]>
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] = { |
d3656344 | 789 | .needs_file = 1, |
f86cd20c | 790 | .file_table = 1, |
d3656344 | 791 | }, |
0463b6c5 | 792 | [IORING_OP_FILES_UPDATE] = { |
d3656344 | 793 | .needs_mm = 1, |
f86cd20c | 794 | .file_table = 1, |
d3656344 | 795 | }, |
0463b6c5 | 796 | [IORING_OP_STATX] = { |
d3656344 | 797 | .needs_mm = 1, |
ff002b30 | 798 | .needs_fs = 1, |
5b0bbee4 | 799 | .file_table = 1, |
d3656344 | 800 | }, |
0463b6c5 | 801 | [IORING_OP_READ] = { |
3a6820f2 JA |
802 | .needs_mm = 1, |
803 | .needs_file = 1, | |
804 | .unbound_nonreg_file = 1, | |
8a72758c | 805 | .pollin = 1, |
bcda7baa | 806 | .buffer_select = 1, |
3a6820f2 | 807 | }, |
0463b6c5 | 808 | [IORING_OP_WRITE] = { |
3a6820f2 JA |
809 | .needs_mm = 1, |
810 | .needs_file = 1, | |
811 | .unbound_nonreg_file = 1, | |
8a72758c | 812 | .pollout = 1, |
3a6820f2 | 813 | }, |
0463b6c5 | 814 | [IORING_OP_FADVISE] = { |
4840e418 JA |
815 | .needs_file = 1, |
816 | }, | |
0463b6c5 | 817 | [IORING_OP_MADVISE] = { |
c1ca757b JA |
818 | .needs_mm = 1, |
819 | }, | |
0463b6c5 | 820 | [IORING_OP_SEND] = { |
fddaface JA |
821 | .needs_mm = 1, |
822 | .needs_file = 1, | |
823 | .unbound_nonreg_file = 1, | |
8a72758c | 824 | .pollout = 1, |
fddaface | 825 | }, |
0463b6c5 | 826 | [IORING_OP_RECV] = { |
fddaface JA |
827 | .needs_mm = 1, |
828 | .needs_file = 1, | |
829 | .unbound_nonreg_file = 1, | |
8a72758c | 830 | .pollin = 1, |
bcda7baa | 831 | .buffer_select = 1, |
fddaface | 832 | }, |
0463b6c5 | 833 | [IORING_OP_OPENAT2] = { |
f86cd20c | 834 | .file_table = 1, |
ff002b30 | 835 | .needs_fs = 1, |
cebdb986 | 836 | }, |
3e4827b0 JA |
837 | [IORING_OP_EPOLL_CTL] = { |
838 | .unbound_nonreg_file = 1, | |
839 | .file_table = 1, | |
840 | }, | |
7d67af2c PB |
841 | [IORING_OP_SPLICE] = { |
842 | .needs_file = 1, | |
843 | .hash_reg_file = 1, | |
844 | .unbound_nonreg_file = 1, | |
ddf0322d JA |
845 | }, |
846 | [IORING_OP_PROVIDE_BUFFERS] = {}, | |
067524e9 | 847 | [IORING_OP_REMOVE_BUFFERS] = {}, |
d3656344 JA |
848 | }; |
849 | ||
561fb04a | 850 | static void io_wq_submit_work(struct io_wq_work **workptr); |
78e19bbe | 851 | static void io_cqring_fill_event(struct io_kiocb *req, long res); |
ec9c02ad | 852 | static void io_put_req(struct io_kiocb *req); |
978db57e | 853 | static void __io_double_put_req(struct io_kiocb *req); |
94ae5e77 JA |
854 | static struct io_kiocb *io_prep_linked_timeout(struct io_kiocb *req); |
855 | static void io_queue_linked_timeout(struct io_kiocb *req); | |
05f3fb3c JA |
856 | static int __io_sqe_files_update(struct io_ring_ctx *ctx, |
857 | struct io_uring_files_update *ip, | |
858 | unsigned nr_args); | |
f86cd20c | 859 | static int io_grab_files(struct io_kiocb *req); |
99bc4c38 | 860 | static void io_cleanup_req(struct io_kiocb *req); |
b41e9852 JA |
861 | static int io_file_get(struct io_submit_state *state, struct io_kiocb *req, |
862 | int fd, struct file **out_file, bool fixed); | |
863 | static void __io_queue_sqe(struct io_kiocb *req, | |
864 | const struct io_uring_sqe *sqe); | |
de0617e4 | 865 | |
2b188cc1 JA |
866 | static struct kmem_cache *req_cachep; |
867 | ||
868 | static const struct file_operations io_uring_fops; | |
869 | ||
870 | struct sock *io_uring_get_socket(struct file *file) | |
871 | { | |
872 | #if defined(CONFIG_UNIX) | |
873 | if (file->f_op == &io_uring_fops) { | |
874 | struct io_ring_ctx *ctx = file->private_data; | |
875 | ||
876 | return ctx->ring_sock->sk; | |
877 | } | |
878 | #endif | |
879 | return NULL; | |
880 | } | |
881 | EXPORT_SYMBOL(io_uring_get_socket); | |
882 | ||
883 | static void io_ring_ctx_ref_free(struct percpu_ref *ref) | |
884 | { | |
885 | struct io_ring_ctx *ctx = container_of(ref, struct io_ring_ctx, refs); | |
886 | ||
206aefde | 887 | complete(&ctx->completions[0]); |
2b188cc1 JA |
888 | } |
889 | ||
890 | static struct io_ring_ctx *io_ring_ctx_alloc(struct io_uring_params *p) | |
891 | { | |
892 | struct io_ring_ctx *ctx; | |
78076bb6 | 893 | int hash_bits; |
2b188cc1 JA |
894 | |
895 | ctx = kzalloc(sizeof(*ctx), GFP_KERNEL); | |
896 | if (!ctx) | |
897 | return NULL; | |
898 | ||
0ddf92e8 JA |
899 | ctx->fallback_req = kmem_cache_alloc(req_cachep, GFP_KERNEL); |
900 | if (!ctx->fallback_req) | |
901 | goto err; | |
902 | ||
206aefde JA |
903 | ctx->completions = kmalloc(2 * sizeof(struct completion), GFP_KERNEL); |
904 | if (!ctx->completions) | |
905 | goto err; | |
906 | ||
78076bb6 JA |
907 | /* |
908 | * Use 5 bits less than the max cq entries, that should give us around | |
909 | * 32 entries per hash list if totally full and uniformly spread. | |
910 | */ | |
911 | hash_bits = ilog2(p->cq_entries); | |
912 | hash_bits -= 5; | |
913 | if (hash_bits <= 0) | |
914 | hash_bits = 1; | |
915 | ctx->cancel_hash_bits = hash_bits; | |
916 | ctx->cancel_hash = kmalloc((1U << hash_bits) * sizeof(struct hlist_head), | |
917 | GFP_KERNEL); | |
918 | if (!ctx->cancel_hash) | |
919 | goto err; | |
920 | __hash_init(ctx->cancel_hash, 1U << hash_bits); | |
921 | ||
21482896 | 922 | if (percpu_ref_init(&ctx->refs, io_ring_ctx_ref_free, |
206aefde JA |
923 | PERCPU_REF_ALLOW_REINIT, GFP_KERNEL)) |
924 | goto err; | |
2b188cc1 JA |
925 | |
926 | ctx->flags = p->flags; | |
927 | init_waitqueue_head(&ctx->cq_wait); | |
1d7bb1d5 | 928 | INIT_LIST_HEAD(&ctx->cq_overflow_list); |
206aefde JA |
929 | init_completion(&ctx->completions[0]); |
930 | init_completion(&ctx->completions[1]); | |
5a2e745d | 931 | idr_init(&ctx->io_buffer_idr); |
071698e1 | 932 | idr_init(&ctx->personality_idr); |
2b188cc1 JA |
933 | mutex_init(&ctx->uring_lock); |
934 | init_waitqueue_head(&ctx->wait); | |
935 | spin_lock_init(&ctx->completion_lock); | |
def596e9 | 936 | INIT_LIST_HEAD(&ctx->poll_list); |
de0617e4 | 937 | INIT_LIST_HEAD(&ctx->defer_list); |
5262f567 | 938 | INIT_LIST_HEAD(&ctx->timeout_list); |
fcb323cc JA |
939 | init_waitqueue_head(&ctx->inflight_wait); |
940 | spin_lock_init(&ctx->inflight_lock); | |
941 | INIT_LIST_HEAD(&ctx->inflight_list); | |
2b188cc1 | 942 | return ctx; |
206aefde | 943 | err: |
0ddf92e8 JA |
944 | if (ctx->fallback_req) |
945 | kmem_cache_free(req_cachep, ctx->fallback_req); | |
206aefde | 946 | kfree(ctx->completions); |
78076bb6 | 947 | kfree(ctx->cancel_hash); |
206aefde JA |
948 | kfree(ctx); |
949 | return NULL; | |
2b188cc1 JA |
950 | } |
951 | ||
9d858b21 | 952 | static inline bool __req_need_defer(struct io_kiocb *req) |
7adf4eaf | 953 | { |
a197f664 JL |
954 | struct io_ring_ctx *ctx = req->ctx; |
955 | ||
31af27c7 PB |
956 | return req->sequence != ctx->cached_cq_tail |
957 | + atomic_read(&ctx->cached_cq_overflow); | |
7adf4eaf JA |
958 | } |
959 | ||
9d858b21 | 960 | static inline bool req_need_defer(struct io_kiocb *req) |
de0617e4 | 961 | { |
87987898 | 962 | if (unlikely(req->flags & REQ_F_IO_DRAIN)) |
9d858b21 | 963 | return __req_need_defer(req); |
de0617e4 | 964 | |
9d858b21 | 965 | return false; |
de0617e4 JA |
966 | } |
967 | ||
7adf4eaf | 968 | static struct io_kiocb *io_get_deferred_req(struct io_ring_ctx *ctx) |
de0617e4 JA |
969 | { |
970 | struct io_kiocb *req; | |
971 | ||
7adf4eaf | 972 | req = list_first_entry_or_null(&ctx->defer_list, struct io_kiocb, list); |
9d858b21 | 973 | if (req && !req_need_defer(req)) { |
de0617e4 JA |
974 | list_del_init(&req->list); |
975 | return req; | |
976 | } | |
977 | ||
978 | return NULL; | |
979 | } | |
980 | ||
5262f567 JA |
981 | static struct io_kiocb *io_get_timeout_req(struct io_ring_ctx *ctx) |
982 | { | |
7adf4eaf JA |
983 | struct io_kiocb *req; |
984 | ||
985 | req = list_first_entry_or_null(&ctx->timeout_list, struct io_kiocb, list); | |
93bd25bb JA |
986 | if (req) { |
987 | if (req->flags & REQ_F_TIMEOUT_NOSEQ) | |
988 | return NULL; | |
fb4b3d3f | 989 | if (!__req_need_defer(req)) { |
93bd25bb JA |
990 | list_del_init(&req->list); |
991 | return req; | |
992 | } | |
7adf4eaf JA |
993 | } |
994 | ||
995 | return NULL; | |
5262f567 JA |
996 | } |
997 | ||
de0617e4 | 998 | static void __io_commit_cqring(struct io_ring_ctx *ctx) |
2b188cc1 | 999 | { |
75b28aff | 1000 | struct io_rings *rings = ctx->rings; |
2b188cc1 | 1001 | |
07910158 PB |
1002 | /* order cqe stores with ring update */ |
1003 | smp_store_release(&rings->cq.tail, ctx->cached_cq_tail); | |
2b188cc1 | 1004 | |
07910158 PB |
1005 | if (wq_has_sleeper(&ctx->cq_wait)) { |
1006 | wake_up_interruptible(&ctx->cq_wait); | |
1007 | kill_fasync(&ctx->cq_fasync, SIGIO, POLL_IN); | |
2b188cc1 JA |
1008 | } |
1009 | } | |
1010 | ||
cccf0ee8 JA |
1011 | static inline void io_req_work_grab_env(struct io_kiocb *req, |
1012 | const struct io_op_def *def) | |
1013 | { | |
1014 | if (!req->work.mm && def->needs_mm) { | |
1015 | mmgrab(current->mm); | |
1016 | req->work.mm = current->mm; | |
2b188cc1 | 1017 | } |
cccf0ee8 JA |
1018 | if (!req->work.creds) |
1019 | req->work.creds = get_current_cred(); | |
ff002b30 JA |
1020 | if (!req->work.fs && def->needs_fs) { |
1021 | spin_lock(¤t->fs->lock); | |
1022 | if (!current->fs->in_exec) { | |
1023 | req->work.fs = current->fs; | |
1024 | req->work.fs->users++; | |
1025 | } else { | |
1026 | req->work.flags |= IO_WQ_WORK_CANCEL; | |
1027 | } | |
1028 | spin_unlock(¤t->fs->lock); | |
1029 | } | |
6ab23144 JA |
1030 | if (!req->work.task_pid) |
1031 | req->work.task_pid = task_pid_vnr(current); | |
2b188cc1 JA |
1032 | } |
1033 | ||
cccf0ee8 | 1034 | static inline void io_req_work_drop_env(struct io_kiocb *req) |
18d9be1a | 1035 | { |
cccf0ee8 JA |
1036 | if (req->work.mm) { |
1037 | mmdrop(req->work.mm); | |
1038 | req->work.mm = NULL; | |
1039 | } | |
1040 | if (req->work.creds) { | |
1041 | put_cred(req->work.creds); | |
1042 | req->work.creds = NULL; | |
1043 | } | |
ff002b30 JA |
1044 | if (req->work.fs) { |
1045 | struct fs_struct *fs = req->work.fs; | |
1046 | ||
1047 | spin_lock(&req->work.fs->lock); | |
1048 | if (--fs->users) | |
1049 | fs = NULL; | |
1050 | spin_unlock(&req->work.fs->lock); | |
1051 | if (fs) | |
1052 | free_fs_struct(fs); | |
1053 | } | |
561fb04a JA |
1054 | } |
1055 | ||
8766dd51 | 1056 | static inline void io_prep_async_work(struct io_kiocb *req, |
94ae5e77 | 1057 | struct io_kiocb **link) |
18d9be1a | 1058 | { |
d3656344 | 1059 | const struct io_op_def *def = &io_op_defs[req->opcode]; |
54a91f3b | 1060 | |
d3656344 JA |
1061 | if (req->flags & REQ_F_ISREG) { |
1062 | if (def->hash_reg_file) | |
8766dd51 | 1063 | io_wq_hash_work(&req->work, file_inode(req->file)); |
d3656344 JA |
1064 | } else { |
1065 | if (def->unbound_nonreg_file) | |
3529d8c2 | 1066 | req->work.flags |= IO_WQ_WORK_UNBOUND; |
54a91f3b | 1067 | } |
cccf0ee8 JA |
1068 | |
1069 | io_req_work_grab_env(req, def); | |
54a91f3b | 1070 | |
94ae5e77 | 1071 | *link = io_prep_linked_timeout(req); |
561fb04a JA |
1072 | } |
1073 | ||
a197f664 | 1074 | static inline void io_queue_async_work(struct io_kiocb *req) |
561fb04a | 1075 | { |
a197f664 | 1076 | struct io_ring_ctx *ctx = req->ctx; |
94ae5e77 | 1077 | struct io_kiocb *link; |
94ae5e77 | 1078 | |
8766dd51 | 1079 | io_prep_async_work(req, &link); |
561fb04a | 1080 | |
8766dd51 PB |
1081 | trace_io_uring_queue_async_work(ctx, io_wq_is_hashed(&req->work), req, |
1082 | &req->work, req->flags); | |
1083 | io_wq_enqueue(ctx->io_wq, &req->work); | |
94ae5e77 JA |
1084 | |
1085 | if (link) | |
1086 | io_queue_linked_timeout(link); | |
18d9be1a JA |
1087 | } |
1088 | ||
5262f567 JA |
1089 | static void io_kill_timeout(struct io_kiocb *req) |
1090 | { | |
1091 | int ret; | |
1092 | ||
2d28390a | 1093 | ret = hrtimer_try_to_cancel(&req->io->timeout.timer); |
5262f567 JA |
1094 | if (ret != -1) { |
1095 | atomic_inc(&req->ctx->cq_timeouts); | |
842f9612 | 1096 | list_del_init(&req->list); |
f0e20b89 | 1097 | req->flags |= REQ_F_COMP_LOCKED; |
78e19bbe | 1098 | io_cqring_fill_event(req, 0); |
ec9c02ad | 1099 | io_put_req(req); |
5262f567 JA |
1100 | } |
1101 | } | |
1102 | ||
1103 | static void io_kill_timeouts(struct io_ring_ctx *ctx) | |
1104 | { | |
1105 | struct io_kiocb *req, *tmp; | |
1106 | ||
1107 | spin_lock_irq(&ctx->completion_lock); | |
1108 | list_for_each_entry_safe(req, tmp, &ctx->timeout_list, list) | |
1109 | io_kill_timeout(req); | |
1110 | spin_unlock_irq(&ctx->completion_lock); | |
1111 | } | |
1112 | ||
de0617e4 JA |
1113 | static void io_commit_cqring(struct io_ring_ctx *ctx) |
1114 | { | |
1115 | struct io_kiocb *req; | |
1116 | ||
5262f567 JA |
1117 | while ((req = io_get_timeout_req(ctx)) != NULL) |
1118 | io_kill_timeout(req); | |
1119 | ||
de0617e4 JA |
1120 | __io_commit_cqring(ctx); |
1121 | ||
87987898 | 1122 | while ((req = io_get_deferred_req(ctx)) != NULL) |
a197f664 | 1123 | io_queue_async_work(req); |
de0617e4 JA |
1124 | } |
1125 | ||
2b188cc1 JA |
1126 | static struct io_uring_cqe *io_get_cqring(struct io_ring_ctx *ctx) |
1127 | { | |
75b28aff | 1128 | struct io_rings *rings = ctx->rings; |
2b188cc1 JA |
1129 | unsigned tail; |
1130 | ||
1131 | tail = ctx->cached_cq_tail; | |
115e12e5 SB |
1132 | /* |
1133 | * writes to the cq entry need to come after reading head; the | |
1134 | * control dependency is enough as we're using WRITE_ONCE to | |
1135 | * fill the cq entry | |
1136 | */ | |
75b28aff | 1137 | if (tail - READ_ONCE(rings->cq.head) == rings->cq_ring_entries) |
2b188cc1 JA |
1138 | return NULL; |
1139 | ||
1140 | ctx->cached_cq_tail++; | |
75b28aff | 1141 | return &rings->cqes[tail & ctx->cq_mask]; |
2b188cc1 JA |
1142 | } |
1143 | ||
f2842ab5 JA |
1144 | static inline bool io_should_trigger_evfd(struct io_ring_ctx *ctx) |
1145 | { | |
f0b493e6 JA |
1146 | if (!ctx->cq_ev_fd) |
1147 | return false; | |
f2842ab5 JA |
1148 | if (!ctx->eventfd_async) |
1149 | return true; | |
b41e9852 | 1150 | return io_wq_current_is_worker(); |
f2842ab5 JA |
1151 | } |
1152 | ||
b41e9852 | 1153 | static void io_cqring_ev_posted(struct io_ring_ctx *ctx) |
1d7bb1d5 JA |
1154 | { |
1155 | if (waitqueue_active(&ctx->wait)) | |
1156 | wake_up(&ctx->wait); | |
1157 | if (waitqueue_active(&ctx->sqo_wait)) | |
1158 | wake_up(&ctx->sqo_wait); | |
b41e9852 | 1159 | if (io_should_trigger_evfd(ctx)) |
1d7bb1d5 JA |
1160 | eventfd_signal(ctx->cq_ev_fd, 1); |
1161 | } | |
1162 | ||
c4a2ed72 JA |
1163 | /* Returns true if there are no backlogged entries after the flush */ |
1164 | static bool io_cqring_overflow_flush(struct io_ring_ctx *ctx, bool force) | |
1d7bb1d5 JA |
1165 | { |
1166 | struct io_rings *rings = ctx->rings; | |
1167 | struct io_uring_cqe *cqe; | |
1168 | struct io_kiocb *req; | |
1169 | unsigned long flags; | |
1170 | LIST_HEAD(list); | |
1171 | ||
1172 | if (!force) { | |
1173 | if (list_empty_careful(&ctx->cq_overflow_list)) | |
c4a2ed72 | 1174 | return true; |
1d7bb1d5 JA |
1175 | if ((ctx->cached_cq_tail - READ_ONCE(rings->cq.head) == |
1176 | rings->cq_ring_entries)) | |
c4a2ed72 | 1177 | return false; |
1d7bb1d5 JA |
1178 | } |
1179 | ||
1180 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
1181 | ||
1182 | /* if force is set, the ring is going away. always drop after that */ | |
1183 | if (force) | |
69b3e546 | 1184 | ctx->cq_overflow_flushed = 1; |
1d7bb1d5 | 1185 | |
c4a2ed72 | 1186 | cqe = NULL; |
1d7bb1d5 JA |
1187 | while (!list_empty(&ctx->cq_overflow_list)) { |
1188 | cqe = io_get_cqring(ctx); | |
1189 | if (!cqe && !force) | |
1190 | break; | |
1191 | ||
1192 | req = list_first_entry(&ctx->cq_overflow_list, struct io_kiocb, | |
1193 | list); | |
1194 | list_move(&req->list, &list); | |
2ca10259 | 1195 | req->flags &= ~REQ_F_OVERFLOW; |
1d7bb1d5 JA |
1196 | if (cqe) { |
1197 | WRITE_ONCE(cqe->user_data, req->user_data); | |
1198 | WRITE_ONCE(cqe->res, req->result); | |
bcda7baa | 1199 | WRITE_ONCE(cqe->flags, req->cflags); |
1d7bb1d5 JA |
1200 | } else { |
1201 | WRITE_ONCE(ctx->rings->cq_overflow, | |
1202 | atomic_inc_return(&ctx->cached_cq_overflow)); | |
1203 | } | |
1204 | } | |
1205 | ||
1206 | io_commit_cqring(ctx); | |
ad3eb2c8 JA |
1207 | if (cqe) { |
1208 | clear_bit(0, &ctx->sq_check_overflow); | |
1209 | clear_bit(0, &ctx->cq_check_overflow); | |
1210 | } | |
1d7bb1d5 JA |
1211 | spin_unlock_irqrestore(&ctx->completion_lock, flags); |
1212 | io_cqring_ev_posted(ctx); | |
1213 | ||
1214 | while (!list_empty(&list)) { | |
1215 | req = list_first_entry(&list, struct io_kiocb, list); | |
1216 | list_del(&req->list); | |
ec9c02ad | 1217 | io_put_req(req); |
1d7bb1d5 | 1218 | } |
c4a2ed72 JA |
1219 | |
1220 | return cqe != NULL; | |
1d7bb1d5 JA |
1221 | } |
1222 | ||
bcda7baa | 1223 | static void __io_cqring_fill_event(struct io_kiocb *req, long res, long cflags) |
2b188cc1 | 1224 | { |
78e19bbe | 1225 | struct io_ring_ctx *ctx = req->ctx; |
2b188cc1 JA |
1226 | struct io_uring_cqe *cqe; |
1227 | ||
78e19bbe | 1228 | trace_io_uring_complete(ctx, req->user_data, res); |
51c3ff62 | 1229 | |
2b188cc1 JA |
1230 | /* |
1231 | * If we can't get a cq entry, userspace overflowed the | |
1232 | * submission (by quite a lot). Increment the overflow count in | |
1233 | * the ring. | |
1234 | */ | |
1235 | cqe = io_get_cqring(ctx); | |
1d7bb1d5 | 1236 | if (likely(cqe)) { |
78e19bbe | 1237 | WRITE_ONCE(cqe->user_data, req->user_data); |
2b188cc1 | 1238 | WRITE_ONCE(cqe->res, res); |
bcda7baa | 1239 | WRITE_ONCE(cqe->flags, cflags); |
1d7bb1d5 | 1240 | } else if (ctx->cq_overflow_flushed) { |
498ccd9e JA |
1241 | WRITE_ONCE(ctx->rings->cq_overflow, |
1242 | atomic_inc_return(&ctx->cached_cq_overflow)); | |
1d7bb1d5 | 1243 | } else { |
ad3eb2c8 JA |
1244 | if (list_empty(&ctx->cq_overflow_list)) { |
1245 | set_bit(0, &ctx->sq_check_overflow); | |
1246 | set_bit(0, &ctx->cq_check_overflow); | |
1247 | } | |
2ca10259 | 1248 | req->flags |= REQ_F_OVERFLOW; |
1d7bb1d5 JA |
1249 | refcount_inc(&req->refs); |
1250 | req->result = res; | |
bcda7baa | 1251 | req->cflags = cflags; |
1d7bb1d5 | 1252 | list_add_tail(&req->list, &ctx->cq_overflow_list); |
2b188cc1 JA |
1253 | } |
1254 | } | |
1255 | ||
bcda7baa JA |
1256 | static void io_cqring_fill_event(struct io_kiocb *req, long res) |
1257 | { | |
1258 | __io_cqring_fill_event(req, res, 0); | |
1259 | } | |
1260 | ||
1261 | static void __io_cqring_add_event(struct io_kiocb *req, long res, long cflags) | |
2b188cc1 | 1262 | { |
78e19bbe | 1263 | struct io_ring_ctx *ctx = req->ctx; |
2b188cc1 JA |
1264 | unsigned long flags; |
1265 | ||
1266 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
bcda7baa | 1267 | __io_cqring_fill_event(req, res, cflags); |
2b188cc1 JA |
1268 | io_commit_cqring(ctx); |
1269 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
1270 | ||
8c838788 | 1271 | io_cqring_ev_posted(ctx); |
2b188cc1 JA |
1272 | } |
1273 | ||
bcda7baa JA |
1274 | static void io_cqring_add_event(struct io_kiocb *req, long res) |
1275 | { | |
1276 | __io_cqring_add_event(req, res, 0); | |
1277 | } | |
1278 | ||
0ddf92e8 JA |
1279 | static inline bool io_is_fallback_req(struct io_kiocb *req) |
1280 | { | |
1281 | return req == (struct io_kiocb *) | |
1282 | ((unsigned long) req->ctx->fallback_req & ~1UL); | |
1283 | } | |
1284 | ||
1285 | static struct io_kiocb *io_get_fallback_req(struct io_ring_ctx *ctx) | |
1286 | { | |
1287 | struct io_kiocb *req; | |
1288 | ||
1289 | req = ctx->fallback_req; | |
dd461af6 | 1290 | if (!test_and_set_bit_lock(0, (unsigned long *) &ctx->fallback_req)) |
0ddf92e8 JA |
1291 | return req; |
1292 | ||
1293 | return NULL; | |
1294 | } | |
1295 | ||
0553b8bd PB |
1296 | static struct io_kiocb *io_alloc_req(struct io_ring_ctx *ctx, |
1297 | struct io_submit_state *state) | |
2b188cc1 | 1298 | { |
fd6fab2c | 1299 | gfp_t gfp = GFP_KERNEL | __GFP_NOWARN; |
2b188cc1 JA |
1300 | struct io_kiocb *req; |
1301 | ||
2579f913 | 1302 | if (!state) { |
fd6fab2c | 1303 | req = kmem_cache_alloc(req_cachep, gfp); |
2579f913 | 1304 | if (unlikely(!req)) |
0ddf92e8 | 1305 | goto fallback; |
2579f913 JA |
1306 | } else if (!state->free_reqs) { |
1307 | size_t sz; | |
1308 | int ret; | |
1309 | ||
1310 | sz = min_t(size_t, state->ios_left, ARRAY_SIZE(state->reqs)); | |
fd6fab2c JA |
1311 | ret = kmem_cache_alloc_bulk(req_cachep, gfp, sz, state->reqs); |
1312 | ||
1313 | /* | |
1314 | * Bulk alloc is all-or-nothing. If we fail to get a batch, | |
1315 | * retry single alloc to be on the safe side. | |
1316 | */ | |
1317 | if (unlikely(ret <= 0)) { | |
1318 | state->reqs[0] = kmem_cache_alloc(req_cachep, gfp); | |
1319 | if (!state->reqs[0]) | |
0ddf92e8 | 1320 | goto fallback; |
fd6fab2c JA |
1321 | ret = 1; |
1322 | } | |
2579f913 | 1323 | state->free_reqs = ret - 1; |
6c8a3134 | 1324 | req = state->reqs[ret - 1]; |
2579f913 | 1325 | } else { |
2579f913 | 1326 | state->free_reqs--; |
6c8a3134 | 1327 | req = state->reqs[state->free_reqs]; |
2b188cc1 JA |
1328 | } |
1329 | ||
2579f913 | 1330 | return req; |
0ddf92e8 | 1331 | fallback: |
0553b8bd | 1332 | return io_get_fallback_req(ctx); |
2b188cc1 JA |
1333 | } |
1334 | ||
8da11c19 PB |
1335 | static inline void io_put_file(struct io_kiocb *req, struct file *file, |
1336 | bool fixed) | |
1337 | { | |
1338 | if (fixed) | |
05589553 | 1339 | percpu_ref_put(req->fixed_file_refs); |
8da11c19 PB |
1340 | else |
1341 | fput(file); | |
1342 | } | |
1343 | ||
c6ca97b3 | 1344 | static void __io_req_aux_free(struct io_kiocb *req) |
2b188cc1 | 1345 | { |
929a3af9 PB |
1346 | if (req->flags & REQ_F_NEED_CLEANUP) |
1347 | io_cleanup_req(req); | |
1348 | ||
96fd84d8 | 1349 | kfree(req->io); |
8da11c19 PB |
1350 | if (req->file) |
1351 | io_put_file(req, req->file, (req->flags & REQ_F_FIXED_FILE)); | |
3537b6a7 JA |
1352 | if (req->task) |
1353 | put_task_struct(req->task); | |
cccf0ee8 JA |
1354 | |
1355 | io_req_work_drop_env(req); | |
def596e9 JA |
1356 | } |
1357 | ||
9e645e11 | 1358 | static void __io_free_req(struct io_kiocb *req) |
2b188cc1 | 1359 | { |
c6ca97b3 | 1360 | __io_req_aux_free(req); |
fcb323cc | 1361 | |
fcb323cc | 1362 | if (req->flags & REQ_F_INFLIGHT) { |
c6ca97b3 | 1363 | struct io_ring_ctx *ctx = req->ctx; |
fcb323cc JA |
1364 | unsigned long flags; |
1365 | ||
1366 | spin_lock_irqsave(&ctx->inflight_lock, flags); | |
1367 | list_del(&req->inflight_entry); | |
1368 | if (waitqueue_active(&ctx->inflight_wait)) | |
1369 | wake_up(&ctx->inflight_wait); | |
1370 | spin_unlock_irqrestore(&ctx->inflight_lock, flags); | |
1371 | } | |
2b85edfc PB |
1372 | |
1373 | percpu_ref_put(&req->ctx->refs); | |
b1e50e54 PB |
1374 | if (likely(!io_is_fallback_req(req))) |
1375 | kmem_cache_free(req_cachep, req); | |
1376 | else | |
dd461af6 | 1377 | clear_bit_unlock(0, (unsigned long *) &req->ctx->fallback_req); |
e65ef56d JA |
1378 | } |
1379 | ||
c6ca97b3 JA |
1380 | struct req_batch { |
1381 | void *reqs[IO_IOPOLL_BATCH]; | |
1382 | int to_free; | |
1383 | int need_iter; | |
1384 | }; | |
1385 | ||
1386 | static void io_free_req_many(struct io_ring_ctx *ctx, struct req_batch *rb) | |
1387 | { | |
1388 | if (!rb->to_free) | |
1389 | return; | |
1390 | if (rb->need_iter) { | |
1391 | int i, inflight = 0; | |
1392 | unsigned long flags; | |
1393 | ||
1394 | for (i = 0; i < rb->to_free; i++) { | |
1395 | struct io_kiocb *req = rb->reqs[i]; | |
1396 | ||
10fef4be | 1397 | if (req->flags & REQ_F_FIXED_FILE) { |
c6ca97b3 | 1398 | req->file = NULL; |
05589553 | 1399 | percpu_ref_put(req->fixed_file_refs); |
10fef4be | 1400 | } |
c6ca97b3 JA |
1401 | if (req->flags & REQ_F_INFLIGHT) |
1402 | inflight++; | |
c6ca97b3 JA |
1403 | __io_req_aux_free(req); |
1404 | } | |
1405 | if (!inflight) | |
1406 | goto do_free; | |
1407 | ||
1408 | spin_lock_irqsave(&ctx->inflight_lock, flags); | |
1409 | for (i = 0; i < rb->to_free; i++) { | |
1410 | struct io_kiocb *req = rb->reqs[i]; | |
1411 | ||
10fef4be | 1412 | if (req->flags & REQ_F_INFLIGHT) { |
c6ca97b3 JA |
1413 | list_del(&req->inflight_entry); |
1414 | if (!--inflight) | |
1415 | break; | |
1416 | } | |
1417 | } | |
1418 | spin_unlock_irqrestore(&ctx->inflight_lock, flags); | |
1419 | ||
1420 | if (waitqueue_active(&ctx->inflight_wait)) | |
1421 | wake_up(&ctx->inflight_wait); | |
1422 | } | |
1423 | do_free: | |
1424 | kmem_cache_free_bulk(req_cachep, rb->to_free, rb->reqs); | |
1425 | percpu_ref_put_many(&ctx->refs, rb->to_free); | |
c6ca97b3 | 1426 | rb->to_free = rb->need_iter = 0; |
e65ef56d JA |
1427 | } |
1428 | ||
a197f664 | 1429 | static bool io_link_cancel_timeout(struct io_kiocb *req) |
2665abfd | 1430 | { |
a197f664 | 1431 | struct io_ring_ctx *ctx = req->ctx; |
2665abfd JA |
1432 | int ret; |
1433 | ||
2d28390a | 1434 | ret = hrtimer_try_to_cancel(&req->io->timeout.timer); |
2665abfd | 1435 | if (ret != -1) { |
78e19bbe | 1436 | io_cqring_fill_event(req, -ECANCELED); |
2665abfd | 1437 | io_commit_cqring(ctx); |
dea3b49c | 1438 | req->flags &= ~REQ_F_LINK_HEAD; |
ec9c02ad | 1439 | io_put_req(req); |
2665abfd JA |
1440 | return true; |
1441 | } | |
1442 | ||
1443 | return false; | |
e65ef56d JA |
1444 | } |
1445 | ||
ba816ad6 | 1446 | static void io_req_link_next(struct io_kiocb *req, struct io_kiocb **nxtptr) |
9e645e11 | 1447 | { |
2665abfd | 1448 | struct io_ring_ctx *ctx = req->ctx; |
2665abfd | 1449 | bool wake_ev = false; |
9e645e11 | 1450 | |
4d7dd462 JA |
1451 | /* Already got next link */ |
1452 | if (req->flags & REQ_F_LINK_NEXT) | |
1453 | return; | |
1454 | ||
9e645e11 JA |
1455 | /* |
1456 | * The list should never be empty when we are called here. But could | |
1457 | * potentially happen if the chain is messed up, check to be on the | |
1458 | * safe side. | |
1459 | */ | |
4493233e PB |
1460 | while (!list_empty(&req->link_list)) { |
1461 | struct io_kiocb *nxt = list_first_entry(&req->link_list, | |
1462 | struct io_kiocb, link_list); | |
94ae5e77 | 1463 | |
4493233e PB |
1464 | if (unlikely((req->flags & REQ_F_LINK_TIMEOUT) && |
1465 | (nxt->flags & REQ_F_TIMEOUT))) { | |
1466 | list_del_init(&nxt->link_list); | |
94ae5e77 | 1467 | wake_ev |= io_link_cancel_timeout(nxt); |
94ae5e77 JA |
1468 | req->flags &= ~REQ_F_LINK_TIMEOUT; |
1469 | continue; | |
1470 | } | |
9e645e11 | 1471 | |
4493233e PB |
1472 | list_del_init(&req->link_list); |
1473 | if (!list_empty(&nxt->link_list)) | |
dea3b49c | 1474 | nxt->flags |= REQ_F_LINK_HEAD; |
b18fdf71 | 1475 | *nxtptr = nxt; |
94ae5e77 | 1476 | break; |
9e645e11 | 1477 | } |
2665abfd | 1478 | |
4d7dd462 | 1479 | req->flags |= REQ_F_LINK_NEXT; |
2665abfd JA |
1480 | if (wake_ev) |
1481 | io_cqring_ev_posted(ctx); | |
9e645e11 JA |
1482 | } |
1483 | ||
1484 | /* | |
dea3b49c | 1485 | * Called if REQ_F_LINK_HEAD is set, and we fail the head request |
9e645e11 JA |
1486 | */ |
1487 | static void io_fail_links(struct io_kiocb *req) | |
1488 | { | |
2665abfd | 1489 | struct io_ring_ctx *ctx = req->ctx; |
2665abfd JA |
1490 | unsigned long flags; |
1491 | ||
1492 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
9e645e11 JA |
1493 | |
1494 | while (!list_empty(&req->link_list)) { | |
4493233e PB |
1495 | struct io_kiocb *link = list_first_entry(&req->link_list, |
1496 | struct io_kiocb, link_list); | |
9e645e11 | 1497 | |
4493233e | 1498 | list_del_init(&link->link_list); |
c826bd7a | 1499 | trace_io_uring_fail_link(req, link); |
2665abfd JA |
1500 | |
1501 | if ((req->flags & REQ_F_LINK_TIMEOUT) && | |
d625c6ee | 1502 | link->opcode == IORING_OP_LINK_TIMEOUT) { |
a197f664 | 1503 | io_link_cancel_timeout(link); |
2665abfd | 1504 | } else { |
78e19bbe | 1505 | io_cqring_fill_event(link, -ECANCELED); |
978db57e | 1506 | __io_double_put_req(link); |
2665abfd | 1507 | } |
5d960724 | 1508 | req->flags &= ~REQ_F_LINK_TIMEOUT; |
9e645e11 | 1509 | } |
2665abfd JA |
1510 | |
1511 | io_commit_cqring(ctx); | |
1512 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
1513 | io_cqring_ev_posted(ctx); | |
9e645e11 JA |
1514 | } |
1515 | ||
4d7dd462 | 1516 | static void io_req_find_next(struct io_kiocb *req, struct io_kiocb **nxt) |
9e645e11 | 1517 | { |
dea3b49c | 1518 | if (likely(!(req->flags & REQ_F_LINK_HEAD))) |
2665abfd | 1519 | return; |
2665abfd | 1520 | |
9e645e11 JA |
1521 | /* |
1522 | * If LINK is set, we have dependent requests in this chain. If we | |
1523 | * didn't fail this request, queue the first one up, moving any other | |
1524 | * dependencies to the next request. In case of failure, fail the rest | |
1525 | * of the chain. | |
1526 | */ | |
2665abfd JA |
1527 | if (req->flags & REQ_F_FAIL_LINK) { |
1528 | io_fail_links(req); | |
7c9e7f0f JA |
1529 | } else if ((req->flags & (REQ_F_LINK_TIMEOUT | REQ_F_COMP_LOCKED)) == |
1530 | REQ_F_LINK_TIMEOUT) { | |
2665abfd JA |
1531 | struct io_ring_ctx *ctx = req->ctx; |
1532 | unsigned long flags; | |
1533 | ||
1534 | /* | |
1535 | * If this is a timeout link, we could be racing with the | |
1536 | * timeout timer. Grab the completion lock for this case to | |
7c9e7f0f | 1537 | * protect against that. |
2665abfd JA |
1538 | */ |
1539 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
1540 | io_req_link_next(req, nxt); | |
1541 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
1542 | } else { | |
1543 | io_req_link_next(req, nxt); | |
9e645e11 | 1544 | } |
4d7dd462 | 1545 | } |
9e645e11 | 1546 | |
c69f8dbe JL |
1547 | static void io_free_req(struct io_kiocb *req) |
1548 | { | |
944e58bf PB |
1549 | struct io_kiocb *nxt = NULL; |
1550 | ||
1551 | io_req_find_next(req, &nxt); | |
70cf9f32 | 1552 | __io_free_req(req); |
944e58bf PB |
1553 | |
1554 | if (nxt) | |
1555 | io_queue_async_work(nxt); | |
c69f8dbe JL |
1556 | } |
1557 | ||
7a743e22 PB |
1558 | static void io_link_work_cb(struct io_wq_work **workptr) |
1559 | { | |
18a542ff PB |
1560 | struct io_kiocb *req = container_of(*workptr, struct io_kiocb, work); |
1561 | struct io_kiocb *link; | |
7a743e22 | 1562 | |
18a542ff | 1563 | link = list_first_entry(&req->link_list, struct io_kiocb, link_list); |
7a743e22 PB |
1564 | io_queue_linked_timeout(link); |
1565 | io_wq_submit_work(workptr); | |
1566 | } | |
1567 | ||
1568 | static void io_wq_assign_next(struct io_wq_work **workptr, struct io_kiocb *nxt) | |
1569 | { | |
1570 | struct io_kiocb *link; | |
8766dd51 PB |
1571 | const struct io_op_def *def = &io_op_defs[nxt->opcode]; |
1572 | ||
1573 | if ((nxt->flags & REQ_F_ISREG) && def->hash_reg_file) | |
1574 | io_wq_hash_work(&nxt->work, file_inode(nxt->file)); | |
7a743e22 PB |
1575 | |
1576 | *workptr = &nxt->work; | |
1577 | link = io_prep_linked_timeout(nxt); | |
18a542ff | 1578 | if (link) |
7a743e22 | 1579 | nxt->work.func = io_link_work_cb; |
7a743e22 PB |
1580 | } |
1581 | ||
ba816ad6 JA |
1582 | /* |
1583 | * Drop reference to request, return next in chain (if there is one) if this | |
1584 | * was the last reference to this request. | |
1585 | */ | |
f9bd67f6 | 1586 | __attribute__((nonnull)) |
ec9c02ad | 1587 | static void io_put_req_find_next(struct io_kiocb *req, struct io_kiocb **nxtptr) |
e65ef56d | 1588 | { |
2a44f467 JA |
1589 | if (refcount_dec_and_test(&req->refs)) { |
1590 | io_req_find_next(req, nxtptr); | |
4d7dd462 | 1591 | __io_free_req(req); |
2a44f467 | 1592 | } |
2b188cc1 JA |
1593 | } |
1594 | ||
e65ef56d JA |
1595 | static void io_put_req(struct io_kiocb *req) |
1596 | { | |
1597 | if (refcount_dec_and_test(&req->refs)) | |
1598 | io_free_req(req); | |
2b188cc1 JA |
1599 | } |
1600 | ||
e9fd9396 PB |
1601 | static void io_steal_work(struct io_kiocb *req, |
1602 | struct io_wq_work **workptr) | |
7a743e22 PB |
1603 | { |
1604 | /* | |
1605 | * It's in an io-wq worker, so there always should be at least | |
1606 | * one reference, which will be dropped in io_put_work() just | |
1607 | * after the current handler returns. | |
1608 | * | |
1609 | * It also means, that if the counter dropped to 1, then there is | |
1610 | * no asynchronous users left, so it's safe to steal the next work. | |
1611 | */ | |
7a743e22 PB |
1612 | if (refcount_read(&req->refs) == 1) { |
1613 | struct io_kiocb *nxt = NULL; | |
1614 | ||
1615 | io_req_find_next(req, &nxt); | |
1616 | if (nxt) | |
1617 | io_wq_assign_next(workptr, nxt); | |
1618 | } | |
1619 | } | |
1620 | ||
978db57e JA |
1621 | /* |
1622 | * Must only be used if we don't need to care about links, usually from | |
1623 | * within the completion handling itself. | |
1624 | */ | |
1625 | static void __io_double_put_req(struct io_kiocb *req) | |
78e19bbe JA |
1626 | { |
1627 | /* drop both submit and complete references */ | |
1628 | if (refcount_sub_and_test(2, &req->refs)) | |
1629 | __io_free_req(req); | |
1630 | } | |
1631 | ||
978db57e JA |
1632 | static void io_double_put_req(struct io_kiocb *req) |
1633 | { | |
1634 | /* drop both submit and complete references */ | |
1635 | if (refcount_sub_and_test(2, &req->refs)) | |
1636 | io_free_req(req); | |
1637 | } | |
1638 | ||
1d7bb1d5 | 1639 | static unsigned io_cqring_events(struct io_ring_ctx *ctx, bool noflush) |
a3a0e43f | 1640 | { |
84f97dc2 JA |
1641 | struct io_rings *rings = ctx->rings; |
1642 | ||
ad3eb2c8 JA |
1643 | if (test_bit(0, &ctx->cq_check_overflow)) { |
1644 | /* | |
1645 | * noflush == true is from the waitqueue handler, just ensure | |
1646 | * we wake up the task, and the next invocation will flush the | |
1647 | * entries. We cannot safely to it from here. | |
1648 | */ | |
1649 | if (noflush && !list_empty(&ctx->cq_overflow_list)) | |
1650 | return -1U; | |
1d7bb1d5 | 1651 | |
ad3eb2c8 JA |
1652 | io_cqring_overflow_flush(ctx, false); |
1653 | } | |
1d7bb1d5 | 1654 | |
a3a0e43f JA |
1655 | /* See comment at the top of this file */ |
1656 | smp_rmb(); | |
ad3eb2c8 | 1657 | return ctx->cached_cq_tail - READ_ONCE(rings->cq.head); |
a3a0e43f JA |
1658 | } |
1659 | ||
fb5ccc98 PB |
1660 | static inline unsigned int io_sqring_entries(struct io_ring_ctx *ctx) |
1661 | { | |
1662 | struct io_rings *rings = ctx->rings; | |
1663 | ||
1664 | /* make sure SQ entry isn't read before tail */ | |
1665 | return smp_load_acquire(&rings->sq.tail) - ctx->cached_sq_head; | |
1666 | } | |
1667 | ||
8237e045 | 1668 | static inline bool io_req_multi_free(struct req_batch *rb, struct io_kiocb *req) |
e94f141b | 1669 | { |
dea3b49c | 1670 | if ((req->flags & REQ_F_LINK_HEAD) || io_is_fallback_req(req)) |
c6ca97b3 | 1671 | return false; |
e94f141b | 1672 | |
c6ca97b3 JA |
1673 | if (!(req->flags & REQ_F_FIXED_FILE) || req->io) |
1674 | rb->need_iter++; | |
1675 | ||
1676 | rb->reqs[rb->to_free++] = req; | |
1677 | if (unlikely(rb->to_free == ARRAY_SIZE(rb->reqs))) | |
1678 | io_free_req_many(req->ctx, rb); | |
1679 | return true; | |
e94f141b JA |
1680 | } |
1681 | ||
bcda7baa JA |
1682 | static int io_put_kbuf(struct io_kiocb *req) |
1683 | { | |
4d954c25 | 1684 | struct io_buffer *kbuf; |
bcda7baa JA |
1685 | int cflags; |
1686 | ||
4d954c25 | 1687 | kbuf = (struct io_buffer *) (unsigned long) req->rw.addr; |
bcda7baa JA |
1688 | cflags = kbuf->bid << IORING_CQE_BUFFER_SHIFT; |
1689 | cflags |= IORING_CQE_F_BUFFER; | |
1690 | req->rw.addr = 0; | |
1691 | kfree(kbuf); | |
1692 | return cflags; | |
1693 | } | |
1694 | ||
def596e9 JA |
1695 | /* |
1696 | * Find and free completed poll iocbs | |
1697 | */ | |
1698 | static void io_iopoll_complete(struct io_ring_ctx *ctx, unsigned int *nr_events, | |
1699 | struct list_head *done) | |
1700 | { | |
8237e045 | 1701 | struct req_batch rb; |
def596e9 | 1702 | struct io_kiocb *req; |
def596e9 | 1703 | |
c6ca97b3 | 1704 | rb.to_free = rb.need_iter = 0; |
def596e9 | 1705 | while (!list_empty(done)) { |
bcda7baa JA |
1706 | int cflags = 0; |
1707 | ||
def596e9 JA |
1708 | req = list_first_entry(done, struct io_kiocb, list); |
1709 | list_del(&req->list); | |
1710 | ||
bcda7baa JA |
1711 | if (req->flags & REQ_F_BUFFER_SELECTED) |
1712 | cflags = io_put_kbuf(req); | |
1713 | ||
1714 | __io_cqring_fill_event(req, req->result, cflags); | |
def596e9 JA |
1715 | (*nr_events)++; |
1716 | ||
8237e045 JA |
1717 | if (refcount_dec_and_test(&req->refs) && |
1718 | !io_req_multi_free(&rb, req)) | |
1719 | io_free_req(req); | |
def596e9 | 1720 | } |
def596e9 | 1721 | |
09bb8394 | 1722 | io_commit_cqring(ctx); |
32b2244a XW |
1723 | if (ctx->flags & IORING_SETUP_SQPOLL) |
1724 | io_cqring_ev_posted(ctx); | |
8237e045 | 1725 | io_free_req_many(ctx, &rb); |
def596e9 JA |
1726 | } |
1727 | ||
581f9810 BM |
1728 | static void io_iopoll_queue(struct list_head *again) |
1729 | { | |
1730 | struct io_kiocb *req; | |
1731 | ||
1732 | do { | |
1733 | req = list_first_entry(again, struct io_kiocb, list); | |
1734 | list_del(&req->list); | |
1735 | refcount_inc(&req->refs); | |
1736 | io_queue_async_work(req); | |
1737 | } while (!list_empty(again)); | |
1738 | } | |
1739 | ||
def596e9 JA |
1740 | static int io_do_iopoll(struct io_ring_ctx *ctx, unsigned int *nr_events, |
1741 | long min) | |
1742 | { | |
1743 | struct io_kiocb *req, *tmp; | |
1744 | LIST_HEAD(done); | |
581f9810 | 1745 | LIST_HEAD(again); |
def596e9 JA |
1746 | bool spin; |
1747 | int ret; | |
1748 | ||
1749 | /* | |
1750 | * Only spin for completions if we don't have multiple devices hanging | |
1751 | * off our complete list, and we're under the requested amount. | |
1752 | */ | |
1753 | spin = !ctx->poll_multi_file && *nr_events < min; | |
1754 | ||
1755 | ret = 0; | |
1756 | list_for_each_entry_safe(req, tmp, &ctx->poll_list, list) { | |
9adbd45d | 1757 | struct kiocb *kiocb = &req->rw.kiocb; |
def596e9 JA |
1758 | |
1759 | /* | |
581f9810 BM |
1760 | * Move completed and retryable entries to our local lists. |
1761 | * If we find a request that requires polling, break out | |
1762 | * and complete those lists first, if we have entries there. | |
def596e9 JA |
1763 | */ |
1764 | if (req->flags & REQ_F_IOPOLL_COMPLETED) { | |
1765 | list_move_tail(&req->list, &done); | |
1766 | continue; | |
1767 | } | |
1768 | if (!list_empty(&done)) | |
1769 | break; | |
1770 | ||
581f9810 BM |
1771 | if (req->result == -EAGAIN) { |
1772 | list_move_tail(&req->list, &again); | |
1773 | continue; | |
1774 | } | |
1775 | if (!list_empty(&again)) | |
1776 | break; | |
1777 | ||
def596e9 JA |
1778 | ret = kiocb->ki_filp->f_op->iopoll(kiocb, spin); |
1779 | if (ret < 0) | |
1780 | break; | |
1781 | ||
1782 | if (ret && spin) | |
1783 | spin = false; | |
1784 | ret = 0; | |
1785 | } | |
1786 | ||
1787 | if (!list_empty(&done)) | |
1788 | io_iopoll_complete(ctx, nr_events, &done); | |
1789 | ||
581f9810 BM |
1790 | if (!list_empty(&again)) |
1791 | io_iopoll_queue(&again); | |
1792 | ||
def596e9 JA |
1793 | return ret; |
1794 | } | |
1795 | ||
1796 | /* | |
d195a66e | 1797 | * Poll for a minimum of 'min' events. Note that if min == 0 we consider that a |
def596e9 JA |
1798 | * non-spinning poll check - we'll still enter the driver poll loop, but only |
1799 | * as a non-spinning completion check. | |
1800 | */ | |
1801 | static int io_iopoll_getevents(struct io_ring_ctx *ctx, unsigned int *nr_events, | |
1802 | long min) | |
1803 | { | |
08f5439f | 1804 | while (!list_empty(&ctx->poll_list) && !need_resched()) { |
def596e9 JA |
1805 | int ret; |
1806 | ||
1807 | ret = io_do_iopoll(ctx, nr_events, min); | |
1808 | if (ret < 0) | |
1809 | return ret; | |
1810 | if (!min || *nr_events >= min) | |
1811 | return 0; | |
1812 | } | |
1813 | ||
1814 | return 1; | |
1815 | } | |
1816 | ||
1817 | /* | |
1818 | * We can't just wait for polled events to come to us, we have to actively | |
1819 | * find and complete them. | |
1820 | */ | |
1821 | static void io_iopoll_reap_events(struct io_ring_ctx *ctx) | |
1822 | { | |
1823 | if (!(ctx->flags & IORING_SETUP_IOPOLL)) | |
1824 | return; | |
1825 | ||
1826 | mutex_lock(&ctx->uring_lock); | |
1827 | while (!list_empty(&ctx->poll_list)) { | |
1828 | unsigned int nr_events = 0; | |
1829 | ||
1830 | io_iopoll_getevents(ctx, &nr_events, 1); | |
08f5439f JA |
1831 | |
1832 | /* | |
1833 | * Ensure we allow local-to-the-cpu processing to take place, | |
1834 | * in this case we need to ensure that we reap all events. | |
1835 | */ | |
1836 | cond_resched(); | |
def596e9 JA |
1837 | } |
1838 | mutex_unlock(&ctx->uring_lock); | |
1839 | } | |
1840 | ||
c7849be9 XW |
1841 | static int io_iopoll_check(struct io_ring_ctx *ctx, unsigned *nr_events, |
1842 | long min) | |
def596e9 | 1843 | { |
2b2ed975 | 1844 | int iters = 0, ret = 0; |
500f9fba | 1845 | |
c7849be9 XW |
1846 | /* |
1847 | * We disallow the app entering submit/complete with polling, but we | |
1848 | * still need to lock the ring to prevent racing with polled issue | |
1849 | * that got punted to a workqueue. | |
1850 | */ | |
1851 | mutex_lock(&ctx->uring_lock); | |
def596e9 JA |
1852 | do { |
1853 | int tmin = 0; | |
1854 | ||
a3a0e43f JA |
1855 | /* |
1856 | * Don't enter poll loop if we already have events pending. | |
1857 | * If we do, we can potentially be spinning for commands that | |
1858 | * already triggered a CQE (eg in error). | |
1859 | */ | |
1d7bb1d5 | 1860 | if (io_cqring_events(ctx, false)) |
a3a0e43f JA |
1861 | break; |
1862 | ||
500f9fba JA |
1863 | /* |
1864 | * If a submit got punted to a workqueue, we can have the | |
1865 | * application entering polling for a command before it gets | |
1866 | * issued. That app will hold the uring_lock for the duration | |
1867 | * of the poll right here, so we need to take a breather every | |
1868 | * now and then to ensure that the issue has a chance to add | |
1869 | * the poll to the issued list. Otherwise we can spin here | |
1870 | * forever, while the workqueue is stuck trying to acquire the | |
1871 | * very same mutex. | |
1872 | */ | |
1873 | if (!(++iters & 7)) { | |
1874 | mutex_unlock(&ctx->uring_lock); | |
1875 | mutex_lock(&ctx->uring_lock); | |
1876 | } | |
1877 | ||
def596e9 JA |
1878 | if (*nr_events < min) |
1879 | tmin = min - *nr_events; | |
1880 | ||
1881 | ret = io_iopoll_getevents(ctx, nr_events, tmin); | |
1882 | if (ret <= 0) | |
1883 | break; | |
1884 | ret = 0; | |
1885 | } while (min && !*nr_events && !need_resched()); | |
1886 | ||
500f9fba | 1887 | mutex_unlock(&ctx->uring_lock); |
def596e9 JA |
1888 | return ret; |
1889 | } | |
1890 | ||
491381ce | 1891 | static void kiocb_end_write(struct io_kiocb *req) |
2b188cc1 | 1892 | { |
491381ce JA |
1893 | /* |
1894 | * Tell lockdep we inherited freeze protection from submission | |
1895 | * thread. | |
1896 | */ | |
1897 | if (req->flags & REQ_F_ISREG) { | |
1898 | struct inode *inode = file_inode(req->file); | |
2b188cc1 | 1899 | |
491381ce | 1900 | __sb_writers_acquired(inode->i_sb, SB_FREEZE_WRITE); |
2b188cc1 | 1901 | } |
491381ce | 1902 | file_end_write(req->file); |
2b188cc1 JA |
1903 | } |
1904 | ||
4e88d6e7 JA |
1905 | static inline void req_set_fail_links(struct io_kiocb *req) |
1906 | { | |
1907 | if ((req->flags & (REQ_F_LINK | REQ_F_HARDLINK)) == REQ_F_LINK) | |
1908 | req->flags |= REQ_F_FAIL_LINK; | |
1909 | } | |
1910 | ||
ba816ad6 | 1911 | static void io_complete_rw_common(struct kiocb *kiocb, long res) |
2b188cc1 | 1912 | { |
9adbd45d | 1913 | struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb); |
bcda7baa | 1914 | int cflags = 0; |
2b188cc1 | 1915 | |
491381ce JA |
1916 | if (kiocb->ki_flags & IOCB_WRITE) |
1917 | kiocb_end_write(req); | |
2b188cc1 | 1918 | |
4e88d6e7 JA |
1919 | if (res != req->result) |
1920 | req_set_fail_links(req); | |
bcda7baa JA |
1921 | if (req->flags & REQ_F_BUFFER_SELECTED) |
1922 | cflags = io_put_kbuf(req); | |
1923 | __io_cqring_add_event(req, res, cflags); | |
ba816ad6 JA |
1924 | } |
1925 | ||
1926 | static void io_complete_rw(struct kiocb *kiocb, long res, long res2) | |
1927 | { | |
9adbd45d | 1928 | struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb); |
ba816ad6 JA |
1929 | |
1930 | io_complete_rw_common(kiocb, res); | |
e65ef56d | 1931 | io_put_req(req); |
2b188cc1 JA |
1932 | } |
1933 | ||
def596e9 JA |
1934 | static void io_complete_rw_iopoll(struct kiocb *kiocb, long res, long res2) |
1935 | { | |
9adbd45d | 1936 | struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb); |
def596e9 | 1937 | |
491381ce JA |
1938 | if (kiocb->ki_flags & IOCB_WRITE) |
1939 | kiocb_end_write(req); | |
def596e9 | 1940 | |
4e88d6e7 JA |
1941 | if (res != req->result) |
1942 | req_set_fail_links(req); | |
9e645e11 | 1943 | req->result = res; |
def596e9 JA |
1944 | if (res != -EAGAIN) |
1945 | req->flags |= REQ_F_IOPOLL_COMPLETED; | |
1946 | } | |
1947 | ||
1948 | /* | |
1949 | * After the iocb has been issued, it's safe to be found on the poll list. | |
1950 | * Adding the kiocb to the list AFTER submission ensures that we don't | |
1951 | * find it from a io_iopoll_getevents() thread before the issuer is done | |
1952 | * accessing the kiocb cookie. | |
1953 | */ | |
1954 | static void io_iopoll_req_issued(struct io_kiocb *req) | |
1955 | { | |
1956 | struct io_ring_ctx *ctx = req->ctx; | |
1957 | ||
1958 | /* | |
1959 | * Track whether we have multiple files in our lists. This will impact | |
1960 | * how we do polling eventually, not spinning if we're on potentially | |
1961 | * different devices. | |
1962 | */ | |
1963 | if (list_empty(&ctx->poll_list)) { | |
1964 | ctx->poll_multi_file = false; | |
1965 | } else if (!ctx->poll_multi_file) { | |
1966 | struct io_kiocb *list_req; | |
1967 | ||
1968 | list_req = list_first_entry(&ctx->poll_list, struct io_kiocb, | |
1969 | list); | |
9adbd45d | 1970 | if (list_req->file != req->file) |
def596e9 JA |
1971 | ctx->poll_multi_file = true; |
1972 | } | |
1973 | ||
1974 | /* | |
1975 | * For fast devices, IO may have already completed. If it has, add | |
1976 | * it to the front so we find it first. | |
1977 | */ | |
1978 | if (req->flags & REQ_F_IOPOLL_COMPLETED) | |
1979 | list_add(&req->list, &ctx->poll_list); | |
1980 | else | |
1981 | list_add_tail(&req->list, &ctx->poll_list); | |
bdcd3eab XW |
1982 | |
1983 | if ((ctx->flags & IORING_SETUP_SQPOLL) && | |
1984 | wq_has_sleeper(&ctx->sqo_wait)) | |
1985 | wake_up(&ctx->sqo_wait); | |
def596e9 JA |
1986 | } |
1987 | ||
3d6770fb | 1988 | static void io_file_put(struct io_submit_state *state) |
9a56a232 | 1989 | { |
3d6770fb | 1990 | if (state->file) { |
9a56a232 JA |
1991 | int diff = state->has_refs - state->used_refs; |
1992 | ||
1993 | if (diff) | |
1994 | fput_many(state->file, diff); | |
1995 | state->file = NULL; | |
1996 | } | |
1997 | } | |
1998 | ||
1999 | /* | |
2000 | * Get as many references to a file as we have IOs left in this submission, | |
2001 | * assuming most submissions are for one file, or at least that each file | |
2002 | * has more than one submission. | |
2003 | */ | |
8da11c19 | 2004 | static struct file *__io_file_get(struct io_submit_state *state, int fd) |
9a56a232 JA |
2005 | { |
2006 | if (!state) | |
2007 | return fget(fd); | |
2008 | ||
2009 | if (state->file) { | |
2010 | if (state->fd == fd) { | |
2011 | state->used_refs++; | |
2012 | state->ios_left--; | |
2013 | return state->file; | |
2014 | } | |
3d6770fb | 2015 | io_file_put(state); |
9a56a232 JA |
2016 | } |
2017 | state->file = fget_many(fd, state->ios_left); | |
2018 | if (!state->file) | |
2019 | return NULL; | |
2020 | ||
2021 | state->fd = fd; | |
2022 | state->has_refs = state->ios_left; | |
2023 | state->used_refs = 1; | |
2024 | state->ios_left--; | |
2025 | return state->file; | |
2026 | } | |
2027 | ||
2b188cc1 JA |
2028 | /* |
2029 | * If we tracked the file through the SCM inflight mechanism, we could support | |
2030 | * any file. For now, just ensure that anything potentially problematic is done | |
2031 | * inline. | |
2032 | */ | |
af197f50 | 2033 | static bool io_file_supports_async(struct file *file, int rw) |
2b188cc1 JA |
2034 | { |
2035 | umode_t mode = file_inode(file)->i_mode; | |
2036 | ||
10d59345 | 2037 | if (S_ISBLK(mode) || S_ISCHR(mode) || S_ISSOCK(mode)) |
2b188cc1 JA |
2038 | return true; |
2039 | if (S_ISREG(mode) && file->f_op != &io_uring_fops) | |
2040 | return true; | |
2041 | ||
af197f50 JA |
2042 | if (!(file->f_mode & FMODE_NOWAIT)) |
2043 | return false; | |
2044 | ||
2045 | if (rw == READ) | |
2046 | return file->f_op->read_iter != NULL; | |
2047 | ||
2048 | return file->f_op->write_iter != NULL; | |
2b188cc1 JA |
2049 | } |
2050 | ||
3529d8c2 JA |
2051 | static int io_prep_rw(struct io_kiocb *req, const struct io_uring_sqe *sqe, |
2052 | bool force_nonblock) | |
2b188cc1 | 2053 | { |
def596e9 | 2054 | struct io_ring_ctx *ctx = req->ctx; |
9adbd45d | 2055 | struct kiocb *kiocb = &req->rw.kiocb; |
09bb8394 JA |
2056 | unsigned ioprio; |
2057 | int ret; | |
2b188cc1 | 2058 | |
491381ce JA |
2059 | if (S_ISREG(file_inode(req->file)->i_mode)) |
2060 | req->flags |= REQ_F_ISREG; | |
2061 | ||
2b188cc1 | 2062 | kiocb->ki_pos = READ_ONCE(sqe->off); |
ba04291e JA |
2063 | if (kiocb->ki_pos == -1 && !(req->file->f_mode & FMODE_STREAM)) { |
2064 | req->flags |= REQ_F_CUR_POS; | |
2065 | kiocb->ki_pos = req->file->f_pos; | |
2066 | } | |
2b188cc1 | 2067 | kiocb->ki_hint = ki_hint_validate(file_write_hint(kiocb->ki_filp)); |
3e577dcd PB |
2068 | kiocb->ki_flags = iocb_flags(kiocb->ki_filp); |
2069 | ret = kiocb_set_rw_flags(kiocb, READ_ONCE(sqe->rw_flags)); | |
2070 | if (unlikely(ret)) | |
2071 | return ret; | |
2b188cc1 JA |
2072 | |
2073 | ioprio = READ_ONCE(sqe->ioprio); | |
2074 | if (ioprio) { | |
2075 | ret = ioprio_check_cap(ioprio); | |
2076 | if (ret) | |
09bb8394 | 2077 | return ret; |
2b188cc1 JA |
2078 | |
2079 | kiocb->ki_ioprio = ioprio; | |
2080 | } else | |
2081 | kiocb->ki_ioprio = get_current_ioprio(); | |
2082 | ||
8449eeda | 2083 | /* don't allow async punt if RWF_NOWAIT was requested */ |
491381ce JA |
2084 | if ((kiocb->ki_flags & IOCB_NOWAIT) || |
2085 | (req->file->f_flags & O_NONBLOCK)) | |
8449eeda SB |
2086 | req->flags |= REQ_F_NOWAIT; |
2087 | ||
2088 | if (force_nonblock) | |
2b188cc1 | 2089 | kiocb->ki_flags |= IOCB_NOWAIT; |
8449eeda | 2090 | |
def596e9 | 2091 | if (ctx->flags & IORING_SETUP_IOPOLL) { |
def596e9 JA |
2092 | if (!(kiocb->ki_flags & IOCB_DIRECT) || |
2093 | !kiocb->ki_filp->f_op->iopoll) | |
09bb8394 | 2094 | return -EOPNOTSUPP; |
2b188cc1 | 2095 | |
def596e9 JA |
2096 | kiocb->ki_flags |= IOCB_HIPRI; |
2097 | kiocb->ki_complete = io_complete_rw_iopoll; | |
6873e0bd | 2098 | req->result = 0; |
def596e9 | 2099 | } else { |
09bb8394 JA |
2100 | if (kiocb->ki_flags & IOCB_HIPRI) |
2101 | return -EINVAL; | |
def596e9 JA |
2102 | kiocb->ki_complete = io_complete_rw; |
2103 | } | |
9adbd45d | 2104 | |
3529d8c2 JA |
2105 | req->rw.addr = READ_ONCE(sqe->addr); |
2106 | req->rw.len = READ_ONCE(sqe->len); | |
bcda7baa | 2107 | /* we own ->private, reuse it for the buffer index / buffer ID */ |
9adbd45d | 2108 | req->rw.kiocb.private = (void *) (unsigned long) |
3529d8c2 | 2109 | READ_ONCE(sqe->buf_index); |
2b188cc1 | 2110 | return 0; |
2b188cc1 JA |
2111 | } |
2112 | ||
2113 | static inline void io_rw_done(struct kiocb *kiocb, ssize_t ret) | |
2114 | { | |
2115 | switch (ret) { | |
2116 | case -EIOCBQUEUED: | |
2117 | break; | |
2118 | case -ERESTARTSYS: | |
2119 | case -ERESTARTNOINTR: | |
2120 | case -ERESTARTNOHAND: | |
2121 | case -ERESTART_RESTARTBLOCK: | |
2122 | /* | |
2123 | * We can't just restart the syscall, since previously | |
2124 | * submitted sqes may already be in progress. Just fail this | |
2125 | * IO with EINTR. | |
2126 | */ | |
2127 | ret = -EINTR; | |
2128 | /* fall through */ | |
2129 | default: | |
2130 | kiocb->ki_complete(kiocb, ret, 0); | |
2131 | } | |
2132 | } | |
2133 | ||
014db007 | 2134 | static void kiocb_done(struct kiocb *kiocb, ssize_t ret) |
ba816ad6 | 2135 | { |
ba04291e JA |
2136 | struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb); |
2137 | ||
2138 | if (req->flags & REQ_F_CUR_POS) | |
2139 | req->file->f_pos = kiocb->ki_pos; | |
bcaec089 | 2140 | if (ret >= 0 && kiocb->ki_complete == io_complete_rw) |
014db007 | 2141 | io_complete_rw(kiocb, ret, 0); |
ba816ad6 JA |
2142 | else |
2143 | io_rw_done(kiocb, ret); | |
2144 | } | |
2145 | ||
9adbd45d | 2146 | static ssize_t io_import_fixed(struct io_kiocb *req, int rw, |
7d009165 | 2147 | struct iov_iter *iter) |
edafccee | 2148 | { |
9adbd45d JA |
2149 | struct io_ring_ctx *ctx = req->ctx; |
2150 | size_t len = req->rw.len; | |
edafccee JA |
2151 | struct io_mapped_ubuf *imu; |
2152 | unsigned index, buf_index; | |
2153 | size_t offset; | |
2154 | u64 buf_addr; | |
2155 | ||
2156 | /* attempt to use fixed buffers without having provided iovecs */ | |
2157 | if (unlikely(!ctx->user_bufs)) | |
2158 | return -EFAULT; | |
2159 | ||
9adbd45d | 2160 | buf_index = (unsigned long) req->rw.kiocb.private; |
edafccee JA |
2161 | if (unlikely(buf_index >= ctx->nr_user_bufs)) |
2162 | return -EFAULT; | |
2163 | ||
2164 | index = array_index_nospec(buf_index, ctx->nr_user_bufs); | |
2165 | imu = &ctx->user_bufs[index]; | |
9adbd45d | 2166 | buf_addr = req->rw.addr; |
edafccee JA |
2167 | |
2168 | /* overflow */ | |
2169 | if (buf_addr + len < buf_addr) | |
2170 | return -EFAULT; | |
2171 | /* not inside the mapped region */ | |
2172 | if (buf_addr < imu->ubuf || buf_addr + len > imu->ubuf + imu->len) | |
2173 | return -EFAULT; | |
2174 | ||
2175 | /* | |
2176 | * May not be a start of buffer, set size appropriately | |
2177 | * and advance us to the beginning. | |
2178 | */ | |
2179 | offset = buf_addr - imu->ubuf; | |
2180 | iov_iter_bvec(iter, rw, imu->bvec, imu->nr_bvecs, offset + len); | |
bd11b3a3 JA |
2181 | |
2182 | if (offset) { | |
2183 | /* | |
2184 | * Don't use iov_iter_advance() here, as it's really slow for | |
2185 | * using the latter parts of a big fixed buffer - it iterates | |
2186 | * over each segment manually. We can cheat a bit here, because | |
2187 | * we know that: | |
2188 | * | |
2189 | * 1) it's a BVEC iter, we set it up | |
2190 | * 2) all bvecs are PAGE_SIZE in size, except potentially the | |
2191 | * first and last bvec | |
2192 | * | |
2193 | * So just find our index, and adjust the iterator afterwards. | |
2194 | * If the offset is within the first bvec (or the whole first | |
2195 | * bvec, just use iov_iter_advance(). This makes it easier | |
2196 | * since we can just skip the first segment, which may not | |
2197 | * be PAGE_SIZE aligned. | |
2198 | */ | |
2199 | const struct bio_vec *bvec = imu->bvec; | |
2200 | ||
2201 | if (offset <= bvec->bv_len) { | |
2202 | iov_iter_advance(iter, offset); | |
2203 | } else { | |
2204 | unsigned long seg_skip; | |
2205 | ||
2206 | /* skip first vec */ | |
2207 | offset -= bvec->bv_len; | |
2208 | seg_skip = 1 + (offset >> PAGE_SHIFT); | |
2209 | ||
2210 | iter->bvec = bvec + seg_skip; | |
2211 | iter->nr_segs -= seg_skip; | |
99c79f66 | 2212 | iter->count -= bvec->bv_len + offset; |
bd11b3a3 | 2213 | iter->iov_offset = offset & ~PAGE_MASK; |
bd11b3a3 JA |
2214 | } |
2215 | } | |
2216 | ||
5e559561 | 2217 | return len; |
edafccee JA |
2218 | } |
2219 | ||
bcda7baa JA |
2220 | static void io_ring_submit_unlock(struct io_ring_ctx *ctx, bool needs_lock) |
2221 | { | |
2222 | if (needs_lock) | |
2223 | mutex_unlock(&ctx->uring_lock); | |
2224 | } | |
2225 | ||
2226 | static void io_ring_submit_lock(struct io_ring_ctx *ctx, bool needs_lock) | |
2227 | { | |
2228 | /* | |
2229 | * "Normal" inline submissions always hold the uring_lock, since we | |
2230 | * grab it from the system call. Same is true for the SQPOLL offload. | |
2231 | * The only exception is when we've detached the request and issue it | |
2232 | * from an async worker thread, grab the lock for that case. | |
2233 | */ | |
2234 | if (needs_lock) | |
2235 | mutex_lock(&ctx->uring_lock); | |
2236 | } | |
2237 | ||
2238 | static struct io_buffer *io_buffer_select(struct io_kiocb *req, size_t *len, | |
2239 | int bgid, struct io_buffer *kbuf, | |
2240 | bool needs_lock) | |
2241 | { | |
2242 | struct io_buffer *head; | |
2243 | ||
2244 | if (req->flags & REQ_F_BUFFER_SELECTED) | |
2245 | return kbuf; | |
2246 | ||
2247 | io_ring_submit_lock(req->ctx, needs_lock); | |
2248 | ||
2249 | lockdep_assert_held(&req->ctx->uring_lock); | |
2250 | ||
2251 | head = idr_find(&req->ctx->io_buffer_idr, bgid); | |
2252 | if (head) { | |
2253 | if (!list_empty(&head->list)) { | |
2254 | kbuf = list_last_entry(&head->list, struct io_buffer, | |
2255 | list); | |
2256 | list_del(&kbuf->list); | |
2257 | } else { | |
2258 | kbuf = head; | |
2259 | idr_remove(&req->ctx->io_buffer_idr, bgid); | |
2260 | } | |
2261 | if (*len > kbuf->len) | |
2262 | *len = kbuf->len; | |
2263 | } else { | |
2264 | kbuf = ERR_PTR(-ENOBUFS); | |
2265 | } | |
2266 | ||
2267 | io_ring_submit_unlock(req->ctx, needs_lock); | |
2268 | ||
2269 | return kbuf; | |
2270 | } | |
2271 | ||
4d954c25 JA |
2272 | static void __user *io_rw_buffer_select(struct io_kiocb *req, size_t *len, |
2273 | bool needs_lock) | |
2274 | { | |
2275 | struct io_buffer *kbuf; | |
2276 | int bgid; | |
2277 | ||
2278 | kbuf = (struct io_buffer *) (unsigned long) req->rw.addr; | |
2279 | bgid = (int) (unsigned long) req->rw.kiocb.private; | |
2280 | kbuf = io_buffer_select(req, len, bgid, kbuf, needs_lock); | |
2281 | if (IS_ERR(kbuf)) | |
2282 | return kbuf; | |
2283 | req->rw.addr = (u64) (unsigned long) kbuf; | |
2284 | req->flags |= REQ_F_BUFFER_SELECTED; | |
2285 | return u64_to_user_ptr(kbuf->addr); | |
2286 | } | |
2287 | ||
2288 | #ifdef CONFIG_COMPAT | |
2289 | static ssize_t io_compat_import(struct io_kiocb *req, struct iovec *iov, | |
2290 | bool needs_lock) | |
2291 | { | |
2292 | struct compat_iovec __user *uiov; | |
2293 | compat_ssize_t clen; | |
2294 | void __user *buf; | |
2295 | ssize_t len; | |
2296 | ||
2297 | uiov = u64_to_user_ptr(req->rw.addr); | |
2298 | if (!access_ok(uiov, sizeof(*uiov))) | |
2299 | return -EFAULT; | |
2300 | if (__get_user(clen, &uiov->iov_len)) | |
2301 | return -EFAULT; | |
2302 | if (clen < 0) | |
2303 | return -EINVAL; | |
2304 | ||
2305 | len = clen; | |
2306 | buf = io_rw_buffer_select(req, &len, needs_lock); | |
2307 | if (IS_ERR(buf)) | |
2308 | return PTR_ERR(buf); | |
2309 | iov[0].iov_base = buf; | |
2310 | iov[0].iov_len = (compat_size_t) len; | |
2311 | return 0; | |
2312 | } | |
2313 | #endif | |
2314 | ||
2315 | static ssize_t __io_iov_buffer_select(struct io_kiocb *req, struct iovec *iov, | |
2316 | bool needs_lock) | |
2317 | { | |
2318 | struct iovec __user *uiov = u64_to_user_ptr(req->rw.addr); | |
2319 | void __user *buf; | |
2320 | ssize_t len; | |
2321 | ||
2322 | if (copy_from_user(iov, uiov, sizeof(*uiov))) | |
2323 | return -EFAULT; | |
2324 | ||
2325 | len = iov[0].iov_len; | |
2326 | if (len < 0) | |
2327 | return -EINVAL; | |
2328 | buf = io_rw_buffer_select(req, &len, needs_lock); | |
2329 | if (IS_ERR(buf)) | |
2330 | return PTR_ERR(buf); | |
2331 | iov[0].iov_base = buf; | |
2332 | iov[0].iov_len = len; | |
2333 | return 0; | |
2334 | } | |
2335 | ||
2336 | static ssize_t io_iov_buffer_select(struct io_kiocb *req, struct iovec *iov, | |
2337 | bool needs_lock) | |
2338 | { | |
2339 | if (req->flags & REQ_F_BUFFER_SELECTED) | |
2340 | return 0; | |
2341 | if (!req->rw.len) | |
2342 | return 0; | |
2343 | else if (req->rw.len > 1) | |
2344 | return -EINVAL; | |
2345 | ||
2346 | #ifdef CONFIG_COMPAT | |
2347 | if (req->ctx->compat) | |
2348 | return io_compat_import(req, iov, needs_lock); | |
2349 | #endif | |
2350 | ||
2351 | return __io_iov_buffer_select(req, iov, needs_lock); | |
2352 | } | |
2353 | ||
cf6fd4bd | 2354 | static ssize_t io_import_iovec(int rw, struct io_kiocb *req, |
bcda7baa JA |
2355 | struct iovec **iovec, struct iov_iter *iter, |
2356 | bool needs_lock) | |
2b188cc1 | 2357 | { |
9adbd45d JA |
2358 | void __user *buf = u64_to_user_ptr(req->rw.addr); |
2359 | size_t sqe_len = req->rw.len; | |
4d954c25 | 2360 | ssize_t ret; |
edafccee JA |
2361 | u8 opcode; |
2362 | ||
d625c6ee | 2363 | opcode = req->opcode; |
7d009165 | 2364 | if (opcode == IORING_OP_READ_FIXED || opcode == IORING_OP_WRITE_FIXED) { |
edafccee | 2365 | *iovec = NULL; |
9adbd45d | 2366 | return io_import_fixed(req, rw, iter); |
edafccee | 2367 | } |
2b188cc1 | 2368 | |
bcda7baa JA |
2369 | /* buffer index only valid with fixed read/write, or buffer select */ |
2370 | if (req->rw.kiocb.private && !(req->flags & REQ_F_BUFFER_SELECT)) | |
9adbd45d JA |
2371 | return -EINVAL; |
2372 | ||
3a6820f2 | 2373 | if (opcode == IORING_OP_READ || opcode == IORING_OP_WRITE) { |
bcda7baa | 2374 | if (req->flags & REQ_F_BUFFER_SELECT) { |
4d954c25 JA |
2375 | buf = io_rw_buffer_select(req, &sqe_len, needs_lock); |
2376 | if (IS_ERR(buf)) { | |
bcda7baa | 2377 | *iovec = NULL; |
4d954c25 | 2378 | return PTR_ERR(buf); |
bcda7baa | 2379 | } |
3f9d6441 | 2380 | req->rw.len = sqe_len; |
bcda7baa JA |
2381 | } |
2382 | ||
3a6820f2 JA |
2383 | ret = import_single_range(rw, buf, sqe_len, *iovec, iter); |
2384 | *iovec = NULL; | |
3a901598 | 2385 | return ret < 0 ? ret : sqe_len; |
3a6820f2 JA |
2386 | } |
2387 | ||
f67676d1 JA |
2388 | if (req->io) { |
2389 | struct io_async_rw *iorw = &req->io->rw; | |
2390 | ||
2391 | *iovec = iorw->iov; | |
2392 | iov_iter_init(iter, rw, *iovec, iorw->nr_segs, iorw->size); | |
2393 | if (iorw->iov == iorw->fast_iov) | |
2394 | *iovec = NULL; | |
2395 | return iorw->size; | |
2396 | } | |
2397 | ||
4d954c25 JA |
2398 | if (req->flags & REQ_F_BUFFER_SELECT) { |
2399 | ret = io_iov_buffer_select(req, *iovec, needs_lock); | |
3f9d6441 JA |
2400 | if (!ret) { |
2401 | ret = (*iovec)->iov_len; | |
2402 | iov_iter_init(iter, rw, *iovec, 1, ret); | |
2403 | } | |
4d954c25 JA |
2404 | *iovec = NULL; |
2405 | return ret; | |
2406 | } | |
2407 | ||
2b188cc1 | 2408 | #ifdef CONFIG_COMPAT |
cf6fd4bd | 2409 | if (req->ctx->compat) |
2b188cc1 JA |
2410 | return compat_import_iovec(rw, buf, sqe_len, UIO_FASTIOV, |
2411 | iovec, iter); | |
2412 | #endif | |
2413 | ||
2414 | return import_iovec(rw, buf, sqe_len, UIO_FASTIOV, iovec, iter); | |
2415 | } | |
2416 | ||
31b51510 | 2417 | /* |
32960613 JA |
2418 | * For files that don't have ->read_iter() and ->write_iter(), handle them |
2419 | * by looping over ->read() or ->write() manually. | |
31b51510 | 2420 | */ |
32960613 JA |
2421 | static ssize_t loop_rw_iter(int rw, struct file *file, struct kiocb *kiocb, |
2422 | struct iov_iter *iter) | |
2423 | { | |
2424 | ssize_t ret = 0; | |
2425 | ||
2426 | /* | |
2427 | * Don't support polled IO through this interface, and we can't | |
2428 | * support non-blocking either. For the latter, this just causes | |
2429 | * the kiocb to be handled from an async context. | |
2430 | */ | |
2431 | if (kiocb->ki_flags & IOCB_HIPRI) | |
2432 | return -EOPNOTSUPP; | |
2433 | if (kiocb->ki_flags & IOCB_NOWAIT) | |
2434 | return -EAGAIN; | |
2435 | ||
2436 | while (iov_iter_count(iter)) { | |
311ae9e1 | 2437 | struct iovec iovec; |
32960613 JA |
2438 | ssize_t nr; |
2439 | ||
311ae9e1 PB |
2440 | if (!iov_iter_is_bvec(iter)) { |
2441 | iovec = iov_iter_iovec(iter); | |
2442 | } else { | |
2443 | /* fixed buffers import bvec */ | |
2444 | iovec.iov_base = kmap(iter->bvec->bv_page) | |
2445 | + iter->iov_offset; | |
2446 | iovec.iov_len = min(iter->count, | |
2447 | iter->bvec->bv_len - iter->iov_offset); | |
2448 | } | |
2449 | ||
32960613 JA |
2450 | if (rw == READ) { |
2451 | nr = file->f_op->read(file, iovec.iov_base, | |
2452 | iovec.iov_len, &kiocb->ki_pos); | |
2453 | } else { | |
2454 | nr = file->f_op->write(file, iovec.iov_base, | |
2455 | iovec.iov_len, &kiocb->ki_pos); | |
2456 | } | |
2457 | ||
311ae9e1 PB |
2458 | if (iov_iter_is_bvec(iter)) |
2459 | kunmap(iter->bvec->bv_page); | |
2460 | ||
32960613 JA |
2461 | if (nr < 0) { |
2462 | if (!ret) | |
2463 | ret = nr; | |
2464 | break; | |
2465 | } | |
2466 | ret += nr; | |
2467 | if (nr != iovec.iov_len) | |
2468 | break; | |
2469 | iov_iter_advance(iter, nr); | |
2470 | } | |
2471 | ||
2472 | return ret; | |
2473 | } | |
2474 | ||
b7bb4f7d | 2475 | static void io_req_map_rw(struct io_kiocb *req, ssize_t io_size, |
f67676d1 JA |
2476 | struct iovec *iovec, struct iovec *fast_iov, |
2477 | struct iov_iter *iter) | |
2478 | { | |
2479 | req->io->rw.nr_segs = iter->nr_segs; | |
2480 | req->io->rw.size = io_size; | |
2481 | req->io->rw.iov = iovec; | |
2482 | if (!req->io->rw.iov) { | |
2483 | req->io->rw.iov = req->io->rw.fast_iov; | |
45097dae XW |
2484 | if (req->io->rw.iov != fast_iov) |
2485 | memcpy(req->io->rw.iov, fast_iov, | |
2486 | sizeof(struct iovec) * iter->nr_segs); | |
99bc4c38 PB |
2487 | } else { |
2488 | req->flags |= REQ_F_NEED_CLEANUP; | |
f67676d1 JA |
2489 | } |
2490 | } | |
2491 | ||
3d9932a8 XW |
2492 | static inline int __io_alloc_async_ctx(struct io_kiocb *req) |
2493 | { | |
2494 | req->io = kmalloc(sizeof(*req->io), GFP_KERNEL); | |
2495 | return req->io == NULL; | |
2496 | } | |
2497 | ||
b7bb4f7d | 2498 | static int io_alloc_async_ctx(struct io_kiocb *req) |
f67676d1 | 2499 | { |
d3656344 JA |
2500 | if (!io_op_defs[req->opcode].async_ctx) |
2501 | return 0; | |
3d9932a8 XW |
2502 | |
2503 | return __io_alloc_async_ctx(req); | |
b7bb4f7d JA |
2504 | } |
2505 | ||
b7bb4f7d JA |
2506 | static int io_setup_async_rw(struct io_kiocb *req, ssize_t io_size, |
2507 | struct iovec *iovec, struct iovec *fast_iov, | |
2508 | struct iov_iter *iter) | |
2509 | { | |
980ad263 | 2510 | if (!io_op_defs[req->opcode].async_ctx) |
74566df3 | 2511 | return 0; |
5d204bcf | 2512 | if (!req->io) { |
3d9932a8 | 2513 | if (__io_alloc_async_ctx(req)) |
5d204bcf | 2514 | return -ENOMEM; |
b7bb4f7d | 2515 | |
5d204bcf JA |
2516 | io_req_map_rw(req, io_size, iovec, fast_iov, iter); |
2517 | } | |
b7bb4f7d | 2518 | return 0; |
f67676d1 JA |
2519 | } |
2520 | ||
3529d8c2 JA |
2521 | static int io_read_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe, |
2522 | bool force_nonblock) | |
f67676d1 | 2523 | { |
3529d8c2 JA |
2524 | struct io_async_ctx *io; |
2525 | struct iov_iter iter; | |
f67676d1 JA |
2526 | ssize_t ret; |
2527 | ||
3529d8c2 JA |
2528 | ret = io_prep_rw(req, sqe, force_nonblock); |
2529 | if (ret) | |
2530 | return ret; | |
f67676d1 | 2531 | |
3529d8c2 JA |
2532 | if (unlikely(!(req->file->f_mode & FMODE_READ))) |
2533 | return -EBADF; | |
f67676d1 | 2534 | |
5f798bea PB |
2535 | /* either don't need iovec imported or already have it */ |
2536 | if (!req->io || req->flags & REQ_F_NEED_CLEANUP) | |
3529d8c2 JA |
2537 | return 0; |
2538 | ||
2539 | io = req->io; | |
2540 | io->rw.iov = io->rw.fast_iov; | |
2541 | req->io = NULL; | |
bcda7baa | 2542 | ret = io_import_iovec(READ, req, &io->rw.iov, &iter, !force_nonblock); |
3529d8c2 JA |
2543 | req->io = io; |
2544 | if (ret < 0) | |
2545 | return ret; | |
2546 | ||
2547 | io_req_map_rw(req, ret, io->rw.iov, io->rw.fast_iov, &iter); | |
2548 | return 0; | |
f67676d1 JA |
2549 | } |
2550 | ||
014db007 | 2551 | static int io_read(struct io_kiocb *req, bool force_nonblock) |
2b188cc1 JA |
2552 | { |
2553 | struct iovec inline_vecs[UIO_FASTIOV], *iovec = inline_vecs; | |
9adbd45d | 2554 | struct kiocb *kiocb = &req->rw.kiocb; |
2b188cc1 | 2555 | struct iov_iter iter; |
31b51510 | 2556 | size_t iov_count; |
f67676d1 | 2557 | ssize_t io_size, ret; |
2b188cc1 | 2558 | |
bcda7baa | 2559 | ret = io_import_iovec(READ, req, &iovec, &iter, !force_nonblock); |
06b76d44 JA |
2560 | if (ret < 0) |
2561 | return ret; | |
2b188cc1 | 2562 | |
fd6c2e4c JA |
2563 | /* Ensure we clear previously set non-block flag */ |
2564 | if (!force_nonblock) | |
29de5f6a | 2565 | kiocb->ki_flags &= ~IOCB_NOWAIT; |
fd6c2e4c | 2566 | |
797f3f53 | 2567 | req->result = 0; |
f67676d1 | 2568 | io_size = ret; |
dea3b49c | 2569 | if (req->flags & REQ_F_LINK_HEAD) |
f67676d1 JA |
2570 | req->result = io_size; |
2571 | ||
2572 | /* | |
2573 | * If the file doesn't support async, mark it as REQ_F_MUST_PUNT so | |
2574 | * we know to async punt it even if it was opened O_NONBLOCK | |
2575 | */ | |
af197f50 | 2576 | if (force_nonblock && !io_file_supports_async(req->file, READ)) |
f67676d1 | 2577 | goto copy_iov; |
9e645e11 | 2578 | |
31b51510 | 2579 | iov_count = iov_iter_count(&iter); |
9adbd45d | 2580 | ret = rw_verify_area(READ, req->file, &kiocb->ki_pos, iov_count); |
2b188cc1 JA |
2581 | if (!ret) { |
2582 | ssize_t ret2; | |
2583 | ||
9adbd45d JA |
2584 | if (req->file->f_op->read_iter) |
2585 | ret2 = call_read_iter(req->file, kiocb, &iter); | |
32960613 | 2586 | else |
9adbd45d | 2587 | ret2 = loop_rw_iter(READ, req->file, kiocb, &iter); |
32960613 | 2588 | |
9d93a3f5 | 2589 | /* Catch -EAGAIN return for forced non-blocking submission */ |
f67676d1 | 2590 | if (!force_nonblock || ret2 != -EAGAIN) { |
014db007 | 2591 | kiocb_done(kiocb, ret2); |
f67676d1 JA |
2592 | } else { |
2593 | copy_iov: | |
b7bb4f7d | 2594 | ret = io_setup_async_rw(req, io_size, iovec, |
f67676d1 JA |
2595 | inline_vecs, &iter); |
2596 | if (ret) | |
2597 | goto out_free; | |
29de5f6a | 2598 | /* any defer here is final, must blocking retry */ |
490e8967 JA |
2599 | if (!(req->flags & REQ_F_NOWAIT) && |
2600 | !file_can_poll(req->file)) | |
29de5f6a | 2601 | req->flags |= REQ_F_MUST_PUNT; |
f67676d1 JA |
2602 | return -EAGAIN; |
2603 | } | |
2b188cc1 | 2604 | } |
f67676d1 | 2605 | out_free: |
1e95081c | 2606 | kfree(iovec); |
99bc4c38 | 2607 | req->flags &= ~REQ_F_NEED_CLEANUP; |
2b188cc1 JA |
2608 | return ret; |
2609 | } | |
2610 | ||
3529d8c2 JA |
2611 | static int io_write_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe, |
2612 | bool force_nonblock) | |
f67676d1 | 2613 | { |
3529d8c2 JA |
2614 | struct io_async_ctx *io; |
2615 | struct iov_iter iter; | |
f67676d1 JA |
2616 | ssize_t ret; |
2617 | ||
3529d8c2 JA |
2618 | ret = io_prep_rw(req, sqe, force_nonblock); |
2619 | if (ret) | |
2620 | return ret; | |
f67676d1 | 2621 | |
3529d8c2 JA |
2622 | if (unlikely(!(req->file->f_mode & FMODE_WRITE))) |
2623 | return -EBADF; | |
f67676d1 | 2624 | |
4ed734b0 JA |
2625 | req->fsize = rlimit(RLIMIT_FSIZE); |
2626 | ||
5f798bea PB |
2627 | /* either don't need iovec imported or already have it */ |
2628 | if (!req->io || req->flags & REQ_F_NEED_CLEANUP) | |
3529d8c2 JA |
2629 | return 0; |
2630 | ||
2631 | io = req->io; | |
2632 | io->rw.iov = io->rw.fast_iov; | |
2633 | req->io = NULL; | |
bcda7baa | 2634 | ret = io_import_iovec(WRITE, req, &io->rw.iov, &iter, !force_nonblock); |
3529d8c2 JA |
2635 | req->io = io; |
2636 | if (ret < 0) | |
2637 | return ret; | |
2638 | ||
2639 | io_req_map_rw(req, ret, io->rw.iov, io->rw.fast_iov, &iter); | |
2640 | return 0; | |
f67676d1 JA |
2641 | } |
2642 | ||
014db007 | 2643 | static int io_write(struct io_kiocb *req, bool force_nonblock) |
2b188cc1 JA |
2644 | { |
2645 | struct iovec inline_vecs[UIO_FASTIOV], *iovec = inline_vecs; | |
9adbd45d | 2646 | struct kiocb *kiocb = &req->rw.kiocb; |
2b188cc1 | 2647 | struct iov_iter iter; |
31b51510 | 2648 | size_t iov_count; |
f67676d1 | 2649 | ssize_t ret, io_size; |
2b188cc1 | 2650 | |
bcda7baa | 2651 | ret = io_import_iovec(WRITE, req, &iovec, &iter, !force_nonblock); |
06b76d44 JA |
2652 | if (ret < 0) |
2653 | return ret; | |
2b188cc1 | 2654 | |
fd6c2e4c JA |
2655 | /* Ensure we clear previously set non-block flag */ |
2656 | if (!force_nonblock) | |
9adbd45d | 2657 | req->rw.kiocb.ki_flags &= ~IOCB_NOWAIT; |
fd6c2e4c | 2658 | |
797f3f53 | 2659 | req->result = 0; |
f67676d1 | 2660 | io_size = ret; |
dea3b49c | 2661 | if (req->flags & REQ_F_LINK_HEAD) |
f67676d1 | 2662 | req->result = io_size; |
9e645e11 | 2663 | |
f67676d1 JA |
2664 | /* |
2665 | * If the file doesn't support async, mark it as REQ_F_MUST_PUNT so | |
2666 | * we know to async punt it even if it was opened O_NONBLOCK | |
2667 | */ | |
af197f50 | 2668 | if (force_nonblock && !io_file_supports_async(req->file, WRITE)) |
f67676d1 | 2669 | goto copy_iov; |
31b51510 | 2670 | |
10d59345 JA |
2671 | /* file path doesn't support NOWAIT for non-direct_IO */ |
2672 | if (force_nonblock && !(kiocb->ki_flags & IOCB_DIRECT) && | |
2673 | (req->flags & REQ_F_ISREG)) | |
f67676d1 | 2674 | goto copy_iov; |
31b51510 | 2675 | |
f67676d1 | 2676 | iov_count = iov_iter_count(&iter); |
9adbd45d | 2677 | ret = rw_verify_area(WRITE, req->file, &kiocb->ki_pos, iov_count); |
2b188cc1 | 2678 | if (!ret) { |
9bf7933f RP |
2679 | ssize_t ret2; |
2680 | ||
2b188cc1 JA |
2681 | /* |
2682 | * Open-code file_start_write here to grab freeze protection, | |
2683 | * which will be released by another thread in | |
2684 | * io_complete_rw(). Fool lockdep by telling it the lock got | |
2685 | * released so that it doesn't complain about the held lock when | |
2686 | * we return to userspace. | |
2687 | */ | |
491381ce | 2688 | if (req->flags & REQ_F_ISREG) { |
9adbd45d | 2689 | __sb_start_write(file_inode(req->file)->i_sb, |
2b188cc1 | 2690 | SB_FREEZE_WRITE, true); |
9adbd45d | 2691 | __sb_writers_release(file_inode(req->file)->i_sb, |
2b188cc1 JA |
2692 | SB_FREEZE_WRITE); |
2693 | } | |
2694 | kiocb->ki_flags |= IOCB_WRITE; | |
9bf7933f | 2695 | |
4ed734b0 JA |
2696 | if (!force_nonblock) |
2697 | current->signal->rlim[RLIMIT_FSIZE].rlim_cur = req->fsize; | |
2698 | ||
9adbd45d JA |
2699 | if (req->file->f_op->write_iter) |
2700 | ret2 = call_write_iter(req->file, kiocb, &iter); | |
32960613 | 2701 | else |
9adbd45d | 2702 | ret2 = loop_rw_iter(WRITE, req->file, kiocb, &iter); |
4ed734b0 JA |
2703 | |
2704 | if (!force_nonblock) | |
2705 | current->signal->rlim[RLIMIT_FSIZE].rlim_cur = RLIM_INFINITY; | |
2706 | ||
faac996c | 2707 | /* |
bff6035d | 2708 | * Raw bdev writes will return -EOPNOTSUPP for IOCB_NOWAIT. Just |
faac996c JA |
2709 | * retry them without IOCB_NOWAIT. |
2710 | */ | |
2711 | if (ret2 == -EOPNOTSUPP && (kiocb->ki_flags & IOCB_NOWAIT)) | |
2712 | ret2 = -EAGAIN; | |
f67676d1 | 2713 | if (!force_nonblock || ret2 != -EAGAIN) { |
014db007 | 2714 | kiocb_done(kiocb, ret2); |
f67676d1 JA |
2715 | } else { |
2716 | copy_iov: | |
b7bb4f7d | 2717 | ret = io_setup_async_rw(req, io_size, iovec, |
f67676d1 JA |
2718 | inline_vecs, &iter); |
2719 | if (ret) | |
2720 | goto out_free; | |
29de5f6a | 2721 | /* any defer here is final, must blocking retry */ |
490e8967 JA |
2722 | if (!file_can_poll(req->file)) |
2723 | req->flags |= REQ_F_MUST_PUNT; | |
f67676d1 JA |
2724 | return -EAGAIN; |
2725 | } | |
2b188cc1 | 2726 | } |
31b51510 | 2727 | out_free: |
99bc4c38 | 2728 | req->flags &= ~REQ_F_NEED_CLEANUP; |
1e95081c | 2729 | kfree(iovec); |
2b188cc1 JA |
2730 | return ret; |
2731 | } | |
2732 | ||
7d67af2c PB |
2733 | static int io_splice_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
2734 | { | |
2735 | struct io_splice* sp = &req->splice; | |
2736 | unsigned int valid_flags = SPLICE_F_FD_IN_FIXED | SPLICE_F_ALL; | |
2737 | int ret; | |
2738 | ||
2739 | if (req->flags & REQ_F_NEED_CLEANUP) | |
2740 | return 0; | |
2741 | ||
2742 | sp->file_in = NULL; | |
2743 | sp->off_in = READ_ONCE(sqe->splice_off_in); | |
2744 | sp->off_out = READ_ONCE(sqe->off); | |
2745 | sp->len = READ_ONCE(sqe->len); | |
2746 | sp->flags = READ_ONCE(sqe->splice_flags); | |
2747 | ||
2748 | if (unlikely(sp->flags & ~valid_flags)) | |
2749 | return -EINVAL; | |
2750 | ||
2751 | ret = io_file_get(NULL, req, READ_ONCE(sqe->splice_fd_in), &sp->file_in, | |
2752 | (sp->flags & SPLICE_F_FD_IN_FIXED)); | |
2753 | if (ret) | |
2754 | return ret; | |
2755 | req->flags |= REQ_F_NEED_CLEANUP; | |
2756 | ||
2757 | if (!S_ISREG(file_inode(sp->file_in)->i_mode)) | |
2758 | req->work.flags |= IO_WQ_WORK_UNBOUND; | |
2759 | ||
2760 | return 0; | |
2761 | } | |
2762 | ||
014db007 | 2763 | static int io_splice(struct io_kiocb *req, bool force_nonblock) |
7d67af2c PB |
2764 | { |
2765 | struct io_splice *sp = &req->splice; | |
2766 | struct file *in = sp->file_in; | |
2767 | struct file *out = sp->file_out; | |
2768 | unsigned int flags = sp->flags & ~SPLICE_F_FD_IN_FIXED; | |
2769 | loff_t *poff_in, *poff_out; | |
2770 | long ret; | |
2771 | ||
2fb3e822 PB |
2772 | if (force_nonblock) |
2773 | return -EAGAIN; | |
7d67af2c PB |
2774 | |
2775 | poff_in = (sp->off_in == -1) ? NULL : &sp->off_in; | |
2776 | poff_out = (sp->off_out == -1) ? NULL : &sp->off_out; | |
2777 | ret = do_splice(in, poff_in, out, poff_out, sp->len, flags); | |
2778 | if (force_nonblock && ret == -EAGAIN) | |
2779 | return -EAGAIN; | |
2780 | ||
2781 | io_put_file(req, in, (sp->flags & SPLICE_F_FD_IN_FIXED)); | |
2782 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
2783 | ||
2784 | io_cqring_add_event(req, ret); | |
2785 | if (ret != sp->len) | |
2786 | req_set_fail_links(req); | |
014db007 | 2787 | io_put_req(req); |
7d67af2c PB |
2788 | return 0; |
2789 | } | |
2790 | ||
2b188cc1 JA |
2791 | /* |
2792 | * IORING_OP_NOP just posts a completion event, nothing else. | |
2793 | */ | |
78e19bbe | 2794 | static int io_nop(struct io_kiocb *req) |
2b188cc1 JA |
2795 | { |
2796 | struct io_ring_ctx *ctx = req->ctx; | |
2b188cc1 | 2797 | |
def596e9 JA |
2798 | if (unlikely(ctx->flags & IORING_SETUP_IOPOLL)) |
2799 | return -EINVAL; | |
2800 | ||
78e19bbe | 2801 | io_cqring_add_event(req, 0); |
e65ef56d | 2802 | io_put_req(req); |
2b188cc1 JA |
2803 | return 0; |
2804 | } | |
2805 | ||
3529d8c2 | 2806 | static int io_prep_fsync(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
c992fe29 | 2807 | { |
6b06314c | 2808 | struct io_ring_ctx *ctx = req->ctx; |
c992fe29 | 2809 | |
09bb8394 JA |
2810 | if (!req->file) |
2811 | return -EBADF; | |
c992fe29 | 2812 | |
6b06314c | 2813 | if (unlikely(ctx->flags & IORING_SETUP_IOPOLL)) |
def596e9 | 2814 | return -EINVAL; |
edafccee | 2815 | if (unlikely(sqe->addr || sqe->ioprio || sqe->buf_index)) |
c992fe29 CH |
2816 | return -EINVAL; |
2817 | ||
8ed8d3c3 JA |
2818 | req->sync.flags = READ_ONCE(sqe->fsync_flags); |
2819 | if (unlikely(req->sync.flags & ~IORING_FSYNC_DATASYNC)) | |
2820 | return -EINVAL; | |
2821 | ||
2822 | req->sync.off = READ_ONCE(sqe->off); | |
2823 | req->sync.len = READ_ONCE(sqe->len); | |
c992fe29 CH |
2824 | return 0; |
2825 | } | |
2826 | ||
8ed8d3c3 JA |
2827 | static bool io_req_cancelled(struct io_kiocb *req) |
2828 | { | |
2829 | if (req->work.flags & IO_WQ_WORK_CANCEL) { | |
2830 | req_set_fail_links(req); | |
2831 | io_cqring_add_event(req, -ECANCELED); | |
2832 | io_put_req(req); | |
2833 | return true; | |
2834 | } | |
2835 | ||
2836 | return false; | |
2837 | } | |
2838 | ||
014db007 | 2839 | static void __io_fsync(struct io_kiocb *req) |
8ed8d3c3 | 2840 | { |
8ed8d3c3 | 2841 | loff_t end = req->sync.off + req->sync.len; |
8ed8d3c3 JA |
2842 | int ret; |
2843 | ||
9adbd45d | 2844 | ret = vfs_fsync_range(req->file, req->sync.off, |
8ed8d3c3 JA |
2845 | end > 0 ? end : LLONG_MAX, |
2846 | req->sync.flags & IORING_FSYNC_DATASYNC); | |
2847 | if (ret < 0) | |
2848 | req_set_fail_links(req); | |
2849 | io_cqring_add_event(req, ret); | |
014db007 | 2850 | io_put_req(req); |
8ed8d3c3 JA |
2851 | } |
2852 | ||
5ea62161 | 2853 | static void io_fsync_finish(struct io_wq_work **workptr) |
c992fe29 | 2854 | { |
5ea62161 | 2855 | struct io_kiocb *req = container_of(*workptr, struct io_kiocb, work); |
5ea62161 PB |
2856 | |
2857 | if (io_req_cancelled(req)) | |
2858 | return; | |
014db007 | 2859 | __io_fsync(req); |
e9fd9396 | 2860 | io_steal_work(req, workptr); |
8ed8d3c3 | 2861 | } |
c992fe29 | 2862 | |
014db007 | 2863 | static int io_fsync(struct io_kiocb *req, bool force_nonblock) |
c992fe29 | 2864 | { |
c992fe29 | 2865 | /* fsync always requires a blocking context */ |
8ed8d3c3 | 2866 | if (force_nonblock) { |
8ed8d3c3 | 2867 | req->work.func = io_fsync_finish; |
c992fe29 | 2868 | return -EAGAIN; |
8ed8d3c3 | 2869 | } |
014db007 | 2870 | __io_fsync(req); |
c992fe29 CH |
2871 | return 0; |
2872 | } | |
2873 | ||
014db007 | 2874 | static void __io_fallocate(struct io_kiocb *req) |
8ed8d3c3 | 2875 | { |
8ed8d3c3 JA |
2876 | int ret; |
2877 | ||
4ed734b0 | 2878 | current->signal->rlim[RLIMIT_FSIZE].rlim_cur = req->fsize; |
d63d1b5e JA |
2879 | ret = vfs_fallocate(req->file, req->sync.mode, req->sync.off, |
2880 | req->sync.len); | |
4ed734b0 | 2881 | current->signal->rlim[RLIMIT_FSIZE].rlim_cur = RLIM_INFINITY; |
8ed8d3c3 JA |
2882 | if (ret < 0) |
2883 | req_set_fail_links(req); | |
2884 | io_cqring_add_event(req, ret); | |
014db007 | 2885 | io_put_req(req); |
5ea62161 PB |
2886 | } |
2887 | ||
2888 | static void io_fallocate_finish(struct io_wq_work **workptr) | |
2889 | { | |
2890 | struct io_kiocb *req = container_of(*workptr, struct io_kiocb, work); | |
5ea62161 | 2891 | |
594506fe PB |
2892 | if (io_req_cancelled(req)) |
2893 | return; | |
014db007 | 2894 | __io_fallocate(req); |
e9fd9396 | 2895 | io_steal_work(req, workptr); |
5d17b4a4 JA |
2896 | } |
2897 | ||
d63d1b5e JA |
2898 | static int io_fallocate_prep(struct io_kiocb *req, |
2899 | const struct io_uring_sqe *sqe) | |
2900 | { | |
2901 | if (sqe->ioprio || sqe->buf_index || sqe->rw_flags) | |
2902 | return -EINVAL; | |
2903 | ||
2904 | req->sync.off = READ_ONCE(sqe->off); | |
2905 | req->sync.len = READ_ONCE(sqe->addr); | |
2906 | req->sync.mode = READ_ONCE(sqe->len); | |
4ed734b0 | 2907 | req->fsize = rlimit(RLIMIT_FSIZE); |
d63d1b5e JA |
2908 | return 0; |
2909 | } | |
2910 | ||
014db007 | 2911 | static int io_fallocate(struct io_kiocb *req, bool force_nonblock) |
5d17b4a4 | 2912 | { |
d63d1b5e | 2913 | /* fallocate always requiring blocking context */ |
8ed8d3c3 | 2914 | if (force_nonblock) { |
d63d1b5e | 2915 | req->work.func = io_fallocate_finish; |
5d17b4a4 | 2916 | return -EAGAIN; |
8ed8d3c3 | 2917 | } |
5d17b4a4 | 2918 | |
014db007 | 2919 | __io_fallocate(req); |
5d17b4a4 JA |
2920 | return 0; |
2921 | } | |
2922 | ||
15b71abe | 2923 | static int io_openat_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
b7bb4f7d | 2924 | { |
f8748881 | 2925 | const char __user *fname; |
15b71abe | 2926 | int ret; |
b7bb4f7d | 2927 | |
15b71abe JA |
2928 | if (sqe->ioprio || sqe->buf_index) |
2929 | return -EINVAL; | |
9c280f90 | 2930 | if (req->flags & REQ_F_FIXED_FILE) |
cf3040ca | 2931 | return -EBADF; |
0bdbdd08 PB |
2932 | if (req->flags & REQ_F_NEED_CLEANUP) |
2933 | return 0; | |
03b1230c | 2934 | |
15b71abe | 2935 | req->open.dfd = READ_ONCE(sqe->fd); |
c12cedf2 | 2936 | req->open.how.mode = READ_ONCE(sqe->len); |
f8748881 | 2937 | fname = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
c12cedf2 | 2938 | req->open.how.flags = READ_ONCE(sqe->open_flags); |
08a1d26e JA |
2939 | if (force_o_largefile()) |
2940 | req->open.how.flags |= O_LARGEFILE; | |
3529d8c2 | 2941 | |
f8748881 | 2942 | req->open.filename = getname(fname); |
15b71abe JA |
2943 | if (IS_ERR(req->open.filename)) { |
2944 | ret = PTR_ERR(req->open.filename); | |
2945 | req->open.filename = NULL; | |
2946 | return ret; | |
2947 | } | |
3529d8c2 | 2948 | |
4022e7af | 2949 | req->open.nofile = rlimit(RLIMIT_NOFILE); |
8fef80bf | 2950 | req->flags |= REQ_F_NEED_CLEANUP; |
15b71abe | 2951 | return 0; |
03b1230c JA |
2952 | } |
2953 | ||
cebdb986 | 2954 | static int io_openat2_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
aa1fa28f | 2955 | { |
cebdb986 JA |
2956 | struct open_how __user *how; |
2957 | const char __user *fname; | |
2958 | size_t len; | |
0fa03c62 JA |
2959 | int ret; |
2960 | ||
cebdb986 | 2961 | if (sqe->ioprio || sqe->buf_index) |
0fa03c62 | 2962 | return -EINVAL; |
9c280f90 | 2963 | if (req->flags & REQ_F_FIXED_FILE) |
cf3040ca | 2964 | return -EBADF; |
0bdbdd08 PB |
2965 | if (req->flags & REQ_F_NEED_CLEANUP) |
2966 | return 0; | |
0fa03c62 | 2967 | |
cebdb986 JA |
2968 | req->open.dfd = READ_ONCE(sqe->fd); |
2969 | fname = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
2970 | how = u64_to_user_ptr(READ_ONCE(sqe->addr2)); | |
2971 | len = READ_ONCE(sqe->len); | |
0fa03c62 | 2972 | |
cebdb986 JA |
2973 | if (len < OPEN_HOW_SIZE_VER0) |
2974 | return -EINVAL; | |
3529d8c2 | 2975 | |
cebdb986 JA |
2976 | ret = copy_struct_from_user(&req->open.how, sizeof(req->open.how), how, |
2977 | len); | |
2978 | if (ret) | |
2979 | return ret; | |
3529d8c2 | 2980 | |
cebdb986 JA |
2981 | if (!(req->open.how.flags & O_PATH) && force_o_largefile()) |
2982 | req->open.how.flags |= O_LARGEFILE; | |
0fa03c62 | 2983 | |
cebdb986 JA |
2984 | req->open.filename = getname(fname); |
2985 | if (IS_ERR(req->open.filename)) { | |
2986 | ret = PTR_ERR(req->open.filename); | |
2987 | req->open.filename = NULL; | |
2988 | return ret; | |
2989 | } | |
2990 | ||
4022e7af | 2991 | req->open.nofile = rlimit(RLIMIT_NOFILE); |
8fef80bf | 2992 | req->flags |= REQ_F_NEED_CLEANUP; |
cebdb986 JA |
2993 | return 0; |
2994 | } | |
2995 | ||
014db007 | 2996 | static int io_openat2(struct io_kiocb *req, bool force_nonblock) |
15b71abe JA |
2997 | { |
2998 | struct open_flags op; | |
15b71abe JA |
2999 | struct file *file; |
3000 | int ret; | |
3001 | ||
f86cd20c | 3002 | if (force_nonblock) |
15b71abe | 3003 | return -EAGAIN; |
15b71abe | 3004 | |
cebdb986 | 3005 | ret = build_open_flags(&req->open.how, &op); |
15b71abe JA |
3006 | if (ret) |
3007 | goto err; | |
3008 | ||
4022e7af | 3009 | ret = __get_unused_fd_flags(req->open.how.flags, req->open.nofile); |
15b71abe JA |
3010 | if (ret < 0) |
3011 | goto err; | |
3012 | ||
3013 | file = do_filp_open(req->open.dfd, req->open.filename, &op); | |
3014 | if (IS_ERR(file)) { | |
3015 | put_unused_fd(ret); | |
3016 | ret = PTR_ERR(file); | |
3017 | } else { | |
3018 | fsnotify_open(file); | |
3019 | fd_install(ret, file); | |
3020 | } | |
3021 | err: | |
3022 | putname(req->open.filename); | |
8fef80bf | 3023 | req->flags &= ~REQ_F_NEED_CLEANUP; |
15b71abe JA |
3024 | if (ret < 0) |
3025 | req_set_fail_links(req); | |
3026 | io_cqring_add_event(req, ret); | |
014db007 | 3027 | io_put_req(req); |
15b71abe JA |
3028 | return 0; |
3029 | } | |
3030 | ||
014db007 | 3031 | static int io_openat(struct io_kiocb *req, bool force_nonblock) |
cebdb986 JA |
3032 | { |
3033 | req->open.how = build_open_how(req->open.how.flags, req->open.how.mode); | |
014db007 | 3034 | return io_openat2(req, force_nonblock); |
cebdb986 JA |
3035 | } |
3036 | ||
067524e9 JA |
3037 | static int io_remove_buffers_prep(struct io_kiocb *req, |
3038 | const struct io_uring_sqe *sqe) | |
3039 | { | |
3040 | struct io_provide_buf *p = &req->pbuf; | |
3041 | u64 tmp; | |
3042 | ||
3043 | if (sqe->ioprio || sqe->rw_flags || sqe->addr || sqe->len || sqe->off) | |
3044 | return -EINVAL; | |
3045 | ||
3046 | tmp = READ_ONCE(sqe->fd); | |
3047 | if (!tmp || tmp > USHRT_MAX) | |
3048 | return -EINVAL; | |
3049 | ||
3050 | memset(p, 0, sizeof(*p)); | |
3051 | p->nbufs = tmp; | |
3052 | p->bgid = READ_ONCE(sqe->buf_group); | |
3053 | return 0; | |
3054 | } | |
3055 | ||
3056 | static int __io_remove_buffers(struct io_ring_ctx *ctx, struct io_buffer *buf, | |
3057 | int bgid, unsigned nbufs) | |
3058 | { | |
3059 | unsigned i = 0; | |
3060 | ||
3061 | /* shouldn't happen */ | |
3062 | if (!nbufs) | |
3063 | return 0; | |
3064 | ||
3065 | /* the head kbuf is the list itself */ | |
3066 | while (!list_empty(&buf->list)) { | |
3067 | struct io_buffer *nxt; | |
3068 | ||
3069 | nxt = list_first_entry(&buf->list, struct io_buffer, list); | |
3070 | list_del(&nxt->list); | |
3071 | kfree(nxt); | |
3072 | if (++i == nbufs) | |
3073 | return i; | |
3074 | } | |
3075 | i++; | |
3076 | kfree(buf); | |
3077 | idr_remove(&ctx->io_buffer_idr, bgid); | |
3078 | ||
3079 | return i; | |
3080 | } | |
3081 | ||
3082 | static int io_remove_buffers(struct io_kiocb *req, bool force_nonblock) | |
3083 | { | |
3084 | struct io_provide_buf *p = &req->pbuf; | |
3085 | struct io_ring_ctx *ctx = req->ctx; | |
3086 | struct io_buffer *head; | |
3087 | int ret = 0; | |
3088 | ||
3089 | io_ring_submit_lock(ctx, !force_nonblock); | |
3090 | ||
3091 | lockdep_assert_held(&ctx->uring_lock); | |
3092 | ||
3093 | ret = -ENOENT; | |
3094 | head = idr_find(&ctx->io_buffer_idr, p->bgid); | |
3095 | if (head) | |
3096 | ret = __io_remove_buffers(ctx, head, p->bgid, p->nbufs); | |
3097 | ||
3098 | io_ring_submit_lock(ctx, !force_nonblock); | |
3099 | if (ret < 0) | |
3100 | req_set_fail_links(req); | |
3101 | io_cqring_add_event(req, ret); | |
3102 | io_put_req(req); | |
3103 | return 0; | |
3104 | } | |
3105 | ||
ddf0322d JA |
3106 | static int io_provide_buffers_prep(struct io_kiocb *req, |
3107 | const struct io_uring_sqe *sqe) | |
3108 | { | |
3109 | struct io_provide_buf *p = &req->pbuf; | |
3110 | u64 tmp; | |
3111 | ||
3112 | if (sqe->ioprio || sqe->rw_flags) | |
3113 | return -EINVAL; | |
3114 | ||
3115 | tmp = READ_ONCE(sqe->fd); | |
3116 | if (!tmp || tmp > USHRT_MAX) | |
3117 | return -E2BIG; | |
3118 | p->nbufs = tmp; | |
3119 | p->addr = READ_ONCE(sqe->addr); | |
3120 | p->len = READ_ONCE(sqe->len); | |
3121 | ||
3122 | if (!access_ok(u64_to_user_ptr(p->addr), p->len)) | |
3123 | return -EFAULT; | |
3124 | ||
3125 | p->bgid = READ_ONCE(sqe->buf_group); | |
3126 | tmp = READ_ONCE(sqe->off); | |
3127 | if (tmp > USHRT_MAX) | |
3128 | return -E2BIG; | |
3129 | p->bid = tmp; | |
3130 | return 0; | |
3131 | } | |
3132 | ||
3133 | static int io_add_buffers(struct io_provide_buf *pbuf, struct io_buffer **head) | |
3134 | { | |
3135 | struct io_buffer *buf; | |
3136 | u64 addr = pbuf->addr; | |
3137 | int i, bid = pbuf->bid; | |
3138 | ||
3139 | for (i = 0; i < pbuf->nbufs; i++) { | |
3140 | buf = kmalloc(sizeof(*buf), GFP_KERNEL); | |
3141 | if (!buf) | |
3142 | break; | |
3143 | ||
3144 | buf->addr = addr; | |
3145 | buf->len = pbuf->len; | |
3146 | buf->bid = bid; | |
3147 | addr += pbuf->len; | |
3148 | bid++; | |
3149 | if (!*head) { | |
3150 | INIT_LIST_HEAD(&buf->list); | |
3151 | *head = buf; | |
3152 | } else { | |
3153 | list_add_tail(&buf->list, &(*head)->list); | |
3154 | } | |
3155 | } | |
3156 | ||
3157 | return i ? i : -ENOMEM; | |
3158 | } | |
3159 | ||
ddf0322d JA |
3160 | static int io_provide_buffers(struct io_kiocb *req, bool force_nonblock) |
3161 | { | |
3162 | struct io_provide_buf *p = &req->pbuf; | |
3163 | struct io_ring_ctx *ctx = req->ctx; | |
3164 | struct io_buffer *head, *list; | |
3165 | int ret = 0; | |
3166 | ||
3167 | io_ring_submit_lock(ctx, !force_nonblock); | |
3168 | ||
3169 | lockdep_assert_held(&ctx->uring_lock); | |
3170 | ||
3171 | list = head = idr_find(&ctx->io_buffer_idr, p->bgid); | |
3172 | ||
3173 | ret = io_add_buffers(p, &head); | |
3174 | if (ret < 0) | |
3175 | goto out; | |
3176 | ||
3177 | if (!list) { | |
3178 | ret = idr_alloc(&ctx->io_buffer_idr, head, p->bgid, p->bgid + 1, | |
3179 | GFP_KERNEL); | |
3180 | if (ret < 0) { | |
067524e9 | 3181 | __io_remove_buffers(ctx, head, p->bgid, -1U); |
ddf0322d JA |
3182 | goto out; |
3183 | } | |
3184 | } | |
3185 | out: | |
3186 | io_ring_submit_unlock(ctx, !force_nonblock); | |
3187 | if (ret < 0) | |
3188 | req_set_fail_links(req); | |
3189 | io_cqring_add_event(req, ret); | |
3190 | io_put_req(req); | |
3191 | return 0; | |
cebdb986 JA |
3192 | } |
3193 | ||
3e4827b0 JA |
3194 | static int io_epoll_ctl_prep(struct io_kiocb *req, |
3195 | const struct io_uring_sqe *sqe) | |
3196 | { | |
3197 | #if defined(CONFIG_EPOLL) | |
3198 | if (sqe->ioprio || sqe->buf_index) | |
3199 | return -EINVAL; | |
3200 | ||
3201 | req->epoll.epfd = READ_ONCE(sqe->fd); | |
3202 | req->epoll.op = READ_ONCE(sqe->len); | |
3203 | req->epoll.fd = READ_ONCE(sqe->off); | |
3204 | ||
3205 | if (ep_op_has_event(req->epoll.op)) { | |
3206 | struct epoll_event __user *ev; | |
3207 | ||
3208 | ev = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
3209 | if (copy_from_user(&req->epoll.event, ev, sizeof(*ev))) | |
3210 | return -EFAULT; | |
3211 | } | |
3212 | ||
3213 | return 0; | |
3214 | #else | |
3215 | return -EOPNOTSUPP; | |
3216 | #endif | |
3217 | } | |
3218 | ||
014db007 | 3219 | static int io_epoll_ctl(struct io_kiocb *req, bool force_nonblock) |
3e4827b0 JA |
3220 | { |
3221 | #if defined(CONFIG_EPOLL) | |
3222 | struct io_epoll *ie = &req->epoll; | |
3223 | int ret; | |
3224 | ||
3225 | ret = do_epoll_ctl(ie->epfd, ie->op, ie->fd, &ie->event, force_nonblock); | |
3226 | if (force_nonblock && ret == -EAGAIN) | |
3227 | return -EAGAIN; | |
3228 | ||
3229 | if (ret < 0) | |
3230 | req_set_fail_links(req); | |
3231 | io_cqring_add_event(req, ret); | |
014db007 | 3232 | io_put_req(req); |
3e4827b0 JA |
3233 | return 0; |
3234 | #else | |
3235 | return -EOPNOTSUPP; | |
3236 | #endif | |
3237 | } | |
3238 | ||
c1ca757b JA |
3239 | static int io_madvise_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
3240 | { | |
3241 | #if defined(CONFIG_ADVISE_SYSCALLS) && defined(CONFIG_MMU) | |
3242 | if (sqe->ioprio || sqe->buf_index || sqe->off) | |
3243 | return -EINVAL; | |
3244 | ||
3245 | req->madvise.addr = READ_ONCE(sqe->addr); | |
3246 | req->madvise.len = READ_ONCE(sqe->len); | |
3247 | req->madvise.advice = READ_ONCE(sqe->fadvise_advice); | |
3248 | return 0; | |
3249 | #else | |
3250 | return -EOPNOTSUPP; | |
3251 | #endif | |
3252 | } | |
3253 | ||
014db007 | 3254 | static int io_madvise(struct io_kiocb *req, bool force_nonblock) |
c1ca757b JA |
3255 | { |
3256 | #if defined(CONFIG_ADVISE_SYSCALLS) && defined(CONFIG_MMU) | |
3257 | struct io_madvise *ma = &req->madvise; | |
3258 | int ret; | |
3259 | ||
3260 | if (force_nonblock) | |
3261 | return -EAGAIN; | |
3262 | ||
3263 | ret = do_madvise(ma->addr, ma->len, ma->advice); | |
3264 | if (ret < 0) | |
3265 | req_set_fail_links(req); | |
3266 | io_cqring_add_event(req, ret); | |
014db007 | 3267 | io_put_req(req); |
c1ca757b JA |
3268 | return 0; |
3269 | #else | |
3270 | return -EOPNOTSUPP; | |
3271 | #endif | |
3272 | } | |
3273 | ||
4840e418 JA |
3274 | static int io_fadvise_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
3275 | { | |
3276 | if (sqe->ioprio || sqe->buf_index || sqe->addr) | |
3277 | return -EINVAL; | |
3278 | ||
3279 | req->fadvise.offset = READ_ONCE(sqe->off); | |
3280 | req->fadvise.len = READ_ONCE(sqe->len); | |
3281 | req->fadvise.advice = READ_ONCE(sqe->fadvise_advice); | |
3282 | return 0; | |
3283 | } | |
3284 | ||
014db007 | 3285 | static int io_fadvise(struct io_kiocb *req, bool force_nonblock) |
4840e418 JA |
3286 | { |
3287 | struct io_fadvise *fa = &req->fadvise; | |
3288 | int ret; | |
3289 | ||
3e69426d JA |
3290 | if (force_nonblock) { |
3291 | switch (fa->advice) { | |
3292 | case POSIX_FADV_NORMAL: | |
3293 | case POSIX_FADV_RANDOM: | |
3294 | case POSIX_FADV_SEQUENTIAL: | |
3295 | break; | |
3296 | default: | |
3297 | return -EAGAIN; | |
3298 | } | |
3299 | } | |
4840e418 JA |
3300 | |
3301 | ret = vfs_fadvise(req->file, fa->offset, fa->len, fa->advice); | |
3302 | if (ret < 0) | |
3303 | req_set_fail_links(req); | |
3304 | io_cqring_add_event(req, ret); | |
014db007 | 3305 | io_put_req(req); |
4840e418 JA |
3306 | return 0; |
3307 | } | |
3308 | ||
eddc7ef5 JA |
3309 | static int io_statx_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
3310 | { | |
f8748881 | 3311 | const char __user *fname; |
eddc7ef5 JA |
3312 | unsigned lookup_flags; |
3313 | int ret; | |
3314 | ||
3315 | if (sqe->ioprio || sqe->buf_index) | |
3316 | return -EINVAL; | |
9c280f90 | 3317 | if (req->flags & REQ_F_FIXED_FILE) |
cf3040ca | 3318 | return -EBADF; |
0bdbdd08 PB |
3319 | if (req->flags & REQ_F_NEED_CLEANUP) |
3320 | return 0; | |
eddc7ef5 JA |
3321 | |
3322 | req->open.dfd = READ_ONCE(sqe->fd); | |
3323 | req->open.mask = READ_ONCE(sqe->len); | |
f8748881 | 3324 | fname = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
eddc7ef5 | 3325 | req->open.buffer = u64_to_user_ptr(READ_ONCE(sqe->addr2)); |
c12cedf2 | 3326 | req->open.how.flags = READ_ONCE(sqe->statx_flags); |
eddc7ef5 | 3327 | |
c12cedf2 | 3328 | if (vfs_stat_set_lookup_flags(&lookup_flags, req->open.how.flags)) |
eddc7ef5 JA |
3329 | return -EINVAL; |
3330 | ||
f8748881 | 3331 | req->open.filename = getname_flags(fname, lookup_flags, NULL); |
eddc7ef5 JA |
3332 | if (IS_ERR(req->open.filename)) { |
3333 | ret = PTR_ERR(req->open.filename); | |
3334 | req->open.filename = NULL; | |
3335 | return ret; | |
3336 | } | |
3337 | ||
8fef80bf | 3338 | req->flags |= REQ_F_NEED_CLEANUP; |
eddc7ef5 JA |
3339 | return 0; |
3340 | } | |
3341 | ||
014db007 | 3342 | static int io_statx(struct io_kiocb *req, bool force_nonblock) |
eddc7ef5 JA |
3343 | { |
3344 | struct io_open *ctx = &req->open; | |
3345 | unsigned lookup_flags; | |
3346 | struct path path; | |
3347 | struct kstat stat; | |
3348 | int ret; | |
3349 | ||
5b0bbee4 JA |
3350 | if (force_nonblock) { |
3351 | /* only need file table for an actual valid fd */ | |
3352 | if (ctx->dfd == -1 || ctx->dfd == AT_FDCWD) | |
3353 | req->flags |= REQ_F_NO_FILE_TABLE; | |
eddc7ef5 | 3354 | return -EAGAIN; |
5b0bbee4 | 3355 | } |
eddc7ef5 | 3356 | |
c12cedf2 | 3357 | if (vfs_stat_set_lookup_flags(&lookup_flags, ctx->how.flags)) |
eddc7ef5 JA |
3358 | return -EINVAL; |
3359 | ||
3360 | retry: | |
3361 | /* filename_lookup() drops it, keep a reference */ | |
3362 | ctx->filename->refcnt++; | |
3363 | ||
3364 | ret = filename_lookup(ctx->dfd, ctx->filename, lookup_flags, &path, | |
3365 | NULL); | |
3366 | if (ret) | |
3367 | goto err; | |
3368 | ||
c12cedf2 | 3369 | ret = vfs_getattr(&path, &stat, ctx->mask, ctx->how.flags); |
eddc7ef5 JA |
3370 | path_put(&path); |
3371 | if (retry_estale(ret, lookup_flags)) { | |
3372 | lookup_flags |= LOOKUP_REVAL; | |
3373 | goto retry; | |
3374 | } | |
3375 | if (!ret) | |
3376 | ret = cp_statx(&stat, ctx->buffer); | |
3377 | err: | |
3378 | putname(ctx->filename); | |
8fef80bf | 3379 | req->flags &= ~REQ_F_NEED_CLEANUP; |
eddc7ef5 JA |
3380 | if (ret < 0) |
3381 | req_set_fail_links(req); | |
3382 | io_cqring_add_event(req, ret); | |
014db007 | 3383 | io_put_req(req); |
eddc7ef5 JA |
3384 | return 0; |
3385 | } | |
3386 | ||
b5dba59e JA |
3387 | static int io_close_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
3388 | { | |
3389 | /* | |
3390 | * If we queue this for async, it must not be cancellable. That would | |
3391 | * leave the 'file' in an undeterminate state. | |
3392 | */ | |
3393 | req->work.flags |= IO_WQ_WORK_NO_CANCEL; | |
3394 | ||
3395 | if (sqe->ioprio || sqe->off || sqe->addr || sqe->len || | |
3396 | sqe->rw_flags || sqe->buf_index) | |
3397 | return -EINVAL; | |
9c280f90 | 3398 | if (req->flags & REQ_F_FIXED_FILE) |
cf3040ca | 3399 | return -EBADF; |
b5dba59e JA |
3400 | |
3401 | req->close.fd = READ_ONCE(sqe->fd); | |
3402 | if (req->file->f_op == &io_uring_fops || | |
b14cca0c | 3403 | req->close.fd == req->ctx->ring_fd) |
b5dba59e JA |
3404 | return -EBADF; |
3405 | ||
3406 | return 0; | |
3407 | } | |
3408 | ||
a93b3331 | 3409 | /* only called when __close_fd_get_file() is done */ |
014db007 | 3410 | static void __io_close_finish(struct io_kiocb *req) |
a93b3331 PB |
3411 | { |
3412 | int ret; | |
3413 | ||
3414 | ret = filp_close(req->close.put_file, req->work.files); | |
3415 | if (ret < 0) | |
3416 | req_set_fail_links(req); | |
3417 | io_cqring_add_event(req, ret); | |
3418 | fput(req->close.put_file); | |
014db007 | 3419 | io_put_req(req); |
a93b3331 PB |
3420 | } |
3421 | ||
b5dba59e JA |
3422 | static void io_close_finish(struct io_wq_work **workptr) |
3423 | { | |
3424 | struct io_kiocb *req = container_of(*workptr, struct io_kiocb, work); | |
b5dba59e | 3425 | |
7fbeb95d | 3426 | /* not cancellable, don't do io_req_cancelled() */ |
014db007 | 3427 | __io_close_finish(req); |
e9fd9396 | 3428 | io_steal_work(req, workptr); |
b5dba59e JA |
3429 | } |
3430 | ||
014db007 | 3431 | static int io_close(struct io_kiocb *req, bool force_nonblock) |
b5dba59e JA |
3432 | { |
3433 | int ret; | |
3434 | ||
3435 | req->close.put_file = NULL; | |
3436 | ret = __close_fd_get_file(req->close.fd, &req->close.put_file); | |
3437 | if (ret < 0) | |
3438 | return ret; | |
3439 | ||
3440 | /* if the file has a flush method, be safe and punt to async */ | |
a2100672 | 3441 | if (req->close.put_file->f_op->flush && force_nonblock) { |
594506fe PB |
3442 | /* submission ref will be dropped, take it for async */ |
3443 | refcount_inc(&req->refs); | |
3444 | ||
a2100672 PB |
3445 | req->work.func = io_close_finish; |
3446 | /* | |
3447 | * Do manual async queue here to avoid grabbing files - we don't | |
3448 | * need the files, and it'll cause io_close_finish() to close | |
3449 | * the file again and cause a double CQE entry for this request | |
3450 | */ | |
3451 | io_queue_async_work(req); | |
3452 | return 0; | |
3453 | } | |
b5dba59e JA |
3454 | |
3455 | /* | |
3456 | * No ->flush(), safely close from here and just punt the | |
3457 | * fput() to async context. | |
3458 | */ | |
014db007 | 3459 | __io_close_finish(req); |
1a417f4e | 3460 | return 0; |
b5dba59e JA |
3461 | } |
3462 | ||
3529d8c2 | 3463 | static int io_prep_sfr(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
5d17b4a4 JA |
3464 | { |
3465 | struct io_ring_ctx *ctx = req->ctx; | |
5d17b4a4 JA |
3466 | |
3467 | if (!req->file) | |
3468 | return -EBADF; | |
5d17b4a4 JA |
3469 | |
3470 | if (unlikely(ctx->flags & IORING_SETUP_IOPOLL)) | |
3471 | return -EINVAL; | |
3472 | if (unlikely(sqe->addr || sqe->ioprio || sqe->buf_index)) | |
3473 | return -EINVAL; | |
3474 | ||
8ed8d3c3 JA |
3475 | req->sync.off = READ_ONCE(sqe->off); |
3476 | req->sync.len = READ_ONCE(sqe->len); | |
3477 | req->sync.flags = READ_ONCE(sqe->sync_range_flags); | |
8ed8d3c3 JA |
3478 | return 0; |
3479 | } | |
3480 | ||
014db007 | 3481 | static void __io_sync_file_range(struct io_kiocb *req) |
8ed8d3c3 | 3482 | { |
8ed8d3c3 JA |
3483 | int ret; |
3484 | ||
9adbd45d | 3485 | ret = sync_file_range(req->file, req->sync.off, req->sync.len, |
8ed8d3c3 JA |
3486 | req->sync.flags); |
3487 | if (ret < 0) | |
3488 | req_set_fail_links(req); | |
3489 | io_cqring_add_event(req, ret); | |
014db007 | 3490 | io_put_req(req); |
5ea62161 PB |
3491 | } |
3492 | ||
3493 | ||
3494 | static void io_sync_file_range_finish(struct io_wq_work **workptr) | |
3495 | { | |
3496 | struct io_kiocb *req = container_of(*workptr, struct io_kiocb, work); | |
5ea62161 PB |
3497 | |
3498 | if (io_req_cancelled(req)) | |
3499 | return; | |
014db007 | 3500 | __io_sync_file_range(req); |
7759a0bf | 3501 | io_steal_work(req, workptr); |
5d17b4a4 JA |
3502 | } |
3503 | ||
014db007 | 3504 | static int io_sync_file_range(struct io_kiocb *req, bool force_nonblock) |
5d17b4a4 | 3505 | { |
5d17b4a4 | 3506 | /* sync_file_range always requires a blocking context */ |
8ed8d3c3 | 3507 | if (force_nonblock) { |
8ed8d3c3 | 3508 | req->work.func = io_sync_file_range_finish; |
5d17b4a4 | 3509 | return -EAGAIN; |
8ed8d3c3 | 3510 | } |
5d17b4a4 | 3511 | |
014db007 | 3512 | __io_sync_file_range(req); |
5d17b4a4 JA |
3513 | return 0; |
3514 | } | |
3515 | ||
469956e8 | 3516 | #if defined(CONFIG_NET) |
02d27d89 PB |
3517 | static int io_setup_async_msg(struct io_kiocb *req, |
3518 | struct io_async_msghdr *kmsg) | |
3519 | { | |
3520 | if (req->io) | |
3521 | return -EAGAIN; | |
3522 | if (io_alloc_async_ctx(req)) { | |
3523 | if (kmsg->iov != kmsg->fast_iov) | |
3524 | kfree(kmsg->iov); | |
3525 | return -ENOMEM; | |
3526 | } | |
3527 | req->flags |= REQ_F_NEED_CLEANUP; | |
3528 | memcpy(&req->io->msg, kmsg, sizeof(*kmsg)); | |
3529 | return -EAGAIN; | |
3530 | } | |
3531 | ||
3529d8c2 | 3532 | static int io_sendmsg_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
03b1230c | 3533 | { |
e47293fd | 3534 | struct io_sr_msg *sr = &req->sr_msg; |
3529d8c2 | 3535 | struct io_async_ctx *io = req->io; |
99bc4c38 | 3536 | int ret; |
03b1230c | 3537 | |
e47293fd JA |
3538 | sr->msg_flags = READ_ONCE(sqe->msg_flags); |
3539 | sr->msg = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
fddaface | 3540 | sr->len = READ_ONCE(sqe->len); |
3529d8c2 | 3541 | |
d8768362 JA |
3542 | #ifdef CONFIG_COMPAT |
3543 | if (req->ctx->compat) | |
3544 | sr->msg_flags |= MSG_CMSG_COMPAT; | |
3545 | #endif | |
3546 | ||
fddaface | 3547 | if (!io || req->opcode == IORING_OP_SEND) |
3529d8c2 | 3548 | return 0; |
5f798bea PB |
3549 | /* iovec is already imported */ |
3550 | if (req->flags & REQ_F_NEED_CLEANUP) | |
3551 | return 0; | |
3529d8c2 | 3552 | |
d9688565 | 3553 | io->msg.iov = io->msg.fast_iov; |
99bc4c38 | 3554 | ret = sendmsg_copy_msghdr(&io->msg.msg, sr->msg, sr->msg_flags, |
e47293fd | 3555 | &io->msg.iov); |
99bc4c38 PB |
3556 | if (!ret) |
3557 | req->flags |= REQ_F_NEED_CLEANUP; | |
3558 | return ret; | |
03b1230c JA |
3559 | } |
3560 | ||
014db007 | 3561 | static int io_sendmsg(struct io_kiocb *req, bool force_nonblock) |
aa1fa28f | 3562 | { |
0b416c3e | 3563 | struct io_async_msghdr *kmsg = NULL; |
0fa03c62 JA |
3564 | struct socket *sock; |
3565 | int ret; | |
3566 | ||
3567 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) | |
3568 | return -EINVAL; | |
3569 | ||
3570 | sock = sock_from_file(req->file, &ret); | |
3571 | if (sock) { | |
b7bb4f7d | 3572 | struct io_async_ctx io; |
0fa03c62 JA |
3573 | unsigned flags; |
3574 | ||
03b1230c | 3575 | if (req->io) { |
0b416c3e | 3576 | kmsg = &req->io->msg; |
b537916c | 3577 | kmsg->msg.msg_name = &req->io->msg.addr; |
0b416c3e JA |
3578 | /* if iov is set, it's allocated already */ |
3579 | if (!kmsg->iov) | |
3580 | kmsg->iov = kmsg->fast_iov; | |
3581 | kmsg->msg.msg_iter.iov = kmsg->iov; | |
03b1230c | 3582 | } else { |
3529d8c2 JA |
3583 | struct io_sr_msg *sr = &req->sr_msg; |
3584 | ||
0b416c3e | 3585 | kmsg = &io.msg; |
b537916c | 3586 | kmsg->msg.msg_name = &io.msg.addr; |
3529d8c2 JA |
3587 | |
3588 | io.msg.iov = io.msg.fast_iov; | |
3589 | ret = sendmsg_copy_msghdr(&io.msg.msg, sr->msg, | |
3590 | sr->msg_flags, &io.msg.iov); | |
03b1230c | 3591 | if (ret) |
3529d8c2 | 3592 | return ret; |
03b1230c | 3593 | } |
0fa03c62 | 3594 | |
e47293fd JA |
3595 | flags = req->sr_msg.msg_flags; |
3596 | if (flags & MSG_DONTWAIT) | |
3597 | req->flags |= REQ_F_NOWAIT; | |
3598 | else if (force_nonblock) | |
3599 | flags |= MSG_DONTWAIT; | |
3600 | ||
0b416c3e | 3601 | ret = __sys_sendmsg_sock(sock, &kmsg->msg, flags); |
02d27d89 PB |
3602 | if (force_nonblock && ret == -EAGAIN) |
3603 | return io_setup_async_msg(req, kmsg); | |
441cdbd5 JA |
3604 | if (ret == -ERESTARTSYS) |
3605 | ret = -EINTR; | |
0fa03c62 JA |
3606 | } |
3607 | ||
1e95081c | 3608 | if (kmsg && kmsg->iov != kmsg->fast_iov) |
0b416c3e | 3609 | kfree(kmsg->iov); |
99bc4c38 | 3610 | req->flags &= ~REQ_F_NEED_CLEANUP; |
78e19bbe | 3611 | io_cqring_add_event(req, ret); |
4e88d6e7 JA |
3612 | if (ret < 0) |
3613 | req_set_fail_links(req); | |
014db007 | 3614 | io_put_req(req); |
5d17b4a4 | 3615 | return 0; |
03b1230c | 3616 | } |
aa1fa28f | 3617 | |
014db007 | 3618 | static int io_send(struct io_kiocb *req, bool force_nonblock) |
fddaface | 3619 | { |
fddaface JA |
3620 | struct socket *sock; |
3621 | int ret; | |
3622 | ||
3623 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) | |
3624 | return -EINVAL; | |
3625 | ||
3626 | sock = sock_from_file(req->file, &ret); | |
3627 | if (sock) { | |
3628 | struct io_sr_msg *sr = &req->sr_msg; | |
3629 | struct msghdr msg; | |
3630 | struct iovec iov; | |
3631 | unsigned flags; | |
3632 | ||
3633 | ret = import_single_range(WRITE, sr->buf, sr->len, &iov, | |
3634 | &msg.msg_iter); | |
3635 | if (ret) | |
3636 | return ret; | |
3637 | ||
3638 | msg.msg_name = NULL; | |
3639 | msg.msg_control = NULL; | |
3640 | msg.msg_controllen = 0; | |
3641 | msg.msg_namelen = 0; | |
3642 | ||
3643 | flags = req->sr_msg.msg_flags; | |
3644 | if (flags & MSG_DONTWAIT) | |
3645 | req->flags |= REQ_F_NOWAIT; | |
3646 | else if (force_nonblock) | |
3647 | flags |= MSG_DONTWAIT; | |
3648 | ||
0b7b21e4 JA |
3649 | msg.msg_flags = flags; |
3650 | ret = sock_sendmsg(sock, &msg); | |
fddaface JA |
3651 | if (force_nonblock && ret == -EAGAIN) |
3652 | return -EAGAIN; | |
3653 | if (ret == -ERESTARTSYS) | |
3654 | ret = -EINTR; | |
3655 | } | |
3656 | ||
3657 | io_cqring_add_event(req, ret); | |
3658 | if (ret < 0) | |
3659 | req_set_fail_links(req); | |
014db007 | 3660 | io_put_req(req); |
fddaface | 3661 | return 0; |
fddaface JA |
3662 | } |
3663 | ||
52de1fe1 JA |
3664 | static int __io_recvmsg_copy_hdr(struct io_kiocb *req, struct io_async_ctx *io) |
3665 | { | |
3666 | struct io_sr_msg *sr = &req->sr_msg; | |
3667 | struct iovec __user *uiov; | |
3668 | size_t iov_len; | |
3669 | int ret; | |
3670 | ||
3671 | ret = __copy_msghdr_from_user(&io->msg.msg, sr->msg, &io->msg.uaddr, | |
3672 | &uiov, &iov_len); | |
3673 | if (ret) | |
3674 | return ret; | |
3675 | ||
3676 | if (req->flags & REQ_F_BUFFER_SELECT) { | |
3677 | if (iov_len > 1) | |
3678 | return -EINVAL; | |
3679 | if (copy_from_user(io->msg.iov, uiov, sizeof(*uiov))) | |
3680 | return -EFAULT; | |
3681 | sr->len = io->msg.iov[0].iov_len; | |
3682 | iov_iter_init(&io->msg.msg.msg_iter, READ, io->msg.iov, 1, | |
3683 | sr->len); | |
3684 | io->msg.iov = NULL; | |
3685 | } else { | |
3686 | ret = import_iovec(READ, uiov, iov_len, UIO_FASTIOV, | |
3687 | &io->msg.iov, &io->msg.msg.msg_iter); | |
3688 | if (ret > 0) | |
3689 | ret = 0; | |
3690 | } | |
3691 | ||
3692 | return ret; | |
3693 | } | |
3694 | ||
3695 | #ifdef CONFIG_COMPAT | |
3696 | static int __io_compat_recvmsg_copy_hdr(struct io_kiocb *req, | |
3697 | struct io_async_ctx *io) | |
3698 | { | |
3699 | struct compat_msghdr __user *msg_compat; | |
3700 | struct io_sr_msg *sr = &req->sr_msg; | |
3701 | struct compat_iovec __user *uiov; | |
3702 | compat_uptr_t ptr; | |
3703 | compat_size_t len; | |
3704 | int ret; | |
3705 | ||
3706 | msg_compat = (struct compat_msghdr __user *) sr->msg; | |
3707 | ret = __get_compat_msghdr(&io->msg.msg, msg_compat, &io->msg.uaddr, | |
3708 | &ptr, &len); | |
3709 | if (ret) | |
3710 | return ret; | |
3711 | ||
3712 | uiov = compat_ptr(ptr); | |
3713 | if (req->flags & REQ_F_BUFFER_SELECT) { | |
3714 | compat_ssize_t clen; | |
3715 | ||
3716 | if (len > 1) | |
3717 | return -EINVAL; | |
3718 | if (!access_ok(uiov, sizeof(*uiov))) | |
3719 | return -EFAULT; | |
3720 | if (__get_user(clen, &uiov->iov_len)) | |
3721 | return -EFAULT; | |
3722 | if (clen < 0) | |
3723 | return -EINVAL; | |
3724 | sr->len = io->msg.iov[0].iov_len; | |
3725 | io->msg.iov = NULL; | |
3726 | } else { | |
3727 | ret = compat_import_iovec(READ, uiov, len, UIO_FASTIOV, | |
3728 | &io->msg.iov, | |
3729 | &io->msg.msg.msg_iter); | |
3730 | if (ret < 0) | |
3731 | return ret; | |
3732 | } | |
3733 | ||
3734 | return 0; | |
3735 | } | |
3736 | #endif | |
3737 | ||
3738 | static int io_recvmsg_copy_hdr(struct io_kiocb *req, struct io_async_ctx *io) | |
3739 | { | |
3740 | io->msg.iov = io->msg.fast_iov; | |
3741 | ||
3742 | #ifdef CONFIG_COMPAT | |
3743 | if (req->ctx->compat) | |
3744 | return __io_compat_recvmsg_copy_hdr(req, io); | |
fddaface | 3745 | #endif |
52de1fe1 JA |
3746 | |
3747 | return __io_recvmsg_copy_hdr(req, io); | |
3748 | } | |
3749 | ||
bcda7baa JA |
3750 | static struct io_buffer *io_recv_buffer_select(struct io_kiocb *req, |
3751 | int *cflags, bool needs_lock) | |
3752 | { | |
3753 | struct io_sr_msg *sr = &req->sr_msg; | |
3754 | struct io_buffer *kbuf; | |
3755 | ||
3756 | if (!(req->flags & REQ_F_BUFFER_SELECT)) | |
3757 | return NULL; | |
3758 | ||
3759 | kbuf = io_buffer_select(req, &sr->len, sr->bgid, sr->kbuf, needs_lock); | |
3760 | if (IS_ERR(kbuf)) | |
3761 | return kbuf; | |
3762 | ||
3763 | sr->kbuf = kbuf; | |
3764 | req->flags |= REQ_F_BUFFER_SELECTED; | |
3765 | ||
3766 | *cflags = kbuf->bid << IORING_CQE_BUFFER_SHIFT; | |
3767 | *cflags |= IORING_CQE_F_BUFFER; | |
3768 | return kbuf; | |
fddaface JA |
3769 | } |
3770 | ||
3529d8c2 JA |
3771 | static int io_recvmsg_prep(struct io_kiocb *req, |
3772 | const struct io_uring_sqe *sqe) | |
aa1fa28f | 3773 | { |
e47293fd | 3774 | struct io_sr_msg *sr = &req->sr_msg; |
3529d8c2 | 3775 | struct io_async_ctx *io = req->io; |
99bc4c38 | 3776 | int ret; |
3529d8c2 JA |
3777 | |
3778 | sr->msg_flags = READ_ONCE(sqe->msg_flags); | |
3779 | sr->msg = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
0b7b21e4 | 3780 | sr->len = READ_ONCE(sqe->len); |
bcda7baa | 3781 | sr->bgid = READ_ONCE(sqe->buf_group); |
06b76d44 | 3782 | |
d8768362 JA |
3783 | #ifdef CONFIG_COMPAT |
3784 | if (req->ctx->compat) | |
3785 | sr->msg_flags |= MSG_CMSG_COMPAT; | |
3786 | #endif | |
3787 | ||
fddaface | 3788 | if (!io || req->opcode == IORING_OP_RECV) |
06b76d44 | 3789 | return 0; |
5f798bea PB |
3790 | /* iovec is already imported */ |
3791 | if (req->flags & REQ_F_NEED_CLEANUP) | |
3792 | return 0; | |
03b1230c | 3793 | |
52de1fe1 | 3794 | ret = io_recvmsg_copy_hdr(req, io); |
99bc4c38 PB |
3795 | if (!ret) |
3796 | req->flags |= REQ_F_NEED_CLEANUP; | |
3797 | return ret; | |
aa1fa28f JA |
3798 | } |
3799 | ||
014db007 | 3800 | static int io_recvmsg(struct io_kiocb *req, bool force_nonblock) |
aa1fa28f | 3801 | { |
0b416c3e | 3802 | struct io_async_msghdr *kmsg = NULL; |
03b1230c | 3803 | struct socket *sock; |
52de1fe1 | 3804 | int ret, cflags = 0; |
03b1230c JA |
3805 | |
3806 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) | |
3807 | return -EINVAL; | |
3808 | ||
3809 | sock = sock_from_file(req->file, &ret); | |
3810 | if (sock) { | |
52de1fe1 | 3811 | struct io_buffer *kbuf; |
b7bb4f7d | 3812 | struct io_async_ctx io; |
03b1230c JA |
3813 | unsigned flags; |
3814 | ||
03b1230c | 3815 | if (req->io) { |
0b416c3e | 3816 | kmsg = &req->io->msg; |
b537916c | 3817 | kmsg->msg.msg_name = &req->io->msg.addr; |
0b416c3e JA |
3818 | /* if iov is set, it's allocated already */ |
3819 | if (!kmsg->iov) | |
3820 | kmsg->iov = kmsg->fast_iov; | |
3821 | kmsg->msg.msg_iter.iov = kmsg->iov; | |
03b1230c | 3822 | } else { |
0b416c3e | 3823 | kmsg = &io.msg; |
b537916c | 3824 | kmsg->msg.msg_name = &io.msg.addr; |
3529d8c2 | 3825 | |
52de1fe1 | 3826 | ret = io_recvmsg_copy_hdr(req, &io); |
03b1230c | 3827 | if (ret) |
3529d8c2 | 3828 | return ret; |
03b1230c JA |
3829 | } |
3830 | ||
52de1fe1 JA |
3831 | kbuf = io_recv_buffer_select(req, &cflags, !force_nonblock); |
3832 | if (IS_ERR(kbuf)) { | |
3833 | return PTR_ERR(kbuf); | |
3834 | } else if (kbuf) { | |
3835 | kmsg->fast_iov[0].iov_base = u64_to_user_ptr(kbuf->addr); | |
3836 | iov_iter_init(&kmsg->msg.msg_iter, READ, kmsg->iov, | |
3837 | 1, req->sr_msg.len); | |
3838 | } | |
3839 | ||
e47293fd JA |
3840 | flags = req->sr_msg.msg_flags; |
3841 | if (flags & MSG_DONTWAIT) | |
3842 | req->flags |= REQ_F_NOWAIT; | |
3843 | else if (force_nonblock) | |
3844 | flags |= MSG_DONTWAIT; | |
3845 | ||
3846 | ret = __sys_recvmsg_sock(sock, &kmsg->msg, req->sr_msg.msg, | |
3847 | kmsg->uaddr, flags); | |
02d27d89 PB |
3848 | if (force_nonblock && ret == -EAGAIN) |
3849 | return io_setup_async_msg(req, kmsg); | |
03b1230c JA |
3850 | if (ret == -ERESTARTSYS) |
3851 | ret = -EINTR; | |
3852 | } | |
3853 | ||
1e95081c | 3854 | if (kmsg && kmsg->iov != kmsg->fast_iov) |
0b416c3e | 3855 | kfree(kmsg->iov); |
99bc4c38 | 3856 | req->flags &= ~REQ_F_NEED_CLEANUP; |
52de1fe1 | 3857 | __io_cqring_add_event(req, ret, cflags); |
4e88d6e7 JA |
3858 | if (ret < 0) |
3859 | req_set_fail_links(req); | |
014db007 | 3860 | io_put_req(req); |
03b1230c | 3861 | return 0; |
0fa03c62 | 3862 | } |
5d17b4a4 | 3863 | |
014db007 | 3864 | static int io_recv(struct io_kiocb *req, bool force_nonblock) |
fddaface | 3865 | { |
bcda7baa | 3866 | struct io_buffer *kbuf = NULL; |
fddaface | 3867 | struct socket *sock; |
bcda7baa | 3868 | int ret, cflags = 0; |
fddaface JA |
3869 | |
3870 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) | |
3871 | return -EINVAL; | |
3872 | ||
3873 | sock = sock_from_file(req->file, &ret); | |
3874 | if (sock) { | |
3875 | struct io_sr_msg *sr = &req->sr_msg; | |
bcda7baa | 3876 | void __user *buf = sr->buf; |
fddaface JA |
3877 | struct msghdr msg; |
3878 | struct iovec iov; | |
3879 | unsigned flags; | |
3880 | ||
bcda7baa JA |
3881 | kbuf = io_recv_buffer_select(req, &cflags, !force_nonblock); |
3882 | if (IS_ERR(kbuf)) | |
3883 | return PTR_ERR(kbuf); | |
3884 | else if (kbuf) | |
3885 | buf = u64_to_user_ptr(kbuf->addr); | |
3886 | ||
3887 | ret = import_single_range(READ, buf, sr->len, &iov, | |
fddaface | 3888 | &msg.msg_iter); |
bcda7baa JA |
3889 | if (ret) { |
3890 | kfree(kbuf); | |
fddaface | 3891 | return ret; |
bcda7baa | 3892 | } |
fddaface | 3893 | |
bcda7baa | 3894 | req->flags |= REQ_F_NEED_CLEANUP; |
fddaface JA |
3895 | msg.msg_name = NULL; |
3896 | msg.msg_control = NULL; | |
3897 | msg.msg_controllen = 0; | |
3898 | msg.msg_namelen = 0; | |
3899 | msg.msg_iocb = NULL; | |
3900 | msg.msg_flags = 0; | |
3901 | ||
3902 | flags = req->sr_msg.msg_flags; | |
3903 | if (flags & MSG_DONTWAIT) | |
3904 | req->flags |= REQ_F_NOWAIT; | |
3905 | else if (force_nonblock) | |
3906 | flags |= MSG_DONTWAIT; | |
3907 | ||
0b7b21e4 | 3908 | ret = sock_recvmsg(sock, &msg, flags); |
fddaface JA |
3909 | if (force_nonblock && ret == -EAGAIN) |
3910 | return -EAGAIN; | |
3911 | if (ret == -ERESTARTSYS) | |
3912 | ret = -EINTR; | |
3913 | } | |
3914 | ||
bcda7baa JA |
3915 | kfree(kbuf); |
3916 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
3917 | __io_cqring_add_event(req, ret, cflags); | |
fddaface JA |
3918 | if (ret < 0) |
3919 | req_set_fail_links(req); | |
014db007 | 3920 | io_put_req(req); |
fddaface | 3921 | return 0; |
fddaface JA |
3922 | } |
3923 | ||
3529d8c2 | 3924 | static int io_accept_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
17f2fe35 | 3925 | { |
8ed8d3c3 JA |
3926 | struct io_accept *accept = &req->accept; |
3927 | ||
17f2fe35 JA |
3928 | if (unlikely(req->ctx->flags & (IORING_SETUP_IOPOLL|IORING_SETUP_SQPOLL))) |
3929 | return -EINVAL; | |
8042d6ce | 3930 | if (sqe->ioprio || sqe->len || sqe->buf_index) |
17f2fe35 JA |
3931 | return -EINVAL; |
3932 | ||
d55e5f5b JA |
3933 | accept->addr = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
3934 | accept->addr_len = u64_to_user_ptr(READ_ONCE(sqe->addr2)); | |
8ed8d3c3 | 3935 | accept->flags = READ_ONCE(sqe->accept_flags); |
09952e3e | 3936 | accept->nofile = rlimit(RLIMIT_NOFILE); |
8ed8d3c3 | 3937 | return 0; |
8ed8d3c3 | 3938 | } |
17f2fe35 | 3939 | |
014db007 | 3940 | static int __io_accept(struct io_kiocb *req, bool force_nonblock) |
8ed8d3c3 JA |
3941 | { |
3942 | struct io_accept *accept = &req->accept; | |
3943 | unsigned file_flags; | |
3944 | int ret; | |
3945 | ||
3946 | file_flags = force_nonblock ? O_NONBLOCK : 0; | |
3947 | ret = __sys_accept4_file(req->file, file_flags, accept->addr, | |
09952e3e JA |
3948 | accept->addr_len, accept->flags, |
3949 | accept->nofile); | |
8ed8d3c3 | 3950 | if (ret == -EAGAIN && force_nonblock) |
17f2fe35 | 3951 | return -EAGAIN; |
8e3cca12 JA |
3952 | if (ret == -ERESTARTSYS) |
3953 | ret = -EINTR; | |
4e88d6e7 JA |
3954 | if (ret < 0) |
3955 | req_set_fail_links(req); | |
78e19bbe | 3956 | io_cqring_add_event(req, ret); |
014db007 | 3957 | io_put_req(req); |
17f2fe35 | 3958 | return 0; |
8ed8d3c3 JA |
3959 | } |
3960 | ||
3961 | static void io_accept_finish(struct io_wq_work **workptr) | |
3962 | { | |
3963 | struct io_kiocb *req = container_of(*workptr, struct io_kiocb, work); | |
8ed8d3c3 JA |
3964 | |
3965 | if (io_req_cancelled(req)) | |
3966 | return; | |
014db007 | 3967 | __io_accept(req, false); |
e9fd9396 | 3968 | io_steal_work(req, workptr); |
8ed8d3c3 | 3969 | } |
8ed8d3c3 | 3970 | |
014db007 | 3971 | static int io_accept(struct io_kiocb *req, bool force_nonblock) |
8ed8d3c3 | 3972 | { |
8ed8d3c3 JA |
3973 | int ret; |
3974 | ||
014db007 | 3975 | ret = __io_accept(req, force_nonblock); |
8ed8d3c3 JA |
3976 | if (ret == -EAGAIN && force_nonblock) { |
3977 | req->work.func = io_accept_finish; | |
8ed8d3c3 JA |
3978 | return -EAGAIN; |
3979 | } | |
3980 | return 0; | |
0fa03c62 | 3981 | } |
5d17b4a4 | 3982 | |
3529d8c2 | 3983 | static int io_connect_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
f499a021 | 3984 | { |
3529d8c2 JA |
3985 | struct io_connect *conn = &req->connect; |
3986 | struct io_async_ctx *io = req->io; | |
f499a021 | 3987 | |
3fbb51c1 JA |
3988 | if (unlikely(req->ctx->flags & (IORING_SETUP_IOPOLL|IORING_SETUP_SQPOLL))) |
3989 | return -EINVAL; | |
3990 | if (sqe->ioprio || sqe->len || sqe->buf_index || sqe->rw_flags) | |
3991 | return -EINVAL; | |
3992 | ||
3529d8c2 JA |
3993 | conn->addr = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
3994 | conn->addr_len = READ_ONCE(sqe->addr2); | |
3995 | ||
3996 | if (!io) | |
3997 | return 0; | |
3998 | ||
3999 | return move_addr_to_kernel(conn->addr, conn->addr_len, | |
3fbb51c1 | 4000 | &io->connect.address); |
f499a021 JA |
4001 | } |
4002 | ||
014db007 | 4003 | static int io_connect(struct io_kiocb *req, bool force_nonblock) |
f8e85cf2 | 4004 | { |
f499a021 | 4005 | struct io_async_ctx __io, *io; |
f8e85cf2 | 4006 | unsigned file_flags; |
3fbb51c1 | 4007 | int ret; |
f8e85cf2 | 4008 | |
f499a021 JA |
4009 | if (req->io) { |
4010 | io = req->io; | |
4011 | } else { | |
3529d8c2 JA |
4012 | ret = move_addr_to_kernel(req->connect.addr, |
4013 | req->connect.addr_len, | |
4014 | &__io.connect.address); | |
f499a021 JA |
4015 | if (ret) |
4016 | goto out; | |
4017 | io = &__io; | |
4018 | } | |
4019 | ||
3fbb51c1 JA |
4020 | file_flags = force_nonblock ? O_NONBLOCK : 0; |
4021 | ||
4022 | ret = __sys_connect_file(req->file, &io->connect.address, | |
4023 | req->connect.addr_len, file_flags); | |
87f80d62 | 4024 | if ((ret == -EAGAIN || ret == -EINPROGRESS) && force_nonblock) { |
b7bb4f7d JA |
4025 | if (req->io) |
4026 | return -EAGAIN; | |
4027 | if (io_alloc_async_ctx(req)) { | |
f499a021 JA |
4028 | ret = -ENOMEM; |
4029 | goto out; | |
4030 | } | |
b7bb4f7d | 4031 | memcpy(&req->io->connect, &__io.connect, sizeof(__io.connect)); |
f8e85cf2 | 4032 | return -EAGAIN; |
f499a021 | 4033 | } |
f8e85cf2 JA |
4034 | if (ret == -ERESTARTSYS) |
4035 | ret = -EINTR; | |
f499a021 | 4036 | out: |
4e88d6e7 JA |
4037 | if (ret < 0) |
4038 | req_set_fail_links(req); | |
f8e85cf2 | 4039 | io_cqring_add_event(req, ret); |
014db007 | 4040 | io_put_req(req); |
f8e85cf2 | 4041 | return 0; |
469956e8 Y |
4042 | } |
4043 | #else /* !CONFIG_NET */ | |
4044 | static int io_sendmsg_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) | |
4045 | { | |
f8e85cf2 | 4046 | return -EOPNOTSUPP; |
f8e85cf2 JA |
4047 | } |
4048 | ||
469956e8 Y |
4049 | static int io_sendmsg(struct io_kiocb *req, bool force_nonblock) |
4050 | { | |
4051 | return -EOPNOTSUPP; | |
4052 | } | |
4053 | ||
4054 | static int io_send(struct io_kiocb *req, bool force_nonblock) | |
4055 | { | |
4056 | return -EOPNOTSUPP; | |
4057 | } | |
4058 | ||
4059 | static int io_recvmsg_prep(struct io_kiocb *req, | |
4060 | const struct io_uring_sqe *sqe) | |
4061 | { | |
4062 | return -EOPNOTSUPP; | |
4063 | } | |
4064 | ||
4065 | static int io_recvmsg(struct io_kiocb *req, bool force_nonblock) | |
4066 | { | |
4067 | return -EOPNOTSUPP; | |
4068 | } | |
4069 | ||
4070 | static int io_recv(struct io_kiocb *req, bool force_nonblock) | |
4071 | { | |
4072 | return -EOPNOTSUPP; | |
4073 | } | |
4074 | ||
4075 | static int io_accept_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) | |
4076 | { | |
4077 | return -EOPNOTSUPP; | |
4078 | } | |
4079 | ||
4080 | static int io_accept(struct io_kiocb *req, bool force_nonblock) | |
4081 | { | |
4082 | return -EOPNOTSUPP; | |
4083 | } | |
4084 | ||
4085 | static int io_connect_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) | |
4086 | { | |
4087 | return -EOPNOTSUPP; | |
4088 | } | |
4089 | ||
4090 | static int io_connect(struct io_kiocb *req, bool force_nonblock) | |
4091 | { | |
f8e85cf2 | 4092 | return -EOPNOTSUPP; |
f8e85cf2 | 4093 | } |
469956e8 | 4094 | #endif /* CONFIG_NET */ |
f8e85cf2 | 4095 | |
d7718a9d JA |
4096 | struct io_poll_table { |
4097 | struct poll_table_struct pt; | |
4098 | struct io_kiocb *req; | |
4099 | int error; | |
4100 | }; | |
4101 | ||
4102 | static void __io_queue_proc(struct io_poll_iocb *poll, struct io_poll_table *pt, | |
4103 | struct wait_queue_head *head) | |
4104 | { | |
4105 | if (unlikely(poll->head)) { | |
4106 | pt->error = -EINVAL; | |
4107 | return; | |
4108 | } | |
4109 | ||
4110 | pt->error = 0; | |
4111 | poll->head = head; | |
4112 | add_wait_queue(head, &poll->wait); | |
4113 | } | |
4114 | ||
4115 | static void io_async_queue_proc(struct file *file, struct wait_queue_head *head, | |
4116 | struct poll_table_struct *p) | |
4117 | { | |
4118 | struct io_poll_table *pt = container_of(p, struct io_poll_table, pt); | |
4119 | ||
4120 | __io_queue_proc(&pt->req->apoll->poll, pt, head); | |
4121 | } | |
4122 | ||
4123 | static int __io_async_wake(struct io_kiocb *req, struct io_poll_iocb *poll, | |
4124 | __poll_t mask, task_work_func_t func) | |
4125 | { | |
4126 | struct task_struct *tsk; | |
aa96bf8a | 4127 | int ret; |
d7718a9d JA |
4128 | |
4129 | /* for instances that support it check for an event match first: */ | |
4130 | if (mask && !(mask & poll->events)) | |
4131 | return 0; | |
4132 | ||
4133 | trace_io_uring_task_add(req->ctx, req->opcode, req->user_data, mask); | |
4134 | ||
4135 | list_del_init(&poll->wait.entry); | |
4136 | ||
4137 | tsk = req->task; | |
4138 | req->result = mask; | |
4139 | init_task_work(&req->task_work, func); | |
4140 | /* | |
aa96bf8a JA |
4141 | * If this fails, then the task is exiting. Punt to one of the io-wq |
4142 | * threads to ensure the work gets run, we can't always rely on exit | |
4143 | * cancelation taking care of this. | |
d7718a9d | 4144 | */ |
aa96bf8a JA |
4145 | ret = task_work_add(tsk, &req->task_work, true); |
4146 | if (unlikely(ret)) { | |
4147 | tsk = io_wq_get_task(req->ctx->io_wq); | |
4148 | task_work_add(tsk, &req->task_work, true); | |
4149 | } | |
d7718a9d JA |
4150 | wake_up_process(tsk); |
4151 | return 1; | |
4152 | } | |
4153 | ||
74ce6ce4 JA |
4154 | static bool io_poll_rewait(struct io_kiocb *req, struct io_poll_iocb *poll) |
4155 | __acquires(&req->ctx->completion_lock) | |
4156 | { | |
4157 | struct io_ring_ctx *ctx = req->ctx; | |
4158 | ||
4159 | if (!req->result && !READ_ONCE(poll->canceled)) { | |
4160 | struct poll_table_struct pt = { ._key = poll->events }; | |
4161 | ||
4162 | req->result = vfs_poll(req->file, &pt) & poll->events; | |
4163 | } | |
4164 | ||
4165 | spin_lock_irq(&ctx->completion_lock); | |
4166 | if (!req->result && !READ_ONCE(poll->canceled)) { | |
4167 | add_wait_queue(poll->head, &poll->wait); | |
4168 | return true; | |
4169 | } | |
4170 | ||
4171 | return false; | |
4172 | } | |
4173 | ||
d7718a9d JA |
4174 | static void io_async_task_func(struct callback_head *cb) |
4175 | { | |
4176 | struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work); | |
4177 | struct async_poll *apoll = req->apoll; | |
4178 | struct io_ring_ctx *ctx = req->ctx; | |
2bae047e | 4179 | bool canceled; |
d7718a9d JA |
4180 | |
4181 | trace_io_uring_task_run(req->ctx, req->opcode, req->user_data); | |
4182 | ||
74ce6ce4 | 4183 | if (io_poll_rewait(req, &apoll->poll)) { |
d7718a9d | 4184 | spin_unlock_irq(&ctx->completion_lock); |
74ce6ce4 | 4185 | return; |
d7718a9d JA |
4186 | } |
4187 | ||
74ce6ce4 JA |
4188 | if (hash_hashed(&req->hash_node)) |
4189 | hash_del(&req->hash_node); | |
4190 | ||
2bae047e JA |
4191 | canceled = READ_ONCE(apoll->poll.canceled); |
4192 | if (canceled) { | |
4193 | io_cqring_fill_event(req, -ECANCELED); | |
4194 | io_commit_cqring(ctx); | |
4195 | } | |
4196 | ||
74ce6ce4 JA |
4197 | spin_unlock_irq(&ctx->completion_lock); |
4198 | ||
44575a67 XW |
4199 | /* restore ->work in case we need to retry again */ |
4200 | memcpy(&req->work, &apoll->work, sizeof(req->work)); | |
4201 | ||
2bae047e JA |
4202 | if (canceled) { |
4203 | kfree(apoll); | |
4204 | io_cqring_ev_posted(ctx); | |
4205 | req_set_fail_links(req); | |
44575a67 | 4206 | io_double_put_req(req); |
2bae047e JA |
4207 | return; |
4208 | } | |
4209 | ||
d7718a9d JA |
4210 | __set_current_state(TASK_RUNNING); |
4211 | mutex_lock(&ctx->uring_lock); | |
4212 | __io_queue_sqe(req, NULL); | |
4213 | mutex_unlock(&ctx->uring_lock); | |
4214 | ||
4215 | kfree(apoll); | |
4216 | } | |
4217 | ||
4218 | static int io_async_wake(struct wait_queue_entry *wait, unsigned mode, int sync, | |
4219 | void *key) | |
4220 | { | |
4221 | struct io_kiocb *req = wait->private; | |
4222 | struct io_poll_iocb *poll = &req->apoll->poll; | |
4223 | ||
4224 | trace_io_uring_poll_wake(req->ctx, req->opcode, req->user_data, | |
4225 | key_to_poll(key)); | |
4226 | ||
4227 | return __io_async_wake(req, poll, key_to_poll(key), io_async_task_func); | |
4228 | } | |
4229 | ||
4230 | static void io_poll_req_insert(struct io_kiocb *req) | |
4231 | { | |
4232 | struct io_ring_ctx *ctx = req->ctx; | |
4233 | struct hlist_head *list; | |
4234 | ||
4235 | list = &ctx->cancel_hash[hash_long(req->user_data, ctx->cancel_hash_bits)]; | |
4236 | hlist_add_head(&req->hash_node, list); | |
4237 | } | |
4238 | ||
4239 | static __poll_t __io_arm_poll_handler(struct io_kiocb *req, | |
4240 | struct io_poll_iocb *poll, | |
4241 | struct io_poll_table *ipt, __poll_t mask, | |
4242 | wait_queue_func_t wake_func) | |
4243 | __acquires(&ctx->completion_lock) | |
4244 | { | |
4245 | struct io_ring_ctx *ctx = req->ctx; | |
4246 | bool cancel = false; | |
4247 | ||
4248 | poll->file = req->file; | |
4249 | poll->head = NULL; | |
4250 | poll->done = poll->canceled = false; | |
4251 | poll->events = mask; | |
4252 | ||
4253 | ipt->pt._key = mask; | |
4254 | ipt->req = req; | |
4255 | ipt->error = -EINVAL; | |
4256 | ||
4257 | INIT_LIST_HEAD(&poll->wait.entry); | |
4258 | init_waitqueue_func_entry(&poll->wait, wake_func); | |
4259 | poll->wait.private = req; | |
4260 | ||
4261 | mask = vfs_poll(req->file, &ipt->pt) & poll->events; | |
4262 | ||
4263 | spin_lock_irq(&ctx->completion_lock); | |
4264 | if (likely(poll->head)) { | |
4265 | spin_lock(&poll->head->lock); | |
4266 | if (unlikely(list_empty(&poll->wait.entry))) { | |
4267 | if (ipt->error) | |
4268 | cancel = true; | |
4269 | ipt->error = 0; | |
4270 | mask = 0; | |
4271 | } | |
4272 | if (mask || ipt->error) | |
4273 | list_del_init(&poll->wait.entry); | |
4274 | else if (cancel) | |
4275 | WRITE_ONCE(poll->canceled, true); | |
4276 | else if (!poll->done) /* actually waiting for an event */ | |
4277 | io_poll_req_insert(req); | |
4278 | spin_unlock(&poll->head->lock); | |
4279 | } | |
4280 | ||
4281 | return mask; | |
4282 | } | |
4283 | ||
4284 | static bool io_arm_poll_handler(struct io_kiocb *req) | |
4285 | { | |
4286 | const struct io_op_def *def = &io_op_defs[req->opcode]; | |
4287 | struct io_ring_ctx *ctx = req->ctx; | |
4288 | struct async_poll *apoll; | |
4289 | struct io_poll_table ipt; | |
4290 | __poll_t mask, ret; | |
4291 | ||
4292 | if (!req->file || !file_can_poll(req->file)) | |
4293 | return false; | |
4294 | if (req->flags & (REQ_F_MUST_PUNT | REQ_F_POLLED)) | |
4295 | return false; | |
4296 | if (!def->pollin && !def->pollout) | |
4297 | return false; | |
4298 | ||
4299 | apoll = kmalloc(sizeof(*apoll), GFP_ATOMIC); | |
4300 | if (unlikely(!apoll)) | |
4301 | return false; | |
4302 | ||
4303 | req->flags |= REQ_F_POLLED; | |
4304 | memcpy(&apoll->work, &req->work, sizeof(req->work)); | |
4305 | ||
3537b6a7 | 4306 | get_task_struct(current); |
d7718a9d JA |
4307 | req->task = current; |
4308 | req->apoll = apoll; | |
4309 | INIT_HLIST_NODE(&req->hash_node); | |
4310 | ||
8755d97a | 4311 | mask = 0; |
d7718a9d | 4312 | if (def->pollin) |
8755d97a | 4313 | mask |= POLLIN | POLLRDNORM; |
d7718a9d JA |
4314 | if (def->pollout) |
4315 | mask |= POLLOUT | POLLWRNORM; | |
4316 | mask |= POLLERR | POLLPRI; | |
4317 | ||
4318 | ipt.pt._qproc = io_async_queue_proc; | |
4319 | ||
4320 | ret = __io_arm_poll_handler(req, &apoll->poll, &ipt, mask, | |
4321 | io_async_wake); | |
4322 | if (ret) { | |
4323 | ipt.error = 0; | |
4324 | apoll->poll.done = true; | |
4325 | spin_unlock_irq(&ctx->completion_lock); | |
4326 | memcpy(&req->work, &apoll->work, sizeof(req->work)); | |
4327 | kfree(apoll); | |
4328 | return false; | |
4329 | } | |
4330 | spin_unlock_irq(&ctx->completion_lock); | |
4331 | trace_io_uring_poll_arm(ctx, req->opcode, req->user_data, mask, | |
4332 | apoll->poll.events); | |
4333 | return true; | |
4334 | } | |
4335 | ||
4336 | static bool __io_poll_remove_one(struct io_kiocb *req, | |
4337 | struct io_poll_iocb *poll) | |
221c5eb2 | 4338 | { |
b41e9852 | 4339 | bool do_complete = false; |
221c5eb2 JA |
4340 | |
4341 | spin_lock(&poll->head->lock); | |
4342 | WRITE_ONCE(poll->canceled, true); | |
392edb45 JA |
4343 | if (!list_empty(&poll->wait.entry)) { |
4344 | list_del_init(&poll->wait.entry); | |
b41e9852 | 4345 | do_complete = true; |
221c5eb2 JA |
4346 | } |
4347 | spin_unlock(&poll->head->lock); | |
d7718a9d JA |
4348 | return do_complete; |
4349 | } | |
4350 | ||
4351 | static bool io_poll_remove_one(struct io_kiocb *req) | |
4352 | { | |
b1f573bd | 4353 | struct async_poll *apoll = NULL; |
d7718a9d JA |
4354 | bool do_complete; |
4355 | ||
4356 | if (req->opcode == IORING_OP_POLL_ADD) { | |
4357 | do_complete = __io_poll_remove_one(req, &req->poll); | |
4358 | } else { | |
b1f573bd | 4359 | apoll = req->apoll; |
d7718a9d JA |
4360 | /* non-poll requests have submit ref still */ |
4361 | do_complete = __io_poll_remove_one(req, &req->apoll->poll); | |
4362 | if (do_complete) | |
4363 | io_put_req(req); | |
4364 | } | |
4365 | ||
78076bb6 | 4366 | hash_del(&req->hash_node); |
d7718a9d | 4367 | |
44575a67 | 4368 | if (do_complete && apoll) { |
b1f573bd XW |
4369 | /* |
4370 | * restore ->work because we need to call io_req_work_drop_env. | |
4371 | */ | |
4372 | memcpy(&req->work, &apoll->work, sizeof(req->work)); | |
4373 | kfree(apoll); | |
4374 | } | |
4375 | ||
b41e9852 JA |
4376 | if (do_complete) { |
4377 | io_cqring_fill_event(req, -ECANCELED); | |
4378 | io_commit_cqring(req->ctx); | |
4379 | req->flags |= REQ_F_COMP_LOCKED; | |
4380 | io_put_req(req); | |
4381 | } | |
4382 | ||
4383 | return do_complete; | |
221c5eb2 JA |
4384 | } |
4385 | ||
4386 | static void io_poll_remove_all(struct io_ring_ctx *ctx) | |
4387 | { | |
78076bb6 | 4388 | struct hlist_node *tmp; |
221c5eb2 | 4389 | struct io_kiocb *req; |
8e2e1faf | 4390 | int posted = 0, i; |
221c5eb2 JA |
4391 | |
4392 | spin_lock_irq(&ctx->completion_lock); | |
78076bb6 JA |
4393 | for (i = 0; i < (1U << ctx->cancel_hash_bits); i++) { |
4394 | struct hlist_head *list; | |
4395 | ||
4396 | list = &ctx->cancel_hash[i]; | |
4397 | hlist_for_each_entry_safe(req, tmp, list, hash_node) | |
8e2e1faf | 4398 | posted += io_poll_remove_one(req); |
221c5eb2 JA |
4399 | } |
4400 | spin_unlock_irq(&ctx->completion_lock); | |
b41e9852 | 4401 | |
8e2e1faf JA |
4402 | if (posted) |
4403 | io_cqring_ev_posted(ctx); | |
221c5eb2 JA |
4404 | } |
4405 | ||
47f46768 JA |
4406 | static int io_poll_cancel(struct io_ring_ctx *ctx, __u64 sqe_addr) |
4407 | { | |
78076bb6 | 4408 | struct hlist_head *list; |
47f46768 JA |
4409 | struct io_kiocb *req; |
4410 | ||
78076bb6 JA |
4411 | list = &ctx->cancel_hash[hash_long(sqe_addr, ctx->cancel_hash_bits)]; |
4412 | hlist_for_each_entry(req, list, hash_node) { | |
b41e9852 JA |
4413 | if (sqe_addr != req->user_data) |
4414 | continue; | |
4415 | if (io_poll_remove_one(req)) | |
eac406c6 | 4416 | return 0; |
b41e9852 | 4417 | return -EALREADY; |
47f46768 JA |
4418 | } |
4419 | ||
4420 | return -ENOENT; | |
4421 | } | |
4422 | ||
3529d8c2 JA |
4423 | static int io_poll_remove_prep(struct io_kiocb *req, |
4424 | const struct io_uring_sqe *sqe) | |
0969e783 | 4425 | { |
0969e783 JA |
4426 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4427 | return -EINVAL; | |
4428 | if (sqe->ioprio || sqe->off || sqe->len || sqe->buf_index || | |
4429 | sqe->poll_events) | |
4430 | return -EINVAL; | |
4431 | ||
4432 | req->poll.addr = READ_ONCE(sqe->addr); | |
0969e783 JA |
4433 | return 0; |
4434 | } | |
4435 | ||
221c5eb2 JA |
4436 | /* |
4437 | * Find a running poll command that matches one specified in sqe->addr, | |
4438 | * and remove it if found. | |
4439 | */ | |
fc4df999 | 4440 | static int io_poll_remove(struct io_kiocb *req) |
221c5eb2 JA |
4441 | { |
4442 | struct io_ring_ctx *ctx = req->ctx; | |
0969e783 | 4443 | u64 addr; |
47f46768 | 4444 | int ret; |
221c5eb2 | 4445 | |
0969e783 | 4446 | addr = req->poll.addr; |
221c5eb2 | 4447 | spin_lock_irq(&ctx->completion_lock); |
0969e783 | 4448 | ret = io_poll_cancel(ctx, addr); |
221c5eb2 JA |
4449 | spin_unlock_irq(&ctx->completion_lock); |
4450 | ||
78e19bbe | 4451 | io_cqring_add_event(req, ret); |
4e88d6e7 JA |
4452 | if (ret < 0) |
4453 | req_set_fail_links(req); | |
e65ef56d | 4454 | io_put_req(req); |
221c5eb2 JA |
4455 | return 0; |
4456 | } | |
4457 | ||
b0dd8a41 | 4458 | static void io_poll_complete(struct io_kiocb *req, __poll_t mask, int error) |
221c5eb2 | 4459 | { |
a197f664 JL |
4460 | struct io_ring_ctx *ctx = req->ctx; |
4461 | ||
8c838788 | 4462 | req->poll.done = true; |
b0a20349 | 4463 | io_cqring_fill_event(req, error ? error : mangle_poll(mask)); |
8c838788 | 4464 | io_commit_cqring(ctx); |
221c5eb2 JA |
4465 | } |
4466 | ||
b41e9852 | 4467 | static void io_poll_task_handler(struct io_kiocb *req, struct io_kiocb **nxt) |
221c5eb2 | 4468 | { |
221c5eb2 | 4469 | struct io_ring_ctx *ctx = req->ctx; |
a6ba632d JA |
4470 | struct io_poll_iocb *poll = &req->poll; |
4471 | ||
74ce6ce4 | 4472 | if (io_poll_rewait(req, poll)) { |
a6ba632d JA |
4473 | spin_unlock_irq(&ctx->completion_lock); |
4474 | return; | |
4475 | } | |
74ce6ce4 | 4476 | |
78076bb6 | 4477 | hash_del(&req->hash_node); |
b41e9852 JA |
4478 | io_poll_complete(req, req->result, 0); |
4479 | req->flags |= REQ_F_COMP_LOCKED; | |
4480 | io_put_req_find_next(req, nxt); | |
221c5eb2 JA |
4481 | spin_unlock_irq(&ctx->completion_lock); |
4482 | ||
8c838788 | 4483 | io_cqring_ev_posted(ctx); |
221c5eb2 JA |
4484 | } |
4485 | ||
b41e9852 | 4486 | static void io_poll_task_func(struct callback_head *cb) |
e94f141b | 4487 | { |
b41e9852 JA |
4488 | struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work); |
4489 | struct io_kiocb *nxt = NULL; | |
e94f141b | 4490 | |
b41e9852 | 4491 | io_poll_task_handler(req, &nxt); |
d7718a9d JA |
4492 | if (nxt) { |
4493 | struct io_ring_ctx *ctx = nxt->ctx; | |
e94f141b | 4494 | |
d7718a9d | 4495 | mutex_lock(&ctx->uring_lock); |
b41e9852 | 4496 | __io_queue_sqe(nxt, NULL); |
d7718a9d | 4497 | mutex_unlock(&ctx->uring_lock); |
e94f141b | 4498 | } |
f0b493e6 JA |
4499 | } |
4500 | ||
221c5eb2 JA |
4501 | static int io_poll_wake(struct wait_queue_entry *wait, unsigned mode, int sync, |
4502 | void *key) | |
4503 | { | |
c2f2eb7d JA |
4504 | struct io_kiocb *req = wait->private; |
4505 | struct io_poll_iocb *poll = &req->poll; | |
221c5eb2 | 4506 | |
d7718a9d | 4507 | return __io_async_wake(req, poll, key_to_poll(key), io_poll_task_func); |
221c5eb2 JA |
4508 | } |
4509 | ||
221c5eb2 JA |
4510 | static void io_poll_queue_proc(struct file *file, struct wait_queue_head *head, |
4511 | struct poll_table_struct *p) | |
4512 | { | |
4513 | struct io_poll_table *pt = container_of(p, struct io_poll_table, pt); | |
4514 | ||
d7718a9d | 4515 | __io_queue_proc(&pt->req->poll, pt, head); |
eac406c6 JA |
4516 | } |
4517 | ||
3529d8c2 | 4518 | static int io_poll_add_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
221c5eb2 JA |
4519 | { |
4520 | struct io_poll_iocb *poll = &req->poll; | |
221c5eb2 | 4521 | u16 events; |
221c5eb2 JA |
4522 | |
4523 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) | |
4524 | return -EINVAL; | |
4525 | if (sqe->addr || sqe->ioprio || sqe->off || sqe->len || sqe->buf_index) | |
4526 | return -EINVAL; | |
09bb8394 JA |
4527 | if (!poll->file) |
4528 | return -EBADF; | |
221c5eb2 | 4529 | |
221c5eb2 JA |
4530 | events = READ_ONCE(sqe->poll_events); |
4531 | poll->events = demangle_poll(events) | EPOLLERR | EPOLLHUP; | |
b41e9852 | 4532 | |
3537b6a7 | 4533 | get_task_struct(current); |
b41e9852 | 4534 | req->task = current; |
0969e783 JA |
4535 | return 0; |
4536 | } | |
4537 | ||
014db007 | 4538 | static int io_poll_add(struct io_kiocb *req) |
0969e783 JA |
4539 | { |
4540 | struct io_poll_iocb *poll = &req->poll; | |
4541 | struct io_ring_ctx *ctx = req->ctx; | |
4542 | struct io_poll_table ipt; | |
0969e783 | 4543 | __poll_t mask; |
0969e783 | 4544 | |
78076bb6 | 4545 | INIT_HLIST_NODE(&req->hash_node); |
36703247 | 4546 | INIT_LIST_HEAD(&req->list); |
d7718a9d | 4547 | ipt.pt._qproc = io_poll_queue_proc; |
36703247 | 4548 | |
d7718a9d JA |
4549 | mask = __io_arm_poll_handler(req, &req->poll, &ipt, poll->events, |
4550 | io_poll_wake); | |
221c5eb2 | 4551 | |
8c838788 | 4552 | if (mask) { /* no async, we'd stolen it */ |
221c5eb2 | 4553 | ipt.error = 0; |
b0dd8a41 | 4554 | io_poll_complete(req, mask, 0); |
221c5eb2 | 4555 | } |
221c5eb2 JA |
4556 | spin_unlock_irq(&ctx->completion_lock); |
4557 | ||
8c838788 JA |
4558 | if (mask) { |
4559 | io_cqring_ev_posted(ctx); | |
014db007 | 4560 | io_put_req(req); |
221c5eb2 | 4561 | } |
8c838788 | 4562 | return ipt.error; |
221c5eb2 JA |
4563 | } |
4564 | ||
5262f567 JA |
4565 | static enum hrtimer_restart io_timeout_fn(struct hrtimer *timer) |
4566 | { | |
ad8a48ac JA |
4567 | struct io_timeout_data *data = container_of(timer, |
4568 | struct io_timeout_data, timer); | |
4569 | struct io_kiocb *req = data->req; | |
4570 | struct io_ring_ctx *ctx = req->ctx; | |
5262f567 JA |
4571 | unsigned long flags; |
4572 | ||
5262f567 JA |
4573 | atomic_inc(&ctx->cq_timeouts); |
4574 | ||
4575 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
ef03681a | 4576 | /* |
11365043 JA |
4577 | * We could be racing with timeout deletion. If the list is empty, |
4578 | * then timeout lookup already found it and will be handling it. | |
ef03681a | 4579 | */ |
842f9612 | 4580 | if (!list_empty(&req->list)) { |
11365043 | 4581 | struct io_kiocb *prev; |
5262f567 | 4582 | |
11365043 JA |
4583 | /* |
4584 | * Adjust the reqs sequence before the current one because it | |
d195a66e | 4585 | * will consume a slot in the cq_ring and the cq_tail |
11365043 JA |
4586 | * pointer will be increased, otherwise other timeout reqs may |
4587 | * return in advance without waiting for enough wait_nr. | |
4588 | */ | |
4589 | prev = req; | |
4590 | list_for_each_entry_continue_reverse(prev, &ctx->timeout_list, list) | |
4591 | prev->sequence++; | |
11365043 | 4592 | list_del_init(&req->list); |
11365043 | 4593 | } |
5262f567 | 4594 | |
78e19bbe | 4595 | io_cqring_fill_event(req, -ETIME); |
5262f567 JA |
4596 | io_commit_cqring(ctx); |
4597 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
4598 | ||
4599 | io_cqring_ev_posted(ctx); | |
4e88d6e7 | 4600 | req_set_fail_links(req); |
5262f567 JA |
4601 | io_put_req(req); |
4602 | return HRTIMER_NORESTART; | |
4603 | } | |
4604 | ||
47f46768 JA |
4605 | static int io_timeout_cancel(struct io_ring_ctx *ctx, __u64 user_data) |
4606 | { | |
4607 | struct io_kiocb *req; | |
4608 | int ret = -ENOENT; | |
4609 | ||
4610 | list_for_each_entry(req, &ctx->timeout_list, list) { | |
4611 | if (user_data == req->user_data) { | |
4612 | list_del_init(&req->list); | |
4613 | ret = 0; | |
4614 | break; | |
4615 | } | |
4616 | } | |
4617 | ||
4618 | if (ret == -ENOENT) | |
4619 | return ret; | |
4620 | ||
2d28390a | 4621 | ret = hrtimer_try_to_cancel(&req->io->timeout.timer); |
47f46768 JA |
4622 | if (ret == -1) |
4623 | return -EALREADY; | |
4624 | ||
4e88d6e7 | 4625 | req_set_fail_links(req); |
47f46768 JA |
4626 | io_cqring_fill_event(req, -ECANCELED); |
4627 | io_put_req(req); | |
4628 | return 0; | |
4629 | } | |
4630 | ||
3529d8c2 JA |
4631 | static int io_timeout_remove_prep(struct io_kiocb *req, |
4632 | const struct io_uring_sqe *sqe) | |
b29472ee | 4633 | { |
b29472ee JA |
4634 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4635 | return -EINVAL; | |
4636 | if (sqe->flags || sqe->ioprio || sqe->buf_index || sqe->len) | |
4637 | return -EINVAL; | |
4638 | ||
4639 | req->timeout.addr = READ_ONCE(sqe->addr); | |
4640 | req->timeout.flags = READ_ONCE(sqe->timeout_flags); | |
4641 | if (req->timeout.flags) | |
4642 | return -EINVAL; | |
4643 | ||
b29472ee JA |
4644 | return 0; |
4645 | } | |
4646 | ||
11365043 JA |
4647 | /* |
4648 | * Remove or update an existing timeout command | |
4649 | */ | |
fc4df999 | 4650 | static int io_timeout_remove(struct io_kiocb *req) |
11365043 JA |
4651 | { |
4652 | struct io_ring_ctx *ctx = req->ctx; | |
47f46768 | 4653 | int ret; |
11365043 | 4654 | |
11365043 | 4655 | spin_lock_irq(&ctx->completion_lock); |
b29472ee | 4656 | ret = io_timeout_cancel(ctx, req->timeout.addr); |
11365043 | 4657 | |
47f46768 | 4658 | io_cqring_fill_event(req, ret); |
11365043 JA |
4659 | io_commit_cqring(ctx); |
4660 | spin_unlock_irq(&ctx->completion_lock); | |
5262f567 | 4661 | io_cqring_ev_posted(ctx); |
4e88d6e7 JA |
4662 | if (ret < 0) |
4663 | req_set_fail_links(req); | |
ec9c02ad | 4664 | io_put_req(req); |
11365043 | 4665 | return 0; |
5262f567 JA |
4666 | } |
4667 | ||
3529d8c2 | 4668 | static int io_timeout_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe, |
2d28390a | 4669 | bool is_timeout_link) |
5262f567 | 4670 | { |
ad8a48ac | 4671 | struct io_timeout_data *data; |
a41525ab | 4672 | unsigned flags; |
5262f567 | 4673 | |
ad8a48ac | 4674 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
5262f567 | 4675 | return -EINVAL; |
ad8a48ac | 4676 | if (sqe->ioprio || sqe->buf_index || sqe->len != 1) |
a41525ab | 4677 | return -EINVAL; |
2d28390a JA |
4678 | if (sqe->off && is_timeout_link) |
4679 | return -EINVAL; | |
a41525ab JA |
4680 | flags = READ_ONCE(sqe->timeout_flags); |
4681 | if (flags & ~IORING_TIMEOUT_ABS) | |
5262f567 | 4682 | return -EINVAL; |
bdf20073 | 4683 | |
26a61679 JA |
4684 | req->timeout.count = READ_ONCE(sqe->off); |
4685 | ||
3529d8c2 | 4686 | if (!req->io && io_alloc_async_ctx(req)) |
26a61679 JA |
4687 | return -ENOMEM; |
4688 | ||
4689 | data = &req->io->timeout; | |
ad8a48ac | 4690 | data->req = req; |
ad8a48ac JA |
4691 | req->flags |= REQ_F_TIMEOUT; |
4692 | ||
4693 | if (get_timespec64(&data->ts, u64_to_user_ptr(sqe->addr))) | |
5262f567 JA |
4694 | return -EFAULT; |
4695 | ||
11365043 | 4696 | if (flags & IORING_TIMEOUT_ABS) |
ad8a48ac | 4697 | data->mode = HRTIMER_MODE_ABS; |
11365043 | 4698 | else |
ad8a48ac | 4699 | data->mode = HRTIMER_MODE_REL; |
11365043 | 4700 | |
ad8a48ac JA |
4701 | hrtimer_init(&data->timer, CLOCK_MONOTONIC, data->mode); |
4702 | return 0; | |
4703 | } | |
4704 | ||
fc4df999 | 4705 | static int io_timeout(struct io_kiocb *req) |
ad8a48ac | 4706 | { |
ad8a48ac JA |
4707 | struct io_ring_ctx *ctx = req->ctx; |
4708 | struct io_timeout_data *data; | |
4709 | struct list_head *entry; | |
4710 | unsigned span = 0; | |
b55ce732 | 4711 | u32 count = req->timeout.count; |
22cad158 | 4712 | u32 seq = req->sequence; |
ad8a48ac | 4713 | |
2d28390a | 4714 | data = &req->io->timeout; |
93bd25bb | 4715 | |
5262f567 JA |
4716 | /* |
4717 | * sqe->off holds how many events that need to occur for this | |
93bd25bb JA |
4718 | * timeout event to be satisfied. If it isn't set, then this is |
4719 | * a pure timeout request, sequence isn't used. | |
5262f567 | 4720 | */ |
93bd25bb JA |
4721 | if (!count) { |
4722 | req->flags |= REQ_F_TIMEOUT_NOSEQ; | |
4723 | spin_lock_irq(&ctx->completion_lock); | |
4724 | entry = ctx->timeout_list.prev; | |
4725 | goto add; | |
4726 | } | |
5262f567 | 4727 | |
22cad158 | 4728 | req->sequence = seq + count; |
5262f567 JA |
4729 | |
4730 | /* | |
4731 | * Insertion sort, ensuring the first entry in the list is always | |
4732 | * the one we need first. | |
4733 | */ | |
5262f567 JA |
4734 | spin_lock_irq(&ctx->completion_lock); |
4735 | list_for_each_prev(entry, &ctx->timeout_list) { | |
4736 | struct io_kiocb *nxt = list_entry(entry, struct io_kiocb, list); | |
22cad158 | 4737 | unsigned nxt_seq; |
5da0fb1a | 4738 | long long tmp, tmp_nxt; |
b55ce732 | 4739 | u32 nxt_offset = nxt->timeout.count; |
5262f567 | 4740 | |
93bd25bb JA |
4741 | if (nxt->flags & REQ_F_TIMEOUT_NOSEQ) |
4742 | continue; | |
4743 | ||
5da0fb1a | 4744 | /* |
22cad158 | 4745 | * Since seq + count can overflow, use type long |
5da0fb1a | 4746 | * long to store it. |
4747 | */ | |
22cad158 PB |
4748 | tmp = (long long)seq + count; |
4749 | nxt_seq = nxt->sequence - nxt_offset; | |
4750 | tmp_nxt = (long long)nxt_seq + nxt_offset; | |
5da0fb1a | 4751 | |
4752 | /* | |
4753 | * cached_sq_head may overflow, and it will never overflow twice | |
4754 | * once there is some timeout req still be valid. | |
4755 | */ | |
22cad158 | 4756 | if (seq < nxt_seq) |
8b07a65a | 4757 | tmp += UINT_MAX; |
5da0fb1a | 4758 | |
a1f58ba4 | 4759 | if (tmp > tmp_nxt) |
5262f567 | 4760 | break; |
a1f58ba4 | 4761 | |
4762 | /* | |
4763 | * Sequence of reqs after the insert one and itself should | |
4764 | * be adjusted because each timeout req consumes a slot. | |
4765 | */ | |
4766 | span++; | |
4767 | nxt->sequence++; | |
5262f567 | 4768 | } |
a1f58ba4 | 4769 | req->sequence -= span; |
93bd25bb | 4770 | add: |
5262f567 | 4771 | list_add(&req->list, entry); |
ad8a48ac JA |
4772 | data->timer.function = io_timeout_fn; |
4773 | hrtimer_start(&data->timer, timespec64_to_ktime(data->ts), data->mode); | |
5262f567 | 4774 | spin_unlock_irq(&ctx->completion_lock); |
5262f567 JA |
4775 | return 0; |
4776 | } | |
5262f567 | 4777 | |
62755e35 JA |
4778 | static bool io_cancel_cb(struct io_wq_work *work, void *data) |
4779 | { | |
4780 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); | |
4781 | ||
4782 | return req->user_data == (unsigned long) data; | |
4783 | } | |
4784 | ||
e977d6d3 | 4785 | static int io_async_cancel_one(struct io_ring_ctx *ctx, void *sqe_addr) |
62755e35 | 4786 | { |
62755e35 | 4787 | enum io_wq_cancel cancel_ret; |
62755e35 JA |
4788 | int ret = 0; |
4789 | ||
62755e35 JA |
4790 | cancel_ret = io_wq_cancel_cb(ctx->io_wq, io_cancel_cb, sqe_addr); |
4791 | switch (cancel_ret) { | |
4792 | case IO_WQ_CANCEL_OK: | |
4793 | ret = 0; | |
4794 | break; | |
4795 | case IO_WQ_CANCEL_RUNNING: | |
4796 | ret = -EALREADY; | |
4797 | break; | |
4798 | case IO_WQ_CANCEL_NOTFOUND: | |
4799 | ret = -ENOENT; | |
4800 | break; | |
4801 | } | |
4802 | ||
e977d6d3 JA |
4803 | return ret; |
4804 | } | |
4805 | ||
47f46768 JA |
4806 | static void io_async_find_and_cancel(struct io_ring_ctx *ctx, |
4807 | struct io_kiocb *req, __u64 sqe_addr, | |
014db007 | 4808 | int success_ret) |
47f46768 JA |
4809 | { |
4810 | unsigned long flags; | |
4811 | int ret; | |
4812 | ||
4813 | ret = io_async_cancel_one(ctx, (void *) (unsigned long) sqe_addr); | |
4814 | if (ret != -ENOENT) { | |
4815 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
4816 | goto done; | |
4817 | } | |
4818 | ||
4819 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
4820 | ret = io_timeout_cancel(ctx, sqe_addr); | |
4821 | if (ret != -ENOENT) | |
4822 | goto done; | |
4823 | ret = io_poll_cancel(ctx, sqe_addr); | |
4824 | done: | |
b0dd8a41 JA |
4825 | if (!ret) |
4826 | ret = success_ret; | |
47f46768 JA |
4827 | io_cqring_fill_event(req, ret); |
4828 | io_commit_cqring(ctx); | |
4829 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
4830 | io_cqring_ev_posted(ctx); | |
4831 | ||
4e88d6e7 JA |
4832 | if (ret < 0) |
4833 | req_set_fail_links(req); | |
014db007 | 4834 | io_put_req(req); |
47f46768 JA |
4835 | } |
4836 | ||
3529d8c2 JA |
4837 | static int io_async_cancel_prep(struct io_kiocb *req, |
4838 | const struct io_uring_sqe *sqe) | |
e977d6d3 | 4839 | { |
fbf23849 | 4840 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
e977d6d3 JA |
4841 | return -EINVAL; |
4842 | if (sqe->flags || sqe->ioprio || sqe->off || sqe->len || | |
4843 | sqe->cancel_flags) | |
4844 | return -EINVAL; | |
4845 | ||
fbf23849 JA |
4846 | req->cancel.addr = READ_ONCE(sqe->addr); |
4847 | return 0; | |
4848 | } | |
4849 | ||
014db007 | 4850 | static int io_async_cancel(struct io_kiocb *req) |
fbf23849 JA |
4851 | { |
4852 | struct io_ring_ctx *ctx = req->ctx; | |
fbf23849 | 4853 | |
014db007 | 4854 | io_async_find_and_cancel(ctx, req, req->cancel.addr, 0); |
5262f567 JA |
4855 | return 0; |
4856 | } | |
4857 | ||
05f3fb3c JA |
4858 | static int io_files_update_prep(struct io_kiocb *req, |
4859 | const struct io_uring_sqe *sqe) | |
4860 | { | |
4861 | if (sqe->flags || sqe->ioprio || sqe->rw_flags) | |
4862 | return -EINVAL; | |
4863 | ||
4864 | req->files_update.offset = READ_ONCE(sqe->off); | |
4865 | req->files_update.nr_args = READ_ONCE(sqe->len); | |
4866 | if (!req->files_update.nr_args) | |
4867 | return -EINVAL; | |
4868 | req->files_update.arg = READ_ONCE(sqe->addr); | |
4869 | return 0; | |
4870 | } | |
4871 | ||
4872 | static int io_files_update(struct io_kiocb *req, bool force_nonblock) | |
fbf23849 JA |
4873 | { |
4874 | struct io_ring_ctx *ctx = req->ctx; | |
05f3fb3c JA |
4875 | struct io_uring_files_update up; |
4876 | int ret; | |
fbf23849 | 4877 | |
f86cd20c | 4878 | if (force_nonblock) |
05f3fb3c | 4879 | return -EAGAIN; |
05f3fb3c JA |
4880 | |
4881 | up.offset = req->files_update.offset; | |
4882 | up.fds = req->files_update.arg; | |
4883 | ||
4884 | mutex_lock(&ctx->uring_lock); | |
4885 | ret = __io_sqe_files_update(ctx, &up, req->files_update.nr_args); | |
4886 | mutex_unlock(&ctx->uring_lock); | |
4887 | ||
4888 | if (ret < 0) | |
4889 | req_set_fail_links(req); | |
4890 | io_cqring_add_event(req, ret); | |
4891 | io_put_req(req); | |
5262f567 JA |
4892 | return 0; |
4893 | } | |
4894 | ||
3529d8c2 JA |
4895 | static int io_req_defer_prep(struct io_kiocb *req, |
4896 | const struct io_uring_sqe *sqe) | |
f67676d1 | 4897 | { |
e781573e | 4898 | ssize_t ret = 0; |
f67676d1 | 4899 | |
f1d96a8f PB |
4900 | if (!sqe) |
4901 | return 0; | |
4902 | ||
f86cd20c JA |
4903 | if (io_op_defs[req->opcode].file_table) { |
4904 | ret = io_grab_files(req); | |
4905 | if (unlikely(ret)) | |
4906 | return ret; | |
4907 | } | |
4908 | ||
cccf0ee8 JA |
4909 | io_req_work_grab_env(req, &io_op_defs[req->opcode]); |
4910 | ||
d625c6ee | 4911 | switch (req->opcode) { |
e781573e JA |
4912 | case IORING_OP_NOP: |
4913 | break; | |
f67676d1 JA |
4914 | case IORING_OP_READV: |
4915 | case IORING_OP_READ_FIXED: | |
3a6820f2 | 4916 | case IORING_OP_READ: |
3529d8c2 | 4917 | ret = io_read_prep(req, sqe, true); |
f67676d1 JA |
4918 | break; |
4919 | case IORING_OP_WRITEV: | |
4920 | case IORING_OP_WRITE_FIXED: | |
3a6820f2 | 4921 | case IORING_OP_WRITE: |
3529d8c2 | 4922 | ret = io_write_prep(req, sqe, true); |
f67676d1 | 4923 | break; |
0969e783 | 4924 | case IORING_OP_POLL_ADD: |
3529d8c2 | 4925 | ret = io_poll_add_prep(req, sqe); |
0969e783 JA |
4926 | break; |
4927 | case IORING_OP_POLL_REMOVE: | |
3529d8c2 | 4928 | ret = io_poll_remove_prep(req, sqe); |
0969e783 | 4929 | break; |
8ed8d3c3 | 4930 | case IORING_OP_FSYNC: |
3529d8c2 | 4931 | ret = io_prep_fsync(req, sqe); |
8ed8d3c3 JA |
4932 | break; |
4933 | case IORING_OP_SYNC_FILE_RANGE: | |
3529d8c2 | 4934 | ret = io_prep_sfr(req, sqe); |
8ed8d3c3 | 4935 | break; |
03b1230c | 4936 | case IORING_OP_SENDMSG: |
fddaface | 4937 | case IORING_OP_SEND: |
3529d8c2 | 4938 | ret = io_sendmsg_prep(req, sqe); |
03b1230c JA |
4939 | break; |
4940 | case IORING_OP_RECVMSG: | |
fddaface | 4941 | case IORING_OP_RECV: |
3529d8c2 | 4942 | ret = io_recvmsg_prep(req, sqe); |
03b1230c | 4943 | break; |
f499a021 | 4944 | case IORING_OP_CONNECT: |
3529d8c2 | 4945 | ret = io_connect_prep(req, sqe); |
f499a021 | 4946 | break; |
2d28390a | 4947 | case IORING_OP_TIMEOUT: |
3529d8c2 | 4948 | ret = io_timeout_prep(req, sqe, false); |
b7bb4f7d | 4949 | break; |
b29472ee | 4950 | case IORING_OP_TIMEOUT_REMOVE: |
3529d8c2 | 4951 | ret = io_timeout_remove_prep(req, sqe); |
b29472ee | 4952 | break; |
fbf23849 | 4953 | case IORING_OP_ASYNC_CANCEL: |
3529d8c2 | 4954 | ret = io_async_cancel_prep(req, sqe); |
fbf23849 | 4955 | break; |
2d28390a | 4956 | case IORING_OP_LINK_TIMEOUT: |
3529d8c2 | 4957 | ret = io_timeout_prep(req, sqe, true); |
b7bb4f7d | 4958 | break; |
8ed8d3c3 | 4959 | case IORING_OP_ACCEPT: |
3529d8c2 | 4960 | ret = io_accept_prep(req, sqe); |
8ed8d3c3 | 4961 | break; |
d63d1b5e JA |
4962 | case IORING_OP_FALLOCATE: |
4963 | ret = io_fallocate_prep(req, sqe); | |
4964 | break; | |
15b71abe JA |
4965 | case IORING_OP_OPENAT: |
4966 | ret = io_openat_prep(req, sqe); | |
4967 | break; | |
b5dba59e JA |
4968 | case IORING_OP_CLOSE: |
4969 | ret = io_close_prep(req, sqe); | |
4970 | break; | |
05f3fb3c JA |
4971 | case IORING_OP_FILES_UPDATE: |
4972 | ret = io_files_update_prep(req, sqe); | |
4973 | break; | |
eddc7ef5 JA |
4974 | case IORING_OP_STATX: |
4975 | ret = io_statx_prep(req, sqe); | |
4976 | break; | |
4840e418 JA |
4977 | case IORING_OP_FADVISE: |
4978 | ret = io_fadvise_prep(req, sqe); | |
4979 | break; | |
c1ca757b JA |
4980 | case IORING_OP_MADVISE: |
4981 | ret = io_madvise_prep(req, sqe); | |
4982 | break; | |
cebdb986 JA |
4983 | case IORING_OP_OPENAT2: |
4984 | ret = io_openat2_prep(req, sqe); | |
4985 | break; | |
3e4827b0 JA |
4986 | case IORING_OP_EPOLL_CTL: |
4987 | ret = io_epoll_ctl_prep(req, sqe); | |
4988 | break; | |
7d67af2c PB |
4989 | case IORING_OP_SPLICE: |
4990 | ret = io_splice_prep(req, sqe); | |
4991 | break; | |
ddf0322d JA |
4992 | case IORING_OP_PROVIDE_BUFFERS: |
4993 | ret = io_provide_buffers_prep(req, sqe); | |
4994 | break; | |
067524e9 JA |
4995 | case IORING_OP_REMOVE_BUFFERS: |
4996 | ret = io_remove_buffers_prep(req, sqe); | |
4997 | break; | |
f67676d1 | 4998 | default: |
e781573e JA |
4999 | printk_once(KERN_WARNING "io_uring: unhandled opcode %d\n", |
5000 | req->opcode); | |
5001 | ret = -EINVAL; | |
b7bb4f7d | 5002 | break; |
f67676d1 JA |
5003 | } |
5004 | ||
b7bb4f7d | 5005 | return ret; |
f67676d1 JA |
5006 | } |
5007 | ||
3529d8c2 | 5008 | static int io_req_defer(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
de0617e4 | 5009 | { |
a197f664 | 5010 | struct io_ring_ctx *ctx = req->ctx; |
f67676d1 | 5011 | int ret; |
de0617e4 | 5012 | |
9d858b21 | 5013 | /* Still need defer if there is pending req in defer list. */ |
4ee36314 | 5014 | if (!req_need_defer(req) && list_empty_careful(&ctx->defer_list)) |
de0617e4 JA |
5015 | return 0; |
5016 | ||
3529d8c2 | 5017 | if (!req->io && io_alloc_async_ctx(req)) |
de0617e4 JA |
5018 | return -EAGAIN; |
5019 | ||
3529d8c2 | 5020 | ret = io_req_defer_prep(req, sqe); |
b7bb4f7d | 5021 | if (ret < 0) |
2d28390a | 5022 | return ret; |
2d28390a | 5023 | |
de0617e4 | 5024 | spin_lock_irq(&ctx->completion_lock); |
9d858b21 | 5025 | if (!req_need_defer(req) && list_empty(&ctx->defer_list)) { |
de0617e4 | 5026 | spin_unlock_irq(&ctx->completion_lock); |
de0617e4 JA |
5027 | return 0; |
5028 | } | |
5029 | ||
915967f6 | 5030 | trace_io_uring_defer(ctx, req, req->user_data); |
de0617e4 JA |
5031 | list_add_tail(&req->list, &ctx->defer_list); |
5032 | spin_unlock_irq(&ctx->completion_lock); | |
5033 | return -EIOCBQUEUED; | |
5034 | } | |
5035 | ||
99bc4c38 PB |
5036 | static void io_cleanup_req(struct io_kiocb *req) |
5037 | { | |
5038 | struct io_async_ctx *io = req->io; | |
5039 | ||
5040 | switch (req->opcode) { | |
5041 | case IORING_OP_READV: | |
5042 | case IORING_OP_READ_FIXED: | |
5043 | case IORING_OP_READ: | |
bcda7baa JA |
5044 | if (req->flags & REQ_F_BUFFER_SELECTED) |
5045 | kfree((void *)(unsigned long)req->rw.addr); | |
5046 | /* fallthrough */ | |
99bc4c38 PB |
5047 | case IORING_OP_WRITEV: |
5048 | case IORING_OP_WRITE_FIXED: | |
5049 | case IORING_OP_WRITE: | |
5050 | if (io->rw.iov != io->rw.fast_iov) | |
5051 | kfree(io->rw.iov); | |
5052 | break; | |
99bc4c38 | 5053 | case IORING_OP_RECVMSG: |
52de1fe1 JA |
5054 | if (req->flags & REQ_F_BUFFER_SELECTED) |
5055 | kfree(req->sr_msg.kbuf); | |
5056 | /* fallthrough */ | |
5057 | case IORING_OP_SENDMSG: | |
99bc4c38 PB |
5058 | if (io->msg.iov != io->msg.fast_iov) |
5059 | kfree(io->msg.iov); | |
5060 | break; | |
bcda7baa JA |
5061 | case IORING_OP_RECV: |
5062 | if (req->flags & REQ_F_BUFFER_SELECTED) | |
5063 | kfree(req->sr_msg.kbuf); | |
5064 | break; | |
8fef80bf PB |
5065 | case IORING_OP_OPENAT: |
5066 | case IORING_OP_OPENAT2: | |
5067 | case IORING_OP_STATX: | |
5068 | putname(req->open.filename); | |
5069 | break; | |
7d67af2c PB |
5070 | case IORING_OP_SPLICE: |
5071 | io_put_file(req, req->splice.file_in, | |
5072 | (req->splice.flags & SPLICE_F_FD_IN_FIXED)); | |
5073 | break; | |
99bc4c38 PB |
5074 | } |
5075 | ||
5076 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
5077 | } | |
5078 | ||
3529d8c2 | 5079 | static int io_issue_sqe(struct io_kiocb *req, const struct io_uring_sqe *sqe, |
014db007 | 5080 | bool force_nonblock) |
2b188cc1 | 5081 | { |
a197f664 | 5082 | struct io_ring_ctx *ctx = req->ctx; |
d625c6ee | 5083 | int ret; |
2b188cc1 | 5084 | |
d625c6ee | 5085 | switch (req->opcode) { |
2b188cc1 | 5086 | case IORING_OP_NOP: |
78e19bbe | 5087 | ret = io_nop(req); |
2b188cc1 JA |
5088 | break; |
5089 | case IORING_OP_READV: | |
edafccee | 5090 | case IORING_OP_READ_FIXED: |
3a6820f2 | 5091 | case IORING_OP_READ: |
3529d8c2 JA |
5092 | if (sqe) { |
5093 | ret = io_read_prep(req, sqe, force_nonblock); | |
5094 | if (ret < 0) | |
5095 | break; | |
5096 | } | |
014db007 | 5097 | ret = io_read(req, force_nonblock); |
edafccee | 5098 | break; |
3529d8c2 | 5099 | case IORING_OP_WRITEV: |
edafccee | 5100 | case IORING_OP_WRITE_FIXED: |
3a6820f2 | 5101 | case IORING_OP_WRITE: |
3529d8c2 JA |
5102 | if (sqe) { |
5103 | ret = io_write_prep(req, sqe, force_nonblock); | |
5104 | if (ret < 0) | |
5105 | break; | |
5106 | } | |
014db007 | 5107 | ret = io_write(req, force_nonblock); |
2b188cc1 | 5108 | break; |
c992fe29 | 5109 | case IORING_OP_FSYNC: |
3529d8c2 JA |
5110 | if (sqe) { |
5111 | ret = io_prep_fsync(req, sqe); | |
5112 | if (ret < 0) | |
5113 | break; | |
5114 | } | |
014db007 | 5115 | ret = io_fsync(req, force_nonblock); |
c992fe29 | 5116 | break; |
221c5eb2 | 5117 | case IORING_OP_POLL_ADD: |
3529d8c2 JA |
5118 | if (sqe) { |
5119 | ret = io_poll_add_prep(req, sqe); | |
5120 | if (ret) | |
5121 | break; | |
5122 | } | |
014db007 | 5123 | ret = io_poll_add(req); |
221c5eb2 JA |
5124 | break; |
5125 | case IORING_OP_POLL_REMOVE: | |
3529d8c2 JA |
5126 | if (sqe) { |
5127 | ret = io_poll_remove_prep(req, sqe); | |
5128 | if (ret < 0) | |
5129 | break; | |
5130 | } | |
fc4df999 | 5131 | ret = io_poll_remove(req); |
221c5eb2 | 5132 | break; |
5d17b4a4 | 5133 | case IORING_OP_SYNC_FILE_RANGE: |
3529d8c2 JA |
5134 | if (sqe) { |
5135 | ret = io_prep_sfr(req, sqe); | |
5136 | if (ret < 0) | |
5137 | break; | |
5138 | } | |
014db007 | 5139 | ret = io_sync_file_range(req, force_nonblock); |
5d17b4a4 | 5140 | break; |
0fa03c62 | 5141 | case IORING_OP_SENDMSG: |
fddaface | 5142 | case IORING_OP_SEND: |
3529d8c2 JA |
5143 | if (sqe) { |
5144 | ret = io_sendmsg_prep(req, sqe); | |
5145 | if (ret < 0) | |
5146 | break; | |
5147 | } | |
fddaface | 5148 | if (req->opcode == IORING_OP_SENDMSG) |
014db007 | 5149 | ret = io_sendmsg(req, force_nonblock); |
fddaface | 5150 | else |
014db007 | 5151 | ret = io_send(req, force_nonblock); |
0fa03c62 | 5152 | break; |
aa1fa28f | 5153 | case IORING_OP_RECVMSG: |
fddaface | 5154 | case IORING_OP_RECV: |
3529d8c2 JA |
5155 | if (sqe) { |
5156 | ret = io_recvmsg_prep(req, sqe); | |
5157 | if (ret) | |
5158 | break; | |
5159 | } | |
fddaface | 5160 | if (req->opcode == IORING_OP_RECVMSG) |
014db007 | 5161 | ret = io_recvmsg(req, force_nonblock); |
fddaface | 5162 | else |
014db007 | 5163 | ret = io_recv(req, force_nonblock); |
aa1fa28f | 5164 | break; |
5262f567 | 5165 | case IORING_OP_TIMEOUT: |
3529d8c2 JA |
5166 | if (sqe) { |
5167 | ret = io_timeout_prep(req, sqe, false); | |
5168 | if (ret) | |
5169 | break; | |
5170 | } | |
fc4df999 | 5171 | ret = io_timeout(req); |
5262f567 | 5172 | break; |
11365043 | 5173 | case IORING_OP_TIMEOUT_REMOVE: |
3529d8c2 JA |
5174 | if (sqe) { |
5175 | ret = io_timeout_remove_prep(req, sqe); | |
5176 | if (ret) | |
5177 | break; | |
5178 | } | |
fc4df999 | 5179 | ret = io_timeout_remove(req); |
11365043 | 5180 | break; |
17f2fe35 | 5181 | case IORING_OP_ACCEPT: |
3529d8c2 JA |
5182 | if (sqe) { |
5183 | ret = io_accept_prep(req, sqe); | |
5184 | if (ret) | |
5185 | break; | |
5186 | } | |
014db007 | 5187 | ret = io_accept(req, force_nonblock); |
17f2fe35 | 5188 | break; |
f8e85cf2 | 5189 | case IORING_OP_CONNECT: |
3529d8c2 JA |
5190 | if (sqe) { |
5191 | ret = io_connect_prep(req, sqe); | |
5192 | if (ret) | |
5193 | break; | |
5194 | } | |
014db007 | 5195 | ret = io_connect(req, force_nonblock); |
f8e85cf2 | 5196 | break; |
62755e35 | 5197 | case IORING_OP_ASYNC_CANCEL: |
3529d8c2 JA |
5198 | if (sqe) { |
5199 | ret = io_async_cancel_prep(req, sqe); | |
5200 | if (ret) | |
5201 | break; | |
5202 | } | |
014db007 | 5203 | ret = io_async_cancel(req); |
62755e35 | 5204 | break; |
d63d1b5e JA |
5205 | case IORING_OP_FALLOCATE: |
5206 | if (sqe) { | |
5207 | ret = io_fallocate_prep(req, sqe); | |
5208 | if (ret) | |
5209 | break; | |
5210 | } | |
014db007 | 5211 | ret = io_fallocate(req, force_nonblock); |
d63d1b5e | 5212 | break; |
15b71abe JA |
5213 | case IORING_OP_OPENAT: |
5214 | if (sqe) { | |
5215 | ret = io_openat_prep(req, sqe); | |
5216 | if (ret) | |
5217 | break; | |
5218 | } | |
014db007 | 5219 | ret = io_openat(req, force_nonblock); |
15b71abe | 5220 | break; |
b5dba59e JA |
5221 | case IORING_OP_CLOSE: |
5222 | if (sqe) { | |
5223 | ret = io_close_prep(req, sqe); | |
5224 | if (ret) | |
5225 | break; | |
5226 | } | |
014db007 | 5227 | ret = io_close(req, force_nonblock); |
b5dba59e | 5228 | break; |
05f3fb3c JA |
5229 | case IORING_OP_FILES_UPDATE: |
5230 | if (sqe) { | |
5231 | ret = io_files_update_prep(req, sqe); | |
5232 | if (ret) | |
5233 | break; | |
5234 | } | |
5235 | ret = io_files_update(req, force_nonblock); | |
5236 | break; | |
eddc7ef5 JA |
5237 | case IORING_OP_STATX: |
5238 | if (sqe) { | |
5239 | ret = io_statx_prep(req, sqe); | |
5240 | if (ret) | |
5241 | break; | |
5242 | } | |
014db007 | 5243 | ret = io_statx(req, force_nonblock); |
eddc7ef5 | 5244 | break; |
4840e418 JA |
5245 | case IORING_OP_FADVISE: |
5246 | if (sqe) { | |
5247 | ret = io_fadvise_prep(req, sqe); | |
5248 | if (ret) | |
5249 | break; | |
5250 | } | |
014db007 | 5251 | ret = io_fadvise(req, force_nonblock); |
4840e418 | 5252 | break; |
c1ca757b JA |
5253 | case IORING_OP_MADVISE: |
5254 | if (sqe) { | |
5255 | ret = io_madvise_prep(req, sqe); | |
5256 | if (ret) | |
5257 | break; | |
5258 | } | |
014db007 | 5259 | ret = io_madvise(req, force_nonblock); |
c1ca757b | 5260 | break; |
cebdb986 JA |
5261 | case IORING_OP_OPENAT2: |
5262 | if (sqe) { | |
5263 | ret = io_openat2_prep(req, sqe); | |
5264 | if (ret) | |
5265 | break; | |
5266 | } | |
014db007 | 5267 | ret = io_openat2(req, force_nonblock); |
cebdb986 | 5268 | break; |
3e4827b0 JA |
5269 | case IORING_OP_EPOLL_CTL: |
5270 | if (sqe) { | |
5271 | ret = io_epoll_ctl_prep(req, sqe); | |
5272 | if (ret) | |
5273 | break; | |
5274 | } | |
014db007 | 5275 | ret = io_epoll_ctl(req, force_nonblock); |
3e4827b0 | 5276 | break; |
7d67af2c PB |
5277 | case IORING_OP_SPLICE: |
5278 | if (sqe) { | |
5279 | ret = io_splice_prep(req, sqe); | |
5280 | if (ret < 0) | |
5281 | break; | |
5282 | } | |
014db007 | 5283 | ret = io_splice(req, force_nonblock); |
7d67af2c | 5284 | break; |
ddf0322d JA |
5285 | case IORING_OP_PROVIDE_BUFFERS: |
5286 | if (sqe) { | |
5287 | ret = io_provide_buffers_prep(req, sqe); | |
5288 | if (ret) | |
5289 | break; | |
5290 | } | |
5291 | ret = io_provide_buffers(req, force_nonblock); | |
5292 | break; | |
067524e9 JA |
5293 | case IORING_OP_REMOVE_BUFFERS: |
5294 | if (sqe) { | |
5295 | ret = io_remove_buffers_prep(req, sqe); | |
5296 | if (ret) | |
5297 | break; | |
5298 | } | |
5299 | ret = io_remove_buffers(req, force_nonblock); | |
3e4827b0 | 5300 | break; |
2b188cc1 JA |
5301 | default: |
5302 | ret = -EINVAL; | |
5303 | break; | |
5304 | } | |
5305 | ||
def596e9 JA |
5306 | if (ret) |
5307 | return ret; | |
5308 | ||
5309 | if (ctx->flags & IORING_SETUP_IOPOLL) { | |
11ba820b JA |
5310 | const bool in_async = io_wq_current_is_worker(); |
5311 | ||
9e645e11 | 5312 | if (req->result == -EAGAIN) |
def596e9 JA |
5313 | return -EAGAIN; |
5314 | ||
11ba820b JA |
5315 | /* workqueue context doesn't hold uring_lock, grab it now */ |
5316 | if (in_async) | |
5317 | mutex_lock(&ctx->uring_lock); | |
5318 | ||
def596e9 | 5319 | io_iopoll_req_issued(req); |
11ba820b JA |
5320 | |
5321 | if (in_async) | |
5322 | mutex_unlock(&ctx->uring_lock); | |
def596e9 JA |
5323 | } |
5324 | ||
5325 | return 0; | |
2b188cc1 JA |
5326 | } |
5327 | ||
561fb04a | 5328 | static void io_wq_submit_work(struct io_wq_work **workptr) |
2b188cc1 | 5329 | { |
561fb04a | 5330 | struct io_wq_work *work = *workptr; |
2b188cc1 | 5331 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); |
561fb04a | 5332 | int ret = 0; |
2b188cc1 | 5333 | |
0c9d5ccd JA |
5334 | /* if NO_CANCEL is set, we must still run the work */ |
5335 | if ((work->flags & (IO_WQ_WORK_CANCEL|IO_WQ_WORK_NO_CANCEL)) == | |
5336 | IO_WQ_WORK_CANCEL) { | |
561fb04a | 5337 | ret = -ECANCELED; |
0c9d5ccd | 5338 | } |
31b51510 | 5339 | |
561fb04a | 5340 | if (!ret) { |
561fb04a | 5341 | do { |
014db007 | 5342 | ret = io_issue_sqe(req, NULL, false); |
561fb04a JA |
5343 | /* |
5344 | * We can get EAGAIN for polled IO even though we're | |
5345 | * forcing a sync submission from here, since we can't | |
5346 | * wait for request slots on the block side. | |
5347 | */ | |
5348 | if (ret != -EAGAIN) | |
5349 | break; | |
5350 | cond_resched(); | |
5351 | } while (1); | |
5352 | } | |
31b51510 | 5353 | |
561fb04a | 5354 | if (ret) { |
4e88d6e7 | 5355 | req_set_fail_links(req); |
78e19bbe | 5356 | io_cqring_add_event(req, ret); |
817869d2 | 5357 | io_put_req(req); |
edafccee | 5358 | } |
2b188cc1 | 5359 | |
e9fd9396 | 5360 | io_steal_work(req, workptr); |
2b188cc1 JA |
5361 | } |
5362 | ||
65e19f54 JA |
5363 | static inline struct file *io_file_from_index(struct io_ring_ctx *ctx, |
5364 | int index) | |
5365 | { | |
5366 | struct fixed_file_table *table; | |
5367 | ||
05f3fb3c JA |
5368 | table = &ctx->file_data->table[index >> IORING_FILE_TABLE_SHIFT]; |
5369 | return table->files[index & IORING_FILE_TABLE_MASK];; | |
65e19f54 JA |
5370 | } |
5371 | ||
8da11c19 PB |
5372 | static int io_file_get(struct io_submit_state *state, struct io_kiocb *req, |
5373 | int fd, struct file **out_file, bool fixed) | |
09bb8394 | 5374 | { |
a197f664 | 5375 | struct io_ring_ctx *ctx = req->ctx; |
8da11c19 | 5376 | struct file *file; |
09bb8394 | 5377 | |
8da11c19 | 5378 | if (fixed) { |
05f3fb3c | 5379 | if (unlikely(!ctx->file_data || |
09bb8394 JA |
5380 | (unsigned) fd >= ctx->nr_user_files)) |
5381 | return -EBADF; | |
b7620121 | 5382 | fd = array_index_nospec(fd, ctx->nr_user_files); |
8da11c19 PB |
5383 | file = io_file_from_index(ctx, fd); |
5384 | if (!file) | |
08a45173 | 5385 | return -EBADF; |
05589553 XW |
5386 | req->fixed_file_refs = ctx->file_data->cur_refs; |
5387 | percpu_ref_get(req->fixed_file_refs); | |
09bb8394 | 5388 | } else { |
c826bd7a | 5389 | trace_io_uring_file_get(ctx, fd); |
8da11c19 PB |
5390 | file = __io_file_get(state, fd); |
5391 | if (unlikely(!file)) | |
09bb8394 JA |
5392 | return -EBADF; |
5393 | } | |
5394 | ||
8da11c19 | 5395 | *out_file = file; |
09bb8394 JA |
5396 | return 0; |
5397 | } | |
5398 | ||
8da11c19 | 5399 | static int io_req_set_file(struct io_submit_state *state, struct io_kiocb *req, |
63ff8223 | 5400 | int fd) |
8da11c19 | 5401 | { |
8da11c19 PB |
5402 | bool fixed; |
5403 | ||
63ff8223 | 5404 | fixed = (req->flags & REQ_F_FIXED_FILE) != 0; |
8da11c19 PB |
5405 | if (unlikely(!fixed && req->needs_fixed_file)) |
5406 | return -EBADF; | |
5407 | ||
5408 | return io_file_get(state, req, fd, &req->file, fixed); | |
5409 | } | |
5410 | ||
a197f664 | 5411 | static int io_grab_files(struct io_kiocb *req) |
fcb323cc JA |
5412 | { |
5413 | int ret = -EBADF; | |
a197f664 | 5414 | struct io_ring_ctx *ctx = req->ctx; |
fcb323cc | 5415 | |
5b0bbee4 | 5416 | if (req->work.files || (req->flags & REQ_F_NO_FILE_TABLE)) |
f86cd20c | 5417 | return 0; |
b14cca0c | 5418 | if (!ctx->ring_file) |
b5dba59e JA |
5419 | return -EBADF; |
5420 | ||
fcb323cc JA |
5421 | rcu_read_lock(); |
5422 | spin_lock_irq(&ctx->inflight_lock); | |
5423 | /* | |
5424 | * We use the f_ops->flush() handler to ensure that we can flush | |
5425 | * out work accessing these files if the fd is closed. Check if | |
5426 | * the fd has changed since we started down this path, and disallow | |
5427 | * this operation if it has. | |
5428 | */ | |
b14cca0c | 5429 | if (fcheck(ctx->ring_fd) == ctx->ring_file) { |
fcb323cc JA |
5430 | list_add(&req->inflight_entry, &ctx->inflight_list); |
5431 | req->flags |= REQ_F_INFLIGHT; | |
5432 | req->work.files = current->files; | |
5433 | ret = 0; | |
5434 | } | |
5435 | spin_unlock_irq(&ctx->inflight_lock); | |
5436 | rcu_read_unlock(); | |
5437 | ||
5438 | return ret; | |
5439 | } | |
5440 | ||
2665abfd | 5441 | static enum hrtimer_restart io_link_timeout_fn(struct hrtimer *timer) |
2b188cc1 | 5442 | { |
ad8a48ac JA |
5443 | struct io_timeout_data *data = container_of(timer, |
5444 | struct io_timeout_data, timer); | |
5445 | struct io_kiocb *req = data->req; | |
2665abfd JA |
5446 | struct io_ring_ctx *ctx = req->ctx; |
5447 | struct io_kiocb *prev = NULL; | |
5448 | unsigned long flags; | |
2665abfd JA |
5449 | |
5450 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
5451 | ||
5452 | /* | |
5453 | * We don't expect the list to be empty, that will only happen if we | |
5454 | * race with the completion of the linked work. | |
5455 | */ | |
4493233e PB |
5456 | if (!list_empty(&req->link_list)) { |
5457 | prev = list_entry(req->link_list.prev, struct io_kiocb, | |
5458 | link_list); | |
5d960724 | 5459 | if (refcount_inc_not_zero(&prev->refs)) { |
4493233e | 5460 | list_del_init(&req->link_list); |
5d960724 JA |
5461 | prev->flags &= ~REQ_F_LINK_TIMEOUT; |
5462 | } else | |
76a46e06 | 5463 | prev = NULL; |
2665abfd JA |
5464 | } |
5465 | ||
5466 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
5467 | ||
5468 | if (prev) { | |
4e88d6e7 | 5469 | req_set_fail_links(prev); |
014db007 | 5470 | io_async_find_and_cancel(ctx, req, prev->user_data, -ETIME); |
76a46e06 | 5471 | io_put_req(prev); |
47f46768 JA |
5472 | } else { |
5473 | io_cqring_add_event(req, -ETIME); | |
5474 | io_put_req(req); | |
2665abfd | 5475 | } |
2665abfd JA |
5476 | return HRTIMER_NORESTART; |
5477 | } | |
5478 | ||
ad8a48ac | 5479 | static void io_queue_linked_timeout(struct io_kiocb *req) |
2665abfd | 5480 | { |
76a46e06 | 5481 | struct io_ring_ctx *ctx = req->ctx; |
2665abfd | 5482 | |
76a46e06 JA |
5483 | /* |
5484 | * If the list is now empty, then our linked request finished before | |
5485 | * we got a chance to setup the timer | |
5486 | */ | |
5487 | spin_lock_irq(&ctx->completion_lock); | |
4493233e | 5488 | if (!list_empty(&req->link_list)) { |
2d28390a | 5489 | struct io_timeout_data *data = &req->io->timeout; |
94ae5e77 | 5490 | |
ad8a48ac JA |
5491 | data->timer.function = io_link_timeout_fn; |
5492 | hrtimer_start(&data->timer, timespec64_to_ktime(data->ts), | |
5493 | data->mode); | |
2665abfd | 5494 | } |
76a46e06 | 5495 | spin_unlock_irq(&ctx->completion_lock); |
2665abfd | 5496 | |
2665abfd | 5497 | /* drop submission reference */ |
76a46e06 JA |
5498 | io_put_req(req); |
5499 | } | |
2665abfd | 5500 | |
ad8a48ac | 5501 | static struct io_kiocb *io_prep_linked_timeout(struct io_kiocb *req) |
2665abfd JA |
5502 | { |
5503 | struct io_kiocb *nxt; | |
5504 | ||
dea3b49c | 5505 | if (!(req->flags & REQ_F_LINK_HEAD)) |
2665abfd | 5506 | return NULL; |
d7718a9d JA |
5507 | /* for polled retry, if flag is set, we already went through here */ |
5508 | if (req->flags & REQ_F_POLLED) | |
5509 | return NULL; | |
2665abfd | 5510 | |
4493233e PB |
5511 | nxt = list_first_entry_or_null(&req->link_list, struct io_kiocb, |
5512 | link_list); | |
d625c6ee | 5513 | if (!nxt || nxt->opcode != IORING_OP_LINK_TIMEOUT) |
76a46e06 | 5514 | return NULL; |
2665abfd | 5515 | |
76a46e06 | 5516 | req->flags |= REQ_F_LINK_TIMEOUT; |
76a46e06 | 5517 | return nxt; |
2665abfd JA |
5518 | } |
5519 | ||
3529d8c2 | 5520 | static void __io_queue_sqe(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
2b188cc1 | 5521 | { |
4a0a7a18 | 5522 | struct io_kiocb *linked_timeout; |
4bc4494e | 5523 | struct io_kiocb *nxt; |
193155c8 | 5524 | const struct cred *old_creds = NULL; |
e0c5c576 | 5525 | int ret; |
2b188cc1 | 5526 | |
4a0a7a18 JA |
5527 | again: |
5528 | linked_timeout = io_prep_linked_timeout(req); | |
5529 | ||
193155c8 JA |
5530 | if (req->work.creds && req->work.creds != current_cred()) { |
5531 | if (old_creds) | |
5532 | revert_creds(old_creds); | |
5533 | if (old_creds == req->work.creds) | |
5534 | old_creds = NULL; /* restored original creds */ | |
5535 | else | |
5536 | old_creds = override_creds(req->work.creds); | |
5537 | } | |
5538 | ||
014db007 | 5539 | ret = io_issue_sqe(req, sqe, true); |
491381ce JA |
5540 | |
5541 | /* | |
5542 | * We async punt it if the file wasn't marked NOWAIT, or if the file | |
5543 | * doesn't support non-blocking read/write attempts | |
5544 | */ | |
5545 | if (ret == -EAGAIN && (!(req->flags & REQ_F_NOWAIT) || | |
5546 | (req->flags & REQ_F_MUST_PUNT))) { | |
d7718a9d JA |
5547 | if (io_arm_poll_handler(req)) { |
5548 | if (linked_timeout) | |
5549 | io_queue_linked_timeout(linked_timeout); | |
4bc4494e | 5550 | goto exit; |
d7718a9d | 5551 | } |
86a761f8 | 5552 | punt: |
f86cd20c | 5553 | if (io_op_defs[req->opcode].file_table) { |
bbad27b2 PB |
5554 | ret = io_grab_files(req); |
5555 | if (ret) | |
5556 | goto err; | |
2b188cc1 | 5557 | } |
bbad27b2 PB |
5558 | |
5559 | /* | |
5560 | * Queued up for async execution, worker will release | |
5561 | * submit reference when the iocb is actually submitted. | |
5562 | */ | |
5563 | io_queue_async_work(req); | |
4bc4494e | 5564 | goto exit; |
2b188cc1 | 5565 | } |
e65ef56d | 5566 | |
fcb323cc | 5567 | err: |
4bc4494e | 5568 | nxt = NULL; |
76a46e06 | 5569 | /* drop submission reference */ |
2a44f467 | 5570 | io_put_req_find_next(req, &nxt); |
e65ef56d | 5571 | |
f9bd67f6 | 5572 | if (linked_timeout) { |
76a46e06 | 5573 | if (!ret) |
f9bd67f6 | 5574 | io_queue_linked_timeout(linked_timeout); |
76a46e06 | 5575 | else |
f9bd67f6 | 5576 | io_put_req(linked_timeout); |
76a46e06 JA |
5577 | } |
5578 | ||
e65ef56d | 5579 | /* and drop final reference, if we failed */ |
9e645e11 | 5580 | if (ret) { |
78e19bbe | 5581 | io_cqring_add_event(req, ret); |
4e88d6e7 | 5582 | req_set_fail_links(req); |
e65ef56d | 5583 | io_put_req(req); |
9e645e11 | 5584 | } |
4a0a7a18 JA |
5585 | if (nxt) { |
5586 | req = nxt; | |
86a761f8 PB |
5587 | |
5588 | if (req->flags & REQ_F_FORCE_ASYNC) | |
5589 | goto punt; | |
4a0a7a18 JA |
5590 | goto again; |
5591 | } | |
4bc4494e | 5592 | exit: |
193155c8 JA |
5593 | if (old_creds) |
5594 | revert_creds(old_creds); | |
2b188cc1 JA |
5595 | } |
5596 | ||
3529d8c2 | 5597 | static void io_queue_sqe(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
4fe2c963 JL |
5598 | { |
5599 | int ret; | |
5600 | ||
3529d8c2 | 5601 | ret = io_req_defer(req, sqe); |
4fe2c963 JL |
5602 | if (ret) { |
5603 | if (ret != -EIOCBQUEUED) { | |
1118591a | 5604 | fail_req: |
78e19bbe | 5605 | io_cqring_add_event(req, ret); |
4e88d6e7 | 5606 | req_set_fail_links(req); |
78e19bbe | 5607 | io_double_put_req(req); |
4fe2c963 | 5608 | } |
2550878f | 5609 | } else if (req->flags & REQ_F_FORCE_ASYNC) { |
1118591a PB |
5610 | ret = io_req_defer_prep(req, sqe); |
5611 | if (unlikely(ret < 0)) | |
5612 | goto fail_req; | |
ce35a47a JA |
5613 | /* |
5614 | * Never try inline submit of IOSQE_ASYNC is set, go straight | |
5615 | * to async execution. | |
5616 | */ | |
5617 | req->work.flags |= IO_WQ_WORK_CONCURRENT; | |
5618 | io_queue_async_work(req); | |
5619 | } else { | |
3529d8c2 | 5620 | __io_queue_sqe(req, sqe); |
ce35a47a | 5621 | } |
4fe2c963 JL |
5622 | } |
5623 | ||
1b4a51b6 | 5624 | static inline void io_queue_link_head(struct io_kiocb *req) |
4fe2c963 | 5625 | { |
94ae5e77 | 5626 | if (unlikely(req->flags & REQ_F_FAIL_LINK)) { |
1b4a51b6 PB |
5627 | io_cqring_add_event(req, -ECANCELED); |
5628 | io_double_put_req(req); | |
5629 | } else | |
3529d8c2 | 5630 | io_queue_sqe(req, NULL); |
4fe2c963 JL |
5631 | } |
5632 | ||
1d4240cc | 5633 | static int io_submit_sqe(struct io_kiocb *req, const struct io_uring_sqe *sqe, |
3529d8c2 | 5634 | struct io_submit_state *state, struct io_kiocb **link) |
9e645e11 | 5635 | { |
a197f664 | 5636 | struct io_ring_ctx *ctx = req->ctx; |
ef4ff581 | 5637 | int ret; |
9e645e11 | 5638 | |
9e645e11 JA |
5639 | /* |
5640 | * If we already have a head request, queue this one for async | |
5641 | * submittal once the head completes. If we don't have a head but | |
5642 | * IOSQE_IO_LINK is set in the sqe, start a new head. This one will be | |
5643 | * submitted sync once the chain is complete. If none of those | |
5644 | * conditions are true (normal request), then just queue it. | |
5645 | */ | |
5646 | if (*link) { | |
9d76377f | 5647 | struct io_kiocb *head = *link; |
4e88d6e7 | 5648 | |
8cdf2193 PB |
5649 | /* |
5650 | * Taking sequential execution of a link, draining both sides | |
5651 | * of the link also fullfils IOSQE_IO_DRAIN semantics for all | |
5652 | * requests in the link. So, it drains the head and the | |
5653 | * next after the link request. The last one is done via | |
5654 | * drain_next flag to persist the effect across calls. | |
5655 | */ | |
ef4ff581 | 5656 | if (req->flags & REQ_F_IO_DRAIN) { |
711be031 PB |
5657 | head->flags |= REQ_F_IO_DRAIN; |
5658 | ctx->drain_next = 1; | |
5659 | } | |
1d4240cc PB |
5660 | if (io_alloc_async_ctx(req)) |
5661 | return -EAGAIN; | |
9e645e11 | 5662 | |
3529d8c2 | 5663 | ret = io_req_defer_prep(req, sqe); |
2d28390a | 5664 | if (ret) { |
4e88d6e7 | 5665 | /* fail even hard links since we don't submit */ |
9d76377f | 5666 | head->flags |= REQ_F_FAIL_LINK; |
1d4240cc | 5667 | return ret; |
2d28390a | 5668 | } |
9d76377f PB |
5669 | trace_io_uring_link(ctx, req, head); |
5670 | list_add_tail(&req->link_list, &head->link_list); | |
32fe525b PB |
5671 | |
5672 | /* last request of a link, enqueue the link */ | |
ef4ff581 | 5673 | if (!(req->flags & (REQ_F_LINK | REQ_F_HARDLINK))) { |
32fe525b PB |
5674 | io_queue_link_head(head); |
5675 | *link = NULL; | |
5676 | } | |
9e645e11 | 5677 | } else { |
711be031 PB |
5678 | if (unlikely(ctx->drain_next)) { |
5679 | req->flags |= REQ_F_IO_DRAIN; | |
ef4ff581 | 5680 | ctx->drain_next = 0; |
711be031 | 5681 | } |
ef4ff581 | 5682 | if (req->flags & (REQ_F_LINK | REQ_F_HARDLINK)) { |
dea3b49c | 5683 | req->flags |= REQ_F_LINK_HEAD; |
711be031 | 5684 | INIT_LIST_HEAD(&req->link_list); |
f1d96a8f | 5685 | |
1d4240cc PB |
5686 | if (io_alloc_async_ctx(req)) |
5687 | return -EAGAIN; | |
5688 | ||
711be031 PB |
5689 | ret = io_req_defer_prep(req, sqe); |
5690 | if (ret) | |
5691 | req->flags |= REQ_F_FAIL_LINK; | |
5692 | *link = req; | |
5693 | } else { | |
5694 | io_queue_sqe(req, sqe); | |
5695 | } | |
9e645e11 | 5696 | } |
2e6e1fde | 5697 | |
1d4240cc | 5698 | return 0; |
9e645e11 JA |
5699 | } |
5700 | ||
9a56a232 JA |
5701 | /* |
5702 | * Batched submission is done, ensure local IO is flushed out. | |
5703 | */ | |
5704 | static void io_submit_state_end(struct io_submit_state *state) | |
5705 | { | |
5706 | blk_finish_plug(&state->plug); | |
3d6770fb | 5707 | io_file_put(state); |
2579f913 | 5708 | if (state->free_reqs) |
6c8a3134 | 5709 | kmem_cache_free_bulk(req_cachep, state->free_reqs, state->reqs); |
9a56a232 JA |
5710 | } |
5711 | ||
5712 | /* | |
5713 | * Start submission side cache. | |
5714 | */ | |
5715 | static void io_submit_state_start(struct io_submit_state *state, | |
22efde59 | 5716 | unsigned int max_ios) |
9a56a232 JA |
5717 | { |
5718 | blk_start_plug(&state->plug); | |
2579f913 | 5719 | state->free_reqs = 0; |
9a56a232 JA |
5720 | state->file = NULL; |
5721 | state->ios_left = max_ios; | |
5722 | } | |
5723 | ||
2b188cc1 JA |
5724 | static void io_commit_sqring(struct io_ring_ctx *ctx) |
5725 | { | |
75b28aff | 5726 | struct io_rings *rings = ctx->rings; |
2b188cc1 | 5727 | |
caf582c6 PB |
5728 | /* |
5729 | * Ensure any loads from the SQEs are done at this point, | |
5730 | * since once we write the new head, the application could | |
5731 | * write new data to them. | |
5732 | */ | |
5733 | smp_store_release(&rings->sq.head, ctx->cached_sq_head); | |
2b188cc1 JA |
5734 | } |
5735 | ||
2b188cc1 | 5736 | /* |
3529d8c2 | 5737 | * Fetch an sqe, if one is available. Note that sqe_ptr will point to memory |
2b188cc1 JA |
5738 | * that is mapped by userspace. This means that care needs to be taken to |
5739 | * ensure that reads are stable, as we cannot rely on userspace always | |
5740 | * being a good citizen. If members of the sqe are validated and then later | |
5741 | * used, it's important that those reads are done through READ_ONCE() to | |
5742 | * prevent a re-load down the line. | |
5743 | */ | |
709b302f | 5744 | static const struct io_uring_sqe *io_get_sqe(struct io_ring_ctx *ctx) |
2b188cc1 | 5745 | { |
75b28aff | 5746 | u32 *sq_array = ctx->sq_array; |
2b188cc1 JA |
5747 | unsigned head; |
5748 | ||
5749 | /* | |
5750 | * The cached sq head (or cq tail) serves two purposes: | |
5751 | * | |
5752 | * 1) allows us to batch the cost of updating the user visible | |
5753 | * head updates. | |
5754 | * 2) allows the kernel side to track the head on its own, even | |
5755 | * though the application is the one updating it. | |
5756 | */ | |
ee7d46d9 | 5757 | head = READ_ONCE(sq_array[ctx->cached_sq_head & ctx->sq_mask]); |
709b302f PB |
5758 | if (likely(head < ctx->sq_entries)) |
5759 | return &ctx->sq_sqes[head]; | |
2b188cc1 JA |
5760 | |
5761 | /* drop invalid entries */ | |
498ccd9e | 5762 | ctx->cached_sq_dropped++; |
ee7d46d9 | 5763 | WRITE_ONCE(ctx->rings->sq_dropped, ctx->cached_sq_dropped); |
709b302f PB |
5764 | return NULL; |
5765 | } | |
5766 | ||
5767 | static inline void io_consume_sqe(struct io_ring_ctx *ctx) | |
5768 | { | |
5769 | ctx->cached_sq_head++; | |
2b188cc1 JA |
5770 | } |
5771 | ||
ef4ff581 PB |
5772 | #define SQE_VALID_FLAGS (IOSQE_FIXED_FILE|IOSQE_IO_DRAIN|IOSQE_IO_LINK| \ |
5773 | IOSQE_IO_HARDLINK | IOSQE_ASYNC | \ | |
5774 | IOSQE_BUFFER_SELECT) | |
5775 | ||
5776 | static int io_init_req(struct io_ring_ctx *ctx, struct io_kiocb *req, | |
5777 | const struct io_uring_sqe *sqe, | |
5778 | struct io_submit_state *state, bool async) | |
0553b8bd | 5779 | { |
ef4ff581 | 5780 | unsigned int sqe_flags; |
63ff8223 | 5781 | int id; |
ef4ff581 | 5782 | |
0553b8bd PB |
5783 | /* |
5784 | * All io need record the previous position, if LINK vs DARIN, | |
5785 | * it can be used to mark the position of the first IO in the | |
5786 | * link list. | |
5787 | */ | |
31af27c7 | 5788 | req->sequence = ctx->cached_sq_head - ctx->cached_sq_dropped; |
0553b8bd PB |
5789 | req->opcode = READ_ONCE(sqe->opcode); |
5790 | req->user_data = READ_ONCE(sqe->user_data); | |
5791 | req->io = NULL; | |
5792 | req->file = NULL; | |
5793 | req->ctx = ctx; | |
5794 | req->flags = 0; | |
5795 | /* one is dropped after submission, the other at completion */ | |
5796 | refcount_set(&req->refs, 2); | |
5797 | req->task = NULL; | |
5798 | req->result = 0; | |
ef4ff581 | 5799 | req->needs_fixed_file = async; |
0553b8bd | 5800 | INIT_IO_WORK(&req->work, io_wq_submit_work); |
ef4ff581 PB |
5801 | |
5802 | if (unlikely(req->opcode >= IORING_OP_LAST)) | |
5803 | return -EINVAL; | |
5804 | ||
5805 | if (io_op_defs[req->opcode].needs_mm && !current->mm) { | |
5806 | if (unlikely(!mmget_not_zero(ctx->sqo_mm))) | |
5807 | return -EFAULT; | |
5808 | use_mm(ctx->sqo_mm); | |
5809 | } | |
5810 | ||
5811 | sqe_flags = READ_ONCE(sqe->flags); | |
5812 | /* enforce forwards compatibility on users */ | |
5813 | if (unlikely(sqe_flags & ~SQE_VALID_FLAGS)) | |
5814 | return -EINVAL; | |
5815 | ||
5816 | if ((sqe_flags & IOSQE_BUFFER_SELECT) && | |
5817 | !io_op_defs[req->opcode].buffer_select) | |
5818 | return -EOPNOTSUPP; | |
5819 | ||
5820 | id = READ_ONCE(sqe->personality); | |
5821 | if (id) { | |
5822 | req->work.creds = idr_find(&ctx->personality_idr, id); | |
5823 | if (unlikely(!req->work.creds)) | |
5824 | return -EINVAL; | |
5825 | get_cred(req->work.creds); | |
5826 | } | |
5827 | ||
5828 | /* same numerical values with corresponding REQ_F_*, safe to copy */ | |
5829 | req->flags |= sqe_flags & (IOSQE_IO_DRAIN | IOSQE_IO_HARDLINK | | |
5830 | IOSQE_ASYNC | IOSQE_FIXED_FILE | | |
5831 | IOSQE_BUFFER_SELECT | IOSQE_IO_LINK); | |
5832 | ||
63ff8223 JA |
5833 | if (!io_op_defs[req->opcode].needs_file) |
5834 | return 0; | |
5835 | ||
5836 | return io_req_set_file(state, req, READ_ONCE(sqe->fd)); | |
0553b8bd PB |
5837 | } |
5838 | ||
fb5ccc98 | 5839 | static int io_submit_sqes(struct io_ring_ctx *ctx, unsigned int nr, |
bf9c2f1c | 5840 | struct file *ring_file, int ring_fd, bool async) |
6c271ce2 JA |
5841 | { |
5842 | struct io_submit_state state, *statep = NULL; | |
9e645e11 | 5843 | struct io_kiocb *link = NULL; |
9e645e11 | 5844 | int i, submitted = 0; |
6c271ce2 | 5845 | |
c4a2ed72 | 5846 | /* if we have a backlog and couldn't flush it all, return BUSY */ |
ad3eb2c8 JA |
5847 | if (test_bit(0, &ctx->sq_check_overflow)) { |
5848 | if (!list_empty(&ctx->cq_overflow_list) && | |
5849 | !io_cqring_overflow_flush(ctx, false)) | |
5850 | return -EBUSY; | |
5851 | } | |
6c271ce2 | 5852 | |
ee7d46d9 PB |
5853 | /* make sure SQ entry isn't read before tail */ |
5854 | nr = min3(nr, ctx->sq_entries, io_sqring_entries(ctx)); | |
9ef4f124 | 5855 | |
2b85edfc PB |
5856 | if (!percpu_ref_tryget_many(&ctx->refs, nr)) |
5857 | return -EAGAIN; | |
6c271ce2 JA |
5858 | |
5859 | if (nr > IO_PLUG_THRESHOLD) { | |
22efde59 | 5860 | io_submit_state_start(&state, nr); |
6c271ce2 JA |
5861 | statep = &state; |
5862 | } | |
5863 | ||
b14cca0c PB |
5864 | ctx->ring_fd = ring_fd; |
5865 | ctx->ring_file = ring_file; | |
5866 | ||
6c271ce2 | 5867 | for (i = 0; i < nr; i++) { |
3529d8c2 | 5868 | const struct io_uring_sqe *sqe; |
196be95c | 5869 | struct io_kiocb *req; |
1cb1edb2 | 5870 | int err; |
fb5ccc98 | 5871 | |
b1e50e54 PB |
5872 | sqe = io_get_sqe(ctx); |
5873 | if (unlikely(!sqe)) { | |
5874 | io_consume_sqe(ctx); | |
5875 | break; | |
5876 | } | |
0553b8bd | 5877 | req = io_alloc_req(ctx, statep); |
196be95c PB |
5878 | if (unlikely(!req)) { |
5879 | if (!submitted) | |
5880 | submitted = -EAGAIN; | |
fb5ccc98 | 5881 | break; |
196be95c | 5882 | } |
fb5ccc98 | 5883 | |
ef4ff581 | 5884 | err = io_init_req(ctx, req, sqe, statep, async); |
709b302f | 5885 | io_consume_sqe(ctx); |
d3656344 JA |
5886 | /* will complete beyond this point, count as submitted */ |
5887 | submitted++; | |
5888 | ||
ef4ff581 | 5889 | if (unlikely(err)) { |
1cb1edb2 PB |
5890 | fail_req: |
5891 | io_cqring_add_event(req, err); | |
d3656344 | 5892 | io_double_put_req(req); |
196be95c PB |
5893 | break; |
5894 | } | |
fb5ccc98 | 5895 | |
354420f7 JA |
5896 | trace_io_uring_submit_sqe(ctx, req->opcode, req->user_data, |
5897 | true, async); | |
1d4240cc PB |
5898 | err = io_submit_sqe(req, sqe, statep, &link); |
5899 | if (err) | |
5900 | goto fail_req; | |
6c271ce2 JA |
5901 | } |
5902 | ||
9466f437 PB |
5903 | if (unlikely(submitted != nr)) { |
5904 | int ref_used = (submitted == -EAGAIN) ? 0 : submitted; | |
5905 | ||
5906 | percpu_ref_put_many(&ctx->refs, nr - ref_used); | |
5907 | } | |
9e645e11 | 5908 | if (link) |
1b4a51b6 | 5909 | io_queue_link_head(link); |
6c271ce2 JA |
5910 | if (statep) |
5911 | io_submit_state_end(&state); | |
5912 | ||
ae9428ca PB |
5913 | /* Commit SQ ring head once we've consumed and submitted all SQEs */ |
5914 | io_commit_sqring(ctx); | |
5915 | ||
6c271ce2 JA |
5916 | return submitted; |
5917 | } | |
5918 | ||
bf9c2f1c PB |
5919 | static inline void io_sq_thread_drop_mm(struct io_ring_ctx *ctx) |
5920 | { | |
5921 | struct mm_struct *mm = current->mm; | |
5922 | ||
5923 | if (mm) { | |
5924 | unuse_mm(mm); | |
5925 | mmput(mm); | |
5926 | } | |
5927 | } | |
5928 | ||
6c271ce2 JA |
5929 | static int io_sq_thread(void *data) |
5930 | { | |
6c271ce2 | 5931 | struct io_ring_ctx *ctx = data; |
181e448d | 5932 | const struct cred *old_cred; |
6c271ce2 JA |
5933 | mm_segment_t old_fs; |
5934 | DEFINE_WAIT(wait); | |
6c271ce2 | 5935 | unsigned long timeout; |
bdcd3eab | 5936 | int ret = 0; |
6c271ce2 | 5937 | |
206aefde | 5938 | complete(&ctx->completions[1]); |
a4c0b3de | 5939 | |
6c271ce2 JA |
5940 | old_fs = get_fs(); |
5941 | set_fs(USER_DS); | |
181e448d | 5942 | old_cred = override_creds(ctx->creds); |
6c271ce2 | 5943 | |
bdcd3eab | 5944 | timeout = jiffies + ctx->sq_thread_idle; |
2bbcd6d3 | 5945 | while (!kthread_should_park()) { |
fb5ccc98 | 5946 | unsigned int to_submit; |
6c271ce2 | 5947 | |
bdcd3eab | 5948 | if (!list_empty(&ctx->poll_list)) { |
6c271ce2 JA |
5949 | unsigned nr_events = 0; |
5950 | ||
bdcd3eab XW |
5951 | mutex_lock(&ctx->uring_lock); |
5952 | if (!list_empty(&ctx->poll_list)) | |
5953 | io_iopoll_getevents(ctx, &nr_events, 0); | |
5954 | else | |
6c271ce2 | 5955 | timeout = jiffies + ctx->sq_thread_idle; |
bdcd3eab | 5956 | mutex_unlock(&ctx->uring_lock); |
6c271ce2 JA |
5957 | } |
5958 | ||
fb5ccc98 | 5959 | to_submit = io_sqring_entries(ctx); |
c1edbf5f JA |
5960 | |
5961 | /* | |
5962 | * If submit got -EBUSY, flag us as needing the application | |
5963 | * to enter the kernel to reap and flush events. | |
5964 | */ | |
5965 | if (!to_submit || ret == -EBUSY) { | |
7143b5ac SG |
5966 | /* |
5967 | * Drop cur_mm before scheduling, we can't hold it for | |
5968 | * long periods (or over schedule()). Do this before | |
5969 | * adding ourselves to the waitqueue, as the unuse/drop | |
5970 | * may sleep. | |
5971 | */ | |
bf9c2f1c | 5972 | io_sq_thread_drop_mm(ctx); |
7143b5ac | 5973 | |
6c271ce2 JA |
5974 | /* |
5975 | * We're polling. If we're within the defined idle | |
5976 | * period, then let us spin without work before going | |
c1edbf5f JA |
5977 | * to sleep. The exception is if we got EBUSY doing |
5978 | * more IO, we should wait for the application to | |
5979 | * reap events and wake us up. | |
6c271ce2 | 5980 | */ |
bdcd3eab | 5981 | if (!list_empty(&ctx->poll_list) || |
df069d80 JA |
5982 | (!time_after(jiffies, timeout) && ret != -EBUSY && |
5983 | !percpu_ref_is_dying(&ctx->refs))) { | |
b41e9852 JA |
5984 | if (current->task_works) |
5985 | task_work_run(); | |
9831a90c | 5986 | cond_resched(); |
6c271ce2 JA |
5987 | continue; |
5988 | } | |
5989 | ||
6c271ce2 JA |
5990 | prepare_to_wait(&ctx->sqo_wait, &wait, |
5991 | TASK_INTERRUPTIBLE); | |
5992 | ||
bdcd3eab XW |
5993 | /* |
5994 | * While doing polled IO, before going to sleep, we need | |
5995 | * to check if there are new reqs added to poll_list, it | |
5996 | * is because reqs may have been punted to io worker and | |
5997 | * will be added to poll_list later, hence check the | |
5998 | * poll_list again. | |
5999 | */ | |
6000 | if ((ctx->flags & IORING_SETUP_IOPOLL) && | |
6001 | !list_empty_careful(&ctx->poll_list)) { | |
6002 | finish_wait(&ctx->sqo_wait, &wait); | |
6003 | continue; | |
6004 | } | |
6005 | ||
6c271ce2 | 6006 | /* Tell userspace we may need a wakeup call */ |
75b28aff | 6007 | ctx->rings->sq_flags |= IORING_SQ_NEED_WAKEUP; |
0d7bae69 SB |
6008 | /* make sure to read SQ tail after writing flags */ |
6009 | smp_mb(); | |
6c271ce2 | 6010 | |
fb5ccc98 | 6011 | to_submit = io_sqring_entries(ctx); |
c1edbf5f | 6012 | if (!to_submit || ret == -EBUSY) { |
2bbcd6d3 | 6013 | if (kthread_should_park()) { |
6c271ce2 JA |
6014 | finish_wait(&ctx->sqo_wait, &wait); |
6015 | break; | |
6016 | } | |
b41e9852 JA |
6017 | if (current->task_works) { |
6018 | task_work_run(); | |
10bea96d | 6019 | finish_wait(&ctx->sqo_wait, &wait); |
b41e9852 JA |
6020 | continue; |
6021 | } | |
6c271ce2 JA |
6022 | if (signal_pending(current)) |
6023 | flush_signals(current); | |
6024 | schedule(); | |
6025 | finish_wait(&ctx->sqo_wait, &wait); | |
6026 | ||
75b28aff | 6027 | ctx->rings->sq_flags &= ~IORING_SQ_NEED_WAKEUP; |
6c271ce2 JA |
6028 | continue; |
6029 | } | |
6030 | finish_wait(&ctx->sqo_wait, &wait); | |
6031 | ||
75b28aff | 6032 | ctx->rings->sq_flags &= ~IORING_SQ_NEED_WAKEUP; |
6c271ce2 JA |
6033 | } |
6034 | ||
8a4955ff | 6035 | mutex_lock(&ctx->uring_lock); |
bf9c2f1c | 6036 | ret = io_submit_sqes(ctx, to_submit, NULL, -1, true); |
8a4955ff | 6037 | mutex_unlock(&ctx->uring_lock); |
bdcd3eab | 6038 | timeout = jiffies + ctx->sq_thread_idle; |
6c271ce2 JA |
6039 | } |
6040 | ||
b41e9852 JA |
6041 | if (current->task_works) |
6042 | task_work_run(); | |
6043 | ||
6c271ce2 | 6044 | set_fs(old_fs); |
bf9c2f1c | 6045 | io_sq_thread_drop_mm(ctx); |
181e448d | 6046 | revert_creds(old_cred); |
06058632 | 6047 | |
2bbcd6d3 | 6048 | kthread_parkme(); |
06058632 | 6049 | |
6c271ce2 JA |
6050 | return 0; |
6051 | } | |
6052 | ||
bda52162 JA |
6053 | struct io_wait_queue { |
6054 | struct wait_queue_entry wq; | |
6055 | struct io_ring_ctx *ctx; | |
6056 | unsigned to_wait; | |
6057 | unsigned nr_timeouts; | |
6058 | }; | |
6059 | ||
1d7bb1d5 | 6060 | static inline bool io_should_wake(struct io_wait_queue *iowq, bool noflush) |
bda52162 JA |
6061 | { |
6062 | struct io_ring_ctx *ctx = iowq->ctx; | |
6063 | ||
6064 | /* | |
d195a66e | 6065 | * Wake up if we have enough events, or if a timeout occurred since we |
bda52162 JA |
6066 | * started waiting. For timeouts, we always want to return to userspace, |
6067 | * regardless of event count. | |
6068 | */ | |
1d7bb1d5 | 6069 | return io_cqring_events(ctx, noflush) >= iowq->to_wait || |
bda52162 JA |
6070 | atomic_read(&ctx->cq_timeouts) != iowq->nr_timeouts; |
6071 | } | |
6072 | ||
6073 | static int io_wake_function(struct wait_queue_entry *curr, unsigned int mode, | |
6074 | int wake_flags, void *key) | |
6075 | { | |
6076 | struct io_wait_queue *iowq = container_of(curr, struct io_wait_queue, | |
6077 | wq); | |
6078 | ||
1d7bb1d5 JA |
6079 | /* use noflush == true, as we can't safely rely on locking context */ |
6080 | if (!io_should_wake(iowq, true)) | |
bda52162 JA |
6081 | return -1; |
6082 | ||
6083 | return autoremove_wake_function(curr, mode, wake_flags, key); | |
6084 | } | |
6085 | ||
2b188cc1 JA |
6086 | /* |
6087 | * Wait until events become available, if we don't already have some. The | |
6088 | * application must reap them itself, as they reside on the shared cq ring. | |
6089 | */ | |
6090 | static int io_cqring_wait(struct io_ring_ctx *ctx, int min_events, | |
6091 | const sigset_t __user *sig, size_t sigsz) | |
6092 | { | |
bda52162 JA |
6093 | struct io_wait_queue iowq = { |
6094 | .wq = { | |
6095 | .private = current, | |
6096 | .func = io_wake_function, | |
6097 | .entry = LIST_HEAD_INIT(iowq.wq.entry), | |
6098 | }, | |
6099 | .ctx = ctx, | |
6100 | .to_wait = min_events, | |
6101 | }; | |
75b28aff | 6102 | struct io_rings *rings = ctx->rings; |
e9ffa5c2 | 6103 | int ret = 0; |
2b188cc1 | 6104 | |
b41e9852 JA |
6105 | do { |
6106 | if (io_cqring_events(ctx, false) >= min_events) | |
6107 | return 0; | |
6108 | if (!current->task_works) | |
6109 | break; | |
6110 | task_work_run(); | |
6111 | } while (1); | |
2b188cc1 JA |
6112 | |
6113 | if (sig) { | |
9e75ad5d AB |
6114 | #ifdef CONFIG_COMPAT |
6115 | if (in_compat_syscall()) | |
6116 | ret = set_compat_user_sigmask((const compat_sigset_t __user *)sig, | |
b772434b | 6117 | sigsz); |
9e75ad5d AB |
6118 | else |
6119 | #endif | |
b772434b | 6120 | ret = set_user_sigmask(sig, sigsz); |
9e75ad5d | 6121 | |
2b188cc1 JA |
6122 | if (ret) |
6123 | return ret; | |
6124 | } | |
6125 | ||
bda52162 | 6126 | iowq.nr_timeouts = atomic_read(&ctx->cq_timeouts); |
c826bd7a | 6127 | trace_io_uring_cqring_wait(ctx, min_events); |
bda52162 JA |
6128 | do { |
6129 | prepare_to_wait_exclusive(&ctx->wait, &iowq.wq, | |
6130 | TASK_INTERRUPTIBLE); | |
b41e9852 JA |
6131 | if (current->task_works) |
6132 | task_work_run(); | |
1d7bb1d5 | 6133 | if (io_should_wake(&iowq, false)) |
bda52162 JA |
6134 | break; |
6135 | schedule(); | |
6136 | if (signal_pending(current)) { | |
e9ffa5c2 | 6137 | ret = -EINTR; |
bda52162 JA |
6138 | break; |
6139 | } | |
6140 | } while (1); | |
6141 | finish_wait(&ctx->wait, &iowq.wq); | |
6142 | ||
e9ffa5c2 | 6143 | restore_saved_sigmask_unless(ret == -EINTR); |
2b188cc1 | 6144 | |
75b28aff | 6145 | return READ_ONCE(rings->cq.head) == READ_ONCE(rings->cq.tail) ? ret : 0; |
2b188cc1 JA |
6146 | } |
6147 | ||
6b06314c JA |
6148 | static void __io_sqe_files_unregister(struct io_ring_ctx *ctx) |
6149 | { | |
6150 | #if defined(CONFIG_UNIX) | |
6151 | if (ctx->ring_sock) { | |
6152 | struct sock *sock = ctx->ring_sock->sk; | |
6153 | struct sk_buff *skb; | |
6154 | ||
6155 | while ((skb = skb_dequeue(&sock->sk_receive_queue)) != NULL) | |
6156 | kfree_skb(skb); | |
6157 | } | |
6158 | #else | |
6159 | int i; | |
6160 | ||
65e19f54 JA |
6161 | for (i = 0; i < ctx->nr_user_files; i++) { |
6162 | struct file *file; | |
6163 | ||
6164 | file = io_file_from_index(ctx, i); | |
6165 | if (file) | |
6166 | fput(file); | |
6167 | } | |
6b06314c JA |
6168 | #endif |
6169 | } | |
6170 | ||
05f3fb3c JA |
6171 | static void io_file_ref_kill(struct percpu_ref *ref) |
6172 | { | |
6173 | struct fixed_file_data *data; | |
6174 | ||
6175 | data = container_of(ref, struct fixed_file_data, refs); | |
6176 | complete(&data->done); | |
6177 | } | |
6178 | ||
6b06314c JA |
6179 | static int io_sqe_files_unregister(struct io_ring_ctx *ctx) |
6180 | { | |
05f3fb3c | 6181 | struct fixed_file_data *data = ctx->file_data; |
05589553 | 6182 | struct fixed_file_ref_node *ref_node = NULL; |
65e19f54 | 6183 | unsigned nr_tables, i; |
05589553 | 6184 | unsigned long flags; |
65e19f54 | 6185 | |
05f3fb3c | 6186 | if (!data) |
6b06314c JA |
6187 | return -ENXIO; |
6188 | ||
05589553 XW |
6189 | spin_lock_irqsave(&data->lock, flags); |
6190 | if (!list_empty(&data->ref_list)) | |
6191 | ref_node = list_first_entry(&data->ref_list, | |
6192 | struct fixed_file_ref_node, node); | |
6193 | spin_unlock_irqrestore(&data->lock, flags); | |
6194 | if (ref_node) | |
6195 | percpu_ref_kill(&ref_node->refs); | |
6196 | ||
6197 | percpu_ref_kill(&data->refs); | |
6198 | ||
6199 | /* wait for all refs nodes to complete */ | |
2faf852d | 6200 | wait_for_completion(&data->done); |
05f3fb3c | 6201 | |
6b06314c | 6202 | __io_sqe_files_unregister(ctx); |
65e19f54 JA |
6203 | nr_tables = DIV_ROUND_UP(ctx->nr_user_files, IORING_MAX_FILES_TABLE); |
6204 | for (i = 0; i < nr_tables; i++) | |
05f3fb3c JA |
6205 | kfree(data->table[i].files); |
6206 | kfree(data->table); | |
05589553 XW |
6207 | percpu_ref_exit(&data->refs); |
6208 | kfree(data); | |
05f3fb3c | 6209 | ctx->file_data = NULL; |
6b06314c JA |
6210 | ctx->nr_user_files = 0; |
6211 | return 0; | |
6212 | } | |
6213 | ||
6c271ce2 JA |
6214 | static void io_sq_thread_stop(struct io_ring_ctx *ctx) |
6215 | { | |
6216 | if (ctx->sqo_thread) { | |
206aefde | 6217 | wait_for_completion(&ctx->completions[1]); |
2bbcd6d3 RP |
6218 | /* |
6219 | * The park is a bit of a work-around, without it we get | |
6220 | * warning spews on shutdown with SQPOLL set and affinity | |
6221 | * set to a single CPU. | |
6222 | */ | |
06058632 | 6223 | kthread_park(ctx->sqo_thread); |
6c271ce2 JA |
6224 | kthread_stop(ctx->sqo_thread); |
6225 | ctx->sqo_thread = NULL; | |
6226 | } | |
6227 | } | |
6228 | ||
6b06314c JA |
6229 | static void io_finish_async(struct io_ring_ctx *ctx) |
6230 | { | |
6c271ce2 JA |
6231 | io_sq_thread_stop(ctx); |
6232 | ||
561fb04a JA |
6233 | if (ctx->io_wq) { |
6234 | io_wq_destroy(ctx->io_wq); | |
6235 | ctx->io_wq = NULL; | |
6b06314c JA |
6236 | } |
6237 | } | |
6238 | ||
6239 | #if defined(CONFIG_UNIX) | |
6b06314c JA |
6240 | /* |
6241 | * Ensure the UNIX gc is aware of our file set, so we are certain that | |
6242 | * the io_uring can be safely unregistered on process exit, even if we have | |
6243 | * loops in the file referencing. | |
6244 | */ | |
6245 | static int __io_sqe_files_scm(struct io_ring_ctx *ctx, int nr, int offset) | |
6246 | { | |
6247 | struct sock *sk = ctx->ring_sock->sk; | |
6248 | struct scm_fp_list *fpl; | |
6249 | struct sk_buff *skb; | |
08a45173 | 6250 | int i, nr_files; |
6b06314c | 6251 | |
6b06314c JA |
6252 | fpl = kzalloc(sizeof(*fpl), GFP_KERNEL); |
6253 | if (!fpl) | |
6254 | return -ENOMEM; | |
6255 | ||
6256 | skb = alloc_skb(0, GFP_KERNEL); | |
6257 | if (!skb) { | |
6258 | kfree(fpl); | |
6259 | return -ENOMEM; | |
6260 | } | |
6261 | ||
6262 | skb->sk = sk; | |
6b06314c | 6263 | |
08a45173 | 6264 | nr_files = 0; |
6b06314c JA |
6265 | fpl->user = get_uid(ctx->user); |
6266 | for (i = 0; i < nr; i++) { | |
65e19f54 JA |
6267 | struct file *file = io_file_from_index(ctx, i + offset); |
6268 | ||
6269 | if (!file) | |
08a45173 | 6270 | continue; |
65e19f54 | 6271 | fpl->fp[nr_files] = get_file(file); |
08a45173 JA |
6272 | unix_inflight(fpl->user, fpl->fp[nr_files]); |
6273 | nr_files++; | |
6b06314c JA |
6274 | } |
6275 | ||
08a45173 JA |
6276 | if (nr_files) { |
6277 | fpl->max = SCM_MAX_FD; | |
6278 | fpl->count = nr_files; | |
6279 | UNIXCB(skb).fp = fpl; | |
05f3fb3c | 6280 | skb->destructor = unix_destruct_scm; |
08a45173 JA |
6281 | refcount_add(skb->truesize, &sk->sk_wmem_alloc); |
6282 | skb_queue_head(&sk->sk_receive_queue, skb); | |
6b06314c | 6283 | |
08a45173 JA |
6284 | for (i = 0; i < nr_files; i++) |
6285 | fput(fpl->fp[i]); | |
6286 | } else { | |
6287 | kfree_skb(skb); | |
6288 | kfree(fpl); | |
6289 | } | |
6b06314c JA |
6290 | |
6291 | return 0; | |
6292 | } | |
6293 | ||
6294 | /* | |
6295 | * If UNIX sockets are enabled, fd passing can cause a reference cycle which | |
6296 | * causes regular reference counting to break down. We rely on the UNIX | |
6297 | * garbage collection to take care of this problem for us. | |
6298 | */ | |
6299 | static int io_sqe_files_scm(struct io_ring_ctx *ctx) | |
6300 | { | |
6301 | unsigned left, total; | |
6302 | int ret = 0; | |
6303 | ||
6304 | total = 0; | |
6305 | left = ctx->nr_user_files; | |
6306 | while (left) { | |
6307 | unsigned this_files = min_t(unsigned, left, SCM_MAX_FD); | |
6b06314c JA |
6308 | |
6309 | ret = __io_sqe_files_scm(ctx, this_files, total); | |
6310 | if (ret) | |
6311 | break; | |
6312 | left -= this_files; | |
6313 | total += this_files; | |
6314 | } | |
6315 | ||
6316 | if (!ret) | |
6317 | return 0; | |
6318 | ||
6319 | while (total < ctx->nr_user_files) { | |
65e19f54 JA |
6320 | struct file *file = io_file_from_index(ctx, total); |
6321 | ||
6322 | if (file) | |
6323 | fput(file); | |
6b06314c JA |
6324 | total++; |
6325 | } | |
6326 | ||
6327 | return ret; | |
6328 | } | |
6329 | #else | |
6330 | static int io_sqe_files_scm(struct io_ring_ctx *ctx) | |
6331 | { | |
6332 | return 0; | |
6333 | } | |
6334 | #endif | |
6335 | ||
65e19f54 JA |
6336 | static int io_sqe_alloc_file_tables(struct io_ring_ctx *ctx, unsigned nr_tables, |
6337 | unsigned nr_files) | |
6338 | { | |
6339 | int i; | |
6340 | ||
6341 | for (i = 0; i < nr_tables; i++) { | |
05f3fb3c | 6342 | struct fixed_file_table *table = &ctx->file_data->table[i]; |
65e19f54 JA |
6343 | unsigned this_files; |
6344 | ||
6345 | this_files = min(nr_files, IORING_MAX_FILES_TABLE); | |
6346 | table->files = kcalloc(this_files, sizeof(struct file *), | |
6347 | GFP_KERNEL); | |
6348 | if (!table->files) | |
6349 | break; | |
6350 | nr_files -= this_files; | |
6351 | } | |
6352 | ||
6353 | if (i == nr_tables) | |
6354 | return 0; | |
6355 | ||
6356 | for (i = 0; i < nr_tables; i++) { | |
05f3fb3c | 6357 | struct fixed_file_table *table = &ctx->file_data->table[i]; |
65e19f54 JA |
6358 | kfree(table->files); |
6359 | } | |
6360 | return 1; | |
6361 | } | |
6362 | ||
05f3fb3c JA |
6363 | static void io_ring_file_put(struct io_ring_ctx *ctx, struct file *file) |
6364 | { | |
6365 | #if defined(CONFIG_UNIX) | |
6366 | struct sock *sock = ctx->ring_sock->sk; | |
6367 | struct sk_buff_head list, *head = &sock->sk_receive_queue; | |
6368 | struct sk_buff *skb; | |
6369 | int i; | |
6370 | ||
6371 | __skb_queue_head_init(&list); | |
6372 | ||
6373 | /* | |
6374 | * Find the skb that holds this file in its SCM_RIGHTS. When found, | |
6375 | * remove this entry and rearrange the file array. | |
6376 | */ | |
6377 | skb = skb_dequeue(head); | |
6378 | while (skb) { | |
6379 | struct scm_fp_list *fp; | |
6380 | ||
6381 | fp = UNIXCB(skb).fp; | |
6382 | for (i = 0; i < fp->count; i++) { | |
6383 | int left; | |
6384 | ||
6385 | if (fp->fp[i] != file) | |
6386 | continue; | |
6387 | ||
6388 | unix_notinflight(fp->user, fp->fp[i]); | |
6389 | left = fp->count - 1 - i; | |
6390 | if (left) { | |
6391 | memmove(&fp->fp[i], &fp->fp[i + 1], | |
6392 | left * sizeof(struct file *)); | |
6393 | } | |
6394 | fp->count--; | |
6395 | if (!fp->count) { | |
6396 | kfree_skb(skb); | |
6397 | skb = NULL; | |
6398 | } else { | |
6399 | __skb_queue_tail(&list, skb); | |
6400 | } | |
6401 | fput(file); | |
6402 | file = NULL; | |
6403 | break; | |
6404 | } | |
6405 | ||
6406 | if (!file) | |
6407 | break; | |
6408 | ||
6409 | __skb_queue_tail(&list, skb); | |
6410 | ||
6411 | skb = skb_dequeue(head); | |
6412 | } | |
6413 | ||
6414 | if (skb_peek(&list)) { | |
6415 | spin_lock_irq(&head->lock); | |
6416 | while ((skb = __skb_dequeue(&list)) != NULL) | |
6417 | __skb_queue_tail(head, skb); | |
6418 | spin_unlock_irq(&head->lock); | |
6419 | } | |
6420 | #else | |
6421 | fput(file); | |
6422 | #endif | |
6423 | } | |
6424 | ||
6425 | struct io_file_put { | |
05589553 | 6426 | struct list_head list; |
05f3fb3c | 6427 | struct file *file; |
05f3fb3c JA |
6428 | }; |
6429 | ||
05589553 | 6430 | static void io_file_put_work(struct work_struct *work) |
65e19f54 | 6431 | { |
05589553 XW |
6432 | struct fixed_file_ref_node *ref_node; |
6433 | struct fixed_file_data *file_data; | |
6434 | struct io_ring_ctx *ctx; | |
05f3fb3c | 6435 | struct io_file_put *pfile, *tmp; |
05589553 | 6436 | unsigned long flags; |
65e19f54 | 6437 | |
05589553 XW |
6438 | ref_node = container_of(work, struct fixed_file_ref_node, work); |
6439 | file_data = ref_node->file_data; | |
6440 | ctx = file_data->ctx; | |
6441 | ||
6442 | list_for_each_entry_safe(pfile, tmp, &ref_node->file_list, list) { | |
6443 | list_del_init(&pfile->list); | |
6444 | io_ring_file_put(ctx, pfile->file); | |
6445 | kfree(pfile); | |
65e19f54 | 6446 | } |
05589553 XW |
6447 | |
6448 | spin_lock_irqsave(&file_data->lock, flags); | |
6449 | list_del_init(&ref_node->node); | |
6450 | spin_unlock_irqrestore(&file_data->lock, flags); | |
6451 | ||
6452 | percpu_ref_exit(&ref_node->refs); | |
6453 | kfree(ref_node); | |
6454 | percpu_ref_put(&file_data->refs); | |
2faf852d | 6455 | } |
65e19f54 | 6456 | |
05589553 | 6457 | static void io_file_data_ref_zero(struct percpu_ref *ref) |
2faf852d | 6458 | { |
05589553 | 6459 | struct fixed_file_ref_node *ref_node; |
65e19f54 | 6460 | |
05589553 XW |
6461 | ref_node = container_of(ref, struct fixed_file_ref_node, refs); |
6462 | ||
6463 | queue_work(system_wq, &ref_node->work); | |
05f3fb3c | 6464 | } |
65e19f54 | 6465 | |
05589553 XW |
6466 | static struct fixed_file_ref_node *alloc_fixed_file_ref_node( |
6467 | struct io_ring_ctx *ctx) | |
05f3fb3c | 6468 | { |
05589553 | 6469 | struct fixed_file_ref_node *ref_node; |
05f3fb3c | 6470 | |
05589553 XW |
6471 | ref_node = kzalloc(sizeof(*ref_node), GFP_KERNEL); |
6472 | if (!ref_node) | |
6473 | return ERR_PTR(-ENOMEM); | |
05f3fb3c | 6474 | |
05589553 XW |
6475 | if (percpu_ref_init(&ref_node->refs, io_file_data_ref_zero, |
6476 | 0, GFP_KERNEL)) { | |
6477 | kfree(ref_node); | |
6478 | return ERR_PTR(-ENOMEM); | |
6479 | } | |
6480 | INIT_LIST_HEAD(&ref_node->node); | |
6481 | INIT_LIST_HEAD(&ref_node->file_list); | |
6482 | INIT_WORK(&ref_node->work, io_file_put_work); | |
6483 | ref_node->file_data = ctx->file_data; | |
6484 | return ref_node; | |
6485 | ||
6486 | } | |
6487 | ||
6488 | static void destroy_fixed_file_ref_node(struct fixed_file_ref_node *ref_node) | |
6489 | { | |
6490 | percpu_ref_exit(&ref_node->refs); | |
6491 | kfree(ref_node); | |
65e19f54 JA |
6492 | } |
6493 | ||
6b06314c JA |
6494 | static int io_sqe_files_register(struct io_ring_ctx *ctx, void __user *arg, |
6495 | unsigned nr_args) | |
6496 | { | |
6497 | __s32 __user *fds = (__s32 __user *) arg; | |
65e19f54 | 6498 | unsigned nr_tables; |
05f3fb3c | 6499 | struct file *file; |
6b06314c JA |
6500 | int fd, ret = 0; |
6501 | unsigned i; | |
05589553 XW |
6502 | struct fixed_file_ref_node *ref_node; |
6503 | unsigned long flags; | |
6b06314c | 6504 | |
05f3fb3c | 6505 | if (ctx->file_data) |
6b06314c JA |
6506 | return -EBUSY; |
6507 | if (!nr_args) | |
6508 | return -EINVAL; | |
6509 | if (nr_args > IORING_MAX_FIXED_FILES) | |
6510 | return -EMFILE; | |
6511 | ||
05f3fb3c JA |
6512 | ctx->file_data = kzalloc(sizeof(*ctx->file_data), GFP_KERNEL); |
6513 | if (!ctx->file_data) | |
6514 | return -ENOMEM; | |
6515 | ctx->file_data->ctx = ctx; | |
6516 | init_completion(&ctx->file_data->done); | |
05589553 | 6517 | INIT_LIST_HEAD(&ctx->file_data->ref_list); |
f7fe9346 | 6518 | spin_lock_init(&ctx->file_data->lock); |
05f3fb3c | 6519 | |
65e19f54 | 6520 | nr_tables = DIV_ROUND_UP(nr_args, IORING_MAX_FILES_TABLE); |
05f3fb3c JA |
6521 | ctx->file_data->table = kcalloc(nr_tables, |
6522 | sizeof(struct fixed_file_table), | |
65e19f54 | 6523 | GFP_KERNEL); |
05f3fb3c JA |
6524 | if (!ctx->file_data->table) { |
6525 | kfree(ctx->file_data); | |
6526 | ctx->file_data = NULL; | |
6b06314c | 6527 | return -ENOMEM; |
05f3fb3c JA |
6528 | } |
6529 | ||
05589553 | 6530 | if (percpu_ref_init(&ctx->file_data->refs, io_file_ref_kill, |
05f3fb3c JA |
6531 | PERCPU_REF_ALLOW_REINIT, GFP_KERNEL)) { |
6532 | kfree(ctx->file_data->table); | |
6533 | kfree(ctx->file_data); | |
6534 | ctx->file_data = NULL; | |
6b06314c | 6535 | return -ENOMEM; |
05f3fb3c | 6536 | } |
6b06314c | 6537 | |
65e19f54 | 6538 | if (io_sqe_alloc_file_tables(ctx, nr_tables, nr_args)) { |
05f3fb3c JA |
6539 | percpu_ref_exit(&ctx->file_data->refs); |
6540 | kfree(ctx->file_data->table); | |
6541 | kfree(ctx->file_data); | |
6542 | ctx->file_data = NULL; | |
65e19f54 JA |
6543 | return -ENOMEM; |
6544 | } | |
6545 | ||
08a45173 | 6546 | for (i = 0; i < nr_args; i++, ctx->nr_user_files++) { |
65e19f54 JA |
6547 | struct fixed_file_table *table; |
6548 | unsigned index; | |
6549 | ||
6b06314c JA |
6550 | ret = -EFAULT; |
6551 | if (copy_from_user(&fd, &fds[i], sizeof(fd))) | |
6552 | break; | |
08a45173 JA |
6553 | /* allow sparse sets */ |
6554 | if (fd == -1) { | |
6555 | ret = 0; | |
6556 | continue; | |
6557 | } | |
6b06314c | 6558 | |
05f3fb3c | 6559 | table = &ctx->file_data->table[i >> IORING_FILE_TABLE_SHIFT]; |
65e19f54 | 6560 | index = i & IORING_FILE_TABLE_MASK; |
05f3fb3c | 6561 | file = fget(fd); |
6b06314c JA |
6562 | |
6563 | ret = -EBADF; | |
05f3fb3c | 6564 | if (!file) |
6b06314c | 6565 | break; |
05f3fb3c | 6566 | |
6b06314c JA |
6567 | /* |
6568 | * Don't allow io_uring instances to be registered. If UNIX | |
6569 | * isn't enabled, then this causes a reference cycle and this | |
6570 | * instance can never get freed. If UNIX is enabled we'll | |
6571 | * handle it just fine, but there's still no point in allowing | |
6572 | * a ring fd as it doesn't support regular read/write anyway. | |
6573 | */ | |
05f3fb3c JA |
6574 | if (file->f_op == &io_uring_fops) { |
6575 | fput(file); | |
6b06314c JA |
6576 | break; |
6577 | } | |
6b06314c | 6578 | ret = 0; |
05f3fb3c | 6579 | table->files[index] = file; |
6b06314c JA |
6580 | } |
6581 | ||
6582 | if (ret) { | |
65e19f54 | 6583 | for (i = 0; i < ctx->nr_user_files; i++) { |
65e19f54 JA |
6584 | file = io_file_from_index(ctx, i); |
6585 | if (file) | |
6586 | fput(file); | |
6587 | } | |
6588 | for (i = 0; i < nr_tables; i++) | |
05f3fb3c | 6589 | kfree(ctx->file_data->table[i].files); |
6b06314c | 6590 | |
05f3fb3c JA |
6591 | kfree(ctx->file_data->table); |
6592 | kfree(ctx->file_data); | |
6593 | ctx->file_data = NULL; | |
6b06314c JA |
6594 | ctx->nr_user_files = 0; |
6595 | return ret; | |
6596 | } | |
6597 | ||
6598 | ret = io_sqe_files_scm(ctx); | |
05589553 | 6599 | if (ret) { |
6b06314c | 6600 | io_sqe_files_unregister(ctx); |
05589553 XW |
6601 | return ret; |
6602 | } | |
6b06314c | 6603 | |
05589553 XW |
6604 | ref_node = alloc_fixed_file_ref_node(ctx); |
6605 | if (IS_ERR(ref_node)) { | |
6606 | io_sqe_files_unregister(ctx); | |
6607 | return PTR_ERR(ref_node); | |
6608 | } | |
6609 | ||
6610 | ctx->file_data->cur_refs = &ref_node->refs; | |
6611 | spin_lock_irqsave(&ctx->file_data->lock, flags); | |
6612 | list_add(&ref_node->node, &ctx->file_data->ref_list); | |
6613 | spin_unlock_irqrestore(&ctx->file_data->lock, flags); | |
6614 | percpu_ref_get(&ctx->file_data->refs); | |
6b06314c JA |
6615 | return ret; |
6616 | } | |
6617 | ||
c3a31e60 JA |
6618 | static int io_sqe_file_register(struct io_ring_ctx *ctx, struct file *file, |
6619 | int index) | |
6620 | { | |
6621 | #if defined(CONFIG_UNIX) | |
6622 | struct sock *sock = ctx->ring_sock->sk; | |
6623 | struct sk_buff_head *head = &sock->sk_receive_queue; | |
6624 | struct sk_buff *skb; | |
6625 | ||
6626 | /* | |
6627 | * See if we can merge this file into an existing skb SCM_RIGHTS | |
6628 | * file set. If there's no room, fall back to allocating a new skb | |
6629 | * and filling it in. | |
6630 | */ | |
6631 | spin_lock_irq(&head->lock); | |
6632 | skb = skb_peek(head); | |
6633 | if (skb) { | |
6634 | struct scm_fp_list *fpl = UNIXCB(skb).fp; | |
6635 | ||
6636 | if (fpl->count < SCM_MAX_FD) { | |
6637 | __skb_unlink(skb, head); | |
6638 | spin_unlock_irq(&head->lock); | |
6639 | fpl->fp[fpl->count] = get_file(file); | |
6640 | unix_inflight(fpl->user, fpl->fp[fpl->count]); | |
6641 | fpl->count++; | |
6642 | spin_lock_irq(&head->lock); | |
6643 | __skb_queue_head(head, skb); | |
6644 | } else { | |
6645 | skb = NULL; | |
6646 | } | |
6647 | } | |
6648 | spin_unlock_irq(&head->lock); | |
6649 | ||
6650 | if (skb) { | |
6651 | fput(file); | |
6652 | return 0; | |
6653 | } | |
6654 | ||
6655 | return __io_sqe_files_scm(ctx, 1, index); | |
6656 | #else | |
6657 | return 0; | |
6658 | #endif | |
6659 | } | |
6660 | ||
a5318d3c | 6661 | static int io_queue_file_removal(struct fixed_file_data *data, |
05589553 | 6662 | struct file *file) |
05f3fb3c | 6663 | { |
a5318d3c | 6664 | struct io_file_put *pfile; |
05589553 XW |
6665 | struct percpu_ref *refs = data->cur_refs; |
6666 | struct fixed_file_ref_node *ref_node; | |
05f3fb3c | 6667 | |
05f3fb3c | 6668 | pfile = kzalloc(sizeof(*pfile), GFP_KERNEL); |
a5318d3c HD |
6669 | if (!pfile) |
6670 | return -ENOMEM; | |
05f3fb3c | 6671 | |
05589553 | 6672 | ref_node = container_of(refs, struct fixed_file_ref_node, refs); |
05f3fb3c | 6673 | pfile->file = file; |
05589553 XW |
6674 | list_add(&pfile->list, &ref_node->file_list); |
6675 | ||
a5318d3c | 6676 | return 0; |
05f3fb3c JA |
6677 | } |
6678 | ||
6679 | static int __io_sqe_files_update(struct io_ring_ctx *ctx, | |
6680 | struct io_uring_files_update *up, | |
6681 | unsigned nr_args) | |
6682 | { | |
6683 | struct fixed_file_data *data = ctx->file_data; | |
05589553 | 6684 | struct fixed_file_ref_node *ref_node; |
05f3fb3c | 6685 | struct file *file; |
c3a31e60 JA |
6686 | __s32 __user *fds; |
6687 | int fd, i, err; | |
6688 | __u32 done; | |
05589553 XW |
6689 | unsigned long flags; |
6690 | bool needs_switch = false; | |
c3a31e60 | 6691 | |
05f3fb3c | 6692 | if (check_add_overflow(up->offset, nr_args, &done)) |
c3a31e60 JA |
6693 | return -EOVERFLOW; |
6694 | if (done > ctx->nr_user_files) | |
6695 | return -EINVAL; | |
6696 | ||
05589553 XW |
6697 | ref_node = alloc_fixed_file_ref_node(ctx); |
6698 | if (IS_ERR(ref_node)) | |
6699 | return PTR_ERR(ref_node); | |
6700 | ||
c3a31e60 | 6701 | done = 0; |
05f3fb3c | 6702 | fds = u64_to_user_ptr(up->fds); |
c3a31e60 | 6703 | while (nr_args) { |
65e19f54 JA |
6704 | struct fixed_file_table *table; |
6705 | unsigned index; | |
6706 | ||
c3a31e60 JA |
6707 | err = 0; |
6708 | if (copy_from_user(&fd, &fds[done], sizeof(fd))) { | |
6709 | err = -EFAULT; | |
6710 | break; | |
6711 | } | |
05f3fb3c JA |
6712 | i = array_index_nospec(up->offset, ctx->nr_user_files); |
6713 | table = &ctx->file_data->table[i >> IORING_FILE_TABLE_SHIFT]; | |
65e19f54 JA |
6714 | index = i & IORING_FILE_TABLE_MASK; |
6715 | if (table->files[index]) { | |
05f3fb3c | 6716 | file = io_file_from_index(ctx, index); |
a5318d3c HD |
6717 | err = io_queue_file_removal(data, file); |
6718 | if (err) | |
6719 | break; | |
65e19f54 | 6720 | table->files[index] = NULL; |
05589553 | 6721 | needs_switch = true; |
c3a31e60 JA |
6722 | } |
6723 | if (fd != -1) { | |
c3a31e60 JA |
6724 | file = fget(fd); |
6725 | if (!file) { | |
6726 | err = -EBADF; | |
6727 | break; | |
6728 | } | |
6729 | /* | |
6730 | * Don't allow io_uring instances to be registered. If | |
6731 | * UNIX isn't enabled, then this causes a reference | |
6732 | * cycle and this instance can never get freed. If UNIX | |
6733 | * is enabled we'll handle it just fine, but there's | |
6734 | * still no point in allowing a ring fd as it doesn't | |
6735 | * support regular read/write anyway. | |
6736 | */ | |
6737 | if (file->f_op == &io_uring_fops) { | |
6738 | fput(file); | |
6739 | err = -EBADF; | |
6740 | break; | |
6741 | } | |
65e19f54 | 6742 | table->files[index] = file; |
c3a31e60 JA |
6743 | err = io_sqe_file_register(ctx, file, i); |
6744 | if (err) | |
6745 | break; | |
6746 | } | |
6747 | nr_args--; | |
6748 | done++; | |
05f3fb3c JA |
6749 | up->offset++; |
6750 | } | |
6751 | ||
05589553 XW |
6752 | if (needs_switch) { |
6753 | percpu_ref_kill(data->cur_refs); | |
6754 | spin_lock_irqsave(&data->lock, flags); | |
6755 | list_add(&ref_node->node, &data->ref_list); | |
6756 | data->cur_refs = &ref_node->refs; | |
6757 | spin_unlock_irqrestore(&data->lock, flags); | |
6758 | percpu_ref_get(&ctx->file_data->refs); | |
6759 | } else | |
6760 | destroy_fixed_file_ref_node(ref_node); | |
c3a31e60 JA |
6761 | |
6762 | return done ? done : err; | |
6763 | } | |
05589553 | 6764 | |
05f3fb3c JA |
6765 | static int io_sqe_files_update(struct io_ring_ctx *ctx, void __user *arg, |
6766 | unsigned nr_args) | |
6767 | { | |
6768 | struct io_uring_files_update up; | |
6769 | ||
6770 | if (!ctx->file_data) | |
6771 | return -ENXIO; | |
6772 | if (!nr_args) | |
6773 | return -EINVAL; | |
6774 | if (copy_from_user(&up, arg, sizeof(up))) | |
6775 | return -EFAULT; | |
6776 | if (up.resv) | |
6777 | return -EINVAL; | |
6778 | ||
6779 | return __io_sqe_files_update(ctx, &up, nr_args); | |
6780 | } | |
c3a31e60 | 6781 | |
e9fd9396 | 6782 | static void io_free_work(struct io_wq_work *work) |
7d723065 JA |
6783 | { |
6784 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); | |
6785 | ||
e9fd9396 | 6786 | /* Consider that io_steal_work() relies on this ref */ |
7d723065 JA |
6787 | io_put_req(req); |
6788 | } | |
6789 | ||
24369c2e PB |
6790 | static int io_init_wq_offload(struct io_ring_ctx *ctx, |
6791 | struct io_uring_params *p) | |
6792 | { | |
6793 | struct io_wq_data data; | |
6794 | struct fd f; | |
6795 | struct io_ring_ctx *ctx_attach; | |
6796 | unsigned int concurrency; | |
6797 | int ret = 0; | |
6798 | ||
6799 | data.user = ctx->user; | |
e9fd9396 | 6800 | data.free_work = io_free_work; |
24369c2e PB |
6801 | |
6802 | if (!(p->flags & IORING_SETUP_ATTACH_WQ)) { | |
6803 | /* Do QD, or 4 * CPUS, whatever is smallest */ | |
6804 | concurrency = min(ctx->sq_entries, 4 * num_online_cpus()); | |
6805 | ||
6806 | ctx->io_wq = io_wq_create(concurrency, &data); | |
6807 | if (IS_ERR(ctx->io_wq)) { | |
6808 | ret = PTR_ERR(ctx->io_wq); | |
6809 | ctx->io_wq = NULL; | |
6810 | } | |
6811 | return ret; | |
6812 | } | |
6813 | ||
6814 | f = fdget(p->wq_fd); | |
6815 | if (!f.file) | |
6816 | return -EBADF; | |
6817 | ||
6818 | if (f.file->f_op != &io_uring_fops) { | |
6819 | ret = -EINVAL; | |
6820 | goto out_fput; | |
6821 | } | |
6822 | ||
6823 | ctx_attach = f.file->private_data; | |
6824 | /* @io_wq is protected by holding the fd */ | |
6825 | if (!io_wq_get(ctx_attach->io_wq, &data)) { | |
6826 | ret = -EINVAL; | |
6827 | goto out_fput; | |
6828 | } | |
6829 | ||
6830 | ctx->io_wq = ctx_attach->io_wq; | |
6831 | out_fput: | |
6832 | fdput(f); | |
6833 | return ret; | |
6834 | } | |
6835 | ||
6c271ce2 JA |
6836 | static int io_sq_offload_start(struct io_ring_ctx *ctx, |
6837 | struct io_uring_params *p) | |
2b188cc1 JA |
6838 | { |
6839 | int ret; | |
6840 | ||
6c271ce2 | 6841 | init_waitqueue_head(&ctx->sqo_wait); |
2b188cc1 JA |
6842 | mmgrab(current->mm); |
6843 | ctx->sqo_mm = current->mm; | |
6844 | ||
6c271ce2 | 6845 | if (ctx->flags & IORING_SETUP_SQPOLL) { |
3ec482d1 JA |
6846 | ret = -EPERM; |
6847 | if (!capable(CAP_SYS_ADMIN)) | |
6848 | goto err; | |
6849 | ||
917257da JA |
6850 | ctx->sq_thread_idle = msecs_to_jiffies(p->sq_thread_idle); |
6851 | if (!ctx->sq_thread_idle) | |
6852 | ctx->sq_thread_idle = HZ; | |
6853 | ||
6c271ce2 | 6854 | if (p->flags & IORING_SETUP_SQ_AFF) { |
44a9bd18 | 6855 | int cpu = p->sq_thread_cpu; |
6c271ce2 | 6856 | |
917257da | 6857 | ret = -EINVAL; |
44a9bd18 JA |
6858 | if (cpu >= nr_cpu_ids) |
6859 | goto err; | |
7889f44d | 6860 | if (!cpu_online(cpu)) |
917257da JA |
6861 | goto err; |
6862 | ||
6c271ce2 JA |
6863 | ctx->sqo_thread = kthread_create_on_cpu(io_sq_thread, |
6864 | ctx, cpu, | |
6865 | "io_uring-sq"); | |
6866 | } else { | |
6867 | ctx->sqo_thread = kthread_create(io_sq_thread, ctx, | |
6868 | "io_uring-sq"); | |
6869 | } | |
6870 | if (IS_ERR(ctx->sqo_thread)) { | |
6871 | ret = PTR_ERR(ctx->sqo_thread); | |
6872 | ctx->sqo_thread = NULL; | |
6873 | goto err; | |
6874 | } | |
6875 | wake_up_process(ctx->sqo_thread); | |
6876 | } else if (p->flags & IORING_SETUP_SQ_AFF) { | |
6877 | /* Can't have SQ_AFF without SQPOLL */ | |
6878 | ret = -EINVAL; | |
6879 | goto err; | |
6880 | } | |
6881 | ||
24369c2e PB |
6882 | ret = io_init_wq_offload(ctx, p); |
6883 | if (ret) | |
2b188cc1 | 6884 | goto err; |
2b188cc1 JA |
6885 | |
6886 | return 0; | |
6887 | err: | |
54a91f3b | 6888 | io_finish_async(ctx); |
2b188cc1 JA |
6889 | mmdrop(ctx->sqo_mm); |
6890 | ctx->sqo_mm = NULL; | |
6891 | return ret; | |
6892 | } | |
6893 | ||
6894 | static void io_unaccount_mem(struct user_struct *user, unsigned long nr_pages) | |
6895 | { | |
6896 | atomic_long_sub(nr_pages, &user->locked_vm); | |
6897 | } | |
6898 | ||
6899 | static int io_account_mem(struct user_struct *user, unsigned long nr_pages) | |
6900 | { | |
6901 | unsigned long page_limit, cur_pages, new_pages; | |
6902 | ||
6903 | /* Don't allow more pages than we can safely lock */ | |
6904 | page_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT; | |
6905 | ||
6906 | do { | |
6907 | cur_pages = atomic_long_read(&user->locked_vm); | |
6908 | new_pages = cur_pages + nr_pages; | |
6909 | if (new_pages > page_limit) | |
6910 | return -ENOMEM; | |
6911 | } while (atomic_long_cmpxchg(&user->locked_vm, cur_pages, | |
6912 | new_pages) != cur_pages); | |
6913 | ||
6914 | return 0; | |
6915 | } | |
6916 | ||
6917 | static void io_mem_free(void *ptr) | |
6918 | { | |
52e04ef4 MR |
6919 | struct page *page; |
6920 | ||
6921 | if (!ptr) | |
6922 | return; | |
2b188cc1 | 6923 | |
52e04ef4 | 6924 | page = virt_to_head_page(ptr); |
2b188cc1 JA |
6925 | if (put_page_testzero(page)) |
6926 | free_compound_page(page); | |
6927 | } | |
6928 | ||
6929 | static void *io_mem_alloc(size_t size) | |
6930 | { | |
6931 | gfp_t gfp_flags = GFP_KERNEL | __GFP_ZERO | __GFP_NOWARN | __GFP_COMP | | |
6932 | __GFP_NORETRY; | |
6933 | ||
6934 | return (void *) __get_free_pages(gfp_flags, get_order(size)); | |
6935 | } | |
6936 | ||
75b28aff HV |
6937 | static unsigned long rings_size(unsigned sq_entries, unsigned cq_entries, |
6938 | size_t *sq_offset) | |
6939 | { | |
6940 | struct io_rings *rings; | |
6941 | size_t off, sq_array_size; | |
6942 | ||
6943 | off = struct_size(rings, cqes, cq_entries); | |
6944 | if (off == SIZE_MAX) | |
6945 | return SIZE_MAX; | |
6946 | ||
6947 | #ifdef CONFIG_SMP | |
6948 | off = ALIGN(off, SMP_CACHE_BYTES); | |
6949 | if (off == 0) | |
6950 | return SIZE_MAX; | |
6951 | #endif | |
6952 | ||
6953 | sq_array_size = array_size(sizeof(u32), sq_entries); | |
6954 | if (sq_array_size == SIZE_MAX) | |
6955 | return SIZE_MAX; | |
6956 | ||
6957 | if (check_add_overflow(off, sq_array_size, &off)) | |
6958 | return SIZE_MAX; | |
6959 | ||
6960 | if (sq_offset) | |
6961 | *sq_offset = off; | |
6962 | ||
6963 | return off; | |
6964 | } | |
6965 | ||
2b188cc1 JA |
6966 | static unsigned long ring_pages(unsigned sq_entries, unsigned cq_entries) |
6967 | { | |
75b28aff | 6968 | size_t pages; |
2b188cc1 | 6969 | |
75b28aff HV |
6970 | pages = (size_t)1 << get_order( |
6971 | rings_size(sq_entries, cq_entries, NULL)); | |
6972 | pages += (size_t)1 << get_order( | |
6973 | array_size(sizeof(struct io_uring_sqe), sq_entries)); | |
2b188cc1 | 6974 | |
75b28aff | 6975 | return pages; |
2b188cc1 JA |
6976 | } |
6977 | ||
edafccee JA |
6978 | static int io_sqe_buffer_unregister(struct io_ring_ctx *ctx) |
6979 | { | |
6980 | int i, j; | |
6981 | ||
6982 | if (!ctx->user_bufs) | |
6983 | return -ENXIO; | |
6984 | ||
6985 | for (i = 0; i < ctx->nr_user_bufs; i++) { | |
6986 | struct io_mapped_ubuf *imu = &ctx->user_bufs[i]; | |
6987 | ||
6988 | for (j = 0; j < imu->nr_bvecs; j++) | |
f1f6a7dd | 6989 | unpin_user_page(imu->bvec[j].bv_page); |
edafccee JA |
6990 | |
6991 | if (ctx->account_mem) | |
6992 | io_unaccount_mem(ctx->user, imu->nr_bvecs); | |
d4ef6475 | 6993 | kvfree(imu->bvec); |
edafccee JA |
6994 | imu->nr_bvecs = 0; |
6995 | } | |
6996 | ||
6997 | kfree(ctx->user_bufs); | |
6998 | ctx->user_bufs = NULL; | |
6999 | ctx->nr_user_bufs = 0; | |
7000 | return 0; | |
7001 | } | |
7002 | ||
7003 | static int io_copy_iov(struct io_ring_ctx *ctx, struct iovec *dst, | |
7004 | void __user *arg, unsigned index) | |
7005 | { | |
7006 | struct iovec __user *src; | |
7007 | ||
7008 | #ifdef CONFIG_COMPAT | |
7009 | if (ctx->compat) { | |
7010 | struct compat_iovec __user *ciovs; | |
7011 | struct compat_iovec ciov; | |
7012 | ||
7013 | ciovs = (struct compat_iovec __user *) arg; | |
7014 | if (copy_from_user(&ciov, &ciovs[index], sizeof(ciov))) | |
7015 | return -EFAULT; | |
7016 | ||
d55e5f5b | 7017 | dst->iov_base = u64_to_user_ptr((u64)ciov.iov_base); |
edafccee JA |
7018 | dst->iov_len = ciov.iov_len; |
7019 | return 0; | |
7020 | } | |
7021 | #endif | |
7022 | src = (struct iovec __user *) arg; | |
7023 | if (copy_from_user(dst, &src[index], sizeof(*dst))) | |
7024 | return -EFAULT; | |
7025 | return 0; | |
7026 | } | |
7027 | ||
7028 | static int io_sqe_buffer_register(struct io_ring_ctx *ctx, void __user *arg, | |
7029 | unsigned nr_args) | |
7030 | { | |
7031 | struct vm_area_struct **vmas = NULL; | |
7032 | struct page **pages = NULL; | |
7033 | int i, j, got_pages = 0; | |
7034 | int ret = -EINVAL; | |
7035 | ||
7036 | if (ctx->user_bufs) | |
7037 | return -EBUSY; | |
7038 | if (!nr_args || nr_args > UIO_MAXIOV) | |
7039 | return -EINVAL; | |
7040 | ||
7041 | ctx->user_bufs = kcalloc(nr_args, sizeof(struct io_mapped_ubuf), | |
7042 | GFP_KERNEL); | |
7043 | if (!ctx->user_bufs) | |
7044 | return -ENOMEM; | |
7045 | ||
7046 | for (i = 0; i < nr_args; i++) { | |
7047 | struct io_mapped_ubuf *imu = &ctx->user_bufs[i]; | |
7048 | unsigned long off, start, end, ubuf; | |
7049 | int pret, nr_pages; | |
7050 | struct iovec iov; | |
7051 | size_t size; | |
7052 | ||
7053 | ret = io_copy_iov(ctx, &iov, arg, i); | |
7054 | if (ret) | |
a278682d | 7055 | goto err; |
edafccee JA |
7056 | |
7057 | /* | |
7058 | * Don't impose further limits on the size and buffer | |
7059 | * constraints here, we'll -EINVAL later when IO is | |
7060 | * submitted if they are wrong. | |
7061 | */ | |
7062 | ret = -EFAULT; | |
7063 | if (!iov.iov_base || !iov.iov_len) | |
7064 | goto err; | |
7065 | ||
7066 | /* arbitrary limit, but we need something */ | |
7067 | if (iov.iov_len > SZ_1G) | |
7068 | goto err; | |
7069 | ||
7070 | ubuf = (unsigned long) iov.iov_base; | |
7071 | end = (ubuf + iov.iov_len + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
7072 | start = ubuf >> PAGE_SHIFT; | |
7073 | nr_pages = end - start; | |
7074 | ||
7075 | if (ctx->account_mem) { | |
7076 | ret = io_account_mem(ctx->user, nr_pages); | |
7077 | if (ret) | |
7078 | goto err; | |
7079 | } | |
7080 | ||
7081 | ret = 0; | |
7082 | if (!pages || nr_pages > got_pages) { | |
7083 | kfree(vmas); | |
7084 | kfree(pages); | |
d4ef6475 | 7085 | pages = kvmalloc_array(nr_pages, sizeof(struct page *), |
edafccee | 7086 | GFP_KERNEL); |
d4ef6475 | 7087 | vmas = kvmalloc_array(nr_pages, |
edafccee JA |
7088 | sizeof(struct vm_area_struct *), |
7089 | GFP_KERNEL); | |
7090 | if (!pages || !vmas) { | |
7091 | ret = -ENOMEM; | |
7092 | if (ctx->account_mem) | |
7093 | io_unaccount_mem(ctx->user, nr_pages); | |
7094 | goto err; | |
7095 | } | |
7096 | got_pages = nr_pages; | |
7097 | } | |
7098 | ||
d4ef6475 | 7099 | imu->bvec = kvmalloc_array(nr_pages, sizeof(struct bio_vec), |
edafccee JA |
7100 | GFP_KERNEL); |
7101 | ret = -ENOMEM; | |
7102 | if (!imu->bvec) { | |
7103 | if (ctx->account_mem) | |
7104 | io_unaccount_mem(ctx->user, nr_pages); | |
7105 | goto err; | |
7106 | } | |
7107 | ||
7108 | ret = 0; | |
7109 | down_read(¤t->mm->mmap_sem); | |
2113b05d | 7110 | pret = pin_user_pages(ubuf, nr_pages, |
932f4a63 IW |
7111 | FOLL_WRITE | FOLL_LONGTERM, |
7112 | pages, vmas); | |
edafccee JA |
7113 | if (pret == nr_pages) { |
7114 | /* don't support file backed memory */ | |
7115 | for (j = 0; j < nr_pages; j++) { | |
7116 | struct vm_area_struct *vma = vmas[j]; | |
7117 | ||
7118 | if (vma->vm_file && | |
7119 | !is_file_hugepages(vma->vm_file)) { | |
7120 | ret = -EOPNOTSUPP; | |
7121 | break; | |
7122 | } | |
7123 | } | |
7124 | } else { | |
7125 | ret = pret < 0 ? pret : -EFAULT; | |
7126 | } | |
7127 | up_read(¤t->mm->mmap_sem); | |
7128 | if (ret) { | |
7129 | /* | |
7130 | * if we did partial map, or found file backed vmas, | |
7131 | * release any pages we did get | |
7132 | */ | |
27c4d3a3 | 7133 | if (pret > 0) |
f1f6a7dd | 7134 | unpin_user_pages(pages, pret); |
edafccee JA |
7135 | if (ctx->account_mem) |
7136 | io_unaccount_mem(ctx->user, nr_pages); | |
d4ef6475 | 7137 | kvfree(imu->bvec); |
edafccee JA |
7138 | goto err; |
7139 | } | |
7140 | ||
7141 | off = ubuf & ~PAGE_MASK; | |
7142 | size = iov.iov_len; | |
7143 | for (j = 0; j < nr_pages; j++) { | |
7144 | size_t vec_len; | |
7145 | ||
7146 | vec_len = min_t(size_t, size, PAGE_SIZE - off); | |
7147 | imu->bvec[j].bv_page = pages[j]; | |
7148 | imu->bvec[j].bv_len = vec_len; | |
7149 | imu->bvec[j].bv_offset = off; | |
7150 | off = 0; | |
7151 | size -= vec_len; | |
7152 | } | |
7153 | /* store original address for later verification */ | |
7154 | imu->ubuf = ubuf; | |
7155 | imu->len = iov.iov_len; | |
7156 | imu->nr_bvecs = nr_pages; | |
7157 | ||
7158 | ctx->nr_user_bufs++; | |
7159 | } | |
d4ef6475 MR |
7160 | kvfree(pages); |
7161 | kvfree(vmas); | |
edafccee JA |
7162 | return 0; |
7163 | err: | |
d4ef6475 MR |
7164 | kvfree(pages); |
7165 | kvfree(vmas); | |
edafccee JA |
7166 | io_sqe_buffer_unregister(ctx); |
7167 | return ret; | |
7168 | } | |
7169 | ||
9b402849 JA |
7170 | static int io_eventfd_register(struct io_ring_ctx *ctx, void __user *arg) |
7171 | { | |
7172 | __s32 __user *fds = arg; | |
7173 | int fd; | |
7174 | ||
7175 | if (ctx->cq_ev_fd) | |
7176 | return -EBUSY; | |
7177 | ||
7178 | if (copy_from_user(&fd, fds, sizeof(*fds))) | |
7179 | return -EFAULT; | |
7180 | ||
7181 | ctx->cq_ev_fd = eventfd_ctx_fdget(fd); | |
7182 | if (IS_ERR(ctx->cq_ev_fd)) { | |
7183 | int ret = PTR_ERR(ctx->cq_ev_fd); | |
7184 | ctx->cq_ev_fd = NULL; | |
7185 | return ret; | |
7186 | } | |
7187 | ||
7188 | return 0; | |
7189 | } | |
7190 | ||
7191 | static int io_eventfd_unregister(struct io_ring_ctx *ctx) | |
7192 | { | |
7193 | if (ctx->cq_ev_fd) { | |
7194 | eventfd_ctx_put(ctx->cq_ev_fd); | |
7195 | ctx->cq_ev_fd = NULL; | |
7196 | return 0; | |
7197 | } | |
7198 | ||
7199 | return -ENXIO; | |
7200 | } | |
7201 | ||
5a2e745d JA |
7202 | static int __io_destroy_buffers(int id, void *p, void *data) |
7203 | { | |
7204 | struct io_ring_ctx *ctx = data; | |
7205 | struct io_buffer *buf = p; | |
7206 | ||
067524e9 | 7207 | __io_remove_buffers(ctx, buf, id, -1U); |
5a2e745d JA |
7208 | return 0; |
7209 | } | |
7210 | ||
7211 | static void io_destroy_buffers(struct io_ring_ctx *ctx) | |
7212 | { | |
7213 | idr_for_each(&ctx->io_buffer_idr, __io_destroy_buffers, ctx); | |
7214 | idr_destroy(&ctx->io_buffer_idr); | |
7215 | } | |
7216 | ||
2b188cc1 JA |
7217 | static void io_ring_ctx_free(struct io_ring_ctx *ctx) |
7218 | { | |
6b06314c | 7219 | io_finish_async(ctx); |
2b188cc1 JA |
7220 | if (ctx->sqo_mm) |
7221 | mmdrop(ctx->sqo_mm); | |
def596e9 JA |
7222 | |
7223 | io_iopoll_reap_events(ctx); | |
edafccee | 7224 | io_sqe_buffer_unregister(ctx); |
6b06314c | 7225 | io_sqe_files_unregister(ctx); |
9b402849 | 7226 | io_eventfd_unregister(ctx); |
5a2e745d | 7227 | io_destroy_buffers(ctx); |
41726c9a | 7228 | idr_destroy(&ctx->personality_idr); |
def596e9 | 7229 | |
2b188cc1 | 7230 | #if defined(CONFIG_UNIX) |
355e8d26 EB |
7231 | if (ctx->ring_sock) { |
7232 | ctx->ring_sock->file = NULL; /* so that iput() is called */ | |
2b188cc1 | 7233 | sock_release(ctx->ring_sock); |
355e8d26 | 7234 | } |
2b188cc1 JA |
7235 | #endif |
7236 | ||
75b28aff | 7237 | io_mem_free(ctx->rings); |
2b188cc1 | 7238 | io_mem_free(ctx->sq_sqes); |
2b188cc1 JA |
7239 | |
7240 | percpu_ref_exit(&ctx->refs); | |
7241 | if (ctx->account_mem) | |
7242 | io_unaccount_mem(ctx->user, | |
7243 | ring_pages(ctx->sq_entries, ctx->cq_entries)); | |
7244 | free_uid(ctx->user); | |
181e448d | 7245 | put_cred(ctx->creds); |
206aefde | 7246 | kfree(ctx->completions); |
78076bb6 | 7247 | kfree(ctx->cancel_hash); |
0ddf92e8 | 7248 | kmem_cache_free(req_cachep, ctx->fallback_req); |
2b188cc1 JA |
7249 | kfree(ctx); |
7250 | } | |
7251 | ||
7252 | static __poll_t io_uring_poll(struct file *file, poll_table *wait) | |
7253 | { | |
7254 | struct io_ring_ctx *ctx = file->private_data; | |
7255 | __poll_t mask = 0; | |
7256 | ||
7257 | poll_wait(file, &ctx->cq_wait, wait); | |
4f7067c3 SB |
7258 | /* |
7259 | * synchronizes with barrier from wq_has_sleeper call in | |
7260 | * io_commit_cqring | |
7261 | */ | |
2b188cc1 | 7262 | smp_rmb(); |
75b28aff HV |
7263 | if (READ_ONCE(ctx->rings->sq.tail) - ctx->cached_sq_head != |
7264 | ctx->rings->sq_ring_entries) | |
2b188cc1 | 7265 | mask |= EPOLLOUT | EPOLLWRNORM; |
63e5d81f | 7266 | if (io_cqring_events(ctx, false)) |
2b188cc1 JA |
7267 | mask |= EPOLLIN | EPOLLRDNORM; |
7268 | ||
7269 | return mask; | |
7270 | } | |
7271 | ||
7272 | static int io_uring_fasync(int fd, struct file *file, int on) | |
7273 | { | |
7274 | struct io_ring_ctx *ctx = file->private_data; | |
7275 | ||
7276 | return fasync_helper(fd, file, on, &ctx->cq_fasync); | |
7277 | } | |
7278 | ||
071698e1 JA |
7279 | static int io_remove_personalities(int id, void *p, void *data) |
7280 | { | |
7281 | struct io_ring_ctx *ctx = data; | |
7282 | const struct cred *cred; | |
7283 | ||
7284 | cred = idr_remove(&ctx->personality_idr, id); | |
7285 | if (cred) | |
7286 | put_cred(cred); | |
7287 | return 0; | |
7288 | } | |
7289 | ||
85faa7b8 JA |
7290 | static void io_ring_exit_work(struct work_struct *work) |
7291 | { | |
7292 | struct io_ring_ctx *ctx; | |
7293 | ||
7294 | ctx = container_of(work, struct io_ring_ctx, exit_work); | |
7295 | if (ctx->rings) | |
7296 | io_cqring_overflow_flush(ctx, true); | |
7297 | ||
7298 | wait_for_completion(&ctx->completions[0]); | |
7299 | io_ring_ctx_free(ctx); | |
7300 | } | |
7301 | ||
2b188cc1 JA |
7302 | static void io_ring_ctx_wait_and_kill(struct io_ring_ctx *ctx) |
7303 | { | |
7304 | mutex_lock(&ctx->uring_lock); | |
7305 | percpu_ref_kill(&ctx->refs); | |
7306 | mutex_unlock(&ctx->uring_lock); | |
7307 | ||
df069d80 JA |
7308 | /* |
7309 | * Wait for sq thread to idle, if we have one. It won't spin on new | |
7310 | * work after we've killed the ctx ref above. This is important to do | |
7311 | * before we cancel existing commands, as the thread could otherwise | |
7312 | * be queueing new work post that. If that's work we need to cancel, | |
7313 | * it could cause shutdown to hang. | |
7314 | */ | |
7315 | while (ctx->sqo_thread && !wq_has_sleeper(&ctx->sqo_wait)) | |
3fd44c86 | 7316 | cond_resched(); |
df069d80 | 7317 | |
5262f567 | 7318 | io_kill_timeouts(ctx); |
221c5eb2 | 7319 | io_poll_remove_all(ctx); |
561fb04a JA |
7320 | |
7321 | if (ctx->io_wq) | |
7322 | io_wq_cancel_all(ctx->io_wq); | |
7323 | ||
def596e9 | 7324 | io_iopoll_reap_events(ctx); |
15dff286 JA |
7325 | /* if we failed setting up the ctx, we might not have any rings */ |
7326 | if (ctx->rings) | |
7327 | io_cqring_overflow_flush(ctx, true); | |
071698e1 | 7328 | idr_for_each(&ctx->personality_idr, io_remove_personalities, ctx); |
85faa7b8 JA |
7329 | INIT_WORK(&ctx->exit_work, io_ring_exit_work); |
7330 | queue_work(system_wq, &ctx->exit_work); | |
2b188cc1 JA |
7331 | } |
7332 | ||
7333 | static int io_uring_release(struct inode *inode, struct file *file) | |
7334 | { | |
7335 | struct io_ring_ctx *ctx = file->private_data; | |
7336 | ||
7337 | file->private_data = NULL; | |
7338 | io_ring_ctx_wait_and_kill(ctx); | |
7339 | return 0; | |
7340 | } | |
7341 | ||
fcb323cc JA |
7342 | static void io_uring_cancel_files(struct io_ring_ctx *ctx, |
7343 | struct files_struct *files) | |
7344 | { | |
fcb323cc | 7345 | while (!list_empty_careful(&ctx->inflight_list)) { |
d8f1b971 XW |
7346 | struct io_kiocb *cancel_req = NULL, *req; |
7347 | DEFINE_WAIT(wait); | |
fcb323cc JA |
7348 | |
7349 | spin_lock_irq(&ctx->inflight_lock); | |
7350 | list_for_each_entry(req, &ctx->inflight_list, inflight_entry) { | |
768134d4 JA |
7351 | if (req->work.files != files) |
7352 | continue; | |
7353 | /* req is being completed, ignore */ | |
7354 | if (!refcount_inc_not_zero(&req->refs)) | |
7355 | continue; | |
7356 | cancel_req = req; | |
7357 | break; | |
fcb323cc | 7358 | } |
768134d4 | 7359 | if (cancel_req) |
fcb323cc | 7360 | prepare_to_wait(&ctx->inflight_wait, &wait, |
768134d4 | 7361 | TASK_UNINTERRUPTIBLE); |
fcb323cc JA |
7362 | spin_unlock_irq(&ctx->inflight_lock); |
7363 | ||
768134d4 JA |
7364 | /* We need to keep going until we don't find a matching req */ |
7365 | if (!cancel_req) | |
fcb323cc | 7366 | break; |
2f6d9b9d | 7367 | |
2ca10259 JA |
7368 | if (cancel_req->flags & REQ_F_OVERFLOW) { |
7369 | spin_lock_irq(&ctx->completion_lock); | |
7370 | list_del(&cancel_req->list); | |
7371 | cancel_req->flags &= ~REQ_F_OVERFLOW; | |
7372 | if (list_empty(&ctx->cq_overflow_list)) { | |
7373 | clear_bit(0, &ctx->sq_check_overflow); | |
7374 | clear_bit(0, &ctx->cq_check_overflow); | |
7375 | } | |
7376 | spin_unlock_irq(&ctx->completion_lock); | |
7377 | ||
7378 | WRITE_ONCE(ctx->rings->cq_overflow, | |
7379 | atomic_inc_return(&ctx->cached_cq_overflow)); | |
7380 | ||
7381 | /* | |
7382 | * Put inflight ref and overflow ref. If that's | |
7383 | * all we had, then we're done with this request. | |
7384 | */ | |
7385 | if (refcount_sub_and_test(2, &cancel_req->refs)) { | |
7386 | io_put_req(cancel_req); | |
d8f1b971 | 7387 | finish_wait(&ctx->inflight_wait, &wait); |
2ca10259 JA |
7388 | continue; |
7389 | } | |
7390 | } | |
7391 | ||
2f6d9b9d BL |
7392 | io_wq_cancel_work(ctx->io_wq, &cancel_req->work); |
7393 | io_put_req(cancel_req); | |
fcb323cc | 7394 | schedule(); |
d8f1b971 | 7395 | finish_wait(&ctx->inflight_wait, &wait); |
fcb323cc JA |
7396 | } |
7397 | } | |
7398 | ||
7399 | static int io_uring_flush(struct file *file, void *data) | |
7400 | { | |
7401 | struct io_ring_ctx *ctx = file->private_data; | |
7402 | ||
7403 | io_uring_cancel_files(ctx, data); | |
6ab23144 JA |
7404 | |
7405 | /* | |
7406 | * If the task is going away, cancel work it may have pending | |
7407 | */ | |
7408 | if (fatal_signal_pending(current) || (current->flags & PF_EXITING)) | |
7409 | io_wq_cancel_pid(ctx->io_wq, task_pid_vnr(current)); | |
7410 | ||
fcb323cc JA |
7411 | return 0; |
7412 | } | |
7413 | ||
6c5c240e RP |
7414 | static void *io_uring_validate_mmap_request(struct file *file, |
7415 | loff_t pgoff, size_t sz) | |
2b188cc1 | 7416 | { |
2b188cc1 | 7417 | struct io_ring_ctx *ctx = file->private_data; |
6c5c240e | 7418 | loff_t offset = pgoff << PAGE_SHIFT; |
2b188cc1 JA |
7419 | struct page *page; |
7420 | void *ptr; | |
7421 | ||
7422 | switch (offset) { | |
7423 | case IORING_OFF_SQ_RING: | |
75b28aff HV |
7424 | case IORING_OFF_CQ_RING: |
7425 | ptr = ctx->rings; | |
2b188cc1 JA |
7426 | break; |
7427 | case IORING_OFF_SQES: | |
7428 | ptr = ctx->sq_sqes; | |
7429 | break; | |
2b188cc1 | 7430 | default: |
6c5c240e | 7431 | return ERR_PTR(-EINVAL); |
2b188cc1 JA |
7432 | } |
7433 | ||
7434 | page = virt_to_head_page(ptr); | |
a50b854e | 7435 | if (sz > page_size(page)) |
6c5c240e RP |
7436 | return ERR_PTR(-EINVAL); |
7437 | ||
7438 | return ptr; | |
7439 | } | |
7440 | ||
7441 | #ifdef CONFIG_MMU | |
7442 | ||
7443 | static int io_uring_mmap(struct file *file, struct vm_area_struct *vma) | |
7444 | { | |
7445 | size_t sz = vma->vm_end - vma->vm_start; | |
7446 | unsigned long pfn; | |
7447 | void *ptr; | |
7448 | ||
7449 | ptr = io_uring_validate_mmap_request(file, vma->vm_pgoff, sz); | |
7450 | if (IS_ERR(ptr)) | |
7451 | return PTR_ERR(ptr); | |
2b188cc1 JA |
7452 | |
7453 | pfn = virt_to_phys(ptr) >> PAGE_SHIFT; | |
7454 | return remap_pfn_range(vma, vma->vm_start, pfn, sz, vma->vm_page_prot); | |
7455 | } | |
7456 | ||
6c5c240e RP |
7457 | #else /* !CONFIG_MMU */ |
7458 | ||
7459 | static int io_uring_mmap(struct file *file, struct vm_area_struct *vma) | |
7460 | { | |
7461 | return vma->vm_flags & (VM_SHARED | VM_MAYSHARE) ? 0 : -EINVAL; | |
7462 | } | |
7463 | ||
7464 | static unsigned int io_uring_nommu_mmap_capabilities(struct file *file) | |
7465 | { | |
7466 | return NOMMU_MAP_DIRECT | NOMMU_MAP_READ | NOMMU_MAP_WRITE; | |
7467 | } | |
7468 | ||
7469 | static unsigned long io_uring_nommu_get_unmapped_area(struct file *file, | |
7470 | unsigned long addr, unsigned long len, | |
7471 | unsigned long pgoff, unsigned long flags) | |
7472 | { | |
7473 | void *ptr; | |
7474 | ||
7475 | ptr = io_uring_validate_mmap_request(file, pgoff, len); | |
7476 | if (IS_ERR(ptr)) | |
7477 | return PTR_ERR(ptr); | |
7478 | ||
7479 | return (unsigned long) ptr; | |
7480 | } | |
7481 | ||
7482 | #endif /* !CONFIG_MMU */ | |
7483 | ||
2b188cc1 JA |
7484 | SYSCALL_DEFINE6(io_uring_enter, unsigned int, fd, u32, to_submit, |
7485 | u32, min_complete, u32, flags, const sigset_t __user *, sig, | |
7486 | size_t, sigsz) | |
7487 | { | |
7488 | struct io_ring_ctx *ctx; | |
7489 | long ret = -EBADF; | |
7490 | int submitted = 0; | |
7491 | struct fd f; | |
7492 | ||
b41e9852 JA |
7493 | if (current->task_works) |
7494 | task_work_run(); | |
7495 | ||
6c271ce2 | 7496 | if (flags & ~(IORING_ENTER_GETEVENTS | IORING_ENTER_SQ_WAKEUP)) |
2b188cc1 JA |
7497 | return -EINVAL; |
7498 | ||
7499 | f = fdget(fd); | |
7500 | if (!f.file) | |
7501 | return -EBADF; | |
7502 | ||
7503 | ret = -EOPNOTSUPP; | |
7504 | if (f.file->f_op != &io_uring_fops) | |
7505 | goto out_fput; | |
7506 | ||
7507 | ret = -ENXIO; | |
7508 | ctx = f.file->private_data; | |
7509 | if (!percpu_ref_tryget(&ctx->refs)) | |
7510 | goto out_fput; | |
7511 | ||
6c271ce2 JA |
7512 | /* |
7513 | * For SQ polling, the thread will do all submissions and completions. | |
7514 | * Just return the requested submit count, and wake the thread if | |
7515 | * we were asked to. | |
7516 | */ | |
b2a9eada | 7517 | ret = 0; |
6c271ce2 | 7518 | if (ctx->flags & IORING_SETUP_SQPOLL) { |
c1edbf5f JA |
7519 | if (!list_empty_careful(&ctx->cq_overflow_list)) |
7520 | io_cqring_overflow_flush(ctx, false); | |
6c271ce2 JA |
7521 | if (flags & IORING_ENTER_SQ_WAKEUP) |
7522 | wake_up(&ctx->sqo_wait); | |
7523 | submitted = to_submit; | |
b2a9eada | 7524 | } else if (to_submit) { |
2b188cc1 | 7525 | mutex_lock(&ctx->uring_lock); |
bf9c2f1c | 7526 | submitted = io_submit_sqes(ctx, to_submit, f.file, fd, false); |
2b188cc1 | 7527 | mutex_unlock(&ctx->uring_lock); |
7c504e65 PB |
7528 | |
7529 | if (submitted != to_submit) | |
7530 | goto out; | |
2b188cc1 JA |
7531 | } |
7532 | if (flags & IORING_ENTER_GETEVENTS) { | |
def596e9 JA |
7533 | unsigned nr_events = 0; |
7534 | ||
2b188cc1 JA |
7535 | min_complete = min(min_complete, ctx->cq_entries); |
7536 | ||
32b2244a XW |
7537 | /* |
7538 | * When SETUP_IOPOLL and SETUP_SQPOLL are both enabled, user | |
7539 | * space applications don't need to do io completion events | |
7540 | * polling again, they can rely on io_sq_thread to do polling | |
7541 | * work, which can reduce cpu usage and uring_lock contention. | |
7542 | */ | |
7543 | if (ctx->flags & IORING_SETUP_IOPOLL && | |
7544 | !(ctx->flags & IORING_SETUP_SQPOLL)) { | |
def596e9 | 7545 | ret = io_iopoll_check(ctx, &nr_events, min_complete); |
def596e9 JA |
7546 | } else { |
7547 | ret = io_cqring_wait(ctx, min_complete, sig, sigsz); | |
7548 | } | |
2b188cc1 JA |
7549 | } |
7550 | ||
7c504e65 | 7551 | out: |
6805b32e | 7552 | percpu_ref_put(&ctx->refs); |
2b188cc1 JA |
7553 | out_fput: |
7554 | fdput(f); | |
7555 | return submitted ? submitted : ret; | |
7556 | } | |
7557 | ||
bebdb65e | 7558 | #ifdef CONFIG_PROC_FS |
87ce955b JA |
7559 | static int io_uring_show_cred(int id, void *p, void *data) |
7560 | { | |
7561 | const struct cred *cred = p; | |
7562 | struct seq_file *m = data; | |
7563 | struct user_namespace *uns = seq_user_ns(m); | |
7564 | struct group_info *gi; | |
7565 | kernel_cap_t cap; | |
7566 | unsigned __capi; | |
7567 | int g; | |
7568 | ||
7569 | seq_printf(m, "%5d\n", id); | |
7570 | seq_put_decimal_ull(m, "\tUid:\t", from_kuid_munged(uns, cred->uid)); | |
7571 | seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->euid)); | |
7572 | seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->suid)); | |
7573 | seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->fsuid)); | |
7574 | seq_put_decimal_ull(m, "\n\tGid:\t", from_kgid_munged(uns, cred->gid)); | |
7575 | seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->egid)); | |
7576 | seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->sgid)); | |
7577 | seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->fsgid)); | |
7578 | seq_puts(m, "\n\tGroups:\t"); | |
7579 | gi = cred->group_info; | |
7580 | for (g = 0; g < gi->ngroups; g++) { | |
7581 | seq_put_decimal_ull(m, g ? " " : "", | |
7582 | from_kgid_munged(uns, gi->gid[g])); | |
7583 | } | |
7584 | seq_puts(m, "\n\tCapEff:\t"); | |
7585 | cap = cred->cap_effective; | |
7586 | CAP_FOR_EACH_U32(__capi) | |
7587 | seq_put_hex_ll(m, NULL, cap.cap[CAP_LAST_U32 - __capi], 8); | |
7588 | seq_putc(m, '\n'); | |
7589 | return 0; | |
7590 | } | |
7591 | ||
7592 | static void __io_uring_show_fdinfo(struct io_ring_ctx *ctx, struct seq_file *m) | |
7593 | { | |
7594 | int i; | |
7595 | ||
7596 | mutex_lock(&ctx->uring_lock); | |
7597 | seq_printf(m, "UserFiles:\t%u\n", ctx->nr_user_files); | |
7598 | for (i = 0; i < ctx->nr_user_files; i++) { | |
7599 | struct fixed_file_table *table; | |
7600 | struct file *f; | |
7601 | ||
7602 | table = &ctx->file_data->table[i >> IORING_FILE_TABLE_SHIFT]; | |
7603 | f = table->files[i & IORING_FILE_TABLE_MASK]; | |
7604 | if (f) | |
7605 | seq_printf(m, "%5u: %s\n", i, file_dentry(f)->d_iname); | |
7606 | else | |
7607 | seq_printf(m, "%5u: <none>\n", i); | |
7608 | } | |
7609 | seq_printf(m, "UserBufs:\t%u\n", ctx->nr_user_bufs); | |
7610 | for (i = 0; i < ctx->nr_user_bufs; i++) { | |
7611 | struct io_mapped_ubuf *buf = &ctx->user_bufs[i]; | |
7612 | ||
7613 | seq_printf(m, "%5u: 0x%llx/%u\n", i, buf->ubuf, | |
7614 | (unsigned int) buf->len); | |
7615 | } | |
7616 | if (!idr_is_empty(&ctx->personality_idr)) { | |
7617 | seq_printf(m, "Personalities:\n"); | |
7618 | idr_for_each(&ctx->personality_idr, io_uring_show_cred, m); | |
7619 | } | |
d7718a9d JA |
7620 | seq_printf(m, "PollList:\n"); |
7621 | spin_lock_irq(&ctx->completion_lock); | |
7622 | for (i = 0; i < (1U << ctx->cancel_hash_bits); i++) { | |
7623 | struct hlist_head *list = &ctx->cancel_hash[i]; | |
7624 | struct io_kiocb *req; | |
7625 | ||
7626 | hlist_for_each_entry(req, list, hash_node) | |
7627 | seq_printf(m, " op=%d, task_works=%d\n", req->opcode, | |
7628 | req->task->task_works != NULL); | |
7629 | } | |
7630 | spin_unlock_irq(&ctx->completion_lock); | |
87ce955b JA |
7631 | mutex_unlock(&ctx->uring_lock); |
7632 | } | |
7633 | ||
7634 | static void io_uring_show_fdinfo(struct seq_file *m, struct file *f) | |
7635 | { | |
7636 | struct io_ring_ctx *ctx = f->private_data; | |
7637 | ||
7638 | if (percpu_ref_tryget(&ctx->refs)) { | |
7639 | __io_uring_show_fdinfo(ctx, m); | |
7640 | percpu_ref_put(&ctx->refs); | |
7641 | } | |
7642 | } | |
bebdb65e | 7643 | #endif |
87ce955b | 7644 | |
2b188cc1 JA |
7645 | static const struct file_operations io_uring_fops = { |
7646 | .release = io_uring_release, | |
fcb323cc | 7647 | .flush = io_uring_flush, |
2b188cc1 | 7648 | .mmap = io_uring_mmap, |
6c5c240e RP |
7649 | #ifndef CONFIG_MMU |
7650 | .get_unmapped_area = io_uring_nommu_get_unmapped_area, | |
7651 | .mmap_capabilities = io_uring_nommu_mmap_capabilities, | |
7652 | #endif | |
2b188cc1 JA |
7653 | .poll = io_uring_poll, |
7654 | .fasync = io_uring_fasync, | |
bebdb65e | 7655 | #ifdef CONFIG_PROC_FS |
87ce955b | 7656 | .show_fdinfo = io_uring_show_fdinfo, |
bebdb65e | 7657 | #endif |
2b188cc1 JA |
7658 | }; |
7659 | ||
7660 | static int io_allocate_scq_urings(struct io_ring_ctx *ctx, | |
7661 | struct io_uring_params *p) | |
7662 | { | |
75b28aff HV |
7663 | struct io_rings *rings; |
7664 | size_t size, sq_array_offset; | |
2b188cc1 | 7665 | |
75b28aff HV |
7666 | size = rings_size(p->sq_entries, p->cq_entries, &sq_array_offset); |
7667 | if (size == SIZE_MAX) | |
7668 | return -EOVERFLOW; | |
7669 | ||
7670 | rings = io_mem_alloc(size); | |
7671 | if (!rings) | |
2b188cc1 JA |
7672 | return -ENOMEM; |
7673 | ||
75b28aff HV |
7674 | ctx->rings = rings; |
7675 | ctx->sq_array = (u32 *)((char *)rings + sq_array_offset); | |
7676 | rings->sq_ring_mask = p->sq_entries - 1; | |
7677 | rings->cq_ring_mask = p->cq_entries - 1; | |
7678 | rings->sq_ring_entries = p->sq_entries; | |
7679 | rings->cq_ring_entries = p->cq_entries; | |
7680 | ctx->sq_mask = rings->sq_ring_mask; | |
7681 | ctx->cq_mask = rings->cq_ring_mask; | |
7682 | ctx->sq_entries = rings->sq_ring_entries; | |
7683 | ctx->cq_entries = rings->cq_ring_entries; | |
2b188cc1 JA |
7684 | |
7685 | size = array_size(sizeof(struct io_uring_sqe), p->sq_entries); | |
eb065d30 JA |
7686 | if (size == SIZE_MAX) { |
7687 | io_mem_free(ctx->rings); | |
7688 | ctx->rings = NULL; | |
2b188cc1 | 7689 | return -EOVERFLOW; |
eb065d30 | 7690 | } |
2b188cc1 JA |
7691 | |
7692 | ctx->sq_sqes = io_mem_alloc(size); | |
eb065d30 JA |
7693 | if (!ctx->sq_sqes) { |
7694 | io_mem_free(ctx->rings); | |
7695 | ctx->rings = NULL; | |
2b188cc1 | 7696 | return -ENOMEM; |
eb065d30 | 7697 | } |
2b188cc1 | 7698 | |
2b188cc1 JA |
7699 | return 0; |
7700 | } | |
7701 | ||
7702 | /* | |
7703 | * Allocate an anonymous fd, this is what constitutes the application | |
7704 | * visible backing of an io_uring instance. The application mmaps this | |
7705 | * fd to gain access to the SQ/CQ ring details. If UNIX sockets are enabled, | |
7706 | * we have to tie this fd to a socket for file garbage collection purposes. | |
7707 | */ | |
7708 | static int io_uring_get_fd(struct io_ring_ctx *ctx) | |
7709 | { | |
7710 | struct file *file; | |
7711 | int ret; | |
7712 | ||
7713 | #if defined(CONFIG_UNIX) | |
7714 | ret = sock_create_kern(&init_net, PF_UNIX, SOCK_RAW, IPPROTO_IP, | |
7715 | &ctx->ring_sock); | |
7716 | if (ret) | |
7717 | return ret; | |
7718 | #endif | |
7719 | ||
7720 | ret = get_unused_fd_flags(O_RDWR | O_CLOEXEC); | |
7721 | if (ret < 0) | |
7722 | goto err; | |
7723 | ||
7724 | file = anon_inode_getfile("[io_uring]", &io_uring_fops, ctx, | |
7725 | O_RDWR | O_CLOEXEC); | |
7726 | if (IS_ERR(file)) { | |
7727 | put_unused_fd(ret); | |
7728 | ret = PTR_ERR(file); | |
7729 | goto err; | |
7730 | } | |
7731 | ||
7732 | #if defined(CONFIG_UNIX) | |
7733 | ctx->ring_sock->file = file; | |
7734 | #endif | |
7735 | fd_install(ret, file); | |
7736 | return ret; | |
7737 | err: | |
7738 | #if defined(CONFIG_UNIX) | |
7739 | sock_release(ctx->ring_sock); | |
7740 | ctx->ring_sock = NULL; | |
7741 | #endif | |
7742 | return ret; | |
7743 | } | |
7744 | ||
7f13657d XW |
7745 | static int io_uring_create(unsigned entries, struct io_uring_params *p, |
7746 | struct io_uring_params __user *params) | |
2b188cc1 JA |
7747 | { |
7748 | struct user_struct *user = NULL; | |
7749 | struct io_ring_ctx *ctx; | |
7750 | bool account_mem; | |
7751 | int ret; | |
7752 | ||
8110c1a6 | 7753 | if (!entries) |
2b188cc1 | 7754 | return -EINVAL; |
8110c1a6 JA |
7755 | if (entries > IORING_MAX_ENTRIES) { |
7756 | if (!(p->flags & IORING_SETUP_CLAMP)) | |
7757 | return -EINVAL; | |
7758 | entries = IORING_MAX_ENTRIES; | |
7759 | } | |
2b188cc1 JA |
7760 | |
7761 | /* | |
7762 | * Use twice as many entries for the CQ ring. It's possible for the | |
7763 | * application to drive a higher depth than the size of the SQ ring, | |
7764 | * since the sqes are only used at submission time. This allows for | |
33a107f0 JA |
7765 | * some flexibility in overcommitting a bit. If the application has |
7766 | * set IORING_SETUP_CQSIZE, it will have passed in the desired number | |
7767 | * of CQ ring entries manually. | |
2b188cc1 JA |
7768 | */ |
7769 | p->sq_entries = roundup_pow_of_two(entries); | |
33a107f0 JA |
7770 | if (p->flags & IORING_SETUP_CQSIZE) { |
7771 | /* | |
7772 | * If IORING_SETUP_CQSIZE is set, we do the same roundup | |
7773 | * to a power-of-two, if it isn't already. We do NOT impose | |
7774 | * any cq vs sq ring sizing. | |
7775 | */ | |
8110c1a6 | 7776 | if (p->cq_entries < p->sq_entries) |
33a107f0 | 7777 | return -EINVAL; |
8110c1a6 JA |
7778 | if (p->cq_entries > IORING_MAX_CQ_ENTRIES) { |
7779 | if (!(p->flags & IORING_SETUP_CLAMP)) | |
7780 | return -EINVAL; | |
7781 | p->cq_entries = IORING_MAX_CQ_ENTRIES; | |
7782 | } | |
33a107f0 JA |
7783 | p->cq_entries = roundup_pow_of_two(p->cq_entries); |
7784 | } else { | |
7785 | p->cq_entries = 2 * p->sq_entries; | |
7786 | } | |
2b188cc1 JA |
7787 | |
7788 | user = get_uid(current_user()); | |
7789 | account_mem = !capable(CAP_IPC_LOCK); | |
7790 | ||
7791 | if (account_mem) { | |
7792 | ret = io_account_mem(user, | |
7793 | ring_pages(p->sq_entries, p->cq_entries)); | |
7794 | if (ret) { | |
7795 | free_uid(user); | |
7796 | return ret; | |
7797 | } | |
7798 | } | |
7799 | ||
7800 | ctx = io_ring_ctx_alloc(p); | |
7801 | if (!ctx) { | |
7802 | if (account_mem) | |
7803 | io_unaccount_mem(user, ring_pages(p->sq_entries, | |
7804 | p->cq_entries)); | |
7805 | free_uid(user); | |
7806 | return -ENOMEM; | |
7807 | } | |
7808 | ctx->compat = in_compat_syscall(); | |
7809 | ctx->account_mem = account_mem; | |
7810 | ctx->user = user; | |
0b8c0ec7 | 7811 | ctx->creds = get_current_cred(); |
2b188cc1 JA |
7812 | |
7813 | ret = io_allocate_scq_urings(ctx, p); | |
7814 | if (ret) | |
7815 | goto err; | |
7816 | ||
6c271ce2 | 7817 | ret = io_sq_offload_start(ctx, p); |
2b188cc1 JA |
7818 | if (ret) |
7819 | goto err; | |
7820 | ||
2b188cc1 | 7821 | memset(&p->sq_off, 0, sizeof(p->sq_off)); |
75b28aff HV |
7822 | p->sq_off.head = offsetof(struct io_rings, sq.head); |
7823 | p->sq_off.tail = offsetof(struct io_rings, sq.tail); | |
7824 | p->sq_off.ring_mask = offsetof(struct io_rings, sq_ring_mask); | |
7825 | p->sq_off.ring_entries = offsetof(struct io_rings, sq_ring_entries); | |
7826 | p->sq_off.flags = offsetof(struct io_rings, sq_flags); | |
7827 | p->sq_off.dropped = offsetof(struct io_rings, sq_dropped); | |
7828 | p->sq_off.array = (char *)ctx->sq_array - (char *)ctx->rings; | |
2b188cc1 JA |
7829 | |
7830 | memset(&p->cq_off, 0, sizeof(p->cq_off)); | |
75b28aff HV |
7831 | p->cq_off.head = offsetof(struct io_rings, cq.head); |
7832 | p->cq_off.tail = offsetof(struct io_rings, cq.tail); | |
7833 | p->cq_off.ring_mask = offsetof(struct io_rings, cq_ring_mask); | |
7834 | p->cq_off.ring_entries = offsetof(struct io_rings, cq_ring_entries); | |
7835 | p->cq_off.overflow = offsetof(struct io_rings, cq_overflow); | |
7836 | p->cq_off.cqes = offsetof(struct io_rings, cqes); | |
ac90f249 | 7837 | |
7f13657d XW |
7838 | p->features = IORING_FEAT_SINGLE_MMAP | IORING_FEAT_NODROP | |
7839 | IORING_FEAT_SUBMIT_STABLE | IORING_FEAT_RW_CUR_POS | | |
7840 | IORING_FEAT_CUR_PERSONALITY | IORING_FEAT_FAST_POLL; | |
7841 | ||
7842 | if (copy_to_user(params, p, sizeof(*p))) { | |
7843 | ret = -EFAULT; | |
7844 | goto err; | |
7845 | } | |
044c1ab3 JA |
7846 | /* |
7847 | * Install ring fd as the very last thing, so we don't risk someone | |
7848 | * having closed it before we finish setup | |
7849 | */ | |
7850 | ret = io_uring_get_fd(ctx); | |
7851 | if (ret < 0) | |
7852 | goto err; | |
7853 | ||
c826bd7a | 7854 | trace_io_uring_create(ret, ctx, p->sq_entries, p->cq_entries, p->flags); |
2b188cc1 JA |
7855 | return ret; |
7856 | err: | |
7857 | io_ring_ctx_wait_and_kill(ctx); | |
7858 | return ret; | |
7859 | } | |
7860 | ||
7861 | /* | |
7862 | * Sets up an aio uring context, and returns the fd. Applications asks for a | |
7863 | * ring size, we return the actual sq/cq ring sizes (among other things) in the | |
7864 | * params structure passed in. | |
7865 | */ | |
7866 | static long io_uring_setup(u32 entries, struct io_uring_params __user *params) | |
7867 | { | |
7868 | struct io_uring_params p; | |
2b188cc1 JA |
7869 | int i; |
7870 | ||
7871 | if (copy_from_user(&p, params, sizeof(p))) | |
7872 | return -EFAULT; | |
7873 | for (i = 0; i < ARRAY_SIZE(p.resv); i++) { | |
7874 | if (p.resv[i]) | |
7875 | return -EINVAL; | |
7876 | } | |
7877 | ||
6c271ce2 | 7878 | if (p.flags & ~(IORING_SETUP_IOPOLL | IORING_SETUP_SQPOLL | |
8110c1a6 | 7879 | IORING_SETUP_SQ_AFF | IORING_SETUP_CQSIZE | |
24369c2e | 7880 | IORING_SETUP_CLAMP | IORING_SETUP_ATTACH_WQ)) |
2b188cc1 JA |
7881 | return -EINVAL; |
7882 | ||
7f13657d | 7883 | return io_uring_create(entries, &p, params); |
2b188cc1 JA |
7884 | } |
7885 | ||
7886 | SYSCALL_DEFINE2(io_uring_setup, u32, entries, | |
7887 | struct io_uring_params __user *, params) | |
7888 | { | |
7889 | return io_uring_setup(entries, params); | |
7890 | } | |
7891 | ||
66f4af93 JA |
7892 | static int io_probe(struct io_ring_ctx *ctx, void __user *arg, unsigned nr_args) |
7893 | { | |
7894 | struct io_uring_probe *p; | |
7895 | size_t size; | |
7896 | int i, ret; | |
7897 | ||
7898 | size = struct_size(p, ops, nr_args); | |
7899 | if (size == SIZE_MAX) | |
7900 | return -EOVERFLOW; | |
7901 | p = kzalloc(size, GFP_KERNEL); | |
7902 | if (!p) | |
7903 | return -ENOMEM; | |
7904 | ||
7905 | ret = -EFAULT; | |
7906 | if (copy_from_user(p, arg, size)) | |
7907 | goto out; | |
7908 | ret = -EINVAL; | |
7909 | if (memchr_inv(p, 0, size)) | |
7910 | goto out; | |
7911 | ||
7912 | p->last_op = IORING_OP_LAST - 1; | |
7913 | if (nr_args > IORING_OP_LAST) | |
7914 | nr_args = IORING_OP_LAST; | |
7915 | ||
7916 | for (i = 0; i < nr_args; i++) { | |
7917 | p->ops[i].op = i; | |
7918 | if (!io_op_defs[i].not_supported) | |
7919 | p->ops[i].flags = IO_URING_OP_SUPPORTED; | |
7920 | } | |
7921 | p->ops_len = i; | |
7922 | ||
7923 | ret = 0; | |
7924 | if (copy_to_user(arg, p, size)) | |
7925 | ret = -EFAULT; | |
7926 | out: | |
7927 | kfree(p); | |
7928 | return ret; | |
7929 | } | |
7930 | ||
071698e1 JA |
7931 | static int io_register_personality(struct io_ring_ctx *ctx) |
7932 | { | |
7933 | const struct cred *creds = get_current_cred(); | |
7934 | int id; | |
7935 | ||
7936 | id = idr_alloc_cyclic(&ctx->personality_idr, (void *) creds, 1, | |
7937 | USHRT_MAX, GFP_KERNEL); | |
7938 | if (id < 0) | |
7939 | put_cred(creds); | |
7940 | return id; | |
7941 | } | |
7942 | ||
7943 | static int io_unregister_personality(struct io_ring_ctx *ctx, unsigned id) | |
7944 | { | |
7945 | const struct cred *old_creds; | |
7946 | ||
7947 | old_creds = idr_remove(&ctx->personality_idr, id); | |
7948 | if (old_creds) { | |
7949 | put_cred(old_creds); | |
7950 | return 0; | |
7951 | } | |
7952 | ||
7953 | return -EINVAL; | |
7954 | } | |
7955 | ||
7956 | static bool io_register_op_must_quiesce(int op) | |
7957 | { | |
7958 | switch (op) { | |
7959 | case IORING_UNREGISTER_FILES: | |
7960 | case IORING_REGISTER_FILES_UPDATE: | |
7961 | case IORING_REGISTER_PROBE: | |
7962 | case IORING_REGISTER_PERSONALITY: | |
7963 | case IORING_UNREGISTER_PERSONALITY: | |
7964 | return false; | |
7965 | default: | |
7966 | return true; | |
7967 | } | |
7968 | } | |
7969 | ||
edafccee JA |
7970 | static int __io_uring_register(struct io_ring_ctx *ctx, unsigned opcode, |
7971 | void __user *arg, unsigned nr_args) | |
b19062a5 JA |
7972 | __releases(ctx->uring_lock) |
7973 | __acquires(ctx->uring_lock) | |
edafccee JA |
7974 | { |
7975 | int ret; | |
7976 | ||
35fa71a0 JA |
7977 | /* |
7978 | * We're inside the ring mutex, if the ref is already dying, then | |
7979 | * someone else killed the ctx or is already going through | |
7980 | * io_uring_register(). | |
7981 | */ | |
7982 | if (percpu_ref_is_dying(&ctx->refs)) | |
7983 | return -ENXIO; | |
7984 | ||
071698e1 | 7985 | if (io_register_op_must_quiesce(opcode)) { |
05f3fb3c | 7986 | percpu_ref_kill(&ctx->refs); |
b19062a5 | 7987 | |
05f3fb3c JA |
7988 | /* |
7989 | * Drop uring mutex before waiting for references to exit. If | |
7990 | * another thread is currently inside io_uring_enter() it might | |
7991 | * need to grab the uring_lock to make progress. If we hold it | |
7992 | * here across the drain wait, then we can deadlock. It's safe | |
7993 | * to drop the mutex here, since no new references will come in | |
7994 | * after we've killed the percpu ref. | |
7995 | */ | |
7996 | mutex_unlock(&ctx->uring_lock); | |
c150368b | 7997 | ret = wait_for_completion_interruptible(&ctx->completions[0]); |
05f3fb3c | 7998 | mutex_lock(&ctx->uring_lock); |
c150368b JA |
7999 | if (ret) { |
8000 | percpu_ref_resurrect(&ctx->refs); | |
8001 | ret = -EINTR; | |
8002 | goto out; | |
8003 | } | |
05f3fb3c | 8004 | } |
edafccee JA |
8005 | |
8006 | switch (opcode) { | |
8007 | case IORING_REGISTER_BUFFERS: | |
8008 | ret = io_sqe_buffer_register(ctx, arg, nr_args); | |
8009 | break; | |
8010 | case IORING_UNREGISTER_BUFFERS: | |
8011 | ret = -EINVAL; | |
8012 | if (arg || nr_args) | |
8013 | break; | |
8014 | ret = io_sqe_buffer_unregister(ctx); | |
8015 | break; | |
6b06314c JA |
8016 | case IORING_REGISTER_FILES: |
8017 | ret = io_sqe_files_register(ctx, arg, nr_args); | |
8018 | break; | |
8019 | case IORING_UNREGISTER_FILES: | |
8020 | ret = -EINVAL; | |
8021 | if (arg || nr_args) | |
8022 | break; | |
8023 | ret = io_sqe_files_unregister(ctx); | |
8024 | break; | |
c3a31e60 JA |
8025 | case IORING_REGISTER_FILES_UPDATE: |
8026 | ret = io_sqe_files_update(ctx, arg, nr_args); | |
8027 | break; | |
9b402849 | 8028 | case IORING_REGISTER_EVENTFD: |
f2842ab5 | 8029 | case IORING_REGISTER_EVENTFD_ASYNC: |
9b402849 JA |
8030 | ret = -EINVAL; |
8031 | if (nr_args != 1) | |
8032 | break; | |
8033 | ret = io_eventfd_register(ctx, arg); | |
f2842ab5 JA |
8034 | if (ret) |
8035 | break; | |
8036 | if (opcode == IORING_REGISTER_EVENTFD_ASYNC) | |
8037 | ctx->eventfd_async = 1; | |
8038 | else | |
8039 | ctx->eventfd_async = 0; | |
9b402849 JA |
8040 | break; |
8041 | case IORING_UNREGISTER_EVENTFD: | |
8042 | ret = -EINVAL; | |
8043 | if (arg || nr_args) | |
8044 | break; | |
8045 | ret = io_eventfd_unregister(ctx); | |
8046 | break; | |
66f4af93 JA |
8047 | case IORING_REGISTER_PROBE: |
8048 | ret = -EINVAL; | |
8049 | if (!arg || nr_args > 256) | |
8050 | break; | |
8051 | ret = io_probe(ctx, arg, nr_args); | |
8052 | break; | |
071698e1 JA |
8053 | case IORING_REGISTER_PERSONALITY: |
8054 | ret = -EINVAL; | |
8055 | if (arg || nr_args) | |
8056 | break; | |
8057 | ret = io_register_personality(ctx); | |
8058 | break; | |
8059 | case IORING_UNREGISTER_PERSONALITY: | |
8060 | ret = -EINVAL; | |
8061 | if (arg) | |
8062 | break; | |
8063 | ret = io_unregister_personality(ctx, nr_args); | |
8064 | break; | |
edafccee JA |
8065 | default: |
8066 | ret = -EINVAL; | |
8067 | break; | |
8068 | } | |
8069 | ||
071698e1 | 8070 | if (io_register_op_must_quiesce(opcode)) { |
05f3fb3c | 8071 | /* bring the ctx back to life */ |
05f3fb3c | 8072 | percpu_ref_reinit(&ctx->refs); |
c150368b JA |
8073 | out: |
8074 | reinit_completion(&ctx->completions[0]); | |
05f3fb3c | 8075 | } |
edafccee JA |
8076 | return ret; |
8077 | } | |
8078 | ||
8079 | SYSCALL_DEFINE4(io_uring_register, unsigned int, fd, unsigned int, opcode, | |
8080 | void __user *, arg, unsigned int, nr_args) | |
8081 | { | |
8082 | struct io_ring_ctx *ctx; | |
8083 | long ret = -EBADF; | |
8084 | struct fd f; | |
8085 | ||
8086 | f = fdget(fd); | |
8087 | if (!f.file) | |
8088 | return -EBADF; | |
8089 | ||
8090 | ret = -EOPNOTSUPP; | |
8091 | if (f.file->f_op != &io_uring_fops) | |
8092 | goto out_fput; | |
8093 | ||
8094 | ctx = f.file->private_data; | |
8095 | ||
8096 | mutex_lock(&ctx->uring_lock); | |
8097 | ret = __io_uring_register(ctx, opcode, arg, nr_args); | |
8098 | mutex_unlock(&ctx->uring_lock); | |
c826bd7a DD |
8099 | trace_io_uring_register(ctx, opcode, ctx->nr_user_files, ctx->nr_user_bufs, |
8100 | ctx->cq_ev_fd != NULL, ret); | |
edafccee JA |
8101 | out_fput: |
8102 | fdput(f); | |
8103 | return ret; | |
8104 | } | |
8105 | ||
2b188cc1 JA |
8106 | static int __init io_uring_init(void) |
8107 | { | |
d7f62e82 SM |
8108 | #define __BUILD_BUG_VERIFY_ELEMENT(stype, eoffset, etype, ename) do { \ |
8109 | BUILD_BUG_ON(offsetof(stype, ename) != eoffset); \ | |
8110 | BUILD_BUG_ON(sizeof(etype) != sizeof_field(stype, ename)); \ | |
8111 | } while (0) | |
8112 | ||
8113 | #define BUILD_BUG_SQE_ELEM(eoffset, etype, ename) \ | |
8114 | __BUILD_BUG_VERIFY_ELEMENT(struct io_uring_sqe, eoffset, etype, ename) | |
8115 | BUILD_BUG_ON(sizeof(struct io_uring_sqe) != 64); | |
8116 | BUILD_BUG_SQE_ELEM(0, __u8, opcode); | |
8117 | BUILD_BUG_SQE_ELEM(1, __u8, flags); | |
8118 | BUILD_BUG_SQE_ELEM(2, __u16, ioprio); | |
8119 | BUILD_BUG_SQE_ELEM(4, __s32, fd); | |
8120 | BUILD_BUG_SQE_ELEM(8, __u64, off); | |
8121 | BUILD_BUG_SQE_ELEM(8, __u64, addr2); | |
8122 | BUILD_BUG_SQE_ELEM(16, __u64, addr); | |
7d67af2c | 8123 | BUILD_BUG_SQE_ELEM(16, __u64, splice_off_in); |
d7f62e82 SM |
8124 | BUILD_BUG_SQE_ELEM(24, __u32, len); |
8125 | BUILD_BUG_SQE_ELEM(28, __kernel_rwf_t, rw_flags); | |
8126 | BUILD_BUG_SQE_ELEM(28, /* compat */ int, rw_flags); | |
8127 | BUILD_BUG_SQE_ELEM(28, /* compat */ __u32, rw_flags); | |
8128 | BUILD_BUG_SQE_ELEM(28, __u32, fsync_flags); | |
8129 | BUILD_BUG_SQE_ELEM(28, __u16, poll_events); | |
8130 | BUILD_BUG_SQE_ELEM(28, __u32, sync_range_flags); | |
8131 | BUILD_BUG_SQE_ELEM(28, __u32, msg_flags); | |
8132 | BUILD_BUG_SQE_ELEM(28, __u32, timeout_flags); | |
8133 | BUILD_BUG_SQE_ELEM(28, __u32, accept_flags); | |
8134 | BUILD_BUG_SQE_ELEM(28, __u32, cancel_flags); | |
8135 | BUILD_BUG_SQE_ELEM(28, __u32, open_flags); | |
8136 | BUILD_BUG_SQE_ELEM(28, __u32, statx_flags); | |
8137 | BUILD_BUG_SQE_ELEM(28, __u32, fadvise_advice); | |
7d67af2c | 8138 | BUILD_BUG_SQE_ELEM(28, __u32, splice_flags); |
d7f62e82 SM |
8139 | BUILD_BUG_SQE_ELEM(32, __u64, user_data); |
8140 | BUILD_BUG_SQE_ELEM(40, __u16, buf_index); | |
8141 | BUILD_BUG_SQE_ELEM(42, __u16, personality); | |
7d67af2c | 8142 | BUILD_BUG_SQE_ELEM(44, __s32, splice_fd_in); |
d7f62e82 | 8143 | |
d3656344 | 8144 | BUILD_BUG_ON(ARRAY_SIZE(io_op_defs) != IORING_OP_LAST); |
84557871 | 8145 | BUILD_BUG_ON(__REQ_F_LAST_BIT >= 8 * sizeof(int)); |
2b188cc1 JA |
8146 | req_cachep = KMEM_CACHE(io_kiocb, SLAB_HWCACHE_ALIGN | SLAB_PANIC); |
8147 | return 0; | |
8148 | }; | |
8149 | __initcall(io_uring_init); |