]>
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; |
2b188cc1 JA |
329 | }; |
330 | ||
09bb8394 JA |
331 | /* |
332 | * First field must be the file pointer in all the | |
333 | * iocb unions! See also 'struct kiocb' in <linux/fs.h> | |
334 | */ | |
221c5eb2 JA |
335 | struct io_poll_iocb { |
336 | struct file *file; | |
0969e783 JA |
337 | union { |
338 | struct wait_queue_head *head; | |
339 | u64 addr; | |
340 | }; | |
221c5eb2 | 341 | __poll_t events; |
8c838788 | 342 | bool done; |
221c5eb2 | 343 | bool canceled; |
392edb45 | 344 | struct wait_queue_entry wait; |
221c5eb2 JA |
345 | }; |
346 | ||
b5dba59e JA |
347 | struct io_close { |
348 | struct file *file; | |
349 | struct file *put_file; | |
350 | int fd; | |
351 | }; | |
352 | ||
ad8a48ac JA |
353 | struct io_timeout_data { |
354 | struct io_kiocb *req; | |
355 | struct hrtimer timer; | |
356 | struct timespec64 ts; | |
357 | enum hrtimer_mode mode; | |
cc42e0ac | 358 | u32 seq_offset; |
ad8a48ac JA |
359 | }; |
360 | ||
8ed8d3c3 JA |
361 | struct io_accept { |
362 | struct file *file; | |
363 | struct sockaddr __user *addr; | |
364 | int __user *addr_len; | |
365 | int flags; | |
09952e3e | 366 | unsigned long nofile; |
8ed8d3c3 JA |
367 | }; |
368 | ||
369 | struct io_sync { | |
370 | struct file *file; | |
371 | loff_t len; | |
372 | loff_t off; | |
373 | int flags; | |
d63d1b5e | 374 | int mode; |
8ed8d3c3 JA |
375 | }; |
376 | ||
fbf23849 JA |
377 | struct io_cancel { |
378 | struct file *file; | |
379 | u64 addr; | |
380 | }; | |
381 | ||
b29472ee JA |
382 | struct io_timeout { |
383 | struct file *file; | |
384 | u64 addr; | |
385 | int flags; | |
26a61679 | 386 | unsigned count; |
b29472ee JA |
387 | }; |
388 | ||
9adbd45d JA |
389 | struct io_rw { |
390 | /* NOTE: kiocb has the file as the first member, so don't do it here */ | |
391 | struct kiocb kiocb; | |
392 | u64 addr; | |
393 | u64 len; | |
394 | }; | |
395 | ||
3fbb51c1 JA |
396 | struct io_connect { |
397 | struct file *file; | |
398 | struct sockaddr __user *addr; | |
399 | int addr_len; | |
400 | }; | |
401 | ||
e47293fd JA |
402 | struct io_sr_msg { |
403 | struct file *file; | |
fddaface JA |
404 | union { |
405 | struct user_msghdr __user *msg; | |
406 | void __user *buf; | |
407 | }; | |
e47293fd | 408 | int msg_flags; |
bcda7baa | 409 | int bgid; |
fddaface | 410 | size_t len; |
bcda7baa | 411 | struct io_buffer *kbuf; |
e47293fd JA |
412 | }; |
413 | ||
15b71abe JA |
414 | struct io_open { |
415 | struct file *file; | |
416 | int dfd; | |
eddc7ef5 | 417 | union { |
eddc7ef5 JA |
418 | unsigned mask; |
419 | }; | |
15b71abe | 420 | struct filename *filename; |
eddc7ef5 | 421 | struct statx __user *buffer; |
c12cedf2 | 422 | struct open_how how; |
4022e7af | 423 | unsigned long nofile; |
15b71abe JA |
424 | }; |
425 | ||
05f3fb3c JA |
426 | struct io_files_update { |
427 | struct file *file; | |
428 | u64 arg; | |
429 | u32 nr_args; | |
430 | u32 offset; | |
431 | }; | |
432 | ||
4840e418 JA |
433 | struct io_fadvise { |
434 | struct file *file; | |
435 | u64 offset; | |
436 | u32 len; | |
437 | u32 advice; | |
438 | }; | |
439 | ||
c1ca757b JA |
440 | struct io_madvise { |
441 | struct file *file; | |
442 | u64 addr; | |
443 | u32 len; | |
444 | u32 advice; | |
445 | }; | |
446 | ||
3e4827b0 JA |
447 | struct io_epoll { |
448 | struct file *file; | |
449 | int epfd; | |
450 | int op; | |
451 | int fd; | |
452 | struct epoll_event event; | |
e47293fd JA |
453 | }; |
454 | ||
7d67af2c PB |
455 | struct io_splice { |
456 | struct file *file_out; | |
457 | struct file *file_in; | |
458 | loff_t off_out; | |
459 | loff_t off_in; | |
460 | u64 len; | |
461 | unsigned int flags; | |
462 | }; | |
463 | ||
ddf0322d JA |
464 | struct io_provide_buf { |
465 | struct file *file; | |
466 | __u64 addr; | |
467 | __s32 len; | |
468 | __u32 bgid; | |
469 | __u16 nbufs; | |
470 | __u16 bid; | |
471 | }; | |
472 | ||
f499a021 JA |
473 | struct io_async_connect { |
474 | struct sockaddr_storage address; | |
475 | }; | |
476 | ||
03b1230c JA |
477 | struct io_async_msghdr { |
478 | struct iovec fast_iov[UIO_FASTIOV]; | |
479 | struct iovec *iov; | |
480 | struct sockaddr __user *uaddr; | |
481 | struct msghdr msg; | |
b537916c | 482 | struct sockaddr_storage addr; |
03b1230c JA |
483 | }; |
484 | ||
f67676d1 JA |
485 | struct io_async_rw { |
486 | struct iovec fast_iov[UIO_FASTIOV]; | |
487 | struct iovec *iov; | |
488 | ssize_t nr_segs; | |
489 | ssize_t size; | |
490 | }; | |
491 | ||
1a6b74fc | 492 | struct io_async_ctx { |
f67676d1 JA |
493 | union { |
494 | struct io_async_rw rw; | |
03b1230c | 495 | struct io_async_msghdr msg; |
f499a021 | 496 | struct io_async_connect connect; |
2d28390a | 497 | struct io_timeout_data timeout; |
f67676d1 | 498 | }; |
1a6b74fc JA |
499 | }; |
500 | ||
6b47ee6e PB |
501 | enum { |
502 | REQ_F_FIXED_FILE_BIT = IOSQE_FIXED_FILE_BIT, | |
503 | REQ_F_IO_DRAIN_BIT = IOSQE_IO_DRAIN_BIT, | |
504 | REQ_F_LINK_BIT = IOSQE_IO_LINK_BIT, | |
505 | REQ_F_HARDLINK_BIT = IOSQE_IO_HARDLINK_BIT, | |
506 | REQ_F_FORCE_ASYNC_BIT = IOSQE_ASYNC_BIT, | |
bcda7baa | 507 | REQ_F_BUFFER_SELECT_BIT = IOSQE_BUFFER_SELECT_BIT, |
6b47ee6e PB |
508 | |
509 | REQ_F_LINK_NEXT_BIT, | |
510 | REQ_F_FAIL_LINK_BIT, | |
511 | REQ_F_INFLIGHT_BIT, | |
512 | REQ_F_CUR_POS_BIT, | |
513 | REQ_F_NOWAIT_BIT, | |
514 | REQ_F_IOPOLL_COMPLETED_BIT, | |
515 | REQ_F_LINK_TIMEOUT_BIT, | |
516 | REQ_F_TIMEOUT_BIT, | |
517 | REQ_F_ISREG_BIT, | |
518 | REQ_F_MUST_PUNT_BIT, | |
519 | REQ_F_TIMEOUT_NOSEQ_BIT, | |
520 | REQ_F_COMP_LOCKED_BIT, | |
99bc4c38 | 521 | REQ_F_NEED_CLEANUP_BIT, |
2ca10259 | 522 | REQ_F_OVERFLOW_BIT, |
d7718a9d | 523 | REQ_F_POLLED_BIT, |
bcda7baa | 524 | REQ_F_BUFFER_SELECTED_BIT, |
84557871 JA |
525 | |
526 | /* not a real bit, just to check we're not overflowing the space */ | |
527 | __REQ_F_LAST_BIT, | |
6b47ee6e PB |
528 | }; |
529 | ||
530 | enum { | |
531 | /* ctx owns file */ | |
532 | REQ_F_FIXED_FILE = BIT(REQ_F_FIXED_FILE_BIT), | |
533 | /* drain existing IO first */ | |
534 | REQ_F_IO_DRAIN = BIT(REQ_F_IO_DRAIN_BIT), | |
535 | /* linked sqes */ | |
536 | REQ_F_LINK = BIT(REQ_F_LINK_BIT), | |
537 | /* doesn't sever on completion < 0 */ | |
538 | REQ_F_HARDLINK = BIT(REQ_F_HARDLINK_BIT), | |
539 | /* IOSQE_ASYNC */ | |
540 | REQ_F_FORCE_ASYNC = BIT(REQ_F_FORCE_ASYNC_BIT), | |
bcda7baa JA |
541 | /* IOSQE_BUFFER_SELECT */ |
542 | REQ_F_BUFFER_SELECT = BIT(REQ_F_BUFFER_SELECT_BIT), | |
6b47ee6e PB |
543 | |
544 | /* already grabbed next link */ | |
545 | REQ_F_LINK_NEXT = BIT(REQ_F_LINK_NEXT_BIT), | |
546 | /* fail rest of links */ | |
547 | REQ_F_FAIL_LINK = BIT(REQ_F_FAIL_LINK_BIT), | |
548 | /* on inflight list */ | |
549 | REQ_F_INFLIGHT = BIT(REQ_F_INFLIGHT_BIT), | |
550 | /* read/write uses file position */ | |
551 | REQ_F_CUR_POS = BIT(REQ_F_CUR_POS_BIT), | |
552 | /* must not punt to workers */ | |
553 | REQ_F_NOWAIT = BIT(REQ_F_NOWAIT_BIT), | |
554 | /* polled IO has completed */ | |
555 | REQ_F_IOPOLL_COMPLETED = BIT(REQ_F_IOPOLL_COMPLETED_BIT), | |
556 | /* has linked timeout */ | |
557 | REQ_F_LINK_TIMEOUT = BIT(REQ_F_LINK_TIMEOUT_BIT), | |
558 | /* timeout request */ | |
559 | REQ_F_TIMEOUT = BIT(REQ_F_TIMEOUT_BIT), | |
560 | /* regular file */ | |
561 | REQ_F_ISREG = BIT(REQ_F_ISREG_BIT), | |
562 | /* must be punted even for NONBLOCK */ | |
563 | REQ_F_MUST_PUNT = BIT(REQ_F_MUST_PUNT_BIT), | |
564 | /* no timeout sequence */ | |
565 | REQ_F_TIMEOUT_NOSEQ = BIT(REQ_F_TIMEOUT_NOSEQ_BIT), | |
566 | /* completion under lock */ | |
567 | REQ_F_COMP_LOCKED = BIT(REQ_F_COMP_LOCKED_BIT), | |
99bc4c38 PB |
568 | /* needs cleanup */ |
569 | REQ_F_NEED_CLEANUP = BIT(REQ_F_NEED_CLEANUP_BIT), | |
2ca10259 JA |
570 | /* in overflow list */ |
571 | REQ_F_OVERFLOW = BIT(REQ_F_OVERFLOW_BIT), | |
d7718a9d JA |
572 | /* already went through poll handler */ |
573 | REQ_F_POLLED = BIT(REQ_F_POLLED_BIT), | |
bcda7baa JA |
574 | /* buffer already selected */ |
575 | REQ_F_BUFFER_SELECTED = BIT(REQ_F_BUFFER_SELECTED_BIT), | |
d7718a9d JA |
576 | }; |
577 | ||
578 | struct async_poll { | |
579 | struct io_poll_iocb poll; | |
580 | struct io_wq_work work; | |
6b47ee6e PB |
581 | }; |
582 | ||
09bb8394 JA |
583 | /* |
584 | * NOTE! Each of the iocb union members has the file pointer | |
585 | * as the first entry in their struct definition. So you can | |
586 | * access the file pointer through any of the sub-structs, | |
587 | * or directly as just 'ki_filp' in this struct. | |
588 | */ | |
2b188cc1 | 589 | struct io_kiocb { |
221c5eb2 | 590 | union { |
09bb8394 | 591 | struct file *file; |
9adbd45d | 592 | struct io_rw rw; |
221c5eb2 | 593 | struct io_poll_iocb poll; |
8ed8d3c3 JA |
594 | struct io_accept accept; |
595 | struct io_sync sync; | |
fbf23849 | 596 | struct io_cancel cancel; |
b29472ee | 597 | struct io_timeout timeout; |
3fbb51c1 | 598 | struct io_connect connect; |
e47293fd | 599 | struct io_sr_msg sr_msg; |
15b71abe | 600 | struct io_open open; |
b5dba59e | 601 | struct io_close close; |
05f3fb3c | 602 | struct io_files_update files_update; |
4840e418 | 603 | struct io_fadvise fadvise; |
c1ca757b | 604 | struct io_madvise madvise; |
3e4827b0 | 605 | struct io_epoll epoll; |
7d67af2c | 606 | struct io_splice splice; |
ddf0322d | 607 | struct io_provide_buf pbuf; |
221c5eb2 | 608 | }; |
2b188cc1 | 609 | |
1a6b74fc | 610 | struct io_async_ctx *io; |
cf6fd4bd | 611 | bool needs_fixed_file; |
d625c6ee | 612 | u8 opcode; |
2b188cc1 JA |
613 | |
614 | struct io_ring_ctx *ctx; | |
d7718a9d | 615 | struct list_head list; |
2b188cc1 | 616 | unsigned int flags; |
c16361c1 | 617 | refcount_t refs; |
3537b6a7 JA |
618 | struct task_struct *task; |
619 | unsigned long fsize; | |
2b188cc1 | 620 | u64 user_data; |
9e645e11 | 621 | u32 result; |
de0617e4 | 622 | u32 sequence; |
2b188cc1 | 623 | |
d7718a9d JA |
624 | struct list_head link_list; |
625 | ||
fcb323cc JA |
626 | struct list_head inflight_entry; |
627 | ||
05589553 XW |
628 | struct percpu_ref *fixed_file_refs; |
629 | ||
b41e9852 JA |
630 | union { |
631 | /* | |
632 | * Only commands that never go async can use the below fields, | |
d7718a9d JA |
633 | * obviously. Right now only IORING_OP_POLL_ADD uses them, and |
634 | * async armed poll handlers for regular commands. The latter | |
635 | * restore the work, if needed. | |
b41e9852 JA |
636 | */ |
637 | struct { | |
b41e9852 | 638 | struct callback_head task_work; |
d7718a9d JA |
639 | struct hlist_node hash_node; |
640 | struct async_poll *apoll; | |
bcda7baa | 641 | int cflags; |
b41e9852 JA |
642 | }; |
643 | struct io_wq_work work; | |
644 | }; | |
2b188cc1 JA |
645 | }; |
646 | ||
647 | #define IO_PLUG_THRESHOLD 2 | |
def596e9 | 648 | #define IO_IOPOLL_BATCH 8 |
2b188cc1 | 649 | |
9a56a232 JA |
650 | struct io_submit_state { |
651 | struct blk_plug plug; | |
652 | ||
2579f913 JA |
653 | /* |
654 | * io_kiocb alloc cache | |
655 | */ | |
656 | void *reqs[IO_IOPOLL_BATCH]; | |
6c8a3134 | 657 | unsigned int free_reqs; |
2579f913 | 658 | |
9a56a232 JA |
659 | /* |
660 | * File reference cache | |
661 | */ | |
662 | struct file *file; | |
663 | unsigned int fd; | |
664 | unsigned int has_refs; | |
665 | unsigned int used_refs; | |
666 | unsigned int ios_left; | |
667 | }; | |
668 | ||
d3656344 JA |
669 | struct io_op_def { |
670 | /* needs req->io allocated for deferral/async */ | |
671 | unsigned async_ctx : 1; | |
672 | /* needs current->mm setup, does mm access */ | |
673 | unsigned needs_mm : 1; | |
674 | /* needs req->file assigned */ | |
675 | unsigned needs_file : 1; | |
676 | /* needs req->file assigned IFF fd is >= 0 */ | |
677 | unsigned fd_non_neg : 1; | |
678 | /* hash wq insertion if file is a regular file */ | |
679 | unsigned hash_reg_file : 1; | |
680 | /* unbound wq insertion if file is a non-regular file */ | |
681 | unsigned unbound_nonreg_file : 1; | |
66f4af93 JA |
682 | /* opcode is not supported by this kernel */ |
683 | unsigned not_supported : 1; | |
f86cd20c JA |
684 | /* needs file table */ |
685 | unsigned file_table : 1; | |
ff002b30 JA |
686 | /* needs ->fs */ |
687 | unsigned needs_fs : 1; | |
8a72758c JA |
688 | /* set if opcode supports polled "wait" */ |
689 | unsigned pollin : 1; | |
690 | unsigned pollout : 1; | |
bcda7baa JA |
691 | /* op supports buffer selection */ |
692 | unsigned buffer_select : 1; | |
d3656344 JA |
693 | }; |
694 | ||
695 | static const struct io_op_def io_op_defs[] = { | |
0463b6c5 PB |
696 | [IORING_OP_NOP] = {}, |
697 | [IORING_OP_READV] = { | |
d3656344 JA |
698 | .async_ctx = 1, |
699 | .needs_mm = 1, | |
700 | .needs_file = 1, | |
701 | .unbound_nonreg_file = 1, | |
8a72758c | 702 | .pollin = 1, |
4d954c25 | 703 | .buffer_select = 1, |
d3656344 | 704 | }, |
0463b6c5 | 705 | [IORING_OP_WRITEV] = { |
d3656344 JA |
706 | .async_ctx = 1, |
707 | .needs_mm = 1, | |
708 | .needs_file = 1, | |
709 | .hash_reg_file = 1, | |
710 | .unbound_nonreg_file = 1, | |
8a72758c | 711 | .pollout = 1, |
d3656344 | 712 | }, |
0463b6c5 | 713 | [IORING_OP_FSYNC] = { |
d3656344 JA |
714 | .needs_file = 1, |
715 | }, | |
0463b6c5 | 716 | [IORING_OP_READ_FIXED] = { |
d3656344 JA |
717 | .needs_file = 1, |
718 | .unbound_nonreg_file = 1, | |
8a72758c | 719 | .pollin = 1, |
d3656344 | 720 | }, |
0463b6c5 | 721 | [IORING_OP_WRITE_FIXED] = { |
d3656344 JA |
722 | .needs_file = 1, |
723 | .hash_reg_file = 1, | |
724 | .unbound_nonreg_file = 1, | |
8a72758c | 725 | .pollout = 1, |
d3656344 | 726 | }, |
0463b6c5 | 727 | [IORING_OP_POLL_ADD] = { |
d3656344 JA |
728 | .needs_file = 1, |
729 | .unbound_nonreg_file = 1, | |
730 | }, | |
0463b6c5 PB |
731 | [IORING_OP_POLL_REMOVE] = {}, |
732 | [IORING_OP_SYNC_FILE_RANGE] = { | |
d3656344 JA |
733 | .needs_file = 1, |
734 | }, | |
0463b6c5 | 735 | [IORING_OP_SENDMSG] = { |
d3656344 JA |
736 | .async_ctx = 1, |
737 | .needs_mm = 1, | |
738 | .needs_file = 1, | |
739 | .unbound_nonreg_file = 1, | |
ff002b30 | 740 | .needs_fs = 1, |
8a72758c | 741 | .pollout = 1, |
d3656344 | 742 | }, |
0463b6c5 | 743 | [IORING_OP_RECVMSG] = { |
d3656344 JA |
744 | .async_ctx = 1, |
745 | .needs_mm = 1, | |
746 | .needs_file = 1, | |
747 | .unbound_nonreg_file = 1, | |
ff002b30 | 748 | .needs_fs = 1, |
8a72758c | 749 | .pollin = 1, |
52de1fe1 | 750 | .buffer_select = 1, |
d3656344 | 751 | }, |
0463b6c5 | 752 | [IORING_OP_TIMEOUT] = { |
d3656344 JA |
753 | .async_ctx = 1, |
754 | .needs_mm = 1, | |
755 | }, | |
0463b6c5 PB |
756 | [IORING_OP_TIMEOUT_REMOVE] = {}, |
757 | [IORING_OP_ACCEPT] = { | |
d3656344 JA |
758 | .needs_mm = 1, |
759 | .needs_file = 1, | |
760 | .unbound_nonreg_file = 1, | |
f86cd20c | 761 | .file_table = 1, |
8a72758c | 762 | .pollin = 1, |
d3656344 | 763 | }, |
0463b6c5 PB |
764 | [IORING_OP_ASYNC_CANCEL] = {}, |
765 | [IORING_OP_LINK_TIMEOUT] = { | |
d3656344 JA |
766 | .async_ctx = 1, |
767 | .needs_mm = 1, | |
768 | }, | |
0463b6c5 | 769 | [IORING_OP_CONNECT] = { |
d3656344 JA |
770 | .async_ctx = 1, |
771 | .needs_mm = 1, | |
772 | .needs_file = 1, | |
773 | .unbound_nonreg_file = 1, | |
8a72758c | 774 | .pollout = 1, |
d3656344 | 775 | }, |
0463b6c5 | 776 | [IORING_OP_FALLOCATE] = { |
d3656344 JA |
777 | .needs_file = 1, |
778 | }, | |
0463b6c5 | 779 | [IORING_OP_OPENAT] = { |
d3656344 JA |
780 | .needs_file = 1, |
781 | .fd_non_neg = 1, | |
f86cd20c | 782 | .file_table = 1, |
ff002b30 | 783 | .needs_fs = 1, |
d3656344 | 784 | }, |
0463b6c5 | 785 | [IORING_OP_CLOSE] = { |
d3656344 | 786 | .needs_file = 1, |
f86cd20c | 787 | .file_table = 1, |
d3656344 | 788 | }, |
0463b6c5 | 789 | [IORING_OP_FILES_UPDATE] = { |
d3656344 | 790 | .needs_mm = 1, |
f86cd20c | 791 | .file_table = 1, |
d3656344 | 792 | }, |
0463b6c5 | 793 | [IORING_OP_STATX] = { |
d3656344 JA |
794 | .needs_mm = 1, |
795 | .needs_file = 1, | |
796 | .fd_non_neg = 1, | |
ff002b30 | 797 | .needs_fs = 1, |
d3656344 | 798 | }, |
0463b6c5 | 799 | [IORING_OP_READ] = { |
3a6820f2 JA |
800 | .needs_mm = 1, |
801 | .needs_file = 1, | |
802 | .unbound_nonreg_file = 1, | |
8a72758c | 803 | .pollin = 1, |
bcda7baa | 804 | .buffer_select = 1, |
3a6820f2 | 805 | }, |
0463b6c5 | 806 | [IORING_OP_WRITE] = { |
3a6820f2 JA |
807 | .needs_mm = 1, |
808 | .needs_file = 1, | |
809 | .unbound_nonreg_file = 1, | |
8a72758c | 810 | .pollout = 1, |
3a6820f2 | 811 | }, |
0463b6c5 | 812 | [IORING_OP_FADVISE] = { |
4840e418 JA |
813 | .needs_file = 1, |
814 | }, | |
0463b6c5 | 815 | [IORING_OP_MADVISE] = { |
c1ca757b JA |
816 | .needs_mm = 1, |
817 | }, | |
0463b6c5 | 818 | [IORING_OP_SEND] = { |
fddaface JA |
819 | .needs_mm = 1, |
820 | .needs_file = 1, | |
821 | .unbound_nonreg_file = 1, | |
8a72758c | 822 | .pollout = 1, |
fddaface | 823 | }, |
0463b6c5 | 824 | [IORING_OP_RECV] = { |
fddaface JA |
825 | .needs_mm = 1, |
826 | .needs_file = 1, | |
827 | .unbound_nonreg_file = 1, | |
8a72758c | 828 | .pollin = 1, |
bcda7baa | 829 | .buffer_select = 1, |
fddaface | 830 | }, |
0463b6c5 | 831 | [IORING_OP_OPENAT2] = { |
cebdb986 JA |
832 | .needs_file = 1, |
833 | .fd_non_neg = 1, | |
f86cd20c | 834 | .file_table = 1, |
ff002b30 | 835 | .needs_fs = 1, |
cebdb986 | 836 | }, |
3e4827b0 JA |
837 | [IORING_OP_EPOLL_CTL] = { |
838 | .unbound_nonreg_file = 1, | |
839 | .file_table = 1, | |
840 | }, | |
7d67af2c PB |
841 | [IORING_OP_SPLICE] = { |
842 | .needs_file = 1, | |
843 | .hash_reg_file = 1, | |
844 | .unbound_nonreg_file = 1, | |
ddf0322d JA |
845 | }, |
846 | [IORING_OP_PROVIDE_BUFFERS] = {}, | |
067524e9 | 847 | [IORING_OP_REMOVE_BUFFERS] = {}, |
d3656344 JA |
848 | }; |
849 | ||
561fb04a | 850 | static void io_wq_submit_work(struct io_wq_work **workptr); |
78e19bbe | 851 | static void io_cqring_fill_event(struct io_kiocb *req, long res); |
ec9c02ad | 852 | static void io_put_req(struct io_kiocb *req); |
978db57e | 853 | static void __io_double_put_req(struct io_kiocb *req); |
94ae5e77 JA |
854 | static struct io_kiocb *io_prep_linked_timeout(struct io_kiocb *req); |
855 | static void io_queue_linked_timeout(struct io_kiocb *req); | |
05f3fb3c JA |
856 | static int __io_sqe_files_update(struct io_ring_ctx *ctx, |
857 | struct io_uring_files_update *ip, | |
858 | unsigned nr_args); | |
f86cd20c | 859 | static int io_grab_files(struct io_kiocb *req); |
99bc4c38 | 860 | static void io_cleanup_req(struct io_kiocb *req); |
b41e9852 JA |
861 | static int io_file_get(struct io_submit_state *state, struct io_kiocb *req, |
862 | int fd, struct file **out_file, bool fixed); | |
863 | static void __io_queue_sqe(struct io_kiocb *req, | |
864 | const struct io_uring_sqe *sqe); | |
de0617e4 | 865 | |
2b188cc1 JA |
866 | static struct kmem_cache *req_cachep; |
867 | ||
868 | static const struct file_operations io_uring_fops; | |
869 | ||
870 | struct sock *io_uring_get_socket(struct file *file) | |
871 | { | |
872 | #if defined(CONFIG_UNIX) | |
873 | if (file->f_op == &io_uring_fops) { | |
874 | struct io_ring_ctx *ctx = file->private_data; | |
875 | ||
876 | return ctx->ring_sock->sk; | |
877 | } | |
878 | #endif | |
879 | return NULL; | |
880 | } | |
881 | EXPORT_SYMBOL(io_uring_get_socket); | |
882 | ||
883 | static void io_ring_ctx_ref_free(struct percpu_ref *ref) | |
884 | { | |
885 | struct io_ring_ctx *ctx = container_of(ref, struct io_ring_ctx, refs); | |
886 | ||
206aefde | 887 | complete(&ctx->completions[0]); |
2b188cc1 JA |
888 | } |
889 | ||
890 | static struct io_ring_ctx *io_ring_ctx_alloc(struct io_uring_params *p) | |
891 | { | |
892 | struct io_ring_ctx *ctx; | |
78076bb6 | 893 | int hash_bits; |
2b188cc1 JA |
894 | |
895 | ctx = kzalloc(sizeof(*ctx), GFP_KERNEL); | |
896 | if (!ctx) | |
897 | return NULL; | |
898 | ||
0ddf92e8 JA |
899 | ctx->fallback_req = kmem_cache_alloc(req_cachep, GFP_KERNEL); |
900 | if (!ctx->fallback_req) | |
901 | goto err; | |
902 | ||
206aefde JA |
903 | ctx->completions = kmalloc(2 * sizeof(struct completion), GFP_KERNEL); |
904 | if (!ctx->completions) | |
905 | goto err; | |
906 | ||
78076bb6 JA |
907 | /* |
908 | * Use 5 bits less than the max cq entries, that should give us around | |
909 | * 32 entries per hash list if totally full and uniformly spread. | |
910 | */ | |
911 | hash_bits = ilog2(p->cq_entries); | |
912 | hash_bits -= 5; | |
913 | if (hash_bits <= 0) | |
914 | hash_bits = 1; | |
915 | ctx->cancel_hash_bits = hash_bits; | |
916 | ctx->cancel_hash = kmalloc((1U << hash_bits) * sizeof(struct hlist_head), | |
917 | GFP_KERNEL); | |
918 | if (!ctx->cancel_hash) | |
919 | goto err; | |
920 | __hash_init(ctx->cancel_hash, 1U << hash_bits); | |
921 | ||
21482896 | 922 | if (percpu_ref_init(&ctx->refs, io_ring_ctx_ref_free, |
206aefde JA |
923 | PERCPU_REF_ALLOW_REINIT, GFP_KERNEL)) |
924 | goto err; | |
2b188cc1 JA |
925 | |
926 | ctx->flags = p->flags; | |
927 | init_waitqueue_head(&ctx->cq_wait); | |
1d7bb1d5 | 928 | INIT_LIST_HEAD(&ctx->cq_overflow_list); |
206aefde JA |
929 | init_completion(&ctx->completions[0]); |
930 | init_completion(&ctx->completions[1]); | |
5a2e745d | 931 | idr_init(&ctx->io_buffer_idr); |
071698e1 | 932 | idr_init(&ctx->personality_idr); |
2b188cc1 JA |
933 | mutex_init(&ctx->uring_lock); |
934 | init_waitqueue_head(&ctx->wait); | |
935 | spin_lock_init(&ctx->completion_lock); | |
def596e9 | 936 | INIT_LIST_HEAD(&ctx->poll_list); |
de0617e4 | 937 | INIT_LIST_HEAD(&ctx->defer_list); |
5262f567 | 938 | INIT_LIST_HEAD(&ctx->timeout_list); |
fcb323cc JA |
939 | init_waitqueue_head(&ctx->inflight_wait); |
940 | spin_lock_init(&ctx->inflight_lock); | |
941 | INIT_LIST_HEAD(&ctx->inflight_list); | |
2b188cc1 | 942 | return ctx; |
206aefde | 943 | err: |
0ddf92e8 JA |
944 | if (ctx->fallback_req) |
945 | kmem_cache_free(req_cachep, ctx->fallback_req); | |
206aefde | 946 | kfree(ctx->completions); |
78076bb6 | 947 | kfree(ctx->cancel_hash); |
206aefde JA |
948 | kfree(ctx); |
949 | return NULL; | |
2b188cc1 JA |
950 | } |
951 | ||
9d858b21 | 952 | static inline bool __req_need_defer(struct io_kiocb *req) |
7adf4eaf | 953 | { |
a197f664 JL |
954 | struct io_ring_ctx *ctx = req->ctx; |
955 | ||
498ccd9e JA |
956 | return req->sequence != ctx->cached_cq_tail + ctx->cached_sq_dropped |
957 | + atomic_read(&ctx->cached_cq_overflow); | |
7adf4eaf JA |
958 | } |
959 | ||
9d858b21 | 960 | static inline bool req_need_defer(struct io_kiocb *req) |
de0617e4 | 961 | { |
87987898 | 962 | if (unlikely(req->flags & REQ_F_IO_DRAIN)) |
9d858b21 | 963 | return __req_need_defer(req); |
de0617e4 | 964 | |
9d858b21 | 965 | return false; |
de0617e4 JA |
966 | } |
967 | ||
7adf4eaf | 968 | static struct io_kiocb *io_get_deferred_req(struct io_ring_ctx *ctx) |
de0617e4 JA |
969 | { |
970 | struct io_kiocb *req; | |
971 | ||
7adf4eaf | 972 | req = list_first_entry_or_null(&ctx->defer_list, struct io_kiocb, list); |
9d858b21 | 973 | if (req && !req_need_defer(req)) { |
de0617e4 JA |
974 | list_del_init(&req->list); |
975 | return req; | |
976 | } | |
977 | ||
978 | return NULL; | |
979 | } | |
980 | ||
5262f567 JA |
981 | static struct io_kiocb *io_get_timeout_req(struct io_ring_ctx *ctx) |
982 | { | |
7adf4eaf JA |
983 | struct io_kiocb *req; |
984 | ||
985 | req = list_first_entry_or_null(&ctx->timeout_list, struct io_kiocb, list); | |
93bd25bb JA |
986 | if (req) { |
987 | if (req->flags & REQ_F_TIMEOUT_NOSEQ) | |
988 | return NULL; | |
fb4b3d3f | 989 | if (!__req_need_defer(req)) { |
93bd25bb JA |
990 | list_del_init(&req->list); |
991 | return req; | |
992 | } | |
7adf4eaf JA |
993 | } |
994 | ||
995 | return NULL; | |
5262f567 JA |
996 | } |
997 | ||
de0617e4 | 998 | static void __io_commit_cqring(struct io_ring_ctx *ctx) |
2b188cc1 | 999 | { |
75b28aff | 1000 | struct io_rings *rings = ctx->rings; |
2b188cc1 | 1001 | |
07910158 PB |
1002 | /* order cqe stores with ring update */ |
1003 | smp_store_release(&rings->cq.tail, ctx->cached_cq_tail); | |
2b188cc1 | 1004 | |
07910158 PB |
1005 | if (wq_has_sleeper(&ctx->cq_wait)) { |
1006 | wake_up_interruptible(&ctx->cq_wait); | |
1007 | kill_fasync(&ctx->cq_fasync, SIGIO, POLL_IN); | |
2b188cc1 JA |
1008 | } |
1009 | } | |
1010 | ||
cccf0ee8 JA |
1011 | static inline void io_req_work_grab_env(struct io_kiocb *req, |
1012 | const struct io_op_def *def) | |
1013 | { | |
1014 | if (!req->work.mm && def->needs_mm) { | |
1015 | mmgrab(current->mm); | |
1016 | req->work.mm = current->mm; | |
2b188cc1 | 1017 | } |
cccf0ee8 JA |
1018 | if (!req->work.creds) |
1019 | req->work.creds = get_current_cred(); | |
ff002b30 JA |
1020 | if (!req->work.fs && def->needs_fs) { |
1021 | spin_lock(¤t->fs->lock); | |
1022 | if (!current->fs->in_exec) { | |
1023 | req->work.fs = current->fs; | |
1024 | req->work.fs->users++; | |
1025 | } else { | |
1026 | req->work.flags |= IO_WQ_WORK_CANCEL; | |
1027 | } | |
1028 | spin_unlock(¤t->fs->lock); | |
1029 | } | |
6ab23144 JA |
1030 | if (!req->work.task_pid) |
1031 | req->work.task_pid = task_pid_vnr(current); | |
2b188cc1 JA |
1032 | } |
1033 | ||
cccf0ee8 | 1034 | static inline void io_req_work_drop_env(struct io_kiocb *req) |
18d9be1a | 1035 | { |
cccf0ee8 JA |
1036 | if (req->work.mm) { |
1037 | mmdrop(req->work.mm); | |
1038 | req->work.mm = NULL; | |
1039 | } | |
1040 | if (req->work.creds) { | |
1041 | put_cred(req->work.creds); | |
1042 | req->work.creds = NULL; | |
1043 | } | |
ff002b30 JA |
1044 | if (req->work.fs) { |
1045 | struct fs_struct *fs = req->work.fs; | |
1046 | ||
1047 | spin_lock(&req->work.fs->lock); | |
1048 | if (--fs->users) | |
1049 | fs = NULL; | |
1050 | spin_unlock(&req->work.fs->lock); | |
1051 | if (fs) | |
1052 | free_fs_struct(fs); | |
1053 | } | |
561fb04a JA |
1054 | } |
1055 | ||
8766dd51 | 1056 | static inline void io_prep_async_work(struct io_kiocb *req, |
94ae5e77 | 1057 | struct io_kiocb **link) |
18d9be1a | 1058 | { |
d3656344 | 1059 | const struct io_op_def *def = &io_op_defs[req->opcode]; |
54a91f3b | 1060 | |
d3656344 JA |
1061 | if (req->flags & REQ_F_ISREG) { |
1062 | if (def->hash_reg_file) | |
8766dd51 | 1063 | io_wq_hash_work(&req->work, file_inode(req->file)); |
d3656344 JA |
1064 | } else { |
1065 | if (def->unbound_nonreg_file) | |
3529d8c2 | 1066 | req->work.flags |= IO_WQ_WORK_UNBOUND; |
54a91f3b | 1067 | } |
cccf0ee8 JA |
1068 | |
1069 | io_req_work_grab_env(req, def); | |
54a91f3b | 1070 | |
94ae5e77 | 1071 | *link = io_prep_linked_timeout(req); |
561fb04a JA |
1072 | } |
1073 | ||
a197f664 | 1074 | static inline void io_queue_async_work(struct io_kiocb *req) |
561fb04a | 1075 | { |
a197f664 | 1076 | struct io_ring_ctx *ctx = req->ctx; |
94ae5e77 | 1077 | struct io_kiocb *link; |
94ae5e77 | 1078 | |
8766dd51 | 1079 | io_prep_async_work(req, &link); |
561fb04a | 1080 | |
8766dd51 PB |
1081 | trace_io_uring_queue_async_work(ctx, io_wq_is_hashed(&req->work), req, |
1082 | &req->work, req->flags); | |
1083 | io_wq_enqueue(ctx->io_wq, &req->work); | |
94ae5e77 JA |
1084 | |
1085 | if (link) | |
1086 | io_queue_linked_timeout(link); | |
18d9be1a JA |
1087 | } |
1088 | ||
5262f567 JA |
1089 | static void io_kill_timeout(struct io_kiocb *req) |
1090 | { | |
1091 | int ret; | |
1092 | ||
2d28390a | 1093 | ret = hrtimer_try_to_cancel(&req->io->timeout.timer); |
5262f567 JA |
1094 | if (ret != -1) { |
1095 | atomic_inc(&req->ctx->cq_timeouts); | |
842f9612 | 1096 | list_del_init(&req->list); |
f0e20b89 | 1097 | req->flags |= REQ_F_COMP_LOCKED; |
78e19bbe | 1098 | io_cqring_fill_event(req, 0); |
ec9c02ad | 1099 | io_put_req(req); |
5262f567 JA |
1100 | } |
1101 | } | |
1102 | ||
1103 | static void io_kill_timeouts(struct io_ring_ctx *ctx) | |
1104 | { | |
1105 | struct io_kiocb *req, *tmp; | |
1106 | ||
1107 | spin_lock_irq(&ctx->completion_lock); | |
1108 | list_for_each_entry_safe(req, tmp, &ctx->timeout_list, list) | |
1109 | io_kill_timeout(req); | |
1110 | spin_unlock_irq(&ctx->completion_lock); | |
1111 | } | |
1112 | ||
de0617e4 JA |
1113 | static void io_commit_cqring(struct io_ring_ctx *ctx) |
1114 | { | |
1115 | struct io_kiocb *req; | |
1116 | ||
5262f567 JA |
1117 | while ((req = io_get_timeout_req(ctx)) != NULL) |
1118 | io_kill_timeout(req); | |
1119 | ||
de0617e4 JA |
1120 | __io_commit_cqring(ctx); |
1121 | ||
87987898 | 1122 | while ((req = io_get_deferred_req(ctx)) != NULL) |
a197f664 | 1123 | io_queue_async_work(req); |
de0617e4 JA |
1124 | } |
1125 | ||
2b188cc1 JA |
1126 | static struct io_uring_cqe *io_get_cqring(struct io_ring_ctx *ctx) |
1127 | { | |
75b28aff | 1128 | struct io_rings *rings = ctx->rings; |
2b188cc1 JA |
1129 | unsigned tail; |
1130 | ||
1131 | tail = ctx->cached_cq_tail; | |
115e12e5 SB |
1132 | /* |
1133 | * writes to the cq entry need to come after reading head; the | |
1134 | * control dependency is enough as we're using WRITE_ONCE to | |
1135 | * fill the cq entry | |
1136 | */ | |
75b28aff | 1137 | if (tail - READ_ONCE(rings->cq.head) == rings->cq_ring_entries) |
2b188cc1 JA |
1138 | return NULL; |
1139 | ||
1140 | ctx->cached_cq_tail++; | |
75b28aff | 1141 | return &rings->cqes[tail & ctx->cq_mask]; |
2b188cc1 JA |
1142 | } |
1143 | ||
f2842ab5 JA |
1144 | static inline bool io_should_trigger_evfd(struct io_ring_ctx *ctx) |
1145 | { | |
f0b493e6 JA |
1146 | if (!ctx->cq_ev_fd) |
1147 | return false; | |
f2842ab5 JA |
1148 | if (!ctx->eventfd_async) |
1149 | return true; | |
b41e9852 | 1150 | return io_wq_current_is_worker(); |
f2842ab5 JA |
1151 | } |
1152 | ||
b41e9852 | 1153 | static void io_cqring_ev_posted(struct io_ring_ctx *ctx) |
1d7bb1d5 JA |
1154 | { |
1155 | if (waitqueue_active(&ctx->wait)) | |
1156 | wake_up(&ctx->wait); | |
1157 | if (waitqueue_active(&ctx->sqo_wait)) | |
1158 | wake_up(&ctx->sqo_wait); | |
b41e9852 | 1159 | if (io_should_trigger_evfd(ctx)) |
1d7bb1d5 JA |
1160 | eventfd_signal(ctx->cq_ev_fd, 1); |
1161 | } | |
1162 | ||
c4a2ed72 JA |
1163 | /* Returns true if there are no backlogged entries after the flush */ |
1164 | static bool io_cqring_overflow_flush(struct io_ring_ctx *ctx, bool force) | |
1d7bb1d5 JA |
1165 | { |
1166 | struct io_rings *rings = ctx->rings; | |
1167 | struct io_uring_cqe *cqe; | |
1168 | struct io_kiocb *req; | |
1169 | unsigned long flags; | |
1170 | LIST_HEAD(list); | |
1171 | ||
1172 | if (!force) { | |
1173 | if (list_empty_careful(&ctx->cq_overflow_list)) | |
c4a2ed72 | 1174 | return true; |
1d7bb1d5 JA |
1175 | if ((ctx->cached_cq_tail - READ_ONCE(rings->cq.head) == |
1176 | rings->cq_ring_entries)) | |
c4a2ed72 | 1177 | return false; |
1d7bb1d5 JA |
1178 | } |
1179 | ||
1180 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
1181 | ||
1182 | /* if force is set, the ring is going away. always drop after that */ | |
1183 | if (force) | |
69b3e546 | 1184 | ctx->cq_overflow_flushed = 1; |
1d7bb1d5 | 1185 | |
c4a2ed72 | 1186 | cqe = NULL; |
1d7bb1d5 JA |
1187 | while (!list_empty(&ctx->cq_overflow_list)) { |
1188 | cqe = io_get_cqring(ctx); | |
1189 | if (!cqe && !force) | |
1190 | break; | |
1191 | ||
1192 | req = list_first_entry(&ctx->cq_overflow_list, struct io_kiocb, | |
1193 | list); | |
1194 | list_move(&req->list, &list); | |
2ca10259 | 1195 | req->flags &= ~REQ_F_OVERFLOW; |
1d7bb1d5 JA |
1196 | if (cqe) { |
1197 | WRITE_ONCE(cqe->user_data, req->user_data); | |
1198 | WRITE_ONCE(cqe->res, req->result); | |
bcda7baa | 1199 | WRITE_ONCE(cqe->flags, req->cflags); |
1d7bb1d5 JA |
1200 | } else { |
1201 | WRITE_ONCE(ctx->rings->cq_overflow, | |
1202 | atomic_inc_return(&ctx->cached_cq_overflow)); | |
1203 | } | |
1204 | } | |
1205 | ||
1206 | io_commit_cqring(ctx); | |
ad3eb2c8 JA |
1207 | if (cqe) { |
1208 | clear_bit(0, &ctx->sq_check_overflow); | |
1209 | clear_bit(0, &ctx->cq_check_overflow); | |
1210 | } | |
1d7bb1d5 JA |
1211 | spin_unlock_irqrestore(&ctx->completion_lock, flags); |
1212 | io_cqring_ev_posted(ctx); | |
1213 | ||
1214 | while (!list_empty(&list)) { | |
1215 | req = list_first_entry(&list, struct io_kiocb, list); | |
1216 | list_del(&req->list); | |
ec9c02ad | 1217 | io_put_req(req); |
1d7bb1d5 | 1218 | } |
c4a2ed72 JA |
1219 | |
1220 | return cqe != NULL; | |
1d7bb1d5 JA |
1221 | } |
1222 | ||
bcda7baa | 1223 | static void __io_cqring_fill_event(struct io_kiocb *req, long res, long cflags) |
2b188cc1 | 1224 | { |
78e19bbe | 1225 | struct io_ring_ctx *ctx = req->ctx; |
2b188cc1 JA |
1226 | struct io_uring_cqe *cqe; |
1227 | ||
78e19bbe | 1228 | trace_io_uring_complete(ctx, req->user_data, res); |
51c3ff62 | 1229 | |
2b188cc1 JA |
1230 | /* |
1231 | * If we can't get a cq entry, userspace overflowed the | |
1232 | * submission (by quite a lot). Increment the overflow count in | |
1233 | * the ring. | |
1234 | */ | |
1235 | cqe = io_get_cqring(ctx); | |
1d7bb1d5 | 1236 | if (likely(cqe)) { |
78e19bbe | 1237 | WRITE_ONCE(cqe->user_data, req->user_data); |
2b188cc1 | 1238 | WRITE_ONCE(cqe->res, res); |
bcda7baa | 1239 | WRITE_ONCE(cqe->flags, cflags); |
1d7bb1d5 | 1240 | } else if (ctx->cq_overflow_flushed) { |
498ccd9e JA |
1241 | WRITE_ONCE(ctx->rings->cq_overflow, |
1242 | atomic_inc_return(&ctx->cached_cq_overflow)); | |
1d7bb1d5 | 1243 | } else { |
ad3eb2c8 JA |
1244 | if (list_empty(&ctx->cq_overflow_list)) { |
1245 | set_bit(0, &ctx->sq_check_overflow); | |
1246 | set_bit(0, &ctx->cq_check_overflow); | |
1247 | } | |
2ca10259 | 1248 | req->flags |= REQ_F_OVERFLOW; |
1d7bb1d5 JA |
1249 | refcount_inc(&req->refs); |
1250 | req->result = res; | |
bcda7baa | 1251 | req->cflags = cflags; |
1d7bb1d5 | 1252 | list_add_tail(&req->list, &ctx->cq_overflow_list); |
2b188cc1 JA |
1253 | } |
1254 | } | |
1255 | ||
bcda7baa JA |
1256 | static void io_cqring_fill_event(struct io_kiocb *req, long res) |
1257 | { | |
1258 | __io_cqring_fill_event(req, res, 0); | |
1259 | } | |
1260 | ||
1261 | static void __io_cqring_add_event(struct io_kiocb *req, long res, long cflags) | |
2b188cc1 | 1262 | { |
78e19bbe | 1263 | struct io_ring_ctx *ctx = req->ctx; |
2b188cc1 JA |
1264 | unsigned long flags; |
1265 | ||
1266 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
bcda7baa | 1267 | __io_cqring_fill_event(req, res, cflags); |
2b188cc1 JA |
1268 | io_commit_cqring(ctx); |
1269 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
1270 | ||
8c838788 | 1271 | io_cqring_ev_posted(ctx); |
2b188cc1 JA |
1272 | } |
1273 | ||
bcda7baa JA |
1274 | static void io_cqring_add_event(struct io_kiocb *req, long res) |
1275 | { | |
1276 | __io_cqring_add_event(req, res, 0); | |
1277 | } | |
1278 | ||
0ddf92e8 JA |
1279 | static inline bool io_is_fallback_req(struct io_kiocb *req) |
1280 | { | |
1281 | return req == (struct io_kiocb *) | |
1282 | ((unsigned long) req->ctx->fallback_req & ~1UL); | |
1283 | } | |
1284 | ||
1285 | static struct io_kiocb *io_get_fallback_req(struct io_ring_ctx *ctx) | |
1286 | { | |
1287 | struct io_kiocb *req; | |
1288 | ||
1289 | req = ctx->fallback_req; | |
1290 | if (!test_and_set_bit_lock(0, (unsigned long *) ctx->fallback_req)) | |
1291 | return req; | |
1292 | ||
1293 | return NULL; | |
1294 | } | |
1295 | ||
2579f913 JA |
1296 | static struct io_kiocb *io_get_req(struct io_ring_ctx *ctx, |
1297 | struct io_submit_state *state) | |
2b188cc1 | 1298 | { |
fd6fab2c | 1299 | gfp_t gfp = GFP_KERNEL | __GFP_NOWARN; |
2b188cc1 JA |
1300 | struct io_kiocb *req; |
1301 | ||
2579f913 | 1302 | if (!state) { |
fd6fab2c | 1303 | req = kmem_cache_alloc(req_cachep, gfp); |
2579f913 | 1304 | if (unlikely(!req)) |
0ddf92e8 | 1305 | goto fallback; |
2579f913 JA |
1306 | } else if (!state->free_reqs) { |
1307 | size_t sz; | |
1308 | int ret; | |
1309 | ||
1310 | sz = min_t(size_t, state->ios_left, ARRAY_SIZE(state->reqs)); | |
fd6fab2c JA |
1311 | ret = kmem_cache_alloc_bulk(req_cachep, gfp, sz, state->reqs); |
1312 | ||
1313 | /* | |
1314 | * Bulk alloc is all-or-nothing. If we fail to get a batch, | |
1315 | * retry single alloc to be on the safe side. | |
1316 | */ | |
1317 | if (unlikely(ret <= 0)) { | |
1318 | state->reqs[0] = kmem_cache_alloc(req_cachep, gfp); | |
1319 | if (!state->reqs[0]) | |
0ddf92e8 | 1320 | goto fallback; |
fd6fab2c JA |
1321 | ret = 1; |
1322 | } | |
2579f913 | 1323 | state->free_reqs = ret - 1; |
6c8a3134 | 1324 | req = state->reqs[ret - 1]; |
2579f913 | 1325 | } else { |
2579f913 | 1326 | state->free_reqs--; |
6c8a3134 | 1327 | req = state->reqs[state->free_reqs]; |
2b188cc1 JA |
1328 | } |
1329 | ||
0ddf92e8 | 1330 | got_it: |
1a6b74fc | 1331 | req->io = NULL; |
60c112b0 | 1332 | req->file = NULL; |
2579f913 JA |
1333 | req->ctx = ctx; |
1334 | req->flags = 0; | |
e65ef56d JA |
1335 | /* one is dropped after submission, the other at completion */ |
1336 | refcount_set(&req->refs, 2); | |
3537b6a7 | 1337 | req->task = NULL; |
9e645e11 | 1338 | req->result = 0; |
561fb04a | 1339 | INIT_IO_WORK(&req->work, io_wq_submit_work); |
2579f913 | 1340 | return req; |
0ddf92e8 JA |
1341 | fallback: |
1342 | req = io_get_fallback_req(ctx); | |
1343 | if (req) | |
1344 | goto got_it; | |
2b188cc1 JA |
1345 | return NULL; |
1346 | } | |
1347 | ||
8da11c19 PB |
1348 | static inline void io_put_file(struct io_kiocb *req, struct file *file, |
1349 | bool fixed) | |
1350 | { | |
1351 | if (fixed) | |
05589553 | 1352 | percpu_ref_put(req->fixed_file_refs); |
8da11c19 PB |
1353 | else |
1354 | fput(file); | |
1355 | } | |
1356 | ||
2b85edfc | 1357 | static void __io_req_do_free(struct io_kiocb *req) |
def596e9 | 1358 | { |
2b85edfc PB |
1359 | if (likely(!io_is_fallback_req(req))) |
1360 | kmem_cache_free(req_cachep, req); | |
1361 | else | |
1362 | clear_bit_unlock(0, (unsigned long *) req->ctx->fallback_req); | |
1363 | } | |
1364 | ||
c6ca97b3 | 1365 | static void __io_req_aux_free(struct io_kiocb *req) |
2b188cc1 | 1366 | { |
929a3af9 PB |
1367 | if (req->flags & REQ_F_NEED_CLEANUP) |
1368 | io_cleanup_req(req); | |
1369 | ||
96fd84d8 | 1370 | kfree(req->io); |
8da11c19 PB |
1371 | if (req->file) |
1372 | io_put_file(req, req->file, (req->flags & REQ_F_FIXED_FILE)); | |
3537b6a7 JA |
1373 | if (req->task) |
1374 | put_task_struct(req->task); | |
cccf0ee8 JA |
1375 | |
1376 | io_req_work_drop_env(req); | |
def596e9 JA |
1377 | } |
1378 | ||
9e645e11 | 1379 | static void __io_free_req(struct io_kiocb *req) |
2b188cc1 | 1380 | { |
c6ca97b3 | 1381 | __io_req_aux_free(req); |
fcb323cc | 1382 | |
fcb323cc | 1383 | if (req->flags & REQ_F_INFLIGHT) { |
c6ca97b3 | 1384 | struct io_ring_ctx *ctx = req->ctx; |
fcb323cc JA |
1385 | unsigned long flags; |
1386 | ||
1387 | spin_lock_irqsave(&ctx->inflight_lock, flags); | |
1388 | list_del(&req->inflight_entry); | |
1389 | if (waitqueue_active(&ctx->inflight_wait)) | |
1390 | wake_up(&ctx->inflight_wait); | |
1391 | spin_unlock_irqrestore(&ctx->inflight_lock, flags); | |
1392 | } | |
2b85edfc PB |
1393 | |
1394 | percpu_ref_put(&req->ctx->refs); | |
1395 | __io_req_do_free(req); | |
e65ef56d JA |
1396 | } |
1397 | ||
c6ca97b3 JA |
1398 | struct req_batch { |
1399 | void *reqs[IO_IOPOLL_BATCH]; | |
1400 | int to_free; | |
1401 | int need_iter; | |
1402 | }; | |
1403 | ||
1404 | static void io_free_req_many(struct io_ring_ctx *ctx, struct req_batch *rb) | |
1405 | { | |
1406 | if (!rb->to_free) | |
1407 | return; | |
1408 | if (rb->need_iter) { | |
1409 | int i, inflight = 0; | |
1410 | unsigned long flags; | |
1411 | ||
1412 | for (i = 0; i < rb->to_free; i++) { | |
1413 | struct io_kiocb *req = rb->reqs[i]; | |
1414 | ||
10fef4be | 1415 | if (req->flags & REQ_F_FIXED_FILE) { |
c6ca97b3 | 1416 | req->file = NULL; |
05589553 | 1417 | percpu_ref_put(req->fixed_file_refs); |
10fef4be | 1418 | } |
c6ca97b3 JA |
1419 | if (req->flags & REQ_F_INFLIGHT) |
1420 | inflight++; | |
c6ca97b3 JA |
1421 | __io_req_aux_free(req); |
1422 | } | |
1423 | if (!inflight) | |
1424 | goto do_free; | |
1425 | ||
1426 | spin_lock_irqsave(&ctx->inflight_lock, flags); | |
1427 | for (i = 0; i < rb->to_free; i++) { | |
1428 | struct io_kiocb *req = rb->reqs[i]; | |
1429 | ||
10fef4be | 1430 | if (req->flags & REQ_F_INFLIGHT) { |
c6ca97b3 JA |
1431 | list_del(&req->inflight_entry); |
1432 | if (!--inflight) | |
1433 | break; | |
1434 | } | |
1435 | } | |
1436 | spin_unlock_irqrestore(&ctx->inflight_lock, flags); | |
1437 | ||
1438 | if (waitqueue_active(&ctx->inflight_wait)) | |
1439 | wake_up(&ctx->inflight_wait); | |
1440 | } | |
1441 | do_free: | |
1442 | kmem_cache_free_bulk(req_cachep, rb->to_free, rb->reqs); | |
1443 | percpu_ref_put_many(&ctx->refs, rb->to_free); | |
c6ca97b3 | 1444 | rb->to_free = rb->need_iter = 0; |
e65ef56d JA |
1445 | } |
1446 | ||
a197f664 | 1447 | static bool io_link_cancel_timeout(struct io_kiocb *req) |
2665abfd | 1448 | { |
a197f664 | 1449 | struct io_ring_ctx *ctx = req->ctx; |
2665abfd JA |
1450 | int ret; |
1451 | ||
2d28390a | 1452 | ret = hrtimer_try_to_cancel(&req->io->timeout.timer); |
2665abfd | 1453 | if (ret != -1) { |
78e19bbe | 1454 | io_cqring_fill_event(req, -ECANCELED); |
2665abfd JA |
1455 | io_commit_cqring(ctx); |
1456 | req->flags &= ~REQ_F_LINK; | |
ec9c02ad | 1457 | io_put_req(req); |
2665abfd JA |
1458 | return true; |
1459 | } | |
1460 | ||
1461 | return false; | |
e65ef56d JA |
1462 | } |
1463 | ||
ba816ad6 | 1464 | static void io_req_link_next(struct io_kiocb *req, struct io_kiocb **nxtptr) |
9e645e11 | 1465 | { |
2665abfd | 1466 | struct io_ring_ctx *ctx = req->ctx; |
2665abfd | 1467 | bool wake_ev = false; |
9e645e11 | 1468 | |
4d7dd462 JA |
1469 | /* Already got next link */ |
1470 | if (req->flags & REQ_F_LINK_NEXT) | |
1471 | return; | |
1472 | ||
9e645e11 JA |
1473 | /* |
1474 | * The list should never be empty when we are called here. But could | |
1475 | * potentially happen if the chain is messed up, check to be on the | |
1476 | * safe side. | |
1477 | */ | |
4493233e PB |
1478 | while (!list_empty(&req->link_list)) { |
1479 | struct io_kiocb *nxt = list_first_entry(&req->link_list, | |
1480 | struct io_kiocb, link_list); | |
94ae5e77 | 1481 | |
4493233e PB |
1482 | if (unlikely((req->flags & REQ_F_LINK_TIMEOUT) && |
1483 | (nxt->flags & REQ_F_TIMEOUT))) { | |
1484 | list_del_init(&nxt->link_list); | |
94ae5e77 | 1485 | wake_ev |= io_link_cancel_timeout(nxt); |
94ae5e77 JA |
1486 | req->flags &= ~REQ_F_LINK_TIMEOUT; |
1487 | continue; | |
1488 | } | |
9e645e11 | 1489 | |
4493233e PB |
1490 | list_del_init(&req->link_list); |
1491 | if (!list_empty(&nxt->link_list)) | |
1492 | nxt->flags |= REQ_F_LINK; | |
b18fdf71 | 1493 | *nxtptr = nxt; |
94ae5e77 | 1494 | break; |
9e645e11 | 1495 | } |
2665abfd | 1496 | |
4d7dd462 | 1497 | req->flags |= REQ_F_LINK_NEXT; |
2665abfd JA |
1498 | if (wake_ev) |
1499 | io_cqring_ev_posted(ctx); | |
9e645e11 JA |
1500 | } |
1501 | ||
1502 | /* | |
1503 | * Called if REQ_F_LINK is set, and we fail the head request | |
1504 | */ | |
1505 | static void io_fail_links(struct io_kiocb *req) | |
1506 | { | |
2665abfd | 1507 | struct io_ring_ctx *ctx = req->ctx; |
2665abfd JA |
1508 | unsigned long flags; |
1509 | ||
1510 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
9e645e11 JA |
1511 | |
1512 | while (!list_empty(&req->link_list)) { | |
4493233e PB |
1513 | struct io_kiocb *link = list_first_entry(&req->link_list, |
1514 | struct io_kiocb, link_list); | |
9e645e11 | 1515 | |
4493233e | 1516 | list_del_init(&link->link_list); |
c826bd7a | 1517 | trace_io_uring_fail_link(req, link); |
2665abfd JA |
1518 | |
1519 | if ((req->flags & REQ_F_LINK_TIMEOUT) && | |
d625c6ee | 1520 | link->opcode == IORING_OP_LINK_TIMEOUT) { |
a197f664 | 1521 | io_link_cancel_timeout(link); |
2665abfd | 1522 | } else { |
78e19bbe | 1523 | io_cqring_fill_event(link, -ECANCELED); |
978db57e | 1524 | __io_double_put_req(link); |
2665abfd | 1525 | } |
5d960724 | 1526 | req->flags &= ~REQ_F_LINK_TIMEOUT; |
9e645e11 | 1527 | } |
2665abfd JA |
1528 | |
1529 | io_commit_cqring(ctx); | |
1530 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
1531 | io_cqring_ev_posted(ctx); | |
9e645e11 JA |
1532 | } |
1533 | ||
4d7dd462 | 1534 | static void io_req_find_next(struct io_kiocb *req, struct io_kiocb **nxt) |
9e645e11 | 1535 | { |
4d7dd462 | 1536 | if (likely(!(req->flags & REQ_F_LINK))) |
2665abfd | 1537 | return; |
2665abfd | 1538 | |
9e645e11 JA |
1539 | /* |
1540 | * If LINK is set, we have dependent requests in this chain. If we | |
1541 | * didn't fail this request, queue the first one up, moving any other | |
1542 | * dependencies to the next request. In case of failure, fail the rest | |
1543 | * of the chain. | |
1544 | */ | |
2665abfd JA |
1545 | if (req->flags & REQ_F_FAIL_LINK) { |
1546 | io_fail_links(req); | |
7c9e7f0f JA |
1547 | } else if ((req->flags & (REQ_F_LINK_TIMEOUT | REQ_F_COMP_LOCKED)) == |
1548 | REQ_F_LINK_TIMEOUT) { | |
2665abfd JA |
1549 | struct io_ring_ctx *ctx = req->ctx; |
1550 | unsigned long flags; | |
1551 | ||
1552 | /* | |
1553 | * If this is a timeout link, we could be racing with the | |
1554 | * timeout timer. Grab the completion lock for this case to | |
7c9e7f0f | 1555 | * protect against that. |
2665abfd JA |
1556 | */ |
1557 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
1558 | io_req_link_next(req, nxt); | |
1559 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
1560 | } else { | |
1561 | io_req_link_next(req, nxt); | |
9e645e11 | 1562 | } |
4d7dd462 | 1563 | } |
9e645e11 | 1564 | |
c69f8dbe JL |
1565 | static void io_free_req(struct io_kiocb *req) |
1566 | { | |
944e58bf PB |
1567 | struct io_kiocb *nxt = NULL; |
1568 | ||
1569 | io_req_find_next(req, &nxt); | |
70cf9f32 | 1570 | __io_free_req(req); |
944e58bf PB |
1571 | |
1572 | if (nxt) | |
1573 | io_queue_async_work(nxt); | |
c69f8dbe JL |
1574 | } |
1575 | ||
7a743e22 PB |
1576 | static void io_link_work_cb(struct io_wq_work **workptr) |
1577 | { | |
18a542ff PB |
1578 | struct io_kiocb *req = container_of(*workptr, struct io_kiocb, work); |
1579 | struct io_kiocb *link; | |
7a743e22 | 1580 | |
18a542ff | 1581 | link = list_first_entry(&req->link_list, struct io_kiocb, link_list); |
7a743e22 PB |
1582 | io_queue_linked_timeout(link); |
1583 | io_wq_submit_work(workptr); | |
1584 | } | |
1585 | ||
1586 | static void io_wq_assign_next(struct io_wq_work **workptr, struct io_kiocb *nxt) | |
1587 | { | |
1588 | struct io_kiocb *link; | |
8766dd51 PB |
1589 | const struct io_op_def *def = &io_op_defs[nxt->opcode]; |
1590 | ||
1591 | if ((nxt->flags & REQ_F_ISREG) && def->hash_reg_file) | |
1592 | io_wq_hash_work(&nxt->work, file_inode(nxt->file)); | |
7a743e22 PB |
1593 | |
1594 | *workptr = &nxt->work; | |
1595 | link = io_prep_linked_timeout(nxt); | |
18a542ff | 1596 | if (link) |
7a743e22 | 1597 | nxt->work.func = io_link_work_cb; |
7a743e22 PB |
1598 | } |
1599 | ||
ba816ad6 JA |
1600 | /* |
1601 | * Drop reference to request, return next in chain (if there is one) if this | |
1602 | * was the last reference to this request. | |
1603 | */ | |
f9bd67f6 | 1604 | __attribute__((nonnull)) |
ec9c02ad | 1605 | static void io_put_req_find_next(struct io_kiocb *req, struct io_kiocb **nxtptr) |
e65ef56d | 1606 | { |
2a44f467 JA |
1607 | if (refcount_dec_and_test(&req->refs)) { |
1608 | io_req_find_next(req, nxtptr); | |
4d7dd462 | 1609 | __io_free_req(req); |
2a44f467 | 1610 | } |
2b188cc1 JA |
1611 | } |
1612 | ||
e65ef56d JA |
1613 | static void io_put_req(struct io_kiocb *req) |
1614 | { | |
1615 | if (refcount_dec_and_test(&req->refs)) | |
1616 | io_free_req(req); | |
2b188cc1 JA |
1617 | } |
1618 | ||
e9fd9396 PB |
1619 | static void io_steal_work(struct io_kiocb *req, |
1620 | struct io_wq_work **workptr) | |
7a743e22 PB |
1621 | { |
1622 | /* | |
1623 | * It's in an io-wq worker, so there always should be at least | |
1624 | * one reference, which will be dropped in io_put_work() just | |
1625 | * after the current handler returns. | |
1626 | * | |
1627 | * It also means, that if the counter dropped to 1, then there is | |
1628 | * no asynchronous users left, so it's safe to steal the next work. | |
1629 | */ | |
7a743e22 PB |
1630 | if (refcount_read(&req->refs) == 1) { |
1631 | struct io_kiocb *nxt = NULL; | |
1632 | ||
1633 | io_req_find_next(req, &nxt); | |
1634 | if (nxt) | |
1635 | io_wq_assign_next(workptr, nxt); | |
1636 | } | |
1637 | } | |
1638 | ||
978db57e JA |
1639 | /* |
1640 | * Must only be used if we don't need to care about links, usually from | |
1641 | * within the completion handling itself. | |
1642 | */ | |
1643 | static void __io_double_put_req(struct io_kiocb *req) | |
78e19bbe JA |
1644 | { |
1645 | /* drop both submit and complete references */ | |
1646 | if (refcount_sub_and_test(2, &req->refs)) | |
1647 | __io_free_req(req); | |
1648 | } | |
1649 | ||
978db57e JA |
1650 | static void io_double_put_req(struct io_kiocb *req) |
1651 | { | |
1652 | /* drop both submit and complete references */ | |
1653 | if (refcount_sub_and_test(2, &req->refs)) | |
1654 | io_free_req(req); | |
1655 | } | |
1656 | ||
1d7bb1d5 | 1657 | static unsigned io_cqring_events(struct io_ring_ctx *ctx, bool noflush) |
a3a0e43f | 1658 | { |
84f97dc2 JA |
1659 | struct io_rings *rings = ctx->rings; |
1660 | ||
ad3eb2c8 JA |
1661 | if (test_bit(0, &ctx->cq_check_overflow)) { |
1662 | /* | |
1663 | * noflush == true is from the waitqueue handler, just ensure | |
1664 | * we wake up the task, and the next invocation will flush the | |
1665 | * entries. We cannot safely to it from here. | |
1666 | */ | |
1667 | if (noflush && !list_empty(&ctx->cq_overflow_list)) | |
1668 | return -1U; | |
1d7bb1d5 | 1669 | |
ad3eb2c8 JA |
1670 | io_cqring_overflow_flush(ctx, false); |
1671 | } | |
1d7bb1d5 | 1672 | |
a3a0e43f JA |
1673 | /* See comment at the top of this file */ |
1674 | smp_rmb(); | |
ad3eb2c8 | 1675 | return ctx->cached_cq_tail - READ_ONCE(rings->cq.head); |
a3a0e43f JA |
1676 | } |
1677 | ||
fb5ccc98 PB |
1678 | static inline unsigned int io_sqring_entries(struct io_ring_ctx *ctx) |
1679 | { | |
1680 | struct io_rings *rings = ctx->rings; | |
1681 | ||
1682 | /* make sure SQ entry isn't read before tail */ | |
1683 | return smp_load_acquire(&rings->sq.tail) - ctx->cached_sq_head; | |
1684 | } | |
1685 | ||
8237e045 | 1686 | static inline bool io_req_multi_free(struct req_batch *rb, struct io_kiocb *req) |
e94f141b | 1687 | { |
c6ca97b3 JA |
1688 | if ((req->flags & REQ_F_LINK) || io_is_fallback_req(req)) |
1689 | return false; | |
e94f141b | 1690 | |
c6ca97b3 JA |
1691 | if (!(req->flags & REQ_F_FIXED_FILE) || req->io) |
1692 | rb->need_iter++; | |
1693 | ||
1694 | rb->reqs[rb->to_free++] = req; | |
1695 | if (unlikely(rb->to_free == ARRAY_SIZE(rb->reqs))) | |
1696 | io_free_req_many(req->ctx, rb); | |
1697 | return true; | |
e94f141b JA |
1698 | } |
1699 | ||
bcda7baa JA |
1700 | static int io_put_kbuf(struct io_kiocb *req) |
1701 | { | |
4d954c25 | 1702 | struct io_buffer *kbuf; |
bcda7baa JA |
1703 | int cflags; |
1704 | ||
4d954c25 | 1705 | kbuf = (struct io_buffer *) (unsigned long) req->rw.addr; |
bcda7baa JA |
1706 | cflags = kbuf->bid << IORING_CQE_BUFFER_SHIFT; |
1707 | cflags |= IORING_CQE_F_BUFFER; | |
1708 | req->rw.addr = 0; | |
1709 | kfree(kbuf); | |
1710 | return cflags; | |
1711 | } | |
1712 | ||
def596e9 JA |
1713 | /* |
1714 | * Find and free completed poll iocbs | |
1715 | */ | |
1716 | static void io_iopoll_complete(struct io_ring_ctx *ctx, unsigned int *nr_events, | |
1717 | struct list_head *done) | |
1718 | { | |
8237e045 | 1719 | struct req_batch rb; |
def596e9 | 1720 | struct io_kiocb *req; |
def596e9 | 1721 | |
c6ca97b3 | 1722 | rb.to_free = rb.need_iter = 0; |
def596e9 | 1723 | while (!list_empty(done)) { |
bcda7baa JA |
1724 | int cflags = 0; |
1725 | ||
def596e9 JA |
1726 | req = list_first_entry(done, struct io_kiocb, list); |
1727 | list_del(&req->list); | |
1728 | ||
bcda7baa JA |
1729 | if (req->flags & REQ_F_BUFFER_SELECTED) |
1730 | cflags = io_put_kbuf(req); | |
1731 | ||
1732 | __io_cqring_fill_event(req, req->result, cflags); | |
def596e9 JA |
1733 | (*nr_events)++; |
1734 | ||
8237e045 JA |
1735 | if (refcount_dec_and_test(&req->refs) && |
1736 | !io_req_multi_free(&rb, req)) | |
1737 | io_free_req(req); | |
def596e9 | 1738 | } |
def596e9 | 1739 | |
09bb8394 | 1740 | io_commit_cqring(ctx); |
32b2244a XW |
1741 | if (ctx->flags & IORING_SETUP_SQPOLL) |
1742 | io_cqring_ev_posted(ctx); | |
8237e045 | 1743 | io_free_req_many(ctx, &rb); |
def596e9 JA |
1744 | } |
1745 | ||
581f9810 BM |
1746 | static void io_iopoll_queue(struct list_head *again) |
1747 | { | |
1748 | struct io_kiocb *req; | |
1749 | ||
1750 | do { | |
1751 | req = list_first_entry(again, struct io_kiocb, list); | |
1752 | list_del(&req->list); | |
1753 | refcount_inc(&req->refs); | |
1754 | io_queue_async_work(req); | |
1755 | } while (!list_empty(again)); | |
1756 | } | |
1757 | ||
def596e9 JA |
1758 | static int io_do_iopoll(struct io_ring_ctx *ctx, unsigned int *nr_events, |
1759 | long min) | |
1760 | { | |
1761 | struct io_kiocb *req, *tmp; | |
1762 | LIST_HEAD(done); | |
581f9810 | 1763 | LIST_HEAD(again); |
def596e9 JA |
1764 | bool spin; |
1765 | int ret; | |
1766 | ||
1767 | /* | |
1768 | * Only spin for completions if we don't have multiple devices hanging | |
1769 | * off our complete list, and we're under the requested amount. | |
1770 | */ | |
1771 | spin = !ctx->poll_multi_file && *nr_events < min; | |
1772 | ||
1773 | ret = 0; | |
1774 | list_for_each_entry_safe(req, tmp, &ctx->poll_list, list) { | |
9adbd45d | 1775 | struct kiocb *kiocb = &req->rw.kiocb; |
def596e9 JA |
1776 | |
1777 | /* | |
581f9810 BM |
1778 | * Move completed and retryable entries to our local lists. |
1779 | * If we find a request that requires polling, break out | |
1780 | * and complete those lists first, if we have entries there. | |
def596e9 JA |
1781 | */ |
1782 | if (req->flags & REQ_F_IOPOLL_COMPLETED) { | |
1783 | list_move_tail(&req->list, &done); | |
1784 | continue; | |
1785 | } | |
1786 | if (!list_empty(&done)) | |
1787 | break; | |
1788 | ||
581f9810 BM |
1789 | if (req->result == -EAGAIN) { |
1790 | list_move_tail(&req->list, &again); | |
1791 | continue; | |
1792 | } | |
1793 | if (!list_empty(&again)) | |
1794 | break; | |
1795 | ||
def596e9 JA |
1796 | ret = kiocb->ki_filp->f_op->iopoll(kiocb, spin); |
1797 | if (ret < 0) | |
1798 | break; | |
1799 | ||
1800 | if (ret && spin) | |
1801 | spin = false; | |
1802 | ret = 0; | |
1803 | } | |
1804 | ||
1805 | if (!list_empty(&done)) | |
1806 | io_iopoll_complete(ctx, nr_events, &done); | |
1807 | ||
581f9810 BM |
1808 | if (!list_empty(&again)) |
1809 | io_iopoll_queue(&again); | |
1810 | ||
def596e9 JA |
1811 | return ret; |
1812 | } | |
1813 | ||
1814 | /* | |
d195a66e | 1815 | * Poll for a minimum of 'min' events. Note that if min == 0 we consider that a |
def596e9 JA |
1816 | * non-spinning poll check - we'll still enter the driver poll loop, but only |
1817 | * as a non-spinning completion check. | |
1818 | */ | |
1819 | static int io_iopoll_getevents(struct io_ring_ctx *ctx, unsigned int *nr_events, | |
1820 | long min) | |
1821 | { | |
08f5439f | 1822 | while (!list_empty(&ctx->poll_list) && !need_resched()) { |
def596e9 JA |
1823 | int ret; |
1824 | ||
1825 | ret = io_do_iopoll(ctx, nr_events, min); | |
1826 | if (ret < 0) | |
1827 | return ret; | |
1828 | if (!min || *nr_events >= min) | |
1829 | return 0; | |
1830 | } | |
1831 | ||
1832 | return 1; | |
1833 | } | |
1834 | ||
1835 | /* | |
1836 | * We can't just wait for polled events to come to us, we have to actively | |
1837 | * find and complete them. | |
1838 | */ | |
1839 | static void io_iopoll_reap_events(struct io_ring_ctx *ctx) | |
1840 | { | |
1841 | if (!(ctx->flags & IORING_SETUP_IOPOLL)) | |
1842 | return; | |
1843 | ||
1844 | mutex_lock(&ctx->uring_lock); | |
1845 | while (!list_empty(&ctx->poll_list)) { | |
1846 | unsigned int nr_events = 0; | |
1847 | ||
1848 | io_iopoll_getevents(ctx, &nr_events, 1); | |
08f5439f JA |
1849 | |
1850 | /* | |
1851 | * Ensure we allow local-to-the-cpu processing to take place, | |
1852 | * in this case we need to ensure that we reap all events. | |
1853 | */ | |
1854 | cond_resched(); | |
def596e9 JA |
1855 | } |
1856 | mutex_unlock(&ctx->uring_lock); | |
1857 | } | |
1858 | ||
c7849be9 XW |
1859 | static int io_iopoll_check(struct io_ring_ctx *ctx, unsigned *nr_events, |
1860 | long min) | |
def596e9 | 1861 | { |
2b2ed975 | 1862 | int iters = 0, ret = 0; |
500f9fba | 1863 | |
c7849be9 XW |
1864 | /* |
1865 | * We disallow the app entering submit/complete with polling, but we | |
1866 | * still need to lock the ring to prevent racing with polled issue | |
1867 | * that got punted to a workqueue. | |
1868 | */ | |
1869 | mutex_lock(&ctx->uring_lock); | |
def596e9 JA |
1870 | do { |
1871 | int tmin = 0; | |
1872 | ||
a3a0e43f JA |
1873 | /* |
1874 | * Don't enter poll loop if we already have events pending. | |
1875 | * If we do, we can potentially be spinning for commands that | |
1876 | * already triggered a CQE (eg in error). | |
1877 | */ | |
1d7bb1d5 | 1878 | if (io_cqring_events(ctx, false)) |
a3a0e43f JA |
1879 | break; |
1880 | ||
500f9fba JA |
1881 | /* |
1882 | * If a submit got punted to a workqueue, we can have the | |
1883 | * application entering polling for a command before it gets | |
1884 | * issued. That app will hold the uring_lock for the duration | |
1885 | * of the poll right here, so we need to take a breather every | |
1886 | * now and then to ensure that the issue has a chance to add | |
1887 | * the poll to the issued list. Otherwise we can spin here | |
1888 | * forever, while the workqueue is stuck trying to acquire the | |
1889 | * very same mutex. | |
1890 | */ | |
1891 | if (!(++iters & 7)) { | |
1892 | mutex_unlock(&ctx->uring_lock); | |
1893 | mutex_lock(&ctx->uring_lock); | |
1894 | } | |
1895 | ||
def596e9 JA |
1896 | if (*nr_events < min) |
1897 | tmin = min - *nr_events; | |
1898 | ||
1899 | ret = io_iopoll_getevents(ctx, nr_events, tmin); | |
1900 | if (ret <= 0) | |
1901 | break; | |
1902 | ret = 0; | |
1903 | } while (min && !*nr_events && !need_resched()); | |
1904 | ||
500f9fba | 1905 | mutex_unlock(&ctx->uring_lock); |
def596e9 JA |
1906 | return ret; |
1907 | } | |
1908 | ||
491381ce | 1909 | static void kiocb_end_write(struct io_kiocb *req) |
2b188cc1 | 1910 | { |
491381ce JA |
1911 | /* |
1912 | * Tell lockdep we inherited freeze protection from submission | |
1913 | * thread. | |
1914 | */ | |
1915 | if (req->flags & REQ_F_ISREG) { | |
1916 | struct inode *inode = file_inode(req->file); | |
2b188cc1 | 1917 | |
491381ce | 1918 | __sb_writers_acquired(inode->i_sb, SB_FREEZE_WRITE); |
2b188cc1 | 1919 | } |
491381ce | 1920 | file_end_write(req->file); |
2b188cc1 JA |
1921 | } |
1922 | ||
4e88d6e7 JA |
1923 | static inline void req_set_fail_links(struct io_kiocb *req) |
1924 | { | |
1925 | if ((req->flags & (REQ_F_LINK | REQ_F_HARDLINK)) == REQ_F_LINK) | |
1926 | req->flags |= REQ_F_FAIL_LINK; | |
1927 | } | |
1928 | ||
ba816ad6 | 1929 | static void io_complete_rw_common(struct kiocb *kiocb, long res) |
2b188cc1 | 1930 | { |
9adbd45d | 1931 | struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb); |
bcda7baa | 1932 | int cflags = 0; |
2b188cc1 | 1933 | |
491381ce JA |
1934 | if (kiocb->ki_flags & IOCB_WRITE) |
1935 | kiocb_end_write(req); | |
2b188cc1 | 1936 | |
4e88d6e7 JA |
1937 | if (res != req->result) |
1938 | req_set_fail_links(req); | |
bcda7baa JA |
1939 | if (req->flags & REQ_F_BUFFER_SELECTED) |
1940 | cflags = io_put_kbuf(req); | |
1941 | __io_cqring_add_event(req, res, cflags); | |
ba816ad6 JA |
1942 | } |
1943 | ||
1944 | static void io_complete_rw(struct kiocb *kiocb, long res, long res2) | |
1945 | { | |
9adbd45d | 1946 | struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb); |
ba816ad6 JA |
1947 | |
1948 | io_complete_rw_common(kiocb, res); | |
e65ef56d | 1949 | io_put_req(req); |
2b188cc1 JA |
1950 | } |
1951 | ||
def596e9 JA |
1952 | static void io_complete_rw_iopoll(struct kiocb *kiocb, long res, long res2) |
1953 | { | |
9adbd45d | 1954 | struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb); |
def596e9 | 1955 | |
491381ce JA |
1956 | if (kiocb->ki_flags & IOCB_WRITE) |
1957 | kiocb_end_write(req); | |
def596e9 | 1958 | |
4e88d6e7 JA |
1959 | if (res != req->result) |
1960 | req_set_fail_links(req); | |
9e645e11 | 1961 | req->result = res; |
def596e9 JA |
1962 | if (res != -EAGAIN) |
1963 | req->flags |= REQ_F_IOPOLL_COMPLETED; | |
1964 | } | |
1965 | ||
1966 | /* | |
1967 | * After the iocb has been issued, it's safe to be found on the poll list. | |
1968 | * Adding the kiocb to the list AFTER submission ensures that we don't | |
1969 | * find it from a io_iopoll_getevents() thread before the issuer is done | |
1970 | * accessing the kiocb cookie. | |
1971 | */ | |
1972 | static void io_iopoll_req_issued(struct io_kiocb *req) | |
1973 | { | |
1974 | struct io_ring_ctx *ctx = req->ctx; | |
1975 | ||
1976 | /* | |
1977 | * Track whether we have multiple files in our lists. This will impact | |
1978 | * how we do polling eventually, not spinning if we're on potentially | |
1979 | * different devices. | |
1980 | */ | |
1981 | if (list_empty(&ctx->poll_list)) { | |
1982 | ctx->poll_multi_file = false; | |
1983 | } else if (!ctx->poll_multi_file) { | |
1984 | struct io_kiocb *list_req; | |
1985 | ||
1986 | list_req = list_first_entry(&ctx->poll_list, struct io_kiocb, | |
1987 | list); | |
9adbd45d | 1988 | if (list_req->file != req->file) |
def596e9 JA |
1989 | ctx->poll_multi_file = true; |
1990 | } | |
1991 | ||
1992 | /* | |
1993 | * For fast devices, IO may have already completed. If it has, add | |
1994 | * it to the front so we find it first. | |
1995 | */ | |
1996 | if (req->flags & REQ_F_IOPOLL_COMPLETED) | |
1997 | list_add(&req->list, &ctx->poll_list); | |
1998 | else | |
1999 | list_add_tail(&req->list, &ctx->poll_list); | |
bdcd3eab XW |
2000 | |
2001 | if ((ctx->flags & IORING_SETUP_SQPOLL) && | |
2002 | wq_has_sleeper(&ctx->sqo_wait)) | |
2003 | wake_up(&ctx->sqo_wait); | |
def596e9 JA |
2004 | } |
2005 | ||
3d6770fb | 2006 | static void io_file_put(struct io_submit_state *state) |
9a56a232 | 2007 | { |
3d6770fb | 2008 | if (state->file) { |
9a56a232 JA |
2009 | int diff = state->has_refs - state->used_refs; |
2010 | ||
2011 | if (diff) | |
2012 | fput_many(state->file, diff); | |
2013 | state->file = NULL; | |
2014 | } | |
2015 | } | |
2016 | ||
2017 | /* | |
2018 | * Get as many references to a file as we have IOs left in this submission, | |
2019 | * assuming most submissions are for one file, or at least that each file | |
2020 | * has more than one submission. | |
2021 | */ | |
8da11c19 | 2022 | static struct file *__io_file_get(struct io_submit_state *state, int fd) |
9a56a232 JA |
2023 | { |
2024 | if (!state) | |
2025 | return fget(fd); | |
2026 | ||
2027 | if (state->file) { | |
2028 | if (state->fd == fd) { | |
2029 | state->used_refs++; | |
2030 | state->ios_left--; | |
2031 | return state->file; | |
2032 | } | |
3d6770fb | 2033 | io_file_put(state); |
9a56a232 JA |
2034 | } |
2035 | state->file = fget_many(fd, state->ios_left); | |
2036 | if (!state->file) | |
2037 | return NULL; | |
2038 | ||
2039 | state->fd = fd; | |
2040 | state->has_refs = state->ios_left; | |
2041 | state->used_refs = 1; | |
2042 | state->ios_left--; | |
2043 | return state->file; | |
2044 | } | |
2045 | ||
2b188cc1 JA |
2046 | /* |
2047 | * If we tracked the file through the SCM inflight mechanism, we could support | |
2048 | * any file. For now, just ensure that anything potentially problematic is done | |
2049 | * inline. | |
2050 | */ | |
2051 | static bool io_file_supports_async(struct file *file) | |
2052 | { | |
2053 | umode_t mode = file_inode(file)->i_mode; | |
2054 | ||
10d59345 | 2055 | if (S_ISBLK(mode) || S_ISCHR(mode) || S_ISSOCK(mode)) |
2b188cc1 JA |
2056 | return true; |
2057 | if (S_ISREG(mode) && file->f_op != &io_uring_fops) | |
2058 | return true; | |
2059 | ||
2060 | return false; | |
2061 | } | |
2062 | ||
3529d8c2 JA |
2063 | static int io_prep_rw(struct io_kiocb *req, const struct io_uring_sqe *sqe, |
2064 | bool force_nonblock) | |
2b188cc1 | 2065 | { |
def596e9 | 2066 | struct io_ring_ctx *ctx = req->ctx; |
9adbd45d | 2067 | struct kiocb *kiocb = &req->rw.kiocb; |
09bb8394 JA |
2068 | unsigned ioprio; |
2069 | int ret; | |
2b188cc1 | 2070 | |
491381ce JA |
2071 | if (S_ISREG(file_inode(req->file)->i_mode)) |
2072 | req->flags |= REQ_F_ISREG; | |
2073 | ||
2b188cc1 | 2074 | kiocb->ki_pos = READ_ONCE(sqe->off); |
ba04291e JA |
2075 | if (kiocb->ki_pos == -1 && !(req->file->f_mode & FMODE_STREAM)) { |
2076 | req->flags |= REQ_F_CUR_POS; | |
2077 | kiocb->ki_pos = req->file->f_pos; | |
2078 | } | |
2b188cc1 | 2079 | kiocb->ki_hint = ki_hint_validate(file_write_hint(kiocb->ki_filp)); |
3e577dcd PB |
2080 | kiocb->ki_flags = iocb_flags(kiocb->ki_filp); |
2081 | ret = kiocb_set_rw_flags(kiocb, READ_ONCE(sqe->rw_flags)); | |
2082 | if (unlikely(ret)) | |
2083 | return ret; | |
2b188cc1 JA |
2084 | |
2085 | ioprio = READ_ONCE(sqe->ioprio); | |
2086 | if (ioprio) { | |
2087 | ret = ioprio_check_cap(ioprio); | |
2088 | if (ret) | |
09bb8394 | 2089 | return ret; |
2b188cc1 JA |
2090 | |
2091 | kiocb->ki_ioprio = ioprio; | |
2092 | } else | |
2093 | kiocb->ki_ioprio = get_current_ioprio(); | |
2094 | ||
8449eeda | 2095 | /* don't allow async punt if RWF_NOWAIT was requested */ |
491381ce JA |
2096 | if ((kiocb->ki_flags & IOCB_NOWAIT) || |
2097 | (req->file->f_flags & O_NONBLOCK)) | |
8449eeda SB |
2098 | req->flags |= REQ_F_NOWAIT; |
2099 | ||
2100 | if (force_nonblock) | |
2b188cc1 | 2101 | kiocb->ki_flags |= IOCB_NOWAIT; |
8449eeda | 2102 | |
def596e9 | 2103 | if (ctx->flags & IORING_SETUP_IOPOLL) { |
def596e9 JA |
2104 | if (!(kiocb->ki_flags & IOCB_DIRECT) || |
2105 | !kiocb->ki_filp->f_op->iopoll) | |
09bb8394 | 2106 | return -EOPNOTSUPP; |
2b188cc1 | 2107 | |
def596e9 JA |
2108 | kiocb->ki_flags |= IOCB_HIPRI; |
2109 | kiocb->ki_complete = io_complete_rw_iopoll; | |
6873e0bd | 2110 | req->result = 0; |
def596e9 | 2111 | } else { |
09bb8394 JA |
2112 | if (kiocb->ki_flags & IOCB_HIPRI) |
2113 | return -EINVAL; | |
def596e9 JA |
2114 | kiocb->ki_complete = io_complete_rw; |
2115 | } | |
9adbd45d | 2116 | |
3529d8c2 JA |
2117 | req->rw.addr = READ_ONCE(sqe->addr); |
2118 | req->rw.len = READ_ONCE(sqe->len); | |
bcda7baa | 2119 | /* we own ->private, reuse it for the buffer index / buffer ID */ |
9adbd45d | 2120 | req->rw.kiocb.private = (void *) (unsigned long) |
3529d8c2 | 2121 | READ_ONCE(sqe->buf_index); |
2b188cc1 | 2122 | return 0; |
2b188cc1 JA |
2123 | } |
2124 | ||
2125 | static inline void io_rw_done(struct kiocb *kiocb, ssize_t ret) | |
2126 | { | |
2127 | switch (ret) { | |
2128 | case -EIOCBQUEUED: | |
2129 | break; | |
2130 | case -ERESTARTSYS: | |
2131 | case -ERESTARTNOINTR: | |
2132 | case -ERESTARTNOHAND: | |
2133 | case -ERESTART_RESTARTBLOCK: | |
2134 | /* | |
2135 | * We can't just restart the syscall, since previously | |
2136 | * submitted sqes may already be in progress. Just fail this | |
2137 | * IO with EINTR. | |
2138 | */ | |
2139 | ret = -EINTR; | |
2140 | /* fall through */ | |
2141 | default: | |
2142 | kiocb->ki_complete(kiocb, ret, 0); | |
2143 | } | |
2144 | } | |
2145 | ||
014db007 | 2146 | static void kiocb_done(struct kiocb *kiocb, ssize_t ret) |
ba816ad6 | 2147 | { |
ba04291e JA |
2148 | struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb); |
2149 | ||
2150 | if (req->flags & REQ_F_CUR_POS) | |
2151 | req->file->f_pos = kiocb->ki_pos; | |
bcaec089 | 2152 | if (ret >= 0 && kiocb->ki_complete == io_complete_rw) |
014db007 | 2153 | io_complete_rw(kiocb, ret, 0); |
ba816ad6 JA |
2154 | else |
2155 | io_rw_done(kiocb, ret); | |
2156 | } | |
2157 | ||
9adbd45d | 2158 | static ssize_t io_import_fixed(struct io_kiocb *req, int rw, |
7d009165 | 2159 | struct iov_iter *iter) |
edafccee | 2160 | { |
9adbd45d JA |
2161 | struct io_ring_ctx *ctx = req->ctx; |
2162 | size_t len = req->rw.len; | |
edafccee JA |
2163 | struct io_mapped_ubuf *imu; |
2164 | unsigned index, buf_index; | |
2165 | size_t offset; | |
2166 | u64 buf_addr; | |
2167 | ||
2168 | /* attempt to use fixed buffers without having provided iovecs */ | |
2169 | if (unlikely(!ctx->user_bufs)) | |
2170 | return -EFAULT; | |
2171 | ||
9adbd45d | 2172 | buf_index = (unsigned long) req->rw.kiocb.private; |
edafccee JA |
2173 | if (unlikely(buf_index >= ctx->nr_user_bufs)) |
2174 | return -EFAULT; | |
2175 | ||
2176 | index = array_index_nospec(buf_index, ctx->nr_user_bufs); | |
2177 | imu = &ctx->user_bufs[index]; | |
9adbd45d | 2178 | buf_addr = req->rw.addr; |
edafccee JA |
2179 | |
2180 | /* overflow */ | |
2181 | if (buf_addr + len < buf_addr) | |
2182 | return -EFAULT; | |
2183 | /* not inside the mapped region */ | |
2184 | if (buf_addr < imu->ubuf || buf_addr + len > imu->ubuf + imu->len) | |
2185 | return -EFAULT; | |
2186 | ||
2187 | /* | |
2188 | * May not be a start of buffer, set size appropriately | |
2189 | * and advance us to the beginning. | |
2190 | */ | |
2191 | offset = buf_addr - imu->ubuf; | |
2192 | iov_iter_bvec(iter, rw, imu->bvec, imu->nr_bvecs, offset + len); | |
bd11b3a3 JA |
2193 | |
2194 | if (offset) { | |
2195 | /* | |
2196 | * Don't use iov_iter_advance() here, as it's really slow for | |
2197 | * using the latter parts of a big fixed buffer - it iterates | |
2198 | * over each segment manually. We can cheat a bit here, because | |
2199 | * we know that: | |
2200 | * | |
2201 | * 1) it's a BVEC iter, we set it up | |
2202 | * 2) all bvecs are PAGE_SIZE in size, except potentially the | |
2203 | * first and last bvec | |
2204 | * | |
2205 | * So just find our index, and adjust the iterator afterwards. | |
2206 | * If the offset is within the first bvec (or the whole first | |
2207 | * bvec, just use iov_iter_advance(). This makes it easier | |
2208 | * since we can just skip the first segment, which may not | |
2209 | * be PAGE_SIZE aligned. | |
2210 | */ | |
2211 | const struct bio_vec *bvec = imu->bvec; | |
2212 | ||
2213 | if (offset <= bvec->bv_len) { | |
2214 | iov_iter_advance(iter, offset); | |
2215 | } else { | |
2216 | unsigned long seg_skip; | |
2217 | ||
2218 | /* skip first vec */ | |
2219 | offset -= bvec->bv_len; | |
2220 | seg_skip = 1 + (offset >> PAGE_SHIFT); | |
2221 | ||
2222 | iter->bvec = bvec + seg_skip; | |
2223 | iter->nr_segs -= seg_skip; | |
99c79f66 | 2224 | iter->count -= bvec->bv_len + offset; |
bd11b3a3 | 2225 | iter->iov_offset = offset & ~PAGE_MASK; |
bd11b3a3 JA |
2226 | } |
2227 | } | |
2228 | ||
5e559561 | 2229 | return len; |
edafccee JA |
2230 | } |
2231 | ||
bcda7baa JA |
2232 | static void io_ring_submit_unlock(struct io_ring_ctx *ctx, bool needs_lock) |
2233 | { | |
2234 | if (needs_lock) | |
2235 | mutex_unlock(&ctx->uring_lock); | |
2236 | } | |
2237 | ||
2238 | static void io_ring_submit_lock(struct io_ring_ctx *ctx, bool needs_lock) | |
2239 | { | |
2240 | /* | |
2241 | * "Normal" inline submissions always hold the uring_lock, since we | |
2242 | * grab it from the system call. Same is true for the SQPOLL offload. | |
2243 | * The only exception is when we've detached the request and issue it | |
2244 | * from an async worker thread, grab the lock for that case. | |
2245 | */ | |
2246 | if (needs_lock) | |
2247 | mutex_lock(&ctx->uring_lock); | |
2248 | } | |
2249 | ||
2250 | static struct io_buffer *io_buffer_select(struct io_kiocb *req, size_t *len, | |
2251 | int bgid, struct io_buffer *kbuf, | |
2252 | bool needs_lock) | |
2253 | { | |
2254 | struct io_buffer *head; | |
2255 | ||
2256 | if (req->flags & REQ_F_BUFFER_SELECTED) | |
2257 | return kbuf; | |
2258 | ||
2259 | io_ring_submit_lock(req->ctx, needs_lock); | |
2260 | ||
2261 | lockdep_assert_held(&req->ctx->uring_lock); | |
2262 | ||
2263 | head = idr_find(&req->ctx->io_buffer_idr, bgid); | |
2264 | if (head) { | |
2265 | if (!list_empty(&head->list)) { | |
2266 | kbuf = list_last_entry(&head->list, struct io_buffer, | |
2267 | list); | |
2268 | list_del(&kbuf->list); | |
2269 | } else { | |
2270 | kbuf = head; | |
2271 | idr_remove(&req->ctx->io_buffer_idr, bgid); | |
2272 | } | |
2273 | if (*len > kbuf->len) | |
2274 | *len = kbuf->len; | |
2275 | } else { | |
2276 | kbuf = ERR_PTR(-ENOBUFS); | |
2277 | } | |
2278 | ||
2279 | io_ring_submit_unlock(req->ctx, needs_lock); | |
2280 | ||
2281 | return kbuf; | |
2282 | } | |
2283 | ||
4d954c25 JA |
2284 | static void __user *io_rw_buffer_select(struct io_kiocb *req, size_t *len, |
2285 | bool needs_lock) | |
2286 | { | |
2287 | struct io_buffer *kbuf; | |
2288 | int bgid; | |
2289 | ||
2290 | kbuf = (struct io_buffer *) (unsigned long) req->rw.addr; | |
2291 | bgid = (int) (unsigned long) req->rw.kiocb.private; | |
2292 | kbuf = io_buffer_select(req, len, bgid, kbuf, needs_lock); | |
2293 | if (IS_ERR(kbuf)) | |
2294 | return kbuf; | |
2295 | req->rw.addr = (u64) (unsigned long) kbuf; | |
2296 | req->flags |= REQ_F_BUFFER_SELECTED; | |
2297 | return u64_to_user_ptr(kbuf->addr); | |
2298 | } | |
2299 | ||
2300 | #ifdef CONFIG_COMPAT | |
2301 | static ssize_t io_compat_import(struct io_kiocb *req, struct iovec *iov, | |
2302 | bool needs_lock) | |
2303 | { | |
2304 | struct compat_iovec __user *uiov; | |
2305 | compat_ssize_t clen; | |
2306 | void __user *buf; | |
2307 | ssize_t len; | |
2308 | ||
2309 | uiov = u64_to_user_ptr(req->rw.addr); | |
2310 | if (!access_ok(uiov, sizeof(*uiov))) | |
2311 | return -EFAULT; | |
2312 | if (__get_user(clen, &uiov->iov_len)) | |
2313 | return -EFAULT; | |
2314 | if (clen < 0) | |
2315 | return -EINVAL; | |
2316 | ||
2317 | len = clen; | |
2318 | buf = io_rw_buffer_select(req, &len, needs_lock); | |
2319 | if (IS_ERR(buf)) | |
2320 | return PTR_ERR(buf); | |
2321 | iov[0].iov_base = buf; | |
2322 | iov[0].iov_len = (compat_size_t) len; | |
2323 | return 0; | |
2324 | } | |
2325 | #endif | |
2326 | ||
2327 | static ssize_t __io_iov_buffer_select(struct io_kiocb *req, struct iovec *iov, | |
2328 | bool needs_lock) | |
2329 | { | |
2330 | struct iovec __user *uiov = u64_to_user_ptr(req->rw.addr); | |
2331 | void __user *buf; | |
2332 | ssize_t len; | |
2333 | ||
2334 | if (copy_from_user(iov, uiov, sizeof(*uiov))) | |
2335 | return -EFAULT; | |
2336 | ||
2337 | len = iov[0].iov_len; | |
2338 | if (len < 0) | |
2339 | return -EINVAL; | |
2340 | buf = io_rw_buffer_select(req, &len, needs_lock); | |
2341 | if (IS_ERR(buf)) | |
2342 | return PTR_ERR(buf); | |
2343 | iov[0].iov_base = buf; | |
2344 | iov[0].iov_len = len; | |
2345 | return 0; | |
2346 | } | |
2347 | ||
2348 | static ssize_t io_iov_buffer_select(struct io_kiocb *req, struct iovec *iov, | |
2349 | bool needs_lock) | |
2350 | { | |
2351 | if (req->flags & REQ_F_BUFFER_SELECTED) | |
2352 | return 0; | |
2353 | if (!req->rw.len) | |
2354 | return 0; | |
2355 | else if (req->rw.len > 1) | |
2356 | return -EINVAL; | |
2357 | ||
2358 | #ifdef CONFIG_COMPAT | |
2359 | if (req->ctx->compat) | |
2360 | return io_compat_import(req, iov, needs_lock); | |
2361 | #endif | |
2362 | ||
2363 | return __io_iov_buffer_select(req, iov, needs_lock); | |
2364 | } | |
2365 | ||
cf6fd4bd | 2366 | static ssize_t io_import_iovec(int rw, struct io_kiocb *req, |
bcda7baa JA |
2367 | struct iovec **iovec, struct iov_iter *iter, |
2368 | bool needs_lock) | |
2b188cc1 | 2369 | { |
9adbd45d JA |
2370 | void __user *buf = u64_to_user_ptr(req->rw.addr); |
2371 | size_t sqe_len = req->rw.len; | |
4d954c25 | 2372 | ssize_t ret; |
edafccee JA |
2373 | u8 opcode; |
2374 | ||
d625c6ee | 2375 | opcode = req->opcode; |
7d009165 | 2376 | if (opcode == IORING_OP_READ_FIXED || opcode == IORING_OP_WRITE_FIXED) { |
edafccee | 2377 | *iovec = NULL; |
9adbd45d | 2378 | return io_import_fixed(req, rw, iter); |
edafccee | 2379 | } |
2b188cc1 | 2380 | |
bcda7baa JA |
2381 | /* buffer index only valid with fixed read/write, or buffer select */ |
2382 | if (req->rw.kiocb.private && !(req->flags & REQ_F_BUFFER_SELECT)) | |
9adbd45d JA |
2383 | return -EINVAL; |
2384 | ||
3a6820f2 | 2385 | if (opcode == IORING_OP_READ || opcode == IORING_OP_WRITE) { |
bcda7baa | 2386 | if (req->flags & REQ_F_BUFFER_SELECT) { |
4d954c25 JA |
2387 | buf = io_rw_buffer_select(req, &sqe_len, needs_lock); |
2388 | if (IS_ERR(buf)) { | |
bcda7baa | 2389 | *iovec = NULL; |
4d954c25 | 2390 | return PTR_ERR(buf); |
bcda7baa | 2391 | } |
3f9d6441 | 2392 | req->rw.len = sqe_len; |
bcda7baa JA |
2393 | } |
2394 | ||
3a6820f2 JA |
2395 | ret = import_single_range(rw, buf, sqe_len, *iovec, iter); |
2396 | *iovec = NULL; | |
3a901598 | 2397 | return ret < 0 ? ret : sqe_len; |
3a6820f2 JA |
2398 | } |
2399 | ||
f67676d1 JA |
2400 | if (req->io) { |
2401 | struct io_async_rw *iorw = &req->io->rw; | |
2402 | ||
2403 | *iovec = iorw->iov; | |
2404 | iov_iter_init(iter, rw, *iovec, iorw->nr_segs, iorw->size); | |
2405 | if (iorw->iov == iorw->fast_iov) | |
2406 | *iovec = NULL; | |
2407 | return iorw->size; | |
2408 | } | |
2409 | ||
4d954c25 JA |
2410 | if (req->flags & REQ_F_BUFFER_SELECT) { |
2411 | ret = io_iov_buffer_select(req, *iovec, needs_lock); | |
3f9d6441 JA |
2412 | if (!ret) { |
2413 | ret = (*iovec)->iov_len; | |
2414 | iov_iter_init(iter, rw, *iovec, 1, ret); | |
2415 | } | |
4d954c25 JA |
2416 | *iovec = NULL; |
2417 | return ret; | |
2418 | } | |
2419 | ||
2b188cc1 | 2420 | #ifdef CONFIG_COMPAT |
cf6fd4bd | 2421 | if (req->ctx->compat) |
2b188cc1 JA |
2422 | return compat_import_iovec(rw, buf, sqe_len, UIO_FASTIOV, |
2423 | iovec, iter); | |
2424 | #endif | |
2425 | ||
2426 | return import_iovec(rw, buf, sqe_len, UIO_FASTIOV, iovec, iter); | |
2427 | } | |
2428 | ||
31b51510 | 2429 | /* |
32960613 JA |
2430 | * For files that don't have ->read_iter() and ->write_iter(), handle them |
2431 | * by looping over ->read() or ->write() manually. | |
31b51510 | 2432 | */ |
32960613 JA |
2433 | static ssize_t loop_rw_iter(int rw, struct file *file, struct kiocb *kiocb, |
2434 | struct iov_iter *iter) | |
2435 | { | |
2436 | ssize_t ret = 0; | |
2437 | ||
2438 | /* | |
2439 | * Don't support polled IO through this interface, and we can't | |
2440 | * support non-blocking either. For the latter, this just causes | |
2441 | * the kiocb to be handled from an async context. | |
2442 | */ | |
2443 | if (kiocb->ki_flags & IOCB_HIPRI) | |
2444 | return -EOPNOTSUPP; | |
2445 | if (kiocb->ki_flags & IOCB_NOWAIT) | |
2446 | return -EAGAIN; | |
2447 | ||
2448 | while (iov_iter_count(iter)) { | |
311ae9e1 | 2449 | struct iovec iovec; |
32960613 JA |
2450 | ssize_t nr; |
2451 | ||
311ae9e1 PB |
2452 | if (!iov_iter_is_bvec(iter)) { |
2453 | iovec = iov_iter_iovec(iter); | |
2454 | } else { | |
2455 | /* fixed buffers import bvec */ | |
2456 | iovec.iov_base = kmap(iter->bvec->bv_page) | |
2457 | + iter->iov_offset; | |
2458 | iovec.iov_len = min(iter->count, | |
2459 | iter->bvec->bv_len - iter->iov_offset); | |
2460 | } | |
2461 | ||
32960613 JA |
2462 | if (rw == READ) { |
2463 | nr = file->f_op->read(file, iovec.iov_base, | |
2464 | iovec.iov_len, &kiocb->ki_pos); | |
2465 | } else { | |
2466 | nr = file->f_op->write(file, iovec.iov_base, | |
2467 | iovec.iov_len, &kiocb->ki_pos); | |
2468 | } | |
2469 | ||
311ae9e1 PB |
2470 | if (iov_iter_is_bvec(iter)) |
2471 | kunmap(iter->bvec->bv_page); | |
2472 | ||
32960613 JA |
2473 | if (nr < 0) { |
2474 | if (!ret) | |
2475 | ret = nr; | |
2476 | break; | |
2477 | } | |
2478 | ret += nr; | |
2479 | if (nr != iovec.iov_len) | |
2480 | break; | |
2481 | iov_iter_advance(iter, nr); | |
2482 | } | |
2483 | ||
2484 | return ret; | |
2485 | } | |
2486 | ||
b7bb4f7d | 2487 | static void io_req_map_rw(struct io_kiocb *req, ssize_t io_size, |
f67676d1 JA |
2488 | struct iovec *iovec, struct iovec *fast_iov, |
2489 | struct iov_iter *iter) | |
2490 | { | |
2491 | req->io->rw.nr_segs = iter->nr_segs; | |
2492 | req->io->rw.size = io_size; | |
2493 | req->io->rw.iov = iovec; | |
2494 | if (!req->io->rw.iov) { | |
2495 | req->io->rw.iov = req->io->rw.fast_iov; | |
45097dae XW |
2496 | if (req->io->rw.iov != fast_iov) |
2497 | memcpy(req->io->rw.iov, fast_iov, | |
2498 | sizeof(struct iovec) * iter->nr_segs); | |
99bc4c38 PB |
2499 | } else { |
2500 | req->flags |= REQ_F_NEED_CLEANUP; | |
f67676d1 JA |
2501 | } |
2502 | } | |
2503 | ||
3d9932a8 XW |
2504 | static inline int __io_alloc_async_ctx(struct io_kiocb *req) |
2505 | { | |
2506 | req->io = kmalloc(sizeof(*req->io), GFP_KERNEL); | |
2507 | return req->io == NULL; | |
2508 | } | |
2509 | ||
b7bb4f7d | 2510 | static int io_alloc_async_ctx(struct io_kiocb *req) |
f67676d1 | 2511 | { |
d3656344 JA |
2512 | if (!io_op_defs[req->opcode].async_ctx) |
2513 | return 0; | |
3d9932a8 XW |
2514 | |
2515 | return __io_alloc_async_ctx(req); | |
b7bb4f7d JA |
2516 | } |
2517 | ||
b7bb4f7d JA |
2518 | static int io_setup_async_rw(struct io_kiocb *req, ssize_t io_size, |
2519 | struct iovec *iovec, struct iovec *fast_iov, | |
2520 | struct iov_iter *iter) | |
2521 | { | |
980ad263 | 2522 | if (!io_op_defs[req->opcode].async_ctx) |
74566df3 | 2523 | return 0; |
5d204bcf | 2524 | if (!req->io) { |
3d9932a8 | 2525 | if (__io_alloc_async_ctx(req)) |
5d204bcf | 2526 | return -ENOMEM; |
b7bb4f7d | 2527 | |
5d204bcf JA |
2528 | io_req_map_rw(req, io_size, iovec, fast_iov, iter); |
2529 | } | |
b7bb4f7d | 2530 | return 0; |
f67676d1 JA |
2531 | } |
2532 | ||
3529d8c2 JA |
2533 | static int io_read_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe, |
2534 | bool force_nonblock) | |
f67676d1 | 2535 | { |
3529d8c2 JA |
2536 | struct io_async_ctx *io; |
2537 | struct iov_iter iter; | |
f67676d1 JA |
2538 | ssize_t ret; |
2539 | ||
3529d8c2 JA |
2540 | ret = io_prep_rw(req, sqe, force_nonblock); |
2541 | if (ret) | |
2542 | return ret; | |
f67676d1 | 2543 | |
3529d8c2 JA |
2544 | if (unlikely(!(req->file->f_mode & FMODE_READ))) |
2545 | return -EBADF; | |
f67676d1 | 2546 | |
5f798bea PB |
2547 | /* either don't need iovec imported or already have it */ |
2548 | if (!req->io || req->flags & REQ_F_NEED_CLEANUP) | |
3529d8c2 JA |
2549 | return 0; |
2550 | ||
2551 | io = req->io; | |
2552 | io->rw.iov = io->rw.fast_iov; | |
2553 | req->io = NULL; | |
bcda7baa | 2554 | ret = io_import_iovec(READ, req, &io->rw.iov, &iter, !force_nonblock); |
3529d8c2 JA |
2555 | req->io = io; |
2556 | if (ret < 0) | |
2557 | return ret; | |
2558 | ||
2559 | io_req_map_rw(req, ret, io->rw.iov, io->rw.fast_iov, &iter); | |
2560 | return 0; | |
f67676d1 JA |
2561 | } |
2562 | ||
014db007 | 2563 | static int io_read(struct io_kiocb *req, bool force_nonblock) |
2b188cc1 JA |
2564 | { |
2565 | struct iovec inline_vecs[UIO_FASTIOV], *iovec = inline_vecs; | |
9adbd45d | 2566 | struct kiocb *kiocb = &req->rw.kiocb; |
2b188cc1 | 2567 | struct iov_iter iter; |
31b51510 | 2568 | size_t iov_count; |
f67676d1 | 2569 | ssize_t io_size, ret; |
2b188cc1 | 2570 | |
bcda7baa | 2571 | ret = io_import_iovec(READ, req, &iovec, &iter, !force_nonblock); |
06b76d44 JA |
2572 | if (ret < 0) |
2573 | return ret; | |
2b188cc1 | 2574 | |
fd6c2e4c JA |
2575 | /* Ensure we clear previously set non-block flag */ |
2576 | if (!force_nonblock) | |
29de5f6a | 2577 | kiocb->ki_flags &= ~IOCB_NOWAIT; |
fd6c2e4c | 2578 | |
797f3f53 | 2579 | req->result = 0; |
f67676d1 | 2580 | io_size = ret; |
9e645e11 | 2581 | if (req->flags & REQ_F_LINK) |
f67676d1 JA |
2582 | req->result = io_size; |
2583 | ||
2584 | /* | |
2585 | * If the file doesn't support async, mark it as REQ_F_MUST_PUNT so | |
2586 | * we know to async punt it even if it was opened O_NONBLOCK | |
2587 | */ | |
29de5f6a | 2588 | if (force_nonblock && !io_file_supports_async(req->file)) |
f67676d1 | 2589 | goto copy_iov; |
9e645e11 | 2590 | |
31b51510 | 2591 | iov_count = iov_iter_count(&iter); |
9adbd45d | 2592 | ret = rw_verify_area(READ, req->file, &kiocb->ki_pos, iov_count); |
2b188cc1 JA |
2593 | if (!ret) { |
2594 | ssize_t ret2; | |
2595 | ||
9adbd45d JA |
2596 | if (req->file->f_op->read_iter) |
2597 | ret2 = call_read_iter(req->file, kiocb, &iter); | |
32960613 | 2598 | else |
9adbd45d | 2599 | ret2 = loop_rw_iter(READ, req->file, kiocb, &iter); |
32960613 | 2600 | |
9d93a3f5 | 2601 | /* Catch -EAGAIN return for forced non-blocking submission */ |
f67676d1 | 2602 | if (!force_nonblock || ret2 != -EAGAIN) { |
014db007 | 2603 | kiocb_done(kiocb, ret2); |
f67676d1 JA |
2604 | } else { |
2605 | copy_iov: | |
b7bb4f7d | 2606 | ret = io_setup_async_rw(req, io_size, iovec, |
f67676d1 JA |
2607 | inline_vecs, &iter); |
2608 | if (ret) | |
2609 | goto out_free; | |
29de5f6a JA |
2610 | /* any defer here is final, must blocking retry */ |
2611 | if (!(req->flags & REQ_F_NOWAIT)) | |
2612 | req->flags |= REQ_F_MUST_PUNT; | |
f67676d1 JA |
2613 | return -EAGAIN; |
2614 | } | |
2b188cc1 | 2615 | } |
f67676d1 | 2616 | out_free: |
1e95081c | 2617 | kfree(iovec); |
99bc4c38 | 2618 | req->flags &= ~REQ_F_NEED_CLEANUP; |
2b188cc1 JA |
2619 | return ret; |
2620 | } | |
2621 | ||
3529d8c2 JA |
2622 | static int io_write_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe, |
2623 | bool force_nonblock) | |
f67676d1 | 2624 | { |
3529d8c2 JA |
2625 | struct io_async_ctx *io; |
2626 | struct iov_iter iter; | |
f67676d1 JA |
2627 | ssize_t ret; |
2628 | ||
3529d8c2 JA |
2629 | ret = io_prep_rw(req, sqe, force_nonblock); |
2630 | if (ret) | |
2631 | return ret; | |
f67676d1 | 2632 | |
3529d8c2 JA |
2633 | if (unlikely(!(req->file->f_mode & FMODE_WRITE))) |
2634 | return -EBADF; | |
f67676d1 | 2635 | |
4ed734b0 JA |
2636 | req->fsize = rlimit(RLIMIT_FSIZE); |
2637 | ||
5f798bea PB |
2638 | /* either don't need iovec imported or already have it */ |
2639 | if (!req->io || req->flags & REQ_F_NEED_CLEANUP) | |
3529d8c2 JA |
2640 | return 0; |
2641 | ||
2642 | io = req->io; | |
2643 | io->rw.iov = io->rw.fast_iov; | |
2644 | req->io = NULL; | |
bcda7baa | 2645 | ret = io_import_iovec(WRITE, req, &io->rw.iov, &iter, !force_nonblock); |
3529d8c2 JA |
2646 | req->io = io; |
2647 | if (ret < 0) | |
2648 | return ret; | |
2649 | ||
2650 | io_req_map_rw(req, ret, io->rw.iov, io->rw.fast_iov, &iter); | |
2651 | return 0; | |
f67676d1 JA |
2652 | } |
2653 | ||
014db007 | 2654 | static int io_write(struct io_kiocb *req, bool force_nonblock) |
2b188cc1 JA |
2655 | { |
2656 | struct iovec inline_vecs[UIO_FASTIOV], *iovec = inline_vecs; | |
9adbd45d | 2657 | struct kiocb *kiocb = &req->rw.kiocb; |
2b188cc1 | 2658 | struct iov_iter iter; |
31b51510 | 2659 | size_t iov_count; |
f67676d1 | 2660 | ssize_t ret, io_size; |
2b188cc1 | 2661 | |
bcda7baa | 2662 | ret = io_import_iovec(WRITE, req, &iovec, &iter, !force_nonblock); |
06b76d44 JA |
2663 | if (ret < 0) |
2664 | return ret; | |
2b188cc1 | 2665 | |
fd6c2e4c JA |
2666 | /* Ensure we clear previously set non-block flag */ |
2667 | if (!force_nonblock) | |
9adbd45d | 2668 | req->rw.kiocb.ki_flags &= ~IOCB_NOWAIT; |
fd6c2e4c | 2669 | |
797f3f53 | 2670 | req->result = 0; |
f67676d1 | 2671 | io_size = ret; |
9e645e11 | 2672 | if (req->flags & REQ_F_LINK) |
f67676d1 | 2673 | req->result = io_size; |
9e645e11 | 2674 | |
f67676d1 JA |
2675 | /* |
2676 | * If the file doesn't support async, mark it as REQ_F_MUST_PUNT so | |
2677 | * we know to async punt it even if it was opened O_NONBLOCK | |
2678 | */ | |
29de5f6a | 2679 | if (force_nonblock && !io_file_supports_async(req->file)) |
f67676d1 | 2680 | goto copy_iov; |
31b51510 | 2681 | |
10d59345 JA |
2682 | /* file path doesn't support NOWAIT for non-direct_IO */ |
2683 | if (force_nonblock && !(kiocb->ki_flags & IOCB_DIRECT) && | |
2684 | (req->flags & REQ_F_ISREG)) | |
f67676d1 | 2685 | goto copy_iov; |
31b51510 | 2686 | |
f67676d1 | 2687 | iov_count = iov_iter_count(&iter); |
9adbd45d | 2688 | ret = rw_verify_area(WRITE, req->file, &kiocb->ki_pos, iov_count); |
2b188cc1 | 2689 | if (!ret) { |
9bf7933f RP |
2690 | ssize_t ret2; |
2691 | ||
2b188cc1 JA |
2692 | /* |
2693 | * Open-code file_start_write here to grab freeze protection, | |
2694 | * which will be released by another thread in | |
2695 | * io_complete_rw(). Fool lockdep by telling it the lock got | |
2696 | * released so that it doesn't complain about the held lock when | |
2697 | * we return to userspace. | |
2698 | */ | |
491381ce | 2699 | if (req->flags & REQ_F_ISREG) { |
9adbd45d | 2700 | __sb_start_write(file_inode(req->file)->i_sb, |
2b188cc1 | 2701 | SB_FREEZE_WRITE, true); |
9adbd45d | 2702 | __sb_writers_release(file_inode(req->file)->i_sb, |
2b188cc1 JA |
2703 | SB_FREEZE_WRITE); |
2704 | } | |
2705 | kiocb->ki_flags |= IOCB_WRITE; | |
9bf7933f | 2706 | |
4ed734b0 JA |
2707 | if (!force_nonblock) |
2708 | current->signal->rlim[RLIMIT_FSIZE].rlim_cur = req->fsize; | |
2709 | ||
9adbd45d JA |
2710 | if (req->file->f_op->write_iter) |
2711 | ret2 = call_write_iter(req->file, kiocb, &iter); | |
32960613 | 2712 | else |
9adbd45d | 2713 | ret2 = loop_rw_iter(WRITE, req->file, kiocb, &iter); |
4ed734b0 JA |
2714 | |
2715 | if (!force_nonblock) | |
2716 | current->signal->rlim[RLIMIT_FSIZE].rlim_cur = RLIM_INFINITY; | |
2717 | ||
faac996c | 2718 | /* |
bff6035d | 2719 | * Raw bdev writes will return -EOPNOTSUPP for IOCB_NOWAIT. Just |
faac996c JA |
2720 | * retry them without IOCB_NOWAIT. |
2721 | */ | |
2722 | if (ret2 == -EOPNOTSUPP && (kiocb->ki_flags & IOCB_NOWAIT)) | |
2723 | ret2 = -EAGAIN; | |
f67676d1 | 2724 | if (!force_nonblock || ret2 != -EAGAIN) { |
014db007 | 2725 | kiocb_done(kiocb, ret2); |
f67676d1 JA |
2726 | } else { |
2727 | copy_iov: | |
b7bb4f7d | 2728 | ret = io_setup_async_rw(req, io_size, iovec, |
f67676d1 JA |
2729 | inline_vecs, &iter); |
2730 | if (ret) | |
2731 | goto out_free; | |
29de5f6a JA |
2732 | /* any defer here is final, must blocking retry */ |
2733 | req->flags |= REQ_F_MUST_PUNT; | |
f67676d1 JA |
2734 | return -EAGAIN; |
2735 | } | |
2b188cc1 | 2736 | } |
31b51510 | 2737 | out_free: |
99bc4c38 | 2738 | req->flags &= ~REQ_F_NEED_CLEANUP; |
1e95081c | 2739 | kfree(iovec); |
2b188cc1 JA |
2740 | return ret; |
2741 | } | |
2742 | ||
7d67af2c PB |
2743 | static int io_splice_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
2744 | { | |
2745 | struct io_splice* sp = &req->splice; | |
2746 | unsigned int valid_flags = SPLICE_F_FD_IN_FIXED | SPLICE_F_ALL; | |
2747 | int ret; | |
2748 | ||
2749 | if (req->flags & REQ_F_NEED_CLEANUP) | |
2750 | return 0; | |
2751 | ||
2752 | sp->file_in = NULL; | |
2753 | sp->off_in = READ_ONCE(sqe->splice_off_in); | |
2754 | sp->off_out = READ_ONCE(sqe->off); | |
2755 | sp->len = READ_ONCE(sqe->len); | |
2756 | sp->flags = READ_ONCE(sqe->splice_flags); | |
2757 | ||
2758 | if (unlikely(sp->flags & ~valid_flags)) | |
2759 | return -EINVAL; | |
2760 | ||
2761 | ret = io_file_get(NULL, req, READ_ONCE(sqe->splice_fd_in), &sp->file_in, | |
2762 | (sp->flags & SPLICE_F_FD_IN_FIXED)); | |
2763 | if (ret) | |
2764 | return ret; | |
2765 | req->flags |= REQ_F_NEED_CLEANUP; | |
2766 | ||
2767 | if (!S_ISREG(file_inode(sp->file_in)->i_mode)) | |
2768 | req->work.flags |= IO_WQ_WORK_UNBOUND; | |
2769 | ||
2770 | return 0; | |
2771 | } | |
2772 | ||
2773 | static bool io_splice_punt(struct file *file) | |
2774 | { | |
2775 | if (get_pipe_info(file)) | |
2776 | return false; | |
2777 | if (!io_file_supports_async(file)) | |
2778 | return true; | |
2779 | return !(file->f_mode & O_NONBLOCK); | |
2780 | } | |
2781 | ||
014db007 | 2782 | static int io_splice(struct io_kiocb *req, bool force_nonblock) |
7d67af2c PB |
2783 | { |
2784 | struct io_splice *sp = &req->splice; | |
2785 | struct file *in = sp->file_in; | |
2786 | struct file *out = sp->file_out; | |
2787 | unsigned int flags = sp->flags & ~SPLICE_F_FD_IN_FIXED; | |
2788 | loff_t *poff_in, *poff_out; | |
2789 | long ret; | |
2790 | ||
2791 | if (force_nonblock) { | |
2792 | if (io_splice_punt(in) || io_splice_punt(out)) | |
2793 | return -EAGAIN; | |
2794 | flags |= SPLICE_F_NONBLOCK; | |
2795 | } | |
2796 | ||
2797 | poff_in = (sp->off_in == -1) ? NULL : &sp->off_in; | |
2798 | poff_out = (sp->off_out == -1) ? NULL : &sp->off_out; | |
2799 | ret = do_splice(in, poff_in, out, poff_out, sp->len, flags); | |
2800 | if (force_nonblock && ret == -EAGAIN) | |
2801 | return -EAGAIN; | |
2802 | ||
2803 | io_put_file(req, in, (sp->flags & SPLICE_F_FD_IN_FIXED)); | |
2804 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
2805 | ||
2806 | io_cqring_add_event(req, ret); | |
2807 | if (ret != sp->len) | |
2808 | req_set_fail_links(req); | |
014db007 | 2809 | io_put_req(req); |
7d67af2c PB |
2810 | return 0; |
2811 | } | |
2812 | ||
2b188cc1 JA |
2813 | /* |
2814 | * IORING_OP_NOP just posts a completion event, nothing else. | |
2815 | */ | |
78e19bbe | 2816 | static int io_nop(struct io_kiocb *req) |
2b188cc1 JA |
2817 | { |
2818 | struct io_ring_ctx *ctx = req->ctx; | |
2b188cc1 | 2819 | |
def596e9 JA |
2820 | if (unlikely(ctx->flags & IORING_SETUP_IOPOLL)) |
2821 | return -EINVAL; | |
2822 | ||
78e19bbe | 2823 | io_cqring_add_event(req, 0); |
e65ef56d | 2824 | io_put_req(req); |
2b188cc1 JA |
2825 | return 0; |
2826 | } | |
2827 | ||
3529d8c2 | 2828 | static int io_prep_fsync(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
c992fe29 | 2829 | { |
6b06314c | 2830 | struct io_ring_ctx *ctx = req->ctx; |
c992fe29 | 2831 | |
09bb8394 JA |
2832 | if (!req->file) |
2833 | return -EBADF; | |
c992fe29 | 2834 | |
6b06314c | 2835 | if (unlikely(ctx->flags & IORING_SETUP_IOPOLL)) |
def596e9 | 2836 | return -EINVAL; |
edafccee | 2837 | if (unlikely(sqe->addr || sqe->ioprio || sqe->buf_index)) |
c992fe29 CH |
2838 | return -EINVAL; |
2839 | ||
8ed8d3c3 JA |
2840 | req->sync.flags = READ_ONCE(sqe->fsync_flags); |
2841 | if (unlikely(req->sync.flags & ~IORING_FSYNC_DATASYNC)) | |
2842 | return -EINVAL; | |
2843 | ||
2844 | req->sync.off = READ_ONCE(sqe->off); | |
2845 | req->sync.len = READ_ONCE(sqe->len); | |
c992fe29 CH |
2846 | return 0; |
2847 | } | |
2848 | ||
8ed8d3c3 JA |
2849 | static bool io_req_cancelled(struct io_kiocb *req) |
2850 | { | |
2851 | if (req->work.flags & IO_WQ_WORK_CANCEL) { | |
2852 | req_set_fail_links(req); | |
2853 | io_cqring_add_event(req, -ECANCELED); | |
2854 | io_put_req(req); | |
2855 | return true; | |
2856 | } | |
2857 | ||
2858 | return false; | |
2859 | } | |
2860 | ||
014db007 | 2861 | static void __io_fsync(struct io_kiocb *req) |
8ed8d3c3 | 2862 | { |
8ed8d3c3 | 2863 | loff_t end = req->sync.off + req->sync.len; |
8ed8d3c3 JA |
2864 | int ret; |
2865 | ||
9adbd45d | 2866 | ret = vfs_fsync_range(req->file, req->sync.off, |
8ed8d3c3 JA |
2867 | end > 0 ? end : LLONG_MAX, |
2868 | req->sync.flags & IORING_FSYNC_DATASYNC); | |
2869 | if (ret < 0) | |
2870 | req_set_fail_links(req); | |
2871 | io_cqring_add_event(req, ret); | |
014db007 | 2872 | io_put_req(req); |
8ed8d3c3 JA |
2873 | } |
2874 | ||
5ea62161 | 2875 | static void io_fsync_finish(struct io_wq_work **workptr) |
c992fe29 | 2876 | { |
5ea62161 | 2877 | struct io_kiocb *req = container_of(*workptr, struct io_kiocb, work); |
5ea62161 PB |
2878 | |
2879 | if (io_req_cancelled(req)) | |
2880 | return; | |
014db007 | 2881 | __io_fsync(req); |
e9fd9396 | 2882 | io_steal_work(req, workptr); |
8ed8d3c3 | 2883 | } |
c992fe29 | 2884 | |
014db007 | 2885 | static int io_fsync(struct io_kiocb *req, bool force_nonblock) |
c992fe29 | 2886 | { |
c992fe29 | 2887 | /* fsync always requires a blocking context */ |
8ed8d3c3 | 2888 | if (force_nonblock) { |
8ed8d3c3 | 2889 | req->work.func = io_fsync_finish; |
c992fe29 | 2890 | return -EAGAIN; |
8ed8d3c3 | 2891 | } |
014db007 | 2892 | __io_fsync(req); |
c992fe29 CH |
2893 | return 0; |
2894 | } | |
2895 | ||
014db007 | 2896 | static void __io_fallocate(struct io_kiocb *req) |
8ed8d3c3 | 2897 | { |
8ed8d3c3 JA |
2898 | int ret; |
2899 | ||
4ed734b0 | 2900 | current->signal->rlim[RLIMIT_FSIZE].rlim_cur = req->fsize; |
d63d1b5e JA |
2901 | ret = vfs_fallocate(req->file, req->sync.mode, req->sync.off, |
2902 | req->sync.len); | |
4ed734b0 | 2903 | current->signal->rlim[RLIMIT_FSIZE].rlim_cur = RLIM_INFINITY; |
8ed8d3c3 JA |
2904 | if (ret < 0) |
2905 | req_set_fail_links(req); | |
2906 | io_cqring_add_event(req, ret); | |
014db007 | 2907 | io_put_req(req); |
5ea62161 PB |
2908 | } |
2909 | ||
2910 | static void io_fallocate_finish(struct io_wq_work **workptr) | |
2911 | { | |
2912 | struct io_kiocb *req = container_of(*workptr, struct io_kiocb, work); | |
5ea62161 | 2913 | |
594506fe PB |
2914 | if (io_req_cancelled(req)) |
2915 | return; | |
014db007 | 2916 | __io_fallocate(req); |
e9fd9396 | 2917 | io_steal_work(req, workptr); |
5d17b4a4 JA |
2918 | } |
2919 | ||
d63d1b5e JA |
2920 | static int io_fallocate_prep(struct io_kiocb *req, |
2921 | const struct io_uring_sqe *sqe) | |
2922 | { | |
2923 | if (sqe->ioprio || sqe->buf_index || sqe->rw_flags) | |
2924 | return -EINVAL; | |
2925 | ||
2926 | req->sync.off = READ_ONCE(sqe->off); | |
2927 | req->sync.len = READ_ONCE(sqe->addr); | |
2928 | req->sync.mode = READ_ONCE(sqe->len); | |
4ed734b0 | 2929 | req->fsize = rlimit(RLIMIT_FSIZE); |
d63d1b5e JA |
2930 | return 0; |
2931 | } | |
2932 | ||
014db007 | 2933 | static int io_fallocate(struct io_kiocb *req, bool force_nonblock) |
5d17b4a4 | 2934 | { |
d63d1b5e | 2935 | /* fallocate always requiring blocking context */ |
8ed8d3c3 | 2936 | if (force_nonblock) { |
d63d1b5e | 2937 | req->work.func = io_fallocate_finish; |
5d17b4a4 | 2938 | return -EAGAIN; |
8ed8d3c3 | 2939 | } |
5d17b4a4 | 2940 | |
014db007 | 2941 | __io_fallocate(req); |
5d17b4a4 JA |
2942 | return 0; |
2943 | } | |
2944 | ||
15b71abe | 2945 | static int io_openat_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
b7bb4f7d | 2946 | { |
f8748881 | 2947 | const char __user *fname; |
15b71abe | 2948 | int ret; |
b7bb4f7d | 2949 | |
15b71abe JA |
2950 | if (sqe->ioprio || sqe->buf_index) |
2951 | return -EINVAL; | |
cf3040ca JA |
2952 | if (sqe->flags & IOSQE_FIXED_FILE) |
2953 | return -EBADF; | |
0bdbdd08 PB |
2954 | if (req->flags & REQ_F_NEED_CLEANUP) |
2955 | return 0; | |
03b1230c | 2956 | |
15b71abe | 2957 | req->open.dfd = READ_ONCE(sqe->fd); |
c12cedf2 | 2958 | req->open.how.mode = READ_ONCE(sqe->len); |
f8748881 | 2959 | fname = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
c12cedf2 | 2960 | req->open.how.flags = READ_ONCE(sqe->open_flags); |
08a1d26e JA |
2961 | if (force_o_largefile()) |
2962 | req->open.how.flags |= O_LARGEFILE; | |
3529d8c2 | 2963 | |
f8748881 | 2964 | req->open.filename = getname(fname); |
15b71abe JA |
2965 | if (IS_ERR(req->open.filename)) { |
2966 | ret = PTR_ERR(req->open.filename); | |
2967 | req->open.filename = NULL; | |
2968 | return ret; | |
2969 | } | |
3529d8c2 | 2970 | |
4022e7af | 2971 | req->open.nofile = rlimit(RLIMIT_NOFILE); |
8fef80bf | 2972 | req->flags |= REQ_F_NEED_CLEANUP; |
15b71abe | 2973 | return 0; |
03b1230c JA |
2974 | } |
2975 | ||
cebdb986 | 2976 | static int io_openat2_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
aa1fa28f | 2977 | { |
cebdb986 JA |
2978 | struct open_how __user *how; |
2979 | const char __user *fname; | |
2980 | size_t len; | |
0fa03c62 JA |
2981 | int ret; |
2982 | ||
cebdb986 | 2983 | if (sqe->ioprio || sqe->buf_index) |
0fa03c62 | 2984 | return -EINVAL; |
cf3040ca JA |
2985 | if (sqe->flags & IOSQE_FIXED_FILE) |
2986 | return -EBADF; | |
0bdbdd08 PB |
2987 | if (req->flags & REQ_F_NEED_CLEANUP) |
2988 | return 0; | |
0fa03c62 | 2989 | |
cebdb986 JA |
2990 | req->open.dfd = READ_ONCE(sqe->fd); |
2991 | fname = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
2992 | how = u64_to_user_ptr(READ_ONCE(sqe->addr2)); | |
2993 | len = READ_ONCE(sqe->len); | |
0fa03c62 | 2994 | |
cebdb986 JA |
2995 | if (len < OPEN_HOW_SIZE_VER0) |
2996 | return -EINVAL; | |
3529d8c2 | 2997 | |
cebdb986 JA |
2998 | ret = copy_struct_from_user(&req->open.how, sizeof(req->open.how), how, |
2999 | len); | |
3000 | if (ret) | |
3001 | return ret; | |
3529d8c2 | 3002 | |
cebdb986 JA |
3003 | if (!(req->open.how.flags & O_PATH) && force_o_largefile()) |
3004 | req->open.how.flags |= O_LARGEFILE; | |
0fa03c62 | 3005 | |
cebdb986 JA |
3006 | req->open.filename = getname(fname); |
3007 | if (IS_ERR(req->open.filename)) { | |
3008 | ret = PTR_ERR(req->open.filename); | |
3009 | req->open.filename = NULL; | |
3010 | return ret; | |
3011 | } | |
3012 | ||
4022e7af | 3013 | req->open.nofile = rlimit(RLIMIT_NOFILE); |
8fef80bf | 3014 | req->flags |= REQ_F_NEED_CLEANUP; |
cebdb986 JA |
3015 | return 0; |
3016 | } | |
3017 | ||
014db007 | 3018 | static int io_openat2(struct io_kiocb *req, bool force_nonblock) |
15b71abe JA |
3019 | { |
3020 | struct open_flags op; | |
15b71abe JA |
3021 | struct file *file; |
3022 | int ret; | |
3023 | ||
f86cd20c | 3024 | if (force_nonblock) |
15b71abe | 3025 | return -EAGAIN; |
15b71abe | 3026 | |
cebdb986 | 3027 | ret = build_open_flags(&req->open.how, &op); |
15b71abe JA |
3028 | if (ret) |
3029 | goto err; | |
3030 | ||
4022e7af | 3031 | ret = __get_unused_fd_flags(req->open.how.flags, req->open.nofile); |
15b71abe JA |
3032 | if (ret < 0) |
3033 | goto err; | |
3034 | ||
3035 | file = do_filp_open(req->open.dfd, req->open.filename, &op); | |
3036 | if (IS_ERR(file)) { | |
3037 | put_unused_fd(ret); | |
3038 | ret = PTR_ERR(file); | |
3039 | } else { | |
3040 | fsnotify_open(file); | |
3041 | fd_install(ret, file); | |
3042 | } | |
3043 | err: | |
3044 | putname(req->open.filename); | |
8fef80bf | 3045 | req->flags &= ~REQ_F_NEED_CLEANUP; |
15b71abe JA |
3046 | if (ret < 0) |
3047 | req_set_fail_links(req); | |
3048 | io_cqring_add_event(req, ret); | |
014db007 | 3049 | io_put_req(req); |
15b71abe JA |
3050 | return 0; |
3051 | } | |
3052 | ||
014db007 | 3053 | static int io_openat(struct io_kiocb *req, bool force_nonblock) |
cebdb986 JA |
3054 | { |
3055 | req->open.how = build_open_how(req->open.how.flags, req->open.how.mode); | |
014db007 | 3056 | return io_openat2(req, force_nonblock); |
cebdb986 JA |
3057 | } |
3058 | ||
067524e9 JA |
3059 | static int io_remove_buffers_prep(struct io_kiocb *req, |
3060 | const struct io_uring_sqe *sqe) | |
3061 | { | |
3062 | struct io_provide_buf *p = &req->pbuf; | |
3063 | u64 tmp; | |
3064 | ||
3065 | if (sqe->ioprio || sqe->rw_flags || sqe->addr || sqe->len || sqe->off) | |
3066 | return -EINVAL; | |
3067 | ||
3068 | tmp = READ_ONCE(sqe->fd); | |
3069 | if (!tmp || tmp > USHRT_MAX) | |
3070 | return -EINVAL; | |
3071 | ||
3072 | memset(p, 0, sizeof(*p)); | |
3073 | p->nbufs = tmp; | |
3074 | p->bgid = READ_ONCE(sqe->buf_group); | |
3075 | return 0; | |
3076 | } | |
3077 | ||
3078 | static int __io_remove_buffers(struct io_ring_ctx *ctx, struct io_buffer *buf, | |
3079 | int bgid, unsigned nbufs) | |
3080 | { | |
3081 | unsigned i = 0; | |
3082 | ||
3083 | /* shouldn't happen */ | |
3084 | if (!nbufs) | |
3085 | return 0; | |
3086 | ||
3087 | /* the head kbuf is the list itself */ | |
3088 | while (!list_empty(&buf->list)) { | |
3089 | struct io_buffer *nxt; | |
3090 | ||
3091 | nxt = list_first_entry(&buf->list, struct io_buffer, list); | |
3092 | list_del(&nxt->list); | |
3093 | kfree(nxt); | |
3094 | if (++i == nbufs) | |
3095 | return i; | |
3096 | } | |
3097 | i++; | |
3098 | kfree(buf); | |
3099 | idr_remove(&ctx->io_buffer_idr, bgid); | |
3100 | ||
3101 | return i; | |
3102 | } | |
3103 | ||
3104 | static int io_remove_buffers(struct io_kiocb *req, bool force_nonblock) | |
3105 | { | |
3106 | struct io_provide_buf *p = &req->pbuf; | |
3107 | struct io_ring_ctx *ctx = req->ctx; | |
3108 | struct io_buffer *head; | |
3109 | int ret = 0; | |
3110 | ||
3111 | io_ring_submit_lock(ctx, !force_nonblock); | |
3112 | ||
3113 | lockdep_assert_held(&ctx->uring_lock); | |
3114 | ||
3115 | ret = -ENOENT; | |
3116 | head = idr_find(&ctx->io_buffer_idr, p->bgid); | |
3117 | if (head) | |
3118 | ret = __io_remove_buffers(ctx, head, p->bgid, p->nbufs); | |
3119 | ||
3120 | io_ring_submit_lock(ctx, !force_nonblock); | |
3121 | if (ret < 0) | |
3122 | req_set_fail_links(req); | |
3123 | io_cqring_add_event(req, ret); | |
3124 | io_put_req(req); | |
3125 | return 0; | |
3126 | } | |
3127 | ||
ddf0322d JA |
3128 | static int io_provide_buffers_prep(struct io_kiocb *req, |
3129 | const struct io_uring_sqe *sqe) | |
3130 | { | |
3131 | struct io_provide_buf *p = &req->pbuf; | |
3132 | u64 tmp; | |
3133 | ||
3134 | if (sqe->ioprio || sqe->rw_flags) | |
3135 | return -EINVAL; | |
3136 | ||
3137 | tmp = READ_ONCE(sqe->fd); | |
3138 | if (!tmp || tmp > USHRT_MAX) | |
3139 | return -E2BIG; | |
3140 | p->nbufs = tmp; | |
3141 | p->addr = READ_ONCE(sqe->addr); | |
3142 | p->len = READ_ONCE(sqe->len); | |
3143 | ||
3144 | if (!access_ok(u64_to_user_ptr(p->addr), p->len)) | |
3145 | return -EFAULT; | |
3146 | ||
3147 | p->bgid = READ_ONCE(sqe->buf_group); | |
3148 | tmp = READ_ONCE(sqe->off); | |
3149 | if (tmp > USHRT_MAX) | |
3150 | return -E2BIG; | |
3151 | p->bid = tmp; | |
3152 | return 0; | |
3153 | } | |
3154 | ||
3155 | static int io_add_buffers(struct io_provide_buf *pbuf, struct io_buffer **head) | |
3156 | { | |
3157 | struct io_buffer *buf; | |
3158 | u64 addr = pbuf->addr; | |
3159 | int i, bid = pbuf->bid; | |
3160 | ||
3161 | for (i = 0; i < pbuf->nbufs; i++) { | |
3162 | buf = kmalloc(sizeof(*buf), GFP_KERNEL); | |
3163 | if (!buf) | |
3164 | break; | |
3165 | ||
3166 | buf->addr = addr; | |
3167 | buf->len = pbuf->len; | |
3168 | buf->bid = bid; | |
3169 | addr += pbuf->len; | |
3170 | bid++; | |
3171 | if (!*head) { | |
3172 | INIT_LIST_HEAD(&buf->list); | |
3173 | *head = buf; | |
3174 | } else { | |
3175 | list_add_tail(&buf->list, &(*head)->list); | |
3176 | } | |
3177 | } | |
3178 | ||
3179 | return i ? i : -ENOMEM; | |
3180 | } | |
3181 | ||
ddf0322d JA |
3182 | static int io_provide_buffers(struct io_kiocb *req, bool force_nonblock) |
3183 | { | |
3184 | struct io_provide_buf *p = &req->pbuf; | |
3185 | struct io_ring_ctx *ctx = req->ctx; | |
3186 | struct io_buffer *head, *list; | |
3187 | int ret = 0; | |
3188 | ||
3189 | io_ring_submit_lock(ctx, !force_nonblock); | |
3190 | ||
3191 | lockdep_assert_held(&ctx->uring_lock); | |
3192 | ||
3193 | list = head = idr_find(&ctx->io_buffer_idr, p->bgid); | |
3194 | ||
3195 | ret = io_add_buffers(p, &head); | |
3196 | if (ret < 0) | |
3197 | goto out; | |
3198 | ||
3199 | if (!list) { | |
3200 | ret = idr_alloc(&ctx->io_buffer_idr, head, p->bgid, p->bgid + 1, | |
3201 | GFP_KERNEL); | |
3202 | if (ret < 0) { | |
067524e9 | 3203 | __io_remove_buffers(ctx, head, p->bgid, -1U); |
ddf0322d JA |
3204 | goto out; |
3205 | } | |
3206 | } | |
3207 | out: | |
3208 | io_ring_submit_unlock(ctx, !force_nonblock); | |
3209 | if (ret < 0) | |
3210 | req_set_fail_links(req); | |
3211 | io_cqring_add_event(req, ret); | |
3212 | io_put_req(req); | |
3213 | return 0; | |
cebdb986 JA |
3214 | } |
3215 | ||
3e4827b0 JA |
3216 | static int io_epoll_ctl_prep(struct io_kiocb *req, |
3217 | const struct io_uring_sqe *sqe) | |
3218 | { | |
3219 | #if defined(CONFIG_EPOLL) | |
3220 | if (sqe->ioprio || sqe->buf_index) | |
3221 | return -EINVAL; | |
3222 | ||
3223 | req->epoll.epfd = READ_ONCE(sqe->fd); | |
3224 | req->epoll.op = READ_ONCE(sqe->len); | |
3225 | req->epoll.fd = READ_ONCE(sqe->off); | |
3226 | ||
3227 | if (ep_op_has_event(req->epoll.op)) { | |
3228 | struct epoll_event __user *ev; | |
3229 | ||
3230 | ev = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
3231 | if (copy_from_user(&req->epoll.event, ev, sizeof(*ev))) | |
3232 | return -EFAULT; | |
3233 | } | |
3234 | ||
3235 | return 0; | |
3236 | #else | |
3237 | return -EOPNOTSUPP; | |
3238 | #endif | |
3239 | } | |
3240 | ||
014db007 | 3241 | static int io_epoll_ctl(struct io_kiocb *req, bool force_nonblock) |
3e4827b0 JA |
3242 | { |
3243 | #if defined(CONFIG_EPOLL) | |
3244 | struct io_epoll *ie = &req->epoll; | |
3245 | int ret; | |
3246 | ||
3247 | ret = do_epoll_ctl(ie->epfd, ie->op, ie->fd, &ie->event, force_nonblock); | |
3248 | if (force_nonblock && ret == -EAGAIN) | |
3249 | return -EAGAIN; | |
3250 | ||
3251 | if (ret < 0) | |
3252 | req_set_fail_links(req); | |
3253 | io_cqring_add_event(req, ret); | |
014db007 | 3254 | io_put_req(req); |
3e4827b0 JA |
3255 | return 0; |
3256 | #else | |
3257 | return -EOPNOTSUPP; | |
3258 | #endif | |
3259 | } | |
3260 | ||
c1ca757b JA |
3261 | static int io_madvise_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
3262 | { | |
3263 | #if defined(CONFIG_ADVISE_SYSCALLS) && defined(CONFIG_MMU) | |
3264 | if (sqe->ioprio || sqe->buf_index || sqe->off) | |
3265 | return -EINVAL; | |
3266 | ||
3267 | req->madvise.addr = READ_ONCE(sqe->addr); | |
3268 | req->madvise.len = READ_ONCE(sqe->len); | |
3269 | req->madvise.advice = READ_ONCE(sqe->fadvise_advice); | |
3270 | return 0; | |
3271 | #else | |
3272 | return -EOPNOTSUPP; | |
3273 | #endif | |
3274 | } | |
3275 | ||
014db007 | 3276 | static int io_madvise(struct io_kiocb *req, bool force_nonblock) |
c1ca757b JA |
3277 | { |
3278 | #if defined(CONFIG_ADVISE_SYSCALLS) && defined(CONFIG_MMU) | |
3279 | struct io_madvise *ma = &req->madvise; | |
3280 | int ret; | |
3281 | ||
3282 | if (force_nonblock) | |
3283 | return -EAGAIN; | |
3284 | ||
3285 | ret = do_madvise(ma->addr, ma->len, ma->advice); | |
3286 | if (ret < 0) | |
3287 | req_set_fail_links(req); | |
3288 | io_cqring_add_event(req, ret); | |
014db007 | 3289 | io_put_req(req); |
c1ca757b JA |
3290 | return 0; |
3291 | #else | |
3292 | return -EOPNOTSUPP; | |
3293 | #endif | |
3294 | } | |
3295 | ||
4840e418 JA |
3296 | static int io_fadvise_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
3297 | { | |
3298 | if (sqe->ioprio || sqe->buf_index || sqe->addr) | |
3299 | return -EINVAL; | |
3300 | ||
3301 | req->fadvise.offset = READ_ONCE(sqe->off); | |
3302 | req->fadvise.len = READ_ONCE(sqe->len); | |
3303 | req->fadvise.advice = READ_ONCE(sqe->fadvise_advice); | |
3304 | return 0; | |
3305 | } | |
3306 | ||
014db007 | 3307 | static int io_fadvise(struct io_kiocb *req, bool force_nonblock) |
4840e418 JA |
3308 | { |
3309 | struct io_fadvise *fa = &req->fadvise; | |
3310 | int ret; | |
3311 | ||
3e69426d JA |
3312 | if (force_nonblock) { |
3313 | switch (fa->advice) { | |
3314 | case POSIX_FADV_NORMAL: | |
3315 | case POSIX_FADV_RANDOM: | |
3316 | case POSIX_FADV_SEQUENTIAL: | |
3317 | break; | |
3318 | default: | |
3319 | return -EAGAIN; | |
3320 | } | |
3321 | } | |
4840e418 JA |
3322 | |
3323 | ret = vfs_fadvise(req->file, fa->offset, fa->len, fa->advice); | |
3324 | if (ret < 0) | |
3325 | req_set_fail_links(req); | |
3326 | io_cqring_add_event(req, ret); | |
014db007 | 3327 | io_put_req(req); |
4840e418 JA |
3328 | return 0; |
3329 | } | |
3330 | ||
eddc7ef5 JA |
3331 | static int io_statx_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
3332 | { | |
f8748881 | 3333 | const char __user *fname; |
eddc7ef5 JA |
3334 | unsigned lookup_flags; |
3335 | int ret; | |
3336 | ||
3337 | if (sqe->ioprio || sqe->buf_index) | |
3338 | return -EINVAL; | |
cf3040ca JA |
3339 | if (sqe->flags & IOSQE_FIXED_FILE) |
3340 | return -EBADF; | |
0bdbdd08 PB |
3341 | if (req->flags & REQ_F_NEED_CLEANUP) |
3342 | return 0; | |
eddc7ef5 JA |
3343 | |
3344 | req->open.dfd = READ_ONCE(sqe->fd); | |
3345 | req->open.mask = READ_ONCE(sqe->len); | |
f8748881 | 3346 | fname = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
eddc7ef5 | 3347 | req->open.buffer = u64_to_user_ptr(READ_ONCE(sqe->addr2)); |
c12cedf2 | 3348 | req->open.how.flags = READ_ONCE(sqe->statx_flags); |
eddc7ef5 | 3349 | |
c12cedf2 | 3350 | if (vfs_stat_set_lookup_flags(&lookup_flags, req->open.how.flags)) |
eddc7ef5 JA |
3351 | return -EINVAL; |
3352 | ||
f8748881 | 3353 | req->open.filename = getname_flags(fname, lookup_flags, NULL); |
eddc7ef5 JA |
3354 | if (IS_ERR(req->open.filename)) { |
3355 | ret = PTR_ERR(req->open.filename); | |
3356 | req->open.filename = NULL; | |
3357 | return ret; | |
3358 | } | |
3359 | ||
8fef80bf | 3360 | req->flags |= REQ_F_NEED_CLEANUP; |
eddc7ef5 JA |
3361 | return 0; |
3362 | } | |
3363 | ||
014db007 | 3364 | static int io_statx(struct io_kiocb *req, bool force_nonblock) |
eddc7ef5 JA |
3365 | { |
3366 | struct io_open *ctx = &req->open; | |
3367 | unsigned lookup_flags; | |
3368 | struct path path; | |
3369 | struct kstat stat; | |
3370 | int ret; | |
3371 | ||
3372 | if (force_nonblock) | |
3373 | return -EAGAIN; | |
3374 | ||
c12cedf2 | 3375 | if (vfs_stat_set_lookup_flags(&lookup_flags, ctx->how.flags)) |
eddc7ef5 JA |
3376 | return -EINVAL; |
3377 | ||
3378 | retry: | |
3379 | /* filename_lookup() drops it, keep a reference */ | |
3380 | ctx->filename->refcnt++; | |
3381 | ||
3382 | ret = filename_lookup(ctx->dfd, ctx->filename, lookup_flags, &path, | |
3383 | NULL); | |
3384 | if (ret) | |
3385 | goto err; | |
3386 | ||
c12cedf2 | 3387 | ret = vfs_getattr(&path, &stat, ctx->mask, ctx->how.flags); |
eddc7ef5 JA |
3388 | path_put(&path); |
3389 | if (retry_estale(ret, lookup_flags)) { | |
3390 | lookup_flags |= LOOKUP_REVAL; | |
3391 | goto retry; | |
3392 | } | |
3393 | if (!ret) | |
3394 | ret = cp_statx(&stat, ctx->buffer); | |
3395 | err: | |
3396 | putname(ctx->filename); | |
8fef80bf | 3397 | req->flags &= ~REQ_F_NEED_CLEANUP; |
eddc7ef5 JA |
3398 | if (ret < 0) |
3399 | req_set_fail_links(req); | |
3400 | io_cqring_add_event(req, ret); | |
014db007 | 3401 | io_put_req(req); |
eddc7ef5 JA |
3402 | return 0; |
3403 | } | |
3404 | ||
b5dba59e JA |
3405 | static int io_close_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
3406 | { | |
3407 | /* | |
3408 | * If we queue this for async, it must not be cancellable. That would | |
3409 | * leave the 'file' in an undeterminate state. | |
3410 | */ | |
3411 | req->work.flags |= IO_WQ_WORK_NO_CANCEL; | |
3412 | ||
3413 | if (sqe->ioprio || sqe->off || sqe->addr || sqe->len || | |
3414 | sqe->rw_flags || sqe->buf_index) | |
3415 | return -EINVAL; | |
3416 | if (sqe->flags & IOSQE_FIXED_FILE) | |
cf3040ca | 3417 | return -EBADF; |
b5dba59e JA |
3418 | |
3419 | req->close.fd = READ_ONCE(sqe->fd); | |
3420 | if (req->file->f_op == &io_uring_fops || | |
b14cca0c | 3421 | req->close.fd == req->ctx->ring_fd) |
b5dba59e JA |
3422 | return -EBADF; |
3423 | ||
3424 | return 0; | |
3425 | } | |
3426 | ||
a93b3331 | 3427 | /* only called when __close_fd_get_file() is done */ |
014db007 | 3428 | static void __io_close_finish(struct io_kiocb *req) |
a93b3331 PB |
3429 | { |
3430 | int ret; | |
3431 | ||
3432 | ret = filp_close(req->close.put_file, req->work.files); | |
3433 | if (ret < 0) | |
3434 | req_set_fail_links(req); | |
3435 | io_cqring_add_event(req, ret); | |
3436 | fput(req->close.put_file); | |
014db007 | 3437 | io_put_req(req); |
a93b3331 PB |
3438 | } |
3439 | ||
b5dba59e JA |
3440 | static void io_close_finish(struct io_wq_work **workptr) |
3441 | { | |
3442 | struct io_kiocb *req = container_of(*workptr, struct io_kiocb, work); | |
b5dba59e | 3443 | |
7fbeb95d | 3444 | /* not cancellable, don't do io_req_cancelled() */ |
014db007 | 3445 | __io_close_finish(req); |
e9fd9396 | 3446 | io_steal_work(req, workptr); |
b5dba59e JA |
3447 | } |
3448 | ||
014db007 | 3449 | static int io_close(struct io_kiocb *req, bool force_nonblock) |
b5dba59e JA |
3450 | { |
3451 | int ret; | |
3452 | ||
3453 | req->close.put_file = NULL; | |
3454 | ret = __close_fd_get_file(req->close.fd, &req->close.put_file); | |
3455 | if (ret < 0) | |
3456 | return ret; | |
3457 | ||
3458 | /* if the file has a flush method, be safe and punt to async */ | |
a2100672 | 3459 | if (req->close.put_file->f_op->flush && force_nonblock) { |
594506fe PB |
3460 | /* submission ref will be dropped, take it for async */ |
3461 | refcount_inc(&req->refs); | |
3462 | ||
a2100672 PB |
3463 | req->work.func = io_close_finish; |
3464 | /* | |
3465 | * Do manual async queue here to avoid grabbing files - we don't | |
3466 | * need the files, and it'll cause io_close_finish() to close | |
3467 | * the file again and cause a double CQE entry for this request | |
3468 | */ | |
3469 | io_queue_async_work(req); | |
3470 | return 0; | |
3471 | } | |
b5dba59e JA |
3472 | |
3473 | /* | |
3474 | * No ->flush(), safely close from here and just punt the | |
3475 | * fput() to async context. | |
3476 | */ | |
014db007 | 3477 | __io_close_finish(req); |
1a417f4e | 3478 | return 0; |
b5dba59e JA |
3479 | } |
3480 | ||
3529d8c2 | 3481 | static int io_prep_sfr(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
5d17b4a4 JA |
3482 | { |
3483 | struct io_ring_ctx *ctx = req->ctx; | |
5d17b4a4 JA |
3484 | |
3485 | if (!req->file) | |
3486 | return -EBADF; | |
5d17b4a4 JA |
3487 | |
3488 | if (unlikely(ctx->flags & IORING_SETUP_IOPOLL)) | |
3489 | return -EINVAL; | |
3490 | if (unlikely(sqe->addr || sqe->ioprio || sqe->buf_index)) | |
3491 | return -EINVAL; | |
3492 | ||
8ed8d3c3 JA |
3493 | req->sync.off = READ_ONCE(sqe->off); |
3494 | req->sync.len = READ_ONCE(sqe->len); | |
3495 | req->sync.flags = READ_ONCE(sqe->sync_range_flags); | |
8ed8d3c3 JA |
3496 | return 0; |
3497 | } | |
3498 | ||
014db007 | 3499 | static void __io_sync_file_range(struct io_kiocb *req) |
8ed8d3c3 | 3500 | { |
8ed8d3c3 JA |
3501 | int ret; |
3502 | ||
9adbd45d | 3503 | ret = sync_file_range(req->file, req->sync.off, req->sync.len, |
8ed8d3c3 JA |
3504 | req->sync.flags); |
3505 | if (ret < 0) | |
3506 | req_set_fail_links(req); | |
3507 | io_cqring_add_event(req, ret); | |
014db007 | 3508 | io_put_req(req); |
5ea62161 PB |
3509 | } |
3510 | ||
3511 | ||
3512 | static void io_sync_file_range_finish(struct io_wq_work **workptr) | |
3513 | { | |
3514 | struct io_kiocb *req = container_of(*workptr, struct io_kiocb, work); | |
5ea62161 PB |
3515 | |
3516 | if (io_req_cancelled(req)) | |
3517 | return; | |
014db007 | 3518 | __io_sync_file_range(req); |
594506fe | 3519 | io_put_req(req); /* put submission ref */ |
5d17b4a4 JA |
3520 | } |
3521 | ||
014db007 | 3522 | static int io_sync_file_range(struct io_kiocb *req, bool force_nonblock) |
5d17b4a4 | 3523 | { |
5d17b4a4 | 3524 | /* sync_file_range always requires a blocking context */ |
8ed8d3c3 | 3525 | if (force_nonblock) { |
8ed8d3c3 | 3526 | req->work.func = io_sync_file_range_finish; |
5d17b4a4 | 3527 | return -EAGAIN; |
8ed8d3c3 | 3528 | } |
5d17b4a4 | 3529 | |
014db007 | 3530 | __io_sync_file_range(req); |
5d17b4a4 JA |
3531 | return 0; |
3532 | } | |
3533 | ||
469956e8 | 3534 | #if defined(CONFIG_NET) |
02d27d89 PB |
3535 | static int io_setup_async_msg(struct io_kiocb *req, |
3536 | struct io_async_msghdr *kmsg) | |
3537 | { | |
3538 | if (req->io) | |
3539 | return -EAGAIN; | |
3540 | if (io_alloc_async_ctx(req)) { | |
3541 | if (kmsg->iov != kmsg->fast_iov) | |
3542 | kfree(kmsg->iov); | |
3543 | return -ENOMEM; | |
3544 | } | |
3545 | req->flags |= REQ_F_NEED_CLEANUP; | |
3546 | memcpy(&req->io->msg, kmsg, sizeof(*kmsg)); | |
3547 | return -EAGAIN; | |
3548 | } | |
3549 | ||
3529d8c2 | 3550 | static int io_sendmsg_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
03b1230c | 3551 | { |
e47293fd | 3552 | struct io_sr_msg *sr = &req->sr_msg; |
3529d8c2 | 3553 | struct io_async_ctx *io = req->io; |
99bc4c38 | 3554 | int ret; |
03b1230c | 3555 | |
e47293fd JA |
3556 | sr->msg_flags = READ_ONCE(sqe->msg_flags); |
3557 | sr->msg = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
fddaface | 3558 | sr->len = READ_ONCE(sqe->len); |
3529d8c2 | 3559 | |
d8768362 JA |
3560 | #ifdef CONFIG_COMPAT |
3561 | if (req->ctx->compat) | |
3562 | sr->msg_flags |= MSG_CMSG_COMPAT; | |
3563 | #endif | |
3564 | ||
fddaface | 3565 | if (!io || req->opcode == IORING_OP_SEND) |
3529d8c2 | 3566 | return 0; |
5f798bea PB |
3567 | /* iovec is already imported */ |
3568 | if (req->flags & REQ_F_NEED_CLEANUP) | |
3569 | return 0; | |
3529d8c2 | 3570 | |
d9688565 | 3571 | io->msg.iov = io->msg.fast_iov; |
99bc4c38 | 3572 | ret = sendmsg_copy_msghdr(&io->msg.msg, sr->msg, sr->msg_flags, |
e47293fd | 3573 | &io->msg.iov); |
99bc4c38 PB |
3574 | if (!ret) |
3575 | req->flags |= REQ_F_NEED_CLEANUP; | |
3576 | return ret; | |
03b1230c JA |
3577 | } |
3578 | ||
014db007 | 3579 | static int io_sendmsg(struct io_kiocb *req, bool force_nonblock) |
aa1fa28f | 3580 | { |
0b416c3e | 3581 | struct io_async_msghdr *kmsg = NULL; |
0fa03c62 JA |
3582 | struct socket *sock; |
3583 | int ret; | |
3584 | ||
3585 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) | |
3586 | return -EINVAL; | |
3587 | ||
3588 | sock = sock_from_file(req->file, &ret); | |
3589 | if (sock) { | |
b7bb4f7d | 3590 | struct io_async_ctx io; |
0fa03c62 JA |
3591 | unsigned flags; |
3592 | ||
03b1230c | 3593 | if (req->io) { |
0b416c3e | 3594 | kmsg = &req->io->msg; |
b537916c | 3595 | kmsg->msg.msg_name = &req->io->msg.addr; |
0b416c3e JA |
3596 | /* if iov is set, it's allocated already */ |
3597 | if (!kmsg->iov) | |
3598 | kmsg->iov = kmsg->fast_iov; | |
3599 | kmsg->msg.msg_iter.iov = kmsg->iov; | |
03b1230c | 3600 | } else { |
3529d8c2 JA |
3601 | struct io_sr_msg *sr = &req->sr_msg; |
3602 | ||
0b416c3e | 3603 | kmsg = &io.msg; |
b537916c | 3604 | kmsg->msg.msg_name = &io.msg.addr; |
3529d8c2 JA |
3605 | |
3606 | io.msg.iov = io.msg.fast_iov; | |
3607 | ret = sendmsg_copy_msghdr(&io.msg.msg, sr->msg, | |
3608 | sr->msg_flags, &io.msg.iov); | |
03b1230c | 3609 | if (ret) |
3529d8c2 | 3610 | return ret; |
03b1230c | 3611 | } |
0fa03c62 | 3612 | |
e47293fd JA |
3613 | flags = req->sr_msg.msg_flags; |
3614 | if (flags & MSG_DONTWAIT) | |
3615 | req->flags |= REQ_F_NOWAIT; | |
3616 | else if (force_nonblock) | |
3617 | flags |= MSG_DONTWAIT; | |
3618 | ||
0b416c3e | 3619 | ret = __sys_sendmsg_sock(sock, &kmsg->msg, flags); |
02d27d89 PB |
3620 | if (force_nonblock && ret == -EAGAIN) |
3621 | return io_setup_async_msg(req, kmsg); | |
441cdbd5 JA |
3622 | if (ret == -ERESTARTSYS) |
3623 | ret = -EINTR; | |
0fa03c62 JA |
3624 | } |
3625 | ||
1e95081c | 3626 | if (kmsg && kmsg->iov != kmsg->fast_iov) |
0b416c3e | 3627 | kfree(kmsg->iov); |
99bc4c38 | 3628 | req->flags &= ~REQ_F_NEED_CLEANUP; |
78e19bbe | 3629 | io_cqring_add_event(req, ret); |
4e88d6e7 JA |
3630 | if (ret < 0) |
3631 | req_set_fail_links(req); | |
014db007 | 3632 | io_put_req(req); |
5d17b4a4 | 3633 | return 0; |
03b1230c | 3634 | } |
aa1fa28f | 3635 | |
014db007 | 3636 | static int io_send(struct io_kiocb *req, bool force_nonblock) |
fddaface | 3637 | { |
fddaface JA |
3638 | struct socket *sock; |
3639 | int ret; | |
3640 | ||
3641 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) | |
3642 | return -EINVAL; | |
3643 | ||
3644 | sock = sock_from_file(req->file, &ret); | |
3645 | if (sock) { | |
3646 | struct io_sr_msg *sr = &req->sr_msg; | |
3647 | struct msghdr msg; | |
3648 | struct iovec iov; | |
3649 | unsigned flags; | |
3650 | ||
3651 | ret = import_single_range(WRITE, sr->buf, sr->len, &iov, | |
3652 | &msg.msg_iter); | |
3653 | if (ret) | |
3654 | return ret; | |
3655 | ||
3656 | msg.msg_name = NULL; | |
3657 | msg.msg_control = NULL; | |
3658 | msg.msg_controllen = 0; | |
3659 | msg.msg_namelen = 0; | |
3660 | ||
3661 | flags = req->sr_msg.msg_flags; | |
3662 | if (flags & MSG_DONTWAIT) | |
3663 | req->flags |= REQ_F_NOWAIT; | |
3664 | else if (force_nonblock) | |
3665 | flags |= MSG_DONTWAIT; | |
3666 | ||
0b7b21e4 JA |
3667 | msg.msg_flags = flags; |
3668 | ret = sock_sendmsg(sock, &msg); | |
fddaface JA |
3669 | if (force_nonblock && ret == -EAGAIN) |
3670 | return -EAGAIN; | |
3671 | if (ret == -ERESTARTSYS) | |
3672 | ret = -EINTR; | |
3673 | } | |
3674 | ||
3675 | io_cqring_add_event(req, ret); | |
3676 | if (ret < 0) | |
3677 | req_set_fail_links(req); | |
014db007 | 3678 | io_put_req(req); |
fddaface | 3679 | return 0; |
fddaface JA |
3680 | } |
3681 | ||
52de1fe1 JA |
3682 | static int __io_recvmsg_copy_hdr(struct io_kiocb *req, struct io_async_ctx *io) |
3683 | { | |
3684 | struct io_sr_msg *sr = &req->sr_msg; | |
3685 | struct iovec __user *uiov; | |
3686 | size_t iov_len; | |
3687 | int ret; | |
3688 | ||
3689 | ret = __copy_msghdr_from_user(&io->msg.msg, sr->msg, &io->msg.uaddr, | |
3690 | &uiov, &iov_len); | |
3691 | if (ret) | |
3692 | return ret; | |
3693 | ||
3694 | if (req->flags & REQ_F_BUFFER_SELECT) { | |
3695 | if (iov_len > 1) | |
3696 | return -EINVAL; | |
3697 | if (copy_from_user(io->msg.iov, uiov, sizeof(*uiov))) | |
3698 | return -EFAULT; | |
3699 | sr->len = io->msg.iov[0].iov_len; | |
3700 | iov_iter_init(&io->msg.msg.msg_iter, READ, io->msg.iov, 1, | |
3701 | sr->len); | |
3702 | io->msg.iov = NULL; | |
3703 | } else { | |
3704 | ret = import_iovec(READ, uiov, iov_len, UIO_FASTIOV, | |
3705 | &io->msg.iov, &io->msg.msg.msg_iter); | |
3706 | if (ret > 0) | |
3707 | ret = 0; | |
3708 | } | |
3709 | ||
3710 | return ret; | |
3711 | } | |
3712 | ||
3713 | #ifdef CONFIG_COMPAT | |
3714 | static int __io_compat_recvmsg_copy_hdr(struct io_kiocb *req, | |
3715 | struct io_async_ctx *io) | |
3716 | { | |
3717 | struct compat_msghdr __user *msg_compat; | |
3718 | struct io_sr_msg *sr = &req->sr_msg; | |
3719 | struct compat_iovec __user *uiov; | |
3720 | compat_uptr_t ptr; | |
3721 | compat_size_t len; | |
3722 | int ret; | |
3723 | ||
3724 | msg_compat = (struct compat_msghdr __user *) sr->msg; | |
3725 | ret = __get_compat_msghdr(&io->msg.msg, msg_compat, &io->msg.uaddr, | |
3726 | &ptr, &len); | |
3727 | if (ret) | |
3728 | return ret; | |
3729 | ||
3730 | uiov = compat_ptr(ptr); | |
3731 | if (req->flags & REQ_F_BUFFER_SELECT) { | |
3732 | compat_ssize_t clen; | |
3733 | ||
3734 | if (len > 1) | |
3735 | return -EINVAL; | |
3736 | if (!access_ok(uiov, sizeof(*uiov))) | |
3737 | return -EFAULT; | |
3738 | if (__get_user(clen, &uiov->iov_len)) | |
3739 | return -EFAULT; | |
3740 | if (clen < 0) | |
3741 | return -EINVAL; | |
3742 | sr->len = io->msg.iov[0].iov_len; | |
3743 | io->msg.iov = NULL; | |
3744 | } else { | |
3745 | ret = compat_import_iovec(READ, uiov, len, UIO_FASTIOV, | |
3746 | &io->msg.iov, | |
3747 | &io->msg.msg.msg_iter); | |
3748 | if (ret < 0) | |
3749 | return ret; | |
3750 | } | |
3751 | ||
3752 | return 0; | |
3753 | } | |
3754 | #endif | |
3755 | ||
3756 | static int io_recvmsg_copy_hdr(struct io_kiocb *req, struct io_async_ctx *io) | |
3757 | { | |
3758 | io->msg.iov = io->msg.fast_iov; | |
3759 | ||
3760 | #ifdef CONFIG_COMPAT | |
3761 | if (req->ctx->compat) | |
3762 | return __io_compat_recvmsg_copy_hdr(req, io); | |
fddaface | 3763 | #endif |
52de1fe1 JA |
3764 | |
3765 | return __io_recvmsg_copy_hdr(req, io); | |
3766 | } | |
3767 | ||
bcda7baa JA |
3768 | static struct io_buffer *io_recv_buffer_select(struct io_kiocb *req, |
3769 | int *cflags, bool needs_lock) | |
3770 | { | |
3771 | struct io_sr_msg *sr = &req->sr_msg; | |
3772 | struct io_buffer *kbuf; | |
3773 | ||
3774 | if (!(req->flags & REQ_F_BUFFER_SELECT)) | |
3775 | return NULL; | |
3776 | ||
3777 | kbuf = io_buffer_select(req, &sr->len, sr->bgid, sr->kbuf, needs_lock); | |
3778 | if (IS_ERR(kbuf)) | |
3779 | return kbuf; | |
3780 | ||
3781 | sr->kbuf = kbuf; | |
3782 | req->flags |= REQ_F_BUFFER_SELECTED; | |
3783 | ||
3784 | *cflags = kbuf->bid << IORING_CQE_BUFFER_SHIFT; | |
3785 | *cflags |= IORING_CQE_F_BUFFER; | |
3786 | return kbuf; | |
fddaface JA |
3787 | } |
3788 | ||
3529d8c2 JA |
3789 | static int io_recvmsg_prep(struct io_kiocb *req, |
3790 | const struct io_uring_sqe *sqe) | |
aa1fa28f | 3791 | { |
e47293fd | 3792 | struct io_sr_msg *sr = &req->sr_msg; |
3529d8c2 | 3793 | struct io_async_ctx *io = req->io; |
99bc4c38 | 3794 | int ret; |
3529d8c2 JA |
3795 | |
3796 | sr->msg_flags = READ_ONCE(sqe->msg_flags); | |
3797 | sr->msg = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
0b7b21e4 | 3798 | sr->len = READ_ONCE(sqe->len); |
bcda7baa | 3799 | sr->bgid = READ_ONCE(sqe->buf_group); |
06b76d44 | 3800 | |
d8768362 JA |
3801 | #ifdef CONFIG_COMPAT |
3802 | if (req->ctx->compat) | |
3803 | sr->msg_flags |= MSG_CMSG_COMPAT; | |
3804 | #endif | |
3805 | ||
fddaface | 3806 | if (!io || req->opcode == IORING_OP_RECV) |
06b76d44 | 3807 | return 0; |
5f798bea PB |
3808 | /* iovec is already imported */ |
3809 | if (req->flags & REQ_F_NEED_CLEANUP) | |
3810 | return 0; | |
03b1230c | 3811 | |
52de1fe1 | 3812 | ret = io_recvmsg_copy_hdr(req, io); |
99bc4c38 PB |
3813 | if (!ret) |
3814 | req->flags |= REQ_F_NEED_CLEANUP; | |
3815 | return ret; | |
aa1fa28f JA |
3816 | } |
3817 | ||
014db007 | 3818 | static int io_recvmsg(struct io_kiocb *req, bool force_nonblock) |
aa1fa28f | 3819 | { |
0b416c3e | 3820 | struct io_async_msghdr *kmsg = NULL; |
03b1230c | 3821 | struct socket *sock; |
52de1fe1 | 3822 | int ret, cflags = 0; |
03b1230c JA |
3823 | |
3824 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) | |
3825 | return -EINVAL; | |
3826 | ||
3827 | sock = sock_from_file(req->file, &ret); | |
3828 | if (sock) { | |
52de1fe1 | 3829 | struct io_buffer *kbuf; |
b7bb4f7d | 3830 | struct io_async_ctx io; |
03b1230c JA |
3831 | unsigned flags; |
3832 | ||
03b1230c | 3833 | if (req->io) { |
0b416c3e | 3834 | kmsg = &req->io->msg; |
b537916c | 3835 | kmsg->msg.msg_name = &req->io->msg.addr; |
0b416c3e JA |
3836 | /* if iov is set, it's allocated already */ |
3837 | if (!kmsg->iov) | |
3838 | kmsg->iov = kmsg->fast_iov; | |
3839 | kmsg->msg.msg_iter.iov = kmsg->iov; | |
03b1230c | 3840 | } else { |
0b416c3e | 3841 | kmsg = &io.msg; |
b537916c | 3842 | kmsg->msg.msg_name = &io.msg.addr; |
3529d8c2 | 3843 | |
52de1fe1 | 3844 | ret = io_recvmsg_copy_hdr(req, &io); |
03b1230c | 3845 | if (ret) |
3529d8c2 | 3846 | return ret; |
03b1230c JA |
3847 | } |
3848 | ||
52de1fe1 JA |
3849 | kbuf = io_recv_buffer_select(req, &cflags, !force_nonblock); |
3850 | if (IS_ERR(kbuf)) { | |
3851 | return PTR_ERR(kbuf); | |
3852 | } else if (kbuf) { | |
3853 | kmsg->fast_iov[0].iov_base = u64_to_user_ptr(kbuf->addr); | |
3854 | iov_iter_init(&kmsg->msg.msg_iter, READ, kmsg->iov, | |
3855 | 1, req->sr_msg.len); | |
3856 | } | |
3857 | ||
e47293fd JA |
3858 | flags = req->sr_msg.msg_flags; |
3859 | if (flags & MSG_DONTWAIT) | |
3860 | req->flags |= REQ_F_NOWAIT; | |
3861 | else if (force_nonblock) | |
3862 | flags |= MSG_DONTWAIT; | |
3863 | ||
3864 | ret = __sys_recvmsg_sock(sock, &kmsg->msg, req->sr_msg.msg, | |
3865 | kmsg->uaddr, flags); | |
02d27d89 PB |
3866 | if (force_nonblock && ret == -EAGAIN) |
3867 | return io_setup_async_msg(req, kmsg); | |
03b1230c JA |
3868 | if (ret == -ERESTARTSYS) |
3869 | ret = -EINTR; | |
3870 | } | |
3871 | ||
1e95081c | 3872 | if (kmsg && kmsg->iov != kmsg->fast_iov) |
0b416c3e | 3873 | kfree(kmsg->iov); |
99bc4c38 | 3874 | req->flags &= ~REQ_F_NEED_CLEANUP; |
52de1fe1 | 3875 | __io_cqring_add_event(req, ret, cflags); |
4e88d6e7 JA |
3876 | if (ret < 0) |
3877 | req_set_fail_links(req); | |
014db007 | 3878 | io_put_req(req); |
03b1230c | 3879 | return 0; |
0fa03c62 | 3880 | } |
5d17b4a4 | 3881 | |
014db007 | 3882 | static int io_recv(struct io_kiocb *req, bool force_nonblock) |
fddaface | 3883 | { |
bcda7baa | 3884 | struct io_buffer *kbuf = NULL; |
fddaface | 3885 | struct socket *sock; |
bcda7baa | 3886 | int ret, cflags = 0; |
fddaface JA |
3887 | |
3888 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) | |
3889 | return -EINVAL; | |
3890 | ||
3891 | sock = sock_from_file(req->file, &ret); | |
3892 | if (sock) { | |
3893 | struct io_sr_msg *sr = &req->sr_msg; | |
bcda7baa | 3894 | void __user *buf = sr->buf; |
fddaface JA |
3895 | struct msghdr msg; |
3896 | struct iovec iov; | |
3897 | unsigned flags; | |
3898 | ||
bcda7baa JA |
3899 | kbuf = io_recv_buffer_select(req, &cflags, !force_nonblock); |
3900 | if (IS_ERR(kbuf)) | |
3901 | return PTR_ERR(kbuf); | |
3902 | else if (kbuf) | |
3903 | buf = u64_to_user_ptr(kbuf->addr); | |
3904 | ||
3905 | ret = import_single_range(READ, buf, sr->len, &iov, | |
fddaface | 3906 | &msg.msg_iter); |
bcda7baa JA |
3907 | if (ret) { |
3908 | kfree(kbuf); | |
fddaface | 3909 | return ret; |
bcda7baa | 3910 | } |
fddaface | 3911 | |
bcda7baa | 3912 | req->flags |= REQ_F_NEED_CLEANUP; |
fddaface JA |
3913 | msg.msg_name = NULL; |
3914 | msg.msg_control = NULL; | |
3915 | msg.msg_controllen = 0; | |
3916 | msg.msg_namelen = 0; | |
3917 | msg.msg_iocb = NULL; | |
3918 | msg.msg_flags = 0; | |
3919 | ||
3920 | flags = req->sr_msg.msg_flags; | |
3921 | if (flags & MSG_DONTWAIT) | |
3922 | req->flags |= REQ_F_NOWAIT; | |
3923 | else if (force_nonblock) | |
3924 | flags |= MSG_DONTWAIT; | |
3925 | ||
0b7b21e4 | 3926 | ret = sock_recvmsg(sock, &msg, flags); |
fddaface JA |
3927 | if (force_nonblock && ret == -EAGAIN) |
3928 | return -EAGAIN; | |
3929 | if (ret == -ERESTARTSYS) | |
3930 | ret = -EINTR; | |
3931 | } | |
3932 | ||
bcda7baa JA |
3933 | kfree(kbuf); |
3934 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
3935 | __io_cqring_add_event(req, ret, cflags); | |
fddaface JA |
3936 | if (ret < 0) |
3937 | req_set_fail_links(req); | |
014db007 | 3938 | io_put_req(req); |
fddaface | 3939 | return 0; |
fddaface JA |
3940 | } |
3941 | ||
3529d8c2 | 3942 | static int io_accept_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
17f2fe35 | 3943 | { |
8ed8d3c3 JA |
3944 | struct io_accept *accept = &req->accept; |
3945 | ||
17f2fe35 JA |
3946 | if (unlikely(req->ctx->flags & (IORING_SETUP_IOPOLL|IORING_SETUP_SQPOLL))) |
3947 | return -EINVAL; | |
8042d6ce | 3948 | if (sqe->ioprio || sqe->len || sqe->buf_index) |
17f2fe35 JA |
3949 | return -EINVAL; |
3950 | ||
d55e5f5b JA |
3951 | accept->addr = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
3952 | accept->addr_len = u64_to_user_ptr(READ_ONCE(sqe->addr2)); | |
8ed8d3c3 | 3953 | accept->flags = READ_ONCE(sqe->accept_flags); |
09952e3e | 3954 | accept->nofile = rlimit(RLIMIT_NOFILE); |
8ed8d3c3 | 3955 | return 0; |
8ed8d3c3 | 3956 | } |
17f2fe35 | 3957 | |
014db007 | 3958 | static int __io_accept(struct io_kiocb *req, bool force_nonblock) |
8ed8d3c3 JA |
3959 | { |
3960 | struct io_accept *accept = &req->accept; | |
3961 | unsigned file_flags; | |
3962 | int ret; | |
3963 | ||
3964 | file_flags = force_nonblock ? O_NONBLOCK : 0; | |
3965 | ret = __sys_accept4_file(req->file, file_flags, accept->addr, | |
09952e3e JA |
3966 | accept->addr_len, accept->flags, |
3967 | accept->nofile); | |
8ed8d3c3 | 3968 | if (ret == -EAGAIN && force_nonblock) |
17f2fe35 | 3969 | return -EAGAIN; |
8e3cca12 JA |
3970 | if (ret == -ERESTARTSYS) |
3971 | ret = -EINTR; | |
4e88d6e7 JA |
3972 | if (ret < 0) |
3973 | req_set_fail_links(req); | |
78e19bbe | 3974 | io_cqring_add_event(req, ret); |
014db007 | 3975 | io_put_req(req); |
17f2fe35 | 3976 | return 0; |
8ed8d3c3 JA |
3977 | } |
3978 | ||
3979 | static void io_accept_finish(struct io_wq_work **workptr) | |
3980 | { | |
3981 | struct io_kiocb *req = container_of(*workptr, struct io_kiocb, work); | |
8ed8d3c3 JA |
3982 | |
3983 | if (io_req_cancelled(req)) | |
3984 | return; | |
014db007 | 3985 | __io_accept(req, false); |
e9fd9396 | 3986 | io_steal_work(req, workptr); |
8ed8d3c3 | 3987 | } |
8ed8d3c3 | 3988 | |
014db007 | 3989 | static int io_accept(struct io_kiocb *req, bool force_nonblock) |
8ed8d3c3 | 3990 | { |
8ed8d3c3 JA |
3991 | int ret; |
3992 | ||
014db007 | 3993 | ret = __io_accept(req, force_nonblock); |
8ed8d3c3 JA |
3994 | if (ret == -EAGAIN && force_nonblock) { |
3995 | req->work.func = io_accept_finish; | |
8ed8d3c3 JA |
3996 | return -EAGAIN; |
3997 | } | |
3998 | return 0; | |
0fa03c62 | 3999 | } |
5d17b4a4 | 4000 | |
3529d8c2 | 4001 | static int io_connect_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
f499a021 | 4002 | { |
3529d8c2 JA |
4003 | struct io_connect *conn = &req->connect; |
4004 | struct io_async_ctx *io = req->io; | |
f499a021 | 4005 | |
3fbb51c1 JA |
4006 | if (unlikely(req->ctx->flags & (IORING_SETUP_IOPOLL|IORING_SETUP_SQPOLL))) |
4007 | return -EINVAL; | |
4008 | if (sqe->ioprio || sqe->len || sqe->buf_index || sqe->rw_flags) | |
4009 | return -EINVAL; | |
4010 | ||
3529d8c2 JA |
4011 | conn->addr = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
4012 | conn->addr_len = READ_ONCE(sqe->addr2); | |
4013 | ||
4014 | if (!io) | |
4015 | return 0; | |
4016 | ||
4017 | return move_addr_to_kernel(conn->addr, conn->addr_len, | |
3fbb51c1 | 4018 | &io->connect.address); |
f499a021 JA |
4019 | } |
4020 | ||
014db007 | 4021 | static int io_connect(struct io_kiocb *req, bool force_nonblock) |
f8e85cf2 | 4022 | { |
f499a021 | 4023 | struct io_async_ctx __io, *io; |
f8e85cf2 | 4024 | unsigned file_flags; |
3fbb51c1 | 4025 | int ret; |
f8e85cf2 | 4026 | |
f499a021 JA |
4027 | if (req->io) { |
4028 | io = req->io; | |
4029 | } else { | |
3529d8c2 JA |
4030 | ret = move_addr_to_kernel(req->connect.addr, |
4031 | req->connect.addr_len, | |
4032 | &__io.connect.address); | |
f499a021 JA |
4033 | if (ret) |
4034 | goto out; | |
4035 | io = &__io; | |
4036 | } | |
4037 | ||
3fbb51c1 JA |
4038 | file_flags = force_nonblock ? O_NONBLOCK : 0; |
4039 | ||
4040 | ret = __sys_connect_file(req->file, &io->connect.address, | |
4041 | req->connect.addr_len, file_flags); | |
87f80d62 | 4042 | if ((ret == -EAGAIN || ret == -EINPROGRESS) && force_nonblock) { |
b7bb4f7d JA |
4043 | if (req->io) |
4044 | return -EAGAIN; | |
4045 | if (io_alloc_async_ctx(req)) { | |
f499a021 JA |
4046 | ret = -ENOMEM; |
4047 | goto out; | |
4048 | } | |
b7bb4f7d | 4049 | memcpy(&req->io->connect, &__io.connect, sizeof(__io.connect)); |
f8e85cf2 | 4050 | return -EAGAIN; |
f499a021 | 4051 | } |
f8e85cf2 JA |
4052 | if (ret == -ERESTARTSYS) |
4053 | ret = -EINTR; | |
f499a021 | 4054 | out: |
4e88d6e7 JA |
4055 | if (ret < 0) |
4056 | req_set_fail_links(req); | |
f8e85cf2 | 4057 | io_cqring_add_event(req, ret); |
014db007 | 4058 | io_put_req(req); |
f8e85cf2 | 4059 | return 0; |
469956e8 Y |
4060 | } |
4061 | #else /* !CONFIG_NET */ | |
4062 | static int io_sendmsg_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) | |
4063 | { | |
f8e85cf2 | 4064 | return -EOPNOTSUPP; |
f8e85cf2 JA |
4065 | } |
4066 | ||
469956e8 Y |
4067 | static int io_sendmsg(struct io_kiocb *req, bool force_nonblock) |
4068 | { | |
4069 | return -EOPNOTSUPP; | |
4070 | } | |
4071 | ||
4072 | static int io_send(struct io_kiocb *req, bool force_nonblock) | |
4073 | { | |
4074 | return -EOPNOTSUPP; | |
4075 | } | |
4076 | ||
4077 | static int io_recvmsg_prep(struct io_kiocb *req, | |
4078 | const struct io_uring_sqe *sqe) | |
4079 | { | |
4080 | return -EOPNOTSUPP; | |
4081 | } | |
4082 | ||
4083 | static int io_recvmsg(struct io_kiocb *req, bool force_nonblock) | |
4084 | { | |
4085 | return -EOPNOTSUPP; | |
4086 | } | |
4087 | ||
4088 | static int io_recv(struct io_kiocb *req, bool force_nonblock) | |
4089 | { | |
4090 | return -EOPNOTSUPP; | |
4091 | } | |
4092 | ||
4093 | static int io_accept_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) | |
4094 | { | |
4095 | return -EOPNOTSUPP; | |
4096 | } | |
4097 | ||
4098 | static int io_accept(struct io_kiocb *req, bool force_nonblock) | |
4099 | { | |
4100 | return -EOPNOTSUPP; | |
4101 | } | |
4102 | ||
4103 | static int io_connect_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) | |
4104 | { | |
4105 | return -EOPNOTSUPP; | |
4106 | } | |
4107 | ||
4108 | static int io_connect(struct io_kiocb *req, bool force_nonblock) | |
4109 | { | |
f8e85cf2 | 4110 | return -EOPNOTSUPP; |
f8e85cf2 | 4111 | } |
469956e8 | 4112 | #endif /* CONFIG_NET */ |
f8e85cf2 | 4113 | |
d7718a9d JA |
4114 | struct io_poll_table { |
4115 | struct poll_table_struct pt; | |
4116 | struct io_kiocb *req; | |
4117 | int error; | |
4118 | }; | |
4119 | ||
4120 | static void __io_queue_proc(struct io_poll_iocb *poll, struct io_poll_table *pt, | |
4121 | struct wait_queue_head *head) | |
4122 | { | |
4123 | if (unlikely(poll->head)) { | |
4124 | pt->error = -EINVAL; | |
4125 | return; | |
4126 | } | |
4127 | ||
4128 | pt->error = 0; | |
4129 | poll->head = head; | |
4130 | add_wait_queue(head, &poll->wait); | |
4131 | } | |
4132 | ||
4133 | static void io_async_queue_proc(struct file *file, struct wait_queue_head *head, | |
4134 | struct poll_table_struct *p) | |
4135 | { | |
4136 | struct io_poll_table *pt = container_of(p, struct io_poll_table, pt); | |
4137 | ||
4138 | __io_queue_proc(&pt->req->apoll->poll, pt, head); | |
4139 | } | |
4140 | ||
4141 | static int __io_async_wake(struct io_kiocb *req, struct io_poll_iocb *poll, | |
4142 | __poll_t mask, task_work_func_t func) | |
4143 | { | |
4144 | struct task_struct *tsk; | |
aa96bf8a | 4145 | int ret; |
d7718a9d JA |
4146 | |
4147 | /* for instances that support it check for an event match first: */ | |
4148 | if (mask && !(mask & poll->events)) | |
4149 | return 0; | |
4150 | ||
4151 | trace_io_uring_task_add(req->ctx, req->opcode, req->user_data, mask); | |
4152 | ||
4153 | list_del_init(&poll->wait.entry); | |
4154 | ||
4155 | tsk = req->task; | |
4156 | req->result = mask; | |
4157 | init_task_work(&req->task_work, func); | |
4158 | /* | |
aa96bf8a JA |
4159 | * If this fails, then the task is exiting. Punt to one of the io-wq |
4160 | * threads to ensure the work gets run, we can't always rely on exit | |
4161 | * cancelation taking care of this. | |
d7718a9d | 4162 | */ |
aa96bf8a JA |
4163 | ret = task_work_add(tsk, &req->task_work, true); |
4164 | if (unlikely(ret)) { | |
4165 | tsk = io_wq_get_task(req->ctx->io_wq); | |
4166 | task_work_add(tsk, &req->task_work, true); | |
4167 | } | |
d7718a9d JA |
4168 | wake_up_process(tsk); |
4169 | return 1; | |
4170 | } | |
4171 | ||
4172 | static void io_async_task_func(struct callback_head *cb) | |
4173 | { | |
4174 | struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work); | |
4175 | struct async_poll *apoll = req->apoll; | |
4176 | struct io_ring_ctx *ctx = req->ctx; | |
4177 | ||
4178 | trace_io_uring_task_run(req->ctx, req->opcode, req->user_data); | |
4179 | ||
4180 | WARN_ON_ONCE(!list_empty(&req->apoll->poll.wait.entry)); | |
4181 | ||
4182 | if (hash_hashed(&req->hash_node)) { | |
4183 | spin_lock_irq(&ctx->completion_lock); | |
4184 | hash_del(&req->hash_node); | |
4185 | spin_unlock_irq(&ctx->completion_lock); | |
4186 | } | |
4187 | ||
4188 | /* restore ->work in case we need to retry again */ | |
4189 | memcpy(&req->work, &apoll->work, sizeof(req->work)); | |
4190 | ||
4191 | __set_current_state(TASK_RUNNING); | |
4192 | mutex_lock(&ctx->uring_lock); | |
4193 | __io_queue_sqe(req, NULL); | |
4194 | mutex_unlock(&ctx->uring_lock); | |
4195 | ||
4196 | kfree(apoll); | |
4197 | } | |
4198 | ||
4199 | static int io_async_wake(struct wait_queue_entry *wait, unsigned mode, int sync, | |
4200 | void *key) | |
4201 | { | |
4202 | struct io_kiocb *req = wait->private; | |
4203 | struct io_poll_iocb *poll = &req->apoll->poll; | |
4204 | ||
4205 | trace_io_uring_poll_wake(req->ctx, req->opcode, req->user_data, | |
4206 | key_to_poll(key)); | |
4207 | ||
4208 | return __io_async_wake(req, poll, key_to_poll(key), io_async_task_func); | |
4209 | } | |
4210 | ||
4211 | static void io_poll_req_insert(struct io_kiocb *req) | |
4212 | { | |
4213 | struct io_ring_ctx *ctx = req->ctx; | |
4214 | struct hlist_head *list; | |
4215 | ||
4216 | list = &ctx->cancel_hash[hash_long(req->user_data, ctx->cancel_hash_bits)]; | |
4217 | hlist_add_head(&req->hash_node, list); | |
4218 | } | |
4219 | ||
4220 | static __poll_t __io_arm_poll_handler(struct io_kiocb *req, | |
4221 | struct io_poll_iocb *poll, | |
4222 | struct io_poll_table *ipt, __poll_t mask, | |
4223 | wait_queue_func_t wake_func) | |
4224 | __acquires(&ctx->completion_lock) | |
4225 | { | |
4226 | struct io_ring_ctx *ctx = req->ctx; | |
4227 | bool cancel = false; | |
4228 | ||
4229 | poll->file = req->file; | |
4230 | poll->head = NULL; | |
4231 | poll->done = poll->canceled = false; | |
4232 | poll->events = mask; | |
4233 | ||
4234 | ipt->pt._key = mask; | |
4235 | ipt->req = req; | |
4236 | ipt->error = -EINVAL; | |
4237 | ||
4238 | INIT_LIST_HEAD(&poll->wait.entry); | |
4239 | init_waitqueue_func_entry(&poll->wait, wake_func); | |
4240 | poll->wait.private = req; | |
4241 | ||
4242 | mask = vfs_poll(req->file, &ipt->pt) & poll->events; | |
4243 | ||
4244 | spin_lock_irq(&ctx->completion_lock); | |
4245 | if (likely(poll->head)) { | |
4246 | spin_lock(&poll->head->lock); | |
4247 | if (unlikely(list_empty(&poll->wait.entry))) { | |
4248 | if (ipt->error) | |
4249 | cancel = true; | |
4250 | ipt->error = 0; | |
4251 | mask = 0; | |
4252 | } | |
4253 | if (mask || ipt->error) | |
4254 | list_del_init(&poll->wait.entry); | |
4255 | else if (cancel) | |
4256 | WRITE_ONCE(poll->canceled, true); | |
4257 | else if (!poll->done) /* actually waiting for an event */ | |
4258 | io_poll_req_insert(req); | |
4259 | spin_unlock(&poll->head->lock); | |
4260 | } | |
4261 | ||
4262 | return mask; | |
4263 | } | |
4264 | ||
4265 | static bool io_arm_poll_handler(struct io_kiocb *req) | |
4266 | { | |
4267 | const struct io_op_def *def = &io_op_defs[req->opcode]; | |
4268 | struct io_ring_ctx *ctx = req->ctx; | |
4269 | struct async_poll *apoll; | |
4270 | struct io_poll_table ipt; | |
4271 | __poll_t mask, ret; | |
4272 | ||
4273 | if (!req->file || !file_can_poll(req->file)) | |
4274 | return false; | |
4275 | if (req->flags & (REQ_F_MUST_PUNT | REQ_F_POLLED)) | |
4276 | return false; | |
4277 | if (!def->pollin && !def->pollout) | |
4278 | return false; | |
4279 | ||
4280 | apoll = kmalloc(sizeof(*apoll), GFP_ATOMIC); | |
4281 | if (unlikely(!apoll)) | |
4282 | return false; | |
4283 | ||
4284 | req->flags |= REQ_F_POLLED; | |
4285 | memcpy(&apoll->work, &req->work, sizeof(req->work)); | |
4286 | ||
3537b6a7 | 4287 | get_task_struct(current); |
d7718a9d JA |
4288 | req->task = current; |
4289 | req->apoll = apoll; | |
4290 | INIT_HLIST_NODE(&req->hash_node); | |
4291 | ||
8755d97a | 4292 | mask = 0; |
d7718a9d | 4293 | if (def->pollin) |
8755d97a | 4294 | mask |= POLLIN | POLLRDNORM; |
d7718a9d JA |
4295 | if (def->pollout) |
4296 | mask |= POLLOUT | POLLWRNORM; | |
4297 | mask |= POLLERR | POLLPRI; | |
4298 | ||
4299 | ipt.pt._qproc = io_async_queue_proc; | |
4300 | ||
4301 | ret = __io_arm_poll_handler(req, &apoll->poll, &ipt, mask, | |
4302 | io_async_wake); | |
4303 | if (ret) { | |
4304 | ipt.error = 0; | |
4305 | apoll->poll.done = true; | |
4306 | spin_unlock_irq(&ctx->completion_lock); | |
4307 | memcpy(&req->work, &apoll->work, sizeof(req->work)); | |
4308 | kfree(apoll); | |
4309 | return false; | |
4310 | } | |
4311 | spin_unlock_irq(&ctx->completion_lock); | |
4312 | trace_io_uring_poll_arm(ctx, req->opcode, req->user_data, mask, | |
4313 | apoll->poll.events); | |
4314 | return true; | |
4315 | } | |
4316 | ||
4317 | static bool __io_poll_remove_one(struct io_kiocb *req, | |
4318 | struct io_poll_iocb *poll) | |
221c5eb2 | 4319 | { |
b41e9852 | 4320 | bool do_complete = false; |
221c5eb2 JA |
4321 | |
4322 | spin_lock(&poll->head->lock); | |
4323 | WRITE_ONCE(poll->canceled, true); | |
392edb45 JA |
4324 | if (!list_empty(&poll->wait.entry)) { |
4325 | list_del_init(&poll->wait.entry); | |
b41e9852 | 4326 | do_complete = true; |
221c5eb2 JA |
4327 | } |
4328 | spin_unlock(&poll->head->lock); | |
d7718a9d JA |
4329 | return do_complete; |
4330 | } | |
4331 | ||
4332 | static bool io_poll_remove_one(struct io_kiocb *req) | |
4333 | { | |
4334 | bool do_complete; | |
4335 | ||
4336 | if (req->opcode == IORING_OP_POLL_ADD) { | |
4337 | do_complete = __io_poll_remove_one(req, &req->poll); | |
4338 | } else { | |
4339 | /* non-poll requests have submit ref still */ | |
4340 | do_complete = __io_poll_remove_one(req, &req->apoll->poll); | |
4341 | if (do_complete) | |
4342 | io_put_req(req); | |
4343 | } | |
4344 | ||
78076bb6 | 4345 | hash_del(&req->hash_node); |
d7718a9d | 4346 | |
b41e9852 JA |
4347 | if (do_complete) { |
4348 | io_cqring_fill_event(req, -ECANCELED); | |
4349 | io_commit_cqring(req->ctx); | |
4350 | req->flags |= REQ_F_COMP_LOCKED; | |
4351 | io_put_req(req); | |
4352 | } | |
4353 | ||
4354 | return do_complete; | |
221c5eb2 JA |
4355 | } |
4356 | ||
4357 | static void io_poll_remove_all(struct io_ring_ctx *ctx) | |
4358 | { | |
78076bb6 | 4359 | struct hlist_node *tmp; |
221c5eb2 | 4360 | struct io_kiocb *req; |
78076bb6 | 4361 | int i; |
221c5eb2 JA |
4362 | |
4363 | spin_lock_irq(&ctx->completion_lock); | |
78076bb6 JA |
4364 | for (i = 0; i < (1U << ctx->cancel_hash_bits); i++) { |
4365 | struct hlist_head *list; | |
4366 | ||
4367 | list = &ctx->cancel_hash[i]; | |
4368 | hlist_for_each_entry_safe(req, tmp, list, hash_node) | |
4369 | io_poll_remove_one(req); | |
221c5eb2 JA |
4370 | } |
4371 | spin_unlock_irq(&ctx->completion_lock); | |
b41e9852 JA |
4372 | |
4373 | io_cqring_ev_posted(ctx); | |
221c5eb2 JA |
4374 | } |
4375 | ||
47f46768 JA |
4376 | static int io_poll_cancel(struct io_ring_ctx *ctx, __u64 sqe_addr) |
4377 | { | |
78076bb6 | 4378 | struct hlist_head *list; |
47f46768 JA |
4379 | struct io_kiocb *req; |
4380 | ||
78076bb6 JA |
4381 | list = &ctx->cancel_hash[hash_long(sqe_addr, ctx->cancel_hash_bits)]; |
4382 | hlist_for_each_entry(req, list, hash_node) { | |
b41e9852 JA |
4383 | if (sqe_addr != req->user_data) |
4384 | continue; | |
4385 | if (io_poll_remove_one(req)) | |
eac406c6 | 4386 | return 0; |
b41e9852 | 4387 | return -EALREADY; |
47f46768 JA |
4388 | } |
4389 | ||
4390 | return -ENOENT; | |
4391 | } | |
4392 | ||
3529d8c2 JA |
4393 | static int io_poll_remove_prep(struct io_kiocb *req, |
4394 | const struct io_uring_sqe *sqe) | |
0969e783 | 4395 | { |
0969e783 JA |
4396 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4397 | return -EINVAL; | |
4398 | if (sqe->ioprio || sqe->off || sqe->len || sqe->buf_index || | |
4399 | sqe->poll_events) | |
4400 | return -EINVAL; | |
4401 | ||
4402 | req->poll.addr = READ_ONCE(sqe->addr); | |
0969e783 JA |
4403 | return 0; |
4404 | } | |
4405 | ||
221c5eb2 JA |
4406 | /* |
4407 | * Find a running poll command that matches one specified in sqe->addr, | |
4408 | * and remove it if found. | |
4409 | */ | |
fc4df999 | 4410 | static int io_poll_remove(struct io_kiocb *req) |
221c5eb2 JA |
4411 | { |
4412 | struct io_ring_ctx *ctx = req->ctx; | |
0969e783 | 4413 | u64 addr; |
47f46768 | 4414 | int ret; |
221c5eb2 | 4415 | |
0969e783 | 4416 | addr = req->poll.addr; |
221c5eb2 | 4417 | spin_lock_irq(&ctx->completion_lock); |
0969e783 | 4418 | ret = io_poll_cancel(ctx, addr); |
221c5eb2 JA |
4419 | spin_unlock_irq(&ctx->completion_lock); |
4420 | ||
78e19bbe | 4421 | io_cqring_add_event(req, ret); |
4e88d6e7 JA |
4422 | if (ret < 0) |
4423 | req_set_fail_links(req); | |
e65ef56d | 4424 | io_put_req(req); |
221c5eb2 JA |
4425 | return 0; |
4426 | } | |
4427 | ||
b0dd8a41 | 4428 | static void io_poll_complete(struct io_kiocb *req, __poll_t mask, int error) |
221c5eb2 | 4429 | { |
a197f664 JL |
4430 | struct io_ring_ctx *ctx = req->ctx; |
4431 | ||
8c838788 | 4432 | req->poll.done = true; |
b0a20349 | 4433 | io_cqring_fill_event(req, error ? error : mangle_poll(mask)); |
8c838788 | 4434 | io_commit_cqring(ctx); |
221c5eb2 JA |
4435 | } |
4436 | ||
b41e9852 | 4437 | static void io_poll_task_handler(struct io_kiocb *req, struct io_kiocb **nxt) |
221c5eb2 | 4438 | { |
221c5eb2 | 4439 | struct io_ring_ctx *ctx = req->ctx; |
a6ba632d JA |
4440 | struct io_poll_iocb *poll = &req->poll; |
4441 | ||
4442 | if (!req->result && !READ_ONCE(poll->canceled)) { | |
4443 | struct poll_table_struct pt = { ._key = poll->events }; | |
4444 | ||
4445 | req->result = vfs_poll(req->file, &pt) & poll->events; | |
4446 | } | |
221c5eb2 | 4447 | |
221c5eb2 | 4448 | spin_lock_irq(&ctx->completion_lock); |
a6ba632d JA |
4449 | if (!req->result && !READ_ONCE(poll->canceled)) { |
4450 | add_wait_queue(poll->head, &poll->wait); | |
4451 | spin_unlock_irq(&ctx->completion_lock); | |
4452 | return; | |
4453 | } | |
78076bb6 | 4454 | hash_del(&req->hash_node); |
b41e9852 JA |
4455 | io_poll_complete(req, req->result, 0); |
4456 | req->flags |= REQ_F_COMP_LOCKED; | |
4457 | io_put_req_find_next(req, nxt); | |
221c5eb2 JA |
4458 | spin_unlock_irq(&ctx->completion_lock); |
4459 | ||
8c838788 | 4460 | io_cqring_ev_posted(ctx); |
221c5eb2 JA |
4461 | } |
4462 | ||
b41e9852 | 4463 | static void io_poll_task_func(struct callback_head *cb) |
e94f141b | 4464 | { |
b41e9852 JA |
4465 | struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work); |
4466 | struct io_kiocb *nxt = NULL; | |
e94f141b | 4467 | |
b41e9852 | 4468 | io_poll_task_handler(req, &nxt); |
d7718a9d JA |
4469 | if (nxt) { |
4470 | struct io_ring_ctx *ctx = nxt->ctx; | |
e94f141b | 4471 | |
d7718a9d | 4472 | mutex_lock(&ctx->uring_lock); |
b41e9852 | 4473 | __io_queue_sqe(nxt, NULL); |
d7718a9d | 4474 | mutex_unlock(&ctx->uring_lock); |
e94f141b | 4475 | } |
f0b493e6 JA |
4476 | } |
4477 | ||
221c5eb2 JA |
4478 | static int io_poll_wake(struct wait_queue_entry *wait, unsigned mode, int sync, |
4479 | void *key) | |
4480 | { | |
c2f2eb7d JA |
4481 | struct io_kiocb *req = wait->private; |
4482 | struct io_poll_iocb *poll = &req->poll; | |
221c5eb2 | 4483 | |
d7718a9d | 4484 | return __io_async_wake(req, poll, key_to_poll(key), io_poll_task_func); |
221c5eb2 JA |
4485 | } |
4486 | ||
221c5eb2 JA |
4487 | static void io_poll_queue_proc(struct file *file, struct wait_queue_head *head, |
4488 | struct poll_table_struct *p) | |
4489 | { | |
4490 | struct io_poll_table *pt = container_of(p, struct io_poll_table, pt); | |
4491 | ||
d7718a9d | 4492 | __io_queue_proc(&pt->req->poll, pt, head); |
eac406c6 JA |
4493 | } |
4494 | ||
3529d8c2 | 4495 | static int io_poll_add_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
221c5eb2 JA |
4496 | { |
4497 | struct io_poll_iocb *poll = &req->poll; | |
221c5eb2 | 4498 | u16 events; |
221c5eb2 JA |
4499 | |
4500 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) | |
4501 | return -EINVAL; | |
4502 | if (sqe->addr || sqe->ioprio || sqe->off || sqe->len || sqe->buf_index) | |
4503 | return -EINVAL; | |
09bb8394 JA |
4504 | if (!poll->file) |
4505 | return -EBADF; | |
221c5eb2 | 4506 | |
221c5eb2 JA |
4507 | events = READ_ONCE(sqe->poll_events); |
4508 | poll->events = demangle_poll(events) | EPOLLERR | EPOLLHUP; | |
b41e9852 | 4509 | |
3537b6a7 | 4510 | get_task_struct(current); |
b41e9852 | 4511 | req->task = current; |
0969e783 JA |
4512 | return 0; |
4513 | } | |
4514 | ||
014db007 | 4515 | static int io_poll_add(struct io_kiocb *req) |
0969e783 JA |
4516 | { |
4517 | struct io_poll_iocb *poll = &req->poll; | |
4518 | struct io_ring_ctx *ctx = req->ctx; | |
4519 | struct io_poll_table ipt; | |
0969e783 | 4520 | __poll_t mask; |
0969e783 | 4521 | |
78076bb6 | 4522 | INIT_HLIST_NODE(&req->hash_node); |
36703247 | 4523 | INIT_LIST_HEAD(&req->list); |
d7718a9d | 4524 | ipt.pt._qproc = io_poll_queue_proc; |
36703247 | 4525 | |
d7718a9d JA |
4526 | mask = __io_arm_poll_handler(req, &req->poll, &ipt, poll->events, |
4527 | io_poll_wake); | |
221c5eb2 | 4528 | |
8c838788 | 4529 | if (mask) { /* no async, we'd stolen it */ |
221c5eb2 | 4530 | ipt.error = 0; |
b0dd8a41 | 4531 | io_poll_complete(req, mask, 0); |
221c5eb2 | 4532 | } |
221c5eb2 JA |
4533 | spin_unlock_irq(&ctx->completion_lock); |
4534 | ||
8c838788 JA |
4535 | if (mask) { |
4536 | io_cqring_ev_posted(ctx); | |
014db007 | 4537 | io_put_req(req); |
221c5eb2 | 4538 | } |
8c838788 | 4539 | return ipt.error; |
221c5eb2 JA |
4540 | } |
4541 | ||
5262f567 JA |
4542 | static enum hrtimer_restart io_timeout_fn(struct hrtimer *timer) |
4543 | { | |
ad8a48ac JA |
4544 | struct io_timeout_data *data = container_of(timer, |
4545 | struct io_timeout_data, timer); | |
4546 | struct io_kiocb *req = data->req; | |
4547 | struct io_ring_ctx *ctx = req->ctx; | |
5262f567 JA |
4548 | unsigned long flags; |
4549 | ||
5262f567 JA |
4550 | atomic_inc(&ctx->cq_timeouts); |
4551 | ||
4552 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
ef03681a | 4553 | /* |
11365043 JA |
4554 | * We could be racing with timeout deletion. If the list is empty, |
4555 | * then timeout lookup already found it and will be handling it. | |
ef03681a | 4556 | */ |
842f9612 | 4557 | if (!list_empty(&req->list)) { |
11365043 | 4558 | struct io_kiocb *prev; |
5262f567 | 4559 | |
11365043 JA |
4560 | /* |
4561 | * Adjust the reqs sequence before the current one because it | |
d195a66e | 4562 | * will consume a slot in the cq_ring and the cq_tail |
11365043 JA |
4563 | * pointer will be increased, otherwise other timeout reqs may |
4564 | * return in advance without waiting for enough wait_nr. | |
4565 | */ | |
4566 | prev = req; | |
4567 | list_for_each_entry_continue_reverse(prev, &ctx->timeout_list, list) | |
4568 | prev->sequence++; | |
11365043 | 4569 | list_del_init(&req->list); |
11365043 | 4570 | } |
5262f567 | 4571 | |
78e19bbe | 4572 | io_cqring_fill_event(req, -ETIME); |
5262f567 JA |
4573 | io_commit_cqring(ctx); |
4574 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
4575 | ||
4576 | io_cqring_ev_posted(ctx); | |
4e88d6e7 | 4577 | req_set_fail_links(req); |
5262f567 JA |
4578 | io_put_req(req); |
4579 | return HRTIMER_NORESTART; | |
4580 | } | |
4581 | ||
47f46768 JA |
4582 | static int io_timeout_cancel(struct io_ring_ctx *ctx, __u64 user_data) |
4583 | { | |
4584 | struct io_kiocb *req; | |
4585 | int ret = -ENOENT; | |
4586 | ||
4587 | list_for_each_entry(req, &ctx->timeout_list, list) { | |
4588 | if (user_data == req->user_data) { | |
4589 | list_del_init(&req->list); | |
4590 | ret = 0; | |
4591 | break; | |
4592 | } | |
4593 | } | |
4594 | ||
4595 | if (ret == -ENOENT) | |
4596 | return ret; | |
4597 | ||
2d28390a | 4598 | ret = hrtimer_try_to_cancel(&req->io->timeout.timer); |
47f46768 JA |
4599 | if (ret == -1) |
4600 | return -EALREADY; | |
4601 | ||
4e88d6e7 | 4602 | req_set_fail_links(req); |
47f46768 JA |
4603 | io_cqring_fill_event(req, -ECANCELED); |
4604 | io_put_req(req); | |
4605 | return 0; | |
4606 | } | |
4607 | ||
3529d8c2 JA |
4608 | static int io_timeout_remove_prep(struct io_kiocb *req, |
4609 | const struct io_uring_sqe *sqe) | |
b29472ee | 4610 | { |
b29472ee JA |
4611 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4612 | return -EINVAL; | |
4613 | if (sqe->flags || sqe->ioprio || sqe->buf_index || sqe->len) | |
4614 | return -EINVAL; | |
4615 | ||
4616 | req->timeout.addr = READ_ONCE(sqe->addr); | |
4617 | req->timeout.flags = READ_ONCE(sqe->timeout_flags); | |
4618 | if (req->timeout.flags) | |
4619 | return -EINVAL; | |
4620 | ||
b29472ee JA |
4621 | return 0; |
4622 | } | |
4623 | ||
11365043 JA |
4624 | /* |
4625 | * Remove or update an existing timeout command | |
4626 | */ | |
fc4df999 | 4627 | static int io_timeout_remove(struct io_kiocb *req) |
11365043 JA |
4628 | { |
4629 | struct io_ring_ctx *ctx = req->ctx; | |
47f46768 | 4630 | int ret; |
11365043 | 4631 | |
11365043 | 4632 | spin_lock_irq(&ctx->completion_lock); |
b29472ee | 4633 | ret = io_timeout_cancel(ctx, req->timeout.addr); |
11365043 | 4634 | |
47f46768 | 4635 | io_cqring_fill_event(req, ret); |
11365043 JA |
4636 | io_commit_cqring(ctx); |
4637 | spin_unlock_irq(&ctx->completion_lock); | |
5262f567 | 4638 | io_cqring_ev_posted(ctx); |
4e88d6e7 JA |
4639 | if (ret < 0) |
4640 | req_set_fail_links(req); | |
ec9c02ad | 4641 | io_put_req(req); |
11365043 | 4642 | return 0; |
5262f567 JA |
4643 | } |
4644 | ||
3529d8c2 | 4645 | static int io_timeout_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe, |
2d28390a | 4646 | bool is_timeout_link) |
5262f567 | 4647 | { |
ad8a48ac | 4648 | struct io_timeout_data *data; |
a41525ab | 4649 | unsigned flags; |
5262f567 | 4650 | |
ad8a48ac | 4651 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
5262f567 | 4652 | return -EINVAL; |
ad8a48ac | 4653 | if (sqe->ioprio || sqe->buf_index || sqe->len != 1) |
a41525ab | 4654 | return -EINVAL; |
2d28390a JA |
4655 | if (sqe->off && is_timeout_link) |
4656 | return -EINVAL; | |
a41525ab JA |
4657 | flags = READ_ONCE(sqe->timeout_flags); |
4658 | if (flags & ~IORING_TIMEOUT_ABS) | |
5262f567 | 4659 | return -EINVAL; |
bdf20073 | 4660 | |
26a61679 JA |
4661 | req->timeout.count = READ_ONCE(sqe->off); |
4662 | ||
3529d8c2 | 4663 | if (!req->io && io_alloc_async_ctx(req)) |
26a61679 JA |
4664 | return -ENOMEM; |
4665 | ||
4666 | data = &req->io->timeout; | |
ad8a48ac | 4667 | data->req = req; |
ad8a48ac JA |
4668 | req->flags |= REQ_F_TIMEOUT; |
4669 | ||
4670 | if (get_timespec64(&data->ts, u64_to_user_ptr(sqe->addr))) | |
5262f567 JA |
4671 | return -EFAULT; |
4672 | ||
11365043 | 4673 | if (flags & IORING_TIMEOUT_ABS) |
ad8a48ac | 4674 | data->mode = HRTIMER_MODE_ABS; |
11365043 | 4675 | else |
ad8a48ac | 4676 | data->mode = HRTIMER_MODE_REL; |
11365043 | 4677 | |
ad8a48ac JA |
4678 | hrtimer_init(&data->timer, CLOCK_MONOTONIC, data->mode); |
4679 | return 0; | |
4680 | } | |
4681 | ||
fc4df999 | 4682 | static int io_timeout(struct io_kiocb *req) |
ad8a48ac JA |
4683 | { |
4684 | unsigned count; | |
4685 | struct io_ring_ctx *ctx = req->ctx; | |
4686 | struct io_timeout_data *data; | |
4687 | struct list_head *entry; | |
4688 | unsigned span = 0; | |
ad8a48ac | 4689 | |
2d28390a | 4690 | data = &req->io->timeout; |
93bd25bb | 4691 | |
5262f567 JA |
4692 | /* |
4693 | * sqe->off holds how many events that need to occur for this | |
93bd25bb JA |
4694 | * timeout event to be satisfied. If it isn't set, then this is |
4695 | * a pure timeout request, sequence isn't used. | |
5262f567 | 4696 | */ |
26a61679 | 4697 | count = req->timeout.count; |
93bd25bb JA |
4698 | if (!count) { |
4699 | req->flags |= REQ_F_TIMEOUT_NOSEQ; | |
4700 | spin_lock_irq(&ctx->completion_lock); | |
4701 | entry = ctx->timeout_list.prev; | |
4702 | goto add; | |
4703 | } | |
5262f567 JA |
4704 | |
4705 | req->sequence = ctx->cached_sq_head + count - 1; | |
2d28390a | 4706 | data->seq_offset = count; |
5262f567 JA |
4707 | |
4708 | /* | |
4709 | * Insertion sort, ensuring the first entry in the list is always | |
4710 | * the one we need first. | |
4711 | */ | |
5262f567 JA |
4712 | spin_lock_irq(&ctx->completion_lock); |
4713 | list_for_each_prev(entry, &ctx->timeout_list) { | |
4714 | struct io_kiocb *nxt = list_entry(entry, struct io_kiocb, list); | |
5da0fb1a | 4715 | unsigned nxt_sq_head; |
4716 | long long tmp, tmp_nxt; | |
2d28390a | 4717 | u32 nxt_offset = nxt->io->timeout.seq_offset; |
5262f567 | 4718 | |
93bd25bb JA |
4719 | if (nxt->flags & REQ_F_TIMEOUT_NOSEQ) |
4720 | continue; | |
4721 | ||
5da0fb1a | 4722 | /* |
4723 | * Since cached_sq_head + count - 1 can overflow, use type long | |
4724 | * long to store it. | |
4725 | */ | |
4726 | tmp = (long long)ctx->cached_sq_head + count - 1; | |
cc42e0ac PB |
4727 | nxt_sq_head = nxt->sequence - nxt_offset + 1; |
4728 | tmp_nxt = (long long)nxt_sq_head + nxt_offset - 1; | |
5da0fb1a | 4729 | |
4730 | /* | |
4731 | * cached_sq_head may overflow, and it will never overflow twice | |
4732 | * once there is some timeout req still be valid. | |
4733 | */ | |
4734 | if (ctx->cached_sq_head < nxt_sq_head) | |
8b07a65a | 4735 | tmp += UINT_MAX; |
5da0fb1a | 4736 | |
a1f58ba4 | 4737 | if (tmp > tmp_nxt) |
5262f567 | 4738 | break; |
a1f58ba4 | 4739 | |
4740 | /* | |
4741 | * Sequence of reqs after the insert one and itself should | |
4742 | * be adjusted because each timeout req consumes a slot. | |
4743 | */ | |
4744 | span++; | |
4745 | nxt->sequence++; | |
5262f567 | 4746 | } |
a1f58ba4 | 4747 | req->sequence -= span; |
93bd25bb | 4748 | add: |
5262f567 | 4749 | list_add(&req->list, entry); |
ad8a48ac JA |
4750 | data->timer.function = io_timeout_fn; |
4751 | hrtimer_start(&data->timer, timespec64_to_ktime(data->ts), data->mode); | |
5262f567 | 4752 | spin_unlock_irq(&ctx->completion_lock); |
5262f567 JA |
4753 | return 0; |
4754 | } | |
5262f567 | 4755 | |
62755e35 JA |
4756 | static bool io_cancel_cb(struct io_wq_work *work, void *data) |
4757 | { | |
4758 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); | |
4759 | ||
4760 | return req->user_data == (unsigned long) data; | |
4761 | } | |
4762 | ||
e977d6d3 | 4763 | static int io_async_cancel_one(struct io_ring_ctx *ctx, void *sqe_addr) |
62755e35 | 4764 | { |
62755e35 | 4765 | enum io_wq_cancel cancel_ret; |
62755e35 JA |
4766 | int ret = 0; |
4767 | ||
62755e35 JA |
4768 | cancel_ret = io_wq_cancel_cb(ctx->io_wq, io_cancel_cb, sqe_addr); |
4769 | switch (cancel_ret) { | |
4770 | case IO_WQ_CANCEL_OK: | |
4771 | ret = 0; | |
4772 | break; | |
4773 | case IO_WQ_CANCEL_RUNNING: | |
4774 | ret = -EALREADY; | |
4775 | break; | |
4776 | case IO_WQ_CANCEL_NOTFOUND: | |
4777 | ret = -ENOENT; | |
4778 | break; | |
4779 | } | |
4780 | ||
e977d6d3 JA |
4781 | return ret; |
4782 | } | |
4783 | ||
47f46768 JA |
4784 | static void io_async_find_and_cancel(struct io_ring_ctx *ctx, |
4785 | struct io_kiocb *req, __u64 sqe_addr, | |
014db007 | 4786 | int success_ret) |
47f46768 JA |
4787 | { |
4788 | unsigned long flags; | |
4789 | int ret; | |
4790 | ||
4791 | ret = io_async_cancel_one(ctx, (void *) (unsigned long) sqe_addr); | |
4792 | if (ret != -ENOENT) { | |
4793 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
4794 | goto done; | |
4795 | } | |
4796 | ||
4797 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
4798 | ret = io_timeout_cancel(ctx, sqe_addr); | |
4799 | if (ret != -ENOENT) | |
4800 | goto done; | |
4801 | ret = io_poll_cancel(ctx, sqe_addr); | |
4802 | done: | |
b0dd8a41 JA |
4803 | if (!ret) |
4804 | ret = success_ret; | |
47f46768 JA |
4805 | io_cqring_fill_event(req, ret); |
4806 | io_commit_cqring(ctx); | |
4807 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
4808 | io_cqring_ev_posted(ctx); | |
4809 | ||
4e88d6e7 JA |
4810 | if (ret < 0) |
4811 | req_set_fail_links(req); | |
014db007 | 4812 | io_put_req(req); |
47f46768 JA |
4813 | } |
4814 | ||
3529d8c2 JA |
4815 | static int io_async_cancel_prep(struct io_kiocb *req, |
4816 | const struct io_uring_sqe *sqe) | |
e977d6d3 | 4817 | { |
fbf23849 | 4818 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
e977d6d3 JA |
4819 | return -EINVAL; |
4820 | if (sqe->flags || sqe->ioprio || sqe->off || sqe->len || | |
4821 | sqe->cancel_flags) | |
4822 | return -EINVAL; | |
4823 | ||
fbf23849 JA |
4824 | req->cancel.addr = READ_ONCE(sqe->addr); |
4825 | return 0; | |
4826 | } | |
4827 | ||
014db007 | 4828 | static int io_async_cancel(struct io_kiocb *req) |
fbf23849 JA |
4829 | { |
4830 | struct io_ring_ctx *ctx = req->ctx; | |
fbf23849 | 4831 | |
014db007 | 4832 | io_async_find_and_cancel(ctx, req, req->cancel.addr, 0); |
5262f567 JA |
4833 | return 0; |
4834 | } | |
4835 | ||
05f3fb3c JA |
4836 | static int io_files_update_prep(struct io_kiocb *req, |
4837 | const struct io_uring_sqe *sqe) | |
4838 | { | |
4839 | if (sqe->flags || sqe->ioprio || sqe->rw_flags) | |
4840 | return -EINVAL; | |
4841 | ||
4842 | req->files_update.offset = READ_ONCE(sqe->off); | |
4843 | req->files_update.nr_args = READ_ONCE(sqe->len); | |
4844 | if (!req->files_update.nr_args) | |
4845 | return -EINVAL; | |
4846 | req->files_update.arg = READ_ONCE(sqe->addr); | |
4847 | return 0; | |
4848 | } | |
4849 | ||
4850 | static int io_files_update(struct io_kiocb *req, bool force_nonblock) | |
fbf23849 JA |
4851 | { |
4852 | struct io_ring_ctx *ctx = req->ctx; | |
05f3fb3c JA |
4853 | struct io_uring_files_update up; |
4854 | int ret; | |
fbf23849 | 4855 | |
f86cd20c | 4856 | if (force_nonblock) |
05f3fb3c | 4857 | return -EAGAIN; |
05f3fb3c JA |
4858 | |
4859 | up.offset = req->files_update.offset; | |
4860 | up.fds = req->files_update.arg; | |
4861 | ||
4862 | mutex_lock(&ctx->uring_lock); | |
4863 | ret = __io_sqe_files_update(ctx, &up, req->files_update.nr_args); | |
4864 | mutex_unlock(&ctx->uring_lock); | |
4865 | ||
4866 | if (ret < 0) | |
4867 | req_set_fail_links(req); | |
4868 | io_cqring_add_event(req, ret); | |
4869 | io_put_req(req); | |
5262f567 JA |
4870 | return 0; |
4871 | } | |
4872 | ||
3529d8c2 JA |
4873 | static int io_req_defer_prep(struct io_kiocb *req, |
4874 | const struct io_uring_sqe *sqe) | |
f67676d1 | 4875 | { |
e781573e | 4876 | ssize_t ret = 0; |
f67676d1 | 4877 | |
f1d96a8f PB |
4878 | if (!sqe) |
4879 | return 0; | |
4880 | ||
f86cd20c JA |
4881 | if (io_op_defs[req->opcode].file_table) { |
4882 | ret = io_grab_files(req); | |
4883 | if (unlikely(ret)) | |
4884 | return ret; | |
4885 | } | |
4886 | ||
cccf0ee8 JA |
4887 | io_req_work_grab_env(req, &io_op_defs[req->opcode]); |
4888 | ||
d625c6ee | 4889 | switch (req->opcode) { |
e781573e JA |
4890 | case IORING_OP_NOP: |
4891 | break; | |
f67676d1 JA |
4892 | case IORING_OP_READV: |
4893 | case IORING_OP_READ_FIXED: | |
3a6820f2 | 4894 | case IORING_OP_READ: |
3529d8c2 | 4895 | ret = io_read_prep(req, sqe, true); |
f67676d1 JA |
4896 | break; |
4897 | case IORING_OP_WRITEV: | |
4898 | case IORING_OP_WRITE_FIXED: | |
3a6820f2 | 4899 | case IORING_OP_WRITE: |
3529d8c2 | 4900 | ret = io_write_prep(req, sqe, true); |
f67676d1 | 4901 | break; |
0969e783 | 4902 | case IORING_OP_POLL_ADD: |
3529d8c2 | 4903 | ret = io_poll_add_prep(req, sqe); |
0969e783 JA |
4904 | break; |
4905 | case IORING_OP_POLL_REMOVE: | |
3529d8c2 | 4906 | ret = io_poll_remove_prep(req, sqe); |
0969e783 | 4907 | break; |
8ed8d3c3 | 4908 | case IORING_OP_FSYNC: |
3529d8c2 | 4909 | ret = io_prep_fsync(req, sqe); |
8ed8d3c3 JA |
4910 | break; |
4911 | case IORING_OP_SYNC_FILE_RANGE: | |
3529d8c2 | 4912 | ret = io_prep_sfr(req, sqe); |
8ed8d3c3 | 4913 | break; |
03b1230c | 4914 | case IORING_OP_SENDMSG: |
fddaface | 4915 | case IORING_OP_SEND: |
3529d8c2 | 4916 | ret = io_sendmsg_prep(req, sqe); |
03b1230c JA |
4917 | break; |
4918 | case IORING_OP_RECVMSG: | |
fddaface | 4919 | case IORING_OP_RECV: |
3529d8c2 | 4920 | ret = io_recvmsg_prep(req, sqe); |
03b1230c | 4921 | break; |
f499a021 | 4922 | case IORING_OP_CONNECT: |
3529d8c2 | 4923 | ret = io_connect_prep(req, sqe); |
f499a021 | 4924 | break; |
2d28390a | 4925 | case IORING_OP_TIMEOUT: |
3529d8c2 | 4926 | ret = io_timeout_prep(req, sqe, false); |
b7bb4f7d | 4927 | break; |
b29472ee | 4928 | case IORING_OP_TIMEOUT_REMOVE: |
3529d8c2 | 4929 | ret = io_timeout_remove_prep(req, sqe); |
b29472ee | 4930 | break; |
fbf23849 | 4931 | case IORING_OP_ASYNC_CANCEL: |
3529d8c2 | 4932 | ret = io_async_cancel_prep(req, sqe); |
fbf23849 | 4933 | break; |
2d28390a | 4934 | case IORING_OP_LINK_TIMEOUT: |
3529d8c2 | 4935 | ret = io_timeout_prep(req, sqe, true); |
b7bb4f7d | 4936 | break; |
8ed8d3c3 | 4937 | case IORING_OP_ACCEPT: |
3529d8c2 | 4938 | ret = io_accept_prep(req, sqe); |
8ed8d3c3 | 4939 | break; |
d63d1b5e JA |
4940 | case IORING_OP_FALLOCATE: |
4941 | ret = io_fallocate_prep(req, sqe); | |
4942 | break; | |
15b71abe JA |
4943 | case IORING_OP_OPENAT: |
4944 | ret = io_openat_prep(req, sqe); | |
4945 | break; | |
b5dba59e JA |
4946 | case IORING_OP_CLOSE: |
4947 | ret = io_close_prep(req, sqe); | |
4948 | break; | |
05f3fb3c JA |
4949 | case IORING_OP_FILES_UPDATE: |
4950 | ret = io_files_update_prep(req, sqe); | |
4951 | break; | |
eddc7ef5 JA |
4952 | case IORING_OP_STATX: |
4953 | ret = io_statx_prep(req, sqe); | |
4954 | break; | |
4840e418 JA |
4955 | case IORING_OP_FADVISE: |
4956 | ret = io_fadvise_prep(req, sqe); | |
4957 | break; | |
c1ca757b JA |
4958 | case IORING_OP_MADVISE: |
4959 | ret = io_madvise_prep(req, sqe); | |
4960 | break; | |
cebdb986 JA |
4961 | case IORING_OP_OPENAT2: |
4962 | ret = io_openat2_prep(req, sqe); | |
4963 | break; | |
3e4827b0 JA |
4964 | case IORING_OP_EPOLL_CTL: |
4965 | ret = io_epoll_ctl_prep(req, sqe); | |
4966 | break; | |
7d67af2c PB |
4967 | case IORING_OP_SPLICE: |
4968 | ret = io_splice_prep(req, sqe); | |
4969 | break; | |
ddf0322d JA |
4970 | case IORING_OP_PROVIDE_BUFFERS: |
4971 | ret = io_provide_buffers_prep(req, sqe); | |
4972 | break; | |
067524e9 JA |
4973 | case IORING_OP_REMOVE_BUFFERS: |
4974 | ret = io_remove_buffers_prep(req, sqe); | |
4975 | break; | |
f67676d1 | 4976 | default: |
e781573e JA |
4977 | printk_once(KERN_WARNING "io_uring: unhandled opcode %d\n", |
4978 | req->opcode); | |
4979 | ret = -EINVAL; | |
b7bb4f7d | 4980 | break; |
f67676d1 JA |
4981 | } |
4982 | ||
b7bb4f7d | 4983 | return ret; |
f67676d1 JA |
4984 | } |
4985 | ||
3529d8c2 | 4986 | static int io_req_defer(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
de0617e4 | 4987 | { |
a197f664 | 4988 | struct io_ring_ctx *ctx = req->ctx; |
f67676d1 | 4989 | int ret; |
de0617e4 | 4990 | |
9d858b21 BL |
4991 | /* Still need defer if there is pending req in defer list. */ |
4992 | if (!req_need_defer(req) && list_empty(&ctx->defer_list)) | |
de0617e4 JA |
4993 | return 0; |
4994 | ||
3529d8c2 | 4995 | if (!req->io && io_alloc_async_ctx(req)) |
de0617e4 JA |
4996 | return -EAGAIN; |
4997 | ||
3529d8c2 | 4998 | ret = io_req_defer_prep(req, sqe); |
b7bb4f7d | 4999 | if (ret < 0) |
2d28390a | 5000 | return ret; |
2d28390a | 5001 | |
de0617e4 | 5002 | spin_lock_irq(&ctx->completion_lock); |
9d858b21 | 5003 | if (!req_need_defer(req) && list_empty(&ctx->defer_list)) { |
de0617e4 | 5004 | spin_unlock_irq(&ctx->completion_lock); |
de0617e4 JA |
5005 | return 0; |
5006 | } | |
5007 | ||
915967f6 | 5008 | trace_io_uring_defer(ctx, req, req->user_data); |
de0617e4 JA |
5009 | list_add_tail(&req->list, &ctx->defer_list); |
5010 | spin_unlock_irq(&ctx->completion_lock); | |
5011 | return -EIOCBQUEUED; | |
5012 | } | |
5013 | ||
99bc4c38 PB |
5014 | static void io_cleanup_req(struct io_kiocb *req) |
5015 | { | |
5016 | struct io_async_ctx *io = req->io; | |
5017 | ||
5018 | switch (req->opcode) { | |
5019 | case IORING_OP_READV: | |
5020 | case IORING_OP_READ_FIXED: | |
5021 | case IORING_OP_READ: | |
bcda7baa JA |
5022 | if (req->flags & REQ_F_BUFFER_SELECTED) |
5023 | kfree((void *)(unsigned long)req->rw.addr); | |
5024 | /* fallthrough */ | |
99bc4c38 PB |
5025 | case IORING_OP_WRITEV: |
5026 | case IORING_OP_WRITE_FIXED: | |
5027 | case IORING_OP_WRITE: | |
5028 | if (io->rw.iov != io->rw.fast_iov) | |
5029 | kfree(io->rw.iov); | |
5030 | break; | |
99bc4c38 | 5031 | case IORING_OP_RECVMSG: |
52de1fe1 JA |
5032 | if (req->flags & REQ_F_BUFFER_SELECTED) |
5033 | kfree(req->sr_msg.kbuf); | |
5034 | /* fallthrough */ | |
5035 | case IORING_OP_SENDMSG: | |
99bc4c38 PB |
5036 | if (io->msg.iov != io->msg.fast_iov) |
5037 | kfree(io->msg.iov); | |
5038 | break; | |
bcda7baa JA |
5039 | case IORING_OP_RECV: |
5040 | if (req->flags & REQ_F_BUFFER_SELECTED) | |
5041 | kfree(req->sr_msg.kbuf); | |
5042 | break; | |
8fef80bf PB |
5043 | case IORING_OP_OPENAT: |
5044 | case IORING_OP_OPENAT2: | |
5045 | case IORING_OP_STATX: | |
5046 | putname(req->open.filename); | |
5047 | break; | |
7d67af2c PB |
5048 | case IORING_OP_SPLICE: |
5049 | io_put_file(req, req->splice.file_in, | |
5050 | (req->splice.flags & SPLICE_F_FD_IN_FIXED)); | |
5051 | break; | |
99bc4c38 PB |
5052 | } |
5053 | ||
5054 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
5055 | } | |
5056 | ||
3529d8c2 | 5057 | static int io_issue_sqe(struct io_kiocb *req, const struct io_uring_sqe *sqe, |
014db007 | 5058 | bool force_nonblock) |
2b188cc1 | 5059 | { |
a197f664 | 5060 | struct io_ring_ctx *ctx = req->ctx; |
d625c6ee | 5061 | int ret; |
2b188cc1 | 5062 | |
d625c6ee | 5063 | switch (req->opcode) { |
2b188cc1 | 5064 | case IORING_OP_NOP: |
78e19bbe | 5065 | ret = io_nop(req); |
2b188cc1 JA |
5066 | break; |
5067 | case IORING_OP_READV: | |
edafccee | 5068 | case IORING_OP_READ_FIXED: |
3a6820f2 | 5069 | case IORING_OP_READ: |
3529d8c2 JA |
5070 | if (sqe) { |
5071 | ret = io_read_prep(req, sqe, force_nonblock); | |
5072 | if (ret < 0) | |
5073 | break; | |
5074 | } | |
014db007 | 5075 | ret = io_read(req, force_nonblock); |
edafccee | 5076 | break; |
3529d8c2 | 5077 | case IORING_OP_WRITEV: |
edafccee | 5078 | case IORING_OP_WRITE_FIXED: |
3a6820f2 | 5079 | case IORING_OP_WRITE: |
3529d8c2 JA |
5080 | if (sqe) { |
5081 | ret = io_write_prep(req, sqe, force_nonblock); | |
5082 | if (ret < 0) | |
5083 | break; | |
5084 | } | |
014db007 | 5085 | ret = io_write(req, force_nonblock); |
2b188cc1 | 5086 | break; |
c992fe29 | 5087 | case IORING_OP_FSYNC: |
3529d8c2 JA |
5088 | if (sqe) { |
5089 | ret = io_prep_fsync(req, sqe); | |
5090 | if (ret < 0) | |
5091 | break; | |
5092 | } | |
014db007 | 5093 | ret = io_fsync(req, force_nonblock); |
c992fe29 | 5094 | break; |
221c5eb2 | 5095 | case IORING_OP_POLL_ADD: |
3529d8c2 JA |
5096 | if (sqe) { |
5097 | ret = io_poll_add_prep(req, sqe); | |
5098 | if (ret) | |
5099 | break; | |
5100 | } | |
014db007 | 5101 | ret = io_poll_add(req); |
221c5eb2 JA |
5102 | break; |
5103 | case IORING_OP_POLL_REMOVE: | |
3529d8c2 JA |
5104 | if (sqe) { |
5105 | ret = io_poll_remove_prep(req, sqe); | |
5106 | if (ret < 0) | |
5107 | break; | |
5108 | } | |
fc4df999 | 5109 | ret = io_poll_remove(req); |
221c5eb2 | 5110 | break; |
5d17b4a4 | 5111 | case IORING_OP_SYNC_FILE_RANGE: |
3529d8c2 JA |
5112 | if (sqe) { |
5113 | ret = io_prep_sfr(req, sqe); | |
5114 | if (ret < 0) | |
5115 | break; | |
5116 | } | |
014db007 | 5117 | ret = io_sync_file_range(req, force_nonblock); |
5d17b4a4 | 5118 | break; |
0fa03c62 | 5119 | case IORING_OP_SENDMSG: |
fddaface | 5120 | case IORING_OP_SEND: |
3529d8c2 JA |
5121 | if (sqe) { |
5122 | ret = io_sendmsg_prep(req, sqe); | |
5123 | if (ret < 0) | |
5124 | break; | |
5125 | } | |
fddaface | 5126 | if (req->opcode == IORING_OP_SENDMSG) |
014db007 | 5127 | ret = io_sendmsg(req, force_nonblock); |
fddaface | 5128 | else |
014db007 | 5129 | ret = io_send(req, force_nonblock); |
0fa03c62 | 5130 | break; |
aa1fa28f | 5131 | case IORING_OP_RECVMSG: |
fddaface | 5132 | case IORING_OP_RECV: |
3529d8c2 JA |
5133 | if (sqe) { |
5134 | ret = io_recvmsg_prep(req, sqe); | |
5135 | if (ret) | |
5136 | break; | |
5137 | } | |
fddaface | 5138 | if (req->opcode == IORING_OP_RECVMSG) |
014db007 | 5139 | ret = io_recvmsg(req, force_nonblock); |
fddaface | 5140 | else |
014db007 | 5141 | ret = io_recv(req, force_nonblock); |
aa1fa28f | 5142 | break; |
5262f567 | 5143 | case IORING_OP_TIMEOUT: |
3529d8c2 JA |
5144 | if (sqe) { |
5145 | ret = io_timeout_prep(req, sqe, false); | |
5146 | if (ret) | |
5147 | break; | |
5148 | } | |
fc4df999 | 5149 | ret = io_timeout(req); |
5262f567 | 5150 | break; |
11365043 | 5151 | case IORING_OP_TIMEOUT_REMOVE: |
3529d8c2 JA |
5152 | if (sqe) { |
5153 | ret = io_timeout_remove_prep(req, sqe); | |
5154 | if (ret) | |
5155 | break; | |
5156 | } | |
fc4df999 | 5157 | ret = io_timeout_remove(req); |
11365043 | 5158 | break; |
17f2fe35 | 5159 | case IORING_OP_ACCEPT: |
3529d8c2 JA |
5160 | if (sqe) { |
5161 | ret = io_accept_prep(req, sqe); | |
5162 | if (ret) | |
5163 | break; | |
5164 | } | |
014db007 | 5165 | ret = io_accept(req, force_nonblock); |
17f2fe35 | 5166 | break; |
f8e85cf2 | 5167 | case IORING_OP_CONNECT: |
3529d8c2 JA |
5168 | if (sqe) { |
5169 | ret = io_connect_prep(req, sqe); | |
5170 | if (ret) | |
5171 | break; | |
5172 | } | |
014db007 | 5173 | ret = io_connect(req, force_nonblock); |
f8e85cf2 | 5174 | break; |
62755e35 | 5175 | case IORING_OP_ASYNC_CANCEL: |
3529d8c2 JA |
5176 | if (sqe) { |
5177 | ret = io_async_cancel_prep(req, sqe); | |
5178 | if (ret) | |
5179 | break; | |
5180 | } | |
014db007 | 5181 | ret = io_async_cancel(req); |
62755e35 | 5182 | break; |
d63d1b5e JA |
5183 | case IORING_OP_FALLOCATE: |
5184 | if (sqe) { | |
5185 | ret = io_fallocate_prep(req, sqe); | |
5186 | if (ret) | |
5187 | break; | |
5188 | } | |
014db007 | 5189 | ret = io_fallocate(req, force_nonblock); |
d63d1b5e | 5190 | break; |
15b71abe JA |
5191 | case IORING_OP_OPENAT: |
5192 | if (sqe) { | |
5193 | ret = io_openat_prep(req, sqe); | |
5194 | if (ret) | |
5195 | break; | |
5196 | } | |
014db007 | 5197 | ret = io_openat(req, force_nonblock); |
15b71abe | 5198 | break; |
b5dba59e JA |
5199 | case IORING_OP_CLOSE: |
5200 | if (sqe) { | |
5201 | ret = io_close_prep(req, sqe); | |
5202 | if (ret) | |
5203 | break; | |
5204 | } | |
014db007 | 5205 | ret = io_close(req, force_nonblock); |
b5dba59e | 5206 | break; |
05f3fb3c JA |
5207 | case IORING_OP_FILES_UPDATE: |
5208 | if (sqe) { | |
5209 | ret = io_files_update_prep(req, sqe); | |
5210 | if (ret) | |
5211 | break; | |
5212 | } | |
5213 | ret = io_files_update(req, force_nonblock); | |
5214 | break; | |
eddc7ef5 JA |
5215 | case IORING_OP_STATX: |
5216 | if (sqe) { | |
5217 | ret = io_statx_prep(req, sqe); | |
5218 | if (ret) | |
5219 | break; | |
5220 | } | |
014db007 | 5221 | ret = io_statx(req, force_nonblock); |
eddc7ef5 | 5222 | break; |
4840e418 JA |
5223 | case IORING_OP_FADVISE: |
5224 | if (sqe) { | |
5225 | ret = io_fadvise_prep(req, sqe); | |
5226 | if (ret) | |
5227 | break; | |
5228 | } | |
014db007 | 5229 | ret = io_fadvise(req, force_nonblock); |
4840e418 | 5230 | break; |
c1ca757b JA |
5231 | case IORING_OP_MADVISE: |
5232 | if (sqe) { | |
5233 | ret = io_madvise_prep(req, sqe); | |
5234 | if (ret) | |
5235 | break; | |
5236 | } | |
014db007 | 5237 | ret = io_madvise(req, force_nonblock); |
c1ca757b | 5238 | break; |
cebdb986 JA |
5239 | case IORING_OP_OPENAT2: |
5240 | if (sqe) { | |
5241 | ret = io_openat2_prep(req, sqe); | |
5242 | if (ret) | |
5243 | break; | |
5244 | } | |
014db007 | 5245 | ret = io_openat2(req, force_nonblock); |
cebdb986 | 5246 | break; |
3e4827b0 JA |
5247 | case IORING_OP_EPOLL_CTL: |
5248 | if (sqe) { | |
5249 | ret = io_epoll_ctl_prep(req, sqe); | |
5250 | if (ret) | |
5251 | break; | |
5252 | } | |
014db007 | 5253 | ret = io_epoll_ctl(req, force_nonblock); |
3e4827b0 | 5254 | break; |
7d67af2c PB |
5255 | case IORING_OP_SPLICE: |
5256 | if (sqe) { | |
5257 | ret = io_splice_prep(req, sqe); | |
5258 | if (ret < 0) | |
5259 | break; | |
5260 | } | |
014db007 | 5261 | ret = io_splice(req, force_nonblock); |
7d67af2c | 5262 | break; |
ddf0322d JA |
5263 | case IORING_OP_PROVIDE_BUFFERS: |
5264 | if (sqe) { | |
5265 | ret = io_provide_buffers_prep(req, sqe); | |
5266 | if (ret) | |
5267 | break; | |
5268 | } | |
5269 | ret = io_provide_buffers(req, force_nonblock); | |
5270 | break; | |
067524e9 JA |
5271 | case IORING_OP_REMOVE_BUFFERS: |
5272 | if (sqe) { | |
5273 | ret = io_remove_buffers_prep(req, sqe); | |
5274 | if (ret) | |
5275 | break; | |
5276 | } | |
5277 | ret = io_remove_buffers(req, force_nonblock); | |
3e4827b0 | 5278 | break; |
2b188cc1 JA |
5279 | default: |
5280 | ret = -EINVAL; | |
5281 | break; | |
5282 | } | |
5283 | ||
def596e9 JA |
5284 | if (ret) |
5285 | return ret; | |
5286 | ||
5287 | if (ctx->flags & IORING_SETUP_IOPOLL) { | |
11ba820b JA |
5288 | const bool in_async = io_wq_current_is_worker(); |
5289 | ||
9e645e11 | 5290 | if (req->result == -EAGAIN) |
def596e9 JA |
5291 | return -EAGAIN; |
5292 | ||
11ba820b JA |
5293 | /* workqueue context doesn't hold uring_lock, grab it now */ |
5294 | if (in_async) | |
5295 | mutex_lock(&ctx->uring_lock); | |
5296 | ||
def596e9 | 5297 | io_iopoll_req_issued(req); |
11ba820b JA |
5298 | |
5299 | if (in_async) | |
5300 | mutex_unlock(&ctx->uring_lock); | |
def596e9 JA |
5301 | } |
5302 | ||
5303 | return 0; | |
2b188cc1 JA |
5304 | } |
5305 | ||
561fb04a | 5306 | static void io_wq_submit_work(struct io_wq_work **workptr) |
2b188cc1 | 5307 | { |
561fb04a | 5308 | struct io_wq_work *work = *workptr; |
2b188cc1 | 5309 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); |
561fb04a | 5310 | int ret = 0; |
2b188cc1 | 5311 | |
0c9d5ccd JA |
5312 | /* if NO_CANCEL is set, we must still run the work */ |
5313 | if ((work->flags & (IO_WQ_WORK_CANCEL|IO_WQ_WORK_NO_CANCEL)) == | |
5314 | IO_WQ_WORK_CANCEL) { | |
561fb04a | 5315 | ret = -ECANCELED; |
0c9d5ccd | 5316 | } |
31b51510 | 5317 | |
561fb04a | 5318 | if (!ret) { |
561fb04a | 5319 | do { |
014db007 | 5320 | ret = io_issue_sqe(req, NULL, false); |
561fb04a JA |
5321 | /* |
5322 | * We can get EAGAIN for polled IO even though we're | |
5323 | * forcing a sync submission from here, since we can't | |
5324 | * wait for request slots on the block side. | |
5325 | */ | |
5326 | if (ret != -EAGAIN) | |
5327 | break; | |
5328 | cond_resched(); | |
5329 | } while (1); | |
5330 | } | |
31b51510 | 5331 | |
561fb04a | 5332 | if (ret) { |
4e88d6e7 | 5333 | req_set_fail_links(req); |
78e19bbe | 5334 | io_cqring_add_event(req, ret); |
817869d2 | 5335 | io_put_req(req); |
edafccee | 5336 | } |
2b188cc1 | 5337 | |
e9fd9396 | 5338 | io_steal_work(req, workptr); |
2b188cc1 JA |
5339 | } |
5340 | ||
15b71abe | 5341 | static int io_req_needs_file(struct io_kiocb *req, int fd) |
9e3aa61a | 5342 | { |
d3656344 | 5343 | if (!io_op_defs[req->opcode].needs_file) |
9e3aa61a | 5344 | return 0; |
0b5faf6b | 5345 | if ((fd == -1 || fd == AT_FDCWD) && io_op_defs[req->opcode].fd_non_neg) |
d3656344 JA |
5346 | return 0; |
5347 | return 1; | |
09bb8394 JA |
5348 | } |
5349 | ||
65e19f54 JA |
5350 | static inline struct file *io_file_from_index(struct io_ring_ctx *ctx, |
5351 | int index) | |
5352 | { | |
5353 | struct fixed_file_table *table; | |
5354 | ||
05f3fb3c JA |
5355 | table = &ctx->file_data->table[index >> IORING_FILE_TABLE_SHIFT]; |
5356 | return table->files[index & IORING_FILE_TABLE_MASK];; | |
65e19f54 JA |
5357 | } |
5358 | ||
8da11c19 PB |
5359 | static int io_file_get(struct io_submit_state *state, struct io_kiocb *req, |
5360 | int fd, struct file **out_file, bool fixed) | |
09bb8394 | 5361 | { |
a197f664 | 5362 | struct io_ring_ctx *ctx = req->ctx; |
8da11c19 | 5363 | struct file *file; |
09bb8394 | 5364 | |
8da11c19 | 5365 | if (fixed) { |
05f3fb3c | 5366 | if (unlikely(!ctx->file_data || |
09bb8394 JA |
5367 | (unsigned) fd >= ctx->nr_user_files)) |
5368 | return -EBADF; | |
b7620121 | 5369 | fd = array_index_nospec(fd, ctx->nr_user_files); |
8da11c19 PB |
5370 | file = io_file_from_index(ctx, fd); |
5371 | if (!file) | |
08a45173 | 5372 | return -EBADF; |
05589553 XW |
5373 | req->fixed_file_refs = ctx->file_data->cur_refs; |
5374 | percpu_ref_get(req->fixed_file_refs); | |
09bb8394 | 5375 | } else { |
c826bd7a | 5376 | trace_io_uring_file_get(ctx, fd); |
8da11c19 PB |
5377 | file = __io_file_get(state, fd); |
5378 | if (unlikely(!file)) | |
09bb8394 JA |
5379 | return -EBADF; |
5380 | } | |
5381 | ||
8da11c19 | 5382 | *out_file = file; |
09bb8394 JA |
5383 | return 0; |
5384 | } | |
5385 | ||
8da11c19 PB |
5386 | static int io_req_set_file(struct io_submit_state *state, struct io_kiocb *req, |
5387 | const struct io_uring_sqe *sqe) | |
5388 | { | |
5389 | unsigned flags; | |
5390 | int fd; | |
5391 | bool fixed; | |
5392 | ||
5393 | flags = READ_ONCE(sqe->flags); | |
5394 | fd = READ_ONCE(sqe->fd); | |
5395 | ||
5396 | if (!io_req_needs_file(req, fd)) | |
5397 | return 0; | |
5398 | ||
5399 | fixed = (flags & IOSQE_FIXED_FILE); | |
5400 | if (unlikely(!fixed && req->needs_fixed_file)) | |
5401 | return -EBADF; | |
5402 | ||
5403 | return io_file_get(state, req, fd, &req->file, fixed); | |
5404 | } | |
5405 | ||
a197f664 | 5406 | static int io_grab_files(struct io_kiocb *req) |
fcb323cc JA |
5407 | { |
5408 | int ret = -EBADF; | |
a197f664 | 5409 | struct io_ring_ctx *ctx = req->ctx; |
fcb323cc | 5410 | |
f86cd20c JA |
5411 | if (req->work.files) |
5412 | return 0; | |
b14cca0c | 5413 | if (!ctx->ring_file) |
b5dba59e JA |
5414 | return -EBADF; |
5415 | ||
fcb323cc JA |
5416 | rcu_read_lock(); |
5417 | spin_lock_irq(&ctx->inflight_lock); | |
5418 | /* | |
5419 | * We use the f_ops->flush() handler to ensure that we can flush | |
5420 | * out work accessing these files if the fd is closed. Check if | |
5421 | * the fd has changed since we started down this path, and disallow | |
5422 | * this operation if it has. | |
5423 | */ | |
b14cca0c | 5424 | if (fcheck(ctx->ring_fd) == ctx->ring_file) { |
fcb323cc JA |
5425 | list_add(&req->inflight_entry, &ctx->inflight_list); |
5426 | req->flags |= REQ_F_INFLIGHT; | |
5427 | req->work.files = current->files; | |
5428 | ret = 0; | |
5429 | } | |
5430 | spin_unlock_irq(&ctx->inflight_lock); | |
5431 | rcu_read_unlock(); | |
5432 | ||
5433 | return ret; | |
5434 | } | |
5435 | ||
2665abfd | 5436 | static enum hrtimer_restart io_link_timeout_fn(struct hrtimer *timer) |
2b188cc1 | 5437 | { |
ad8a48ac JA |
5438 | struct io_timeout_data *data = container_of(timer, |
5439 | struct io_timeout_data, timer); | |
5440 | struct io_kiocb *req = data->req; | |
2665abfd JA |
5441 | struct io_ring_ctx *ctx = req->ctx; |
5442 | struct io_kiocb *prev = NULL; | |
5443 | unsigned long flags; | |
2665abfd JA |
5444 | |
5445 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
5446 | ||
5447 | /* | |
5448 | * We don't expect the list to be empty, that will only happen if we | |
5449 | * race with the completion of the linked work. | |
5450 | */ | |
4493233e PB |
5451 | if (!list_empty(&req->link_list)) { |
5452 | prev = list_entry(req->link_list.prev, struct io_kiocb, | |
5453 | link_list); | |
5d960724 | 5454 | if (refcount_inc_not_zero(&prev->refs)) { |
4493233e | 5455 | list_del_init(&req->link_list); |
5d960724 JA |
5456 | prev->flags &= ~REQ_F_LINK_TIMEOUT; |
5457 | } else | |
76a46e06 | 5458 | prev = NULL; |
2665abfd JA |
5459 | } |
5460 | ||
5461 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
5462 | ||
5463 | if (prev) { | |
4e88d6e7 | 5464 | req_set_fail_links(prev); |
014db007 | 5465 | io_async_find_and_cancel(ctx, req, prev->user_data, -ETIME); |
76a46e06 | 5466 | io_put_req(prev); |
47f46768 JA |
5467 | } else { |
5468 | io_cqring_add_event(req, -ETIME); | |
5469 | io_put_req(req); | |
2665abfd | 5470 | } |
2665abfd JA |
5471 | return HRTIMER_NORESTART; |
5472 | } | |
5473 | ||
ad8a48ac | 5474 | static void io_queue_linked_timeout(struct io_kiocb *req) |
2665abfd | 5475 | { |
76a46e06 | 5476 | struct io_ring_ctx *ctx = req->ctx; |
2665abfd | 5477 | |
76a46e06 JA |
5478 | /* |
5479 | * If the list is now empty, then our linked request finished before | |
5480 | * we got a chance to setup the timer | |
5481 | */ | |
5482 | spin_lock_irq(&ctx->completion_lock); | |
4493233e | 5483 | if (!list_empty(&req->link_list)) { |
2d28390a | 5484 | struct io_timeout_data *data = &req->io->timeout; |
94ae5e77 | 5485 | |
ad8a48ac JA |
5486 | data->timer.function = io_link_timeout_fn; |
5487 | hrtimer_start(&data->timer, timespec64_to_ktime(data->ts), | |
5488 | data->mode); | |
2665abfd | 5489 | } |
76a46e06 | 5490 | spin_unlock_irq(&ctx->completion_lock); |
2665abfd | 5491 | |
2665abfd | 5492 | /* drop submission reference */ |
76a46e06 JA |
5493 | io_put_req(req); |
5494 | } | |
2665abfd | 5495 | |
ad8a48ac | 5496 | static struct io_kiocb *io_prep_linked_timeout(struct io_kiocb *req) |
2665abfd JA |
5497 | { |
5498 | struct io_kiocb *nxt; | |
5499 | ||
5500 | if (!(req->flags & REQ_F_LINK)) | |
5501 | return NULL; | |
d7718a9d JA |
5502 | /* for polled retry, if flag is set, we already went through here */ |
5503 | if (req->flags & REQ_F_POLLED) | |
5504 | return NULL; | |
2665abfd | 5505 | |
4493233e PB |
5506 | nxt = list_first_entry_or_null(&req->link_list, struct io_kiocb, |
5507 | link_list); | |
d625c6ee | 5508 | if (!nxt || nxt->opcode != IORING_OP_LINK_TIMEOUT) |
76a46e06 | 5509 | return NULL; |
2665abfd | 5510 | |
76a46e06 | 5511 | req->flags |= REQ_F_LINK_TIMEOUT; |
76a46e06 | 5512 | return nxt; |
2665abfd JA |
5513 | } |
5514 | ||
3529d8c2 | 5515 | static void __io_queue_sqe(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
2b188cc1 | 5516 | { |
4a0a7a18 | 5517 | struct io_kiocb *linked_timeout; |
4bc4494e | 5518 | struct io_kiocb *nxt; |
193155c8 | 5519 | const struct cred *old_creds = NULL; |
e0c5c576 | 5520 | int ret; |
2b188cc1 | 5521 | |
4a0a7a18 JA |
5522 | again: |
5523 | linked_timeout = io_prep_linked_timeout(req); | |
5524 | ||
193155c8 JA |
5525 | if (req->work.creds && req->work.creds != current_cred()) { |
5526 | if (old_creds) | |
5527 | revert_creds(old_creds); | |
5528 | if (old_creds == req->work.creds) | |
5529 | old_creds = NULL; /* restored original creds */ | |
5530 | else | |
5531 | old_creds = override_creds(req->work.creds); | |
5532 | } | |
5533 | ||
014db007 | 5534 | ret = io_issue_sqe(req, sqe, true); |
491381ce JA |
5535 | |
5536 | /* | |
5537 | * We async punt it if the file wasn't marked NOWAIT, or if the file | |
5538 | * doesn't support non-blocking read/write attempts | |
5539 | */ | |
5540 | if (ret == -EAGAIN && (!(req->flags & REQ_F_NOWAIT) || | |
5541 | (req->flags & REQ_F_MUST_PUNT))) { | |
d7718a9d JA |
5542 | if (io_arm_poll_handler(req)) { |
5543 | if (linked_timeout) | |
5544 | io_queue_linked_timeout(linked_timeout); | |
4bc4494e | 5545 | goto exit; |
d7718a9d | 5546 | } |
86a761f8 | 5547 | punt: |
f86cd20c | 5548 | if (io_op_defs[req->opcode].file_table) { |
bbad27b2 PB |
5549 | ret = io_grab_files(req); |
5550 | if (ret) | |
5551 | goto err; | |
2b188cc1 | 5552 | } |
bbad27b2 PB |
5553 | |
5554 | /* | |
5555 | * Queued up for async execution, worker will release | |
5556 | * submit reference when the iocb is actually submitted. | |
5557 | */ | |
5558 | io_queue_async_work(req); | |
4bc4494e | 5559 | goto exit; |
2b188cc1 | 5560 | } |
e65ef56d | 5561 | |
fcb323cc | 5562 | err: |
4bc4494e | 5563 | nxt = NULL; |
76a46e06 | 5564 | /* drop submission reference */ |
2a44f467 | 5565 | io_put_req_find_next(req, &nxt); |
e65ef56d | 5566 | |
f9bd67f6 | 5567 | if (linked_timeout) { |
76a46e06 | 5568 | if (!ret) |
f9bd67f6 | 5569 | io_queue_linked_timeout(linked_timeout); |
76a46e06 | 5570 | else |
f9bd67f6 | 5571 | io_put_req(linked_timeout); |
76a46e06 JA |
5572 | } |
5573 | ||
e65ef56d | 5574 | /* and drop final reference, if we failed */ |
9e645e11 | 5575 | if (ret) { |
78e19bbe | 5576 | io_cqring_add_event(req, ret); |
4e88d6e7 | 5577 | req_set_fail_links(req); |
e65ef56d | 5578 | io_put_req(req); |
9e645e11 | 5579 | } |
4a0a7a18 JA |
5580 | if (nxt) { |
5581 | req = nxt; | |
86a761f8 PB |
5582 | |
5583 | if (req->flags & REQ_F_FORCE_ASYNC) | |
5584 | goto punt; | |
4a0a7a18 JA |
5585 | goto again; |
5586 | } | |
4bc4494e | 5587 | exit: |
193155c8 JA |
5588 | if (old_creds) |
5589 | revert_creds(old_creds); | |
2b188cc1 JA |
5590 | } |
5591 | ||
3529d8c2 | 5592 | static void io_queue_sqe(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
4fe2c963 JL |
5593 | { |
5594 | int ret; | |
5595 | ||
3529d8c2 | 5596 | ret = io_req_defer(req, sqe); |
4fe2c963 JL |
5597 | if (ret) { |
5598 | if (ret != -EIOCBQUEUED) { | |
1118591a | 5599 | fail_req: |
78e19bbe | 5600 | io_cqring_add_event(req, ret); |
4e88d6e7 | 5601 | req_set_fail_links(req); |
78e19bbe | 5602 | io_double_put_req(req); |
4fe2c963 | 5603 | } |
2550878f | 5604 | } else if (req->flags & REQ_F_FORCE_ASYNC) { |
1118591a PB |
5605 | ret = io_req_defer_prep(req, sqe); |
5606 | if (unlikely(ret < 0)) | |
5607 | goto fail_req; | |
ce35a47a JA |
5608 | /* |
5609 | * Never try inline submit of IOSQE_ASYNC is set, go straight | |
5610 | * to async execution. | |
5611 | */ | |
5612 | req->work.flags |= IO_WQ_WORK_CONCURRENT; | |
5613 | io_queue_async_work(req); | |
5614 | } else { | |
3529d8c2 | 5615 | __io_queue_sqe(req, sqe); |
ce35a47a | 5616 | } |
4fe2c963 JL |
5617 | } |
5618 | ||
1b4a51b6 | 5619 | static inline void io_queue_link_head(struct io_kiocb *req) |
4fe2c963 | 5620 | { |
94ae5e77 | 5621 | if (unlikely(req->flags & REQ_F_FAIL_LINK)) { |
1b4a51b6 PB |
5622 | io_cqring_add_event(req, -ECANCELED); |
5623 | io_double_put_req(req); | |
5624 | } else | |
3529d8c2 | 5625 | io_queue_sqe(req, NULL); |
4fe2c963 JL |
5626 | } |
5627 | ||
4e88d6e7 | 5628 | #define SQE_VALID_FLAGS (IOSQE_FIXED_FILE|IOSQE_IO_DRAIN|IOSQE_IO_LINK| \ |
bcda7baa JA |
5629 | IOSQE_IO_HARDLINK | IOSQE_ASYNC | \ |
5630 | IOSQE_BUFFER_SELECT) | |
9e645e11 | 5631 | |
3529d8c2 JA |
5632 | static bool io_submit_sqe(struct io_kiocb *req, const struct io_uring_sqe *sqe, |
5633 | struct io_submit_state *state, struct io_kiocb **link) | |
9e645e11 | 5634 | { |
a197f664 | 5635 | struct io_ring_ctx *ctx = req->ctx; |
32fe525b | 5636 | unsigned int sqe_flags; |
75c6a039 | 5637 | int ret, id; |
9e645e11 | 5638 | |
32fe525b | 5639 | sqe_flags = READ_ONCE(sqe->flags); |
9e645e11 JA |
5640 | |
5641 | /* enforce forwards compatibility on users */ | |
32fe525b | 5642 | if (unlikely(sqe_flags & ~SQE_VALID_FLAGS)) { |
9e645e11 | 5643 | ret = -EINVAL; |
196be95c | 5644 | goto err_req; |
9e645e11 JA |
5645 | } |
5646 | ||
bcda7baa JA |
5647 | if ((sqe_flags & IOSQE_BUFFER_SELECT) && |
5648 | !io_op_defs[req->opcode].buffer_select) { | |
5649 | ret = -EOPNOTSUPP; | |
5650 | goto err_req; | |
5651 | } | |
5652 | ||
75c6a039 JA |
5653 | id = READ_ONCE(sqe->personality); |
5654 | if (id) { | |
193155c8 JA |
5655 | req->work.creds = idr_find(&ctx->personality_idr, id); |
5656 | if (unlikely(!req->work.creds)) { | |
75c6a039 JA |
5657 | ret = -EINVAL; |
5658 | goto err_req; | |
5659 | } | |
193155c8 | 5660 | get_cred(req->work.creds); |
75c6a039 JA |
5661 | } |
5662 | ||
6b47ee6e | 5663 | /* same numerical values with corresponding REQ_F_*, safe to copy */ |
8da11c19 | 5664 | req->flags |= sqe_flags & (IOSQE_IO_DRAIN | IOSQE_IO_HARDLINK | |
bcda7baa JA |
5665 | IOSQE_ASYNC | IOSQE_FIXED_FILE | |
5666 | IOSQE_BUFFER_SELECT); | |
9e645e11 | 5667 | |
3529d8c2 | 5668 | ret = io_req_set_file(state, req, sqe); |
9e645e11 JA |
5669 | if (unlikely(ret)) { |
5670 | err_req: | |
78e19bbe JA |
5671 | io_cqring_add_event(req, ret); |
5672 | io_double_put_req(req); | |
2e6e1fde | 5673 | return false; |
9e645e11 JA |
5674 | } |
5675 | ||
9e645e11 JA |
5676 | /* |
5677 | * If we already have a head request, queue this one for async | |
5678 | * submittal once the head completes. If we don't have a head but | |
5679 | * IOSQE_IO_LINK is set in the sqe, start a new head. This one will be | |
5680 | * submitted sync once the chain is complete. If none of those | |
5681 | * conditions are true (normal request), then just queue it. | |
5682 | */ | |
5683 | if (*link) { | |
9d76377f | 5684 | struct io_kiocb *head = *link; |
4e88d6e7 | 5685 | |
8cdf2193 PB |
5686 | /* |
5687 | * Taking sequential execution of a link, draining both sides | |
5688 | * of the link also fullfils IOSQE_IO_DRAIN semantics for all | |
5689 | * requests in the link. So, it drains the head and the | |
5690 | * next after the link request. The last one is done via | |
5691 | * drain_next flag to persist the effect across calls. | |
5692 | */ | |
711be031 PB |
5693 | if (sqe_flags & IOSQE_IO_DRAIN) { |
5694 | head->flags |= REQ_F_IO_DRAIN; | |
5695 | ctx->drain_next = 1; | |
5696 | } | |
b7bb4f7d | 5697 | if (io_alloc_async_ctx(req)) { |
9e645e11 JA |
5698 | ret = -EAGAIN; |
5699 | goto err_req; | |
5700 | } | |
5701 | ||
3529d8c2 | 5702 | ret = io_req_defer_prep(req, sqe); |
2d28390a | 5703 | if (ret) { |
4e88d6e7 | 5704 | /* fail even hard links since we don't submit */ |
9d76377f | 5705 | head->flags |= REQ_F_FAIL_LINK; |
f67676d1 | 5706 | goto err_req; |
2d28390a | 5707 | } |
9d76377f PB |
5708 | trace_io_uring_link(ctx, req, head); |
5709 | list_add_tail(&req->link_list, &head->link_list); | |
32fe525b PB |
5710 | |
5711 | /* last request of a link, enqueue the link */ | |
5712 | if (!(sqe_flags & (IOSQE_IO_LINK|IOSQE_IO_HARDLINK))) { | |
5713 | io_queue_link_head(head); | |
5714 | *link = NULL; | |
5715 | } | |
9e645e11 | 5716 | } else { |
711be031 PB |
5717 | if (unlikely(ctx->drain_next)) { |
5718 | req->flags |= REQ_F_IO_DRAIN; | |
5719 | req->ctx->drain_next = 0; | |
5720 | } | |
5721 | if (sqe_flags & (IOSQE_IO_LINK|IOSQE_IO_HARDLINK)) { | |
5722 | req->flags |= REQ_F_LINK; | |
711be031 | 5723 | INIT_LIST_HEAD(&req->link_list); |
f1d96a8f PB |
5724 | |
5725 | if (io_alloc_async_ctx(req)) { | |
5726 | ret = -EAGAIN; | |
5727 | goto err_req; | |
5728 | } | |
711be031 PB |
5729 | ret = io_req_defer_prep(req, sqe); |
5730 | if (ret) | |
5731 | req->flags |= REQ_F_FAIL_LINK; | |
5732 | *link = req; | |
5733 | } else { | |
5734 | io_queue_sqe(req, sqe); | |
5735 | } | |
9e645e11 | 5736 | } |
2e6e1fde PB |
5737 | |
5738 | return true; | |
9e645e11 JA |
5739 | } |
5740 | ||
9a56a232 JA |
5741 | /* |
5742 | * Batched submission is done, ensure local IO is flushed out. | |
5743 | */ | |
5744 | static void io_submit_state_end(struct io_submit_state *state) | |
5745 | { | |
5746 | blk_finish_plug(&state->plug); | |
3d6770fb | 5747 | io_file_put(state); |
2579f913 | 5748 | if (state->free_reqs) |
6c8a3134 | 5749 | kmem_cache_free_bulk(req_cachep, state->free_reqs, state->reqs); |
9a56a232 JA |
5750 | } |
5751 | ||
5752 | /* | |
5753 | * Start submission side cache. | |
5754 | */ | |
5755 | static void io_submit_state_start(struct io_submit_state *state, | |
22efde59 | 5756 | unsigned int max_ios) |
9a56a232 JA |
5757 | { |
5758 | blk_start_plug(&state->plug); | |
2579f913 | 5759 | state->free_reqs = 0; |
9a56a232 JA |
5760 | state->file = NULL; |
5761 | state->ios_left = max_ios; | |
5762 | } | |
5763 | ||
2b188cc1 JA |
5764 | static void io_commit_sqring(struct io_ring_ctx *ctx) |
5765 | { | |
75b28aff | 5766 | struct io_rings *rings = ctx->rings; |
2b188cc1 | 5767 | |
caf582c6 PB |
5768 | /* |
5769 | * Ensure any loads from the SQEs are done at this point, | |
5770 | * since once we write the new head, the application could | |
5771 | * write new data to them. | |
5772 | */ | |
5773 | smp_store_release(&rings->sq.head, ctx->cached_sq_head); | |
2b188cc1 JA |
5774 | } |
5775 | ||
2b188cc1 | 5776 | /* |
3529d8c2 | 5777 | * Fetch an sqe, if one is available. Note that sqe_ptr will point to memory |
2b188cc1 JA |
5778 | * that is mapped by userspace. This means that care needs to be taken to |
5779 | * ensure that reads are stable, as we cannot rely on userspace always | |
5780 | * being a good citizen. If members of the sqe are validated and then later | |
5781 | * used, it's important that those reads are done through READ_ONCE() to | |
5782 | * prevent a re-load down the line. | |
5783 | */ | |
3529d8c2 JA |
5784 | static bool io_get_sqring(struct io_ring_ctx *ctx, struct io_kiocb *req, |
5785 | const struct io_uring_sqe **sqe_ptr) | |
2b188cc1 | 5786 | { |
75b28aff | 5787 | u32 *sq_array = ctx->sq_array; |
2b188cc1 JA |
5788 | unsigned head; |
5789 | ||
5790 | /* | |
5791 | * The cached sq head (or cq tail) serves two purposes: | |
5792 | * | |
5793 | * 1) allows us to batch the cost of updating the user visible | |
5794 | * head updates. | |
5795 | * 2) allows the kernel side to track the head on its own, even | |
5796 | * though the application is the one updating it. | |
5797 | */ | |
ee7d46d9 | 5798 | head = READ_ONCE(sq_array[ctx->cached_sq_head & ctx->sq_mask]); |
9835d6fa | 5799 | if (likely(head < ctx->sq_entries)) { |
cf6fd4bd PB |
5800 | /* |
5801 | * All io need record the previous position, if LINK vs DARIN, | |
5802 | * it can be used to mark the position of the first IO in the | |
5803 | * link list. | |
5804 | */ | |
5805 | req->sequence = ctx->cached_sq_head; | |
3529d8c2 JA |
5806 | *sqe_ptr = &ctx->sq_sqes[head]; |
5807 | req->opcode = READ_ONCE((*sqe_ptr)->opcode); | |
5808 | req->user_data = READ_ONCE((*sqe_ptr)->user_data); | |
2b188cc1 JA |
5809 | ctx->cached_sq_head++; |
5810 | return true; | |
5811 | } | |
5812 | ||
5813 | /* drop invalid entries */ | |
5814 | ctx->cached_sq_head++; | |
498ccd9e | 5815 | ctx->cached_sq_dropped++; |
ee7d46d9 | 5816 | WRITE_ONCE(ctx->rings->sq_dropped, ctx->cached_sq_dropped); |
2b188cc1 JA |
5817 | return false; |
5818 | } | |
5819 | ||
fb5ccc98 | 5820 | static int io_submit_sqes(struct io_ring_ctx *ctx, unsigned int nr, |
ae9428ca PB |
5821 | struct file *ring_file, int ring_fd, |
5822 | struct mm_struct **mm, bool async) | |
6c271ce2 JA |
5823 | { |
5824 | struct io_submit_state state, *statep = NULL; | |
9e645e11 | 5825 | struct io_kiocb *link = NULL; |
9e645e11 | 5826 | int i, submitted = 0; |
95a1b3ff | 5827 | bool mm_fault = false; |
6c271ce2 | 5828 | |
c4a2ed72 | 5829 | /* if we have a backlog and couldn't flush it all, return BUSY */ |
ad3eb2c8 JA |
5830 | if (test_bit(0, &ctx->sq_check_overflow)) { |
5831 | if (!list_empty(&ctx->cq_overflow_list) && | |
5832 | !io_cqring_overflow_flush(ctx, false)) | |
5833 | return -EBUSY; | |
5834 | } | |
6c271ce2 | 5835 | |
ee7d46d9 PB |
5836 | /* make sure SQ entry isn't read before tail */ |
5837 | nr = min3(nr, ctx->sq_entries, io_sqring_entries(ctx)); | |
9ef4f124 | 5838 | |
2b85edfc PB |
5839 | if (!percpu_ref_tryget_many(&ctx->refs, nr)) |
5840 | return -EAGAIN; | |
6c271ce2 JA |
5841 | |
5842 | if (nr > IO_PLUG_THRESHOLD) { | |
22efde59 | 5843 | io_submit_state_start(&state, nr); |
6c271ce2 JA |
5844 | statep = &state; |
5845 | } | |
5846 | ||
b14cca0c PB |
5847 | ctx->ring_fd = ring_fd; |
5848 | ctx->ring_file = ring_file; | |
5849 | ||
6c271ce2 | 5850 | for (i = 0; i < nr; i++) { |
3529d8c2 | 5851 | const struct io_uring_sqe *sqe; |
196be95c | 5852 | struct io_kiocb *req; |
1cb1edb2 | 5853 | int err; |
fb5ccc98 | 5854 | |
196be95c PB |
5855 | req = io_get_req(ctx, statep); |
5856 | if (unlikely(!req)) { | |
5857 | if (!submitted) | |
5858 | submitted = -EAGAIN; | |
fb5ccc98 | 5859 | break; |
196be95c | 5860 | } |
3529d8c2 | 5861 | if (!io_get_sqring(ctx, req, &sqe)) { |
2b85edfc | 5862 | __io_req_do_free(req); |
196be95c PB |
5863 | break; |
5864 | } | |
fb5ccc98 | 5865 | |
d3656344 JA |
5866 | /* will complete beyond this point, count as submitted */ |
5867 | submitted++; | |
5868 | ||
5869 | if (unlikely(req->opcode >= IORING_OP_LAST)) { | |
1cb1edb2 PB |
5870 | err = -EINVAL; |
5871 | fail_req: | |
5872 | io_cqring_add_event(req, err); | |
d3656344 | 5873 | io_double_put_req(req); |
196be95c PB |
5874 | break; |
5875 | } | |
fb5ccc98 | 5876 | |
d3656344 | 5877 | if (io_op_defs[req->opcode].needs_mm && !*mm) { |
95a1b3ff | 5878 | mm_fault = mm_fault || !mmget_not_zero(ctx->sqo_mm); |
1cb1edb2 PB |
5879 | if (unlikely(mm_fault)) { |
5880 | err = -EFAULT; | |
5881 | goto fail_req; | |
95a1b3ff | 5882 | } |
1cb1edb2 PB |
5883 | use_mm(ctx->sqo_mm); |
5884 | *mm = ctx->sqo_mm; | |
9e645e11 | 5885 | } |
9e645e11 | 5886 | |
cf6fd4bd | 5887 | req->needs_fixed_file = async; |
354420f7 JA |
5888 | trace_io_uring_submit_sqe(ctx, req->opcode, req->user_data, |
5889 | true, async); | |
3529d8c2 | 5890 | if (!io_submit_sqe(req, sqe, statep, &link)) |
2e6e1fde | 5891 | break; |
6c271ce2 JA |
5892 | } |
5893 | ||
9466f437 PB |
5894 | if (unlikely(submitted != nr)) { |
5895 | int ref_used = (submitted == -EAGAIN) ? 0 : submitted; | |
5896 | ||
5897 | percpu_ref_put_many(&ctx->refs, nr - ref_used); | |
5898 | } | |
9e645e11 | 5899 | if (link) |
1b4a51b6 | 5900 | io_queue_link_head(link); |
6c271ce2 JA |
5901 | if (statep) |
5902 | io_submit_state_end(&state); | |
5903 | ||
ae9428ca PB |
5904 | /* Commit SQ ring head once we've consumed and submitted all SQEs */ |
5905 | io_commit_sqring(ctx); | |
5906 | ||
6c271ce2 JA |
5907 | return submitted; |
5908 | } | |
5909 | ||
5910 | static int io_sq_thread(void *data) | |
5911 | { | |
6c271ce2 JA |
5912 | struct io_ring_ctx *ctx = data; |
5913 | struct mm_struct *cur_mm = NULL; | |
181e448d | 5914 | const struct cred *old_cred; |
6c271ce2 JA |
5915 | mm_segment_t old_fs; |
5916 | DEFINE_WAIT(wait); | |
6c271ce2 | 5917 | unsigned long timeout; |
bdcd3eab | 5918 | int ret = 0; |
6c271ce2 | 5919 | |
206aefde | 5920 | complete(&ctx->completions[1]); |
a4c0b3de | 5921 | |
6c271ce2 JA |
5922 | old_fs = get_fs(); |
5923 | set_fs(USER_DS); | |
181e448d | 5924 | old_cred = override_creds(ctx->creds); |
6c271ce2 | 5925 | |
bdcd3eab | 5926 | timeout = jiffies + ctx->sq_thread_idle; |
2bbcd6d3 | 5927 | while (!kthread_should_park()) { |
fb5ccc98 | 5928 | unsigned int to_submit; |
6c271ce2 | 5929 | |
bdcd3eab | 5930 | if (!list_empty(&ctx->poll_list)) { |
6c271ce2 JA |
5931 | unsigned nr_events = 0; |
5932 | ||
bdcd3eab XW |
5933 | mutex_lock(&ctx->uring_lock); |
5934 | if (!list_empty(&ctx->poll_list)) | |
5935 | io_iopoll_getevents(ctx, &nr_events, 0); | |
5936 | else | |
6c271ce2 | 5937 | timeout = jiffies + ctx->sq_thread_idle; |
bdcd3eab | 5938 | mutex_unlock(&ctx->uring_lock); |
6c271ce2 JA |
5939 | } |
5940 | ||
fb5ccc98 | 5941 | to_submit = io_sqring_entries(ctx); |
c1edbf5f JA |
5942 | |
5943 | /* | |
5944 | * If submit got -EBUSY, flag us as needing the application | |
5945 | * to enter the kernel to reap and flush events. | |
5946 | */ | |
5947 | if (!to_submit || ret == -EBUSY) { | |
7143b5ac SG |
5948 | /* |
5949 | * Drop cur_mm before scheduling, we can't hold it for | |
5950 | * long periods (or over schedule()). Do this before | |
5951 | * adding ourselves to the waitqueue, as the unuse/drop | |
5952 | * may sleep. | |
5953 | */ | |
5954 | if (cur_mm) { | |
5955 | unuse_mm(cur_mm); | |
5956 | mmput(cur_mm); | |
5957 | cur_mm = NULL; | |
5958 | } | |
5959 | ||
6c271ce2 JA |
5960 | /* |
5961 | * We're polling. If we're within the defined idle | |
5962 | * period, then let us spin without work before going | |
c1edbf5f JA |
5963 | * to sleep. The exception is if we got EBUSY doing |
5964 | * more IO, we should wait for the application to | |
5965 | * reap events and wake us up. | |
6c271ce2 | 5966 | */ |
bdcd3eab | 5967 | if (!list_empty(&ctx->poll_list) || |
df069d80 JA |
5968 | (!time_after(jiffies, timeout) && ret != -EBUSY && |
5969 | !percpu_ref_is_dying(&ctx->refs))) { | |
b41e9852 JA |
5970 | if (current->task_works) |
5971 | task_work_run(); | |
9831a90c | 5972 | cond_resched(); |
6c271ce2 JA |
5973 | continue; |
5974 | } | |
5975 | ||
6c271ce2 JA |
5976 | prepare_to_wait(&ctx->sqo_wait, &wait, |
5977 | TASK_INTERRUPTIBLE); | |
5978 | ||
bdcd3eab XW |
5979 | /* |
5980 | * While doing polled IO, before going to sleep, we need | |
5981 | * to check if there are new reqs added to poll_list, it | |
5982 | * is because reqs may have been punted to io worker and | |
5983 | * will be added to poll_list later, hence check the | |
5984 | * poll_list again. | |
5985 | */ | |
5986 | if ((ctx->flags & IORING_SETUP_IOPOLL) && | |
5987 | !list_empty_careful(&ctx->poll_list)) { | |
5988 | finish_wait(&ctx->sqo_wait, &wait); | |
5989 | continue; | |
5990 | } | |
5991 | ||
6c271ce2 | 5992 | /* Tell userspace we may need a wakeup call */ |
75b28aff | 5993 | ctx->rings->sq_flags |= IORING_SQ_NEED_WAKEUP; |
0d7bae69 SB |
5994 | /* make sure to read SQ tail after writing flags */ |
5995 | smp_mb(); | |
6c271ce2 | 5996 | |
fb5ccc98 | 5997 | to_submit = io_sqring_entries(ctx); |
c1edbf5f | 5998 | if (!to_submit || ret == -EBUSY) { |
2bbcd6d3 | 5999 | if (kthread_should_park()) { |
6c271ce2 JA |
6000 | finish_wait(&ctx->sqo_wait, &wait); |
6001 | break; | |
6002 | } | |
b41e9852 JA |
6003 | if (current->task_works) { |
6004 | task_work_run(); | |
10bea96d | 6005 | finish_wait(&ctx->sqo_wait, &wait); |
b41e9852 JA |
6006 | continue; |
6007 | } | |
6c271ce2 JA |
6008 | if (signal_pending(current)) |
6009 | flush_signals(current); | |
6010 | schedule(); | |
6011 | finish_wait(&ctx->sqo_wait, &wait); | |
6012 | ||
75b28aff | 6013 | ctx->rings->sq_flags &= ~IORING_SQ_NEED_WAKEUP; |
6c271ce2 JA |
6014 | continue; |
6015 | } | |
6016 | finish_wait(&ctx->sqo_wait, &wait); | |
6017 | ||
75b28aff | 6018 | ctx->rings->sq_flags &= ~IORING_SQ_NEED_WAKEUP; |
6c271ce2 JA |
6019 | } |
6020 | ||
8a4955ff | 6021 | mutex_lock(&ctx->uring_lock); |
1d7bb1d5 | 6022 | ret = io_submit_sqes(ctx, to_submit, NULL, -1, &cur_mm, true); |
8a4955ff | 6023 | mutex_unlock(&ctx->uring_lock); |
bdcd3eab | 6024 | timeout = jiffies + ctx->sq_thread_idle; |
6c271ce2 JA |
6025 | } |
6026 | ||
b41e9852 JA |
6027 | if (current->task_works) |
6028 | task_work_run(); | |
6029 | ||
6c271ce2 JA |
6030 | set_fs(old_fs); |
6031 | if (cur_mm) { | |
6032 | unuse_mm(cur_mm); | |
6033 | mmput(cur_mm); | |
6034 | } | |
181e448d | 6035 | revert_creds(old_cred); |
06058632 | 6036 | |
2bbcd6d3 | 6037 | kthread_parkme(); |
06058632 | 6038 | |
6c271ce2 JA |
6039 | return 0; |
6040 | } | |
6041 | ||
bda52162 JA |
6042 | struct io_wait_queue { |
6043 | struct wait_queue_entry wq; | |
6044 | struct io_ring_ctx *ctx; | |
6045 | unsigned to_wait; | |
6046 | unsigned nr_timeouts; | |
6047 | }; | |
6048 | ||
1d7bb1d5 | 6049 | static inline bool io_should_wake(struct io_wait_queue *iowq, bool noflush) |
bda52162 JA |
6050 | { |
6051 | struct io_ring_ctx *ctx = iowq->ctx; | |
6052 | ||
6053 | /* | |
d195a66e | 6054 | * Wake up if we have enough events, or if a timeout occurred since we |
bda52162 JA |
6055 | * started waiting. For timeouts, we always want to return to userspace, |
6056 | * regardless of event count. | |
6057 | */ | |
1d7bb1d5 | 6058 | return io_cqring_events(ctx, noflush) >= iowq->to_wait || |
bda52162 JA |
6059 | atomic_read(&ctx->cq_timeouts) != iowq->nr_timeouts; |
6060 | } | |
6061 | ||
6062 | static int io_wake_function(struct wait_queue_entry *curr, unsigned int mode, | |
6063 | int wake_flags, void *key) | |
6064 | { | |
6065 | struct io_wait_queue *iowq = container_of(curr, struct io_wait_queue, | |
6066 | wq); | |
6067 | ||
1d7bb1d5 JA |
6068 | /* use noflush == true, as we can't safely rely on locking context */ |
6069 | if (!io_should_wake(iowq, true)) | |
bda52162 JA |
6070 | return -1; |
6071 | ||
6072 | return autoremove_wake_function(curr, mode, wake_flags, key); | |
6073 | } | |
6074 | ||
2b188cc1 JA |
6075 | /* |
6076 | * Wait until events become available, if we don't already have some. The | |
6077 | * application must reap them itself, as they reside on the shared cq ring. | |
6078 | */ | |
6079 | static int io_cqring_wait(struct io_ring_ctx *ctx, int min_events, | |
6080 | const sigset_t __user *sig, size_t sigsz) | |
6081 | { | |
bda52162 JA |
6082 | struct io_wait_queue iowq = { |
6083 | .wq = { | |
6084 | .private = current, | |
6085 | .func = io_wake_function, | |
6086 | .entry = LIST_HEAD_INIT(iowq.wq.entry), | |
6087 | }, | |
6088 | .ctx = ctx, | |
6089 | .to_wait = min_events, | |
6090 | }; | |
75b28aff | 6091 | struct io_rings *rings = ctx->rings; |
e9ffa5c2 | 6092 | int ret = 0; |
2b188cc1 | 6093 | |
b41e9852 JA |
6094 | do { |
6095 | if (io_cqring_events(ctx, false) >= min_events) | |
6096 | return 0; | |
6097 | if (!current->task_works) | |
6098 | break; | |
6099 | task_work_run(); | |
6100 | } while (1); | |
2b188cc1 JA |
6101 | |
6102 | if (sig) { | |
9e75ad5d AB |
6103 | #ifdef CONFIG_COMPAT |
6104 | if (in_compat_syscall()) | |
6105 | ret = set_compat_user_sigmask((const compat_sigset_t __user *)sig, | |
b772434b | 6106 | sigsz); |
9e75ad5d AB |
6107 | else |
6108 | #endif | |
b772434b | 6109 | ret = set_user_sigmask(sig, sigsz); |
9e75ad5d | 6110 | |
2b188cc1 JA |
6111 | if (ret) |
6112 | return ret; | |
6113 | } | |
6114 | ||
bda52162 | 6115 | iowq.nr_timeouts = atomic_read(&ctx->cq_timeouts); |
c826bd7a | 6116 | trace_io_uring_cqring_wait(ctx, min_events); |
bda52162 JA |
6117 | do { |
6118 | prepare_to_wait_exclusive(&ctx->wait, &iowq.wq, | |
6119 | TASK_INTERRUPTIBLE); | |
b41e9852 JA |
6120 | if (current->task_works) |
6121 | task_work_run(); | |
1d7bb1d5 | 6122 | if (io_should_wake(&iowq, false)) |
bda52162 JA |
6123 | break; |
6124 | schedule(); | |
6125 | if (signal_pending(current)) { | |
e9ffa5c2 | 6126 | ret = -EINTR; |
bda52162 JA |
6127 | break; |
6128 | } | |
6129 | } while (1); | |
6130 | finish_wait(&ctx->wait, &iowq.wq); | |
6131 | ||
e9ffa5c2 | 6132 | restore_saved_sigmask_unless(ret == -EINTR); |
2b188cc1 | 6133 | |
75b28aff | 6134 | return READ_ONCE(rings->cq.head) == READ_ONCE(rings->cq.tail) ? ret : 0; |
2b188cc1 JA |
6135 | } |
6136 | ||
6b06314c JA |
6137 | static void __io_sqe_files_unregister(struct io_ring_ctx *ctx) |
6138 | { | |
6139 | #if defined(CONFIG_UNIX) | |
6140 | if (ctx->ring_sock) { | |
6141 | struct sock *sock = ctx->ring_sock->sk; | |
6142 | struct sk_buff *skb; | |
6143 | ||
6144 | while ((skb = skb_dequeue(&sock->sk_receive_queue)) != NULL) | |
6145 | kfree_skb(skb); | |
6146 | } | |
6147 | #else | |
6148 | int i; | |
6149 | ||
65e19f54 JA |
6150 | for (i = 0; i < ctx->nr_user_files; i++) { |
6151 | struct file *file; | |
6152 | ||
6153 | file = io_file_from_index(ctx, i); | |
6154 | if (file) | |
6155 | fput(file); | |
6156 | } | |
6b06314c JA |
6157 | #endif |
6158 | } | |
6159 | ||
05f3fb3c JA |
6160 | static void io_file_ref_kill(struct percpu_ref *ref) |
6161 | { | |
6162 | struct fixed_file_data *data; | |
6163 | ||
6164 | data = container_of(ref, struct fixed_file_data, refs); | |
6165 | complete(&data->done); | |
6166 | } | |
6167 | ||
6b06314c JA |
6168 | static int io_sqe_files_unregister(struct io_ring_ctx *ctx) |
6169 | { | |
05f3fb3c | 6170 | struct fixed_file_data *data = ctx->file_data; |
05589553 | 6171 | struct fixed_file_ref_node *ref_node = NULL; |
65e19f54 | 6172 | unsigned nr_tables, i; |
05589553 | 6173 | unsigned long flags; |
65e19f54 | 6174 | |
05f3fb3c | 6175 | if (!data) |
6b06314c JA |
6176 | return -ENXIO; |
6177 | ||
05589553 XW |
6178 | spin_lock_irqsave(&data->lock, flags); |
6179 | if (!list_empty(&data->ref_list)) | |
6180 | ref_node = list_first_entry(&data->ref_list, | |
6181 | struct fixed_file_ref_node, node); | |
6182 | spin_unlock_irqrestore(&data->lock, flags); | |
6183 | if (ref_node) | |
6184 | percpu_ref_kill(&ref_node->refs); | |
6185 | ||
6186 | percpu_ref_kill(&data->refs); | |
6187 | ||
6188 | /* wait for all refs nodes to complete */ | |
2faf852d | 6189 | wait_for_completion(&data->done); |
05f3fb3c | 6190 | |
6b06314c | 6191 | __io_sqe_files_unregister(ctx); |
65e19f54 JA |
6192 | nr_tables = DIV_ROUND_UP(ctx->nr_user_files, IORING_MAX_FILES_TABLE); |
6193 | for (i = 0; i < nr_tables; i++) | |
05f3fb3c JA |
6194 | kfree(data->table[i].files); |
6195 | kfree(data->table); | |
05589553 XW |
6196 | percpu_ref_exit(&data->refs); |
6197 | kfree(data); | |
05f3fb3c | 6198 | ctx->file_data = NULL; |
6b06314c JA |
6199 | ctx->nr_user_files = 0; |
6200 | return 0; | |
6201 | } | |
6202 | ||
6c271ce2 JA |
6203 | static void io_sq_thread_stop(struct io_ring_ctx *ctx) |
6204 | { | |
6205 | if (ctx->sqo_thread) { | |
206aefde | 6206 | wait_for_completion(&ctx->completions[1]); |
2bbcd6d3 RP |
6207 | /* |
6208 | * The park is a bit of a work-around, without it we get | |
6209 | * warning spews on shutdown with SQPOLL set and affinity | |
6210 | * set to a single CPU. | |
6211 | */ | |
06058632 | 6212 | kthread_park(ctx->sqo_thread); |
6c271ce2 JA |
6213 | kthread_stop(ctx->sqo_thread); |
6214 | ctx->sqo_thread = NULL; | |
6215 | } | |
6216 | } | |
6217 | ||
6b06314c JA |
6218 | static void io_finish_async(struct io_ring_ctx *ctx) |
6219 | { | |
6c271ce2 JA |
6220 | io_sq_thread_stop(ctx); |
6221 | ||
561fb04a JA |
6222 | if (ctx->io_wq) { |
6223 | io_wq_destroy(ctx->io_wq); | |
6224 | ctx->io_wq = NULL; | |
6b06314c JA |
6225 | } |
6226 | } | |
6227 | ||
6228 | #if defined(CONFIG_UNIX) | |
6b06314c JA |
6229 | /* |
6230 | * Ensure the UNIX gc is aware of our file set, so we are certain that | |
6231 | * the io_uring can be safely unregistered on process exit, even if we have | |
6232 | * loops in the file referencing. | |
6233 | */ | |
6234 | static int __io_sqe_files_scm(struct io_ring_ctx *ctx, int nr, int offset) | |
6235 | { | |
6236 | struct sock *sk = ctx->ring_sock->sk; | |
6237 | struct scm_fp_list *fpl; | |
6238 | struct sk_buff *skb; | |
08a45173 | 6239 | int i, nr_files; |
6b06314c | 6240 | |
6b06314c JA |
6241 | fpl = kzalloc(sizeof(*fpl), GFP_KERNEL); |
6242 | if (!fpl) | |
6243 | return -ENOMEM; | |
6244 | ||
6245 | skb = alloc_skb(0, GFP_KERNEL); | |
6246 | if (!skb) { | |
6247 | kfree(fpl); | |
6248 | return -ENOMEM; | |
6249 | } | |
6250 | ||
6251 | skb->sk = sk; | |
6b06314c | 6252 | |
08a45173 | 6253 | nr_files = 0; |
6b06314c JA |
6254 | fpl->user = get_uid(ctx->user); |
6255 | for (i = 0; i < nr; i++) { | |
65e19f54 JA |
6256 | struct file *file = io_file_from_index(ctx, i + offset); |
6257 | ||
6258 | if (!file) | |
08a45173 | 6259 | continue; |
65e19f54 | 6260 | fpl->fp[nr_files] = get_file(file); |
08a45173 JA |
6261 | unix_inflight(fpl->user, fpl->fp[nr_files]); |
6262 | nr_files++; | |
6b06314c JA |
6263 | } |
6264 | ||
08a45173 JA |
6265 | if (nr_files) { |
6266 | fpl->max = SCM_MAX_FD; | |
6267 | fpl->count = nr_files; | |
6268 | UNIXCB(skb).fp = fpl; | |
05f3fb3c | 6269 | skb->destructor = unix_destruct_scm; |
08a45173 JA |
6270 | refcount_add(skb->truesize, &sk->sk_wmem_alloc); |
6271 | skb_queue_head(&sk->sk_receive_queue, skb); | |
6b06314c | 6272 | |
08a45173 JA |
6273 | for (i = 0; i < nr_files; i++) |
6274 | fput(fpl->fp[i]); | |
6275 | } else { | |
6276 | kfree_skb(skb); | |
6277 | kfree(fpl); | |
6278 | } | |
6b06314c JA |
6279 | |
6280 | return 0; | |
6281 | } | |
6282 | ||
6283 | /* | |
6284 | * If UNIX sockets are enabled, fd passing can cause a reference cycle which | |
6285 | * causes regular reference counting to break down. We rely on the UNIX | |
6286 | * garbage collection to take care of this problem for us. | |
6287 | */ | |
6288 | static int io_sqe_files_scm(struct io_ring_ctx *ctx) | |
6289 | { | |
6290 | unsigned left, total; | |
6291 | int ret = 0; | |
6292 | ||
6293 | total = 0; | |
6294 | left = ctx->nr_user_files; | |
6295 | while (left) { | |
6296 | unsigned this_files = min_t(unsigned, left, SCM_MAX_FD); | |
6b06314c JA |
6297 | |
6298 | ret = __io_sqe_files_scm(ctx, this_files, total); | |
6299 | if (ret) | |
6300 | break; | |
6301 | left -= this_files; | |
6302 | total += this_files; | |
6303 | } | |
6304 | ||
6305 | if (!ret) | |
6306 | return 0; | |
6307 | ||
6308 | while (total < ctx->nr_user_files) { | |
65e19f54 JA |
6309 | struct file *file = io_file_from_index(ctx, total); |
6310 | ||
6311 | if (file) | |
6312 | fput(file); | |
6b06314c JA |
6313 | total++; |
6314 | } | |
6315 | ||
6316 | return ret; | |
6317 | } | |
6318 | #else | |
6319 | static int io_sqe_files_scm(struct io_ring_ctx *ctx) | |
6320 | { | |
6321 | return 0; | |
6322 | } | |
6323 | #endif | |
6324 | ||
65e19f54 JA |
6325 | static int io_sqe_alloc_file_tables(struct io_ring_ctx *ctx, unsigned nr_tables, |
6326 | unsigned nr_files) | |
6327 | { | |
6328 | int i; | |
6329 | ||
6330 | for (i = 0; i < nr_tables; i++) { | |
05f3fb3c | 6331 | struct fixed_file_table *table = &ctx->file_data->table[i]; |
65e19f54 JA |
6332 | unsigned this_files; |
6333 | ||
6334 | this_files = min(nr_files, IORING_MAX_FILES_TABLE); | |
6335 | table->files = kcalloc(this_files, sizeof(struct file *), | |
6336 | GFP_KERNEL); | |
6337 | if (!table->files) | |
6338 | break; | |
6339 | nr_files -= this_files; | |
6340 | } | |
6341 | ||
6342 | if (i == nr_tables) | |
6343 | return 0; | |
6344 | ||
6345 | for (i = 0; i < nr_tables; i++) { | |
05f3fb3c | 6346 | struct fixed_file_table *table = &ctx->file_data->table[i]; |
65e19f54 JA |
6347 | kfree(table->files); |
6348 | } | |
6349 | return 1; | |
6350 | } | |
6351 | ||
05f3fb3c JA |
6352 | static void io_ring_file_put(struct io_ring_ctx *ctx, struct file *file) |
6353 | { | |
6354 | #if defined(CONFIG_UNIX) | |
6355 | struct sock *sock = ctx->ring_sock->sk; | |
6356 | struct sk_buff_head list, *head = &sock->sk_receive_queue; | |
6357 | struct sk_buff *skb; | |
6358 | int i; | |
6359 | ||
6360 | __skb_queue_head_init(&list); | |
6361 | ||
6362 | /* | |
6363 | * Find the skb that holds this file in its SCM_RIGHTS. When found, | |
6364 | * remove this entry and rearrange the file array. | |
6365 | */ | |
6366 | skb = skb_dequeue(head); | |
6367 | while (skb) { | |
6368 | struct scm_fp_list *fp; | |
6369 | ||
6370 | fp = UNIXCB(skb).fp; | |
6371 | for (i = 0; i < fp->count; i++) { | |
6372 | int left; | |
6373 | ||
6374 | if (fp->fp[i] != file) | |
6375 | continue; | |
6376 | ||
6377 | unix_notinflight(fp->user, fp->fp[i]); | |
6378 | left = fp->count - 1 - i; | |
6379 | if (left) { | |
6380 | memmove(&fp->fp[i], &fp->fp[i + 1], | |
6381 | left * sizeof(struct file *)); | |
6382 | } | |
6383 | fp->count--; | |
6384 | if (!fp->count) { | |
6385 | kfree_skb(skb); | |
6386 | skb = NULL; | |
6387 | } else { | |
6388 | __skb_queue_tail(&list, skb); | |
6389 | } | |
6390 | fput(file); | |
6391 | file = NULL; | |
6392 | break; | |
6393 | } | |
6394 | ||
6395 | if (!file) | |
6396 | break; | |
6397 | ||
6398 | __skb_queue_tail(&list, skb); | |
6399 | ||
6400 | skb = skb_dequeue(head); | |
6401 | } | |
6402 | ||
6403 | if (skb_peek(&list)) { | |
6404 | spin_lock_irq(&head->lock); | |
6405 | while ((skb = __skb_dequeue(&list)) != NULL) | |
6406 | __skb_queue_tail(head, skb); | |
6407 | spin_unlock_irq(&head->lock); | |
6408 | } | |
6409 | #else | |
6410 | fput(file); | |
6411 | #endif | |
6412 | } | |
6413 | ||
6414 | struct io_file_put { | |
05589553 | 6415 | struct list_head list; |
05f3fb3c | 6416 | struct file *file; |
05f3fb3c JA |
6417 | }; |
6418 | ||
05589553 | 6419 | static void io_file_put_work(struct work_struct *work) |
65e19f54 | 6420 | { |
05589553 XW |
6421 | struct fixed_file_ref_node *ref_node; |
6422 | struct fixed_file_data *file_data; | |
6423 | struct io_ring_ctx *ctx; | |
05f3fb3c | 6424 | struct io_file_put *pfile, *tmp; |
05589553 | 6425 | unsigned long flags; |
65e19f54 | 6426 | |
05589553 XW |
6427 | ref_node = container_of(work, struct fixed_file_ref_node, work); |
6428 | file_data = ref_node->file_data; | |
6429 | ctx = file_data->ctx; | |
6430 | ||
6431 | list_for_each_entry_safe(pfile, tmp, &ref_node->file_list, list) { | |
6432 | list_del_init(&pfile->list); | |
6433 | io_ring_file_put(ctx, pfile->file); | |
6434 | kfree(pfile); | |
65e19f54 | 6435 | } |
05589553 XW |
6436 | |
6437 | spin_lock_irqsave(&file_data->lock, flags); | |
6438 | list_del_init(&ref_node->node); | |
6439 | spin_unlock_irqrestore(&file_data->lock, flags); | |
6440 | ||
6441 | percpu_ref_exit(&ref_node->refs); | |
6442 | kfree(ref_node); | |
6443 | percpu_ref_put(&file_data->refs); | |
2faf852d | 6444 | } |
65e19f54 | 6445 | |
05589553 | 6446 | static void io_file_data_ref_zero(struct percpu_ref *ref) |
2faf852d | 6447 | { |
05589553 | 6448 | struct fixed_file_ref_node *ref_node; |
65e19f54 | 6449 | |
05589553 XW |
6450 | ref_node = container_of(ref, struct fixed_file_ref_node, refs); |
6451 | ||
6452 | queue_work(system_wq, &ref_node->work); | |
05f3fb3c | 6453 | } |
65e19f54 | 6454 | |
05589553 XW |
6455 | static struct fixed_file_ref_node *alloc_fixed_file_ref_node( |
6456 | struct io_ring_ctx *ctx) | |
05f3fb3c | 6457 | { |
05589553 | 6458 | struct fixed_file_ref_node *ref_node; |
05f3fb3c | 6459 | |
05589553 XW |
6460 | ref_node = kzalloc(sizeof(*ref_node), GFP_KERNEL); |
6461 | if (!ref_node) | |
6462 | return ERR_PTR(-ENOMEM); | |
05f3fb3c | 6463 | |
05589553 XW |
6464 | if (percpu_ref_init(&ref_node->refs, io_file_data_ref_zero, |
6465 | 0, GFP_KERNEL)) { | |
6466 | kfree(ref_node); | |
6467 | return ERR_PTR(-ENOMEM); | |
6468 | } | |
6469 | INIT_LIST_HEAD(&ref_node->node); | |
6470 | INIT_LIST_HEAD(&ref_node->file_list); | |
6471 | INIT_WORK(&ref_node->work, io_file_put_work); | |
6472 | ref_node->file_data = ctx->file_data; | |
6473 | return ref_node; | |
6474 | ||
6475 | } | |
6476 | ||
6477 | static void destroy_fixed_file_ref_node(struct fixed_file_ref_node *ref_node) | |
6478 | { | |
6479 | percpu_ref_exit(&ref_node->refs); | |
6480 | kfree(ref_node); | |
65e19f54 JA |
6481 | } |
6482 | ||
6b06314c JA |
6483 | static int io_sqe_files_register(struct io_ring_ctx *ctx, void __user *arg, |
6484 | unsigned nr_args) | |
6485 | { | |
6486 | __s32 __user *fds = (__s32 __user *) arg; | |
65e19f54 | 6487 | unsigned nr_tables; |
05f3fb3c | 6488 | struct file *file; |
6b06314c JA |
6489 | int fd, ret = 0; |
6490 | unsigned i; | |
05589553 XW |
6491 | struct fixed_file_ref_node *ref_node; |
6492 | unsigned long flags; | |
6b06314c | 6493 | |
05f3fb3c | 6494 | if (ctx->file_data) |
6b06314c JA |
6495 | return -EBUSY; |
6496 | if (!nr_args) | |
6497 | return -EINVAL; | |
6498 | if (nr_args > IORING_MAX_FIXED_FILES) | |
6499 | return -EMFILE; | |
6500 | ||
05f3fb3c JA |
6501 | ctx->file_data = kzalloc(sizeof(*ctx->file_data), GFP_KERNEL); |
6502 | if (!ctx->file_data) | |
6503 | return -ENOMEM; | |
6504 | ctx->file_data->ctx = ctx; | |
6505 | init_completion(&ctx->file_data->done); | |
05589553 | 6506 | INIT_LIST_HEAD(&ctx->file_data->ref_list); |
f7fe9346 | 6507 | spin_lock_init(&ctx->file_data->lock); |
05f3fb3c | 6508 | |
65e19f54 | 6509 | nr_tables = DIV_ROUND_UP(nr_args, IORING_MAX_FILES_TABLE); |
05f3fb3c JA |
6510 | ctx->file_data->table = kcalloc(nr_tables, |
6511 | sizeof(struct fixed_file_table), | |
65e19f54 | 6512 | GFP_KERNEL); |
05f3fb3c JA |
6513 | if (!ctx->file_data->table) { |
6514 | kfree(ctx->file_data); | |
6515 | ctx->file_data = NULL; | |
6b06314c | 6516 | return -ENOMEM; |
05f3fb3c JA |
6517 | } |
6518 | ||
05589553 | 6519 | if (percpu_ref_init(&ctx->file_data->refs, io_file_ref_kill, |
05f3fb3c JA |
6520 | PERCPU_REF_ALLOW_REINIT, GFP_KERNEL)) { |
6521 | kfree(ctx->file_data->table); | |
6522 | kfree(ctx->file_data); | |
6523 | ctx->file_data = NULL; | |
6b06314c | 6524 | return -ENOMEM; |
05f3fb3c | 6525 | } |
6b06314c | 6526 | |
65e19f54 | 6527 | if (io_sqe_alloc_file_tables(ctx, nr_tables, nr_args)) { |
05f3fb3c JA |
6528 | percpu_ref_exit(&ctx->file_data->refs); |
6529 | kfree(ctx->file_data->table); | |
6530 | kfree(ctx->file_data); | |
6531 | ctx->file_data = NULL; | |
65e19f54 JA |
6532 | return -ENOMEM; |
6533 | } | |
6534 | ||
08a45173 | 6535 | for (i = 0; i < nr_args; i++, ctx->nr_user_files++) { |
65e19f54 JA |
6536 | struct fixed_file_table *table; |
6537 | unsigned index; | |
6538 | ||
6b06314c JA |
6539 | ret = -EFAULT; |
6540 | if (copy_from_user(&fd, &fds[i], sizeof(fd))) | |
6541 | break; | |
08a45173 JA |
6542 | /* allow sparse sets */ |
6543 | if (fd == -1) { | |
6544 | ret = 0; | |
6545 | continue; | |
6546 | } | |
6b06314c | 6547 | |
05f3fb3c | 6548 | table = &ctx->file_data->table[i >> IORING_FILE_TABLE_SHIFT]; |
65e19f54 | 6549 | index = i & IORING_FILE_TABLE_MASK; |
05f3fb3c | 6550 | file = fget(fd); |
6b06314c JA |
6551 | |
6552 | ret = -EBADF; | |
05f3fb3c | 6553 | if (!file) |
6b06314c | 6554 | break; |
05f3fb3c | 6555 | |
6b06314c JA |
6556 | /* |
6557 | * Don't allow io_uring instances to be registered. If UNIX | |
6558 | * isn't enabled, then this causes a reference cycle and this | |
6559 | * instance can never get freed. If UNIX is enabled we'll | |
6560 | * handle it just fine, but there's still no point in allowing | |
6561 | * a ring fd as it doesn't support regular read/write anyway. | |
6562 | */ | |
05f3fb3c JA |
6563 | if (file->f_op == &io_uring_fops) { |
6564 | fput(file); | |
6b06314c JA |
6565 | break; |
6566 | } | |
6b06314c | 6567 | ret = 0; |
05f3fb3c | 6568 | table->files[index] = file; |
6b06314c JA |
6569 | } |
6570 | ||
6571 | if (ret) { | |
65e19f54 | 6572 | for (i = 0; i < ctx->nr_user_files; i++) { |
65e19f54 JA |
6573 | file = io_file_from_index(ctx, i); |
6574 | if (file) | |
6575 | fput(file); | |
6576 | } | |
6577 | for (i = 0; i < nr_tables; i++) | |
05f3fb3c | 6578 | kfree(ctx->file_data->table[i].files); |
6b06314c | 6579 | |
05f3fb3c JA |
6580 | kfree(ctx->file_data->table); |
6581 | kfree(ctx->file_data); | |
6582 | ctx->file_data = NULL; | |
6b06314c JA |
6583 | ctx->nr_user_files = 0; |
6584 | return ret; | |
6585 | } | |
6586 | ||
6587 | ret = io_sqe_files_scm(ctx); | |
05589553 | 6588 | if (ret) { |
6b06314c | 6589 | io_sqe_files_unregister(ctx); |
05589553 XW |
6590 | return ret; |
6591 | } | |
6b06314c | 6592 | |
05589553 XW |
6593 | ref_node = alloc_fixed_file_ref_node(ctx); |
6594 | if (IS_ERR(ref_node)) { | |
6595 | io_sqe_files_unregister(ctx); | |
6596 | return PTR_ERR(ref_node); | |
6597 | } | |
6598 | ||
6599 | ctx->file_data->cur_refs = &ref_node->refs; | |
6600 | spin_lock_irqsave(&ctx->file_data->lock, flags); | |
6601 | list_add(&ref_node->node, &ctx->file_data->ref_list); | |
6602 | spin_unlock_irqrestore(&ctx->file_data->lock, flags); | |
6603 | percpu_ref_get(&ctx->file_data->refs); | |
6b06314c JA |
6604 | return ret; |
6605 | } | |
6606 | ||
c3a31e60 JA |
6607 | static int io_sqe_file_register(struct io_ring_ctx *ctx, struct file *file, |
6608 | int index) | |
6609 | { | |
6610 | #if defined(CONFIG_UNIX) | |
6611 | struct sock *sock = ctx->ring_sock->sk; | |
6612 | struct sk_buff_head *head = &sock->sk_receive_queue; | |
6613 | struct sk_buff *skb; | |
6614 | ||
6615 | /* | |
6616 | * See if we can merge this file into an existing skb SCM_RIGHTS | |
6617 | * file set. If there's no room, fall back to allocating a new skb | |
6618 | * and filling it in. | |
6619 | */ | |
6620 | spin_lock_irq(&head->lock); | |
6621 | skb = skb_peek(head); | |
6622 | if (skb) { | |
6623 | struct scm_fp_list *fpl = UNIXCB(skb).fp; | |
6624 | ||
6625 | if (fpl->count < SCM_MAX_FD) { | |
6626 | __skb_unlink(skb, head); | |
6627 | spin_unlock_irq(&head->lock); | |
6628 | fpl->fp[fpl->count] = get_file(file); | |
6629 | unix_inflight(fpl->user, fpl->fp[fpl->count]); | |
6630 | fpl->count++; | |
6631 | spin_lock_irq(&head->lock); | |
6632 | __skb_queue_head(head, skb); | |
6633 | } else { | |
6634 | skb = NULL; | |
6635 | } | |
6636 | } | |
6637 | spin_unlock_irq(&head->lock); | |
6638 | ||
6639 | if (skb) { | |
6640 | fput(file); | |
6641 | return 0; | |
6642 | } | |
6643 | ||
6644 | return __io_sqe_files_scm(ctx, 1, index); | |
6645 | #else | |
6646 | return 0; | |
6647 | #endif | |
6648 | } | |
6649 | ||
a5318d3c | 6650 | static int io_queue_file_removal(struct fixed_file_data *data, |
05589553 | 6651 | struct file *file) |
05f3fb3c | 6652 | { |
a5318d3c | 6653 | struct io_file_put *pfile; |
05589553 XW |
6654 | struct percpu_ref *refs = data->cur_refs; |
6655 | struct fixed_file_ref_node *ref_node; | |
05f3fb3c | 6656 | |
05f3fb3c | 6657 | pfile = kzalloc(sizeof(*pfile), GFP_KERNEL); |
a5318d3c HD |
6658 | if (!pfile) |
6659 | return -ENOMEM; | |
05f3fb3c | 6660 | |
05589553 | 6661 | ref_node = container_of(refs, struct fixed_file_ref_node, refs); |
05f3fb3c | 6662 | pfile->file = file; |
05589553 XW |
6663 | list_add(&pfile->list, &ref_node->file_list); |
6664 | ||
a5318d3c | 6665 | return 0; |
05f3fb3c JA |
6666 | } |
6667 | ||
6668 | static int __io_sqe_files_update(struct io_ring_ctx *ctx, | |
6669 | struct io_uring_files_update *up, | |
6670 | unsigned nr_args) | |
6671 | { | |
6672 | struct fixed_file_data *data = ctx->file_data; | |
05589553 | 6673 | struct fixed_file_ref_node *ref_node; |
05f3fb3c | 6674 | struct file *file; |
c3a31e60 JA |
6675 | __s32 __user *fds; |
6676 | int fd, i, err; | |
6677 | __u32 done; | |
05589553 XW |
6678 | unsigned long flags; |
6679 | bool needs_switch = false; | |
c3a31e60 | 6680 | |
05f3fb3c | 6681 | if (check_add_overflow(up->offset, nr_args, &done)) |
c3a31e60 JA |
6682 | return -EOVERFLOW; |
6683 | if (done > ctx->nr_user_files) | |
6684 | return -EINVAL; | |
6685 | ||
05589553 XW |
6686 | ref_node = alloc_fixed_file_ref_node(ctx); |
6687 | if (IS_ERR(ref_node)) | |
6688 | return PTR_ERR(ref_node); | |
6689 | ||
c3a31e60 | 6690 | done = 0; |
05f3fb3c | 6691 | fds = u64_to_user_ptr(up->fds); |
c3a31e60 | 6692 | while (nr_args) { |
65e19f54 JA |
6693 | struct fixed_file_table *table; |
6694 | unsigned index; | |
6695 | ||
c3a31e60 JA |
6696 | err = 0; |
6697 | if (copy_from_user(&fd, &fds[done], sizeof(fd))) { | |
6698 | err = -EFAULT; | |
6699 | break; | |
6700 | } | |
05f3fb3c JA |
6701 | i = array_index_nospec(up->offset, ctx->nr_user_files); |
6702 | table = &ctx->file_data->table[i >> IORING_FILE_TABLE_SHIFT]; | |
65e19f54 JA |
6703 | index = i & IORING_FILE_TABLE_MASK; |
6704 | if (table->files[index]) { | |
05f3fb3c | 6705 | file = io_file_from_index(ctx, index); |
a5318d3c HD |
6706 | err = io_queue_file_removal(data, file); |
6707 | if (err) | |
6708 | break; | |
65e19f54 | 6709 | table->files[index] = NULL; |
05589553 | 6710 | needs_switch = true; |
c3a31e60 JA |
6711 | } |
6712 | if (fd != -1) { | |
c3a31e60 JA |
6713 | file = fget(fd); |
6714 | if (!file) { | |
6715 | err = -EBADF; | |
6716 | break; | |
6717 | } | |
6718 | /* | |
6719 | * Don't allow io_uring instances to be registered. If | |
6720 | * UNIX isn't enabled, then this causes a reference | |
6721 | * cycle and this instance can never get freed. If UNIX | |
6722 | * is enabled we'll handle it just fine, but there's | |
6723 | * still no point in allowing a ring fd as it doesn't | |
6724 | * support regular read/write anyway. | |
6725 | */ | |
6726 | if (file->f_op == &io_uring_fops) { | |
6727 | fput(file); | |
6728 | err = -EBADF; | |
6729 | break; | |
6730 | } | |
65e19f54 | 6731 | table->files[index] = file; |
c3a31e60 JA |
6732 | err = io_sqe_file_register(ctx, file, i); |
6733 | if (err) | |
6734 | break; | |
6735 | } | |
6736 | nr_args--; | |
6737 | done++; | |
05f3fb3c JA |
6738 | up->offset++; |
6739 | } | |
6740 | ||
05589553 XW |
6741 | if (needs_switch) { |
6742 | percpu_ref_kill(data->cur_refs); | |
6743 | spin_lock_irqsave(&data->lock, flags); | |
6744 | list_add(&ref_node->node, &data->ref_list); | |
6745 | data->cur_refs = &ref_node->refs; | |
6746 | spin_unlock_irqrestore(&data->lock, flags); | |
6747 | percpu_ref_get(&ctx->file_data->refs); | |
6748 | } else | |
6749 | destroy_fixed_file_ref_node(ref_node); | |
c3a31e60 JA |
6750 | |
6751 | return done ? done : err; | |
6752 | } | |
05589553 | 6753 | |
05f3fb3c JA |
6754 | static int io_sqe_files_update(struct io_ring_ctx *ctx, void __user *arg, |
6755 | unsigned nr_args) | |
6756 | { | |
6757 | struct io_uring_files_update up; | |
6758 | ||
6759 | if (!ctx->file_data) | |
6760 | return -ENXIO; | |
6761 | if (!nr_args) | |
6762 | return -EINVAL; | |
6763 | if (copy_from_user(&up, arg, sizeof(up))) | |
6764 | return -EFAULT; | |
6765 | if (up.resv) | |
6766 | return -EINVAL; | |
6767 | ||
6768 | return __io_sqe_files_update(ctx, &up, nr_args); | |
6769 | } | |
c3a31e60 | 6770 | |
e9fd9396 | 6771 | static void io_free_work(struct io_wq_work *work) |
7d723065 JA |
6772 | { |
6773 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); | |
6774 | ||
e9fd9396 | 6775 | /* Consider that io_steal_work() relies on this ref */ |
7d723065 JA |
6776 | io_put_req(req); |
6777 | } | |
6778 | ||
24369c2e PB |
6779 | static int io_init_wq_offload(struct io_ring_ctx *ctx, |
6780 | struct io_uring_params *p) | |
6781 | { | |
6782 | struct io_wq_data data; | |
6783 | struct fd f; | |
6784 | struct io_ring_ctx *ctx_attach; | |
6785 | unsigned int concurrency; | |
6786 | int ret = 0; | |
6787 | ||
6788 | data.user = ctx->user; | |
e9fd9396 | 6789 | data.free_work = io_free_work; |
24369c2e PB |
6790 | |
6791 | if (!(p->flags & IORING_SETUP_ATTACH_WQ)) { | |
6792 | /* Do QD, or 4 * CPUS, whatever is smallest */ | |
6793 | concurrency = min(ctx->sq_entries, 4 * num_online_cpus()); | |
6794 | ||
6795 | ctx->io_wq = io_wq_create(concurrency, &data); | |
6796 | if (IS_ERR(ctx->io_wq)) { | |
6797 | ret = PTR_ERR(ctx->io_wq); | |
6798 | ctx->io_wq = NULL; | |
6799 | } | |
6800 | return ret; | |
6801 | } | |
6802 | ||
6803 | f = fdget(p->wq_fd); | |
6804 | if (!f.file) | |
6805 | return -EBADF; | |
6806 | ||
6807 | if (f.file->f_op != &io_uring_fops) { | |
6808 | ret = -EINVAL; | |
6809 | goto out_fput; | |
6810 | } | |
6811 | ||
6812 | ctx_attach = f.file->private_data; | |
6813 | /* @io_wq is protected by holding the fd */ | |
6814 | if (!io_wq_get(ctx_attach->io_wq, &data)) { | |
6815 | ret = -EINVAL; | |
6816 | goto out_fput; | |
6817 | } | |
6818 | ||
6819 | ctx->io_wq = ctx_attach->io_wq; | |
6820 | out_fput: | |
6821 | fdput(f); | |
6822 | return ret; | |
6823 | } | |
6824 | ||
6c271ce2 JA |
6825 | static int io_sq_offload_start(struct io_ring_ctx *ctx, |
6826 | struct io_uring_params *p) | |
2b188cc1 JA |
6827 | { |
6828 | int ret; | |
6829 | ||
6c271ce2 | 6830 | init_waitqueue_head(&ctx->sqo_wait); |
2b188cc1 JA |
6831 | mmgrab(current->mm); |
6832 | ctx->sqo_mm = current->mm; | |
6833 | ||
6c271ce2 | 6834 | if (ctx->flags & IORING_SETUP_SQPOLL) { |
3ec482d1 JA |
6835 | ret = -EPERM; |
6836 | if (!capable(CAP_SYS_ADMIN)) | |
6837 | goto err; | |
6838 | ||
917257da JA |
6839 | ctx->sq_thread_idle = msecs_to_jiffies(p->sq_thread_idle); |
6840 | if (!ctx->sq_thread_idle) | |
6841 | ctx->sq_thread_idle = HZ; | |
6842 | ||
6c271ce2 | 6843 | if (p->flags & IORING_SETUP_SQ_AFF) { |
44a9bd18 | 6844 | int cpu = p->sq_thread_cpu; |
6c271ce2 | 6845 | |
917257da | 6846 | ret = -EINVAL; |
44a9bd18 JA |
6847 | if (cpu >= nr_cpu_ids) |
6848 | goto err; | |
7889f44d | 6849 | if (!cpu_online(cpu)) |
917257da JA |
6850 | goto err; |
6851 | ||
6c271ce2 JA |
6852 | ctx->sqo_thread = kthread_create_on_cpu(io_sq_thread, |
6853 | ctx, cpu, | |
6854 | "io_uring-sq"); | |
6855 | } else { | |
6856 | ctx->sqo_thread = kthread_create(io_sq_thread, ctx, | |
6857 | "io_uring-sq"); | |
6858 | } | |
6859 | if (IS_ERR(ctx->sqo_thread)) { | |
6860 | ret = PTR_ERR(ctx->sqo_thread); | |
6861 | ctx->sqo_thread = NULL; | |
6862 | goto err; | |
6863 | } | |
6864 | wake_up_process(ctx->sqo_thread); | |
6865 | } else if (p->flags & IORING_SETUP_SQ_AFF) { | |
6866 | /* Can't have SQ_AFF without SQPOLL */ | |
6867 | ret = -EINVAL; | |
6868 | goto err; | |
6869 | } | |
6870 | ||
24369c2e PB |
6871 | ret = io_init_wq_offload(ctx, p); |
6872 | if (ret) | |
2b188cc1 | 6873 | goto err; |
2b188cc1 JA |
6874 | |
6875 | return 0; | |
6876 | err: | |
54a91f3b | 6877 | io_finish_async(ctx); |
2b188cc1 JA |
6878 | mmdrop(ctx->sqo_mm); |
6879 | ctx->sqo_mm = NULL; | |
6880 | return ret; | |
6881 | } | |
6882 | ||
6883 | static void io_unaccount_mem(struct user_struct *user, unsigned long nr_pages) | |
6884 | { | |
6885 | atomic_long_sub(nr_pages, &user->locked_vm); | |
6886 | } | |
6887 | ||
6888 | static int io_account_mem(struct user_struct *user, unsigned long nr_pages) | |
6889 | { | |
6890 | unsigned long page_limit, cur_pages, new_pages; | |
6891 | ||
6892 | /* Don't allow more pages than we can safely lock */ | |
6893 | page_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT; | |
6894 | ||
6895 | do { | |
6896 | cur_pages = atomic_long_read(&user->locked_vm); | |
6897 | new_pages = cur_pages + nr_pages; | |
6898 | if (new_pages > page_limit) | |
6899 | return -ENOMEM; | |
6900 | } while (atomic_long_cmpxchg(&user->locked_vm, cur_pages, | |
6901 | new_pages) != cur_pages); | |
6902 | ||
6903 | return 0; | |
6904 | } | |
6905 | ||
6906 | static void io_mem_free(void *ptr) | |
6907 | { | |
52e04ef4 MR |
6908 | struct page *page; |
6909 | ||
6910 | if (!ptr) | |
6911 | return; | |
2b188cc1 | 6912 | |
52e04ef4 | 6913 | page = virt_to_head_page(ptr); |
2b188cc1 JA |
6914 | if (put_page_testzero(page)) |
6915 | free_compound_page(page); | |
6916 | } | |
6917 | ||
6918 | static void *io_mem_alloc(size_t size) | |
6919 | { | |
6920 | gfp_t gfp_flags = GFP_KERNEL | __GFP_ZERO | __GFP_NOWARN | __GFP_COMP | | |
6921 | __GFP_NORETRY; | |
6922 | ||
6923 | return (void *) __get_free_pages(gfp_flags, get_order(size)); | |
6924 | } | |
6925 | ||
75b28aff HV |
6926 | static unsigned long rings_size(unsigned sq_entries, unsigned cq_entries, |
6927 | size_t *sq_offset) | |
6928 | { | |
6929 | struct io_rings *rings; | |
6930 | size_t off, sq_array_size; | |
6931 | ||
6932 | off = struct_size(rings, cqes, cq_entries); | |
6933 | if (off == SIZE_MAX) | |
6934 | return SIZE_MAX; | |
6935 | ||
6936 | #ifdef CONFIG_SMP | |
6937 | off = ALIGN(off, SMP_CACHE_BYTES); | |
6938 | if (off == 0) | |
6939 | return SIZE_MAX; | |
6940 | #endif | |
6941 | ||
6942 | sq_array_size = array_size(sizeof(u32), sq_entries); | |
6943 | if (sq_array_size == SIZE_MAX) | |
6944 | return SIZE_MAX; | |
6945 | ||
6946 | if (check_add_overflow(off, sq_array_size, &off)) | |
6947 | return SIZE_MAX; | |
6948 | ||
6949 | if (sq_offset) | |
6950 | *sq_offset = off; | |
6951 | ||
6952 | return off; | |
6953 | } | |
6954 | ||
2b188cc1 JA |
6955 | static unsigned long ring_pages(unsigned sq_entries, unsigned cq_entries) |
6956 | { | |
75b28aff | 6957 | size_t pages; |
2b188cc1 | 6958 | |
75b28aff HV |
6959 | pages = (size_t)1 << get_order( |
6960 | rings_size(sq_entries, cq_entries, NULL)); | |
6961 | pages += (size_t)1 << get_order( | |
6962 | array_size(sizeof(struct io_uring_sqe), sq_entries)); | |
2b188cc1 | 6963 | |
75b28aff | 6964 | return pages; |
2b188cc1 JA |
6965 | } |
6966 | ||
edafccee JA |
6967 | static int io_sqe_buffer_unregister(struct io_ring_ctx *ctx) |
6968 | { | |
6969 | int i, j; | |
6970 | ||
6971 | if (!ctx->user_bufs) | |
6972 | return -ENXIO; | |
6973 | ||
6974 | for (i = 0; i < ctx->nr_user_bufs; i++) { | |
6975 | struct io_mapped_ubuf *imu = &ctx->user_bufs[i]; | |
6976 | ||
6977 | for (j = 0; j < imu->nr_bvecs; j++) | |
f1f6a7dd | 6978 | unpin_user_page(imu->bvec[j].bv_page); |
edafccee JA |
6979 | |
6980 | if (ctx->account_mem) | |
6981 | io_unaccount_mem(ctx->user, imu->nr_bvecs); | |
d4ef6475 | 6982 | kvfree(imu->bvec); |
edafccee JA |
6983 | imu->nr_bvecs = 0; |
6984 | } | |
6985 | ||
6986 | kfree(ctx->user_bufs); | |
6987 | ctx->user_bufs = NULL; | |
6988 | ctx->nr_user_bufs = 0; | |
6989 | return 0; | |
6990 | } | |
6991 | ||
6992 | static int io_copy_iov(struct io_ring_ctx *ctx, struct iovec *dst, | |
6993 | void __user *arg, unsigned index) | |
6994 | { | |
6995 | struct iovec __user *src; | |
6996 | ||
6997 | #ifdef CONFIG_COMPAT | |
6998 | if (ctx->compat) { | |
6999 | struct compat_iovec __user *ciovs; | |
7000 | struct compat_iovec ciov; | |
7001 | ||
7002 | ciovs = (struct compat_iovec __user *) arg; | |
7003 | if (copy_from_user(&ciov, &ciovs[index], sizeof(ciov))) | |
7004 | return -EFAULT; | |
7005 | ||
d55e5f5b | 7006 | dst->iov_base = u64_to_user_ptr((u64)ciov.iov_base); |
edafccee JA |
7007 | dst->iov_len = ciov.iov_len; |
7008 | return 0; | |
7009 | } | |
7010 | #endif | |
7011 | src = (struct iovec __user *) arg; | |
7012 | if (copy_from_user(dst, &src[index], sizeof(*dst))) | |
7013 | return -EFAULT; | |
7014 | return 0; | |
7015 | } | |
7016 | ||
7017 | static int io_sqe_buffer_register(struct io_ring_ctx *ctx, void __user *arg, | |
7018 | unsigned nr_args) | |
7019 | { | |
7020 | struct vm_area_struct **vmas = NULL; | |
7021 | struct page **pages = NULL; | |
7022 | int i, j, got_pages = 0; | |
7023 | int ret = -EINVAL; | |
7024 | ||
7025 | if (ctx->user_bufs) | |
7026 | return -EBUSY; | |
7027 | if (!nr_args || nr_args > UIO_MAXIOV) | |
7028 | return -EINVAL; | |
7029 | ||
7030 | ctx->user_bufs = kcalloc(nr_args, sizeof(struct io_mapped_ubuf), | |
7031 | GFP_KERNEL); | |
7032 | if (!ctx->user_bufs) | |
7033 | return -ENOMEM; | |
7034 | ||
7035 | for (i = 0; i < nr_args; i++) { | |
7036 | struct io_mapped_ubuf *imu = &ctx->user_bufs[i]; | |
7037 | unsigned long off, start, end, ubuf; | |
7038 | int pret, nr_pages; | |
7039 | struct iovec iov; | |
7040 | size_t size; | |
7041 | ||
7042 | ret = io_copy_iov(ctx, &iov, arg, i); | |
7043 | if (ret) | |
a278682d | 7044 | goto err; |
edafccee JA |
7045 | |
7046 | /* | |
7047 | * Don't impose further limits on the size and buffer | |
7048 | * constraints here, we'll -EINVAL later when IO is | |
7049 | * submitted if they are wrong. | |
7050 | */ | |
7051 | ret = -EFAULT; | |
7052 | if (!iov.iov_base || !iov.iov_len) | |
7053 | goto err; | |
7054 | ||
7055 | /* arbitrary limit, but we need something */ | |
7056 | if (iov.iov_len > SZ_1G) | |
7057 | goto err; | |
7058 | ||
7059 | ubuf = (unsigned long) iov.iov_base; | |
7060 | end = (ubuf + iov.iov_len + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
7061 | start = ubuf >> PAGE_SHIFT; | |
7062 | nr_pages = end - start; | |
7063 | ||
7064 | if (ctx->account_mem) { | |
7065 | ret = io_account_mem(ctx->user, nr_pages); | |
7066 | if (ret) | |
7067 | goto err; | |
7068 | } | |
7069 | ||
7070 | ret = 0; | |
7071 | if (!pages || nr_pages > got_pages) { | |
7072 | kfree(vmas); | |
7073 | kfree(pages); | |
d4ef6475 | 7074 | pages = kvmalloc_array(nr_pages, sizeof(struct page *), |
edafccee | 7075 | GFP_KERNEL); |
d4ef6475 | 7076 | vmas = kvmalloc_array(nr_pages, |
edafccee JA |
7077 | sizeof(struct vm_area_struct *), |
7078 | GFP_KERNEL); | |
7079 | if (!pages || !vmas) { | |
7080 | ret = -ENOMEM; | |
7081 | if (ctx->account_mem) | |
7082 | io_unaccount_mem(ctx->user, nr_pages); | |
7083 | goto err; | |
7084 | } | |
7085 | got_pages = nr_pages; | |
7086 | } | |
7087 | ||
d4ef6475 | 7088 | imu->bvec = kvmalloc_array(nr_pages, sizeof(struct bio_vec), |
edafccee JA |
7089 | GFP_KERNEL); |
7090 | ret = -ENOMEM; | |
7091 | if (!imu->bvec) { | |
7092 | if (ctx->account_mem) | |
7093 | io_unaccount_mem(ctx->user, nr_pages); | |
7094 | goto err; | |
7095 | } | |
7096 | ||
7097 | ret = 0; | |
7098 | down_read(¤t->mm->mmap_sem); | |
2113b05d | 7099 | pret = pin_user_pages(ubuf, nr_pages, |
932f4a63 IW |
7100 | FOLL_WRITE | FOLL_LONGTERM, |
7101 | pages, vmas); | |
edafccee JA |
7102 | if (pret == nr_pages) { |
7103 | /* don't support file backed memory */ | |
7104 | for (j = 0; j < nr_pages; j++) { | |
7105 | struct vm_area_struct *vma = vmas[j]; | |
7106 | ||
7107 | if (vma->vm_file && | |
7108 | !is_file_hugepages(vma->vm_file)) { | |
7109 | ret = -EOPNOTSUPP; | |
7110 | break; | |
7111 | } | |
7112 | } | |
7113 | } else { | |
7114 | ret = pret < 0 ? pret : -EFAULT; | |
7115 | } | |
7116 | up_read(¤t->mm->mmap_sem); | |
7117 | if (ret) { | |
7118 | /* | |
7119 | * if we did partial map, or found file backed vmas, | |
7120 | * release any pages we did get | |
7121 | */ | |
27c4d3a3 | 7122 | if (pret > 0) |
f1f6a7dd | 7123 | unpin_user_pages(pages, pret); |
edafccee JA |
7124 | if (ctx->account_mem) |
7125 | io_unaccount_mem(ctx->user, nr_pages); | |
d4ef6475 | 7126 | kvfree(imu->bvec); |
edafccee JA |
7127 | goto err; |
7128 | } | |
7129 | ||
7130 | off = ubuf & ~PAGE_MASK; | |
7131 | size = iov.iov_len; | |
7132 | for (j = 0; j < nr_pages; j++) { | |
7133 | size_t vec_len; | |
7134 | ||
7135 | vec_len = min_t(size_t, size, PAGE_SIZE - off); | |
7136 | imu->bvec[j].bv_page = pages[j]; | |
7137 | imu->bvec[j].bv_len = vec_len; | |
7138 | imu->bvec[j].bv_offset = off; | |
7139 | off = 0; | |
7140 | size -= vec_len; | |
7141 | } | |
7142 | /* store original address for later verification */ | |
7143 | imu->ubuf = ubuf; | |
7144 | imu->len = iov.iov_len; | |
7145 | imu->nr_bvecs = nr_pages; | |
7146 | ||
7147 | ctx->nr_user_bufs++; | |
7148 | } | |
d4ef6475 MR |
7149 | kvfree(pages); |
7150 | kvfree(vmas); | |
edafccee JA |
7151 | return 0; |
7152 | err: | |
d4ef6475 MR |
7153 | kvfree(pages); |
7154 | kvfree(vmas); | |
edafccee JA |
7155 | io_sqe_buffer_unregister(ctx); |
7156 | return ret; | |
7157 | } | |
7158 | ||
9b402849 JA |
7159 | static int io_eventfd_register(struct io_ring_ctx *ctx, void __user *arg) |
7160 | { | |
7161 | __s32 __user *fds = arg; | |
7162 | int fd; | |
7163 | ||
7164 | if (ctx->cq_ev_fd) | |
7165 | return -EBUSY; | |
7166 | ||
7167 | if (copy_from_user(&fd, fds, sizeof(*fds))) | |
7168 | return -EFAULT; | |
7169 | ||
7170 | ctx->cq_ev_fd = eventfd_ctx_fdget(fd); | |
7171 | if (IS_ERR(ctx->cq_ev_fd)) { | |
7172 | int ret = PTR_ERR(ctx->cq_ev_fd); | |
7173 | ctx->cq_ev_fd = NULL; | |
7174 | return ret; | |
7175 | } | |
7176 | ||
7177 | return 0; | |
7178 | } | |
7179 | ||
7180 | static int io_eventfd_unregister(struct io_ring_ctx *ctx) | |
7181 | { | |
7182 | if (ctx->cq_ev_fd) { | |
7183 | eventfd_ctx_put(ctx->cq_ev_fd); | |
7184 | ctx->cq_ev_fd = NULL; | |
7185 | return 0; | |
7186 | } | |
7187 | ||
7188 | return -ENXIO; | |
7189 | } | |
7190 | ||
5a2e745d JA |
7191 | static int __io_destroy_buffers(int id, void *p, void *data) |
7192 | { | |
7193 | struct io_ring_ctx *ctx = data; | |
7194 | struct io_buffer *buf = p; | |
7195 | ||
067524e9 | 7196 | __io_remove_buffers(ctx, buf, id, -1U); |
5a2e745d JA |
7197 | return 0; |
7198 | } | |
7199 | ||
7200 | static void io_destroy_buffers(struct io_ring_ctx *ctx) | |
7201 | { | |
7202 | idr_for_each(&ctx->io_buffer_idr, __io_destroy_buffers, ctx); | |
7203 | idr_destroy(&ctx->io_buffer_idr); | |
7204 | } | |
7205 | ||
2b188cc1 JA |
7206 | static void io_ring_ctx_free(struct io_ring_ctx *ctx) |
7207 | { | |
6b06314c | 7208 | io_finish_async(ctx); |
2b188cc1 JA |
7209 | if (ctx->sqo_mm) |
7210 | mmdrop(ctx->sqo_mm); | |
def596e9 JA |
7211 | |
7212 | io_iopoll_reap_events(ctx); | |
edafccee | 7213 | io_sqe_buffer_unregister(ctx); |
6b06314c | 7214 | io_sqe_files_unregister(ctx); |
9b402849 | 7215 | io_eventfd_unregister(ctx); |
5a2e745d | 7216 | io_destroy_buffers(ctx); |
41726c9a | 7217 | idr_destroy(&ctx->personality_idr); |
def596e9 | 7218 | |
2b188cc1 | 7219 | #if defined(CONFIG_UNIX) |
355e8d26 EB |
7220 | if (ctx->ring_sock) { |
7221 | ctx->ring_sock->file = NULL; /* so that iput() is called */ | |
2b188cc1 | 7222 | sock_release(ctx->ring_sock); |
355e8d26 | 7223 | } |
2b188cc1 JA |
7224 | #endif |
7225 | ||
75b28aff | 7226 | io_mem_free(ctx->rings); |
2b188cc1 | 7227 | io_mem_free(ctx->sq_sqes); |
2b188cc1 JA |
7228 | |
7229 | percpu_ref_exit(&ctx->refs); | |
7230 | if (ctx->account_mem) | |
7231 | io_unaccount_mem(ctx->user, | |
7232 | ring_pages(ctx->sq_entries, ctx->cq_entries)); | |
7233 | free_uid(ctx->user); | |
181e448d | 7234 | put_cred(ctx->creds); |
206aefde | 7235 | kfree(ctx->completions); |
78076bb6 | 7236 | kfree(ctx->cancel_hash); |
0ddf92e8 | 7237 | kmem_cache_free(req_cachep, ctx->fallback_req); |
2b188cc1 JA |
7238 | kfree(ctx); |
7239 | } | |
7240 | ||
7241 | static __poll_t io_uring_poll(struct file *file, poll_table *wait) | |
7242 | { | |
7243 | struct io_ring_ctx *ctx = file->private_data; | |
7244 | __poll_t mask = 0; | |
7245 | ||
7246 | poll_wait(file, &ctx->cq_wait, wait); | |
4f7067c3 SB |
7247 | /* |
7248 | * synchronizes with barrier from wq_has_sleeper call in | |
7249 | * io_commit_cqring | |
7250 | */ | |
2b188cc1 | 7251 | smp_rmb(); |
75b28aff HV |
7252 | if (READ_ONCE(ctx->rings->sq.tail) - ctx->cached_sq_head != |
7253 | ctx->rings->sq_ring_entries) | |
2b188cc1 | 7254 | mask |= EPOLLOUT | EPOLLWRNORM; |
63e5d81f | 7255 | if (io_cqring_events(ctx, false)) |
2b188cc1 JA |
7256 | mask |= EPOLLIN | EPOLLRDNORM; |
7257 | ||
7258 | return mask; | |
7259 | } | |
7260 | ||
7261 | static int io_uring_fasync(int fd, struct file *file, int on) | |
7262 | { | |
7263 | struct io_ring_ctx *ctx = file->private_data; | |
7264 | ||
7265 | return fasync_helper(fd, file, on, &ctx->cq_fasync); | |
7266 | } | |
7267 | ||
071698e1 JA |
7268 | static int io_remove_personalities(int id, void *p, void *data) |
7269 | { | |
7270 | struct io_ring_ctx *ctx = data; | |
7271 | const struct cred *cred; | |
7272 | ||
7273 | cred = idr_remove(&ctx->personality_idr, id); | |
7274 | if (cred) | |
7275 | put_cred(cred); | |
7276 | return 0; | |
7277 | } | |
7278 | ||
2b188cc1 JA |
7279 | static void io_ring_ctx_wait_and_kill(struct io_ring_ctx *ctx) |
7280 | { | |
7281 | mutex_lock(&ctx->uring_lock); | |
7282 | percpu_ref_kill(&ctx->refs); | |
7283 | mutex_unlock(&ctx->uring_lock); | |
7284 | ||
df069d80 JA |
7285 | /* |
7286 | * Wait for sq thread to idle, if we have one. It won't spin on new | |
7287 | * work after we've killed the ctx ref above. This is important to do | |
7288 | * before we cancel existing commands, as the thread could otherwise | |
7289 | * be queueing new work post that. If that's work we need to cancel, | |
7290 | * it could cause shutdown to hang. | |
7291 | */ | |
7292 | while (ctx->sqo_thread && !wq_has_sleeper(&ctx->sqo_wait)) | |
7293 | cpu_relax(); | |
7294 | ||
5262f567 | 7295 | io_kill_timeouts(ctx); |
221c5eb2 | 7296 | io_poll_remove_all(ctx); |
561fb04a JA |
7297 | |
7298 | if (ctx->io_wq) | |
7299 | io_wq_cancel_all(ctx->io_wq); | |
7300 | ||
def596e9 | 7301 | io_iopoll_reap_events(ctx); |
15dff286 JA |
7302 | /* if we failed setting up the ctx, we might not have any rings */ |
7303 | if (ctx->rings) | |
7304 | io_cqring_overflow_flush(ctx, true); | |
071698e1 | 7305 | idr_for_each(&ctx->personality_idr, io_remove_personalities, ctx); |
206aefde | 7306 | wait_for_completion(&ctx->completions[0]); |
2b188cc1 JA |
7307 | io_ring_ctx_free(ctx); |
7308 | } | |
7309 | ||
7310 | static int io_uring_release(struct inode *inode, struct file *file) | |
7311 | { | |
7312 | struct io_ring_ctx *ctx = file->private_data; | |
7313 | ||
7314 | file->private_data = NULL; | |
7315 | io_ring_ctx_wait_and_kill(ctx); | |
7316 | return 0; | |
7317 | } | |
7318 | ||
fcb323cc JA |
7319 | static void io_uring_cancel_files(struct io_ring_ctx *ctx, |
7320 | struct files_struct *files) | |
7321 | { | |
7322 | struct io_kiocb *req; | |
7323 | DEFINE_WAIT(wait); | |
7324 | ||
7325 | while (!list_empty_careful(&ctx->inflight_list)) { | |
768134d4 | 7326 | struct io_kiocb *cancel_req = NULL; |
fcb323cc JA |
7327 | |
7328 | spin_lock_irq(&ctx->inflight_lock); | |
7329 | list_for_each_entry(req, &ctx->inflight_list, inflight_entry) { | |
768134d4 JA |
7330 | if (req->work.files != files) |
7331 | continue; | |
7332 | /* req is being completed, ignore */ | |
7333 | if (!refcount_inc_not_zero(&req->refs)) | |
7334 | continue; | |
7335 | cancel_req = req; | |
7336 | break; | |
fcb323cc | 7337 | } |
768134d4 | 7338 | if (cancel_req) |
fcb323cc | 7339 | prepare_to_wait(&ctx->inflight_wait, &wait, |
768134d4 | 7340 | TASK_UNINTERRUPTIBLE); |
fcb323cc JA |
7341 | spin_unlock_irq(&ctx->inflight_lock); |
7342 | ||
768134d4 JA |
7343 | /* We need to keep going until we don't find a matching req */ |
7344 | if (!cancel_req) | |
fcb323cc | 7345 | break; |
2f6d9b9d | 7346 | |
2ca10259 JA |
7347 | if (cancel_req->flags & REQ_F_OVERFLOW) { |
7348 | spin_lock_irq(&ctx->completion_lock); | |
7349 | list_del(&cancel_req->list); | |
7350 | cancel_req->flags &= ~REQ_F_OVERFLOW; | |
7351 | if (list_empty(&ctx->cq_overflow_list)) { | |
7352 | clear_bit(0, &ctx->sq_check_overflow); | |
7353 | clear_bit(0, &ctx->cq_check_overflow); | |
7354 | } | |
7355 | spin_unlock_irq(&ctx->completion_lock); | |
7356 | ||
7357 | WRITE_ONCE(ctx->rings->cq_overflow, | |
7358 | atomic_inc_return(&ctx->cached_cq_overflow)); | |
7359 | ||
7360 | /* | |
7361 | * Put inflight ref and overflow ref. If that's | |
7362 | * all we had, then we're done with this request. | |
7363 | */ | |
7364 | if (refcount_sub_and_test(2, &cancel_req->refs)) { | |
7365 | io_put_req(cancel_req); | |
7366 | continue; | |
7367 | } | |
7368 | } | |
7369 | ||
2f6d9b9d BL |
7370 | io_wq_cancel_work(ctx->io_wq, &cancel_req->work); |
7371 | io_put_req(cancel_req); | |
fcb323cc JA |
7372 | schedule(); |
7373 | } | |
768134d4 | 7374 | finish_wait(&ctx->inflight_wait, &wait); |
fcb323cc JA |
7375 | } |
7376 | ||
7377 | static int io_uring_flush(struct file *file, void *data) | |
7378 | { | |
7379 | struct io_ring_ctx *ctx = file->private_data; | |
7380 | ||
7381 | io_uring_cancel_files(ctx, data); | |
6ab23144 JA |
7382 | |
7383 | /* | |
7384 | * If the task is going away, cancel work it may have pending | |
7385 | */ | |
7386 | if (fatal_signal_pending(current) || (current->flags & PF_EXITING)) | |
7387 | io_wq_cancel_pid(ctx->io_wq, task_pid_vnr(current)); | |
7388 | ||
fcb323cc JA |
7389 | return 0; |
7390 | } | |
7391 | ||
6c5c240e RP |
7392 | static void *io_uring_validate_mmap_request(struct file *file, |
7393 | loff_t pgoff, size_t sz) | |
2b188cc1 | 7394 | { |
2b188cc1 | 7395 | struct io_ring_ctx *ctx = file->private_data; |
6c5c240e | 7396 | loff_t offset = pgoff << PAGE_SHIFT; |
2b188cc1 JA |
7397 | struct page *page; |
7398 | void *ptr; | |
7399 | ||
7400 | switch (offset) { | |
7401 | case IORING_OFF_SQ_RING: | |
75b28aff HV |
7402 | case IORING_OFF_CQ_RING: |
7403 | ptr = ctx->rings; | |
2b188cc1 JA |
7404 | break; |
7405 | case IORING_OFF_SQES: | |
7406 | ptr = ctx->sq_sqes; | |
7407 | break; | |
2b188cc1 | 7408 | default: |
6c5c240e | 7409 | return ERR_PTR(-EINVAL); |
2b188cc1 JA |
7410 | } |
7411 | ||
7412 | page = virt_to_head_page(ptr); | |
a50b854e | 7413 | if (sz > page_size(page)) |
6c5c240e RP |
7414 | return ERR_PTR(-EINVAL); |
7415 | ||
7416 | return ptr; | |
7417 | } | |
7418 | ||
7419 | #ifdef CONFIG_MMU | |
7420 | ||
7421 | static int io_uring_mmap(struct file *file, struct vm_area_struct *vma) | |
7422 | { | |
7423 | size_t sz = vma->vm_end - vma->vm_start; | |
7424 | unsigned long pfn; | |
7425 | void *ptr; | |
7426 | ||
7427 | ptr = io_uring_validate_mmap_request(file, vma->vm_pgoff, sz); | |
7428 | if (IS_ERR(ptr)) | |
7429 | return PTR_ERR(ptr); | |
2b188cc1 JA |
7430 | |
7431 | pfn = virt_to_phys(ptr) >> PAGE_SHIFT; | |
7432 | return remap_pfn_range(vma, vma->vm_start, pfn, sz, vma->vm_page_prot); | |
7433 | } | |
7434 | ||
6c5c240e RP |
7435 | #else /* !CONFIG_MMU */ |
7436 | ||
7437 | static int io_uring_mmap(struct file *file, struct vm_area_struct *vma) | |
7438 | { | |
7439 | return vma->vm_flags & (VM_SHARED | VM_MAYSHARE) ? 0 : -EINVAL; | |
7440 | } | |
7441 | ||
7442 | static unsigned int io_uring_nommu_mmap_capabilities(struct file *file) | |
7443 | { | |
7444 | return NOMMU_MAP_DIRECT | NOMMU_MAP_READ | NOMMU_MAP_WRITE; | |
7445 | } | |
7446 | ||
7447 | static unsigned long io_uring_nommu_get_unmapped_area(struct file *file, | |
7448 | unsigned long addr, unsigned long len, | |
7449 | unsigned long pgoff, unsigned long flags) | |
7450 | { | |
7451 | void *ptr; | |
7452 | ||
7453 | ptr = io_uring_validate_mmap_request(file, pgoff, len); | |
7454 | if (IS_ERR(ptr)) | |
7455 | return PTR_ERR(ptr); | |
7456 | ||
7457 | return (unsigned long) ptr; | |
7458 | } | |
7459 | ||
7460 | #endif /* !CONFIG_MMU */ | |
7461 | ||
2b188cc1 JA |
7462 | SYSCALL_DEFINE6(io_uring_enter, unsigned int, fd, u32, to_submit, |
7463 | u32, min_complete, u32, flags, const sigset_t __user *, sig, | |
7464 | size_t, sigsz) | |
7465 | { | |
7466 | struct io_ring_ctx *ctx; | |
7467 | long ret = -EBADF; | |
7468 | int submitted = 0; | |
7469 | struct fd f; | |
7470 | ||
b41e9852 JA |
7471 | if (current->task_works) |
7472 | task_work_run(); | |
7473 | ||
6c271ce2 | 7474 | if (flags & ~(IORING_ENTER_GETEVENTS | IORING_ENTER_SQ_WAKEUP)) |
2b188cc1 JA |
7475 | return -EINVAL; |
7476 | ||
7477 | f = fdget(fd); | |
7478 | if (!f.file) | |
7479 | return -EBADF; | |
7480 | ||
7481 | ret = -EOPNOTSUPP; | |
7482 | if (f.file->f_op != &io_uring_fops) | |
7483 | goto out_fput; | |
7484 | ||
7485 | ret = -ENXIO; | |
7486 | ctx = f.file->private_data; | |
7487 | if (!percpu_ref_tryget(&ctx->refs)) | |
7488 | goto out_fput; | |
7489 | ||
6c271ce2 JA |
7490 | /* |
7491 | * For SQ polling, the thread will do all submissions and completions. | |
7492 | * Just return the requested submit count, and wake the thread if | |
7493 | * we were asked to. | |
7494 | */ | |
b2a9eada | 7495 | ret = 0; |
6c271ce2 | 7496 | if (ctx->flags & IORING_SETUP_SQPOLL) { |
c1edbf5f JA |
7497 | if (!list_empty_careful(&ctx->cq_overflow_list)) |
7498 | io_cqring_overflow_flush(ctx, false); | |
6c271ce2 JA |
7499 | if (flags & IORING_ENTER_SQ_WAKEUP) |
7500 | wake_up(&ctx->sqo_wait); | |
7501 | submitted = to_submit; | |
b2a9eada | 7502 | } else if (to_submit) { |
ae9428ca | 7503 | struct mm_struct *cur_mm; |
2b188cc1 JA |
7504 | |
7505 | mutex_lock(&ctx->uring_lock); | |
ae9428ca PB |
7506 | /* already have mm, so io_submit_sqes() won't try to grab it */ |
7507 | cur_mm = ctx->sqo_mm; | |
7508 | submitted = io_submit_sqes(ctx, to_submit, f.file, fd, | |
7509 | &cur_mm, false); | |
2b188cc1 | 7510 | mutex_unlock(&ctx->uring_lock); |
7c504e65 PB |
7511 | |
7512 | if (submitted != to_submit) | |
7513 | goto out; | |
2b188cc1 JA |
7514 | } |
7515 | if (flags & IORING_ENTER_GETEVENTS) { | |
def596e9 JA |
7516 | unsigned nr_events = 0; |
7517 | ||
2b188cc1 JA |
7518 | min_complete = min(min_complete, ctx->cq_entries); |
7519 | ||
32b2244a XW |
7520 | /* |
7521 | * When SETUP_IOPOLL and SETUP_SQPOLL are both enabled, user | |
7522 | * space applications don't need to do io completion events | |
7523 | * polling again, they can rely on io_sq_thread to do polling | |
7524 | * work, which can reduce cpu usage and uring_lock contention. | |
7525 | */ | |
7526 | if (ctx->flags & IORING_SETUP_IOPOLL && | |
7527 | !(ctx->flags & IORING_SETUP_SQPOLL)) { | |
def596e9 | 7528 | ret = io_iopoll_check(ctx, &nr_events, min_complete); |
def596e9 JA |
7529 | } else { |
7530 | ret = io_cqring_wait(ctx, min_complete, sig, sigsz); | |
7531 | } | |
2b188cc1 JA |
7532 | } |
7533 | ||
7c504e65 | 7534 | out: |
6805b32e | 7535 | percpu_ref_put(&ctx->refs); |
2b188cc1 JA |
7536 | out_fput: |
7537 | fdput(f); | |
7538 | return submitted ? submitted : ret; | |
7539 | } | |
7540 | ||
bebdb65e | 7541 | #ifdef CONFIG_PROC_FS |
87ce955b JA |
7542 | static int io_uring_show_cred(int id, void *p, void *data) |
7543 | { | |
7544 | const struct cred *cred = p; | |
7545 | struct seq_file *m = data; | |
7546 | struct user_namespace *uns = seq_user_ns(m); | |
7547 | struct group_info *gi; | |
7548 | kernel_cap_t cap; | |
7549 | unsigned __capi; | |
7550 | int g; | |
7551 | ||
7552 | seq_printf(m, "%5d\n", id); | |
7553 | seq_put_decimal_ull(m, "\tUid:\t", from_kuid_munged(uns, cred->uid)); | |
7554 | seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->euid)); | |
7555 | seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->suid)); | |
7556 | seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->fsuid)); | |
7557 | seq_put_decimal_ull(m, "\n\tGid:\t", from_kgid_munged(uns, cred->gid)); | |
7558 | seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->egid)); | |
7559 | seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->sgid)); | |
7560 | seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->fsgid)); | |
7561 | seq_puts(m, "\n\tGroups:\t"); | |
7562 | gi = cred->group_info; | |
7563 | for (g = 0; g < gi->ngroups; g++) { | |
7564 | seq_put_decimal_ull(m, g ? " " : "", | |
7565 | from_kgid_munged(uns, gi->gid[g])); | |
7566 | } | |
7567 | seq_puts(m, "\n\tCapEff:\t"); | |
7568 | cap = cred->cap_effective; | |
7569 | CAP_FOR_EACH_U32(__capi) | |
7570 | seq_put_hex_ll(m, NULL, cap.cap[CAP_LAST_U32 - __capi], 8); | |
7571 | seq_putc(m, '\n'); | |
7572 | return 0; | |
7573 | } | |
7574 | ||
7575 | static void __io_uring_show_fdinfo(struct io_ring_ctx *ctx, struct seq_file *m) | |
7576 | { | |
7577 | int i; | |
7578 | ||
7579 | mutex_lock(&ctx->uring_lock); | |
7580 | seq_printf(m, "UserFiles:\t%u\n", ctx->nr_user_files); | |
7581 | for (i = 0; i < ctx->nr_user_files; i++) { | |
7582 | struct fixed_file_table *table; | |
7583 | struct file *f; | |
7584 | ||
7585 | table = &ctx->file_data->table[i >> IORING_FILE_TABLE_SHIFT]; | |
7586 | f = table->files[i & IORING_FILE_TABLE_MASK]; | |
7587 | if (f) | |
7588 | seq_printf(m, "%5u: %s\n", i, file_dentry(f)->d_iname); | |
7589 | else | |
7590 | seq_printf(m, "%5u: <none>\n", i); | |
7591 | } | |
7592 | seq_printf(m, "UserBufs:\t%u\n", ctx->nr_user_bufs); | |
7593 | for (i = 0; i < ctx->nr_user_bufs; i++) { | |
7594 | struct io_mapped_ubuf *buf = &ctx->user_bufs[i]; | |
7595 | ||
7596 | seq_printf(m, "%5u: 0x%llx/%u\n", i, buf->ubuf, | |
7597 | (unsigned int) buf->len); | |
7598 | } | |
7599 | if (!idr_is_empty(&ctx->personality_idr)) { | |
7600 | seq_printf(m, "Personalities:\n"); | |
7601 | idr_for_each(&ctx->personality_idr, io_uring_show_cred, m); | |
7602 | } | |
d7718a9d JA |
7603 | seq_printf(m, "PollList:\n"); |
7604 | spin_lock_irq(&ctx->completion_lock); | |
7605 | for (i = 0; i < (1U << ctx->cancel_hash_bits); i++) { | |
7606 | struct hlist_head *list = &ctx->cancel_hash[i]; | |
7607 | struct io_kiocb *req; | |
7608 | ||
7609 | hlist_for_each_entry(req, list, hash_node) | |
7610 | seq_printf(m, " op=%d, task_works=%d\n", req->opcode, | |
7611 | req->task->task_works != NULL); | |
7612 | } | |
7613 | spin_unlock_irq(&ctx->completion_lock); | |
87ce955b JA |
7614 | mutex_unlock(&ctx->uring_lock); |
7615 | } | |
7616 | ||
7617 | static void io_uring_show_fdinfo(struct seq_file *m, struct file *f) | |
7618 | { | |
7619 | struct io_ring_ctx *ctx = f->private_data; | |
7620 | ||
7621 | if (percpu_ref_tryget(&ctx->refs)) { | |
7622 | __io_uring_show_fdinfo(ctx, m); | |
7623 | percpu_ref_put(&ctx->refs); | |
7624 | } | |
7625 | } | |
bebdb65e | 7626 | #endif |
87ce955b | 7627 | |
2b188cc1 JA |
7628 | static const struct file_operations io_uring_fops = { |
7629 | .release = io_uring_release, | |
fcb323cc | 7630 | .flush = io_uring_flush, |
2b188cc1 | 7631 | .mmap = io_uring_mmap, |
6c5c240e RP |
7632 | #ifndef CONFIG_MMU |
7633 | .get_unmapped_area = io_uring_nommu_get_unmapped_area, | |
7634 | .mmap_capabilities = io_uring_nommu_mmap_capabilities, | |
7635 | #endif | |
2b188cc1 JA |
7636 | .poll = io_uring_poll, |
7637 | .fasync = io_uring_fasync, | |
bebdb65e | 7638 | #ifdef CONFIG_PROC_FS |
87ce955b | 7639 | .show_fdinfo = io_uring_show_fdinfo, |
bebdb65e | 7640 | #endif |
2b188cc1 JA |
7641 | }; |
7642 | ||
7643 | static int io_allocate_scq_urings(struct io_ring_ctx *ctx, | |
7644 | struct io_uring_params *p) | |
7645 | { | |
75b28aff HV |
7646 | struct io_rings *rings; |
7647 | size_t size, sq_array_offset; | |
2b188cc1 | 7648 | |
75b28aff HV |
7649 | size = rings_size(p->sq_entries, p->cq_entries, &sq_array_offset); |
7650 | if (size == SIZE_MAX) | |
7651 | return -EOVERFLOW; | |
7652 | ||
7653 | rings = io_mem_alloc(size); | |
7654 | if (!rings) | |
2b188cc1 JA |
7655 | return -ENOMEM; |
7656 | ||
75b28aff HV |
7657 | ctx->rings = rings; |
7658 | ctx->sq_array = (u32 *)((char *)rings + sq_array_offset); | |
7659 | rings->sq_ring_mask = p->sq_entries - 1; | |
7660 | rings->cq_ring_mask = p->cq_entries - 1; | |
7661 | rings->sq_ring_entries = p->sq_entries; | |
7662 | rings->cq_ring_entries = p->cq_entries; | |
7663 | ctx->sq_mask = rings->sq_ring_mask; | |
7664 | ctx->cq_mask = rings->cq_ring_mask; | |
7665 | ctx->sq_entries = rings->sq_ring_entries; | |
7666 | ctx->cq_entries = rings->cq_ring_entries; | |
2b188cc1 JA |
7667 | |
7668 | size = array_size(sizeof(struct io_uring_sqe), p->sq_entries); | |
eb065d30 JA |
7669 | if (size == SIZE_MAX) { |
7670 | io_mem_free(ctx->rings); | |
7671 | ctx->rings = NULL; | |
2b188cc1 | 7672 | return -EOVERFLOW; |
eb065d30 | 7673 | } |
2b188cc1 JA |
7674 | |
7675 | ctx->sq_sqes = io_mem_alloc(size); | |
eb065d30 JA |
7676 | if (!ctx->sq_sqes) { |
7677 | io_mem_free(ctx->rings); | |
7678 | ctx->rings = NULL; | |
2b188cc1 | 7679 | return -ENOMEM; |
eb065d30 | 7680 | } |
2b188cc1 | 7681 | |
2b188cc1 JA |
7682 | return 0; |
7683 | } | |
7684 | ||
7685 | /* | |
7686 | * Allocate an anonymous fd, this is what constitutes the application | |
7687 | * visible backing of an io_uring instance. The application mmaps this | |
7688 | * fd to gain access to the SQ/CQ ring details. If UNIX sockets are enabled, | |
7689 | * we have to tie this fd to a socket for file garbage collection purposes. | |
7690 | */ | |
7691 | static int io_uring_get_fd(struct io_ring_ctx *ctx) | |
7692 | { | |
7693 | struct file *file; | |
7694 | int ret; | |
7695 | ||
7696 | #if defined(CONFIG_UNIX) | |
7697 | ret = sock_create_kern(&init_net, PF_UNIX, SOCK_RAW, IPPROTO_IP, | |
7698 | &ctx->ring_sock); | |
7699 | if (ret) | |
7700 | return ret; | |
7701 | #endif | |
7702 | ||
7703 | ret = get_unused_fd_flags(O_RDWR | O_CLOEXEC); | |
7704 | if (ret < 0) | |
7705 | goto err; | |
7706 | ||
7707 | file = anon_inode_getfile("[io_uring]", &io_uring_fops, ctx, | |
7708 | O_RDWR | O_CLOEXEC); | |
7709 | if (IS_ERR(file)) { | |
7710 | put_unused_fd(ret); | |
7711 | ret = PTR_ERR(file); | |
7712 | goto err; | |
7713 | } | |
7714 | ||
7715 | #if defined(CONFIG_UNIX) | |
7716 | ctx->ring_sock->file = file; | |
7717 | #endif | |
7718 | fd_install(ret, file); | |
7719 | return ret; | |
7720 | err: | |
7721 | #if defined(CONFIG_UNIX) | |
7722 | sock_release(ctx->ring_sock); | |
7723 | ctx->ring_sock = NULL; | |
7724 | #endif | |
7725 | return ret; | |
7726 | } | |
7727 | ||
7728 | static int io_uring_create(unsigned entries, struct io_uring_params *p) | |
7729 | { | |
7730 | struct user_struct *user = NULL; | |
7731 | struct io_ring_ctx *ctx; | |
7732 | bool account_mem; | |
7733 | int ret; | |
7734 | ||
8110c1a6 | 7735 | if (!entries) |
2b188cc1 | 7736 | return -EINVAL; |
8110c1a6 JA |
7737 | if (entries > IORING_MAX_ENTRIES) { |
7738 | if (!(p->flags & IORING_SETUP_CLAMP)) | |
7739 | return -EINVAL; | |
7740 | entries = IORING_MAX_ENTRIES; | |
7741 | } | |
2b188cc1 JA |
7742 | |
7743 | /* | |
7744 | * Use twice as many entries for the CQ ring. It's possible for the | |
7745 | * application to drive a higher depth than the size of the SQ ring, | |
7746 | * since the sqes are only used at submission time. This allows for | |
33a107f0 JA |
7747 | * some flexibility in overcommitting a bit. If the application has |
7748 | * set IORING_SETUP_CQSIZE, it will have passed in the desired number | |
7749 | * of CQ ring entries manually. | |
2b188cc1 JA |
7750 | */ |
7751 | p->sq_entries = roundup_pow_of_two(entries); | |
33a107f0 JA |
7752 | if (p->flags & IORING_SETUP_CQSIZE) { |
7753 | /* | |
7754 | * If IORING_SETUP_CQSIZE is set, we do the same roundup | |
7755 | * to a power-of-two, if it isn't already. We do NOT impose | |
7756 | * any cq vs sq ring sizing. | |
7757 | */ | |
8110c1a6 | 7758 | if (p->cq_entries < p->sq_entries) |
33a107f0 | 7759 | return -EINVAL; |
8110c1a6 JA |
7760 | if (p->cq_entries > IORING_MAX_CQ_ENTRIES) { |
7761 | if (!(p->flags & IORING_SETUP_CLAMP)) | |
7762 | return -EINVAL; | |
7763 | p->cq_entries = IORING_MAX_CQ_ENTRIES; | |
7764 | } | |
33a107f0 JA |
7765 | p->cq_entries = roundup_pow_of_two(p->cq_entries); |
7766 | } else { | |
7767 | p->cq_entries = 2 * p->sq_entries; | |
7768 | } | |
2b188cc1 JA |
7769 | |
7770 | user = get_uid(current_user()); | |
7771 | account_mem = !capable(CAP_IPC_LOCK); | |
7772 | ||
7773 | if (account_mem) { | |
7774 | ret = io_account_mem(user, | |
7775 | ring_pages(p->sq_entries, p->cq_entries)); | |
7776 | if (ret) { | |
7777 | free_uid(user); | |
7778 | return ret; | |
7779 | } | |
7780 | } | |
7781 | ||
7782 | ctx = io_ring_ctx_alloc(p); | |
7783 | if (!ctx) { | |
7784 | if (account_mem) | |
7785 | io_unaccount_mem(user, ring_pages(p->sq_entries, | |
7786 | p->cq_entries)); | |
7787 | free_uid(user); | |
7788 | return -ENOMEM; | |
7789 | } | |
7790 | ctx->compat = in_compat_syscall(); | |
7791 | ctx->account_mem = account_mem; | |
7792 | ctx->user = user; | |
0b8c0ec7 | 7793 | ctx->creds = get_current_cred(); |
2b188cc1 JA |
7794 | |
7795 | ret = io_allocate_scq_urings(ctx, p); | |
7796 | if (ret) | |
7797 | goto err; | |
7798 | ||
6c271ce2 | 7799 | ret = io_sq_offload_start(ctx, p); |
2b188cc1 JA |
7800 | if (ret) |
7801 | goto err; | |
7802 | ||
2b188cc1 | 7803 | memset(&p->sq_off, 0, sizeof(p->sq_off)); |
75b28aff HV |
7804 | p->sq_off.head = offsetof(struct io_rings, sq.head); |
7805 | p->sq_off.tail = offsetof(struct io_rings, sq.tail); | |
7806 | p->sq_off.ring_mask = offsetof(struct io_rings, sq_ring_mask); | |
7807 | p->sq_off.ring_entries = offsetof(struct io_rings, sq_ring_entries); | |
7808 | p->sq_off.flags = offsetof(struct io_rings, sq_flags); | |
7809 | p->sq_off.dropped = offsetof(struct io_rings, sq_dropped); | |
7810 | p->sq_off.array = (char *)ctx->sq_array - (char *)ctx->rings; | |
2b188cc1 JA |
7811 | |
7812 | memset(&p->cq_off, 0, sizeof(p->cq_off)); | |
75b28aff HV |
7813 | p->cq_off.head = offsetof(struct io_rings, cq.head); |
7814 | p->cq_off.tail = offsetof(struct io_rings, cq.tail); | |
7815 | p->cq_off.ring_mask = offsetof(struct io_rings, cq_ring_mask); | |
7816 | p->cq_off.ring_entries = offsetof(struct io_rings, cq_ring_entries); | |
7817 | p->cq_off.overflow = offsetof(struct io_rings, cq_overflow); | |
7818 | p->cq_off.cqes = offsetof(struct io_rings, cqes); | |
ac90f249 | 7819 | |
044c1ab3 JA |
7820 | /* |
7821 | * Install ring fd as the very last thing, so we don't risk someone | |
7822 | * having closed it before we finish setup | |
7823 | */ | |
7824 | ret = io_uring_get_fd(ctx); | |
7825 | if (ret < 0) | |
7826 | goto err; | |
7827 | ||
da8c9690 | 7828 | p->features = IORING_FEAT_SINGLE_MMAP | IORING_FEAT_NODROP | |
cccf0ee8 | 7829 | IORING_FEAT_SUBMIT_STABLE | IORING_FEAT_RW_CUR_POS | |
d7718a9d | 7830 | IORING_FEAT_CUR_PERSONALITY | IORING_FEAT_FAST_POLL; |
c826bd7a | 7831 | trace_io_uring_create(ret, ctx, p->sq_entries, p->cq_entries, p->flags); |
2b188cc1 JA |
7832 | return ret; |
7833 | err: | |
7834 | io_ring_ctx_wait_and_kill(ctx); | |
7835 | return ret; | |
7836 | } | |
7837 | ||
7838 | /* | |
7839 | * Sets up an aio uring context, and returns the fd. Applications asks for a | |
7840 | * ring size, we return the actual sq/cq ring sizes (among other things) in the | |
7841 | * params structure passed in. | |
7842 | */ | |
7843 | static long io_uring_setup(u32 entries, struct io_uring_params __user *params) | |
7844 | { | |
7845 | struct io_uring_params p; | |
7846 | long ret; | |
7847 | int i; | |
7848 | ||
7849 | if (copy_from_user(&p, params, sizeof(p))) | |
7850 | return -EFAULT; | |
7851 | for (i = 0; i < ARRAY_SIZE(p.resv); i++) { | |
7852 | if (p.resv[i]) | |
7853 | return -EINVAL; | |
7854 | } | |
7855 | ||
6c271ce2 | 7856 | if (p.flags & ~(IORING_SETUP_IOPOLL | IORING_SETUP_SQPOLL | |
8110c1a6 | 7857 | IORING_SETUP_SQ_AFF | IORING_SETUP_CQSIZE | |
24369c2e | 7858 | IORING_SETUP_CLAMP | IORING_SETUP_ATTACH_WQ)) |
2b188cc1 JA |
7859 | return -EINVAL; |
7860 | ||
7861 | ret = io_uring_create(entries, &p); | |
7862 | if (ret < 0) | |
7863 | return ret; | |
7864 | ||
7865 | if (copy_to_user(params, &p, sizeof(p))) | |
7866 | return -EFAULT; | |
7867 | ||
7868 | return ret; | |
7869 | } | |
7870 | ||
7871 | SYSCALL_DEFINE2(io_uring_setup, u32, entries, | |
7872 | struct io_uring_params __user *, params) | |
7873 | { | |
7874 | return io_uring_setup(entries, params); | |
7875 | } | |
7876 | ||
66f4af93 JA |
7877 | static int io_probe(struct io_ring_ctx *ctx, void __user *arg, unsigned nr_args) |
7878 | { | |
7879 | struct io_uring_probe *p; | |
7880 | size_t size; | |
7881 | int i, ret; | |
7882 | ||
7883 | size = struct_size(p, ops, nr_args); | |
7884 | if (size == SIZE_MAX) | |
7885 | return -EOVERFLOW; | |
7886 | p = kzalloc(size, GFP_KERNEL); | |
7887 | if (!p) | |
7888 | return -ENOMEM; | |
7889 | ||
7890 | ret = -EFAULT; | |
7891 | if (copy_from_user(p, arg, size)) | |
7892 | goto out; | |
7893 | ret = -EINVAL; | |
7894 | if (memchr_inv(p, 0, size)) | |
7895 | goto out; | |
7896 | ||
7897 | p->last_op = IORING_OP_LAST - 1; | |
7898 | if (nr_args > IORING_OP_LAST) | |
7899 | nr_args = IORING_OP_LAST; | |
7900 | ||
7901 | for (i = 0; i < nr_args; i++) { | |
7902 | p->ops[i].op = i; | |
7903 | if (!io_op_defs[i].not_supported) | |
7904 | p->ops[i].flags = IO_URING_OP_SUPPORTED; | |
7905 | } | |
7906 | p->ops_len = i; | |
7907 | ||
7908 | ret = 0; | |
7909 | if (copy_to_user(arg, p, size)) | |
7910 | ret = -EFAULT; | |
7911 | out: | |
7912 | kfree(p); | |
7913 | return ret; | |
7914 | } | |
7915 | ||
071698e1 JA |
7916 | static int io_register_personality(struct io_ring_ctx *ctx) |
7917 | { | |
7918 | const struct cred *creds = get_current_cred(); | |
7919 | int id; | |
7920 | ||
7921 | id = idr_alloc_cyclic(&ctx->personality_idr, (void *) creds, 1, | |
7922 | USHRT_MAX, GFP_KERNEL); | |
7923 | if (id < 0) | |
7924 | put_cred(creds); | |
7925 | return id; | |
7926 | } | |
7927 | ||
7928 | static int io_unregister_personality(struct io_ring_ctx *ctx, unsigned id) | |
7929 | { | |
7930 | const struct cred *old_creds; | |
7931 | ||
7932 | old_creds = idr_remove(&ctx->personality_idr, id); | |
7933 | if (old_creds) { | |
7934 | put_cred(old_creds); | |
7935 | return 0; | |
7936 | } | |
7937 | ||
7938 | return -EINVAL; | |
7939 | } | |
7940 | ||
7941 | static bool io_register_op_must_quiesce(int op) | |
7942 | { | |
7943 | switch (op) { | |
7944 | case IORING_UNREGISTER_FILES: | |
7945 | case IORING_REGISTER_FILES_UPDATE: | |
7946 | case IORING_REGISTER_PROBE: | |
7947 | case IORING_REGISTER_PERSONALITY: | |
7948 | case IORING_UNREGISTER_PERSONALITY: | |
7949 | return false; | |
7950 | default: | |
7951 | return true; | |
7952 | } | |
7953 | } | |
7954 | ||
edafccee JA |
7955 | static int __io_uring_register(struct io_ring_ctx *ctx, unsigned opcode, |
7956 | void __user *arg, unsigned nr_args) | |
b19062a5 JA |
7957 | __releases(ctx->uring_lock) |
7958 | __acquires(ctx->uring_lock) | |
edafccee JA |
7959 | { |
7960 | int ret; | |
7961 | ||
35fa71a0 JA |
7962 | /* |
7963 | * We're inside the ring mutex, if the ref is already dying, then | |
7964 | * someone else killed the ctx or is already going through | |
7965 | * io_uring_register(). | |
7966 | */ | |
7967 | if (percpu_ref_is_dying(&ctx->refs)) | |
7968 | return -ENXIO; | |
7969 | ||
071698e1 | 7970 | if (io_register_op_must_quiesce(opcode)) { |
05f3fb3c | 7971 | percpu_ref_kill(&ctx->refs); |
b19062a5 | 7972 | |
05f3fb3c JA |
7973 | /* |
7974 | * Drop uring mutex before waiting for references to exit. If | |
7975 | * another thread is currently inside io_uring_enter() it might | |
7976 | * need to grab the uring_lock to make progress. If we hold it | |
7977 | * here across the drain wait, then we can deadlock. It's safe | |
7978 | * to drop the mutex here, since no new references will come in | |
7979 | * after we've killed the percpu ref. | |
7980 | */ | |
7981 | mutex_unlock(&ctx->uring_lock); | |
c150368b | 7982 | ret = wait_for_completion_interruptible(&ctx->completions[0]); |
05f3fb3c | 7983 | mutex_lock(&ctx->uring_lock); |
c150368b JA |
7984 | if (ret) { |
7985 | percpu_ref_resurrect(&ctx->refs); | |
7986 | ret = -EINTR; | |
7987 | goto out; | |
7988 | } | |
05f3fb3c | 7989 | } |
edafccee JA |
7990 | |
7991 | switch (opcode) { | |
7992 | case IORING_REGISTER_BUFFERS: | |
7993 | ret = io_sqe_buffer_register(ctx, arg, nr_args); | |
7994 | break; | |
7995 | case IORING_UNREGISTER_BUFFERS: | |
7996 | ret = -EINVAL; | |
7997 | if (arg || nr_args) | |
7998 | break; | |
7999 | ret = io_sqe_buffer_unregister(ctx); | |
8000 | break; | |
6b06314c JA |
8001 | case IORING_REGISTER_FILES: |
8002 | ret = io_sqe_files_register(ctx, arg, nr_args); | |
8003 | break; | |
8004 | case IORING_UNREGISTER_FILES: | |
8005 | ret = -EINVAL; | |
8006 | if (arg || nr_args) | |
8007 | break; | |
8008 | ret = io_sqe_files_unregister(ctx); | |
8009 | break; | |
c3a31e60 JA |
8010 | case IORING_REGISTER_FILES_UPDATE: |
8011 | ret = io_sqe_files_update(ctx, arg, nr_args); | |
8012 | break; | |
9b402849 | 8013 | case IORING_REGISTER_EVENTFD: |
f2842ab5 | 8014 | case IORING_REGISTER_EVENTFD_ASYNC: |
9b402849 JA |
8015 | ret = -EINVAL; |
8016 | if (nr_args != 1) | |
8017 | break; | |
8018 | ret = io_eventfd_register(ctx, arg); | |
f2842ab5 JA |
8019 | if (ret) |
8020 | break; | |
8021 | if (opcode == IORING_REGISTER_EVENTFD_ASYNC) | |
8022 | ctx->eventfd_async = 1; | |
8023 | else | |
8024 | ctx->eventfd_async = 0; | |
9b402849 JA |
8025 | break; |
8026 | case IORING_UNREGISTER_EVENTFD: | |
8027 | ret = -EINVAL; | |
8028 | if (arg || nr_args) | |
8029 | break; | |
8030 | ret = io_eventfd_unregister(ctx); | |
8031 | break; | |
66f4af93 JA |
8032 | case IORING_REGISTER_PROBE: |
8033 | ret = -EINVAL; | |
8034 | if (!arg || nr_args > 256) | |
8035 | break; | |
8036 | ret = io_probe(ctx, arg, nr_args); | |
8037 | break; | |
071698e1 JA |
8038 | case IORING_REGISTER_PERSONALITY: |
8039 | ret = -EINVAL; | |
8040 | if (arg || nr_args) | |
8041 | break; | |
8042 | ret = io_register_personality(ctx); | |
8043 | break; | |
8044 | case IORING_UNREGISTER_PERSONALITY: | |
8045 | ret = -EINVAL; | |
8046 | if (arg) | |
8047 | break; | |
8048 | ret = io_unregister_personality(ctx, nr_args); | |
8049 | break; | |
edafccee JA |
8050 | default: |
8051 | ret = -EINVAL; | |
8052 | break; | |
8053 | } | |
8054 | ||
071698e1 | 8055 | if (io_register_op_must_quiesce(opcode)) { |
05f3fb3c | 8056 | /* bring the ctx back to life */ |
05f3fb3c | 8057 | percpu_ref_reinit(&ctx->refs); |
c150368b JA |
8058 | out: |
8059 | reinit_completion(&ctx->completions[0]); | |
05f3fb3c | 8060 | } |
edafccee JA |
8061 | return ret; |
8062 | } | |
8063 | ||
8064 | SYSCALL_DEFINE4(io_uring_register, unsigned int, fd, unsigned int, opcode, | |
8065 | void __user *, arg, unsigned int, nr_args) | |
8066 | { | |
8067 | struct io_ring_ctx *ctx; | |
8068 | long ret = -EBADF; | |
8069 | struct fd f; | |
8070 | ||
8071 | f = fdget(fd); | |
8072 | if (!f.file) | |
8073 | return -EBADF; | |
8074 | ||
8075 | ret = -EOPNOTSUPP; | |
8076 | if (f.file->f_op != &io_uring_fops) | |
8077 | goto out_fput; | |
8078 | ||
8079 | ctx = f.file->private_data; | |
8080 | ||
8081 | mutex_lock(&ctx->uring_lock); | |
8082 | ret = __io_uring_register(ctx, opcode, arg, nr_args); | |
8083 | mutex_unlock(&ctx->uring_lock); | |
c826bd7a DD |
8084 | trace_io_uring_register(ctx, opcode, ctx->nr_user_files, ctx->nr_user_bufs, |
8085 | ctx->cq_ev_fd != NULL, ret); | |
edafccee JA |
8086 | out_fput: |
8087 | fdput(f); | |
8088 | return ret; | |
8089 | } | |
8090 | ||
2b188cc1 JA |
8091 | static int __init io_uring_init(void) |
8092 | { | |
d7f62e82 SM |
8093 | #define __BUILD_BUG_VERIFY_ELEMENT(stype, eoffset, etype, ename) do { \ |
8094 | BUILD_BUG_ON(offsetof(stype, ename) != eoffset); \ | |
8095 | BUILD_BUG_ON(sizeof(etype) != sizeof_field(stype, ename)); \ | |
8096 | } while (0) | |
8097 | ||
8098 | #define BUILD_BUG_SQE_ELEM(eoffset, etype, ename) \ | |
8099 | __BUILD_BUG_VERIFY_ELEMENT(struct io_uring_sqe, eoffset, etype, ename) | |
8100 | BUILD_BUG_ON(sizeof(struct io_uring_sqe) != 64); | |
8101 | BUILD_BUG_SQE_ELEM(0, __u8, opcode); | |
8102 | BUILD_BUG_SQE_ELEM(1, __u8, flags); | |
8103 | BUILD_BUG_SQE_ELEM(2, __u16, ioprio); | |
8104 | BUILD_BUG_SQE_ELEM(4, __s32, fd); | |
8105 | BUILD_BUG_SQE_ELEM(8, __u64, off); | |
8106 | BUILD_BUG_SQE_ELEM(8, __u64, addr2); | |
8107 | BUILD_BUG_SQE_ELEM(16, __u64, addr); | |
7d67af2c | 8108 | BUILD_BUG_SQE_ELEM(16, __u64, splice_off_in); |
d7f62e82 SM |
8109 | BUILD_BUG_SQE_ELEM(24, __u32, len); |
8110 | BUILD_BUG_SQE_ELEM(28, __kernel_rwf_t, rw_flags); | |
8111 | BUILD_BUG_SQE_ELEM(28, /* compat */ int, rw_flags); | |
8112 | BUILD_BUG_SQE_ELEM(28, /* compat */ __u32, rw_flags); | |
8113 | BUILD_BUG_SQE_ELEM(28, __u32, fsync_flags); | |
8114 | BUILD_BUG_SQE_ELEM(28, __u16, poll_events); | |
8115 | BUILD_BUG_SQE_ELEM(28, __u32, sync_range_flags); | |
8116 | BUILD_BUG_SQE_ELEM(28, __u32, msg_flags); | |
8117 | BUILD_BUG_SQE_ELEM(28, __u32, timeout_flags); | |
8118 | BUILD_BUG_SQE_ELEM(28, __u32, accept_flags); | |
8119 | BUILD_BUG_SQE_ELEM(28, __u32, cancel_flags); | |
8120 | BUILD_BUG_SQE_ELEM(28, __u32, open_flags); | |
8121 | BUILD_BUG_SQE_ELEM(28, __u32, statx_flags); | |
8122 | BUILD_BUG_SQE_ELEM(28, __u32, fadvise_advice); | |
7d67af2c | 8123 | BUILD_BUG_SQE_ELEM(28, __u32, splice_flags); |
d7f62e82 SM |
8124 | BUILD_BUG_SQE_ELEM(32, __u64, user_data); |
8125 | BUILD_BUG_SQE_ELEM(40, __u16, buf_index); | |
8126 | BUILD_BUG_SQE_ELEM(42, __u16, personality); | |
7d67af2c | 8127 | BUILD_BUG_SQE_ELEM(44, __s32, splice_fd_in); |
d7f62e82 | 8128 | |
d3656344 | 8129 | BUILD_BUG_ON(ARRAY_SIZE(io_op_defs) != IORING_OP_LAST); |
84557871 | 8130 | BUILD_BUG_ON(__REQ_F_LAST_BIT >= 8 * sizeof(int)); |
2b188cc1 JA |
8131 | req_cachep = KMEM_CACHE(io_kiocb, SLAB_HWCACHE_ALIGN | SLAB_PANIC); |
8132 | return 0; | |
8133 | }; | |
8134 | __initcall(io_uring_init); |