]>
Commit | Line | Data |
---|---|---|
2b188cc1 JA |
1 | // SPDX-License-Identifier: GPL-2.0 |
2 | /* | |
3 | * Shared application/kernel submission and completion ring pairs, for | |
4 | * supporting fast/efficient IO. | |
5 | * | |
6 | * A note on the read/write ordering memory barriers that are matched between | |
1e84b97b SB |
7 | * the application and kernel side. |
8 | * | |
9 | * After the application reads the CQ ring tail, it must use an | |
10 | * appropriate smp_rmb() to pair with the smp_wmb() the kernel uses | |
11 | * before writing the tail (using smp_load_acquire to read the tail will | |
12 | * do). It also needs a smp_mb() before updating CQ head (ordering the | |
13 | * entry load(s) with the head store), pairing with an implicit barrier | |
14 | * through a control-dependency in io_get_cqring (smp_store_release to | |
15 | * store head will do). Failure to do so could lead to reading invalid | |
16 | * CQ entries. | |
17 | * | |
18 | * Likewise, the application must use an appropriate smp_wmb() before | |
19 | * writing the SQ tail (ordering SQ entry stores with the tail store), | |
20 | * which pairs with smp_load_acquire in io_get_sqring (smp_store_release | |
21 | * to store the tail will do). And it needs a barrier ordering the SQ | |
22 | * head load before writing new SQ entries (smp_load_acquire to read | |
23 | * head will do). | |
24 | * | |
25 | * When using the SQ poll thread (IORING_SETUP_SQPOLL), the application | |
26 | * needs to check the SQ flags for IORING_SQ_NEED_WAKEUP *after* | |
27 | * updating the SQ tail; a full memory barrier smp_mb() is needed | |
28 | * between. | |
2b188cc1 JA |
29 | * |
30 | * Also see the examples in the liburing library: | |
31 | * | |
32 | * git://git.kernel.dk/liburing | |
33 | * | |
34 | * io_uring also uses READ/WRITE_ONCE() for _any_ store or load that happens | |
35 | * from data shared between the kernel and application. This is done both | |
36 | * for ordering purposes, but also to ensure that once a value is loaded from | |
37 | * data that the application could potentially modify, it remains stable. | |
38 | * | |
39 | * Copyright (C) 2018-2019 Jens Axboe | |
c992fe29 | 40 | * Copyright (c) 2018-2019 Christoph Hellwig |
2b188cc1 JA |
41 | */ |
42 | #include <linux/kernel.h> | |
43 | #include <linux/init.h> | |
44 | #include <linux/errno.h> | |
45 | #include <linux/syscalls.h> | |
46 | #include <linux/compat.h> | |
52de1fe1 | 47 | #include <net/compat.h> |
2b188cc1 JA |
48 | #include <linux/refcount.h> |
49 | #include <linux/uio.h> | |
6b47ee6e | 50 | #include <linux/bits.h> |
2b188cc1 JA |
51 | |
52 | #include <linux/sched/signal.h> | |
53 | #include <linux/fs.h> | |
54 | #include <linux/file.h> | |
55 | #include <linux/fdtable.h> | |
56 | #include <linux/mm.h> | |
57 | #include <linux/mman.h> | |
58 | #include <linux/mmu_context.h> | |
59 | #include <linux/percpu.h> | |
60 | #include <linux/slab.h> | |
6c271ce2 | 61 | #include <linux/kthread.h> |
2b188cc1 | 62 | #include <linux/blkdev.h> |
edafccee | 63 | #include <linux/bvec.h> |
2b188cc1 JA |
64 | #include <linux/net.h> |
65 | #include <net/sock.h> | |
66 | #include <net/af_unix.h> | |
6b06314c | 67 | #include <net/scm.h> |
2b188cc1 JA |
68 | #include <linux/anon_inodes.h> |
69 | #include <linux/sched/mm.h> | |
70 | #include <linux/uaccess.h> | |
71 | #include <linux/nospec.h> | |
edafccee JA |
72 | #include <linux/sizes.h> |
73 | #include <linux/hugetlb.h> | |
aa4c3967 | 74 | #include <linux/highmem.h> |
15b71abe JA |
75 | #include <linux/namei.h> |
76 | #include <linux/fsnotify.h> | |
4840e418 | 77 | #include <linux/fadvise.h> |
3e4827b0 | 78 | #include <linux/eventpoll.h> |
ff002b30 | 79 | #include <linux/fs_struct.h> |
7d67af2c | 80 | #include <linux/splice.h> |
b41e9852 | 81 | #include <linux/task_work.h> |
2b188cc1 | 82 | |
c826bd7a DD |
83 | #define CREATE_TRACE_POINTS |
84 | #include <trace/events/io_uring.h> | |
85 | ||
2b188cc1 JA |
86 | #include <uapi/linux/io_uring.h> |
87 | ||
88 | #include "internal.h" | |
561fb04a | 89 | #include "io-wq.h" |
2b188cc1 | 90 | |
5277deaa | 91 | #define IORING_MAX_ENTRIES 32768 |
33a107f0 | 92 | #define IORING_MAX_CQ_ENTRIES (2 * IORING_MAX_ENTRIES) |
65e19f54 JA |
93 | |
94 | /* | |
95 | * Shift of 9 is 512 entries, or exactly one page on 64-bit archs | |
96 | */ | |
97 | #define IORING_FILE_TABLE_SHIFT 9 | |
98 | #define IORING_MAX_FILES_TABLE (1U << IORING_FILE_TABLE_SHIFT) | |
99 | #define IORING_FILE_TABLE_MASK (IORING_MAX_FILES_TABLE - 1) | |
100 | #define IORING_MAX_FIXED_FILES (64 * IORING_MAX_FILES_TABLE) | |
2b188cc1 JA |
101 | |
102 | struct io_uring { | |
103 | u32 head ____cacheline_aligned_in_smp; | |
104 | u32 tail ____cacheline_aligned_in_smp; | |
105 | }; | |
106 | ||
1e84b97b | 107 | /* |
75b28aff HV |
108 | * This data is shared with the application through the mmap at offsets |
109 | * IORING_OFF_SQ_RING and IORING_OFF_CQ_RING. | |
1e84b97b SB |
110 | * |
111 | * The offsets to the member fields are published through struct | |
112 | * io_sqring_offsets when calling io_uring_setup. | |
113 | */ | |
75b28aff | 114 | struct io_rings { |
1e84b97b SB |
115 | /* |
116 | * Head and tail offsets into the ring; the offsets need to be | |
117 | * masked to get valid indices. | |
118 | * | |
75b28aff HV |
119 | * The kernel controls head of the sq ring and the tail of the cq ring, |
120 | * and the application controls tail of the sq ring and the head of the | |
121 | * cq ring. | |
1e84b97b | 122 | */ |
75b28aff | 123 | struct io_uring sq, cq; |
1e84b97b | 124 | /* |
75b28aff | 125 | * Bitmasks to apply to head and tail offsets (constant, equals |
1e84b97b SB |
126 | * ring_entries - 1) |
127 | */ | |
75b28aff HV |
128 | u32 sq_ring_mask, cq_ring_mask; |
129 | /* Ring sizes (constant, power of 2) */ | |
130 | u32 sq_ring_entries, cq_ring_entries; | |
1e84b97b SB |
131 | /* |
132 | * Number of invalid entries dropped by the kernel due to | |
133 | * invalid index stored in array | |
134 | * | |
135 | * Written by the kernel, shouldn't be modified by the | |
136 | * application (i.e. get number of "new events" by comparing to | |
137 | * cached value). | |
138 | * | |
139 | * After a new SQ head value was read by the application this | |
140 | * counter includes all submissions that were dropped reaching | |
141 | * the new SQ head (and possibly more). | |
142 | */ | |
75b28aff | 143 | u32 sq_dropped; |
1e84b97b SB |
144 | /* |
145 | * Runtime flags | |
146 | * | |
147 | * Written by the kernel, shouldn't be modified by the | |
148 | * application. | |
149 | * | |
150 | * The application needs a full memory barrier before checking | |
151 | * for IORING_SQ_NEED_WAKEUP after updating the sq tail. | |
152 | */ | |
75b28aff | 153 | u32 sq_flags; |
1e84b97b SB |
154 | /* |
155 | * Number of completion events lost because the queue was full; | |
156 | * this should be avoided by the application by making sure | |
0b4295b5 | 157 | * there are not more requests pending than there is space in |
1e84b97b SB |
158 | * the completion queue. |
159 | * | |
160 | * Written by the kernel, shouldn't be modified by the | |
161 | * application (i.e. get number of "new events" by comparing to | |
162 | * cached value). | |
163 | * | |
164 | * As completion events come in out of order this counter is not | |
165 | * ordered with any other data. | |
166 | */ | |
75b28aff | 167 | u32 cq_overflow; |
1e84b97b SB |
168 | /* |
169 | * Ring buffer of completion events. | |
170 | * | |
171 | * The kernel writes completion events fresh every time they are | |
172 | * produced, so the application is allowed to modify pending | |
173 | * entries. | |
174 | */ | |
75b28aff | 175 | struct io_uring_cqe cqes[] ____cacheline_aligned_in_smp; |
2b188cc1 JA |
176 | }; |
177 | ||
edafccee JA |
178 | struct io_mapped_ubuf { |
179 | u64 ubuf; | |
180 | size_t len; | |
181 | struct bio_vec *bvec; | |
182 | unsigned int nr_bvecs; | |
183 | }; | |
184 | ||
65e19f54 JA |
185 | struct fixed_file_table { |
186 | struct file **files; | |
31b51510 JA |
187 | }; |
188 | ||
05589553 XW |
189 | struct fixed_file_ref_node { |
190 | struct percpu_ref refs; | |
191 | struct list_head node; | |
192 | struct list_head file_list; | |
193 | struct fixed_file_data *file_data; | |
194 | struct work_struct work; | |
195 | }; | |
196 | ||
05f3fb3c JA |
197 | struct fixed_file_data { |
198 | struct fixed_file_table *table; | |
199 | struct io_ring_ctx *ctx; | |
200 | ||
05589553 | 201 | struct percpu_ref *cur_refs; |
05f3fb3c | 202 | struct percpu_ref refs; |
05f3fb3c | 203 | struct completion done; |
05589553 XW |
204 | struct list_head ref_list; |
205 | spinlock_t lock; | |
05f3fb3c JA |
206 | }; |
207 | ||
5a2e745d JA |
208 | struct io_buffer { |
209 | struct list_head list; | |
210 | __u64 addr; | |
211 | __s32 len; | |
212 | __u16 bid; | |
213 | }; | |
214 | ||
2b188cc1 JA |
215 | struct io_ring_ctx { |
216 | struct { | |
217 | struct percpu_ref refs; | |
218 | } ____cacheline_aligned_in_smp; | |
219 | ||
220 | struct { | |
221 | unsigned int flags; | |
e1d85334 RD |
222 | unsigned int compat: 1; |
223 | unsigned int account_mem: 1; | |
224 | unsigned int cq_overflow_flushed: 1; | |
225 | unsigned int drain_next: 1; | |
226 | unsigned int eventfd_async: 1; | |
2b188cc1 | 227 | |
75b28aff HV |
228 | /* |
229 | * Ring buffer of indices into array of io_uring_sqe, which is | |
230 | * mmapped by the application using the IORING_OFF_SQES offset. | |
231 | * | |
232 | * This indirection could e.g. be used to assign fixed | |
233 | * io_uring_sqe entries to operations and only submit them to | |
234 | * the queue when needed. | |
235 | * | |
236 | * The kernel modifies neither the indices array nor the entries | |
237 | * array. | |
238 | */ | |
239 | u32 *sq_array; | |
2b188cc1 JA |
240 | unsigned cached_sq_head; |
241 | unsigned sq_entries; | |
242 | unsigned sq_mask; | |
6c271ce2 | 243 | unsigned sq_thread_idle; |
498ccd9e | 244 | unsigned cached_sq_dropped; |
206aefde | 245 | atomic_t cached_cq_overflow; |
ad3eb2c8 | 246 | unsigned long sq_check_overflow; |
de0617e4 JA |
247 | |
248 | struct list_head defer_list; | |
5262f567 | 249 | struct list_head timeout_list; |
1d7bb1d5 | 250 | struct list_head cq_overflow_list; |
fcb323cc JA |
251 | |
252 | wait_queue_head_t inflight_wait; | |
ad3eb2c8 | 253 | struct io_uring_sqe *sq_sqes; |
2b188cc1 JA |
254 | } ____cacheline_aligned_in_smp; |
255 | ||
206aefde JA |
256 | struct io_rings *rings; |
257 | ||
2b188cc1 | 258 | /* IO offload */ |
561fb04a | 259 | struct io_wq *io_wq; |
6c271ce2 | 260 | struct task_struct *sqo_thread; /* if using sq thread polling */ |
2b188cc1 | 261 | struct mm_struct *sqo_mm; |
6c271ce2 | 262 | wait_queue_head_t sqo_wait; |
75b28aff | 263 | |
6b06314c JA |
264 | /* |
265 | * If used, fixed file set. Writers must ensure that ->refs is dead, | |
266 | * readers must ensure that ->refs is alive as long as the file* is | |
267 | * used. Only updated through io_uring_register(2). | |
268 | */ | |
05f3fb3c | 269 | struct fixed_file_data *file_data; |
6b06314c | 270 | unsigned nr_user_files; |
b14cca0c PB |
271 | int ring_fd; |
272 | struct file *ring_file; | |
6b06314c | 273 | |
edafccee JA |
274 | /* if used, fixed mapped user buffers */ |
275 | unsigned nr_user_bufs; | |
276 | struct io_mapped_ubuf *user_bufs; | |
277 | ||
2b188cc1 JA |
278 | struct user_struct *user; |
279 | ||
0b8c0ec7 | 280 | const struct cred *creds; |
181e448d | 281 | |
206aefde JA |
282 | /* 0 is for ctx quiesce/reinit/free, 1 is for sqo_thread started */ |
283 | struct completion *completions; | |
284 | ||
0ddf92e8 JA |
285 | /* if all else fails... */ |
286 | struct io_kiocb *fallback_req; | |
287 | ||
206aefde JA |
288 | #if defined(CONFIG_UNIX) |
289 | struct socket *ring_sock; | |
290 | #endif | |
291 | ||
5a2e745d JA |
292 | struct idr io_buffer_idr; |
293 | ||
071698e1 JA |
294 | struct idr personality_idr; |
295 | ||
206aefde JA |
296 | struct { |
297 | unsigned cached_cq_tail; | |
298 | unsigned cq_entries; | |
299 | unsigned cq_mask; | |
300 | atomic_t cq_timeouts; | |
ad3eb2c8 | 301 | unsigned long cq_check_overflow; |
206aefde JA |
302 | struct wait_queue_head cq_wait; |
303 | struct fasync_struct *cq_fasync; | |
304 | struct eventfd_ctx *cq_ev_fd; | |
305 | } ____cacheline_aligned_in_smp; | |
2b188cc1 JA |
306 | |
307 | struct { | |
308 | struct mutex uring_lock; | |
309 | wait_queue_head_t wait; | |
310 | } ____cacheline_aligned_in_smp; | |
311 | ||
312 | struct { | |
313 | spinlock_t completion_lock; | |
e94f141b | 314 | |
def596e9 JA |
315 | /* |
316 | * ->poll_list is protected by the ctx->uring_lock for | |
317 | * io_uring instances that don't use IORING_SETUP_SQPOLL. | |
318 | * For SQPOLL, only the single threaded io_sq_thread() will | |
319 | * manipulate the list, hence no extra locking is needed there. | |
320 | */ | |
321 | struct list_head poll_list; | |
78076bb6 JA |
322 | struct hlist_head *cancel_hash; |
323 | unsigned cancel_hash_bits; | |
e94f141b | 324 | bool poll_multi_file; |
31b51510 | 325 | |
fcb323cc JA |
326 | spinlock_t inflight_lock; |
327 | struct list_head inflight_list; | |
2b188cc1 | 328 | } ____cacheline_aligned_in_smp; |
85faa7b8 JA |
329 | |
330 | struct work_struct exit_work; | |
2b188cc1 JA |
331 | }; |
332 | ||
09bb8394 JA |
333 | /* |
334 | * First field must be the file pointer in all the | |
335 | * iocb unions! See also 'struct kiocb' in <linux/fs.h> | |
336 | */ | |
221c5eb2 JA |
337 | struct io_poll_iocb { |
338 | struct file *file; | |
0969e783 JA |
339 | union { |
340 | struct wait_queue_head *head; | |
341 | u64 addr; | |
342 | }; | |
221c5eb2 | 343 | __poll_t events; |
8c838788 | 344 | bool done; |
221c5eb2 | 345 | bool canceled; |
392edb45 | 346 | struct wait_queue_entry wait; |
221c5eb2 JA |
347 | }; |
348 | ||
b5dba59e JA |
349 | struct io_close { |
350 | struct file *file; | |
351 | struct file *put_file; | |
352 | int fd; | |
353 | }; | |
354 | ||
ad8a48ac JA |
355 | struct io_timeout_data { |
356 | struct io_kiocb *req; | |
357 | struct hrtimer timer; | |
358 | struct timespec64 ts; | |
359 | enum hrtimer_mode mode; | |
360 | }; | |
361 | ||
8ed8d3c3 JA |
362 | struct io_accept { |
363 | struct file *file; | |
364 | struct sockaddr __user *addr; | |
365 | int __user *addr_len; | |
366 | int flags; | |
09952e3e | 367 | unsigned long nofile; |
8ed8d3c3 JA |
368 | }; |
369 | ||
370 | struct io_sync { | |
371 | struct file *file; | |
372 | loff_t len; | |
373 | loff_t off; | |
374 | int flags; | |
d63d1b5e | 375 | int mode; |
8ed8d3c3 JA |
376 | }; |
377 | ||
fbf23849 JA |
378 | struct io_cancel { |
379 | struct file *file; | |
380 | u64 addr; | |
381 | }; | |
382 | ||
b29472ee JA |
383 | struct io_timeout { |
384 | struct file *file; | |
385 | u64 addr; | |
386 | int flags; | |
b55ce732 | 387 | u32 count; |
b29472ee JA |
388 | }; |
389 | ||
9adbd45d JA |
390 | struct io_rw { |
391 | /* NOTE: kiocb has the file as the first member, so don't do it here */ | |
392 | struct kiocb kiocb; | |
393 | u64 addr; | |
394 | u64 len; | |
395 | }; | |
396 | ||
3fbb51c1 JA |
397 | struct io_connect { |
398 | struct file *file; | |
399 | struct sockaddr __user *addr; | |
400 | int addr_len; | |
401 | }; | |
402 | ||
e47293fd JA |
403 | struct io_sr_msg { |
404 | struct file *file; | |
fddaface JA |
405 | union { |
406 | struct user_msghdr __user *msg; | |
407 | void __user *buf; | |
408 | }; | |
e47293fd | 409 | int msg_flags; |
bcda7baa | 410 | int bgid; |
fddaface | 411 | size_t len; |
bcda7baa | 412 | struct io_buffer *kbuf; |
e47293fd JA |
413 | }; |
414 | ||
15b71abe JA |
415 | struct io_open { |
416 | struct file *file; | |
417 | int dfd; | |
eddc7ef5 | 418 | union { |
eddc7ef5 JA |
419 | unsigned mask; |
420 | }; | |
15b71abe | 421 | struct filename *filename; |
eddc7ef5 | 422 | struct statx __user *buffer; |
c12cedf2 | 423 | struct open_how how; |
4022e7af | 424 | unsigned long nofile; |
15b71abe JA |
425 | }; |
426 | ||
05f3fb3c JA |
427 | struct io_files_update { |
428 | struct file *file; | |
429 | u64 arg; | |
430 | u32 nr_args; | |
431 | u32 offset; | |
432 | }; | |
433 | ||
4840e418 JA |
434 | struct io_fadvise { |
435 | struct file *file; | |
436 | u64 offset; | |
437 | u32 len; | |
438 | u32 advice; | |
439 | }; | |
440 | ||
c1ca757b JA |
441 | struct io_madvise { |
442 | struct file *file; | |
443 | u64 addr; | |
444 | u32 len; | |
445 | u32 advice; | |
446 | }; | |
447 | ||
3e4827b0 JA |
448 | struct io_epoll { |
449 | struct file *file; | |
450 | int epfd; | |
451 | int op; | |
452 | int fd; | |
453 | struct epoll_event event; | |
e47293fd JA |
454 | }; |
455 | ||
7d67af2c PB |
456 | struct io_splice { |
457 | struct file *file_out; | |
458 | struct file *file_in; | |
459 | loff_t off_out; | |
460 | loff_t off_in; | |
461 | u64 len; | |
462 | unsigned int flags; | |
463 | }; | |
464 | ||
ddf0322d JA |
465 | struct io_provide_buf { |
466 | struct file *file; | |
467 | __u64 addr; | |
468 | __s32 len; | |
469 | __u32 bgid; | |
470 | __u16 nbufs; | |
471 | __u16 bid; | |
472 | }; | |
473 | ||
f499a021 JA |
474 | struct io_async_connect { |
475 | struct sockaddr_storage address; | |
476 | }; | |
477 | ||
03b1230c JA |
478 | struct io_async_msghdr { |
479 | struct iovec fast_iov[UIO_FASTIOV]; | |
480 | struct iovec *iov; | |
481 | struct sockaddr __user *uaddr; | |
482 | struct msghdr msg; | |
b537916c | 483 | struct sockaddr_storage addr; |
03b1230c JA |
484 | }; |
485 | ||
f67676d1 JA |
486 | struct io_async_rw { |
487 | struct iovec fast_iov[UIO_FASTIOV]; | |
488 | struct iovec *iov; | |
489 | ssize_t nr_segs; | |
490 | ssize_t size; | |
491 | }; | |
492 | ||
1a6b74fc | 493 | struct io_async_ctx { |
f67676d1 JA |
494 | union { |
495 | struct io_async_rw rw; | |
03b1230c | 496 | struct io_async_msghdr msg; |
f499a021 | 497 | struct io_async_connect connect; |
2d28390a | 498 | struct io_timeout_data timeout; |
f67676d1 | 499 | }; |
1a6b74fc JA |
500 | }; |
501 | ||
6b47ee6e PB |
502 | enum { |
503 | REQ_F_FIXED_FILE_BIT = IOSQE_FIXED_FILE_BIT, | |
504 | REQ_F_IO_DRAIN_BIT = IOSQE_IO_DRAIN_BIT, | |
505 | REQ_F_LINK_BIT = IOSQE_IO_LINK_BIT, | |
506 | REQ_F_HARDLINK_BIT = IOSQE_IO_HARDLINK_BIT, | |
507 | REQ_F_FORCE_ASYNC_BIT = IOSQE_ASYNC_BIT, | |
bcda7baa | 508 | REQ_F_BUFFER_SELECT_BIT = IOSQE_BUFFER_SELECT_BIT, |
6b47ee6e | 509 | |
dea3b49c | 510 | REQ_F_LINK_HEAD_BIT, |
6b47ee6e PB |
511 | REQ_F_LINK_NEXT_BIT, |
512 | REQ_F_FAIL_LINK_BIT, | |
513 | REQ_F_INFLIGHT_BIT, | |
514 | REQ_F_CUR_POS_BIT, | |
515 | REQ_F_NOWAIT_BIT, | |
516 | REQ_F_IOPOLL_COMPLETED_BIT, | |
517 | REQ_F_LINK_TIMEOUT_BIT, | |
518 | REQ_F_TIMEOUT_BIT, | |
519 | REQ_F_ISREG_BIT, | |
520 | REQ_F_MUST_PUNT_BIT, | |
521 | REQ_F_TIMEOUT_NOSEQ_BIT, | |
522 | REQ_F_COMP_LOCKED_BIT, | |
99bc4c38 | 523 | REQ_F_NEED_CLEANUP_BIT, |
2ca10259 | 524 | REQ_F_OVERFLOW_BIT, |
d7718a9d | 525 | REQ_F_POLLED_BIT, |
bcda7baa | 526 | REQ_F_BUFFER_SELECTED_BIT, |
5b0bbee4 | 527 | REQ_F_NO_FILE_TABLE_BIT, |
84557871 JA |
528 | |
529 | /* not a real bit, just to check we're not overflowing the space */ | |
530 | __REQ_F_LAST_BIT, | |
6b47ee6e PB |
531 | }; |
532 | ||
533 | enum { | |
534 | /* ctx owns file */ | |
535 | REQ_F_FIXED_FILE = BIT(REQ_F_FIXED_FILE_BIT), | |
536 | /* drain existing IO first */ | |
537 | REQ_F_IO_DRAIN = BIT(REQ_F_IO_DRAIN_BIT), | |
538 | /* linked sqes */ | |
539 | REQ_F_LINK = BIT(REQ_F_LINK_BIT), | |
540 | /* doesn't sever on completion < 0 */ | |
541 | REQ_F_HARDLINK = BIT(REQ_F_HARDLINK_BIT), | |
542 | /* IOSQE_ASYNC */ | |
543 | REQ_F_FORCE_ASYNC = BIT(REQ_F_FORCE_ASYNC_BIT), | |
bcda7baa JA |
544 | /* IOSQE_BUFFER_SELECT */ |
545 | REQ_F_BUFFER_SELECT = BIT(REQ_F_BUFFER_SELECT_BIT), | |
6b47ee6e | 546 | |
dea3b49c PB |
547 | /* head of a link */ |
548 | REQ_F_LINK_HEAD = BIT(REQ_F_LINK_HEAD_BIT), | |
6b47ee6e PB |
549 | /* already grabbed next link */ |
550 | REQ_F_LINK_NEXT = BIT(REQ_F_LINK_NEXT_BIT), | |
551 | /* fail rest of links */ | |
552 | REQ_F_FAIL_LINK = BIT(REQ_F_FAIL_LINK_BIT), | |
553 | /* on inflight list */ | |
554 | REQ_F_INFLIGHT = BIT(REQ_F_INFLIGHT_BIT), | |
555 | /* read/write uses file position */ | |
556 | REQ_F_CUR_POS = BIT(REQ_F_CUR_POS_BIT), | |
557 | /* must not punt to workers */ | |
558 | REQ_F_NOWAIT = BIT(REQ_F_NOWAIT_BIT), | |
559 | /* polled IO has completed */ | |
560 | REQ_F_IOPOLL_COMPLETED = BIT(REQ_F_IOPOLL_COMPLETED_BIT), | |
561 | /* has linked timeout */ | |
562 | REQ_F_LINK_TIMEOUT = BIT(REQ_F_LINK_TIMEOUT_BIT), | |
563 | /* timeout request */ | |
564 | REQ_F_TIMEOUT = BIT(REQ_F_TIMEOUT_BIT), | |
565 | /* regular file */ | |
566 | REQ_F_ISREG = BIT(REQ_F_ISREG_BIT), | |
567 | /* must be punted even for NONBLOCK */ | |
568 | REQ_F_MUST_PUNT = BIT(REQ_F_MUST_PUNT_BIT), | |
569 | /* no timeout sequence */ | |
570 | REQ_F_TIMEOUT_NOSEQ = BIT(REQ_F_TIMEOUT_NOSEQ_BIT), | |
571 | /* completion under lock */ | |
572 | REQ_F_COMP_LOCKED = BIT(REQ_F_COMP_LOCKED_BIT), | |
99bc4c38 PB |
573 | /* needs cleanup */ |
574 | REQ_F_NEED_CLEANUP = BIT(REQ_F_NEED_CLEANUP_BIT), | |
2ca10259 JA |
575 | /* in overflow list */ |
576 | REQ_F_OVERFLOW = BIT(REQ_F_OVERFLOW_BIT), | |
d7718a9d JA |
577 | /* already went through poll handler */ |
578 | REQ_F_POLLED = BIT(REQ_F_POLLED_BIT), | |
bcda7baa JA |
579 | /* buffer already selected */ |
580 | REQ_F_BUFFER_SELECTED = BIT(REQ_F_BUFFER_SELECTED_BIT), | |
5b0bbee4 JA |
581 | /* doesn't need file table for this request */ |
582 | REQ_F_NO_FILE_TABLE = BIT(REQ_F_NO_FILE_TABLE_BIT), | |
d7718a9d JA |
583 | }; |
584 | ||
585 | struct async_poll { | |
586 | struct io_poll_iocb poll; | |
587 | struct io_wq_work work; | |
6b47ee6e PB |
588 | }; |
589 | ||
09bb8394 JA |
590 | /* |
591 | * NOTE! Each of the iocb union members has the file pointer | |
592 | * as the first entry in their struct definition. So you can | |
593 | * access the file pointer through any of the sub-structs, | |
594 | * or directly as just 'ki_filp' in this struct. | |
595 | */ | |
2b188cc1 | 596 | struct io_kiocb { |
221c5eb2 | 597 | union { |
09bb8394 | 598 | struct file *file; |
9adbd45d | 599 | struct io_rw rw; |
221c5eb2 | 600 | struct io_poll_iocb poll; |
8ed8d3c3 JA |
601 | struct io_accept accept; |
602 | struct io_sync sync; | |
fbf23849 | 603 | struct io_cancel cancel; |
b29472ee | 604 | struct io_timeout timeout; |
3fbb51c1 | 605 | struct io_connect connect; |
e47293fd | 606 | struct io_sr_msg sr_msg; |
15b71abe | 607 | struct io_open open; |
b5dba59e | 608 | struct io_close close; |
05f3fb3c | 609 | struct io_files_update files_update; |
4840e418 | 610 | struct io_fadvise fadvise; |
c1ca757b | 611 | struct io_madvise madvise; |
3e4827b0 | 612 | struct io_epoll epoll; |
7d67af2c | 613 | struct io_splice splice; |
ddf0322d | 614 | struct io_provide_buf pbuf; |
221c5eb2 | 615 | }; |
2b188cc1 | 616 | |
1a6b74fc | 617 | struct io_async_ctx *io; |
c398ecb3 | 618 | int cflags; |
cf6fd4bd | 619 | bool needs_fixed_file; |
d625c6ee | 620 | u8 opcode; |
2b188cc1 JA |
621 | |
622 | struct io_ring_ctx *ctx; | |
d7718a9d | 623 | struct list_head list; |
2b188cc1 | 624 | unsigned int flags; |
c16361c1 | 625 | refcount_t refs; |
3537b6a7 JA |
626 | struct task_struct *task; |
627 | unsigned long fsize; | |
2b188cc1 | 628 | u64 user_data; |
9e645e11 | 629 | u32 result; |
de0617e4 | 630 | u32 sequence; |
2b188cc1 | 631 | |
d7718a9d JA |
632 | struct list_head link_list; |
633 | ||
fcb323cc JA |
634 | struct list_head inflight_entry; |
635 | ||
05589553 XW |
636 | struct percpu_ref *fixed_file_refs; |
637 | ||
b41e9852 JA |
638 | union { |
639 | /* | |
640 | * Only commands that never go async can use the below fields, | |
d7718a9d JA |
641 | * obviously. Right now only IORING_OP_POLL_ADD uses them, and |
642 | * async armed poll handlers for regular commands. The latter | |
643 | * restore the work, if needed. | |
b41e9852 JA |
644 | */ |
645 | struct { | |
b41e9852 | 646 | struct callback_head task_work; |
d7718a9d JA |
647 | struct hlist_node hash_node; |
648 | struct async_poll *apoll; | |
b41e9852 JA |
649 | }; |
650 | struct io_wq_work work; | |
651 | }; | |
2b188cc1 JA |
652 | }; |
653 | ||
654 | #define IO_PLUG_THRESHOLD 2 | |
def596e9 | 655 | #define IO_IOPOLL_BATCH 8 |
2b188cc1 | 656 | |
9a56a232 JA |
657 | struct io_submit_state { |
658 | struct blk_plug plug; | |
659 | ||
2579f913 JA |
660 | /* |
661 | * io_kiocb alloc cache | |
662 | */ | |
663 | void *reqs[IO_IOPOLL_BATCH]; | |
6c8a3134 | 664 | unsigned int free_reqs; |
2579f913 | 665 | |
9a56a232 JA |
666 | /* |
667 | * File reference cache | |
668 | */ | |
669 | struct file *file; | |
670 | unsigned int fd; | |
671 | unsigned int has_refs; | |
672 | unsigned int used_refs; | |
673 | unsigned int ios_left; | |
674 | }; | |
675 | ||
d3656344 JA |
676 | struct io_op_def { |
677 | /* needs req->io allocated for deferral/async */ | |
678 | unsigned async_ctx : 1; | |
679 | /* needs current->mm setup, does mm access */ | |
680 | unsigned needs_mm : 1; | |
681 | /* needs req->file assigned */ | |
682 | unsigned needs_file : 1; | |
d3656344 JA |
683 | /* hash wq insertion if file is a regular file */ |
684 | unsigned hash_reg_file : 1; | |
685 | /* unbound wq insertion if file is a non-regular file */ | |
686 | unsigned unbound_nonreg_file : 1; | |
66f4af93 JA |
687 | /* opcode is not supported by this kernel */ |
688 | unsigned not_supported : 1; | |
f86cd20c JA |
689 | /* needs file table */ |
690 | unsigned file_table : 1; | |
ff002b30 JA |
691 | /* needs ->fs */ |
692 | unsigned needs_fs : 1; | |
8a72758c JA |
693 | /* set if opcode supports polled "wait" */ |
694 | unsigned pollin : 1; | |
695 | unsigned pollout : 1; | |
bcda7baa JA |
696 | /* op supports buffer selection */ |
697 | unsigned buffer_select : 1; | |
d3656344 JA |
698 | }; |
699 | ||
700 | static const struct io_op_def io_op_defs[] = { | |
0463b6c5 PB |
701 | [IORING_OP_NOP] = {}, |
702 | [IORING_OP_READV] = { | |
d3656344 JA |
703 | .async_ctx = 1, |
704 | .needs_mm = 1, | |
705 | .needs_file = 1, | |
706 | .unbound_nonreg_file = 1, | |
8a72758c | 707 | .pollin = 1, |
4d954c25 | 708 | .buffer_select = 1, |
d3656344 | 709 | }, |
0463b6c5 | 710 | [IORING_OP_WRITEV] = { |
d3656344 JA |
711 | .async_ctx = 1, |
712 | .needs_mm = 1, | |
713 | .needs_file = 1, | |
714 | .hash_reg_file = 1, | |
715 | .unbound_nonreg_file = 1, | |
8a72758c | 716 | .pollout = 1, |
d3656344 | 717 | }, |
0463b6c5 | 718 | [IORING_OP_FSYNC] = { |
d3656344 JA |
719 | .needs_file = 1, |
720 | }, | |
0463b6c5 | 721 | [IORING_OP_READ_FIXED] = { |
d3656344 JA |
722 | .needs_file = 1, |
723 | .unbound_nonreg_file = 1, | |
8a72758c | 724 | .pollin = 1, |
d3656344 | 725 | }, |
0463b6c5 | 726 | [IORING_OP_WRITE_FIXED] = { |
d3656344 JA |
727 | .needs_file = 1, |
728 | .hash_reg_file = 1, | |
729 | .unbound_nonreg_file = 1, | |
8a72758c | 730 | .pollout = 1, |
d3656344 | 731 | }, |
0463b6c5 | 732 | [IORING_OP_POLL_ADD] = { |
d3656344 JA |
733 | .needs_file = 1, |
734 | .unbound_nonreg_file = 1, | |
735 | }, | |
0463b6c5 PB |
736 | [IORING_OP_POLL_REMOVE] = {}, |
737 | [IORING_OP_SYNC_FILE_RANGE] = { | |
d3656344 JA |
738 | .needs_file = 1, |
739 | }, | |
0463b6c5 | 740 | [IORING_OP_SENDMSG] = { |
d3656344 JA |
741 | .async_ctx = 1, |
742 | .needs_mm = 1, | |
743 | .needs_file = 1, | |
744 | .unbound_nonreg_file = 1, | |
ff002b30 | 745 | .needs_fs = 1, |
8a72758c | 746 | .pollout = 1, |
d3656344 | 747 | }, |
0463b6c5 | 748 | [IORING_OP_RECVMSG] = { |
d3656344 JA |
749 | .async_ctx = 1, |
750 | .needs_mm = 1, | |
751 | .needs_file = 1, | |
752 | .unbound_nonreg_file = 1, | |
ff002b30 | 753 | .needs_fs = 1, |
8a72758c | 754 | .pollin = 1, |
52de1fe1 | 755 | .buffer_select = 1, |
d3656344 | 756 | }, |
0463b6c5 | 757 | [IORING_OP_TIMEOUT] = { |
d3656344 JA |
758 | .async_ctx = 1, |
759 | .needs_mm = 1, | |
760 | }, | |
0463b6c5 PB |
761 | [IORING_OP_TIMEOUT_REMOVE] = {}, |
762 | [IORING_OP_ACCEPT] = { | |
d3656344 JA |
763 | .needs_mm = 1, |
764 | .needs_file = 1, | |
765 | .unbound_nonreg_file = 1, | |
f86cd20c | 766 | .file_table = 1, |
8a72758c | 767 | .pollin = 1, |
d3656344 | 768 | }, |
0463b6c5 PB |
769 | [IORING_OP_ASYNC_CANCEL] = {}, |
770 | [IORING_OP_LINK_TIMEOUT] = { | |
d3656344 JA |
771 | .async_ctx = 1, |
772 | .needs_mm = 1, | |
773 | }, | |
0463b6c5 | 774 | [IORING_OP_CONNECT] = { |
d3656344 JA |
775 | .async_ctx = 1, |
776 | .needs_mm = 1, | |
777 | .needs_file = 1, | |
778 | .unbound_nonreg_file = 1, | |
8a72758c | 779 | .pollout = 1, |
d3656344 | 780 | }, |
0463b6c5 | 781 | [IORING_OP_FALLOCATE] = { |
d3656344 JA |
782 | .needs_file = 1, |
783 | }, | |
0463b6c5 | 784 | [IORING_OP_OPENAT] = { |
f86cd20c | 785 | .file_table = 1, |
ff002b30 | 786 | .needs_fs = 1, |
d3656344 | 787 | }, |
0463b6c5 | 788 | [IORING_OP_CLOSE] = { |
f86cd20c | 789 | .file_table = 1, |
d3656344 | 790 | }, |
0463b6c5 | 791 | [IORING_OP_FILES_UPDATE] = { |
d3656344 | 792 | .needs_mm = 1, |
f86cd20c | 793 | .file_table = 1, |
d3656344 | 794 | }, |
0463b6c5 | 795 | [IORING_OP_STATX] = { |
d3656344 | 796 | .needs_mm = 1, |
ff002b30 | 797 | .needs_fs = 1, |
5b0bbee4 | 798 | .file_table = 1, |
d3656344 | 799 | }, |
0463b6c5 | 800 | [IORING_OP_READ] = { |
3a6820f2 JA |
801 | .needs_mm = 1, |
802 | .needs_file = 1, | |
803 | .unbound_nonreg_file = 1, | |
8a72758c | 804 | .pollin = 1, |
bcda7baa | 805 | .buffer_select = 1, |
3a6820f2 | 806 | }, |
0463b6c5 | 807 | [IORING_OP_WRITE] = { |
3a6820f2 JA |
808 | .needs_mm = 1, |
809 | .needs_file = 1, | |
810 | .unbound_nonreg_file = 1, | |
8a72758c | 811 | .pollout = 1, |
3a6820f2 | 812 | }, |
0463b6c5 | 813 | [IORING_OP_FADVISE] = { |
4840e418 JA |
814 | .needs_file = 1, |
815 | }, | |
0463b6c5 | 816 | [IORING_OP_MADVISE] = { |
c1ca757b JA |
817 | .needs_mm = 1, |
818 | }, | |
0463b6c5 | 819 | [IORING_OP_SEND] = { |
fddaface JA |
820 | .needs_mm = 1, |
821 | .needs_file = 1, | |
822 | .unbound_nonreg_file = 1, | |
8a72758c | 823 | .pollout = 1, |
fddaface | 824 | }, |
0463b6c5 | 825 | [IORING_OP_RECV] = { |
fddaface JA |
826 | .needs_mm = 1, |
827 | .needs_file = 1, | |
828 | .unbound_nonreg_file = 1, | |
8a72758c | 829 | .pollin = 1, |
bcda7baa | 830 | .buffer_select = 1, |
fddaface | 831 | }, |
0463b6c5 | 832 | [IORING_OP_OPENAT2] = { |
f86cd20c | 833 | .file_table = 1, |
ff002b30 | 834 | .needs_fs = 1, |
cebdb986 | 835 | }, |
3e4827b0 JA |
836 | [IORING_OP_EPOLL_CTL] = { |
837 | .unbound_nonreg_file = 1, | |
838 | .file_table = 1, | |
839 | }, | |
7d67af2c PB |
840 | [IORING_OP_SPLICE] = { |
841 | .needs_file = 1, | |
842 | .hash_reg_file = 1, | |
843 | .unbound_nonreg_file = 1, | |
ddf0322d JA |
844 | }, |
845 | [IORING_OP_PROVIDE_BUFFERS] = {}, | |
067524e9 | 846 | [IORING_OP_REMOVE_BUFFERS] = {}, |
d3656344 JA |
847 | }; |
848 | ||
561fb04a | 849 | static void io_wq_submit_work(struct io_wq_work **workptr); |
78e19bbe | 850 | static void io_cqring_fill_event(struct io_kiocb *req, long res); |
ec9c02ad | 851 | static void io_put_req(struct io_kiocb *req); |
978db57e | 852 | static void __io_double_put_req(struct io_kiocb *req); |
94ae5e77 JA |
853 | static struct io_kiocb *io_prep_linked_timeout(struct io_kiocb *req); |
854 | static void io_queue_linked_timeout(struct io_kiocb *req); | |
05f3fb3c JA |
855 | static int __io_sqe_files_update(struct io_ring_ctx *ctx, |
856 | struct io_uring_files_update *ip, | |
857 | unsigned nr_args); | |
f86cd20c | 858 | static int io_grab_files(struct io_kiocb *req); |
99bc4c38 | 859 | static void io_cleanup_req(struct io_kiocb *req); |
b41e9852 JA |
860 | static int io_file_get(struct io_submit_state *state, struct io_kiocb *req, |
861 | int fd, struct file **out_file, bool fixed); | |
862 | static void __io_queue_sqe(struct io_kiocb *req, | |
863 | const struct io_uring_sqe *sqe); | |
de0617e4 | 864 | |
2b188cc1 JA |
865 | static struct kmem_cache *req_cachep; |
866 | ||
867 | static const struct file_operations io_uring_fops; | |
868 | ||
869 | struct sock *io_uring_get_socket(struct file *file) | |
870 | { | |
871 | #if defined(CONFIG_UNIX) | |
872 | if (file->f_op == &io_uring_fops) { | |
873 | struct io_ring_ctx *ctx = file->private_data; | |
874 | ||
875 | return ctx->ring_sock->sk; | |
876 | } | |
877 | #endif | |
878 | return NULL; | |
879 | } | |
880 | EXPORT_SYMBOL(io_uring_get_socket); | |
881 | ||
882 | static void io_ring_ctx_ref_free(struct percpu_ref *ref) | |
883 | { | |
884 | struct io_ring_ctx *ctx = container_of(ref, struct io_ring_ctx, refs); | |
885 | ||
206aefde | 886 | complete(&ctx->completions[0]); |
2b188cc1 JA |
887 | } |
888 | ||
889 | static struct io_ring_ctx *io_ring_ctx_alloc(struct io_uring_params *p) | |
890 | { | |
891 | struct io_ring_ctx *ctx; | |
78076bb6 | 892 | int hash_bits; |
2b188cc1 JA |
893 | |
894 | ctx = kzalloc(sizeof(*ctx), GFP_KERNEL); | |
895 | if (!ctx) | |
896 | return NULL; | |
897 | ||
0ddf92e8 JA |
898 | ctx->fallback_req = kmem_cache_alloc(req_cachep, GFP_KERNEL); |
899 | if (!ctx->fallback_req) | |
900 | goto err; | |
901 | ||
206aefde JA |
902 | ctx->completions = kmalloc(2 * sizeof(struct completion), GFP_KERNEL); |
903 | if (!ctx->completions) | |
904 | goto err; | |
905 | ||
78076bb6 JA |
906 | /* |
907 | * Use 5 bits less than the max cq entries, that should give us around | |
908 | * 32 entries per hash list if totally full and uniformly spread. | |
909 | */ | |
910 | hash_bits = ilog2(p->cq_entries); | |
911 | hash_bits -= 5; | |
912 | if (hash_bits <= 0) | |
913 | hash_bits = 1; | |
914 | ctx->cancel_hash_bits = hash_bits; | |
915 | ctx->cancel_hash = kmalloc((1U << hash_bits) * sizeof(struct hlist_head), | |
916 | GFP_KERNEL); | |
917 | if (!ctx->cancel_hash) | |
918 | goto err; | |
919 | __hash_init(ctx->cancel_hash, 1U << hash_bits); | |
920 | ||
21482896 | 921 | if (percpu_ref_init(&ctx->refs, io_ring_ctx_ref_free, |
206aefde JA |
922 | PERCPU_REF_ALLOW_REINIT, GFP_KERNEL)) |
923 | goto err; | |
2b188cc1 JA |
924 | |
925 | ctx->flags = p->flags; | |
926 | init_waitqueue_head(&ctx->cq_wait); | |
1d7bb1d5 | 927 | INIT_LIST_HEAD(&ctx->cq_overflow_list); |
206aefde JA |
928 | init_completion(&ctx->completions[0]); |
929 | init_completion(&ctx->completions[1]); | |
5a2e745d | 930 | idr_init(&ctx->io_buffer_idr); |
071698e1 | 931 | idr_init(&ctx->personality_idr); |
2b188cc1 JA |
932 | mutex_init(&ctx->uring_lock); |
933 | init_waitqueue_head(&ctx->wait); | |
934 | spin_lock_init(&ctx->completion_lock); | |
def596e9 | 935 | INIT_LIST_HEAD(&ctx->poll_list); |
de0617e4 | 936 | INIT_LIST_HEAD(&ctx->defer_list); |
5262f567 | 937 | INIT_LIST_HEAD(&ctx->timeout_list); |
fcb323cc JA |
938 | init_waitqueue_head(&ctx->inflight_wait); |
939 | spin_lock_init(&ctx->inflight_lock); | |
940 | INIT_LIST_HEAD(&ctx->inflight_list); | |
2b188cc1 | 941 | return ctx; |
206aefde | 942 | err: |
0ddf92e8 JA |
943 | if (ctx->fallback_req) |
944 | kmem_cache_free(req_cachep, ctx->fallback_req); | |
206aefde | 945 | kfree(ctx->completions); |
78076bb6 | 946 | kfree(ctx->cancel_hash); |
206aefde JA |
947 | kfree(ctx); |
948 | return NULL; | |
2b188cc1 JA |
949 | } |
950 | ||
9d858b21 | 951 | static inline bool __req_need_defer(struct io_kiocb *req) |
7adf4eaf | 952 | { |
a197f664 JL |
953 | struct io_ring_ctx *ctx = req->ctx; |
954 | ||
31af27c7 PB |
955 | return req->sequence != ctx->cached_cq_tail |
956 | + atomic_read(&ctx->cached_cq_overflow); | |
7adf4eaf JA |
957 | } |
958 | ||
9d858b21 | 959 | static inline bool req_need_defer(struct io_kiocb *req) |
de0617e4 | 960 | { |
87987898 | 961 | if (unlikely(req->flags & REQ_F_IO_DRAIN)) |
9d858b21 | 962 | return __req_need_defer(req); |
de0617e4 | 963 | |
9d858b21 | 964 | return false; |
de0617e4 JA |
965 | } |
966 | ||
7adf4eaf | 967 | static struct io_kiocb *io_get_deferred_req(struct io_ring_ctx *ctx) |
de0617e4 JA |
968 | { |
969 | struct io_kiocb *req; | |
970 | ||
7adf4eaf | 971 | req = list_first_entry_or_null(&ctx->defer_list, struct io_kiocb, list); |
9d858b21 | 972 | if (req && !req_need_defer(req)) { |
de0617e4 JA |
973 | list_del_init(&req->list); |
974 | return req; | |
975 | } | |
976 | ||
977 | return NULL; | |
978 | } | |
979 | ||
5262f567 JA |
980 | static struct io_kiocb *io_get_timeout_req(struct io_ring_ctx *ctx) |
981 | { | |
7adf4eaf JA |
982 | struct io_kiocb *req; |
983 | ||
984 | req = list_first_entry_or_null(&ctx->timeout_list, struct io_kiocb, list); | |
93bd25bb JA |
985 | if (req) { |
986 | if (req->flags & REQ_F_TIMEOUT_NOSEQ) | |
987 | return NULL; | |
fb4b3d3f | 988 | if (!__req_need_defer(req)) { |
93bd25bb JA |
989 | list_del_init(&req->list); |
990 | return req; | |
991 | } | |
7adf4eaf JA |
992 | } |
993 | ||
994 | return NULL; | |
5262f567 JA |
995 | } |
996 | ||
de0617e4 | 997 | static void __io_commit_cqring(struct io_ring_ctx *ctx) |
2b188cc1 | 998 | { |
75b28aff | 999 | struct io_rings *rings = ctx->rings; |
2b188cc1 | 1000 | |
07910158 PB |
1001 | /* order cqe stores with ring update */ |
1002 | smp_store_release(&rings->cq.tail, ctx->cached_cq_tail); | |
2b188cc1 | 1003 | |
07910158 PB |
1004 | if (wq_has_sleeper(&ctx->cq_wait)) { |
1005 | wake_up_interruptible(&ctx->cq_wait); | |
1006 | kill_fasync(&ctx->cq_fasync, SIGIO, POLL_IN); | |
2b188cc1 JA |
1007 | } |
1008 | } | |
1009 | ||
cccf0ee8 JA |
1010 | static inline void io_req_work_grab_env(struct io_kiocb *req, |
1011 | const struct io_op_def *def) | |
1012 | { | |
1013 | if (!req->work.mm && def->needs_mm) { | |
1014 | mmgrab(current->mm); | |
1015 | req->work.mm = current->mm; | |
2b188cc1 | 1016 | } |
cccf0ee8 JA |
1017 | if (!req->work.creds) |
1018 | req->work.creds = get_current_cred(); | |
ff002b30 JA |
1019 | if (!req->work.fs && def->needs_fs) { |
1020 | spin_lock(¤t->fs->lock); | |
1021 | if (!current->fs->in_exec) { | |
1022 | req->work.fs = current->fs; | |
1023 | req->work.fs->users++; | |
1024 | } else { | |
1025 | req->work.flags |= IO_WQ_WORK_CANCEL; | |
1026 | } | |
1027 | spin_unlock(¤t->fs->lock); | |
1028 | } | |
6ab23144 JA |
1029 | if (!req->work.task_pid) |
1030 | req->work.task_pid = task_pid_vnr(current); | |
2b188cc1 JA |
1031 | } |
1032 | ||
cccf0ee8 | 1033 | static inline void io_req_work_drop_env(struct io_kiocb *req) |
18d9be1a | 1034 | { |
cccf0ee8 JA |
1035 | if (req->work.mm) { |
1036 | mmdrop(req->work.mm); | |
1037 | req->work.mm = NULL; | |
1038 | } | |
1039 | if (req->work.creds) { | |
1040 | put_cred(req->work.creds); | |
1041 | req->work.creds = NULL; | |
1042 | } | |
ff002b30 JA |
1043 | if (req->work.fs) { |
1044 | struct fs_struct *fs = req->work.fs; | |
1045 | ||
1046 | spin_lock(&req->work.fs->lock); | |
1047 | if (--fs->users) | |
1048 | fs = NULL; | |
1049 | spin_unlock(&req->work.fs->lock); | |
1050 | if (fs) | |
1051 | free_fs_struct(fs); | |
1052 | } | |
561fb04a JA |
1053 | } |
1054 | ||
8766dd51 | 1055 | static inline void io_prep_async_work(struct io_kiocb *req, |
94ae5e77 | 1056 | struct io_kiocb **link) |
18d9be1a | 1057 | { |
d3656344 | 1058 | const struct io_op_def *def = &io_op_defs[req->opcode]; |
54a91f3b | 1059 | |
d3656344 JA |
1060 | if (req->flags & REQ_F_ISREG) { |
1061 | if (def->hash_reg_file) | |
8766dd51 | 1062 | io_wq_hash_work(&req->work, file_inode(req->file)); |
d3656344 JA |
1063 | } else { |
1064 | if (def->unbound_nonreg_file) | |
3529d8c2 | 1065 | req->work.flags |= IO_WQ_WORK_UNBOUND; |
54a91f3b | 1066 | } |
cccf0ee8 JA |
1067 | |
1068 | io_req_work_grab_env(req, def); | |
54a91f3b | 1069 | |
94ae5e77 | 1070 | *link = io_prep_linked_timeout(req); |
561fb04a JA |
1071 | } |
1072 | ||
a197f664 | 1073 | static inline void io_queue_async_work(struct io_kiocb *req) |
561fb04a | 1074 | { |
a197f664 | 1075 | struct io_ring_ctx *ctx = req->ctx; |
94ae5e77 | 1076 | struct io_kiocb *link; |
94ae5e77 | 1077 | |
8766dd51 | 1078 | io_prep_async_work(req, &link); |
561fb04a | 1079 | |
8766dd51 PB |
1080 | trace_io_uring_queue_async_work(ctx, io_wq_is_hashed(&req->work), req, |
1081 | &req->work, req->flags); | |
1082 | io_wq_enqueue(ctx->io_wq, &req->work); | |
94ae5e77 JA |
1083 | |
1084 | if (link) | |
1085 | io_queue_linked_timeout(link); | |
18d9be1a JA |
1086 | } |
1087 | ||
5262f567 JA |
1088 | static void io_kill_timeout(struct io_kiocb *req) |
1089 | { | |
1090 | int ret; | |
1091 | ||
2d28390a | 1092 | ret = hrtimer_try_to_cancel(&req->io->timeout.timer); |
5262f567 JA |
1093 | if (ret != -1) { |
1094 | atomic_inc(&req->ctx->cq_timeouts); | |
842f9612 | 1095 | list_del_init(&req->list); |
f0e20b89 | 1096 | req->flags |= REQ_F_COMP_LOCKED; |
78e19bbe | 1097 | io_cqring_fill_event(req, 0); |
ec9c02ad | 1098 | io_put_req(req); |
5262f567 JA |
1099 | } |
1100 | } | |
1101 | ||
1102 | static void io_kill_timeouts(struct io_ring_ctx *ctx) | |
1103 | { | |
1104 | struct io_kiocb *req, *tmp; | |
1105 | ||
1106 | spin_lock_irq(&ctx->completion_lock); | |
1107 | list_for_each_entry_safe(req, tmp, &ctx->timeout_list, list) | |
1108 | io_kill_timeout(req); | |
1109 | spin_unlock_irq(&ctx->completion_lock); | |
1110 | } | |
1111 | ||
de0617e4 JA |
1112 | static void io_commit_cqring(struct io_ring_ctx *ctx) |
1113 | { | |
1114 | struct io_kiocb *req; | |
1115 | ||
5262f567 JA |
1116 | while ((req = io_get_timeout_req(ctx)) != NULL) |
1117 | io_kill_timeout(req); | |
1118 | ||
de0617e4 JA |
1119 | __io_commit_cqring(ctx); |
1120 | ||
87987898 | 1121 | while ((req = io_get_deferred_req(ctx)) != NULL) |
a197f664 | 1122 | io_queue_async_work(req); |
de0617e4 JA |
1123 | } |
1124 | ||
2b188cc1 JA |
1125 | static struct io_uring_cqe *io_get_cqring(struct io_ring_ctx *ctx) |
1126 | { | |
75b28aff | 1127 | struct io_rings *rings = ctx->rings; |
2b188cc1 JA |
1128 | unsigned tail; |
1129 | ||
1130 | tail = ctx->cached_cq_tail; | |
115e12e5 SB |
1131 | /* |
1132 | * writes to the cq entry need to come after reading head; the | |
1133 | * control dependency is enough as we're using WRITE_ONCE to | |
1134 | * fill the cq entry | |
1135 | */ | |
75b28aff | 1136 | if (tail - READ_ONCE(rings->cq.head) == rings->cq_ring_entries) |
2b188cc1 JA |
1137 | return NULL; |
1138 | ||
1139 | ctx->cached_cq_tail++; | |
75b28aff | 1140 | return &rings->cqes[tail & ctx->cq_mask]; |
2b188cc1 JA |
1141 | } |
1142 | ||
f2842ab5 JA |
1143 | static inline bool io_should_trigger_evfd(struct io_ring_ctx *ctx) |
1144 | { | |
f0b493e6 JA |
1145 | if (!ctx->cq_ev_fd) |
1146 | return false; | |
f2842ab5 JA |
1147 | if (!ctx->eventfd_async) |
1148 | return true; | |
b41e9852 | 1149 | return io_wq_current_is_worker(); |
f2842ab5 JA |
1150 | } |
1151 | ||
b41e9852 | 1152 | static void io_cqring_ev_posted(struct io_ring_ctx *ctx) |
1d7bb1d5 JA |
1153 | { |
1154 | if (waitqueue_active(&ctx->wait)) | |
1155 | wake_up(&ctx->wait); | |
1156 | if (waitqueue_active(&ctx->sqo_wait)) | |
1157 | wake_up(&ctx->sqo_wait); | |
b41e9852 | 1158 | if (io_should_trigger_evfd(ctx)) |
1d7bb1d5 JA |
1159 | eventfd_signal(ctx->cq_ev_fd, 1); |
1160 | } | |
1161 | ||
c4a2ed72 JA |
1162 | /* Returns true if there are no backlogged entries after the flush */ |
1163 | static bool io_cqring_overflow_flush(struct io_ring_ctx *ctx, bool force) | |
1d7bb1d5 JA |
1164 | { |
1165 | struct io_rings *rings = ctx->rings; | |
1166 | struct io_uring_cqe *cqe; | |
1167 | struct io_kiocb *req; | |
1168 | unsigned long flags; | |
1169 | LIST_HEAD(list); | |
1170 | ||
1171 | if (!force) { | |
1172 | if (list_empty_careful(&ctx->cq_overflow_list)) | |
c4a2ed72 | 1173 | return true; |
1d7bb1d5 JA |
1174 | if ((ctx->cached_cq_tail - READ_ONCE(rings->cq.head) == |
1175 | rings->cq_ring_entries)) | |
c4a2ed72 | 1176 | return false; |
1d7bb1d5 JA |
1177 | } |
1178 | ||
1179 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
1180 | ||
1181 | /* if force is set, the ring is going away. always drop after that */ | |
1182 | if (force) | |
69b3e546 | 1183 | ctx->cq_overflow_flushed = 1; |
1d7bb1d5 | 1184 | |
c4a2ed72 | 1185 | cqe = NULL; |
1d7bb1d5 JA |
1186 | while (!list_empty(&ctx->cq_overflow_list)) { |
1187 | cqe = io_get_cqring(ctx); | |
1188 | if (!cqe && !force) | |
1189 | break; | |
1190 | ||
1191 | req = list_first_entry(&ctx->cq_overflow_list, struct io_kiocb, | |
1192 | list); | |
1193 | list_move(&req->list, &list); | |
2ca10259 | 1194 | req->flags &= ~REQ_F_OVERFLOW; |
1d7bb1d5 JA |
1195 | if (cqe) { |
1196 | WRITE_ONCE(cqe->user_data, req->user_data); | |
1197 | WRITE_ONCE(cqe->res, req->result); | |
bcda7baa | 1198 | WRITE_ONCE(cqe->flags, req->cflags); |
1d7bb1d5 JA |
1199 | } else { |
1200 | WRITE_ONCE(ctx->rings->cq_overflow, | |
1201 | atomic_inc_return(&ctx->cached_cq_overflow)); | |
1202 | } | |
1203 | } | |
1204 | ||
1205 | io_commit_cqring(ctx); | |
ad3eb2c8 JA |
1206 | if (cqe) { |
1207 | clear_bit(0, &ctx->sq_check_overflow); | |
1208 | clear_bit(0, &ctx->cq_check_overflow); | |
1209 | } | |
1d7bb1d5 JA |
1210 | spin_unlock_irqrestore(&ctx->completion_lock, flags); |
1211 | io_cqring_ev_posted(ctx); | |
1212 | ||
1213 | while (!list_empty(&list)) { | |
1214 | req = list_first_entry(&list, struct io_kiocb, list); | |
1215 | list_del(&req->list); | |
ec9c02ad | 1216 | io_put_req(req); |
1d7bb1d5 | 1217 | } |
c4a2ed72 JA |
1218 | |
1219 | return cqe != NULL; | |
1d7bb1d5 JA |
1220 | } |
1221 | ||
bcda7baa | 1222 | static void __io_cqring_fill_event(struct io_kiocb *req, long res, long cflags) |
2b188cc1 | 1223 | { |
78e19bbe | 1224 | struct io_ring_ctx *ctx = req->ctx; |
2b188cc1 JA |
1225 | struct io_uring_cqe *cqe; |
1226 | ||
78e19bbe | 1227 | trace_io_uring_complete(ctx, req->user_data, res); |
51c3ff62 | 1228 | |
2b188cc1 JA |
1229 | /* |
1230 | * If we can't get a cq entry, userspace overflowed the | |
1231 | * submission (by quite a lot). Increment the overflow count in | |
1232 | * the ring. | |
1233 | */ | |
1234 | cqe = io_get_cqring(ctx); | |
1d7bb1d5 | 1235 | if (likely(cqe)) { |
78e19bbe | 1236 | WRITE_ONCE(cqe->user_data, req->user_data); |
2b188cc1 | 1237 | WRITE_ONCE(cqe->res, res); |
bcda7baa | 1238 | WRITE_ONCE(cqe->flags, cflags); |
1d7bb1d5 | 1239 | } else if (ctx->cq_overflow_flushed) { |
498ccd9e JA |
1240 | WRITE_ONCE(ctx->rings->cq_overflow, |
1241 | atomic_inc_return(&ctx->cached_cq_overflow)); | |
1d7bb1d5 | 1242 | } else { |
ad3eb2c8 JA |
1243 | if (list_empty(&ctx->cq_overflow_list)) { |
1244 | set_bit(0, &ctx->sq_check_overflow); | |
1245 | set_bit(0, &ctx->cq_check_overflow); | |
1246 | } | |
2ca10259 | 1247 | req->flags |= REQ_F_OVERFLOW; |
1d7bb1d5 JA |
1248 | refcount_inc(&req->refs); |
1249 | req->result = res; | |
bcda7baa | 1250 | req->cflags = cflags; |
1d7bb1d5 | 1251 | list_add_tail(&req->list, &ctx->cq_overflow_list); |
2b188cc1 JA |
1252 | } |
1253 | } | |
1254 | ||
bcda7baa JA |
1255 | static void io_cqring_fill_event(struct io_kiocb *req, long res) |
1256 | { | |
1257 | __io_cqring_fill_event(req, res, 0); | |
1258 | } | |
1259 | ||
1260 | static void __io_cqring_add_event(struct io_kiocb *req, long res, long cflags) | |
2b188cc1 | 1261 | { |
78e19bbe | 1262 | struct io_ring_ctx *ctx = req->ctx; |
2b188cc1 JA |
1263 | unsigned long flags; |
1264 | ||
1265 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
bcda7baa | 1266 | __io_cqring_fill_event(req, res, cflags); |
2b188cc1 JA |
1267 | io_commit_cqring(ctx); |
1268 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
1269 | ||
8c838788 | 1270 | io_cqring_ev_posted(ctx); |
2b188cc1 JA |
1271 | } |
1272 | ||
bcda7baa JA |
1273 | static void io_cqring_add_event(struct io_kiocb *req, long res) |
1274 | { | |
1275 | __io_cqring_add_event(req, res, 0); | |
1276 | } | |
1277 | ||
0ddf92e8 JA |
1278 | static inline bool io_is_fallback_req(struct io_kiocb *req) |
1279 | { | |
1280 | return req == (struct io_kiocb *) | |
1281 | ((unsigned long) req->ctx->fallback_req & ~1UL); | |
1282 | } | |
1283 | ||
1284 | static struct io_kiocb *io_get_fallback_req(struct io_ring_ctx *ctx) | |
1285 | { | |
1286 | struct io_kiocb *req; | |
1287 | ||
1288 | req = ctx->fallback_req; | |
dd461af6 | 1289 | if (!test_and_set_bit_lock(0, (unsigned long *) &ctx->fallback_req)) |
0ddf92e8 JA |
1290 | return req; |
1291 | ||
1292 | return NULL; | |
1293 | } | |
1294 | ||
0553b8bd PB |
1295 | static struct io_kiocb *io_alloc_req(struct io_ring_ctx *ctx, |
1296 | struct io_submit_state *state) | |
2b188cc1 | 1297 | { |
fd6fab2c | 1298 | gfp_t gfp = GFP_KERNEL | __GFP_NOWARN; |
2b188cc1 JA |
1299 | struct io_kiocb *req; |
1300 | ||
2579f913 | 1301 | if (!state) { |
fd6fab2c | 1302 | req = kmem_cache_alloc(req_cachep, gfp); |
2579f913 | 1303 | if (unlikely(!req)) |
0ddf92e8 | 1304 | goto fallback; |
2579f913 JA |
1305 | } else if (!state->free_reqs) { |
1306 | size_t sz; | |
1307 | int ret; | |
1308 | ||
1309 | sz = min_t(size_t, state->ios_left, ARRAY_SIZE(state->reqs)); | |
fd6fab2c JA |
1310 | ret = kmem_cache_alloc_bulk(req_cachep, gfp, sz, state->reqs); |
1311 | ||
1312 | /* | |
1313 | * Bulk alloc is all-or-nothing. If we fail to get a batch, | |
1314 | * retry single alloc to be on the safe side. | |
1315 | */ | |
1316 | if (unlikely(ret <= 0)) { | |
1317 | state->reqs[0] = kmem_cache_alloc(req_cachep, gfp); | |
1318 | if (!state->reqs[0]) | |
0ddf92e8 | 1319 | goto fallback; |
fd6fab2c JA |
1320 | ret = 1; |
1321 | } | |
2579f913 | 1322 | state->free_reqs = ret - 1; |
6c8a3134 | 1323 | req = state->reqs[ret - 1]; |
2579f913 | 1324 | } else { |
2579f913 | 1325 | state->free_reqs--; |
6c8a3134 | 1326 | req = state->reqs[state->free_reqs]; |
2b188cc1 JA |
1327 | } |
1328 | ||
2579f913 | 1329 | return req; |
0ddf92e8 | 1330 | fallback: |
0553b8bd | 1331 | return io_get_fallback_req(ctx); |
2b188cc1 JA |
1332 | } |
1333 | ||
8da11c19 PB |
1334 | static inline void io_put_file(struct io_kiocb *req, struct file *file, |
1335 | bool fixed) | |
1336 | { | |
1337 | if (fixed) | |
05589553 | 1338 | percpu_ref_put(req->fixed_file_refs); |
8da11c19 PB |
1339 | else |
1340 | fput(file); | |
1341 | } | |
1342 | ||
c6ca97b3 | 1343 | static void __io_req_aux_free(struct io_kiocb *req) |
2b188cc1 | 1344 | { |
929a3af9 PB |
1345 | if (req->flags & REQ_F_NEED_CLEANUP) |
1346 | io_cleanup_req(req); | |
1347 | ||
96fd84d8 | 1348 | kfree(req->io); |
8da11c19 PB |
1349 | if (req->file) |
1350 | io_put_file(req, req->file, (req->flags & REQ_F_FIXED_FILE)); | |
3537b6a7 JA |
1351 | if (req->task) |
1352 | put_task_struct(req->task); | |
cccf0ee8 JA |
1353 | |
1354 | io_req_work_drop_env(req); | |
def596e9 JA |
1355 | } |
1356 | ||
9e645e11 | 1357 | static void __io_free_req(struct io_kiocb *req) |
2b188cc1 | 1358 | { |
c6ca97b3 | 1359 | __io_req_aux_free(req); |
fcb323cc | 1360 | |
fcb323cc | 1361 | if (req->flags & REQ_F_INFLIGHT) { |
c6ca97b3 | 1362 | struct io_ring_ctx *ctx = req->ctx; |
fcb323cc JA |
1363 | unsigned long flags; |
1364 | ||
1365 | spin_lock_irqsave(&ctx->inflight_lock, flags); | |
1366 | list_del(&req->inflight_entry); | |
1367 | if (waitqueue_active(&ctx->inflight_wait)) | |
1368 | wake_up(&ctx->inflight_wait); | |
1369 | spin_unlock_irqrestore(&ctx->inflight_lock, flags); | |
1370 | } | |
2b85edfc PB |
1371 | |
1372 | percpu_ref_put(&req->ctx->refs); | |
b1e50e54 PB |
1373 | if (likely(!io_is_fallback_req(req))) |
1374 | kmem_cache_free(req_cachep, req); | |
1375 | else | |
dd461af6 | 1376 | clear_bit_unlock(0, (unsigned long *) &req->ctx->fallback_req); |
e65ef56d JA |
1377 | } |
1378 | ||
c6ca97b3 JA |
1379 | struct req_batch { |
1380 | void *reqs[IO_IOPOLL_BATCH]; | |
1381 | int to_free; | |
1382 | int need_iter; | |
1383 | }; | |
1384 | ||
1385 | static void io_free_req_many(struct io_ring_ctx *ctx, struct req_batch *rb) | |
1386 | { | |
1387 | if (!rb->to_free) | |
1388 | return; | |
1389 | if (rb->need_iter) { | |
1390 | int i, inflight = 0; | |
1391 | unsigned long flags; | |
1392 | ||
1393 | for (i = 0; i < rb->to_free; i++) { | |
1394 | struct io_kiocb *req = rb->reqs[i]; | |
1395 | ||
10fef4be | 1396 | if (req->flags & REQ_F_FIXED_FILE) { |
c6ca97b3 | 1397 | req->file = NULL; |
05589553 | 1398 | percpu_ref_put(req->fixed_file_refs); |
10fef4be | 1399 | } |
c6ca97b3 JA |
1400 | if (req->flags & REQ_F_INFLIGHT) |
1401 | inflight++; | |
c6ca97b3 JA |
1402 | __io_req_aux_free(req); |
1403 | } | |
1404 | if (!inflight) | |
1405 | goto do_free; | |
1406 | ||
1407 | spin_lock_irqsave(&ctx->inflight_lock, flags); | |
1408 | for (i = 0; i < rb->to_free; i++) { | |
1409 | struct io_kiocb *req = rb->reqs[i]; | |
1410 | ||
10fef4be | 1411 | if (req->flags & REQ_F_INFLIGHT) { |
c6ca97b3 JA |
1412 | list_del(&req->inflight_entry); |
1413 | if (!--inflight) | |
1414 | break; | |
1415 | } | |
1416 | } | |
1417 | spin_unlock_irqrestore(&ctx->inflight_lock, flags); | |
1418 | ||
1419 | if (waitqueue_active(&ctx->inflight_wait)) | |
1420 | wake_up(&ctx->inflight_wait); | |
1421 | } | |
1422 | do_free: | |
1423 | kmem_cache_free_bulk(req_cachep, rb->to_free, rb->reqs); | |
1424 | percpu_ref_put_many(&ctx->refs, rb->to_free); | |
c6ca97b3 | 1425 | rb->to_free = rb->need_iter = 0; |
e65ef56d JA |
1426 | } |
1427 | ||
a197f664 | 1428 | static bool io_link_cancel_timeout(struct io_kiocb *req) |
2665abfd | 1429 | { |
a197f664 | 1430 | struct io_ring_ctx *ctx = req->ctx; |
2665abfd JA |
1431 | int ret; |
1432 | ||
2d28390a | 1433 | ret = hrtimer_try_to_cancel(&req->io->timeout.timer); |
2665abfd | 1434 | if (ret != -1) { |
78e19bbe | 1435 | io_cqring_fill_event(req, -ECANCELED); |
2665abfd | 1436 | io_commit_cqring(ctx); |
dea3b49c | 1437 | req->flags &= ~REQ_F_LINK_HEAD; |
ec9c02ad | 1438 | io_put_req(req); |
2665abfd JA |
1439 | return true; |
1440 | } | |
1441 | ||
1442 | return false; | |
e65ef56d JA |
1443 | } |
1444 | ||
ba816ad6 | 1445 | static void io_req_link_next(struct io_kiocb *req, struct io_kiocb **nxtptr) |
9e645e11 | 1446 | { |
2665abfd | 1447 | struct io_ring_ctx *ctx = req->ctx; |
2665abfd | 1448 | bool wake_ev = false; |
9e645e11 | 1449 | |
4d7dd462 JA |
1450 | /* Already got next link */ |
1451 | if (req->flags & REQ_F_LINK_NEXT) | |
1452 | return; | |
1453 | ||
9e645e11 JA |
1454 | /* |
1455 | * The list should never be empty when we are called here. But could | |
1456 | * potentially happen if the chain is messed up, check to be on the | |
1457 | * safe side. | |
1458 | */ | |
4493233e PB |
1459 | while (!list_empty(&req->link_list)) { |
1460 | struct io_kiocb *nxt = list_first_entry(&req->link_list, | |
1461 | struct io_kiocb, link_list); | |
94ae5e77 | 1462 | |
4493233e PB |
1463 | if (unlikely((req->flags & REQ_F_LINK_TIMEOUT) && |
1464 | (nxt->flags & REQ_F_TIMEOUT))) { | |
1465 | list_del_init(&nxt->link_list); | |
94ae5e77 | 1466 | wake_ev |= io_link_cancel_timeout(nxt); |
94ae5e77 JA |
1467 | req->flags &= ~REQ_F_LINK_TIMEOUT; |
1468 | continue; | |
1469 | } | |
9e645e11 | 1470 | |
4493233e PB |
1471 | list_del_init(&req->link_list); |
1472 | if (!list_empty(&nxt->link_list)) | |
dea3b49c | 1473 | nxt->flags |= REQ_F_LINK_HEAD; |
b18fdf71 | 1474 | *nxtptr = nxt; |
94ae5e77 | 1475 | break; |
9e645e11 | 1476 | } |
2665abfd | 1477 | |
4d7dd462 | 1478 | req->flags |= REQ_F_LINK_NEXT; |
2665abfd JA |
1479 | if (wake_ev) |
1480 | io_cqring_ev_posted(ctx); | |
9e645e11 JA |
1481 | } |
1482 | ||
1483 | /* | |
dea3b49c | 1484 | * Called if REQ_F_LINK_HEAD is set, and we fail the head request |
9e645e11 JA |
1485 | */ |
1486 | static void io_fail_links(struct io_kiocb *req) | |
1487 | { | |
2665abfd | 1488 | struct io_ring_ctx *ctx = req->ctx; |
2665abfd JA |
1489 | unsigned long flags; |
1490 | ||
1491 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
9e645e11 JA |
1492 | |
1493 | while (!list_empty(&req->link_list)) { | |
4493233e PB |
1494 | struct io_kiocb *link = list_first_entry(&req->link_list, |
1495 | struct io_kiocb, link_list); | |
9e645e11 | 1496 | |
4493233e | 1497 | list_del_init(&link->link_list); |
c826bd7a | 1498 | trace_io_uring_fail_link(req, link); |
2665abfd JA |
1499 | |
1500 | if ((req->flags & REQ_F_LINK_TIMEOUT) && | |
d625c6ee | 1501 | link->opcode == IORING_OP_LINK_TIMEOUT) { |
a197f664 | 1502 | io_link_cancel_timeout(link); |
2665abfd | 1503 | } else { |
78e19bbe | 1504 | io_cqring_fill_event(link, -ECANCELED); |
978db57e | 1505 | __io_double_put_req(link); |
2665abfd | 1506 | } |
5d960724 | 1507 | req->flags &= ~REQ_F_LINK_TIMEOUT; |
9e645e11 | 1508 | } |
2665abfd JA |
1509 | |
1510 | io_commit_cqring(ctx); | |
1511 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
1512 | io_cqring_ev_posted(ctx); | |
9e645e11 JA |
1513 | } |
1514 | ||
4d7dd462 | 1515 | static void io_req_find_next(struct io_kiocb *req, struct io_kiocb **nxt) |
9e645e11 | 1516 | { |
dea3b49c | 1517 | if (likely(!(req->flags & REQ_F_LINK_HEAD))) |
2665abfd | 1518 | return; |
2665abfd | 1519 | |
9e645e11 JA |
1520 | /* |
1521 | * If LINK is set, we have dependent requests in this chain. If we | |
1522 | * didn't fail this request, queue the first one up, moving any other | |
1523 | * dependencies to the next request. In case of failure, fail the rest | |
1524 | * of the chain. | |
1525 | */ | |
2665abfd JA |
1526 | if (req->flags & REQ_F_FAIL_LINK) { |
1527 | io_fail_links(req); | |
7c9e7f0f JA |
1528 | } else if ((req->flags & (REQ_F_LINK_TIMEOUT | REQ_F_COMP_LOCKED)) == |
1529 | REQ_F_LINK_TIMEOUT) { | |
2665abfd JA |
1530 | struct io_ring_ctx *ctx = req->ctx; |
1531 | unsigned long flags; | |
1532 | ||
1533 | /* | |
1534 | * If this is a timeout link, we could be racing with the | |
1535 | * timeout timer. Grab the completion lock for this case to | |
7c9e7f0f | 1536 | * protect against that. |
2665abfd JA |
1537 | */ |
1538 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
1539 | io_req_link_next(req, nxt); | |
1540 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
1541 | } else { | |
1542 | io_req_link_next(req, nxt); | |
9e645e11 | 1543 | } |
4d7dd462 | 1544 | } |
9e645e11 | 1545 | |
c69f8dbe JL |
1546 | static void io_free_req(struct io_kiocb *req) |
1547 | { | |
944e58bf PB |
1548 | struct io_kiocb *nxt = NULL; |
1549 | ||
1550 | io_req_find_next(req, &nxt); | |
70cf9f32 | 1551 | __io_free_req(req); |
944e58bf PB |
1552 | |
1553 | if (nxt) | |
1554 | io_queue_async_work(nxt); | |
c69f8dbe JL |
1555 | } |
1556 | ||
7a743e22 PB |
1557 | static void io_link_work_cb(struct io_wq_work **workptr) |
1558 | { | |
18a542ff PB |
1559 | struct io_kiocb *req = container_of(*workptr, struct io_kiocb, work); |
1560 | struct io_kiocb *link; | |
7a743e22 | 1561 | |
18a542ff | 1562 | link = list_first_entry(&req->link_list, struct io_kiocb, link_list); |
7a743e22 PB |
1563 | io_queue_linked_timeout(link); |
1564 | io_wq_submit_work(workptr); | |
1565 | } | |
1566 | ||
1567 | static void io_wq_assign_next(struct io_wq_work **workptr, struct io_kiocb *nxt) | |
1568 | { | |
1569 | struct io_kiocb *link; | |
8766dd51 PB |
1570 | const struct io_op_def *def = &io_op_defs[nxt->opcode]; |
1571 | ||
1572 | if ((nxt->flags & REQ_F_ISREG) && def->hash_reg_file) | |
1573 | io_wq_hash_work(&nxt->work, file_inode(nxt->file)); | |
7a743e22 PB |
1574 | |
1575 | *workptr = &nxt->work; | |
1576 | link = io_prep_linked_timeout(nxt); | |
18a542ff | 1577 | if (link) |
7a743e22 | 1578 | nxt->work.func = io_link_work_cb; |
7a743e22 PB |
1579 | } |
1580 | ||
ba816ad6 JA |
1581 | /* |
1582 | * Drop reference to request, return next in chain (if there is one) if this | |
1583 | * was the last reference to this request. | |
1584 | */ | |
f9bd67f6 | 1585 | __attribute__((nonnull)) |
ec9c02ad | 1586 | static void io_put_req_find_next(struct io_kiocb *req, struct io_kiocb **nxtptr) |
e65ef56d | 1587 | { |
2a44f467 JA |
1588 | if (refcount_dec_and_test(&req->refs)) { |
1589 | io_req_find_next(req, nxtptr); | |
4d7dd462 | 1590 | __io_free_req(req); |
2a44f467 | 1591 | } |
2b188cc1 JA |
1592 | } |
1593 | ||
e65ef56d JA |
1594 | static void io_put_req(struct io_kiocb *req) |
1595 | { | |
1596 | if (refcount_dec_and_test(&req->refs)) | |
1597 | io_free_req(req); | |
2b188cc1 JA |
1598 | } |
1599 | ||
e9fd9396 PB |
1600 | static void io_steal_work(struct io_kiocb *req, |
1601 | struct io_wq_work **workptr) | |
7a743e22 PB |
1602 | { |
1603 | /* | |
1604 | * It's in an io-wq worker, so there always should be at least | |
1605 | * one reference, which will be dropped in io_put_work() just | |
1606 | * after the current handler returns. | |
1607 | * | |
1608 | * It also means, that if the counter dropped to 1, then there is | |
1609 | * no asynchronous users left, so it's safe to steal the next work. | |
1610 | */ | |
7a743e22 PB |
1611 | if (refcount_read(&req->refs) == 1) { |
1612 | struct io_kiocb *nxt = NULL; | |
1613 | ||
1614 | io_req_find_next(req, &nxt); | |
1615 | if (nxt) | |
1616 | io_wq_assign_next(workptr, nxt); | |
1617 | } | |
1618 | } | |
1619 | ||
978db57e JA |
1620 | /* |
1621 | * Must only be used if we don't need to care about links, usually from | |
1622 | * within the completion handling itself. | |
1623 | */ | |
1624 | static void __io_double_put_req(struct io_kiocb *req) | |
78e19bbe JA |
1625 | { |
1626 | /* drop both submit and complete references */ | |
1627 | if (refcount_sub_and_test(2, &req->refs)) | |
1628 | __io_free_req(req); | |
1629 | } | |
1630 | ||
978db57e JA |
1631 | static void io_double_put_req(struct io_kiocb *req) |
1632 | { | |
1633 | /* drop both submit and complete references */ | |
1634 | if (refcount_sub_and_test(2, &req->refs)) | |
1635 | io_free_req(req); | |
1636 | } | |
1637 | ||
1d7bb1d5 | 1638 | static unsigned io_cqring_events(struct io_ring_ctx *ctx, bool noflush) |
a3a0e43f | 1639 | { |
84f97dc2 JA |
1640 | struct io_rings *rings = ctx->rings; |
1641 | ||
ad3eb2c8 JA |
1642 | if (test_bit(0, &ctx->cq_check_overflow)) { |
1643 | /* | |
1644 | * noflush == true is from the waitqueue handler, just ensure | |
1645 | * we wake up the task, and the next invocation will flush the | |
1646 | * entries. We cannot safely to it from here. | |
1647 | */ | |
1648 | if (noflush && !list_empty(&ctx->cq_overflow_list)) | |
1649 | return -1U; | |
1d7bb1d5 | 1650 | |
ad3eb2c8 JA |
1651 | io_cqring_overflow_flush(ctx, false); |
1652 | } | |
1d7bb1d5 | 1653 | |
a3a0e43f JA |
1654 | /* See comment at the top of this file */ |
1655 | smp_rmb(); | |
ad3eb2c8 | 1656 | return ctx->cached_cq_tail - READ_ONCE(rings->cq.head); |
a3a0e43f JA |
1657 | } |
1658 | ||
fb5ccc98 PB |
1659 | static inline unsigned int io_sqring_entries(struct io_ring_ctx *ctx) |
1660 | { | |
1661 | struct io_rings *rings = ctx->rings; | |
1662 | ||
1663 | /* make sure SQ entry isn't read before tail */ | |
1664 | return smp_load_acquire(&rings->sq.tail) - ctx->cached_sq_head; | |
1665 | } | |
1666 | ||
8237e045 | 1667 | static inline bool io_req_multi_free(struct req_batch *rb, struct io_kiocb *req) |
e94f141b | 1668 | { |
dea3b49c | 1669 | if ((req->flags & REQ_F_LINK_HEAD) || io_is_fallback_req(req)) |
c6ca97b3 | 1670 | return false; |
e94f141b | 1671 | |
c6ca97b3 JA |
1672 | if (!(req->flags & REQ_F_FIXED_FILE) || req->io) |
1673 | rb->need_iter++; | |
1674 | ||
1675 | rb->reqs[rb->to_free++] = req; | |
1676 | if (unlikely(rb->to_free == ARRAY_SIZE(rb->reqs))) | |
1677 | io_free_req_many(req->ctx, rb); | |
1678 | return true; | |
e94f141b JA |
1679 | } |
1680 | ||
bcda7baa JA |
1681 | static int io_put_kbuf(struct io_kiocb *req) |
1682 | { | |
4d954c25 | 1683 | struct io_buffer *kbuf; |
bcda7baa JA |
1684 | int cflags; |
1685 | ||
4d954c25 | 1686 | kbuf = (struct io_buffer *) (unsigned long) req->rw.addr; |
bcda7baa JA |
1687 | cflags = kbuf->bid << IORING_CQE_BUFFER_SHIFT; |
1688 | cflags |= IORING_CQE_F_BUFFER; | |
1689 | req->rw.addr = 0; | |
1690 | kfree(kbuf); | |
1691 | return cflags; | |
1692 | } | |
1693 | ||
def596e9 JA |
1694 | /* |
1695 | * Find and free completed poll iocbs | |
1696 | */ | |
1697 | static void io_iopoll_complete(struct io_ring_ctx *ctx, unsigned int *nr_events, | |
1698 | struct list_head *done) | |
1699 | { | |
8237e045 | 1700 | struct req_batch rb; |
def596e9 | 1701 | struct io_kiocb *req; |
def596e9 | 1702 | |
c6ca97b3 | 1703 | rb.to_free = rb.need_iter = 0; |
def596e9 | 1704 | while (!list_empty(done)) { |
bcda7baa JA |
1705 | int cflags = 0; |
1706 | ||
def596e9 JA |
1707 | req = list_first_entry(done, struct io_kiocb, list); |
1708 | list_del(&req->list); | |
1709 | ||
bcda7baa JA |
1710 | if (req->flags & REQ_F_BUFFER_SELECTED) |
1711 | cflags = io_put_kbuf(req); | |
1712 | ||
1713 | __io_cqring_fill_event(req, req->result, cflags); | |
def596e9 JA |
1714 | (*nr_events)++; |
1715 | ||
8237e045 JA |
1716 | if (refcount_dec_and_test(&req->refs) && |
1717 | !io_req_multi_free(&rb, req)) | |
1718 | io_free_req(req); | |
def596e9 | 1719 | } |
def596e9 | 1720 | |
09bb8394 | 1721 | io_commit_cqring(ctx); |
32b2244a XW |
1722 | if (ctx->flags & IORING_SETUP_SQPOLL) |
1723 | io_cqring_ev_posted(ctx); | |
8237e045 | 1724 | io_free_req_many(ctx, &rb); |
def596e9 JA |
1725 | } |
1726 | ||
581f9810 BM |
1727 | static void io_iopoll_queue(struct list_head *again) |
1728 | { | |
1729 | struct io_kiocb *req; | |
1730 | ||
1731 | do { | |
1732 | req = list_first_entry(again, struct io_kiocb, list); | |
1733 | list_del(&req->list); | |
1734 | refcount_inc(&req->refs); | |
1735 | io_queue_async_work(req); | |
1736 | } while (!list_empty(again)); | |
1737 | } | |
1738 | ||
def596e9 JA |
1739 | static int io_do_iopoll(struct io_ring_ctx *ctx, unsigned int *nr_events, |
1740 | long min) | |
1741 | { | |
1742 | struct io_kiocb *req, *tmp; | |
1743 | LIST_HEAD(done); | |
581f9810 | 1744 | LIST_HEAD(again); |
def596e9 JA |
1745 | bool spin; |
1746 | int ret; | |
1747 | ||
1748 | /* | |
1749 | * Only spin for completions if we don't have multiple devices hanging | |
1750 | * off our complete list, and we're under the requested amount. | |
1751 | */ | |
1752 | spin = !ctx->poll_multi_file && *nr_events < min; | |
1753 | ||
1754 | ret = 0; | |
1755 | list_for_each_entry_safe(req, tmp, &ctx->poll_list, list) { | |
9adbd45d | 1756 | struct kiocb *kiocb = &req->rw.kiocb; |
def596e9 JA |
1757 | |
1758 | /* | |
581f9810 BM |
1759 | * Move completed and retryable entries to our local lists. |
1760 | * If we find a request that requires polling, break out | |
1761 | * and complete those lists first, if we have entries there. | |
def596e9 JA |
1762 | */ |
1763 | if (req->flags & REQ_F_IOPOLL_COMPLETED) { | |
1764 | list_move_tail(&req->list, &done); | |
1765 | continue; | |
1766 | } | |
1767 | if (!list_empty(&done)) | |
1768 | break; | |
1769 | ||
581f9810 BM |
1770 | if (req->result == -EAGAIN) { |
1771 | list_move_tail(&req->list, &again); | |
1772 | continue; | |
1773 | } | |
1774 | if (!list_empty(&again)) | |
1775 | break; | |
1776 | ||
def596e9 JA |
1777 | ret = kiocb->ki_filp->f_op->iopoll(kiocb, spin); |
1778 | if (ret < 0) | |
1779 | break; | |
1780 | ||
1781 | if (ret && spin) | |
1782 | spin = false; | |
1783 | ret = 0; | |
1784 | } | |
1785 | ||
1786 | if (!list_empty(&done)) | |
1787 | io_iopoll_complete(ctx, nr_events, &done); | |
1788 | ||
581f9810 BM |
1789 | if (!list_empty(&again)) |
1790 | io_iopoll_queue(&again); | |
1791 | ||
def596e9 JA |
1792 | return ret; |
1793 | } | |
1794 | ||
1795 | /* | |
d195a66e | 1796 | * Poll for a minimum of 'min' events. Note that if min == 0 we consider that a |
def596e9 JA |
1797 | * non-spinning poll check - we'll still enter the driver poll loop, but only |
1798 | * as a non-spinning completion check. | |
1799 | */ | |
1800 | static int io_iopoll_getevents(struct io_ring_ctx *ctx, unsigned int *nr_events, | |
1801 | long min) | |
1802 | { | |
08f5439f | 1803 | while (!list_empty(&ctx->poll_list) && !need_resched()) { |
def596e9 JA |
1804 | int ret; |
1805 | ||
1806 | ret = io_do_iopoll(ctx, nr_events, min); | |
1807 | if (ret < 0) | |
1808 | return ret; | |
1809 | if (!min || *nr_events >= min) | |
1810 | return 0; | |
1811 | } | |
1812 | ||
1813 | return 1; | |
1814 | } | |
1815 | ||
1816 | /* | |
1817 | * We can't just wait for polled events to come to us, we have to actively | |
1818 | * find and complete them. | |
1819 | */ | |
1820 | static void io_iopoll_reap_events(struct io_ring_ctx *ctx) | |
1821 | { | |
1822 | if (!(ctx->flags & IORING_SETUP_IOPOLL)) | |
1823 | return; | |
1824 | ||
1825 | mutex_lock(&ctx->uring_lock); | |
1826 | while (!list_empty(&ctx->poll_list)) { | |
1827 | unsigned int nr_events = 0; | |
1828 | ||
1829 | io_iopoll_getevents(ctx, &nr_events, 1); | |
08f5439f JA |
1830 | |
1831 | /* | |
1832 | * Ensure we allow local-to-the-cpu processing to take place, | |
1833 | * in this case we need to ensure that we reap all events. | |
1834 | */ | |
1835 | cond_resched(); | |
def596e9 JA |
1836 | } |
1837 | mutex_unlock(&ctx->uring_lock); | |
1838 | } | |
1839 | ||
c7849be9 XW |
1840 | static int io_iopoll_check(struct io_ring_ctx *ctx, unsigned *nr_events, |
1841 | long min) | |
def596e9 | 1842 | { |
2b2ed975 | 1843 | int iters = 0, ret = 0; |
500f9fba | 1844 | |
c7849be9 XW |
1845 | /* |
1846 | * We disallow the app entering submit/complete with polling, but we | |
1847 | * still need to lock the ring to prevent racing with polled issue | |
1848 | * that got punted to a workqueue. | |
1849 | */ | |
1850 | mutex_lock(&ctx->uring_lock); | |
def596e9 JA |
1851 | do { |
1852 | int tmin = 0; | |
1853 | ||
a3a0e43f JA |
1854 | /* |
1855 | * Don't enter poll loop if we already have events pending. | |
1856 | * If we do, we can potentially be spinning for commands that | |
1857 | * already triggered a CQE (eg in error). | |
1858 | */ | |
1d7bb1d5 | 1859 | if (io_cqring_events(ctx, false)) |
a3a0e43f JA |
1860 | break; |
1861 | ||
500f9fba JA |
1862 | /* |
1863 | * If a submit got punted to a workqueue, we can have the | |
1864 | * application entering polling for a command before it gets | |
1865 | * issued. That app will hold the uring_lock for the duration | |
1866 | * of the poll right here, so we need to take a breather every | |
1867 | * now and then to ensure that the issue has a chance to add | |
1868 | * the poll to the issued list. Otherwise we can spin here | |
1869 | * forever, while the workqueue is stuck trying to acquire the | |
1870 | * very same mutex. | |
1871 | */ | |
1872 | if (!(++iters & 7)) { | |
1873 | mutex_unlock(&ctx->uring_lock); | |
1874 | mutex_lock(&ctx->uring_lock); | |
1875 | } | |
1876 | ||
def596e9 JA |
1877 | if (*nr_events < min) |
1878 | tmin = min - *nr_events; | |
1879 | ||
1880 | ret = io_iopoll_getevents(ctx, nr_events, tmin); | |
1881 | if (ret <= 0) | |
1882 | break; | |
1883 | ret = 0; | |
1884 | } while (min && !*nr_events && !need_resched()); | |
1885 | ||
500f9fba | 1886 | mutex_unlock(&ctx->uring_lock); |
def596e9 JA |
1887 | return ret; |
1888 | } | |
1889 | ||
491381ce | 1890 | static void kiocb_end_write(struct io_kiocb *req) |
2b188cc1 | 1891 | { |
491381ce JA |
1892 | /* |
1893 | * Tell lockdep we inherited freeze protection from submission | |
1894 | * thread. | |
1895 | */ | |
1896 | if (req->flags & REQ_F_ISREG) { | |
1897 | struct inode *inode = file_inode(req->file); | |
2b188cc1 | 1898 | |
491381ce | 1899 | __sb_writers_acquired(inode->i_sb, SB_FREEZE_WRITE); |
2b188cc1 | 1900 | } |
491381ce | 1901 | file_end_write(req->file); |
2b188cc1 JA |
1902 | } |
1903 | ||
4e88d6e7 JA |
1904 | static inline void req_set_fail_links(struct io_kiocb *req) |
1905 | { | |
1906 | if ((req->flags & (REQ_F_LINK | REQ_F_HARDLINK)) == REQ_F_LINK) | |
1907 | req->flags |= REQ_F_FAIL_LINK; | |
1908 | } | |
1909 | ||
ba816ad6 | 1910 | static void io_complete_rw_common(struct kiocb *kiocb, long res) |
2b188cc1 | 1911 | { |
9adbd45d | 1912 | struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb); |
bcda7baa | 1913 | int cflags = 0; |
2b188cc1 | 1914 | |
491381ce JA |
1915 | if (kiocb->ki_flags & IOCB_WRITE) |
1916 | kiocb_end_write(req); | |
2b188cc1 | 1917 | |
4e88d6e7 JA |
1918 | if (res != req->result) |
1919 | req_set_fail_links(req); | |
bcda7baa JA |
1920 | if (req->flags & REQ_F_BUFFER_SELECTED) |
1921 | cflags = io_put_kbuf(req); | |
1922 | __io_cqring_add_event(req, res, cflags); | |
ba816ad6 JA |
1923 | } |
1924 | ||
1925 | static void io_complete_rw(struct kiocb *kiocb, long res, long res2) | |
1926 | { | |
9adbd45d | 1927 | struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb); |
ba816ad6 JA |
1928 | |
1929 | io_complete_rw_common(kiocb, res); | |
e65ef56d | 1930 | io_put_req(req); |
2b188cc1 JA |
1931 | } |
1932 | ||
def596e9 JA |
1933 | static void io_complete_rw_iopoll(struct kiocb *kiocb, long res, long res2) |
1934 | { | |
9adbd45d | 1935 | struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb); |
def596e9 | 1936 | |
491381ce JA |
1937 | if (kiocb->ki_flags & IOCB_WRITE) |
1938 | kiocb_end_write(req); | |
def596e9 | 1939 | |
4e88d6e7 JA |
1940 | if (res != req->result) |
1941 | req_set_fail_links(req); | |
9e645e11 | 1942 | req->result = res; |
def596e9 JA |
1943 | if (res != -EAGAIN) |
1944 | req->flags |= REQ_F_IOPOLL_COMPLETED; | |
1945 | } | |
1946 | ||
1947 | /* | |
1948 | * After the iocb has been issued, it's safe to be found on the poll list. | |
1949 | * Adding the kiocb to the list AFTER submission ensures that we don't | |
1950 | * find it from a io_iopoll_getevents() thread before the issuer is done | |
1951 | * accessing the kiocb cookie. | |
1952 | */ | |
1953 | static void io_iopoll_req_issued(struct io_kiocb *req) | |
1954 | { | |
1955 | struct io_ring_ctx *ctx = req->ctx; | |
1956 | ||
1957 | /* | |
1958 | * Track whether we have multiple files in our lists. This will impact | |
1959 | * how we do polling eventually, not spinning if we're on potentially | |
1960 | * different devices. | |
1961 | */ | |
1962 | if (list_empty(&ctx->poll_list)) { | |
1963 | ctx->poll_multi_file = false; | |
1964 | } else if (!ctx->poll_multi_file) { | |
1965 | struct io_kiocb *list_req; | |
1966 | ||
1967 | list_req = list_first_entry(&ctx->poll_list, struct io_kiocb, | |
1968 | list); | |
9adbd45d | 1969 | if (list_req->file != req->file) |
def596e9 JA |
1970 | ctx->poll_multi_file = true; |
1971 | } | |
1972 | ||
1973 | /* | |
1974 | * For fast devices, IO may have already completed. If it has, add | |
1975 | * it to the front so we find it first. | |
1976 | */ | |
1977 | if (req->flags & REQ_F_IOPOLL_COMPLETED) | |
1978 | list_add(&req->list, &ctx->poll_list); | |
1979 | else | |
1980 | list_add_tail(&req->list, &ctx->poll_list); | |
bdcd3eab XW |
1981 | |
1982 | if ((ctx->flags & IORING_SETUP_SQPOLL) && | |
1983 | wq_has_sleeper(&ctx->sqo_wait)) | |
1984 | wake_up(&ctx->sqo_wait); | |
def596e9 JA |
1985 | } |
1986 | ||
3d6770fb | 1987 | static void io_file_put(struct io_submit_state *state) |
9a56a232 | 1988 | { |
3d6770fb | 1989 | if (state->file) { |
9a56a232 JA |
1990 | int diff = state->has_refs - state->used_refs; |
1991 | ||
1992 | if (diff) | |
1993 | fput_many(state->file, diff); | |
1994 | state->file = NULL; | |
1995 | } | |
1996 | } | |
1997 | ||
1998 | /* | |
1999 | * Get as many references to a file as we have IOs left in this submission, | |
2000 | * assuming most submissions are for one file, or at least that each file | |
2001 | * has more than one submission. | |
2002 | */ | |
8da11c19 | 2003 | static struct file *__io_file_get(struct io_submit_state *state, int fd) |
9a56a232 JA |
2004 | { |
2005 | if (!state) | |
2006 | return fget(fd); | |
2007 | ||
2008 | if (state->file) { | |
2009 | if (state->fd == fd) { | |
2010 | state->used_refs++; | |
2011 | state->ios_left--; | |
2012 | return state->file; | |
2013 | } | |
3d6770fb | 2014 | io_file_put(state); |
9a56a232 JA |
2015 | } |
2016 | state->file = fget_many(fd, state->ios_left); | |
2017 | if (!state->file) | |
2018 | return NULL; | |
2019 | ||
2020 | state->fd = fd; | |
2021 | state->has_refs = state->ios_left; | |
2022 | state->used_refs = 1; | |
2023 | state->ios_left--; | |
2024 | return state->file; | |
2025 | } | |
2026 | ||
2b188cc1 JA |
2027 | /* |
2028 | * If we tracked the file through the SCM inflight mechanism, we could support | |
2029 | * any file. For now, just ensure that anything potentially problematic is done | |
2030 | * inline. | |
2031 | */ | |
af197f50 | 2032 | static bool io_file_supports_async(struct file *file, int rw) |
2b188cc1 JA |
2033 | { |
2034 | umode_t mode = file_inode(file)->i_mode; | |
2035 | ||
10d59345 | 2036 | if (S_ISBLK(mode) || S_ISCHR(mode) || S_ISSOCK(mode)) |
2b188cc1 JA |
2037 | return true; |
2038 | if (S_ISREG(mode) && file->f_op != &io_uring_fops) | |
2039 | return true; | |
2040 | ||
af197f50 JA |
2041 | if (!(file->f_mode & FMODE_NOWAIT)) |
2042 | return false; | |
2043 | ||
2044 | if (rw == READ) | |
2045 | return file->f_op->read_iter != NULL; | |
2046 | ||
2047 | return file->f_op->write_iter != NULL; | |
2b188cc1 JA |
2048 | } |
2049 | ||
3529d8c2 JA |
2050 | static int io_prep_rw(struct io_kiocb *req, const struct io_uring_sqe *sqe, |
2051 | bool force_nonblock) | |
2b188cc1 | 2052 | { |
def596e9 | 2053 | struct io_ring_ctx *ctx = req->ctx; |
9adbd45d | 2054 | struct kiocb *kiocb = &req->rw.kiocb; |
09bb8394 JA |
2055 | unsigned ioprio; |
2056 | int ret; | |
2b188cc1 | 2057 | |
491381ce JA |
2058 | if (S_ISREG(file_inode(req->file)->i_mode)) |
2059 | req->flags |= REQ_F_ISREG; | |
2060 | ||
2b188cc1 | 2061 | kiocb->ki_pos = READ_ONCE(sqe->off); |
ba04291e JA |
2062 | if (kiocb->ki_pos == -1 && !(req->file->f_mode & FMODE_STREAM)) { |
2063 | req->flags |= REQ_F_CUR_POS; | |
2064 | kiocb->ki_pos = req->file->f_pos; | |
2065 | } | |
2b188cc1 | 2066 | kiocb->ki_hint = ki_hint_validate(file_write_hint(kiocb->ki_filp)); |
3e577dcd PB |
2067 | kiocb->ki_flags = iocb_flags(kiocb->ki_filp); |
2068 | ret = kiocb_set_rw_flags(kiocb, READ_ONCE(sqe->rw_flags)); | |
2069 | if (unlikely(ret)) | |
2070 | return ret; | |
2b188cc1 JA |
2071 | |
2072 | ioprio = READ_ONCE(sqe->ioprio); | |
2073 | if (ioprio) { | |
2074 | ret = ioprio_check_cap(ioprio); | |
2075 | if (ret) | |
09bb8394 | 2076 | return ret; |
2b188cc1 JA |
2077 | |
2078 | kiocb->ki_ioprio = ioprio; | |
2079 | } else | |
2080 | kiocb->ki_ioprio = get_current_ioprio(); | |
2081 | ||
8449eeda | 2082 | /* don't allow async punt if RWF_NOWAIT was requested */ |
491381ce JA |
2083 | if ((kiocb->ki_flags & IOCB_NOWAIT) || |
2084 | (req->file->f_flags & O_NONBLOCK)) | |
8449eeda SB |
2085 | req->flags |= REQ_F_NOWAIT; |
2086 | ||
2087 | if (force_nonblock) | |
2b188cc1 | 2088 | kiocb->ki_flags |= IOCB_NOWAIT; |
8449eeda | 2089 | |
def596e9 | 2090 | if (ctx->flags & IORING_SETUP_IOPOLL) { |
def596e9 JA |
2091 | if (!(kiocb->ki_flags & IOCB_DIRECT) || |
2092 | !kiocb->ki_filp->f_op->iopoll) | |
09bb8394 | 2093 | return -EOPNOTSUPP; |
2b188cc1 | 2094 | |
def596e9 JA |
2095 | kiocb->ki_flags |= IOCB_HIPRI; |
2096 | kiocb->ki_complete = io_complete_rw_iopoll; | |
6873e0bd | 2097 | req->result = 0; |
def596e9 | 2098 | } else { |
09bb8394 JA |
2099 | if (kiocb->ki_flags & IOCB_HIPRI) |
2100 | return -EINVAL; | |
def596e9 JA |
2101 | kiocb->ki_complete = io_complete_rw; |
2102 | } | |
9adbd45d | 2103 | |
3529d8c2 JA |
2104 | req->rw.addr = READ_ONCE(sqe->addr); |
2105 | req->rw.len = READ_ONCE(sqe->len); | |
bcda7baa | 2106 | /* we own ->private, reuse it for the buffer index / buffer ID */ |
9adbd45d | 2107 | req->rw.kiocb.private = (void *) (unsigned long) |
3529d8c2 | 2108 | READ_ONCE(sqe->buf_index); |
2b188cc1 | 2109 | return 0; |
2b188cc1 JA |
2110 | } |
2111 | ||
2112 | static inline void io_rw_done(struct kiocb *kiocb, ssize_t ret) | |
2113 | { | |
2114 | switch (ret) { | |
2115 | case -EIOCBQUEUED: | |
2116 | break; | |
2117 | case -ERESTARTSYS: | |
2118 | case -ERESTARTNOINTR: | |
2119 | case -ERESTARTNOHAND: | |
2120 | case -ERESTART_RESTARTBLOCK: | |
2121 | /* | |
2122 | * We can't just restart the syscall, since previously | |
2123 | * submitted sqes may already be in progress. Just fail this | |
2124 | * IO with EINTR. | |
2125 | */ | |
2126 | ret = -EINTR; | |
2127 | /* fall through */ | |
2128 | default: | |
2129 | kiocb->ki_complete(kiocb, ret, 0); | |
2130 | } | |
2131 | } | |
2132 | ||
014db007 | 2133 | static void kiocb_done(struct kiocb *kiocb, ssize_t ret) |
ba816ad6 | 2134 | { |
ba04291e JA |
2135 | struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb); |
2136 | ||
2137 | if (req->flags & REQ_F_CUR_POS) | |
2138 | req->file->f_pos = kiocb->ki_pos; | |
bcaec089 | 2139 | if (ret >= 0 && kiocb->ki_complete == io_complete_rw) |
014db007 | 2140 | io_complete_rw(kiocb, ret, 0); |
ba816ad6 JA |
2141 | else |
2142 | io_rw_done(kiocb, ret); | |
2143 | } | |
2144 | ||
9adbd45d | 2145 | static ssize_t io_import_fixed(struct io_kiocb *req, int rw, |
7d009165 | 2146 | struct iov_iter *iter) |
edafccee | 2147 | { |
9adbd45d JA |
2148 | struct io_ring_ctx *ctx = req->ctx; |
2149 | size_t len = req->rw.len; | |
edafccee JA |
2150 | struct io_mapped_ubuf *imu; |
2151 | unsigned index, buf_index; | |
2152 | size_t offset; | |
2153 | u64 buf_addr; | |
2154 | ||
2155 | /* attempt to use fixed buffers without having provided iovecs */ | |
2156 | if (unlikely(!ctx->user_bufs)) | |
2157 | return -EFAULT; | |
2158 | ||
9adbd45d | 2159 | buf_index = (unsigned long) req->rw.kiocb.private; |
edafccee JA |
2160 | if (unlikely(buf_index >= ctx->nr_user_bufs)) |
2161 | return -EFAULT; | |
2162 | ||
2163 | index = array_index_nospec(buf_index, ctx->nr_user_bufs); | |
2164 | imu = &ctx->user_bufs[index]; | |
9adbd45d | 2165 | buf_addr = req->rw.addr; |
edafccee JA |
2166 | |
2167 | /* overflow */ | |
2168 | if (buf_addr + len < buf_addr) | |
2169 | return -EFAULT; | |
2170 | /* not inside the mapped region */ | |
2171 | if (buf_addr < imu->ubuf || buf_addr + len > imu->ubuf + imu->len) | |
2172 | return -EFAULT; | |
2173 | ||
2174 | /* | |
2175 | * May not be a start of buffer, set size appropriately | |
2176 | * and advance us to the beginning. | |
2177 | */ | |
2178 | offset = buf_addr - imu->ubuf; | |
2179 | iov_iter_bvec(iter, rw, imu->bvec, imu->nr_bvecs, offset + len); | |
bd11b3a3 JA |
2180 | |
2181 | if (offset) { | |
2182 | /* | |
2183 | * Don't use iov_iter_advance() here, as it's really slow for | |
2184 | * using the latter parts of a big fixed buffer - it iterates | |
2185 | * over each segment manually. We can cheat a bit here, because | |
2186 | * we know that: | |
2187 | * | |
2188 | * 1) it's a BVEC iter, we set it up | |
2189 | * 2) all bvecs are PAGE_SIZE in size, except potentially the | |
2190 | * first and last bvec | |
2191 | * | |
2192 | * So just find our index, and adjust the iterator afterwards. | |
2193 | * If the offset is within the first bvec (or the whole first | |
2194 | * bvec, just use iov_iter_advance(). This makes it easier | |
2195 | * since we can just skip the first segment, which may not | |
2196 | * be PAGE_SIZE aligned. | |
2197 | */ | |
2198 | const struct bio_vec *bvec = imu->bvec; | |
2199 | ||
2200 | if (offset <= bvec->bv_len) { | |
2201 | iov_iter_advance(iter, offset); | |
2202 | } else { | |
2203 | unsigned long seg_skip; | |
2204 | ||
2205 | /* skip first vec */ | |
2206 | offset -= bvec->bv_len; | |
2207 | seg_skip = 1 + (offset >> PAGE_SHIFT); | |
2208 | ||
2209 | iter->bvec = bvec + seg_skip; | |
2210 | iter->nr_segs -= seg_skip; | |
99c79f66 | 2211 | iter->count -= bvec->bv_len + offset; |
bd11b3a3 | 2212 | iter->iov_offset = offset & ~PAGE_MASK; |
bd11b3a3 JA |
2213 | } |
2214 | } | |
2215 | ||
5e559561 | 2216 | return len; |
edafccee JA |
2217 | } |
2218 | ||
bcda7baa JA |
2219 | static void io_ring_submit_unlock(struct io_ring_ctx *ctx, bool needs_lock) |
2220 | { | |
2221 | if (needs_lock) | |
2222 | mutex_unlock(&ctx->uring_lock); | |
2223 | } | |
2224 | ||
2225 | static void io_ring_submit_lock(struct io_ring_ctx *ctx, bool needs_lock) | |
2226 | { | |
2227 | /* | |
2228 | * "Normal" inline submissions always hold the uring_lock, since we | |
2229 | * grab it from the system call. Same is true for the SQPOLL offload. | |
2230 | * The only exception is when we've detached the request and issue it | |
2231 | * from an async worker thread, grab the lock for that case. | |
2232 | */ | |
2233 | if (needs_lock) | |
2234 | mutex_lock(&ctx->uring_lock); | |
2235 | } | |
2236 | ||
2237 | static struct io_buffer *io_buffer_select(struct io_kiocb *req, size_t *len, | |
2238 | int bgid, struct io_buffer *kbuf, | |
2239 | bool needs_lock) | |
2240 | { | |
2241 | struct io_buffer *head; | |
2242 | ||
2243 | if (req->flags & REQ_F_BUFFER_SELECTED) | |
2244 | return kbuf; | |
2245 | ||
2246 | io_ring_submit_lock(req->ctx, needs_lock); | |
2247 | ||
2248 | lockdep_assert_held(&req->ctx->uring_lock); | |
2249 | ||
2250 | head = idr_find(&req->ctx->io_buffer_idr, bgid); | |
2251 | if (head) { | |
2252 | if (!list_empty(&head->list)) { | |
2253 | kbuf = list_last_entry(&head->list, struct io_buffer, | |
2254 | list); | |
2255 | list_del(&kbuf->list); | |
2256 | } else { | |
2257 | kbuf = head; | |
2258 | idr_remove(&req->ctx->io_buffer_idr, bgid); | |
2259 | } | |
2260 | if (*len > kbuf->len) | |
2261 | *len = kbuf->len; | |
2262 | } else { | |
2263 | kbuf = ERR_PTR(-ENOBUFS); | |
2264 | } | |
2265 | ||
2266 | io_ring_submit_unlock(req->ctx, needs_lock); | |
2267 | ||
2268 | return kbuf; | |
2269 | } | |
2270 | ||
4d954c25 JA |
2271 | static void __user *io_rw_buffer_select(struct io_kiocb *req, size_t *len, |
2272 | bool needs_lock) | |
2273 | { | |
2274 | struct io_buffer *kbuf; | |
2275 | int bgid; | |
2276 | ||
2277 | kbuf = (struct io_buffer *) (unsigned long) req->rw.addr; | |
2278 | bgid = (int) (unsigned long) req->rw.kiocb.private; | |
2279 | kbuf = io_buffer_select(req, len, bgid, kbuf, needs_lock); | |
2280 | if (IS_ERR(kbuf)) | |
2281 | return kbuf; | |
2282 | req->rw.addr = (u64) (unsigned long) kbuf; | |
2283 | req->flags |= REQ_F_BUFFER_SELECTED; | |
2284 | return u64_to_user_ptr(kbuf->addr); | |
2285 | } | |
2286 | ||
2287 | #ifdef CONFIG_COMPAT | |
2288 | static ssize_t io_compat_import(struct io_kiocb *req, struct iovec *iov, | |
2289 | bool needs_lock) | |
2290 | { | |
2291 | struct compat_iovec __user *uiov; | |
2292 | compat_ssize_t clen; | |
2293 | void __user *buf; | |
2294 | ssize_t len; | |
2295 | ||
2296 | uiov = u64_to_user_ptr(req->rw.addr); | |
2297 | if (!access_ok(uiov, sizeof(*uiov))) | |
2298 | return -EFAULT; | |
2299 | if (__get_user(clen, &uiov->iov_len)) | |
2300 | return -EFAULT; | |
2301 | if (clen < 0) | |
2302 | return -EINVAL; | |
2303 | ||
2304 | len = clen; | |
2305 | buf = io_rw_buffer_select(req, &len, needs_lock); | |
2306 | if (IS_ERR(buf)) | |
2307 | return PTR_ERR(buf); | |
2308 | iov[0].iov_base = buf; | |
2309 | iov[0].iov_len = (compat_size_t) len; | |
2310 | return 0; | |
2311 | } | |
2312 | #endif | |
2313 | ||
2314 | static ssize_t __io_iov_buffer_select(struct io_kiocb *req, struct iovec *iov, | |
2315 | bool needs_lock) | |
2316 | { | |
2317 | struct iovec __user *uiov = u64_to_user_ptr(req->rw.addr); | |
2318 | void __user *buf; | |
2319 | ssize_t len; | |
2320 | ||
2321 | if (copy_from_user(iov, uiov, sizeof(*uiov))) | |
2322 | return -EFAULT; | |
2323 | ||
2324 | len = iov[0].iov_len; | |
2325 | if (len < 0) | |
2326 | return -EINVAL; | |
2327 | buf = io_rw_buffer_select(req, &len, needs_lock); | |
2328 | if (IS_ERR(buf)) | |
2329 | return PTR_ERR(buf); | |
2330 | iov[0].iov_base = buf; | |
2331 | iov[0].iov_len = len; | |
2332 | return 0; | |
2333 | } | |
2334 | ||
2335 | static ssize_t io_iov_buffer_select(struct io_kiocb *req, struct iovec *iov, | |
2336 | bool needs_lock) | |
2337 | { | |
2338 | if (req->flags & REQ_F_BUFFER_SELECTED) | |
2339 | return 0; | |
2340 | if (!req->rw.len) | |
2341 | return 0; | |
2342 | else if (req->rw.len > 1) | |
2343 | return -EINVAL; | |
2344 | ||
2345 | #ifdef CONFIG_COMPAT | |
2346 | if (req->ctx->compat) | |
2347 | return io_compat_import(req, iov, needs_lock); | |
2348 | #endif | |
2349 | ||
2350 | return __io_iov_buffer_select(req, iov, needs_lock); | |
2351 | } | |
2352 | ||
cf6fd4bd | 2353 | static ssize_t io_import_iovec(int rw, struct io_kiocb *req, |
bcda7baa JA |
2354 | struct iovec **iovec, struct iov_iter *iter, |
2355 | bool needs_lock) | |
2b188cc1 | 2356 | { |
9adbd45d JA |
2357 | void __user *buf = u64_to_user_ptr(req->rw.addr); |
2358 | size_t sqe_len = req->rw.len; | |
4d954c25 | 2359 | ssize_t ret; |
edafccee JA |
2360 | u8 opcode; |
2361 | ||
d625c6ee | 2362 | opcode = req->opcode; |
7d009165 | 2363 | if (opcode == IORING_OP_READ_FIXED || opcode == IORING_OP_WRITE_FIXED) { |
edafccee | 2364 | *iovec = NULL; |
9adbd45d | 2365 | return io_import_fixed(req, rw, iter); |
edafccee | 2366 | } |
2b188cc1 | 2367 | |
bcda7baa JA |
2368 | /* buffer index only valid with fixed read/write, or buffer select */ |
2369 | if (req->rw.kiocb.private && !(req->flags & REQ_F_BUFFER_SELECT)) | |
9adbd45d JA |
2370 | return -EINVAL; |
2371 | ||
3a6820f2 | 2372 | if (opcode == IORING_OP_READ || opcode == IORING_OP_WRITE) { |
bcda7baa | 2373 | if (req->flags & REQ_F_BUFFER_SELECT) { |
4d954c25 JA |
2374 | buf = io_rw_buffer_select(req, &sqe_len, needs_lock); |
2375 | if (IS_ERR(buf)) { | |
bcda7baa | 2376 | *iovec = NULL; |
4d954c25 | 2377 | return PTR_ERR(buf); |
bcda7baa | 2378 | } |
3f9d6441 | 2379 | req->rw.len = sqe_len; |
bcda7baa JA |
2380 | } |
2381 | ||
3a6820f2 JA |
2382 | ret = import_single_range(rw, buf, sqe_len, *iovec, iter); |
2383 | *iovec = NULL; | |
3a901598 | 2384 | return ret < 0 ? ret : sqe_len; |
3a6820f2 JA |
2385 | } |
2386 | ||
f67676d1 JA |
2387 | if (req->io) { |
2388 | struct io_async_rw *iorw = &req->io->rw; | |
2389 | ||
2390 | *iovec = iorw->iov; | |
2391 | iov_iter_init(iter, rw, *iovec, iorw->nr_segs, iorw->size); | |
2392 | if (iorw->iov == iorw->fast_iov) | |
2393 | *iovec = NULL; | |
2394 | return iorw->size; | |
2395 | } | |
2396 | ||
4d954c25 JA |
2397 | if (req->flags & REQ_F_BUFFER_SELECT) { |
2398 | ret = io_iov_buffer_select(req, *iovec, needs_lock); | |
3f9d6441 JA |
2399 | if (!ret) { |
2400 | ret = (*iovec)->iov_len; | |
2401 | iov_iter_init(iter, rw, *iovec, 1, ret); | |
2402 | } | |
4d954c25 JA |
2403 | *iovec = NULL; |
2404 | return ret; | |
2405 | } | |
2406 | ||
2b188cc1 | 2407 | #ifdef CONFIG_COMPAT |
cf6fd4bd | 2408 | if (req->ctx->compat) |
2b188cc1 JA |
2409 | return compat_import_iovec(rw, buf, sqe_len, UIO_FASTIOV, |
2410 | iovec, iter); | |
2411 | #endif | |
2412 | ||
2413 | return import_iovec(rw, buf, sqe_len, UIO_FASTIOV, iovec, iter); | |
2414 | } | |
2415 | ||
31b51510 | 2416 | /* |
32960613 JA |
2417 | * For files that don't have ->read_iter() and ->write_iter(), handle them |
2418 | * by looping over ->read() or ->write() manually. | |
31b51510 | 2419 | */ |
32960613 JA |
2420 | static ssize_t loop_rw_iter(int rw, struct file *file, struct kiocb *kiocb, |
2421 | struct iov_iter *iter) | |
2422 | { | |
2423 | ssize_t ret = 0; | |
2424 | ||
2425 | /* | |
2426 | * Don't support polled IO through this interface, and we can't | |
2427 | * support non-blocking either. For the latter, this just causes | |
2428 | * the kiocb to be handled from an async context. | |
2429 | */ | |
2430 | if (kiocb->ki_flags & IOCB_HIPRI) | |
2431 | return -EOPNOTSUPP; | |
2432 | if (kiocb->ki_flags & IOCB_NOWAIT) | |
2433 | return -EAGAIN; | |
2434 | ||
2435 | while (iov_iter_count(iter)) { | |
311ae9e1 | 2436 | struct iovec iovec; |
32960613 JA |
2437 | ssize_t nr; |
2438 | ||
311ae9e1 PB |
2439 | if (!iov_iter_is_bvec(iter)) { |
2440 | iovec = iov_iter_iovec(iter); | |
2441 | } else { | |
2442 | /* fixed buffers import bvec */ | |
2443 | iovec.iov_base = kmap(iter->bvec->bv_page) | |
2444 | + iter->iov_offset; | |
2445 | iovec.iov_len = min(iter->count, | |
2446 | iter->bvec->bv_len - iter->iov_offset); | |
2447 | } | |
2448 | ||
32960613 JA |
2449 | if (rw == READ) { |
2450 | nr = file->f_op->read(file, iovec.iov_base, | |
2451 | iovec.iov_len, &kiocb->ki_pos); | |
2452 | } else { | |
2453 | nr = file->f_op->write(file, iovec.iov_base, | |
2454 | iovec.iov_len, &kiocb->ki_pos); | |
2455 | } | |
2456 | ||
311ae9e1 PB |
2457 | if (iov_iter_is_bvec(iter)) |
2458 | kunmap(iter->bvec->bv_page); | |
2459 | ||
32960613 JA |
2460 | if (nr < 0) { |
2461 | if (!ret) | |
2462 | ret = nr; | |
2463 | break; | |
2464 | } | |
2465 | ret += nr; | |
2466 | if (nr != iovec.iov_len) | |
2467 | break; | |
2468 | iov_iter_advance(iter, nr); | |
2469 | } | |
2470 | ||
2471 | return ret; | |
2472 | } | |
2473 | ||
b7bb4f7d | 2474 | static void io_req_map_rw(struct io_kiocb *req, ssize_t io_size, |
f67676d1 JA |
2475 | struct iovec *iovec, struct iovec *fast_iov, |
2476 | struct iov_iter *iter) | |
2477 | { | |
2478 | req->io->rw.nr_segs = iter->nr_segs; | |
2479 | req->io->rw.size = io_size; | |
2480 | req->io->rw.iov = iovec; | |
2481 | if (!req->io->rw.iov) { | |
2482 | req->io->rw.iov = req->io->rw.fast_iov; | |
45097dae XW |
2483 | if (req->io->rw.iov != fast_iov) |
2484 | memcpy(req->io->rw.iov, fast_iov, | |
2485 | sizeof(struct iovec) * iter->nr_segs); | |
99bc4c38 PB |
2486 | } else { |
2487 | req->flags |= REQ_F_NEED_CLEANUP; | |
f67676d1 JA |
2488 | } |
2489 | } | |
2490 | ||
3d9932a8 XW |
2491 | static inline int __io_alloc_async_ctx(struct io_kiocb *req) |
2492 | { | |
2493 | req->io = kmalloc(sizeof(*req->io), GFP_KERNEL); | |
2494 | return req->io == NULL; | |
2495 | } | |
2496 | ||
b7bb4f7d | 2497 | static int io_alloc_async_ctx(struct io_kiocb *req) |
f67676d1 | 2498 | { |
d3656344 JA |
2499 | if (!io_op_defs[req->opcode].async_ctx) |
2500 | return 0; | |
3d9932a8 XW |
2501 | |
2502 | return __io_alloc_async_ctx(req); | |
b7bb4f7d JA |
2503 | } |
2504 | ||
b7bb4f7d JA |
2505 | static int io_setup_async_rw(struct io_kiocb *req, ssize_t io_size, |
2506 | struct iovec *iovec, struct iovec *fast_iov, | |
2507 | struct iov_iter *iter) | |
2508 | { | |
980ad263 | 2509 | if (!io_op_defs[req->opcode].async_ctx) |
74566df3 | 2510 | return 0; |
5d204bcf | 2511 | if (!req->io) { |
3d9932a8 | 2512 | if (__io_alloc_async_ctx(req)) |
5d204bcf | 2513 | return -ENOMEM; |
b7bb4f7d | 2514 | |
5d204bcf JA |
2515 | io_req_map_rw(req, io_size, iovec, fast_iov, iter); |
2516 | } | |
b7bb4f7d | 2517 | return 0; |
f67676d1 JA |
2518 | } |
2519 | ||
3529d8c2 JA |
2520 | static int io_read_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe, |
2521 | bool force_nonblock) | |
f67676d1 | 2522 | { |
3529d8c2 JA |
2523 | struct io_async_ctx *io; |
2524 | struct iov_iter iter; | |
f67676d1 JA |
2525 | ssize_t ret; |
2526 | ||
3529d8c2 JA |
2527 | ret = io_prep_rw(req, sqe, force_nonblock); |
2528 | if (ret) | |
2529 | return ret; | |
f67676d1 | 2530 | |
3529d8c2 JA |
2531 | if (unlikely(!(req->file->f_mode & FMODE_READ))) |
2532 | return -EBADF; | |
f67676d1 | 2533 | |
5f798bea PB |
2534 | /* either don't need iovec imported or already have it */ |
2535 | if (!req->io || req->flags & REQ_F_NEED_CLEANUP) | |
3529d8c2 JA |
2536 | return 0; |
2537 | ||
2538 | io = req->io; | |
2539 | io->rw.iov = io->rw.fast_iov; | |
2540 | req->io = NULL; | |
bcda7baa | 2541 | ret = io_import_iovec(READ, req, &io->rw.iov, &iter, !force_nonblock); |
3529d8c2 JA |
2542 | req->io = io; |
2543 | if (ret < 0) | |
2544 | return ret; | |
2545 | ||
2546 | io_req_map_rw(req, ret, io->rw.iov, io->rw.fast_iov, &iter); | |
2547 | return 0; | |
f67676d1 JA |
2548 | } |
2549 | ||
014db007 | 2550 | static int io_read(struct io_kiocb *req, bool force_nonblock) |
2b188cc1 JA |
2551 | { |
2552 | struct iovec inline_vecs[UIO_FASTIOV], *iovec = inline_vecs; | |
9adbd45d | 2553 | struct kiocb *kiocb = &req->rw.kiocb; |
2b188cc1 | 2554 | struct iov_iter iter; |
31b51510 | 2555 | size_t iov_count; |
f67676d1 | 2556 | ssize_t io_size, ret; |
2b188cc1 | 2557 | |
bcda7baa | 2558 | ret = io_import_iovec(READ, req, &iovec, &iter, !force_nonblock); |
06b76d44 JA |
2559 | if (ret < 0) |
2560 | return ret; | |
2b188cc1 | 2561 | |
fd6c2e4c JA |
2562 | /* Ensure we clear previously set non-block flag */ |
2563 | if (!force_nonblock) | |
29de5f6a | 2564 | kiocb->ki_flags &= ~IOCB_NOWAIT; |
fd6c2e4c | 2565 | |
797f3f53 | 2566 | req->result = 0; |
f67676d1 | 2567 | io_size = ret; |
dea3b49c | 2568 | if (req->flags & REQ_F_LINK_HEAD) |
f67676d1 JA |
2569 | req->result = io_size; |
2570 | ||
2571 | /* | |
2572 | * If the file doesn't support async, mark it as REQ_F_MUST_PUNT so | |
2573 | * we know to async punt it even if it was opened O_NONBLOCK | |
2574 | */ | |
af197f50 | 2575 | if (force_nonblock && !io_file_supports_async(req->file, READ)) |
f67676d1 | 2576 | goto copy_iov; |
9e645e11 | 2577 | |
31b51510 | 2578 | iov_count = iov_iter_count(&iter); |
9adbd45d | 2579 | ret = rw_verify_area(READ, req->file, &kiocb->ki_pos, iov_count); |
2b188cc1 JA |
2580 | if (!ret) { |
2581 | ssize_t ret2; | |
2582 | ||
9adbd45d JA |
2583 | if (req->file->f_op->read_iter) |
2584 | ret2 = call_read_iter(req->file, kiocb, &iter); | |
32960613 | 2585 | else |
9adbd45d | 2586 | ret2 = loop_rw_iter(READ, req->file, kiocb, &iter); |
32960613 | 2587 | |
9d93a3f5 | 2588 | /* Catch -EAGAIN return for forced non-blocking submission */ |
f67676d1 | 2589 | if (!force_nonblock || ret2 != -EAGAIN) { |
014db007 | 2590 | kiocb_done(kiocb, ret2); |
f67676d1 JA |
2591 | } else { |
2592 | copy_iov: | |
b7bb4f7d | 2593 | ret = io_setup_async_rw(req, io_size, iovec, |
f67676d1 JA |
2594 | inline_vecs, &iter); |
2595 | if (ret) | |
2596 | goto out_free; | |
29de5f6a | 2597 | /* any defer here is final, must blocking retry */ |
490e8967 JA |
2598 | if (!(req->flags & REQ_F_NOWAIT) && |
2599 | !file_can_poll(req->file)) | |
29de5f6a | 2600 | req->flags |= REQ_F_MUST_PUNT; |
f67676d1 JA |
2601 | return -EAGAIN; |
2602 | } | |
2b188cc1 | 2603 | } |
f67676d1 | 2604 | out_free: |
1e95081c | 2605 | kfree(iovec); |
99bc4c38 | 2606 | req->flags &= ~REQ_F_NEED_CLEANUP; |
2b188cc1 JA |
2607 | return ret; |
2608 | } | |
2609 | ||
3529d8c2 JA |
2610 | static int io_write_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe, |
2611 | bool force_nonblock) | |
f67676d1 | 2612 | { |
3529d8c2 JA |
2613 | struct io_async_ctx *io; |
2614 | struct iov_iter iter; | |
f67676d1 JA |
2615 | ssize_t ret; |
2616 | ||
3529d8c2 JA |
2617 | ret = io_prep_rw(req, sqe, force_nonblock); |
2618 | if (ret) | |
2619 | return ret; | |
f67676d1 | 2620 | |
3529d8c2 JA |
2621 | if (unlikely(!(req->file->f_mode & FMODE_WRITE))) |
2622 | return -EBADF; | |
f67676d1 | 2623 | |
4ed734b0 JA |
2624 | req->fsize = rlimit(RLIMIT_FSIZE); |
2625 | ||
5f798bea PB |
2626 | /* either don't need iovec imported or already have it */ |
2627 | if (!req->io || req->flags & REQ_F_NEED_CLEANUP) | |
3529d8c2 JA |
2628 | return 0; |
2629 | ||
2630 | io = req->io; | |
2631 | io->rw.iov = io->rw.fast_iov; | |
2632 | req->io = NULL; | |
bcda7baa | 2633 | ret = io_import_iovec(WRITE, req, &io->rw.iov, &iter, !force_nonblock); |
3529d8c2 JA |
2634 | req->io = io; |
2635 | if (ret < 0) | |
2636 | return ret; | |
2637 | ||
2638 | io_req_map_rw(req, ret, io->rw.iov, io->rw.fast_iov, &iter); | |
2639 | return 0; | |
f67676d1 JA |
2640 | } |
2641 | ||
014db007 | 2642 | static int io_write(struct io_kiocb *req, bool force_nonblock) |
2b188cc1 JA |
2643 | { |
2644 | struct iovec inline_vecs[UIO_FASTIOV], *iovec = inline_vecs; | |
9adbd45d | 2645 | struct kiocb *kiocb = &req->rw.kiocb; |
2b188cc1 | 2646 | struct iov_iter iter; |
31b51510 | 2647 | size_t iov_count; |
f67676d1 | 2648 | ssize_t ret, io_size; |
2b188cc1 | 2649 | |
bcda7baa | 2650 | ret = io_import_iovec(WRITE, req, &iovec, &iter, !force_nonblock); |
06b76d44 JA |
2651 | if (ret < 0) |
2652 | return ret; | |
2b188cc1 | 2653 | |
fd6c2e4c JA |
2654 | /* Ensure we clear previously set non-block flag */ |
2655 | if (!force_nonblock) | |
9adbd45d | 2656 | req->rw.kiocb.ki_flags &= ~IOCB_NOWAIT; |
fd6c2e4c | 2657 | |
797f3f53 | 2658 | req->result = 0; |
f67676d1 | 2659 | io_size = ret; |
dea3b49c | 2660 | if (req->flags & REQ_F_LINK_HEAD) |
f67676d1 | 2661 | req->result = io_size; |
9e645e11 | 2662 | |
f67676d1 JA |
2663 | /* |
2664 | * If the file doesn't support async, mark it as REQ_F_MUST_PUNT so | |
2665 | * we know to async punt it even if it was opened O_NONBLOCK | |
2666 | */ | |
af197f50 | 2667 | if (force_nonblock && !io_file_supports_async(req->file, WRITE)) |
f67676d1 | 2668 | goto copy_iov; |
31b51510 | 2669 | |
10d59345 JA |
2670 | /* file path doesn't support NOWAIT for non-direct_IO */ |
2671 | if (force_nonblock && !(kiocb->ki_flags & IOCB_DIRECT) && | |
2672 | (req->flags & REQ_F_ISREG)) | |
f67676d1 | 2673 | goto copy_iov; |
31b51510 | 2674 | |
f67676d1 | 2675 | iov_count = iov_iter_count(&iter); |
9adbd45d | 2676 | ret = rw_verify_area(WRITE, req->file, &kiocb->ki_pos, iov_count); |
2b188cc1 | 2677 | if (!ret) { |
9bf7933f RP |
2678 | ssize_t ret2; |
2679 | ||
2b188cc1 JA |
2680 | /* |
2681 | * Open-code file_start_write here to grab freeze protection, | |
2682 | * which will be released by another thread in | |
2683 | * io_complete_rw(). Fool lockdep by telling it the lock got | |
2684 | * released so that it doesn't complain about the held lock when | |
2685 | * we return to userspace. | |
2686 | */ | |
491381ce | 2687 | if (req->flags & REQ_F_ISREG) { |
9adbd45d | 2688 | __sb_start_write(file_inode(req->file)->i_sb, |
2b188cc1 | 2689 | SB_FREEZE_WRITE, true); |
9adbd45d | 2690 | __sb_writers_release(file_inode(req->file)->i_sb, |
2b188cc1 JA |
2691 | SB_FREEZE_WRITE); |
2692 | } | |
2693 | kiocb->ki_flags |= IOCB_WRITE; | |
9bf7933f | 2694 | |
4ed734b0 JA |
2695 | if (!force_nonblock) |
2696 | current->signal->rlim[RLIMIT_FSIZE].rlim_cur = req->fsize; | |
2697 | ||
9adbd45d JA |
2698 | if (req->file->f_op->write_iter) |
2699 | ret2 = call_write_iter(req->file, kiocb, &iter); | |
32960613 | 2700 | else |
9adbd45d | 2701 | ret2 = loop_rw_iter(WRITE, req->file, kiocb, &iter); |
4ed734b0 JA |
2702 | |
2703 | if (!force_nonblock) | |
2704 | current->signal->rlim[RLIMIT_FSIZE].rlim_cur = RLIM_INFINITY; | |
2705 | ||
faac996c | 2706 | /* |
bff6035d | 2707 | * Raw bdev writes will return -EOPNOTSUPP for IOCB_NOWAIT. Just |
faac996c JA |
2708 | * retry them without IOCB_NOWAIT. |
2709 | */ | |
2710 | if (ret2 == -EOPNOTSUPP && (kiocb->ki_flags & IOCB_NOWAIT)) | |
2711 | ret2 = -EAGAIN; | |
f67676d1 | 2712 | if (!force_nonblock || ret2 != -EAGAIN) { |
014db007 | 2713 | kiocb_done(kiocb, ret2); |
f67676d1 JA |
2714 | } else { |
2715 | copy_iov: | |
b7bb4f7d | 2716 | ret = io_setup_async_rw(req, io_size, iovec, |
f67676d1 JA |
2717 | inline_vecs, &iter); |
2718 | if (ret) | |
2719 | goto out_free; | |
29de5f6a | 2720 | /* any defer here is final, must blocking retry */ |
490e8967 JA |
2721 | if (!file_can_poll(req->file)) |
2722 | req->flags |= REQ_F_MUST_PUNT; | |
f67676d1 JA |
2723 | return -EAGAIN; |
2724 | } | |
2b188cc1 | 2725 | } |
31b51510 | 2726 | out_free: |
99bc4c38 | 2727 | req->flags &= ~REQ_F_NEED_CLEANUP; |
1e95081c | 2728 | kfree(iovec); |
2b188cc1 JA |
2729 | return ret; |
2730 | } | |
2731 | ||
7d67af2c PB |
2732 | static int io_splice_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
2733 | { | |
2734 | struct io_splice* sp = &req->splice; | |
2735 | unsigned int valid_flags = SPLICE_F_FD_IN_FIXED | SPLICE_F_ALL; | |
2736 | int ret; | |
2737 | ||
2738 | if (req->flags & REQ_F_NEED_CLEANUP) | |
2739 | return 0; | |
2740 | ||
2741 | sp->file_in = NULL; | |
2742 | sp->off_in = READ_ONCE(sqe->splice_off_in); | |
2743 | sp->off_out = READ_ONCE(sqe->off); | |
2744 | sp->len = READ_ONCE(sqe->len); | |
2745 | sp->flags = READ_ONCE(sqe->splice_flags); | |
2746 | ||
2747 | if (unlikely(sp->flags & ~valid_flags)) | |
2748 | return -EINVAL; | |
2749 | ||
2750 | ret = io_file_get(NULL, req, READ_ONCE(sqe->splice_fd_in), &sp->file_in, | |
2751 | (sp->flags & SPLICE_F_FD_IN_FIXED)); | |
2752 | if (ret) | |
2753 | return ret; | |
2754 | req->flags |= REQ_F_NEED_CLEANUP; | |
2755 | ||
2756 | if (!S_ISREG(file_inode(sp->file_in)->i_mode)) | |
2757 | req->work.flags |= IO_WQ_WORK_UNBOUND; | |
2758 | ||
2759 | return 0; | |
2760 | } | |
2761 | ||
014db007 | 2762 | static int io_splice(struct io_kiocb *req, bool force_nonblock) |
7d67af2c PB |
2763 | { |
2764 | struct io_splice *sp = &req->splice; | |
2765 | struct file *in = sp->file_in; | |
2766 | struct file *out = sp->file_out; | |
2767 | unsigned int flags = sp->flags & ~SPLICE_F_FD_IN_FIXED; | |
2768 | loff_t *poff_in, *poff_out; | |
2769 | long ret; | |
2770 | ||
2fb3e822 PB |
2771 | if (force_nonblock) |
2772 | return -EAGAIN; | |
7d67af2c PB |
2773 | |
2774 | poff_in = (sp->off_in == -1) ? NULL : &sp->off_in; | |
2775 | poff_out = (sp->off_out == -1) ? NULL : &sp->off_out; | |
2776 | ret = do_splice(in, poff_in, out, poff_out, sp->len, flags); | |
2777 | if (force_nonblock && ret == -EAGAIN) | |
2778 | return -EAGAIN; | |
2779 | ||
2780 | io_put_file(req, in, (sp->flags & SPLICE_F_FD_IN_FIXED)); | |
2781 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
2782 | ||
2783 | io_cqring_add_event(req, ret); | |
2784 | if (ret != sp->len) | |
2785 | req_set_fail_links(req); | |
014db007 | 2786 | io_put_req(req); |
7d67af2c PB |
2787 | return 0; |
2788 | } | |
2789 | ||
2b188cc1 JA |
2790 | /* |
2791 | * IORING_OP_NOP just posts a completion event, nothing else. | |
2792 | */ | |
78e19bbe | 2793 | static int io_nop(struct io_kiocb *req) |
2b188cc1 JA |
2794 | { |
2795 | struct io_ring_ctx *ctx = req->ctx; | |
2b188cc1 | 2796 | |
def596e9 JA |
2797 | if (unlikely(ctx->flags & IORING_SETUP_IOPOLL)) |
2798 | return -EINVAL; | |
2799 | ||
78e19bbe | 2800 | io_cqring_add_event(req, 0); |
e65ef56d | 2801 | io_put_req(req); |
2b188cc1 JA |
2802 | return 0; |
2803 | } | |
2804 | ||
3529d8c2 | 2805 | static int io_prep_fsync(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
c992fe29 | 2806 | { |
6b06314c | 2807 | struct io_ring_ctx *ctx = req->ctx; |
c992fe29 | 2808 | |
09bb8394 JA |
2809 | if (!req->file) |
2810 | return -EBADF; | |
c992fe29 | 2811 | |
6b06314c | 2812 | if (unlikely(ctx->flags & IORING_SETUP_IOPOLL)) |
def596e9 | 2813 | return -EINVAL; |
edafccee | 2814 | if (unlikely(sqe->addr || sqe->ioprio || sqe->buf_index)) |
c992fe29 CH |
2815 | return -EINVAL; |
2816 | ||
8ed8d3c3 JA |
2817 | req->sync.flags = READ_ONCE(sqe->fsync_flags); |
2818 | if (unlikely(req->sync.flags & ~IORING_FSYNC_DATASYNC)) | |
2819 | return -EINVAL; | |
2820 | ||
2821 | req->sync.off = READ_ONCE(sqe->off); | |
2822 | req->sync.len = READ_ONCE(sqe->len); | |
c992fe29 CH |
2823 | return 0; |
2824 | } | |
2825 | ||
8ed8d3c3 JA |
2826 | static bool io_req_cancelled(struct io_kiocb *req) |
2827 | { | |
2828 | if (req->work.flags & IO_WQ_WORK_CANCEL) { | |
2829 | req_set_fail_links(req); | |
2830 | io_cqring_add_event(req, -ECANCELED); | |
2831 | io_put_req(req); | |
2832 | return true; | |
2833 | } | |
2834 | ||
2835 | return false; | |
2836 | } | |
2837 | ||
014db007 | 2838 | static void __io_fsync(struct io_kiocb *req) |
8ed8d3c3 | 2839 | { |
8ed8d3c3 | 2840 | loff_t end = req->sync.off + req->sync.len; |
8ed8d3c3 JA |
2841 | int ret; |
2842 | ||
9adbd45d | 2843 | ret = vfs_fsync_range(req->file, req->sync.off, |
8ed8d3c3 JA |
2844 | end > 0 ? end : LLONG_MAX, |
2845 | req->sync.flags & IORING_FSYNC_DATASYNC); | |
2846 | if (ret < 0) | |
2847 | req_set_fail_links(req); | |
2848 | io_cqring_add_event(req, ret); | |
014db007 | 2849 | io_put_req(req); |
8ed8d3c3 JA |
2850 | } |
2851 | ||
5ea62161 | 2852 | static void io_fsync_finish(struct io_wq_work **workptr) |
c992fe29 | 2853 | { |
5ea62161 | 2854 | struct io_kiocb *req = container_of(*workptr, struct io_kiocb, work); |
5ea62161 PB |
2855 | |
2856 | if (io_req_cancelled(req)) | |
2857 | return; | |
014db007 | 2858 | __io_fsync(req); |
e9fd9396 | 2859 | io_steal_work(req, workptr); |
8ed8d3c3 | 2860 | } |
c992fe29 | 2861 | |
014db007 | 2862 | static int io_fsync(struct io_kiocb *req, bool force_nonblock) |
c992fe29 | 2863 | { |
c992fe29 | 2864 | /* fsync always requires a blocking context */ |
8ed8d3c3 | 2865 | if (force_nonblock) { |
8ed8d3c3 | 2866 | req->work.func = io_fsync_finish; |
c992fe29 | 2867 | return -EAGAIN; |
8ed8d3c3 | 2868 | } |
014db007 | 2869 | __io_fsync(req); |
c992fe29 CH |
2870 | return 0; |
2871 | } | |
2872 | ||
014db007 | 2873 | static void __io_fallocate(struct io_kiocb *req) |
8ed8d3c3 | 2874 | { |
8ed8d3c3 JA |
2875 | int ret; |
2876 | ||
4ed734b0 | 2877 | current->signal->rlim[RLIMIT_FSIZE].rlim_cur = req->fsize; |
d63d1b5e JA |
2878 | ret = vfs_fallocate(req->file, req->sync.mode, req->sync.off, |
2879 | req->sync.len); | |
4ed734b0 | 2880 | current->signal->rlim[RLIMIT_FSIZE].rlim_cur = RLIM_INFINITY; |
8ed8d3c3 JA |
2881 | if (ret < 0) |
2882 | req_set_fail_links(req); | |
2883 | io_cqring_add_event(req, ret); | |
014db007 | 2884 | io_put_req(req); |
5ea62161 PB |
2885 | } |
2886 | ||
2887 | static void io_fallocate_finish(struct io_wq_work **workptr) | |
2888 | { | |
2889 | struct io_kiocb *req = container_of(*workptr, struct io_kiocb, work); | |
5ea62161 | 2890 | |
594506fe PB |
2891 | if (io_req_cancelled(req)) |
2892 | return; | |
014db007 | 2893 | __io_fallocate(req); |
e9fd9396 | 2894 | io_steal_work(req, workptr); |
5d17b4a4 JA |
2895 | } |
2896 | ||
d63d1b5e JA |
2897 | static int io_fallocate_prep(struct io_kiocb *req, |
2898 | const struct io_uring_sqe *sqe) | |
2899 | { | |
2900 | if (sqe->ioprio || sqe->buf_index || sqe->rw_flags) | |
2901 | return -EINVAL; | |
2902 | ||
2903 | req->sync.off = READ_ONCE(sqe->off); | |
2904 | req->sync.len = READ_ONCE(sqe->addr); | |
2905 | req->sync.mode = READ_ONCE(sqe->len); | |
4ed734b0 | 2906 | req->fsize = rlimit(RLIMIT_FSIZE); |
d63d1b5e JA |
2907 | return 0; |
2908 | } | |
2909 | ||
014db007 | 2910 | static int io_fallocate(struct io_kiocb *req, bool force_nonblock) |
5d17b4a4 | 2911 | { |
d63d1b5e | 2912 | /* fallocate always requiring blocking context */ |
8ed8d3c3 | 2913 | if (force_nonblock) { |
d63d1b5e | 2914 | req->work.func = io_fallocate_finish; |
5d17b4a4 | 2915 | return -EAGAIN; |
8ed8d3c3 | 2916 | } |
5d17b4a4 | 2917 | |
014db007 | 2918 | __io_fallocate(req); |
5d17b4a4 JA |
2919 | return 0; |
2920 | } | |
2921 | ||
15b71abe | 2922 | static int io_openat_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
b7bb4f7d | 2923 | { |
f8748881 | 2924 | const char __user *fname; |
15b71abe | 2925 | int ret; |
b7bb4f7d | 2926 | |
15b71abe JA |
2927 | if (sqe->ioprio || sqe->buf_index) |
2928 | return -EINVAL; | |
9c280f90 | 2929 | if (req->flags & REQ_F_FIXED_FILE) |
cf3040ca | 2930 | return -EBADF; |
0bdbdd08 PB |
2931 | if (req->flags & REQ_F_NEED_CLEANUP) |
2932 | return 0; | |
03b1230c | 2933 | |
15b71abe | 2934 | req->open.dfd = READ_ONCE(sqe->fd); |
c12cedf2 | 2935 | req->open.how.mode = READ_ONCE(sqe->len); |
f8748881 | 2936 | fname = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
c12cedf2 | 2937 | req->open.how.flags = READ_ONCE(sqe->open_flags); |
08a1d26e JA |
2938 | if (force_o_largefile()) |
2939 | req->open.how.flags |= O_LARGEFILE; | |
3529d8c2 | 2940 | |
f8748881 | 2941 | req->open.filename = getname(fname); |
15b71abe JA |
2942 | if (IS_ERR(req->open.filename)) { |
2943 | ret = PTR_ERR(req->open.filename); | |
2944 | req->open.filename = NULL; | |
2945 | return ret; | |
2946 | } | |
3529d8c2 | 2947 | |
4022e7af | 2948 | req->open.nofile = rlimit(RLIMIT_NOFILE); |
8fef80bf | 2949 | req->flags |= REQ_F_NEED_CLEANUP; |
15b71abe | 2950 | return 0; |
03b1230c JA |
2951 | } |
2952 | ||
cebdb986 | 2953 | static int io_openat2_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
aa1fa28f | 2954 | { |
cebdb986 JA |
2955 | struct open_how __user *how; |
2956 | const char __user *fname; | |
2957 | size_t len; | |
0fa03c62 JA |
2958 | int ret; |
2959 | ||
cebdb986 | 2960 | if (sqe->ioprio || sqe->buf_index) |
0fa03c62 | 2961 | return -EINVAL; |
9c280f90 | 2962 | if (req->flags & REQ_F_FIXED_FILE) |
cf3040ca | 2963 | return -EBADF; |
0bdbdd08 PB |
2964 | if (req->flags & REQ_F_NEED_CLEANUP) |
2965 | return 0; | |
0fa03c62 | 2966 | |
cebdb986 JA |
2967 | req->open.dfd = READ_ONCE(sqe->fd); |
2968 | fname = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
2969 | how = u64_to_user_ptr(READ_ONCE(sqe->addr2)); | |
2970 | len = READ_ONCE(sqe->len); | |
0fa03c62 | 2971 | |
cebdb986 JA |
2972 | if (len < OPEN_HOW_SIZE_VER0) |
2973 | return -EINVAL; | |
3529d8c2 | 2974 | |
cebdb986 JA |
2975 | ret = copy_struct_from_user(&req->open.how, sizeof(req->open.how), how, |
2976 | len); | |
2977 | if (ret) | |
2978 | return ret; | |
3529d8c2 | 2979 | |
cebdb986 JA |
2980 | if (!(req->open.how.flags & O_PATH) && force_o_largefile()) |
2981 | req->open.how.flags |= O_LARGEFILE; | |
0fa03c62 | 2982 | |
cebdb986 JA |
2983 | req->open.filename = getname(fname); |
2984 | if (IS_ERR(req->open.filename)) { | |
2985 | ret = PTR_ERR(req->open.filename); | |
2986 | req->open.filename = NULL; | |
2987 | return ret; | |
2988 | } | |
2989 | ||
4022e7af | 2990 | req->open.nofile = rlimit(RLIMIT_NOFILE); |
8fef80bf | 2991 | req->flags |= REQ_F_NEED_CLEANUP; |
cebdb986 JA |
2992 | return 0; |
2993 | } | |
2994 | ||
014db007 | 2995 | static int io_openat2(struct io_kiocb *req, bool force_nonblock) |
15b71abe JA |
2996 | { |
2997 | struct open_flags op; | |
15b71abe JA |
2998 | struct file *file; |
2999 | int ret; | |
3000 | ||
f86cd20c | 3001 | if (force_nonblock) |
15b71abe | 3002 | return -EAGAIN; |
15b71abe | 3003 | |
cebdb986 | 3004 | ret = build_open_flags(&req->open.how, &op); |
15b71abe JA |
3005 | if (ret) |
3006 | goto err; | |
3007 | ||
4022e7af | 3008 | ret = __get_unused_fd_flags(req->open.how.flags, req->open.nofile); |
15b71abe JA |
3009 | if (ret < 0) |
3010 | goto err; | |
3011 | ||
3012 | file = do_filp_open(req->open.dfd, req->open.filename, &op); | |
3013 | if (IS_ERR(file)) { | |
3014 | put_unused_fd(ret); | |
3015 | ret = PTR_ERR(file); | |
3016 | } else { | |
3017 | fsnotify_open(file); | |
3018 | fd_install(ret, file); | |
3019 | } | |
3020 | err: | |
3021 | putname(req->open.filename); | |
8fef80bf | 3022 | req->flags &= ~REQ_F_NEED_CLEANUP; |
15b71abe JA |
3023 | if (ret < 0) |
3024 | req_set_fail_links(req); | |
3025 | io_cqring_add_event(req, ret); | |
014db007 | 3026 | io_put_req(req); |
15b71abe JA |
3027 | return 0; |
3028 | } | |
3029 | ||
014db007 | 3030 | static int io_openat(struct io_kiocb *req, bool force_nonblock) |
cebdb986 JA |
3031 | { |
3032 | req->open.how = build_open_how(req->open.how.flags, req->open.how.mode); | |
014db007 | 3033 | return io_openat2(req, force_nonblock); |
cebdb986 JA |
3034 | } |
3035 | ||
067524e9 JA |
3036 | static int io_remove_buffers_prep(struct io_kiocb *req, |
3037 | const struct io_uring_sqe *sqe) | |
3038 | { | |
3039 | struct io_provide_buf *p = &req->pbuf; | |
3040 | u64 tmp; | |
3041 | ||
3042 | if (sqe->ioprio || sqe->rw_flags || sqe->addr || sqe->len || sqe->off) | |
3043 | return -EINVAL; | |
3044 | ||
3045 | tmp = READ_ONCE(sqe->fd); | |
3046 | if (!tmp || tmp > USHRT_MAX) | |
3047 | return -EINVAL; | |
3048 | ||
3049 | memset(p, 0, sizeof(*p)); | |
3050 | p->nbufs = tmp; | |
3051 | p->bgid = READ_ONCE(sqe->buf_group); | |
3052 | return 0; | |
3053 | } | |
3054 | ||
3055 | static int __io_remove_buffers(struct io_ring_ctx *ctx, struct io_buffer *buf, | |
3056 | int bgid, unsigned nbufs) | |
3057 | { | |
3058 | unsigned i = 0; | |
3059 | ||
3060 | /* shouldn't happen */ | |
3061 | if (!nbufs) | |
3062 | return 0; | |
3063 | ||
3064 | /* the head kbuf is the list itself */ | |
3065 | while (!list_empty(&buf->list)) { | |
3066 | struct io_buffer *nxt; | |
3067 | ||
3068 | nxt = list_first_entry(&buf->list, struct io_buffer, list); | |
3069 | list_del(&nxt->list); | |
3070 | kfree(nxt); | |
3071 | if (++i == nbufs) | |
3072 | return i; | |
3073 | } | |
3074 | i++; | |
3075 | kfree(buf); | |
3076 | idr_remove(&ctx->io_buffer_idr, bgid); | |
3077 | ||
3078 | return i; | |
3079 | } | |
3080 | ||
3081 | static int io_remove_buffers(struct io_kiocb *req, bool force_nonblock) | |
3082 | { | |
3083 | struct io_provide_buf *p = &req->pbuf; | |
3084 | struct io_ring_ctx *ctx = req->ctx; | |
3085 | struct io_buffer *head; | |
3086 | int ret = 0; | |
3087 | ||
3088 | io_ring_submit_lock(ctx, !force_nonblock); | |
3089 | ||
3090 | lockdep_assert_held(&ctx->uring_lock); | |
3091 | ||
3092 | ret = -ENOENT; | |
3093 | head = idr_find(&ctx->io_buffer_idr, p->bgid); | |
3094 | if (head) | |
3095 | ret = __io_remove_buffers(ctx, head, p->bgid, p->nbufs); | |
3096 | ||
3097 | io_ring_submit_lock(ctx, !force_nonblock); | |
3098 | if (ret < 0) | |
3099 | req_set_fail_links(req); | |
3100 | io_cqring_add_event(req, ret); | |
3101 | io_put_req(req); | |
3102 | return 0; | |
3103 | } | |
3104 | ||
ddf0322d JA |
3105 | static int io_provide_buffers_prep(struct io_kiocb *req, |
3106 | const struct io_uring_sqe *sqe) | |
3107 | { | |
3108 | struct io_provide_buf *p = &req->pbuf; | |
3109 | u64 tmp; | |
3110 | ||
3111 | if (sqe->ioprio || sqe->rw_flags) | |
3112 | return -EINVAL; | |
3113 | ||
3114 | tmp = READ_ONCE(sqe->fd); | |
3115 | if (!tmp || tmp > USHRT_MAX) | |
3116 | return -E2BIG; | |
3117 | p->nbufs = tmp; | |
3118 | p->addr = READ_ONCE(sqe->addr); | |
3119 | p->len = READ_ONCE(sqe->len); | |
3120 | ||
3121 | if (!access_ok(u64_to_user_ptr(p->addr), p->len)) | |
3122 | return -EFAULT; | |
3123 | ||
3124 | p->bgid = READ_ONCE(sqe->buf_group); | |
3125 | tmp = READ_ONCE(sqe->off); | |
3126 | if (tmp > USHRT_MAX) | |
3127 | return -E2BIG; | |
3128 | p->bid = tmp; | |
3129 | return 0; | |
3130 | } | |
3131 | ||
3132 | static int io_add_buffers(struct io_provide_buf *pbuf, struct io_buffer **head) | |
3133 | { | |
3134 | struct io_buffer *buf; | |
3135 | u64 addr = pbuf->addr; | |
3136 | int i, bid = pbuf->bid; | |
3137 | ||
3138 | for (i = 0; i < pbuf->nbufs; i++) { | |
3139 | buf = kmalloc(sizeof(*buf), GFP_KERNEL); | |
3140 | if (!buf) | |
3141 | break; | |
3142 | ||
3143 | buf->addr = addr; | |
3144 | buf->len = pbuf->len; | |
3145 | buf->bid = bid; | |
3146 | addr += pbuf->len; | |
3147 | bid++; | |
3148 | if (!*head) { | |
3149 | INIT_LIST_HEAD(&buf->list); | |
3150 | *head = buf; | |
3151 | } else { | |
3152 | list_add_tail(&buf->list, &(*head)->list); | |
3153 | } | |
3154 | } | |
3155 | ||
3156 | return i ? i : -ENOMEM; | |
3157 | } | |
3158 | ||
ddf0322d JA |
3159 | static int io_provide_buffers(struct io_kiocb *req, bool force_nonblock) |
3160 | { | |
3161 | struct io_provide_buf *p = &req->pbuf; | |
3162 | struct io_ring_ctx *ctx = req->ctx; | |
3163 | struct io_buffer *head, *list; | |
3164 | int ret = 0; | |
3165 | ||
3166 | io_ring_submit_lock(ctx, !force_nonblock); | |
3167 | ||
3168 | lockdep_assert_held(&ctx->uring_lock); | |
3169 | ||
3170 | list = head = idr_find(&ctx->io_buffer_idr, p->bgid); | |
3171 | ||
3172 | ret = io_add_buffers(p, &head); | |
3173 | if (ret < 0) | |
3174 | goto out; | |
3175 | ||
3176 | if (!list) { | |
3177 | ret = idr_alloc(&ctx->io_buffer_idr, head, p->bgid, p->bgid + 1, | |
3178 | GFP_KERNEL); | |
3179 | if (ret < 0) { | |
067524e9 | 3180 | __io_remove_buffers(ctx, head, p->bgid, -1U); |
ddf0322d JA |
3181 | goto out; |
3182 | } | |
3183 | } | |
3184 | out: | |
3185 | io_ring_submit_unlock(ctx, !force_nonblock); | |
3186 | if (ret < 0) | |
3187 | req_set_fail_links(req); | |
3188 | io_cqring_add_event(req, ret); | |
3189 | io_put_req(req); | |
3190 | return 0; | |
cebdb986 JA |
3191 | } |
3192 | ||
3e4827b0 JA |
3193 | static int io_epoll_ctl_prep(struct io_kiocb *req, |
3194 | const struct io_uring_sqe *sqe) | |
3195 | { | |
3196 | #if defined(CONFIG_EPOLL) | |
3197 | if (sqe->ioprio || sqe->buf_index) | |
3198 | return -EINVAL; | |
3199 | ||
3200 | req->epoll.epfd = READ_ONCE(sqe->fd); | |
3201 | req->epoll.op = READ_ONCE(sqe->len); | |
3202 | req->epoll.fd = READ_ONCE(sqe->off); | |
3203 | ||
3204 | if (ep_op_has_event(req->epoll.op)) { | |
3205 | struct epoll_event __user *ev; | |
3206 | ||
3207 | ev = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
3208 | if (copy_from_user(&req->epoll.event, ev, sizeof(*ev))) | |
3209 | return -EFAULT; | |
3210 | } | |
3211 | ||
3212 | return 0; | |
3213 | #else | |
3214 | return -EOPNOTSUPP; | |
3215 | #endif | |
3216 | } | |
3217 | ||
014db007 | 3218 | static int io_epoll_ctl(struct io_kiocb *req, bool force_nonblock) |
3e4827b0 JA |
3219 | { |
3220 | #if defined(CONFIG_EPOLL) | |
3221 | struct io_epoll *ie = &req->epoll; | |
3222 | int ret; | |
3223 | ||
3224 | ret = do_epoll_ctl(ie->epfd, ie->op, ie->fd, &ie->event, force_nonblock); | |
3225 | if (force_nonblock && ret == -EAGAIN) | |
3226 | return -EAGAIN; | |
3227 | ||
3228 | if (ret < 0) | |
3229 | req_set_fail_links(req); | |
3230 | io_cqring_add_event(req, ret); | |
014db007 | 3231 | io_put_req(req); |
3e4827b0 JA |
3232 | return 0; |
3233 | #else | |
3234 | return -EOPNOTSUPP; | |
3235 | #endif | |
3236 | } | |
3237 | ||
c1ca757b JA |
3238 | static int io_madvise_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
3239 | { | |
3240 | #if defined(CONFIG_ADVISE_SYSCALLS) && defined(CONFIG_MMU) | |
3241 | if (sqe->ioprio || sqe->buf_index || sqe->off) | |
3242 | return -EINVAL; | |
3243 | ||
3244 | req->madvise.addr = READ_ONCE(sqe->addr); | |
3245 | req->madvise.len = READ_ONCE(sqe->len); | |
3246 | req->madvise.advice = READ_ONCE(sqe->fadvise_advice); | |
3247 | return 0; | |
3248 | #else | |
3249 | return -EOPNOTSUPP; | |
3250 | #endif | |
3251 | } | |
3252 | ||
014db007 | 3253 | static int io_madvise(struct io_kiocb *req, bool force_nonblock) |
c1ca757b JA |
3254 | { |
3255 | #if defined(CONFIG_ADVISE_SYSCALLS) && defined(CONFIG_MMU) | |
3256 | struct io_madvise *ma = &req->madvise; | |
3257 | int ret; | |
3258 | ||
3259 | if (force_nonblock) | |
3260 | return -EAGAIN; | |
3261 | ||
3262 | ret = do_madvise(ma->addr, ma->len, ma->advice); | |
3263 | if (ret < 0) | |
3264 | req_set_fail_links(req); | |
3265 | io_cqring_add_event(req, ret); | |
014db007 | 3266 | io_put_req(req); |
c1ca757b JA |
3267 | return 0; |
3268 | #else | |
3269 | return -EOPNOTSUPP; | |
3270 | #endif | |
3271 | } | |
3272 | ||
4840e418 JA |
3273 | static int io_fadvise_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
3274 | { | |
3275 | if (sqe->ioprio || sqe->buf_index || sqe->addr) | |
3276 | return -EINVAL; | |
3277 | ||
3278 | req->fadvise.offset = READ_ONCE(sqe->off); | |
3279 | req->fadvise.len = READ_ONCE(sqe->len); | |
3280 | req->fadvise.advice = READ_ONCE(sqe->fadvise_advice); | |
3281 | return 0; | |
3282 | } | |
3283 | ||
014db007 | 3284 | static int io_fadvise(struct io_kiocb *req, bool force_nonblock) |
4840e418 JA |
3285 | { |
3286 | struct io_fadvise *fa = &req->fadvise; | |
3287 | int ret; | |
3288 | ||
3e69426d JA |
3289 | if (force_nonblock) { |
3290 | switch (fa->advice) { | |
3291 | case POSIX_FADV_NORMAL: | |
3292 | case POSIX_FADV_RANDOM: | |
3293 | case POSIX_FADV_SEQUENTIAL: | |
3294 | break; | |
3295 | default: | |
3296 | return -EAGAIN; | |
3297 | } | |
3298 | } | |
4840e418 JA |
3299 | |
3300 | ret = vfs_fadvise(req->file, fa->offset, fa->len, fa->advice); | |
3301 | if (ret < 0) | |
3302 | req_set_fail_links(req); | |
3303 | io_cqring_add_event(req, ret); | |
014db007 | 3304 | io_put_req(req); |
4840e418 JA |
3305 | return 0; |
3306 | } | |
3307 | ||
eddc7ef5 JA |
3308 | static int io_statx_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
3309 | { | |
f8748881 | 3310 | const char __user *fname; |
eddc7ef5 JA |
3311 | unsigned lookup_flags; |
3312 | int ret; | |
3313 | ||
3314 | if (sqe->ioprio || sqe->buf_index) | |
3315 | return -EINVAL; | |
9c280f90 | 3316 | if (req->flags & REQ_F_FIXED_FILE) |
cf3040ca | 3317 | return -EBADF; |
0bdbdd08 PB |
3318 | if (req->flags & REQ_F_NEED_CLEANUP) |
3319 | return 0; | |
eddc7ef5 JA |
3320 | |
3321 | req->open.dfd = READ_ONCE(sqe->fd); | |
3322 | req->open.mask = READ_ONCE(sqe->len); | |
f8748881 | 3323 | fname = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
eddc7ef5 | 3324 | req->open.buffer = u64_to_user_ptr(READ_ONCE(sqe->addr2)); |
c12cedf2 | 3325 | req->open.how.flags = READ_ONCE(sqe->statx_flags); |
eddc7ef5 | 3326 | |
c12cedf2 | 3327 | if (vfs_stat_set_lookup_flags(&lookup_flags, req->open.how.flags)) |
eddc7ef5 JA |
3328 | return -EINVAL; |
3329 | ||
f8748881 | 3330 | req->open.filename = getname_flags(fname, lookup_flags, NULL); |
eddc7ef5 JA |
3331 | if (IS_ERR(req->open.filename)) { |
3332 | ret = PTR_ERR(req->open.filename); | |
3333 | req->open.filename = NULL; | |
3334 | return ret; | |
3335 | } | |
3336 | ||
8fef80bf | 3337 | req->flags |= REQ_F_NEED_CLEANUP; |
eddc7ef5 JA |
3338 | return 0; |
3339 | } | |
3340 | ||
014db007 | 3341 | static int io_statx(struct io_kiocb *req, bool force_nonblock) |
eddc7ef5 JA |
3342 | { |
3343 | struct io_open *ctx = &req->open; | |
3344 | unsigned lookup_flags; | |
3345 | struct path path; | |
3346 | struct kstat stat; | |
3347 | int ret; | |
3348 | ||
5b0bbee4 JA |
3349 | if (force_nonblock) { |
3350 | /* only need file table for an actual valid fd */ | |
3351 | if (ctx->dfd == -1 || ctx->dfd == AT_FDCWD) | |
3352 | req->flags |= REQ_F_NO_FILE_TABLE; | |
eddc7ef5 | 3353 | return -EAGAIN; |
5b0bbee4 | 3354 | } |
eddc7ef5 | 3355 | |
c12cedf2 | 3356 | if (vfs_stat_set_lookup_flags(&lookup_flags, ctx->how.flags)) |
eddc7ef5 JA |
3357 | return -EINVAL; |
3358 | ||
3359 | retry: | |
3360 | /* filename_lookup() drops it, keep a reference */ | |
3361 | ctx->filename->refcnt++; | |
3362 | ||
3363 | ret = filename_lookup(ctx->dfd, ctx->filename, lookup_flags, &path, | |
3364 | NULL); | |
3365 | if (ret) | |
3366 | goto err; | |
3367 | ||
c12cedf2 | 3368 | ret = vfs_getattr(&path, &stat, ctx->mask, ctx->how.flags); |
eddc7ef5 JA |
3369 | path_put(&path); |
3370 | if (retry_estale(ret, lookup_flags)) { | |
3371 | lookup_flags |= LOOKUP_REVAL; | |
3372 | goto retry; | |
3373 | } | |
3374 | if (!ret) | |
3375 | ret = cp_statx(&stat, ctx->buffer); | |
3376 | err: | |
3377 | putname(ctx->filename); | |
8fef80bf | 3378 | req->flags &= ~REQ_F_NEED_CLEANUP; |
eddc7ef5 JA |
3379 | if (ret < 0) |
3380 | req_set_fail_links(req); | |
3381 | io_cqring_add_event(req, ret); | |
014db007 | 3382 | io_put_req(req); |
eddc7ef5 JA |
3383 | return 0; |
3384 | } | |
3385 | ||
b5dba59e JA |
3386 | static int io_close_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
3387 | { | |
3388 | /* | |
3389 | * If we queue this for async, it must not be cancellable. That would | |
3390 | * leave the 'file' in an undeterminate state. | |
3391 | */ | |
3392 | req->work.flags |= IO_WQ_WORK_NO_CANCEL; | |
3393 | ||
3394 | if (sqe->ioprio || sqe->off || sqe->addr || sqe->len || | |
3395 | sqe->rw_flags || sqe->buf_index) | |
3396 | return -EINVAL; | |
9c280f90 | 3397 | if (req->flags & REQ_F_FIXED_FILE) |
cf3040ca | 3398 | return -EBADF; |
b5dba59e JA |
3399 | |
3400 | req->close.fd = READ_ONCE(sqe->fd); | |
b5dba59e JA |
3401 | return 0; |
3402 | } | |
3403 | ||
a93b3331 | 3404 | /* only called when __close_fd_get_file() is done */ |
014db007 | 3405 | static void __io_close_finish(struct io_kiocb *req) |
a93b3331 PB |
3406 | { |
3407 | int ret; | |
3408 | ||
3409 | ret = filp_close(req->close.put_file, req->work.files); | |
3410 | if (ret < 0) | |
3411 | req_set_fail_links(req); | |
3412 | io_cqring_add_event(req, ret); | |
3413 | fput(req->close.put_file); | |
014db007 | 3414 | io_put_req(req); |
a93b3331 PB |
3415 | } |
3416 | ||
b5dba59e JA |
3417 | static void io_close_finish(struct io_wq_work **workptr) |
3418 | { | |
3419 | struct io_kiocb *req = container_of(*workptr, struct io_kiocb, work); | |
b5dba59e | 3420 | |
7fbeb95d | 3421 | /* not cancellable, don't do io_req_cancelled() */ |
014db007 | 3422 | __io_close_finish(req); |
e9fd9396 | 3423 | io_steal_work(req, workptr); |
b5dba59e JA |
3424 | } |
3425 | ||
014db007 | 3426 | static int io_close(struct io_kiocb *req, bool force_nonblock) |
b5dba59e JA |
3427 | { |
3428 | int ret; | |
3429 | ||
3430 | req->close.put_file = NULL; | |
3431 | ret = __close_fd_get_file(req->close.fd, &req->close.put_file); | |
904fbcb1 JA |
3432 | if (ret < 0) { |
3433 | if (ret == -ENOENT) | |
3434 | ret = -EBADF; | |
b5dba59e | 3435 | return ret; |
904fbcb1 | 3436 | } |
b5dba59e JA |
3437 | |
3438 | /* if the file has a flush method, be safe and punt to async */ | |
a2100672 | 3439 | if (req->close.put_file->f_op->flush && force_nonblock) { |
594506fe PB |
3440 | /* submission ref will be dropped, take it for async */ |
3441 | refcount_inc(&req->refs); | |
3442 | ||
a2100672 PB |
3443 | req->work.func = io_close_finish; |
3444 | /* | |
3445 | * Do manual async queue here to avoid grabbing files - we don't | |
3446 | * need the files, and it'll cause io_close_finish() to close | |
3447 | * the file again and cause a double CQE entry for this request | |
3448 | */ | |
3449 | io_queue_async_work(req); | |
3450 | return 0; | |
3451 | } | |
b5dba59e JA |
3452 | |
3453 | /* | |
3454 | * No ->flush(), safely close from here and just punt the | |
3455 | * fput() to async context. | |
3456 | */ | |
014db007 | 3457 | __io_close_finish(req); |
1a417f4e | 3458 | return 0; |
b5dba59e JA |
3459 | } |
3460 | ||
3529d8c2 | 3461 | static int io_prep_sfr(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
5d17b4a4 JA |
3462 | { |
3463 | struct io_ring_ctx *ctx = req->ctx; | |
5d17b4a4 JA |
3464 | |
3465 | if (!req->file) | |
3466 | return -EBADF; | |
5d17b4a4 JA |
3467 | |
3468 | if (unlikely(ctx->flags & IORING_SETUP_IOPOLL)) | |
3469 | return -EINVAL; | |
3470 | if (unlikely(sqe->addr || sqe->ioprio || sqe->buf_index)) | |
3471 | return -EINVAL; | |
3472 | ||
8ed8d3c3 JA |
3473 | req->sync.off = READ_ONCE(sqe->off); |
3474 | req->sync.len = READ_ONCE(sqe->len); | |
3475 | req->sync.flags = READ_ONCE(sqe->sync_range_flags); | |
8ed8d3c3 JA |
3476 | return 0; |
3477 | } | |
3478 | ||
014db007 | 3479 | static void __io_sync_file_range(struct io_kiocb *req) |
8ed8d3c3 | 3480 | { |
8ed8d3c3 JA |
3481 | int ret; |
3482 | ||
9adbd45d | 3483 | ret = sync_file_range(req->file, req->sync.off, req->sync.len, |
8ed8d3c3 JA |
3484 | req->sync.flags); |
3485 | if (ret < 0) | |
3486 | req_set_fail_links(req); | |
3487 | io_cqring_add_event(req, ret); | |
014db007 | 3488 | io_put_req(req); |
5ea62161 PB |
3489 | } |
3490 | ||
3491 | ||
3492 | static void io_sync_file_range_finish(struct io_wq_work **workptr) | |
3493 | { | |
3494 | struct io_kiocb *req = container_of(*workptr, struct io_kiocb, work); | |
5ea62161 PB |
3495 | |
3496 | if (io_req_cancelled(req)) | |
3497 | return; | |
014db007 | 3498 | __io_sync_file_range(req); |
7759a0bf | 3499 | io_steal_work(req, workptr); |
5d17b4a4 JA |
3500 | } |
3501 | ||
014db007 | 3502 | static int io_sync_file_range(struct io_kiocb *req, bool force_nonblock) |
5d17b4a4 | 3503 | { |
5d17b4a4 | 3504 | /* sync_file_range always requires a blocking context */ |
8ed8d3c3 | 3505 | if (force_nonblock) { |
8ed8d3c3 | 3506 | req->work.func = io_sync_file_range_finish; |
5d17b4a4 | 3507 | return -EAGAIN; |
8ed8d3c3 | 3508 | } |
5d17b4a4 | 3509 | |
014db007 | 3510 | __io_sync_file_range(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 | |
e47293fd JA |
3536 | sr->msg_flags = READ_ONCE(sqe->msg_flags); |
3537 | sr->msg = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
fddaface | 3538 | sr->len = READ_ONCE(sqe->len); |
3529d8c2 | 3539 | |
d8768362 JA |
3540 | #ifdef CONFIG_COMPAT |
3541 | if (req->ctx->compat) | |
3542 | sr->msg_flags |= MSG_CMSG_COMPAT; | |
3543 | #endif | |
3544 | ||
fddaface | 3545 | if (!io || req->opcode == IORING_OP_SEND) |
3529d8c2 | 3546 | return 0; |
5f798bea PB |
3547 | /* iovec is already imported */ |
3548 | if (req->flags & REQ_F_NEED_CLEANUP) | |
3549 | return 0; | |
3529d8c2 | 3550 | |
d9688565 | 3551 | io->msg.iov = io->msg.fast_iov; |
99bc4c38 | 3552 | ret = sendmsg_copy_msghdr(&io->msg.msg, sr->msg, sr->msg_flags, |
e47293fd | 3553 | &io->msg.iov); |
99bc4c38 PB |
3554 | if (!ret) |
3555 | req->flags |= REQ_F_NEED_CLEANUP; | |
3556 | return ret; | |
03b1230c JA |
3557 | } |
3558 | ||
014db007 | 3559 | static int io_sendmsg(struct io_kiocb *req, bool force_nonblock) |
aa1fa28f | 3560 | { |
0b416c3e | 3561 | struct io_async_msghdr *kmsg = NULL; |
0fa03c62 JA |
3562 | struct socket *sock; |
3563 | int ret; | |
3564 | ||
3565 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) | |
3566 | return -EINVAL; | |
3567 | ||
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 | ||
3621 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) | |
3622 | return -EINVAL; | |
3623 | ||
3624 | sock = sock_from_file(req->file, &ret); | |
3625 | if (sock) { | |
3626 | struct io_sr_msg *sr = &req->sr_msg; | |
3627 | struct msghdr msg; | |
3628 | struct iovec iov; | |
3629 | unsigned flags; | |
3630 | ||
3631 | ret = import_single_range(WRITE, sr->buf, sr->len, &iov, | |
3632 | &msg.msg_iter); | |
3633 | if (ret) | |
3634 | return ret; | |
3635 | ||
3636 | msg.msg_name = NULL; | |
3637 | msg.msg_control = NULL; | |
3638 | msg.msg_controllen = 0; | |
3639 | msg.msg_namelen = 0; | |
3640 | ||
3641 | flags = req->sr_msg.msg_flags; | |
3642 | if (flags & MSG_DONTWAIT) | |
3643 | req->flags |= REQ_F_NOWAIT; | |
3644 | else if (force_nonblock) | |
3645 | flags |= MSG_DONTWAIT; | |
3646 | ||
0b7b21e4 JA |
3647 | msg.msg_flags = flags; |
3648 | ret = sock_sendmsg(sock, &msg); | |
fddaface JA |
3649 | if (force_nonblock && ret == -EAGAIN) |
3650 | return -EAGAIN; | |
3651 | if (ret == -ERESTARTSYS) | |
3652 | ret = -EINTR; | |
3653 | } | |
3654 | ||
3655 | io_cqring_add_event(req, ret); | |
3656 | if (ret < 0) | |
3657 | req_set_fail_links(req); | |
014db007 | 3658 | io_put_req(req); |
fddaface | 3659 | return 0; |
fddaface JA |
3660 | } |
3661 | ||
52de1fe1 JA |
3662 | static int __io_recvmsg_copy_hdr(struct io_kiocb *req, struct io_async_ctx *io) |
3663 | { | |
3664 | struct io_sr_msg *sr = &req->sr_msg; | |
3665 | struct iovec __user *uiov; | |
3666 | size_t iov_len; | |
3667 | int ret; | |
3668 | ||
3669 | ret = __copy_msghdr_from_user(&io->msg.msg, sr->msg, &io->msg.uaddr, | |
3670 | &uiov, &iov_len); | |
3671 | if (ret) | |
3672 | return ret; | |
3673 | ||
3674 | if (req->flags & REQ_F_BUFFER_SELECT) { | |
3675 | if (iov_len > 1) | |
3676 | return -EINVAL; | |
3677 | if (copy_from_user(io->msg.iov, uiov, sizeof(*uiov))) | |
3678 | return -EFAULT; | |
3679 | sr->len = io->msg.iov[0].iov_len; | |
3680 | iov_iter_init(&io->msg.msg.msg_iter, READ, io->msg.iov, 1, | |
3681 | sr->len); | |
3682 | io->msg.iov = NULL; | |
3683 | } else { | |
3684 | ret = import_iovec(READ, uiov, iov_len, UIO_FASTIOV, | |
3685 | &io->msg.iov, &io->msg.msg.msg_iter); | |
3686 | if (ret > 0) | |
3687 | ret = 0; | |
3688 | } | |
3689 | ||
3690 | return ret; | |
3691 | } | |
3692 | ||
3693 | #ifdef CONFIG_COMPAT | |
3694 | static int __io_compat_recvmsg_copy_hdr(struct io_kiocb *req, | |
3695 | struct io_async_ctx *io) | |
3696 | { | |
3697 | struct compat_msghdr __user *msg_compat; | |
3698 | struct io_sr_msg *sr = &req->sr_msg; | |
3699 | struct compat_iovec __user *uiov; | |
3700 | compat_uptr_t ptr; | |
3701 | compat_size_t len; | |
3702 | int ret; | |
3703 | ||
3704 | msg_compat = (struct compat_msghdr __user *) sr->msg; | |
3705 | ret = __get_compat_msghdr(&io->msg.msg, msg_compat, &io->msg.uaddr, | |
3706 | &ptr, &len); | |
3707 | if (ret) | |
3708 | return ret; | |
3709 | ||
3710 | uiov = compat_ptr(ptr); | |
3711 | if (req->flags & REQ_F_BUFFER_SELECT) { | |
3712 | compat_ssize_t clen; | |
3713 | ||
3714 | if (len > 1) | |
3715 | return -EINVAL; | |
3716 | if (!access_ok(uiov, sizeof(*uiov))) | |
3717 | return -EFAULT; | |
3718 | if (__get_user(clen, &uiov->iov_len)) | |
3719 | return -EFAULT; | |
3720 | if (clen < 0) | |
3721 | return -EINVAL; | |
3722 | sr->len = io->msg.iov[0].iov_len; | |
3723 | io->msg.iov = NULL; | |
3724 | } else { | |
3725 | ret = compat_import_iovec(READ, uiov, len, UIO_FASTIOV, | |
3726 | &io->msg.iov, | |
3727 | &io->msg.msg.msg_iter); | |
3728 | if (ret < 0) | |
3729 | return ret; | |
3730 | } | |
3731 | ||
3732 | return 0; | |
3733 | } | |
3734 | #endif | |
3735 | ||
3736 | static int io_recvmsg_copy_hdr(struct io_kiocb *req, struct io_async_ctx *io) | |
3737 | { | |
3738 | io->msg.iov = io->msg.fast_iov; | |
3739 | ||
3740 | #ifdef CONFIG_COMPAT | |
3741 | if (req->ctx->compat) | |
3742 | return __io_compat_recvmsg_copy_hdr(req, io); | |
fddaface | 3743 | #endif |
52de1fe1 JA |
3744 | |
3745 | return __io_recvmsg_copy_hdr(req, io); | |
3746 | } | |
3747 | ||
bcda7baa JA |
3748 | static struct io_buffer *io_recv_buffer_select(struct io_kiocb *req, |
3749 | int *cflags, bool needs_lock) | |
3750 | { | |
3751 | struct io_sr_msg *sr = &req->sr_msg; | |
3752 | struct io_buffer *kbuf; | |
3753 | ||
3754 | if (!(req->flags & REQ_F_BUFFER_SELECT)) | |
3755 | return NULL; | |
3756 | ||
3757 | kbuf = io_buffer_select(req, &sr->len, sr->bgid, sr->kbuf, needs_lock); | |
3758 | if (IS_ERR(kbuf)) | |
3759 | return kbuf; | |
3760 | ||
3761 | sr->kbuf = kbuf; | |
3762 | req->flags |= REQ_F_BUFFER_SELECTED; | |
3763 | ||
3764 | *cflags = kbuf->bid << IORING_CQE_BUFFER_SHIFT; | |
3765 | *cflags |= IORING_CQE_F_BUFFER; | |
3766 | return kbuf; | |
fddaface JA |
3767 | } |
3768 | ||
3529d8c2 JA |
3769 | static int io_recvmsg_prep(struct io_kiocb *req, |
3770 | const struct io_uring_sqe *sqe) | |
aa1fa28f | 3771 | { |
e47293fd | 3772 | struct io_sr_msg *sr = &req->sr_msg; |
3529d8c2 | 3773 | struct io_async_ctx *io = req->io; |
99bc4c38 | 3774 | int ret; |
3529d8c2 JA |
3775 | |
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 JA |
3803 | |
3804 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) | |
3805 | return -EINVAL; | |
3806 | ||
3807 | sock = sock_from_file(req->file, &ret); | |
3808 | if (sock) { | |
52de1fe1 | 3809 | struct io_buffer *kbuf; |
b7bb4f7d | 3810 | struct io_async_ctx io; |
03b1230c JA |
3811 | unsigned flags; |
3812 | ||
03b1230c | 3813 | if (req->io) { |
0b416c3e | 3814 | kmsg = &req->io->msg; |
b537916c | 3815 | kmsg->msg.msg_name = &req->io->msg.addr; |
0b416c3e JA |
3816 | /* if iov is set, it's allocated already */ |
3817 | if (!kmsg->iov) | |
3818 | kmsg->iov = kmsg->fast_iov; | |
3819 | kmsg->msg.msg_iter.iov = kmsg->iov; | |
03b1230c | 3820 | } else { |
0b416c3e | 3821 | kmsg = &io.msg; |
b537916c | 3822 | kmsg->msg.msg_name = &io.msg.addr; |
3529d8c2 | 3823 | |
52de1fe1 | 3824 | ret = io_recvmsg_copy_hdr(req, &io); |
03b1230c | 3825 | if (ret) |
3529d8c2 | 3826 | return ret; |
03b1230c JA |
3827 | } |
3828 | ||
52de1fe1 JA |
3829 | kbuf = io_recv_buffer_select(req, &cflags, !force_nonblock); |
3830 | if (IS_ERR(kbuf)) { | |
3831 | return PTR_ERR(kbuf); | |
3832 | } else if (kbuf) { | |
3833 | kmsg->fast_iov[0].iov_base = u64_to_user_ptr(kbuf->addr); | |
3834 | iov_iter_init(&kmsg->msg.msg_iter, READ, kmsg->iov, | |
3835 | 1, req->sr_msg.len); | |
3836 | } | |
3837 | ||
e47293fd JA |
3838 | flags = req->sr_msg.msg_flags; |
3839 | if (flags & MSG_DONTWAIT) | |
3840 | req->flags |= REQ_F_NOWAIT; | |
3841 | else if (force_nonblock) | |
3842 | flags |= MSG_DONTWAIT; | |
3843 | ||
3844 | ret = __sys_recvmsg_sock(sock, &kmsg->msg, req->sr_msg.msg, | |
3845 | kmsg->uaddr, flags); | |
02d27d89 PB |
3846 | if (force_nonblock && ret == -EAGAIN) |
3847 | return io_setup_async_msg(req, kmsg); | |
03b1230c JA |
3848 | if (ret == -ERESTARTSYS) |
3849 | ret = -EINTR; | |
3850 | } | |
3851 | ||
1e95081c | 3852 | if (kmsg && kmsg->iov != kmsg->fast_iov) |
0b416c3e | 3853 | kfree(kmsg->iov); |
99bc4c38 | 3854 | req->flags &= ~REQ_F_NEED_CLEANUP; |
52de1fe1 | 3855 | __io_cqring_add_event(req, ret, cflags); |
4e88d6e7 JA |
3856 | if (ret < 0) |
3857 | req_set_fail_links(req); | |
014db007 | 3858 | io_put_req(req); |
03b1230c | 3859 | return 0; |
0fa03c62 | 3860 | } |
5d17b4a4 | 3861 | |
014db007 | 3862 | static int io_recv(struct io_kiocb *req, bool force_nonblock) |
fddaface | 3863 | { |
bcda7baa | 3864 | struct io_buffer *kbuf = NULL; |
fddaface | 3865 | struct socket *sock; |
bcda7baa | 3866 | int ret, cflags = 0; |
fddaface JA |
3867 | |
3868 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) | |
3869 | return -EINVAL; | |
3870 | ||
3871 | sock = sock_from_file(req->file, &ret); | |
3872 | if (sock) { | |
3873 | struct io_sr_msg *sr = &req->sr_msg; | |
bcda7baa | 3874 | void __user *buf = sr->buf; |
fddaface JA |
3875 | struct msghdr msg; |
3876 | struct iovec iov; | |
3877 | unsigned flags; | |
3878 | ||
bcda7baa JA |
3879 | kbuf = io_recv_buffer_select(req, &cflags, !force_nonblock); |
3880 | if (IS_ERR(kbuf)) | |
3881 | return PTR_ERR(kbuf); | |
3882 | else if (kbuf) | |
3883 | buf = u64_to_user_ptr(kbuf->addr); | |
3884 | ||
3885 | ret = import_single_range(READ, buf, sr->len, &iov, | |
fddaface | 3886 | &msg.msg_iter); |
bcda7baa JA |
3887 | if (ret) { |
3888 | kfree(kbuf); | |
fddaface | 3889 | return ret; |
bcda7baa | 3890 | } |
fddaface | 3891 | |
bcda7baa | 3892 | req->flags |= REQ_F_NEED_CLEANUP; |
fddaface JA |
3893 | msg.msg_name = NULL; |
3894 | msg.msg_control = NULL; | |
3895 | msg.msg_controllen = 0; | |
3896 | msg.msg_namelen = 0; | |
3897 | msg.msg_iocb = NULL; | |
3898 | msg.msg_flags = 0; | |
3899 | ||
3900 | flags = req->sr_msg.msg_flags; | |
3901 | if (flags & MSG_DONTWAIT) | |
3902 | req->flags |= REQ_F_NOWAIT; | |
3903 | else if (force_nonblock) | |
3904 | flags |= MSG_DONTWAIT; | |
3905 | ||
0b7b21e4 | 3906 | ret = sock_recvmsg(sock, &msg, flags); |
fddaface JA |
3907 | if (force_nonblock && ret == -EAGAIN) |
3908 | return -EAGAIN; | |
3909 | if (ret == -ERESTARTSYS) | |
3910 | ret = -EINTR; | |
3911 | } | |
3912 | ||
bcda7baa JA |
3913 | kfree(kbuf); |
3914 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
3915 | __io_cqring_add_event(req, ret, cflags); | |
fddaface JA |
3916 | if (ret < 0) |
3917 | req_set_fail_links(req); | |
014db007 | 3918 | io_put_req(req); |
fddaface | 3919 | return 0; |
fddaface JA |
3920 | } |
3921 | ||
3529d8c2 | 3922 | static int io_accept_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
17f2fe35 | 3923 | { |
8ed8d3c3 JA |
3924 | struct io_accept *accept = &req->accept; |
3925 | ||
17f2fe35 JA |
3926 | if (unlikely(req->ctx->flags & (IORING_SETUP_IOPOLL|IORING_SETUP_SQPOLL))) |
3927 | return -EINVAL; | |
8042d6ce | 3928 | if (sqe->ioprio || sqe->len || sqe->buf_index) |
17f2fe35 JA |
3929 | return -EINVAL; |
3930 | ||
d55e5f5b JA |
3931 | accept->addr = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
3932 | accept->addr_len = u64_to_user_ptr(READ_ONCE(sqe->addr2)); | |
8ed8d3c3 | 3933 | accept->flags = READ_ONCE(sqe->accept_flags); |
09952e3e | 3934 | accept->nofile = rlimit(RLIMIT_NOFILE); |
8ed8d3c3 | 3935 | return 0; |
8ed8d3c3 | 3936 | } |
17f2fe35 | 3937 | |
014db007 | 3938 | static int __io_accept(struct io_kiocb *req, bool force_nonblock) |
8ed8d3c3 JA |
3939 | { |
3940 | struct io_accept *accept = &req->accept; | |
3941 | unsigned file_flags; | |
3942 | int ret; | |
3943 | ||
3944 | file_flags = force_nonblock ? O_NONBLOCK : 0; | |
3945 | ret = __sys_accept4_file(req->file, file_flags, accept->addr, | |
09952e3e JA |
3946 | accept->addr_len, accept->flags, |
3947 | accept->nofile); | |
8ed8d3c3 | 3948 | if (ret == -EAGAIN && force_nonblock) |
17f2fe35 | 3949 | return -EAGAIN; |
8e3cca12 JA |
3950 | if (ret == -ERESTARTSYS) |
3951 | ret = -EINTR; | |
4e88d6e7 JA |
3952 | if (ret < 0) |
3953 | req_set_fail_links(req); | |
78e19bbe | 3954 | io_cqring_add_event(req, ret); |
014db007 | 3955 | io_put_req(req); |
17f2fe35 | 3956 | return 0; |
8ed8d3c3 JA |
3957 | } |
3958 | ||
3959 | static void io_accept_finish(struct io_wq_work **workptr) | |
3960 | { | |
3961 | struct io_kiocb *req = container_of(*workptr, struct io_kiocb, work); | |
8ed8d3c3 JA |
3962 | |
3963 | if (io_req_cancelled(req)) | |
3964 | return; | |
014db007 | 3965 | __io_accept(req, false); |
e9fd9396 | 3966 | io_steal_work(req, workptr); |
8ed8d3c3 | 3967 | } |
8ed8d3c3 | 3968 | |
014db007 | 3969 | static int io_accept(struct io_kiocb *req, bool force_nonblock) |
8ed8d3c3 | 3970 | { |
8ed8d3c3 JA |
3971 | int ret; |
3972 | ||
014db007 | 3973 | ret = __io_accept(req, force_nonblock); |
8ed8d3c3 JA |
3974 | if (ret == -EAGAIN && force_nonblock) { |
3975 | req->work.func = io_accept_finish; | |
8ed8d3c3 JA |
3976 | return -EAGAIN; |
3977 | } | |
3978 | return 0; | |
0fa03c62 | 3979 | } |
5d17b4a4 | 3980 | |
3529d8c2 | 3981 | static int io_connect_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
f499a021 | 3982 | { |
3529d8c2 JA |
3983 | struct io_connect *conn = &req->connect; |
3984 | struct io_async_ctx *io = req->io; | |
f499a021 | 3985 | |
3fbb51c1 JA |
3986 | if (unlikely(req->ctx->flags & (IORING_SETUP_IOPOLL|IORING_SETUP_SQPOLL))) |
3987 | return -EINVAL; | |
3988 | if (sqe->ioprio || sqe->len || sqe->buf_index || sqe->rw_flags) | |
3989 | return -EINVAL; | |
3990 | ||
3529d8c2 JA |
3991 | conn->addr = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
3992 | conn->addr_len = READ_ONCE(sqe->addr2); | |
3993 | ||
3994 | if (!io) | |
3995 | return 0; | |
3996 | ||
3997 | return move_addr_to_kernel(conn->addr, conn->addr_len, | |
3fbb51c1 | 3998 | &io->connect.address); |
f499a021 JA |
3999 | } |
4000 | ||
014db007 | 4001 | static int io_connect(struct io_kiocb *req, bool force_nonblock) |
f8e85cf2 | 4002 | { |
f499a021 | 4003 | struct io_async_ctx __io, *io; |
f8e85cf2 | 4004 | unsigned file_flags; |
3fbb51c1 | 4005 | int ret; |
f8e85cf2 | 4006 | |
f499a021 JA |
4007 | if (req->io) { |
4008 | io = req->io; | |
4009 | } else { | |
3529d8c2 JA |
4010 | ret = move_addr_to_kernel(req->connect.addr, |
4011 | req->connect.addr_len, | |
4012 | &__io.connect.address); | |
f499a021 JA |
4013 | if (ret) |
4014 | goto out; | |
4015 | io = &__io; | |
4016 | } | |
4017 | ||
3fbb51c1 JA |
4018 | file_flags = force_nonblock ? O_NONBLOCK : 0; |
4019 | ||
4020 | ret = __sys_connect_file(req->file, &io->connect.address, | |
4021 | req->connect.addr_len, file_flags); | |
87f80d62 | 4022 | if ((ret == -EAGAIN || ret == -EINPROGRESS) && force_nonblock) { |
b7bb4f7d JA |
4023 | if (req->io) |
4024 | return -EAGAIN; | |
4025 | if (io_alloc_async_ctx(req)) { | |
f499a021 JA |
4026 | ret = -ENOMEM; |
4027 | goto out; | |
4028 | } | |
b7bb4f7d | 4029 | memcpy(&req->io->connect, &__io.connect, sizeof(__io.connect)); |
f8e85cf2 | 4030 | return -EAGAIN; |
f499a021 | 4031 | } |
f8e85cf2 JA |
4032 | if (ret == -ERESTARTSYS) |
4033 | ret = -EINTR; | |
f499a021 | 4034 | out: |
4e88d6e7 JA |
4035 | if (ret < 0) |
4036 | req_set_fail_links(req); | |
f8e85cf2 | 4037 | io_cqring_add_event(req, ret); |
014db007 | 4038 | io_put_req(req); |
f8e85cf2 | 4039 | return 0; |
469956e8 Y |
4040 | } |
4041 | #else /* !CONFIG_NET */ | |
4042 | static int io_sendmsg_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) | |
4043 | { | |
f8e85cf2 | 4044 | return -EOPNOTSUPP; |
f8e85cf2 JA |
4045 | } |
4046 | ||
469956e8 Y |
4047 | static int io_sendmsg(struct io_kiocb *req, bool force_nonblock) |
4048 | { | |
4049 | return -EOPNOTSUPP; | |
4050 | } | |
4051 | ||
4052 | static int io_send(struct io_kiocb *req, bool force_nonblock) | |
4053 | { | |
4054 | return -EOPNOTSUPP; | |
4055 | } | |
4056 | ||
4057 | static int io_recvmsg_prep(struct io_kiocb *req, | |
4058 | const struct io_uring_sqe *sqe) | |
4059 | { | |
4060 | return -EOPNOTSUPP; | |
4061 | } | |
4062 | ||
4063 | static int io_recvmsg(struct io_kiocb *req, bool force_nonblock) | |
4064 | { | |
4065 | return -EOPNOTSUPP; | |
4066 | } | |
4067 | ||
4068 | static int io_recv(struct io_kiocb *req, bool force_nonblock) | |
4069 | { | |
4070 | return -EOPNOTSUPP; | |
4071 | } | |
4072 | ||
4073 | static int io_accept_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) | |
4074 | { | |
4075 | return -EOPNOTSUPP; | |
4076 | } | |
4077 | ||
4078 | static int io_accept(struct io_kiocb *req, bool force_nonblock) | |
4079 | { | |
4080 | return -EOPNOTSUPP; | |
4081 | } | |
4082 | ||
4083 | static int io_connect_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) | |
4084 | { | |
4085 | return -EOPNOTSUPP; | |
4086 | } | |
4087 | ||
4088 | static int io_connect(struct io_kiocb *req, bool force_nonblock) | |
4089 | { | |
f8e85cf2 | 4090 | return -EOPNOTSUPP; |
f8e85cf2 | 4091 | } |
469956e8 | 4092 | #endif /* CONFIG_NET */ |
f8e85cf2 | 4093 | |
d7718a9d JA |
4094 | struct io_poll_table { |
4095 | struct poll_table_struct pt; | |
4096 | struct io_kiocb *req; | |
4097 | int error; | |
4098 | }; | |
4099 | ||
4100 | static void __io_queue_proc(struct io_poll_iocb *poll, struct io_poll_table *pt, | |
4101 | struct wait_queue_head *head) | |
4102 | { | |
4103 | if (unlikely(poll->head)) { | |
4104 | pt->error = -EINVAL; | |
4105 | return; | |
4106 | } | |
4107 | ||
4108 | pt->error = 0; | |
4109 | poll->head = head; | |
4110 | add_wait_queue(head, &poll->wait); | |
4111 | } | |
4112 | ||
4113 | static void io_async_queue_proc(struct file *file, struct wait_queue_head *head, | |
4114 | struct poll_table_struct *p) | |
4115 | { | |
4116 | struct io_poll_table *pt = container_of(p, struct io_poll_table, pt); | |
4117 | ||
4118 | __io_queue_proc(&pt->req->apoll->poll, pt, head); | |
4119 | } | |
4120 | ||
4121 | static int __io_async_wake(struct io_kiocb *req, struct io_poll_iocb *poll, | |
4122 | __poll_t mask, task_work_func_t func) | |
4123 | { | |
4124 | struct task_struct *tsk; | |
aa96bf8a | 4125 | int ret; |
d7718a9d JA |
4126 | |
4127 | /* for instances that support it check for an event match first: */ | |
4128 | if (mask && !(mask & poll->events)) | |
4129 | return 0; | |
4130 | ||
4131 | trace_io_uring_task_add(req->ctx, req->opcode, req->user_data, mask); | |
4132 | ||
4133 | list_del_init(&poll->wait.entry); | |
4134 | ||
4135 | tsk = req->task; | |
4136 | req->result = mask; | |
4137 | init_task_work(&req->task_work, func); | |
4138 | /* | |
aa96bf8a JA |
4139 | * If this fails, then the task is exiting. Punt to one of the io-wq |
4140 | * threads to ensure the work gets run, we can't always rely on exit | |
4141 | * cancelation taking care of this. | |
d7718a9d | 4142 | */ |
aa96bf8a JA |
4143 | ret = task_work_add(tsk, &req->task_work, true); |
4144 | if (unlikely(ret)) { | |
4145 | tsk = io_wq_get_task(req->ctx->io_wq); | |
4146 | task_work_add(tsk, &req->task_work, true); | |
4147 | } | |
d7718a9d JA |
4148 | wake_up_process(tsk); |
4149 | return 1; | |
4150 | } | |
4151 | ||
74ce6ce4 JA |
4152 | static bool io_poll_rewait(struct io_kiocb *req, struct io_poll_iocb *poll) |
4153 | __acquires(&req->ctx->completion_lock) | |
4154 | { | |
4155 | struct io_ring_ctx *ctx = req->ctx; | |
4156 | ||
4157 | if (!req->result && !READ_ONCE(poll->canceled)) { | |
4158 | struct poll_table_struct pt = { ._key = poll->events }; | |
4159 | ||
4160 | req->result = vfs_poll(req->file, &pt) & poll->events; | |
4161 | } | |
4162 | ||
4163 | spin_lock_irq(&ctx->completion_lock); | |
4164 | if (!req->result && !READ_ONCE(poll->canceled)) { | |
4165 | add_wait_queue(poll->head, &poll->wait); | |
4166 | return true; | |
4167 | } | |
4168 | ||
4169 | return false; | |
4170 | } | |
4171 | ||
d7718a9d JA |
4172 | static void io_async_task_func(struct callback_head *cb) |
4173 | { | |
4174 | struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work); | |
4175 | struct async_poll *apoll = req->apoll; | |
4176 | struct io_ring_ctx *ctx = req->ctx; | |
2bae047e | 4177 | bool canceled; |
d7718a9d JA |
4178 | |
4179 | trace_io_uring_task_run(req->ctx, req->opcode, req->user_data); | |
4180 | ||
74ce6ce4 | 4181 | if (io_poll_rewait(req, &apoll->poll)) { |
d7718a9d | 4182 | spin_unlock_irq(&ctx->completion_lock); |
74ce6ce4 | 4183 | return; |
d7718a9d JA |
4184 | } |
4185 | ||
74ce6ce4 JA |
4186 | if (hash_hashed(&req->hash_node)) |
4187 | hash_del(&req->hash_node); | |
4188 | ||
2bae047e JA |
4189 | canceled = READ_ONCE(apoll->poll.canceled); |
4190 | if (canceled) { | |
4191 | io_cqring_fill_event(req, -ECANCELED); | |
4192 | io_commit_cqring(ctx); | |
4193 | } | |
4194 | ||
74ce6ce4 JA |
4195 | spin_unlock_irq(&ctx->completion_lock); |
4196 | ||
44575a67 XW |
4197 | /* restore ->work in case we need to retry again */ |
4198 | memcpy(&req->work, &apoll->work, sizeof(req->work)); | |
4199 | ||
2bae047e JA |
4200 | if (canceled) { |
4201 | kfree(apoll); | |
4202 | io_cqring_ev_posted(ctx); | |
4203 | req_set_fail_links(req); | |
44575a67 | 4204 | io_double_put_req(req); |
2bae047e JA |
4205 | return; |
4206 | } | |
4207 | ||
d7718a9d JA |
4208 | __set_current_state(TASK_RUNNING); |
4209 | mutex_lock(&ctx->uring_lock); | |
4210 | __io_queue_sqe(req, NULL); | |
4211 | mutex_unlock(&ctx->uring_lock); | |
4212 | ||
4213 | kfree(apoll); | |
4214 | } | |
4215 | ||
4216 | static int io_async_wake(struct wait_queue_entry *wait, unsigned mode, int sync, | |
4217 | void *key) | |
4218 | { | |
4219 | struct io_kiocb *req = wait->private; | |
4220 | struct io_poll_iocb *poll = &req->apoll->poll; | |
4221 | ||
4222 | trace_io_uring_poll_wake(req->ctx, req->opcode, req->user_data, | |
4223 | key_to_poll(key)); | |
4224 | ||
4225 | return __io_async_wake(req, poll, key_to_poll(key), io_async_task_func); | |
4226 | } | |
4227 | ||
4228 | static void io_poll_req_insert(struct io_kiocb *req) | |
4229 | { | |
4230 | struct io_ring_ctx *ctx = req->ctx; | |
4231 | struct hlist_head *list; | |
4232 | ||
4233 | list = &ctx->cancel_hash[hash_long(req->user_data, ctx->cancel_hash_bits)]; | |
4234 | hlist_add_head(&req->hash_node, list); | |
4235 | } | |
4236 | ||
4237 | static __poll_t __io_arm_poll_handler(struct io_kiocb *req, | |
4238 | struct io_poll_iocb *poll, | |
4239 | struct io_poll_table *ipt, __poll_t mask, | |
4240 | wait_queue_func_t wake_func) | |
4241 | __acquires(&ctx->completion_lock) | |
4242 | { | |
4243 | struct io_ring_ctx *ctx = req->ctx; | |
4244 | bool cancel = false; | |
4245 | ||
4246 | poll->file = req->file; | |
4247 | poll->head = NULL; | |
4248 | poll->done = poll->canceled = false; | |
4249 | poll->events = mask; | |
4250 | ||
4251 | ipt->pt._key = mask; | |
4252 | ipt->req = req; | |
4253 | ipt->error = -EINVAL; | |
4254 | ||
4255 | INIT_LIST_HEAD(&poll->wait.entry); | |
4256 | init_waitqueue_func_entry(&poll->wait, wake_func); | |
4257 | poll->wait.private = req; | |
4258 | ||
4259 | mask = vfs_poll(req->file, &ipt->pt) & poll->events; | |
4260 | ||
4261 | spin_lock_irq(&ctx->completion_lock); | |
4262 | if (likely(poll->head)) { | |
4263 | spin_lock(&poll->head->lock); | |
4264 | if (unlikely(list_empty(&poll->wait.entry))) { | |
4265 | if (ipt->error) | |
4266 | cancel = true; | |
4267 | ipt->error = 0; | |
4268 | mask = 0; | |
4269 | } | |
4270 | if (mask || ipt->error) | |
4271 | list_del_init(&poll->wait.entry); | |
4272 | else if (cancel) | |
4273 | WRITE_ONCE(poll->canceled, true); | |
4274 | else if (!poll->done) /* actually waiting for an event */ | |
4275 | io_poll_req_insert(req); | |
4276 | spin_unlock(&poll->head->lock); | |
4277 | } | |
4278 | ||
4279 | return mask; | |
4280 | } | |
4281 | ||
4282 | static bool io_arm_poll_handler(struct io_kiocb *req) | |
4283 | { | |
4284 | const struct io_op_def *def = &io_op_defs[req->opcode]; | |
4285 | struct io_ring_ctx *ctx = req->ctx; | |
4286 | struct async_poll *apoll; | |
4287 | struct io_poll_table ipt; | |
4288 | __poll_t mask, ret; | |
4289 | ||
4290 | if (!req->file || !file_can_poll(req->file)) | |
4291 | return false; | |
4292 | if (req->flags & (REQ_F_MUST_PUNT | REQ_F_POLLED)) | |
4293 | return false; | |
4294 | if (!def->pollin && !def->pollout) | |
4295 | return false; | |
4296 | ||
4297 | apoll = kmalloc(sizeof(*apoll), GFP_ATOMIC); | |
4298 | if (unlikely(!apoll)) | |
4299 | return false; | |
4300 | ||
4301 | req->flags |= REQ_F_POLLED; | |
4302 | memcpy(&apoll->work, &req->work, sizeof(req->work)); | |
4303 | ||
3537b6a7 | 4304 | get_task_struct(current); |
d7718a9d JA |
4305 | req->task = current; |
4306 | req->apoll = apoll; | |
4307 | INIT_HLIST_NODE(&req->hash_node); | |
4308 | ||
8755d97a | 4309 | mask = 0; |
d7718a9d | 4310 | if (def->pollin) |
8755d97a | 4311 | mask |= POLLIN | POLLRDNORM; |
d7718a9d JA |
4312 | if (def->pollout) |
4313 | mask |= POLLOUT | POLLWRNORM; | |
4314 | mask |= POLLERR | POLLPRI; | |
4315 | ||
4316 | ipt.pt._qproc = io_async_queue_proc; | |
4317 | ||
4318 | ret = __io_arm_poll_handler(req, &apoll->poll, &ipt, mask, | |
4319 | io_async_wake); | |
4320 | if (ret) { | |
4321 | ipt.error = 0; | |
4322 | apoll->poll.done = true; | |
4323 | spin_unlock_irq(&ctx->completion_lock); | |
4324 | memcpy(&req->work, &apoll->work, sizeof(req->work)); | |
4325 | kfree(apoll); | |
4326 | return false; | |
4327 | } | |
4328 | spin_unlock_irq(&ctx->completion_lock); | |
4329 | trace_io_uring_poll_arm(ctx, req->opcode, req->user_data, mask, | |
4330 | apoll->poll.events); | |
4331 | return true; | |
4332 | } | |
4333 | ||
4334 | static bool __io_poll_remove_one(struct io_kiocb *req, | |
4335 | struct io_poll_iocb *poll) | |
221c5eb2 | 4336 | { |
b41e9852 | 4337 | bool do_complete = false; |
221c5eb2 JA |
4338 | |
4339 | spin_lock(&poll->head->lock); | |
4340 | WRITE_ONCE(poll->canceled, true); | |
392edb45 JA |
4341 | if (!list_empty(&poll->wait.entry)) { |
4342 | list_del_init(&poll->wait.entry); | |
b41e9852 | 4343 | do_complete = true; |
221c5eb2 JA |
4344 | } |
4345 | spin_unlock(&poll->head->lock); | |
d7718a9d JA |
4346 | return do_complete; |
4347 | } | |
4348 | ||
4349 | static bool io_poll_remove_one(struct io_kiocb *req) | |
4350 | { | |
b1f573bd | 4351 | struct async_poll *apoll = NULL; |
d7718a9d JA |
4352 | bool do_complete; |
4353 | ||
4354 | if (req->opcode == IORING_OP_POLL_ADD) { | |
4355 | do_complete = __io_poll_remove_one(req, &req->poll); | |
4356 | } else { | |
b1f573bd | 4357 | apoll = req->apoll; |
d7718a9d JA |
4358 | /* non-poll requests have submit ref still */ |
4359 | do_complete = __io_poll_remove_one(req, &req->apoll->poll); | |
4360 | if (do_complete) | |
4361 | io_put_req(req); | |
4362 | } | |
4363 | ||
78076bb6 | 4364 | hash_del(&req->hash_node); |
d7718a9d | 4365 | |
44575a67 | 4366 | if (do_complete && apoll) { |
b1f573bd XW |
4367 | /* |
4368 | * restore ->work because we need to call io_req_work_drop_env. | |
4369 | */ | |
4370 | memcpy(&req->work, &apoll->work, sizeof(req->work)); | |
4371 | kfree(apoll); | |
4372 | } | |
4373 | ||
b41e9852 JA |
4374 | if (do_complete) { |
4375 | io_cqring_fill_event(req, -ECANCELED); | |
4376 | io_commit_cqring(req->ctx); | |
4377 | req->flags |= REQ_F_COMP_LOCKED; | |
4378 | io_put_req(req); | |
4379 | } | |
4380 | ||
4381 | return do_complete; | |
221c5eb2 JA |
4382 | } |
4383 | ||
4384 | static void io_poll_remove_all(struct io_ring_ctx *ctx) | |
4385 | { | |
78076bb6 | 4386 | struct hlist_node *tmp; |
221c5eb2 | 4387 | struct io_kiocb *req; |
8e2e1faf | 4388 | int posted = 0, i; |
221c5eb2 JA |
4389 | |
4390 | spin_lock_irq(&ctx->completion_lock); | |
78076bb6 JA |
4391 | for (i = 0; i < (1U << ctx->cancel_hash_bits); i++) { |
4392 | struct hlist_head *list; | |
4393 | ||
4394 | list = &ctx->cancel_hash[i]; | |
4395 | hlist_for_each_entry_safe(req, tmp, list, hash_node) | |
8e2e1faf | 4396 | posted += io_poll_remove_one(req); |
221c5eb2 JA |
4397 | } |
4398 | spin_unlock_irq(&ctx->completion_lock); | |
b41e9852 | 4399 | |
8e2e1faf JA |
4400 | if (posted) |
4401 | io_cqring_ev_posted(ctx); | |
221c5eb2 JA |
4402 | } |
4403 | ||
47f46768 JA |
4404 | static int io_poll_cancel(struct io_ring_ctx *ctx, __u64 sqe_addr) |
4405 | { | |
78076bb6 | 4406 | struct hlist_head *list; |
47f46768 JA |
4407 | struct io_kiocb *req; |
4408 | ||
78076bb6 JA |
4409 | list = &ctx->cancel_hash[hash_long(sqe_addr, ctx->cancel_hash_bits)]; |
4410 | hlist_for_each_entry(req, list, hash_node) { | |
b41e9852 JA |
4411 | if (sqe_addr != req->user_data) |
4412 | continue; | |
4413 | if (io_poll_remove_one(req)) | |
eac406c6 | 4414 | return 0; |
b41e9852 | 4415 | return -EALREADY; |
47f46768 JA |
4416 | } |
4417 | ||
4418 | return -ENOENT; | |
4419 | } | |
4420 | ||
3529d8c2 JA |
4421 | static int io_poll_remove_prep(struct io_kiocb *req, |
4422 | const struct io_uring_sqe *sqe) | |
0969e783 | 4423 | { |
0969e783 JA |
4424 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4425 | return -EINVAL; | |
4426 | if (sqe->ioprio || sqe->off || sqe->len || sqe->buf_index || | |
4427 | sqe->poll_events) | |
4428 | return -EINVAL; | |
4429 | ||
4430 | req->poll.addr = READ_ONCE(sqe->addr); | |
0969e783 JA |
4431 | return 0; |
4432 | } | |
4433 | ||
221c5eb2 JA |
4434 | /* |
4435 | * Find a running poll command that matches one specified in sqe->addr, | |
4436 | * and remove it if found. | |
4437 | */ | |
fc4df999 | 4438 | static int io_poll_remove(struct io_kiocb *req) |
221c5eb2 JA |
4439 | { |
4440 | struct io_ring_ctx *ctx = req->ctx; | |
0969e783 | 4441 | u64 addr; |
47f46768 | 4442 | int ret; |
221c5eb2 | 4443 | |
0969e783 | 4444 | addr = req->poll.addr; |
221c5eb2 | 4445 | spin_lock_irq(&ctx->completion_lock); |
0969e783 | 4446 | ret = io_poll_cancel(ctx, addr); |
221c5eb2 JA |
4447 | spin_unlock_irq(&ctx->completion_lock); |
4448 | ||
78e19bbe | 4449 | io_cqring_add_event(req, ret); |
4e88d6e7 JA |
4450 | if (ret < 0) |
4451 | req_set_fail_links(req); | |
e65ef56d | 4452 | io_put_req(req); |
221c5eb2 JA |
4453 | return 0; |
4454 | } | |
4455 | ||
b0dd8a41 | 4456 | static void io_poll_complete(struct io_kiocb *req, __poll_t mask, int error) |
221c5eb2 | 4457 | { |
a197f664 JL |
4458 | struct io_ring_ctx *ctx = req->ctx; |
4459 | ||
8c838788 | 4460 | req->poll.done = true; |
b0a20349 | 4461 | io_cqring_fill_event(req, error ? error : mangle_poll(mask)); |
8c838788 | 4462 | io_commit_cqring(ctx); |
221c5eb2 JA |
4463 | } |
4464 | ||
b41e9852 | 4465 | static void io_poll_task_handler(struct io_kiocb *req, struct io_kiocb **nxt) |
221c5eb2 | 4466 | { |
221c5eb2 | 4467 | struct io_ring_ctx *ctx = req->ctx; |
a6ba632d JA |
4468 | struct io_poll_iocb *poll = &req->poll; |
4469 | ||
74ce6ce4 | 4470 | if (io_poll_rewait(req, poll)) { |
a6ba632d JA |
4471 | spin_unlock_irq(&ctx->completion_lock); |
4472 | return; | |
4473 | } | |
74ce6ce4 | 4474 | |
78076bb6 | 4475 | hash_del(&req->hash_node); |
b41e9852 JA |
4476 | io_poll_complete(req, req->result, 0); |
4477 | req->flags |= REQ_F_COMP_LOCKED; | |
4478 | io_put_req_find_next(req, nxt); | |
221c5eb2 JA |
4479 | spin_unlock_irq(&ctx->completion_lock); |
4480 | ||
8c838788 | 4481 | io_cqring_ev_posted(ctx); |
221c5eb2 JA |
4482 | } |
4483 | ||
b41e9852 | 4484 | static void io_poll_task_func(struct callback_head *cb) |
e94f141b | 4485 | { |
b41e9852 JA |
4486 | struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work); |
4487 | struct io_kiocb *nxt = NULL; | |
e94f141b | 4488 | |
b41e9852 | 4489 | io_poll_task_handler(req, &nxt); |
d7718a9d JA |
4490 | if (nxt) { |
4491 | struct io_ring_ctx *ctx = nxt->ctx; | |
e94f141b | 4492 | |
d7718a9d | 4493 | mutex_lock(&ctx->uring_lock); |
b41e9852 | 4494 | __io_queue_sqe(nxt, NULL); |
d7718a9d | 4495 | mutex_unlock(&ctx->uring_lock); |
e94f141b | 4496 | } |
f0b493e6 JA |
4497 | } |
4498 | ||
221c5eb2 JA |
4499 | static int io_poll_wake(struct wait_queue_entry *wait, unsigned mode, int sync, |
4500 | void *key) | |
4501 | { | |
c2f2eb7d JA |
4502 | struct io_kiocb *req = wait->private; |
4503 | struct io_poll_iocb *poll = &req->poll; | |
221c5eb2 | 4504 | |
d7718a9d | 4505 | return __io_async_wake(req, poll, key_to_poll(key), io_poll_task_func); |
221c5eb2 JA |
4506 | } |
4507 | ||
221c5eb2 JA |
4508 | static void io_poll_queue_proc(struct file *file, struct wait_queue_head *head, |
4509 | struct poll_table_struct *p) | |
4510 | { | |
4511 | struct io_poll_table *pt = container_of(p, struct io_poll_table, pt); | |
4512 | ||
d7718a9d | 4513 | __io_queue_proc(&pt->req->poll, pt, head); |
eac406c6 JA |
4514 | } |
4515 | ||
3529d8c2 | 4516 | static int io_poll_add_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
221c5eb2 JA |
4517 | { |
4518 | struct io_poll_iocb *poll = &req->poll; | |
221c5eb2 | 4519 | u16 events; |
221c5eb2 JA |
4520 | |
4521 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) | |
4522 | return -EINVAL; | |
4523 | if (sqe->addr || sqe->ioprio || sqe->off || sqe->len || sqe->buf_index) | |
4524 | return -EINVAL; | |
09bb8394 JA |
4525 | if (!poll->file) |
4526 | return -EBADF; | |
221c5eb2 | 4527 | |
221c5eb2 JA |
4528 | events = READ_ONCE(sqe->poll_events); |
4529 | poll->events = demangle_poll(events) | EPOLLERR | EPOLLHUP; | |
b41e9852 | 4530 | |
3537b6a7 | 4531 | get_task_struct(current); |
b41e9852 | 4532 | req->task = current; |
0969e783 JA |
4533 | return 0; |
4534 | } | |
4535 | ||
014db007 | 4536 | static int io_poll_add(struct io_kiocb *req) |
0969e783 JA |
4537 | { |
4538 | struct io_poll_iocb *poll = &req->poll; | |
4539 | struct io_ring_ctx *ctx = req->ctx; | |
4540 | struct io_poll_table ipt; | |
0969e783 | 4541 | __poll_t mask; |
0969e783 | 4542 | |
78076bb6 | 4543 | INIT_HLIST_NODE(&req->hash_node); |
36703247 | 4544 | INIT_LIST_HEAD(&req->list); |
d7718a9d | 4545 | ipt.pt._qproc = io_poll_queue_proc; |
36703247 | 4546 | |
d7718a9d JA |
4547 | mask = __io_arm_poll_handler(req, &req->poll, &ipt, poll->events, |
4548 | io_poll_wake); | |
221c5eb2 | 4549 | |
8c838788 | 4550 | if (mask) { /* no async, we'd stolen it */ |
221c5eb2 | 4551 | ipt.error = 0; |
b0dd8a41 | 4552 | io_poll_complete(req, mask, 0); |
221c5eb2 | 4553 | } |
221c5eb2 JA |
4554 | spin_unlock_irq(&ctx->completion_lock); |
4555 | ||
8c838788 JA |
4556 | if (mask) { |
4557 | io_cqring_ev_posted(ctx); | |
014db007 | 4558 | io_put_req(req); |
221c5eb2 | 4559 | } |
8c838788 | 4560 | return ipt.error; |
221c5eb2 JA |
4561 | } |
4562 | ||
5262f567 JA |
4563 | static enum hrtimer_restart io_timeout_fn(struct hrtimer *timer) |
4564 | { | |
ad8a48ac JA |
4565 | struct io_timeout_data *data = container_of(timer, |
4566 | struct io_timeout_data, timer); | |
4567 | struct io_kiocb *req = data->req; | |
4568 | struct io_ring_ctx *ctx = req->ctx; | |
5262f567 JA |
4569 | unsigned long flags; |
4570 | ||
5262f567 JA |
4571 | atomic_inc(&ctx->cq_timeouts); |
4572 | ||
4573 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
ef03681a | 4574 | /* |
11365043 JA |
4575 | * We could be racing with timeout deletion. If the list is empty, |
4576 | * then timeout lookup already found it and will be handling it. | |
ef03681a | 4577 | */ |
842f9612 | 4578 | if (!list_empty(&req->list)) { |
11365043 | 4579 | struct io_kiocb *prev; |
5262f567 | 4580 | |
11365043 JA |
4581 | /* |
4582 | * Adjust the reqs sequence before the current one because it | |
d195a66e | 4583 | * will consume a slot in the cq_ring and the cq_tail |
11365043 JA |
4584 | * pointer will be increased, otherwise other timeout reqs may |
4585 | * return in advance without waiting for enough wait_nr. | |
4586 | */ | |
4587 | prev = req; | |
4588 | list_for_each_entry_continue_reverse(prev, &ctx->timeout_list, list) | |
4589 | prev->sequence++; | |
11365043 | 4590 | list_del_init(&req->list); |
11365043 | 4591 | } |
5262f567 | 4592 | |
78e19bbe | 4593 | io_cqring_fill_event(req, -ETIME); |
5262f567 JA |
4594 | io_commit_cqring(ctx); |
4595 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
4596 | ||
4597 | io_cqring_ev_posted(ctx); | |
4e88d6e7 | 4598 | req_set_fail_links(req); |
5262f567 JA |
4599 | io_put_req(req); |
4600 | return HRTIMER_NORESTART; | |
4601 | } | |
4602 | ||
47f46768 JA |
4603 | static int io_timeout_cancel(struct io_ring_ctx *ctx, __u64 user_data) |
4604 | { | |
4605 | struct io_kiocb *req; | |
4606 | int ret = -ENOENT; | |
4607 | ||
4608 | list_for_each_entry(req, &ctx->timeout_list, list) { | |
4609 | if (user_data == req->user_data) { | |
4610 | list_del_init(&req->list); | |
4611 | ret = 0; | |
4612 | break; | |
4613 | } | |
4614 | } | |
4615 | ||
4616 | if (ret == -ENOENT) | |
4617 | return ret; | |
4618 | ||
2d28390a | 4619 | ret = hrtimer_try_to_cancel(&req->io->timeout.timer); |
47f46768 JA |
4620 | if (ret == -1) |
4621 | return -EALREADY; | |
4622 | ||
4e88d6e7 | 4623 | req_set_fail_links(req); |
47f46768 JA |
4624 | io_cqring_fill_event(req, -ECANCELED); |
4625 | io_put_req(req); | |
4626 | return 0; | |
4627 | } | |
4628 | ||
3529d8c2 JA |
4629 | static int io_timeout_remove_prep(struct io_kiocb *req, |
4630 | const struct io_uring_sqe *sqe) | |
b29472ee | 4631 | { |
b29472ee JA |
4632 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4633 | return -EINVAL; | |
4634 | if (sqe->flags || sqe->ioprio || sqe->buf_index || sqe->len) | |
4635 | return -EINVAL; | |
4636 | ||
4637 | req->timeout.addr = READ_ONCE(sqe->addr); | |
4638 | req->timeout.flags = READ_ONCE(sqe->timeout_flags); | |
4639 | if (req->timeout.flags) | |
4640 | return -EINVAL; | |
4641 | ||
b29472ee JA |
4642 | return 0; |
4643 | } | |
4644 | ||
11365043 JA |
4645 | /* |
4646 | * Remove or update an existing timeout command | |
4647 | */ | |
fc4df999 | 4648 | static int io_timeout_remove(struct io_kiocb *req) |
11365043 JA |
4649 | { |
4650 | struct io_ring_ctx *ctx = req->ctx; | |
47f46768 | 4651 | int ret; |
11365043 | 4652 | |
11365043 | 4653 | spin_lock_irq(&ctx->completion_lock); |
b29472ee | 4654 | ret = io_timeout_cancel(ctx, req->timeout.addr); |
11365043 | 4655 | |
47f46768 | 4656 | io_cqring_fill_event(req, ret); |
11365043 JA |
4657 | io_commit_cqring(ctx); |
4658 | spin_unlock_irq(&ctx->completion_lock); | |
5262f567 | 4659 | io_cqring_ev_posted(ctx); |
4e88d6e7 JA |
4660 | if (ret < 0) |
4661 | req_set_fail_links(req); | |
ec9c02ad | 4662 | io_put_req(req); |
11365043 | 4663 | return 0; |
5262f567 JA |
4664 | } |
4665 | ||
3529d8c2 | 4666 | static int io_timeout_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe, |
2d28390a | 4667 | bool is_timeout_link) |
5262f567 | 4668 | { |
ad8a48ac | 4669 | struct io_timeout_data *data; |
a41525ab | 4670 | unsigned flags; |
5262f567 | 4671 | |
ad8a48ac | 4672 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
5262f567 | 4673 | return -EINVAL; |
ad8a48ac | 4674 | if (sqe->ioprio || sqe->buf_index || sqe->len != 1) |
a41525ab | 4675 | return -EINVAL; |
2d28390a JA |
4676 | if (sqe->off && is_timeout_link) |
4677 | return -EINVAL; | |
a41525ab JA |
4678 | flags = READ_ONCE(sqe->timeout_flags); |
4679 | if (flags & ~IORING_TIMEOUT_ABS) | |
5262f567 | 4680 | return -EINVAL; |
bdf20073 | 4681 | |
26a61679 JA |
4682 | req->timeout.count = READ_ONCE(sqe->off); |
4683 | ||
3529d8c2 | 4684 | if (!req->io && io_alloc_async_ctx(req)) |
26a61679 JA |
4685 | return -ENOMEM; |
4686 | ||
4687 | data = &req->io->timeout; | |
ad8a48ac | 4688 | data->req = req; |
ad8a48ac JA |
4689 | req->flags |= REQ_F_TIMEOUT; |
4690 | ||
4691 | if (get_timespec64(&data->ts, u64_to_user_ptr(sqe->addr))) | |
5262f567 JA |
4692 | return -EFAULT; |
4693 | ||
11365043 | 4694 | if (flags & IORING_TIMEOUT_ABS) |
ad8a48ac | 4695 | data->mode = HRTIMER_MODE_ABS; |
11365043 | 4696 | else |
ad8a48ac | 4697 | data->mode = HRTIMER_MODE_REL; |
11365043 | 4698 | |
ad8a48ac JA |
4699 | hrtimer_init(&data->timer, CLOCK_MONOTONIC, data->mode); |
4700 | return 0; | |
4701 | } | |
4702 | ||
fc4df999 | 4703 | static int io_timeout(struct io_kiocb *req) |
ad8a48ac | 4704 | { |
ad8a48ac JA |
4705 | struct io_ring_ctx *ctx = req->ctx; |
4706 | struct io_timeout_data *data; | |
4707 | struct list_head *entry; | |
4708 | unsigned span = 0; | |
b55ce732 | 4709 | u32 count = req->timeout.count; |
22cad158 | 4710 | u32 seq = req->sequence; |
ad8a48ac | 4711 | |
2d28390a | 4712 | data = &req->io->timeout; |
93bd25bb | 4713 | |
5262f567 JA |
4714 | /* |
4715 | * sqe->off holds how many events that need to occur for this | |
93bd25bb JA |
4716 | * timeout event to be satisfied. If it isn't set, then this is |
4717 | * a pure timeout request, sequence isn't used. | |
5262f567 | 4718 | */ |
93bd25bb JA |
4719 | if (!count) { |
4720 | req->flags |= REQ_F_TIMEOUT_NOSEQ; | |
4721 | spin_lock_irq(&ctx->completion_lock); | |
4722 | entry = ctx->timeout_list.prev; | |
4723 | goto add; | |
4724 | } | |
5262f567 | 4725 | |
22cad158 | 4726 | req->sequence = seq + count; |
5262f567 JA |
4727 | |
4728 | /* | |
4729 | * Insertion sort, ensuring the first entry in the list is always | |
4730 | * the one we need first. | |
4731 | */ | |
5262f567 JA |
4732 | spin_lock_irq(&ctx->completion_lock); |
4733 | list_for_each_prev(entry, &ctx->timeout_list) { | |
4734 | struct io_kiocb *nxt = list_entry(entry, struct io_kiocb, list); | |
22cad158 | 4735 | unsigned nxt_seq; |
5da0fb1a | 4736 | long long tmp, tmp_nxt; |
b55ce732 | 4737 | u32 nxt_offset = nxt->timeout.count; |
5262f567 | 4738 | |
93bd25bb JA |
4739 | if (nxt->flags & REQ_F_TIMEOUT_NOSEQ) |
4740 | continue; | |
4741 | ||
5da0fb1a | 4742 | /* |
22cad158 | 4743 | * Since seq + count can overflow, use type long |
5da0fb1a | 4744 | * long to store it. |
4745 | */ | |
22cad158 PB |
4746 | tmp = (long long)seq + count; |
4747 | nxt_seq = nxt->sequence - nxt_offset; | |
4748 | tmp_nxt = (long long)nxt_seq + nxt_offset; | |
5da0fb1a | 4749 | |
4750 | /* | |
4751 | * cached_sq_head may overflow, and it will never overflow twice | |
4752 | * once there is some timeout req still be valid. | |
4753 | */ | |
22cad158 | 4754 | if (seq < nxt_seq) |
8b07a65a | 4755 | tmp += UINT_MAX; |
5da0fb1a | 4756 | |
a1f58ba4 | 4757 | if (tmp > tmp_nxt) |
5262f567 | 4758 | break; |
a1f58ba4 | 4759 | |
4760 | /* | |
4761 | * Sequence of reqs after the insert one and itself should | |
4762 | * be adjusted because each timeout req consumes a slot. | |
4763 | */ | |
4764 | span++; | |
4765 | nxt->sequence++; | |
5262f567 | 4766 | } |
a1f58ba4 | 4767 | req->sequence -= span; |
93bd25bb | 4768 | add: |
5262f567 | 4769 | list_add(&req->list, entry); |
ad8a48ac JA |
4770 | data->timer.function = io_timeout_fn; |
4771 | hrtimer_start(&data->timer, timespec64_to_ktime(data->ts), data->mode); | |
5262f567 | 4772 | spin_unlock_irq(&ctx->completion_lock); |
5262f567 JA |
4773 | return 0; |
4774 | } | |
5262f567 | 4775 | |
62755e35 JA |
4776 | static bool io_cancel_cb(struct io_wq_work *work, void *data) |
4777 | { | |
4778 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); | |
4779 | ||
4780 | return req->user_data == (unsigned long) data; | |
4781 | } | |
4782 | ||
e977d6d3 | 4783 | static int io_async_cancel_one(struct io_ring_ctx *ctx, void *sqe_addr) |
62755e35 | 4784 | { |
62755e35 | 4785 | enum io_wq_cancel cancel_ret; |
62755e35 JA |
4786 | int ret = 0; |
4787 | ||
62755e35 JA |
4788 | cancel_ret = io_wq_cancel_cb(ctx->io_wq, io_cancel_cb, sqe_addr); |
4789 | switch (cancel_ret) { | |
4790 | case IO_WQ_CANCEL_OK: | |
4791 | ret = 0; | |
4792 | break; | |
4793 | case IO_WQ_CANCEL_RUNNING: | |
4794 | ret = -EALREADY; | |
4795 | break; | |
4796 | case IO_WQ_CANCEL_NOTFOUND: | |
4797 | ret = -ENOENT; | |
4798 | break; | |
4799 | } | |
4800 | ||
e977d6d3 JA |
4801 | return ret; |
4802 | } | |
4803 | ||
47f46768 JA |
4804 | static void io_async_find_and_cancel(struct io_ring_ctx *ctx, |
4805 | struct io_kiocb *req, __u64 sqe_addr, | |
014db007 | 4806 | int success_ret) |
47f46768 JA |
4807 | { |
4808 | unsigned long flags; | |
4809 | int ret; | |
4810 | ||
4811 | ret = io_async_cancel_one(ctx, (void *) (unsigned long) sqe_addr); | |
4812 | if (ret != -ENOENT) { | |
4813 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
4814 | goto done; | |
4815 | } | |
4816 | ||
4817 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
4818 | ret = io_timeout_cancel(ctx, sqe_addr); | |
4819 | if (ret != -ENOENT) | |
4820 | goto done; | |
4821 | ret = io_poll_cancel(ctx, sqe_addr); | |
4822 | done: | |
b0dd8a41 JA |
4823 | if (!ret) |
4824 | ret = success_ret; | |
47f46768 JA |
4825 | io_cqring_fill_event(req, ret); |
4826 | io_commit_cqring(ctx); | |
4827 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
4828 | io_cqring_ev_posted(ctx); | |
4829 | ||
4e88d6e7 JA |
4830 | if (ret < 0) |
4831 | req_set_fail_links(req); | |
014db007 | 4832 | io_put_req(req); |
47f46768 JA |
4833 | } |
4834 | ||
3529d8c2 JA |
4835 | static int io_async_cancel_prep(struct io_kiocb *req, |
4836 | const struct io_uring_sqe *sqe) | |
e977d6d3 | 4837 | { |
fbf23849 | 4838 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
e977d6d3 JA |
4839 | return -EINVAL; |
4840 | if (sqe->flags || sqe->ioprio || sqe->off || sqe->len || | |
4841 | sqe->cancel_flags) | |
4842 | return -EINVAL; | |
4843 | ||
fbf23849 JA |
4844 | req->cancel.addr = READ_ONCE(sqe->addr); |
4845 | return 0; | |
4846 | } | |
4847 | ||
014db007 | 4848 | static int io_async_cancel(struct io_kiocb *req) |
fbf23849 JA |
4849 | { |
4850 | struct io_ring_ctx *ctx = req->ctx; | |
fbf23849 | 4851 | |
014db007 | 4852 | io_async_find_and_cancel(ctx, req, req->cancel.addr, 0); |
5262f567 JA |
4853 | return 0; |
4854 | } | |
4855 | ||
05f3fb3c JA |
4856 | static int io_files_update_prep(struct io_kiocb *req, |
4857 | const struct io_uring_sqe *sqe) | |
4858 | { | |
4859 | if (sqe->flags || sqe->ioprio || sqe->rw_flags) | |
4860 | return -EINVAL; | |
4861 | ||
4862 | req->files_update.offset = READ_ONCE(sqe->off); | |
4863 | req->files_update.nr_args = READ_ONCE(sqe->len); | |
4864 | if (!req->files_update.nr_args) | |
4865 | return -EINVAL; | |
4866 | req->files_update.arg = READ_ONCE(sqe->addr); | |
4867 | return 0; | |
4868 | } | |
4869 | ||
4870 | static int io_files_update(struct io_kiocb *req, bool force_nonblock) | |
fbf23849 JA |
4871 | { |
4872 | struct io_ring_ctx *ctx = req->ctx; | |
05f3fb3c JA |
4873 | struct io_uring_files_update up; |
4874 | int ret; | |
fbf23849 | 4875 | |
f86cd20c | 4876 | if (force_nonblock) |
05f3fb3c | 4877 | return -EAGAIN; |
05f3fb3c JA |
4878 | |
4879 | up.offset = req->files_update.offset; | |
4880 | up.fds = req->files_update.arg; | |
4881 | ||
4882 | mutex_lock(&ctx->uring_lock); | |
4883 | ret = __io_sqe_files_update(ctx, &up, req->files_update.nr_args); | |
4884 | mutex_unlock(&ctx->uring_lock); | |
4885 | ||
4886 | if (ret < 0) | |
4887 | req_set_fail_links(req); | |
4888 | io_cqring_add_event(req, ret); | |
4889 | io_put_req(req); | |
5262f567 JA |
4890 | return 0; |
4891 | } | |
4892 | ||
3529d8c2 JA |
4893 | static int io_req_defer_prep(struct io_kiocb *req, |
4894 | const struct io_uring_sqe *sqe) | |
f67676d1 | 4895 | { |
e781573e | 4896 | ssize_t ret = 0; |
f67676d1 | 4897 | |
f1d96a8f PB |
4898 | if (!sqe) |
4899 | return 0; | |
4900 | ||
f86cd20c JA |
4901 | if (io_op_defs[req->opcode].file_table) { |
4902 | ret = io_grab_files(req); | |
4903 | if (unlikely(ret)) | |
4904 | return ret; | |
4905 | } | |
4906 | ||
cccf0ee8 JA |
4907 | io_req_work_grab_env(req, &io_op_defs[req->opcode]); |
4908 | ||
d625c6ee | 4909 | switch (req->opcode) { |
e781573e JA |
4910 | case IORING_OP_NOP: |
4911 | break; | |
f67676d1 JA |
4912 | case IORING_OP_READV: |
4913 | case IORING_OP_READ_FIXED: | |
3a6820f2 | 4914 | case IORING_OP_READ: |
3529d8c2 | 4915 | ret = io_read_prep(req, sqe, true); |
f67676d1 JA |
4916 | break; |
4917 | case IORING_OP_WRITEV: | |
4918 | case IORING_OP_WRITE_FIXED: | |
3a6820f2 | 4919 | case IORING_OP_WRITE: |
3529d8c2 | 4920 | ret = io_write_prep(req, sqe, true); |
f67676d1 | 4921 | break; |
0969e783 | 4922 | case IORING_OP_POLL_ADD: |
3529d8c2 | 4923 | ret = io_poll_add_prep(req, sqe); |
0969e783 JA |
4924 | break; |
4925 | case IORING_OP_POLL_REMOVE: | |
3529d8c2 | 4926 | ret = io_poll_remove_prep(req, sqe); |
0969e783 | 4927 | break; |
8ed8d3c3 | 4928 | case IORING_OP_FSYNC: |
3529d8c2 | 4929 | ret = io_prep_fsync(req, sqe); |
8ed8d3c3 JA |
4930 | break; |
4931 | case IORING_OP_SYNC_FILE_RANGE: | |
3529d8c2 | 4932 | ret = io_prep_sfr(req, sqe); |
8ed8d3c3 | 4933 | break; |
03b1230c | 4934 | case IORING_OP_SENDMSG: |
fddaface | 4935 | case IORING_OP_SEND: |
3529d8c2 | 4936 | ret = io_sendmsg_prep(req, sqe); |
03b1230c JA |
4937 | break; |
4938 | case IORING_OP_RECVMSG: | |
fddaface | 4939 | case IORING_OP_RECV: |
3529d8c2 | 4940 | ret = io_recvmsg_prep(req, sqe); |
03b1230c | 4941 | break; |
f499a021 | 4942 | case IORING_OP_CONNECT: |
3529d8c2 | 4943 | ret = io_connect_prep(req, sqe); |
f499a021 | 4944 | break; |
2d28390a | 4945 | case IORING_OP_TIMEOUT: |
3529d8c2 | 4946 | ret = io_timeout_prep(req, sqe, false); |
b7bb4f7d | 4947 | break; |
b29472ee | 4948 | case IORING_OP_TIMEOUT_REMOVE: |
3529d8c2 | 4949 | ret = io_timeout_remove_prep(req, sqe); |
b29472ee | 4950 | break; |
fbf23849 | 4951 | case IORING_OP_ASYNC_CANCEL: |
3529d8c2 | 4952 | ret = io_async_cancel_prep(req, sqe); |
fbf23849 | 4953 | break; |
2d28390a | 4954 | case IORING_OP_LINK_TIMEOUT: |
3529d8c2 | 4955 | ret = io_timeout_prep(req, sqe, true); |
b7bb4f7d | 4956 | break; |
8ed8d3c3 | 4957 | case IORING_OP_ACCEPT: |
3529d8c2 | 4958 | ret = io_accept_prep(req, sqe); |
8ed8d3c3 | 4959 | break; |
d63d1b5e JA |
4960 | case IORING_OP_FALLOCATE: |
4961 | ret = io_fallocate_prep(req, sqe); | |
4962 | break; | |
15b71abe JA |
4963 | case IORING_OP_OPENAT: |
4964 | ret = io_openat_prep(req, sqe); | |
4965 | break; | |
b5dba59e JA |
4966 | case IORING_OP_CLOSE: |
4967 | ret = io_close_prep(req, sqe); | |
4968 | break; | |
05f3fb3c JA |
4969 | case IORING_OP_FILES_UPDATE: |
4970 | ret = io_files_update_prep(req, sqe); | |
4971 | break; | |
eddc7ef5 JA |
4972 | case IORING_OP_STATX: |
4973 | ret = io_statx_prep(req, sqe); | |
4974 | break; | |
4840e418 JA |
4975 | case IORING_OP_FADVISE: |
4976 | ret = io_fadvise_prep(req, sqe); | |
4977 | break; | |
c1ca757b JA |
4978 | case IORING_OP_MADVISE: |
4979 | ret = io_madvise_prep(req, sqe); | |
4980 | break; | |
cebdb986 JA |
4981 | case IORING_OP_OPENAT2: |
4982 | ret = io_openat2_prep(req, sqe); | |
4983 | break; | |
3e4827b0 JA |
4984 | case IORING_OP_EPOLL_CTL: |
4985 | ret = io_epoll_ctl_prep(req, sqe); | |
4986 | break; | |
7d67af2c PB |
4987 | case IORING_OP_SPLICE: |
4988 | ret = io_splice_prep(req, sqe); | |
4989 | break; | |
ddf0322d JA |
4990 | case IORING_OP_PROVIDE_BUFFERS: |
4991 | ret = io_provide_buffers_prep(req, sqe); | |
4992 | break; | |
067524e9 JA |
4993 | case IORING_OP_REMOVE_BUFFERS: |
4994 | ret = io_remove_buffers_prep(req, sqe); | |
4995 | break; | |
f67676d1 | 4996 | default: |
e781573e JA |
4997 | printk_once(KERN_WARNING "io_uring: unhandled opcode %d\n", |
4998 | req->opcode); | |
4999 | ret = -EINVAL; | |
b7bb4f7d | 5000 | break; |
f67676d1 JA |
5001 | } |
5002 | ||
b7bb4f7d | 5003 | return ret; |
f67676d1 JA |
5004 | } |
5005 | ||
3529d8c2 | 5006 | static int io_req_defer(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
de0617e4 | 5007 | { |
a197f664 | 5008 | struct io_ring_ctx *ctx = req->ctx; |
f67676d1 | 5009 | int ret; |
de0617e4 | 5010 | |
9d858b21 | 5011 | /* Still need defer if there is pending req in defer list. */ |
4ee36314 | 5012 | if (!req_need_defer(req) && list_empty_careful(&ctx->defer_list)) |
de0617e4 JA |
5013 | return 0; |
5014 | ||
3529d8c2 | 5015 | if (!req->io && io_alloc_async_ctx(req)) |
de0617e4 JA |
5016 | return -EAGAIN; |
5017 | ||
3529d8c2 | 5018 | ret = io_req_defer_prep(req, sqe); |
b7bb4f7d | 5019 | if (ret < 0) |
2d28390a | 5020 | return ret; |
2d28390a | 5021 | |
de0617e4 | 5022 | spin_lock_irq(&ctx->completion_lock); |
9d858b21 | 5023 | if (!req_need_defer(req) && list_empty(&ctx->defer_list)) { |
de0617e4 | 5024 | spin_unlock_irq(&ctx->completion_lock); |
de0617e4 JA |
5025 | return 0; |
5026 | } | |
5027 | ||
915967f6 | 5028 | trace_io_uring_defer(ctx, req, req->user_data); |
de0617e4 JA |
5029 | list_add_tail(&req->list, &ctx->defer_list); |
5030 | spin_unlock_irq(&ctx->completion_lock); | |
5031 | return -EIOCBQUEUED; | |
5032 | } | |
5033 | ||
99bc4c38 PB |
5034 | static void io_cleanup_req(struct io_kiocb *req) |
5035 | { | |
5036 | struct io_async_ctx *io = req->io; | |
5037 | ||
5038 | switch (req->opcode) { | |
5039 | case IORING_OP_READV: | |
5040 | case IORING_OP_READ_FIXED: | |
5041 | case IORING_OP_READ: | |
bcda7baa JA |
5042 | if (req->flags & REQ_F_BUFFER_SELECTED) |
5043 | kfree((void *)(unsigned long)req->rw.addr); | |
5044 | /* fallthrough */ | |
99bc4c38 PB |
5045 | case IORING_OP_WRITEV: |
5046 | case IORING_OP_WRITE_FIXED: | |
5047 | case IORING_OP_WRITE: | |
5048 | if (io->rw.iov != io->rw.fast_iov) | |
5049 | kfree(io->rw.iov); | |
5050 | break; | |
99bc4c38 | 5051 | case IORING_OP_RECVMSG: |
52de1fe1 JA |
5052 | if (req->flags & REQ_F_BUFFER_SELECTED) |
5053 | kfree(req->sr_msg.kbuf); | |
5054 | /* fallthrough */ | |
5055 | case IORING_OP_SENDMSG: | |
99bc4c38 PB |
5056 | if (io->msg.iov != io->msg.fast_iov) |
5057 | kfree(io->msg.iov); | |
5058 | break; | |
bcda7baa JA |
5059 | case IORING_OP_RECV: |
5060 | if (req->flags & REQ_F_BUFFER_SELECTED) | |
5061 | kfree(req->sr_msg.kbuf); | |
5062 | break; | |
8fef80bf PB |
5063 | case IORING_OP_OPENAT: |
5064 | case IORING_OP_OPENAT2: | |
5065 | case IORING_OP_STATX: | |
5066 | putname(req->open.filename); | |
5067 | break; | |
7d67af2c PB |
5068 | case IORING_OP_SPLICE: |
5069 | io_put_file(req, req->splice.file_in, | |
5070 | (req->splice.flags & SPLICE_F_FD_IN_FIXED)); | |
5071 | break; | |
99bc4c38 PB |
5072 | } |
5073 | ||
5074 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
5075 | } | |
5076 | ||
3529d8c2 | 5077 | static int io_issue_sqe(struct io_kiocb *req, const struct io_uring_sqe *sqe, |
014db007 | 5078 | bool force_nonblock) |
2b188cc1 | 5079 | { |
a197f664 | 5080 | struct io_ring_ctx *ctx = req->ctx; |
d625c6ee | 5081 | int ret; |
2b188cc1 | 5082 | |
d625c6ee | 5083 | switch (req->opcode) { |
2b188cc1 | 5084 | case IORING_OP_NOP: |
78e19bbe | 5085 | ret = io_nop(req); |
2b188cc1 JA |
5086 | break; |
5087 | case IORING_OP_READV: | |
edafccee | 5088 | case IORING_OP_READ_FIXED: |
3a6820f2 | 5089 | case IORING_OP_READ: |
3529d8c2 JA |
5090 | if (sqe) { |
5091 | ret = io_read_prep(req, sqe, force_nonblock); | |
5092 | if (ret < 0) | |
5093 | break; | |
5094 | } | |
014db007 | 5095 | ret = io_read(req, force_nonblock); |
edafccee | 5096 | break; |
3529d8c2 | 5097 | case IORING_OP_WRITEV: |
edafccee | 5098 | case IORING_OP_WRITE_FIXED: |
3a6820f2 | 5099 | case IORING_OP_WRITE: |
3529d8c2 JA |
5100 | if (sqe) { |
5101 | ret = io_write_prep(req, sqe, force_nonblock); | |
5102 | if (ret < 0) | |
5103 | break; | |
5104 | } | |
014db007 | 5105 | ret = io_write(req, force_nonblock); |
2b188cc1 | 5106 | break; |
c992fe29 | 5107 | case IORING_OP_FSYNC: |
3529d8c2 JA |
5108 | if (sqe) { |
5109 | ret = io_prep_fsync(req, sqe); | |
5110 | if (ret < 0) | |
5111 | break; | |
5112 | } | |
014db007 | 5113 | ret = io_fsync(req, force_nonblock); |
c992fe29 | 5114 | break; |
221c5eb2 | 5115 | case IORING_OP_POLL_ADD: |
3529d8c2 JA |
5116 | if (sqe) { |
5117 | ret = io_poll_add_prep(req, sqe); | |
5118 | if (ret) | |
5119 | break; | |
5120 | } | |
014db007 | 5121 | ret = io_poll_add(req); |
221c5eb2 JA |
5122 | break; |
5123 | case IORING_OP_POLL_REMOVE: | |
3529d8c2 JA |
5124 | if (sqe) { |
5125 | ret = io_poll_remove_prep(req, sqe); | |
5126 | if (ret < 0) | |
5127 | break; | |
5128 | } | |
fc4df999 | 5129 | ret = io_poll_remove(req); |
221c5eb2 | 5130 | break; |
5d17b4a4 | 5131 | case IORING_OP_SYNC_FILE_RANGE: |
3529d8c2 JA |
5132 | if (sqe) { |
5133 | ret = io_prep_sfr(req, sqe); | |
5134 | if (ret < 0) | |
5135 | break; | |
5136 | } | |
014db007 | 5137 | ret = io_sync_file_range(req, force_nonblock); |
5d17b4a4 | 5138 | break; |
0fa03c62 | 5139 | case IORING_OP_SENDMSG: |
fddaface | 5140 | case IORING_OP_SEND: |
3529d8c2 JA |
5141 | if (sqe) { |
5142 | ret = io_sendmsg_prep(req, sqe); | |
5143 | if (ret < 0) | |
5144 | break; | |
5145 | } | |
fddaface | 5146 | if (req->opcode == IORING_OP_SENDMSG) |
014db007 | 5147 | ret = io_sendmsg(req, force_nonblock); |
fddaface | 5148 | else |
014db007 | 5149 | ret = io_send(req, force_nonblock); |
0fa03c62 | 5150 | break; |
aa1fa28f | 5151 | case IORING_OP_RECVMSG: |
fddaface | 5152 | case IORING_OP_RECV: |
3529d8c2 JA |
5153 | if (sqe) { |
5154 | ret = io_recvmsg_prep(req, sqe); | |
5155 | if (ret) | |
5156 | break; | |
5157 | } | |
fddaface | 5158 | if (req->opcode == IORING_OP_RECVMSG) |
014db007 | 5159 | ret = io_recvmsg(req, force_nonblock); |
fddaface | 5160 | else |
014db007 | 5161 | ret = io_recv(req, force_nonblock); |
aa1fa28f | 5162 | break; |
5262f567 | 5163 | case IORING_OP_TIMEOUT: |
3529d8c2 JA |
5164 | if (sqe) { |
5165 | ret = io_timeout_prep(req, sqe, false); | |
5166 | if (ret) | |
5167 | break; | |
5168 | } | |
fc4df999 | 5169 | ret = io_timeout(req); |
5262f567 | 5170 | break; |
11365043 | 5171 | case IORING_OP_TIMEOUT_REMOVE: |
3529d8c2 JA |
5172 | if (sqe) { |
5173 | ret = io_timeout_remove_prep(req, sqe); | |
5174 | if (ret) | |
5175 | break; | |
5176 | } | |
fc4df999 | 5177 | ret = io_timeout_remove(req); |
11365043 | 5178 | break; |
17f2fe35 | 5179 | case IORING_OP_ACCEPT: |
3529d8c2 JA |
5180 | if (sqe) { |
5181 | ret = io_accept_prep(req, sqe); | |
5182 | if (ret) | |
5183 | break; | |
5184 | } | |
014db007 | 5185 | ret = io_accept(req, force_nonblock); |
17f2fe35 | 5186 | break; |
f8e85cf2 | 5187 | case IORING_OP_CONNECT: |
3529d8c2 JA |
5188 | if (sqe) { |
5189 | ret = io_connect_prep(req, sqe); | |
5190 | if (ret) | |
5191 | break; | |
5192 | } | |
014db007 | 5193 | ret = io_connect(req, force_nonblock); |
f8e85cf2 | 5194 | break; |
62755e35 | 5195 | case IORING_OP_ASYNC_CANCEL: |
3529d8c2 JA |
5196 | if (sqe) { |
5197 | ret = io_async_cancel_prep(req, sqe); | |
5198 | if (ret) | |
5199 | break; | |
5200 | } | |
014db007 | 5201 | ret = io_async_cancel(req); |
62755e35 | 5202 | break; |
d63d1b5e JA |
5203 | case IORING_OP_FALLOCATE: |
5204 | if (sqe) { | |
5205 | ret = io_fallocate_prep(req, sqe); | |
5206 | if (ret) | |
5207 | break; | |
5208 | } | |
014db007 | 5209 | ret = io_fallocate(req, force_nonblock); |
d63d1b5e | 5210 | break; |
15b71abe JA |
5211 | case IORING_OP_OPENAT: |
5212 | if (sqe) { | |
5213 | ret = io_openat_prep(req, sqe); | |
5214 | if (ret) | |
5215 | break; | |
5216 | } | |
014db007 | 5217 | ret = io_openat(req, force_nonblock); |
15b71abe | 5218 | break; |
b5dba59e JA |
5219 | case IORING_OP_CLOSE: |
5220 | if (sqe) { | |
5221 | ret = io_close_prep(req, sqe); | |
5222 | if (ret) | |
5223 | break; | |
5224 | } | |
014db007 | 5225 | ret = io_close(req, force_nonblock); |
b5dba59e | 5226 | break; |
05f3fb3c JA |
5227 | case IORING_OP_FILES_UPDATE: |
5228 | if (sqe) { | |
5229 | ret = io_files_update_prep(req, sqe); | |
5230 | if (ret) | |
5231 | break; | |
5232 | } | |
5233 | ret = io_files_update(req, force_nonblock); | |
5234 | break; | |
eddc7ef5 JA |
5235 | case IORING_OP_STATX: |
5236 | if (sqe) { | |
5237 | ret = io_statx_prep(req, sqe); | |
5238 | if (ret) | |
5239 | break; | |
5240 | } | |
014db007 | 5241 | ret = io_statx(req, force_nonblock); |
eddc7ef5 | 5242 | break; |
4840e418 JA |
5243 | case IORING_OP_FADVISE: |
5244 | if (sqe) { | |
5245 | ret = io_fadvise_prep(req, sqe); | |
5246 | if (ret) | |
5247 | break; | |
5248 | } | |
014db007 | 5249 | ret = io_fadvise(req, force_nonblock); |
4840e418 | 5250 | break; |
c1ca757b JA |
5251 | case IORING_OP_MADVISE: |
5252 | if (sqe) { | |
5253 | ret = io_madvise_prep(req, sqe); | |
5254 | if (ret) | |
5255 | break; | |
5256 | } | |
014db007 | 5257 | ret = io_madvise(req, force_nonblock); |
c1ca757b | 5258 | break; |
cebdb986 JA |
5259 | case IORING_OP_OPENAT2: |
5260 | if (sqe) { | |
5261 | ret = io_openat2_prep(req, sqe); | |
5262 | if (ret) | |
5263 | break; | |
5264 | } | |
014db007 | 5265 | ret = io_openat2(req, force_nonblock); |
cebdb986 | 5266 | break; |
3e4827b0 JA |
5267 | case IORING_OP_EPOLL_CTL: |
5268 | if (sqe) { | |
5269 | ret = io_epoll_ctl_prep(req, sqe); | |
5270 | if (ret) | |
5271 | break; | |
5272 | } | |
014db007 | 5273 | ret = io_epoll_ctl(req, force_nonblock); |
3e4827b0 | 5274 | break; |
7d67af2c PB |
5275 | case IORING_OP_SPLICE: |
5276 | if (sqe) { | |
5277 | ret = io_splice_prep(req, sqe); | |
5278 | if (ret < 0) | |
5279 | break; | |
5280 | } | |
014db007 | 5281 | ret = io_splice(req, force_nonblock); |
7d67af2c | 5282 | break; |
ddf0322d JA |
5283 | case IORING_OP_PROVIDE_BUFFERS: |
5284 | if (sqe) { | |
5285 | ret = io_provide_buffers_prep(req, sqe); | |
5286 | if (ret) | |
5287 | break; | |
5288 | } | |
5289 | ret = io_provide_buffers(req, force_nonblock); | |
5290 | break; | |
067524e9 JA |
5291 | case IORING_OP_REMOVE_BUFFERS: |
5292 | if (sqe) { | |
5293 | ret = io_remove_buffers_prep(req, sqe); | |
5294 | if (ret) | |
5295 | break; | |
5296 | } | |
5297 | ret = io_remove_buffers(req, force_nonblock); | |
3e4827b0 | 5298 | break; |
2b188cc1 JA |
5299 | default: |
5300 | ret = -EINVAL; | |
5301 | break; | |
5302 | } | |
5303 | ||
def596e9 JA |
5304 | if (ret) |
5305 | return ret; | |
5306 | ||
5307 | if (ctx->flags & IORING_SETUP_IOPOLL) { | |
11ba820b JA |
5308 | const bool in_async = io_wq_current_is_worker(); |
5309 | ||
9e645e11 | 5310 | if (req->result == -EAGAIN) |
def596e9 JA |
5311 | return -EAGAIN; |
5312 | ||
11ba820b JA |
5313 | /* workqueue context doesn't hold uring_lock, grab it now */ |
5314 | if (in_async) | |
5315 | mutex_lock(&ctx->uring_lock); | |
5316 | ||
def596e9 | 5317 | io_iopoll_req_issued(req); |
11ba820b JA |
5318 | |
5319 | if (in_async) | |
5320 | mutex_unlock(&ctx->uring_lock); | |
def596e9 JA |
5321 | } |
5322 | ||
5323 | return 0; | |
2b188cc1 JA |
5324 | } |
5325 | ||
561fb04a | 5326 | static void io_wq_submit_work(struct io_wq_work **workptr) |
2b188cc1 | 5327 | { |
561fb04a | 5328 | struct io_wq_work *work = *workptr; |
2b188cc1 | 5329 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); |
561fb04a | 5330 | int ret = 0; |
2b188cc1 | 5331 | |
0c9d5ccd JA |
5332 | /* if NO_CANCEL is set, we must still run the work */ |
5333 | if ((work->flags & (IO_WQ_WORK_CANCEL|IO_WQ_WORK_NO_CANCEL)) == | |
5334 | IO_WQ_WORK_CANCEL) { | |
561fb04a | 5335 | ret = -ECANCELED; |
0c9d5ccd | 5336 | } |
31b51510 | 5337 | |
561fb04a | 5338 | if (!ret) { |
561fb04a | 5339 | do { |
014db007 | 5340 | ret = io_issue_sqe(req, NULL, false); |
561fb04a JA |
5341 | /* |
5342 | * We can get EAGAIN for polled IO even though we're | |
5343 | * forcing a sync submission from here, since we can't | |
5344 | * wait for request slots on the block side. | |
5345 | */ | |
5346 | if (ret != -EAGAIN) | |
5347 | break; | |
5348 | cond_resched(); | |
5349 | } while (1); | |
5350 | } | |
31b51510 | 5351 | |
561fb04a | 5352 | if (ret) { |
4e88d6e7 | 5353 | req_set_fail_links(req); |
78e19bbe | 5354 | io_cqring_add_event(req, ret); |
817869d2 | 5355 | io_put_req(req); |
edafccee | 5356 | } |
2b188cc1 | 5357 | |
e9fd9396 | 5358 | io_steal_work(req, workptr); |
2b188cc1 JA |
5359 | } |
5360 | ||
65e19f54 JA |
5361 | static inline struct file *io_file_from_index(struct io_ring_ctx *ctx, |
5362 | int index) | |
5363 | { | |
5364 | struct fixed_file_table *table; | |
5365 | ||
05f3fb3c JA |
5366 | table = &ctx->file_data->table[index >> IORING_FILE_TABLE_SHIFT]; |
5367 | return table->files[index & IORING_FILE_TABLE_MASK];; | |
65e19f54 JA |
5368 | } |
5369 | ||
8da11c19 PB |
5370 | static int io_file_get(struct io_submit_state *state, struct io_kiocb *req, |
5371 | int fd, struct file **out_file, bool fixed) | |
09bb8394 | 5372 | { |
a197f664 | 5373 | struct io_ring_ctx *ctx = req->ctx; |
8da11c19 | 5374 | struct file *file; |
09bb8394 | 5375 | |
8da11c19 | 5376 | if (fixed) { |
05f3fb3c | 5377 | if (unlikely(!ctx->file_data || |
09bb8394 JA |
5378 | (unsigned) fd >= ctx->nr_user_files)) |
5379 | return -EBADF; | |
b7620121 | 5380 | fd = array_index_nospec(fd, ctx->nr_user_files); |
8da11c19 PB |
5381 | file = io_file_from_index(ctx, fd); |
5382 | if (!file) | |
08a45173 | 5383 | return -EBADF; |
05589553 XW |
5384 | req->fixed_file_refs = ctx->file_data->cur_refs; |
5385 | percpu_ref_get(req->fixed_file_refs); | |
09bb8394 | 5386 | } else { |
c826bd7a | 5387 | trace_io_uring_file_get(ctx, fd); |
8da11c19 PB |
5388 | file = __io_file_get(state, fd); |
5389 | if (unlikely(!file)) | |
09bb8394 JA |
5390 | return -EBADF; |
5391 | } | |
5392 | ||
8da11c19 | 5393 | *out_file = file; |
09bb8394 JA |
5394 | return 0; |
5395 | } | |
5396 | ||
8da11c19 | 5397 | static int io_req_set_file(struct io_submit_state *state, struct io_kiocb *req, |
63ff8223 | 5398 | int fd) |
8da11c19 | 5399 | { |
8da11c19 PB |
5400 | bool fixed; |
5401 | ||
63ff8223 | 5402 | fixed = (req->flags & REQ_F_FIXED_FILE) != 0; |
8da11c19 PB |
5403 | if (unlikely(!fixed && req->needs_fixed_file)) |
5404 | return -EBADF; | |
5405 | ||
5406 | return io_file_get(state, req, fd, &req->file, fixed); | |
5407 | } | |
5408 | ||
a197f664 | 5409 | static int io_grab_files(struct io_kiocb *req) |
fcb323cc JA |
5410 | { |
5411 | int ret = -EBADF; | |
a197f664 | 5412 | struct io_ring_ctx *ctx = req->ctx; |
fcb323cc | 5413 | |
5b0bbee4 | 5414 | if (req->work.files || (req->flags & REQ_F_NO_FILE_TABLE)) |
f86cd20c | 5415 | return 0; |
b14cca0c | 5416 | if (!ctx->ring_file) |
b5dba59e JA |
5417 | return -EBADF; |
5418 | ||
fcb323cc JA |
5419 | rcu_read_lock(); |
5420 | spin_lock_irq(&ctx->inflight_lock); | |
5421 | /* | |
5422 | * We use the f_ops->flush() handler to ensure that we can flush | |
5423 | * out work accessing these files if the fd is closed. Check if | |
5424 | * the fd has changed since we started down this path, and disallow | |
5425 | * this operation if it has. | |
5426 | */ | |
b14cca0c | 5427 | if (fcheck(ctx->ring_fd) == ctx->ring_file) { |
fcb323cc JA |
5428 | list_add(&req->inflight_entry, &ctx->inflight_list); |
5429 | req->flags |= REQ_F_INFLIGHT; | |
5430 | req->work.files = current->files; | |
5431 | ret = 0; | |
5432 | } | |
5433 | spin_unlock_irq(&ctx->inflight_lock); | |
5434 | rcu_read_unlock(); | |
5435 | ||
5436 | return ret; | |
5437 | } | |
5438 | ||
2665abfd | 5439 | static enum hrtimer_restart io_link_timeout_fn(struct hrtimer *timer) |
2b188cc1 | 5440 | { |
ad8a48ac JA |
5441 | struct io_timeout_data *data = container_of(timer, |
5442 | struct io_timeout_data, timer); | |
5443 | struct io_kiocb *req = data->req; | |
2665abfd JA |
5444 | struct io_ring_ctx *ctx = req->ctx; |
5445 | struct io_kiocb *prev = NULL; | |
5446 | unsigned long flags; | |
2665abfd JA |
5447 | |
5448 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
5449 | ||
5450 | /* | |
5451 | * We don't expect the list to be empty, that will only happen if we | |
5452 | * race with the completion of the linked work. | |
5453 | */ | |
4493233e PB |
5454 | if (!list_empty(&req->link_list)) { |
5455 | prev = list_entry(req->link_list.prev, struct io_kiocb, | |
5456 | link_list); | |
5d960724 | 5457 | if (refcount_inc_not_zero(&prev->refs)) { |
4493233e | 5458 | list_del_init(&req->link_list); |
5d960724 JA |
5459 | prev->flags &= ~REQ_F_LINK_TIMEOUT; |
5460 | } else | |
76a46e06 | 5461 | prev = NULL; |
2665abfd JA |
5462 | } |
5463 | ||
5464 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
5465 | ||
5466 | if (prev) { | |
4e88d6e7 | 5467 | req_set_fail_links(prev); |
014db007 | 5468 | io_async_find_and_cancel(ctx, req, prev->user_data, -ETIME); |
76a46e06 | 5469 | io_put_req(prev); |
47f46768 JA |
5470 | } else { |
5471 | io_cqring_add_event(req, -ETIME); | |
5472 | io_put_req(req); | |
2665abfd | 5473 | } |
2665abfd JA |
5474 | return HRTIMER_NORESTART; |
5475 | } | |
5476 | ||
ad8a48ac | 5477 | static void io_queue_linked_timeout(struct io_kiocb *req) |
2665abfd | 5478 | { |
76a46e06 | 5479 | struct io_ring_ctx *ctx = req->ctx; |
2665abfd | 5480 | |
76a46e06 JA |
5481 | /* |
5482 | * If the list is now empty, then our linked request finished before | |
5483 | * we got a chance to setup the timer | |
5484 | */ | |
5485 | spin_lock_irq(&ctx->completion_lock); | |
4493233e | 5486 | if (!list_empty(&req->link_list)) { |
2d28390a | 5487 | struct io_timeout_data *data = &req->io->timeout; |
94ae5e77 | 5488 | |
ad8a48ac JA |
5489 | data->timer.function = io_link_timeout_fn; |
5490 | hrtimer_start(&data->timer, timespec64_to_ktime(data->ts), | |
5491 | data->mode); | |
2665abfd | 5492 | } |
76a46e06 | 5493 | spin_unlock_irq(&ctx->completion_lock); |
2665abfd | 5494 | |
2665abfd | 5495 | /* drop submission reference */ |
76a46e06 JA |
5496 | io_put_req(req); |
5497 | } | |
2665abfd | 5498 | |
ad8a48ac | 5499 | static struct io_kiocb *io_prep_linked_timeout(struct io_kiocb *req) |
2665abfd JA |
5500 | { |
5501 | struct io_kiocb *nxt; | |
5502 | ||
dea3b49c | 5503 | if (!(req->flags & REQ_F_LINK_HEAD)) |
2665abfd | 5504 | return NULL; |
d7718a9d JA |
5505 | /* for polled retry, if flag is set, we already went through here */ |
5506 | if (req->flags & REQ_F_POLLED) | |
5507 | return NULL; | |
2665abfd | 5508 | |
4493233e PB |
5509 | nxt = list_first_entry_or_null(&req->link_list, struct io_kiocb, |
5510 | link_list); | |
d625c6ee | 5511 | if (!nxt || nxt->opcode != IORING_OP_LINK_TIMEOUT) |
76a46e06 | 5512 | return NULL; |
2665abfd | 5513 | |
76a46e06 | 5514 | req->flags |= REQ_F_LINK_TIMEOUT; |
76a46e06 | 5515 | return nxt; |
2665abfd JA |
5516 | } |
5517 | ||
3529d8c2 | 5518 | static void __io_queue_sqe(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
2b188cc1 | 5519 | { |
4a0a7a18 | 5520 | struct io_kiocb *linked_timeout; |
4bc4494e | 5521 | struct io_kiocb *nxt; |
193155c8 | 5522 | const struct cred *old_creds = NULL; |
e0c5c576 | 5523 | int ret; |
2b188cc1 | 5524 | |
4a0a7a18 JA |
5525 | again: |
5526 | linked_timeout = io_prep_linked_timeout(req); | |
5527 | ||
193155c8 JA |
5528 | if (req->work.creds && req->work.creds != current_cred()) { |
5529 | if (old_creds) | |
5530 | revert_creds(old_creds); | |
5531 | if (old_creds == req->work.creds) | |
5532 | old_creds = NULL; /* restored original creds */ | |
5533 | else | |
5534 | old_creds = override_creds(req->work.creds); | |
5535 | } | |
5536 | ||
014db007 | 5537 | ret = io_issue_sqe(req, sqe, true); |
491381ce JA |
5538 | |
5539 | /* | |
5540 | * We async punt it if the file wasn't marked NOWAIT, or if the file | |
5541 | * doesn't support non-blocking read/write attempts | |
5542 | */ | |
5543 | if (ret == -EAGAIN && (!(req->flags & REQ_F_NOWAIT) || | |
5544 | (req->flags & REQ_F_MUST_PUNT))) { | |
d7718a9d JA |
5545 | if (io_arm_poll_handler(req)) { |
5546 | if (linked_timeout) | |
5547 | io_queue_linked_timeout(linked_timeout); | |
4bc4494e | 5548 | goto exit; |
d7718a9d | 5549 | } |
86a761f8 | 5550 | punt: |
f86cd20c | 5551 | if (io_op_defs[req->opcode].file_table) { |
bbad27b2 PB |
5552 | ret = io_grab_files(req); |
5553 | if (ret) | |
5554 | goto err; | |
2b188cc1 | 5555 | } |
bbad27b2 PB |
5556 | |
5557 | /* | |
5558 | * Queued up for async execution, worker will release | |
5559 | * submit reference when the iocb is actually submitted. | |
5560 | */ | |
5561 | io_queue_async_work(req); | |
4bc4494e | 5562 | goto exit; |
2b188cc1 | 5563 | } |
e65ef56d | 5564 | |
fcb323cc | 5565 | err: |
4bc4494e | 5566 | nxt = NULL; |
76a46e06 | 5567 | /* drop submission reference */ |
2a44f467 | 5568 | io_put_req_find_next(req, &nxt); |
e65ef56d | 5569 | |
f9bd67f6 | 5570 | if (linked_timeout) { |
76a46e06 | 5571 | if (!ret) |
f9bd67f6 | 5572 | io_queue_linked_timeout(linked_timeout); |
76a46e06 | 5573 | else |
f9bd67f6 | 5574 | io_put_req(linked_timeout); |
76a46e06 JA |
5575 | } |
5576 | ||
e65ef56d | 5577 | /* and drop final reference, if we failed */ |
9e645e11 | 5578 | if (ret) { |
78e19bbe | 5579 | io_cqring_add_event(req, ret); |
4e88d6e7 | 5580 | req_set_fail_links(req); |
e65ef56d | 5581 | io_put_req(req); |
9e645e11 | 5582 | } |
4a0a7a18 JA |
5583 | if (nxt) { |
5584 | req = nxt; | |
86a761f8 PB |
5585 | |
5586 | if (req->flags & REQ_F_FORCE_ASYNC) | |
5587 | goto punt; | |
4a0a7a18 JA |
5588 | goto again; |
5589 | } | |
4bc4494e | 5590 | exit: |
193155c8 JA |
5591 | if (old_creds) |
5592 | revert_creds(old_creds); | |
2b188cc1 JA |
5593 | } |
5594 | ||
3529d8c2 | 5595 | static void io_queue_sqe(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
4fe2c963 JL |
5596 | { |
5597 | int ret; | |
5598 | ||
3529d8c2 | 5599 | ret = io_req_defer(req, sqe); |
4fe2c963 JL |
5600 | if (ret) { |
5601 | if (ret != -EIOCBQUEUED) { | |
1118591a | 5602 | fail_req: |
78e19bbe | 5603 | io_cqring_add_event(req, ret); |
4e88d6e7 | 5604 | req_set_fail_links(req); |
78e19bbe | 5605 | io_double_put_req(req); |
4fe2c963 | 5606 | } |
2550878f | 5607 | } else if (req->flags & REQ_F_FORCE_ASYNC) { |
1118591a PB |
5608 | ret = io_req_defer_prep(req, sqe); |
5609 | if (unlikely(ret < 0)) | |
5610 | goto fail_req; | |
ce35a47a JA |
5611 | /* |
5612 | * Never try inline submit of IOSQE_ASYNC is set, go straight | |
5613 | * to async execution. | |
5614 | */ | |
5615 | req->work.flags |= IO_WQ_WORK_CONCURRENT; | |
5616 | io_queue_async_work(req); | |
5617 | } else { | |
3529d8c2 | 5618 | __io_queue_sqe(req, sqe); |
ce35a47a | 5619 | } |
4fe2c963 JL |
5620 | } |
5621 | ||
1b4a51b6 | 5622 | static inline void io_queue_link_head(struct io_kiocb *req) |
4fe2c963 | 5623 | { |
94ae5e77 | 5624 | if (unlikely(req->flags & REQ_F_FAIL_LINK)) { |
1b4a51b6 PB |
5625 | io_cqring_add_event(req, -ECANCELED); |
5626 | io_double_put_req(req); | |
5627 | } else | |
3529d8c2 | 5628 | io_queue_sqe(req, NULL); |
4fe2c963 JL |
5629 | } |
5630 | ||
1d4240cc | 5631 | static int io_submit_sqe(struct io_kiocb *req, const struct io_uring_sqe *sqe, |
3529d8c2 | 5632 | struct io_submit_state *state, struct io_kiocb **link) |
9e645e11 | 5633 | { |
a197f664 | 5634 | struct io_ring_ctx *ctx = req->ctx; |
ef4ff581 | 5635 | int ret; |
9e645e11 | 5636 | |
9e645e11 JA |
5637 | /* |
5638 | * If we already have a head request, queue this one for async | |
5639 | * submittal once the head completes. If we don't have a head but | |
5640 | * IOSQE_IO_LINK is set in the sqe, start a new head. This one will be | |
5641 | * submitted sync once the chain is complete. If none of those | |
5642 | * conditions are true (normal request), then just queue it. | |
5643 | */ | |
5644 | if (*link) { | |
9d76377f | 5645 | struct io_kiocb *head = *link; |
4e88d6e7 | 5646 | |
8cdf2193 PB |
5647 | /* |
5648 | * Taking sequential execution of a link, draining both sides | |
5649 | * of the link also fullfils IOSQE_IO_DRAIN semantics for all | |
5650 | * requests in the link. So, it drains the head and the | |
5651 | * next after the link request. The last one is done via | |
5652 | * drain_next flag to persist the effect across calls. | |
5653 | */ | |
ef4ff581 | 5654 | if (req->flags & REQ_F_IO_DRAIN) { |
711be031 PB |
5655 | head->flags |= REQ_F_IO_DRAIN; |
5656 | ctx->drain_next = 1; | |
5657 | } | |
1d4240cc PB |
5658 | if (io_alloc_async_ctx(req)) |
5659 | return -EAGAIN; | |
9e645e11 | 5660 | |
3529d8c2 | 5661 | ret = io_req_defer_prep(req, sqe); |
2d28390a | 5662 | if (ret) { |
4e88d6e7 | 5663 | /* fail even hard links since we don't submit */ |
9d76377f | 5664 | head->flags |= REQ_F_FAIL_LINK; |
1d4240cc | 5665 | return ret; |
2d28390a | 5666 | } |
9d76377f PB |
5667 | trace_io_uring_link(ctx, req, head); |
5668 | list_add_tail(&req->link_list, &head->link_list); | |
32fe525b PB |
5669 | |
5670 | /* last request of a link, enqueue the link */ | |
ef4ff581 | 5671 | if (!(req->flags & (REQ_F_LINK | REQ_F_HARDLINK))) { |
32fe525b PB |
5672 | io_queue_link_head(head); |
5673 | *link = NULL; | |
5674 | } | |
9e645e11 | 5675 | } else { |
711be031 PB |
5676 | if (unlikely(ctx->drain_next)) { |
5677 | req->flags |= REQ_F_IO_DRAIN; | |
ef4ff581 | 5678 | ctx->drain_next = 0; |
711be031 | 5679 | } |
ef4ff581 | 5680 | if (req->flags & (REQ_F_LINK | REQ_F_HARDLINK)) { |
dea3b49c | 5681 | req->flags |= REQ_F_LINK_HEAD; |
711be031 | 5682 | INIT_LIST_HEAD(&req->link_list); |
f1d96a8f | 5683 | |
1d4240cc PB |
5684 | if (io_alloc_async_ctx(req)) |
5685 | return -EAGAIN; | |
5686 | ||
711be031 PB |
5687 | ret = io_req_defer_prep(req, sqe); |
5688 | if (ret) | |
5689 | req->flags |= REQ_F_FAIL_LINK; | |
5690 | *link = req; | |
5691 | } else { | |
5692 | io_queue_sqe(req, sqe); | |
5693 | } | |
9e645e11 | 5694 | } |
2e6e1fde | 5695 | |
1d4240cc | 5696 | return 0; |
9e645e11 JA |
5697 | } |
5698 | ||
9a56a232 JA |
5699 | /* |
5700 | * Batched submission is done, ensure local IO is flushed out. | |
5701 | */ | |
5702 | static void io_submit_state_end(struct io_submit_state *state) | |
5703 | { | |
5704 | blk_finish_plug(&state->plug); | |
3d6770fb | 5705 | io_file_put(state); |
2579f913 | 5706 | if (state->free_reqs) |
6c8a3134 | 5707 | kmem_cache_free_bulk(req_cachep, state->free_reqs, state->reqs); |
9a56a232 JA |
5708 | } |
5709 | ||
5710 | /* | |
5711 | * Start submission side cache. | |
5712 | */ | |
5713 | static void io_submit_state_start(struct io_submit_state *state, | |
22efde59 | 5714 | unsigned int max_ios) |
9a56a232 JA |
5715 | { |
5716 | blk_start_plug(&state->plug); | |
2579f913 | 5717 | state->free_reqs = 0; |
9a56a232 JA |
5718 | state->file = NULL; |
5719 | state->ios_left = max_ios; | |
5720 | } | |
5721 | ||
2b188cc1 JA |
5722 | static void io_commit_sqring(struct io_ring_ctx *ctx) |
5723 | { | |
75b28aff | 5724 | struct io_rings *rings = ctx->rings; |
2b188cc1 | 5725 | |
caf582c6 PB |
5726 | /* |
5727 | * Ensure any loads from the SQEs are done at this point, | |
5728 | * since once we write the new head, the application could | |
5729 | * write new data to them. | |
5730 | */ | |
5731 | smp_store_release(&rings->sq.head, ctx->cached_sq_head); | |
2b188cc1 JA |
5732 | } |
5733 | ||
2b188cc1 | 5734 | /* |
3529d8c2 | 5735 | * Fetch an sqe, if one is available. Note that sqe_ptr will point to memory |
2b188cc1 JA |
5736 | * that is mapped by userspace. This means that care needs to be taken to |
5737 | * ensure that reads are stable, as we cannot rely on userspace always | |
5738 | * being a good citizen. If members of the sqe are validated and then later | |
5739 | * used, it's important that those reads are done through READ_ONCE() to | |
5740 | * prevent a re-load down the line. | |
5741 | */ | |
709b302f | 5742 | static const struct io_uring_sqe *io_get_sqe(struct io_ring_ctx *ctx) |
2b188cc1 | 5743 | { |
75b28aff | 5744 | u32 *sq_array = ctx->sq_array; |
2b188cc1 JA |
5745 | unsigned head; |
5746 | ||
5747 | /* | |
5748 | * The cached sq head (or cq tail) serves two purposes: | |
5749 | * | |
5750 | * 1) allows us to batch the cost of updating the user visible | |
5751 | * head updates. | |
5752 | * 2) allows the kernel side to track the head on its own, even | |
5753 | * though the application is the one updating it. | |
5754 | */ | |
ee7d46d9 | 5755 | head = READ_ONCE(sq_array[ctx->cached_sq_head & ctx->sq_mask]); |
709b302f PB |
5756 | if (likely(head < ctx->sq_entries)) |
5757 | return &ctx->sq_sqes[head]; | |
2b188cc1 JA |
5758 | |
5759 | /* drop invalid entries */ | |
498ccd9e | 5760 | ctx->cached_sq_dropped++; |
ee7d46d9 | 5761 | WRITE_ONCE(ctx->rings->sq_dropped, ctx->cached_sq_dropped); |
709b302f PB |
5762 | return NULL; |
5763 | } | |
5764 | ||
5765 | static inline void io_consume_sqe(struct io_ring_ctx *ctx) | |
5766 | { | |
5767 | ctx->cached_sq_head++; | |
2b188cc1 JA |
5768 | } |
5769 | ||
ef4ff581 PB |
5770 | #define SQE_VALID_FLAGS (IOSQE_FIXED_FILE|IOSQE_IO_DRAIN|IOSQE_IO_LINK| \ |
5771 | IOSQE_IO_HARDLINK | IOSQE_ASYNC | \ | |
5772 | IOSQE_BUFFER_SELECT) | |
5773 | ||
5774 | static int io_init_req(struct io_ring_ctx *ctx, struct io_kiocb *req, | |
5775 | const struct io_uring_sqe *sqe, | |
5776 | struct io_submit_state *state, bool async) | |
0553b8bd | 5777 | { |
ef4ff581 | 5778 | unsigned int sqe_flags; |
63ff8223 | 5779 | int id; |
ef4ff581 | 5780 | |
0553b8bd PB |
5781 | /* |
5782 | * All io need record the previous position, if LINK vs DARIN, | |
5783 | * it can be used to mark the position of the first IO in the | |
5784 | * link list. | |
5785 | */ | |
31af27c7 | 5786 | req->sequence = ctx->cached_sq_head - ctx->cached_sq_dropped; |
0553b8bd PB |
5787 | req->opcode = READ_ONCE(sqe->opcode); |
5788 | req->user_data = READ_ONCE(sqe->user_data); | |
5789 | req->io = NULL; | |
5790 | req->file = NULL; | |
5791 | req->ctx = ctx; | |
5792 | req->flags = 0; | |
5793 | /* one is dropped after submission, the other at completion */ | |
5794 | refcount_set(&req->refs, 2); | |
5795 | req->task = NULL; | |
5796 | req->result = 0; | |
ef4ff581 | 5797 | req->needs_fixed_file = async; |
0553b8bd | 5798 | INIT_IO_WORK(&req->work, io_wq_submit_work); |
ef4ff581 PB |
5799 | |
5800 | if (unlikely(req->opcode >= IORING_OP_LAST)) | |
5801 | return -EINVAL; | |
5802 | ||
5803 | if (io_op_defs[req->opcode].needs_mm && !current->mm) { | |
5804 | if (unlikely(!mmget_not_zero(ctx->sqo_mm))) | |
5805 | return -EFAULT; | |
5806 | use_mm(ctx->sqo_mm); | |
5807 | } | |
5808 | ||
5809 | sqe_flags = READ_ONCE(sqe->flags); | |
5810 | /* enforce forwards compatibility on users */ | |
5811 | if (unlikely(sqe_flags & ~SQE_VALID_FLAGS)) | |
5812 | return -EINVAL; | |
5813 | ||
5814 | if ((sqe_flags & IOSQE_BUFFER_SELECT) && | |
5815 | !io_op_defs[req->opcode].buffer_select) | |
5816 | return -EOPNOTSUPP; | |
5817 | ||
5818 | id = READ_ONCE(sqe->personality); | |
5819 | if (id) { | |
5820 | req->work.creds = idr_find(&ctx->personality_idr, id); | |
5821 | if (unlikely(!req->work.creds)) | |
5822 | return -EINVAL; | |
5823 | get_cred(req->work.creds); | |
5824 | } | |
5825 | ||
5826 | /* same numerical values with corresponding REQ_F_*, safe to copy */ | |
5827 | req->flags |= sqe_flags & (IOSQE_IO_DRAIN | IOSQE_IO_HARDLINK | | |
5828 | IOSQE_ASYNC | IOSQE_FIXED_FILE | | |
5829 | IOSQE_BUFFER_SELECT | IOSQE_IO_LINK); | |
5830 | ||
63ff8223 JA |
5831 | if (!io_op_defs[req->opcode].needs_file) |
5832 | return 0; | |
5833 | ||
5834 | return io_req_set_file(state, req, READ_ONCE(sqe->fd)); | |
0553b8bd PB |
5835 | } |
5836 | ||
fb5ccc98 | 5837 | static int io_submit_sqes(struct io_ring_ctx *ctx, unsigned int nr, |
bf9c2f1c | 5838 | struct file *ring_file, int ring_fd, bool async) |
6c271ce2 JA |
5839 | { |
5840 | struct io_submit_state state, *statep = NULL; | |
9e645e11 | 5841 | struct io_kiocb *link = NULL; |
9e645e11 | 5842 | int i, submitted = 0; |
6c271ce2 | 5843 | |
c4a2ed72 | 5844 | /* if we have a backlog and couldn't flush it all, return BUSY */ |
ad3eb2c8 JA |
5845 | if (test_bit(0, &ctx->sq_check_overflow)) { |
5846 | if (!list_empty(&ctx->cq_overflow_list) && | |
5847 | !io_cqring_overflow_flush(ctx, false)) | |
5848 | return -EBUSY; | |
5849 | } | |
6c271ce2 | 5850 | |
ee7d46d9 PB |
5851 | /* make sure SQ entry isn't read before tail */ |
5852 | nr = min3(nr, ctx->sq_entries, io_sqring_entries(ctx)); | |
9ef4f124 | 5853 | |
2b85edfc PB |
5854 | if (!percpu_ref_tryget_many(&ctx->refs, nr)) |
5855 | return -EAGAIN; | |
6c271ce2 JA |
5856 | |
5857 | if (nr > IO_PLUG_THRESHOLD) { | |
22efde59 | 5858 | io_submit_state_start(&state, nr); |
6c271ce2 JA |
5859 | statep = &state; |
5860 | } | |
5861 | ||
b14cca0c PB |
5862 | ctx->ring_fd = ring_fd; |
5863 | ctx->ring_file = ring_file; | |
5864 | ||
6c271ce2 | 5865 | for (i = 0; i < nr; i++) { |
3529d8c2 | 5866 | const struct io_uring_sqe *sqe; |
196be95c | 5867 | struct io_kiocb *req; |
1cb1edb2 | 5868 | int err; |
fb5ccc98 | 5869 | |
b1e50e54 PB |
5870 | sqe = io_get_sqe(ctx); |
5871 | if (unlikely(!sqe)) { | |
5872 | io_consume_sqe(ctx); | |
5873 | break; | |
5874 | } | |
0553b8bd | 5875 | req = io_alloc_req(ctx, statep); |
196be95c PB |
5876 | if (unlikely(!req)) { |
5877 | if (!submitted) | |
5878 | submitted = -EAGAIN; | |
fb5ccc98 | 5879 | break; |
196be95c | 5880 | } |
fb5ccc98 | 5881 | |
ef4ff581 | 5882 | err = io_init_req(ctx, req, sqe, statep, async); |
709b302f | 5883 | io_consume_sqe(ctx); |
d3656344 JA |
5884 | /* will complete beyond this point, count as submitted */ |
5885 | submitted++; | |
5886 | ||
ef4ff581 | 5887 | if (unlikely(err)) { |
1cb1edb2 PB |
5888 | fail_req: |
5889 | io_cqring_add_event(req, err); | |
d3656344 | 5890 | io_double_put_req(req); |
196be95c PB |
5891 | break; |
5892 | } | |
fb5ccc98 | 5893 | |
354420f7 JA |
5894 | trace_io_uring_submit_sqe(ctx, req->opcode, req->user_data, |
5895 | true, async); | |
1d4240cc PB |
5896 | err = io_submit_sqe(req, sqe, statep, &link); |
5897 | if (err) | |
5898 | goto fail_req; | |
6c271ce2 JA |
5899 | } |
5900 | ||
9466f437 PB |
5901 | if (unlikely(submitted != nr)) { |
5902 | int ref_used = (submitted == -EAGAIN) ? 0 : submitted; | |
5903 | ||
5904 | percpu_ref_put_many(&ctx->refs, nr - ref_used); | |
5905 | } | |
9e645e11 | 5906 | if (link) |
1b4a51b6 | 5907 | io_queue_link_head(link); |
6c271ce2 JA |
5908 | if (statep) |
5909 | io_submit_state_end(&state); | |
5910 | ||
ae9428ca PB |
5911 | /* Commit SQ ring head once we've consumed and submitted all SQEs */ |
5912 | io_commit_sqring(ctx); | |
5913 | ||
6c271ce2 JA |
5914 | return submitted; |
5915 | } | |
5916 | ||
bf9c2f1c PB |
5917 | static inline void io_sq_thread_drop_mm(struct io_ring_ctx *ctx) |
5918 | { | |
5919 | struct mm_struct *mm = current->mm; | |
5920 | ||
5921 | if (mm) { | |
5922 | unuse_mm(mm); | |
5923 | mmput(mm); | |
5924 | } | |
5925 | } | |
5926 | ||
6c271ce2 JA |
5927 | static int io_sq_thread(void *data) |
5928 | { | |
6c271ce2 | 5929 | struct io_ring_ctx *ctx = data; |
181e448d | 5930 | const struct cred *old_cred; |
6c271ce2 JA |
5931 | mm_segment_t old_fs; |
5932 | DEFINE_WAIT(wait); | |
6c271ce2 | 5933 | unsigned long timeout; |
bdcd3eab | 5934 | int ret = 0; |
6c271ce2 | 5935 | |
206aefde | 5936 | complete(&ctx->completions[1]); |
a4c0b3de | 5937 | |
6c271ce2 JA |
5938 | old_fs = get_fs(); |
5939 | set_fs(USER_DS); | |
181e448d | 5940 | old_cred = override_creds(ctx->creds); |
6c271ce2 | 5941 | |
bdcd3eab | 5942 | timeout = jiffies + ctx->sq_thread_idle; |
2bbcd6d3 | 5943 | while (!kthread_should_park()) { |
fb5ccc98 | 5944 | unsigned int to_submit; |
6c271ce2 | 5945 | |
bdcd3eab | 5946 | if (!list_empty(&ctx->poll_list)) { |
6c271ce2 JA |
5947 | unsigned nr_events = 0; |
5948 | ||
bdcd3eab XW |
5949 | mutex_lock(&ctx->uring_lock); |
5950 | if (!list_empty(&ctx->poll_list)) | |
5951 | io_iopoll_getevents(ctx, &nr_events, 0); | |
5952 | else | |
6c271ce2 | 5953 | timeout = jiffies + ctx->sq_thread_idle; |
bdcd3eab | 5954 | mutex_unlock(&ctx->uring_lock); |
6c271ce2 JA |
5955 | } |
5956 | ||
fb5ccc98 | 5957 | to_submit = io_sqring_entries(ctx); |
c1edbf5f JA |
5958 | |
5959 | /* | |
5960 | * If submit got -EBUSY, flag us as needing the application | |
5961 | * to enter the kernel to reap and flush events. | |
5962 | */ | |
5963 | if (!to_submit || ret == -EBUSY) { | |
7143b5ac SG |
5964 | /* |
5965 | * Drop cur_mm before scheduling, we can't hold it for | |
5966 | * long periods (or over schedule()). Do this before | |
5967 | * adding ourselves to the waitqueue, as the unuse/drop | |
5968 | * may sleep. | |
5969 | */ | |
bf9c2f1c | 5970 | io_sq_thread_drop_mm(ctx); |
7143b5ac | 5971 | |
6c271ce2 JA |
5972 | /* |
5973 | * We're polling. If we're within the defined idle | |
5974 | * period, then let us spin without work before going | |
c1edbf5f JA |
5975 | * to sleep. The exception is if we got EBUSY doing |
5976 | * more IO, we should wait for the application to | |
5977 | * reap events and wake us up. | |
6c271ce2 | 5978 | */ |
bdcd3eab | 5979 | if (!list_empty(&ctx->poll_list) || |
df069d80 JA |
5980 | (!time_after(jiffies, timeout) && ret != -EBUSY && |
5981 | !percpu_ref_is_dying(&ctx->refs))) { | |
b41e9852 JA |
5982 | if (current->task_works) |
5983 | task_work_run(); | |
9831a90c | 5984 | cond_resched(); |
6c271ce2 JA |
5985 | continue; |
5986 | } | |
5987 | ||
6c271ce2 JA |
5988 | prepare_to_wait(&ctx->sqo_wait, &wait, |
5989 | TASK_INTERRUPTIBLE); | |
5990 | ||
bdcd3eab XW |
5991 | /* |
5992 | * While doing polled IO, before going to sleep, we need | |
5993 | * to check if there are new reqs added to poll_list, it | |
5994 | * is because reqs may have been punted to io worker and | |
5995 | * will be added to poll_list later, hence check the | |
5996 | * poll_list again. | |
5997 | */ | |
5998 | if ((ctx->flags & IORING_SETUP_IOPOLL) && | |
5999 | !list_empty_careful(&ctx->poll_list)) { | |
6000 | finish_wait(&ctx->sqo_wait, &wait); | |
6001 | continue; | |
6002 | } | |
6003 | ||
6c271ce2 | 6004 | /* Tell userspace we may need a wakeup call */ |
75b28aff | 6005 | ctx->rings->sq_flags |= IORING_SQ_NEED_WAKEUP; |
0d7bae69 SB |
6006 | /* make sure to read SQ tail after writing flags */ |
6007 | smp_mb(); | |
6c271ce2 | 6008 | |
fb5ccc98 | 6009 | to_submit = io_sqring_entries(ctx); |
c1edbf5f | 6010 | if (!to_submit || ret == -EBUSY) { |
2bbcd6d3 | 6011 | if (kthread_should_park()) { |
6c271ce2 JA |
6012 | finish_wait(&ctx->sqo_wait, &wait); |
6013 | break; | |
6014 | } | |
b41e9852 JA |
6015 | if (current->task_works) { |
6016 | task_work_run(); | |
10bea96d | 6017 | finish_wait(&ctx->sqo_wait, &wait); |
b41e9852 JA |
6018 | continue; |
6019 | } | |
6c271ce2 JA |
6020 | if (signal_pending(current)) |
6021 | flush_signals(current); | |
6022 | schedule(); | |
6023 | finish_wait(&ctx->sqo_wait, &wait); | |
6024 | ||
75b28aff | 6025 | ctx->rings->sq_flags &= ~IORING_SQ_NEED_WAKEUP; |
6c271ce2 JA |
6026 | continue; |
6027 | } | |
6028 | finish_wait(&ctx->sqo_wait, &wait); | |
6029 | ||
75b28aff | 6030 | ctx->rings->sq_flags &= ~IORING_SQ_NEED_WAKEUP; |
6c271ce2 JA |
6031 | } |
6032 | ||
8a4955ff | 6033 | mutex_lock(&ctx->uring_lock); |
bf9c2f1c | 6034 | ret = io_submit_sqes(ctx, to_submit, NULL, -1, true); |
8a4955ff | 6035 | mutex_unlock(&ctx->uring_lock); |
bdcd3eab | 6036 | timeout = jiffies + ctx->sq_thread_idle; |
6c271ce2 JA |
6037 | } |
6038 | ||
b41e9852 JA |
6039 | if (current->task_works) |
6040 | task_work_run(); | |
6041 | ||
6c271ce2 | 6042 | set_fs(old_fs); |
bf9c2f1c | 6043 | io_sq_thread_drop_mm(ctx); |
181e448d | 6044 | revert_creds(old_cred); |
06058632 | 6045 | |
2bbcd6d3 | 6046 | kthread_parkme(); |
06058632 | 6047 | |
6c271ce2 JA |
6048 | return 0; |
6049 | } | |
6050 | ||
bda52162 JA |
6051 | struct io_wait_queue { |
6052 | struct wait_queue_entry wq; | |
6053 | struct io_ring_ctx *ctx; | |
6054 | unsigned to_wait; | |
6055 | unsigned nr_timeouts; | |
6056 | }; | |
6057 | ||
1d7bb1d5 | 6058 | static inline bool io_should_wake(struct io_wait_queue *iowq, bool noflush) |
bda52162 JA |
6059 | { |
6060 | struct io_ring_ctx *ctx = iowq->ctx; | |
6061 | ||
6062 | /* | |
d195a66e | 6063 | * Wake up if we have enough events, or if a timeout occurred since we |
bda52162 JA |
6064 | * started waiting. For timeouts, we always want to return to userspace, |
6065 | * regardless of event count. | |
6066 | */ | |
1d7bb1d5 | 6067 | return io_cqring_events(ctx, noflush) >= iowq->to_wait || |
bda52162 JA |
6068 | atomic_read(&ctx->cq_timeouts) != iowq->nr_timeouts; |
6069 | } | |
6070 | ||
6071 | static int io_wake_function(struct wait_queue_entry *curr, unsigned int mode, | |
6072 | int wake_flags, void *key) | |
6073 | { | |
6074 | struct io_wait_queue *iowq = container_of(curr, struct io_wait_queue, | |
6075 | wq); | |
6076 | ||
1d7bb1d5 JA |
6077 | /* use noflush == true, as we can't safely rely on locking context */ |
6078 | if (!io_should_wake(iowq, true)) | |
bda52162 JA |
6079 | return -1; |
6080 | ||
6081 | return autoremove_wake_function(curr, mode, wake_flags, key); | |
6082 | } | |
6083 | ||
2b188cc1 JA |
6084 | /* |
6085 | * Wait until events become available, if we don't already have some. The | |
6086 | * application must reap them itself, as they reside on the shared cq ring. | |
6087 | */ | |
6088 | static int io_cqring_wait(struct io_ring_ctx *ctx, int min_events, | |
6089 | const sigset_t __user *sig, size_t sigsz) | |
6090 | { | |
bda52162 JA |
6091 | struct io_wait_queue iowq = { |
6092 | .wq = { | |
6093 | .private = current, | |
6094 | .func = io_wake_function, | |
6095 | .entry = LIST_HEAD_INIT(iowq.wq.entry), | |
6096 | }, | |
6097 | .ctx = ctx, | |
6098 | .to_wait = min_events, | |
6099 | }; | |
75b28aff | 6100 | struct io_rings *rings = ctx->rings; |
e9ffa5c2 | 6101 | int ret = 0; |
2b188cc1 | 6102 | |
b41e9852 JA |
6103 | do { |
6104 | if (io_cqring_events(ctx, false) >= min_events) | |
6105 | return 0; | |
6106 | if (!current->task_works) | |
6107 | break; | |
6108 | task_work_run(); | |
6109 | } while (1); | |
2b188cc1 JA |
6110 | |
6111 | if (sig) { | |
9e75ad5d AB |
6112 | #ifdef CONFIG_COMPAT |
6113 | if (in_compat_syscall()) | |
6114 | ret = set_compat_user_sigmask((const compat_sigset_t __user *)sig, | |
b772434b | 6115 | sigsz); |
9e75ad5d AB |
6116 | else |
6117 | #endif | |
b772434b | 6118 | ret = set_user_sigmask(sig, sigsz); |
9e75ad5d | 6119 | |
2b188cc1 JA |
6120 | if (ret) |
6121 | return ret; | |
6122 | } | |
6123 | ||
bda52162 | 6124 | iowq.nr_timeouts = atomic_read(&ctx->cq_timeouts); |
c826bd7a | 6125 | trace_io_uring_cqring_wait(ctx, min_events); |
bda52162 JA |
6126 | do { |
6127 | prepare_to_wait_exclusive(&ctx->wait, &iowq.wq, | |
6128 | TASK_INTERRUPTIBLE); | |
b41e9852 JA |
6129 | if (current->task_works) |
6130 | task_work_run(); | |
1d7bb1d5 | 6131 | if (io_should_wake(&iowq, false)) |
bda52162 JA |
6132 | break; |
6133 | schedule(); | |
6134 | if (signal_pending(current)) { | |
e9ffa5c2 | 6135 | ret = -EINTR; |
bda52162 JA |
6136 | break; |
6137 | } | |
6138 | } while (1); | |
6139 | finish_wait(&ctx->wait, &iowq.wq); | |
6140 | ||
e9ffa5c2 | 6141 | restore_saved_sigmask_unless(ret == -EINTR); |
2b188cc1 | 6142 | |
75b28aff | 6143 | return READ_ONCE(rings->cq.head) == READ_ONCE(rings->cq.tail) ? ret : 0; |
2b188cc1 JA |
6144 | } |
6145 | ||
6b06314c JA |
6146 | static void __io_sqe_files_unregister(struct io_ring_ctx *ctx) |
6147 | { | |
6148 | #if defined(CONFIG_UNIX) | |
6149 | if (ctx->ring_sock) { | |
6150 | struct sock *sock = ctx->ring_sock->sk; | |
6151 | struct sk_buff *skb; | |
6152 | ||
6153 | while ((skb = skb_dequeue(&sock->sk_receive_queue)) != NULL) | |
6154 | kfree_skb(skb); | |
6155 | } | |
6156 | #else | |
6157 | int i; | |
6158 | ||
65e19f54 JA |
6159 | for (i = 0; i < ctx->nr_user_files; i++) { |
6160 | struct file *file; | |
6161 | ||
6162 | file = io_file_from_index(ctx, i); | |
6163 | if (file) | |
6164 | fput(file); | |
6165 | } | |
6b06314c JA |
6166 | #endif |
6167 | } | |
6168 | ||
05f3fb3c JA |
6169 | static void io_file_ref_kill(struct percpu_ref *ref) |
6170 | { | |
6171 | struct fixed_file_data *data; | |
6172 | ||
6173 | data = container_of(ref, struct fixed_file_data, refs); | |
6174 | complete(&data->done); | |
6175 | } | |
6176 | ||
6b06314c JA |
6177 | static int io_sqe_files_unregister(struct io_ring_ctx *ctx) |
6178 | { | |
05f3fb3c | 6179 | struct fixed_file_data *data = ctx->file_data; |
05589553 | 6180 | struct fixed_file_ref_node *ref_node = NULL; |
65e19f54 | 6181 | unsigned nr_tables, i; |
05589553 | 6182 | unsigned long flags; |
65e19f54 | 6183 | |
05f3fb3c | 6184 | if (!data) |
6b06314c JA |
6185 | return -ENXIO; |
6186 | ||
05589553 XW |
6187 | spin_lock_irqsave(&data->lock, flags); |
6188 | if (!list_empty(&data->ref_list)) | |
6189 | ref_node = list_first_entry(&data->ref_list, | |
6190 | struct fixed_file_ref_node, node); | |
6191 | spin_unlock_irqrestore(&data->lock, flags); | |
6192 | if (ref_node) | |
6193 | percpu_ref_kill(&ref_node->refs); | |
6194 | ||
6195 | percpu_ref_kill(&data->refs); | |
6196 | ||
6197 | /* wait for all refs nodes to complete */ | |
2faf852d | 6198 | wait_for_completion(&data->done); |
05f3fb3c | 6199 | |
6b06314c | 6200 | __io_sqe_files_unregister(ctx); |
65e19f54 JA |
6201 | nr_tables = DIV_ROUND_UP(ctx->nr_user_files, IORING_MAX_FILES_TABLE); |
6202 | for (i = 0; i < nr_tables; i++) | |
05f3fb3c JA |
6203 | kfree(data->table[i].files); |
6204 | kfree(data->table); | |
05589553 XW |
6205 | percpu_ref_exit(&data->refs); |
6206 | kfree(data); | |
05f3fb3c | 6207 | ctx->file_data = NULL; |
6b06314c JA |
6208 | ctx->nr_user_files = 0; |
6209 | return 0; | |
6210 | } | |
6211 | ||
6c271ce2 JA |
6212 | static void io_sq_thread_stop(struct io_ring_ctx *ctx) |
6213 | { | |
6214 | if (ctx->sqo_thread) { | |
206aefde | 6215 | wait_for_completion(&ctx->completions[1]); |
2bbcd6d3 RP |
6216 | /* |
6217 | * The park is a bit of a work-around, without it we get | |
6218 | * warning spews on shutdown with SQPOLL set and affinity | |
6219 | * set to a single CPU. | |
6220 | */ | |
06058632 | 6221 | kthread_park(ctx->sqo_thread); |
6c271ce2 JA |
6222 | kthread_stop(ctx->sqo_thread); |
6223 | ctx->sqo_thread = NULL; | |
6224 | } | |
6225 | } | |
6226 | ||
6b06314c JA |
6227 | static void io_finish_async(struct io_ring_ctx *ctx) |
6228 | { | |
6c271ce2 JA |
6229 | io_sq_thread_stop(ctx); |
6230 | ||
561fb04a JA |
6231 | if (ctx->io_wq) { |
6232 | io_wq_destroy(ctx->io_wq); | |
6233 | ctx->io_wq = NULL; | |
6b06314c JA |
6234 | } |
6235 | } | |
6236 | ||
6237 | #if defined(CONFIG_UNIX) | |
6b06314c JA |
6238 | /* |
6239 | * Ensure the UNIX gc is aware of our file set, so we are certain that | |
6240 | * the io_uring can be safely unregistered on process exit, even if we have | |
6241 | * loops in the file referencing. | |
6242 | */ | |
6243 | static int __io_sqe_files_scm(struct io_ring_ctx *ctx, int nr, int offset) | |
6244 | { | |
6245 | struct sock *sk = ctx->ring_sock->sk; | |
6246 | struct scm_fp_list *fpl; | |
6247 | struct sk_buff *skb; | |
08a45173 | 6248 | int i, nr_files; |
6b06314c | 6249 | |
6b06314c JA |
6250 | fpl = kzalloc(sizeof(*fpl), GFP_KERNEL); |
6251 | if (!fpl) | |
6252 | return -ENOMEM; | |
6253 | ||
6254 | skb = alloc_skb(0, GFP_KERNEL); | |
6255 | if (!skb) { | |
6256 | kfree(fpl); | |
6257 | return -ENOMEM; | |
6258 | } | |
6259 | ||
6260 | skb->sk = sk; | |
6b06314c | 6261 | |
08a45173 | 6262 | nr_files = 0; |
6b06314c JA |
6263 | fpl->user = get_uid(ctx->user); |
6264 | for (i = 0; i < nr; i++) { | |
65e19f54 JA |
6265 | struct file *file = io_file_from_index(ctx, i + offset); |
6266 | ||
6267 | if (!file) | |
08a45173 | 6268 | continue; |
65e19f54 | 6269 | fpl->fp[nr_files] = get_file(file); |
08a45173 JA |
6270 | unix_inflight(fpl->user, fpl->fp[nr_files]); |
6271 | nr_files++; | |
6b06314c JA |
6272 | } |
6273 | ||
08a45173 JA |
6274 | if (nr_files) { |
6275 | fpl->max = SCM_MAX_FD; | |
6276 | fpl->count = nr_files; | |
6277 | UNIXCB(skb).fp = fpl; | |
05f3fb3c | 6278 | skb->destructor = unix_destruct_scm; |
08a45173 JA |
6279 | refcount_add(skb->truesize, &sk->sk_wmem_alloc); |
6280 | skb_queue_head(&sk->sk_receive_queue, skb); | |
6b06314c | 6281 | |
08a45173 JA |
6282 | for (i = 0; i < nr_files; i++) |
6283 | fput(fpl->fp[i]); | |
6284 | } else { | |
6285 | kfree_skb(skb); | |
6286 | kfree(fpl); | |
6287 | } | |
6b06314c JA |
6288 | |
6289 | return 0; | |
6290 | } | |
6291 | ||
6292 | /* | |
6293 | * If UNIX sockets are enabled, fd passing can cause a reference cycle which | |
6294 | * causes regular reference counting to break down. We rely on the UNIX | |
6295 | * garbage collection to take care of this problem for us. | |
6296 | */ | |
6297 | static int io_sqe_files_scm(struct io_ring_ctx *ctx) | |
6298 | { | |
6299 | unsigned left, total; | |
6300 | int ret = 0; | |
6301 | ||
6302 | total = 0; | |
6303 | left = ctx->nr_user_files; | |
6304 | while (left) { | |
6305 | unsigned this_files = min_t(unsigned, left, SCM_MAX_FD); | |
6b06314c JA |
6306 | |
6307 | ret = __io_sqe_files_scm(ctx, this_files, total); | |
6308 | if (ret) | |
6309 | break; | |
6310 | left -= this_files; | |
6311 | total += this_files; | |
6312 | } | |
6313 | ||
6314 | if (!ret) | |
6315 | return 0; | |
6316 | ||
6317 | while (total < ctx->nr_user_files) { | |
65e19f54 JA |
6318 | struct file *file = io_file_from_index(ctx, total); |
6319 | ||
6320 | if (file) | |
6321 | fput(file); | |
6b06314c JA |
6322 | total++; |
6323 | } | |
6324 | ||
6325 | return ret; | |
6326 | } | |
6327 | #else | |
6328 | static int io_sqe_files_scm(struct io_ring_ctx *ctx) | |
6329 | { | |
6330 | return 0; | |
6331 | } | |
6332 | #endif | |
6333 | ||
65e19f54 JA |
6334 | static int io_sqe_alloc_file_tables(struct io_ring_ctx *ctx, unsigned nr_tables, |
6335 | unsigned nr_files) | |
6336 | { | |
6337 | int i; | |
6338 | ||
6339 | for (i = 0; i < nr_tables; i++) { | |
05f3fb3c | 6340 | struct fixed_file_table *table = &ctx->file_data->table[i]; |
65e19f54 JA |
6341 | unsigned this_files; |
6342 | ||
6343 | this_files = min(nr_files, IORING_MAX_FILES_TABLE); | |
6344 | table->files = kcalloc(this_files, sizeof(struct file *), | |
6345 | GFP_KERNEL); | |
6346 | if (!table->files) | |
6347 | break; | |
6348 | nr_files -= this_files; | |
6349 | } | |
6350 | ||
6351 | if (i == nr_tables) | |
6352 | return 0; | |
6353 | ||
6354 | for (i = 0; i < nr_tables; i++) { | |
05f3fb3c | 6355 | struct fixed_file_table *table = &ctx->file_data->table[i]; |
65e19f54 JA |
6356 | kfree(table->files); |
6357 | } | |
6358 | return 1; | |
6359 | } | |
6360 | ||
05f3fb3c JA |
6361 | static void io_ring_file_put(struct io_ring_ctx *ctx, struct file *file) |
6362 | { | |
6363 | #if defined(CONFIG_UNIX) | |
6364 | struct sock *sock = ctx->ring_sock->sk; | |
6365 | struct sk_buff_head list, *head = &sock->sk_receive_queue; | |
6366 | struct sk_buff *skb; | |
6367 | int i; | |
6368 | ||
6369 | __skb_queue_head_init(&list); | |
6370 | ||
6371 | /* | |
6372 | * Find the skb that holds this file in its SCM_RIGHTS. When found, | |
6373 | * remove this entry and rearrange the file array. | |
6374 | */ | |
6375 | skb = skb_dequeue(head); | |
6376 | while (skb) { | |
6377 | struct scm_fp_list *fp; | |
6378 | ||
6379 | fp = UNIXCB(skb).fp; | |
6380 | for (i = 0; i < fp->count; i++) { | |
6381 | int left; | |
6382 | ||
6383 | if (fp->fp[i] != file) | |
6384 | continue; | |
6385 | ||
6386 | unix_notinflight(fp->user, fp->fp[i]); | |
6387 | left = fp->count - 1 - i; | |
6388 | if (left) { | |
6389 | memmove(&fp->fp[i], &fp->fp[i + 1], | |
6390 | left * sizeof(struct file *)); | |
6391 | } | |
6392 | fp->count--; | |
6393 | if (!fp->count) { | |
6394 | kfree_skb(skb); | |
6395 | skb = NULL; | |
6396 | } else { | |
6397 | __skb_queue_tail(&list, skb); | |
6398 | } | |
6399 | fput(file); | |
6400 | file = NULL; | |
6401 | break; | |
6402 | } | |
6403 | ||
6404 | if (!file) | |
6405 | break; | |
6406 | ||
6407 | __skb_queue_tail(&list, skb); | |
6408 | ||
6409 | skb = skb_dequeue(head); | |
6410 | } | |
6411 | ||
6412 | if (skb_peek(&list)) { | |
6413 | spin_lock_irq(&head->lock); | |
6414 | while ((skb = __skb_dequeue(&list)) != NULL) | |
6415 | __skb_queue_tail(head, skb); | |
6416 | spin_unlock_irq(&head->lock); | |
6417 | } | |
6418 | #else | |
6419 | fput(file); | |
6420 | #endif | |
6421 | } | |
6422 | ||
6423 | struct io_file_put { | |
05589553 | 6424 | struct list_head list; |
05f3fb3c | 6425 | struct file *file; |
05f3fb3c JA |
6426 | }; |
6427 | ||
05589553 | 6428 | static void io_file_put_work(struct work_struct *work) |
65e19f54 | 6429 | { |
05589553 XW |
6430 | struct fixed_file_ref_node *ref_node; |
6431 | struct fixed_file_data *file_data; | |
6432 | struct io_ring_ctx *ctx; | |
05f3fb3c | 6433 | struct io_file_put *pfile, *tmp; |
05589553 | 6434 | unsigned long flags; |
65e19f54 | 6435 | |
05589553 XW |
6436 | ref_node = container_of(work, struct fixed_file_ref_node, work); |
6437 | file_data = ref_node->file_data; | |
6438 | ctx = file_data->ctx; | |
6439 | ||
6440 | list_for_each_entry_safe(pfile, tmp, &ref_node->file_list, list) { | |
6441 | list_del_init(&pfile->list); | |
6442 | io_ring_file_put(ctx, pfile->file); | |
6443 | kfree(pfile); | |
65e19f54 | 6444 | } |
05589553 XW |
6445 | |
6446 | spin_lock_irqsave(&file_data->lock, flags); | |
6447 | list_del_init(&ref_node->node); | |
6448 | spin_unlock_irqrestore(&file_data->lock, flags); | |
6449 | ||
6450 | percpu_ref_exit(&ref_node->refs); | |
6451 | kfree(ref_node); | |
6452 | percpu_ref_put(&file_data->refs); | |
2faf852d | 6453 | } |
65e19f54 | 6454 | |
05589553 | 6455 | static void io_file_data_ref_zero(struct percpu_ref *ref) |
2faf852d | 6456 | { |
05589553 | 6457 | struct fixed_file_ref_node *ref_node; |
65e19f54 | 6458 | |
05589553 XW |
6459 | ref_node = container_of(ref, struct fixed_file_ref_node, refs); |
6460 | ||
6461 | queue_work(system_wq, &ref_node->work); | |
05f3fb3c | 6462 | } |
65e19f54 | 6463 | |
05589553 XW |
6464 | static struct fixed_file_ref_node *alloc_fixed_file_ref_node( |
6465 | struct io_ring_ctx *ctx) | |
05f3fb3c | 6466 | { |
05589553 | 6467 | struct fixed_file_ref_node *ref_node; |
05f3fb3c | 6468 | |
05589553 XW |
6469 | ref_node = kzalloc(sizeof(*ref_node), GFP_KERNEL); |
6470 | if (!ref_node) | |
6471 | return ERR_PTR(-ENOMEM); | |
05f3fb3c | 6472 | |
05589553 XW |
6473 | if (percpu_ref_init(&ref_node->refs, io_file_data_ref_zero, |
6474 | 0, GFP_KERNEL)) { | |
6475 | kfree(ref_node); | |
6476 | return ERR_PTR(-ENOMEM); | |
6477 | } | |
6478 | INIT_LIST_HEAD(&ref_node->node); | |
6479 | INIT_LIST_HEAD(&ref_node->file_list); | |
6480 | INIT_WORK(&ref_node->work, io_file_put_work); | |
6481 | ref_node->file_data = ctx->file_data; | |
6482 | return ref_node; | |
6483 | ||
6484 | } | |
6485 | ||
6486 | static void destroy_fixed_file_ref_node(struct fixed_file_ref_node *ref_node) | |
6487 | { | |
6488 | percpu_ref_exit(&ref_node->refs); | |
6489 | kfree(ref_node); | |
65e19f54 JA |
6490 | } |
6491 | ||
6b06314c JA |
6492 | static int io_sqe_files_register(struct io_ring_ctx *ctx, void __user *arg, |
6493 | unsigned nr_args) | |
6494 | { | |
6495 | __s32 __user *fds = (__s32 __user *) arg; | |
65e19f54 | 6496 | unsigned nr_tables; |
05f3fb3c | 6497 | struct file *file; |
6b06314c JA |
6498 | int fd, ret = 0; |
6499 | unsigned i; | |
05589553 XW |
6500 | struct fixed_file_ref_node *ref_node; |
6501 | unsigned long flags; | |
6b06314c | 6502 | |
05f3fb3c | 6503 | if (ctx->file_data) |
6b06314c JA |
6504 | return -EBUSY; |
6505 | if (!nr_args) | |
6506 | return -EINVAL; | |
6507 | if (nr_args > IORING_MAX_FIXED_FILES) | |
6508 | return -EMFILE; | |
6509 | ||
05f3fb3c JA |
6510 | ctx->file_data = kzalloc(sizeof(*ctx->file_data), GFP_KERNEL); |
6511 | if (!ctx->file_data) | |
6512 | return -ENOMEM; | |
6513 | ctx->file_data->ctx = ctx; | |
6514 | init_completion(&ctx->file_data->done); | |
05589553 | 6515 | INIT_LIST_HEAD(&ctx->file_data->ref_list); |
f7fe9346 | 6516 | spin_lock_init(&ctx->file_data->lock); |
05f3fb3c | 6517 | |
65e19f54 | 6518 | nr_tables = DIV_ROUND_UP(nr_args, IORING_MAX_FILES_TABLE); |
05f3fb3c JA |
6519 | ctx->file_data->table = kcalloc(nr_tables, |
6520 | sizeof(struct fixed_file_table), | |
65e19f54 | 6521 | GFP_KERNEL); |
05f3fb3c JA |
6522 | if (!ctx->file_data->table) { |
6523 | kfree(ctx->file_data); | |
6524 | ctx->file_data = NULL; | |
6b06314c | 6525 | return -ENOMEM; |
05f3fb3c JA |
6526 | } |
6527 | ||
05589553 | 6528 | if (percpu_ref_init(&ctx->file_data->refs, io_file_ref_kill, |
05f3fb3c JA |
6529 | PERCPU_REF_ALLOW_REINIT, GFP_KERNEL)) { |
6530 | kfree(ctx->file_data->table); | |
6531 | kfree(ctx->file_data); | |
6532 | ctx->file_data = NULL; | |
6b06314c | 6533 | return -ENOMEM; |
05f3fb3c | 6534 | } |
6b06314c | 6535 | |
65e19f54 | 6536 | if (io_sqe_alloc_file_tables(ctx, nr_tables, nr_args)) { |
05f3fb3c JA |
6537 | percpu_ref_exit(&ctx->file_data->refs); |
6538 | kfree(ctx->file_data->table); | |
6539 | kfree(ctx->file_data); | |
6540 | ctx->file_data = NULL; | |
65e19f54 JA |
6541 | return -ENOMEM; |
6542 | } | |
6543 | ||
08a45173 | 6544 | for (i = 0; i < nr_args; i++, ctx->nr_user_files++) { |
65e19f54 JA |
6545 | struct fixed_file_table *table; |
6546 | unsigned index; | |
6547 | ||
6b06314c JA |
6548 | ret = -EFAULT; |
6549 | if (copy_from_user(&fd, &fds[i], sizeof(fd))) | |
6550 | break; | |
08a45173 JA |
6551 | /* allow sparse sets */ |
6552 | if (fd == -1) { | |
6553 | ret = 0; | |
6554 | continue; | |
6555 | } | |
6b06314c | 6556 | |
05f3fb3c | 6557 | table = &ctx->file_data->table[i >> IORING_FILE_TABLE_SHIFT]; |
65e19f54 | 6558 | index = i & IORING_FILE_TABLE_MASK; |
05f3fb3c | 6559 | file = fget(fd); |
6b06314c JA |
6560 | |
6561 | ret = -EBADF; | |
05f3fb3c | 6562 | if (!file) |
6b06314c | 6563 | break; |
05f3fb3c | 6564 | |
6b06314c JA |
6565 | /* |
6566 | * Don't allow io_uring instances to be registered. If UNIX | |
6567 | * isn't enabled, then this causes a reference cycle and this | |
6568 | * instance can never get freed. If UNIX is enabled we'll | |
6569 | * handle it just fine, but there's still no point in allowing | |
6570 | * a ring fd as it doesn't support regular read/write anyway. | |
6571 | */ | |
05f3fb3c JA |
6572 | if (file->f_op == &io_uring_fops) { |
6573 | fput(file); | |
6b06314c JA |
6574 | break; |
6575 | } | |
6b06314c | 6576 | ret = 0; |
05f3fb3c | 6577 | table->files[index] = file; |
6b06314c JA |
6578 | } |
6579 | ||
6580 | if (ret) { | |
65e19f54 | 6581 | for (i = 0; i < ctx->nr_user_files; i++) { |
65e19f54 JA |
6582 | file = io_file_from_index(ctx, i); |
6583 | if (file) | |
6584 | fput(file); | |
6585 | } | |
6586 | for (i = 0; i < nr_tables; i++) | |
05f3fb3c | 6587 | kfree(ctx->file_data->table[i].files); |
6b06314c | 6588 | |
05f3fb3c JA |
6589 | kfree(ctx->file_data->table); |
6590 | kfree(ctx->file_data); | |
6591 | ctx->file_data = NULL; | |
6b06314c JA |
6592 | ctx->nr_user_files = 0; |
6593 | return ret; | |
6594 | } | |
6595 | ||
6596 | ret = io_sqe_files_scm(ctx); | |
05589553 | 6597 | if (ret) { |
6b06314c | 6598 | io_sqe_files_unregister(ctx); |
05589553 XW |
6599 | return ret; |
6600 | } | |
6b06314c | 6601 | |
05589553 XW |
6602 | ref_node = alloc_fixed_file_ref_node(ctx); |
6603 | if (IS_ERR(ref_node)) { | |
6604 | io_sqe_files_unregister(ctx); | |
6605 | return PTR_ERR(ref_node); | |
6606 | } | |
6607 | ||
6608 | ctx->file_data->cur_refs = &ref_node->refs; | |
6609 | spin_lock_irqsave(&ctx->file_data->lock, flags); | |
6610 | list_add(&ref_node->node, &ctx->file_data->ref_list); | |
6611 | spin_unlock_irqrestore(&ctx->file_data->lock, flags); | |
6612 | percpu_ref_get(&ctx->file_data->refs); | |
6b06314c JA |
6613 | return ret; |
6614 | } | |
6615 | ||
c3a31e60 JA |
6616 | static int io_sqe_file_register(struct io_ring_ctx *ctx, struct file *file, |
6617 | int index) | |
6618 | { | |
6619 | #if defined(CONFIG_UNIX) | |
6620 | struct sock *sock = ctx->ring_sock->sk; | |
6621 | struct sk_buff_head *head = &sock->sk_receive_queue; | |
6622 | struct sk_buff *skb; | |
6623 | ||
6624 | /* | |
6625 | * See if we can merge this file into an existing skb SCM_RIGHTS | |
6626 | * file set. If there's no room, fall back to allocating a new skb | |
6627 | * and filling it in. | |
6628 | */ | |
6629 | spin_lock_irq(&head->lock); | |
6630 | skb = skb_peek(head); | |
6631 | if (skb) { | |
6632 | struct scm_fp_list *fpl = UNIXCB(skb).fp; | |
6633 | ||
6634 | if (fpl->count < SCM_MAX_FD) { | |
6635 | __skb_unlink(skb, head); | |
6636 | spin_unlock_irq(&head->lock); | |
6637 | fpl->fp[fpl->count] = get_file(file); | |
6638 | unix_inflight(fpl->user, fpl->fp[fpl->count]); | |
6639 | fpl->count++; | |
6640 | spin_lock_irq(&head->lock); | |
6641 | __skb_queue_head(head, skb); | |
6642 | } else { | |
6643 | skb = NULL; | |
6644 | } | |
6645 | } | |
6646 | spin_unlock_irq(&head->lock); | |
6647 | ||
6648 | if (skb) { | |
6649 | fput(file); | |
6650 | return 0; | |
6651 | } | |
6652 | ||
6653 | return __io_sqe_files_scm(ctx, 1, index); | |
6654 | #else | |
6655 | return 0; | |
6656 | #endif | |
6657 | } | |
6658 | ||
a5318d3c | 6659 | static int io_queue_file_removal(struct fixed_file_data *data, |
05589553 | 6660 | struct file *file) |
05f3fb3c | 6661 | { |
a5318d3c | 6662 | struct io_file_put *pfile; |
05589553 XW |
6663 | struct percpu_ref *refs = data->cur_refs; |
6664 | struct fixed_file_ref_node *ref_node; | |
05f3fb3c | 6665 | |
05f3fb3c | 6666 | pfile = kzalloc(sizeof(*pfile), GFP_KERNEL); |
a5318d3c HD |
6667 | if (!pfile) |
6668 | return -ENOMEM; | |
05f3fb3c | 6669 | |
05589553 | 6670 | ref_node = container_of(refs, struct fixed_file_ref_node, refs); |
05f3fb3c | 6671 | pfile->file = file; |
05589553 XW |
6672 | list_add(&pfile->list, &ref_node->file_list); |
6673 | ||
a5318d3c | 6674 | return 0; |
05f3fb3c JA |
6675 | } |
6676 | ||
6677 | static int __io_sqe_files_update(struct io_ring_ctx *ctx, | |
6678 | struct io_uring_files_update *up, | |
6679 | unsigned nr_args) | |
6680 | { | |
6681 | struct fixed_file_data *data = ctx->file_data; | |
05589553 | 6682 | struct fixed_file_ref_node *ref_node; |
05f3fb3c | 6683 | struct file *file; |
c3a31e60 JA |
6684 | __s32 __user *fds; |
6685 | int fd, i, err; | |
6686 | __u32 done; | |
05589553 XW |
6687 | unsigned long flags; |
6688 | bool needs_switch = false; | |
c3a31e60 | 6689 | |
05f3fb3c | 6690 | if (check_add_overflow(up->offset, nr_args, &done)) |
c3a31e60 JA |
6691 | return -EOVERFLOW; |
6692 | if (done > ctx->nr_user_files) | |
6693 | return -EINVAL; | |
6694 | ||
05589553 XW |
6695 | ref_node = alloc_fixed_file_ref_node(ctx); |
6696 | if (IS_ERR(ref_node)) | |
6697 | return PTR_ERR(ref_node); | |
6698 | ||
c3a31e60 | 6699 | done = 0; |
05f3fb3c | 6700 | fds = u64_to_user_ptr(up->fds); |
c3a31e60 | 6701 | while (nr_args) { |
65e19f54 JA |
6702 | struct fixed_file_table *table; |
6703 | unsigned index; | |
6704 | ||
c3a31e60 JA |
6705 | err = 0; |
6706 | if (copy_from_user(&fd, &fds[done], sizeof(fd))) { | |
6707 | err = -EFAULT; | |
6708 | break; | |
6709 | } | |
05f3fb3c JA |
6710 | i = array_index_nospec(up->offset, ctx->nr_user_files); |
6711 | table = &ctx->file_data->table[i >> IORING_FILE_TABLE_SHIFT]; | |
65e19f54 JA |
6712 | index = i & IORING_FILE_TABLE_MASK; |
6713 | if (table->files[index]) { | |
05f3fb3c | 6714 | file = io_file_from_index(ctx, index); |
a5318d3c HD |
6715 | err = io_queue_file_removal(data, file); |
6716 | if (err) | |
6717 | break; | |
65e19f54 | 6718 | table->files[index] = NULL; |
05589553 | 6719 | needs_switch = true; |
c3a31e60 JA |
6720 | } |
6721 | if (fd != -1) { | |
c3a31e60 JA |
6722 | file = fget(fd); |
6723 | if (!file) { | |
6724 | err = -EBADF; | |
6725 | break; | |
6726 | } | |
6727 | /* | |
6728 | * Don't allow io_uring instances to be registered. If | |
6729 | * UNIX isn't enabled, then this causes a reference | |
6730 | * cycle and this instance can never get freed. If UNIX | |
6731 | * is enabled we'll handle it just fine, but there's | |
6732 | * still no point in allowing a ring fd as it doesn't | |
6733 | * support regular read/write anyway. | |
6734 | */ | |
6735 | if (file->f_op == &io_uring_fops) { | |
6736 | fput(file); | |
6737 | err = -EBADF; | |
6738 | break; | |
6739 | } | |
65e19f54 | 6740 | table->files[index] = file; |
c3a31e60 JA |
6741 | err = io_sqe_file_register(ctx, file, i); |
6742 | if (err) | |
6743 | break; | |
6744 | } | |
6745 | nr_args--; | |
6746 | done++; | |
05f3fb3c JA |
6747 | up->offset++; |
6748 | } | |
6749 | ||
05589553 XW |
6750 | if (needs_switch) { |
6751 | percpu_ref_kill(data->cur_refs); | |
6752 | spin_lock_irqsave(&data->lock, flags); | |
6753 | list_add(&ref_node->node, &data->ref_list); | |
6754 | data->cur_refs = &ref_node->refs; | |
6755 | spin_unlock_irqrestore(&data->lock, flags); | |
6756 | percpu_ref_get(&ctx->file_data->refs); | |
6757 | } else | |
6758 | destroy_fixed_file_ref_node(ref_node); | |
c3a31e60 JA |
6759 | |
6760 | return done ? done : err; | |
6761 | } | |
05589553 | 6762 | |
05f3fb3c JA |
6763 | static int io_sqe_files_update(struct io_ring_ctx *ctx, void __user *arg, |
6764 | unsigned nr_args) | |
6765 | { | |
6766 | struct io_uring_files_update up; | |
6767 | ||
6768 | if (!ctx->file_data) | |
6769 | return -ENXIO; | |
6770 | if (!nr_args) | |
6771 | return -EINVAL; | |
6772 | if (copy_from_user(&up, arg, sizeof(up))) | |
6773 | return -EFAULT; | |
6774 | if (up.resv) | |
6775 | return -EINVAL; | |
6776 | ||
6777 | return __io_sqe_files_update(ctx, &up, nr_args); | |
6778 | } | |
c3a31e60 | 6779 | |
e9fd9396 | 6780 | static void io_free_work(struct io_wq_work *work) |
7d723065 JA |
6781 | { |
6782 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); | |
6783 | ||
e9fd9396 | 6784 | /* Consider that io_steal_work() relies on this ref */ |
7d723065 JA |
6785 | io_put_req(req); |
6786 | } | |
6787 | ||
24369c2e PB |
6788 | static int io_init_wq_offload(struct io_ring_ctx *ctx, |
6789 | struct io_uring_params *p) | |
6790 | { | |
6791 | struct io_wq_data data; | |
6792 | struct fd f; | |
6793 | struct io_ring_ctx *ctx_attach; | |
6794 | unsigned int concurrency; | |
6795 | int ret = 0; | |
6796 | ||
6797 | data.user = ctx->user; | |
e9fd9396 | 6798 | data.free_work = io_free_work; |
24369c2e PB |
6799 | |
6800 | if (!(p->flags & IORING_SETUP_ATTACH_WQ)) { | |
6801 | /* Do QD, or 4 * CPUS, whatever is smallest */ | |
6802 | concurrency = min(ctx->sq_entries, 4 * num_online_cpus()); | |
6803 | ||
6804 | ctx->io_wq = io_wq_create(concurrency, &data); | |
6805 | if (IS_ERR(ctx->io_wq)) { | |
6806 | ret = PTR_ERR(ctx->io_wq); | |
6807 | ctx->io_wq = NULL; | |
6808 | } | |
6809 | return ret; | |
6810 | } | |
6811 | ||
6812 | f = fdget(p->wq_fd); | |
6813 | if (!f.file) | |
6814 | return -EBADF; | |
6815 | ||
6816 | if (f.file->f_op != &io_uring_fops) { | |
6817 | ret = -EINVAL; | |
6818 | goto out_fput; | |
6819 | } | |
6820 | ||
6821 | ctx_attach = f.file->private_data; | |
6822 | /* @io_wq is protected by holding the fd */ | |
6823 | if (!io_wq_get(ctx_attach->io_wq, &data)) { | |
6824 | ret = -EINVAL; | |
6825 | goto out_fput; | |
6826 | } | |
6827 | ||
6828 | ctx->io_wq = ctx_attach->io_wq; | |
6829 | out_fput: | |
6830 | fdput(f); | |
6831 | return ret; | |
6832 | } | |
6833 | ||
6c271ce2 JA |
6834 | static int io_sq_offload_start(struct io_ring_ctx *ctx, |
6835 | struct io_uring_params *p) | |
2b188cc1 JA |
6836 | { |
6837 | int ret; | |
6838 | ||
6c271ce2 | 6839 | init_waitqueue_head(&ctx->sqo_wait); |
2b188cc1 JA |
6840 | mmgrab(current->mm); |
6841 | ctx->sqo_mm = current->mm; | |
6842 | ||
6c271ce2 | 6843 | if (ctx->flags & IORING_SETUP_SQPOLL) { |
3ec482d1 JA |
6844 | ret = -EPERM; |
6845 | if (!capable(CAP_SYS_ADMIN)) | |
6846 | goto err; | |
6847 | ||
917257da JA |
6848 | ctx->sq_thread_idle = msecs_to_jiffies(p->sq_thread_idle); |
6849 | if (!ctx->sq_thread_idle) | |
6850 | ctx->sq_thread_idle = HZ; | |
6851 | ||
6c271ce2 | 6852 | if (p->flags & IORING_SETUP_SQ_AFF) { |
44a9bd18 | 6853 | int cpu = p->sq_thread_cpu; |
6c271ce2 | 6854 | |
917257da | 6855 | ret = -EINVAL; |
44a9bd18 JA |
6856 | if (cpu >= nr_cpu_ids) |
6857 | goto err; | |
7889f44d | 6858 | if (!cpu_online(cpu)) |
917257da JA |
6859 | goto err; |
6860 | ||
6c271ce2 JA |
6861 | ctx->sqo_thread = kthread_create_on_cpu(io_sq_thread, |
6862 | ctx, cpu, | |
6863 | "io_uring-sq"); | |
6864 | } else { | |
6865 | ctx->sqo_thread = kthread_create(io_sq_thread, ctx, | |
6866 | "io_uring-sq"); | |
6867 | } | |
6868 | if (IS_ERR(ctx->sqo_thread)) { | |
6869 | ret = PTR_ERR(ctx->sqo_thread); | |
6870 | ctx->sqo_thread = NULL; | |
6871 | goto err; | |
6872 | } | |
6873 | wake_up_process(ctx->sqo_thread); | |
6874 | } else if (p->flags & IORING_SETUP_SQ_AFF) { | |
6875 | /* Can't have SQ_AFF without SQPOLL */ | |
6876 | ret = -EINVAL; | |
6877 | goto err; | |
6878 | } | |
6879 | ||
24369c2e PB |
6880 | ret = io_init_wq_offload(ctx, p); |
6881 | if (ret) | |
2b188cc1 | 6882 | goto err; |
2b188cc1 JA |
6883 | |
6884 | return 0; | |
6885 | err: | |
54a91f3b | 6886 | io_finish_async(ctx); |
2b188cc1 JA |
6887 | mmdrop(ctx->sqo_mm); |
6888 | ctx->sqo_mm = NULL; | |
6889 | return ret; | |
6890 | } | |
6891 | ||
6892 | static void io_unaccount_mem(struct user_struct *user, unsigned long nr_pages) | |
6893 | { | |
6894 | atomic_long_sub(nr_pages, &user->locked_vm); | |
6895 | } | |
6896 | ||
6897 | static int io_account_mem(struct user_struct *user, unsigned long nr_pages) | |
6898 | { | |
6899 | unsigned long page_limit, cur_pages, new_pages; | |
6900 | ||
6901 | /* Don't allow more pages than we can safely lock */ | |
6902 | page_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT; | |
6903 | ||
6904 | do { | |
6905 | cur_pages = atomic_long_read(&user->locked_vm); | |
6906 | new_pages = cur_pages + nr_pages; | |
6907 | if (new_pages > page_limit) | |
6908 | return -ENOMEM; | |
6909 | } while (atomic_long_cmpxchg(&user->locked_vm, cur_pages, | |
6910 | new_pages) != cur_pages); | |
6911 | ||
6912 | return 0; | |
6913 | } | |
6914 | ||
6915 | static void io_mem_free(void *ptr) | |
6916 | { | |
52e04ef4 MR |
6917 | struct page *page; |
6918 | ||
6919 | if (!ptr) | |
6920 | return; | |
2b188cc1 | 6921 | |
52e04ef4 | 6922 | page = virt_to_head_page(ptr); |
2b188cc1 JA |
6923 | if (put_page_testzero(page)) |
6924 | free_compound_page(page); | |
6925 | } | |
6926 | ||
6927 | static void *io_mem_alloc(size_t size) | |
6928 | { | |
6929 | gfp_t gfp_flags = GFP_KERNEL | __GFP_ZERO | __GFP_NOWARN | __GFP_COMP | | |
6930 | __GFP_NORETRY; | |
6931 | ||
6932 | return (void *) __get_free_pages(gfp_flags, get_order(size)); | |
6933 | } | |
6934 | ||
75b28aff HV |
6935 | static unsigned long rings_size(unsigned sq_entries, unsigned cq_entries, |
6936 | size_t *sq_offset) | |
6937 | { | |
6938 | struct io_rings *rings; | |
6939 | size_t off, sq_array_size; | |
6940 | ||
6941 | off = struct_size(rings, cqes, cq_entries); | |
6942 | if (off == SIZE_MAX) | |
6943 | return SIZE_MAX; | |
6944 | ||
6945 | #ifdef CONFIG_SMP | |
6946 | off = ALIGN(off, SMP_CACHE_BYTES); | |
6947 | if (off == 0) | |
6948 | return SIZE_MAX; | |
6949 | #endif | |
6950 | ||
6951 | sq_array_size = array_size(sizeof(u32), sq_entries); | |
6952 | if (sq_array_size == SIZE_MAX) | |
6953 | return SIZE_MAX; | |
6954 | ||
6955 | if (check_add_overflow(off, sq_array_size, &off)) | |
6956 | return SIZE_MAX; | |
6957 | ||
6958 | if (sq_offset) | |
6959 | *sq_offset = off; | |
6960 | ||
6961 | return off; | |
6962 | } | |
6963 | ||
2b188cc1 JA |
6964 | static unsigned long ring_pages(unsigned sq_entries, unsigned cq_entries) |
6965 | { | |
75b28aff | 6966 | size_t pages; |
2b188cc1 | 6967 | |
75b28aff HV |
6968 | pages = (size_t)1 << get_order( |
6969 | rings_size(sq_entries, cq_entries, NULL)); | |
6970 | pages += (size_t)1 << get_order( | |
6971 | array_size(sizeof(struct io_uring_sqe), sq_entries)); | |
2b188cc1 | 6972 | |
75b28aff | 6973 | return pages; |
2b188cc1 JA |
6974 | } |
6975 | ||
edafccee JA |
6976 | static int io_sqe_buffer_unregister(struct io_ring_ctx *ctx) |
6977 | { | |
6978 | int i, j; | |
6979 | ||
6980 | if (!ctx->user_bufs) | |
6981 | return -ENXIO; | |
6982 | ||
6983 | for (i = 0; i < ctx->nr_user_bufs; i++) { | |
6984 | struct io_mapped_ubuf *imu = &ctx->user_bufs[i]; | |
6985 | ||
6986 | for (j = 0; j < imu->nr_bvecs; j++) | |
f1f6a7dd | 6987 | unpin_user_page(imu->bvec[j].bv_page); |
edafccee JA |
6988 | |
6989 | if (ctx->account_mem) | |
6990 | io_unaccount_mem(ctx->user, imu->nr_bvecs); | |
d4ef6475 | 6991 | kvfree(imu->bvec); |
edafccee JA |
6992 | imu->nr_bvecs = 0; |
6993 | } | |
6994 | ||
6995 | kfree(ctx->user_bufs); | |
6996 | ctx->user_bufs = NULL; | |
6997 | ctx->nr_user_bufs = 0; | |
6998 | return 0; | |
6999 | } | |
7000 | ||
7001 | static int io_copy_iov(struct io_ring_ctx *ctx, struct iovec *dst, | |
7002 | void __user *arg, unsigned index) | |
7003 | { | |
7004 | struct iovec __user *src; | |
7005 | ||
7006 | #ifdef CONFIG_COMPAT | |
7007 | if (ctx->compat) { | |
7008 | struct compat_iovec __user *ciovs; | |
7009 | struct compat_iovec ciov; | |
7010 | ||
7011 | ciovs = (struct compat_iovec __user *) arg; | |
7012 | if (copy_from_user(&ciov, &ciovs[index], sizeof(ciov))) | |
7013 | return -EFAULT; | |
7014 | ||
d55e5f5b | 7015 | dst->iov_base = u64_to_user_ptr((u64)ciov.iov_base); |
edafccee JA |
7016 | dst->iov_len = ciov.iov_len; |
7017 | return 0; | |
7018 | } | |
7019 | #endif | |
7020 | src = (struct iovec __user *) arg; | |
7021 | if (copy_from_user(dst, &src[index], sizeof(*dst))) | |
7022 | return -EFAULT; | |
7023 | return 0; | |
7024 | } | |
7025 | ||
7026 | static int io_sqe_buffer_register(struct io_ring_ctx *ctx, void __user *arg, | |
7027 | unsigned nr_args) | |
7028 | { | |
7029 | struct vm_area_struct **vmas = NULL; | |
7030 | struct page **pages = NULL; | |
7031 | int i, j, got_pages = 0; | |
7032 | int ret = -EINVAL; | |
7033 | ||
7034 | if (ctx->user_bufs) | |
7035 | return -EBUSY; | |
7036 | if (!nr_args || nr_args > UIO_MAXIOV) | |
7037 | return -EINVAL; | |
7038 | ||
7039 | ctx->user_bufs = kcalloc(nr_args, sizeof(struct io_mapped_ubuf), | |
7040 | GFP_KERNEL); | |
7041 | if (!ctx->user_bufs) | |
7042 | return -ENOMEM; | |
7043 | ||
7044 | for (i = 0; i < nr_args; i++) { | |
7045 | struct io_mapped_ubuf *imu = &ctx->user_bufs[i]; | |
7046 | unsigned long off, start, end, ubuf; | |
7047 | int pret, nr_pages; | |
7048 | struct iovec iov; | |
7049 | size_t size; | |
7050 | ||
7051 | ret = io_copy_iov(ctx, &iov, arg, i); | |
7052 | if (ret) | |
a278682d | 7053 | goto err; |
edafccee JA |
7054 | |
7055 | /* | |
7056 | * Don't impose further limits on the size and buffer | |
7057 | * constraints here, we'll -EINVAL later when IO is | |
7058 | * submitted if they are wrong. | |
7059 | */ | |
7060 | ret = -EFAULT; | |
7061 | if (!iov.iov_base || !iov.iov_len) | |
7062 | goto err; | |
7063 | ||
7064 | /* arbitrary limit, but we need something */ | |
7065 | if (iov.iov_len > SZ_1G) | |
7066 | goto err; | |
7067 | ||
7068 | ubuf = (unsigned long) iov.iov_base; | |
7069 | end = (ubuf + iov.iov_len + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
7070 | start = ubuf >> PAGE_SHIFT; | |
7071 | nr_pages = end - start; | |
7072 | ||
7073 | if (ctx->account_mem) { | |
7074 | ret = io_account_mem(ctx->user, nr_pages); | |
7075 | if (ret) | |
7076 | goto err; | |
7077 | } | |
7078 | ||
7079 | ret = 0; | |
7080 | if (!pages || nr_pages > got_pages) { | |
7081 | kfree(vmas); | |
7082 | kfree(pages); | |
d4ef6475 | 7083 | pages = kvmalloc_array(nr_pages, sizeof(struct page *), |
edafccee | 7084 | GFP_KERNEL); |
d4ef6475 | 7085 | vmas = kvmalloc_array(nr_pages, |
edafccee JA |
7086 | sizeof(struct vm_area_struct *), |
7087 | GFP_KERNEL); | |
7088 | if (!pages || !vmas) { | |
7089 | ret = -ENOMEM; | |
7090 | if (ctx->account_mem) | |
7091 | io_unaccount_mem(ctx->user, nr_pages); | |
7092 | goto err; | |
7093 | } | |
7094 | got_pages = nr_pages; | |
7095 | } | |
7096 | ||
d4ef6475 | 7097 | imu->bvec = kvmalloc_array(nr_pages, sizeof(struct bio_vec), |
edafccee JA |
7098 | GFP_KERNEL); |
7099 | ret = -ENOMEM; | |
7100 | if (!imu->bvec) { | |
7101 | if (ctx->account_mem) | |
7102 | io_unaccount_mem(ctx->user, nr_pages); | |
7103 | goto err; | |
7104 | } | |
7105 | ||
7106 | ret = 0; | |
7107 | down_read(¤t->mm->mmap_sem); | |
2113b05d | 7108 | pret = pin_user_pages(ubuf, nr_pages, |
932f4a63 IW |
7109 | FOLL_WRITE | FOLL_LONGTERM, |
7110 | pages, vmas); | |
edafccee JA |
7111 | if (pret == nr_pages) { |
7112 | /* don't support file backed memory */ | |
7113 | for (j = 0; j < nr_pages; j++) { | |
7114 | struct vm_area_struct *vma = vmas[j]; | |
7115 | ||
7116 | if (vma->vm_file && | |
7117 | !is_file_hugepages(vma->vm_file)) { | |
7118 | ret = -EOPNOTSUPP; | |
7119 | break; | |
7120 | } | |
7121 | } | |
7122 | } else { | |
7123 | ret = pret < 0 ? pret : -EFAULT; | |
7124 | } | |
7125 | up_read(¤t->mm->mmap_sem); | |
7126 | if (ret) { | |
7127 | /* | |
7128 | * if we did partial map, or found file backed vmas, | |
7129 | * release any pages we did get | |
7130 | */ | |
27c4d3a3 | 7131 | if (pret > 0) |
f1f6a7dd | 7132 | unpin_user_pages(pages, pret); |
edafccee JA |
7133 | if (ctx->account_mem) |
7134 | io_unaccount_mem(ctx->user, nr_pages); | |
d4ef6475 | 7135 | kvfree(imu->bvec); |
edafccee JA |
7136 | goto err; |
7137 | } | |
7138 | ||
7139 | off = ubuf & ~PAGE_MASK; | |
7140 | size = iov.iov_len; | |
7141 | for (j = 0; j < nr_pages; j++) { | |
7142 | size_t vec_len; | |
7143 | ||
7144 | vec_len = min_t(size_t, size, PAGE_SIZE - off); | |
7145 | imu->bvec[j].bv_page = pages[j]; | |
7146 | imu->bvec[j].bv_len = vec_len; | |
7147 | imu->bvec[j].bv_offset = off; | |
7148 | off = 0; | |
7149 | size -= vec_len; | |
7150 | } | |
7151 | /* store original address for later verification */ | |
7152 | imu->ubuf = ubuf; | |
7153 | imu->len = iov.iov_len; | |
7154 | imu->nr_bvecs = nr_pages; | |
7155 | ||
7156 | ctx->nr_user_bufs++; | |
7157 | } | |
d4ef6475 MR |
7158 | kvfree(pages); |
7159 | kvfree(vmas); | |
edafccee JA |
7160 | return 0; |
7161 | err: | |
d4ef6475 MR |
7162 | kvfree(pages); |
7163 | kvfree(vmas); | |
edafccee JA |
7164 | io_sqe_buffer_unregister(ctx); |
7165 | return ret; | |
7166 | } | |
7167 | ||
9b402849 JA |
7168 | static int io_eventfd_register(struct io_ring_ctx *ctx, void __user *arg) |
7169 | { | |
7170 | __s32 __user *fds = arg; | |
7171 | int fd; | |
7172 | ||
7173 | if (ctx->cq_ev_fd) | |
7174 | return -EBUSY; | |
7175 | ||
7176 | if (copy_from_user(&fd, fds, sizeof(*fds))) | |
7177 | return -EFAULT; | |
7178 | ||
7179 | ctx->cq_ev_fd = eventfd_ctx_fdget(fd); | |
7180 | if (IS_ERR(ctx->cq_ev_fd)) { | |
7181 | int ret = PTR_ERR(ctx->cq_ev_fd); | |
7182 | ctx->cq_ev_fd = NULL; | |
7183 | return ret; | |
7184 | } | |
7185 | ||
7186 | return 0; | |
7187 | } | |
7188 | ||
7189 | static int io_eventfd_unregister(struct io_ring_ctx *ctx) | |
7190 | { | |
7191 | if (ctx->cq_ev_fd) { | |
7192 | eventfd_ctx_put(ctx->cq_ev_fd); | |
7193 | ctx->cq_ev_fd = NULL; | |
7194 | return 0; | |
7195 | } | |
7196 | ||
7197 | return -ENXIO; | |
7198 | } | |
7199 | ||
5a2e745d JA |
7200 | static int __io_destroy_buffers(int id, void *p, void *data) |
7201 | { | |
7202 | struct io_ring_ctx *ctx = data; | |
7203 | struct io_buffer *buf = p; | |
7204 | ||
067524e9 | 7205 | __io_remove_buffers(ctx, buf, id, -1U); |
5a2e745d JA |
7206 | return 0; |
7207 | } | |
7208 | ||
7209 | static void io_destroy_buffers(struct io_ring_ctx *ctx) | |
7210 | { | |
7211 | idr_for_each(&ctx->io_buffer_idr, __io_destroy_buffers, ctx); | |
7212 | idr_destroy(&ctx->io_buffer_idr); | |
7213 | } | |
7214 | ||
2b188cc1 JA |
7215 | static void io_ring_ctx_free(struct io_ring_ctx *ctx) |
7216 | { | |
6b06314c | 7217 | io_finish_async(ctx); |
2b188cc1 JA |
7218 | if (ctx->sqo_mm) |
7219 | mmdrop(ctx->sqo_mm); | |
def596e9 JA |
7220 | |
7221 | io_iopoll_reap_events(ctx); | |
edafccee | 7222 | io_sqe_buffer_unregister(ctx); |
6b06314c | 7223 | io_sqe_files_unregister(ctx); |
9b402849 | 7224 | io_eventfd_unregister(ctx); |
5a2e745d | 7225 | io_destroy_buffers(ctx); |
41726c9a | 7226 | idr_destroy(&ctx->personality_idr); |
def596e9 | 7227 | |
2b188cc1 | 7228 | #if defined(CONFIG_UNIX) |
355e8d26 EB |
7229 | if (ctx->ring_sock) { |
7230 | ctx->ring_sock->file = NULL; /* so that iput() is called */ | |
2b188cc1 | 7231 | sock_release(ctx->ring_sock); |
355e8d26 | 7232 | } |
2b188cc1 JA |
7233 | #endif |
7234 | ||
75b28aff | 7235 | io_mem_free(ctx->rings); |
2b188cc1 | 7236 | io_mem_free(ctx->sq_sqes); |
2b188cc1 JA |
7237 | |
7238 | percpu_ref_exit(&ctx->refs); | |
7239 | if (ctx->account_mem) | |
7240 | io_unaccount_mem(ctx->user, | |
7241 | ring_pages(ctx->sq_entries, ctx->cq_entries)); | |
7242 | free_uid(ctx->user); | |
181e448d | 7243 | put_cred(ctx->creds); |
206aefde | 7244 | kfree(ctx->completions); |
78076bb6 | 7245 | kfree(ctx->cancel_hash); |
0ddf92e8 | 7246 | kmem_cache_free(req_cachep, ctx->fallback_req); |
2b188cc1 JA |
7247 | kfree(ctx); |
7248 | } | |
7249 | ||
7250 | static __poll_t io_uring_poll(struct file *file, poll_table *wait) | |
7251 | { | |
7252 | struct io_ring_ctx *ctx = file->private_data; | |
7253 | __poll_t mask = 0; | |
7254 | ||
7255 | poll_wait(file, &ctx->cq_wait, wait); | |
4f7067c3 SB |
7256 | /* |
7257 | * synchronizes with barrier from wq_has_sleeper call in | |
7258 | * io_commit_cqring | |
7259 | */ | |
2b188cc1 | 7260 | smp_rmb(); |
75b28aff HV |
7261 | if (READ_ONCE(ctx->rings->sq.tail) - ctx->cached_sq_head != |
7262 | ctx->rings->sq_ring_entries) | |
2b188cc1 | 7263 | mask |= EPOLLOUT | EPOLLWRNORM; |
63e5d81f | 7264 | if (io_cqring_events(ctx, false)) |
2b188cc1 JA |
7265 | mask |= EPOLLIN | EPOLLRDNORM; |
7266 | ||
7267 | return mask; | |
7268 | } | |
7269 | ||
7270 | static int io_uring_fasync(int fd, struct file *file, int on) | |
7271 | { | |
7272 | struct io_ring_ctx *ctx = file->private_data; | |
7273 | ||
7274 | return fasync_helper(fd, file, on, &ctx->cq_fasync); | |
7275 | } | |
7276 | ||
071698e1 JA |
7277 | static int io_remove_personalities(int id, void *p, void *data) |
7278 | { | |
7279 | struct io_ring_ctx *ctx = data; | |
7280 | const struct cred *cred; | |
7281 | ||
7282 | cred = idr_remove(&ctx->personality_idr, id); | |
7283 | if (cred) | |
7284 | put_cred(cred); | |
7285 | return 0; | |
7286 | } | |
7287 | ||
85faa7b8 JA |
7288 | static void io_ring_exit_work(struct work_struct *work) |
7289 | { | |
7290 | struct io_ring_ctx *ctx; | |
7291 | ||
7292 | ctx = container_of(work, struct io_ring_ctx, exit_work); | |
7293 | if (ctx->rings) | |
7294 | io_cqring_overflow_flush(ctx, true); | |
7295 | ||
7296 | wait_for_completion(&ctx->completions[0]); | |
7297 | io_ring_ctx_free(ctx); | |
7298 | } | |
7299 | ||
2b188cc1 JA |
7300 | static void io_ring_ctx_wait_and_kill(struct io_ring_ctx *ctx) |
7301 | { | |
7302 | mutex_lock(&ctx->uring_lock); | |
7303 | percpu_ref_kill(&ctx->refs); | |
7304 | mutex_unlock(&ctx->uring_lock); | |
7305 | ||
df069d80 JA |
7306 | /* |
7307 | * Wait for sq thread to idle, if we have one. It won't spin on new | |
7308 | * work after we've killed the ctx ref above. This is important to do | |
7309 | * before we cancel existing commands, as the thread could otherwise | |
7310 | * be queueing new work post that. If that's work we need to cancel, | |
7311 | * it could cause shutdown to hang. | |
7312 | */ | |
7313 | while (ctx->sqo_thread && !wq_has_sleeper(&ctx->sqo_wait)) | |
3fd44c86 | 7314 | cond_resched(); |
df069d80 | 7315 | |
5262f567 | 7316 | io_kill_timeouts(ctx); |
221c5eb2 | 7317 | io_poll_remove_all(ctx); |
561fb04a JA |
7318 | |
7319 | if (ctx->io_wq) | |
7320 | io_wq_cancel_all(ctx->io_wq); | |
7321 | ||
def596e9 | 7322 | io_iopoll_reap_events(ctx); |
15dff286 JA |
7323 | /* if we failed setting up the ctx, we might not have any rings */ |
7324 | if (ctx->rings) | |
7325 | io_cqring_overflow_flush(ctx, true); | |
071698e1 | 7326 | idr_for_each(&ctx->personality_idr, io_remove_personalities, ctx); |
85faa7b8 JA |
7327 | INIT_WORK(&ctx->exit_work, io_ring_exit_work); |
7328 | queue_work(system_wq, &ctx->exit_work); | |
2b188cc1 JA |
7329 | } |
7330 | ||
7331 | static int io_uring_release(struct inode *inode, struct file *file) | |
7332 | { | |
7333 | struct io_ring_ctx *ctx = file->private_data; | |
7334 | ||
7335 | file->private_data = NULL; | |
7336 | io_ring_ctx_wait_and_kill(ctx); | |
7337 | return 0; | |
7338 | } | |
7339 | ||
fcb323cc JA |
7340 | static void io_uring_cancel_files(struct io_ring_ctx *ctx, |
7341 | struct files_struct *files) | |
7342 | { | |
fcb323cc | 7343 | while (!list_empty_careful(&ctx->inflight_list)) { |
d8f1b971 XW |
7344 | struct io_kiocb *cancel_req = NULL, *req; |
7345 | DEFINE_WAIT(wait); | |
fcb323cc JA |
7346 | |
7347 | spin_lock_irq(&ctx->inflight_lock); | |
7348 | list_for_each_entry(req, &ctx->inflight_list, inflight_entry) { | |
768134d4 JA |
7349 | if (req->work.files != files) |
7350 | continue; | |
7351 | /* req is being completed, ignore */ | |
7352 | if (!refcount_inc_not_zero(&req->refs)) | |
7353 | continue; | |
7354 | cancel_req = req; | |
7355 | break; | |
fcb323cc | 7356 | } |
768134d4 | 7357 | if (cancel_req) |
fcb323cc | 7358 | prepare_to_wait(&ctx->inflight_wait, &wait, |
768134d4 | 7359 | TASK_UNINTERRUPTIBLE); |
fcb323cc JA |
7360 | spin_unlock_irq(&ctx->inflight_lock); |
7361 | ||
768134d4 JA |
7362 | /* We need to keep going until we don't find a matching req */ |
7363 | if (!cancel_req) | |
fcb323cc | 7364 | break; |
2f6d9b9d | 7365 | |
2ca10259 JA |
7366 | if (cancel_req->flags & REQ_F_OVERFLOW) { |
7367 | spin_lock_irq(&ctx->completion_lock); | |
7368 | list_del(&cancel_req->list); | |
7369 | cancel_req->flags &= ~REQ_F_OVERFLOW; | |
7370 | if (list_empty(&ctx->cq_overflow_list)) { | |
7371 | clear_bit(0, &ctx->sq_check_overflow); | |
7372 | clear_bit(0, &ctx->cq_check_overflow); | |
7373 | } | |
7374 | spin_unlock_irq(&ctx->completion_lock); | |
7375 | ||
7376 | WRITE_ONCE(ctx->rings->cq_overflow, | |
7377 | atomic_inc_return(&ctx->cached_cq_overflow)); | |
7378 | ||
7379 | /* | |
7380 | * Put inflight ref and overflow ref. If that's | |
7381 | * all we had, then we're done with this request. | |
7382 | */ | |
7383 | if (refcount_sub_and_test(2, &cancel_req->refs)) { | |
7384 | io_put_req(cancel_req); | |
d8f1b971 | 7385 | finish_wait(&ctx->inflight_wait, &wait); |
2ca10259 JA |
7386 | continue; |
7387 | } | |
7388 | } | |
7389 | ||
2f6d9b9d BL |
7390 | io_wq_cancel_work(ctx->io_wq, &cancel_req->work); |
7391 | io_put_req(cancel_req); | |
fcb323cc | 7392 | schedule(); |
d8f1b971 | 7393 | finish_wait(&ctx->inflight_wait, &wait); |
fcb323cc JA |
7394 | } |
7395 | } | |
7396 | ||
7397 | static int io_uring_flush(struct file *file, void *data) | |
7398 | { | |
7399 | struct io_ring_ctx *ctx = file->private_data; | |
7400 | ||
7401 | io_uring_cancel_files(ctx, data); | |
6ab23144 JA |
7402 | |
7403 | /* | |
7404 | * If the task is going away, cancel work it may have pending | |
7405 | */ | |
7406 | if (fatal_signal_pending(current) || (current->flags & PF_EXITING)) | |
7407 | io_wq_cancel_pid(ctx->io_wq, task_pid_vnr(current)); | |
7408 | ||
fcb323cc JA |
7409 | return 0; |
7410 | } | |
7411 | ||
6c5c240e RP |
7412 | static void *io_uring_validate_mmap_request(struct file *file, |
7413 | loff_t pgoff, size_t sz) | |
2b188cc1 | 7414 | { |
2b188cc1 | 7415 | struct io_ring_ctx *ctx = file->private_data; |
6c5c240e | 7416 | loff_t offset = pgoff << PAGE_SHIFT; |
2b188cc1 JA |
7417 | struct page *page; |
7418 | void *ptr; | |
7419 | ||
7420 | switch (offset) { | |
7421 | case IORING_OFF_SQ_RING: | |
75b28aff HV |
7422 | case IORING_OFF_CQ_RING: |
7423 | ptr = ctx->rings; | |
2b188cc1 JA |
7424 | break; |
7425 | case IORING_OFF_SQES: | |
7426 | ptr = ctx->sq_sqes; | |
7427 | break; | |
2b188cc1 | 7428 | default: |
6c5c240e | 7429 | return ERR_PTR(-EINVAL); |
2b188cc1 JA |
7430 | } |
7431 | ||
7432 | page = virt_to_head_page(ptr); | |
a50b854e | 7433 | if (sz > page_size(page)) |
6c5c240e RP |
7434 | return ERR_PTR(-EINVAL); |
7435 | ||
7436 | return ptr; | |
7437 | } | |
7438 | ||
7439 | #ifdef CONFIG_MMU | |
7440 | ||
7441 | static int io_uring_mmap(struct file *file, struct vm_area_struct *vma) | |
7442 | { | |
7443 | size_t sz = vma->vm_end - vma->vm_start; | |
7444 | unsigned long pfn; | |
7445 | void *ptr; | |
7446 | ||
7447 | ptr = io_uring_validate_mmap_request(file, vma->vm_pgoff, sz); | |
7448 | if (IS_ERR(ptr)) | |
7449 | return PTR_ERR(ptr); | |
2b188cc1 JA |
7450 | |
7451 | pfn = virt_to_phys(ptr) >> PAGE_SHIFT; | |
7452 | return remap_pfn_range(vma, vma->vm_start, pfn, sz, vma->vm_page_prot); | |
7453 | } | |
7454 | ||
6c5c240e RP |
7455 | #else /* !CONFIG_MMU */ |
7456 | ||
7457 | static int io_uring_mmap(struct file *file, struct vm_area_struct *vma) | |
7458 | { | |
7459 | return vma->vm_flags & (VM_SHARED | VM_MAYSHARE) ? 0 : -EINVAL; | |
7460 | } | |
7461 | ||
7462 | static unsigned int io_uring_nommu_mmap_capabilities(struct file *file) | |
7463 | { | |
7464 | return NOMMU_MAP_DIRECT | NOMMU_MAP_READ | NOMMU_MAP_WRITE; | |
7465 | } | |
7466 | ||
7467 | static unsigned long io_uring_nommu_get_unmapped_area(struct file *file, | |
7468 | unsigned long addr, unsigned long len, | |
7469 | unsigned long pgoff, unsigned long flags) | |
7470 | { | |
7471 | void *ptr; | |
7472 | ||
7473 | ptr = io_uring_validate_mmap_request(file, pgoff, len); | |
7474 | if (IS_ERR(ptr)) | |
7475 | return PTR_ERR(ptr); | |
7476 | ||
7477 | return (unsigned long) ptr; | |
7478 | } | |
7479 | ||
7480 | #endif /* !CONFIG_MMU */ | |
7481 | ||
2b188cc1 JA |
7482 | SYSCALL_DEFINE6(io_uring_enter, unsigned int, fd, u32, to_submit, |
7483 | u32, min_complete, u32, flags, const sigset_t __user *, sig, | |
7484 | size_t, sigsz) | |
7485 | { | |
7486 | struct io_ring_ctx *ctx; | |
7487 | long ret = -EBADF; | |
7488 | int submitted = 0; | |
7489 | struct fd f; | |
7490 | ||
b41e9852 JA |
7491 | if (current->task_works) |
7492 | task_work_run(); | |
7493 | ||
6c271ce2 | 7494 | if (flags & ~(IORING_ENTER_GETEVENTS | IORING_ENTER_SQ_WAKEUP)) |
2b188cc1 JA |
7495 | return -EINVAL; |
7496 | ||
7497 | f = fdget(fd); | |
7498 | if (!f.file) | |
7499 | return -EBADF; | |
7500 | ||
7501 | ret = -EOPNOTSUPP; | |
7502 | if (f.file->f_op != &io_uring_fops) | |
7503 | goto out_fput; | |
7504 | ||
7505 | ret = -ENXIO; | |
7506 | ctx = f.file->private_data; | |
7507 | if (!percpu_ref_tryget(&ctx->refs)) | |
7508 | goto out_fput; | |
7509 | ||
6c271ce2 JA |
7510 | /* |
7511 | * For SQ polling, the thread will do all submissions and completions. | |
7512 | * Just return the requested submit count, and wake the thread if | |
7513 | * we were asked to. | |
7514 | */ | |
b2a9eada | 7515 | ret = 0; |
6c271ce2 | 7516 | if (ctx->flags & IORING_SETUP_SQPOLL) { |
c1edbf5f JA |
7517 | if (!list_empty_careful(&ctx->cq_overflow_list)) |
7518 | io_cqring_overflow_flush(ctx, false); | |
6c271ce2 JA |
7519 | if (flags & IORING_ENTER_SQ_WAKEUP) |
7520 | wake_up(&ctx->sqo_wait); | |
7521 | submitted = to_submit; | |
b2a9eada | 7522 | } else if (to_submit) { |
2b188cc1 | 7523 | mutex_lock(&ctx->uring_lock); |
bf9c2f1c | 7524 | submitted = io_submit_sqes(ctx, to_submit, f.file, fd, false); |
2b188cc1 | 7525 | mutex_unlock(&ctx->uring_lock); |
7c504e65 PB |
7526 | |
7527 | if (submitted != to_submit) | |
7528 | goto out; | |
2b188cc1 JA |
7529 | } |
7530 | if (flags & IORING_ENTER_GETEVENTS) { | |
def596e9 JA |
7531 | unsigned nr_events = 0; |
7532 | ||
2b188cc1 JA |
7533 | min_complete = min(min_complete, ctx->cq_entries); |
7534 | ||
32b2244a XW |
7535 | /* |
7536 | * When SETUP_IOPOLL and SETUP_SQPOLL are both enabled, user | |
7537 | * space applications don't need to do io completion events | |
7538 | * polling again, they can rely on io_sq_thread to do polling | |
7539 | * work, which can reduce cpu usage and uring_lock contention. | |
7540 | */ | |
7541 | if (ctx->flags & IORING_SETUP_IOPOLL && | |
7542 | !(ctx->flags & IORING_SETUP_SQPOLL)) { | |
def596e9 | 7543 | ret = io_iopoll_check(ctx, &nr_events, min_complete); |
def596e9 JA |
7544 | } else { |
7545 | ret = io_cqring_wait(ctx, min_complete, sig, sigsz); | |
7546 | } | |
2b188cc1 JA |
7547 | } |
7548 | ||
7c504e65 | 7549 | out: |
6805b32e | 7550 | percpu_ref_put(&ctx->refs); |
2b188cc1 JA |
7551 | out_fput: |
7552 | fdput(f); | |
7553 | return submitted ? submitted : ret; | |
7554 | } | |
7555 | ||
bebdb65e | 7556 | #ifdef CONFIG_PROC_FS |
87ce955b JA |
7557 | static int io_uring_show_cred(int id, void *p, void *data) |
7558 | { | |
7559 | const struct cred *cred = p; | |
7560 | struct seq_file *m = data; | |
7561 | struct user_namespace *uns = seq_user_ns(m); | |
7562 | struct group_info *gi; | |
7563 | kernel_cap_t cap; | |
7564 | unsigned __capi; | |
7565 | int g; | |
7566 | ||
7567 | seq_printf(m, "%5d\n", id); | |
7568 | seq_put_decimal_ull(m, "\tUid:\t", from_kuid_munged(uns, cred->uid)); | |
7569 | seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->euid)); | |
7570 | seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->suid)); | |
7571 | seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->fsuid)); | |
7572 | seq_put_decimal_ull(m, "\n\tGid:\t", from_kgid_munged(uns, cred->gid)); | |
7573 | seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->egid)); | |
7574 | seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->sgid)); | |
7575 | seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->fsgid)); | |
7576 | seq_puts(m, "\n\tGroups:\t"); | |
7577 | gi = cred->group_info; | |
7578 | for (g = 0; g < gi->ngroups; g++) { | |
7579 | seq_put_decimal_ull(m, g ? " " : "", | |
7580 | from_kgid_munged(uns, gi->gid[g])); | |
7581 | } | |
7582 | seq_puts(m, "\n\tCapEff:\t"); | |
7583 | cap = cred->cap_effective; | |
7584 | CAP_FOR_EACH_U32(__capi) | |
7585 | seq_put_hex_ll(m, NULL, cap.cap[CAP_LAST_U32 - __capi], 8); | |
7586 | seq_putc(m, '\n'); | |
7587 | return 0; | |
7588 | } | |
7589 | ||
7590 | static void __io_uring_show_fdinfo(struct io_ring_ctx *ctx, struct seq_file *m) | |
7591 | { | |
7592 | int i; | |
7593 | ||
7594 | mutex_lock(&ctx->uring_lock); | |
7595 | seq_printf(m, "UserFiles:\t%u\n", ctx->nr_user_files); | |
7596 | for (i = 0; i < ctx->nr_user_files; i++) { | |
7597 | struct fixed_file_table *table; | |
7598 | struct file *f; | |
7599 | ||
7600 | table = &ctx->file_data->table[i >> IORING_FILE_TABLE_SHIFT]; | |
7601 | f = table->files[i & IORING_FILE_TABLE_MASK]; | |
7602 | if (f) | |
7603 | seq_printf(m, "%5u: %s\n", i, file_dentry(f)->d_iname); | |
7604 | else | |
7605 | seq_printf(m, "%5u: <none>\n", i); | |
7606 | } | |
7607 | seq_printf(m, "UserBufs:\t%u\n", ctx->nr_user_bufs); | |
7608 | for (i = 0; i < ctx->nr_user_bufs; i++) { | |
7609 | struct io_mapped_ubuf *buf = &ctx->user_bufs[i]; | |
7610 | ||
7611 | seq_printf(m, "%5u: 0x%llx/%u\n", i, buf->ubuf, | |
7612 | (unsigned int) buf->len); | |
7613 | } | |
7614 | if (!idr_is_empty(&ctx->personality_idr)) { | |
7615 | seq_printf(m, "Personalities:\n"); | |
7616 | idr_for_each(&ctx->personality_idr, io_uring_show_cred, m); | |
7617 | } | |
d7718a9d JA |
7618 | seq_printf(m, "PollList:\n"); |
7619 | spin_lock_irq(&ctx->completion_lock); | |
7620 | for (i = 0; i < (1U << ctx->cancel_hash_bits); i++) { | |
7621 | struct hlist_head *list = &ctx->cancel_hash[i]; | |
7622 | struct io_kiocb *req; | |
7623 | ||
7624 | hlist_for_each_entry(req, list, hash_node) | |
7625 | seq_printf(m, " op=%d, task_works=%d\n", req->opcode, | |
7626 | req->task->task_works != NULL); | |
7627 | } | |
7628 | spin_unlock_irq(&ctx->completion_lock); | |
87ce955b JA |
7629 | mutex_unlock(&ctx->uring_lock); |
7630 | } | |
7631 | ||
7632 | static void io_uring_show_fdinfo(struct seq_file *m, struct file *f) | |
7633 | { | |
7634 | struct io_ring_ctx *ctx = f->private_data; | |
7635 | ||
7636 | if (percpu_ref_tryget(&ctx->refs)) { | |
7637 | __io_uring_show_fdinfo(ctx, m); | |
7638 | percpu_ref_put(&ctx->refs); | |
7639 | } | |
7640 | } | |
bebdb65e | 7641 | #endif |
87ce955b | 7642 | |
2b188cc1 JA |
7643 | static const struct file_operations io_uring_fops = { |
7644 | .release = io_uring_release, | |
fcb323cc | 7645 | .flush = io_uring_flush, |
2b188cc1 | 7646 | .mmap = io_uring_mmap, |
6c5c240e RP |
7647 | #ifndef CONFIG_MMU |
7648 | .get_unmapped_area = io_uring_nommu_get_unmapped_area, | |
7649 | .mmap_capabilities = io_uring_nommu_mmap_capabilities, | |
7650 | #endif | |
2b188cc1 JA |
7651 | .poll = io_uring_poll, |
7652 | .fasync = io_uring_fasync, | |
bebdb65e | 7653 | #ifdef CONFIG_PROC_FS |
87ce955b | 7654 | .show_fdinfo = io_uring_show_fdinfo, |
bebdb65e | 7655 | #endif |
2b188cc1 JA |
7656 | }; |
7657 | ||
7658 | static int io_allocate_scq_urings(struct io_ring_ctx *ctx, | |
7659 | struct io_uring_params *p) | |
7660 | { | |
75b28aff HV |
7661 | struct io_rings *rings; |
7662 | size_t size, sq_array_offset; | |
2b188cc1 | 7663 | |
75b28aff HV |
7664 | size = rings_size(p->sq_entries, p->cq_entries, &sq_array_offset); |
7665 | if (size == SIZE_MAX) | |
7666 | return -EOVERFLOW; | |
7667 | ||
7668 | rings = io_mem_alloc(size); | |
7669 | if (!rings) | |
2b188cc1 JA |
7670 | return -ENOMEM; |
7671 | ||
75b28aff HV |
7672 | ctx->rings = rings; |
7673 | ctx->sq_array = (u32 *)((char *)rings + sq_array_offset); | |
7674 | rings->sq_ring_mask = p->sq_entries - 1; | |
7675 | rings->cq_ring_mask = p->cq_entries - 1; | |
7676 | rings->sq_ring_entries = p->sq_entries; | |
7677 | rings->cq_ring_entries = p->cq_entries; | |
7678 | ctx->sq_mask = rings->sq_ring_mask; | |
7679 | ctx->cq_mask = rings->cq_ring_mask; | |
7680 | ctx->sq_entries = rings->sq_ring_entries; | |
7681 | ctx->cq_entries = rings->cq_ring_entries; | |
2b188cc1 JA |
7682 | |
7683 | size = array_size(sizeof(struct io_uring_sqe), p->sq_entries); | |
eb065d30 JA |
7684 | if (size == SIZE_MAX) { |
7685 | io_mem_free(ctx->rings); | |
7686 | ctx->rings = NULL; | |
2b188cc1 | 7687 | return -EOVERFLOW; |
eb065d30 | 7688 | } |
2b188cc1 JA |
7689 | |
7690 | ctx->sq_sqes = io_mem_alloc(size); | |
eb065d30 JA |
7691 | if (!ctx->sq_sqes) { |
7692 | io_mem_free(ctx->rings); | |
7693 | ctx->rings = NULL; | |
2b188cc1 | 7694 | return -ENOMEM; |
eb065d30 | 7695 | } |
2b188cc1 | 7696 | |
2b188cc1 JA |
7697 | return 0; |
7698 | } | |
7699 | ||
7700 | /* | |
7701 | * Allocate an anonymous fd, this is what constitutes the application | |
7702 | * visible backing of an io_uring instance. The application mmaps this | |
7703 | * fd to gain access to the SQ/CQ ring details. If UNIX sockets are enabled, | |
7704 | * we have to tie this fd to a socket for file garbage collection purposes. | |
7705 | */ | |
7706 | static int io_uring_get_fd(struct io_ring_ctx *ctx) | |
7707 | { | |
7708 | struct file *file; | |
7709 | int ret; | |
7710 | ||
7711 | #if defined(CONFIG_UNIX) | |
7712 | ret = sock_create_kern(&init_net, PF_UNIX, SOCK_RAW, IPPROTO_IP, | |
7713 | &ctx->ring_sock); | |
7714 | if (ret) | |
7715 | return ret; | |
7716 | #endif | |
7717 | ||
7718 | ret = get_unused_fd_flags(O_RDWR | O_CLOEXEC); | |
7719 | if (ret < 0) | |
7720 | goto err; | |
7721 | ||
7722 | file = anon_inode_getfile("[io_uring]", &io_uring_fops, ctx, | |
7723 | O_RDWR | O_CLOEXEC); | |
7724 | if (IS_ERR(file)) { | |
7725 | put_unused_fd(ret); | |
7726 | ret = PTR_ERR(file); | |
7727 | goto err; | |
7728 | } | |
7729 | ||
7730 | #if defined(CONFIG_UNIX) | |
7731 | ctx->ring_sock->file = file; | |
7732 | #endif | |
7733 | fd_install(ret, file); | |
7734 | return ret; | |
7735 | err: | |
7736 | #if defined(CONFIG_UNIX) | |
7737 | sock_release(ctx->ring_sock); | |
7738 | ctx->ring_sock = NULL; | |
7739 | #endif | |
7740 | return ret; | |
7741 | } | |
7742 | ||
7f13657d XW |
7743 | static int io_uring_create(unsigned entries, struct io_uring_params *p, |
7744 | struct io_uring_params __user *params) | |
2b188cc1 JA |
7745 | { |
7746 | struct user_struct *user = NULL; | |
7747 | struct io_ring_ctx *ctx; | |
7748 | bool account_mem; | |
7749 | int ret; | |
7750 | ||
8110c1a6 | 7751 | if (!entries) |
2b188cc1 | 7752 | return -EINVAL; |
8110c1a6 JA |
7753 | if (entries > IORING_MAX_ENTRIES) { |
7754 | if (!(p->flags & IORING_SETUP_CLAMP)) | |
7755 | return -EINVAL; | |
7756 | entries = IORING_MAX_ENTRIES; | |
7757 | } | |
2b188cc1 JA |
7758 | |
7759 | /* | |
7760 | * Use twice as many entries for the CQ ring. It's possible for the | |
7761 | * application to drive a higher depth than the size of the SQ ring, | |
7762 | * since the sqes are only used at submission time. This allows for | |
33a107f0 JA |
7763 | * some flexibility in overcommitting a bit. If the application has |
7764 | * set IORING_SETUP_CQSIZE, it will have passed in the desired number | |
7765 | * of CQ ring entries manually. | |
2b188cc1 JA |
7766 | */ |
7767 | p->sq_entries = roundup_pow_of_two(entries); | |
33a107f0 JA |
7768 | if (p->flags & IORING_SETUP_CQSIZE) { |
7769 | /* | |
7770 | * If IORING_SETUP_CQSIZE is set, we do the same roundup | |
7771 | * to a power-of-two, if it isn't already. We do NOT impose | |
7772 | * any cq vs sq ring sizing. | |
7773 | */ | |
8110c1a6 | 7774 | if (p->cq_entries < p->sq_entries) |
33a107f0 | 7775 | return -EINVAL; |
8110c1a6 JA |
7776 | if (p->cq_entries > IORING_MAX_CQ_ENTRIES) { |
7777 | if (!(p->flags & IORING_SETUP_CLAMP)) | |
7778 | return -EINVAL; | |
7779 | p->cq_entries = IORING_MAX_CQ_ENTRIES; | |
7780 | } | |
33a107f0 JA |
7781 | p->cq_entries = roundup_pow_of_two(p->cq_entries); |
7782 | } else { | |
7783 | p->cq_entries = 2 * p->sq_entries; | |
7784 | } | |
2b188cc1 JA |
7785 | |
7786 | user = get_uid(current_user()); | |
7787 | account_mem = !capable(CAP_IPC_LOCK); | |
7788 | ||
7789 | if (account_mem) { | |
7790 | ret = io_account_mem(user, | |
7791 | ring_pages(p->sq_entries, p->cq_entries)); | |
7792 | if (ret) { | |
7793 | free_uid(user); | |
7794 | return ret; | |
7795 | } | |
7796 | } | |
7797 | ||
7798 | ctx = io_ring_ctx_alloc(p); | |
7799 | if (!ctx) { | |
7800 | if (account_mem) | |
7801 | io_unaccount_mem(user, ring_pages(p->sq_entries, | |
7802 | p->cq_entries)); | |
7803 | free_uid(user); | |
7804 | return -ENOMEM; | |
7805 | } | |
7806 | ctx->compat = in_compat_syscall(); | |
7807 | ctx->account_mem = account_mem; | |
7808 | ctx->user = user; | |
0b8c0ec7 | 7809 | ctx->creds = get_current_cred(); |
2b188cc1 JA |
7810 | |
7811 | ret = io_allocate_scq_urings(ctx, p); | |
7812 | if (ret) | |
7813 | goto err; | |
7814 | ||
6c271ce2 | 7815 | ret = io_sq_offload_start(ctx, p); |
2b188cc1 JA |
7816 | if (ret) |
7817 | goto err; | |
7818 | ||
2b188cc1 | 7819 | memset(&p->sq_off, 0, sizeof(p->sq_off)); |
75b28aff HV |
7820 | p->sq_off.head = offsetof(struct io_rings, sq.head); |
7821 | p->sq_off.tail = offsetof(struct io_rings, sq.tail); | |
7822 | p->sq_off.ring_mask = offsetof(struct io_rings, sq_ring_mask); | |
7823 | p->sq_off.ring_entries = offsetof(struct io_rings, sq_ring_entries); | |
7824 | p->sq_off.flags = offsetof(struct io_rings, sq_flags); | |
7825 | p->sq_off.dropped = offsetof(struct io_rings, sq_dropped); | |
7826 | p->sq_off.array = (char *)ctx->sq_array - (char *)ctx->rings; | |
2b188cc1 JA |
7827 | |
7828 | memset(&p->cq_off, 0, sizeof(p->cq_off)); | |
75b28aff HV |
7829 | p->cq_off.head = offsetof(struct io_rings, cq.head); |
7830 | p->cq_off.tail = offsetof(struct io_rings, cq.tail); | |
7831 | p->cq_off.ring_mask = offsetof(struct io_rings, cq_ring_mask); | |
7832 | p->cq_off.ring_entries = offsetof(struct io_rings, cq_ring_entries); | |
7833 | p->cq_off.overflow = offsetof(struct io_rings, cq_overflow); | |
7834 | p->cq_off.cqes = offsetof(struct io_rings, cqes); | |
ac90f249 | 7835 | |
7f13657d XW |
7836 | p->features = IORING_FEAT_SINGLE_MMAP | IORING_FEAT_NODROP | |
7837 | IORING_FEAT_SUBMIT_STABLE | IORING_FEAT_RW_CUR_POS | | |
7838 | IORING_FEAT_CUR_PERSONALITY | IORING_FEAT_FAST_POLL; | |
7839 | ||
7840 | if (copy_to_user(params, p, sizeof(*p))) { | |
7841 | ret = -EFAULT; | |
7842 | goto err; | |
7843 | } | |
044c1ab3 JA |
7844 | /* |
7845 | * Install ring fd as the very last thing, so we don't risk someone | |
7846 | * having closed it before we finish setup | |
7847 | */ | |
7848 | ret = io_uring_get_fd(ctx); | |
7849 | if (ret < 0) | |
7850 | goto err; | |
7851 | ||
c826bd7a | 7852 | trace_io_uring_create(ret, ctx, p->sq_entries, p->cq_entries, p->flags); |
2b188cc1 JA |
7853 | return ret; |
7854 | err: | |
7855 | io_ring_ctx_wait_and_kill(ctx); | |
7856 | return ret; | |
7857 | } | |
7858 | ||
7859 | /* | |
7860 | * Sets up an aio uring context, and returns the fd. Applications asks for a | |
7861 | * ring size, we return the actual sq/cq ring sizes (among other things) in the | |
7862 | * params structure passed in. | |
7863 | */ | |
7864 | static long io_uring_setup(u32 entries, struct io_uring_params __user *params) | |
7865 | { | |
7866 | struct io_uring_params p; | |
2b188cc1 JA |
7867 | int i; |
7868 | ||
7869 | if (copy_from_user(&p, params, sizeof(p))) | |
7870 | return -EFAULT; | |
7871 | for (i = 0; i < ARRAY_SIZE(p.resv); i++) { | |
7872 | if (p.resv[i]) | |
7873 | return -EINVAL; | |
7874 | } | |
7875 | ||
6c271ce2 | 7876 | if (p.flags & ~(IORING_SETUP_IOPOLL | IORING_SETUP_SQPOLL | |
8110c1a6 | 7877 | IORING_SETUP_SQ_AFF | IORING_SETUP_CQSIZE | |
24369c2e | 7878 | IORING_SETUP_CLAMP | IORING_SETUP_ATTACH_WQ)) |
2b188cc1 JA |
7879 | return -EINVAL; |
7880 | ||
7f13657d | 7881 | return io_uring_create(entries, &p, params); |
2b188cc1 JA |
7882 | } |
7883 | ||
7884 | SYSCALL_DEFINE2(io_uring_setup, u32, entries, | |
7885 | struct io_uring_params __user *, params) | |
7886 | { | |
7887 | return io_uring_setup(entries, params); | |
7888 | } | |
7889 | ||
66f4af93 JA |
7890 | static int io_probe(struct io_ring_ctx *ctx, void __user *arg, unsigned nr_args) |
7891 | { | |
7892 | struct io_uring_probe *p; | |
7893 | size_t size; | |
7894 | int i, ret; | |
7895 | ||
7896 | size = struct_size(p, ops, nr_args); | |
7897 | if (size == SIZE_MAX) | |
7898 | return -EOVERFLOW; | |
7899 | p = kzalloc(size, GFP_KERNEL); | |
7900 | if (!p) | |
7901 | return -ENOMEM; | |
7902 | ||
7903 | ret = -EFAULT; | |
7904 | if (copy_from_user(p, arg, size)) | |
7905 | goto out; | |
7906 | ret = -EINVAL; | |
7907 | if (memchr_inv(p, 0, size)) | |
7908 | goto out; | |
7909 | ||
7910 | p->last_op = IORING_OP_LAST - 1; | |
7911 | if (nr_args > IORING_OP_LAST) | |
7912 | nr_args = IORING_OP_LAST; | |
7913 | ||
7914 | for (i = 0; i < nr_args; i++) { | |
7915 | p->ops[i].op = i; | |
7916 | if (!io_op_defs[i].not_supported) | |
7917 | p->ops[i].flags = IO_URING_OP_SUPPORTED; | |
7918 | } | |
7919 | p->ops_len = i; | |
7920 | ||
7921 | ret = 0; | |
7922 | if (copy_to_user(arg, p, size)) | |
7923 | ret = -EFAULT; | |
7924 | out: | |
7925 | kfree(p); | |
7926 | return ret; | |
7927 | } | |
7928 | ||
071698e1 JA |
7929 | static int io_register_personality(struct io_ring_ctx *ctx) |
7930 | { | |
7931 | const struct cred *creds = get_current_cred(); | |
7932 | int id; | |
7933 | ||
7934 | id = idr_alloc_cyclic(&ctx->personality_idr, (void *) creds, 1, | |
7935 | USHRT_MAX, GFP_KERNEL); | |
7936 | if (id < 0) | |
7937 | put_cred(creds); | |
7938 | return id; | |
7939 | } | |
7940 | ||
7941 | static int io_unregister_personality(struct io_ring_ctx *ctx, unsigned id) | |
7942 | { | |
7943 | const struct cred *old_creds; | |
7944 | ||
7945 | old_creds = idr_remove(&ctx->personality_idr, id); | |
7946 | if (old_creds) { | |
7947 | put_cred(old_creds); | |
7948 | return 0; | |
7949 | } | |
7950 | ||
7951 | return -EINVAL; | |
7952 | } | |
7953 | ||
7954 | static bool io_register_op_must_quiesce(int op) | |
7955 | { | |
7956 | switch (op) { | |
7957 | case IORING_UNREGISTER_FILES: | |
7958 | case IORING_REGISTER_FILES_UPDATE: | |
7959 | case IORING_REGISTER_PROBE: | |
7960 | case IORING_REGISTER_PERSONALITY: | |
7961 | case IORING_UNREGISTER_PERSONALITY: | |
7962 | return false; | |
7963 | default: | |
7964 | return true; | |
7965 | } | |
7966 | } | |
7967 | ||
edafccee JA |
7968 | static int __io_uring_register(struct io_ring_ctx *ctx, unsigned opcode, |
7969 | void __user *arg, unsigned nr_args) | |
b19062a5 JA |
7970 | __releases(ctx->uring_lock) |
7971 | __acquires(ctx->uring_lock) | |
edafccee JA |
7972 | { |
7973 | int ret; | |
7974 | ||
35fa71a0 JA |
7975 | /* |
7976 | * We're inside the ring mutex, if the ref is already dying, then | |
7977 | * someone else killed the ctx or is already going through | |
7978 | * io_uring_register(). | |
7979 | */ | |
7980 | if (percpu_ref_is_dying(&ctx->refs)) | |
7981 | return -ENXIO; | |
7982 | ||
071698e1 | 7983 | if (io_register_op_must_quiesce(opcode)) { |
05f3fb3c | 7984 | percpu_ref_kill(&ctx->refs); |
b19062a5 | 7985 | |
05f3fb3c JA |
7986 | /* |
7987 | * Drop uring mutex before waiting for references to exit. If | |
7988 | * another thread is currently inside io_uring_enter() it might | |
7989 | * need to grab the uring_lock to make progress. If we hold it | |
7990 | * here across the drain wait, then we can deadlock. It's safe | |
7991 | * to drop the mutex here, since no new references will come in | |
7992 | * after we've killed the percpu ref. | |
7993 | */ | |
7994 | mutex_unlock(&ctx->uring_lock); | |
c150368b | 7995 | ret = wait_for_completion_interruptible(&ctx->completions[0]); |
05f3fb3c | 7996 | mutex_lock(&ctx->uring_lock); |
c150368b JA |
7997 | if (ret) { |
7998 | percpu_ref_resurrect(&ctx->refs); | |
7999 | ret = -EINTR; | |
8000 | goto out; | |
8001 | } | |
05f3fb3c | 8002 | } |
edafccee JA |
8003 | |
8004 | switch (opcode) { | |
8005 | case IORING_REGISTER_BUFFERS: | |
8006 | ret = io_sqe_buffer_register(ctx, arg, nr_args); | |
8007 | break; | |
8008 | case IORING_UNREGISTER_BUFFERS: | |
8009 | ret = -EINVAL; | |
8010 | if (arg || nr_args) | |
8011 | break; | |
8012 | ret = io_sqe_buffer_unregister(ctx); | |
8013 | break; | |
6b06314c JA |
8014 | case IORING_REGISTER_FILES: |
8015 | ret = io_sqe_files_register(ctx, arg, nr_args); | |
8016 | break; | |
8017 | case IORING_UNREGISTER_FILES: | |
8018 | ret = -EINVAL; | |
8019 | if (arg || nr_args) | |
8020 | break; | |
8021 | ret = io_sqe_files_unregister(ctx); | |
8022 | break; | |
c3a31e60 JA |
8023 | case IORING_REGISTER_FILES_UPDATE: |
8024 | ret = io_sqe_files_update(ctx, arg, nr_args); | |
8025 | break; | |
9b402849 | 8026 | case IORING_REGISTER_EVENTFD: |
f2842ab5 | 8027 | case IORING_REGISTER_EVENTFD_ASYNC: |
9b402849 JA |
8028 | ret = -EINVAL; |
8029 | if (nr_args != 1) | |
8030 | break; | |
8031 | ret = io_eventfd_register(ctx, arg); | |
f2842ab5 JA |
8032 | if (ret) |
8033 | break; | |
8034 | if (opcode == IORING_REGISTER_EVENTFD_ASYNC) | |
8035 | ctx->eventfd_async = 1; | |
8036 | else | |
8037 | ctx->eventfd_async = 0; | |
9b402849 JA |
8038 | break; |
8039 | case IORING_UNREGISTER_EVENTFD: | |
8040 | ret = -EINVAL; | |
8041 | if (arg || nr_args) | |
8042 | break; | |
8043 | ret = io_eventfd_unregister(ctx); | |
8044 | break; | |
66f4af93 JA |
8045 | case IORING_REGISTER_PROBE: |
8046 | ret = -EINVAL; | |
8047 | if (!arg || nr_args > 256) | |
8048 | break; | |
8049 | ret = io_probe(ctx, arg, nr_args); | |
8050 | break; | |
071698e1 JA |
8051 | case IORING_REGISTER_PERSONALITY: |
8052 | ret = -EINVAL; | |
8053 | if (arg || nr_args) | |
8054 | break; | |
8055 | ret = io_register_personality(ctx); | |
8056 | break; | |
8057 | case IORING_UNREGISTER_PERSONALITY: | |
8058 | ret = -EINVAL; | |
8059 | if (arg) | |
8060 | break; | |
8061 | ret = io_unregister_personality(ctx, nr_args); | |
8062 | break; | |
edafccee JA |
8063 | default: |
8064 | ret = -EINVAL; | |
8065 | break; | |
8066 | } | |
8067 | ||
071698e1 | 8068 | if (io_register_op_must_quiesce(opcode)) { |
05f3fb3c | 8069 | /* bring the ctx back to life */ |
05f3fb3c | 8070 | percpu_ref_reinit(&ctx->refs); |
c150368b JA |
8071 | out: |
8072 | reinit_completion(&ctx->completions[0]); | |
05f3fb3c | 8073 | } |
edafccee JA |
8074 | return ret; |
8075 | } | |
8076 | ||
8077 | SYSCALL_DEFINE4(io_uring_register, unsigned int, fd, unsigned int, opcode, | |
8078 | void __user *, arg, unsigned int, nr_args) | |
8079 | { | |
8080 | struct io_ring_ctx *ctx; | |
8081 | long ret = -EBADF; | |
8082 | struct fd f; | |
8083 | ||
8084 | f = fdget(fd); | |
8085 | if (!f.file) | |
8086 | return -EBADF; | |
8087 | ||
8088 | ret = -EOPNOTSUPP; | |
8089 | if (f.file->f_op != &io_uring_fops) | |
8090 | goto out_fput; | |
8091 | ||
8092 | ctx = f.file->private_data; | |
8093 | ||
8094 | mutex_lock(&ctx->uring_lock); | |
8095 | ret = __io_uring_register(ctx, opcode, arg, nr_args); | |
8096 | mutex_unlock(&ctx->uring_lock); | |
c826bd7a DD |
8097 | trace_io_uring_register(ctx, opcode, ctx->nr_user_files, ctx->nr_user_bufs, |
8098 | ctx->cq_ev_fd != NULL, ret); | |
edafccee JA |
8099 | out_fput: |
8100 | fdput(f); | |
8101 | return ret; | |
8102 | } | |
8103 | ||
2b188cc1 JA |
8104 | static int __init io_uring_init(void) |
8105 | { | |
d7f62e82 SM |
8106 | #define __BUILD_BUG_VERIFY_ELEMENT(stype, eoffset, etype, ename) do { \ |
8107 | BUILD_BUG_ON(offsetof(stype, ename) != eoffset); \ | |
8108 | BUILD_BUG_ON(sizeof(etype) != sizeof_field(stype, ename)); \ | |
8109 | } while (0) | |
8110 | ||
8111 | #define BUILD_BUG_SQE_ELEM(eoffset, etype, ename) \ | |
8112 | __BUILD_BUG_VERIFY_ELEMENT(struct io_uring_sqe, eoffset, etype, ename) | |
8113 | BUILD_BUG_ON(sizeof(struct io_uring_sqe) != 64); | |
8114 | BUILD_BUG_SQE_ELEM(0, __u8, opcode); | |
8115 | BUILD_BUG_SQE_ELEM(1, __u8, flags); | |
8116 | BUILD_BUG_SQE_ELEM(2, __u16, ioprio); | |
8117 | BUILD_BUG_SQE_ELEM(4, __s32, fd); | |
8118 | BUILD_BUG_SQE_ELEM(8, __u64, off); | |
8119 | BUILD_BUG_SQE_ELEM(8, __u64, addr2); | |
8120 | BUILD_BUG_SQE_ELEM(16, __u64, addr); | |
7d67af2c | 8121 | BUILD_BUG_SQE_ELEM(16, __u64, splice_off_in); |
d7f62e82 SM |
8122 | BUILD_BUG_SQE_ELEM(24, __u32, len); |
8123 | BUILD_BUG_SQE_ELEM(28, __kernel_rwf_t, rw_flags); | |
8124 | BUILD_BUG_SQE_ELEM(28, /* compat */ int, rw_flags); | |
8125 | BUILD_BUG_SQE_ELEM(28, /* compat */ __u32, rw_flags); | |
8126 | BUILD_BUG_SQE_ELEM(28, __u32, fsync_flags); | |
8127 | BUILD_BUG_SQE_ELEM(28, __u16, poll_events); | |
8128 | BUILD_BUG_SQE_ELEM(28, __u32, sync_range_flags); | |
8129 | BUILD_BUG_SQE_ELEM(28, __u32, msg_flags); | |
8130 | BUILD_BUG_SQE_ELEM(28, __u32, timeout_flags); | |
8131 | BUILD_BUG_SQE_ELEM(28, __u32, accept_flags); | |
8132 | BUILD_BUG_SQE_ELEM(28, __u32, cancel_flags); | |
8133 | BUILD_BUG_SQE_ELEM(28, __u32, open_flags); | |
8134 | BUILD_BUG_SQE_ELEM(28, __u32, statx_flags); | |
8135 | BUILD_BUG_SQE_ELEM(28, __u32, fadvise_advice); | |
7d67af2c | 8136 | BUILD_BUG_SQE_ELEM(28, __u32, splice_flags); |
d7f62e82 SM |
8137 | BUILD_BUG_SQE_ELEM(32, __u64, user_data); |
8138 | BUILD_BUG_SQE_ELEM(40, __u16, buf_index); | |
8139 | BUILD_BUG_SQE_ELEM(42, __u16, personality); | |
7d67af2c | 8140 | BUILD_BUG_SQE_ELEM(44, __s32, splice_fd_in); |
d7f62e82 | 8141 | |
d3656344 | 8142 | BUILD_BUG_ON(ARRAY_SIZE(io_op_defs) != IORING_OP_LAST); |
84557871 | 8143 | BUILD_BUG_ON(__REQ_F_LAST_BIT >= 8 * sizeof(int)); |
2b188cc1 JA |
8144 | req_cachep = KMEM_CACHE(io_kiocb, SLAB_HWCACHE_ALIGN | SLAB_PANIC); |
8145 | return 0; | |
8146 | }; | |
8147 | __initcall(io_uring_init); |