]>
Commit | Line | Data |
---|---|---|
2b188cc1 JA |
1 | // SPDX-License-Identifier: GPL-2.0 |
2 | /* | |
3 | * Shared application/kernel submission and completion ring pairs, for | |
4 | * supporting fast/efficient IO. | |
5 | * | |
6 | * A note on the read/write ordering memory barriers that are matched between | |
1e84b97b SB |
7 | * the application and kernel side. |
8 | * | |
9 | * After the application reads the CQ ring tail, it must use an | |
10 | * appropriate smp_rmb() to pair with the smp_wmb() the kernel uses | |
11 | * before writing the tail (using smp_load_acquire to read the tail will | |
12 | * do). It also needs a smp_mb() before updating CQ head (ordering the | |
13 | * entry load(s) with the head store), pairing with an implicit barrier | |
14 | * through a control-dependency in io_get_cqring (smp_store_release to | |
15 | * store head will do). Failure to do so could lead to reading invalid | |
16 | * CQ entries. | |
17 | * | |
18 | * Likewise, the application must use an appropriate smp_wmb() before | |
19 | * writing the SQ tail (ordering SQ entry stores with the tail store), | |
20 | * which pairs with smp_load_acquire in io_get_sqring (smp_store_release | |
21 | * to store the tail will do). And it needs a barrier ordering the SQ | |
22 | * head load before writing new SQ entries (smp_load_acquire to read | |
23 | * head will do). | |
24 | * | |
25 | * When using the SQ poll thread (IORING_SETUP_SQPOLL), the application | |
26 | * needs to check the SQ flags for IORING_SQ_NEED_WAKEUP *after* | |
27 | * updating the SQ tail; a full memory barrier smp_mb() is needed | |
28 | * between. | |
2b188cc1 JA |
29 | * |
30 | * Also see the examples in the liburing library: | |
31 | * | |
32 | * git://git.kernel.dk/liburing | |
33 | * | |
34 | * io_uring also uses READ/WRITE_ONCE() for _any_ store or load that happens | |
35 | * from data shared between the kernel and application. This is done both | |
36 | * for ordering purposes, but also to ensure that once a value is loaded from | |
37 | * data that the application could potentially modify, it remains stable. | |
38 | * | |
39 | * Copyright (C) 2018-2019 Jens Axboe | |
c992fe29 | 40 | * Copyright (c) 2018-2019 Christoph Hellwig |
2b188cc1 JA |
41 | */ |
42 | #include <linux/kernel.h> | |
43 | #include <linux/init.h> | |
44 | #include <linux/errno.h> | |
45 | #include <linux/syscalls.h> | |
46 | #include <linux/compat.h> | |
52de1fe1 | 47 | #include <net/compat.h> |
2b188cc1 JA |
48 | #include <linux/refcount.h> |
49 | #include <linux/uio.h> | |
6b47ee6e | 50 | #include <linux/bits.h> |
2b188cc1 JA |
51 | |
52 | #include <linux/sched/signal.h> | |
53 | #include <linux/fs.h> | |
54 | #include <linux/file.h> | |
55 | #include <linux/fdtable.h> | |
56 | #include <linux/mm.h> | |
57 | #include <linux/mman.h> | |
2b188cc1 JA |
58 | #include <linux/percpu.h> |
59 | #include <linux/slab.h> | |
6c271ce2 | 60 | #include <linux/kthread.h> |
2b188cc1 | 61 | #include <linux/blkdev.h> |
edafccee | 62 | #include <linux/bvec.h> |
2b188cc1 JA |
63 | #include <linux/net.h> |
64 | #include <net/sock.h> | |
65 | #include <net/af_unix.h> | |
6b06314c | 66 | #include <net/scm.h> |
2b188cc1 JA |
67 | #include <linux/anon_inodes.h> |
68 | #include <linux/sched/mm.h> | |
69 | #include <linux/uaccess.h> | |
70 | #include <linux/nospec.h> | |
edafccee JA |
71 | #include <linux/sizes.h> |
72 | #include <linux/hugetlb.h> | |
aa4c3967 | 73 | #include <linux/highmem.h> |
15b71abe JA |
74 | #include <linux/namei.h> |
75 | #include <linux/fsnotify.h> | |
4840e418 | 76 | #include <linux/fadvise.h> |
3e4827b0 | 77 | #include <linux/eventpoll.h> |
ff002b30 | 78 | #include <linux/fs_struct.h> |
7d67af2c | 79 | #include <linux/splice.h> |
b41e9852 | 80 | #include <linux/task_work.h> |
2b188cc1 | 81 | |
c826bd7a DD |
82 | #define CREATE_TRACE_POINTS |
83 | #include <trace/events/io_uring.h> | |
84 | ||
2b188cc1 JA |
85 | #include <uapi/linux/io_uring.h> |
86 | ||
87 | #include "internal.h" | |
561fb04a | 88 | #include "io-wq.h" |
2b188cc1 | 89 | |
5277deaa | 90 | #define IORING_MAX_ENTRIES 32768 |
33a107f0 | 91 | #define IORING_MAX_CQ_ENTRIES (2 * IORING_MAX_ENTRIES) |
65e19f54 JA |
92 | |
93 | /* | |
94 | * Shift of 9 is 512 entries, or exactly one page on 64-bit archs | |
95 | */ | |
96 | #define IORING_FILE_TABLE_SHIFT 9 | |
97 | #define IORING_MAX_FILES_TABLE (1U << IORING_FILE_TABLE_SHIFT) | |
98 | #define IORING_FILE_TABLE_MASK (IORING_MAX_FILES_TABLE - 1) | |
99 | #define IORING_MAX_FIXED_FILES (64 * IORING_MAX_FILES_TABLE) | |
2b188cc1 JA |
100 | |
101 | struct io_uring { | |
102 | u32 head ____cacheline_aligned_in_smp; | |
103 | u32 tail ____cacheline_aligned_in_smp; | |
104 | }; | |
105 | ||
1e84b97b | 106 | /* |
75b28aff HV |
107 | * This data is shared with the application through the mmap at offsets |
108 | * IORING_OFF_SQ_RING and IORING_OFF_CQ_RING. | |
1e84b97b SB |
109 | * |
110 | * The offsets to the member fields are published through struct | |
111 | * io_sqring_offsets when calling io_uring_setup. | |
112 | */ | |
75b28aff | 113 | struct io_rings { |
1e84b97b SB |
114 | /* |
115 | * Head and tail offsets into the ring; the offsets need to be | |
116 | * masked to get valid indices. | |
117 | * | |
75b28aff HV |
118 | * The kernel controls head of the sq ring and the tail of the cq ring, |
119 | * and the application controls tail of the sq ring and the head of the | |
120 | * cq ring. | |
1e84b97b | 121 | */ |
75b28aff | 122 | struct io_uring sq, cq; |
1e84b97b | 123 | /* |
75b28aff | 124 | * Bitmasks to apply to head and tail offsets (constant, equals |
1e84b97b SB |
125 | * ring_entries - 1) |
126 | */ | |
75b28aff HV |
127 | u32 sq_ring_mask, cq_ring_mask; |
128 | /* Ring sizes (constant, power of 2) */ | |
129 | u32 sq_ring_entries, cq_ring_entries; | |
1e84b97b SB |
130 | /* |
131 | * Number of invalid entries dropped by the kernel due to | |
132 | * invalid index stored in array | |
133 | * | |
134 | * Written by the kernel, shouldn't be modified by the | |
135 | * application (i.e. get number of "new events" by comparing to | |
136 | * cached value). | |
137 | * | |
138 | * After a new SQ head value was read by the application this | |
139 | * counter includes all submissions that were dropped reaching | |
140 | * the new SQ head (and possibly more). | |
141 | */ | |
75b28aff | 142 | u32 sq_dropped; |
1e84b97b | 143 | /* |
0d9b5b3a | 144 | * Runtime SQ flags |
1e84b97b SB |
145 | * |
146 | * Written by the kernel, shouldn't be modified by the | |
147 | * application. | |
148 | * | |
149 | * The application needs a full memory barrier before checking | |
150 | * for IORING_SQ_NEED_WAKEUP after updating the sq tail. | |
151 | */ | |
75b28aff | 152 | u32 sq_flags; |
0d9b5b3a SG |
153 | /* |
154 | * Runtime CQ flags | |
155 | * | |
156 | * Written by the application, shouldn't be modified by the | |
157 | * kernel. | |
158 | */ | |
159 | u32 cq_flags; | |
1e84b97b SB |
160 | /* |
161 | * Number of completion events lost because the queue was full; | |
162 | * this should be avoided by the application by making sure | |
0b4295b5 | 163 | * there are not more requests pending than there is space in |
1e84b97b SB |
164 | * the completion queue. |
165 | * | |
166 | * Written by the kernel, shouldn't be modified by the | |
167 | * application (i.e. get number of "new events" by comparing to | |
168 | * cached value). | |
169 | * | |
170 | * As completion events come in out of order this counter is not | |
171 | * ordered with any other data. | |
172 | */ | |
75b28aff | 173 | u32 cq_overflow; |
1e84b97b SB |
174 | /* |
175 | * Ring buffer of completion events. | |
176 | * | |
177 | * The kernel writes completion events fresh every time they are | |
178 | * produced, so the application is allowed to modify pending | |
179 | * entries. | |
180 | */ | |
75b28aff | 181 | struct io_uring_cqe cqes[] ____cacheline_aligned_in_smp; |
2b188cc1 JA |
182 | }; |
183 | ||
edafccee JA |
184 | struct io_mapped_ubuf { |
185 | u64 ubuf; | |
186 | size_t len; | |
187 | struct bio_vec *bvec; | |
188 | unsigned int nr_bvecs; | |
189 | }; | |
190 | ||
65e19f54 JA |
191 | struct fixed_file_table { |
192 | struct file **files; | |
31b51510 JA |
193 | }; |
194 | ||
05589553 XW |
195 | struct fixed_file_ref_node { |
196 | struct percpu_ref refs; | |
197 | struct list_head node; | |
198 | struct list_head file_list; | |
199 | struct fixed_file_data *file_data; | |
4a38aed2 | 200 | struct llist_node llist; |
05589553 XW |
201 | }; |
202 | ||
05f3fb3c JA |
203 | struct fixed_file_data { |
204 | struct fixed_file_table *table; | |
205 | struct io_ring_ctx *ctx; | |
206 | ||
05589553 | 207 | struct percpu_ref *cur_refs; |
05f3fb3c | 208 | struct percpu_ref refs; |
05f3fb3c | 209 | struct completion done; |
05589553 XW |
210 | struct list_head ref_list; |
211 | spinlock_t lock; | |
05f3fb3c JA |
212 | }; |
213 | ||
5a2e745d JA |
214 | struct io_buffer { |
215 | struct list_head list; | |
216 | __u64 addr; | |
217 | __s32 len; | |
218 | __u16 bid; | |
219 | }; | |
220 | ||
2b188cc1 JA |
221 | struct io_ring_ctx { |
222 | struct { | |
223 | struct percpu_ref refs; | |
224 | } ____cacheline_aligned_in_smp; | |
225 | ||
226 | struct { | |
227 | unsigned int flags; | |
e1d85334 RD |
228 | unsigned int compat: 1; |
229 | unsigned int account_mem: 1; | |
230 | unsigned int cq_overflow_flushed: 1; | |
231 | unsigned int drain_next: 1; | |
232 | unsigned int eventfd_async: 1; | |
2b188cc1 | 233 | |
75b28aff HV |
234 | /* |
235 | * Ring buffer of indices into array of io_uring_sqe, which is | |
236 | * mmapped by the application using the IORING_OFF_SQES offset. | |
237 | * | |
238 | * This indirection could e.g. be used to assign fixed | |
239 | * io_uring_sqe entries to operations and only submit them to | |
240 | * the queue when needed. | |
241 | * | |
242 | * The kernel modifies neither the indices array nor the entries | |
243 | * array. | |
244 | */ | |
245 | u32 *sq_array; | |
2b188cc1 JA |
246 | unsigned cached_sq_head; |
247 | unsigned sq_entries; | |
248 | unsigned sq_mask; | |
6c271ce2 | 249 | unsigned sq_thread_idle; |
498ccd9e | 250 | unsigned cached_sq_dropped; |
206aefde | 251 | atomic_t cached_cq_overflow; |
ad3eb2c8 | 252 | unsigned long sq_check_overflow; |
de0617e4 JA |
253 | |
254 | struct list_head defer_list; | |
5262f567 | 255 | struct list_head timeout_list; |
1d7bb1d5 | 256 | struct list_head cq_overflow_list; |
fcb323cc JA |
257 | |
258 | wait_queue_head_t inflight_wait; | |
ad3eb2c8 | 259 | struct io_uring_sqe *sq_sqes; |
2b188cc1 JA |
260 | } ____cacheline_aligned_in_smp; |
261 | ||
206aefde JA |
262 | struct io_rings *rings; |
263 | ||
2b188cc1 | 264 | /* IO offload */ |
561fb04a | 265 | struct io_wq *io_wq; |
6c271ce2 | 266 | struct task_struct *sqo_thread; /* if using sq thread polling */ |
2b188cc1 | 267 | struct mm_struct *sqo_mm; |
6c271ce2 | 268 | wait_queue_head_t sqo_wait; |
75b28aff | 269 | |
6b06314c JA |
270 | /* |
271 | * If used, fixed file set. Writers must ensure that ->refs is dead, | |
272 | * readers must ensure that ->refs is alive as long as the file* is | |
273 | * used. Only updated through io_uring_register(2). | |
274 | */ | |
05f3fb3c | 275 | struct fixed_file_data *file_data; |
6b06314c | 276 | unsigned nr_user_files; |
b14cca0c PB |
277 | int ring_fd; |
278 | struct file *ring_file; | |
6b06314c | 279 | |
edafccee JA |
280 | /* if used, fixed mapped user buffers */ |
281 | unsigned nr_user_bufs; | |
282 | struct io_mapped_ubuf *user_bufs; | |
283 | ||
2b188cc1 JA |
284 | struct user_struct *user; |
285 | ||
0b8c0ec7 | 286 | const struct cred *creds; |
181e448d | 287 | |
0f158b4c JA |
288 | struct completion ref_comp; |
289 | struct completion sq_thread_comp; | |
206aefde | 290 | |
0ddf92e8 JA |
291 | /* if all else fails... */ |
292 | struct io_kiocb *fallback_req; | |
293 | ||
206aefde JA |
294 | #if defined(CONFIG_UNIX) |
295 | struct socket *ring_sock; | |
296 | #endif | |
297 | ||
5a2e745d JA |
298 | struct idr io_buffer_idr; |
299 | ||
071698e1 JA |
300 | struct idr personality_idr; |
301 | ||
206aefde JA |
302 | struct { |
303 | unsigned cached_cq_tail; | |
304 | unsigned cq_entries; | |
305 | unsigned cq_mask; | |
306 | atomic_t cq_timeouts; | |
ad3eb2c8 | 307 | unsigned long cq_check_overflow; |
206aefde JA |
308 | struct wait_queue_head cq_wait; |
309 | struct fasync_struct *cq_fasync; | |
310 | struct eventfd_ctx *cq_ev_fd; | |
311 | } ____cacheline_aligned_in_smp; | |
2b188cc1 JA |
312 | |
313 | struct { | |
314 | struct mutex uring_lock; | |
315 | wait_queue_head_t wait; | |
316 | } ____cacheline_aligned_in_smp; | |
317 | ||
318 | struct { | |
319 | spinlock_t completion_lock; | |
e94f141b | 320 | |
def596e9 JA |
321 | /* |
322 | * ->poll_list is protected by the ctx->uring_lock for | |
323 | * io_uring instances that don't use IORING_SETUP_SQPOLL. | |
324 | * For SQPOLL, only the single threaded io_sq_thread() will | |
325 | * manipulate the list, hence no extra locking is needed there. | |
326 | */ | |
327 | struct list_head poll_list; | |
78076bb6 JA |
328 | struct hlist_head *cancel_hash; |
329 | unsigned cancel_hash_bits; | |
e94f141b | 330 | bool poll_multi_file; |
31b51510 | 331 | |
fcb323cc JA |
332 | spinlock_t inflight_lock; |
333 | struct list_head inflight_list; | |
2b188cc1 | 334 | } ____cacheline_aligned_in_smp; |
85faa7b8 | 335 | |
4a38aed2 JA |
336 | struct delayed_work file_put_work; |
337 | struct llist_head file_put_llist; | |
338 | ||
85faa7b8 | 339 | struct work_struct exit_work; |
2b188cc1 JA |
340 | }; |
341 | ||
09bb8394 JA |
342 | /* |
343 | * First field must be the file pointer in all the | |
344 | * iocb unions! See also 'struct kiocb' in <linux/fs.h> | |
345 | */ | |
221c5eb2 JA |
346 | struct io_poll_iocb { |
347 | struct file *file; | |
0969e783 JA |
348 | union { |
349 | struct wait_queue_head *head; | |
350 | u64 addr; | |
351 | }; | |
221c5eb2 | 352 | __poll_t events; |
8c838788 | 353 | bool done; |
221c5eb2 | 354 | bool canceled; |
392edb45 | 355 | struct wait_queue_entry wait; |
221c5eb2 JA |
356 | }; |
357 | ||
b5dba59e JA |
358 | struct io_close { |
359 | struct file *file; | |
360 | struct file *put_file; | |
361 | int fd; | |
362 | }; | |
363 | ||
ad8a48ac JA |
364 | struct io_timeout_data { |
365 | struct io_kiocb *req; | |
366 | struct hrtimer timer; | |
367 | struct timespec64 ts; | |
368 | enum hrtimer_mode mode; | |
369 | }; | |
370 | ||
8ed8d3c3 JA |
371 | struct io_accept { |
372 | struct file *file; | |
373 | struct sockaddr __user *addr; | |
374 | int __user *addr_len; | |
375 | int flags; | |
09952e3e | 376 | unsigned long nofile; |
8ed8d3c3 JA |
377 | }; |
378 | ||
379 | struct io_sync { | |
380 | struct file *file; | |
381 | loff_t len; | |
382 | loff_t off; | |
383 | int flags; | |
d63d1b5e | 384 | int mode; |
8ed8d3c3 JA |
385 | }; |
386 | ||
fbf23849 JA |
387 | struct io_cancel { |
388 | struct file *file; | |
389 | u64 addr; | |
390 | }; | |
391 | ||
b29472ee JA |
392 | struct io_timeout { |
393 | struct file *file; | |
394 | u64 addr; | |
395 | int flags; | |
bfe68a22 PB |
396 | u32 off; |
397 | u32 target_seq; | |
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; | |
15b71abe | 428 | struct filename *filename; |
c12cedf2 | 429 | struct open_how how; |
4022e7af | 430 | unsigned long nofile; |
15b71abe JA |
431 | }; |
432 | ||
05f3fb3c JA |
433 | struct io_files_update { |
434 | struct file *file; | |
435 | u64 arg; | |
436 | u32 nr_args; | |
437 | u32 offset; | |
438 | }; | |
439 | ||
4840e418 JA |
440 | struct io_fadvise { |
441 | struct file *file; | |
442 | u64 offset; | |
443 | u32 len; | |
444 | u32 advice; | |
445 | }; | |
446 | ||
c1ca757b JA |
447 | struct io_madvise { |
448 | struct file *file; | |
449 | u64 addr; | |
450 | u32 len; | |
451 | u32 advice; | |
452 | }; | |
453 | ||
3e4827b0 JA |
454 | struct io_epoll { |
455 | struct file *file; | |
456 | int epfd; | |
457 | int op; | |
458 | int fd; | |
459 | struct epoll_event event; | |
e47293fd JA |
460 | }; |
461 | ||
7d67af2c PB |
462 | struct io_splice { |
463 | struct file *file_out; | |
464 | struct file *file_in; | |
465 | loff_t off_out; | |
466 | loff_t off_in; | |
467 | u64 len; | |
468 | unsigned int flags; | |
469 | }; | |
470 | ||
ddf0322d JA |
471 | struct io_provide_buf { |
472 | struct file *file; | |
473 | __u64 addr; | |
474 | __s32 len; | |
475 | __u32 bgid; | |
476 | __u16 nbufs; | |
477 | __u16 bid; | |
478 | }; | |
479 | ||
1d9e1288 BM |
480 | struct io_statx { |
481 | struct file *file; | |
482 | int dfd; | |
483 | unsigned int mask; | |
484 | unsigned int flags; | |
e62753e4 | 485 | const char __user *filename; |
1d9e1288 BM |
486 | struct statx __user *buffer; |
487 | }; | |
488 | ||
f499a021 JA |
489 | struct io_async_connect { |
490 | struct sockaddr_storage address; | |
491 | }; | |
492 | ||
03b1230c JA |
493 | struct io_async_msghdr { |
494 | struct iovec fast_iov[UIO_FASTIOV]; | |
495 | struct iovec *iov; | |
496 | struct sockaddr __user *uaddr; | |
497 | struct msghdr msg; | |
b537916c | 498 | struct sockaddr_storage addr; |
03b1230c JA |
499 | }; |
500 | ||
f67676d1 JA |
501 | struct io_async_rw { |
502 | struct iovec fast_iov[UIO_FASTIOV]; | |
503 | struct iovec *iov; | |
504 | ssize_t nr_segs; | |
505 | ssize_t size; | |
506 | }; | |
507 | ||
1a6b74fc | 508 | struct io_async_ctx { |
f67676d1 JA |
509 | union { |
510 | struct io_async_rw rw; | |
03b1230c | 511 | struct io_async_msghdr msg; |
f499a021 | 512 | struct io_async_connect connect; |
2d28390a | 513 | struct io_timeout_data timeout; |
f67676d1 | 514 | }; |
1a6b74fc JA |
515 | }; |
516 | ||
6b47ee6e PB |
517 | enum { |
518 | REQ_F_FIXED_FILE_BIT = IOSQE_FIXED_FILE_BIT, | |
519 | REQ_F_IO_DRAIN_BIT = IOSQE_IO_DRAIN_BIT, | |
520 | REQ_F_LINK_BIT = IOSQE_IO_LINK_BIT, | |
521 | REQ_F_HARDLINK_BIT = IOSQE_IO_HARDLINK_BIT, | |
522 | REQ_F_FORCE_ASYNC_BIT = IOSQE_ASYNC_BIT, | |
bcda7baa | 523 | REQ_F_BUFFER_SELECT_BIT = IOSQE_BUFFER_SELECT_BIT, |
6b47ee6e | 524 | |
dea3b49c | 525 | REQ_F_LINK_HEAD_BIT, |
6b47ee6e PB |
526 | REQ_F_LINK_NEXT_BIT, |
527 | REQ_F_FAIL_LINK_BIT, | |
528 | REQ_F_INFLIGHT_BIT, | |
529 | REQ_F_CUR_POS_BIT, | |
530 | REQ_F_NOWAIT_BIT, | |
6b47ee6e PB |
531 | REQ_F_LINK_TIMEOUT_BIT, |
532 | REQ_F_TIMEOUT_BIT, | |
533 | REQ_F_ISREG_BIT, | |
534 | REQ_F_MUST_PUNT_BIT, | |
535 | REQ_F_TIMEOUT_NOSEQ_BIT, | |
536 | REQ_F_COMP_LOCKED_BIT, | |
99bc4c38 | 537 | REQ_F_NEED_CLEANUP_BIT, |
2ca10259 | 538 | REQ_F_OVERFLOW_BIT, |
d7718a9d | 539 | REQ_F_POLLED_BIT, |
bcda7baa | 540 | REQ_F_BUFFER_SELECTED_BIT, |
5b0bbee4 | 541 | REQ_F_NO_FILE_TABLE_BIT, |
d4c81f38 | 542 | REQ_F_QUEUE_TIMEOUT_BIT, |
7cdaf587 | 543 | REQ_F_WORK_INITIALIZED_BIT, |
4dd2824d | 544 | REQ_F_TASK_PINNED_BIT, |
84557871 JA |
545 | |
546 | /* not a real bit, just to check we're not overflowing the space */ | |
547 | __REQ_F_LAST_BIT, | |
6b47ee6e PB |
548 | }; |
549 | ||
550 | enum { | |
551 | /* ctx owns file */ | |
552 | REQ_F_FIXED_FILE = BIT(REQ_F_FIXED_FILE_BIT), | |
553 | /* drain existing IO first */ | |
554 | REQ_F_IO_DRAIN = BIT(REQ_F_IO_DRAIN_BIT), | |
555 | /* linked sqes */ | |
556 | REQ_F_LINK = BIT(REQ_F_LINK_BIT), | |
557 | /* doesn't sever on completion < 0 */ | |
558 | REQ_F_HARDLINK = BIT(REQ_F_HARDLINK_BIT), | |
559 | /* IOSQE_ASYNC */ | |
560 | REQ_F_FORCE_ASYNC = BIT(REQ_F_FORCE_ASYNC_BIT), | |
bcda7baa JA |
561 | /* IOSQE_BUFFER_SELECT */ |
562 | REQ_F_BUFFER_SELECT = BIT(REQ_F_BUFFER_SELECT_BIT), | |
6b47ee6e | 563 | |
dea3b49c PB |
564 | /* head of a link */ |
565 | REQ_F_LINK_HEAD = BIT(REQ_F_LINK_HEAD_BIT), | |
6b47ee6e PB |
566 | /* already grabbed next link */ |
567 | REQ_F_LINK_NEXT = BIT(REQ_F_LINK_NEXT_BIT), | |
568 | /* fail rest of links */ | |
569 | REQ_F_FAIL_LINK = BIT(REQ_F_FAIL_LINK_BIT), | |
570 | /* on inflight list */ | |
571 | REQ_F_INFLIGHT = BIT(REQ_F_INFLIGHT_BIT), | |
572 | /* read/write uses file position */ | |
573 | REQ_F_CUR_POS = BIT(REQ_F_CUR_POS_BIT), | |
574 | /* must not punt to workers */ | |
575 | REQ_F_NOWAIT = BIT(REQ_F_NOWAIT_BIT), | |
6b47ee6e PB |
576 | /* has linked timeout */ |
577 | REQ_F_LINK_TIMEOUT = BIT(REQ_F_LINK_TIMEOUT_BIT), | |
578 | /* timeout request */ | |
579 | REQ_F_TIMEOUT = BIT(REQ_F_TIMEOUT_BIT), | |
580 | /* regular file */ | |
581 | REQ_F_ISREG = BIT(REQ_F_ISREG_BIT), | |
582 | /* must be punted even for NONBLOCK */ | |
583 | REQ_F_MUST_PUNT = BIT(REQ_F_MUST_PUNT_BIT), | |
584 | /* no timeout sequence */ | |
585 | REQ_F_TIMEOUT_NOSEQ = BIT(REQ_F_TIMEOUT_NOSEQ_BIT), | |
586 | /* completion under lock */ | |
587 | REQ_F_COMP_LOCKED = BIT(REQ_F_COMP_LOCKED_BIT), | |
99bc4c38 PB |
588 | /* needs cleanup */ |
589 | REQ_F_NEED_CLEANUP = BIT(REQ_F_NEED_CLEANUP_BIT), | |
2ca10259 JA |
590 | /* in overflow list */ |
591 | REQ_F_OVERFLOW = BIT(REQ_F_OVERFLOW_BIT), | |
d7718a9d JA |
592 | /* already went through poll handler */ |
593 | REQ_F_POLLED = BIT(REQ_F_POLLED_BIT), | |
bcda7baa JA |
594 | /* buffer already selected */ |
595 | REQ_F_BUFFER_SELECTED = BIT(REQ_F_BUFFER_SELECTED_BIT), | |
5b0bbee4 JA |
596 | /* doesn't need file table for this request */ |
597 | REQ_F_NO_FILE_TABLE = BIT(REQ_F_NO_FILE_TABLE_BIT), | |
d4c81f38 PB |
598 | /* needs to queue linked timeout */ |
599 | REQ_F_QUEUE_TIMEOUT = BIT(REQ_F_QUEUE_TIMEOUT_BIT), | |
7cdaf587 XW |
600 | /* io_wq_work is initialized */ |
601 | REQ_F_WORK_INITIALIZED = BIT(REQ_F_WORK_INITIALIZED_BIT), | |
4dd2824d PB |
602 | /* req->task is refcounted */ |
603 | REQ_F_TASK_PINNED = BIT(REQ_F_TASK_PINNED_BIT), | |
d7718a9d JA |
604 | }; |
605 | ||
606 | struct async_poll { | |
607 | struct io_poll_iocb poll; | |
608 | struct io_wq_work work; | |
6b47ee6e PB |
609 | }; |
610 | ||
09bb8394 JA |
611 | /* |
612 | * NOTE! Each of the iocb union members has the file pointer | |
613 | * as the first entry in their struct definition. So you can | |
614 | * access the file pointer through any of the sub-structs, | |
615 | * or directly as just 'ki_filp' in this struct. | |
616 | */ | |
2b188cc1 | 617 | struct io_kiocb { |
221c5eb2 | 618 | union { |
09bb8394 | 619 | struct file *file; |
9adbd45d | 620 | struct io_rw rw; |
221c5eb2 | 621 | struct io_poll_iocb poll; |
8ed8d3c3 JA |
622 | struct io_accept accept; |
623 | struct io_sync sync; | |
fbf23849 | 624 | struct io_cancel cancel; |
b29472ee | 625 | struct io_timeout timeout; |
3fbb51c1 | 626 | struct io_connect connect; |
e47293fd | 627 | struct io_sr_msg sr_msg; |
15b71abe | 628 | struct io_open open; |
b5dba59e | 629 | struct io_close close; |
05f3fb3c | 630 | struct io_files_update files_update; |
4840e418 | 631 | struct io_fadvise fadvise; |
c1ca757b | 632 | struct io_madvise madvise; |
3e4827b0 | 633 | struct io_epoll epoll; |
7d67af2c | 634 | struct io_splice splice; |
ddf0322d | 635 | struct io_provide_buf pbuf; |
1d9e1288 | 636 | struct io_statx statx; |
221c5eb2 | 637 | }; |
2b188cc1 | 638 | |
1a6b74fc | 639 | struct io_async_ctx *io; |
c398ecb3 | 640 | int cflags; |
d625c6ee | 641 | u8 opcode; |
65a6543d XW |
642 | /* polled IO has completed */ |
643 | u8 iopoll_completed; | |
2b188cc1 | 644 | |
4f4eeba8 BM |
645 | u16 buf_index; |
646 | ||
2b188cc1 | 647 | struct io_ring_ctx *ctx; |
d7718a9d | 648 | struct list_head list; |
2b188cc1 | 649 | unsigned int flags; |
c16361c1 | 650 | refcount_t refs; |
3537b6a7 JA |
651 | struct task_struct *task; |
652 | unsigned long fsize; | |
2b188cc1 | 653 | u64 user_data; |
9e645e11 | 654 | u32 result; |
de0617e4 | 655 | u32 sequence; |
2b188cc1 | 656 | |
d7718a9d JA |
657 | struct list_head link_list; |
658 | ||
fcb323cc JA |
659 | struct list_head inflight_entry; |
660 | ||
05589553 XW |
661 | struct percpu_ref *fixed_file_refs; |
662 | ||
b41e9852 JA |
663 | union { |
664 | /* | |
665 | * Only commands that never go async can use the below fields, | |
d7718a9d JA |
666 | * obviously. Right now only IORING_OP_POLL_ADD uses them, and |
667 | * async armed poll handlers for regular commands. The latter | |
668 | * restore the work, if needed. | |
b41e9852 JA |
669 | */ |
670 | struct { | |
b41e9852 | 671 | struct callback_head task_work; |
d7718a9d JA |
672 | struct hlist_node hash_node; |
673 | struct async_poll *apoll; | |
b41e9852 JA |
674 | }; |
675 | struct io_wq_work work; | |
676 | }; | |
2b188cc1 JA |
677 | }; |
678 | ||
679 | #define IO_PLUG_THRESHOLD 2 | |
def596e9 | 680 | #define IO_IOPOLL_BATCH 8 |
2b188cc1 | 681 | |
9a56a232 JA |
682 | struct io_submit_state { |
683 | struct blk_plug plug; | |
684 | ||
2579f913 JA |
685 | /* |
686 | * io_kiocb alloc cache | |
687 | */ | |
688 | void *reqs[IO_IOPOLL_BATCH]; | |
6c8a3134 | 689 | unsigned int free_reqs; |
2579f913 | 690 | |
9a56a232 JA |
691 | /* |
692 | * File reference cache | |
693 | */ | |
694 | struct file *file; | |
695 | unsigned int fd; | |
696 | unsigned int has_refs; | |
697 | unsigned int used_refs; | |
698 | unsigned int ios_left; | |
699 | }; | |
700 | ||
d3656344 JA |
701 | struct io_op_def { |
702 | /* needs req->io allocated for deferral/async */ | |
703 | unsigned async_ctx : 1; | |
704 | /* needs current->mm setup, does mm access */ | |
705 | unsigned needs_mm : 1; | |
706 | /* needs req->file assigned */ | |
707 | unsigned needs_file : 1; | |
fd2206e4 JA |
708 | /* don't fail if file grab fails */ |
709 | unsigned needs_file_no_error : 1; | |
d3656344 JA |
710 | /* hash wq insertion if file is a regular file */ |
711 | unsigned hash_reg_file : 1; | |
712 | /* unbound wq insertion if file is a non-regular file */ | |
713 | unsigned unbound_nonreg_file : 1; | |
66f4af93 JA |
714 | /* opcode is not supported by this kernel */ |
715 | unsigned not_supported : 1; | |
f86cd20c JA |
716 | /* needs file table */ |
717 | unsigned file_table : 1; | |
ff002b30 JA |
718 | /* needs ->fs */ |
719 | unsigned needs_fs : 1; | |
8a72758c JA |
720 | /* set if opcode supports polled "wait" */ |
721 | unsigned pollin : 1; | |
722 | unsigned pollout : 1; | |
bcda7baa JA |
723 | /* op supports buffer selection */ |
724 | unsigned buffer_select : 1; | |
d3656344 JA |
725 | }; |
726 | ||
727 | static const struct io_op_def io_op_defs[] = { | |
0463b6c5 PB |
728 | [IORING_OP_NOP] = {}, |
729 | [IORING_OP_READV] = { | |
d3656344 JA |
730 | .async_ctx = 1, |
731 | .needs_mm = 1, | |
732 | .needs_file = 1, | |
733 | .unbound_nonreg_file = 1, | |
8a72758c | 734 | .pollin = 1, |
4d954c25 | 735 | .buffer_select = 1, |
d3656344 | 736 | }, |
0463b6c5 | 737 | [IORING_OP_WRITEV] = { |
d3656344 JA |
738 | .async_ctx = 1, |
739 | .needs_mm = 1, | |
740 | .needs_file = 1, | |
741 | .hash_reg_file = 1, | |
742 | .unbound_nonreg_file = 1, | |
8a72758c | 743 | .pollout = 1, |
d3656344 | 744 | }, |
0463b6c5 | 745 | [IORING_OP_FSYNC] = { |
d3656344 JA |
746 | .needs_file = 1, |
747 | }, | |
0463b6c5 | 748 | [IORING_OP_READ_FIXED] = { |
d3656344 JA |
749 | .needs_file = 1, |
750 | .unbound_nonreg_file = 1, | |
8a72758c | 751 | .pollin = 1, |
d3656344 | 752 | }, |
0463b6c5 | 753 | [IORING_OP_WRITE_FIXED] = { |
d3656344 JA |
754 | .needs_file = 1, |
755 | .hash_reg_file = 1, | |
756 | .unbound_nonreg_file = 1, | |
8a72758c | 757 | .pollout = 1, |
d3656344 | 758 | }, |
0463b6c5 | 759 | [IORING_OP_POLL_ADD] = { |
d3656344 JA |
760 | .needs_file = 1, |
761 | .unbound_nonreg_file = 1, | |
762 | }, | |
0463b6c5 PB |
763 | [IORING_OP_POLL_REMOVE] = {}, |
764 | [IORING_OP_SYNC_FILE_RANGE] = { | |
d3656344 JA |
765 | .needs_file = 1, |
766 | }, | |
0463b6c5 | 767 | [IORING_OP_SENDMSG] = { |
d3656344 JA |
768 | .async_ctx = 1, |
769 | .needs_mm = 1, | |
770 | .needs_file = 1, | |
771 | .unbound_nonreg_file = 1, | |
ff002b30 | 772 | .needs_fs = 1, |
8a72758c | 773 | .pollout = 1, |
d3656344 | 774 | }, |
0463b6c5 | 775 | [IORING_OP_RECVMSG] = { |
d3656344 JA |
776 | .async_ctx = 1, |
777 | .needs_mm = 1, | |
778 | .needs_file = 1, | |
779 | .unbound_nonreg_file = 1, | |
ff002b30 | 780 | .needs_fs = 1, |
8a72758c | 781 | .pollin = 1, |
52de1fe1 | 782 | .buffer_select = 1, |
d3656344 | 783 | }, |
0463b6c5 | 784 | [IORING_OP_TIMEOUT] = { |
d3656344 JA |
785 | .async_ctx = 1, |
786 | .needs_mm = 1, | |
787 | }, | |
0463b6c5 PB |
788 | [IORING_OP_TIMEOUT_REMOVE] = {}, |
789 | [IORING_OP_ACCEPT] = { | |
d3656344 JA |
790 | .needs_mm = 1, |
791 | .needs_file = 1, | |
792 | .unbound_nonreg_file = 1, | |
f86cd20c | 793 | .file_table = 1, |
8a72758c | 794 | .pollin = 1, |
d3656344 | 795 | }, |
0463b6c5 PB |
796 | [IORING_OP_ASYNC_CANCEL] = {}, |
797 | [IORING_OP_LINK_TIMEOUT] = { | |
d3656344 JA |
798 | .async_ctx = 1, |
799 | .needs_mm = 1, | |
800 | }, | |
0463b6c5 | 801 | [IORING_OP_CONNECT] = { |
d3656344 JA |
802 | .async_ctx = 1, |
803 | .needs_mm = 1, | |
804 | .needs_file = 1, | |
805 | .unbound_nonreg_file = 1, | |
8a72758c | 806 | .pollout = 1, |
d3656344 | 807 | }, |
0463b6c5 | 808 | [IORING_OP_FALLOCATE] = { |
d3656344 JA |
809 | .needs_file = 1, |
810 | }, | |
0463b6c5 | 811 | [IORING_OP_OPENAT] = { |
f86cd20c | 812 | .file_table = 1, |
ff002b30 | 813 | .needs_fs = 1, |
d3656344 | 814 | }, |
0463b6c5 | 815 | [IORING_OP_CLOSE] = { |
fd2206e4 JA |
816 | .needs_file = 1, |
817 | .needs_file_no_error = 1, | |
f86cd20c | 818 | .file_table = 1, |
d3656344 | 819 | }, |
0463b6c5 | 820 | [IORING_OP_FILES_UPDATE] = { |
d3656344 | 821 | .needs_mm = 1, |
f86cd20c | 822 | .file_table = 1, |
d3656344 | 823 | }, |
0463b6c5 | 824 | [IORING_OP_STATX] = { |
d3656344 | 825 | .needs_mm = 1, |
ff002b30 | 826 | .needs_fs = 1, |
5b0bbee4 | 827 | .file_table = 1, |
d3656344 | 828 | }, |
0463b6c5 | 829 | [IORING_OP_READ] = { |
3a6820f2 JA |
830 | .needs_mm = 1, |
831 | .needs_file = 1, | |
832 | .unbound_nonreg_file = 1, | |
8a72758c | 833 | .pollin = 1, |
bcda7baa | 834 | .buffer_select = 1, |
3a6820f2 | 835 | }, |
0463b6c5 | 836 | [IORING_OP_WRITE] = { |
3a6820f2 JA |
837 | .needs_mm = 1, |
838 | .needs_file = 1, | |
839 | .unbound_nonreg_file = 1, | |
8a72758c | 840 | .pollout = 1, |
3a6820f2 | 841 | }, |
0463b6c5 | 842 | [IORING_OP_FADVISE] = { |
4840e418 JA |
843 | .needs_file = 1, |
844 | }, | |
0463b6c5 | 845 | [IORING_OP_MADVISE] = { |
c1ca757b JA |
846 | .needs_mm = 1, |
847 | }, | |
0463b6c5 | 848 | [IORING_OP_SEND] = { |
fddaface JA |
849 | .needs_mm = 1, |
850 | .needs_file = 1, | |
851 | .unbound_nonreg_file = 1, | |
8a72758c | 852 | .pollout = 1, |
fddaface | 853 | }, |
0463b6c5 | 854 | [IORING_OP_RECV] = { |
fddaface JA |
855 | .needs_mm = 1, |
856 | .needs_file = 1, | |
857 | .unbound_nonreg_file = 1, | |
8a72758c | 858 | .pollin = 1, |
bcda7baa | 859 | .buffer_select = 1, |
fddaface | 860 | }, |
0463b6c5 | 861 | [IORING_OP_OPENAT2] = { |
f86cd20c | 862 | .file_table = 1, |
ff002b30 | 863 | .needs_fs = 1, |
cebdb986 | 864 | }, |
3e4827b0 JA |
865 | [IORING_OP_EPOLL_CTL] = { |
866 | .unbound_nonreg_file = 1, | |
867 | .file_table = 1, | |
868 | }, | |
7d67af2c PB |
869 | [IORING_OP_SPLICE] = { |
870 | .needs_file = 1, | |
871 | .hash_reg_file = 1, | |
872 | .unbound_nonreg_file = 1, | |
ddf0322d JA |
873 | }, |
874 | [IORING_OP_PROVIDE_BUFFERS] = {}, | |
067524e9 | 875 | [IORING_OP_REMOVE_BUFFERS] = {}, |
f2a8d5c7 PB |
876 | [IORING_OP_TEE] = { |
877 | .needs_file = 1, | |
878 | .hash_reg_file = 1, | |
879 | .unbound_nonreg_file = 1, | |
880 | }, | |
d3656344 JA |
881 | }; |
882 | ||
561fb04a | 883 | static void io_wq_submit_work(struct io_wq_work **workptr); |
78e19bbe | 884 | static void io_cqring_fill_event(struct io_kiocb *req, long res); |
ec9c02ad | 885 | static void io_put_req(struct io_kiocb *req); |
978db57e | 886 | static void __io_double_put_req(struct io_kiocb *req); |
94ae5e77 JA |
887 | static struct io_kiocb *io_prep_linked_timeout(struct io_kiocb *req); |
888 | static void io_queue_linked_timeout(struct io_kiocb *req); | |
05f3fb3c JA |
889 | static int __io_sqe_files_update(struct io_ring_ctx *ctx, |
890 | struct io_uring_files_update *ip, | |
891 | unsigned nr_args); | |
f86cd20c | 892 | static int io_grab_files(struct io_kiocb *req); |
d60b5fbc | 893 | static void io_complete_rw_common(struct kiocb *kiocb, long res); |
99bc4c38 | 894 | static void io_cleanup_req(struct io_kiocb *req); |
b41e9852 JA |
895 | static int io_file_get(struct io_submit_state *state, struct io_kiocb *req, |
896 | int fd, struct file **out_file, bool fixed); | |
897 | static void __io_queue_sqe(struct io_kiocb *req, | |
898 | const struct io_uring_sqe *sqe); | |
de0617e4 | 899 | |
2b188cc1 JA |
900 | static struct kmem_cache *req_cachep; |
901 | ||
902 | static const struct file_operations io_uring_fops; | |
903 | ||
904 | struct sock *io_uring_get_socket(struct file *file) | |
905 | { | |
906 | #if defined(CONFIG_UNIX) | |
907 | if (file->f_op == &io_uring_fops) { | |
908 | struct io_ring_ctx *ctx = file->private_data; | |
909 | ||
910 | return ctx->ring_sock->sk; | |
911 | } | |
912 | #endif | |
913 | return NULL; | |
914 | } | |
915 | EXPORT_SYMBOL(io_uring_get_socket); | |
916 | ||
4dd2824d PB |
917 | static void io_get_req_task(struct io_kiocb *req) |
918 | { | |
919 | if (req->flags & REQ_F_TASK_PINNED) | |
920 | return; | |
921 | get_task_struct(req->task); | |
922 | req->flags |= REQ_F_TASK_PINNED; | |
923 | } | |
924 | ||
925 | /* not idempotent -- it doesn't clear REQ_F_TASK_PINNED */ | |
926 | static void __io_put_req_task(struct io_kiocb *req) | |
927 | { | |
928 | if (req->flags & REQ_F_TASK_PINNED) | |
929 | put_task_struct(req->task); | |
930 | } | |
931 | ||
4a38aed2 JA |
932 | static void io_file_put_work(struct work_struct *work); |
933 | ||
7cdaf587 XW |
934 | /* |
935 | * Note: must call io_req_init_async() for the first time you | |
936 | * touch any members of io_wq_work. | |
937 | */ | |
938 | static inline void io_req_init_async(struct io_kiocb *req) | |
939 | { | |
940 | if (req->flags & REQ_F_WORK_INITIALIZED) | |
941 | return; | |
942 | ||
943 | memset(&req->work, 0, sizeof(req->work)); | |
944 | req->flags |= REQ_F_WORK_INITIALIZED; | |
945 | } | |
946 | ||
0cdaf760 PB |
947 | static inline bool io_async_submit(struct io_ring_ctx *ctx) |
948 | { | |
949 | return ctx->flags & IORING_SETUP_SQPOLL; | |
950 | } | |
951 | ||
2b188cc1 JA |
952 | static void io_ring_ctx_ref_free(struct percpu_ref *ref) |
953 | { | |
954 | struct io_ring_ctx *ctx = container_of(ref, struct io_ring_ctx, refs); | |
955 | ||
0f158b4c | 956 | complete(&ctx->ref_comp); |
2b188cc1 JA |
957 | } |
958 | ||
959 | static struct io_ring_ctx *io_ring_ctx_alloc(struct io_uring_params *p) | |
960 | { | |
961 | struct io_ring_ctx *ctx; | |
78076bb6 | 962 | int hash_bits; |
2b188cc1 JA |
963 | |
964 | ctx = kzalloc(sizeof(*ctx), GFP_KERNEL); | |
965 | if (!ctx) | |
966 | return NULL; | |
967 | ||
0ddf92e8 JA |
968 | ctx->fallback_req = kmem_cache_alloc(req_cachep, GFP_KERNEL); |
969 | if (!ctx->fallback_req) | |
970 | goto err; | |
971 | ||
78076bb6 JA |
972 | /* |
973 | * Use 5 bits less than the max cq entries, that should give us around | |
974 | * 32 entries per hash list if totally full and uniformly spread. | |
975 | */ | |
976 | hash_bits = ilog2(p->cq_entries); | |
977 | hash_bits -= 5; | |
978 | if (hash_bits <= 0) | |
979 | hash_bits = 1; | |
980 | ctx->cancel_hash_bits = hash_bits; | |
981 | ctx->cancel_hash = kmalloc((1U << hash_bits) * sizeof(struct hlist_head), | |
982 | GFP_KERNEL); | |
983 | if (!ctx->cancel_hash) | |
984 | goto err; | |
985 | __hash_init(ctx->cancel_hash, 1U << hash_bits); | |
986 | ||
21482896 | 987 | if (percpu_ref_init(&ctx->refs, io_ring_ctx_ref_free, |
206aefde JA |
988 | PERCPU_REF_ALLOW_REINIT, GFP_KERNEL)) |
989 | goto err; | |
2b188cc1 JA |
990 | |
991 | ctx->flags = p->flags; | |
583863ed | 992 | init_waitqueue_head(&ctx->sqo_wait); |
2b188cc1 | 993 | init_waitqueue_head(&ctx->cq_wait); |
1d7bb1d5 | 994 | INIT_LIST_HEAD(&ctx->cq_overflow_list); |
0f158b4c JA |
995 | init_completion(&ctx->ref_comp); |
996 | init_completion(&ctx->sq_thread_comp); | |
5a2e745d | 997 | idr_init(&ctx->io_buffer_idr); |
071698e1 | 998 | idr_init(&ctx->personality_idr); |
2b188cc1 JA |
999 | mutex_init(&ctx->uring_lock); |
1000 | init_waitqueue_head(&ctx->wait); | |
1001 | spin_lock_init(&ctx->completion_lock); | |
def596e9 | 1002 | INIT_LIST_HEAD(&ctx->poll_list); |
de0617e4 | 1003 | INIT_LIST_HEAD(&ctx->defer_list); |
5262f567 | 1004 | INIT_LIST_HEAD(&ctx->timeout_list); |
fcb323cc JA |
1005 | init_waitqueue_head(&ctx->inflight_wait); |
1006 | spin_lock_init(&ctx->inflight_lock); | |
1007 | INIT_LIST_HEAD(&ctx->inflight_list); | |
4a38aed2 JA |
1008 | INIT_DELAYED_WORK(&ctx->file_put_work, io_file_put_work); |
1009 | init_llist_head(&ctx->file_put_llist); | |
2b188cc1 | 1010 | return ctx; |
206aefde | 1011 | err: |
0ddf92e8 JA |
1012 | if (ctx->fallback_req) |
1013 | kmem_cache_free(req_cachep, ctx->fallback_req); | |
78076bb6 | 1014 | kfree(ctx->cancel_hash); |
206aefde JA |
1015 | kfree(ctx); |
1016 | return NULL; | |
2b188cc1 JA |
1017 | } |
1018 | ||
9d858b21 | 1019 | static inline bool __req_need_defer(struct io_kiocb *req) |
7adf4eaf | 1020 | { |
a197f664 JL |
1021 | struct io_ring_ctx *ctx = req->ctx; |
1022 | ||
31af27c7 PB |
1023 | return req->sequence != ctx->cached_cq_tail |
1024 | + atomic_read(&ctx->cached_cq_overflow); | |
7adf4eaf JA |
1025 | } |
1026 | ||
9d858b21 | 1027 | static inline bool req_need_defer(struct io_kiocb *req) |
de0617e4 | 1028 | { |
87987898 | 1029 | if (unlikely(req->flags & REQ_F_IO_DRAIN)) |
9d858b21 | 1030 | return __req_need_defer(req); |
de0617e4 | 1031 | |
9d858b21 | 1032 | return false; |
de0617e4 JA |
1033 | } |
1034 | ||
de0617e4 | 1035 | static void __io_commit_cqring(struct io_ring_ctx *ctx) |
2b188cc1 | 1036 | { |
75b28aff | 1037 | struct io_rings *rings = ctx->rings; |
2b188cc1 | 1038 | |
07910158 PB |
1039 | /* order cqe stores with ring update */ |
1040 | smp_store_release(&rings->cq.tail, ctx->cached_cq_tail); | |
2b188cc1 | 1041 | |
07910158 PB |
1042 | if (wq_has_sleeper(&ctx->cq_wait)) { |
1043 | wake_up_interruptible(&ctx->cq_wait); | |
1044 | kill_fasync(&ctx->cq_fasync, SIGIO, POLL_IN); | |
2b188cc1 JA |
1045 | } |
1046 | } | |
1047 | ||
cccf0ee8 JA |
1048 | static inline void io_req_work_grab_env(struct io_kiocb *req, |
1049 | const struct io_op_def *def) | |
1050 | { | |
1051 | if (!req->work.mm && def->needs_mm) { | |
1052 | mmgrab(current->mm); | |
1053 | req->work.mm = current->mm; | |
2b188cc1 | 1054 | } |
cccf0ee8 JA |
1055 | if (!req->work.creds) |
1056 | req->work.creds = get_current_cred(); | |
ff002b30 JA |
1057 | if (!req->work.fs && def->needs_fs) { |
1058 | spin_lock(¤t->fs->lock); | |
1059 | if (!current->fs->in_exec) { | |
1060 | req->work.fs = current->fs; | |
1061 | req->work.fs->users++; | |
1062 | } else { | |
1063 | req->work.flags |= IO_WQ_WORK_CANCEL; | |
1064 | } | |
1065 | spin_unlock(¤t->fs->lock); | |
1066 | } | |
2b188cc1 JA |
1067 | } |
1068 | ||
cccf0ee8 | 1069 | static inline void io_req_work_drop_env(struct io_kiocb *req) |
18d9be1a | 1070 | { |
7cdaf587 XW |
1071 | if (!(req->flags & REQ_F_WORK_INITIALIZED)) |
1072 | return; | |
1073 | ||
cccf0ee8 JA |
1074 | if (req->work.mm) { |
1075 | mmdrop(req->work.mm); | |
1076 | req->work.mm = NULL; | |
1077 | } | |
1078 | if (req->work.creds) { | |
1079 | put_cred(req->work.creds); | |
1080 | req->work.creds = NULL; | |
1081 | } | |
ff002b30 JA |
1082 | if (req->work.fs) { |
1083 | struct fs_struct *fs = req->work.fs; | |
1084 | ||
1085 | spin_lock(&req->work.fs->lock); | |
1086 | if (--fs->users) | |
1087 | fs = NULL; | |
1088 | spin_unlock(&req->work.fs->lock); | |
1089 | if (fs) | |
1090 | free_fs_struct(fs); | |
1091 | } | |
561fb04a JA |
1092 | } |
1093 | ||
8766dd51 | 1094 | static inline void io_prep_async_work(struct io_kiocb *req, |
94ae5e77 | 1095 | struct io_kiocb **link) |
18d9be1a | 1096 | { |
d3656344 | 1097 | const struct io_op_def *def = &io_op_defs[req->opcode]; |
54a91f3b | 1098 | |
d3656344 JA |
1099 | if (req->flags & REQ_F_ISREG) { |
1100 | if (def->hash_reg_file) | |
8766dd51 | 1101 | io_wq_hash_work(&req->work, file_inode(req->file)); |
d3656344 JA |
1102 | } else { |
1103 | if (def->unbound_nonreg_file) | |
3529d8c2 | 1104 | req->work.flags |= IO_WQ_WORK_UNBOUND; |
54a91f3b | 1105 | } |
cccf0ee8 | 1106 | |
59960b9d | 1107 | io_req_init_async(req); |
cccf0ee8 | 1108 | io_req_work_grab_env(req, def); |
54a91f3b | 1109 | |
94ae5e77 | 1110 | *link = io_prep_linked_timeout(req); |
561fb04a JA |
1111 | } |
1112 | ||
a197f664 | 1113 | static inline void io_queue_async_work(struct io_kiocb *req) |
561fb04a | 1114 | { |
a197f664 | 1115 | struct io_ring_ctx *ctx = req->ctx; |
94ae5e77 | 1116 | struct io_kiocb *link; |
94ae5e77 | 1117 | |
8766dd51 | 1118 | io_prep_async_work(req, &link); |
561fb04a | 1119 | |
8766dd51 PB |
1120 | trace_io_uring_queue_async_work(ctx, io_wq_is_hashed(&req->work), req, |
1121 | &req->work, req->flags); | |
1122 | io_wq_enqueue(ctx->io_wq, &req->work); | |
94ae5e77 JA |
1123 | |
1124 | if (link) | |
1125 | io_queue_linked_timeout(link); | |
18d9be1a JA |
1126 | } |
1127 | ||
5262f567 JA |
1128 | static void io_kill_timeout(struct io_kiocb *req) |
1129 | { | |
1130 | int ret; | |
1131 | ||
2d28390a | 1132 | ret = hrtimer_try_to_cancel(&req->io->timeout.timer); |
5262f567 JA |
1133 | if (ret != -1) { |
1134 | atomic_inc(&req->ctx->cq_timeouts); | |
842f9612 | 1135 | list_del_init(&req->list); |
f0e20b89 | 1136 | req->flags |= REQ_F_COMP_LOCKED; |
78e19bbe | 1137 | io_cqring_fill_event(req, 0); |
ec9c02ad | 1138 | io_put_req(req); |
5262f567 JA |
1139 | } |
1140 | } | |
1141 | ||
1142 | static void io_kill_timeouts(struct io_ring_ctx *ctx) | |
1143 | { | |
1144 | struct io_kiocb *req, *tmp; | |
1145 | ||
1146 | spin_lock_irq(&ctx->completion_lock); | |
1147 | list_for_each_entry_safe(req, tmp, &ctx->timeout_list, list) | |
1148 | io_kill_timeout(req); | |
1149 | spin_unlock_irq(&ctx->completion_lock); | |
1150 | } | |
1151 | ||
04518945 | 1152 | static void __io_queue_deferred(struct io_ring_ctx *ctx) |
de0617e4 | 1153 | { |
04518945 PB |
1154 | do { |
1155 | struct io_kiocb *req = list_first_entry(&ctx->defer_list, | |
1156 | struct io_kiocb, list); | |
de0617e4 | 1157 | |
04518945 PB |
1158 | if (req_need_defer(req)) |
1159 | break; | |
1160 | list_del_init(&req->list); | |
1161 | io_queue_async_work(req); | |
1162 | } while (!list_empty(&ctx->defer_list)); | |
1163 | } | |
1164 | ||
360428f8 | 1165 | static void io_flush_timeouts(struct io_ring_ctx *ctx) |
de0617e4 | 1166 | { |
360428f8 PB |
1167 | while (!list_empty(&ctx->timeout_list)) { |
1168 | struct io_kiocb *req = list_first_entry(&ctx->timeout_list, | |
1169 | struct io_kiocb, list); | |
de0617e4 | 1170 | |
360428f8 PB |
1171 | if (req->flags & REQ_F_TIMEOUT_NOSEQ) |
1172 | break; | |
bfe68a22 PB |
1173 | if (req->timeout.target_seq != ctx->cached_cq_tail |
1174 | - atomic_read(&ctx->cq_timeouts)) | |
360428f8 | 1175 | break; |
bfe68a22 | 1176 | |
360428f8 | 1177 | list_del_init(&req->list); |
5262f567 | 1178 | io_kill_timeout(req); |
360428f8 PB |
1179 | } |
1180 | } | |
5262f567 | 1181 | |
360428f8 PB |
1182 | static void io_commit_cqring(struct io_ring_ctx *ctx) |
1183 | { | |
1184 | io_flush_timeouts(ctx); | |
de0617e4 JA |
1185 | __io_commit_cqring(ctx); |
1186 | ||
04518945 PB |
1187 | if (unlikely(!list_empty(&ctx->defer_list))) |
1188 | __io_queue_deferred(ctx); | |
de0617e4 JA |
1189 | } |
1190 | ||
2b188cc1 JA |
1191 | static struct io_uring_cqe *io_get_cqring(struct io_ring_ctx *ctx) |
1192 | { | |
75b28aff | 1193 | struct io_rings *rings = ctx->rings; |
2b188cc1 JA |
1194 | unsigned tail; |
1195 | ||
1196 | tail = ctx->cached_cq_tail; | |
115e12e5 SB |
1197 | /* |
1198 | * writes to the cq entry need to come after reading head; the | |
1199 | * control dependency is enough as we're using WRITE_ONCE to | |
1200 | * fill the cq entry | |
1201 | */ | |
75b28aff | 1202 | if (tail - READ_ONCE(rings->cq.head) == rings->cq_ring_entries) |
2b188cc1 JA |
1203 | return NULL; |
1204 | ||
1205 | ctx->cached_cq_tail++; | |
75b28aff | 1206 | return &rings->cqes[tail & ctx->cq_mask]; |
2b188cc1 JA |
1207 | } |
1208 | ||
f2842ab5 JA |
1209 | static inline bool io_should_trigger_evfd(struct io_ring_ctx *ctx) |
1210 | { | |
f0b493e6 JA |
1211 | if (!ctx->cq_ev_fd) |
1212 | return false; | |
7e55a19c SG |
1213 | if (READ_ONCE(ctx->rings->cq_flags) & IORING_CQ_EVENTFD_DISABLED) |
1214 | return false; | |
f2842ab5 JA |
1215 | if (!ctx->eventfd_async) |
1216 | return true; | |
b41e9852 | 1217 | return io_wq_current_is_worker(); |
f2842ab5 JA |
1218 | } |
1219 | ||
b41e9852 | 1220 | static void io_cqring_ev_posted(struct io_ring_ctx *ctx) |
1d7bb1d5 JA |
1221 | { |
1222 | if (waitqueue_active(&ctx->wait)) | |
1223 | wake_up(&ctx->wait); | |
1224 | if (waitqueue_active(&ctx->sqo_wait)) | |
1225 | wake_up(&ctx->sqo_wait); | |
b41e9852 | 1226 | if (io_should_trigger_evfd(ctx)) |
1d7bb1d5 JA |
1227 | eventfd_signal(ctx->cq_ev_fd, 1); |
1228 | } | |
1229 | ||
c4a2ed72 JA |
1230 | /* Returns true if there are no backlogged entries after the flush */ |
1231 | static bool io_cqring_overflow_flush(struct io_ring_ctx *ctx, bool force) | |
1d7bb1d5 JA |
1232 | { |
1233 | struct io_rings *rings = ctx->rings; | |
1234 | struct io_uring_cqe *cqe; | |
1235 | struct io_kiocb *req; | |
1236 | unsigned long flags; | |
1237 | LIST_HEAD(list); | |
1238 | ||
1239 | if (!force) { | |
1240 | if (list_empty_careful(&ctx->cq_overflow_list)) | |
c4a2ed72 | 1241 | return true; |
1d7bb1d5 JA |
1242 | if ((ctx->cached_cq_tail - READ_ONCE(rings->cq.head) == |
1243 | rings->cq_ring_entries)) | |
c4a2ed72 | 1244 | return false; |
1d7bb1d5 JA |
1245 | } |
1246 | ||
1247 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
1248 | ||
1249 | /* if force is set, the ring is going away. always drop after that */ | |
1250 | if (force) | |
69b3e546 | 1251 | ctx->cq_overflow_flushed = 1; |
1d7bb1d5 | 1252 | |
c4a2ed72 | 1253 | cqe = NULL; |
1d7bb1d5 JA |
1254 | while (!list_empty(&ctx->cq_overflow_list)) { |
1255 | cqe = io_get_cqring(ctx); | |
1256 | if (!cqe && !force) | |
1257 | break; | |
1258 | ||
1259 | req = list_first_entry(&ctx->cq_overflow_list, struct io_kiocb, | |
1260 | list); | |
1261 | list_move(&req->list, &list); | |
2ca10259 | 1262 | req->flags &= ~REQ_F_OVERFLOW; |
1d7bb1d5 JA |
1263 | if (cqe) { |
1264 | WRITE_ONCE(cqe->user_data, req->user_data); | |
1265 | WRITE_ONCE(cqe->res, req->result); | |
bcda7baa | 1266 | WRITE_ONCE(cqe->flags, req->cflags); |
1d7bb1d5 JA |
1267 | } else { |
1268 | WRITE_ONCE(ctx->rings->cq_overflow, | |
1269 | atomic_inc_return(&ctx->cached_cq_overflow)); | |
1270 | } | |
1271 | } | |
1272 | ||
1273 | io_commit_cqring(ctx); | |
ad3eb2c8 JA |
1274 | if (cqe) { |
1275 | clear_bit(0, &ctx->sq_check_overflow); | |
1276 | clear_bit(0, &ctx->cq_check_overflow); | |
1277 | } | |
1d7bb1d5 JA |
1278 | spin_unlock_irqrestore(&ctx->completion_lock, flags); |
1279 | io_cqring_ev_posted(ctx); | |
1280 | ||
1281 | while (!list_empty(&list)) { | |
1282 | req = list_first_entry(&list, struct io_kiocb, list); | |
1283 | list_del(&req->list); | |
ec9c02ad | 1284 | io_put_req(req); |
1d7bb1d5 | 1285 | } |
c4a2ed72 JA |
1286 | |
1287 | return cqe != NULL; | |
1d7bb1d5 JA |
1288 | } |
1289 | ||
bcda7baa | 1290 | static void __io_cqring_fill_event(struct io_kiocb *req, long res, long cflags) |
2b188cc1 | 1291 | { |
78e19bbe | 1292 | struct io_ring_ctx *ctx = req->ctx; |
2b188cc1 JA |
1293 | struct io_uring_cqe *cqe; |
1294 | ||
78e19bbe | 1295 | trace_io_uring_complete(ctx, req->user_data, res); |
51c3ff62 | 1296 | |
2b188cc1 JA |
1297 | /* |
1298 | * If we can't get a cq entry, userspace overflowed the | |
1299 | * submission (by quite a lot). Increment the overflow count in | |
1300 | * the ring. | |
1301 | */ | |
1302 | cqe = io_get_cqring(ctx); | |
1d7bb1d5 | 1303 | if (likely(cqe)) { |
78e19bbe | 1304 | WRITE_ONCE(cqe->user_data, req->user_data); |
2b188cc1 | 1305 | WRITE_ONCE(cqe->res, res); |
bcda7baa | 1306 | WRITE_ONCE(cqe->flags, cflags); |
1d7bb1d5 | 1307 | } else if (ctx->cq_overflow_flushed) { |
498ccd9e JA |
1308 | WRITE_ONCE(ctx->rings->cq_overflow, |
1309 | atomic_inc_return(&ctx->cached_cq_overflow)); | |
1d7bb1d5 | 1310 | } else { |
ad3eb2c8 JA |
1311 | if (list_empty(&ctx->cq_overflow_list)) { |
1312 | set_bit(0, &ctx->sq_check_overflow); | |
1313 | set_bit(0, &ctx->cq_check_overflow); | |
1314 | } | |
2ca10259 | 1315 | req->flags |= REQ_F_OVERFLOW; |
1d7bb1d5 JA |
1316 | refcount_inc(&req->refs); |
1317 | req->result = res; | |
bcda7baa | 1318 | req->cflags = cflags; |
1d7bb1d5 | 1319 | list_add_tail(&req->list, &ctx->cq_overflow_list); |
2b188cc1 JA |
1320 | } |
1321 | } | |
1322 | ||
bcda7baa JA |
1323 | static void io_cqring_fill_event(struct io_kiocb *req, long res) |
1324 | { | |
1325 | __io_cqring_fill_event(req, res, 0); | |
1326 | } | |
1327 | ||
1328 | static void __io_cqring_add_event(struct io_kiocb *req, long res, long cflags) | |
2b188cc1 | 1329 | { |
78e19bbe | 1330 | struct io_ring_ctx *ctx = req->ctx; |
2b188cc1 JA |
1331 | unsigned long flags; |
1332 | ||
1333 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
bcda7baa | 1334 | __io_cqring_fill_event(req, res, cflags); |
2b188cc1 JA |
1335 | io_commit_cqring(ctx); |
1336 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
1337 | ||
8c838788 | 1338 | io_cqring_ev_posted(ctx); |
2b188cc1 JA |
1339 | } |
1340 | ||
bcda7baa JA |
1341 | static void io_cqring_add_event(struct io_kiocb *req, long res) |
1342 | { | |
1343 | __io_cqring_add_event(req, res, 0); | |
1344 | } | |
1345 | ||
0ddf92e8 JA |
1346 | static inline bool io_is_fallback_req(struct io_kiocb *req) |
1347 | { | |
1348 | return req == (struct io_kiocb *) | |
1349 | ((unsigned long) req->ctx->fallback_req & ~1UL); | |
1350 | } | |
1351 | ||
1352 | static struct io_kiocb *io_get_fallback_req(struct io_ring_ctx *ctx) | |
1353 | { | |
1354 | struct io_kiocb *req; | |
1355 | ||
1356 | req = ctx->fallback_req; | |
dd461af6 | 1357 | if (!test_and_set_bit_lock(0, (unsigned long *) &ctx->fallback_req)) |
0ddf92e8 JA |
1358 | return req; |
1359 | ||
1360 | return NULL; | |
1361 | } | |
1362 | ||
0553b8bd PB |
1363 | static struct io_kiocb *io_alloc_req(struct io_ring_ctx *ctx, |
1364 | struct io_submit_state *state) | |
2b188cc1 | 1365 | { |
fd6fab2c | 1366 | gfp_t gfp = GFP_KERNEL | __GFP_NOWARN; |
2b188cc1 JA |
1367 | struct io_kiocb *req; |
1368 | ||
2579f913 | 1369 | if (!state) { |
fd6fab2c | 1370 | req = kmem_cache_alloc(req_cachep, gfp); |
2579f913 | 1371 | if (unlikely(!req)) |
0ddf92e8 | 1372 | goto fallback; |
2579f913 JA |
1373 | } else if (!state->free_reqs) { |
1374 | size_t sz; | |
1375 | int ret; | |
1376 | ||
1377 | sz = min_t(size_t, state->ios_left, ARRAY_SIZE(state->reqs)); | |
fd6fab2c JA |
1378 | ret = kmem_cache_alloc_bulk(req_cachep, gfp, sz, state->reqs); |
1379 | ||
1380 | /* | |
1381 | * Bulk alloc is all-or-nothing. If we fail to get a batch, | |
1382 | * retry single alloc to be on the safe side. | |
1383 | */ | |
1384 | if (unlikely(ret <= 0)) { | |
1385 | state->reqs[0] = kmem_cache_alloc(req_cachep, gfp); | |
1386 | if (!state->reqs[0]) | |
0ddf92e8 | 1387 | goto fallback; |
fd6fab2c JA |
1388 | ret = 1; |
1389 | } | |
2579f913 | 1390 | state->free_reqs = ret - 1; |
6c8a3134 | 1391 | req = state->reqs[ret - 1]; |
2579f913 | 1392 | } else { |
2579f913 | 1393 | state->free_reqs--; |
6c8a3134 | 1394 | req = state->reqs[state->free_reqs]; |
2b188cc1 JA |
1395 | } |
1396 | ||
2579f913 | 1397 | return req; |
0ddf92e8 | 1398 | fallback: |
0553b8bd | 1399 | return io_get_fallback_req(ctx); |
2b188cc1 JA |
1400 | } |
1401 | ||
8da11c19 PB |
1402 | static inline void io_put_file(struct io_kiocb *req, struct file *file, |
1403 | bool fixed) | |
1404 | { | |
1405 | if (fixed) | |
05589553 | 1406 | percpu_ref_put(req->fixed_file_refs); |
8da11c19 PB |
1407 | else |
1408 | fput(file); | |
1409 | } | |
1410 | ||
c6ca97b3 | 1411 | static void __io_req_aux_free(struct io_kiocb *req) |
2b188cc1 | 1412 | { |
929a3af9 PB |
1413 | if (req->flags & REQ_F_NEED_CLEANUP) |
1414 | io_cleanup_req(req); | |
1415 | ||
96fd84d8 | 1416 | kfree(req->io); |
8da11c19 PB |
1417 | if (req->file) |
1418 | io_put_file(req, req->file, (req->flags & REQ_F_FIXED_FILE)); | |
4dd2824d | 1419 | __io_put_req_task(req); |
cccf0ee8 | 1420 | io_req_work_drop_env(req); |
def596e9 JA |
1421 | } |
1422 | ||
9e645e11 | 1423 | static void __io_free_req(struct io_kiocb *req) |
2b188cc1 | 1424 | { |
c6ca97b3 | 1425 | __io_req_aux_free(req); |
fcb323cc | 1426 | |
fcb323cc | 1427 | if (req->flags & REQ_F_INFLIGHT) { |
c6ca97b3 | 1428 | struct io_ring_ctx *ctx = req->ctx; |
fcb323cc JA |
1429 | unsigned long flags; |
1430 | ||
1431 | spin_lock_irqsave(&ctx->inflight_lock, flags); | |
1432 | list_del(&req->inflight_entry); | |
1433 | if (waitqueue_active(&ctx->inflight_wait)) | |
1434 | wake_up(&ctx->inflight_wait); | |
1435 | spin_unlock_irqrestore(&ctx->inflight_lock, flags); | |
1436 | } | |
2b85edfc PB |
1437 | |
1438 | percpu_ref_put(&req->ctx->refs); | |
b1e50e54 PB |
1439 | if (likely(!io_is_fallback_req(req))) |
1440 | kmem_cache_free(req_cachep, req); | |
1441 | else | |
dd461af6 | 1442 | clear_bit_unlock(0, (unsigned long *) &req->ctx->fallback_req); |
e65ef56d JA |
1443 | } |
1444 | ||
c6ca97b3 JA |
1445 | struct req_batch { |
1446 | void *reqs[IO_IOPOLL_BATCH]; | |
1447 | int to_free; | |
1448 | int need_iter; | |
1449 | }; | |
1450 | ||
1451 | static void io_free_req_many(struct io_ring_ctx *ctx, struct req_batch *rb) | |
1452 | { | |
1453 | if (!rb->to_free) | |
1454 | return; | |
1455 | if (rb->need_iter) { | |
1456 | int i, inflight = 0; | |
1457 | unsigned long flags; | |
1458 | ||
1459 | for (i = 0; i < rb->to_free; i++) { | |
1460 | struct io_kiocb *req = rb->reqs[i]; | |
1461 | ||
c6ca97b3 JA |
1462 | if (req->flags & REQ_F_INFLIGHT) |
1463 | inflight++; | |
c6ca97b3 JA |
1464 | __io_req_aux_free(req); |
1465 | } | |
1466 | if (!inflight) | |
1467 | goto do_free; | |
1468 | ||
1469 | spin_lock_irqsave(&ctx->inflight_lock, flags); | |
1470 | for (i = 0; i < rb->to_free; i++) { | |
1471 | struct io_kiocb *req = rb->reqs[i]; | |
1472 | ||
10fef4be | 1473 | if (req->flags & REQ_F_INFLIGHT) { |
c6ca97b3 JA |
1474 | list_del(&req->inflight_entry); |
1475 | if (!--inflight) | |
1476 | break; | |
1477 | } | |
1478 | } | |
1479 | spin_unlock_irqrestore(&ctx->inflight_lock, flags); | |
1480 | ||
1481 | if (waitqueue_active(&ctx->inflight_wait)) | |
1482 | wake_up(&ctx->inflight_wait); | |
1483 | } | |
1484 | do_free: | |
1485 | kmem_cache_free_bulk(req_cachep, rb->to_free, rb->reqs); | |
1486 | percpu_ref_put_many(&ctx->refs, rb->to_free); | |
c6ca97b3 | 1487 | rb->to_free = rb->need_iter = 0; |
e65ef56d JA |
1488 | } |
1489 | ||
a197f664 | 1490 | static bool io_link_cancel_timeout(struct io_kiocb *req) |
2665abfd | 1491 | { |
a197f664 | 1492 | struct io_ring_ctx *ctx = req->ctx; |
2665abfd JA |
1493 | int ret; |
1494 | ||
2d28390a | 1495 | ret = hrtimer_try_to_cancel(&req->io->timeout.timer); |
2665abfd | 1496 | if (ret != -1) { |
78e19bbe | 1497 | io_cqring_fill_event(req, -ECANCELED); |
2665abfd | 1498 | io_commit_cqring(ctx); |
dea3b49c | 1499 | req->flags &= ~REQ_F_LINK_HEAD; |
ec9c02ad | 1500 | io_put_req(req); |
2665abfd JA |
1501 | return true; |
1502 | } | |
1503 | ||
1504 | return false; | |
e65ef56d JA |
1505 | } |
1506 | ||
ba816ad6 | 1507 | static void io_req_link_next(struct io_kiocb *req, struct io_kiocb **nxtptr) |
9e645e11 | 1508 | { |
2665abfd | 1509 | struct io_ring_ctx *ctx = req->ctx; |
2665abfd | 1510 | bool wake_ev = false; |
9e645e11 | 1511 | |
4d7dd462 JA |
1512 | /* Already got next link */ |
1513 | if (req->flags & REQ_F_LINK_NEXT) | |
1514 | return; | |
1515 | ||
9e645e11 JA |
1516 | /* |
1517 | * The list should never be empty when we are called here. But could | |
1518 | * potentially happen if the chain is messed up, check to be on the | |
1519 | * safe side. | |
1520 | */ | |
4493233e PB |
1521 | while (!list_empty(&req->link_list)) { |
1522 | struct io_kiocb *nxt = list_first_entry(&req->link_list, | |
1523 | struct io_kiocb, link_list); | |
94ae5e77 | 1524 | |
4493233e PB |
1525 | if (unlikely((req->flags & REQ_F_LINK_TIMEOUT) && |
1526 | (nxt->flags & REQ_F_TIMEOUT))) { | |
1527 | list_del_init(&nxt->link_list); | |
94ae5e77 | 1528 | wake_ev |= io_link_cancel_timeout(nxt); |
94ae5e77 JA |
1529 | req->flags &= ~REQ_F_LINK_TIMEOUT; |
1530 | continue; | |
1531 | } | |
9e645e11 | 1532 | |
4493233e PB |
1533 | list_del_init(&req->link_list); |
1534 | if (!list_empty(&nxt->link_list)) | |
dea3b49c | 1535 | nxt->flags |= REQ_F_LINK_HEAD; |
b18fdf71 | 1536 | *nxtptr = nxt; |
94ae5e77 | 1537 | break; |
9e645e11 | 1538 | } |
2665abfd | 1539 | |
4d7dd462 | 1540 | req->flags |= REQ_F_LINK_NEXT; |
2665abfd JA |
1541 | if (wake_ev) |
1542 | io_cqring_ev_posted(ctx); | |
9e645e11 JA |
1543 | } |
1544 | ||
1545 | /* | |
dea3b49c | 1546 | * Called if REQ_F_LINK_HEAD is set, and we fail the head request |
9e645e11 JA |
1547 | */ |
1548 | static void io_fail_links(struct io_kiocb *req) | |
1549 | { | |
2665abfd | 1550 | struct io_ring_ctx *ctx = req->ctx; |
2665abfd JA |
1551 | unsigned long flags; |
1552 | ||
1553 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
9e645e11 JA |
1554 | |
1555 | while (!list_empty(&req->link_list)) { | |
4493233e PB |
1556 | struct io_kiocb *link = list_first_entry(&req->link_list, |
1557 | struct io_kiocb, link_list); | |
9e645e11 | 1558 | |
4493233e | 1559 | list_del_init(&link->link_list); |
c826bd7a | 1560 | trace_io_uring_fail_link(req, link); |
2665abfd JA |
1561 | |
1562 | if ((req->flags & REQ_F_LINK_TIMEOUT) && | |
d625c6ee | 1563 | link->opcode == IORING_OP_LINK_TIMEOUT) { |
a197f664 | 1564 | io_link_cancel_timeout(link); |
2665abfd | 1565 | } else { |
78e19bbe | 1566 | io_cqring_fill_event(link, -ECANCELED); |
978db57e | 1567 | __io_double_put_req(link); |
2665abfd | 1568 | } |
5d960724 | 1569 | req->flags &= ~REQ_F_LINK_TIMEOUT; |
9e645e11 | 1570 | } |
2665abfd JA |
1571 | |
1572 | io_commit_cqring(ctx); | |
1573 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
1574 | io_cqring_ev_posted(ctx); | |
9e645e11 JA |
1575 | } |
1576 | ||
4d7dd462 | 1577 | static void io_req_find_next(struct io_kiocb *req, struct io_kiocb **nxt) |
9e645e11 | 1578 | { |
dea3b49c | 1579 | if (likely(!(req->flags & REQ_F_LINK_HEAD))) |
2665abfd | 1580 | return; |
2665abfd | 1581 | |
9e645e11 JA |
1582 | /* |
1583 | * If LINK is set, we have dependent requests in this chain. If we | |
1584 | * didn't fail this request, queue the first one up, moving any other | |
1585 | * dependencies to the next request. In case of failure, fail the rest | |
1586 | * of the chain. | |
1587 | */ | |
2665abfd JA |
1588 | if (req->flags & REQ_F_FAIL_LINK) { |
1589 | io_fail_links(req); | |
7c9e7f0f JA |
1590 | } else if ((req->flags & (REQ_F_LINK_TIMEOUT | REQ_F_COMP_LOCKED)) == |
1591 | REQ_F_LINK_TIMEOUT) { | |
2665abfd JA |
1592 | struct io_ring_ctx *ctx = req->ctx; |
1593 | unsigned long flags; | |
1594 | ||
1595 | /* | |
1596 | * If this is a timeout link, we could be racing with the | |
1597 | * timeout timer. Grab the completion lock for this case to | |
7c9e7f0f | 1598 | * protect against that. |
2665abfd JA |
1599 | */ |
1600 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
1601 | io_req_link_next(req, nxt); | |
1602 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
1603 | } else { | |
1604 | io_req_link_next(req, nxt); | |
9e645e11 | 1605 | } |
4d7dd462 | 1606 | } |
9e645e11 | 1607 | |
c69f8dbe JL |
1608 | static void io_free_req(struct io_kiocb *req) |
1609 | { | |
944e58bf PB |
1610 | struct io_kiocb *nxt = NULL; |
1611 | ||
1612 | io_req_find_next(req, &nxt); | |
70cf9f32 | 1613 | __io_free_req(req); |
944e58bf PB |
1614 | |
1615 | if (nxt) | |
1616 | io_queue_async_work(nxt); | |
c69f8dbe JL |
1617 | } |
1618 | ||
7a743e22 PB |
1619 | static void io_wq_assign_next(struct io_wq_work **workptr, struct io_kiocb *nxt) |
1620 | { | |
1621 | struct io_kiocb *link; | |
8766dd51 PB |
1622 | const struct io_op_def *def = &io_op_defs[nxt->opcode]; |
1623 | ||
1624 | if ((nxt->flags & REQ_F_ISREG) && def->hash_reg_file) | |
1625 | io_wq_hash_work(&nxt->work, file_inode(nxt->file)); | |
7a743e22 PB |
1626 | |
1627 | *workptr = &nxt->work; | |
1628 | link = io_prep_linked_timeout(nxt); | |
18a542ff | 1629 | if (link) |
d4c81f38 | 1630 | nxt->flags |= REQ_F_QUEUE_TIMEOUT; |
7a743e22 PB |
1631 | } |
1632 | ||
ba816ad6 JA |
1633 | /* |
1634 | * Drop reference to request, return next in chain (if there is one) if this | |
1635 | * was the last reference to this request. | |
1636 | */ | |
f9bd67f6 | 1637 | __attribute__((nonnull)) |
ec9c02ad | 1638 | static void io_put_req_find_next(struct io_kiocb *req, struct io_kiocb **nxtptr) |
e65ef56d | 1639 | { |
2a44f467 JA |
1640 | if (refcount_dec_and_test(&req->refs)) { |
1641 | io_req_find_next(req, nxtptr); | |
4d7dd462 | 1642 | __io_free_req(req); |
2a44f467 | 1643 | } |
2b188cc1 JA |
1644 | } |
1645 | ||
e65ef56d JA |
1646 | static void io_put_req(struct io_kiocb *req) |
1647 | { | |
1648 | if (refcount_dec_and_test(&req->refs)) | |
1649 | io_free_req(req); | |
2b188cc1 JA |
1650 | } |
1651 | ||
e9fd9396 PB |
1652 | static void io_steal_work(struct io_kiocb *req, |
1653 | struct io_wq_work **workptr) | |
7a743e22 PB |
1654 | { |
1655 | /* | |
1656 | * It's in an io-wq worker, so there always should be at least | |
1657 | * one reference, which will be dropped in io_put_work() just | |
1658 | * after the current handler returns. | |
1659 | * | |
1660 | * It also means, that if the counter dropped to 1, then there is | |
1661 | * no asynchronous users left, so it's safe to steal the next work. | |
1662 | */ | |
7a743e22 PB |
1663 | if (refcount_read(&req->refs) == 1) { |
1664 | struct io_kiocb *nxt = NULL; | |
1665 | ||
1666 | io_req_find_next(req, &nxt); | |
1667 | if (nxt) | |
1668 | io_wq_assign_next(workptr, nxt); | |
1669 | } | |
1670 | } | |
1671 | ||
978db57e JA |
1672 | /* |
1673 | * Must only be used if we don't need to care about links, usually from | |
1674 | * within the completion handling itself. | |
1675 | */ | |
1676 | static void __io_double_put_req(struct io_kiocb *req) | |
78e19bbe JA |
1677 | { |
1678 | /* drop both submit and complete references */ | |
1679 | if (refcount_sub_and_test(2, &req->refs)) | |
1680 | __io_free_req(req); | |
1681 | } | |
1682 | ||
978db57e JA |
1683 | static void io_double_put_req(struct io_kiocb *req) |
1684 | { | |
1685 | /* drop both submit and complete references */ | |
1686 | if (refcount_sub_and_test(2, &req->refs)) | |
1687 | io_free_req(req); | |
1688 | } | |
1689 | ||
1d7bb1d5 | 1690 | static unsigned io_cqring_events(struct io_ring_ctx *ctx, bool noflush) |
a3a0e43f | 1691 | { |
84f97dc2 JA |
1692 | struct io_rings *rings = ctx->rings; |
1693 | ||
ad3eb2c8 JA |
1694 | if (test_bit(0, &ctx->cq_check_overflow)) { |
1695 | /* | |
1696 | * noflush == true is from the waitqueue handler, just ensure | |
1697 | * we wake up the task, and the next invocation will flush the | |
1698 | * entries. We cannot safely to it from here. | |
1699 | */ | |
1700 | if (noflush && !list_empty(&ctx->cq_overflow_list)) | |
1701 | return -1U; | |
1d7bb1d5 | 1702 | |
ad3eb2c8 JA |
1703 | io_cqring_overflow_flush(ctx, false); |
1704 | } | |
1d7bb1d5 | 1705 | |
a3a0e43f JA |
1706 | /* See comment at the top of this file */ |
1707 | smp_rmb(); | |
ad3eb2c8 | 1708 | return ctx->cached_cq_tail - READ_ONCE(rings->cq.head); |
a3a0e43f JA |
1709 | } |
1710 | ||
fb5ccc98 PB |
1711 | static inline unsigned int io_sqring_entries(struct io_ring_ctx *ctx) |
1712 | { | |
1713 | struct io_rings *rings = ctx->rings; | |
1714 | ||
1715 | /* make sure SQ entry isn't read before tail */ | |
1716 | return smp_load_acquire(&rings->sq.tail) - ctx->cached_sq_head; | |
1717 | } | |
1718 | ||
8237e045 | 1719 | static inline bool io_req_multi_free(struct req_batch *rb, struct io_kiocb *req) |
e94f141b | 1720 | { |
dea3b49c | 1721 | if ((req->flags & REQ_F_LINK_HEAD) || io_is_fallback_req(req)) |
c6ca97b3 | 1722 | return false; |
e94f141b | 1723 | |
9d9e88a2 | 1724 | if (req->file || req->io) |
c6ca97b3 JA |
1725 | rb->need_iter++; |
1726 | ||
1727 | rb->reqs[rb->to_free++] = req; | |
1728 | if (unlikely(rb->to_free == ARRAY_SIZE(rb->reqs))) | |
1729 | io_free_req_many(req->ctx, rb); | |
1730 | return true; | |
e94f141b JA |
1731 | } |
1732 | ||
bcda7baa JA |
1733 | static int io_put_kbuf(struct io_kiocb *req) |
1734 | { | |
4d954c25 | 1735 | struct io_buffer *kbuf; |
bcda7baa JA |
1736 | int cflags; |
1737 | ||
4d954c25 | 1738 | kbuf = (struct io_buffer *) (unsigned long) req->rw.addr; |
bcda7baa JA |
1739 | cflags = kbuf->bid << IORING_CQE_BUFFER_SHIFT; |
1740 | cflags |= IORING_CQE_F_BUFFER; | |
1741 | req->rw.addr = 0; | |
1742 | kfree(kbuf); | |
1743 | return cflags; | |
1744 | } | |
1745 | ||
bbde017a XW |
1746 | static void io_iopoll_queue(struct list_head *again) |
1747 | { | |
1748 | struct io_kiocb *req; | |
1749 | ||
1750 | do { | |
1751 | req = list_first_entry(again, struct io_kiocb, list); | |
1752 | list_del(&req->list); | |
d60b5fbc PB |
1753 | |
1754 | /* shouldn't happen unless io_uring is dying, cancel reqs */ | |
1755 | if (unlikely(!current->mm)) { | |
1756 | io_complete_rw_common(&req->rw.kiocb, -EAGAIN); | |
1757 | io_put_req(req); | |
1758 | continue; | |
1759 | } | |
1760 | ||
bbde017a XW |
1761 | refcount_inc(&req->refs); |
1762 | io_queue_async_work(req); | |
1763 | } while (!list_empty(again)); | |
1764 | } | |
1765 | ||
def596e9 JA |
1766 | /* |
1767 | * Find and free completed poll iocbs | |
1768 | */ | |
1769 | static void io_iopoll_complete(struct io_ring_ctx *ctx, unsigned int *nr_events, | |
1770 | struct list_head *done) | |
1771 | { | |
8237e045 | 1772 | struct req_batch rb; |
def596e9 | 1773 | struct io_kiocb *req; |
bbde017a XW |
1774 | LIST_HEAD(again); |
1775 | ||
1776 | /* order with ->result store in io_complete_rw_iopoll() */ | |
1777 | smp_rmb(); | |
def596e9 | 1778 | |
c6ca97b3 | 1779 | rb.to_free = rb.need_iter = 0; |
def596e9 | 1780 | while (!list_empty(done)) { |
bcda7baa JA |
1781 | int cflags = 0; |
1782 | ||
def596e9 | 1783 | req = list_first_entry(done, struct io_kiocb, list); |
bbde017a XW |
1784 | if (READ_ONCE(req->result) == -EAGAIN) { |
1785 | req->iopoll_completed = 0; | |
1786 | list_move_tail(&req->list, &again); | |
1787 | continue; | |
1788 | } | |
def596e9 JA |
1789 | list_del(&req->list); |
1790 | ||
bcda7baa JA |
1791 | if (req->flags & REQ_F_BUFFER_SELECTED) |
1792 | cflags = io_put_kbuf(req); | |
1793 | ||
1794 | __io_cqring_fill_event(req, req->result, cflags); | |
def596e9 JA |
1795 | (*nr_events)++; |
1796 | ||
8237e045 JA |
1797 | if (refcount_dec_and_test(&req->refs) && |
1798 | !io_req_multi_free(&rb, req)) | |
1799 | io_free_req(req); | |
def596e9 | 1800 | } |
def596e9 | 1801 | |
09bb8394 | 1802 | io_commit_cqring(ctx); |
32b2244a XW |
1803 | if (ctx->flags & IORING_SETUP_SQPOLL) |
1804 | io_cqring_ev_posted(ctx); | |
8237e045 | 1805 | io_free_req_many(ctx, &rb); |
581f9810 | 1806 | |
bbde017a XW |
1807 | if (!list_empty(&again)) |
1808 | io_iopoll_queue(&again); | |
581f9810 BM |
1809 | } |
1810 | ||
def596e9 JA |
1811 | static int io_do_iopoll(struct io_ring_ctx *ctx, unsigned int *nr_events, |
1812 | long min) | |
1813 | { | |
1814 | struct io_kiocb *req, *tmp; | |
1815 | LIST_HEAD(done); | |
1816 | bool spin; | |
1817 | int ret; | |
1818 | ||
1819 | /* | |
1820 | * Only spin for completions if we don't have multiple devices hanging | |
1821 | * off our complete list, and we're under the requested amount. | |
1822 | */ | |
1823 | spin = !ctx->poll_multi_file && *nr_events < min; | |
1824 | ||
1825 | ret = 0; | |
1826 | list_for_each_entry_safe(req, tmp, &ctx->poll_list, list) { | |
9adbd45d | 1827 | struct kiocb *kiocb = &req->rw.kiocb; |
def596e9 JA |
1828 | |
1829 | /* | |
581f9810 BM |
1830 | * Move completed and retryable entries to our local lists. |
1831 | * If we find a request that requires polling, break out | |
1832 | * and complete those lists first, if we have entries there. | |
def596e9 | 1833 | */ |
65a6543d | 1834 | if (READ_ONCE(req->iopoll_completed)) { |
def596e9 JA |
1835 | list_move_tail(&req->list, &done); |
1836 | continue; | |
1837 | } | |
1838 | if (!list_empty(&done)) | |
1839 | break; | |
1840 | ||
1841 | ret = kiocb->ki_filp->f_op->iopoll(kiocb, spin); | |
1842 | if (ret < 0) | |
1843 | break; | |
1844 | ||
1845 | if (ret && spin) | |
1846 | spin = false; | |
1847 | ret = 0; | |
1848 | } | |
1849 | ||
1850 | if (!list_empty(&done)) | |
1851 | io_iopoll_complete(ctx, nr_events, &done); | |
1852 | ||
1853 | return ret; | |
1854 | } | |
1855 | ||
1856 | /* | |
d195a66e | 1857 | * Poll for a minimum of 'min' events. Note that if min == 0 we consider that a |
def596e9 JA |
1858 | * non-spinning poll check - we'll still enter the driver poll loop, but only |
1859 | * as a non-spinning completion check. | |
1860 | */ | |
1861 | static int io_iopoll_getevents(struct io_ring_ctx *ctx, unsigned int *nr_events, | |
1862 | long min) | |
1863 | { | |
08f5439f | 1864 | while (!list_empty(&ctx->poll_list) && !need_resched()) { |
def596e9 JA |
1865 | int ret; |
1866 | ||
1867 | ret = io_do_iopoll(ctx, nr_events, min); | |
1868 | if (ret < 0) | |
1869 | return ret; | |
1870 | if (!min || *nr_events >= min) | |
1871 | return 0; | |
1872 | } | |
1873 | ||
1874 | return 1; | |
1875 | } | |
1876 | ||
1877 | /* | |
1878 | * We can't just wait for polled events to come to us, we have to actively | |
1879 | * find and complete them. | |
1880 | */ | |
1881 | static void io_iopoll_reap_events(struct io_ring_ctx *ctx) | |
1882 | { | |
1883 | if (!(ctx->flags & IORING_SETUP_IOPOLL)) | |
1884 | return; | |
1885 | ||
1886 | mutex_lock(&ctx->uring_lock); | |
1887 | while (!list_empty(&ctx->poll_list)) { | |
1888 | unsigned int nr_events = 0; | |
1889 | ||
1890 | io_iopoll_getevents(ctx, &nr_events, 1); | |
08f5439f JA |
1891 | |
1892 | /* | |
1893 | * Ensure we allow local-to-the-cpu processing to take place, | |
1894 | * in this case we need to ensure that we reap all events. | |
1895 | */ | |
1896 | cond_resched(); | |
def596e9 JA |
1897 | } |
1898 | mutex_unlock(&ctx->uring_lock); | |
1899 | } | |
1900 | ||
c7849be9 XW |
1901 | static int io_iopoll_check(struct io_ring_ctx *ctx, unsigned *nr_events, |
1902 | long min) | |
def596e9 | 1903 | { |
2b2ed975 | 1904 | int iters = 0, ret = 0; |
500f9fba | 1905 | |
c7849be9 XW |
1906 | /* |
1907 | * We disallow the app entering submit/complete with polling, but we | |
1908 | * still need to lock the ring to prevent racing with polled issue | |
1909 | * that got punted to a workqueue. | |
1910 | */ | |
1911 | mutex_lock(&ctx->uring_lock); | |
def596e9 JA |
1912 | do { |
1913 | int tmin = 0; | |
1914 | ||
a3a0e43f JA |
1915 | /* |
1916 | * Don't enter poll loop if we already have events pending. | |
1917 | * If we do, we can potentially be spinning for commands that | |
1918 | * already triggered a CQE (eg in error). | |
1919 | */ | |
1d7bb1d5 | 1920 | if (io_cqring_events(ctx, false)) |
a3a0e43f JA |
1921 | break; |
1922 | ||
500f9fba JA |
1923 | /* |
1924 | * If a submit got punted to a workqueue, we can have the | |
1925 | * application entering polling for a command before it gets | |
1926 | * issued. That app will hold the uring_lock for the duration | |
1927 | * of the poll right here, so we need to take a breather every | |
1928 | * now and then to ensure that the issue has a chance to add | |
1929 | * the poll to the issued list. Otherwise we can spin here | |
1930 | * forever, while the workqueue is stuck trying to acquire the | |
1931 | * very same mutex. | |
1932 | */ | |
1933 | if (!(++iters & 7)) { | |
1934 | mutex_unlock(&ctx->uring_lock); | |
1935 | mutex_lock(&ctx->uring_lock); | |
1936 | } | |
1937 | ||
def596e9 JA |
1938 | if (*nr_events < min) |
1939 | tmin = min - *nr_events; | |
1940 | ||
1941 | ret = io_iopoll_getevents(ctx, nr_events, tmin); | |
1942 | if (ret <= 0) | |
1943 | break; | |
1944 | ret = 0; | |
1945 | } while (min && !*nr_events && !need_resched()); | |
1946 | ||
500f9fba | 1947 | mutex_unlock(&ctx->uring_lock); |
def596e9 JA |
1948 | return ret; |
1949 | } | |
1950 | ||
491381ce | 1951 | static void kiocb_end_write(struct io_kiocb *req) |
2b188cc1 | 1952 | { |
491381ce JA |
1953 | /* |
1954 | * Tell lockdep we inherited freeze protection from submission | |
1955 | * thread. | |
1956 | */ | |
1957 | if (req->flags & REQ_F_ISREG) { | |
1958 | struct inode *inode = file_inode(req->file); | |
2b188cc1 | 1959 | |
491381ce | 1960 | __sb_writers_acquired(inode->i_sb, SB_FREEZE_WRITE); |
2b188cc1 | 1961 | } |
491381ce | 1962 | file_end_write(req->file); |
2b188cc1 JA |
1963 | } |
1964 | ||
4e88d6e7 JA |
1965 | static inline void req_set_fail_links(struct io_kiocb *req) |
1966 | { | |
1967 | if ((req->flags & (REQ_F_LINK | REQ_F_HARDLINK)) == REQ_F_LINK) | |
1968 | req->flags |= REQ_F_FAIL_LINK; | |
1969 | } | |
1970 | ||
ba816ad6 | 1971 | static void io_complete_rw_common(struct kiocb *kiocb, long res) |
2b188cc1 | 1972 | { |
9adbd45d | 1973 | struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb); |
bcda7baa | 1974 | int cflags = 0; |
2b188cc1 | 1975 | |
491381ce JA |
1976 | if (kiocb->ki_flags & IOCB_WRITE) |
1977 | kiocb_end_write(req); | |
2b188cc1 | 1978 | |
4e88d6e7 JA |
1979 | if (res != req->result) |
1980 | req_set_fail_links(req); | |
bcda7baa JA |
1981 | if (req->flags & REQ_F_BUFFER_SELECTED) |
1982 | cflags = io_put_kbuf(req); | |
1983 | __io_cqring_add_event(req, res, cflags); | |
ba816ad6 JA |
1984 | } |
1985 | ||
1986 | static void io_complete_rw(struct kiocb *kiocb, long res, long res2) | |
1987 | { | |
9adbd45d | 1988 | struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb); |
ba816ad6 JA |
1989 | |
1990 | io_complete_rw_common(kiocb, res); | |
e65ef56d | 1991 | io_put_req(req); |
2b188cc1 JA |
1992 | } |
1993 | ||
def596e9 JA |
1994 | static void io_complete_rw_iopoll(struct kiocb *kiocb, long res, long res2) |
1995 | { | |
9adbd45d | 1996 | struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb); |
def596e9 | 1997 | |
491381ce JA |
1998 | if (kiocb->ki_flags & IOCB_WRITE) |
1999 | kiocb_end_write(req); | |
def596e9 | 2000 | |
2d7d6792 | 2001 | if (res != -EAGAIN && res != req->result) |
4e88d6e7 | 2002 | req_set_fail_links(req); |
bbde017a XW |
2003 | |
2004 | WRITE_ONCE(req->result, res); | |
2005 | /* order with io_poll_complete() checking ->result */ | |
cd664b0e PB |
2006 | smp_wmb(); |
2007 | WRITE_ONCE(req->iopoll_completed, 1); | |
def596e9 JA |
2008 | } |
2009 | ||
2010 | /* | |
2011 | * After the iocb has been issued, it's safe to be found on the poll list. | |
2012 | * Adding the kiocb to the list AFTER submission ensures that we don't | |
2013 | * find it from a io_iopoll_getevents() thread before the issuer is done | |
2014 | * accessing the kiocb cookie. | |
2015 | */ | |
2016 | static void io_iopoll_req_issued(struct io_kiocb *req) | |
2017 | { | |
2018 | struct io_ring_ctx *ctx = req->ctx; | |
2019 | ||
2020 | /* | |
2021 | * Track whether we have multiple files in our lists. This will impact | |
2022 | * how we do polling eventually, not spinning if we're on potentially | |
2023 | * different devices. | |
2024 | */ | |
2025 | if (list_empty(&ctx->poll_list)) { | |
2026 | ctx->poll_multi_file = false; | |
2027 | } else if (!ctx->poll_multi_file) { | |
2028 | struct io_kiocb *list_req; | |
2029 | ||
2030 | list_req = list_first_entry(&ctx->poll_list, struct io_kiocb, | |
2031 | list); | |
9adbd45d | 2032 | if (list_req->file != req->file) |
def596e9 JA |
2033 | ctx->poll_multi_file = true; |
2034 | } | |
2035 | ||
2036 | /* | |
2037 | * For fast devices, IO may have already completed. If it has, add | |
2038 | * it to the front so we find it first. | |
2039 | */ | |
65a6543d | 2040 | if (READ_ONCE(req->iopoll_completed)) |
def596e9 JA |
2041 | list_add(&req->list, &ctx->poll_list); |
2042 | else | |
2043 | list_add_tail(&req->list, &ctx->poll_list); | |
bdcd3eab XW |
2044 | |
2045 | if ((ctx->flags & IORING_SETUP_SQPOLL) && | |
2046 | wq_has_sleeper(&ctx->sqo_wait)) | |
2047 | wake_up(&ctx->sqo_wait); | |
def596e9 JA |
2048 | } |
2049 | ||
9f13c35b | 2050 | static void __io_state_file_put(struct io_submit_state *state) |
9a56a232 | 2051 | { |
9f13c35b | 2052 | int diff = state->has_refs - state->used_refs; |
9a56a232 | 2053 | |
9f13c35b PB |
2054 | if (diff) |
2055 | fput_many(state->file, diff); | |
2056 | state->file = NULL; | |
2057 | } | |
2058 | ||
2059 | static inline void io_state_file_put(struct io_submit_state *state) | |
2060 | { | |
2061 | if (state->file) | |
2062 | __io_state_file_put(state); | |
9a56a232 JA |
2063 | } |
2064 | ||
2065 | /* | |
2066 | * Get as many references to a file as we have IOs left in this submission, | |
2067 | * assuming most submissions are for one file, or at least that each file | |
2068 | * has more than one submission. | |
2069 | */ | |
8da11c19 | 2070 | static struct file *__io_file_get(struct io_submit_state *state, int fd) |
9a56a232 JA |
2071 | { |
2072 | if (!state) | |
2073 | return fget(fd); | |
2074 | ||
2075 | if (state->file) { | |
2076 | if (state->fd == fd) { | |
2077 | state->used_refs++; | |
2078 | state->ios_left--; | |
2079 | return state->file; | |
2080 | } | |
9f13c35b | 2081 | __io_state_file_put(state); |
9a56a232 JA |
2082 | } |
2083 | state->file = fget_many(fd, state->ios_left); | |
2084 | if (!state->file) | |
2085 | return NULL; | |
2086 | ||
2087 | state->fd = fd; | |
2088 | state->has_refs = state->ios_left; | |
2089 | state->used_refs = 1; | |
2090 | state->ios_left--; | |
2091 | return state->file; | |
2092 | } | |
2093 | ||
2b188cc1 JA |
2094 | /* |
2095 | * If we tracked the file through the SCM inflight mechanism, we could support | |
2096 | * any file. For now, just ensure that anything potentially problematic is done | |
2097 | * inline. | |
2098 | */ | |
af197f50 | 2099 | static bool io_file_supports_async(struct file *file, int rw) |
2b188cc1 JA |
2100 | { |
2101 | umode_t mode = file_inode(file)->i_mode; | |
2102 | ||
10d59345 | 2103 | if (S_ISBLK(mode) || S_ISCHR(mode) || S_ISSOCK(mode)) |
2b188cc1 JA |
2104 | return true; |
2105 | if (S_ISREG(mode) && file->f_op != &io_uring_fops) | |
2106 | return true; | |
2107 | ||
c5b85625 JA |
2108 | /* any ->read/write should understand O_NONBLOCK */ |
2109 | if (file->f_flags & O_NONBLOCK) | |
2110 | return true; | |
2111 | ||
af197f50 JA |
2112 | if (!(file->f_mode & FMODE_NOWAIT)) |
2113 | return false; | |
2114 | ||
2115 | if (rw == READ) | |
2116 | return file->f_op->read_iter != NULL; | |
2117 | ||
2118 | return file->f_op->write_iter != NULL; | |
2b188cc1 JA |
2119 | } |
2120 | ||
3529d8c2 JA |
2121 | static int io_prep_rw(struct io_kiocb *req, const struct io_uring_sqe *sqe, |
2122 | bool force_nonblock) | |
2b188cc1 | 2123 | { |
def596e9 | 2124 | struct io_ring_ctx *ctx = req->ctx; |
9adbd45d | 2125 | struct kiocb *kiocb = &req->rw.kiocb; |
09bb8394 JA |
2126 | unsigned ioprio; |
2127 | int ret; | |
2b188cc1 | 2128 | |
491381ce JA |
2129 | if (S_ISREG(file_inode(req->file)->i_mode)) |
2130 | req->flags |= REQ_F_ISREG; | |
2131 | ||
2b188cc1 | 2132 | kiocb->ki_pos = READ_ONCE(sqe->off); |
ba04291e JA |
2133 | if (kiocb->ki_pos == -1 && !(req->file->f_mode & FMODE_STREAM)) { |
2134 | req->flags |= REQ_F_CUR_POS; | |
2135 | kiocb->ki_pos = req->file->f_pos; | |
2136 | } | |
2b188cc1 | 2137 | kiocb->ki_hint = ki_hint_validate(file_write_hint(kiocb->ki_filp)); |
3e577dcd PB |
2138 | kiocb->ki_flags = iocb_flags(kiocb->ki_filp); |
2139 | ret = kiocb_set_rw_flags(kiocb, READ_ONCE(sqe->rw_flags)); | |
2140 | if (unlikely(ret)) | |
2141 | return ret; | |
2b188cc1 JA |
2142 | |
2143 | ioprio = READ_ONCE(sqe->ioprio); | |
2144 | if (ioprio) { | |
2145 | ret = ioprio_check_cap(ioprio); | |
2146 | if (ret) | |
09bb8394 | 2147 | return ret; |
2b188cc1 JA |
2148 | |
2149 | kiocb->ki_ioprio = ioprio; | |
2150 | } else | |
2151 | kiocb->ki_ioprio = get_current_ioprio(); | |
2152 | ||
8449eeda | 2153 | /* don't allow async punt if RWF_NOWAIT was requested */ |
c5b85625 | 2154 | if (kiocb->ki_flags & IOCB_NOWAIT) |
8449eeda SB |
2155 | req->flags |= REQ_F_NOWAIT; |
2156 | ||
2157 | if (force_nonblock) | |
2b188cc1 | 2158 | kiocb->ki_flags |= IOCB_NOWAIT; |
8449eeda | 2159 | |
def596e9 | 2160 | if (ctx->flags & IORING_SETUP_IOPOLL) { |
def596e9 JA |
2161 | if (!(kiocb->ki_flags & IOCB_DIRECT) || |
2162 | !kiocb->ki_filp->f_op->iopoll) | |
09bb8394 | 2163 | return -EOPNOTSUPP; |
2b188cc1 | 2164 | |
def596e9 JA |
2165 | kiocb->ki_flags |= IOCB_HIPRI; |
2166 | kiocb->ki_complete = io_complete_rw_iopoll; | |
6873e0bd | 2167 | req->result = 0; |
65a6543d | 2168 | req->iopoll_completed = 0; |
def596e9 | 2169 | } else { |
09bb8394 JA |
2170 | if (kiocb->ki_flags & IOCB_HIPRI) |
2171 | return -EINVAL; | |
def596e9 JA |
2172 | kiocb->ki_complete = io_complete_rw; |
2173 | } | |
9adbd45d | 2174 | |
3529d8c2 JA |
2175 | req->rw.addr = READ_ONCE(sqe->addr); |
2176 | req->rw.len = READ_ONCE(sqe->len); | |
4f4eeba8 | 2177 | req->buf_index = READ_ONCE(sqe->buf_index); |
2b188cc1 | 2178 | return 0; |
2b188cc1 JA |
2179 | } |
2180 | ||
2181 | static inline void io_rw_done(struct kiocb *kiocb, ssize_t ret) | |
2182 | { | |
2183 | switch (ret) { | |
2184 | case -EIOCBQUEUED: | |
2185 | break; | |
2186 | case -ERESTARTSYS: | |
2187 | case -ERESTARTNOINTR: | |
2188 | case -ERESTARTNOHAND: | |
2189 | case -ERESTART_RESTARTBLOCK: | |
2190 | /* | |
2191 | * We can't just restart the syscall, since previously | |
2192 | * submitted sqes may already be in progress. Just fail this | |
2193 | * IO with EINTR. | |
2194 | */ | |
2195 | ret = -EINTR; | |
2196 | /* fall through */ | |
2197 | default: | |
2198 | kiocb->ki_complete(kiocb, ret, 0); | |
2199 | } | |
2200 | } | |
2201 | ||
014db007 | 2202 | static void kiocb_done(struct kiocb *kiocb, ssize_t ret) |
ba816ad6 | 2203 | { |
ba04291e JA |
2204 | struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb); |
2205 | ||
2206 | if (req->flags & REQ_F_CUR_POS) | |
2207 | req->file->f_pos = kiocb->ki_pos; | |
bcaec089 | 2208 | if (ret >= 0 && kiocb->ki_complete == io_complete_rw) |
014db007 | 2209 | io_complete_rw(kiocb, ret, 0); |
ba816ad6 JA |
2210 | else |
2211 | io_rw_done(kiocb, ret); | |
2212 | } | |
2213 | ||
9adbd45d | 2214 | static ssize_t io_import_fixed(struct io_kiocb *req, int rw, |
7d009165 | 2215 | struct iov_iter *iter) |
edafccee | 2216 | { |
9adbd45d JA |
2217 | struct io_ring_ctx *ctx = req->ctx; |
2218 | size_t len = req->rw.len; | |
edafccee | 2219 | struct io_mapped_ubuf *imu; |
4f4eeba8 | 2220 | u16 index, buf_index; |
edafccee JA |
2221 | size_t offset; |
2222 | u64 buf_addr; | |
2223 | ||
2224 | /* attempt to use fixed buffers without having provided iovecs */ | |
2225 | if (unlikely(!ctx->user_bufs)) | |
2226 | return -EFAULT; | |
2227 | ||
4f4eeba8 | 2228 | buf_index = req->buf_index; |
edafccee JA |
2229 | if (unlikely(buf_index >= ctx->nr_user_bufs)) |
2230 | return -EFAULT; | |
2231 | ||
2232 | index = array_index_nospec(buf_index, ctx->nr_user_bufs); | |
2233 | imu = &ctx->user_bufs[index]; | |
9adbd45d | 2234 | buf_addr = req->rw.addr; |
edafccee JA |
2235 | |
2236 | /* overflow */ | |
2237 | if (buf_addr + len < buf_addr) | |
2238 | return -EFAULT; | |
2239 | /* not inside the mapped region */ | |
2240 | if (buf_addr < imu->ubuf || buf_addr + len > imu->ubuf + imu->len) | |
2241 | return -EFAULT; | |
2242 | ||
2243 | /* | |
2244 | * May not be a start of buffer, set size appropriately | |
2245 | * and advance us to the beginning. | |
2246 | */ | |
2247 | offset = buf_addr - imu->ubuf; | |
2248 | iov_iter_bvec(iter, rw, imu->bvec, imu->nr_bvecs, offset + len); | |
bd11b3a3 JA |
2249 | |
2250 | if (offset) { | |
2251 | /* | |
2252 | * Don't use iov_iter_advance() here, as it's really slow for | |
2253 | * using the latter parts of a big fixed buffer - it iterates | |
2254 | * over each segment manually. We can cheat a bit here, because | |
2255 | * we know that: | |
2256 | * | |
2257 | * 1) it's a BVEC iter, we set it up | |
2258 | * 2) all bvecs are PAGE_SIZE in size, except potentially the | |
2259 | * first and last bvec | |
2260 | * | |
2261 | * So just find our index, and adjust the iterator afterwards. | |
2262 | * If the offset is within the first bvec (or the whole first | |
2263 | * bvec, just use iov_iter_advance(). This makes it easier | |
2264 | * since we can just skip the first segment, which may not | |
2265 | * be PAGE_SIZE aligned. | |
2266 | */ | |
2267 | const struct bio_vec *bvec = imu->bvec; | |
2268 | ||
2269 | if (offset <= bvec->bv_len) { | |
2270 | iov_iter_advance(iter, offset); | |
2271 | } else { | |
2272 | unsigned long seg_skip; | |
2273 | ||
2274 | /* skip first vec */ | |
2275 | offset -= bvec->bv_len; | |
2276 | seg_skip = 1 + (offset >> PAGE_SHIFT); | |
2277 | ||
2278 | iter->bvec = bvec + seg_skip; | |
2279 | iter->nr_segs -= seg_skip; | |
99c79f66 | 2280 | iter->count -= bvec->bv_len + offset; |
bd11b3a3 | 2281 | iter->iov_offset = offset & ~PAGE_MASK; |
bd11b3a3 JA |
2282 | } |
2283 | } | |
2284 | ||
5e559561 | 2285 | return len; |
edafccee JA |
2286 | } |
2287 | ||
bcda7baa JA |
2288 | static void io_ring_submit_unlock(struct io_ring_ctx *ctx, bool needs_lock) |
2289 | { | |
2290 | if (needs_lock) | |
2291 | mutex_unlock(&ctx->uring_lock); | |
2292 | } | |
2293 | ||
2294 | static void io_ring_submit_lock(struct io_ring_ctx *ctx, bool needs_lock) | |
2295 | { | |
2296 | /* | |
2297 | * "Normal" inline submissions always hold the uring_lock, since we | |
2298 | * grab it from the system call. Same is true for the SQPOLL offload. | |
2299 | * The only exception is when we've detached the request and issue it | |
2300 | * from an async worker thread, grab the lock for that case. | |
2301 | */ | |
2302 | if (needs_lock) | |
2303 | mutex_lock(&ctx->uring_lock); | |
2304 | } | |
2305 | ||
2306 | static struct io_buffer *io_buffer_select(struct io_kiocb *req, size_t *len, | |
2307 | int bgid, struct io_buffer *kbuf, | |
2308 | bool needs_lock) | |
2309 | { | |
2310 | struct io_buffer *head; | |
2311 | ||
2312 | if (req->flags & REQ_F_BUFFER_SELECTED) | |
2313 | return kbuf; | |
2314 | ||
2315 | io_ring_submit_lock(req->ctx, needs_lock); | |
2316 | ||
2317 | lockdep_assert_held(&req->ctx->uring_lock); | |
2318 | ||
2319 | head = idr_find(&req->ctx->io_buffer_idr, bgid); | |
2320 | if (head) { | |
2321 | if (!list_empty(&head->list)) { | |
2322 | kbuf = list_last_entry(&head->list, struct io_buffer, | |
2323 | list); | |
2324 | list_del(&kbuf->list); | |
2325 | } else { | |
2326 | kbuf = head; | |
2327 | idr_remove(&req->ctx->io_buffer_idr, bgid); | |
2328 | } | |
2329 | if (*len > kbuf->len) | |
2330 | *len = kbuf->len; | |
2331 | } else { | |
2332 | kbuf = ERR_PTR(-ENOBUFS); | |
2333 | } | |
2334 | ||
2335 | io_ring_submit_unlock(req->ctx, needs_lock); | |
2336 | ||
2337 | return kbuf; | |
2338 | } | |
2339 | ||
4d954c25 JA |
2340 | static void __user *io_rw_buffer_select(struct io_kiocb *req, size_t *len, |
2341 | bool needs_lock) | |
2342 | { | |
2343 | struct io_buffer *kbuf; | |
4f4eeba8 | 2344 | u16 bgid; |
4d954c25 JA |
2345 | |
2346 | kbuf = (struct io_buffer *) (unsigned long) req->rw.addr; | |
4f4eeba8 | 2347 | bgid = req->buf_index; |
4d954c25 JA |
2348 | kbuf = io_buffer_select(req, len, bgid, kbuf, needs_lock); |
2349 | if (IS_ERR(kbuf)) | |
2350 | return kbuf; | |
2351 | req->rw.addr = (u64) (unsigned long) kbuf; | |
2352 | req->flags |= REQ_F_BUFFER_SELECTED; | |
2353 | return u64_to_user_ptr(kbuf->addr); | |
2354 | } | |
2355 | ||
2356 | #ifdef CONFIG_COMPAT | |
2357 | static ssize_t io_compat_import(struct io_kiocb *req, struct iovec *iov, | |
2358 | bool needs_lock) | |
2359 | { | |
2360 | struct compat_iovec __user *uiov; | |
2361 | compat_ssize_t clen; | |
2362 | void __user *buf; | |
2363 | ssize_t len; | |
2364 | ||
2365 | uiov = u64_to_user_ptr(req->rw.addr); | |
2366 | if (!access_ok(uiov, sizeof(*uiov))) | |
2367 | return -EFAULT; | |
2368 | if (__get_user(clen, &uiov->iov_len)) | |
2369 | return -EFAULT; | |
2370 | if (clen < 0) | |
2371 | return -EINVAL; | |
2372 | ||
2373 | len = clen; | |
2374 | buf = io_rw_buffer_select(req, &len, needs_lock); | |
2375 | if (IS_ERR(buf)) | |
2376 | return PTR_ERR(buf); | |
2377 | iov[0].iov_base = buf; | |
2378 | iov[0].iov_len = (compat_size_t) len; | |
2379 | return 0; | |
2380 | } | |
2381 | #endif | |
2382 | ||
2383 | static ssize_t __io_iov_buffer_select(struct io_kiocb *req, struct iovec *iov, | |
2384 | bool needs_lock) | |
2385 | { | |
2386 | struct iovec __user *uiov = u64_to_user_ptr(req->rw.addr); | |
2387 | void __user *buf; | |
2388 | ssize_t len; | |
2389 | ||
2390 | if (copy_from_user(iov, uiov, sizeof(*uiov))) | |
2391 | return -EFAULT; | |
2392 | ||
2393 | len = iov[0].iov_len; | |
2394 | if (len < 0) | |
2395 | return -EINVAL; | |
2396 | buf = io_rw_buffer_select(req, &len, needs_lock); | |
2397 | if (IS_ERR(buf)) | |
2398 | return PTR_ERR(buf); | |
2399 | iov[0].iov_base = buf; | |
2400 | iov[0].iov_len = len; | |
2401 | return 0; | |
2402 | } | |
2403 | ||
2404 | static ssize_t io_iov_buffer_select(struct io_kiocb *req, struct iovec *iov, | |
2405 | bool needs_lock) | |
2406 | { | |
dddb3e26 JA |
2407 | if (req->flags & REQ_F_BUFFER_SELECTED) { |
2408 | struct io_buffer *kbuf; | |
2409 | ||
2410 | kbuf = (struct io_buffer *) (unsigned long) req->rw.addr; | |
2411 | iov[0].iov_base = u64_to_user_ptr(kbuf->addr); | |
2412 | iov[0].iov_len = kbuf->len; | |
4d954c25 | 2413 | return 0; |
dddb3e26 | 2414 | } |
4d954c25 JA |
2415 | if (!req->rw.len) |
2416 | return 0; | |
2417 | else if (req->rw.len > 1) | |
2418 | return -EINVAL; | |
2419 | ||
2420 | #ifdef CONFIG_COMPAT | |
2421 | if (req->ctx->compat) | |
2422 | return io_compat_import(req, iov, needs_lock); | |
2423 | #endif | |
2424 | ||
2425 | return __io_iov_buffer_select(req, iov, needs_lock); | |
2426 | } | |
2427 | ||
cf6fd4bd | 2428 | static ssize_t io_import_iovec(int rw, struct io_kiocb *req, |
bcda7baa JA |
2429 | struct iovec **iovec, struct iov_iter *iter, |
2430 | bool needs_lock) | |
2b188cc1 | 2431 | { |
9adbd45d JA |
2432 | void __user *buf = u64_to_user_ptr(req->rw.addr); |
2433 | size_t sqe_len = req->rw.len; | |
4d954c25 | 2434 | ssize_t ret; |
edafccee JA |
2435 | u8 opcode; |
2436 | ||
d625c6ee | 2437 | opcode = req->opcode; |
7d009165 | 2438 | if (opcode == IORING_OP_READ_FIXED || opcode == IORING_OP_WRITE_FIXED) { |
edafccee | 2439 | *iovec = NULL; |
9adbd45d | 2440 | return io_import_fixed(req, rw, iter); |
edafccee | 2441 | } |
2b188cc1 | 2442 | |
bcda7baa | 2443 | /* buffer index only valid with fixed read/write, or buffer select */ |
4f4eeba8 | 2444 | if (req->buf_index && !(req->flags & REQ_F_BUFFER_SELECT)) |
9adbd45d JA |
2445 | return -EINVAL; |
2446 | ||
3a6820f2 | 2447 | if (opcode == IORING_OP_READ || opcode == IORING_OP_WRITE) { |
bcda7baa | 2448 | if (req->flags & REQ_F_BUFFER_SELECT) { |
4d954c25 JA |
2449 | buf = io_rw_buffer_select(req, &sqe_len, needs_lock); |
2450 | if (IS_ERR(buf)) { | |
bcda7baa | 2451 | *iovec = NULL; |
4d954c25 | 2452 | return PTR_ERR(buf); |
bcda7baa | 2453 | } |
3f9d6441 | 2454 | req->rw.len = sqe_len; |
bcda7baa JA |
2455 | } |
2456 | ||
3a6820f2 JA |
2457 | ret = import_single_range(rw, buf, sqe_len, *iovec, iter); |
2458 | *iovec = NULL; | |
3a901598 | 2459 | return ret < 0 ? ret : sqe_len; |
3a6820f2 JA |
2460 | } |
2461 | ||
f67676d1 JA |
2462 | if (req->io) { |
2463 | struct io_async_rw *iorw = &req->io->rw; | |
2464 | ||
2465 | *iovec = iorw->iov; | |
2466 | iov_iter_init(iter, rw, *iovec, iorw->nr_segs, iorw->size); | |
2467 | if (iorw->iov == iorw->fast_iov) | |
2468 | *iovec = NULL; | |
2469 | return iorw->size; | |
2470 | } | |
2471 | ||
4d954c25 JA |
2472 | if (req->flags & REQ_F_BUFFER_SELECT) { |
2473 | ret = io_iov_buffer_select(req, *iovec, needs_lock); | |
3f9d6441 JA |
2474 | if (!ret) { |
2475 | ret = (*iovec)->iov_len; | |
2476 | iov_iter_init(iter, rw, *iovec, 1, ret); | |
2477 | } | |
4d954c25 JA |
2478 | *iovec = NULL; |
2479 | return ret; | |
2480 | } | |
2481 | ||
2b188cc1 | 2482 | #ifdef CONFIG_COMPAT |
cf6fd4bd | 2483 | if (req->ctx->compat) |
2b188cc1 JA |
2484 | return compat_import_iovec(rw, buf, sqe_len, UIO_FASTIOV, |
2485 | iovec, iter); | |
2486 | #endif | |
2487 | ||
2488 | return import_iovec(rw, buf, sqe_len, UIO_FASTIOV, iovec, iter); | |
2489 | } | |
2490 | ||
31b51510 | 2491 | /* |
32960613 JA |
2492 | * For files that don't have ->read_iter() and ->write_iter(), handle them |
2493 | * by looping over ->read() or ->write() manually. | |
31b51510 | 2494 | */ |
32960613 JA |
2495 | static ssize_t loop_rw_iter(int rw, struct file *file, struct kiocb *kiocb, |
2496 | struct iov_iter *iter) | |
2497 | { | |
2498 | ssize_t ret = 0; | |
2499 | ||
2500 | /* | |
2501 | * Don't support polled IO through this interface, and we can't | |
2502 | * support non-blocking either. For the latter, this just causes | |
2503 | * the kiocb to be handled from an async context. | |
2504 | */ | |
2505 | if (kiocb->ki_flags & IOCB_HIPRI) | |
2506 | return -EOPNOTSUPP; | |
2507 | if (kiocb->ki_flags & IOCB_NOWAIT) | |
2508 | return -EAGAIN; | |
2509 | ||
2510 | while (iov_iter_count(iter)) { | |
311ae9e1 | 2511 | struct iovec iovec; |
32960613 JA |
2512 | ssize_t nr; |
2513 | ||
311ae9e1 PB |
2514 | if (!iov_iter_is_bvec(iter)) { |
2515 | iovec = iov_iter_iovec(iter); | |
2516 | } else { | |
2517 | /* fixed buffers import bvec */ | |
2518 | iovec.iov_base = kmap(iter->bvec->bv_page) | |
2519 | + iter->iov_offset; | |
2520 | iovec.iov_len = min(iter->count, | |
2521 | iter->bvec->bv_len - iter->iov_offset); | |
2522 | } | |
2523 | ||
32960613 JA |
2524 | if (rw == READ) { |
2525 | nr = file->f_op->read(file, iovec.iov_base, | |
2526 | iovec.iov_len, &kiocb->ki_pos); | |
2527 | } else { | |
2528 | nr = file->f_op->write(file, iovec.iov_base, | |
2529 | iovec.iov_len, &kiocb->ki_pos); | |
2530 | } | |
2531 | ||
311ae9e1 PB |
2532 | if (iov_iter_is_bvec(iter)) |
2533 | kunmap(iter->bvec->bv_page); | |
2534 | ||
32960613 JA |
2535 | if (nr < 0) { |
2536 | if (!ret) | |
2537 | ret = nr; | |
2538 | break; | |
2539 | } | |
2540 | ret += nr; | |
2541 | if (nr != iovec.iov_len) | |
2542 | break; | |
2543 | iov_iter_advance(iter, nr); | |
2544 | } | |
2545 | ||
2546 | return ret; | |
2547 | } | |
2548 | ||
b7bb4f7d | 2549 | static void io_req_map_rw(struct io_kiocb *req, ssize_t io_size, |
f67676d1 JA |
2550 | struct iovec *iovec, struct iovec *fast_iov, |
2551 | struct iov_iter *iter) | |
2552 | { | |
2553 | req->io->rw.nr_segs = iter->nr_segs; | |
2554 | req->io->rw.size = io_size; | |
2555 | req->io->rw.iov = iovec; | |
2556 | if (!req->io->rw.iov) { | |
2557 | req->io->rw.iov = req->io->rw.fast_iov; | |
45097dae XW |
2558 | if (req->io->rw.iov != fast_iov) |
2559 | memcpy(req->io->rw.iov, fast_iov, | |
2560 | sizeof(struct iovec) * iter->nr_segs); | |
99bc4c38 PB |
2561 | } else { |
2562 | req->flags |= REQ_F_NEED_CLEANUP; | |
f67676d1 JA |
2563 | } |
2564 | } | |
2565 | ||
3d9932a8 XW |
2566 | static inline int __io_alloc_async_ctx(struct io_kiocb *req) |
2567 | { | |
2568 | req->io = kmalloc(sizeof(*req->io), GFP_KERNEL); | |
2569 | return req->io == NULL; | |
2570 | } | |
2571 | ||
b7bb4f7d | 2572 | static int io_alloc_async_ctx(struct io_kiocb *req) |
f67676d1 | 2573 | { |
d3656344 JA |
2574 | if (!io_op_defs[req->opcode].async_ctx) |
2575 | return 0; | |
3d9932a8 XW |
2576 | |
2577 | return __io_alloc_async_ctx(req); | |
b7bb4f7d JA |
2578 | } |
2579 | ||
b7bb4f7d JA |
2580 | static int io_setup_async_rw(struct io_kiocb *req, ssize_t io_size, |
2581 | struct iovec *iovec, struct iovec *fast_iov, | |
2582 | struct iov_iter *iter) | |
2583 | { | |
980ad263 | 2584 | if (!io_op_defs[req->opcode].async_ctx) |
74566df3 | 2585 | return 0; |
5d204bcf | 2586 | if (!req->io) { |
3d9932a8 | 2587 | if (__io_alloc_async_ctx(req)) |
5d204bcf | 2588 | return -ENOMEM; |
b7bb4f7d | 2589 | |
5d204bcf JA |
2590 | io_req_map_rw(req, io_size, iovec, fast_iov, iter); |
2591 | } | |
b7bb4f7d | 2592 | return 0; |
f67676d1 JA |
2593 | } |
2594 | ||
3529d8c2 JA |
2595 | static int io_read_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe, |
2596 | bool force_nonblock) | |
f67676d1 | 2597 | { |
3529d8c2 JA |
2598 | struct io_async_ctx *io; |
2599 | struct iov_iter iter; | |
f67676d1 JA |
2600 | ssize_t ret; |
2601 | ||
3529d8c2 JA |
2602 | ret = io_prep_rw(req, sqe, force_nonblock); |
2603 | if (ret) | |
2604 | return ret; | |
f67676d1 | 2605 | |
3529d8c2 JA |
2606 | if (unlikely(!(req->file->f_mode & FMODE_READ))) |
2607 | return -EBADF; | |
f67676d1 | 2608 | |
5f798bea PB |
2609 | /* either don't need iovec imported or already have it */ |
2610 | if (!req->io || req->flags & REQ_F_NEED_CLEANUP) | |
3529d8c2 JA |
2611 | return 0; |
2612 | ||
2613 | io = req->io; | |
2614 | io->rw.iov = io->rw.fast_iov; | |
2615 | req->io = NULL; | |
bcda7baa | 2616 | ret = io_import_iovec(READ, req, &io->rw.iov, &iter, !force_nonblock); |
3529d8c2 JA |
2617 | req->io = io; |
2618 | if (ret < 0) | |
2619 | return ret; | |
2620 | ||
2621 | io_req_map_rw(req, ret, io->rw.iov, io->rw.fast_iov, &iter); | |
2622 | return 0; | |
f67676d1 JA |
2623 | } |
2624 | ||
014db007 | 2625 | static int io_read(struct io_kiocb *req, bool force_nonblock) |
2b188cc1 JA |
2626 | { |
2627 | struct iovec inline_vecs[UIO_FASTIOV], *iovec = inline_vecs; | |
9adbd45d | 2628 | struct kiocb *kiocb = &req->rw.kiocb; |
2b188cc1 | 2629 | struct iov_iter iter; |
31b51510 | 2630 | size_t iov_count; |
f67676d1 | 2631 | ssize_t io_size, ret; |
2b188cc1 | 2632 | |
bcda7baa | 2633 | ret = io_import_iovec(READ, req, &iovec, &iter, !force_nonblock); |
06b76d44 JA |
2634 | if (ret < 0) |
2635 | return ret; | |
2b188cc1 | 2636 | |
fd6c2e4c JA |
2637 | /* Ensure we clear previously set non-block flag */ |
2638 | if (!force_nonblock) | |
29de5f6a | 2639 | kiocb->ki_flags &= ~IOCB_NOWAIT; |
fd6c2e4c | 2640 | |
797f3f53 | 2641 | req->result = 0; |
f67676d1 | 2642 | io_size = ret; |
dea3b49c | 2643 | if (req->flags & REQ_F_LINK_HEAD) |
f67676d1 JA |
2644 | req->result = io_size; |
2645 | ||
2646 | /* | |
2647 | * If the file doesn't support async, mark it as REQ_F_MUST_PUNT so | |
2648 | * we know to async punt it even if it was opened O_NONBLOCK | |
2649 | */ | |
af197f50 | 2650 | if (force_nonblock && !io_file_supports_async(req->file, READ)) |
f67676d1 | 2651 | goto copy_iov; |
9e645e11 | 2652 | |
31b51510 | 2653 | iov_count = iov_iter_count(&iter); |
9adbd45d | 2654 | ret = rw_verify_area(READ, req->file, &kiocb->ki_pos, iov_count); |
2b188cc1 JA |
2655 | if (!ret) { |
2656 | ssize_t ret2; | |
2657 | ||
9adbd45d JA |
2658 | if (req->file->f_op->read_iter) |
2659 | ret2 = call_read_iter(req->file, kiocb, &iter); | |
32960613 | 2660 | else |
9adbd45d | 2661 | ret2 = loop_rw_iter(READ, req->file, kiocb, &iter); |
32960613 | 2662 | |
9d93a3f5 | 2663 | /* Catch -EAGAIN return for forced non-blocking submission */ |
f67676d1 | 2664 | if (!force_nonblock || ret2 != -EAGAIN) { |
014db007 | 2665 | kiocb_done(kiocb, ret2); |
f67676d1 JA |
2666 | } else { |
2667 | copy_iov: | |
b7bb4f7d | 2668 | ret = io_setup_async_rw(req, io_size, iovec, |
f67676d1 JA |
2669 | inline_vecs, &iter); |
2670 | if (ret) | |
2671 | goto out_free; | |
29de5f6a | 2672 | /* any defer here is final, must blocking retry */ |
490e8967 JA |
2673 | if (!(req->flags & REQ_F_NOWAIT) && |
2674 | !file_can_poll(req->file)) | |
29de5f6a | 2675 | req->flags |= REQ_F_MUST_PUNT; |
f67676d1 JA |
2676 | return -EAGAIN; |
2677 | } | |
2b188cc1 | 2678 | } |
f67676d1 | 2679 | out_free: |
6f2cc166 XW |
2680 | if (!(req->flags & REQ_F_NEED_CLEANUP)) |
2681 | kfree(iovec); | |
2b188cc1 JA |
2682 | return ret; |
2683 | } | |
2684 | ||
3529d8c2 JA |
2685 | static int io_write_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe, |
2686 | bool force_nonblock) | |
f67676d1 | 2687 | { |
3529d8c2 JA |
2688 | struct io_async_ctx *io; |
2689 | struct iov_iter iter; | |
f67676d1 JA |
2690 | ssize_t ret; |
2691 | ||
3529d8c2 JA |
2692 | ret = io_prep_rw(req, sqe, force_nonblock); |
2693 | if (ret) | |
2694 | return ret; | |
f67676d1 | 2695 | |
3529d8c2 JA |
2696 | if (unlikely(!(req->file->f_mode & FMODE_WRITE))) |
2697 | return -EBADF; | |
f67676d1 | 2698 | |
4ed734b0 JA |
2699 | req->fsize = rlimit(RLIMIT_FSIZE); |
2700 | ||
5f798bea PB |
2701 | /* either don't need iovec imported or already have it */ |
2702 | if (!req->io || req->flags & REQ_F_NEED_CLEANUP) | |
3529d8c2 JA |
2703 | return 0; |
2704 | ||
2705 | io = req->io; | |
2706 | io->rw.iov = io->rw.fast_iov; | |
2707 | req->io = NULL; | |
bcda7baa | 2708 | ret = io_import_iovec(WRITE, req, &io->rw.iov, &iter, !force_nonblock); |
3529d8c2 JA |
2709 | req->io = io; |
2710 | if (ret < 0) | |
2711 | return ret; | |
2712 | ||
2713 | io_req_map_rw(req, ret, io->rw.iov, io->rw.fast_iov, &iter); | |
2714 | return 0; | |
f67676d1 JA |
2715 | } |
2716 | ||
014db007 | 2717 | static int io_write(struct io_kiocb *req, bool force_nonblock) |
2b188cc1 JA |
2718 | { |
2719 | struct iovec inline_vecs[UIO_FASTIOV], *iovec = inline_vecs; | |
9adbd45d | 2720 | struct kiocb *kiocb = &req->rw.kiocb; |
2b188cc1 | 2721 | struct iov_iter iter; |
31b51510 | 2722 | size_t iov_count; |
f67676d1 | 2723 | ssize_t ret, io_size; |
2b188cc1 | 2724 | |
bcda7baa | 2725 | ret = io_import_iovec(WRITE, req, &iovec, &iter, !force_nonblock); |
06b76d44 JA |
2726 | if (ret < 0) |
2727 | return ret; | |
2b188cc1 | 2728 | |
fd6c2e4c JA |
2729 | /* Ensure we clear previously set non-block flag */ |
2730 | if (!force_nonblock) | |
9adbd45d | 2731 | req->rw.kiocb.ki_flags &= ~IOCB_NOWAIT; |
fd6c2e4c | 2732 | |
797f3f53 | 2733 | req->result = 0; |
f67676d1 | 2734 | io_size = ret; |
dea3b49c | 2735 | if (req->flags & REQ_F_LINK_HEAD) |
f67676d1 | 2736 | req->result = io_size; |
9e645e11 | 2737 | |
f67676d1 JA |
2738 | /* |
2739 | * If the file doesn't support async, mark it as REQ_F_MUST_PUNT so | |
2740 | * we know to async punt it even if it was opened O_NONBLOCK | |
2741 | */ | |
af197f50 | 2742 | if (force_nonblock && !io_file_supports_async(req->file, WRITE)) |
f67676d1 | 2743 | goto copy_iov; |
31b51510 | 2744 | |
10d59345 JA |
2745 | /* file path doesn't support NOWAIT for non-direct_IO */ |
2746 | if (force_nonblock && !(kiocb->ki_flags & IOCB_DIRECT) && | |
2747 | (req->flags & REQ_F_ISREG)) | |
f67676d1 | 2748 | goto copy_iov; |
31b51510 | 2749 | |
f67676d1 | 2750 | iov_count = iov_iter_count(&iter); |
9adbd45d | 2751 | ret = rw_verify_area(WRITE, req->file, &kiocb->ki_pos, iov_count); |
2b188cc1 | 2752 | if (!ret) { |
9bf7933f RP |
2753 | ssize_t ret2; |
2754 | ||
2b188cc1 JA |
2755 | /* |
2756 | * Open-code file_start_write here to grab freeze protection, | |
2757 | * which will be released by another thread in | |
2758 | * io_complete_rw(). Fool lockdep by telling it the lock got | |
2759 | * released so that it doesn't complain about the held lock when | |
2760 | * we return to userspace. | |
2761 | */ | |
491381ce | 2762 | if (req->flags & REQ_F_ISREG) { |
9adbd45d | 2763 | __sb_start_write(file_inode(req->file)->i_sb, |
2b188cc1 | 2764 | SB_FREEZE_WRITE, true); |
9adbd45d | 2765 | __sb_writers_release(file_inode(req->file)->i_sb, |
2b188cc1 JA |
2766 | SB_FREEZE_WRITE); |
2767 | } | |
2768 | kiocb->ki_flags |= IOCB_WRITE; | |
9bf7933f | 2769 | |
4ed734b0 JA |
2770 | if (!force_nonblock) |
2771 | current->signal->rlim[RLIMIT_FSIZE].rlim_cur = req->fsize; | |
2772 | ||
9adbd45d JA |
2773 | if (req->file->f_op->write_iter) |
2774 | ret2 = call_write_iter(req->file, kiocb, &iter); | |
32960613 | 2775 | else |
9adbd45d | 2776 | ret2 = loop_rw_iter(WRITE, req->file, kiocb, &iter); |
4ed734b0 JA |
2777 | |
2778 | if (!force_nonblock) | |
2779 | current->signal->rlim[RLIMIT_FSIZE].rlim_cur = RLIM_INFINITY; | |
2780 | ||
faac996c | 2781 | /* |
bff6035d | 2782 | * Raw bdev writes will return -EOPNOTSUPP for IOCB_NOWAIT. Just |
faac996c JA |
2783 | * retry them without IOCB_NOWAIT. |
2784 | */ | |
2785 | if (ret2 == -EOPNOTSUPP && (kiocb->ki_flags & IOCB_NOWAIT)) | |
2786 | ret2 = -EAGAIN; | |
f67676d1 | 2787 | if (!force_nonblock || ret2 != -EAGAIN) { |
014db007 | 2788 | kiocb_done(kiocb, ret2); |
f67676d1 JA |
2789 | } else { |
2790 | copy_iov: | |
b7bb4f7d | 2791 | ret = io_setup_async_rw(req, io_size, iovec, |
f67676d1 JA |
2792 | inline_vecs, &iter); |
2793 | if (ret) | |
2794 | goto out_free; | |
29de5f6a | 2795 | /* any defer here is final, must blocking retry */ |
c5b85625 JA |
2796 | if (!(req->flags & REQ_F_NOWAIT) && |
2797 | !file_can_poll(req->file)) | |
490e8967 | 2798 | req->flags |= REQ_F_MUST_PUNT; |
f67676d1 JA |
2799 | return -EAGAIN; |
2800 | } | |
2b188cc1 | 2801 | } |
31b51510 | 2802 | out_free: |
6f2cc166 XW |
2803 | if (!(req->flags & REQ_F_NEED_CLEANUP)) |
2804 | kfree(iovec); | |
2b188cc1 JA |
2805 | return ret; |
2806 | } | |
2807 | ||
f2a8d5c7 PB |
2808 | static int __io_splice_prep(struct io_kiocb *req, |
2809 | const struct io_uring_sqe *sqe) | |
7d67af2c PB |
2810 | { |
2811 | struct io_splice* sp = &req->splice; | |
2812 | unsigned int valid_flags = SPLICE_F_FD_IN_FIXED | SPLICE_F_ALL; | |
2813 | int ret; | |
2814 | ||
2815 | if (req->flags & REQ_F_NEED_CLEANUP) | |
2816 | return 0; | |
3232dd02 PB |
2817 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
2818 | return -EINVAL; | |
7d67af2c PB |
2819 | |
2820 | sp->file_in = NULL; | |
7d67af2c PB |
2821 | sp->len = READ_ONCE(sqe->len); |
2822 | sp->flags = READ_ONCE(sqe->splice_flags); | |
2823 | ||
2824 | if (unlikely(sp->flags & ~valid_flags)) | |
2825 | return -EINVAL; | |
2826 | ||
2827 | ret = io_file_get(NULL, req, READ_ONCE(sqe->splice_fd_in), &sp->file_in, | |
2828 | (sp->flags & SPLICE_F_FD_IN_FIXED)); | |
2829 | if (ret) | |
2830 | return ret; | |
2831 | req->flags |= REQ_F_NEED_CLEANUP; | |
2832 | ||
7cdaf587 XW |
2833 | if (!S_ISREG(file_inode(sp->file_in)->i_mode)) { |
2834 | /* | |
2835 | * Splice operation will be punted aync, and here need to | |
2836 | * modify io_wq_work.flags, so initialize io_wq_work firstly. | |
2837 | */ | |
2838 | io_req_init_async(req); | |
7d67af2c | 2839 | req->work.flags |= IO_WQ_WORK_UNBOUND; |
7cdaf587 | 2840 | } |
7d67af2c PB |
2841 | |
2842 | return 0; | |
2843 | } | |
2844 | ||
f2a8d5c7 PB |
2845 | static int io_tee_prep(struct io_kiocb *req, |
2846 | const struct io_uring_sqe *sqe) | |
2847 | { | |
2848 | if (READ_ONCE(sqe->splice_off_in) || READ_ONCE(sqe->off)) | |
2849 | return -EINVAL; | |
2850 | return __io_splice_prep(req, sqe); | |
2851 | } | |
2852 | ||
2853 | static int io_tee(struct io_kiocb *req, bool force_nonblock) | |
2854 | { | |
2855 | struct io_splice *sp = &req->splice; | |
2856 | struct file *in = sp->file_in; | |
2857 | struct file *out = sp->file_out; | |
2858 | unsigned int flags = sp->flags & ~SPLICE_F_FD_IN_FIXED; | |
2859 | long ret = 0; | |
2860 | ||
2861 | if (force_nonblock) | |
2862 | return -EAGAIN; | |
2863 | if (sp->len) | |
2864 | ret = do_tee(in, out, sp->len, flags); | |
2865 | ||
2866 | io_put_file(req, in, (sp->flags & SPLICE_F_FD_IN_FIXED)); | |
2867 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
2868 | ||
2869 | io_cqring_add_event(req, ret); | |
2870 | if (ret != sp->len) | |
2871 | req_set_fail_links(req); | |
2872 | io_put_req(req); | |
2873 | return 0; | |
2874 | } | |
2875 | ||
2876 | static int io_splice_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) | |
2877 | { | |
2878 | struct io_splice* sp = &req->splice; | |
2879 | ||
2880 | sp->off_in = READ_ONCE(sqe->splice_off_in); | |
2881 | sp->off_out = READ_ONCE(sqe->off); | |
2882 | return __io_splice_prep(req, sqe); | |
2883 | } | |
2884 | ||
014db007 | 2885 | static int io_splice(struct io_kiocb *req, bool force_nonblock) |
7d67af2c PB |
2886 | { |
2887 | struct io_splice *sp = &req->splice; | |
2888 | struct file *in = sp->file_in; | |
2889 | struct file *out = sp->file_out; | |
2890 | unsigned int flags = sp->flags & ~SPLICE_F_FD_IN_FIXED; | |
2891 | loff_t *poff_in, *poff_out; | |
c9687426 | 2892 | long ret = 0; |
7d67af2c | 2893 | |
2fb3e822 PB |
2894 | if (force_nonblock) |
2895 | return -EAGAIN; | |
7d67af2c PB |
2896 | |
2897 | poff_in = (sp->off_in == -1) ? NULL : &sp->off_in; | |
2898 | poff_out = (sp->off_out == -1) ? NULL : &sp->off_out; | |
c9687426 | 2899 | |
948a7749 | 2900 | if (sp->len) |
c9687426 | 2901 | ret = do_splice(in, poff_in, out, poff_out, sp->len, flags); |
7d67af2c PB |
2902 | |
2903 | io_put_file(req, in, (sp->flags & SPLICE_F_FD_IN_FIXED)); | |
2904 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
2905 | ||
2906 | io_cqring_add_event(req, ret); | |
2907 | if (ret != sp->len) | |
2908 | req_set_fail_links(req); | |
014db007 | 2909 | io_put_req(req); |
7d67af2c PB |
2910 | return 0; |
2911 | } | |
2912 | ||
2b188cc1 JA |
2913 | /* |
2914 | * IORING_OP_NOP just posts a completion event, nothing else. | |
2915 | */ | |
78e19bbe | 2916 | static int io_nop(struct io_kiocb *req) |
2b188cc1 JA |
2917 | { |
2918 | struct io_ring_ctx *ctx = req->ctx; | |
2b188cc1 | 2919 | |
def596e9 JA |
2920 | if (unlikely(ctx->flags & IORING_SETUP_IOPOLL)) |
2921 | return -EINVAL; | |
2922 | ||
78e19bbe | 2923 | io_cqring_add_event(req, 0); |
e65ef56d | 2924 | io_put_req(req); |
2b188cc1 JA |
2925 | return 0; |
2926 | } | |
2927 | ||
3529d8c2 | 2928 | static int io_prep_fsync(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
c992fe29 | 2929 | { |
6b06314c | 2930 | struct io_ring_ctx *ctx = req->ctx; |
c992fe29 | 2931 | |
09bb8394 JA |
2932 | if (!req->file) |
2933 | return -EBADF; | |
c992fe29 | 2934 | |
6b06314c | 2935 | if (unlikely(ctx->flags & IORING_SETUP_IOPOLL)) |
def596e9 | 2936 | return -EINVAL; |
edafccee | 2937 | if (unlikely(sqe->addr || sqe->ioprio || sqe->buf_index)) |
c992fe29 CH |
2938 | return -EINVAL; |
2939 | ||
8ed8d3c3 JA |
2940 | req->sync.flags = READ_ONCE(sqe->fsync_flags); |
2941 | if (unlikely(req->sync.flags & ~IORING_FSYNC_DATASYNC)) | |
2942 | return -EINVAL; | |
2943 | ||
2944 | req->sync.off = READ_ONCE(sqe->off); | |
2945 | req->sync.len = READ_ONCE(sqe->len); | |
c992fe29 CH |
2946 | return 0; |
2947 | } | |
2948 | ||
ac45abc0 | 2949 | static int io_fsync(struct io_kiocb *req, bool force_nonblock) |
8ed8d3c3 | 2950 | { |
8ed8d3c3 | 2951 | loff_t end = req->sync.off + req->sync.len; |
8ed8d3c3 JA |
2952 | int ret; |
2953 | ||
ac45abc0 PB |
2954 | /* fsync always requires a blocking context */ |
2955 | if (force_nonblock) | |
2956 | return -EAGAIN; | |
2957 | ||
9adbd45d | 2958 | ret = vfs_fsync_range(req->file, req->sync.off, |
8ed8d3c3 JA |
2959 | end > 0 ? end : LLONG_MAX, |
2960 | req->sync.flags & IORING_FSYNC_DATASYNC); | |
2961 | if (ret < 0) | |
2962 | req_set_fail_links(req); | |
2963 | io_cqring_add_event(req, ret); | |
014db007 | 2964 | io_put_req(req); |
c992fe29 CH |
2965 | return 0; |
2966 | } | |
2967 | ||
d63d1b5e JA |
2968 | static int io_fallocate_prep(struct io_kiocb *req, |
2969 | const struct io_uring_sqe *sqe) | |
2970 | { | |
2971 | if (sqe->ioprio || sqe->buf_index || sqe->rw_flags) | |
2972 | return -EINVAL; | |
3232dd02 PB |
2973 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
2974 | return -EINVAL; | |
d63d1b5e JA |
2975 | |
2976 | req->sync.off = READ_ONCE(sqe->off); | |
2977 | req->sync.len = READ_ONCE(sqe->addr); | |
2978 | req->sync.mode = READ_ONCE(sqe->len); | |
4ed734b0 | 2979 | req->fsize = rlimit(RLIMIT_FSIZE); |
d63d1b5e JA |
2980 | return 0; |
2981 | } | |
2982 | ||
014db007 | 2983 | static int io_fallocate(struct io_kiocb *req, bool force_nonblock) |
5d17b4a4 | 2984 | { |
ac45abc0 PB |
2985 | int ret; |
2986 | ||
d63d1b5e | 2987 | /* fallocate always requiring blocking context */ |
ac45abc0 | 2988 | if (force_nonblock) |
5d17b4a4 JA |
2989 | return -EAGAIN; |
2990 | ||
ac45abc0 PB |
2991 | current->signal->rlim[RLIMIT_FSIZE].rlim_cur = req->fsize; |
2992 | ret = vfs_fallocate(req->file, req->sync.mode, req->sync.off, | |
2993 | req->sync.len); | |
2994 | current->signal->rlim[RLIMIT_FSIZE].rlim_cur = RLIM_INFINITY; | |
2995 | if (ret < 0) | |
2996 | req_set_fail_links(req); | |
2997 | io_cqring_add_event(req, ret); | |
2998 | io_put_req(req); | |
5d17b4a4 JA |
2999 | return 0; |
3000 | } | |
3001 | ||
ec65fea5 | 3002 | static int __io_openat_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
b7bb4f7d | 3003 | { |
f8748881 | 3004 | const char __user *fname; |
15b71abe | 3005 | int ret; |
b7bb4f7d | 3006 | |
3232dd02 | 3007 | if (unlikely(req->ctx->flags & (IORING_SETUP_IOPOLL|IORING_SETUP_SQPOLL))) |
15b71abe | 3008 | return -EINVAL; |
ec65fea5 | 3009 | if (unlikely(sqe->ioprio || sqe->buf_index)) |
15b71abe | 3010 | return -EINVAL; |
ec65fea5 | 3011 | if (unlikely(req->flags & REQ_F_FIXED_FILE)) |
cf3040ca | 3012 | return -EBADF; |
03b1230c | 3013 | |
ec65fea5 PB |
3014 | /* open.how should be already initialised */ |
3015 | if (!(req->open.how.flags & O_PATH) && force_o_largefile()) | |
08a1d26e | 3016 | req->open.how.flags |= O_LARGEFILE; |
3529d8c2 | 3017 | |
25e72d10 PB |
3018 | req->open.dfd = READ_ONCE(sqe->fd); |
3019 | fname = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
f8748881 | 3020 | req->open.filename = getname(fname); |
15b71abe JA |
3021 | if (IS_ERR(req->open.filename)) { |
3022 | ret = PTR_ERR(req->open.filename); | |
3023 | req->open.filename = NULL; | |
3024 | return ret; | |
3025 | } | |
4022e7af | 3026 | req->open.nofile = rlimit(RLIMIT_NOFILE); |
8fef80bf | 3027 | req->flags |= REQ_F_NEED_CLEANUP; |
15b71abe | 3028 | return 0; |
03b1230c JA |
3029 | } |
3030 | ||
ec65fea5 PB |
3031 | static int io_openat_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
3032 | { | |
3033 | u64 flags, mode; | |
3034 | ||
3035 | if (req->flags & REQ_F_NEED_CLEANUP) | |
3036 | return 0; | |
3037 | mode = READ_ONCE(sqe->len); | |
3038 | flags = READ_ONCE(sqe->open_flags); | |
3039 | req->open.how = build_open_how(flags, mode); | |
3040 | return __io_openat_prep(req, sqe); | |
3041 | } | |
3042 | ||
cebdb986 | 3043 | static int io_openat2_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
aa1fa28f | 3044 | { |
cebdb986 | 3045 | struct open_how __user *how; |
cebdb986 | 3046 | size_t len; |
0fa03c62 JA |
3047 | int ret; |
3048 | ||
0bdbdd08 PB |
3049 | if (req->flags & REQ_F_NEED_CLEANUP) |
3050 | return 0; | |
cebdb986 JA |
3051 | how = u64_to_user_ptr(READ_ONCE(sqe->addr2)); |
3052 | len = READ_ONCE(sqe->len); | |
cebdb986 JA |
3053 | if (len < OPEN_HOW_SIZE_VER0) |
3054 | return -EINVAL; | |
3529d8c2 | 3055 | |
cebdb986 JA |
3056 | ret = copy_struct_from_user(&req->open.how, sizeof(req->open.how), how, |
3057 | len); | |
3058 | if (ret) | |
3059 | return ret; | |
3529d8c2 | 3060 | |
ec65fea5 | 3061 | return __io_openat_prep(req, sqe); |
cebdb986 JA |
3062 | } |
3063 | ||
014db007 | 3064 | static int io_openat2(struct io_kiocb *req, bool force_nonblock) |
15b71abe JA |
3065 | { |
3066 | struct open_flags op; | |
15b71abe JA |
3067 | struct file *file; |
3068 | int ret; | |
3069 | ||
f86cd20c | 3070 | if (force_nonblock) |
15b71abe | 3071 | return -EAGAIN; |
15b71abe | 3072 | |
cebdb986 | 3073 | ret = build_open_flags(&req->open.how, &op); |
15b71abe JA |
3074 | if (ret) |
3075 | goto err; | |
3076 | ||
4022e7af | 3077 | ret = __get_unused_fd_flags(req->open.how.flags, req->open.nofile); |
15b71abe JA |
3078 | if (ret < 0) |
3079 | goto err; | |
3080 | ||
3081 | file = do_filp_open(req->open.dfd, req->open.filename, &op); | |
3082 | if (IS_ERR(file)) { | |
3083 | put_unused_fd(ret); | |
3084 | ret = PTR_ERR(file); | |
3085 | } else { | |
3086 | fsnotify_open(file); | |
3087 | fd_install(ret, file); | |
3088 | } | |
3089 | err: | |
3090 | putname(req->open.filename); | |
8fef80bf | 3091 | req->flags &= ~REQ_F_NEED_CLEANUP; |
15b71abe JA |
3092 | if (ret < 0) |
3093 | req_set_fail_links(req); | |
3094 | io_cqring_add_event(req, ret); | |
014db007 | 3095 | io_put_req(req); |
15b71abe JA |
3096 | return 0; |
3097 | } | |
3098 | ||
014db007 | 3099 | static int io_openat(struct io_kiocb *req, bool force_nonblock) |
cebdb986 | 3100 | { |
014db007 | 3101 | return io_openat2(req, force_nonblock); |
cebdb986 JA |
3102 | } |
3103 | ||
067524e9 JA |
3104 | static int io_remove_buffers_prep(struct io_kiocb *req, |
3105 | const struct io_uring_sqe *sqe) | |
3106 | { | |
3107 | struct io_provide_buf *p = &req->pbuf; | |
3108 | u64 tmp; | |
3109 | ||
3110 | if (sqe->ioprio || sqe->rw_flags || sqe->addr || sqe->len || sqe->off) | |
3111 | return -EINVAL; | |
3112 | ||
3113 | tmp = READ_ONCE(sqe->fd); | |
3114 | if (!tmp || tmp > USHRT_MAX) | |
3115 | return -EINVAL; | |
3116 | ||
3117 | memset(p, 0, sizeof(*p)); | |
3118 | p->nbufs = tmp; | |
3119 | p->bgid = READ_ONCE(sqe->buf_group); | |
3120 | return 0; | |
3121 | } | |
3122 | ||
3123 | static int __io_remove_buffers(struct io_ring_ctx *ctx, struct io_buffer *buf, | |
3124 | int bgid, unsigned nbufs) | |
3125 | { | |
3126 | unsigned i = 0; | |
3127 | ||
3128 | /* shouldn't happen */ | |
3129 | if (!nbufs) | |
3130 | return 0; | |
3131 | ||
3132 | /* the head kbuf is the list itself */ | |
3133 | while (!list_empty(&buf->list)) { | |
3134 | struct io_buffer *nxt; | |
3135 | ||
3136 | nxt = list_first_entry(&buf->list, struct io_buffer, list); | |
3137 | list_del(&nxt->list); | |
3138 | kfree(nxt); | |
3139 | if (++i == nbufs) | |
3140 | return i; | |
3141 | } | |
3142 | i++; | |
3143 | kfree(buf); | |
3144 | idr_remove(&ctx->io_buffer_idr, bgid); | |
3145 | ||
3146 | return i; | |
3147 | } | |
3148 | ||
3149 | static int io_remove_buffers(struct io_kiocb *req, bool force_nonblock) | |
3150 | { | |
3151 | struct io_provide_buf *p = &req->pbuf; | |
3152 | struct io_ring_ctx *ctx = req->ctx; | |
3153 | struct io_buffer *head; | |
3154 | int ret = 0; | |
3155 | ||
3156 | io_ring_submit_lock(ctx, !force_nonblock); | |
3157 | ||
3158 | lockdep_assert_held(&ctx->uring_lock); | |
3159 | ||
3160 | ret = -ENOENT; | |
3161 | head = idr_find(&ctx->io_buffer_idr, p->bgid); | |
3162 | if (head) | |
3163 | ret = __io_remove_buffers(ctx, head, p->bgid, p->nbufs); | |
3164 | ||
3165 | io_ring_submit_lock(ctx, !force_nonblock); | |
3166 | if (ret < 0) | |
3167 | req_set_fail_links(req); | |
3168 | io_cqring_add_event(req, ret); | |
3169 | io_put_req(req); | |
3170 | return 0; | |
3171 | } | |
3172 | ||
ddf0322d JA |
3173 | static int io_provide_buffers_prep(struct io_kiocb *req, |
3174 | const struct io_uring_sqe *sqe) | |
3175 | { | |
3176 | struct io_provide_buf *p = &req->pbuf; | |
3177 | u64 tmp; | |
3178 | ||
3179 | if (sqe->ioprio || sqe->rw_flags) | |
3180 | return -EINVAL; | |
3181 | ||
3182 | tmp = READ_ONCE(sqe->fd); | |
3183 | if (!tmp || tmp > USHRT_MAX) | |
3184 | return -E2BIG; | |
3185 | p->nbufs = tmp; | |
3186 | p->addr = READ_ONCE(sqe->addr); | |
3187 | p->len = READ_ONCE(sqe->len); | |
3188 | ||
efe68c1c | 3189 | if (!access_ok(u64_to_user_ptr(p->addr), (p->len * p->nbufs))) |
ddf0322d JA |
3190 | return -EFAULT; |
3191 | ||
3192 | p->bgid = READ_ONCE(sqe->buf_group); | |
3193 | tmp = READ_ONCE(sqe->off); | |
3194 | if (tmp > USHRT_MAX) | |
3195 | return -E2BIG; | |
3196 | p->bid = tmp; | |
3197 | return 0; | |
3198 | } | |
3199 | ||
3200 | static int io_add_buffers(struct io_provide_buf *pbuf, struct io_buffer **head) | |
3201 | { | |
3202 | struct io_buffer *buf; | |
3203 | u64 addr = pbuf->addr; | |
3204 | int i, bid = pbuf->bid; | |
3205 | ||
3206 | for (i = 0; i < pbuf->nbufs; i++) { | |
3207 | buf = kmalloc(sizeof(*buf), GFP_KERNEL); | |
3208 | if (!buf) | |
3209 | break; | |
3210 | ||
3211 | buf->addr = addr; | |
3212 | buf->len = pbuf->len; | |
3213 | buf->bid = bid; | |
3214 | addr += pbuf->len; | |
3215 | bid++; | |
3216 | if (!*head) { | |
3217 | INIT_LIST_HEAD(&buf->list); | |
3218 | *head = buf; | |
3219 | } else { | |
3220 | list_add_tail(&buf->list, &(*head)->list); | |
3221 | } | |
3222 | } | |
3223 | ||
3224 | return i ? i : -ENOMEM; | |
3225 | } | |
3226 | ||
ddf0322d JA |
3227 | static int io_provide_buffers(struct io_kiocb *req, bool force_nonblock) |
3228 | { | |
3229 | struct io_provide_buf *p = &req->pbuf; | |
3230 | struct io_ring_ctx *ctx = req->ctx; | |
3231 | struct io_buffer *head, *list; | |
3232 | int ret = 0; | |
3233 | ||
3234 | io_ring_submit_lock(ctx, !force_nonblock); | |
3235 | ||
3236 | lockdep_assert_held(&ctx->uring_lock); | |
3237 | ||
3238 | list = head = idr_find(&ctx->io_buffer_idr, p->bgid); | |
3239 | ||
3240 | ret = io_add_buffers(p, &head); | |
3241 | if (ret < 0) | |
3242 | goto out; | |
3243 | ||
3244 | if (!list) { | |
3245 | ret = idr_alloc(&ctx->io_buffer_idr, head, p->bgid, p->bgid + 1, | |
3246 | GFP_KERNEL); | |
3247 | if (ret < 0) { | |
067524e9 | 3248 | __io_remove_buffers(ctx, head, p->bgid, -1U); |
ddf0322d JA |
3249 | goto out; |
3250 | } | |
3251 | } | |
3252 | out: | |
3253 | io_ring_submit_unlock(ctx, !force_nonblock); | |
3254 | if (ret < 0) | |
3255 | req_set_fail_links(req); | |
3256 | io_cqring_add_event(req, ret); | |
3257 | io_put_req(req); | |
3258 | return 0; | |
cebdb986 JA |
3259 | } |
3260 | ||
3e4827b0 JA |
3261 | static int io_epoll_ctl_prep(struct io_kiocb *req, |
3262 | const struct io_uring_sqe *sqe) | |
3263 | { | |
3264 | #if defined(CONFIG_EPOLL) | |
3265 | if (sqe->ioprio || sqe->buf_index) | |
3266 | return -EINVAL; | |
3232dd02 PB |
3267 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
3268 | return -EINVAL; | |
3e4827b0 JA |
3269 | |
3270 | req->epoll.epfd = READ_ONCE(sqe->fd); | |
3271 | req->epoll.op = READ_ONCE(sqe->len); | |
3272 | req->epoll.fd = READ_ONCE(sqe->off); | |
3273 | ||
3274 | if (ep_op_has_event(req->epoll.op)) { | |
3275 | struct epoll_event __user *ev; | |
3276 | ||
3277 | ev = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
3278 | if (copy_from_user(&req->epoll.event, ev, sizeof(*ev))) | |
3279 | return -EFAULT; | |
3280 | } | |
3281 | ||
3282 | return 0; | |
3283 | #else | |
3284 | return -EOPNOTSUPP; | |
3285 | #endif | |
3286 | } | |
3287 | ||
014db007 | 3288 | static int io_epoll_ctl(struct io_kiocb *req, bool force_nonblock) |
3e4827b0 JA |
3289 | { |
3290 | #if defined(CONFIG_EPOLL) | |
3291 | struct io_epoll *ie = &req->epoll; | |
3292 | int ret; | |
3293 | ||
3294 | ret = do_epoll_ctl(ie->epfd, ie->op, ie->fd, &ie->event, force_nonblock); | |
3295 | if (force_nonblock && ret == -EAGAIN) | |
3296 | return -EAGAIN; | |
3297 | ||
3298 | if (ret < 0) | |
3299 | req_set_fail_links(req); | |
3300 | io_cqring_add_event(req, ret); | |
014db007 | 3301 | io_put_req(req); |
3e4827b0 JA |
3302 | return 0; |
3303 | #else | |
3304 | return -EOPNOTSUPP; | |
3305 | #endif | |
3306 | } | |
3307 | ||
c1ca757b JA |
3308 | static int io_madvise_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
3309 | { | |
3310 | #if defined(CONFIG_ADVISE_SYSCALLS) && defined(CONFIG_MMU) | |
3311 | if (sqe->ioprio || sqe->buf_index || sqe->off) | |
3312 | return -EINVAL; | |
3232dd02 PB |
3313 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
3314 | return -EINVAL; | |
c1ca757b JA |
3315 | |
3316 | req->madvise.addr = READ_ONCE(sqe->addr); | |
3317 | req->madvise.len = READ_ONCE(sqe->len); | |
3318 | req->madvise.advice = READ_ONCE(sqe->fadvise_advice); | |
3319 | return 0; | |
3320 | #else | |
3321 | return -EOPNOTSUPP; | |
3322 | #endif | |
3323 | } | |
3324 | ||
014db007 | 3325 | static int io_madvise(struct io_kiocb *req, bool force_nonblock) |
c1ca757b JA |
3326 | { |
3327 | #if defined(CONFIG_ADVISE_SYSCALLS) && defined(CONFIG_MMU) | |
3328 | struct io_madvise *ma = &req->madvise; | |
3329 | int ret; | |
3330 | ||
3331 | if (force_nonblock) | |
3332 | return -EAGAIN; | |
3333 | ||
3334 | ret = do_madvise(ma->addr, ma->len, ma->advice); | |
3335 | if (ret < 0) | |
3336 | req_set_fail_links(req); | |
3337 | io_cqring_add_event(req, ret); | |
014db007 | 3338 | io_put_req(req); |
c1ca757b JA |
3339 | return 0; |
3340 | #else | |
3341 | return -EOPNOTSUPP; | |
3342 | #endif | |
3343 | } | |
3344 | ||
4840e418 JA |
3345 | static int io_fadvise_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
3346 | { | |
3347 | if (sqe->ioprio || sqe->buf_index || sqe->addr) | |
3348 | return -EINVAL; | |
3232dd02 PB |
3349 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
3350 | return -EINVAL; | |
4840e418 JA |
3351 | |
3352 | req->fadvise.offset = READ_ONCE(sqe->off); | |
3353 | req->fadvise.len = READ_ONCE(sqe->len); | |
3354 | req->fadvise.advice = READ_ONCE(sqe->fadvise_advice); | |
3355 | return 0; | |
3356 | } | |
3357 | ||
014db007 | 3358 | static int io_fadvise(struct io_kiocb *req, bool force_nonblock) |
4840e418 JA |
3359 | { |
3360 | struct io_fadvise *fa = &req->fadvise; | |
3361 | int ret; | |
3362 | ||
3e69426d JA |
3363 | if (force_nonblock) { |
3364 | switch (fa->advice) { | |
3365 | case POSIX_FADV_NORMAL: | |
3366 | case POSIX_FADV_RANDOM: | |
3367 | case POSIX_FADV_SEQUENTIAL: | |
3368 | break; | |
3369 | default: | |
3370 | return -EAGAIN; | |
3371 | } | |
3372 | } | |
4840e418 JA |
3373 | |
3374 | ret = vfs_fadvise(req->file, fa->offset, fa->len, fa->advice); | |
3375 | if (ret < 0) | |
3376 | req_set_fail_links(req); | |
3377 | io_cqring_add_event(req, ret); | |
014db007 | 3378 | io_put_req(req); |
4840e418 JA |
3379 | return 0; |
3380 | } | |
3381 | ||
eddc7ef5 JA |
3382 | static int io_statx_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
3383 | { | |
3232dd02 PB |
3384 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
3385 | return -EINVAL; | |
eddc7ef5 JA |
3386 | if (sqe->ioprio || sqe->buf_index) |
3387 | return -EINVAL; | |
9c280f90 | 3388 | if (req->flags & REQ_F_FIXED_FILE) |
cf3040ca | 3389 | return -EBADF; |
eddc7ef5 | 3390 | |
1d9e1288 BM |
3391 | req->statx.dfd = READ_ONCE(sqe->fd); |
3392 | req->statx.mask = READ_ONCE(sqe->len); | |
e62753e4 | 3393 | req->statx.filename = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
1d9e1288 BM |
3394 | req->statx.buffer = u64_to_user_ptr(READ_ONCE(sqe->addr2)); |
3395 | req->statx.flags = READ_ONCE(sqe->statx_flags); | |
eddc7ef5 JA |
3396 | |
3397 | return 0; | |
3398 | } | |
3399 | ||
014db007 | 3400 | static int io_statx(struct io_kiocb *req, bool force_nonblock) |
eddc7ef5 | 3401 | { |
1d9e1288 | 3402 | struct io_statx *ctx = &req->statx; |
eddc7ef5 JA |
3403 | int ret; |
3404 | ||
5b0bbee4 JA |
3405 | if (force_nonblock) { |
3406 | /* only need file table for an actual valid fd */ | |
3407 | if (ctx->dfd == -1 || ctx->dfd == AT_FDCWD) | |
3408 | req->flags |= REQ_F_NO_FILE_TABLE; | |
eddc7ef5 | 3409 | return -EAGAIN; |
5b0bbee4 | 3410 | } |
eddc7ef5 | 3411 | |
e62753e4 BM |
3412 | ret = do_statx(ctx->dfd, ctx->filename, ctx->flags, ctx->mask, |
3413 | ctx->buffer); | |
eddc7ef5 | 3414 | |
eddc7ef5 JA |
3415 | if (ret < 0) |
3416 | req_set_fail_links(req); | |
3417 | io_cqring_add_event(req, ret); | |
014db007 | 3418 | io_put_req(req); |
eddc7ef5 JA |
3419 | return 0; |
3420 | } | |
3421 | ||
b5dba59e JA |
3422 | static int io_close_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
3423 | { | |
3424 | /* | |
3425 | * If we queue this for async, it must not be cancellable. That would | |
7cdaf587 XW |
3426 | * leave the 'file' in an undeterminate state, and here need to modify |
3427 | * io_wq_work.flags, so initialize io_wq_work firstly. | |
b5dba59e | 3428 | */ |
7cdaf587 | 3429 | io_req_init_async(req); |
b5dba59e JA |
3430 | req->work.flags |= IO_WQ_WORK_NO_CANCEL; |
3431 | ||
3232dd02 PB |
3432 | if (unlikely(req->ctx->flags & (IORING_SETUP_IOPOLL|IORING_SETUP_SQPOLL))) |
3433 | return -EINVAL; | |
b5dba59e JA |
3434 | if (sqe->ioprio || sqe->off || sqe->addr || sqe->len || |
3435 | sqe->rw_flags || sqe->buf_index) | |
3436 | return -EINVAL; | |
9c280f90 | 3437 | if (req->flags & REQ_F_FIXED_FILE) |
cf3040ca | 3438 | return -EBADF; |
b5dba59e JA |
3439 | |
3440 | req->close.fd = READ_ONCE(sqe->fd); | |
fd2206e4 JA |
3441 | if ((req->file && req->file->f_op == &io_uring_fops) || |
3442 | req->close.fd == req->ctx->ring_fd) | |
3443 | return -EBADF; | |
b5dba59e | 3444 | |
3af73b28 | 3445 | req->close.put_file = NULL; |
b5dba59e | 3446 | return 0; |
b5dba59e JA |
3447 | } |
3448 | ||
014db007 | 3449 | static int io_close(struct io_kiocb *req, bool force_nonblock) |
b5dba59e | 3450 | { |
3af73b28 | 3451 | struct io_close *close = &req->close; |
b5dba59e JA |
3452 | int ret; |
3453 | ||
3af73b28 PB |
3454 | /* might be already done during nonblock submission */ |
3455 | if (!close->put_file) { | |
3456 | ret = __close_fd_get_file(close->fd, &close->put_file); | |
3457 | if (ret < 0) | |
3458 | return (ret == -ENOENT) ? -EBADF : ret; | |
3459 | } | |
b5dba59e JA |
3460 | |
3461 | /* if the file has a flush method, be safe and punt to async */ | |
3af73b28 | 3462 | if (close->put_file->f_op->flush && force_nonblock) { |
0bf0eefd PB |
3463 | /* avoid grabbing files - we don't need the files */ |
3464 | req->flags |= REQ_F_NO_FILE_TABLE | REQ_F_MUST_PUNT; | |
0bf0eefd | 3465 | return -EAGAIN; |
a2100672 | 3466 | } |
b5dba59e | 3467 | |
3af73b28 PB |
3468 | /* No ->flush() or already async, safely close from here */ |
3469 | ret = filp_close(close->put_file, req->work.files); | |
3470 | if (ret < 0) | |
3471 | req_set_fail_links(req); | |
3472 | io_cqring_add_event(req, ret); | |
3473 | fput(close->put_file); | |
3474 | close->put_file = NULL; | |
3475 | io_put_req(req); | |
1a417f4e | 3476 | return 0; |
b5dba59e JA |
3477 | } |
3478 | ||
3529d8c2 | 3479 | static int io_prep_sfr(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
5d17b4a4 JA |
3480 | { |
3481 | struct io_ring_ctx *ctx = req->ctx; | |
5d17b4a4 JA |
3482 | |
3483 | if (!req->file) | |
3484 | return -EBADF; | |
5d17b4a4 JA |
3485 | |
3486 | if (unlikely(ctx->flags & IORING_SETUP_IOPOLL)) | |
3487 | return -EINVAL; | |
3488 | if (unlikely(sqe->addr || sqe->ioprio || sqe->buf_index)) | |
3489 | return -EINVAL; | |
3490 | ||
8ed8d3c3 JA |
3491 | req->sync.off = READ_ONCE(sqe->off); |
3492 | req->sync.len = READ_ONCE(sqe->len); | |
3493 | req->sync.flags = READ_ONCE(sqe->sync_range_flags); | |
8ed8d3c3 JA |
3494 | return 0; |
3495 | } | |
3496 | ||
ac45abc0 | 3497 | static int io_sync_file_range(struct io_kiocb *req, bool force_nonblock) |
8ed8d3c3 | 3498 | { |
8ed8d3c3 JA |
3499 | int ret; |
3500 | ||
ac45abc0 PB |
3501 | /* sync_file_range always requires a blocking context */ |
3502 | if (force_nonblock) | |
3503 | return -EAGAIN; | |
3504 | ||
9adbd45d | 3505 | ret = sync_file_range(req->file, req->sync.off, req->sync.len, |
8ed8d3c3 JA |
3506 | req->sync.flags); |
3507 | if (ret < 0) | |
3508 | req_set_fail_links(req); | |
3509 | io_cqring_add_event(req, ret); | |
014db007 | 3510 | io_put_req(req); |
5d17b4a4 JA |
3511 | return 0; |
3512 | } | |
3513 | ||
469956e8 | 3514 | #if defined(CONFIG_NET) |
02d27d89 PB |
3515 | static int io_setup_async_msg(struct io_kiocb *req, |
3516 | struct io_async_msghdr *kmsg) | |
3517 | { | |
3518 | if (req->io) | |
3519 | return -EAGAIN; | |
3520 | if (io_alloc_async_ctx(req)) { | |
3521 | if (kmsg->iov != kmsg->fast_iov) | |
3522 | kfree(kmsg->iov); | |
3523 | return -ENOMEM; | |
3524 | } | |
3525 | req->flags |= REQ_F_NEED_CLEANUP; | |
3526 | memcpy(&req->io->msg, kmsg, sizeof(*kmsg)); | |
3527 | return -EAGAIN; | |
3528 | } | |
3529 | ||
3529d8c2 | 3530 | static int io_sendmsg_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
03b1230c | 3531 | { |
e47293fd | 3532 | struct io_sr_msg *sr = &req->sr_msg; |
3529d8c2 | 3533 | struct io_async_ctx *io = req->io; |
99bc4c38 | 3534 | int ret; |
03b1230c | 3535 | |
d2b6f48b PB |
3536 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
3537 | return -EINVAL; | |
3538 | ||
e47293fd JA |
3539 | sr->msg_flags = READ_ONCE(sqe->msg_flags); |
3540 | sr->msg = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
fddaface | 3541 | sr->len = READ_ONCE(sqe->len); |
3529d8c2 | 3542 | |
d8768362 JA |
3543 | #ifdef CONFIG_COMPAT |
3544 | if (req->ctx->compat) | |
3545 | sr->msg_flags |= MSG_CMSG_COMPAT; | |
3546 | #endif | |
3547 | ||
fddaface | 3548 | if (!io || req->opcode == IORING_OP_SEND) |
3529d8c2 | 3549 | return 0; |
5f798bea PB |
3550 | /* iovec is already imported */ |
3551 | if (req->flags & REQ_F_NEED_CLEANUP) | |
3552 | return 0; | |
3529d8c2 | 3553 | |
d9688565 | 3554 | io->msg.iov = io->msg.fast_iov; |
99bc4c38 | 3555 | ret = sendmsg_copy_msghdr(&io->msg.msg, sr->msg, sr->msg_flags, |
e47293fd | 3556 | &io->msg.iov); |
99bc4c38 PB |
3557 | if (!ret) |
3558 | req->flags |= REQ_F_NEED_CLEANUP; | |
3559 | return ret; | |
03b1230c JA |
3560 | } |
3561 | ||
014db007 | 3562 | static int io_sendmsg(struct io_kiocb *req, bool force_nonblock) |
aa1fa28f | 3563 | { |
0b416c3e | 3564 | struct io_async_msghdr *kmsg = NULL; |
0fa03c62 JA |
3565 | struct socket *sock; |
3566 | int ret; | |
3567 | ||
0fa03c62 JA |
3568 | sock = sock_from_file(req->file, &ret); |
3569 | if (sock) { | |
b7bb4f7d | 3570 | struct io_async_ctx io; |
0fa03c62 JA |
3571 | unsigned flags; |
3572 | ||
03b1230c | 3573 | if (req->io) { |
0b416c3e | 3574 | kmsg = &req->io->msg; |
b537916c | 3575 | kmsg->msg.msg_name = &req->io->msg.addr; |
0b416c3e JA |
3576 | /* if iov is set, it's allocated already */ |
3577 | if (!kmsg->iov) | |
3578 | kmsg->iov = kmsg->fast_iov; | |
3579 | kmsg->msg.msg_iter.iov = kmsg->iov; | |
03b1230c | 3580 | } else { |
3529d8c2 JA |
3581 | struct io_sr_msg *sr = &req->sr_msg; |
3582 | ||
0b416c3e | 3583 | kmsg = &io.msg; |
b537916c | 3584 | kmsg->msg.msg_name = &io.msg.addr; |
3529d8c2 JA |
3585 | |
3586 | io.msg.iov = io.msg.fast_iov; | |
3587 | ret = sendmsg_copy_msghdr(&io.msg.msg, sr->msg, | |
3588 | sr->msg_flags, &io.msg.iov); | |
03b1230c | 3589 | if (ret) |
3529d8c2 | 3590 | return ret; |
03b1230c | 3591 | } |
0fa03c62 | 3592 | |
e47293fd JA |
3593 | flags = req->sr_msg.msg_flags; |
3594 | if (flags & MSG_DONTWAIT) | |
3595 | req->flags |= REQ_F_NOWAIT; | |
3596 | else if (force_nonblock) | |
3597 | flags |= MSG_DONTWAIT; | |
3598 | ||
0b416c3e | 3599 | ret = __sys_sendmsg_sock(sock, &kmsg->msg, flags); |
02d27d89 PB |
3600 | if (force_nonblock && ret == -EAGAIN) |
3601 | return io_setup_async_msg(req, kmsg); | |
441cdbd5 JA |
3602 | if (ret == -ERESTARTSYS) |
3603 | ret = -EINTR; | |
0fa03c62 JA |
3604 | } |
3605 | ||
1e95081c | 3606 | if (kmsg && kmsg->iov != kmsg->fast_iov) |
0b416c3e | 3607 | kfree(kmsg->iov); |
99bc4c38 | 3608 | req->flags &= ~REQ_F_NEED_CLEANUP; |
78e19bbe | 3609 | io_cqring_add_event(req, ret); |
4e88d6e7 JA |
3610 | if (ret < 0) |
3611 | req_set_fail_links(req); | |
014db007 | 3612 | io_put_req(req); |
5d17b4a4 | 3613 | return 0; |
03b1230c | 3614 | } |
aa1fa28f | 3615 | |
014db007 | 3616 | static int io_send(struct io_kiocb *req, bool force_nonblock) |
fddaface | 3617 | { |
fddaface JA |
3618 | struct socket *sock; |
3619 | int ret; | |
3620 | ||
fddaface JA |
3621 | sock = sock_from_file(req->file, &ret); |
3622 | if (sock) { | |
3623 | struct io_sr_msg *sr = &req->sr_msg; | |
3624 | struct msghdr msg; | |
3625 | struct iovec iov; | |
3626 | unsigned flags; | |
3627 | ||
3628 | ret = import_single_range(WRITE, sr->buf, sr->len, &iov, | |
3629 | &msg.msg_iter); | |
3630 | if (ret) | |
3631 | return ret; | |
3632 | ||
3633 | msg.msg_name = NULL; | |
3634 | msg.msg_control = NULL; | |
3635 | msg.msg_controllen = 0; | |
3636 | msg.msg_namelen = 0; | |
3637 | ||
3638 | flags = req->sr_msg.msg_flags; | |
3639 | if (flags & MSG_DONTWAIT) | |
3640 | req->flags |= REQ_F_NOWAIT; | |
3641 | else if (force_nonblock) | |
3642 | flags |= MSG_DONTWAIT; | |
3643 | ||
0b7b21e4 JA |
3644 | msg.msg_flags = flags; |
3645 | ret = sock_sendmsg(sock, &msg); | |
fddaface JA |
3646 | if (force_nonblock && ret == -EAGAIN) |
3647 | return -EAGAIN; | |
3648 | if (ret == -ERESTARTSYS) | |
3649 | ret = -EINTR; | |
3650 | } | |
3651 | ||
3652 | io_cqring_add_event(req, ret); | |
3653 | if (ret < 0) | |
3654 | req_set_fail_links(req); | |
014db007 | 3655 | io_put_req(req); |
fddaface | 3656 | return 0; |
fddaface JA |
3657 | } |
3658 | ||
52de1fe1 JA |
3659 | static int __io_recvmsg_copy_hdr(struct io_kiocb *req, struct io_async_ctx *io) |
3660 | { | |
3661 | struct io_sr_msg *sr = &req->sr_msg; | |
3662 | struct iovec __user *uiov; | |
3663 | size_t iov_len; | |
3664 | int ret; | |
3665 | ||
3666 | ret = __copy_msghdr_from_user(&io->msg.msg, sr->msg, &io->msg.uaddr, | |
3667 | &uiov, &iov_len); | |
3668 | if (ret) | |
3669 | return ret; | |
3670 | ||
3671 | if (req->flags & REQ_F_BUFFER_SELECT) { | |
3672 | if (iov_len > 1) | |
3673 | return -EINVAL; | |
3674 | if (copy_from_user(io->msg.iov, uiov, sizeof(*uiov))) | |
3675 | return -EFAULT; | |
3676 | sr->len = io->msg.iov[0].iov_len; | |
3677 | iov_iter_init(&io->msg.msg.msg_iter, READ, io->msg.iov, 1, | |
3678 | sr->len); | |
3679 | io->msg.iov = NULL; | |
3680 | } else { | |
3681 | ret = import_iovec(READ, uiov, iov_len, UIO_FASTIOV, | |
3682 | &io->msg.iov, &io->msg.msg.msg_iter); | |
3683 | if (ret > 0) | |
3684 | ret = 0; | |
3685 | } | |
3686 | ||
3687 | return ret; | |
3688 | } | |
3689 | ||
3690 | #ifdef CONFIG_COMPAT | |
3691 | static int __io_compat_recvmsg_copy_hdr(struct io_kiocb *req, | |
3692 | struct io_async_ctx *io) | |
3693 | { | |
3694 | struct compat_msghdr __user *msg_compat; | |
3695 | struct io_sr_msg *sr = &req->sr_msg; | |
3696 | struct compat_iovec __user *uiov; | |
3697 | compat_uptr_t ptr; | |
3698 | compat_size_t len; | |
3699 | int ret; | |
3700 | ||
3701 | msg_compat = (struct compat_msghdr __user *) sr->msg; | |
3702 | ret = __get_compat_msghdr(&io->msg.msg, msg_compat, &io->msg.uaddr, | |
3703 | &ptr, &len); | |
3704 | if (ret) | |
3705 | return ret; | |
3706 | ||
3707 | uiov = compat_ptr(ptr); | |
3708 | if (req->flags & REQ_F_BUFFER_SELECT) { | |
3709 | compat_ssize_t clen; | |
3710 | ||
3711 | if (len > 1) | |
3712 | return -EINVAL; | |
3713 | if (!access_ok(uiov, sizeof(*uiov))) | |
3714 | return -EFAULT; | |
3715 | if (__get_user(clen, &uiov->iov_len)) | |
3716 | return -EFAULT; | |
3717 | if (clen < 0) | |
3718 | return -EINVAL; | |
3719 | sr->len = io->msg.iov[0].iov_len; | |
3720 | io->msg.iov = NULL; | |
3721 | } else { | |
3722 | ret = compat_import_iovec(READ, uiov, len, UIO_FASTIOV, | |
3723 | &io->msg.iov, | |
3724 | &io->msg.msg.msg_iter); | |
3725 | if (ret < 0) | |
3726 | return ret; | |
3727 | } | |
3728 | ||
3729 | return 0; | |
3730 | } | |
3731 | #endif | |
3732 | ||
3733 | static int io_recvmsg_copy_hdr(struct io_kiocb *req, struct io_async_ctx *io) | |
3734 | { | |
3735 | io->msg.iov = io->msg.fast_iov; | |
3736 | ||
3737 | #ifdef CONFIG_COMPAT | |
3738 | if (req->ctx->compat) | |
3739 | return __io_compat_recvmsg_copy_hdr(req, io); | |
fddaface | 3740 | #endif |
52de1fe1 JA |
3741 | |
3742 | return __io_recvmsg_copy_hdr(req, io); | |
3743 | } | |
3744 | ||
bcda7baa JA |
3745 | static struct io_buffer *io_recv_buffer_select(struct io_kiocb *req, |
3746 | int *cflags, bool needs_lock) | |
3747 | { | |
3748 | struct io_sr_msg *sr = &req->sr_msg; | |
3749 | struct io_buffer *kbuf; | |
3750 | ||
3751 | if (!(req->flags & REQ_F_BUFFER_SELECT)) | |
3752 | return NULL; | |
3753 | ||
3754 | kbuf = io_buffer_select(req, &sr->len, sr->bgid, sr->kbuf, needs_lock); | |
3755 | if (IS_ERR(kbuf)) | |
3756 | return kbuf; | |
3757 | ||
3758 | sr->kbuf = kbuf; | |
3759 | req->flags |= REQ_F_BUFFER_SELECTED; | |
3760 | ||
3761 | *cflags = kbuf->bid << IORING_CQE_BUFFER_SHIFT; | |
3762 | *cflags |= IORING_CQE_F_BUFFER; | |
3763 | return kbuf; | |
fddaface JA |
3764 | } |
3765 | ||
3529d8c2 JA |
3766 | static int io_recvmsg_prep(struct io_kiocb *req, |
3767 | const struct io_uring_sqe *sqe) | |
aa1fa28f | 3768 | { |
e47293fd | 3769 | struct io_sr_msg *sr = &req->sr_msg; |
3529d8c2 | 3770 | struct io_async_ctx *io = req->io; |
99bc4c38 | 3771 | int ret; |
3529d8c2 | 3772 | |
d2b6f48b PB |
3773 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
3774 | return -EINVAL; | |
3775 | ||
3529d8c2 JA |
3776 | sr->msg_flags = READ_ONCE(sqe->msg_flags); |
3777 | sr->msg = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
0b7b21e4 | 3778 | sr->len = READ_ONCE(sqe->len); |
bcda7baa | 3779 | sr->bgid = READ_ONCE(sqe->buf_group); |
06b76d44 | 3780 | |
d8768362 JA |
3781 | #ifdef CONFIG_COMPAT |
3782 | if (req->ctx->compat) | |
3783 | sr->msg_flags |= MSG_CMSG_COMPAT; | |
3784 | #endif | |
3785 | ||
fddaface | 3786 | if (!io || req->opcode == IORING_OP_RECV) |
06b76d44 | 3787 | return 0; |
5f798bea PB |
3788 | /* iovec is already imported */ |
3789 | if (req->flags & REQ_F_NEED_CLEANUP) | |
3790 | return 0; | |
03b1230c | 3791 | |
52de1fe1 | 3792 | ret = io_recvmsg_copy_hdr(req, io); |
99bc4c38 PB |
3793 | if (!ret) |
3794 | req->flags |= REQ_F_NEED_CLEANUP; | |
3795 | return ret; | |
aa1fa28f JA |
3796 | } |
3797 | ||
014db007 | 3798 | static int io_recvmsg(struct io_kiocb *req, bool force_nonblock) |
aa1fa28f | 3799 | { |
0b416c3e | 3800 | struct io_async_msghdr *kmsg = NULL; |
03b1230c | 3801 | struct socket *sock; |
52de1fe1 | 3802 | int ret, cflags = 0; |
03b1230c | 3803 | |
03b1230c JA |
3804 | sock = sock_from_file(req->file, &ret); |
3805 | if (sock) { | |
52de1fe1 | 3806 | struct io_buffer *kbuf; |
b7bb4f7d | 3807 | struct io_async_ctx io; |
03b1230c JA |
3808 | unsigned flags; |
3809 | ||
03b1230c | 3810 | if (req->io) { |
0b416c3e | 3811 | kmsg = &req->io->msg; |
b537916c | 3812 | kmsg->msg.msg_name = &req->io->msg.addr; |
0b416c3e JA |
3813 | /* if iov is set, it's allocated already */ |
3814 | if (!kmsg->iov) | |
3815 | kmsg->iov = kmsg->fast_iov; | |
3816 | kmsg->msg.msg_iter.iov = kmsg->iov; | |
03b1230c | 3817 | } else { |
0b416c3e | 3818 | kmsg = &io.msg; |
b537916c | 3819 | kmsg->msg.msg_name = &io.msg.addr; |
3529d8c2 | 3820 | |
52de1fe1 | 3821 | ret = io_recvmsg_copy_hdr(req, &io); |
03b1230c | 3822 | if (ret) |
3529d8c2 | 3823 | return ret; |
03b1230c JA |
3824 | } |
3825 | ||
52de1fe1 JA |
3826 | kbuf = io_recv_buffer_select(req, &cflags, !force_nonblock); |
3827 | if (IS_ERR(kbuf)) { | |
3828 | return PTR_ERR(kbuf); | |
3829 | } else if (kbuf) { | |
3830 | kmsg->fast_iov[0].iov_base = u64_to_user_ptr(kbuf->addr); | |
3831 | iov_iter_init(&kmsg->msg.msg_iter, READ, kmsg->iov, | |
3832 | 1, req->sr_msg.len); | |
3833 | } | |
3834 | ||
e47293fd JA |
3835 | flags = req->sr_msg.msg_flags; |
3836 | if (flags & MSG_DONTWAIT) | |
3837 | req->flags |= REQ_F_NOWAIT; | |
3838 | else if (force_nonblock) | |
3839 | flags |= MSG_DONTWAIT; | |
3840 | ||
3841 | ret = __sys_recvmsg_sock(sock, &kmsg->msg, req->sr_msg.msg, | |
3842 | kmsg->uaddr, flags); | |
02d27d89 PB |
3843 | if (force_nonblock && ret == -EAGAIN) |
3844 | return io_setup_async_msg(req, kmsg); | |
03b1230c JA |
3845 | if (ret == -ERESTARTSYS) |
3846 | ret = -EINTR; | |
3847 | } | |
3848 | ||
1e95081c | 3849 | if (kmsg && kmsg->iov != kmsg->fast_iov) |
0b416c3e | 3850 | kfree(kmsg->iov); |
99bc4c38 | 3851 | req->flags &= ~REQ_F_NEED_CLEANUP; |
52de1fe1 | 3852 | __io_cqring_add_event(req, ret, cflags); |
4e88d6e7 JA |
3853 | if (ret < 0) |
3854 | req_set_fail_links(req); | |
014db007 | 3855 | io_put_req(req); |
03b1230c | 3856 | return 0; |
0fa03c62 | 3857 | } |
5d17b4a4 | 3858 | |
014db007 | 3859 | static int io_recv(struct io_kiocb *req, bool force_nonblock) |
fddaface | 3860 | { |
bcda7baa | 3861 | struct io_buffer *kbuf = NULL; |
fddaface | 3862 | struct socket *sock; |
bcda7baa | 3863 | int ret, cflags = 0; |
fddaface | 3864 | |
fddaface JA |
3865 | sock = sock_from_file(req->file, &ret); |
3866 | if (sock) { | |
3867 | struct io_sr_msg *sr = &req->sr_msg; | |
bcda7baa | 3868 | void __user *buf = sr->buf; |
fddaface JA |
3869 | struct msghdr msg; |
3870 | struct iovec iov; | |
3871 | unsigned flags; | |
3872 | ||
bcda7baa JA |
3873 | kbuf = io_recv_buffer_select(req, &cflags, !force_nonblock); |
3874 | if (IS_ERR(kbuf)) | |
3875 | return PTR_ERR(kbuf); | |
3876 | else if (kbuf) | |
3877 | buf = u64_to_user_ptr(kbuf->addr); | |
3878 | ||
3879 | ret = import_single_range(READ, buf, sr->len, &iov, | |
fddaface | 3880 | &msg.msg_iter); |
bcda7baa JA |
3881 | if (ret) { |
3882 | kfree(kbuf); | |
fddaface | 3883 | return ret; |
bcda7baa | 3884 | } |
fddaface | 3885 | |
bcda7baa | 3886 | req->flags |= REQ_F_NEED_CLEANUP; |
fddaface JA |
3887 | msg.msg_name = NULL; |
3888 | msg.msg_control = NULL; | |
3889 | msg.msg_controllen = 0; | |
3890 | msg.msg_namelen = 0; | |
3891 | msg.msg_iocb = NULL; | |
3892 | msg.msg_flags = 0; | |
3893 | ||
3894 | flags = req->sr_msg.msg_flags; | |
3895 | if (flags & MSG_DONTWAIT) | |
3896 | req->flags |= REQ_F_NOWAIT; | |
3897 | else if (force_nonblock) | |
3898 | flags |= MSG_DONTWAIT; | |
3899 | ||
0b7b21e4 | 3900 | ret = sock_recvmsg(sock, &msg, flags); |
fddaface JA |
3901 | if (force_nonblock && ret == -EAGAIN) |
3902 | return -EAGAIN; | |
3903 | if (ret == -ERESTARTSYS) | |
3904 | ret = -EINTR; | |
3905 | } | |
3906 | ||
bcda7baa JA |
3907 | kfree(kbuf); |
3908 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
3909 | __io_cqring_add_event(req, ret, cflags); | |
fddaface JA |
3910 | if (ret < 0) |
3911 | req_set_fail_links(req); | |
014db007 | 3912 | io_put_req(req); |
fddaface | 3913 | return 0; |
fddaface JA |
3914 | } |
3915 | ||
3529d8c2 | 3916 | static int io_accept_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
17f2fe35 | 3917 | { |
8ed8d3c3 JA |
3918 | struct io_accept *accept = &req->accept; |
3919 | ||
17f2fe35 JA |
3920 | if (unlikely(req->ctx->flags & (IORING_SETUP_IOPOLL|IORING_SETUP_SQPOLL))) |
3921 | return -EINVAL; | |
8042d6ce | 3922 | if (sqe->ioprio || sqe->len || sqe->buf_index) |
17f2fe35 JA |
3923 | return -EINVAL; |
3924 | ||
d55e5f5b JA |
3925 | accept->addr = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
3926 | accept->addr_len = u64_to_user_ptr(READ_ONCE(sqe->addr2)); | |
8ed8d3c3 | 3927 | accept->flags = READ_ONCE(sqe->accept_flags); |
09952e3e | 3928 | accept->nofile = rlimit(RLIMIT_NOFILE); |
8ed8d3c3 | 3929 | return 0; |
8ed8d3c3 | 3930 | } |
17f2fe35 | 3931 | |
ac45abc0 | 3932 | static int io_accept(struct io_kiocb *req, bool force_nonblock) |
8ed8d3c3 JA |
3933 | { |
3934 | struct io_accept *accept = &req->accept; | |
ac45abc0 | 3935 | unsigned int file_flags = force_nonblock ? O_NONBLOCK : 0; |
8ed8d3c3 JA |
3936 | int ret; |
3937 | ||
e697deed JX |
3938 | if (req->file->f_flags & O_NONBLOCK) |
3939 | req->flags |= REQ_F_NOWAIT; | |
3940 | ||
8ed8d3c3 | 3941 | ret = __sys_accept4_file(req->file, file_flags, accept->addr, |
09952e3e JA |
3942 | accept->addr_len, accept->flags, |
3943 | accept->nofile); | |
8ed8d3c3 | 3944 | if (ret == -EAGAIN && force_nonblock) |
17f2fe35 | 3945 | return -EAGAIN; |
ac45abc0 PB |
3946 | if (ret < 0) { |
3947 | if (ret == -ERESTARTSYS) | |
3948 | ret = -EINTR; | |
4e88d6e7 | 3949 | req_set_fail_links(req); |
ac45abc0 | 3950 | } |
78e19bbe | 3951 | io_cqring_add_event(req, ret); |
014db007 | 3952 | io_put_req(req); |
17f2fe35 | 3953 | return 0; |
8ed8d3c3 JA |
3954 | } |
3955 | ||
3529d8c2 | 3956 | static int io_connect_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
f499a021 | 3957 | { |
3529d8c2 JA |
3958 | struct io_connect *conn = &req->connect; |
3959 | struct io_async_ctx *io = req->io; | |
f499a021 | 3960 | |
3fbb51c1 JA |
3961 | if (unlikely(req->ctx->flags & (IORING_SETUP_IOPOLL|IORING_SETUP_SQPOLL))) |
3962 | return -EINVAL; | |
3963 | if (sqe->ioprio || sqe->len || sqe->buf_index || sqe->rw_flags) | |
3964 | return -EINVAL; | |
3965 | ||
3529d8c2 JA |
3966 | conn->addr = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
3967 | conn->addr_len = READ_ONCE(sqe->addr2); | |
3968 | ||
3969 | if (!io) | |
3970 | return 0; | |
3971 | ||
3972 | return move_addr_to_kernel(conn->addr, conn->addr_len, | |
3fbb51c1 | 3973 | &io->connect.address); |
f499a021 JA |
3974 | } |
3975 | ||
014db007 | 3976 | static int io_connect(struct io_kiocb *req, bool force_nonblock) |
f8e85cf2 | 3977 | { |
f499a021 | 3978 | struct io_async_ctx __io, *io; |
f8e85cf2 | 3979 | unsigned file_flags; |
3fbb51c1 | 3980 | int ret; |
f8e85cf2 | 3981 | |
f499a021 JA |
3982 | if (req->io) { |
3983 | io = req->io; | |
3984 | } else { | |
3529d8c2 JA |
3985 | ret = move_addr_to_kernel(req->connect.addr, |
3986 | req->connect.addr_len, | |
3987 | &__io.connect.address); | |
f499a021 JA |
3988 | if (ret) |
3989 | goto out; | |
3990 | io = &__io; | |
3991 | } | |
3992 | ||
3fbb51c1 JA |
3993 | file_flags = force_nonblock ? O_NONBLOCK : 0; |
3994 | ||
3995 | ret = __sys_connect_file(req->file, &io->connect.address, | |
3996 | req->connect.addr_len, file_flags); | |
87f80d62 | 3997 | if ((ret == -EAGAIN || ret == -EINPROGRESS) && force_nonblock) { |
b7bb4f7d JA |
3998 | if (req->io) |
3999 | return -EAGAIN; | |
4000 | if (io_alloc_async_ctx(req)) { | |
f499a021 JA |
4001 | ret = -ENOMEM; |
4002 | goto out; | |
4003 | } | |
b7bb4f7d | 4004 | memcpy(&req->io->connect, &__io.connect, sizeof(__io.connect)); |
f8e85cf2 | 4005 | return -EAGAIN; |
f499a021 | 4006 | } |
f8e85cf2 JA |
4007 | if (ret == -ERESTARTSYS) |
4008 | ret = -EINTR; | |
f499a021 | 4009 | out: |
4e88d6e7 JA |
4010 | if (ret < 0) |
4011 | req_set_fail_links(req); | |
f8e85cf2 | 4012 | io_cqring_add_event(req, ret); |
014db007 | 4013 | io_put_req(req); |
f8e85cf2 | 4014 | return 0; |
469956e8 Y |
4015 | } |
4016 | #else /* !CONFIG_NET */ | |
4017 | static int io_sendmsg_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) | |
4018 | { | |
f8e85cf2 | 4019 | return -EOPNOTSUPP; |
f8e85cf2 JA |
4020 | } |
4021 | ||
469956e8 Y |
4022 | static int io_sendmsg(struct io_kiocb *req, bool force_nonblock) |
4023 | { | |
4024 | return -EOPNOTSUPP; | |
4025 | } | |
4026 | ||
4027 | static int io_send(struct io_kiocb *req, bool force_nonblock) | |
4028 | { | |
4029 | return -EOPNOTSUPP; | |
4030 | } | |
4031 | ||
4032 | static int io_recvmsg_prep(struct io_kiocb *req, | |
4033 | const struct io_uring_sqe *sqe) | |
4034 | { | |
4035 | return -EOPNOTSUPP; | |
4036 | } | |
4037 | ||
4038 | static int io_recvmsg(struct io_kiocb *req, bool force_nonblock) | |
4039 | { | |
4040 | return -EOPNOTSUPP; | |
4041 | } | |
4042 | ||
4043 | static int io_recv(struct io_kiocb *req, bool force_nonblock) | |
4044 | { | |
4045 | return -EOPNOTSUPP; | |
4046 | } | |
4047 | ||
4048 | static int io_accept_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) | |
4049 | { | |
4050 | return -EOPNOTSUPP; | |
4051 | } | |
4052 | ||
4053 | static int io_accept(struct io_kiocb *req, bool force_nonblock) | |
4054 | { | |
4055 | return -EOPNOTSUPP; | |
4056 | } | |
4057 | ||
4058 | static int io_connect_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) | |
4059 | { | |
4060 | return -EOPNOTSUPP; | |
4061 | } | |
4062 | ||
4063 | static int io_connect(struct io_kiocb *req, bool force_nonblock) | |
4064 | { | |
f8e85cf2 | 4065 | return -EOPNOTSUPP; |
f8e85cf2 | 4066 | } |
469956e8 | 4067 | #endif /* CONFIG_NET */ |
f8e85cf2 | 4068 | |
d7718a9d JA |
4069 | struct io_poll_table { |
4070 | struct poll_table_struct pt; | |
4071 | struct io_kiocb *req; | |
4072 | int error; | |
4073 | }; | |
4074 | ||
ce593a6c JA |
4075 | static int io_req_task_work_add(struct io_kiocb *req, struct callback_head *cb, |
4076 | int notify) | |
4077 | { | |
4078 | struct task_struct *tsk = req->task; | |
4079 | int ret; | |
4080 | ||
4081 | if (req->ctx->flags & IORING_SETUP_SQPOLL) | |
4082 | notify = 0; | |
4083 | ||
4084 | ret = task_work_add(tsk, cb, notify); | |
4085 | if (!ret) | |
4086 | wake_up_process(tsk); | |
4087 | return ret; | |
4088 | } | |
4089 | ||
d7718a9d JA |
4090 | static int __io_async_wake(struct io_kiocb *req, struct io_poll_iocb *poll, |
4091 | __poll_t mask, task_work_func_t func) | |
4092 | { | |
4093 | struct task_struct *tsk; | |
aa96bf8a | 4094 | int ret; |
d7718a9d JA |
4095 | |
4096 | /* for instances that support it check for an event match first: */ | |
4097 | if (mask && !(mask & poll->events)) | |
4098 | return 0; | |
4099 | ||
4100 | trace_io_uring_task_add(req->ctx, req->opcode, req->user_data, mask); | |
4101 | ||
4102 | list_del_init(&poll->wait.entry); | |
4103 | ||
4104 | tsk = req->task; | |
4105 | req->result = mask; | |
4106 | init_task_work(&req->task_work, func); | |
4107 | /* | |
e3aabf95 JA |
4108 | * If this fails, then the task is exiting. When a task exits, the |
4109 | * work gets canceled, so just cancel this request as well instead | |
4110 | * of executing it. We can't safely execute it anyway, as we may not | |
4111 | * have the needed state needed for it anyway. | |
d7718a9d | 4112 | */ |
ce593a6c | 4113 | ret = io_req_task_work_add(req, &req->task_work, TWA_SIGNAL); |
aa96bf8a | 4114 | if (unlikely(ret)) { |
e3aabf95 | 4115 | WRITE_ONCE(poll->canceled, true); |
aa96bf8a | 4116 | tsk = io_wq_get_task(req->ctx->io_wq); |
ce593a6c JA |
4117 | task_work_add(tsk, &req->task_work, 0); |
4118 | wake_up_process(tsk); | |
aa96bf8a | 4119 | } |
d7718a9d JA |
4120 | return 1; |
4121 | } | |
4122 | ||
74ce6ce4 JA |
4123 | static bool io_poll_rewait(struct io_kiocb *req, struct io_poll_iocb *poll) |
4124 | __acquires(&req->ctx->completion_lock) | |
4125 | { | |
4126 | struct io_ring_ctx *ctx = req->ctx; | |
4127 | ||
4128 | if (!req->result && !READ_ONCE(poll->canceled)) { | |
4129 | struct poll_table_struct pt = { ._key = poll->events }; | |
4130 | ||
4131 | req->result = vfs_poll(req->file, &pt) & poll->events; | |
4132 | } | |
4133 | ||
4134 | spin_lock_irq(&ctx->completion_lock); | |
4135 | if (!req->result && !READ_ONCE(poll->canceled)) { | |
4136 | add_wait_queue(poll->head, &poll->wait); | |
4137 | return true; | |
4138 | } | |
4139 | ||
4140 | return false; | |
4141 | } | |
4142 | ||
18bceab1 JA |
4143 | static void io_poll_remove_double(struct io_kiocb *req) |
4144 | { | |
4145 | struct io_poll_iocb *poll = (struct io_poll_iocb *) req->io; | |
4146 | ||
4147 | lockdep_assert_held(&req->ctx->completion_lock); | |
4148 | ||
4149 | if (poll && poll->head) { | |
4150 | struct wait_queue_head *head = poll->head; | |
4151 | ||
4152 | spin_lock(&head->lock); | |
4153 | list_del_init(&poll->wait.entry); | |
4154 | if (poll->wait.private) | |
4155 | refcount_dec(&req->refs); | |
4156 | poll->head = NULL; | |
4157 | spin_unlock(&head->lock); | |
4158 | } | |
4159 | } | |
4160 | ||
4161 | static void io_poll_complete(struct io_kiocb *req, __poll_t mask, int error) | |
4162 | { | |
4163 | struct io_ring_ctx *ctx = req->ctx; | |
4164 | ||
4165 | io_poll_remove_double(req); | |
4166 | req->poll.done = true; | |
4167 | io_cqring_fill_event(req, error ? error : mangle_poll(mask)); | |
4168 | io_commit_cqring(ctx); | |
4169 | } | |
4170 | ||
4171 | static void io_poll_task_handler(struct io_kiocb *req, struct io_kiocb **nxt) | |
4172 | { | |
4173 | struct io_ring_ctx *ctx = req->ctx; | |
4174 | ||
4175 | if (io_poll_rewait(req, &req->poll)) { | |
4176 | spin_unlock_irq(&ctx->completion_lock); | |
4177 | return; | |
4178 | } | |
4179 | ||
4180 | hash_del(&req->hash_node); | |
4181 | io_poll_complete(req, req->result, 0); | |
4182 | req->flags |= REQ_F_COMP_LOCKED; | |
4183 | io_put_req_find_next(req, nxt); | |
4184 | spin_unlock_irq(&ctx->completion_lock); | |
4185 | ||
4186 | io_cqring_ev_posted(ctx); | |
4187 | } | |
4188 | ||
4189 | static void io_poll_task_func(struct callback_head *cb) | |
4190 | { | |
4191 | struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work); | |
4192 | struct io_kiocb *nxt = NULL; | |
4193 | ||
4194 | io_poll_task_handler(req, &nxt); | |
4195 | if (nxt) { | |
4196 | struct io_ring_ctx *ctx = nxt->ctx; | |
4197 | ||
4198 | mutex_lock(&ctx->uring_lock); | |
4199 | __io_queue_sqe(nxt, NULL); | |
4200 | mutex_unlock(&ctx->uring_lock); | |
4201 | } | |
4202 | } | |
4203 | ||
4204 | static int io_poll_double_wake(struct wait_queue_entry *wait, unsigned mode, | |
4205 | int sync, void *key) | |
4206 | { | |
4207 | struct io_kiocb *req = wait->private; | |
4208 | struct io_poll_iocb *poll = (struct io_poll_iocb *) req->io; | |
4209 | __poll_t mask = key_to_poll(key); | |
4210 | ||
4211 | /* for instances that support it check for an event match first: */ | |
4212 | if (mask && !(mask & poll->events)) | |
4213 | return 0; | |
4214 | ||
4215 | if (req->poll.head) { | |
4216 | bool done; | |
4217 | ||
4218 | spin_lock(&req->poll.head->lock); | |
4219 | done = list_empty(&req->poll.wait.entry); | |
4220 | if (!done) | |
4221 | list_del_init(&req->poll.wait.entry); | |
4222 | spin_unlock(&req->poll.head->lock); | |
4223 | if (!done) | |
4224 | __io_async_wake(req, poll, mask, io_poll_task_func); | |
4225 | } | |
4226 | refcount_dec(&req->refs); | |
4227 | return 1; | |
4228 | } | |
4229 | ||
4230 | static void io_init_poll_iocb(struct io_poll_iocb *poll, __poll_t events, | |
4231 | wait_queue_func_t wake_func) | |
4232 | { | |
4233 | poll->head = NULL; | |
4234 | poll->done = false; | |
4235 | poll->canceled = false; | |
4236 | poll->events = events; | |
4237 | INIT_LIST_HEAD(&poll->wait.entry); | |
4238 | init_waitqueue_func_entry(&poll->wait, wake_func); | |
4239 | } | |
4240 | ||
4241 | static void __io_queue_proc(struct io_poll_iocb *poll, struct io_poll_table *pt, | |
4242 | struct wait_queue_head *head) | |
4243 | { | |
4244 | struct io_kiocb *req = pt->req; | |
4245 | ||
4246 | /* | |
4247 | * If poll->head is already set, it's because the file being polled | |
4248 | * uses multiple waitqueues for poll handling (eg one for read, one | |
4249 | * for write). Setup a separate io_poll_iocb if this happens. | |
4250 | */ | |
4251 | if (unlikely(poll->head)) { | |
4252 | /* already have a 2nd entry, fail a third attempt */ | |
4253 | if (req->io) { | |
4254 | pt->error = -EINVAL; | |
4255 | return; | |
4256 | } | |
4257 | poll = kmalloc(sizeof(*poll), GFP_ATOMIC); | |
4258 | if (!poll) { | |
4259 | pt->error = -ENOMEM; | |
4260 | return; | |
4261 | } | |
4262 | io_init_poll_iocb(poll, req->poll.events, io_poll_double_wake); | |
4263 | refcount_inc(&req->refs); | |
4264 | poll->wait.private = req; | |
4265 | req->io = (void *) poll; | |
4266 | } | |
4267 | ||
4268 | pt->error = 0; | |
4269 | poll->head = head; | |
4270 | add_wait_queue(head, &poll->wait); | |
4271 | } | |
4272 | ||
4273 | static void io_async_queue_proc(struct file *file, struct wait_queue_head *head, | |
4274 | struct poll_table_struct *p) | |
4275 | { | |
4276 | struct io_poll_table *pt = container_of(p, struct io_poll_table, pt); | |
4277 | ||
4278 | __io_queue_proc(&pt->req->apoll->poll, pt, head); | |
4279 | } | |
4280 | ||
9d8426a0 JA |
4281 | static void io_sq_thread_drop_mm(struct io_ring_ctx *ctx) |
4282 | { | |
4283 | struct mm_struct *mm = current->mm; | |
4284 | ||
4285 | if (mm) { | |
4286 | kthread_unuse_mm(mm); | |
4287 | mmput(mm); | |
4288 | } | |
4289 | } | |
4290 | ||
4291 | static int io_sq_thread_acquire_mm(struct io_ring_ctx *ctx, | |
4292 | struct io_kiocb *req) | |
4293 | { | |
4294 | if (io_op_defs[req->opcode].needs_mm && !current->mm) { | |
4295 | if (unlikely(!mmget_not_zero(ctx->sqo_mm))) | |
4296 | return -EFAULT; | |
4297 | kthread_use_mm(ctx->sqo_mm); | |
4298 | } | |
4299 | ||
4300 | return 0; | |
4301 | } | |
4302 | ||
d7718a9d JA |
4303 | static void io_async_task_func(struct callback_head *cb) |
4304 | { | |
4305 | struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work); | |
4306 | struct async_poll *apoll = req->apoll; | |
4307 | struct io_ring_ctx *ctx = req->ctx; | |
31067255 | 4308 | bool canceled = false; |
d7718a9d JA |
4309 | |
4310 | trace_io_uring_task_run(req->ctx, req->opcode, req->user_data); | |
4311 | ||
74ce6ce4 | 4312 | if (io_poll_rewait(req, &apoll->poll)) { |
d7718a9d | 4313 | spin_unlock_irq(&ctx->completion_lock); |
74ce6ce4 | 4314 | return; |
d7718a9d JA |
4315 | } |
4316 | ||
31067255 JA |
4317 | /* If req is still hashed, it cannot have been canceled. Don't check. */ |
4318 | if (hash_hashed(&req->hash_node)) { | |
74ce6ce4 | 4319 | hash_del(&req->hash_node); |
31067255 JA |
4320 | } else { |
4321 | canceled = READ_ONCE(apoll->poll.canceled); | |
4322 | if (canceled) { | |
4323 | io_cqring_fill_event(req, -ECANCELED); | |
4324 | io_commit_cqring(ctx); | |
4325 | } | |
2bae047e JA |
4326 | } |
4327 | ||
74ce6ce4 JA |
4328 | spin_unlock_irq(&ctx->completion_lock); |
4329 | ||
44575a67 | 4330 | /* restore ->work in case we need to retry again */ |
405a5d2b XW |
4331 | if (req->flags & REQ_F_WORK_INITIALIZED) |
4332 | memcpy(&req->work, &apoll->work, sizeof(req->work)); | |
31067255 | 4333 | kfree(apoll); |
44575a67 | 4334 | |
31067255 JA |
4335 | if (!canceled) { |
4336 | __set_current_state(TASK_RUNNING); | |
9d8426a0 JA |
4337 | if (io_sq_thread_acquire_mm(ctx, req)) { |
4338 | io_cqring_add_event(req, -EFAULT); | |
4339 | goto end_req; | |
4340 | } | |
31067255 JA |
4341 | mutex_lock(&ctx->uring_lock); |
4342 | __io_queue_sqe(req, NULL); | |
4343 | mutex_unlock(&ctx->uring_lock); | |
4344 | } else { | |
2bae047e | 4345 | io_cqring_ev_posted(ctx); |
9d8426a0 | 4346 | end_req: |
2bae047e | 4347 | req_set_fail_links(req); |
44575a67 | 4348 | io_double_put_req(req); |
2bae047e | 4349 | } |
d7718a9d JA |
4350 | } |
4351 | ||
4352 | static int io_async_wake(struct wait_queue_entry *wait, unsigned mode, int sync, | |
4353 | void *key) | |
4354 | { | |
4355 | struct io_kiocb *req = wait->private; | |
4356 | struct io_poll_iocb *poll = &req->apoll->poll; | |
4357 | ||
4358 | trace_io_uring_poll_wake(req->ctx, req->opcode, req->user_data, | |
4359 | key_to_poll(key)); | |
4360 | ||
4361 | return __io_async_wake(req, poll, key_to_poll(key), io_async_task_func); | |
4362 | } | |
4363 | ||
4364 | static void io_poll_req_insert(struct io_kiocb *req) | |
4365 | { | |
4366 | struct io_ring_ctx *ctx = req->ctx; | |
4367 | struct hlist_head *list; | |
4368 | ||
4369 | list = &ctx->cancel_hash[hash_long(req->user_data, ctx->cancel_hash_bits)]; | |
4370 | hlist_add_head(&req->hash_node, list); | |
4371 | } | |
4372 | ||
4373 | static __poll_t __io_arm_poll_handler(struct io_kiocb *req, | |
4374 | struct io_poll_iocb *poll, | |
4375 | struct io_poll_table *ipt, __poll_t mask, | |
4376 | wait_queue_func_t wake_func) | |
4377 | __acquires(&ctx->completion_lock) | |
4378 | { | |
4379 | struct io_ring_ctx *ctx = req->ctx; | |
4380 | bool cancel = false; | |
4381 | ||
4382 | poll->file = req->file; | |
18bceab1 JA |
4383 | io_init_poll_iocb(poll, mask, wake_func); |
4384 | poll->wait.private = req; | |
d7718a9d JA |
4385 | |
4386 | ipt->pt._key = mask; | |
4387 | ipt->req = req; | |
4388 | ipt->error = -EINVAL; | |
4389 | ||
d7718a9d JA |
4390 | mask = vfs_poll(req->file, &ipt->pt) & poll->events; |
4391 | ||
4392 | spin_lock_irq(&ctx->completion_lock); | |
4393 | if (likely(poll->head)) { | |
4394 | spin_lock(&poll->head->lock); | |
4395 | if (unlikely(list_empty(&poll->wait.entry))) { | |
4396 | if (ipt->error) | |
4397 | cancel = true; | |
4398 | ipt->error = 0; | |
4399 | mask = 0; | |
4400 | } | |
4401 | if (mask || ipt->error) | |
4402 | list_del_init(&poll->wait.entry); | |
4403 | else if (cancel) | |
4404 | WRITE_ONCE(poll->canceled, true); | |
4405 | else if (!poll->done) /* actually waiting for an event */ | |
4406 | io_poll_req_insert(req); | |
4407 | spin_unlock(&poll->head->lock); | |
4408 | } | |
4409 | ||
4410 | return mask; | |
4411 | } | |
4412 | ||
4413 | static bool io_arm_poll_handler(struct io_kiocb *req) | |
4414 | { | |
4415 | const struct io_op_def *def = &io_op_defs[req->opcode]; | |
4416 | struct io_ring_ctx *ctx = req->ctx; | |
4417 | struct async_poll *apoll; | |
4418 | struct io_poll_table ipt; | |
4419 | __poll_t mask, ret; | |
18bceab1 | 4420 | bool had_io; |
d7718a9d JA |
4421 | |
4422 | if (!req->file || !file_can_poll(req->file)) | |
4423 | return false; | |
4424 | if (req->flags & (REQ_F_MUST_PUNT | REQ_F_POLLED)) | |
4425 | return false; | |
4426 | if (!def->pollin && !def->pollout) | |
4427 | return false; | |
4428 | ||
4429 | apoll = kmalloc(sizeof(*apoll), GFP_ATOMIC); | |
4430 | if (unlikely(!apoll)) | |
4431 | return false; | |
4432 | ||
4433 | req->flags |= REQ_F_POLLED; | |
405a5d2b XW |
4434 | if (req->flags & REQ_F_WORK_INITIALIZED) |
4435 | memcpy(&apoll->work, &req->work, sizeof(req->work)); | |
18bceab1 | 4436 | had_io = req->io != NULL; |
d7718a9d | 4437 | |
4dd2824d | 4438 | io_get_req_task(req); |
d7718a9d JA |
4439 | req->apoll = apoll; |
4440 | INIT_HLIST_NODE(&req->hash_node); | |
4441 | ||
8755d97a | 4442 | mask = 0; |
d7718a9d | 4443 | if (def->pollin) |
8755d97a | 4444 | mask |= POLLIN | POLLRDNORM; |
d7718a9d JA |
4445 | if (def->pollout) |
4446 | mask |= POLLOUT | POLLWRNORM; | |
4447 | mask |= POLLERR | POLLPRI; | |
4448 | ||
4449 | ipt.pt._qproc = io_async_queue_proc; | |
4450 | ||
4451 | ret = __io_arm_poll_handler(req, &apoll->poll, &ipt, mask, | |
4452 | io_async_wake); | |
4453 | if (ret) { | |
4454 | ipt.error = 0; | |
18bceab1 JA |
4455 | /* only remove double add if we did it here */ |
4456 | if (!had_io) | |
4457 | io_poll_remove_double(req); | |
d7718a9d | 4458 | spin_unlock_irq(&ctx->completion_lock); |
405a5d2b XW |
4459 | if (req->flags & REQ_F_WORK_INITIALIZED) |
4460 | memcpy(&req->work, &apoll->work, sizeof(req->work)); | |
d7718a9d JA |
4461 | kfree(apoll); |
4462 | return false; | |
4463 | } | |
4464 | spin_unlock_irq(&ctx->completion_lock); | |
4465 | trace_io_uring_poll_arm(ctx, req->opcode, req->user_data, mask, | |
4466 | apoll->poll.events); | |
4467 | return true; | |
4468 | } | |
4469 | ||
4470 | static bool __io_poll_remove_one(struct io_kiocb *req, | |
4471 | struct io_poll_iocb *poll) | |
221c5eb2 | 4472 | { |
b41e9852 | 4473 | bool do_complete = false; |
221c5eb2 JA |
4474 | |
4475 | spin_lock(&poll->head->lock); | |
4476 | WRITE_ONCE(poll->canceled, true); | |
392edb45 JA |
4477 | if (!list_empty(&poll->wait.entry)) { |
4478 | list_del_init(&poll->wait.entry); | |
b41e9852 | 4479 | do_complete = true; |
221c5eb2 JA |
4480 | } |
4481 | spin_unlock(&poll->head->lock); | |
3bfa5bcb | 4482 | hash_del(&req->hash_node); |
d7718a9d JA |
4483 | return do_complete; |
4484 | } | |
4485 | ||
4486 | static bool io_poll_remove_one(struct io_kiocb *req) | |
4487 | { | |
4488 | bool do_complete; | |
4489 | ||
4490 | if (req->opcode == IORING_OP_POLL_ADD) { | |
18bceab1 | 4491 | io_poll_remove_double(req); |
d7718a9d JA |
4492 | do_complete = __io_poll_remove_one(req, &req->poll); |
4493 | } else { | |
3bfa5bcb JA |
4494 | struct async_poll *apoll = req->apoll; |
4495 | ||
d7718a9d | 4496 | /* non-poll requests have submit ref still */ |
3bfa5bcb JA |
4497 | do_complete = __io_poll_remove_one(req, &apoll->poll); |
4498 | if (do_complete) { | |
d7718a9d | 4499 | io_put_req(req); |
3bfa5bcb JA |
4500 | /* |
4501 | * restore ->work because we will call | |
4502 | * io_req_work_drop_env below when dropping the | |
4503 | * final reference. | |
4504 | */ | |
405a5d2b XW |
4505 | if (req->flags & REQ_F_WORK_INITIALIZED) |
4506 | memcpy(&req->work, &apoll->work, | |
4507 | sizeof(req->work)); | |
3bfa5bcb JA |
4508 | kfree(apoll); |
4509 | } | |
b1f573bd XW |
4510 | } |
4511 | ||
b41e9852 JA |
4512 | if (do_complete) { |
4513 | io_cqring_fill_event(req, -ECANCELED); | |
4514 | io_commit_cqring(req->ctx); | |
4515 | req->flags |= REQ_F_COMP_LOCKED; | |
4516 | io_put_req(req); | |
4517 | } | |
4518 | ||
4519 | return do_complete; | |
221c5eb2 JA |
4520 | } |
4521 | ||
4522 | static void io_poll_remove_all(struct io_ring_ctx *ctx) | |
4523 | { | |
78076bb6 | 4524 | struct hlist_node *tmp; |
221c5eb2 | 4525 | struct io_kiocb *req; |
8e2e1faf | 4526 | int posted = 0, i; |
221c5eb2 JA |
4527 | |
4528 | spin_lock_irq(&ctx->completion_lock); | |
78076bb6 JA |
4529 | for (i = 0; i < (1U << ctx->cancel_hash_bits); i++) { |
4530 | struct hlist_head *list; | |
4531 | ||
4532 | list = &ctx->cancel_hash[i]; | |
4533 | hlist_for_each_entry_safe(req, tmp, list, hash_node) | |
8e2e1faf | 4534 | posted += io_poll_remove_one(req); |
221c5eb2 JA |
4535 | } |
4536 | spin_unlock_irq(&ctx->completion_lock); | |
b41e9852 | 4537 | |
8e2e1faf JA |
4538 | if (posted) |
4539 | io_cqring_ev_posted(ctx); | |
221c5eb2 JA |
4540 | } |
4541 | ||
47f46768 JA |
4542 | static int io_poll_cancel(struct io_ring_ctx *ctx, __u64 sqe_addr) |
4543 | { | |
78076bb6 | 4544 | struct hlist_head *list; |
47f46768 JA |
4545 | struct io_kiocb *req; |
4546 | ||
78076bb6 JA |
4547 | list = &ctx->cancel_hash[hash_long(sqe_addr, ctx->cancel_hash_bits)]; |
4548 | hlist_for_each_entry(req, list, hash_node) { | |
b41e9852 JA |
4549 | if (sqe_addr != req->user_data) |
4550 | continue; | |
4551 | if (io_poll_remove_one(req)) | |
eac406c6 | 4552 | return 0; |
b41e9852 | 4553 | return -EALREADY; |
47f46768 JA |
4554 | } |
4555 | ||
4556 | return -ENOENT; | |
4557 | } | |
4558 | ||
3529d8c2 JA |
4559 | static int io_poll_remove_prep(struct io_kiocb *req, |
4560 | const struct io_uring_sqe *sqe) | |
0969e783 | 4561 | { |
0969e783 JA |
4562 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4563 | return -EINVAL; | |
4564 | if (sqe->ioprio || sqe->off || sqe->len || sqe->buf_index || | |
4565 | sqe->poll_events) | |
4566 | return -EINVAL; | |
4567 | ||
4568 | req->poll.addr = READ_ONCE(sqe->addr); | |
0969e783 JA |
4569 | return 0; |
4570 | } | |
4571 | ||
221c5eb2 JA |
4572 | /* |
4573 | * Find a running poll command that matches one specified in sqe->addr, | |
4574 | * and remove it if found. | |
4575 | */ | |
fc4df999 | 4576 | static int io_poll_remove(struct io_kiocb *req) |
221c5eb2 JA |
4577 | { |
4578 | struct io_ring_ctx *ctx = req->ctx; | |
0969e783 | 4579 | u64 addr; |
47f46768 | 4580 | int ret; |
221c5eb2 | 4581 | |
0969e783 | 4582 | addr = req->poll.addr; |
221c5eb2 | 4583 | spin_lock_irq(&ctx->completion_lock); |
0969e783 | 4584 | ret = io_poll_cancel(ctx, addr); |
221c5eb2 JA |
4585 | spin_unlock_irq(&ctx->completion_lock); |
4586 | ||
78e19bbe | 4587 | io_cqring_add_event(req, ret); |
4e88d6e7 JA |
4588 | if (ret < 0) |
4589 | req_set_fail_links(req); | |
e65ef56d | 4590 | io_put_req(req); |
221c5eb2 JA |
4591 | return 0; |
4592 | } | |
4593 | ||
221c5eb2 JA |
4594 | static int io_poll_wake(struct wait_queue_entry *wait, unsigned mode, int sync, |
4595 | void *key) | |
4596 | { | |
c2f2eb7d JA |
4597 | struct io_kiocb *req = wait->private; |
4598 | struct io_poll_iocb *poll = &req->poll; | |
221c5eb2 | 4599 | |
d7718a9d | 4600 | return __io_async_wake(req, poll, key_to_poll(key), io_poll_task_func); |
221c5eb2 JA |
4601 | } |
4602 | ||
221c5eb2 JA |
4603 | static void io_poll_queue_proc(struct file *file, struct wait_queue_head *head, |
4604 | struct poll_table_struct *p) | |
4605 | { | |
4606 | struct io_poll_table *pt = container_of(p, struct io_poll_table, pt); | |
4607 | ||
d7718a9d | 4608 | __io_queue_proc(&pt->req->poll, pt, head); |
eac406c6 JA |
4609 | } |
4610 | ||
3529d8c2 | 4611 | static int io_poll_add_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
221c5eb2 JA |
4612 | { |
4613 | struct io_poll_iocb *poll = &req->poll; | |
221c5eb2 | 4614 | u16 events; |
221c5eb2 JA |
4615 | |
4616 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) | |
4617 | return -EINVAL; | |
4618 | if (sqe->addr || sqe->ioprio || sqe->off || sqe->len || sqe->buf_index) | |
4619 | return -EINVAL; | |
09bb8394 JA |
4620 | if (!poll->file) |
4621 | return -EBADF; | |
221c5eb2 | 4622 | |
221c5eb2 JA |
4623 | events = READ_ONCE(sqe->poll_events); |
4624 | poll->events = demangle_poll(events) | EPOLLERR | EPOLLHUP; | |
b41e9852 | 4625 | |
4dd2824d | 4626 | io_get_req_task(req); |
0969e783 JA |
4627 | return 0; |
4628 | } | |
4629 | ||
014db007 | 4630 | static int io_poll_add(struct io_kiocb *req) |
0969e783 JA |
4631 | { |
4632 | struct io_poll_iocb *poll = &req->poll; | |
4633 | struct io_ring_ctx *ctx = req->ctx; | |
4634 | struct io_poll_table ipt; | |
0969e783 | 4635 | __poll_t mask; |
0969e783 | 4636 | |
78076bb6 | 4637 | INIT_HLIST_NODE(&req->hash_node); |
36703247 | 4638 | INIT_LIST_HEAD(&req->list); |
d7718a9d | 4639 | ipt.pt._qproc = io_poll_queue_proc; |
36703247 | 4640 | |
d7718a9d JA |
4641 | mask = __io_arm_poll_handler(req, &req->poll, &ipt, poll->events, |
4642 | io_poll_wake); | |
221c5eb2 | 4643 | |
8c838788 | 4644 | if (mask) { /* no async, we'd stolen it */ |
221c5eb2 | 4645 | ipt.error = 0; |
b0dd8a41 | 4646 | io_poll_complete(req, mask, 0); |
221c5eb2 | 4647 | } |
221c5eb2 JA |
4648 | spin_unlock_irq(&ctx->completion_lock); |
4649 | ||
8c838788 JA |
4650 | if (mask) { |
4651 | io_cqring_ev_posted(ctx); | |
014db007 | 4652 | io_put_req(req); |
221c5eb2 | 4653 | } |
8c838788 | 4654 | return ipt.error; |
221c5eb2 JA |
4655 | } |
4656 | ||
5262f567 JA |
4657 | static enum hrtimer_restart io_timeout_fn(struct hrtimer *timer) |
4658 | { | |
ad8a48ac JA |
4659 | struct io_timeout_data *data = container_of(timer, |
4660 | struct io_timeout_data, timer); | |
4661 | struct io_kiocb *req = data->req; | |
4662 | struct io_ring_ctx *ctx = req->ctx; | |
5262f567 JA |
4663 | unsigned long flags; |
4664 | ||
5262f567 JA |
4665 | atomic_inc(&ctx->cq_timeouts); |
4666 | ||
4667 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
ef03681a | 4668 | /* |
11365043 JA |
4669 | * We could be racing with timeout deletion. If the list is empty, |
4670 | * then timeout lookup already found it and will be handling it. | |
ef03681a | 4671 | */ |
bfe68a22 | 4672 | if (!list_empty(&req->list)) |
11365043 | 4673 | list_del_init(&req->list); |
5262f567 | 4674 | |
78e19bbe | 4675 | io_cqring_fill_event(req, -ETIME); |
5262f567 JA |
4676 | io_commit_cqring(ctx); |
4677 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
4678 | ||
4679 | io_cqring_ev_posted(ctx); | |
4e88d6e7 | 4680 | req_set_fail_links(req); |
5262f567 JA |
4681 | io_put_req(req); |
4682 | return HRTIMER_NORESTART; | |
4683 | } | |
4684 | ||
47f46768 JA |
4685 | static int io_timeout_cancel(struct io_ring_ctx *ctx, __u64 user_data) |
4686 | { | |
4687 | struct io_kiocb *req; | |
4688 | int ret = -ENOENT; | |
4689 | ||
4690 | list_for_each_entry(req, &ctx->timeout_list, list) { | |
4691 | if (user_data == req->user_data) { | |
4692 | list_del_init(&req->list); | |
4693 | ret = 0; | |
4694 | break; | |
4695 | } | |
4696 | } | |
4697 | ||
4698 | if (ret == -ENOENT) | |
4699 | return ret; | |
4700 | ||
2d28390a | 4701 | ret = hrtimer_try_to_cancel(&req->io->timeout.timer); |
47f46768 JA |
4702 | if (ret == -1) |
4703 | return -EALREADY; | |
4704 | ||
4e88d6e7 | 4705 | req_set_fail_links(req); |
47f46768 JA |
4706 | io_cqring_fill_event(req, -ECANCELED); |
4707 | io_put_req(req); | |
4708 | return 0; | |
4709 | } | |
4710 | ||
3529d8c2 JA |
4711 | static int io_timeout_remove_prep(struct io_kiocb *req, |
4712 | const struct io_uring_sqe *sqe) | |
b29472ee | 4713 | { |
b29472ee JA |
4714 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4715 | return -EINVAL; | |
4716 | if (sqe->flags || sqe->ioprio || sqe->buf_index || sqe->len) | |
4717 | return -EINVAL; | |
4718 | ||
4719 | req->timeout.addr = READ_ONCE(sqe->addr); | |
4720 | req->timeout.flags = READ_ONCE(sqe->timeout_flags); | |
4721 | if (req->timeout.flags) | |
4722 | return -EINVAL; | |
4723 | ||
b29472ee JA |
4724 | return 0; |
4725 | } | |
4726 | ||
11365043 JA |
4727 | /* |
4728 | * Remove or update an existing timeout command | |
4729 | */ | |
fc4df999 | 4730 | static int io_timeout_remove(struct io_kiocb *req) |
11365043 JA |
4731 | { |
4732 | struct io_ring_ctx *ctx = req->ctx; | |
47f46768 | 4733 | int ret; |
11365043 | 4734 | |
11365043 | 4735 | spin_lock_irq(&ctx->completion_lock); |
b29472ee | 4736 | ret = io_timeout_cancel(ctx, req->timeout.addr); |
11365043 | 4737 | |
47f46768 | 4738 | io_cqring_fill_event(req, ret); |
11365043 JA |
4739 | io_commit_cqring(ctx); |
4740 | spin_unlock_irq(&ctx->completion_lock); | |
5262f567 | 4741 | io_cqring_ev_posted(ctx); |
4e88d6e7 JA |
4742 | if (ret < 0) |
4743 | req_set_fail_links(req); | |
ec9c02ad | 4744 | io_put_req(req); |
11365043 | 4745 | return 0; |
5262f567 JA |
4746 | } |
4747 | ||
3529d8c2 | 4748 | static int io_timeout_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe, |
2d28390a | 4749 | bool is_timeout_link) |
5262f567 | 4750 | { |
ad8a48ac | 4751 | struct io_timeout_data *data; |
a41525ab | 4752 | unsigned flags; |
56080b02 | 4753 | u32 off = READ_ONCE(sqe->off); |
5262f567 | 4754 | |
ad8a48ac | 4755 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
5262f567 | 4756 | return -EINVAL; |
ad8a48ac | 4757 | if (sqe->ioprio || sqe->buf_index || sqe->len != 1) |
a41525ab | 4758 | return -EINVAL; |
56080b02 | 4759 | if (off && is_timeout_link) |
2d28390a | 4760 | return -EINVAL; |
a41525ab JA |
4761 | flags = READ_ONCE(sqe->timeout_flags); |
4762 | if (flags & ~IORING_TIMEOUT_ABS) | |
5262f567 | 4763 | return -EINVAL; |
bdf20073 | 4764 | |
bfe68a22 | 4765 | req->timeout.off = off; |
26a61679 | 4766 | |
3529d8c2 | 4767 | if (!req->io && io_alloc_async_ctx(req)) |
26a61679 JA |
4768 | return -ENOMEM; |
4769 | ||
4770 | data = &req->io->timeout; | |
ad8a48ac | 4771 | data->req = req; |
ad8a48ac JA |
4772 | req->flags |= REQ_F_TIMEOUT; |
4773 | ||
4774 | if (get_timespec64(&data->ts, u64_to_user_ptr(sqe->addr))) | |
5262f567 JA |
4775 | return -EFAULT; |
4776 | ||
11365043 | 4777 | if (flags & IORING_TIMEOUT_ABS) |
ad8a48ac | 4778 | data->mode = HRTIMER_MODE_ABS; |
11365043 | 4779 | else |
ad8a48ac | 4780 | data->mode = HRTIMER_MODE_REL; |
11365043 | 4781 | |
ad8a48ac JA |
4782 | hrtimer_init(&data->timer, CLOCK_MONOTONIC, data->mode); |
4783 | return 0; | |
4784 | } | |
4785 | ||
fc4df999 | 4786 | static int io_timeout(struct io_kiocb *req) |
ad8a48ac | 4787 | { |
ad8a48ac | 4788 | struct io_ring_ctx *ctx = req->ctx; |
bfe68a22 | 4789 | struct io_timeout_data *data = &req->io->timeout; |
ad8a48ac | 4790 | struct list_head *entry; |
bfe68a22 | 4791 | u32 tail, off = req->timeout.off; |
ad8a48ac | 4792 | |
733f5c95 | 4793 | spin_lock_irq(&ctx->completion_lock); |
93bd25bb | 4794 | |
5262f567 JA |
4795 | /* |
4796 | * sqe->off holds how many events that need to occur for this | |
93bd25bb JA |
4797 | * timeout event to be satisfied. If it isn't set, then this is |
4798 | * a pure timeout request, sequence isn't used. | |
5262f567 | 4799 | */ |
bfe68a22 | 4800 | if (!off) { |
93bd25bb | 4801 | req->flags |= REQ_F_TIMEOUT_NOSEQ; |
93bd25bb JA |
4802 | entry = ctx->timeout_list.prev; |
4803 | goto add; | |
4804 | } | |
5262f567 | 4805 | |
bfe68a22 PB |
4806 | tail = ctx->cached_cq_tail - atomic_read(&ctx->cq_timeouts); |
4807 | req->timeout.target_seq = tail + off; | |
5262f567 JA |
4808 | |
4809 | /* | |
4810 | * Insertion sort, ensuring the first entry in the list is always | |
4811 | * the one we need first. | |
4812 | */ | |
5262f567 JA |
4813 | list_for_each_prev(entry, &ctx->timeout_list) { |
4814 | struct io_kiocb *nxt = list_entry(entry, struct io_kiocb, list); | |
5262f567 | 4815 | |
93bd25bb JA |
4816 | if (nxt->flags & REQ_F_TIMEOUT_NOSEQ) |
4817 | continue; | |
bfe68a22 PB |
4818 | /* nxt.seq is behind @tail, otherwise would've been completed */ |
4819 | if (off >= nxt->timeout.target_seq - tail) | |
5262f567 JA |
4820 | break; |
4821 | } | |
93bd25bb | 4822 | add: |
5262f567 | 4823 | list_add(&req->list, entry); |
ad8a48ac JA |
4824 | data->timer.function = io_timeout_fn; |
4825 | hrtimer_start(&data->timer, timespec64_to_ktime(data->ts), data->mode); | |
5262f567 | 4826 | spin_unlock_irq(&ctx->completion_lock); |
5262f567 JA |
4827 | return 0; |
4828 | } | |
5262f567 | 4829 | |
62755e35 JA |
4830 | static bool io_cancel_cb(struct io_wq_work *work, void *data) |
4831 | { | |
4832 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); | |
4833 | ||
4834 | return req->user_data == (unsigned long) data; | |
4835 | } | |
4836 | ||
e977d6d3 | 4837 | static int io_async_cancel_one(struct io_ring_ctx *ctx, void *sqe_addr) |
62755e35 | 4838 | { |
62755e35 | 4839 | enum io_wq_cancel cancel_ret; |
62755e35 JA |
4840 | int ret = 0; |
4841 | ||
4f26bda1 | 4842 | cancel_ret = io_wq_cancel_cb(ctx->io_wq, io_cancel_cb, sqe_addr, false); |
62755e35 JA |
4843 | switch (cancel_ret) { |
4844 | case IO_WQ_CANCEL_OK: | |
4845 | ret = 0; | |
4846 | break; | |
4847 | case IO_WQ_CANCEL_RUNNING: | |
4848 | ret = -EALREADY; | |
4849 | break; | |
4850 | case IO_WQ_CANCEL_NOTFOUND: | |
4851 | ret = -ENOENT; | |
4852 | break; | |
4853 | } | |
4854 | ||
e977d6d3 JA |
4855 | return ret; |
4856 | } | |
4857 | ||
47f46768 JA |
4858 | static void io_async_find_and_cancel(struct io_ring_ctx *ctx, |
4859 | struct io_kiocb *req, __u64 sqe_addr, | |
014db007 | 4860 | int success_ret) |
47f46768 JA |
4861 | { |
4862 | unsigned long flags; | |
4863 | int ret; | |
4864 | ||
4865 | ret = io_async_cancel_one(ctx, (void *) (unsigned long) sqe_addr); | |
4866 | if (ret != -ENOENT) { | |
4867 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
4868 | goto done; | |
4869 | } | |
4870 | ||
4871 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
4872 | ret = io_timeout_cancel(ctx, sqe_addr); | |
4873 | if (ret != -ENOENT) | |
4874 | goto done; | |
4875 | ret = io_poll_cancel(ctx, sqe_addr); | |
4876 | done: | |
b0dd8a41 JA |
4877 | if (!ret) |
4878 | ret = success_ret; | |
47f46768 JA |
4879 | io_cqring_fill_event(req, ret); |
4880 | io_commit_cqring(ctx); | |
4881 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
4882 | io_cqring_ev_posted(ctx); | |
4883 | ||
4e88d6e7 JA |
4884 | if (ret < 0) |
4885 | req_set_fail_links(req); | |
014db007 | 4886 | io_put_req(req); |
47f46768 JA |
4887 | } |
4888 | ||
3529d8c2 JA |
4889 | static int io_async_cancel_prep(struct io_kiocb *req, |
4890 | const struct io_uring_sqe *sqe) | |
e977d6d3 | 4891 | { |
fbf23849 | 4892 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
e977d6d3 JA |
4893 | return -EINVAL; |
4894 | if (sqe->flags || sqe->ioprio || sqe->off || sqe->len || | |
4895 | sqe->cancel_flags) | |
4896 | return -EINVAL; | |
4897 | ||
fbf23849 JA |
4898 | req->cancel.addr = READ_ONCE(sqe->addr); |
4899 | return 0; | |
4900 | } | |
4901 | ||
014db007 | 4902 | static int io_async_cancel(struct io_kiocb *req) |
fbf23849 JA |
4903 | { |
4904 | struct io_ring_ctx *ctx = req->ctx; | |
fbf23849 | 4905 | |
014db007 | 4906 | io_async_find_and_cancel(ctx, req, req->cancel.addr, 0); |
5262f567 JA |
4907 | return 0; |
4908 | } | |
4909 | ||
05f3fb3c JA |
4910 | static int io_files_update_prep(struct io_kiocb *req, |
4911 | const struct io_uring_sqe *sqe) | |
4912 | { | |
4913 | if (sqe->flags || sqe->ioprio || sqe->rw_flags) | |
4914 | return -EINVAL; | |
4915 | ||
4916 | req->files_update.offset = READ_ONCE(sqe->off); | |
4917 | req->files_update.nr_args = READ_ONCE(sqe->len); | |
4918 | if (!req->files_update.nr_args) | |
4919 | return -EINVAL; | |
4920 | req->files_update.arg = READ_ONCE(sqe->addr); | |
4921 | return 0; | |
4922 | } | |
4923 | ||
4924 | static int io_files_update(struct io_kiocb *req, bool force_nonblock) | |
fbf23849 JA |
4925 | { |
4926 | struct io_ring_ctx *ctx = req->ctx; | |
05f3fb3c JA |
4927 | struct io_uring_files_update up; |
4928 | int ret; | |
fbf23849 | 4929 | |
f86cd20c | 4930 | if (force_nonblock) |
05f3fb3c | 4931 | return -EAGAIN; |
05f3fb3c JA |
4932 | |
4933 | up.offset = req->files_update.offset; | |
4934 | up.fds = req->files_update.arg; | |
4935 | ||
4936 | mutex_lock(&ctx->uring_lock); | |
4937 | ret = __io_sqe_files_update(ctx, &up, req->files_update.nr_args); | |
4938 | mutex_unlock(&ctx->uring_lock); | |
4939 | ||
4940 | if (ret < 0) | |
4941 | req_set_fail_links(req); | |
4942 | io_cqring_add_event(req, ret); | |
4943 | io_put_req(req); | |
5262f567 JA |
4944 | return 0; |
4945 | } | |
4946 | ||
3529d8c2 JA |
4947 | static int io_req_defer_prep(struct io_kiocb *req, |
4948 | const struct io_uring_sqe *sqe) | |
f67676d1 | 4949 | { |
e781573e | 4950 | ssize_t ret = 0; |
f67676d1 | 4951 | |
f1d96a8f PB |
4952 | if (!sqe) |
4953 | return 0; | |
4954 | ||
7cdaf587 XW |
4955 | io_req_init_async(req); |
4956 | ||
f86cd20c JA |
4957 | if (io_op_defs[req->opcode].file_table) { |
4958 | ret = io_grab_files(req); | |
4959 | if (unlikely(ret)) | |
4960 | return ret; | |
4961 | } | |
4962 | ||
cccf0ee8 JA |
4963 | io_req_work_grab_env(req, &io_op_defs[req->opcode]); |
4964 | ||
d625c6ee | 4965 | switch (req->opcode) { |
e781573e JA |
4966 | case IORING_OP_NOP: |
4967 | break; | |
f67676d1 JA |
4968 | case IORING_OP_READV: |
4969 | case IORING_OP_READ_FIXED: | |
3a6820f2 | 4970 | case IORING_OP_READ: |
3529d8c2 | 4971 | ret = io_read_prep(req, sqe, true); |
f67676d1 JA |
4972 | break; |
4973 | case IORING_OP_WRITEV: | |
4974 | case IORING_OP_WRITE_FIXED: | |
3a6820f2 | 4975 | case IORING_OP_WRITE: |
3529d8c2 | 4976 | ret = io_write_prep(req, sqe, true); |
f67676d1 | 4977 | break; |
0969e783 | 4978 | case IORING_OP_POLL_ADD: |
3529d8c2 | 4979 | ret = io_poll_add_prep(req, sqe); |
0969e783 JA |
4980 | break; |
4981 | case IORING_OP_POLL_REMOVE: | |
3529d8c2 | 4982 | ret = io_poll_remove_prep(req, sqe); |
0969e783 | 4983 | break; |
8ed8d3c3 | 4984 | case IORING_OP_FSYNC: |
3529d8c2 | 4985 | ret = io_prep_fsync(req, sqe); |
8ed8d3c3 JA |
4986 | break; |
4987 | case IORING_OP_SYNC_FILE_RANGE: | |
3529d8c2 | 4988 | ret = io_prep_sfr(req, sqe); |
8ed8d3c3 | 4989 | break; |
03b1230c | 4990 | case IORING_OP_SENDMSG: |
fddaface | 4991 | case IORING_OP_SEND: |
3529d8c2 | 4992 | ret = io_sendmsg_prep(req, sqe); |
03b1230c JA |
4993 | break; |
4994 | case IORING_OP_RECVMSG: | |
fddaface | 4995 | case IORING_OP_RECV: |
3529d8c2 | 4996 | ret = io_recvmsg_prep(req, sqe); |
03b1230c | 4997 | break; |
f499a021 | 4998 | case IORING_OP_CONNECT: |
3529d8c2 | 4999 | ret = io_connect_prep(req, sqe); |
f499a021 | 5000 | break; |
2d28390a | 5001 | case IORING_OP_TIMEOUT: |
3529d8c2 | 5002 | ret = io_timeout_prep(req, sqe, false); |
b7bb4f7d | 5003 | break; |
b29472ee | 5004 | case IORING_OP_TIMEOUT_REMOVE: |
3529d8c2 | 5005 | ret = io_timeout_remove_prep(req, sqe); |
b29472ee | 5006 | break; |
fbf23849 | 5007 | case IORING_OP_ASYNC_CANCEL: |
3529d8c2 | 5008 | ret = io_async_cancel_prep(req, sqe); |
fbf23849 | 5009 | break; |
2d28390a | 5010 | case IORING_OP_LINK_TIMEOUT: |
3529d8c2 | 5011 | ret = io_timeout_prep(req, sqe, true); |
b7bb4f7d | 5012 | break; |
8ed8d3c3 | 5013 | case IORING_OP_ACCEPT: |
3529d8c2 | 5014 | ret = io_accept_prep(req, sqe); |
8ed8d3c3 | 5015 | break; |
d63d1b5e JA |
5016 | case IORING_OP_FALLOCATE: |
5017 | ret = io_fallocate_prep(req, sqe); | |
5018 | break; | |
15b71abe JA |
5019 | case IORING_OP_OPENAT: |
5020 | ret = io_openat_prep(req, sqe); | |
5021 | break; | |
b5dba59e JA |
5022 | case IORING_OP_CLOSE: |
5023 | ret = io_close_prep(req, sqe); | |
5024 | break; | |
05f3fb3c JA |
5025 | case IORING_OP_FILES_UPDATE: |
5026 | ret = io_files_update_prep(req, sqe); | |
5027 | break; | |
eddc7ef5 JA |
5028 | case IORING_OP_STATX: |
5029 | ret = io_statx_prep(req, sqe); | |
5030 | break; | |
4840e418 JA |
5031 | case IORING_OP_FADVISE: |
5032 | ret = io_fadvise_prep(req, sqe); | |
5033 | break; | |
c1ca757b JA |
5034 | case IORING_OP_MADVISE: |
5035 | ret = io_madvise_prep(req, sqe); | |
5036 | break; | |
cebdb986 JA |
5037 | case IORING_OP_OPENAT2: |
5038 | ret = io_openat2_prep(req, sqe); | |
5039 | break; | |
3e4827b0 JA |
5040 | case IORING_OP_EPOLL_CTL: |
5041 | ret = io_epoll_ctl_prep(req, sqe); | |
5042 | break; | |
7d67af2c PB |
5043 | case IORING_OP_SPLICE: |
5044 | ret = io_splice_prep(req, sqe); | |
5045 | break; | |
ddf0322d JA |
5046 | case IORING_OP_PROVIDE_BUFFERS: |
5047 | ret = io_provide_buffers_prep(req, sqe); | |
5048 | break; | |
067524e9 JA |
5049 | case IORING_OP_REMOVE_BUFFERS: |
5050 | ret = io_remove_buffers_prep(req, sqe); | |
5051 | break; | |
f2a8d5c7 PB |
5052 | case IORING_OP_TEE: |
5053 | ret = io_tee_prep(req, sqe); | |
5054 | break; | |
f67676d1 | 5055 | default: |
e781573e JA |
5056 | printk_once(KERN_WARNING "io_uring: unhandled opcode %d\n", |
5057 | req->opcode); | |
5058 | ret = -EINVAL; | |
b7bb4f7d | 5059 | break; |
f67676d1 JA |
5060 | } |
5061 | ||
b7bb4f7d | 5062 | return ret; |
f67676d1 JA |
5063 | } |
5064 | ||
3529d8c2 | 5065 | static int io_req_defer(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
de0617e4 | 5066 | { |
a197f664 | 5067 | struct io_ring_ctx *ctx = req->ctx; |
f67676d1 | 5068 | int ret; |
de0617e4 | 5069 | |
9d858b21 | 5070 | /* Still need defer if there is pending req in defer list. */ |
4ee36314 | 5071 | if (!req_need_defer(req) && list_empty_careful(&ctx->defer_list)) |
de0617e4 JA |
5072 | return 0; |
5073 | ||
650b5481 PB |
5074 | if (!req->io) { |
5075 | if (io_alloc_async_ctx(req)) | |
5076 | return -EAGAIN; | |
5077 | ret = io_req_defer_prep(req, sqe); | |
5078 | if (ret < 0) | |
5079 | return ret; | |
5080 | } | |
2d28390a | 5081 | |
de0617e4 | 5082 | spin_lock_irq(&ctx->completion_lock); |
9d858b21 | 5083 | if (!req_need_defer(req) && list_empty(&ctx->defer_list)) { |
de0617e4 | 5084 | spin_unlock_irq(&ctx->completion_lock); |
de0617e4 JA |
5085 | return 0; |
5086 | } | |
5087 | ||
915967f6 | 5088 | trace_io_uring_defer(ctx, req, req->user_data); |
de0617e4 JA |
5089 | list_add_tail(&req->list, &ctx->defer_list); |
5090 | spin_unlock_irq(&ctx->completion_lock); | |
5091 | return -EIOCBQUEUED; | |
5092 | } | |
5093 | ||
99bc4c38 PB |
5094 | static void io_cleanup_req(struct io_kiocb *req) |
5095 | { | |
5096 | struct io_async_ctx *io = req->io; | |
5097 | ||
5098 | switch (req->opcode) { | |
5099 | case IORING_OP_READV: | |
5100 | case IORING_OP_READ_FIXED: | |
5101 | case IORING_OP_READ: | |
bcda7baa JA |
5102 | if (req->flags & REQ_F_BUFFER_SELECTED) |
5103 | kfree((void *)(unsigned long)req->rw.addr); | |
5104 | /* fallthrough */ | |
99bc4c38 PB |
5105 | case IORING_OP_WRITEV: |
5106 | case IORING_OP_WRITE_FIXED: | |
5107 | case IORING_OP_WRITE: | |
5108 | if (io->rw.iov != io->rw.fast_iov) | |
5109 | kfree(io->rw.iov); | |
5110 | break; | |
99bc4c38 | 5111 | case IORING_OP_RECVMSG: |
52de1fe1 JA |
5112 | if (req->flags & REQ_F_BUFFER_SELECTED) |
5113 | kfree(req->sr_msg.kbuf); | |
5114 | /* fallthrough */ | |
5115 | case IORING_OP_SENDMSG: | |
99bc4c38 PB |
5116 | if (io->msg.iov != io->msg.fast_iov) |
5117 | kfree(io->msg.iov); | |
5118 | break; | |
bcda7baa JA |
5119 | case IORING_OP_RECV: |
5120 | if (req->flags & REQ_F_BUFFER_SELECTED) | |
5121 | kfree(req->sr_msg.kbuf); | |
5122 | break; | |
8fef80bf PB |
5123 | case IORING_OP_OPENAT: |
5124 | case IORING_OP_OPENAT2: | |
8fef80bf | 5125 | break; |
7d67af2c | 5126 | case IORING_OP_SPLICE: |
f2a8d5c7 | 5127 | case IORING_OP_TEE: |
7d67af2c PB |
5128 | io_put_file(req, req->splice.file_in, |
5129 | (req->splice.flags & SPLICE_F_FD_IN_FIXED)); | |
5130 | break; | |
99bc4c38 PB |
5131 | } |
5132 | ||
5133 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
5134 | } | |
5135 | ||
3529d8c2 | 5136 | static int io_issue_sqe(struct io_kiocb *req, const struct io_uring_sqe *sqe, |
014db007 | 5137 | bool force_nonblock) |
2b188cc1 | 5138 | { |
a197f664 | 5139 | struct io_ring_ctx *ctx = req->ctx; |
d625c6ee | 5140 | int ret; |
2b188cc1 | 5141 | |
d625c6ee | 5142 | switch (req->opcode) { |
2b188cc1 | 5143 | case IORING_OP_NOP: |
78e19bbe | 5144 | ret = io_nop(req); |
2b188cc1 JA |
5145 | break; |
5146 | case IORING_OP_READV: | |
edafccee | 5147 | case IORING_OP_READ_FIXED: |
3a6820f2 | 5148 | case IORING_OP_READ: |
3529d8c2 JA |
5149 | if (sqe) { |
5150 | ret = io_read_prep(req, sqe, force_nonblock); | |
5151 | if (ret < 0) | |
5152 | break; | |
5153 | } | |
014db007 | 5154 | ret = io_read(req, force_nonblock); |
edafccee | 5155 | break; |
3529d8c2 | 5156 | case IORING_OP_WRITEV: |
edafccee | 5157 | case IORING_OP_WRITE_FIXED: |
3a6820f2 | 5158 | case IORING_OP_WRITE: |
3529d8c2 JA |
5159 | if (sqe) { |
5160 | ret = io_write_prep(req, sqe, force_nonblock); | |
5161 | if (ret < 0) | |
5162 | break; | |
5163 | } | |
014db007 | 5164 | ret = io_write(req, force_nonblock); |
2b188cc1 | 5165 | break; |
c992fe29 | 5166 | case IORING_OP_FSYNC: |
3529d8c2 JA |
5167 | if (sqe) { |
5168 | ret = io_prep_fsync(req, sqe); | |
5169 | if (ret < 0) | |
5170 | break; | |
5171 | } | |
014db007 | 5172 | ret = io_fsync(req, force_nonblock); |
c992fe29 | 5173 | break; |
221c5eb2 | 5174 | case IORING_OP_POLL_ADD: |
3529d8c2 JA |
5175 | if (sqe) { |
5176 | ret = io_poll_add_prep(req, sqe); | |
5177 | if (ret) | |
5178 | break; | |
5179 | } | |
014db007 | 5180 | ret = io_poll_add(req); |
221c5eb2 JA |
5181 | break; |
5182 | case IORING_OP_POLL_REMOVE: | |
3529d8c2 JA |
5183 | if (sqe) { |
5184 | ret = io_poll_remove_prep(req, sqe); | |
5185 | if (ret < 0) | |
5186 | break; | |
5187 | } | |
fc4df999 | 5188 | ret = io_poll_remove(req); |
221c5eb2 | 5189 | break; |
5d17b4a4 | 5190 | case IORING_OP_SYNC_FILE_RANGE: |
3529d8c2 JA |
5191 | if (sqe) { |
5192 | ret = io_prep_sfr(req, sqe); | |
5193 | if (ret < 0) | |
5194 | break; | |
5195 | } | |
014db007 | 5196 | ret = io_sync_file_range(req, force_nonblock); |
5d17b4a4 | 5197 | break; |
0fa03c62 | 5198 | case IORING_OP_SENDMSG: |
fddaface | 5199 | case IORING_OP_SEND: |
3529d8c2 JA |
5200 | if (sqe) { |
5201 | ret = io_sendmsg_prep(req, sqe); | |
5202 | if (ret < 0) | |
5203 | break; | |
5204 | } | |
fddaface | 5205 | if (req->opcode == IORING_OP_SENDMSG) |
014db007 | 5206 | ret = io_sendmsg(req, force_nonblock); |
fddaface | 5207 | else |
014db007 | 5208 | ret = io_send(req, force_nonblock); |
0fa03c62 | 5209 | break; |
aa1fa28f | 5210 | case IORING_OP_RECVMSG: |
fddaface | 5211 | case IORING_OP_RECV: |
3529d8c2 JA |
5212 | if (sqe) { |
5213 | ret = io_recvmsg_prep(req, sqe); | |
5214 | if (ret) | |
5215 | break; | |
5216 | } | |
fddaface | 5217 | if (req->opcode == IORING_OP_RECVMSG) |
014db007 | 5218 | ret = io_recvmsg(req, force_nonblock); |
fddaface | 5219 | else |
014db007 | 5220 | ret = io_recv(req, force_nonblock); |
aa1fa28f | 5221 | break; |
5262f567 | 5222 | case IORING_OP_TIMEOUT: |
3529d8c2 JA |
5223 | if (sqe) { |
5224 | ret = io_timeout_prep(req, sqe, false); | |
5225 | if (ret) | |
5226 | break; | |
5227 | } | |
fc4df999 | 5228 | ret = io_timeout(req); |
5262f567 | 5229 | break; |
11365043 | 5230 | case IORING_OP_TIMEOUT_REMOVE: |
3529d8c2 JA |
5231 | if (sqe) { |
5232 | ret = io_timeout_remove_prep(req, sqe); | |
5233 | if (ret) | |
5234 | break; | |
5235 | } | |
fc4df999 | 5236 | ret = io_timeout_remove(req); |
11365043 | 5237 | break; |
17f2fe35 | 5238 | case IORING_OP_ACCEPT: |
3529d8c2 JA |
5239 | if (sqe) { |
5240 | ret = io_accept_prep(req, sqe); | |
5241 | if (ret) | |
5242 | break; | |
5243 | } | |
014db007 | 5244 | ret = io_accept(req, force_nonblock); |
17f2fe35 | 5245 | break; |
f8e85cf2 | 5246 | case IORING_OP_CONNECT: |
3529d8c2 JA |
5247 | if (sqe) { |
5248 | ret = io_connect_prep(req, sqe); | |
5249 | if (ret) | |
5250 | break; | |
5251 | } | |
014db007 | 5252 | ret = io_connect(req, force_nonblock); |
f8e85cf2 | 5253 | break; |
62755e35 | 5254 | case IORING_OP_ASYNC_CANCEL: |
3529d8c2 JA |
5255 | if (sqe) { |
5256 | ret = io_async_cancel_prep(req, sqe); | |
5257 | if (ret) | |
5258 | break; | |
5259 | } | |
014db007 | 5260 | ret = io_async_cancel(req); |
62755e35 | 5261 | break; |
d63d1b5e JA |
5262 | case IORING_OP_FALLOCATE: |
5263 | if (sqe) { | |
5264 | ret = io_fallocate_prep(req, sqe); | |
5265 | if (ret) | |
5266 | break; | |
5267 | } | |
014db007 | 5268 | ret = io_fallocate(req, force_nonblock); |
d63d1b5e | 5269 | break; |
15b71abe JA |
5270 | case IORING_OP_OPENAT: |
5271 | if (sqe) { | |
5272 | ret = io_openat_prep(req, sqe); | |
5273 | if (ret) | |
5274 | break; | |
5275 | } | |
014db007 | 5276 | ret = io_openat(req, force_nonblock); |
15b71abe | 5277 | break; |
b5dba59e JA |
5278 | case IORING_OP_CLOSE: |
5279 | if (sqe) { | |
5280 | ret = io_close_prep(req, sqe); | |
5281 | if (ret) | |
5282 | break; | |
5283 | } | |
014db007 | 5284 | ret = io_close(req, force_nonblock); |
b5dba59e | 5285 | break; |
05f3fb3c JA |
5286 | case IORING_OP_FILES_UPDATE: |
5287 | if (sqe) { | |
5288 | ret = io_files_update_prep(req, sqe); | |
5289 | if (ret) | |
5290 | break; | |
5291 | } | |
5292 | ret = io_files_update(req, force_nonblock); | |
5293 | break; | |
eddc7ef5 JA |
5294 | case IORING_OP_STATX: |
5295 | if (sqe) { | |
5296 | ret = io_statx_prep(req, sqe); | |
5297 | if (ret) | |
5298 | break; | |
5299 | } | |
014db007 | 5300 | ret = io_statx(req, force_nonblock); |
eddc7ef5 | 5301 | break; |
4840e418 JA |
5302 | case IORING_OP_FADVISE: |
5303 | if (sqe) { | |
5304 | ret = io_fadvise_prep(req, sqe); | |
5305 | if (ret) | |
5306 | break; | |
5307 | } | |
014db007 | 5308 | ret = io_fadvise(req, force_nonblock); |
4840e418 | 5309 | break; |
c1ca757b JA |
5310 | case IORING_OP_MADVISE: |
5311 | if (sqe) { | |
5312 | ret = io_madvise_prep(req, sqe); | |
5313 | if (ret) | |
5314 | break; | |
5315 | } | |
014db007 | 5316 | ret = io_madvise(req, force_nonblock); |
c1ca757b | 5317 | break; |
cebdb986 JA |
5318 | case IORING_OP_OPENAT2: |
5319 | if (sqe) { | |
5320 | ret = io_openat2_prep(req, sqe); | |
5321 | if (ret) | |
5322 | break; | |
5323 | } | |
014db007 | 5324 | ret = io_openat2(req, force_nonblock); |
cebdb986 | 5325 | break; |
3e4827b0 JA |
5326 | case IORING_OP_EPOLL_CTL: |
5327 | if (sqe) { | |
5328 | ret = io_epoll_ctl_prep(req, sqe); | |
5329 | if (ret) | |
5330 | break; | |
5331 | } | |
014db007 | 5332 | ret = io_epoll_ctl(req, force_nonblock); |
3e4827b0 | 5333 | break; |
7d67af2c PB |
5334 | case IORING_OP_SPLICE: |
5335 | if (sqe) { | |
5336 | ret = io_splice_prep(req, sqe); | |
5337 | if (ret < 0) | |
5338 | break; | |
5339 | } | |
014db007 | 5340 | ret = io_splice(req, force_nonblock); |
7d67af2c | 5341 | break; |
ddf0322d JA |
5342 | case IORING_OP_PROVIDE_BUFFERS: |
5343 | if (sqe) { | |
5344 | ret = io_provide_buffers_prep(req, sqe); | |
5345 | if (ret) | |
5346 | break; | |
5347 | } | |
5348 | ret = io_provide_buffers(req, force_nonblock); | |
5349 | break; | |
067524e9 JA |
5350 | case IORING_OP_REMOVE_BUFFERS: |
5351 | if (sqe) { | |
5352 | ret = io_remove_buffers_prep(req, sqe); | |
5353 | if (ret) | |
5354 | break; | |
5355 | } | |
5356 | ret = io_remove_buffers(req, force_nonblock); | |
3e4827b0 | 5357 | break; |
f2a8d5c7 PB |
5358 | case IORING_OP_TEE: |
5359 | if (sqe) { | |
5360 | ret = io_tee_prep(req, sqe); | |
5361 | if (ret < 0) | |
5362 | break; | |
5363 | } | |
5364 | ret = io_tee(req, force_nonblock); | |
5365 | break; | |
2b188cc1 JA |
5366 | default: |
5367 | ret = -EINVAL; | |
5368 | break; | |
5369 | } | |
5370 | ||
def596e9 JA |
5371 | if (ret) |
5372 | return ret; | |
5373 | ||
b532576e JA |
5374 | /* If the op doesn't have a file, we're not polling for it */ |
5375 | if ((ctx->flags & IORING_SETUP_IOPOLL) && req->file) { | |
11ba820b JA |
5376 | const bool in_async = io_wq_current_is_worker(); |
5377 | ||
11ba820b JA |
5378 | /* workqueue context doesn't hold uring_lock, grab it now */ |
5379 | if (in_async) | |
5380 | mutex_lock(&ctx->uring_lock); | |
5381 | ||
def596e9 | 5382 | io_iopoll_req_issued(req); |
11ba820b JA |
5383 | |
5384 | if (in_async) | |
5385 | mutex_unlock(&ctx->uring_lock); | |
def596e9 JA |
5386 | } |
5387 | ||
5388 | return 0; | |
2b188cc1 JA |
5389 | } |
5390 | ||
d4c81f38 PB |
5391 | static void io_arm_async_linked_timeout(struct io_kiocb *req) |
5392 | { | |
5393 | struct io_kiocb *link; | |
5394 | ||
5395 | /* link head's timeout is queued in io_queue_async_work() */ | |
5396 | if (!(req->flags & REQ_F_QUEUE_TIMEOUT)) | |
5397 | return; | |
5398 | ||
5399 | link = list_first_entry(&req->link_list, struct io_kiocb, link_list); | |
5400 | io_queue_linked_timeout(link); | |
5401 | } | |
5402 | ||
561fb04a | 5403 | static void io_wq_submit_work(struct io_wq_work **workptr) |
2b188cc1 | 5404 | { |
561fb04a | 5405 | struct io_wq_work *work = *workptr; |
2b188cc1 | 5406 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); |
561fb04a | 5407 | int ret = 0; |
2b188cc1 | 5408 | |
d4c81f38 PB |
5409 | io_arm_async_linked_timeout(req); |
5410 | ||
0c9d5ccd JA |
5411 | /* if NO_CANCEL is set, we must still run the work */ |
5412 | if ((work->flags & (IO_WQ_WORK_CANCEL|IO_WQ_WORK_NO_CANCEL)) == | |
5413 | IO_WQ_WORK_CANCEL) { | |
561fb04a | 5414 | ret = -ECANCELED; |
0c9d5ccd | 5415 | } |
31b51510 | 5416 | |
561fb04a | 5417 | if (!ret) { |
561fb04a | 5418 | do { |
014db007 | 5419 | ret = io_issue_sqe(req, NULL, false); |
561fb04a JA |
5420 | /* |
5421 | * We can get EAGAIN for polled IO even though we're | |
5422 | * forcing a sync submission from here, since we can't | |
5423 | * wait for request slots on the block side. | |
5424 | */ | |
5425 | if (ret != -EAGAIN) | |
5426 | break; | |
5427 | cond_resched(); | |
5428 | } while (1); | |
5429 | } | |
31b51510 | 5430 | |
561fb04a | 5431 | if (ret) { |
4e88d6e7 | 5432 | req_set_fail_links(req); |
78e19bbe | 5433 | io_cqring_add_event(req, ret); |
817869d2 | 5434 | io_put_req(req); |
edafccee | 5435 | } |
2b188cc1 | 5436 | |
e9fd9396 | 5437 | io_steal_work(req, workptr); |
2b188cc1 JA |
5438 | } |
5439 | ||
65e19f54 JA |
5440 | static inline struct file *io_file_from_index(struct io_ring_ctx *ctx, |
5441 | int index) | |
5442 | { | |
5443 | struct fixed_file_table *table; | |
5444 | ||
05f3fb3c | 5445 | table = &ctx->file_data->table[index >> IORING_FILE_TABLE_SHIFT]; |
84695089 | 5446 | return table->files[index & IORING_FILE_TABLE_MASK]; |
65e19f54 JA |
5447 | } |
5448 | ||
8da11c19 PB |
5449 | static int io_file_get(struct io_submit_state *state, struct io_kiocb *req, |
5450 | int fd, struct file **out_file, bool fixed) | |
09bb8394 | 5451 | { |
a197f664 | 5452 | struct io_ring_ctx *ctx = req->ctx; |
8da11c19 | 5453 | struct file *file; |
09bb8394 | 5454 | |
8da11c19 | 5455 | if (fixed) { |
05f3fb3c | 5456 | if (unlikely(!ctx->file_data || |
09bb8394 JA |
5457 | (unsigned) fd >= ctx->nr_user_files)) |
5458 | return -EBADF; | |
b7620121 | 5459 | fd = array_index_nospec(fd, ctx->nr_user_files); |
8da11c19 | 5460 | file = io_file_from_index(ctx, fd); |
fd2206e4 JA |
5461 | if (file) { |
5462 | req->fixed_file_refs = ctx->file_data->cur_refs; | |
5463 | percpu_ref_get(req->fixed_file_refs); | |
5464 | } | |
09bb8394 | 5465 | } else { |
c826bd7a | 5466 | trace_io_uring_file_get(ctx, fd); |
8da11c19 | 5467 | file = __io_file_get(state, fd); |
09bb8394 JA |
5468 | } |
5469 | ||
fd2206e4 JA |
5470 | if (file || io_op_defs[req->opcode].needs_file_no_error) { |
5471 | *out_file = file; | |
5472 | return 0; | |
5473 | } | |
5474 | return -EBADF; | |
09bb8394 JA |
5475 | } |
5476 | ||
8da11c19 | 5477 | static int io_req_set_file(struct io_submit_state *state, struct io_kiocb *req, |
63ff8223 | 5478 | int fd) |
8da11c19 | 5479 | { |
8da11c19 PB |
5480 | bool fixed; |
5481 | ||
63ff8223 | 5482 | fixed = (req->flags & REQ_F_FIXED_FILE) != 0; |
0cdaf760 | 5483 | if (unlikely(!fixed && io_async_submit(req->ctx))) |
8da11c19 PB |
5484 | return -EBADF; |
5485 | ||
5486 | return io_file_get(state, req, fd, &req->file, fixed); | |
5487 | } | |
5488 | ||
a197f664 | 5489 | static int io_grab_files(struct io_kiocb *req) |
fcb323cc JA |
5490 | { |
5491 | int ret = -EBADF; | |
a197f664 | 5492 | struct io_ring_ctx *ctx = req->ctx; |
fcb323cc | 5493 | |
5b0bbee4 | 5494 | if (req->work.files || (req->flags & REQ_F_NO_FILE_TABLE)) |
f86cd20c | 5495 | return 0; |
b14cca0c | 5496 | if (!ctx->ring_file) |
b5dba59e JA |
5497 | return -EBADF; |
5498 | ||
fcb323cc JA |
5499 | rcu_read_lock(); |
5500 | spin_lock_irq(&ctx->inflight_lock); | |
5501 | /* | |
5502 | * We use the f_ops->flush() handler to ensure that we can flush | |
5503 | * out work accessing these files if the fd is closed. Check if | |
5504 | * the fd has changed since we started down this path, and disallow | |
5505 | * this operation if it has. | |
5506 | */ | |
b14cca0c | 5507 | if (fcheck(ctx->ring_fd) == ctx->ring_file) { |
fcb323cc JA |
5508 | list_add(&req->inflight_entry, &ctx->inflight_list); |
5509 | req->flags |= REQ_F_INFLIGHT; | |
5510 | req->work.files = current->files; | |
5511 | ret = 0; | |
5512 | } | |
5513 | spin_unlock_irq(&ctx->inflight_lock); | |
5514 | rcu_read_unlock(); | |
5515 | ||
5516 | return ret; | |
5517 | } | |
5518 | ||
2665abfd | 5519 | static enum hrtimer_restart io_link_timeout_fn(struct hrtimer *timer) |
2b188cc1 | 5520 | { |
ad8a48ac JA |
5521 | struct io_timeout_data *data = container_of(timer, |
5522 | struct io_timeout_data, timer); | |
5523 | struct io_kiocb *req = data->req; | |
2665abfd JA |
5524 | struct io_ring_ctx *ctx = req->ctx; |
5525 | struct io_kiocb *prev = NULL; | |
5526 | unsigned long flags; | |
2665abfd JA |
5527 | |
5528 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
5529 | ||
5530 | /* | |
5531 | * We don't expect the list to be empty, that will only happen if we | |
5532 | * race with the completion of the linked work. | |
5533 | */ | |
4493233e PB |
5534 | if (!list_empty(&req->link_list)) { |
5535 | prev = list_entry(req->link_list.prev, struct io_kiocb, | |
5536 | link_list); | |
5d960724 | 5537 | if (refcount_inc_not_zero(&prev->refs)) { |
4493233e | 5538 | list_del_init(&req->link_list); |
5d960724 JA |
5539 | prev->flags &= ~REQ_F_LINK_TIMEOUT; |
5540 | } else | |
76a46e06 | 5541 | prev = NULL; |
2665abfd JA |
5542 | } |
5543 | ||
5544 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
5545 | ||
5546 | if (prev) { | |
4e88d6e7 | 5547 | req_set_fail_links(prev); |
014db007 | 5548 | io_async_find_and_cancel(ctx, req, prev->user_data, -ETIME); |
76a46e06 | 5549 | io_put_req(prev); |
47f46768 JA |
5550 | } else { |
5551 | io_cqring_add_event(req, -ETIME); | |
5552 | io_put_req(req); | |
2665abfd | 5553 | } |
2665abfd JA |
5554 | return HRTIMER_NORESTART; |
5555 | } | |
5556 | ||
ad8a48ac | 5557 | static void io_queue_linked_timeout(struct io_kiocb *req) |
2665abfd | 5558 | { |
76a46e06 | 5559 | struct io_ring_ctx *ctx = req->ctx; |
2665abfd | 5560 | |
76a46e06 JA |
5561 | /* |
5562 | * If the list is now empty, then our linked request finished before | |
5563 | * we got a chance to setup the timer | |
5564 | */ | |
5565 | spin_lock_irq(&ctx->completion_lock); | |
4493233e | 5566 | if (!list_empty(&req->link_list)) { |
2d28390a | 5567 | struct io_timeout_data *data = &req->io->timeout; |
94ae5e77 | 5568 | |
ad8a48ac JA |
5569 | data->timer.function = io_link_timeout_fn; |
5570 | hrtimer_start(&data->timer, timespec64_to_ktime(data->ts), | |
5571 | data->mode); | |
2665abfd | 5572 | } |
76a46e06 | 5573 | spin_unlock_irq(&ctx->completion_lock); |
2665abfd | 5574 | |
2665abfd | 5575 | /* drop submission reference */ |
76a46e06 JA |
5576 | io_put_req(req); |
5577 | } | |
2665abfd | 5578 | |
ad8a48ac | 5579 | static struct io_kiocb *io_prep_linked_timeout(struct io_kiocb *req) |
2665abfd JA |
5580 | { |
5581 | struct io_kiocb *nxt; | |
5582 | ||
dea3b49c | 5583 | if (!(req->flags & REQ_F_LINK_HEAD)) |
2665abfd | 5584 | return NULL; |
d7718a9d JA |
5585 | /* for polled retry, if flag is set, we already went through here */ |
5586 | if (req->flags & REQ_F_POLLED) | |
5587 | return NULL; | |
2665abfd | 5588 | |
4493233e PB |
5589 | nxt = list_first_entry_or_null(&req->link_list, struct io_kiocb, |
5590 | link_list); | |
d625c6ee | 5591 | if (!nxt || nxt->opcode != IORING_OP_LINK_TIMEOUT) |
76a46e06 | 5592 | return NULL; |
2665abfd | 5593 | |
76a46e06 | 5594 | req->flags |= REQ_F_LINK_TIMEOUT; |
76a46e06 | 5595 | return nxt; |
2665abfd JA |
5596 | } |
5597 | ||
3529d8c2 | 5598 | static void __io_queue_sqe(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
2b188cc1 | 5599 | { |
4a0a7a18 | 5600 | struct io_kiocb *linked_timeout; |
4bc4494e | 5601 | struct io_kiocb *nxt; |
193155c8 | 5602 | const struct cred *old_creds = NULL; |
e0c5c576 | 5603 | int ret; |
2b188cc1 | 5604 | |
4a0a7a18 JA |
5605 | again: |
5606 | linked_timeout = io_prep_linked_timeout(req); | |
5607 | ||
7cdaf587 XW |
5608 | if ((req->flags & REQ_F_WORK_INITIALIZED) && req->work.creds && |
5609 | req->work.creds != current_cred()) { | |
193155c8 JA |
5610 | if (old_creds) |
5611 | revert_creds(old_creds); | |
5612 | if (old_creds == req->work.creds) | |
5613 | old_creds = NULL; /* restored original creds */ | |
5614 | else | |
5615 | old_creds = override_creds(req->work.creds); | |
5616 | } | |
5617 | ||
014db007 | 5618 | ret = io_issue_sqe(req, sqe, true); |
491381ce JA |
5619 | |
5620 | /* | |
5621 | * We async punt it if the file wasn't marked NOWAIT, or if the file | |
5622 | * doesn't support non-blocking read/write attempts | |
5623 | */ | |
5624 | if (ret == -EAGAIN && (!(req->flags & REQ_F_NOWAIT) || | |
5625 | (req->flags & REQ_F_MUST_PUNT))) { | |
d7718a9d JA |
5626 | if (io_arm_poll_handler(req)) { |
5627 | if (linked_timeout) | |
5628 | io_queue_linked_timeout(linked_timeout); | |
4bc4494e | 5629 | goto exit; |
d7718a9d | 5630 | } |
86a761f8 | 5631 | punt: |
7cdaf587 XW |
5632 | io_req_init_async(req); |
5633 | ||
f86cd20c | 5634 | if (io_op_defs[req->opcode].file_table) { |
bbad27b2 PB |
5635 | ret = io_grab_files(req); |
5636 | if (ret) | |
5637 | goto err; | |
2b188cc1 | 5638 | } |
bbad27b2 PB |
5639 | |
5640 | /* | |
5641 | * Queued up for async execution, worker will release | |
5642 | * submit reference when the iocb is actually submitted. | |
5643 | */ | |
5644 | io_queue_async_work(req); | |
4bc4494e | 5645 | goto exit; |
2b188cc1 | 5646 | } |
e65ef56d | 5647 | |
fcb323cc | 5648 | err: |
4bc4494e | 5649 | nxt = NULL; |
76a46e06 | 5650 | /* drop submission reference */ |
2a44f467 | 5651 | io_put_req_find_next(req, &nxt); |
e65ef56d | 5652 | |
f9bd67f6 | 5653 | if (linked_timeout) { |
76a46e06 | 5654 | if (!ret) |
f9bd67f6 | 5655 | io_queue_linked_timeout(linked_timeout); |
76a46e06 | 5656 | else |
f9bd67f6 | 5657 | io_put_req(linked_timeout); |
76a46e06 JA |
5658 | } |
5659 | ||
e65ef56d | 5660 | /* and drop final reference, if we failed */ |
9e645e11 | 5661 | if (ret) { |
78e19bbe | 5662 | io_cqring_add_event(req, ret); |
4e88d6e7 | 5663 | req_set_fail_links(req); |
e65ef56d | 5664 | io_put_req(req); |
9e645e11 | 5665 | } |
4a0a7a18 JA |
5666 | if (nxt) { |
5667 | req = nxt; | |
86a761f8 PB |
5668 | |
5669 | if (req->flags & REQ_F_FORCE_ASYNC) | |
5670 | goto punt; | |
4a0a7a18 JA |
5671 | goto again; |
5672 | } | |
4bc4494e | 5673 | exit: |
193155c8 JA |
5674 | if (old_creds) |
5675 | revert_creds(old_creds); | |
2b188cc1 JA |
5676 | } |
5677 | ||
3529d8c2 | 5678 | static void io_queue_sqe(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
4fe2c963 JL |
5679 | { |
5680 | int ret; | |
5681 | ||
3529d8c2 | 5682 | ret = io_req_defer(req, sqe); |
4fe2c963 JL |
5683 | if (ret) { |
5684 | if (ret != -EIOCBQUEUED) { | |
1118591a | 5685 | fail_req: |
78e19bbe | 5686 | io_cqring_add_event(req, ret); |
4e88d6e7 | 5687 | req_set_fail_links(req); |
78e19bbe | 5688 | io_double_put_req(req); |
4fe2c963 | 5689 | } |
2550878f | 5690 | } else if (req->flags & REQ_F_FORCE_ASYNC) { |
bd2ab18a PB |
5691 | if (!req->io) { |
5692 | ret = -EAGAIN; | |
5693 | if (io_alloc_async_ctx(req)) | |
5694 | goto fail_req; | |
5695 | ret = io_req_defer_prep(req, sqe); | |
5696 | if (unlikely(ret < 0)) | |
5697 | goto fail_req; | |
5698 | } | |
5699 | ||
ce35a47a JA |
5700 | /* |
5701 | * Never try inline submit of IOSQE_ASYNC is set, go straight | |
5702 | * to async execution. | |
5703 | */ | |
5704 | req->work.flags |= IO_WQ_WORK_CONCURRENT; | |
5705 | io_queue_async_work(req); | |
5706 | } else { | |
3529d8c2 | 5707 | __io_queue_sqe(req, sqe); |
ce35a47a | 5708 | } |
4fe2c963 JL |
5709 | } |
5710 | ||
1b4a51b6 | 5711 | static inline void io_queue_link_head(struct io_kiocb *req) |
4fe2c963 | 5712 | { |
94ae5e77 | 5713 | if (unlikely(req->flags & REQ_F_FAIL_LINK)) { |
1b4a51b6 PB |
5714 | io_cqring_add_event(req, -ECANCELED); |
5715 | io_double_put_req(req); | |
5716 | } else | |
3529d8c2 | 5717 | io_queue_sqe(req, NULL); |
4fe2c963 JL |
5718 | } |
5719 | ||
1d4240cc | 5720 | static int io_submit_sqe(struct io_kiocb *req, const struct io_uring_sqe *sqe, |
7d01bd74 | 5721 | struct io_kiocb **link) |
9e645e11 | 5722 | { |
a197f664 | 5723 | struct io_ring_ctx *ctx = req->ctx; |
ef4ff581 | 5724 | int ret; |
9e645e11 | 5725 | |
9e645e11 JA |
5726 | /* |
5727 | * If we already have a head request, queue this one for async | |
5728 | * submittal once the head completes. If we don't have a head but | |
5729 | * IOSQE_IO_LINK is set in the sqe, start a new head. This one will be | |
5730 | * submitted sync once the chain is complete. If none of those | |
5731 | * conditions are true (normal request), then just queue it. | |
5732 | */ | |
5733 | if (*link) { | |
9d76377f | 5734 | struct io_kiocb *head = *link; |
4e88d6e7 | 5735 | |
8cdf2193 PB |
5736 | /* |
5737 | * Taking sequential execution of a link, draining both sides | |
5738 | * of the link also fullfils IOSQE_IO_DRAIN semantics for all | |
5739 | * requests in the link. So, it drains the head and the | |
5740 | * next after the link request. The last one is done via | |
5741 | * drain_next flag to persist the effect across calls. | |
5742 | */ | |
ef4ff581 | 5743 | if (req->flags & REQ_F_IO_DRAIN) { |
711be031 PB |
5744 | head->flags |= REQ_F_IO_DRAIN; |
5745 | ctx->drain_next = 1; | |
5746 | } | |
1d4240cc PB |
5747 | if (io_alloc_async_ctx(req)) |
5748 | return -EAGAIN; | |
9e645e11 | 5749 | |
3529d8c2 | 5750 | ret = io_req_defer_prep(req, sqe); |
2d28390a | 5751 | if (ret) { |
4e88d6e7 | 5752 | /* fail even hard links since we don't submit */ |
9d76377f | 5753 | head->flags |= REQ_F_FAIL_LINK; |
1d4240cc | 5754 | return ret; |
2d28390a | 5755 | } |
9d76377f PB |
5756 | trace_io_uring_link(ctx, req, head); |
5757 | list_add_tail(&req->link_list, &head->link_list); | |
32fe525b PB |
5758 | |
5759 | /* last request of a link, enqueue the link */ | |
ef4ff581 | 5760 | if (!(req->flags & (REQ_F_LINK | REQ_F_HARDLINK))) { |
32fe525b PB |
5761 | io_queue_link_head(head); |
5762 | *link = NULL; | |
5763 | } | |
9e645e11 | 5764 | } else { |
711be031 PB |
5765 | if (unlikely(ctx->drain_next)) { |
5766 | req->flags |= REQ_F_IO_DRAIN; | |
ef4ff581 | 5767 | ctx->drain_next = 0; |
711be031 | 5768 | } |
ef4ff581 | 5769 | if (req->flags & (REQ_F_LINK | REQ_F_HARDLINK)) { |
dea3b49c | 5770 | req->flags |= REQ_F_LINK_HEAD; |
711be031 | 5771 | INIT_LIST_HEAD(&req->link_list); |
f1d96a8f | 5772 | |
1d4240cc PB |
5773 | if (io_alloc_async_ctx(req)) |
5774 | return -EAGAIN; | |
5775 | ||
711be031 PB |
5776 | ret = io_req_defer_prep(req, sqe); |
5777 | if (ret) | |
5778 | req->flags |= REQ_F_FAIL_LINK; | |
5779 | *link = req; | |
5780 | } else { | |
5781 | io_queue_sqe(req, sqe); | |
5782 | } | |
9e645e11 | 5783 | } |
2e6e1fde | 5784 | |
1d4240cc | 5785 | return 0; |
9e645e11 JA |
5786 | } |
5787 | ||
9a56a232 JA |
5788 | /* |
5789 | * Batched submission is done, ensure local IO is flushed out. | |
5790 | */ | |
5791 | static void io_submit_state_end(struct io_submit_state *state) | |
5792 | { | |
5793 | blk_finish_plug(&state->plug); | |
9f13c35b | 5794 | io_state_file_put(state); |
2579f913 | 5795 | if (state->free_reqs) |
6c8a3134 | 5796 | kmem_cache_free_bulk(req_cachep, state->free_reqs, state->reqs); |
9a56a232 JA |
5797 | } |
5798 | ||
5799 | /* | |
5800 | * Start submission side cache. | |
5801 | */ | |
5802 | static void io_submit_state_start(struct io_submit_state *state, | |
22efde59 | 5803 | unsigned int max_ios) |
9a56a232 JA |
5804 | { |
5805 | blk_start_plug(&state->plug); | |
2579f913 | 5806 | state->free_reqs = 0; |
9a56a232 JA |
5807 | state->file = NULL; |
5808 | state->ios_left = max_ios; | |
5809 | } | |
5810 | ||
2b188cc1 JA |
5811 | static void io_commit_sqring(struct io_ring_ctx *ctx) |
5812 | { | |
75b28aff | 5813 | struct io_rings *rings = ctx->rings; |
2b188cc1 | 5814 | |
caf582c6 PB |
5815 | /* |
5816 | * Ensure any loads from the SQEs are done at this point, | |
5817 | * since once we write the new head, the application could | |
5818 | * write new data to them. | |
5819 | */ | |
5820 | smp_store_release(&rings->sq.head, ctx->cached_sq_head); | |
2b188cc1 JA |
5821 | } |
5822 | ||
2b188cc1 | 5823 | /* |
3529d8c2 | 5824 | * Fetch an sqe, if one is available. Note that sqe_ptr will point to memory |
2b188cc1 JA |
5825 | * that is mapped by userspace. This means that care needs to be taken to |
5826 | * ensure that reads are stable, as we cannot rely on userspace always | |
5827 | * being a good citizen. If members of the sqe are validated and then later | |
5828 | * used, it's important that those reads are done through READ_ONCE() to | |
5829 | * prevent a re-load down the line. | |
5830 | */ | |
709b302f | 5831 | static const struct io_uring_sqe *io_get_sqe(struct io_ring_ctx *ctx) |
2b188cc1 | 5832 | { |
75b28aff | 5833 | u32 *sq_array = ctx->sq_array; |
2b188cc1 JA |
5834 | unsigned head; |
5835 | ||
5836 | /* | |
5837 | * The cached sq head (or cq tail) serves two purposes: | |
5838 | * | |
5839 | * 1) allows us to batch the cost of updating the user visible | |
5840 | * head updates. | |
5841 | * 2) allows the kernel side to track the head on its own, even | |
5842 | * though the application is the one updating it. | |
5843 | */ | |
ee7d46d9 | 5844 | head = READ_ONCE(sq_array[ctx->cached_sq_head & ctx->sq_mask]); |
709b302f PB |
5845 | if (likely(head < ctx->sq_entries)) |
5846 | return &ctx->sq_sqes[head]; | |
2b188cc1 JA |
5847 | |
5848 | /* drop invalid entries */ | |
498ccd9e | 5849 | ctx->cached_sq_dropped++; |
ee7d46d9 | 5850 | WRITE_ONCE(ctx->rings->sq_dropped, ctx->cached_sq_dropped); |
709b302f PB |
5851 | return NULL; |
5852 | } | |
5853 | ||
5854 | static inline void io_consume_sqe(struct io_ring_ctx *ctx) | |
5855 | { | |
5856 | ctx->cached_sq_head++; | |
2b188cc1 JA |
5857 | } |
5858 | ||
ef4ff581 PB |
5859 | #define SQE_VALID_FLAGS (IOSQE_FIXED_FILE|IOSQE_IO_DRAIN|IOSQE_IO_LINK| \ |
5860 | IOSQE_IO_HARDLINK | IOSQE_ASYNC | \ | |
5861 | IOSQE_BUFFER_SELECT) | |
5862 | ||
5863 | static int io_init_req(struct io_ring_ctx *ctx, struct io_kiocb *req, | |
5864 | const struct io_uring_sqe *sqe, | |
0cdaf760 | 5865 | struct io_submit_state *state) |
0553b8bd | 5866 | { |
ef4ff581 | 5867 | unsigned int sqe_flags; |
63ff8223 | 5868 | int id; |
ef4ff581 | 5869 | |
0553b8bd PB |
5870 | /* |
5871 | * All io need record the previous position, if LINK vs DARIN, | |
5872 | * it can be used to mark the position of the first IO in the | |
5873 | * link list. | |
5874 | */ | |
31af27c7 | 5875 | req->sequence = ctx->cached_sq_head - ctx->cached_sq_dropped; |
0553b8bd PB |
5876 | req->opcode = READ_ONCE(sqe->opcode); |
5877 | req->user_data = READ_ONCE(sqe->user_data); | |
5878 | req->io = NULL; | |
5879 | req->file = NULL; | |
5880 | req->ctx = ctx; | |
5881 | req->flags = 0; | |
5882 | /* one is dropped after submission, the other at completion */ | |
5883 | refcount_set(&req->refs, 2); | |
4dd2824d | 5884 | req->task = current; |
0553b8bd | 5885 | req->result = 0; |
ef4ff581 PB |
5886 | |
5887 | if (unlikely(req->opcode >= IORING_OP_LAST)) | |
5888 | return -EINVAL; | |
5889 | ||
9d8426a0 JA |
5890 | if (unlikely(io_sq_thread_acquire_mm(ctx, req))) |
5891 | return -EFAULT; | |
ef4ff581 PB |
5892 | |
5893 | sqe_flags = READ_ONCE(sqe->flags); | |
5894 | /* enforce forwards compatibility on users */ | |
5895 | if (unlikely(sqe_flags & ~SQE_VALID_FLAGS)) | |
5896 | return -EINVAL; | |
5897 | ||
5898 | if ((sqe_flags & IOSQE_BUFFER_SELECT) && | |
5899 | !io_op_defs[req->opcode].buffer_select) | |
5900 | return -EOPNOTSUPP; | |
5901 | ||
5902 | id = READ_ONCE(sqe->personality); | |
5903 | if (id) { | |
7cdaf587 | 5904 | io_req_init_async(req); |
ef4ff581 PB |
5905 | req->work.creds = idr_find(&ctx->personality_idr, id); |
5906 | if (unlikely(!req->work.creds)) | |
5907 | return -EINVAL; | |
5908 | get_cred(req->work.creds); | |
5909 | } | |
5910 | ||
5911 | /* same numerical values with corresponding REQ_F_*, safe to copy */ | |
c11368a5 | 5912 | req->flags |= sqe_flags; |
ef4ff581 | 5913 | |
63ff8223 JA |
5914 | if (!io_op_defs[req->opcode].needs_file) |
5915 | return 0; | |
5916 | ||
5917 | return io_req_set_file(state, req, READ_ONCE(sqe->fd)); | |
0553b8bd PB |
5918 | } |
5919 | ||
fb5ccc98 | 5920 | static int io_submit_sqes(struct io_ring_ctx *ctx, unsigned int nr, |
0cdaf760 | 5921 | struct file *ring_file, int ring_fd) |
6c271ce2 JA |
5922 | { |
5923 | struct io_submit_state state, *statep = NULL; | |
9e645e11 | 5924 | struct io_kiocb *link = NULL; |
9e645e11 | 5925 | int i, submitted = 0; |
6c271ce2 | 5926 | |
c4a2ed72 | 5927 | /* if we have a backlog and couldn't flush it all, return BUSY */ |
ad3eb2c8 JA |
5928 | if (test_bit(0, &ctx->sq_check_overflow)) { |
5929 | if (!list_empty(&ctx->cq_overflow_list) && | |
5930 | !io_cqring_overflow_flush(ctx, false)) | |
5931 | return -EBUSY; | |
5932 | } | |
6c271ce2 | 5933 | |
ee7d46d9 PB |
5934 | /* make sure SQ entry isn't read before tail */ |
5935 | nr = min3(nr, ctx->sq_entries, io_sqring_entries(ctx)); | |
9ef4f124 | 5936 | |
2b85edfc PB |
5937 | if (!percpu_ref_tryget_many(&ctx->refs, nr)) |
5938 | return -EAGAIN; | |
6c271ce2 JA |
5939 | |
5940 | if (nr > IO_PLUG_THRESHOLD) { | |
22efde59 | 5941 | io_submit_state_start(&state, nr); |
6c271ce2 JA |
5942 | statep = &state; |
5943 | } | |
5944 | ||
b14cca0c PB |
5945 | ctx->ring_fd = ring_fd; |
5946 | ctx->ring_file = ring_file; | |
5947 | ||
6c271ce2 | 5948 | for (i = 0; i < nr; i++) { |
3529d8c2 | 5949 | const struct io_uring_sqe *sqe; |
196be95c | 5950 | struct io_kiocb *req; |
1cb1edb2 | 5951 | int err; |
fb5ccc98 | 5952 | |
b1e50e54 PB |
5953 | sqe = io_get_sqe(ctx); |
5954 | if (unlikely(!sqe)) { | |
5955 | io_consume_sqe(ctx); | |
5956 | break; | |
5957 | } | |
0553b8bd | 5958 | req = io_alloc_req(ctx, statep); |
196be95c PB |
5959 | if (unlikely(!req)) { |
5960 | if (!submitted) | |
5961 | submitted = -EAGAIN; | |
fb5ccc98 | 5962 | break; |
196be95c | 5963 | } |
fb5ccc98 | 5964 | |
0cdaf760 | 5965 | err = io_init_req(ctx, req, sqe, statep); |
709b302f | 5966 | io_consume_sqe(ctx); |
d3656344 JA |
5967 | /* will complete beyond this point, count as submitted */ |
5968 | submitted++; | |
5969 | ||
ef4ff581 | 5970 | if (unlikely(err)) { |
1cb1edb2 PB |
5971 | fail_req: |
5972 | io_cqring_add_event(req, err); | |
d3656344 | 5973 | io_double_put_req(req); |
196be95c PB |
5974 | break; |
5975 | } | |
fb5ccc98 | 5976 | |
354420f7 | 5977 | trace_io_uring_submit_sqe(ctx, req->opcode, req->user_data, |
0cdaf760 | 5978 | true, io_async_submit(ctx)); |
7d01bd74 | 5979 | err = io_submit_sqe(req, sqe, &link); |
1d4240cc PB |
5980 | if (err) |
5981 | goto fail_req; | |
6c271ce2 JA |
5982 | } |
5983 | ||
9466f437 PB |
5984 | if (unlikely(submitted != nr)) { |
5985 | int ref_used = (submitted == -EAGAIN) ? 0 : submitted; | |
5986 | ||
5987 | percpu_ref_put_many(&ctx->refs, nr - ref_used); | |
5988 | } | |
9e645e11 | 5989 | if (link) |
1b4a51b6 | 5990 | io_queue_link_head(link); |
6c271ce2 JA |
5991 | if (statep) |
5992 | io_submit_state_end(&state); | |
5993 | ||
ae9428ca PB |
5994 | /* Commit SQ ring head once we've consumed and submitted all SQEs */ |
5995 | io_commit_sqring(ctx); | |
5996 | ||
6c271ce2 JA |
5997 | return submitted; |
5998 | } | |
5999 | ||
6000 | static int io_sq_thread(void *data) | |
6001 | { | |
6c271ce2 | 6002 | struct io_ring_ctx *ctx = data; |
181e448d | 6003 | const struct cred *old_cred; |
6c271ce2 | 6004 | DEFINE_WAIT(wait); |
6c271ce2 | 6005 | unsigned long timeout; |
bdcd3eab | 6006 | int ret = 0; |
6c271ce2 | 6007 | |
0f158b4c | 6008 | complete(&ctx->sq_thread_comp); |
a4c0b3de | 6009 | |
181e448d | 6010 | old_cred = override_creds(ctx->creds); |
6c271ce2 | 6011 | |
bdcd3eab | 6012 | timeout = jiffies + ctx->sq_thread_idle; |
2bbcd6d3 | 6013 | while (!kthread_should_park()) { |
fb5ccc98 | 6014 | unsigned int to_submit; |
6c271ce2 | 6015 | |
bdcd3eab | 6016 | if (!list_empty(&ctx->poll_list)) { |
6c271ce2 JA |
6017 | unsigned nr_events = 0; |
6018 | ||
bdcd3eab XW |
6019 | mutex_lock(&ctx->uring_lock); |
6020 | if (!list_empty(&ctx->poll_list)) | |
6021 | io_iopoll_getevents(ctx, &nr_events, 0); | |
6022 | else | |
6c271ce2 | 6023 | timeout = jiffies + ctx->sq_thread_idle; |
bdcd3eab | 6024 | mutex_unlock(&ctx->uring_lock); |
6c271ce2 JA |
6025 | } |
6026 | ||
fb5ccc98 | 6027 | to_submit = io_sqring_entries(ctx); |
c1edbf5f JA |
6028 | |
6029 | /* | |
6030 | * If submit got -EBUSY, flag us as needing the application | |
6031 | * to enter the kernel to reap and flush events. | |
6032 | */ | |
b772f07a | 6033 | if (!to_submit || ret == -EBUSY || need_resched()) { |
7143b5ac SG |
6034 | /* |
6035 | * Drop cur_mm before scheduling, we can't hold it for | |
6036 | * long periods (or over schedule()). Do this before | |
6037 | * adding ourselves to the waitqueue, as the unuse/drop | |
6038 | * may sleep. | |
6039 | */ | |
bf9c2f1c | 6040 | io_sq_thread_drop_mm(ctx); |
7143b5ac | 6041 | |
6c271ce2 JA |
6042 | /* |
6043 | * We're polling. If we're within the defined idle | |
6044 | * period, then let us spin without work before going | |
c1edbf5f JA |
6045 | * to sleep. The exception is if we got EBUSY doing |
6046 | * more IO, we should wait for the application to | |
6047 | * reap events and wake us up. | |
6c271ce2 | 6048 | */ |
b772f07a | 6049 | if (!list_empty(&ctx->poll_list) || need_resched() || |
df069d80 JA |
6050 | (!time_after(jiffies, timeout) && ret != -EBUSY && |
6051 | !percpu_ref_is_dying(&ctx->refs))) { | |
b41e9852 JA |
6052 | if (current->task_works) |
6053 | task_work_run(); | |
9831a90c | 6054 | cond_resched(); |
6c271ce2 JA |
6055 | continue; |
6056 | } | |
6057 | ||
6c271ce2 JA |
6058 | prepare_to_wait(&ctx->sqo_wait, &wait, |
6059 | TASK_INTERRUPTIBLE); | |
6060 | ||
bdcd3eab XW |
6061 | /* |
6062 | * While doing polled IO, before going to sleep, we need | |
6063 | * to check if there are new reqs added to poll_list, it | |
6064 | * is because reqs may have been punted to io worker and | |
6065 | * will be added to poll_list later, hence check the | |
6066 | * poll_list again. | |
6067 | */ | |
6068 | if ((ctx->flags & IORING_SETUP_IOPOLL) && | |
6069 | !list_empty_careful(&ctx->poll_list)) { | |
6070 | finish_wait(&ctx->sqo_wait, &wait); | |
6071 | continue; | |
6072 | } | |
6073 | ||
6c271ce2 | 6074 | /* Tell userspace we may need a wakeup call */ |
75b28aff | 6075 | ctx->rings->sq_flags |= IORING_SQ_NEED_WAKEUP; |
0d7bae69 SB |
6076 | /* make sure to read SQ tail after writing flags */ |
6077 | smp_mb(); | |
6c271ce2 | 6078 | |
fb5ccc98 | 6079 | to_submit = io_sqring_entries(ctx); |
c1edbf5f | 6080 | if (!to_submit || ret == -EBUSY) { |
2bbcd6d3 | 6081 | if (kthread_should_park()) { |
6c271ce2 JA |
6082 | finish_wait(&ctx->sqo_wait, &wait); |
6083 | break; | |
6084 | } | |
b41e9852 JA |
6085 | if (current->task_works) { |
6086 | task_work_run(); | |
10bea96d | 6087 | finish_wait(&ctx->sqo_wait, &wait); |
b41e9852 JA |
6088 | continue; |
6089 | } | |
6c271ce2 JA |
6090 | if (signal_pending(current)) |
6091 | flush_signals(current); | |
6092 | schedule(); | |
6093 | finish_wait(&ctx->sqo_wait, &wait); | |
6094 | ||
75b28aff | 6095 | ctx->rings->sq_flags &= ~IORING_SQ_NEED_WAKEUP; |
d4ae271d | 6096 | ret = 0; |
6c271ce2 JA |
6097 | continue; |
6098 | } | |
6099 | finish_wait(&ctx->sqo_wait, &wait); | |
6100 | ||
75b28aff | 6101 | ctx->rings->sq_flags &= ~IORING_SQ_NEED_WAKEUP; |
6c271ce2 JA |
6102 | } |
6103 | ||
8a4955ff | 6104 | mutex_lock(&ctx->uring_lock); |
6b668c9b XW |
6105 | if (likely(!percpu_ref_is_dying(&ctx->refs))) |
6106 | ret = io_submit_sqes(ctx, to_submit, NULL, -1); | |
8a4955ff | 6107 | mutex_unlock(&ctx->uring_lock); |
bdcd3eab | 6108 | timeout = jiffies + ctx->sq_thread_idle; |
6c271ce2 JA |
6109 | } |
6110 | ||
b41e9852 JA |
6111 | if (current->task_works) |
6112 | task_work_run(); | |
6113 | ||
bf9c2f1c | 6114 | io_sq_thread_drop_mm(ctx); |
181e448d | 6115 | revert_creds(old_cred); |
06058632 | 6116 | |
2bbcd6d3 | 6117 | kthread_parkme(); |
06058632 | 6118 | |
6c271ce2 JA |
6119 | return 0; |
6120 | } | |
6121 | ||
bda52162 JA |
6122 | struct io_wait_queue { |
6123 | struct wait_queue_entry wq; | |
6124 | struct io_ring_ctx *ctx; | |
6125 | unsigned to_wait; | |
6126 | unsigned nr_timeouts; | |
6127 | }; | |
6128 | ||
1d7bb1d5 | 6129 | static inline bool io_should_wake(struct io_wait_queue *iowq, bool noflush) |
bda52162 JA |
6130 | { |
6131 | struct io_ring_ctx *ctx = iowq->ctx; | |
6132 | ||
6133 | /* | |
d195a66e | 6134 | * Wake up if we have enough events, or if a timeout occurred since we |
bda52162 JA |
6135 | * started waiting. For timeouts, we always want to return to userspace, |
6136 | * regardless of event count. | |
6137 | */ | |
1d7bb1d5 | 6138 | return io_cqring_events(ctx, noflush) >= iowq->to_wait || |
bda52162 JA |
6139 | atomic_read(&ctx->cq_timeouts) != iowq->nr_timeouts; |
6140 | } | |
6141 | ||
6142 | static int io_wake_function(struct wait_queue_entry *curr, unsigned int mode, | |
6143 | int wake_flags, void *key) | |
6144 | { | |
6145 | struct io_wait_queue *iowq = container_of(curr, struct io_wait_queue, | |
6146 | wq); | |
6147 | ||
1d7bb1d5 JA |
6148 | /* use noflush == true, as we can't safely rely on locking context */ |
6149 | if (!io_should_wake(iowq, true)) | |
bda52162 JA |
6150 | return -1; |
6151 | ||
6152 | return autoremove_wake_function(curr, mode, wake_flags, key); | |
6153 | } | |
6154 | ||
2b188cc1 JA |
6155 | /* |
6156 | * Wait until events become available, if we don't already have some. The | |
6157 | * application must reap them itself, as they reside on the shared cq ring. | |
6158 | */ | |
6159 | static int io_cqring_wait(struct io_ring_ctx *ctx, int min_events, | |
6160 | const sigset_t __user *sig, size_t sigsz) | |
6161 | { | |
bda52162 JA |
6162 | struct io_wait_queue iowq = { |
6163 | .wq = { | |
6164 | .private = current, | |
6165 | .func = io_wake_function, | |
6166 | .entry = LIST_HEAD_INIT(iowq.wq.entry), | |
6167 | }, | |
6168 | .ctx = ctx, | |
6169 | .to_wait = min_events, | |
6170 | }; | |
75b28aff | 6171 | struct io_rings *rings = ctx->rings; |
e9ffa5c2 | 6172 | int ret = 0; |
2b188cc1 | 6173 | |
b41e9852 JA |
6174 | do { |
6175 | if (io_cqring_events(ctx, false) >= min_events) | |
6176 | return 0; | |
6177 | if (!current->task_works) | |
6178 | break; | |
6179 | task_work_run(); | |
6180 | } while (1); | |
2b188cc1 JA |
6181 | |
6182 | if (sig) { | |
9e75ad5d AB |
6183 | #ifdef CONFIG_COMPAT |
6184 | if (in_compat_syscall()) | |
6185 | ret = set_compat_user_sigmask((const compat_sigset_t __user *)sig, | |
b772434b | 6186 | sigsz); |
9e75ad5d AB |
6187 | else |
6188 | #endif | |
b772434b | 6189 | ret = set_user_sigmask(sig, sigsz); |
9e75ad5d | 6190 | |
2b188cc1 JA |
6191 | if (ret) |
6192 | return ret; | |
6193 | } | |
6194 | ||
bda52162 | 6195 | iowq.nr_timeouts = atomic_read(&ctx->cq_timeouts); |
c826bd7a | 6196 | trace_io_uring_cqring_wait(ctx, min_events); |
bda52162 JA |
6197 | do { |
6198 | prepare_to_wait_exclusive(&ctx->wait, &iowq.wq, | |
6199 | TASK_INTERRUPTIBLE); | |
ce593a6c | 6200 | /* make sure we run task_work before checking for signals */ |
b41e9852 JA |
6201 | if (current->task_works) |
6202 | task_work_run(); | |
bda52162 | 6203 | if (signal_pending(current)) { |
ce593a6c | 6204 | ret = -ERESTARTSYS; |
bda52162 JA |
6205 | break; |
6206 | } | |
ce593a6c JA |
6207 | if (io_should_wake(&iowq, false)) |
6208 | break; | |
6209 | schedule(); | |
bda52162 JA |
6210 | } while (1); |
6211 | finish_wait(&ctx->wait, &iowq.wq); | |
6212 | ||
ce593a6c | 6213 | restore_saved_sigmask_unless(ret == -ERESTARTSYS); |
2b188cc1 | 6214 | |
75b28aff | 6215 | return READ_ONCE(rings->cq.head) == READ_ONCE(rings->cq.tail) ? ret : 0; |
2b188cc1 JA |
6216 | } |
6217 | ||
6b06314c JA |
6218 | static void __io_sqe_files_unregister(struct io_ring_ctx *ctx) |
6219 | { | |
6220 | #if defined(CONFIG_UNIX) | |
6221 | if (ctx->ring_sock) { | |
6222 | struct sock *sock = ctx->ring_sock->sk; | |
6223 | struct sk_buff *skb; | |
6224 | ||
6225 | while ((skb = skb_dequeue(&sock->sk_receive_queue)) != NULL) | |
6226 | kfree_skb(skb); | |
6227 | } | |
6228 | #else | |
6229 | int i; | |
6230 | ||
65e19f54 JA |
6231 | for (i = 0; i < ctx->nr_user_files; i++) { |
6232 | struct file *file; | |
6233 | ||
6234 | file = io_file_from_index(ctx, i); | |
6235 | if (file) | |
6236 | fput(file); | |
6237 | } | |
6b06314c JA |
6238 | #endif |
6239 | } | |
6240 | ||
05f3fb3c JA |
6241 | static void io_file_ref_kill(struct percpu_ref *ref) |
6242 | { | |
6243 | struct fixed_file_data *data; | |
6244 | ||
6245 | data = container_of(ref, struct fixed_file_data, refs); | |
6246 | complete(&data->done); | |
6247 | } | |
6248 | ||
6b06314c JA |
6249 | static int io_sqe_files_unregister(struct io_ring_ctx *ctx) |
6250 | { | |
05f3fb3c | 6251 | struct fixed_file_data *data = ctx->file_data; |
05589553 | 6252 | struct fixed_file_ref_node *ref_node = NULL; |
65e19f54 JA |
6253 | unsigned nr_tables, i; |
6254 | ||
05f3fb3c | 6255 | if (!data) |
6b06314c JA |
6256 | return -ENXIO; |
6257 | ||
6a4d07cd | 6258 | spin_lock(&data->lock); |
05589553 XW |
6259 | if (!list_empty(&data->ref_list)) |
6260 | ref_node = list_first_entry(&data->ref_list, | |
6261 | struct fixed_file_ref_node, node); | |
6a4d07cd | 6262 | spin_unlock(&data->lock); |
05589553 XW |
6263 | if (ref_node) |
6264 | percpu_ref_kill(&ref_node->refs); | |
6265 | ||
6266 | percpu_ref_kill(&data->refs); | |
6267 | ||
6268 | /* wait for all refs nodes to complete */ | |
4a38aed2 | 6269 | flush_delayed_work(&ctx->file_put_work); |
2faf852d | 6270 | wait_for_completion(&data->done); |
05f3fb3c | 6271 | |
6b06314c | 6272 | __io_sqe_files_unregister(ctx); |
65e19f54 JA |
6273 | nr_tables = DIV_ROUND_UP(ctx->nr_user_files, IORING_MAX_FILES_TABLE); |
6274 | for (i = 0; i < nr_tables; i++) | |
05f3fb3c JA |
6275 | kfree(data->table[i].files); |
6276 | kfree(data->table); | |
05589553 XW |
6277 | percpu_ref_exit(&data->refs); |
6278 | kfree(data); | |
05f3fb3c | 6279 | ctx->file_data = NULL; |
6b06314c JA |
6280 | ctx->nr_user_files = 0; |
6281 | return 0; | |
6282 | } | |
6283 | ||
6c271ce2 JA |
6284 | static void io_sq_thread_stop(struct io_ring_ctx *ctx) |
6285 | { | |
6286 | if (ctx->sqo_thread) { | |
0f158b4c | 6287 | wait_for_completion(&ctx->sq_thread_comp); |
2bbcd6d3 RP |
6288 | /* |
6289 | * The park is a bit of a work-around, without it we get | |
6290 | * warning spews on shutdown with SQPOLL set and affinity | |
6291 | * set to a single CPU. | |
6292 | */ | |
06058632 | 6293 | kthread_park(ctx->sqo_thread); |
6c271ce2 JA |
6294 | kthread_stop(ctx->sqo_thread); |
6295 | ctx->sqo_thread = NULL; | |
6296 | } | |
6297 | } | |
6298 | ||
6b06314c JA |
6299 | static void io_finish_async(struct io_ring_ctx *ctx) |
6300 | { | |
6c271ce2 JA |
6301 | io_sq_thread_stop(ctx); |
6302 | ||
561fb04a JA |
6303 | if (ctx->io_wq) { |
6304 | io_wq_destroy(ctx->io_wq); | |
6305 | ctx->io_wq = NULL; | |
6b06314c JA |
6306 | } |
6307 | } | |
6308 | ||
6309 | #if defined(CONFIG_UNIX) | |
6b06314c JA |
6310 | /* |
6311 | * Ensure the UNIX gc is aware of our file set, so we are certain that | |
6312 | * the io_uring can be safely unregistered on process exit, even if we have | |
6313 | * loops in the file referencing. | |
6314 | */ | |
6315 | static int __io_sqe_files_scm(struct io_ring_ctx *ctx, int nr, int offset) | |
6316 | { | |
6317 | struct sock *sk = ctx->ring_sock->sk; | |
6318 | struct scm_fp_list *fpl; | |
6319 | struct sk_buff *skb; | |
08a45173 | 6320 | int i, nr_files; |
6b06314c | 6321 | |
6b06314c JA |
6322 | fpl = kzalloc(sizeof(*fpl), GFP_KERNEL); |
6323 | if (!fpl) | |
6324 | return -ENOMEM; | |
6325 | ||
6326 | skb = alloc_skb(0, GFP_KERNEL); | |
6327 | if (!skb) { | |
6328 | kfree(fpl); | |
6329 | return -ENOMEM; | |
6330 | } | |
6331 | ||
6332 | skb->sk = sk; | |
6b06314c | 6333 | |
08a45173 | 6334 | nr_files = 0; |
6b06314c JA |
6335 | fpl->user = get_uid(ctx->user); |
6336 | for (i = 0; i < nr; i++) { | |
65e19f54 JA |
6337 | struct file *file = io_file_from_index(ctx, i + offset); |
6338 | ||
6339 | if (!file) | |
08a45173 | 6340 | continue; |
65e19f54 | 6341 | fpl->fp[nr_files] = get_file(file); |
08a45173 JA |
6342 | unix_inflight(fpl->user, fpl->fp[nr_files]); |
6343 | nr_files++; | |
6b06314c JA |
6344 | } |
6345 | ||
08a45173 JA |
6346 | if (nr_files) { |
6347 | fpl->max = SCM_MAX_FD; | |
6348 | fpl->count = nr_files; | |
6349 | UNIXCB(skb).fp = fpl; | |
05f3fb3c | 6350 | skb->destructor = unix_destruct_scm; |
08a45173 JA |
6351 | refcount_add(skb->truesize, &sk->sk_wmem_alloc); |
6352 | skb_queue_head(&sk->sk_receive_queue, skb); | |
6b06314c | 6353 | |
08a45173 JA |
6354 | for (i = 0; i < nr_files; i++) |
6355 | fput(fpl->fp[i]); | |
6356 | } else { | |
6357 | kfree_skb(skb); | |
6358 | kfree(fpl); | |
6359 | } | |
6b06314c JA |
6360 | |
6361 | return 0; | |
6362 | } | |
6363 | ||
6364 | /* | |
6365 | * If UNIX sockets are enabled, fd passing can cause a reference cycle which | |
6366 | * causes regular reference counting to break down. We rely on the UNIX | |
6367 | * garbage collection to take care of this problem for us. | |
6368 | */ | |
6369 | static int io_sqe_files_scm(struct io_ring_ctx *ctx) | |
6370 | { | |
6371 | unsigned left, total; | |
6372 | int ret = 0; | |
6373 | ||
6374 | total = 0; | |
6375 | left = ctx->nr_user_files; | |
6376 | while (left) { | |
6377 | unsigned this_files = min_t(unsigned, left, SCM_MAX_FD); | |
6b06314c JA |
6378 | |
6379 | ret = __io_sqe_files_scm(ctx, this_files, total); | |
6380 | if (ret) | |
6381 | break; | |
6382 | left -= this_files; | |
6383 | total += this_files; | |
6384 | } | |
6385 | ||
6386 | if (!ret) | |
6387 | return 0; | |
6388 | ||
6389 | while (total < ctx->nr_user_files) { | |
65e19f54 JA |
6390 | struct file *file = io_file_from_index(ctx, total); |
6391 | ||
6392 | if (file) | |
6393 | fput(file); | |
6b06314c JA |
6394 | total++; |
6395 | } | |
6396 | ||
6397 | return ret; | |
6398 | } | |
6399 | #else | |
6400 | static int io_sqe_files_scm(struct io_ring_ctx *ctx) | |
6401 | { | |
6402 | return 0; | |
6403 | } | |
6404 | #endif | |
6405 | ||
65e19f54 JA |
6406 | static int io_sqe_alloc_file_tables(struct io_ring_ctx *ctx, unsigned nr_tables, |
6407 | unsigned nr_files) | |
6408 | { | |
6409 | int i; | |
6410 | ||
6411 | for (i = 0; i < nr_tables; i++) { | |
05f3fb3c | 6412 | struct fixed_file_table *table = &ctx->file_data->table[i]; |
65e19f54 JA |
6413 | unsigned this_files; |
6414 | ||
6415 | this_files = min(nr_files, IORING_MAX_FILES_TABLE); | |
6416 | table->files = kcalloc(this_files, sizeof(struct file *), | |
6417 | GFP_KERNEL); | |
6418 | if (!table->files) | |
6419 | break; | |
6420 | nr_files -= this_files; | |
6421 | } | |
6422 | ||
6423 | if (i == nr_tables) | |
6424 | return 0; | |
6425 | ||
6426 | for (i = 0; i < nr_tables; i++) { | |
05f3fb3c | 6427 | struct fixed_file_table *table = &ctx->file_data->table[i]; |
65e19f54 JA |
6428 | kfree(table->files); |
6429 | } | |
6430 | return 1; | |
6431 | } | |
6432 | ||
05f3fb3c JA |
6433 | static void io_ring_file_put(struct io_ring_ctx *ctx, struct file *file) |
6434 | { | |
6435 | #if defined(CONFIG_UNIX) | |
6436 | struct sock *sock = ctx->ring_sock->sk; | |
6437 | struct sk_buff_head list, *head = &sock->sk_receive_queue; | |
6438 | struct sk_buff *skb; | |
6439 | int i; | |
6440 | ||
6441 | __skb_queue_head_init(&list); | |
6442 | ||
6443 | /* | |
6444 | * Find the skb that holds this file in its SCM_RIGHTS. When found, | |
6445 | * remove this entry and rearrange the file array. | |
6446 | */ | |
6447 | skb = skb_dequeue(head); | |
6448 | while (skb) { | |
6449 | struct scm_fp_list *fp; | |
6450 | ||
6451 | fp = UNIXCB(skb).fp; | |
6452 | for (i = 0; i < fp->count; i++) { | |
6453 | int left; | |
6454 | ||
6455 | if (fp->fp[i] != file) | |
6456 | continue; | |
6457 | ||
6458 | unix_notinflight(fp->user, fp->fp[i]); | |
6459 | left = fp->count - 1 - i; | |
6460 | if (left) { | |
6461 | memmove(&fp->fp[i], &fp->fp[i + 1], | |
6462 | left * sizeof(struct file *)); | |
6463 | } | |
6464 | fp->count--; | |
6465 | if (!fp->count) { | |
6466 | kfree_skb(skb); | |
6467 | skb = NULL; | |
6468 | } else { | |
6469 | __skb_queue_tail(&list, skb); | |
6470 | } | |
6471 | fput(file); | |
6472 | file = NULL; | |
6473 | break; | |
6474 | } | |
6475 | ||
6476 | if (!file) | |
6477 | break; | |
6478 | ||
6479 | __skb_queue_tail(&list, skb); | |
6480 | ||
6481 | skb = skb_dequeue(head); | |
6482 | } | |
6483 | ||
6484 | if (skb_peek(&list)) { | |
6485 | spin_lock_irq(&head->lock); | |
6486 | while ((skb = __skb_dequeue(&list)) != NULL) | |
6487 | __skb_queue_tail(head, skb); | |
6488 | spin_unlock_irq(&head->lock); | |
6489 | } | |
6490 | #else | |
6491 | fput(file); | |
6492 | #endif | |
6493 | } | |
6494 | ||
6495 | struct io_file_put { | |
05589553 | 6496 | struct list_head list; |
05f3fb3c | 6497 | struct file *file; |
05f3fb3c JA |
6498 | }; |
6499 | ||
4a38aed2 | 6500 | static void __io_file_put_work(struct fixed_file_ref_node *ref_node) |
65e19f54 | 6501 | { |
4a38aed2 JA |
6502 | struct fixed_file_data *file_data = ref_node->file_data; |
6503 | struct io_ring_ctx *ctx = file_data->ctx; | |
05f3fb3c | 6504 | struct io_file_put *pfile, *tmp; |
05589553 XW |
6505 | |
6506 | list_for_each_entry_safe(pfile, tmp, &ref_node->file_list, list) { | |
6a4d07cd | 6507 | list_del(&pfile->list); |
05589553 XW |
6508 | io_ring_file_put(ctx, pfile->file); |
6509 | kfree(pfile); | |
65e19f54 | 6510 | } |
05589553 | 6511 | |
6a4d07cd JA |
6512 | spin_lock(&file_data->lock); |
6513 | list_del(&ref_node->node); | |
6514 | spin_unlock(&file_data->lock); | |
05589553 XW |
6515 | |
6516 | percpu_ref_exit(&ref_node->refs); | |
6517 | kfree(ref_node); | |
6518 | percpu_ref_put(&file_data->refs); | |
2faf852d | 6519 | } |
65e19f54 | 6520 | |
4a38aed2 JA |
6521 | static void io_file_put_work(struct work_struct *work) |
6522 | { | |
6523 | struct io_ring_ctx *ctx; | |
6524 | struct llist_node *node; | |
6525 | ||
6526 | ctx = container_of(work, struct io_ring_ctx, file_put_work.work); | |
6527 | node = llist_del_all(&ctx->file_put_llist); | |
6528 | ||
6529 | while (node) { | |
6530 | struct fixed_file_ref_node *ref_node; | |
6531 | struct llist_node *next = node->next; | |
6532 | ||
6533 | ref_node = llist_entry(node, struct fixed_file_ref_node, llist); | |
6534 | __io_file_put_work(ref_node); | |
6535 | node = next; | |
6536 | } | |
6537 | } | |
6538 | ||
05589553 | 6539 | static void io_file_data_ref_zero(struct percpu_ref *ref) |
2faf852d | 6540 | { |
05589553 | 6541 | struct fixed_file_ref_node *ref_node; |
4a38aed2 JA |
6542 | struct io_ring_ctx *ctx; |
6543 | bool first_add; | |
6544 | int delay = HZ; | |
65e19f54 | 6545 | |
05589553 | 6546 | ref_node = container_of(ref, struct fixed_file_ref_node, refs); |
4a38aed2 | 6547 | ctx = ref_node->file_data->ctx; |
05589553 | 6548 | |
4a38aed2 JA |
6549 | if (percpu_ref_is_dying(&ctx->file_data->refs)) |
6550 | delay = 0; | |
05589553 | 6551 | |
4a38aed2 JA |
6552 | first_add = llist_add(&ref_node->llist, &ctx->file_put_llist); |
6553 | if (!delay) | |
6554 | mod_delayed_work(system_wq, &ctx->file_put_work, 0); | |
6555 | else if (first_add) | |
6556 | queue_delayed_work(system_wq, &ctx->file_put_work, delay); | |
05f3fb3c | 6557 | } |
65e19f54 | 6558 | |
05589553 XW |
6559 | static struct fixed_file_ref_node *alloc_fixed_file_ref_node( |
6560 | struct io_ring_ctx *ctx) | |
05f3fb3c | 6561 | { |
05589553 | 6562 | struct fixed_file_ref_node *ref_node; |
05f3fb3c | 6563 | |
05589553 XW |
6564 | ref_node = kzalloc(sizeof(*ref_node), GFP_KERNEL); |
6565 | if (!ref_node) | |
6566 | return ERR_PTR(-ENOMEM); | |
05f3fb3c | 6567 | |
05589553 XW |
6568 | if (percpu_ref_init(&ref_node->refs, io_file_data_ref_zero, |
6569 | 0, GFP_KERNEL)) { | |
6570 | kfree(ref_node); | |
6571 | return ERR_PTR(-ENOMEM); | |
6572 | } | |
6573 | INIT_LIST_HEAD(&ref_node->node); | |
6574 | INIT_LIST_HEAD(&ref_node->file_list); | |
05589553 XW |
6575 | ref_node->file_data = ctx->file_data; |
6576 | return ref_node; | |
05589553 XW |
6577 | } |
6578 | ||
6579 | static void destroy_fixed_file_ref_node(struct fixed_file_ref_node *ref_node) | |
6580 | { | |
6581 | percpu_ref_exit(&ref_node->refs); | |
6582 | kfree(ref_node); | |
65e19f54 JA |
6583 | } |
6584 | ||
6b06314c JA |
6585 | static int io_sqe_files_register(struct io_ring_ctx *ctx, void __user *arg, |
6586 | unsigned nr_args) | |
6587 | { | |
6588 | __s32 __user *fds = (__s32 __user *) arg; | |
65e19f54 | 6589 | unsigned nr_tables; |
05f3fb3c | 6590 | struct file *file; |
6b06314c JA |
6591 | int fd, ret = 0; |
6592 | unsigned i; | |
05589553 | 6593 | struct fixed_file_ref_node *ref_node; |
6b06314c | 6594 | |
05f3fb3c | 6595 | if (ctx->file_data) |
6b06314c JA |
6596 | return -EBUSY; |
6597 | if (!nr_args) | |
6598 | return -EINVAL; | |
6599 | if (nr_args > IORING_MAX_FIXED_FILES) | |
6600 | return -EMFILE; | |
6601 | ||
05f3fb3c JA |
6602 | ctx->file_data = kzalloc(sizeof(*ctx->file_data), GFP_KERNEL); |
6603 | if (!ctx->file_data) | |
6604 | return -ENOMEM; | |
6605 | ctx->file_data->ctx = ctx; | |
6606 | init_completion(&ctx->file_data->done); | |
05589553 | 6607 | INIT_LIST_HEAD(&ctx->file_data->ref_list); |
f7fe9346 | 6608 | spin_lock_init(&ctx->file_data->lock); |
05f3fb3c | 6609 | |
65e19f54 | 6610 | nr_tables = DIV_ROUND_UP(nr_args, IORING_MAX_FILES_TABLE); |
05f3fb3c JA |
6611 | ctx->file_data->table = kcalloc(nr_tables, |
6612 | sizeof(struct fixed_file_table), | |
65e19f54 | 6613 | GFP_KERNEL); |
05f3fb3c JA |
6614 | if (!ctx->file_data->table) { |
6615 | kfree(ctx->file_data); | |
6616 | ctx->file_data = NULL; | |
6b06314c | 6617 | return -ENOMEM; |
05f3fb3c JA |
6618 | } |
6619 | ||
05589553 | 6620 | if (percpu_ref_init(&ctx->file_data->refs, io_file_ref_kill, |
05f3fb3c JA |
6621 | PERCPU_REF_ALLOW_REINIT, GFP_KERNEL)) { |
6622 | kfree(ctx->file_data->table); | |
6623 | kfree(ctx->file_data); | |
6624 | ctx->file_data = NULL; | |
6b06314c | 6625 | return -ENOMEM; |
05f3fb3c | 6626 | } |
6b06314c | 6627 | |
65e19f54 | 6628 | if (io_sqe_alloc_file_tables(ctx, nr_tables, nr_args)) { |
05f3fb3c JA |
6629 | percpu_ref_exit(&ctx->file_data->refs); |
6630 | kfree(ctx->file_data->table); | |
6631 | kfree(ctx->file_data); | |
6632 | ctx->file_data = NULL; | |
65e19f54 JA |
6633 | return -ENOMEM; |
6634 | } | |
6635 | ||
08a45173 | 6636 | for (i = 0; i < nr_args; i++, ctx->nr_user_files++) { |
65e19f54 JA |
6637 | struct fixed_file_table *table; |
6638 | unsigned index; | |
6639 | ||
6b06314c JA |
6640 | ret = -EFAULT; |
6641 | if (copy_from_user(&fd, &fds[i], sizeof(fd))) | |
6642 | break; | |
08a45173 JA |
6643 | /* allow sparse sets */ |
6644 | if (fd == -1) { | |
6645 | ret = 0; | |
6646 | continue; | |
6647 | } | |
6b06314c | 6648 | |
05f3fb3c | 6649 | table = &ctx->file_data->table[i >> IORING_FILE_TABLE_SHIFT]; |
65e19f54 | 6650 | index = i & IORING_FILE_TABLE_MASK; |
05f3fb3c | 6651 | file = fget(fd); |
6b06314c JA |
6652 | |
6653 | ret = -EBADF; | |
05f3fb3c | 6654 | if (!file) |
6b06314c | 6655 | break; |
05f3fb3c | 6656 | |
6b06314c JA |
6657 | /* |
6658 | * Don't allow io_uring instances to be registered. If UNIX | |
6659 | * isn't enabled, then this causes a reference cycle and this | |
6660 | * instance can never get freed. If UNIX is enabled we'll | |
6661 | * handle it just fine, but there's still no point in allowing | |
6662 | * a ring fd as it doesn't support regular read/write anyway. | |
6663 | */ | |
05f3fb3c JA |
6664 | if (file->f_op == &io_uring_fops) { |
6665 | fput(file); | |
6b06314c JA |
6666 | break; |
6667 | } | |
6b06314c | 6668 | ret = 0; |
05f3fb3c | 6669 | table->files[index] = file; |
6b06314c JA |
6670 | } |
6671 | ||
6672 | if (ret) { | |
65e19f54 | 6673 | for (i = 0; i < ctx->nr_user_files; i++) { |
65e19f54 JA |
6674 | file = io_file_from_index(ctx, i); |
6675 | if (file) | |
6676 | fput(file); | |
6677 | } | |
6678 | for (i = 0; i < nr_tables; i++) | |
05f3fb3c | 6679 | kfree(ctx->file_data->table[i].files); |
6b06314c | 6680 | |
05f3fb3c JA |
6681 | kfree(ctx->file_data->table); |
6682 | kfree(ctx->file_data); | |
6683 | ctx->file_data = NULL; | |
6b06314c JA |
6684 | ctx->nr_user_files = 0; |
6685 | return ret; | |
6686 | } | |
6687 | ||
6688 | ret = io_sqe_files_scm(ctx); | |
05589553 | 6689 | if (ret) { |
6b06314c | 6690 | io_sqe_files_unregister(ctx); |
05589553 XW |
6691 | return ret; |
6692 | } | |
6b06314c | 6693 | |
05589553 XW |
6694 | ref_node = alloc_fixed_file_ref_node(ctx); |
6695 | if (IS_ERR(ref_node)) { | |
6696 | io_sqe_files_unregister(ctx); | |
6697 | return PTR_ERR(ref_node); | |
6698 | } | |
6699 | ||
6700 | ctx->file_data->cur_refs = &ref_node->refs; | |
6a4d07cd | 6701 | spin_lock(&ctx->file_data->lock); |
05589553 | 6702 | list_add(&ref_node->node, &ctx->file_data->ref_list); |
6a4d07cd | 6703 | spin_unlock(&ctx->file_data->lock); |
05589553 | 6704 | percpu_ref_get(&ctx->file_data->refs); |
6b06314c JA |
6705 | return ret; |
6706 | } | |
6707 | ||
c3a31e60 JA |
6708 | static int io_sqe_file_register(struct io_ring_ctx *ctx, struct file *file, |
6709 | int index) | |
6710 | { | |
6711 | #if defined(CONFIG_UNIX) | |
6712 | struct sock *sock = ctx->ring_sock->sk; | |
6713 | struct sk_buff_head *head = &sock->sk_receive_queue; | |
6714 | struct sk_buff *skb; | |
6715 | ||
6716 | /* | |
6717 | * See if we can merge this file into an existing skb SCM_RIGHTS | |
6718 | * file set. If there's no room, fall back to allocating a new skb | |
6719 | * and filling it in. | |
6720 | */ | |
6721 | spin_lock_irq(&head->lock); | |
6722 | skb = skb_peek(head); | |
6723 | if (skb) { | |
6724 | struct scm_fp_list *fpl = UNIXCB(skb).fp; | |
6725 | ||
6726 | if (fpl->count < SCM_MAX_FD) { | |
6727 | __skb_unlink(skb, head); | |
6728 | spin_unlock_irq(&head->lock); | |
6729 | fpl->fp[fpl->count] = get_file(file); | |
6730 | unix_inflight(fpl->user, fpl->fp[fpl->count]); | |
6731 | fpl->count++; | |
6732 | spin_lock_irq(&head->lock); | |
6733 | __skb_queue_head(head, skb); | |
6734 | } else { | |
6735 | skb = NULL; | |
6736 | } | |
6737 | } | |
6738 | spin_unlock_irq(&head->lock); | |
6739 | ||
6740 | if (skb) { | |
6741 | fput(file); | |
6742 | return 0; | |
6743 | } | |
6744 | ||
6745 | return __io_sqe_files_scm(ctx, 1, index); | |
6746 | #else | |
6747 | return 0; | |
6748 | #endif | |
6749 | } | |
6750 | ||
a5318d3c | 6751 | static int io_queue_file_removal(struct fixed_file_data *data, |
05589553 | 6752 | struct file *file) |
05f3fb3c | 6753 | { |
a5318d3c | 6754 | struct io_file_put *pfile; |
05589553 XW |
6755 | struct percpu_ref *refs = data->cur_refs; |
6756 | struct fixed_file_ref_node *ref_node; | |
05f3fb3c | 6757 | |
05f3fb3c | 6758 | pfile = kzalloc(sizeof(*pfile), GFP_KERNEL); |
a5318d3c HD |
6759 | if (!pfile) |
6760 | return -ENOMEM; | |
05f3fb3c | 6761 | |
05589553 | 6762 | ref_node = container_of(refs, struct fixed_file_ref_node, refs); |
05f3fb3c | 6763 | pfile->file = file; |
05589553 XW |
6764 | list_add(&pfile->list, &ref_node->file_list); |
6765 | ||
a5318d3c | 6766 | return 0; |
05f3fb3c JA |
6767 | } |
6768 | ||
6769 | static int __io_sqe_files_update(struct io_ring_ctx *ctx, | |
6770 | struct io_uring_files_update *up, | |
6771 | unsigned nr_args) | |
6772 | { | |
6773 | struct fixed_file_data *data = ctx->file_data; | |
05589553 | 6774 | struct fixed_file_ref_node *ref_node; |
05f3fb3c | 6775 | struct file *file; |
c3a31e60 JA |
6776 | __s32 __user *fds; |
6777 | int fd, i, err; | |
6778 | __u32 done; | |
05589553 | 6779 | bool needs_switch = false; |
c3a31e60 | 6780 | |
05f3fb3c | 6781 | if (check_add_overflow(up->offset, nr_args, &done)) |
c3a31e60 JA |
6782 | return -EOVERFLOW; |
6783 | if (done > ctx->nr_user_files) | |
6784 | return -EINVAL; | |
6785 | ||
05589553 XW |
6786 | ref_node = alloc_fixed_file_ref_node(ctx); |
6787 | if (IS_ERR(ref_node)) | |
6788 | return PTR_ERR(ref_node); | |
6789 | ||
c3a31e60 | 6790 | done = 0; |
05f3fb3c | 6791 | fds = u64_to_user_ptr(up->fds); |
c3a31e60 | 6792 | while (nr_args) { |
65e19f54 JA |
6793 | struct fixed_file_table *table; |
6794 | unsigned index; | |
6795 | ||
c3a31e60 JA |
6796 | err = 0; |
6797 | if (copy_from_user(&fd, &fds[done], sizeof(fd))) { | |
6798 | err = -EFAULT; | |
6799 | break; | |
6800 | } | |
05f3fb3c JA |
6801 | i = array_index_nospec(up->offset, ctx->nr_user_files); |
6802 | table = &ctx->file_data->table[i >> IORING_FILE_TABLE_SHIFT]; | |
65e19f54 JA |
6803 | index = i & IORING_FILE_TABLE_MASK; |
6804 | if (table->files[index]) { | |
05f3fb3c | 6805 | file = io_file_from_index(ctx, index); |
a5318d3c HD |
6806 | err = io_queue_file_removal(data, file); |
6807 | if (err) | |
6808 | break; | |
65e19f54 | 6809 | table->files[index] = NULL; |
05589553 | 6810 | needs_switch = true; |
c3a31e60 JA |
6811 | } |
6812 | if (fd != -1) { | |
c3a31e60 JA |
6813 | file = fget(fd); |
6814 | if (!file) { | |
6815 | err = -EBADF; | |
6816 | break; | |
6817 | } | |
6818 | /* | |
6819 | * Don't allow io_uring instances to be registered. If | |
6820 | * UNIX isn't enabled, then this causes a reference | |
6821 | * cycle and this instance can never get freed. If UNIX | |
6822 | * is enabled we'll handle it just fine, but there's | |
6823 | * still no point in allowing a ring fd as it doesn't | |
6824 | * support regular read/write anyway. | |
6825 | */ | |
6826 | if (file->f_op == &io_uring_fops) { | |
6827 | fput(file); | |
6828 | err = -EBADF; | |
6829 | break; | |
6830 | } | |
65e19f54 | 6831 | table->files[index] = file; |
c3a31e60 JA |
6832 | err = io_sqe_file_register(ctx, file, i); |
6833 | if (err) | |
6834 | break; | |
6835 | } | |
6836 | nr_args--; | |
6837 | done++; | |
05f3fb3c JA |
6838 | up->offset++; |
6839 | } | |
6840 | ||
05589553 XW |
6841 | if (needs_switch) { |
6842 | percpu_ref_kill(data->cur_refs); | |
6a4d07cd | 6843 | spin_lock(&data->lock); |
05589553 XW |
6844 | list_add(&ref_node->node, &data->ref_list); |
6845 | data->cur_refs = &ref_node->refs; | |
6a4d07cd | 6846 | spin_unlock(&data->lock); |
05589553 XW |
6847 | percpu_ref_get(&ctx->file_data->refs); |
6848 | } else | |
6849 | destroy_fixed_file_ref_node(ref_node); | |
c3a31e60 JA |
6850 | |
6851 | return done ? done : err; | |
6852 | } | |
05589553 | 6853 | |
05f3fb3c JA |
6854 | static int io_sqe_files_update(struct io_ring_ctx *ctx, void __user *arg, |
6855 | unsigned nr_args) | |
6856 | { | |
6857 | struct io_uring_files_update up; | |
6858 | ||
6859 | if (!ctx->file_data) | |
6860 | return -ENXIO; | |
6861 | if (!nr_args) | |
6862 | return -EINVAL; | |
6863 | if (copy_from_user(&up, arg, sizeof(up))) | |
6864 | return -EFAULT; | |
6865 | if (up.resv) | |
6866 | return -EINVAL; | |
6867 | ||
6868 | return __io_sqe_files_update(ctx, &up, nr_args); | |
6869 | } | |
c3a31e60 | 6870 | |
e9fd9396 | 6871 | static void io_free_work(struct io_wq_work *work) |
7d723065 JA |
6872 | { |
6873 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); | |
6874 | ||
e9fd9396 | 6875 | /* Consider that io_steal_work() relies on this ref */ |
7d723065 JA |
6876 | io_put_req(req); |
6877 | } | |
6878 | ||
24369c2e PB |
6879 | static int io_init_wq_offload(struct io_ring_ctx *ctx, |
6880 | struct io_uring_params *p) | |
6881 | { | |
6882 | struct io_wq_data data; | |
6883 | struct fd f; | |
6884 | struct io_ring_ctx *ctx_attach; | |
6885 | unsigned int concurrency; | |
6886 | int ret = 0; | |
6887 | ||
6888 | data.user = ctx->user; | |
e9fd9396 | 6889 | data.free_work = io_free_work; |
f5fa38c5 | 6890 | data.do_work = io_wq_submit_work; |
24369c2e PB |
6891 | |
6892 | if (!(p->flags & IORING_SETUP_ATTACH_WQ)) { | |
6893 | /* Do QD, or 4 * CPUS, whatever is smallest */ | |
6894 | concurrency = min(ctx->sq_entries, 4 * num_online_cpus()); | |
6895 | ||
6896 | ctx->io_wq = io_wq_create(concurrency, &data); | |
6897 | if (IS_ERR(ctx->io_wq)) { | |
6898 | ret = PTR_ERR(ctx->io_wq); | |
6899 | ctx->io_wq = NULL; | |
6900 | } | |
6901 | return ret; | |
6902 | } | |
6903 | ||
6904 | f = fdget(p->wq_fd); | |
6905 | if (!f.file) | |
6906 | return -EBADF; | |
6907 | ||
6908 | if (f.file->f_op != &io_uring_fops) { | |
6909 | ret = -EINVAL; | |
6910 | goto out_fput; | |
6911 | } | |
6912 | ||
6913 | ctx_attach = f.file->private_data; | |
6914 | /* @io_wq is protected by holding the fd */ | |
6915 | if (!io_wq_get(ctx_attach->io_wq, &data)) { | |
6916 | ret = -EINVAL; | |
6917 | goto out_fput; | |
6918 | } | |
6919 | ||
6920 | ctx->io_wq = ctx_attach->io_wq; | |
6921 | out_fput: | |
6922 | fdput(f); | |
6923 | return ret; | |
6924 | } | |
6925 | ||
6c271ce2 JA |
6926 | static int io_sq_offload_start(struct io_ring_ctx *ctx, |
6927 | struct io_uring_params *p) | |
2b188cc1 JA |
6928 | { |
6929 | int ret; | |
6930 | ||
6931 | mmgrab(current->mm); | |
6932 | ctx->sqo_mm = current->mm; | |
6933 | ||
6c271ce2 | 6934 | if (ctx->flags & IORING_SETUP_SQPOLL) { |
3ec482d1 JA |
6935 | ret = -EPERM; |
6936 | if (!capable(CAP_SYS_ADMIN)) | |
6937 | goto err; | |
6938 | ||
917257da JA |
6939 | ctx->sq_thread_idle = msecs_to_jiffies(p->sq_thread_idle); |
6940 | if (!ctx->sq_thread_idle) | |
6941 | ctx->sq_thread_idle = HZ; | |
6942 | ||
6c271ce2 | 6943 | if (p->flags & IORING_SETUP_SQ_AFF) { |
44a9bd18 | 6944 | int cpu = p->sq_thread_cpu; |
6c271ce2 | 6945 | |
917257da | 6946 | ret = -EINVAL; |
44a9bd18 JA |
6947 | if (cpu >= nr_cpu_ids) |
6948 | goto err; | |
7889f44d | 6949 | if (!cpu_online(cpu)) |
917257da JA |
6950 | goto err; |
6951 | ||
6c271ce2 JA |
6952 | ctx->sqo_thread = kthread_create_on_cpu(io_sq_thread, |
6953 | ctx, cpu, | |
6954 | "io_uring-sq"); | |
6955 | } else { | |
6956 | ctx->sqo_thread = kthread_create(io_sq_thread, ctx, | |
6957 | "io_uring-sq"); | |
6958 | } | |
6959 | if (IS_ERR(ctx->sqo_thread)) { | |
6960 | ret = PTR_ERR(ctx->sqo_thread); | |
6961 | ctx->sqo_thread = NULL; | |
6962 | goto err; | |
6963 | } | |
6964 | wake_up_process(ctx->sqo_thread); | |
6965 | } else if (p->flags & IORING_SETUP_SQ_AFF) { | |
6966 | /* Can't have SQ_AFF without SQPOLL */ | |
6967 | ret = -EINVAL; | |
6968 | goto err; | |
6969 | } | |
6970 | ||
24369c2e PB |
6971 | ret = io_init_wq_offload(ctx, p); |
6972 | if (ret) | |
2b188cc1 | 6973 | goto err; |
2b188cc1 JA |
6974 | |
6975 | return 0; | |
6976 | err: | |
54a91f3b | 6977 | io_finish_async(ctx); |
2b188cc1 JA |
6978 | mmdrop(ctx->sqo_mm); |
6979 | ctx->sqo_mm = NULL; | |
6980 | return ret; | |
6981 | } | |
6982 | ||
6983 | static void io_unaccount_mem(struct user_struct *user, unsigned long nr_pages) | |
6984 | { | |
6985 | atomic_long_sub(nr_pages, &user->locked_vm); | |
6986 | } | |
6987 | ||
6988 | static int io_account_mem(struct user_struct *user, unsigned long nr_pages) | |
6989 | { | |
6990 | unsigned long page_limit, cur_pages, new_pages; | |
6991 | ||
6992 | /* Don't allow more pages than we can safely lock */ | |
6993 | page_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT; | |
6994 | ||
6995 | do { | |
6996 | cur_pages = atomic_long_read(&user->locked_vm); | |
6997 | new_pages = cur_pages + nr_pages; | |
6998 | if (new_pages > page_limit) | |
6999 | return -ENOMEM; | |
7000 | } while (atomic_long_cmpxchg(&user->locked_vm, cur_pages, | |
7001 | new_pages) != cur_pages); | |
7002 | ||
7003 | return 0; | |
7004 | } | |
7005 | ||
7006 | static void io_mem_free(void *ptr) | |
7007 | { | |
52e04ef4 MR |
7008 | struct page *page; |
7009 | ||
7010 | if (!ptr) | |
7011 | return; | |
2b188cc1 | 7012 | |
52e04ef4 | 7013 | page = virt_to_head_page(ptr); |
2b188cc1 JA |
7014 | if (put_page_testzero(page)) |
7015 | free_compound_page(page); | |
7016 | } | |
7017 | ||
7018 | static void *io_mem_alloc(size_t size) | |
7019 | { | |
7020 | gfp_t gfp_flags = GFP_KERNEL | __GFP_ZERO | __GFP_NOWARN | __GFP_COMP | | |
7021 | __GFP_NORETRY; | |
7022 | ||
7023 | return (void *) __get_free_pages(gfp_flags, get_order(size)); | |
7024 | } | |
7025 | ||
75b28aff HV |
7026 | static unsigned long rings_size(unsigned sq_entries, unsigned cq_entries, |
7027 | size_t *sq_offset) | |
7028 | { | |
7029 | struct io_rings *rings; | |
7030 | size_t off, sq_array_size; | |
7031 | ||
7032 | off = struct_size(rings, cqes, cq_entries); | |
7033 | if (off == SIZE_MAX) | |
7034 | return SIZE_MAX; | |
7035 | ||
7036 | #ifdef CONFIG_SMP | |
7037 | off = ALIGN(off, SMP_CACHE_BYTES); | |
7038 | if (off == 0) | |
7039 | return SIZE_MAX; | |
7040 | #endif | |
7041 | ||
7042 | sq_array_size = array_size(sizeof(u32), sq_entries); | |
7043 | if (sq_array_size == SIZE_MAX) | |
7044 | return SIZE_MAX; | |
7045 | ||
7046 | if (check_add_overflow(off, sq_array_size, &off)) | |
7047 | return SIZE_MAX; | |
7048 | ||
7049 | if (sq_offset) | |
7050 | *sq_offset = off; | |
7051 | ||
7052 | return off; | |
7053 | } | |
7054 | ||
2b188cc1 JA |
7055 | static unsigned long ring_pages(unsigned sq_entries, unsigned cq_entries) |
7056 | { | |
75b28aff | 7057 | size_t pages; |
2b188cc1 | 7058 | |
75b28aff HV |
7059 | pages = (size_t)1 << get_order( |
7060 | rings_size(sq_entries, cq_entries, NULL)); | |
7061 | pages += (size_t)1 << get_order( | |
7062 | array_size(sizeof(struct io_uring_sqe), sq_entries)); | |
2b188cc1 | 7063 | |
75b28aff | 7064 | return pages; |
2b188cc1 JA |
7065 | } |
7066 | ||
edafccee JA |
7067 | static int io_sqe_buffer_unregister(struct io_ring_ctx *ctx) |
7068 | { | |
7069 | int i, j; | |
7070 | ||
7071 | if (!ctx->user_bufs) | |
7072 | return -ENXIO; | |
7073 | ||
7074 | for (i = 0; i < ctx->nr_user_bufs; i++) { | |
7075 | struct io_mapped_ubuf *imu = &ctx->user_bufs[i]; | |
7076 | ||
7077 | for (j = 0; j < imu->nr_bvecs; j++) | |
f1f6a7dd | 7078 | unpin_user_page(imu->bvec[j].bv_page); |
edafccee JA |
7079 | |
7080 | if (ctx->account_mem) | |
7081 | io_unaccount_mem(ctx->user, imu->nr_bvecs); | |
d4ef6475 | 7082 | kvfree(imu->bvec); |
edafccee JA |
7083 | imu->nr_bvecs = 0; |
7084 | } | |
7085 | ||
7086 | kfree(ctx->user_bufs); | |
7087 | ctx->user_bufs = NULL; | |
7088 | ctx->nr_user_bufs = 0; | |
7089 | return 0; | |
7090 | } | |
7091 | ||
7092 | static int io_copy_iov(struct io_ring_ctx *ctx, struct iovec *dst, | |
7093 | void __user *arg, unsigned index) | |
7094 | { | |
7095 | struct iovec __user *src; | |
7096 | ||
7097 | #ifdef CONFIG_COMPAT | |
7098 | if (ctx->compat) { | |
7099 | struct compat_iovec __user *ciovs; | |
7100 | struct compat_iovec ciov; | |
7101 | ||
7102 | ciovs = (struct compat_iovec __user *) arg; | |
7103 | if (copy_from_user(&ciov, &ciovs[index], sizeof(ciov))) | |
7104 | return -EFAULT; | |
7105 | ||
d55e5f5b | 7106 | dst->iov_base = u64_to_user_ptr((u64)ciov.iov_base); |
edafccee JA |
7107 | dst->iov_len = ciov.iov_len; |
7108 | return 0; | |
7109 | } | |
7110 | #endif | |
7111 | src = (struct iovec __user *) arg; | |
7112 | if (copy_from_user(dst, &src[index], sizeof(*dst))) | |
7113 | return -EFAULT; | |
7114 | return 0; | |
7115 | } | |
7116 | ||
7117 | static int io_sqe_buffer_register(struct io_ring_ctx *ctx, void __user *arg, | |
7118 | unsigned nr_args) | |
7119 | { | |
7120 | struct vm_area_struct **vmas = NULL; | |
7121 | struct page **pages = NULL; | |
7122 | int i, j, got_pages = 0; | |
7123 | int ret = -EINVAL; | |
7124 | ||
7125 | if (ctx->user_bufs) | |
7126 | return -EBUSY; | |
7127 | if (!nr_args || nr_args > UIO_MAXIOV) | |
7128 | return -EINVAL; | |
7129 | ||
7130 | ctx->user_bufs = kcalloc(nr_args, sizeof(struct io_mapped_ubuf), | |
7131 | GFP_KERNEL); | |
7132 | if (!ctx->user_bufs) | |
7133 | return -ENOMEM; | |
7134 | ||
7135 | for (i = 0; i < nr_args; i++) { | |
7136 | struct io_mapped_ubuf *imu = &ctx->user_bufs[i]; | |
7137 | unsigned long off, start, end, ubuf; | |
7138 | int pret, nr_pages; | |
7139 | struct iovec iov; | |
7140 | size_t size; | |
7141 | ||
7142 | ret = io_copy_iov(ctx, &iov, arg, i); | |
7143 | if (ret) | |
a278682d | 7144 | goto err; |
edafccee JA |
7145 | |
7146 | /* | |
7147 | * Don't impose further limits on the size and buffer | |
7148 | * constraints here, we'll -EINVAL later when IO is | |
7149 | * submitted if they are wrong. | |
7150 | */ | |
7151 | ret = -EFAULT; | |
7152 | if (!iov.iov_base || !iov.iov_len) | |
7153 | goto err; | |
7154 | ||
7155 | /* arbitrary limit, but we need something */ | |
7156 | if (iov.iov_len > SZ_1G) | |
7157 | goto err; | |
7158 | ||
7159 | ubuf = (unsigned long) iov.iov_base; | |
7160 | end = (ubuf + iov.iov_len + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
7161 | start = ubuf >> PAGE_SHIFT; | |
7162 | nr_pages = end - start; | |
7163 | ||
7164 | if (ctx->account_mem) { | |
7165 | ret = io_account_mem(ctx->user, nr_pages); | |
7166 | if (ret) | |
7167 | goto err; | |
7168 | } | |
7169 | ||
7170 | ret = 0; | |
7171 | if (!pages || nr_pages > got_pages) { | |
a8c73c1a DE |
7172 | kvfree(vmas); |
7173 | kvfree(pages); | |
d4ef6475 | 7174 | pages = kvmalloc_array(nr_pages, sizeof(struct page *), |
edafccee | 7175 | GFP_KERNEL); |
d4ef6475 | 7176 | vmas = kvmalloc_array(nr_pages, |
edafccee JA |
7177 | sizeof(struct vm_area_struct *), |
7178 | GFP_KERNEL); | |
7179 | if (!pages || !vmas) { | |
7180 | ret = -ENOMEM; | |
7181 | if (ctx->account_mem) | |
7182 | io_unaccount_mem(ctx->user, nr_pages); | |
7183 | goto err; | |
7184 | } | |
7185 | got_pages = nr_pages; | |
7186 | } | |
7187 | ||
d4ef6475 | 7188 | imu->bvec = kvmalloc_array(nr_pages, sizeof(struct bio_vec), |
edafccee JA |
7189 | GFP_KERNEL); |
7190 | ret = -ENOMEM; | |
7191 | if (!imu->bvec) { | |
7192 | if (ctx->account_mem) | |
7193 | io_unaccount_mem(ctx->user, nr_pages); | |
7194 | goto err; | |
7195 | } | |
7196 | ||
7197 | ret = 0; | |
d8ed45c5 | 7198 | mmap_read_lock(current->mm); |
2113b05d | 7199 | pret = pin_user_pages(ubuf, nr_pages, |
932f4a63 IW |
7200 | FOLL_WRITE | FOLL_LONGTERM, |
7201 | pages, vmas); | |
edafccee JA |
7202 | if (pret == nr_pages) { |
7203 | /* don't support file backed memory */ | |
7204 | for (j = 0; j < nr_pages; j++) { | |
7205 | struct vm_area_struct *vma = vmas[j]; | |
7206 | ||
7207 | if (vma->vm_file && | |
7208 | !is_file_hugepages(vma->vm_file)) { | |
7209 | ret = -EOPNOTSUPP; | |
7210 | break; | |
7211 | } | |
7212 | } | |
7213 | } else { | |
7214 | ret = pret < 0 ? pret : -EFAULT; | |
7215 | } | |
d8ed45c5 | 7216 | mmap_read_unlock(current->mm); |
edafccee JA |
7217 | if (ret) { |
7218 | /* | |
7219 | * if we did partial map, or found file backed vmas, | |
7220 | * release any pages we did get | |
7221 | */ | |
27c4d3a3 | 7222 | if (pret > 0) |
f1f6a7dd | 7223 | unpin_user_pages(pages, pret); |
edafccee JA |
7224 | if (ctx->account_mem) |
7225 | io_unaccount_mem(ctx->user, nr_pages); | |
d4ef6475 | 7226 | kvfree(imu->bvec); |
edafccee JA |
7227 | goto err; |
7228 | } | |
7229 | ||
7230 | off = ubuf & ~PAGE_MASK; | |
7231 | size = iov.iov_len; | |
7232 | for (j = 0; j < nr_pages; j++) { | |
7233 | size_t vec_len; | |
7234 | ||
7235 | vec_len = min_t(size_t, size, PAGE_SIZE - off); | |
7236 | imu->bvec[j].bv_page = pages[j]; | |
7237 | imu->bvec[j].bv_len = vec_len; | |
7238 | imu->bvec[j].bv_offset = off; | |
7239 | off = 0; | |
7240 | size -= vec_len; | |
7241 | } | |
7242 | /* store original address for later verification */ | |
7243 | imu->ubuf = ubuf; | |
7244 | imu->len = iov.iov_len; | |
7245 | imu->nr_bvecs = nr_pages; | |
7246 | ||
7247 | ctx->nr_user_bufs++; | |
7248 | } | |
d4ef6475 MR |
7249 | kvfree(pages); |
7250 | kvfree(vmas); | |
edafccee JA |
7251 | return 0; |
7252 | err: | |
d4ef6475 MR |
7253 | kvfree(pages); |
7254 | kvfree(vmas); | |
edafccee JA |
7255 | io_sqe_buffer_unregister(ctx); |
7256 | return ret; | |
7257 | } | |
7258 | ||
9b402849 JA |
7259 | static int io_eventfd_register(struct io_ring_ctx *ctx, void __user *arg) |
7260 | { | |
7261 | __s32 __user *fds = arg; | |
7262 | int fd; | |
7263 | ||
7264 | if (ctx->cq_ev_fd) | |
7265 | return -EBUSY; | |
7266 | ||
7267 | if (copy_from_user(&fd, fds, sizeof(*fds))) | |
7268 | return -EFAULT; | |
7269 | ||
7270 | ctx->cq_ev_fd = eventfd_ctx_fdget(fd); | |
7271 | if (IS_ERR(ctx->cq_ev_fd)) { | |
7272 | int ret = PTR_ERR(ctx->cq_ev_fd); | |
7273 | ctx->cq_ev_fd = NULL; | |
7274 | return ret; | |
7275 | } | |
7276 | ||
7277 | return 0; | |
7278 | } | |
7279 | ||
7280 | static int io_eventfd_unregister(struct io_ring_ctx *ctx) | |
7281 | { | |
7282 | if (ctx->cq_ev_fd) { | |
7283 | eventfd_ctx_put(ctx->cq_ev_fd); | |
7284 | ctx->cq_ev_fd = NULL; | |
7285 | return 0; | |
7286 | } | |
7287 | ||
7288 | return -ENXIO; | |
7289 | } | |
7290 | ||
5a2e745d JA |
7291 | static int __io_destroy_buffers(int id, void *p, void *data) |
7292 | { | |
7293 | struct io_ring_ctx *ctx = data; | |
7294 | struct io_buffer *buf = p; | |
7295 | ||
067524e9 | 7296 | __io_remove_buffers(ctx, buf, id, -1U); |
5a2e745d JA |
7297 | return 0; |
7298 | } | |
7299 | ||
7300 | static void io_destroy_buffers(struct io_ring_ctx *ctx) | |
7301 | { | |
7302 | idr_for_each(&ctx->io_buffer_idr, __io_destroy_buffers, ctx); | |
7303 | idr_destroy(&ctx->io_buffer_idr); | |
7304 | } | |
7305 | ||
2b188cc1 JA |
7306 | static void io_ring_ctx_free(struct io_ring_ctx *ctx) |
7307 | { | |
6b06314c | 7308 | io_finish_async(ctx); |
2b188cc1 JA |
7309 | if (ctx->sqo_mm) |
7310 | mmdrop(ctx->sqo_mm); | |
def596e9 JA |
7311 | |
7312 | io_iopoll_reap_events(ctx); | |
edafccee | 7313 | io_sqe_buffer_unregister(ctx); |
6b06314c | 7314 | io_sqe_files_unregister(ctx); |
9b402849 | 7315 | io_eventfd_unregister(ctx); |
5a2e745d | 7316 | io_destroy_buffers(ctx); |
41726c9a | 7317 | idr_destroy(&ctx->personality_idr); |
def596e9 | 7318 | |
2b188cc1 | 7319 | #if defined(CONFIG_UNIX) |
355e8d26 EB |
7320 | if (ctx->ring_sock) { |
7321 | ctx->ring_sock->file = NULL; /* so that iput() is called */ | |
2b188cc1 | 7322 | sock_release(ctx->ring_sock); |
355e8d26 | 7323 | } |
2b188cc1 JA |
7324 | #endif |
7325 | ||
75b28aff | 7326 | io_mem_free(ctx->rings); |
2b188cc1 | 7327 | io_mem_free(ctx->sq_sqes); |
2b188cc1 JA |
7328 | |
7329 | percpu_ref_exit(&ctx->refs); | |
7330 | if (ctx->account_mem) | |
7331 | io_unaccount_mem(ctx->user, | |
7332 | ring_pages(ctx->sq_entries, ctx->cq_entries)); | |
7333 | free_uid(ctx->user); | |
181e448d | 7334 | put_cred(ctx->creds); |
78076bb6 | 7335 | kfree(ctx->cancel_hash); |
0ddf92e8 | 7336 | kmem_cache_free(req_cachep, ctx->fallback_req); |
2b188cc1 JA |
7337 | kfree(ctx); |
7338 | } | |
7339 | ||
7340 | static __poll_t io_uring_poll(struct file *file, poll_table *wait) | |
7341 | { | |
7342 | struct io_ring_ctx *ctx = file->private_data; | |
7343 | __poll_t mask = 0; | |
7344 | ||
7345 | poll_wait(file, &ctx->cq_wait, wait); | |
4f7067c3 SB |
7346 | /* |
7347 | * synchronizes with barrier from wq_has_sleeper call in | |
7348 | * io_commit_cqring | |
7349 | */ | |
2b188cc1 | 7350 | smp_rmb(); |
75b28aff HV |
7351 | if (READ_ONCE(ctx->rings->sq.tail) - ctx->cached_sq_head != |
7352 | ctx->rings->sq_ring_entries) | |
2b188cc1 | 7353 | mask |= EPOLLOUT | EPOLLWRNORM; |
63e5d81f | 7354 | if (io_cqring_events(ctx, false)) |
2b188cc1 JA |
7355 | mask |= EPOLLIN | EPOLLRDNORM; |
7356 | ||
7357 | return mask; | |
7358 | } | |
7359 | ||
7360 | static int io_uring_fasync(int fd, struct file *file, int on) | |
7361 | { | |
7362 | struct io_ring_ctx *ctx = file->private_data; | |
7363 | ||
7364 | return fasync_helper(fd, file, on, &ctx->cq_fasync); | |
7365 | } | |
7366 | ||
071698e1 JA |
7367 | static int io_remove_personalities(int id, void *p, void *data) |
7368 | { | |
7369 | struct io_ring_ctx *ctx = data; | |
7370 | const struct cred *cred; | |
7371 | ||
7372 | cred = idr_remove(&ctx->personality_idr, id); | |
7373 | if (cred) | |
7374 | put_cred(cred); | |
7375 | return 0; | |
7376 | } | |
7377 | ||
85faa7b8 JA |
7378 | static void io_ring_exit_work(struct work_struct *work) |
7379 | { | |
7380 | struct io_ring_ctx *ctx; | |
7381 | ||
7382 | ctx = container_of(work, struct io_ring_ctx, exit_work); | |
7383 | if (ctx->rings) | |
7384 | io_cqring_overflow_flush(ctx, true); | |
7385 | ||
56952e91 JA |
7386 | /* |
7387 | * If we're doing polled IO and end up having requests being | |
7388 | * submitted async (out-of-line), then completions can come in while | |
7389 | * we're waiting for refs to drop. We need to reap these manually, | |
7390 | * as nobody else will be looking for them. | |
7391 | */ | |
7392 | while (!wait_for_completion_timeout(&ctx->ref_comp, HZ/20)) { | |
7393 | io_iopoll_reap_events(ctx); | |
7394 | if (ctx->rings) | |
7395 | io_cqring_overflow_flush(ctx, true); | |
7396 | } | |
85faa7b8 JA |
7397 | io_ring_ctx_free(ctx); |
7398 | } | |
7399 | ||
2b188cc1 JA |
7400 | static void io_ring_ctx_wait_and_kill(struct io_ring_ctx *ctx) |
7401 | { | |
7402 | mutex_lock(&ctx->uring_lock); | |
7403 | percpu_ref_kill(&ctx->refs); | |
7404 | mutex_unlock(&ctx->uring_lock); | |
7405 | ||
5262f567 | 7406 | io_kill_timeouts(ctx); |
221c5eb2 | 7407 | io_poll_remove_all(ctx); |
561fb04a JA |
7408 | |
7409 | if (ctx->io_wq) | |
7410 | io_wq_cancel_all(ctx->io_wq); | |
7411 | ||
def596e9 | 7412 | io_iopoll_reap_events(ctx); |
15dff286 JA |
7413 | /* if we failed setting up the ctx, we might not have any rings */ |
7414 | if (ctx->rings) | |
7415 | io_cqring_overflow_flush(ctx, true); | |
071698e1 | 7416 | idr_for_each(&ctx->personality_idr, io_remove_personalities, ctx); |
85faa7b8 JA |
7417 | INIT_WORK(&ctx->exit_work, io_ring_exit_work); |
7418 | queue_work(system_wq, &ctx->exit_work); | |
2b188cc1 JA |
7419 | } |
7420 | ||
7421 | static int io_uring_release(struct inode *inode, struct file *file) | |
7422 | { | |
7423 | struct io_ring_ctx *ctx = file->private_data; | |
7424 | ||
7425 | file->private_data = NULL; | |
7426 | io_ring_ctx_wait_and_kill(ctx); | |
7427 | return 0; | |
7428 | } | |
7429 | ||
67c4d9e6 PB |
7430 | static bool io_wq_files_match(struct io_wq_work *work, void *data) |
7431 | { | |
7432 | struct files_struct *files = data; | |
7433 | ||
7434 | return work->files == files; | |
7435 | } | |
7436 | ||
fcb323cc JA |
7437 | static void io_uring_cancel_files(struct io_ring_ctx *ctx, |
7438 | struct files_struct *files) | |
7439 | { | |
67c4d9e6 PB |
7440 | if (list_empty_careful(&ctx->inflight_list)) |
7441 | return; | |
7442 | ||
7443 | /* cancel all at once, should be faster than doing it one by one*/ | |
7444 | io_wq_cancel_cb(ctx->io_wq, io_wq_files_match, files, true); | |
7445 | ||
fcb323cc | 7446 | while (!list_empty_careful(&ctx->inflight_list)) { |
d8f1b971 XW |
7447 | struct io_kiocb *cancel_req = NULL, *req; |
7448 | DEFINE_WAIT(wait); | |
fcb323cc JA |
7449 | |
7450 | spin_lock_irq(&ctx->inflight_lock); | |
7451 | list_for_each_entry(req, &ctx->inflight_list, inflight_entry) { | |
768134d4 JA |
7452 | if (req->work.files != files) |
7453 | continue; | |
7454 | /* req is being completed, ignore */ | |
7455 | if (!refcount_inc_not_zero(&req->refs)) | |
7456 | continue; | |
7457 | cancel_req = req; | |
7458 | break; | |
fcb323cc | 7459 | } |
768134d4 | 7460 | if (cancel_req) |
fcb323cc | 7461 | prepare_to_wait(&ctx->inflight_wait, &wait, |
768134d4 | 7462 | TASK_UNINTERRUPTIBLE); |
fcb323cc JA |
7463 | spin_unlock_irq(&ctx->inflight_lock); |
7464 | ||
768134d4 JA |
7465 | /* We need to keep going until we don't find a matching req */ |
7466 | if (!cancel_req) | |
fcb323cc | 7467 | break; |
2f6d9b9d | 7468 | |
2ca10259 JA |
7469 | if (cancel_req->flags & REQ_F_OVERFLOW) { |
7470 | spin_lock_irq(&ctx->completion_lock); | |
7471 | list_del(&cancel_req->list); | |
7472 | cancel_req->flags &= ~REQ_F_OVERFLOW; | |
7473 | if (list_empty(&ctx->cq_overflow_list)) { | |
7474 | clear_bit(0, &ctx->sq_check_overflow); | |
7475 | clear_bit(0, &ctx->cq_check_overflow); | |
7476 | } | |
7477 | spin_unlock_irq(&ctx->completion_lock); | |
7478 | ||
7479 | WRITE_ONCE(ctx->rings->cq_overflow, | |
7480 | atomic_inc_return(&ctx->cached_cq_overflow)); | |
7481 | ||
7482 | /* | |
7483 | * Put inflight ref and overflow ref. If that's | |
7484 | * all we had, then we're done with this request. | |
7485 | */ | |
7486 | if (refcount_sub_and_test(2, &cancel_req->refs)) { | |
4518a3cc | 7487 | io_free_req(cancel_req); |
d8f1b971 | 7488 | finish_wait(&ctx->inflight_wait, &wait); |
2ca10259 JA |
7489 | continue; |
7490 | } | |
7b53d598 PB |
7491 | } else { |
7492 | io_wq_cancel_work(ctx->io_wq, &cancel_req->work); | |
7493 | io_put_req(cancel_req); | |
2ca10259 JA |
7494 | } |
7495 | ||
fcb323cc | 7496 | schedule(); |
d8f1b971 | 7497 | finish_wait(&ctx->inflight_wait, &wait); |
fcb323cc JA |
7498 | } |
7499 | } | |
7500 | ||
801dd57b | 7501 | static bool io_cancel_task_cb(struct io_wq_work *work, void *data) |
44e728b8 | 7502 | { |
801dd57b PB |
7503 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); |
7504 | struct task_struct *task = data; | |
44e728b8 | 7505 | |
801dd57b | 7506 | return req->task == task; |
44e728b8 PB |
7507 | } |
7508 | ||
fcb323cc JA |
7509 | static int io_uring_flush(struct file *file, void *data) |
7510 | { | |
7511 | struct io_ring_ctx *ctx = file->private_data; | |
7512 | ||
7513 | io_uring_cancel_files(ctx, data); | |
6ab23144 JA |
7514 | |
7515 | /* | |
7516 | * If the task is going away, cancel work it may have pending | |
7517 | */ | |
801dd57b PB |
7518 | if (fatal_signal_pending(current) || (current->flags & PF_EXITING)) |
7519 | io_wq_cancel_cb(ctx->io_wq, io_cancel_task_cb, current, true); | |
6ab23144 | 7520 | |
fcb323cc JA |
7521 | return 0; |
7522 | } | |
7523 | ||
6c5c240e RP |
7524 | static void *io_uring_validate_mmap_request(struct file *file, |
7525 | loff_t pgoff, size_t sz) | |
2b188cc1 | 7526 | { |
2b188cc1 | 7527 | struct io_ring_ctx *ctx = file->private_data; |
6c5c240e | 7528 | loff_t offset = pgoff << PAGE_SHIFT; |
2b188cc1 JA |
7529 | struct page *page; |
7530 | void *ptr; | |
7531 | ||
7532 | switch (offset) { | |
7533 | case IORING_OFF_SQ_RING: | |
75b28aff HV |
7534 | case IORING_OFF_CQ_RING: |
7535 | ptr = ctx->rings; | |
2b188cc1 JA |
7536 | break; |
7537 | case IORING_OFF_SQES: | |
7538 | ptr = ctx->sq_sqes; | |
7539 | break; | |
2b188cc1 | 7540 | default: |
6c5c240e | 7541 | return ERR_PTR(-EINVAL); |
2b188cc1 JA |
7542 | } |
7543 | ||
7544 | page = virt_to_head_page(ptr); | |
a50b854e | 7545 | if (sz > page_size(page)) |
6c5c240e RP |
7546 | return ERR_PTR(-EINVAL); |
7547 | ||
7548 | return ptr; | |
7549 | } | |
7550 | ||
7551 | #ifdef CONFIG_MMU | |
7552 | ||
7553 | static int io_uring_mmap(struct file *file, struct vm_area_struct *vma) | |
7554 | { | |
7555 | size_t sz = vma->vm_end - vma->vm_start; | |
7556 | unsigned long pfn; | |
7557 | void *ptr; | |
7558 | ||
7559 | ptr = io_uring_validate_mmap_request(file, vma->vm_pgoff, sz); | |
7560 | if (IS_ERR(ptr)) | |
7561 | return PTR_ERR(ptr); | |
2b188cc1 JA |
7562 | |
7563 | pfn = virt_to_phys(ptr) >> PAGE_SHIFT; | |
7564 | return remap_pfn_range(vma, vma->vm_start, pfn, sz, vma->vm_page_prot); | |
7565 | } | |
7566 | ||
6c5c240e RP |
7567 | #else /* !CONFIG_MMU */ |
7568 | ||
7569 | static int io_uring_mmap(struct file *file, struct vm_area_struct *vma) | |
7570 | { | |
7571 | return vma->vm_flags & (VM_SHARED | VM_MAYSHARE) ? 0 : -EINVAL; | |
7572 | } | |
7573 | ||
7574 | static unsigned int io_uring_nommu_mmap_capabilities(struct file *file) | |
7575 | { | |
7576 | return NOMMU_MAP_DIRECT | NOMMU_MAP_READ | NOMMU_MAP_WRITE; | |
7577 | } | |
7578 | ||
7579 | static unsigned long io_uring_nommu_get_unmapped_area(struct file *file, | |
7580 | unsigned long addr, unsigned long len, | |
7581 | unsigned long pgoff, unsigned long flags) | |
7582 | { | |
7583 | void *ptr; | |
7584 | ||
7585 | ptr = io_uring_validate_mmap_request(file, pgoff, len); | |
7586 | if (IS_ERR(ptr)) | |
7587 | return PTR_ERR(ptr); | |
7588 | ||
7589 | return (unsigned long) ptr; | |
7590 | } | |
7591 | ||
7592 | #endif /* !CONFIG_MMU */ | |
7593 | ||
2b188cc1 JA |
7594 | SYSCALL_DEFINE6(io_uring_enter, unsigned int, fd, u32, to_submit, |
7595 | u32, min_complete, u32, flags, const sigset_t __user *, sig, | |
7596 | size_t, sigsz) | |
7597 | { | |
7598 | struct io_ring_ctx *ctx; | |
7599 | long ret = -EBADF; | |
7600 | int submitted = 0; | |
7601 | struct fd f; | |
7602 | ||
b41e9852 JA |
7603 | if (current->task_works) |
7604 | task_work_run(); | |
7605 | ||
6c271ce2 | 7606 | if (flags & ~(IORING_ENTER_GETEVENTS | IORING_ENTER_SQ_WAKEUP)) |
2b188cc1 JA |
7607 | return -EINVAL; |
7608 | ||
7609 | f = fdget(fd); | |
7610 | if (!f.file) | |
7611 | return -EBADF; | |
7612 | ||
7613 | ret = -EOPNOTSUPP; | |
7614 | if (f.file->f_op != &io_uring_fops) | |
7615 | goto out_fput; | |
7616 | ||
7617 | ret = -ENXIO; | |
7618 | ctx = f.file->private_data; | |
7619 | if (!percpu_ref_tryget(&ctx->refs)) | |
7620 | goto out_fput; | |
7621 | ||
6c271ce2 JA |
7622 | /* |
7623 | * For SQ polling, the thread will do all submissions and completions. | |
7624 | * Just return the requested submit count, and wake the thread if | |
7625 | * we were asked to. | |
7626 | */ | |
b2a9eada | 7627 | ret = 0; |
6c271ce2 | 7628 | if (ctx->flags & IORING_SETUP_SQPOLL) { |
c1edbf5f JA |
7629 | if (!list_empty_careful(&ctx->cq_overflow_list)) |
7630 | io_cqring_overflow_flush(ctx, false); | |
6c271ce2 JA |
7631 | if (flags & IORING_ENTER_SQ_WAKEUP) |
7632 | wake_up(&ctx->sqo_wait); | |
7633 | submitted = to_submit; | |
b2a9eada | 7634 | } else if (to_submit) { |
2b188cc1 | 7635 | mutex_lock(&ctx->uring_lock); |
0cdaf760 | 7636 | submitted = io_submit_sqes(ctx, to_submit, f.file, fd); |
2b188cc1 | 7637 | mutex_unlock(&ctx->uring_lock); |
7c504e65 PB |
7638 | |
7639 | if (submitted != to_submit) | |
7640 | goto out; | |
2b188cc1 JA |
7641 | } |
7642 | if (flags & IORING_ENTER_GETEVENTS) { | |
def596e9 JA |
7643 | unsigned nr_events = 0; |
7644 | ||
2b188cc1 JA |
7645 | min_complete = min(min_complete, ctx->cq_entries); |
7646 | ||
32b2244a XW |
7647 | /* |
7648 | * When SETUP_IOPOLL and SETUP_SQPOLL are both enabled, user | |
7649 | * space applications don't need to do io completion events | |
7650 | * polling again, they can rely on io_sq_thread to do polling | |
7651 | * work, which can reduce cpu usage and uring_lock contention. | |
7652 | */ | |
7653 | if (ctx->flags & IORING_SETUP_IOPOLL && | |
7654 | !(ctx->flags & IORING_SETUP_SQPOLL)) { | |
def596e9 | 7655 | ret = io_iopoll_check(ctx, &nr_events, min_complete); |
def596e9 JA |
7656 | } else { |
7657 | ret = io_cqring_wait(ctx, min_complete, sig, sigsz); | |
7658 | } | |
2b188cc1 JA |
7659 | } |
7660 | ||
7c504e65 | 7661 | out: |
6805b32e | 7662 | percpu_ref_put(&ctx->refs); |
2b188cc1 JA |
7663 | out_fput: |
7664 | fdput(f); | |
7665 | return submitted ? submitted : ret; | |
7666 | } | |
7667 | ||
bebdb65e | 7668 | #ifdef CONFIG_PROC_FS |
87ce955b JA |
7669 | static int io_uring_show_cred(int id, void *p, void *data) |
7670 | { | |
7671 | const struct cred *cred = p; | |
7672 | struct seq_file *m = data; | |
7673 | struct user_namespace *uns = seq_user_ns(m); | |
7674 | struct group_info *gi; | |
7675 | kernel_cap_t cap; | |
7676 | unsigned __capi; | |
7677 | int g; | |
7678 | ||
7679 | seq_printf(m, "%5d\n", id); | |
7680 | seq_put_decimal_ull(m, "\tUid:\t", from_kuid_munged(uns, cred->uid)); | |
7681 | seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->euid)); | |
7682 | seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->suid)); | |
7683 | seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->fsuid)); | |
7684 | seq_put_decimal_ull(m, "\n\tGid:\t", from_kgid_munged(uns, cred->gid)); | |
7685 | seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->egid)); | |
7686 | seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->sgid)); | |
7687 | seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->fsgid)); | |
7688 | seq_puts(m, "\n\tGroups:\t"); | |
7689 | gi = cred->group_info; | |
7690 | for (g = 0; g < gi->ngroups; g++) { | |
7691 | seq_put_decimal_ull(m, g ? " " : "", | |
7692 | from_kgid_munged(uns, gi->gid[g])); | |
7693 | } | |
7694 | seq_puts(m, "\n\tCapEff:\t"); | |
7695 | cap = cred->cap_effective; | |
7696 | CAP_FOR_EACH_U32(__capi) | |
7697 | seq_put_hex_ll(m, NULL, cap.cap[CAP_LAST_U32 - __capi], 8); | |
7698 | seq_putc(m, '\n'); | |
7699 | return 0; | |
7700 | } | |
7701 | ||
7702 | static void __io_uring_show_fdinfo(struct io_ring_ctx *ctx, struct seq_file *m) | |
7703 | { | |
7704 | int i; | |
7705 | ||
7706 | mutex_lock(&ctx->uring_lock); | |
7707 | seq_printf(m, "UserFiles:\t%u\n", ctx->nr_user_files); | |
7708 | for (i = 0; i < ctx->nr_user_files; i++) { | |
7709 | struct fixed_file_table *table; | |
7710 | struct file *f; | |
7711 | ||
7712 | table = &ctx->file_data->table[i >> IORING_FILE_TABLE_SHIFT]; | |
7713 | f = table->files[i & IORING_FILE_TABLE_MASK]; | |
7714 | if (f) | |
7715 | seq_printf(m, "%5u: %s\n", i, file_dentry(f)->d_iname); | |
7716 | else | |
7717 | seq_printf(m, "%5u: <none>\n", i); | |
7718 | } | |
7719 | seq_printf(m, "UserBufs:\t%u\n", ctx->nr_user_bufs); | |
7720 | for (i = 0; i < ctx->nr_user_bufs; i++) { | |
7721 | struct io_mapped_ubuf *buf = &ctx->user_bufs[i]; | |
7722 | ||
7723 | seq_printf(m, "%5u: 0x%llx/%u\n", i, buf->ubuf, | |
7724 | (unsigned int) buf->len); | |
7725 | } | |
7726 | if (!idr_is_empty(&ctx->personality_idr)) { | |
7727 | seq_printf(m, "Personalities:\n"); | |
7728 | idr_for_each(&ctx->personality_idr, io_uring_show_cred, m); | |
7729 | } | |
d7718a9d JA |
7730 | seq_printf(m, "PollList:\n"); |
7731 | spin_lock_irq(&ctx->completion_lock); | |
7732 | for (i = 0; i < (1U << ctx->cancel_hash_bits); i++) { | |
7733 | struct hlist_head *list = &ctx->cancel_hash[i]; | |
7734 | struct io_kiocb *req; | |
7735 | ||
7736 | hlist_for_each_entry(req, list, hash_node) | |
7737 | seq_printf(m, " op=%d, task_works=%d\n", req->opcode, | |
7738 | req->task->task_works != NULL); | |
7739 | } | |
7740 | spin_unlock_irq(&ctx->completion_lock); | |
87ce955b JA |
7741 | mutex_unlock(&ctx->uring_lock); |
7742 | } | |
7743 | ||
7744 | static void io_uring_show_fdinfo(struct seq_file *m, struct file *f) | |
7745 | { | |
7746 | struct io_ring_ctx *ctx = f->private_data; | |
7747 | ||
7748 | if (percpu_ref_tryget(&ctx->refs)) { | |
7749 | __io_uring_show_fdinfo(ctx, m); | |
7750 | percpu_ref_put(&ctx->refs); | |
7751 | } | |
7752 | } | |
bebdb65e | 7753 | #endif |
87ce955b | 7754 | |
2b188cc1 JA |
7755 | static const struct file_operations io_uring_fops = { |
7756 | .release = io_uring_release, | |
fcb323cc | 7757 | .flush = io_uring_flush, |
2b188cc1 | 7758 | .mmap = io_uring_mmap, |
6c5c240e RP |
7759 | #ifndef CONFIG_MMU |
7760 | .get_unmapped_area = io_uring_nommu_get_unmapped_area, | |
7761 | .mmap_capabilities = io_uring_nommu_mmap_capabilities, | |
7762 | #endif | |
2b188cc1 JA |
7763 | .poll = io_uring_poll, |
7764 | .fasync = io_uring_fasync, | |
bebdb65e | 7765 | #ifdef CONFIG_PROC_FS |
87ce955b | 7766 | .show_fdinfo = io_uring_show_fdinfo, |
bebdb65e | 7767 | #endif |
2b188cc1 JA |
7768 | }; |
7769 | ||
7770 | static int io_allocate_scq_urings(struct io_ring_ctx *ctx, | |
7771 | struct io_uring_params *p) | |
7772 | { | |
75b28aff HV |
7773 | struct io_rings *rings; |
7774 | size_t size, sq_array_offset; | |
2b188cc1 | 7775 | |
75b28aff HV |
7776 | size = rings_size(p->sq_entries, p->cq_entries, &sq_array_offset); |
7777 | if (size == SIZE_MAX) | |
7778 | return -EOVERFLOW; | |
7779 | ||
7780 | rings = io_mem_alloc(size); | |
7781 | if (!rings) | |
2b188cc1 JA |
7782 | return -ENOMEM; |
7783 | ||
75b28aff HV |
7784 | ctx->rings = rings; |
7785 | ctx->sq_array = (u32 *)((char *)rings + sq_array_offset); | |
7786 | rings->sq_ring_mask = p->sq_entries - 1; | |
7787 | rings->cq_ring_mask = p->cq_entries - 1; | |
7788 | rings->sq_ring_entries = p->sq_entries; | |
7789 | rings->cq_ring_entries = p->cq_entries; | |
7790 | ctx->sq_mask = rings->sq_ring_mask; | |
7791 | ctx->cq_mask = rings->cq_ring_mask; | |
7792 | ctx->sq_entries = rings->sq_ring_entries; | |
7793 | ctx->cq_entries = rings->cq_ring_entries; | |
2b188cc1 JA |
7794 | |
7795 | size = array_size(sizeof(struct io_uring_sqe), p->sq_entries); | |
eb065d30 JA |
7796 | if (size == SIZE_MAX) { |
7797 | io_mem_free(ctx->rings); | |
7798 | ctx->rings = NULL; | |
2b188cc1 | 7799 | return -EOVERFLOW; |
eb065d30 | 7800 | } |
2b188cc1 JA |
7801 | |
7802 | ctx->sq_sqes = io_mem_alloc(size); | |
eb065d30 JA |
7803 | if (!ctx->sq_sqes) { |
7804 | io_mem_free(ctx->rings); | |
7805 | ctx->rings = NULL; | |
2b188cc1 | 7806 | return -ENOMEM; |
eb065d30 | 7807 | } |
2b188cc1 | 7808 | |
2b188cc1 JA |
7809 | return 0; |
7810 | } | |
7811 | ||
7812 | /* | |
7813 | * Allocate an anonymous fd, this is what constitutes the application | |
7814 | * visible backing of an io_uring instance. The application mmaps this | |
7815 | * fd to gain access to the SQ/CQ ring details. If UNIX sockets are enabled, | |
7816 | * we have to tie this fd to a socket for file garbage collection purposes. | |
7817 | */ | |
7818 | static int io_uring_get_fd(struct io_ring_ctx *ctx) | |
7819 | { | |
7820 | struct file *file; | |
7821 | int ret; | |
7822 | ||
7823 | #if defined(CONFIG_UNIX) | |
7824 | ret = sock_create_kern(&init_net, PF_UNIX, SOCK_RAW, IPPROTO_IP, | |
7825 | &ctx->ring_sock); | |
7826 | if (ret) | |
7827 | return ret; | |
7828 | #endif | |
7829 | ||
7830 | ret = get_unused_fd_flags(O_RDWR | O_CLOEXEC); | |
7831 | if (ret < 0) | |
7832 | goto err; | |
7833 | ||
7834 | file = anon_inode_getfile("[io_uring]", &io_uring_fops, ctx, | |
7835 | O_RDWR | O_CLOEXEC); | |
7836 | if (IS_ERR(file)) { | |
7837 | put_unused_fd(ret); | |
7838 | ret = PTR_ERR(file); | |
7839 | goto err; | |
7840 | } | |
7841 | ||
7842 | #if defined(CONFIG_UNIX) | |
7843 | ctx->ring_sock->file = file; | |
7844 | #endif | |
7845 | fd_install(ret, file); | |
7846 | return ret; | |
7847 | err: | |
7848 | #if defined(CONFIG_UNIX) | |
7849 | sock_release(ctx->ring_sock); | |
7850 | ctx->ring_sock = NULL; | |
7851 | #endif | |
7852 | return ret; | |
7853 | } | |
7854 | ||
7f13657d XW |
7855 | static int io_uring_create(unsigned entries, struct io_uring_params *p, |
7856 | struct io_uring_params __user *params) | |
2b188cc1 JA |
7857 | { |
7858 | struct user_struct *user = NULL; | |
7859 | struct io_ring_ctx *ctx; | |
7860 | bool account_mem; | |
7861 | int ret; | |
7862 | ||
8110c1a6 | 7863 | if (!entries) |
2b188cc1 | 7864 | return -EINVAL; |
8110c1a6 JA |
7865 | if (entries > IORING_MAX_ENTRIES) { |
7866 | if (!(p->flags & IORING_SETUP_CLAMP)) | |
7867 | return -EINVAL; | |
7868 | entries = IORING_MAX_ENTRIES; | |
7869 | } | |
2b188cc1 JA |
7870 | |
7871 | /* | |
7872 | * Use twice as many entries for the CQ ring. It's possible for the | |
7873 | * application to drive a higher depth than the size of the SQ ring, | |
7874 | * since the sqes are only used at submission time. This allows for | |
33a107f0 JA |
7875 | * some flexibility in overcommitting a bit. If the application has |
7876 | * set IORING_SETUP_CQSIZE, it will have passed in the desired number | |
7877 | * of CQ ring entries manually. | |
2b188cc1 JA |
7878 | */ |
7879 | p->sq_entries = roundup_pow_of_two(entries); | |
33a107f0 JA |
7880 | if (p->flags & IORING_SETUP_CQSIZE) { |
7881 | /* | |
7882 | * If IORING_SETUP_CQSIZE is set, we do the same roundup | |
7883 | * to a power-of-two, if it isn't already. We do NOT impose | |
7884 | * any cq vs sq ring sizing. | |
7885 | */ | |
8110c1a6 | 7886 | if (p->cq_entries < p->sq_entries) |
33a107f0 | 7887 | return -EINVAL; |
8110c1a6 JA |
7888 | if (p->cq_entries > IORING_MAX_CQ_ENTRIES) { |
7889 | if (!(p->flags & IORING_SETUP_CLAMP)) | |
7890 | return -EINVAL; | |
7891 | p->cq_entries = IORING_MAX_CQ_ENTRIES; | |
7892 | } | |
33a107f0 JA |
7893 | p->cq_entries = roundup_pow_of_two(p->cq_entries); |
7894 | } else { | |
7895 | p->cq_entries = 2 * p->sq_entries; | |
7896 | } | |
2b188cc1 JA |
7897 | |
7898 | user = get_uid(current_user()); | |
7899 | account_mem = !capable(CAP_IPC_LOCK); | |
7900 | ||
7901 | if (account_mem) { | |
7902 | ret = io_account_mem(user, | |
7903 | ring_pages(p->sq_entries, p->cq_entries)); | |
7904 | if (ret) { | |
7905 | free_uid(user); | |
7906 | return ret; | |
7907 | } | |
7908 | } | |
7909 | ||
7910 | ctx = io_ring_ctx_alloc(p); | |
7911 | if (!ctx) { | |
7912 | if (account_mem) | |
7913 | io_unaccount_mem(user, ring_pages(p->sq_entries, | |
7914 | p->cq_entries)); | |
7915 | free_uid(user); | |
7916 | return -ENOMEM; | |
7917 | } | |
7918 | ctx->compat = in_compat_syscall(); | |
7919 | ctx->account_mem = account_mem; | |
7920 | ctx->user = user; | |
0b8c0ec7 | 7921 | ctx->creds = get_current_cred(); |
2b188cc1 JA |
7922 | |
7923 | ret = io_allocate_scq_urings(ctx, p); | |
7924 | if (ret) | |
7925 | goto err; | |
7926 | ||
6c271ce2 | 7927 | ret = io_sq_offload_start(ctx, p); |
2b188cc1 JA |
7928 | if (ret) |
7929 | goto err; | |
7930 | ||
2b188cc1 | 7931 | memset(&p->sq_off, 0, sizeof(p->sq_off)); |
75b28aff HV |
7932 | p->sq_off.head = offsetof(struct io_rings, sq.head); |
7933 | p->sq_off.tail = offsetof(struct io_rings, sq.tail); | |
7934 | p->sq_off.ring_mask = offsetof(struct io_rings, sq_ring_mask); | |
7935 | p->sq_off.ring_entries = offsetof(struct io_rings, sq_ring_entries); | |
7936 | p->sq_off.flags = offsetof(struct io_rings, sq_flags); | |
7937 | p->sq_off.dropped = offsetof(struct io_rings, sq_dropped); | |
7938 | p->sq_off.array = (char *)ctx->sq_array - (char *)ctx->rings; | |
2b188cc1 JA |
7939 | |
7940 | memset(&p->cq_off, 0, sizeof(p->cq_off)); | |
75b28aff HV |
7941 | p->cq_off.head = offsetof(struct io_rings, cq.head); |
7942 | p->cq_off.tail = offsetof(struct io_rings, cq.tail); | |
7943 | p->cq_off.ring_mask = offsetof(struct io_rings, cq_ring_mask); | |
7944 | p->cq_off.ring_entries = offsetof(struct io_rings, cq_ring_entries); | |
7945 | p->cq_off.overflow = offsetof(struct io_rings, cq_overflow); | |
7946 | p->cq_off.cqes = offsetof(struct io_rings, cqes); | |
0d9b5b3a | 7947 | p->cq_off.flags = offsetof(struct io_rings, cq_flags); |
ac90f249 | 7948 | |
7f13657d XW |
7949 | p->features = IORING_FEAT_SINGLE_MMAP | IORING_FEAT_NODROP | |
7950 | IORING_FEAT_SUBMIT_STABLE | IORING_FEAT_RW_CUR_POS | | |
7951 | IORING_FEAT_CUR_PERSONALITY | IORING_FEAT_FAST_POLL; | |
7952 | ||
7953 | if (copy_to_user(params, p, sizeof(*p))) { | |
7954 | ret = -EFAULT; | |
7955 | goto err; | |
7956 | } | |
044c1ab3 JA |
7957 | /* |
7958 | * Install ring fd as the very last thing, so we don't risk someone | |
7959 | * having closed it before we finish setup | |
7960 | */ | |
7961 | ret = io_uring_get_fd(ctx); | |
7962 | if (ret < 0) | |
7963 | goto err; | |
7964 | ||
c826bd7a | 7965 | trace_io_uring_create(ret, ctx, p->sq_entries, p->cq_entries, p->flags); |
2b188cc1 JA |
7966 | return ret; |
7967 | err: | |
7968 | io_ring_ctx_wait_and_kill(ctx); | |
7969 | return ret; | |
7970 | } | |
7971 | ||
7972 | /* | |
7973 | * Sets up an aio uring context, and returns the fd. Applications asks for a | |
7974 | * ring size, we return the actual sq/cq ring sizes (among other things) in the | |
7975 | * params structure passed in. | |
7976 | */ | |
7977 | static long io_uring_setup(u32 entries, struct io_uring_params __user *params) | |
7978 | { | |
7979 | struct io_uring_params p; | |
2b188cc1 JA |
7980 | int i; |
7981 | ||
7982 | if (copy_from_user(&p, params, sizeof(p))) | |
7983 | return -EFAULT; | |
7984 | for (i = 0; i < ARRAY_SIZE(p.resv); i++) { | |
7985 | if (p.resv[i]) | |
7986 | return -EINVAL; | |
7987 | } | |
7988 | ||
6c271ce2 | 7989 | if (p.flags & ~(IORING_SETUP_IOPOLL | IORING_SETUP_SQPOLL | |
8110c1a6 | 7990 | IORING_SETUP_SQ_AFF | IORING_SETUP_CQSIZE | |
24369c2e | 7991 | IORING_SETUP_CLAMP | IORING_SETUP_ATTACH_WQ)) |
2b188cc1 JA |
7992 | return -EINVAL; |
7993 | ||
7f13657d | 7994 | return io_uring_create(entries, &p, params); |
2b188cc1 JA |
7995 | } |
7996 | ||
7997 | SYSCALL_DEFINE2(io_uring_setup, u32, entries, | |
7998 | struct io_uring_params __user *, params) | |
7999 | { | |
8000 | return io_uring_setup(entries, params); | |
8001 | } | |
8002 | ||
66f4af93 JA |
8003 | static int io_probe(struct io_ring_ctx *ctx, void __user *arg, unsigned nr_args) |
8004 | { | |
8005 | struct io_uring_probe *p; | |
8006 | size_t size; | |
8007 | int i, ret; | |
8008 | ||
8009 | size = struct_size(p, ops, nr_args); | |
8010 | if (size == SIZE_MAX) | |
8011 | return -EOVERFLOW; | |
8012 | p = kzalloc(size, GFP_KERNEL); | |
8013 | if (!p) | |
8014 | return -ENOMEM; | |
8015 | ||
8016 | ret = -EFAULT; | |
8017 | if (copy_from_user(p, arg, size)) | |
8018 | goto out; | |
8019 | ret = -EINVAL; | |
8020 | if (memchr_inv(p, 0, size)) | |
8021 | goto out; | |
8022 | ||
8023 | p->last_op = IORING_OP_LAST - 1; | |
8024 | if (nr_args > IORING_OP_LAST) | |
8025 | nr_args = IORING_OP_LAST; | |
8026 | ||
8027 | for (i = 0; i < nr_args; i++) { | |
8028 | p->ops[i].op = i; | |
8029 | if (!io_op_defs[i].not_supported) | |
8030 | p->ops[i].flags = IO_URING_OP_SUPPORTED; | |
8031 | } | |
8032 | p->ops_len = i; | |
8033 | ||
8034 | ret = 0; | |
8035 | if (copy_to_user(arg, p, size)) | |
8036 | ret = -EFAULT; | |
8037 | out: | |
8038 | kfree(p); | |
8039 | return ret; | |
8040 | } | |
8041 | ||
071698e1 JA |
8042 | static int io_register_personality(struct io_ring_ctx *ctx) |
8043 | { | |
8044 | const struct cred *creds = get_current_cred(); | |
8045 | int id; | |
8046 | ||
8047 | id = idr_alloc_cyclic(&ctx->personality_idr, (void *) creds, 1, | |
8048 | USHRT_MAX, GFP_KERNEL); | |
8049 | if (id < 0) | |
8050 | put_cred(creds); | |
8051 | return id; | |
8052 | } | |
8053 | ||
8054 | static int io_unregister_personality(struct io_ring_ctx *ctx, unsigned id) | |
8055 | { | |
8056 | const struct cred *old_creds; | |
8057 | ||
8058 | old_creds = idr_remove(&ctx->personality_idr, id); | |
8059 | if (old_creds) { | |
8060 | put_cred(old_creds); | |
8061 | return 0; | |
8062 | } | |
8063 | ||
8064 | return -EINVAL; | |
8065 | } | |
8066 | ||
8067 | static bool io_register_op_must_quiesce(int op) | |
8068 | { | |
8069 | switch (op) { | |
8070 | case IORING_UNREGISTER_FILES: | |
8071 | case IORING_REGISTER_FILES_UPDATE: | |
8072 | case IORING_REGISTER_PROBE: | |
8073 | case IORING_REGISTER_PERSONALITY: | |
8074 | case IORING_UNREGISTER_PERSONALITY: | |
8075 | return false; | |
8076 | default: | |
8077 | return true; | |
8078 | } | |
8079 | } | |
8080 | ||
edafccee JA |
8081 | static int __io_uring_register(struct io_ring_ctx *ctx, unsigned opcode, |
8082 | void __user *arg, unsigned nr_args) | |
b19062a5 JA |
8083 | __releases(ctx->uring_lock) |
8084 | __acquires(ctx->uring_lock) | |
edafccee JA |
8085 | { |
8086 | int ret; | |
8087 | ||
35fa71a0 JA |
8088 | /* |
8089 | * We're inside the ring mutex, if the ref is already dying, then | |
8090 | * someone else killed the ctx or is already going through | |
8091 | * io_uring_register(). | |
8092 | */ | |
8093 | if (percpu_ref_is_dying(&ctx->refs)) | |
8094 | return -ENXIO; | |
8095 | ||
071698e1 | 8096 | if (io_register_op_must_quiesce(opcode)) { |
05f3fb3c | 8097 | percpu_ref_kill(&ctx->refs); |
b19062a5 | 8098 | |
05f3fb3c JA |
8099 | /* |
8100 | * Drop uring mutex before waiting for references to exit. If | |
8101 | * another thread is currently inside io_uring_enter() it might | |
8102 | * need to grab the uring_lock to make progress. If we hold it | |
8103 | * here across the drain wait, then we can deadlock. It's safe | |
8104 | * to drop the mutex here, since no new references will come in | |
8105 | * after we've killed the percpu ref. | |
8106 | */ | |
8107 | mutex_unlock(&ctx->uring_lock); | |
0f158b4c | 8108 | ret = wait_for_completion_interruptible(&ctx->ref_comp); |
05f3fb3c | 8109 | mutex_lock(&ctx->uring_lock); |
c150368b JA |
8110 | if (ret) { |
8111 | percpu_ref_resurrect(&ctx->refs); | |
8112 | ret = -EINTR; | |
8113 | goto out; | |
8114 | } | |
05f3fb3c | 8115 | } |
edafccee JA |
8116 | |
8117 | switch (opcode) { | |
8118 | case IORING_REGISTER_BUFFERS: | |
8119 | ret = io_sqe_buffer_register(ctx, arg, nr_args); | |
8120 | break; | |
8121 | case IORING_UNREGISTER_BUFFERS: | |
8122 | ret = -EINVAL; | |
8123 | if (arg || nr_args) | |
8124 | break; | |
8125 | ret = io_sqe_buffer_unregister(ctx); | |
8126 | break; | |
6b06314c JA |
8127 | case IORING_REGISTER_FILES: |
8128 | ret = io_sqe_files_register(ctx, arg, nr_args); | |
8129 | break; | |
8130 | case IORING_UNREGISTER_FILES: | |
8131 | ret = -EINVAL; | |
8132 | if (arg || nr_args) | |
8133 | break; | |
8134 | ret = io_sqe_files_unregister(ctx); | |
8135 | break; | |
c3a31e60 JA |
8136 | case IORING_REGISTER_FILES_UPDATE: |
8137 | ret = io_sqe_files_update(ctx, arg, nr_args); | |
8138 | break; | |
9b402849 | 8139 | case IORING_REGISTER_EVENTFD: |
f2842ab5 | 8140 | case IORING_REGISTER_EVENTFD_ASYNC: |
9b402849 JA |
8141 | ret = -EINVAL; |
8142 | if (nr_args != 1) | |
8143 | break; | |
8144 | ret = io_eventfd_register(ctx, arg); | |
f2842ab5 JA |
8145 | if (ret) |
8146 | break; | |
8147 | if (opcode == IORING_REGISTER_EVENTFD_ASYNC) | |
8148 | ctx->eventfd_async = 1; | |
8149 | else | |
8150 | ctx->eventfd_async = 0; | |
9b402849 JA |
8151 | break; |
8152 | case IORING_UNREGISTER_EVENTFD: | |
8153 | ret = -EINVAL; | |
8154 | if (arg || nr_args) | |
8155 | break; | |
8156 | ret = io_eventfd_unregister(ctx); | |
8157 | break; | |
66f4af93 JA |
8158 | case IORING_REGISTER_PROBE: |
8159 | ret = -EINVAL; | |
8160 | if (!arg || nr_args > 256) | |
8161 | break; | |
8162 | ret = io_probe(ctx, arg, nr_args); | |
8163 | break; | |
071698e1 JA |
8164 | case IORING_REGISTER_PERSONALITY: |
8165 | ret = -EINVAL; | |
8166 | if (arg || nr_args) | |
8167 | break; | |
8168 | ret = io_register_personality(ctx); | |
8169 | break; | |
8170 | case IORING_UNREGISTER_PERSONALITY: | |
8171 | ret = -EINVAL; | |
8172 | if (arg) | |
8173 | break; | |
8174 | ret = io_unregister_personality(ctx, nr_args); | |
8175 | break; | |
edafccee JA |
8176 | default: |
8177 | ret = -EINVAL; | |
8178 | break; | |
8179 | } | |
8180 | ||
071698e1 | 8181 | if (io_register_op_must_quiesce(opcode)) { |
05f3fb3c | 8182 | /* bring the ctx back to life */ |
05f3fb3c | 8183 | percpu_ref_reinit(&ctx->refs); |
c150368b | 8184 | out: |
0f158b4c | 8185 | reinit_completion(&ctx->ref_comp); |
05f3fb3c | 8186 | } |
edafccee JA |
8187 | return ret; |
8188 | } | |
8189 | ||
8190 | SYSCALL_DEFINE4(io_uring_register, unsigned int, fd, unsigned int, opcode, | |
8191 | void __user *, arg, unsigned int, nr_args) | |
8192 | { | |
8193 | struct io_ring_ctx *ctx; | |
8194 | long ret = -EBADF; | |
8195 | struct fd f; | |
8196 | ||
8197 | f = fdget(fd); | |
8198 | if (!f.file) | |
8199 | return -EBADF; | |
8200 | ||
8201 | ret = -EOPNOTSUPP; | |
8202 | if (f.file->f_op != &io_uring_fops) | |
8203 | goto out_fput; | |
8204 | ||
8205 | ctx = f.file->private_data; | |
8206 | ||
8207 | mutex_lock(&ctx->uring_lock); | |
8208 | ret = __io_uring_register(ctx, opcode, arg, nr_args); | |
8209 | mutex_unlock(&ctx->uring_lock); | |
c826bd7a DD |
8210 | trace_io_uring_register(ctx, opcode, ctx->nr_user_files, ctx->nr_user_bufs, |
8211 | ctx->cq_ev_fd != NULL, ret); | |
edafccee JA |
8212 | out_fput: |
8213 | fdput(f); | |
8214 | return ret; | |
8215 | } | |
8216 | ||
2b188cc1 JA |
8217 | static int __init io_uring_init(void) |
8218 | { | |
d7f62e82 SM |
8219 | #define __BUILD_BUG_VERIFY_ELEMENT(stype, eoffset, etype, ename) do { \ |
8220 | BUILD_BUG_ON(offsetof(stype, ename) != eoffset); \ | |
8221 | BUILD_BUG_ON(sizeof(etype) != sizeof_field(stype, ename)); \ | |
8222 | } while (0) | |
8223 | ||
8224 | #define BUILD_BUG_SQE_ELEM(eoffset, etype, ename) \ | |
8225 | __BUILD_BUG_VERIFY_ELEMENT(struct io_uring_sqe, eoffset, etype, ename) | |
8226 | BUILD_BUG_ON(sizeof(struct io_uring_sqe) != 64); | |
8227 | BUILD_BUG_SQE_ELEM(0, __u8, opcode); | |
8228 | BUILD_BUG_SQE_ELEM(1, __u8, flags); | |
8229 | BUILD_BUG_SQE_ELEM(2, __u16, ioprio); | |
8230 | BUILD_BUG_SQE_ELEM(4, __s32, fd); | |
8231 | BUILD_BUG_SQE_ELEM(8, __u64, off); | |
8232 | BUILD_BUG_SQE_ELEM(8, __u64, addr2); | |
8233 | BUILD_BUG_SQE_ELEM(16, __u64, addr); | |
7d67af2c | 8234 | BUILD_BUG_SQE_ELEM(16, __u64, splice_off_in); |
d7f62e82 SM |
8235 | BUILD_BUG_SQE_ELEM(24, __u32, len); |
8236 | BUILD_BUG_SQE_ELEM(28, __kernel_rwf_t, rw_flags); | |
8237 | BUILD_BUG_SQE_ELEM(28, /* compat */ int, rw_flags); | |
8238 | BUILD_BUG_SQE_ELEM(28, /* compat */ __u32, rw_flags); | |
8239 | BUILD_BUG_SQE_ELEM(28, __u32, fsync_flags); | |
8240 | BUILD_BUG_SQE_ELEM(28, __u16, poll_events); | |
8241 | BUILD_BUG_SQE_ELEM(28, __u32, sync_range_flags); | |
8242 | BUILD_BUG_SQE_ELEM(28, __u32, msg_flags); | |
8243 | BUILD_BUG_SQE_ELEM(28, __u32, timeout_flags); | |
8244 | BUILD_BUG_SQE_ELEM(28, __u32, accept_flags); | |
8245 | BUILD_BUG_SQE_ELEM(28, __u32, cancel_flags); | |
8246 | BUILD_BUG_SQE_ELEM(28, __u32, open_flags); | |
8247 | BUILD_BUG_SQE_ELEM(28, __u32, statx_flags); | |
8248 | BUILD_BUG_SQE_ELEM(28, __u32, fadvise_advice); | |
7d67af2c | 8249 | BUILD_BUG_SQE_ELEM(28, __u32, splice_flags); |
d7f62e82 SM |
8250 | BUILD_BUG_SQE_ELEM(32, __u64, user_data); |
8251 | BUILD_BUG_SQE_ELEM(40, __u16, buf_index); | |
8252 | BUILD_BUG_SQE_ELEM(42, __u16, personality); | |
7d67af2c | 8253 | BUILD_BUG_SQE_ELEM(44, __s32, splice_fd_in); |
d7f62e82 | 8254 | |
d3656344 | 8255 | BUILD_BUG_ON(ARRAY_SIZE(io_op_defs) != IORING_OP_LAST); |
84557871 | 8256 | BUILD_BUG_ON(__REQ_F_LAST_BIT >= 8 * sizeof(int)); |
2b188cc1 JA |
8257 | req_cachep = KMEM_CACHE(io_kiocb, SLAB_HWCACHE_ALIGN | SLAB_PANIC); |
8258 | return 0; | |
8259 | }; | |
8260 | __initcall(io_uring_init); |