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> |
2b188cc1 | 82 | |
c826bd7a DD |
83 | #define CREATE_TRACE_POINTS |
84 | #include <trace/events/io_uring.h> | |
85 | ||
2b188cc1 JA |
86 | #include <uapi/linux/io_uring.h> |
87 | ||
88 | #include "internal.h" | |
561fb04a | 89 | #include "io-wq.h" |
2b188cc1 | 90 | |
5277deaa | 91 | #define IORING_MAX_ENTRIES 32768 |
33a107f0 | 92 | #define IORING_MAX_CQ_ENTRIES (2 * IORING_MAX_ENTRIES) |
65e19f54 JA |
93 | |
94 | /* | |
95 | * Shift of 9 is 512 entries, or exactly one page on 64-bit archs | |
96 | */ | |
97 | #define IORING_FILE_TABLE_SHIFT 9 | |
98 | #define IORING_MAX_FILES_TABLE (1U << IORING_FILE_TABLE_SHIFT) | |
99 | #define IORING_FILE_TABLE_MASK (IORING_MAX_FILES_TABLE - 1) | |
100 | #define IORING_MAX_FIXED_FILES (64 * IORING_MAX_FILES_TABLE) | |
2b188cc1 JA |
101 | |
102 | struct io_uring { | |
103 | u32 head ____cacheline_aligned_in_smp; | |
104 | u32 tail ____cacheline_aligned_in_smp; | |
105 | }; | |
106 | ||
1e84b97b | 107 | /* |
75b28aff HV |
108 | * This data is shared with the application through the mmap at offsets |
109 | * IORING_OFF_SQ_RING and IORING_OFF_CQ_RING. | |
1e84b97b SB |
110 | * |
111 | * The offsets to the member fields are published through struct | |
112 | * io_sqring_offsets when calling io_uring_setup. | |
113 | */ | |
75b28aff | 114 | struct io_rings { |
1e84b97b SB |
115 | /* |
116 | * Head and tail offsets into the ring; the offsets need to be | |
117 | * masked to get valid indices. | |
118 | * | |
75b28aff HV |
119 | * The kernel controls head of the sq ring and the tail of the cq ring, |
120 | * and the application controls tail of the sq ring and the head of the | |
121 | * cq ring. | |
1e84b97b | 122 | */ |
75b28aff | 123 | struct io_uring sq, cq; |
1e84b97b | 124 | /* |
75b28aff | 125 | * Bitmasks to apply to head and tail offsets (constant, equals |
1e84b97b SB |
126 | * ring_entries - 1) |
127 | */ | |
75b28aff HV |
128 | u32 sq_ring_mask, cq_ring_mask; |
129 | /* Ring sizes (constant, power of 2) */ | |
130 | u32 sq_ring_entries, cq_ring_entries; | |
1e84b97b SB |
131 | /* |
132 | * Number of invalid entries dropped by the kernel due to | |
133 | * invalid index stored in array | |
134 | * | |
135 | * Written by the kernel, shouldn't be modified by the | |
136 | * application (i.e. get number of "new events" by comparing to | |
137 | * cached value). | |
138 | * | |
139 | * After a new SQ head value was read by the application this | |
140 | * counter includes all submissions that were dropped reaching | |
141 | * the new SQ head (and possibly more). | |
142 | */ | |
75b28aff | 143 | u32 sq_dropped; |
1e84b97b | 144 | /* |
0d9b5b3a | 145 | * Runtime SQ flags |
1e84b97b SB |
146 | * |
147 | * Written by the kernel, shouldn't be modified by the | |
148 | * application. | |
149 | * | |
150 | * The application needs a full memory barrier before checking | |
151 | * for IORING_SQ_NEED_WAKEUP after updating the sq tail. | |
152 | */ | |
75b28aff | 153 | u32 sq_flags; |
0d9b5b3a SG |
154 | /* |
155 | * Runtime CQ flags | |
156 | * | |
157 | * Written by the application, shouldn't be modified by the | |
158 | * kernel. | |
159 | */ | |
160 | u32 cq_flags; | |
1e84b97b SB |
161 | /* |
162 | * Number of completion events lost because the queue was full; | |
163 | * this should be avoided by the application by making sure | |
0b4295b5 | 164 | * there are not more requests pending than there is space in |
1e84b97b SB |
165 | * the completion queue. |
166 | * | |
167 | * Written by the kernel, shouldn't be modified by the | |
168 | * application (i.e. get number of "new events" by comparing to | |
169 | * cached value). | |
170 | * | |
171 | * As completion events come in out of order this counter is not | |
172 | * ordered with any other data. | |
173 | */ | |
75b28aff | 174 | u32 cq_overflow; |
1e84b97b SB |
175 | /* |
176 | * Ring buffer of completion events. | |
177 | * | |
178 | * The kernel writes completion events fresh every time they are | |
179 | * produced, so the application is allowed to modify pending | |
180 | * entries. | |
181 | */ | |
75b28aff | 182 | struct io_uring_cqe cqes[] ____cacheline_aligned_in_smp; |
2b188cc1 JA |
183 | }; |
184 | ||
edafccee JA |
185 | struct io_mapped_ubuf { |
186 | u64 ubuf; | |
187 | size_t len; | |
188 | struct bio_vec *bvec; | |
189 | unsigned int nr_bvecs; | |
190 | }; | |
191 | ||
65e19f54 JA |
192 | struct fixed_file_table { |
193 | struct file **files; | |
31b51510 JA |
194 | }; |
195 | ||
05589553 XW |
196 | struct fixed_file_ref_node { |
197 | struct percpu_ref refs; | |
198 | struct list_head node; | |
199 | struct list_head file_list; | |
200 | struct fixed_file_data *file_data; | |
4a38aed2 | 201 | struct llist_node llist; |
05589553 XW |
202 | }; |
203 | ||
05f3fb3c JA |
204 | struct fixed_file_data { |
205 | struct fixed_file_table *table; | |
206 | struct io_ring_ctx *ctx; | |
207 | ||
05589553 | 208 | struct percpu_ref *cur_refs; |
05f3fb3c | 209 | struct percpu_ref refs; |
05f3fb3c | 210 | struct completion done; |
05589553 XW |
211 | struct list_head ref_list; |
212 | spinlock_t lock; | |
05f3fb3c JA |
213 | }; |
214 | ||
5a2e745d JA |
215 | struct io_buffer { |
216 | struct list_head list; | |
217 | __u64 addr; | |
218 | __s32 len; | |
219 | __u16 bid; | |
220 | }; | |
221 | ||
2b188cc1 JA |
222 | struct io_ring_ctx { |
223 | struct { | |
224 | struct percpu_ref refs; | |
225 | } ____cacheline_aligned_in_smp; | |
226 | ||
227 | struct { | |
228 | unsigned int flags; | |
e1d85334 | 229 | unsigned int compat: 1; |
aad5d8da | 230 | unsigned int limit_mem: 1; |
e1d85334 RD |
231 | unsigned int cq_overflow_flushed: 1; |
232 | unsigned int drain_next: 1; | |
233 | unsigned int eventfd_async: 1; | |
2b188cc1 | 234 | |
75b28aff HV |
235 | /* |
236 | * Ring buffer of indices into array of io_uring_sqe, which is | |
237 | * mmapped by the application using the IORING_OFF_SQES offset. | |
238 | * | |
239 | * This indirection could e.g. be used to assign fixed | |
240 | * io_uring_sqe entries to operations and only submit them to | |
241 | * the queue when needed. | |
242 | * | |
243 | * The kernel modifies neither the indices array nor the entries | |
244 | * array. | |
245 | */ | |
246 | u32 *sq_array; | |
2b188cc1 JA |
247 | unsigned cached_sq_head; |
248 | unsigned sq_entries; | |
249 | unsigned sq_mask; | |
6c271ce2 | 250 | unsigned sq_thread_idle; |
498ccd9e | 251 | unsigned cached_sq_dropped; |
206aefde | 252 | atomic_t cached_cq_overflow; |
ad3eb2c8 | 253 | unsigned long sq_check_overflow; |
de0617e4 JA |
254 | |
255 | struct list_head defer_list; | |
5262f567 | 256 | struct list_head timeout_list; |
1d7bb1d5 | 257 | struct list_head cq_overflow_list; |
fcb323cc JA |
258 | |
259 | wait_queue_head_t inflight_wait; | |
ad3eb2c8 | 260 | struct io_uring_sqe *sq_sqes; |
2b188cc1 JA |
261 | } ____cacheline_aligned_in_smp; |
262 | ||
206aefde JA |
263 | struct io_rings *rings; |
264 | ||
2b188cc1 | 265 | /* IO offload */ |
561fb04a | 266 | struct io_wq *io_wq; |
6c271ce2 | 267 | struct task_struct *sqo_thread; /* if using sq thread polling */ |
2b188cc1 | 268 | struct mm_struct *sqo_mm; |
6c271ce2 | 269 | wait_queue_head_t sqo_wait; |
75b28aff | 270 | |
6b06314c JA |
271 | /* |
272 | * If used, fixed file set. Writers must ensure that ->refs is dead, | |
273 | * readers must ensure that ->refs is alive as long as the file* is | |
274 | * used. Only updated through io_uring_register(2). | |
275 | */ | |
05f3fb3c | 276 | struct fixed_file_data *file_data; |
6b06314c | 277 | unsigned nr_user_files; |
b14cca0c PB |
278 | int ring_fd; |
279 | struct file *ring_file; | |
6b06314c | 280 | |
edafccee JA |
281 | /* if used, fixed mapped user buffers */ |
282 | unsigned nr_user_bufs; | |
283 | struct io_mapped_ubuf *user_bufs; | |
284 | ||
2b188cc1 JA |
285 | struct user_struct *user; |
286 | ||
0b8c0ec7 | 287 | const struct cred *creds; |
181e448d | 288 | |
0f158b4c JA |
289 | struct completion ref_comp; |
290 | struct completion sq_thread_comp; | |
206aefde | 291 | |
0ddf92e8 JA |
292 | /* if all else fails... */ |
293 | struct io_kiocb *fallback_req; | |
294 | ||
206aefde JA |
295 | #if defined(CONFIG_UNIX) |
296 | struct socket *ring_sock; | |
297 | #endif | |
298 | ||
5a2e745d JA |
299 | struct idr io_buffer_idr; |
300 | ||
071698e1 JA |
301 | struct idr personality_idr; |
302 | ||
206aefde JA |
303 | struct { |
304 | unsigned cached_cq_tail; | |
305 | unsigned cq_entries; | |
306 | unsigned cq_mask; | |
307 | atomic_t cq_timeouts; | |
ad3eb2c8 | 308 | unsigned long cq_check_overflow; |
206aefde JA |
309 | struct wait_queue_head cq_wait; |
310 | struct fasync_struct *cq_fasync; | |
311 | struct eventfd_ctx *cq_ev_fd; | |
312 | } ____cacheline_aligned_in_smp; | |
2b188cc1 JA |
313 | |
314 | struct { | |
315 | struct mutex uring_lock; | |
316 | wait_queue_head_t wait; | |
317 | } ____cacheline_aligned_in_smp; | |
318 | ||
319 | struct { | |
320 | spinlock_t completion_lock; | |
e94f141b | 321 | |
def596e9 | 322 | /* |
540e32a0 | 323 | * ->iopoll_list is protected by the ctx->uring_lock for |
def596e9 JA |
324 | * io_uring instances that don't use IORING_SETUP_SQPOLL. |
325 | * For SQPOLL, only the single threaded io_sq_thread() will | |
326 | * manipulate the list, hence no extra locking is needed there. | |
327 | */ | |
540e32a0 | 328 | struct list_head iopoll_list; |
78076bb6 JA |
329 | struct hlist_head *cancel_hash; |
330 | unsigned cancel_hash_bits; | |
e94f141b | 331 | bool poll_multi_file; |
31b51510 | 332 | |
fcb323cc JA |
333 | spinlock_t inflight_lock; |
334 | struct list_head inflight_list; | |
2b188cc1 | 335 | } ____cacheline_aligned_in_smp; |
85faa7b8 | 336 | |
4a38aed2 JA |
337 | struct delayed_work file_put_work; |
338 | struct llist_head file_put_llist; | |
339 | ||
85faa7b8 | 340 | struct work_struct exit_work; |
2b188cc1 JA |
341 | }; |
342 | ||
09bb8394 JA |
343 | /* |
344 | * First field must be the file pointer in all the | |
345 | * iocb unions! See also 'struct kiocb' in <linux/fs.h> | |
346 | */ | |
221c5eb2 JA |
347 | struct io_poll_iocb { |
348 | struct file *file; | |
0969e783 JA |
349 | union { |
350 | struct wait_queue_head *head; | |
351 | u64 addr; | |
352 | }; | |
221c5eb2 | 353 | __poll_t events; |
8c838788 | 354 | bool done; |
221c5eb2 | 355 | bool canceled; |
392edb45 | 356 | struct wait_queue_entry wait; |
221c5eb2 JA |
357 | }; |
358 | ||
b5dba59e JA |
359 | struct io_close { |
360 | struct file *file; | |
361 | struct file *put_file; | |
362 | int fd; | |
363 | }; | |
364 | ||
ad8a48ac JA |
365 | struct io_timeout_data { |
366 | struct io_kiocb *req; | |
367 | struct hrtimer timer; | |
368 | struct timespec64 ts; | |
369 | enum hrtimer_mode mode; | |
370 | }; | |
371 | ||
8ed8d3c3 JA |
372 | struct io_accept { |
373 | struct file *file; | |
374 | struct sockaddr __user *addr; | |
375 | int __user *addr_len; | |
376 | int flags; | |
09952e3e | 377 | unsigned long nofile; |
8ed8d3c3 JA |
378 | }; |
379 | ||
380 | struct io_sync { | |
381 | struct file *file; | |
382 | loff_t len; | |
383 | loff_t off; | |
384 | int flags; | |
d63d1b5e | 385 | int mode; |
8ed8d3c3 JA |
386 | }; |
387 | ||
fbf23849 JA |
388 | struct io_cancel { |
389 | struct file *file; | |
390 | u64 addr; | |
391 | }; | |
392 | ||
b29472ee JA |
393 | struct io_timeout { |
394 | struct file *file; | |
395 | u64 addr; | |
396 | int flags; | |
bfe68a22 PB |
397 | u32 off; |
398 | u32 target_seq; | |
135fcde8 | 399 | struct list_head list; |
b29472ee JA |
400 | }; |
401 | ||
9adbd45d JA |
402 | struct io_rw { |
403 | /* NOTE: kiocb has the file as the first member, so don't do it here */ | |
404 | struct kiocb kiocb; | |
405 | u64 addr; | |
406 | u64 len; | |
407 | }; | |
408 | ||
3fbb51c1 JA |
409 | struct io_connect { |
410 | struct file *file; | |
411 | struct sockaddr __user *addr; | |
412 | int addr_len; | |
413 | }; | |
414 | ||
e47293fd JA |
415 | struct io_sr_msg { |
416 | struct file *file; | |
fddaface | 417 | union { |
270a5940 | 418 | struct user_msghdr __user *umsg; |
fddaface JA |
419 | void __user *buf; |
420 | }; | |
e47293fd | 421 | int msg_flags; |
bcda7baa | 422 | int bgid; |
fddaface | 423 | size_t len; |
bcda7baa | 424 | struct io_buffer *kbuf; |
e47293fd JA |
425 | }; |
426 | ||
15b71abe JA |
427 | struct io_open { |
428 | struct file *file; | |
429 | int dfd; | |
15b71abe | 430 | struct filename *filename; |
c12cedf2 | 431 | struct open_how how; |
4022e7af | 432 | unsigned long nofile; |
15b71abe JA |
433 | }; |
434 | ||
05f3fb3c JA |
435 | struct io_files_update { |
436 | struct file *file; | |
437 | u64 arg; | |
438 | u32 nr_args; | |
439 | u32 offset; | |
440 | }; | |
441 | ||
4840e418 JA |
442 | struct io_fadvise { |
443 | struct file *file; | |
444 | u64 offset; | |
445 | u32 len; | |
446 | u32 advice; | |
447 | }; | |
448 | ||
c1ca757b JA |
449 | struct io_madvise { |
450 | struct file *file; | |
451 | u64 addr; | |
452 | u32 len; | |
453 | u32 advice; | |
454 | }; | |
455 | ||
3e4827b0 JA |
456 | struct io_epoll { |
457 | struct file *file; | |
458 | int epfd; | |
459 | int op; | |
460 | int fd; | |
461 | struct epoll_event event; | |
e47293fd JA |
462 | }; |
463 | ||
7d67af2c PB |
464 | struct io_splice { |
465 | struct file *file_out; | |
466 | struct file *file_in; | |
467 | loff_t off_out; | |
468 | loff_t off_in; | |
469 | u64 len; | |
470 | unsigned int flags; | |
471 | }; | |
472 | ||
ddf0322d JA |
473 | struct io_provide_buf { |
474 | struct file *file; | |
475 | __u64 addr; | |
476 | __s32 len; | |
477 | __u32 bgid; | |
478 | __u16 nbufs; | |
479 | __u16 bid; | |
480 | }; | |
481 | ||
1d9e1288 BM |
482 | struct io_statx { |
483 | struct file *file; | |
484 | int dfd; | |
485 | unsigned int mask; | |
486 | unsigned int flags; | |
e62753e4 | 487 | const char __user *filename; |
1d9e1288 BM |
488 | struct statx __user *buffer; |
489 | }; | |
490 | ||
3ca405eb PB |
491 | struct io_completion { |
492 | struct file *file; | |
493 | struct list_head list; | |
0f7e466b | 494 | int cflags; |
3ca405eb PB |
495 | }; |
496 | ||
f499a021 JA |
497 | struct io_async_connect { |
498 | struct sockaddr_storage address; | |
499 | }; | |
500 | ||
03b1230c JA |
501 | struct io_async_msghdr { |
502 | struct iovec fast_iov[UIO_FASTIOV]; | |
503 | struct iovec *iov; | |
504 | struct sockaddr __user *uaddr; | |
505 | struct msghdr msg; | |
b537916c | 506 | struct sockaddr_storage addr; |
03b1230c JA |
507 | }; |
508 | ||
f67676d1 JA |
509 | struct io_async_rw { |
510 | struct iovec fast_iov[UIO_FASTIOV]; | |
511 | struct iovec *iov; | |
512 | ssize_t nr_segs; | |
513 | ssize_t size; | |
bcf5a063 | 514 | struct wait_page_queue wpq; |
f67676d1 JA |
515 | }; |
516 | ||
1a6b74fc | 517 | struct io_async_ctx { |
f67676d1 JA |
518 | union { |
519 | struct io_async_rw rw; | |
03b1230c | 520 | struct io_async_msghdr msg; |
f499a021 | 521 | struct io_async_connect connect; |
2d28390a | 522 | struct io_timeout_data timeout; |
f67676d1 | 523 | }; |
1a6b74fc JA |
524 | }; |
525 | ||
6b47ee6e PB |
526 | enum { |
527 | REQ_F_FIXED_FILE_BIT = IOSQE_FIXED_FILE_BIT, | |
528 | REQ_F_IO_DRAIN_BIT = IOSQE_IO_DRAIN_BIT, | |
529 | REQ_F_LINK_BIT = IOSQE_IO_LINK_BIT, | |
530 | REQ_F_HARDLINK_BIT = IOSQE_IO_HARDLINK_BIT, | |
531 | REQ_F_FORCE_ASYNC_BIT = IOSQE_ASYNC_BIT, | |
bcda7baa | 532 | REQ_F_BUFFER_SELECT_BIT = IOSQE_BUFFER_SELECT_BIT, |
6b47ee6e | 533 | |
dea3b49c | 534 | REQ_F_LINK_HEAD_BIT, |
6b47ee6e PB |
535 | REQ_F_FAIL_LINK_BIT, |
536 | REQ_F_INFLIGHT_BIT, | |
537 | REQ_F_CUR_POS_BIT, | |
538 | REQ_F_NOWAIT_BIT, | |
6b47ee6e | 539 | REQ_F_LINK_TIMEOUT_BIT, |
6b47ee6e | 540 | REQ_F_ISREG_BIT, |
6b47ee6e | 541 | REQ_F_COMP_LOCKED_BIT, |
99bc4c38 | 542 | REQ_F_NEED_CLEANUP_BIT, |
2ca10259 | 543 | REQ_F_OVERFLOW_BIT, |
d7718a9d | 544 | REQ_F_POLLED_BIT, |
bcda7baa | 545 | REQ_F_BUFFER_SELECTED_BIT, |
5b0bbee4 | 546 | REQ_F_NO_FILE_TABLE_BIT, |
7cdaf587 | 547 | REQ_F_WORK_INITIALIZED_BIT, |
4dd2824d | 548 | REQ_F_TASK_PINNED_BIT, |
84557871 JA |
549 | |
550 | /* not a real bit, just to check we're not overflowing the space */ | |
551 | __REQ_F_LAST_BIT, | |
6b47ee6e PB |
552 | }; |
553 | ||
554 | enum { | |
555 | /* ctx owns file */ | |
556 | REQ_F_FIXED_FILE = BIT(REQ_F_FIXED_FILE_BIT), | |
557 | /* drain existing IO first */ | |
558 | REQ_F_IO_DRAIN = BIT(REQ_F_IO_DRAIN_BIT), | |
559 | /* linked sqes */ | |
560 | REQ_F_LINK = BIT(REQ_F_LINK_BIT), | |
561 | /* doesn't sever on completion < 0 */ | |
562 | REQ_F_HARDLINK = BIT(REQ_F_HARDLINK_BIT), | |
563 | /* IOSQE_ASYNC */ | |
564 | REQ_F_FORCE_ASYNC = BIT(REQ_F_FORCE_ASYNC_BIT), | |
bcda7baa JA |
565 | /* IOSQE_BUFFER_SELECT */ |
566 | REQ_F_BUFFER_SELECT = BIT(REQ_F_BUFFER_SELECT_BIT), | |
6b47ee6e | 567 | |
dea3b49c PB |
568 | /* head of a link */ |
569 | REQ_F_LINK_HEAD = BIT(REQ_F_LINK_HEAD_BIT), | |
6b47ee6e PB |
570 | /* fail rest of links */ |
571 | REQ_F_FAIL_LINK = BIT(REQ_F_FAIL_LINK_BIT), | |
572 | /* on inflight list */ | |
573 | REQ_F_INFLIGHT = BIT(REQ_F_INFLIGHT_BIT), | |
574 | /* read/write uses file position */ | |
575 | REQ_F_CUR_POS = BIT(REQ_F_CUR_POS_BIT), | |
576 | /* must not punt to workers */ | |
577 | REQ_F_NOWAIT = BIT(REQ_F_NOWAIT_BIT), | |
6b47ee6e PB |
578 | /* has linked timeout */ |
579 | REQ_F_LINK_TIMEOUT = BIT(REQ_F_LINK_TIMEOUT_BIT), | |
6b47ee6e PB |
580 | /* regular file */ |
581 | REQ_F_ISREG = BIT(REQ_F_ISREG_BIT), | |
6b47ee6e PB |
582 | /* completion under lock */ |
583 | REQ_F_COMP_LOCKED = BIT(REQ_F_COMP_LOCKED_BIT), | |
99bc4c38 PB |
584 | /* needs cleanup */ |
585 | REQ_F_NEED_CLEANUP = BIT(REQ_F_NEED_CLEANUP_BIT), | |
2ca10259 JA |
586 | /* in overflow list */ |
587 | REQ_F_OVERFLOW = BIT(REQ_F_OVERFLOW_BIT), | |
d7718a9d JA |
588 | /* already went through poll handler */ |
589 | REQ_F_POLLED = BIT(REQ_F_POLLED_BIT), | |
bcda7baa JA |
590 | /* buffer already selected */ |
591 | REQ_F_BUFFER_SELECTED = BIT(REQ_F_BUFFER_SELECTED_BIT), | |
5b0bbee4 JA |
592 | /* doesn't need file table for this request */ |
593 | REQ_F_NO_FILE_TABLE = BIT(REQ_F_NO_FILE_TABLE_BIT), | |
7cdaf587 XW |
594 | /* io_wq_work is initialized */ |
595 | REQ_F_WORK_INITIALIZED = BIT(REQ_F_WORK_INITIALIZED_BIT), | |
4dd2824d PB |
596 | /* req->task is refcounted */ |
597 | REQ_F_TASK_PINNED = BIT(REQ_F_TASK_PINNED_BIT), | |
d7718a9d JA |
598 | }; |
599 | ||
600 | struct async_poll { | |
601 | struct io_poll_iocb poll; | |
807abcb0 | 602 | struct io_poll_iocb *double_poll; |
d7718a9d | 603 | struct io_wq_work work; |
6b47ee6e PB |
604 | }; |
605 | ||
09bb8394 JA |
606 | /* |
607 | * NOTE! Each of the iocb union members has the file pointer | |
608 | * as the first entry in their struct definition. So you can | |
609 | * access the file pointer through any of the sub-structs, | |
610 | * or directly as just 'ki_filp' in this struct. | |
611 | */ | |
2b188cc1 | 612 | struct io_kiocb { |
221c5eb2 | 613 | union { |
09bb8394 | 614 | struct file *file; |
9adbd45d | 615 | struct io_rw rw; |
221c5eb2 | 616 | struct io_poll_iocb poll; |
8ed8d3c3 JA |
617 | struct io_accept accept; |
618 | struct io_sync sync; | |
fbf23849 | 619 | struct io_cancel cancel; |
b29472ee | 620 | struct io_timeout timeout; |
3fbb51c1 | 621 | struct io_connect connect; |
e47293fd | 622 | struct io_sr_msg sr_msg; |
15b71abe | 623 | struct io_open open; |
b5dba59e | 624 | struct io_close close; |
05f3fb3c | 625 | struct io_files_update files_update; |
4840e418 | 626 | struct io_fadvise fadvise; |
c1ca757b | 627 | struct io_madvise madvise; |
3e4827b0 | 628 | struct io_epoll epoll; |
7d67af2c | 629 | struct io_splice splice; |
ddf0322d | 630 | struct io_provide_buf pbuf; |
1d9e1288 | 631 | struct io_statx statx; |
3ca405eb PB |
632 | /* use only after cleaning per-op data, see io_clean_op() */ |
633 | struct io_completion compl; | |
221c5eb2 | 634 | }; |
2b188cc1 | 635 | |
1a6b74fc | 636 | struct io_async_ctx *io; |
d625c6ee | 637 | u8 opcode; |
65a6543d XW |
638 | /* polled IO has completed */ |
639 | u8 iopoll_completed; | |
2b188cc1 | 640 | |
4f4eeba8 | 641 | u16 buf_index; |
9cf7c104 | 642 | u32 result; |
4f4eeba8 | 643 | |
2b188cc1 | 644 | struct io_ring_ctx *ctx; |
2b188cc1 | 645 | unsigned int flags; |
c16361c1 | 646 | refcount_t refs; |
3537b6a7 JA |
647 | struct task_struct *task; |
648 | unsigned long fsize; | |
2b188cc1 JA |
649 | u64 user_data; |
650 | ||
d7718a9d JA |
651 | struct list_head link_list; |
652 | ||
d21ffe7e PB |
653 | /* |
654 | * 1. used with ctx->iopoll_list with reads/writes | |
655 | * 2. to track reqs with ->files (see io_op_def::file_table) | |
656 | */ | |
fcb323cc JA |
657 | struct list_head inflight_entry; |
658 | ||
05589553 XW |
659 | struct percpu_ref *fixed_file_refs; |
660 | ||
b41e9852 JA |
661 | union { |
662 | /* | |
663 | * Only commands that never go async can use the below fields, | |
d7718a9d JA |
664 | * obviously. Right now only IORING_OP_POLL_ADD uses them, and |
665 | * async armed poll handlers for regular commands. The latter | |
666 | * restore the work, if needed. | |
b41e9852 JA |
667 | */ |
668 | struct { | |
d7718a9d JA |
669 | struct hlist_node hash_node; |
670 | struct async_poll *apoll; | |
b41e9852 JA |
671 | }; |
672 | struct io_wq_work work; | |
673 | }; | |
8ef77766 | 674 | struct callback_head task_work; |
2b188cc1 JA |
675 | }; |
676 | ||
27dc8338 PB |
677 | struct io_defer_entry { |
678 | struct list_head list; | |
679 | struct io_kiocb *req; | |
9cf7c104 | 680 | u32 seq; |
27dc8338 PB |
681 | }; |
682 | ||
def596e9 | 683 | #define IO_IOPOLL_BATCH 8 |
2b188cc1 | 684 | |
013538bd JA |
685 | struct io_comp_state { |
686 | unsigned int nr; | |
687 | struct list_head list; | |
688 | struct io_ring_ctx *ctx; | |
689 | }; | |
690 | ||
9a56a232 JA |
691 | struct io_submit_state { |
692 | struct blk_plug plug; | |
693 | ||
2579f913 JA |
694 | /* |
695 | * io_kiocb alloc cache | |
696 | */ | |
697 | void *reqs[IO_IOPOLL_BATCH]; | |
6c8a3134 | 698 | unsigned int free_reqs; |
2579f913 | 699 | |
013538bd JA |
700 | /* |
701 | * Batch completion logic | |
702 | */ | |
703 | struct io_comp_state comp; | |
704 | ||
9a56a232 JA |
705 | /* |
706 | * File reference cache | |
707 | */ | |
708 | struct file *file; | |
709 | unsigned int fd; | |
710 | unsigned int has_refs; | |
711 | unsigned int used_refs; | |
712 | unsigned int ios_left; | |
713 | }; | |
714 | ||
d3656344 JA |
715 | struct io_op_def { |
716 | /* needs req->io allocated for deferral/async */ | |
717 | unsigned async_ctx : 1; | |
718 | /* needs current->mm setup, does mm access */ | |
719 | unsigned needs_mm : 1; | |
720 | /* needs req->file assigned */ | |
721 | unsigned needs_file : 1; | |
fd2206e4 JA |
722 | /* don't fail if file grab fails */ |
723 | unsigned needs_file_no_error : 1; | |
d3656344 JA |
724 | /* hash wq insertion if file is a regular file */ |
725 | unsigned hash_reg_file : 1; | |
726 | /* unbound wq insertion if file is a non-regular file */ | |
727 | unsigned unbound_nonreg_file : 1; | |
66f4af93 JA |
728 | /* opcode is not supported by this kernel */ |
729 | unsigned not_supported : 1; | |
f86cd20c JA |
730 | /* needs file table */ |
731 | unsigned file_table : 1; | |
ff002b30 JA |
732 | /* needs ->fs */ |
733 | unsigned needs_fs : 1; | |
8a72758c JA |
734 | /* set if opcode supports polled "wait" */ |
735 | unsigned pollin : 1; | |
736 | unsigned pollout : 1; | |
bcda7baa JA |
737 | /* op supports buffer selection */ |
738 | unsigned buffer_select : 1; | |
d3656344 JA |
739 | }; |
740 | ||
741 | static const struct io_op_def io_op_defs[] = { | |
0463b6c5 PB |
742 | [IORING_OP_NOP] = {}, |
743 | [IORING_OP_READV] = { | |
d3656344 JA |
744 | .async_ctx = 1, |
745 | .needs_mm = 1, | |
746 | .needs_file = 1, | |
747 | .unbound_nonreg_file = 1, | |
8a72758c | 748 | .pollin = 1, |
4d954c25 | 749 | .buffer_select = 1, |
d3656344 | 750 | }, |
0463b6c5 | 751 | [IORING_OP_WRITEV] = { |
d3656344 JA |
752 | .async_ctx = 1, |
753 | .needs_mm = 1, | |
754 | .needs_file = 1, | |
755 | .hash_reg_file = 1, | |
756 | .unbound_nonreg_file = 1, | |
8a72758c | 757 | .pollout = 1, |
d3656344 | 758 | }, |
0463b6c5 | 759 | [IORING_OP_FSYNC] = { |
d3656344 JA |
760 | .needs_file = 1, |
761 | }, | |
0463b6c5 | 762 | [IORING_OP_READ_FIXED] = { |
d3656344 JA |
763 | .needs_file = 1, |
764 | .unbound_nonreg_file = 1, | |
8a72758c | 765 | .pollin = 1, |
d3656344 | 766 | }, |
0463b6c5 | 767 | [IORING_OP_WRITE_FIXED] = { |
d3656344 JA |
768 | .needs_file = 1, |
769 | .hash_reg_file = 1, | |
770 | .unbound_nonreg_file = 1, | |
8a72758c | 771 | .pollout = 1, |
d3656344 | 772 | }, |
0463b6c5 | 773 | [IORING_OP_POLL_ADD] = { |
d3656344 JA |
774 | .needs_file = 1, |
775 | .unbound_nonreg_file = 1, | |
776 | }, | |
0463b6c5 PB |
777 | [IORING_OP_POLL_REMOVE] = {}, |
778 | [IORING_OP_SYNC_FILE_RANGE] = { | |
d3656344 JA |
779 | .needs_file = 1, |
780 | }, | |
0463b6c5 | 781 | [IORING_OP_SENDMSG] = { |
d3656344 JA |
782 | .async_ctx = 1, |
783 | .needs_mm = 1, | |
784 | .needs_file = 1, | |
785 | .unbound_nonreg_file = 1, | |
ff002b30 | 786 | .needs_fs = 1, |
8a72758c | 787 | .pollout = 1, |
d3656344 | 788 | }, |
0463b6c5 | 789 | [IORING_OP_RECVMSG] = { |
d3656344 JA |
790 | .async_ctx = 1, |
791 | .needs_mm = 1, | |
792 | .needs_file = 1, | |
793 | .unbound_nonreg_file = 1, | |
ff002b30 | 794 | .needs_fs = 1, |
8a72758c | 795 | .pollin = 1, |
52de1fe1 | 796 | .buffer_select = 1, |
d3656344 | 797 | }, |
0463b6c5 | 798 | [IORING_OP_TIMEOUT] = { |
d3656344 JA |
799 | .async_ctx = 1, |
800 | .needs_mm = 1, | |
801 | }, | |
0463b6c5 PB |
802 | [IORING_OP_TIMEOUT_REMOVE] = {}, |
803 | [IORING_OP_ACCEPT] = { | |
d3656344 JA |
804 | .needs_mm = 1, |
805 | .needs_file = 1, | |
806 | .unbound_nonreg_file = 1, | |
f86cd20c | 807 | .file_table = 1, |
8a72758c | 808 | .pollin = 1, |
d3656344 | 809 | }, |
0463b6c5 PB |
810 | [IORING_OP_ASYNC_CANCEL] = {}, |
811 | [IORING_OP_LINK_TIMEOUT] = { | |
d3656344 JA |
812 | .async_ctx = 1, |
813 | .needs_mm = 1, | |
814 | }, | |
0463b6c5 | 815 | [IORING_OP_CONNECT] = { |
d3656344 JA |
816 | .async_ctx = 1, |
817 | .needs_mm = 1, | |
818 | .needs_file = 1, | |
819 | .unbound_nonreg_file = 1, | |
8a72758c | 820 | .pollout = 1, |
d3656344 | 821 | }, |
0463b6c5 | 822 | [IORING_OP_FALLOCATE] = { |
d3656344 JA |
823 | .needs_file = 1, |
824 | }, | |
0463b6c5 | 825 | [IORING_OP_OPENAT] = { |
f86cd20c | 826 | .file_table = 1, |
ff002b30 | 827 | .needs_fs = 1, |
d3656344 | 828 | }, |
0463b6c5 | 829 | [IORING_OP_CLOSE] = { |
fd2206e4 JA |
830 | .needs_file = 1, |
831 | .needs_file_no_error = 1, | |
f86cd20c | 832 | .file_table = 1, |
d3656344 | 833 | }, |
0463b6c5 | 834 | [IORING_OP_FILES_UPDATE] = { |
d3656344 | 835 | .needs_mm = 1, |
f86cd20c | 836 | .file_table = 1, |
d3656344 | 837 | }, |
0463b6c5 | 838 | [IORING_OP_STATX] = { |
d3656344 | 839 | .needs_mm = 1, |
ff002b30 | 840 | .needs_fs = 1, |
5b0bbee4 | 841 | .file_table = 1, |
d3656344 | 842 | }, |
0463b6c5 | 843 | [IORING_OP_READ] = { |
3a6820f2 JA |
844 | .needs_mm = 1, |
845 | .needs_file = 1, | |
846 | .unbound_nonreg_file = 1, | |
8a72758c | 847 | .pollin = 1, |
bcda7baa | 848 | .buffer_select = 1, |
3a6820f2 | 849 | }, |
0463b6c5 | 850 | [IORING_OP_WRITE] = { |
3a6820f2 JA |
851 | .needs_mm = 1, |
852 | .needs_file = 1, | |
853 | .unbound_nonreg_file = 1, | |
8a72758c | 854 | .pollout = 1, |
3a6820f2 | 855 | }, |
0463b6c5 | 856 | [IORING_OP_FADVISE] = { |
4840e418 JA |
857 | .needs_file = 1, |
858 | }, | |
0463b6c5 | 859 | [IORING_OP_MADVISE] = { |
c1ca757b JA |
860 | .needs_mm = 1, |
861 | }, | |
0463b6c5 | 862 | [IORING_OP_SEND] = { |
fddaface JA |
863 | .needs_mm = 1, |
864 | .needs_file = 1, | |
865 | .unbound_nonreg_file = 1, | |
8a72758c | 866 | .pollout = 1, |
fddaface | 867 | }, |
0463b6c5 | 868 | [IORING_OP_RECV] = { |
fddaface JA |
869 | .needs_mm = 1, |
870 | .needs_file = 1, | |
871 | .unbound_nonreg_file = 1, | |
8a72758c | 872 | .pollin = 1, |
bcda7baa | 873 | .buffer_select = 1, |
fddaface | 874 | }, |
0463b6c5 | 875 | [IORING_OP_OPENAT2] = { |
f86cd20c | 876 | .file_table = 1, |
ff002b30 | 877 | .needs_fs = 1, |
cebdb986 | 878 | }, |
3e4827b0 JA |
879 | [IORING_OP_EPOLL_CTL] = { |
880 | .unbound_nonreg_file = 1, | |
881 | .file_table = 1, | |
882 | }, | |
7d67af2c PB |
883 | [IORING_OP_SPLICE] = { |
884 | .needs_file = 1, | |
885 | .hash_reg_file = 1, | |
886 | .unbound_nonreg_file = 1, | |
ddf0322d JA |
887 | }, |
888 | [IORING_OP_PROVIDE_BUFFERS] = {}, | |
067524e9 | 889 | [IORING_OP_REMOVE_BUFFERS] = {}, |
f2a8d5c7 PB |
890 | [IORING_OP_TEE] = { |
891 | .needs_file = 1, | |
892 | .hash_reg_file = 1, | |
893 | .unbound_nonreg_file = 1, | |
894 | }, | |
d3656344 JA |
895 | }; |
896 | ||
2e0464d4 BM |
897 | enum io_mem_account { |
898 | ACCT_LOCKED, | |
899 | ACCT_PINNED, | |
900 | }; | |
901 | ||
f3a6fa22 | 902 | static bool io_rw_reissue(struct io_kiocb *req, long res); |
78e19bbe | 903 | static void io_cqring_fill_event(struct io_kiocb *req, long res); |
ec9c02ad | 904 | static void io_put_req(struct io_kiocb *req); |
c40f6379 | 905 | static void io_double_put_req(struct io_kiocb *req); |
978db57e | 906 | static void __io_double_put_req(struct io_kiocb *req); |
94ae5e77 JA |
907 | static struct io_kiocb *io_prep_linked_timeout(struct io_kiocb *req); |
908 | static void io_queue_linked_timeout(struct io_kiocb *req); | |
05f3fb3c JA |
909 | static int __io_sqe_files_update(struct io_ring_ctx *ctx, |
910 | struct io_uring_files_update *ip, | |
911 | unsigned nr_args); | |
f86cd20c | 912 | static int io_grab_files(struct io_kiocb *req); |
2237d765 JA |
913 | static void io_complete_rw_common(struct kiocb *kiocb, long res, |
914 | struct io_comp_state *cs); | |
3ca405eb | 915 | static void __io_clean_op(struct io_kiocb *req); |
b41e9852 JA |
916 | static int io_file_get(struct io_submit_state *state, struct io_kiocb *req, |
917 | int fd, struct file **out_file, bool fixed); | |
918 | static void __io_queue_sqe(struct io_kiocb *req, | |
f13fad7b JA |
919 | const struct io_uring_sqe *sqe, |
920 | struct io_comp_state *cs); | |
4349f30e | 921 | static void io_file_put_work(struct work_struct *work); |
de0617e4 | 922 | |
b63534c4 JA |
923 | static ssize_t io_import_iovec(int rw, struct io_kiocb *req, |
924 | struct iovec **iovec, struct iov_iter *iter, | |
925 | bool needs_lock); | |
926 | static int io_setup_async_rw(struct io_kiocb *req, ssize_t io_size, | |
927 | struct iovec *iovec, struct iovec *fast_iov, | |
928 | struct iov_iter *iter); | |
929 | ||
2b188cc1 JA |
930 | static struct kmem_cache *req_cachep; |
931 | ||
932 | static const struct file_operations io_uring_fops; | |
933 | ||
934 | struct sock *io_uring_get_socket(struct file *file) | |
935 | { | |
936 | #if defined(CONFIG_UNIX) | |
937 | if (file->f_op == &io_uring_fops) { | |
938 | struct io_ring_ctx *ctx = file->private_data; | |
939 | ||
940 | return ctx->ring_sock->sk; | |
941 | } | |
942 | #endif | |
943 | return NULL; | |
944 | } | |
945 | EXPORT_SYMBOL(io_uring_get_socket); | |
946 | ||
4dd2824d PB |
947 | static void io_get_req_task(struct io_kiocb *req) |
948 | { | |
949 | if (req->flags & REQ_F_TASK_PINNED) | |
950 | return; | |
951 | get_task_struct(req->task); | |
952 | req->flags |= REQ_F_TASK_PINNED; | |
953 | } | |
954 | ||
3ca405eb PB |
955 | static inline void io_clean_op(struct io_kiocb *req) |
956 | { | |
957 | if (req->flags & REQ_F_NEED_CLEANUP) | |
958 | __io_clean_op(req); | |
959 | } | |
960 | ||
4dd2824d PB |
961 | /* not idempotent -- it doesn't clear REQ_F_TASK_PINNED */ |
962 | static void __io_put_req_task(struct io_kiocb *req) | |
963 | { | |
964 | if (req->flags & REQ_F_TASK_PINNED) | |
965 | put_task_struct(req->task); | |
966 | } | |
967 | ||
4349f30e | 968 | static void io_sq_thread_drop_mm(void) |
c40f6379 JA |
969 | { |
970 | struct mm_struct *mm = current->mm; | |
971 | ||
972 | if (mm) { | |
973 | kthread_unuse_mm(mm); | |
974 | mmput(mm); | |
975 | } | |
976 | } | |
977 | ||
978 | static int __io_sq_thread_acquire_mm(struct io_ring_ctx *ctx) | |
979 | { | |
980 | if (!current->mm) { | |
8eb06d7e | 981 | if (unlikely(!ctx->sqo_mm || !mmget_not_zero(ctx->sqo_mm))) |
c40f6379 JA |
982 | return -EFAULT; |
983 | kthread_use_mm(ctx->sqo_mm); | |
984 | } | |
985 | ||
986 | return 0; | |
987 | } | |
988 | ||
989 | static int io_sq_thread_acquire_mm(struct io_ring_ctx *ctx, | |
990 | struct io_kiocb *req) | |
991 | { | |
992 | if (!io_op_defs[req->opcode].needs_mm) | |
993 | return 0; | |
994 | return __io_sq_thread_acquire_mm(ctx); | |
995 | } | |
996 | ||
997 | static inline void req_set_fail_links(struct io_kiocb *req) | |
998 | { | |
999 | if ((req->flags & (REQ_F_LINK | REQ_F_HARDLINK)) == REQ_F_LINK) | |
1000 | req->flags |= REQ_F_FAIL_LINK; | |
1001 | } | |
1002 | ||
7cdaf587 XW |
1003 | /* |
1004 | * Note: must call io_req_init_async() for the first time you | |
1005 | * touch any members of io_wq_work. | |
1006 | */ | |
1007 | static inline void io_req_init_async(struct io_kiocb *req) | |
1008 | { | |
1009 | if (req->flags & REQ_F_WORK_INITIALIZED) | |
1010 | return; | |
1011 | ||
1012 | memset(&req->work, 0, sizeof(req->work)); | |
1013 | req->flags |= REQ_F_WORK_INITIALIZED; | |
1014 | } | |
1015 | ||
0cdaf760 PB |
1016 | static inline bool io_async_submit(struct io_ring_ctx *ctx) |
1017 | { | |
1018 | return ctx->flags & IORING_SETUP_SQPOLL; | |
1019 | } | |
1020 | ||
2b188cc1 JA |
1021 | static void io_ring_ctx_ref_free(struct percpu_ref *ref) |
1022 | { | |
1023 | struct io_ring_ctx *ctx = container_of(ref, struct io_ring_ctx, refs); | |
1024 | ||
0f158b4c | 1025 | complete(&ctx->ref_comp); |
2b188cc1 JA |
1026 | } |
1027 | ||
8eb7e2d0 PB |
1028 | static inline bool io_is_timeout_noseq(struct io_kiocb *req) |
1029 | { | |
1030 | return !req->timeout.off; | |
1031 | } | |
1032 | ||
2b188cc1 JA |
1033 | static struct io_ring_ctx *io_ring_ctx_alloc(struct io_uring_params *p) |
1034 | { | |
1035 | struct io_ring_ctx *ctx; | |
78076bb6 | 1036 | int hash_bits; |
2b188cc1 JA |
1037 | |
1038 | ctx = kzalloc(sizeof(*ctx), GFP_KERNEL); | |
1039 | if (!ctx) | |
1040 | return NULL; | |
1041 | ||
0ddf92e8 JA |
1042 | ctx->fallback_req = kmem_cache_alloc(req_cachep, GFP_KERNEL); |
1043 | if (!ctx->fallback_req) | |
1044 | goto err; | |
1045 | ||
78076bb6 JA |
1046 | /* |
1047 | * Use 5 bits less than the max cq entries, that should give us around | |
1048 | * 32 entries per hash list if totally full and uniformly spread. | |
1049 | */ | |
1050 | hash_bits = ilog2(p->cq_entries); | |
1051 | hash_bits -= 5; | |
1052 | if (hash_bits <= 0) | |
1053 | hash_bits = 1; | |
1054 | ctx->cancel_hash_bits = hash_bits; | |
1055 | ctx->cancel_hash = kmalloc((1U << hash_bits) * sizeof(struct hlist_head), | |
1056 | GFP_KERNEL); | |
1057 | if (!ctx->cancel_hash) | |
1058 | goto err; | |
1059 | __hash_init(ctx->cancel_hash, 1U << hash_bits); | |
1060 | ||
21482896 | 1061 | if (percpu_ref_init(&ctx->refs, io_ring_ctx_ref_free, |
206aefde JA |
1062 | PERCPU_REF_ALLOW_REINIT, GFP_KERNEL)) |
1063 | goto err; | |
2b188cc1 JA |
1064 | |
1065 | ctx->flags = p->flags; | |
583863ed | 1066 | init_waitqueue_head(&ctx->sqo_wait); |
2b188cc1 | 1067 | init_waitqueue_head(&ctx->cq_wait); |
1d7bb1d5 | 1068 | INIT_LIST_HEAD(&ctx->cq_overflow_list); |
0f158b4c JA |
1069 | init_completion(&ctx->ref_comp); |
1070 | init_completion(&ctx->sq_thread_comp); | |
5a2e745d | 1071 | idr_init(&ctx->io_buffer_idr); |
071698e1 | 1072 | idr_init(&ctx->personality_idr); |
2b188cc1 JA |
1073 | mutex_init(&ctx->uring_lock); |
1074 | init_waitqueue_head(&ctx->wait); | |
1075 | spin_lock_init(&ctx->completion_lock); | |
540e32a0 | 1076 | INIT_LIST_HEAD(&ctx->iopoll_list); |
de0617e4 | 1077 | INIT_LIST_HEAD(&ctx->defer_list); |
5262f567 | 1078 | INIT_LIST_HEAD(&ctx->timeout_list); |
fcb323cc JA |
1079 | init_waitqueue_head(&ctx->inflight_wait); |
1080 | spin_lock_init(&ctx->inflight_lock); | |
1081 | INIT_LIST_HEAD(&ctx->inflight_list); | |
4a38aed2 JA |
1082 | INIT_DELAYED_WORK(&ctx->file_put_work, io_file_put_work); |
1083 | init_llist_head(&ctx->file_put_llist); | |
2b188cc1 | 1084 | return ctx; |
206aefde | 1085 | err: |
0ddf92e8 JA |
1086 | if (ctx->fallback_req) |
1087 | kmem_cache_free(req_cachep, ctx->fallback_req); | |
78076bb6 | 1088 | kfree(ctx->cancel_hash); |
206aefde JA |
1089 | kfree(ctx); |
1090 | return NULL; | |
2b188cc1 JA |
1091 | } |
1092 | ||
9cf7c104 | 1093 | static bool req_need_defer(struct io_kiocb *req, u32 seq) |
de0617e4 | 1094 | { |
2bc9930e JA |
1095 | if (unlikely(req->flags & REQ_F_IO_DRAIN)) { |
1096 | struct io_ring_ctx *ctx = req->ctx; | |
1097 | ||
9cf7c104 PB |
1098 | return seq != ctx->cached_cq_tail |
1099 | + atomic_read(&ctx->cached_cq_overflow); | |
2bc9930e | 1100 | } |
de0617e4 | 1101 | |
9d858b21 | 1102 | return false; |
de0617e4 JA |
1103 | } |
1104 | ||
de0617e4 | 1105 | static void __io_commit_cqring(struct io_ring_ctx *ctx) |
2b188cc1 | 1106 | { |
75b28aff | 1107 | struct io_rings *rings = ctx->rings; |
2b188cc1 | 1108 | |
07910158 PB |
1109 | /* order cqe stores with ring update */ |
1110 | smp_store_release(&rings->cq.tail, ctx->cached_cq_tail); | |
2b188cc1 | 1111 | |
07910158 PB |
1112 | if (wq_has_sleeper(&ctx->cq_wait)) { |
1113 | wake_up_interruptible(&ctx->cq_wait); | |
1114 | kill_fasync(&ctx->cq_fasync, SIGIO, POLL_IN); | |
2b188cc1 JA |
1115 | } |
1116 | } | |
1117 | ||
dca9cf8b | 1118 | static void io_req_clean_work(struct io_kiocb *req) |
18d9be1a | 1119 | { |
7cdaf587 XW |
1120 | if (!(req->flags & REQ_F_WORK_INITIALIZED)) |
1121 | return; | |
1122 | ||
cccf0ee8 JA |
1123 | if (req->work.mm) { |
1124 | mmdrop(req->work.mm); | |
1125 | req->work.mm = NULL; | |
1126 | } | |
1127 | if (req->work.creds) { | |
1128 | put_cred(req->work.creds); | |
1129 | req->work.creds = NULL; | |
1130 | } | |
ff002b30 JA |
1131 | if (req->work.fs) { |
1132 | struct fs_struct *fs = req->work.fs; | |
1133 | ||
1134 | spin_lock(&req->work.fs->lock); | |
1135 | if (--fs->users) | |
1136 | fs = NULL; | |
1137 | spin_unlock(&req->work.fs->lock); | |
1138 | if (fs) | |
1139 | free_fs_struct(fs); | |
1140 | } | |
561fb04a JA |
1141 | } |
1142 | ||
cbdcb435 | 1143 | static void io_prep_async_work(struct io_kiocb *req) |
18d9be1a | 1144 | { |
d3656344 | 1145 | const struct io_op_def *def = &io_op_defs[req->opcode]; |
54a91f3b | 1146 | |
16d59803 PB |
1147 | io_req_init_async(req); |
1148 | ||
d3656344 JA |
1149 | if (req->flags & REQ_F_ISREG) { |
1150 | if (def->hash_reg_file) | |
8766dd51 | 1151 | io_wq_hash_work(&req->work, file_inode(req->file)); |
d3656344 JA |
1152 | } else { |
1153 | if (def->unbound_nonreg_file) | |
3529d8c2 | 1154 | req->work.flags |= IO_WQ_WORK_UNBOUND; |
54a91f3b | 1155 | } |
dca9cf8b PB |
1156 | if (!req->work.mm && def->needs_mm) { |
1157 | mmgrab(current->mm); | |
1158 | req->work.mm = current->mm; | |
1159 | } | |
1160 | if (!req->work.creds) | |
1161 | req->work.creds = get_current_cred(); | |
1162 | if (!req->work.fs && def->needs_fs) { | |
1163 | spin_lock(¤t->fs->lock); | |
1164 | if (!current->fs->in_exec) { | |
1165 | req->work.fs = current->fs; | |
1166 | req->work.fs->users++; | |
1167 | } else { | |
1168 | req->work.flags |= IO_WQ_WORK_CANCEL; | |
1169 | } | |
1170 | spin_unlock(¤t->fs->lock); | |
1171 | } | |
561fb04a JA |
1172 | } |
1173 | ||
cbdcb435 | 1174 | static void io_prep_async_link(struct io_kiocb *req) |
561fb04a | 1175 | { |
cbdcb435 | 1176 | struct io_kiocb *cur; |
94ae5e77 | 1177 | |
cbdcb435 PB |
1178 | io_prep_async_work(req); |
1179 | if (req->flags & REQ_F_LINK_HEAD) | |
1180 | list_for_each_entry(cur, &req->link_list, link_list) | |
1181 | io_prep_async_work(cur); | |
1182 | } | |
1183 | ||
1184 | static void __io_queue_async_work(struct io_kiocb *req) | |
1185 | { | |
1186 | struct io_ring_ctx *ctx = req->ctx; | |
1187 | struct io_kiocb *link = io_prep_linked_timeout(req); | |
561fb04a | 1188 | |
8766dd51 PB |
1189 | trace_io_uring_queue_async_work(ctx, io_wq_is_hashed(&req->work), req, |
1190 | &req->work, req->flags); | |
1191 | io_wq_enqueue(ctx->io_wq, &req->work); | |
94ae5e77 JA |
1192 | |
1193 | if (link) | |
1194 | io_queue_linked_timeout(link); | |
18d9be1a JA |
1195 | } |
1196 | ||
cbdcb435 PB |
1197 | static void io_queue_async_work(struct io_kiocb *req) |
1198 | { | |
1199 | /* init ->work of the whole link before punting */ | |
1200 | io_prep_async_link(req); | |
1201 | __io_queue_async_work(req); | |
1202 | } | |
1203 | ||
5262f567 JA |
1204 | static void io_kill_timeout(struct io_kiocb *req) |
1205 | { | |
1206 | int ret; | |
1207 | ||
2d28390a | 1208 | ret = hrtimer_try_to_cancel(&req->io->timeout.timer); |
5262f567 JA |
1209 | if (ret != -1) { |
1210 | atomic_inc(&req->ctx->cq_timeouts); | |
135fcde8 | 1211 | list_del_init(&req->timeout.list); |
f0e20b89 | 1212 | req->flags |= REQ_F_COMP_LOCKED; |
78e19bbe | 1213 | io_cqring_fill_event(req, 0); |
ec9c02ad | 1214 | io_put_req(req); |
5262f567 JA |
1215 | } |
1216 | } | |
1217 | ||
1218 | static void io_kill_timeouts(struct io_ring_ctx *ctx) | |
1219 | { | |
1220 | struct io_kiocb *req, *tmp; | |
1221 | ||
1222 | spin_lock_irq(&ctx->completion_lock); | |
135fcde8 | 1223 | list_for_each_entry_safe(req, tmp, &ctx->timeout_list, timeout.list) |
5262f567 JA |
1224 | io_kill_timeout(req); |
1225 | spin_unlock_irq(&ctx->completion_lock); | |
1226 | } | |
1227 | ||
04518945 | 1228 | static void __io_queue_deferred(struct io_ring_ctx *ctx) |
de0617e4 | 1229 | { |
04518945 | 1230 | do { |
27dc8338 PB |
1231 | struct io_defer_entry *de = list_first_entry(&ctx->defer_list, |
1232 | struct io_defer_entry, list); | |
de0617e4 | 1233 | |
9cf7c104 | 1234 | if (req_need_defer(de->req, de->seq)) |
04518945 | 1235 | break; |
27dc8338 | 1236 | list_del_init(&de->list); |
cbdcb435 | 1237 | /* punt-init is done before queueing for defer */ |
27dc8338 PB |
1238 | __io_queue_async_work(de->req); |
1239 | kfree(de); | |
04518945 PB |
1240 | } while (!list_empty(&ctx->defer_list)); |
1241 | } | |
1242 | ||
360428f8 | 1243 | static void io_flush_timeouts(struct io_ring_ctx *ctx) |
de0617e4 | 1244 | { |
360428f8 PB |
1245 | while (!list_empty(&ctx->timeout_list)) { |
1246 | struct io_kiocb *req = list_first_entry(&ctx->timeout_list, | |
135fcde8 | 1247 | struct io_kiocb, timeout.list); |
de0617e4 | 1248 | |
8eb7e2d0 | 1249 | if (io_is_timeout_noseq(req)) |
360428f8 | 1250 | break; |
bfe68a22 PB |
1251 | if (req->timeout.target_seq != ctx->cached_cq_tail |
1252 | - atomic_read(&ctx->cq_timeouts)) | |
360428f8 | 1253 | break; |
bfe68a22 | 1254 | |
135fcde8 | 1255 | list_del_init(&req->timeout.list); |
5262f567 | 1256 | io_kill_timeout(req); |
360428f8 PB |
1257 | } |
1258 | } | |
5262f567 | 1259 | |
360428f8 PB |
1260 | static void io_commit_cqring(struct io_ring_ctx *ctx) |
1261 | { | |
1262 | io_flush_timeouts(ctx); | |
de0617e4 JA |
1263 | __io_commit_cqring(ctx); |
1264 | ||
04518945 PB |
1265 | if (unlikely(!list_empty(&ctx->defer_list))) |
1266 | __io_queue_deferred(ctx); | |
de0617e4 JA |
1267 | } |
1268 | ||
2b188cc1 JA |
1269 | static struct io_uring_cqe *io_get_cqring(struct io_ring_ctx *ctx) |
1270 | { | |
75b28aff | 1271 | struct io_rings *rings = ctx->rings; |
2b188cc1 JA |
1272 | unsigned tail; |
1273 | ||
1274 | tail = ctx->cached_cq_tail; | |
115e12e5 SB |
1275 | /* |
1276 | * writes to the cq entry need to come after reading head; the | |
1277 | * control dependency is enough as we're using WRITE_ONCE to | |
1278 | * fill the cq entry | |
1279 | */ | |
75b28aff | 1280 | if (tail - READ_ONCE(rings->cq.head) == rings->cq_ring_entries) |
2b188cc1 JA |
1281 | return NULL; |
1282 | ||
1283 | ctx->cached_cq_tail++; | |
75b28aff | 1284 | return &rings->cqes[tail & ctx->cq_mask]; |
2b188cc1 JA |
1285 | } |
1286 | ||
f2842ab5 JA |
1287 | static inline bool io_should_trigger_evfd(struct io_ring_ctx *ctx) |
1288 | { | |
f0b493e6 JA |
1289 | if (!ctx->cq_ev_fd) |
1290 | return false; | |
7e55a19c SG |
1291 | if (READ_ONCE(ctx->rings->cq_flags) & IORING_CQ_EVENTFD_DISABLED) |
1292 | return false; | |
f2842ab5 JA |
1293 | if (!ctx->eventfd_async) |
1294 | return true; | |
b41e9852 | 1295 | return io_wq_current_is_worker(); |
f2842ab5 JA |
1296 | } |
1297 | ||
b41e9852 | 1298 | static void io_cqring_ev_posted(struct io_ring_ctx *ctx) |
1d7bb1d5 JA |
1299 | { |
1300 | if (waitqueue_active(&ctx->wait)) | |
1301 | wake_up(&ctx->wait); | |
1302 | if (waitqueue_active(&ctx->sqo_wait)) | |
1303 | wake_up(&ctx->sqo_wait); | |
b41e9852 | 1304 | if (io_should_trigger_evfd(ctx)) |
1d7bb1d5 JA |
1305 | eventfd_signal(ctx->cq_ev_fd, 1); |
1306 | } | |
1307 | ||
c4a2ed72 JA |
1308 | /* Returns true if there are no backlogged entries after the flush */ |
1309 | static bool io_cqring_overflow_flush(struct io_ring_ctx *ctx, bool force) | |
1d7bb1d5 JA |
1310 | { |
1311 | struct io_rings *rings = ctx->rings; | |
1312 | struct io_uring_cqe *cqe; | |
1313 | struct io_kiocb *req; | |
1314 | unsigned long flags; | |
1315 | LIST_HEAD(list); | |
1316 | ||
1317 | if (!force) { | |
1318 | if (list_empty_careful(&ctx->cq_overflow_list)) | |
c4a2ed72 | 1319 | return true; |
1d7bb1d5 JA |
1320 | if ((ctx->cached_cq_tail - READ_ONCE(rings->cq.head) == |
1321 | rings->cq_ring_entries)) | |
c4a2ed72 | 1322 | return false; |
1d7bb1d5 JA |
1323 | } |
1324 | ||
1325 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
1326 | ||
1327 | /* if force is set, the ring is going away. always drop after that */ | |
1328 | if (force) | |
69b3e546 | 1329 | ctx->cq_overflow_flushed = 1; |
1d7bb1d5 | 1330 | |
c4a2ed72 | 1331 | cqe = NULL; |
1d7bb1d5 JA |
1332 | while (!list_empty(&ctx->cq_overflow_list)) { |
1333 | cqe = io_get_cqring(ctx); | |
1334 | if (!cqe && !force) | |
1335 | break; | |
1336 | ||
1337 | req = list_first_entry(&ctx->cq_overflow_list, struct io_kiocb, | |
40d8ddd4 PB |
1338 | compl.list); |
1339 | list_move(&req->compl.list, &list); | |
2ca10259 | 1340 | req->flags &= ~REQ_F_OVERFLOW; |
1d7bb1d5 JA |
1341 | if (cqe) { |
1342 | WRITE_ONCE(cqe->user_data, req->user_data); | |
1343 | WRITE_ONCE(cqe->res, req->result); | |
0f7e466b | 1344 | WRITE_ONCE(cqe->flags, req->compl.cflags); |
1d7bb1d5 JA |
1345 | } else { |
1346 | WRITE_ONCE(ctx->rings->cq_overflow, | |
1347 | atomic_inc_return(&ctx->cached_cq_overflow)); | |
1348 | } | |
1349 | } | |
1350 | ||
1351 | io_commit_cqring(ctx); | |
ad3eb2c8 JA |
1352 | if (cqe) { |
1353 | clear_bit(0, &ctx->sq_check_overflow); | |
1354 | clear_bit(0, &ctx->cq_check_overflow); | |
6d5f9049 | 1355 | ctx->rings->sq_flags &= ~IORING_SQ_CQ_OVERFLOW; |
ad3eb2c8 | 1356 | } |
1d7bb1d5 JA |
1357 | spin_unlock_irqrestore(&ctx->completion_lock, flags); |
1358 | io_cqring_ev_posted(ctx); | |
1359 | ||
1360 | while (!list_empty(&list)) { | |
40d8ddd4 PB |
1361 | req = list_first_entry(&list, struct io_kiocb, compl.list); |
1362 | list_del(&req->compl.list); | |
ec9c02ad | 1363 | io_put_req(req); |
1d7bb1d5 | 1364 | } |
c4a2ed72 JA |
1365 | |
1366 | return cqe != NULL; | |
1d7bb1d5 JA |
1367 | } |
1368 | ||
bcda7baa | 1369 | static void __io_cqring_fill_event(struct io_kiocb *req, long res, long cflags) |
2b188cc1 | 1370 | { |
78e19bbe | 1371 | struct io_ring_ctx *ctx = req->ctx; |
2b188cc1 JA |
1372 | struct io_uring_cqe *cqe; |
1373 | ||
78e19bbe | 1374 | trace_io_uring_complete(ctx, req->user_data, res); |
51c3ff62 | 1375 | |
2b188cc1 JA |
1376 | /* |
1377 | * If we can't get a cq entry, userspace overflowed the | |
1378 | * submission (by quite a lot). Increment the overflow count in | |
1379 | * the ring. | |
1380 | */ | |
1381 | cqe = io_get_cqring(ctx); | |
1d7bb1d5 | 1382 | if (likely(cqe)) { |
78e19bbe | 1383 | WRITE_ONCE(cqe->user_data, req->user_data); |
2b188cc1 | 1384 | WRITE_ONCE(cqe->res, res); |
bcda7baa | 1385 | WRITE_ONCE(cqe->flags, cflags); |
1d7bb1d5 | 1386 | } else if (ctx->cq_overflow_flushed) { |
498ccd9e JA |
1387 | WRITE_ONCE(ctx->rings->cq_overflow, |
1388 | atomic_inc_return(&ctx->cached_cq_overflow)); | |
1d7bb1d5 | 1389 | } else { |
ad3eb2c8 JA |
1390 | if (list_empty(&ctx->cq_overflow_list)) { |
1391 | set_bit(0, &ctx->sq_check_overflow); | |
1392 | set_bit(0, &ctx->cq_check_overflow); | |
6d5f9049 | 1393 | ctx->rings->sq_flags |= IORING_SQ_CQ_OVERFLOW; |
ad3eb2c8 | 1394 | } |
40d8ddd4 | 1395 | io_clean_op(req); |
2ca10259 | 1396 | req->flags |= REQ_F_OVERFLOW; |
1d7bb1d5 | 1397 | req->result = res; |
0f7e466b | 1398 | req->compl.cflags = cflags; |
40d8ddd4 PB |
1399 | refcount_inc(&req->refs); |
1400 | list_add_tail(&req->compl.list, &ctx->cq_overflow_list); | |
2b188cc1 JA |
1401 | } |
1402 | } | |
1403 | ||
bcda7baa JA |
1404 | static void io_cqring_fill_event(struct io_kiocb *req, long res) |
1405 | { | |
1406 | __io_cqring_fill_event(req, res, 0); | |
1407 | } | |
1408 | ||
e1e16097 | 1409 | static void io_cqring_add_event(struct io_kiocb *req, long res, long cflags) |
2b188cc1 | 1410 | { |
78e19bbe | 1411 | struct io_ring_ctx *ctx = req->ctx; |
2b188cc1 JA |
1412 | unsigned long flags; |
1413 | ||
1414 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
bcda7baa | 1415 | __io_cqring_fill_event(req, res, cflags); |
2b188cc1 JA |
1416 | io_commit_cqring(ctx); |
1417 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
1418 | ||
8c838788 | 1419 | io_cqring_ev_posted(ctx); |
2b188cc1 JA |
1420 | } |
1421 | ||
229a7b63 | 1422 | static void io_submit_flush_completions(struct io_comp_state *cs) |
bcda7baa | 1423 | { |
229a7b63 JA |
1424 | struct io_ring_ctx *ctx = cs->ctx; |
1425 | ||
1426 | spin_lock_irq(&ctx->completion_lock); | |
1427 | while (!list_empty(&cs->list)) { | |
1428 | struct io_kiocb *req; | |
1429 | ||
3ca405eb PB |
1430 | req = list_first_entry(&cs->list, struct io_kiocb, compl.list); |
1431 | list_del(&req->compl.list); | |
0f7e466b | 1432 | __io_cqring_fill_event(req, req->result, req->compl.cflags); |
229a7b63 JA |
1433 | if (!(req->flags & REQ_F_LINK_HEAD)) { |
1434 | req->flags |= REQ_F_COMP_LOCKED; | |
1435 | io_put_req(req); | |
1436 | } else { | |
1437 | spin_unlock_irq(&ctx->completion_lock); | |
1438 | io_put_req(req); | |
1439 | spin_lock_irq(&ctx->completion_lock); | |
1440 | } | |
1441 | } | |
1442 | io_commit_cqring(ctx); | |
1443 | spin_unlock_irq(&ctx->completion_lock); | |
1444 | ||
1445 | io_cqring_ev_posted(ctx); | |
1446 | cs->nr = 0; | |
1447 | } | |
1448 | ||
1449 | static void __io_req_complete(struct io_kiocb *req, long res, unsigned cflags, | |
1450 | struct io_comp_state *cs) | |
1451 | { | |
1452 | if (!cs) { | |
1453 | io_cqring_add_event(req, res, cflags); | |
1454 | io_put_req(req); | |
1455 | } else { | |
3ca405eb | 1456 | io_clean_op(req); |
229a7b63 | 1457 | req->result = res; |
0f7e466b | 1458 | req->compl.cflags = cflags; |
3ca405eb | 1459 | list_add_tail(&req->compl.list, &cs->list); |
229a7b63 JA |
1460 | if (++cs->nr >= 32) |
1461 | io_submit_flush_completions(cs); | |
1462 | } | |
e1e16097 JA |
1463 | } |
1464 | ||
1465 | static void io_req_complete(struct io_kiocb *req, long res) | |
1466 | { | |
229a7b63 | 1467 | __io_req_complete(req, res, 0, NULL); |
bcda7baa JA |
1468 | } |
1469 | ||
0ddf92e8 JA |
1470 | static inline bool io_is_fallback_req(struct io_kiocb *req) |
1471 | { | |
1472 | return req == (struct io_kiocb *) | |
1473 | ((unsigned long) req->ctx->fallback_req & ~1UL); | |
1474 | } | |
1475 | ||
1476 | static struct io_kiocb *io_get_fallback_req(struct io_ring_ctx *ctx) | |
1477 | { | |
1478 | struct io_kiocb *req; | |
1479 | ||
1480 | req = ctx->fallback_req; | |
dd461af6 | 1481 | if (!test_and_set_bit_lock(0, (unsigned long *) &ctx->fallback_req)) |
0ddf92e8 JA |
1482 | return req; |
1483 | ||
1484 | return NULL; | |
1485 | } | |
1486 | ||
0553b8bd PB |
1487 | static struct io_kiocb *io_alloc_req(struct io_ring_ctx *ctx, |
1488 | struct io_submit_state *state) | |
2b188cc1 | 1489 | { |
fd6fab2c | 1490 | gfp_t gfp = GFP_KERNEL | __GFP_NOWARN; |
2b188cc1 JA |
1491 | struct io_kiocb *req; |
1492 | ||
f6b6c7d6 | 1493 | if (!state->free_reqs) { |
2579f913 JA |
1494 | size_t sz; |
1495 | int ret; | |
1496 | ||
1497 | sz = min_t(size_t, state->ios_left, ARRAY_SIZE(state->reqs)); | |
fd6fab2c JA |
1498 | ret = kmem_cache_alloc_bulk(req_cachep, gfp, sz, state->reqs); |
1499 | ||
1500 | /* | |
1501 | * Bulk alloc is all-or-nothing. If we fail to get a batch, | |
1502 | * retry single alloc to be on the safe side. | |
1503 | */ | |
1504 | if (unlikely(ret <= 0)) { | |
1505 | state->reqs[0] = kmem_cache_alloc(req_cachep, gfp); | |
1506 | if (!state->reqs[0]) | |
0ddf92e8 | 1507 | goto fallback; |
fd6fab2c JA |
1508 | ret = 1; |
1509 | } | |
2579f913 | 1510 | state->free_reqs = ret - 1; |
6c8a3134 | 1511 | req = state->reqs[ret - 1]; |
2579f913 | 1512 | } else { |
2579f913 | 1513 | state->free_reqs--; |
6c8a3134 | 1514 | req = state->reqs[state->free_reqs]; |
2b188cc1 JA |
1515 | } |
1516 | ||
2579f913 | 1517 | return req; |
0ddf92e8 | 1518 | fallback: |
0553b8bd | 1519 | return io_get_fallback_req(ctx); |
2b188cc1 JA |
1520 | } |
1521 | ||
8da11c19 PB |
1522 | static inline void io_put_file(struct io_kiocb *req, struct file *file, |
1523 | bool fixed) | |
1524 | { | |
1525 | if (fixed) | |
05589553 | 1526 | percpu_ref_put(req->fixed_file_refs); |
8da11c19 PB |
1527 | else |
1528 | fput(file); | |
1529 | } | |
1530 | ||
e6543a81 | 1531 | static void io_dismantle_req(struct io_kiocb *req) |
2b188cc1 | 1532 | { |
3ca405eb | 1533 | io_clean_op(req); |
929a3af9 | 1534 | |
5acbbc8e JA |
1535 | if (req->io) |
1536 | kfree(req->io); | |
8da11c19 PB |
1537 | if (req->file) |
1538 | io_put_file(req, req->file, (req->flags & REQ_F_FIXED_FILE)); | |
4dd2824d | 1539 | __io_put_req_task(req); |
dca9cf8b | 1540 | io_req_clean_work(req); |
fcb323cc | 1541 | |
fcb323cc | 1542 | if (req->flags & REQ_F_INFLIGHT) { |
c6ca97b3 | 1543 | struct io_ring_ctx *ctx = req->ctx; |
fcb323cc JA |
1544 | unsigned long flags; |
1545 | ||
1546 | spin_lock_irqsave(&ctx->inflight_lock, flags); | |
1547 | list_del(&req->inflight_entry); | |
1548 | if (waitqueue_active(&ctx->inflight_wait)) | |
1549 | wake_up(&ctx->inflight_wait); | |
1550 | spin_unlock_irqrestore(&ctx->inflight_lock, flags); | |
1551 | } | |
e6543a81 | 1552 | } |
2b85edfc | 1553 | |
e6543a81 PB |
1554 | static void __io_free_req(struct io_kiocb *req) |
1555 | { | |
ecfc5177 PB |
1556 | struct io_ring_ctx *ctx; |
1557 | ||
e6543a81 | 1558 | io_dismantle_req(req); |
ecfc5177 | 1559 | ctx = req->ctx; |
b1e50e54 PB |
1560 | if (likely(!io_is_fallback_req(req))) |
1561 | kmem_cache_free(req_cachep, req); | |
1562 | else | |
ecfc5177 PB |
1563 | clear_bit_unlock(0, (unsigned long *) &ctx->fallback_req); |
1564 | percpu_ref_put(&ctx->refs); | |
e65ef56d JA |
1565 | } |
1566 | ||
a197f664 | 1567 | static bool io_link_cancel_timeout(struct io_kiocb *req) |
2665abfd | 1568 | { |
a197f664 | 1569 | struct io_ring_ctx *ctx = req->ctx; |
2665abfd JA |
1570 | int ret; |
1571 | ||
2d28390a | 1572 | ret = hrtimer_try_to_cancel(&req->io->timeout.timer); |
2665abfd | 1573 | if (ret != -1) { |
78e19bbe | 1574 | io_cqring_fill_event(req, -ECANCELED); |
2665abfd | 1575 | io_commit_cqring(ctx); |
dea3b49c | 1576 | req->flags &= ~REQ_F_LINK_HEAD; |
ec9c02ad | 1577 | io_put_req(req); |
2665abfd JA |
1578 | return true; |
1579 | } | |
1580 | ||
1581 | return false; | |
e65ef56d JA |
1582 | } |
1583 | ||
ab0b6451 | 1584 | static bool __io_kill_linked_timeout(struct io_kiocb *req) |
9e645e11 | 1585 | { |
7c86ffee | 1586 | struct io_kiocb *link; |
ab0b6451 | 1587 | bool wake_ev; |
7c86ffee PB |
1588 | |
1589 | if (list_empty(&req->link_list)) | |
ab0b6451 | 1590 | return false; |
7c86ffee PB |
1591 | link = list_first_entry(&req->link_list, struct io_kiocb, link_list); |
1592 | if (link->opcode != IORING_OP_LINK_TIMEOUT) | |
ab0b6451 | 1593 | return false; |
7c86ffee PB |
1594 | |
1595 | list_del_init(&link->link_list); | |
1596 | wake_ev = io_link_cancel_timeout(link); | |
1597 | req->flags &= ~REQ_F_LINK_TIMEOUT; | |
ab0b6451 JA |
1598 | return wake_ev; |
1599 | } | |
1600 | ||
1601 | static void io_kill_linked_timeout(struct io_kiocb *req) | |
1602 | { | |
1603 | struct io_ring_ctx *ctx = req->ctx; | |
1604 | bool wake_ev; | |
1605 | ||
1606 | if (!(req->flags & REQ_F_COMP_LOCKED)) { | |
1607 | unsigned long flags; | |
1608 | ||
1609 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
1610 | wake_ev = __io_kill_linked_timeout(req); | |
7c86ffee | 1611 | spin_unlock_irqrestore(&ctx->completion_lock, flags); |
ab0b6451 JA |
1612 | } else { |
1613 | wake_ev = __io_kill_linked_timeout(req); | |
1614 | } | |
1615 | ||
7c86ffee PB |
1616 | if (wake_ev) |
1617 | io_cqring_ev_posted(ctx); | |
1618 | } | |
1619 | ||
9b5f7bd9 | 1620 | static struct io_kiocb *io_req_link_next(struct io_kiocb *req) |
7c86ffee PB |
1621 | { |
1622 | struct io_kiocb *nxt; | |
9e645e11 JA |
1623 | |
1624 | /* | |
1625 | * The list should never be empty when we are called here. But could | |
1626 | * potentially happen if the chain is messed up, check to be on the | |
1627 | * safe side. | |
1628 | */ | |
7c86ffee | 1629 | if (unlikely(list_empty(&req->link_list))) |
9b5f7bd9 | 1630 | return NULL; |
2665abfd | 1631 | |
7c86ffee PB |
1632 | nxt = list_first_entry(&req->link_list, struct io_kiocb, link_list); |
1633 | list_del_init(&req->link_list); | |
1634 | if (!list_empty(&nxt->link_list)) | |
1635 | nxt->flags |= REQ_F_LINK_HEAD; | |
9b5f7bd9 | 1636 | return nxt; |
9e645e11 JA |
1637 | } |
1638 | ||
1639 | /* | |
dea3b49c | 1640 | * Called if REQ_F_LINK_HEAD is set, and we fail the head request |
9e645e11 | 1641 | */ |
7c86ffee | 1642 | static void __io_fail_links(struct io_kiocb *req) |
9e645e11 | 1643 | { |
2665abfd | 1644 | struct io_ring_ctx *ctx = req->ctx; |
9e645e11 JA |
1645 | |
1646 | while (!list_empty(&req->link_list)) { | |
4493233e PB |
1647 | struct io_kiocb *link = list_first_entry(&req->link_list, |
1648 | struct io_kiocb, link_list); | |
9e645e11 | 1649 | |
4493233e | 1650 | list_del_init(&link->link_list); |
c826bd7a | 1651 | trace_io_uring_fail_link(req, link); |
2665abfd | 1652 | |
7c86ffee PB |
1653 | io_cqring_fill_event(link, -ECANCELED); |
1654 | __io_double_put_req(link); | |
5d960724 | 1655 | req->flags &= ~REQ_F_LINK_TIMEOUT; |
9e645e11 | 1656 | } |
2665abfd JA |
1657 | |
1658 | io_commit_cqring(ctx); | |
7c86ffee PB |
1659 | io_cqring_ev_posted(ctx); |
1660 | } | |
1661 | ||
1662 | static void io_fail_links(struct io_kiocb *req) | |
1663 | { | |
1664 | struct io_ring_ctx *ctx = req->ctx; | |
1665 | ||
1666 | if (!(req->flags & REQ_F_COMP_LOCKED)) { | |
1667 | unsigned long flags; | |
1668 | ||
1669 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
1670 | __io_fail_links(req); | |
1671 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
1672 | } else { | |
1673 | __io_fail_links(req); | |
1674 | } | |
1675 | ||
2665abfd | 1676 | io_cqring_ev_posted(ctx); |
9e645e11 JA |
1677 | } |
1678 | ||
3fa5e0f3 | 1679 | static struct io_kiocb *__io_req_find_next(struct io_kiocb *req) |
9e645e11 | 1680 | { |
9b0d911a | 1681 | req->flags &= ~REQ_F_LINK_HEAD; |
7c86ffee PB |
1682 | if (req->flags & REQ_F_LINK_TIMEOUT) |
1683 | io_kill_linked_timeout(req); | |
1684 | ||
9e645e11 JA |
1685 | /* |
1686 | * If LINK is set, we have dependent requests in this chain. If we | |
1687 | * didn't fail this request, queue the first one up, moving any other | |
1688 | * dependencies to the next request. In case of failure, fail the rest | |
1689 | * of the chain. | |
1690 | */ | |
9b5f7bd9 PB |
1691 | if (likely(!(req->flags & REQ_F_FAIL_LINK))) |
1692 | return io_req_link_next(req); | |
1693 | io_fail_links(req); | |
1694 | return NULL; | |
4d7dd462 | 1695 | } |
9e645e11 | 1696 | |
3fa5e0f3 PB |
1697 | static struct io_kiocb *io_req_find_next(struct io_kiocb *req) |
1698 | { | |
1699 | if (likely(!(req->flags & REQ_F_LINK_HEAD))) | |
1700 | return NULL; | |
1701 | return __io_req_find_next(req); | |
1702 | } | |
1703 | ||
c2c4c83c JA |
1704 | static int io_req_task_work_add(struct io_kiocb *req, struct callback_head *cb) |
1705 | { | |
1706 | struct task_struct *tsk = req->task; | |
1707 | struct io_ring_ctx *ctx = req->ctx; | |
1708 | int ret, notify = TWA_RESUME; | |
1709 | ||
1710 | /* | |
1711 | * SQPOLL kernel thread doesn't need notification, just a wakeup. | |
1712 | * If we're not using an eventfd, then TWA_RESUME is always fine, | |
1713 | * as we won't have dependencies between request completions for | |
1714 | * other kernel wait conditions. | |
1715 | */ | |
1716 | if (ctx->flags & IORING_SETUP_SQPOLL) | |
1717 | notify = 0; | |
1718 | else if (ctx->cq_ev_fd) | |
1719 | notify = TWA_SIGNAL; | |
1720 | ||
1721 | ret = task_work_add(tsk, cb, notify); | |
1722 | if (!ret) | |
1723 | wake_up_process(tsk); | |
1724 | return ret; | |
1725 | } | |
1726 | ||
c40f6379 JA |
1727 | static void __io_req_task_cancel(struct io_kiocb *req, int error) |
1728 | { | |
1729 | struct io_ring_ctx *ctx = req->ctx; | |
1730 | ||
1731 | spin_lock_irq(&ctx->completion_lock); | |
1732 | io_cqring_fill_event(req, error); | |
1733 | io_commit_cqring(ctx); | |
1734 | spin_unlock_irq(&ctx->completion_lock); | |
1735 | ||
1736 | io_cqring_ev_posted(ctx); | |
1737 | req_set_fail_links(req); | |
1738 | io_double_put_req(req); | |
1739 | } | |
1740 | ||
1741 | static void io_req_task_cancel(struct callback_head *cb) | |
1742 | { | |
1743 | struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work); | |
1744 | ||
1745 | __io_req_task_cancel(req, -ECANCELED); | |
1746 | } | |
1747 | ||
1748 | static void __io_req_task_submit(struct io_kiocb *req) | |
1749 | { | |
1750 | struct io_ring_ctx *ctx = req->ctx; | |
1751 | ||
c40f6379 JA |
1752 | if (!__io_sq_thread_acquire_mm(ctx)) { |
1753 | mutex_lock(&ctx->uring_lock); | |
1754 | __io_queue_sqe(req, NULL, NULL); | |
1755 | mutex_unlock(&ctx->uring_lock); | |
1756 | } else { | |
1757 | __io_req_task_cancel(req, -EFAULT); | |
1758 | } | |
1759 | } | |
1760 | ||
1761 | static void io_req_task_submit(struct callback_head *cb) | |
1762 | { | |
1763 | struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work); | |
1764 | ||
1765 | __io_req_task_submit(req); | |
1766 | } | |
1767 | ||
1768 | static void io_req_task_queue(struct io_kiocb *req) | |
1769 | { | |
c40f6379 JA |
1770 | int ret; |
1771 | ||
1772 | init_task_work(&req->task_work, io_req_task_submit); | |
1773 | ||
c2c4c83c | 1774 | ret = io_req_task_work_add(req, &req->task_work); |
c40f6379 | 1775 | if (unlikely(ret)) { |
c2c4c83c JA |
1776 | struct task_struct *tsk; |
1777 | ||
c40f6379 JA |
1778 | init_task_work(&req->task_work, io_req_task_cancel); |
1779 | tsk = io_wq_get_task(req->ctx->io_wq); | |
c2c4c83c JA |
1780 | task_work_add(tsk, &req->task_work, 0); |
1781 | wake_up_process(tsk); | |
c40f6379 | 1782 | } |
c40f6379 JA |
1783 | } |
1784 | ||
c3524383 | 1785 | static void io_queue_next(struct io_kiocb *req) |
c69f8dbe | 1786 | { |
9b5f7bd9 | 1787 | struct io_kiocb *nxt = io_req_find_next(req); |
944e58bf | 1788 | |
906a8c3f PB |
1789 | if (nxt) |
1790 | io_req_task_queue(nxt); | |
c69f8dbe JL |
1791 | } |
1792 | ||
c3524383 PB |
1793 | static void io_free_req(struct io_kiocb *req) |
1794 | { | |
1795 | io_queue_next(req); | |
1796 | __io_free_req(req); | |
1797 | } | |
1798 | ||
2d6500d4 PB |
1799 | struct req_batch { |
1800 | void *reqs[IO_IOPOLL_BATCH]; | |
1801 | int to_free; | |
1802 | }; | |
1803 | ||
1804 | static void __io_req_free_batch_flush(struct io_ring_ctx *ctx, | |
1805 | struct req_batch *rb) | |
1806 | { | |
1807 | kmem_cache_free_bulk(req_cachep, rb->to_free, rb->reqs); | |
1808 | percpu_ref_put_many(&ctx->refs, rb->to_free); | |
1809 | rb->to_free = 0; | |
1810 | } | |
1811 | ||
1812 | static void io_req_free_batch_finish(struct io_ring_ctx *ctx, | |
1813 | struct req_batch *rb) | |
1814 | { | |
1815 | if (rb->to_free) | |
1816 | __io_req_free_batch_flush(ctx, rb); | |
1817 | } | |
1818 | ||
1819 | static void io_req_free_batch(struct req_batch *rb, struct io_kiocb *req) | |
1820 | { | |
1821 | if (unlikely(io_is_fallback_req(req))) { | |
1822 | io_free_req(req); | |
1823 | return; | |
1824 | } | |
1825 | if (req->flags & REQ_F_LINK_HEAD) | |
1826 | io_queue_next(req); | |
1827 | ||
1828 | io_dismantle_req(req); | |
1829 | rb->reqs[rb->to_free++] = req; | |
1830 | if (unlikely(rb->to_free == ARRAY_SIZE(rb->reqs))) | |
1831 | __io_req_free_batch_flush(req->ctx, rb); | |
1832 | } | |
1833 | ||
ba816ad6 JA |
1834 | /* |
1835 | * Drop reference to request, return next in chain (if there is one) if this | |
1836 | * was the last reference to this request. | |
1837 | */ | |
9b5f7bd9 | 1838 | static struct io_kiocb *io_put_req_find_next(struct io_kiocb *req) |
e65ef56d | 1839 | { |
9b5f7bd9 PB |
1840 | struct io_kiocb *nxt = NULL; |
1841 | ||
2a44f467 | 1842 | if (refcount_dec_and_test(&req->refs)) { |
9b5f7bd9 | 1843 | nxt = io_req_find_next(req); |
4d7dd462 | 1844 | __io_free_req(req); |
2a44f467 | 1845 | } |
9b5f7bd9 | 1846 | return nxt; |
2b188cc1 JA |
1847 | } |
1848 | ||
e65ef56d JA |
1849 | static void io_put_req(struct io_kiocb *req) |
1850 | { | |
1851 | if (refcount_dec_and_test(&req->refs)) | |
1852 | io_free_req(req); | |
2b188cc1 JA |
1853 | } |
1854 | ||
f4db7182 | 1855 | static struct io_wq_work *io_steal_work(struct io_kiocb *req) |
7a743e22 | 1856 | { |
6df1db6b | 1857 | struct io_kiocb *nxt; |
f4db7182 | 1858 | |
7a743e22 | 1859 | /* |
f4db7182 PB |
1860 | * A ref is owned by io-wq in which context we're. So, if that's the |
1861 | * last one, it's safe to steal next work. False negatives are Ok, | |
1862 | * it just will be re-punted async in io_put_work() | |
7a743e22 | 1863 | */ |
f4db7182 PB |
1864 | if (refcount_read(&req->refs) != 1) |
1865 | return NULL; | |
7a743e22 | 1866 | |
9b5f7bd9 | 1867 | nxt = io_req_find_next(req); |
6df1db6b | 1868 | return nxt ? &nxt->work : NULL; |
7a743e22 PB |
1869 | } |
1870 | ||
978db57e JA |
1871 | /* |
1872 | * Must only be used if we don't need to care about links, usually from | |
1873 | * within the completion handling itself. | |
1874 | */ | |
1875 | static void __io_double_put_req(struct io_kiocb *req) | |
78e19bbe JA |
1876 | { |
1877 | /* drop both submit and complete references */ | |
1878 | if (refcount_sub_and_test(2, &req->refs)) | |
1879 | __io_free_req(req); | |
1880 | } | |
1881 | ||
978db57e JA |
1882 | static void io_double_put_req(struct io_kiocb *req) |
1883 | { | |
1884 | /* drop both submit and complete references */ | |
1885 | if (refcount_sub_and_test(2, &req->refs)) | |
1886 | io_free_req(req); | |
1887 | } | |
1888 | ||
1d7bb1d5 | 1889 | static unsigned io_cqring_events(struct io_ring_ctx *ctx, bool noflush) |
a3a0e43f | 1890 | { |
84f97dc2 JA |
1891 | struct io_rings *rings = ctx->rings; |
1892 | ||
ad3eb2c8 JA |
1893 | if (test_bit(0, &ctx->cq_check_overflow)) { |
1894 | /* | |
1895 | * noflush == true is from the waitqueue handler, just ensure | |
1896 | * we wake up the task, and the next invocation will flush the | |
1897 | * entries. We cannot safely to it from here. | |
1898 | */ | |
1899 | if (noflush && !list_empty(&ctx->cq_overflow_list)) | |
1900 | return -1U; | |
1d7bb1d5 | 1901 | |
ad3eb2c8 JA |
1902 | io_cqring_overflow_flush(ctx, false); |
1903 | } | |
1d7bb1d5 | 1904 | |
a3a0e43f JA |
1905 | /* See comment at the top of this file */ |
1906 | smp_rmb(); | |
ad3eb2c8 | 1907 | return ctx->cached_cq_tail - READ_ONCE(rings->cq.head); |
a3a0e43f JA |
1908 | } |
1909 | ||
fb5ccc98 PB |
1910 | static inline unsigned int io_sqring_entries(struct io_ring_ctx *ctx) |
1911 | { | |
1912 | struct io_rings *rings = ctx->rings; | |
1913 | ||
1914 | /* make sure SQ entry isn't read before tail */ | |
1915 | return smp_load_acquire(&rings->sq.tail) - ctx->cached_sq_head; | |
1916 | } | |
1917 | ||
bcda7baa JA |
1918 | static int io_put_kbuf(struct io_kiocb *req) |
1919 | { | |
4d954c25 | 1920 | struct io_buffer *kbuf; |
bcda7baa JA |
1921 | int cflags; |
1922 | ||
4d954c25 | 1923 | kbuf = (struct io_buffer *) (unsigned long) req->rw.addr; |
bcda7baa JA |
1924 | cflags = kbuf->bid << IORING_CQE_BUFFER_SHIFT; |
1925 | cflags |= IORING_CQE_F_BUFFER; | |
1926 | req->rw.addr = 0; | |
1927 | kfree(kbuf); | |
1928 | return cflags; | |
1929 | } | |
1930 | ||
4c6e277c JA |
1931 | static inline bool io_run_task_work(void) |
1932 | { | |
1933 | if (current->task_works) { | |
1934 | __set_current_state(TASK_RUNNING); | |
1935 | task_work_run(); | |
1936 | return true; | |
1937 | } | |
1938 | ||
1939 | return false; | |
1940 | } | |
1941 | ||
bbde017a XW |
1942 | static void io_iopoll_queue(struct list_head *again) |
1943 | { | |
1944 | struct io_kiocb *req; | |
1945 | ||
1946 | do { | |
d21ffe7e PB |
1947 | req = list_first_entry(again, struct io_kiocb, inflight_entry); |
1948 | list_del(&req->inflight_entry); | |
cf2f5425 | 1949 | if (!io_rw_reissue(req, -EAGAIN)) |
2237d765 | 1950 | io_complete_rw_common(&req->rw.kiocb, -EAGAIN, NULL); |
bbde017a XW |
1951 | } while (!list_empty(again)); |
1952 | } | |
1953 | ||
def596e9 JA |
1954 | /* |
1955 | * Find and free completed poll iocbs | |
1956 | */ | |
1957 | static void io_iopoll_complete(struct io_ring_ctx *ctx, unsigned int *nr_events, | |
1958 | struct list_head *done) | |
1959 | { | |
8237e045 | 1960 | struct req_batch rb; |
def596e9 | 1961 | struct io_kiocb *req; |
bbde017a XW |
1962 | LIST_HEAD(again); |
1963 | ||
1964 | /* order with ->result store in io_complete_rw_iopoll() */ | |
1965 | smp_rmb(); | |
def596e9 | 1966 | |
2757a23e | 1967 | rb.to_free = 0; |
def596e9 | 1968 | while (!list_empty(done)) { |
bcda7baa JA |
1969 | int cflags = 0; |
1970 | ||
d21ffe7e | 1971 | req = list_first_entry(done, struct io_kiocb, inflight_entry); |
bbde017a XW |
1972 | if (READ_ONCE(req->result) == -EAGAIN) { |
1973 | req->iopoll_completed = 0; | |
d21ffe7e | 1974 | list_move_tail(&req->inflight_entry, &again); |
bbde017a XW |
1975 | continue; |
1976 | } | |
d21ffe7e | 1977 | list_del(&req->inflight_entry); |
def596e9 | 1978 | |
bcda7baa JA |
1979 | if (req->flags & REQ_F_BUFFER_SELECTED) |
1980 | cflags = io_put_kbuf(req); | |
1981 | ||
1982 | __io_cqring_fill_event(req, req->result, cflags); | |
def596e9 JA |
1983 | (*nr_events)++; |
1984 | ||
c3524383 | 1985 | if (refcount_dec_and_test(&req->refs)) |
2d6500d4 | 1986 | io_req_free_batch(&rb, req); |
def596e9 | 1987 | } |
def596e9 | 1988 | |
09bb8394 | 1989 | io_commit_cqring(ctx); |
32b2244a XW |
1990 | if (ctx->flags & IORING_SETUP_SQPOLL) |
1991 | io_cqring_ev_posted(ctx); | |
2d6500d4 | 1992 | io_req_free_batch_finish(ctx, &rb); |
581f9810 | 1993 | |
bbde017a XW |
1994 | if (!list_empty(&again)) |
1995 | io_iopoll_queue(&again); | |
581f9810 BM |
1996 | } |
1997 | ||
def596e9 JA |
1998 | static int io_do_iopoll(struct io_ring_ctx *ctx, unsigned int *nr_events, |
1999 | long min) | |
2000 | { | |
2001 | struct io_kiocb *req, *tmp; | |
2002 | LIST_HEAD(done); | |
2003 | bool spin; | |
2004 | int ret; | |
2005 | ||
2006 | /* | |
2007 | * Only spin for completions if we don't have multiple devices hanging | |
2008 | * off our complete list, and we're under the requested amount. | |
2009 | */ | |
2010 | spin = !ctx->poll_multi_file && *nr_events < min; | |
2011 | ||
2012 | ret = 0; | |
d21ffe7e | 2013 | list_for_each_entry_safe(req, tmp, &ctx->iopoll_list, inflight_entry) { |
9adbd45d | 2014 | struct kiocb *kiocb = &req->rw.kiocb; |
def596e9 JA |
2015 | |
2016 | /* | |
581f9810 BM |
2017 | * Move completed and retryable entries to our local lists. |
2018 | * If we find a request that requires polling, break out | |
2019 | * and complete those lists first, if we have entries there. | |
def596e9 | 2020 | */ |
65a6543d | 2021 | if (READ_ONCE(req->iopoll_completed)) { |
d21ffe7e | 2022 | list_move_tail(&req->inflight_entry, &done); |
def596e9 JA |
2023 | continue; |
2024 | } | |
2025 | if (!list_empty(&done)) | |
2026 | break; | |
2027 | ||
2028 | ret = kiocb->ki_filp->f_op->iopoll(kiocb, spin); | |
2029 | if (ret < 0) | |
2030 | break; | |
2031 | ||
3aadc23e PB |
2032 | /* iopoll may have completed current req */ |
2033 | if (READ_ONCE(req->iopoll_completed)) | |
d21ffe7e | 2034 | list_move_tail(&req->inflight_entry, &done); |
3aadc23e | 2035 | |
def596e9 JA |
2036 | if (ret && spin) |
2037 | spin = false; | |
2038 | ret = 0; | |
2039 | } | |
2040 | ||
2041 | if (!list_empty(&done)) | |
2042 | io_iopoll_complete(ctx, nr_events, &done); | |
2043 | ||
2044 | return ret; | |
2045 | } | |
2046 | ||
2047 | /* | |
d195a66e | 2048 | * Poll for a minimum of 'min' events. Note that if min == 0 we consider that a |
def596e9 JA |
2049 | * non-spinning poll check - we'll still enter the driver poll loop, but only |
2050 | * as a non-spinning completion check. | |
2051 | */ | |
2052 | static int io_iopoll_getevents(struct io_ring_ctx *ctx, unsigned int *nr_events, | |
2053 | long min) | |
2054 | { | |
540e32a0 | 2055 | while (!list_empty(&ctx->iopoll_list) && !need_resched()) { |
def596e9 JA |
2056 | int ret; |
2057 | ||
2058 | ret = io_do_iopoll(ctx, nr_events, min); | |
2059 | if (ret < 0) | |
2060 | return ret; | |
eba0a4dd | 2061 | if (*nr_events >= min) |
def596e9 JA |
2062 | return 0; |
2063 | } | |
2064 | ||
2065 | return 1; | |
2066 | } | |
2067 | ||
2068 | /* | |
2069 | * We can't just wait for polled events to come to us, we have to actively | |
2070 | * find and complete them. | |
2071 | */ | |
b2edc0a7 | 2072 | static void io_iopoll_try_reap_events(struct io_ring_ctx *ctx) |
def596e9 JA |
2073 | { |
2074 | if (!(ctx->flags & IORING_SETUP_IOPOLL)) | |
2075 | return; | |
2076 | ||
2077 | mutex_lock(&ctx->uring_lock); | |
540e32a0 | 2078 | while (!list_empty(&ctx->iopoll_list)) { |
def596e9 JA |
2079 | unsigned int nr_events = 0; |
2080 | ||
b2edc0a7 | 2081 | io_do_iopoll(ctx, &nr_events, 0); |
08f5439f | 2082 | |
b2edc0a7 PB |
2083 | /* let it sleep and repeat later if can't complete a request */ |
2084 | if (nr_events == 0) | |
2085 | break; | |
08f5439f JA |
2086 | /* |
2087 | * Ensure we allow local-to-the-cpu processing to take place, | |
2088 | * in this case we need to ensure that we reap all events. | |
3fcee5a6 | 2089 | * Also let task_work, etc. to progress by releasing the mutex |
08f5439f | 2090 | */ |
3fcee5a6 PB |
2091 | if (need_resched()) { |
2092 | mutex_unlock(&ctx->uring_lock); | |
2093 | cond_resched(); | |
2094 | mutex_lock(&ctx->uring_lock); | |
2095 | } | |
def596e9 JA |
2096 | } |
2097 | mutex_unlock(&ctx->uring_lock); | |
2098 | } | |
2099 | ||
7668b92a | 2100 | static int io_iopoll_check(struct io_ring_ctx *ctx, long min) |
def596e9 | 2101 | { |
7668b92a | 2102 | unsigned int nr_events = 0; |
2b2ed975 | 2103 | int iters = 0, ret = 0; |
500f9fba | 2104 | |
c7849be9 XW |
2105 | /* |
2106 | * We disallow the app entering submit/complete with polling, but we | |
2107 | * still need to lock the ring to prevent racing with polled issue | |
2108 | * that got punted to a workqueue. | |
2109 | */ | |
2110 | mutex_lock(&ctx->uring_lock); | |
def596e9 | 2111 | do { |
a3a0e43f JA |
2112 | /* |
2113 | * Don't enter poll loop if we already have events pending. | |
2114 | * If we do, we can potentially be spinning for commands that | |
2115 | * already triggered a CQE (eg in error). | |
2116 | */ | |
1d7bb1d5 | 2117 | if (io_cqring_events(ctx, false)) |
a3a0e43f JA |
2118 | break; |
2119 | ||
500f9fba JA |
2120 | /* |
2121 | * If a submit got punted to a workqueue, we can have the | |
2122 | * application entering polling for a command before it gets | |
2123 | * issued. That app will hold the uring_lock for the duration | |
2124 | * of the poll right here, so we need to take a breather every | |
2125 | * now and then to ensure that the issue has a chance to add | |
2126 | * the poll to the issued list. Otherwise we can spin here | |
2127 | * forever, while the workqueue is stuck trying to acquire the | |
2128 | * very same mutex. | |
2129 | */ | |
2130 | if (!(++iters & 7)) { | |
2131 | mutex_unlock(&ctx->uring_lock); | |
4c6e277c | 2132 | io_run_task_work(); |
500f9fba JA |
2133 | mutex_lock(&ctx->uring_lock); |
2134 | } | |
2135 | ||
7668b92a | 2136 | ret = io_iopoll_getevents(ctx, &nr_events, min); |
def596e9 JA |
2137 | if (ret <= 0) |
2138 | break; | |
2139 | ret = 0; | |
7668b92a | 2140 | } while (min && !nr_events && !need_resched()); |
def596e9 | 2141 | |
500f9fba | 2142 | mutex_unlock(&ctx->uring_lock); |
def596e9 JA |
2143 | return ret; |
2144 | } | |
2145 | ||
491381ce | 2146 | static void kiocb_end_write(struct io_kiocb *req) |
2b188cc1 | 2147 | { |
491381ce JA |
2148 | /* |
2149 | * Tell lockdep we inherited freeze protection from submission | |
2150 | * thread. | |
2151 | */ | |
2152 | if (req->flags & REQ_F_ISREG) { | |
2153 | struct inode *inode = file_inode(req->file); | |
2b188cc1 | 2154 | |
491381ce | 2155 | __sb_writers_acquired(inode->i_sb, SB_FREEZE_WRITE); |
2b188cc1 | 2156 | } |
491381ce | 2157 | file_end_write(req->file); |
2b188cc1 JA |
2158 | } |
2159 | ||
a1d7c393 JA |
2160 | static void io_complete_rw_common(struct kiocb *kiocb, long res, |
2161 | struct io_comp_state *cs) | |
2b188cc1 | 2162 | { |
9adbd45d | 2163 | struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb); |
bcda7baa | 2164 | int cflags = 0; |
2b188cc1 | 2165 | |
491381ce JA |
2166 | if (kiocb->ki_flags & IOCB_WRITE) |
2167 | kiocb_end_write(req); | |
2b188cc1 | 2168 | |
4e88d6e7 JA |
2169 | if (res != req->result) |
2170 | req_set_fail_links(req); | |
bcda7baa JA |
2171 | if (req->flags & REQ_F_BUFFER_SELECTED) |
2172 | cflags = io_put_kbuf(req); | |
a1d7c393 | 2173 | __io_req_complete(req, res, cflags, cs); |
ba816ad6 JA |
2174 | } |
2175 | ||
b63534c4 JA |
2176 | #ifdef CONFIG_BLOCK |
2177 | static bool io_resubmit_prep(struct io_kiocb *req, int error) | |
2178 | { | |
2179 | struct iovec inline_vecs[UIO_FASTIOV], *iovec = inline_vecs; | |
2180 | ssize_t ret = -ECANCELED; | |
2181 | struct iov_iter iter; | |
2182 | int rw; | |
2183 | ||
2184 | if (error) { | |
2185 | ret = error; | |
2186 | goto end_req; | |
2187 | } | |
2188 | ||
2189 | switch (req->opcode) { | |
2190 | case IORING_OP_READV: | |
2191 | case IORING_OP_READ_FIXED: | |
2192 | case IORING_OP_READ: | |
2193 | rw = READ; | |
2194 | break; | |
2195 | case IORING_OP_WRITEV: | |
2196 | case IORING_OP_WRITE_FIXED: | |
2197 | case IORING_OP_WRITE: | |
2198 | rw = WRITE; | |
2199 | break; | |
2200 | default: | |
2201 | printk_once(KERN_WARNING "io_uring: bad opcode in resubmit %d\n", | |
2202 | req->opcode); | |
2203 | goto end_req; | |
2204 | } | |
2205 | ||
2206 | ret = io_import_iovec(rw, req, &iovec, &iter, false); | |
2207 | if (ret < 0) | |
2208 | goto end_req; | |
2209 | ret = io_setup_async_rw(req, ret, iovec, inline_vecs, &iter); | |
2210 | if (!ret) | |
2211 | return true; | |
2212 | kfree(iovec); | |
2213 | end_req: | |
b63534c4 | 2214 | req_set_fail_links(req); |
e1e16097 | 2215 | io_req_complete(req, ret); |
b63534c4 JA |
2216 | return false; |
2217 | } | |
2218 | ||
2219 | static void io_rw_resubmit(struct callback_head *cb) | |
2220 | { | |
2221 | struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work); | |
2222 | struct io_ring_ctx *ctx = req->ctx; | |
2223 | int err; | |
2224 | ||
b63534c4 JA |
2225 | err = io_sq_thread_acquire_mm(ctx, req); |
2226 | ||
2227 | if (io_resubmit_prep(req, err)) { | |
2228 | refcount_inc(&req->refs); | |
2229 | io_queue_async_work(req); | |
2230 | } | |
2231 | } | |
2232 | #endif | |
2233 | ||
2234 | static bool io_rw_reissue(struct io_kiocb *req, long res) | |
2235 | { | |
2236 | #ifdef CONFIG_BLOCK | |
b63534c4 JA |
2237 | int ret; |
2238 | ||
2239 | if ((res != -EAGAIN && res != -EOPNOTSUPP) || io_wq_current_is_worker()) | |
2240 | return false; | |
2241 | ||
b63534c4 | 2242 | init_task_work(&req->task_work, io_rw_resubmit); |
c2c4c83c JA |
2243 | ret = io_req_task_work_add(req, &req->task_work); |
2244 | if (!ret) | |
b63534c4 JA |
2245 | return true; |
2246 | #endif | |
2247 | return false; | |
2248 | } | |
2249 | ||
a1d7c393 JA |
2250 | static void __io_complete_rw(struct io_kiocb *req, long res, long res2, |
2251 | struct io_comp_state *cs) | |
2252 | { | |
2253 | if (!io_rw_reissue(req, res)) | |
2254 | io_complete_rw_common(&req->rw.kiocb, res, cs); | |
2255 | } | |
2256 | ||
ba816ad6 JA |
2257 | static void io_complete_rw(struct kiocb *kiocb, long res, long res2) |
2258 | { | |
9adbd45d | 2259 | struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb); |
ba816ad6 | 2260 | |
a1d7c393 | 2261 | __io_complete_rw(req, res, res2, NULL); |
2b188cc1 JA |
2262 | } |
2263 | ||
def596e9 JA |
2264 | static void io_complete_rw_iopoll(struct kiocb *kiocb, long res, long res2) |
2265 | { | |
9adbd45d | 2266 | struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb); |
def596e9 | 2267 | |
491381ce JA |
2268 | if (kiocb->ki_flags & IOCB_WRITE) |
2269 | kiocb_end_write(req); | |
def596e9 | 2270 | |
2d7d6792 | 2271 | if (res != -EAGAIN && res != req->result) |
4e88d6e7 | 2272 | req_set_fail_links(req); |
bbde017a XW |
2273 | |
2274 | WRITE_ONCE(req->result, res); | |
2275 | /* order with io_poll_complete() checking ->result */ | |
cd664b0e PB |
2276 | smp_wmb(); |
2277 | WRITE_ONCE(req->iopoll_completed, 1); | |
def596e9 JA |
2278 | } |
2279 | ||
2280 | /* | |
2281 | * After the iocb has been issued, it's safe to be found on the poll list. | |
2282 | * Adding the kiocb to the list AFTER submission ensures that we don't | |
2283 | * find it from a io_iopoll_getevents() thread before the issuer is done | |
2284 | * accessing the kiocb cookie. | |
2285 | */ | |
2286 | static void io_iopoll_req_issued(struct io_kiocb *req) | |
2287 | { | |
2288 | struct io_ring_ctx *ctx = req->ctx; | |
2289 | ||
2290 | /* | |
2291 | * Track whether we have multiple files in our lists. This will impact | |
2292 | * how we do polling eventually, not spinning if we're on potentially | |
2293 | * different devices. | |
2294 | */ | |
540e32a0 | 2295 | if (list_empty(&ctx->iopoll_list)) { |
def596e9 JA |
2296 | ctx->poll_multi_file = false; |
2297 | } else if (!ctx->poll_multi_file) { | |
2298 | struct io_kiocb *list_req; | |
2299 | ||
540e32a0 | 2300 | list_req = list_first_entry(&ctx->iopoll_list, struct io_kiocb, |
d21ffe7e | 2301 | inflight_entry); |
9adbd45d | 2302 | if (list_req->file != req->file) |
def596e9 JA |
2303 | ctx->poll_multi_file = true; |
2304 | } | |
2305 | ||
2306 | /* | |
2307 | * For fast devices, IO may have already completed. If it has, add | |
2308 | * it to the front so we find it first. | |
2309 | */ | |
65a6543d | 2310 | if (READ_ONCE(req->iopoll_completed)) |
d21ffe7e | 2311 | list_add(&req->inflight_entry, &ctx->iopoll_list); |
def596e9 | 2312 | else |
d21ffe7e | 2313 | list_add_tail(&req->inflight_entry, &ctx->iopoll_list); |
bdcd3eab XW |
2314 | |
2315 | if ((ctx->flags & IORING_SETUP_SQPOLL) && | |
2316 | wq_has_sleeper(&ctx->sqo_wait)) | |
2317 | wake_up(&ctx->sqo_wait); | |
def596e9 JA |
2318 | } |
2319 | ||
9f13c35b | 2320 | static void __io_state_file_put(struct io_submit_state *state) |
9a56a232 | 2321 | { |
9f13c35b | 2322 | int diff = state->has_refs - state->used_refs; |
9a56a232 | 2323 | |
9f13c35b PB |
2324 | if (diff) |
2325 | fput_many(state->file, diff); | |
2326 | state->file = NULL; | |
2327 | } | |
2328 | ||
2329 | static inline void io_state_file_put(struct io_submit_state *state) | |
2330 | { | |
2331 | if (state->file) | |
2332 | __io_state_file_put(state); | |
9a56a232 JA |
2333 | } |
2334 | ||
2335 | /* | |
2336 | * Get as many references to a file as we have IOs left in this submission, | |
2337 | * assuming most submissions are for one file, or at least that each file | |
2338 | * has more than one submission. | |
2339 | */ | |
8da11c19 | 2340 | static struct file *__io_file_get(struct io_submit_state *state, int fd) |
9a56a232 JA |
2341 | { |
2342 | if (!state) | |
2343 | return fget(fd); | |
2344 | ||
2345 | if (state->file) { | |
2346 | if (state->fd == fd) { | |
2347 | state->used_refs++; | |
2348 | state->ios_left--; | |
2349 | return state->file; | |
2350 | } | |
9f13c35b | 2351 | __io_state_file_put(state); |
9a56a232 JA |
2352 | } |
2353 | state->file = fget_many(fd, state->ios_left); | |
2354 | if (!state->file) | |
2355 | return NULL; | |
2356 | ||
2357 | state->fd = fd; | |
2358 | state->has_refs = state->ios_left; | |
2359 | state->used_refs = 1; | |
2360 | state->ios_left--; | |
2361 | return state->file; | |
2362 | } | |
2363 | ||
4503b767 JA |
2364 | static bool io_bdev_nowait(struct block_device *bdev) |
2365 | { | |
2366 | #ifdef CONFIG_BLOCK | |
2367 | return !bdev || queue_is_mq(bdev_get_queue(bdev)); | |
2368 | #else | |
2369 | return true; | |
2370 | #endif | |
2371 | } | |
2372 | ||
2b188cc1 JA |
2373 | /* |
2374 | * If we tracked the file through the SCM inflight mechanism, we could support | |
2375 | * any file. For now, just ensure that anything potentially problematic is done | |
2376 | * inline. | |
2377 | */ | |
af197f50 | 2378 | static bool io_file_supports_async(struct file *file, int rw) |
2b188cc1 JA |
2379 | { |
2380 | umode_t mode = file_inode(file)->i_mode; | |
2381 | ||
4503b767 JA |
2382 | if (S_ISBLK(mode)) { |
2383 | if (io_bdev_nowait(file->f_inode->i_bdev)) | |
2384 | return true; | |
2385 | return false; | |
2386 | } | |
2387 | if (S_ISCHR(mode) || S_ISSOCK(mode)) | |
2b188cc1 | 2388 | return true; |
4503b767 JA |
2389 | if (S_ISREG(mode)) { |
2390 | if (io_bdev_nowait(file->f_inode->i_sb->s_bdev) && | |
2391 | file->f_op != &io_uring_fops) | |
2392 | return true; | |
2393 | return false; | |
2394 | } | |
2b188cc1 | 2395 | |
c5b85625 JA |
2396 | /* any ->read/write should understand O_NONBLOCK */ |
2397 | if (file->f_flags & O_NONBLOCK) | |
2398 | return true; | |
2399 | ||
af197f50 JA |
2400 | if (!(file->f_mode & FMODE_NOWAIT)) |
2401 | return false; | |
2402 | ||
2403 | if (rw == READ) | |
2404 | return file->f_op->read_iter != NULL; | |
2405 | ||
2406 | return file->f_op->write_iter != NULL; | |
2b188cc1 JA |
2407 | } |
2408 | ||
3529d8c2 JA |
2409 | static int io_prep_rw(struct io_kiocb *req, const struct io_uring_sqe *sqe, |
2410 | bool force_nonblock) | |
2b188cc1 | 2411 | { |
def596e9 | 2412 | struct io_ring_ctx *ctx = req->ctx; |
9adbd45d | 2413 | struct kiocb *kiocb = &req->rw.kiocb; |
09bb8394 JA |
2414 | unsigned ioprio; |
2415 | int ret; | |
2b188cc1 | 2416 | |
491381ce JA |
2417 | if (S_ISREG(file_inode(req->file)->i_mode)) |
2418 | req->flags |= REQ_F_ISREG; | |
2419 | ||
2b188cc1 | 2420 | kiocb->ki_pos = READ_ONCE(sqe->off); |
ba04291e JA |
2421 | if (kiocb->ki_pos == -1 && !(req->file->f_mode & FMODE_STREAM)) { |
2422 | req->flags |= REQ_F_CUR_POS; | |
2423 | kiocb->ki_pos = req->file->f_pos; | |
2424 | } | |
2b188cc1 | 2425 | kiocb->ki_hint = ki_hint_validate(file_write_hint(kiocb->ki_filp)); |
3e577dcd PB |
2426 | kiocb->ki_flags = iocb_flags(kiocb->ki_filp); |
2427 | ret = kiocb_set_rw_flags(kiocb, READ_ONCE(sqe->rw_flags)); | |
2428 | if (unlikely(ret)) | |
2429 | return ret; | |
2b188cc1 JA |
2430 | |
2431 | ioprio = READ_ONCE(sqe->ioprio); | |
2432 | if (ioprio) { | |
2433 | ret = ioprio_check_cap(ioprio); | |
2434 | if (ret) | |
09bb8394 | 2435 | return ret; |
2b188cc1 JA |
2436 | |
2437 | kiocb->ki_ioprio = ioprio; | |
2438 | } else | |
2439 | kiocb->ki_ioprio = get_current_ioprio(); | |
2440 | ||
8449eeda | 2441 | /* don't allow async punt if RWF_NOWAIT was requested */ |
c5b85625 | 2442 | if (kiocb->ki_flags & IOCB_NOWAIT) |
8449eeda SB |
2443 | req->flags |= REQ_F_NOWAIT; |
2444 | ||
b63534c4 JA |
2445 | if (kiocb->ki_flags & IOCB_DIRECT) |
2446 | io_get_req_task(req); | |
2447 | ||
8449eeda | 2448 | if (force_nonblock) |
2b188cc1 | 2449 | kiocb->ki_flags |= IOCB_NOWAIT; |
8449eeda | 2450 | |
def596e9 | 2451 | if (ctx->flags & IORING_SETUP_IOPOLL) { |
def596e9 JA |
2452 | if (!(kiocb->ki_flags & IOCB_DIRECT) || |
2453 | !kiocb->ki_filp->f_op->iopoll) | |
09bb8394 | 2454 | return -EOPNOTSUPP; |
2b188cc1 | 2455 | |
def596e9 JA |
2456 | kiocb->ki_flags |= IOCB_HIPRI; |
2457 | kiocb->ki_complete = io_complete_rw_iopoll; | |
65a6543d | 2458 | req->iopoll_completed = 0; |
f3a6fa22 | 2459 | io_get_req_task(req); |
def596e9 | 2460 | } else { |
09bb8394 JA |
2461 | if (kiocb->ki_flags & IOCB_HIPRI) |
2462 | return -EINVAL; | |
def596e9 JA |
2463 | kiocb->ki_complete = io_complete_rw; |
2464 | } | |
9adbd45d | 2465 | |
3529d8c2 JA |
2466 | req->rw.addr = READ_ONCE(sqe->addr); |
2467 | req->rw.len = READ_ONCE(sqe->len); | |
4f4eeba8 | 2468 | req->buf_index = READ_ONCE(sqe->buf_index); |
2b188cc1 | 2469 | return 0; |
2b188cc1 JA |
2470 | } |
2471 | ||
2472 | static inline void io_rw_done(struct kiocb *kiocb, ssize_t ret) | |
2473 | { | |
2474 | switch (ret) { | |
2475 | case -EIOCBQUEUED: | |
2476 | break; | |
2477 | case -ERESTARTSYS: | |
2478 | case -ERESTARTNOINTR: | |
2479 | case -ERESTARTNOHAND: | |
2480 | case -ERESTART_RESTARTBLOCK: | |
2481 | /* | |
2482 | * We can't just restart the syscall, since previously | |
2483 | * submitted sqes may already be in progress. Just fail this | |
2484 | * IO with EINTR. | |
2485 | */ | |
2486 | ret = -EINTR; | |
2487 | /* fall through */ | |
2488 | default: | |
2489 | kiocb->ki_complete(kiocb, ret, 0); | |
2490 | } | |
2491 | } | |
2492 | ||
a1d7c393 JA |
2493 | static void kiocb_done(struct kiocb *kiocb, ssize_t ret, |
2494 | struct io_comp_state *cs) | |
ba816ad6 | 2495 | { |
ba04291e JA |
2496 | struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb); |
2497 | ||
2498 | if (req->flags & REQ_F_CUR_POS) | |
2499 | req->file->f_pos = kiocb->ki_pos; | |
bcaec089 | 2500 | if (ret >= 0 && kiocb->ki_complete == io_complete_rw) |
a1d7c393 | 2501 | __io_complete_rw(req, ret, 0, cs); |
ba816ad6 JA |
2502 | else |
2503 | io_rw_done(kiocb, ret); | |
2504 | } | |
2505 | ||
9adbd45d | 2506 | static ssize_t io_import_fixed(struct io_kiocb *req, int rw, |
7d009165 | 2507 | struct iov_iter *iter) |
edafccee | 2508 | { |
9adbd45d JA |
2509 | struct io_ring_ctx *ctx = req->ctx; |
2510 | size_t len = req->rw.len; | |
edafccee | 2511 | struct io_mapped_ubuf *imu; |
4f4eeba8 | 2512 | u16 index, buf_index; |
edafccee JA |
2513 | size_t offset; |
2514 | u64 buf_addr; | |
2515 | ||
2516 | /* attempt to use fixed buffers without having provided iovecs */ | |
2517 | if (unlikely(!ctx->user_bufs)) | |
2518 | return -EFAULT; | |
2519 | ||
4f4eeba8 | 2520 | buf_index = req->buf_index; |
edafccee JA |
2521 | if (unlikely(buf_index >= ctx->nr_user_bufs)) |
2522 | return -EFAULT; | |
2523 | ||
2524 | index = array_index_nospec(buf_index, ctx->nr_user_bufs); | |
2525 | imu = &ctx->user_bufs[index]; | |
9adbd45d | 2526 | buf_addr = req->rw.addr; |
edafccee JA |
2527 | |
2528 | /* overflow */ | |
2529 | if (buf_addr + len < buf_addr) | |
2530 | return -EFAULT; | |
2531 | /* not inside the mapped region */ | |
2532 | if (buf_addr < imu->ubuf || buf_addr + len > imu->ubuf + imu->len) | |
2533 | return -EFAULT; | |
2534 | ||
2535 | /* | |
2536 | * May not be a start of buffer, set size appropriately | |
2537 | * and advance us to the beginning. | |
2538 | */ | |
2539 | offset = buf_addr - imu->ubuf; | |
2540 | iov_iter_bvec(iter, rw, imu->bvec, imu->nr_bvecs, offset + len); | |
bd11b3a3 JA |
2541 | |
2542 | if (offset) { | |
2543 | /* | |
2544 | * Don't use iov_iter_advance() here, as it's really slow for | |
2545 | * using the latter parts of a big fixed buffer - it iterates | |
2546 | * over each segment manually. We can cheat a bit here, because | |
2547 | * we know that: | |
2548 | * | |
2549 | * 1) it's a BVEC iter, we set it up | |
2550 | * 2) all bvecs are PAGE_SIZE in size, except potentially the | |
2551 | * first and last bvec | |
2552 | * | |
2553 | * So just find our index, and adjust the iterator afterwards. | |
2554 | * If the offset is within the first bvec (or the whole first | |
2555 | * bvec, just use iov_iter_advance(). This makes it easier | |
2556 | * since we can just skip the first segment, which may not | |
2557 | * be PAGE_SIZE aligned. | |
2558 | */ | |
2559 | const struct bio_vec *bvec = imu->bvec; | |
2560 | ||
2561 | if (offset <= bvec->bv_len) { | |
2562 | iov_iter_advance(iter, offset); | |
2563 | } else { | |
2564 | unsigned long seg_skip; | |
2565 | ||
2566 | /* skip first vec */ | |
2567 | offset -= bvec->bv_len; | |
2568 | seg_skip = 1 + (offset >> PAGE_SHIFT); | |
2569 | ||
2570 | iter->bvec = bvec + seg_skip; | |
2571 | iter->nr_segs -= seg_skip; | |
99c79f66 | 2572 | iter->count -= bvec->bv_len + offset; |
bd11b3a3 | 2573 | iter->iov_offset = offset & ~PAGE_MASK; |
bd11b3a3 JA |
2574 | } |
2575 | } | |
2576 | ||
5e559561 | 2577 | return len; |
edafccee JA |
2578 | } |
2579 | ||
bcda7baa JA |
2580 | static void io_ring_submit_unlock(struct io_ring_ctx *ctx, bool needs_lock) |
2581 | { | |
2582 | if (needs_lock) | |
2583 | mutex_unlock(&ctx->uring_lock); | |
2584 | } | |
2585 | ||
2586 | static void io_ring_submit_lock(struct io_ring_ctx *ctx, bool needs_lock) | |
2587 | { | |
2588 | /* | |
2589 | * "Normal" inline submissions always hold the uring_lock, since we | |
2590 | * grab it from the system call. Same is true for the SQPOLL offload. | |
2591 | * The only exception is when we've detached the request and issue it | |
2592 | * from an async worker thread, grab the lock for that case. | |
2593 | */ | |
2594 | if (needs_lock) | |
2595 | mutex_lock(&ctx->uring_lock); | |
2596 | } | |
2597 | ||
2598 | static struct io_buffer *io_buffer_select(struct io_kiocb *req, size_t *len, | |
2599 | int bgid, struct io_buffer *kbuf, | |
2600 | bool needs_lock) | |
2601 | { | |
2602 | struct io_buffer *head; | |
2603 | ||
2604 | if (req->flags & REQ_F_BUFFER_SELECTED) | |
2605 | return kbuf; | |
2606 | ||
2607 | io_ring_submit_lock(req->ctx, needs_lock); | |
2608 | ||
2609 | lockdep_assert_held(&req->ctx->uring_lock); | |
2610 | ||
2611 | head = idr_find(&req->ctx->io_buffer_idr, bgid); | |
2612 | if (head) { | |
2613 | if (!list_empty(&head->list)) { | |
2614 | kbuf = list_last_entry(&head->list, struct io_buffer, | |
2615 | list); | |
2616 | list_del(&kbuf->list); | |
2617 | } else { | |
2618 | kbuf = head; | |
2619 | idr_remove(&req->ctx->io_buffer_idr, bgid); | |
2620 | } | |
2621 | if (*len > kbuf->len) | |
2622 | *len = kbuf->len; | |
2623 | } else { | |
2624 | kbuf = ERR_PTR(-ENOBUFS); | |
2625 | } | |
2626 | ||
2627 | io_ring_submit_unlock(req->ctx, needs_lock); | |
2628 | ||
2629 | return kbuf; | |
2630 | } | |
2631 | ||
4d954c25 JA |
2632 | static void __user *io_rw_buffer_select(struct io_kiocb *req, size_t *len, |
2633 | bool needs_lock) | |
2634 | { | |
2635 | struct io_buffer *kbuf; | |
4f4eeba8 | 2636 | u16 bgid; |
4d954c25 JA |
2637 | |
2638 | kbuf = (struct io_buffer *) (unsigned long) req->rw.addr; | |
4f4eeba8 | 2639 | bgid = req->buf_index; |
4d954c25 JA |
2640 | kbuf = io_buffer_select(req, len, bgid, kbuf, needs_lock); |
2641 | if (IS_ERR(kbuf)) | |
2642 | return kbuf; | |
2643 | req->rw.addr = (u64) (unsigned long) kbuf; | |
2644 | req->flags |= REQ_F_BUFFER_SELECTED; | |
2645 | return u64_to_user_ptr(kbuf->addr); | |
2646 | } | |
2647 | ||
2648 | #ifdef CONFIG_COMPAT | |
2649 | static ssize_t io_compat_import(struct io_kiocb *req, struct iovec *iov, | |
2650 | bool needs_lock) | |
2651 | { | |
2652 | struct compat_iovec __user *uiov; | |
2653 | compat_ssize_t clen; | |
2654 | void __user *buf; | |
2655 | ssize_t len; | |
2656 | ||
2657 | uiov = u64_to_user_ptr(req->rw.addr); | |
2658 | if (!access_ok(uiov, sizeof(*uiov))) | |
2659 | return -EFAULT; | |
2660 | if (__get_user(clen, &uiov->iov_len)) | |
2661 | return -EFAULT; | |
2662 | if (clen < 0) | |
2663 | return -EINVAL; | |
2664 | ||
2665 | len = clen; | |
2666 | buf = io_rw_buffer_select(req, &len, needs_lock); | |
2667 | if (IS_ERR(buf)) | |
2668 | return PTR_ERR(buf); | |
2669 | iov[0].iov_base = buf; | |
2670 | iov[0].iov_len = (compat_size_t) len; | |
2671 | return 0; | |
2672 | } | |
2673 | #endif | |
2674 | ||
2675 | static ssize_t __io_iov_buffer_select(struct io_kiocb *req, struct iovec *iov, | |
2676 | bool needs_lock) | |
2677 | { | |
2678 | struct iovec __user *uiov = u64_to_user_ptr(req->rw.addr); | |
2679 | void __user *buf; | |
2680 | ssize_t len; | |
2681 | ||
2682 | if (copy_from_user(iov, uiov, sizeof(*uiov))) | |
2683 | return -EFAULT; | |
2684 | ||
2685 | len = iov[0].iov_len; | |
2686 | if (len < 0) | |
2687 | return -EINVAL; | |
2688 | buf = io_rw_buffer_select(req, &len, needs_lock); | |
2689 | if (IS_ERR(buf)) | |
2690 | return PTR_ERR(buf); | |
2691 | iov[0].iov_base = buf; | |
2692 | iov[0].iov_len = len; | |
2693 | return 0; | |
2694 | } | |
2695 | ||
2696 | static ssize_t io_iov_buffer_select(struct io_kiocb *req, struct iovec *iov, | |
2697 | bool needs_lock) | |
2698 | { | |
dddb3e26 JA |
2699 | if (req->flags & REQ_F_BUFFER_SELECTED) { |
2700 | struct io_buffer *kbuf; | |
2701 | ||
2702 | kbuf = (struct io_buffer *) (unsigned long) req->rw.addr; | |
2703 | iov[0].iov_base = u64_to_user_ptr(kbuf->addr); | |
2704 | iov[0].iov_len = kbuf->len; | |
4d954c25 | 2705 | return 0; |
dddb3e26 | 2706 | } |
4d954c25 JA |
2707 | if (!req->rw.len) |
2708 | return 0; | |
2709 | else if (req->rw.len > 1) | |
2710 | return -EINVAL; | |
2711 | ||
2712 | #ifdef CONFIG_COMPAT | |
2713 | if (req->ctx->compat) | |
2714 | return io_compat_import(req, iov, needs_lock); | |
2715 | #endif | |
2716 | ||
2717 | return __io_iov_buffer_select(req, iov, needs_lock); | |
2718 | } | |
2719 | ||
cf6fd4bd | 2720 | static ssize_t io_import_iovec(int rw, struct io_kiocb *req, |
bcda7baa JA |
2721 | struct iovec **iovec, struct iov_iter *iter, |
2722 | bool needs_lock) | |
2b188cc1 | 2723 | { |
9adbd45d JA |
2724 | void __user *buf = u64_to_user_ptr(req->rw.addr); |
2725 | size_t sqe_len = req->rw.len; | |
4d954c25 | 2726 | ssize_t ret; |
edafccee JA |
2727 | u8 opcode; |
2728 | ||
d625c6ee | 2729 | opcode = req->opcode; |
7d009165 | 2730 | if (opcode == IORING_OP_READ_FIXED || opcode == IORING_OP_WRITE_FIXED) { |
edafccee | 2731 | *iovec = NULL; |
9adbd45d | 2732 | return io_import_fixed(req, rw, iter); |
edafccee | 2733 | } |
2b188cc1 | 2734 | |
bcda7baa | 2735 | /* buffer index only valid with fixed read/write, or buffer select */ |
4f4eeba8 | 2736 | if (req->buf_index && !(req->flags & REQ_F_BUFFER_SELECT)) |
9adbd45d JA |
2737 | return -EINVAL; |
2738 | ||
3a6820f2 | 2739 | if (opcode == IORING_OP_READ || opcode == IORING_OP_WRITE) { |
bcda7baa | 2740 | if (req->flags & REQ_F_BUFFER_SELECT) { |
4d954c25 JA |
2741 | buf = io_rw_buffer_select(req, &sqe_len, needs_lock); |
2742 | if (IS_ERR(buf)) { | |
bcda7baa | 2743 | *iovec = NULL; |
4d954c25 | 2744 | return PTR_ERR(buf); |
bcda7baa | 2745 | } |
3f9d6441 | 2746 | req->rw.len = sqe_len; |
bcda7baa JA |
2747 | } |
2748 | ||
3a6820f2 JA |
2749 | ret = import_single_range(rw, buf, sqe_len, *iovec, iter); |
2750 | *iovec = NULL; | |
3a901598 | 2751 | return ret < 0 ? ret : sqe_len; |
3a6820f2 JA |
2752 | } |
2753 | ||
f67676d1 JA |
2754 | if (req->io) { |
2755 | struct io_async_rw *iorw = &req->io->rw; | |
2756 | ||
252917c3 PB |
2757 | iov_iter_init(iter, rw, iorw->iov, iorw->nr_segs, iorw->size); |
2758 | *iovec = NULL; | |
f67676d1 JA |
2759 | return iorw->size; |
2760 | } | |
2761 | ||
4d954c25 JA |
2762 | if (req->flags & REQ_F_BUFFER_SELECT) { |
2763 | ret = io_iov_buffer_select(req, *iovec, needs_lock); | |
3f9d6441 JA |
2764 | if (!ret) { |
2765 | ret = (*iovec)->iov_len; | |
2766 | iov_iter_init(iter, rw, *iovec, 1, ret); | |
2767 | } | |
4d954c25 JA |
2768 | *iovec = NULL; |
2769 | return ret; | |
2770 | } | |
2771 | ||
2b188cc1 | 2772 | #ifdef CONFIG_COMPAT |
cf6fd4bd | 2773 | if (req->ctx->compat) |
2b188cc1 JA |
2774 | return compat_import_iovec(rw, buf, sqe_len, UIO_FASTIOV, |
2775 | iovec, iter); | |
2776 | #endif | |
2777 | ||
2778 | return import_iovec(rw, buf, sqe_len, UIO_FASTIOV, iovec, iter); | |
2779 | } | |
2780 | ||
31b51510 | 2781 | /* |
32960613 JA |
2782 | * For files that don't have ->read_iter() and ->write_iter(), handle them |
2783 | * by looping over ->read() or ->write() manually. | |
31b51510 | 2784 | */ |
32960613 JA |
2785 | static ssize_t loop_rw_iter(int rw, struct file *file, struct kiocb *kiocb, |
2786 | struct iov_iter *iter) | |
2787 | { | |
2788 | ssize_t ret = 0; | |
2789 | ||
2790 | /* | |
2791 | * Don't support polled IO through this interface, and we can't | |
2792 | * support non-blocking either. For the latter, this just causes | |
2793 | * the kiocb to be handled from an async context. | |
2794 | */ | |
2795 | if (kiocb->ki_flags & IOCB_HIPRI) | |
2796 | return -EOPNOTSUPP; | |
2797 | if (kiocb->ki_flags & IOCB_NOWAIT) | |
2798 | return -EAGAIN; | |
2799 | ||
2800 | while (iov_iter_count(iter)) { | |
311ae9e1 | 2801 | struct iovec iovec; |
32960613 JA |
2802 | ssize_t nr; |
2803 | ||
311ae9e1 PB |
2804 | if (!iov_iter_is_bvec(iter)) { |
2805 | iovec = iov_iter_iovec(iter); | |
2806 | } else { | |
2807 | /* fixed buffers import bvec */ | |
2808 | iovec.iov_base = kmap(iter->bvec->bv_page) | |
2809 | + iter->iov_offset; | |
2810 | iovec.iov_len = min(iter->count, | |
2811 | iter->bvec->bv_len - iter->iov_offset); | |
2812 | } | |
2813 | ||
32960613 JA |
2814 | if (rw == READ) { |
2815 | nr = file->f_op->read(file, iovec.iov_base, | |
2816 | iovec.iov_len, &kiocb->ki_pos); | |
2817 | } else { | |
2818 | nr = file->f_op->write(file, iovec.iov_base, | |
2819 | iovec.iov_len, &kiocb->ki_pos); | |
2820 | } | |
2821 | ||
311ae9e1 PB |
2822 | if (iov_iter_is_bvec(iter)) |
2823 | kunmap(iter->bvec->bv_page); | |
2824 | ||
32960613 JA |
2825 | if (nr < 0) { |
2826 | if (!ret) | |
2827 | ret = nr; | |
2828 | break; | |
2829 | } | |
2830 | ret += nr; | |
2831 | if (nr != iovec.iov_len) | |
2832 | break; | |
2833 | iov_iter_advance(iter, nr); | |
2834 | } | |
2835 | ||
2836 | return ret; | |
2837 | } | |
2838 | ||
b7bb4f7d | 2839 | static void io_req_map_rw(struct io_kiocb *req, ssize_t io_size, |
f67676d1 JA |
2840 | struct iovec *iovec, struct iovec *fast_iov, |
2841 | struct iov_iter *iter) | |
2842 | { | |
b64e3444 PB |
2843 | struct io_async_rw *rw = &req->io->rw; |
2844 | ||
2845 | rw->nr_segs = iter->nr_segs; | |
2846 | rw->size = io_size; | |
2847 | if (!iovec) { | |
2848 | rw->iov = rw->fast_iov; | |
2849 | if (rw->iov != fast_iov) | |
2850 | memcpy(rw->iov, fast_iov, | |
45097dae | 2851 | sizeof(struct iovec) * iter->nr_segs); |
99bc4c38 | 2852 | } else { |
b64e3444 | 2853 | rw->iov = iovec; |
99bc4c38 | 2854 | req->flags |= REQ_F_NEED_CLEANUP; |
f67676d1 JA |
2855 | } |
2856 | } | |
2857 | ||
3d9932a8 XW |
2858 | static inline int __io_alloc_async_ctx(struct io_kiocb *req) |
2859 | { | |
2860 | req->io = kmalloc(sizeof(*req->io), GFP_KERNEL); | |
2861 | return req->io == NULL; | |
2862 | } | |
2863 | ||
b7bb4f7d | 2864 | static int io_alloc_async_ctx(struct io_kiocb *req) |
f67676d1 | 2865 | { |
d3656344 JA |
2866 | if (!io_op_defs[req->opcode].async_ctx) |
2867 | return 0; | |
3d9932a8 XW |
2868 | |
2869 | return __io_alloc_async_ctx(req); | |
b7bb4f7d JA |
2870 | } |
2871 | ||
b7bb4f7d JA |
2872 | static int io_setup_async_rw(struct io_kiocb *req, ssize_t io_size, |
2873 | struct iovec *iovec, struct iovec *fast_iov, | |
2874 | struct iov_iter *iter) | |
2875 | { | |
980ad263 | 2876 | if (!io_op_defs[req->opcode].async_ctx) |
74566df3 | 2877 | return 0; |
5d204bcf | 2878 | if (!req->io) { |
3d9932a8 | 2879 | if (__io_alloc_async_ctx(req)) |
5d204bcf | 2880 | return -ENOMEM; |
b7bb4f7d | 2881 | |
5d204bcf JA |
2882 | io_req_map_rw(req, io_size, iovec, fast_iov, iter); |
2883 | } | |
b7bb4f7d | 2884 | return 0; |
f67676d1 JA |
2885 | } |
2886 | ||
c3e330a4 PB |
2887 | static inline int io_rw_prep_async(struct io_kiocb *req, int rw, |
2888 | bool force_nonblock) | |
2889 | { | |
2890 | struct io_async_ctx *io = req->io; | |
2891 | struct iov_iter iter; | |
2892 | ssize_t ret; | |
2893 | ||
2894 | io->rw.iov = io->rw.fast_iov; | |
2895 | req->io = NULL; | |
2896 | ret = io_import_iovec(rw, req, &io->rw.iov, &iter, !force_nonblock); | |
2897 | req->io = io; | |
2898 | if (unlikely(ret < 0)) | |
2899 | return ret; | |
2900 | ||
2901 | io_req_map_rw(req, ret, io->rw.iov, io->rw.fast_iov, &iter); | |
2902 | return 0; | |
2903 | } | |
2904 | ||
3529d8c2 JA |
2905 | static int io_read_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe, |
2906 | bool force_nonblock) | |
f67676d1 JA |
2907 | { |
2908 | ssize_t ret; | |
2909 | ||
3529d8c2 JA |
2910 | ret = io_prep_rw(req, sqe, force_nonblock); |
2911 | if (ret) | |
2912 | return ret; | |
f67676d1 | 2913 | |
3529d8c2 JA |
2914 | if (unlikely(!(req->file->f_mode & FMODE_READ))) |
2915 | return -EBADF; | |
f67676d1 | 2916 | |
5f798bea PB |
2917 | /* either don't need iovec imported or already have it */ |
2918 | if (!req->io || req->flags & REQ_F_NEED_CLEANUP) | |
3529d8c2 | 2919 | return 0; |
c3e330a4 | 2920 | return io_rw_prep_async(req, READ, force_nonblock); |
f67676d1 JA |
2921 | } |
2922 | ||
bcf5a063 JA |
2923 | static int io_async_buf_func(struct wait_queue_entry *wait, unsigned mode, |
2924 | int sync, void *arg) | |
2925 | { | |
2926 | struct wait_page_queue *wpq; | |
2927 | struct io_kiocb *req = wait->private; | |
bcf5a063 | 2928 | struct wait_page_key *key = arg; |
bcf5a063 JA |
2929 | int ret; |
2930 | ||
2931 | wpq = container_of(wait, struct wait_page_queue, wait); | |
2932 | ||
2933 | ret = wake_page_match(wpq, key); | |
2934 | if (ret != 1) | |
2935 | return ret; | |
2936 | ||
2937 | list_del_init(&wait->entry); | |
2938 | ||
e7375122 | 2939 | init_task_work(&req->task_work, io_req_task_submit); |
bcf5a063 JA |
2940 | /* submit ref gets dropped, acquire a new one */ |
2941 | refcount_inc(&req->refs); | |
e7375122 | 2942 | ret = io_req_task_work_add(req, &req->task_work); |
bcf5a063 | 2943 | if (unlikely(ret)) { |
c2c4c83c JA |
2944 | struct task_struct *tsk; |
2945 | ||
bcf5a063 | 2946 | /* queue just for cancelation */ |
e7375122 | 2947 | init_task_work(&req->task_work, io_req_task_cancel); |
bcf5a063 | 2948 | tsk = io_wq_get_task(req->ctx->io_wq); |
e7375122 | 2949 | task_work_add(tsk, &req->task_work, 0); |
c2c4c83c | 2950 | wake_up_process(tsk); |
bcf5a063 | 2951 | } |
bcf5a063 JA |
2952 | return 1; |
2953 | } | |
2954 | ||
2955 | static bool io_rw_should_retry(struct io_kiocb *req) | |
2956 | { | |
2957 | struct kiocb *kiocb = &req->rw.kiocb; | |
2958 | int ret; | |
2959 | ||
2960 | /* never retry for NOWAIT, we just complete with -EAGAIN */ | |
2961 | if (req->flags & REQ_F_NOWAIT) | |
2962 | return false; | |
2963 | ||
2964 | /* already tried, or we're doing O_DIRECT */ | |
2965 | if (kiocb->ki_flags & (IOCB_DIRECT | IOCB_WAITQ)) | |
2966 | return false; | |
2967 | /* | |
2968 | * just use poll if we can, and don't attempt if the fs doesn't | |
2969 | * support callback based unlocks | |
2970 | */ | |
2971 | if (file_can_poll(req->file) || !(req->file->f_mode & FMODE_BUF_RASYNC)) | |
2972 | return false; | |
2973 | ||
2974 | /* | |
2975 | * If request type doesn't require req->io to defer in general, | |
2976 | * we need to allocate it here | |
2977 | */ | |
2978 | if (!req->io && __io_alloc_async_ctx(req)) | |
2979 | return false; | |
2980 | ||
2981 | ret = kiocb_wait_page_queue_init(kiocb, &req->io->rw.wpq, | |
2982 | io_async_buf_func, req); | |
2983 | if (!ret) { | |
2984 | io_get_req_task(req); | |
2985 | return true; | |
2986 | } | |
2987 | ||
2988 | return false; | |
2989 | } | |
2990 | ||
2991 | static int io_iter_do_read(struct io_kiocb *req, struct iov_iter *iter) | |
2992 | { | |
2993 | if (req->file->f_op->read_iter) | |
2994 | return call_read_iter(req->file, &req->rw.kiocb, iter); | |
2995 | return loop_rw_iter(READ, req->file, &req->rw.kiocb, iter); | |
2996 | } | |
2997 | ||
a1d7c393 JA |
2998 | static int io_read(struct io_kiocb *req, bool force_nonblock, |
2999 | struct io_comp_state *cs) | |
2b188cc1 JA |
3000 | { |
3001 | struct iovec inline_vecs[UIO_FASTIOV], *iovec = inline_vecs; | |
9adbd45d | 3002 | struct kiocb *kiocb = &req->rw.kiocb; |
2b188cc1 | 3003 | struct iov_iter iter; |
31b51510 | 3004 | size_t iov_count; |
f67676d1 | 3005 | ssize_t io_size, ret; |
2b188cc1 | 3006 | |
bcda7baa | 3007 | ret = io_import_iovec(READ, req, &iovec, &iter, !force_nonblock); |
06b76d44 JA |
3008 | if (ret < 0) |
3009 | return ret; | |
2b188cc1 | 3010 | |
fd6c2e4c JA |
3011 | /* Ensure we clear previously set non-block flag */ |
3012 | if (!force_nonblock) | |
29de5f6a | 3013 | kiocb->ki_flags &= ~IOCB_NOWAIT; |
fd6c2e4c | 3014 | |
f67676d1 | 3015 | io_size = ret; |
6795c5ab | 3016 | req->result = io_size; |
f67676d1 | 3017 | |
24c74678 | 3018 | /* If the file doesn't support async, just async punt */ |
af197f50 | 3019 | if (force_nonblock && !io_file_supports_async(req->file, READ)) |
f67676d1 | 3020 | goto copy_iov; |
9e645e11 | 3021 | |
31b51510 | 3022 | iov_count = iov_iter_count(&iter); |
9adbd45d | 3023 | ret = rw_verify_area(READ, req->file, &kiocb->ki_pos, iov_count); |
2b188cc1 | 3024 | if (!ret) { |
b63534c4 | 3025 | unsigned long nr_segs = iter.nr_segs; |
4503b767 | 3026 | ssize_t ret2 = 0; |
2b188cc1 | 3027 | |
bcf5a063 | 3028 | ret2 = io_iter_do_read(req, &iter); |
32960613 | 3029 | |
9d93a3f5 | 3030 | /* Catch -EAGAIN return for forced non-blocking submission */ |
4503b767 | 3031 | if (!force_nonblock || (ret2 != -EAGAIN && ret2 != -EIO)) { |
a1d7c393 | 3032 | kiocb_done(kiocb, ret2, cs); |
f67676d1 | 3033 | } else { |
b63534c4 JA |
3034 | iter.count = iov_count; |
3035 | iter.nr_segs = nr_segs; | |
f67676d1 | 3036 | copy_iov: |
b7bb4f7d | 3037 | ret = io_setup_async_rw(req, io_size, iovec, |
f67676d1 JA |
3038 | inline_vecs, &iter); |
3039 | if (ret) | |
3040 | goto out_free; | |
252917c3 PB |
3041 | /* it's copied and will be cleaned with ->io */ |
3042 | iovec = NULL; | |
bcf5a063 JA |
3043 | /* if we can retry, do so with the callbacks armed */ |
3044 | if (io_rw_should_retry(req)) { | |
3045 | ret2 = io_iter_do_read(req, &iter); | |
3046 | if (ret2 == -EIOCBQUEUED) { | |
3047 | goto out_free; | |
3048 | } else if (ret2 != -EAGAIN) { | |
a1d7c393 | 3049 | kiocb_done(kiocb, ret2, cs); |
bcf5a063 JA |
3050 | goto out_free; |
3051 | } | |
3052 | } | |
3053 | kiocb->ki_flags &= ~IOCB_WAITQ; | |
f67676d1 JA |
3054 | return -EAGAIN; |
3055 | } | |
2b188cc1 | 3056 | } |
f67676d1 | 3057 | out_free: |
252917c3 | 3058 | if (iovec) |
6f2cc166 | 3059 | kfree(iovec); |
2b188cc1 JA |
3060 | return ret; |
3061 | } | |
3062 | ||
3529d8c2 JA |
3063 | static int io_write_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe, |
3064 | bool force_nonblock) | |
f67676d1 JA |
3065 | { |
3066 | ssize_t ret; | |
3067 | ||
3529d8c2 JA |
3068 | ret = io_prep_rw(req, sqe, force_nonblock); |
3069 | if (ret) | |
3070 | return ret; | |
f67676d1 | 3071 | |
3529d8c2 JA |
3072 | if (unlikely(!(req->file->f_mode & FMODE_WRITE))) |
3073 | return -EBADF; | |
f67676d1 | 3074 | |
4ed734b0 JA |
3075 | req->fsize = rlimit(RLIMIT_FSIZE); |
3076 | ||
5f798bea PB |
3077 | /* either don't need iovec imported or already have it */ |
3078 | if (!req->io || req->flags & REQ_F_NEED_CLEANUP) | |
3529d8c2 | 3079 | return 0; |
c3e330a4 | 3080 | return io_rw_prep_async(req, WRITE, force_nonblock); |
f67676d1 JA |
3081 | } |
3082 | ||
a1d7c393 JA |
3083 | static int io_write(struct io_kiocb *req, bool force_nonblock, |
3084 | struct io_comp_state *cs) | |
2b188cc1 JA |
3085 | { |
3086 | struct iovec inline_vecs[UIO_FASTIOV], *iovec = inline_vecs; | |
9adbd45d | 3087 | struct kiocb *kiocb = &req->rw.kiocb; |
2b188cc1 | 3088 | struct iov_iter iter; |
31b51510 | 3089 | size_t iov_count; |
f67676d1 | 3090 | ssize_t ret, io_size; |
2b188cc1 | 3091 | |
bcda7baa | 3092 | ret = io_import_iovec(WRITE, req, &iovec, &iter, !force_nonblock); |
06b76d44 JA |
3093 | if (ret < 0) |
3094 | return ret; | |
2b188cc1 | 3095 | |
fd6c2e4c JA |
3096 | /* Ensure we clear previously set non-block flag */ |
3097 | if (!force_nonblock) | |
9adbd45d | 3098 | req->rw.kiocb.ki_flags &= ~IOCB_NOWAIT; |
fd6c2e4c | 3099 | |
f67676d1 | 3100 | io_size = ret; |
6795c5ab | 3101 | req->result = io_size; |
9e645e11 | 3102 | |
24c74678 | 3103 | /* If the file doesn't support async, just async punt */ |
af197f50 | 3104 | if (force_nonblock && !io_file_supports_async(req->file, WRITE)) |
f67676d1 | 3105 | goto copy_iov; |
31b51510 | 3106 | |
10d59345 JA |
3107 | /* file path doesn't support NOWAIT for non-direct_IO */ |
3108 | if (force_nonblock && !(kiocb->ki_flags & IOCB_DIRECT) && | |
3109 | (req->flags & REQ_F_ISREG)) | |
f67676d1 | 3110 | goto copy_iov; |
31b51510 | 3111 | |
f67676d1 | 3112 | iov_count = iov_iter_count(&iter); |
9adbd45d | 3113 | ret = rw_verify_area(WRITE, req->file, &kiocb->ki_pos, iov_count); |
2b188cc1 | 3114 | if (!ret) { |
b63534c4 | 3115 | unsigned long nr_segs = iter.nr_segs; |
9bf7933f RP |
3116 | ssize_t ret2; |
3117 | ||
2b188cc1 JA |
3118 | /* |
3119 | * Open-code file_start_write here to grab freeze protection, | |
3120 | * which will be released by another thread in | |
3121 | * io_complete_rw(). Fool lockdep by telling it the lock got | |
3122 | * released so that it doesn't complain about the held lock when | |
3123 | * we return to userspace. | |
3124 | */ | |
491381ce | 3125 | if (req->flags & REQ_F_ISREG) { |
9adbd45d | 3126 | __sb_start_write(file_inode(req->file)->i_sb, |
2b188cc1 | 3127 | SB_FREEZE_WRITE, true); |
9adbd45d | 3128 | __sb_writers_release(file_inode(req->file)->i_sb, |
2b188cc1 JA |
3129 | SB_FREEZE_WRITE); |
3130 | } | |
3131 | kiocb->ki_flags |= IOCB_WRITE; | |
9bf7933f | 3132 | |
4ed734b0 JA |
3133 | if (!force_nonblock) |
3134 | current->signal->rlim[RLIMIT_FSIZE].rlim_cur = req->fsize; | |
3135 | ||
9adbd45d JA |
3136 | if (req->file->f_op->write_iter) |
3137 | ret2 = call_write_iter(req->file, kiocb, &iter); | |
32960613 | 3138 | else |
9adbd45d | 3139 | ret2 = loop_rw_iter(WRITE, req->file, kiocb, &iter); |
4ed734b0 JA |
3140 | |
3141 | if (!force_nonblock) | |
3142 | current->signal->rlim[RLIMIT_FSIZE].rlim_cur = RLIM_INFINITY; | |
3143 | ||
faac996c | 3144 | /* |
bff6035d | 3145 | * Raw bdev writes will return -EOPNOTSUPP for IOCB_NOWAIT. Just |
faac996c JA |
3146 | * retry them without IOCB_NOWAIT. |
3147 | */ | |
3148 | if (ret2 == -EOPNOTSUPP && (kiocb->ki_flags & IOCB_NOWAIT)) | |
3149 | ret2 = -EAGAIN; | |
f67676d1 | 3150 | if (!force_nonblock || ret2 != -EAGAIN) { |
a1d7c393 | 3151 | kiocb_done(kiocb, ret2, cs); |
f67676d1 | 3152 | } else { |
b63534c4 JA |
3153 | iter.count = iov_count; |
3154 | iter.nr_segs = nr_segs; | |
f67676d1 | 3155 | copy_iov: |
b7bb4f7d | 3156 | ret = io_setup_async_rw(req, io_size, iovec, |
f67676d1 JA |
3157 | inline_vecs, &iter); |
3158 | if (ret) | |
3159 | goto out_free; | |
252917c3 PB |
3160 | /* it's copied and will be cleaned with ->io */ |
3161 | iovec = NULL; | |
f67676d1 JA |
3162 | return -EAGAIN; |
3163 | } | |
2b188cc1 | 3164 | } |
31b51510 | 3165 | out_free: |
252917c3 | 3166 | if (iovec) |
6f2cc166 | 3167 | kfree(iovec); |
2b188cc1 JA |
3168 | return ret; |
3169 | } | |
3170 | ||
f2a8d5c7 PB |
3171 | static int __io_splice_prep(struct io_kiocb *req, |
3172 | const struct io_uring_sqe *sqe) | |
7d67af2c PB |
3173 | { |
3174 | struct io_splice* sp = &req->splice; | |
3175 | unsigned int valid_flags = SPLICE_F_FD_IN_FIXED | SPLICE_F_ALL; | |
3176 | int ret; | |
3177 | ||
3178 | if (req->flags & REQ_F_NEED_CLEANUP) | |
3179 | return 0; | |
3232dd02 PB |
3180 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
3181 | return -EINVAL; | |
7d67af2c PB |
3182 | |
3183 | sp->file_in = NULL; | |
7d67af2c PB |
3184 | sp->len = READ_ONCE(sqe->len); |
3185 | sp->flags = READ_ONCE(sqe->splice_flags); | |
3186 | ||
3187 | if (unlikely(sp->flags & ~valid_flags)) | |
3188 | return -EINVAL; | |
3189 | ||
3190 | ret = io_file_get(NULL, req, READ_ONCE(sqe->splice_fd_in), &sp->file_in, | |
3191 | (sp->flags & SPLICE_F_FD_IN_FIXED)); | |
3192 | if (ret) | |
3193 | return ret; | |
3194 | req->flags |= REQ_F_NEED_CLEANUP; | |
3195 | ||
7cdaf587 XW |
3196 | if (!S_ISREG(file_inode(sp->file_in)->i_mode)) { |
3197 | /* | |
3198 | * Splice operation will be punted aync, and here need to | |
3199 | * modify io_wq_work.flags, so initialize io_wq_work firstly. | |
3200 | */ | |
3201 | io_req_init_async(req); | |
7d67af2c | 3202 | req->work.flags |= IO_WQ_WORK_UNBOUND; |
7cdaf587 | 3203 | } |
7d67af2c PB |
3204 | |
3205 | return 0; | |
3206 | } | |
3207 | ||
f2a8d5c7 PB |
3208 | static int io_tee_prep(struct io_kiocb *req, |
3209 | const struct io_uring_sqe *sqe) | |
3210 | { | |
3211 | if (READ_ONCE(sqe->splice_off_in) || READ_ONCE(sqe->off)) | |
3212 | return -EINVAL; | |
3213 | return __io_splice_prep(req, sqe); | |
3214 | } | |
3215 | ||
3216 | static int io_tee(struct io_kiocb *req, bool force_nonblock) | |
3217 | { | |
3218 | struct io_splice *sp = &req->splice; | |
3219 | struct file *in = sp->file_in; | |
3220 | struct file *out = sp->file_out; | |
3221 | unsigned int flags = sp->flags & ~SPLICE_F_FD_IN_FIXED; | |
3222 | long ret = 0; | |
3223 | ||
3224 | if (force_nonblock) | |
3225 | return -EAGAIN; | |
3226 | if (sp->len) | |
3227 | ret = do_tee(in, out, sp->len, flags); | |
3228 | ||
3229 | io_put_file(req, in, (sp->flags & SPLICE_F_FD_IN_FIXED)); | |
3230 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
3231 | ||
f2a8d5c7 PB |
3232 | if (ret != sp->len) |
3233 | req_set_fail_links(req); | |
e1e16097 | 3234 | io_req_complete(req, ret); |
f2a8d5c7 PB |
3235 | return 0; |
3236 | } | |
3237 | ||
3238 | static int io_splice_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) | |
3239 | { | |
3240 | struct io_splice* sp = &req->splice; | |
3241 | ||
3242 | sp->off_in = READ_ONCE(sqe->splice_off_in); | |
3243 | sp->off_out = READ_ONCE(sqe->off); | |
3244 | return __io_splice_prep(req, sqe); | |
3245 | } | |
3246 | ||
014db007 | 3247 | static int io_splice(struct io_kiocb *req, bool force_nonblock) |
7d67af2c PB |
3248 | { |
3249 | struct io_splice *sp = &req->splice; | |
3250 | struct file *in = sp->file_in; | |
3251 | struct file *out = sp->file_out; | |
3252 | unsigned int flags = sp->flags & ~SPLICE_F_FD_IN_FIXED; | |
3253 | loff_t *poff_in, *poff_out; | |
c9687426 | 3254 | long ret = 0; |
7d67af2c | 3255 | |
2fb3e822 PB |
3256 | if (force_nonblock) |
3257 | return -EAGAIN; | |
7d67af2c PB |
3258 | |
3259 | poff_in = (sp->off_in == -1) ? NULL : &sp->off_in; | |
3260 | poff_out = (sp->off_out == -1) ? NULL : &sp->off_out; | |
c9687426 | 3261 | |
948a7749 | 3262 | if (sp->len) |
c9687426 | 3263 | ret = do_splice(in, poff_in, out, poff_out, sp->len, flags); |
7d67af2c PB |
3264 | |
3265 | io_put_file(req, in, (sp->flags & SPLICE_F_FD_IN_FIXED)); | |
3266 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
3267 | ||
7d67af2c PB |
3268 | if (ret != sp->len) |
3269 | req_set_fail_links(req); | |
e1e16097 | 3270 | io_req_complete(req, ret); |
7d67af2c PB |
3271 | return 0; |
3272 | } | |
3273 | ||
2b188cc1 JA |
3274 | /* |
3275 | * IORING_OP_NOP just posts a completion event, nothing else. | |
3276 | */ | |
229a7b63 | 3277 | static int io_nop(struct io_kiocb *req, struct io_comp_state *cs) |
2b188cc1 JA |
3278 | { |
3279 | struct io_ring_ctx *ctx = req->ctx; | |
2b188cc1 | 3280 | |
def596e9 JA |
3281 | if (unlikely(ctx->flags & IORING_SETUP_IOPOLL)) |
3282 | return -EINVAL; | |
3283 | ||
229a7b63 | 3284 | __io_req_complete(req, 0, 0, cs); |
2b188cc1 JA |
3285 | return 0; |
3286 | } | |
3287 | ||
3529d8c2 | 3288 | static int io_prep_fsync(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
c992fe29 | 3289 | { |
6b06314c | 3290 | struct io_ring_ctx *ctx = req->ctx; |
c992fe29 | 3291 | |
09bb8394 JA |
3292 | if (!req->file) |
3293 | return -EBADF; | |
c992fe29 | 3294 | |
6b06314c | 3295 | if (unlikely(ctx->flags & IORING_SETUP_IOPOLL)) |
def596e9 | 3296 | return -EINVAL; |
edafccee | 3297 | if (unlikely(sqe->addr || sqe->ioprio || sqe->buf_index)) |
c992fe29 CH |
3298 | return -EINVAL; |
3299 | ||
8ed8d3c3 JA |
3300 | req->sync.flags = READ_ONCE(sqe->fsync_flags); |
3301 | if (unlikely(req->sync.flags & ~IORING_FSYNC_DATASYNC)) | |
3302 | return -EINVAL; | |
3303 | ||
3304 | req->sync.off = READ_ONCE(sqe->off); | |
3305 | req->sync.len = READ_ONCE(sqe->len); | |
c992fe29 CH |
3306 | return 0; |
3307 | } | |
3308 | ||
ac45abc0 | 3309 | static int io_fsync(struct io_kiocb *req, bool force_nonblock) |
8ed8d3c3 | 3310 | { |
8ed8d3c3 | 3311 | loff_t end = req->sync.off + req->sync.len; |
8ed8d3c3 JA |
3312 | int ret; |
3313 | ||
ac45abc0 PB |
3314 | /* fsync always requires a blocking context */ |
3315 | if (force_nonblock) | |
3316 | return -EAGAIN; | |
3317 | ||
9adbd45d | 3318 | ret = vfs_fsync_range(req->file, req->sync.off, |
8ed8d3c3 JA |
3319 | end > 0 ? end : LLONG_MAX, |
3320 | req->sync.flags & IORING_FSYNC_DATASYNC); | |
3321 | if (ret < 0) | |
3322 | req_set_fail_links(req); | |
e1e16097 | 3323 | io_req_complete(req, ret); |
c992fe29 CH |
3324 | return 0; |
3325 | } | |
3326 | ||
d63d1b5e JA |
3327 | static int io_fallocate_prep(struct io_kiocb *req, |
3328 | const struct io_uring_sqe *sqe) | |
3329 | { | |
3330 | if (sqe->ioprio || sqe->buf_index || sqe->rw_flags) | |
3331 | return -EINVAL; | |
3232dd02 PB |
3332 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
3333 | return -EINVAL; | |
d63d1b5e JA |
3334 | |
3335 | req->sync.off = READ_ONCE(sqe->off); | |
3336 | req->sync.len = READ_ONCE(sqe->addr); | |
3337 | req->sync.mode = READ_ONCE(sqe->len); | |
4ed734b0 | 3338 | req->fsize = rlimit(RLIMIT_FSIZE); |
d63d1b5e JA |
3339 | return 0; |
3340 | } | |
3341 | ||
014db007 | 3342 | static int io_fallocate(struct io_kiocb *req, bool force_nonblock) |
5d17b4a4 | 3343 | { |
ac45abc0 PB |
3344 | int ret; |
3345 | ||
d63d1b5e | 3346 | /* fallocate always requiring blocking context */ |
ac45abc0 | 3347 | if (force_nonblock) |
5d17b4a4 JA |
3348 | return -EAGAIN; |
3349 | ||
ac45abc0 PB |
3350 | current->signal->rlim[RLIMIT_FSIZE].rlim_cur = req->fsize; |
3351 | ret = vfs_fallocate(req->file, req->sync.mode, req->sync.off, | |
3352 | req->sync.len); | |
3353 | current->signal->rlim[RLIMIT_FSIZE].rlim_cur = RLIM_INFINITY; | |
3354 | if (ret < 0) | |
3355 | req_set_fail_links(req); | |
e1e16097 | 3356 | io_req_complete(req, ret); |
5d17b4a4 JA |
3357 | return 0; |
3358 | } | |
3359 | ||
ec65fea5 | 3360 | static int __io_openat_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
b7bb4f7d | 3361 | { |
f8748881 | 3362 | const char __user *fname; |
15b71abe | 3363 | int ret; |
b7bb4f7d | 3364 | |
3232dd02 | 3365 | if (unlikely(req->ctx->flags & (IORING_SETUP_IOPOLL|IORING_SETUP_SQPOLL))) |
15b71abe | 3366 | return -EINVAL; |
ec65fea5 | 3367 | if (unlikely(sqe->ioprio || sqe->buf_index)) |
15b71abe | 3368 | return -EINVAL; |
ec65fea5 | 3369 | if (unlikely(req->flags & REQ_F_FIXED_FILE)) |
cf3040ca | 3370 | return -EBADF; |
03b1230c | 3371 | |
ec65fea5 PB |
3372 | /* open.how should be already initialised */ |
3373 | if (!(req->open.how.flags & O_PATH) && force_o_largefile()) | |
08a1d26e | 3374 | req->open.how.flags |= O_LARGEFILE; |
3529d8c2 | 3375 | |
25e72d10 PB |
3376 | req->open.dfd = READ_ONCE(sqe->fd); |
3377 | fname = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
f8748881 | 3378 | req->open.filename = getname(fname); |
15b71abe JA |
3379 | if (IS_ERR(req->open.filename)) { |
3380 | ret = PTR_ERR(req->open.filename); | |
3381 | req->open.filename = NULL; | |
3382 | return ret; | |
3383 | } | |
4022e7af | 3384 | req->open.nofile = rlimit(RLIMIT_NOFILE); |
8fef80bf | 3385 | req->flags |= REQ_F_NEED_CLEANUP; |
15b71abe | 3386 | return 0; |
03b1230c JA |
3387 | } |
3388 | ||
ec65fea5 PB |
3389 | static int io_openat_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
3390 | { | |
3391 | u64 flags, mode; | |
3392 | ||
3393 | if (req->flags & REQ_F_NEED_CLEANUP) | |
3394 | return 0; | |
3395 | mode = READ_ONCE(sqe->len); | |
3396 | flags = READ_ONCE(sqe->open_flags); | |
3397 | req->open.how = build_open_how(flags, mode); | |
3398 | return __io_openat_prep(req, sqe); | |
3399 | } | |
3400 | ||
cebdb986 | 3401 | static int io_openat2_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
aa1fa28f | 3402 | { |
cebdb986 | 3403 | struct open_how __user *how; |
cebdb986 | 3404 | size_t len; |
0fa03c62 JA |
3405 | int ret; |
3406 | ||
0bdbdd08 PB |
3407 | if (req->flags & REQ_F_NEED_CLEANUP) |
3408 | return 0; | |
cebdb986 JA |
3409 | how = u64_to_user_ptr(READ_ONCE(sqe->addr2)); |
3410 | len = READ_ONCE(sqe->len); | |
cebdb986 JA |
3411 | if (len < OPEN_HOW_SIZE_VER0) |
3412 | return -EINVAL; | |
3529d8c2 | 3413 | |
cebdb986 JA |
3414 | ret = copy_struct_from_user(&req->open.how, sizeof(req->open.how), how, |
3415 | len); | |
3416 | if (ret) | |
3417 | return ret; | |
3529d8c2 | 3418 | |
ec65fea5 | 3419 | return __io_openat_prep(req, sqe); |
cebdb986 JA |
3420 | } |
3421 | ||
014db007 | 3422 | static int io_openat2(struct io_kiocb *req, bool force_nonblock) |
15b71abe JA |
3423 | { |
3424 | struct open_flags op; | |
15b71abe JA |
3425 | struct file *file; |
3426 | int ret; | |
3427 | ||
f86cd20c | 3428 | if (force_nonblock) |
15b71abe | 3429 | return -EAGAIN; |
15b71abe | 3430 | |
cebdb986 | 3431 | ret = build_open_flags(&req->open.how, &op); |
15b71abe JA |
3432 | if (ret) |
3433 | goto err; | |
3434 | ||
4022e7af | 3435 | ret = __get_unused_fd_flags(req->open.how.flags, req->open.nofile); |
15b71abe JA |
3436 | if (ret < 0) |
3437 | goto err; | |
3438 | ||
3439 | file = do_filp_open(req->open.dfd, req->open.filename, &op); | |
3440 | if (IS_ERR(file)) { | |
3441 | put_unused_fd(ret); | |
3442 | ret = PTR_ERR(file); | |
3443 | } else { | |
3444 | fsnotify_open(file); | |
3445 | fd_install(ret, file); | |
3446 | } | |
3447 | err: | |
3448 | putname(req->open.filename); | |
8fef80bf | 3449 | req->flags &= ~REQ_F_NEED_CLEANUP; |
15b71abe JA |
3450 | if (ret < 0) |
3451 | req_set_fail_links(req); | |
e1e16097 | 3452 | io_req_complete(req, ret); |
15b71abe JA |
3453 | return 0; |
3454 | } | |
3455 | ||
014db007 | 3456 | static int io_openat(struct io_kiocb *req, bool force_nonblock) |
cebdb986 | 3457 | { |
014db007 | 3458 | return io_openat2(req, force_nonblock); |
cebdb986 JA |
3459 | } |
3460 | ||
067524e9 JA |
3461 | static int io_remove_buffers_prep(struct io_kiocb *req, |
3462 | const struct io_uring_sqe *sqe) | |
3463 | { | |
3464 | struct io_provide_buf *p = &req->pbuf; | |
3465 | u64 tmp; | |
3466 | ||
3467 | if (sqe->ioprio || sqe->rw_flags || sqe->addr || sqe->len || sqe->off) | |
3468 | return -EINVAL; | |
3469 | ||
3470 | tmp = READ_ONCE(sqe->fd); | |
3471 | if (!tmp || tmp > USHRT_MAX) | |
3472 | return -EINVAL; | |
3473 | ||
3474 | memset(p, 0, sizeof(*p)); | |
3475 | p->nbufs = tmp; | |
3476 | p->bgid = READ_ONCE(sqe->buf_group); | |
3477 | return 0; | |
3478 | } | |
3479 | ||
3480 | static int __io_remove_buffers(struct io_ring_ctx *ctx, struct io_buffer *buf, | |
3481 | int bgid, unsigned nbufs) | |
3482 | { | |
3483 | unsigned i = 0; | |
3484 | ||
3485 | /* shouldn't happen */ | |
3486 | if (!nbufs) | |
3487 | return 0; | |
3488 | ||
3489 | /* the head kbuf is the list itself */ | |
3490 | while (!list_empty(&buf->list)) { | |
3491 | struct io_buffer *nxt; | |
3492 | ||
3493 | nxt = list_first_entry(&buf->list, struct io_buffer, list); | |
3494 | list_del(&nxt->list); | |
3495 | kfree(nxt); | |
3496 | if (++i == nbufs) | |
3497 | return i; | |
3498 | } | |
3499 | i++; | |
3500 | kfree(buf); | |
3501 | idr_remove(&ctx->io_buffer_idr, bgid); | |
3502 | ||
3503 | return i; | |
3504 | } | |
3505 | ||
229a7b63 JA |
3506 | static int io_remove_buffers(struct io_kiocb *req, bool force_nonblock, |
3507 | struct io_comp_state *cs) | |
067524e9 JA |
3508 | { |
3509 | struct io_provide_buf *p = &req->pbuf; | |
3510 | struct io_ring_ctx *ctx = req->ctx; | |
3511 | struct io_buffer *head; | |
3512 | int ret = 0; | |
3513 | ||
3514 | io_ring_submit_lock(ctx, !force_nonblock); | |
3515 | ||
3516 | lockdep_assert_held(&ctx->uring_lock); | |
3517 | ||
3518 | ret = -ENOENT; | |
3519 | head = idr_find(&ctx->io_buffer_idr, p->bgid); | |
3520 | if (head) | |
3521 | ret = __io_remove_buffers(ctx, head, p->bgid, p->nbufs); | |
3522 | ||
3523 | io_ring_submit_lock(ctx, !force_nonblock); | |
3524 | if (ret < 0) | |
3525 | req_set_fail_links(req); | |
229a7b63 | 3526 | __io_req_complete(req, ret, 0, cs); |
067524e9 JA |
3527 | return 0; |
3528 | } | |
3529 | ||
ddf0322d JA |
3530 | static int io_provide_buffers_prep(struct io_kiocb *req, |
3531 | const struct io_uring_sqe *sqe) | |
3532 | { | |
3533 | struct io_provide_buf *p = &req->pbuf; | |
3534 | u64 tmp; | |
3535 | ||
3536 | if (sqe->ioprio || sqe->rw_flags) | |
3537 | return -EINVAL; | |
3538 | ||
3539 | tmp = READ_ONCE(sqe->fd); | |
3540 | if (!tmp || tmp > USHRT_MAX) | |
3541 | return -E2BIG; | |
3542 | p->nbufs = tmp; | |
3543 | p->addr = READ_ONCE(sqe->addr); | |
3544 | p->len = READ_ONCE(sqe->len); | |
3545 | ||
efe68c1c | 3546 | if (!access_ok(u64_to_user_ptr(p->addr), (p->len * p->nbufs))) |
ddf0322d JA |
3547 | return -EFAULT; |
3548 | ||
3549 | p->bgid = READ_ONCE(sqe->buf_group); | |
3550 | tmp = READ_ONCE(sqe->off); | |
3551 | if (tmp > USHRT_MAX) | |
3552 | return -E2BIG; | |
3553 | p->bid = tmp; | |
3554 | return 0; | |
3555 | } | |
3556 | ||
3557 | static int io_add_buffers(struct io_provide_buf *pbuf, struct io_buffer **head) | |
3558 | { | |
3559 | struct io_buffer *buf; | |
3560 | u64 addr = pbuf->addr; | |
3561 | int i, bid = pbuf->bid; | |
3562 | ||
3563 | for (i = 0; i < pbuf->nbufs; i++) { | |
3564 | buf = kmalloc(sizeof(*buf), GFP_KERNEL); | |
3565 | if (!buf) | |
3566 | break; | |
3567 | ||
3568 | buf->addr = addr; | |
3569 | buf->len = pbuf->len; | |
3570 | buf->bid = bid; | |
3571 | addr += pbuf->len; | |
3572 | bid++; | |
3573 | if (!*head) { | |
3574 | INIT_LIST_HEAD(&buf->list); | |
3575 | *head = buf; | |
3576 | } else { | |
3577 | list_add_tail(&buf->list, &(*head)->list); | |
3578 | } | |
3579 | } | |
3580 | ||
3581 | return i ? i : -ENOMEM; | |
3582 | } | |
3583 | ||
229a7b63 JA |
3584 | static int io_provide_buffers(struct io_kiocb *req, bool force_nonblock, |
3585 | struct io_comp_state *cs) | |
ddf0322d JA |
3586 | { |
3587 | struct io_provide_buf *p = &req->pbuf; | |
3588 | struct io_ring_ctx *ctx = req->ctx; | |
3589 | struct io_buffer *head, *list; | |
3590 | int ret = 0; | |
3591 | ||
3592 | io_ring_submit_lock(ctx, !force_nonblock); | |
3593 | ||
3594 | lockdep_assert_held(&ctx->uring_lock); | |
3595 | ||
3596 | list = head = idr_find(&ctx->io_buffer_idr, p->bgid); | |
3597 | ||
3598 | ret = io_add_buffers(p, &head); | |
3599 | if (ret < 0) | |
3600 | goto out; | |
3601 | ||
3602 | if (!list) { | |
3603 | ret = idr_alloc(&ctx->io_buffer_idr, head, p->bgid, p->bgid + 1, | |
3604 | GFP_KERNEL); | |
3605 | if (ret < 0) { | |
067524e9 | 3606 | __io_remove_buffers(ctx, head, p->bgid, -1U); |
ddf0322d JA |
3607 | goto out; |
3608 | } | |
3609 | } | |
3610 | out: | |
3611 | io_ring_submit_unlock(ctx, !force_nonblock); | |
3612 | if (ret < 0) | |
3613 | req_set_fail_links(req); | |
229a7b63 | 3614 | __io_req_complete(req, ret, 0, cs); |
ddf0322d | 3615 | return 0; |
cebdb986 JA |
3616 | } |
3617 | ||
3e4827b0 JA |
3618 | static int io_epoll_ctl_prep(struct io_kiocb *req, |
3619 | const struct io_uring_sqe *sqe) | |
3620 | { | |
3621 | #if defined(CONFIG_EPOLL) | |
3622 | if (sqe->ioprio || sqe->buf_index) | |
3623 | return -EINVAL; | |
3232dd02 PB |
3624 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
3625 | return -EINVAL; | |
3e4827b0 JA |
3626 | |
3627 | req->epoll.epfd = READ_ONCE(sqe->fd); | |
3628 | req->epoll.op = READ_ONCE(sqe->len); | |
3629 | req->epoll.fd = READ_ONCE(sqe->off); | |
3630 | ||
3631 | if (ep_op_has_event(req->epoll.op)) { | |
3632 | struct epoll_event __user *ev; | |
3633 | ||
3634 | ev = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
3635 | if (copy_from_user(&req->epoll.event, ev, sizeof(*ev))) | |
3636 | return -EFAULT; | |
3637 | } | |
3638 | ||
3639 | return 0; | |
3640 | #else | |
3641 | return -EOPNOTSUPP; | |
3642 | #endif | |
3643 | } | |
3644 | ||
229a7b63 JA |
3645 | static int io_epoll_ctl(struct io_kiocb *req, bool force_nonblock, |
3646 | struct io_comp_state *cs) | |
3e4827b0 JA |
3647 | { |
3648 | #if defined(CONFIG_EPOLL) | |
3649 | struct io_epoll *ie = &req->epoll; | |
3650 | int ret; | |
3651 | ||
3652 | ret = do_epoll_ctl(ie->epfd, ie->op, ie->fd, &ie->event, force_nonblock); | |
3653 | if (force_nonblock && ret == -EAGAIN) | |
3654 | return -EAGAIN; | |
3655 | ||
3656 | if (ret < 0) | |
3657 | req_set_fail_links(req); | |
229a7b63 | 3658 | __io_req_complete(req, ret, 0, cs); |
3e4827b0 JA |
3659 | return 0; |
3660 | #else | |
3661 | return -EOPNOTSUPP; | |
3662 | #endif | |
3663 | } | |
3664 | ||
c1ca757b JA |
3665 | static int io_madvise_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
3666 | { | |
3667 | #if defined(CONFIG_ADVISE_SYSCALLS) && defined(CONFIG_MMU) | |
3668 | if (sqe->ioprio || sqe->buf_index || sqe->off) | |
3669 | return -EINVAL; | |
3232dd02 PB |
3670 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
3671 | return -EINVAL; | |
c1ca757b JA |
3672 | |
3673 | req->madvise.addr = READ_ONCE(sqe->addr); | |
3674 | req->madvise.len = READ_ONCE(sqe->len); | |
3675 | req->madvise.advice = READ_ONCE(sqe->fadvise_advice); | |
3676 | return 0; | |
3677 | #else | |
3678 | return -EOPNOTSUPP; | |
3679 | #endif | |
3680 | } | |
3681 | ||
014db007 | 3682 | static int io_madvise(struct io_kiocb *req, bool force_nonblock) |
c1ca757b JA |
3683 | { |
3684 | #if defined(CONFIG_ADVISE_SYSCALLS) && defined(CONFIG_MMU) | |
3685 | struct io_madvise *ma = &req->madvise; | |
3686 | int ret; | |
3687 | ||
3688 | if (force_nonblock) | |
3689 | return -EAGAIN; | |
3690 | ||
3691 | ret = do_madvise(ma->addr, ma->len, ma->advice); | |
3692 | if (ret < 0) | |
3693 | req_set_fail_links(req); | |
e1e16097 | 3694 | io_req_complete(req, ret); |
c1ca757b JA |
3695 | return 0; |
3696 | #else | |
3697 | return -EOPNOTSUPP; | |
3698 | #endif | |
3699 | } | |
3700 | ||
4840e418 JA |
3701 | static int io_fadvise_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
3702 | { | |
3703 | if (sqe->ioprio || sqe->buf_index || sqe->addr) | |
3704 | return -EINVAL; | |
3232dd02 PB |
3705 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
3706 | return -EINVAL; | |
4840e418 JA |
3707 | |
3708 | req->fadvise.offset = READ_ONCE(sqe->off); | |
3709 | req->fadvise.len = READ_ONCE(sqe->len); | |
3710 | req->fadvise.advice = READ_ONCE(sqe->fadvise_advice); | |
3711 | return 0; | |
3712 | } | |
3713 | ||
014db007 | 3714 | static int io_fadvise(struct io_kiocb *req, bool force_nonblock) |
4840e418 JA |
3715 | { |
3716 | struct io_fadvise *fa = &req->fadvise; | |
3717 | int ret; | |
3718 | ||
3e69426d JA |
3719 | if (force_nonblock) { |
3720 | switch (fa->advice) { | |
3721 | case POSIX_FADV_NORMAL: | |
3722 | case POSIX_FADV_RANDOM: | |
3723 | case POSIX_FADV_SEQUENTIAL: | |
3724 | break; | |
3725 | default: | |
3726 | return -EAGAIN; | |
3727 | } | |
3728 | } | |
4840e418 JA |
3729 | |
3730 | ret = vfs_fadvise(req->file, fa->offset, fa->len, fa->advice); | |
3731 | if (ret < 0) | |
3732 | req_set_fail_links(req); | |
e1e16097 | 3733 | io_req_complete(req, ret); |
4840e418 JA |
3734 | return 0; |
3735 | } | |
3736 | ||
eddc7ef5 JA |
3737 | static int io_statx_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
3738 | { | |
3232dd02 PB |
3739 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
3740 | return -EINVAL; | |
eddc7ef5 JA |
3741 | if (sqe->ioprio || sqe->buf_index) |
3742 | return -EINVAL; | |
9c280f90 | 3743 | if (req->flags & REQ_F_FIXED_FILE) |
cf3040ca | 3744 | return -EBADF; |
eddc7ef5 | 3745 | |
1d9e1288 BM |
3746 | req->statx.dfd = READ_ONCE(sqe->fd); |
3747 | req->statx.mask = READ_ONCE(sqe->len); | |
e62753e4 | 3748 | req->statx.filename = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
1d9e1288 BM |
3749 | req->statx.buffer = u64_to_user_ptr(READ_ONCE(sqe->addr2)); |
3750 | req->statx.flags = READ_ONCE(sqe->statx_flags); | |
eddc7ef5 JA |
3751 | |
3752 | return 0; | |
3753 | } | |
3754 | ||
014db007 | 3755 | static int io_statx(struct io_kiocb *req, bool force_nonblock) |
eddc7ef5 | 3756 | { |
1d9e1288 | 3757 | struct io_statx *ctx = &req->statx; |
eddc7ef5 JA |
3758 | int ret; |
3759 | ||
5b0bbee4 JA |
3760 | if (force_nonblock) { |
3761 | /* only need file table for an actual valid fd */ | |
3762 | if (ctx->dfd == -1 || ctx->dfd == AT_FDCWD) | |
3763 | req->flags |= REQ_F_NO_FILE_TABLE; | |
eddc7ef5 | 3764 | return -EAGAIN; |
5b0bbee4 | 3765 | } |
eddc7ef5 | 3766 | |
e62753e4 BM |
3767 | ret = do_statx(ctx->dfd, ctx->filename, ctx->flags, ctx->mask, |
3768 | ctx->buffer); | |
eddc7ef5 | 3769 | |
eddc7ef5 JA |
3770 | if (ret < 0) |
3771 | req_set_fail_links(req); | |
e1e16097 | 3772 | io_req_complete(req, ret); |
eddc7ef5 JA |
3773 | return 0; |
3774 | } | |
3775 | ||
b5dba59e JA |
3776 | static int io_close_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
3777 | { | |
3778 | /* | |
3779 | * If we queue this for async, it must not be cancellable. That would | |
7cdaf587 XW |
3780 | * leave the 'file' in an undeterminate state, and here need to modify |
3781 | * io_wq_work.flags, so initialize io_wq_work firstly. | |
b5dba59e | 3782 | */ |
7cdaf587 | 3783 | io_req_init_async(req); |
b5dba59e JA |
3784 | req->work.flags |= IO_WQ_WORK_NO_CANCEL; |
3785 | ||
3232dd02 PB |
3786 | if (unlikely(req->ctx->flags & (IORING_SETUP_IOPOLL|IORING_SETUP_SQPOLL))) |
3787 | return -EINVAL; | |
b5dba59e JA |
3788 | if (sqe->ioprio || sqe->off || sqe->addr || sqe->len || |
3789 | sqe->rw_flags || sqe->buf_index) | |
3790 | return -EINVAL; | |
9c280f90 | 3791 | if (req->flags & REQ_F_FIXED_FILE) |
cf3040ca | 3792 | return -EBADF; |
b5dba59e JA |
3793 | |
3794 | req->close.fd = READ_ONCE(sqe->fd); | |
fd2206e4 JA |
3795 | if ((req->file && req->file->f_op == &io_uring_fops) || |
3796 | req->close.fd == req->ctx->ring_fd) | |
3797 | return -EBADF; | |
b5dba59e | 3798 | |
3af73b28 | 3799 | req->close.put_file = NULL; |
b5dba59e | 3800 | return 0; |
b5dba59e JA |
3801 | } |
3802 | ||
229a7b63 JA |
3803 | static int io_close(struct io_kiocb *req, bool force_nonblock, |
3804 | struct io_comp_state *cs) | |
b5dba59e | 3805 | { |
3af73b28 | 3806 | struct io_close *close = &req->close; |
b5dba59e JA |
3807 | int ret; |
3808 | ||
3af73b28 PB |
3809 | /* might be already done during nonblock submission */ |
3810 | if (!close->put_file) { | |
3811 | ret = __close_fd_get_file(close->fd, &close->put_file); | |
3812 | if (ret < 0) | |
3813 | return (ret == -ENOENT) ? -EBADF : ret; | |
3814 | } | |
b5dba59e JA |
3815 | |
3816 | /* if the file has a flush method, be safe and punt to async */ | |
3af73b28 | 3817 | if (close->put_file->f_op->flush && force_nonblock) { |
24c74678 PB |
3818 | /* was never set, but play safe */ |
3819 | req->flags &= ~REQ_F_NOWAIT; | |
0bf0eefd | 3820 | /* avoid grabbing files - we don't need the files */ |
24c74678 | 3821 | req->flags |= REQ_F_NO_FILE_TABLE; |
0bf0eefd | 3822 | return -EAGAIN; |
a2100672 | 3823 | } |
b5dba59e | 3824 | |
3af73b28 PB |
3825 | /* No ->flush() or already async, safely close from here */ |
3826 | ret = filp_close(close->put_file, req->work.files); | |
3827 | if (ret < 0) | |
3828 | req_set_fail_links(req); | |
3af73b28 PB |
3829 | fput(close->put_file); |
3830 | close->put_file = NULL; | |
229a7b63 | 3831 | __io_req_complete(req, ret, 0, cs); |
1a417f4e | 3832 | return 0; |
b5dba59e JA |
3833 | } |
3834 | ||
3529d8c2 | 3835 | static int io_prep_sfr(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
5d17b4a4 JA |
3836 | { |
3837 | struct io_ring_ctx *ctx = req->ctx; | |
5d17b4a4 JA |
3838 | |
3839 | if (!req->file) | |
3840 | return -EBADF; | |
5d17b4a4 JA |
3841 | |
3842 | if (unlikely(ctx->flags & IORING_SETUP_IOPOLL)) | |
3843 | return -EINVAL; | |
3844 | if (unlikely(sqe->addr || sqe->ioprio || sqe->buf_index)) | |
3845 | return -EINVAL; | |
3846 | ||
8ed8d3c3 JA |
3847 | req->sync.off = READ_ONCE(sqe->off); |
3848 | req->sync.len = READ_ONCE(sqe->len); | |
3849 | req->sync.flags = READ_ONCE(sqe->sync_range_flags); | |
8ed8d3c3 JA |
3850 | return 0; |
3851 | } | |
3852 | ||
ac45abc0 | 3853 | static int io_sync_file_range(struct io_kiocb *req, bool force_nonblock) |
8ed8d3c3 | 3854 | { |
8ed8d3c3 JA |
3855 | int ret; |
3856 | ||
ac45abc0 PB |
3857 | /* sync_file_range always requires a blocking context */ |
3858 | if (force_nonblock) | |
3859 | return -EAGAIN; | |
3860 | ||
9adbd45d | 3861 | ret = sync_file_range(req->file, req->sync.off, req->sync.len, |
8ed8d3c3 JA |
3862 | req->sync.flags); |
3863 | if (ret < 0) | |
3864 | req_set_fail_links(req); | |
e1e16097 | 3865 | io_req_complete(req, ret); |
5d17b4a4 JA |
3866 | return 0; |
3867 | } | |
3868 | ||
469956e8 | 3869 | #if defined(CONFIG_NET) |
02d27d89 PB |
3870 | static int io_setup_async_msg(struct io_kiocb *req, |
3871 | struct io_async_msghdr *kmsg) | |
3872 | { | |
3873 | if (req->io) | |
3874 | return -EAGAIN; | |
3875 | if (io_alloc_async_ctx(req)) { | |
3876 | if (kmsg->iov != kmsg->fast_iov) | |
3877 | kfree(kmsg->iov); | |
3878 | return -ENOMEM; | |
3879 | } | |
3880 | req->flags |= REQ_F_NEED_CLEANUP; | |
3881 | memcpy(&req->io->msg, kmsg, sizeof(*kmsg)); | |
3882 | return -EAGAIN; | |
3883 | } | |
3884 | ||
2ae523ed PB |
3885 | static int io_sendmsg_copy_hdr(struct io_kiocb *req, |
3886 | struct io_async_msghdr *iomsg) | |
3887 | { | |
3888 | iomsg->iov = iomsg->fast_iov; | |
3889 | iomsg->msg.msg_name = &iomsg->addr; | |
3890 | return sendmsg_copy_msghdr(&iomsg->msg, req->sr_msg.umsg, | |
3891 | req->sr_msg.msg_flags, &iomsg->iov); | |
3892 | } | |
3893 | ||
3529d8c2 | 3894 | static int io_sendmsg_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
03b1230c | 3895 | { |
e47293fd | 3896 | struct io_sr_msg *sr = &req->sr_msg; |
3529d8c2 | 3897 | struct io_async_ctx *io = req->io; |
99bc4c38 | 3898 | int ret; |
03b1230c | 3899 | |
d2b6f48b PB |
3900 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
3901 | return -EINVAL; | |
3902 | ||
e47293fd | 3903 | sr->msg_flags = READ_ONCE(sqe->msg_flags); |
270a5940 | 3904 | sr->umsg = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
fddaface | 3905 | sr->len = READ_ONCE(sqe->len); |
3529d8c2 | 3906 | |
d8768362 JA |
3907 | #ifdef CONFIG_COMPAT |
3908 | if (req->ctx->compat) | |
3909 | sr->msg_flags |= MSG_CMSG_COMPAT; | |
3910 | #endif | |
3911 | ||
fddaface | 3912 | if (!io || req->opcode == IORING_OP_SEND) |
3529d8c2 | 3913 | return 0; |
5f798bea PB |
3914 | /* iovec is already imported */ |
3915 | if (req->flags & REQ_F_NEED_CLEANUP) | |
3916 | return 0; | |
3529d8c2 | 3917 | |
2ae523ed | 3918 | ret = io_sendmsg_copy_hdr(req, &io->msg); |
99bc4c38 PB |
3919 | if (!ret) |
3920 | req->flags |= REQ_F_NEED_CLEANUP; | |
3921 | return ret; | |
03b1230c JA |
3922 | } |
3923 | ||
229a7b63 JA |
3924 | static int io_sendmsg(struct io_kiocb *req, bool force_nonblock, |
3925 | struct io_comp_state *cs) | |
aa1fa28f | 3926 | { |
0b416c3e | 3927 | struct io_async_msghdr *kmsg = NULL; |
0fa03c62 JA |
3928 | struct socket *sock; |
3929 | int ret; | |
3930 | ||
0fa03c62 JA |
3931 | sock = sock_from_file(req->file, &ret); |
3932 | if (sock) { | |
1400e697 | 3933 | struct io_async_msghdr iomsg; |
0fa03c62 JA |
3934 | unsigned flags; |
3935 | ||
03b1230c | 3936 | if (req->io) { |
0b416c3e | 3937 | kmsg = &req->io->msg; |
b537916c | 3938 | kmsg->msg.msg_name = &req->io->msg.addr; |
0b416c3e JA |
3939 | /* if iov is set, it's allocated already */ |
3940 | if (!kmsg->iov) | |
3941 | kmsg->iov = kmsg->fast_iov; | |
3942 | kmsg->msg.msg_iter.iov = kmsg->iov; | |
03b1230c | 3943 | } else { |
2ae523ed | 3944 | ret = io_sendmsg_copy_hdr(req, &iomsg); |
03b1230c | 3945 | if (ret) |
3529d8c2 | 3946 | return ret; |
1400e697 | 3947 | kmsg = &iomsg; |
03b1230c | 3948 | } |
0fa03c62 | 3949 | |
e47293fd JA |
3950 | flags = req->sr_msg.msg_flags; |
3951 | if (flags & MSG_DONTWAIT) | |
3952 | req->flags |= REQ_F_NOWAIT; | |
3953 | else if (force_nonblock) | |
3954 | flags |= MSG_DONTWAIT; | |
3955 | ||
0b416c3e | 3956 | ret = __sys_sendmsg_sock(sock, &kmsg->msg, flags); |
02d27d89 PB |
3957 | if (force_nonblock && ret == -EAGAIN) |
3958 | return io_setup_async_msg(req, kmsg); | |
441cdbd5 JA |
3959 | if (ret == -ERESTARTSYS) |
3960 | ret = -EINTR; | |
0fa03c62 JA |
3961 | } |
3962 | ||
1e95081c | 3963 | if (kmsg && kmsg->iov != kmsg->fast_iov) |
0b416c3e | 3964 | kfree(kmsg->iov); |
99bc4c38 | 3965 | req->flags &= ~REQ_F_NEED_CLEANUP; |
4e88d6e7 JA |
3966 | if (ret < 0) |
3967 | req_set_fail_links(req); | |
229a7b63 | 3968 | __io_req_complete(req, ret, 0, cs); |
5d17b4a4 | 3969 | return 0; |
03b1230c | 3970 | } |
aa1fa28f | 3971 | |
229a7b63 JA |
3972 | static int io_send(struct io_kiocb *req, bool force_nonblock, |
3973 | struct io_comp_state *cs) | |
fddaface | 3974 | { |
fddaface JA |
3975 | struct socket *sock; |
3976 | int ret; | |
3977 | ||
fddaface JA |
3978 | sock = sock_from_file(req->file, &ret); |
3979 | if (sock) { | |
3980 | struct io_sr_msg *sr = &req->sr_msg; | |
3981 | struct msghdr msg; | |
3982 | struct iovec iov; | |
3983 | unsigned flags; | |
3984 | ||
3985 | ret = import_single_range(WRITE, sr->buf, sr->len, &iov, | |
3986 | &msg.msg_iter); | |
3987 | if (ret) | |
3988 | return ret; | |
3989 | ||
3990 | msg.msg_name = NULL; | |
3991 | msg.msg_control = NULL; | |
3992 | msg.msg_controllen = 0; | |
3993 | msg.msg_namelen = 0; | |
3994 | ||
3995 | flags = req->sr_msg.msg_flags; | |
3996 | if (flags & MSG_DONTWAIT) | |
3997 | req->flags |= REQ_F_NOWAIT; | |
3998 | else if (force_nonblock) | |
3999 | flags |= MSG_DONTWAIT; | |
4000 | ||
0b7b21e4 JA |
4001 | msg.msg_flags = flags; |
4002 | ret = sock_sendmsg(sock, &msg); | |
fddaface JA |
4003 | if (force_nonblock && ret == -EAGAIN) |
4004 | return -EAGAIN; | |
4005 | if (ret == -ERESTARTSYS) | |
4006 | ret = -EINTR; | |
4007 | } | |
4008 | ||
fddaface JA |
4009 | if (ret < 0) |
4010 | req_set_fail_links(req); | |
229a7b63 | 4011 | __io_req_complete(req, ret, 0, cs); |
fddaface | 4012 | return 0; |
fddaface JA |
4013 | } |
4014 | ||
1400e697 PB |
4015 | static int __io_recvmsg_copy_hdr(struct io_kiocb *req, |
4016 | struct io_async_msghdr *iomsg) | |
52de1fe1 JA |
4017 | { |
4018 | struct io_sr_msg *sr = &req->sr_msg; | |
4019 | struct iovec __user *uiov; | |
4020 | size_t iov_len; | |
4021 | int ret; | |
4022 | ||
1400e697 PB |
4023 | ret = __copy_msghdr_from_user(&iomsg->msg, sr->umsg, |
4024 | &iomsg->uaddr, &uiov, &iov_len); | |
52de1fe1 JA |
4025 | if (ret) |
4026 | return ret; | |
4027 | ||
4028 | if (req->flags & REQ_F_BUFFER_SELECT) { | |
4029 | if (iov_len > 1) | |
4030 | return -EINVAL; | |
1400e697 | 4031 | if (copy_from_user(iomsg->iov, uiov, sizeof(*uiov))) |
52de1fe1 | 4032 | return -EFAULT; |
1400e697 PB |
4033 | sr->len = iomsg->iov[0].iov_len; |
4034 | iov_iter_init(&iomsg->msg.msg_iter, READ, iomsg->iov, 1, | |
52de1fe1 | 4035 | sr->len); |
1400e697 | 4036 | iomsg->iov = NULL; |
52de1fe1 JA |
4037 | } else { |
4038 | ret = import_iovec(READ, uiov, iov_len, UIO_FASTIOV, | |
1400e697 | 4039 | &iomsg->iov, &iomsg->msg.msg_iter); |
52de1fe1 JA |
4040 | if (ret > 0) |
4041 | ret = 0; | |
4042 | } | |
4043 | ||
4044 | return ret; | |
4045 | } | |
4046 | ||
4047 | #ifdef CONFIG_COMPAT | |
4048 | static int __io_compat_recvmsg_copy_hdr(struct io_kiocb *req, | |
1400e697 | 4049 | struct io_async_msghdr *iomsg) |
52de1fe1 JA |
4050 | { |
4051 | struct compat_msghdr __user *msg_compat; | |
4052 | struct io_sr_msg *sr = &req->sr_msg; | |
4053 | struct compat_iovec __user *uiov; | |
4054 | compat_uptr_t ptr; | |
4055 | compat_size_t len; | |
4056 | int ret; | |
4057 | ||
270a5940 | 4058 | msg_compat = (struct compat_msghdr __user *) sr->umsg; |
1400e697 | 4059 | ret = __get_compat_msghdr(&iomsg->msg, msg_compat, &iomsg->uaddr, |
52de1fe1 JA |
4060 | &ptr, &len); |
4061 | if (ret) | |
4062 | return ret; | |
4063 | ||
4064 | uiov = compat_ptr(ptr); | |
4065 | if (req->flags & REQ_F_BUFFER_SELECT) { | |
4066 | compat_ssize_t clen; | |
4067 | ||
4068 | if (len > 1) | |
4069 | return -EINVAL; | |
4070 | if (!access_ok(uiov, sizeof(*uiov))) | |
4071 | return -EFAULT; | |
4072 | if (__get_user(clen, &uiov->iov_len)) | |
4073 | return -EFAULT; | |
4074 | if (clen < 0) | |
4075 | return -EINVAL; | |
1400e697 PB |
4076 | sr->len = iomsg->iov[0].iov_len; |
4077 | iomsg->iov = NULL; | |
52de1fe1 JA |
4078 | } else { |
4079 | ret = compat_import_iovec(READ, uiov, len, UIO_FASTIOV, | |
1400e697 PB |
4080 | &iomsg->iov, |
4081 | &iomsg->msg.msg_iter); | |
52de1fe1 JA |
4082 | if (ret < 0) |
4083 | return ret; | |
4084 | } | |
4085 | ||
4086 | return 0; | |
4087 | } | |
4088 | #endif | |
4089 | ||
1400e697 PB |
4090 | static int io_recvmsg_copy_hdr(struct io_kiocb *req, |
4091 | struct io_async_msghdr *iomsg) | |
52de1fe1 | 4092 | { |
1400e697 PB |
4093 | iomsg->msg.msg_name = &iomsg->addr; |
4094 | iomsg->iov = iomsg->fast_iov; | |
52de1fe1 JA |
4095 | |
4096 | #ifdef CONFIG_COMPAT | |
4097 | if (req->ctx->compat) | |
1400e697 | 4098 | return __io_compat_recvmsg_copy_hdr(req, iomsg); |
fddaface | 4099 | #endif |
52de1fe1 | 4100 | |
1400e697 | 4101 | return __io_recvmsg_copy_hdr(req, iomsg); |
52de1fe1 JA |
4102 | } |
4103 | ||
bcda7baa JA |
4104 | static struct io_buffer *io_recv_buffer_select(struct io_kiocb *req, |
4105 | int *cflags, bool needs_lock) | |
4106 | { | |
4107 | struct io_sr_msg *sr = &req->sr_msg; | |
4108 | struct io_buffer *kbuf; | |
4109 | ||
4110 | if (!(req->flags & REQ_F_BUFFER_SELECT)) | |
4111 | return NULL; | |
4112 | ||
4113 | kbuf = io_buffer_select(req, &sr->len, sr->bgid, sr->kbuf, needs_lock); | |
4114 | if (IS_ERR(kbuf)) | |
4115 | return kbuf; | |
4116 | ||
4117 | sr->kbuf = kbuf; | |
4118 | req->flags |= REQ_F_BUFFER_SELECTED; | |
4119 | ||
4120 | *cflags = kbuf->bid << IORING_CQE_BUFFER_SHIFT; | |
4121 | *cflags |= IORING_CQE_F_BUFFER; | |
4122 | return kbuf; | |
fddaface JA |
4123 | } |
4124 | ||
3529d8c2 JA |
4125 | static int io_recvmsg_prep(struct io_kiocb *req, |
4126 | const struct io_uring_sqe *sqe) | |
aa1fa28f | 4127 | { |
e47293fd | 4128 | struct io_sr_msg *sr = &req->sr_msg; |
3529d8c2 | 4129 | struct io_async_ctx *io = req->io; |
99bc4c38 | 4130 | int ret; |
3529d8c2 | 4131 | |
d2b6f48b PB |
4132 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4133 | return -EINVAL; | |
4134 | ||
3529d8c2 | 4135 | sr->msg_flags = READ_ONCE(sqe->msg_flags); |
270a5940 | 4136 | sr->umsg = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
0b7b21e4 | 4137 | sr->len = READ_ONCE(sqe->len); |
bcda7baa | 4138 | sr->bgid = READ_ONCE(sqe->buf_group); |
06b76d44 | 4139 | |
d8768362 JA |
4140 | #ifdef CONFIG_COMPAT |
4141 | if (req->ctx->compat) | |
4142 | sr->msg_flags |= MSG_CMSG_COMPAT; | |
4143 | #endif | |
4144 | ||
fddaface | 4145 | if (!io || req->opcode == IORING_OP_RECV) |
06b76d44 | 4146 | return 0; |
5f798bea PB |
4147 | /* iovec is already imported */ |
4148 | if (req->flags & REQ_F_NEED_CLEANUP) | |
4149 | return 0; | |
03b1230c | 4150 | |
1400e697 | 4151 | ret = io_recvmsg_copy_hdr(req, &io->msg); |
99bc4c38 PB |
4152 | if (!ret) |
4153 | req->flags |= REQ_F_NEED_CLEANUP; | |
4154 | return ret; | |
aa1fa28f JA |
4155 | } |
4156 | ||
229a7b63 JA |
4157 | static int io_recvmsg(struct io_kiocb *req, bool force_nonblock, |
4158 | struct io_comp_state *cs) | |
aa1fa28f | 4159 | { |
0b416c3e | 4160 | struct io_async_msghdr *kmsg = NULL; |
03b1230c | 4161 | struct socket *sock; |
52de1fe1 | 4162 | int ret, cflags = 0; |
03b1230c | 4163 | |
03b1230c JA |
4164 | sock = sock_from_file(req->file, &ret); |
4165 | if (sock) { | |
52de1fe1 | 4166 | struct io_buffer *kbuf; |
1400e697 | 4167 | struct io_async_msghdr iomsg; |
03b1230c JA |
4168 | unsigned flags; |
4169 | ||
03b1230c | 4170 | if (req->io) { |
0b416c3e | 4171 | kmsg = &req->io->msg; |
b537916c | 4172 | kmsg->msg.msg_name = &req->io->msg.addr; |
0b416c3e JA |
4173 | /* if iov is set, it's allocated already */ |
4174 | if (!kmsg->iov) | |
4175 | kmsg->iov = kmsg->fast_iov; | |
4176 | kmsg->msg.msg_iter.iov = kmsg->iov; | |
03b1230c | 4177 | } else { |
1400e697 | 4178 | ret = io_recvmsg_copy_hdr(req, &iomsg); |
03b1230c | 4179 | if (ret) |
3529d8c2 | 4180 | return ret; |
1400e697 | 4181 | kmsg = &iomsg; |
03b1230c JA |
4182 | } |
4183 | ||
52de1fe1 JA |
4184 | kbuf = io_recv_buffer_select(req, &cflags, !force_nonblock); |
4185 | if (IS_ERR(kbuf)) { | |
4186 | return PTR_ERR(kbuf); | |
4187 | } else if (kbuf) { | |
4188 | kmsg->fast_iov[0].iov_base = u64_to_user_ptr(kbuf->addr); | |
4189 | iov_iter_init(&kmsg->msg.msg_iter, READ, kmsg->iov, | |
4190 | 1, req->sr_msg.len); | |
4191 | } | |
4192 | ||
e47293fd JA |
4193 | flags = req->sr_msg.msg_flags; |
4194 | if (flags & MSG_DONTWAIT) | |
4195 | req->flags |= REQ_F_NOWAIT; | |
4196 | else if (force_nonblock) | |
4197 | flags |= MSG_DONTWAIT; | |
4198 | ||
270a5940 | 4199 | ret = __sys_recvmsg_sock(sock, &kmsg->msg, req->sr_msg.umsg, |
e47293fd | 4200 | kmsg->uaddr, flags); |
681fda8d PB |
4201 | if (force_nonblock && ret == -EAGAIN) { |
4202 | ret = io_setup_async_msg(req, kmsg); | |
4203 | if (ret != -EAGAIN) | |
4204 | kfree(kbuf); | |
4205 | return ret; | |
4206 | } | |
03b1230c JA |
4207 | if (ret == -ERESTARTSYS) |
4208 | ret = -EINTR; | |
681fda8d PB |
4209 | if (kbuf) |
4210 | kfree(kbuf); | |
03b1230c JA |
4211 | } |
4212 | ||
1e95081c | 4213 | if (kmsg && kmsg->iov != kmsg->fast_iov) |
0b416c3e | 4214 | kfree(kmsg->iov); |
99bc4c38 | 4215 | req->flags &= ~REQ_F_NEED_CLEANUP; |
4e88d6e7 JA |
4216 | if (ret < 0) |
4217 | req_set_fail_links(req); | |
229a7b63 | 4218 | __io_req_complete(req, ret, cflags, cs); |
03b1230c | 4219 | return 0; |
0fa03c62 | 4220 | } |
5d17b4a4 | 4221 | |
229a7b63 JA |
4222 | static int io_recv(struct io_kiocb *req, bool force_nonblock, |
4223 | struct io_comp_state *cs) | |
fddaface | 4224 | { |
bcda7baa | 4225 | struct io_buffer *kbuf = NULL; |
fddaface | 4226 | struct socket *sock; |
bcda7baa | 4227 | int ret, cflags = 0; |
fddaface | 4228 | |
fddaface JA |
4229 | sock = sock_from_file(req->file, &ret); |
4230 | if (sock) { | |
4231 | struct io_sr_msg *sr = &req->sr_msg; | |
bcda7baa | 4232 | void __user *buf = sr->buf; |
fddaface JA |
4233 | struct msghdr msg; |
4234 | struct iovec iov; | |
4235 | unsigned flags; | |
4236 | ||
bcda7baa JA |
4237 | kbuf = io_recv_buffer_select(req, &cflags, !force_nonblock); |
4238 | if (IS_ERR(kbuf)) | |
4239 | return PTR_ERR(kbuf); | |
4240 | else if (kbuf) | |
4241 | buf = u64_to_user_ptr(kbuf->addr); | |
4242 | ||
4243 | ret = import_single_range(READ, buf, sr->len, &iov, | |
fddaface | 4244 | &msg.msg_iter); |
bcda7baa JA |
4245 | if (ret) { |
4246 | kfree(kbuf); | |
fddaface | 4247 | return ret; |
bcda7baa | 4248 | } |
fddaface | 4249 | |
bcda7baa | 4250 | req->flags |= REQ_F_NEED_CLEANUP; |
fddaface JA |
4251 | msg.msg_name = NULL; |
4252 | msg.msg_control = NULL; | |
4253 | msg.msg_controllen = 0; | |
4254 | msg.msg_namelen = 0; | |
4255 | msg.msg_iocb = NULL; | |
4256 | msg.msg_flags = 0; | |
4257 | ||
4258 | flags = req->sr_msg.msg_flags; | |
4259 | if (flags & MSG_DONTWAIT) | |
4260 | req->flags |= REQ_F_NOWAIT; | |
4261 | else if (force_nonblock) | |
4262 | flags |= MSG_DONTWAIT; | |
4263 | ||
0b7b21e4 | 4264 | ret = sock_recvmsg(sock, &msg, flags); |
fddaface JA |
4265 | if (force_nonblock && ret == -EAGAIN) |
4266 | return -EAGAIN; | |
4267 | if (ret == -ERESTARTSYS) | |
4268 | ret = -EINTR; | |
4269 | } | |
4270 | ||
bcda7baa JA |
4271 | kfree(kbuf); |
4272 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
fddaface JA |
4273 | if (ret < 0) |
4274 | req_set_fail_links(req); | |
229a7b63 | 4275 | __io_req_complete(req, ret, cflags, cs); |
fddaface | 4276 | return 0; |
fddaface JA |
4277 | } |
4278 | ||
3529d8c2 | 4279 | static int io_accept_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
17f2fe35 | 4280 | { |
8ed8d3c3 JA |
4281 | struct io_accept *accept = &req->accept; |
4282 | ||
17f2fe35 JA |
4283 | if (unlikely(req->ctx->flags & (IORING_SETUP_IOPOLL|IORING_SETUP_SQPOLL))) |
4284 | return -EINVAL; | |
8042d6ce | 4285 | if (sqe->ioprio || sqe->len || sqe->buf_index) |
17f2fe35 JA |
4286 | return -EINVAL; |
4287 | ||
d55e5f5b JA |
4288 | accept->addr = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
4289 | accept->addr_len = u64_to_user_ptr(READ_ONCE(sqe->addr2)); | |
8ed8d3c3 | 4290 | accept->flags = READ_ONCE(sqe->accept_flags); |
09952e3e | 4291 | accept->nofile = rlimit(RLIMIT_NOFILE); |
8ed8d3c3 | 4292 | return 0; |
8ed8d3c3 | 4293 | } |
17f2fe35 | 4294 | |
229a7b63 JA |
4295 | static int io_accept(struct io_kiocb *req, bool force_nonblock, |
4296 | struct io_comp_state *cs) | |
8ed8d3c3 JA |
4297 | { |
4298 | struct io_accept *accept = &req->accept; | |
ac45abc0 | 4299 | unsigned int file_flags = force_nonblock ? O_NONBLOCK : 0; |
8ed8d3c3 JA |
4300 | int ret; |
4301 | ||
e697deed JX |
4302 | if (req->file->f_flags & O_NONBLOCK) |
4303 | req->flags |= REQ_F_NOWAIT; | |
4304 | ||
8ed8d3c3 | 4305 | ret = __sys_accept4_file(req->file, file_flags, accept->addr, |
09952e3e JA |
4306 | accept->addr_len, accept->flags, |
4307 | accept->nofile); | |
8ed8d3c3 | 4308 | if (ret == -EAGAIN && force_nonblock) |
17f2fe35 | 4309 | return -EAGAIN; |
ac45abc0 PB |
4310 | if (ret < 0) { |
4311 | if (ret == -ERESTARTSYS) | |
4312 | ret = -EINTR; | |
4e88d6e7 | 4313 | req_set_fail_links(req); |
ac45abc0 | 4314 | } |
229a7b63 | 4315 | __io_req_complete(req, ret, 0, cs); |
17f2fe35 | 4316 | return 0; |
8ed8d3c3 JA |
4317 | } |
4318 | ||
3529d8c2 | 4319 | static int io_connect_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
f499a021 | 4320 | { |
3529d8c2 JA |
4321 | struct io_connect *conn = &req->connect; |
4322 | struct io_async_ctx *io = req->io; | |
f499a021 | 4323 | |
3fbb51c1 JA |
4324 | if (unlikely(req->ctx->flags & (IORING_SETUP_IOPOLL|IORING_SETUP_SQPOLL))) |
4325 | return -EINVAL; | |
4326 | if (sqe->ioprio || sqe->len || sqe->buf_index || sqe->rw_flags) | |
4327 | return -EINVAL; | |
4328 | ||
3529d8c2 JA |
4329 | conn->addr = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
4330 | conn->addr_len = READ_ONCE(sqe->addr2); | |
4331 | ||
4332 | if (!io) | |
4333 | return 0; | |
4334 | ||
4335 | return move_addr_to_kernel(conn->addr, conn->addr_len, | |
3fbb51c1 | 4336 | &io->connect.address); |
f499a021 JA |
4337 | } |
4338 | ||
229a7b63 JA |
4339 | static int io_connect(struct io_kiocb *req, bool force_nonblock, |
4340 | struct io_comp_state *cs) | |
f8e85cf2 | 4341 | { |
f499a021 | 4342 | struct io_async_ctx __io, *io; |
f8e85cf2 | 4343 | unsigned file_flags; |
3fbb51c1 | 4344 | int ret; |
f8e85cf2 | 4345 | |
f499a021 JA |
4346 | if (req->io) { |
4347 | io = req->io; | |
4348 | } else { | |
3529d8c2 JA |
4349 | ret = move_addr_to_kernel(req->connect.addr, |
4350 | req->connect.addr_len, | |
4351 | &__io.connect.address); | |
f499a021 JA |
4352 | if (ret) |
4353 | goto out; | |
4354 | io = &__io; | |
4355 | } | |
4356 | ||
3fbb51c1 JA |
4357 | file_flags = force_nonblock ? O_NONBLOCK : 0; |
4358 | ||
4359 | ret = __sys_connect_file(req->file, &io->connect.address, | |
4360 | req->connect.addr_len, file_flags); | |
87f80d62 | 4361 | if ((ret == -EAGAIN || ret == -EINPROGRESS) && force_nonblock) { |
b7bb4f7d JA |
4362 | if (req->io) |
4363 | return -EAGAIN; | |
4364 | if (io_alloc_async_ctx(req)) { | |
f499a021 JA |
4365 | ret = -ENOMEM; |
4366 | goto out; | |
4367 | } | |
b7bb4f7d | 4368 | memcpy(&req->io->connect, &__io.connect, sizeof(__io.connect)); |
f8e85cf2 | 4369 | return -EAGAIN; |
f499a021 | 4370 | } |
f8e85cf2 JA |
4371 | if (ret == -ERESTARTSYS) |
4372 | ret = -EINTR; | |
f499a021 | 4373 | out: |
4e88d6e7 JA |
4374 | if (ret < 0) |
4375 | req_set_fail_links(req); | |
229a7b63 | 4376 | __io_req_complete(req, ret, 0, cs); |
f8e85cf2 | 4377 | return 0; |
469956e8 Y |
4378 | } |
4379 | #else /* !CONFIG_NET */ | |
4380 | static int io_sendmsg_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) | |
4381 | { | |
f8e85cf2 | 4382 | return -EOPNOTSUPP; |
f8e85cf2 JA |
4383 | } |
4384 | ||
1e16c2f9 RD |
4385 | static int io_sendmsg(struct io_kiocb *req, bool force_nonblock, |
4386 | struct io_comp_state *cs) | |
469956e8 Y |
4387 | { |
4388 | return -EOPNOTSUPP; | |
4389 | } | |
4390 | ||
1e16c2f9 RD |
4391 | static int io_send(struct io_kiocb *req, bool force_nonblock, |
4392 | struct io_comp_state *cs) | |
469956e8 Y |
4393 | { |
4394 | return -EOPNOTSUPP; | |
4395 | } | |
4396 | ||
4397 | static int io_recvmsg_prep(struct io_kiocb *req, | |
4398 | const struct io_uring_sqe *sqe) | |
4399 | { | |
4400 | return -EOPNOTSUPP; | |
4401 | } | |
4402 | ||
1e16c2f9 RD |
4403 | static int io_recvmsg(struct io_kiocb *req, bool force_nonblock, |
4404 | struct io_comp_state *cs) | |
469956e8 Y |
4405 | { |
4406 | return -EOPNOTSUPP; | |
4407 | } | |
4408 | ||
1e16c2f9 RD |
4409 | static int io_recv(struct io_kiocb *req, bool force_nonblock, |
4410 | struct io_comp_state *cs) | |
469956e8 Y |
4411 | { |
4412 | return -EOPNOTSUPP; | |
4413 | } | |
4414 | ||
4415 | static int io_accept_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) | |
4416 | { | |
4417 | return -EOPNOTSUPP; | |
4418 | } | |
4419 | ||
1e16c2f9 RD |
4420 | static int io_accept(struct io_kiocb *req, bool force_nonblock, |
4421 | struct io_comp_state *cs) | |
469956e8 Y |
4422 | { |
4423 | return -EOPNOTSUPP; | |
4424 | } | |
4425 | ||
4426 | static int io_connect_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) | |
4427 | { | |
4428 | return -EOPNOTSUPP; | |
4429 | } | |
4430 | ||
1e16c2f9 RD |
4431 | static int io_connect(struct io_kiocb *req, bool force_nonblock, |
4432 | struct io_comp_state *cs) | |
469956e8 | 4433 | { |
f8e85cf2 | 4434 | return -EOPNOTSUPP; |
f8e85cf2 | 4435 | } |
469956e8 | 4436 | #endif /* CONFIG_NET */ |
f8e85cf2 | 4437 | |
d7718a9d JA |
4438 | struct io_poll_table { |
4439 | struct poll_table_struct pt; | |
4440 | struct io_kiocb *req; | |
4441 | int error; | |
4442 | }; | |
4443 | ||
d7718a9d JA |
4444 | static int __io_async_wake(struct io_kiocb *req, struct io_poll_iocb *poll, |
4445 | __poll_t mask, task_work_func_t func) | |
4446 | { | |
aa96bf8a | 4447 | int ret; |
d7718a9d JA |
4448 | |
4449 | /* for instances that support it check for an event match first: */ | |
4450 | if (mask && !(mask & poll->events)) | |
4451 | return 0; | |
4452 | ||
4453 | trace_io_uring_task_add(req->ctx, req->opcode, req->user_data, mask); | |
4454 | ||
4455 | list_del_init(&poll->wait.entry); | |
4456 | ||
d7718a9d JA |
4457 | req->result = mask; |
4458 | init_task_work(&req->task_work, func); | |
4459 | /* | |
e3aabf95 JA |
4460 | * If this fails, then the task is exiting. When a task exits, the |
4461 | * work gets canceled, so just cancel this request as well instead | |
4462 | * of executing it. We can't safely execute it anyway, as we may not | |
4463 | * have the needed state needed for it anyway. | |
d7718a9d | 4464 | */ |
b7db41c9 | 4465 | ret = io_req_task_work_add(req, &req->task_work); |
aa96bf8a | 4466 | if (unlikely(ret)) { |
c2c4c83c JA |
4467 | struct task_struct *tsk; |
4468 | ||
e3aabf95 | 4469 | WRITE_ONCE(poll->canceled, true); |
aa96bf8a | 4470 | tsk = io_wq_get_task(req->ctx->io_wq); |
ce593a6c JA |
4471 | task_work_add(tsk, &req->task_work, 0); |
4472 | wake_up_process(tsk); | |
aa96bf8a | 4473 | } |
d7718a9d JA |
4474 | return 1; |
4475 | } | |
4476 | ||
74ce6ce4 JA |
4477 | static bool io_poll_rewait(struct io_kiocb *req, struct io_poll_iocb *poll) |
4478 | __acquires(&req->ctx->completion_lock) | |
4479 | { | |
4480 | struct io_ring_ctx *ctx = req->ctx; | |
4481 | ||
4482 | if (!req->result && !READ_ONCE(poll->canceled)) { | |
4483 | struct poll_table_struct pt = { ._key = poll->events }; | |
4484 | ||
4485 | req->result = vfs_poll(req->file, &pt) & poll->events; | |
4486 | } | |
4487 | ||
4488 | spin_lock_irq(&ctx->completion_lock); | |
4489 | if (!req->result && !READ_ONCE(poll->canceled)) { | |
4490 | add_wait_queue(poll->head, &poll->wait); | |
4491 | return true; | |
4492 | } | |
4493 | ||
4494 | return false; | |
4495 | } | |
4496 | ||
807abcb0 | 4497 | static void io_poll_remove_double(struct io_kiocb *req, void *data) |
18bceab1 | 4498 | { |
807abcb0 | 4499 | struct io_poll_iocb *poll = data; |
18bceab1 JA |
4500 | |
4501 | lockdep_assert_held(&req->ctx->completion_lock); | |
4502 | ||
4503 | if (poll && poll->head) { | |
4504 | struct wait_queue_head *head = poll->head; | |
4505 | ||
4506 | spin_lock(&head->lock); | |
4507 | list_del_init(&poll->wait.entry); | |
4508 | if (poll->wait.private) | |
4509 | refcount_dec(&req->refs); | |
4510 | poll->head = NULL; | |
4511 | spin_unlock(&head->lock); | |
4512 | } | |
4513 | } | |
4514 | ||
4515 | static void io_poll_complete(struct io_kiocb *req, __poll_t mask, int error) | |
4516 | { | |
4517 | struct io_ring_ctx *ctx = req->ctx; | |
4518 | ||
807abcb0 | 4519 | io_poll_remove_double(req, req->io); |
18bceab1 JA |
4520 | req->poll.done = true; |
4521 | io_cqring_fill_event(req, error ? error : mangle_poll(mask)); | |
4522 | io_commit_cqring(ctx); | |
4523 | } | |
4524 | ||
4525 | static void io_poll_task_handler(struct io_kiocb *req, struct io_kiocb **nxt) | |
4526 | { | |
4527 | struct io_ring_ctx *ctx = req->ctx; | |
4528 | ||
4529 | if (io_poll_rewait(req, &req->poll)) { | |
4530 | spin_unlock_irq(&ctx->completion_lock); | |
4531 | return; | |
4532 | } | |
4533 | ||
4534 | hash_del(&req->hash_node); | |
4535 | io_poll_complete(req, req->result, 0); | |
4536 | req->flags |= REQ_F_COMP_LOCKED; | |
9b5f7bd9 | 4537 | *nxt = io_put_req_find_next(req); |
18bceab1 JA |
4538 | spin_unlock_irq(&ctx->completion_lock); |
4539 | ||
4540 | io_cqring_ev_posted(ctx); | |
4541 | } | |
4542 | ||
4543 | static void io_poll_task_func(struct callback_head *cb) | |
4544 | { | |
4545 | struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work); | |
4546 | struct io_kiocb *nxt = NULL; | |
4547 | ||
4548 | io_poll_task_handler(req, &nxt); | |
ea1164e5 PB |
4549 | if (nxt) |
4550 | __io_req_task_submit(nxt); | |
18bceab1 JA |
4551 | } |
4552 | ||
4553 | static int io_poll_double_wake(struct wait_queue_entry *wait, unsigned mode, | |
4554 | int sync, void *key) | |
4555 | { | |
4556 | struct io_kiocb *req = wait->private; | |
807abcb0 | 4557 | struct io_poll_iocb *poll = req->apoll->double_poll; |
18bceab1 JA |
4558 | __poll_t mask = key_to_poll(key); |
4559 | ||
4560 | /* for instances that support it check for an event match first: */ | |
4561 | if (mask && !(mask & poll->events)) | |
4562 | return 0; | |
4563 | ||
807abcb0 | 4564 | if (poll && poll->head) { |
18bceab1 JA |
4565 | bool done; |
4566 | ||
807abcb0 JA |
4567 | spin_lock(&poll->head->lock); |
4568 | done = list_empty(&poll->wait.entry); | |
18bceab1 | 4569 | if (!done) |
807abcb0 JA |
4570 | list_del_init(&poll->wait.entry); |
4571 | spin_unlock(&poll->head->lock); | |
18bceab1 JA |
4572 | if (!done) |
4573 | __io_async_wake(req, poll, mask, io_poll_task_func); | |
4574 | } | |
4575 | refcount_dec(&req->refs); | |
4576 | return 1; | |
4577 | } | |
4578 | ||
4579 | static void io_init_poll_iocb(struct io_poll_iocb *poll, __poll_t events, | |
4580 | wait_queue_func_t wake_func) | |
4581 | { | |
4582 | poll->head = NULL; | |
4583 | poll->done = false; | |
4584 | poll->canceled = false; | |
4585 | poll->events = events; | |
4586 | INIT_LIST_HEAD(&poll->wait.entry); | |
4587 | init_waitqueue_func_entry(&poll->wait, wake_func); | |
4588 | } | |
4589 | ||
4590 | static void __io_queue_proc(struct io_poll_iocb *poll, struct io_poll_table *pt, | |
807abcb0 JA |
4591 | struct wait_queue_head *head, |
4592 | struct io_poll_iocb **poll_ptr) | |
18bceab1 JA |
4593 | { |
4594 | struct io_kiocb *req = pt->req; | |
4595 | ||
4596 | /* | |
4597 | * If poll->head is already set, it's because the file being polled | |
4598 | * uses multiple waitqueues for poll handling (eg one for read, one | |
4599 | * for write). Setup a separate io_poll_iocb if this happens. | |
4600 | */ | |
4601 | if (unlikely(poll->head)) { | |
4602 | /* already have a 2nd entry, fail a third attempt */ | |
807abcb0 | 4603 | if (*poll_ptr) { |
18bceab1 JA |
4604 | pt->error = -EINVAL; |
4605 | return; | |
4606 | } | |
4607 | poll = kmalloc(sizeof(*poll), GFP_ATOMIC); | |
4608 | if (!poll) { | |
4609 | pt->error = -ENOMEM; | |
4610 | return; | |
4611 | } | |
4612 | io_init_poll_iocb(poll, req->poll.events, io_poll_double_wake); | |
4613 | refcount_inc(&req->refs); | |
4614 | poll->wait.private = req; | |
807abcb0 | 4615 | *poll_ptr = poll; |
18bceab1 JA |
4616 | } |
4617 | ||
4618 | pt->error = 0; | |
4619 | poll->head = head; | |
a31eb4a2 JX |
4620 | |
4621 | if (poll->events & EPOLLEXCLUSIVE) | |
4622 | add_wait_queue_exclusive(head, &poll->wait); | |
4623 | else | |
4624 | add_wait_queue(head, &poll->wait); | |
18bceab1 JA |
4625 | } |
4626 | ||
4627 | static void io_async_queue_proc(struct file *file, struct wait_queue_head *head, | |
4628 | struct poll_table_struct *p) | |
4629 | { | |
4630 | struct io_poll_table *pt = container_of(p, struct io_poll_table, pt); | |
807abcb0 | 4631 | struct async_poll *apoll = pt->req->apoll; |
18bceab1 | 4632 | |
807abcb0 | 4633 | __io_queue_proc(&apoll->poll, pt, head, &apoll->double_poll); |
18bceab1 JA |
4634 | } |
4635 | ||
d7718a9d JA |
4636 | static void io_async_task_func(struct callback_head *cb) |
4637 | { | |
4638 | struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work); | |
4639 | struct async_poll *apoll = req->apoll; | |
4640 | struct io_ring_ctx *ctx = req->ctx; | |
4641 | ||
4642 | trace_io_uring_task_run(req->ctx, req->opcode, req->user_data); | |
4643 | ||
74ce6ce4 | 4644 | if (io_poll_rewait(req, &apoll->poll)) { |
d7718a9d | 4645 | spin_unlock_irq(&ctx->completion_lock); |
74ce6ce4 | 4646 | return; |
d7718a9d JA |
4647 | } |
4648 | ||
31067255 | 4649 | /* If req is still hashed, it cannot have been canceled. Don't check. */ |
0be0b0e3 | 4650 | if (hash_hashed(&req->hash_node)) |
74ce6ce4 | 4651 | hash_del(&req->hash_node); |
2bae047e | 4652 | |
807abcb0 | 4653 | io_poll_remove_double(req, apoll->double_poll); |
74ce6ce4 JA |
4654 | spin_unlock_irq(&ctx->completion_lock); |
4655 | ||
44575a67 | 4656 | /* restore ->work in case we need to retry again */ |
405a5d2b XW |
4657 | if (req->flags & REQ_F_WORK_INITIALIZED) |
4658 | memcpy(&req->work, &apoll->work, sizeof(req->work)); | |
44575a67 | 4659 | |
0be0b0e3 PB |
4660 | if (!READ_ONCE(apoll->poll.canceled)) |
4661 | __io_req_task_submit(req); | |
4662 | else | |
4663 | __io_req_task_cancel(req, -ECANCELED); | |
aa340845 | 4664 | |
807abcb0 | 4665 | kfree(apoll->double_poll); |
aa340845 | 4666 | kfree(apoll); |
d7718a9d JA |
4667 | } |
4668 | ||
4669 | static int io_async_wake(struct wait_queue_entry *wait, unsigned mode, int sync, | |
4670 | void *key) | |
4671 | { | |
4672 | struct io_kiocb *req = wait->private; | |
4673 | struct io_poll_iocb *poll = &req->apoll->poll; | |
4674 | ||
4675 | trace_io_uring_poll_wake(req->ctx, req->opcode, req->user_data, | |
4676 | key_to_poll(key)); | |
4677 | ||
4678 | return __io_async_wake(req, poll, key_to_poll(key), io_async_task_func); | |
4679 | } | |
4680 | ||
4681 | static void io_poll_req_insert(struct io_kiocb *req) | |
4682 | { | |
4683 | struct io_ring_ctx *ctx = req->ctx; | |
4684 | struct hlist_head *list; | |
4685 | ||
4686 | list = &ctx->cancel_hash[hash_long(req->user_data, ctx->cancel_hash_bits)]; | |
4687 | hlist_add_head(&req->hash_node, list); | |
4688 | } | |
4689 | ||
4690 | static __poll_t __io_arm_poll_handler(struct io_kiocb *req, | |
4691 | struct io_poll_iocb *poll, | |
4692 | struct io_poll_table *ipt, __poll_t mask, | |
4693 | wait_queue_func_t wake_func) | |
4694 | __acquires(&ctx->completion_lock) | |
4695 | { | |
4696 | struct io_ring_ctx *ctx = req->ctx; | |
4697 | bool cancel = false; | |
4698 | ||
18bceab1 | 4699 | io_init_poll_iocb(poll, mask, wake_func); |
b90cd197 | 4700 | poll->file = req->file; |
18bceab1 | 4701 | poll->wait.private = req; |
d7718a9d JA |
4702 | |
4703 | ipt->pt._key = mask; | |
4704 | ipt->req = req; | |
4705 | ipt->error = -EINVAL; | |
4706 | ||
d7718a9d JA |
4707 | mask = vfs_poll(req->file, &ipt->pt) & poll->events; |
4708 | ||
4709 | spin_lock_irq(&ctx->completion_lock); | |
4710 | if (likely(poll->head)) { | |
4711 | spin_lock(&poll->head->lock); | |
4712 | if (unlikely(list_empty(&poll->wait.entry))) { | |
4713 | if (ipt->error) | |
4714 | cancel = true; | |
4715 | ipt->error = 0; | |
4716 | mask = 0; | |
4717 | } | |
4718 | if (mask || ipt->error) | |
4719 | list_del_init(&poll->wait.entry); | |
4720 | else if (cancel) | |
4721 | WRITE_ONCE(poll->canceled, true); | |
4722 | else if (!poll->done) /* actually waiting for an event */ | |
4723 | io_poll_req_insert(req); | |
4724 | spin_unlock(&poll->head->lock); | |
4725 | } | |
4726 | ||
4727 | return mask; | |
4728 | } | |
4729 | ||
4730 | static bool io_arm_poll_handler(struct io_kiocb *req) | |
4731 | { | |
4732 | const struct io_op_def *def = &io_op_defs[req->opcode]; | |
4733 | struct io_ring_ctx *ctx = req->ctx; | |
4734 | struct async_poll *apoll; | |
4735 | struct io_poll_table ipt; | |
4736 | __poll_t mask, ret; | |
4737 | ||
4738 | if (!req->file || !file_can_poll(req->file)) | |
4739 | return false; | |
24c74678 | 4740 | if (req->flags & REQ_F_POLLED) |
d7718a9d JA |
4741 | return false; |
4742 | if (!def->pollin && !def->pollout) | |
4743 | return false; | |
4744 | ||
4745 | apoll = kmalloc(sizeof(*apoll), GFP_ATOMIC); | |
4746 | if (unlikely(!apoll)) | |
4747 | return false; | |
807abcb0 | 4748 | apoll->double_poll = NULL; |
d7718a9d JA |
4749 | |
4750 | req->flags |= REQ_F_POLLED; | |
405a5d2b XW |
4751 | if (req->flags & REQ_F_WORK_INITIALIZED) |
4752 | memcpy(&apoll->work, &req->work, sizeof(req->work)); | |
d7718a9d | 4753 | |
4dd2824d | 4754 | io_get_req_task(req); |
d7718a9d JA |
4755 | req->apoll = apoll; |
4756 | INIT_HLIST_NODE(&req->hash_node); | |
4757 | ||
8755d97a | 4758 | mask = 0; |
d7718a9d | 4759 | if (def->pollin) |
8755d97a | 4760 | mask |= POLLIN | POLLRDNORM; |
d7718a9d JA |
4761 | if (def->pollout) |
4762 | mask |= POLLOUT | POLLWRNORM; | |
4763 | mask |= POLLERR | POLLPRI; | |
4764 | ||
4765 | ipt.pt._qproc = io_async_queue_proc; | |
4766 | ||
4767 | ret = __io_arm_poll_handler(req, &apoll->poll, &ipt, mask, | |
4768 | io_async_wake); | |
4769 | if (ret) { | |
807abcb0 | 4770 | io_poll_remove_double(req, apoll->double_poll); |
d7718a9d | 4771 | spin_unlock_irq(&ctx->completion_lock); |
405a5d2b XW |
4772 | if (req->flags & REQ_F_WORK_INITIALIZED) |
4773 | memcpy(&req->work, &apoll->work, sizeof(req->work)); | |
807abcb0 | 4774 | kfree(apoll->double_poll); |
d7718a9d JA |
4775 | kfree(apoll); |
4776 | return false; | |
4777 | } | |
4778 | spin_unlock_irq(&ctx->completion_lock); | |
4779 | trace_io_uring_poll_arm(ctx, req->opcode, req->user_data, mask, | |
4780 | apoll->poll.events); | |
4781 | return true; | |
4782 | } | |
4783 | ||
4784 | static bool __io_poll_remove_one(struct io_kiocb *req, | |
4785 | struct io_poll_iocb *poll) | |
221c5eb2 | 4786 | { |
b41e9852 | 4787 | bool do_complete = false; |
221c5eb2 JA |
4788 | |
4789 | spin_lock(&poll->head->lock); | |
4790 | WRITE_ONCE(poll->canceled, true); | |
392edb45 JA |
4791 | if (!list_empty(&poll->wait.entry)) { |
4792 | list_del_init(&poll->wait.entry); | |
b41e9852 | 4793 | do_complete = true; |
221c5eb2 JA |
4794 | } |
4795 | spin_unlock(&poll->head->lock); | |
3bfa5bcb | 4796 | hash_del(&req->hash_node); |
d7718a9d JA |
4797 | return do_complete; |
4798 | } | |
4799 | ||
4800 | static bool io_poll_remove_one(struct io_kiocb *req) | |
4801 | { | |
4802 | bool do_complete; | |
4803 | ||
4804 | if (req->opcode == IORING_OP_POLL_ADD) { | |
807abcb0 | 4805 | io_poll_remove_double(req, req->io); |
d7718a9d JA |
4806 | do_complete = __io_poll_remove_one(req, &req->poll); |
4807 | } else { | |
3bfa5bcb JA |
4808 | struct async_poll *apoll = req->apoll; |
4809 | ||
807abcb0 JA |
4810 | io_poll_remove_double(req, apoll->double_poll); |
4811 | ||
d7718a9d | 4812 | /* non-poll requests have submit ref still */ |
3bfa5bcb JA |
4813 | do_complete = __io_poll_remove_one(req, &apoll->poll); |
4814 | if (do_complete) { | |
d7718a9d | 4815 | io_put_req(req); |
3bfa5bcb JA |
4816 | /* |
4817 | * restore ->work because we will call | |
dca9cf8b | 4818 | * io_req_clean_work below when dropping the |
3bfa5bcb JA |
4819 | * final reference. |
4820 | */ | |
405a5d2b XW |
4821 | if (req->flags & REQ_F_WORK_INITIALIZED) |
4822 | memcpy(&req->work, &apoll->work, | |
4823 | sizeof(req->work)); | |
807abcb0 | 4824 | kfree(apoll->double_poll); |
3bfa5bcb JA |
4825 | kfree(apoll); |
4826 | } | |
b1f573bd XW |
4827 | } |
4828 | ||
b41e9852 JA |
4829 | if (do_complete) { |
4830 | io_cqring_fill_event(req, -ECANCELED); | |
4831 | io_commit_cqring(req->ctx); | |
4832 | req->flags |= REQ_F_COMP_LOCKED; | |
4833 | io_put_req(req); | |
4834 | } | |
4835 | ||
4836 | return do_complete; | |
221c5eb2 JA |
4837 | } |
4838 | ||
4839 | static void io_poll_remove_all(struct io_ring_ctx *ctx) | |
4840 | { | |
78076bb6 | 4841 | struct hlist_node *tmp; |
221c5eb2 | 4842 | struct io_kiocb *req; |
8e2e1faf | 4843 | int posted = 0, i; |
221c5eb2 JA |
4844 | |
4845 | spin_lock_irq(&ctx->completion_lock); | |
78076bb6 JA |
4846 | for (i = 0; i < (1U << ctx->cancel_hash_bits); i++) { |
4847 | struct hlist_head *list; | |
4848 | ||
4849 | list = &ctx->cancel_hash[i]; | |
4850 | hlist_for_each_entry_safe(req, tmp, list, hash_node) | |
8e2e1faf | 4851 | posted += io_poll_remove_one(req); |
221c5eb2 JA |
4852 | } |
4853 | spin_unlock_irq(&ctx->completion_lock); | |
b41e9852 | 4854 | |
8e2e1faf JA |
4855 | if (posted) |
4856 | io_cqring_ev_posted(ctx); | |
221c5eb2 JA |
4857 | } |
4858 | ||
47f46768 JA |
4859 | static int io_poll_cancel(struct io_ring_ctx *ctx, __u64 sqe_addr) |
4860 | { | |
78076bb6 | 4861 | struct hlist_head *list; |
47f46768 JA |
4862 | struct io_kiocb *req; |
4863 | ||
78076bb6 JA |
4864 | list = &ctx->cancel_hash[hash_long(sqe_addr, ctx->cancel_hash_bits)]; |
4865 | hlist_for_each_entry(req, list, hash_node) { | |
b41e9852 JA |
4866 | if (sqe_addr != req->user_data) |
4867 | continue; | |
4868 | if (io_poll_remove_one(req)) | |
eac406c6 | 4869 | return 0; |
b41e9852 | 4870 | return -EALREADY; |
47f46768 JA |
4871 | } |
4872 | ||
4873 | return -ENOENT; | |
4874 | } | |
4875 | ||
3529d8c2 JA |
4876 | static int io_poll_remove_prep(struct io_kiocb *req, |
4877 | const struct io_uring_sqe *sqe) | |
0969e783 | 4878 | { |
0969e783 JA |
4879 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4880 | return -EINVAL; | |
4881 | if (sqe->ioprio || sqe->off || sqe->len || sqe->buf_index || | |
4882 | sqe->poll_events) | |
4883 | return -EINVAL; | |
4884 | ||
4885 | req->poll.addr = READ_ONCE(sqe->addr); | |
0969e783 JA |
4886 | return 0; |
4887 | } | |
4888 | ||
221c5eb2 JA |
4889 | /* |
4890 | * Find a running poll command that matches one specified in sqe->addr, | |
4891 | * and remove it if found. | |
4892 | */ | |
fc4df999 | 4893 | static int io_poll_remove(struct io_kiocb *req) |
221c5eb2 JA |
4894 | { |
4895 | struct io_ring_ctx *ctx = req->ctx; | |
0969e783 | 4896 | u64 addr; |
47f46768 | 4897 | int ret; |
221c5eb2 | 4898 | |
0969e783 | 4899 | addr = req->poll.addr; |
221c5eb2 | 4900 | spin_lock_irq(&ctx->completion_lock); |
0969e783 | 4901 | ret = io_poll_cancel(ctx, addr); |
221c5eb2 JA |
4902 | spin_unlock_irq(&ctx->completion_lock); |
4903 | ||
4e88d6e7 JA |
4904 | if (ret < 0) |
4905 | req_set_fail_links(req); | |
e1e16097 | 4906 | io_req_complete(req, ret); |
221c5eb2 JA |
4907 | return 0; |
4908 | } | |
4909 | ||
221c5eb2 JA |
4910 | static int io_poll_wake(struct wait_queue_entry *wait, unsigned mode, int sync, |
4911 | void *key) | |
4912 | { | |
c2f2eb7d JA |
4913 | struct io_kiocb *req = wait->private; |
4914 | struct io_poll_iocb *poll = &req->poll; | |
221c5eb2 | 4915 | |
d7718a9d | 4916 | return __io_async_wake(req, poll, key_to_poll(key), io_poll_task_func); |
221c5eb2 JA |
4917 | } |
4918 | ||
221c5eb2 JA |
4919 | static void io_poll_queue_proc(struct file *file, struct wait_queue_head *head, |
4920 | struct poll_table_struct *p) | |
4921 | { | |
4922 | struct io_poll_table *pt = container_of(p, struct io_poll_table, pt); | |
4923 | ||
807abcb0 | 4924 | __io_queue_proc(&pt->req->poll, pt, head, (struct io_poll_iocb **) &pt->req->io); |
eac406c6 JA |
4925 | } |
4926 | ||
3529d8c2 | 4927 | static int io_poll_add_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
221c5eb2 JA |
4928 | { |
4929 | struct io_poll_iocb *poll = &req->poll; | |
5769a351 | 4930 | u32 events; |
221c5eb2 JA |
4931 | |
4932 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) | |
4933 | return -EINVAL; | |
4934 | if (sqe->addr || sqe->ioprio || sqe->off || sqe->len || sqe->buf_index) | |
4935 | return -EINVAL; | |
09bb8394 JA |
4936 | if (!poll->file) |
4937 | return -EBADF; | |
221c5eb2 | 4938 | |
5769a351 JX |
4939 | events = READ_ONCE(sqe->poll32_events); |
4940 | #ifdef __BIG_ENDIAN | |
4941 | events = swahw32(events); | |
4942 | #endif | |
a31eb4a2 JX |
4943 | poll->events = demangle_poll(events) | EPOLLERR | EPOLLHUP | |
4944 | (events & EPOLLEXCLUSIVE); | |
b41e9852 | 4945 | |
4dd2824d | 4946 | io_get_req_task(req); |
0969e783 JA |
4947 | return 0; |
4948 | } | |
4949 | ||
014db007 | 4950 | static int io_poll_add(struct io_kiocb *req) |
0969e783 JA |
4951 | { |
4952 | struct io_poll_iocb *poll = &req->poll; | |
4953 | struct io_ring_ctx *ctx = req->ctx; | |
4954 | struct io_poll_table ipt; | |
0969e783 | 4955 | __poll_t mask; |
0969e783 | 4956 | |
d5e16d8e | 4957 | /* ->work is in union with hash_node and others */ |
dca9cf8b | 4958 | io_req_clean_work(req); |
d5e16d8e PB |
4959 | req->flags &= ~REQ_F_WORK_INITIALIZED; |
4960 | ||
78076bb6 | 4961 | INIT_HLIST_NODE(&req->hash_node); |
d7718a9d | 4962 | ipt.pt._qproc = io_poll_queue_proc; |
36703247 | 4963 | |
d7718a9d JA |
4964 | mask = __io_arm_poll_handler(req, &req->poll, &ipt, poll->events, |
4965 | io_poll_wake); | |
221c5eb2 | 4966 | |
8c838788 | 4967 | if (mask) { /* no async, we'd stolen it */ |
221c5eb2 | 4968 | ipt.error = 0; |
b0dd8a41 | 4969 | io_poll_complete(req, mask, 0); |
221c5eb2 | 4970 | } |
221c5eb2 JA |
4971 | spin_unlock_irq(&ctx->completion_lock); |
4972 | ||
8c838788 JA |
4973 | if (mask) { |
4974 | io_cqring_ev_posted(ctx); | |
014db007 | 4975 | io_put_req(req); |
221c5eb2 | 4976 | } |
8c838788 | 4977 | return ipt.error; |
221c5eb2 JA |
4978 | } |
4979 | ||
5262f567 JA |
4980 | static enum hrtimer_restart io_timeout_fn(struct hrtimer *timer) |
4981 | { | |
ad8a48ac JA |
4982 | struct io_timeout_data *data = container_of(timer, |
4983 | struct io_timeout_data, timer); | |
4984 | struct io_kiocb *req = data->req; | |
4985 | struct io_ring_ctx *ctx = req->ctx; | |
5262f567 JA |
4986 | unsigned long flags; |
4987 | ||
5262f567 JA |
4988 | atomic_inc(&ctx->cq_timeouts); |
4989 | ||
4990 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
ef03681a | 4991 | /* |
11365043 JA |
4992 | * We could be racing with timeout deletion. If the list is empty, |
4993 | * then timeout lookup already found it and will be handling it. | |
ef03681a | 4994 | */ |
135fcde8 PB |
4995 | if (!list_empty(&req->timeout.list)) |
4996 | list_del_init(&req->timeout.list); | |
5262f567 | 4997 | |
78e19bbe | 4998 | io_cqring_fill_event(req, -ETIME); |
5262f567 JA |
4999 | io_commit_cqring(ctx); |
5000 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
5001 | ||
5002 | io_cqring_ev_posted(ctx); | |
4e88d6e7 | 5003 | req_set_fail_links(req); |
5262f567 JA |
5004 | io_put_req(req); |
5005 | return HRTIMER_NORESTART; | |
5006 | } | |
5007 | ||
47f46768 JA |
5008 | static int io_timeout_cancel(struct io_ring_ctx *ctx, __u64 user_data) |
5009 | { | |
5010 | struct io_kiocb *req; | |
5011 | int ret = -ENOENT; | |
5012 | ||
135fcde8 | 5013 | list_for_each_entry(req, &ctx->timeout_list, timeout.list) { |
47f46768 | 5014 | if (user_data == req->user_data) { |
135fcde8 | 5015 | list_del_init(&req->timeout.list); |
47f46768 JA |
5016 | ret = 0; |
5017 | break; | |
5018 | } | |
5019 | } | |
5020 | ||
5021 | if (ret == -ENOENT) | |
5022 | return ret; | |
5023 | ||
2d28390a | 5024 | ret = hrtimer_try_to_cancel(&req->io->timeout.timer); |
47f46768 JA |
5025 | if (ret == -1) |
5026 | return -EALREADY; | |
5027 | ||
4e88d6e7 | 5028 | req_set_fail_links(req); |
47f46768 JA |
5029 | io_cqring_fill_event(req, -ECANCELED); |
5030 | io_put_req(req); | |
5031 | return 0; | |
5032 | } | |
5033 | ||
3529d8c2 JA |
5034 | static int io_timeout_remove_prep(struct io_kiocb *req, |
5035 | const struct io_uring_sqe *sqe) | |
b29472ee | 5036 | { |
b29472ee JA |
5037 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
5038 | return -EINVAL; | |
61710e43 DA |
5039 | if (unlikely(req->flags & (REQ_F_FIXED_FILE | REQ_F_BUFFER_SELECT))) |
5040 | return -EINVAL; | |
5041 | if (sqe->ioprio || sqe->buf_index || sqe->len) | |
b29472ee JA |
5042 | return -EINVAL; |
5043 | ||
5044 | req->timeout.addr = READ_ONCE(sqe->addr); | |
5045 | req->timeout.flags = READ_ONCE(sqe->timeout_flags); | |
5046 | if (req->timeout.flags) | |
5047 | return -EINVAL; | |
5048 | ||
b29472ee JA |
5049 | return 0; |
5050 | } | |
5051 | ||
11365043 JA |
5052 | /* |
5053 | * Remove or update an existing timeout command | |
5054 | */ | |
fc4df999 | 5055 | static int io_timeout_remove(struct io_kiocb *req) |
11365043 JA |
5056 | { |
5057 | struct io_ring_ctx *ctx = req->ctx; | |
47f46768 | 5058 | int ret; |
11365043 | 5059 | |
11365043 | 5060 | spin_lock_irq(&ctx->completion_lock); |
b29472ee | 5061 | ret = io_timeout_cancel(ctx, req->timeout.addr); |
11365043 | 5062 | |
47f46768 | 5063 | io_cqring_fill_event(req, ret); |
11365043 JA |
5064 | io_commit_cqring(ctx); |
5065 | spin_unlock_irq(&ctx->completion_lock); | |
5262f567 | 5066 | io_cqring_ev_posted(ctx); |
4e88d6e7 JA |
5067 | if (ret < 0) |
5068 | req_set_fail_links(req); | |
ec9c02ad | 5069 | io_put_req(req); |
11365043 | 5070 | return 0; |
5262f567 JA |
5071 | } |
5072 | ||
3529d8c2 | 5073 | static int io_timeout_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe, |
2d28390a | 5074 | bool is_timeout_link) |
5262f567 | 5075 | { |
ad8a48ac | 5076 | struct io_timeout_data *data; |
a41525ab | 5077 | unsigned flags; |
56080b02 | 5078 | u32 off = READ_ONCE(sqe->off); |
5262f567 | 5079 | |
ad8a48ac | 5080 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
5262f567 | 5081 | return -EINVAL; |
ad8a48ac | 5082 | if (sqe->ioprio || sqe->buf_index || sqe->len != 1) |
a41525ab | 5083 | return -EINVAL; |
56080b02 | 5084 | if (off && is_timeout_link) |
2d28390a | 5085 | return -EINVAL; |
a41525ab JA |
5086 | flags = READ_ONCE(sqe->timeout_flags); |
5087 | if (flags & ~IORING_TIMEOUT_ABS) | |
5262f567 | 5088 | return -EINVAL; |
bdf20073 | 5089 | |
bfe68a22 | 5090 | req->timeout.off = off; |
26a61679 | 5091 | |
3529d8c2 | 5092 | if (!req->io && io_alloc_async_ctx(req)) |
26a61679 JA |
5093 | return -ENOMEM; |
5094 | ||
5095 | data = &req->io->timeout; | |
ad8a48ac | 5096 | data->req = req; |
ad8a48ac JA |
5097 | |
5098 | if (get_timespec64(&data->ts, u64_to_user_ptr(sqe->addr))) | |
5262f567 JA |
5099 | return -EFAULT; |
5100 | ||
11365043 | 5101 | if (flags & IORING_TIMEOUT_ABS) |
ad8a48ac | 5102 | data->mode = HRTIMER_MODE_ABS; |
11365043 | 5103 | else |
ad8a48ac | 5104 | data->mode = HRTIMER_MODE_REL; |
11365043 | 5105 | |
ad8a48ac JA |
5106 | hrtimer_init(&data->timer, CLOCK_MONOTONIC, data->mode); |
5107 | return 0; | |
5108 | } | |
5109 | ||
fc4df999 | 5110 | static int io_timeout(struct io_kiocb *req) |
ad8a48ac | 5111 | { |
ad8a48ac | 5112 | struct io_ring_ctx *ctx = req->ctx; |
bfe68a22 | 5113 | struct io_timeout_data *data = &req->io->timeout; |
ad8a48ac | 5114 | struct list_head *entry; |
bfe68a22 | 5115 | u32 tail, off = req->timeout.off; |
ad8a48ac | 5116 | |
733f5c95 | 5117 | spin_lock_irq(&ctx->completion_lock); |
93bd25bb | 5118 | |
5262f567 JA |
5119 | /* |
5120 | * sqe->off holds how many events that need to occur for this | |
93bd25bb JA |
5121 | * timeout event to be satisfied. If it isn't set, then this is |
5122 | * a pure timeout request, sequence isn't used. | |
5262f567 | 5123 | */ |
8eb7e2d0 | 5124 | if (io_is_timeout_noseq(req)) { |
93bd25bb JA |
5125 | entry = ctx->timeout_list.prev; |
5126 | goto add; | |
5127 | } | |
5262f567 | 5128 | |
bfe68a22 PB |
5129 | tail = ctx->cached_cq_tail - atomic_read(&ctx->cq_timeouts); |
5130 | req->timeout.target_seq = tail + off; | |
5262f567 JA |
5131 | |
5132 | /* | |
5133 | * Insertion sort, ensuring the first entry in the list is always | |
5134 | * the one we need first. | |
5135 | */ | |
5262f567 | 5136 | list_for_each_prev(entry, &ctx->timeout_list) { |
135fcde8 PB |
5137 | struct io_kiocb *nxt = list_entry(entry, struct io_kiocb, |
5138 | timeout.list); | |
5262f567 | 5139 | |
8eb7e2d0 | 5140 | if (io_is_timeout_noseq(nxt)) |
93bd25bb | 5141 | continue; |
bfe68a22 PB |
5142 | /* nxt.seq is behind @tail, otherwise would've been completed */ |
5143 | if (off >= nxt->timeout.target_seq - tail) | |
5262f567 JA |
5144 | break; |
5145 | } | |
93bd25bb | 5146 | add: |
135fcde8 | 5147 | list_add(&req->timeout.list, entry); |
ad8a48ac JA |
5148 | data->timer.function = io_timeout_fn; |
5149 | hrtimer_start(&data->timer, timespec64_to_ktime(data->ts), data->mode); | |
5262f567 | 5150 | spin_unlock_irq(&ctx->completion_lock); |
5262f567 JA |
5151 | return 0; |
5152 | } | |
5262f567 | 5153 | |
62755e35 JA |
5154 | static bool io_cancel_cb(struct io_wq_work *work, void *data) |
5155 | { | |
5156 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); | |
5157 | ||
5158 | return req->user_data == (unsigned long) data; | |
5159 | } | |
5160 | ||
e977d6d3 | 5161 | static int io_async_cancel_one(struct io_ring_ctx *ctx, void *sqe_addr) |
62755e35 | 5162 | { |
62755e35 | 5163 | enum io_wq_cancel cancel_ret; |
62755e35 JA |
5164 | int ret = 0; |
5165 | ||
4f26bda1 | 5166 | cancel_ret = io_wq_cancel_cb(ctx->io_wq, io_cancel_cb, sqe_addr, false); |
62755e35 JA |
5167 | switch (cancel_ret) { |
5168 | case IO_WQ_CANCEL_OK: | |
5169 | ret = 0; | |
5170 | break; | |
5171 | case IO_WQ_CANCEL_RUNNING: | |
5172 | ret = -EALREADY; | |
5173 | break; | |
5174 | case IO_WQ_CANCEL_NOTFOUND: | |
5175 | ret = -ENOENT; | |
5176 | break; | |
5177 | } | |
5178 | ||
e977d6d3 JA |
5179 | return ret; |
5180 | } | |
5181 | ||
47f46768 JA |
5182 | static void io_async_find_and_cancel(struct io_ring_ctx *ctx, |
5183 | struct io_kiocb *req, __u64 sqe_addr, | |
014db007 | 5184 | int success_ret) |
47f46768 JA |
5185 | { |
5186 | unsigned long flags; | |
5187 | int ret; | |
5188 | ||
5189 | ret = io_async_cancel_one(ctx, (void *) (unsigned long) sqe_addr); | |
5190 | if (ret != -ENOENT) { | |
5191 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
5192 | goto done; | |
5193 | } | |
5194 | ||
5195 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
5196 | ret = io_timeout_cancel(ctx, sqe_addr); | |
5197 | if (ret != -ENOENT) | |
5198 | goto done; | |
5199 | ret = io_poll_cancel(ctx, sqe_addr); | |
5200 | done: | |
b0dd8a41 JA |
5201 | if (!ret) |
5202 | ret = success_ret; | |
47f46768 JA |
5203 | io_cqring_fill_event(req, ret); |
5204 | io_commit_cqring(ctx); | |
5205 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
5206 | io_cqring_ev_posted(ctx); | |
5207 | ||
4e88d6e7 JA |
5208 | if (ret < 0) |
5209 | req_set_fail_links(req); | |
014db007 | 5210 | io_put_req(req); |
47f46768 JA |
5211 | } |
5212 | ||
3529d8c2 JA |
5213 | static int io_async_cancel_prep(struct io_kiocb *req, |
5214 | const struct io_uring_sqe *sqe) | |
e977d6d3 | 5215 | { |
fbf23849 | 5216 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
e977d6d3 | 5217 | return -EINVAL; |
61710e43 DA |
5218 | if (unlikely(req->flags & (REQ_F_FIXED_FILE | REQ_F_BUFFER_SELECT))) |
5219 | return -EINVAL; | |
5220 | if (sqe->ioprio || sqe->off || sqe->len || sqe->cancel_flags) | |
e977d6d3 JA |
5221 | return -EINVAL; |
5222 | ||
fbf23849 JA |
5223 | req->cancel.addr = READ_ONCE(sqe->addr); |
5224 | return 0; | |
5225 | } | |
5226 | ||
014db007 | 5227 | static int io_async_cancel(struct io_kiocb *req) |
fbf23849 JA |
5228 | { |
5229 | struct io_ring_ctx *ctx = req->ctx; | |
fbf23849 | 5230 | |
014db007 | 5231 | io_async_find_and_cancel(ctx, req, req->cancel.addr, 0); |
5262f567 JA |
5232 | return 0; |
5233 | } | |
5234 | ||
05f3fb3c JA |
5235 | static int io_files_update_prep(struct io_kiocb *req, |
5236 | const struct io_uring_sqe *sqe) | |
5237 | { | |
61710e43 DA |
5238 | if (unlikely(req->flags & (REQ_F_FIXED_FILE | REQ_F_BUFFER_SELECT))) |
5239 | return -EINVAL; | |
5240 | if (sqe->ioprio || sqe->rw_flags) | |
05f3fb3c JA |
5241 | return -EINVAL; |
5242 | ||
5243 | req->files_update.offset = READ_ONCE(sqe->off); | |
5244 | req->files_update.nr_args = READ_ONCE(sqe->len); | |
5245 | if (!req->files_update.nr_args) | |
5246 | return -EINVAL; | |
5247 | req->files_update.arg = READ_ONCE(sqe->addr); | |
5248 | return 0; | |
5249 | } | |
5250 | ||
229a7b63 JA |
5251 | static int io_files_update(struct io_kiocb *req, bool force_nonblock, |
5252 | struct io_comp_state *cs) | |
fbf23849 JA |
5253 | { |
5254 | struct io_ring_ctx *ctx = req->ctx; | |
05f3fb3c JA |
5255 | struct io_uring_files_update up; |
5256 | int ret; | |
fbf23849 | 5257 | |
f86cd20c | 5258 | if (force_nonblock) |
05f3fb3c | 5259 | return -EAGAIN; |
05f3fb3c JA |
5260 | |
5261 | up.offset = req->files_update.offset; | |
5262 | up.fds = req->files_update.arg; | |
5263 | ||
5264 | mutex_lock(&ctx->uring_lock); | |
5265 | ret = __io_sqe_files_update(ctx, &up, req->files_update.nr_args); | |
5266 | mutex_unlock(&ctx->uring_lock); | |
5267 | ||
5268 | if (ret < 0) | |
5269 | req_set_fail_links(req); | |
229a7b63 | 5270 | __io_req_complete(req, ret, 0, cs); |
5262f567 JA |
5271 | return 0; |
5272 | } | |
5273 | ||
3529d8c2 | 5274 | static int io_req_defer_prep(struct io_kiocb *req, |
debb85f4 | 5275 | const struct io_uring_sqe *sqe) |
f67676d1 | 5276 | { |
e781573e | 5277 | ssize_t ret = 0; |
f67676d1 | 5278 | |
f1d96a8f PB |
5279 | if (!sqe) |
5280 | return 0; | |
5281 | ||
327d6d96 PB |
5282 | if (io_alloc_async_ctx(req)) |
5283 | return -EAGAIN; | |
5284 | ||
710c2bfb | 5285 | if (io_op_defs[req->opcode].file_table) { |
c40f6379 | 5286 | io_req_init_async(req); |
710c2bfb PB |
5287 | ret = io_grab_files(req); |
5288 | if (unlikely(ret)) | |
5289 | return ret; | |
5290 | } | |
7cdaf587 | 5291 | |
d625c6ee | 5292 | switch (req->opcode) { |
e781573e JA |
5293 | case IORING_OP_NOP: |
5294 | break; | |
f67676d1 JA |
5295 | case IORING_OP_READV: |
5296 | case IORING_OP_READ_FIXED: | |
3a6820f2 | 5297 | case IORING_OP_READ: |
3529d8c2 | 5298 | ret = io_read_prep(req, sqe, true); |
f67676d1 JA |
5299 | break; |
5300 | case IORING_OP_WRITEV: | |
5301 | case IORING_OP_WRITE_FIXED: | |
3a6820f2 | 5302 | case IORING_OP_WRITE: |
3529d8c2 | 5303 | ret = io_write_prep(req, sqe, true); |
f67676d1 | 5304 | break; |
0969e783 | 5305 | case IORING_OP_POLL_ADD: |
3529d8c2 | 5306 | ret = io_poll_add_prep(req, sqe); |
0969e783 JA |
5307 | break; |
5308 | case IORING_OP_POLL_REMOVE: | |
3529d8c2 | 5309 | ret = io_poll_remove_prep(req, sqe); |
0969e783 | 5310 | break; |
8ed8d3c3 | 5311 | case IORING_OP_FSYNC: |
3529d8c2 | 5312 | ret = io_prep_fsync(req, sqe); |
8ed8d3c3 JA |
5313 | break; |
5314 | case IORING_OP_SYNC_FILE_RANGE: | |
3529d8c2 | 5315 | ret = io_prep_sfr(req, sqe); |
8ed8d3c3 | 5316 | break; |
03b1230c | 5317 | case IORING_OP_SENDMSG: |
fddaface | 5318 | case IORING_OP_SEND: |
3529d8c2 | 5319 | ret = io_sendmsg_prep(req, sqe); |
03b1230c JA |
5320 | break; |
5321 | case IORING_OP_RECVMSG: | |
fddaface | 5322 | case IORING_OP_RECV: |
3529d8c2 | 5323 | ret = io_recvmsg_prep(req, sqe); |
03b1230c | 5324 | break; |
f499a021 | 5325 | case IORING_OP_CONNECT: |
3529d8c2 | 5326 | ret = io_connect_prep(req, sqe); |
f499a021 | 5327 | break; |
2d28390a | 5328 | case IORING_OP_TIMEOUT: |
3529d8c2 | 5329 | ret = io_timeout_prep(req, sqe, false); |
b7bb4f7d | 5330 | break; |
b29472ee | 5331 | case IORING_OP_TIMEOUT_REMOVE: |
3529d8c2 | 5332 | ret = io_timeout_remove_prep(req, sqe); |
b29472ee | 5333 | break; |
fbf23849 | 5334 | case IORING_OP_ASYNC_CANCEL: |
3529d8c2 | 5335 | ret = io_async_cancel_prep(req, sqe); |
fbf23849 | 5336 | break; |
2d28390a | 5337 | case IORING_OP_LINK_TIMEOUT: |
3529d8c2 | 5338 | ret = io_timeout_prep(req, sqe, true); |
b7bb4f7d | 5339 | break; |
8ed8d3c3 | 5340 | case IORING_OP_ACCEPT: |
3529d8c2 | 5341 | ret = io_accept_prep(req, sqe); |
8ed8d3c3 | 5342 | break; |
d63d1b5e JA |
5343 | case IORING_OP_FALLOCATE: |
5344 | ret = io_fallocate_prep(req, sqe); | |
5345 | break; | |
15b71abe JA |
5346 | case IORING_OP_OPENAT: |
5347 | ret = io_openat_prep(req, sqe); | |
5348 | break; | |
b5dba59e JA |
5349 | case IORING_OP_CLOSE: |
5350 | ret = io_close_prep(req, sqe); | |
5351 | break; | |
05f3fb3c JA |
5352 | case IORING_OP_FILES_UPDATE: |
5353 | ret = io_files_update_prep(req, sqe); | |
5354 | break; | |
eddc7ef5 JA |
5355 | case IORING_OP_STATX: |
5356 | ret = io_statx_prep(req, sqe); | |
5357 | break; | |
4840e418 JA |
5358 | case IORING_OP_FADVISE: |
5359 | ret = io_fadvise_prep(req, sqe); | |
5360 | break; | |
c1ca757b JA |
5361 | case IORING_OP_MADVISE: |
5362 | ret = io_madvise_prep(req, sqe); | |
5363 | break; | |
cebdb986 JA |
5364 | case IORING_OP_OPENAT2: |
5365 | ret = io_openat2_prep(req, sqe); | |
5366 | break; | |
3e4827b0 JA |
5367 | case IORING_OP_EPOLL_CTL: |
5368 | ret = io_epoll_ctl_prep(req, sqe); | |
5369 | break; | |
7d67af2c PB |
5370 | case IORING_OP_SPLICE: |
5371 | ret = io_splice_prep(req, sqe); | |
5372 | break; | |
ddf0322d JA |
5373 | case IORING_OP_PROVIDE_BUFFERS: |
5374 | ret = io_provide_buffers_prep(req, sqe); | |
5375 | break; | |
067524e9 JA |
5376 | case IORING_OP_REMOVE_BUFFERS: |
5377 | ret = io_remove_buffers_prep(req, sqe); | |
5378 | break; | |
f2a8d5c7 PB |
5379 | case IORING_OP_TEE: |
5380 | ret = io_tee_prep(req, sqe); | |
5381 | break; | |
f67676d1 | 5382 | default: |
e781573e JA |
5383 | printk_once(KERN_WARNING "io_uring: unhandled opcode %d\n", |
5384 | req->opcode); | |
5385 | ret = -EINVAL; | |
b7bb4f7d | 5386 | break; |
f67676d1 JA |
5387 | } |
5388 | ||
b7bb4f7d | 5389 | return ret; |
f67676d1 JA |
5390 | } |
5391 | ||
9cf7c104 PB |
5392 | static u32 io_get_sequence(struct io_kiocb *req) |
5393 | { | |
5394 | struct io_kiocb *pos; | |
5395 | struct io_ring_ctx *ctx = req->ctx; | |
5396 | u32 total_submitted, nr_reqs = 1; | |
5397 | ||
5398 | if (req->flags & REQ_F_LINK_HEAD) | |
5399 | list_for_each_entry(pos, &req->link_list, link_list) | |
5400 | nr_reqs++; | |
5401 | ||
5402 | total_submitted = ctx->cached_sq_head - ctx->cached_sq_dropped; | |
5403 | return total_submitted - nr_reqs; | |
5404 | } | |
5405 | ||
3529d8c2 | 5406 | static int io_req_defer(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
de0617e4 | 5407 | { |
a197f664 | 5408 | struct io_ring_ctx *ctx = req->ctx; |
27dc8338 | 5409 | struct io_defer_entry *de; |
f67676d1 | 5410 | int ret; |
9cf7c104 | 5411 | u32 seq; |
de0617e4 | 5412 | |
9d858b21 | 5413 | /* Still need defer if there is pending req in defer list. */ |
9cf7c104 PB |
5414 | if (likely(list_empty_careful(&ctx->defer_list) && |
5415 | !(req->flags & REQ_F_IO_DRAIN))) | |
5416 | return 0; | |
5417 | ||
5418 | seq = io_get_sequence(req); | |
5419 | /* Still a chance to pass the sequence check */ | |
5420 | if (!req_need_defer(req, seq) && list_empty_careful(&ctx->defer_list)) | |
de0617e4 JA |
5421 | return 0; |
5422 | ||
650b5481 | 5423 | if (!req->io) { |
debb85f4 | 5424 | ret = io_req_defer_prep(req, sqe); |
327d6d96 | 5425 | if (ret) |
650b5481 PB |
5426 | return ret; |
5427 | } | |
cbdcb435 | 5428 | io_prep_async_link(req); |
27dc8338 PB |
5429 | de = kmalloc(sizeof(*de), GFP_KERNEL); |
5430 | if (!de) | |
5431 | return -ENOMEM; | |
2d28390a | 5432 | |
de0617e4 | 5433 | spin_lock_irq(&ctx->completion_lock); |
9cf7c104 | 5434 | if (!req_need_defer(req, seq) && list_empty(&ctx->defer_list)) { |
de0617e4 | 5435 | spin_unlock_irq(&ctx->completion_lock); |
27dc8338 | 5436 | kfree(de); |
de0617e4 JA |
5437 | return 0; |
5438 | } | |
5439 | ||
915967f6 | 5440 | trace_io_uring_defer(ctx, req, req->user_data); |
27dc8338 | 5441 | de->req = req; |
9cf7c104 | 5442 | de->seq = seq; |
27dc8338 | 5443 | list_add_tail(&de->list, &ctx->defer_list); |
de0617e4 JA |
5444 | spin_unlock_irq(&ctx->completion_lock); |
5445 | return -EIOCBQUEUED; | |
5446 | } | |
5447 | ||
3ca405eb | 5448 | static void __io_clean_op(struct io_kiocb *req) |
99bc4c38 PB |
5449 | { |
5450 | struct io_async_ctx *io = req->io; | |
5451 | ||
5452 | switch (req->opcode) { | |
5453 | case IORING_OP_READV: | |
5454 | case IORING_OP_READ_FIXED: | |
5455 | case IORING_OP_READ: | |
bcda7baa JA |
5456 | if (req->flags & REQ_F_BUFFER_SELECTED) |
5457 | kfree((void *)(unsigned long)req->rw.addr); | |
5458 | /* fallthrough */ | |
99bc4c38 PB |
5459 | case IORING_OP_WRITEV: |
5460 | case IORING_OP_WRITE_FIXED: | |
5461 | case IORING_OP_WRITE: | |
5462 | if (io->rw.iov != io->rw.fast_iov) | |
5463 | kfree(io->rw.iov); | |
5464 | break; | |
99bc4c38 | 5465 | case IORING_OP_RECVMSG: |
52de1fe1 JA |
5466 | if (req->flags & REQ_F_BUFFER_SELECTED) |
5467 | kfree(req->sr_msg.kbuf); | |
5468 | /* fallthrough */ | |
5469 | case IORING_OP_SENDMSG: | |
99bc4c38 PB |
5470 | if (io->msg.iov != io->msg.fast_iov) |
5471 | kfree(io->msg.iov); | |
5472 | break; | |
bcda7baa JA |
5473 | case IORING_OP_RECV: |
5474 | if (req->flags & REQ_F_BUFFER_SELECTED) | |
5475 | kfree(req->sr_msg.kbuf); | |
5476 | break; | |
7d67af2c | 5477 | case IORING_OP_SPLICE: |
f2a8d5c7 | 5478 | case IORING_OP_TEE: |
7d67af2c PB |
5479 | io_put_file(req, req->splice.file_in, |
5480 | (req->splice.flags & SPLICE_F_FD_IN_FIXED)); | |
5481 | break; | |
99bc4c38 PB |
5482 | } |
5483 | ||
5484 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
5485 | } | |
5486 | ||
3529d8c2 | 5487 | static int io_issue_sqe(struct io_kiocb *req, const struct io_uring_sqe *sqe, |
f13fad7b | 5488 | bool force_nonblock, struct io_comp_state *cs) |
2b188cc1 | 5489 | { |
a197f664 | 5490 | struct io_ring_ctx *ctx = req->ctx; |
d625c6ee | 5491 | int ret; |
2b188cc1 | 5492 | |
d625c6ee | 5493 | switch (req->opcode) { |
2b188cc1 | 5494 | case IORING_OP_NOP: |
229a7b63 | 5495 | ret = io_nop(req, cs); |
2b188cc1 JA |
5496 | break; |
5497 | case IORING_OP_READV: | |
edafccee | 5498 | case IORING_OP_READ_FIXED: |
3a6820f2 | 5499 | case IORING_OP_READ: |
3529d8c2 JA |
5500 | if (sqe) { |
5501 | ret = io_read_prep(req, sqe, force_nonblock); | |
5502 | if (ret < 0) | |
5503 | break; | |
5504 | } | |
a1d7c393 | 5505 | ret = io_read(req, force_nonblock, cs); |
edafccee | 5506 | break; |
3529d8c2 | 5507 | case IORING_OP_WRITEV: |
edafccee | 5508 | case IORING_OP_WRITE_FIXED: |
3a6820f2 | 5509 | case IORING_OP_WRITE: |
3529d8c2 JA |
5510 | if (sqe) { |
5511 | ret = io_write_prep(req, sqe, force_nonblock); | |
5512 | if (ret < 0) | |
5513 | break; | |
5514 | } | |
a1d7c393 | 5515 | ret = io_write(req, force_nonblock, cs); |
2b188cc1 | 5516 | break; |
c992fe29 | 5517 | case IORING_OP_FSYNC: |
3529d8c2 JA |
5518 | if (sqe) { |
5519 | ret = io_prep_fsync(req, sqe); | |
5520 | if (ret < 0) | |
5521 | break; | |
5522 | } | |
014db007 | 5523 | ret = io_fsync(req, force_nonblock); |
c992fe29 | 5524 | break; |
221c5eb2 | 5525 | case IORING_OP_POLL_ADD: |
3529d8c2 JA |
5526 | if (sqe) { |
5527 | ret = io_poll_add_prep(req, sqe); | |
5528 | if (ret) | |
5529 | break; | |
5530 | } | |
014db007 | 5531 | ret = io_poll_add(req); |
221c5eb2 JA |
5532 | break; |
5533 | case IORING_OP_POLL_REMOVE: | |
3529d8c2 JA |
5534 | if (sqe) { |
5535 | ret = io_poll_remove_prep(req, sqe); | |
5536 | if (ret < 0) | |
5537 | break; | |
5538 | } | |
fc4df999 | 5539 | ret = io_poll_remove(req); |
221c5eb2 | 5540 | break; |
5d17b4a4 | 5541 | case IORING_OP_SYNC_FILE_RANGE: |
3529d8c2 JA |
5542 | if (sqe) { |
5543 | ret = io_prep_sfr(req, sqe); | |
5544 | if (ret < 0) | |
5545 | break; | |
5546 | } | |
014db007 | 5547 | ret = io_sync_file_range(req, force_nonblock); |
5d17b4a4 | 5548 | break; |
0fa03c62 | 5549 | case IORING_OP_SENDMSG: |
fddaface | 5550 | case IORING_OP_SEND: |
3529d8c2 JA |
5551 | if (sqe) { |
5552 | ret = io_sendmsg_prep(req, sqe); | |
5553 | if (ret < 0) | |
5554 | break; | |
5555 | } | |
fddaface | 5556 | if (req->opcode == IORING_OP_SENDMSG) |
229a7b63 | 5557 | ret = io_sendmsg(req, force_nonblock, cs); |
fddaface | 5558 | else |
229a7b63 | 5559 | ret = io_send(req, force_nonblock, cs); |
0fa03c62 | 5560 | break; |
aa1fa28f | 5561 | case IORING_OP_RECVMSG: |
fddaface | 5562 | case IORING_OP_RECV: |
3529d8c2 JA |
5563 | if (sqe) { |
5564 | ret = io_recvmsg_prep(req, sqe); | |
5565 | if (ret) | |
5566 | break; | |
5567 | } | |
fddaface | 5568 | if (req->opcode == IORING_OP_RECVMSG) |
229a7b63 | 5569 | ret = io_recvmsg(req, force_nonblock, cs); |
fddaface | 5570 | else |
229a7b63 | 5571 | ret = io_recv(req, force_nonblock, cs); |
aa1fa28f | 5572 | break; |
5262f567 | 5573 | case IORING_OP_TIMEOUT: |
3529d8c2 JA |
5574 | if (sqe) { |
5575 | ret = io_timeout_prep(req, sqe, false); | |
5576 | if (ret) | |
5577 | break; | |
5578 | } | |
fc4df999 | 5579 | ret = io_timeout(req); |
5262f567 | 5580 | break; |
11365043 | 5581 | case IORING_OP_TIMEOUT_REMOVE: |
3529d8c2 JA |
5582 | if (sqe) { |
5583 | ret = io_timeout_remove_prep(req, sqe); | |
5584 | if (ret) | |
5585 | break; | |
5586 | } | |
fc4df999 | 5587 | ret = io_timeout_remove(req); |
11365043 | 5588 | break; |
17f2fe35 | 5589 | case IORING_OP_ACCEPT: |
3529d8c2 JA |
5590 | if (sqe) { |
5591 | ret = io_accept_prep(req, sqe); | |
5592 | if (ret) | |
5593 | break; | |
5594 | } | |
229a7b63 | 5595 | ret = io_accept(req, force_nonblock, cs); |
17f2fe35 | 5596 | break; |
f8e85cf2 | 5597 | case IORING_OP_CONNECT: |
3529d8c2 JA |
5598 | if (sqe) { |
5599 | ret = io_connect_prep(req, sqe); | |
5600 | if (ret) | |
5601 | break; | |
5602 | } | |
229a7b63 | 5603 | ret = io_connect(req, force_nonblock, cs); |
f8e85cf2 | 5604 | break; |
62755e35 | 5605 | case IORING_OP_ASYNC_CANCEL: |
3529d8c2 JA |
5606 | if (sqe) { |
5607 | ret = io_async_cancel_prep(req, sqe); | |
5608 | if (ret) | |
5609 | break; | |
5610 | } | |
014db007 | 5611 | ret = io_async_cancel(req); |
62755e35 | 5612 | break; |
d63d1b5e JA |
5613 | case IORING_OP_FALLOCATE: |
5614 | if (sqe) { | |
5615 | ret = io_fallocate_prep(req, sqe); | |
5616 | if (ret) | |
5617 | break; | |
5618 | } | |
014db007 | 5619 | ret = io_fallocate(req, force_nonblock); |
d63d1b5e | 5620 | break; |
15b71abe JA |
5621 | case IORING_OP_OPENAT: |
5622 | if (sqe) { | |
5623 | ret = io_openat_prep(req, sqe); | |
5624 | if (ret) | |
5625 | break; | |
5626 | } | |
014db007 | 5627 | ret = io_openat(req, force_nonblock); |
15b71abe | 5628 | break; |
b5dba59e JA |
5629 | case IORING_OP_CLOSE: |
5630 | if (sqe) { | |
5631 | ret = io_close_prep(req, sqe); | |
5632 | if (ret) | |
5633 | break; | |
5634 | } | |
229a7b63 | 5635 | ret = io_close(req, force_nonblock, cs); |
b5dba59e | 5636 | break; |
05f3fb3c JA |
5637 | case IORING_OP_FILES_UPDATE: |
5638 | if (sqe) { | |
5639 | ret = io_files_update_prep(req, sqe); | |
5640 | if (ret) | |
5641 | break; | |
5642 | } | |
229a7b63 | 5643 | ret = io_files_update(req, force_nonblock, cs); |
05f3fb3c | 5644 | break; |
eddc7ef5 JA |
5645 | case IORING_OP_STATX: |
5646 | if (sqe) { | |
5647 | ret = io_statx_prep(req, sqe); | |
5648 | if (ret) | |
5649 | break; | |
5650 | } | |
014db007 | 5651 | ret = io_statx(req, force_nonblock); |
eddc7ef5 | 5652 | break; |
4840e418 JA |
5653 | case IORING_OP_FADVISE: |
5654 | if (sqe) { | |
5655 | ret = io_fadvise_prep(req, sqe); | |
5656 | if (ret) | |
5657 | break; | |
5658 | } | |
014db007 | 5659 | ret = io_fadvise(req, force_nonblock); |
4840e418 | 5660 | break; |
c1ca757b JA |
5661 | case IORING_OP_MADVISE: |
5662 | if (sqe) { | |
5663 | ret = io_madvise_prep(req, sqe); | |
5664 | if (ret) | |
5665 | break; | |
5666 | } | |
014db007 | 5667 | ret = io_madvise(req, force_nonblock); |
c1ca757b | 5668 | break; |
cebdb986 JA |
5669 | case IORING_OP_OPENAT2: |
5670 | if (sqe) { | |
5671 | ret = io_openat2_prep(req, sqe); | |
5672 | if (ret) | |
5673 | break; | |
5674 | } | |
014db007 | 5675 | ret = io_openat2(req, force_nonblock); |
cebdb986 | 5676 | break; |
3e4827b0 JA |
5677 | case IORING_OP_EPOLL_CTL: |
5678 | if (sqe) { | |
5679 | ret = io_epoll_ctl_prep(req, sqe); | |
5680 | if (ret) | |
5681 | break; | |
5682 | } | |
229a7b63 | 5683 | ret = io_epoll_ctl(req, force_nonblock, cs); |
3e4827b0 | 5684 | break; |
7d67af2c PB |
5685 | case IORING_OP_SPLICE: |
5686 | if (sqe) { | |
5687 | ret = io_splice_prep(req, sqe); | |
5688 | if (ret < 0) | |
5689 | break; | |
5690 | } | |
014db007 | 5691 | ret = io_splice(req, force_nonblock); |
7d67af2c | 5692 | break; |
ddf0322d JA |
5693 | case IORING_OP_PROVIDE_BUFFERS: |
5694 | if (sqe) { | |
5695 | ret = io_provide_buffers_prep(req, sqe); | |
5696 | if (ret) | |
5697 | break; | |
5698 | } | |
229a7b63 | 5699 | ret = io_provide_buffers(req, force_nonblock, cs); |
ddf0322d | 5700 | break; |
067524e9 JA |
5701 | case IORING_OP_REMOVE_BUFFERS: |
5702 | if (sqe) { | |
5703 | ret = io_remove_buffers_prep(req, sqe); | |
5704 | if (ret) | |
5705 | break; | |
5706 | } | |
229a7b63 | 5707 | ret = io_remove_buffers(req, force_nonblock, cs); |
3e4827b0 | 5708 | break; |
f2a8d5c7 PB |
5709 | case IORING_OP_TEE: |
5710 | if (sqe) { | |
5711 | ret = io_tee_prep(req, sqe); | |
5712 | if (ret < 0) | |
5713 | break; | |
5714 | } | |
5715 | ret = io_tee(req, force_nonblock); | |
5716 | break; | |
2b188cc1 JA |
5717 | default: |
5718 | ret = -EINVAL; | |
5719 | break; | |
5720 | } | |
5721 | ||
def596e9 JA |
5722 | if (ret) |
5723 | return ret; | |
5724 | ||
b532576e JA |
5725 | /* If the op doesn't have a file, we're not polling for it */ |
5726 | if ((ctx->flags & IORING_SETUP_IOPOLL) && req->file) { | |
11ba820b JA |
5727 | const bool in_async = io_wq_current_is_worker(); |
5728 | ||
11ba820b JA |
5729 | /* workqueue context doesn't hold uring_lock, grab it now */ |
5730 | if (in_async) | |
5731 | mutex_lock(&ctx->uring_lock); | |
5732 | ||
def596e9 | 5733 | io_iopoll_req_issued(req); |
11ba820b JA |
5734 | |
5735 | if (in_async) | |
5736 | mutex_unlock(&ctx->uring_lock); | |
def596e9 JA |
5737 | } |
5738 | ||
5739 | return 0; | |
2b188cc1 JA |
5740 | } |
5741 | ||
f4db7182 | 5742 | static struct io_wq_work *io_wq_submit_work(struct io_wq_work *work) |
2b188cc1 JA |
5743 | { |
5744 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); | |
6df1db6b | 5745 | struct io_kiocb *timeout; |
561fb04a | 5746 | int ret = 0; |
2b188cc1 | 5747 | |
6df1db6b PB |
5748 | timeout = io_prep_linked_timeout(req); |
5749 | if (timeout) | |
5750 | io_queue_linked_timeout(timeout); | |
d4c81f38 | 5751 | |
0c9d5ccd JA |
5752 | /* if NO_CANCEL is set, we must still run the work */ |
5753 | if ((work->flags & (IO_WQ_WORK_CANCEL|IO_WQ_WORK_NO_CANCEL)) == | |
5754 | IO_WQ_WORK_CANCEL) { | |
561fb04a | 5755 | ret = -ECANCELED; |
0c9d5ccd | 5756 | } |
31b51510 | 5757 | |
561fb04a | 5758 | if (!ret) { |
561fb04a | 5759 | do { |
f13fad7b | 5760 | ret = io_issue_sqe(req, NULL, false, NULL); |
561fb04a JA |
5761 | /* |
5762 | * We can get EAGAIN for polled IO even though we're | |
5763 | * forcing a sync submission from here, since we can't | |
5764 | * wait for request slots on the block side. | |
5765 | */ | |
5766 | if (ret != -EAGAIN) | |
5767 | break; | |
5768 | cond_resched(); | |
5769 | } while (1); | |
5770 | } | |
31b51510 | 5771 | |
561fb04a | 5772 | if (ret) { |
4e88d6e7 | 5773 | req_set_fail_links(req); |
e1e16097 | 5774 | io_req_complete(req, ret); |
edafccee | 5775 | } |
2b188cc1 | 5776 | |
f4db7182 | 5777 | return io_steal_work(req); |
2b188cc1 JA |
5778 | } |
5779 | ||
65e19f54 JA |
5780 | static inline struct file *io_file_from_index(struct io_ring_ctx *ctx, |
5781 | int index) | |
5782 | { | |
5783 | struct fixed_file_table *table; | |
5784 | ||
05f3fb3c | 5785 | table = &ctx->file_data->table[index >> IORING_FILE_TABLE_SHIFT]; |
84695089 | 5786 | return table->files[index & IORING_FILE_TABLE_MASK]; |
65e19f54 JA |
5787 | } |
5788 | ||
8da11c19 PB |
5789 | static int io_file_get(struct io_submit_state *state, struct io_kiocb *req, |
5790 | int fd, struct file **out_file, bool fixed) | |
09bb8394 | 5791 | { |
a197f664 | 5792 | struct io_ring_ctx *ctx = req->ctx; |
8da11c19 | 5793 | struct file *file; |
09bb8394 | 5794 | |
8da11c19 | 5795 | if (fixed) { |
05f3fb3c | 5796 | if (unlikely(!ctx->file_data || |
09bb8394 JA |
5797 | (unsigned) fd >= ctx->nr_user_files)) |
5798 | return -EBADF; | |
b7620121 | 5799 | fd = array_index_nospec(fd, ctx->nr_user_files); |
8da11c19 | 5800 | file = io_file_from_index(ctx, fd); |
fd2206e4 JA |
5801 | if (file) { |
5802 | req->fixed_file_refs = ctx->file_data->cur_refs; | |
5803 | percpu_ref_get(req->fixed_file_refs); | |
5804 | } | |
09bb8394 | 5805 | } else { |
c826bd7a | 5806 | trace_io_uring_file_get(ctx, fd); |
8da11c19 | 5807 | file = __io_file_get(state, fd); |
09bb8394 JA |
5808 | } |
5809 | ||
fd2206e4 JA |
5810 | if (file || io_op_defs[req->opcode].needs_file_no_error) { |
5811 | *out_file = file; | |
5812 | return 0; | |
5813 | } | |
5814 | return -EBADF; | |
09bb8394 JA |
5815 | } |
5816 | ||
8da11c19 | 5817 | static int io_req_set_file(struct io_submit_state *state, struct io_kiocb *req, |
63ff8223 | 5818 | int fd) |
8da11c19 | 5819 | { |
8da11c19 PB |
5820 | bool fixed; |
5821 | ||
63ff8223 | 5822 | fixed = (req->flags & REQ_F_FIXED_FILE) != 0; |
0cdaf760 | 5823 | if (unlikely(!fixed && io_async_submit(req->ctx))) |
8da11c19 PB |
5824 | return -EBADF; |
5825 | ||
5826 | return io_file_get(state, req, fd, &req->file, fixed); | |
5827 | } | |
5828 | ||
a197f664 | 5829 | static int io_grab_files(struct io_kiocb *req) |
fcb323cc JA |
5830 | { |
5831 | int ret = -EBADF; | |
a197f664 | 5832 | struct io_ring_ctx *ctx = req->ctx; |
fcb323cc | 5833 | |
5b0bbee4 | 5834 | if (req->work.files || (req->flags & REQ_F_NO_FILE_TABLE)) |
f86cd20c | 5835 | return 0; |
b14cca0c | 5836 | if (!ctx->ring_file) |
b5dba59e JA |
5837 | return -EBADF; |
5838 | ||
fcb323cc JA |
5839 | rcu_read_lock(); |
5840 | spin_lock_irq(&ctx->inflight_lock); | |
5841 | /* | |
5842 | * We use the f_ops->flush() handler to ensure that we can flush | |
5843 | * out work accessing these files if the fd is closed. Check if | |
5844 | * the fd has changed since we started down this path, and disallow | |
5845 | * this operation if it has. | |
5846 | */ | |
b14cca0c | 5847 | if (fcheck(ctx->ring_fd) == ctx->ring_file) { |
fcb323cc JA |
5848 | list_add(&req->inflight_entry, &ctx->inflight_list); |
5849 | req->flags |= REQ_F_INFLIGHT; | |
5850 | req->work.files = current->files; | |
5851 | ret = 0; | |
5852 | } | |
5853 | spin_unlock_irq(&ctx->inflight_lock); | |
5854 | rcu_read_unlock(); | |
5855 | ||
5856 | return ret; | |
5857 | } | |
5858 | ||
2665abfd | 5859 | static enum hrtimer_restart io_link_timeout_fn(struct hrtimer *timer) |
2b188cc1 | 5860 | { |
ad8a48ac JA |
5861 | struct io_timeout_data *data = container_of(timer, |
5862 | struct io_timeout_data, timer); | |
5863 | struct io_kiocb *req = data->req; | |
2665abfd JA |
5864 | struct io_ring_ctx *ctx = req->ctx; |
5865 | struct io_kiocb *prev = NULL; | |
5866 | unsigned long flags; | |
2665abfd JA |
5867 | |
5868 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
5869 | ||
5870 | /* | |
5871 | * We don't expect the list to be empty, that will only happen if we | |
5872 | * race with the completion of the linked work. | |
5873 | */ | |
4493233e PB |
5874 | if (!list_empty(&req->link_list)) { |
5875 | prev = list_entry(req->link_list.prev, struct io_kiocb, | |
5876 | link_list); | |
5d960724 | 5877 | if (refcount_inc_not_zero(&prev->refs)) { |
4493233e | 5878 | list_del_init(&req->link_list); |
5d960724 JA |
5879 | prev->flags &= ~REQ_F_LINK_TIMEOUT; |
5880 | } else | |
76a46e06 | 5881 | prev = NULL; |
2665abfd JA |
5882 | } |
5883 | ||
5884 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
5885 | ||
5886 | if (prev) { | |
4e88d6e7 | 5887 | req_set_fail_links(prev); |
014db007 | 5888 | io_async_find_and_cancel(ctx, req, prev->user_data, -ETIME); |
76a46e06 | 5889 | io_put_req(prev); |
47f46768 | 5890 | } else { |
e1e16097 | 5891 | io_req_complete(req, -ETIME); |
2665abfd | 5892 | } |
2665abfd JA |
5893 | return HRTIMER_NORESTART; |
5894 | } | |
5895 | ||
ad8a48ac | 5896 | static void io_queue_linked_timeout(struct io_kiocb *req) |
2665abfd | 5897 | { |
76a46e06 | 5898 | struct io_ring_ctx *ctx = req->ctx; |
2665abfd | 5899 | |
76a46e06 JA |
5900 | /* |
5901 | * If the list is now empty, then our linked request finished before | |
5902 | * we got a chance to setup the timer | |
5903 | */ | |
5904 | spin_lock_irq(&ctx->completion_lock); | |
4493233e | 5905 | if (!list_empty(&req->link_list)) { |
2d28390a | 5906 | struct io_timeout_data *data = &req->io->timeout; |
94ae5e77 | 5907 | |
ad8a48ac JA |
5908 | data->timer.function = io_link_timeout_fn; |
5909 | hrtimer_start(&data->timer, timespec64_to_ktime(data->ts), | |
5910 | data->mode); | |
2665abfd | 5911 | } |
76a46e06 | 5912 | spin_unlock_irq(&ctx->completion_lock); |
2665abfd | 5913 | |
2665abfd | 5914 | /* drop submission reference */ |
76a46e06 JA |
5915 | io_put_req(req); |
5916 | } | |
2665abfd | 5917 | |
ad8a48ac | 5918 | static struct io_kiocb *io_prep_linked_timeout(struct io_kiocb *req) |
2665abfd JA |
5919 | { |
5920 | struct io_kiocb *nxt; | |
5921 | ||
dea3b49c | 5922 | if (!(req->flags & REQ_F_LINK_HEAD)) |
2665abfd | 5923 | return NULL; |
6df1db6b | 5924 | if (req->flags & REQ_F_LINK_TIMEOUT) |
d7718a9d | 5925 | return NULL; |
2665abfd | 5926 | |
4493233e PB |
5927 | nxt = list_first_entry_or_null(&req->link_list, struct io_kiocb, |
5928 | link_list); | |
d625c6ee | 5929 | if (!nxt || nxt->opcode != IORING_OP_LINK_TIMEOUT) |
76a46e06 | 5930 | return NULL; |
2665abfd | 5931 | |
76a46e06 | 5932 | req->flags |= REQ_F_LINK_TIMEOUT; |
76a46e06 | 5933 | return nxt; |
2665abfd JA |
5934 | } |
5935 | ||
f13fad7b JA |
5936 | static void __io_queue_sqe(struct io_kiocb *req, const struct io_uring_sqe *sqe, |
5937 | struct io_comp_state *cs) | |
2b188cc1 | 5938 | { |
4a0a7a18 | 5939 | struct io_kiocb *linked_timeout; |
4bc4494e | 5940 | struct io_kiocb *nxt; |
193155c8 | 5941 | const struct cred *old_creds = NULL; |
e0c5c576 | 5942 | int ret; |
2b188cc1 | 5943 | |
4a0a7a18 JA |
5944 | again: |
5945 | linked_timeout = io_prep_linked_timeout(req); | |
5946 | ||
7cdaf587 XW |
5947 | if ((req->flags & REQ_F_WORK_INITIALIZED) && req->work.creds && |
5948 | req->work.creds != current_cred()) { | |
193155c8 JA |
5949 | if (old_creds) |
5950 | revert_creds(old_creds); | |
5951 | if (old_creds == req->work.creds) | |
5952 | old_creds = NULL; /* restored original creds */ | |
5953 | else | |
5954 | old_creds = override_creds(req->work.creds); | |
5955 | } | |
5956 | ||
f13fad7b | 5957 | ret = io_issue_sqe(req, sqe, true, cs); |
491381ce JA |
5958 | |
5959 | /* | |
5960 | * We async punt it if the file wasn't marked NOWAIT, or if the file | |
5961 | * doesn't support non-blocking read/write attempts | |
5962 | */ | |
24c74678 | 5963 | if (ret == -EAGAIN && !(req->flags & REQ_F_NOWAIT)) { |
d7718a9d JA |
5964 | if (io_arm_poll_handler(req)) { |
5965 | if (linked_timeout) | |
5966 | io_queue_linked_timeout(linked_timeout); | |
4bc4494e | 5967 | goto exit; |
d7718a9d | 5968 | } |
86a761f8 | 5969 | punt: |
7cdaf587 XW |
5970 | io_req_init_async(req); |
5971 | ||
f86cd20c | 5972 | if (io_op_defs[req->opcode].file_table) { |
bbad27b2 PB |
5973 | ret = io_grab_files(req); |
5974 | if (ret) | |
5975 | goto err; | |
2b188cc1 | 5976 | } |
bbad27b2 PB |
5977 | |
5978 | /* | |
5979 | * Queued up for async execution, worker will release | |
5980 | * submit reference when the iocb is actually submitted. | |
5981 | */ | |
5982 | io_queue_async_work(req); | |
4bc4494e | 5983 | goto exit; |
2b188cc1 | 5984 | } |
e65ef56d | 5985 | |
652532ad | 5986 | if (unlikely(ret)) { |
fcb323cc | 5987 | err: |
652532ad PB |
5988 | /* un-prep timeout, so it'll be killed as any other linked */ |
5989 | req->flags &= ~REQ_F_LINK_TIMEOUT; | |
4e88d6e7 | 5990 | req_set_fail_links(req); |
652532ad | 5991 | io_put_req(req); |
e1e16097 | 5992 | io_req_complete(req, ret); |
652532ad | 5993 | goto exit; |
9e645e11 | 5994 | } |
652532ad PB |
5995 | |
5996 | /* drop submission reference */ | |
5997 | nxt = io_put_req_find_next(req); | |
5998 | if (linked_timeout) | |
5999 | io_queue_linked_timeout(linked_timeout); | |
6000 | ||
4a0a7a18 JA |
6001 | if (nxt) { |
6002 | req = nxt; | |
86a761f8 PB |
6003 | |
6004 | if (req->flags & REQ_F_FORCE_ASYNC) | |
6005 | goto punt; | |
4a0a7a18 JA |
6006 | goto again; |
6007 | } | |
4bc4494e | 6008 | exit: |
193155c8 JA |
6009 | if (old_creds) |
6010 | revert_creds(old_creds); | |
2b188cc1 JA |
6011 | } |
6012 | ||
f13fad7b JA |
6013 | static void io_queue_sqe(struct io_kiocb *req, const struct io_uring_sqe *sqe, |
6014 | struct io_comp_state *cs) | |
4fe2c963 JL |
6015 | { |
6016 | int ret; | |
6017 | ||
3529d8c2 | 6018 | ret = io_req_defer(req, sqe); |
4fe2c963 JL |
6019 | if (ret) { |
6020 | if (ret != -EIOCBQUEUED) { | |
1118591a | 6021 | fail_req: |
4e88d6e7 | 6022 | req_set_fail_links(req); |
e1e16097 JA |
6023 | io_put_req(req); |
6024 | io_req_complete(req, ret); | |
4fe2c963 | 6025 | } |
2550878f | 6026 | } else if (req->flags & REQ_F_FORCE_ASYNC) { |
bd2ab18a | 6027 | if (!req->io) { |
debb85f4 | 6028 | ret = io_req_defer_prep(req, sqe); |
327d6d96 | 6029 | if (unlikely(ret)) |
bd2ab18a PB |
6030 | goto fail_req; |
6031 | } | |
6032 | ||
ce35a47a JA |
6033 | /* |
6034 | * Never try inline submit of IOSQE_ASYNC is set, go straight | |
6035 | * to async execution. | |
6036 | */ | |
3e863ea3 | 6037 | io_req_init_async(req); |
ce35a47a JA |
6038 | req->work.flags |= IO_WQ_WORK_CONCURRENT; |
6039 | io_queue_async_work(req); | |
6040 | } else { | |
f13fad7b | 6041 | __io_queue_sqe(req, sqe, cs); |
ce35a47a | 6042 | } |
4fe2c963 JL |
6043 | } |
6044 | ||
f13fad7b JA |
6045 | static inline void io_queue_link_head(struct io_kiocb *req, |
6046 | struct io_comp_state *cs) | |
4fe2c963 | 6047 | { |
94ae5e77 | 6048 | if (unlikely(req->flags & REQ_F_FAIL_LINK)) { |
e1e16097 JA |
6049 | io_put_req(req); |
6050 | io_req_complete(req, -ECANCELED); | |
1b4a51b6 | 6051 | } else |
f13fad7b | 6052 | io_queue_sqe(req, NULL, cs); |
4fe2c963 JL |
6053 | } |
6054 | ||
1d4240cc | 6055 | static int io_submit_sqe(struct io_kiocb *req, const struct io_uring_sqe *sqe, |
f13fad7b | 6056 | struct io_kiocb **link, struct io_comp_state *cs) |
9e645e11 | 6057 | { |
a197f664 | 6058 | struct io_ring_ctx *ctx = req->ctx; |
ef4ff581 | 6059 | int ret; |
9e645e11 | 6060 | |
9e645e11 JA |
6061 | /* |
6062 | * If we already have a head request, queue this one for async | |
6063 | * submittal once the head completes. If we don't have a head but | |
6064 | * IOSQE_IO_LINK is set in the sqe, start a new head. This one will be | |
6065 | * submitted sync once the chain is complete. If none of those | |
6066 | * conditions are true (normal request), then just queue it. | |
6067 | */ | |
6068 | if (*link) { | |
9d76377f | 6069 | struct io_kiocb *head = *link; |
4e88d6e7 | 6070 | |
8cdf2193 PB |
6071 | /* |
6072 | * Taking sequential execution of a link, draining both sides | |
6073 | * of the link also fullfils IOSQE_IO_DRAIN semantics for all | |
6074 | * requests in the link. So, it drains the head and the | |
6075 | * next after the link request. The last one is done via | |
6076 | * drain_next flag to persist the effect across calls. | |
6077 | */ | |
ef4ff581 | 6078 | if (req->flags & REQ_F_IO_DRAIN) { |
711be031 PB |
6079 | head->flags |= REQ_F_IO_DRAIN; |
6080 | ctx->drain_next = 1; | |
6081 | } | |
debb85f4 | 6082 | ret = io_req_defer_prep(req, sqe); |
327d6d96 | 6083 | if (unlikely(ret)) { |
4e88d6e7 | 6084 | /* fail even hard links since we don't submit */ |
9d76377f | 6085 | head->flags |= REQ_F_FAIL_LINK; |
1d4240cc | 6086 | return ret; |
2d28390a | 6087 | } |
9d76377f | 6088 | trace_io_uring_link(ctx, req, head); |
c40f6379 | 6089 | io_get_req_task(req); |
9d76377f | 6090 | list_add_tail(&req->link_list, &head->link_list); |
32fe525b PB |
6091 | |
6092 | /* last request of a link, enqueue the link */ | |
ef4ff581 | 6093 | if (!(req->flags & (REQ_F_LINK | REQ_F_HARDLINK))) { |
f13fad7b | 6094 | io_queue_link_head(head, cs); |
32fe525b PB |
6095 | *link = NULL; |
6096 | } | |
9e645e11 | 6097 | } else { |
711be031 PB |
6098 | if (unlikely(ctx->drain_next)) { |
6099 | req->flags |= REQ_F_IO_DRAIN; | |
ef4ff581 | 6100 | ctx->drain_next = 0; |
711be031 | 6101 | } |
ef4ff581 | 6102 | if (req->flags & (REQ_F_LINK | REQ_F_HARDLINK)) { |
dea3b49c | 6103 | req->flags |= REQ_F_LINK_HEAD; |
711be031 | 6104 | INIT_LIST_HEAD(&req->link_list); |
f1d96a8f | 6105 | |
debb85f4 | 6106 | ret = io_req_defer_prep(req, sqe); |
327d6d96 | 6107 | if (unlikely(ret)) |
711be031 PB |
6108 | req->flags |= REQ_F_FAIL_LINK; |
6109 | *link = req; | |
6110 | } else { | |
f13fad7b | 6111 | io_queue_sqe(req, sqe, cs); |
711be031 | 6112 | } |
9e645e11 | 6113 | } |
2e6e1fde | 6114 | |
1d4240cc | 6115 | return 0; |
9e645e11 JA |
6116 | } |
6117 | ||
9a56a232 JA |
6118 | /* |
6119 | * Batched submission is done, ensure local IO is flushed out. | |
6120 | */ | |
6121 | static void io_submit_state_end(struct io_submit_state *state) | |
6122 | { | |
f13fad7b JA |
6123 | if (!list_empty(&state->comp.list)) |
6124 | io_submit_flush_completions(&state->comp); | |
9a56a232 | 6125 | blk_finish_plug(&state->plug); |
9f13c35b | 6126 | io_state_file_put(state); |
2579f913 | 6127 | if (state->free_reqs) |
6c8a3134 | 6128 | kmem_cache_free_bulk(req_cachep, state->free_reqs, state->reqs); |
9a56a232 JA |
6129 | } |
6130 | ||
6131 | /* | |
6132 | * Start submission side cache. | |
6133 | */ | |
6134 | static void io_submit_state_start(struct io_submit_state *state, | |
013538bd | 6135 | struct io_ring_ctx *ctx, unsigned int max_ios) |
9a56a232 JA |
6136 | { |
6137 | blk_start_plug(&state->plug); | |
b63534c4 JA |
6138 | #ifdef CONFIG_BLOCK |
6139 | state->plug.nowait = true; | |
6140 | #endif | |
013538bd JA |
6141 | state->comp.nr = 0; |
6142 | INIT_LIST_HEAD(&state->comp.list); | |
6143 | state->comp.ctx = ctx; | |
2579f913 | 6144 | state->free_reqs = 0; |
9a56a232 JA |
6145 | state->file = NULL; |
6146 | state->ios_left = max_ios; | |
6147 | } | |
6148 | ||
2b188cc1 JA |
6149 | static void io_commit_sqring(struct io_ring_ctx *ctx) |
6150 | { | |
75b28aff | 6151 | struct io_rings *rings = ctx->rings; |
2b188cc1 | 6152 | |
caf582c6 PB |
6153 | /* |
6154 | * Ensure any loads from the SQEs are done at this point, | |
6155 | * since once we write the new head, the application could | |
6156 | * write new data to them. | |
6157 | */ | |
6158 | smp_store_release(&rings->sq.head, ctx->cached_sq_head); | |
2b188cc1 JA |
6159 | } |
6160 | ||
2b188cc1 | 6161 | /* |
3529d8c2 | 6162 | * Fetch an sqe, if one is available. Note that sqe_ptr will point to memory |
2b188cc1 JA |
6163 | * that is mapped by userspace. This means that care needs to be taken to |
6164 | * ensure that reads are stable, as we cannot rely on userspace always | |
6165 | * being a good citizen. If members of the sqe are validated and then later | |
6166 | * used, it's important that those reads are done through READ_ONCE() to | |
6167 | * prevent a re-load down the line. | |
6168 | */ | |
709b302f | 6169 | static const struct io_uring_sqe *io_get_sqe(struct io_ring_ctx *ctx) |
2b188cc1 | 6170 | { |
75b28aff | 6171 | u32 *sq_array = ctx->sq_array; |
2b188cc1 JA |
6172 | unsigned head; |
6173 | ||
6174 | /* | |
6175 | * The cached sq head (or cq tail) serves two purposes: | |
6176 | * | |
6177 | * 1) allows us to batch the cost of updating the user visible | |
6178 | * head updates. | |
6179 | * 2) allows the kernel side to track the head on its own, even | |
6180 | * though the application is the one updating it. | |
6181 | */ | |
ee7d46d9 | 6182 | head = READ_ONCE(sq_array[ctx->cached_sq_head & ctx->sq_mask]); |
709b302f PB |
6183 | if (likely(head < ctx->sq_entries)) |
6184 | return &ctx->sq_sqes[head]; | |
2b188cc1 JA |
6185 | |
6186 | /* drop invalid entries */ | |
498ccd9e | 6187 | ctx->cached_sq_dropped++; |
ee7d46d9 | 6188 | WRITE_ONCE(ctx->rings->sq_dropped, ctx->cached_sq_dropped); |
709b302f PB |
6189 | return NULL; |
6190 | } | |
6191 | ||
6192 | static inline void io_consume_sqe(struct io_ring_ctx *ctx) | |
6193 | { | |
6194 | ctx->cached_sq_head++; | |
2b188cc1 JA |
6195 | } |
6196 | ||
ef4ff581 PB |
6197 | #define SQE_VALID_FLAGS (IOSQE_FIXED_FILE|IOSQE_IO_DRAIN|IOSQE_IO_LINK| \ |
6198 | IOSQE_IO_HARDLINK | IOSQE_ASYNC | \ | |
6199 | IOSQE_BUFFER_SELECT) | |
6200 | ||
6201 | static int io_init_req(struct io_ring_ctx *ctx, struct io_kiocb *req, | |
6202 | const struct io_uring_sqe *sqe, | |
0cdaf760 | 6203 | struct io_submit_state *state) |
0553b8bd | 6204 | { |
ef4ff581 | 6205 | unsigned int sqe_flags; |
63ff8223 | 6206 | int id; |
ef4ff581 | 6207 | |
0553b8bd PB |
6208 | req->opcode = READ_ONCE(sqe->opcode); |
6209 | req->user_data = READ_ONCE(sqe->user_data); | |
6210 | req->io = NULL; | |
6211 | req->file = NULL; | |
6212 | req->ctx = ctx; | |
6213 | req->flags = 0; | |
6214 | /* one is dropped after submission, the other at completion */ | |
6215 | refcount_set(&req->refs, 2); | |
4dd2824d | 6216 | req->task = current; |
0553b8bd | 6217 | req->result = 0; |
ef4ff581 PB |
6218 | |
6219 | if (unlikely(req->opcode >= IORING_OP_LAST)) | |
6220 | return -EINVAL; | |
6221 | ||
9d8426a0 JA |
6222 | if (unlikely(io_sq_thread_acquire_mm(ctx, req))) |
6223 | return -EFAULT; | |
ef4ff581 PB |
6224 | |
6225 | sqe_flags = READ_ONCE(sqe->flags); | |
6226 | /* enforce forwards compatibility on users */ | |
6227 | if (unlikely(sqe_flags & ~SQE_VALID_FLAGS)) | |
6228 | return -EINVAL; | |
6229 | ||
6230 | if ((sqe_flags & IOSQE_BUFFER_SELECT) && | |
6231 | !io_op_defs[req->opcode].buffer_select) | |
6232 | return -EOPNOTSUPP; | |
6233 | ||
6234 | id = READ_ONCE(sqe->personality); | |
6235 | if (id) { | |
7cdaf587 | 6236 | io_req_init_async(req); |
ef4ff581 PB |
6237 | req->work.creds = idr_find(&ctx->personality_idr, id); |
6238 | if (unlikely(!req->work.creds)) | |
6239 | return -EINVAL; | |
6240 | get_cred(req->work.creds); | |
6241 | } | |
6242 | ||
6243 | /* same numerical values with corresponding REQ_F_*, safe to copy */ | |
c11368a5 | 6244 | req->flags |= sqe_flags; |
ef4ff581 | 6245 | |
63ff8223 JA |
6246 | if (!io_op_defs[req->opcode].needs_file) |
6247 | return 0; | |
6248 | ||
6249 | return io_req_set_file(state, req, READ_ONCE(sqe->fd)); | |
0553b8bd PB |
6250 | } |
6251 | ||
fb5ccc98 | 6252 | static int io_submit_sqes(struct io_ring_ctx *ctx, unsigned int nr, |
0cdaf760 | 6253 | struct file *ring_file, int ring_fd) |
6c271ce2 | 6254 | { |
ac8691c4 | 6255 | struct io_submit_state state; |
9e645e11 | 6256 | struct io_kiocb *link = NULL; |
9e645e11 | 6257 | int i, submitted = 0; |
6c271ce2 | 6258 | |
c4a2ed72 | 6259 | /* if we have a backlog and couldn't flush it all, return BUSY */ |
ad3eb2c8 JA |
6260 | if (test_bit(0, &ctx->sq_check_overflow)) { |
6261 | if (!list_empty(&ctx->cq_overflow_list) && | |
6262 | !io_cqring_overflow_flush(ctx, false)) | |
6263 | return -EBUSY; | |
6264 | } | |
6c271ce2 | 6265 | |
ee7d46d9 PB |
6266 | /* make sure SQ entry isn't read before tail */ |
6267 | nr = min3(nr, ctx->sq_entries, io_sqring_entries(ctx)); | |
9ef4f124 | 6268 | |
2b85edfc PB |
6269 | if (!percpu_ref_tryget_many(&ctx->refs, nr)) |
6270 | return -EAGAIN; | |
6c271ce2 | 6271 | |
013538bd | 6272 | io_submit_state_start(&state, ctx, nr); |
6c271ce2 | 6273 | |
b14cca0c PB |
6274 | ctx->ring_fd = ring_fd; |
6275 | ctx->ring_file = ring_file; | |
6276 | ||
6c271ce2 | 6277 | for (i = 0; i < nr; i++) { |
3529d8c2 | 6278 | const struct io_uring_sqe *sqe; |
196be95c | 6279 | struct io_kiocb *req; |
1cb1edb2 | 6280 | int err; |
fb5ccc98 | 6281 | |
b1e50e54 PB |
6282 | sqe = io_get_sqe(ctx); |
6283 | if (unlikely(!sqe)) { | |
6284 | io_consume_sqe(ctx); | |
6285 | break; | |
6286 | } | |
ac8691c4 | 6287 | req = io_alloc_req(ctx, &state); |
196be95c PB |
6288 | if (unlikely(!req)) { |
6289 | if (!submitted) | |
6290 | submitted = -EAGAIN; | |
fb5ccc98 | 6291 | break; |
196be95c | 6292 | } |
fb5ccc98 | 6293 | |
ac8691c4 | 6294 | err = io_init_req(ctx, req, sqe, &state); |
709b302f | 6295 | io_consume_sqe(ctx); |
d3656344 JA |
6296 | /* will complete beyond this point, count as submitted */ |
6297 | submitted++; | |
6298 | ||
ef4ff581 | 6299 | if (unlikely(err)) { |
1cb1edb2 | 6300 | fail_req: |
e1e16097 JA |
6301 | io_put_req(req); |
6302 | io_req_complete(req, err); | |
196be95c PB |
6303 | break; |
6304 | } | |
fb5ccc98 | 6305 | |
354420f7 | 6306 | trace_io_uring_submit_sqe(ctx, req->opcode, req->user_data, |
0cdaf760 | 6307 | true, io_async_submit(ctx)); |
f13fad7b | 6308 | err = io_submit_sqe(req, sqe, &link, &state.comp); |
1d4240cc PB |
6309 | if (err) |
6310 | goto fail_req; | |
6c271ce2 JA |
6311 | } |
6312 | ||
9466f437 PB |
6313 | if (unlikely(submitted != nr)) { |
6314 | int ref_used = (submitted == -EAGAIN) ? 0 : submitted; | |
6315 | ||
6316 | percpu_ref_put_many(&ctx->refs, nr - ref_used); | |
6317 | } | |
9e645e11 | 6318 | if (link) |
f13fad7b | 6319 | io_queue_link_head(link, &state.comp); |
ac8691c4 | 6320 | io_submit_state_end(&state); |
6c271ce2 | 6321 | |
ae9428ca PB |
6322 | /* Commit SQ ring head once we've consumed and submitted all SQEs */ |
6323 | io_commit_sqring(ctx); | |
6324 | ||
6c271ce2 JA |
6325 | return submitted; |
6326 | } | |
6327 | ||
6328 | static int io_sq_thread(void *data) | |
6329 | { | |
6c271ce2 | 6330 | struct io_ring_ctx *ctx = data; |
181e448d | 6331 | const struct cred *old_cred; |
6c271ce2 | 6332 | DEFINE_WAIT(wait); |
6c271ce2 | 6333 | unsigned long timeout; |
bdcd3eab | 6334 | int ret = 0; |
6c271ce2 | 6335 | |
0f158b4c | 6336 | complete(&ctx->sq_thread_comp); |
a4c0b3de | 6337 | |
181e448d | 6338 | old_cred = override_creds(ctx->creds); |
6c271ce2 | 6339 | |
bdcd3eab | 6340 | timeout = jiffies + ctx->sq_thread_idle; |
2bbcd6d3 | 6341 | while (!kthread_should_park()) { |
fb5ccc98 | 6342 | unsigned int to_submit; |
6c271ce2 | 6343 | |
540e32a0 | 6344 | if (!list_empty(&ctx->iopoll_list)) { |
6c271ce2 JA |
6345 | unsigned nr_events = 0; |
6346 | ||
bdcd3eab | 6347 | mutex_lock(&ctx->uring_lock); |
540e32a0 | 6348 | if (!list_empty(&ctx->iopoll_list) && !need_resched()) |
9dedd563 | 6349 | io_do_iopoll(ctx, &nr_events, 0); |
bdcd3eab | 6350 | else |
6c271ce2 | 6351 | timeout = jiffies + ctx->sq_thread_idle; |
bdcd3eab | 6352 | mutex_unlock(&ctx->uring_lock); |
6c271ce2 JA |
6353 | } |
6354 | ||
fb5ccc98 | 6355 | to_submit = io_sqring_entries(ctx); |
c1edbf5f JA |
6356 | |
6357 | /* | |
6358 | * If submit got -EBUSY, flag us as needing the application | |
6359 | * to enter the kernel to reap and flush events. | |
6360 | */ | |
b772f07a | 6361 | if (!to_submit || ret == -EBUSY || need_resched()) { |
7143b5ac SG |
6362 | /* |
6363 | * Drop cur_mm before scheduling, we can't hold it for | |
6364 | * long periods (or over schedule()). Do this before | |
6365 | * adding ourselves to the waitqueue, as the unuse/drop | |
6366 | * may sleep. | |
6367 | */ | |
4349f30e | 6368 | io_sq_thread_drop_mm(); |
7143b5ac | 6369 | |
6c271ce2 JA |
6370 | /* |
6371 | * We're polling. If we're within the defined idle | |
6372 | * period, then let us spin without work before going | |
c1edbf5f JA |
6373 | * to sleep. The exception is if we got EBUSY doing |
6374 | * more IO, we should wait for the application to | |
6375 | * reap events and wake us up. | |
6c271ce2 | 6376 | */ |
540e32a0 | 6377 | if (!list_empty(&ctx->iopoll_list) || need_resched() || |
df069d80 JA |
6378 | (!time_after(jiffies, timeout) && ret != -EBUSY && |
6379 | !percpu_ref_is_dying(&ctx->refs))) { | |
4c6e277c | 6380 | io_run_task_work(); |
9831a90c | 6381 | cond_resched(); |
6c271ce2 JA |
6382 | continue; |
6383 | } | |
6384 | ||
6c271ce2 JA |
6385 | prepare_to_wait(&ctx->sqo_wait, &wait, |
6386 | TASK_INTERRUPTIBLE); | |
6387 | ||
bdcd3eab XW |
6388 | /* |
6389 | * While doing polled IO, before going to sleep, we need | |
540e32a0 PB |
6390 | * to check if there are new reqs added to iopoll_list, |
6391 | * it is because reqs may have been punted to io worker | |
6392 | * and will be added to iopoll_list later, hence check | |
6393 | * the iopoll_list again. | |
bdcd3eab XW |
6394 | */ |
6395 | if ((ctx->flags & IORING_SETUP_IOPOLL) && | |
540e32a0 | 6396 | !list_empty_careful(&ctx->iopoll_list)) { |
bdcd3eab XW |
6397 | finish_wait(&ctx->sqo_wait, &wait); |
6398 | continue; | |
6399 | } | |
6400 | ||
6c271ce2 | 6401 | /* Tell userspace we may need a wakeup call */ |
6d5f9049 | 6402 | spin_lock_irq(&ctx->completion_lock); |
75b28aff | 6403 | ctx->rings->sq_flags |= IORING_SQ_NEED_WAKEUP; |
6d5f9049 | 6404 | spin_unlock_irq(&ctx->completion_lock); |
6c271ce2 | 6405 | |
fb5ccc98 | 6406 | to_submit = io_sqring_entries(ctx); |
c1edbf5f | 6407 | if (!to_submit || ret == -EBUSY) { |
2bbcd6d3 | 6408 | if (kthread_should_park()) { |
6c271ce2 JA |
6409 | finish_wait(&ctx->sqo_wait, &wait); |
6410 | break; | |
6411 | } | |
4c6e277c | 6412 | if (io_run_task_work()) { |
10bea96d | 6413 | finish_wait(&ctx->sqo_wait, &wait); |
b41e9852 JA |
6414 | continue; |
6415 | } | |
6c271ce2 JA |
6416 | if (signal_pending(current)) |
6417 | flush_signals(current); | |
6418 | schedule(); | |
6419 | finish_wait(&ctx->sqo_wait, &wait); | |
6420 | ||
6d5f9049 | 6421 | spin_lock_irq(&ctx->completion_lock); |
75b28aff | 6422 | ctx->rings->sq_flags &= ~IORING_SQ_NEED_WAKEUP; |
6d5f9049 | 6423 | spin_unlock_irq(&ctx->completion_lock); |
d4ae271d | 6424 | ret = 0; |
6c271ce2 JA |
6425 | continue; |
6426 | } | |
6427 | finish_wait(&ctx->sqo_wait, &wait); | |
6428 | ||
6d5f9049 | 6429 | spin_lock_irq(&ctx->completion_lock); |
75b28aff | 6430 | ctx->rings->sq_flags &= ~IORING_SQ_NEED_WAKEUP; |
6d5f9049 | 6431 | spin_unlock_irq(&ctx->completion_lock); |
6c271ce2 JA |
6432 | } |
6433 | ||
8a4955ff | 6434 | mutex_lock(&ctx->uring_lock); |
6b668c9b XW |
6435 | if (likely(!percpu_ref_is_dying(&ctx->refs))) |
6436 | ret = io_submit_sqes(ctx, to_submit, NULL, -1); | |
8a4955ff | 6437 | mutex_unlock(&ctx->uring_lock); |
bdcd3eab | 6438 | timeout = jiffies + ctx->sq_thread_idle; |
6c271ce2 JA |
6439 | } |
6440 | ||
4c6e277c | 6441 | io_run_task_work(); |
b41e9852 | 6442 | |
4349f30e | 6443 | io_sq_thread_drop_mm(); |
181e448d | 6444 | revert_creds(old_cred); |
06058632 | 6445 | |
2bbcd6d3 | 6446 | kthread_parkme(); |
06058632 | 6447 | |
6c271ce2 JA |
6448 | return 0; |
6449 | } | |
6450 | ||
bda52162 JA |
6451 | struct io_wait_queue { |
6452 | struct wait_queue_entry wq; | |
6453 | struct io_ring_ctx *ctx; | |
6454 | unsigned to_wait; | |
6455 | unsigned nr_timeouts; | |
6456 | }; | |
6457 | ||
1d7bb1d5 | 6458 | static inline bool io_should_wake(struct io_wait_queue *iowq, bool noflush) |
bda52162 JA |
6459 | { |
6460 | struct io_ring_ctx *ctx = iowq->ctx; | |
6461 | ||
6462 | /* | |
d195a66e | 6463 | * Wake up if we have enough events, or if a timeout occurred since we |
bda52162 JA |
6464 | * started waiting. For timeouts, we always want to return to userspace, |
6465 | * regardless of event count. | |
6466 | */ | |
1d7bb1d5 | 6467 | return io_cqring_events(ctx, noflush) >= iowq->to_wait || |
bda52162 JA |
6468 | atomic_read(&ctx->cq_timeouts) != iowq->nr_timeouts; |
6469 | } | |
6470 | ||
6471 | static int io_wake_function(struct wait_queue_entry *curr, unsigned int mode, | |
6472 | int wake_flags, void *key) | |
6473 | { | |
6474 | struct io_wait_queue *iowq = container_of(curr, struct io_wait_queue, | |
6475 | wq); | |
6476 | ||
1d7bb1d5 JA |
6477 | /* use noflush == true, as we can't safely rely on locking context */ |
6478 | if (!io_should_wake(iowq, true)) | |
bda52162 JA |
6479 | return -1; |
6480 | ||
6481 | return autoremove_wake_function(curr, mode, wake_flags, key); | |
6482 | } | |
6483 | ||
2b188cc1 JA |
6484 | /* |
6485 | * Wait until events become available, if we don't already have some. The | |
6486 | * application must reap them itself, as they reside on the shared cq ring. | |
6487 | */ | |
6488 | static int io_cqring_wait(struct io_ring_ctx *ctx, int min_events, | |
6489 | const sigset_t __user *sig, size_t sigsz) | |
6490 | { | |
bda52162 JA |
6491 | struct io_wait_queue iowq = { |
6492 | .wq = { | |
6493 | .private = current, | |
6494 | .func = io_wake_function, | |
6495 | .entry = LIST_HEAD_INIT(iowq.wq.entry), | |
6496 | }, | |
6497 | .ctx = ctx, | |
6498 | .to_wait = min_events, | |
6499 | }; | |
75b28aff | 6500 | struct io_rings *rings = ctx->rings; |
e9ffa5c2 | 6501 | int ret = 0; |
2b188cc1 | 6502 | |
b41e9852 JA |
6503 | do { |
6504 | if (io_cqring_events(ctx, false) >= min_events) | |
6505 | return 0; | |
4c6e277c | 6506 | if (!io_run_task_work()) |
b41e9852 | 6507 | break; |
b41e9852 | 6508 | } while (1); |
2b188cc1 JA |
6509 | |
6510 | if (sig) { | |
9e75ad5d AB |
6511 | #ifdef CONFIG_COMPAT |
6512 | if (in_compat_syscall()) | |
6513 | ret = set_compat_user_sigmask((const compat_sigset_t __user *)sig, | |
b772434b | 6514 | sigsz); |
9e75ad5d AB |
6515 | else |
6516 | #endif | |
b772434b | 6517 | ret = set_user_sigmask(sig, sigsz); |
9e75ad5d | 6518 | |
2b188cc1 JA |
6519 | if (ret) |
6520 | return ret; | |
6521 | } | |
6522 | ||
bda52162 | 6523 | iowq.nr_timeouts = atomic_read(&ctx->cq_timeouts); |
c826bd7a | 6524 | trace_io_uring_cqring_wait(ctx, min_events); |
bda52162 JA |
6525 | do { |
6526 | prepare_to_wait_exclusive(&ctx->wait, &iowq.wq, | |
6527 | TASK_INTERRUPTIBLE); | |
ce593a6c | 6528 | /* make sure we run task_work before checking for signals */ |
4c6e277c JA |
6529 | if (io_run_task_work()) |
6530 | continue; | |
bda52162 | 6531 | if (signal_pending(current)) { |
b7db41c9 JA |
6532 | if (current->jobctl & JOBCTL_TASK_WORK) { |
6533 | spin_lock_irq(¤t->sighand->siglock); | |
6534 | current->jobctl &= ~JOBCTL_TASK_WORK; | |
6535 | recalc_sigpending(); | |
6536 | spin_unlock_irq(¤t->sighand->siglock); | |
6537 | continue; | |
6538 | } | |
e9ffa5c2 | 6539 | ret = -EINTR; |
bda52162 JA |
6540 | break; |
6541 | } | |
ce593a6c JA |
6542 | if (io_should_wake(&iowq, false)) |
6543 | break; | |
6544 | schedule(); | |
bda52162 JA |
6545 | } while (1); |
6546 | finish_wait(&ctx->wait, &iowq.wq); | |
6547 | ||
e9ffa5c2 | 6548 | restore_saved_sigmask_unless(ret == -EINTR); |
2b188cc1 | 6549 | |
75b28aff | 6550 | return READ_ONCE(rings->cq.head) == READ_ONCE(rings->cq.tail) ? ret : 0; |
2b188cc1 JA |
6551 | } |
6552 | ||
6b06314c JA |
6553 | static void __io_sqe_files_unregister(struct io_ring_ctx *ctx) |
6554 | { | |
6555 | #if defined(CONFIG_UNIX) | |
6556 | if (ctx->ring_sock) { | |
6557 | struct sock *sock = ctx->ring_sock->sk; | |
6558 | struct sk_buff *skb; | |
6559 | ||
6560 | while ((skb = skb_dequeue(&sock->sk_receive_queue)) != NULL) | |
6561 | kfree_skb(skb); | |
6562 | } | |
6563 | #else | |
6564 | int i; | |
6565 | ||
65e19f54 JA |
6566 | for (i = 0; i < ctx->nr_user_files; i++) { |
6567 | struct file *file; | |
6568 | ||
6569 | file = io_file_from_index(ctx, i); | |
6570 | if (file) | |
6571 | fput(file); | |
6572 | } | |
6b06314c JA |
6573 | #endif |
6574 | } | |
6575 | ||
05f3fb3c JA |
6576 | static void io_file_ref_kill(struct percpu_ref *ref) |
6577 | { | |
6578 | struct fixed_file_data *data; | |
6579 | ||
6580 | data = container_of(ref, struct fixed_file_data, refs); | |
6581 | complete(&data->done); | |
6582 | } | |
6583 | ||
6b06314c JA |
6584 | static int io_sqe_files_unregister(struct io_ring_ctx *ctx) |
6585 | { | |
05f3fb3c | 6586 | struct fixed_file_data *data = ctx->file_data; |
05589553 | 6587 | struct fixed_file_ref_node *ref_node = NULL; |
65e19f54 JA |
6588 | unsigned nr_tables, i; |
6589 | ||
05f3fb3c | 6590 | if (!data) |
6b06314c JA |
6591 | return -ENXIO; |
6592 | ||
6a4d07cd | 6593 | spin_lock(&data->lock); |
05589553 XW |
6594 | if (!list_empty(&data->ref_list)) |
6595 | ref_node = list_first_entry(&data->ref_list, | |
6596 | struct fixed_file_ref_node, node); | |
6a4d07cd | 6597 | spin_unlock(&data->lock); |
05589553 XW |
6598 | if (ref_node) |
6599 | percpu_ref_kill(&ref_node->refs); | |
6600 | ||
6601 | percpu_ref_kill(&data->refs); | |
6602 | ||
6603 | /* wait for all refs nodes to complete */ | |
4a38aed2 | 6604 | flush_delayed_work(&ctx->file_put_work); |
2faf852d | 6605 | wait_for_completion(&data->done); |
05f3fb3c | 6606 | |
6b06314c | 6607 | __io_sqe_files_unregister(ctx); |
65e19f54 JA |
6608 | nr_tables = DIV_ROUND_UP(ctx->nr_user_files, IORING_MAX_FILES_TABLE); |
6609 | for (i = 0; i < nr_tables; i++) | |
05f3fb3c JA |
6610 | kfree(data->table[i].files); |
6611 | kfree(data->table); | |
05589553 XW |
6612 | percpu_ref_exit(&data->refs); |
6613 | kfree(data); | |
05f3fb3c | 6614 | ctx->file_data = NULL; |
6b06314c JA |
6615 | ctx->nr_user_files = 0; |
6616 | return 0; | |
6617 | } | |
6618 | ||
6c271ce2 JA |
6619 | static void io_sq_thread_stop(struct io_ring_ctx *ctx) |
6620 | { | |
6621 | if (ctx->sqo_thread) { | |
0f158b4c | 6622 | wait_for_completion(&ctx->sq_thread_comp); |
2bbcd6d3 RP |
6623 | /* |
6624 | * The park is a bit of a work-around, without it we get | |
6625 | * warning spews on shutdown with SQPOLL set and affinity | |
6626 | * set to a single CPU. | |
6627 | */ | |
06058632 | 6628 | kthread_park(ctx->sqo_thread); |
6c271ce2 JA |
6629 | kthread_stop(ctx->sqo_thread); |
6630 | ctx->sqo_thread = NULL; | |
6631 | } | |
6632 | } | |
6633 | ||
6b06314c JA |
6634 | static void io_finish_async(struct io_ring_ctx *ctx) |
6635 | { | |
6c271ce2 JA |
6636 | io_sq_thread_stop(ctx); |
6637 | ||
561fb04a JA |
6638 | if (ctx->io_wq) { |
6639 | io_wq_destroy(ctx->io_wq); | |
6640 | ctx->io_wq = NULL; | |
6b06314c JA |
6641 | } |
6642 | } | |
6643 | ||
6644 | #if defined(CONFIG_UNIX) | |
6b06314c JA |
6645 | /* |
6646 | * Ensure the UNIX gc is aware of our file set, so we are certain that | |
6647 | * the io_uring can be safely unregistered on process exit, even if we have | |
6648 | * loops in the file referencing. | |
6649 | */ | |
6650 | static int __io_sqe_files_scm(struct io_ring_ctx *ctx, int nr, int offset) | |
6651 | { | |
6652 | struct sock *sk = ctx->ring_sock->sk; | |
6653 | struct scm_fp_list *fpl; | |
6654 | struct sk_buff *skb; | |
08a45173 | 6655 | int i, nr_files; |
6b06314c | 6656 | |
6b06314c JA |
6657 | fpl = kzalloc(sizeof(*fpl), GFP_KERNEL); |
6658 | if (!fpl) | |
6659 | return -ENOMEM; | |
6660 | ||
6661 | skb = alloc_skb(0, GFP_KERNEL); | |
6662 | if (!skb) { | |
6663 | kfree(fpl); | |
6664 | return -ENOMEM; | |
6665 | } | |
6666 | ||
6667 | skb->sk = sk; | |
6b06314c | 6668 | |
08a45173 | 6669 | nr_files = 0; |
6b06314c JA |
6670 | fpl->user = get_uid(ctx->user); |
6671 | for (i = 0; i < nr; i++) { | |
65e19f54 JA |
6672 | struct file *file = io_file_from_index(ctx, i + offset); |
6673 | ||
6674 | if (!file) | |
08a45173 | 6675 | continue; |
65e19f54 | 6676 | fpl->fp[nr_files] = get_file(file); |
08a45173 JA |
6677 | unix_inflight(fpl->user, fpl->fp[nr_files]); |
6678 | nr_files++; | |
6b06314c JA |
6679 | } |
6680 | ||
08a45173 JA |
6681 | if (nr_files) { |
6682 | fpl->max = SCM_MAX_FD; | |
6683 | fpl->count = nr_files; | |
6684 | UNIXCB(skb).fp = fpl; | |
05f3fb3c | 6685 | skb->destructor = unix_destruct_scm; |
08a45173 JA |
6686 | refcount_add(skb->truesize, &sk->sk_wmem_alloc); |
6687 | skb_queue_head(&sk->sk_receive_queue, skb); | |
6b06314c | 6688 | |
08a45173 JA |
6689 | for (i = 0; i < nr_files; i++) |
6690 | fput(fpl->fp[i]); | |
6691 | } else { | |
6692 | kfree_skb(skb); | |
6693 | kfree(fpl); | |
6694 | } | |
6b06314c JA |
6695 | |
6696 | return 0; | |
6697 | } | |
6698 | ||
6699 | /* | |
6700 | * If UNIX sockets are enabled, fd passing can cause a reference cycle which | |
6701 | * causes regular reference counting to break down. We rely on the UNIX | |
6702 | * garbage collection to take care of this problem for us. | |
6703 | */ | |
6704 | static int io_sqe_files_scm(struct io_ring_ctx *ctx) | |
6705 | { | |
6706 | unsigned left, total; | |
6707 | int ret = 0; | |
6708 | ||
6709 | total = 0; | |
6710 | left = ctx->nr_user_files; | |
6711 | while (left) { | |
6712 | unsigned this_files = min_t(unsigned, left, SCM_MAX_FD); | |
6b06314c JA |
6713 | |
6714 | ret = __io_sqe_files_scm(ctx, this_files, total); | |
6715 | if (ret) | |
6716 | break; | |
6717 | left -= this_files; | |
6718 | total += this_files; | |
6719 | } | |
6720 | ||
6721 | if (!ret) | |
6722 | return 0; | |
6723 | ||
6724 | while (total < ctx->nr_user_files) { | |
65e19f54 JA |
6725 | struct file *file = io_file_from_index(ctx, total); |
6726 | ||
6727 | if (file) | |
6728 | fput(file); | |
6b06314c JA |
6729 | total++; |
6730 | } | |
6731 | ||
6732 | return ret; | |
6733 | } | |
6734 | #else | |
6735 | static int io_sqe_files_scm(struct io_ring_ctx *ctx) | |
6736 | { | |
6737 | return 0; | |
6738 | } | |
6739 | #endif | |
6740 | ||
65e19f54 JA |
6741 | static int io_sqe_alloc_file_tables(struct io_ring_ctx *ctx, unsigned nr_tables, |
6742 | unsigned nr_files) | |
6743 | { | |
6744 | int i; | |
6745 | ||
6746 | for (i = 0; i < nr_tables; i++) { | |
05f3fb3c | 6747 | struct fixed_file_table *table = &ctx->file_data->table[i]; |
65e19f54 JA |
6748 | unsigned this_files; |
6749 | ||
6750 | this_files = min(nr_files, IORING_MAX_FILES_TABLE); | |
6751 | table->files = kcalloc(this_files, sizeof(struct file *), | |
6752 | GFP_KERNEL); | |
6753 | if (!table->files) | |
6754 | break; | |
6755 | nr_files -= this_files; | |
6756 | } | |
6757 | ||
6758 | if (i == nr_tables) | |
6759 | return 0; | |
6760 | ||
6761 | for (i = 0; i < nr_tables; i++) { | |
05f3fb3c | 6762 | struct fixed_file_table *table = &ctx->file_data->table[i]; |
65e19f54 JA |
6763 | kfree(table->files); |
6764 | } | |
6765 | return 1; | |
6766 | } | |
6767 | ||
05f3fb3c JA |
6768 | static void io_ring_file_put(struct io_ring_ctx *ctx, struct file *file) |
6769 | { | |
6770 | #if defined(CONFIG_UNIX) | |
6771 | struct sock *sock = ctx->ring_sock->sk; | |
6772 | struct sk_buff_head list, *head = &sock->sk_receive_queue; | |
6773 | struct sk_buff *skb; | |
6774 | int i; | |
6775 | ||
6776 | __skb_queue_head_init(&list); | |
6777 | ||
6778 | /* | |
6779 | * Find the skb that holds this file in its SCM_RIGHTS. When found, | |
6780 | * remove this entry and rearrange the file array. | |
6781 | */ | |
6782 | skb = skb_dequeue(head); | |
6783 | while (skb) { | |
6784 | struct scm_fp_list *fp; | |
6785 | ||
6786 | fp = UNIXCB(skb).fp; | |
6787 | for (i = 0; i < fp->count; i++) { | |
6788 | int left; | |
6789 | ||
6790 | if (fp->fp[i] != file) | |
6791 | continue; | |
6792 | ||
6793 | unix_notinflight(fp->user, fp->fp[i]); | |
6794 | left = fp->count - 1 - i; | |
6795 | if (left) { | |
6796 | memmove(&fp->fp[i], &fp->fp[i + 1], | |
6797 | left * sizeof(struct file *)); | |
6798 | } | |
6799 | fp->count--; | |
6800 | if (!fp->count) { | |
6801 | kfree_skb(skb); | |
6802 | skb = NULL; | |
6803 | } else { | |
6804 | __skb_queue_tail(&list, skb); | |
6805 | } | |
6806 | fput(file); | |
6807 | file = NULL; | |
6808 | break; | |
6809 | } | |
6810 | ||
6811 | if (!file) | |
6812 | break; | |
6813 | ||
6814 | __skb_queue_tail(&list, skb); | |
6815 | ||
6816 | skb = skb_dequeue(head); | |
6817 | } | |
6818 | ||
6819 | if (skb_peek(&list)) { | |
6820 | spin_lock_irq(&head->lock); | |
6821 | while ((skb = __skb_dequeue(&list)) != NULL) | |
6822 | __skb_queue_tail(head, skb); | |
6823 | spin_unlock_irq(&head->lock); | |
6824 | } | |
6825 | #else | |
6826 | fput(file); | |
6827 | #endif | |
6828 | } | |
6829 | ||
6830 | struct io_file_put { | |
05589553 | 6831 | struct list_head list; |
05f3fb3c | 6832 | struct file *file; |
05f3fb3c JA |
6833 | }; |
6834 | ||
4a38aed2 | 6835 | static void __io_file_put_work(struct fixed_file_ref_node *ref_node) |
65e19f54 | 6836 | { |
4a38aed2 JA |
6837 | struct fixed_file_data *file_data = ref_node->file_data; |
6838 | struct io_ring_ctx *ctx = file_data->ctx; | |
05f3fb3c | 6839 | struct io_file_put *pfile, *tmp; |
05589553 XW |
6840 | |
6841 | list_for_each_entry_safe(pfile, tmp, &ref_node->file_list, list) { | |
6a4d07cd | 6842 | list_del(&pfile->list); |
05589553 XW |
6843 | io_ring_file_put(ctx, pfile->file); |
6844 | kfree(pfile); | |
65e19f54 | 6845 | } |
05589553 | 6846 | |
6a4d07cd JA |
6847 | spin_lock(&file_data->lock); |
6848 | list_del(&ref_node->node); | |
6849 | spin_unlock(&file_data->lock); | |
05589553 XW |
6850 | |
6851 | percpu_ref_exit(&ref_node->refs); | |
6852 | kfree(ref_node); | |
6853 | percpu_ref_put(&file_data->refs); | |
2faf852d | 6854 | } |
65e19f54 | 6855 | |
4a38aed2 JA |
6856 | static void io_file_put_work(struct work_struct *work) |
6857 | { | |
6858 | struct io_ring_ctx *ctx; | |
6859 | struct llist_node *node; | |
6860 | ||
6861 | ctx = container_of(work, struct io_ring_ctx, file_put_work.work); | |
6862 | node = llist_del_all(&ctx->file_put_llist); | |
6863 | ||
6864 | while (node) { | |
6865 | struct fixed_file_ref_node *ref_node; | |
6866 | struct llist_node *next = node->next; | |
6867 | ||
6868 | ref_node = llist_entry(node, struct fixed_file_ref_node, llist); | |
6869 | __io_file_put_work(ref_node); | |
6870 | node = next; | |
6871 | } | |
6872 | } | |
6873 | ||
05589553 | 6874 | static void io_file_data_ref_zero(struct percpu_ref *ref) |
2faf852d | 6875 | { |
05589553 | 6876 | struct fixed_file_ref_node *ref_node; |
4a38aed2 JA |
6877 | struct io_ring_ctx *ctx; |
6878 | bool first_add; | |
6879 | int delay = HZ; | |
65e19f54 | 6880 | |
05589553 | 6881 | ref_node = container_of(ref, struct fixed_file_ref_node, refs); |
4a38aed2 | 6882 | ctx = ref_node->file_data->ctx; |
05589553 | 6883 | |
4a38aed2 JA |
6884 | if (percpu_ref_is_dying(&ctx->file_data->refs)) |
6885 | delay = 0; | |
05589553 | 6886 | |
4a38aed2 JA |
6887 | first_add = llist_add(&ref_node->llist, &ctx->file_put_llist); |
6888 | if (!delay) | |
6889 | mod_delayed_work(system_wq, &ctx->file_put_work, 0); | |
6890 | else if (first_add) | |
6891 | queue_delayed_work(system_wq, &ctx->file_put_work, delay); | |
05f3fb3c | 6892 | } |
65e19f54 | 6893 | |
05589553 XW |
6894 | static struct fixed_file_ref_node *alloc_fixed_file_ref_node( |
6895 | struct io_ring_ctx *ctx) | |
05f3fb3c | 6896 | { |
05589553 | 6897 | struct fixed_file_ref_node *ref_node; |
05f3fb3c | 6898 | |
05589553 XW |
6899 | ref_node = kzalloc(sizeof(*ref_node), GFP_KERNEL); |
6900 | if (!ref_node) | |
6901 | return ERR_PTR(-ENOMEM); | |
05f3fb3c | 6902 | |
05589553 XW |
6903 | if (percpu_ref_init(&ref_node->refs, io_file_data_ref_zero, |
6904 | 0, GFP_KERNEL)) { | |
6905 | kfree(ref_node); | |
6906 | return ERR_PTR(-ENOMEM); | |
6907 | } | |
6908 | INIT_LIST_HEAD(&ref_node->node); | |
6909 | INIT_LIST_HEAD(&ref_node->file_list); | |
05589553 XW |
6910 | ref_node->file_data = ctx->file_data; |
6911 | return ref_node; | |
05589553 XW |
6912 | } |
6913 | ||
6914 | static void destroy_fixed_file_ref_node(struct fixed_file_ref_node *ref_node) | |
6915 | { | |
6916 | percpu_ref_exit(&ref_node->refs); | |
6917 | kfree(ref_node); | |
65e19f54 JA |
6918 | } |
6919 | ||
6b06314c JA |
6920 | static int io_sqe_files_register(struct io_ring_ctx *ctx, void __user *arg, |
6921 | unsigned nr_args) | |
6922 | { | |
6923 | __s32 __user *fds = (__s32 __user *) arg; | |
65e19f54 | 6924 | unsigned nr_tables; |
05f3fb3c | 6925 | struct file *file; |
6b06314c JA |
6926 | int fd, ret = 0; |
6927 | unsigned i; | |
05589553 | 6928 | struct fixed_file_ref_node *ref_node; |
6b06314c | 6929 | |
05f3fb3c | 6930 | if (ctx->file_data) |
6b06314c JA |
6931 | return -EBUSY; |
6932 | if (!nr_args) | |
6933 | return -EINVAL; | |
6934 | if (nr_args > IORING_MAX_FIXED_FILES) | |
6935 | return -EMFILE; | |
6936 | ||
05f3fb3c JA |
6937 | ctx->file_data = kzalloc(sizeof(*ctx->file_data), GFP_KERNEL); |
6938 | if (!ctx->file_data) | |
6939 | return -ENOMEM; | |
6940 | ctx->file_data->ctx = ctx; | |
6941 | init_completion(&ctx->file_data->done); | |
05589553 | 6942 | INIT_LIST_HEAD(&ctx->file_data->ref_list); |
f7fe9346 | 6943 | spin_lock_init(&ctx->file_data->lock); |
05f3fb3c | 6944 | |
65e19f54 | 6945 | nr_tables = DIV_ROUND_UP(nr_args, IORING_MAX_FILES_TABLE); |
05f3fb3c JA |
6946 | ctx->file_data->table = kcalloc(nr_tables, |
6947 | sizeof(struct fixed_file_table), | |
65e19f54 | 6948 | GFP_KERNEL); |
05f3fb3c JA |
6949 | if (!ctx->file_data->table) { |
6950 | kfree(ctx->file_data); | |
6951 | ctx->file_data = NULL; | |
6b06314c | 6952 | return -ENOMEM; |
05f3fb3c JA |
6953 | } |
6954 | ||
05589553 | 6955 | if (percpu_ref_init(&ctx->file_data->refs, io_file_ref_kill, |
05f3fb3c JA |
6956 | PERCPU_REF_ALLOW_REINIT, GFP_KERNEL)) { |
6957 | kfree(ctx->file_data->table); | |
6958 | kfree(ctx->file_data); | |
6959 | ctx->file_data = NULL; | |
6b06314c | 6960 | return -ENOMEM; |
05f3fb3c | 6961 | } |
6b06314c | 6962 | |
65e19f54 | 6963 | if (io_sqe_alloc_file_tables(ctx, nr_tables, nr_args)) { |
05f3fb3c JA |
6964 | percpu_ref_exit(&ctx->file_data->refs); |
6965 | kfree(ctx->file_data->table); | |
6966 | kfree(ctx->file_data); | |
6967 | ctx->file_data = NULL; | |
65e19f54 JA |
6968 | return -ENOMEM; |
6969 | } | |
6970 | ||
08a45173 | 6971 | for (i = 0; i < nr_args; i++, ctx->nr_user_files++) { |
65e19f54 JA |
6972 | struct fixed_file_table *table; |
6973 | unsigned index; | |
6974 | ||
6b06314c JA |
6975 | ret = -EFAULT; |
6976 | if (copy_from_user(&fd, &fds[i], sizeof(fd))) | |
6977 | break; | |
08a45173 JA |
6978 | /* allow sparse sets */ |
6979 | if (fd == -1) { | |
6980 | ret = 0; | |
6981 | continue; | |
6982 | } | |
6b06314c | 6983 | |
05f3fb3c | 6984 | table = &ctx->file_data->table[i >> IORING_FILE_TABLE_SHIFT]; |
65e19f54 | 6985 | index = i & IORING_FILE_TABLE_MASK; |
05f3fb3c | 6986 | file = fget(fd); |
6b06314c JA |
6987 | |
6988 | ret = -EBADF; | |
05f3fb3c | 6989 | if (!file) |
6b06314c | 6990 | break; |
05f3fb3c | 6991 | |
6b06314c JA |
6992 | /* |
6993 | * Don't allow io_uring instances to be registered. If UNIX | |
6994 | * isn't enabled, then this causes a reference cycle and this | |
6995 | * instance can never get freed. If UNIX is enabled we'll | |
6996 | * handle it just fine, but there's still no point in allowing | |
6997 | * a ring fd as it doesn't support regular read/write anyway. | |
6998 | */ | |
05f3fb3c JA |
6999 | if (file->f_op == &io_uring_fops) { |
7000 | fput(file); | |
6b06314c JA |
7001 | break; |
7002 | } | |
6b06314c | 7003 | ret = 0; |
05f3fb3c | 7004 | table->files[index] = file; |
6b06314c JA |
7005 | } |
7006 | ||
7007 | if (ret) { | |
65e19f54 | 7008 | for (i = 0; i < ctx->nr_user_files; i++) { |
65e19f54 JA |
7009 | file = io_file_from_index(ctx, i); |
7010 | if (file) | |
7011 | fput(file); | |
7012 | } | |
7013 | for (i = 0; i < nr_tables; i++) | |
05f3fb3c | 7014 | kfree(ctx->file_data->table[i].files); |
6b06314c | 7015 | |
667e57da | 7016 | percpu_ref_exit(&ctx->file_data->refs); |
05f3fb3c JA |
7017 | kfree(ctx->file_data->table); |
7018 | kfree(ctx->file_data); | |
7019 | ctx->file_data = NULL; | |
6b06314c JA |
7020 | ctx->nr_user_files = 0; |
7021 | return ret; | |
7022 | } | |
7023 | ||
7024 | ret = io_sqe_files_scm(ctx); | |
05589553 | 7025 | if (ret) { |
6b06314c | 7026 | io_sqe_files_unregister(ctx); |
05589553 XW |
7027 | return ret; |
7028 | } | |
6b06314c | 7029 | |
05589553 XW |
7030 | ref_node = alloc_fixed_file_ref_node(ctx); |
7031 | if (IS_ERR(ref_node)) { | |
7032 | io_sqe_files_unregister(ctx); | |
7033 | return PTR_ERR(ref_node); | |
7034 | } | |
7035 | ||
7036 | ctx->file_data->cur_refs = &ref_node->refs; | |
6a4d07cd | 7037 | spin_lock(&ctx->file_data->lock); |
05589553 | 7038 | list_add(&ref_node->node, &ctx->file_data->ref_list); |
6a4d07cd | 7039 | spin_unlock(&ctx->file_data->lock); |
05589553 | 7040 | percpu_ref_get(&ctx->file_data->refs); |
6b06314c JA |
7041 | return ret; |
7042 | } | |
7043 | ||
c3a31e60 JA |
7044 | static int io_sqe_file_register(struct io_ring_ctx *ctx, struct file *file, |
7045 | int index) | |
7046 | { | |
7047 | #if defined(CONFIG_UNIX) | |
7048 | struct sock *sock = ctx->ring_sock->sk; | |
7049 | struct sk_buff_head *head = &sock->sk_receive_queue; | |
7050 | struct sk_buff *skb; | |
7051 | ||
7052 | /* | |
7053 | * See if we can merge this file into an existing skb SCM_RIGHTS | |
7054 | * file set. If there's no room, fall back to allocating a new skb | |
7055 | * and filling it in. | |
7056 | */ | |
7057 | spin_lock_irq(&head->lock); | |
7058 | skb = skb_peek(head); | |
7059 | if (skb) { | |
7060 | struct scm_fp_list *fpl = UNIXCB(skb).fp; | |
7061 | ||
7062 | if (fpl->count < SCM_MAX_FD) { | |
7063 | __skb_unlink(skb, head); | |
7064 | spin_unlock_irq(&head->lock); | |
7065 | fpl->fp[fpl->count] = get_file(file); | |
7066 | unix_inflight(fpl->user, fpl->fp[fpl->count]); | |
7067 | fpl->count++; | |
7068 | spin_lock_irq(&head->lock); | |
7069 | __skb_queue_head(head, skb); | |
7070 | } else { | |
7071 | skb = NULL; | |
7072 | } | |
7073 | } | |
7074 | spin_unlock_irq(&head->lock); | |
7075 | ||
7076 | if (skb) { | |
7077 | fput(file); | |
7078 | return 0; | |
7079 | } | |
7080 | ||
7081 | return __io_sqe_files_scm(ctx, 1, index); | |
7082 | #else | |
7083 | return 0; | |
7084 | #endif | |
7085 | } | |
7086 | ||
a5318d3c | 7087 | static int io_queue_file_removal(struct fixed_file_data *data, |
05589553 | 7088 | struct file *file) |
05f3fb3c | 7089 | { |
a5318d3c | 7090 | struct io_file_put *pfile; |
05589553 XW |
7091 | struct percpu_ref *refs = data->cur_refs; |
7092 | struct fixed_file_ref_node *ref_node; | |
05f3fb3c | 7093 | |
05f3fb3c | 7094 | pfile = kzalloc(sizeof(*pfile), GFP_KERNEL); |
a5318d3c HD |
7095 | if (!pfile) |
7096 | return -ENOMEM; | |
05f3fb3c | 7097 | |
05589553 | 7098 | ref_node = container_of(refs, struct fixed_file_ref_node, refs); |
05f3fb3c | 7099 | pfile->file = file; |
05589553 XW |
7100 | list_add(&pfile->list, &ref_node->file_list); |
7101 | ||
a5318d3c | 7102 | return 0; |
05f3fb3c JA |
7103 | } |
7104 | ||
7105 | static int __io_sqe_files_update(struct io_ring_ctx *ctx, | |
7106 | struct io_uring_files_update *up, | |
7107 | unsigned nr_args) | |
7108 | { | |
7109 | struct fixed_file_data *data = ctx->file_data; | |
05589553 | 7110 | struct fixed_file_ref_node *ref_node; |
05f3fb3c | 7111 | struct file *file; |
c3a31e60 JA |
7112 | __s32 __user *fds; |
7113 | int fd, i, err; | |
7114 | __u32 done; | |
05589553 | 7115 | bool needs_switch = false; |
c3a31e60 | 7116 | |
05f3fb3c | 7117 | if (check_add_overflow(up->offset, nr_args, &done)) |
c3a31e60 JA |
7118 | return -EOVERFLOW; |
7119 | if (done > ctx->nr_user_files) | |
7120 | return -EINVAL; | |
7121 | ||
05589553 XW |
7122 | ref_node = alloc_fixed_file_ref_node(ctx); |
7123 | if (IS_ERR(ref_node)) | |
7124 | return PTR_ERR(ref_node); | |
7125 | ||
c3a31e60 | 7126 | done = 0; |
05f3fb3c | 7127 | fds = u64_to_user_ptr(up->fds); |
c3a31e60 | 7128 | while (nr_args) { |
65e19f54 JA |
7129 | struct fixed_file_table *table; |
7130 | unsigned index; | |
7131 | ||
c3a31e60 JA |
7132 | err = 0; |
7133 | if (copy_from_user(&fd, &fds[done], sizeof(fd))) { | |
7134 | err = -EFAULT; | |
7135 | break; | |
7136 | } | |
05f3fb3c JA |
7137 | i = array_index_nospec(up->offset, ctx->nr_user_files); |
7138 | table = &ctx->file_data->table[i >> IORING_FILE_TABLE_SHIFT]; | |
65e19f54 JA |
7139 | index = i & IORING_FILE_TABLE_MASK; |
7140 | if (table->files[index]) { | |
05f3fb3c | 7141 | file = io_file_from_index(ctx, index); |
a5318d3c HD |
7142 | err = io_queue_file_removal(data, file); |
7143 | if (err) | |
7144 | break; | |
65e19f54 | 7145 | table->files[index] = NULL; |
05589553 | 7146 | needs_switch = true; |
c3a31e60 JA |
7147 | } |
7148 | if (fd != -1) { | |
c3a31e60 JA |
7149 | file = fget(fd); |
7150 | if (!file) { | |
7151 | err = -EBADF; | |
7152 | break; | |
7153 | } | |
7154 | /* | |
7155 | * Don't allow io_uring instances to be registered. If | |
7156 | * UNIX isn't enabled, then this causes a reference | |
7157 | * cycle and this instance can never get freed. If UNIX | |
7158 | * is enabled we'll handle it just fine, but there's | |
7159 | * still no point in allowing a ring fd as it doesn't | |
7160 | * support regular read/write anyway. | |
7161 | */ | |
7162 | if (file->f_op == &io_uring_fops) { | |
7163 | fput(file); | |
7164 | err = -EBADF; | |
7165 | break; | |
7166 | } | |
65e19f54 | 7167 | table->files[index] = file; |
c3a31e60 | 7168 | err = io_sqe_file_register(ctx, file, i); |
f3bd9dae YY |
7169 | if (err) { |
7170 | fput(file); | |
c3a31e60 | 7171 | break; |
f3bd9dae | 7172 | } |
c3a31e60 JA |
7173 | } |
7174 | nr_args--; | |
7175 | done++; | |
05f3fb3c JA |
7176 | up->offset++; |
7177 | } | |
7178 | ||
05589553 XW |
7179 | if (needs_switch) { |
7180 | percpu_ref_kill(data->cur_refs); | |
6a4d07cd | 7181 | spin_lock(&data->lock); |
05589553 XW |
7182 | list_add(&ref_node->node, &data->ref_list); |
7183 | data->cur_refs = &ref_node->refs; | |
6a4d07cd | 7184 | spin_unlock(&data->lock); |
05589553 XW |
7185 | percpu_ref_get(&ctx->file_data->refs); |
7186 | } else | |
7187 | destroy_fixed_file_ref_node(ref_node); | |
c3a31e60 JA |
7188 | |
7189 | return done ? done : err; | |
7190 | } | |
05589553 | 7191 | |
05f3fb3c JA |
7192 | static int io_sqe_files_update(struct io_ring_ctx *ctx, void __user *arg, |
7193 | unsigned nr_args) | |
7194 | { | |
7195 | struct io_uring_files_update up; | |
7196 | ||
7197 | if (!ctx->file_data) | |
7198 | return -ENXIO; | |
7199 | if (!nr_args) | |
7200 | return -EINVAL; | |
7201 | if (copy_from_user(&up, arg, sizeof(up))) | |
7202 | return -EFAULT; | |
7203 | if (up.resv) | |
7204 | return -EINVAL; | |
7205 | ||
7206 | return __io_sqe_files_update(ctx, &up, nr_args); | |
7207 | } | |
c3a31e60 | 7208 | |
e9fd9396 | 7209 | static void io_free_work(struct io_wq_work *work) |
7d723065 JA |
7210 | { |
7211 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); | |
7212 | ||
e9fd9396 | 7213 | /* Consider that io_steal_work() relies on this ref */ |
7d723065 JA |
7214 | io_put_req(req); |
7215 | } | |
7216 | ||
24369c2e PB |
7217 | static int io_init_wq_offload(struct io_ring_ctx *ctx, |
7218 | struct io_uring_params *p) | |
7219 | { | |
7220 | struct io_wq_data data; | |
7221 | struct fd f; | |
7222 | struct io_ring_ctx *ctx_attach; | |
7223 | unsigned int concurrency; | |
7224 | int ret = 0; | |
7225 | ||
7226 | data.user = ctx->user; | |
e9fd9396 | 7227 | data.free_work = io_free_work; |
f5fa38c5 | 7228 | data.do_work = io_wq_submit_work; |
24369c2e PB |
7229 | |
7230 | if (!(p->flags & IORING_SETUP_ATTACH_WQ)) { | |
7231 | /* Do QD, or 4 * CPUS, whatever is smallest */ | |
7232 | concurrency = min(ctx->sq_entries, 4 * num_online_cpus()); | |
7233 | ||
7234 | ctx->io_wq = io_wq_create(concurrency, &data); | |
7235 | if (IS_ERR(ctx->io_wq)) { | |
7236 | ret = PTR_ERR(ctx->io_wq); | |
7237 | ctx->io_wq = NULL; | |
7238 | } | |
7239 | return ret; | |
7240 | } | |
7241 | ||
7242 | f = fdget(p->wq_fd); | |
7243 | if (!f.file) | |
7244 | return -EBADF; | |
7245 | ||
7246 | if (f.file->f_op != &io_uring_fops) { | |
7247 | ret = -EINVAL; | |
7248 | goto out_fput; | |
7249 | } | |
7250 | ||
7251 | ctx_attach = f.file->private_data; | |
7252 | /* @io_wq is protected by holding the fd */ | |
7253 | if (!io_wq_get(ctx_attach->io_wq, &data)) { | |
7254 | ret = -EINVAL; | |
7255 | goto out_fput; | |
7256 | } | |
7257 | ||
7258 | ctx->io_wq = ctx_attach->io_wq; | |
7259 | out_fput: | |
7260 | fdput(f); | |
7261 | return ret; | |
7262 | } | |
7263 | ||
6c271ce2 JA |
7264 | static int io_sq_offload_start(struct io_ring_ctx *ctx, |
7265 | struct io_uring_params *p) | |
2b188cc1 JA |
7266 | { |
7267 | int ret; | |
7268 | ||
6c271ce2 | 7269 | if (ctx->flags & IORING_SETUP_SQPOLL) { |
8eb06d7e PB |
7270 | mmgrab(current->mm); |
7271 | ctx->sqo_mm = current->mm; | |
7272 | ||
3ec482d1 JA |
7273 | ret = -EPERM; |
7274 | if (!capable(CAP_SYS_ADMIN)) | |
7275 | goto err; | |
7276 | ||
917257da JA |
7277 | ctx->sq_thread_idle = msecs_to_jiffies(p->sq_thread_idle); |
7278 | if (!ctx->sq_thread_idle) | |
7279 | ctx->sq_thread_idle = HZ; | |
7280 | ||
6c271ce2 | 7281 | if (p->flags & IORING_SETUP_SQ_AFF) { |
44a9bd18 | 7282 | int cpu = p->sq_thread_cpu; |
6c271ce2 | 7283 | |
917257da | 7284 | ret = -EINVAL; |
44a9bd18 JA |
7285 | if (cpu >= nr_cpu_ids) |
7286 | goto err; | |
7889f44d | 7287 | if (!cpu_online(cpu)) |
917257da JA |
7288 | goto err; |
7289 | ||
6c271ce2 JA |
7290 | ctx->sqo_thread = kthread_create_on_cpu(io_sq_thread, |
7291 | ctx, cpu, | |
7292 | "io_uring-sq"); | |
7293 | } else { | |
7294 | ctx->sqo_thread = kthread_create(io_sq_thread, ctx, | |
7295 | "io_uring-sq"); | |
7296 | } | |
7297 | if (IS_ERR(ctx->sqo_thread)) { | |
7298 | ret = PTR_ERR(ctx->sqo_thread); | |
7299 | ctx->sqo_thread = NULL; | |
7300 | goto err; | |
7301 | } | |
7302 | wake_up_process(ctx->sqo_thread); | |
7303 | } else if (p->flags & IORING_SETUP_SQ_AFF) { | |
7304 | /* Can't have SQ_AFF without SQPOLL */ | |
7305 | ret = -EINVAL; | |
7306 | goto err; | |
7307 | } | |
7308 | ||
24369c2e PB |
7309 | ret = io_init_wq_offload(ctx, p); |
7310 | if (ret) | |
2b188cc1 | 7311 | goto err; |
2b188cc1 JA |
7312 | |
7313 | return 0; | |
7314 | err: | |
54a91f3b | 7315 | io_finish_async(ctx); |
8eb06d7e PB |
7316 | if (ctx->sqo_mm) { |
7317 | mmdrop(ctx->sqo_mm); | |
7318 | ctx->sqo_mm = NULL; | |
7319 | } | |
2b188cc1 JA |
7320 | return ret; |
7321 | } | |
7322 | ||
a087e2b5 BM |
7323 | static inline void __io_unaccount_mem(struct user_struct *user, |
7324 | unsigned long nr_pages) | |
2b188cc1 JA |
7325 | { |
7326 | atomic_long_sub(nr_pages, &user->locked_vm); | |
7327 | } | |
7328 | ||
a087e2b5 BM |
7329 | static inline int __io_account_mem(struct user_struct *user, |
7330 | unsigned long nr_pages) | |
2b188cc1 JA |
7331 | { |
7332 | unsigned long page_limit, cur_pages, new_pages; | |
7333 | ||
7334 | /* Don't allow more pages than we can safely lock */ | |
7335 | page_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT; | |
7336 | ||
7337 | do { | |
7338 | cur_pages = atomic_long_read(&user->locked_vm); | |
7339 | new_pages = cur_pages + nr_pages; | |
7340 | if (new_pages > page_limit) | |
7341 | return -ENOMEM; | |
7342 | } while (atomic_long_cmpxchg(&user->locked_vm, cur_pages, | |
7343 | new_pages) != cur_pages); | |
7344 | ||
7345 | return 0; | |
7346 | } | |
7347 | ||
2e0464d4 BM |
7348 | static void io_unaccount_mem(struct io_ring_ctx *ctx, unsigned long nr_pages, |
7349 | enum io_mem_account acct) | |
a087e2b5 | 7350 | { |
aad5d8da | 7351 | if (ctx->limit_mem) |
a087e2b5 | 7352 | __io_unaccount_mem(ctx->user, nr_pages); |
30975825 | 7353 | |
2e0464d4 BM |
7354 | if (ctx->sqo_mm) { |
7355 | if (acct == ACCT_LOCKED) | |
7356 | ctx->sqo_mm->locked_vm -= nr_pages; | |
7357 | else if (acct == ACCT_PINNED) | |
7358 | atomic64_sub(nr_pages, &ctx->sqo_mm->pinned_vm); | |
7359 | } | |
a087e2b5 BM |
7360 | } |
7361 | ||
2e0464d4 BM |
7362 | static int io_account_mem(struct io_ring_ctx *ctx, unsigned long nr_pages, |
7363 | enum io_mem_account acct) | |
a087e2b5 | 7364 | { |
30975825 BM |
7365 | int ret; |
7366 | ||
7367 | if (ctx->limit_mem) { | |
7368 | ret = __io_account_mem(ctx->user, nr_pages); | |
7369 | if (ret) | |
7370 | return ret; | |
7371 | } | |
7372 | ||
2e0464d4 BM |
7373 | if (ctx->sqo_mm) { |
7374 | if (acct == ACCT_LOCKED) | |
7375 | ctx->sqo_mm->locked_vm += nr_pages; | |
7376 | else if (acct == ACCT_PINNED) | |
7377 | atomic64_add(nr_pages, &ctx->sqo_mm->pinned_vm); | |
7378 | } | |
a087e2b5 BM |
7379 | |
7380 | return 0; | |
7381 | } | |
7382 | ||
2b188cc1 JA |
7383 | static void io_mem_free(void *ptr) |
7384 | { | |
52e04ef4 MR |
7385 | struct page *page; |
7386 | ||
7387 | if (!ptr) | |
7388 | return; | |
2b188cc1 | 7389 | |
52e04ef4 | 7390 | page = virt_to_head_page(ptr); |
2b188cc1 JA |
7391 | if (put_page_testzero(page)) |
7392 | free_compound_page(page); | |
7393 | } | |
7394 | ||
7395 | static void *io_mem_alloc(size_t size) | |
7396 | { | |
7397 | gfp_t gfp_flags = GFP_KERNEL | __GFP_ZERO | __GFP_NOWARN | __GFP_COMP | | |
7398 | __GFP_NORETRY; | |
7399 | ||
7400 | return (void *) __get_free_pages(gfp_flags, get_order(size)); | |
7401 | } | |
7402 | ||
75b28aff HV |
7403 | static unsigned long rings_size(unsigned sq_entries, unsigned cq_entries, |
7404 | size_t *sq_offset) | |
7405 | { | |
7406 | struct io_rings *rings; | |
7407 | size_t off, sq_array_size; | |
7408 | ||
7409 | off = struct_size(rings, cqes, cq_entries); | |
7410 | if (off == SIZE_MAX) | |
7411 | return SIZE_MAX; | |
7412 | ||
7413 | #ifdef CONFIG_SMP | |
7414 | off = ALIGN(off, SMP_CACHE_BYTES); | |
7415 | if (off == 0) | |
7416 | return SIZE_MAX; | |
7417 | #endif | |
7418 | ||
b36200f5 DV |
7419 | if (sq_offset) |
7420 | *sq_offset = off; | |
7421 | ||
75b28aff HV |
7422 | sq_array_size = array_size(sizeof(u32), sq_entries); |
7423 | if (sq_array_size == SIZE_MAX) | |
7424 | return SIZE_MAX; | |
7425 | ||
7426 | if (check_add_overflow(off, sq_array_size, &off)) | |
7427 | return SIZE_MAX; | |
7428 | ||
75b28aff HV |
7429 | return off; |
7430 | } | |
7431 | ||
2b188cc1 JA |
7432 | static unsigned long ring_pages(unsigned sq_entries, unsigned cq_entries) |
7433 | { | |
75b28aff | 7434 | size_t pages; |
2b188cc1 | 7435 | |
75b28aff HV |
7436 | pages = (size_t)1 << get_order( |
7437 | rings_size(sq_entries, cq_entries, NULL)); | |
7438 | pages += (size_t)1 << get_order( | |
7439 | array_size(sizeof(struct io_uring_sqe), sq_entries)); | |
2b188cc1 | 7440 | |
75b28aff | 7441 | return pages; |
2b188cc1 JA |
7442 | } |
7443 | ||
edafccee JA |
7444 | static int io_sqe_buffer_unregister(struct io_ring_ctx *ctx) |
7445 | { | |
7446 | int i, j; | |
7447 | ||
7448 | if (!ctx->user_bufs) | |
7449 | return -ENXIO; | |
7450 | ||
7451 | for (i = 0; i < ctx->nr_user_bufs; i++) { | |
7452 | struct io_mapped_ubuf *imu = &ctx->user_bufs[i]; | |
7453 | ||
7454 | for (j = 0; j < imu->nr_bvecs; j++) | |
f1f6a7dd | 7455 | unpin_user_page(imu->bvec[j].bv_page); |
edafccee | 7456 | |
2e0464d4 | 7457 | io_unaccount_mem(ctx, imu->nr_bvecs, ACCT_PINNED); |
d4ef6475 | 7458 | kvfree(imu->bvec); |
edafccee JA |
7459 | imu->nr_bvecs = 0; |
7460 | } | |
7461 | ||
7462 | kfree(ctx->user_bufs); | |
7463 | ctx->user_bufs = NULL; | |
7464 | ctx->nr_user_bufs = 0; | |
7465 | return 0; | |
7466 | } | |
7467 | ||
7468 | static int io_copy_iov(struct io_ring_ctx *ctx, struct iovec *dst, | |
7469 | void __user *arg, unsigned index) | |
7470 | { | |
7471 | struct iovec __user *src; | |
7472 | ||
7473 | #ifdef CONFIG_COMPAT | |
7474 | if (ctx->compat) { | |
7475 | struct compat_iovec __user *ciovs; | |
7476 | struct compat_iovec ciov; | |
7477 | ||
7478 | ciovs = (struct compat_iovec __user *) arg; | |
7479 | if (copy_from_user(&ciov, &ciovs[index], sizeof(ciov))) | |
7480 | return -EFAULT; | |
7481 | ||
d55e5f5b | 7482 | dst->iov_base = u64_to_user_ptr((u64)ciov.iov_base); |
edafccee JA |
7483 | dst->iov_len = ciov.iov_len; |
7484 | return 0; | |
7485 | } | |
7486 | #endif | |
7487 | src = (struct iovec __user *) arg; | |
7488 | if (copy_from_user(dst, &src[index], sizeof(*dst))) | |
7489 | return -EFAULT; | |
7490 | return 0; | |
7491 | } | |
7492 | ||
7493 | static int io_sqe_buffer_register(struct io_ring_ctx *ctx, void __user *arg, | |
7494 | unsigned nr_args) | |
7495 | { | |
7496 | struct vm_area_struct **vmas = NULL; | |
7497 | struct page **pages = NULL; | |
7498 | int i, j, got_pages = 0; | |
7499 | int ret = -EINVAL; | |
7500 | ||
7501 | if (ctx->user_bufs) | |
7502 | return -EBUSY; | |
7503 | if (!nr_args || nr_args > UIO_MAXIOV) | |
7504 | return -EINVAL; | |
7505 | ||
7506 | ctx->user_bufs = kcalloc(nr_args, sizeof(struct io_mapped_ubuf), | |
7507 | GFP_KERNEL); | |
7508 | if (!ctx->user_bufs) | |
7509 | return -ENOMEM; | |
7510 | ||
7511 | for (i = 0; i < nr_args; i++) { | |
7512 | struct io_mapped_ubuf *imu = &ctx->user_bufs[i]; | |
7513 | unsigned long off, start, end, ubuf; | |
7514 | int pret, nr_pages; | |
7515 | struct iovec iov; | |
7516 | size_t size; | |
7517 | ||
7518 | ret = io_copy_iov(ctx, &iov, arg, i); | |
7519 | if (ret) | |
a278682d | 7520 | goto err; |
edafccee JA |
7521 | |
7522 | /* | |
7523 | * Don't impose further limits on the size and buffer | |
7524 | * constraints here, we'll -EINVAL later when IO is | |
7525 | * submitted if they are wrong. | |
7526 | */ | |
7527 | ret = -EFAULT; | |
7528 | if (!iov.iov_base || !iov.iov_len) | |
7529 | goto err; | |
7530 | ||
7531 | /* arbitrary limit, but we need something */ | |
7532 | if (iov.iov_len > SZ_1G) | |
7533 | goto err; | |
7534 | ||
7535 | ubuf = (unsigned long) iov.iov_base; | |
7536 | end = (ubuf + iov.iov_len + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
7537 | start = ubuf >> PAGE_SHIFT; | |
7538 | nr_pages = end - start; | |
7539 | ||
2e0464d4 | 7540 | ret = io_account_mem(ctx, nr_pages, ACCT_PINNED); |
a087e2b5 BM |
7541 | if (ret) |
7542 | goto err; | |
edafccee JA |
7543 | |
7544 | ret = 0; | |
7545 | if (!pages || nr_pages > got_pages) { | |
a8c73c1a DE |
7546 | kvfree(vmas); |
7547 | kvfree(pages); | |
d4ef6475 | 7548 | pages = kvmalloc_array(nr_pages, sizeof(struct page *), |
edafccee | 7549 | GFP_KERNEL); |
d4ef6475 | 7550 | vmas = kvmalloc_array(nr_pages, |
edafccee JA |
7551 | sizeof(struct vm_area_struct *), |
7552 | GFP_KERNEL); | |
7553 | if (!pages || !vmas) { | |
7554 | ret = -ENOMEM; | |
2e0464d4 | 7555 | io_unaccount_mem(ctx, nr_pages, ACCT_PINNED); |
edafccee JA |
7556 | goto err; |
7557 | } | |
7558 | got_pages = nr_pages; | |
7559 | } | |
7560 | ||
d4ef6475 | 7561 | imu->bvec = kvmalloc_array(nr_pages, sizeof(struct bio_vec), |
edafccee JA |
7562 | GFP_KERNEL); |
7563 | ret = -ENOMEM; | |
7564 | if (!imu->bvec) { | |
2e0464d4 | 7565 | io_unaccount_mem(ctx, nr_pages, ACCT_PINNED); |
edafccee JA |
7566 | goto err; |
7567 | } | |
7568 | ||
7569 | ret = 0; | |
d8ed45c5 | 7570 | mmap_read_lock(current->mm); |
2113b05d | 7571 | pret = pin_user_pages(ubuf, nr_pages, |
932f4a63 IW |
7572 | FOLL_WRITE | FOLL_LONGTERM, |
7573 | pages, vmas); | |
edafccee JA |
7574 | if (pret == nr_pages) { |
7575 | /* don't support file backed memory */ | |
7576 | for (j = 0; j < nr_pages; j++) { | |
7577 | struct vm_area_struct *vma = vmas[j]; | |
7578 | ||
7579 | if (vma->vm_file && | |
7580 | !is_file_hugepages(vma->vm_file)) { | |
7581 | ret = -EOPNOTSUPP; | |
7582 | break; | |
7583 | } | |
7584 | } | |
7585 | } else { | |
7586 | ret = pret < 0 ? pret : -EFAULT; | |
7587 | } | |
d8ed45c5 | 7588 | mmap_read_unlock(current->mm); |
edafccee JA |
7589 | if (ret) { |
7590 | /* | |
7591 | * if we did partial map, or found file backed vmas, | |
7592 | * release any pages we did get | |
7593 | */ | |
27c4d3a3 | 7594 | if (pret > 0) |
f1f6a7dd | 7595 | unpin_user_pages(pages, pret); |
2e0464d4 | 7596 | io_unaccount_mem(ctx, nr_pages, ACCT_PINNED); |
d4ef6475 | 7597 | kvfree(imu->bvec); |
edafccee JA |
7598 | goto err; |
7599 | } | |
7600 | ||
7601 | off = ubuf & ~PAGE_MASK; | |
7602 | size = iov.iov_len; | |
7603 | for (j = 0; j < nr_pages; j++) { | |
7604 | size_t vec_len; | |
7605 | ||
7606 | vec_len = min_t(size_t, size, PAGE_SIZE - off); | |
7607 | imu->bvec[j].bv_page = pages[j]; | |
7608 | imu->bvec[j].bv_len = vec_len; | |
7609 | imu->bvec[j].bv_offset = off; | |
7610 | off = 0; | |
7611 | size -= vec_len; | |
7612 | } | |
7613 | /* store original address for later verification */ | |
7614 | imu->ubuf = ubuf; | |
7615 | imu->len = iov.iov_len; | |
7616 | imu->nr_bvecs = nr_pages; | |
7617 | ||
7618 | ctx->nr_user_bufs++; | |
7619 | } | |
d4ef6475 MR |
7620 | kvfree(pages); |
7621 | kvfree(vmas); | |
edafccee JA |
7622 | return 0; |
7623 | err: | |
d4ef6475 MR |
7624 | kvfree(pages); |
7625 | kvfree(vmas); | |
edafccee JA |
7626 | io_sqe_buffer_unregister(ctx); |
7627 | return ret; | |
7628 | } | |
7629 | ||
9b402849 JA |
7630 | static int io_eventfd_register(struct io_ring_ctx *ctx, void __user *arg) |
7631 | { | |
7632 | __s32 __user *fds = arg; | |
7633 | int fd; | |
7634 | ||
7635 | if (ctx->cq_ev_fd) | |
7636 | return -EBUSY; | |
7637 | ||
7638 | if (copy_from_user(&fd, fds, sizeof(*fds))) | |
7639 | return -EFAULT; | |
7640 | ||
7641 | ctx->cq_ev_fd = eventfd_ctx_fdget(fd); | |
7642 | if (IS_ERR(ctx->cq_ev_fd)) { | |
7643 | int ret = PTR_ERR(ctx->cq_ev_fd); | |
7644 | ctx->cq_ev_fd = NULL; | |
7645 | return ret; | |
7646 | } | |
7647 | ||
7648 | return 0; | |
7649 | } | |
7650 | ||
7651 | static int io_eventfd_unregister(struct io_ring_ctx *ctx) | |
7652 | { | |
7653 | if (ctx->cq_ev_fd) { | |
7654 | eventfd_ctx_put(ctx->cq_ev_fd); | |
7655 | ctx->cq_ev_fd = NULL; | |
7656 | return 0; | |
7657 | } | |
7658 | ||
7659 | return -ENXIO; | |
7660 | } | |
7661 | ||
5a2e745d JA |
7662 | static int __io_destroy_buffers(int id, void *p, void *data) |
7663 | { | |
7664 | struct io_ring_ctx *ctx = data; | |
7665 | struct io_buffer *buf = p; | |
7666 | ||
067524e9 | 7667 | __io_remove_buffers(ctx, buf, id, -1U); |
5a2e745d JA |
7668 | return 0; |
7669 | } | |
7670 | ||
7671 | static void io_destroy_buffers(struct io_ring_ctx *ctx) | |
7672 | { | |
7673 | idr_for_each(&ctx->io_buffer_idr, __io_destroy_buffers, ctx); | |
7674 | idr_destroy(&ctx->io_buffer_idr); | |
7675 | } | |
7676 | ||
2b188cc1 JA |
7677 | static void io_ring_ctx_free(struct io_ring_ctx *ctx) |
7678 | { | |
6b06314c | 7679 | io_finish_async(ctx); |
30975825 | 7680 | if (ctx->sqo_mm) { |
2b188cc1 | 7681 | mmdrop(ctx->sqo_mm); |
30975825 BM |
7682 | ctx->sqo_mm = NULL; |
7683 | } | |
def596e9 | 7684 | |
edafccee | 7685 | io_sqe_buffer_unregister(ctx); |
6b06314c | 7686 | io_sqe_files_unregister(ctx); |
9b402849 | 7687 | io_eventfd_unregister(ctx); |
5a2e745d | 7688 | io_destroy_buffers(ctx); |
41726c9a | 7689 | idr_destroy(&ctx->personality_idr); |
def596e9 | 7690 | |
2b188cc1 | 7691 | #if defined(CONFIG_UNIX) |
355e8d26 EB |
7692 | if (ctx->ring_sock) { |
7693 | ctx->ring_sock->file = NULL; /* so that iput() is called */ | |
2b188cc1 | 7694 | sock_release(ctx->ring_sock); |
355e8d26 | 7695 | } |
2b188cc1 JA |
7696 | #endif |
7697 | ||
75b28aff | 7698 | io_mem_free(ctx->rings); |
2b188cc1 | 7699 | io_mem_free(ctx->sq_sqes); |
2b188cc1 JA |
7700 | |
7701 | percpu_ref_exit(&ctx->refs); | |
2b188cc1 | 7702 | free_uid(ctx->user); |
181e448d | 7703 | put_cred(ctx->creds); |
78076bb6 | 7704 | kfree(ctx->cancel_hash); |
0ddf92e8 | 7705 | kmem_cache_free(req_cachep, ctx->fallback_req); |
2b188cc1 JA |
7706 | kfree(ctx); |
7707 | } | |
7708 | ||
7709 | static __poll_t io_uring_poll(struct file *file, poll_table *wait) | |
7710 | { | |
7711 | struct io_ring_ctx *ctx = file->private_data; | |
7712 | __poll_t mask = 0; | |
7713 | ||
7714 | poll_wait(file, &ctx->cq_wait, wait); | |
4f7067c3 SB |
7715 | /* |
7716 | * synchronizes with barrier from wq_has_sleeper call in | |
7717 | * io_commit_cqring | |
7718 | */ | |
2b188cc1 | 7719 | smp_rmb(); |
75b28aff HV |
7720 | if (READ_ONCE(ctx->rings->sq.tail) - ctx->cached_sq_head != |
7721 | ctx->rings->sq_ring_entries) | |
2b188cc1 | 7722 | mask |= EPOLLOUT | EPOLLWRNORM; |
63e5d81f | 7723 | if (io_cqring_events(ctx, false)) |
2b188cc1 JA |
7724 | mask |= EPOLLIN | EPOLLRDNORM; |
7725 | ||
7726 | return mask; | |
7727 | } | |
7728 | ||
7729 | static int io_uring_fasync(int fd, struct file *file, int on) | |
7730 | { | |
7731 | struct io_ring_ctx *ctx = file->private_data; | |
7732 | ||
7733 | return fasync_helper(fd, file, on, &ctx->cq_fasync); | |
7734 | } | |
7735 | ||
071698e1 JA |
7736 | static int io_remove_personalities(int id, void *p, void *data) |
7737 | { | |
7738 | struct io_ring_ctx *ctx = data; | |
7739 | const struct cred *cred; | |
7740 | ||
7741 | cred = idr_remove(&ctx->personality_idr, id); | |
7742 | if (cred) | |
7743 | put_cred(cred); | |
7744 | return 0; | |
7745 | } | |
7746 | ||
85faa7b8 JA |
7747 | static void io_ring_exit_work(struct work_struct *work) |
7748 | { | |
b2edc0a7 PB |
7749 | struct io_ring_ctx *ctx = container_of(work, struct io_ring_ctx, |
7750 | exit_work); | |
85faa7b8 | 7751 | |
56952e91 JA |
7752 | /* |
7753 | * If we're doing polled IO and end up having requests being | |
7754 | * submitted async (out-of-line), then completions can come in while | |
7755 | * we're waiting for refs to drop. We need to reap these manually, | |
7756 | * as nobody else will be looking for them. | |
7757 | */ | |
b2edc0a7 | 7758 | do { |
56952e91 JA |
7759 | if (ctx->rings) |
7760 | io_cqring_overflow_flush(ctx, true); | |
b2edc0a7 PB |
7761 | io_iopoll_try_reap_events(ctx); |
7762 | } while (!wait_for_completion_timeout(&ctx->ref_comp, HZ/20)); | |
85faa7b8 JA |
7763 | io_ring_ctx_free(ctx); |
7764 | } | |
7765 | ||
2b188cc1 JA |
7766 | static void io_ring_ctx_wait_and_kill(struct io_ring_ctx *ctx) |
7767 | { | |
7768 | mutex_lock(&ctx->uring_lock); | |
7769 | percpu_ref_kill(&ctx->refs); | |
7770 | mutex_unlock(&ctx->uring_lock); | |
7771 | ||
5262f567 | 7772 | io_kill_timeouts(ctx); |
221c5eb2 | 7773 | io_poll_remove_all(ctx); |
561fb04a JA |
7774 | |
7775 | if (ctx->io_wq) | |
7776 | io_wq_cancel_all(ctx->io_wq); | |
7777 | ||
15dff286 JA |
7778 | /* if we failed setting up the ctx, we might not have any rings */ |
7779 | if (ctx->rings) | |
7780 | io_cqring_overflow_flush(ctx, true); | |
b2edc0a7 | 7781 | io_iopoll_try_reap_events(ctx); |
071698e1 | 7782 | idr_for_each(&ctx->personality_idr, io_remove_personalities, ctx); |
309fc03a JA |
7783 | |
7784 | /* | |
7785 | * Do this upfront, so we won't have a grace period where the ring | |
7786 | * is closed but resources aren't reaped yet. This can cause | |
7787 | * spurious failure in setting up a new ring. | |
7788 | */ | |
760618f7 JA |
7789 | io_unaccount_mem(ctx, ring_pages(ctx->sq_entries, ctx->cq_entries), |
7790 | ACCT_LOCKED); | |
309fc03a | 7791 | |
85faa7b8 JA |
7792 | INIT_WORK(&ctx->exit_work, io_ring_exit_work); |
7793 | queue_work(system_wq, &ctx->exit_work); | |
2b188cc1 JA |
7794 | } |
7795 | ||
7796 | static int io_uring_release(struct inode *inode, struct file *file) | |
7797 | { | |
7798 | struct io_ring_ctx *ctx = file->private_data; | |
7799 | ||
7800 | file->private_data = NULL; | |
7801 | io_ring_ctx_wait_and_kill(ctx); | |
7802 | return 0; | |
7803 | } | |
7804 | ||
67c4d9e6 PB |
7805 | static bool io_wq_files_match(struct io_wq_work *work, void *data) |
7806 | { | |
7807 | struct files_struct *files = data; | |
7808 | ||
7809 | return work->files == files; | |
7810 | } | |
7811 | ||
fcb323cc JA |
7812 | static void io_uring_cancel_files(struct io_ring_ctx *ctx, |
7813 | struct files_struct *files) | |
7814 | { | |
67c4d9e6 PB |
7815 | if (list_empty_careful(&ctx->inflight_list)) |
7816 | return; | |
7817 | ||
7818 | /* cancel all at once, should be faster than doing it one by one*/ | |
7819 | io_wq_cancel_cb(ctx->io_wq, io_wq_files_match, files, true); | |
7820 | ||
fcb323cc | 7821 | while (!list_empty_careful(&ctx->inflight_list)) { |
d8f1b971 XW |
7822 | struct io_kiocb *cancel_req = NULL, *req; |
7823 | DEFINE_WAIT(wait); | |
fcb323cc JA |
7824 | |
7825 | spin_lock_irq(&ctx->inflight_lock); | |
7826 | list_for_each_entry(req, &ctx->inflight_list, inflight_entry) { | |
768134d4 JA |
7827 | if (req->work.files != files) |
7828 | continue; | |
7829 | /* req is being completed, ignore */ | |
7830 | if (!refcount_inc_not_zero(&req->refs)) | |
7831 | continue; | |
7832 | cancel_req = req; | |
7833 | break; | |
fcb323cc | 7834 | } |
768134d4 | 7835 | if (cancel_req) |
fcb323cc | 7836 | prepare_to_wait(&ctx->inflight_wait, &wait, |
768134d4 | 7837 | TASK_UNINTERRUPTIBLE); |
fcb323cc JA |
7838 | spin_unlock_irq(&ctx->inflight_lock); |
7839 | ||
768134d4 JA |
7840 | /* We need to keep going until we don't find a matching req */ |
7841 | if (!cancel_req) | |
fcb323cc | 7842 | break; |
2f6d9b9d | 7843 | |
2ca10259 JA |
7844 | if (cancel_req->flags & REQ_F_OVERFLOW) { |
7845 | spin_lock_irq(&ctx->completion_lock); | |
40d8ddd4 | 7846 | list_del(&cancel_req->compl.list); |
2ca10259 JA |
7847 | cancel_req->flags &= ~REQ_F_OVERFLOW; |
7848 | if (list_empty(&ctx->cq_overflow_list)) { | |
7849 | clear_bit(0, &ctx->sq_check_overflow); | |
7850 | clear_bit(0, &ctx->cq_check_overflow); | |
6d5f9049 | 7851 | ctx->rings->sq_flags &= ~IORING_SQ_CQ_OVERFLOW; |
2ca10259 JA |
7852 | } |
7853 | spin_unlock_irq(&ctx->completion_lock); | |
7854 | ||
7855 | WRITE_ONCE(ctx->rings->cq_overflow, | |
7856 | atomic_inc_return(&ctx->cached_cq_overflow)); | |
7857 | ||
7858 | /* | |
7859 | * Put inflight ref and overflow ref. If that's | |
7860 | * all we had, then we're done with this request. | |
7861 | */ | |
7862 | if (refcount_sub_and_test(2, &cancel_req->refs)) { | |
4518a3cc | 7863 | io_free_req(cancel_req); |
d8f1b971 | 7864 | finish_wait(&ctx->inflight_wait, &wait); |
2ca10259 JA |
7865 | continue; |
7866 | } | |
7b53d598 PB |
7867 | } else { |
7868 | io_wq_cancel_work(ctx->io_wq, &cancel_req->work); | |
7869 | io_put_req(cancel_req); | |
2ca10259 JA |
7870 | } |
7871 | ||
fcb323cc | 7872 | schedule(); |
d8f1b971 | 7873 | finish_wait(&ctx->inflight_wait, &wait); |
fcb323cc JA |
7874 | } |
7875 | } | |
7876 | ||
801dd57b | 7877 | static bool io_cancel_task_cb(struct io_wq_work *work, void *data) |
44e728b8 | 7878 | { |
801dd57b PB |
7879 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); |
7880 | struct task_struct *task = data; | |
44e728b8 | 7881 | |
801dd57b | 7882 | return req->task == task; |
44e728b8 PB |
7883 | } |
7884 | ||
fcb323cc JA |
7885 | static int io_uring_flush(struct file *file, void *data) |
7886 | { | |
7887 | struct io_ring_ctx *ctx = file->private_data; | |
7888 | ||
7889 | io_uring_cancel_files(ctx, data); | |
6ab23144 JA |
7890 | |
7891 | /* | |
7892 | * If the task is going away, cancel work it may have pending | |
7893 | */ | |
801dd57b PB |
7894 | if (fatal_signal_pending(current) || (current->flags & PF_EXITING)) |
7895 | io_wq_cancel_cb(ctx->io_wq, io_cancel_task_cb, current, true); | |
6ab23144 | 7896 | |
fcb323cc JA |
7897 | return 0; |
7898 | } | |
7899 | ||
6c5c240e RP |
7900 | static void *io_uring_validate_mmap_request(struct file *file, |
7901 | loff_t pgoff, size_t sz) | |
2b188cc1 | 7902 | { |
2b188cc1 | 7903 | struct io_ring_ctx *ctx = file->private_data; |
6c5c240e | 7904 | loff_t offset = pgoff << PAGE_SHIFT; |
2b188cc1 JA |
7905 | struct page *page; |
7906 | void *ptr; | |
7907 | ||
7908 | switch (offset) { | |
7909 | case IORING_OFF_SQ_RING: | |
75b28aff HV |
7910 | case IORING_OFF_CQ_RING: |
7911 | ptr = ctx->rings; | |
2b188cc1 JA |
7912 | break; |
7913 | case IORING_OFF_SQES: | |
7914 | ptr = ctx->sq_sqes; | |
7915 | break; | |
2b188cc1 | 7916 | default: |
6c5c240e | 7917 | return ERR_PTR(-EINVAL); |
2b188cc1 JA |
7918 | } |
7919 | ||
7920 | page = virt_to_head_page(ptr); | |
a50b854e | 7921 | if (sz > page_size(page)) |
6c5c240e RP |
7922 | return ERR_PTR(-EINVAL); |
7923 | ||
7924 | return ptr; | |
7925 | } | |
7926 | ||
7927 | #ifdef CONFIG_MMU | |
7928 | ||
7929 | static int io_uring_mmap(struct file *file, struct vm_area_struct *vma) | |
7930 | { | |
7931 | size_t sz = vma->vm_end - vma->vm_start; | |
7932 | unsigned long pfn; | |
7933 | void *ptr; | |
7934 | ||
7935 | ptr = io_uring_validate_mmap_request(file, vma->vm_pgoff, sz); | |
7936 | if (IS_ERR(ptr)) | |
7937 | return PTR_ERR(ptr); | |
2b188cc1 JA |
7938 | |
7939 | pfn = virt_to_phys(ptr) >> PAGE_SHIFT; | |
7940 | return remap_pfn_range(vma, vma->vm_start, pfn, sz, vma->vm_page_prot); | |
7941 | } | |
7942 | ||
6c5c240e RP |
7943 | #else /* !CONFIG_MMU */ |
7944 | ||
7945 | static int io_uring_mmap(struct file *file, struct vm_area_struct *vma) | |
7946 | { | |
7947 | return vma->vm_flags & (VM_SHARED | VM_MAYSHARE) ? 0 : -EINVAL; | |
7948 | } | |
7949 | ||
7950 | static unsigned int io_uring_nommu_mmap_capabilities(struct file *file) | |
7951 | { | |
7952 | return NOMMU_MAP_DIRECT | NOMMU_MAP_READ | NOMMU_MAP_WRITE; | |
7953 | } | |
7954 | ||
7955 | static unsigned long io_uring_nommu_get_unmapped_area(struct file *file, | |
7956 | unsigned long addr, unsigned long len, | |
7957 | unsigned long pgoff, unsigned long flags) | |
7958 | { | |
7959 | void *ptr; | |
7960 | ||
7961 | ptr = io_uring_validate_mmap_request(file, pgoff, len); | |
7962 | if (IS_ERR(ptr)) | |
7963 | return PTR_ERR(ptr); | |
7964 | ||
7965 | return (unsigned long) ptr; | |
7966 | } | |
7967 | ||
7968 | #endif /* !CONFIG_MMU */ | |
7969 | ||
2b188cc1 JA |
7970 | SYSCALL_DEFINE6(io_uring_enter, unsigned int, fd, u32, to_submit, |
7971 | u32, min_complete, u32, flags, const sigset_t __user *, sig, | |
7972 | size_t, sigsz) | |
7973 | { | |
7974 | struct io_ring_ctx *ctx; | |
7975 | long ret = -EBADF; | |
7976 | int submitted = 0; | |
7977 | struct fd f; | |
7978 | ||
4c6e277c | 7979 | io_run_task_work(); |
b41e9852 | 7980 | |
6c271ce2 | 7981 | if (flags & ~(IORING_ENTER_GETEVENTS | IORING_ENTER_SQ_WAKEUP)) |
2b188cc1 JA |
7982 | return -EINVAL; |
7983 | ||
7984 | f = fdget(fd); | |
7985 | if (!f.file) | |
7986 | return -EBADF; | |
7987 | ||
7988 | ret = -EOPNOTSUPP; | |
7989 | if (f.file->f_op != &io_uring_fops) | |
7990 | goto out_fput; | |
7991 | ||
7992 | ret = -ENXIO; | |
7993 | ctx = f.file->private_data; | |
7994 | if (!percpu_ref_tryget(&ctx->refs)) | |
7995 | goto out_fput; | |
7996 | ||
6c271ce2 JA |
7997 | /* |
7998 | * For SQ polling, the thread will do all submissions and completions. | |
7999 | * Just return the requested submit count, and wake the thread if | |
8000 | * we were asked to. | |
8001 | */ | |
b2a9eada | 8002 | ret = 0; |
6c271ce2 | 8003 | if (ctx->flags & IORING_SETUP_SQPOLL) { |
c1edbf5f JA |
8004 | if (!list_empty_careful(&ctx->cq_overflow_list)) |
8005 | io_cqring_overflow_flush(ctx, false); | |
6c271ce2 JA |
8006 | if (flags & IORING_ENTER_SQ_WAKEUP) |
8007 | wake_up(&ctx->sqo_wait); | |
8008 | submitted = to_submit; | |
b2a9eada | 8009 | } else if (to_submit) { |
2b188cc1 | 8010 | mutex_lock(&ctx->uring_lock); |
0cdaf760 | 8011 | submitted = io_submit_sqes(ctx, to_submit, f.file, fd); |
2b188cc1 | 8012 | mutex_unlock(&ctx->uring_lock); |
7c504e65 PB |
8013 | |
8014 | if (submitted != to_submit) | |
8015 | goto out; | |
2b188cc1 JA |
8016 | } |
8017 | if (flags & IORING_ENTER_GETEVENTS) { | |
8018 | min_complete = min(min_complete, ctx->cq_entries); | |
8019 | ||
32b2244a XW |
8020 | /* |
8021 | * When SETUP_IOPOLL and SETUP_SQPOLL are both enabled, user | |
8022 | * space applications don't need to do io completion events | |
8023 | * polling again, they can rely on io_sq_thread to do polling | |
8024 | * work, which can reduce cpu usage and uring_lock contention. | |
8025 | */ | |
8026 | if (ctx->flags & IORING_SETUP_IOPOLL && | |
8027 | !(ctx->flags & IORING_SETUP_SQPOLL)) { | |
7668b92a | 8028 | ret = io_iopoll_check(ctx, min_complete); |
def596e9 JA |
8029 | } else { |
8030 | ret = io_cqring_wait(ctx, min_complete, sig, sigsz); | |
8031 | } | |
2b188cc1 JA |
8032 | } |
8033 | ||
7c504e65 | 8034 | out: |
6805b32e | 8035 | percpu_ref_put(&ctx->refs); |
2b188cc1 JA |
8036 | out_fput: |
8037 | fdput(f); | |
8038 | return submitted ? submitted : ret; | |
8039 | } | |
8040 | ||
bebdb65e | 8041 | #ifdef CONFIG_PROC_FS |
87ce955b JA |
8042 | static int io_uring_show_cred(int id, void *p, void *data) |
8043 | { | |
8044 | const struct cred *cred = p; | |
8045 | struct seq_file *m = data; | |
8046 | struct user_namespace *uns = seq_user_ns(m); | |
8047 | struct group_info *gi; | |
8048 | kernel_cap_t cap; | |
8049 | unsigned __capi; | |
8050 | int g; | |
8051 | ||
8052 | seq_printf(m, "%5d\n", id); | |
8053 | seq_put_decimal_ull(m, "\tUid:\t", from_kuid_munged(uns, cred->uid)); | |
8054 | seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->euid)); | |
8055 | seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->suid)); | |
8056 | seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->fsuid)); | |
8057 | seq_put_decimal_ull(m, "\n\tGid:\t", from_kgid_munged(uns, cred->gid)); | |
8058 | seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->egid)); | |
8059 | seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->sgid)); | |
8060 | seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->fsgid)); | |
8061 | seq_puts(m, "\n\tGroups:\t"); | |
8062 | gi = cred->group_info; | |
8063 | for (g = 0; g < gi->ngroups; g++) { | |
8064 | seq_put_decimal_ull(m, g ? " " : "", | |
8065 | from_kgid_munged(uns, gi->gid[g])); | |
8066 | } | |
8067 | seq_puts(m, "\n\tCapEff:\t"); | |
8068 | cap = cred->cap_effective; | |
8069 | CAP_FOR_EACH_U32(__capi) | |
8070 | seq_put_hex_ll(m, NULL, cap.cap[CAP_LAST_U32 - __capi], 8); | |
8071 | seq_putc(m, '\n'); | |
8072 | return 0; | |
8073 | } | |
8074 | ||
8075 | static void __io_uring_show_fdinfo(struct io_ring_ctx *ctx, struct seq_file *m) | |
8076 | { | |
8077 | int i; | |
8078 | ||
8079 | mutex_lock(&ctx->uring_lock); | |
8080 | seq_printf(m, "UserFiles:\t%u\n", ctx->nr_user_files); | |
8081 | for (i = 0; i < ctx->nr_user_files; i++) { | |
8082 | struct fixed_file_table *table; | |
8083 | struct file *f; | |
8084 | ||
8085 | table = &ctx->file_data->table[i >> IORING_FILE_TABLE_SHIFT]; | |
8086 | f = table->files[i & IORING_FILE_TABLE_MASK]; | |
8087 | if (f) | |
8088 | seq_printf(m, "%5u: %s\n", i, file_dentry(f)->d_iname); | |
8089 | else | |
8090 | seq_printf(m, "%5u: <none>\n", i); | |
8091 | } | |
8092 | seq_printf(m, "UserBufs:\t%u\n", ctx->nr_user_bufs); | |
8093 | for (i = 0; i < ctx->nr_user_bufs; i++) { | |
8094 | struct io_mapped_ubuf *buf = &ctx->user_bufs[i]; | |
8095 | ||
8096 | seq_printf(m, "%5u: 0x%llx/%u\n", i, buf->ubuf, | |
8097 | (unsigned int) buf->len); | |
8098 | } | |
8099 | if (!idr_is_empty(&ctx->personality_idr)) { | |
8100 | seq_printf(m, "Personalities:\n"); | |
8101 | idr_for_each(&ctx->personality_idr, io_uring_show_cred, m); | |
8102 | } | |
d7718a9d JA |
8103 | seq_printf(m, "PollList:\n"); |
8104 | spin_lock_irq(&ctx->completion_lock); | |
8105 | for (i = 0; i < (1U << ctx->cancel_hash_bits); i++) { | |
8106 | struct hlist_head *list = &ctx->cancel_hash[i]; | |
8107 | struct io_kiocb *req; | |
8108 | ||
8109 | hlist_for_each_entry(req, list, hash_node) | |
8110 | seq_printf(m, " op=%d, task_works=%d\n", req->opcode, | |
8111 | req->task->task_works != NULL); | |
8112 | } | |
8113 | spin_unlock_irq(&ctx->completion_lock); | |
87ce955b JA |
8114 | mutex_unlock(&ctx->uring_lock); |
8115 | } | |
8116 | ||
8117 | static void io_uring_show_fdinfo(struct seq_file *m, struct file *f) | |
8118 | { | |
8119 | struct io_ring_ctx *ctx = f->private_data; | |
8120 | ||
8121 | if (percpu_ref_tryget(&ctx->refs)) { | |
8122 | __io_uring_show_fdinfo(ctx, m); | |
8123 | percpu_ref_put(&ctx->refs); | |
8124 | } | |
8125 | } | |
bebdb65e | 8126 | #endif |
87ce955b | 8127 | |
2b188cc1 JA |
8128 | static const struct file_operations io_uring_fops = { |
8129 | .release = io_uring_release, | |
fcb323cc | 8130 | .flush = io_uring_flush, |
2b188cc1 | 8131 | .mmap = io_uring_mmap, |
6c5c240e RP |
8132 | #ifndef CONFIG_MMU |
8133 | .get_unmapped_area = io_uring_nommu_get_unmapped_area, | |
8134 | .mmap_capabilities = io_uring_nommu_mmap_capabilities, | |
8135 | #endif | |
2b188cc1 JA |
8136 | .poll = io_uring_poll, |
8137 | .fasync = io_uring_fasync, | |
bebdb65e | 8138 | #ifdef CONFIG_PROC_FS |
87ce955b | 8139 | .show_fdinfo = io_uring_show_fdinfo, |
bebdb65e | 8140 | #endif |
2b188cc1 JA |
8141 | }; |
8142 | ||
8143 | static int io_allocate_scq_urings(struct io_ring_ctx *ctx, | |
8144 | struct io_uring_params *p) | |
8145 | { | |
75b28aff HV |
8146 | struct io_rings *rings; |
8147 | size_t size, sq_array_offset; | |
2b188cc1 | 8148 | |
75b28aff HV |
8149 | size = rings_size(p->sq_entries, p->cq_entries, &sq_array_offset); |
8150 | if (size == SIZE_MAX) | |
8151 | return -EOVERFLOW; | |
8152 | ||
8153 | rings = io_mem_alloc(size); | |
8154 | if (!rings) | |
2b188cc1 JA |
8155 | return -ENOMEM; |
8156 | ||
75b28aff HV |
8157 | ctx->rings = rings; |
8158 | ctx->sq_array = (u32 *)((char *)rings + sq_array_offset); | |
8159 | rings->sq_ring_mask = p->sq_entries - 1; | |
8160 | rings->cq_ring_mask = p->cq_entries - 1; | |
8161 | rings->sq_ring_entries = p->sq_entries; | |
8162 | rings->cq_ring_entries = p->cq_entries; | |
8163 | ctx->sq_mask = rings->sq_ring_mask; | |
8164 | ctx->cq_mask = rings->cq_ring_mask; | |
8165 | ctx->sq_entries = rings->sq_ring_entries; | |
8166 | ctx->cq_entries = rings->cq_ring_entries; | |
2b188cc1 JA |
8167 | |
8168 | size = array_size(sizeof(struct io_uring_sqe), p->sq_entries); | |
eb065d30 JA |
8169 | if (size == SIZE_MAX) { |
8170 | io_mem_free(ctx->rings); | |
8171 | ctx->rings = NULL; | |
2b188cc1 | 8172 | return -EOVERFLOW; |
eb065d30 | 8173 | } |
2b188cc1 JA |
8174 | |
8175 | ctx->sq_sqes = io_mem_alloc(size); | |
eb065d30 JA |
8176 | if (!ctx->sq_sqes) { |
8177 | io_mem_free(ctx->rings); | |
8178 | ctx->rings = NULL; | |
2b188cc1 | 8179 | return -ENOMEM; |
eb065d30 | 8180 | } |
2b188cc1 | 8181 | |
2b188cc1 JA |
8182 | return 0; |
8183 | } | |
8184 | ||
8185 | /* | |
8186 | * Allocate an anonymous fd, this is what constitutes the application | |
8187 | * visible backing of an io_uring instance. The application mmaps this | |
8188 | * fd to gain access to the SQ/CQ ring details. If UNIX sockets are enabled, | |
8189 | * we have to tie this fd to a socket for file garbage collection purposes. | |
8190 | */ | |
8191 | static int io_uring_get_fd(struct io_ring_ctx *ctx) | |
8192 | { | |
8193 | struct file *file; | |
8194 | int ret; | |
8195 | ||
8196 | #if defined(CONFIG_UNIX) | |
8197 | ret = sock_create_kern(&init_net, PF_UNIX, SOCK_RAW, IPPROTO_IP, | |
8198 | &ctx->ring_sock); | |
8199 | if (ret) | |
8200 | return ret; | |
8201 | #endif | |
8202 | ||
8203 | ret = get_unused_fd_flags(O_RDWR | O_CLOEXEC); | |
8204 | if (ret < 0) | |
8205 | goto err; | |
8206 | ||
8207 | file = anon_inode_getfile("[io_uring]", &io_uring_fops, ctx, | |
8208 | O_RDWR | O_CLOEXEC); | |
8209 | if (IS_ERR(file)) { | |
8210 | put_unused_fd(ret); | |
8211 | ret = PTR_ERR(file); | |
8212 | goto err; | |
8213 | } | |
8214 | ||
8215 | #if defined(CONFIG_UNIX) | |
8216 | ctx->ring_sock->file = file; | |
8217 | #endif | |
8218 | fd_install(ret, file); | |
8219 | return ret; | |
8220 | err: | |
8221 | #if defined(CONFIG_UNIX) | |
8222 | sock_release(ctx->ring_sock); | |
8223 | ctx->ring_sock = NULL; | |
8224 | #endif | |
8225 | return ret; | |
8226 | } | |
8227 | ||
7f13657d XW |
8228 | static int io_uring_create(unsigned entries, struct io_uring_params *p, |
8229 | struct io_uring_params __user *params) | |
2b188cc1 JA |
8230 | { |
8231 | struct user_struct *user = NULL; | |
8232 | struct io_ring_ctx *ctx; | |
aad5d8da | 8233 | bool limit_mem; |
2b188cc1 JA |
8234 | int ret; |
8235 | ||
8110c1a6 | 8236 | if (!entries) |
2b188cc1 | 8237 | return -EINVAL; |
8110c1a6 JA |
8238 | if (entries > IORING_MAX_ENTRIES) { |
8239 | if (!(p->flags & IORING_SETUP_CLAMP)) | |
8240 | return -EINVAL; | |
8241 | entries = IORING_MAX_ENTRIES; | |
8242 | } | |
2b188cc1 JA |
8243 | |
8244 | /* | |
8245 | * Use twice as many entries for the CQ ring. It's possible for the | |
8246 | * application to drive a higher depth than the size of the SQ ring, | |
8247 | * since the sqes are only used at submission time. This allows for | |
33a107f0 JA |
8248 | * some flexibility in overcommitting a bit. If the application has |
8249 | * set IORING_SETUP_CQSIZE, it will have passed in the desired number | |
8250 | * of CQ ring entries manually. | |
2b188cc1 JA |
8251 | */ |
8252 | p->sq_entries = roundup_pow_of_two(entries); | |
33a107f0 JA |
8253 | if (p->flags & IORING_SETUP_CQSIZE) { |
8254 | /* | |
8255 | * If IORING_SETUP_CQSIZE is set, we do the same roundup | |
8256 | * to a power-of-two, if it isn't already. We do NOT impose | |
8257 | * any cq vs sq ring sizing. | |
8258 | */ | |
8110c1a6 | 8259 | if (p->cq_entries < p->sq_entries) |
33a107f0 | 8260 | return -EINVAL; |
8110c1a6 JA |
8261 | if (p->cq_entries > IORING_MAX_CQ_ENTRIES) { |
8262 | if (!(p->flags & IORING_SETUP_CLAMP)) | |
8263 | return -EINVAL; | |
8264 | p->cq_entries = IORING_MAX_CQ_ENTRIES; | |
8265 | } | |
33a107f0 JA |
8266 | p->cq_entries = roundup_pow_of_two(p->cq_entries); |
8267 | } else { | |
8268 | p->cq_entries = 2 * p->sq_entries; | |
8269 | } | |
2b188cc1 JA |
8270 | |
8271 | user = get_uid(current_user()); | |
aad5d8da | 8272 | limit_mem = !capable(CAP_IPC_LOCK); |
2b188cc1 | 8273 | |
aad5d8da | 8274 | if (limit_mem) { |
a087e2b5 | 8275 | ret = __io_account_mem(user, |
2b188cc1 JA |
8276 | ring_pages(p->sq_entries, p->cq_entries)); |
8277 | if (ret) { | |
8278 | free_uid(user); | |
8279 | return ret; | |
8280 | } | |
8281 | } | |
8282 | ||
8283 | ctx = io_ring_ctx_alloc(p); | |
8284 | if (!ctx) { | |
aad5d8da | 8285 | if (limit_mem) |
a087e2b5 | 8286 | __io_unaccount_mem(user, ring_pages(p->sq_entries, |
2b188cc1 JA |
8287 | p->cq_entries)); |
8288 | free_uid(user); | |
8289 | return -ENOMEM; | |
8290 | } | |
8291 | ctx->compat = in_compat_syscall(); | |
2b188cc1 | 8292 | ctx->user = user; |
0b8c0ec7 | 8293 | ctx->creds = get_current_cred(); |
2b188cc1 JA |
8294 | |
8295 | ret = io_allocate_scq_urings(ctx, p); | |
8296 | if (ret) | |
8297 | goto err; | |
8298 | ||
6c271ce2 | 8299 | ret = io_sq_offload_start(ctx, p); |
2b188cc1 JA |
8300 | if (ret) |
8301 | goto err; | |
8302 | ||
2b188cc1 | 8303 | memset(&p->sq_off, 0, sizeof(p->sq_off)); |
75b28aff HV |
8304 | p->sq_off.head = offsetof(struct io_rings, sq.head); |
8305 | p->sq_off.tail = offsetof(struct io_rings, sq.tail); | |
8306 | p->sq_off.ring_mask = offsetof(struct io_rings, sq_ring_mask); | |
8307 | p->sq_off.ring_entries = offsetof(struct io_rings, sq_ring_entries); | |
8308 | p->sq_off.flags = offsetof(struct io_rings, sq_flags); | |
8309 | p->sq_off.dropped = offsetof(struct io_rings, sq_dropped); | |
8310 | p->sq_off.array = (char *)ctx->sq_array - (char *)ctx->rings; | |
2b188cc1 JA |
8311 | |
8312 | memset(&p->cq_off, 0, sizeof(p->cq_off)); | |
75b28aff HV |
8313 | p->cq_off.head = offsetof(struct io_rings, cq.head); |
8314 | p->cq_off.tail = offsetof(struct io_rings, cq.tail); | |
8315 | p->cq_off.ring_mask = offsetof(struct io_rings, cq_ring_mask); | |
8316 | p->cq_off.ring_entries = offsetof(struct io_rings, cq_ring_entries); | |
8317 | p->cq_off.overflow = offsetof(struct io_rings, cq_overflow); | |
8318 | p->cq_off.cqes = offsetof(struct io_rings, cqes); | |
0d9b5b3a | 8319 | p->cq_off.flags = offsetof(struct io_rings, cq_flags); |
ac90f249 | 8320 | |
7f13657d XW |
8321 | p->features = IORING_FEAT_SINGLE_MMAP | IORING_FEAT_NODROP | |
8322 | IORING_FEAT_SUBMIT_STABLE | IORING_FEAT_RW_CUR_POS | | |
5769a351 JX |
8323 | IORING_FEAT_CUR_PERSONALITY | IORING_FEAT_FAST_POLL | |
8324 | IORING_FEAT_POLL_32BITS; | |
7f13657d XW |
8325 | |
8326 | if (copy_to_user(params, p, sizeof(*p))) { | |
8327 | ret = -EFAULT; | |
8328 | goto err; | |
8329 | } | |
044c1ab3 JA |
8330 | /* |
8331 | * Install ring fd as the very last thing, so we don't risk someone | |
8332 | * having closed it before we finish setup | |
8333 | */ | |
8334 | ret = io_uring_get_fd(ctx); | |
8335 | if (ret < 0) | |
8336 | goto err; | |
8337 | ||
c826bd7a | 8338 | trace_io_uring_create(ret, ctx, p->sq_entries, p->cq_entries, p->flags); |
2e0464d4 BM |
8339 | io_account_mem(ctx, ring_pages(p->sq_entries, p->cq_entries), |
8340 | ACCT_LOCKED); | |
30975825 | 8341 | ctx->limit_mem = limit_mem; |
2b188cc1 JA |
8342 | return ret; |
8343 | err: | |
8344 | io_ring_ctx_wait_and_kill(ctx); | |
8345 | return ret; | |
8346 | } | |
8347 | ||
8348 | /* | |
8349 | * Sets up an aio uring context, and returns the fd. Applications asks for a | |
8350 | * ring size, we return the actual sq/cq ring sizes (among other things) in the | |
8351 | * params structure passed in. | |
8352 | */ | |
8353 | static long io_uring_setup(u32 entries, struct io_uring_params __user *params) | |
8354 | { | |
8355 | struct io_uring_params p; | |
2b188cc1 JA |
8356 | int i; |
8357 | ||
8358 | if (copy_from_user(&p, params, sizeof(p))) | |
8359 | return -EFAULT; | |
8360 | for (i = 0; i < ARRAY_SIZE(p.resv); i++) { | |
8361 | if (p.resv[i]) | |
8362 | return -EINVAL; | |
8363 | } | |
8364 | ||
6c271ce2 | 8365 | if (p.flags & ~(IORING_SETUP_IOPOLL | IORING_SETUP_SQPOLL | |
8110c1a6 | 8366 | IORING_SETUP_SQ_AFF | IORING_SETUP_CQSIZE | |
24369c2e | 8367 | IORING_SETUP_CLAMP | IORING_SETUP_ATTACH_WQ)) |
2b188cc1 JA |
8368 | return -EINVAL; |
8369 | ||
7f13657d | 8370 | return io_uring_create(entries, &p, params); |
2b188cc1 JA |
8371 | } |
8372 | ||
8373 | SYSCALL_DEFINE2(io_uring_setup, u32, entries, | |
8374 | struct io_uring_params __user *, params) | |
8375 | { | |
8376 | return io_uring_setup(entries, params); | |
8377 | } | |
8378 | ||
66f4af93 JA |
8379 | static int io_probe(struct io_ring_ctx *ctx, void __user *arg, unsigned nr_args) |
8380 | { | |
8381 | struct io_uring_probe *p; | |
8382 | size_t size; | |
8383 | int i, ret; | |
8384 | ||
8385 | size = struct_size(p, ops, nr_args); | |
8386 | if (size == SIZE_MAX) | |
8387 | return -EOVERFLOW; | |
8388 | p = kzalloc(size, GFP_KERNEL); | |
8389 | if (!p) | |
8390 | return -ENOMEM; | |
8391 | ||
8392 | ret = -EFAULT; | |
8393 | if (copy_from_user(p, arg, size)) | |
8394 | goto out; | |
8395 | ret = -EINVAL; | |
8396 | if (memchr_inv(p, 0, size)) | |
8397 | goto out; | |
8398 | ||
8399 | p->last_op = IORING_OP_LAST - 1; | |
8400 | if (nr_args > IORING_OP_LAST) | |
8401 | nr_args = IORING_OP_LAST; | |
8402 | ||
8403 | for (i = 0; i < nr_args; i++) { | |
8404 | p->ops[i].op = i; | |
8405 | if (!io_op_defs[i].not_supported) | |
8406 | p->ops[i].flags = IO_URING_OP_SUPPORTED; | |
8407 | } | |
8408 | p->ops_len = i; | |
8409 | ||
8410 | ret = 0; | |
8411 | if (copy_to_user(arg, p, size)) | |
8412 | ret = -EFAULT; | |
8413 | out: | |
8414 | kfree(p); | |
8415 | return ret; | |
8416 | } | |
8417 | ||
071698e1 JA |
8418 | static int io_register_personality(struct io_ring_ctx *ctx) |
8419 | { | |
8420 | const struct cred *creds = get_current_cred(); | |
8421 | int id; | |
8422 | ||
8423 | id = idr_alloc_cyclic(&ctx->personality_idr, (void *) creds, 1, | |
8424 | USHRT_MAX, GFP_KERNEL); | |
8425 | if (id < 0) | |
8426 | put_cred(creds); | |
8427 | return id; | |
8428 | } | |
8429 | ||
8430 | static int io_unregister_personality(struct io_ring_ctx *ctx, unsigned id) | |
8431 | { | |
8432 | const struct cred *old_creds; | |
8433 | ||
8434 | old_creds = idr_remove(&ctx->personality_idr, id); | |
8435 | if (old_creds) { | |
8436 | put_cred(old_creds); | |
8437 | return 0; | |
8438 | } | |
8439 | ||
8440 | return -EINVAL; | |
8441 | } | |
8442 | ||
8443 | static bool io_register_op_must_quiesce(int op) | |
8444 | { | |
8445 | switch (op) { | |
8446 | case IORING_UNREGISTER_FILES: | |
8447 | case IORING_REGISTER_FILES_UPDATE: | |
8448 | case IORING_REGISTER_PROBE: | |
8449 | case IORING_REGISTER_PERSONALITY: | |
8450 | case IORING_UNREGISTER_PERSONALITY: | |
8451 | return false; | |
8452 | default: | |
8453 | return true; | |
8454 | } | |
8455 | } | |
8456 | ||
edafccee JA |
8457 | static int __io_uring_register(struct io_ring_ctx *ctx, unsigned opcode, |
8458 | void __user *arg, unsigned nr_args) | |
b19062a5 JA |
8459 | __releases(ctx->uring_lock) |
8460 | __acquires(ctx->uring_lock) | |
edafccee JA |
8461 | { |
8462 | int ret; | |
8463 | ||
35fa71a0 JA |
8464 | /* |
8465 | * We're inside the ring mutex, if the ref is already dying, then | |
8466 | * someone else killed the ctx or is already going through | |
8467 | * io_uring_register(). | |
8468 | */ | |
8469 | if (percpu_ref_is_dying(&ctx->refs)) | |
8470 | return -ENXIO; | |
8471 | ||
071698e1 | 8472 | if (io_register_op_must_quiesce(opcode)) { |
05f3fb3c | 8473 | percpu_ref_kill(&ctx->refs); |
b19062a5 | 8474 | |
05f3fb3c JA |
8475 | /* |
8476 | * Drop uring mutex before waiting for references to exit. If | |
8477 | * another thread is currently inside io_uring_enter() it might | |
8478 | * need to grab the uring_lock to make progress. If we hold it | |
8479 | * here across the drain wait, then we can deadlock. It's safe | |
8480 | * to drop the mutex here, since no new references will come in | |
8481 | * after we've killed the percpu ref. | |
8482 | */ | |
8483 | mutex_unlock(&ctx->uring_lock); | |
0f158b4c | 8484 | ret = wait_for_completion_interruptible(&ctx->ref_comp); |
05f3fb3c | 8485 | mutex_lock(&ctx->uring_lock); |
c150368b JA |
8486 | if (ret) { |
8487 | percpu_ref_resurrect(&ctx->refs); | |
8488 | ret = -EINTR; | |
8489 | goto out; | |
8490 | } | |
05f3fb3c | 8491 | } |
edafccee JA |
8492 | |
8493 | switch (opcode) { | |
8494 | case IORING_REGISTER_BUFFERS: | |
8495 | ret = io_sqe_buffer_register(ctx, arg, nr_args); | |
8496 | break; | |
8497 | case IORING_UNREGISTER_BUFFERS: | |
8498 | ret = -EINVAL; | |
8499 | if (arg || nr_args) | |
8500 | break; | |
8501 | ret = io_sqe_buffer_unregister(ctx); | |
8502 | break; | |
6b06314c JA |
8503 | case IORING_REGISTER_FILES: |
8504 | ret = io_sqe_files_register(ctx, arg, nr_args); | |
8505 | break; | |
8506 | case IORING_UNREGISTER_FILES: | |
8507 | ret = -EINVAL; | |
8508 | if (arg || nr_args) | |
8509 | break; | |
8510 | ret = io_sqe_files_unregister(ctx); | |
8511 | break; | |
c3a31e60 JA |
8512 | case IORING_REGISTER_FILES_UPDATE: |
8513 | ret = io_sqe_files_update(ctx, arg, nr_args); | |
8514 | break; | |
9b402849 | 8515 | case IORING_REGISTER_EVENTFD: |
f2842ab5 | 8516 | case IORING_REGISTER_EVENTFD_ASYNC: |
9b402849 JA |
8517 | ret = -EINVAL; |
8518 | if (nr_args != 1) | |
8519 | break; | |
8520 | ret = io_eventfd_register(ctx, arg); | |
f2842ab5 JA |
8521 | if (ret) |
8522 | break; | |
8523 | if (opcode == IORING_REGISTER_EVENTFD_ASYNC) | |
8524 | ctx->eventfd_async = 1; | |
8525 | else | |
8526 | ctx->eventfd_async = 0; | |
9b402849 JA |
8527 | break; |
8528 | case IORING_UNREGISTER_EVENTFD: | |
8529 | ret = -EINVAL; | |
8530 | if (arg || nr_args) | |
8531 | break; | |
8532 | ret = io_eventfd_unregister(ctx); | |
8533 | break; | |
66f4af93 JA |
8534 | case IORING_REGISTER_PROBE: |
8535 | ret = -EINVAL; | |
8536 | if (!arg || nr_args > 256) | |
8537 | break; | |
8538 | ret = io_probe(ctx, arg, nr_args); | |
8539 | break; | |
071698e1 JA |
8540 | case IORING_REGISTER_PERSONALITY: |
8541 | ret = -EINVAL; | |
8542 | if (arg || nr_args) | |
8543 | break; | |
8544 | ret = io_register_personality(ctx); | |
8545 | break; | |
8546 | case IORING_UNREGISTER_PERSONALITY: | |
8547 | ret = -EINVAL; | |
8548 | if (arg) | |
8549 | break; | |
8550 | ret = io_unregister_personality(ctx, nr_args); | |
8551 | break; | |
edafccee JA |
8552 | default: |
8553 | ret = -EINVAL; | |
8554 | break; | |
8555 | } | |
8556 | ||
071698e1 | 8557 | if (io_register_op_must_quiesce(opcode)) { |
05f3fb3c | 8558 | /* bring the ctx back to life */ |
05f3fb3c | 8559 | percpu_ref_reinit(&ctx->refs); |
c150368b | 8560 | out: |
0f158b4c | 8561 | reinit_completion(&ctx->ref_comp); |
05f3fb3c | 8562 | } |
edafccee JA |
8563 | return ret; |
8564 | } | |
8565 | ||
8566 | SYSCALL_DEFINE4(io_uring_register, unsigned int, fd, unsigned int, opcode, | |
8567 | void __user *, arg, unsigned int, nr_args) | |
8568 | { | |
8569 | struct io_ring_ctx *ctx; | |
8570 | long ret = -EBADF; | |
8571 | struct fd f; | |
8572 | ||
8573 | f = fdget(fd); | |
8574 | if (!f.file) | |
8575 | return -EBADF; | |
8576 | ||
8577 | ret = -EOPNOTSUPP; | |
8578 | if (f.file->f_op != &io_uring_fops) | |
8579 | goto out_fput; | |
8580 | ||
8581 | ctx = f.file->private_data; | |
8582 | ||
8583 | mutex_lock(&ctx->uring_lock); | |
8584 | ret = __io_uring_register(ctx, opcode, arg, nr_args); | |
8585 | mutex_unlock(&ctx->uring_lock); | |
c826bd7a DD |
8586 | trace_io_uring_register(ctx, opcode, ctx->nr_user_files, ctx->nr_user_bufs, |
8587 | ctx->cq_ev_fd != NULL, ret); | |
edafccee JA |
8588 | out_fput: |
8589 | fdput(f); | |
8590 | return ret; | |
8591 | } | |
8592 | ||
2b188cc1 JA |
8593 | static int __init io_uring_init(void) |
8594 | { | |
d7f62e82 SM |
8595 | #define __BUILD_BUG_VERIFY_ELEMENT(stype, eoffset, etype, ename) do { \ |
8596 | BUILD_BUG_ON(offsetof(stype, ename) != eoffset); \ | |
8597 | BUILD_BUG_ON(sizeof(etype) != sizeof_field(stype, ename)); \ | |
8598 | } while (0) | |
8599 | ||
8600 | #define BUILD_BUG_SQE_ELEM(eoffset, etype, ename) \ | |
8601 | __BUILD_BUG_VERIFY_ELEMENT(struct io_uring_sqe, eoffset, etype, ename) | |
8602 | BUILD_BUG_ON(sizeof(struct io_uring_sqe) != 64); | |
8603 | BUILD_BUG_SQE_ELEM(0, __u8, opcode); | |
8604 | BUILD_BUG_SQE_ELEM(1, __u8, flags); | |
8605 | BUILD_BUG_SQE_ELEM(2, __u16, ioprio); | |
8606 | BUILD_BUG_SQE_ELEM(4, __s32, fd); | |
8607 | BUILD_BUG_SQE_ELEM(8, __u64, off); | |
8608 | BUILD_BUG_SQE_ELEM(8, __u64, addr2); | |
8609 | BUILD_BUG_SQE_ELEM(16, __u64, addr); | |
7d67af2c | 8610 | BUILD_BUG_SQE_ELEM(16, __u64, splice_off_in); |
d7f62e82 SM |
8611 | BUILD_BUG_SQE_ELEM(24, __u32, len); |
8612 | BUILD_BUG_SQE_ELEM(28, __kernel_rwf_t, rw_flags); | |
8613 | BUILD_BUG_SQE_ELEM(28, /* compat */ int, rw_flags); | |
8614 | BUILD_BUG_SQE_ELEM(28, /* compat */ __u32, rw_flags); | |
8615 | BUILD_BUG_SQE_ELEM(28, __u32, fsync_flags); | |
5769a351 JX |
8616 | BUILD_BUG_SQE_ELEM(28, /* compat */ __u16, poll_events); |
8617 | BUILD_BUG_SQE_ELEM(28, __u32, poll32_events); | |
d7f62e82 SM |
8618 | BUILD_BUG_SQE_ELEM(28, __u32, sync_range_flags); |
8619 | BUILD_BUG_SQE_ELEM(28, __u32, msg_flags); | |
8620 | BUILD_BUG_SQE_ELEM(28, __u32, timeout_flags); | |
8621 | BUILD_BUG_SQE_ELEM(28, __u32, accept_flags); | |
8622 | BUILD_BUG_SQE_ELEM(28, __u32, cancel_flags); | |
8623 | BUILD_BUG_SQE_ELEM(28, __u32, open_flags); | |
8624 | BUILD_BUG_SQE_ELEM(28, __u32, statx_flags); | |
8625 | BUILD_BUG_SQE_ELEM(28, __u32, fadvise_advice); | |
7d67af2c | 8626 | BUILD_BUG_SQE_ELEM(28, __u32, splice_flags); |
d7f62e82 SM |
8627 | BUILD_BUG_SQE_ELEM(32, __u64, user_data); |
8628 | BUILD_BUG_SQE_ELEM(40, __u16, buf_index); | |
8629 | BUILD_BUG_SQE_ELEM(42, __u16, personality); | |
7d67af2c | 8630 | BUILD_BUG_SQE_ELEM(44, __s32, splice_fd_in); |
d7f62e82 | 8631 | |
d3656344 | 8632 | BUILD_BUG_ON(ARRAY_SIZE(io_op_defs) != IORING_OP_LAST); |
84557871 | 8633 | BUILD_BUG_ON(__REQ_F_LAST_BIT >= 8 * sizeof(int)); |
2b188cc1 JA |
8634 | req_cachep = KMEM_CACHE(io_kiocb, SLAB_HWCACHE_ALIGN | SLAB_PANIC); |
8635 | return 0; | |
8636 | }; | |
8637 | __initcall(io_uring_init); |