]>
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 | 621 | |
4f4eeba8 BM |
622 | u16 buf_index; |
623 | ||
2b188cc1 | 624 | struct io_ring_ctx *ctx; |
d7718a9d | 625 | struct list_head list; |
2b188cc1 | 626 | unsigned int flags; |
c16361c1 | 627 | refcount_t refs; |
3537b6a7 JA |
628 | struct task_struct *task; |
629 | unsigned long fsize; | |
2b188cc1 | 630 | u64 user_data; |
9e645e11 | 631 | u32 result; |
de0617e4 | 632 | u32 sequence; |
2b188cc1 | 633 | |
d7718a9d JA |
634 | struct list_head link_list; |
635 | ||
fcb323cc JA |
636 | struct list_head inflight_entry; |
637 | ||
05589553 XW |
638 | struct percpu_ref *fixed_file_refs; |
639 | ||
b41e9852 JA |
640 | union { |
641 | /* | |
642 | * Only commands that never go async can use the below fields, | |
d7718a9d JA |
643 | * obviously. Right now only IORING_OP_POLL_ADD uses them, and |
644 | * async armed poll handlers for regular commands. The latter | |
645 | * restore the work, if needed. | |
b41e9852 JA |
646 | */ |
647 | struct { | |
b41e9852 | 648 | struct callback_head task_work; |
d7718a9d JA |
649 | struct hlist_node hash_node; |
650 | struct async_poll *apoll; | |
b41e9852 JA |
651 | }; |
652 | struct io_wq_work work; | |
653 | }; | |
2b188cc1 JA |
654 | }; |
655 | ||
656 | #define IO_PLUG_THRESHOLD 2 | |
def596e9 | 657 | #define IO_IOPOLL_BATCH 8 |
2b188cc1 | 658 | |
9a56a232 JA |
659 | struct io_submit_state { |
660 | struct blk_plug plug; | |
661 | ||
2579f913 JA |
662 | /* |
663 | * io_kiocb alloc cache | |
664 | */ | |
665 | void *reqs[IO_IOPOLL_BATCH]; | |
6c8a3134 | 666 | unsigned int free_reqs; |
2579f913 | 667 | |
9a56a232 JA |
668 | /* |
669 | * File reference cache | |
670 | */ | |
671 | struct file *file; | |
672 | unsigned int fd; | |
673 | unsigned int has_refs; | |
674 | unsigned int used_refs; | |
675 | unsigned int ios_left; | |
676 | }; | |
677 | ||
d3656344 JA |
678 | struct io_op_def { |
679 | /* needs req->io allocated for deferral/async */ | |
680 | unsigned async_ctx : 1; | |
681 | /* needs current->mm setup, does mm access */ | |
682 | unsigned needs_mm : 1; | |
683 | /* needs req->file assigned */ | |
684 | unsigned needs_file : 1; | |
d3656344 JA |
685 | /* hash wq insertion if file is a regular file */ |
686 | unsigned hash_reg_file : 1; | |
687 | /* unbound wq insertion if file is a non-regular file */ | |
688 | unsigned unbound_nonreg_file : 1; | |
66f4af93 JA |
689 | /* opcode is not supported by this kernel */ |
690 | unsigned not_supported : 1; | |
f86cd20c JA |
691 | /* needs file table */ |
692 | unsigned file_table : 1; | |
ff002b30 JA |
693 | /* needs ->fs */ |
694 | unsigned needs_fs : 1; | |
8a72758c JA |
695 | /* set if opcode supports polled "wait" */ |
696 | unsigned pollin : 1; | |
697 | unsigned pollout : 1; | |
bcda7baa JA |
698 | /* op supports buffer selection */ |
699 | unsigned buffer_select : 1; | |
d3656344 JA |
700 | }; |
701 | ||
702 | static const struct io_op_def io_op_defs[] = { | |
0463b6c5 PB |
703 | [IORING_OP_NOP] = {}, |
704 | [IORING_OP_READV] = { | |
d3656344 JA |
705 | .async_ctx = 1, |
706 | .needs_mm = 1, | |
707 | .needs_file = 1, | |
708 | .unbound_nonreg_file = 1, | |
8a72758c | 709 | .pollin = 1, |
4d954c25 | 710 | .buffer_select = 1, |
d3656344 | 711 | }, |
0463b6c5 | 712 | [IORING_OP_WRITEV] = { |
d3656344 JA |
713 | .async_ctx = 1, |
714 | .needs_mm = 1, | |
715 | .needs_file = 1, | |
716 | .hash_reg_file = 1, | |
717 | .unbound_nonreg_file = 1, | |
8a72758c | 718 | .pollout = 1, |
d3656344 | 719 | }, |
0463b6c5 | 720 | [IORING_OP_FSYNC] = { |
d3656344 JA |
721 | .needs_file = 1, |
722 | }, | |
0463b6c5 | 723 | [IORING_OP_READ_FIXED] = { |
d3656344 JA |
724 | .needs_file = 1, |
725 | .unbound_nonreg_file = 1, | |
8a72758c | 726 | .pollin = 1, |
d3656344 | 727 | }, |
0463b6c5 | 728 | [IORING_OP_WRITE_FIXED] = { |
d3656344 JA |
729 | .needs_file = 1, |
730 | .hash_reg_file = 1, | |
731 | .unbound_nonreg_file = 1, | |
8a72758c | 732 | .pollout = 1, |
d3656344 | 733 | }, |
0463b6c5 | 734 | [IORING_OP_POLL_ADD] = { |
d3656344 JA |
735 | .needs_file = 1, |
736 | .unbound_nonreg_file = 1, | |
737 | }, | |
0463b6c5 PB |
738 | [IORING_OP_POLL_REMOVE] = {}, |
739 | [IORING_OP_SYNC_FILE_RANGE] = { | |
d3656344 JA |
740 | .needs_file = 1, |
741 | }, | |
0463b6c5 | 742 | [IORING_OP_SENDMSG] = { |
d3656344 JA |
743 | .async_ctx = 1, |
744 | .needs_mm = 1, | |
745 | .needs_file = 1, | |
746 | .unbound_nonreg_file = 1, | |
ff002b30 | 747 | .needs_fs = 1, |
8a72758c | 748 | .pollout = 1, |
d3656344 | 749 | }, |
0463b6c5 | 750 | [IORING_OP_RECVMSG] = { |
d3656344 JA |
751 | .async_ctx = 1, |
752 | .needs_mm = 1, | |
753 | .needs_file = 1, | |
754 | .unbound_nonreg_file = 1, | |
ff002b30 | 755 | .needs_fs = 1, |
8a72758c | 756 | .pollin = 1, |
52de1fe1 | 757 | .buffer_select = 1, |
d3656344 | 758 | }, |
0463b6c5 | 759 | [IORING_OP_TIMEOUT] = { |
d3656344 JA |
760 | .async_ctx = 1, |
761 | .needs_mm = 1, | |
762 | }, | |
0463b6c5 PB |
763 | [IORING_OP_TIMEOUT_REMOVE] = {}, |
764 | [IORING_OP_ACCEPT] = { | |
d3656344 JA |
765 | .needs_mm = 1, |
766 | .needs_file = 1, | |
767 | .unbound_nonreg_file = 1, | |
f86cd20c | 768 | .file_table = 1, |
8a72758c | 769 | .pollin = 1, |
d3656344 | 770 | }, |
0463b6c5 PB |
771 | [IORING_OP_ASYNC_CANCEL] = {}, |
772 | [IORING_OP_LINK_TIMEOUT] = { | |
d3656344 JA |
773 | .async_ctx = 1, |
774 | .needs_mm = 1, | |
775 | }, | |
0463b6c5 | 776 | [IORING_OP_CONNECT] = { |
d3656344 JA |
777 | .async_ctx = 1, |
778 | .needs_mm = 1, | |
779 | .needs_file = 1, | |
780 | .unbound_nonreg_file = 1, | |
8a72758c | 781 | .pollout = 1, |
d3656344 | 782 | }, |
0463b6c5 | 783 | [IORING_OP_FALLOCATE] = { |
d3656344 JA |
784 | .needs_file = 1, |
785 | }, | |
0463b6c5 | 786 | [IORING_OP_OPENAT] = { |
f86cd20c | 787 | .file_table = 1, |
ff002b30 | 788 | .needs_fs = 1, |
d3656344 | 789 | }, |
0463b6c5 | 790 | [IORING_OP_CLOSE] = { |
d3656344 | 791 | .needs_file = 1, |
f86cd20c | 792 | .file_table = 1, |
d3656344 | 793 | }, |
0463b6c5 | 794 | [IORING_OP_FILES_UPDATE] = { |
d3656344 | 795 | .needs_mm = 1, |
f86cd20c | 796 | .file_table = 1, |
d3656344 | 797 | }, |
0463b6c5 | 798 | [IORING_OP_STATX] = { |
d3656344 | 799 | .needs_mm = 1, |
ff002b30 | 800 | .needs_fs = 1, |
5b0bbee4 | 801 | .file_table = 1, |
d3656344 | 802 | }, |
0463b6c5 | 803 | [IORING_OP_READ] = { |
3a6820f2 JA |
804 | .needs_mm = 1, |
805 | .needs_file = 1, | |
806 | .unbound_nonreg_file = 1, | |
8a72758c | 807 | .pollin = 1, |
bcda7baa | 808 | .buffer_select = 1, |
3a6820f2 | 809 | }, |
0463b6c5 | 810 | [IORING_OP_WRITE] = { |
3a6820f2 JA |
811 | .needs_mm = 1, |
812 | .needs_file = 1, | |
813 | .unbound_nonreg_file = 1, | |
8a72758c | 814 | .pollout = 1, |
3a6820f2 | 815 | }, |
0463b6c5 | 816 | [IORING_OP_FADVISE] = { |
4840e418 JA |
817 | .needs_file = 1, |
818 | }, | |
0463b6c5 | 819 | [IORING_OP_MADVISE] = { |
c1ca757b JA |
820 | .needs_mm = 1, |
821 | }, | |
0463b6c5 | 822 | [IORING_OP_SEND] = { |
fddaface JA |
823 | .needs_mm = 1, |
824 | .needs_file = 1, | |
825 | .unbound_nonreg_file = 1, | |
8a72758c | 826 | .pollout = 1, |
fddaface | 827 | }, |
0463b6c5 | 828 | [IORING_OP_RECV] = { |
fddaface JA |
829 | .needs_mm = 1, |
830 | .needs_file = 1, | |
831 | .unbound_nonreg_file = 1, | |
8a72758c | 832 | .pollin = 1, |
bcda7baa | 833 | .buffer_select = 1, |
fddaface | 834 | }, |
0463b6c5 | 835 | [IORING_OP_OPENAT2] = { |
f86cd20c | 836 | .file_table = 1, |
ff002b30 | 837 | .needs_fs = 1, |
cebdb986 | 838 | }, |
3e4827b0 JA |
839 | [IORING_OP_EPOLL_CTL] = { |
840 | .unbound_nonreg_file = 1, | |
841 | .file_table = 1, | |
842 | }, | |
7d67af2c PB |
843 | [IORING_OP_SPLICE] = { |
844 | .needs_file = 1, | |
845 | .hash_reg_file = 1, | |
846 | .unbound_nonreg_file = 1, | |
ddf0322d JA |
847 | }, |
848 | [IORING_OP_PROVIDE_BUFFERS] = {}, | |
067524e9 | 849 | [IORING_OP_REMOVE_BUFFERS] = {}, |
d3656344 JA |
850 | }; |
851 | ||
561fb04a | 852 | static void io_wq_submit_work(struct io_wq_work **workptr); |
78e19bbe | 853 | static void io_cqring_fill_event(struct io_kiocb *req, long res); |
ec9c02ad | 854 | static void io_put_req(struct io_kiocb *req); |
978db57e | 855 | static void __io_double_put_req(struct io_kiocb *req); |
94ae5e77 JA |
856 | static struct io_kiocb *io_prep_linked_timeout(struct io_kiocb *req); |
857 | static void io_queue_linked_timeout(struct io_kiocb *req); | |
05f3fb3c JA |
858 | static int __io_sqe_files_update(struct io_ring_ctx *ctx, |
859 | struct io_uring_files_update *ip, | |
860 | unsigned nr_args); | |
f86cd20c | 861 | static int io_grab_files(struct io_kiocb *req); |
99bc4c38 | 862 | static void io_cleanup_req(struct io_kiocb *req); |
b41e9852 JA |
863 | static int io_file_get(struct io_submit_state *state, struct io_kiocb *req, |
864 | int fd, struct file **out_file, bool fixed); | |
865 | static void __io_queue_sqe(struct io_kiocb *req, | |
866 | const struct io_uring_sqe *sqe); | |
de0617e4 | 867 | |
2b188cc1 JA |
868 | static struct kmem_cache *req_cachep; |
869 | ||
870 | static const struct file_operations io_uring_fops; | |
871 | ||
872 | struct sock *io_uring_get_socket(struct file *file) | |
873 | { | |
874 | #if defined(CONFIG_UNIX) | |
875 | if (file->f_op == &io_uring_fops) { | |
876 | struct io_ring_ctx *ctx = file->private_data; | |
877 | ||
878 | return ctx->ring_sock->sk; | |
879 | } | |
880 | #endif | |
881 | return NULL; | |
882 | } | |
883 | EXPORT_SYMBOL(io_uring_get_socket); | |
884 | ||
885 | static void io_ring_ctx_ref_free(struct percpu_ref *ref) | |
886 | { | |
887 | struct io_ring_ctx *ctx = container_of(ref, struct io_ring_ctx, refs); | |
888 | ||
206aefde | 889 | complete(&ctx->completions[0]); |
2b188cc1 JA |
890 | } |
891 | ||
892 | static struct io_ring_ctx *io_ring_ctx_alloc(struct io_uring_params *p) | |
893 | { | |
894 | struct io_ring_ctx *ctx; | |
78076bb6 | 895 | int hash_bits; |
2b188cc1 JA |
896 | |
897 | ctx = kzalloc(sizeof(*ctx), GFP_KERNEL); | |
898 | if (!ctx) | |
899 | return NULL; | |
900 | ||
0ddf92e8 JA |
901 | ctx->fallback_req = kmem_cache_alloc(req_cachep, GFP_KERNEL); |
902 | if (!ctx->fallback_req) | |
903 | goto err; | |
904 | ||
206aefde JA |
905 | ctx->completions = kmalloc(2 * sizeof(struct completion), GFP_KERNEL); |
906 | if (!ctx->completions) | |
907 | goto err; | |
908 | ||
78076bb6 JA |
909 | /* |
910 | * Use 5 bits less than the max cq entries, that should give us around | |
911 | * 32 entries per hash list if totally full and uniformly spread. | |
912 | */ | |
913 | hash_bits = ilog2(p->cq_entries); | |
914 | hash_bits -= 5; | |
915 | if (hash_bits <= 0) | |
916 | hash_bits = 1; | |
917 | ctx->cancel_hash_bits = hash_bits; | |
918 | ctx->cancel_hash = kmalloc((1U << hash_bits) * sizeof(struct hlist_head), | |
919 | GFP_KERNEL); | |
920 | if (!ctx->cancel_hash) | |
921 | goto err; | |
922 | __hash_init(ctx->cancel_hash, 1U << hash_bits); | |
923 | ||
21482896 | 924 | if (percpu_ref_init(&ctx->refs, io_ring_ctx_ref_free, |
206aefde JA |
925 | PERCPU_REF_ALLOW_REINIT, GFP_KERNEL)) |
926 | goto err; | |
2b188cc1 JA |
927 | |
928 | ctx->flags = p->flags; | |
583863ed | 929 | init_waitqueue_head(&ctx->sqo_wait); |
2b188cc1 | 930 | init_waitqueue_head(&ctx->cq_wait); |
1d7bb1d5 | 931 | INIT_LIST_HEAD(&ctx->cq_overflow_list); |
206aefde JA |
932 | init_completion(&ctx->completions[0]); |
933 | init_completion(&ctx->completions[1]); | |
5a2e745d | 934 | idr_init(&ctx->io_buffer_idr); |
071698e1 | 935 | idr_init(&ctx->personality_idr); |
2b188cc1 JA |
936 | mutex_init(&ctx->uring_lock); |
937 | init_waitqueue_head(&ctx->wait); | |
938 | spin_lock_init(&ctx->completion_lock); | |
def596e9 | 939 | INIT_LIST_HEAD(&ctx->poll_list); |
de0617e4 | 940 | INIT_LIST_HEAD(&ctx->defer_list); |
5262f567 | 941 | INIT_LIST_HEAD(&ctx->timeout_list); |
fcb323cc JA |
942 | init_waitqueue_head(&ctx->inflight_wait); |
943 | spin_lock_init(&ctx->inflight_lock); | |
944 | INIT_LIST_HEAD(&ctx->inflight_list); | |
2b188cc1 | 945 | return ctx; |
206aefde | 946 | err: |
0ddf92e8 JA |
947 | if (ctx->fallback_req) |
948 | kmem_cache_free(req_cachep, ctx->fallback_req); | |
206aefde | 949 | kfree(ctx->completions); |
78076bb6 | 950 | kfree(ctx->cancel_hash); |
206aefde JA |
951 | kfree(ctx); |
952 | return NULL; | |
2b188cc1 JA |
953 | } |
954 | ||
9d858b21 | 955 | static inline bool __req_need_defer(struct io_kiocb *req) |
7adf4eaf | 956 | { |
a197f664 JL |
957 | struct io_ring_ctx *ctx = req->ctx; |
958 | ||
31af27c7 PB |
959 | return req->sequence != ctx->cached_cq_tail |
960 | + atomic_read(&ctx->cached_cq_overflow); | |
7adf4eaf JA |
961 | } |
962 | ||
9d858b21 | 963 | static inline bool req_need_defer(struct io_kiocb *req) |
de0617e4 | 964 | { |
87987898 | 965 | if (unlikely(req->flags & REQ_F_IO_DRAIN)) |
9d858b21 | 966 | return __req_need_defer(req); |
de0617e4 | 967 | |
9d858b21 | 968 | return false; |
de0617e4 JA |
969 | } |
970 | ||
7adf4eaf | 971 | static struct io_kiocb *io_get_deferred_req(struct io_ring_ctx *ctx) |
de0617e4 JA |
972 | { |
973 | struct io_kiocb *req; | |
974 | ||
7adf4eaf | 975 | req = list_first_entry_or_null(&ctx->defer_list, struct io_kiocb, list); |
9d858b21 | 976 | if (req && !req_need_defer(req)) { |
de0617e4 JA |
977 | list_del_init(&req->list); |
978 | return req; | |
979 | } | |
980 | ||
981 | return NULL; | |
982 | } | |
983 | ||
5262f567 JA |
984 | static struct io_kiocb *io_get_timeout_req(struct io_ring_ctx *ctx) |
985 | { | |
7adf4eaf JA |
986 | struct io_kiocb *req; |
987 | ||
988 | req = list_first_entry_or_null(&ctx->timeout_list, struct io_kiocb, list); | |
93bd25bb JA |
989 | if (req) { |
990 | if (req->flags & REQ_F_TIMEOUT_NOSEQ) | |
991 | return NULL; | |
fb4b3d3f | 992 | if (!__req_need_defer(req)) { |
93bd25bb JA |
993 | list_del_init(&req->list); |
994 | return req; | |
995 | } | |
7adf4eaf JA |
996 | } |
997 | ||
998 | return NULL; | |
5262f567 JA |
999 | } |
1000 | ||
de0617e4 | 1001 | static void __io_commit_cqring(struct io_ring_ctx *ctx) |
2b188cc1 | 1002 | { |
75b28aff | 1003 | struct io_rings *rings = ctx->rings; |
2b188cc1 | 1004 | |
07910158 PB |
1005 | /* order cqe stores with ring update */ |
1006 | smp_store_release(&rings->cq.tail, ctx->cached_cq_tail); | |
2b188cc1 | 1007 | |
07910158 PB |
1008 | if (wq_has_sleeper(&ctx->cq_wait)) { |
1009 | wake_up_interruptible(&ctx->cq_wait); | |
1010 | kill_fasync(&ctx->cq_fasync, SIGIO, POLL_IN); | |
2b188cc1 JA |
1011 | } |
1012 | } | |
1013 | ||
cccf0ee8 JA |
1014 | static inline void io_req_work_grab_env(struct io_kiocb *req, |
1015 | const struct io_op_def *def) | |
1016 | { | |
1017 | if (!req->work.mm && def->needs_mm) { | |
1018 | mmgrab(current->mm); | |
1019 | req->work.mm = current->mm; | |
2b188cc1 | 1020 | } |
cccf0ee8 JA |
1021 | if (!req->work.creds) |
1022 | req->work.creds = get_current_cred(); | |
ff002b30 JA |
1023 | if (!req->work.fs && def->needs_fs) { |
1024 | spin_lock(¤t->fs->lock); | |
1025 | if (!current->fs->in_exec) { | |
1026 | req->work.fs = current->fs; | |
1027 | req->work.fs->users++; | |
1028 | } else { | |
1029 | req->work.flags |= IO_WQ_WORK_CANCEL; | |
1030 | } | |
1031 | spin_unlock(¤t->fs->lock); | |
1032 | } | |
6ab23144 JA |
1033 | if (!req->work.task_pid) |
1034 | req->work.task_pid = task_pid_vnr(current); | |
2b188cc1 JA |
1035 | } |
1036 | ||
cccf0ee8 | 1037 | static inline void io_req_work_drop_env(struct io_kiocb *req) |
18d9be1a | 1038 | { |
cccf0ee8 JA |
1039 | if (req->work.mm) { |
1040 | mmdrop(req->work.mm); | |
1041 | req->work.mm = NULL; | |
1042 | } | |
1043 | if (req->work.creds) { | |
1044 | put_cred(req->work.creds); | |
1045 | req->work.creds = NULL; | |
1046 | } | |
ff002b30 JA |
1047 | if (req->work.fs) { |
1048 | struct fs_struct *fs = req->work.fs; | |
1049 | ||
1050 | spin_lock(&req->work.fs->lock); | |
1051 | if (--fs->users) | |
1052 | fs = NULL; | |
1053 | spin_unlock(&req->work.fs->lock); | |
1054 | if (fs) | |
1055 | free_fs_struct(fs); | |
1056 | } | |
561fb04a JA |
1057 | } |
1058 | ||
8766dd51 | 1059 | static inline void io_prep_async_work(struct io_kiocb *req, |
94ae5e77 | 1060 | struct io_kiocb **link) |
18d9be1a | 1061 | { |
d3656344 | 1062 | const struct io_op_def *def = &io_op_defs[req->opcode]; |
54a91f3b | 1063 | |
d3656344 JA |
1064 | if (req->flags & REQ_F_ISREG) { |
1065 | if (def->hash_reg_file) | |
8766dd51 | 1066 | io_wq_hash_work(&req->work, file_inode(req->file)); |
d3656344 JA |
1067 | } else { |
1068 | if (def->unbound_nonreg_file) | |
3529d8c2 | 1069 | req->work.flags |= IO_WQ_WORK_UNBOUND; |
54a91f3b | 1070 | } |
cccf0ee8 JA |
1071 | |
1072 | io_req_work_grab_env(req, def); | |
54a91f3b | 1073 | |
94ae5e77 | 1074 | *link = io_prep_linked_timeout(req); |
561fb04a JA |
1075 | } |
1076 | ||
a197f664 | 1077 | static inline void io_queue_async_work(struct io_kiocb *req) |
561fb04a | 1078 | { |
a197f664 | 1079 | struct io_ring_ctx *ctx = req->ctx; |
94ae5e77 | 1080 | struct io_kiocb *link; |
94ae5e77 | 1081 | |
8766dd51 | 1082 | io_prep_async_work(req, &link); |
561fb04a | 1083 | |
8766dd51 PB |
1084 | trace_io_uring_queue_async_work(ctx, io_wq_is_hashed(&req->work), req, |
1085 | &req->work, req->flags); | |
1086 | io_wq_enqueue(ctx->io_wq, &req->work); | |
94ae5e77 JA |
1087 | |
1088 | if (link) | |
1089 | io_queue_linked_timeout(link); | |
18d9be1a JA |
1090 | } |
1091 | ||
5262f567 JA |
1092 | static void io_kill_timeout(struct io_kiocb *req) |
1093 | { | |
1094 | int ret; | |
1095 | ||
2d28390a | 1096 | ret = hrtimer_try_to_cancel(&req->io->timeout.timer); |
5262f567 JA |
1097 | if (ret != -1) { |
1098 | atomic_inc(&req->ctx->cq_timeouts); | |
842f9612 | 1099 | list_del_init(&req->list); |
f0e20b89 | 1100 | req->flags |= REQ_F_COMP_LOCKED; |
78e19bbe | 1101 | io_cqring_fill_event(req, 0); |
ec9c02ad | 1102 | io_put_req(req); |
5262f567 JA |
1103 | } |
1104 | } | |
1105 | ||
1106 | static void io_kill_timeouts(struct io_ring_ctx *ctx) | |
1107 | { | |
1108 | struct io_kiocb *req, *tmp; | |
1109 | ||
1110 | spin_lock_irq(&ctx->completion_lock); | |
1111 | list_for_each_entry_safe(req, tmp, &ctx->timeout_list, list) | |
1112 | io_kill_timeout(req); | |
1113 | spin_unlock_irq(&ctx->completion_lock); | |
1114 | } | |
1115 | ||
de0617e4 JA |
1116 | static void io_commit_cqring(struct io_ring_ctx *ctx) |
1117 | { | |
1118 | struct io_kiocb *req; | |
1119 | ||
5262f567 JA |
1120 | while ((req = io_get_timeout_req(ctx)) != NULL) |
1121 | io_kill_timeout(req); | |
1122 | ||
de0617e4 JA |
1123 | __io_commit_cqring(ctx); |
1124 | ||
87987898 | 1125 | while ((req = io_get_deferred_req(ctx)) != NULL) |
a197f664 | 1126 | io_queue_async_work(req); |
de0617e4 JA |
1127 | } |
1128 | ||
2b188cc1 JA |
1129 | static struct io_uring_cqe *io_get_cqring(struct io_ring_ctx *ctx) |
1130 | { | |
75b28aff | 1131 | struct io_rings *rings = ctx->rings; |
2b188cc1 JA |
1132 | unsigned tail; |
1133 | ||
1134 | tail = ctx->cached_cq_tail; | |
115e12e5 SB |
1135 | /* |
1136 | * writes to the cq entry need to come after reading head; the | |
1137 | * control dependency is enough as we're using WRITE_ONCE to | |
1138 | * fill the cq entry | |
1139 | */ | |
75b28aff | 1140 | if (tail - READ_ONCE(rings->cq.head) == rings->cq_ring_entries) |
2b188cc1 JA |
1141 | return NULL; |
1142 | ||
1143 | ctx->cached_cq_tail++; | |
75b28aff | 1144 | return &rings->cqes[tail & ctx->cq_mask]; |
2b188cc1 JA |
1145 | } |
1146 | ||
f2842ab5 JA |
1147 | static inline bool io_should_trigger_evfd(struct io_ring_ctx *ctx) |
1148 | { | |
f0b493e6 JA |
1149 | if (!ctx->cq_ev_fd) |
1150 | return false; | |
f2842ab5 JA |
1151 | if (!ctx->eventfd_async) |
1152 | return true; | |
b41e9852 | 1153 | return io_wq_current_is_worker(); |
f2842ab5 JA |
1154 | } |
1155 | ||
b41e9852 | 1156 | static void io_cqring_ev_posted(struct io_ring_ctx *ctx) |
1d7bb1d5 JA |
1157 | { |
1158 | if (waitqueue_active(&ctx->wait)) | |
1159 | wake_up(&ctx->wait); | |
1160 | if (waitqueue_active(&ctx->sqo_wait)) | |
1161 | wake_up(&ctx->sqo_wait); | |
b41e9852 | 1162 | if (io_should_trigger_evfd(ctx)) |
1d7bb1d5 JA |
1163 | eventfd_signal(ctx->cq_ev_fd, 1); |
1164 | } | |
1165 | ||
c4a2ed72 JA |
1166 | /* Returns true if there are no backlogged entries after the flush */ |
1167 | static bool io_cqring_overflow_flush(struct io_ring_ctx *ctx, bool force) | |
1d7bb1d5 JA |
1168 | { |
1169 | struct io_rings *rings = ctx->rings; | |
1170 | struct io_uring_cqe *cqe; | |
1171 | struct io_kiocb *req; | |
1172 | unsigned long flags; | |
1173 | LIST_HEAD(list); | |
1174 | ||
1175 | if (!force) { | |
1176 | if (list_empty_careful(&ctx->cq_overflow_list)) | |
c4a2ed72 | 1177 | return true; |
1d7bb1d5 JA |
1178 | if ((ctx->cached_cq_tail - READ_ONCE(rings->cq.head) == |
1179 | rings->cq_ring_entries)) | |
c4a2ed72 | 1180 | return false; |
1d7bb1d5 JA |
1181 | } |
1182 | ||
1183 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
1184 | ||
1185 | /* if force is set, the ring is going away. always drop after that */ | |
1186 | if (force) | |
69b3e546 | 1187 | ctx->cq_overflow_flushed = 1; |
1d7bb1d5 | 1188 | |
c4a2ed72 | 1189 | cqe = NULL; |
1d7bb1d5 JA |
1190 | while (!list_empty(&ctx->cq_overflow_list)) { |
1191 | cqe = io_get_cqring(ctx); | |
1192 | if (!cqe && !force) | |
1193 | break; | |
1194 | ||
1195 | req = list_first_entry(&ctx->cq_overflow_list, struct io_kiocb, | |
1196 | list); | |
1197 | list_move(&req->list, &list); | |
2ca10259 | 1198 | req->flags &= ~REQ_F_OVERFLOW; |
1d7bb1d5 JA |
1199 | if (cqe) { |
1200 | WRITE_ONCE(cqe->user_data, req->user_data); | |
1201 | WRITE_ONCE(cqe->res, req->result); | |
bcda7baa | 1202 | WRITE_ONCE(cqe->flags, req->cflags); |
1d7bb1d5 JA |
1203 | } else { |
1204 | WRITE_ONCE(ctx->rings->cq_overflow, | |
1205 | atomic_inc_return(&ctx->cached_cq_overflow)); | |
1206 | } | |
1207 | } | |
1208 | ||
1209 | io_commit_cqring(ctx); | |
ad3eb2c8 JA |
1210 | if (cqe) { |
1211 | clear_bit(0, &ctx->sq_check_overflow); | |
1212 | clear_bit(0, &ctx->cq_check_overflow); | |
1213 | } | |
1d7bb1d5 JA |
1214 | spin_unlock_irqrestore(&ctx->completion_lock, flags); |
1215 | io_cqring_ev_posted(ctx); | |
1216 | ||
1217 | while (!list_empty(&list)) { | |
1218 | req = list_first_entry(&list, struct io_kiocb, list); | |
1219 | list_del(&req->list); | |
ec9c02ad | 1220 | io_put_req(req); |
1d7bb1d5 | 1221 | } |
c4a2ed72 JA |
1222 | |
1223 | return cqe != NULL; | |
1d7bb1d5 JA |
1224 | } |
1225 | ||
bcda7baa | 1226 | static void __io_cqring_fill_event(struct io_kiocb *req, long res, long cflags) |
2b188cc1 | 1227 | { |
78e19bbe | 1228 | struct io_ring_ctx *ctx = req->ctx; |
2b188cc1 JA |
1229 | struct io_uring_cqe *cqe; |
1230 | ||
78e19bbe | 1231 | trace_io_uring_complete(ctx, req->user_data, res); |
51c3ff62 | 1232 | |
2b188cc1 JA |
1233 | /* |
1234 | * If we can't get a cq entry, userspace overflowed the | |
1235 | * submission (by quite a lot). Increment the overflow count in | |
1236 | * the ring. | |
1237 | */ | |
1238 | cqe = io_get_cqring(ctx); | |
1d7bb1d5 | 1239 | if (likely(cqe)) { |
78e19bbe | 1240 | WRITE_ONCE(cqe->user_data, req->user_data); |
2b188cc1 | 1241 | WRITE_ONCE(cqe->res, res); |
bcda7baa | 1242 | WRITE_ONCE(cqe->flags, cflags); |
1d7bb1d5 | 1243 | } else if (ctx->cq_overflow_flushed) { |
498ccd9e JA |
1244 | WRITE_ONCE(ctx->rings->cq_overflow, |
1245 | atomic_inc_return(&ctx->cached_cq_overflow)); | |
1d7bb1d5 | 1246 | } else { |
ad3eb2c8 JA |
1247 | if (list_empty(&ctx->cq_overflow_list)) { |
1248 | set_bit(0, &ctx->sq_check_overflow); | |
1249 | set_bit(0, &ctx->cq_check_overflow); | |
1250 | } | |
2ca10259 | 1251 | req->flags |= REQ_F_OVERFLOW; |
1d7bb1d5 JA |
1252 | refcount_inc(&req->refs); |
1253 | req->result = res; | |
bcda7baa | 1254 | req->cflags = cflags; |
1d7bb1d5 | 1255 | list_add_tail(&req->list, &ctx->cq_overflow_list); |
2b188cc1 JA |
1256 | } |
1257 | } | |
1258 | ||
bcda7baa JA |
1259 | static void io_cqring_fill_event(struct io_kiocb *req, long res) |
1260 | { | |
1261 | __io_cqring_fill_event(req, res, 0); | |
1262 | } | |
1263 | ||
1264 | static void __io_cqring_add_event(struct io_kiocb *req, long res, long cflags) | |
2b188cc1 | 1265 | { |
78e19bbe | 1266 | struct io_ring_ctx *ctx = req->ctx; |
2b188cc1 JA |
1267 | unsigned long flags; |
1268 | ||
1269 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
bcda7baa | 1270 | __io_cqring_fill_event(req, res, cflags); |
2b188cc1 JA |
1271 | io_commit_cqring(ctx); |
1272 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
1273 | ||
8c838788 | 1274 | io_cqring_ev_posted(ctx); |
2b188cc1 JA |
1275 | } |
1276 | ||
bcda7baa JA |
1277 | static void io_cqring_add_event(struct io_kiocb *req, long res) |
1278 | { | |
1279 | __io_cqring_add_event(req, res, 0); | |
1280 | } | |
1281 | ||
0ddf92e8 JA |
1282 | static inline bool io_is_fallback_req(struct io_kiocb *req) |
1283 | { | |
1284 | return req == (struct io_kiocb *) | |
1285 | ((unsigned long) req->ctx->fallback_req & ~1UL); | |
1286 | } | |
1287 | ||
1288 | static struct io_kiocb *io_get_fallback_req(struct io_ring_ctx *ctx) | |
1289 | { | |
1290 | struct io_kiocb *req; | |
1291 | ||
1292 | req = ctx->fallback_req; | |
dd461af6 | 1293 | if (!test_and_set_bit_lock(0, (unsigned long *) &ctx->fallback_req)) |
0ddf92e8 JA |
1294 | return req; |
1295 | ||
1296 | return NULL; | |
1297 | } | |
1298 | ||
0553b8bd PB |
1299 | static struct io_kiocb *io_alloc_req(struct io_ring_ctx *ctx, |
1300 | struct io_submit_state *state) | |
2b188cc1 | 1301 | { |
fd6fab2c | 1302 | gfp_t gfp = GFP_KERNEL | __GFP_NOWARN; |
2b188cc1 JA |
1303 | struct io_kiocb *req; |
1304 | ||
2579f913 | 1305 | if (!state) { |
fd6fab2c | 1306 | req = kmem_cache_alloc(req_cachep, gfp); |
2579f913 | 1307 | if (unlikely(!req)) |
0ddf92e8 | 1308 | goto fallback; |
2579f913 JA |
1309 | } else if (!state->free_reqs) { |
1310 | size_t sz; | |
1311 | int ret; | |
1312 | ||
1313 | sz = min_t(size_t, state->ios_left, ARRAY_SIZE(state->reqs)); | |
fd6fab2c JA |
1314 | ret = kmem_cache_alloc_bulk(req_cachep, gfp, sz, state->reqs); |
1315 | ||
1316 | /* | |
1317 | * Bulk alloc is all-or-nothing. If we fail to get a batch, | |
1318 | * retry single alloc to be on the safe side. | |
1319 | */ | |
1320 | if (unlikely(ret <= 0)) { | |
1321 | state->reqs[0] = kmem_cache_alloc(req_cachep, gfp); | |
1322 | if (!state->reqs[0]) | |
0ddf92e8 | 1323 | goto fallback; |
fd6fab2c JA |
1324 | ret = 1; |
1325 | } | |
2579f913 | 1326 | state->free_reqs = ret - 1; |
6c8a3134 | 1327 | req = state->reqs[ret - 1]; |
2579f913 | 1328 | } else { |
2579f913 | 1329 | state->free_reqs--; |
6c8a3134 | 1330 | req = state->reqs[state->free_reqs]; |
2b188cc1 JA |
1331 | } |
1332 | ||
2579f913 | 1333 | return req; |
0ddf92e8 | 1334 | fallback: |
0553b8bd | 1335 | return io_get_fallback_req(ctx); |
2b188cc1 JA |
1336 | } |
1337 | ||
8da11c19 PB |
1338 | static inline void io_put_file(struct io_kiocb *req, struct file *file, |
1339 | bool fixed) | |
1340 | { | |
1341 | if (fixed) | |
05589553 | 1342 | percpu_ref_put(req->fixed_file_refs); |
8da11c19 PB |
1343 | else |
1344 | fput(file); | |
1345 | } | |
1346 | ||
c6ca97b3 | 1347 | static void __io_req_aux_free(struct io_kiocb *req) |
2b188cc1 | 1348 | { |
929a3af9 PB |
1349 | if (req->flags & REQ_F_NEED_CLEANUP) |
1350 | io_cleanup_req(req); | |
1351 | ||
96fd84d8 | 1352 | kfree(req->io); |
8da11c19 PB |
1353 | if (req->file) |
1354 | io_put_file(req, req->file, (req->flags & REQ_F_FIXED_FILE)); | |
3537b6a7 JA |
1355 | if (req->task) |
1356 | put_task_struct(req->task); | |
cccf0ee8 JA |
1357 | |
1358 | io_req_work_drop_env(req); | |
def596e9 JA |
1359 | } |
1360 | ||
9e645e11 | 1361 | static void __io_free_req(struct io_kiocb *req) |
2b188cc1 | 1362 | { |
c6ca97b3 | 1363 | __io_req_aux_free(req); |
fcb323cc | 1364 | |
fcb323cc | 1365 | if (req->flags & REQ_F_INFLIGHT) { |
c6ca97b3 | 1366 | struct io_ring_ctx *ctx = req->ctx; |
fcb323cc JA |
1367 | unsigned long flags; |
1368 | ||
1369 | spin_lock_irqsave(&ctx->inflight_lock, flags); | |
1370 | list_del(&req->inflight_entry); | |
1371 | if (waitqueue_active(&ctx->inflight_wait)) | |
1372 | wake_up(&ctx->inflight_wait); | |
1373 | spin_unlock_irqrestore(&ctx->inflight_lock, flags); | |
1374 | } | |
2b85edfc PB |
1375 | |
1376 | percpu_ref_put(&req->ctx->refs); | |
b1e50e54 PB |
1377 | if (likely(!io_is_fallback_req(req))) |
1378 | kmem_cache_free(req_cachep, req); | |
1379 | else | |
dd461af6 | 1380 | clear_bit_unlock(0, (unsigned long *) &req->ctx->fallback_req); |
e65ef56d JA |
1381 | } |
1382 | ||
c6ca97b3 JA |
1383 | struct req_batch { |
1384 | void *reqs[IO_IOPOLL_BATCH]; | |
1385 | int to_free; | |
1386 | int need_iter; | |
1387 | }; | |
1388 | ||
1389 | static void io_free_req_many(struct io_ring_ctx *ctx, struct req_batch *rb) | |
1390 | { | |
1391 | if (!rb->to_free) | |
1392 | return; | |
1393 | if (rb->need_iter) { | |
1394 | int i, inflight = 0; | |
1395 | unsigned long flags; | |
1396 | ||
1397 | for (i = 0; i < rb->to_free; i++) { | |
1398 | struct io_kiocb *req = rb->reqs[i]; | |
1399 | ||
c6ca97b3 JA |
1400 | if (req->flags & REQ_F_INFLIGHT) |
1401 | inflight++; | |
c6ca97b3 JA |
1402 | __io_req_aux_free(req); |
1403 | } | |
1404 | if (!inflight) | |
1405 | goto do_free; | |
1406 | ||
1407 | spin_lock_irqsave(&ctx->inflight_lock, flags); | |
1408 | for (i = 0; i < rb->to_free; i++) { | |
1409 | struct io_kiocb *req = rb->reqs[i]; | |
1410 | ||
10fef4be | 1411 | if (req->flags & REQ_F_INFLIGHT) { |
c6ca97b3 JA |
1412 | list_del(&req->inflight_entry); |
1413 | if (!--inflight) | |
1414 | break; | |
1415 | } | |
1416 | } | |
1417 | spin_unlock_irqrestore(&ctx->inflight_lock, flags); | |
1418 | ||
1419 | if (waitqueue_active(&ctx->inflight_wait)) | |
1420 | wake_up(&ctx->inflight_wait); | |
1421 | } | |
1422 | do_free: | |
1423 | kmem_cache_free_bulk(req_cachep, rb->to_free, rb->reqs); | |
1424 | percpu_ref_put_many(&ctx->refs, rb->to_free); | |
c6ca97b3 | 1425 | rb->to_free = rb->need_iter = 0; |
e65ef56d JA |
1426 | } |
1427 | ||
a197f664 | 1428 | static bool io_link_cancel_timeout(struct io_kiocb *req) |
2665abfd | 1429 | { |
a197f664 | 1430 | struct io_ring_ctx *ctx = req->ctx; |
2665abfd JA |
1431 | int ret; |
1432 | ||
2d28390a | 1433 | ret = hrtimer_try_to_cancel(&req->io->timeout.timer); |
2665abfd | 1434 | if (ret != -1) { |
78e19bbe | 1435 | io_cqring_fill_event(req, -ECANCELED); |
2665abfd | 1436 | io_commit_cqring(ctx); |
dea3b49c | 1437 | req->flags &= ~REQ_F_LINK_HEAD; |
ec9c02ad | 1438 | io_put_req(req); |
2665abfd JA |
1439 | return true; |
1440 | } | |
1441 | ||
1442 | return false; | |
e65ef56d JA |
1443 | } |
1444 | ||
ba816ad6 | 1445 | static void io_req_link_next(struct io_kiocb *req, struct io_kiocb **nxtptr) |
9e645e11 | 1446 | { |
2665abfd | 1447 | struct io_ring_ctx *ctx = req->ctx; |
2665abfd | 1448 | bool wake_ev = false; |
9e645e11 | 1449 | |
4d7dd462 JA |
1450 | /* Already got next link */ |
1451 | if (req->flags & REQ_F_LINK_NEXT) | |
1452 | return; | |
1453 | ||
9e645e11 JA |
1454 | /* |
1455 | * The list should never be empty when we are called here. But could | |
1456 | * potentially happen if the chain is messed up, check to be on the | |
1457 | * safe side. | |
1458 | */ | |
4493233e PB |
1459 | while (!list_empty(&req->link_list)) { |
1460 | struct io_kiocb *nxt = list_first_entry(&req->link_list, | |
1461 | struct io_kiocb, link_list); | |
94ae5e77 | 1462 | |
4493233e PB |
1463 | if (unlikely((req->flags & REQ_F_LINK_TIMEOUT) && |
1464 | (nxt->flags & REQ_F_TIMEOUT))) { | |
1465 | list_del_init(&nxt->link_list); | |
94ae5e77 | 1466 | wake_ev |= io_link_cancel_timeout(nxt); |
94ae5e77 JA |
1467 | req->flags &= ~REQ_F_LINK_TIMEOUT; |
1468 | continue; | |
1469 | } | |
9e645e11 | 1470 | |
4493233e PB |
1471 | list_del_init(&req->link_list); |
1472 | if (!list_empty(&nxt->link_list)) | |
dea3b49c | 1473 | nxt->flags |= REQ_F_LINK_HEAD; |
b18fdf71 | 1474 | *nxtptr = nxt; |
94ae5e77 | 1475 | break; |
9e645e11 | 1476 | } |
2665abfd | 1477 | |
4d7dd462 | 1478 | req->flags |= REQ_F_LINK_NEXT; |
2665abfd JA |
1479 | if (wake_ev) |
1480 | io_cqring_ev_posted(ctx); | |
9e645e11 JA |
1481 | } |
1482 | ||
1483 | /* | |
dea3b49c | 1484 | * Called if REQ_F_LINK_HEAD is set, and we fail the head request |
9e645e11 JA |
1485 | */ |
1486 | static void io_fail_links(struct io_kiocb *req) | |
1487 | { | |
2665abfd | 1488 | struct io_ring_ctx *ctx = req->ctx; |
2665abfd JA |
1489 | unsigned long flags; |
1490 | ||
1491 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
9e645e11 JA |
1492 | |
1493 | while (!list_empty(&req->link_list)) { | |
4493233e PB |
1494 | struct io_kiocb *link = list_first_entry(&req->link_list, |
1495 | struct io_kiocb, link_list); | |
9e645e11 | 1496 | |
4493233e | 1497 | list_del_init(&link->link_list); |
c826bd7a | 1498 | trace_io_uring_fail_link(req, link); |
2665abfd JA |
1499 | |
1500 | if ((req->flags & REQ_F_LINK_TIMEOUT) && | |
d625c6ee | 1501 | link->opcode == IORING_OP_LINK_TIMEOUT) { |
a197f664 | 1502 | io_link_cancel_timeout(link); |
2665abfd | 1503 | } else { |
78e19bbe | 1504 | io_cqring_fill_event(link, -ECANCELED); |
978db57e | 1505 | __io_double_put_req(link); |
2665abfd | 1506 | } |
5d960724 | 1507 | req->flags &= ~REQ_F_LINK_TIMEOUT; |
9e645e11 | 1508 | } |
2665abfd JA |
1509 | |
1510 | io_commit_cqring(ctx); | |
1511 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
1512 | io_cqring_ev_posted(ctx); | |
9e645e11 JA |
1513 | } |
1514 | ||
4d7dd462 | 1515 | static void io_req_find_next(struct io_kiocb *req, struct io_kiocb **nxt) |
9e645e11 | 1516 | { |
dea3b49c | 1517 | if (likely(!(req->flags & REQ_F_LINK_HEAD))) |
2665abfd | 1518 | return; |
2665abfd | 1519 | |
9e645e11 JA |
1520 | /* |
1521 | * If LINK is set, we have dependent requests in this chain. If we | |
1522 | * didn't fail this request, queue the first one up, moving any other | |
1523 | * dependencies to the next request. In case of failure, fail the rest | |
1524 | * of the chain. | |
1525 | */ | |
2665abfd JA |
1526 | if (req->flags & REQ_F_FAIL_LINK) { |
1527 | io_fail_links(req); | |
7c9e7f0f JA |
1528 | } else if ((req->flags & (REQ_F_LINK_TIMEOUT | REQ_F_COMP_LOCKED)) == |
1529 | REQ_F_LINK_TIMEOUT) { | |
2665abfd JA |
1530 | struct io_ring_ctx *ctx = req->ctx; |
1531 | unsigned long flags; | |
1532 | ||
1533 | /* | |
1534 | * If this is a timeout link, we could be racing with the | |
1535 | * timeout timer. Grab the completion lock for this case to | |
7c9e7f0f | 1536 | * protect against that. |
2665abfd JA |
1537 | */ |
1538 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
1539 | io_req_link_next(req, nxt); | |
1540 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
1541 | } else { | |
1542 | io_req_link_next(req, nxt); | |
9e645e11 | 1543 | } |
4d7dd462 | 1544 | } |
9e645e11 | 1545 | |
c69f8dbe JL |
1546 | static void io_free_req(struct io_kiocb *req) |
1547 | { | |
944e58bf PB |
1548 | struct io_kiocb *nxt = NULL; |
1549 | ||
1550 | io_req_find_next(req, &nxt); | |
70cf9f32 | 1551 | __io_free_req(req); |
944e58bf PB |
1552 | |
1553 | if (nxt) | |
1554 | io_queue_async_work(nxt); | |
c69f8dbe JL |
1555 | } |
1556 | ||
7a743e22 PB |
1557 | static void io_link_work_cb(struct io_wq_work **workptr) |
1558 | { | |
18a542ff PB |
1559 | struct io_kiocb *req = container_of(*workptr, struct io_kiocb, work); |
1560 | struct io_kiocb *link; | |
7a743e22 | 1561 | |
18a542ff | 1562 | link = list_first_entry(&req->link_list, struct io_kiocb, link_list); |
7a743e22 PB |
1563 | io_queue_linked_timeout(link); |
1564 | io_wq_submit_work(workptr); | |
1565 | } | |
1566 | ||
1567 | static void io_wq_assign_next(struct io_wq_work **workptr, struct io_kiocb *nxt) | |
1568 | { | |
1569 | struct io_kiocb *link; | |
8766dd51 PB |
1570 | const struct io_op_def *def = &io_op_defs[nxt->opcode]; |
1571 | ||
1572 | if ((nxt->flags & REQ_F_ISREG) && def->hash_reg_file) | |
1573 | io_wq_hash_work(&nxt->work, file_inode(nxt->file)); | |
7a743e22 PB |
1574 | |
1575 | *workptr = &nxt->work; | |
1576 | link = io_prep_linked_timeout(nxt); | |
18a542ff | 1577 | if (link) |
7a743e22 | 1578 | nxt->work.func = io_link_work_cb; |
7a743e22 PB |
1579 | } |
1580 | ||
ba816ad6 JA |
1581 | /* |
1582 | * Drop reference to request, return next in chain (if there is one) if this | |
1583 | * was the last reference to this request. | |
1584 | */ | |
f9bd67f6 | 1585 | __attribute__((nonnull)) |
ec9c02ad | 1586 | static void io_put_req_find_next(struct io_kiocb *req, struct io_kiocb **nxtptr) |
e65ef56d | 1587 | { |
2a44f467 JA |
1588 | if (refcount_dec_and_test(&req->refs)) { |
1589 | io_req_find_next(req, nxtptr); | |
4d7dd462 | 1590 | __io_free_req(req); |
2a44f467 | 1591 | } |
2b188cc1 JA |
1592 | } |
1593 | ||
e65ef56d JA |
1594 | static void io_put_req(struct io_kiocb *req) |
1595 | { | |
1596 | if (refcount_dec_and_test(&req->refs)) | |
1597 | io_free_req(req); | |
2b188cc1 JA |
1598 | } |
1599 | ||
e9fd9396 PB |
1600 | static void io_steal_work(struct io_kiocb *req, |
1601 | struct io_wq_work **workptr) | |
7a743e22 PB |
1602 | { |
1603 | /* | |
1604 | * It's in an io-wq worker, so there always should be at least | |
1605 | * one reference, which will be dropped in io_put_work() just | |
1606 | * after the current handler returns. | |
1607 | * | |
1608 | * It also means, that if the counter dropped to 1, then there is | |
1609 | * no asynchronous users left, so it's safe to steal the next work. | |
1610 | */ | |
7a743e22 PB |
1611 | if (refcount_read(&req->refs) == 1) { |
1612 | struct io_kiocb *nxt = NULL; | |
1613 | ||
1614 | io_req_find_next(req, &nxt); | |
1615 | if (nxt) | |
1616 | io_wq_assign_next(workptr, nxt); | |
1617 | } | |
1618 | } | |
1619 | ||
978db57e JA |
1620 | /* |
1621 | * Must only be used if we don't need to care about links, usually from | |
1622 | * within the completion handling itself. | |
1623 | */ | |
1624 | static void __io_double_put_req(struct io_kiocb *req) | |
78e19bbe JA |
1625 | { |
1626 | /* drop both submit and complete references */ | |
1627 | if (refcount_sub_and_test(2, &req->refs)) | |
1628 | __io_free_req(req); | |
1629 | } | |
1630 | ||
978db57e JA |
1631 | static void io_double_put_req(struct io_kiocb *req) |
1632 | { | |
1633 | /* drop both submit and complete references */ | |
1634 | if (refcount_sub_and_test(2, &req->refs)) | |
1635 | io_free_req(req); | |
1636 | } | |
1637 | ||
1d7bb1d5 | 1638 | static unsigned io_cqring_events(struct io_ring_ctx *ctx, bool noflush) |
a3a0e43f | 1639 | { |
84f97dc2 JA |
1640 | struct io_rings *rings = ctx->rings; |
1641 | ||
ad3eb2c8 JA |
1642 | if (test_bit(0, &ctx->cq_check_overflow)) { |
1643 | /* | |
1644 | * noflush == true is from the waitqueue handler, just ensure | |
1645 | * we wake up the task, and the next invocation will flush the | |
1646 | * entries. We cannot safely to it from here. | |
1647 | */ | |
1648 | if (noflush && !list_empty(&ctx->cq_overflow_list)) | |
1649 | return -1U; | |
1d7bb1d5 | 1650 | |
ad3eb2c8 JA |
1651 | io_cqring_overflow_flush(ctx, false); |
1652 | } | |
1d7bb1d5 | 1653 | |
a3a0e43f JA |
1654 | /* See comment at the top of this file */ |
1655 | smp_rmb(); | |
ad3eb2c8 | 1656 | return ctx->cached_cq_tail - READ_ONCE(rings->cq.head); |
a3a0e43f JA |
1657 | } |
1658 | ||
fb5ccc98 PB |
1659 | static inline unsigned int io_sqring_entries(struct io_ring_ctx *ctx) |
1660 | { | |
1661 | struct io_rings *rings = ctx->rings; | |
1662 | ||
1663 | /* make sure SQ entry isn't read before tail */ | |
1664 | return smp_load_acquire(&rings->sq.tail) - ctx->cached_sq_head; | |
1665 | } | |
1666 | ||
8237e045 | 1667 | static inline bool io_req_multi_free(struct req_batch *rb, struct io_kiocb *req) |
e94f141b | 1668 | { |
dea3b49c | 1669 | if ((req->flags & REQ_F_LINK_HEAD) || io_is_fallback_req(req)) |
c6ca97b3 | 1670 | return false; |
e94f141b | 1671 | |
9d9e88a2 | 1672 | if (req->file || req->io) |
c6ca97b3 JA |
1673 | rb->need_iter++; |
1674 | ||
1675 | rb->reqs[rb->to_free++] = req; | |
1676 | if (unlikely(rb->to_free == ARRAY_SIZE(rb->reqs))) | |
1677 | io_free_req_many(req->ctx, rb); | |
1678 | return true; | |
e94f141b JA |
1679 | } |
1680 | ||
bcda7baa JA |
1681 | static int io_put_kbuf(struct io_kiocb *req) |
1682 | { | |
4d954c25 | 1683 | struct io_buffer *kbuf; |
bcda7baa JA |
1684 | int cflags; |
1685 | ||
4d954c25 | 1686 | kbuf = (struct io_buffer *) (unsigned long) req->rw.addr; |
bcda7baa JA |
1687 | cflags = kbuf->bid << IORING_CQE_BUFFER_SHIFT; |
1688 | cflags |= IORING_CQE_F_BUFFER; | |
1689 | req->rw.addr = 0; | |
1690 | kfree(kbuf); | |
1691 | return cflags; | |
1692 | } | |
1693 | ||
def596e9 JA |
1694 | /* |
1695 | * Find and free completed poll iocbs | |
1696 | */ | |
1697 | static void io_iopoll_complete(struct io_ring_ctx *ctx, unsigned int *nr_events, | |
1698 | struct list_head *done) | |
1699 | { | |
8237e045 | 1700 | struct req_batch rb; |
def596e9 | 1701 | struct io_kiocb *req; |
def596e9 | 1702 | |
c6ca97b3 | 1703 | rb.to_free = rb.need_iter = 0; |
def596e9 | 1704 | while (!list_empty(done)) { |
bcda7baa JA |
1705 | int cflags = 0; |
1706 | ||
def596e9 JA |
1707 | req = list_first_entry(done, struct io_kiocb, list); |
1708 | list_del(&req->list); | |
1709 | ||
bcda7baa JA |
1710 | if (req->flags & REQ_F_BUFFER_SELECTED) |
1711 | cflags = io_put_kbuf(req); | |
1712 | ||
1713 | __io_cqring_fill_event(req, req->result, cflags); | |
def596e9 JA |
1714 | (*nr_events)++; |
1715 | ||
8237e045 JA |
1716 | if (refcount_dec_and_test(&req->refs) && |
1717 | !io_req_multi_free(&rb, req)) | |
1718 | io_free_req(req); | |
def596e9 | 1719 | } |
def596e9 | 1720 | |
09bb8394 | 1721 | io_commit_cqring(ctx); |
32b2244a XW |
1722 | if (ctx->flags & IORING_SETUP_SQPOLL) |
1723 | io_cqring_ev_posted(ctx); | |
8237e045 | 1724 | io_free_req_many(ctx, &rb); |
def596e9 JA |
1725 | } |
1726 | ||
581f9810 BM |
1727 | static void io_iopoll_queue(struct list_head *again) |
1728 | { | |
1729 | struct io_kiocb *req; | |
1730 | ||
1731 | do { | |
1732 | req = list_first_entry(again, struct io_kiocb, list); | |
1733 | list_del(&req->list); | |
1734 | refcount_inc(&req->refs); | |
1735 | io_queue_async_work(req); | |
1736 | } while (!list_empty(again)); | |
1737 | } | |
1738 | ||
def596e9 JA |
1739 | static int io_do_iopoll(struct io_ring_ctx *ctx, unsigned int *nr_events, |
1740 | long min) | |
1741 | { | |
1742 | struct io_kiocb *req, *tmp; | |
1743 | LIST_HEAD(done); | |
581f9810 | 1744 | LIST_HEAD(again); |
def596e9 JA |
1745 | bool spin; |
1746 | int ret; | |
1747 | ||
1748 | /* | |
1749 | * Only spin for completions if we don't have multiple devices hanging | |
1750 | * off our complete list, and we're under the requested amount. | |
1751 | */ | |
1752 | spin = !ctx->poll_multi_file && *nr_events < min; | |
1753 | ||
1754 | ret = 0; | |
1755 | list_for_each_entry_safe(req, tmp, &ctx->poll_list, list) { | |
9adbd45d | 1756 | struct kiocb *kiocb = &req->rw.kiocb; |
def596e9 JA |
1757 | |
1758 | /* | |
581f9810 BM |
1759 | * Move completed and retryable entries to our local lists. |
1760 | * If we find a request that requires polling, break out | |
1761 | * and complete those lists first, if we have entries there. | |
def596e9 JA |
1762 | */ |
1763 | if (req->flags & REQ_F_IOPOLL_COMPLETED) { | |
1764 | list_move_tail(&req->list, &done); | |
1765 | continue; | |
1766 | } | |
1767 | if (!list_empty(&done)) | |
1768 | break; | |
1769 | ||
581f9810 BM |
1770 | if (req->result == -EAGAIN) { |
1771 | list_move_tail(&req->list, &again); | |
1772 | continue; | |
1773 | } | |
1774 | if (!list_empty(&again)) | |
1775 | break; | |
1776 | ||
def596e9 JA |
1777 | ret = kiocb->ki_filp->f_op->iopoll(kiocb, spin); |
1778 | if (ret < 0) | |
1779 | break; | |
1780 | ||
1781 | if (ret && spin) | |
1782 | spin = false; | |
1783 | ret = 0; | |
1784 | } | |
1785 | ||
1786 | if (!list_empty(&done)) | |
1787 | io_iopoll_complete(ctx, nr_events, &done); | |
1788 | ||
581f9810 BM |
1789 | if (!list_empty(&again)) |
1790 | io_iopoll_queue(&again); | |
1791 | ||
def596e9 JA |
1792 | return ret; |
1793 | } | |
1794 | ||
1795 | /* | |
d195a66e | 1796 | * Poll for a minimum of 'min' events. Note that if min == 0 we consider that a |
def596e9 JA |
1797 | * non-spinning poll check - we'll still enter the driver poll loop, but only |
1798 | * as a non-spinning completion check. | |
1799 | */ | |
1800 | static int io_iopoll_getevents(struct io_ring_ctx *ctx, unsigned int *nr_events, | |
1801 | long min) | |
1802 | { | |
08f5439f | 1803 | while (!list_empty(&ctx->poll_list) && !need_resched()) { |
def596e9 JA |
1804 | int ret; |
1805 | ||
1806 | ret = io_do_iopoll(ctx, nr_events, min); | |
1807 | if (ret < 0) | |
1808 | return ret; | |
1809 | if (!min || *nr_events >= min) | |
1810 | return 0; | |
1811 | } | |
1812 | ||
1813 | return 1; | |
1814 | } | |
1815 | ||
1816 | /* | |
1817 | * We can't just wait for polled events to come to us, we have to actively | |
1818 | * find and complete them. | |
1819 | */ | |
1820 | static void io_iopoll_reap_events(struct io_ring_ctx *ctx) | |
1821 | { | |
1822 | if (!(ctx->flags & IORING_SETUP_IOPOLL)) | |
1823 | return; | |
1824 | ||
1825 | mutex_lock(&ctx->uring_lock); | |
1826 | while (!list_empty(&ctx->poll_list)) { | |
1827 | unsigned int nr_events = 0; | |
1828 | ||
1829 | io_iopoll_getevents(ctx, &nr_events, 1); | |
08f5439f JA |
1830 | |
1831 | /* | |
1832 | * Ensure we allow local-to-the-cpu processing to take place, | |
1833 | * in this case we need to ensure that we reap all events. | |
1834 | */ | |
1835 | cond_resched(); | |
def596e9 JA |
1836 | } |
1837 | mutex_unlock(&ctx->uring_lock); | |
1838 | } | |
1839 | ||
c7849be9 XW |
1840 | static int io_iopoll_check(struct io_ring_ctx *ctx, unsigned *nr_events, |
1841 | long min) | |
def596e9 | 1842 | { |
2b2ed975 | 1843 | int iters = 0, ret = 0; |
500f9fba | 1844 | |
c7849be9 XW |
1845 | /* |
1846 | * We disallow the app entering submit/complete with polling, but we | |
1847 | * still need to lock the ring to prevent racing with polled issue | |
1848 | * that got punted to a workqueue. | |
1849 | */ | |
1850 | mutex_lock(&ctx->uring_lock); | |
def596e9 JA |
1851 | do { |
1852 | int tmin = 0; | |
1853 | ||
a3a0e43f JA |
1854 | /* |
1855 | * Don't enter poll loop if we already have events pending. | |
1856 | * If we do, we can potentially be spinning for commands that | |
1857 | * already triggered a CQE (eg in error). | |
1858 | */ | |
1d7bb1d5 | 1859 | if (io_cqring_events(ctx, false)) |
a3a0e43f JA |
1860 | break; |
1861 | ||
500f9fba JA |
1862 | /* |
1863 | * If a submit got punted to a workqueue, we can have the | |
1864 | * application entering polling for a command before it gets | |
1865 | * issued. That app will hold the uring_lock for the duration | |
1866 | * of the poll right here, so we need to take a breather every | |
1867 | * now and then to ensure that the issue has a chance to add | |
1868 | * the poll to the issued list. Otherwise we can spin here | |
1869 | * forever, while the workqueue is stuck trying to acquire the | |
1870 | * very same mutex. | |
1871 | */ | |
1872 | if (!(++iters & 7)) { | |
1873 | mutex_unlock(&ctx->uring_lock); | |
1874 | mutex_lock(&ctx->uring_lock); | |
1875 | } | |
1876 | ||
def596e9 JA |
1877 | if (*nr_events < min) |
1878 | tmin = min - *nr_events; | |
1879 | ||
1880 | ret = io_iopoll_getevents(ctx, nr_events, tmin); | |
1881 | if (ret <= 0) | |
1882 | break; | |
1883 | ret = 0; | |
1884 | } while (min && !*nr_events && !need_resched()); | |
1885 | ||
500f9fba | 1886 | mutex_unlock(&ctx->uring_lock); |
def596e9 JA |
1887 | return ret; |
1888 | } | |
1889 | ||
491381ce | 1890 | static void kiocb_end_write(struct io_kiocb *req) |
2b188cc1 | 1891 | { |
491381ce JA |
1892 | /* |
1893 | * Tell lockdep we inherited freeze protection from submission | |
1894 | * thread. | |
1895 | */ | |
1896 | if (req->flags & REQ_F_ISREG) { | |
1897 | struct inode *inode = file_inode(req->file); | |
2b188cc1 | 1898 | |
491381ce | 1899 | __sb_writers_acquired(inode->i_sb, SB_FREEZE_WRITE); |
2b188cc1 | 1900 | } |
491381ce | 1901 | file_end_write(req->file); |
2b188cc1 JA |
1902 | } |
1903 | ||
4e88d6e7 JA |
1904 | static inline void req_set_fail_links(struct io_kiocb *req) |
1905 | { | |
1906 | if ((req->flags & (REQ_F_LINK | REQ_F_HARDLINK)) == REQ_F_LINK) | |
1907 | req->flags |= REQ_F_FAIL_LINK; | |
1908 | } | |
1909 | ||
ba816ad6 | 1910 | static void io_complete_rw_common(struct kiocb *kiocb, long res) |
2b188cc1 | 1911 | { |
9adbd45d | 1912 | struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb); |
bcda7baa | 1913 | int cflags = 0; |
2b188cc1 | 1914 | |
491381ce JA |
1915 | if (kiocb->ki_flags & IOCB_WRITE) |
1916 | kiocb_end_write(req); | |
2b188cc1 | 1917 | |
4e88d6e7 JA |
1918 | if (res != req->result) |
1919 | req_set_fail_links(req); | |
bcda7baa JA |
1920 | if (req->flags & REQ_F_BUFFER_SELECTED) |
1921 | cflags = io_put_kbuf(req); | |
1922 | __io_cqring_add_event(req, res, cflags); | |
ba816ad6 JA |
1923 | } |
1924 | ||
1925 | static void io_complete_rw(struct kiocb *kiocb, long res, long res2) | |
1926 | { | |
9adbd45d | 1927 | struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb); |
ba816ad6 JA |
1928 | |
1929 | io_complete_rw_common(kiocb, res); | |
e65ef56d | 1930 | io_put_req(req); |
2b188cc1 JA |
1931 | } |
1932 | ||
def596e9 JA |
1933 | static void io_complete_rw_iopoll(struct kiocb *kiocb, long res, long res2) |
1934 | { | |
9adbd45d | 1935 | struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb); |
def596e9 | 1936 | |
491381ce JA |
1937 | if (kiocb->ki_flags & IOCB_WRITE) |
1938 | kiocb_end_write(req); | |
def596e9 | 1939 | |
4e88d6e7 JA |
1940 | if (res != req->result) |
1941 | req_set_fail_links(req); | |
9e645e11 | 1942 | req->result = res; |
def596e9 JA |
1943 | if (res != -EAGAIN) |
1944 | req->flags |= REQ_F_IOPOLL_COMPLETED; | |
1945 | } | |
1946 | ||
1947 | /* | |
1948 | * After the iocb has been issued, it's safe to be found on the poll list. | |
1949 | * Adding the kiocb to the list AFTER submission ensures that we don't | |
1950 | * find it from a io_iopoll_getevents() thread before the issuer is done | |
1951 | * accessing the kiocb cookie. | |
1952 | */ | |
1953 | static void io_iopoll_req_issued(struct io_kiocb *req) | |
1954 | { | |
1955 | struct io_ring_ctx *ctx = req->ctx; | |
1956 | ||
1957 | /* | |
1958 | * Track whether we have multiple files in our lists. This will impact | |
1959 | * how we do polling eventually, not spinning if we're on potentially | |
1960 | * different devices. | |
1961 | */ | |
1962 | if (list_empty(&ctx->poll_list)) { | |
1963 | ctx->poll_multi_file = false; | |
1964 | } else if (!ctx->poll_multi_file) { | |
1965 | struct io_kiocb *list_req; | |
1966 | ||
1967 | list_req = list_first_entry(&ctx->poll_list, struct io_kiocb, | |
1968 | list); | |
9adbd45d | 1969 | if (list_req->file != req->file) |
def596e9 JA |
1970 | ctx->poll_multi_file = true; |
1971 | } | |
1972 | ||
1973 | /* | |
1974 | * For fast devices, IO may have already completed. If it has, add | |
1975 | * it to the front so we find it first. | |
1976 | */ | |
1977 | if (req->flags & REQ_F_IOPOLL_COMPLETED) | |
1978 | list_add(&req->list, &ctx->poll_list); | |
1979 | else | |
1980 | list_add_tail(&req->list, &ctx->poll_list); | |
bdcd3eab XW |
1981 | |
1982 | if ((ctx->flags & IORING_SETUP_SQPOLL) && | |
1983 | wq_has_sleeper(&ctx->sqo_wait)) | |
1984 | wake_up(&ctx->sqo_wait); | |
def596e9 JA |
1985 | } |
1986 | ||
3d6770fb | 1987 | static void io_file_put(struct io_submit_state *state) |
9a56a232 | 1988 | { |
3d6770fb | 1989 | if (state->file) { |
9a56a232 JA |
1990 | int diff = state->has_refs - state->used_refs; |
1991 | ||
1992 | if (diff) | |
1993 | fput_many(state->file, diff); | |
1994 | state->file = NULL; | |
1995 | } | |
1996 | } | |
1997 | ||
1998 | /* | |
1999 | * Get as many references to a file as we have IOs left in this submission, | |
2000 | * assuming most submissions are for one file, or at least that each file | |
2001 | * has more than one submission. | |
2002 | */ | |
8da11c19 | 2003 | static struct file *__io_file_get(struct io_submit_state *state, int fd) |
9a56a232 JA |
2004 | { |
2005 | if (!state) | |
2006 | return fget(fd); | |
2007 | ||
2008 | if (state->file) { | |
2009 | if (state->fd == fd) { | |
2010 | state->used_refs++; | |
2011 | state->ios_left--; | |
2012 | return state->file; | |
2013 | } | |
3d6770fb | 2014 | io_file_put(state); |
9a56a232 JA |
2015 | } |
2016 | state->file = fget_many(fd, state->ios_left); | |
2017 | if (!state->file) | |
2018 | return NULL; | |
2019 | ||
2020 | state->fd = fd; | |
2021 | state->has_refs = state->ios_left; | |
2022 | state->used_refs = 1; | |
2023 | state->ios_left--; | |
2024 | return state->file; | |
2025 | } | |
2026 | ||
2b188cc1 JA |
2027 | /* |
2028 | * If we tracked the file through the SCM inflight mechanism, we could support | |
2029 | * any file. For now, just ensure that anything potentially problematic is done | |
2030 | * inline. | |
2031 | */ | |
af197f50 | 2032 | static bool io_file_supports_async(struct file *file, int rw) |
2b188cc1 JA |
2033 | { |
2034 | umode_t mode = file_inode(file)->i_mode; | |
2035 | ||
10d59345 | 2036 | if (S_ISBLK(mode) || S_ISCHR(mode) || S_ISSOCK(mode)) |
2b188cc1 JA |
2037 | return true; |
2038 | if (S_ISREG(mode) && file->f_op != &io_uring_fops) | |
2039 | return true; | |
2040 | ||
af197f50 JA |
2041 | if (!(file->f_mode & FMODE_NOWAIT)) |
2042 | return false; | |
2043 | ||
2044 | if (rw == READ) | |
2045 | return file->f_op->read_iter != NULL; | |
2046 | ||
2047 | return file->f_op->write_iter != NULL; | |
2b188cc1 JA |
2048 | } |
2049 | ||
3529d8c2 JA |
2050 | static int io_prep_rw(struct io_kiocb *req, const struct io_uring_sqe *sqe, |
2051 | bool force_nonblock) | |
2b188cc1 | 2052 | { |
def596e9 | 2053 | struct io_ring_ctx *ctx = req->ctx; |
9adbd45d | 2054 | struct kiocb *kiocb = &req->rw.kiocb; |
09bb8394 JA |
2055 | unsigned ioprio; |
2056 | int ret; | |
2b188cc1 | 2057 | |
491381ce JA |
2058 | if (S_ISREG(file_inode(req->file)->i_mode)) |
2059 | req->flags |= REQ_F_ISREG; | |
2060 | ||
2b188cc1 | 2061 | kiocb->ki_pos = READ_ONCE(sqe->off); |
ba04291e JA |
2062 | if (kiocb->ki_pos == -1 && !(req->file->f_mode & FMODE_STREAM)) { |
2063 | req->flags |= REQ_F_CUR_POS; | |
2064 | kiocb->ki_pos = req->file->f_pos; | |
2065 | } | |
2b188cc1 | 2066 | kiocb->ki_hint = ki_hint_validate(file_write_hint(kiocb->ki_filp)); |
3e577dcd PB |
2067 | kiocb->ki_flags = iocb_flags(kiocb->ki_filp); |
2068 | ret = kiocb_set_rw_flags(kiocb, READ_ONCE(sqe->rw_flags)); | |
2069 | if (unlikely(ret)) | |
2070 | return ret; | |
2b188cc1 JA |
2071 | |
2072 | ioprio = READ_ONCE(sqe->ioprio); | |
2073 | if (ioprio) { | |
2074 | ret = ioprio_check_cap(ioprio); | |
2075 | if (ret) | |
09bb8394 | 2076 | return ret; |
2b188cc1 JA |
2077 | |
2078 | kiocb->ki_ioprio = ioprio; | |
2079 | } else | |
2080 | kiocb->ki_ioprio = get_current_ioprio(); | |
2081 | ||
8449eeda | 2082 | /* don't allow async punt if RWF_NOWAIT was requested */ |
491381ce JA |
2083 | if ((kiocb->ki_flags & IOCB_NOWAIT) || |
2084 | (req->file->f_flags & O_NONBLOCK)) | |
8449eeda SB |
2085 | req->flags |= REQ_F_NOWAIT; |
2086 | ||
2087 | if (force_nonblock) | |
2b188cc1 | 2088 | kiocb->ki_flags |= IOCB_NOWAIT; |
8449eeda | 2089 | |
def596e9 | 2090 | if (ctx->flags & IORING_SETUP_IOPOLL) { |
def596e9 JA |
2091 | if (!(kiocb->ki_flags & IOCB_DIRECT) || |
2092 | !kiocb->ki_filp->f_op->iopoll) | |
09bb8394 | 2093 | return -EOPNOTSUPP; |
2b188cc1 | 2094 | |
def596e9 JA |
2095 | kiocb->ki_flags |= IOCB_HIPRI; |
2096 | kiocb->ki_complete = io_complete_rw_iopoll; | |
6873e0bd | 2097 | req->result = 0; |
def596e9 | 2098 | } else { |
09bb8394 JA |
2099 | if (kiocb->ki_flags & IOCB_HIPRI) |
2100 | return -EINVAL; | |
def596e9 JA |
2101 | kiocb->ki_complete = io_complete_rw; |
2102 | } | |
9adbd45d | 2103 | |
3529d8c2 JA |
2104 | req->rw.addr = READ_ONCE(sqe->addr); |
2105 | req->rw.len = READ_ONCE(sqe->len); | |
4f4eeba8 | 2106 | req->buf_index = READ_ONCE(sqe->buf_index); |
2b188cc1 | 2107 | return 0; |
2b188cc1 JA |
2108 | } |
2109 | ||
2110 | static inline void io_rw_done(struct kiocb *kiocb, ssize_t ret) | |
2111 | { | |
2112 | switch (ret) { | |
2113 | case -EIOCBQUEUED: | |
2114 | break; | |
2115 | case -ERESTARTSYS: | |
2116 | case -ERESTARTNOINTR: | |
2117 | case -ERESTARTNOHAND: | |
2118 | case -ERESTART_RESTARTBLOCK: | |
2119 | /* | |
2120 | * We can't just restart the syscall, since previously | |
2121 | * submitted sqes may already be in progress. Just fail this | |
2122 | * IO with EINTR. | |
2123 | */ | |
2124 | ret = -EINTR; | |
2125 | /* fall through */ | |
2126 | default: | |
2127 | kiocb->ki_complete(kiocb, ret, 0); | |
2128 | } | |
2129 | } | |
2130 | ||
014db007 | 2131 | static void kiocb_done(struct kiocb *kiocb, ssize_t ret) |
ba816ad6 | 2132 | { |
ba04291e JA |
2133 | struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb); |
2134 | ||
2135 | if (req->flags & REQ_F_CUR_POS) | |
2136 | req->file->f_pos = kiocb->ki_pos; | |
bcaec089 | 2137 | if (ret >= 0 && kiocb->ki_complete == io_complete_rw) |
014db007 | 2138 | io_complete_rw(kiocb, ret, 0); |
ba816ad6 JA |
2139 | else |
2140 | io_rw_done(kiocb, ret); | |
2141 | } | |
2142 | ||
9adbd45d | 2143 | static ssize_t io_import_fixed(struct io_kiocb *req, int rw, |
7d009165 | 2144 | struct iov_iter *iter) |
edafccee | 2145 | { |
9adbd45d JA |
2146 | struct io_ring_ctx *ctx = req->ctx; |
2147 | size_t len = req->rw.len; | |
edafccee | 2148 | struct io_mapped_ubuf *imu; |
4f4eeba8 | 2149 | u16 index, buf_index; |
edafccee JA |
2150 | size_t offset; |
2151 | u64 buf_addr; | |
2152 | ||
2153 | /* attempt to use fixed buffers without having provided iovecs */ | |
2154 | if (unlikely(!ctx->user_bufs)) | |
2155 | return -EFAULT; | |
2156 | ||
4f4eeba8 | 2157 | buf_index = req->buf_index; |
edafccee JA |
2158 | if (unlikely(buf_index >= ctx->nr_user_bufs)) |
2159 | return -EFAULT; | |
2160 | ||
2161 | index = array_index_nospec(buf_index, ctx->nr_user_bufs); | |
2162 | imu = &ctx->user_bufs[index]; | |
9adbd45d | 2163 | buf_addr = req->rw.addr; |
edafccee JA |
2164 | |
2165 | /* overflow */ | |
2166 | if (buf_addr + len < buf_addr) | |
2167 | return -EFAULT; | |
2168 | /* not inside the mapped region */ | |
2169 | if (buf_addr < imu->ubuf || buf_addr + len > imu->ubuf + imu->len) | |
2170 | return -EFAULT; | |
2171 | ||
2172 | /* | |
2173 | * May not be a start of buffer, set size appropriately | |
2174 | * and advance us to the beginning. | |
2175 | */ | |
2176 | offset = buf_addr - imu->ubuf; | |
2177 | iov_iter_bvec(iter, rw, imu->bvec, imu->nr_bvecs, offset + len); | |
bd11b3a3 JA |
2178 | |
2179 | if (offset) { | |
2180 | /* | |
2181 | * Don't use iov_iter_advance() here, as it's really slow for | |
2182 | * using the latter parts of a big fixed buffer - it iterates | |
2183 | * over each segment manually. We can cheat a bit here, because | |
2184 | * we know that: | |
2185 | * | |
2186 | * 1) it's a BVEC iter, we set it up | |
2187 | * 2) all bvecs are PAGE_SIZE in size, except potentially the | |
2188 | * first and last bvec | |
2189 | * | |
2190 | * So just find our index, and adjust the iterator afterwards. | |
2191 | * If the offset is within the first bvec (or the whole first | |
2192 | * bvec, just use iov_iter_advance(). This makes it easier | |
2193 | * since we can just skip the first segment, which may not | |
2194 | * be PAGE_SIZE aligned. | |
2195 | */ | |
2196 | const struct bio_vec *bvec = imu->bvec; | |
2197 | ||
2198 | if (offset <= bvec->bv_len) { | |
2199 | iov_iter_advance(iter, offset); | |
2200 | } else { | |
2201 | unsigned long seg_skip; | |
2202 | ||
2203 | /* skip first vec */ | |
2204 | offset -= bvec->bv_len; | |
2205 | seg_skip = 1 + (offset >> PAGE_SHIFT); | |
2206 | ||
2207 | iter->bvec = bvec + seg_skip; | |
2208 | iter->nr_segs -= seg_skip; | |
99c79f66 | 2209 | iter->count -= bvec->bv_len + offset; |
bd11b3a3 | 2210 | iter->iov_offset = offset & ~PAGE_MASK; |
bd11b3a3 JA |
2211 | } |
2212 | } | |
2213 | ||
5e559561 | 2214 | return len; |
edafccee JA |
2215 | } |
2216 | ||
bcda7baa JA |
2217 | static void io_ring_submit_unlock(struct io_ring_ctx *ctx, bool needs_lock) |
2218 | { | |
2219 | if (needs_lock) | |
2220 | mutex_unlock(&ctx->uring_lock); | |
2221 | } | |
2222 | ||
2223 | static void io_ring_submit_lock(struct io_ring_ctx *ctx, bool needs_lock) | |
2224 | { | |
2225 | /* | |
2226 | * "Normal" inline submissions always hold the uring_lock, since we | |
2227 | * grab it from the system call. Same is true for the SQPOLL offload. | |
2228 | * The only exception is when we've detached the request and issue it | |
2229 | * from an async worker thread, grab the lock for that case. | |
2230 | */ | |
2231 | if (needs_lock) | |
2232 | mutex_lock(&ctx->uring_lock); | |
2233 | } | |
2234 | ||
2235 | static struct io_buffer *io_buffer_select(struct io_kiocb *req, size_t *len, | |
2236 | int bgid, struct io_buffer *kbuf, | |
2237 | bool needs_lock) | |
2238 | { | |
2239 | struct io_buffer *head; | |
2240 | ||
2241 | if (req->flags & REQ_F_BUFFER_SELECTED) | |
2242 | return kbuf; | |
2243 | ||
2244 | io_ring_submit_lock(req->ctx, needs_lock); | |
2245 | ||
2246 | lockdep_assert_held(&req->ctx->uring_lock); | |
2247 | ||
2248 | head = idr_find(&req->ctx->io_buffer_idr, bgid); | |
2249 | if (head) { | |
2250 | if (!list_empty(&head->list)) { | |
2251 | kbuf = list_last_entry(&head->list, struct io_buffer, | |
2252 | list); | |
2253 | list_del(&kbuf->list); | |
2254 | } else { | |
2255 | kbuf = head; | |
2256 | idr_remove(&req->ctx->io_buffer_idr, bgid); | |
2257 | } | |
2258 | if (*len > kbuf->len) | |
2259 | *len = kbuf->len; | |
2260 | } else { | |
2261 | kbuf = ERR_PTR(-ENOBUFS); | |
2262 | } | |
2263 | ||
2264 | io_ring_submit_unlock(req->ctx, needs_lock); | |
2265 | ||
2266 | return kbuf; | |
2267 | } | |
2268 | ||
4d954c25 JA |
2269 | static void __user *io_rw_buffer_select(struct io_kiocb *req, size_t *len, |
2270 | bool needs_lock) | |
2271 | { | |
2272 | struct io_buffer *kbuf; | |
4f4eeba8 | 2273 | u16 bgid; |
4d954c25 JA |
2274 | |
2275 | kbuf = (struct io_buffer *) (unsigned long) req->rw.addr; | |
4f4eeba8 | 2276 | bgid = req->buf_index; |
4d954c25 JA |
2277 | kbuf = io_buffer_select(req, len, bgid, kbuf, needs_lock); |
2278 | if (IS_ERR(kbuf)) | |
2279 | return kbuf; | |
2280 | req->rw.addr = (u64) (unsigned long) kbuf; | |
2281 | req->flags |= REQ_F_BUFFER_SELECTED; | |
2282 | return u64_to_user_ptr(kbuf->addr); | |
2283 | } | |
2284 | ||
2285 | #ifdef CONFIG_COMPAT | |
2286 | static ssize_t io_compat_import(struct io_kiocb *req, struct iovec *iov, | |
2287 | bool needs_lock) | |
2288 | { | |
2289 | struct compat_iovec __user *uiov; | |
2290 | compat_ssize_t clen; | |
2291 | void __user *buf; | |
2292 | ssize_t len; | |
2293 | ||
2294 | uiov = u64_to_user_ptr(req->rw.addr); | |
2295 | if (!access_ok(uiov, sizeof(*uiov))) | |
2296 | return -EFAULT; | |
2297 | if (__get_user(clen, &uiov->iov_len)) | |
2298 | return -EFAULT; | |
2299 | if (clen < 0) | |
2300 | return -EINVAL; | |
2301 | ||
2302 | len = clen; | |
2303 | buf = io_rw_buffer_select(req, &len, needs_lock); | |
2304 | if (IS_ERR(buf)) | |
2305 | return PTR_ERR(buf); | |
2306 | iov[0].iov_base = buf; | |
2307 | iov[0].iov_len = (compat_size_t) len; | |
2308 | return 0; | |
2309 | } | |
2310 | #endif | |
2311 | ||
2312 | static ssize_t __io_iov_buffer_select(struct io_kiocb *req, struct iovec *iov, | |
2313 | bool needs_lock) | |
2314 | { | |
2315 | struct iovec __user *uiov = u64_to_user_ptr(req->rw.addr); | |
2316 | void __user *buf; | |
2317 | ssize_t len; | |
2318 | ||
2319 | if (copy_from_user(iov, uiov, sizeof(*uiov))) | |
2320 | return -EFAULT; | |
2321 | ||
2322 | len = iov[0].iov_len; | |
2323 | if (len < 0) | |
2324 | return -EINVAL; | |
2325 | buf = io_rw_buffer_select(req, &len, needs_lock); | |
2326 | if (IS_ERR(buf)) | |
2327 | return PTR_ERR(buf); | |
2328 | iov[0].iov_base = buf; | |
2329 | iov[0].iov_len = len; | |
2330 | return 0; | |
2331 | } | |
2332 | ||
2333 | static ssize_t io_iov_buffer_select(struct io_kiocb *req, struct iovec *iov, | |
2334 | bool needs_lock) | |
2335 | { | |
2336 | if (req->flags & REQ_F_BUFFER_SELECTED) | |
2337 | return 0; | |
2338 | if (!req->rw.len) | |
2339 | return 0; | |
2340 | else if (req->rw.len > 1) | |
2341 | return -EINVAL; | |
2342 | ||
2343 | #ifdef CONFIG_COMPAT | |
2344 | if (req->ctx->compat) | |
2345 | return io_compat_import(req, iov, needs_lock); | |
2346 | #endif | |
2347 | ||
2348 | return __io_iov_buffer_select(req, iov, needs_lock); | |
2349 | } | |
2350 | ||
cf6fd4bd | 2351 | static ssize_t io_import_iovec(int rw, struct io_kiocb *req, |
bcda7baa JA |
2352 | struct iovec **iovec, struct iov_iter *iter, |
2353 | bool needs_lock) | |
2b188cc1 | 2354 | { |
9adbd45d JA |
2355 | void __user *buf = u64_to_user_ptr(req->rw.addr); |
2356 | size_t sqe_len = req->rw.len; | |
4d954c25 | 2357 | ssize_t ret; |
edafccee JA |
2358 | u8 opcode; |
2359 | ||
d625c6ee | 2360 | opcode = req->opcode; |
7d009165 | 2361 | if (opcode == IORING_OP_READ_FIXED || opcode == IORING_OP_WRITE_FIXED) { |
edafccee | 2362 | *iovec = NULL; |
9adbd45d | 2363 | return io_import_fixed(req, rw, iter); |
edafccee | 2364 | } |
2b188cc1 | 2365 | |
bcda7baa | 2366 | /* buffer index only valid with fixed read/write, or buffer select */ |
4f4eeba8 | 2367 | if (req->buf_index && !(req->flags & REQ_F_BUFFER_SELECT)) |
9adbd45d JA |
2368 | return -EINVAL; |
2369 | ||
3a6820f2 | 2370 | if (opcode == IORING_OP_READ || opcode == IORING_OP_WRITE) { |
bcda7baa | 2371 | if (req->flags & REQ_F_BUFFER_SELECT) { |
4d954c25 JA |
2372 | buf = io_rw_buffer_select(req, &sqe_len, needs_lock); |
2373 | if (IS_ERR(buf)) { | |
bcda7baa | 2374 | *iovec = NULL; |
4d954c25 | 2375 | return PTR_ERR(buf); |
bcda7baa | 2376 | } |
3f9d6441 | 2377 | req->rw.len = sqe_len; |
bcda7baa JA |
2378 | } |
2379 | ||
3a6820f2 JA |
2380 | ret = import_single_range(rw, buf, sqe_len, *iovec, iter); |
2381 | *iovec = NULL; | |
3a901598 | 2382 | return ret < 0 ? ret : sqe_len; |
3a6820f2 JA |
2383 | } |
2384 | ||
f67676d1 JA |
2385 | if (req->io) { |
2386 | struct io_async_rw *iorw = &req->io->rw; | |
2387 | ||
2388 | *iovec = iorw->iov; | |
2389 | iov_iter_init(iter, rw, *iovec, iorw->nr_segs, iorw->size); | |
2390 | if (iorw->iov == iorw->fast_iov) | |
2391 | *iovec = NULL; | |
2392 | return iorw->size; | |
2393 | } | |
2394 | ||
4d954c25 JA |
2395 | if (req->flags & REQ_F_BUFFER_SELECT) { |
2396 | ret = io_iov_buffer_select(req, *iovec, needs_lock); | |
3f9d6441 JA |
2397 | if (!ret) { |
2398 | ret = (*iovec)->iov_len; | |
2399 | iov_iter_init(iter, rw, *iovec, 1, ret); | |
2400 | } | |
4d954c25 JA |
2401 | *iovec = NULL; |
2402 | return ret; | |
2403 | } | |
2404 | ||
2b188cc1 | 2405 | #ifdef CONFIG_COMPAT |
cf6fd4bd | 2406 | if (req->ctx->compat) |
2b188cc1 JA |
2407 | return compat_import_iovec(rw, buf, sqe_len, UIO_FASTIOV, |
2408 | iovec, iter); | |
2409 | #endif | |
2410 | ||
2411 | return import_iovec(rw, buf, sqe_len, UIO_FASTIOV, iovec, iter); | |
2412 | } | |
2413 | ||
31b51510 | 2414 | /* |
32960613 JA |
2415 | * For files that don't have ->read_iter() and ->write_iter(), handle them |
2416 | * by looping over ->read() or ->write() manually. | |
31b51510 | 2417 | */ |
32960613 JA |
2418 | static ssize_t loop_rw_iter(int rw, struct file *file, struct kiocb *kiocb, |
2419 | struct iov_iter *iter) | |
2420 | { | |
2421 | ssize_t ret = 0; | |
2422 | ||
2423 | /* | |
2424 | * Don't support polled IO through this interface, and we can't | |
2425 | * support non-blocking either. For the latter, this just causes | |
2426 | * the kiocb to be handled from an async context. | |
2427 | */ | |
2428 | if (kiocb->ki_flags & IOCB_HIPRI) | |
2429 | return -EOPNOTSUPP; | |
2430 | if (kiocb->ki_flags & IOCB_NOWAIT) | |
2431 | return -EAGAIN; | |
2432 | ||
2433 | while (iov_iter_count(iter)) { | |
311ae9e1 | 2434 | struct iovec iovec; |
32960613 JA |
2435 | ssize_t nr; |
2436 | ||
311ae9e1 PB |
2437 | if (!iov_iter_is_bvec(iter)) { |
2438 | iovec = iov_iter_iovec(iter); | |
2439 | } else { | |
2440 | /* fixed buffers import bvec */ | |
2441 | iovec.iov_base = kmap(iter->bvec->bv_page) | |
2442 | + iter->iov_offset; | |
2443 | iovec.iov_len = min(iter->count, | |
2444 | iter->bvec->bv_len - iter->iov_offset); | |
2445 | } | |
2446 | ||
32960613 JA |
2447 | if (rw == READ) { |
2448 | nr = file->f_op->read(file, iovec.iov_base, | |
2449 | iovec.iov_len, &kiocb->ki_pos); | |
2450 | } else { | |
2451 | nr = file->f_op->write(file, iovec.iov_base, | |
2452 | iovec.iov_len, &kiocb->ki_pos); | |
2453 | } | |
2454 | ||
311ae9e1 PB |
2455 | if (iov_iter_is_bvec(iter)) |
2456 | kunmap(iter->bvec->bv_page); | |
2457 | ||
32960613 JA |
2458 | if (nr < 0) { |
2459 | if (!ret) | |
2460 | ret = nr; | |
2461 | break; | |
2462 | } | |
2463 | ret += nr; | |
2464 | if (nr != iovec.iov_len) | |
2465 | break; | |
2466 | iov_iter_advance(iter, nr); | |
2467 | } | |
2468 | ||
2469 | return ret; | |
2470 | } | |
2471 | ||
b7bb4f7d | 2472 | static void io_req_map_rw(struct io_kiocb *req, ssize_t io_size, |
f67676d1 JA |
2473 | struct iovec *iovec, struct iovec *fast_iov, |
2474 | struct iov_iter *iter) | |
2475 | { | |
2476 | req->io->rw.nr_segs = iter->nr_segs; | |
2477 | req->io->rw.size = io_size; | |
2478 | req->io->rw.iov = iovec; | |
2479 | if (!req->io->rw.iov) { | |
2480 | req->io->rw.iov = req->io->rw.fast_iov; | |
45097dae XW |
2481 | if (req->io->rw.iov != fast_iov) |
2482 | memcpy(req->io->rw.iov, fast_iov, | |
2483 | sizeof(struct iovec) * iter->nr_segs); | |
99bc4c38 PB |
2484 | } else { |
2485 | req->flags |= REQ_F_NEED_CLEANUP; | |
f67676d1 JA |
2486 | } |
2487 | } | |
2488 | ||
3d9932a8 XW |
2489 | static inline int __io_alloc_async_ctx(struct io_kiocb *req) |
2490 | { | |
2491 | req->io = kmalloc(sizeof(*req->io), GFP_KERNEL); | |
2492 | return req->io == NULL; | |
2493 | } | |
2494 | ||
b7bb4f7d | 2495 | static int io_alloc_async_ctx(struct io_kiocb *req) |
f67676d1 | 2496 | { |
d3656344 JA |
2497 | if (!io_op_defs[req->opcode].async_ctx) |
2498 | return 0; | |
3d9932a8 XW |
2499 | |
2500 | return __io_alloc_async_ctx(req); | |
b7bb4f7d JA |
2501 | } |
2502 | ||
b7bb4f7d JA |
2503 | static int io_setup_async_rw(struct io_kiocb *req, ssize_t io_size, |
2504 | struct iovec *iovec, struct iovec *fast_iov, | |
2505 | struct iov_iter *iter) | |
2506 | { | |
980ad263 | 2507 | if (!io_op_defs[req->opcode].async_ctx) |
74566df3 | 2508 | return 0; |
5d204bcf | 2509 | if (!req->io) { |
3d9932a8 | 2510 | if (__io_alloc_async_ctx(req)) |
5d204bcf | 2511 | return -ENOMEM; |
b7bb4f7d | 2512 | |
5d204bcf JA |
2513 | io_req_map_rw(req, io_size, iovec, fast_iov, iter); |
2514 | } | |
b7bb4f7d | 2515 | return 0; |
f67676d1 JA |
2516 | } |
2517 | ||
3529d8c2 JA |
2518 | static int io_read_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe, |
2519 | bool force_nonblock) | |
f67676d1 | 2520 | { |
3529d8c2 JA |
2521 | struct io_async_ctx *io; |
2522 | struct iov_iter iter; | |
f67676d1 JA |
2523 | ssize_t ret; |
2524 | ||
3529d8c2 JA |
2525 | ret = io_prep_rw(req, sqe, force_nonblock); |
2526 | if (ret) | |
2527 | return ret; | |
f67676d1 | 2528 | |
3529d8c2 JA |
2529 | if (unlikely(!(req->file->f_mode & FMODE_READ))) |
2530 | return -EBADF; | |
f67676d1 | 2531 | |
5f798bea PB |
2532 | /* either don't need iovec imported or already have it */ |
2533 | if (!req->io || req->flags & REQ_F_NEED_CLEANUP) | |
3529d8c2 JA |
2534 | return 0; |
2535 | ||
2536 | io = req->io; | |
2537 | io->rw.iov = io->rw.fast_iov; | |
2538 | req->io = NULL; | |
bcda7baa | 2539 | ret = io_import_iovec(READ, req, &io->rw.iov, &iter, !force_nonblock); |
3529d8c2 JA |
2540 | req->io = io; |
2541 | if (ret < 0) | |
2542 | return ret; | |
2543 | ||
2544 | io_req_map_rw(req, ret, io->rw.iov, io->rw.fast_iov, &iter); | |
2545 | return 0; | |
f67676d1 JA |
2546 | } |
2547 | ||
014db007 | 2548 | static int io_read(struct io_kiocb *req, bool force_nonblock) |
2b188cc1 JA |
2549 | { |
2550 | struct iovec inline_vecs[UIO_FASTIOV], *iovec = inline_vecs; | |
9adbd45d | 2551 | struct kiocb *kiocb = &req->rw.kiocb; |
2b188cc1 | 2552 | struct iov_iter iter; |
31b51510 | 2553 | size_t iov_count; |
f67676d1 | 2554 | ssize_t io_size, ret; |
2b188cc1 | 2555 | |
bcda7baa | 2556 | ret = io_import_iovec(READ, req, &iovec, &iter, !force_nonblock); |
06b76d44 JA |
2557 | if (ret < 0) |
2558 | return ret; | |
2b188cc1 | 2559 | |
fd6c2e4c JA |
2560 | /* Ensure we clear previously set non-block flag */ |
2561 | if (!force_nonblock) | |
29de5f6a | 2562 | kiocb->ki_flags &= ~IOCB_NOWAIT; |
fd6c2e4c | 2563 | |
797f3f53 | 2564 | req->result = 0; |
f67676d1 | 2565 | io_size = ret; |
dea3b49c | 2566 | if (req->flags & REQ_F_LINK_HEAD) |
f67676d1 JA |
2567 | req->result = io_size; |
2568 | ||
2569 | /* | |
2570 | * If the file doesn't support async, mark it as REQ_F_MUST_PUNT so | |
2571 | * we know to async punt it even if it was opened O_NONBLOCK | |
2572 | */ | |
af197f50 | 2573 | if (force_nonblock && !io_file_supports_async(req->file, READ)) |
f67676d1 | 2574 | goto copy_iov; |
9e645e11 | 2575 | |
31b51510 | 2576 | iov_count = iov_iter_count(&iter); |
9adbd45d | 2577 | ret = rw_verify_area(READ, req->file, &kiocb->ki_pos, iov_count); |
2b188cc1 JA |
2578 | if (!ret) { |
2579 | ssize_t ret2; | |
2580 | ||
9adbd45d JA |
2581 | if (req->file->f_op->read_iter) |
2582 | ret2 = call_read_iter(req->file, kiocb, &iter); | |
32960613 | 2583 | else |
9adbd45d | 2584 | ret2 = loop_rw_iter(READ, req->file, kiocb, &iter); |
32960613 | 2585 | |
9d93a3f5 | 2586 | /* Catch -EAGAIN return for forced non-blocking submission */ |
f67676d1 | 2587 | if (!force_nonblock || ret2 != -EAGAIN) { |
014db007 | 2588 | kiocb_done(kiocb, ret2); |
f67676d1 JA |
2589 | } else { |
2590 | copy_iov: | |
b7bb4f7d | 2591 | ret = io_setup_async_rw(req, io_size, iovec, |
f67676d1 JA |
2592 | inline_vecs, &iter); |
2593 | if (ret) | |
2594 | goto out_free; | |
29de5f6a | 2595 | /* any defer here is final, must blocking retry */ |
490e8967 JA |
2596 | if (!(req->flags & REQ_F_NOWAIT) && |
2597 | !file_can_poll(req->file)) | |
29de5f6a | 2598 | req->flags |= REQ_F_MUST_PUNT; |
f67676d1 JA |
2599 | return -EAGAIN; |
2600 | } | |
2b188cc1 | 2601 | } |
f67676d1 | 2602 | out_free: |
1e95081c | 2603 | kfree(iovec); |
99bc4c38 | 2604 | req->flags &= ~REQ_F_NEED_CLEANUP; |
2b188cc1 JA |
2605 | return ret; |
2606 | } | |
2607 | ||
3529d8c2 JA |
2608 | static int io_write_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe, |
2609 | bool force_nonblock) | |
f67676d1 | 2610 | { |
3529d8c2 JA |
2611 | struct io_async_ctx *io; |
2612 | struct iov_iter iter; | |
f67676d1 JA |
2613 | ssize_t ret; |
2614 | ||
3529d8c2 JA |
2615 | ret = io_prep_rw(req, sqe, force_nonblock); |
2616 | if (ret) | |
2617 | return ret; | |
f67676d1 | 2618 | |
3529d8c2 JA |
2619 | if (unlikely(!(req->file->f_mode & FMODE_WRITE))) |
2620 | return -EBADF; | |
f67676d1 | 2621 | |
4ed734b0 JA |
2622 | req->fsize = rlimit(RLIMIT_FSIZE); |
2623 | ||
5f798bea PB |
2624 | /* either don't need iovec imported or already have it */ |
2625 | if (!req->io || req->flags & REQ_F_NEED_CLEANUP) | |
3529d8c2 JA |
2626 | return 0; |
2627 | ||
2628 | io = req->io; | |
2629 | io->rw.iov = io->rw.fast_iov; | |
2630 | req->io = NULL; | |
bcda7baa | 2631 | ret = io_import_iovec(WRITE, req, &io->rw.iov, &iter, !force_nonblock); |
3529d8c2 JA |
2632 | req->io = io; |
2633 | if (ret < 0) | |
2634 | return ret; | |
2635 | ||
2636 | io_req_map_rw(req, ret, io->rw.iov, io->rw.fast_iov, &iter); | |
2637 | return 0; | |
f67676d1 JA |
2638 | } |
2639 | ||
014db007 | 2640 | static int io_write(struct io_kiocb *req, bool force_nonblock) |
2b188cc1 JA |
2641 | { |
2642 | struct iovec inline_vecs[UIO_FASTIOV], *iovec = inline_vecs; | |
9adbd45d | 2643 | struct kiocb *kiocb = &req->rw.kiocb; |
2b188cc1 | 2644 | struct iov_iter iter; |
31b51510 | 2645 | size_t iov_count; |
f67676d1 | 2646 | ssize_t ret, io_size; |
2b188cc1 | 2647 | |
bcda7baa | 2648 | ret = io_import_iovec(WRITE, req, &iovec, &iter, !force_nonblock); |
06b76d44 JA |
2649 | if (ret < 0) |
2650 | return ret; | |
2b188cc1 | 2651 | |
fd6c2e4c JA |
2652 | /* Ensure we clear previously set non-block flag */ |
2653 | if (!force_nonblock) | |
9adbd45d | 2654 | req->rw.kiocb.ki_flags &= ~IOCB_NOWAIT; |
fd6c2e4c | 2655 | |
797f3f53 | 2656 | req->result = 0; |
f67676d1 | 2657 | io_size = ret; |
dea3b49c | 2658 | if (req->flags & REQ_F_LINK_HEAD) |
f67676d1 | 2659 | req->result = io_size; |
9e645e11 | 2660 | |
f67676d1 JA |
2661 | /* |
2662 | * If the file doesn't support async, mark it as REQ_F_MUST_PUNT so | |
2663 | * we know to async punt it even if it was opened O_NONBLOCK | |
2664 | */ | |
af197f50 | 2665 | if (force_nonblock && !io_file_supports_async(req->file, WRITE)) |
f67676d1 | 2666 | goto copy_iov; |
31b51510 | 2667 | |
10d59345 JA |
2668 | /* file path doesn't support NOWAIT for non-direct_IO */ |
2669 | if (force_nonblock && !(kiocb->ki_flags & IOCB_DIRECT) && | |
2670 | (req->flags & REQ_F_ISREG)) | |
f67676d1 | 2671 | goto copy_iov; |
31b51510 | 2672 | |
f67676d1 | 2673 | iov_count = iov_iter_count(&iter); |
9adbd45d | 2674 | ret = rw_verify_area(WRITE, req->file, &kiocb->ki_pos, iov_count); |
2b188cc1 | 2675 | if (!ret) { |
9bf7933f RP |
2676 | ssize_t ret2; |
2677 | ||
2b188cc1 JA |
2678 | /* |
2679 | * Open-code file_start_write here to grab freeze protection, | |
2680 | * which will be released by another thread in | |
2681 | * io_complete_rw(). Fool lockdep by telling it the lock got | |
2682 | * released so that it doesn't complain about the held lock when | |
2683 | * we return to userspace. | |
2684 | */ | |
491381ce | 2685 | if (req->flags & REQ_F_ISREG) { |
9adbd45d | 2686 | __sb_start_write(file_inode(req->file)->i_sb, |
2b188cc1 | 2687 | SB_FREEZE_WRITE, true); |
9adbd45d | 2688 | __sb_writers_release(file_inode(req->file)->i_sb, |
2b188cc1 JA |
2689 | SB_FREEZE_WRITE); |
2690 | } | |
2691 | kiocb->ki_flags |= IOCB_WRITE; | |
9bf7933f | 2692 | |
4ed734b0 JA |
2693 | if (!force_nonblock) |
2694 | current->signal->rlim[RLIMIT_FSIZE].rlim_cur = req->fsize; | |
2695 | ||
9adbd45d JA |
2696 | if (req->file->f_op->write_iter) |
2697 | ret2 = call_write_iter(req->file, kiocb, &iter); | |
32960613 | 2698 | else |
9adbd45d | 2699 | ret2 = loop_rw_iter(WRITE, req->file, kiocb, &iter); |
4ed734b0 JA |
2700 | |
2701 | if (!force_nonblock) | |
2702 | current->signal->rlim[RLIMIT_FSIZE].rlim_cur = RLIM_INFINITY; | |
2703 | ||
faac996c | 2704 | /* |
bff6035d | 2705 | * Raw bdev writes will return -EOPNOTSUPP for IOCB_NOWAIT. Just |
faac996c JA |
2706 | * retry them without IOCB_NOWAIT. |
2707 | */ | |
2708 | if (ret2 == -EOPNOTSUPP && (kiocb->ki_flags & IOCB_NOWAIT)) | |
2709 | ret2 = -EAGAIN; | |
f67676d1 | 2710 | if (!force_nonblock || ret2 != -EAGAIN) { |
014db007 | 2711 | kiocb_done(kiocb, ret2); |
f67676d1 JA |
2712 | } else { |
2713 | copy_iov: | |
b7bb4f7d | 2714 | ret = io_setup_async_rw(req, io_size, iovec, |
f67676d1 JA |
2715 | inline_vecs, &iter); |
2716 | if (ret) | |
2717 | goto out_free; | |
29de5f6a | 2718 | /* any defer here is final, must blocking retry */ |
490e8967 JA |
2719 | if (!file_can_poll(req->file)) |
2720 | req->flags |= REQ_F_MUST_PUNT; | |
f67676d1 JA |
2721 | return -EAGAIN; |
2722 | } | |
2b188cc1 | 2723 | } |
31b51510 | 2724 | out_free: |
99bc4c38 | 2725 | req->flags &= ~REQ_F_NEED_CLEANUP; |
1e95081c | 2726 | kfree(iovec); |
2b188cc1 JA |
2727 | return ret; |
2728 | } | |
2729 | ||
7d67af2c PB |
2730 | static int io_splice_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
2731 | { | |
2732 | struct io_splice* sp = &req->splice; | |
2733 | unsigned int valid_flags = SPLICE_F_FD_IN_FIXED | SPLICE_F_ALL; | |
2734 | int ret; | |
2735 | ||
2736 | if (req->flags & REQ_F_NEED_CLEANUP) | |
2737 | return 0; | |
2738 | ||
2739 | sp->file_in = NULL; | |
2740 | sp->off_in = READ_ONCE(sqe->splice_off_in); | |
2741 | sp->off_out = READ_ONCE(sqe->off); | |
2742 | sp->len = READ_ONCE(sqe->len); | |
2743 | sp->flags = READ_ONCE(sqe->splice_flags); | |
2744 | ||
2745 | if (unlikely(sp->flags & ~valid_flags)) | |
2746 | return -EINVAL; | |
2747 | ||
2748 | ret = io_file_get(NULL, req, READ_ONCE(sqe->splice_fd_in), &sp->file_in, | |
2749 | (sp->flags & SPLICE_F_FD_IN_FIXED)); | |
2750 | if (ret) | |
2751 | return ret; | |
2752 | req->flags |= REQ_F_NEED_CLEANUP; | |
2753 | ||
2754 | if (!S_ISREG(file_inode(sp->file_in)->i_mode)) | |
2755 | req->work.flags |= IO_WQ_WORK_UNBOUND; | |
2756 | ||
2757 | return 0; | |
2758 | } | |
2759 | ||
014db007 | 2760 | static int io_splice(struct io_kiocb *req, bool force_nonblock) |
7d67af2c PB |
2761 | { |
2762 | struct io_splice *sp = &req->splice; | |
2763 | struct file *in = sp->file_in; | |
2764 | struct file *out = sp->file_out; | |
2765 | unsigned int flags = sp->flags & ~SPLICE_F_FD_IN_FIXED; | |
2766 | loff_t *poff_in, *poff_out; | |
c9687426 | 2767 | long ret = 0; |
7d67af2c | 2768 | |
2fb3e822 PB |
2769 | if (force_nonblock) |
2770 | return -EAGAIN; | |
7d67af2c PB |
2771 | |
2772 | poff_in = (sp->off_in == -1) ? NULL : &sp->off_in; | |
2773 | poff_out = (sp->off_out == -1) ? NULL : &sp->off_out; | |
c9687426 | 2774 | |
948a7749 | 2775 | if (sp->len) |
c9687426 | 2776 | ret = do_splice(in, poff_in, out, poff_out, sp->len, flags); |
7d67af2c PB |
2777 | |
2778 | io_put_file(req, in, (sp->flags & SPLICE_F_FD_IN_FIXED)); | |
2779 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
2780 | ||
2781 | io_cqring_add_event(req, ret); | |
2782 | if (ret != sp->len) | |
2783 | req_set_fail_links(req); | |
014db007 | 2784 | io_put_req(req); |
7d67af2c PB |
2785 | return 0; |
2786 | } | |
2787 | ||
2b188cc1 JA |
2788 | /* |
2789 | * IORING_OP_NOP just posts a completion event, nothing else. | |
2790 | */ | |
78e19bbe | 2791 | static int io_nop(struct io_kiocb *req) |
2b188cc1 JA |
2792 | { |
2793 | struct io_ring_ctx *ctx = req->ctx; | |
2b188cc1 | 2794 | |
def596e9 JA |
2795 | if (unlikely(ctx->flags & IORING_SETUP_IOPOLL)) |
2796 | return -EINVAL; | |
2797 | ||
78e19bbe | 2798 | io_cqring_add_event(req, 0); |
e65ef56d | 2799 | io_put_req(req); |
2b188cc1 JA |
2800 | return 0; |
2801 | } | |
2802 | ||
3529d8c2 | 2803 | static int io_prep_fsync(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
c992fe29 | 2804 | { |
6b06314c | 2805 | struct io_ring_ctx *ctx = req->ctx; |
c992fe29 | 2806 | |
09bb8394 JA |
2807 | if (!req->file) |
2808 | return -EBADF; | |
c992fe29 | 2809 | |
6b06314c | 2810 | if (unlikely(ctx->flags & IORING_SETUP_IOPOLL)) |
def596e9 | 2811 | return -EINVAL; |
edafccee | 2812 | if (unlikely(sqe->addr || sqe->ioprio || sqe->buf_index)) |
c992fe29 CH |
2813 | return -EINVAL; |
2814 | ||
8ed8d3c3 JA |
2815 | req->sync.flags = READ_ONCE(sqe->fsync_flags); |
2816 | if (unlikely(req->sync.flags & ~IORING_FSYNC_DATASYNC)) | |
2817 | return -EINVAL; | |
2818 | ||
2819 | req->sync.off = READ_ONCE(sqe->off); | |
2820 | req->sync.len = READ_ONCE(sqe->len); | |
c992fe29 CH |
2821 | return 0; |
2822 | } | |
2823 | ||
8ed8d3c3 JA |
2824 | static bool io_req_cancelled(struct io_kiocb *req) |
2825 | { | |
2826 | if (req->work.flags & IO_WQ_WORK_CANCEL) { | |
2827 | req_set_fail_links(req); | |
2828 | io_cqring_add_event(req, -ECANCELED); | |
2829 | io_put_req(req); | |
2830 | return true; | |
2831 | } | |
2832 | ||
2833 | return false; | |
2834 | } | |
2835 | ||
014db007 | 2836 | static void __io_fsync(struct io_kiocb *req) |
8ed8d3c3 | 2837 | { |
8ed8d3c3 | 2838 | loff_t end = req->sync.off + req->sync.len; |
8ed8d3c3 JA |
2839 | int ret; |
2840 | ||
9adbd45d | 2841 | ret = vfs_fsync_range(req->file, req->sync.off, |
8ed8d3c3 JA |
2842 | end > 0 ? end : LLONG_MAX, |
2843 | req->sync.flags & IORING_FSYNC_DATASYNC); | |
2844 | if (ret < 0) | |
2845 | req_set_fail_links(req); | |
2846 | io_cqring_add_event(req, ret); | |
014db007 | 2847 | io_put_req(req); |
8ed8d3c3 JA |
2848 | } |
2849 | ||
5ea62161 | 2850 | static void io_fsync_finish(struct io_wq_work **workptr) |
c992fe29 | 2851 | { |
5ea62161 | 2852 | struct io_kiocb *req = container_of(*workptr, struct io_kiocb, work); |
5ea62161 PB |
2853 | |
2854 | if (io_req_cancelled(req)) | |
2855 | return; | |
014db007 | 2856 | __io_fsync(req); |
e9fd9396 | 2857 | io_steal_work(req, workptr); |
8ed8d3c3 | 2858 | } |
c992fe29 | 2859 | |
014db007 | 2860 | static int io_fsync(struct io_kiocb *req, bool force_nonblock) |
c992fe29 | 2861 | { |
c992fe29 | 2862 | /* fsync always requires a blocking context */ |
8ed8d3c3 | 2863 | if (force_nonblock) { |
8ed8d3c3 | 2864 | req->work.func = io_fsync_finish; |
c992fe29 | 2865 | return -EAGAIN; |
8ed8d3c3 | 2866 | } |
014db007 | 2867 | __io_fsync(req); |
c992fe29 CH |
2868 | return 0; |
2869 | } | |
2870 | ||
014db007 | 2871 | static void __io_fallocate(struct io_kiocb *req) |
8ed8d3c3 | 2872 | { |
8ed8d3c3 JA |
2873 | int ret; |
2874 | ||
4ed734b0 | 2875 | current->signal->rlim[RLIMIT_FSIZE].rlim_cur = req->fsize; |
d63d1b5e JA |
2876 | ret = vfs_fallocate(req->file, req->sync.mode, req->sync.off, |
2877 | req->sync.len); | |
4ed734b0 | 2878 | current->signal->rlim[RLIMIT_FSIZE].rlim_cur = RLIM_INFINITY; |
8ed8d3c3 JA |
2879 | if (ret < 0) |
2880 | req_set_fail_links(req); | |
2881 | io_cqring_add_event(req, ret); | |
014db007 | 2882 | io_put_req(req); |
5ea62161 PB |
2883 | } |
2884 | ||
2885 | static void io_fallocate_finish(struct io_wq_work **workptr) | |
2886 | { | |
2887 | struct io_kiocb *req = container_of(*workptr, struct io_kiocb, work); | |
5ea62161 | 2888 | |
594506fe PB |
2889 | if (io_req_cancelled(req)) |
2890 | return; | |
014db007 | 2891 | __io_fallocate(req); |
e9fd9396 | 2892 | io_steal_work(req, workptr); |
5d17b4a4 JA |
2893 | } |
2894 | ||
d63d1b5e JA |
2895 | static int io_fallocate_prep(struct io_kiocb *req, |
2896 | const struct io_uring_sqe *sqe) | |
2897 | { | |
2898 | if (sqe->ioprio || sqe->buf_index || sqe->rw_flags) | |
2899 | return -EINVAL; | |
2900 | ||
2901 | req->sync.off = READ_ONCE(sqe->off); | |
2902 | req->sync.len = READ_ONCE(sqe->addr); | |
2903 | req->sync.mode = READ_ONCE(sqe->len); | |
4ed734b0 | 2904 | req->fsize = rlimit(RLIMIT_FSIZE); |
d63d1b5e JA |
2905 | return 0; |
2906 | } | |
2907 | ||
014db007 | 2908 | static int io_fallocate(struct io_kiocb *req, bool force_nonblock) |
5d17b4a4 | 2909 | { |
d63d1b5e | 2910 | /* fallocate always requiring blocking context */ |
8ed8d3c3 | 2911 | if (force_nonblock) { |
d63d1b5e | 2912 | req->work.func = io_fallocate_finish; |
5d17b4a4 | 2913 | return -EAGAIN; |
8ed8d3c3 | 2914 | } |
5d17b4a4 | 2915 | |
014db007 | 2916 | __io_fallocate(req); |
5d17b4a4 JA |
2917 | return 0; |
2918 | } | |
2919 | ||
15b71abe | 2920 | static int io_openat_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
b7bb4f7d | 2921 | { |
f8748881 | 2922 | const char __user *fname; |
15b71abe | 2923 | int ret; |
b7bb4f7d | 2924 | |
15b71abe JA |
2925 | if (sqe->ioprio || sqe->buf_index) |
2926 | return -EINVAL; | |
9c280f90 | 2927 | if (req->flags & REQ_F_FIXED_FILE) |
cf3040ca | 2928 | return -EBADF; |
0bdbdd08 PB |
2929 | if (req->flags & REQ_F_NEED_CLEANUP) |
2930 | return 0; | |
03b1230c | 2931 | |
15b71abe | 2932 | req->open.dfd = READ_ONCE(sqe->fd); |
c12cedf2 | 2933 | req->open.how.mode = READ_ONCE(sqe->len); |
f8748881 | 2934 | fname = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
c12cedf2 | 2935 | req->open.how.flags = READ_ONCE(sqe->open_flags); |
08a1d26e JA |
2936 | if (force_o_largefile()) |
2937 | req->open.how.flags |= O_LARGEFILE; | |
3529d8c2 | 2938 | |
f8748881 | 2939 | req->open.filename = getname(fname); |
15b71abe JA |
2940 | if (IS_ERR(req->open.filename)) { |
2941 | ret = PTR_ERR(req->open.filename); | |
2942 | req->open.filename = NULL; | |
2943 | return ret; | |
2944 | } | |
3529d8c2 | 2945 | |
4022e7af | 2946 | req->open.nofile = rlimit(RLIMIT_NOFILE); |
8fef80bf | 2947 | req->flags |= REQ_F_NEED_CLEANUP; |
15b71abe | 2948 | return 0; |
03b1230c JA |
2949 | } |
2950 | ||
cebdb986 | 2951 | static int io_openat2_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
aa1fa28f | 2952 | { |
cebdb986 JA |
2953 | struct open_how __user *how; |
2954 | const char __user *fname; | |
2955 | size_t len; | |
0fa03c62 JA |
2956 | int ret; |
2957 | ||
cebdb986 | 2958 | if (sqe->ioprio || sqe->buf_index) |
0fa03c62 | 2959 | return -EINVAL; |
9c280f90 | 2960 | if (req->flags & REQ_F_FIXED_FILE) |
cf3040ca | 2961 | return -EBADF; |
0bdbdd08 PB |
2962 | if (req->flags & REQ_F_NEED_CLEANUP) |
2963 | return 0; | |
0fa03c62 | 2964 | |
cebdb986 JA |
2965 | req->open.dfd = READ_ONCE(sqe->fd); |
2966 | fname = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
2967 | how = u64_to_user_ptr(READ_ONCE(sqe->addr2)); | |
2968 | len = READ_ONCE(sqe->len); | |
0fa03c62 | 2969 | |
cebdb986 JA |
2970 | if (len < OPEN_HOW_SIZE_VER0) |
2971 | return -EINVAL; | |
3529d8c2 | 2972 | |
cebdb986 JA |
2973 | ret = copy_struct_from_user(&req->open.how, sizeof(req->open.how), how, |
2974 | len); | |
2975 | if (ret) | |
2976 | return ret; | |
3529d8c2 | 2977 | |
cebdb986 JA |
2978 | if (!(req->open.how.flags & O_PATH) && force_o_largefile()) |
2979 | req->open.how.flags |= O_LARGEFILE; | |
0fa03c62 | 2980 | |
cebdb986 JA |
2981 | req->open.filename = getname(fname); |
2982 | if (IS_ERR(req->open.filename)) { | |
2983 | ret = PTR_ERR(req->open.filename); | |
2984 | req->open.filename = NULL; | |
2985 | return ret; | |
2986 | } | |
2987 | ||
4022e7af | 2988 | req->open.nofile = rlimit(RLIMIT_NOFILE); |
8fef80bf | 2989 | req->flags |= REQ_F_NEED_CLEANUP; |
cebdb986 JA |
2990 | return 0; |
2991 | } | |
2992 | ||
014db007 | 2993 | static int io_openat2(struct io_kiocb *req, bool force_nonblock) |
15b71abe JA |
2994 | { |
2995 | struct open_flags op; | |
15b71abe JA |
2996 | struct file *file; |
2997 | int ret; | |
2998 | ||
f86cd20c | 2999 | if (force_nonblock) |
15b71abe | 3000 | return -EAGAIN; |
15b71abe | 3001 | |
cebdb986 | 3002 | ret = build_open_flags(&req->open.how, &op); |
15b71abe JA |
3003 | if (ret) |
3004 | goto err; | |
3005 | ||
4022e7af | 3006 | ret = __get_unused_fd_flags(req->open.how.flags, req->open.nofile); |
15b71abe JA |
3007 | if (ret < 0) |
3008 | goto err; | |
3009 | ||
3010 | file = do_filp_open(req->open.dfd, req->open.filename, &op); | |
3011 | if (IS_ERR(file)) { | |
3012 | put_unused_fd(ret); | |
3013 | ret = PTR_ERR(file); | |
3014 | } else { | |
3015 | fsnotify_open(file); | |
3016 | fd_install(ret, file); | |
3017 | } | |
3018 | err: | |
3019 | putname(req->open.filename); | |
8fef80bf | 3020 | req->flags &= ~REQ_F_NEED_CLEANUP; |
15b71abe JA |
3021 | if (ret < 0) |
3022 | req_set_fail_links(req); | |
3023 | io_cqring_add_event(req, ret); | |
014db007 | 3024 | io_put_req(req); |
15b71abe JA |
3025 | return 0; |
3026 | } | |
3027 | ||
014db007 | 3028 | static int io_openat(struct io_kiocb *req, bool force_nonblock) |
cebdb986 JA |
3029 | { |
3030 | req->open.how = build_open_how(req->open.how.flags, req->open.how.mode); | |
014db007 | 3031 | return io_openat2(req, force_nonblock); |
cebdb986 JA |
3032 | } |
3033 | ||
067524e9 JA |
3034 | static int io_remove_buffers_prep(struct io_kiocb *req, |
3035 | const struct io_uring_sqe *sqe) | |
3036 | { | |
3037 | struct io_provide_buf *p = &req->pbuf; | |
3038 | u64 tmp; | |
3039 | ||
3040 | if (sqe->ioprio || sqe->rw_flags || sqe->addr || sqe->len || sqe->off) | |
3041 | return -EINVAL; | |
3042 | ||
3043 | tmp = READ_ONCE(sqe->fd); | |
3044 | if (!tmp || tmp > USHRT_MAX) | |
3045 | return -EINVAL; | |
3046 | ||
3047 | memset(p, 0, sizeof(*p)); | |
3048 | p->nbufs = tmp; | |
3049 | p->bgid = READ_ONCE(sqe->buf_group); | |
3050 | return 0; | |
3051 | } | |
3052 | ||
3053 | static int __io_remove_buffers(struct io_ring_ctx *ctx, struct io_buffer *buf, | |
3054 | int bgid, unsigned nbufs) | |
3055 | { | |
3056 | unsigned i = 0; | |
3057 | ||
3058 | /* shouldn't happen */ | |
3059 | if (!nbufs) | |
3060 | return 0; | |
3061 | ||
3062 | /* the head kbuf is the list itself */ | |
3063 | while (!list_empty(&buf->list)) { | |
3064 | struct io_buffer *nxt; | |
3065 | ||
3066 | nxt = list_first_entry(&buf->list, struct io_buffer, list); | |
3067 | list_del(&nxt->list); | |
3068 | kfree(nxt); | |
3069 | if (++i == nbufs) | |
3070 | return i; | |
3071 | } | |
3072 | i++; | |
3073 | kfree(buf); | |
3074 | idr_remove(&ctx->io_buffer_idr, bgid); | |
3075 | ||
3076 | return i; | |
3077 | } | |
3078 | ||
3079 | static int io_remove_buffers(struct io_kiocb *req, bool force_nonblock) | |
3080 | { | |
3081 | struct io_provide_buf *p = &req->pbuf; | |
3082 | struct io_ring_ctx *ctx = req->ctx; | |
3083 | struct io_buffer *head; | |
3084 | int ret = 0; | |
3085 | ||
3086 | io_ring_submit_lock(ctx, !force_nonblock); | |
3087 | ||
3088 | lockdep_assert_held(&ctx->uring_lock); | |
3089 | ||
3090 | ret = -ENOENT; | |
3091 | head = idr_find(&ctx->io_buffer_idr, p->bgid); | |
3092 | if (head) | |
3093 | ret = __io_remove_buffers(ctx, head, p->bgid, p->nbufs); | |
3094 | ||
3095 | io_ring_submit_lock(ctx, !force_nonblock); | |
3096 | if (ret < 0) | |
3097 | req_set_fail_links(req); | |
3098 | io_cqring_add_event(req, ret); | |
3099 | io_put_req(req); | |
3100 | return 0; | |
3101 | } | |
3102 | ||
ddf0322d JA |
3103 | static int io_provide_buffers_prep(struct io_kiocb *req, |
3104 | const struct io_uring_sqe *sqe) | |
3105 | { | |
3106 | struct io_provide_buf *p = &req->pbuf; | |
3107 | u64 tmp; | |
3108 | ||
3109 | if (sqe->ioprio || sqe->rw_flags) | |
3110 | return -EINVAL; | |
3111 | ||
3112 | tmp = READ_ONCE(sqe->fd); | |
3113 | if (!tmp || tmp > USHRT_MAX) | |
3114 | return -E2BIG; | |
3115 | p->nbufs = tmp; | |
3116 | p->addr = READ_ONCE(sqe->addr); | |
3117 | p->len = READ_ONCE(sqe->len); | |
3118 | ||
3119 | if (!access_ok(u64_to_user_ptr(p->addr), p->len)) | |
3120 | return -EFAULT; | |
3121 | ||
3122 | p->bgid = READ_ONCE(sqe->buf_group); | |
3123 | tmp = READ_ONCE(sqe->off); | |
3124 | if (tmp > USHRT_MAX) | |
3125 | return -E2BIG; | |
3126 | p->bid = tmp; | |
3127 | return 0; | |
3128 | } | |
3129 | ||
3130 | static int io_add_buffers(struct io_provide_buf *pbuf, struct io_buffer **head) | |
3131 | { | |
3132 | struct io_buffer *buf; | |
3133 | u64 addr = pbuf->addr; | |
3134 | int i, bid = pbuf->bid; | |
3135 | ||
3136 | for (i = 0; i < pbuf->nbufs; i++) { | |
3137 | buf = kmalloc(sizeof(*buf), GFP_KERNEL); | |
3138 | if (!buf) | |
3139 | break; | |
3140 | ||
3141 | buf->addr = addr; | |
3142 | buf->len = pbuf->len; | |
3143 | buf->bid = bid; | |
3144 | addr += pbuf->len; | |
3145 | bid++; | |
3146 | if (!*head) { | |
3147 | INIT_LIST_HEAD(&buf->list); | |
3148 | *head = buf; | |
3149 | } else { | |
3150 | list_add_tail(&buf->list, &(*head)->list); | |
3151 | } | |
3152 | } | |
3153 | ||
3154 | return i ? i : -ENOMEM; | |
3155 | } | |
3156 | ||
ddf0322d JA |
3157 | static int io_provide_buffers(struct io_kiocb *req, bool force_nonblock) |
3158 | { | |
3159 | struct io_provide_buf *p = &req->pbuf; | |
3160 | struct io_ring_ctx *ctx = req->ctx; | |
3161 | struct io_buffer *head, *list; | |
3162 | int ret = 0; | |
3163 | ||
3164 | io_ring_submit_lock(ctx, !force_nonblock); | |
3165 | ||
3166 | lockdep_assert_held(&ctx->uring_lock); | |
3167 | ||
3168 | list = head = idr_find(&ctx->io_buffer_idr, p->bgid); | |
3169 | ||
3170 | ret = io_add_buffers(p, &head); | |
3171 | if (ret < 0) | |
3172 | goto out; | |
3173 | ||
3174 | if (!list) { | |
3175 | ret = idr_alloc(&ctx->io_buffer_idr, head, p->bgid, p->bgid + 1, | |
3176 | GFP_KERNEL); | |
3177 | if (ret < 0) { | |
067524e9 | 3178 | __io_remove_buffers(ctx, head, p->bgid, -1U); |
ddf0322d JA |
3179 | goto out; |
3180 | } | |
3181 | } | |
3182 | out: | |
3183 | io_ring_submit_unlock(ctx, !force_nonblock); | |
3184 | if (ret < 0) | |
3185 | req_set_fail_links(req); | |
3186 | io_cqring_add_event(req, ret); | |
3187 | io_put_req(req); | |
3188 | return 0; | |
cebdb986 JA |
3189 | } |
3190 | ||
3e4827b0 JA |
3191 | static int io_epoll_ctl_prep(struct io_kiocb *req, |
3192 | const struct io_uring_sqe *sqe) | |
3193 | { | |
3194 | #if defined(CONFIG_EPOLL) | |
3195 | if (sqe->ioprio || sqe->buf_index) | |
3196 | return -EINVAL; | |
3197 | ||
3198 | req->epoll.epfd = READ_ONCE(sqe->fd); | |
3199 | req->epoll.op = READ_ONCE(sqe->len); | |
3200 | req->epoll.fd = READ_ONCE(sqe->off); | |
3201 | ||
3202 | if (ep_op_has_event(req->epoll.op)) { | |
3203 | struct epoll_event __user *ev; | |
3204 | ||
3205 | ev = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
3206 | if (copy_from_user(&req->epoll.event, ev, sizeof(*ev))) | |
3207 | return -EFAULT; | |
3208 | } | |
3209 | ||
3210 | return 0; | |
3211 | #else | |
3212 | return -EOPNOTSUPP; | |
3213 | #endif | |
3214 | } | |
3215 | ||
014db007 | 3216 | static int io_epoll_ctl(struct io_kiocb *req, bool force_nonblock) |
3e4827b0 JA |
3217 | { |
3218 | #if defined(CONFIG_EPOLL) | |
3219 | struct io_epoll *ie = &req->epoll; | |
3220 | int ret; | |
3221 | ||
3222 | ret = do_epoll_ctl(ie->epfd, ie->op, ie->fd, &ie->event, force_nonblock); | |
3223 | if (force_nonblock && ret == -EAGAIN) | |
3224 | return -EAGAIN; | |
3225 | ||
3226 | if (ret < 0) | |
3227 | req_set_fail_links(req); | |
3228 | io_cqring_add_event(req, ret); | |
014db007 | 3229 | io_put_req(req); |
3e4827b0 JA |
3230 | return 0; |
3231 | #else | |
3232 | return -EOPNOTSUPP; | |
3233 | #endif | |
3234 | } | |
3235 | ||
c1ca757b JA |
3236 | static int io_madvise_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
3237 | { | |
3238 | #if defined(CONFIG_ADVISE_SYSCALLS) && defined(CONFIG_MMU) | |
3239 | if (sqe->ioprio || sqe->buf_index || sqe->off) | |
3240 | return -EINVAL; | |
3241 | ||
3242 | req->madvise.addr = READ_ONCE(sqe->addr); | |
3243 | req->madvise.len = READ_ONCE(sqe->len); | |
3244 | req->madvise.advice = READ_ONCE(sqe->fadvise_advice); | |
3245 | return 0; | |
3246 | #else | |
3247 | return -EOPNOTSUPP; | |
3248 | #endif | |
3249 | } | |
3250 | ||
014db007 | 3251 | static int io_madvise(struct io_kiocb *req, bool force_nonblock) |
c1ca757b JA |
3252 | { |
3253 | #if defined(CONFIG_ADVISE_SYSCALLS) && defined(CONFIG_MMU) | |
3254 | struct io_madvise *ma = &req->madvise; | |
3255 | int ret; | |
3256 | ||
3257 | if (force_nonblock) | |
3258 | return -EAGAIN; | |
3259 | ||
3260 | ret = do_madvise(ma->addr, ma->len, ma->advice); | |
3261 | if (ret < 0) | |
3262 | req_set_fail_links(req); | |
3263 | io_cqring_add_event(req, ret); | |
014db007 | 3264 | io_put_req(req); |
c1ca757b JA |
3265 | return 0; |
3266 | #else | |
3267 | return -EOPNOTSUPP; | |
3268 | #endif | |
3269 | } | |
3270 | ||
4840e418 JA |
3271 | static int io_fadvise_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
3272 | { | |
3273 | if (sqe->ioprio || sqe->buf_index || sqe->addr) | |
3274 | return -EINVAL; | |
3275 | ||
3276 | req->fadvise.offset = READ_ONCE(sqe->off); | |
3277 | req->fadvise.len = READ_ONCE(sqe->len); | |
3278 | req->fadvise.advice = READ_ONCE(sqe->fadvise_advice); | |
3279 | return 0; | |
3280 | } | |
3281 | ||
014db007 | 3282 | static int io_fadvise(struct io_kiocb *req, bool force_nonblock) |
4840e418 JA |
3283 | { |
3284 | struct io_fadvise *fa = &req->fadvise; | |
3285 | int ret; | |
3286 | ||
3e69426d JA |
3287 | if (force_nonblock) { |
3288 | switch (fa->advice) { | |
3289 | case POSIX_FADV_NORMAL: | |
3290 | case POSIX_FADV_RANDOM: | |
3291 | case POSIX_FADV_SEQUENTIAL: | |
3292 | break; | |
3293 | default: | |
3294 | return -EAGAIN; | |
3295 | } | |
3296 | } | |
4840e418 JA |
3297 | |
3298 | ret = vfs_fadvise(req->file, fa->offset, fa->len, fa->advice); | |
3299 | if (ret < 0) | |
3300 | req_set_fail_links(req); | |
3301 | io_cqring_add_event(req, ret); | |
014db007 | 3302 | io_put_req(req); |
4840e418 JA |
3303 | return 0; |
3304 | } | |
3305 | ||
eddc7ef5 JA |
3306 | static int io_statx_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
3307 | { | |
f8748881 | 3308 | const char __user *fname; |
eddc7ef5 JA |
3309 | unsigned lookup_flags; |
3310 | int ret; | |
3311 | ||
3312 | if (sqe->ioprio || sqe->buf_index) | |
3313 | return -EINVAL; | |
9c280f90 | 3314 | if (req->flags & REQ_F_FIXED_FILE) |
cf3040ca | 3315 | return -EBADF; |
0bdbdd08 PB |
3316 | if (req->flags & REQ_F_NEED_CLEANUP) |
3317 | return 0; | |
eddc7ef5 JA |
3318 | |
3319 | req->open.dfd = READ_ONCE(sqe->fd); | |
3320 | req->open.mask = READ_ONCE(sqe->len); | |
f8748881 | 3321 | fname = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
eddc7ef5 | 3322 | req->open.buffer = u64_to_user_ptr(READ_ONCE(sqe->addr2)); |
c12cedf2 | 3323 | req->open.how.flags = READ_ONCE(sqe->statx_flags); |
eddc7ef5 | 3324 | |
c12cedf2 | 3325 | if (vfs_stat_set_lookup_flags(&lookup_flags, req->open.how.flags)) |
eddc7ef5 JA |
3326 | return -EINVAL; |
3327 | ||
f8748881 | 3328 | req->open.filename = getname_flags(fname, lookup_flags, NULL); |
eddc7ef5 JA |
3329 | if (IS_ERR(req->open.filename)) { |
3330 | ret = PTR_ERR(req->open.filename); | |
3331 | req->open.filename = NULL; | |
3332 | return ret; | |
3333 | } | |
3334 | ||
8fef80bf | 3335 | req->flags |= REQ_F_NEED_CLEANUP; |
eddc7ef5 JA |
3336 | return 0; |
3337 | } | |
3338 | ||
014db007 | 3339 | static int io_statx(struct io_kiocb *req, bool force_nonblock) |
eddc7ef5 JA |
3340 | { |
3341 | struct io_open *ctx = &req->open; | |
3342 | unsigned lookup_flags; | |
3343 | struct path path; | |
3344 | struct kstat stat; | |
3345 | int ret; | |
3346 | ||
5b0bbee4 JA |
3347 | if (force_nonblock) { |
3348 | /* only need file table for an actual valid fd */ | |
3349 | if (ctx->dfd == -1 || ctx->dfd == AT_FDCWD) | |
3350 | req->flags |= REQ_F_NO_FILE_TABLE; | |
eddc7ef5 | 3351 | return -EAGAIN; |
5b0bbee4 | 3352 | } |
eddc7ef5 | 3353 | |
c12cedf2 | 3354 | if (vfs_stat_set_lookup_flags(&lookup_flags, ctx->how.flags)) |
eddc7ef5 JA |
3355 | return -EINVAL; |
3356 | ||
3357 | retry: | |
3358 | /* filename_lookup() drops it, keep a reference */ | |
3359 | ctx->filename->refcnt++; | |
3360 | ||
3361 | ret = filename_lookup(ctx->dfd, ctx->filename, lookup_flags, &path, | |
3362 | NULL); | |
3363 | if (ret) | |
3364 | goto err; | |
3365 | ||
c12cedf2 | 3366 | ret = vfs_getattr(&path, &stat, ctx->mask, ctx->how.flags); |
eddc7ef5 JA |
3367 | path_put(&path); |
3368 | if (retry_estale(ret, lookup_flags)) { | |
3369 | lookup_flags |= LOOKUP_REVAL; | |
3370 | goto retry; | |
3371 | } | |
3372 | if (!ret) | |
3373 | ret = cp_statx(&stat, ctx->buffer); | |
3374 | err: | |
3375 | putname(ctx->filename); | |
8fef80bf | 3376 | req->flags &= ~REQ_F_NEED_CLEANUP; |
eddc7ef5 JA |
3377 | if (ret < 0) |
3378 | req_set_fail_links(req); | |
3379 | io_cqring_add_event(req, ret); | |
014db007 | 3380 | io_put_req(req); |
eddc7ef5 JA |
3381 | return 0; |
3382 | } | |
3383 | ||
b5dba59e JA |
3384 | static int io_close_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
3385 | { | |
3386 | /* | |
3387 | * If we queue this for async, it must not be cancellable. That would | |
3388 | * leave the 'file' in an undeterminate state. | |
3389 | */ | |
3390 | req->work.flags |= IO_WQ_WORK_NO_CANCEL; | |
3391 | ||
3392 | if (sqe->ioprio || sqe->off || sqe->addr || sqe->len || | |
3393 | sqe->rw_flags || sqe->buf_index) | |
3394 | return -EINVAL; | |
9c280f90 | 3395 | if (req->flags & REQ_F_FIXED_FILE) |
cf3040ca | 3396 | return -EBADF; |
b5dba59e JA |
3397 | |
3398 | req->close.fd = READ_ONCE(sqe->fd); | |
3399 | if (req->file->f_op == &io_uring_fops || | |
b14cca0c | 3400 | req->close.fd == req->ctx->ring_fd) |
b5dba59e JA |
3401 | return -EBADF; |
3402 | ||
3403 | return 0; | |
3404 | } | |
3405 | ||
a93b3331 | 3406 | /* only called when __close_fd_get_file() is done */ |
014db007 | 3407 | static void __io_close_finish(struct io_kiocb *req) |
a93b3331 PB |
3408 | { |
3409 | int ret; | |
3410 | ||
3411 | ret = filp_close(req->close.put_file, req->work.files); | |
3412 | if (ret < 0) | |
3413 | req_set_fail_links(req); | |
3414 | io_cqring_add_event(req, ret); | |
3415 | fput(req->close.put_file); | |
014db007 | 3416 | io_put_req(req); |
a93b3331 PB |
3417 | } |
3418 | ||
b5dba59e JA |
3419 | static void io_close_finish(struct io_wq_work **workptr) |
3420 | { | |
3421 | struct io_kiocb *req = container_of(*workptr, struct io_kiocb, work); | |
b5dba59e | 3422 | |
7fbeb95d | 3423 | /* not cancellable, don't do io_req_cancelled() */ |
014db007 | 3424 | __io_close_finish(req); |
e9fd9396 | 3425 | io_steal_work(req, workptr); |
b5dba59e JA |
3426 | } |
3427 | ||
014db007 | 3428 | static int io_close(struct io_kiocb *req, bool force_nonblock) |
b5dba59e JA |
3429 | { |
3430 | int ret; | |
3431 | ||
3432 | req->close.put_file = NULL; | |
3433 | ret = __close_fd_get_file(req->close.fd, &req->close.put_file); | |
3434 | if (ret < 0) | |
3435 | return ret; | |
3436 | ||
3437 | /* if the file has a flush method, be safe and punt to async */ | |
a2100672 | 3438 | if (req->close.put_file->f_op->flush && force_nonblock) { |
594506fe PB |
3439 | /* submission ref will be dropped, take it for async */ |
3440 | refcount_inc(&req->refs); | |
3441 | ||
a2100672 PB |
3442 | req->work.func = io_close_finish; |
3443 | /* | |
3444 | * Do manual async queue here to avoid grabbing files - we don't | |
3445 | * need the files, and it'll cause io_close_finish() to close | |
3446 | * the file again and cause a double CQE entry for this request | |
3447 | */ | |
3448 | io_queue_async_work(req); | |
3449 | return 0; | |
3450 | } | |
b5dba59e JA |
3451 | |
3452 | /* | |
3453 | * No ->flush(), safely close from here and just punt the | |
3454 | * fput() to async context. | |
3455 | */ | |
014db007 | 3456 | __io_close_finish(req); |
1a417f4e | 3457 | return 0; |
b5dba59e JA |
3458 | } |
3459 | ||
3529d8c2 | 3460 | static int io_prep_sfr(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
5d17b4a4 JA |
3461 | { |
3462 | struct io_ring_ctx *ctx = req->ctx; | |
5d17b4a4 JA |
3463 | |
3464 | if (!req->file) | |
3465 | return -EBADF; | |
5d17b4a4 JA |
3466 | |
3467 | if (unlikely(ctx->flags & IORING_SETUP_IOPOLL)) | |
3468 | return -EINVAL; | |
3469 | if (unlikely(sqe->addr || sqe->ioprio || sqe->buf_index)) | |
3470 | return -EINVAL; | |
3471 | ||
8ed8d3c3 JA |
3472 | req->sync.off = READ_ONCE(sqe->off); |
3473 | req->sync.len = READ_ONCE(sqe->len); | |
3474 | req->sync.flags = READ_ONCE(sqe->sync_range_flags); | |
8ed8d3c3 JA |
3475 | return 0; |
3476 | } | |
3477 | ||
014db007 | 3478 | static void __io_sync_file_range(struct io_kiocb *req) |
8ed8d3c3 | 3479 | { |
8ed8d3c3 JA |
3480 | int ret; |
3481 | ||
9adbd45d | 3482 | ret = sync_file_range(req->file, req->sync.off, req->sync.len, |
8ed8d3c3 JA |
3483 | req->sync.flags); |
3484 | if (ret < 0) | |
3485 | req_set_fail_links(req); | |
3486 | io_cqring_add_event(req, ret); | |
014db007 | 3487 | io_put_req(req); |
5ea62161 PB |
3488 | } |
3489 | ||
3490 | ||
3491 | static void io_sync_file_range_finish(struct io_wq_work **workptr) | |
3492 | { | |
3493 | struct io_kiocb *req = container_of(*workptr, struct io_kiocb, work); | |
5ea62161 PB |
3494 | |
3495 | if (io_req_cancelled(req)) | |
3496 | return; | |
014db007 | 3497 | __io_sync_file_range(req); |
7759a0bf | 3498 | io_steal_work(req, workptr); |
5d17b4a4 JA |
3499 | } |
3500 | ||
014db007 | 3501 | static int io_sync_file_range(struct io_kiocb *req, bool force_nonblock) |
5d17b4a4 | 3502 | { |
5d17b4a4 | 3503 | /* sync_file_range always requires a blocking context */ |
8ed8d3c3 | 3504 | if (force_nonblock) { |
8ed8d3c3 | 3505 | req->work.func = io_sync_file_range_finish; |
5d17b4a4 | 3506 | return -EAGAIN; |
8ed8d3c3 | 3507 | } |
5d17b4a4 | 3508 | |
014db007 | 3509 | __io_sync_file_range(req); |
5d17b4a4 JA |
3510 | return 0; |
3511 | } | |
3512 | ||
469956e8 | 3513 | #if defined(CONFIG_NET) |
02d27d89 PB |
3514 | static int io_setup_async_msg(struct io_kiocb *req, |
3515 | struct io_async_msghdr *kmsg) | |
3516 | { | |
3517 | if (req->io) | |
3518 | return -EAGAIN; | |
3519 | if (io_alloc_async_ctx(req)) { | |
3520 | if (kmsg->iov != kmsg->fast_iov) | |
3521 | kfree(kmsg->iov); | |
3522 | return -ENOMEM; | |
3523 | } | |
3524 | req->flags |= REQ_F_NEED_CLEANUP; | |
3525 | memcpy(&req->io->msg, kmsg, sizeof(*kmsg)); | |
3526 | return -EAGAIN; | |
3527 | } | |
3528 | ||
3529d8c2 | 3529 | static int io_sendmsg_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
03b1230c | 3530 | { |
e47293fd | 3531 | struct io_sr_msg *sr = &req->sr_msg; |
3529d8c2 | 3532 | struct io_async_ctx *io = req->io; |
99bc4c38 | 3533 | int ret; |
03b1230c | 3534 | |
e47293fd JA |
3535 | sr->msg_flags = READ_ONCE(sqe->msg_flags); |
3536 | sr->msg = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
fddaface | 3537 | sr->len = READ_ONCE(sqe->len); |
3529d8c2 | 3538 | |
d8768362 JA |
3539 | #ifdef CONFIG_COMPAT |
3540 | if (req->ctx->compat) | |
3541 | sr->msg_flags |= MSG_CMSG_COMPAT; | |
3542 | #endif | |
3543 | ||
fddaface | 3544 | if (!io || req->opcode == IORING_OP_SEND) |
3529d8c2 | 3545 | return 0; |
5f798bea PB |
3546 | /* iovec is already imported */ |
3547 | if (req->flags & REQ_F_NEED_CLEANUP) | |
3548 | return 0; | |
3529d8c2 | 3549 | |
d9688565 | 3550 | io->msg.iov = io->msg.fast_iov; |
99bc4c38 | 3551 | ret = sendmsg_copy_msghdr(&io->msg.msg, sr->msg, sr->msg_flags, |
e47293fd | 3552 | &io->msg.iov); |
99bc4c38 PB |
3553 | if (!ret) |
3554 | req->flags |= REQ_F_NEED_CLEANUP; | |
3555 | return ret; | |
03b1230c JA |
3556 | } |
3557 | ||
014db007 | 3558 | static int io_sendmsg(struct io_kiocb *req, bool force_nonblock) |
aa1fa28f | 3559 | { |
0b416c3e | 3560 | struct io_async_msghdr *kmsg = NULL; |
0fa03c62 JA |
3561 | struct socket *sock; |
3562 | int ret; | |
3563 | ||
3564 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) | |
3565 | return -EINVAL; | |
3566 | ||
3567 | sock = sock_from_file(req->file, &ret); | |
3568 | if (sock) { | |
b7bb4f7d | 3569 | struct io_async_ctx io; |
0fa03c62 JA |
3570 | unsigned flags; |
3571 | ||
03b1230c | 3572 | if (req->io) { |
0b416c3e | 3573 | kmsg = &req->io->msg; |
b537916c | 3574 | kmsg->msg.msg_name = &req->io->msg.addr; |
0b416c3e JA |
3575 | /* if iov is set, it's allocated already */ |
3576 | if (!kmsg->iov) | |
3577 | kmsg->iov = kmsg->fast_iov; | |
3578 | kmsg->msg.msg_iter.iov = kmsg->iov; | |
03b1230c | 3579 | } else { |
3529d8c2 JA |
3580 | struct io_sr_msg *sr = &req->sr_msg; |
3581 | ||
0b416c3e | 3582 | kmsg = &io.msg; |
b537916c | 3583 | kmsg->msg.msg_name = &io.msg.addr; |
3529d8c2 JA |
3584 | |
3585 | io.msg.iov = io.msg.fast_iov; | |
3586 | ret = sendmsg_copy_msghdr(&io.msg.msg, sr->msg, | |
3587 | sr->msg_flags, &io.msg.iov); | |
03b1230c | 3588 | if (ret) |
3529d8c2 | 3589 | return ret; |
03b1230c | 3590 | } |
0fa03c62 | 3591 | |
e47293fd JA |
3592 | flags = req->sr_msg.msg_flags; |
3593 | if (flags & MSG_DONTWAIT) | |
3594 | req->flags |= REQ_F_NOWAIT; | |
3595 | else if (force_nonblock) | |
3596 | flags |= MSG_DONTWAIT; | |
3597 | ||
0b416c3e | 3598 | ret = __sys_sendmsg_sock(sock, &kmsg->msg, flags); |
02d27d89 PB |
3599 | if (force_nonblock && ret == -EAGAIN) |
3600 | return io_setup_async_msg(req, kmsg); | |
441cdbd5 JA |
3601 | if (ret == -ERESTARTSYS) |
3602 | ret = -EINTR; | |
0fa03c62 JA |
3603 | } |
3604 | ||
1e95081c | 3605 | if (kmsg && kmsg->iov != kmsg->fast_iov) |
0b416c3e | 3606 | kfree(kmsg->iov); |
99bc4c38 | 3607 | req->flags &= ~REQ_F_NEED_CLEANUP; |
78e19bbe | 3608 | io_cqring_add_event(req, ret); |
4e88d6e7 JA |
3609 | if (ret < 0) |
3610 | req_set_fail_links(req); | |
014db007 | 3611 | io_put_req(req); |
5d17b4a4 | 3612 | return 0; |
03b1230c | 3613 | } |
aa1fa28f | 3614 | |
014db007 | 3615 | static int io_send(struct io_kiocb *req, bool force_nonblock) |
fddaface | 3616 | { |
fddaface JA |
3617 | struct socket *sock; |
3618 | int ret; | |
3619 | ||
3620 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) | |
3621 | return -EINVAL; | |
3622 | ||
3623 | sock = sock_from_file(req->file, &ret); | |
3624 | if (sock) { | |
3625 | struct io_sr_msg *sr = &req->sr_msg; | |
3626 | struct msghdr msg; | |
3627 | struct iovec iov; | |
3628 | unsigned flags; | |
3629 | ||
3630 | ret = import_single_range(WRITE, sr->buf, sr->len, &iov, | |
3631 | &msg.msg_iter); | |
3632 | if (ret) | |
3633 | return ret; | |
3634 | ||
3635 | msg.msg_name = NULL; | |
3636 | msg.msg_control = NULL; | |
3637 | msg.msg_controllen = 0; | |
3638 | msg.msg_namelen = 0; | |
3639 | ||
3640 | flags = req->sr_msg.msg_flags; | |
3641 | if (flags & MSG_DONTWAIT) | |
3642 | req->flags |= REQ_F_NOWAIT; | |
3643 | else if (force_nonblock) | |
3644 | flags |= MSG_DONTWAIT; | |
3645 | ||
0b7b21e4 JA |
3646 | msg.msg_flags = flags; |
3647 | ret = sock_sendmsg(sock, &msg); | |
fddaface JA |
3648 | if (force_nonblock && ret == -EAGAIN) |
3649 | return -EAGAIN; | |
3650 | if (ret == -ERESTARTSYS) | |
3651 | ret = -EINTR; | |
3652 | } | |
3653 | ||
3654 | io_cqring_add_event(req, ret); | |
3655 | if (ret < 0) | |
3656 | req_set_fail_links(req); | |
014db007 | 3657 | io_put_req(req); |
fddaface | 3658 | return 0; |
fddaface JA |
3659 | } |
3660 | ||
52de1fe1 JA |
3661 | static int __io_recvmsg_copy_hdr(struct io_kiocb *req, struct io_async_ctx *io) |
3662 | { | |
3663 | struct io_sr_msg *sr = &req->sr_msg; | |
3664 | struct iovec __user *uiov; | |
3665 | size_t iov_len; | |
3666 | int ret; | |
3667 | ||
3668 | ret = __copy_msghdr_from_user(&io->msg.msg, sr->msg, &io->msg.uaddr, | |
3669 | &uiov, &iov_len); | |
3670 | if (ret) | |
3671 | return ret; | |
3672 | ||
3673 | if (req->flags & REQ_F_BUFFER_SELECT) { | |
3674 | if (iov_len > 1) | |
3675 | return -EINVAL; | |
3676 | if (copy_from_user(io->msg.iov, uiov, sizeof(*uiov))) | |
3677 | return -EFAULT; | |
3678 | sr->len = io->msg.iov[0].iov_len; | |
3679 | iov_iter_init(&io->msg.msg.msg_iter, READ, io->msg.iov, 1, | |
3680 | sr->len); | |
3681 | io->msg.iov = NULL; | |
3682 | } else { | |
3683 | ret = import_iovec(READ, uiov, iov_len, UIO_FASTIOV, | |
3684 | &io->msg.iov, &io->msg.msg.msg_iter); | |
3685 | if (ret > 0) | |
3686 | ret = 0; | |
3687 | } | |
3688 | ||
3689 | return ret; | |
3690 | } | |
3691 | ||
3692 | #ifdef CONFIG_COMPAT | |
3693 | static int __io_compat_recvmsg_copy_hdr(struct io_kiocb *req, | |
3694 | struct io_async_ctx *io) | |
3695 | { | |
3696 | struct compat_msghdr __user *msg_compat; | |
3697 | struct io_sr_msg *sr = &req->sr_msg; | |
3698 | struct compat_iovec __user *uiov; | |
3699 | compat_uptr_t ptr; | |
3700 | compat_size_t len; | |
3701 | int ret; | |
3702 | ||
3703 | msg_compat = (struct compat_msghdr __user *) sr->msg; | |
3704 | ret = __get_compat_msghdr(&io->msg.msg, msg_compat, &io->msg.uaddr, | |
3705 | &ptr, &len); | |
3706 | if (ret) | |
3707 | return ret; | |
3708 | ||
3709 | uiov = compat_ptr(ptr); | |
3710 | if (req->flags & REQ_F_BUFFER_SELECT) { | |
3711 | compat_ssize_t clen; | |
3712 | ||
3713 | if (len > 1) | |
3714 | return -EINVAL; | |
3715 | if (!access_ok(uiov, sizeof(*uiov))) | |
3716 | return -EFAULT; | |
3717 | if (__get_user(clen, &uiov->iov_len)) | |
3718 | return -EFAULT; | |
3719 | if (clen < 0) | |
3720 | return -EINVAL; | |
3721 | sr->len = io->msg.iov[0].iov_len; | |
3722 | io->msg.iov = NULL; | |
3723 | } else { | |
3724 | ret = compat_import_iovec(READ, uiov, len, UIO_FASTIOV, | |
3725 | &io->msg.iov, | |
3726 | &io->msg.msg.msg_iter); | |
3727 | if (ret < 0) | |
3728 | return ret; | |
3729 | } | |
3730 | ||
3731 | return 0; | |
3732 | } | |
3733 | #endif | |
3734 | ||
3735 | static int io_recvmsg_copy_hdr(struct io_kiocb *req, struct io_async_ctx *io) | |
3736 | { | |
3737 | io->msg.iov = io->msg.fast_iov; | |
3738 | ||
3739 | #ifdef CONFIG_COMPAT | |
3740 | if (req->ctx->compat) | |
3741 | return __io_compat_recvmsg_copy_hdr(req, io); | |
fddaface | 3742 | #endif |
52de1fe1 JA |
3743 | |
3744 | return __io_recvmsg_copy_hdr(req, io); | |
3745 | } | |
3746 | ||
bcda7baa JA |
3747 | static struct io_buffer *io_recv_buffer_select(struct io_kiocb *req, |
3748 | int *cflags, bool needs_lock) | |
3749 | { | |
3750 | struct io_sr_msg *sr = &req->sr_msg; | |
3751 | struct io_buffer *kbuf; | |
3752 | ||
3753 | if (!(req->flags & REQ_F_BUFFER_SELECT)) | |
3754 | return NULL; | |
3755 | ||
3756 | kbuf = io_buffer_select(req, &sr->len, sr->bgid, sr->kbuf, needs_lock); | |
3757 | if (IS_ERR(kbuf)) | |
3758 | return kbuf; | |
3759 | ||
3760 | sr->kbuf = kbuf; | |
3761 | req->flags |= REQ_F_BUFFER_SELECTED; | |
3762 | ||
3763 | *cflags = kbuf->bid << IORING_CQE_BUFFER_SHIFT; | |
3764 | *cflags |= IORING_CQE_F_BUFFER; | |
3765 | return kbuf; | |
fddaface JA |
3766 | } |
3767 | ||
3529d8c2 JA |
3768 | static int io_recvmsg_prep(struct io_kiocb *req, |
3769 | const struct io_uring_sqe *sqe) | |
aa1fa28f | 3770 | { |
e47293fd | 3771 | struct io_sr_msg *sr = &req->sr_msg; |
3529d8c2 | 3772 | struct io_async_ctx *io = req->io; |
99bc4c38 | 3773 | int ret; |
3529d8c2 JA |
3774 | |
3775 | sr->msg_flags = READ_ONCE(sqe->msg_flags); | |
3776 | sr->msg = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
0b7b21e4 | 3777 | sr->len = READ_ONCE(sqe->len); |
bcda7baa | 3778 | sr->bgid = READ_ONCE(sqe->buf_group); |
06b76d44 | 3779 | |
d8768362 JA |
3780 | #ifdef CONFIG_COMPAT |
3781 | if (req->ctx->compat) | |
3782 | sr->msg_flags |= MSG_CMSG_COMPAT; | |
3783 | #endif | |
3784 | ||
fddaface | 3785 | if (!io || req->opcode == IORING_OP_RECV) |
06b76d44 | 3786 | return 0; |
5f798bea PB |
3787 | /* iovec is already imported */ |
3788 | if (req->flags & REQ_F_NEED_CLEANUP) | |
3789 | return 0; | |
03b1230c | 3790 | |
52de1fe1 | 3791 | ret = io_recvmsg_copy_hdr(req, io); |
99bc4c38 PB |
3792 | if (!ret) |
3793 | req->flags |= REQ_F_NEED_CLEANUP; | |
3794 | return ret; | |
aa1fa28f JA |
3795 | } |
3796 | ||
014db007 | 3797 | static int io_recvmsg(struct io_kiocb *req, bool force_nonblock) |
aa1fa28f | 3798 | { |
0b416c3e | 3799 | struct io_async_msghdr *kmsg = NULL; |
03b1230c | 3800 | struct socket *sock; |
52de1fe1 | 3801 | int ret, cflags = 0; |
03b1230c JA |
3802 | |
3803 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) | |
3804 | return -EINVAL; | |
3805 | ||
3806 | sock = sock_from_file(req->file, &ret); | |
3807 | if (sock) { | |
52de1fe1 | 3808 | struct io_buffer *kbuf; |
b7bb4f7d | 3809 | struct io_async_ctx io; |
03b1230c JA |
3810 | unsigned flags; |
3811 | ||
03b1230c | 3812 | if (req->io) { |
0b416c3e | 3813 | kmsg = &req->io->msg; |
b537916c | 3814 | kmsg->msg.msg_name = &req->io->msg.addr; |
0b416c3e JA |
3815 | /* if iov is set, it's allocated already */ |
3816 | if (!kmsg->iov) | |
3817 | kmsg->iov = kmsg->fast_iov; | |
3818 | kmsg->msg.msg_iter.iov = kmsg->iov; | |
03b1230c | 3819 | } else { |
0b416c3e | 3820 | kmsg = &io.msg; |
b537916c | 3821 | kmsg->msg.msg_name = &io.msg.addr; |
3529d8c2 | 3822 | |
52de1fe1 | 3823 | ret = io_recvmsg_copy_hdr(req, &io); |
03b1230c | 3824 | if (ret) |
3529d8c2 | 3825 | return ret; |
03b1230c JA |
3826 | } |
3827 | ||
52de1fe1 JA |
3828 | kbuf = io_recv_buffer_select(req, &cflags, !force_nonblock); |
3829 | if (IS_ERR(kbuf)) { | |
3830 | return PTR_ERR(kbuf); | |
3831 | } else if (kbuf) { | |
3832 | kmsg->fast_iov[0].iov_base = u64_to_user_ptr(kbuf->addr); | |
3833 | iov_iter_init(&kmsg->msg.msg_iter, READ, kmsg->iov, | |
3834 | 1, req->sr_msg.len); | |
3835 | } | |
3836 | ||
e47293fd JA |
3837 | flags = req->sr_msg.msg_flags; |
3838 | if (flags & MSG_DONTWAIT) | |
3839 | req->flags |= REQ_F_NOWAIT; | |
3840 | else if (force_nonblock) | |
3841 | flags |= MSG_DONTWAIT; | |
3842 | ||
3843 | ret = __sys_recvmsg_sock(sock, &kmsg->msg, req->sr_msg.msg, | |
3844 | kmsg->uaddr, flags); | |
02d27d89 PB |
3845 | if (force_nonblock && ret == -EAGAIN) |
3846 | return io_setup_async_msg(req, kmsg); | |
03b1230c JA |
3847 | if (ret == -ERESTARTSYS) |
3848 | ret = -EINTR; | |
3849 | } | |
3850 | ||
1e95081c | 3851 | if (kmsg && kmsg->iov != kmsg->fast_iov) |
0b416c3e | 3852 | kfree(kmsg->iov); |
99bc4c38 | 3853 | req->flags &= ~REQ_F_NEED_CLEANUP; |
52de1fe1 | 3854 | __io_cqring_add_event(req, ret, cflags); |
4e88d6e7 JA |
3855 | if (ret < 0) |
3856 | req_set_fail_links(req); | |
014db007 | 3857 | io_put_req(req); |
03b1230c | 3858 | return 0; |
0fa03c62 | 3859 | } |
5d17b4a4 | 3860 | |
014db007 | 3861 | static int io_recv(struct io_kiocb *req, bool force_nonblock) |
fddaface | 3862 | { |
bcda7baa | 3863 | struct io_buffer *kbuf = NULL; |
fddaface | 3864 | struct socket *sock; |
bcda7baa | 3865 | int ret, cflags = 0; |
fddaface JA |
3866 | |
3867 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) | |
3868 | return -EINVAL; | |
3869 | ||
3870 | sock = sock_from_file(req->file, &ret); | |
3871 | if (sock) { | |
3872 | struct io_sr_msg *sr = &req->sr_msg; | |
bcda7baa | 3873 | void __user *buf = sr->buf; |
fddaface JA |
3874 | struct msghdr msg; |
3875 | struct iovec iov; | |
3876 | unsigned flags; | |
3877 | ||
bcda7baa JA |
3878 | kbuf = io_recv_buffer_select(req, &cflags, !force_nonblock); |
3879 | if (IS_ERR(kbuf)) | |
3880 | return PTR_ERR(kbuf); | |
3881 | else if (kbuf) | |
3882 | buf = u64_to_user_ptr(kbuf->addr); | |
3883 | ||
3884 | ret = import_single_range(READ, buf, sr->len, &iov, | |
fddaface | 3885 | &msg.msg_iter); |
bcda7baa JA |
3886 | if (ret) { |
3887 | kfree(kbuf); | |
fddaface | 3888 | return ret; |
bcda7baa | 3889 | } |
fddaface | 3890 | |
bcda7baa | 3891 | req->flags |= REQ_F_NEED_CLEANUP; |
fddaface JA |
3892 | msg.msg_name = NULL; |
3893 | msg.msg_control = NULL; | |
3894 | msg.msg_controllen = 0; | |
3895 | msg.msg_namelen = 0; | |
3896 | msg.msg_iocb = NULL; | |
3897 | msg.msg_flags = 0; | |
3898 | ||
3899 | flags = req->sr_msg.msg_flags; | |
3900 | if (flags & MSG_DONTWAIT) | |
3901 | req->flags |= REQ_F_NOWAIT; | |
3902 | else if (force_nonblock) | |
3903 | flags |= MSG_DONTWAIT; | |
3904 | ||
0b7b21e4 | 3905 | ret = sock_recvmsg(sock, &msg, flags); |
fddaface JA |
3906 | if (force_nonblock && ret == -EAGAIN) |
3907 | return -EAGAIN; | |
3908 | if (ret == -ERESTARTSYS) | |
3909 | ret = -EINTR; | |
3910 | } | |
3911 | ||
bcda7baa JA |
3912 | kfree(kbuf); |
3913 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
3914 | __io_cqring_add_event(req, ret, cflags); | |
fddaface JA |
3915 | if (ret < 0) |
3916 | req_set_fail_links(req); | |
014db007 | 3917 | io_put_req(req); |
fddaface | 3918 | return 0; |
fddaface JA |
3919 | } |
3920 | ||
3529d8c2 | 3921 | static int io_accept_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
17f2fe35 | 3922 | { |
8ed8d3c3 JA |
3923 | struct io_accept *accept = &req->accept; |
3924 | ||
17f2fe35 JA |
3925 | if (unlikely(req->ctx->flags & (IORING_SETUP_IOPOLL|IORING_SETUP_SQPOLL))) |
3926 | return -EINVAL; | |
8042d6ce | 3927 | if (sqe->ioprio || sqe->len || sqe->buf_index) |
17f2fe35 JA |
3928 | return -EINVAL; |
3929 | ||
d55e5f5b JA |
3930 | accept->addr = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
3931 | accept->addr_len = u64_to_user_ptr(READ_ONCE(sqe->addr2)); | |
8ed8d3c3 | 3932 | accept->flags = READ_ONCE(sqe->accept_flags); |
09952e3e | 3933 | accept->nofile = rlimit(RLIMIT_NOFILE); |
8ed8d3c3 | 3934 | return 0; |
8ed8d3c3 | 3935 | } |
17f2fe35 | 3936 | |
014db007 | 3937 | static int __io_accept(struct io_kiocb *req, bool force_nonblock) |
8ed8d3c3 JA |
3938 | { |
3939 | struct io_accept *accept = &req->accept; | |
3940 | unsigned file_flags; | |
3941 | int ret; | |
3942 | ||
3943 | file_flags = force_nonblock ? O_NONBLOCK : 0; | |
3944 | ret = __sys_accept4_file(req->file, file_flags, accept->addr, | |
09952e3e JA |
3945 | accept->addr_len, accept->flags, |
3946 | accept->nofile); | |
8ed8d3c3 | 3947 | if (ret == -EAGAIN && force_nonblock) |
17f2fe35 | 3948 | return -EAGAIN; |
8e3cca12 JA |
3949 | if (ret == -ERESTARTSYS) |
3950 | ret = -EINTR; | |
4e88d6e7 JA |
3951 | if (ret < 0) |
3952 | req_set_fail_links(req); | |
78e19bbe | 3953 | io_cqring_add_event(req, ret); |
014db007 | 3954 | io_put_req(req); |
17f2fe35 | 3955 | return 0; |
8ed8d3c3 JA |
3956 | } |
3957 | ||
3958 | static void io_accept_finish(struct io_wq_work **workptr) | |
3959 | { | |
3960 | struct io_kiocb *req = container_of(*workptr, struct io_kiocb, work); | |
8ed8d3c3 JA |
3961 | |
3962 | if (io_req_cancelled(req)) | |
3963 | return; | |
014db007 | 3964 | __io_accept(req, false); |
e9fd9396 | 3965 | io_steal_work(req, workptr); |
8ed8d3c3 | 3966 | } |
8ed8d3c3 | 3967 | |
014db007 | 3968 | static int io_accept(struct io_kiocb *req, bool force_nonblock) |
8ed8d3c3 | 3969 | { |
8ed8d3c3 JA |
3970 | int ret; |
3971 | ||
014db007 | 3972 | ret = __io_accept(req, force_nonblock); |
8ed8d3c3 JA |
3973 | if (ret == -EAGAIN && force_nonblock) { |
3974 | req->work.func = io_accept_finish; | |
8ed8d3c3 JA |
3975 | return -EAGAIN; |
3976 | } | |
3977 | return 0; | |
0fa03c62 | 3978 | } |
5d17b4a4 | 3979 | |
3529d8c2 | 3980 | static int io_connect_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
f499a021 | 3981 | { |
3529d8c2 JA |
3982 | struct io_connect *conn = &req->connect; |
3983 | struct io_async_ctx *io = req->io; | |
f499a021 | 3984 | |
3fbb51c1 JA |
3985 | if (unlikely(req->ctx->flags & (IORING_SETUP_IOPOLL|IORING_SETUP_SQPOLL))) |
3986 | return -EINVAL; | |
3987 | if (sqe->ioprio || sqe->len || sqe->buf_index || sqe->rw_flags) | |
3988 | return -EINVAL; | |
3989 | ||
3529d8c2 JA |
3990 | conn->addr = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
3991 | conn->addr_len = READ_ONCE(sqe->addr2); | |
3992 | ||
3993 | if (!io) | |
3994 | return 0; | |
3995 | ||
3996 | return move_addr_to_kernel(conn->addr, conn->addr_len, | |
3fbb51c1 | 3997 | &io->connect.address); |
f499a021 JA |
3998 | } |
3999 | ||
014db007 | 4000 | static int io_connect(struct io_kiocb *req, bool force_nonblock) |
f8e85cf2 | 4001 | { |
f499a021 | 4002 | struct io_async_ctx __io, *io; |
f8e85cf2 | 4003 | unsigned file_flags; |
3fbb51c1 | 4004 | int ret; |
f8e85cf2 | 4005 | |
f499a021 JA |
4006 | if (req->io) { |
4007 | io = req->io; | |
4008 | } else { | |
3529d8c2 JA |
4009 | ret = move_addr_to_kernel(req->connect.addr, |
4010 | req->connect.addr_len, | |
4011 | &__io.connect.address); | |
f499a021 JA |
4012 | if (ret) |
4013 | goto out; | |
4014 | io = &__io; | |
4015 | } | |
4016 | ||
3fbb51c1 JA |
4017 | file_flags = force_nonblock ? O_NONBLOCK : 0; |
4018 | ||
4019 | ret = __sys_connect_file(req->file, &io->connect.address, | |
4020 | req->connect.addr_len, file_flags); | |
87f80d62 | 4021 | if ((ret == -EAGAIN || ret == -EINPROGRESS) && force_nonblock) { |
b7bb4f7d JA |
4022 | if (req->io) |
4023 | return -EAGAIN; | |
4024 | if (io_alloc_async_ctx(req)) { | |
f499a021 JA |
4025 | ret = -ENOMEM; |
4026 | goto out; | |
4027 | } | |
b7bb4f7d | 4028 | memcpy(&req->io->connect, &__io.connect, sizeof(__io.connect)); |
f8e85cf2 | 4029 | return -EAGAIN; |
f499a021 | 4030 | } |
f8e85cf2 JA |
4031 | if (ret == -ERESTARTSYS) |
4032 | ret = -EINTR; | |
f499a021 | 4033 | out: |
4e88d6e7 JA |
4034 | if (ret < 0) |
4035 | req_set_fail_links(req); | |
f8e85cf2 | 4036 | io_cqring_add_event(req, ret); |
014db007 | 4037 | io_put_req(req); |
f8e85cf2 | 4038 | return 0; |
469956e8 Y |
4039 | } |
4040 | #else /* !CONFIG_NET */ | |
4041 | static int io_sendmsg_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) | |
4042 | { | |
f8e85cf2 | 4043 | return -EOPNOTSUPP; |
f8e85cf2 JA |
4044 | } |
4045 | ||
469956e8 Y |
4046 | static int io_sendmsg(struct io_kiocb *req, bool force_nonblock) |
4047 | { | |
4048 | return -EOPNOTSUPP; | |
4049 | } | |
4050 | ||
4051 | static int io_send(struct io_kiocb *req, bool force_nonblock) | |
4052 | { | |
4053 | return -EOPNOTSUPP; | |
4054 | } | |
4055 | ||
4056 | static int io_recvmsg_prep(struct io_kiocb *req, | |
4057 | const struct io_uring_sqe *sqe) | |
4058 | { | |
4059 | return -EOPNOTSUPP; | |
4060 | } | |
4061 | ||
4062 | static int io_recvmsg(struct io_kiocb *req, bool force_nonblock) | |
4063 | { | |
4064 | return -EOPNOTSUPP; | |
4065 | } | |
4066 | ||
4067 | static int io_recv(struct io_kiocb *req, bool force_nonblock) | |
4068 | { | |
4069 | return -EOPNOTSUPP; | |
4070 | } | |
4071 | ||
4072 | static int io_accept_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) | |
4073 | { | |
4074 | return -EOPNOTSUPP; | |
4075 | } | |
4076 | ||
4077 | static int io_accept(struct io_kiocb *req, bool force_nonblock) | |
4078 | { | |
4079 | return -EOPNOTSUPP; | |
4080 | } | |
4081 | ||
4082 | static int io_connect_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) | |
4083 | { | |
4084 | return -EOPNOTSUPP; | |
4085 | } | |
4086 | ||
4087 | static int io_connect(struct io_kiocb *req, bool force_nonblock) | |
4088 | { | |
f8e85cf2 | 4089 | return -EOPNOTSUPP; |
f8e85cf2 | 4090 | } |
469956e8 | 4091 | #endif /* CONFIG_NET */ |
f8e85cf2 | 4092 | |
d7718a9d JA |
4093 | struct io_poll_table { |
4094 | struct poll_table_struct pt; | |
4095 | struct io_kiocb *req; | |
4096 | int error; | |
4097 | }; | |
4098 | ||
4099 | static void __io_queue_proc(struct io_poll_iocb *poll, struct io_poll_table *pt, | |
4100 | struct wait_queue_head *head) | |
4101 | { | |
4102 | if (unlikely(poll->head)) { | |
4103 | pt->error = -EINVAL; | |
4104 | return; | |
4105 | } | |
4106 | ||
4107 | pt->error = 0; | |
4108 | poll->head = head; | |
4109 | add_wait_queue(head, &poll->wait); | |
4110 | } | |
4111 | ||
4112 | static void io_async_queue_proc(struct file *file, struct wait_queue_head *head, | |
4113 | struct poll_table_struct *p) | |
4114 | { | |
4115 | struct io_poll_table *pt = container_of(p, struct io_poll_table, pt); | |
4116 | ||
4117 | __io_queue_proc(&pt->req->apoll->poll, pt, head); | |
4118 | } | |
4119 | ||
4120 | static int __io_async_wake(struct io_kiocb *req, struct io_poll_iocb *poll, | |
4121 | __poll_t mask, task_work_func_t func) | |
4122 | { | |
4123 | struct task_struct *tsk; | |
aa96bf8a | 4124 | int ret; |
d7718a9d JA |
4125 | |
4126 | /* for instances that support it check for an event match first: */ | |
4127 | if (mask && !(mask & poll->events)) | |
4128 | return 0; | |
4129 | ||
4130 | trace_io_uring_task_add(req->ctx, req->opcode, req->user_data, mask); | |
4131 | ||
4132 | list_del_init(&poll->wait.entry); | |
4133 | ||
4134 | tsk = req->task; | |
4135 | req->result = mask; | |
4136 | init_task_work(&req->task_work, func); | |
4137 | /* | |
e3aabf95 JA |
4138 | * If this fails, then the task is exiting. When a task exits, the |
4139 | * work gets canceled, so just cancel this request as well instead | |
4140 | * of executing it. We can't safely execute it anyway, as we may not | |
4141 | * have the needed state needed for it anyway. | |
d7718a9d | 4142 | */ |
aa96bf8a JA |
4143 | ret = task_work_add(tsk, &req->task_work, true); |
4144 | if (unlikely(ret)) { | |
e3aabf95 | 4145 | WRITE_ONCE(poll->canceled, true); |
aa96bf8a JA |
4146 | tsk = io_wq_get_task(req->ctx->io_wq); |
4147 | task_work_add(tsk, &req->task_work, true); | |
4148 | } | |
d7718a9d JA |
4149 | wake_up_process(tsk); |
4150 | return 1; | |
4151 | } | |
4152 | ||
74ce6ce4 JA |
4153 | static bool io_poll_rewait(struct io_kiocb *req, struct io_poll_iocb *poll) |
4154 | __acquires(&req->ctx->completion_lock) | |
4155 | { | |
4156 | struct io_ring_ctx *ctx = req->ctx; | |
4157 | ||
4158 | if (!req->result && !READ_ONCE(poll->canceled)) { | |
4159 | struct poll_table_struct pt = { ._key = poll->events }; | |
4160 | ||
4161 | req->result = vfs_poll(req->file, &pt) & poll->events; | |
4162 | } | |
4163 | ||
4164 | spin_lock_irq(&ctx->completion_lock); | |
4165 | if (!req->result && !READ_ONCE(poll->canceled)) { | |
4166 | add_wait_queue(poll->head, &poll->wait); | |
4167 | return true; | |
4168 | } | |
4169 | ||
4170 | return false; | |
4171 | } | |
4172 | ||
d7718a9d JA |
4173 | static void io_async_task_func(struct callback_head *cb) |
4174 | { | |
4175 | struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work); | |
4176 | struct async_poll *apoll = req->apoll; | |
4177 | struct io_ring_ctx *ctx = req->ctx; | |
2bae047e | 4178 | bool canceled; |
d7718a9d JA |
4179 | |
4180 | trace_io_uring_task_run(req->ctx, req->opcode, req->user_data); | |
4181 | ||
74ce6ce4 | 4182 | if (io_poll_rewait(req, &apoll->poll)) { |
d7718a9d | 4183 | spin_unlock_irq(&ctx->completion_lock); |
74ce6ce4 | 4184 | return; |
d7718a9d JA |
4185 | } |
4186 | ||
74ce6ce4 JA |
4187 | if (hash_hashed(&req->hash_node)) |
4188 | hash_del(&req->hash_node); | |
4189 | ||
2bae047e JA |
4190 | canceled = READ_ONCE(apoll->poll.canceled); |
4191 | if (canceled) { | |
4192 | io_cqring_fill_event(req, -ECANCELED); | |
4193 | io_commit_cqring(ctx); | |
4194 | } | |
4195 | ||
74ce6ce4 JA |
4196 | spin_unlock_irq(&ctx->completion_lock); |
4197 | ||
44575a67 XW |
4198 | /* restore ->work in case we need to retry again */ |
4199 | memcpy(&req->work, &apoll->work, sizeof(req->work)); | |
4200 | ||
2bae047e JA |
4201 | if (canceled) { |
4202 | kfree(apoll); | |
4203 | io_cqring_ev_posted(ctx); | |
4204 | req_set_fail_links(req); | |
44575a67 | 4205 | io_double_put_req(req); |
2bae047e JA |
4206 | return; |
4207 | } | |
4208 | ||
d7718a9d JA |
4209 | __set_current_state(TASK_RUNNING); |
4210 | mutex_lock(&ctx->uring_lock); | |
4211 | __io_queue_sqe(req, NULL); | |
4212 | mutex_unlock(&ctx->uring_lock); | |
4213 | ||
4214 | kfree(apoll); | |
4215 | } | |
4216 | ||
4217 | static int io_async_wake(struct wait_queue_entry *wait, unsigned mode, int sync, | |
4218 | void *key) | |
4219 | { | |
4220 | struct io_kiocb *req = wait->private; | |
4221 | struct io_poll_iocb *poll = &req->apoll->poll; | |
4222 | ||
4223 | trace_io_uring_poll_wake(req->ctx, req->opcode, req->user_data, | |
4224 | key_to_poll(key)); | |
4225 | ||
4226 | return __io_async_wake(req, poll, key_to_poll(key), io_async_task_func); | |
4227 | } | |
4228 | ||
4229 | static void io_poll_req_insert(struct io_kiocb *req) | |
4230 | { | |
4231 | struct io_ring_ctx *ctx = req->ctx; | |
4232 | struct hlist_head *list; | |
4233 | ||
4234 | list = &ctx->cancel_hash[hash_long(req->user_data, ctx->cancel_hash_bits)]; | |
4235 | hlist_add_head(&req->hash_node, list); | |
4236 | } | |
4237 | ||
4238 | static __poll_t __io_arm_poll_handler(struct io_kiocb *req, | |
4239 | struct io_poll_iocb *poll, | |
4240 | struct io_poll_table *ipt, __poll_t mask, | |
4241 | wait_queue_func_t wake_func) | |
4242 | __acquires(&ctx->completion_lock) | |
4243 | { | |
4244 | struct io_ring_ctx *ctx = req->ctx; | |
4245 | bool cancel = false; | |
4246 | ||
4247 | poll->file = req->file; | |
4248 | poll->head = NULL; | |
4249 | poll->done = poll->canceled = false; | |
4250 | poll->events = mask; | |
4251 | ||
4252 | ipt->pt._key = mask; | |
4253 | ipt->req = req; | |
4254 | ipt->error = -EINVAL; | |
4255 | ||
4256 | INIT_LIST_HEAD(&poll->wait.entry); | |
4257 | init_waitqueue_func_entry(&poll->wait, wake_func); | |
4258 | poll->wait.private = req; | |
4259 | ||
4260 | mask = vfs_poll(req->file, &ipt->pt) & poll->events; | |
4261 | ||
4262 | spin_lock_irq(&ctx->completion_lock); | |
4263 | if (likely(poll->head)) { | |
4264 | spin_lock(&poll->head->lock); | |
4265 | if (unlikely(list_empty(&poll->wait.entry))) { | |
4266 | if (ipt->error) | |
4267 | cancel = true; | |
4268 | ipt->error = 0; | |
4269 | mask = 0; | |
4270 | } | |
4271 | if (mask || ipt->error) | |
4272 | list_del_init(&poll->wait.entry); | |
4273 | else if (cancel) | |
4274 | WRITE_ONCE(poll->canceled, true); | |
4275 | else if (!poll->done) /* actually waiting for an event */ | |
4276 | io_poll_req_insert(req); | |
4277 | spin_unlock(&poll->head->lock); | |
4278 | } | |
4279 | ||
4280 | return mask; | |
4281 | } | |
4282 | ||
4283 | static bool io_arm_poll_handler(struct io_kiocb *req) | |
4284 | { | |
4285 | const struct io_op_def *def = &io_op_defs[req->opcode]; | |
4286 | struct io_ring_ctx *ctx = req->ctx; | |
4287 | struct async_poll *apoll; | |
4288 | struct io_poll_table ipt; | |
4289 | __poll_t mask, ret; | |
4290 | ||
4291 | if (!req->file || !file_can_poll(req->file)) | |
4292 | return false; | |
4293 | if (req->flags & (REQ_F_MUST_PUNT | REQ_F_POLLED)) | |
4294 | return false; | |
4295 | if (!def->pollin && !def->pollout) | |
4296 | return false; | |
4297 | ||
4298 | apoll = kmalloc(sizeof(*apoll), GFP_ATOMIC); | |
4299 | if (unlikely(!apoll)) | |
4300 | return false; | |
4301 | ||
4302 | req->flags |= REQ_F_POLLED; | |
4303 | memcpy(&apoll->work, &req->work, sizeof(req->work)); | |
4304 | ||
3537b6a7 | 4305 | get_task_struct(current); |
d7718a9d JA |
4306 | req->task = current; |
4307 | req->apoll = apoll; | |
4308 | INIT_HLIST_NODE(&req->hash_node); | |
4309 | ||
8755d97a | 4310 | mask = 0; |
d7718a9d | 4311 | if (def->pollin) |
8755d97a | 4312 | mask |= POLLIN | POLLRDNORM; |
d7718a9d JA |
4313 | if (def->pollout) |
4314 | mask |= POLLOUT | POLLWRNORM; | |
4315 | mask |= POLLERR | POLLPRI; | |
4316 | ||
4317 | ipt.pt._qproc = io_async_queue_proc; | |
4318 | ||
4319 | ret = __io_arm_poll_handler(req, &apoll->poll, &ipt, mask, | |
4320 | io_async_wake); | |
4321 | if (ret) { | |
4322 | ipt.error = 0; | |
4323 | apoll->poll.done = true; | |
4324 | spin_unlock_irq(&ctx->completion_lock); | |
4325 | memcpy(&req->work, &apoll->work, sizeof(req->work)); | |
4326 | kfree(apoll); | |
4327 | return false; | |
4328 | } | |
4329 | spin_unlock_irq(&ctx->completion_lock); | |
4330 | trace_io_uring_poll_arm(ctx, req->opcode, req->user_data, mask, | |
4331 | apoll->poll.events); | |
4332 | return true; | |
4333 | } | |
4334 | ||
4335 | static bool __io_poll_remove_one(struct io_kiocb *req, | |
4336 | struct io_poll_iocb *poll) | |
221c5eb2 | 4337 | { |
b41e9852 | 4338 | bool do_complete = false; |
221c5eb2 JA |
4339 | |
4340 | spin_lock(&poll->head->lock); | |
4341 | WRITE_ONCE(poll->canceled, true); | |
392edb45 JA |
4342 | if (!list_empty(&poll->wait.entry)) { |
4343 | list_del_init(&poll->wait.entry); | |
b41e9852 | 4344 | do_complete = true; |
221c5eb2 JA |
4345 | } |
4346 | spin_unlock(&poll->head->lock); | |
d7718a9d JA |
4347 | return do_complete; |
4348 | } | |
4349 | ||
4350 | static bool io_poll_remove_one(struct io_kiocb *req) | |
4351 | { | |
b1f573bd | 4352 | struct async_poll *apoll = NULL; |
d7718a9d JA |
4353 | bool do_complete; |
4354 | ||
4355 | if (req->opcode == IORING_OP_POLL_ADD) { | |
4356 | do_complete = __io_poll_remove_one(req, &req->poll); | |
4357 | } else { | |
b1f573bd | 4358 | apoll = req->apoll; |
d7718a9d JA |
4359 | /* non-poll requests have submit ref still */ |
4360 | do_complete = __io_poll_remove_one(req, &req->apoll->poll); | |
4361 | if (do_complete) | |
4362 | io_put_req(req); | |
4363 | } | |
4364 | ||
78076bb6 | 4365 | hash_del(&req->hash_node); |
d7718a9d | 4366 | |
44575a67 | 4367 | if (do_complete && apoll) { |
b1f573bd XW |
4368 | /* |
4369 | * restore ->work because we need to call io_req_work_drop_env. | |
4370 | */ | |
4371 | memcpy(&req->work, &apoll->work, sizeof(req->work)); | |
4372 | kfree(apoll); | |
4373 | } | |
4374 | ||
b41e9852 JA |
4375 | if (do_complete) { |
4376 | io_cqring_fill_event(req, -ECANCELED); | |
4377 | io_commit_cqring(req->ctx); | |
4378 | req->flags |= REQ_F_COMP_LOCKED; | |
4379 | io_put_req(req); | |
4380 | } | |
4381 | ||
4382 | return do_complete; | |
221c5eb2 JA |
4383 | } |
4384 | ||
4385 | static void io_poll_remove_all(struct io_ring_ctx *ctx) | |
4386 | { | |
78076bb6 | 4387 | struct hlist_node *tmp; |
221c5eb2 | 4388 | struct io_kiocb *req; |
8e2e1faf | 4389 | int posted = 0, i; |
221c5eb2 JA |
4390 | |
4391 | spin_lock_irq(&ctx->completion_lock); | |
78076bb6 JA |
4392 | for (i = 0; i < (1U << ctx->cancel_hash_bits); i++) { |
4393 | struct hlist_head *list; | |
4394 | ||
4395 | list = &ctx->cancel_hash[i]; | |
4396 | hlist_for_each_entry_safe(req, tmp, list, hash_node) | |
8e2e1faf | 4397 | posted += io_poll_remove_one(req); |
221c5eb2 JA |
4398 | } |
4399 | spin_unlock_irq(&ctx->completion_lock); | |
b41e9852 | 4400 | |
8e2e1faf JA |
4401 | if (posted) |
4402 | io_cqring_ev_posted(ctx); | |
221c5eb2 JA |
4403 | } |
4404 | ||
47f46768 JA |
4405 | static int io_poll_cancel(struct io_ring_ctx *ctx, __u64 sqe_addr) |
4406 | { | |
78076bb6 | 4407 | struct hlist_head *list; |
47f46768 JA |
4408 | struct io_kiocb *req; |
4409 | ||
78076bb6 JA |
4410 | list = &ctx->cancel_hash[hash_long(sqe_addr, ctx->cancel_hash_bits)]; |
4411 | hlist_for_each_entry(req, list, hash_node) { | |
b41e9852 JA |
4412 | if (sqe_addr != req->user_data) |
4413 | continue; | |
4414 | if (io_poll_remove_one(req)) | |
eac406c6 | 4415 | return 0; |
b41e9852 | 4416 | return -EALREADY; |
47f46768 JA |
4417 | } |
4418 | ||
4419 | return -ENOENT; | |
4420 | } | |
4421 | ||
3529d8c2 JA |
4422 | static int io_poll_remove_prep(struct io_kiocb *req, |
4423 | const struct io_uring_sqe *sqe) | |
0969e783 | 4424 | { |
0969e783 JA |
4425 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4426 | return -EINVAL; | |
4427 | if (sqe->ioprio || sqe->off || sqe->len || sqe->buf_index || | |
4428 | sqe->poll_events) | |
4429 | return -EINVAL; | |
4430 | ||
4431 | req->poll.addr = READ_ONCE(sqe->addr); | |
0969e783 JA |
4432 | return 0; |
4433 | } | |
4434 | ||
221c5eb2 JA |
4435 | /* |
4436 | * Find a running poll command that matches one specified in sqe->addr, | |
4437 | * and remove it if found. | |
4438 | */ | |
fc4df999 | 4439 | static int io_poll_remove(struct io_kiocb *req) |
221c5eb2 JA |
4440 | { |
4441 | struct io_ring_ctx *ctx = req->ctx; | |
0969e783 | 4442 | u64 addr; |
47f46768 | 4443 | int ret; |
221c5eb2 | 4444 | |
0969e783 | 4445 | addr = req->poll.addr; |
221c5eb2 | 4446 | spin_lock_irq(&ctx->completion_lock); |
0969e783 | 4447 | ret = io_poll_cancel(ctx, addr); |
221c5eb2 JA |
4448 | spin_unlock_irq(&ctx->completion_lock); |
4449 | ||
78e19bbe | 4450 | io_cqring_add_event(req, ret); |
4e88d6e7 JA |
4451 | if (ret < 0) |
4452 | req_set_fail_links(req); | |
e65ef56d | 4453 | io_put_req(req); |
221c5eb2 JA |
4454 | return 0; |
4455 | } | |
4456 | ||
b0dd8a41 | 4457 | static void io_poll_complete(struct io_kiocb *req, __poll_t mask, int error) |
221c5eb2 | 4458 | { |
a197f664 JL |
4459 | struct io_ring_ctx *ctx = req->ctx; |
4460 | ||
8c838788 | 4461 | req->poll.done = true; |
b0a20349 | 4462 | io_cqring_fill_event(req, error ? error : mangle_poll(mask)); |
8c838788 | 4463 | io_commit_cqring(ctx); |
221c5eb2 JA |
4464 | } |
4465 | ||
b41e9852 | 4466 | static void io_poll_task_handler(struct io_kiocb *req, struct io_kiocb **nxt) |
221c5eb2 | 4467 | { |
221c5eb2 | 4468 | struct io_ring_ctx *ctx = req->ctx; |
a6ba632d JA |
4469 | struct io_poll_iocb *poll = &req->poll; |
4470 | ||
74ce6ce4 | 4471 | if (io_poll_rewait(req, poll)) { |
a6ba632d JA |
4472 | spin_unlock_irq(&ctx->completion_lock); |
4473 | return; | |
4474 | } | |
74ce6ce4 | 4475 | |
78076bb6 | 4476 | hash_del(&req->hash_node); |
b41e9852 JA |
4477 | io_poll_complete(req, req->result, 0); |
4478 | req->flags |= REQ_F_COMP_LOCKED; | |
4479 | io_put_req_find_next(req, nxt); | |
221c5eb2 JA |
4480 | spin_unlock_irq(&ctx->completion_lock); |
4481 | ||
8c838788 | 4482 | io_cqring_ev_posted(ctx); |
221c5eb2 JA |
4483 | } |
4484 | ||
b41e9852 | 4485 | static void io_poll_task_func(struct callback_head *cb) |
e94f141b | 4486 | { |
b41e9852 JA |
4487 | struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work); |
4488 | struct io_kiocb *nxt = NULL; | |
e94f141b | 4489 | |
b41e9852 | 4490 | io_poll_task_handler(req, &nxt); |
d7718a9d JA |
4491 | if (nxt) { |
4492 | struct io_ring_ctx *ctx = nxt->ctx; | |
e94f141b | 4493 | |
d7718a9d | 4494 | mutex_lock(&ctx->uring_lock); |
b41e9852 | 4495 | __io_queue_sqe(nxt, NULL); |
d7718a9d | 4496 | mutex_unlock(&ctx->uring_lock); |
e94f141b | 4497 | } |
f0b493e6 JA |
4498 | } |
4499 | ||
221c5eb2 JA |
4500 | static int io_poll_wake(struct wait_queue_entry *wait, unsigned mode, int sync, |
4501 | void *key) | |
4502 | { | |
c2f2eb7d JA |
4503 | struct io_kiocb *req = wait->private; |
4504 | struct io_poll_iocb *poll = &req->poll; | |
221c5eb2 | 4505 | |
d7718a9d | 4506 | return __io_async_wake(req, poll, key_to_poll(key), io_poll_task_func); |
221c5eb2 JA |
4507 | } |
4508 | ||
221c5eb2 JA |
4509 | static void io_poll_queue_proc(struct file *file, struct wait_queue_head *head, |
4510 | struct poll_table_struct *p) | |
4511 | { | |
4512 | struct io_poll_table *pt = container_of(p, struct io_poll_table, pt); | |
4513 | ||
d7718a9d | 4514 | __io_queue_proc(&pt->req->poll, pt, head); |
eac406c6 JA |
4515 | } |
4516 | ||
3529d8c2 | 4517 | static int io_poll_add_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
221c5eb2 JA |
4518 | { |
4519 | struct io_poll_iocb *poll = &req->poll; | |
221c5eb2 | 4520 | u16 events; |
221c5eb2 JA |
4521 | |
4522 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) | |
4523 | return -EINVAL; | |
4524 | if (sqe->addr || sqe->ioprio || sqe->off || sqe->len || sqe->buf_index) | |
4525 | return -EINVAL; | |
09bb8394 JA |
4526 | if (!poll->file) |
4527 | return -EBADF; | |
221c5eb2 | 4528 | |
221c5eb2 JA |
4529 | events = READ_ONCE(sqe->poll_events); |
4530 | poll->events = demangle_poll(events) | EPOLLERR | EPOLLHUP; | |
b41e9852 | 4531 | |
3537b6a7 | 4532 | get_task_struct(current); |
b41e9852 | 4533 | req->task = current; |
0969e783 JA |
4534 | return 0; |
4535 | } | |
4536 | ||
014db007 | 4537 | static int io_poll_add(struct io_kiocb *req) |
0969e783 JA |
4538 | { |
4539 | struct io_poll_iocb *poll = &req->poll; | |
4540 | struct io_ring_ctx *ctx = req->ctx; | |
4541 | struct io_poll_table ipt; | |
0969e783 | 4542 | __poll_t mask; |
0969e783 | 4543 | |
78076bb6 | 4544 | INIT_HLIST_NODE(&req->hash_node); |
36703247 | 4545 | INIT_LIST_HEAD(&req->list); |
d7718a9d | 4546 | ipt.pt._qproc = io_poll_queue_proc; |
36703247 | 4547 | |
d7718a9d JA |
4548 | mask = __io_arm_poll_handler(req, &req->poll, &ipt, poll->events, |
4549 | io_poll_wake); | |
221c5eb2 | 4550 | |
8c838788 | 4551 | if (mask) { /* no async, we'd stolen it */ |
221c5eb2 | 4552 | ipt.error = 0; |
b0dd8a41 | 4553 | io_poll_complete(req, mask, 0); |
221c5eb2 | 4554 | } |
221c5eb2 JA |
4555 | spin_unlock_irq(&ctx->completion_lock); |
4556 | ||
8c838788 JA |
4557 | if (mask) { |
4558 | io_cqring_ev_posted(ctx); | |
014db007 | 4559 | io_put_req(req); |
221c5eb2 | 4560 | } |
8c838788 | 4561 | return ipt.error; |
221c5eb2 JA |
4562 | } |
4563 | ||
5262f567 JA |
4564 | static enum hrtimer_restart io_timeout_fn(struct hrtimer *timer) |
4565 | { | |
ad8a48ac JA |
4566 | struct io_timeout_data *data = container_of(timer, |
4567 | struct io_timeout_data, timer); | |
4568 | struct io_kiocb *req = data->req; | |
4569 | struct io_ring_ctx *ctx = req->ctx; | |
5262f567 JA |
4570 | unsigned long flags; |
4571 | ||
5262f567 JA |
4572 | atomic_inc(&ctx->cq_timeouts); |
4573 | ||
4574 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
ef03681a | 4575 | /* |
11365043 JA |
4576 | * We could be racing with timeout deletion. If the list is empty, |
4577 | * then timeout lookup already found it and will be handling it. | |
ef03681a | 4578 | */ |
842f9612 | 4579 | if (!list_empty(&req->list)) { |
11365043 | 4580 | struct io_kiocb *prev; |
5262f567 | 4581 | |
11365043 JA |
4582 | /* |
4583 | * Adjust the reqs sequence before the current one because it | |
d195a66e | 4584 | * will consume a slot in the cq_ring and the cq_tail |
11365043 JA |
4585 | * pointer will be increased, otherwise other timeout reqs may |
4586 | * return in advance without waiting for enough wait_nr. | |
4587 | */ | |
4588 | prev = req; | |
4589 | list_for_each_entry_continue_reverse(prev, &ctx->timeout_list, list) | |
4590 | prev->sequence++; | |
11365043 | 4591 | list_del_init(&req->list); |
11365043 | 4592 | } |
5262f567 | 4593 | |
78e19bbe | 4594 | io_cqring_fill_event(req, -ETIME); |
5262f567 JA |
4595 | io_commit_cqring(ctx); |
4596 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
4597 | ||
4598 | io_cqring_ev_posted(ctx); | |
4e88d6e7 | 4599 | req_set_fail_links(req); |
5262f567 JA |
4600 | io_put_req(req); |
4601 | return HRTIMER_NORESTART; | |
4602 | } | |
4603 | ||
47f46768 JA |
4604 | static int io_timeout_cancel(struct io_ring_ctx *ctx, __u64 user_data) |
4605 | { | |
4606 | struct io_kiocb *req; | |
4607 | int ret = -ENOENT; | |
4608 | ||
4609 | list_for_each_entry(req, &ctx->timeout_list, list) { | |
4610 | if (user_data == req->user_data) { | |
4611 | list_del_init(&req->list); | |
4612 | ret = 0; | |
4613 | break; | |
4614 | } | |
4615 | } | |
4616 | ||
4617 | if (ret == -ENOENT) | |
4618 | return ret; | |
4619 | ||
2d28390a | 4620 | ret = hrtimer_try_to_cancel(&req->io->timeout.timer); |
47f46768 JA |
4621 | if (ret == -1) |
4622 | return -EALREADY; | |
4623 | ||
4e88d6e7 | 4624 | req_set_fail_links(req); |
47f46768 JA |
4625 | io_cqring_fill_event(req, -ECANCELED); |
4626 | io_put_req(req); | |
4627 | return 0; | |
4628 | } | |
4629 | ||
3529d8c2 JA |
4630 | static int io_timeout_remove_prep(struct io_kiocb *req, |
4631 | const struct io_uring_sqe *sqe) | |
b29472ee | 4632 | { |
b29472ee JA |
4633 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4634 | return -EINVAL; | |
4635 | if (sqe->flags || sqe->ioprio || sqe->buf_index || sqe->len) | |
4636 | return -EINVAL; | |
4637 | ||
4638 | req->timeout.addr = READ_ONCE(sqe->addr); | |
4639 | req->timeout.flags = READ_ONCE(sqe->timeout_flags); | |
4640 | if (req->timeout.flags) | |
4641 | return -EINVAL; | |
4642 | ||
b29472ee JA |
4643 | return 0; |
4644 | } | |
4645 | ||
11365043 JA |
4646 | /* |
4647 | * Remove or update an existing timeout command | |
4648 | */ | |
fc4df999 | 4649 | static int io_timeout_remove(struct io_kiocb *req) |
11365043 JA |
4650 | { |
4651 | struct io_ring_ctx *ctx = req->ctx; | |
47f46768 | 4652 | int ret; |
11365043 | 4653 | |
11365043 | 4654 | spin_lock_irq(&ctx->completion_lock); |
b29472ee | 4655 | ret = io_timeout_cancel(ctx, req->timeout.addr); |
11365043 | 4656 | |
47f46768 | 4657 | io_cqring_fill_event(req, ret); |
11365043 JA |
4658 | io_commit_cqring(ctx); |
4659 | spin_unlock_irq(&ctx->completion_lock); | |
5262f567 | 4660 | io_cqring_ev_posted(ctx); |
4e88d6e7 JA |
4661 | if (ret < 0) |
4662 | req_set_fail_links(req); | |
ec9c02ad | 4663 | io_put_req(req); |
11365043 | 4664 | return 0; |
5262f567 JA |
4665 | } |
4666 | ||
3529d8c2 | 4667 | static int io_timeout_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe, |
2d28390a | 4668 | bool is_timeout_link) |
5262f567 | 4669 | { |
ad8a48ac | 4670 | struct io_timeout_data *data; |
a41525ab | 4671 | unsigned flags; |
5262f567 | 4672 | |
ad8a48ac | 4673 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
5262f567 | 4674 | return -EINVAL; |
ad8a48ac | 4675 | if (sqe->ioprio || sqe->buf_index || sqe->len != 1) |
a41525ab | 4676 | return -EINVAL; |
2d28390a JA |
4677 | if (sqe->off && is_timeout_link) |
4678 | return -EINVAL; | |
a41525ab JA |
4679 | flags = READ_ONCE(sqe->timeout_flags); |
4680 | if (flags & ~IORING_TIMEOUT_ABS) | |
5262f567 | 4681 | return -EINVAL; |
bdf20073 | 4682 | |
26a61679 JA |
4683 | req->timeout.count = READ_ONCE(sqe->off); |
4684 | ||
3529d8c2 | 4685 | if (!req->io && io_alloc_async_ctx(req)) |
26a61679 JA |
4686 | return -ENOMEM; |
4687 | ||
4688 | data = &req->io->timeout; | |
ad8a48ac | 4689 | data->req = req; |
ad8a48ac JA |
4690 | req->flags |= REQ_F_TIMEOUT; |
4691 | ||
4692 | if (get_timespec64(&data->ts, u64_to_user_ptr(sqe->addr))) | |
5262f567 JA |
4693 | return -EFAULT; |
4694 | ||
11365043 | 4695 | if (flags & IORING_TIMEOUT_ABS) |
ad8a48ac | 4696 | data->mode = HRTIMER_MODE_ABS; |
11365043 | 4697 | else |
ad8a48ac | 4698 | data->mode = HRTIMER_MODE_REL; |
11365043 | 4699 | |
ad8a48ac JA |
4700 | hrtimer_init(&data->timer, CLOCK_MONOTONIC, data->mode); |
4701 | return 0; | |
4702 | } | |
4703 | ||
fc4df999 | 4704 | static int io_timeout(struct io_kiocb *req) |
ad8a48ac | 4705 | { |
ad8a48ac JA |
4706 | struct io_ring_ctx *ctx = req->ctx; |
4707 | struct io_timeout_data *data; | |
4708 | struct list_head *entry; | |
4709 | unsigned span = 0; | |
b55ce732 | 4710 | u32 count = req->timeout.count; |
22cad158 | 4711 | u32 seq = req->sequence; |
ad8a48ac | 4712 | |
2d28390a | 4713 | data = &req->io->timeout; |
93bd25bb | 4714 | |
5262f567 JA |
4715 | /* |
4716 | * sqe->off holds how many events that need to occur for this | |
93bd25bb JA |
4717 | * timeout event to be satisfied. If it isn't set, then this is |
4718 | * a pure timeout request, sequence isn't used. | |
5262f567 | 4719 | */ |
93bd25bb JA |
4720 | if (!count) { |
4721 | req->flags |= REQ_F_TIMEOUT_NOSEQ; | |
4722 | spin_lock_irq(&ctx->completion_lock); | |
4723 | entry = ctx->timeout_list.prev; | |
4724 | goto add; | |
4725 | } | |
5262f567 | 4726 | |
22cad158 | 4727 | req->sequence = seq + count; |
5262f567 JA |
4728 | |
4729 | /* | |
4730 | * Insertion sort, ensuring the first entry in the list is always | |
4731 | * the one we need first. | |
4732 | */ | |
5262f567 JA |
4733 | spin_lock_irq(&ctx->completion_lock); |
4734 | list_for_each_prev(entry, &ctx->timeout_list) { | |
4735 | struct io_kiocb *nxt = list_entry(entry, struct io_kiocb, list); | |
22cad158 | 4736 | unsigned nxt_seq; |
5da0fb1a | 4737 | long long tmp, tmp_nxt; |
b55ce732 | 4738 | u32 nxt_offset = nxt->timeout.count; |
5262f567 | 4739 | |
93bd25bb JA |
4740 | if (nxt->flags & REQ_F_TIMEOUT_NOSEQ) |
4741 | continue; | |
4742 | ||
5da0fb1a | 4743 | /* |
22cad158 | 4744 | * Since seq + count can overflow, use type long |
5da0fb1a | 4745 | * long to store it. |
4746 | */ | |
22cad158 PB |
4747 | tmp = (long long)seq + count; |
4748 | nxt_seq = nxt->sequence - nxt_offset; | |
4749 | tmp_nxt = (long long)nxt_seq + nxt_offset; | |
5da0fb1a | 4750 | |
4751 | /* | |
4752 | * cached_sq_head may overflow, and it will never overflow twice | |
4753 | * once there is some timeout req still be valid. | |
4754 | */ | |
22cad158 | 4755 | if (seq < nxt_seq) |
8b07a65a | 4756 | tmp += UINT_MAX; |
5da0fb1a | 4757 | |
a1f58ba4 | 4758 | if (tmp > tmp_nxt) |
5262f567 | 4759 | break; |
a1f58ba4 | 4760 | |
4761 | /* | |
4762 | * Sequence of reqs after the insert one and itself should | |
4763 | * be adjusted because each timeout req consumes a slot. | |
4764 | */ | |
4765 | span++; | |
4766 | nxt->sequence++; | |
5262f567 | 4767 | } |
a1f58ba4 | 4768 | req->sequence -= span; |
93bd25bb | 4769 | add: |
5262f567 | 4770 | list_add(&req->list, entry); |
ad8a48ac JA |
4771 | data->timer.function = io_timeout_fn; |
4772 | hrtimer_start(&data->timer, timespec64_to_ktime(data->ts), data->mode); | |
5262f567 | 4773 | spin_unlock_irq(&ctx->completion_lock); |
5262f567 JA |
4774 | return 0; |
4775 | } | |
5262f567 | 4776 | |
62755e35 JA |
4777 | static bool io_cancel_cb(struct io_wq_work *work, void *data) |
4778 | { | |
4779 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); | |
4780 | ||
4781 | return req->user_data == (unsigned long) data; | |
4782 | } | |
4783 | ||
e977d6d3 | 4784 | static int io_async_cancel_one(struct io_ring_ctx *ctx, void *sqe_addr) |
62755e35 | 4785 | { |
62755e35 | 4786 | enum io_wq_cancel cancel_ret; |
62755e35 JA |
4787 | int ret = 0; |
4788 | ||
62755e35 JA |
4789 | cancel_ret = io_wq_cancel_cb(ctx->io_wq, io_cancel_cb, sqe_addr); |
4790 | switch (cancel_ret) { | |
4791 | case IO_WQ_CANCEL_OK: | |
4792 | ret = 0; | |
4793 | break; | |
4794 | case IO_WQ_CANCEL_RUNNING: | |
4795 | ret = -EALREADY; | |
4796 | break; | |
4797 | case IO_WQ_CANCEL_NOTFOUND: | |
4798 | ret = -ENOENT; | |
4799 | break; | |
4800 | } | |
4801 | ||
e977d6d3 JA |
4802 | return ret; |
4803 | } | |
4804 | ||
47f46768 JA |
4805 | static void io_async_find_and_cancel(struct io_ring_ctx *ctx, |
4806 | struct io_kiocb *req, __u64 sqe_addr, | |
014db007 | 4807 | int success_ret) |
47f46768 JA |
4808 | { |
4809 | unsigned long flags; | |
4810 | int ret; | |
4811 | ||
4812 | ret = io_async_cancel_one(ctx, (void *) (unsigned long) sqe_addr); | |
4813 | if (ret != -ENOENT) { | |
4814 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
4815 | goto done; | |
4816 | } | |
4817 | ||
4818 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
4819 | ret = io_timeout_cancel(ctx, sqe_addr); | |
4820 | if (ret != -ENOENT) | |
4821 | goto done; | |
4822 | ret = io_poll_cancel(ctx, sqe_addr); | |
4823 | done: | |
b0dd8a41 JA |
4824 | if (!ret) |
4825 | ret = success_ret; | |
47f46768 JA |
4826 | io_cqring_fill_event(req, ret); |
4827 | io_commit_cqring(ctx); | |
4828 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
4829 | io_cqring_ev_posted(ctx); | |
4830 | ||
4e88d6e7 JA |
4831 | if (ret < 0) |
4832 | req_set_fail_links(req); | |
014db007 | 4833 | io_put_req(req); |
47f46768 JA |
4834 | } |
4835 | ||
3529d8c2 JA |
4836 | static int io_async_cancel_prep(struct io_kiocb *req, |
4837 | const struct io_uring_sqe *sqe) | |
e977d6d3 | 4838 | { |
fbf23849 | 4839 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
e977d6d3 JA |
4840 | return -EINVAL; |
4841 | if (sqe->flags || sqe->ioprio || sqe->off || sqe->len || | |
4842 | sqe->cancel_flags) | |
4843 | return -EINVAL; | |
4844 | ||
fbf23849 JA |
4845 | req->cancel.addr = READ_ONCE(sqe->addr); |
4846 | return 0; | |
4847 | } | |
4848 | ||
014db007 | 4849 | static int io_async_cancel(struct io_kiocb *req) |
fbf23849 JA |
4850 | { |
4851 | struct io_ring_ctx *ctx = req->ctx; | |
fbf23849 | 4852 | |
014db007 | 4853 | io_async_find_and_cancel(ctx, req, req->cancel.addr, 0); |
5262f567 JA |
4854 | return 0; |
4855 | } | |
4856 | ||
05f3fb3c JA |
4857 | static int io_files_update_prep(struct io_kiocb *req, |
4858 | const struct io_uring_sqe *sqe) | |
4859 | { | |
4860 | if (sqe->flags || sqe->ioprio || sqe->rw_flags) | |
4861 | return -EINVAL; | |
4862 | ||
4863 | req->files_update.offset = READ_ONCE(sqe->off); | |
4864 | req->files_update.nr_args = READ_ONCE(sqe->len); | |
4865 | if (!req->files_update.nr_args) | |
4866 | return -EINVAL; | |
4867 | req->files_update.arg = READ_ONCE(sqe->addr); | |
4868 | return 0; | |
4869 | } | |
4870 | ||
4871 | static int io_files_update(struct io_kiocb *req, bool force_nonblock) | |
fbf23849 JA |
4872 | { |
4873 | struct io_ring_ctx *ctx = req->ctx; | |
05f3fb3c JA |
4874 | struct io_uring_files_update up; |
4875 | int ret; | |
fbf23849 | 4876 | |
f86cd20c | 4877 | if (force_nonblock) |
05f3fb3c | 4878 | return -EAGAIN; |
05f3fb3c JA |
4879 | |
4880 | up.offset = req->files_update.offset; | |
4881 | up.fds = req->files_update.arg; | |
4882 | ||
4883 | mutex_lock(&ctx->uring_lock); | |
4884 | ret = __io_sqe_files_update(ctx, &up, req->files_update.nr_args); | |
4885 | mutex_unlock(&ctx->uring_lock); | |
4886 | ||
4887 | if (ret < 0) | |
4888 | req_set_fail_links(req); | |
4889 | io_cqring_add_event(req, ret); | |
4890 | io_put_req(req); | |
5262f567 JA |
4891 | return 0; |
4892 | } | |
4893 | ||
3529d8c2 JA |
4894 | static int io_req_defer_prep(struct io_kiocb *req, |
4895 | const struct io_uring_sqe *sqe) | |
f67676d1 | 4896 | { |
e781573e | 4897 | ssize_t ret = 0; |
f67676d1 | 4898 | |
f1d96a8f PB |
4899 | if (!sqe) |
4900 | return 0; | |
4901 | ||
f86cd20c JA |
4902 | if (io_op_defs[req->opcode].file_table) { |
4903 | ret = io_grab_files(req); | |
4904 | if (unlikely(ret)) | |
4905 | return ret; | |
4906 | } | |
4907 | ||
cccf0ee8 JA |
4908 | io_req_work_grab_env(req, &io_op_defs[req->opcode]); |
4909 | ||
d625c6ee | 4910 | switch (req->opcode) { |
e781573e JA |
4911 | case IORING_OP_NOP: |
4912 | break; | |
f67676d1 JA |
4913 | case IORING_OP_READV: |
4914 | case IORING_OP_READ_FIXED: | |
3a6820f2 | 4915 | case IORING_OP_READ: |
3529d8c2 | 4916 | ret = io_read_prep(req, sqe, true); |
f67676d1 JA |
4917 | break; |
4918 | case IORING_OP_WRITEV: | |
4919 | case IORING_OP_WRITE_FIXED: | |
3a6820f2 | 4920 | case IORING_OP_WRITE: |
3529d8c2 | 4921 | ret = io_write_prep(req, sqe, true); |
f67676d1 | 4922 | break; |
0969e783 | 4923 | case IORING_OP_POLL_ADD: |
3529d8c2 | 4924 | ret = io_poll_add_prep(req, sqe); |
0969e783 JA |
4925 | break; |
4926 | case IORING_OP_POLL_REMOVE: | |
3529d8c2 | 4927 | ret = io_poll_remove_prep(req, sqe); |
0969e783 | 4928 | break; |
8ed8d3c3 | 4929 | case IORING_OP_FSYNC: |
3529d8c2 | 4930 | ret = io_prep_fsync(req, sqe); |
8ed8d3c3 JA |
4931 | break; |
4932 | case IORING_OP_SYNC_FILE_RANGE: | |
3529d8c2 | 4933 | ret = io_prep_sfr(req, sqe); |
8ed8d3c3 | 4934 | break; |
03b1230c | 4935 | case IORING_OP_SENDMSG: |
fddaface | 4936 | case IORING_OP_SEND: |
3529d8c2 | 4937 | ret = io_sendmsg_prep(req, sqe); |
03b1230c JA |
4938 | break; |
4939 | case IORING_OP_RECVMSG: | |
fddaface | 4940 | case IORING_OP_RECV: |
3529d8c2 | 4941 | ret = io_recvmsg_prep(req, sqe); |
03b1230c | 4942 | break; |
f499a021 | 4943 | case IORING_OP_CONNECT: |
3529d8c2 | 4944 | ret = io_connect_prep(req, sqe); |
f499a021 | 4945 | break; |
2d28390a | 4946 | case IORING_OP_TIMEOUT: |
3529d8c2 | 4947 | ret = io_timeout_prep(req, sqe, false); |
b7bb4f7d | 4948 | break; |
b29472ee | 4949 | case IORING_OP_TIMEOUT_REMOVE: |
3529d8c2 | 4950 | ret = io_timeout_remove_prep(req, sqe); |
b29472ee | 4951 | break; |
fbf23849 | 4952 | case IORING_OP_ASYNC_CANCEL: |
3529d8c2 | 4953 | ret = io_async_cancel_prep(req, sqe); |
fbf23849 | 4954 | break; |
2d28390a | 4955 | case IORING_OP_LINK_TIMEOUT: |
3529d8c2 | 4956 | ret = io_timeout_prep(req, sqe, true); |
b7bb4f7d | 4957 | break; |
8ed8d3c3 | 4958 | case IORING_OP_ACCEPT: |
3529d8c2 | 4959 | ret = io_accept_prep(req, sqe); |
8ed8d3c3 | 4960 | break; |
d63d1b5e JA |
4961 | case IORING_OP_FALLOCATE: |
4962 | ret = io_fallocate_prep(req, sqe); | |
4963 | break; | |
15b71abe JA |
4964 | case IORING_OP_OPENAT: |
4965 | ret = io_openat_prep(req, sqe); | |
4966 | break; | |
b5dba59e JA |
4967 | case IORING_OP_CLOSE: |
4968 | ret = io_close_prep(req, sqe); | |
4969 | break; | |
05f3fb3c JA |
4970 | case IORING_OP_FILES_UPDATE: |
4971 | ret = io_files_update_prep(req, sqe); | |
4972 | break; | |
eddc7ef5 JA |
4973 | case IORING_OP_STATX: |
4974 | ret = io_statx_prep(req, sqe); | |
4975 | break; | |
4840e418 JA |
4976 | case IORING_OP_FADVISE: |
4977 | ret = io_fadvise_prep(req, sqe); | |
4978 | break; | |
c1ca757b JA |
4979 | case IORING_OP_MADVISE: |
4980 | ret = io_madvise_prep(req, sqe); | |
4981 | break; | |
cebdb986 JA |
4982 | case IORING_OP_OPENAT2: |
4983 | ret = io_openat2_prep(req, sqe); | |
4984 | break; | |
3e4827b0 JA |
4985 | case IORING_OP_EPOLL_CTL: |
4986 | ret = io_epoll_ctl_prep(req, sqe); | |
4987 | break; | |
7d67af2c PB |
4988 | case IORING_OP_SPLICE: |
4989 | ret = io_splice_prep(req, sqe); | |
4990 | break; | |
ddf0322d JA |
4991 | case IORING_OP_PROVIDE_BUFFERS: |
4992 | ret = io_provide_buffers_prep(req, sqe); | |
4993 | break; | |
067524e9 JA |
4994 | case IORING_OP_REMOVE_BUFFERS: |
4995 | ret = io_remove_buffers_prep(req, sqe); | |
4996 | break; | |
f67676d1 | 4997 | default: |
e781573e JA |
4998 | printk_once(KERN_WARNING "io_uring: unhandled opcode %d\n", |
4999 | req->opcode); | |
5000 | ret = -EINVAL; | |
b7bb4f7d | 5001 | break; |
f67676d1 JA |
5002 | } |
5003 | ||
b7bb4f7d | 5004 | return ret; |
f67676d1 JA |
5005 | } |
5006 | ||
3529d8c2 | 5007 | static int io_req_defer(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
de0617e4 | 5008 | { |
a197f664 | 5009 | struct io_ring_ctx *ctx = req->ctx; |
f67676d1 | 5010 | int ret; |
de0617e4 | 5011 | |
9d858b21 | 5012 | /* Still need defer if there is pending req in defer list. */ |
4ee36314 | 5013 | if (!req_need_defer(req) && list_empty_careful(&ctx->defer_list)) |
de0617e4 JA |
5014 | return 0; |
5015 | ||
650b5481 PB |
5016 | if (!req->io) { |
5017 | if (io_alloc_async_ctx(req)) | |
5018 | return -EAGAIN; | |
5019 | ret = io_req_defer_prep(req, sqe); | |
5020 | if (ret < 0) | |
5021 | return ret; | |
5022 | } | |
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 | ||
b532576e JA |
5309 | /* If the op doesn't have a file, we're not polling for it */ |
5310 | if ((ctx->flags & IORING_SETUP_IOPOLL) && req->file) { | |
11ba820b JA |
5311 | const bool in_async = io_wq_current_is_worker(); |
5312 | ||
9e645e11 | 5313 | if (req->result == -EAGAIN) |
def596e9 JA |
5314 | return -EAGAIN; |
5315 | ||
11ba820b JA |
5316 | /* workqueue context doesn't hold uring_lock, grab it now */ |
5317 | if (in_async) | |
5318 | mutex_lock(&ctx->uring_lock); | |
5319 | ||
def596e9 | 5320 | io_iopoll_req_issued(req); |
11ba820b JA |
5321 | |
5322 | if (in_async) | |
5323 | mutex_unlock(&ctx->uring_lock); | |
def596e9 JA |
5324 | } |
5325 | ||
5326 | return 0; | |
2b188cc1 JA |
5327 | } |
5328 | ||
561fb04a | 5329 | static void io_wq_submit_work(struct io_wq_work **workptr) |
2b188cc1 | 5330 | { |
561fb04a | 5331 | struct io_wq_work *work = *workptr; |
2b188cc1 | 5332 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); |
561fb04a | 5333 | int ret = 0; |
2b188cc1 | 5334 | |
0c9d5ccd JA |
5335 | /* if NO_CANCEL is set, we must still run the work */ |
5336 | if ((work->flags & (IO_WQ_WORK_CANCEL|IO_WQ_WORK_NO_CANCEL)) == | |
5337 | IO_WQ_WORK_CANCEL) { | |
561fb04a | 5338 | ret = -ECANCELED; |
0c9d5ccd | 5339 | } |
31b51510 | 5340 | |
561fb04a | 5341 | if (!ret) { |
561fb04a | 5342 | do { |
014db007 | 5343 | ret = io_issue_sqe(req, NULL, false); |
561fb04a JA |
5344 | /* |
5345 | * We can get EAGAIN for polled IO even though we're | |
5346 | * forcing a sync submission from here, since we can't | |
5347 | * wait for request slots on the block side. | |
5348 | */ | |
5349 | if (ret != -EAGAIN) | |
5350 | break; | |
5351 | cond_resched(); | |
5352 | } while (1); | |
5353 | } | |
31b51510 | 5354 | |
561fb04a | 5355 | if (ret) { |
4e88d6e7 | 5356 | req_set_fail_links(req); |
78e19bbe | 5357 | io_cqring_add_event(req, ret); |
817869d2 | 5358 | io_put_req(req); |
edafccee | 5359 | } |
2b188cc1 | 5360 | |
e9fd9396 | 5361 | io_steal_work(req, workptr); |
2b188cc1 JA |
5362 | } |
5363 | ||
65e19f54 JA |
5364 | static inline struct file *io_file_from_index(struct io_ring_ctx *ctx, |
5365 | int index) | |
5366 | { | |
5367 | struct fixed_file_table *table; | |
5368 | ||
05f3fb3c JA |
5369 | table = &ctx->file_data->table[index >> IORING_FILE_TABLE_SHIFT]; |
5370 | return table->files[index & IORING_FILE_TABLE_MASK];; | |
65e19f54 JA |
5371 | } |
5372 | ||
8da11c19 PB |
5373 | static int io_file_get(struct io_submit_state *state, struct io_kiocb *req, |
5374 | int fd, struct file **out_file, bool fixed) | |
09bb8394 | 5375 | { |
a197f664 | 5376 | struct io_ring_ctx *ctx = req->ctx; |
8da11c19 | 5377 | struct file *file; |
09bb8394 | 5378 | |
8da11c19 | 5379 | if (fixed) { |
05f3fb3c | 5380 | if (unlikely(!ctx->file_data || |
09bb8394 JA |
5381 | (unsigned) fd >= ctx->nr_user_files)) |
5382 | return -EBADF; | |
b7620121 | 5383 | fd = array_index_nospec(fd, ctx->nr_user_files); |
8da11c19 PB |
5384 | file = io_file_from_index(ctx, fd); |
5385 | if (!file) | |
08a45173 | 5386 | return -EBADF; |
05589553 XW |
5387 | req->fixed_file_refs = ctx->file_data->cur_refs; |
5388 | percpu_ref_get(req->fixed_file_refs); | |
09bb8394 | 5389 | } else { |
c826bd7a | 5390 | trace_io_uring_file_get(ctx, fd); |
8da11c19 PB |
5391 | file = __io_file_get(state, fd); |
5392 | if (unlikely(!file)) | |
09bb8394 JA |
5393 | return -EBADF; |
5394 | } | |
5395 | ||
8da11c19 | 5396 | *out_file = file; |
09bb8394 JA |
5397 | return 0; |
5398 | } | |
5399 | ||
8da11c19 | 5400 | static int io_req_set_file(struct io_submit_state *state, struct io_kiocb *req, |
63ff8223 | 5401 | int fd) |
8da11c19 | 5402 | { |
8da11c19 PB |
5403 | bool fixed; |
5404 | ||
63ff8223 | 5405 | fixed = (req->flags & REQ_F_FIXED_FILE) != 0; |
8da11c19 PB |
5406 | if (unlikely(!fixed && req->needs_fixed_file)) |
5407 | return -EBADF; | |
5408 | ||
5409 | return io_file_get(state, req, fd, &req->file, fixed); | |
5410 | } | |
5411 | ||
a197f664 | 5412 | static int io_grab_files(struct io_kiocb *req) |
fcb323cc JA |
5413 | { |
5414 | int ret = -EBADF; | |
a197f664 | 5415 | struct io_ring_ctx *ctx = req->ctx; |
fcb323cc | 5416 | |
5b0bbee4 | 5417 | if (req->work.files || (req->flags & REQ_F_NO_FILE_TABLE)) |
f86cd20c | 5418 | return 0; |
b14cca0c | 5419 | if (!ctx->ring_file) |
b5dba59e JA |
5420 | return -EBADF; |
5421 | ||
fcb323cc JA |
5422 | rcu_read_lock(); |
5423 | spin_lock_irq(&ctx->inflight_lock); | |
5424 | /* | |
5425 | * We use the f_ops->flush() handler to ensure that we can flush | |
5426 | * out work accessing these files if the fd is closed. Check if | |
5427 | * the fd has changed since we started down this path, and disallow | |
5428 | * this operation if it has. | |
5429 | */ | |
b14cca0c | 5430 | if (fcheck(ctx->ring_fd) == ctx->ring_file) { |
fcb323cc JA |
5431 | list_add(&req->inflight_entry, &ctx->inflight_list); |
5432 | req->flags |= REQ_F_INFLIGHT; | |
5433 | req->work.files = current->files; | |
5434 | ret = 0; | |
5435 | } | |
5436 | spin_unlock_irq(&ctx->inflight_lock); | |
5437 | rcu_read_unlock(); | |
5438 | ||
5439 | return ret; | |
5440 | } | |
5441 | ||
2665abfd | 5442 | static enum hrtimer_restart io_link_timeout_fn(struct hrtimer *timer) |
2b188cc1 | 5443 | { |
ad8a48ac JA |
5444 | struct io_timeout_data *data = container_of(timer, |
5445 | struct io_timeout_data, timer); | |
5446 | struct io_kiocb *req = data->req; | |
2665abfd JA |
5447 | struct io_ring_ctx *ctx = req->ctx; |
5448 | struct io_kiocb *prev = NULL; | |
5449 | unsigned long flags; | |
2665abfd JA |
5450 | |
5451 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
5452 | ||
5453 | /* | |
5454 | * We don't expect the list to be empty, that will only happen if we | |
5455 | * race with the completion of the linked work. | |
5456 | */ | |
4493233e PB |
5457 | if (!list_empty(&req->link_list)) { |
5458 | prev = list_entry(req->link_list.prev, struct io_kiocb, | |
5459 | link_list); | |
5d960724 | 5460 | if (refcount_inc_not_zero(&prev->refs)) { |
4493233e | 5461 | list_del_init(&req->link_list); |
5d960724 JA |
5462 | prev->flags &= ~REQ_F_LINK_TIMEOUT; |
5463 | } else | |
76a46e06 | 5464 | prev = NULL; |
2665abfd JA |
5465 | } |
5466 | ||
5467 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
5468 | ||
5469 | if (prev) { | |
4e88d6e7 | 5470 | req_set_fail_links(prev); |
014db007 | 5471 | io_async_find_and_cancel(ctx, req, prev->user_data, -ETIME); |
76a46e06 | 5472 | io_put_req(prev); |
47f46768 JA |
5473 | } else { |
5474 | io_cqring_add_event(req, -ETIME); | |
5475 | io_put_req(req); | |
2665abfd | 5476 | } |
2665abfd JA |
5477 | return HRTIMER_NORESTART; |
5478 | } | |
5479 | ||
ad8a48ac | 5480 | static void io_queue_linked_timeout(struct io_kiocb *req) |
2665abfd | 5481 | { |
76a46e06 | 5482 | struct io_ring_ctx *ctx = req->ctx; |
2665abfd | 5483 | |
76a46e06 JA |
5484 | /* |
5485 | * If the list is now empty, then our linked request finished before | |
5486 | * we got a chance to setup the timer | |
5487 | */ | |
5488 | spin_lock_irq(&ctx->completion_lock); | |
4493233e | 5489 | if (!list_empty(&req->link_list)) { |
2d28390a | 5490 | struct io_timeout_data *data = &req->io->timeout; |
94ae5e77 | 5491 | |
ad8a48ac JA |
5492 | data->timer.function = io_link_timeout_fn; |
5493 | hrtimer_start(&data->timer, timespec64_to_ktime(data->ts), | |
5494 | data->mode); | |
2665abfd | 5495 | } |
76a46e06 | 5496 | spin_unlock_irq(&ctx->completion_lock); |
2665abfd | 5497 | |
2665abfd | 5498 | /* drop submission reference */ |
76a46e06 JA |
5499 | io_put_req(req); |
5500 | } | |
2665abfd | 5501 | |
ad8a48ac | 5502 | static struct io_kiocb *io_prep_linked_timeout(struct io_kiocb *req) |
2665abfd JA |
5503 | { |
5504 | struct io_kiocb *nxt; | |
5505 | ||
dea3b49c | 5506 | if (!(req->flags & REQ_F_LINK_HEAD)) |
2665abfd | 5507 | return NULL; |
d7718a9d JA |
5508 | /* for polled retry, if flag is set, we already went through here */ |
5509 | if (req->flags & REQ_F_POLLED) | |
5510 | return NULL; | |
2665abfd | 5511 | |
4493233e PB |
5512 | nxt = list_first_entry_or_null(&req->link_list, struct io_kiocb, |
5513 | link_list); | |
d625c6ee | 5514 | if (!nxt || nxt->opcode != IORING_OP_LINK_TIMEOUT) |
76a46e06 | 5515 | return NULL; |
2665abfd | 5516 | |
76a46e06 | 5517 | req->flags |= REQ_F_LINK_TIMEOUT; |
76a46e06 | 5518 | return nxt; |
2665abfd JA |
5519 | } |
5520 | ||
3529d8c2 | 5521 | static void __io_queue_sqe(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
2b188cc1 | 5522 | { |
4a0a7a18 | 5523 | struct io_kiocb *linked_timeout; |
4bc4494e | 5524 | struct io_kiocb *nxt; |
193155c8 | 5525 | const struct cred *old_creds = NULL; |
e0c5c576 | 5526 | int ret; |
2b188cc1 | 5527 | |
4a0a7a18 JA |
5528 | again: |
5529 | linked_timeout = io_prep_linked_timeout(req); | |
5530 | ||
193155c8 JA |
5531 | if (req->work.creds && req->work.creds != current_cred()) { |
5532 | if (old_creds) | |
5533 | revert_creds(old_creds); | |
5534 | if (old_creds == req->work.creds) | |
5535 | old_creds = NULL; /* restored original creds */ | |
5536 | else | |
5537 | old_creds = override_creds(req->work.creds); | |
5538 | } | |
5539 | ||
014db007 | 5540 | ret = io_issue_sqe(req, sqe, true); |
491381ce JA |
5541 | |
5542 | /* | |
5543 | * We async punt it if the file wasn't marked NOWAIT, or if the file | |
5544 | * doesn't support non-blocking read/write attempts | |
5545 | */ | |
5546 | if (ret == -EAGAIN && (!(req->flags & REQ_F_NOWAIT) || | |
5547 | (req->flags & REQ_F_MUST_PUNT))) { | |
d7718a9d JA |
5548 | if (io_arm_poll_handler(req)) { |
5549 | if (linked_timeout) | |
5550 | io_queue_linked_timeout(linked_timeout); | |
4bc4494e | 5551 | goto exit; |
d7718a9d | 5552 | } |
86a761f8 | 5553 | punt: |
f86cd20c | 5554 | if (io_op_defs[req->opcode].file_table) { |
bbad27b2 PB |
5555 | ret = io_grab_files(req); |
5556 | if (ret) | |
5557 | goto err; | |
2b188cc1 | 5558 | } |
bbad27b2 PB |
5559 | |
5560 | /* | |
5561 | * Queued up for async execution, worker will release | |
5562 | * submit reference when the iocb is actually submitted. | |
5563 | */ | |
5564 | io_queue_async_work(req); | |
4bc4494e | 5565 | goto exit; |
2b188cc1 | 5566 | } |
e65ef56d | 5567 | |
fcb323cc | 5568 | err: |
4bc4494e | 5569 | nxt = NULL; |
76a46e06 | 5570 | /* drop submission reference */ |
2a44f467 | 5571 | io_put_req_find_next(req, &nxt); |
e65ef56d | 5572 | |
f9bd67f6 | 5573 | if (linked_timeout) { |
76a46e06 | 5574 | if (!ret) |
f9bd67f6 | 5575 | io_queue_linked_timeout(linked_timeout); |
76a46e06 | 5576 | else |
f9bd67f6 | 5577 | io_put_req(linked_timeout); |
76a46e06 JA |
5578 | } |
5579 | ||
e65ef56d | 5580 | /* and drop final reference, if we failed */ |
9e645e11 | 5581 | if (ret) { |
78e19bbe | 5582 | io_cqring_add_event(req, ret); |
4e88d6e7 | 5583 | req_set_fail_links(req); |
e65ef56d | 5584 | io_put_req(req); |
9e645e11 | 5585 | } |
4a0a7a18 JA |
5586 | if (nxt) { |
5587 | req = nxt; | |
86a761f8 PB |
5588 | |
5589 | if (req->flags & REQ_F_FORCE_ASYNC) | |
5590 | goto punt; | |
4a0a7a18 JA |
5591 | goto again; |
5592 | } | |
4bc4494e | 5593 | exit: |
193155c8 JA |
5594 | if (old_creds) |
5595 | revert_creds(old_creds); | |
2b188cc1 JA |
5596 | } |
5597 | ||
3529d8c2 | 5598 | static void io_queue_sqe(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
4fe2c963 JL |
5599 | { |
5600 | int ret; | |
5601 | ||
3529d8c2 | 5602 | ret = io_req_defer(req, sqe); |
4fe2c963 JL |
5603 | if (ret) { |
5604 | if (ret != -EIOCBQUEUED) { | |
1118591a | 5605 | fail_req: |
78e19bbe | 5606 | io_cqring_add_event(req, ret); |
4e88d6e7 | 5607 | req_set_fail_links(req); |
78e19bbe | 5608 | io_double_put_req(req); |
4fe2c963 | 5609 | } |
2550878f | 5610 | } else if (req->flags & REQ_F_FORCE_ASYNC) { |
bd2ab18a PB |
5611 | if (!req->io) { |
5612 | ret = -EAGAIN; | |
5613 | if (io_alloc_async_ctx(req)) | |
5614 | goto fail_req; | |
5615 | ret = io_req_defer_prep(req, sqe); | |
5616 | if (unlikely(ret < 0)) | |
5617 | goto fail_req; | |
5618 | } | |
5619 | ||
ce35a47a JA |
5620 | /* |
5621 | * Never try inline submit of IOSQE_ASYNC is set, go straight | |
5622 | * to async execution. | |
5623 | */ | |
5624 | req->work.flags |= IO_WQ_WORK_CONCURRENT; | |
5625 | io_queue_async_work(req); | |
5626 | } else { | |
3529d8c2 | 5627 | __io_queue_sqe(req, sqe); |
ce35a47a | 5628 | } |
4fe2c963 JL |
5629 | } |
5630 | ||
1b4a51b6 | 5631 | static inline void io_queue_link_head(struct io_kiocb *req) |
4fe2c963 | 5632 | { |
94ae5e77 | 5633 | if (unlikely(req->flags & REQ_F_FAIL_LINK)) { |
1b4a51b6 PB |
5634 | io_cqring_add_event(req, -ECANCELED); |
5635 | io_double_put_req(req); | |
5636 | } else | |
3529d8c2 | 5637 | io_queue_sqe(req, NULL); |
4fe2c963 JL |
5638 | } |
5639 | ||
1d4240cc | 5640 | static int io_submit_sqe(struct io_kiocb *req, const struct io_uring_sqe *sqe, |
3529d8c2 | 5641 | struct io_submit_state *state, struct io_kiocb **link) |
9e645e11 | 5642 | { |
a197f664 | 5643 | struct io_ring_ctx *ctx = req->ctx; |
ef4ff581 | 5644 | int ret; |
9e645e11 | 5645 | |
9e645e11 JA |
5646 | /* |
5647 | * If we already have a head request, queue this one for async | |
5648 | * submittal once the head completes. If we don't have a head but | |
5649 | * IOSQE_IO_LINK is set in the sqe, start a new head. This one will be | |
5650 | * submitted sync once the chain is complete. If none of those | |
5651 | * conditions are true (normal request), then just queue it. | |
5652 | */ | |
5653 | if (*link) { | |
9d76377f | 5654 | struct io_kiocb *head = *link; |
4e88d6e7 | 5655 | |
8cdf2193 PB |
5656 | /* |
5657 | * Taking sequential execution of a link, draining both sides | |
5658 | * of the link also fullfils IOSQE_IO_DRAIN semantics for all | |
5659 | * requests in the link. So, it drains the head and the | |
5660 | * next after the link request. The last one is done via | |
5661 | * drain_next flag to persist the effect across calls. | |
5662 | */ | |
ef4ff581 | 5663 | if (req->flags & REQ_F_IO_DRAIN) { |
711be031 PB |
5664 | head->flags |= REQ_F_IO_DRAIN; |
5665 | ctx->drain_next = 1; | |
5666 | } | |
1d4240cc PB |
5667 | if (io_alloc_async_ctx(req)) |
5668 | return -EAGAIN; | |
9e645e11 | 5669 | |
3529d8c2 | 5670 | ret = io_req_defer_prep(req, sqe); |
2d28390a | 5671 | if (ret) { |
4e88d6e7 | 5672 | /* fail even hard links since we don't submit */ |
9d76377f | 5673 | head->flags |= REQ_F_FAIL_LINK; |
1d4240cc | 5674 | return ret; |
2d28390a | 5675 | } |
9d76377f PB |
5676 | trace_io_uring_link(ctx, req, head); |
5677 | list_add_tail(&req->link_list, &head->link_list); | |
32fe525b PB |
5678 | |
5679 | /* last request of a link, enqueue the link */ | |
ef4ff581 | 5680 | if (!(req->flags & (REQ_F_LINK | REQ_F_HARDLINK))) { |
32fe525b PB |
5681 | io_queue_link_head(head); |
5682 | *link = NULL; | |
5683 | } | |
9e645e11 | 5684 | } else { |
711be031 PB |
5685 | if (unlikely(ctx->drain_next)) { |
5686 | req->flags |= REQ_F_IO_DRAIN; | |
ef4ff581 | 5687 | ctx->drain_next = 0; |
711be031 | 5688 | } |
ef4ff581 | 5689 | if (req->flags & (REQ_F_LINK | REQ_F_HARDLINK)) { |
dea3b49c | 5690 | req->flags |= REQ_F_LINK_HEAD; |
711be031 | 5691 | INIT_LIST_HEAD(&req->link_list); |
f1d96a8f | 5692 | |
1d4240cc PB |
5693 | if (io_alloc_async_ctx(req)) |
5694 | return -EAGAIN; | |
5695 | ||
711be031 PB |
5696 | ret = io_req_defer_prep(req, sqe); |
5697 | if (ret) | |
5698 | req->flags |= REQ_F_FAIL_LINK; | |
5699 | *link = req; | |
5700 | } else { | |
5701 | io_queue_sqe(req, sqe); | |
5702 | } | |
9e645e11 | 5703 | } |
2e6e1fde | 5704 | |
1d4240cc | 5705 | return 0; |
9e645e11 JA |
5706 | } |
5707 | ||
9a56a232 JA |
5708 | /* |
5709 | * Batched submission is done, ensure local IO is flushed out. | |
5710 | */ | |
5711 | static void io_submit_state_end(struct io_submit_state *state) | |
5712 | { | |
5713 | blk_finish_plug(&state->plug); | |
3d6770fb | 5714 | io_file_put(state); |
2579f913 | 5715 | if (state->free_reqs) |
6c8a3134 | 5716 | kmem_cache_free_bulk(req_cachep, state->free_reqs, state->reqs); |
9a56a232 JA |
5717 | } |
5718 | ||
5719 | /* | |
5720 | * Start submission side cache. | |
5721 | */ | |
5722 | static void io_submit_state_start(struct io_submit_state *state, | |
22efde59 | 5723 | unsigned int max_ios) |
9a56a232 JA |
5724 | { |
5725 | blk_start_plug(&state->plug); | |
2579f913 | 5726 | state->free_reqs = 0; |
9a56a232 JA |
5727 | state->file = NULL; |
5728 | state->ios_left = max_ios; | |
5729 | } | |
5730 | ||
2b188cc1 JA |
5731 | static void io_commit_sqring(struct io_ring_ctx *ctx) |
5732 | { | |
75b28aff | 5733 | struct io_rings *rings = ctx->rings; |
2b188cc1 | 5734 | |
caf582c6 PB |
5735 | /* |
5736 | * Ensure any loads from the SQEs are done at this point, | |
5737 | * since once we write the new head, the application could | |
5738 | * write new data to them. | |
5739 | */ | |
5740 | smp_store_release(&rings->sq.head, ctx->cached_sq_head); | |
2b188cc1 JA |
5741 | } |
5742 | ||
2b188cc1 | 5743 | /* |
3529d8c2 | 5744 | * Fetch an sqe, if one is available. Note that sqe_ptr will point to memory |
2b188cc1 JA |
5745 | * that is mapped by userspace. This means that care needs to be taken to |
5746 | * ensure that reads are stable, as we cannot rely on userspace always | |
5747 | * being a good citizen. If members of the sqe are validated and then later | |
5748 | * used, it's important that those reads are done through READ_ONCE() to | |
5749 | * prevent a re-load down the line. | |
5750 | */ | |
709b302f | 5751 | static const struct io_uring_sqe *io_get_sqe(struct io_ring_ctx *ctx) |
2b188cc1 | 5752 | { |
75b28aff | 5753 | u32 *sq_array = ctx->sq_array; |
2b188cc1 JA |
5754 | unsigned head; |
5755 | ||
5756 | /* | |
5757 | * The cached sq head (or cq tail) serves two purposes: | |
5758 | * | |
5759 | * 1) allows us to batch the cost of updating the user visible | |
5760 | * head updates. | |
5761 | * 2) allows the kernel side to track the head on its own, even | |
5762 | * though the application is the one updating it. | |
5763 | */ | |
ee7d46d9 | 5764 | head = READ_ONCE(sq_array[ctx->cached_sq_head & ctx->sq_mask]); |
709b302f PB |
5765 | if (likely(head < ctx->sq_entries)) |
5766 | return &ctx->sq_sqes[head]; | |
2b188cc1 JA |
5767 | |
5768 | /* drop invalid entries */ | |
498ccd9e | 5769 | ctx->cached_sq_dropped++; |
ee7d46d9 | 5770 | WRITE_ONCE(ctx->rings->sq_dropped, ctx->cached_sq_dropped); |
709b302f PB |
5771 | return NULL; |
5772 | } | |
5773 | ||
5774 | static inline void io_consume_sqe(struct io_ring_ctx *ctx) | |
5775 | { | |
5776 | ctx->cached_sq_head++; | |
2b188cc1 JA |
5777 | } |
5778 | ||
ef4ff581 PB |
5779 | #define SQE_VALID_FLAGS (IOSQE_FIXED_FILE|IOSQE_IO_DRAIN|IOSQE_IO_LINK| \ |
5780 | IOSQE_IO_HARDLINK | IOSQE_ASYNC | \ | |
5781 | IOSQE_BUFFER_SELECT) | |
5782 | ||
5783 | static int io_init_req(struct io_ring_ctx *ctx, struct io_kiocb *req, | |
5784 | const struct io_uring_sqe *sqe, | |
5785 | struct io_submit_state *state, bool async) | |
0553b8bd | 5786 | { |
ef4ff581 | 5787 | unsigned int sqe_flags; |
63ff8223 | 5788 | int id; |
ef4ff581 | 5789 | |
0553b8bd PB |
5790 | /* |
5791 | * All io need record the previous position, if LINK vs DARIN, | |
5792 | * it can be used to mark the position of the first IO in the | |
5793 | * link list. | |
5794 | */ | |
31af27c7 | 5795 | req->sequence = ctx->cached_sq_head - ctx->cached_sq_dropped; |
0553b8bd PB |
5796 | req->opcode = READ_ONCE(sqe->opcode); |
5797 | req->user_data = READ_ONCE(sqe->user_data); | |
5798 | req->io = NULL; | |
5799 | req->file = NULL; | |
5800 | req->ctx = ctx; | |
5801 | req->flags = 0; | |
5802 | /* one is dropped after submission, the other at completion */ | |
5803 | refcount_set(&req->refs, 2); | |
5804 | req->task = NULL; | |
5805 | req->result = 0; | |
ef4ff581 | 5806 | req->needs_fixed_file = async; |
0553b8bd | 5807 | INIT_IO_WORK(&req->work, io_wq_submit_work); |
ef4ff581 PB |
5808 | |
5809 | if (unlikely(req->opcode >= IORING_OP_LAST)) | |
5810 | return -EINVAL; | |
5811 | ||
5812 | if (io_op_defs[req->opcode].needs_mm && !current->mm) { | |
5813 | if (unlikely(!mmget_not_zero(ctx->sqo_mm))) | |
5814 | return -EFAULT; | |
5815 | use_mm(ctx->sqo_mm); | |
5816 | } | |
5817 | ||
5818 | sqe_flags = READ_ONCE(sqe->flags); | |
5819 | /* enforce forwards compatibility on users */ | |
5820 | if (unlikely(sqe_flags & ~SQE_VALID_FLAGS)) | |
5821 | return -EINVAL; | |
5822 | ||
5823 | if ((sqe_flags & IOSQE_BUFFER_SELECT) && | |
5824 | !io_op_defs[req->opcode].buffer_select) | |
5825 | return -EOPNOTSUPP; | |
5826 | ||
5827 | id = READ_ONCE(sqe->personality); | |
5828 | if (id) { | |
5829 | req->work.creds = idr_find(&ctx->personality_idr, id); | |
5830 | if (unlikely(!req->work.creds)) | |
5831 | return -EINVAL; | |
5832 | get_cred(req->work.creds); | |
5833 | } | |
5834 | ||
5835 | /* same numerical values with corresponding REQ_F_*, safe to copy */ | |
5836 | req->flags |= sqe_flags & (IOSQE_IO_DRAIN | IOSQE_IO_HARDLINK | | |
5837 | IOSQE_ASYNC | IOSQE_FIXED_FILE | | |
5838 | IOSQE_BUFFER_SELECT | IOSQE_IO_LINK); | |
5839 | ||
63ff8223 JA |
5840 | if (!io_op_defs[req->opcode].needs_file) |
5841 | return 0; | |
5842 | ||
5843 | return io_req_set_file(state, req, READ_ONCE(sqe->fd)); | |
0553b8bd PB |
5844 | } |
5845 | ||
fb5ccc98 | 5846 | static int io_submit_sqes(struct io_ring_ctx *ctx, unsigned int nr, |
bf9c2f1c | 5847 | struct file *ring_file, int ring_fd, bool async) |
6c271ce2 JA |
5848 | { |
5849 | struct io_submit_state state, *statep = NULL; | |
9e645e11 | 5850 | struct io_kiocb *link = NULL; |
9e645e11 | 5851 | int i, submitted = 0; |
6c271ce2 | 5852 | |
c4a2ed72 | 5853 | /* if we have a backlog and couldn't flush it all, return BUSY */ |
ad3eb2c8 JA |
5854 | if (test_bit(0, &ctx->sq_check_overflow)) { |
5855 | if (!list_empty(&ctx->cq_overflow_list) && | |
5856 | !io_cqring_overflow_flush(ctx, false)) | |
5857 | return -EBUSY; | |
5858 | } | |
6c271ce2 | 5859 | |
ee7d46d9 PB |
5860 | /* make sure SQ entry isn't read before tail */ |
5861 | nr = min3(nr, ctx->sq_entries, io_sqring_entries(ctx)); | |
9ef4f124 | 5862 | |
2b85edfc PB |
5863 | if (!percpu_ref_tryget_many(&ctx->refs, nr)) |
5864 | return -EAGAIN; | |
6c271ce2 JA |
5865 | |
5866 | if (nr > IO_PLUG_THRESHOLD) { | |
22efde59 | 5867 | io_submit_state_start(&state, nr); |
6c271ce2 JA |
5868 | statep = &state; |
5869 | } | |
5870 | ||
b14cca0c PB |
5871 | ctx->ring_fd = ring_fd; |
5872 | ctx->ring_file = ring_file; | |
5873 | ||
6c271ce2 | 5874 | for (i = 0; i < nr; i++) { |
3529d8c2 | 5875 | const struct io_uring_sqe *sqe; |
196be95c | 5876 | struct io_kiocb *req; |
1cb1edb2 | 5877 | int err; |
fb5ccc98 | 5878 | |
b1e50e54 PB |
5879 | sqe = io_get_sqe(ctx); |
5880 | if (unlikely(!sqe)) { | |
5881 | io_consume_sqe(ctx); | |
5882 | break; | |
5883 | } | |
0553b8bd | 5884 | req = io_alloc_req(ctx, statep); |
196be95c PB |
5885 | if (unlikely(!req)) { |
5886 | if (!submitted) | |
5887 | submitted = -EAGAIN; | |
fb5ccc98 | 5888 | break; |
196be95c | 5889 | } |
fb5ccc98 | 5890 | |
ef4ff581 | 5891 | err = io_init_req(ctx, req, sqe, statep, async); |
709b302f | 5892 | io_consume_sqe(ctx); |
d3656344 JA |
5893 | /* will complete beyond this point, count as submitted */ |
5894 | submitted++; | |
5895 | ||
ef4ff581 | 5896 | if (unlikely(err)) { |
1cb1edb2 PB |
5897 | fail_req: |
5898 | io_cqring_add_event(req, err); | |
d3656344 | 5899 | io_double_put_req(req); |
196be95c PB |
5900 | break; |
5901 | } | |
fb5ccc98 | 5902 | |
354420f7 JA |
5903 | trace_io_uring_submit_sqe(ctx, req->opcode, req->user_data, |
5904 | true, async); | |
1d4240cc PB |
5905 | err = io_submit_sqe(req, sqe, statep, &link); |
5906 | if (err) | |
5907 | goto fail_req; | |
6c271ce2 JA |
5908 | } |
5909 | ||
9466f437 PB |
5910 | if (unlikely(submitted != nr)) { |
5911 | int ref_used = (submitted == -EAGAIN) ? 0 : submitted; | |
5912 | ||
5913 | percpu_ref_put_many(&ctx->refs, nr - ref_used); | |
5914 | } | |
9e645e11 | 5915 | if (link) |
1b4a51b6 | 5916 | io_queue_link_head(link); |
6c271ce2 JA |
5917 | if (statep) |
5918 | io_submit_state_end(&state); | |
5919 | ||
ae9428ca PB |
5920 | /* Commit SQ ring head once we've consumed and submitted all SQEs */ |
5921 | io_commit_sqring(ctx); | |
5922 | ||
6c271ce2 JA |
5923 | return submitted; |
5924 | } | |
5925 | ||
bf9c2f1c PB |
5926 | static inline void io_sq_thread_drop_mm(struct io_ring_ctx *ctx) |
5927 | { | |
5928 | struct mm_struct *mm = current->mm; | |
5929 | ||
5930 | if (mm) { | |
5931 | unuse_mm(mm); | |
5932 | mmput(mm); | |
5933 | } | |
5934 | } | |
5935 | ||
6c271ce2 JA |
5936 | static int io_sq_thread(void *data) |
5937 | { | |
6c271ce2 | 5938 | struct io_ring_ctx *ctx = data; |
181e448d | 5939 | const struct cred *old_cred; |
6c271ce2 JA |
5940 | mm_segment_t old_fs; |
5941 | DEFINE_WAIT(wait); | |
6c271ce2 | 5942 | unsigned long timeout; |
bdcd3eab | 5943 | int ret = 0; |
6c271ce2 | 5944 | |
206aefde | 5945 | complete(&ctx->completions[1]); |
a4c0b3de | 5946 | |
6c271ce2 JA |
5947 | old_fs = get_fs(); |
5948 | set_fs(USER_DS); | |
181e448d | 5949 | old_cred = override_creds(ctx->creds); |
6c271ce2 | 5950 | |
bdcd3eab | 5951 | timeout = jiffies + ctx->sq_thread_idle; |
2bbcd6d3 | 5952 | while (!kthread_should_park()) { |
fb5ccc98 | 5953 | unsigned int to_submit; |
6c271ce2 | 5954 | |
bdcd3eab | 5955 | if (!list_empty(&ctx->poll_list)) { |
6c271ce2 JA |
5956 | unsigned nr_events = 0; |
5957 | ||
bdcd3eab XW |
5958 | mutex_lock(&ctx->uring_lock); |
5959 | if (!list_empty(&ctx->poll_list)) | |
5960 | io_iopoll_getevents(ctx, &nr_events, 0); | |
5961 | else | |
6c271ce2 | 5962 | timeout = jiffies + ctx->sq_thread_idle; |
bdcd3eab | 5963 | mutex_unlock(&ctx->uring_lock); |
6c271ce2 JA |
5964 | } |
5965 | ||
fb5ccc98 | 5966 | to_submit = io_sqring_entries(ctx); |
c1edbf5f JA |
5967 | |
5968 | /* | |
5969 | * If submit got -EBUSY, flag us as needing the application | |
5970 | * to enter the kernel to reap and flush events. | |
5971 | */ | |
5972 | if (!to_submit || ret == -EBUSY) { | |
7143b5ac SG |
5973 | /* |
5974 | * Drop cur_mm before scheduling, we can't hold it for | |
5975 | * long periods (or over schedule()). Do this before | |
5976 | * adding ourselves to the waitqueue, as the unuse/drop | |
5977 | * may sleep. | |
5978 | */ | |
bf9c2f1c | 5979 | io_sq_thread_drop_mm(ctx); |
7143b5ac | 5980 | |
6c271ce2 JA |
5981 | /* |
5982 | * We're polling. If we're within the defined idle | |
5983 | * period, then let us spin without work before going | |
c1edbf5f JA |
5984 | * to sleep. The exception is if we got EBUSY doing |
5985 | * more IO, we should wait for the application to | |
5986 | * reap events and wake us up. | |
6c271ce2 | 5987 | */ |
bdcd3eab | 5988 | if (!list_empty(&ctx->poll_list) || |
df069d80 JA |
5989 | (!time_after(jiffies, timeout) && ret != -EBUSY && |
5990 | !percpu_ref_is_dying(&ctx->refs))) { | |
b41e9852 JA |
5991 | if (current->task_works) |
5992 | task_work_run(); | |
9831a90c | 5993 | cond_resched(); |
6c271ce2 JA |
5994 | continue; |
5995 | } | |
5996 | ||
6c271ce2 JA |
5997 | prepare_to_wait(&ctx->sqo_wait, &wait, |
5998 | TASK_INTERRUPTIBLE); | |
5999 | ||
bdcd3eab XW |
6000 | /* |
6001 | * While doing polled IO, before going to sleep, we need | |
6002 | * to check if there are new reqs added to poll_list, it | |
6003 | * is because reqs may have been punted to io worker and | |
6004 | * will be added to poll_list later, hence check the | |
6005 | * poll_list again. | |
6006 | */ | |
6007 | if ((ctx->flags & IORING_SETUP_IOPOLL) && | |
6008 | !list_empty_careful(&ctx->poll_list)) { | |
6009 | finish_wait(&ctx->sqo_wait, &wait); | |
6010 | continue; | |
6011 | } | |
6012 | ||
6c271ce2 | 6013 | /* Tell userspace we may need a wakeup call */ |
75b28aff | 6014 | ctx->rings->sq_flags |= IORING_SQ_NEED_WAKEUP; |
0d7bae69 SB |
6015 | /* make sure to read SQ tail after writing flags */ |
6016 | smp_mb(); | |
6c271ce2 | 6017 | |
fb5ccc98 | 6018 | to_submit = io_sqring_entries(ctx); |
c1edbf5f | 6019 | if (!to_submit || ret == -EBUSY) { |
2bbcd6d3 | 6020 | if (kthread_should_park()) { |
6c271ce2 JA |
6021 | finish_wait(&ctx->sqo_wait, &wait); |
6022 | break; | |
6023 | } | |
b41e9852 JA |
6024 | if (current->task_works) { |
6025 | task_work_run(); | |
10bea96d | 6026 | finish_wait(&ctx->sqo_wait, &wait); |
b41e9852 JA |
6027 | continue; |
6028 | } | |
6c271ce2 JA |
6029 | if (signal_pending(current)) |
6030 | flush_signals(current); | |
6031 | schedule(); | |
6032 | finish_wait(&ctx->sqo_wait, &wait); | |
6033 | ||
75b28aff | 6034 | ctx->rings->sq_flags &= ~IORING_SQ_NEED_WAKEUP; |
6c271ce2 JA |
6035 | continue; |
6036 | } | |
6037 | finish_wait(&ctx->sqo_wait, &wait); | |
6038 | ||
75b28aff | 6039 | ctx->rings->sq_flags &= ~IORING_SQ_NEED_WAKEUP; |
6c271ce2 JA |
6040 | } |
6041 | ||
8a4955ff | 6042 | mutex_lock(&ctx->uring_lock); |
bf9c2f1c | 6043 | ret = io_submit_sqes(ctx, to_submit, NULL, -1, true); |
8a4955ff | 6044 | mutex_unlock(&ctx->uring_lock); |
bdcd3eab | 6045 | timeout = jiffies + ctx->sq_thread_idle; |
6c271ce2 JA |
6046 | } |
6047 | ||
b41e9852 JA |
6048 | if (current->task_works) |
6049 | task_work_run(); | |
6050 | ||
6c271ce2 | 6051 | set_fs(old_fs); |
bf9c2f1c | 6052 | io_sq_thread_drop_mm(ctx); |
181e448d | 6053 | revert_creds(old_cred); |
06058632 | 6054 | |
2bbcd6d3 | 6055 | kthread_parkme(); |
06058632 | 6056 | |
6c271ce2 JA |
6057 | return 0; |
6058 | } | |
6059 | ||
bda52162 JA |
6060 | struct io_wait_queue { |
6061 | struct wait_queue_entry wq; | |
6062 | struct io_ring_ctx *ctx; | |
6063 | unsigned to_wait; | |
6064 | unsigned nr_timeouts; | |
6065 | }; | |
6066 | ||
1d7bb1d5 | 6067 | static inline bool io_should_wake(struct io_wait_queue *iowq, bool noflush) |
bda52162 JA |
6068 | { |
6069 | struct io_ring_ctx *ctx = iowq->ctx; | |
6070 | ||
6071 | /* | |
d195a66e | 6072 | * Wake up if we have enough events, or if a timeout occurred since we |
bda52162 JA |
6073 | * started waiting. For timeouts, we always want to return to userspace, |
6074 | * regardless of event count. | |
6075 | */ | |
1d7bb1d5 | 6076 | return io_cqring_events(ctx, noflush) >= iowq->to_wait || |
bda52162 JA |
6077 | atomic_read(&ctx->cq_timeouts) != iowq->nr_timeouts; |
6078 | } | |
6079 | ||
6080 | static int io_wake_function(struct wait_queue_entry *curr, unsigned int mode, | |
6081 | int wake_flags, void *key) | |
6082 | { | |
6083 | struct io_wait_queue *iowq = container_of(curr, struct io_wait_queue, | |
6084 | wq); | |
6085 | ||
1d7bb1d5 JA |
6086 | /* use noflush == true, as we can't safely rely on locking context */ |
6087 | if (!io_should_wake(iowq, true)) | |
bda52162 JA |
6088 | return -1; |
6089 | ||
6090 | return autoremove_wake_function(curr, mode, wake_flags, key); | |
6091 | } | |
6092 | ||
2b188cc1 JA |
6093 | /* |
6094 | * Wait until events become available, if we don't already have some. The | |
6095 | * application must reap them itself, as they reside on the shared cq ring. | |
6096 | */ | |
6097 | static int io_cqring_wait(struct io_ring_ctx *ctx, int min_events, | |
6098 | const sigset_t __user *sig, size_t sigsz) | |
6099 | { | |
bda52162 JA |
6100 | struct io_wait_queue iowq = { |
6101 | .wq = { | |
6102 | .private = current, | |
6103 | .func = io_wake_function, | |
6104 | .entry = LIST_HEAD_INIT(iowq.wq.entry), | |
6105 | }, | |
6106 | .ctx = ctx, | |
6107 | .to_wait = min_events, | |
6108 | }; | |
75b28aff | 6109 | struct io_rings *rings = ctx->rings; |
e9ffa5c2 | 6110 | int ret = 0; |
2b188cc1 | 6111 | |
b41e9852 JA |
6112 | do { |
6113 | if (io_cqring_events(ctx, false) >= min_events) | |
6114 | return 0; | |
6115 | if (!current->task_works) | |
6116 | break; | |
6117 | task_work_run(); | |
6118 | } while (1); | |
2b188cc1 JA |
6119 | |
6120 | if (sig) { | |
9e75ad5d AB |
6121 | #ifdef CONFIG_COMPAT |
6122 | if (in_compat_syscall()) | |
6123 | ret = set_compat_user_sigmask((const compat_sigset_t __user *)sig, | |
b772434b | 6124 | sigsz); |
9e75ad5d AB |
6125 | else |
6126 | #endif | |
b772434b | 6127 | ret = set_user_sigmask(sig, sigsz); |
9e75ad5d | 6128 | |
2b188cc1 JA |
6129 | if (ret) |
6130 | return ret; | |
6131 | } | |
6132 | ||
bda52162 | 6133 | iowq.nr_timeouts = atomic_read(&ctx->cq_timeouts); |
c826bd7a | 6134 | trace_io_uring_cqring_wait(ctx, min_events); |
bda52162 JA |
6135 | do { |
6136 | prepare_to_wait_exclusive(&ctx->wait, &iowq.wq, | |
6137 | TASK_INTERRUPTIBLE); | |
b41e9852 JA |
6138 | if (current->task_works) |
6139 | task_work_run(); | |
1d7bb1d5 | 6140 | if (io_should_wake(&iowq, false)) |
bda52162 JA |
6141 | break; |
6142 | schedule(); | |
6143 | if (signal_pending(current)) { | |
e9ffa5c2 | 6144 | ret = -EINTR; |
bda52162 JA |
6145 | break; |
6146 | } | |
6147 | } while (1); | |
6148 | finish_wait(&ctx->wait, &iowq.wq); | |
6149 | ||
e9ffa5c2 | 6150 | restore_saved_sigmask_unless(ret == -EINTR); |
2b188cc1 | 6151 | |
75b28aff | 6152 | return READ_ONCE(rings->cq.head) == READ_ONCE(rings->cq.tail) ? ret : 0; |
2b188cc1 JA |
6153 | } |
6154 | ||
6b06314c JA |
6155 | static void __io_sqe_files_unregister(struct io_ring_ctx *ctx) |
6156 | { | |
6157 | #if defined(CONFIG_UNIX) | |
6158 | if (ctx->ring_sock) { | |
6159 | struct sock *sock = ctx->ring_sock->sk; | |
6160 | struct sk_buff *skb; | |
6161 | ||
6162 | while ((skb = skb_dequeue(&sock->sk_receive_queue)) != NULL) | |
6163 | kfree_skb(skb); | |
6164 | } | |
6165 | #else | |
6166 | int i; | |
6167 | ||
65e19f54 JA |
6168 | for (i = 0; i < ctx->nr_user_files; i++) { |
6169 | struct file *file; | |
6170 | ||
6171 | file = io_file_from_index(ctx, i); | |
6172 | if (file) | |
6173 | fput(file); | |
6174 | } | |
6b06314c JA |
6175 | #endif |
6176 | } | |
6177 | ||
05f3fb3c JA |
6178 | static void io_file_ref_kill(struct percpu_ref *ref) |
6179 | { | |
6180 | struct fixed_file_data *data; | |
6181 | ||
6182 | data = container_of(ref, struct fixed_file_data, refs); | |
6183 | complete(&data->done); | |
6184 | } | |
6185 | ||
6b06314c JA |
6186 | static int io_sqe_files_unregister(struct io_ring_ctx *ctx) |
6187 | { | |
05f3fb3c | 6188 | struct fixed_file_data *data = ctx->file_data; |
05589553 | 6189 | struct fixed_file_ref_node *ref_node = NULL; |
65e19f54 | 6190 | unsigned nr_tables, i; |
05589553 | 6191 | unsigned long flags; |
65e19f54 | 6192 | |
05f3fb3c | 6193 | if (!data) |
6b06314c JA |
6194 | return -ENXIO; |
6195 | ||
05589553 XW |
6196 | spin_lock_irqsave(&data->lock, flags); |
6197 | if (!list_empty(&data->ref_list)) | |
6198 | ref_node = list_first_entry(&data->ref_list, | |
6199 | struct fixed_file_ref_node, node); | |
6200 | spin_unlock_irqrestore(&data->lock, flags); | |
6201 | if (ref_node) | |
6202 | percpu_ref_kill(&ref_node->refs); | |
6203 | ||
6204 | percpu_ref_kill(&data->refs); | |
6205 | ||
6206 | /* wait for all refs nodes to complete */ | |
2faf852d | 6207 | wait_for_completion(&data->done); |
05f3fb3c | 6208 | |
6b06314c | 6209 | __io_sqe_files_unregister(ctx); |
65e19f54 JA |
6210 | nr_tables = DIV_ROUND_UP(ctx->nr_user_files, IORING_MAX_FILES_TABLE); |
6211 | for (i = 0; i < nr_tables; i++) | |
05f3fb3c JA |
6212 | kfree(data->table[i].files); |
6213 | kfree(data->table); | |
05589553 XW |
6214 | percpu_ref_exit(&data->refs); |
6215 | kfree(data); | |
05f3fb3c | 6216 | ctx->file_data = NULL; |
6b06314c JA |
6217 | ctx->nr_user_files = 0; |
6218 | return 0; | |
6219 | } | |
6220 | ||
6c271ce2 JA |
6221 | static void io_sq_thread_stop(struct io_ring_ctx *ctx) |
6222 | { | |
6223 | if (ctx->sqo_thread) { | |
206aefde | 6224 | wait_for_completion(&ctx->completions[1]); |
2bbcd6d3 RP |
6225 | /* |
6226 | * The park is a bit of a work-around, without it we get | |
6227 | * warning spews on shutdown with SQPOLL set and affinity | |
6228 | * set to a single CPU. | |
6229 | */ | |
06058632 | 6230 | kthread_park(ctx->sqo_thread); |
6c271ce2 JA |
6231 | kthread_stop(ctx->sqo_thread); |
6232 | ctx->sqo_thread = NULL; | |
6233 | } | |
6234 | } | |
6235 | ||
6b06314c JA |
6236 | static void io_finish_async(struct io_ring_ctx *ctx) |
6237 | { | |
6c271ce2 JA |
6238 | io_sq_thread_stop(ctx); |
6239 | ||
561fb04a JA |
6240 | if (ctx->io_wq) { |
6241 | io_wq_destroy(ctx->io_wq); | |
6242 | ctx->io_wq = NULL; | |
6b06314c JA |
6243 | } |
6244 | } | |
6245 | ||
6246 | #if defined(CONFIG_UNIX) | |
6b06314c JA |
6247 | /* |
6248 | * Ensure the UNIX gc is aware of our file set, so we are certain that | |
6249 | * the io_uring can be safely unregistered on process exit, even if we have | |
6250 | * loops in the file referencing. | |
6251 | */ | |
6252 | static int __io_sqe_files_scm(struct io_ring_ctx *ctx, int nr, int offset) | |
6253 | { | |
6254 | struct sock *sk = ctx->ring_sock->sk; | |
6255 | struct scm_fp_list *fpl; | |
6256 | struct sk_buff *skb; | |
08a45173 | 6257 | int i, nr_files; |
6b06314c | 6258 | |
6b06314c JA |
6259 | fpl = kzalloc(sizeof(*fpl), GFP_KERNEL); |
6260 | if (!fpl) | |
6261 | return -ENOMEM; | |
6262 | ||
6263 | skb = alloc_skb(0, GFP_KERNEL); | |
6264 | if (!skb) { | |
6265 | kfree(fpl); | |
6266 | return -ENOMEM; | |
6267 | } | |
6268 | ||
6269 | skb->sk = sk; | |
6b06314c | 6270 | |
08a45173 | 6271 | nr_files = 0; |
6b06314c JA |
6272 | fpl->user = get_uid(ctx->user); |
6273 | for (i = 0; i < nr; i++) { | |
65e19f54 JA |
6274 | struct file *file = io_file_from_index(ctx, i + offset); |
6275 | ||
6276 | if (!file) | |
08a45173 | 6277 | continue; |
65e19f54 | 6278 | fpl->fp[nr_files] = get_file(file); |
08a45173 JA |
6279 | unix_inflight(fpl->user, fpl->fp[nr_files]); |
6280 | nr_files++; | |
6b06314c JA |
6281 | } |
6282 | ||
08a45173 JA |
6283 | if (nr_files) { |
6284 | fpl->max = SCM_MAX_FD; | |
6285 | fpl->count = nr_files; | |
6286 | UNIXCB(skb).fp = fpl; | |
05f3fb3c | 6287 | skb->destructor = unix_destruct_scm; |
08a45173 JA |
6288 | refcount_add(skb->truesize, &sk->sk_wmem_alloc); |
6289 | skb_queue_head(&sk->sk_receive_queue, skb); | |
6b06314c | 6290 | |
08a45173 JA |
6291 | for (i = 0; i < nr_files; i++) |
6292 | fput(fpl->fp[i]); | |
6293 | } else { | |
6294 | kfree_skb(skb); | |
6295 | kfree(fpl); | |
6296 | } | |
6b06314c JA |
6297 | |
6298 | return 0; | |
6299 | } | |
6300 | ||
6301 | /* | |
6302 | * If UNIX sockets are enabled, fd passing can cause a reference cycle which | |
6303 | * causes regular reference counting to break down. We rely on the UNIX | |
6304 | * garbage collection to take care of this problem for us. | |
6305 | */ | |
6306 | static int io_sqe_files_scm(struct io_ring_ctx *ctx) | |
6307 | { | |
6308 | unsigned left, total; | |
6309 | int ret = 0; | |
6310 | ||
6311 | total = 0; | |
6312 | left = ctx->nr_user_files; | |
6313 | while (left) { | |
6314 | unsigned this_files = min_t(unsigned, left, SCM_MAX_FD); | |
6b06314c JA |
6315 | |
6316 | ret = __io_sqe_files_scm(ctx, this_files, total); | |
6317 | if (ret) | |
6318 | break; | |
6319 | left -= this_files; | |
6320 | total += this_files; | |
6321 | } | |
6322 | ||
6323 | if (!ret) | |
6324 | return 0; | |
6325 | ||
6326 | while (total < ctx->nr_user_files) { | |
65e19f54 JA |
6327 | struct file *file = io_file_from_index(ctx, total); |
6328 | ||
6329 | if (file) | |
6330 | fput(file); | |
6b06314c JA |
6331 | total++; |
6332 | } | |
6333 | ||
6334 | return ret; | |
6335 | } | |
6336 | #else | |
6337 | static int io_sqe_files_scm(struct io_ring_ctx *ctx) | |
6338 | { | |
6339 | return 0; | |
6340 | } | |
6341 | #endif | |
6342 | ||
65e19f54 JA |
6343 | static int io_sqe_alloc_file_tables(struct io_ring_ctx *ctx, unsigned nr_tables, |
6344 | unsigned nr_files) | |
6345 | { | |
6346 | int i; | |
6347 | ||
6348 | for (i = 0; i < nr_tables; i++) { | |
05f3fb3c | 6349 | struct fixed_file_table *table = &ctx->file_data->table[i]; |
65e19f54 JA |
6350 | unsigned this_files; |
6351 | ||
6352 | this_files = min(nr_files, IORING_MAX_FILES_TABLE); | |
6353 | table->files = kcalloc(this_files, sizeof(struct file *), | |
6354 | GFP_KERNEL); | |
6355 | if (!table->files) | |
6356 | break; | |
6357 | nr_files -= this_files; | |
6358 | } | |
6359 | ||
6360 | if (i == nr_tables) | |
6361 | return 0; | |
6362 | ||
6363 | for (i = 0; i < nr_tables; i++) { | |
05f3fb3c | 6364 | struct fixed_file_table *table = &ctx->file_data->table[i]; |
65e19f54 JA |
6365 | kfree(table->files); |
6366 | } | |
6367 | return 1; | |
6368 | } | |
6369 | ||
05f3fb3c JA |
6370 | static void io_ring_file_put(struct io_ring_ctx *ctx, struct file *file) |
6371 | { | |
6372 | #if defined(CONFIG_UNIX) | |
6373 | struct sock *sock = ctx->ring_sock->sk; | |
6374 | struct sk_buff_head list, *head = &sock->sk_receive_queue; | |
6375 | struct sk_buff *skb; | |
6376 | int i; | |
6377 | ||
6378 | __skb_queue_head_init(&list); | |
6379 | ||
6380 | /* | |
6381 | * Find the skb that holds this file in its SCM_RIGHTS. When found, | |
6382 | * remove this entry and rearrange the file array. | |
6383 | */ | |
6384 | skb = skb_dequeue(head); | |
6385 | while (skb) { | |
6386 | struct scm_fp_list *fp; | |
6387 | ||
6388 | fp = UNIXCB(skb).fp; | |
6389 | for (i = 0; i < fp->count; i++) { | |
6390 | int left; | |
6391 | ||
6392 | if (fp->fp[i] != file) | |
6393 | continue; | |
6394 | ||
6395 | unix_notinflight(fp->user, fp->fp[i]); | |
6396 | left = fp->count - 1 - i; | |
6397 | if (left) { | |
6398 | memmove(&fp->fp[i], &fp->fp[i + 1], | |
6399 | left * sizeof(struct file *)); | |
6400 | } | |
6401 | fp->count--; | |
6402 | if (!fp->count) { | |
6403 | kfree_skb(skb); | |
6404 | skb = NULL; | |
6405 | } else { | |
6406 | __skb_queue_tail(&list, skb); | |
6407 | } | |
6408 | fput(file); | |
6409 | file = NULL; | |
6410 | break; | |
6411 | } | |
6412 | ||
6413 | if (!file) | |
6414 | break; | |
6415 | ||
6416 | __skb_queue_tail(&list, skb); | |
6417 | ||
6418 | skb = skb_dequeue(head); | |
6419 | } | |
6420 | ||
6421 | if (skb_peek(&list)) { | |
6422 | spin_lock_irq(&head->lock); | |
6423 | while ((skb = __skb_dequeue(&list)) != NULL) | |
6424 | __skb_queue_tail(head, skb); | |
6425 | spin_unlock_irq(&head->lock); | |
6426 | } | |
6427 | #else | |
6428 | fput(file); | |
6429 | #endif | |
6430 | } | |
6431 | ||
6432 | struct io_file_put { | |
05589553 | 6433 | struct list_head list; |
05f3fb3c | 6434 | struct file *file; |
05f3fb3c JA |
6435 | }; |
6436 | ||
05589553 | 6437 | static void io_file_put_work(struct work_struct *work) |
65e19f54 | 6438 | { |
05589553 XW |
6439 | struct fixed_file_ref_node *ref_node; |
6440 | struct fixed_file_data *file_data; | |
6441 | struct io_ring_ctx *ctx; | |
05f3fb3c | 6442 | struct io_file_put *pfile, *tmp; |
05589553 | 6443 | unsigned long flags; |
65e19f54 | 6444 | |
05589553 XW |
6445 | ref_node = container_of(work, struct fixed_file_ref_node, work); |
6446 | file_data = ref_node->file_data; | |
6447 | ctx = file_data->ctx; | |
6448 | ||
6449 | list_for_each_entry_safe(pfile, tmp, &ref_node->file_list, list) { | |
6450 | list_del_init(&pfile->list); | |
6451 | io_ring_file_put(ctx, pfile->file); | |
6452 | kfree(pfile); | |
65e19f54 | 6453 | } |
05589553 XW |
6454 | |
6455 | spin_lock_irqsave(&file_data->lock, flags); | |
6456 | list_del_init(&ref_node->node); | |
6457 | spin_unlock_irqrestore(&file_data->lock, flags); | |
6458 | ||
6459 | percpu_ref_exit(&ref_node->refs); | |
6460 | kfree(ref_node); | |
6461 | percpu_ref_put(&file_data->refs); | |
2faf852d | 6462 | } |
65e19f54 | 6463 | |
05589553 | 6464 | static void io_file_data_ref_zero(struct percpu_ref *ref) |
2faf852d | 6465 | { |
05589553 | 6466 | struct fixed_file_ref_node *ref_node; |
65e19f54 | 6467 | |
05589553 XW |
6468 | ref_node = container_of(ref, struct fixed_file_ref_node, refs); |
6469 | ||
6470 | queue_work(system_wq, &ref_node->work); | |
05f3fb3c | 6471 | } |
65e19f54 | 6472 | |
05589553 XW |
6473 | static struct fixed_file_ref_node *alloc_fixed_file_ref_node( |
6474 | struct io_ring_ctx *ctx) | |
05f3fb3c | 6475 | { |
05589553 | 6476 | struct fixed_file_ref_node *ref_node; |
05f3fb3c | 6477 | |
05589553 XW |
6478 | ref_node = kzalloc(sizeof(*ref_node), GFP_KERNEL); |
6479 | if (!ref_node) | |
6480 | return ERR_PTR(-ENOMEM); | |
05f3fb3c | 6481 | |
05589553 XW |
6482 | if (percpu_ref_init(&ref_node->refs, io_file_data_ref_zero, |
6483 | 0, GFP_KERNEL)) { | |
6484 | kfree(ref_node); | |
6485 | return ERR_PTR(-ENOMEM); | |
6486 | } | |
6487 | INIT_LIST_HEAD(&ref_node->node); | |
6488 | INIT_LIST_HEAD(&ref_node->file_list); | |
6489 | INIT_WORK(&ref_node->work, io_file_put_work); | |
6490 | ref_node->file_data = ctx->file_data; | |
6491 | return ref_node; | |
6492 | ||
6493 | } | |
6494 | ||
6495 | static void destroy_fixed_file_ref_node(struct fixed_file_ref_node *ref_node) | |
6496 | { | |
6497 | percpu_ref_exit(&ref_node->refs); | |
6498 | kfree(ref_node); | |
65e19f54 JA |
6499 | } |
6500 | ||
6b06314c JA |
6501 | static int io_sqe_files_register(struct io_ring_ctx *ctx, void __user *arg, |
6502 | unsigned nr_args) | |
6503 | { | |
6504 | __s32 __user *fds = (__s32 __user *) arg; | |
65e19f54 | 6505 | unsigned nr_tables; |
05f3fb3c | 6506 | struct file *file; |
6b06314c JA |
6507 | int fd, ret = 0; |
6508 | unsigned i; | |
05589553 XW |
6509 | struct fixed_file_ref_node *ref_node; |
6510 | unsigned long flags; | |
6b06314c | 6511 | |
05f3fb3c | 6512 | if (ctx->file_data) |
6b06314c JA |
6513 | return -EBUSY; |
6514 | if (!nr_args) | |
6515 | return -EINVAL; | |
6516 | if (nr_args > IORING_MAX_FIXED_FILES) | |
6517 | return -EMFILE; | |
6518 | ||
05f3fb3c JA |
6519 | ctx->file_data = kzalloc(sizeof(*ctx->file_data), GFP_KERNEL); |
6520 | if (!ctx->file_data) | |
6521 | return -ENOMEM; | |
6522 | ctx->file_data->ctx = ctx; | |
6523 | init_completion(&ctx->file_data->done); | |
05589553 | 6524 | INIT_LIST_HEAD(&ctx->file_data->ref_list); |
f7fe9346 | 6525 | spin_lock_init(&ctx->file_data->lock); |
05f3fb3c | 6526 | |
65e19f54 | 6527 | nr_tables = DIV_ROUND_UP(nr_args, IORING_MAX_FILES_TABLE); |
05f3fb3c JA |
6528 | ctx->file_data->table = kcalloc(nr_tables, |
6529 | sizeof(struct fixed_file_table), | |
65e19f54 | 6530 | GFP_KERNEL); |
05f3fb3c JA |
6531 | if (!ctx->file_data->table) { |
6532 | kfree(ctx->file_data); | |
6533 | ctx->file_data = NULL; | |
6b06314c | 6534 | return -ENOMEM; |
05f3fb3c JA |
6535 | } |
6536 | ||
05589553 | 6537 | if (percpu_ref_init(&ctx->file_data->refs, io_file_ref_kill, |
05f3fb3c JA |
6538 | PERCPU_REF_ALLOW_REINIT, GFP_KERNEL)) { |
6539 | kfree(ctx->file_data->table); | |
6540 | kfree(ctx->file_data); | |
6541 | ctx->file_data = NULL; | |
6b06314c | 6542 | return -ENOMEM; |
05f3fb3c | 6543 | } |
6b06314c | 6544 | |
65e19f54 | 6545 | if (io_sqe_alloc_file_tables(ctx, nr_tables, nr_args)) { |
05f3fb3c JA |
6546 | percpu_ref_exit(&ctx->file_data->refs); |
6547 | kfree(ctx->file_data->table); | |
6548 | kfree(ctx->file_data); | |
6549 | ctx->file_data = NULL; | |
65e19f54 JA |
6550 | return -ENOMEM; |
6551 | } | |
6552 | ||
08a45173 | 6553 | for (i = 0; i < nr_args; i++, ctx->nr_user_files++) { |
65e19f54 JA |
6554 | struct fixed_file_table *table; |
6555 | unsigned index; | |
6556 | ||
6b06314c JA |
6557 | ret = -EFAULT; |
6558 | if (copy_from_user(&fd, &fds[i], sizeof(fd))) | |
6559 | break; | |
08a45173 JA |
6560 | /* allow sparse sets */ |
6561 | if (fd == -1) { | |
6562 | ret = 0; | |
6563 | continue; | |
6564 | } | |
6b06314c | 6565 | |
05f3fb3c | 6566 | table = &ctx->file_data->table[i >> IORING_FILE_TABLE_SHIFT]; |
65e19f54 | 6567 | index = i & IORING_FILE_TABLE_MASK; |
05f3fb3c | 6568 | file = fget(fd); |
6b06314c JA |
6569 | |
6570 | ret = -EBADF; | |
05f3fb3c | 6571 | if (!file) |
6b06314c | 6572 | break; |
05f3fb3c | 6573 | |
6b06314c JA |
6574 | /* |
6575 | * Don't allow io_uring instances to be registered. If UNIX | |
6576 | * isn't enabled, then this causes a reference cycle and this | |
6577 | * instance can never get freed. If UNIX is enabled we'll | |
6578 | * handle it just fine, but there's still no point in allowing | |
6579 | * a ring fd as it doesn't support regular read/write anyway. | |
6580 | */ | |
05f3fb3c JA |
6581 | if (file->f_op == &io_uring_fops) { |
6582 | fput(file); | |
6b06314c JA |
6583 | break; |
6584 | } | |
6b06314c | 6585 | ret = 0; |
05f3fb3c | 6586 | table->files[index] = file; |
6b06314c JA |
6587 | } |
6588 | ||
6589 | if (ret) { | |
65e19f54 | 6590 | for (i = 0; i < ctx->nr_user_files; i++) { |
65e19f54 JA |
6591 | file = io_file_from_index(ctx, i); |
6592 | if (file) | |
6593 | fput(file); | |
6594 | } | |
6595 | for (i = 0; i < nr_tables; i++) | |
05f3fb3c | 6596 | kfree(ctx->file_data->table[i].files); |
6b06314c | 6597 | |
05f3fb3c JA |
6598 | kfree(ctx->file_data->table); |
6599 | kfree(ctx->file_data); | |
6600 | ctx->file_data = NULL; | |
6b06314c JA |
6601 | ctx->nr_user_files = 0; |
6602 | return ret; | |
6603 | } | |
6604 | ||
6605 | ret = io_sqe_files_scm(ctx); | |
05589553 | 6606 | if (ret) { |
6b06314c | 6607 | io_sqe_files_unregister(ctx); |
05589553 XW |
6608 | return ret; |
6609 | } | |
6b06314c | 6610 | |
05589553 XW |
6611 | ref_node = alloc_fixed_file_ref_node(ctx); |
6612 | if (IS_ERR(ref_node)) { | |
6613 | io_sqe_files_unregister(ctx); | |
6614 | return PTR_ERR(ref_node); | |
6615 | } | |
6616 | ||
6617 | ctx->file_data->cur_refs = &ref_node->refs; | |
6618 | spin_lock_irqsave(&ctx->file_data->lock, flags); | |
6619 | list_add(&ref_node->node, &ctx->file_data->ref_list); | |
6620 | spin_unlock_irqrestore(&ctx->file_data->lock, flags); | |
6621 | percpu_ref_get(&ctx->file_data->refs); | |
6b06314c JA |
6622 | return ret; |
6623 | } | |
6624 | ||
c3a31e60 JA |
6625 | static int io_sqe_file_register(struct io_ring_ctx *ctx, struct file *file, |
6626 | int index) | |
6627 | { | |
6628 | #if defined(CONFIG_UNIX) | |
6629 | struct sock *sock = ctx->ring_sock->sk; | |
6630 | struct sk_buff_head *head = &sock->sk_receive_queue; | |
6631 | struct sk_buff *skb; | |
6632 | ||
6633 | /* | |
6634 | * See if we can merge this file into an existing skb SCM_RIGHTS | |
6635 | * file set. If there's no room, fall back to allocating a new skb | |
6636 | * and filling it in. | |
6637 | */ | |
6638 | spin_lock_irq(&head->lock); | |
6639 | skb = skb_peek(head); | |
6640 | if (skb) { | |
6641 | struct scm_fp_list *fpl = UNIXCB(skb).fp; | |
6642 | ||
6643 | if (fpl->count < SCM_MAX_FD) { | |
6644 | __skb_unlink(skb, head); | |
6645 | spin_unlock_irq(&head->lock); | |
6646 | fpl->fp[fpl->count] = get_file(file); | |
6647 | unix_inflight(fpl->user, fpl->fp[fpl->count]); | |
6648 | fpl->count++; | |
6649 | spin_lock_irq(&head->lock); | |
6650 | __skb_queue_head(head, skb); | |
6651 | } else { | |
6652 | skb = NULL; | |
6653 | } | |
6654 | } | |
6655 | spin_unlock_irq(&head->lock); | |
6656 | ||
6657 | if (skb) { | |
6658 | fput(file); | |
6659 | return 0; | |
6660 | } | |
6661 | ||
6662 | return __io_sqe_files_scm(ctx, 1, index); | |
6663 | #else | |
6664 | return 0; | |
6665 | #endif | |
6666 | } | |
6667 | ||
a5318d3c | 6668 | static int io_queue_file_removal(struct fixed_file_data *data, |
05589553 | 6669 | struct file *file) |
05f3fb3c | 6670 | { |
a5318d3c | 6671 | struct io_file_put *pfile; |
05589553 XW |
6672 | struct percpu_ref *refs = data->cur_refs; |
6673 | struct fixed_file_ref_node *ref_node; | |
05f3fb3c | 6674 | |
05f3fb3c | 6675 | pfile = kzalloc(sizeof(*pfile), GFP_KERNEL); |
a5318d3c HD |
6676 | if (!pfile) |
6677 | return -ENOMEM; | |
05f3fb3c | 6678 | |
05589553 | 6679 | ref_node = container_of(refs, struct fixed_file_ref_node, refs); |
05f3fb3c | 6680 | pfile->file = file; |
05589553 XW |
6681 | list_add(&pfile->list, &ref_node->file_list); |
6682 | ||
a5318d3c | 6683 | return 0; |
05f3fb3c JA |
6684 | } |
6685 | ||
6686 | static int __io_sqe_files_update(struct io_ring_ctx *ctx, | |
6687 | struct io_uring_files_update *up, | |
6688 | unsigned nr_args) | |
6689 | { | |
6690 | struct fixed_file_data *data = ctx->file_data; | |
05589553 | 6691 | struct fixed_file_ref_node *ref_node; |
05f3fb3c | 6692 | struct file *file; |
c3a31e60 JA |
6693 | __s32 __user *fds; |
6694 | int fd, i, err; | |
6695 | __u32 done; | |
05589553 XW |
6696 | unsigned long flags; |
6697 | bool needs_switch = false; | |
c3a31e60 | 6698 | |
05f3fb3c | 6699 | if (check_add_overflow(up->offset, nr_args, &done)) |
c3a31e60 JA |
6700 | return -EOVERFLOW; |
6701 | if (done > ctx->nr_user_files) | |
6702 | return -EINVAL; | |
6703 | ||
05589553 XW |
6704 | ref_node = alloc_fixed_file_ref_node(ctx); |
6705 | if (IS_ERR(ref_node)) | |
6706 | return PTR_ERR(ref_node); | |
6707 | ||
c3a31e60 | 6708 | done = 0; |
05f3fb3c | 6709 | fds = u64_to_user_ptr(up->fds); |
c3a31e60 | 6710 | while (nr_args) { |
65e19f54 JA |
6711 | struct fixed_file_table *table; |
6712 | unsigned index; | |
6713 | ||
c3a31e60 JA |
6714 | err = 0; |
6715 | if (copy_from_user(&fd, &fds[done], sizeof(fd))) { | |
6716 | err = -EFAULT; | |
6717 | break; | |
6718 | } | |
05f3fb3c JA |
6719 | i = array_index_nospec(up->offset, ctx->nr_user_files); |
6720 | table = &ctx->file_data->table[i >> IORING_FILE_TABLE_SHIFT]; | |
65e19f54 JA |
6721 | index = i & IORING_FILE_TABLE_MASK; |
6722 | if (table->files[index]) { | |
05f3fb3c | 6723 | file = io_file_from_index(ctx, index); |
a5318d3c HD |
6724 | err = io_queue_file_removal(data, file); |
6725 | if (err) | |
6726 | break; | |
65e19f54 | 6727 | table->files[index] = NULL; |
05589553 | 6728 | needs_switch = true; |
c3a31e60 JA |
6729 | } |
6730 | if (fd != -1) { | |
c3a31e60 JA |
6731 | file = fget(fd); |
6732 | if (!file) { | |
6733 | err = -EBADF; | |
6734 | break; | |
6735 | } | |
6736 | /* | |
6737 | * Don't allow io_uring instances to be registered. If | |
6738 | * UNIX isn't enabled, then this causes a reference | |
6739 | * cycle and this instance can never get freed. If UNIX | |
6740 | * is enabled we'll handle it just fine, but there's | |
6741 | * still no point in allowing a ring fd as it doesn't | |
6742 | * support regular read/write anyway. | |
6743 | */ | |
6744 | if (file->f_op == &io_uring_fops) { | |
6745 | fput(file); | |
6746 | err = -EBADF; | |
6747 | break; | |
6748 | } | |
65e19f54 | 6749 | table->files[index] = file; |
c3a31e60 JA |
6750 | err = io_sqe_file_register(ctx, file, i); |
6751 | if (err) | |
6752 | break; | |
6753 | } | |
6754 | nr_args--; | |
6755 | done++; | |
05f3fb3c JA |
6756 | up->offset++; |
6757 | } | |
6758 | ||
05589553 XW |
6759 | if (needs_switch) { |
6760 | percpu_ref_kill(data->cur_refs); | |
6761 | spin_lock_irqsave(&data->lock, flags); | |
6762 | list_add(&ref_node->node, &data->ref_list); | |
6763 | data->cur_refs = &ref_node->refs; | |
6764 | spin_unlock_irqrestore(&data->lock, flags); | |
6765 | percpu_ref_get(&ctx->file_data->refs); | |
6766 | } else | |
6767 | destroy_fixed_file_ref_node(ref_node); | |
c3a31e60 JA |
6768 | |
6769 | return done ? done : err; | |
6770 | } | |
05589553 | 6771 | |
05f3fb3c JA |
6772 | static int io_sqe_files_update(struct io_ring_ctx *ctx, void __user *arg, |
6773 | unsigned nr_args) | |
6774 | { | |
6775 | struct io_uring_files_update up; | |
6776 | ||
6777 | if (!ctx->file_data) | |
6778 | return -ENXIO; | |
6779 | if (!nr_args) | |
6780 | return -EINVAL; | |
6781 | if (copy_from_user(&up, arg, sizeof(up))) | |
6782 | return -EFAULT; | |
6783 | if (up.resv) | |
6784 | return -EINVAL; | |
6785 | ||
6786 | return __io_sqe_files_update(ctx, &up, nr_args); | |
6787 | } | |
c3a31e60 | 6788 | |
e9fd9396 | 6789 | static void io_free_work(struct io_wq_work *work) |
7d723065 JA |
6790 | { |
6791 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); | |
6792 | ||
e9fd9396 | 6793 | /* Consider that io_steal_work() relies on this ref */ |
7d723065 JA |
6794 | io_put_req(req); |
6795 | } | |
6796 | ||
24369c2e PB |
6797 | static int io_init_wq_offload(struct io_ring_ctx *ctx, |
6798 | struct io_uring_params *p) | |
6799 | { | |
6800 | struct io_wq_data data; | |
6801 | struct fd f; | |
6802 | struct io_ring_ctx *ctx_attach; | |
6803 | unsigned int concurrency; | |
6804 | int ret = 0; | |
6805 | ||
6806 | data.user = ctx->user; | |
e9fd9396 | 6807 | data.free_work = io_free_work; |
24369c2e PB |
6808 | |
6809 | if (!(p->flags & IORING_SETUP_ATTACH_WQ)) { | |
6810 | /* Do QD, or 4 * CPUS, whatever is smallest */ | |
6811 | concurrency = min(ctx->sq_entries, 4 * num_online_cpus()); | |
6812 | ||
6813 | ctx->io_wq = io_wq_create(concurrency, &data); | |
6814 | if (IS_ERR(ctx->io_wq)) { | |
6815 | ret = PTR_ERR(ctx->io_wq); | |
6816 | ctx->io_wq = NULL; | |
6817 | } | |
6818 | return ret; | |
6819 | } | |
6820 | ||
6821 | f = fdget(p->wq_fd); | |
6822 | if (!f.file) | |
6823 | return -EBADF; | |
6824 | ||
6825 | if (f.file->f_op != &io_uring_fops) { | |
6826 | ret = -EINVAL; | |
6827 | goto out_fput; | |
6828 | } | |
6829 | ||
6830 | ctx_attach = f.file->private_data; | |
6831 | /* @io_wq is protected by holding the fd */ | |
6832 | if (!io_wq_get(ctx_attach->io_wq, &data)) { | |
6833 | ret = -EINVAL; | |
6834 | goto out_fput; | |
6835 | } | |
6836 | ||
6837 | ctx->io_wq = ctx_attach->io_wq; | |
6838 | out_fput: | |
6839 | fdput(f); | |
6840 | return ret; | |
6841 | } | |
6842 | ||
6c271ce2 JA |
6843 | static int io_sq_offload_start(struct io_ring_ctx *ctx, |
6844 | struct io_uring_params *p) | |
2b188cc1 JA |
6845 | { |
6846 | int ret; | |
6847 | ||
6848 | mmgrab(current->mm); | |
6849 | ctx->sqo_mm = current->mm; | |
6850 | ||
6c271ce2 | 6851 | if (ctx->flags & IORING_SETUP_SQPOLL) { |
3ec482d1 JA |
6852 | ret = -EPERM; |
6853 | if (!capable(CAP_SYS_ADMIN)) | |
6854 | goto err; | |
6855 | ||
917257da JA |
6856 | ctx->sq_thread_idle = msecs_to_jiffies(p->sq_thread_idle); |
6857 | if (!ctx->sq_thread_idle) | |
6858 | ctx->sq_thread_idle = HZ; | |
6859 | ||
6c271ce2 | 6860 | if (p->flags & IORING_SETUP_SQ_AFF) { |
44a9bd18 | 6861 | int cpu = p->sq_thread_cpu; |
6c271ce2 | 6862 | |
917257da | 6863 | ret = -EINVAL; |
44a9bd18 JA |
6864 | if (cpu >= nr_cpu_ids) |
6865 | goto err; | |
7889f44d | 6866 | if (!cpu_online(cpu)) |
917257da JA |
6867 | goto err; |
6868 | ||
6c271ce2 JA |
6869 | ctx->sqo_thread = kthread_create_on_cpu(io_sq_thread, |
6870 | ctx, cpu, | |
6871 | "io_uring-sq"); | |
6872 | } else { | |
6873 | ctx->sqo_thread = kthread_create(io_sq_thread, ctx, | |
6874 | "io_uring-sq"); | |
6875 | } | |
6876 | if (IS_ERR(ctx->sqo_thread)) { | |
6877 | ret = PTR_ERR(ctx->sqo_thread); | |
6878 | ctx->sqo_thread = NULL; | |
6879 | goto err; | |
6880 | } | |
6881 | wake_up_process(ctx->sqo_thread); | |
6882 | } else if (p->flags & IORING_SETUP_SQ_AFF) { | |
6883 | /* Can't have SQ_AFF without SQPOLL */ | |
6884 | ret = -EINVAL; | |
6885 | goto err; | |
6886 | } | |
6887 | ||
24369c2e PB |
6888 | ret = io_init_wq_offload(ctx, p); |
6889 | if (ret) | |
2b188cc1 | 6890 | goto err; |
2b188cc1 JA |
6891 | |
6892 | return 0; | |
6893 | err: | |
54a91f3b | 6894 | io_finish_async(ctx); |
2b188cc1 JA |
6895 | mmdrop(ctx->sqo_mm); |
6896 | ctx->sqo_mm = NULL; | |
6897 | return ret; | |
6898 | } | |
6899 | ||
6900 | static void io_unaccount_mem(struct user_struct *user, unsigned long nr_pages) | |
6901 | { | |
6902 | atomic_long_sub(nr_pages, &user->locked_vm); | |
6903 | } | |
6904 | ||
6905 | static int io_account_mem(struct user_struct *user, unsigned long nr_pages) | |
6906 | { | |
6907 | unsigned long page_limit, cur_pages, new_pages; | |
6908 | ||
6909 | /* Don't allow more pages than we can safely lock */ | |
6910 | page_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT; | |
6911 | ||
6912 | do { | |
6913 | cur_pages = atomic_long_read(&user->locked_vm); | |
6914 | new_pages = cur_pages + nr_pages; | |
6915 | if (new_pages > page_limit) | |
6916 | return -ENOMEM; | |
6917 | } while (atomic_long_cmpxchg(&user->locked_vm, cur_pages, | |
6918 | new_pages) != cur_pages); | |
6919 | ||
6920 | return 0; | |
6921 | } | |
6922 | ||
6923 | static void io_mem_free(void *ptr) | |
6924 | { | |
52e04ef4 MR |
6925 | struct page *page; |
6926 | ||
6927 | if (!ptr) | |
6928 | return; | |
2b188cc1 | 6929 | |
52e04ef4 | 6930 | page = virt_to_head_page(ptr); |
2b188cc1 JA |
6931 | if (put_page_testzero(page)) |
6932 | free_compound_page(page); | |
6933 | } | |
6934 | ||
6935 | static void *io_mem_alloc(size_t size) | |
6936 | { | |
6937 | gfp_t gfp_flags = GFP_KERNEL | __GFP_ZERO | __GFP_NOWARN | __GFP_COMP | | |
6938 | __GFP_NORETRY; | |
6939 | ||
6940 | return (void *) __get_free_pages(gfp_flags, get_order(size)); | |
6941 | } | |
6942 | ||
75b28aff HV |
6943 | static unsigned long rings_size(unsigned sq_entries, unsigned cq_entries, |
6944 | size_t *sq_offset) | |
6945 | { | |
6946 | struct io_rings *rings; | |
6947 | size_t off, sq_array_size; | |
6948 | ||
6949 | off = struct_size(rings, cqes, cq_entries); | |
6950 | if (off == SIZE_MAX) | |
6951 | return SIZE_MAX; | |
6952 | ||
6953 | #ifdef CONFIG_SMP | |
6954 | off = ALIGN(off, SMP_CACHE_BYTES); | |
6955 | if (off == 0) | |
6956 | return SIZE_MAX; | |
6957 | #endif | |
6958 | ||
6959 | sq_array_size = array_size(sizeof(u32), sq_entries); | |
6960 | if (sq_array_size == SIZE_MAX) | |
6961 | return SIZE_MAX; | |
6962 | ||
6963 | if (check_add_overflow(off, sq_array_size, &off)) | |
6964 | return SIZE_MAX; | |
6965 | ||
6966 | if (sq_offset) | |
6967 | *sq_offset = off; | |
6968 | ||
6969 | return off; | |
6970 | } | |
6971 | ||
2b188cc1 JA |
6972 | static unsigned long ring_pages(unsigned sq_entries, unsigned cq_entries) |
6973 | { | |
75b28aff | 6974 | size_t pages; |
2b188cc1 | 6975 | |
75b28aff HV |
6976 | pages = (size_t)1 << get_order( |
6977 | rings_size(sq_entries, cq_entries, NULL)); | |
6978 | pages += (size_t)1 << get_order( | |
6979 | array_size(sizeof(struct io_uring_sqe), sq_entries)); | |
2b188cc1 | 6980 | |
75b28aff | 6981 | return pages; |
2b188cc1 JA |
6982 | } |
6983 | ||
edafccee JA |
6984 | static int io_sqe_buffer_unregister(struct io_ring_ctx *ctx) |
6985 | { | |
6986 | int i, j; | |
6987 | ||
6988 | if (!ctx->user_bufs) | |
6989 | return -ENXIO; | |
6990 | ||
6991 | for (i = 0; i < ctx->nr_user_bufs; i++) { | |
6992 | struct io_mapped_ubuf *imu = &ctx->user_bufs[i]; | |
6993 | ||
6994 | for (j = 0; j < imu->nr_bvecs; j++) | |
f1f6a7dd | 6995 | unpin_user_page(imu->bvec[j].bv_page); |
edafccee JA |
6996 | |
6997 | if (ctx->account_mem) | |
6998 | io_unaccount_mem(ctx->user, imu->nr_bvecs); | |
d4ef6475 | 6999 | kvfree(imu->bvec); |
edafccee JA |
7000 | imu->nr_bvecs = 0; |
7001 | } | |
7002 | ||
7003 | kfree(ctx->user_bufs); | |
7004 | ctx->user_bufs = NULL; | |
7005 | ctx->nr_user_bufs = 0; | |
7006 | return 0; | |
7007 | } | |
7008 | ||
7009 | static int io_copy_iov(struct io_ring_ctx *ctx, struct iovec *dst, | |
7010 | void __user *arg, unsigned index) | |
7011 | { | |
7012 | struct iovec __user *src; | |
7013 | ||
7014 | #ifdef CONFIG_COMPAT | |
7015 | if (ctx->compat) { | |
7016 | struct compat_iovec __user *ciovs; | |
7017 | struct compat_iovec ciov; | |
7018 | ||
7019 | ciovs = (struct compat_iovec __user *) arg; | |
7020 | if (copy_from_user(&ciov, &ciovs[index], sizeof(ciov))) | |
7021 | return -EFAULT; | |
7022 | ||
d55e5f5b | 7023 | dst->iov_base = u64_to_user_ptr((u64)ciov.iov_base); |
edafccee JA |
7024 | dst->iov_len = ciov.iov_len; |
7025 | return 0; | |
7026 | } | |
7027 | #endif | |
7028 | src = (struct iovec __user *) arg; | |
7029 | if (copy_from_user(dst, &src[index], sizeof(*dst))) | |
7030 | return -EFAULT; | |
7031 | return 0; | |
7032 | } | |
7033 | ||
7034 | static int io_sqe_buffer_register(struct io_ring_ctx *ctx, void __user *arg, | |
7035 | unsigned nr_args) | |
7036 | { | |
7037 | struct vm_area_struct **vmas = NULL; | |
7038 | struct page **pages = NULL; | |
7039 | int i, j, got_pages = 0; | |
7040 | int ret = -EINVAL; | |
7041 | ||
7042 | if (ctx->user_bufs) | |
7043 | return -EBUSY; | |
7044 | if (!nr_args || nr_args > UIO_MAXIOV) | |
7045 | return -EINVAL; | |
7046 | ||
7047 | ctx->user_bufs = kcalloc(nr_args, sizeof(struct io_mapped_ubuf), | |
7048 | GFP_KERNEL); | |
7049 | if (!ctx->user_bufs) | |
7050 | return -ENOMEM; | |
7051 | ||
7052 | for (i = 0; i < nr_args; i++) { | |
7053 | struct io_mapped_ubuf *imu = &ctx->user_bufs[i]; | |
7054 | unsigned long off, start, end, ubuf; | |
7055 | int pret, nr_pages; | |
7056 | struct iovec iov; | |
7057 | size_t size; | |
7058 | ||
7059 | ret = io_copy_iov(ctx, &iov, arg, i); | |
7060 | if (ret) | |
a278682d | 7061 | goto err; |
edafccee JA |
7062 | |
7063 | /* | |
7064 | * Don't impose further limits on the size and buffer | |
7065 | * constraints here, we'll -EINVAL later when IO is | |
7066 | * submitted if they are wrong. | |
7067 | */ | |
7068 | ret = -EFAULT; | |
7069 | if (!iov.iov_base || !iov.iov_len) | |
7070 | goto err; | |
7071 | ||
7072 | /* arbitrary limit, but we need something */ | |
7073 | if (iov.iov_len > SZ_1G) | |
7074 | goto err; | |
7075 | ||
7076 | ubuf = (unsigned long) iov.iov_base; | |
7077 | end = (ubuf + iov.iov_len + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
7078 | start = ubuf >> PAGE_SHIFT; | |
7079 | nr_pages = end - start; | |
7080 | ||
7081 | if (ctx->account_mem) { | |
7082 | ret = io_account_mem(ctx->user, nr_pages); | |
7083 | if (ret) | |
7084 | goto err; | |
7085 | } | |
7086 | ||
7087 | ret = 0; | |
7088 | if (!pages || nr_pages > got_pages) { | |
7089 | kfree(vmas); | |
7090 | kfree(pages); | |
d4ef6475 | 7091 | pages = kvmalloc_array(nr_pages, sizeof(struct page *), |
edafccee | 7092 | GFP_KERNEL); |
d4ef6475 | 7093 | vmas = kvmalloc_array(nr_pages, |
edafccee JA |
7094 | sizeof(struct vm_area_struct *), |
7095 | GFP_KERNEL); | |
7096 | if (!pages || !vmas) { | |
7097 | ret = -ENOMEM; | |
7098 | if (ctx->account_mem) | |
7099 | io_unaccount_mem(ctx->user, nr_pages); | |
7100 | goto err; | |
7101 | } | |
7102 | got_pages = nr_pages; | |
7103 | } | |
7104 | ||
d4ef6475 | 7105 | imu->bvec = kvmalloc_array(nr_pages, sizeof(struct bio_vec), |
edafccee JA |
7106 | GFP_KERNEL); |
7107 | ret = -ENOMEM; | |
7108 | if (!imu->bvec) { | |
7109 | if (ctx->account_mem) | |
7110 | io_unaccount_mem(ctx->user, nr_pages); | |
7111 | goto err; | |
7112 | } | |
7113 | ||
7114 | ret = 0; | |
7115 | down_read(¤t->mm->mmap_sem); | |
2113b05d | 7116 | pret = pin_user_pages(ubuf, nr_pages, |
932f4a63 IW |
7117 | FOLL_WRITE | FOLL_LONGTERM, |
7118 | pages, vmas); | |
edafccee JA |
7119 | if (pret == nr_pages) { |
7120 | /* don't support file backed memory */ | |
7121 | for (j = 0; j < nr_pages; j++) { | |
7122 | struct vm_area_struct *vma = vmas[j]; | |
7123 | ||
7124 | if (vma->vm_file && | |
7125 | !is_file_hugepages(vma->vm_file)) { | |
7126 | ret = -EOPNOTSUPP; | |
7127 | break; | |
7128 | } | |
7129 | } | |
7130 | } else { | |
7131 | ret = pret < 0 ? pret : -EFAULT; | |
7132 | } | |
7133 | up_read(¤t->mm->mmap_sem); | |
7134 | if (ret) { | |
7135 | /* | |
7136 | * if we did partial map, or found file backed vmas, | |
7137 | * release any pages we did get | |
7138 | */ | |
27c4d3a3 | 7139 | if (pret > 0) |
f1f6a7dd | 7140 | unpin_user_pages(pages, pret); |
edafccee JA |
7141 | if (ctx->account_mem) |
7142 | io_unaccount_mem(ctx->user, nr_pages); | |
d4ef6475 | 7143 | kvfree(imu->bvec); |
edafccee JA |
7144 | goto err; |
7145 | } | |
7146 | ||
7147 | off = ubuf & ~PAGE_MASK; | |
7148 | size = iov.iov_len; | |
7149 | for (j = 0; j < nr_pages; j++) { | |
7150 | size_t vec_len; | |
7151 | ||
7152 | vec_len = min_t(size_t, size, PAGE_SIZE - off); | |
7153 | imu->bvec[j].bv_page = pages[j]; | |
7154 | imu->bvec[j].bv_len = vec_len; | |
7155 | imu->bvec[j].bv_offset = off; | |
7156 | off = 0; | |
7157 | size -= vec_len; | |
7158 | } | |
7159 | /* store original address for later verification */ | |
7160 | imu->ubuf = ubuf; | |
7161 | imu->len = iov.iov_len; | |
7162 | imu->nr_bvecs = nr_pages; | |
7163 | ||
7164 | ctx->nr_user_bufs++; | |
7165 | } | |
d4ef6475 MR |
7166 | kvfree(pages); |
7167 | kvfree(vmas); | |
edafccee JA |
7168 | return 0; |
7169 | err: | |
d4ef6475 MR |
7170 | kvfree(pages); |
7171 | kvfree(vmas); | |
edafccee JA |
7172 | io_sqe_buffer_unregister(ctx); |
7173 | return ret; | |
7174 | } | |
7175 | ||
9b402849 JA |
7176 | static int io_eventfd_register(struct io_ring_ctx *ctx, void __user *arg) |
7177 | { | |
7178 | __s32 __user *fds = arg; | |
7179 | int fd; | |
7180 | ||
7181 | if (ctx->cq_ev_fd) | |
7182 | return -EBUSY; | |
7183 | ||
7184 | if (copy_from_user(&fd, fds, sizeof(*fds))) | |
7185 | return -EFAULT; | |
7186 | ||
7187 | ctx->cq_ev_fd = eventfd_ctx_fdget(fd); | |
7188 | if (IS_ERR(ctx->cq_ev_fd)) { | |
7189 | int ret = PTR_ERR(ctx->cq_ev_fd); | |
7190 | ctx->cq_ev_fd = NULL; | |
7191 | return ret; | |
7192 | } | |
7193 | ||
7194 | return 0; | |
7195 | } | |
7196 | ||
7197 | static int io_eventfd_unregister(struct io_ring_ctx *ctx) | |
7198 | { | |
7199 | if (ctx->cq_ev_fd) { | |
7200 | eventfd_ctx_put(ctx->cq_ev_fd); | |
7201 | ctx->cq_ev_fd = NULL; | |
7202 | return 0; | |
7203 | } | |
7204 | ||
7205 | return -ENXIO; | |
7206 | } | |
7207 | ||
5a2e745d JA |
7208 | static int __io_destroy_buffers(int id, void *p, void *data) |
7209 | { | |
7210 | struct io_ring_ctx *ctx = data; | |
7211 | struct io_buffer *buf = p; | |
7212 | ||
067524e9 | 7213 | __io_remove_buffers(ctx, buf, id, -1U); |
5a2e745d JA |
7214 | return 0; |
7215 | } | |
7216 | ||
7217 | static void io_destroy_buffers(struct io_ring_ctx *ctx) | |
7218 | { | |
7219 | idr_for_each(&ctx->io_buffer_idr, __io_destroy_buffers, ctx); | |
7220 | idr_destroy(&ctx->io_buffer_idr); | |
7221 | } | |
7222 | ||
2b188cc1 JA |
7223 | static void io_ring_ctx_free(struct io_ring_ctx *ctx) |
7224 | { | |
6b06314c | 7225 | io_finish_async(ctx); |
2b188cc1 JA |
7226 | if (ctx->sqo_mm) |
7227 | mmdrop(ctx->sqo_mm); | |
def596e9 JA |
7228 | |
7229 | io_iopoll_reap_events(ctx); | |
edafccee | 7230 | io_sqe_buffer_unregister(ctx); |
6b06314c | 7231 | io_sqe_files_unregister(ctx); |
9b402849 | 7232 | io_eventfd_unregister(ctx); |
5a2e745d | 7233 | io_destroy_buffers(ctx); |
41726c9a | 7234 | idr_destroy(&ctx->personality_idr); |
def596e9 | 7235 | |
2b188cc1 | 7236 | #if defined(CONFIG_UNIX) |
355e8d26 EB |
7237 | if (ctx->ring_sock) { |
7238 | ctx->ring_sock->file = NULL; /* so that iput() is called */ | |
2b188cc1 | 7239 | sock_release(ctx->ring_sock); |
355e8d26 | 7240 | } |
2b188cc1 JA |
7241 | #endif |
7242 | ||
75b28aff | 7243 | io_mem_free(ctx->rings); |
2b188cc1 | 7244 | io_mem_free(ctx->sq_sqes); |
2b188cc1 JA |
7245 | |
7246 | percpu_ref_exit(&ctx->refs); | |
7247 | if (ctx->account_mem) | |
7248 | io_unaccount_mem(ctx->user, | |
7249 | ring_pages(ctx->sq_entries, ctx->cq_entries)); | |
7250 | free_uid(ctx->user); | |
181e448d | 7251 | put_cred(ctx->creds); |
206aefde | 7252 | kfree(ctx->completions); |
78076bb6 | 7253 | kfree(ctx->cancel_hash); |
0ddf92e8 | 7254 | kmem_cache_free(req_cachep, ctx->fallback_req); |
2b188cc1 JA |
7255 | kfree(ctx); |
7256 | } | |
7257 | ||
7258 | static __poll_t io_uring_poll(struct file *file, poll_table *wait) | |
7259 | { | |
7260 | struct io_ring_ctx *ctx = file->private_data; | |
7261 | __poll_t mask = 0; | |
7262 | ||
7263 | poll_wait(file, &ctx->cq_wait, wait); | |
4f7067c3 SB |
7264 | /* |
7265 | * synchronizes with barrier from wq_has_sleeper call in | |
7266 | * io_commit_cqring | |
7267 | */ | |
2b188cc1 | 7268 | smp_rmb(); |
75b28aff HV |
7269 | if (READ_ONCE(ctx->rings->sq.tail) - ctx->cached_sq_head != |
7270 | ctx->rings->sq_ring_entries) | |
2b188cc1 | 7271 | mask |= EPOLLOUT | EPOLLWRNORM; |
63e5d81f | 7272 | if (io_cqring_events(ctx, false)) |
2b188cc1 JA |
7273 | mask |= EPOLLIN | EPOLLRDNORM; |
7274 | ||
7275 | return mask; | |
7276 | } | |
7277 | ||
7278 | static int io_uring_fasync(int fd, struct file *file, int on) | |
7279 | { | |
7280 | struct io_ring_ctx *ctx = file->private_data; | |
7281 | ||
7282 | return fasync_helper(fd, file, on, &ctx->cq_fasync); | |
7283 | } | |
7284 | ||
071698e1 JA |
7285 | static int io_remove_personalities(int id, void *p, void *data) |
7286 | { | |
7287 | struct io_ring_ctx *ctx = data; | |
7288 | const struct cred *cred; | |
7289 | ||
7290 | cred = idr_remove(&ctx->personality_idr, id); | |
7291 | if (cred) | |
7292 | put_cred(cred); | |
7293 | return 0; | |
7294 | } | |
7295 | ||
85faa7b8 JA |
7296 | static void io_ring_exit_work(struct work_struct *work) |
7297 | { | |
7298 | struct io_ring_ctx *ctx; | |
7299 | ||
7300 | ctx = container_of(work, struct io_ring_ctx, exit_work); | |
7301 | if (ctx->rings) | |
7302 | io_cqring_overflow_flush(ctx, true); | |
7303 | ||
7304 | wait_for_completion(&ctx->completions[0]); | |
7305 | io_ring_ctx_free(ctx); | |
7306 | } | |
7307 | ||
2b188cc1 JA |
7308 | static void io_ring_ctx_wait_and_kill(struct io_ring_ctx *ctx) |
7309 | { | |
7310 | mutex_lock(&ctx->uring_lock); | |
7311 | percpu_ref_kill(&ctx->refs); | |
7312 | mutex_unlock(&ctx->uring_lock); | |
7313 | ||
df069d80 JA |
7314 | /* |
7315 | * Wait for sq thread to idle, if we have one. It won't spin on new | |
7316 | * work after we've killed the ctx ref above. This is important to do | |
7317 | * before we cancel existing commands, as the thread could otherwise | |
7318 | * be queueing new work post that. If that's work we need to cancel, | |
7319 | * it could cause shutdown to hang. | |
7320 | */ | |
7321 | while (ctx->sqo_thread && !wq_has_sleeper(&ctx->sqo_wait)) | |
3fd44c86 | 7322 | cond_resched(); |
df069d80 | 7323 | |
5262f567 | 7324 | io_kill_timeouts(ctx); |
221c5eb2 | 7325 | io_poll_remove_all(ctx); |
561fb04a JA |
7326 | |
7327 | if (ctx->io_wq) | |
7328 | io_wq_cancel_all(ctx->io_wq); | |
7329 | ||
def596e9 | 7330 | io_iopoll_reap_events(ctx); |
15dff286 JA |
7331 | /* if we failed setting up the ctx, we might not have any rings */ |
7332 | if (ctx->rings) | |
7333 | io_cqring_overflow_flush(ctx, true); | |
071698e1 | 7334 | idr_for_each(&ctx->personality_idr, io_remove_personalities, ctx); |
85faa7b8 JA |
7335 | INIT_WORK(&ctx->exit_work, io_ring_exit_work); |
7336 | queue_work(system_wq, &ctx->exit_work); | |
2b188cc1 JA |
7337 | } |
7338 | ||
7339 | static int io_uring_release(struct inode *inode, struct file *file) | |
7340 | { | |
7341 | struct io_ring_ctx *ctx = file->private_data; | |
7342 | ||
7343 | file->private_data = NULL; | |
7344 | io_ring_ctx_wait_and_kill(ctx); | |
7345 | return 0; | |
7346 | } | |
7347 | ||
fcb323cc JA |
7348 | static void io_uring_cancel_files(struct io_ring_ctx *ctx, |
7349 | struct files_struct *files) | |
7350 | { | |
fcb323cc | 7351 | while (!list_empty_careful(&ctx->inflight_list)) { |
d8f1b971 XW |
7352 | struct io_kiocb *cancel_req = NULL, *req; |
7353 | DEFINE_WAIT(wait); | |
fcb323cc JA |
7354 | |
7355 | spin_lock_irq(&ctx->inflight_lock); | |
7356 | list_for_each_entry(req, &ctx->inflight_list, inflight_entry) { | |
768134d4 JA |
7357 | if (req->work.files != files) |
7358 | continue; | |
7359 | /* req is being completed, ignore */ | |
7360 | if (!refcount_inc_not_zero(&req->refs)) | |
7361 | continue; | |
7362 | cancel_req = req; | |
7363 | break; | |
fcb323cc | 7364 | } |
768134d4 | 7365 | if (cancel_req) |
fcb323cc | 7366 | prepare_to_wait(&ctx->inflight_wait, &wait, |
768134d4 | 7367 | TASK_UNINTERRUPTIBLE); |
fcb323cc JA |
7368 | spin_unlock_irq(&ctx->inflight_lock); |
7369 | ||
768134d4 JA |
7370 | /* We need to keep going until we don't find a matching req */ |
7371 | if (!cancel_req) | |
fcb323cc | 7372 | break; |
2f6d9b9d | 7373 | |
2ca10259 JA |
7374 | if (cancel_req->flags & REQ_F_OVERFLOW) { |
7375 | spin_lock_irq(&ctx->completion_lock); | |
7376 | list_del(&cancel_req->list); | |
7377 | cancel_req->flags &= ~REQ_F_OVERFLOW; | |
7378 | if (list_empty(&ctx->cq_overflow_list)) { | |
7379 | clear_bit(0, &ctx->sq_check_overflow); | |
7380 | clear_bit(0, &ctx->cq_check_overflow); | |
7381 | } | |
7382 | spin_unlock_irq(&ctx->completion_lock); | |
7383 | ||
7384 | WRITE_ONCE(ctx->rings->cq_overflow, | |
7385 | atomic_inc_return(&ctx->cached_cq_overflow)); | |
7386 | ||
7387 | /* | |
7388 | * Put inflight ref and overflow ref. If that's | |
7389 | * all we had, then we're done with this request. | |
7390 | */ | |
7391 | if (refcount_sub_and_test(2, &cancel_req->refs)) { | |
7392 | io_put_req(cancel_req); | |
d8f1b971 | 7393 | finish_wait(&ctx->inflight_wait, &wait); |
2ca10259 JA |
7394 | continue; |
7395 | } | |
7396 | } | |
7397 | ||
2f6d9b9d BL |
7398 | io_wq_cancel_work(ctx->io_wq, &cancel_req->work); |
7399 | io_put_req(cancel_req); | |
fcb323cc | 7400 | schedule(); |
d8f1b971 | 7401 | finish_wait(&ctx->inflight_wait, &wait); |
fcb323cc JA |
7402 | } |
7403 | } | |
7404 | ||
7405 | static int io_uring_flush(struct file *file, void *data) | |
7406 | { | |
7407 | struct io_ring_ctx *ctx = file->private_data; | |
7408 | ||
7409 | io_uring_cancel_files(ctx, data); | |
6ab23144 JA |
7410 | |
7411 | /* | |
7412 | * If the task is going away, cancel work it may have pending | |
7413 | */ | |
7414 | if (fatal_signal_pending(current) || (current->flags & PF_EXITING)) | |
7415 | io_wq_cancel_pid(ctx->io_wq, task_pid_vnr(current)); | |
7416 | ||
fcb323cc JA |
7417 | return 0; |
7418 | } | |
7419 | ||
6c5c240e RP |
7420 | static void *io_uring_validate_mmap_request(struct file *file, |
7421 | loff_t pgoff, size_t sz) | |
2b188cc1 | 7422 | { |
2b188cc1 | 7423 | struct io_ring_ctx *ctx = file->private_data; |
6c5c240e | 7424 | loff_t offset = pgoff << PAGE_SHIFT; |
2b188cc1 JA |
7425 | struct page *page; |
7426 | void *ptr; | |
7427 | ||
7428 | switch (offset) { | |
7429 | case IORING_OFF_SQ_RING: | |
75b28aff HV |
7430 | case IORING_OFF_CQ_RING: |
7431 | ptr = ctx->rings; | |
2b188cc1 JA |
7432 | break; |
7433 | case IORING_OFF_SQES: | |
7434 | ptr = ctx->sq_sqes; | |
7435 | break; | |
2b188cc1 | 7436 | default: |
6c5c240e | 7437 | return ERR_PTR(-EINVAL); |
2b188cc1 JA |
7438 | } |
7439 | ||
7440 | page = virt_to_head_page(ptr); | |
a50b854e | 7441 | if (sz > page_size(page)) |
6c5c240e RP |
7442 | return ERR_PTR(-EINVAL); |
7443 | ||
7444 | return ptr; | |
7445 | } | |
7446 | ||
7447 | #ifdef CONFIG_MMU | |
7448 | ||
7449 | static int io_uring_mmap(struct file *file, struct vm_area_struct *vma) | |
7450 | { | |
7451 | size_t sz = vma->vm_end - vma->vm_start; | |
7452 | unsigned long pfn; | |
7453 | void *ptr; | |
7454 | ||
7455 | ptr = io_uring_validate_mmap_request(file, vma->vm_pgoff, sz); | |
7456 | if (IS_ERR(ptr)) | |
7457 | return PTR_ERR(ptr); | |
2b188cc1 JA |
7458 | |
7459 | pfn = virt_to_phys(ptr) >> PAGE_SHIFT; | |
7460 | return remap_pfn_range(vma, vma->vm_start, pfn, sz, vma->vm_page_prot); | |
7461 | } | |
7462 | ||
6c5c240e RP |
7463 | #else /* !CONFIG_MMU */ |
7464 | ||
7465 | static int io_uring_mmap(struct file *file, struct vm_area_struct *vma) | |
7466 | { | |
7467 | return vma->vm_flags & (VM_SHARED | VM_MAYSHARE) ? 0 : -EINVAL; | |
7468 | } | |
7469 | ||
7470 | static unsigned int io_uring_nommu_mmap_capabilities(struct file *file) | |
7471 | { | |
7472 | return NOMMU_MAP_DIRECT | NOMMU_MAP_READ | NOMMU_MAP_WRITE; | |
7473 | } | |
7474 | ||
7475 | static unsigned long io_uring_nommu_get_unmapped_area(struct file *file, | |
7476 | unsigned long addr, unsigned long len, | |
7477 | unsigned long pgoff, unsigned long flags) | |
7478 | { | |
7479 | void *ptr; | |
7480 | ||
7481 | ptr = io_uring_validate_mmap_request(file, pgoff, len); | |
7482 | if (IS_ERR(ptr)) | |
7483 | return PTR_ERR(ptr); | |
7484 | ||
7485 | return (unsigned long) ptr; | |
7486 | } | |
7487 | ||
7488 | #endif /* !CONFIG_MMU */ | |
7489 | ||
2b188cc1 JA |
7490 | SYSCALL_DEFINE6(io_uring_enter, unsigned int, fd, u32, to_submit, |
7491 | u32, min_complete, u32, flags, const sigset_t __user *, sig, | |
7492 | size_t, sigsz) | |
7493 | { | |
7494 | struct io_ring_ctx *ctx; | |
7495 | long ret = -EBADF; | |
7496 | int submitted = 0; | |
7497 | struct fd f; | |
7498 | ||
b41e9852 JA |
7499 | if (current->task_works) |
7500 | task_work_run(); | |
7501 | ||
6c271ce2 | 7502 | if (flags & ~(IORING_ENTER_GETEVENTS | IORING_ENTER_SQ_WAKEUP)) |
2b188cc1 JA |
7503 | return -EINVAL; |
7504 | ||
7505 | f = fdget(fd); | |
7506 | if (!f.file) | |
7507 | return -EBADF; | |
7508 | ||
7509 | ret = -EOPNOTSUPP; | |
7510 | if (f.file->f_op != &io_uring_fops) | |
7511 | goto out_fput; | |
7512 | ||
7513 | ret = -ENXIO; | |
7514 | ctx = f.file->private_data; | |
7515 | if (!percpu_ref_tryget(&ctx->refs)) | |
7516 | goto out_fput; | |
7517 | ||
6c271ce2 JA |
7518 | /* |
7519 | * For SQ polling, the thread will do all submissions and completions. | |
7520 | * Just return the requested submit count, and wake the thread if | |
7521 | * we were asked to. | |
7522 | */ | |
b2a9eada | 7523 | ret = 0; |
6c271ce2 | 7524 | if (ctx->flags & IORING_SETUP_SQPOLL) { |
c1edbf5f JA |
7525 | if (!list_empty_careful(&ctx->cq_overflow_list)) |
7526 | io_cqring_overflow_flush(ctx, false); | |
6c271ce2 JA |
7527 | if (flags & IORING_ENTER_SQ_WAKEUP) |
7528 | wake_up(&ctx->sqo_wait); | |
7529 | submitted = to_submit; | |
b2a9eada | 7530 | } else if (to_submit) { |
2b188cc1 | 7531 | mutex_lock(&ctx->uring_lock); |
bf9c2f1c | 7532 | submitted = io_submit_sqes(ctx, to_submit, f.file, fd, false); |
2b188cc1 | 7533 | mutex_unlock(&ctx->uring_lock); |
7c504e65 PB |
7534 | |
7535 | if (submitted != to_submit) | |
7536 | goto out; | |
2b188cc1 JA |
7537 | } |
7538 | if (flags & IORING_ENTER_GETEVENTS) { | |
def596e9 JA |
7539 | unsigned nr_events = 0; |
7540 | ||
2b188cc1 JA |
7541 | min_complete = min(min_complete, ctx->cq_entries); |
7542 | ||
32b2244a XW |
7543 | /* |
7544 | * When SETUP_IOPOLL and SETUP_SQPOLL are both enabled, user | |
7545 | * space applications don't need to do io completion events | |
7546 | * polling again, they can rely on io_sq_thread to do polling | |
7547 | * work, which can reduce cpu usage and uring_lock contention. | |
7548 | */ | |
7549 | if (ctx->flags & IORING_SETUP_IOPOLL && | |
7550 | !(ctx->flags & IORING_SETUP_SQPOLL)) { | |
def596e9 | 7551 | ret = io_iopoll_check(ctx, &nr_events, min_complete); |
def596e9 JA |
7552 | } else { |
7553 | ret = io_cqring_wait(ctx, min_complete, sig, sigsz); | |
7554 | } | |
2b188cc1 JA |
7555 | } |
7556 | ||
7c504e65 | 7557 | out: |
6805b32e | 7558 | percpu_ref_put(&ctx->refs); |
2b188cc1 JA |
7559 | out_fput: |
7560 | fdput(f); | |
7561 | return submitted ? submitted : ret; | |
7562 | } | |
7563 | ||
bebdb65e | 7564 | #ifdef CONFIG_PROC_FS |
87ce955b JA |
7565 | static int io_uring_show_cred(int id, void *p, void *data) |
7566 | { | |
7567 | const struct cred *cred = p; | |
7568 | struct seq_file *m = data; | |
7569 | struct user_namespace *uns = seq_user_ns(m); | |
7570 | struct group_info *gi; | |
7571 | kernel_cap_t cap; | |
7572 | unsigned __capi; | |
7573 | int g; | |
7574 | ||
7575 | seq_printf(m, "%5d\n", id); | |
7576 | seq_put_decimal_ull(m, "\tUid:\t", from_kuid_munged(uns, cred->uid)); | |
7577 | seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->euid)); | |
7578 | seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->suid)); | |
7579 | seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->fsuid)); | |
7580 | seq_put_decimal_ull(m, "\n\tGid:\t", from_kgid_munged(uns, cred->gid)); | |
7581 | seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->egid)); | |
7582 | seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->sgid)); | |
7583 | seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->fsgid)); | |
7584 | seq_puts(m, "\n\tGroups:\t"); | |
7585 | gi = cred->group_info; | |
7586 | for (g = 0; g < gi->ngroups; g++) { | |
7587 | seq_put_decimal_ull(m, g ? " " : "", | |
7588 | from_kgid_munged(uns, gi->gid[g])); | |
7589 | } | |
7590 | seq_puts(m, "\n\tCapEff:\t"); | |
7591 | cap = cred->cap_effective; | |
7592 | CAP_FOR_EACH_U32(__capi) | |
7593 | seq_put_hex_ll(m, NULL, cap.cap[CAP_LAST_U32 - __capi], 8); | |
7594 | seq_putc(m, '\n'); | |
7595 | return 0; | |
7596 | } | |
7597 | ||
7598 | static void __io_uring_show_fdinfo(struct io_ring_ctx *ctx, struct seq_file *m) | |
7599 | { | |
7600 | int i; | |
7601 | ||
7602 | mutex_lock(&ctx->uring_lock); | |
7603 | seq_printf(m, "UserFiles:\t%u\n", ctx->nr_user_files); | |
7604 | for (i = 0; i < ctx->nr_user_files; i++) { | |
7605 | struct fixed_file_table *table; | |
7606 | struct file *f; | |
7607 | ||
7608 | table = &ctx->file_data->table[i >> IORING_FILE_TABLE_SHIFT]; | |
7609 | f = table->files[i & IORING_FILE_TABLE_MASK]; | |
7610 | if (f) | |
7611 | seq_printf(m, "%5u: %s\n", i, file_dentry(f)->d_iname); | |
7612 | else | |
7613 | seq_printf(m, "%5u: <none>\n", i); | |
7614 | } | |
7615 | seq_printf(m, "UserBufs:\t%u\n", ctx->nr_user_bufs); | |
7616 | for (i = 0; i < ctx->nr_user_bufs; i++) { | |
7617 | struct io_mapped_ubuf *buf = &ctx->user_bufs[i]; | |
7618 | ||
7619 | seq_printf(m, "%5u: 0x%llx/%u\n", i, buf->ubuf, | |
7620 | (unsigned int) buf->len); | |
7621 | } | |
7622 | if (!idr_is_empty(&ctx->personality_idr)) { | |
7623 | seq_printf(m, "Personalities:\n"); | |
7624 | idr_for_each(&ctx->personality_idr, io_uring_show_cred, m); | |
7625 | } | |
d7718a9d JA |
7626 | seq_printf(m, "PollList:\n"); |
7627 | spin_lock_irq(&ctx->completion_lock); | |
7628 | for (i = 0; i < (1U << ctx->cancel_hash_bits); i++) { | |
7629 | struct hlist_head *list = &ctx->cancel_hash[i]; | |
7630 | struct io_kiocb *req; | |
7631 | ||
7632 | hlist_for_each_entry(req, list, hash_node) | |
7633 | seq_printf(m, " op=%d, task_works=%d\n", req->opcode, | |
7634 | req->task->task_works != NULL); | |
7635 | } | |
7636 | spin_unlock_irq(&ctx->completion_lock); | |
87ce955b JA |
7637 | mutex_unlock(&ctx->uring_lock); |
7638 | } | |
7639 | ||
7640 | static void io_uring_show_fdinfo(struct seq_file *m, struct file *f) | |
7641 | { | |
7642 | struct io_ring_ctx *ctx = f->private_data; | |
7643 | ||
7644 | if (percpu_ref_tryget(&ctx->refs)) { | |
7645 | __io_uring_show_fdinfo(ctx, m); | |
7646 | percpu_ref_put(&ctx->refs); | |
7647 | } | |
7648 | } | |
bebdb65e | 7649 | #endif |
87ce955b | 7650 | |
2b188cc1 JA |
7651 | static const struct file_operations io_uring_fops = { |
7652 | .release = io_uring_release, | |
fcb323cc | 7653 | .flush = io_uring_flush, |
2b188cc1 | 7654 | .mmap = io_uring_mmap, |
6c5c240e RP |
7655 | #ifndef CONFIG_MMU |
7656 | .get_unmapped_area = io_uring_nommu_get_unmapped_area, | |
7657 | .mmap_capabilities = io_uring_nommu_mmap_capabilities, | |
7658 | #endif | |
2b188cc1 JA |
7659 | .poll = io_uring_poll, |
7660 | .fasync = io_uring_fasync, | |
bebdb65e | 7661 | #ifdef CONFIG_PROC_FS |
87ce955b | 7662 | .show_fdinfo = io_uring_show_fdinfo, |
bebdb65e | 7663 | #endif |
2b188cc1 JA |
7664 | }; |
7665 | ||
7666 | static int io_allocate_scq_urings(struct io_ring_ctx *ctx, | |
7667 | struct io_uring_params *p) | |
7668 | { | |
75b28aff HV |
7669 | struct io_rings *rings; |
7670 | size_t size, sq_array_offset; | |
2b188cc1 | 7671 | |
75b28aff HV |
7672 | size = rings_size(p->sq_entries, p->cq_entries, &sq_array_offset); |
7673 | if (size == SIZE_MAX) | |
7674 | return -EOVERFLOW; | |
7675 | ||
7676 | rings = io_mem_alloc(size); | |
7677 | if (!rings) | |
2b188cc1 JA |
7678 | return -ENOMEM; |
7679 | ||
75b28aff HV |
7680 | ctx->rings = rings; |
7681 | ctx->sq_array = (u32 *)((char *)rings + sq_array_offset); | |
7682 | rings->sq_ring_mask = p->sq_entries - 1; | |
7683 | rings->cq_ring_mask = p->cq_entries - 1; | |
7684 | rings->sq_ring_entries = p->sq_entries; | |
7685 | rings->cq_ring_entries = p->cq_entries; | |
7686 | ctx->sq_mask = rings->sq_ring_mask; | |
7687 | ctx->cq_mask = rings->cq_ring_mask; | |
7688 | ctx->sq_entries = rings->sq_ring_entries; | |
7689 | ctx->cq_entries = rings->cq_ring_entries; | |
2b188cc1 JA |
7690 | |
7691 | size = array_size(sizeof(struct io_uring_sqe), p->sq_entries); | |
eb065d30 JA |
7692 | if (size == SIZE_MAX) { |
7693 | io_mem_free(ctx->rings); | |
7694 | ctx->rings = NULL; | |
2b188cc1 | 7695 | return -EOVERFLOW; |
eb065d30 | 7696 | } |
2b188cc1 JA |
7697 | |
7698 | ctx->sq_sqes = io_mem_alloc(size); | |
eb065d30 JA |
7699 | if (!ctx->sq_sqes) { |
7700 | io_mem_free(ctx->rings); | |
7701 | ctx->rings = NULL; | |
2b188cc1 | 7702 | return -ENOMEM; |
eb065d30 | 7703 | } |
2b188cc1 | 7704 | |
2b188cc1 JA |
7705 | return 0; |
7706 | } | |
7707 | ||
7708 | /* | |
7709 | * Allocate an anonymous fd, this is what constitutes the application | |
7710 | * visible backing of an io_uring instance. The application mmaps this | |
7711 | * fd to gain access to the SQ/CQ ring details. If UNIX sockets are enabled, | |
7712 | * we have to tie this fd to a socket for file garbage collection purposes. | |
7713 | */ | |
7714 | static int io_uring_get_fd(struct io_ring_ctx *ctx) | |
7715 | { | |
7716 | struct file *file; | |
7717 | int ret; | |
7718 | ||
7719 | #if defined(CONFIG_UNIX) | |
7720 | ret = sock_create_kern(&init_net, PF_UNIX, SOCK_RAW, IPPROTO_IP, | |
7721 | &ctx->ring_sock); | |
7722 | if (ret) | |
7723 | return ret; | |
7724 | #endif | |
7725 | ||
7726 | ret = get_unused_fd_flags(O_RDWR | O_CLOEXEC); | |
7727 | if (ret < 0) | |
7728 | goto err; | |
7729 | ||
7730 | file = anon_inode_getfile("[io_uring]", &io_uring_fops, ctx, | |
7731 | O_RDWR | O_CLOEXEC); | |
7732 | if (IS_ERR(file)) { | |
7733 | put_unused_fd(ret); | |
7734 | ret = PTR_ERR(file); | |
7735 | goto err; | |
7736 | } | |
7737 | ||
7738 | #if defined(CONFIG_UNIX) | |
7739 | ctx->ring_sock->file = file; | |
7740 | #endif | |
7741 | fd_install(ret, file); | |
7742 | return ret; | |
7743 | err: | |
7744 | #if defined(CONFIG_UNIX) | |
7745 | sock_release(ctx->ring_sock); | |
7746 | ctx->ring_sock = NULL; | |
7747 | #endif | |
7748 | return ret; | |
7749 | } | |
7750 | ||
7f13657d XW |
7751 | static int io_uring_create(unsigned entries, struct io_uring_params *p, |
7752 | struct io_uring_params __user *params) | |
2b188cc1 JA |
7753 | { |
7754 | struct user_struct *user = NULL; | |
7755 | struct io_ring_ctx *ctx; | |
7756 | bool account_mem; | |
7757 | int ret; | |
7758 | ||
8110c1a6 | 7759 | if (!entries) |
2b188cc1 | 7760 | return -EINVAL; |
8110c1a6 JA |
7761 | if (entries > IORING_MAX_ENTRIES) { |
7762 | if (!(p->flags & IORING_SETUP_CLAMP)) | |
7763 | return -EINVAL; | |
7764 | entries = IORING_MAX_ENTRIES; | |
7765 | } | |
2b188cc1 JA |
7766 | |
7767 | /* | |
7768 | * Use twice as many entries for the CQ ring. It's possible for the | |
7769 | * application to drive a higher depth than the size of the SQ ring, | |
7770 | * since the sqes are only used at submission time. This allows for | |
33a107f0 JA |
7771 | * some flexibility in overcommitting a bit. If the application has |
7772 | * set IORING_SETUP_CQSIZE, it will have passed in the desired number | |
7773 | * of CQ ring entries manually. | |
2b188cc1 JA |
7774 | */ |
7775 | p->sq_entries = roundup_pow_of_two(entries); | |
33a107f0 JA |
7776 | if (p->flags & IORING_SETUP_CQSIZE) { |
7777 | /* | |
7778 | * If IORING_SETUP_CQSIZE is set, we do the same roundup | |
7779 | * to a power-of-two, if it isn't already. We do NOT impose | |
7780 | * any cq vs sq ring sizing. | |
7781 | */ | |
8110c1a6 | 7782 | if (p->cq_entries < p->sq_entries) |
33a107f0 | 7783 | return -EINVAL; |
8110c1a6 JA |
7784 | if (p->cq_entries > IORING_MAX_CQ_ENTRIES) { |
7785 | if (!(p->flags & IORING_SETUP_CLAMP)) | |
7786 | return -EINVAL; | |
7787 | p->cq_entries = IORING_MAX_CQ_ENTRIES; | |
7788 | } | |
33a107f0 JA |
7789 | p->cq_entries = roundup_pow_of_two(p->cq_entries); |
7790 | } else { | |
7791 | p->cq_entries = 2 * p->sq_entries; | |
7792 | } | |
2b188cc1 JA |
7793 | |
7794 | user = get_uid(current_user()); | |
7795 | account_mem = !capable(CAP_IPC_LOCK); | |
7796 | ||
7797 | if (account_mem) { | |
7798 | ret = io_account_mem(user, | |
7799 | ring_pages(p->sq_entries, p->cq_entries)); | |
7800 | if (ret) { | |
7801 | free_uid(user); | |
7802 | return ret; | |
7803 | } | |
7804 | } | |
7805 | ||
7806 | ctx = io_ring_ctx_alloc(p); | |
7807 | if (!ctx) { | |
7808 | if (account_mem) | |
7809 | io_unaccount_mem(user, ring_pages(p->sq_entries, | |
7810 | p->cq_entries)); | |
7811 | free_uid(user); | |
7812 | return -ENOMEM; | |
7813 | } | |
7814 | ctx->compat = in_compat_syscall(); | |
7815 | ctx->account_mem = account_mem; | |
7816 | ctx->user = user; | |
0b8c0ec7 | 7817 | ctx->creds = get_current_cred(); |
2b188cc1 JA |
7818 | |
7819 | ret = io_allocate_scq_urings(ctx, p); | |
7820 | if (ret) | |
7821 | goto err; | |
7822 | ||
6c271ce2 | 7823 | ret = io_sq_offload_start(ctx, p); |
2b188cc1 JA |
7824 | if (ret) |
7825 | goto err; | |
7826 | ||
2b188cc1 | 7827 | memset(&p->sq_off, 0, sizeof(p->sq_off)); |
75b28aff HV |
7828 | p->sq_off.head = offsetof(struct io_rings, sq.head); |
7829 | p->sq_off.tail = offsetof(struct io_rings, sq.tail); | |
7830 | p->sq_off.ring_mask = offsetof(struct io_rings, sq_ring_mask); | |
7831 | p->sq_off.ring_entries = offsetof(struct io_rings, sq_ring_entries); | |
7832 | p->sq_off.flags = offsetof(struct io_rings, sq_flags); | |
7833 | p->sq_off.dropped = offsetof(struct io_rings, sq_dropped); | |
7834 | p->sq_off.array = (char *)ctx->sq_array - (char *)ctx->rings; | |
2b188cc1 JA |
7835 | |
7836 | memset(&p->cq_off, 0, sizeof(p->cq_off)); | |
75b28aff HV |
7837 | p->cq_off.head = offsetof(struct io_rings, cq.head); |
7838 | p->cq_off.tail = offsetof(struct io_rings, cq.tail); | |
7839 | p->cq_off.ring_mask = offsetof(struct io_rings, cq_ring_mask); | |
7840 | p->cq_off.ring_entries = offsetof(struct io_rings, cq_ring_entries); | |
7841 | p->cq_off.overflow = offsetof(struct io_rings, cq_overflow); | |
7842 | p->cq_off.cqes = offsetof(struct io_rings, cqes); | |
ac90f249 | 7843 | |
7f13657d XW |
7844 | p->features = IORING_FEAT_SINGLE_MMAP | IORING_FEAT_NODROP | |
7845 | IORING_FEAT_SUBMIT_STABLE | IORING_FEAT_RW_CUR_POS | | |
7846 | IORING_FEAT_CUR_PERSONALITY | IORING_FEAT_FAST_POLL; | |
7847 | ||
7848 | if (copy_to_user(params, p, sizeof(*p))) { | |
7849 | ret = -EFAULT; | |
7850 | goto err; | |
7851 | } | |
044c1ab3 JA |
7852 | /* |
7853 | * Install ring fd as the very last thing, so we don't risk someone | |
7854 | * having closed it before we finish setup | |
7855 | */ | |
7856 | ret = io_uring_get_fd(ctx); | |
7857 | if (ret < 0) | |
7858 | goto err; | |
7859 | ||
c826bd7a | 7860 | trace_io_uring_create(ret, ctx, p->sq_entries, p->cq_entries, p->flags); |
2b188cc1 JA |
7861 | return ret; |
7862 | err: | |
7863 | io_ring_ctx_wait_and_kill(ctx); | |
7864 | return ret; | |
7865 | } | |
7866 | ||
7867 | /* | |
7868 | * Sets up an aio uring context, and returns the fd. Applications asks for a | |
7869 | * ring size, we return the actual sq/cq ring sizes (among other things) in the | |
7870 | * params structure passed in. | |
7871 | */ | |
7872 | static long io_uring_setup(u32 entries, struct io_uring_params __user *params) | |
7873 | { | |
7874 | struct io_uring_params p; | |
2b188cc1 JA |
7875 | int i; |
7876 | ||
7877 | if (copy_from_user(&p, params, sizeof(p))) | |
7878 | return -EFAULT; | |
7879 | for (i = 0; i < ARRAY_SIZE(p.resv); i++) { | |
7880 | if (p.resv[i]) | |
7881 | return -EINVAL; | |
7882 | } | |
7883 | ||
6c271ce2 | 7884 | if (p.flags & ~(IORING_SETUP_IOPOLL | IORING_SETUP_SQPOLL | |
8110c1a6 | 7885 | IORING_SETUP_SQ_AFF | IORING_SETUP_CQSIZE | |
24369c2e | 7886 | IORING_SETUP_CLAMP | IORING_SETUP_ATTACH_WQ)) |
2b188cc1 JA |
7887 | return -EINVAL; |
7888 | ||
7f13657d | 7889 | return io_uring_create(entries, &p, params); |
2b188cc1 JA |
7890 | } |
7891 | ||
7892 | SYSCALL_DEFINE2(io_uring_setup, u32, entries, | |
7893 | struct io_uring_params __user *, params) | |
7894 | { | |
7895 | return io_uring_setup(entries, params); | |
7896 | } | |
7897 | ||
66f4af93 JA |
7898 | static int io_probe(struct io_ring_ctx *ctx, void __user *arg, unsigned nr_args) |
7899 | { | |
7900 | struct io_uring_probe *p; | |
7901 | size_t size; | |
7902 | int i, ret; | |
7903 | ||
7904 | size = struct_size(p, ops, nr_args); | |
7905 | if (size == SIZE_MAX) | |
7906 | return -EOVERFLOW; | |
7907 | p = kzalloc(size, GFP_KERNEL); | |
7908 | if (!p) | |
7909 | return -ENOMEM; | |
7910 | ||
7911 | ret = -EFAULT; | |
7912 | if (copy_from_user(p, arg, size)) | |
7913 | goto out; | |
7914 | ret = -EINVAL; | |
7915 | if (memchr_inv(p, 0, size)) | |
7916 | goto out; | |
7917 | ||
7918 | p->last_op = IORING_OP_LAST - 1; | |
7919 | if (nr_args > IORING_OP_LAST) | |
7920 | nr_args = IORING_OP_LAST; | |
7921 | ||
7922 | for (i = 0; i < nr_args; i++) { | |
7923 | p->ops[i].op = i; | |
7924 | if (!io_op_defs[i].not_supported) | |
7925 | p->ops[i].flags = IO_URING_OP_SUPPORTED; | |
7926 | } | |
7927 | p->ops_len = i; | |
7928 | ||
7929 | ret = 0; | |
7930 | if (copy_to_user(arg, p, size)) | |
7931 | ret = -EFAULT; | |
7932 | out: | |
7933 | kfree(p); | |
7934 | return ret; | |
7935 | } | |
7936 | ||
071698e1 JA |
7937 | static int io_register_personality(struct io_ring_ctx *ctx) |
7938 | { | |
7939 | const struct cred *creds = get_current_cred(); | |
7940 | int id; | |
7941 | ||
7942 | id = idr_alloc_cyclic(&ctx->personality_idr, (void *) creds, 1, | |
7943 | USHRT_MAX, GFP_KERNEL); | |
7944 | if (id < 0) | |
7945 | put_cred(creds); | |
7946 | return id; | |
7947 | } | |
7948 | ||
7949 | static int io_unregister_personality(struct io_ring_ctx *ctx, unsigned id) | |
7950 | { | |
7951 | const struct cred *old_creds; | |
7952 | ||
7953 | old_creds = idr_remove(&ctx->personality_idr, id); | |
7954 | if (old_creds) { | |
7955 | put_cred(old_creds); | |
7956 | return 0; | |
7957 | } | |
7958 | ||
7959 | return -EINVAL; | |
7960 | } | |
7961 | ||
7962 | static bool io_register_op_must_quiesce(int op) | |
7963 | { | |
7964 | switch (op) { | |
7965 | case IORING_UNREGISTER_FILES: | |
7966 | case IORING_REGISTER_FILES_UPDATE: | |
7967 | case IORING_REGISTER_PROBE: | |
7968 | case IORING_REGISTER_PERSONALITY: | |
7969 | case IORING_UNREGISTER_PERSONALITY: | |
7970 | return false; | |
7971 | default: | |
7972 | return true; | |
7973 | } | |
7974 | } | |
7975 | ||
edafccee JA |
7976 | static int __io_uring_register(struct io_ring_ctx *ctx, unsigned opcode, |
7977 | void __user *arg, unsigned nr_args) | |
b19062a5 JA |
7978 | __releases(ctx->uring_lock) |
7979 | __acquires(ctx->uring_lock) | |
edafccee JA |
7980 | { |
7981 | int ret; | |
7982 | ||
35fa71a0 JA |
7983 | /* |
7984 | * We're inside the ring mutex, if the ref is already dying, then | |
7985 | * someone else killed the ctx or is already going through | |
7986 | * io_uring_register(). | |
7987 | */ | |
7988 | if (percpu_ref_is_dying(&ctx->refs)) | |
7989 | return -ENXIO; | |
7990 | ||
071698e1 | 7991 | if (io_register_op_must_quiesce(opcode)) { |
05f3fb3c | 7992 | percpu_ref_kill(&ctx->refs); |
b19062a5 | 7993 | |
05f3fb3c JA |
7994 | /* |
7995 | * Drop uring mutex before waiting for references to exit. If | |
7996 | * another thread is currently inside io_uring_enter() it might | |
7997 | * need to grab the uring_lock to make progress. If we hold it | |
7998 | * here across the drain wait, then we can deadlock. It's safe | |
7999 | * to drop the mutex here, since no new references will come in | |
8000 | * after we've killed the percpu ref. | |
8001 | */ | |
8002 | mutex_unlock(&ctx->uring_lock); | |
c150368b | 8003 | ret = wait_for_completion_interruptible(&ctx->completions[0]); |
05f3fb3c | 8004 | mutex_lock(&ctx->uring_lock); |
c150368b JA |
8005 | if (ret) { |
8006 | percpu_ref_resurrect(&ctx->refs); | |
8007 | ret = -EINTR; | |
8008 | goto out; | |
8009 | } | |
05f3fb3c | 8010 | } |
edafccee JA |
8011 | |
8012 | switch (opcode) { | |
8013 | case IORING_REGISTER_BUFFERS: | |
8014 | ret = io_sqe_buffer_register(ctx, arg, nr_args); | |
8015 | break; | |
8016 | case IORING_UNREGISTER_BUFFERS: | |
8017 | ret = -EINVAL; | |
8018 | if (arg || nr_args) | |
8019 | break; | |
8020 | ret = io_sqe_buffer_unregister(ctx); | |
8021 | break; | |
6b06314c JA |
8022 | case IORING_REGISTER_FILES: |
8023 | ret = io_sqe_files_register(ctx, arg, nr_args); | |
8024 | break; | |
8025 | case IORING_UNREGISTER_FILES: | |
8026 | ret = -EINVAL; | |
8027 | if (arg || nr_args) | |
8028 | break; | |
8029 | ret = io_sqe_files_unregister(ctx); | |
8030 | break; | |
c3a31e60 JA |
8031 | case IORING_REGISTER_FILES_UPDATE: |
8032 | ret = io_sqe_files_update(ctx, arg, nr_args); | |
8033 | break; | |
9b402849 | 8034 | case IORING_REGISTER_EVENTFD: |
f2842ab5 | 8035 | case IORING_REGISTER_EVENTFD_ASYNC: |
9b402849 JA |
8036 | ret = -EINVAL; |
8037 | if (nr_args != 1) | |
8038 | break; | |
8039 | ret = io_eventfd_register(ctx, arg); | |
f2842ab5 JA |
8040 | if (ret) |
8041 | break; | |
8042 | if (opcode == IORING_REGISTER_EVENTFD_ASYNC) | |
8043 | ctx->eventfd_async = 1; | |
8044 | else | |
8045 | ctx->eventfd_async = 0; | |
9b402849 JA |
8046 | break; |
8047 | case IORING_UNREGISTER_EVENTFD: | |
8048 | ret = -EINVAL; | |
8049 | if (arg || nr_args) | |
8050 | break; | |
8051 | ret = io_eventfd_unregister(ctx); | |
8052 | break; | |
66f4af93 JA |
8053 | case IORING_REGISTER_PROBE: |
8054 | ret = -EINVAL; | |
8055 | if (!arg || nr_args > 256) | |
8056 | break; | |
8057 | ret = io_probe(ctx, arg, nr_args); | |
8058 | break; | |
071698e1 JA |
8059 | case IORING_REGISTER_PERSONALITY: |
8060 | ret = -EINVAL; | |
8061 | if (arg || nr_args) | |
8062 | break; | |
8063 | ret = io_register_personality(ctx); | |
8064 | break; | |
8065 | case IORING_UNREGISTER_PERSONALITY: | |
8066 | ret = -EINVAL; | |
8067 | if (arg) | |
8068 | break; | |
8069 | ret = io_unregister_personality(ctx, nr_args); | |
8070 | break; | |
edafccee JA |
8071 | default: |
8072 | ret = -EINVAL; | |
8073 | break; | |
8074 | } | |
8075 | ||
071698e1 | 8076 | if (io_register_op_must_quiesce(opcode)) { |
05f3fb3c | 8077 | /* bring the ctx back to life */ |
05f3fb3c | 8078 | percpu_ref_reinit(&ctx->refs); |
c150368b JA |
8079 | out: |
8080 | reinit_completion(&ctx->completions[0]); | |
05f3fb3c | 8081 | } |
edafccee JA |
8082 | return ret; |
8083 | } | |
8084 | ||
8085 | SYSCALL_DEFINE4(io_uring_register, unsigned int, fd, unsigned int, opcode, | |
8086 | void __user *, arg, unsigned int, nr_args) | |
8087 | { | |
8088 | struct io_ring_ctx *ctx; | |
8089 | long ret = -EBADF; | |
8090 | struct fd f; | |
8091 | ||
8092 | f = fdget(fd); | |
8093 | if (!f.file) | |
8094 | return -EBADF; | |
8095 | ||
8096 | ret = -EOPNOTSUPP; | |
8097 | if (f.file->f_op != &io_uring_fops) | |
8098 | goto out_fput; | |
8099 | ||
8100 | ctx = f.file->private_data; | |
8101 | ||
8102 | mutex_lock(&ctx->uring_lock); | |
8103 | ret = __io_uring_register(ctx, opcode, arg, nr_args); | |
8104 | mutex_unlock(&ctx->uring_lock); | |
c826bd7a DD |
8105 | trace_io_uring_register(ctx, opcode, ctx->nr_user_files, ctx->nr_user_bufs, |
8106 | ctx->cq_ev_fd != NULL, ret); | |
edafccee JA |
8107 | out_fput: |
8108 | fdput(f); | |
8109 | return ret; | |
8110 | } | |
8111 | ||
2b188cc1 JA |
8112 | static int __init io_uring_init(void) |
8113 | { | |
d7f62e82 SM |
8114 | #define __BUILD_BUG_VERIFY_ELEMENT(stype, eoffset, etype, ename) do { \ |
8115 | BUILD_BUG_ON(offsetof(stype, ename) != eoffset); \ | |
8116 | BUILD_BUG_ON(sizeof(etype) != sizeof_field(stype, ename)); \ | |
8117 | } while (0) | |
8118 | ||
8119 | #define BUILD_BUG_SQE_ELEM(eoffset, etype, ename) \ | |
8120 | __BUILD_BUG_VERIFY_ELEMENT(struct io_uring_sqe, eoffset, etype, ename) | |
8121 | BUILD_BUG_ON(sizeof(struct io_uring_sqe) != 64); | |
8122 | BUILD_BUG_SQE_ELEM(0, __u8, opcode); | |
8123 | BUILD_BUG_SQE_ELEM(1, __u8, flags); | |
8124 | BUILD_BUG_SQE_ELEM(2, __u16, ioprio); | |
8125 | BUILD_BUG_SQE_ELEM(4, __s32, fd); | |
8126 | BUILD_BUG_SQE_ELEM(8, __u64, off); | |
8127 | BUILD_BUG_SQE_ELEM(8, __u64, addr2); | |
8128 | BUILD_BUG_SQE_ELEM(16, __u64, addr); | |
7d67af2c | 8129 | BUILD_BUG_SQE_ELEM(16, __u64, splice_off_in); |
d7f62e82 SM |
8130 | BUILD_BUG_SQE_ELEM(24, __u32, len); |
8131 | BUILD_BUG_SQE_ELEM(28, __kernel_rwf_t, rw_flags); | |
8132 | BUILD_BUG_SQE_ELEM(28, /* compat */ int, rw_flags); | |
8133 | BUILD_BUG_SQE_ELEM(28, /* compat */ __u32, rw_flags); | |
8134 | BUILD_BUG_SQE_ELEM(28, __u32, fsync_flags); | |
8135 | BUILD_BUG_SQE_ELEM(28, __u16, poll_events); | |
8136 | BUILD_BUG_SQE_ELEM(28, __u32, sync_range_flags); | |
8137 | BUILD_BUG_SQE_ELEM(28, __u32, msg_flags); | |
8138 | BUILD_BUG_SQE_ELEM(28, __u32, timeout_flags); | |
8139 | BUILD_BUG_SQE_ELEM(28, __u32, accept_flags); | |
8140 | BUILD_BUG_SQE_ELEM(28, __u32, cancel_flags); | |
8141 | BUILD_BUG_SQE_ELEM(28, __u32, open_flags); | |
8142 | BUILD_BUG_SQE_ELEM(28, __u32, statx_flags); | |
8143 | BUILD_BUG_SQE_ELEM(28, __u32, fadvise_advice); | |
7d67af2c | 8144 | BUILD_BUG_SQE_ELEM(28, __u32, splice_flags); |
d7f62e82 SM |
8145 | BUILD_BUG_SQE_ELEM(32, __u64, user_data); |
8146 | BUILD_BUG_SQE_ELEM(40, __u16, buf_index); | |
8147 | BUILD_BUG_SQE_ELEM(42, __u16, personality); | |
7d67af2c | 8148 | BUILD_BUG_SQE_ELEM(44, __s32, splice_fd_in); |
d7f62e82 | 8149 | |
d3656344 | 8150 | BUILD_BUG_ON(ARRAY_SIZE(io_op_defs) != IORING_OP_LAST); |
84557871 | 8151 | BUILD_BUG_ON(__REQ_F_LAST_BIT >= 8 * sizeof(int)); |
2b188cc1 JA |
8152 | req_cachep = KMEM_CACHE(io_kiocb, SLAB_HWCACHE_ALIGN | SLAB_PANIC); |
8153 | return 0; | |
8154 | }; | |
8155 | __initcall(io_uring_init); |