]>
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> | |
58 | #include <linux/mmu_context.h> | |
59 | #include <linux/percpu.h> | |
60 | #include <linux/slab.h> | |
6c271ce2 | 61 | #include <linux/kthread.h> |
2b188cc1 | 62 | #include <linux/blkdev.h> |
edafccee | 63 | #include <linux/bvec.h> |
2b188cc1 JA |
64 | #include <linux/net.h> |
65 | #include <net/sock.h> | |
66 | #include <net/af_unix.h> | |
6b06314c | 67 | #include <net/scm.h> |
2b188cc1 JA |
68 | #include <linux/anon_inodes.h> |
69 | #include <linux/sched/mm.h> | |
70 | #include <linux/uaccess.h> | |
71 | #include <linux/nospec.h> | |
edafccee JA |
72 | #include <linux/sizes.h> |
73 | #include <linux/hugetlb.h> | |
aa4c3967 | 74 | #include <linux/highmem.h> |
15b71abe JA |
75 | #include <linux/namei.h> |
76 | #include <linux/fsnotify.h> | |
4840e418 | 77 | #include <linux/fadvise.h> |
3e4827b0 | 78 | #include <linux/eventpoll.h> |
ff002b30 | 79 | #include <linux/fs_struct.h> |
7d67af2c | 80 | #include <linux/splice.h> |
b41e9852 | 81 | #include <linux/task_work.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 RD |
229 | unsigned int compat: 1; |
230 | unsigned int account_mem: 1; | |
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 JA |
322 | /* |
323 | * ->poll_list is protected by the ctx->uring_lock for | |
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 | */ | |
328 | struct list_head poll_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; | |
b55ce732 | 397 | u32 count; |
b29472ee JA |
398 | }; |
399 | ||
9adbd45d JA |
400 | struct io_rw { |
401 | /* NOTE: kiocb has the file as the first member, so don't do it here */ | |
402 | struct kiocb kiocb; | |
403 | u64 addr; | |
404 | u64 len; | |
405 | }; | |
406 | ||
3fbb51c1 JA |
407 | struct io_connect { |
408 | struct file *file; | |
409 | struct sockaddr __user *addr; | |
410 | int addr_len; | |
411 | }; | |
412 | ||
e47293fd JA |
413 | struct io_sr_msg { |
414 | struct file *file; | |
fddaface JA |
415 | union { |
416 | struct user_msghdr __user *msg; | |
417 | void __user *buf; | |
418 | }; | |
e47293fd | 419 | int msg_flags; |
bcda7baa | 420 | int bgid; |
fddaface | 421 | size_t len; |
bcda7baa | 422 | struct io_buffer *kbuf; |
e47293fd JA |
423 | }; |
424 | ||
15b71abe JA |
425 | struct io_open { |
426 | struct file *file; | |
427 | int dfd; | |
eddc7ef5 | 428 | union { |
eddc7ef5 JA |
429 | unsigned mask; |
430 | }; | |
15b71abe | 431 | struct filename *filename; |
eddc7ef5 | 432 | struct statx __user *buffer; |
c12cedf2 | 433 | struct open_how how; |
4022e7af | 434 | unsigned long nofile; |
15b71abe JA |
435 | }; |
436 | ||
05f3fb3c JA |
437 | struct io_files_update { |
438 | struct file *file; | |
439 | u64 arg; | |
440 | u32 nr_args; | |
441 | u32 offset; | |
442 | }; | |
443 | ||
4840e418 JA |
444 | struct io_fadvise { |
445 | struct file *file; | |
446 | u64 offset; | |
447 | u32 len; | |
448 | u32 advice; | |
449 | }; | |
450 | ||
c1ca757b JA |
451 | struct io_madvise { |
452 | struct file *file; | |
453 | u64 addr; | |
454 | u32 len; | |
455 | u32 advice; | |
456 | }; | |
457 | ||
3e4827b0 JA |
458 | struct io_epoll { |
459 | struct file *file; | |
460 | int epfd; | |
461 | int op; | |
462 | int fd; | |
463 | struct epoll_event event; | |
e47293fd JA |
464 | }; |
465 | ||
7d67af2c PB |
466 | struct io_splice { |
467 | struct file *file_out; | |
468 | struct file *file_in; | |
469 | loff_t off_out; | |
470 | loff_t off_in; | |
471 | u64 len; | |
472 | unsigned int flags; | |
473 | }; | |
474 | ||
ddf0322d JA |
475 | struct io_provide_buf { |
476 | struct file *file; | |
477 | __u64 addr; | |
478 | __s32 len; | |
479 | __u32 bgid; | |
480 | __u16 nbufs; | |
481 | __u16 bid; | |
482 | }; | |
483 | ||
f499a021 JA |
484 | struct io_async_connect { |
485 | struct sockaddr_storage address; | |
486 | }; | |
487 | ||
03b1230c JA |
488 | struct io_async_msghdr { |
489 | struct iovec fast_iov[UIO_FASTIOV]; | |
490 | struct iovec *iov; | |
491 | struct sockaddr __user *uaddr; | |
492 | struct msghdr msg; | |
b537916c | 493 | struct sockaddr_storage addr; |
03b1230c JA |
494 | }; |
495 | ||
f67676d1 JA |
496 | struct io_async_rw { |
497 | struct iovec fast_iov[UIO_FASTIOV]; | |
498 | struct iovec *iov; | |
499 | ssize_t nr_segs; | |
500 | ssize_t size; | |
501 | }; | |
502 | ||
1a6b74fc | 503 | struct io_async_ctx { |
f67676d1 JA |
504 | union { |
505 | struct io_async_rw rw; | |
03b1230c | 506 | struct io_async_msghdr msg; |
f499a021 | 507 | struct io_async_connect connect; |
2d28390a | 508 | struct io_timeout_data timeout; |
f67676d1 | 509 | }; |
1a6b74fc JA |
510 | }; |
511 | ||
6b47ee6e PB |
512 | enum { |
513 | REQ_F_FIXED_FILE_BIT = IOSQE_FIXED_FILE_BIT, | |
514 | REQ_F_IO_DRAIN_BIT = IOSQE_IO_DRAIN_BIT, | |
515 | REQ_F_LINK_BIT = IOSQE_IO_LINK_BIT, | |
516 | REQ_F_HARDLINK_BIT = IOSQE_IO_HARDLINK_BIT, | |
517 | REQ_F_FORCE_ASYNC_BIT = IOSQE_ASYNC_BIT, | |
bcda7baa | 518 | REQ_F_BUFFER_SELECT_BIT = IOSQE_BUFFER_SELECT_BIT, |
6b47ee6e | 519 | |
dea3b49c | 520 | REQ_F_LINK_HEAD_BIT, |
6b47ee6e PB |
521 | REQ_F_LINK_NEXT_BIT, |
522 | REQ_F_FAIL_LINK_BIT, | |
523 | REQ_F_INFLIGHT_BIT, | |
524 | REQ_F_CUR_POS_BIT, | |
525 | REQ_F_NOWAIT_BIT, | |
526 | REQ_F_IOPOLL_COMPLETED_BIT, | |
527 | REQ_F_LINK_TIMEOUT_BIT, | |
528 | REQ_F_TIMEOUT_BIT, | |
529 | REQ_F_ISREG_BIT, | |
530 | REQ_F_MUST_PUNT_BIT, | |
531 | REQ_F_TIMEOUT_NOSEQ_BIT, | |
532 | REQ_F_COMP_LOCKED_BIT, | |
99bc4c38 | 533 | REQ_F_NEED_CLEANUP_BIT, |
2ca10259 | 534 | REQ_F_OVERFLOW_BIT, |
d7718a9d | 535 | REQ_F_POLLED_BIT, |
bcda7baa | 536 | REQ_F_BUFFER_SELECTED_BIT, |
5b0bbee4 | 537 | REQ_F_NO_FILE_TABLE_BIT, |
84557871 JA |
538 | |
539 | /* not a real bit, just to check we're not overflowing the space */ | |
540 | __REQ_F_LAST_BIT, | |
6b47ee6e PB |
541 | }; |
542 | ||
543 | enum { | |
544 | /* ctx owns file */ | |
545 | REQ_F_FIXED_FILE = BIT(REQ_F_FIXED_FILE_BIT), | |
546 | /* drain existing IO first */ | |
547 | REQ_F_IO_DRAIN = BIT(REQ_F_IO_DRAIN_BIT), | |
548 | /* linked sqes */ | |
549 | REQ_F_LINK = BIT(REQ_F_LINK_BIT), | |
550 | /* doesn't sever on completion < 0 */ | |
551 | REQ_F_HARDLINK = BIT(REQ_F_HARDLINK_BIT), | |
552 | /* IOSQE_ASYNC */ | |
553 | REQ_F_FORCE_ASYNC = BIT(REQ_F_FORCE_ASYNC_BIT), | |
bcda7baa JA |
554 | /* IOSQE_BUFFER_SELECT */ |
555 | REQ_F_BUFFER_SELECT = BIT(REQ_F_BUFFER_SELECT_BIT), | |
6b47ee6e | 556 | |
dea3b49c PB |
557 | /* head of a link */ |
558 | REQ_F_LINK_HEAD = BIT(REQ_F_LINK_HEAD_BIT), | |
6b47ee6e PB |
559 | /* already grabbed next link */ |
560 | REQ_F_LINK_NEXT = BIT(REQ_F_LINK_NEXT_BIT), | |
561 | /* fail rest of links */ | |
562 | REQ_F_FAIL_LINK = BIT(REQ_F_FAIL_LINK_BIT), | |
563 | /* on inflight list */ | |
564 | REQ_F_INFLIGHT = BIT(REQ_F_INFLIGHT_BIT), | |
565 | /* read/write uses file position */ | |
566 | REQ_F_CUR_POS = BIT(REQ_F_CUR_POS_BIT), | |
567 | /* must not punt to workers */ | |
568 | REQ_F_NOWAIT = BIT(REQ_F_NOWAIT_BIT), | |
569 | /* polled IO has completed */ | |
570 | REQ_F_IOPOLL_COMPLETED = BIT(REQ_F_IOPOLL_COMPLETED_BIT), | |
571 | /* has linked timeout */ | |
572 | REQ_F_LINK_TIMEOUT = BIT(REQ_F_LINK_TIMEOUT_BIT), | |
573 | /* timeout request */ | |
574 | REQ_F_TIMEOUT = BIT(REQ_F_TIMEOUT_BIT), | |
575 | /* regular file */ | |
576 | REQ_F_ISREG = BIT(REQ_F_ISREG_BIT), | |
577 | /* must be punted even for NONBLOCK */ | |
578 | REQ_F_MUST_PUNT = BIT(REQ_F_MUST_PUNT_BIT), | |
579 | /* no timeout sequence */ | |
580 | REQ_F_TIMEOUT_NOSEQ = BIT(REQ_F_TIMEOUT_NOSEQ_BIT), | |
581 | /* completion under lock */ | |
582 | REQ_F_COMP_LOCKED = BIT(REQ_F_COMP_LOCKED_BIT), | |
99bc4c38 PB |
583 | /* needs cleanup */ |
584 | REQ_F_NEED_CLEANUP = BIT(REQ_F_NEED_CLEANUP_BIT), | |
2ca10259 JA |
585 | /* in overflow list */ |
586 | REQ_F_OVERFLOW = BIT(REQ_F_OVERFLOW_BIT), | |
d7718a9d JA |
587 | /* already went through poll handler */ |
588 | REQ_F_POLLED = BIT(REQ_F_POLLED_BIT), | |
bcda7baa JA |
589 | /* buffer already selected */ |
590 | REQ_F_BUFFER_SELECTED = BIT(REQ_F_BUFFER_SELECTED_BIT), | |
5b0bbee4 JA |
591 | /* doesn't need file table for this request */ |
592 | REQ_F_NO_FILE_TABLE = BIT(REQ_F_NO_FILE_TABLE_BIT), | |
d7718a9d JA |
593 | }; |
594 | ||
595 | struct async_poll { | |
596 | struct io_poll_iocb poll; | |
597 | struct io_wq_work work; | |
6b47ee6e PB |
598 | }; |
599 | ||
09bb8394 JA |
600 | /* |
601 | * NOTE! Each of the iocb union members has the file pointer | |
602 | * as the first entry in their struct definition. So you can | |
603 | * access the file pointer through any of the sub-structs, | |
604 | * or directly as just 'ki_filp' in this struct. | |
605 | */ | |
2b188cc1 | 606 | struct io_kiocb { |
221c5eb2 | 607 | union { |
09bb8394 | 608 | struct file *file; |
9adbd45d | 609 | struct io_rw rw; |
221c5eb2 | 610 | struct io_poll_iocb poll; |
8ed8d3c3 JA |
611 | struct io_accept accept; |
612 | struct io_sync sync; | |
fbf23849 | 613 | struct io_cancel cancel; |
b29472ee | 614 | struct io_timeout timeout; |
3fbb51c1 | 615 | struct io_connect connect; |
e47293fd | 616 | struct io_sr_msg sr_msg; |
15b71abe | 617 | struct io_open open; |
b5dba59e | 618 | struct io_close close; |
05f3fb3c | 619 | struct io_files_update files_update; |
4840e418 | 620 | struct io_fadvise fadvise; |
c1ca757b | 621 | struct io_madvise madvise; |
3e4827b0 | 622 | struct io_epoll epoll; |
7d67af2c | 623 | struct io_splice splice; |
ddf0322d | 624 | struct io_provide_buf pbuf; |
221c5eb2 | 625 | }; |
2b188cc1 | 626 | |
1a6b74fc | 627 | struct io_async_ctx *io; |
c398ecb3 | 628 | int cflags; |
d625c6ee | 629 | u8 opcode; |
2b188cc1 JA |
630 | |
631 | struct io_ring_ctx *ctx; | |
d7718a9d | 632 | struct list_head list; |
2b188cc1 | 633 | unsigned int flags; |
c16361c1 | 634 | refcount_t refs; |
3537b6a7 JA |
635 | struct task_struct *task; |
636 | unsigned long fsize; | |
2b188cc1 | 637 | u64 user_data; |
9e645e11 | 638 | u32 result; |
de0617e4 | 639 | u32 sequence; |
2b188cc1 | 640 | |
d7718a9d JA |
641 | struct list_head link_list; |
642 | ||
fcb323cc JA |
643 | struct list_head inflight_entry; |
644 | ||
05589553 XW |
645 | struct percpu_ref *fixed_file_refs; |
646 | ||
b41e9852 JA |
647 | union { |
648 | /* | |
649 | * Only commands that never go async can use the below fields, | |
d7718a9d JA |
650 | * obviously. Right now only IORING_OP_POLL_ADD uses them, and |
651 | * async armed poll handlers for regular commands. The latter | |
652 | * restore the work, if needed. | |
b41e9852 JA |
653 | */ |
654 | struct { | |
b41e9852 | 655 | struct callback_head task_work; |
d7718a9d JA |
656 | struct hlist_node hash_node; |
657 | struct async_poll *apoll; | |
b41e9852 JA |
658 | }; |
659 | struct io_wq_work work; | |
660 | }; | |
2b188cc1 JA |
661 | }; |
662 | ||
663 | #define IO_PLUG_THRESHOLD 2 | |
def596e9 | 664 | #define IO_IOPOLL_BATCH 8 |
2b188cc1 | 665 | |
9a56a232 JA |
666 | struct io_submit_state { |
667 | struct blk_plug plug; | |
668 | ||
2579f913 JA |
669 | /* |
670 | * io_kiocb alloc cache | |
671 | */ | |
672 | void *reqs[IO_IOPOLL_BATCH]; | |
6c8a3134 | 673 | unsigned int free_reqs; |
2579f913 | 674 | |
9a56a232 JA |
675 | /* |
676 | * File reference cache | |
677 | */ | |
678 | struct file *file; | |
679 | unsigned int fd; | |
680 | unsigned int has_refs; | |
681 | unsigned int used_refs; | |
682 | unsigned int ios_left; | |
683 | }; | |
684 | ||
d3656344 JA |
685 | struct io_op_def { |
686 | /* needs req->io allocated for deferral/async */ | |
687 | unsigned async_ctx : 1; | |
688 | /* needs current->mm setup, does mm access */ | |
689 | unsigned needs_mm : 1; | |
690 | /* needs req->file assigned */ | |
691 | unsigned needs_file : 1; | |
d3656344 JA |
692 | /* hash wq insertion if file is a regular file */ |
693 | unsigned hash_reg_file : 1; | |
694 | /* unbound wq insertion if file is a non-regular file */ | |
695 | unsigned unbound_nonreg_file : 1; | |
66f4af93 JA |
696 | /* opcode is not supported by this kernel */ |
697 | unsigned not_supported : 1; | |
f86cd20c JA |
698 | /* needs file table */ |
699 | unsigned file_table : 1; | |
ff002b30 JA |
700 | /* needs ->fs */ |
701 | unsigned needs_fs : 1; | |
8a72758c JA |
702 | /* set if opcode supports polled "wait" */ |
703 | unsigned pollin : 1; | |
704 | unsigned pollout : 1; | |
bcda7baa JA |
705 | /* op supports buffer selection */ |
706 | unsigned buffer_select : 1; | |
d3656344 JA |
707 | }; |
708 | ||
709 | static const struct io_op_def io_op_defs[] = { | |
0463b6c5 PB |
710 | [IORING_OP_NOP] = {}, |
711 | [IORING_OP_READV] = { | |
d3656344 JA |
712 | .async_ctx = 1, |
713 | .needs_mm = 1, | |
714 | .needs_file = 1, | |
715 | .unbound_nonreg_file = 1, | |
8a72758c | 716 | .pollin = 1, |
4d954c25 | 717 | .buffer_select = 1, |
d3656344 | 718 | }, |
0463b6c5 | 719 | [IORING_OP_WRITEV] = { |
d3656344 JA |
720 | .async_ctx = 1, |
721 | .needs_mm = 1, | |
722 | .needs_file = 1, | |
723 | .hash_reg_file = 1, | |
724 | .unbound_nonreg_file = 1, | |
8a72758c | 725 | .pollout = 1, |
d3656344 | 726 | }, |
0463b6c5 | 727 | [IORING_OP_FSYNC] = { |
d3656344 JA |
728 | .needs_file = 1, |
729 | }, | |
0463b6c5 | 730 | [IORING_OP_READ_FIXED] = { |
d3656344 JA |
731 | .needs_file = 1, |
732 | .unbound_nonreg_file = 1, | |
8a72758c | 733 | .pollin = 1, |
d3656344 | 734 | }, |
0463b6c5 | 735 | [IORING_OP_WRITE_FIXED] = { |
d3656344 JA |
736 | .needs_file = 1, |
737 | .hash_reg_file = 1, | |
738 | .unbound_nonreg_file = 1, | |
8a72758c | 739 | .pollout = 1, |
d3656344 | 740 | }, |
0463b6c5 | 741 | [IORING_OP_POLL_ADD] = { |
d3656344 JA |
742 | .needs_file = 1, |
743 | .unbound_nonreg_file = 1, | |
744 | }, | |
0463b6c5 PB |
745 | [IORING_OP_POLL_REMOVE] = {}, |
746 | [IORING_OP_SYNC_FILE_RANGE] = { | |
d3656344 JA |
747 | .needs_file = 1, |
748 | }, | |
0463b6c5 | 749 | [IORING_OP_SENDMSG] = { |
d3656344 JA |
750 | .async_ctx = 1, |
751 | .needs_mm = 1, | |
752 | .needs_file = 1, | |
753 | .unbound_nonreg_file = 1, | |
ff002b30 | 754 | .needs_fs = 1, |
8a72758c | 755 | .pollout = 1, |
d3656344 | 756 | }, |
0463b6c5 | 757 | [IORING_OP_RECVMSG] = { |
d3656344 JA |
758 | .async_ctx = 1, |
759 | .needs_mm = 1, | |
760 | .needs_file = 1, | |
761 | .unbound_nonreg_file = 1, | |
ff002b30 | 762 | .needs_fs = 1, |
8a72758c | 763 | .pollin = 1, |
52de1fe1 | 764 | .buffer_select = 1, |
d3656344 | 765 | }, |
0463b6c5 | 766 | [IORING_OP_TIMEOUT] = { |
d3656344 JA |
767 | .async_ctx = 1, |
768 | .needs_mm = 1, | |
769 | }, | |
0463b6c5 PB |
770 | [IORING_OP_TIMEOUT_REMOVE] = {}, |
771 | [IORING_OP_ACCEPT] = { | |
d3656344 JA |
772 | .needs_mm = 1, |
773 | .needs_file = 1, | |
774 | .unbound_nonreg_file = 1, | |
f86cd20c | 775 | .file_table = 1, |
8a72758c | 776 | .pollin = 1, |
d3656344 | 777 | }, |
0463b6c5 PB |
778 | [IORING_OP_ASYNC_CANCEL] = {}, |
779 | [IORING_OP_LINK_TIMEOUT] = { | |
d3656344 JA |
780 | .async_ctx = 1, |
781 | .needs_mm = 1, | |
782 | }, | |
0463b6c5 | 783 | [IORING_OP_CONNECT] = { |
d3656344 JA |
784 | .async_ctx = 1, |
785 | .needs_mm = 1, | |
786 | .needs_file = 1, | |
787 | .unbound_nonreg_file = 1, | |
8a72758c | 788 | .pollout = 1, |
d3656344 | 789 | }, |
0463b6c5 | 790 | [IORING_OP_FALLOCATE] = { |
d3656344 JA |
791 | .needs_file = 1, |
792 | }, | |
0463b6c5 | 793 | [IORING_OP_OPENAT] = { |
f86cd20c | 794 | .file_table = 1, |
ff002b30 | 795 | .needs_fs = 1, |
d3656344 | 796 | }, |
0463b6c5 | 797 | [IORING_OP_CLOSE] = { |
f86cd20c | 798 | .file_table = 1, |
d3656344 | 799 | }, |
0463b6c5 | 800 | [IORING_OP_FILES_UPDATE] = { |
d3656344 | 801 | .needs_mm = 1, |
f86cd20c | 802 | .file_table = 1, |
d3656344 | 803 | }, |
0463b6c5 | 804 | [IORING_OP_STATX] = { |
d3656344 | 805 | .needs_mm = 1, |
ff002b30 | 806 | .needs_fs = 1, |
5b0bbee4 | 807 | .file_table = 1, |
d3656344 | 808 | }, |
0463b6c5 | 809 | [IORING_OP_READ] = { |
3a6820f2 JA |
810 | .needs_mm = 1, |
811 | .needs_file = 1, | |
812 | .unbound_nonreg_file = 1, | |
8a72758c | 813 | .pollin = 1, |
bcda7baa | 814 | .buffer_select = 1, |
3a6820f2 | 815 | }, |
0463b6c5 | 816 | [IORING_OP_WRITE] = { |
3a6820f2 JA |
817 | .needs_mm = 1, |
818 | .needs_file = 1, | |
819 | .unbound_nonreg_file = 1, | |
8a72758c | 820 | .pollout = 1, |
3a6820f2 | 821 | }, |
0463b6c5 | 822 | [IORING_OP_FADVISE] = { |
4840e418 JA |
823 | .needs_file = 1, |
824 | }, | |
0463b6c5 | 825 | [IORING_OP_MADVISE] = { |
c1ca757b JA |
826 | .needs_mm = 1, |
827 | }, | |
0463b6c5 | 828 | [IORING_OP_SEND] = { |
fddaface JA |
829 | .needs_mm = 1, |
830 | .needs_file = 1, | |
831 | .unbound_nonreg_file = 1, | |
8a72758c | 832 | .pollout = 1, |
fddaface | 833 | }, |
0463b6c5 | 834 | [IORING_OP_RECV] = { |
fddaface JA |
835 | .needs_mm = 1, |
836 | .needs_file = 1, | |
837 | .unbound_nonreg_file = 1, | |
8a72758c | 838 | .pollin = 1, |
bcda7baa | 839 | .buffer_select = 1, |
fddaface | 840 | }, |
0463b6c5 | 841 | [IORING_OP_OPENAT2] = { |
f86cd20c | 842 | .file_table = 1, |
ff002b30 | 843 | .needs_fs = 1, |
cebdb986 | 844 | }, |
3e4827b0 JA |
845 | [IORING_OP_EPOLL_CTL] = { |
846 | .unbound_nonreg_file = 1, | |
847 | .file_table = 1, | |
848 | }, | |
7d67af2c PB |
849 | [IORING_OP_SPLICE] = { |
850 | .needs_file = 1, | |
851 | .hash_reg_file = 1, | |
852 | .unbound_nonreg_file = 1, | |
ddf0322d JA |
853 | }, |
854 | [IORING_OP_PROVIDE_BUFFERS] = {}, | |
067524e9 | 855 | [IORING_OP_REMOVE_BUFFERS] = {}, |
f2a8d5c7 PB |
856 | [IORING_OP_TEE] = { |
857 | .needs_file = 1, | |
858 | .hash_reg_file = 1, | |
859 | .unbound_nonreg_file = 1, | |
860 | }, | |
d3656344 JA |
861 | }; |
862 | ||
561fb04a | 863 | static void io_wq_submit_work(struct io_wq_work **workptr); |
78e19bbe | 864 | static void io_cqring_fill_event(struct io_kiocb *req, long res); |
ec9c02ad | 865 | static void io_put_req(struct io_kiocb *req); |
978db57e | 866 | static void __io_double_put_req(struct io_kiocb *req); |
94ae5e77 JA |
867 | static struct io_kiocb *io_prep_linked_timeout(struct io_kiocb *req); |
868 | static void io_queue_linked_timeout(struct io_kiocb *req); | |
05f3fb3c JA |
869 | static int __io_sqe_files_update(struct io_ring_ctx *ctx, |
870 | struct io_uring_files_update *ip, | |
871 | unsigned nr_args); | |
f86cd20c | 872 | static int io_grab_files(struct io_kiocb *req); |
99bc4c38 | 873 | static void io_cleanup_req(struct io_kiocb *req); |
b41e9852 JA |
874 | static int io_file_get(struct io_submit_state *state, struct io_kiocb *req, |
875 | int fd, struct file **out_file, bool fixed); | |
876 | static void __io_queue_sqe(struct io_kiocb *req, | |
877 | const struct io_uring_sqe *sqe); | |
de0617e4 | 878 | |
2b188cc1 JA |
879 | static struct kmem_cache *req_cachep; |
880 | ||
881 | static const struct file_operations io_uring_fops; | |
882 | ||
883 | struct sock *io_uring_get_socket(struct file *file) | |
884 | { | |
885 | #if defined(CONFIG_UNIX) | |
886 | if (file->f_op == &io_uring_fops) { | |
887 | struct io_ring_ctx *ctx = file->private_data; | |
888 | ||
889 | return ctx->ring_sock->sk; | |
890 | } | |
891 | #endif | |
892 | return NULL; | |
893 | } | |
894 | EXPORT_SYMBOL(io_uring_get_socket); | |
895 | ||
4a38aed2 JA |
896 | static void io_file_put_work(struct work_struct *work); |
897 | ||
0cdaf760 PB |
898 | static inline bool io_async_submit(struct io_ring_ctx *ctx) |
899 | { | |
900 | return ctx->flags & IORING_SETUP_SQPOLL; | |
901 | } | |
902 | ||
2b188cc1 JA |
903 | static void io_ring_ctx_ref_free(struct percpu_ref *ref) |
904 | { | |
905 | struct io_ring_ctx *ctx = container_of(ref, struct io_ring_ctx, refs); | |
906 | ||
0f158b4c | 907 | complete(&ctx->ref_comp); |
2b188cc1 JA |
908 | } |
909 | ||
910 | static struct io_ring_ctx *io_ring_ctx_alloc(struct io_uring_params *p) | |
911 | { | |
912 | struct io_ring_ctx *ctx; | |
78076bb6 | 913 | int hash_bits; |
2b188cc1 JA |
914 | |
915 | ctx = kzalloc(sizeof(*ctx), GFP_KERNEL); | |
916 | if (!ctx) | |
917 | return NULL; | |
918 | ||
0ddf92e8 JA |
919 | ctx->fallback_req = kmem_cache_alloc(req_cachep, GFP_KERNEL); |
920 | if (!ctx->fallback_req) | |
921 | goto err; | |
922 | ||
78076bb6 JA |
923 | /* |
924 | * Use 5 bits less than the max cq entries, that should give us around | |
925 | * 32 entries per hash list if totally full and uniformly spread. | |
926 | */ | |
927 | hash_bits = ilog2(p->cq_entries); | |
928 | hash_bits -= 5; | |
929 | if (hash_bits <= 0) | |
930 | hash_bits = 1; | |
931 | ctx->cancel_hash_bits = hash_bits; | |
932 | ctx->cancel_hash = kmalloc((1U << hash_bits) * sizeof(struct hlist_head), | |
933 | GFP_KERNEL); | |
934 | if (!ctx->cancel_hash) | |
935 | goto err; | |
936 | __hash_init(ctx->cancel_hash, 1U << hash_bits); | |
937 | ||
21482896 | 938 | if (percpu_ref_init(&ctx->refs, io_ring_ctx_ref_free, |
206aefde JA |
939 | PERCPU_REF_ALLOW_REINIT, GFP_KERNEL)) |
940 | goto err; | |
2b188cc1 JA |
941 | |
942 | ctx->flags = p->flags; | |
943 | init_waitqueue_head(&ctx->cq_wait); | |
1d7bb1d5 | 944 | INIT_LIST_HEAD(&ctx->cq_overflow_list); |
0f158b4c JA |
945 | init_completion(&ctx->ref_comp); |
946 | init_completion(&ctx->sq_thread_comp); | |
5a2e745d | 947 | idr_init(&ctx->io_buffer_idr); |
071698e1 | 948 | idr_init(&ctx->personality_idr); |
2b188cc1 JA |
949 | mutex_init(&ctx->uring_lock); |
950 | init_waitqueue_head(&ctx->wait); | |
951 | spin_lock_init(&ctx->completion_lock); | |
def596e9 | 952 | INIT_LIST_HEAD(&ctx->poll_list); |
de0617e4 | 953 | INIT_LIST_HEAD(&ctx->defer_list); |
5262f567 | 954 | INIT_LIST_HEAD(&ctx->timeout_list); |
fcb323cc JA |
955 | init_waitqueue_head(&ctx->inflight_wait); |
956 | spin_lock_init(&ctx->inflight_lock); | |
957 | INIT_LIST_HEAD(&ctx->inflight_list); | |
4a38aed2 JA |
958 | INIT_DELAYED_WORK(&ctx->file_put_work, io_file_put_work); |
959 | init_llist_head(&ctx->file_put_llist); | |
2b188cc1 | 960 | return ctx; |
206aefde | 961 | err: |
0ddf92e8 JA |
962 | if (ctx->fallback_req) |
963 | kmem_cache_free(req_cachep, ctx->fallback_req); | |
78076bb6 | 964 | kfree(ctx->cancel_hash); |
206aefde JA |
965 | kfree(ctx); |
966 | return NULL; | |
2b188cc1 JA |
967 | } |
968 | ||
9d858b21 | 969 | static inline bool __req_need_defer(struct io_kiocb *req) |
7adf4eaf | 970 | { |
a197f664 JL |
971 | struct io_ring_ctx *ctx = req->ctx; |
972 | ||
31af27c7 PB |
973 | return req->sequence != ctx->cached_cq_tail |
974 | + atomic_read(&ctx->cached_cq_overflow); | |
7adf4eaf JA |
975 | } |
976 | ||
9d858b21 | 977 | static inline bool req_need_defer(struct io_kiocb *req) |
de0617e4 | 978 | { |
87987898 | 979 | if (unlikely(req->flags & REQ_F_IO_DRAIN)) |
9d858b21 | 980 | return __req_need_defer(req); |
de0617e4 | 981 | |
9d858b21 | 982 | return false; |
de0617e4 JA |
983 | } |
984 | ||
7adf4eaf | 985 | static struct io_kiocb *io_get_deferred_req(struct io_ring_ctx *ctx) |
de0617e4 JA |
986 | { |
987 | struct io_kiocb *req; | |
988 | ||
7adf4eaf | 989 | req = list_first_entry_or_null(&ctx->defer_list, struct io_kiocb, list); |
9d858b21 | 990 | if (req && !req_need_defer(req)) { |
de0617e4 JA |
991 | list_del_init(&req->list); |
992 | return req; | |
993 | } | |
994 | ||
995 | return NULL; | |
996 | } | |
997 | ||
5262f567 JA |
998 | static struct io_kiocb *io_get_timeout_req(struct io_ring_ctx *ctx) |
999 | { | |
7adf4eaf JA |
1000 | struct io_kiocb *req; |
1001 | ||
1002 | req = list_first_entry_or_null(&ctx->timeout_list, struct io_kiocb, list); | |
93bd25bb JA |
1003 | if (req) { |
1004 | if (req->flags & REQ_F_TIMEOUT_NOSEQ) | |
1005 | return NULL; | |
fb4b3d3f | 1006 | if (!__req_need_defer(req)) { |
93bd25bb JA |
1007 | list_del_init(&req->list); |
1008 | return req; | |
1009 | } | |
7adf4eaf JA |
1010 | } |
1011 | ||
1012 | return NULL; | |
5262f567 JA |
1013 | } |
1014 | ||
de0617e4 | 1015 | static void __io_commit_cqring(struct io_ring_ctx *ctx) |
2b188cc1 | 1016 | { |
75b28aff | 1017 | struct io_rings *rings = ctx->rings; |
2b188cc1 | 1018 | |
07910158 PB |
1019 | /* order cqe stores with ring update */ |
1020 | smp_store_release(&rings->cq.tail, ctx->cached_cq_tail); | |
2b188cc1 | 1021 | |
07910158 PB |
1022 | if (wq_has_sleeper(&ctx->cq_wait)) { |
1023 | wake_up_interruptible(&ctx->cq_wait); | |
1024 | kill_fasync(&ctx->cq_fasync, SIGIO, POLL_IN); | |
2b188cc1 JA |
1025 | } |
1026 | } | |
1027 | ||
cccf0ee8 JA |
1028 | static inline void io_req_work_grab_env(struct io_kiocb *req, |
1029 | const struct io_op_def *def) | |
1030 | { | |
1031 | if (!req->work.mm && def->needs_mm) { | |
1032 | mmgrab(current->mm); | |
1033 | req->work.mm = current->mm; | |
2b188cc1 | 1034 | } |
cccf0ee8 JA |
1035 | if (!req->work.creds) |
1036 | req->work.creds = get_current_cred(); | |
ff002b30 JA |
1037 | if (!req->work.fs && def->needs_fs) { |
1038 | spin_lock(¤t->fs->lock); | |
1039 | if (!current->fs->in_exec) { | |
1040 | req->work.fs = current->fs; | |
1041 | req->work.fs->users++; | |
1042 | } else { | |
1043 | req->work.flags |= IO_WQ_WORK_CANCEL; | |
1044 | } | |
1045 | spin_unlock(¤t->fs->lock); | |
1046 | } | |
6ab23144 JA |
1047 | if (!req->work.task_pid) |
1048 | req->work.task_pid = task_pid_vnr(current); | |
2b188cc1 JA |
1049 | } |
1050 | ||
cccf0ee8 | 1051 | static inline void io_req_work_drop_env(struct io_kiocb *req) |
18d9be1a | 1052 | { |
cccf0ee8 JA |
1053 | if (req->work.mm) { |
1054 | mmdrop(req->work.mm); | |
1055 | req->work.mm = NULL; | |
1056 | } | |
1057 | if (req->work.creds) { | |
1058 | put_cred(req->work.creds); | |
1059 | req->work.creds = NULL; | |
1060 | } | |
ff002b30 JA |
1061 | if (req->work.fs) { |
1062 | struct fs_struct *fs = req->work.fs; | |
1063 | ||
1064 | spin_lock(&req->work.fs->lock); | |
1065 | if (--fs->users) | |
1066 | fs = NULL; | |
1067 | spin_unlock(&req->work.fs->lock); | |
1068 | if (fs) | |
1069 | free_fs_struct(fs); | |
1070 | } | |
561fb04a JA |
1071 | } |
1072 | ||
8766dd51 | 1073 | static inline void io_prep_async_work(struct io_kiocb *req, |
94ae5e77 | 1074 | struct io_kiocb **link) |
18d9be1a | 1075 | { |
d3656344 | 1076 | const struct io_op_def *def = &io_op_defs[req->opcode]; |
54a91f3b | 1077 | |
d3656344 JA |
1078 | if (req->flags & REQ_F_ISREG) { |
1079 | if (def->hash_reg_file) | |
8766dd51 | 1080 | io_wq_hash_work(&req->work, file_inode(req->file)); |
d3656344 JA |
1081 | } else { |
1082 | if (def->unbound_nonreg_file) | |
3529d8c2 | 1083 | req->work.flags |= IO_WQ_WORK_UNBOUND; |
54a91f3b | 1084 | } |
cccf0ee8 JA |
1085 | |
1086 | io_req_work_grab_env(req, def); | |
54a91f3b | 1087 | |
94ae5e77 | 1088 | *link = io_prep_linked_timeout(req); |
561fb04a JA |
1089 | } |
1090 | ||
a197f664 | 1091 | static inline void io_queue_async_work(struct io_kiocb *req) |
561fb04a | 1092 | { |
a197f664 | 1093 | struct io_ring_ctx *ctx = req->ctx; |
94ae5e77 | 1094 | struct io_kiocb *link; |
94ae5e77 | 1095 | |
8766dd51 | 1096 | io_prep_async_work(req, &link); |
561fb04a | 1097 | |
8766dd51 PB |
1098 | trace_io_uring_queue_async_work(ctx, io_wq_is_hashed(&req->work), req, |
1099 | &req->work, req->flags); | |
1100 | io_wq_enqueue(ctx->io_wq, &req->work); | |
94ae5e77 JA |
1101 | |
1102 | if (link) | |
1103 | io_queue_linked_timeout(link); | |
18d9be1a JA |
1104 | } |
1105 | ||
5262f567 JA |
1106 | static void io_kill_timeout(struct io_kiocb *req) |
1107 | { | |
1108 | int ret; | |
1109 | ||
2d28390a | 1110 | ret = hrtimer_try_to_cancel(&req->io->timeout.timer); |
5262f567 JA |
1111 | if (ret != -1) { |
1112 | atomic_inc(&req->ctx->cq_timeouts); | |
842f9612 | 1113 | list_del_init(&req->list); |
f0e20b89 | 1114 | req->flags |= REQ_F_COMP_LOCKED; |
78e19bbe | 1115 | io_cqring_fill_event(req, 0); |
ec9c02ad | 1116 | io_put_req(req); |
5262f567 JA |
1117 | } |
1118 | } | |
1119 | ||
1120 | static void io_kill_timeouts(struct io_ring_ctx *ctx) | |
1121 | { | |
1122 | struct io_kiocb *req, *tmp; | |
1123 | ||
1124 | spin_lock_irq(&ctx->completion_lock); | |
1125 | list_for_each_entry_safe(req, tmp, &ctx->timeout_list, list) | |
1126 | io_kill_timeout(req); | |
1127 | spin_unlock_irq(&ctx->completion_lock); | |
1128 | } | |
1129 | ||
de0617e4 JA |
1130 | static void io_commit_cqring(struct io_ring_ctx *ctx) |
1131 | { | |
1132 | struct io_kiocb *req; | |
1133 | ||
5262f567 JA |
1134 | while ((req = io_get_timeout_req(ctx)) != NULL) |
1135 | io_kill_timeout(req); | |
1136 | ||
de0617e4 JA |
1137 | __io_commit_cqring(ctx); |
1138 | ||
87987898 | 1139 | while ((req = io_get_deferred_req(ctx)) != NULL) |
a197f664 | 1140 | io_queue_async_work(req); |
de0617e4 JA |
1141 | } |
1142 | ||
2b188cc1 JA |
1143 | static struct io_uring_cqe *io_get_cqring(struct io_ring_ctx *ctx) |
1144 | { | |
75b28aff | 1145 | struct io_rings *rings = ctx->rings; |
2b188cc1 JA |
1146 | unsigned tail; |
1147 | ||
1148 | tail = ctx->cached_cq_tail; | |
115e12e5 SB |
1149 | /* |
1150 | * writes to the cq entry need to come after reading head; the | |
1151 | * control dependency is enough as we're using WRITE_ONCE to | |
1152 | * fill the cq entry | |
1153 | */ | |
75b28aff | 1154 | if (tail - READ_ONCE(rings->cq.head) == rings->cq_ring_entries) |
2b188cc1 JA |
1155 | return NULL; |
1156 | ||
1157 | ctx->cached_cq_tail++; | |
75b28aff | 1158 | return &rings->cqes[tail & ctx->cq_mask]; |
2b188cc1 JA |
1159 | } |
1160 | ||
f2842ab5 JA |
1161 | static inline bool io_should_trigger_evfd(struct io_ring_ctx *ctx) |
1162 | { | |
f0b493e6 JA |
1163 | if (!ctx->cq_ev_fd) |
1164 | return false; | |
7e55a19c SG |
1165 | if (READ_ONCE(ctx->rings->cq_flags) & IORING_CQ_EVENTFD_DISABLED) |
1166 | return false; | |
f2842ab5 JA |
1167 | if (!ctx->eventfd_async) |
1168 | return true; | |
b41e9852 | 1169 | return io_wq_current_is_worker(); |
f2842ab5 JA |
1170 | } |
1171 | ||
b41e9852 | 1172 | static void io_cqring_ev_posted(struct io_ring_ctx *ctx) |
1d7bb1d5 JA |
1173 | { |
1174 | if (waitqueue_active(&ctx->wait)) | |
1175 | wake_up(&ctx->wait); | |
1176 | if (waitqueue_active(&ctx->sqo_wait)) | |
1177 | wake_up(&ctx->sqo_wait); | |
b41e9852 | 1178 | if (io_should_trigger_evfd(ctx)) |
1d7bb1d5 JA |
1179 | eventfd_signal(ctx->cq_ev_fd, 1); |
1180 | } | |
1181 | ||
c4a2ed72 JA |
1182 | /* Returns true if there are no backlogged entries after the flush */ |
1183 | static bool io_cqring_overflow_flush(struct io_ring_ctx *ctx, bool force) | |
1d7bb1d5 JA |
1184 | { |
1185 | struct io_rings *rings = ctx->rings; | |
1186 | struct io_uring_cqe *cqe; | |
1187 | struct io_kiocb *req; | |
1188 | unsigned long flags; | |
1189 | LIST_HEAD(list); | |
1190 | ||
1191 | if (!force) { | |
1192 | if (list_empty_careful(&ctx->cq_overflow_list)) | |
c4a2ed72 | 1193 | return true; |
1d7bb1d5 JA |
1194 | if ((ctx->cached_cq_tail - READ_ONCE(rings->cq.head) == |
1195 | rings->cq_ring_entries)) | |
c4a2ed72 | 1196 | return false; |
1d7bb1d5 JA |
1197 | } |
1198 | ||
1199 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
1200 | ||
1201 | /* if force is set, the ring is going away. always drop after that */ | |
1202 | if (force) | |
69b3e546 | 1203 | ctx->cq_overflow_flushed = 1; |
1d7bb1d5 | 1204 | |
c4a2ed72 | 1205 | cqe = NULL; |
1d7bb1d5 JA |
1206 | while (!list_empty(&ctx->cq_overflow_list)) { |
1207 | cqe = io_get_cqring(ctx); | |
1208 | if (!cqe && !force) | |
1209 | break; | |
1210 | ||
1211 | req = list_first_entry(&ctx->cq_overflow_list, struct io_kiocb, | |
1212 | list); | |
1213 | list_move(&req->list, &list); | |
2ca10259 | 1214 | req->flags &= ~REQ_F_OVERFLOW; |
1d7bb1d5 JA |
1215 | if (cqe) { |
1216 | WRITE_ONCE(cqe->user_data, req->user_data); | |
1217 | WRITE_ONCE(cqe->res, req->result); | |
bcda7baa | 1218 | WRITE_ONCE(cqe->flags, req->cflags); |
1d7bb1d5 JA |
1219 | } else { |
1220 | WRITE_ONCE(ctx->rings->cq_overflow, | |
1221 | atomic_inc_return(&ctx->cached_cq_overflow)); | |
1222 | } | |
1223 | } | |
1224 | ||
1225 | io_commit_cqring(ctx); | |
ad3eb2c8 JA |
1226 | if (cqe) { |
1227 | clear_bit(0, &ctx->sq_check_overflow); | |
1228 | clear_bit(0, &ctx->cq_check_overflow); | |
1229 | } | |
1d7bb1d5 JA |
1230 | spin_unlock_irqrestore(&ctx->completion_lock, flags); |
1231 | io_cqring_ev_posted(ctx); | |
1232 | ||
1233 | while (!list_empty(&list)) { | |
1234 | req = list_first_entry(&list, struct io_kiocb, list); | |
1235 | list_del(&req->list); | |
ec9c02ad | 1236 | io_put_req(req); |
1d7bb1d5 | 1237 | } |
c4a2ed72 JA |
1238 | |
1239 | return cqe != NULL; | |
1d7bb1d5 JA |
1240 | } |
1241 | ||
bcda7baa | 1242 | static void __io_cqring_fill_event(struct io_kiocb *req, long res, long cflags) |
2b188cc1 | 1243 | { |
78e19bbe | 1244 | struct io_ring_ctx *ctx = req->ctx; |
2b188cc1 JA |
1245 | struct io_uring_cqe *cqe; |
1246 | ||
78e19bbe | 1247 | trace_io_uring_complete(ctx, req->user_data, res); |
51c3ff62 | 1248 | |
2b188cc1 JA |
1249 | /* |
1250 | * If we can't get a cq entry, userspace overflowed the | |
1251 | * submission (by quite a lot). Increment the overflow count in | |
1252 | * the ring. | |
1253 | */ | |
1254 | cqe = io_get_cqring(ctx); | |
1d7bb1d5 | 1255 | if (likely(cqe)) { |
78e19bbe | 1256 | WRITE_ONCE(cqe->user_data, req->user_data); |
2b188cc1 | 1257 | WRITE_ONCE(cqe->res, res); |
bcda7baa | 1258 | WRITE_ONCE(cqe->flags, cflags); |
1d7bb1d5 | 1259 | } else if (ctx->cq_overflow_flushed) { |
498ccd9e JA |
1260 | WRITE_ONCE(ctx->rings->cq_overflow, |
1261 | atomic_inc_return(&ctx->cached_cq_overflow)); | |
1d7bb1d5 | 1262 | } else { |
ad3eb2c8 JA |
1263 | if (list_empty(&ctx->cq_overflow_list)) { |
1264 | set_bit(0, &ctx->sq_check_overflow); | |
1265 | set_bit(0, &ctx->cq_check_overflow); | |
1266 | } | |
2ca10259 | 1267 | req->flags |= REQ_F_OVERFLOW; |
1d7bb1d5 JA |
1268 | refcount_inc(&req->refs); |
1269 | req->result = res; | |
bcda7baa | 1270 | req->cflags = cflags; |
1d7bb1d5 | 1271 | list_add_tail(&req->list, &ctx->cq_overflow_list); |
2b188cc1 JA |
1272 | } |
1273 | } | |
1274 | ||
bcda7baa JA |
1275 | static void io_cqring_fill_event(struct io_kiocb *req, long res) |
1276 | { | |
1277 | __io_cqring_fill_event(req, res, 0); | |
1278 | } | |
1279 | ||
1280 | static void __io_cqring_add_event(struct io_kiocb *req, long res, long cflags) | |
2b188cc1 | 1281 | { |
78e19bbe | 1282 | struct io_ring_ctx *ctx = req->ctx; |
2b188cc1 JA |
1283 | unsigned long flags; |
1284 | ||
1285 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
bcda7baa | 1286 | __io_cqring_fill_event(req, res, cflags); |
2b188cc1 JA |
1287 | io_commit_cqring(ctx); |
1288 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
1289 | ||
8c838788 | 1290 | io_cqring_ev_posted(ctx); |
2b188cc1 JA |
1291 | } |
1292 | ||
bcda7baa JA |
1293 | static void io_cqring_add_event(struct io_kiocb *req, long res) |
1294 | { | |
1295 | __io_cqring_add_event(req, res, 0); | |
1296 | } | |
1297 | ||
0ddf92e8 JA |
1298 | static inline bool io_is_fallback_req(struct io_kiocb *req) |
1299 | { | |
1300 | return req == (struct io_kiocb *) | |
1301 | ((unsigned long) req->ctx->fallback_req & ~1UL); | |
1302 | } | |
1303 | ||
1304 | static struct io_kiocb *io_get_fallback_req(struct io_ring_ctx *ctx) | |
1305 | { | |
1306 | struct io_kiocb *req; | |
1307 | ||
1308 | req = ctx->fallback_req; | |
dd461af6 | 1309 | if (!test_and_set_bit_lock(0, (unsigned long *) &ctx->fallback_req)) |
0ddf92e8 JA |
1310 | return req; |
1311 | ||
1312 | return NULL; | |
1313 | } | |
1314 | ||
0553b8bd PB |
1315 | static struct io_kiocb *io_alloc_req(struct io_ring_ctx *ctx, |
1316 | struct io_submit_state *state) | |
2b188cc1 | 1317 | { |
fd6fab2c | 1318 | gfp_t gfp = GFP_KERNEL | __GFP_NOWARN; |
2b188cc1 JA |
1319 | struct io_kiocb *req; |
1320 | ||
2579f913 | 1321 | if (!state) { |
fd6fab2c | 1322 | req = kmem_cache_alloc(req_cachep, gfp); |
2579f913 | 1323 | if (unlikely(!req)) |
0ddf92e8 | 1324 | goto fallback; |
2579f913 JA |
1325 | } else if (!state->free_reqs) { |
1326 | size_t sz; | |
1327 | int ret; | |
1328 | ||
1329 | sz = min_t(size_t, state->ios_left, ARRAY_SIZE(state->reqs)); | |
fd6fab2c JA |
1330 | ret = kmem_cache_alloc_bulk(req_cachep, gfp, sz, state->reqs); |
1331 | ||
1332 | /* | |
1333 | * Bulk alloc is all-or-nothing. If we fail to get a batch, | |
1334 | * retry single alloc to be on the safe side. | |
1335 | */ | |
1336 | if (unlikely(ret <= 0)) { | |
1337 | state->reqs[0] = kmem_cache_alloc(req_cachep, gfp); | |
1338 | if (!state->reqs[0]) | |
0ddf92e8 | 1339 | goto fallback; |
fd6fab2c JA |
1340 | ret = 1; |
1341 | } | |
2579f913 | 1342 | state->free_reqs = ret - 1; |
6c8a3134 | 1343 | req = state->reqs[ret - 1]; |
2579f913 | 1344 | } else { |
2579f913 | 1345 | state->free_reqs--; |
6c8a3134 | 1346 | req = state->reqs[state->free_reqs]; |
2b188cc1 JA |
1347 | } |
1348 | ||
2579f913 | 1349 | return req; |
0ddf92e8 | 1350 | fallback: |
0553b8bd | 1351 | return io_get_fallback_req(ctx); |
2b188cc1 JA |
1352 | } |
1353 | ||
8da11c19 PB |
1354 | static inline void io_put_file(struct io_kiocb *req, struct file *file, |
1355 | bool fixed) | |
1356 | { | |
1357 | if (fixed) | |
05589553 | 1358 | percpu_ref_put(req->fixed_file_refs); |
8da11c19 PB |
1359 | else |
1360 | fput(file); | |
1361 | } | |
1362 | ||
c6ca97b3 | 1363 | static void __io_req_aux_free(struct io_kiocb *req) |
2b188cc1 | 1364 | { |
929a3af9 PB |
1365 | if (req->flags & REQ_F_NEED_CLEANUP) |
1366 | io_cleanup_req(req); | |
1367 | ||
96fd84d8 | 1368 | kfree(req->io); |
8da11c19 PB |
1369 | if (req->file) |
1370 | io_put_file(req, req->file, (req->flags & REQ_F_FIXED_FILE)); | |
3537b6a7 JA |
1371 | if (req->task) |
1372 | put_task_struct(req->task); | |
cccf0ee8 JA |
1373 | |
1374 | io_req_work_drop_env(req); | |
def596e9 JA |
1375 | } |
1376 | ||
9e645e11 | 1377 | static void __io_free_req(struct io_kiocb *req) |
2b188cc1 | 1378 | { |
c6ca97b3 | 1379 | __io_req_aux_free(req); |
fcb323cc | 1380 | |
fcb323cc | 1381 | if (req->flags & REQ_F_INFLIGHT) { |
c6ca97b3 | 1382 | struct io_ring_ctx *ctx = req->ctx; |
fcb323cc JA |
1383 | unsigned long flags; |
1384 | ||
1385 | spin_lock_irqsave(&ctx->inflight_lock, flags); | |
1386 | list_del(&req->inflight_entry); | |
1387 | if (waitqueue_active(&ctx->inflight_wait)) | |
1388 | wake_up(&ctx->inflight_wait); | |
1389 | spin_unlock_irqrestore(&ctx->inflight_lock, flags); | |
1390 | } | |
2b85edfc PB |
1391 | |
1392 | percpu_ref_put(&req->ctx->refs); | |
b1e50e54 PB |
1393 | if (likely(!io_is_fallback_req(req))) |
1394 | kmem_cache_free(req_cachep, req); | |
1395 | else | |
dd461af6 | 1396 | clear_bit_unlock(0, (unsigned long *) &req->ctx->fallback_req); |
e65ef56d JA |
1397 | } |
1398 | ||
c6ca97b3 JA |
1399 | struct req_batch { |
1400 | void *reqs[IO_IOPOLL_BATCH]; | |
1401 | int to_free; | |
1402 | int need_iter; | |
1403 | }; | |
1404 | ||
1405 | static void io_free_req_many(struct io_ring_ctx *ctx, struct req_batch *rb) | |
1406 | { | |
1407 | if (!rb->to_free) | |
1408 | return; | |
1409 | if (rb->need_iter) { | |
1410 | int i, inflight = 0; | |
1411 | unsigned long flags; | |
1412 | ||
1413 | for (i = 0; i < rb->to_free; i++) { | |
1414 | struct io_kiocb *req = rb->reqs[i]; | |
1415 | ||
10fef4be | 1416 | if (req->flags & REQ_F_FIXED_FILE) { |
c6ca97b3 | 1417 | req->file = NULL; |
05589553 | 1418 | percpu_ref_put(req->fixed_file_refs); |
10fef4be | 1419 | } |
c6ca97b3 JA |
1420 | if (req->flags & REQ_F_INFLIGHT) |
1421 | inflight++; | |
c6ca97b3 JA |
1422 | __io_req_aux_free(req); |
1423 | } | |
1424 | if (!inflight) | |
1425 | goto do_free; | |
1426 | ||
1427 | spin_lock_irqsave(&ctx->inflight_lock, flags); | |
1428 | for (i = 0; i < rb->to_free; i++) { | |
1429 | struct io_kiocb *req = rb->reqs[i]; | |
1430 | ||
10fef4be | 1431 | if (req->flags & REQ_F_INFLIGHT) { |
c6ca97b3 JA |
1432 | list_del(&req->inflight_entry); |
1433 | if (!--inflight) | |
1434 | break; | |
1435 | } | |
1436 | } | |
1437 | spin_unlock_irqrestore(&ctx->inflight_lock, flags); | |
1438 | ||
1439 | if (waitqueue_active(&ctx->inflight_wait)) | |
1440 | wake_up(&ctx->inflight_wait); | |
1441 | } | |
1442 | do_free: | |
1443 | kmem_cache_free_bulk(req_cachep, rb->to_free, rb->reqs); | |
1444 | percpu_ref_put_many(&ctx->refs, rb->to_free); | |
c6ca97b3 | 1445 | rb->to_free = rb->need_iter = 0; |
e65ef56d JA |
1446 | } |
1447 | ||
a197f664 | 1448 | static bool io_link_cancel_timeout(struct io_kiocb *req) |
2665abfd | 1449 | { |
a197f664 | 1450 | struct io_ring_ctx *ctx = req->ctx; |
2665abfd JA |
1451 | int ret; |
1452 | ||
2d28390a | 1453 | ret = hrtimer_try_to_cancel(&req->io->timeout.timer); |
2665abfd | 1454 | if (ret != -1) { |
78e19bbe | 1455 | io_cqring_fill_event(req, -ECANCELED); |
2665abfd | 1456 | io_commit_cqring(ctx); |
dea3b49c | 1457 | req->flags &= ~REQ_F_LINK_HEAD; |
ec9c02ad | 1458 | io_put_req(req); |
2665abfd JA |
1459 | return true; |
1460 | } | |
1461 | ||
1462 | return false; | |
e65ef56d JA |
1463 | } |
1464 | ||
ba816ad6 | 1465 | static void io_req_link_next(struct io_kiocb *req, struct io_kiocb **nxtptr) |
9e645e11 | 1466 | { |
2665abfd | 1467 | struct io_ring_ctx *ctx = req->ctx; |
2665abfd | 1468 | bool wake_ev = false; |
9e645e11 | 1469 | |
4d7dd462 JA |
1470 | /* Already got next link */ |
1471 | if (req->flags & REQ_F_LINK_NEXT) | |
1472 | return; | |
1473 | ||
9e645e11 JA |
1474 | /* |
1475 | * The list should never be empty when we are called here. But could | |
1476 | * potentially happen if the chain is messed up, check to be on the | |
1477 | * safe side. | |
1478 | */ | |
4493233e PB |
1479 | while (!list_empty(&req->link_list)) { |
1480 | struct io_kiocb *nxt = list_first_entry(&req->link_list, | |
1481 | struct io_kiocb, link_list); | |
94ae5e77 | 1482 | |
4493233e PB |
1483 | if (unlikely((req->flags & REQ_F_LINK_TIMEOUT) && |
1484 | (nxt->flags & REQ_F_TIMEOUT))) { | |
1485 | list_del_init(&nxt->link_list); | |
94ae5e77 | 1486 | wake_ev |= io_link_cancel_timeout(nxt); |
94ae5e77 JA |
1487 | req->flags &= ~REQ_F_LINK_TIMEOUT; |
1488 | continue; | |
1489 | } | |
9e645e11 | 1490 | |
4493233e PB |
1491 | list_del_init(&req->link_list); |
1492 | if (!list_empty(&nxt->link_list)) | |
dea3b49c | 1493 | nxt->flags |= REQ_F_LINK_HEAD; |
b18fdf71 | 1494 | *nxtptr = nxt; |
94ae5e77 | 1495 | break; |
9e645e11 | 1496 | } |
2665abfd | 1497 | |
4d7dd462 | 1498 | req->flags |= REQ_F_LINK_NEXT; |
2665abfd JA |
1499 | if (wake_ev) |
1500 | io_cqring_ev_posted(ctx); | |
9e645e11 JA |
1501 | } |
1502 | ||
1503 | /* | |
dea3b49c | 1504 | * Called if REQ_F_LINK_HEAD is set, and we fail the head request |
9e645e11 JA |
1505 | */ |
1506 | static void io_fail_links(struct io_kiocb *req) | |
1507 | { | |
2665abfd | 1508 | struct io_ring_ctx *ctx = req->ctx; |
2665abfd JA |
1509 | unsigned long flags; |
1510 | ||
1511 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
9e645e11 JA |
1512 | |
1513 | while (!list_empty(&req->link_list)) { | |
4493233e PB |
1514 | struct io_kiocb *link = list_first_entry(&req->link_list, |
1515 | struct io_kiocb, link_list); | |
9e645e11 | 1516 | |
4493233e | 1517 | list_del_init(&link->link_list); |
c826bd7a | 1518 | trace_io_uring_fail_link(req, link); |
2665abfd JA |
1519 | |
1520 | if ((req->flags & REQ_F_LINK_TIMEOUT) && | |
d625c6ee | 1521 | link->opcode == IORING_OP_LINK_TIMEOUT) { |
a197f664 | 1522 | io_link_cancel_timeout(link); |
2665abfd | 1523 | } else { |
78e19bbe | 1524 | io_cqring_fill_event(link, -ECANCELED); |
978db57e | 1525 | __io_double_put_req(link); |
2665abfd | 1526 | } |
5d960724 | 1527 | req->flags &= ~REQ_F_LINK_TIMEOUT; |
9e645e11 | 1528 | } |
2665abfd JA |
1529 | |
1530 | io_commit_cqring(ctx); | |
1531 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
1532 | io_cqring_ev_posted(ctx); | |
9e645e11 JA |
1533 | } |
1534 | ||
4d7dd462 | 1535 | static void io_req_find_next(struct io_kiocb *req, struct io_kiocb **nxt) |
9e645e11 | 1536 | { |
dea3b49c | 1537 | if (likely(!(req->flags & REQ_F_LINK_HEAD))) |
2665abfd | 1538 | return; |
2665abfd | 1539 | |
9e645e11 JA |
1540 | /* |
1541 | * If LINK is set, we have dependent requests in this chain. If we | |
1542 | * didn't fail this request, queue the first one up, moving any other | |
1543 | * dependencies to the next request. In case of failure, fail the rest | |
1544 | * of the chain. | |
1545 | */ | |
2665abfd JA |
1546 | if (req->flags & REQ_F_FAIL_LINK) { |
1547 | io_fail_links(req); | |
7c9e7f0f JA |
1548 | } else if ((req->flags & (REQ_F_LINK_TIMEOUT | REQ_F_COMP_LOCKED)) == |
1549 | REQ_F_LINK_TIMEOUT) { | |
2665abfd JA |
1550 | struct io_ring_ctx *ctx = req->ctx; |
1551 | unsigned long flags; | |
1552 | ||
1553 | /* | |
1554 | * If this is a timeout link, we could be racing with the | |
1555 | * timeout timer. Grab the completion lock for this case to | |
7c9e7f0f | 1556 | * protect against that. |
2665abfd JA |
1557 | */ |
1558 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
1559 | io_req_link_next(req, nxt); | |
1560 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
1561 | } else { | |
1562 | io_req_link_next(req, nxt); | |
9e645e11 | 1563 | } |
4d7dd462 | 1564 | } |
9e645e11 | 1565 | |
c69f8dbe JL |
1566 | static void io_free_req(struct io_kiocb *req) |
1567 | { | |
944e58bf PB |
1568 | struct io_kiocb *nxt = NULL; |
1569 | ||
1570 | io_req_find_next(req, &nxt); | |
70cf9f32 | 1571 | __io_free_req(req); |
944e58bf PB |
1572 | |
1573 | if (nxt) | |
1574 | io_queue_async_work(nxt); | |
c69f8dbe JL |
1575 | } |
1576 | ||
7a743e22 PB |
1577 | static void io_link_work_cb(struct io_wq_work **workptr) |
1578 | { | |
18a542ff PB |
1579 | struct io_kiocb *req = container_of(*workptr, struct io_kiocb, work); |
1580 | struct io_kiocb *link; | |
7a743e22 | 1581 | |
18a542ff | 1582 | link = list_first_entry(&req->link_list, struct io_kiocb, link_list); |
7a743e22 PB |
1583 | io_queue_linked_timeout(link); |
1584 | io_wq_submit_work(workptr); | |
1585 | } | |
1586 | ||
1587 | static void io_wq_assign_next(struct io_wq_work **workptr, struct io_kiocb *nxt) | |
1588 | { | |
1589 | struct io_kiocb *link; | |
8766dd51 PB |
1590 | const struct io_op_def *def = &io_op_defs[nxt->opcode]; |
1591 | ||
1592 | if ((nxt->flags & REQ_F_ISREG) && def->hash_reg_file) | |
1593 | io_wq_hash_work(&nxt->work, file_inode(nxt->file)); | |
7a743e22 PB |
1594 | |
1595 | *workptr = &nxt->work; | |
1596 | link = io_prep_linked_timeout(nxt); | |
18a542ff | 1597 | if (link) |
7a743e22 | 1598 | nxt->work.func = io_link_work_cb; |
7a743e22 PB |
1599 | } |
1600 | ||
ba816ad6 JA |
1601 | /* |
1602 | * Drop reference to request, return next in chain (if there is one) if this | |
1603 | * was the last reference to this request. | |
1604 | */ | |
f9bd67f6 | 1605 | __attribute__((nonnull)) |
ec9c02ad | 1606 | static void io_put_req_find_next(struct io_kiocb *req, struct io_kiocb **nxtptr) |
e65ef56d | 1607 | { |
2a44f467 JA |
1608 | if (refcount_dec_and_test(&req->refs)) { |
1609 | io_req_find_next(req, nxtptr); | |
4d7dd462 | 1610 | __io_free_req(req); |
2a44f467 | 1611 | } |
2b188cc1 JA |
1612 | } |
1613 | ||
e65ef56d JA |
1614 | static void io_put_req(struct io_kiocb *req) |
1615 | { | |
1616 | if (refcount_dec_and_test(&req->refs)) | |
1617 | io_free_req(req); | |
2b188cc1 JA |
1618 | } |
1619 | ||
e9fd9396 PB |
1620 | static void io_steal_work(struct io_kiocb *req, |
1621 | struct io_wq_work **workptr) | |
7a743e22 PB |
1622 | { |
1623 | /* | |
1624 | * It's in an io-wq worker, so there always should be at least | |
1625 | * one reference, which will be dropped in io_put_work() just | |
1626 | * after the current handler returns. | |
1627 | * | |
1628 | * It also means, that if the counter dropped to 1, then there is | |
1629 | * no asynchronous users left, so it's safe to steal the next work. | |
1630 | */ | |
7a743e22 PB |
1631 | if (refcount_read(&req->refs) == 1) { |
1632 | struct io_kiocb *nxt = NULL; | |
1633 | ||
1634 | io_req_find_next(req, &nxt); | |
1635 | if (nxt) | |
1636 | io_wq_assign_next(workptr, nxt); | |
1637 | } | |
1638 | } | |
1639 | ||
978db57e JA |
1640 | /* |
1641 | * Must only be used if we don't need to care about links, usually from | |
1642 | * within the completion handling itself. | |
1643 | */ | |
1644 | static void __io_double_put_req(struct io_kiocb *req) | |
78e19bbe JA |
1645 | { |
1646 | /* drop both submit and complete references */ | |
1647 | if (refcount_sub_and_test(2, &req->refs)) | |
1648 | __io_free_req(req); | |
1649 | } | |
1650 | ||
978db57e JA |
1651 | static void io_double_put_req(struct io_kiocb *req) |
1652 | { | |
1653 | /* drop both submit and complete references */ | |
1654 | if (refcount_sub_and_test(2, &req->refs)) | |
1655 | io_free_req(req); | |
1656 | } | |
1657 | ||
1d7bb1d5 | 1658 | static unsigned io_cqring_events(struct io_ring_ctx *ctx, bool noflush) |
a3a0e43f | 1659 | { |
84f97dc2 JA |
1660 | struct io_rings *rings = ctx->rings; |
1661 | ||
ad3eb2c8 JA |
1662 | if (test_bit(0, &ctx->cq_check_overflow)) { |
1663 | /* | |
1664 | * noflush == true is from the waitqueue handler, just ensure | |
1665 | * we wake up the task, and the next invocation will flush the | |
1666 | * entries. We cannot safely to it from here. | |
1667 | */ | |
1668 | if (noflush && !list_empty(&ctx->cq_overflow_list)) | |
1669 | return -1U; | |
1d7bb1d5 | 1670 | |
ad3eb2c8 JA |
1671 | io_cqring_overflow_flush(ctx, false); |
1672 | } | |
1d7bb1d5 | 1673 | |
a3a0e43f JA |
1674 | /* See comment at the top of this file */ |
1675 | smp_rmb(); | |
ad3eb2c8 | 1676 | return ctx->cached_cq_tail - READ_ONCE(rings->cq.head); |
a3a0e43f JA |
1677 | } |
1678 | ||
fb5ccc98 PB |
1679 | static inline unsigned int io_sqring_entries(struct io_ring_ctx *ctx) |
1680 | { | |
1681 | struct io_rings *rings = ctx->rings; | |
1682 | ||
1683 | /* make sure SQ entry isn't read before tail */ | |
1684 | return smp_load_acquire(&rings->sq.tail) - ctx->cached_sq_head; | |
1685 | } | |
1686 | ||
8237e045 | 1687 | static inline bool io_req_multi_free(struct req_batch *rb, struct io_kiocb *req) |
e94f141b | 1688 | { |
dea3b49c | 1689 | if ((req->flags & REQ_F_LINK_HEAD) || io_is_fallback_req(req)) |
c6ca97b3 | 1690 | return false; |
e94f141b | 1691 | |
c6ca97b3 JA |
1692 | if (!(req->flags & REQ_F_FIXED_FILE) || req->io) |
1693 | rb->need_iter++; | |
1694 | ||
1695 | rb->reqs[rb->to_free++] = req; | |
1696 | if (unlikely(rb->to_free == ARRAY_SIZE(rb->reqs))) | |
1697 | io_free_req_many(req->ctx, rb); | |
1698 | return true; | |
e94f141b JA |
1699 | } |
1700 | ||
bcda7baa JA |
1701 | static int io_put_kbuf(struct io_kiocb *req) |
1702 | { | |
4d954c25 | 1703 | struct io_buffer *kbuf; |
bcda7baa JA |
1704 | int cflags; |
1705 | ||
4d954c25 | 1706 | kbuf = (struct io_buffer *) (unsigned long) req->rw.addr; |
bcda7baa JA |
1707 | cflags = kbuf->bid << IORING_CQE_BUFFER_SHIFT; |
1708 | cflags |= IORING_CQE_F_BUFFER; | |
1709 | req->rw.addr = 0; | |
1710 | kfree(kbuf); | |
1711 | return cflags; | |
1712 | } | |
1713 | ||
def596e9 JA |
1714 | /* |
1715 | * Find and free completed poll iocbs | |
1716 | */ | |
1717 | static void io_iopoll_complete(struct io_ring_ctx *ctx, unsigned int *nr_events, | |
1718 | struct list_head *done) | |
1719 | { | |
8237e045 | 1720 | struct req_batch rb; |
def596e9 | 1721 | struct io_kiocb *req; |
def596e9 | 1722 | |
c6ca97b3 | 1723 | rb.to_free = rb.need_iter = 0; |
def596e9 | 1724 | while (!list_empty(done)) { |
bcda7baa JA |
1725 | int cflags = 0; |
1726 | ||
def596e9 JA |
1727 | req = list_first_entry(done, struct io_kiocb, list); |
1728 | list_del(&req->list); | |
1729 | ||
bcda7baa JA |
1730 | if (req->flags & REQ_F_BUFFER_SELECTED) |
1731 | cflags = io_put_kbuf(req); | |
1732 | ||
1733 | __io_cqring_fill_event(req, req->result, cflags); | |
def596e9 JA |
1734 | (*nr_events)++; |
1735 | ||
8237e045 JA |
1736 | if (refcount_dec_and_test(&req->refs) && |
1737 | !io_req_multi_free(&rb, req)) | |
1738 | io_free_req(req); | |
def596e9 | 1739 | } |
def596e9 | 1740 | |
09bb8394 | 1741 | io_commit_cqring(ctx); |
32b2244a XW |
1742 | if (ctx->flags & IORING_SETUP_SQPOLL) |
1743 | io_cqring_ev_posted(ctx); | |
8237e045 | 1744 | io_free_req_many(ctx, &rb); |
def596e9 JA |
1745 | } |
1746 | ||
581f9810 BM |
1747 | static void io_iopoll_queue(struct list_head *again) |
1748 | { | |
1749 | struct io_kiocb *req; | |
1750 | ||
1751 | do { | |
1752 | req = list_first_entry(again, struct io_kiocb, list); | |
1753 | list_del(&req->list); | |
1754 | refcount_inc(&req->refs); | |
1755 | io_queue_async_work(req); | |
1756 | } while (!list_empty(again)); | |
1757 | } | |
1758 | ||
def596e9 JA |
1759 | static int io_do_iopoll(struct io_ring_ctx *ctx, unsigned int *nr_events, |
1760 | long min) | |
1761 | { | |
1762 | struct io_kiocb *req, *tmp; | |
1763 | LIST_HEAD(done); | |
581f9810 | 1764 | LIST_HEAD(again); |
def596e9 JA |
1765 | bool spin; |
1766 | int ret; | |
1767 | ||
1768 | /* | |
1769 | * Only spin for completions if we don't have multiple devices hanging | |
1770 | * off our complete list, and we're under the requested amount. | |
1771 | */ | |
1772 | spin = !ctx->poll_multi_file && *nr_events < min; | |
1773 | ||
1774 | ret = 0; | |
1775 | list_for_each_entry_safe(req, tmp, &ctx->poll_list, list) { | |
9adbd45d | 1776 | struct kiocb *kiocb = &req->rw.kiocb; |
def596e9 JA |
1777 | |
1778 | /* | |
581f9810 BM |
1779 | * Move completed and retryable entries to our local lists. |
1780 | * If we find a request that requires polling, break out | |
1781 | * and complete those lists first, if we have entries there. | |
def596e9 JA |
1782 | */ |
1783 | if (req->flags & REQ_F_IOPOLL_COMPLETED) { | |
1784 | list_move_tail(&req->list, &done); | |
1785 | continue; | |
1786 | } | |
1787 | if (!list_empty(&done)) | |
1788 | break; | |
1789 | ||
581f9810 BM |
1790 | if (req->result == -EAGAIN) { |
1791 | list_move_tail(&req->list, &again); | |
1792 | continue; | |
1793 | } | |
1794 | if (!list_empty(&again)) | |
1795 | break; | |
1796 | ||
def596e9 JA |
1797 | ret = kiocb->ki_filp->f_op->iopoll(kiocb, spin); |
1798 | if (ret < 0) | |
1799 | break; | |
1800 | ||
1801 | if (ret && spin) | |
1802 | spin = false; | |
1803 | ret = 0; | |
1804 | } | |
1805 | ||
1806 | if (!list_empty(&done)) | |
1807 | io_iopoll_complete(ctx, nr_events, &done); | |
1808 | ||
581f9810 BM |
1809 | if (!list_empty(&again)) |
1810 | io_iopoll_queue(&again); | |
1811 | ||
def596e9 JA |
1812 | return ret; |
1813 | } | |
1814 | ||
1815 | /* | |
d195a66e | 1816 | * Poll for a minimum of 'min' events. Note that if min == 0 we consider that a |
def596e9 JA |
1817 | * non-spinning poll check - we'll still enter the driver poll loop, but only |
1818 | * as a non-spinning completion check. | |
1819 | */ | |
1820 | static int io_iopoll_getevents(struct io_ring_ctx *ctx, unsigned int *nr_events, | |
1821 | long min) | |
1822 | { | |
08f5439f | 1823 | while (!list_empty(&ctx->poll_list) && !need_resched()) { |
def596e9 JA |
1824 | int ret; |
1825 | ||
1826 | ret = io_do_iopoll(ctx, nr_events, min); | |
1827 | if (ret < 0) | |
1828 | return ret; | |
1829 | if (!min || *nr_events >= min) | |
1830 | return 0; | |
1831 | } | |
1832 | ||
1833 | return 1; | |
1834 | } | |
1835 | ||
1836 | /* | |
1837 | * We can't just wait for polled events to come to us, we have to actively | |
1838 | * find and complete them. | |
1839 | */ | |
1840 | static void io_iopoll_reap_events(struct io_ring_ctx *ctx) | |
1841 | { | |
1842 | if (!(ctx->flags & IORING_SETUP_IOPOLL)) | |
1843 | return; | |
1844 | ||
1845 | mutex_lock(&ctx->uring_lock); | |
1846 | while (!list_empty(&ctx->poll_list)) { | |
1847 | unsigned int nr_events = 0; | |
1848 | ||
1849 | io_iopoll_getevents(ctx, &nr_events, 1); | |
08f5439f JA |
1850 | |
1851 | /* | |
1852 | * Ensure we allow local-to-the-cpu processing to take place, | |
1853 | * in this case we need to ensure that we reap all events. | |
1854 | */ | |
1855 | cond_resched(); | |
def596e9 JA |
1856 | } |
1857 | mutex_unlock(&ctx->uring_lock); | |
1858 | } | |
1859 | ||
c7849be9 XW |
1860 | static int io_iopoll_check(struct io_ring_ctx *ctx, unsigned *nr_events, |
1861 | long min) | |
def596e9 | 1862 | { |
2b2ed975 | 1863 | int iters = 0, ret = 0; |
500f9fba | 1864 | |
c7849be9 XW |
1865 | /* |
1866 | * We disallow the app entering submit/complete with polling, but we | |
1867 | * still need to lock the ring to prevent racing with polled issue | |
1868 | * that got punted to a workqueue. | |
1869 | */ | |
1870 | mutex_lock(&ctx->uring_lock); | |
def596e9 JA |
1871 | do { |
1872 | int tmin = 0; | |
1873 | ||
a3a0e43f JA |
1874 | /* |
1875 | * Don't enter poll loop if we already have events pending. | |
1876 | * If we do, we can potentially be spinning for commands that | |
1877 | * already triggered a CQE (eg in error). | |
1878 | */ | |
1d7bb1d5 | 1879 | if (io_cqring_events(ctx, false)) |
a3a0e43f JA |
1880 | break; |
1881 | ||
500f9fba JA |
1882 | /* |
1883 | * If a submit got punted to a workqueue, we can have the | |
1884 | * application entering polling for a command before it gets | |
1885 | * issued. That app will hold the uring_lock for the duration | |
1886 | * of the poll right here, so we need to take a breather every | |
1887 | * now and then to ensure that the issue has a chance to add | |
1888 | * the poll to the issued list. Otherwise we can spin here | |
1889 | * forever, while the workqueue is stuck trying to acquire the | |
1890 | * very same mutex. | |
1891 | */ | |
1892 | if (!(++iters & 7)) { | |
1893 | mutex_unlock(&ctx->uring_lock); | |
1894 | mutex_lock(&ctx->uring_lock); | |
1895 | } | |
1896 | ||
def596e9 JA |
1897 | if (*nr_events < min) |
1898 | tmin = min - *nr_events; | |
1899 | ||
1900 | ret = io_iopoll_getevents(ctx, nr_events, tmin); | |
1901 | if (ret <= 0) | |
1902 | break; | |
1903 | ret = 0; | |
1904 | } while (min && !*nr_events && !need_resched()); | |
1905 | ||
500f9fba | 1906 | mutex_unlock(&ctx->uring_lock); |
def596e9 JA |
1907 | return ret; |
1908 | } | |
1909 | ||
491381ce | 1910 | static void kiocb_end_write(struct io_kiocb *req) |
2b188cc1 | 1911 | { |
491381ce JA |
1912 | /* |
1913 | * Tell lockdep we inherited freeze protection from submission | |
1914 | * thread. | |
1915 | */ | |
1916 | if (req->flags & REQ_F_ISREG) { | |
1917 | struct inode *inode = file_inode(req->file); | |
2b188cc1 | 1918 | |
491381ce | 1919 | __sb_writers_acquired(inode->i_sb, SB_FREEZE_WRITE); |
2b188cc1 | 1920 | } |
491381ce | 1921 | file_end_write(req->file); |
2b188cc1 JA |
1922 | } |
1923 | ||
4e88d6e7 JA |
1924 | static inline void req_set_fail_links(struct io_kiocb *req) |
1925 | { | |
1926 | if ((req->flags & (REQ_F_LINK | REQ_F_HARDLINK)) == REQ_F_LINK) | |
1927 | req->flags |= REQ_F_FAIL_LINK; | |
1928 | } | |
1929 | ||
ba816ad6 | 1930 | static void io_complete_rw_common(struct kiocb *kiocb, long res) |
2b188cc1 | 1931 | { |
9adbd45d | 1932 | struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb); |
bcda7baa | 1933 | int cflags = 0; |
2b188cc1 | 1934 | |
491381ce JA |
1935 | if (kiocb->ki_flags & IOCB_WRITE) |
1936 | kiocb_end_write(req); | |
2b188cc1 | 1937 | |
4e88d6e7 JA |
1938 | if (res != req->result) |
1939 | req_set_fail_links(req); | |
bcda7baa JA |
1940 | if (req->flags & REQ_F_BUFFER_SELECTED) |
1941 | cflags = io_put_kbuf(req); | |
1942 | __io_cqring_add_event(req, res, cflags); | |
ba816ad6 JA |
1943 | } |
1944 | ||
1945 | static void io_complete_rw(struct kiocb *kiocb, long res, long res2) | |
1946 | { | |
9adbd45d | 1947 | struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb); |
ba816ad6 JA |
1948 | |
1949 | io_complete_rw_common(kiocb, res); | |
e65ef56d | 1950 | io_put_req(req); |
2b188cc1 JA |
1951 | } |
1952 | ||
def596e9 JA |
1953 | static void io_complete_rw_iopoll(struct kiocb *kiocb, long res, long res2) |
1954 | { | |
9adbd45d | 1955 | struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb); |
def596e9 | 1956 | |
491381ce JA |
1957 | if (kiocb->ki_flags & IOCB_WRITE) |
1958 | kiocb_end_write(req); | |
def596e9 | 1959 | |
4e88d6e7 JA |
1960 | if (res != req->result) |
1961 | req_set_fail_links(req); | |
9e645e11 | 1962 | req->result = res; |
def596e9 JA |
1963 | if (res != -EAGAIN) |
1964 | req->flags |= REQ_F_IOPOLL_COMPLETED; | |
1965 | } | |
1966 | ||
1967 | /* | |
1968 | * After the iocb has been issued, it's safe to be found on the poll list. | |
1969 | * Adding the kiocb to the list AFTER submission ensures that we don't | |
1970 | * find it from a io_iopoll_getevents() thread before the issuer is done | |
1971 | * accessing the kiocb cookie. | |
1972 | */ | |
1973 | static void io_iopoll_req_issued(struct io_kiocb *req) | |
1974 | { | |
1975 | struct io_ring_ctx *ctx = req->ctx; | |
1976 | ||
1977 | /* | |
1978 | * Track whether we have multiple files in our lists. This will impact | |
1979 | * how we do polling eventually, not spinning if we're on potentially | |
1980 | * different devices. | |
1981 | */ | |
1982 | if (list_empty(&ctx->poll_list)) { | |
1983 | ctx->poll_multi_file = false; | |
1984 | } else if (!ctx->poll_multi_file) { | |
1985 | struct io_kiocb *list_req; | |
1986 | ||
1987 | list_req = list_first_entry(&ctx->poll_list, struct io_kiocb, | |
1988 | list); | |
9adbd45d | 1989 | if (list_req->file != req->file) |
def596e9 JA |
1990 | ctx->poll_multi_file = true; |
1991 | } | |
1992 | ||
1993 | /* | |
1994 | * For fast devices, IO may have already completed. If it has, add | |
1995 | * it to the front so we find it first. | |
1996 | */ | |
1997 | if (req->flags & REQ_F_IOPOLL_COMPLETED) | |
1998 | list_add(&req->list, &ctx->poll_list); | |
1999 | else | |
2000 | list_add_tail(&req->list, &ctx->poll_list); | |
bdcd3eab XW |
2001 | |
2002 | if ((ctx->flags & IORING_SETUP_SQPOLL) && | |
2003 | wq_has_sleeper(&ctx->sqo_wait)) | |
2004 | wake_up(&ctx->sqo_wait); | |
def596e9 JA |
2005 | } |
2006 | ||
9f13c35b | 2007 | static void __io_state_file_put(struct io_submit_state *state) |
9a56a232 | 2008 | { |
9f13c35b | 2009 | int diff = state->has_refs - state->used_refs; |
9a56a232 | 2010 | |
9f13c35b PB |
2011 | if (diff) |
2012 | fput_many(state->file, diff); | |
2013 | state->file = NULL; | |
2014 | } | |
2015 | ||
2016 | static inline void io_state_file_put(struct io_submit_state *state) | |
2017 | { | |
2018 | if (state->file) | |
2019 | __io_state_file_put(state); | |
9a56a232 JA |
2020 | } |
2021 | ||
2022 | /* | |
2023 | * Get as many references to a file as we have IOs left in this submission, | |
2024 | * assuming most submissions are for one file, or at least that each file | |
2025 | * has more than one submission. | |
2026 | */ | |
8da11c19 | 2027 | static struct file *__io_file_get(struct io_submit_state *state, int fd) |
9a56a232 JA |
2028 | { |
2029 | if (!state) | |
2030 | return fget(fd); | |
2031 | ||
2032 | if (state->file) { | |
2033 | if (state->fd == fd) { | |
2034 | state->used_refs++; | |
2035 | state->ios_left--; | |
2036 | return state->file; | |
2037 | } | |
9f13c35b | 2038 | __io_state_file_put(state); |
9a56a232 JA |
2039 | } |
2040 | state->file = fget_many(fd, state->ios_left); | |
2041 | if (!state->file) | |
2042 | return NULL; | |
2043 | ||
2044 | state->fd = fd; | |
2045 | state->has_refs = state->ios_left; | |
2046 | state->used_refs = 1; | |
2047 | state->ios_left--; | |
2048 | return state->file; | |
2049 | } | |
2050 | ||
2b188cc1 JA |
2051 | /* |
2052 | * If we tracked the file through the SCM inflight mechanism, we could support | |
2053 | * any file. For now, just ensure that anything potentially problematic is done | |
2054 | * inline. | |
2055 | */ | |
af197f50 | 2056 | static bool io_file_supports_async(struct file *file, int rw) |
2b188cc1 JA |
2057 | { |
2058 | umode_t mode = file_inode(file)->i_mode; | |
2059 | ||
10d59345 | 2060 | if (S_ISBLK(mode) || S_ISCHR(mode) || S_ISSOCK(mode)) |
2b188cc1 JA |
2061 | return true; |
2062 | if (S_ISREG(mode) && file->f_op != &io_uring_fops) | |
2063 | return true; | |
2064 | ||
af197f50 JA |
2065 | if (!(file->f_mode & FMODE_NOWAIT)) |
2066 | return false; | |
2067 | ||
2068 | if (rw == READ) | |
2069 | return file->f_op->read_iter != NULL; | |
2070 | ||
2071 | return file->f_op->write_iter != NULL; | |
2b188cc1 JA |
2072 | } |
2073 | ||
3529d8c2 JA |
2074 | static int io_prep_rw(struct io_kiocb *req, const struct io_uring_sqe *sqe, |
2075 | bool force_nonblock) | |
2b188cc1 | 2076 | { |
def596e9 | 2077 | struct io_ring_ctx *ctx = req->ctx; |
9adbd45d | 2078 | struct kiocb *kiocb = &req->rw.kiocb; |
09bb8394 JA |
2079 | unsigned ioprio; |
2080 | int ret; | |
2b188cc1 | 2081 | |
491381ce JA |
2082 | if (S_ISREG(file_inode(req->file)->i_mode)) |
2083 | req->flags |= REQ_F_ISREG; | |
2084 | ||
2b188cc1 | 2085 | kiocb->ki_pos = READ_ONCE(sqe->off); |
ba04291e JA |
2086 | if (kiocb->ki_pos == -1 && !(req->file->f_mode & FMODE_STREAM)) { |
2087 | req->flags |= REQ_F_CUR_POS; | |
2088 | kiocb->ki_pos = req->file->f_pos; | |
2089 | } | |
2b188cc1 | 2090 | kiocb->ki_hint = ki_hint_validate(file_write_hint(kiocb->ki_filp)); |
3e577dcd PB |
2091 | kiocb->ki_flags = iocb_flags(kiocb->ki_filp); |
2092 | ret = kiocb_set_rw_flags(kiocb, READ_ONCE(sqe->rw_flags)); | |
2093 | if (unlikely(ret)) | |
2094 | return ret; | |
2b188cc1 JA |
2095 | |
2096 | ioprio = READ_ONCE(sqe->ioprio); | |
2097 | if (ioprio) { | |
2098 | ret = ioprio_check_cap(ioprio); | |
2099 | if (ret) | |
09bb8394 | 2100 | return ret; |
2b188cc1 JA |
2101 | |
2102 | kiocb->ki_ioprio = ioprio; | |
2103 | } else | |
2104 | kiocb->ki_ioprio = get_current_ioprio(); | |
2105 | ||
8449eeda | 2106 | /* don't allow async punt if RWF_NOWAIT was requested */ |
491381ce JA |
2107 | if ((kiocb->ki_flags & IOCB_NOWAIT) || |
2108 | (req->file->f_flags & O_NONBLOCK)) | |
8449eeda SB |
2109 | req->flags |= REQ_F_NOWAIT; |
2110 | ||
2111 | if (force_nonblock) | |
2b188cc1 | 2112 | kiocb->ki_flags |= IOCB_NOWAIT; |
8449eeda | 2113 | |
def596e9 | 2114 | if (ctx->flags & IORING_SETUP_IOPOLL) { |
def596e9 JA |
2115 | if (!(kiocb->ki_flags & IOCB_DIRECT) || |
2116 | !kiocb->ki_filp->f_op->iopoll) | |
09bb8394 | 2117 | return -EOPNOTSUPP; |
2b188cc1 | 2118 | |
def596e9 JA |
2119 | kiocb->ki_flags |= IOCB_HIPRI; |
2120 | kiocb->ki_complete = io_complete_rw_iopoll; | |
6873e0bd | 2121 | req->result = 0; |
def596e9 | 2122 | } else { |
09bb8394 JA |
2123 | if (kiocb->ki_flags & IOCB_HIPRI) |
2124 | return -EINVAL; | |
def596e9 JA |
2125 | kiocb->ki_complete = io_complete_rw; |
2126 | } | |
9adbd45d | 2127 | |
3529d8c2 JA |
2128 | req->rw.addr = READ_ONCE(sqe->addr); |
2129 | req->rw.len = READ_ONCE(sqe->len); | |
bcda7baa | 2130 | /* we own ->private, reuse it for the buffer index / buffer ID */ |
9adbd45d | 2131 | req->rw.kiocb.private = (void *) (unsigned long) |
3529d8c2 | 2132 | READ_ONCE(sqe->buf_index); |
2b188cc1 | 2133 | return 0; |
2b188cc1 JA |
2134 | } |
2135 | ||
2136 | static inline void io_rw_done(struct kiocb *kiocb, ssize_t ret) | |
2137 | { | |
2138 | switch (ret) { | |
2139 | case -EIOCBQUEUED: | |
2140 | break; | |
2141 | case -ERESTARTSYS: | |
2142 | case -ERESTARTNOINTR: | |
2143 | case -ERESTARTNOHAND: | |
2144 | case -ERESTART_RESTARTBLOCK: | |
2145 | /* | |
2146 | * We can't just restart the syscall, since previously | |
2147 | * submitted sqes may already be in progress. Just fail this | |
2148 | * IO with EINTR. | |
2149 | */ | |
2150 | ret = -EINTR; | |
2151 | /* fall through */ | |
2152 | default: | |
2153 | kiocb->ki_complete(kiocb, ret, 0); | |
2154 | } | |
2155 | } | |
2156 | ||
014db007 | 2157 | static void kiocb_done(struct kiocb *kiocb, ssize_t ret) |
ba816ad6 | 2158 | { |
ba04291e JA |
2159 | struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb); |
2160 | ||
2161 | if (req->flags & REQ_F_CUR_POS) | |
2162 | req->file->f_pos = kiocb->ki_pos; | |
bcaec089 | 2163 | if (ret >= 0 && kiocb->ki_complete == io_complete_rw) |
014db007 | 2164 | io_complete_rw(kiocb, ret, 0); |
ba816ad6 JA |
2165 | else |
2166 | io_rw_done(kiocb, ret); | |
2167 | } | |
2168 | ||
9adbd45d | 2169 | static ssize_t io_import_fixed(struct io_kiocb *req, int rw, |
7d009165 | 2170 | struct iov_iter *iter) |
edafccee | 2171 | { |
9adbd45d JA |
2172 | struct io_ring_ctx *ctx = req->ctx; |
2173 | size_t len = req->rw.len; | |
edafccee JA |
2174 | struct io_mapped_ubuf *imu; |
2175 | unsigned index, buf_index; | |
2176 | size_t offset; | |
2177 | u64 buf_addr; | |
2178 | ||
2179 | /* attempt to use fixed buffers without having provided iovecs */ | |
2180 | if (unlikely(!ctx->user_bufs)) | |
2181 | return -EFAULT; | |
2182 | ||
9adbd45d | 2183 | buf_index = (unsigned long) req->rw.kiocb.private; |
edafccee JA |
2184 | if (unlikely(buf_index >= ctx->nr_user_bufs)) |
2185 | return -EFAULT; | |
2186 | ||
2187 | index = array_index_nospec(buf_index, ctx->nr_user_bufs); | |
2188 | imu = &ctx->user_bufs[index]; | |
9adbd45d | 2189 | buf_addr = req->rw.addr; |
edafccee JA |
2190 | |
2191 | /* overflow */ | |
2192 | if (buf_addr + len < buf_addr) | |
2193 | return -EFAULT; | |
2194 | /* not inside the mapped region */ | |
2195 | if (buf_addr < imu->ubuf || buf_addr + len > imu->ubuf + imu->len) | |
2196 | return -EFAULT; | |
2197 | ||
2198 | /* | |
2199 | * May not be a start of buffer, set size appropriately | |
2200 | * and advance us to the beginning. | |
2201 | */ | |
2202 | offset = buf_addr - imu->ubuf; | |
2203 | iov_iter_bvec(iter, rw, imu->bvec, imu->nr_bvecs, offset + len); | |
bd11b3a3 JA |
2204 | |
2205 | if (offset) { | |
2206 | /* | |
2207 | * Don't use iov_iter_advance() here, as it's really slow for | |
2208 | * using the latter parts of a big fixed buffer - it iterates | |
2209 | * over each segment manually. We can cheat a bit here, because | |
2210 | * we know that: | |
2211 | * | |
2212 | * 1) it's a BVEC iter, we set it up | |
2213 | * 2) all bvecs are PAGE_SIZE in size, except potentially the | |
2214 | * first and last bvec | |
2215 | * | |
2216 | * So just find our index, and adjust the iterator afterwards. | |
2217 | * If the offset is within the first bvec (or the whole first | |
2218 | * bvec, just use iov_iter_advance(). This makes it easier | |
2219 | * since we can just skip the first segment, which may not | |
2220 | * be PAGE_SIZE aligned. | |
2221 | */ | |
2222 | const struct bio_vec *bvec = imu->bvec; | |
2223 | ||
2224 | if (offset <= bvec->bv_len) { | |
2225 | iov_iter_advance(iter, offset); | |
2226 | } else { | |
2227 | unsigned long seg_skip; | |
2228 | ||
2229 | /* skip first vec */ | |
2230 | offset -= bvec->bv_len; | |
2231 | seg_skip = 1 + (offset >> PAGE_SHIFT); | |
2232 | ||
2233 | iter->bvec = bvec + seg_skip; | |
2234 | iter->nr_segs -= seg_skip; | |
99c79f66 | 2235 | iter->count -= bvec->bv_len + offset; |
bd11b3a3 | 2236 | iter->iov_offset = offset & ~PAGE_MASK; |
bd11b3a3 JA |
2237 | } |
2238 | } | |
2239 | ||
5e559561 | 2240 | return len; |
edafccee JA |
2241 | } |
2242 | ||
bcda7baa JA |
2243 | static void io_ring_submit_unlock(struct io_ring_ctx *ctx, bool needs_lock) |
2244 | { | |
2245 | if (needs_lock) | |
2246 | mutex_unlock(&ctx->uring_lock); | |
2247 | } | |
2248 | ||
2249 | static void io_ring_submit_lock(struct io_ring_ctx *ctx, bool needs_lock) | |
2250 | { | |
2251 | /* | |
2252 | * "Normal" inline submissions always hold the uring_lock, since we | |
2253 | * grab it from the system call. Same is true for the SQPOLL offload. | |
2254 | * The only exception is when we've detached the request and issue it | |
2255 | * from an async worker thread, grab the lock for that case. | |
2256 | */ | |
2257 | if (needs_lock) | |
2258 | mutex_lock(&ctx->uring_lock); | |
2259 | } | |
2260 | ||
2261 | static struct io_buffer *io_buffer_select(struct io_kiocb *req, size_t *len, | |
2262 | int bgid, struct io_buffer *kbuf, | |
2263 | bool needs_lock) | |
2264 | { | |
2265 | struct io_buffer *head; | |
2266 | ||
2267 | if (req->flags & REQ_F_BUFFER_SELECTED) | |
2268 | return kbuf; | |
2269 | ||
2270 | io_ring_submit_lock(req->ctx, needs_lock); | |
2271 | ||
2272 | lockdep_assert_held(&req->ctx->uring_lock); | |
2273 | ||
2274 | head = idr_find(&req->ctx->io_buffer_idr, bgid); | |
2275 | if (head) { | |
2276 | if (!list_empty(&head->list)) { | |
2277 | kbuf = list_last_entry(&head->list, struct io_buffer, | |
2278 | list); | |
2279 | list_del(&kbuf->list); | |
2280 | } else { | |
2281 | kbuf = head; | |
2282 | idr_remove(&req->ctx->io_buffer_idr, bgid); | |
2283 | } | |
2284 | if (*len > kbuf->len) | |
2285 | *len = kbuf->len; | |
2286 | } else { | |
2287 | kbuf = ERR_PTR(-ENOBUFS); | |
2288 | } | |
2289 | ||
2290 | io_ring_submit_unlock(req->ctx, needs_lock); | |
2291 | ||
2292 | return kbuf; | |
2293 | } | |
2294 | ||
4d954c25 JA |
2295 | static void __user *io_rw_buffer_select(struct io_kiocb *req, size_t *len, |
2296 | bool needs_lock) | |
2297 | { | |
2298 | struct io_buffer *kbuf; | |
2299 | int bgid; | |
2300 | ||
2301 | kbuf = (struct io_buffer *) (unsigned long) req->rw.addr; | |
2302 | bgid = (int) (unsigned long) req->rw.kiocb.private; | |
2303 | kbuf = io_buffer_select(req, len, bgid, kbuf, needs_lock); | |
2304 | if (IS_ERR(kbuf)) | |
2305 | return kbuf; | |
2306 | req->rw.addr = (u64) (unsigned long) kbuf; | |
2307 | req->flags |= REQ_F_BUFFER_SELECTED; | |
2308 | return u64_to_user_ptr(kbuf->addr); | |
2309 | } | |
2310 | ||
2311 | #ifdef CONFIG_COMPAT | |
2312 | static ssize_t io_compat_import(struct io_kiocb *req, struct iovec *iov, | |
2313 | bool needs_lock) | |
2314 | { | |
2315 | struct compat_iovec __user *uiov; | |
2316 | compat_ssize_t clen; | |
2317 | void __user *buf; | |
2318 | ssize_t len; | |
2319 | ||
2320 | uiov = u64_to_user_ptr(req->rw.addr); | |
2321 | if (!access_ok(uiov, sizeof(*uiov))) | |
2322 | return -EFAULT; | |
2323 | if (__get_user(clen, &uiov->iov_len)) | |
2324 | return -EFAULT; | |
2325 | if (clen < 0) | |
2326 | return -EINVAL; | |
2327 | ||
2328 | len = clen; | |
2329 | buf = io_rw_buffer_select(req, &len, needs_lock); | |
2330 | if (IS_ERR(buf)) | |
2331 | return PTR_ERR(buf); | |
2332 | iov[0].iov_base = buf; | |
2333 | iov[0].iov_len = (compat_size_t) len; | |
2334 | return 0; | |
2335 | } | |
2336 | #endif | |
2337 | ||
2338 | static ssize_t __io_iov_buffer_select(struct io_kiocb *req, struct iovec *iov, | |
2339 | bool needs_lock) | |
2340 | { | |
2341 | struct iovec __user *uiov = u64_to_user_ptr(req->rw.addr); | |
2342 | void __user *buf; | |
2343 | ssize_t len; | |
2344 | ||
2345 | if (copy_from_user(iov, uiov, sizeof(*uiov))) | |
2346 | return -EFAULT; | |
2347 | ||
2348 | len = iov[0].iov_len; | |
2349 | if (len < 0) | |
2350 | return -EINVAL; | |
2351 | buf = io_rw_buffer_select(req, &len, needs_lock); | |
2352 | if (IS_ERR(buf)) | |
2353 | return PTR_ERR(buf); | |
2354 | iov[0].iov_base = buf; | |
2355 | iov[0].iov_len = len; | |
2356 | return 0; | |
2357 | } | |
2358 | ||
2359 | static ssize_t io_iov_buffer_select(struct io_kiocb *req, struct iovec *iov, | |
2360 | bool needs_lock) | |
2361 | { | |
2362 | if (req->flags & REQ_F_BUFFER_SELECTED) | |
2363 | return 0; | |
2364 | if (!req->rw.len) | |
2365 | return 0; | |
2366 | else if (req->rw.len > 1) | |
2367 | return -EINVAL; | |
2368 | ||
2369 | #ifdef CONFIG_COMPAT | |
2370 | if (req->ctx->compat) | |
2371 | return io_compat_import(req, iov, needs_lock); | |
2372 | #endif | |
2373 | ||
2374 | return __io_iov_buffer_select(req, iov, needs_lock); | |
2375 | } | |
2376 | ||
cf6fd4bd | 2377 | static ssize_t io_import_iovec(int rw, struct io_kiocb *req, |
bcda7baa JA |
2378 | struct iovec **iovec, struct iov_iter *iter, |
2379 | bool needs_lock) | |
2b188cc1 | 2380 | { |
9adbd45d JA |
2381 | void __user *buf = u64_to_user_ptr(req->rw.addr); |
2382 | size_t sqe_len = req->rw.len; | |
4d954c25 | 2383 | ssize_t ret; |
edafccee JA |
2384 | u8 opcode; |
2385 | ||
d625c6ee | 2386 | opcode = req->opcode; |
7d009165 | 2387 | if (opcode == IORING_OP_READ_FIXED || opcode == IORING_OP_WRITE_FIXED) { |
edafccee | 2388 | *iovec = NULL; |
9adbd45d | 2389 | return io_import_fixed(req, rw, iter); |
edafccee | 2390 | } |
2b188cc1 | 2391 | |
bcda7baa JA |
2392 | /* buffer index only valid with fixed read/write, or buffer select */ |
2393 | if (req->rw.kiocb.private && !(req->flags & REQ_F_BUFFER_SELECT)) | |
9adbd45d JA |
2394 | return -EINVAL; |
2395 | ||
3a6820f2 | 2396 | if (opcode == IORING_OP_READ || opcode == IORING_OP_WRITE) { |
bcda7baa | 2397 | if (req->flags & REQ_F_BUFFER_SELECT) { |
4d954c25 JA |
2398 | buf = io_rw_buffer_select(req, &sqe_len, needs_lock); |
2399 | if (IS_ERR(buf)) { | |
bcda7baa | 2400 | *iovec = NULL; |
4d954c25 | 2401 | return PTR_ERR(buf); |
bcda7baa | 2402 | } |
3f9d6441 | 2403 | req->rw.len = sqe_len; |
bcda7baa JA |
2404 | } |
2405 | ||
3a6820f2 JA |
2406 | ret = import_single_range(rw, buf, sqe_len, *iovec, iter); |
2407 | *iovec = NULL; | |
3a901598 | 2408 | return ret < 0 ? ret : sqe_len; |
3a6820f2 JA |
2409 | } |
2410 | ||
f67676d1 JA |
2411 | if (req->io) { |
2412 | struct io_async_rw *iorw = &req->io->rw; | |
2413 | ||
2414 | *iovec = iorw->iov; | |
2415 | iov_iter_init(iter, rw, *iovec, iorw->nr_segs, iorw->size); | |
2416 | if (iorw->iov == iorw->fast_iov) | |
2417 | *iovec = NULL; | |
2418 | return iorw->size; | |
2419 | } | |
2420 | ||
4d954c25 JA |
2421 | if (req->flags & REQ_F_BUFFER_SELECT) { |
2422 | ret = io_iov_buffer_select(req, *iovec, needs_lock); | |
3f9d6441 JA |
2423 | if (!ret) { |
2424 | ret = (*iovec)->iov_len; | |
2425 | iov_iter_init(iter, rw, *iovec, 1, ret); | |
2426 | } | |
4d954c25 JA |
2427 | *iovec = NULL; |
2428 | return ret; | |
2429 | } | |
2430 | ||
2b188cc1 | 2431 | #ifdef CONFIG_COMPAT |
cf6fd4bd | 2432 | if (req->ctx->compat) |
2b188cc1 JA |
2433 | return compat_import_iovec(rw, buf, sqe_len, UIO_FASTIOV, |
2434 | iovec, iter); | |
2435 | #endif | |
2436 | ||
2437 | return import_iovec(rw, buf, sqe_len, UIO_FASTIOV, iovec, iter); | |
2438 | } | |
2439 | ||
31b51510 | 2440 | /* |
32960613 JA |
2441 | * For files that don't have ->read_iter() and ->write_iter(), handle them |
2442 | * by looping over ->read() or ->write() manually. | |
31b51510 | 2443 | */ |
32960613 JA |
2444 | static ssize_t loop_rw_iter(int rw, struct file *file, struct kiocb *kiocb, |
2445 | struct iov_iter *iter) | |
2446 | { | |
2447 | ssize_t ret = 0; | |
2448 | ||
2449 | /* | |
2450 | * Don't support polled IO through this interface, and we can't | |
2451 | * support non-blocking either. For the latter, this just causes | |
2452 | * the kiocb to be handled from an async context. | |
2453 | */ | |
2454 | if (kiocb->ki_flags & IOCB_HIPRI) | |
2455 | return -EOPNOTSUPP; | |
2456 | if (kiocb->ki_flags & IOCB_NOWAIT) | |
2457 | return -EAGAIN; | |
2458 | ||
2459 | while (iov_iter_count(iter)) { | |
311ae9e1 | 2460 | struct iovec iovec; |
32960613 JA |
2461 | ssize_t nr; |
2462 | ||
311ae9e1 PB |
2463 | if (!iov_iter_is_bvec(iter)) { |
2464 | iovec = iov_iter_iovec(iter); | |
2465 | } else { | |
2466 | /* fixed buffers import bvec */ | |
2467 | iovec.iov_base = kmap(iter->bvec->bv_page) | |
2468 | + iter->iov_offset; | |
2469 | iovec.iov_len = min(iter->count, | |
2470 | iter->bvec->bv_len - iter->iov_offset); | |
2471 | } | |
2472 | ||
32960613 JA |
2473 | if (rw == READ) { |
2474 | nr = file->f_op->read(file, iovec.iov_base, | |
2475 | iovec.iov_len, &kiocb->ki_pos); | |
2476 | } else { | |
2477 | nr = file->f_op->write(file, iovec.iov_base, | |
2478 | iovec.iov_len, &kiocb->ki_pos); | |
2479 | } | |
2480 | ||
311ae9e1 PB |
2481 | if (iov_iter_is_bvec(iter)) |
2482 | kunmap(iter->bvec->bv_page); | |
2483 | ||
32960613 JA |
2484 | if (nr < 0) { |
2485 | if (!ret) | |
2486 | ret = nr; | |
2487 | break; | |
2488 | } | |
2489 | ret += nr; | |
2490 | if (nr != iovec.iov_len) | |
2491 | break; | |
2492 | iov_iter_advance(iter, nr); | |
2493 | } | |
2494 | ||
2495 | return ret; | |
2496 | } | |
2497 | ||
b7bb4f7d | 2498 | static void io_req_map_rw(struct io_kiocb *req, ssize_t io_size, |
f67676d1 JA |
2499 | struct iovec *iovec, struct iovec *fast_iov, |
2500 | struct iov_iter *iter) | |
2501 | { | |
2502 | req->io->rw.nr_segs = iter->nr_segs; | |
2503 | req->io->rw.size = io_size; | |
2504 | req->io->rw.iov = iovec; | |
2505 | if (!req->io->rw.iov) { | |
2506 | req->io->rw.iov = req->io->rw.fast_iov; | |
45097dae XW |
2507 | if (req->io->rw.iov != fast_iov) |
2508 | memcpy(req->io->rw.iov, fast_iov, | |
2509 | sizeof(struct iovec) * iter->nr_segs); | |
99bc4c38 PB |
2510 | } else { |
2511 | req->flags |= REQ_F_NEED_CLEANUP; | |
f67676d1 JA |
2512 | } |
2513 | } | |
2514 | ||
3d9932a8 XW |
2515 | static inline int __io_alloc_async_ctx(struct io_kiocb *req) |
2516 | { | |
2517 | req->io = kmalloc(sizeof(*req->io), GFP_KERNEL); | |
2518 | return req->io == NULL; | |
2519 | } | |
2520 | ||
b7bb4f7d | 2521 | static int io_alloc_async_ctx(struct io_kiocb *req) |
f67676d1 | 2522 | { |
d3656344 JA |
2523 | if (!io_op_defs[req->opcode].async_ctx) |
2524 | return 0; | |
3d9932a8 XW |
2525 | |
2526 | return __io_alloc_async_ctx(req); | |
b7bb4f7d JA |
2527 | } |
2528 | ||
b7bb4f7d JA |
2529 | static int io_setup_async_rw(struct io_kiocb *req, ssize_t io_size, |
2530 | struct iovec *iovec, struct iovec *fast_iov, | |
2531 | struct iov_iter *iter) | |
2532 | { | |
980ad263 | 2533 | if (!io_op_defs[req->opcode].async_ctx) |
74566df3 | 2534 | return 0; |
5d204bcf | 2535 | if (!req->io) { |
3d9932a8 | 2536 | if (__io_alloc_async_ctx(req)) |
5d204bcf | 2537 | return -ENOMEM; |
b7bb4f7d | 2538 | |
5d204bcf JA |
2539 | io_req_map_rw(req, io_size, iovec, fast_iov, iter); |
2540 | } | |
b7bb4f7d | 2541 | return 0; |
f67676d1 JA |
2542 | } |
2543 | ||
3529d8c2 JA |
2544 | static int io_read_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe, |
2545 | bool force_nonblock) | |
f67676d1 | 2546 | { |
3529d8c2 JA |
2547 | struct io_async_ctx *io; |
2548 | struct iov_iter iter; | |
f67676d1 JA |
2549 | ssize_t ret; |
2550 | ||
3529d8c2 JA |
2551 | ret = io_prep_rw(req, sqe, force_nonblock); |
2552 | if (ret) | |
2553 | return ret; | |
f67676d1 | 2554 | |
3529d8c2 JA |
2555 | if (unlikely(!(req->file->f_mode & FMODE_READ))) |
2556 | return -EBADF; | |
f67676d1 | 2557 | |
5f798bea PB |
2558 | /* either don't need iovec imported or already have it */ |
2559 | if (!req->io || req->flags & REQ_F_NEED_CLEANUP) | |
3529d8c2 JA |
2560 | return 0; |
2561 | ||
2562 | io = req->io; | |
2563 | io->rw.iov = io->rw.fast_iov; | |
2564 | req->io = NULL; | |
bcda7baa | 2565 | ret = io_import_iovec(READ, req, &io->rw.iov, &iter, !force_nonblock); |
3529d8c2 JA |
2566 | req->io = io; |
2567 | if (ret < 0) | |
2568 | return ret; | |
2569 | ||
2570 | io_req_map_rw(req, ret, io->rw.iov, io->rw.fast_iov, &iter); | |
2571 | return 0; | |
f67676d1 JA |
2572 | } |
2573 | ||
014db007 | 2574 | static int io_read(struct io_kiocb *req, bool force_nonblock) |
2b188cc1 JA |
2575 | { |
2576 | struct iovec inline_vecs[UIO_FASTIOV], *iovec = inline_vecs; | |
9adbd45d | 2577 | struct kiocb *kiocb = &req->rw.kiocb; |
2b188cc1 | 2578 | struct iov_iter iter; |
31b51510 | 2579 | size_t iov_count; |
f67676d1 | 2580 | ssize_t io_size, ret; |
2b188cc1 | 2581 | |
bcda7baa | 2582 | ret = io_import_iovec(READ, req, &iovec, &iter, !force_nonblock); |
06b76d44 JA |
2583 | if (ret < 0) |
2584 | return ret; | |
2b188cc1 | 2585 | |
fd6c2e4c JA |
2586 | /* Ensure we clear previously set non-block flag */ |
2587 | if (!force_nonblock) | |
29de5f6a | 2588 | kiocb->ki_flags &= ~IOCB_NOWAIT; |
fd6c2e4c | 2589 | |
797f3f53 | 2590 | req->result = 0; |
f67676d1 | 2591 | io_size = ret; |
dea3b49c | 2592 | if (req->flags & REQ_F_LINK_HEAD) |
f67676d1 JA |
2593 | req->result = io_size; |
2594 | ||
2595 | /* | |
2596 | * If the file doesn't support async, mark it as REQ_F_MUST_PUNT so | |
2597 | * we know to async punt it even if it was opened O_NONBLOCK | |
2598 | */ | |
af197f50 | 2599 | if (force_nonblock && !io_file_supports_async(req->file, READ)) |
f67676d1 | 2600 | goto copy_iov; |
9e645e11 | 2601 | |
31b51510 | 2602 | iov_count = iov_iter_count(&iter); |
9adbd45d | 2603 | ret = rw_verify_area(READ, req->file, &kiocb->ki_pos, iov_count); |
2b188cc1 JA |
2604 | if (!ret) { |
2605 | ssize_t ret2; | |
2606 | ||
9adbd45d JA |
2607 | if (req->file->f_op->read_iter) |
2608 | ret2 = call_read_iter(req->file, kiocb, &iter); | |
32960613 | 2609 | else |
9adbd45d | 2610 | ret2 = loop_rw_iter(READ, req->file, kiocb, &iter); |
32960613 | 2611 | |
9d93a3f5 | 2612 | /* Catch -EAGAIN return for forced non-blocking submission */ |
f67676d1 | 2613 | if (!force_nonblock || ret2 != -EAGAIN) { |
014db007 | 2614 | kiocb_done(kiocb, ret2); |
f67676d1 JA |
2615 | } else { |
2616 | copy_iov: | |
b7bb4f7d | 2617 | ret = io_setup_async_rw(req, io_size, iovec, |
f67676d1 JA |
2618 | inline_vecs, &iter); |
2619 | if (ret) | |
2620 | goto out_free; | |
29de5f6a | 2621 | /* any defer here is final, must blocking retry */ |
490e8967 JA |
2622 | if (!(req->flags & REQ_F_NOWAIT) && |
2623 | !file_can_poll(req->file)) | |
29de5f6a | 2624 | req->flags |= REQ_F_MUST_PUNT; |
f67676d1 JA |
2625 | return -EAGAIN; |
2626 | } | |
2b188cc1 | 2627 | } |
f67676d1 | 2628 | out_free: |
1e95081c | 2629 | kfree(iovec); |
99bc4c38 | 2630 | req->flags &= ~REQ_F_NEED_CLEANUP; |
2b188cc1 JA |
2631 | return ret; |
2632 | } | |
2633 | ||
3529d8c2 JA |
2634 | static int io_write_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe, |
2635 | bool force_nonblock) | |
f67676d1 | 2636 | { |
3529d8c2 JA |
2637 | struct io_async_ctx *io; |
2638 | struct iov_iter iter; | |
f67676d1 JA |
2639 | ssize_t ret; |
2640 | ||
3529d8c2 JA |
2641 | ret = io_prep_rw(req, sqe, force_nonblock); |
2642 | if (ret) | |
2643 | return ret; | |
f67676d1 | 2644 | |
3529d8c2 JA |
2645 | if (unlikely(!(req->file->f_mode & FMODE_WRITE))) |
2646 | return -EBADF; | |
f67676d1 | 2647 | |
4ed734b0 JA |
2648 | req->fsize = rlimit(RLIMIT_FSIZE); |
2649 | ||
5f798bea PB |
2650 | /* either don't need iovec imported or already have it */ |
2651 | if (!req->io || req->flags & REQ_F_NEED_CLEANUP) | |
3529d8c2 JA |
2652 | return 0; |
2653 | ||
2654 | io = req->io; | |
2655 | io->rw.iov = io->rw.fast_iov; | |
2656 | req->io = NULL; | |
bcda7baa | 2657 | ret = io_import_iovec(WRITE, req, &io->rw.iov, &iter, !force_nonblock); |
3529d8c2 JA |
2658 | req->io = io; |
2659 | if (ret < 0) | |
2660 | return ret; | |
2661 | ||
2662 | io_req_map_rw(req, ret, io->rw.iov, io->rw.fast_iov, &iter); | |
2663 | return 0; | |
f67676d1 JA |
2664 | } |
2665 | ||
014db007 | 2666 | static int io_write(struct io_kiocb *req, bool force_nonblock) |
2b188cc1 JA |
2667 | { |
2668 | struct iovec inline_vecs[UIO_FASTIOV], *iovec = inline_vecs; | |
9adbd45d | 2669 | struct kiocb *kiocb = &req->rw.kiocb; |
2b188cc1 | 2670 | struct iov_iter iter; |
31b51510 | 2671 | size_t iov_count; |
f67676d1 | 2672 | ssize_t ret, io_size; |
2b188cc1 | 2673 | |
bcda7baa | 2674 | ret = io_import_iovec(WRITE, req, &iovec, &iter, !force_nonblock); |
06b76d44 JA |
2675 | if (ret < 0) |
2676 | return ret; | |
2b188cc1 | 2677 | |
fd6c2e4c JA |
2678 | /* Ensure we clear previously set non-block flag */ |
2679 | if (!force_nonblock) | |
9adbd45d | 2680 | req->rw.kiocb.ki_flags &= ~IOCB_NOWAIT; |
fd6c2e4c | 2681 | |
797f3f53 | 2682 | req->result = 0; |
f67676d1 | 2683 | io_size = ret; |
dea3b49c | 2684 | if (req->flags & REQ_F_LINK_HEAD) |
f67676d1 | 2685 | req->result = io_size; |
9e645e11 | 2686 | |
f67676d1 JA |
2687 | /* |
2688 | * If the file doesn't support async, mark it as REQ_F_MUST_PUNT so | |
2689 | * we know to async punt it even if it was opened O_NONBLOCK | |
2690 | */ | |
af197f50 | 2691 | if (force_nonblock && !io_file_supports_async(req->file, WRITE)) |
f67676d1 | 2692 | goto copy_iov; |
31b51510 | 2693 | |
10d59345 JA |
2694 | /* file path doesn't support NOWAIT for non-direct_IO */ |
2695 | if (force_nonblock && !(kiocb->ki_flags & IOCB_DIRECT) && | |
2696 | (req->flags & REQ_F_ISREG)) | |
f67676d1 | 2697 | goto copy_iov; |
31b51510 | 2698 | |
f67676d1 | 2699 | iov_count = iov_iter_count(&iter); |
9adbd45d | 2700 | ret = rw_verify_area(WRITE, req->file, &kiocb->ki_pos, iov_count); |
2b188cc1 | 2701 | if (!ret) { |
9bf7933f RP |
2702 | ssize_t ret2; |
2703 | ||
2b188cc1 JA |
2704 | /* |
2705 | * Open-code file_start_write here to grab freeze protection, | |
2706 | * which will be released by another thread in | |
2707 | * io_complete_rw(). Fool lockdep by telling it the lock got | |
2708 | * released so that it doesn't complain about the held lock when | |
2709 | * we return to userspace. | |
2710 | */ | |
491381ce | 2711 | if (req->flags & REQ_F_ISREG) { |
9adbd45d | 2712 | __sb_start_write(file_inode(req->file)->i_sb, |
2b188cc1 | 2713 | SB_FREEZE_WRITE, true); |
9adbd45d | 2714 | __sb_writers_release(file_inode(req->file)->i_sb, |
2b188cc1 JA |
2715 | SB_FREEZE_WRITE); |
2716 | } | |
2717 | kiocb->ki_flags |= IOCB_WRITE; | |
9bf7933f | 2718 | |
4ed734b0 JA |
2719 | if (!force_nonblock) |
2720 | current->signal->rlim[RLIMIT_FSIZE].rlim_cur = req->fsize; | |
2721 | ||
9adbd45d JA |
2722 | if (req->file->f_op->write_iter) |
2723 | ret2 = call_write_iter(req->file, kiocb, &iter); | |
32960613 | 2724 | else |
9adbd45d | 2725 | ret2 = loop_rw_iter(WRITE, req->file, kiocb, &iter); |
4ed734b0 JA |
2726 | |
2727 | if (!force_nonblock) | |
2728 | current->signal->rlim[RLIMIT_FSIZE].rlim_cur = RLIM_INFINITY; | |
2729 | ||
faac996c | 2730 | /* |
bff6035d | 2731 | * Raw bdev writes will return -EOPNOTSUPP for IOCB_NOWAIT. Just |
faac996c JA |
2732 | * retry them without IOCB_NOWAIT. |
2733 | */ | |
2734 | if (ret2 == -EOPNOTSUPP && (kiocb->ki_flags & IOCB_NOWAIT)) | |
2735 | ret2 = -EAGAIN; | |
f67676d1 | 2736 | if (!force_nonblock || ret2 != -EAGAIN) { |
014db007 | 2737 | kiocb_done(kiocb, ret2); |
f67676d1 JA |
2738 | } else { |
2739 | copy_iov: | |
b7bb4f7d | 2740 | ret = io_setup_async_rw(req, io_size, iovec, |
f67676d1 JA |
2741 | inline_vecs, &iter); |
2742 | if (ret) | |
2743 | goto out_free; | |
29de5f6a | 2744 | /* any defer here is final, must blocking retry */ |
490e8967 JA |
2745 | if (!file_can_poll(req->file)) |
2746 | req->flags |= REQ_F_MUST_PUNT; | |
f67676d1 JA |
2747 | return -EAGAIN; |
2748 | } | |
2b188cc1 | 2749 | } |
31b51510 | 2750 | out_free: |
99bc4c38 | 2751 | req->flags &= ~REQ_F_NEED_CLEANUP; |
1e95081c | 2752 | kfree(iovec); |
2b188cc1 JA |
2753 | return ret; |
2754 | } | |
2755 | ||
f2a8d5c7 PB |
2756 | static int __io_splice_prep(struct io_kiocb *req, |
2757 | const struct io_uring_sqe *sqe) | |
7d67af2c PB |
2758 | { |
2759 | struct io_splice* sp = &req->splice; | |
2760 | unsigned int valid_flags = SPLICE_F_FD_IN_FIXED | SPLICE_F_ALL; | |
2761 | int ret; | |
2762 | ||
2763 | if (req->flags & REQ_F_NEED_CLEANUP) | |
2764 | return 0; | |
2765 | ||
2766 | sp->file_in = NULL; | |
7d67af2c PB |
2767 | sp->len = READ_ONCE(sqe->len); |
2768 | sp->flags = READ_ONCE(sqe->splice_flags); | |
2769 | ||
2770 | if (unlikely(sp->flags & ~valid_flags)) | |
2771 | return -EINVAL; | |
2772 | ||
2773 | ret = io_file_get(NULL, req, READ_ONCE(sqe->splice_fd_in), &sp->file_in, | |
2774 | (sp->flags & SPLICE_F_FD_IN_FIXED)); | |
2775 | if (ret) | |
2776 | return ret; | |
2777 | req->flags |= REQ_F_NEED_CLEANUP; | |
2778 | ||
2779 | if (!S_ISREG(file_inode(sp->file_in)->i_mode)) | |
2780 | req->work.flags |= IO_WQ_WORK_UNBOUND; | |
2781 | ||
2782 | return 0; | |
2783 | } | |
2784 | ||
f2a8d5c7 PB |
2785 | static int io_tee_prep(struct io_kiocb *req, |
2786 | const struct io_uring_sqe *sqe) | |
2787 | { | |
2788 | if (READ_ONCE(sqe->splice_off_in) || READ_ONCE(sqe->off)) | |
2789 | return -EINVAL; | |
2790 | return __io_splice_prep(req, sqe); | |
2791 | } | |
2792 | ||
2793 | static int io_tee(struct io_kiocb *req, bool force_nonblock) | |
2794 | { | |
2795 | struct io_splice *sp = &req->splice; | |
2796 | struct file *in = sp->file_in; | |
2797 | struct file *out = sp->file_out; | |
2798 | unsigned int flags = sp->flags & ~SPLICE_F_FD_IN_FIXED; | |
2799 | long ret = 0; | |
2800 | ||
2801 | if (force_nonblock) | |
2802 | return -EAGAIN; | |
2803 | if (sp->len) | |
2804 | ret = do_tee(in, out, sp->len, flags); | |
2805 | ||
2806 | io_put_file(req, in, (sp->flags & SPLICE_F_FD_IN_FIXED)); | |
2807 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
2808 | ||
2809 | io_cqring_add_event(req, ret); | |
2810 | if (ret != sp->len) | |
2811 | req_set_fail_links(req); | |
2812 | io_put_req(req); | |
2813 | return 0; | |
2814 | } | |
2815 | ||
2816 | static int io_splice_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) | |
2817 | { | |
2818 | struct io_splice* sp = &req->splice; | |
2819 | ||
2820 | sp->off_in = READ_ONCE(sqe->splice_off_in); | |
2821 | sp->off_out = READ_ONCE(sqe->off); | |
2822 | return __io_splice_prep(req, sqe); | |
2823 | } | |
2824 | ||
014db007 | 2825 | static int io_splice(struct io_kiocb *req, bool force_nonblock) |
7d67af2c PB |
2826 | { |
2827 | struct io_splice *sp = &req->splice; | |
2828 | struct file *in = sp->file_in; | |
2829 | struct file *out = sp->file_out; | |
2830 | unsigned int flags = sp->flags & ~SPLICE_F_FD_IN_FIXED; | |
2831 | loff_t *poff_in, *poff_out; | |
2832 | long ret; | |
2833 | ||
2fb3e822 PB |
2834 | if (force_nonblock) |
2835 | return -EAGAIN; | |
7d67af2c PB |
2836 | |
2837 | poff_in = (sp->off_in == -1) ? NULL : &sp->off_in; | |
2838 | poff_out = (sp->off_out == -1) ? NULL : &sp->off_out; | |
2839 | ret = do_splice(in, poff_in, out, poff_out, sp->len, flags); | |
2840 | if (force_nonblock && ret == -EAGAIN) | |
2841 | return -EAGAIN; | |
2842 | ||
2843 | io_put_file(req, in, (sp->flags & SPLICE_F_FD_IN_FIXED)); | |
2844 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
2845 | ||
2846 | io_cqring_add_event(req, ret); | |
2847 | if (ret != sp->len) | |
2848 | req_set_fail_links(req); | |
014db007 | 2849 | io_put_req(req); |
7d67af2c PB |
2850 | return 0; |
2851 | } | |
2852 | ||
2b188cc1 JA |
2853 | /* |
2854 | * IORING_OP_NOP just posts a completion event, nothing else. | |
2855 | */ | |
78e19bbe | 2856 | static int io_nop(struct io_kiocb *req) |
2b188cc1 JA |
2857 | { |
2858 | struct io_ring_ctx *ctx = req->ctx; | |
2b188cc1 | 2859 | |
def596e9 JA |
2860 | if (unlikely(ctx->flags & IORING_SETUP_IOPOLL)) |
2861 | return -EINVAL; | |
2862 | ||
78e19bbe | 2863 | io_cqring_add_event(req, 0); |
e65ef56d | 2864 | io_put_req(req); |
2b188cc1 JA |
2865 | return 0; |
2866 | } | |
2867 | ||
3529d8c2 | 2868 | static int io_prep_fsync(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
c992fe29 | 2869 | { |
6b06314c | 2870 | struct io_ring_ctx *ctx = req->ctx; |
c992fe29 | 2871 | |
09bb8394 JA |
2872 | if (!req->file) |
2873 | return -EBADF; | |
c992fe29 | 2874 | |
6b06314c | 2875 | if (unlikely(ctx->flags & IORING_SETUP_IOPOLL)) |
def596e9 | 2876 | return -EINVAL; |
edafccee | 2877 | if (unlikely(sqe->addr || sqe->ioprio || sqe->buf_index)) |
c992fe29 CH |
2878 | return -EINVAL; |
2879 | ||
8ed8d3c3 JA |
2880 | req->sync.flags = READ_ONCE(sqe->fsync_flags); |
2881 | if (unlikely(req->sync.flags & ~IORING_FSYNC_DATASYNC)) | |
2882 | return -EINVAL; | |
2883 | ||
2884 | req->sync.off = READ_ONCE(sqe->off); | |
2885 | req->sync.len = READ_ONCE(sqe->len); | |
c992fe29 CH |
2886 | return 0; |
2887 | } | |
2888 | ||
8ed8d3c3 JA |
2889 | static bool io_req_cancelled(struct io_kiocb *req) |
2890 | { | |
2891 | if (req->work.flags & IO_WQ_WORK_CANCEL) { | |
2892 | req_set_fail_links(req); | |
2893 | io_cqring_add_event(req, -ECANCELED); | |
2894 | io_put_req(req); | |
2895 | return true; | |
2896 | } | |
2897 | ||
2898 | return false; | |
2899 | } | |
2900 | ||
014db007 | 2901 | static void __io_fsync(struct io_kiocb *req) |
8ed8d3c3 | 2902 | { |
8ed8d3c3 | 2903 | loff_t end = req->sync.off + req->sync.len; |
8ed8d3c3 JA |
2904 | int ret; |
2905 | ||
9adbd45d | 2906 | ret = vfs_fsync_range(req->file, req->sync.off, |
8ed8d3c3 JA |
2907 | end > 0 ? end : LLONG_MAX, |
2908 | req->sync.flags & IORING_FSYNC_DATASYNC); | |
2909 | if (ret < 0) | |
2910 | req_set_fail_links(req); | |
2911 | io_cqring_add_event(req, ret); | |
014db007 | 2912 | io_put_req(req); |
8ed8d3c3 JA |
2913 | } |
2914 | ||
5ea62161 | 2915 | static void io_fsync_finish(struct io_wq_work **workptr) |
c992fe29 | 2916 | { |
5ea62161 | 2917 | struct io_kiocb *req = container_of(*workptr, struct io_kiocb, work); |
5ea62161 PB |
2918 | |
2919 | if (io_req_cancelled(req)) | |
2920 | return; | |
014db007 | 2921 | __io_fsync(req); |
e9fd9396 | 2922 | io_steal_work(req, workptr); |
8ed8d3c3 | 2923 | } |
c992fe29 | 2924 | |
014db007 | 2925 | static int io_fsync(struct io_kiocb *req, bool force_nonblock) |
c992fe29 | 2926 | { |
c992fe29 | 2927 | /* fsync always requires a blocking context */ |
8ed8d3c3 | 2928 | if (force_nonblock) { |
8ed8d3c3 | 2929 | req->work.func = io_fsync_finish; |
c992fe29 | 2930 | return -EAGAIN; |
8ed8d3c3 | 2931 | } |
014db007 | 2932 | __io_fsync(req); |
c992fe29 CH |
2933 | return 0; |
2934 | } | |
2935 | ||
014db007 | 2936 | static void __io_fallocate(struct io_kiocb *req) |
8ed8d3c3 | 2937 | { |
8ed8d3c3 JA |
2938 | int ret; |
2939 | ||
4ed734b0 | 2940 | current->signal->rlim[RLIMIT_FSIZE].rlim_cur = req->fsize; |
d63d1b5e JA |
2941 | ret = vfs_fallocate(req->file, req->sync.mode, req->sync.off, |
2942 | req->sync.len); | |
4ed734b0 | 2943 | current->signal->rlim[RLIMIT_FSIZE].rlim_cur = RLIM_INFINITY; |
8ed8d3c3 JA |
2944 | if (ret < 0) |
2945 | req_set_fail_links(req); | |
2946 | io_cqring_add_event(req, ret); | |
014db007 | 2947 | io_put_req(req); |
5ea62161 PB |
2948 | } |
2949 | ||
2950 | static void io_fallocate_finish(struct io_wq_work **workptr) | |
2951 | { | |
2952 | struct io_kiocb *req = container_of(*workptr, struct io_kiocb, work); | |
5ea62161 | 2953 | |
594506fe PB |
2954 | if (io_req_cancelled(req)) |
2955 | return; | |
014db007 | 2956 | __io_fallocate(req); |
e9fd9396 | 2957 | io_steal_work(req, workptr); |
5d17b4a4 JA |
2958 | } |
2959 | ||
d63d1b5e JA |
2960 | static int io_fallocate_prep(struct io_kiocb *req, |
2961 | const struct io_uring_sqe *sqe) | |
2962 | { | |
2963 | if (sqe->ioprio || sqe->buf_index || sqe->rw_flags) | |
2964 | return -EINVAL; | |
2965 | ||
2966 | req->sync.off = READ_ONCE(sqe->off); | |
2967 | req->sync.len = READ_ONCE(sqe->addr); | |
2968 | req->sync.mode = READ_ONCE(sqe->len); | |
4ed734b0 | 2969 | req->fsize = rlimit(RLIMIT_FSIZE); |
d63d1b5e JA |
2970 | return 0; |
2971 | } | |
2972 | ||
014db007 | 2973 | static int io_fallocate(struct io_kiocb *req, bool force_nonblock) |
5d17b4a4 | 2974 | { |
d63d1b5e | 2975 | /* fallocate always requiring blocking context */ |
8ed8d3c3 | 2976 | if (force_nonblock) { |
d63d1b5e | 2977 | req->work.func = io_fallocate_finish; |
5d17b4a4 | 2978 | return -EAGAIN; |
8ed8d3c3 | 2979 | } |
5d17b4a4 | 2980 | |
014db007 | 2981 | __io_fallocate(req); |
5d17b4a4 JA |
2982 | return 0; |
2983 | } | |
2984 | ||
15b71abe | 2985 | static int io_openat_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
b7bb4f7d | 2986 | { |
f8748881 | 2987 | const char __user *fname; |
15b71abe | 2988 | int ret; |
b7bb4f7d | 2989 | |
15b71abe JA |
2990 | if (sqe->ioprio || sqe->buf_index) |
2991 | return -EINVAL; | |
9c280f90 | 2992 | if (req->flags & REQ_F_FIXED_FILE) |
cf3040ca | 2993 | return -EBADF; |
0bdbdd08 PB |
2994 | if (req->flags & REQ_F_NEED_CLEANUP) |
2995 | return 0; | |
03b1230c | 2996 | |
15b71abe | 2997 | req->open.dfd = READ_ONCE(sqe->fd); |
c12cedf2 | 2998 | req->open.how.mode = READ_ONCE(sqe->len); |
f8748881 | 2999 | fname = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
c12cedf2 | 3000 | req->open.how.flags = READ_ONCE(sqe->open_flags); |
08a1d26e JA |
3001 | if (force_o_largefile()) |
3002 | req->open.how.flags |= O_LARGEFILE; | |
3529d8c2 | 3003 | |
f8748881 | 3004 | req->open.filename = getname(fname); |
15b71abe JA |
3005 | if (IS_ERR(req->open.filename)) { |
3006 | ret = PTR_ERR(req->open.filename); | |
3007 | req->open.filename = NULL; | |
3008 | return ret; | |
3009 | } | |
3529d8c2 | 3010 | |
4022e7af | 3011 | req->open.nofile = rlimit(RLIMIT_NOFILE); |
8fef80bf | 3012 | req->flags |= REQ_F_NEED_CLEANUP; |
15b71abe | 3013 | return 0; |
03b1230c JA |
3014 | } |
3015 | ||
cebdb986 | 3016 | static int io_openat2_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
aa1fa28f | 3017 | { |
cebdb986 JA |
3018 | struct open_how __user *how; |
3019 | const char __user *fname; | |
3020 | size_t len; | |
0fa03c62 JA |
3021 | int ret; |
3022 | ||
cebdb986 | 3023 | if (sqe->ioprio || sqe->buf_index) |
0fa03c62 | 3024 | return -EINVAL; |
9c280f90 | 3025 | if (req->flags & REQ_F_FIXED_FILE) |
cf3040ca | 3026 | return -EBADF; |
0bdbdd08 PB |
3027 | if (req->flags & REQ_F_NEED_CLEANUP) |
3028 | return 0; | |
0fa03c62 | 3029 | |
cebdb986 JA |
3030 | req->open.dfd = READ_ONCE(sqe->fd); |
3031 | fname = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
3032 | how = u64_to_user_ptr(READ_ONCE(sqe->addr2)); | |
3033 | len = READ_ONCE(sqe->len); | |
0fa03c62 | 3034 | |
cebdb986 JA |
3035 | if (len < OPEN_HOW_SIZE_VER0) |
3036 | return -EINVAL; | |
3529d8c2 | 3037 | |
cebdb986 JA |
3038 | ret = copy_struct_from_user(&req->open.how, sizeof(req->open.how), how, |
3039 | len); | |
3040 | if (ret) | |
3041 | return ret; | |
3529d8c2 | 3042 | |
cebdb986 JA |
3043 | if (!(req->open.how.flags & O_PATH) && force_o_largefile()) |
3044 | req->open.how.flags |= O_LARGEFILE; | |
0fa03c62 | 3045 | |
cebdb986 JA |
3046 | req->open.filename = getname(fname); |
3047 | if (IS_ERR(req->open.filename)) { | |
3048 | ret = PTR_ERR(req->open.filename); | |
3049 | req->open.filename = NULL; | |
3050 | return ret; | |
3051 | } | |
3052 | ||
4022e7af | 3053 | req->open.nofile = rlimit(RLIMIT_NOFILE); |
8fef80bf | 3054 | req->flags |= REQ_F_NEED_CLEANUP; |
cebdb986 JA |
3055 | return 0; |
3056 | } | |
3057 | ||
014db007 | 3058 | static int io_openat2(struct io_kiocb *req, bool force_nonblock) |
15b71abe JA |
3059 | { |
3060 | struct open_flags op; | |
15b71abe JA |
3061 | struct file *file; |
3062 | int ret; | |
3063 | ||
f86cd20c | 3064 | if (force_nonblock) |
15b71abe | 3065 | return -EAGAIN; |
15b71abe | 3066 | |
cebdb986 | 3067 | ret = build_open_flags(&req->open.how, &op); |
15b71abe JA |
3068 | if (ret) |
3069 | goto err; | |
3070 | ||
4022e7af | 3071 | ret = __get_unused_fd_flags(req->open.how.flags, req->open.nofile); |
15b71abe JA |
3072 | if (ret < 0) |
3073 | goto err; | |
3074 | ||
3075 | file = do_filp_open(req->open.dfd, req->open.filename, &op); | |
3076 | if (IS_ERR(file)) { | |
3077 | put_unused_fd(ret); | |
3078 | ret = PTR_ERR(file); | |
3079 | } else { | |
3080 | fsnotify_open(file); | |
3081 | fd_install(ret, file); | |
3082 | } | |
3083 | err: | |
3084 | putname(req->open.filename); | |
8fef80bf | 3085 | req->flags &= ~REQ_F_NEED_CLEANUP; |
15b71abe JA |
3086 | if (ret < 0) |
3087 | req_set_fail_links(req); | |
3088 | io_cqring_add_event(req, ret); | |
014db007 | 3089 | io_put_req(req); |
15b71abe JA |
3090 | return 0; |
3091 | } | |
3092 | ||
014db007 | 3093 | static int io_openat(struct io_kiocb *req, bool force_nonblock) |
cebdb986 JA |
3094 | { |
3095 | req->open.how = build_open_how(req->open.how.flags, req->open.how.mode); | |
014db007 | 3096 | return io_openat2(req, force_nonblock); |
cebdb986 JA |
3097 | } |
3098 | ||
067524e9 JA |
3099 | static int io_remove_buffers_prep(struct io_kiocb *req, |
3100 | const struct io_uring_sqe *sqe) | |
3101 | { | |
3102 | struct io_provide_buf *p = &req->pbuf; | |
3103 | u64 tmp; | |
3104 | ||
3105 | if (sqe->ioprio || sqe->rw_flags || sqe->addr || sqe->len || sqe->off) | |
3106 | return -EINVAL; | |
3107 | ||
3108 | tmp = READ_ONCE(sqe->fd); | |
3109 | if (!tmp || tmp > USHRT_MAX) | |
3110 | return -EINVAL; | |
3111 | ||
3112 | memset(p, 0, sizeof(*p)); | |
3113 | p->nbufs = tmp; | |
3114 | p->bgid = READ_ONCE(sqe->buf_group); | |
3115 | return 0; | |
3116 | } | |
3117 | ||
3118 | static int __io_remove_buffers(struct io_ring_ctx *ctx, struct io_buffer *buf, | |
3119 | int bgid, unsigned nbufs) | |
3120 | { | |
3121 | unsigned i = 0; | |
3122 | ||
3123 | /* shouldn't happen */ | |
3124 | if (!nbufs) | |
3125 | return 0; | |
3126 | ||
3127 | /* the head kbuf is the list itself */ | |
3128 | while (!list_empty(&buf->list)) { | |
3129 | struct io_buffer *nxt; | |
3130 | ||
3131 | nxt = list_first_entry(&buf->list, struct io_buffer, list); | |
3132 | list_del(&nxt->list); | |
3133 | kfree(nxt); | |
3134 | if (++i == nbufs) | |
3135 | return i; | |
3136 | } | |
3137 | i++; | |
3138 | kfree(buf); | |
3139 | idr_remove(&ctx->io_buffer_idr, bgid); | |
3140 | ||
3141 | return i; | |
3142 | } | |
3143 | ||
3144 | static int io_remove_buffers(struct io_kiocb *req, bool force_nonblock) | |
3145 | { | |
3146 | struct io_provide_buf *p = &req->pbuf; | |
3147 | struct io_ring_ctx *ctx = req->ctx; | |
3148 | struct io_buffer *head; | |
3149 | int ret = 0; | |
3150 | ||
3151 | io_ring_submit_lock(ctx, !force_nonblock); | |
3152 | ||
3153 | lockdep_assert_held(&ctx->uring_lock); | |
3154 | ||
3155 | ret = -ENOENT; | |
3156 | head = idr_find(&ctx->io_buffer_idr, p->bgid); | |
3157 | if (head) | |
3158 | ret = __io_remove_buffers(ctx, head, p->bgid, p->nbufs); | |
3159 | ||
3160 | io_ring_submit_lock(ctx, !force_nonblock); | |
3161 | if (ret < 0) | |
3162 | req_set_fail_links(req); | |
3163 | io_cqring_add_event(req, ret); | |
3164 | io_put_req(req); | |
3165 | return 0; | |
3166 | } | |
3167 | ||
ddf0322d JA |
3168 | static int io_provide_buffers_prep(struct io_kiocb *req, |
3169 | const struct io_uring_sqe *sqe) | |
3170 | { | |
3171 | struct io_provide_buf *p = &req->pbuf; | |
3172 | u64 tmp; | |
3173 | ||
3174 | if (sqe->ioprio || sqe->rw_flags) | |
3175 | return -EINVAL; | |
3176 | ||
3177 | tmp = READ_ONCE(sqe->fd); | |
3178 | if (!tmp || tmp > USHRT_MAX) | |
3179 | return -E2BIG; | |
3180 | p->nbufs = tmp; | |
3181 | p->addr = READ_ONCE(sqe->addr); | |
3182 | p->len = READ_ONCE(sqe->len); | |
3183 | ||
3184 | if (!access_ok(u64_to_user_ptr(p->addr), p->len)) | |
3185 | return -EFAULT; | |
3186 | ||
3187 | p->bgid = READ_ONCE(sqe->buf_group); | |
3188 | tmp = READ_ONCE(sqe->off); | |
3189 | if (tmp > USHRT_MAX) | |
3190 | return -E2BIG; | |
3191 | p->bid = tmp; | |
3192 | return 0; | |
3193 | } | |
3194 | ||
3195 | static int io_add_buffers(struct io_provide_buf *pbuf, struct io_buffer **head) | |
3196 | { | |
3197 | struct io_buffer *buf; | |
3198 | u64 addr = pbuf->addr; | |
3199 | int i, bid = pbuf->bid; | |
3200 | ||
3201 | for (i = 0; i < pbuf->nbufs; i++) { | |
3202 | buf = kmalloc(sizeof(*buf), GFP_KERNEL); | |
3203 | if (!buf) | |
3204 | break; | |
3205 | ||
3206 | buf->addr = addr; | |
3207 | buf->len = pbuf->len; | |
3208 | buf->bid = bid; | |
3209 | addr += pbuf->len; | |
3210 | bid++; | |
3211 | if (!*head) { | |
3212 | INIT_LIST_HEAD(&buf->list); | |
3213 | *head = buf; | |
3214 | } else { | |
3215 | list_add_tail(&buf->list, &(*head)->list); | |
3216 | } | |
3217 | } | |
3218 | ||
3219 | return i ? i : -ENOMEM; | |
3220 | } | |
3221 | ||
ddf0322d JA |
3222 | static int io_provide_buffers(struct io_kiocb *req, bool force_nonblock) |
3223 | { | |
3224 | struct io_provide_buf *p = &req->pbuf; | |
3225 | struct io_ring_ctx *ctx = req->ctx; | |
3226 | struct io_buffer *head, *list; | |
3227 | int ret = 0; | |
3228 | ||
3229 | io_ring_submit_lock(ctx, !force_nonblock); | |
3230 | ||
3231 | lockdep_assert_held(&ctx->uring_lock); | |
3232 | ||
3233 | list = head = idr_find(&ctx->io_buffer_idr, p->bgid); | |
3234 | ||
3235 | ret = io_add_buffers(p, &head); | |
3236 | if (ret < 0) | |
3237 | goto out; | |
3238 | ||
3239 | if (!list) { | |
3240 | ret = idr_alloc(&ctx->io_buffer_idr, head, p->bgid, p->bgid + 1, | |
3241 | GFP_KERNEL); | |
3242 | if (ret < 0) { | |
067524e9 | 3243 | __io_remove_buffers(ctx, head, p->bgid, -1U); |
ddf0322d JA |
3244 | goto out; |
3245 | } | |
3246 | } | |
3247 | out: | |
3248 | io_ring_submit_unlock(ctx, !force_nonblock); | |
3249 | if (ret < 0) | |
3250 | req_set_fail_links(req); | |
3251 | io_cqring_add_event(req, ret); | |
3252 | io_put_req(req); | |
3253 | return 0; | |
cebdb986 JA |
3254 | } |
3255 | ||
3e4827b0 JA |
3256 | static int io_epoll_ctl_prep(struct io_kiocb *req, |
3257 | const struct io_uring_sqe *sqe) | |
3258 | { | |
3259 | #if defined(CONFIG_EPOLL) | |
3260 | if (sqe->ioprio || sqe->buf_index) | |
3261 | return -EINVAL; | |
3262 | ||
3263 | req->epoll.epfd = READ_ONCE(sqe->fd); | |
3264 | req->epoll.op = READ_ONCE(sqe->len); | |
3265 | req->epoll.fd = READ_ONCE(sqe->off); | |
3266 | ||
3267 | if (ep_op_has_event(req->epoll.op)) { | |
3268 | struct epoll_event __user *ev; | |
3269 | ||
3270 | ev = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
3271 | if (copy_from_user(&req->epoll.event, ev, sizeof(*ev))) | |
3272 | return -EFAULT; | |
3273 | } | |
3274 | ||
3275 | return 0; | |
3276 | #else | |
3277 | return -EOPNOTSUPP; | |
3278 | #endif | |
3279 | } | |
3280 | ||
014db007 | 3281 | static int io_epoll_ctl(struct io_kiocb *req, bool force_nonblock) |
3e4827b0 JA |
3282 | { |
3283 | #if defined(CONFIG_EPOLL) | |
3284 | struct io_epoll *ie = &req->epoll; | |
3285 | int ret; | |
3286 | ||
3287 | ret = do_epoll_ctl(ie->epfd, ie->op, ie->fd, &ie->event, force_nonblock); | |
3288 | if (force_nonblock && ret == -EAGAIN) | |
3289 | return -EAGAIN; | |
3290 | ||
3291 | if (ret < 0) | |
3292 | req_set_fail_links(req); | |
3293 | io_cqring_add_event(req, ret); | |
014db007 | 3294 | io_put_req(req); |
3e4827b0 JA |
3295 | return 0; |
3296 | #else | |
3297 | return -EOPNOTSUPP; | |
3298 | #endif | |
3299 | } | |
3300 | ||
c1ca757b JA |
3301 | static int io_madvise_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
3302 | { | |
3303 | #if defined(CONFIG_ADVISE_SYSCALLS) && defined(CONFIG_MMU) | |
3304 | if (sqe->ioprio || sqe->buf_index || sqe->off) | |
3305 | return -EINVAL; | |
3306 | ||
3307 | req->madvise.addr = READ_ONCE(sqe->addr); | |
3308 | req->madvise.len = READ_ONCE(sqe->len); | |
3309 | req->madvise.advice = READ_ONCE(sqe->fadvise_advice); | |
3310 | return 0; | |
3311 | #else | |
3312 | return -EOPNOTSUPP; | |
3313 | #endif | |
3314 | } | |
3315 | ||
014db007 | 3316 | static int io_madvise(struct io_kiocb *req, bool force_nonblock) |
c1ca757b JA |
3317 | { |
3318 | #if defined(CONFIG_ADVISE_SYSCALLS) && defined(CONFIG_MMU) | |
3319 | struct io_madvise *ma = &req->madvise; | |
3320 | int ret; | |
3321 | ||
3322 | if (force_nonblock) | |
3323 | return -EAGAIN; | |
3324 | ||
3325 | ret = do_madvise(ma->addr, ma->len, ma->advice); | |
3326 | if (ret < 0) | |
3327 | req_set_fail_links(req); | |
3328 | io_cqring_add_event(req, ret); | |
014db007 | 3329 | io_put_req(req); |
c1ca757b JA |
3330 | return 0; |
3331 | #else | |
3332 | return -EOPNOTSUPP; | |
3333 | #endif | |
3334 | } | |
3335 | ||
4840e418 JA |
3336 | static int io_fadvise_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
3337 | { | |
3338 | if (sqe->ioprio || sqe->buf_index || sqe->addr) | |
3339 | return -EINVAL; | |
3340 | ||
3341 | req->fadvise.offset = READ_ONCE(sqe->off); | |
3342 | req->fadvise.len = READ_ONCE(sqe->len); | |
3343 | req->fadvise.advice = READ_ONCE(sqe->fadvise_advice); | |
3344 | return 0; | |
3345 | } | |
3346 | ||
014db007 | 3347 | static int io_fadvise(struct io_kiocb *req, bool force_nonblock) |
4840e418 JA |
3348 | { |
3349 | struct io_fadvise *fa = &req->fadvise; | |
3350 | int ret; | |
3351 | ||
3e69426d JA |
3352 | if (force_nonblock) { |
3353 | switch (fa->advice) { | |
3354 | case POSIX_FADV_NORMAL: | |
3355 | case POSIX_FADV_RANDOM: | |
3356 | case POSIX_FADV_SEQUENTIAL: | |
3357 | break; | |
3358 | default: | |
3359 | return -EAGAIN; | |
3360 | } | |
3361 | } | |
4840e418 JA |
3362 | |
3363 | ret = vfs_fadvise(req->file, fa->offset, fa->len, fa->advice); | |
3364 | if (ret < 0) | |
3365 | req_set_fail_links(req); | |
3366 | io_cqring_add_event(req, ret); | |
014db007 | 3367 | io_put_req(req); |
4840e418 JA |
3368 | return 0; |
3369 | } | |
3370 | ||
eddc7ef5 JA |
3371 | static int io_statx_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
3372 | { | |
f8748881 | 3373 | const char __user *fname; |
eddc7ef5 JA |
3374 | unsigned lookup_flags; |
3375 | int ret; | |
3376 | ||
3377 | if (sqe->ioprio || sqe->buf_index) | |
3378 | return -EINVAL; | |
9c280f90 | 3379 | if (req->flags & REQ_F_FIXED_FILE) |
cf3040ca | 3380 | return -EBADF; |
0bdbdd08 PB |
3381 | if (req->flags & REQ_F_NEED_CLEANUP) |
3382 | return 0; | |
eddc7ef5 JA |
3383 | |
3384 | req->open.dfd = READ_ONCE(sqe->fd); | |
3385 | req->open.mask = READ_ONCE(sqe->len); | |
f8748881 | 3386 | fname = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
eddc7ef5 | 3387 | req->open.buffer = u64_to_user_ptr(READ_ONCE(sqe->addr2)); |
c12cedf2 | 3388 | req->open.how.flags = READ_ONCE(sqe->statx_flags); |
eddc7ef5 | 3389 | |
c12cedf2 | 3390 | if (vfs_stat_set_lookup_flags(&lookup_flags, req->open.how.flags)) |
eddc7ef5 JA |
3391 | return -EINVAL; |
3392 | ||
f8748881 | 3393 | req->open.filename = getname_flags(fname, lookup_flags, NULL); |
eddc7ef5 JA |
3394 | if (IS_ERR(req->open.filename)) { |
3395 | ret = PTR_ERR(req->open.filename); | |
3396 | req->open.filename = NULL; | |
3397 | return ret; | |
3398 | } | |
3399 | ||
8fef80bf | 3400 | req->flags |= REQ_F_NEED_CLEANUP; |
eddc7ef5 JA |
3401 | return 0; |
3402 | } | |
3403 | ||
014db007 | 3404 | static int io_statx(struct io_kiocb *req, bool force_nonblock) |
eddc7ef5 JA |
3405 | { |
3406 | struct io_open *ctx = &req->open; | |
3407 | unsigned lookup_flags; | |
3408 | struct path path; | |
3409 | struct kstat stat; | |
3410 | int ret; | |
3411 | ||
5b0bbee4 JA |
3412 | if (force_nonblock) { |
3413 | /* only need file table for an actual valid fd */ | |
3414 | if (ctx->dfd == -1 || ctx->dfd == AT_FDCWD) | |
3415 | req->flags |= REQ_F_NO_FILE_TABLE; | |
eddc7ef5 | 3416 | return -EAGAIN; |
5b0bbee4 | 3417 | } |
eddc7ef5 | 3418 | |
c12cedf2 | 3419 | if (vfs_stat_set_lookup_flags(&lookup_flags, ctx->how.flags)) |
eddc7ef5 JA |
3420 | return -EINVAL; |
3421 | ||
3422 | retry: | |
3423 | /* filename_lookup() drops it, keep a reference */ | |
3424 | ctx->filename->refcnt++; | |
3425 | ||
3426 | ret = filename_lookup(ctx->dfd, ctx->filename, lookup_flags, &path, | |
3427 | NULL); | |
3428 | if (ret) | |
3429 | goto err; | |
3430 | ||
c12cedf2 | 3431 | ret = vfs_getattr(&path, &stat, ctx->mask, ctx->how.flags); |
eddc7ef5 JA |
3432 | path_put(&path); |
3433 | if (retry_estale(ret, lookup_flags)) { | |
3434 | lookup_flags |= LOOKUP_REVAL; | |
3435 | goto retry; | |
3436 | } | |
3437 | if (!ret) | |
3438 | ret = cp_statx(&stat, ctx->buffer); | |
3439 | err: | |
3440 | putname(ctx->filename); | |
8fef80bf | 3441 | req->flags &= ~REQ_F_NEED_CLEANUP; |
eddc7ef5 JA |
3442 | if (ret < 0) |
3443 | req_set_fail_links(req); | |
3444 | io_cqring_add_event(req, ret); | |
014db007 | 3445 | io_put_req(req); |
eddc7ef5 JA |
3446 | return 0; |
3447 | } | |
3448 | ||
b5dba59e JA |
3449 | static int io_close_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
3450 | { | |
3451 | /* | |
3452 | * If we queue this for async, it must not be cancellable. That would | |
3453 | * leave the 'file' in an undeterminate state. | |
3454 | */ | |
3455 | req->work.flags |= IO_WQ_WORK_NO_CANCEL; | |
3456 | ||
3457 | if (sqe->ioprio || sqe->off || sqe->addr || sqe->len || | |
3458 | sqe->rw_flags || sqe->buf_index) | |
3459 | return -EINVAL; | |
9c280f90 | 3460 | if (req->flags & REQ_F_FIXED_FILE) |
cf3040ca | 3461 | return -EBADF; |
b5dba59e JA |
3462 | |
3463 | req->close.fd = READ_ONCE(sqe->fd); | |
b5dba59e JA |
3464 | return 0; |
3465 | } | |
3466 | ||
a93b3331 | 3467 | /* only called when __close_fd_get_file() is done */ |
014db007 | 3468 | static void __io_close_finish(struct io_kiocb *req) |
a93b3331 PB |
3469 | { |
3470 | int ret; | |
3471 | ||
3472 | ret = filp_close(req->close.put_file, req->work.files); | |
3473 | if (ret < 0) | |
3474 | req_set_fail_links(req); | |
3475 | io_cqring_add_event(req, ret); | |
3476 | fput(req->close.put_file); | |
014db007 | 3477 | io_put_req(req); |
a93b3331 PB |
3478 | } |
3479 | ||
b5dba59e JA |
3480 | static void io_close_finish(struct io_wq_work **workptr) |
3481 | { | |
3482 | struct io_kiocb *req = container_of(*workptr, struct io_kiocb, work); | |
b5dba59e | 3483 | |
7fbeb95d | 3484 | /* not cancellable, don't do io_req_cancelled() */ |
014db007 | 3485 | __io_close_finish(req); |
e9fd9396 | 3486 | io_steal_work(req, workptr); |
b5dba59e JA |
3487 | } |
3488 | ||
014db007 | 3489 | static int io_close(struct io_kiocb *req, bool force_nonblock) |
b5dba59e JA |
3490 | { |
3491 | int ret; | |
3492 | ||
3493 | req->close.put_file = NULL; | |
3494 | ret = __close_fd_get_file(req->close.fd, &req->close.put_file); | |
904fbcb1 JA |
3495 | if (ret < 0) { |
3496 | if (ret == -ENOENT) | |
3497 | ret = -EBADF; | |
b5dba59e | 3498 | return ret; |
904fbcb1 | 3499 | } |
b5dba59e JA |
3500 | |
3501 | /* if the file has a flush method, be safe and punt to async */ | |
a2100672 | 3502 | if (req->close.put_file->f_op->flush && force_nonblock) { |
594506fe PB |
3503 | /* submission ref will be dropped, take it for async */ |
3504 | refcount_inc(&req->refs); | |
3505 | ||
a2100672 PB |
3506 | req->work.func = io_close_finish; |
3507 | /* | |
3508 | * Do manual async queue here to avoid grabbing files - we don't | |
3509 | * need the files, and it'll cause io_close_finish() to close | |
3510 | * the file again and cause a double CQE entry for this request | |
3511 | */ | |
3512 | io_queue_async_work(req); | |
3513 | return 0; | |
3514 | } | |
b5dba59e JA |
3515 | |
3516 | /* | |
3517 | * No ->flush(), safely close from here and just punt the | |
3518 | * fput() to async context. | |
3519 | */ | |
014db007 | 3520 | __io_close_finish(req); |
1a417f4e | 3521 | return 0; |
b5dba59e JA |
3522 | } |
3523 | ||
3529d8c2 | 3524 | static int io_prep_sfr(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
5d17b4a4 JA |
3525 | { |
3526 | struct io_ring_ctx *ctx = req->ctx; | |
5d17b4a4 JA |
3527 | |
3528 | if (!req->file) | |
3529 | return -EBADF; | |
5d17b4a4 JA |
3530 | |
3531 | if (unlikely(ctx->flags & IORING_SETUP_IOPOLL)) | |
3532 | return -EINVAL; | |
3533 | if (unlikely(sqe->addr || sqe->ioprio || sqe->buf_index)) | |
3534 | return -EINVAL; | |
3535 | ||
8ed8d3c3 JA |
3536 | req->sync.off = READ_ONCE(sqe->off); |
3537 | req->sync.len = READ_ONCE(sqe->len); | |
3538 | req->sync.flags = READ_ONCE(sqe->sync_range_flags); | |
8ed8d3c3 JA |
3539 | return 0; |
3540 | } | |
3541 | ||
014db007 | 3542 | static void __io_sync_file_range(struct io_kiocb *req) |
8ed8d3c3 | 3543 | { |
8ed8d3c3 JA |
3544 | int ret; |
3545 | ||
9adbd45d | 3546 | ret = sync_file_range(req->file, req->sync.off, req->sync.len, |
8ed8d3c3 JA |
3547 | req->sync.flags); |
3548 | if (ret < 0) | |
3549 | req_set_fail_links(req); | |
3550 | io_cqring_add_event(req, ret); | |
014db007 | 3551 | io_put_req(req); |
5ea62161 PB |
3552 | } |
3553 | ||
3554 | ||
3555 | static void io_sync_file_range_finish(struct io_wq_work **workptr) | |
3556 | { | |
3557 | struct io_kiocb *req = container_of(*workptr, struct io_kiocb, work); | |
5ea62161 PB |
3558 | |
3559 | if (io_req_cancelled(req)) | |
3560 | return; | |
014db007 | 3561 | __io_sync_file_range(req); |
7759a0bf | 3562 | io_steal_work(req, workptr); |
5d17b4a4 JA |
3563 | } |
3564 | ||
014db007 | 3565 | static int io_sync_file_range(struct io_kiocb *req, bool force_nonblock) |
5d17b4a4 | 3566 | { |
5d17b4a4 | 3567 | /* sync_file_range always requires a blocking context */ |
8ed8d3c3 | 3568 | if (force_nonblock) { |
8ed8d3c3 | 3569 | req->work.func = io_sync_file_range_finish; |
5d17b4a4 | 3570 | return -EAGAIN; |
8ed8d3c3 | 3571 | } |
5d17b4a4 | 3572 | |
014db007 | 3573 | __io_sync_file_range(req); |
5d17b4a4 JA |
3574 | return 0; |
3575 | } | |
3576 | ||
469956e8 | 3577 | #if defined(CONFIG_NET) |
02d27d89 PB |
3578 | static int io_setup_async_msg(struct io_kiocb *req, |
3579 | struct io_async_msghdr *kmsg) | |
3580 | { | |
3581 | if (req->io) | |
3582 | return -EAGAIN; | |
3583 | if (io_alloc_async_ctx(req)) { | |
3584 | if (kmsg->iov != kmsg->fast_iov) | |
3585 | kfree(kmsg->iov); | |
3586 | return -ENOMEM; | |
3587 | } | |
3588 | req->flags |= REQ_F_NEED_CLEANUP; | |
3589 | memcpy(&req->io->msg, kmsg, sizeof(*kmsg)); | |
3590 | return -EAGAIN; | |
3591 | } | |
3592 | ||
3529d8c2 | 3593 | static int io_sendmsg_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
03b1230c | 3594 | { |
e47293fd | 3595 | struct io_sr_msg *sr = &req->sr_msg; |
3529d8c2 | 3596 | struct io_async_ctx *io = req->io; |
99bc4c38 | 3597 | int ret; |
03b1230c | 3598 | |
e47293fd JA |
3599 | sr->msg_flags = READ_ONCE(sqe->msg_flags); |
3600 | sr->msg = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
fddaface | 3601 | sr->len = READ_ONCE(sqe->len); |
3529d8c2 | 3602 | |
d8768362 JA |
3603 | #ifdef CONFIG_COMPAT |
3604 | if (req->ctx->compat) | |
3605 | sr->msg_flags |= MSG_CMSG_COMPAT; | |
3606 | #endif | |
3607 | ||
fddaface | 3608 | if (!io || req->opcode == IORING_OP_SEND) |
3529d8c2 | 3609 | return 0; |
5f798bea PB |
3610 | /* iovec is already imported */ |
3611 | if (req->flags & REQ_F_NEED_CLEANUP) | |
3612 | return 0; | |
3529d8c2 | 3613 | |
d9688565 | 3614 | io->msg.iov = io->msg.fast_iov; |
99bc4c38 | 3615 | ret = sendmsg_copy_msghdr(&io->msg.msg, sr->msg, sr->msg_flags, |
e47293fd | 3616 | &io->msg.iov); |
99bc4c38 PB |
3617 | if (!ret) |
3618 | req->flags |= REQ_F_NEED_CLEANUP; | |
3619 | return ret; | |
03b1230c JA |
3620 | } |
3621 | ||
014db007 | 3622 | static int io_sendmsg(struct io_kiocb *req, bool force_nonblock) |
aa1fa28f | 3623 | { |
0b416c3e | 3624 | struct io_async_msghdr *kmsg = NULL; |
0fa03c62 JA |
3625 | struct socket *sock; |
3626 | int ret; | |
3627 | ||
3628 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) | |
3629 | return -EINVAL; | |
3630 | ||
3631 | sock = sock_from_file(req->file, &ret); | |
3632 | if (sock) { | |
b7bb4f7d | 3633 | struct io_async_ctx io; |
0fa03c62 JA |
3634 | unsigned flags; |
3635 | ||
03b1230c | 3636 | if (req->io) { |
0b416c3e | 3637 | kmsg = &req->io->msg; |
b537916c | 3638 | kmsg->msg.msg_name = &req->io->msg.addr; |
0b416c3e JA |
3639 | /* if iov is set, it's allocated already */ |
3640 | if (!kmsg->iov) | |
3641 | kmsg->iov = kmsg->fast_iov; | |
3642 | kmsg->msg.msg_iter.iov = kmsg->iov; | |
03b1230c | 3643 | } else { |
3529d8c2 JA |
3644 | struct io_sr_msg *sr = &req->sr_msg; |
3645 | ||
0b416c3e | 3646 | kmsg = &io.msg; |
b537916c | 3647 | kmsg->msg.msg_name = &io.msg.addr; |
3529d8c2 JA |
3648 | |
3649 | io.msg.iov = io.msg.fast_iov; | |
3650 | ret = sendmsg_copy_msghdr(&io.msg.msg, sr->msg, | |
3651 | sr->msg_flags, &io.msg.iov); | |
03b1230c | 3652 | if (ret) |
3529d8c2 | 3653 | return ret; |
03b1230c | 3654 | } |
0fa03c62 | 3655 | |
e47293fd JA |
3656 | flags = req->sr_msg.msg_flags; |
3657 | if (flags & MSG_DONTWAIT) | |
3658 | req->flags |= REQ_F_NOWAIT; | |
3659 | else if (force_nonblock) | |
3660 | flags |= MSG_DONTWAIT; | |
3661 | ||
0b416c3e | 3662 | ret = __sys_sendmsg_sock(sock, &kmsg->msg, flags); |
02d27d89 PB |
3663 | if (force_nonblock && ret == -EAGAIN) |
3664 | return io_setup_async_msg(req, kmsg); | |
441cdbd5 JA |
3665 | if (ret == -ERESTARTSYS) |
3666 | ret = -EINTR; | |
0fa03c62 JA |
3667 | } |
3668 | ||
1e95081c | 3669 | if (kmsg && kmsg->iov != kmsg->fast_iov) |
0b416c3e | 3670 | kfree(kmsg->iov); |
99bc4c38 | 3671 | req->flags &= ~REQ_F_NEED_CLEANUP; |
78e19bbe | 3672 | io_cqring_add_event(req, ret); |
4e88d6e7 JA |
3673 | if (ret < 0) |
3674 | req_set_fail_links(req); | |
014db007 | 3675 | io_put_req(req); |
5d17b4a4 | 3676 | return 0; |
03b1230c | 3677 | } |
aa1fa28f | 3678 | |
014db007 | 3679 | static int io_send(struct io_kiocb *req, bool force_nonblock) |
fddaface | 3680 | { |
fddaface JA |
3681 | struct socket *sock; |
3682 | int ret; | |
3683 | ||
3684 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) | |
3685 | return -EINVAL; | |
3686 | ||
3687 | sock = sock_from_file(req->file, &ret); | |
3688 | if (sock) { | |
3689 | struct io_sr_msg *sr = &req->sr_msg; | |
3690 | struct msghdr msg; | |
3691 | struct iovec iov; | |
3692 | unsigned flags; | |
3693 | ||
3694 | ret = import_single_range(WRITE, sr->buf, sr->len, &iov, | |
3695 | &msg.msg_iter); | |
3696 | if (ret) | |
3697 | return ret; | |
3698 | ||
3699 | msg.msg_name = NULL; | |
3700 | msg.msg_control = NULL; | |
3701 | msg.msg_controllen = 0; | |
3702 | msg.msg_namelen = 0; | |
3703 | ||
3704 | flags = req->sr_msg.msg_flags; | |
3705 | if (flags & MSG_DONTWAIT) | |
3706 | req->flags |= REQ_F_NOWAIT; | |
3707 | else if (force_nonblock) | |
3708 | flags |= MSG_DONTWAIT; | |
3709 | ||
0b7b21e4 JA |
3710 | msg.msg_flags = flags; |
3711 | ret = sock_sendmsg(sock, &msg); | |
fddaface JA |
3712 | if (force_nonblock && ret == -EAGAIN) |
3713 | return -EAGAIN; | |
3714 | if (ret == -ERESTARTSYS) | |
3715 | ret = -EINTR; | |
3716 | } | |
3717 | ||
3718 | io_cqring_add_event(req, ret); | |
3719 | if (ret < 0) | |
3720 | req_set_fail_links(req); | |
014db007 | 3721 | io_put_req(req); |
fddaface | 3722 | return 0; |
fddaface JA |
3723 | } |
3724 | ||
52de1fe1 JA |
3725 | static int __io_recvmsg_copy_hdr(struct io_kiocb *req, struct io_async_ctx *io) |
3726 | { | |
3727 | struct io_sr_msg *sr = &req->sr_msg; | |
3728 | struct iovec __user *uiov; | |
3729 | size_t iov_len; | |
3730 | int ret; | |
3731 | ||
3732 | ret = __copy_msghdr_from_user(&io->msg.msg, sr->msg, &io->msg.uaddr, | |
3733 | &uiov, &iov_len); | |
3734 | if (ret) | |
3735 | return ret; | |
3736 | ||
3737 | if (req->flags & REQ_F_BUFFER_SELECT) { | |
3738 | if (iov_len > 1) | |
3739 | return -EINVAL; | |
3740 | if (copy_from_user(io->msg.iov, uiov, sizeof(*uiov))) | |
3741 | return -EFAULT; | |
3742 | sr->len = io->msg.iov[0].iov_len; | |
3743 | iov_iter_init(&io->msg.msg.msg_iter, READ, io->msg.iov, 1, | |
3744 | sr->len); | |
3745 | io->msg.iov = NULL; | |
3746 | } else { | |
3747 | ret = import_iovec(READ, uiov, iov_len, UIO_FASTIOV, | |
3748 | &io->msg.iov, &io->msg.msg.msg_iter); | |
3749 | if (ret > 0) | |
3750 | ret = 0; | |
3751 | } | |
3752 | ||
3753 | return ret; | |
3754 | } | |
3755 | ||
3756 | #ifdef CONFIG_COMPAT | |
3757 | static int __io_compat_recvmsg_copy_hdr(struct io_kiocb *req, | |
3758 | struct io_async_ctx *io) | |
3759 | { | |
3760 | struct compat_msghdr __user *msg_compat; | |
3761 | struct io_sr_msg *sr = &req->sr_msg; | |
3762 | struct compat_iovec __user *uiov; | |
3763 | compat_uptr_t ptr; | |
3764 | compat_size_t len; | |
3765 | int ret; | |
3766 | ||
3767 | msg_compat = (struct compat_msghdr __user *) sr->msg; | |
3768 | ret = __get_compat_msghdr(&io->msg.msg, msg_compat, &io->msg.uaddr, | |
3769 | &ptr, &len); | |
3770 | if (ret) | |
3771 | return ret; | |
3772 | ||
3773 | uiov = compat_ptr(ptr); | |
3774 | if (req->flags & REQ_F_BUFFER_SELECT) { | |
3775 | compat_ssize_t clen; | |
3776 | ||
3777 | if (len > 1) | |
3778 | return -EINVAL; | |
3779 | if (!access_ok(uiov, sizeof(*uiov))) | |
3780 | return -EFAULT; | |
3781 | if (__get_user(clen, &uiov->iov_len)) | |
3782 | return -EFAULT; | |
3783 | if (clen < 0) | |
3784 | return -EINVAL; | |
3785 | sr->len = io->msg.iov[0].iov_len; | |
3786 | io->msg.iov = NULL; | |
3787 | } else { | |
3788 | ret = compat_import_iovec(READ, uiov, len, UIO_FASTIOV, | |
3789 | &io->msg.iov, | |
3790 | &io->msg.msg.msg_iter); | |
3791 | if (ret < 0) | |
3792 | return ret; | |
3793 | } | |
3794 | ||
3795 | return 0; | |
3796 | } | |
3797 | #endif | |
3798 | ||
3799 | static int io_recvmsg_copy_hdr(struct io_kiocb *req, struct io_async_ctx *io) | |
3800 | { | |
3801 | io->msg.iov = io->msg.fast_iov; | |
3802 | ||
3803 | #ifdef CONFIG_COMPAT | |
3804 | if (req->ctx->compat) | |
3805 | return __io_compat_recvmsg_copy_hdr(req, io); | |
fddaface | 3806 | #endif |
52de1fe1 JA |
3807 | |
3808 | return __io_recvmsg_copy_hdr(req, io); | |
3809 | } | |
3810 | ||
bcda7baa JA |
3811 | static struct io_buffer *io_recv_buffer_select(struct io_kiocb *req, |
3812 | int *cflags, bool needs_lock) | |
3813 | { | |
3814 | struct io_sr_msg *sr = &req->sr_msg; | |
3815 | struct io_buffer *kbuf; | |
3816 | ||
3817 | if (!(req->flags & REQ_F_BUFFER_SELECT)) | |
3818 | return NULL; | |
3819 | ||
3820 | kbuf = io_buffer_select(req, &sr->len, sr->bgid, sr->kbuf, needs_lock); | |
3821 | if (IS_ERR(kbuf)) | |
3822 | return kbuf; | |
3823 | ||
3824 | sr->kbuf = kbuf; | |
3825 | req->flags |= REQ_F_BUFFER_SELECTED; | |
3826 | ||
3827 | *cflags = kbuf->bid << IORING_CQE_BUFFER_SHIFT; | |
3828 | *cflags |= IORING_CQE_F_BUFFER; | |
3829 | return kbuf; | |
fddaface JA |
3830 | } |
3831 | ||
3529d8c2 JA |
3832 | static int io_recvmsg_prep(struct io_kiocb *req, |
3833 | const struct io_uring_sqe *sqe) | |
aa1fa28f | 3834 | { |
e47293fd | 3835 | struct io_sr_msg *sr = &req->sr_msg; |
3529d8c2 | 3836 | struct io_async_ctx *io = req->io; |
99bc4c38 | 3837 | int ret; |
3529d8c2 JA |
3838 | |
3839 | sr->msg_flags = READ_ONCE(sqe->msg_flags); | |
3840 | sr->msg = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
0b7b21e4 | 3841 | sr->len = READ_ONCE(sqe->len); |
bcda7baa | 3842 | sr->bgid = READ_ONCE(sqe->buf_group); |
06b76d44 | 3843 | |
d8768362 JA |
3844 | #ifdef CONFIG_COMPAT |
3845 | if (req->ctx->compat) | |
3846 | sr->msg_flags |= MSG_CMSG_COMPAT; | |
3847 | #endif | |
3848 | ||
fddaface | 3849 | if (!io || req->opcode == IORING_OP_RECV) |
06b76d44 | 3850 | return 0; |
5f798bea PB |
3851 | /* iovec is already imported */ |
3852 | if (req->flags & REQ_F_NEED_CLEANUP) | |
3853 | return 0; | |
03b1230c | 3854 | |
52de1fe1 | 3855 | ret = io_recvmsg_copy_hdr(req, io); |
99bc4c38 PB |
3856 | if (!ret) |
3857 | req->flags |= REQ_F_NEED_CLEANUP; | |
3858 | return ret; | |
aa1fa28f JA |
3859 | } |
3860 | ||
014db007 | 3861 | static int io_recvmsg(struct io_kiocb *req, bool force_nonblock) |
aa1fa28f | 3862 | { |
0b416c3e | 3863 | struct io_async_msghdr *kmsg = NULL; |
03b1230c | 3864 | struct socket *sock; |
52de1fe1 | 3865 | int ret, cflags = 0; |
03b1230c JA |
3866 | |
3867 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) | |
3868 | return -EINVAL; | |
3869 | ||
3870 | sock = sock_from_file(req->file, &ret); | |
3871 | if (sock) { | |
52de1fe1 | 3872 | struct io_buffer *kbuf; |
b7bb4f7d | 3873 | struct io_async_ctx io; |
03b1230c JA |
3874 | unsigned flags; |
3875 | ||
03b1230c | 3876 | if (req->io) { |
0b416c3e | 3877 | kmsg = &req->io->msg; |
b537916c | 3878 | kmsg->msg.msg_name = &req->io->msg.addr; |
0b416c3e JA |
3879 | /* if iov is set, it's allocated already */ |
3880 | if (!kmsg->iov) | |
3881 | kmsg->iov = kmsg->fast_iov; | |
3882 | kmsg->msg.msg_iter.iov = kmsg->iov; | |
03b1230c | 3883 | } else { |
0b416c3e | 3884 | kmsg = &io.msg; |
b537916c | 3885 | kmsg->msg.msg_name = &io.msg.addr; |
3529d8c2 | 3886 | |
52de1fe1 | 3887 | ret = io_recvmsg_copy_hdr(req, &io); |
03b1230c | 3888 | if (ret) |
3529d8c2 | 3889 | return ret; |
03b1230c JA |
3890 | } |
3891 | ||
52de1fe1 JA |
3892 | kbuf = io_recv_buffer_select(req, &cflags, !force_nonblock); |
3893 | if (IS_ERR(kbuf)) { | |
3894 | return PTR_ERR(kbuf); | |
3895 | } else if (kbuf) { | |
3896 | kmsg->fast_iov[0].iov_base = u64_to_user_ptr(kbuf->addr); | |
3897 | iov_iter_init(&kmsg->msg.msg_iter, READ, kmsg->iov, | |
3898 | 1, req->sr_msg.len); | |
3899 | } | |
3900 | ||
e47293fd JA |
3901 | flags = req->sr_msg.msg_flags; |
3902 | if (flags & MSG_DONTWAIT) | |
3903 | req->flags |= REQ_F_NOWAIT; | |
3904 | else if (force_nonblock) | |
3905 | flags |= MSG_DONTWAIT; | |
3906 | ||
3907 | ret = __sys_recvmsg_sock(sock, &kmsg->msg, req->sr_msg.msg, | |
3908 | kmsg->uaddr, flags); | |
02d27d89 PB |
3909 | if (force_nonblock && ret == -EAGAIN) |
3910 | return io_setup_async_msg(req, kmsg); | |
03b1230c JA |
3911 | if (ret == -ERESTARTSYS) |
3912 | ret = -EINTR; | |
3913 | } | |
3914 | ||
1e95081c | 3915 | if (kmsg && kmsg->iov != kmsg->fast_iov) |
0b416c3e | 3916 | kfree(kmsg->iov); |
99bc4c38 | 3917 | req->flags &= ~REQ_F_NEED_CLEANUP; |
52de1fe1 | 3918 | __io_cqring_add_event(req, ret, cflags); |
4e88d6e7 JA |
3919 | if (ret < 0) |
3920 | req_set_fail_links(req); | |
014db007 | 3921 | io_put_req(req); |
03b1230c | 3922 | return 0; |
0fa03c62 | 3923 | } |
5d17b4a4 | 3924 | |
014db007 | 3925 | static int io_recv(struct io_kiocb *req, bool force_nonblock) |
fddaface | 3926 | { |
bcda7baa | 3927 | struct io_buffer *kbuf = NULL; |
fddaface | 3928 | struct socket *sock; |
bcda7baa | 3929 | int ret, cflags = 0; |
fddaface JA |
3930 | |
3931 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) | |
3932 | return -EINVAL; | |
3933 | ||
3934 | sock = sock_from_file(req->file, &ret); | |
3935 | if (sock) { | |
3936 | struct io_sr_msg *sr = &req->sr_msg; | |
bcda7baa | 3937 | void __user *buf = sr->buf; |
fddaface JA |
3938 | struct msghdr msg; |
3939 | struct iovec iov; | |
3940 | unsigned flags; | |
3941 | ||
bcda7baa JA |
3942 | kbuf = io_recv_buffer_select(req, &cflags, !force_nonblock); |
3943 | if (IS_ERR(kbuf)) | |
3944 | return PTR_ERR(kbuf); | |
3945 | else if (kbuf) | |
3946 | buf = u64_to_user_ptr(kbuf->addr); | |
3947 | ||
3948 | ret = import_single_range(READ, buf, sr->len, &iov, | |
fddaface | 3949 | &msg.msg_iter); |
bcda7baa JA |
3950 | if (ret) { |
3951 | kfree(kbuf); | |
fddaface | 3952 | return ret; |
bcda7baa | 3953 | } |
fddaface | 3954 | |
bcda7baa | 3955 | req->flags |= REQ_F_NEED_CLEANUP; |
fddaface JA |
3956 | msg.msg_name = NULL; |
3957 | msg.msg_control = NULL; | |
3958 | msg.msg_controllen = 0; | |
3959 | msg.msg_namelen = 0; | |
3960 | msg.msg_iocb = NULL; | |
3961 | msg.msg_flags = 0; | |
3962 | ||
3963 | flags = req->sr_msg.msg_flags; | |
3964 | if (flags & MSG_DONTWAIT) | |
3965 | req->flags |= REQ_F_NOWAIT; | |
3966 | else if (force_nonblock) | |
3967 | flags |= MSG_DONTWAIT; | |
3968 | ||
0b7b21e4 | 3969 | ret = sock_recvmsg(sock, &msg, flags); |
fddaface JA |
3970 | if (force_nonblock && ret == -EAGAIN) |
3971 | return -EAGAIN; | |
3972 | if (ret == -ERESTARTSYS) | |
3973 | ret = -EINTR; | |
3974 | } | |
3975 | ||
bcda7baa JA |
3976 | kfree(kbuf); |
3977 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
3978 | __io_cqring_add_event(req, ret, cflags); | |
fddaface JA |
3979 | if (ret < 0) |
3980 | req_set_fail_links(req); | |
014db007 | 3981 | io_put_req(req); |
fddaface | 3982 | return 0; |
fddaface JA |
3983 | } |
3984 | ||
3529d8c2 | 3985 | static int io_accept_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
17f2fe35 | 3986 | { |
8ed8d3c3 JA |
3987 | struct io_accept *accept = &req->accept; |
3988 | ||
17f2fe35 JA |
3989 | if (unlikely(req->ctx->flags & (IORING_SETUP_IOPOLL|IORING_SETUP_SQPOLL))) |
3990 | return -EINVAL; | |
8042d6ce | 3991 | if (sqe->ioprio || sqe->len || sqe->buf_index) |
17f2fe35 JA |
3992 | return -EINVAL; |
3993 | ||
d55e5f5b JA |
3994 | accept->addr = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
3995 | accept->addr_len = u64_to_user_ptr(READ_ONCE(sqe->addr2)); | |
8ed8d3c3 | 3996 | accept->flags = READ_ONCE(sqe->accept_flags); |
09952e3e | 3997 | accept->nofile = rlimit(RLIMIT_NOFILE); |
8ed8d3c3 | 3998 | return 0; |
8ed8d3c3 | 3999 | } |
17f2fe35 | 4000 | |
014db007 | 4001 | static int __io_accept(struct io_kiocb *req, bool force_nonblock) |
8ed8d3c3 JA |
4002 | { |
4003 | struct io_accept *accept = &req->accept; | |
4004 | unsigned file_flags; | |
4005 | int ret; | |
4006 | ||
4007 | file_flags = force_nonblock ? O_NONBLOCK : 0; | |
4008 | ret = __sys_accept4_file(req->file, file_flags, accept->addr, | |
09952e3e JA |
4009 | accept->addr_len, accept->flags, |
4010 | accept->nofile); | |
8ed8d3c3 | 4011 | if (ret == -EAGAIN && force_nonblock) |
17f2fe35 | 4012 | return -EAGAIN; |
8e3cca12 JA |
4013 | if (ret == -ERESTARTSYS) |
4014 | ret = -EINTR; | |
4e88d6e7 JA |
4015 | if (ret < 0) |
4016 | req_set_fail_links(req); | |
78e19bbe | 4017 | io_cqring_add_event(req, ret); |
014db007 | 4018 | io_put_req(req); |
17f2fe35 | 4019 | return 0; |
8ed8d3c3 JA |
4020 | } |
4021 | ||
4022 | static void io_accept_finish(struct io_wq_work **workptr) | |
4023 | { | |
4024 | struct io_kiocb *req = container_of(*workptr, struct io_kiocb, work); | |
8ed8d3c3 JA |
4025 | |
4026 | if (io_req_cancelled(req)) | |
4027 | return; | |
014db007 | 4028 | __io_accept(req, false); |
e9fd9396 | 4029 | io_steal_work(req, workptr); |
8ed8d3c3 | 4030 | } |
8ed8d3c3 | 4031 | |
014db007 | 4032 | static int io_accept(struct io_kiocb *req, bool force_nonblock) |
8ed8d3c3 | 4033 | { |
8ed8d3c3 JA |
4034 | int ret; |
4035 | ||
014db007 | 4036 | ret = __io_accept(req, force_nonblock); |
8ed8d3c3 JA |
4037 | if (ret == -EAGAIN && force_nonblock) { |
4038 | req->work.func = io_accept_finish; | |
8ed8d3c3 JA |
4039 | return -EAGAIN; |
4040 | } | |
4041 | return 0; | |
0fa03c62 | 4042 | } |
5d17b4a4 | 4043 | |
3529d8c2 | 4044 | static int io_connect_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
f499a021 | 4045 | { |
3529d8c2 JA |
4046 | struct io_connect *conn = &req->connect; |
4047 | struct io_async_ctx *io = req->io; | |
f499a021 | 4048 | |
3fbb51c1 JA |
4049 | if (unlikely(req->ctx->flags & (IORING_SETUP_IOPOLL|IORING_SETUP_SQPOLL))) |
4050 | return -EINVAL; | |
4051 | if (sqe->ioprio || sqe->len || sqe->buf_index || sqe->rw_flags) | |
4052 | return -EINVAL; | |
4053 | ||
3529d8c2 JA |
4054 | conn->addr = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
4055 | conn->addr_len = READ_ONCE(sqe->addr2); | |
4056 | ||
4057 | if (!io) | |
4058 | return 0; | |
4059 | ||
4060 | return move_addr_to_kernel(conn->addr, conn->addr_len, | |
3fbb51c1 | 4061 | &io->connect.address); |
f499a021 JA |
4062 | } |
4063 | ||
014db007 | 4064 | static int io_connect(struct io_kiocb *req, bool force_nonblock) |
f8e85cf2 | 4065 | { |
f499a021 | 4066 | struct io_async_ctx __io, *io; |
f8e85cf2 | 4067 | unsigned file_flags; |
3fbb51c1 | 4068 | int ret; |
f8e85cf2 | 4069 | |
f499a021 JA |
4070 | if (req->io) { |
4071 | io = req->io; | |
4072 | } else { | |
3529d8c2 JA |
4073 | ret = move_addr_to_kernel(req->connect.addr, |
4074 | req->connect.addr_len, | |
4075 | &__io.connect.address); | |
f499a021 JA |
4076 | if (ret) |
4077 | goto out; | |
4078 | io = &__io; | |
4079 | } | |
4080 | ||
3fbb51c1 JA |
4081 | file_flags = force_nonblock ? O_NONBLOCK : 0; |
4082 | ||
4083 | ret = __sys_connect_file(req->file, &io->connect.address, | |
4084 | req->connect.addr_len, file_flags); | |
87f80d62 | 4085 | if ((ret == -EAGAIN || ret == -EINPROGRESS) && force_nonblock) { |
b7bb4f7d JA |
4086 | if (req->io) |
4087 | return -EAGAIN; | |
4088 | if (io_alloc_async_ctx(req)) { | |
f499a021 JA |
4089 | ret = -ENOMEM; |
4090 | goto out; | |
4091 | } | |
b7bb4f7d | 4092 | memcpy(&req->io->connect, &__io.connect, sizeof(__io.connect)); |
f8e85cf2 | 4093 | return -EAGAIN; |
f499a021 | 4094 | } |
f8e85cf2 JA |
4095 | if (ret == -ERESTARTSYS) |
4096 | ret = -EINTR; | |
f499a021 | 4097 | out: |
4e88d6e7 JA |
4098 | if (ret < 0) |
4099 | req_set_fail_links(req); | |
f8e85cf2 | 4100 | io_cqring_add_event(req, ret); |
014db007 | 4101 | io_put_req(req); |
f8e85cf2 | 4102 | return 0; |
469956e8 Y |
4103 | } |
4104 | #else /* !CONFIG_NET */ | |
4105 | static int io_sendmsg_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) | |
4106 | { | |
f8e85cf2 | 4107 | return -EOPNOTSUPP; |
f8e85cf2 JA |
4108 | } |
4109 | ||
469956e8 Y |
4110 | static int io_sendmsg(struct io_kiocb *req, bool force_nonblock) |
4111 | { | |
4112 | return -EOPNOTSUPP; | |
4113 | } | |
4114 | ||
4115 | static int io_send(struct io_kiocb *req, bool force_nonblock) | |
4116 | { | |
4117 | return -EOPNOTSUPP; | |
4118 | } | |
4119 | ||
4120 | static int io_recvmsg_prep(struct io_kiocb *req, | |
4121 | const struct io_uring_sqe *sqe) | |
4122 | { | |
4123 | return -EOPNOTSUPP; | |
4124 | } | |
4125 | ||
4126 | static int io_recvmsg(struct io_kiocb *req, bool force_nonblock) | |
4127 | { | |
4128 | return -EOPNOTSUPP; | |
4129 | } | |
4130 | ||
4131 | static int io_recv(struct io_kiocb *req, bool force_nonblock) | |
4132 | { | |
4133 | return -EOPNOTSUPP; | |
4134 | } | |
4135 | ||
4136 | static int io_accept_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) | |
4137 | { | |
4138 | return -EOPNOTSUPP; | |
4139 | } | |
4140 | ||
4141 | static int io_accept(struct io_kiocb *req, bool force_nonblock) | |
4142 | { | |
4143 | return -EOPNOTSUPP; | |
4144 | } | |
4145 | ||
4146 | static int io_connect_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) | |
4147 | { | |
4148 | return -EOPNOTSUPP; | |
4149 | } | |
4150 | ||
4151 | static int io_connect(struct io_kiocb *req, bool force_nonblock) | |
4152 | { | |
f8e85cf2 | 4153 | return -EOPNOTSUPP; |
f8e85cf2 | 4154 | } |
469956e8 | 4155 | #endif /* CONFIG_NET */ |
f8e85cf2 | 4156 | |
d7718a9d JA |
4157 | struct io_poll_table { |
4158 | struct poll_table_struct pt; | |
4159 | struct io_kiocb *req; | |
4160 | int error; | |
4161 | }; | |
4162 | ||
d7718a9d JA |
4163 | static int __io_async_wake(struct io_kiocb *req, struct io_poll_iocb *poll, |
4164 | __poll_t mask, task_work_func_t func) | |
4165 | { | |
4166 | struct task_struct *tsk; | |
aa96bf8a | 4167 | int ret; |
d7718a9d JA |
4168 | |
4169 | /* for instances that support it check for an event match first: */ | |
4170 | if (mask && !(mask & poll->events)) | |
4171 | return 0; | |
4172 | ||
4173 | trace_io_uring_task_add(req->ctx, req->opcode, req->user_data, mask); | |
4174 | ||
4175 | list_del_init(&poll->wait.entry); | |
4176 | ||
4177 | tsk = req->task; | |
4178 | req->result = mask; | |
4179 | init_task_work(&req->task_work, func); | |
4180 | /* | |
aa96bf8a JA |
4181 | * If this fails, then the task is exiting. Punt to one of the io-wq |
4182 | * threads to ensure the work gets run, we can't always rely on exit | |
4183 | * cancelation taking care of this. | |
d7718a9d | 4184 | */ |
aa96bf8a JA |
4185 | ret = task_work_add(tsk, &req->task_work, true); |
4186 | if (unlikely(ret)) { | |
4187 | tsk = io_wq_get_task(req->ctx->io_wq); | |
4188 | task_work_add(tsk, &req->task_work, true); | |
4189 | } | |
d7718a9d JA |
4190 | wake_up_process(tsk); |
4191 | return 1; | |
4192 | } | |
4193 | ||
74ce6ce4 JA |
4194 | static bool io_poll_rewait(struct io_kiocb *req, struct io_poll_iocb *poll) |
4195 | __acquires(&req->ctx->completion_lock) | |
4196 | { | |
4197 | struct io_ring_ctx *ctx = req->ctx; | |
4198 | ||
4199 | if (!req->result && !READ_ONCE(poll->canceled)) { | |
4200 | struct poll_table_struct pt = { ._key = poll->events }; | |
4201 | ||
4202 | req->result = vfs_poll(req->file, &pt) & poll->events; | |
4203 | } | |
4204 | ||
4205 | spin_lock_irq(&ctx->completion_lock); | |
4206 | if (!req->result && !READ_ONCE(poll->canceled)) { | |
4207 | add_wait_queue(poll->head, &poll->wait); | |
4208 | return true; | |
4209 | } | |
4210 | ||
4211 | return false; | |
4212 | } | |
4213 | ||
18bceab1 JA |
4214 | static void io_poll_remove_double(struct io_kiocb *req) |
4215 | { | |
4216 | struct io_poll_iocb *poll = (struct io_poll_iocb *) req->io; | |
4217 | ||
4218 | lockdep_assert_held(&req->ctx->completion_lock); | |
4219 | ||
4220 | if (poll && poll->head) { | |
4221 | struct wait_queue_head *head = poll->head; | |
4222 | ||
4223 | spin_lock(&head->lock); | |
4224 | list_del_init(&poll->wait.entry); | |
4225 | if (poll->wait.private) | |
4226 | refcount_dec(&req->refs); | |
4227 | poll->head = NULL; | |
4228 | spin_unlock(&head->lock); | |
4229 | } | |
4230 | } | |
4231 | ||
4232 | static void io_poll_complete(struct io_kiocb *req, __poll_t mask, int error) | |
4233 | { | |
4234 | struct io_ring_ctx *ctx = req->ctx; | |
4235 | ||
4236 | io_poll_remove_double(req); | |
4237 | req->poll.done = true; | |
4238 | io_cqring_fill_event(req, error ? error : mangle_poll(mask)); | |
4239 | io_commit_cqring(ctx); | |
4240 | } | |
4241 | ||
4242 | static void io_poll_task_handler(struct io_kiocb *req, struct io_kiocb **nxt) | |
4243 | { | |
4244 | struct io_ring_ctx *ctx = req->ctx; | |
4245 | ||
4246 | if (io_poll_rewait(req, &req->poll)) { | |
4247 | spin_unlock_irq(&ctx->completion_lock); | |
4248 | return; | |
4249 | } | |
4250 | ||
4251 | hash_del(&req->hash_node); | |
4252 | io_poll_complete(req, req->result, 0); | |
4253 | req->flags |= REQ_F_COMP_LOCKED; | |
4254 | io_put_req_find_next(req, nxt); | |
4255 | spin_unlock_irq(&ctx->completion_lock); | |
4256 | ||
4257 | io_cqring_ev_posted(ctx); | |
4258 | } | |
4259 | ||
4260 | static void io_poll_task_func(struct callback_head *cb) | |
4261 | { | |
4262 | struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work); | |
4263 | struct io_kiocb *nxt = NULL; | |
4264 | ||
4265 | io_poll_task_handler(req, &nxt); | |
4266 | if (nxt) { | |
4267 | struct io_ring_ctx *ctx = nxt->ctx; | |
4268 | ||
4269 | mutex_lock(&ctx->uring_lock); | |
4270 | __io_queue_sqe(nxt, NULL); | |
4271 | mutex_unlock(&ctx->uring_lock); | |
4272 | } | |
4273 | } | |
4274 | ||
4275 | static int io_poll_double_wake(struct wait_queue_entry *wait, unsigned mode, | |
4276 | int sync, void *key) | |
4277 | { | |
4278 | struct io_kiocb *req = wait->private; | |
4279 | struct io_poll_iocb *poll = (struct io_poll_iocb *) req->io; | |
4280 | __poll_t mask = key_to_poll(key); | |
4281 | ||
4282 | /* for instances that support it check for an event match first: */ | |
4283 | if (mask && !(mask & poll->events)) | |
4284 | return 0; | |
4285 | ||
4286 | if (req->poll.head) { | |
4287 | bool done; | |
4288 | ||
4289 | spin_lock(&req->poll.head->lock); | |
4290 | done = list_empty(&req->poll.wait.entry); | |
4291 | if (!done) | |
4292 | list_del_init(&req->poll.wait.entry); | |
4293 | spin_unlock(&req->poll.head->lock); | |
4294 | if (!done) | |
4295 | __io_async_wake(req, poll, mask, io_poll_task_func); | |
4296 | } | |
4297 | refcount_dec(&req->refs); | |
4298 | return 1; | |
4299 | } | |
4300 | ||
4301 | static void io_init_poll_iocb(struct io_poll_iocb *poll, __poll_t events, | |
4302 | wait_queue_func_t wake_func) | |
4303 | { | |
4304 | poll->head = NULL; | |
4305 | poll->done = false; | |
4306 | poll->canceled = false; | |
4307 | poll->events = events; | |
4308 | INIT_LIST_HEAD(&poll->wait.entry); | |
4309 | init_waitqueue_func_entry(&poll->wait, wake_func); | |
4310 | } | |
4311 | ||
4312 | static void __io_queue_proc(struct io_poll_iocb *poll, struct io_poll_table *pt, | |
4313 | struct wait_queue_head *head) | |
4314 | { | |
4315 | struct io_kiocb *req = pt->req; | |
4316 | ||
4317 | /* | |
4318 | * If poll->head is already set, it's because the file being polled | |
4319 | * uses multiple waitqueues for poll handling (eg one for read, one | |
4320 | * for write). Setup a separate io_poll_iocb if this happens. | |
4321 | */ | |
4322 | if (unlikely(poll->head)) { | |
4323 | /* already have a 2nd entry, fail a third attempt */ | |
4324 | if (req->io) { | |
4325 | pt->error = -EINVAL; | |
4326 | return; | |
4327 | } | |
4328 | poll = kmalloc(sizeof(*poll), GFP_ATOMIC); | |
4329 | if (!poll) { | |
4330 | pt->error = -ENOMEM; | |
4331 | return; | |
4332 | } | |
4333 | io_init_poll_iocb(poll, req->poll.events, io_poll_double_wake); | |
4334 | refcount_inc(&req->refs); | |
4335 | poll->wait.private = req; | |
4336 | req->io = (void *) poll; | |
4337 | } | |
4338 | ||
4339 | pt->error = 0; | |
4340 | poll->head = head; | |
4341 | add_wait_queue(head, &poll->wait); | |
4342 | } | |
4343 | ||
4344 | static void io_async_queue_proc(struct file *file, struct wait_queue_head *head, | |
4345 | struct poll_table_struct *p) | |
4346 | { | |
4347 | struct io_poll_table *pt = container_of(p, struct io_poll_table, pt); | |
4348 | ||
4349 | __io_queue_proc(&pt->req->apoll->poll, pt, head); | |
4350 | } | |
4351 | ||
d7718a9d JA |
4352 | static void io_async_task_func(struct callback_head *cb) |
4353 | { | |
4354 | struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work); | |
4355 | struct async_poll *apoll = req->apoll; | |
4356 | struct io_ring_ctx *ctx = req->ctx; | |
31067255 | 4357 | bool canceled = false; |
d7718a9d JA |
4358 | |
4359 | trace_io_uring_task_run(req->ctx, req->opcode, req->user_data); | |
4360 | ||
74ce6ce4 | 4361 | if (io_poll_rewait(req, &apoll->poll)) { |
d7718a9d | 4362 | spin_unlock_irq(&ctx->completion_lock); |
74ce6ce4 | 4363 | return; |
d7718a9d JA |
4364 | } |
4365 | ||
31067255 JA |
4366 | /* If req is still hashed, it cannot have been canceled. Don't check. */ |
4367 | if (hash_hashed(&req->hash_node)) { | |
74ce6ce4 | 4368 | hash_del(&req->hash_node); |
31067255 JA |
4369 | } else { |
4370 | canceled = READ_ONCE(apoll->poll.canceled); | |
4371 | if (canceled) { | |
4372 | io_cqring_fill_event(req, -ECANCELED); | |
4373 | io_commit_cqring(ctx); | |
4374 | } | |
2bae047e JA |
4375 | } |
4376 | ||
74ce6ce4 JA |
4377 | spin_unlock_irq(&ctx->completion_lock); |
4378 | ||
44575a67 XW |
4379 | /* restore ->work in case we need to retry again */ |
4380 | memcpy(&req->work, &apoll->work, sizeof(req->work)); | |
31067255 | 4381 | kfree(apoll); |
44575a67 | 4382 | |
31067255 JA |
4383 | if (!canceled) { |
4384 | __set_current_state(TASK_RUNNING); | |
4385 | mutex_lock(&ctx->uring_lock); | |
4386 | __io_queue_sqe(req, NULL); | |
4387 | mutex_unlock(&ctx->uring_lock); | |
4388 | } else { | |
2bae047e JA |
4389 | io_cqring_ev_posted(ctx); |
4390 | req_set_fail_links(req); | |
44575a67 | 4391 | io_double_put_req(req); |
2bae047e | 4392 | } |
d7718a9d JA |
4393 | } |
4394 | ||
4395 | static int io_async_wake(struct wait_queue_entry *wait, unsigned mode, int sync, | |
4396 | void *key) | |
4397 | { | |
4398 | struct io_kiocb *req = wait->private; | |
4399 | struct io_poll_iocb *poll = &req->apoll->poll; | |
4400 | ||
4401 | trace_io_uring_poll_wake(req->ctx, req->opcode, req->user_data, | |
4402 | key_to_poll(key)); | |
4403 | ||
4404 | return __io_async_wake(req, poll, key_to_poll(key), io_async_task_func); | |
4405 | } | |
4406 | ||
4407 | static void io_poll_req_insert(struct io_kiocb *req) | |
4408 | { | |
4409 | struct io_ring_ctx *ctx = req->ctx; | |
4410 | struct hlist_head *list; | |
4411 | ||
4412 | list = &ctx->cancel_hash[hash_long(req->user_data, ctx->cancel_hash_bits)]; | |
4413 | hlist_add_head(&req->hash_node, list); | |
4414 | } | |
4415 | ||
4416 | static __poll_t __io_arm_poll_handler(struct io_kiocb *req, | |
4417 | struct io_poll_iocb *poll, | |
4418 | struct io_poll_table *ipt, __poll_t mask, | |
4419 | wait_queue_func_t wake_func) | |
4420 | __acquires(&ctx->completion_lock) | |
4421 | { | |
4422 | struct io_ring_ctx *ctx = req->ctx; | |
4423 | bool cancel = false; | |
4424 | ||
4425 | poll->file = req->file; | |
18bceab1 JA |
4426 | io_init_poll_iocb(poll, mask, wake_func); |
4427 | poll->wait.private = req; | |
d7718a9d JA |
4428 | |
4429 | ipt->pt._key = mask; | |
4430 | ipt->req = req; | |
4431 | ipt->error = -EINVAL; | |
4432 | ||
d7718a9d JA |
4433 | mask = vfs_poll(req->file, &ipt->pt) & poll->events; |
4434 | ||
4435 | spin_lock_irq(&ctx->completion_lock); | |
4436 | if (likely(poll->head)) { | |
4437 | spin_lock(&poll->head->lock); | |
4438 | if (unlikely(list_empty(&poll->wait.entry))) { | |
4439 | if (ipt->error) | |
4440 | cancel = true; | |
4441 | ipt->error = 0; | |
4442 | mask = 0; | |
4443 | } | |
4444 | if (mask || ipt->error) | |
4445 | list_del_init(&poll->wait.entry); | |
4446 | else if (cancel) | |
4447 | WRITE_ONCE(poll->canceled, true); | |
4448 | else if (!poll->done) /* actually waiting for an event */ | |
4449 | io_poll_req_insert(req); | |
4450 | spin_unlock(&poll->head->lock); | |
4451 | } | |
4452 | ||
4453 | return mask; | |
4454 | } | |
4455 | ||
4456 | static bool io_arm_poll_handler(struct io_kiocb *req) | |
4457 | { | |
4458 | const struct io_op_def *def = &io_op_defs[req->opcode]; | |
4459 | struct io_ring_ctx *ctx = req->ctx; | |
4460 | struct async_poll *apoll; | |
4461 | struct io_poll_table ipt; | |
4462 | __poll_t mask, ret; | |
18bceab1 | 4463 | bool had_io; |
d7718a9d JA |
4464 | |
4465 | if (!req->file || !file_can_poll(req->file)) | |
4466 | return false; | |
4467 | if (req->flags & (REQ_F_MUST_PUNT | REQ_F_POLLED)) | |
4468 | return false; | |
4469 | if (!def->pollin && !def->pollout) | |
4470 | return false; | |
4471 | ||
4472 | apoll = kmalloc(sizeof(*apoll), GFP_ATOMIC); | |
4473 | if (unlikely(!apoll)) | |
4474 | return false; | |
4475 | ||
4476 | req->flags |= REQ_F_POLLED; | |
4477 | memcpy(&apoll->work, &req->work, sizeof(req->work)); | |
18bceab1 | 4478 | had_io = req->io != NULL; |
d7718a9d | 4479 | |
3537b6a7 | 4480 | get_task_struct(current); |
d7718a9d JA |
4481 | req->task = current; |
4482 | req->apoll = apoll; | |
4483 | INIT_HLIST_NODE(&req->hash_node); | |
4484 | ||
8755d97a | 4485 | mask = 0; |
d7718a9d | 4486 | if (def->pollin) |
8755d97a | 4487 | mask |= POLLIN | POLLRDNORM; |
d7718a9d JA |
4488 | if (def->pollout) |
4489 | mask |= POLLOUT | POLLWRNORM; | |
4490 | mask |= POLLERR | POLLPRI; | |
4491 | ||
4492 | ipt.pt._qproc = io_async_queue_proc; | |
4493 | ||
4494 | ret = __io_arm_poll_handler(req, &apoll->poll, &ipt, mask, | |
4495 | io_async_wake); | |
4496 | if (ret) { | |
4497 | ipt.error = 0; | |
18bceab1 JA |
4498 | /* only remove double add if we did it here */ |
4499 | if (!had_io) | |
4500 | io_poll_remove_double(req); | |
d7718a9d JA |
4501 | spin_unlock_irq(&ctx->completion_lock); |
4502 | memcpy(&req->work, &apoll->work, sizeof(req->work)); | |
4503 | kfree(apoll); | |
4504 | return false; | |
4505 | } | |
4506 | spin_unlock_irq(&ctx->completion_lock); | |
4507 | trace_io_uring_poll_arm(ctx, req->opcode, req->user_data, mask, | |
4508 | apoll->poll.events); | |
4509 | return true; | |
4510 | } | |
4511 | ||
4512 | static bool __io_poll_remove_one(struct io_kiocb *req, | |
4513 | struct io_poll_iocb *poll) | |
221c5eb2 | 4514 | { |
b41e9852 | 4515 | bool do_complete = false; |
221c5eb2 JA |
4516 | |
4517 | spin_lock(&poll->head->lock); | |
4518 | WRITE_ONCE(poll->canceled, true); | |
392edb45 JA |
4519 | if (!list_empty(&poll->wait.entry)) { |
4520 | list_del_init(&poll->wait.entry); | |
b41e9852 | 4521 | do_complete = true; |
221c5eb2 JA |
4522 | } |
4523 | spin_unlock(&poll->head->lock); | |
3bfa5bcb | 4524 | hash_del(&req->hash_node); |
d7718a9d JA |
4525 | return do_complete; |
4526 | } | |
4527 | ||
4528 | static bool io_poll_remove_one(struct io_kiocb *req) | |
4529 | { | |
4530 | bool do_complete; | |
4531 | ||
4532 | if (req->opcode == IORING_OP_POLL_ADD) { | |
18bceab1 | 4533 | io_poll_remove_double(req); |
d7718a9d JA |
4534 | do_complete = __io_poll_remove_one(req, &req->poll); |
4535 | } else { | |
3bfa5bcb JA |
4536 | struct async_poll *apoll = req->apoll; |
4537 | ||
d7718a9d | 4538 | /* non-poll requests have submit ref still */ |
3bfa5bcb JA |
4539 | do_complete = __io_poll_remove_one(req, &apoll->poll); |
4540 | if (do_complete) { | |
d7718a9d | 4541 | io_put_req(req); |
3bfa5bcb JA |
4542 | /* |
4543 | * restore ->work because we will call | |
4544 | * io_req_work_drop_env below when dropping the | |
4545 | * final reference. | |
4546 | */ | |
4547 | memcpy(&req->work, &apoll->work, sizeof(req->work)); | |
4548 | kfree(apoll); | |
4549 | } | |
b1f573bd XW |
4550 | } |
4551 | ||
b41e9852 JA |
4552 | if (do_complete) { |
4553 | io_cqring_fill_event(req, -ECANCELED); | |
4554 | io_commit_cqring(req->ctx); | |
4555 | req->flags |= REQ_F_COMP_LOCKED; | |
4556 | io_put_req(req); | |
4557 | } | |
4558 | ||
4559 | return do_complete; | |
221c5eb2 JA |
4560 | } |
4561 | ||
4562 | static void io_poll_remove_all(struct io_ring_ctx *ctx) | |
4563 | { | |
78076bb6 | 4564 | struct hlist_node *tmp; |
221c5eb2 | 4565 | struct io_kiocb *req; |
8e2e1faf | 4566 | int posted = 0, i; |
221c5eb2 JA |
4567 | |
4568 | spin_lock_irq(&ctx->completion_lock); | |
78076bb6 JA |
4569 | for (i = 0; i < (1U << ctx->cancel_hash_bits); i++) { |
4570 | struct hlist_head *list; | |
4571 | ||
4572 | list = &ctx->cancel_hash[i]; | |
4573 | hlist_for_each_entry_safe(req, tmp, list, hash_node) | |
8e2e1faf | 4574 | posted += io_poll_remove_one(req); |
221c5eb2 JA |
4575 | } |
4576 | spin_unlock_irq(&ctx->completion_lock); | |
b41e9852 | 4577 | |
8e2e1faf JA |
4578 | if (posted) |
4579 | io_cqring_ev_posted(ctx); | |
221c5eb2 JA |
4580 | } |
4581 | ||
47f46768 JA |
4582 | static int io_poll_cancel(struct io_ring_ctx *ctx, __u64 sqe_addr) |
4583 | { | |
78076bb6 | 4584 | struct hlist_head *list; |
47f46768 JA |
4585 | struct io_kiocb *req; |
4586 | ||
78076bb6 JA |
4587 | list = &ctx->cancel_hash[hash_long(sqe_addr, ctx->cancel_hash_bits)]; |
4588 | hlist_for_each_entry(req, list, hash_node) { | |
b41e9852 JA |
4589 | if (sqe_addr != req->user_data) |
4590 | continue; | |
4591 | if (io_poll_remove_one(req)) | |
eac406c6 | 4592 | return 0; |
b41e9852 | 4593 | return -EALREADY; |
47f46768 JA |
4594 | } |
4595 | ||
4596 | return -ENOENT; | |
4597 | } | |
4598 | ||
3529d8c2 JA |
4599 | static int io_poll_remove_prep(struct io_kiocb *req, |
4600 | const struct io_uring_sqe *sqe) | |
0969e783 | 4601 | { |
0969e783 JA |
4602 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4603 | return -EINVAL; | |
4604 | if (sqe->ioprio || sqe->off || sqe->len || sqe->buf_index || | |
4605 | sqe->poll_events) | |
4606 | return -EINVAL; | |
4607 | ||
4608 | req->poll.addr = READ_ONCE(sqe->addr); | |
0969e783 JA |
4609 | return 0; |
4610 | } | |
4611 | ||
221c5eb2 JA |
4612 | /* |
4613 | * Find a running poll command that matches one specified in sqe->addr, | |
4614 | * and remove it if found. | |
4615 | */ | |
fc4df999 | 4616 | static int io_poll_remove(struct io_kiocb *req) |
221c5eb2 JA |
4617 | { |
4618 | struct io_ring_ctx *ctx = req->ctx; | |
0969e783 | 4619 | u64 addr; |
47f46768 | 4620 | int ret; |
221c5eb2 | 4621 | |
0969e783 | 4622 | addr = req->poll.addr; |
221c5eb2 | 4623 | spin_lock_irq(&ctx->completion_lock); |
0969e783 | 4624 | ret = io_poll_cancel(ctx, addr); |
221c5eb2 JA |
4625 | spin_unlock_irq(&ctx->completion_lock); |
4626 | ||
78e19bbe | 4627 | io_cqring_add_event(req, ret); |
4e88d6e7 JA |
4628 | if (ret < 0) |
4629 | req_set_fail_links(req); | |
e65ef56d | 4630 | io_put_req(req); |
221c5eb2 JA |
4631 | return 0; |
4632 | } | |
4633 | ||
221c5eb2 JA |
4634 | static int io_poll_wake(struct wait_queue_entry *wait, unsigned mode, int sync, |
4635 | void *key) | |
4636 | { | |
c2f2eb7d JA |
4637 | struct io_kiocb *req = wait->private; |
4638 | struct io_poll_iocb *poll = &req->poll; | |
221c5eb2 | 4639 | |
d7718a9d | 4640 | return __io_async_wake(req, poll, key_to_poll(key), io_poll_task_func); |
221c5eb2 JA |
4641 | } |
4642 | ||
221c5eb2 JA |
4643 | static void io_poll_queue_proc(struct file *file, struct wait_queue_head *head, |
4644 | struct poll_table_struct *p) | |
4645 | { | |
4646 | struct io_poll_table *pt = container_of(p, struct io_poll_table, pt); | |
4647 | ||
d7718a9d | 4648 | __io_queue_proc(&pt->req->poll, pt, head); |
eac406c6 JA |
4649 | } |
4650 | ||
3529d8c2 | 4651 | static int io_poll_add_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
221c5eb2 JA |
4652 | { |
4653 | struct io_poll_iocb *poll = &req->poll; | |
221c5eb2 | 4654 | u16 events; |
221c5eb2 JA |
4655 | |
4656 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) | |
4657 | return -EINVAL; | |
4658 | if (sqe->addr || sqe->ioprio || sqe->off || sqe->len || sqe->buf_index) | |
4659 | return -EINVAL; | |
09bb8394 JA |
4660 | if (!poll->file) |
4661 | return -EBADF; | |
221c5eb2 | 4662 | |
221c5eb2 JA |
4663 | events = READ_ONCE(sqe->poll_events); |
4664 | poll->events = demangle_poll(events) | EPOLLERR | EPOLLHUP; | |
b41e9852 | 4665 | |
3537b6a7 | 4666 | get_task_struct(current); |
b41e9852 | 4667 | req->task = current; |
0969e783 JA |
4668 | return 0; |
4669 | } | |
4670 | ||
014db007 | 4671 | static int io_poll_add(struct io_kiocb *req) |
0969e783 JA |
4672 | { |
4673 | struct io_poll_iocb *poll = &req->poll; | |
4674 | struct io_ring_ctx *ctx = req->ctx; | |
4675 | struct io_poll_table ipt; | |
0969e783 | 4676 | __poll_t mask; |
0969e783 | 4677 | |
78076bb6 | 4678 | INIT_HLIST_NODE(&req->hash_node); |
36703247 | 4679 | INIT_LIST_HEAD(&req->list); |
d7718a9d | 4680 | ipt.pt._qproc = io_poll_queue_proc; |
36703247 | 4681 | |
d7718a9d JA |
4682 | mask = __io_arm_poll_handler(req, &req->poll, &ipt, poll->events, |
4683 | io_poll_wake); | |
221c5eb2 | 4684 | |
8c838788 | 4685 | if (mask) { /* no async, we'd stolen it */ |
221c5eb2 | 4686 | ipt.error = 0; |
b0dd8a41 | 4687 | io_poll_complete(req, mask, 0); |
221c5eb2 | 4688 | } |
221c5eb2 JA |
4689 | spin_unlock_irq(&ctx->completion_lock); |
4690 | ||
8c838788 JA |
4691 | if (mask) { |
4692 | io_cqring_ev_posted(ctx); | |
014db007 | 4693 | io_put_req(req); |
221c5eb2 | 4694 | } |
8c838788 | 4695 | return ipt.error; |
221c5eb2 JA |
4696 | } |
4697 | ||
5262f567 JA |
4698 | static enum hrtimer_restart io_timeout_fn(struct hrtimer *timer) |
4699 | { | |
ad8a48ac JA |
4700 | struct io_timeout_data *data = container_of(timer, |
4701 | struct io_timeout_data, timer); | |
4702 | struct io_kiocb *req = data->req; | |
4703 | struct io_ring_ctx *ctx = req->ctx; | |
5262f567 JA |
4704 | unsigned long flags; |
4705 | ||
5262f567 JA |
4706 | atomic_inc(&ctx->cq_timeouts); |
4707 | ||
4708 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
ef03681a | 4709 | /* |
11365043 JA |
4710 | * We could be racing with timeout deletion. If the list is empty, |
4711 | * then timeout lookup already found it and will be handling it. | |
ef03681a | 4712 | */ |
842f9612 | 4713 | if (!list_empty(&req->list)) { |
11365043 | 4714 | struct io_kiocb *prev; |
5262f567 | 4715 | |
11365043 JA |
4716 | /* |
4717 | * Adjust the reqs sequence before the current one because it | |
d195a66e | 4718 | * will consume a slot in the cq_ring and the cq_tail |
11365043 JA |
4719 | * pointer will be increased, otherwise other timeout reqs may |
4720 | * return in advance without waiting for enough wait_nr. | |
4721 | */ | |
4722 | prev = req; | |
4723 | list_for_each_entry_continue_reverse(prev, &ctx->timeout_list, list) | |
4724 | prev->sequence++; | |
11365043 | 4725 | list_del_init(&req->list); |
11365043 | 4726 | } |
5262f567 | 4727 | |
78e19bbe | 4728 | io_cqring_fill_event(req, -ETIME); |
5262f567 JA |
4729 | io_commit_cqring(ctx); |
4730 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
4731 | ||
4732 | io_cqring_ev_posted(ctx); | |
4e88d6e7 | 4733 | req_set_fail_links(req); |
5262f567 JA |
4734 | io_put_req(req); |
4735 | return HRTIMER_NORESTART; | |
4736 | } | |
4737 | ||
47f46768 JA |
4738 | static int io_timeout_cancel(struct io_ring_ctx *ctx, __u64 user_data) |
4739 | { | |
4740 | struct io_kiocb *req; | |
4741 | int ret = -ENOENT; | |
4742 | ||
4743 | list_for_each_entry(req, &ctx->timeout_list, list) { | |
4744 | if (user_data == req->user_data) { | |
4745 | list_del_init(&req->list); | |
4746 | ret = 0; | |
4747 | break; | |
4748 | } | |
4749 | } | |
4750 | ||
4751 | if (ret == -ENOENT) | |
4752 | return ret; | |
4753 | ||
2d28390a | 4754 | ret = hrtimer_try_to_cancel(&req->io->timeout.timer); |
47f46768 JA |
4755 | if (ret == -1) |
4756 | return -EALREADY; | |
4757 | ||
4e88d6e7 | 4758 | req_set_fail_links(req); |
47f46768 JA |
4759 | io_cqring_fill_event(req, -ECANCELED); |
4760 | io_put_req(req); | |
4761 | return 0; | |
4762 | } | |
4763 | ||
3529d8c2 JA |
4764 | static int io_timeout_remove_prep(struct io_kiocb *req, |
4765 | const struct io_uring_sqe *sqe) | |
b29472ee | 4766 | { |
b29472ee JA |
4767 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4768 | return -EINVAL; | |
4769 | if (sqe->flags || sqe->ioprio || sqe->buf_index || sqe->len) | |
4770 | return -EINVAL; | |
4771 | ||
4772 | req->timeout.addr = READ_ONCE(sqe->addr); | |
4773 | req->timeout.flags = READ_ONCE(sqe->timeout_flags); | |
4774 | if (req->timeout.flags) | |
4775 | return -EINVAL; | |
4776 | ||
b29472ee JA |
4777 | return 0; |
4778 | } | |
4779 | ||
11365043 JA |
4780 | /* |
4781 | * Remove or update an existing timeout command | |
4782 | */ | |
fc4df999 | 4783 | static int io_timeout_remove(struct io_kiocb *req) |
11365043 JA |
4784 | { |
4785 | struct io_ring_ctx *ctx = req->ctx; | |
47f46768 | 4786 | int ret; |
11365043 | 4787 | |
11365043 | 4788 | spin_lock_irq(&ctx->completion_lock); |
b29472ee | 4789 | ret = io_timeout_cancel(ctx, req->timeout.addr); |
11365043 | 4790 | |
47f46768 | 4791 | io_cqring_fill_event(req, ret); |
11365043 JA |
4792 | io_commit_cqring(ctx); |
4793 | spin_unlock_irq(&ctx->completion_lock); | |
5262f567 | 4794 | io_cqring_ev_posted(ctx); |
4e88d6e7 JA |
4795 | if (ret < 0) |
4796 | req_set_fail_links(req); | |
ec9c02ad | 4797 | io_put_req(req); |
11365043 | 4798 | return 0; |
5262f567 JA |
4799 | } |
4800 | ||
3529d8c2 | 4801 | static int io_timeout_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe, |
2d28390a | 4802 | bool is_timeout_link) |
5262f567 | 4803 | { |
ad8a48ac | 4804 | struct io_timeout_data *data; |
a41525ab | 4805 | unsigned flags; |
56080b02 | 4806 | u32 off = READ_ONCE(sqe->off); |
5262f567 | 4807 | |
ad8a48ac | 4808 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
5262f567 | 4809 | return -EINVAL; |
ad8a48ac | 4810 | if (sqe->ioprio || sqe->buf_index || sqe->len != 1) |
a41525ab | 4811 | return -EINVAL; |
56080b02 | 4812 | if (off && is_timeout_link) |
2d28390a | 4813 | return -EINVAL; |
a41525ab JA |
4814 | flags = READ_ONCE(sqe->timeout_flags); |
4815 | if (flags & ~IORING_TIMEOUT_ABS) | |
5262f567 | 4816 | return -EINVAL; |
bdf20073 | 4817 | |
56080b02 | 4818 | req->timeout.count = off; |
26a61679 | 4819 | |
3529d8c2 | 4820 | if (!req->io && io_alloc_async_ctx(req)) |
26a61679 JA |
4821 | return -ENOMEM; |
4822 | ||
4823 | data = &req->io->timeout; | |
ad8a48ac | 4824 | data->req = req; |
ad8a48ac JA |
4825 | req->flags |= REQ_F_TIMEOUT; |
4826 | ||
4827 | if (get_timespec64(&data->ts, u64_to_user_ptr(sqe->addr))) | |
5262f567 JA |
4828 | return -EFAULT; |
4829 | ||
11365043 | 4830 | if (flags & IORING_TIMEOUT_ABS) |
ad8a48ac | 4831 | data->mode = HRTIMER_MODE_ABS; |
11365043 | 4832 | else |
ad8a48ac | 4833 | data->mode = HRTIMER_MODE_REL; |
11365043 | 4834 | |
ad8a48ac JA |
4835 | hrtimer_init(&data->timer, CLOCK_MONOTONIC, data->mode); |
4836 | return 0; | |
4837 | } | |
4838 | ||
fc4df999 | 4839 | static int io_timeout(struct io_kiocb *req) |
ad8a48ac | 4840 | { |
ad8a48ac JA |
4841 | struct io_ring_ctx *ctx = req->ctx; |
4842 | struct io_timeout_data *data; | |
4843 | struct list_head *entry; | |
4844 | unsigned span = 0; | |
b55ce732 | 4845 | u32 count = req->timeout.count; |
22cad158 | 4846 | u32 seq = req->sequence; |
ad8a48ac | 4847 | |
2d28390a | 4848 | data = &req->io->timeout; |
733f5c95 | 4849 | spin_lock_irq(&ctx->completion_lock); |
93bd25bb | 4850 | |
5262f567 JA |
4851 | /* |
4852 | * sqe->off holds how many events that need to occur for this | |
93bd25bb JA |
4853 | * timeout event to be satisfied. If it isn't set, then this is |
4854 | * a pure timeout request, sequence isn't used. | |
5262f567 | 4855 | */ |
93bd25bb JA |
4856 | if (!count) { |
4857 | req->flags |= REQ_F_TIMEOUT_NOSEQ; | |
93bd25bb JA |
4858 | entry = ctx->timeout_list.prev; |
4859 | goto add; | |
4860 | } | |
5262f567 | 4861 | |
22cad158 | 4862 | req->sequence = seq + count; |
5262f567 JA |
4863 | |
4864 | /* | |
4865 | * Insertion sort, ensuring the first entry in the list is always | |
4866 | * the one we need first. | |
4867 | */ | |
5262f567 JA |
4868 | list_for_each_prev(entry, &ctx->timeout_list) { |
4869 | struct io_kiocb *nxt = list_entry(entry, struct io_kiocb, list); | |
22cad158 | 4870 | unsigned nxt_seq; |
5da0fb1a | 4871 | long long tmp, tmp_nxt; |
b55ce732 | 4872 | u32 nxt_offset = nxt->timeout.count; |
5262f567 | 4873 | |
93bd25bb JA |
4874 | if (nxt->flags & REQ_F_TIMEOUT_NOSEQ) |
4875 | continue; | |
4876 | ||
5da0fb1a | 4877 | /* |
22cad158 | 4878 | * Since seq + count can overflow, use type long |
5da0fb1a | 4879 | * long to store it. |
4880 | */ | |
22cad158 PB |
4881 | tmp = (long long)seq + count; |
4882 | nxt_seq = nxt->sequence - nxt_offset; | |
4883 | tmp_nxt = (long long)nxt_seq + nxt_offset; | |
5da0fb1a | 4884 | |
4885 | /* | |
4886 | * cached_sq_head may overflow, and it will never overflow twice | |
4887 | * once there is some timeout req still be valid. | |
4888 | */ | |
22cad158 | 4889 | if (seq < nxt_seq) |
8b07a65a | 4890 | tmp += UINT_MAX; |
5da0fb1a | 4891 | |
a1f58ba4 | 4892 | if (tmp > tmp_nxt) |
5262f567 | 4893 | break; |
a1f58ba4 | 4894 | |
4895 | /* | |
4896 | * Sequence of reqs after the insert one and itself should | |
4897 | * be adjusted because each timeout req consumes a slot. | |
4898 | */ | |
4899 | span++; | |
4900 | nxt->sequence++; | |
5262f567 | 4901 | } |
a1f58ba4 | 4902 | req->sequence -= span; |
93bd25bb | 4903 | add: |
5262f567 | 4904 | list_add(&req->list, entry); |
ad8a48ac JA |
4905 | data->timer.function = io_timeout_fn; |
4906 | hrtimer_start(&data->timer, timespec64_to_ktime(data->ts), data->mode); | |
5262f567 | 4907 | spin_unlock_irq(&ctx->completion_lock); |
5262f567 JA |
4908 | return 0; |
4909 | } | |
5262f567 | 4910 | |
62755e35 JA |
4911 | static bool io_cancel_cb(struct io_wq_work *work, void *data) |
4912 | { | |
4913 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); | |
4914 | ||
4915 | return req->user_data == (unsigned long) data; | |
4916 | } | |
4917 | ||
e977d6d3 | 4918 | static int io_async_cancel_one(struct io_ring_ctx *ctx, void *sqe_addr) |
62755e35 | 4919 | { |
62755e35 | 4920 | enum io_wq_cancel cancel_ret; |
62755e35 JA |
4921 | int ret = 0; |
4922 | ||
62755e35 JA |
4923 | cancel_ret = io_wq_cancel_cb(ctx->io_wq, io_cancel_cb, sqe_addr); |
4924 | switch (cancel_ret) { | |
4925 | case IO_WQ_CANCEL_OK: | |
4926 | ret = 0; | |
4927 | break; | |
4928 | case IO_WQ_CANCEL_RUNNING: | |
4929 | ret = -EALREADY; | |
4930 | break; | |
4931 | case IO_WQ_CANCEL_NOTFOUND: | |
4932 | ret = -ENOENT; | |
4933 | break; | |
4934 | } | |
4935 | ||
e977d6d3 JA |
4936 | return ret; |
4937 | } | |
4938 | ||
47f46768 JA |
4939 | static void io_async_find_and_cancel(struct io_ring_ctx *ctx, |
4940 | struct io_kiocb *req, __u64 sqe_addr, | |
014db007 | 4941 | int success_ret) |
47f46768 JA |
4942 | { |
4943 | unsigned long flags; | |
4944 | int ret; | |
4945 | ||
4946 | ret = io_async_cancel_one(ctx, (void *) (unsigned long) sqe_addr); | |
4947 | if (ret != -ENOENT) { | |
4948 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
4949 | goto done; | |
4950 | } | |
4951 | ||
4952 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
4953 | ret = io_timeout_cancel(ctx, sqe_addr); | |
4954 | if (ret != -ENOENT) | |
4955 | goto done; | |
4956 | ret = io_poll_cancel(ctx, sqe_addr); | |
4957 | done: | |
b0dd8a41 JA |
4958 | if (!ret) |
4959 | ret = success_ret; | |
47f46768 JA |
4960 | io_cqring_fill_event(req, ret); |
4961 | io_commit_cqring(ctx); | |
4962 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
4963 | io_cqring_ev_posted(ctx); | |
4964 | ||
4e88d6e7 JA |
4965 | if (ret < 0) |
4966 | req_set_fail_links(req); | |
014db007 | 4967 | io_put_req(req); |
47f46768 JA |
4968 | } |
4969 | ||
3529d8c2 JA |
4970 | static int io_async_cancel_prep(struct io_kiocb *req, |
4971 | const struct io_uring_sqe *sqe) | |
e977d6d3 | 4972 | { |
fbf23849 | 4973 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
e977d6d3 JA |
4974 | return -EINVAL; |
4975 | if (sqe->flags || sqe->ioprio || sqe->off || sqe->len || | |
4976 | sqe->cancel_flags) | |
4977 | return -EINVAL; | |
4978 | ||
fbf23849 JA |
4979 | req->cancel.addr = READ_ONCE(sqe->addr); |
4980 | return 0; | |
4981 | } | |
4982 | ||
014db007 | 4983 | static int io_async_cancel(struct io_kiocb *req) |
fbf23849 JA |
4984 | { |
4985 | struct io_ring_ctx *ctx = req->ctx; | |
fbf23849 | 4986 | |
014db007 | 4987 | io_async_find_and_cancel(ctx, req, req->cancel.addr, 0); |
5262f567 JA |
4988 | return 0; |
4989 | } | |
4990 | ||
05f3fb3c JA |
4991 | static int io_files_update_prep(struct io_kiocb *req, |
4992 | const struct io_uring_sqe *sqe) | |
4993 | { | |
4994 | if (sqe->flags || sqe->ioprio || sqe->rw_flags) | |
4995 | return -EINVAL; | |
4996 | ||
4997 | req->files_update.offset = READ_ONCE(sqe->off); | |
4998 | req->files_update.nr_args = READ_ONCE(sqe->len); | |
4999 | if (!req->files_update.nr_args) | |
5000 | return -EINVAL; | |
5001 | req->files_update.arg = READ_ONCE(sqe->addr); | |
5002 | return 0; | |
5003 | } | |
5004 | ||
5005 | static int io_files_update(struct io_kiocb *req, bool force_nonblock) | |
fbf23849 JA |
5006 | { |
5007 | struct io_ring_ctx *ctx = req->ctx; | |
05f3fb3c JA |
5008 | struct io_uring_files_update up; |
5009 | int ret; | |
fbf23849 | 5010 | |
f86cd20c | 5011 | if (force_nonblock) |
05f3fb3c | 5012 | return -EAGAIN; |
05f3fb3c JA |
5013 | |
5014 | up.offset = req->files_update.offset; | |
5015 | up.fds = req->files_update.arg; | |
5016 | ||
5017 | mutex_lock(&ctx->uring_lock); | |
5018 | ret = __io_sqe_files_update(ctx, &up, req->files_update.nr_args); | |
5019 | mutex_unlock(&ctx->uring_lock); | |
5020 | ||
5021 | if (ret < 0) | |
5022 | req_set_fail_links(req); | |
5023 | io_cqring_add_event(req, ret); | |
5024 | io_put_req(req); | |
5262f567 JA |
5025 | return 0; |
5026 | } | |
5027 | ||
3529d8c2 JA |
5028 | static int io_req_defer_prep(struct io_kiocb *req, |
5029 | const struct io_uring_sqe *sqe) | |
f67676d1 | 5030 | { |
e781573e | 5031 | ssize_t ret = 0; |
f67676d1 | 5032 | |
f1d96a8f PB |
5033 | if (!sqe) |
5034 | return 0; | |
5035 | ||
f86cd20c JA |
5036 | if (io_op_defs[req->opcode].file_table) { |
5037 | ret = io_grab_files(req); | |
5038 | if (unlikely(ret)) | |
5039 | return ret; | |
5040 | } | |
5041 | ||
cccf0ee8 JA |
5042 | io_req_work_grab_env(req, &io_op_defs[req->opcode]); |
5043 | ||
d625c6ee | 5044 | switch (req->opcode) { |
e781573e JA |
5045 | case IORING_OP_NOP: |
5046 | break; | |
f67676d1 JA |
5047 | case IORING_OP_READV: |
5048 | case IORING_OP_READ_FIXED: | |
3a6820f2 | 5049 | case IORING_OP_READ: |
3529d8c2 | 5050 | ret = io_read_prep(req, sqe, true); |
f67676d1 JA |
5051 | break; |
5052 | case IORING_OP_WRITEV: | |
5053 | case IORING_OP_WRITE_FIXED: | |
3a6820f2 | 5054 | case IORING_OP_WRITE: |
3529d8c2 | 5055 | ret = io_write_prep(req, sqe, true); |
f67676d1 | 5056 | break; |
0969e783 | 5057 | case IORING_OP_POLL_ADD: |
3529d8c2 | 5058 | ret = io_poll_add_prep(req, sqe); |
0969e783 JA |
5059 | break; |
5060 | case IORING_OP_POLL_REMOVE: | |
3529d8c2 | 5061 | ret = io_poll_remove_prep(req, sqe); |
0969e783 | 5062 | break; |
8ed8d3c3 | 5063 | case IORING_OP_FSYNC: |
3529d8c2 | 5064 | ret = io_prep_fsync(req, sqe); |
8ed8d3c3 JA |
5065 | break; |
5066 | case IORING_OP_SYNC_FILE_RANGE: | |
3529d8c2 | 5067 | ret = io_prep_sfr(req, sqe); |
8ed8d3c3 | 5068 | break; |
03b1230c | 5069 | case IORING_OP_SENDMSG: |
fddaface | 5070 | case IORING_OP_SEND: |
3529d8c2 | 5071 | ret = io_sendmsg_prep(req, sqe); |
03b1230c JA |
5072 | break; |
5073 | case IORING_OP_RECVMSG: | |
fddaface | 5074 | case IORING_OP_RECV: |
3529d8c2 | 5075 | ret = io_recvmsg_prep(req, sqe); |
03b1230c | 5076 | break; |
f499a021 | 5077 | case IORING_OP_CONNECT: |
3529d8c2 | 5078 | ret = io_connect_prep(req, sqe); |
f499a021 | 5079 | break; |
2d28390a | 5080 | case IORING_OP_TIMEOUT: |
3529d8c2 | 5081 | ret = io_timeout_prep(req, sqe, false); |
b7bb4f7d | 5082 | break; |
b29472ee | 5083 | case IORING_OP_TIMEOUT_REMOVE: |
3529d8c2 | 5084 | ret = io_timeout_remove_prep(req, sqe); |
b29472ee | 5085 | break; |
fbf23849 | 5086 | case IORING_OP_ASYNC_CANCEL: |
3529d8c2 | 5087 | ret = io_async_cancel_prep(req, sqe); |
fbf23849 | 5088 | break; |
2d28390a | 5089 | case IORING_OP_LINK_TIMEOUT: |
3529d8c2 | 5090 | ret = io_timeout_prep(req, sqe, true); |
b7bb4f7d | 5091 | break; |
8ed8d3c3 | 5092 | case IORING_OP_ACCEPT: |
3529d8c2 | 5093 | ret = io_accept_prep(req, sqe); |
8ed8d3c3 | 5094 | break; |
d63d1b5e JA |
5095 | case IORING_OP_FALLOCATE: |
5096 | ret = io_fallocate_prep(req, sqe); | |
5097 | break; | |
15b71abe JA |
5098 | case IORING_OP_OPENAT: |
5099 | ret = io_openat_prep(req, sqe); | |
5100 | break; | |
b5dba59e JA |
5101 | case IORING_OP_CLOSE: |
5102 | ret = io_close_prep(req, sqe); | |
5103 | break; | |
05f3fb3c JA |
5104 | case IORING_OP_FILES_UPDATE: |
5105 | ret = io_files_update_prep(req, sqe); | |
5106 | break; | |
eddc7ef5 JA |
5107 | case IORING_OP_STATX: |
5108 | ret = io_statx_prep(req, sqe); | |
5109 | break; | |
4840e418 JA |
5110 | case IORING_OP_FADVISE: |
5111 | ret = io_fadvise_prep(req, sqe); | |
5112 | break; | |
c1ca757b JA |
5113 | case IORING_OP_MADVISE: |
5114 | ret = io_madvise_prep(req, sqe); | |
5115 | break; | |
cebdb986 JA |
5116 | case IORING_OP_OPENAT2: |
5117 | ret = io_openat2_prep(req, sqe); | |
5118 | break; | |
3e4827b0 JA |
5119 | case IORING_OP_EPOLL_CTL: |
5120 | ret = io_epoll_ctl_prep(req, sqe); | |
5121 | break; | |
7d67af2c PB |
5122 | case IORING_OP_SPLICE: |
5123 | ret = io_splice_prep(req, sqe); | |
5124 | break; | |
ddf0322d JA |
5125 | case IORING_OP_PROVIDE_BUFFERS: |
5126 | ret = io_provide_buffers_prep(req, sqe); | |
5127 | break; | |
067524e9 JA |
5128 | case IORING_OP_REMOVE_BUFFERS: |
5129 | ret = io_remove_buffers_prep(req, sqe); | |
5130 | break; | |
f2a8d5c7 PB |
5131 | case IORING_OP_TEE: |
5132 | ret = io_tee_prep(req, sqe); | |
5133 | break; | |
f67676d1 | 5134 | default: |
e781573e JA |
5135 | printk_once(KERN_WARNING "io_uring: unhandled opcode %d\n", |
5136 | req->opcode); | |
5137 | ret = -EINVAL; | |
b7bb4f7d | 5138 | break; |
f67676d1 JA |
5139 | } |
5140 | ||
b7bb4f7d | 5141 | return ret; |
f67676d1 JA |
5142 | } |
5143 | ||
3529d8c2 | 5144 | static int io_req_defer(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
de0617e4 | 5145 | { |
a197f664 | 5146 | struct io_ring_ctx *ctx = req->ctx; |
f67676d1 | 5147 | int ret; |
de0617e4 | 5148 | |
9d858b21 | 5149 | /* Still need defer if there is pending req in defer list. */ |
4ee36314 | 5150 | if (!req_need_defer(req) && list_empty_careful(&ctx->defer_list)) |
de0617e4 JA |
5151 | return 0; |
5152 | ||
3529d8c2 | 5153 | if (!req->io && io_alloc_async_ctx(req)) |
de0617e4 JA |
5154 | return -EAGAIN; |
5155 | ||
3529d8c2 | 5156 | ret = io_req_defer_prep(req, sqe); |
b7bb4f7d | 5157 | if (ret < 0) |
2d28390a | 5158 | return ret; |
2d28390a | 5159 | |
de0617e4 | 5160 | spin_lock_irq(&ctx->completion_lock); |
9d858b21 | 5161 | if (!req_need_defer(req) && list_empty(&ctx->defer_list)) { |
de0617e4 | 5162 | spin_unlock_irq(&ctx->completion_lock); |
de0617e4 JA |
5163 | return 0; |
5164 | } | |
5165 | ||
915967f6 | 5166 | trace_io_uring_defer(ctx, req, req->user_data); |
de0617e4 JA |
5167 | list_add_tail(&req->list, &ctx->defer_list); |
5168 | spin_unlock_irq(&ctx->completion_lock); | |
5169 | return -EIOCBQUEUED; | |
5170 | } | |
5171 | ||
99bc4c38 PB |
5172 | static void io_cleanup_req(struct io_kiocb *req) |
5173 | { | |
5174 | struct io_async_ctx *io = req->io; | |
5175 | ||
5176 | switch (req->opcode) { | |
5177 | case IORING_OP_READV: | |
5178 | case IORING_OP_READ_FIXED: | |
5179 | case IORING_OP_READ: | |
bcda7baa JA |
5180 | if (req->flags & REQ_F_BUFFER_SELECTED) |
5181 | kfree((void *)(unsigned long)req->rw.addr); | |
5182 | /* fallthrough */ | |
99bc4c38 PB |
5183 | case IORING_OP_WRITEV: |
5184 | case IORING_OP_WRITE_FIXED: | |
5185 | case IORING_OP_WRITE: | |
5186 | if (io->rw.iov != io->rw.fast_iov) | |
5187 | kfree(io->rw.iov); | |
5188 | break; | |
99bc4c38 | 5189 | case IORING_OP_RECVMSG: |
52de1fe1 JA |
5190 | if (req->flags & REQ_F_BUFFER_SELECTED) |
5191 | kfree(req->sr_msg.kbuf); | |
5192 | /* fallthrough */ | |
5193 | case IORING_OP_SENDMSG: | |
99bc4c38 PB |
5194 | if (io->msg.iov != io->msg.fast_iov) |
5195 | kfree(io->msg.iov); | |
5196 | break; | |
bcda7baa JA |
5197 | case IORING_OP_RECV: |
5198 | if (req->flags & REQ_F_BUFFER_SELECTED) | |
5199 | kfree(req->sr_msg.kbuf); | |
5200 | break; | |
8fef80bf PB |
5201 | case IORING_OP_OPENAT: |
5202 | case IORING_OP_OPENAT2: | |
5203 | case IORING_OP_STATX: | |
5204 | putname(req->open.filename); | |
5205 | break; | |
7d67af2c | 5206 | case IORING_OP_SPLICE: |
f2a8d5c7 | 5207 | case IORING_OP_TEE: |
7d67af2c PB |
5208 | io_put_file(req, req->splice.file_in, |
5209 | (req->splice.flags & SPLICE_F_FD_IN_FIXED)); | |
5210 | break; | |
99bc4c38 PB |
5211 | } |
5212 | ||
5213 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
5214 | } | |
5215 | ||
3529d8c2 | 5216 | static int io_issue_sqe(struct io_kiocb *req, const struct io_uring_sqe *sqe, |
014db007 | 5217 | bool force_nonblock) |
2b188cc1 | 5218 | { |
a197f664 | 5219 | struct io_ring_ctx *ctx = req->ctx; |
d625c6ee | 5220 | int ret; |
2b188cc1 | 5221 | |
d625c6ee | 5222 | switch (req->opcode) { |
2b188cc1 | 5223 | case IORING_OP_NOP: |
78e19bbe | 5224 | ret = io_nop(req); |
2b188cc1 JA |
5225 | break; |
5226 | case IORING_OP_READV: | |
edafccee | 5227 | case IORING_OP_READ_FIXED: |
3a6820f2 | 5228 | case IORING_OP_READ: |
3529d8c2 JA |
5229 | if (sqe) { |
5230 | ret = io_read_prep(req, sqe, force_nonblock); | |
5231 | if (ret < 0) | |
5232 | break; | |
5233 | } | |
014db007 | 5234 | ret = io_read(req, force_nonblock); |
edafccee | 5235 | break; |
3529d8c2 | 5236 | case IORING_OP_WRITEV: |
edafccee | 5237 | case IORING_OP_WRITE_FIXED: |
3a6820f2 | 5238 | case IORING_OP_WRITE: |
3529d8c2 JA |
5239 | if (sqe) { |
5240 | ret = io_write_prep(req, sqe, force_nonblock); | |
5241 | if (ret < 0) | |
5242 | break; | |
5243 | } | |
014db007 | 5244 | ret = io_write(req, force_nonblock); |
2b188cc1 | 5245 | break; |
c992fe29 | 5246 | case IORING_OP_FSYNC: |
3529d8c2 JA |
5247 | if (sqe) { |
5248 | ret = io_prep_fsync(req, sqe); | |
5249 | if (ret < 0) | |
5250 | break; | |
5251 | } | |
014db007 | 5252 | ret = io_fsync(req, force_nonblock); |
c992fe29 | 5253 | break; |
221c5eb2 | 5254 | case IORING_OP_POLL_ADD: |
3529d8c2 JA |
5255 | if (sqe) { |
5256 | ret = io_poll_add_prep(req, sqe); | |
5257 | if (ret) | |
5258 | break; | |
5259 | } | |
014db007 | 5260 | ret = io_poll_add(req); |
221c5eb2 JA |
5261 | break; |
5262 | case IORING_OP_POLL_REMOVE: | |
3529d8c2 JA |
5263 | if (sqe) { |
5264 | ret = io_poll_remove_prep(req, sqe); | |
5265 | if (ret < 0) | |
5266 | break; | |
5267 | } | |
fc4df999 | 5268 | ret = io_poll_remove(req); |
221c5eb2 | 5269 | break; |
5d17b4a4 | 5270 | case IORING_OP_SYNC_FILE_RANGE: |
3529d8c2 JA |
5271 | if (sqe) { |
5272 | ret = io_prep_sfr(req, sqe); | |
5273 | if (ret < 0) | |
5274 | break; | |
5275 | } | |
014db007 | 5276 | ret = io_sync_file_range(req, force_nonblock); |
5d17b4a4 | 5277 | break; |
0fa03c62 | 5278 | case IORING_OP_SENDMSG: |
fddaface | 5279 | case IORING_OP_SEND: |
3529d8c2 JA |
5280 | if (sqe) { |
5281 | ret = io_sendmsg_prep(req, sqe); | |
5282 | if (ret < 0) | |
5283 | break; | |
5284 | } | |
fddaface | 5285 | if (req->opcode == IORING_OP_SENDMSG) |
014db007 | 5286 | ret = io_sendmsg(req, force_nonblock); |
fddaface | 5287 | else |
014db007 | 5288 | ret = io_send(req, force_nonblock); |
0fa03c62 | 5289 | break; |
aa1fa28f | 5290 | case IORING_OP_RECVMSG: |
fddaface | 5291 | case IORING_OP_RECV: |
3529d8c2 JA |
5292 | if (sqe) { |
5293 | ret = io_recvmsg_prep(req, sqe); | |
5294 | if (ret) | |
5295 | break; | |
5296 | } | |
fddaface | 5297 | if (req->opcode == IORING_OP_RECVMSG) |
014db007 | 5298 | ret = io_recvmsg(req, force_nonblock); |
fddaface | 5299 | else |
014db007 | 5300 | ret = io_recv(req, force_nonblock); |
aa1fa28f | 5301 | break; |
5262f567 | 5302 | case IORING_OP_TIMEOUT: |
3529d8c2 JA |
5303 | if (sqe) { |
5304 | ret = io_timeout_prep(req, sqe, false); | |
5305 | if (ret) | |
5306 | break; | |
5307 | } | |
fc4df999 | 5308 | ret = io_timeout(req); |
5262f567 | 5309 | break; |
11365043 | 5310 | case IORING_OP_TIMEOUT_REMOVE: |
3529d8c2 JA |
5311 | if (sqe) { |
5312 | ret = io_timeout_remove_prep(req, sqe); | |
5313 | if (ret) | |
5314 | break; | |
5315 | } | |
fc4df999 | 5316 | ret = io_timeout_remove(req); |
11365043 | 5317 | break; |
17f2fe35 | 5318 | case IORING_OP_ACCEPT: |
3529d8c2 JA |
5319 | if (sqe) { |
5320 | ret = io_accept_prep(req, sqe); | |
5321 | if (ret) | |
5322 | break; | |
5323 | } | |
014db007 | 5324 | ret = io_accept(req, force_nonblock); |
17f2fe35 | 5325 | break; |
f8e85cf2 | 5326 | case IORING_OP_CONNECT: |
3529d8c2 JA |
5327 | if (sqe) { |
5328 | ret = io_connect_prep(req, sqe); | |
5329 | if (ret) | |
5330 | break; | |
5331 | } | |
014db007 | 5332 | ret = io_connect(req, force_nonblock); |
f8e85cf2 | 5333 | break; |
62755e35 | 5334 | case IORING_OP_ASYNC_CANCEL: |
3529d8c2 JA |
5335 | if (sqe) { |
5336 | ret = io_async_cancel_prep(req, sqe); | |
5337 | if (ret) | |
5338 | break; | |
5339 | } | |
014db007 | 5340 | ret = io_async_cancel(req); |
62755e35 | 5341 | break; |
d63d1b5e JA |
5342 | case IORING_OP_FALLOCATE: |
5343 | if (sqe) { | |
5344 | ret = io_fallocate_prep(req, sqe); | |
5345 | if (ret) | |
5346 | break; | |
5347 | } | |
014db007 | 5348 | ret = io_fallocate(req, force_nonblock); |
d63d1b5e | 5349 | break; |
15b71abe JA |
5350 | case IORING_OP_OPENAT: |
5351 | if (sqe) { | |
5352 | ret = io_openat_prep(req, sqe); | |
5353 | if (ret) | |
5354 | break; | |
5355 | } | |
014db007 | 5356 | ret = io_openat(req, force_nonblock); |
15b71abe | 5357 | break; |
b5dba59e JA |
5358 | case IORING_OP_CLOSE: |
5359 | if (sqe) { | |
5360 | ret = io_close_prep(req, sqe); | |
5361 | if (ret) | |
5362 | break; | |
5363 | } | |
014db007 | 5364 | ret = io_close(req, force_nonblock); |
b5dba59e | 5365 | break; |
05f3fb3c JA |
5366 | case IORING_OP_FILES_UPDATE: |
5367 | if (sqe) { | |
5368 | ret = io_files_update_prep(req, sqe); | |
5369 | if (ret) | |
5370 | break; | |
5371 | } | |
5372 | ret = io_files_update(req, force_nonblock); | |
5373 | break; | |
eddc7ef5 JA |
5374 | case IORING_OP_STATX: |
5375 | if (sqe) { | |
5376 | ret = io_statx_prep(req, sqe); | |
5377 | if (ret) | |
5378 | break; | |
5379 | } | |
014db007 | 5380 | ret = io_statx(req, force_nonblock); |
eddc7ef5 | 5381 | break; |
4840e418 JA |
5382 | case IORING_OP_FADVISE: |
5383 | if (sqe) { | |
5384 | ret = io_fadvise_prep(req, sqe); | |
5385 | if (ret) | |
5386 | break; | |
5387 | } | |
014db007 | 5388 | ret = io_fadvise(req, force_nonblock); |
4840e418 | 5389 | break; |
c1ca757b JA |
5390 | case IORING_OP_MADVISE: |
5391 | if (sqe) { | |
5392 | ret = io_madvise_prep(req, sqe); | |
5393 | if (ret) | |
5394 | break; | |
5395 | } | |
014db007 | 5396 | ret = io_madvise(req, force_nonblock); |
c1ca757b | 5397 | break; |
cebdb986 JA |
5398 | case IORING_OP_OPENAT2: |
5399 | if (sqe) { | |
5400 | ret = io_openat2_prep(req, sqe); | |
5401 | if (ret) | |
5402 | break; | |
5403 | } | |
014db007 | 5404 | ret = io_openat2(req, force_nonblock); |
cebdb986 | 5405 | break; |
3e4827b0 JA |
5406 | case IORING_OP_EPOLL_CTL: |
5407 | if (sqe) { | |
5408 | ret = io_epoll_ctl_prep(req, sqe); | |
5409 | if (ret) | |
5410 | break; | |
5411 | } | |
014db007 | 5412 | ret = io_epoll_ctl(req, force_nonblock); |
3e4827b0 | 5413 | break; |
7d67af2c PB |
5414 | case IORING_OP_SPLICE: |
5415 | if (sqe) { | |
5416 | ret = io_splice_prep(req, sqe); | |
5417 | if (ret < 0) | |
5418 | break; | |
5419 | } | |
014db007 | 5420 | ret = io_splice(req, force_nonblock); |
7d67af2c | 5421 | break; |
ddf0322d JA |
5422 | case IORING_OP_PROVIDE_BUFFERS: |
5423 | if (sqe) { | |
5424 | ret = io_provide_buffers_prep(req, sqe); | |
5425 | if (ret) | |
5426 | break; | |
5427 | } | |
5428 | ret = io_provide_buffers(req, force_nonblock); | |
5429 | break; | |
067524e9 JA |
5430 | case IORING_OP_REMOVE_BUFFERS: |
5431 | if (sqe) { | |
5432 | ret = io_remove_buffers_prep(req, sqe); | |
5433 | if (ret) | |
5434 | break; | |
5435 | } | |
5436 | ret = io_remove_buffers(req, force_nonblock); | |
3e4827b0 | 5437 | break; |
f2a8d5c7 PB |
5438 | case IORING_OP_TEE: |
5439 | if (sqe) { | |
5440 | ret = io_tee_prep(req, sqe); | |
5441 | if (ret < 0) | |
5442 | break; | |
5443 | } | |
5444 | ret = io_tee(req, force_nonblock); | |
5445 | break; | |
2b188cc1 JA |
5446 | default: |
5447 | ret = -EINVAL; | |
5448 | break; | |
5449 | } | |
5450 | ||
def596e9 JA |
5451 | if (ret) |
5452 | return ret; | |
5453 | ||
5454 | if (ctx->flags & IORING_SETUP_IOPOLL) { | |
11ba820b JA |
5455 | const bool in_async = io_wq_current_is_worker(); |
5456 | ||
9e645e11 | 5457 | if (req->result == -EAGAIN) |
def596e9 JA |
5458 | return -EAGAIN; |
5459 | ||
11ba820b JA |
5460 | /* workqueue context doesn't hold uring_lock, grab it now */ |
5461 | if (in_async) | |
5462 | mutex_lock(&ctx->uring_lock); | |
5463 | ||
def596e9 | 5464 | io_iopoll_req_issued(req); |
11ba820b JA |
5465 | |
5466 | if (in_async) | |
5467 | mutex_unlock(&ctx->uring_lock); | |
def596e9 JA |
5468 | } |
5469 | ||
5470 | return 0; | |
2b188cc1 JA |
5471 | } |
5472 | ||
561fb04a | 5473 | static void io_wq_submit_work(struct io_wq_work **workptr) |
2b188cc1 | 5474 | { |
561fb04a | 5475 | struct io_wq_work *work = *workptr; |
2b188cc1 | 5476 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); |
561fb04a | 5477 | int ret = 0; |
2b188cc1 | 5478 | |
0c9d5ccd JA |
5479 | /* if NO_CANCEL is set, we must still run the work */ |
5480 | if ((work->flags & (IO_WQ_WORK_CANCEL|IO_WQ_WORK_NO_CANCEL)) == | |
5481 | IO_WQ_WORK_CANCEL) { | |
561fb04a | 5482 | ret = -ECANCELED; |
0c9d5ccd | 5483 | } |
31b51510 | 5484 | |
561fb04a | 5485 | if (!ret) { |
561fb04a | 5486 | do { |
014db007 | 5487 | ret = io_issue_sqe(req, NULL, false); |
561fb04a JA |
5488 | /* |
5489 | * We can get EAGAIN for polled IO even though we're | |
5490 | * forcing a sync submission from here, since we can't | |
5491 | * wait for request slots on the block side. | |
5492 | */ | |
5493 | if (ret != -EAGAIN) | |
5494 | break; | |
5495 | cond_resched(); | |
5496 | } while (1); | |
5497 | } | |
31b51510 | 5498 | |
561fb04a | 5499 | if (ret) { |
4e88d6e7 | 5500 | req_set_fail_links(req); |
78e19bbe | 5501 | io_cqring_add_event(req, ret); |
817869d2 | 5502 | io_put_req(req); |
edafccee | 5503 | } |
2b188cc1 | 5504 | |
e9fd9396 | 5505 | io_steal_work(req, workptr); |
2b188cc1 JA |
5506 | } |
5507 | ||
65e19f54 JA |
5508 | static inline struct file *io_file_from_index(struct io_ring_ctx *ctx, |
5509 | int index) | |
5510 | { | |
5511 | struct fixed_file_table *table; | |
5512 | ||
05f3fb3c | 5513 | table = &ctx->file_data->table[index >> IORING_FILE_TABLE_SHIFT]; |
84695089 | 5514 | return table->files[index & IORING_FILE_TABLE_MASK]; |
65e19f54 JA |
5515 | } |
5516 | ||
8da11c19 PB |
5517 | static int io_file_get(struct io_submit_state *state, struct io_kiocb *req, |
5518 | int fd, struct file **out_file, bool fixed) | |
09bb8394 | 5519 | { |
a197f664 | 5520 | struct io_ring_ctx *ctx = req->ctx; |
8da11c19 | 5521 | struct file *file; |
09bb8394 | 5522 | |
8da11c19 | 5523 | if (fixed) { |
05f3fb3c | 5524 | if (unlikely(!ctx->file_data || |
09bb8394 JA |
5525 | (unsigned) fd >= ctx->nr_user_files)) |
5526 | return -EBADF; | |
b7620121 | 5527 | fd = array_index_nospec(fd, ctx->nr_user_files); |
8da11c19 PB |
5528 | file = io_file_from_index(ctx, fd); |
5529 | if (!file) | |
08a45173 | 5530 | return -EBADF; |
05589553 XW |
5531 | req->fixed_file_refs = ctx->file_data->cur_refs; |
5532 | percpu_ref_get(req->fixed_file_refs); | |
09bb8394 | 5533 | } else { |
c826bd7a | 5534 | trace_io_uring_file_get(ctx, fd); |
8da11c19 PB |
5535 | file = __io_file_get(state, fd); |
5536 | if (unlikely(!file)) | |
09bb8394 JA |
5537 | return -EBADF; |
5538 | } | |
5539 | ||
8da11c19 | 5540 | *out_file = file; |
09bb8394 JA |
5541 | return 0; |
5542 | } | |
5543 | ||
8da11c19 | 5544 | static int io_req_set_file(struct io_submit_state *state, struct io_kiocb *req, |
63ff8223 | 5545 | int fd) |
8da11c19 | 5546 | { |
8da11c19 PB |
5547 | bool fixed; |
5548 | ||
63ff8223 | 5549 | fixed = (req->flags & REQ_F_FIXED_FILE) != 0; |
0cdaf760 | 5550 | if (unlikely(!fixed && io_async_submit(req->ctx))) |
8da11c19 PB |
5551 | return -EBADF; |
5552 | ||
5553 | return io_file_get(state, req, fd, &req->file, fixed); | |
5554 | } | |
5555 | ||
a197f664 | 5556 | static int io_grab_files(struct io_kiocb *req) |
fcb323cc JA |
5557 | { |
5558 | int ret = -EBADF; | |
a197f664 | 5559 | struct io_ring_ctx *ctx = req->ctx; |
fcb323cc | 5560 | |
5b0bbee4 | 5561 | if (req->work.files || (req->flags & REQ_F_NO_FILE_TABLE)) |
f86cd20c | 5562 | return 0; |
b14cca0c | 5563 | if (!ctx->ring_file) |
b5dba59e JA |
5564 | return -EBADF; |
5565 | ||
fcb323cc JA |
5566 | rcu_read_lock(); |
5567 | spin_lock_irq(&ctx->inflight_lock); | |
5568 | /* | |
5569 | * We use the f_ops->flush() handler to ensure that we can flush | |
5570 | * out work accessing these files if the fd is closed. Check if | |
5571 | * the fd has changed since we started down this path, and disallow | |
5572 | * this operation if it has. | |
5573 | */ | |
b14cca0c | 5574 | if (fcheck(ctx->ring_fd) == ctx->ring_file) { |
fcb323cc JA |
5575 | list_add(&req->inflight_entry, &ctx->inflight_list); |
5576 | req->flags |= REQ_F_INFLIGHT; | |
5577 | req->work.files = current->files; | |
5578 | ret = 0; | |
5579 | } | |
5580 | spin_unlock_irq(&ctx->inflight_lock); | |
5581 | rcu_read_unlock(); | |
5582 | ||
5583 | return ret; | |
5584 | } | |
5585 | ||
2665abfd | 5586 | static enum hrtimer_restart io_link_timeout_fn(struct hrtimer *timer) |
2b188cc1 | 5587 | { |
ad8a48ac JA |
5588 | struct io_timeout_data *data = container_of(timer, |
5589 | struct io_timeout_data, timer); | |
5590 | struct io_kiocb *req = data->req; | |
2665abfd JA |
5591 | struct io_ring_ctx *ctx = req->ctx; |
5592 | struct io_kiocb *prev = NULL; | |
5593 | unsigned long flags; | |
2665abfd JA |
5594 | |
5595 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
5596 | ||
5597 | /* | |
5598 | * We don't expect the list to be empty, that will only happen if we | |
5599 | * race with the completion of the linked work. | |
5600 | */ | |
4493233e PB |
5601 | if (!list_empty(&req->link_list)) { |
5602 | prev = list_entry(req->link_list.prev, struct io_kiocb, | |
5603 | link_list); | |
5d960724 | 5604 | if (refcount_inc_not_zero(&prev->refs)) { |
4493233e | 5605 | list_del_init(&req->link_list); |
5d960724 JA |
5606 | prev->flags &= ~REQ_F_LINK_TIMEOUT; |
5607 | } else | |
76a46e06 | 5608 | prev = NULL; |
2665abfd JA |
5609 | } |
5610 | ||
5611 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
5612 | ||
5613 | if (prev) { | |
4e88d6e7 | 5614 | req_set_fail_links(prev); |
014db007 | 5615 | io_async_find_and_cancel(ctx, req, prev->user_data, -ETIME); |
76a46e06 | 5616 | io_put_req(prev); |
47f46768 JA |
5617 | } else { |
5618 | io_cqring_add_event(req, -ETIME); | |
5619 | io_put_req(req); | |
2665abfd | 5620 | } |
2665abfd JA |
5621 | return HRTIMER_NORESTART; |
5622 | } | |
5623 | ||
ad8a48ac | 5624 | static void io_queue_linked_timeout(struct io_kiocb *req) |
2665abfd | 5625 | { |
76a46e06 | 5626 | struct io_ring_ctx *ctx = req->ctx; |
2665abfd | 5627 | |
76a46e06 JA |
5628 | /* |
5629 | * If the list is now empty, then our linked request finished before | |
5630 | * we got a chance to setup the timer | |
5631 | */ | |
5632 | spin_lock_irq(&ctx->completion_lock); | |
4493233e | 5633 | if (!list_empty(&req->link_list)) { |
2d28390a | 5634 | struct io_timeout_data *data = &req->io->timeout; |
94ae5e77 | 5635 | |
ad8a48ac JA |
5636 | data->timer.function = io_link_timeout_fn; |
5637 | hrtimer_start(&data->timer, timespec64_to_ktime(data->ts), | |
5638 | data->mode); | |
2665abfd | 5639 | } |
76a46e06 | 5640 | spin_unlock_irq(&ctx->completion_lock); |
2665abfd | 5641 | |
2665abfd | 5642 | /* drop submission reference */ |
76a46e06 JA |
5643 | io_put_req(req); |
5644 | } | |
2665abfd | 5645 | |
ad8a48ac | 5646 | static struct io_kiocb *io_prep_linked_timeout(struct io_kiocb *req) |
2665abfd JA |
5647 | { |
5648 | struct io_kiocb *nxt; | |
5649 | ||
dea3b49c | 5650 | if (!(req->flags & REQ_F_LINK_HEAD)) |
2665abfd | 5651 | return NULL; |
d7718a9d JA |
5652 | /* for polled retry, if flag is set, we already went through here */ |
5653 | if (req->flags & REQ_F_POLLED) | |
5654 | return NULL; | |
2665abfd | 5655 | |
4493233e PB |
5656 | nxt = list_first_entry_or_null(&req->link_list, struct io_kiocb, |
5657 | link_list); | |
d625c6ee | 5658 | if (!nxt || nxt->opcode != IORING_OP_LINK_TIMEOUT) |
76a46e06 | 5659 | return NULL; |
2665abfd | 5660 | |
76a46e06 | 5661 | req->flags |= REQ_F_LINK_TIMEOUT; |
76a46e06 | 5662 | return nxt; |
2665abfd JA |
5663 | } |
5664 | ||
3529d8c2 | 5665 | static void __io_queue_sqe(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
2b188cc1 | 5666 | { |
4a0a7a18 | 5667 | struct io_kiocb *linked_timeout; |
4bc4494e | 5668 | struct io_kiocb *nxt; |
193155c8 | 5669 | const struct cred *old_creds = NULL; |
e0c5c576 | 5670 | int ret; |
2b188cc1 | 5671 | |
4a0a7a18 JA |
5672 | again: |
5673 | linked_timeout = io_prep_linked_timeout(req); | |
5674 | ||
193155c8 JA |
5675 | if (req->work.creds && req->work.creds != current_cred()) { |
5676 | if (old_creds) | |
5677 | revert_creds(old_creds); | |
5678 | if (old_creds == req->work.creds) | |
5679 | old_creds = NULL; /* restored original creds */ | |
5680 | else | |
5681 | old_creds = override_creds(req->work.creds); | |
5682 | } | |
5683 | ||
014db007 | 5684 | ret = io_issue_sqe(req, sqe, true); |
491381ce JA |
5685 | |
5686 | /* | |
5687 | * We async punt it if the file wasn't marked NOWAIT, or if the file | |
5688 | * doesn't support non-blocking read/write attempts | |
5689 | */ | |
5690 | if (ret == -EAGAIN && (!(req->flags & REQ_F_NOWAIT) || | |
5691 | (req->flags & REQ_F_MUST_PUNT))) { | |
d7718a9d JA |
5692 | if (io_arm_poll_handler(req)) { |
5693 | if (linked_timeout) | |
5694 | io_queue_linked_timeout(linked_timeout); | |
4bc4494e | 5695 | goto exit; |
d7718a9d | 5696 | } |
86a761f8 | 5697 | punt: |
f86cd20c | 5698 | if (io_op_defs[req->opcode].file_table) { |
bbad27b2 PB |
5699 | ret = io_grab_files(req); |
5700 | if (ret) | |
5701 | goto err; | |
2b188cc1 | 5702 | } |
bbad27b2 PB |
5703 | |
5704 | /* | |
5705 | * Queued up for async execution, worker will release | |
5706 | * submit reference when the iocb is actually submitted. | |
5707 | */ | |
5708 | io_queue_async_work(req); | |
4bc4494e | 5709 | goto exit; |
2b188cc1 | 5710 | } |
e65ef56d | 5711 | |
fcb323cc | 5712 | err: |
4bc4494e | 5713 | nxt = NULL; |
76a46e06 | 5714 | /* drop submission reference */ |
2a44f467 | 5715 | io_put_req_find_next(req, &nxt); |
e65ef56d | 5716 | |
f9bd67f6 | 5717 | if (linked_timeout) { |
76a46e06 | 5718 | if (!ret) |
f9bd67f6 | 5719 | io_queue_linked_timeout(linked_timeout); |
76a46e06 | 5720 | else |
f9bd67f6 | 5721 | io_put_req(linked_timeout); |
76a46e06 JA |
5722 | } |
5723 | ||
e65ef56d | 5724 | /* and drop final reference, if we failed */ |
9e645e11 | 5725 | if (ret) { |
78e19bbe | 5726 | io_cqring_add_event(req, ret); |
4e88d6e7 | 5727 | req_set_fail_links(req); |
e65ef56d | 5728 | io_put_req(req); |
9e645e11 | 5729 | } |
4a0a7a18 JA |
5730 | if (nxt) { |
5731 | req = nxt; | |
86a761f8 PB |
5732 | |
5733 | if (req->flags & REQ_F_FORCE_ASYNC) | |
5734 | goto punt; | |
4a0a7a18 JA |
5735 | goto again; |
5736 | } | |
4bc4494e | 5737 | exit: |
193155c8 JA |
5738 | if (old_creds) |
5739 | revert_creds(old_creds); | |
2b188cc1 JA |
5740 | } |
5741 | ||
3529d8c2 | 5742 | static void io_queue_sqe(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
4fe2c963 JL |
5743 | { |
5744 | int ret; | |
5745 | ||
3529d8c2 | 5746 | ret = io_req_defer(req, sqe); |
4fe2c963 JL |
5747 | if (ret) { |
5748 | if (ret != -EIOCBQUEUED) { | |
1118591a | 5749 | fail_req: |
78e19bbe | 5750 | io_cqring_add_event(req, ret); |
4e88d6e7 | 5751 | req_set_fail_links(req); |
78e19bbe | 5752 | io_double_put_req(req); |
4fe2c963 | 5753 | } |
2550878f | 5754 | } else if (req->flags & REQ_F_FORCE_ASYNC) { |
1118591a PB |
5755 | ret = io_req_defer_prep(req, sqe); |
5756 | if (unlikely(ret < 0)) | |
5757 | goto fail_req; | |
ce35a47a JA |
5758 | /* |
5759 | * Never try inline submit of IOSQE_ASYNC is set, go straight | |
5760 | * to async execution. | |
5761 | */ | |
5762 | req->work.flags |= IO_WQ_WORK_CONCURRENT; | |
5763 | io_queue_async_work(req); | |
5764 | } else { | |
3529d8c2 | 5765 | __io_queue_sqe(req, sqe); |
ce35a47a | 5766 | } |
4fe2c963 JL |
5767 | } |
5768 | ||
1b4a51b6 | 5769 | static inline void io_queue_link_head(struct io_kiocb *req) |
4fe2c963 | 5770 | { |
94ae5e77 | 5771 | if (unlikely(req->flags & REQ_F_FAIL_LINK)) { |
1b4a51b6 PB |
5772 | io_cqring_add_event(req, -ECANCELED); |
5773 | io_double_put_req(req); | |
5774 | } else | |
3529d8c2 | 5775 | io_queue_sqe(req, NULL); |
4fe2c963 JL |
5776 | } |
5777 | ||
1d4240cc | 5778 | static int io_submit_sqe(struct io_kiocb *req, const struct io_uring_sqe *sqe, |
7d01bd74 | 5779 | struct io_kiocb **link) |
9e645e11 | 5780 | { |
a197f664 | 5781 | struct io_ring_ctx *ctx = req->ctx; |
ef4ff581 | 5782 | int ret; |
9e645e11 | 5783 | |
9e645e11 JA |
5784 | /* |
5785 | * If we already have a head request, queue this one for async | |
5786 | * submittal once the head completes. If we don't have a head but | |
5787 | * IOSQE_IO_LINK is set in the sqe, start a new head. This one will be | |
5788 | * submitted sync once the chain is complete. If none of those | |
5789 | * conditions are true (normal request), then just queue it. | |
5790 | */ | |
5791 | if (*link) { | |
9d76377f | 5792 | struct io_kiocb *head = *link; |
4e88d6e7 | 5793 | |
8cdf2193 PB |
5794 | /* |
5795 | * Taking sequential execution of a link, draining both sides | |
5796 | * of the link also fullfils IOSQE_IO_DRAIN semantics for all | |
5797 | * requests in the link. So, it drains the head and the | |
5798 | * next after the link request. The last one is done via | |
5799 | * drain_next flag to persist the effect across calls. | |
5800 | */ | |
ef4ff581 | 5801 | if (req->flags & REQ_F_IO_DRAIN) { |
711be031 PB |
5802 | head->flags |= REQ_F_IO_DRAIN; |
5803 | ctx->drain_next = 1; | |
5804 | } | |
1d4240cc PB |
5805 | if (io_alloc_async_ctx(req)) |
5806 | return -EAGAIN; | |
9e645e11 | 5807 | |
3529d8c2 | 5808 | ret = io_req_defer_prep(req, sqe); |
2d28390a | 5809 | if (ret) { |
4e88d6e7 | 5810 | /* fail even hard links since we don't submit */ |
9d76377f | 5811 | head->flags |= REQ_F_FAIL_LINK; |
1d4240cc | 5812 | return ret; |
2d28390a | 5813 | } |
9d76377f PB |
5814 | trace_io_uring_link(ctx, req, head); |
5815 | list_add_tail(&req->link_list, &head->link_list); | |
32fe525b PB |
5816 | |
5817 | /* last request of a link, enqueue the link */ | |
ef4ff581 | 5818 | if (!(req->flags & (REQ_F_LINK | REQ_F_HARDLINK))) { |
32fe525b PB |
5819 | io_queue_link_head(head); |
5820 | *link = NULL; | |
5821 | } | |
9e645e11 | 5822 | } else { |
711be031 PB |
5823 | if (unlikely(ctx->drain_next)) { |
5824 | req->flags |= REQ_F_IO_DRAIN; | |
ef4ff581 | 5825 | ctx->drain_next = 0; |
711be031 | 5826 | } |
ef4ff581 | 5827 | if (req->flags & (REQ_F_LINK | REQ_F_HARDLINK)) { |
dea3b49c | 5828 | req->flags |= REQ_F_LINK_HEAD; |
711be031 | 5829 | INIT_LIST_HEAD(&req->link_list); |
f1d96a8f | 5830 | |
1d4240cc PB |
5831 | if (io_alloc_async_ctx(req)) |
5832 | return -EAGAIN; | |
5833 | ||
711be031 PB |
5834 | ret = io_req_defer_prep(req, sqe); |
5835 | if (ret) | |
5836 | req->flags |= REQ_F_FAIL_LINK; | |
5837 | *link = req; | |
5838 | } else { | |
5839 | io_queue_sqe(req, sqe); | |
5840 | } | |
9e645e11 | 5841 | } |
2e6e1fde | 5842 | |
1d4240cc | 5843 | return 0; |
9e645e11 JA |
5844 | } |
5845 | ||
9a56a232 JA |
5846 | /* |
5847 | * Batched submission is done, ensure local IO is flushed out. | |
5848 | */ | |
5849 | static void io_submit_state_end(struct io_submit_state *state) | |
5850 | { | |
5851 | blk_finish_plug(&state->plug); | |
9f13c35b | 5852 | io_state_file_put(state); |
2579f913 | 5853 | if (state->free_reqs) |
6c8a3134 | 5854 | kmem_cache_free_bulk(req_cachep, state->free_reqs, state->reqs); |
9a56a232 JA |
5855 | } |
5856 | ||
5857 | /* | |
5858 | * Start submission side cache. | |
5859 | */ | |
5860 | static void io_submit_state_start(struct io_submit_state *state, | |
22efde59 | 5861 | unsigned int max_ios) |
9a56a232 JA |
5862 | { |
5863 | blk_start_plug(&state->plug); | |
2579f913 | 5864 | state->free_reqs = 0; |
9a56a232 JA |
5865 | state->file = NULL; |
5866 | state->ios_left = max_ios; | |
5867 | } | |
5868 | ||
2b188cc1 JA |
5869 | static void io_commit_sqring(struct io_ring_ctx *ctx) |
5870 | { | |
75b28aff | 5871 | struct io_rings *rings = ctx->rings; |
2b188cc1 | 5872 | |
caf582c6 PB |
5873 | /* |
5874 | * Ensure any loads from the SQEs are done at this point, | |
5875 | * since once we write the new head, the application could | |
5876 | * write new data to them. | |
5877 | */ | |
5878 | smp_store_release(&rings->sq.head, ctx->cached_sq_head); | |
2b188cc1 JA |
5879 | } |
5880 | ||
2b188cc1 | 5881 | /* |
3529d8c2 | 5882 | * Fetch an sqe, if one is available. Note that sqe_ptr will point to memory |
2b188cc1 JA |
5883 | * that is mapped by userspace. This means that care needs to be taken to |
5884 | * ensure that reads are stable, as we cannot rely on userspace always | |
5885 | * being a good citizen. If members of the sqe are validated and then later | |
5886 | * used, it's important that those reads are done through READ_ONCE() to | |
5887 | * prevent a re-load down the line. | |
5888 | */ | |
709b302f | 5889 | static const struct io_uring_sqe *io_get_sqe(struct io_ring_ctx *ctx) |
2b188cc1 | 5890 | { |
75b28aff | 5891 | u32 *sq_array = ctx->sq_array; |
2b188cc1 JA |
5892 | unsigned head; |
5893 | ||
5894 | /* | |
5895 | * The cached sq head (or cq tail) serves two purposes: | |
5896 | * | |
5897 | * 1) allows us to batch the cost of updating the user visible | |
5898 | * head updates. | |
5899 | * 2) allows the kernel side to track the head on its own, even | |
5900 | * though the application is the one updating it. | |
5901 | */ | |
ee7d46d9 | 5902 | head = READ_ONCE(sq_array[ctx->cached_sq_head & ctx->sq_mask]); |
709b302f PB |
5903 | if (likely(head < ctx->sq_entries)) |
5904 | return &ctx->sq_sqes[head]; | |
2b188cc1 JA |
5905 | |
5906 | /* drop invalid entries */ | |
498ccd9e | 5907 | ctx->cached_sq_dropped++; |
ee7d46d9 | 5908 | WRITE_ONCE(ctx->rings->sq_dropped, ctx->cached_sq_dropped); |
709b302f PB |
5909 | return NULL; |
5910 | } | |
5911 | ||
5912 | static inline void io_consume_sqe(struct io_ring_ctx *ctx) | |
5913 | { | |
5914 | ctx->cached_sq_head++; | |
2b188cc1 JA |
5915 | } |
5916 | ||
ef4ff581 PB |
5917 | #define SQE_VALID_FLAGS (IOSQE_FIXED_FILE|IOSQE_IO_DRAIN|IOSQE_IO_LINK| \ |
5918 | IOSQE_IO_HARDLINK | IOSQE_ASYNC | \ | |
5919 | IOSQE_BUFFER_SELECT) | |
5920 | ||
5921 | static int io_init_req(struct io_ring_ctx *ctx, struct io_kiocb *req, | |
5922 | const struct io_uring_sqe *sqe, | |
0cdaf760 | 5923 | struct io_submit_state *state) |
0553b8bd | 5924 | { |
ef4ff581 | 5925 | unsigned int sqe_flags; |
63ff8223 | 5926 | int id; |
ef4ff581 | 5927 | |
0553b8bd PB |
5928 | /* |
5929 | * All io need record the previous position, if LINK vs DARIN, | |
5930 | * it can be used to mark the position of the first IO in the | |
5931 | * link list. | |
5932 | */ | |
31af27c7 | 5933 | req->sequence = ctx->cached_sq_head - ctx->cached_sq_dropped; |
0553b8bd PB |
5934 | req->opcode = READ_ONCE(sqe->opcode); |
5935 | req->user_data = READ_ONCE(sqe->user_data); | |
5936 | req->io = NULL; | |
5937 | req->file = NULL; | |
5938 | req->ctx = ctx; | |
5939 | req->flags = 0; | |
5940 | /* one is dropped after submission, the other at completion */ | |
5941 | refcount_set(&req->refs, 2); | |
5942 | req->task = NULL; | |
5943 | req->result = 0; | |
5944 | INIT_IO_WORK(&req->work, io_wq_submit_work); | |
ef4ff581 PB |
5945 | |
5946 | if (unlikely(req->opcode >= IORING_OP_LAST)) | |
5947 | return -EINVAL; | |
5948 | ||
5949 | if (io_op_defs[req->opcode].needs_mm && !current->mm) { | |
5950 | if (unlikely(!mmget_not_zero(ctx->sqo_mm))) | |
5951 | return -EFAULT; | |
5952 | use_mm(ctx->sqo_mm); | |
5953 | } | |
5954 | ||
5955 | sqe_flags = READ_ONCE(sqe->flags); | |
5956 | /* enforce forwards compatibility on users */ | |
5957 | if (unlikely(sqe_flags & ~SQE_VALID_FLAGS)) | |
5958 | return -EINVAL; | |
5959 | ||
5960 | if ((sqe_flags & IOSQE_BUFFER_SELECT) && | |
5961 | !io_op_defs[req->opcode].buffer_select) | |
5962 | return -EOPNOTSUPP; | |
5963 | ||
5964 | id = READ_ONCE(sqe->personality); | |
5965 | if (id) { | |
5966 | req->work.creds = idr_find(&ctx->personality_idr, id); | |
5967 | if (unlikely(!req->work.creds)) | |
5968 | return -EINVAL; | |
5969 | get_cred(req->work.creds); | |
5970 | } | |
5971 | ||
5972 | /* same numerical values with corresponding REQ_F_*, safe to copy */ | |
c11368a5 | 5973 | req->flags |= sqe_flags; |
ef4ff581 | 5974 | |
63ff8223 JA |
5975 | if (!io_op_defs[req->opcode].needs_file) |
5976 | return 0; | |
5977 | ||
5978 | return io_req_set_file(state, req, READ_ONCE(sqe->fd)); | |
0553b8bd PB |
5979 | } |
5980 | ||
fb5ccc98 | 5981 | static int io_submit_sqes(struct io_ring_ctx *ctx, unsigned int nr, |
0cdaf760 | 5982 | struct file *ring_file, int ring_fd) |
6c271ce2 JA |
5983 | { |
5984 | struct io_submit_state state, *statep = NULL; | |
9e645e11 | 5985 | struct io_kiocb *link = NULL; |
9e645e11 | 5986 | int i, submitted = 0; |
6c271ce2 | 5987 | |
c4a2ed72 | 5988 | /* if we have a backlog and couldn't flush it all, return BUSY */ |
ad3eb2c8 JA |
5989 | if (test_bit(0, &ctx->sq_check_overflow)) { |
5990 | if (!list_empty(&ctx->cq_overflow_list) && | |
5991 | !io_cqring_overflow_flush(ctx, false)) | |
5992 | return -EBUSY; | |
5993 | } | |
6c271ce2 | 5994 | |
ee7d46d9 PB |
5995 | /* make sure SQ entry isn't read before tail */ |
5996 | nr = min3(nr, ctx->sq_entries, io_sqring_entries(ctx)); | |
9ef4f124 | 5997 | |
2b85edfc PB |
5998 | if (!percpu_ref_tryget_many(&ctx->refs, nr)) |
5999 | return -EAGAIN; | |
6c271ce2 JA |
6000 | |
6001 | if (nr > IO_PLUG_THRESHOLD) { | |
22efde59 | 6002 | io_submit_state_start(&state, nr); |
6c271ce2 JA |
6003 | statep = &state; |
6004 | } | |
6005 | ||
b14cca0c PB |
6006 | ctx->ring_fd = ring_fd; |
6007 | ctx->ring_file = ring_file; | |
6008 | ||
6c271ce2 | 6009 | for (i = 0; i < nr; i++) { |
3529d8c2 | 6010 | const struct io_uring_sqe *sqe; |
196be95c | 6011 | struct io_kiocb *req; |
1cb1edb2 | 6012 | int err; |
fb5ccc98 | 6013 | |
b1e50e54 PB |
6014 | sqe = io_get_sqe(ctx); |
6015 | if (unlikely(!sqe)) { | |
6016 | io_consume_sqe(ctx); | |
6017 | break; | |
6018 | } | |
0553b8bd | 6019 | req = io_alloc_req(ctx, statep); |
196be95c PB |
6020 | if (unlikely(!req)) { |
6021 | if (!submitted) | |
6022 | submitted = -EAGAIN; | |
fb5ccc98 | 6023 | break; |
196be95c | 6024 | } |
fb5ccc98 | 6025 | |
0cdaf760 | 6026 | err = io_init_req(ctx, req, sqe, statep); |
709b302f | 6027 | io_consume_sqe(ctx); |
d3656344 JA |
6028 | /* will complete beyond this point, count as submitted */ |
6029 | submitted++; | |
6030 | ||
ef4ff581 | 6031 | if (unlikely(err)) { |
1cb1edb2 PB |
6032 | fail_req: |
6033 | io_cqring_add_event(req, err); | |
d3656344 | 6034 | io_double_put_req(req); |
196be95c PB |
6035 | break; |
6036 | } | |
fb5ccc98 | 6037 | |
354420f7 | 6038 | trace_io_uring_submit_sqe(ctx, req->opcode, req->user_data, |
0cdaf760 | 6039 | true, io_async_submit(ctx)); |
7d01bd74 | 6040 | err = io_submit_sqe(req, sqe, &link); |
1d4240cc PB |
6041 | if (err) |
6042 | goto fail_req; | |
6c271ce2 JA |
6043 | } |
6044 | ||
9466f437 PB |
6045 | if (unlikely(submitted != nr)) { |
6046 | int ref_used = (submitted == -EAGAIN) ? 0 : submitted; | |
6047 | ||
6048 | percpu_ref_put_many(&ctx->refs, nr - ref_used); | |
6049 | } | |
9e645e11 | 6050 | if (link) |
1b4a51b6 | 6051 | io_queue_link_head(link); |
6c271ce2 JA |
6052 | if (statep) |
6053 | io_submit_state_end(&state); | |
6054 | ||
ae9428ca PB |
6055 | /* Commit SQ ring head once we've consumed and submitted all SQEs */ |
6056 | io_commit_sqring(ctx); | |
6057 | ||
6c271ce2 JA |
6058 | return submitted; |
6059 | } | |
6060 | ||
bf9c2f1c PB |
6061 | static inline void io_sq_thread_drop_mm(struct io_ring_ctx *ctx) |
6062 | { | |
6063 | struct mm_struct *mm = current->mm; | |
6064 | ||
6065 | if (mm) { | |
6066 | unuse_mm(mm); | |
6067 | mmput(mm); | |
6068 | } | |
6069 | } | |
6070 | ||
6c271ce2 JA |
6071 | static int io_sq_thread(void *data) |
6072 | { | |
6c271ce2 | 6073 | struct io_ring_ctx *ctx = data; |
181e448d | 6074 | const struct cred *old_cred; |
6c271ce2 JA |
6075 | mm_segment_t old_fs; |
6076 | DEFINE_WAIT(wait); | |
6c271ce2 | 6077 | unsigned long timeout; |
bdcd3eab | 6078 | int ret = 0; |
6c271ce2 | 6079 | |
0f158b4c | 6080 | complete(&ctx->sq_thread_comp); |
a4c0b3de | 6081 | |
6c271ce2 JA |
6082 | old_fs = get_fs(); |
6083 | set_fs(USER_DS); | |
181e448d | 6084 | old_cred = override_creds(ctx->creds); |
6c271ce2 | 6085 | |
bdcd3eab | 6086 | timeout = jiffies + ctx->sq_thread_idle; |
2bbcd6d3 | 6087 | while (!kthread_should_park()) { |
fb5ccc98 | 6088 | unsigned int to_submit; |
6c271ce2 | 6089 | |
bdcd3eab | 6090 | if (!list_empty(&ctx->poll_list)) { |
6c271ce2 JA |
6091 | unsigned nr_events = 0; |
6092 | ||
bdcd3eab XW |
6093 | mutex_lock(&ctx->uring_lock); |
6094 | if (!list_empty(&ctx->poll_list)) | |
6095 | io_iopoll_getevents(ctx, &nr_events, 0); | |
6096 | else | |
6c271ce2 | 6097 | timeout = jiffies + ctx->sq_thread_idle; |
bdcd3eab | 6098 | mutex_unlock(&ctx->uring_lock); |
6c271ce2 JA |
6099 | } |
6100 | ||
fb5ccc98 | 6101 | to_submit = io_sqring_entries(ctx); |
c1edbf5f JA |
6102 | |
6103 | /* | |
6104 | * If submit got -EBUSY, flag us as needing the application | |
6105 | * to enter the kernel to reap and flush events. | |
6106 | */ | |
6107 | if (!to_submit || ret == -EBUSY) { | |
7143b5ac SG |
6108 | /* |
6109 | * Drop cur_mm before scheduling, we can't hold it for | |
6110 | * long periods (or over schedule()). Do this before | |
6111 | * adding ourselves to the waitqueue, as the unuse/drop | |
6112 | * may sleep. | |
6113 | */ | |
bf9c2f1c | 6114 | io_sq_thread_drop_mm(ctx); |
7143b5ac | 6115 | |
6c271ce2 JA |
6116 | /* |
6117 | * We're polling. If we're within the defined idle | |
6118 | * period, then let us spin without work before going | |
c1edbf5f JA |
6119 | * to sleep. The exception is if we got EBUSY doing |
6120 | * more IO, we should wait for the application to | |
6121 | * reap events and wake us up. | |
6c271ce2 | 6122 | */ |
bdcd3eab | 6123 | if (!list_empty(&ctx->poll_list) || |
df069d80 JA |
6124 | (!time_after(jiffies, timeout) && ret != -EBUSY && |
6125 | !percpu_ref_is_dying(&ctx->refs))) { | |
b41e9852 JA |
6126 | if (current->task_works) |
6127 | task_work_run(); | |
9831a90c | 6128 | cond_resched(); |
6c271ce2 JA |
6129 | continue; |
6130 | } | |
6131 | ||
6c271ce2 JA |
6132 | prepare_to_wait(&ctx->sqo_wait, &wait, |
6133 | TASK_INTERRUPTIBLE); | |
6134 | ||
bdcd3eab XW |
6135 | /* |
6136 | * While doing polled IO, before going to sleep, we need | |
6137 | * to check if there are new reqs added to poll_list, it | |
6138 | * is because reqs may have been punted to io worker and | |
6139 | * will be added to poll_list later, hence check the | |
6140 | * poll_list again. | |
6141 | */ | |
6142 | if ((ctx->flags & IORING_SETUP_IOPOLL) && | |
6143 | !list_empty_careful(&ctx->poll_list)) { | |
6144 | finish_wait(&ctx->sqo_wait, &wait); | |
6145 | continue; | |
6146 | } | |
6147 | ||
6c271ce2 | 6148 | /* Tell userspace we may need a wakeup call */ |
75b28aff | 6149 | ctx->rings->sq_flags |= IORING_SQ_NEED_WAKEUP; |
0d7bae69 SB |
6150 | /* make sure to read SQ tail after writing flags */ |
6151 | smp_mb(); | |
6c271ce2 | 6152 | |
fb5ccc98 | 6153 | to_submit = io_sqring_entries(ctx); |
c1edbf5f | 6154 | if (!to_submit || ret == -EBUSY) { |
2bbcd6d3 | 6155 | if (kthread_should_park()) { |
6c271ce2 JA |
6156 | finish_wait(&ctx->sqo_wait, &wait); |
6157 | break; | |
6158 | } | |
b41e9852 JA |
6159 | if (current->task_works) { |
6160 | task_work_run(); | |
10bea96d | 6161 | finish_wait(&ctx->sqo_wait, &wait); |
b41e9852 JA |
6162 | continue; |
6163 | } | |
6c271ce2 JA |
6164 | if (signal_pending(current)) |
6165 | flush_signals(current); | |
6166 | schedule(); | |
6167 | finish_wait(&ctx->sqo_wait, &wait); | |
6168 | ||
75b28aff | 6169 | ctx->rings->sq_flags &= ~IORING_SQ_NEED_WAKEUP; |
6c271ce2 JA |
6170 | continue; |
6171 | } | |
6172 | finish_wait(&ctx->sqo_wait, &wait); | |
6173 | ||
75b28aff | 6174 | ctx->rings->sq_flags &= ~IORING_SQ_NEED_WAKEUP; |
6c271ce2 JA |
6175 | } |
6176 | ||
8a4955ff | 6177 | mutex_lock(&ctx->uring_lock); |
6b668c9b XW |
6178 | if (likely(!percpu_ref_is_dying(&ctx->refs))) |
6179 | ret = io_submit_sqes(ctx, to_submit, NULL, -1); | |
8a4955ff | 6180 | mutex_unlock(&ctx->uring_lock); |
bdcd3eab | 6181 | timeout = jiffies + ctx->sq_thread_idle; |
6c271ce2 JA |
6182 | } |
6183 | ||
b41e9852 JA |
6184 | if (current->task_works) |
6185 | task_work_run(); | |
6186 | ||
6c271ce2 | 6187 | set_fs(old_fs); |
bf9c2f1c | 6188 | io_sq_thread_drop_mm(ctx); |
181e448d | 6189 | revert_creds(old_cred); |
06058632 | 6190 | |
2bbcd6d3 | 6191 | kthread_parkme(); |
06058632 | 6192 | |
6c271ce2 JA |
6193 | return 0; |
6194 | } | |
6195 | ||
bda52162 JA |
6196 | struct io_wait_queue { |
6197 | struct wait_queue_entry wq; | |
6198 | struct io_ring_ctx *ctx; | |
6199 | unsigned to_wait; | |
6200 | unsigned nr_timeouts; | |
6201 | }; | |
6202 | ||
1d7bb1d5 | 6203 | static inline bool io_should_wake(struct io_wait_queue *iowq, bool noflush) |
bda52162 JA |
6204 | { |
6205 | struct io_ring_ctx *ctx = iowq->ctx; | |
6206 | ||
6207 | /* | |
d195a66e | 6208 | * Wake up if we have enough events, or if a timeout occurred since we |
bda52162 JA |
6209 | * started waiting. For timeouts, we always want to return to userspace, |
6210 | * regardless of event count. | |
6211 | */ | |
1d7bb1d5 | 6212 | return io_cqring_events(ctx, noflush) >= iowq->to_wait || |
bda52162 JA |
6213 | atomic_read(&ctx->cq_timeouts) != iowq->nr_timeouts; |
6214 | } | |
6215 | ||
6216 | static int io_wake_function(struct wait_queue_entry *curr, unsigned int mode, | |
6217 | int wake_flags, void *key) | |
6218 | { | |
6219 | struct io_wait_queue *iowq = container_of(curr, struct io_wait_queue, | |
6220 | wq); | |
6221 | ||
1d7bb1d5 JA |
6222 | /* use noflush == true, as we can't safely rely on locking context */ |
6223 | if (!io_should_wake(iowq, true)) | |
bda52162 JA |
6224 | return -1; |
6225 | ||
6226 | return autoremove_wake_function(curr, mode, wake_flags, key); | |
6227 | } | |
6228 | ||
2b188cc1 JA |
6229 | /* |
6230 | * Wait until events become available, if we don't already have some. The | |
6231 | * application must reap them itself, as they reside on the shared cq ring. | |
6232 | */ | |
6233 | static int io_cqring_wait(struct io_ring_ctx *ctx, int min_events, | |
6234 | const sigset_t __user *sig, size_t sigsz) | |
6235 | { | |
bda52162 JA |
6236 | struct io_wait_queue iowq = { |
6237 | .wq = { | |
6238 | .private = current, | |
6239 | .func = io_wake_function, | |
6240 | .entry = LIST_HEAD_INIT(iowq.wq.entry), | |
6241 | }, | |
6242 | .ctx = ctx, | |
6243 | .to_wait = min_events, | |
6244 | }; | |
75b28aff | 6245 | struct io_rings *rings = ctx->rings; |
e9ffa5c2 | 6246 | int ret = 0; |
2b188cc1 | 6247 | |
b41e9852 JA |
6248 | do { |
6249 | if (io_cqring_events(ctx, false) >= min_events) | |
6250 | return 0; | |
6251 | if (!current->task_works) | |
6252 | break; | |
6253 | task_work_run(); | |
6254 | } while (1); | |
2b188cc1 JA |
6255 | |
6256 | if (sig) { | |
9e75ad5d AB |
6257 | #ifdef CONFIG_COMPAT |
6258 | if (in_compat_syscall()) | |
6259 | ret = set_compat_user_sigmask((const compat_sigset_t __user *)sig, | |
b772434b | 6260 | sigsz); |
9e75ad5d AB |
6261 | else |
6262 | #endif | |
b772434b | 6263 | ret = set_user_sigmask(sig, sigsz); |
9e75ad5d | 6264 | |
2b188cc1 JA |
6265 | if (ret) |
6266 | return ret; | |
6267 | } | |
6268 | ||
bda52162 | 6269 | iowq.nr_timeouts = atomic_read(&ctx->cq_timeouts); |
c826bd7a | 6270 | trace_io_uring_cqring_wait(ctx, min_events); |
bda52162 JA |
6271 | do { |
6272 | prepare_to_wait_exclusive(&ctx->wait, &iowq.wq, | |
6273 | TASK_INTERRUPTIBLE); | |
b41e9852 JA |
6274 | if (current->task_works) |
6275 | task_work_run(); | |
1d7bb1d5 | 6276 | if (io_should_wake(&iowq, false)) |
bda52162 JA |
6277 | break; |
6278 | schedule(); | |
6279 | if (signal_pending(current)) { | |
e9ffa5c2 | 6280 | ret = -EINTR; |
bda52162 JA |
6281 | break; |
6282 | } | |
6283 | } while (1); | |
6284 | finish_wait(&ctx->wait, &iowq.wq); | |
6285 | ||
e9ffa5c2 | 6286 | restore_saved_sigmask_unless(ret == -EINTR); |
2b188cc1 | 6287 | |
75b28aff | 6288 | return READ_ONCE(rings->cq.head) == READ_ONCE(rings->cq.tail) ? ret : 0; |
2b188cc1 JA |
6289 | } |
6290 | ||
6b06314c JA |
6291 | static void __io_sqe_files_unregister(struct io_ring_ctx *ctx) |
6292 | { | |
6293 | #if defined(CONFIG_UNIX) | |
6294 | if (ctx->ring_sock) { | |
6295 | struct sock *sock = ctx->ring_sock->sk; | |
6296 | struct sk_buff *skb; | |
6297 | ||
6298 | while ((skb = skb_dequeue(&sock->sk_receive_queue)) != NULL) | |
6299 | kfree_skb(skb); | |
6300 | } | |
6301 | #else | |
6302 | int i; | |
6303 | ||
65e19f54 JA |
6304 | for (i = 0; i < ctx->nr_user_files; i++) { |
6305 | struct file *file; | |
6306 | ||
6307 | file = io_file_from_index(ctx, i); | |
6308 | if (file) | |
6309 | fput(file); | |
6310 | } | |
6b06314c JA |
6311 | #endif |
6312 | } | |
6313 | ||
05f3fb3c JA |
6314 | static void io_file_ref_kill(struct percpu_ref *ref) |
6315 | { | |
6316 | struct fixed_file_data *data; | |
6317 | ||
6318 | data = container_of(ref, struct fixed_file_data, refs); | |
6319 | complete(&data->done); | |
6320 | } | |
6321 | ||
6b06314c JA |
6322 | static int io_sqe_files_unregister(struct io_ring_ctx *ctx) |
6323 | { | |
05f3fb3c | 6324 | struct fixed_file_data *data = ctx->file_data; |
05589553 | 6325 | struct fixed_file_ref_node *ref_node = NULL; |
65e19f54 JA |
6326 | unsigned nr_tables, i; |
6327 | ||
05f3fb3c | 6328 | if (!data) |
6b06314c JA |
6329 | return -ENXIO; |
6330 | ||
6a4d07cd | 6331 | spin_lock(&data->lock); |
05589553 XW |
6332 | if (!list_empty(&data->ref_list)) |
6333 | ref_node = list_first_entry(&data->ref_list, | |
6334 | struct fixed_file_ref_node, node); | |
6a4d07cd | 6335 | spin_unlock(&data->lock); |
05589553 XW |
6336 | if (ref_node) |
6337 | percpu_ref_kill(&ref_node->refs); | |
6338 | ||
6339 | percpu_ref_kill(&data->refs); | |
6340 | ||
6341 | /* wait for all refs nodes to complete */ | |
4a38aed2 | 6342 | flush_delayed_work(&ctx->file_put_work); |
2faf852d | 6343 | wait_for_completion(&data->done); |
05f3fb3c | 6344 | |
6b06314c | 6345 | __io_sqe_files_unregister(ctx); |
65e19f54 JA |
6346 | nr_tables = DIV_ROUND_UP(ctx->nr_user_files, IORING_MAX_FILES_TABLE); |
6347 | for (i = 0; i < nr_tables; i++) | |
05f3fb3c JA |
6348 | kfree(data->table[i].files); |
6349 | kfree(data->table); | |
05589553 XW |
6350 | percpu_ref_exit(&data->refs); |
6351 | kfree(data); | |
05f3fb3c | 6352 | ctx->file_data = NULL; |
6b06314c JA |
6353 | ctx->nr_user_files = 0; |
6354 | return 0; | |
6355 | } | |
6356 | ||
6c271ce2 JA |
6357 | static void io_sq_thread_stop(struct io_ring_ctx *ctx) |
6358 | { | |
6359 | if (ctx->sqo_thread) { | |
0f158b4c | 6360 | wait_for_completion(&ctx->sq_thread_comp); |
2bbcd6d3 RP |
6361 | /* |
6362 | * The park is a bit of a work-around, without it we get | |
6363 | * warning spews on shutdown with SQPOLL set and affinity | |
6364 | * set to a single CPU. | |
6365 | */ | |
06058632 | 6366 | kthread_park(ctx->sqo_thread); |
6c271ce2 JA |
6367 | kthread_stop(ctx->sqo_thread); |
6368 | ctx->sqo_thread = NULL; | |
6369 | } | |
6370 | } | |
6371 | ||
6b06314c JA |
6372 | static void io_finish_async(struct io_ring_ctx *ctx) |
6373 | { | |
6c271ce2 JA |
6374 | io_sq_thread_stop(ctx); |
6375 | ||
561fb04a JA |
6376 | if (ctx->io_wq) { |
6377 | io_wq_destroy(ctx->io_wq); | |
6378 | ctx->io_wq = NULL; | |
6b06314c JA |
6379 | } |
6380 | } | |
6381 | ||
6382 | #if defined(CONFIG_UNIX) | |
6b06314c JA |
6383 | /* |
6384 | * Ensure the UNIX gc is aware of our file set, so we are certain that | |
6385 | * the io_uring can be safely unregistered on process exit, even if we have | |
6386 | * loops in the file referencing. | |
6387 | */ | |
6388 | static int __io_sqe_files_scm(struct io_ring_ctx *ctx, int nr, int offset) | |
6389 | { | |
6390 | struct sock *sk = ctx->ring_sock->sk; | |
6391 | struct scm_fp_list *fpl; | |
6392 | struct sk_buff *skb; | |
08a45173 | 6393 | int i, nr_files; |
6b06314c | 6394 | |
6b06314c JA |
6395 | fpl = kzalloc(sizeof(*fpl), GFP_KERNEL); |
6396 | if (!fpl) | |
6397 | return -ENOMEM; | |
6398 | ||
6399 | skb = alloc_skb(0, GFP_KERNEL); | |
6400 | if (!skb) { | |
6401 | kfree(fpl); | |
6402 | return -ENOMEM; | |
6403 | } | |
6404 | ||
6405 | skb->sk = sk; | |
6b06314c | 6406 | |
08a45173 | 6407 | nr_files = 0; |
6b06314c JA |
6408 | fpl->user = get_uid(ctx->user); |
6409 | for (i = 0; i < nr; i++) { | |
65e19f54 JA |
6410 | struct file *file = io_file_from_index(ctx, i + offset); |
6411 | ||
6412 | if (!file) | |
08a45173 | 6413 | continue; |
65e19f54 | 6414 | fpl->fp[nr_files] = get_file(file); |
08a45173 JA |
6415 | unix_inflight(fpl->user, fpl->fp[nr_files]); |
6416 | nr_files++; | |
6b06314c JA |
6417 | } |
6418 | ||
08a45173 JA |
6419 | if (nr_files) { |
6420 | fpl->max = SCM_MAX_FD; | |
6421 | fpl->count = nr_files; | |
6422 | UNIXCB(skb).fp = fpl; | |
05f3fb3c | 6423 | skb->destructor = unix_destruct_scm; |
08a45173 JA |
6424 | refcount_add(skb->truesize, &sk->sk_wmem_alloc); |
6425 | skb_queue_head(&sk->sk_receive_queue, skb); | |
6b06314c | 6426 | |
08a45173 JA |
6427 | for (i = 0; i < nr_files; i++) |
6428 | fput(fpl->fp[i]); | |
6429 | } else { | |
6430 | kfree_skb(skb); | |
6431 | kfree(fpl); | |
6432 | } | |
6b06314c JA |
6433 | |
6434 | return 0; | |
6435 | } | |
6436 | ||
6437 | /* | |
6438 | * If UNIX sockets are enabled, fd passing can cause a reference cycle which | |
6439 | * causes regular reference counting to break down. We rely on the UNIX | |
6440 | * garbage collection to take care of this problem for us. | |
6441 | */ | |
6442 | static int io_sqe_files_scm(struct io_ring_ctx *ctx) | |
6443 | { | |
6444 | unsigned left, total; | |
6445 | int ret = 0; | |
6446 | ||
6447 | total = 0; | |
6448 | left = ctx->nr_user_files; | |
6449 | while (left) { | |
6450 | unsigned this_files = min_t(unsigned, left, SCM_MAX_FD); | |
6b06314c JA |
6451 | |
6452 | ret = __io_sqe_files_scm(ctx, this_files, total); | |
6453 | if (ret) | |
6454 | break; | |
6455 | left -= this_files; | |
6456 | total += this_files; | |
6457 | } | |
6458 | ||
6459 | if (!ret) | |
6460 | return 0; | |
6461 | ||
6462 | while (total < ctx->nr_user_files) { | |
65e19f54 JA |
6463 | struct file *file = io_file_from_index(ctx, total); |
6464 | ||
6465 | if (file) | |
6466 | fput(file); | |
6b06314c JA |
6467 | total++; |
6468 | } | |
6469 | ||
6470 | return ret; | |
6471 | } | |
6472 | #else | |
6473 | static int io_sqe_files_scm(struct io_ring_ctx *ctx) | |
6474 | { | |
6475 | return 0; | |
6476 | } | |
6477 | #endif | |
6478 | ||
65e19f54 JA |
6479 | static int io_sqe_alloc_file_tables(struct io_ring_ctx *ctx, unsigned nr_tables, |
6480 | unsigned nr_files) | |
6481 | { | |
6482 | int i; | |
6483 | ||
6484 | for (i = 0; i < nr_tables; i++) { | |
05f3fb3c | 6485 | struct fixed_file_table *table = &ctx->file_data->table[i]; |
65e19f54 JA |
6486 | unsigned this_files; |
6487 | ||
6488 | this_files = min(nr_files, IORING_MAX_FILES_TABLE); | |
6489 | table->files = kcalloc(this_files, sizeof(struct file *), | |
6490 | GFP_KERNEL); | |
6491 | if (!table->files) | |
6492 | break; | |
6493 | nr_files -= this_files; | |
6494 | } | |
6495 | ||
6496 | if (i == nr_tables) | |
6497 | return 0; | |
6498 | ||
6499 | for (i = 0; i < nr_tables; i++) { | |
05f3fb3c | 6500 | struct fixed_file_table *table = &ctx->file_data->table[i]; |
65e19f54 JA |
6501 | kfree(table->files); |
6502 | } | |
6503 | return 1; | |
6504 | } | |
6505 | ||
05f3fb3c JA |
6506 | static void io_ring_file_put(struct io_ring_ctx *ctx, struct file *file) |
6507 | { | |
6508 | #if defined(CONFIG_UNIX) | |
6509 | struct sock *sock = ctx->ring_sock->sk; | |
6510 | struct sk_buff_head list, *head = &sock->sk_receive_queue; | |
6511 | struct sk_buff *skb; | |
6512 | int i; | |
6513 | ||
6514 | __skb_queue_head_init(&list); | |
6515 | ||
6516 | /* | |
6517 | * Find the skb that holds this file in its SCM_RIGHTS. When found, | |
6518 | * remove this entry and rearrange the file array. | |
6519 | */ | |
6520 | skb = skb_dequeue(head); | |
6521 | while (skb) { | |
6522 | struct scm_fp_list *fp; | |
6523 | ||
6524 | fp = UNIXCB(skb).fp; | |
6525 | for (i = 0; i < fp->count; i++) { | |
6526 | int left; | |
6527 | ||
6528 | if (fp->fp[i] != file) | |
6529 | continue; | |
6530 | ||
6531 | unix_notinflight(fp->user, fp->fp[i]); | |
6532 | left = fp->count - 1 - i; | |
6533 | if (left) { | |
6534 | memmove(&fp->fp[i], &fp->fp[i + 1], | |
6535 | left * sizeof(struct file *)); | |
6536 | } | |
6537 | fp->count--; | |
6538 | if (!fp->count) { | |
6539 | kfree_skb(skb); | |
6540 | skb = NULL; | |
6541 | } else { | |
6542 | __skb_queue_tail(&list, skb); | |
6543 | } | |
6544 | fput(file); | |
6545 | file = NULL; | |
6546 | break; | |
6547 | } | |
6548 | ||
6549 | if (!file) | |
6550 | break; | |
6551 | ||
6552 | __skb_queue_tail(&list, skb); | |
6553 | ||
6554 | skb = skb_dequeue(head); | |
6555 | } | |
6556 | ||
6557 | if (skb_peek(&list)) { | |
6558 | spin_lock_irq(&head->lock); | |
6559 | while ((skb = __skb_dequeue(&list)) != NULL) | |
6560 | __skb_queue_tail(head, skb); | |
6561 | spin_unlock_irq(&head->lock); | |
6562 | } | |
6563 | #else | |
6564 | fput(file); | |
6565 | #endif | |
6566 | } | |
6567 | ||
6568 | struct io_file_put { | |
05589553 | 6569 | struct list_head list; |
05f3fb3c | 6570 | struct file *file; |
05f3fb3c JA |
6571 | }; |
6572 | ||
4a38aed2 | 6573 | static void __io_file_put_work(struct fixed_file_ref_node *ref_node) |
65e19f54 | 6574 | { |
4a38aed2 JA |
6575 | struct fixed_file_data *file_data = ref_node->file_data; |
6576 | struct io_ring_ctx *ctx = file_data->ctx; | |
05f3fb3c | 6577 | struct io_file_put *pfile, *tmp; |
65e19f54 | 6578 | |
05589553 | 6579 | list_for_each_entry_safe(pfile, tmp, &ref_node->file_list, list) { |
6a4d07cd | 6580 | list_del(&pfile->list); |
05589553 XW |
6581 | io_ring_file_put(ctx, pfile->file); |
6582 | kfree(pfile); | |
65e19f54 | 6583 | } |
05589553 | 6584 | |
6a4d07cd JA |
6585 | spin_lock(&file_data->lock); |
6586 | list_del(&ref_node->node); | |
6587 | spin_unlock(&file_data->lock); | |
05589553 XW |
6588 | |
6589 | percpu_ref_exit(&ref_node->refs); | |
6590 | kfree(ref_node); | |
6591 | percpu_ref_put(&file_data->refs); | |
2faf852d | 6592 | } |
65e19f54 | 6593 | |
4a38aed2 JA |
6594 | static void io_file_put_work(struct work_struct *work) |
6595 | { | |
6596 | struct io_ring_ctx *ctx; | |
6597 | struct llist_node *node; | |
6598 | ||
6599 | ctx = container_of(work, struct io_ring_ctx, file_put_work.work); | |
6600 | node = llist_del_all(&ctx->file_put_llist); | |
6601 | ||
6602 | while (node) { | |
6603 | struct fixed_file_ref_node *ref_node; | |
6604 | struct llist_node *next = node->next; | |
6605 | ||
6606 | ref_node = llist_entry(node, struct fixed_file_ref_node, llist); | |
6607 | __io_file_put_work(ref_node); | |
6608 | node = next; | |
6609 | } | |
6610 | } | |
6611 | ||
05589553 | 6612 | static void io_file_data_ref_zero(struct percpu_ref *ref) |
2faf852d | 6613 | { |
05589553 | 6614 | struct fixed_file_ref_node *ref_node; |
4a38aed2 JA |
6615 | struct io_ring_ctx *ctx; |
6616 | bool first_add; | |
6617 | int delay = HZ; | |
65e19f54 | 6618 | |
05589553 | 6619 | ref_node = container_of(ref, struct fixed_file_ref_node, refs); |
4a38aed2 | 6620 | ctx = ref_node->file_data->ctx; |
05589553 | 6621 | |
4a38aed2 JA |
6622 | if (percpu_ref_is_dying(&ctx->file_data->refs)) |
6623 | delay = 0; | |
6624 | ||
6625 | first_add = llist_add(&ref_node->llist, &ctx->file_put_llist); | |
6626 | if (!delay) | |
6627 | mod_delayed_work(system_wq, &ctx->file_put_work, 0); | |
6628 | else if (first_add) | |
6629 | queue_delayed_work(system_wq, &ctx->file_put_work, delay); | |
05f3fb3c | 6630 | } |
65e19f54 | 6631 | |
05589553 XW |
6632 | static struct fixed_file_ref_node *alloc_fixed_file_ref_node( |
6633 | struct io_ring_ctx *ctx) | |
05f3fb3c | 6634 | { |
05589553 | 6635 | struct fixed_file_ref_node *ref_node; |
05f3fb3c | 6636 | |
05589553 XW |
6637 | ref_node = kzalloc(sizeof(*ref_node), GFP_KERNEL); |
6638 | if (!ref_node) | |
6639 | return ERR_PTR(-ENOMEM); | |
05f3fb3c | 6640 | |
05589553 XW |
6641 | if (percpu_ref_init(&ref_node->refs, io_file_data_ref_zero, |
6642 | 0, GFP_KERNEL)) { | |
6643 | kfree(ref_node); | |
6644 | return ERR_PTR(-ENOMEM); | |
6645 | } | |
6646 | INIT_LIST_HEAD(&ref_node->node); | |
6647 | INIT_LIST_HEAD(&ref_node->file_list); | |
05589553 XW |
6648 | ref_node->file_data = ctx->file_data; |
6649 | return ref_node; | |
05589553 XW |
6650 | } |
6651 | ||
6652 | static void destroy_fixed_file_ref_node(struct fixed_file_ref_node *ref_node) | |
6653 | { | |
6654 | percpu_ref_exit(&ref_node->refs); | |
6655 | kfree(ref_node); | |
65e19f54 JA |
6656 | } |
6657 | ||
6b06314c JA |
6658 | static int io_sqe_files_register(struct io_ring_ctx *ctx, void __user *arg, |
6659 | unsigned nr_args) | |
6660 | { | |
6661 | __s32 __user *fds = (__s32 __user *) arg; | |
65e19f54 | 6662 | unsigned nr_tables; |
05f3fb3c | 6663 | struct file *file; |
6b06314c JA |
6664 | int fd, ret = 0; |
6665 | unsigned i; | |
05589553 | 6666 | struct fixed_file_ref_node *ref_node; |
6b06314c | 6667 | |
05f3fb3c | 6668 | if (ctx->file_data) |
6b06314c JA |
6669 | return -EBUSY; |
6670 | if (!nr_args) | |
6671 | return -EINVAL; | |
6672 | if (nr_args > IORING_MAX_FIXED_FILES) | |
6673 | return -EMFILE; | |
6674 | ||
05f3fb3c JA |
6675 | ctx->file_data = kzalloc(sizeof(*ctx->file_data), GFP_KERNEL); |
6676 | if (!ctx->file_data) | |
6677 | return -ENOMEM; | |
6678 | ctx->file_data->ctx = ctx; | |
6679 | init_completion(&ctx->file_data->done); | |
05589553 | 6680 | INIT_LIST_HEAD(&ctx->file_data->ref_list); |
f7fe9346 | 6681 | spin_lock_init(&ctx->file_data->lock); |
05f3fb3c | 6682 | |
65e19f54 | 6683 | nr_tables = DIV_ROUND_UP(nr_args, IORING_MAX_FILES_TABLE); |
05f3fb3c JA |
6684 | ctx->file_data->table = kcalloc(nr_tables, |
6685 | sizeof(struct fixed_file_table), | |
65e19f54 | 6686 | GFP_KERNEL); |
05f3fb3c JA |
6687 | if (!ctx->file_data->table) { |
6688 | kfree(ctx->file_data); | |
6689 | ctx->file_data = NULL; | |
6b06314c | 6690 | return -ENOMEM; |
05f3fb3c JA |
6691 | } |
6692 | ||
05589553 | 6693 | if (percpu_ref_init(&ctx->file_data->refs, io_file_ref_kill, |
05f3fb3c JA |
6694 | PERCPU_REF_ALLOW_REINIT, GFP_KERNEL)) { |
6695 | kfree(ctx->file_data->table); | |
6696 | kfree(ctx->file_data); | |
6697 | ctx->file_data = NULL; | |
6b06314c | 6698 | return -ENOMEM; |
05f3fb3c | 6699 | } |
6b06314c | 6700 | |
65e19f54 | 6701 | if (io_sqe_alloc_file_tables(ctx, nr_tables, nr_args)) { |
05f3fb3c JA |
6702 | percpu_ref_exit(&ctx->file_data->refs); |
6703 | kfree(ctx->file_data->table); | |
6704 | kfree(ctx->file_data); | |
6705 | ctx->file_data = NULL; | |
65e19f54 JA |
6706 | return -ENOMEM; |
6707 | } | |
6708 | ||
08a45173 | 6709 | for (i = 0; i < nr_args; i++, ctx->nr_user_files++) { |
65e19f54 JA |
6710 | struct fixed_file_table *table; |
6711 | unsigned index; | |
6712 | ||
6b06314c JA |
6713 | ret = -EFAULT; |
6714 | if (copy_from_user(&fd, &fds[i], sizeof(fd))) | |
6715 | break; | |
08a45173 JA |
6716 | /* allow sparse sets */ |
6717 | if (fd == -1) { | |
6718 | ret = 0; | |
6719 | continue; | |
6720 | } | |
6b06314c | 6721 | |
05f3fb3c | 6722 | table = &ctx->file_data->table[i >> IORING_FILE_TABLE_SHIFT]; |
65e19f54 | 6723 | index = i & IORING_FILE_TABLE_MASK; |
05f3fb3c | 6724 | file = fget(fd); |
6b06314c JA |
6725 | |
6726 | ret = -EBADF; | |
05f3fb3c | 6727 | if (!file) |
6b06314c | 6728 | break; |
05f3fb3c | 6729 | |
6b06314c JA |
6730 | /* |
6731 | * Don't allow io_uring instances to be registered. If UNIX | |
6732 | * isn't enabled, then this causes a reference cycle and this | |
6733 | * instance can never get freed. If UNIX is enabled we'll | |
6734 | * handle it just fine, but there's still no point in allowing | |
6735 | * a ring fd as it doesn't support regular read/write anyway. | |
6736 | */ | |
05f3fb3c JA |
6737 | if (file->f_op == &io_uring_fops) { |
6738 | fput(file); | |
6b06314c JA |
6739 | break; |
6740 | } | |
6b06314c | 6741 | ret = 0; |
05f3fb3c | 6742 | table->files[index] = file; |
6b06314c JA |
6743 | } |
6744 | ||
6745 | if (ret) { | |
65e19f54 | 6746 | for (i = 0; i < ctx->nr_user_files; i++) { |
65e19f54 JA |
6747 | file = io_file_from_index(ctx, i); |
6748 | if (file) | |
6749 | fput(file); | |
6750 | } | |
6751 | for (i = 0; i < nr_tables; i++) | |
05f3fb3c | 6752 | kfree(ctx->file_data->table[i].files); |
6b06314c | 6753 | |
05f3fb3c JA |
6754 | kfree(ctx->file_data->table); |
6755 | kfree(ctx->file_data); | |
6756 | ctx->file_data = NULL; | |
6b06314c JA |
6757 | ctx->nr_user_files = 0; |
6758 | return ret; | |
6759 | } | |
6760 | ||
6761 | ret = io_sqe_files_scm(ctx); | |
05589553 | 6762 | if (ret) { |
6b06314c | 6763 | io_sqe_files_unregister(ctx); |
05589553 XW |
6764 | return ret; |
6765 | } | |
6b06314c | 6766 | |
05589553 XW |
6767 | ref_node = alloc_fixed_file_ref_node(ctx); |
6768 | if (IS_ERR(ref_node)) { | |
6769 | io_sqe_files_unregister(ctx); | |
6770 | return PTR_ERR(ref_node); | |
6771 | } | |
6772 | ||
6773 | ctx->file_data->cur_refs = &ref_node->refs; | |
6a4d07cd | 6774 | spin_lock(&ctx->file_data->lock); |
05589553 | 6775 | list_add(&ref_node->node, &ctx->file_data->ref_list); |
6a4d07cd | 6776 | spin_unlock(&ctx->file_data->lock); |
05589553 | 6777 | percpu_ref_get(&ctx->file_data->refs); |
6b06314c JA |
6778 | return ret; |
6779 | } | |
6780 | ||
c3a31e60 JA |
6781 | static int io_sqe_file_register(struct io_ring_ctx *ctx, struct file *file, |
6782 | int index) | |
6783 | { | |
6784 | #if defined(CONFIG_UNIX) | |
6785 | struct sock *sock = ctx->ring_sock->sk; | |
6786 | struct sk_buff_head *head = &sock->sk_receive_queue; | |
6787 | struct sk_buff *skb; | |
6788 | ||
6789 | /* | |
6790 | * See if we can merge this file into an existing skb SCM_RIGHTS | |
6791 | * file set. If there's no room, fall back to allocating a new skb | |
6792 | * and filling it in. | |
6793 | */ | |
6794 | spin_lock_irq(&head->lock); | |
6795 | skb = skb_peek(head); | |
6796 | if (skb) { | |
6797 | struct scm_fp_list *fpl = UNIXCB(skb).fp; | |
6798 | ||
6799 | if (fpl->count < SCM_MAX_FD) { | |
6800 | __skb_unlink(skb, head); | |
6801 | spin_unlock_irq(&head->lock); | |
6802 | fpl->fp[fpl->count] = get_file(file); | |
6803 | unix_inflight(fpl->user, fpl->fp[fpl->count]); | |
6804 | fpl->count++; | |
6805 | spin_lock_irq(&head->lock); | |
6806 | __skb_queue_head(head, skb); | |
6807 | } else { | |
6808 | skb = NULL; | |
6809 | } | |
6810 | } | |
6811 | spin_unlock_irq(&head->lock); | |
6812 | ||
6813 | if (skb) { | |
6814 | fput(file); | |
6815 | return 0; | |
6816 | } | |
6817 | ||
6818 | return __io_sqe_files_scm(ctx, 1, index); | |
6819 | #else | |
6820 | return 0; | |
6821 | #endif | |
6822 | } | |
6823 | ||
a5318d3c | 6824 | static int io_queue_file_removal(struct fixed_file_data *data, |
05589553 | 6825 | struct file *file) |
05f3fb3c | 6826 | { |
a5318d3c | 6827 | struct io_file_put *pfile; |
05589553 XW |
6828 | struct percpu_ref *refs = data->cur_refs; |
6829 | struct fixed_file_ref_node *ref_node; | |
05f3fb3c | 6830 | |
05f3fb3c | 6831 | pfile = kzalloc(sizeof(*pfile), GFP_KERNEL); |
a5318d3c HD |
6832 | if (!pfile) |
6833 | return -ENOMEM; | |
05f3fb3c | 6834 | |
05589553 | 6835 | ref_node = container_of(refs, struct fixed_file_ref_node, refs); |
05f3fb3c | 6836 | pfile->file = file; |
05589553 XW |
6837 | list_add(&pfile->list, &ref_node->file_list); |
6838 | ||
a5318d3c | 6839 | return 0; |
05f3fb3c JA |
6840 | } |
6841 | ||
6842 | static int __io_sqe_files_update(struct io_ring_ctx *ctx, | |
6843 | struct io_uring_files_update *up, | |
6844 | unsigned nr_args) | |
6845 | { | |
6846 | struct fixed_file_data *data = ctx->file_data; | |
05589553 | 6847 | struct fixed_file_ref_node *ref_node; |
05f3fb3c | 6848 | struct file *file; |
c3a31e60 JA |
6849 | __s32 __user *fds; |
6850 | int fd, i, err; | |
6851 | __u32 done; | |
05589553 | 6852 | bool needs_switch = false; |
c3a31e60 | 6853 | |
05f3fb3c | 6854 | if (check_add_overflow(up->offset, nr_args, &done)) |
c3a31e60 JA |
6855 | return -EOVERFLOW; |
6856 | if (done > ctx->nr_user_files) | |
6857 | return -EINVAL; | |
6858 | ||
05589553 XW |
6859 | ref_node = alloc_fixed_file_ref_node(ctx); |
6860 | if (IS_ERR(ref_node)) | |
6861 | return PTR_ERR(ref_node); | |
6862 | ||
c3a31e60 | 6863 | done = 0; |
05f3fb3c | 6864 | fds = u64_to_user_ptr(up->fds); |
c3a31e60 | 6865 | while (nr_args) { |
65e19f54 JA |
6866 | struct fixed_file_table *table; |
6867 | unsigned index; | |
6868 | ||
c3a31e60 JA |
6869 | err = 0; |
6870 | if (copy_from_user(&fd, &fds[done], sizeof(fd))) { | |
6871 | err = -EFAULT; | |
6872 | break; | |
6873 | } | |
05f3fb3c JA |
6874 | i = array_index_nospec(up->offset, ctx->nr_user_files); |
6875 | table = &ctx->file_data->table[i >> IORING_FILE_TABLE_SHIFT]; | |
65e19f54 JA |
6876 | index = i & IORING_FILE_TABLE_MASK; |
6877 | if (table->files[index]) { | |
05f3fb3c | 6878 | file = io_file_from_index(ctx, index); |
a5318d3c HD |
6879 | err = io_queue_file_removal(data, file); |
6880 | if (err) | |
6881 | break; | |
65e19f54 | 6882 | table->files[index] = NULL; |
05589553 | 6883 | needs_switch = true; |
c3a31e60 JA |
6884 | } |
6885 | if (fd != -1) { | |
c3a31e60 JA |
6886 | file = fget(fd); |
6887 | if (!file) { | |
6888 | err = -EBADF; | |
6889 | break; | |
6890 | } | |
6891 | /* | |
6892 | * Don't allow io_uring instances to be registered. If | |
6893 | * UNIX isn't enabled, then this causes a reference | |
6894 | * cycle and this instance can never get freed. If UNIX | |
6895 | * is enabled we'll handle it just fine, but there's | |
6896 | * still no point in allowing a ring fd as it doesn't | |
6897 | * support regular read/write anyway. | |
6898 | */ | |
6899 | if (file->f_op == &io_uring_fops) { | |
6900 | fput(file); | |
6901 | err = -EBADF; | |
6902 | break; | |
6903 | } | |
65e19f54 | 6904 | table->files[index] = file; |
c3a31e60 JA |
6905 | err = io_sqe_file_register(ctx, file, i); |
6906 | if (err) | |
6907 | break; | |
6908 | } | |
6909 | nr_args--; | |
6910 | done++; | |
05f3fb3c JA |
6911 | up->offset++; |
6912 | } | |
6913 | ||
05589553 XW |
6914 | if (needs_switch) { |
6915 | percpu_ref_kill(data->cur_refs); | |
6a4d07cd | 6916 | spin_lock(&data->lock); |
05589553 XW |
6917 | list_add(&ref_node->node, &data->ref_list); |
6918 | data->cur_refs = &ref_node->refs; | |
6a4d07cd | 6919 | spin_unlock(&data->lock); |
05589553 XW |
6920 | percpu_ref_get(&ctx->file_data->refs); |
6921 | } else | |
6922 | destroy_fixed_file_ref_node(ref_node); | |
c3a31e60 JA |
6923 | |
6924 | return done ? done : err; | |
6925 | } | |
05589553 | 6926 | |
05f3fb3c JA |
6927 | static int io_sqe_files_update(struct io_ring_ctx *ctx, void __user *arg, |
6928 | unsigned nr_args) | |
6929 | { | |
6930 | struct io_uring_files_update up; | |
6931 | ||
6932 | if (!ctx->file_data) | |
6933 | return -ENXIO; | |
6934 | if (!nr_args) | |
6935 | return -EINVAL; | |
6936 | if (copy_from_user(&up, arg, sizeof(up))) | |
6937 | return -EFAULT; | |
6938 | if (up.resv) | |
6939 | return -EINVAL; | |
6940 | ||
6941 | return __io_sqe_files_update(ctx, &up, nr_args); | |
6942 | } | |
c3a31e60 | 6943 | |
e9fd9396 | 6944 | static void io_free_work(struct io_wq_work *work) |
7d723065 JA |
6945 | { |
6946 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); | |
6947 | ||
e9fd9396 | 6948 | /* Consider that io_steal_work() relies on this ref */ |
7d723065 JA |
6949 | io_put_req(req); |
6950 | } | |
6951 | ||
24369c2e PB |
6952 | static int io_init_wq_offload(struct io_ring_ctx *ctx, |
6953 | struct io_uring_params *p) | |
6954 | { | |
6955 | struct io_wq_data data; | |
6956 | struct fd f; | |
6957 | struct io_ring_ctx *ctx_attach; | |
6958 | unsigned int concurrency; | |
6959 | int ret = 0; | |
6960 | ||
6961 | data.user = ctx->user; | |
e9fd9396 | 6962 | data.free_work = io_free_work; |
24369c2e PB |
6963 | |
6964 | if (!(p->flags & IORING_SETUP_ATTACH_WQ)) { | |
6965 | /* Do QD, or 4 * CPUS, whatever is smallest */ | |
6966 | concurrency = min(ctx->sq_entries, 4 * num_online_cpus()); | |
6967 | ||
6968 | ctx->io_wq = io_wq_create(concurrency, &data); | |
6969 | if (IS_ERR(ctx->io_wq)) { | |
6970 | ret = PTR_ERR(ctx->io_wq); | |
6971 | ctx->io_wq = NULL; | |
6972 | } | |
6973 | return ret; | |
6974 | } | |
6975 | ||
6976 | f = fdget(p->wq_fd); | |
6977 | if (!f.file) | |
6978 | return -EBADF; | |
6979 | ||
6980 | if (f.file->f_op != &io_uring_fops) { | |
6981 | ret = -EINVAL; | |
6982 | goto out_fput; | |
6983 | } | |
6984 | ||
6985 | ctx_attach = f.file->private_data; | |
6986 | /* @io_wq is protected by holding the fd */ | |
6987 | if (!io_wq_get(ctx_attach->io_wq, &data)) { | |
6988 | ret = -EINVAL; | |
6989 | goto out_fput; | |
6990 | } | |
6991 | ||
6992 | ctx->io_wq = ctx_attach->io_wq; | |
6993 | out_fput: | |
6994 | fdput(f); | |
6995 | return ret; | |
6996 | } | |
6997 | ||
6c271ce2 JA |
6998 | static int io_sq_offload_start(struct io_ring_ctx *ctx, |
6999 | struct io_uring_params *p) | |
2b188cc1 JA |
7000 | { |
7001 | int ret; | |
7002 | ||
6c271ce2 | 7003 | init_waitqueue_head(&ctx->sqo_wait); |
2b188cc1 JA |
7004 | mmgrab(current->mm); |
7005 | ctx->sqo_mm = current->mm; | |
7006 | ||
6c271ce2 | 7007 | if (ctx->flags & IORING_SETUP_SQPOLL) { |
3ec482d1 JA |
7008 | ret = -EPERM; |
7009 | if (!capable(CAP_SYS_ADMIN)) | |
7010 | goto err; | |
7011 | ||
917257da JA |
7012 | ctx->sq_thread_idle = msecs_to_jiffies(p->sq_thread_idle); |
7013 | if (!ctx->sq_thread_idle) | |
7014 | ctx->sq_thread_idle = HZ; | |
7015 | ||
6c271ce2 | 7016 | if (p->flags & IORING_SETUP_SQ_AFF) { |
44a9bd18 | 7017 | int cpu = p->sq_thread_cpu; |
6c271ce2 | 7018 | |
917257da | 7019 | ret = -EINVAL; |
44a9bd18 JA |
7020 | if (cpu >= nr_cpu_ids) |
7021 | goto err; | |
7889f44d | 7022 | if (!cpu_online(cpu)) |
917257da JA |
7023 | goto err; |
7024 | ||
6c271ce2 JA |
7025 | ctx->sqo_thread = kthread_create_on_cpu(io_sq_thread, |
7026 | ctx, cpu, | |
7027 | "io_uring-sq"); | |
7028 | } else { | |
7029 | ctx->sqo_thread = kthread_create(io_sq_thread, ctx, | |
7030 | "io_uring-sq"); | |
7031 | } | |
7032 | if (IS_ERR(ctx->sqo_thread)) { | |
7033 | ret = PTR_ERR(ctx->sqo_thread); | |
7034 | ctx->sqo_thread = NULL; | |
7035 | goto err; | |
7036 | } | |
7037 | wake_up_process(ctx->sqo_thread); | |
7038 | } else if (p->flags & IORING_SETUP_SQ_AFF) { | |
7039 | /* Can't have SQ_AFF without SQPOLL */ | |
7040 | ret = -EINVAL; | |
7041 | goto err; | |
7042 | } | |
7043 | ||
24369c2e PB |
7044 | ret = io_init_wq_offload(ctx, p); |
7045 | if (ret) | |
2b188cc1 | 7046 | goto err; |
2b188cc1 JA |
7047 | |
7048 | return 0; | |
7049 | err: | |
54a91f3b | 7050 | io_finish_async(ctx); |
2b188cc1 JA |
7051 | mmdrop(ctx->sqo_mm); |
7052 | ctx->sqo_mm = NULL; | |
7053 | return ret; | |
7054 | } | |
7055 | ||
7056 | static void io_unaccount_mem(struct user_struct *user, unsigned long nr_pages) | |
7057 | { | |
7058 | atomic_long_sub(nr_pages, &user->locked_vm); | |
7059 | } | |
7060 | ||
7061 | static int io_account_mem(struct user_struct *user, unsigned long nr_pages) | |
7062 | { | |
7063 | unsigned long page_limit, cur_pages, new_pages; | |
7064 | ||
7065 | /* Don't allow more pages than we can safely lock */ | |
7066 | page_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT; | |
7067 | ||
7068 | do { | |
7069 | cur_pages = atomic_long_read(&user->locked_vm); | |
7070 | new_pages = cur_pages + nr_pages; | |
7071 | if (new_pages > page_limit) | |
7072 | return -ENOMEM; | |
7073 | } while (atomic_long_cmpxchg(&user->locked_vm, cur_pages, | |
7074 | new_pages) != cur_pages); | |
7075 | ||
7076 | return 0; | |
7077 | } | |
7078 | ||
7079 | static void io_mem_free(void *ptr) | |
7080 | { | |
52e04ef4 MR |
7081 | struct page *page; |
7082 | ||
7083 | if (!ptr) | |
7084 | return; | |
2b188cc1 | 7085 | |
52e04ef4 | 7086 | page = virt_to_head_page(ptr); |
2b188cc1 JA |
7087 | if (put_page_testzero(page)) |
7088 | free_compound_page(page); | |
7089 | } | |
7090 | ||
7091 | static void *io_mem_alloc(size_t size) | |
7092 | { | |
7093 | gfp_t gfp_flags = GFP_KERNEL | __GFP_ZERO | __GFP_NOWARN | __GFP_COMP | | |
7094 | __GFP_NORETRY; | |
7095 | ||
7096 | return (void *) __get_free_pages(gfp_flags, get_order(size)); | |
7097 | } | |
7098 | ||
75b28aff HV |
7099 | static unsigned long rings_size(unsigned sq_entries, unsigned cq_entries, |
7100 | size_t *sq_offset) | |
7101 | { | |
7102 | struct io_rings *rings; | |
7103 | size_t off, sq_array_size; | |
7104 | ||
7105 | off = struct_size(rings, cqes, cq_entries); | |
7106 | if (off == SIZE_MAX) | |
7107 | return SIZE_MAX; | |
7108 | ||
7109 | #ifdef CONFIG_SMP | |
7110 | off = ALIGN(off, SMP_CACHE_BYTES); | |
7111 | if (off == 0) | |
7112 | return SIZE_MAX; | |
7113 | #endif | |
7114 | ||
7115 | sq_array_size = array_size(sizeof(u32), sq_entries); | |
7116 | if (sq_array_size == SIZE_MAX) | |
7117 | return SIZE_MAX; | |
7118 | ||
7119 | if (check_add_overflow(off, sq_array_size, &off)) | |
7120 | return SIZE_MAX; | |
7121 | ||
7122 | if (sq_offset) | |
7123 | *sq_offset = off; | |
7124 | ||
7125 | return off; | |
7126 | } | |
7127 | ||
2b188cc1 JA |
7128 | static unsigned long ring_pages(unsigned sq_entries, unsigned cq_entries) |
7129 | { | |
75b28aff | 7130 | size_t pages; |
2b188cc1 | 7131 | |
75b28aff HV |
7132 | pages = (size_t)1 << get_order( |
7133 | rings_size(sq_entries, cq_entries, NULL)); | |
7134 | pages += (size_t)1 << get_order( | |
7135 | array_size(sizeof(struct io_uring_sqe), sq_entries)); | |
2b188cc1 | 7136 | |
75b28aff | 7137 | return pages; |
2b188cc1 JA |
7138 | } |
7139 | ||
edafccee JA |
7140 | static int io_sqe_buffer_unregister(struct io_ring_ctx *ctx) |
7141 | { | |
7142 | int i, j; | |
7143 | ||
7144 | if (!ctx->user_bufs) | |
7145 | return -ENXIO; | |
7146 | ||
7147 | for (i = 0; i < ctx->nr_user_bufs; i++) { | |
7148 | struct io_mapped_ubuf *imu = &ctx->user_bufs[i]; | |
7149 | ||
7150 | for (j = 0; j < imu->nr_bvecs; j++) | |
f1f6a7dd | 7151 | unpin_user_page(imu->bvec[j].bv_page); |
edafccee JA |
7152 | |
7153 | if (ctx->account_mem) | |
7154 | io_unaccount_mem(ctx->user, imu->nr_bvecs); | |
d4ef6475 | 7155 | kvfree(imu->bvec); |
edafccee JA |
7156 | imu->nr_bvecs = 0; |
7157 | } | |
7158 | ||
7159 | kfree(ctx->user_bufs); | |
7160 | ctx->user_bufs = NULL; | |
7161 | ctx->nr_user_bufs = 0; | |
7162 | return 0; | |
7163 | } | |
7164 | ||
7165 | static int io_copy_iov(struct io_ring_ctx *ctx, struct iovec *dst, | |
7166 | void __user *arg, unsigned index) | |
7167 | { | |
7168 | struct iovec __user *src; | |
7169 | ||
7170 | #ifdef CONFIG_COMPAT | |
7171 | if (ctx->compat) { | |
7172 | struct compat_iovec __user *ciovs; | |
7173 | struct compat_iovec ciov; | |
7174 | ||
7175 | ciovs = (struct compat_iovec __user *) arg; | |
7176 | if (copy_from_user(&ciov, &ciovs[index], sizeof(ciov))) | |
7177 | return -EFAULT; | |
7178 | ||
d55e5f5b | 7179 | dst->iov_base = u64_to_user_ptr((u64)ciov.iov_base); |
edafccee JA |
7180 | dst->iov_len = ciov.iov_len; |
7181 | return 0; | |
7182 | } | |
7183 | #endif | |
7184 | src = (struct iovec __user *) arg; | |
7185 | if (copy_from_user(dst, &src[index], sizeof(*dst))) | |
7186 | return -EFAULT; | |
7187 | return 0; | |
7188 | } | |
7189 | ||
7190 | static int io_sqe_buffer_register(struct io_ring_ctx *ctx, void __user *arg, | |
7191 | unsigned nr_args) | |
7192 | { | |
7193 | struct vm_area_struct **vmas = NULL; | |
7194 | struct page **pages = NULL; | |
7195 | int i, j, got_pages = 0; | |
7196 | int ret = -EINVAL; | |
7197 | ||
7198 | if (ctx->user_bufs) | |
7199 | return -EBUSY; | |
7200 | if (!nr_args || nr_args > UIO_MAXIOV) | |
7201 | return -EINVAL; | |
7202 | ||
7203 | ctx->user_bufs = kcalloc(nr_args, sizeof(struct io_mapped_ubuf), | |
7204 | GFP_KERNEL); | |
7205 | if (!ctx->user_bufs) | |
7206 | return -ENOMEM; | |
7207 | ||
7208 | for (i = 0; i < nr_args; i++) { | |
7209 | struct io_mapped_ubuf *imu = &ctx->user_bufs[i]; | |
7210 | unsigned long off, start, end, ubuf; | |
7211 | int pret, nr_pages; | |
7212 | struct iovec iov; | |
7213 | size_t size; | |
7214 | ||
7215 | ret = io_copy_iov(ctx, &iov, arg, i); | |
7216 | if (ret) | |
a278682d | 7217 | goto err; |
edafccee JA |
7218 | |
7219 | /* | |
7220 | * Don't impose further limits on the size and buffer | |
7221 | * constraints here, we'll -EINVAL later when IO is | |
7222 | * submitted if they are wrong. | |
7223 | */ | |
7224 | ret = -EFAULT; | |
7225 | if (!iov.iov_base || !iov.iov_len) | |
7226 | goto err; | |
7227 | ||
7228 | /* arbitrary limit, but we need something */ | |
7229 | if (iov.iov_len > SZ_1G) | |
7230 | goto err; | |
7231 | ||
7232 | ubuf = (unsigned long) iov.iov_base; | |
7233 | end = (ubuf + iov.iov_len + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
7234 | start = ubuf >> PAGE_SHIFT; | |
7235 | nr_pages = end - start; | |
7236 | ||
7237 | if (ctx->account_mem) { | |
7238 | ret = io_account_mem(ctx->user, nr_pages); | |
7239 | if (ret) | |
7240 | goto err; | |
7241 | } | |
7242 | ||
7243 | ret = 0; | |
7244 | if (!pages || nr_pages > got_pages) { | |
7245 | kfree(vmas); | |
7246 | kfree(pages); | |
d4ef6475 | 7247 | pages = kvmalloc_array(nr_pages, sizeof(struct page *), |
edafccee | 7248 | GFP_KERNEL); |
d4ef6475 | 7249 | vmas = kvmalloc_array(nr_pages, |
edafccee JA |
7250 | sizeof(struct vm_area_struct *), |
7251 | GFP_KERNEL); | |
7252 | if (!pages || !vmas) { | |
7253 | ret = -ENOMEM; | |
7254 | if (ctx->account_mem) | |
7255 | io_unaccount_mem(ctx->user, nr_pages); | |
7256 | goto err; | |
7257 | } | |
7258 | got_pages = nr_pages; | |
7259 | } | |
7260 | ||
d4ef6475 | 7261 | imu->bvec = kvmalloc_array(nr_pages, sizeof(struct bio_vec), |
edafccee JA |
7262 | GFP_KERNEL); |
7263 | ret = -ENOMEM; | |
7264 | if (!imu->bvec) { | |
7265 | if (ctx->account_mem) | |
7266 | io_unaccount_mem(ctx->user, nr_pages); | |
7267 | goto err; | |
7268 | } | |
7269 | ||
7270 | ret = 0; | |
7271 | down_read(¤t->mm->mmap_sem); | |
2113b05d | 7272 | pret = pin_user_pages(ubuf, nr_pages, |
932f4a63 IW |
7273 | FOLL_WRITE | FOLL_LONGTERM, |
7274 | pages, vmas); | |
edafccee JA |
7275 | if (pret == nr_pages) { |
7276 | /* don't support file backed memory */ | |
7277 | for (j = 0; j < nr_pages; j++) { | |
7278 | struct vm_area_struct *vma = vmas[j]; | |
7279 | ||
7280 | if (vma->vm_file && | |
7281 | !is_file_hugepages(vma->vm_file)) { | |
7282 | ret = -EOPNOTSUPP; | |
7283 | break; | |
7284 | } | |
7285 | } | |
7286 | } else { | |
7287 | ret = pret < 0 ? pret : -EFAULT; | |
7288 | } | |
7289 | up_read(¤t->mm->mmap_sem); | |
7290 | if (ret) { | |
7291 | /* | |
7292 | * if we did partial map, or found file backed vmas, | |
7293 | * release any pages we did get | |
7294 | */ | |
27c4d3a3 | 7295 | if (pret > 0) |
f1f6a7dd | 7296 | unpin_user_pages(pages, pret); |
edafccee JA |
7297 | if (ctx->account_mem) |
7298 | io_unaccount_mem(ctx->user, nr_pages); | |
d4ef6475 | 7299 | kvfree(imu->bvec); |
edafccee JA |
7300 | goto err; |
7301 | } | |
7302 | ||
7303 | off = ubuf & ~PAGE_MASK; | |
7304 | size = iov.iov_len; | |
7305 | for (j = 0; j < nr_pages; j++) { | |
7306 | size_t vec_len; | |
7307 | ||
7308 | vec_len = min_t(size_t, size, PAGE_SIZE - off); | |
7309 | imu->bvec[j].bv_page = pages[j]; | |
7310 | imu->bvec[j].bv_len = vec_len; | |
7311 | imu->bvec[j].bv_offset = off; | |
7312 | off = 0; | |
7313 | size -= vec_len; | |
7314 | } | |
7315 | /* store original address for later verification */ | |
7316 | imu->ubuf = ubuf; | |
7317 | imu->len = iov.iov_len; | |
7318 | imu->nr_bvecs = nr_pages; | |
7319 | ||
7320 | ctx->nr_user_bufs++; | |
7321 | } | |
d4ef6475 MR |
7322 | kvfree(pages); |
7323 | kvfree(vmas); | |
edafccee JA |
7324 | return 0; |
7325 | err: | |
d4ef6475 MR |
7326 | kvfree(pages); |
7327 | kvfree(vmas); | |
edafccee JA |
7328 | io_sqe_buffer_unregister(ctx); |
7329 | return ret; | |
7330 | } | |
7331 | ||
9b402849 JA |
7332 | static int io_eventfd_register(struct io_ring_ctx *ctx, void __user *arg) |
7333 | { | |
7334 | __s32 __user *fds = arg; | |
7335 | int fd; | |
7336 | ||
7337 | if (ctx->cq_ev_fd) | |
7338 | return -EBUSY; | |
7339 | ||
7340 | if (copy_from_user(&fd, fds, sizeof(*fds))) | |
7341 | return -EFAULT; | |
7342 | ||
7343 | ctx->cq_ev_fd = eventfd_ctx_fdget(fd); | |
7344 | if (IS_ERR(ctx->cq_ev_fd)) { | |
7345 | int ret = PTR_ERR(ctx->cq_ev_fd); | |
7346 | ctx->cq_ev_fd = NULL; | |
7347 | return ret; | |
7348 | } | |
7349 | ||
7350 | return 0; | |
7351 | } | |
7352 | ||
7353 | static int io_eventfd_unregister(struct io_ring_ctx *ctx) | |
7354 | { | |
7355 | if (ctx->cq_ev_fd) { | |
7356 | eventfd_ctx_put(ctx->cq_ev_fd); | |
7357 | ctx->cq_ev_fd = NULL; | |
7358 | return 0; | |
7359 | } | |
7360 | ||
7361 | return -ENXIO; | |
7362 | } | |
7363 | ||
5a2e745d JA |
7364 | static int __io_destroy_buffers(int id, void *p, void *data) |
7365 | { | |
7366 | struct io_ring_ctx *ctx = data; | |
7367 | struct io_buffer *buf = p; | |
7368 | ||
067524e9 | 7369 | __io_remove_buffers(ctx, buf, id, -1U); |
5a2e745d JA |
7370 | return 0; |
7371 | } | |
7372 | ||
7373 | static void io_destroy_buffers(struct io_ring_ctx *ctx) | |
7374 | { | |
7375 | idr_for_each(&ctx->io_buffer_idr, __io_destroy_buffers, ctx); | |
7376 | idr_destroy(&ctx->io_buffer_idr); | |
7377 | } | |
7378 | ||
2b188cc1 JA |
7379 | static void io_ring_ctx_free(struct io_ring_ctx *ctx) |
7380 | { | |
6b06314c | 7381 | io_finish_async(ctx); |
2b188cc1 JA |
7382 | if (ctx->sqo_mm) |
7383 | mmdrop(ctx->sqo_mm); | |
def596e9 JA |
7384 | |
7385 | io_iopoll_reap_events(ctx); | |
edafccee | 7386 | io_sqe_buffer_unregister(ctx); |
6b06314c | 7387 | io_sqe_files_unregister(ctx); |
9b402849 | 7388 | io_eventfd_unregister(ctx); |
5a2e745d | 7389 | io_destroy_buffers(ctx); |
41726c9a | 7390 | idr_destroy(&ctx->personality_idr); |
def596e9 | 7391 | |
2b188cc1 | 7392 | #if defined(CONFIG_UNIX) |
355e8d26 EB |
7393 | if (ctx->ring_sock) { |
7394 | ctx->ring_sock->file = NULL; /* so that iput() is called */ | |
2b188cc1 | 7395 | sock_release(ctx->ring_sock); |
355e8d26 | 7396 | } |
2b188cc1 JA |
7397 | #endif |
7398 | ||
75b28aff | 7399 | io_mem_free(ctx->rings); |
2b188cc1 | 7400 | io_mem_free(ctx->sq_sqes); |
2b188cc1 JA |
7401 | |
7402 | percpu_ref_exit(&ctx->refs); | |
7403 | if (ctx->account_mem) | |
7404 | io_unaccount_mem(ctx->user, | |
7405 | ring_pages(ctx->sq_entries, ctx->cq_entries)); | |
7406 | free_uid(ctx->user); | |
181e448d | 7407 | put_cred(ctx->creds); |
78076bb6 | 7408 | kfree(ctx->cancel_hash); |
0ddf92e8 | 7409 | kmem_cache_free(req_cachep, ctx->fallback_req); |
2b188cc1 JA |
7410 | kfree(ctx); |
7411 | } | |
7412 | ||
7413 | static __poll_t io_uring_poll(struct file *file, poll_table *wait) | |
7414 | { | |
7415 | struct io_ring_ctx *ctx = file->private_data; | |
7416 | __poll_t mask = 0; | |
7417 | ||
7418 | poll_wait(file, &ctx->cq_wait, wait); | |
4f7067c3 SB |
7419 | /* |
7420 | * synchronizes with barrier from wq_has_sleeper call in | |
7421 | * io_commit_cqring | |
7422 | */ | |
2b188cc1 | 7423 | smp_rmb(); |
75b28aff HV |
7424 | if (READ_ONCE(ctx->rings->sq.tail) - ctx->cached_sq_head != |
7425 | ctx->rings->sq_ring_entries) | |
2b188cc1 | 7426 | mask |= EPOLLOUT | EPOLLWRNORM; |
63e5d81f | 7427 | if (io_cqring_events(ctx, false)) |
2b188cc1 JA |
7428 | mask |= EPOLLIN | EPOLLRDNORM; |
7429 | ||
7430 | return mask; | |
7431 | } | |
7432 | ||
7433 | static int io_uring_fasync(int fd, struct file *file, int on) | |
7434 | { | |
7435 | struct io_ring_ctx *ctx = file->private_data; | |
7436 | ||
7437 | return fasync_helper(fd, file, on, &ctx->cq_fasync); | |
7438 | } | |
7439 | ||
071698e1 JA |
7440 | static int io_remove_personalities(int id, void *p, void *data) |
7441 | { | |
7442 | struct io_ring_ctx *ctx = data; | |
7443 | const struct cred *cred; | |
7444 | ||
7445 | cred = idr_remove(&ctx->personality_idr, id); | |
7446 | if (cred) | |
7447 | put_cred(cred); | |
7448 | return 0; | |
7449 | } | |
7450 | ||
85faa7b8 JA |
7451 | static void io_ring_exit_work(struct work_struct *work) |
7452 | { | |
7453 | struct io_ring_ctx *ctx; | |
7454 | ||
7455 | ctx = container_of(work, struct io_ring_ctx, exit_work); | |
7456 | if (ctx->rings) | |
7457 | io_cqring_overflow_flush(ctx, true); | |
7458 | ||
0f158b4c | 7459 | wait_for_completion(&ctx->ref_comp); |
85faa7b8 JA |
7460 | io_ring_ctx_free(ctx); |
7461 | } | |
7462 | ||
2b188cc1 JA |
7463 | static void io_ring_ctx_wait_and_kill(struct io_ring_ctx *ctx) |
7464 | { | |
7465 | mutex_lock(&ctx->uring_lock); | |
7466 | percpu_ref_kill(&ctx->refs); | |
7467 | mutex_unlock(&ctx->uring_lock); | |
7468 | ||
5262f567 | 7469 | io_kill_timeouts(ctx); |
221c5eb2 | 7470 | io_poll_remove_all(ctx); |
561fb04a JA |
7471 | |
7472 | if (ctx->io_wq) | |
7473 | io_wq_cancel_all(ctx->io_wq); | |
7474 | ||
def596e9 | 7475 | io_iopoll_reap_events(ctx); |
15dff286 JA |
7476 | /* if we failed setting up the ctx, we might not have any rings */ |
7477 | if (ctx->rings) | |
7478 | io_cqring_overflow_flush(ctx, true); | |
071698e1 | 7479 | idr_for_each(&ctx->personality_idr, io_remove_personalities, ctx); |
85faa7b8 JA |
7480 | INIT_WORK(&ctx->exit_work, io_ring_exit_work); |
7481 | queue_work(system_wq, &ctx->exit_work); | |
2b188cc1 JA |
7482 | } |
7483 | ||
7484 | static int io_uring_release(struct inode *inode, struct file *file) | |
7485 | { | |
7486 | struct io_ring_ctx *ctx = file->private_data; | |
7487 | ||
7488 | file->private_data = NULL; | |
7489 | io_ring_ctx_wait_and_kill(ctx); | |
7490 | return 0; | |
7491 | } | |
7492 | ||
fcb323cc JA |
7493 | static void io_uring_cancel_files(struct io_ring_ctx *ctx, |
7494 | struct files_struct *files) | |
7495 | { | |
fcb323cc | 7496 | while (!list_empty_careful(&ctx->inflight_list)) { |
d8f1b971 XW |
7497 | struct io_kiocb *cancel_req = NULL, *req; |
7498 | DEFINE_WAIT(wait); | |
fcb323cc JA |
7499 | |
7500 | spin_lock_irq(&ctx->inflight_lock); | |
7501 | list_for_each_entry(req, &ctx->inflight_list, inflight_entry) { | |
768134d4 JA |
7502 | if (req->work.files != files) |
7503 | continue; | |
7504 | /* req is being completed, ignore */ | |
7505 | if (!refcount_inc_not_zero(&req->refs)) | |
7506 | continue; | |
7507 | cancel_req = req; | |
7508 | break; | |
fcb323cc | 7509 | } |
768134d4 | 7510 | if (cancel_req) |
fcb323cc | 7511 | prepare_to_wait(&ctx->inflight_wait, &wait, |
768134d4 | 7512 | TASK_UNINTERRUPTIBLE); |
fcb323cc JA |
7513 | spin_unlock_irq(&ctx->inflight_lock); |
7514 | ||
768134d4 JA |
7515 | /* We need to keep going until we don't find a matching req */ |
7516 | if (!cancel_req) | |
fcb323cc | 7517 | break; |
2f6d9b9d | 7518 | |
2ca10259 JA |
7519 | if (cancel_req->flags & REQ_F_OVERFLOW) { |
7520 | spin_lock_irq(&ctx->completion_lock); | |
7521 | list_del(&cancel_req->list); | |
7522 | cancel_req->flags &= ~REQ_F_OVERFLOW; | |
7523 | if (list_empty(&ctx->cq_overflow_list)) { | |
7524 | clear_bit(0, &ctx->sq_check_overflow); | |
7525 | clear_bit(0, &ctx->cq_check_overflow); | |
7526 | } | |
7527 | spin_unlock_irq(&ctx->completion_lock); | |
7528 | ||
7529 | WRITE_ONCE(ctx->rings->cq_overflow, | |
7530 | atomic_inc_return(&ctx->cached_cq_overflow)); | |
7531 | ||
7532 | /* | |
7533 | * Put inflight ref and overflow ref. If that's | |
7534 | * all we had, then we're done with this request. | |
7535 | */ | |
7536 | if (refcount_sub_and_test(2, &cancel_req->refs)) { | |
4518a3cc | 7537 | io_free_req(cancel_req); |
d8f1b971 | 7538 | finish_wait(&ctx->inflight_wait, &wait); |
2ca10259 JA |
7539 | continue; |
7540 | } | |
7541 | } | |
7542 | ||
2f6d9b9d BL |
7543 | io_wq_cancel_work(ctx->io_wq, &cancel_req->work); |
7544 | io_put_req(cancel_req); | |
fcb323cc | 7545 | schedule(); |
d8f1b971 | 7546 | finish_wait(&ctx->inflight_wait, &wait); |
fcb323cc JA |
7547 | } |
7548 | } | |
7549 | ||
7550 | static int io_uring_flush(struct file *file, void *data) | |
7551 | { | |
7552 | struct io_ring_ctx *ctx = file->private_data; | |
7553 | ||
7554 | io_uring_cancel_files(ctx, data); | |
6ab23144 JA |
7555 | |
7556 | /* | |
7557 | * If the task is going away, cancel work it may have pending | |
7558 | */ | |
7559 | if (fatal_signal_pending(current) || (current->flags & PF_EXITING)) | |
7560 | io_wq_cancel_pid(ctx->io_wq, task_pid_vnr(current)); | |
7561 | ||
fcb323cc JA |
7562 | return 0; |
7563 | } | |
7564 | ||
6c5c240e RP |
7565 | static void *io_uring_validate_mmap_request(struct file *file, |
7566 | loff_t pgoff, size_t sz) | |
2b188cc1 | 7567 | { |
2b188cc1 | 7568 | struct io_ring_ctx *ctx = file->private_data; |
6c5c240e | 7569 | loff_t offset = pgoff << PAGE_SHIFT; |
2b188cc1 JA |
7570 | struct page *page; |
7571 | void *ptr; | |
7572 | ||
7573 | switch (offset) { | |
7574 | case IORING_OFF_SQ_RING: | |
75b28aff HV |
7575 | case IORING_OFF_CQ_RING: |
7576 | ptr = ctx->rings; | |
2b188cc1 JA |
7577 | break; |
7578 | case IORING_OFF_SQES: | |
7579 | ptr = ctx->sq_sqes; | |
7580 | break; | |
2b188cc1 | 7581 | default: |
6c5c240e | 7582 | return ERR_PTR(-EINVAL); |
2b188cc1 JA |
7583 | } |
7584 | ||
7585 | page = virt_to_head_page(ptr); | |
a50b854e | 7586 | if (sz > page_size(page)) |
6c5c240e RP |
7587 | return ERR_PTR(-EINVAL); |
7588 | ||
7589 | return ptr; | |
7590 | } | |
7591 | ||
7592 | #ifdef CONFIG_MMU | |
7593 | ||
7594 | static int io_uring_mmap(struct file *file, struct vm_area_struct *vma) | |
7595 | { | |
7596 | size_t sz = vma->vm_end - vma->vm_start; | |
7597 | unsigned long pfn; | |
7598 | void *ptr; | |
7599 | ||
7600 | ptr = io_uring_validate_mmap_request(file, vma->vm_pgoff, sz); | |
7601 | if (IS_ERR(ptr)) | |
7602 | return PTR_ERR(ptr); | |
2b188cc1 JA |
7603 | |
7604 | pfn = virt_to_phys(ptr) >> PAGE_SHIFT; | |
7605 | return remap_pfn_range(vma, vma->vm_start, pfn, sz, vma->vm_page_prot); | |
7606 | } | |
7607 | ||
6c5c240e RP |
7608 | #else /* !CONFIG_MMU */ |
7609 | ||
7610 | static int io_uring_mmap(struct file *file, struct vm_area_struct *vma) | |
7611 | { | |
7612 | return vma->vm_flags & (VM_SHARED | VM_MAYSHARE) ? 0 : -EINVAL; | |
7613 | } | |
7614 | ||
7615 | static unsigned int io_uring_nommu_mmap_capabilities(struct file *file) | |
7616 | { | |
7617 | return NOMMU_MAP_DIRECT | NOMMU_MAP_READ | NOMMU_MAP_WRITE; | |
7618 | } | |
7619 | ||
7620 | static unsigned long io_uring_nommu_get_unmapped_area(struct file *file, | |
7621 | unsigned long addr, unsigned long len, | |
7622 | unsigned long pgoff, unsigned long flags) | |
7623 | { | |
7624 | void *ptr; | |
7625 | ||
7626 | ptr = io_uring_validate_mmap_request(file, pgoff, len); | |
7627 | if (IS_ERR(ptr)) | |
7628 | return PTR_ERR(ptr); | |
7629 | ||
7630 | return (unsigned long) ptr; | |
7631 | } | |
7632 | ||
7633 | #endif /* !CONFIG_MMU */ | |
7634 | ||
2b188cc1 JA |
7635 | SYSCALL_DEFINE6(io_uring_enter, unsigned int, fd, u32, to_submit, |
7636 | u32, min_complete, u32, flags, const sigset_t __user *, sig, | |
7637 | size_t, sigsz) | |
7638 | { | |
7639 | struct io_ring_ctx *ctx; | |
7640 | long ret = -EBADF; | |
7641 | int submitted = 0; | |
7642 | struct fd f; | |
7643 | ||
b41e9852 JA |
7644 | if (current->task_works) |
7645 | task_work_run(); | |
7646 | ||
6c271ce2 | 7647 | if (flags & ~(IORING_ENTER_GETEVENTS | IORING_ENTER_SQ_WAKEUP)) |
2b188cc1 JA |
7648 | return -EINVAL; |
7649 | ||
7650 | f = fdget(fd); | |
7651 | if (!f.file) | |
7652 | return -EBADF; | |
7653 | ||
7654 | ret = -EOPNOTSUPP; | |
7655 | if (f.file->f_op != &io_uring_fops) | |
7656 | goto out_fput; | |
7657 | ||
7658 | ret = -ENXIO; | |
7659 | ctx = f.file->private_data; | |
7660 | if (!percpu_ref_tryget(&ctx->refs)) | |
7661 | goto out_fput; | |
7662 | ||
6c271ce2 JA |
7663 | /* |
7664 | * For SQ polling, the thread will do all submissions and completions. | |
7665 | * Just return the requested submit count, and wake the thread if | |
7666 | * we were asked to. | |
7667 | */ | |
b2a9eada | 7668 | ret = 0; |
6c271ce2 | 7669 | if (ctx->flags & IORING_SETUP_SQPOLL) { |
c1edbf5f JA |
7670 | if (!list_empty_careful(&ctx->cq_overflow_list)) |
7671 | io_cqring_overflow_flush(ctx, false); | |
6c271ce2 JA |
7672 | if (flags & IORING_ENTER_SQ_WAKEUP) |
7673 | wake_up(&ctx->sqo_wait); | |
7674 | submitted = to_submit; | |
b2a9eada | 7675 | } else if (to_submit) { |
2b188cc1 | 7676 | mutex_lock(&ctx->uring_lock); |
0cdaf760 | 7677 | submitted = io_submit_sqes(ctx, to_submit, f.file, fd); |
2b188cc1 | 7678 | mutex_unlock(&ctx->uring_lock); |
7c504e65 PB |
7679 | |
7680 | if (submitted != to_submit) | |
7681 | goto out; | |
2b188cc1 JA |
7682 | } |
7683 | if (flags & IORING_ENTER_GETEVENTS) { | |
def596e9 JA |
7684 | unsigned nr_events = 0; |
7685 | ||
2b188cc1 JA |
7686 | min_complete = min(min_complete, ctx->cq_entries); |
7687 | ||
32b2244a XW |
7688 | /* |
7689 | * When SETUP_IOPOLL and SETUP_SQPOLL are both enabled, user | |
7690 | * space applications don't need to do io completion events | |
7691 | * polling again, they can rely on io_sq_thread to do polling | |
7692 | * work, which can reduce cpu usage and uring_lock contention. | |
7693 | */ | |
7694 | if (ctx->flags & IORING_SETUP_IOPOLL && | |
7695 | !(ctx->flags & IORING_SETUP_SQPOLL)) { | |
def596e9 | 7696 | ret = io_iopoll_check(ctx, &nr_events, min_complete); |
def596e9 JA |
7697 | } else { |
7698 | ret = io_cqring_wait(ctx, min_complete, sig, sigsz); | |
7699 | } | |
2b188cc1 JA |
7700 | } |
7701 | ||
7c504e65 | 7702 | out: |
6805b32e | 7703 | percpu_ref_put(&ctx->refs); |
2b188cc1 JA |
7704 | out_fput: |
7705 | fdput(f); | |
7706 | return submitted ? submitted : ret; | |
7707 | } | |
7708 | ||
bebdb65e | 7709 | #ifdef CONFIG_PROC_FS |
87ce955b JA |
7710 | static int io_uring_show_cred(int id, void *p, void *data) |
7711 | { | |
7712 | const struct cred *cred = p; | |
7713 | struct seq_file *m = data; | |
7714 | struct user_namespace *uns = seq_user_ns(m); | |
7715 | struct group_info *gi; | |
7716 | kernel_cap_t cap; | |
7717 | unsigned __capi; | |
7718 | int g; | |
7719 | ||
7720 | seq_printf(m, "%5d\n", id); | |
7721 | seq_put_decimal_ull(m, "\tUid:\t", from_kuid_munged(uns, cred->uid)); | |
7722 | seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->euid)); | |
7723 | seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->suid)); | |
7724 | seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->fsuid)); | |
7725 | seq_put_decimal_ull(m, "\n\tGid:\t", from_kgid_munged(uns, cred->gid)); | |
7726 | seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->egid)); | |
7727 | seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->sgid)); | |
7728 | seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->fsgid)); | |
7729 | seq_puts(m, "\n\tGroups:\t"); | |
7730 | gi = cred->group_info; | |
7731 | for (g = 0; g < gi->ngroups; g++) { | |
7732 | seq_put_decimal_ull(m, g ? " " : "", | |
7733 | from_kgid_munged(uns, gi->gid[g])); | |
7734 | } | |
7735 | seq_puts(m, "\n\tCapEff:\t"); | |
7736 | cap = cred->cap_effective; | |
7737 | CAP_FOR_EACH_U32(__capi) | |
7738 | seq_put_hex_ll(m, NULL, cap.cap[CAP_LAST_U32 - __capi], 8); | |
7739 | seq_putc(m, '\n'); | |
7740 | return 0; | |
7741 | } | |
7742 | ||
7743 | static void __io_uring_show_fdinfo(struct io_ring_ctx *ctx, struct seq_file *m) | |
7744 | { | |
7745 | int i; | |
7746 | ||
7747 | mutex_lock(&ctx->uring_lock); | |
7748 | seq_printf(m, "UserFiles:\t%u\n", ctx->nr_user_files); | |
7749 | for (i = 0; i < ctx->nr_user_files; i++) { | |
7750 | struct fixed_file_table *table; | |
7751 | struct file *f; | |
7752 | ||
7753 | table = &ctx->file_data->table[i >> IORING_FILE_TABLE_SHIFT]; | |
7754 | f = table->files[i & IORING_FILE_TABLE_MASK]; | |
7755 | if (f) | |
7756 | seq_printf(m, "%5u: %s\n", i, file_dentry(f)->d_iname); | |
7757 | else | |
7758 | seq_printf(m, "%5u: <none>\n", i); | |
7759 | } | |
7760 | seq_printf(m, "UserBufs:\t%u\n", ctx->nr_user_bufs); | |
7761 | for (i = 0; i < ctx->nr_user_bufs; i++) { | |
7762 | struct io_mapped_ubuf *buf = &ctx->user_bufs[i]; | |
7763 | ||
7764 | seq_printf(m, "%5u: 0x%llx/%u\n", i, buf->ubuf, | |
7765 | (unsigned int) buf->len); | |
7766 | } | |
7767 | if (!idr_is_empty(&ctx->personality_idr)) { | |
7768 | seq_printf(m, "Personalities:\n"); | |
7769 | idr_for_each(&ctx->personality_idr, io_uring_show_cred, m); | |
7770 | } | |
d7718a9d JA |
7771 | seq_printf(m, "PollList:\n"); |
7772 | spin_lock_irq(&ctx->completion_lock); | |
7773 | for (i = 0; i < (1U << ctx->cancel_hash_bits); i++) { | |
7774 | struct hlist_head *list = &ctx->cancel_hash[i]; | |
7775 | struct io_kiocb *req; | |
7776 | ||
7777 | hlist_for_each_entry(req, list, hash_node) | |
7778 | seq_printf(m, " op=%d, task_works=%d\n", req->opcode, | |
7779 | req->task->task_works != NULL); | |
7780 | } | |
7781 | spin_unlock_irq(&ctx->completion_lock); | |
87ce955b JA |
7782 | mutex_unlock(&ctx->uring_lock); |
7783 | } | |
7784 | ||
7785 | static void io_uring_show_fdinfo(struct seq_file *m, struct file *f) | |
7786 | { | |
7787 | struct io_ring_ctx *ctx = f->private_data; | |
7788 | ||
7789 | if (percpu_ref_tryget(&ctx->refs)) { | |
7790 | __io_uring_show_fdinfo(ctx, m); | |
7791 | percpu_ref_put(&ctx->refs); | |
7792 | } | |
7793 | } | |
bebdb65e | 7794 | #endif |
87ce955b | 7795 | |
2b188cc1 JA |
7796 | static const struct file_operations io_uring_fops = { |
7797 | .release = io_uring_release, | |
fcb323cc | 7798 | .flush = io_uring_flush, |
2b188cc1 | 7799 | .mmap = io_uring_mmap, |
6c5c240e RP |
7800 | #ifndef CONFIG_MMU |
7801 | .get_unmapped_area = io_uring_nommu_get_unmapped_area, | |
7802 | .mmap_capabilities = io_uring_nommu_mmap_capabilities, | |
7803 | #endif | |
2b188cc1 JA |
7804 | .poll = io_uring_poll, |
7805 | .fasync = io_uring_fasync, | |
bebdb65e | 7806 | #ifdef CONFIG_PROC_FS |
87ce955b | 7807 | .show_fdinfo = io_uring_show_fdinfo, |
bebdb65e | 7808 | #endif |
2b188cc1 JA |
7809 | }; |
7810 | ||
7811 | static int io_allocate_scq_urings(struct io_ring_ctx *ctx, | |
7812 | struct io_uring_params *p) | |
7813 | { | |
75b28aff HV |
7814 | struct io_rings *rings; |
7815 | size_t size, sq_array_offset; | |
2b188cc1 | 7816 | |
75b28aff HV |
7817 | size = rings_size(p->sq_entries, p->cq_entries, &sq_array_offset); |
7818 | if (size == SIZE_MAX) | |
7819 | return -EOVERFLOW; | |
7820 | ||
7821 | rings = io_mem_alloc(size); | |
7822 | if (!rings) | |
2b188cc1 JA |
7823 | return -ENOMEM; |
7824 | ||
75b28aff HV |
7825 | ctx->rings = rings; |
7826 | ctx->sq_array = (u32 *)((char *)rings + sq_array_offset); | |
7827 | rings->sq_ring_mask = p->sq_entries - 1; | |
7828 | rings->cq_ring_mask = p->cq_entries - 1; | |
7829 | rings->sq_ring_entries = p->sq_entries; | |
7830 | rings->cq_ring_entries = p->cq_entries; | |
7831 | ctx->sq_mask = rings->sq_ring_mask; | |
7832 | ctx->cq_mask = rings->cq_ring_mask; | |
7833 | ctx->sq_entries = rings->sq_ring_entries; | |
7834 | ctx->cq_entries = rings->cq_ring_entries; | |
2b188cc1 JA |
7835 | |
7836 | size = array_size(sizeof(struct io_uring_sqe), p->sq_entries); | |
eb065d30 JA |
7837 | if (size == SIZE_MAX) { |
7838 | io_mem_free(ctx->rings); | |
7839 | ctx->rings = NULL; | |
2b188cc1 | 7840 | return -EOVERFLOW; |
eb065d30 | 7841 | } |
2b188cc1 JA |
7842 | |
7843 | ctx->sq_sqes = io_mem_alloc(size); | |
eb065d30 JA |
7844 | if (!ctx->sq_sqes) { |
7845 | io_mem_free(ctx->rings); | |
7846 | ctx->rings = NULL; | |
2b188cc1 | 7847 | return -ENOMEM; |
eb065d30 | 7848 | } |
2b188cc1 | 7849 | |
2b188cc1 JA |
7850 | return 0; |
7851 | } | |
7852 | ||
7853 | /* | |
7854 | * Allocate an anonymous fd, this is what constitutes the application | |
7855 | * visible backing of an io_uring instance. The application mmaps this | |
7856 | * fd to gain access to the SQ/CQ ring details. If UNIX sockets are enabled, | |
7857 | * we have to tie this fd to a socket for file garbage collection purposes. | |
7858 | */ | |
7859 | static int io_uring_get_fd(struct io_ring_ctx *ctx) | |
7860 | { | |
7861 | struct file *file; | |
7862 | int ret; | |
7863 | ||
7864 | #if defined(CONFIG_UNIX) | |
7865 | ret = sock_create_kern(&init_net, PF_UNIX, SOCK_RAW, IPPROTO_IP, | |
7866 | &ctx->ring_sock); | |
7867 | if (ret) | |
7868 | return ret; | |
7869 | #endif | |
7870 | ||
7871 | ret = get_unused_fd_flags(O_RDWR | O_CLOEXEC); | |
7872 | if (ret < 0) | |
7873 | goto err; | |
7874 | ||
7875 | file = anon_inode_getfile("[io_uring]", &io_uring_fops, ctx, | |
7876 | O_RDWR | O_CLOEXEC); | |
7877 | if (IS_ERR(file)) { | |
7878 | put_unused_fd(ret); | |
7879 | ret = PTR_ERR(file); | |
7880 | goto err; | |
7881 | } | |
7882 | ||
7883 | #if defined(CONFIG_UNIX) | |
7884 | ctx->ring_sock->file = file; | |
7885 | #endif | |
7886 | fd_install(ret, file); | |
7887 | return ret; | |
7888 | err: | |
7889 | #if defined(CONFIG_UNIX) | |
7890 | sock_release(ctx->ring_sock); | |
7891 | ctx->ring_sock = NULL; | |
7892 | #endif | |
7893 | return ret; | |
7894 | } | |
7895 | ||
7f13657d XW |
7896 | static int io_uring_create(unsigned entries, struct io_uring_params *p, |
7897 | struct io_uring_params __user *params) | |
2b188cc1 JA |
7898 | { |
7899 | struct user_struct *user = NULL; | |
7900 | struct io_ring_ctx *ctx; | |
7901 | bool account_mem; | |
7902 | int ret; | |
7903 | ||
8110c1a6 | 7904 | if (!entries) |
2b188cc1 | 7905 | return -EINVAL; |
8110c1a6 JA |
7906 | if (entries > IORING_MAX_ENTRIES) { |
7907 | if (!(p->flags & IORING_SETUP_CLAMP)) | |
7908 | return -EINVAL; | |
7909 | entries = IORING_MAX_ENTRIES; | |
7910 | } | |
2b188cc1 JA |
7911 | |
7912 | /* | |
7913 | * Use twice as many entries for the CQ ring. It's possible for the | |
7914 | * application to drive a higher depth than the size of the SQ ring, | |
7915 | * since the sqes are only used at submission time. This allows for | |
33a107f0 JA |
7916 | * some flexibility in overcommitting a bit. If the application has |
7917 | * set IORING_SETUP_CQSIZE, it will have passed in the desired number | |
7918 | * of CQ ring entries manually. | |
2b188cc1 JA |
7919 | */ |
7920 | p->sq_entries = roundup_pow_of_two(entries); | |
33a107f0 JA |
7921 | if (p->flags & IORING_SETUP_CQSIZE) { |
7922 | /* | |
7923 | * If IORING_SETUP_CQSIZE is set, we do the same roundup | |
7924 | * to a power-of-two, if it isn't already. We do NOT impose | |
7925 | * any cq vs sq ring sizing. | |
7926 | */ | |
8110c1a6 | 7927 | if (p->cq_entries < p->sq_entries) |
33a107f0 | 7928 | return -EINVAL; |
8110c1a6 JA |
7929 | if (p->cq_entries > IORING_MAX_CQ_ENTRIES) { |
7930 | if (!(p->flags & IORING_SETUP_CLAMP)) | |
7931 | return -EINVAL; | |
7932 | p->cq_entries = IORING_MAX_CQ_ENTRIES; | |
7933 | } | |
33a107f0 JA |
7934 | p->cq_entries = roundup_pow_of_two(p->cq_entries); |
7935 | } else { | |
7936 | p->cq_entries = 2 * p->sq_entries; | |
7937 | } | |
2b188cc1 JA |
7938 | |
7939 | user = get_uid(current_user()); | |
7940 | account_mem = !capable(CAP_IPC_LOCK); | |
7941 | ||
7942 | if (account_mem) { | |
7943 | ret = io_account_mem(user, | |
7944 | ring_pages(p->sq_entries, p->cq_entries)); | |
7945 | if (ret) { | |
7946 | free_uid(user); | |
7947 | return ret; | |
7948 | } | |
7949 | } | |
7950 | ||
7951 | ctx = io_ring_ctx_alloc(p); | |
7952 | if (!ctx) { | |
7953 | if (account_mem) | |
7954 | io_unaccount_mem(user, ring_pages(p->sq_entries, | |
7955 | p->cq_entries)); | |
7956 | free_uid(user); | |
7957 | return -ENOMEM; | |
7958 | } | |
7959 | ctx->compat = in_compat_syscall(); | |
7960 | ctx->account_mem = account_mem; | |
7961 | ctx->user = user; | |
0b8c0ec7 | 7962 | ctx->creds = get_current_cred(); |
2b188cc1 JA |
7963 | |
7964 | ret = io_allocate_scq_urings(ctx, p); | |
7965 | if (ret) | |
7966 | goto err; | |
7967 | ||
6c271ce2 | 7968 | ret = io_sq_offload_start(ctx, p); |
2b188cc1 JA |
7969 | if (ret) |
7970 | goto err; | |
7971 | ||
2b188cc1 | 7972 | memset(&p->sq_off, 0, sizeof(p->sq_off)); |
75b28aff HV |
7973 | p->sq_off.head = offsetof(struct io_rings, sq.head); |
7974 | p->sq_off.tail = offsetof(struct io_rings, sq.tail); | |
7975 | p->sq_off.ring_mask = offsetof(struct io_rings, sq_ring_mask); | |
7976 | p->sq_off.ring_entries = offsetof(struct io_rings, sq_ring_entries); | |
7977 | p->sq_off.flags = offsetof(struct io_rings, sq_flags); | |
7978 | p->sq_off.dropped = offsetof(struct io_rings, sq_dropped); | |
7979 | p->sq_off.array = (char *)ctx->sq_array - (char *)ctx->rings; | |
2b188cc1 JA |
7980 | |
7981 | memset(&p->cq_off, 0, sizeof(p->cq_off)); | |
75b28aff HV |
7982 | p->cq_off.head = offsetof(struct io_rings, cq.head); |
7983 | p->cq_off.tail = offsetof(struct io_rings, cq.tail); | |
7984 | p->cq_off.ring_mask = offsetof(struct io_rings, cq_ring_mask); | |
7985 | p->cq_off.ring_entries = offsetof(struct io_rings, cq_ring_entries); | |
7986 | p->cq_off.overflow = offsetof(struct io_rings, cq_overflow); | |
7987 | p->cq_off.cqes = offsetof(struct io_rings, cqes); | |
0d9b5b3a | 7988 | p->cq_off.flags = offsetof(struct io_rings, cq_flags); |
ac90f249 | 7989 | |
7f13657d XW |
7990 | p->features = IORING_FEAT_SINGLE_MMAP | IORING_FEAT_NODROP | |
7991 | IORING_FEAT_SUBMIT_STABLE | IORING_FEAT_RW_CUR_POS | | |
7992 | IORING_FEAT_CUR_PERSONALITY | IORING_FEAT_FAST_POLL; | |
7993 | ||
7994 | if (copy_to_user(params, p, sizeof(*p))) { | |
7995 | ret = -EFAULT; | |
7996 | goto err; | |
7997 | } | |
044c1ab3 JA |
7998 | /* |
7999 | * Install ring fd as the very last thing, so we don't risk someone | |
8000 | * having closed it before we finish setup | |
8001 | */ | |
8002 | ret = io_uring_get_fd(ctx); | |
8003 | if (ret < 0) | |
8004 | goto err; | |
8005 | ||
c826bd7a | 8006 | trace_io_uring_create(ret, ctx, p->sq_entries, p->cq_entries, p->flags); |
2b188cc1 JA |
8007 | return ret; |
8008 | err: | |
8009 | io_ring_ctx_wait_and_kill(ctx); | |
8010 | return ret; | |
8011 | } | |
8012 | ||
8013 | /* | |
8014 | * Sets up an aio uring context, and returns the fd. Applications asks for a | |
8015 | * ring size, we return the actual sq/cq ring sizes (among other things) in the | |
8016 | * params structure passed in. | |
8017 | */ | |
8018 | static long io_uring_setup(u32 entries, struct io_uring_params __user *params) | |
8019 | { | |
8020 | struct io_uring_params p; | |
2b188cc1 JA |
8021 | int i; |
8022 | ||
8023 | if (copy_from_user(&p, params, sizeof(p))) | |
8024 | return -EFAULT; | |
8025 | for (i = 0; i < ARRAY_SIZE(p.resv); i++) { | |
8026 | if (p.resv[i]) | |
8027 | return -EINVAL; | |
8028 | } | |
8029 | ||
6c271ce2 | 8030 | if (p.flags & ~(IORING_SETUP_IOPOLL | IORING_SETUP_SQPOLL | |
8110c1a6 | 8031 | IORING_SETUP_SQ_AFF | IORING_SETUP_CQSIZE | |
24369c2e | 8032 | IORING_SETUP_CLAMP | IORING_SETUP_ATTACH_WQ)) |
2b188cc1 JA |
8033 | return -EINVAL; |
8034 | ||
7f13657d | 8035 | return io_uring_create(entries, &p, params); |
2b188cc1 JA |
8036 | } |
8037 | ||
8038 | SYSCALL_DEFINE2(io_uring_setup, u32, entries, | |
8039 | struct io_uring_params __user *, params) | |
8040 | { | |
8041 | return io_uring_setup(entries, params); | |
8042 | } | |
8043 | ||
66f4af93 JA |
8044 | static int io_probe(struct io_ring_ctx *ctx, void __user *arg, unsigned nr_args) |
8045 | { | |
8046 | struct io_uring_probe *p; | |
8047 | size_t size; | |
8048 | int i, ret; | |
8049 | ||
8050 | size = struct_size(p, ops, nr_args); | |
8051 | if (size == SIZE_MAX) | |
8052 | return -EOVERFLOW; | |
8053 | p = kzalloc(size, GFP_KERNEL); | |
8054 | if (!p) | |
8055 | return -ENOMEM; | |
8056 | ||
8057 | ret = -EFAULT; | |
8058 | if (copy_from_user(p, arg, size)) | |
8059 | goto out; | |
8060 | ret = -EINVAL; | |
8061 | if (memchr_inv(p, 0, size)) | |
8062 | goto out; | |
8063 | ||
8064 | p->last_op = IORING_OP_LAST - 1; | |
8065 | if (nr_args > IORING_OP_LAST) | |
8066 | nr_args = IORING_OP_LAST; | |
8067 | ||
8068 | for (i = 0; i < nr_args; i++) { | |
8069 | p->ops[i].op = i; | |
8070 | if (!io_op_defs[i].not_supported) | |
8071 | p->ops[i].flags = IO_URING_OP_SUPPORTED; | |
8072 | } | |
8073 | p->ops_len = i; | |
8074 | ||
8075 | ret = 0; | |
8076 | if (copy_to_user(arg, p, size)) | |
8077 | ret = -EFAULT; | |
8078 | out: | |
8079 | kfree(p); | |
8080 | return ret; | |
8081 | } | |
8082 | ||
071698e1 JA |
8083 | static int io_register_personality(struct io_ring_ctx *ctx) |
8084 | { | |
8085 | const struct cred *creds = get_current_cred(); | |
8086 | int id; | |
8087 | ||
8088 | id = idr_alloc_cyclic(&ctx->personality_idr, (void *) creds, 1, | |
8089 | USHRT_MAX, GFP_KERNEL); | |
8090 | if (id < 0) | |
8091 | put_cred(creds); | |
8092 | return id; | |
8093 | } | |
8094 | ||
8095 | static int io_unregister_personality(struct io_ring_ctx *ctx, unsigned id) | |
8096 | { | |
8097 | const struct cred *old_creds; | |
8098 | ||
8099 | old_creds = idr_remove(&ctx->personality_idr, id); | |
8100 | if (old_creds) { | |
8101 | put_cred(old_creds); | |
8102 | return 0; | |
8103 | } | |
8104 | ||
8105 | return -EINVAL; | |
8106 | } | |
8107 | ||
8108 | static bool io_register_op_must_quiesce(int op) | |
8109 | { | |
8110 | switch (op) { | |
8111 | case IORING_UNREGISTER_FILES: | |
8112 | case IORING_REGISTER_FILES_UPDATE: | |
8113 | case IORING_REGISTER_PROBE: | |
8114 | case IORING_REGISTER_PERSONALITY: | |
8115 | case IORING_UNREGISTER_PERSONALITY: | |
8116 | return false; | |
8117 | default: | |
8118 | return true; | |
8119 | } | |
8120 | } | |
8121 | ||
edafccee JA |
8122 | static int __io_uring_register(struct io_ring_ctx *ctx, unsigned opcode, |
8123 | void __user *arg, unsigned nr_args) | |
b19062a5 JA |
8124 | __releases(ctx->uring_lock) |
8125 | __acquires(ctx->uring_lock) | |
edafccee JA |
8126 | { |
8127 | int ret; | |
8128 | ||
35fa71a0 JA |
8129 | /* |
8130 | * We're inside the ring mutex, if the ref is already dying, then | |
8131 | * someone else killed the ctx or is already going through | |
8132 | * io_uring_register(). | |
8133 | */ | |
8134 | if (percpu_ref_is_dying(&ctx->refs)) | |
8135 | return -ENXIO; | |
8136 | ||
071698e1 | 8137 | if (io_register_op_must_quiesce(opcode)) { |
05f3fb3c | 8138 | percpu_ref_kill(&ctx->refs); |
b19062a5 | 8139 | |
05f3fb3c JA |
8140 | /* |
8141 | * Drop uring mutex before waiting for references to exit. If | |
8142 | * another thread is currently inside io_uring_enter() it might | |
8143 | * need to grab the uring_lock to make progress. If we hold it | |
8144 | * here across the drain wait, then we can deadlock. It's safe | |
8145 | * to drop the mutex here, since no new references will come in | |
8146 | * after we've killed the percpu ref. | |
8147 | */ | |
8148 | mutex_unlock(&ctx->uring_lock); | |
0f158b4c | 8149 | ret = wait_for_completion_interruptible(&ctx->ref_comp); |
05f3fb3c | 8150 | mutex_lock(&ctx->uring_lock); |
c150368b JA |
8151 | if (ret) { |
8152 | percpu_ref_resurrect(&ctx->refs); | |
8153 | ret = -EINTR; | |
8154 | goto out; | |
8155 | } | |
05f3fb3c | 8156 | } |
edafccee JA |
8157 | |
8158 | switch (opcode) { | |
8159 | case IORING_REGISTER_BUFFERS: | |
8160 | ret = io_sqe_buffer_register(ctx, arg, nr_args); | |
8161 | break; | |
8162 | case IORING_UNREGISTER_BUFFERS: | |
8163 | ret = -EINVAL; | |
8164 | if (arg || nr_args) | |
8165 | break; | |
8166 | ret = io_sqe_buffer_unregister(ctx); | |
8167 | break; | |
6b06314c JA |
8168 | case IORING_REGISTER_FILES: |
8169 | ret = io_sqe_files_register(ctx, arg, nr_args); | |
8170 | break; | |
8171 | case IORING_UNREGISTER_FILES: | |
8172 | ret = -EINVAL; | |
8173 | if (arg || nr_args) | |
8174 | break; | |
8175 | ret = io_sqe_files_unregister(ctx); | |
8176 | break; | |
c3a31e60 JA |
8177 | case IORING_REGISTER_FILES_UPDATE: |
8178 | ret = io_sqe_files_update(ctx, arg, nr_args); | |
8179 | break; | |
9b402849 | 8180 | case IORING_REGISTER_EVENTFD: |
f2842ab5 | 8181 | case IORING_REGISTER_EVENTFD_ASYNC: |
9b402849 JA |
8182 | ret = -EINVAL; |
8183 | if (nr_args != 1) | |
8184 | break; | |
8185 | ret = io_eventfd_register(ctx, arg); | |
f2842ab5 JA |
8186 | if (ret) |
8187 | break; | |
8188 | if (opcode == IORING_REGISTER_EVENTFD_ASYNC) | |
8189 | ctx->eventfd_async = 1; | |
8190 | else | |
8191 | ctx->eventfd_async = 0; | |
9b402849 JA |
8192 | break; |
8193 | case IORING_UNREGISTER_EVENTFD: | |
8194 | ret = -EINVAL; | |
8195 | if (arg || nr_args) | |
8196 | break; | |
8197 | ret = io_eventfd_unregister(ctx); | |
8198 | break; | |
66f4af93 JA |
8199 | case IORING_REGISTER_PROBE: |
8200 | ret = -EINVAL; | |
8201 | if (!arg || nr_args > 256) | |
8202 | break; | |
8203 | ret = io_probe(ctx, arg, nr_args); | |
8204 | break; | |
071698e1 JA |
8205 | case IORING_REGISTER_PERSONALITY: |
8206 | ret = -EINVAL; | |
8207 | if (arg || nr_args) | |
8208 | break; | |
8209 | ret = io_register_personality(ctx); | |
8210 | break; | |
8211 | case IORING_UNREGISTER_PERSONALITY: | |
8212 | ret = -EINVAL; | |
8213 | if (arg) | |
8214 | break; | |
8215 | ret = io_unregister_personality(ctx, nr_args); | |
8216 | break; | |
edafccee JA |
8217 | default: |
8218 | ret = -EINVAL; | |
8219 | break; | |
8220 | } | |
8221 | ||
071698e1 | 8222 | if (io_register_op_must_quiesce(opcode)) { |
05f3fb3c | 8223 | /* bring the ctx back to life */ |
05f3fb3c | 8224 | percpu_ref_reinit(&ctx->refs); |
c150368b | 8225 | out: |
0f158b4c | 8226 | reinit_completion(&ctx->ref_comp); |
05f3fb3c | 8227 | } |
edafccee JA |
8228 | return ret; |
8229 | } | |
8230 | ||
8231 | SYSCALL_DEFINE4(io_uring_register, unsigned int, fd, unsigned int, opcode, | |
8232 | void __user *, arg, unsigned int, nr_args) | |
8233 | { | |
8234 | struct io_ring_ctx *ctx; | |
8235 | long ret = -EBADF; | |
8236 | struct fd f; | |
8237 | ||
8238 | f = fdget(fd); | |
8239 | if (!f.file) | |
8240 | return -EBADF; | |
8241 | ||
8242 | ret = -EOPNOTSUPP; | |
8243 | if (f.file->f_op != &io_uring_fops) | |
8244 | goto out_fput; | |
8245 | ||
8246 | ctx = f.file->private_data; | |
8247 | ||
8248 | mutex_lock(&ctx->uring_lock); | |
8249 | ret = __io_uring_register(ctx, opcode, arg, nr_args); | |
8250 | mutex_unlock(&ctx->uring_lock); | |
c826bd7a DD |
8251 | trace_io_uring_register(ctx, opcode, ctx->nr_user_files, ctx->nr_user_bufs, |
8252 | ctx->cq_ev_fd != NULL, ret); | |
edafccee JA |
8253 | out_fput: |
8254 | fdput(f); | |
8255 | return ret; | |
8256 | } | |
8257 | ||
2b188cc1 JA |
8258 | static int __init io_uring_init(void) |
8259 | { | |
d7f62e82 SM |
8260 | #define __BUILD_BUG_VERIFY_ELEMENT(stype, eoffset, etype, ename) do { \ |
8261 | BUILD_BUG_ON(offsetof(stype, ename) != eoffset); \ | |
8262 | BUILD_BUG_ON(sizeof(etype) != sizeof_field(stype, ename)); \ | |
8263 | } while (0) | |
8264 | ||
8265 | #define BUILD_BUG_SQE_ELEM(eoffset, etype, ename) \ | |
8266 | __BUILD_BUG_VERIFY_ELEMENT(struct io_uring_sqe, eoffset, etype, ename) | |
8267 | BUILD_BUG_ON(sizeof(struct io_uring_sqe) != 64); | |
8268 | BUILD_BUG_SQE_ELEM(0, __u8, opcode); | |
8269 | BUILD_BUG_SQE_ELEM(1, __u8, flags); | |
8270 | BUILD_BUG_SQE_ELEM(2, __u16, ioprio); | |
8271 | BUILD_BUG_SQE_ELEM(4, __s32, fd); | |
8272 | BUILD_BUG_SQE_ELEM(8, __u64, off); | |
8273 | BUILD_BUG_SQE_ELEM(8, __u64, addr2); | |
8274 | BUILD_BUG_SQE_ELEM(16, __u64, addr); | |
7d67af2c | 8275 | BUILD_BUG_SQE_ELEM(16, __u64, splice_off_in); |
d7f62e82 SM |
8276 | BUILD_BUG_SQE_ELEM(24, __u32, len); |
8277 | BUILD_BUG_SQE_ELEM(28, __kernel_rwf_t, rw_flags); | |
8278 | BUILD_BUG_SQE_ELEM(28, /* compat */ int, rw_flags); | |
8279 | BUILD_BUG_SQE_ELEM(28, /* compat */ __u32, rw_flags); | |
8280 | BUILD_BUG_SQE_ELEM(28, __u32, fsync_flags); | |
8281 | BUILD_BUG_SQE_ELEM(28, __u16, poll_events); | |
8282 | BUILD_BUG_SQE_ELEM(28, __u32, sync_range_flags); | |
8283 | BUILD_BUG_SQE_ELEM(28, __u32, msg_flags); | |
8284 | BUILD_BUG_SQE_ELEM(28, __u32, timeout_flags); | |
8285 | BUILD_BUG_SQE_ELEM(28, __u32, accept_flags); | |
8286 | BUILD_BUG_SQE_ELEM(28, __u32, cancel_flags); | |
8287 | BUILD_BUG_SQE_ELEM(28, __u32, open_flags); | |
8288 | BUILD_BUG_SQE_ELEM(28, __u32, statx_flags); | |
8289 | BUILD_BUG_SQE_ELEM(28, __u32, fadvise_advice); | |
7d67af2c | 8290 | BUILD_BUG_SQE_ELEM(28, __u32, splice_flags); |
d7f62e82 SM |
8291 | BUILD_BUG_SQE_ELEM(32, __u64, user_data); |
8292 | BUILD_BUG_SQE_ELEM(40, __u16, buf_index); | |
8293 | BUILD_BUG_SQE_ELEM(42, __u16, personality); | |
7d67af2c | 8294 | BUILD_BUG_SQE_ELEM(44, __s32, splice_fd_in); |
d7f62e82 | 8295 | |
d3656344 | 8296 | BUILD_BUG_ON(ARRAY_SIZE(io_op_defs) != IORING_OP_LAST); |
84557871 | 8297 | BUILD_BUG_ON(__REQ_F_LAST_BIT >= 8 * sizeof(int)); |
2b188cc1 JA |
8298 | req_cachep = KMEM_CACHE(io_kiocb, SLAB_HWCACHE_ALIGN | SLAB_PANIC); |
8299 | return 0; | |
8300 | }; | |
8301 | __initcall(io_uring_init); |