking. This synchronizes the thread vs users, | |
6698 | * the users are synchronized on the sqd->ctx_lock. | |
6699 | */ | |
6700 | if (kthread_should_park()) | |
6701 | kthread_parkme(); | |
6c271ce2 | 6702 | |
69fb2131 JA |
6703 | if (unlikely(!list_empty(&sqd->ctx_new_list))) |
6704 | io_sqd_init_new(sqd); | |
c8d1ba58 | 6705 | |
e95eee2d JA |
6706 | cap_entries = !list_is_singular(&sqd->ctx_list); |
6707 | ||
69fb2131 JA |
6708 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) { |
6709 | if (current->cred != ctx->creds) { | |
6710 | if (old_cred) | |
6711 | revert_creds(old_cred); | |
6712 | old_cred = override_creds(ctx->creds); | |
6713 | } | |
91d8f519 | 6714 | io_sq_thread_associate_blkcg(ctx, &cur_css); |
c8d1ba58 | 6715 | |
e95eee2d | 6716 | ret |= __io_sq_thread(ctx, start_jiffies, cap_entries); |
69fb2131 JA |
6717 | |
6718 | io_sq_thread_drop_mm(); | |
6719 | } | |
6720 | ||
6721 | if (ret & SQT_SPIN) { | |
c8d1ba58 JA |
6722 | io_run_task_work(); |
6723 | cond_resched(); | |
69fb2131 JA |
6724 | } else if (ret == SQT_IDLE) { |
6725 | if (kthread_should_park()) | |
6726 | continue; | |
6727 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) | |
6728 | io_ring_set_wakeup_flag(ctx); | |
6729 | schedule(); | |
6730 | start_jiffies = jiffies; | |
6731 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) | |
6732 | io_ring_clear_wakeup_flag(ctx); | |
c8d1ba58 | 6733 | } |
6c271ce2 JA |
6734 | } |
6735 | ||
4c6e277c | 6736 | io_run_task_work(); |
b41e9852 | 6737 | |
91d8f519 DZ |
6738 | if (cur_css) |
6739 | io_sq_thread_unassociate_blkcg(); | |
69fb2131 JA |
6740 | if (old_cred) |
6741 | revert_creds(old_cred); | |
06058632 | 6742 | |
2bbcd6d3 | 6743 | kthread_parkme(); |
06058632 | 6744 | |
6c271ce2 JA |
6745 | return 0; |
6746 | } | |
6747 | ||
bda52162 JA |
6748 | struct io_wait_queue { |
6749 | struct wait_queue_entry wq; | |
6750 | struct io_ring_ctx *ctx; | |
6751 | unsigned to_wait; | |
6752 | unsigned nr_timeouts; | |
6753 | }; | |
6754 | ||
1d7bb1d5 | 6755 | static inline bool io_should_wake(struct io_wait_queue *iowq, bool noflush) |
bda52162 JA |
6756 | { |
6757 | struct io_ring_ctx *ctx = iowq->ctx; | |
6758 | ||
6759 | /* | |
d195a66e | 6760 | * Wake up if we have enough events, or if a timeout occurred since we |
bda52162 JA |
6761 | * started waiting. For timeouts, we always want to return to userspace, |
6762 | * regardless of event count. | |
6763 | */ | |
1d7bb1d5 | 6764 | return io_cqring_events(ctx, noflush) >= iowq->to_wait || |
bda52162 JA |
6765 | atomic_read(&ctx->cq_timeouts) != iowq->nr_timeouts; |
6766 | } | |
6767 | ||
6768 | static int io_wake_function(struct wait_queue_entry *curr, unsigned int mode, | |
6769 | int wake_flags, void *key) | |
6770 | { | |
6771 | struct io_wait_queue *iowq = container_of(curr, struct io_wait_queue, | |
6772 | wq); | |
6773 | ||
1d7bb1d5 JA |
6774 | /* use noflush == true, as we can't safely rely on locking context */ |
6775 | if (!io_should_wake(iowq, true)) | |
bda52162 JA |
6776 | return -1; |
6777 | ||
6778 | return autoremove_wake_function(curr, mode, wake_flags, key); | |
6779 | } | |
6780 | ||
af9c1a44 JA |
6781 | static int io_run_task_work_sig(void) |
6782 | { | |
6783 | if (io_run_task_work()) | |
6784 | return 1; | |
6785 | if (!signal_pending(current)) | |
6786 | return 0; | |
6787 | if (current->jobctl & JOBCTL_TASK_WORK) { | |
6788 | spin_lock_irq(¤t->sighand->siglock); | |
6789 | current->jobctl &= ~JOBCTL_TASK_WORK; | |
6790 | recalc_sigpending(); | |
6791 | spin_unlock_irq(¤t->sighand->siglock); | |
6792 | return 1; | |
6793 | } | |
6794 | return -EINTR; | |
6795 | } | |
6796 | ||
2b188cc1 JA |
6797 | /* |
6798 | * Wait until events become available, if we don't already have some. The | |
6799 | * application must reap them itself, as they reside on the shared cq ring. | |
6800 | */ | |
6801 | static int io_cqring_wait(struct io_ring_ctx *ctx, int min_events, | |
6802 | const sigset_t __user *sig, size_t sigsz) | |
6803 | { | |
bda52162 JA |
6804 | struct io_wait_queue iowq = { |
6805 | .wq = { | |
6806 | .private = current, | |
6807 | .func = io_wake_function, | |
6808 | .entry = LIST_HEAD_INIT(iowq.wq.entry), | |
6809 | }, | |
6810 | .ctx = ctx, | |
6811 | .to_wait = min_events, | |
6812 | }; | |
75b28aff | 6813 | struct io_rings *rings = ctx->rings; |
e9ffa5c2 | 6814 | int ret = 0; |
2b188cc1 | 6815 | |
b41e9852 JA |
6816 | do { |
6817 | if (io_cqring_events(ctx, false) >= min_events) | |
6818 | return 0; | |
4c6e277c | 6819 | if (!io_run_task_work()) |
b41e9852 | 6820 | break; |
b41e9852 | 6821 | } while (1); |
2b188cc1 JA |
6822 | |
6823 | if (sig) { | |
9e75ad5d AB |
6824 | #ifdef CONFIG_COMPAT |
6825 | if (in_compat_syscall()) | |
6826 | ret = set_compat_user_sigmask((const compat_sigset_t __user *)sig, | |
b772434b | 6827 | sigsz); |
9e75ad5d AB |
6828 | else |
6829 | #endif | |
b772434b | 6830 | ret = set_user_sigmask(sig, sigsz); |
9e75ad5d | 6831 | |
2b188cc1 JA |
6832 | if (ret) |
6833 | return ret; | |
6834 | } | |
6835 | ||
bda52162 | 6836 | iowq.nr_timeouts = atomic_read(&ctx->cq_timeouts); |
c826bd7a | 6837 | trace_io_uring_cqring_wait(ctx, min_events); |
bda52162 JA |
6838 | do { |
6839 | prepare_to_wait_exclusive(&ctx->wait, &iowq.wq, | |
6840 | TASK_INTERRUPTIBLE); | |
ce593a6c | 6841 | /* make sure we run task_work before checking for signals */ |
af9c1a44 JA |
6842 | ret = io_run_task_work_sig(); |
6843 | if (ret > 0) | |
4c6e277c | 6844 | continue; |
af9c1a44 | 6845 | else if (ret < 0) |
bda52162 | 6846 | break; |
ce593a6c JA |
6847 | if (io_should_wake(&iowq, false)) |
6848 | break; | |
6849 | schedule(); | |
bda52162 JA |
6850 | } while (1); |
6851 | finish_wait(&ctx->wait, &iowq.wq); | |
6852 | ||
b7db41c9 | 6853 | restore_saved_sigmask_unless(ret == -EINTR); |
2b188cc1 | 6854 | |
75b28aff | 6855 | return READ_ONCE(rings->cq.head) == READ_ONCE(rings->cq.tail) ? ret : 0; |
2b188cc1 JA |
6856 | } |
6857 | ||
6b06314c JA |
6858 | static void __io_sqe_files_unregister(struct io_ring_ctx *ctx) |
6859 | { | |
6860 | #if defined(CONFIG_UNIX) | |
6861 | if (ctx->ring_sock) { | |
6862 | struct sock *sock = ctx->ring_sock->sk; | |
6863 | struct sk_buff *skb; | |
6864 | ||
6865 | while ((skb = skb_dequeue(&sock->sk_receive_queue)) != NULL) | |
6866 | kfree_skb(skb); | |
6867 | } | |
6868 | #else | |
6869 | int i; | |
6870 | ||
65e19f54 JA |
6871 | for (i = 0; i < ctx->nr_user_files; i++) { |
6872 | struct file *file; | |
6873 | ||
6874 | file = io_file_from_index(ctx, i); | |
6875 | if (file) | |
6876 | fput(file); | |
6877 | } | |
6b06314c JA |
6878 | #endif |
6879 | } | |
6880 | ||
05f3fb3c JA |
6881 | static void io_file_ref_kill(struct percpu_ref *ref) |
6882 | { | |
6883 | struct fixed_file_data *data; | |
6884 | ||
6885 | data = container_of(ref, struct fixed_file_data, refs); | |
6886 | complete(&data->done); | |
6887 | } | |
6888 | ||
6b06314c JA |
6889 | static int io_sqe_files_unregister(struct io_ring_ctx *ctx) |
6890 | { | |
05f3fb3c | 6891 | struct fixed_file_data *data = ctx->file_data; |
05589553 | 6892 | struct fixed_file_ref_node *ref_node = NULL; |
65e19f54 JA |
6893 | unsigned nr_tables, i; |
6894 | ||
05f3fb3c | 6895 | if (!data) |
6b06314c JA |
6896 | return -ENXIO; |
6897 | ||
6a4d07cd | 6898 | spin_lock(&data->lock); |
05589553 XW |
6899 | if (!list_empty(&data->ref_list)) |
6900 | ref_node = list_first_entry(&data->ref_list, | |
6901 | struct fixed_file_ref_node, node); | |
6a4d07cd | 6902 | spin_unlock(&data->lock); |
05589553 XW |
6903 | if (ref_node) |
6904 | percpu_ref_kill(&ref_node->refs); | |
6905 | ||
6906 | percpu_ref_kill(&data->refs); | |
6907 | ||
6908 | /* wait for all refs nodes to complete */ | |
4a38aed2 | 6909 | flush_delayed_work(&ctx->file_put_work); |
2faf852d | 6910 | wait_for_completion(&data->done); |
05f3fb3c | 6911 | |
6b06314c | 6912 | __io_sqe_files_unregister(ctx); |
65e19f54 JA |
6913 | nr_tables = DIV_ROUND_UP(ctx->nr_user_files, IORING_MAX_FILES_TABLE); |
6914 | for (i = 0; i < nr_tables; i++) | |
05f3fb3c JA |
6915 | kfree(data->table[i].files); |
6916 | kfree(data->table); | |
05589553 XW |
6917 | percpu_ref_exit(&data->refs); |
6918 | kfree(data); | |
05f3fb3c | 6919 | ctx->file_data = NULL; |
6b06314c JA |
6920 | ctx->nr_user_files = 0; |
6921 | return 0; | |
6922 | } | |
6923 | ||
534ca6d6 | 6924 | static void io_put_sq_data(struct io_sq_data *sqd) |
6c271ce2 | 6925 | { |
534ca6d6 | 6926 | if (refcount_dec_and_test(&sqd->refs)) { |
2bbcd6d3 RP |
6927 | /* |
6928 | * The park is a bit of a work-around, without it we get | |
6929 | * warning spews on shutdown with SQPOLL set and affinity | |
6930 | * set to a single CPU. | |
6931 | */ | |
534ca6d6 JA |
6932 | if (sqd->thread) { |
6933 | kthread_park(sqd->thread); | |
6934 | kthread_stop(sqd->thread); | |
6935 | } | |
6936 | ||
6937 | kfree(sqd); | |
6938 | } | |
6939 | } | |
6940 | ||
aa06165d JA |
6941 | static struct io_sq_data *io_attach_sq_data(struct io_uring_params *p) |
6942 | { | |
6943 | struct io_ring_ctx *ctx_attach; | |
6944 | struct io_sq_data *sqd; | |
6945 | struct fd f; | |
6946 | ||
6947 | f = fdget(p->wq_fd); | |
6948 | if (!f.file) | |
6949 | return ERR_PTR(-ENXIO); | |
6950 | if (f.file->f_op != &io_uring_fops) { | |
6951 | fdput(f); | |
6952 | return ERR_PTR(-EINVAL); | |
6953 | } | |
6954 | ||
6955 | ctx_attach = f.file->private_data; | |
6956 | sqd = ctx_attach->sq_data; | |
6957 | if (!sqd) { | |
6958 | fdput(f); | |
6959 | return ERR_PTR(-EINVAL); | |
6960 | } | |
6961 | ||
6962 | refcount_inc(&sqd->refs); | |
6963 | fdput(f); | |
6964 | return sqd; | |
6965 | } | |
6966 | ||
534ca6d6 JA |
6967 | static struct io_sq_data *io_get_sq_data(struct io_uring_params *p) |
6968 | { | |
6969 | struct io_sq_data *sqd; | |
6970 | ||
aa06165d JA |
6971 | if (p->flags & IORING_SETUP_ATTACH_WQ) |
6972 | return io_attach_sq_data(p); | |
6973 | ||
534ca6d6 JA |
6974 | sqd = kzalloc(sizeof(*sqd), GFP_KERNEL); |
6975 | if (!sqd) | |
6976 | return ERR_PTR(-ENOMEM); | |
6977 | ||
6978 | refcount_set(&sqd->refs, 1); | |
69fb2131 JA |
6979 | INIT_LIST_HEAD(&sqd->ctx_list); |
6980 | INIT_LIST_HEAD(&sqd->ctx_new_list); | |
6981 | mutex_init(&sqd->ctx_lock); | |
6982 | mutex_init(&sqd->lock); | |
534ca6d6 JA |
6983 | init_waitqueue_head(&sqd->wait); |
6984 | return sqd; | |
6985 | } | |
6986 | ||
69fb2131 JA |
6987 | static void io_sq_thread_unpark(struct io_sq_data *sqd) |
6988 | __releases(&sqd->lock) | |
6989 | { | |
6990 | if (!sqd->thread) | |
6991 | return; | |
6992 | kthread_unpark(sqd->thread); | |
6993 | mutex_unlock(&sqd->lock); | |
6994 | } | |
6995 | ||
6996 | static void io_sq_thread_park(struct io_sq_data *sqd) | |
6997 | __acquires(&sqd->lock) | |
6998 | { | |
6999 | if (!sqd->thread) | |
7000 | return; | |
7001 | mutex_lock(&sqd->lock); | |
7002 | kthread_park(sqd->thread); | |
7003 | } | |
7004 | ||
534ca6d6 JA |
7005 | static void io_sq_thread_stop(struct io_ring_ctx *ctx) |
7006 | { | |
7007 | struct io_sq_data *sqd = ctx->sq_data; | |
7008 | ||
7009 | if (sqd) { | |
7010 | if (sqd->thread) { | |
7011 | /* | |
7012 | * We may arrive here from the error branch in | |
7013 | * io_sq_offload_create() where the kthread is created | |
7014 | * without being waked up, thus wake it up now to make | |
7015 | * sure the wait will complete. | |
7016 | */ | |
7017 | wake_up_process(sqd->thread); | |
7018 | wait_for_completion(&ctx->sq_thread_comp); | |
69fb2131 JA |
7019 | |
7020 | io_sq_thread_park(sqd); | |
7021 | } | |
7022 | ||
7023 | mutex_lock(&sqd->ctx_lock); | |
7024 | list_del(&ctx->sqd_list); | |
7025 | mutex_unlock(&sqd->ctx_lock); | |
7026 | ||
7027 | if (sqd->thread) { | |
7028 | finish_wait(&sqd->wait, &ctx->sqo_wait_entry); | |
7029 | io_sq_thread_unpark(sqd); | |
534ca6d6 JA |
7030 | } |
7031 | ||
7032 | io_put_sq_data(sqd); | |
7033 | ctx->sq_data = NULL; | |
6c271ce2 JA |
7034 | } |
7035 | } | |
7036 | ||
6b06314c JA |
7037 | static void io_finish_async(struct io_ring_ctx *ctx) |
7038 | { | |
6c271ce2 JA |
7039 | io_sq_thread_stop(ctx); |
7040 | ||
561fb04a JA |
7041 | if (ctx->io_wq) { |
7042 | io_wq_destroy(ctx->io_wq); | |
7043 | ctx->io_wq = NULL; | |
6b06314c JA |
7044 | } |
7045 | } | |
7046 | ||
7047 | #if defined(CONFIG_UNIX) | |
6b06314c JA |
7048 | /* |
7049 | * Ensure the UNIX gc is aware of our file set, so we are certain that | |
7050 | * the io_uring can be safely unregistered on process exit, even if we have | |
7051 | * loops in the file referencing. | |
7052 | */ | |
7053 | static int __io_sqe_files_scm(struct io_ring_ctx *ctx, int nr, int offset) | |
7054 | { | |
7055 | struct sock *sk = ctx->ring_sock->sk; | |
7056 | struct scm_fp_list *fpl; | |
7057 | struct sk_buff *skb; | |
08a45173 | 7058 | int i, nr_files; |
6b06314c | 7059 | |
6b06314c JA |
7060 | fpl = kzalloc(sizeof(*fpl), GFP_KERNEL); |
7061 | if (!fpl) | |
7062 | return -ENOMEM; | |
7063 | ||
7064 | skb = alloc_skb(0, GFP_KERNEL); | |
7065 | if (!skb) { | |
7066 | kfree(fpl); | |
7067 | return -ENOMEM; | |
7068 | } | |
7069 | ||
7070 | skb->sk = sk; | |
6b06314c | 7071 | |
08a45173 | 7072 | nr_files = 0; |
6b06314c JA |
7073 | fpl->user = get_uid(ctx->user); |
7074 | for (i = 0; i < nr; i++) { | |
65e19f54 JA |
7075 | struct file *file = io_file_from_index(ctx, i + offset); |
7076 | ||
7077 | if (!file) | |
08a45173 | 7078 | continue; |
65e19f54 | 7079 | fpl->fp[nr_files] = get_file(file); |
08a45173 JA |
7080 | unix_inflight(fpl->user, fpl->fp[nr_files]); |
7081 | nr_files++; | |
6b06314c JA |
7082 | } |
7083 | ||
08a45173 JA |
7084 | if (nr_files) { |
7085 | fpl->max = SCM_MAX_FD; | |
7086 | fpl->count = nr_files; | |
7087 | UNIXCB(skb).fp = fpl; | |
05f3fb3c | 7088 | skb->destructor = unix_destruct_scm; |
08a45173 JA |
7089 | refcount_add(skb->truesize, &sk->sk_wmem_alloc); |
7090 | skb_queue_head(&sk->sk_receive_queue, skb); | |
6b06314c | 7091 | |
08a45173 JA |
7092 | for (i = 0; i < nr_files; i++) |
7093 | fput(fpl->fp[i]); | |
7094 | } else { | |
7095 | kfree_skb(skb); | |
7096 | kfree(fpl); | |
7097 | } | |
6b06314c JA |
7098 | |
7099 | return 0; | |
7100 | } | |
7101 | ||
7102 | /* | |
7103 | * If UNIX sockets are enabled, fd passing can cause a reference cycle which | |
7104 | * causes regular reference counting to break down. We rely on the UNIX | |
7105 | * garbage collection to take care of this problem for us. | |
7106 | */ | |
7107 | static int io_sqe_files_scm(struct io_ring_ctx *ctx) | |
7108 | { | |
7109 | unsigned left, total; | |
7110 | int ret = 0; | |
7111 | ||
7112 | total = 0; | |
7113 | left = ctx->nr_user_files; | |
7114 | while (left) { | |
7115 | unsigned this_files = min_t(unsigned, left, SCM_MAX_FD); | |
6b06314c JA |
7116 | |
7117 | ret = __io_sqe_files_scm(ctx, this_files, total); | |
7118 | if (ret) | |
7119 | break; | |
7120 | left -= this_files; | |
7121 | total += this_files; | |
7122 | } | |
7123 | ||
7124 | if (!ret) | |
7125 | return 0; | |
7126 | ||
7127 | while (total < ctx->nr_user_files) { | |
65e19f54 JA |
7128 | struct file *file = io_file_from_index(ctx, total); |
7129 | ||
7130 | if (file) | |
7131 | fput(file); | |
6b06314c JA |
7132 | total++; |
7133 | } | |
7134 | ||
7135 | return ret; | |
7136 | } | |
7137 | #else | |
7138 | static int io_sqe_files_scm(struct io_ring_ctx *ctx) | |
7139 | { | |
7140 | return 0; | |
7141 | } | |
7142 | #endif | |
7143 | ||
65e19f54 JA |
7144 | static int io_sqe_alloc_file_tables(struct io_ring_ctx *ctx, unsigned nr_tables, |
7145 | unsigned nr_files) | |
7146 | { | |
7147 | int i; | |
7148 | ||
7149 | for (i = 0; i < nr_tables; i++) { | |
05f3fb3c | 7150 | struct fixed_file_table *table = &ctx->file_data->table[i]; |
65e19f54 JA |
7151 | unsigned this_files; |
7152 | ||
7153 | this_files = min(nr_files, IORING_MAX_FILES_TABLE); | |
7154 | table->files = kcalloc(this_files, sizeof(struct file *), | |
7155 | GFP_KERNEL); | |
7156 | if (!table->files) | |
7157 | break; | |
7158 | nr_files -= this_files; | |
7159 | } | |
7160 | ||
7161 | if (i == nr_tables) | |
7162 | return 0; | |
7163 | ||
7164 | for (i = 0; i < nr_tables; i++) { | |
05f3fb3c | 7165 | struct fixed_file_table *table = &ctx->file_data->table[i]; |
65e19f54 JA |
7166 | kfree(table->files); |
7167 | } | |
7168 | return 1; | |
7169 | } | |
7170 | ||
05f3fb3c JA |
7171 | static void io_ring_file_put(struct io_ring_ctx *ctx, struct file *file) |
7172 | { | |
7173 | #if defined(CONFIG_UNIX) | |
7174 | struct sock *sock = ctx->ring_sock->sk; | |
7175 | struct sk_buff_head list, *head = &sock->sk_receive_queue; | |
7176 | struct sk_buff *skb; | |
7177 | int i; | |
7178 | ||
7179 | __skb_queue_head_init(&list); | |
7180 | ||
7181 | /* | |
7182 | * Find the skb that holds this file in its SCM_RIGHTS. When found, | |
7183 | * remove this entry and rearrange the file array. | |
7184 | */ | |
7185 | skb = skb_dequeue(head); | |
7186 | while (skb) { | |
7187 | struct scm_fp_list *fp; | |
7188 | ||
7189 | fp = UNIXCB(skb).fp; | |
7190 | for (i = 0; i < fp->count; i++) { | |
7191 | int left; | |
7192 | ||
7193 | if (fp->fp[i] != file) | |
7194 | continue; | |
7195 | ||
7196 | unix_notinflight(fp->user, fp->fp[i]); | |
7197 | left = fp->count - 1 - i; | |
7198 | if (left) { | |
7199 | memmove(&fp->fp[i], &fp->fp[i + 1], | |
7200 | left * sizeof(struct file *)); | |
7201 | } | |
7202 | fp->count--; | |
7203 | if (!fp->count) { | |
7204 | kfree_skb(skb); | |
7205 | skb = NULL; | |
7206 | } else { | |
7207 | __skb_queue_tail(&list, skb); | |
7208 | } | |
7209 | fput(file); | |
7210 | file = NULL; | |
7211 | break; | |
7212 | } | |
7213 | ||
7214 | if (!file) | |
7215 | break; | |
7216 | ||
7217 | __skb_queue_tail(&list, skb); | |
7218 | ||
7219 | skb = skb_dequeue(head); | |
7220 | } | |
7221 | ||
7222 | if (skb_peek(&list)) { | |
7223 | spin_lock_irq(&head->lock); | |
7224 | while ((skb = __skb_dequeue(&list)) != NULL) | |
7225 | __skb_queue_tail(head, skb); | |
7226 | spin_unlock_irq(&head->lock); | |
7227 | } | |
7228 | #else | |
7229 | fput(file); | |
7230 | #endif | |
7231 | } | |
7232 | ||
7233 | struct io_file_put { | |
05589553 | 7234 | struct list_head list; |
05f3fb3c | 7235 | struct file *file; |
05f3fb3c JA |
7236 | }; |
7237 | ||
4a38aed2 | 7238 | static void __io_file_put_work(struct fixed_file_ref_node *ref_node) |
65e19f54 | 7239 | { |
4a38aed2 JA |
7240 | struct fixed_file_data *file_data = ref_node->file_data; |
7241 | struct io_ring_ctx *ctx = file_data->ctx; | |
05f3fb3c | 7242 | struct io_file_put *pfile, *tmp; |
05589553 XW |
7243 | |
7244 | list_for_each_entry_safe(pfile, tmp, &ref_node->file_list, list) { | |
6a4d07cd | 7245 | list_del(&pfile->list); |
05589553 XW |
7246 | io_ring_file_put(ctx, pfile->file); |
7247 | kfree(pfile); | |
65e19f54 | 7248 | } |
05589553 | 7249 | |
6a4d07cd JA |
7250 | spin_lock(&file_data->lock); |
7251 | list_del(&ref_node->node); | |
7252 | spin_unlock(&file_data->lock); | |
05589553 XW |
7253 | |
7254 | percpu_ref_exit(&ref_node->refs); | |
7255 | kfree(ref_node); | |
7256 | percpu_ref_put(&file_data->refs); | |
2faf852d | 7257 | } |
65e19f54 | 7258 | |
4a38aed2 JA |
7259 | static void io_file_put_work(struct work_struct *work) |
7260 | { | |
7261 | struct io_ring_ctx *ctx; | |
7262 | struct llist_node *node; | |
7263 | ||
7264 | ctx = container_of(work, struct io_ring_ctx, file_put_work.work); | |
7265 | node = llist_del_all(&ctx->file_put_llist); | |
7266 | ||
7267 | while (node) { | |
7268 | struct fixed_file_ref_node *ref_node; | |
7269 | struct llist_node *next = node->next; | |
7270 | ||
7271 | ref_node = llist_entry(node, struct fixed_file_ref_node, llist); | |
7272 | __io_file_put_work(ref_node); | |
7273 | node = next; | |
7274 | } | |
7275 | } | |
7276 | ||
05589553 | 7277 | static void io_file_data_ref_zero(struct percpu_ref *ref) |
2faf852d | 7278 | { |
05589553 | 7279 | struct fixed_file_ref_node *ref_node; |
4a38aed2 JA |
7280 | struct io_ring_ctx *ctx; |
7281 | bool first_add; | |
7282 | int delay = HZ; | |
65e19f54 | 7283 | |
05589553 | 7284 | ref_node = container_of(ref, struct fixed_file_ref_node, refs); |
4a38aed2 | 7285 | ctx = ref_node->file_data->ctx; |
05589553 | 7286 | |
4a38aed2 JA |
7287 | if (percpu_ref_is_dying(&ctx->file_data->refs)) |
7288 | delay = 0; | |
05589553 | 7289 | |
4a38aed2 JA |
7290 | first_add = llist_add(&ref_node->llist, &ctx->file_put_llist); |
7291 | if (!delay) | |
7292 | mod_delayed_work(system_wq, &ctx->file_put_work, 0); | |
7293 | else if (first_add) | |
7294 | queue_delayed_work(system_wq, &ctx->file_put_work, delay); | |
05f3fb3c | 7295 | } |
65e19f54 | 7296 | |
05589553 XW |
7297 | static struct fixed_file_ref_node *alloc_fixed_file_ref_node( |
7298 | struct io_ring_ctx *ctx) | |
05f3fb3c | 7299 | { |
05589553 | 7300 | struct fixed_file_ref_node *ref_node; |
05f3fb3c | 7301 | |
05589553 XW |
7302 | ref_node = kzalloc(sizeof(*ref_node), GFP_KERNEL); |
7303 | if (!ref_node) | |
7304 | return ERR_PTR(-ENOMEM); | |
05f3fb3c | 7305 | |
05589553 XW |
7306 | if (percpu_ref_init(&ref_node->refs, io_file_data_ref_zero, |
7307 | 0, GFP_KERNEL)) { | |
7308 | kfree(ref_node); | |
7309 | return ERR_PTR(-ENOMEM); | |
7310 | } | |
7311 | INIT_LIST_HEAD(&ref_node->node); | |
7312 | INIT_LIST_HEAD(&ref_node->file_list); | |
05589553 XW |
7313 | ref_node->file_data = ctx->file_data; |
7314 | return ref_node; | |
05589553 XW |
7315 | } |
7316 | ||
7317 | static void destroy_fixed_file_ref_node(struct fixed_file_ref_node *ref_node) | |
7318 | { | |
7319 | percpu_ref_exit(&ref_node->refs); | |
7320 | kfree(ref_node); | |
65e19f54 JA |
7321 | } |
7322 | ||
6b06314c JA |
7323 | static int io_sqe_files_register(struct io_ring_ctx *ctx, void __user *arg, |
7324 | unsigned nr_args) | |
7325 | { | |
7326 | __s32 __user *fds = (__s32 __user *) arg; | |
65e19f54 | 7327 | unsigned nr_tables; |
05f3fb3c | 7328 | struct file *file; |
6b06314c JA |
7329 | int fd, ret = 0; |
7330 | unsigned i; | |
05589553 | 7331 | struct fixed_file_ref_node *ref_node; |
6b06314c | 7332 | |
05f3fb3c | 7333 | if (ctx->file_data) |
6b06314c JA |
7334 | return -EBUSY; |
7335 | if (!nr_args) | |
7336 | return -EINVAL; | |
7337 | if (nr_args > IORING_MAX_FIXED_FILES) | |
7338 | return -EMFILE; | |
7339 | ||
05f3fb3c JA |
7340 | ctx->file_data = kzalloc(sizeof(*ctx->file_data), GFP_KERNEL); |
7341 | if (!ctx->file_data) | |
7342 | return -ENOMEM; | |
7343 | ctx->file_data->ctx = ctx; | |
7344 | init_completion(&ctx->file_data->done); | |
05589553 | 7345 | INIT_LIST_HEAD(&ctx->file_data->ref_list); |
f7fe9346 | 7346 | spin_lock_init(&ctx->file_data->lock); |
05f3fb3c | 7347 | |
65e19f54 | 7348 | nr_tables = DIV_ROUND_UP(nr_args, IORING_MAX_FILES_TABLE); |
05f3fb3c JA |
7349 | ctx->file_data->table = kcalloc(nr_tables, |
7350 | sizeof(struct fixed_file_table), | |
65e19f54 | 7351 | GFP_KERNEL); |
05f3fb3c JA |
7352 | if (!ctx->file_data->table) { |
7353 | kfree(ctx->file_data); | |
7354 | ctx->file_data = NULL; | |
6b06314c | 7355 | return -ENOMEM; |
05f3fb3c JA |
7356 | } |
7357 | ||
05589553 | 7358 | if (percpu_ref_init(&ctx->file_data->refs, io_file_ref_kill, |
05f3fb3c JA |
7359 | PERCPU_REF_ALLOW_REINIT, GFP_KERNEL)) { |
7360 | kfree(ctx->file_data->table); | |
7361 | kfree(ctx->file_data); | |
7362 | ctx->file_data = NULL; | |
6b06314c | 7363 | return -ENOMEM; |
05f3fb3c | 7364 | } |
6b06314c | 7365 | |
65e19f54 | 7366 | if (io_sqe_alloc_file_tables(ctx, nr_tables, nr_args)) { |
05f3fb3c JA |
7367 | percpu_ref_exit(&ctx->file_data->refs); |
7368 | kfree(ctx->file_data->table); | |
7369 | kfree(ctx->file_data); | |
7370 | ctx->file_data = NULL; | |
65e19f54 JA |
7371 | return -ENOMEM; |
7372 | } | |
7373 | ||
08a45173 | 7374 | for (i = 0; i < nr_args; i++, ctx->nr_user_files++) { |
65e19f54 JA |
7375 | struct fixed_file_table *table; |
7376 | unsigned index; | |
7377 | ||
6b06314c JA |
7378 | ret = -EFAULT; |
7379 | if (copy_from_user(&fd, &fds[i], sizeof(fd))) | |
7380 | break; | |
08a45173 JA |
7381 | /* allow sparse sets */ |
7382 | if (fd == -1) { | |
7383 | ret = 0; | |
7384 | continue; | |
7385 | } | |
6b06314c | 7386 | |
05f3fb3c | 7387 | table = &ctx->file_data->table[i >> IORING_FILE_TABLE_SHIFT]; |
65e19f54 | 7388 | index = i & IORING_FILE_TABLE_MASK; |
05f3fb3c | 7389 | file = fget(fd); |
6b06314c JA |
7390 | |
7391 | ret = -EBADF; | |
05f3fb3c | 7392 | if (!file) |
6b06314c | 7393 | break; |
05f3fb3c | 7394 | |
6b06314c JA |
7395 | /* |
7396 | * Don't allow io_uring instances to be registered. If UNIX | |
7397 | * isn't enabled, then this causes a reference cycle and this | |
7398 | * instance can never get freed. If UNIX is enabled we'll | |
7399 | * handle it just fine, but there's still no point in allowing | |
7400 | * a ring fd as it doesn't support regular read/write anyway. | |
7401 | */ | |
05f3fb3c JA |
7402 | if (file->f_op == &io_uring_fops) { |
7403 | fput(file); | |
6b06314c JA |
7404 | break; |
7405 | } | |
6b06314c | 7406 | ret = 0; |
05f3fb3c | 7407 | table->files[index] = file; |
6b06314c JA |
7408 | } |
7409 | ||
7410 | if (ret) { | |
65e19f54 | 7411 | for (i = 0; i < ctx->nr_user_files; i++) { |
65e19f54 JA |
7412 | file = io_file_from_index(ctx, i); |
7413 | if (file) | |
7414 | fput(file); | |
7415 | } | |
7416 | for (i = 0; i < nr_tables; i++) | |
05f3fb3c | 7417 | kfree(ctx->file_data->table[i].files); |
6b06314c | 7418 | |
667e57da | 7419 | percpu_ref_exit(&ctx->file_data->refs); |
05f3fb3c JA |
7420 | kfree(ctx->file_data->table); |
7421 | kfree(ctx->file_data); | |
7422 | ctx->file_data = NULL; | |
6b06314c JA |
7423 | ctx->nr_user_files = 0; |
7424 | return ret; | |
7425 | } | |
7426 | ||
7427 | ret = io_sqe_files_scm(ctx); | |
05589553 | 7428 | if (ret) { |
6b06314c | 7429 | io_sqe_files_unregister(ctx); |
05589553 XW |
7430 | return ret; |
7431 | } | |
6b06314c | 7432 | |
05589553 XW |
7433 | ref_node = alloc_fixed_file_ref_node(ctx); |
7434 | if (IS_ERR(ref_node)) { | |
7435 | io_sqe_files_unregister(ctx); | |
7436 | return PTR_ERR(ref_node); | |
7437 | } | |
7438 | ||
7439 | ctx->file_data->cur_refs = &ref_node->refs; | |
6a4d07cd | 7440 | spin_lock(&ctx->file_data->lock); |
05589553 | 7441 | list_add(&ref_node->node, &ctx->file_data->ref_list); |
6a4d07cd | 7442 | spin_unlock(&ctx->file_data->lock); |
05589553 | 7443 | percpu_ref_get(&ctx->file_data->refs); |
6b06314c JA |
7444 | return ret; |
7445 | } | |
7446 | ||
c3a31e60 JA |
7447 | static int io_sqe_file_register(struct io_ring_ctx *ctx, struct file *file, |
7448 | int index) | |
7449 | { | |
7450 | #if defined(CONFIG_UNIX) | |
7451 | struct sock *sock = ctx->ring_sock->sk; | |
7452 | struct sk_buff_head *head = &sock->sk_receive_queue; | |
7453 | struct sk_buff *skb; | |
7454 | ||
7455 | /* | |
7456 | * See if we can merge this file into an existing skb SCM_RIGHTS | |
7457 | * file set. If there's no room, fall back to allocating a new skb | |
7458 | * and filling it in. | |
7459 | */ | |
7460 | spin_lock_irq(&head->lock); | |
7461 | skb = skb_peek(head); | |
7462 | if (skb) { | |
7463 | struct scm_fp_list *fpl = UNIXCB(skb).fp; | |
7464 | ||
7465 | if (fpl->count < SCM_MAX_FD) { | |
7466 | __skb_unlink(skb, head); | |
7467 | spin_unlock_irq(&head->lock); | |
7468 | fpl->fp[fpl->count] = get_file(file); | |
7469 | unix_inflight(fpl->user, fpl->fp[fpl->count]); | |
7470 | fpl->count++; | |
7471 | spin_lock_irq(&head->lock); | |
7472 | __skb_queue_head(head, skb); | |
7473 | } else { | |
7474 | skb = NULL; | |
7475 | } | |
7476 | } | |
7477 | spin_unlock_irq(&head->lock); | |
7478 | ||
7479 | if (skb) { | |
7480 | fput(file); | |
7481 | return 0; | |
7482 | } | |
7483 | ||
7484 | return __io_sqe_files_scm(ctx, 1, index); | |
7485 | #else | |
7486 | return 0; | |
7487 | #endif | |
7488 | } | |
7489 | ||
a5318d3c | 7490 | static int io_queue_file_removal(struct fixed_file_data *data, |
05589553 | 7491 | struct file *file) |
05f3fb3c | 7492 | { |
a5318d3c | 7493 | struct io_file_put *pfile; |
05589553 XW |
7494 | struct percpu_ref *refs = data->cur_refs; |
7495 | struct fixed_file_ref_node *ref_node; | |
05f3fb3c | 7496 | |
05f3fb3c | 7497 | pfile = kzalloc(sizeof(*pfile), GFP_KERNEL); |
a5318d3c HD |
7498 | if (!pfile) |
7499 | return -ENOMEM; | |
05f3fb3c | 7500 | |
05589553 | 7501 | ref_node = container_of(refs, struct fixed_file_ref_node, refs); |
05f3fb3c | 7502 | pfile->file = file; |
05589553 XW |
7503 | list_add(&pfile->list, &ref_node->file_list); |
7504 | ||
a5318d3c | 7505 | return 0; |
05f3fb3c JA |
7506 | } |
7507 | ||
7508 | static int __io_sqe_files_update(struct io_ring_ctx *ctx, | |
7509 | struct io_uring_files_update *up, | |
7510 | unsigned nr_args) | |
7511 | { | |
7512 | struct fixed_file_data *data = ctx->file_data; | |
05589553 | 7513 | struct fixed_file_ref_node *ref_node; |
05f3fb3c | 7514 | struct file *file; |
c3a31e60 JA |
7515 | __s32 __user *fds; |
7516 | int fd, i, err; | |
7517 | __u32 done; | |
05589553 | 7518 | bool needs_switch = false; |
c3a31e60 | 7519 | |
05f3fb3c | 7520 | if (check_add_overflow(up->offset, nr_args, &done)) |
c3a31e60 JA |
7521 | return -EOVERFLOW; |
7522 | if (done > ctx->nr_user_files) | |
7523 | return -EINVAL; | |
7524 | ||
05589553 XW |
7525 | ref_node = alloc_fixed_file_ref_node(ctx); |
7526 | if (IS_ERR(ref_node)) | |
7527 | return PTR_ERR(ref_node); | |
7528 | ||
c3a31e60 | 7529 | done = 0; |
05f3fb3c | 7530 | fds = u64_to_user_ptr(up->fds); |
c3a31e60 | 7531 | while (nr_args) { |
65e19f54 JA |
7532 | struct fixed_file_table *table; |
7533 | unsigned index; | |
7534 | ||
c3a31e60 JA |
7535 | err = 0; |
7536 | if (copy_from_user(&fd, &fds[done], sizeof(fd))) { | |
7537 | err = -EFAULT; | |
7538 | break; | |
7539 | } | |
05f3fb3c JA |
7540 | i = array_index_nospec(up->offset, ctx->nr_user_files); |
7541 | table = &ctx->file_data->table[i >> IORING_FILE_TABLE_SHIFT]; | |
65e19f54 JA |
7542 | index = i & IORING_FILE_TABLE_MASK; |
7543 | if (table->files[index]) { | |
98dfd502 | 7544 | file = table->files[index]; |
a5318d3c HD |
7545 | err = io_queue_file_removal(data, file); |
7546 | if (err) | |
7547 | break; | |
65e19f54 | 7548 | table->files[index] = NULL; |
05589553 | 7549 | needs_switch = true; |
c3a31e60 JA |
7550 | } |
7551 | if (fd != -1) { | |
c3a31e60 JA |
7552 | file = fget(fd); |
7553 | if (!file) { | |
7554 | err = -EBADF; | |
7555 | break; | |
7556 | } | |
7557 | /* | |
7558 | * Don't allow io_uring instances to be registered. If | |
7559 | * UNIX isn't enabled, then this causes a reference | |
7560 | * cycle and this instance can never get freed. If UNIX | |
7561 | * is enabled we'll handle it just fine, but there's | |
7562 | * still no point in allowing a ring fd as it doesn't | |
7563 | * support regular read/write anyway. | |
7564 | */ | |
7565 | if (file->f_op == &io_uring_fops) { | |
7566 | fput(file); | |
7567 | err = -EBADF; | |
7568 | break; | |
7569 | } | |
65e19f54 | 7570 | table->files[index] = file; |
c3a31e60 | 7571 | err = io_sqe_file_register(ctx, file, i); |
f3bd9dae | 7572 | if (err) { |
95d1c8e5 | 7573 | table->files[index] = NULL; |
f3bd9dae | 7574 | fput(file); |
c3a31e60 | 7575 | break; |
f3bd9dae | 7576 | } |
c3a31e60 JA |
7577 | } |
7578 | nr_args--; | |
7579 | done++; | |
05f3fb3c JA |
7580 | up->offset++; |
7581 | } | |
7582 | ||
05589553 XW |
7583 | if (needs_switch) { |
7584 | percpu_ref_kill(data->cur_refs); | |
6a4d07cd | 7585 | spin_lock(&data->lock); |
05589553 XW |
7586 | list_add(&ref_node->node, &data->ref_list); |
7587 | data->cur_refs = &ref_node->refs; | |
6a4d07cd | 7588 | spin_unlock(&data->lock); |
05589553 XW |
7589 | percpu_ref_get(&ctx->file_data->refs); |
7590 | } else | |
7591 | destroy_fixed_file_ref_node(ref_node); | |
c3a31e60 JA |
7592 | |
7593 | return done ? done : err; | |
7594 | } | |
05589553 | 7595 | |
05f3fb3c JA |
7596 | static int io_sqe_files_update(struct io_ring_ctx *ctx, void __user *arg, |
7597 | unsigned nr_args) | |
7598 | { | |
7599 | struct io_uring_files_update up; | |
7600 | ||
7601 | if (!ctx->file_data) | |
7602 | return -ENXIO; | |
7603 | if (!nr_args) | |
7604 | return -EINVAL; | |
7605 | if (copy_from_user(&up, arg, sizeof(up))) | |
7606 | return -EFAULT; | |
7607 | if (up.resv) | |
7608 | return -EINVAL; | |
7609 | ||
7610 | return __io_sqe_files_update(ctx, &up, nr_args); | |
7611 | } | |
c3a31e60 | 7612 | |
e9fd9396 | 7613 | static void io_free_work(struct io_wq_work *work) |
7d723065 JA |
7614 | { |
7615 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); | |
7616 | ||
e9fd9396 | 7617 | /* Consider that io_steal_work() relies on this ref */ |
7d723065 JA |
7618 | io_put_req(req); |
7619 | } | |
7620 | ||
24369c2e PB |
7621 | static int io_init_wq_offload(struct io_ring_ctx *ctx, |
7622 | struct io_uring_params *p) | |
7623 | { | |
7624 | struct io_wq_data data; | |
7625 | struct fd f; | |
7626 | struct io_ring_ctx *ctx_attach; | |
7627 | unsigned int concurrency; | |
7628 | int ret = 0; | |
7629 | ||
7630 | data.user = ctx->user; | |
e9fd9396 | 7631 | data.free_work = io_free_work; |
f5fa38c5 | 7632 | data.do_work = io_wq_submit_work; |
24369c2e PB |
7633 | |
7634 | if (!(p->flags & IORING_SETUP_ATTACH_WQ)) { | |
7635 | /* Do QD, or 4 * CPUS, whatever is smallest */ | |
7636 | concurrency = min(ctx->sq_entries, 4 * num_online_cpus()); | |
7637 | ||
7638 | ctx->io_wq = io_wq_create(concurrency, &data); | |
7639 | if (IS_ERR(ctx->io_wq)) { | |
7640 | ret = PTR_ERR(ctx->io_wq); | |
7641 | ctx->io_wq = NULL; | |
7642 | } | |
7643 | return ret; | |
7644 | } | |
7645 | ||
7646 | f = fdget(p->wq_fd); | |
7647 | if (!f.file) | |
7648 | return -EBADF; | |
7649 | ||
7650 | if (f.file->f_op != &io_uring_fops) { | |
7651 | ret = -EINVAL; | |
7652 | goto out_fput; | |
7653 | } | |
7654 | ||
7655 | ctx_attach = f.file->private_data; | |
7656 | /* @io_wq is protected by holding the fd */ | |
7657 | if (!io_wq_get(ctx_attach->io_wq, &data)) { | |
7658 | ret = -EINVAL; | |
7659 | goto out_fput; | |
7660 | } | |
7661 | ||
7662 | ctx->io_wq = ctx_attach->io_wq; | |
7663 | out_fput: | |
7664 | fdput(f); | |
7665 | return ret; | |
7666 | } | |
7667 | ||
0f212204 JA |
7668 | static int io_uring_alloc_task_context(struct task_struct *task) |
7669 | { | |
7670 | struct io_uring_task *tctx; | |
7671 | ||
7672 | tctx = kmalloc(sizeof(*tctx), GFP_KERNEL); | |
7673 | if (unlikely(!tctx)) | |
7674 | return -ENOMEM; | |
7675 | ||
7676 | xa_init(&tctx->xa); | |
7677 | init_waitqueue_head(&tctx->wait); | |
7678 | tctx->last = NULL; | |
7679 | tctx->in_idle = 0; | |
7680 | atomic_long_set(&tctx->req_issue, 0); | |
7681 | atomic_long_set(&tctx->req_complete, 0); | |
7682 | task->io_uring = tctx; | |
7683 | return 0; | |
7684 | } | |
7685 | ||
7686 | void __io_uring_free(struct task_struct *tsk) | |
7687 | { | |
7688 | struct io_uring_task *tctx = tsk->io_uring; | |
7689 | ||
7690 | WARN_ON_ONCE(!xa_empty(&tctx->xa)); | |
0f212204 JA |
7691 | kfree(tctx); |
7692 | tsk->io_uring = NULL; | |
7693 | } | |
7694 | ||
7e84e1c7 SG |
7695 | static int io_sq_offload_create(struct io_ring_ctx *ctx, |
7696 | struct io_uring_params *p) | |
2b188cc1 JA |
7697 | { |
7698 | int ret; | |
7699 | ||
6c271ce2 | 7700 | if (ctx->flags & IORING_SETUP_SQPOLL) { |
534ca6d6 JA |
7701 | struct io_sq_data *sqd; |
7702 | ||
3ec482d1 JA |
7703 | ret = -EPERM; |
7704 | if (!capable(CAP_SYS_ADMIN)) | |
7705 | goto err; | |
7706 | ||
534ca6d6 JA |
7707 | sqd = io_get_sq_data(p); |
7708 | if (IS_ERR(sqd)) { | |
7709 | ret = PTR_ERR(sqd); | |
7710 | goto err; | |
7711 | } | |
69fb2131 | 7712 | |
534ca6d6 | 7713 | ctx->sq_data = sqd; |
69fb2131 JA |
7714 | io_sq_thread_park(sqd); |
7715 | mutex_lock(&sqd->ctx_lock); | |
7716 | list_add(&ctx->sqd_list, &sqd->ctx_new_list); | |
7717 | mutex_unlock(&sqd->ctx_lock); | |
7718 | io_sq_thread_unpark(sqd); | |
534ca6d6 | 7719 | |
917257da JA |
7720 | ctx->sq_thread_idle = msecs_to_jiffies(p->sq_thread_idle); |
7721 | if (!ctx->sq_thread_idle) | |
7722 | ctx->sq_thread_idle = HZ; | |
7723 | ||
aa06165d JA |
7724 | if (sqd->thread) |
7725 | goto done; | |
7726 | ||
6c271ce2 | 7727 | if (p->flags & IORING_SETUP_SQ_AFF) { |
44a9bd18 | 7728 | int cpu = p->sq_thread_cpu; |
6c271ce2 | 7729 | |
917257da | 7730 | ret = -EINVAL; |
44a9bd18 JA |
7731 | if (cpu >= nr_cpu_ids) |
7732 | goto err; | |
7889f44d | 7733 | if (!cpu_online(cpu)) |
917257da JA |
7734 | goto err; |
7735 | ||
69fb2131 | 7736 | sqd->thread = kthread_create_on_cpu(io_sq_thread, sqd, |
534ca6d6 | 7737 | cpu, "io_uring-sq"); |
6c271ce2 | 7738 | } else { |
69fb2131 | 7739 | sqd->thread = kthread_create(io_sq_thread, sqd, |
6c271ce2 JA |
7740 | "io_uring-sq"); |
7741 | } | |
534ca6d6 JA |
7742 | if (IS_ERR(sqd->thread)) { |
7743 | ret = PTR_ERR(sqd->thread); | |
7744 | sqd->thread = NULL; | |
6c271ce2 JA |
7745 | goto err; |
7746 | } | |
534ca6d6 | 7747 | ret = io_uring_alloc_task_context(sqd->thread); |
0f212204 JA |
7748 | if (ret) |
7749 | goto err; | |
6c271ce2 JA |
7750 | } else if (p->flags & IORING_SETUP_SQ_AFF) { |
7751 | /* Can't have SQ_AFF without SQPOLL */ | |
7752 | ret = -EINVAL; | |
7753 | goto err; | |
7754 | } | |
7755 | ||
aa06165d | 7756 | done: |
24369c2e PB |
7757 | ret = io_init_wq_offload(ctx, p); |
7758 | if (ret) | |
2b188cc1 | 7759 | goto err; |
2b188cc1 JA |
7760 | |
7761 | return 0; | |
7762 | err: | |
54a91f3b | 7763 | io_finish_async(ctx); |
2b188cc1 JA |
7764 | return ret; |
7765 | } | |
7766 | ||
7e84e1c7 SG |
7767 | static void io_sq_offload_start(struct io_ring_ctx *ctx) |
7768 | { | |
534ca6d6 JA |
7769 | struct io_sq_data *sqd = ctx->sq_data; |
7770 | ||
7771 | if ((ctx->flags & IORING_SETUP_SQPOLL) && sqd->thread) | |
7772 | wake_up_process(sqd->thread); | |
7e84e1c7 SG |
7773 | } |
7774 | ||
a087e2b5 BM |
7775 | static inline void __io_unaccount_mem(struct user_struct *user, |
7776 | unsigned long nr_pages) | |
2b188cc1 JA |
7777 | { |
7778 | atomic_long_sub(nr_pages, &user->locked_vm); | |
7779 | } | |
7780 | ||
a087e2b5 BM |
7781 | static inline int __io_account_mem(struct user_struct *user, |
7782 | unsigned long nr_pages) | |
2b188cc1 JA |
7783 | { |
7784 | unsigned long page_limit, cur_pages, new_pages; | |
7785 | ||
7786 | /* Don't allow more pages than we can safely lock */ | |
7787 | page_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT; | |
7788 | ||
7789 | do { | |
7790 | cur_pages = atomic_long_read(&user->locked_vm); | |
7791 | new_pages = cur_pages + nr_pages; | |
7792 | if (new_pages > page_limit) | |
7793 | return -ENOMEM; | |
7794 | } while (atomic_long_cmpxchg(&user->locked_vm, cur_pages, | |
7795 | new_pages) != cur_pages); | |
7796 | ||
7797 | return 0; | |
7798 | } | |
7799 | ||
2e0464d4 BM |
7800 | static void io_unaccount_mem(struct io_ring_ctx *ctx, unsigned long nr_pages, |
7801 | enum io_mem_account acct) | |
a087e2b5 | 7802 | { |
aad5d8da | 7803 | if (ctx->limit_mem) |
a087e2b5 | 7804 | __io_unaccount_mem(ctx->user, nr_pages); |
30975825 | 7805 | |
2aede0e4 | 7806 | if (ctx->mm_account) { |
2e0464d4 | 7807 | if (acct == ACCT_LOCKED) |
2aede0e4 | 7808 | ctx->mm_account->locked_vm -= nr_pages; |
2e0464d4 | 7809 | else if (acct == ACCT_PINNED) |
2aede0e4 | 7810 | atomic64_sub(nr_pages, &ctx->mm_account->pinned_vm); |
2e0464d4 | 7811 | } |
a087e2b5 BM |
7812 | } |
7813 | ||
2e0464d4 BM |
7814 | static int io_account_mem(struct io_ring_ctx *ctx, unsigned long nr_pages, |
7815 | enum io_mem_account acct) | |
a087e2b5 | 7816 | { |
30975825 BM |
7817 | int ret; |
7818 | ||
7819 | if (ctx->limit_mem) { | |
7820 | ret = __io_account_mem(ctx->user, nr_pages); | |
7821 | if (ret) | |
7822 | return ret; | |
7823 | } | |
7824 | ||
2aede0e4 | 7825 | if (ctx->mm_account) { |
2e0464d4 | 7826 | if (acct == ACCT_LOCKED) |
2aede0e4 | 7827 | ctx->mm_account->locked_vm += nr_pages; |
2e0464d4 | 7828 | else if (acct == ACCT_PINNED) |
2aede0e4 | 7829 | atomic64_add(nr_pages, &ctx->mm_account->pinned_vm); |
2e0464d4 | 7830 | } |
a087e2b5 BM |
7831 | |
7832 | return 0; | |
7833 | } | |
7834 | ||
2b188cc1 JA |
7835 | static void io_mem_free(void *ptr) |
7836 | { | |
52e04ef4 MR |
7837 | struct page *page; |
7838 | ||
7839 | if (!ptr) | |
7840 | return; | |
2b188cc1 | 7841 | |
52e04ef4 | 7842 | page = virt_to_head_page(ptr); |
2b188cc1 JA |
7843 | if (put_page_testzero(page)) |
7844 | free_compound_page(page); | |
7845 | } | |
7846 | ||
7847 | static void *io_mem_alloc(size_t size) | |
7848 | { | |
7849 | gfp_t gfp_flags = GFP_KERNEL | __GFP_ZERO | __GFP_NOWARN | __GFP_COMP | | |
7850 | __GFP_NORETRY; | |
7851 | ||
7852 | return (void *) __get_free_pages(gfp_flags, get_order(size)); | |
7853 | } | |
7854 | ||
75b28aff HV |
7855 | static unsigned long rings_size(unsigned sq_entries, unsigned cq_entries, |
7856 | size_t *sq_offset) | |
7857 | { | |
7858 | struct io_rings *rings; | |
7859 | size_t off, sq_array_size; | |
7860 | ||
7861 | off = struct_size(rings, cqes, cq_entries); | |
7862 | if (off == SIZE_MAX) | |
7863 | return SIZE_MAX; | |
7864 | ||
7865 | #ifdef CONFIG_SMP | |
7866 | off = ALIGN(off, SMP_CACHE_BYTES); | |
7867 | if (off == 0) | |
7868 | return SIZE_MAX; | |
7869 | #endif | |
7870 | ||
b36200f5 DV |
7871 | if (sq_offset) |
7872 | *sq_offset = off; | |
7873 | ||
75b28aff HV |
7874 | sq_array_size = array_size(sizeof(u32), sq_entries); |
7875 | if (sq_array_size == SIZE_MAX) | |
7876 | return SIZE_MAX; | |
7877 | ||
7878 | if (check_add_overflow(off, sq_array_size, &off)) | |
7879 | return SIZE_MAX; | |
7880 | ||
75b28aff HV |
7881 | return off; |
7882 | } | |
7883 | ||
2b188cc1 JA |
7884 | static unsigned long ring_pages(unsigned sq_entries, unsigned cq_entries) |
7885 | { | |
75b28aff | 7886 | size_t pages; |
2b188cc1 | 7887 | |
75b28aff HV |
7888 | pages = (size_t)1 << get_order( |
7889 | rings_size(sq_entries, cq_entries, NULL)); | |
7890 | pages += (size_t)1 << get_order( | |
7891 | array_size(sizeof(struct io_uring_sqe), sq_entries)); | |
2b188cc1 | 7892 | |
75b28aff | 7893 | return pages; |
2b188cc1 JA |
7894 | } |
7895 | ||
edafccee JA |
7896 | static int io_sqe_buffer_unregister(struct io_ring_ctx *ctx) |
7897 | { | |
7898 | int i, j; | |
7899 | ||
7900 | if (!ctx->user_bufs) | |
7901 | return -ENXIO; | |
7902 | ||
7903 | for (i = 0; i < ctx->nr_user_bufs; i++) { | |
7904 | struct io_mapped_ubuf *imu = &ctx->user_bufs[i]; | |
7905 | ||
7906 | for (j = 0; j < imu->nr_bvecs; j++) | |
f1f6a7dd | 7907 | unpin_user_page(imu->bvec[j].bv_page); |
edafccee | 7908 | |
de293938 JA |
7909 | if (imu->acct_pages) |
7910 | io_unaccount_mem(ctx, imu->acct_pages, ACCT_PINNED); | |
d4ef6475 | 7911 | kvfree(imu->bvec); |
edafccee JA |
7912 | imu->nr_bvecs = 0; |
7913 | } | |
7914 | ||
7915 | kfree(ctx->user_bufs); | |
7916 | ctx->user_bufs = NULL; | |
7917 | ctx->nr_user_bufs = 0; | |
7918 | return 0; | |
7919 | } | |
7920 | ||
7921 | static int io_copy_iov(struct io_ring_ctx *ctx, struct iovec *dst, | |
7922 | void __user *arg, unsigned index) | |
7923 | { | |
7924 | struct iovec __user *src; | |
7925 | ||
7926 | #ifdef CONFIG_COMPAT | |
7927 | if (ctx->compat) { | |
7928 | struct compat_iovec __user *ciovs; | |
7929 | struct compat_iovec ciov; | |
7930 | ||
7931 | ciovs = (struct compat_iovec __user *) arg; | |
7932 | if (copy_from_user(&ciov, &ciovs[index], sizeof(ciov))) | |
7933 | return -EFAULT; | |
7934 | ||
d55e5f5b | 7935 | dst->iov_base = u64_to_user_ptr((u64)ciov.iov_base); |
edafccee JA |
7936 | dst->iov_len = ciov.iov_len; |
7937 | return 0; | |
7938 | } | |
7939 | #endif | |
7940 | src = (struct iovec __user *) arg; | |
7941 | if (copy_from_user(dst, &src[index], sizeof(*dst))) | |
7942 | return -EFAULT; | |
7943 | return 0; | |
7944 | } | |
7945 | ||
de293938 JA |
7946 | /* |
7947 | * Not super efficient, but this is just a registration time. And we do cache | |
7948 | * the last compound head, so generally we'll only do a full search if we don't | |
7949 | * match that one. | |
7950 | * | |
7951 | * We check if the given compound head page has already been accounted, to | |
7952 | * avoid double accounting it. This allows us to account the full size of the | |
7953 | * page, not just the constituent pages of a huge page. | |
7954 | */ | |
7955 | static bool headpage_already_acct(struct io_ring_ctx *ctx, struct page **pages, | |
7956 | int nr_pages, struct page *hpage) | |
7957 | { | |
7958 | int i, j; | |
7959 | ||
7960 | /* check current page array */ | |
7961 | for (i = 0; i < nr_pages; i++) { | |
7962 | if (!PageCompound(pages[i])) | |
7963 | continue; | |
7964 | if (compound_head(pages[i]) == hpage) | |
7965 | return true; | |
7966 | } | |
7967 | ||
7968 | /* check previously registered pages */ | |
7969 | for (i = 0; i < ctx->nr_user_bufs; i++) { | |
7970 | struct io_mapped_ubuf *imu = &ctx->user_bufs[i]; | |
7971 | ||
7972 | for (j = 0; j < imu->nr_bvecs; j++) { | |
7973 | if (!PageCompound(imu->bvec[j].bv_page)) | |
7974 | continue; | |
7975 | if (compound_head(imu->bvec[j].bv_page) == hpage) | |
7976 | return true; | |
7977 | } | |
7978 | } | |
7979 | ||
7980 | return false; | |
7981 | } | |
7982 | ||
7983 | static int io_buffer_account_pin(struct io_ring_ctx *ctx, struct page **pages, | |
7984 | int nr_pages, struct io_mapped_ubuf *imu, | |
7985 | struct page **last_hpage) | |
7986 | { | |
7987 | int i, ret; | |
7988 | ||
7989 | for (i = 0; i < nr_pages; i++) { | |
7990 | if (!PageCompound(pages[i])) { | |
7991 | imu->acct_pages++; | |
7992 | } else { | |
7993 | struct page *hpage; | |
7994 | ||
7995 | hpage = compound_head(pages[i]); | |
7996 | if (hpage == *last_hpage) | |
7997 | continue; | |
7998 | *last_hpage = hpage; | |
7999 | if (headpage_already_acct(ctx, pages, i, hpage)) | |
8000 | continue; | |
8001 | imu->acct_pages += page_size(hpage) >> PAGE_SHIFT; | |
8002 | } | |
8003 | } | |
8004 | ||
8005 | if (!imu->acct_pages) | |
8006 | return 0; | |
8007 | ||
8008 | ret = io_account_mem(ctx, imu->acct_pages, ACCT_PINNED); | |
8009 | if (ret) | |
8010 | imu->acct_pages = 0; | |
8011 | return ret; | |
8012 | } | |
8013 | ||
edafccee JA |
8014 | static int io_sqe_buffer_register(struct io_ring_ctx *ctx, void __user *arg, |
8015 | unsigned nr_args) | |
8016 | { | |
8017 | struct vm_area_struct **vmas = NULL; | |
8018 | struct page **pages = NULL; | |
de293938 | 8019 | struct page *last_hpage = NULL; |
edafccee JA |
8020 | int i, j, got_pages = 0; |
8021 | int ret = -EINVAL; | |
8022 | ||
8023 | if (ctx->user_bufs) | |
8024 | return -EBUSY; | |
8025 | if (!nr_args || nr_args > UIO_MAXIOV) | |
8026 | return -EINVAL; | |
8027 | ||
8028 | ctx->user_bufs = kcalloc(nr_args, sizeof(struct io_mapped_ubuf), | |
8029 | GFP_KERNEL); | |
8030 | if (!ctx->user_bufs) | |
8031 | return -ENOMEM; | |
8032 | ||
8033 | for (i = 0; i < nr_args; i++) { | |
8034 | struct io_mapped_ubuf *imu = &ctx->user_bufs[i]; | |
8035 | unsigned long off, start, end, ubuf; | |
8036 | int pret, nr_pages; | |
8037 | struct iovec iov; | |
8038 | size_t size; | |
8039 | ||
8040 | ret = io_copy_iov(ctx, &iov, arg, i); | |
8041 | if (ret) | |
a278682d | 8042 | goto err; |
edafccee JA |
8043 | |
8044 | /* | |
8045 | * Don't impose further limits on the size and buffer | |
8046 | * constraints here, we'll -EINVAL later when IO is | |
8047 | * submitted if they are wrong. | |
8048 | */ | |
8049 | ret = -EFAULT; | |
8050 | if (!iov.iov_base || !iov.iov_len) | |
8051 | goto err; | |
8052 | ||
8053 | /* arbitrary limit, but we need something */ | |
8054 | if (iov.iov_len > SZ_1G) | |
8055 | goto err; | |
8056 | ||
8057 | ubuf = (unsigned long) iov.iov_base; | |
8058 | end = (ubuf + iov.iov_len + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
8059 | start = ubuf >> PAGE_SHIFT; | |
8060 | nr_pages = end - start; | |
8061 | ||
edafccee JA |
8062 | ret = 0; |
8063 | if (!pages || nr_pages > got_pages) { | |
a8c73c1a DE |
8064 | kvfree(vmas); |
8065 | kvfree(pages); | |
d4ef6475 | 8066 | pages = kvmalloc_array(nr_pages, sizeof(struct page *), |
edafccee | 8067 | GFP_KERNEL); |
d4ef6475 | 8068 | vmas = kvmalloc_array(nr_pages, |
edafccee JA |
8069 | sizeof(struct vm_area_struct *), |
8070 | GFP_KERNEL); | |
8071 | if (!pages || !vmas) { | |
8072 | ret = -ENOMEM; | |
edafccee JA |
8073 | goto err; |
8074 | } | |
8075 | got_pages = nr_pages; | |
8076 | } | |
8077 | ||
d4ef6475 | 8078 | imu->bvec = kvmalloc_array(nr_pages, sizeof(struct bio_vec), |
edafccee JA |
8079 | GFP_KERNEL); |
8080 | ret = -ENOMEM; | |
de293938 | 8081 | if (!imu->bvec) |
edafccee | 8082 | goto err; |
edafccee JA |
8083 | |
8084 | ret = 0; | |
d8ed45c5 | 8085 | mmap_read_lock(current->mm); |
2113b05d | 8086 | pret = pin_user_pages(ubuf, nr_pages, |
932f4a63 IW |
8087 | FOLL_WRITE | FOLL_LONGTERM, |
8088 | pages, vmas); | |
edafccee JA |
8089 | if (pret == nr_pages) { |
8090 | /* don't support file backed memory */ | |
8091 | for (j = 0; j < nr_pages; j++) { | |
8092 | struct vm_area_struct *vma = vmas[j]; | |
8093 | ||
8094 | if (vma->vm_file && | |
8095 | !is_file_hugepages(vma->vm_file)) { | |
8096 | ret = -EOPNOTSUPP; | |
8097 | break; | |
8098 | } | |
8099 | } | |
8100 | } else { | |
8101 | ret = pret < 0 ? pret : -EFAULT; | |
8102 | } | |
d8ed45c5 | 8103 | mmap_read_unlock(current->mm); |
edafccee JA |
8104 | if (ret) { |
8105 | /* | |
8106 | * if we did partial map, or found file backed vmas, | |
8107 | * release any pages we did get | |
8108 | */ | |
27c4d3a3 | 8109 | if (pret > 0) |
f1f6a7dd | 8110 | unpin_user_pages(pages, pret); |
de293938 JA |
8111 | kvfree(imu->bvec); |
8112 | goto err; | |
8113 | } | |
8114 | ||
8115 | ret = io_buffer_account_pin(ctx, pages, pret, imu, &last_hpage); | |
8116 | if (ret) { | |
8117 | unpin_user_pages(pages, pret); | |
d4ef6475 | 8118 | kvfree(imu->bvec); |
edafccee JA |
8119 | goto err; |
8120 | } | |
8121 | ||
8122 | off = ubuf & ~PAGE_MASK; | |
8123 | size = iov.iov_len; | |
8124 | for (j = 0; j < nr_pages; j++) { | |
8125 | size_t vec_len; | |
8126 | ||
8127 | vec_len = min_t(size_t, size, PAGE_SIZE - off); | |
8128 | imu->bvec[j].bv_page = pages[j]; | |
8129 | imu->bvec[j].bv_len = vec_len; | |
8130 | imu->bvec[j].bv_offset = off; | |
8131 | off = 0; | |
8132 | size -= vec_len; | |
8133 | } | |
8134 | /* store original address for later verification */ | |
8135 | imu->ubuf = ubuf; | |
8136 | imu->len = iov.iov_len; | |
8137 | imu->nr_bvecs = nr_pages; | |
8138 | ||
8139 | ctx->nr_user_bufs++; | |
8140 | } | |
d4ef6475 MR |
8141 | kvfree(pages); |
8142 | kvfree(vmas); | |
edafccee JA |
8143 | return 0; |
8144 | err: | |
d4ef6475 MR |
8145 | kvfree(pages); |
8146 | kvfree(vmas); | |
edafccee JA |
8147 | io_sqe_buffer_unregister(ctx); |
8148 | return ret; | |
8149 | } | |
8150 | ||
9b402849 JA |
8151 | static int io_eventfd_register(struct io_ring_ctx *ctx, void __user *arg) |
8152 | { | |
8153 | __s32 __user *fds = arg; | |
8154 | int fd; | |
8155 | ||
8156 | if (ctx->cq_ev_fd) | |
8157 | return -EBUSY; | |
8158 | ||
8159 | if (copy_from_user(&fd, fds, sizeof(*fds))) | |
8160 | return -EFAULT; | |
8161 | ||
8162 | ctx->cq_ev_fd = eventfd_ctx_fdget(fd); | |
8163 | if (IS_ERR(ctx->cq_ev_fd)) { | |
8164 | int ret = PTR_ERR(ctx->cq_ev_fd); | |
8165 | ctx->cq_ev_fd = NULL; | |
8166 | return ret; | |
8167 | } | |
8168 | ||
8169 | return 0; | |
8170 | } | |
8171 | ||
8172 | static int io_eventfd_unregister(struct io_ring_ctx *ctx) | |
8173 | { | |
8174 | if (ctx->cq_ev_fd) { | |
8175 | eventfd_ctx_put(ctx->cq_ev_fd); | |
8176 | ctx->cq_ev_fd = NULL; | |
8177 | return 0; | |
8178 | } | |
8179 | ||
8180 | return -ENXIO; | |
8181 | } | |
8182 | ||
5a2e745d JA |
8183 | static int __io_destroy_buffers(int id, void *p, void *data) |
8184 | { | |
8185 | struct io_ring_ctx *ctx = data; | |
8186 | struct io_buffer *buf = p; | |
8187 | ||
067524e9 | 8188 | __io_remove_buffers(ctx, buf, id, -1U); |
5a2e745d JA |
8189 | return 0; |
8190 | } | |
8191 | ||
8192 | static void io_destroy_buffers(struct io_ring_ctx *ctx) | |
8193 | { | |
8194 | idr_for_each(&ctx->io_buffer_idr, __io_destroy_buffers, ctx); | |
8195 | idr_destroy(&ctx->io_buffer_idr); | |
8196 | } | |
8197 | ||
2b188cc1 JA |
8198 | static void io_ring_ctx_free(struct io_ring_ctx *ctx) |
8199 | { | |
6b06314c | 8200 | io_finish_async(ctx); |
5dbcad51 | 8201 | io_sqe_buffer_unregister(ctx); |
2aede0e4 JA |
8202 | |
8203 | if (ctx->sqo_task) { | |
8204 | put_task_struct(ctx->sqo_task); | |
8205 | ctx->sqo_task = NULL; | |
8206 | mmdrop(ctx->mm_account); | |
8207 | ctx->mm_account = NULL; | |
30975825 | 8208 | } |
def596e9 | 8209 | |
91d8f519 DZ |
8210 | #ifdef CONFIG_BLK_CGROUP |
8211 | if (ctx->sqo_blkcg_css) | |
8212 | css_put(ctx->sqo_blkcg_css); | |
8213 | #endif | |
8214 | ||
6b06314c | 8215 | io_sqe_files_unregister(ctx); |
9b402849 | 8216 | io_eventfd_unregister(ctx); |
5a2e745d | 8217 | io_destroy_buffers(ctx); |
41726c9a | 8218 | idr_destroy(&ctx->personality_idr); |
def596e9 | 8219 | |
2b188cc1 | 8220 | #if defined(CONFIG_UNIX) |
355e8d26 EB |
8221 | if (ctx->ring_sock) { |
8222 | ctx->ring_sock->file = NULL; /* so that iput() is called */ | |
2b188cc1 | 8223 | sock_release(ctx->ring_sock); |
355e8d26 | 8224 | } |
2b188cc1 JA |
8225 | #endif |
8226 | ||
75b28aff | 8227 | io_mem_free(ctx->rings); |
2b188cc1 | 8228 | io_mem_free(ctx->sq_sqes); |
2b188cc1 JA |
8229 | |
8230 | percpu_ref_exit(&ctx->refs); | |
2b188cc1 | 8231 | free_uid(ctx->user); |
181e448d | 8232 | put_cred(ctx->creds); |
78076bb6 | 8233 | kfree(ctx->cancel_hash); |
0ddf92e8 | 8234 | kmem_cache_free(req_cachep, ctx->fallback_req); |
2b188cc1 JA |
8235 | kfree(ctx); |
8236 | } | |
8237 | ||
8238 | static __poll_t io_uring_poll(struct file *file, poll_table *wait) | |
8239 | { | |
8240 | struct io_ring_ctx *ctx = file->private_data; | |
8241 | __poll_t mask = 0; | |
8242 | ||
8243 | poll_wait(file, &ctx->cq_wait, wait); | |
4f7067c3 SB |
8244 | /* |
8245 | * synchronizes with barrier from wq_has_sleeper call in | |
8246 | * io_commit_cqring | |
8247 | */ | |
2b188cc1 | 8248 | smp_rmb(); |
90554200 | 8249 | if (!io_sqring_full(ctx)) |
2b188cc1 | 8250 | mask |= EPOLLOUT | EPOLLWRNORM; |
63e5d81f | 8251 | if (io_cqring_events(ctx, false)) |
2b188cc1 JA |
8252 | mask |= EPOLLIN | EPOLLRDNORM; |
8253 | ||
8254 | return mask; | |
8255 | } | |
8256 | ||
8257 | static int io_uring_fasync(int fd, struct file *file, int on) | |
8258 | { | |
8259 | struct io_ring_ctx *ctx = file->private_data; | |
8260 | ||
8261 | return fasync_helper(fd, file, on, &ctx->cq_fasync); | |
8262 | } | |
8263 | ||
071698e1 JA |
8264 | static int io_remove_personalities(int id, void *p, void *data) |
8265 | { | |
8266 | struct io_ring_ctx *ctx = data; | |
8267 | const struct cred *cred; | |
8268 | ||
8269 | cred = idr_remove(&ctx->personality_idr, id); | |
8270 | if (cred) | |
8271 | put_cred(cred); | |
8272 | return 0; | |
8273 | } | |
8274 | ||
85faa7b8 JA |
8275 | static void io_ring_exit_work(struct work_struct *work) |
8276 | { | |
b2edc0a7 PB |
8277 | struct io_ring_ctx *ctx = container_of(work, struct io_ring_ctx, |
8278 | exit_work); | |
85faa7b8 | 8279 | |
56952e91 JA |
8280 | /* |
8281 | * If we're doing polled IO and end up having requests being | |
8282 | * submitted async (out-of-line), then completions can come in while | |
8283 | * we're waiting for refs to drop. We need to reap these manually, | |
8284 | * as nobody else will be looking for them. | |
8285 | */ | |
b2edc0a7 | 8286 | do { |
56952e91 | 8287 | if (ctx->rings) |
e6c8aa9a | 8288 | io_cqring_overflow_flush(ctx, true, NULL, NULL); |
b2edc0a7 PB |
8289 | io_iopoll_try_reap_events(ctx); |
8290 | } while (!wait_for_completion_timeout(&ctx->ref_comp, HZ/20)); | |
85faa7b8 JA |
8291 | io_ring_ctx_free(ctx); |
8292 | } | |
8293 | ||
2b188cc1 JA |
8294 | static void io_ring_ctx_wait_and_kill(struct io_ring_ctx *ctx) |
8295 | { | |
8296 | mutex_lock(&ctx->uring_lock); | |
8297 | percpu_ref_kill(&ctx->refs); | |
8298 | mutex_unlock(&ctx->uring_lock); | |
8299 | ||
f3606e3a JA |
8300 | io_kill_timeouts(ctx, NULL); |
8301 | io_poll_remove_all(ctx, NULL); | |
561fb04a JA |
8302 | |
8303 | if (ctx->io_wq) | |
8304 | io_wq_cancel_all(ctx->io_wq); | |
8305 | ||
15dff286 JA |
8306 | /* if we failed setting up the ctx, we might not have any rings */ |
8307 | if (ctx->rings) | |
e6c8aa9a | 8308 | io_cqring_overflow_flush(ctx, true, NULL, NULL); |
b2edc0a7 | 8309 | io_iopoll_try_reap_events(ctx); |
071698e1 | 8310 | idr_for_each(&ctx->personality_idr, io_remove_personalities, ctx); |
309fc03a JA |
8311 | |
8312 | /* | |
8313 | * Do this upfront, so we won't have a grace period where the ring | |
8314 | * is closed but resources aren't reaped yet. This can cause | |
8315 | * spurious failure in setting up a new ring. | |
8316 | */ | |
760618f7 JA |
8317 | io_unaccount_mem(ctx, ring_pages(ctx->sq_entries, ctx->cq_entries), |
8318 | ACCT_LOCKED); | |
309fc03a | 8319 | |
85faa7b8 | 8320 | INIT_WORK(&ctx->exit_work, io_ring_exit_work); |
fc666777 JA |
8321 | /* |
8322 | * Use system_unbound_wq to avoid spawning tons of event kworkers | |
8323 | * if we're exiting a ton of rings at the same time. It just adds | |
8324 | * noise and overhead, there's no discernable change in runtime | |
8325 | * over using system_wq. | |
8326 | */ | |
8327 | queue_work(system_unbound_wq, &ctx->exit_work); | |
2b188cc1 JA |
8328 | } |
8329 | ||
8330 | static int io_uring_release(struct inode *inode, struct file *file) | |
8331 | { | |
8332 | struct io_ring_ctx *ctx = file->private_data; | |
8333 | ||
8334 | file->private_data = NULL; | |
8335 | io_ring_ctx_wait_and_kill(ctx); | |
8336 | return 0; | |
8337 | } | |
8338 | ||
67c4d9e6 PB |
8339 | static bool io_wq_files_match(struct io_wq_work *work, void *data) |
8340 | { | |
8341 | struct files_struct *files = data; | |
8342 | ||
0f212204 | 8343 | return !files || work->files == files; |
67c4d9e6 PB |
8344 | } |
8345 | ||
f254ac04 JA |
8346 | /* |
8347 | * Returns true if 'preq' is the link parent of 'req' | |
8348 | */ | |
8349 | static bool io_match_link(struct io_kiocb *preq, struct io_kiocb *req) | |
8350 | { | |
8351 | struct io_kiocb *link; | |
8352 | ||
8353 | if (!(preq->flags & REQ_F_LINK_HEAD)) | |
8354 | return false; | |
8355 | ||
8356 | list_for_each_entry(link, &preq->link_list, link_list) { | |
8357 | if (link == req) | |
8358 | return true; | |
8359 | } | |
8360 | ||
8361 | return false; | |
8362 | } | |
8363 | ||
c127a2a1 PB |
8364 | static bool io_match_link_files(struct io_kiocb *req, |
8365 | struct files_struct *files) | |
8366 | { | |
8367 | struct io_kiocb *link; | |
8368 | ||
8369 | if (io_match_files(req, files)) | |
8370 | return true; | |
8371 | if (req->flags & REQ_F_LINK_HEAD) { | |
8372 | list_for_each_entry(link, &req->link_list, link_list) { | |
8373 | if (io_match_files(link, files)) | |
8374 | return true; | |
8375 | } | |
8376 | } | |
8377 | return false; | |
8378 | } | |
8379 | ||
f254ac04 JA |
8380 | /* |
8381 | * We're looking to cancel 'req' because it's holding on to our files, but | |
8382 | * 'req' could be a link to another request. See if it is, and cancel that | |
8383 | * parent request if so. | |
8384 | */ | |
8385 | static bool io_poll_remove_link(struct io_ring_ctx *ctx, struct io_kiocb *req) | |
8386 | { | |
8387 | struct hlist_node *tmp; | |
8388 | struct io_kiocb *preq; | |
8389 | bool found = false; | |
8390 | int i; | |
8391 | ||
8392 | spin_lock_irq(&ctx->completion_lock); | |
8393 | for (i = 0; i < (1U << ctx->cancel_hash_bits); i++) { | |
8394 | struct hlist_head *list; | |
8395 | ||
8396 | list = &ctx->cancel_hash[i]; | |
8397 | hlist_for_each_entry_safe(preq, tmp, list, hash_node) { | |
8398 | found = io_match_link(preq, req); | |
8399 | if (found) { | |
8400 | io_poll_remove_one(preq); | |
8401 | break; | |
8402 | } | |
8403 | } | |
8404 | } | |
8405 | spin_unlock_irq(&ctx->completion_lock); | |
8406 | return found; | |
8407 | } | |
8408 | ||
8409 | static bool io_timeout_remove_link(struct io_ring_ctx *ctx, | |
8410 | struct io_kiocb *req) | |
8411 | { | |
8412 | struct io_kiocb *preq; | |
8413 | bool found = false; | |
8414 | ||
8415 | spin_lock_irq(&ctx->completion_lock); | |
8416 | list_for_each_entry(preq, &ctx->timeout_list, timeout.list) { | |
8417 | found = io_match_link(preq, req); | |
8418 | if (found) { | |
8419 | __io_timeout_cancel(preq); | |
8420 | break; | |
8421 | } | |
8422 | } | |
8423 | spin_unlock_irq(&ctx->completion_lock); | |
8424 | return found; | |
8425 | } | |
8426 | ||
b711d4ea JA |
8427 | static bool io_cancel_link_cb(struct io_wq_work *work, void *data) |
8428 | { | |
8429 | return io_match_link(container_of(work, struct io_kiocb, work), data); | |
8430 | } | |
8431 | ||
8432 | static void io_attempt_cancel(struct io_ring_ctx *ctx, struct io_kiocb *req) | |
8433 | { | |
8434 | enum io_wq_cancel cret; | |
8435 | ||
8436 | /* cancel this particular work, if it's running */ | |
8437 | cret = io_wq_cancel_work(ctx->io_wq, &req->work); | |
8438 | if (cret != IO_WQ_CANCEL_NOTFOUND) | |
8439 | return; | |
8440 | ||
8441 | /* find links that hold this pending, cancel those */ | |
8442 | cret = io_wq_cancel_cb(ctx->io_wq, io_cancel_link_cb, req, true); | |
8443 | if (cret != IO_WQ_CANCEL_NOTFOUND) | |
8444 | return; | |
8445 | ||
8446 | /* if we have a poll link holding this pending, cancel that */ | |
8447 | if (io_poll_remove_link(ctx, req)) | |
8448 | return; | |
8449 | ||
8450 | /* final option, timeout link is holding this req pending */ | |
8451 | io_timeout_remove_link(ctx, req); | |
8452 | } | |
8453 | ||
b7ddce3c PB |
8454 | static void io_cancel_defer_files(struct io_ring_ctx *ctx, |
8455 | struct files_struct *files) | |
8456 | { | |
8457 | struct io_defer_entry *de = NULL; | |
8458 | LIST_HEAD(list); | |
8459 | ||
8460 | spin_lock_irq(&ctx->completion_lock); | |
8461 | list_for_each_entry_reverse(de, &ctx->defer_list, list) { | |
c127a2a1 | 8462 | if (io_match_link_files(de->req, files)) { |
b7ddce3c PB |
8463 | list_cut_position(&list, &ctx->defer_list, &de->list); |
8464 | break; | |
8465 | } | |
8466 | } | |
8467 | spin_unlock_irq(&ctx->completion_lock); | |
8468 | ||
8469 | while (!list_empty(&list)) { | |
8470 | de = list_first_entry(&list, struct io_defer_entry, list); | |
8471 | list_del_init(&de->list); | |
8472 | req_set_fail_links(de->req); | |
8473 | io_put_req(de->req); | |
8474 | io_req_complete(de->req, -ECANCELED); | |
8475 | kfree(de); | |
8476 | } | |
8477 | } | |
8478 | ||
76e1b642 JA |
8479 | /* |
8480 | * Returns true if we found and killed one or more files pinning requests | |
8481 | */ | |
8482 | static bool io_uring_cancel_files(struct io_ring_ctx *ctx, | |
fcb323cc JA |
8483 | struct files_struct *files) |
8484 | { | |
67c4d9e6 | 8485 | if (list_empty_careful(&ctx->inflight_list)) |
76e1b642 | 8486 | return false; |
67c4d9e6 | 8487 | |
b7ddce3c | 8488 | io_cancel_defer_files(ctx, files); |
67c4d9e6 PB |
8489 | /* cancel all at once, should be faster than doing it one by one*/ |
8490 | io_wq_cancel_cb(ctx->io_wq, io_wq_files_match, files, true); | |
8491 | ||
fcb323cc | 8492 | while (!list_empty_careful(&ctx->inflight_list)) { |
d8f1b971 XW |
8493 | struct io_kiocb *cancel_req = NULL, *req; |
8494 | DEFINE_WAIT(wait); | |
fcb323cc JA |
8495 | |
8496 | spin_lock_irq(&ctx->inflight_lock); | |
8497 | list_for_each_entry(req, &ctx->inflight_list, inflight_entry) { | |
0f212204 | 8498 | if (files && req->work.files != files) |
768134d4 JA |
8499 | continue; |
8500 | /* req is being completed, ignore */ | |
8501 | if (!refcount_inc_not_zero(&req->refs)) | |
8502 | continue; | |
8503 | cancel_req = req; | |
8504 | break; | |
fcb323cc | 8505 | } |
768134d4 | 8506 | if (cancel_req) |
fcb323cc | 8507 | prepare_to_wait(&ctx->inflight_wait, &wait, |
768134d4 | 8508 | TASK_UNINTERRUPTIBLE); |
fcb323cc JA |
8509 | spin_unlock_irq(&ctx->inflight_lock); |
8510 | ||
768134d4 JA |
8511 | /* We need to keep going until we don't find a matching req */ |
8512 | if (!cancel_req) | |
fcb323cc | 8513 | break; |
bb175342 PB |
8514 | /* cancel this request, or head link requests */ |
8515 | io_attempt_cancel(ctx, cancel_req); | |
8516 | io_put_req(cancel_req); | |
6200b0ae JA |
8517 | /* cancellations _may_ trigger task work */ |
8518 | io_run_task_work(); | |
fcb323cc | 8519 | schedule(); |
d8f1b971 | 8520 | finish_wait(&ctx->inflight_wait, &wait); |
fcb323cc | 8521 | } |
76e1b642 JA |
8522 | |
8523 | return true; | |
fcb323cc JA |
8524 | } |
8525 | ||
801dd57b | 8526 | static bool io_cancel_task_cb(struct io_wq_work *work, void *data) |
44e728b8 | 8527 | { |
801dd57b PB |
8528 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); |
8529 | struct task_struct *task = data; | |
44e728b8 | 8530 | |
f3606e3a | 8531 | return io_task_match(req, task); |
44e728b8 PB |
8532 | } |
8533 | ||
0f212204 JA |
8534 | static bool __io_uring_cancel_task_requests(struct io_ring_ctx *ctx, |
8535 | struct task_struct *task, | |
8536 | struct files_struct *files) | |
8537 | { | |
8538 | bool ret; | |
8539 | ||
8540 | ret = io_uring_cancel_files(ctx, files); | |
8541 | if (!files) { | |
8542 | enum io_wq_cancel cret; | |
8543 | ||
8544 | cret = io_wq_cancel_cb(ctx->io_wq, io_cancel_task_cb, task, true); | |
8545 | if (cret != IO_WQ_CANCEL_NOTFOUND) | |
8546 | ret = true; | |
8547 | ||
8548 | /* SQPOLL thread does its own polling */ | |
8549 | if (!(ctx->flags & IORING_SETUP_SQPOLL)) { | |
8550 | while (!list_empty_careful(&ctx->iopoll_list)) { | |
8551 | io_iopoll_try_reap_events(ctx); | |
8552 | ret = true; | |
8553 | } | |
8554 | } | |
8555 | ||
8556 | ret |= io_poll_remove_all(ctx, task); | |
8557 | ret |= io_kill_timeouts(ctx, task); | |
8558 | } | |
8559 | ||
8560 | return ret; | |
8561 | } | |
8562 | ||
8563 | /* | |
8564 | * We need to iteratively cancel requests, in case a request has dependent | |
8565 | * hard links. These persist even for failure of cancelations, hence keep | |
8566 | * looping until none are found. | |
8567 | */ | |
8568 | static void io_uring_cancel_task_requests(struct io_ring_ctx *ctx, | |
8569 | struct files_struct *files) | |
8570 | { | |
8571 | struct task_struct *task = current; | |
8572 | ||
534ca6d6 JA |
8573 | if ((ctx->flags & IORING_SETUP_SQPOLL) && ctx->sq_data) |
8574 | task = ctx->sq_data->thread; | |
0f212204 JA |
8575 | |
8576 | io_cqring_overflow_flush(ctx, true, task, files); | |
8577 | ||
8578 | while (__io_uring_cancel_task_requests(ctx, task, files)) { | |
8579 | io_run_task_work(); | |
8580 | cond_resched(); | |
8581 | } | |
8582 | } | |
8583 | ||
8584 | /* | |
8585 | * Note that this task has used io_uring. We use it for cancelation purposes. | |
8586 | */ | |
8587 | static int io_uring_add_task_file(struct file *file) | |
8588 | { | |
8589 | if (unlikely(!current->io_uring)) { | |
8590 | int ret; | |
8591 | ||
8592 | ret = io_uring_alloc_task_context(current); | |
8593 | if (unlikely(ret)) | |
8594 | return ret; | |
8595 | } | |
8596 | if (current->io_uring->last != file) { | |
8597 | XA_STATE(xas, ¤t->io_uring->xa, (unsigned long) file); | |
8598 | void *old; | |
8599 | ||
8600 | rcu_read_lock(); | |
8601 | old = xas_load(&xas); | |
8602 | if (old != file) { | |
8603 | get_file(file); | |
8604 | xas_lock(&xas); | |
8605 | xas_store(&xas, file); | |
8606 | xas_unlock(&xas); | |
8607 | } | |
8608 | rcu_read_unlock(); | |
8609 | current->io_uring->last = file; | |
8610 | } | |
8611 | ||
8612 | return 0; | |
8613 | } | |
8614 | ||
8615 | /* | |
8616 | * Remove this io_uring_file -> task mapping. | |
8617 | */ | |
8618 | static void io_uring_del_task_file(struct file *file) | |
8619 | { | |
8620 | struct io_uring_task *tctx = current->io_uring; | |
8621 | XA_STATE(xas, &tctx->xa, (unsigned long) file); | |
8622 | ||
8623 | if (tctx->last == file) | |
8624 | tctx->last = NULL; | |
8625 | ||
8626 | xas_lock(&xas); | |
8627 | file = xas_store(&xas, NULL); | |
8628 | xas_unlock(&xas); | |
8629 | ||
8630 | if (file) | |
8631 | fput(file); | |
8632 | } | |
8633 | ||
8634 | static void __io_uring_attempt_task_drop(struct file *file) | |
8635 | { | |
8636 | XA_STATE(xas, ¤t->io_uring->xa, (unsigned long) file); | |
8637 | struct file *old; | |
8638 | ||
8639 | rcu_read_lock(); | |
8640 | old = xas_load(&xas); | |
8641 | rcu_read_unlock(); | |
8642 | ||
8643 | if (old == file) | |
8644 | io_uring_del_task_file(file); | |
8645 | } | |
8646 | ||
8647 | /* | |
8648 | * Drop task note for this file if we're the only ones that hold it after | |
8649 | * pending fput() | |
8650 | */ | |
8651 | static void io_uring_attempt_task_drop(struct file *file, bool exiting) | |
8652 | { | |
8653 | if (!current->io_uring) | |
8654 | return; | |
8655 | /* | |
8656 | * fput() is pending, will be 2 if the only other ref is our potential | |
8657 | * task file note. If the task is exiting, drop regardless of count. | |
8658 | */ | |
8659 | if (!exiting && atomic_long_read(&file->f_count) != 2) | |
8660 | return; | |
8661 | ||
8662 | __io_uring_attempt_task_drop(file); | |
8663 | } | |
8664 | ||
8665 | void __io_uring_files_cancel(struct files_struct *files) | |
8666 | { | |
8667 | struct io_uring_task *tctx = current->io_uring; | |
8668 | XA_STATE(xas, &tctx->xa, 0); | |
8669 | ||
8670 | /* make sure overflow events are dropped */ | |
8671 | tctx->in_idle = true; | |
8672 | ||
8673 | do { | |
8674 | struct io_ring_ctx *ctx; | |
8675 | struct file *file; | |
8676 | ||
8677 | xas_lock(&xas); | |
8678 | file = xas_next_entry(&xas, ULONG_MAX); | |
8679 | xas_unlock(&xas); | |
8680 | ||
8681 | if (!file) | |
8682 | break; | |
8683 | ||
8684 | ctx = file->private_data; | |
8685 | ||
8686 | io_uring_cancel_task_requests(ctx, files); | |
8687 | if (files) | |
8688 | io_uring_del_task_file(file); | |
8689 | } while (1); | |
8690 | } | |
8691 | ||
8692 | static inline bool io_uring_task_idle(struct io_uring_task *tctx) | |
8693 | { | |
8694 | return atomic_long_read(&tctx->req_issue) == | |
8695 | atomic_long_read(&tctx->req_complete); | |
8696 | } | |
8697 | ||
8698 | /* | |
8699 | * Find any io_uring fd that this task has registered or done IO on, and cancel | |
8700 | * requests. | |
8701 | */ | |
8702 | void __io_uring_task_cancel(void) | |
8703 | { | |
8704 | struct io_uring_task *tctx = current->io_uring; | |
8705 | DEFINE_WAIT(wait); | |
8706 | long completions; | |
8707 | ||
8708 | /* make sure overflow events are dropped */ | |
8709 | tctx->in_idle = true; | |
8710 | ||
8711 | while (!io_uring_task_idle(tctx)) { | |
8712 | /* read completions before cancelations */ | |
8713 | completions = atomic_long_read(&tctx->req_complete); | |
8714 | __io_uring_files_cancel(NULL); | |
8715 | ||
8716 | prepare_to_wait(&tctx->wait, &wait, TASK_UNINTERRUPTIBLE); | |
8717 | ||
8718 | /* | |
8719 | * If we've seen completions, retry. This avoids a race where | |
8720 | * a completion comes in before we did prepare_to_wait(). | |
8721 | */ | |
8722 | if (completions != atomic_long_read(&tctx->req_complete)) | |
8723 | continue; | |
8724 | if (io_uring_task_idle(tctx)) | |
8725 | break; | |
8726 | schedule(); | |
8727 | } | |
8728 | ||
8729 | finish_wait(&tctx->wait, &wait); | |
8730 | tctx->in_idle = false; | |
8731 | } | |
8732 | ||
fcb323cc JA |
8733 | static int io_uring_flush(struct file *file, void *data) |
8734 | { | |
8735 | struct io_ring_ctx *ctx = file->private_data; | |
8736 | ||
6ab23144 JA |
8737 | /* |
8738 | * If the task is going away, cancel work it may have pending | |
8739 | */ | |
801dd57b | 8740 | if (fatal_signal_pending(current) || (current->flags & PF_EXITING)) |
0f212204 | 8741 | data = NULL; |
6ab23144 | 8742 | |
0f212204 JA |
8743 | io_uring_cancel_task_requests(ctx, data); |
8744 | io_uring_attempt_task_drop(file, !data); | |
fcb323cc JA |
8745 | return 0; |
8746 | } | |
8747 | ||
6c5c240e RP |
8748 | static void *io_uring_validate_mmap_request(struct file *file, |
8749 | loff_t pgoff, size_t sz) | |
2b188cc1 | 8750 | { |
2b188cc1 | 8751 | struct io_ring_ctx *ctx = file->private_data; |
6c5c240e | 8752 | loff_t offset = pgoff << PAGE_SHIFT; |
2b188cc1 JA |
8753 | struct page *page; |
8754 | void *ptr; | |
8755 | ||
8756 | switch (offset) { | |
8757 | case IORING_OFF_SQ_RING: | |
75b28aff HV |
8758 | case IORING_OFF_CQ_RING: |
8759 | ptr = ctx->rings; | |
2b188cc1 JA |
8760 | break; |
8761 | case IORING_OFF_SQES: | |
8762 | ptr = ctx->sq_sqes; | |
8763 | break; | |
2b188cc1 | 8764 | default: |
6c5c240e | 8765 | return ERR_PTR(-EINVAL); |
2b188cc1 JA |
8766 | } |
8767 | ||
8768 | page = virt_to_head_page(ptr); | |
a50b854e | 8769 | if (sz > page_size(page)) |
6c5c240e RP |
8770 | return ERR_PTR(-EINVAL); |
8771 | ||
8772 | return ptr; | |
8773 | } | |
8774 | ||
8775 | #ifdef CONFIG_MMU | |
8776 | ||
8777 | static int io_uring_mmap(struct file *file, struct vm_area_struct *vma) | |
8778 | { | |
8779 | size_t sz = vma->vm_end - vma->vm_start; | |
8780 | unsigned long pfn; | |
8781 | void *ptr; | |
8782 | ||
8783 | ptr = io_uring_validate_mmap_request(file, vma->vm_pgoff, sz); | |
8784 | if (IS_ERR(ptr)) | |
8785 | return PTR_ERR(ptr); | |
2b188cc1 JA |
8786 | |
8787 | pfn = virt_to_phys(ptr) >> PAGE_SHIFT; | |
8788 | return remap_pfn_range(vma, vma->vm_start, pfn, sz, vma->vm_page_prot); | |
8789 | } | |
8790 | ||
6c5c240e RP |
8791 | #else /* !CONFIG_MMU */ |
8792 | ||
8793 | static int io_uring_mmap(struct file *file, struct vm_area_struct *vma) | |
8794 | { | |
8795 | return vma->vm_flags & (VM_SHARED | VM_MAYSHARE) ? 0 : -EINVAL; | |
8796 | } | |
8797 | ||
8798 | static unsigned int io_uring_nommu_mmap_capabilities(struct file *file) | |
8799 | { | |
8800 | return NOMMU_MAP_DIRECT | NOMMU_MAP_READ | NOMMU_MAP_WRITE; | |
8801 | } | |
8802 | ||
8803 | static unsigned long io_uring_nommu_get_unmapped_area(struct file *file, | |
8804 | unsigned long addr, unsigned long len, | |
8805 | unsigned long pgoff, unsigned long flags) | |
8806 | { | |
8807 | void *ptr; | |
8808 | ||
8809 | ptr = io_uring_validate_mmap_request(file, pgoff, len); | |
8810 | if (IS_ERR(ptr)) | |
8811 | return PTR_ERR(ptr); | |
8812 | ||
8813 | return (unsigned long) ptr; | |
8814 | } | |
8815 | ||
8816 | #endif /* !CONFIG_MMU */ | |
8817 | ||
90554200 JA |
8818 | static void io_sqpoll_wait_sq(struct io_ring_ctx *ctx) |
8819 | { | |
8820 | DEFINE_WAIT(wait); | |
8821 | ||
8822 | do { | |
8823 | if (!io_sqring_full(ctx)) | |
8824 | break; | |
8825 | ||
8826 | prepare_to_wait(&ctx->sqo_sq_wait, &wait, TASK_INTERRUPTIBLE); | |
8827 | ||
8828 | if (!io_sqring_full(ctx)) | |
8829 | break; | |
8830 | ||
8831 | schedule(); | |
8832 | } while (!signal_pending(current)); | |
8833 | ||
8834 | finish_wait(&ctx->sqo_sq_wait, &wait); | |
8835 | } | |
8836 | ||
2b188cc1 JA |
8837 | SYSCALL_DEFINE6(io_uring_enter, unsigned int, fd, u32, to_submit, |
8838 | u32, min_complete, u32, flags, const sigset_t __user *, sig, | |
8839 | size_t, sigsz) | |
8840 | { | |
8841 | struct io_ring_ctx *ctx; | |
8842 | long ret = -EBADF; | |
8843 | int submitted = 0; | |
8844 | struct fd f; | |
8845 | ||
4c6e277c | 8846 | io_run_task_work(); |
b41e9852 | 8847 | |
90554200 JA |
8848 | if (flags & ~(IORING_ENTER_GETEVENTS | IORING_ENTER_SQ_WAKEUP | |
8849 | IORING_ENTER_SQ_WAIT)) | |
2b188cc1 JA |
8850 | return -EINVAL; |
8851 | ||
8852 | f = fdget(fd); | |
8853 | if (!f.file) | |
8854 | return -EBADF; | |
8855 | ||
8856 | ret = -EOPNOTSUPP; | |
8857 | if (f.file->f_op != &io_uring_fops) | |
8858 | goto out_fput; | |
8859 | ||
8860 | ret = -ENXIO; | |
8861 | ctx = f.file->private_data; | |
8862 | if (!percpu_ref_tryget(&ctx->refs)) | |
8863 | goto out_fput; | |
8864 | ||
7e84e1c7 SG |
8865 | ret = -EBADFD; |
8866 | if (ctx->flags & IORING_SETUP_R_DISABLED) | |
8867 | goto out; | |
8868 | ||
6c271ce2 JA |
8869 | /* |
8870 | * For SQ polling, the thread will do all submissions and completions. | |
8871 | * Just return the requested submit count, and wake the thread if | |
8872 | * we were asked to. | |
8873 | */ | |
b2a9eada | 8874 | ret = 0; |
6c271ce2 | 8875 | if (ctx->flags & IORING_SETUP_SQPOLL) { |
c1edbf5f | 8876 | if (!list_empty_careful(&ctx->cq_overflow_list)) |
e6c8aa9a | 8877 | io_cqring_overflow_flush(ctx, false, NULL, NULL); |
6c271ce2 | 8878 | if (flags & IORING_ENTER_SQ_WAKEUP) |
534ca6d6 | 8879 | wake_up(&ctx->sq_data->wait); |
90554200 JA |
8880 | if (flags & IORING_ENTER_SQ_WAIT) |
8881 | io_sqpoll_wait_sq(ctx); | |
6c271ce2 | 8882 | submitted = to_submit; |
b2a9eada | 8883 | } else if (to_submit) { |
0f212204 JA |
8884 | ret = io_uring_add_task_file(f.file); |
8885 | if (unlikely(ret)) | |
8886 | goto out; | |
2b188cc1 | 8887 | mutex_lock(&ctx->uring_lock); |
0f212204 | 8888 | submitted = io_submit_sqes(ctx, to_submit); |
2b188cc1 | 8889 | mutex_unlock(&ctx->uring_lock); |
7c504e65 PB |
8890 | |
8891 | if (submitted != to_submit) | |
8892 | goto out; | |
2b188cc1 JA |
8893 | } |
8894 | if (flags & IORING_ENTER_GETEVENTS) { | |
8895 | min_complete = min(min_complete, ctx->cq_entries); | |
8896 | ||
32b2244a XW |
8897 | /* |
8898 | * When SETUP_IOPOLL and SETUP_SQPOLL are both enabled, user | |
8899 | * space applications don't need to do io completion events | |
8900 | * polling again, they can rely on io_sq_thread to do polling | |
8901 | * work, which can reduce cpu usage and uring_lock contention. | |
8902 | */ | |
8903 | if (ctx->flags & IORING_SETUP_IOPOLL && | |
8904 | !(ctx->flags & IORING_SETUP_SQPOLL)) { | |
7668b92a | 8905 | ret = io_iopoll_check(ctx, min_complete); |
def596e9 JA |
8906 | } else { |
8907 | ret = io_cqring_wait(ctx, min_complete, sig, sigsz); | |
8908 | } | |
2b188cc1 JA |
8909 | } |
8910 | ||
7c504e65 | 8911 | out: |
6805b32e | 8912 | percpu_ref_put(&ctx->refs); |
2b188cc1 JA |
8913 | out_fput: |
8914 | fdput(f); | |
8915 | return submitted ? submitted : ret; | |
8916 | } | |
8917 | ||
bebdb65e | 8918 | #ifdef CONFIG_PROC_FS |
87ce955b JA |
8919 | static int io_uring_show_cred(int id, void *p, void *data) |
8920 | { | |
8921 | const struct cred *cred = p; | |
8922 | struct seq_file *m = data; | |
8923 | struct user_namespace *uns = seq_user_ns(m); | |
8924 | struct group_info *gi; | |
8925 | kernel_cap_t cap; | |
8926 | unsigned __capi; | |
8927 | int g; | |
8928 | ||
8929 | seq_printf(m, "%5d\n", id); | |
8930 | seq_put_decimal_ull(m, "\tUid:\t", from_kuid_munged(uns, cred->uid)); | |
8931 | seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->euid)); | |
8932 | seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->suid)); | |
8933 | seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->fsuid)); | |
8934 | seq_put_decimal_ull(m, "\n\tGid:\t", from_kgid_munged(uns, cred->gid)); | |
8935 | seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->egid)); | |
8936 | seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->sgid)); | |
8937 | seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->fsgid)); | |
8938 | seq_puts(m, "\n\tGroups:\t"); | |
8939 | gi = cred->group_info; | |
8940 | for (g = 0; g < gi->ngroups; g++) { | |
8941 | seq_put_decimal_ull(m, g ? " " : "", | |
8942 | from_kgid_munged(uns, gi->gid[g])); | |
8943 | } | |
8944 | seq_puts(m, "\n\tCapEff:\t"); | |
8945 | cap = cred->cap_effective; | |
8946 | CAP_FOR_EACH_U32(__capi) | |
8947 | seq_put_hex_ll(m, NULL, cap.cap[CAP_LAST_U32 - __capi], 8); | |
8948 | seq_putc(m, '\n'); | |
8949 | return 0; | |
8950 | } | |
8951 | ||
8952 | static void __io_uring_show_fdinfo(struct io_ring_ctx *ctx, struct seq_file *m) | |
8953 | { | |
dbbe9c64 | 8954 | struct io_sq_data *sq = NULL; |
fad8e0de | 8955 | bool has_lock; |
87ce955b JA |
8956 | int i; |
8957 | ||
fad8e0de JA |
8958 | /* |
8959 | * Avoid ABBA deadlock between the seq lock and the io_uring mutex, | |
8960 | * since fdinfo case grabs it in the opposite direction of normal use | |
8961 | * cases. If we fail to get the lock, we just don't iterate any | |
8962 | * structures that could be going away outside the io_uring mutex. | |
8963 | */ | |
8964 | has_lock = mutex_trylock(&ctx->uring_lock); | |
8965 | ||
dbbe9c64 JQ |
8966 | if (has_lock && (ctx->flags & IORING_SETUP_SQPOLL)) |
8967 | sq = ctx->sq_data; | |
8968 | ||
8969 | seq_printf(m, "SqThread:\t%d\n", sq ? task_pid_nr(sq->thread) : -1); | |
8970 | seq_printf(m, "SqThreadCpu:\t%d\n", sq ? task_cpu(sq->thread) : -1); | |
87ce955b | 8971 | seq_printf(m, "UserFiles:\t%u\n", ctx->nr_user_files); |
fad8e0de | 8972 | for (i = 0; has_lock && i < ctx->nr_user_files; i++) { |
87ce955b JA |
8973 | struct fixed_file_table *table; |
8974 | struct file *f; | |
8975 | ||
8976 | table = &ctx->file_data->table[i >> IORING_FILE_TABLE_SHIFT]; | |
8977 | f = table->files[i & IORING_FILE_TABLE_MASK]; | |
8978 | if (f) | |
8979 | seq_printf(m, "%5u: %s\n", i, file_dentry(f)->d_iname); | |
8980 | else | |
8981 | seq_printf(m, "%5u: <none>\n", i); | |
8982 | } | |
8983 | seq_printf(m, "UserBufs:\t%u\n", ctx->nr_user_bufs); | |
fad8e0de | 8984 | for (i = 0; has_lock && i < ctx->nr_user_bufs; i++) { |
87ce955b JA |
8985 | struct io_mapped_ubuf *buf = &ctx->user_bufs[i]; |
8986 | ||
8987 | seq_printf(m, "%5u: 0x%llx/%u\n", i, buf->ubuf, | |
8988 | (unsigned int) buf->len); | |
8989 | } | |
fad8e0de | 8990 | if (has_lock && !idr_is_empty(&ctx->personality_idr)) { |
87ce955b JA |
8991 | seq_printf(m, "Personalities:\n"); |
8992 | idr_for_each(&ctx->personality_idr, io_uring_show_cred, m); | |
8993 | } | |
d7718a9d JA |
8994 | seq_printf(m, "PollList:\n"); |
8995 | spin_lock_irq(&ctx->completion_lock); | |
8996 | for (i = 0; i < (1U << ctx->cancel_hash_bits); i++) { | |
8997 | struct hlist_head *list = &ctx->cancel_hash[i]; | |
8998 | struct io_kiocb *req; | |
8999 | ||
9000 | hlist_for_each_entry(req, list, hash_node) | |
9001 | seq_printf(m, " op=%d, task_works=%d\n", req->opcode, | |
9002 | req->task->task_works != NULL); | |
9003 | } | |
9004 | spin_unlock_irq(&ctx->completion_lock); | |
fad8e0de JA |
9005 | if (has_lock) |
9006 | mutex_unlock(&ctx->uring_lock); | |
87ce955b JA |
9007 | } |
9008 | ||
9009 | static void io_uring_show_fdinfo(struct seq_file *m, struct file *f) | |
9010 | { | |
9011 | struct io_ring_ctx *ctx = f->private_data; | |
9012 | ||
9013 | if (percpu_ref_tryget(&ctx->refs)) { | |
9014 | __io_uring_show_fdinfo(ctx, m); | |
9015 | percpu_ref_put(&ctx->refs); | |
9016 | } | |
9017 | } | |
bebdb65e | 9018 | #endif |
87ce955b | 9019 | |
2b188cc1 JA |
9020 | static const struct file_operations io_uring_fops = { |
9021 | .release = io_uring_release, | |
fcb323cc | 9022 | .flush = io_uring_flush, |
2b188cc1 | 9023 | .mmap = io_uring_mmap, |
6c5c240e RP |
9024 | #ifndef CONFIG_MMU |
9025 | .get_unmapped_area = io_uring_nommu_get_unmapped_area, | |
9026 | .mmap_capabilities = io_uring_nommu_mmap_capabilities, | |
9027 | #endif | |
2b188cc1 JA |
9028 | .poll = io_uring_poll, |
9029 | .fasync = io_uring_fasync, | |
bebdb65e | 9030 | #ifdef CONFIG_PROC_FS |
87ce955b | 9031 | .show_fdinfo = io_uring_show_fdinfo, |
bebdb65e | 9032 | #endif |
2b188cc1 JA |
9033 | }; |
9034 | ||
9035 | static int io_allocate_scq_urings(struct io_ring_ctx *ctx, | |
9036 | struct io_uring_params *p) | |
9037 | { | |
75b28aff HV |
9038 | struct io_rings *rings; |
9039 | size_t size, sq_array_offset; | |
2b188cc1 | 9040 | |
bd740481 JA |
9041 | /* make sure these are sane, as we already accounted them */ |
9042 | ctx->sq_entries = p->sq_entries; | |
9043 | ctx->cq_entries = p->cq_entries; | |
9044 | ||
75b28aff HV |
9045 | size = rings_size(p->sq_entries, p->cq_entries, &sq_array_offset); |
9046 | if (size == SIZE_MAX) | |
9047 | return -EOVERFLOW; | |
9048 | ||
9049 | rings = io_mem_alloc(size); | |
9050 | if (!rings) | |
2b188cc1 JA |
9051 | return -ENOMEM; |
9052 | ||
75b28aff HV |
9053 | ctx->rings = rings; |
9054 | ctx->sq_array = (u32 *)((char *)rings + sq_array_offset); | |
9055 | rings->sq_ring_mask = p->sq_entries - 1; | |
9056 | rings->cq_ring_mask = p->cq_entries - 1; | |
9057 | rings->sq_ring_entries = p->sq_entries; | |
9058 | rings->cq_ring_entries = p->cq_entries; | |
9059 | ctx->sq_mask = rings->sq_ring_mask; | |
9060 | ctx->cq_mask = rings->cq_ring_mask; | |
2b188cc1 JA |
9061 | |
9062 | size = array_size(sizeof(struct io_uring_sqe), p->sq_entries); | |
eb065d30 JA |
9063 | if (size == SIZE_MAX) { |
9064 | io_mem_free(ctx->rings); | |
9065 | ctx->rings = NULL; | |
2b188cc1 | 9066 | return -EOVERFLOW; |
eb065d30 | 9067 | } |
2b188cc1 JA |
9068 | |
9069 | ctx->sq_sqes = io_mem_alloc(size); | |
eb065d30 JA |
9070 | if (!ctx->sq_sqes) { |
9071 | io_mem_free(ctx->rings); | |
9072 | ctx->rings = NULL; | |
2b188cc1 | 9073 | return -ENOMEM; |
eb065d30 | 9074 | } |
2b188cc1 | 9075 | |
2b188cc1 JA |
9076 | return 0; |
9077 | } | |
9078 | ||
9079 | /* | |
9080 | * Allocate an anonymous fd, this is what constitutes the application | |
9081 | * visible backing of an io_uring instance. The application mmaps this | |
9082 | * fd to gain access to the SQ/CQ ring details. If UNIX sockets are enabled, | |
9083 | * we have to tie this fd to a socket for file garbage collection purposes. | |
9084 | */ | |
9085 | static int io_uring_get_fd(struct io_ring_ctx *ctx) | |
9086 | { | |
9087 | struct file *file; | |
9088 | int ret; | |
9089 | ||
9090 | #if defined(CONFIG_UNIX) | |
9091 | ret = sock_create_kern(&init_net, PF_UNIX, SOCK_RAW, IPPROTO_IP, | |
9092 | &ctx->ring_sock); | |
9093 | if (ret) | |
9094 | return ret; | |
9095 | #endif | |
9096 | ||
9097 | ret = get_unused_fd_flags(O_RDWR | O_CLOEXEC); | |
9098 | if (ret < 0) | |
9099 | goto err; | |
9100 | ||
9101 | file = anon_inode_getfile("[io_uring]", &io_uring_fops, ctx, | |
9102 | O_RDWR | O_CLOEXEC); | |
9103 | if (IS_ERR(file)) { | |
0f212204 | 9104 | err_fd: |
2b188cc1 JA |
9105 | put_unused_fd(ret); |
9106 | ret = PTR_ERR(file); | |
9107 | goto err; | |
9108 | } | |
9109 | ||
9110 | #if defined(CONFIG_UNIX) | |
9111 | ctx->ring_sock->file = file; | |
9112 | #endif | |
0f212204 JA |
9113 | if (unlikely(io_uring_add_task_file(file))) { |
9114 | file = ERR_PTR(-ENOMEM); | |
9115 | goto err_fd; | |
9116 | } | |
2b188cc1 JA |
9117 | fd_install(ret, file); |
9118 | return ret; | |
9119 | err: | |
9120 | #if defined(CONFIG_UNIX) | |
9121 | sock_release(ctx->ring_sock); | |
9122 | ctx->ring_sock = NULL; | |
9123 | #endif | |
9124 | return ret; | |
9125 | } | |
9126 | ||
7f13657d XW |
9127 | static int io_uring_create(unsigned entries, struct io_uring_params *p, |
9128 | struct io_uring_params __user *params) | |
2b188cc1 JA |
9129 | { |
9130 | struct user_struct *user = NULL; | |
9131 | struct io_ring_ctx *ctx; | |
aad5d8da | 9132 | bool limit_mem; |
2b188cc1 JA |
9133 | int ret; |
9134 | ||
8110c1a6 | 9135 | if (!entries) |
2b188cc1 | 9136 | return -EINVAL; |
8110c1a6 JA |
9137 | if (entries > IORING_MAX_ENTRIES) { |
9138 | if (!(p->flags & IORING_SETUP_CLAMP)) | |
9139 | return -EINVAL; | |
9140 | entries = IORING_MAX_ENTRIES; | |
9141 | } | |
2b188cc1 JA |
9142 | |
9143 | /* | |
9144 | * Use twice as many entries for the CQ ring. It's possible for the | |
9145 | * application to drive a higher depth than the size of the SQ ring, | |
9146 | * since the sqes are only used at submission time. This allows for | |
33a107f0 JA |
9147 | * some flexibility in overcommitting a bit. If the application has |
9148 | * set IORING_SETUP_CQSIZE, it will have passed in the desired number | |
9149 | * of CQ ring entries manually. | |
2b188cc1 JA |
9150 | */ |
9151 | p->sq_entries = roundup_pow_of_two(entries); | |
33a107f0 JA |
9152 | if (p->flags & IORING_SETUP_CQSIZE) { |
9153 | /* | |
9154 | * If IORING_SETUP_CQSIZE is set, we do the same roundup | |
9155 | * to a power-of-two, if it isn't already. We do NOT impose | |
9156 | * any cq vs sq ring sizing. | |
9157 | */ | |
8110c1a6 | 9158 | if (p->cq_entries < p->sq_entries) |
33a107f0 | 9159 | return -EINVAL; |
8110c1a6 JA |
9160 | if (p->cq_entries > IORING_MAX_CQ_ENTRIES) { |
9161 | if (!(p->flags & IORING_SETUP_CLAMP)) | |
9162 | return -EINVAL; | |
9163 | p->cq_entries = IORING_MAX_CQ_ENTRIES; | |
9164 | } | |
33a107f0 JA |
9165 | p->cq_entries = roundup_pow_of_two(p->cq_entries); |
9166 | } else { | |
9167 | p->cq_entries = 2 * p->sq_entries; | |
9168 | } | |
2b188cc1 JA |
9169 | |
9170 | user = get_uid(current_user()); | |
aad5d8da | 9171 | limit_mem = !capable(CAP_IPC_LOCK); |
2b188cc1 | 9172 | |
aad5d8da | 9173 | if (limit_mem) { |
a087e2b5 | 9174 | ret = __io_account_mem(user, |
2b188cc1 JA |
9175 | ring_pages(p->sq_entries, p->cq_entries)); |
9176 | if (ret) { | |
9177 | free_uid(user); | |
9178 | return ret; | |
9179 | } | |
9180 | } | |
9181 | ||
9182 | ctx = io_ring_ctx_alloc(p); | |
9183 | if (!ctx) { | |
aad5d8da | 9184 | if (limit_mem) |
a087e2b5 | 9185 | __io_unaccount_mem(user, ring_pages(p->sq_entries, |
2b188cc1 JA |
9186 | p->cq_entries)); |
9187 | free_uid(user); | |
9188 | return -ENOMEM; | |
9189 | } | |
9190 | ctx->compat = in_compat_syscall(); | |
2b188cc1 | 9191 | ctx->user = user; |
0b8c0ec7 | 9192 | ctx->creds = get_current_cred(); |
2b188cc1 | 9193 | |
2aede0e4 JA |
9194 | ctx->sqo_task = get_task_struct(current); |
9195 | ||
9196 | /* | |
9197 | * This is just grabbed for accounting purposes. When a process exits, | |
9198 | * the mm is exited and dropped before the files, hence we need to hang | |
9199 | * on to this mm purely for the purposes of being able to unaccount | |
9200 | * memory (locked/pinned vm). It's not used for anything else. | |
9201 | */ | |
6b7898eb | 9202 | mmgrab(current->mm); |
2aede0e4 | 9203 | ctx->mm_account = current->mm; |
6b7898eb | 9204 | |
91d8f519 DZ |
9205 | #ifdef CONFIG_BLK_CGROUP |
9206 | /* | |
9207 | * The sq thread will belong to the original cgroup it was inited in. | |
9208 | * If the cgroup goes offline (e.g. disabling the io controller), then | |
9209 | * issued bios will be associated with the closest cgroup later in the | |
9210 | * block layer. | |
9211 | */ | |
9212 | rcu_read_lock(); | |
9213 | ctx->sqo_blkcg_css = blkcg_css(); | |
9214 | ret = css_tryget_online(ctx->sqo_blkcg_css); | |
9215 | rcu_read_unlock(); | |
9216 | if (!ret) { | |
9217 | /* don't init against a dying cgroup, have the user try again */ | |
9218 | ctx->sqo_blkcg_css = NULL; | |
9219 | ret = -ENODEV; | |
9220 | goto err; | |
9221 | } | |
9222 | #endif | |
9223 | ||
f74441e6 JA |
9224 | /* |
9225 | * Account memory _before_ installing the file descriptor. Once | |
9226 | * the descriptor is installed, it can get closed at any time. Also | |
9227 | * do this before hitting the general error path, as ring freeing | |
9228 | * will un-account as well. | |
9229 | */ | |
9230 | io_account_mem(ctx, ring_pages(p->sq_entries, p->cq_entries), | |
9231 | ACCT_LOCKED); | |
9232 | ctx->limit_mem = limit_mem; | |
9233 | ||
2b188cc1 JA |
9234 | ret = io_allocate_scq_urings(ctx, p); |
9235 | if (ret) | |
9236 | goto err; | |
9237 | ||
7e84e1c7 | 9238 | ret = io_sq_offload_create(ctx, p); |
2b188cc1 JA |
9239 | if (ret) |
9240 | goto err; | |
9241 | ||
7e84e1c7 SG |
9242 | if (!(p->flags & IORING_SETUP_R_DISABLED)) |
9243 | io_sq_offload_start(ctx); | |
9244 | ||
2b188cc1 | 9245 | memset(&p->sq_off, 0, sizeof(p->sq_off)); |
75b28aff HV |
9246 | p->sq_off.head = offsetof(struct io_rings, sq.head); |
9247 | p->sq_off.tail = offsetof(struct io_rings, sq.tail); | |
9248 | p->sq_off.ring_mask = offsetof(struct io_rings, sq_ring_mask); | |
9249 | p->sq_off.ring_entries = offsetof(struct io_rings, sq_ring_entries); | |
9250 | p->sq_off.flags = offsetof(struct io_rings, sq_flags); | |
9251 | p->sq_off.dropped = offsetof(struct io_rings, sq_dropped); | |
9252 | p->sq_off.array = (char *)ctx->sq_array - (char *)ctx->rings; | |
2b188cc1 JA |
9253 | |
9254 | memset(&p->cq_off, 0, sizeof(p->cq_off)); | |
75b28aff HV |
9255 | p->cq_off.head = offsetof(struct io_rings, cq.head); |
9256 | p->cq_off.tail = offsetof(struct io_rings, cq.tail); | |
9257 | p->cq_off.ring_mask = offsetof(struct io_rings, cq_ring_mask); | |
9258 | p->cq_off.ring_entries = offsetof(struct io_rings, cq_ring_entries); | |
9259 | p->cq_off.overflow = offsetof(struct io_rings, cq_overflow); | |
9260 | p->cq_off.cqes = offsetof(struct io_rings, cqes); | |
0d9b5b3a | 9261 | p->cq_off.flags = offsetof(struct io_rings, cq_flags); |
ac90f249 | 9262 | |
7f13657d XW |
9263 | p->features = IORING_FEAT_SINGLE_MMAP | IORING_FEAT_NODROP | |
9264 | IORING_FEAT_SUBMIT_STABLE | IORING_FEAT_RW_CUR_POS | | |
5769a351 JX |
9265 | IORING_FEAT_CUR_PERSONALITY | IORING_FEAT_FAST_POLL | |
9266 | IORING_FEAT_POLL_32BITS; | |
7f13657d XW |
9267 | |
9268 | if (copy_to_user(params, p, sizeof(*p))) { | |
9269 | ret = -EFAULT; | |
9270 | goto err; | |
9271 | } | |
d1719f70 | 9272 | |
044c1ab3 JA |
9273 | /* |
9274 | * Install ring fd as the very last thing, so we don't risk someone | |
9275 | * having closed it before we finish setup | |
9276 | */ | |
9277 | ret = io_uring_get_fd(ctx); | |
9278 | if (ret < 0) | |
9279 | goto err; | |
9280 | ||
c826bd7a | 9281 | trace_io_uring_create(ret, ctx, p->sq_entries, p->cq_entries, p->flags); |
2b188cc1 JA |
9282 | return ret; |
9283 | err: | |
9284 | io_ring_ctx_wait_and_kill(ctx); | |
9285 | return ret; | |
9286 | } | |
9287 | ||
9288 | /* | |
9289 | * Sets up an aio uring context, and returns the fd. Applications asks for a | |
9290 | * ring size, we return the actual sq/cq ring sizes (among other things) in the | |
9291 | * params structure passed in. | |
9292 | */ | |
9293 | static long io_uring_setup(u32 entries, struct io_uring_params __user *params) | |
9294 | { | |
9295 | struct io_uring_params p; | |
2b188cc1 JA |
9296 | int i; |
9297 | ||
9298 | if (copy_from_user(&p, params, sizeof(p))) | |
9299 | return -EFAULT; | |
9300 | for (i = 0; i < ARRAY_SIZE(p.resv); i++) { | |
9301 | if (p.resv[i]) | |
9302 | return -EINVAL; | |
9303 | } | |
9304 | ||
6c271ce2 | 9305 | if (p.flags & ~(IORING_SETUP_IOPOLL | IORING_SETUP_SQPOLL | |
8110c1a6 | 9306 | IORING_SETUP_SQ_AFF | IORING_SETUP_CQSIZE | |
7e84e1c7 SG |
9307 | IORING_SETUP_CLAMP | IORING_SETUP_ATTACH_WQ | |
9308 | IORING_SETUP_R_DISABLED)) | |
2b188cc1 JA |
9309 | return -EINVAL; |
9310 | ||
7f13657d | 9311 | return io_uring_create(entries, &p, params); |
2b188cc1 JA |
9312 | } |
9313 | ||
9314 | SYSCALL_DEFINE2(io_uring_setup, u32, entries, | |
9315 | struct io_uring_params __user *, params) | |
9316 | { | |
9317 | return io_uring_setup(entries, params); | |
9318 | } | |
9319 | ||
66f4af93 JA |
9320 | static int io_probe(struct io_ring_ctx *ctx, void __user *arg, unsigned nr_args) |
9321 | { | |
9322 | struct io_uring_probe *p; | |
9323 | size_t size; | |
9324 | int i, ret; | |
9325 | ||
9326 | size = struct_size(p, ops, nr_args); | |
9327 | if (size == SIZE_MAX) | |
9328 | return -EOVERFLOW; | |
9329 | p = kzalloc(size, GFP_KERNEL); | |
9330 | if (!p) | |
9331 | return -ENOMEM; | |
9332 | ||
9333 | ret = -EFAULT; | |
9334 | if (copy_from_user(p, arg, size)) | |
9335 | goto out; | |
9336 | ret = -EINVAL; | |
9337 | if (memchr_inv(p, 0, size)) | |
9338 | goto out; | |
9339 | ||
9340 | p->last_op = IORING_OP_LAST - 1; | |
9341 | if (nr_args > IORING_OP_LAST) | |
9342 | nr_args = IORING_OP_LAST; | |
9343 | ||
9344 | for (i = 0; i < nr_args; i++) { | |
9345 | p->ops[i].op = i; | |
9346 | if (!io_op_defs[i].not_supported) | |
9347 | p->ops[i].flags = IO_URING_OP_SUPPORTED; | |
9348 | } | |
9349 | p->ops_len = i; | |
9350 | ||
9351 | ret = 0; | |
9352 | if (copy_to_user(arg, p, size)) | |
9353 | ret = -EFAULT; | |
9354 | out: | |
9355 | kfree(p); | |
9356 | return ret; | |
9357 | } | |
9358 | ||
071698e1 JA |
9359 | static int io_register_personality(struct io_ring_ctx *ctx) |
9360 | { | |
9361 | const struct cred *creds = get_current_cred(); | |
9362 | int id; | |
9363 | ||
9364 | id = idr_alloc_cyclic(&ctx->personality_idr, (void *) creds, 1, | |
9365 | USHRT_MAX, GFP_KERNEL); | |
9366 | if (id < 0) | |
9367 | put_cred(creds); | |
9368 | return id; | |
9369 | } | |
9370 | ||
9371 | static int io_unregister_personality(struct io_ring_ctx *ctx, unsigned id) | |
9372 | { | |
9373 | const struct cred *old_creds; | |
9374 | ||
9375 | old_creds = idr_remove(&ctx->personality_idr, id); | |
9376 | if (old_creds) { | |
9377 | put_cred(old_creds); | |
9378 | return 0; | |
9379 | } | |
9380 | ||
9381 | return -EINVAL; | |
9382 | } | |
9383 | ||
21b55dbc SG |
9384 | static int io_register_restrictions(struct io_ring_ctx *ctx, void __user *arg, |
9385 | unsigned int nr_args) | |
9386 | { | |
9387 | struct io_uring_restriction *res; | |
9388 | size_t size; | |
9389 | int i, ret; | |
9390 | ||
7e84e1c7 SG |
9391 | /* Restrictions allowed only if rings started disabled */ |
9392 | if (!(ctx->flags & IORING_SETUP_R_DISABLED)) | |
9393 | return -EBADFD; | |
9394 | ||
21b55dbc | 9395 | /* We allow only a single restrictions registration */ |
7e84e1c7 | 9396 | if (ctx->restrictions.registered) |
21b55dbc SG |
9397 | return -EBUSY; |
9398 | ||
9399 | if (!arg || nr_args > IORING_MAX_RESTRICTIONS) | |
9400 | return -EINVAL; | |
9401 | ||
9402 | size = array_size(nr_args, sizeof(*res)); | |
9403 | if (size == SIZE_MAX) | |
9404 | return -EOVERFLOW; | |
9405 | ||
9406 | res = memdup_user(arg, size); | |
9407 | if (IS_ERR(res)) | |
9408 | return PTR_ERR(res); | |
9409 | ||
9410 | ret = 0; | |
9411 | ||
9412 | for (i = 0; i < nr_args; i++) { | |
9413 | switch (res[i].opcode) { | |
9414 | case IORING_RESTRICTION_REGISTER_OP: | |
9415 | if (res[i].register_op >= IORING_REGISTER_LAST) { | |
9416 | ret = -EINVAL; | |
9417 | goto out; | |
9418 | } | |
9419 | ||
9420 | __set_bit(res[i].register_op, | |
9421 | ctx->restrictions.register_op); | |
9422 | break; | |
9423 | case IORING_RESTRICTION_SQE_OP: | |
9424 | if (res[i].sqe_op >= IORING_OP_LAST) { | |
9425 | ret = -EINVAL; | |
9426 | goto out; | |
9427 | } | |
9428 | ||
9429 | __set_bit(res[i].sqe_op, ctx->restrictions.sqe_op); | |
9430 | break; | |
9431 | case IORING_RESTRICTION_SQE_FLAGS_ALLOWED: | |
9432 | ctx->restrictions.sqe_flags_allowed = res[i].sqe_flags; | |
9433 | break; | |
9434 | case IORING_RESTRICTION_SQE_FLAGS_REQUIRED: | |
9435 | ctx->restrictions.sqe_flags_required = res[i].sqe_flags; | |
9436 | break; | |
9437 | default: | |
9438 | ret = -EINVAL; | |
9439 | goto out; | |
9440 | } | |
9441 | } | |
9442 | ||
9443 | out: | |
9444 | /* Reset all restrictions if an error happened */ | |
9445 | if (ret != 0) | |
9446 | memset(&ctx->restrictions, 0, sizeof(ctx->restrictions)); | |
9447 | else | |
7e84e1c7 | 9448 | ctx->restrictions.registered = true; |
21b55dbc SG |
9449 | |
9450 | kfree(res); | |
9451 | return ret; | |
9452 | } | |
9453 | ||
7e84e1c7 SG |
9454 | static int io_register_enable_rings(struct io_ring_ctx *ctx) |
9455 | { | |
9456 | if (!(ctx->flags & IORING_SETUP_R_DISABLED)) | |
9457 | return -EBADFD; | |
9458 | ||
9459 | if (ctx->restrictions.registered) | |
9460 | ctx->restricted = 1; | |
9461 | ||
9462 | ctx->flags &= ~IORING_SETUP_R_DISABLED; | |
9463 | ||
9464 | io_sq_offload_start(ctx); | |
9465 | ||
9466 | return 0; | |
9467 | } | |
9468 | ||
071698e1 JA |
9469 | static bool io_register_op_must_quiesce(int op) |
9470 | { | |
9471 | switch (op) { | |
9472 | case IORING_UNREGISTER_FILES: | |
9473 | case IORING_REGISTER_FILES_UPDATE: | |
9474 | case IORING_REGISTER_PROBE: | |
9475 | case IORING_REGISTER_PERSONALITY: | |
9476 | case IORING_UNREGISTER_PERSONALITY: | |
9477 | return false; | |
9478 | default: | |
9479 | return true; | |
9480 | } | |
9481 | } | |
9482 | ||
edafccee JA |
9483 | static int __io_uring_register(struct io_ring_ctx *ctx, unsigned opcode, |
9484 | void __user *arg, unsigned nr_args) | |
b19062a5 JA |
9485 | __releases(ctx->uring_lock) |
9486 | __acquires(ctx->uring_lock) | |
edafccee JA |
9487 | { |
9488 | int ret; | |
9489 | ||
35fa71a0 JA |
9490 | /* |
9491 | * We're inside the ring mutex, if the ref is already dying, then | |
9492 | * someone else killed the ctx or is already going through | |
9493 | * io_uring_register(). | |
9494 | */ | |
9495 | if (percpu_ref_is_dying(&ctx->refs)) | |
9496 | return -ENXIO; | |
9497 | ||
071698e1 | 9498 | if (io_register_op_must_quiesce(opcode)) { |
05f3fb3c | 9499 | percpu_ref_kill(&ctx->refs); |
b19062a5 | 9500 | |
05f3fb3c JA |
9501 | /* |
9502 | * Drop uring mutex before waiting for references to exit. If | |
9503 | * another thread is currently inside io_uring_enter() it might | |
9504 | * need to grab the uring_lock to make progress. If we hold it | |
9505 | * here across the drain wait, then we can deadlock. It's safe | |
9506 | * to drop the mutex here, since no new references will come in | |
9507 | * after we've killed the percpu ref. | |
9508 | */ | |
9509 | mutex_unlock(&ctx->uring_lock); | |
af9c1a44 JA |
9510 | do { |
9511 | ret = wait_for_completion_interruptible(&ctx->ref_comp); | |
9512 | if (!ret) | |
9513 | break; | |
ed6930c9 JA |
9514 | ret = io_run_task_work_sig(); |
9515 | if (ret < 0) | |
9516 | break; | |
af9c1a44 JA |
9517 | } while (1); |
9518 | ||
05f3fb3c | 9519 | mutex_lock(&ctx->uring_lock); |
af9c1a44 | 9520 | |
c150368b JA |
9521 | if (ret) { |
9522 | percpu_ref_resurrect(&ctx->refs); | |
21b55dbc SG |
9523 | goto out_quiesce; |
9524 | } | |
9525 | } | |
9526 | ||
9527 | if (ctx->restricted) { | |
9528 | if (opcode >= IORING_REGISTER_LAST) { | |
9529 | ret = -EINVAL; | |
9530 | goto out; | |
9531 | } | |
9532 | ||
9533 | if (!test_bit(opcode, ctx->restrictions.register_op)) { | |
9534 | ret = -EACCES; | |
c150368b JA |
9535 | goto out; |
9536 | } | |
05f3fb3c | 9537 | } |
edafccee JA |
9538 | |
9539 | switch (opcode) { | |
9540 | case IORING_REGISTER_BUFFERS: | |
9541 | ret = io_sqe_buffer_register(ctx, arg, nr_args); | |
9542 | break; | |
9543 | case IORING_UNREGISTER_BUFFERS: | |
9544 | ret = -EINVAL; | |
9545 | if (arg || nr_args) | |
9546 | break; | |
9547 | ret = io_sqe_buffer_unregister(ctx); | |
9548 | break; | |
6b06314c JA |
9549 | case IORING_REGISTER_FILES: |
9550 | ret = io_sqe_files_register(ctx, arg, nr_args); | |
9551 | break; | |
9552 | case IORING_UNREGISTER_FILES: | |
9553 | ret = -EINVAL; | |
9554 | if (arg || nr_args) | |
9555 | break; | |
9556 | ret = io_sqe_files_unregister(ctx); | |
9557 | break; | |
c3a31e60 JA |
9558 | case IORING_REGISTER_FILES_UPDATE: |
9559 | ret = io_sqe_files_update(ctx, arg, nr_args); | |
9560 | break; | |
9b402849 | 9561 | case IORING_REGISTER_EVENTFD: |
f2842ab5 | 9562 | case IORING_REGISTER_EVENTFD_ASYNC: |
9b402849 JA |
9563 | ret = -EINVAL; |
9564 | if (nr_args != 1) | |
9565 | break; | |
9566 | ret = io_eventfd_register(ctx, arg); | |
f2842ab5 JA |
9567 | if (ret) |
9568 | break; | |
9569 | if (opcode == IORING_REGISTER_EVENTFD_ASYNC) | |
9570 | ctx->eventfd_async = 1; | |
9571 | else | |
9572 | ctx->eventfd_async = 0; | |
9b402849 JA |
9573 | break; |
9574 | case IORING_UNREGISTER_EVENTFD: | |
9575 | ret = -EINVAL; | |
9576 | if (arg || nr_args) | |
9577 | break; | |
9578 | ret = io_eventfd_unregister(ctx); | |
9579 | break; | |
66f4af93 JA |
9580 | case IORING_REGISTER_PROBE: |
9581 | ret = -EINVAL; | |
9582 | if (!arg || nr_args > 256) | |
9583 | break; | |
9584 | ret = io_probe(ctx, arg, nr_args); | |
9585 | break; | |
071698e1 JA |
9586 | case IORING_REGISTER_PERSONALITY: |
9587 | ret = -EINVAL; | |
9588 | if (arg || nr_args) | |
9589 | break; | |
9590 | ret = io_register_personality(ctx); | |
9591 | break; | |
9592 | case IORING_UNREGISTER_PERSONALITY: | |
9593 | ret = -EINVAL; | |
9594 | if (arg) | |
9595 | break; | |
9596 | ret = io_unregister_personality(ctx, nr_args); | |
9597 | break; | |
7e84e1c7 SG |
9598 | case IORING_REGISTER_ENABLE_RINGS: |
9599 | ret = -EINVAL; | |
9600 | if (arg || nr_args) | |
9601 | break; | |
9602 | ret = io_register_enable_rings(ctx); | |
9603 | break; | |
21b55dbc SG |
9604 | case IORING_REGISTER_RESTRICTIONS: |
9605 | ret = io_register_restrictions(ctx, arg, nr_args); | |
9606 | break; | |
edafccee JA |
9607 | default: |
9608 | ret = -EINVAL; | |
9609 | break; | |
9610 | } | |
9611 | ||
21b55dbc | 9612 | out: |
071698e1 | 9613 | if (io_register_op_must_quiesce(opcode)) { |
05f3fb3c | 9614 | /* bring the ctx back to life */ |
05f3fb3c | 9615 | percpu_ref_reinit(&ctx->refs); |
21b55dbc | 9616 | out_quiesce: |
0f158b4c | 9617 | reinit_completion(&ctx->ref_comp); |
05f3fb3c | 9618 | } |
edafccee JA |
9619 | return ret; |
9620 | } | |
9621 | ||
9622 | SYSCALL_DEFINE4(io_uring_register, unsigned int, fd, unsigned int, opcode, | |
9623 | void __user *, arg, unsigned int, nr_args) | |
9624 | { | |
9625 | struct io_ring_ctx *ctx; | |
9626 | long ret = -EBADF; | |
9627 | struct fd f; | |
9628 | ||
9629 | f = fdget(fd); | |
9630 | if (!f.file) | |
9631 | return -EBADF; | |
9632 | ||
9633 | ret = -EOPNOTSUPP; | |
9634 | if (f.file->f_op != &io_uring_fops) | |
9635 | goto out_fput; | |
9636 | ||
9637 | ctx = f.file->private_data; | |
9638 | ||
9639 | mutex_lock(&ctx->uring_lock); | |
9640 | ret = __io_uring_register(ctx, opcode, arg, nr_args); | |
9641 | mutex_unlock(&ctx->uring_lock); | |
c826bd7a DD |
9642 | trace_io_uring_register(ctx, opcode, ctx->nr_user_files, ctx->nr_user_bufs, |
9643 | ctx->cq_ev_fd != NULL, ret); | |
edafccee JA |
9644 | out_fput: |
9645 | fdput(f); | |
9646 | return ret; | |
9647 | } | |
9648 | ||
2b188cc1 JA |
9649 | static int __init io_uring_init(void) |
9650 | { | |
d7f62e82 SM |
9651 | #define __BUILD_BUG_VERIFY_ELEMENT(stype, eoffset, etype, ename) do { \ |
9652 | BUILD_BUG_ON(offsetof(stype, ename) != eoffset); \ | |
9653 | BUILD_BUG_ON(sizeof(etype) != sizeof_field(stype, ename)); \ | |
9654 | } while (0) | |
9655 | ||
9656 | #define BUILD_BUG_SQE_ELEM(eoffset, etype, ename) \ | |
9657 | __BUILD_BUG_VERIFY_ELEMENT(struct io_uring_sqe, eoffset, etype, ename) | |
9658 | BUILD_BUG_ON(sizeof(struct io_uring_sqe) != 64); | |
9659 | BUILD_BUG_SQE_ELEM(0, __u8, opcode); | |
9660 | BUILD_BUG_SQE_ELEM(1, __u8, flags); | |
9661 | BUILD_BUG_SQE_ELEM(2, __u16, ioprio); | |
9662 | BUILD_BUG_SQE_ELEM(4, __s32, fd); | |
9663 | BUILD_BUG_SQE_ELEM(8, __u64, off); | |
9664 | BUILD_BUG_SQE_ELEM(8, __u64, addr2); | |
9665 | BUILD_BUG_SQE_ELEM(16, __u64, addr); | |
7d67af2c | 9666 | BUILD_BUG_SQE_ELEM(16, __u64, splice_off_in); |
d7f62e82 SM |
9667 | BUILD_BUG_SQE_ELEM(24, __u32, len); |
9668 | BUILD_BUG_SQE_ELEM(28, __kernel_rwf_t, rw_flags); | |
9669 | BUILD_BUG_SQE_ELEM(28, /* compat */ int, rw_flags); | |
9670 | BUILD_BUG_SQE_ELEM(28, /* compat */ __u32, rw_flags); | |
9671 | BUILD_BUG_SQE_ELEM(28, __u32, fsync_flags); | |
5769a351 JX |
9672 | BUILD_BUG_SQE_ELEM(28, /* compat */ __u16, poll_events); |
9673 | BUILD_BUG_SQE_ELEM(28, __u32, poll32_events); | |
d7f62e82 SM |
9674 | BUILD_BUG_SQE_ELEM(28, __u32, sync_range_flags); |
9675 | BUILD_BUG_SQE_ELEM(28, __u32, msg_flags); | |
9676 | BUILD_BUG_SQE_ELEM(28, __u32, timeout_flags); | |
9677 | BUILD_BUG_SQE_ELEM(28, __u32, accept_flags); | |
9678 | BUILD_BUG_SQE_ELEM(28, __u32, cancel_flags); | |
9679 | BUILD_BUG_SQE_ELEM(28, __u32, open_flags); | |
9680 | BUILD_BUG_SQE_ELEM(28, __u32, statx_flags); | |
9681 | BUILD_BUG_SQE_ELEM(28, __u32, fadvise_advice); | |
7d67af2c | 9682 | BUILD_BUG_SQE_ELEM(28, __u32, splice_flags); |
d7f62e82 SM |
9683 | BUILD_BUG_SQE_ELEM(32, __u64, user_data); |
9684 | BUILD_BUG_SQE_ELEM(40, __u16, buf_index); | |
9685 | BUILD_BUG_SQE_ELEM(42, __u16, personality); | |
7d67af2c | 9686 | BUILD_BUG_SQE_ELEM(44, __s32, splice_fd_in); |
d7f62e82 | 9687 | |
d3656344 | 9688 | BUILD_BUG_ON(ARRAY_SIZE(io_op_defs) != IORING_OP_LAST); |
84557871 | 9689 | BUILD_BUG_ON(__REQ_F_LAST_BIT >= 8 * sizeof(int)); |
2b188cc1 JA |
9690 | req_cachep = KMEM_CACHE(io_kiocb, SLAB_HWCACHE_ALIGN | SLAB_PANIC); |
9691 | return 0; | |
9692 | }; | |
9693 | __initcall(io_uring_init); |